712 files changed, 87354 insertions, 43428 deletions

diff --git a/lib/Target/AArch64/AArch64.h b/lib/Target/AArch64/AArch64.h
new file mode 100644
index 000000000000..4de4faa58182
--- /dev/null
+++ b/lib/Target/AArch64/AArch64.h
@@ -0,0 +1,42 @@
+//==-- AArch64.h - Top-level interface for AArch64 representation -*- C++ -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the entry points for global functions defined in the LLVM
+// AArch64 back-end.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_TARGET_AARCH64_H
+#define LLVM_TARGET_AARCH64_H
+
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "llvm/Target/TargetMachine.h"
+
+namespace llvm {
+
+class AArch64AsmPrinter;
+class FunctionPass;
+class AArch64TargetMachine;
+class MachineInstr;
+class MCInst;
+
+FunctionPass *createAArch64ISelDAG(AArch64TargetMachine &TM,
+                                   CodeGenOpt::Level OptLevel);
+
+FunctionPass *createAArch64CleanupLocalDynamicTLSPass();
+
+FunctionPass *createAArch64BranchFixupPass();
+
+void LowerAArch64MachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI,
+                                      AArch64AsmPrinter &AP);
+
+
+}
+
+#endif
diff --git a/lib/Target/AArch64/AArch64.td b/lib/Target/AArch64/AArch64.td
new file mode 100644
index 000000000000..e17052b4a565
--- /dev/null
+++ b/lib/Target/AArch64/AArch64.td
@@ -0,0 +1,70 @@
+//===- AArch64.td - Describe the AArch64 Target Machine -------*- tblgen -*-==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This is the top level entry point for the AArch64 target.
+//
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// Target-independent interfaces
+//===----------------------------------------------------------------------===//
+
+include "llvm/Target/Target.td"
+
+//===----------------------------------------------------------------------===//
+// AArch64 Subtarget features.
+//
+
+def FeatureNEON : SubtargetFeature<"neon", "HasNEON", "true",
+  "Enable Advanced SIMD instructions">;
+
+def FeatureCrypto : SubtargetFeature<"crypto", "HasCrypto", "true",
+  "Enable cryptographic instructions">;
+
+//===----------------------------------------------------------------------===//
+// AArch64 Processors
+//
+
+include "AArch64Schedule.td"
+
+def : Processor<"generic", GenericItineraries, [FeatureNEON, FeatureCrypto]>;
+
+//===----------------------------------------------------------------------===//
+// Register File Description
+//===----------------------------------------------------------------------===//
+
+include "AArch64RegisterInfo.td"
+
+include "AArch64CallingConv.td"
+
+//===----------------------------------------------------------------------===//
+// Instruction Descriptions
+//===----------------------------------------------------------------------===//
+
+include "AArch64InstrInfo.td"
+
+def AArch64InstrInfo : InstrInfo;
+
+//===----------------------------------------------------------------------===//
+// Assembly printer
+//===----------------------------------------------------------------------===//
+
+def A64InstPrinter : AsmWriter {
+  string AsmWriterClassName = "InstPrinter";
+  bit isMCAsmWriter = 1;
+}
+
+//===----------------------------------------------------------------------===//
+// Declare the target which we are implementing
+//===----------------------------------------------------------------------===//
+
+def AArch64 : Target {
+  let InstructionSet = AArch64InstrInfo;
+  let AssemblyWriters = [A64InstPrinter];
+}
diff --git a/lib/Target/AArch64/AArch64AsmPrinter.cpp b/lib/Target/AArch64/AArch64AsmPrinter.cpp
new file mode 100644
index 000000000000..47ebb826e0d0
--- /dev/null
+++ b/lib/Target/AArch64/AArch64AsmPrinter.cpp
@@ -0,0 +1,347 @@
+//===-- AArch64AsmPrinter.cpp - Print machine code to an AArch64 .s file --===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains a printer that converts from our internal representation
+// of machine-dependent LLVM code to GAS-format AArch64 assembly language.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "asm-printer"
+#include "AArch64AsmPrinter.h"
+#include "InstPrinter/AArch64InstPrinter.h"
+#include "llvm/DebugInfo.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/CodeGen/MachineModuleInfoImpls.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCSymbol.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Target/Mangler.h"
+
+using namespace llvm;
+
+MachineLocation
+AArch64AsmPrinter::getDebugValueLocation(const MachineInstr *MI) const {
+  // See emitFrameIndexDebugValue in InstrInfo for where this instruction is
+  // expected to be created.
+  assert(MI->getNumOperands() == 4 && MI->getOperand(0).isReg()
+         && MI->getOperand(1).isImm() && "unexpected custom DBG_VALUE");
+  return MachineLocation(MI->getOperand(0).getReg(),
+                         MI->getOperand(1).getImm());
+}
+
+/// Try to print a floating-point register as if it belonged to a specified
+/// register-class. For example the inline asm operand modifier "b" requires its
+/// argument to be printed as "bN".
+static bool printModifiedFPRAsmOperand(const MachineOperand &MO,
+                                       const TargetRegisterInfo *TRI,
+                                       const TargetRegisterClass &RegClass,
+                                       raw_ostream &O) {
+  if (!MO.isReg())
+    return true;
+
+  for (MCRegAliasIterator AR(MO.getReg(), TRI, true); AR.isValid(); ++AR) {
+    if (RegClass.contains(*AR)) {
+      O << AArch64InstPrinter::getRegisterName(*AR);
+      return false;
+    }
+  }
+  return true;
+}
+
+/// Implements the 'w' and 'x' inline asm operand modifiers, which print a GPR
+/// with the obvious type and an immediate 0 as either wzr or xzr.
+static bool printModifiedGPRAsmOperand(const MachineOperand &MO,
+                                       const TargetRegisterInfo *TRI,
+                                       const TargetRegisterClass &RegClass,
+                                       raw_ostream &O) {
+  char Prefix = &RegClass == &AArch64::GPR32RegClass ? 'w' : 'x';
+
+  if (MO.isImm() && MO.getImm() == 0) {
+    O << Prefix << "zr";
+    return false;
+  } else if (MO.isReg()) {
+    if (MO.getReg() == AArch64::XSP || MO.getReg() == AArch64::WSP) {
+      O << (Prefix == 'x' ? "sp" : "wsp");
+      return false;
+    }
+
+    for (MCRegAliasIterator AR(MO.getReg(), TRI, true); AR.isValid(); ++AR) {
+      if (RegClass.contains(*AR)) {
+        O << AArch64InstPrinter::getRegisterName(*AR);
+        return false;
+      }
+    }
+  }
+
+  return true;
+}
+
+bool AArch64AsmPrinter::printSymbolicAddress(const MachineOperand &MO,
+                                             bool PrintImmediatePrefix,
+                                             StringRef Suffix, raw_ostream &O) {
+  StringRef Name;
+  StringRef Modifier;
+  switch (MO.getType()) {
+  default:
+    llvm_unreachable("Unexpected operand for symbolic address constraint");
+  case MachineOperand::MO_GlobalAddress:
+    Name = Mang->getSymbol(MO.getGlobal())->getName();
+
+    // Global variables may be accessed either via a GOT or in various fun and
+    // interesting TLS-model specific ways. Set the prefix modifier as
+    // appropriate here.
+    if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(MO.getGlobal())) {
+      Reloc::Model RelocM = TM.getRelocationModel();
+      if (GV->isThreadLocal()) {
+        switch (TM.getTLSModel(GV)) {
+        case TLSModel::GeneralDynamic:
+          Modifier = "tlsdesc";
+          break;
+        case TLSModel::LocalDynamic:
+          Modifier = "dtprel";
+          break;
+        case TLSModel::InitialExec:
+          Modifier = "gottprel";
+          break;
+        case TLSModel::LocalExec:
+          Modifier = "tprel";
+          break;
+        }
+      } else if (Subtarget->GVIsIndirectSymbol(GV, RelocM)) {
+        Modifier = "got";
+      }
+    }
+    break;
+  case MachineOperand::MO_BlockAddress:
+    Name = GetBlockAddressSymbol(MO.getBlockAddress())->getName();
+    break;
+  case MachineOperand::MO_ExternalSymbol:
+    Name = MO.getSymbolName();
+    break;
+  case MachineOperand::MO_ConstantPoolIndex:
+    Name = GetCPISymbol(MO.getIndex())->getName();
+    break;
+  }
+
+  // Some instructions (notably ADRP) don't take the # prefix for
+  // immediates. Only print it if asked to.
+  if (PrintImmediatePrefix)
+    O << '#';
+
+  // Only need the joining "_" if both the prefix and the suffix are
+  // non-null. This little block simply takes care of the four possibly
+  // combinations involved there.
+  if (Modifier == "" && Suffix == "")
+    O << Name;
+  else if (Modifier == "" && Suffix != "")
+    O << ":" << Suffix << ':' << Name;
+  else if (Modifier != "" && Suffix == "")
+    O << ":" << Modifier << ':' << Name;
+  else
+    O << ":" << Modifier << '_' << Suffix << ':' << Name;
+
+  return false;
+}
+
+bool AArch64AsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNum,
+                                        unsigned AsmVariant,
+                                        const char *ExtraCode, raw_ostream &O) {
+  const TargetRegisterInfo *TRI = MF->getTarget().getRegisterInfo();
+  if (!ExtraCode || !ExtraCode[0]) {
+    // There's actually no operand modifier, which leads to a slightly eclectic
+    // set of behaviour which we have to handle here.
+    const MachineOperand &MO = MI->getOperand(OpNum);
+    switch (MO.getType()) {
+    default:
+      llvm_unreachable("Unexpected operand for inline assembly");
+    case MachineOperand::MO_Register:
+      // GCC prints the unmodified operand of a 'w' constraint as the vector
+      // register. Technically, we could allocate the argument as a VPR128, but
+      // that leads to extremely dodgy copies being generated to get the data
+      // there.
+      if (printModifiedFPRAsmOperand(MO, TRI, AArch64::VPR128RegClass, O))
+        O << AArch64InstPrinter::getRegisterName(MO.getReg());
+      break;
+    case MachineOperand::MO_Immediate:
+      O << '#' << MO.getImm();
+      break;
+    case MachineOperand::MO_FPImmediate:
+      assert(MO.getFPImm()->isExactlyValue(0.0) && "Only FP 0.0 expected");
+      O << "#0.0";
+      break;
+    case MachineOperand::MO_BlockAddress:
+    case MachineOperand::MO_ConstantPoolIndex:
+    case MachineOperand::MO_GlobalAddress:
+    case MachineOperand::MO_ExternalSymbol:
+      return printSymbolicAddress(MO, false, "", O);
+    }
+    return false;
+  }
+
+  // We have a real modifier to handle.
+  switch(ExtraCode[0]) {
+  default:
+    // See if this is a generic operand
+    return AsmPrinter::PrintAsmOperand(MI, OpNum, AsmVariant, ExtraCode, O);
+  case 'c': // Don't print "#" before an immediate operand.
+    if (!MI->getOperand(OpNum).isImm())
+      return true;
+    O << MI->getOperand(OpNum).getImm();
+    return false;
+  case 'w':
+    // Output 32-bit general register operand, constant zero as wzr, or stack
+    // pointer as wsp. Ignored when used with other operand types.
+    return printModifiedGPRAsmOperand(MI->getOperand(OpNum), TRI,
+                                      AArch64::GPR32RegClass, O);
+  case 'x':
+    // Output 64-bit general register operand, constant zero as xzr, or stack
+    // pointer as sp. Ignored when used with other operand types.
+    return printModifiedGPRAsmOperand(MI->getOperand(OpNum), TRI,
+                                      AArch64::GPR64RegClass, O);
+  case 'H':
+    // Output higher numbered of a 64-bit general register pair
+  case 'Q':
+    // Output least significant register of a 64-bit general register pair
+  case 'R':
+    // Output most significant register of a 64-bit general register pair
+
+    // FIXME note: these three operand modifiers will require, to some extent,
+    // adding a paired GPR64 register class. Initial investigation suggests that
+    // assertions are hit unless it has a type and is made legal for that type
+    // in ISelLowering. After that step is made, the number of modifications
+    // needed explodes (operation legality, calling conventions, stores, reg
+    // copies ...).
+    llvm_unreachable("FIXME: Unimplemented register pairs");
+  case 'b':
+    // Output 8-bit FP/SIMD scalar register operand, prefixed with b.
+    return printModifiedFPRAsmOperand(MI->getOperand(OpNum), TRI,
+                                      AArch64::FPR8RegClass, O);
+  case 'h':
+    // Output 16-bit FP/SIMD scalar register operand, prefixed with h.
+    return printModifiedFPRAsmOperand(MI->getOperand(OpNum), TRI,
+                                      AArch64::FPR16RegClass, O);
+  case 's':
+    // Output 32-bit FP/SIMD scalar register operand, prefixed with s.
+    return printModifiedFPRAsmOperand(MI->getOperand(OpNum), TRI,
+                                      AArch64::FPR32RegClass, O);
+  case 'd':
+    // Output 64-bit FP/SIMD scalar register operand, prefixed with d.
+    return printModifiedFPRAsmOperand(MI->getOperand(OpNum), TRI,
+                                      AArch64::FPR64RegClass, O);
+  case 'q':
+    // Output 128-bit FP/SIMD scalar register operand, prefixed with q.
+    return printModifiedFPRAsmOperand(MI->getOperand(OpNum), TRI,
+                                      AArch64::FPR128RegClass, O);
+  case 'A':
+    // Output symbolic address with appropriate relocation modifier (also
+    // suitable for ADRP).
+    return printSymbolicAddress(MI->getOperand(OpNum), false, "", O);
+  case 'L':
+    // Output bits 11:0 of symbolic address with appropriate :lo12: relocation
+    // modifier.
+    return printSymbolicAddress(MI->getOperand(OpNum), true, "lo12", O);
+  case 'G':
+    // Output bits 23:12 of symbolic address with appropriate :hi12: relocation
+    // modifier (currently only for TLS local exec).
+    return printSymbolicAddress(MI->getOperand(OpNum), true, "hi12", O);
+  }
+
+
+}
+
+bool AArch64AsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI,
+                                              unsigned OpNum,
+                                              unsigned AsmVariant,
+                                              const char *ExtraCode,
+                                              raw_ostream &O) {
+  // Currently both the memory constraints (m and Q) behave the same and amount
+  // to the address as a single register. In future, we may allow "m" to provide
+  // both a base and an offset.
+  const MachineOperand &MO = MI->getOperand(OpNum);
+  assert(MO.isReg() && "unexpected inline assembly memory operand");
+  O << '[' << AArch64InstPrinter::getRegisterName(MO.getReg()) << ']';
+  return false;
+}
+
+void AArch64AsmPrinter::PrintDebugValueComment(const MachineInstr *MI,
+                                               raw_ostream &OS) {
+  unsigned NOps = MI->getNumOperands();
+  assert(NOps==4);
+  OS << '\t' << MAI->getCommentString() << "DEBUG_VALUE: ";
+  // cast away const; DIetc do not take const operands for some reason.
+  DIVariable V(const_cast<MDNode *>(MI->getOperand(NOps-1).getMetadata()));
+  OS << V.getName();
+  OS << " <- ";
+  // Frame address.  Currently handles register +- offset only.
+  assert(MI->getOperand(0).isReg() && MI->getOperand(1).isImm());
+  OS << '[' << AArch64InstPrinter::getRegisterName(MI->getOperand(0).getReg());
+  OS << '+' << MI->getOperand(1).getImm();
+  OS << ']';
+  OS << "+" << MI->getOperand(NOps - 2).getImm();
+}
+
+
+#include "AArch64GenMCPseudoLowering.inc"
+
+void AArch64AsmPrinter::EmitInstruction(const MachineInstr *MI) {
+  // Do any auto-generated pseudo lowerings.
+  if (emitPseudoExpansionLowering(OutStreamer, MI))
+    return;
+
+  switch (MI->getOpcode()) {
+  case AArch64::DBG_VALUE: {
+    if (isVerbose() && OutStreamer.hasRawTextSupport()) {
+      SmallString<128> TmpStr;
+      raw_svector_ostream OS(TmpStr);
+      PrintDebugValueComment(MI, OS);
+      OutStreamer.EmitRawText(StringRef(OS.str()));
+    }
+    return;
+  }
+  }
+
+  MCInst TmpInst;
+  LowerAArch64MachineInstrToMCInst(MI, TmpInst, *this);
+  OutStreamer.EmitInstruction(TmpInst);
+}
+
+void AArch64AsmPrinter::EmitEndOfAsmFile(Module &M) {
+  if (Subtarget->isTargetELF()) {
+    const TargetLoweringObjectFileELF &TLOFELF =
+      static_cast<const TargetLoweringObjectFileELF &>(getObjFileLowering());
+
+    MachineModuleInfoELF &MMIELF = MMI->getObjFileInfo<MachineModuleInfoELF>();
+
+    // Output stubs for external and common global variables.
+    MachineModuleInfoELF::SymbolListTy Stubs = MMIELF.GetGVStubList();
+    if (!Stubs.empty()) {
+      OutStreamer.SwitchSection(TLOFELF.getDataRelSection());
+      const DataLayout *TD = TM.getDataLayout();
+
+      for (unsigned i = 0, e = Stubs.size(); i != e; ++i) {
+        OutStreamer.EmitLabel(Stubs[i].first);
+        OutStreamer.EmitSymbolValue(Stubs[i].second.getPointer(),
+                                    TD->getPointerSize(0), 0);
+      }
+      Stubs.clear();
+    }
+  }
+}
+
+bool AArch64AsmPrinter::runOnMachineFunction(MachineFunction &MF) {
+  return AsmPrinter::runOnMachineFunction(MF);
+}
+
+// Force static initialization.
+extern "C" void LLVMInitializeAArch64AsmPrinter() {
+    RegisterAsmPrinter<AArch64AsmPrinter> X(TheAArch64Target);
+}
+
diff --git a/lib/Target/AArch64/AArch64AsmPrinter.h b/lib/Target/AArch64/AArch64AsmPrinter.h
new file mode 100644
index 000000000000..af0c9fed066f
--- /dev/null
+++ b/lib/Target/AArch64/AArch64AsmPrinter.h
@@ -0,0 +1,80 @@
+// AArch64AsmPrinter.h - Print machine code to an AArch64 .s file -*- C++ -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the AArch64 assembly printer class.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_AARCH64ASMPRINTER_H
+#define LLVM_AARCH64ASMPRINTER_H
+
+#include "AArch64.h"
+#include "AArch64TargetMachine.h"
+#include "llvm/CodeGen/AsmPrinter.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/Support/Compiler.h"
+
+namespace llvm {
+
+class MCOperand;
+
+class LLVM_LIBRARY_VISIBILITY AArch64AsmPrinter : public AsmPrinter {
+
+  /// Subtarget - Keep a pointer to the AArch64Subtarget around so that we can
+  /// make the right decision when printing asm code for different targets.
+  const AArch64Subtarget *Subtarget;
+
+  // emitPseudoExpansionLowering - tblgen'erated.
+  bool emitPseudoExpansionLowering(MCStreamer &OutStreamer,
+                                   const MachineInstr *MI);
+
+  public:
+  explicit AArch64AsmPrinter(TargetMachine &TM, MCStreamer &Streamer)
+    : AsmPrinter(TM, Streamer) {
+    Subtarget = &TM.getSubtarget<AArch64Subtarget>();
+  }
+
+  bool lowerOperand(const MachineOperand &MO, MCOperand &MCOp) const;
+
+  MCOperand lowerSymbolOperand(const MachineOperand &MO,
+                               const MCSymbol *Sym) const;
+
+  void EmitInstruction(const MachineInstr *MI);
+  void EmitEndOfAsmFile(Module &M);
+
+  bool PrintAsmOperand(const MachineInstr *MI, unsigned OpNum,
+                       unsigned AsmVariant, const char *ExtraCode,
+                       raw_ostream &O);
+  bool PrintAsmMemoryOperand(const MachineInstr *MI, unsigned OpNum,
+                             unsigned AsmVariant, const char *ExtraCode,
+                             raw_ostream &O);
+
+  void PrintDebugValueComment(const MachineInstr *MI, raw_ostream &OS);
+
+  /// printSymbolicAddress - Given some kind of reasonably bare symbolic
+  /// reference, print out the appropriate asm string to represent it. If
+  /// appropriate, a relocation-specifier will be produced, composed of a
+  /// general class derived from the MO parameter and an instruction-specific
+  /// suffix, provided in Suffix. E.g. ":got_lo12:" if a Suffix of "lo12" is
+  /// given.
+  bool printSymbolicAddress(const MachineOperand &MO,
+                            bool PrintImmediatePrefix,
+                            StringRef Suffix, raw_ostream &O);
+
+  MachineLocation getDebugValueLocation(const MachineInstr *MI) const;
+
+  virtual const char *getPassName() const {
+    return "AArch64 Assembly Printer";
+  }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF);
+};
+} // end namespace llvm
+
+#endif
diff --git a/lib/Target/AArch64/AArch64BranchFixupPass.cpp b/lib/Target/AArch64/AArch64BranchFixupPass.cpp
new file mode 100644
index 000000000000..71233ba5c3dc
--- /dev/null
+++ b/lib/Target/AArch64/AArch64BranchFixupPass.cpp
@@ -0,0 +1,600 @@
+//===-- AArch64BranchFixupPass.cpp - AArch64 branch fixup -----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains a pass that fixes AArch64 branches which have ended up out
+// of range for their immediate operands.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "aarch64-branch-fixup"
+#include "AArch64.h"
+#include "AArch64InstrInfo.h"
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/Format.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/Statistic.h"
+using namespace llvm;
+
+STATISTIC(NumSplit,      "Number of uncond branches inserted");
+STATISTIC(NumCBrFixed,   "Number of cond branches fixed");
+
+/// Return the worst case padding that could result from unknown offset bits.
+/// This does not include alignment padding caused by known offset bits.
+///
+/// @param LogAlign log2(alignment)
+/// @param KnownBits Number of known low offset bits.
+static inline unsigned UnknownPadding(unsigned LogAlign, unsigned KnownBits) {
+  if (KnownBits < LogAlign)
+    return (1u << LogAlign) - (1u << KnownBits);
+  return 0;
+}
+
+namespace {
+  /// Due to limited PC-relative displacements, conditional branches to distant
+  /// blocks may need converting into an unconditional equivalent. For example:
+  ///     tbz w1, #0, far_away
+  /// becomes
+  ///     tbnz w1, #0, skip
+  ///     b far_away
+  ///   skip:
+  class AArch64BranchFixup : public MachineFunctionPass {
+    /// Information about the offset and size of a single basic block.
+    struct BasicBlockInfo {
+      /// Distance from the beginning of the function to the beginning of this
+      /// basic block.
+      ///
+      /// Offsets are computed assuming worst case padding before an aligned
+      /// block. This means that subtracting basic block offsets always gives a
+      /// conservative estimate of the real distance which may be smaller.
+      ///
+      /// Because worst case padding is used, the computed offset of an aligned
+      /// block may not actually be aligned.
+      unsigned Offset;
+
+      /// Size of the basic block in bytes.  If the block contains inline
+      /// assembly, this is a worst case estimate.
+      ///
+      /// The size does not include any alignment padding whether from the
+      /// beginning of the block, or from an aligned jump table at the end.
+      unsigned Size;
+
+      /// The number of low bits in Offset that are known to be exact.  The
+      /// remaining bits of Offset are an upper bound.
+      uint8_t KnownBits;
+
+      /// When non-zero, the block contains instructions (inline asm) of unknown
+      /// size.  The real size may be smaller than Size bytes by a multiple of 1
+      /// << Unalign.
+      uint8_t Unalign;
+
+      BasicBlockInfo() : Offset(0), Size(0), KnownBits(0), Unalign(0) {}
+
+      /// Compute the number of known offset bits internally to this block.
+      /// This number should be used to predict worst case padding when
+      /// splitting the block.
+      unsigned internalKnownBits() const {
+        unsigned Bits = Unalign ? Unalign : KnownBits;
+        // If the block size isn't a multiple of the known bits, assume the
+        // worst case padding.
+        if (Size & ((1u << Bits) - 1))
+          Bits = CountTrailingZeros_32(Size);
+        return Bits;
+      }
+
+      /// Compute the offset immediately following this block.  If LogAlign is
+      /// specified, return the offset the successor block will get if it has
+      /// this alignment.
+      unsigned postOffset(unsigned LogAlign = 0) const {
+        unsigned PO = Offset + Size;
+        if (!LogAlign)
+          return PO;
+        // Add alignment padding from the terminator.
+        return PO + UnknownPadding(LogAlign, internalKnownBits());
+      }
+
+      /// Compute the number of known low bits of postOffset.  If this block
+      /// contains inline asm, the number of known bits drops to the
+      /// instruction alignment.  An aligned terminator may increase the number
+      /// of know bits.
+      /// If LogAlign is given, also consider the alignment of the next block.
+      unsigned postKnownBits(unsigned LogAlign = 0) const {
+        return std::max(LogAlign, internalKnownBits());
+      }
+    };
+
+    std::vector<BasicBlockInfo> BBInfo;
+
+    /// One per immediate branch, keeping the machine instruction pointer,
+    /// conditional or unconditional, the max displacement, and (if IsCond is
+    /// true) the corresponding inverted branch opcode.
+    struct ImmBranch {
+      MachineInstr *MI;
+      unsigned OffsetBits : 31;
+      bool IsCond : 1;
+      ImmBranch(MachineInstr *mi, unsigned offsetbits, bool cond)
+        : MI(mi), OffsetBits(offsetbits), IsCond(cond) {}
+    };
+
+    /// Keep track of all the immediate branch instructions.
+    ///
+    std::vector<ImmBranch> ImmBranches;
+
+    MachineFunction *MF;
+    const AArch64InstrInfo *TII;
+  public:
+    static char ID;
+    AArch64BranchFixup() : MachineFunctionPass(ID) {}
+
+    virtual bool runOnMachineFunction(MachineFunction &MF);
+
+    virtual const char *getPassName() const {
+      return "AArch64 branch fixup pass";
+    }
+
+  private:
+    void initializeFunctionInfo();
+    MachineBasicBlock *splitBlockBeforeInstr(MachineInstr *MI);
+    void adjustBBOffsetsAfter(MachineBasicBlock *BB);
+    bool isBBInRange(MachineInstr *MI, MachineBasicBlock *BB,
+                     unsigned OffsetBits);
+    bool fixupImmediateBr(ImmBranch &Br);
+    bool fixupConditionalBr(ImmBranch &Br);
+
+    void computeBlockSize(MachineBasicBlock *MBB);
+    unsigned getOffsetOf(MachineInstr *MI) const;
+    void dumpBBs();
+    void verify();
+  };
+  char AArch64BranchFixup::ID = 0;
+}
+
+/// check BBOffsets
+void AArch64BranchFixup::verify() {
+#ifndef NDEBUG
+  for (MachineFunction::iterator MBBI = MF->begin(), E = MF->end();
+       MBBI != E; ++MBBI) {
+    MachineBasicBlock *MBB = MBBI;
+    unsigned MBBId = MBB->getNumber();
+    assert(!MBBId || BBInfo[MBBId - 1].postOffset() <= BBInfo[MBBId].Offset);
+  }
+#endif
+}
+
+/// print block size and offset information - debugging
+void AArch64BranchFixup::dumpBBs() {
+  DEBUG({
+    for (unsigned J = 0, E = BBInfo.size(); J !=E; ++J) {
+      const BasicBlockInfo &BBI = BBInfo[J];
+      dbgs() << format("%08x BB#%u\t", BBI.Offset, J)
+             << " kb=" << unsigned(BBI.KnownBits)
+             << " ua=" << unsigned(BBI.Unalign)
+             << format(" size=%#x\n", BBInfo[J].Size);
+    }
+  });
+}
+
+/// Returns an instance of the branch fixup pass.
+FunctionPass *llvm::createAArch64BranchFixupPass() {
+  return new AArch64BranchFixup();
+}
+
+bool AArch64BranchFixup::runOnMachineFunction(MachineFunction &mf) {
+  MF = &mf;
+  DEBUG(dbgs() << "***** AArch64BranchFixup ******");
+  TII = (const AArch64InstrInfo*)MF->getTarget().getInstrInfo();
+
+  // This pass invalidates liveness information when it splits basic blocks.
+  MF->getRegInfo().invalidateLiveness();
+
+  // Renumber all of the machine basic blocks in the function, guaranteeing that
+  // the numbers agree with the position of the block in the function.
+  MF->RenumberBlocks();
+
+  // Do the initial scan of the function, building up information about the
+  // sizes of each block and location of each immediate branch.
+  initializeFunctionInfo();
+
+  // Iteratively fix up branches until there is no change.
+  unsigned NoBRIters = 0;
+  bool MadeChange = false;
+  while (true) {
+    DEBUG(dbgs() << "Beginning iteration #" << NoBRIters << '\n');
+    bool BRChange = false;
+    for (unsigned i = 0, e = ImmBranches.size(); i != e; ++i)
+      BRChange |= fixupImmediateBr(ImmBranches[i]);
+    if (BRChange && ++NoBRIters > 30)
+      report_fatal_error("Branch Fix Up pass failed to converge!");
+    DEBUG(dumpBBs());
+
+    if (!BRChange)
+      break;
+    MadeChange = true;
+  }
+
+  // After a while, this might be made debug-only, but it is not expensive.
+  verify();
+
+  DEBUG(dbgs() << '\n'; dumpBBs());
+
+  BBInfo.clear();
+  ImmBranches.clear();
+
+  return MadeChange;
+}
+
+/// Return true if the specified basic block can fallthrough into the block
+/// immediately after it.
+static bool BBHasFallthrough(MachineBasicBlock *MBB) {
+  // Get the next machine basic block in the function.
+  MachineFunction::iterator MBBI = MBB;
+  // Can't fall off end of function.
+  if (llvm::next(MBBI) == MBB->getParent()->end())
+    return false;
+
+  MachineBasicBlock *NextBB = llvm::next(MBBI);
+  for (MachineBasicBlock::succ_iterator I = MBB->succ_begin(),
+       E = MBB->succ_end(); I != E; ++I)
+    if (*I == NextBB)
+      return true;
+
+  return false;
+}
+
+/// Do the initial scan of the function, building up information about the sizes
+/// of each block, and each immediate branch.
+void AArch64BranchFixup::initializeFunctionInfo() {
+  BBInfo.clear();
+  BBInfo.resize(MF->getNumBlockIDs());
+
+  // First thing, compute the size of all basic blocks, and see if the function
+  // has any inline assembly in it. If so, we have to be conservative about
+  // alignment assumptions, as we don't know for sure the size of any
+  // instructions in the inline assembly.
+  for (MachineFunction::iterator I = MF->begin(), E = MF->end(); I != E; ++I)
+    computeBlockSize(I);
+
+  // The known bits of the entry block offset are determined by the function
+  // alignment.
+  BBInfo.front().KnownBits = MF->getAlignment();
+
+  // Compute block offsets and known bits.
+  adjustBBOffsetsAfter(MF->begin());
+
+  // Now go back through the instructions and build up our data structures.
+  for (MachineFunction::iterator MBBI = MF->begin(), E = MF->end();
+       MBBI != E; ++MBBI) {
+    MachineBasicBlock &MBB = *MBBI;
+
+    for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
+         I != E; ++I) {
+      if (I->isDebugValue())
+        continue;
+
+      int Opc = I->getOpcode();
+      if (I->isBranch()) {
+        bool IsCond = false;
+
+        // The offsets encoded in instructions here scale by the instruction
+        // size (4 bytes), effectively increasing their range by 2 bits.
+        unsigned Bits = 0;
+        switch (Opc) {
+        default:
+          continue;  // Ignore other JT branches
+        case AArch64::TBZxii:
+        case AArch64::TBZwii:
+        case AArch64::TBNZxii:
+        case AArch64::TBNZwii:
+          IsCond = true;
+          Bits = 14 + 2;
+          break;
+        case AArch64::Bcc:
+        case AArch64::CBZx:
+        case AArch64::CBZw:
+        case AArch64::CBNZx:
+        case AArch64::CBNZw:
+          IsCond = true;
+          Bits = 19 + 2;
+          break;
+        case AArch64::Bimm:
+          Bits = 26 + 2;
+          break;
+        }
+
+        // Record this immediate branch.
+        ImmBranches.push_back(ImmBranch(I, Bits, IsCond));
+      }
+    }
+  }
+}
+
+/// Compute the size and some alignment information for MBB.  This function
+/// updates BBInfo directly.
+void AArch64BranchFixup::computeBlockSize(MachineBasicBlock *MBB) {
+  BasicBlockInfo &BBI = BBInfo[MBB->getNumber()];
+  BBI.Size = 0;
+  BBI.Unalign = 0;
+
+  for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end(); I != E;
+       ++I) {
+    BBI.Size += TII->getInstSizeInBytes(*I);
+    // For inline asm, GetInstSizeInBytes returns a conservative estimate.
+    // The actual size may be smaller, but still a multiple of the instr size.
+    if (I->isInlineAsm())
+      BBI.Unalign = 2;
+  }
+}
+
+/// Return the current offset of the specified machine instruction from the
+/// start of the function.  This offset changes as stuff is moved around inside
+/// the function.
+unsigned AArch64BranchFixup::getOffsetOf(MachineInstr *MI) const {
+  MachineBasicBlock *MBB = MI->getParent();
+
+  // The offset is composed of two things: the sum of the sizes of all MBB's
+  // before this instruction's block, and the offset from the start of the block
+  // it is in.
+  unsigned Offset = BBInfo[MBB->getNumber()].Offset;
+
+  // Sum instructions before MI in MBB.
+  for (MachineBasicBlock::iterator I = MBB->begin(); &*I != MI; ++I) {
+    assert(I != MBB->end() && "Didn't find MI in its own basic block?");
+    Offset += TII->getInstSizeInBytes(*I);
+  }
+  return Offset;
+}
+
+/// Split the basic block containing MI into two blocks, which are joined by
+/// an unconditional branch.  Update data structures and renumber blocks to
+/// account for this change and returns the newly created block.
+MachineBasicBlock *
+AArch64BranchFixup::splitBlockBeforeInstr(MachineInstr *MI) {
+  MachineBasicBlock *OrigBB = MI->getParent();
+
+  // Create a new MBB for the code after the OrigBB.
+  MachineBasicBlock *NewBB =
+    MF->CreateMachineBasicBlock(OrigBB->getBasicBlock());
+  MachineFunction::iterator MBBI = OrigBB; ++MBBI;
+  MF->insert(MBBI, NewBB);
+
+  // Splice the instructions starting with MI over to NewBB.
+  NewBB->splice(NewBB->end(), OrigBB, MI, OrigBB->end());
+
+  // Add an unconditional branch from OrigBB to NewBB.
+  // Note the new unconditional branch is not being recorded.
+  // There doesn't seem to be meaningful DebugInfo available; this doesn't
+  // correspond to anything in the source.
+  BuildMI(OrigBB, DebugLoc(), TII->get(AArch64::Bimm)).addMBB(NewBB);
+  ++NumSplit;
+
+  // Update the CFG.  All succs of OrigBB are now succs of NewBB.
+  NewBB->transferSuccessors(OrigBB);
+
+  // OrigBB branches to NewBB.
+  OrigBB->addSuccessor(NewBB);
+
+  // Update internal data structures to account for the newly inserted MBB.
+  MF->RenumberBlocks(NewBB);
+
+  // Insert an entry into BBInfo to align it properly with the (newly
+  // renumbered) block numbers.
+  BBInfo.insert(BBInfo.begin() + NewBB->getNumber(), BasicBlockInfo());
+
+  // Figure out how large the OrigBB is.  As the first half of the original
+  // block, it cannot contain a tablejump.  The size includes
+  // the new jump we added.  (It should be possible to do this without
+  // recounting everything, but it's very confusing, and this is rarely
+  // executed.)
+  computeBlockSize(OrigBB);
+
+  // Figure out how large the NewMBB is.  As the second half of the original
+  // block, it may contain a tablejump.
+  computeBlockSize(NewBB);
+
+  // All BBOffsets following these blocks must be modified.
+  adjustBBOffsetsAfter(OrigBB);
+
+  return NewBB;
+}
+
+void AArch64BranchFixup::adjustBBOffsetsAfter(MachineBasicBlock *BB) {
+  unsigned BBNum = BB->getNumber();
+  for(unsigned i = BBNum + 1, e = MF->getNumBlockIDs(); i < e; ++i) {
+    // Get the offset and known bits at the end of the layout predecessor.
+    // Include the alignment of the current block.
+    unsigned LogAlign = MF->getBlockNumbered(i)->getAlignment();
+    unsigned Offset = BBInfo[i - 1].postOffset(LogAlign);
+    unsigned KnownBits = BBInfo[i - 1].postKnownBits(LogAlign);
+
+    // This is where block i begins.  Stop if the offset is already correct,
+    // and we have updated 2 blocks.  This is the maximum number of blocks
+    // changed before calling this function.
+    if (i > BBNum + 2 &&
+        BBInfo[i].Offset == Offset &&
+        BBInfo[i].KnownBits == KnownBits)
+      break;
+
+    BBInfo[i].Offset = Offset;
+    BBInfo[i].KnownBits = KnownBits;
+  }
+}
+
+/// Returns true if the distance between specific MI and specific BB can fit in
+/// MI's displacement field.
+bool AArch64BranchFixup::isBBInRange(MachineInstr *MI,
+                                     MachineBasicBlock *DestBB,
+                                     unsigned OffsetBits) {
+  int64_t BrOffset   = getOffsetOf(MI);
+  int64_t DestOffset = BBInfo[DestBB->getNumber()].Offset;
+
+  DEBUG(dbgs() << "Branch of destination BB#" << DestBB->getNumber()
+               << " from BB#" << MI->getParent()->getNumber()
+               << " bits available=" << OffsetBits
+               << " from " << getOffsetOf(MI) << " to " << DestOffset
+               << " offset " << int(DestOffset-BrOffset) << "\t" << *MI);
+
+  return isIntN(OffsetBits, DestOffset - BrOffset);
+}
+
+/// Fix up an immediate branch whose destination is too far away to fit in its
+/// displacement field.
+bool AArch64BranchFixup::fixupImmediateBr(ImmBranch &Br) {
+  MachineInstr *MI = Br.MI;
+  MachineBasicBlock *DestBB = 0;
+  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+    if (MI->getOperand(i).isMBB()) {
+      DestBB = MI->getOperand(i).getMBB();
+      break;
+    }
+  }
+  assert(DestBB && "Branch with no destination BB?");
+
+  // Check to see if the DestBB is already in-range.
+  if (isBBInRange(MI, DestBB, Br.OffsetBits))
+    return false;
+
+  assert(Br.IsCond && "Only conditional branches should need fixup");
+  return fixupConditionalBr(Br);
+}
+
+/// Fix up a conditional branch whose destination is too far away to fit in its
+/// displacement field. It is converted to an inverse conditional branch + an
+/// unconditional branch to the destination.
+bool
+AArch64BranchFixup::fixupConditionalBr(ImmBranch &Br) {
+  MachineInstr *MI = Br.MI;
+  MachineBasicBlock *MBB = MI->getParent();
+  unsigned CondBrMBBOperand = 0;
+
+  // The general idea is to add an unconditional branch to the destination and
+  // invert the conditional branch to jump over it. Complications occur around
+  // fallthrough and unreachable ends to the block.
+  //   b.lt L1
+  //   =>
+  //   b.ge L2
+  //   b   L1
+  // L2:
+
+  // First we invert the conditional branch, by creating a replacement if
+  // necessary. This if statement contains all the special handling of different
+  // branch types.
+  if (MI->getOpcode() == AArch64::Bcc) {
+    // The basic block is operand number 1 for Bcc
+    CondBrMBBOperand = 1;
+
+    A64CC::CondCodes CC = (A64CC::CondCodes)MI->getOperand(0).getImm();
+    CC = A64InvertCondCode(CC);
+    MI->getOperand(0).setImm(CC);
+  } else {
+    MachineInstrBuilder InvertedMI;
+    int InvertedOpcode;
+    switch (MI->getOpcode()) {
+    default: llvm_unreachable("Unknown branch type");
+    case AArch64::TBZxii: InvertedOpcode = AArch64::TBNZxii; break;
+    case AArch64::TBZwii: InvertedOpcode = AArch64::TBNZwii; break;
+    case AArch64::TBNZxii: InvertedOpcode = AArch64::TBZxii; break;
+    case AArch64::TBNZwii: InvertedOpcode = AArch64::TBZwii; break;
+    case AArch64::CBZx: InvertedOpcode = AArch64::CBNZx; break;
+    case AArch64::CBZw: InvertedOpcode = AArch64::CBNZw; break;
+    case AArch64::CBNZx: InvertedOpcode = AArch64::CBZx; break;
+    case AArch64::CBNZw: InvertedOpcode = AArch64::CBZw; break;
+    }
+
+    InvertedMI = BuildMI(*MBB, MI, MI->getDebugLoc(), TII->get(InvertedOpcode));
+    for (unsigned i = 0, e= MI->getNumOperands(); i != e; ++i) {
+      InvertedMI.addOperand(MI->getOperand(i));
+      if (MI->getOperand(i).isMBB())
+        CondBrMBBOperand = i;
+    }
+
+    MI->eraseFromParent();
+    MI = Br.MI = InvertedMI;
+  }
+
+  // If the branch is at the end of its MBB and that has a fall-through block,
+  // direct the updated conditional branch to the fall-through
+  // block. Otherwise, split the MBB before the next instruction.
+  MachineInstr *BMI = &MBB->back();
+  bool NeedSplit = (BMI != MI) || !BBHasFallthrough(MBB);
+
+  ++NumCBrFixed;
+  if (BMI != MI) {
+    if (llvm::next(MachineBasicBlock::iterator(MI)) == prior(MBB->end()) &&
+        BMI->getOpcode() == AArch64::Bimm) {
+      // Last MI in the BB is an unconditional branch. We can swap destinations:
+      // b.eq L1 (temporarily b.ne L1 after first change)
+      // b   L2
+      // =>
+      // b.ne L2
+      // b   L1
+      MachineBasicBlock *NewDest = BMI->getOperand(0).getMBB();
+      if (isBBInRange(MI, NewDest, Br.OffsetBits)) {
+        DEBUG(dbgs() << "  Invert Bcc condition and swap its destination with "
+                     << *BMI);
+        MachineBasicBlock *DestBB = MI->getOperand(CondBrMBBOperand).getMBB();
+        BMI->getOperand(0).setMBB(DestBB);
+        MI->getOperand(CondBrMBBOperand).setMBB(NewDest);
+        return true;
+      }
+    }
+  }
+
+  if (NeedSplit) {
+    MachineBasicBlock::iterator MBBI = MI; ++MBBI;
+    splitBlockBeforeInstr(MBBI);
+    // No need for the branch to the next block. We're adding an unconditional
+    // branch to the destination.
+    int delta = TII->getInstSizeInBytes(MBB->back());
+    BBInfo[MBB->getNumber()].Size -= delta;
+    MBB->back().eraseFromParent();
+    // BBInfo[SplitBB].Offset is wrong temporarily, fixed below
+  }
+
+  // After splitting and removing the unconditional branch from the original BB,
+  // the structure is now:
+  // oldbb:
+  //   [things]
+  //   b.invertedCC L1
+  // splitbb/fallthroughbb:
+  //   [old b L2/real continuation]
+  //
+  // We now have to change the conditional branch to point to splitbb and add an
+  // unconditional branch after it to L1, giving the final structure:
+  // oldbb:
+  //   [things]
+  //   b.invertedCC splitbb
+  //   b L1
+  // splitbb/fallthroughbb:
+  //   [old b L2/real continuation]
+  MachineBasicBlock *NextBB = llvm::next(MachineFunction::iterator(MBB));
+
+  DEBUG(dbgs() << "  Insert B to BB#"
+               << MI->getOperand(CondBrMBBOperand).getMBB()->getNumber()
+               << " also invert condition and change dest. to BB#"
+               << NextBB->getNumber() << "\n");
+
+  // Insert a new unconditional branch and fixup the destination of the
+  // conditional one.  Also update the ImmBranch as well as adding a new entry
+  // for the new branch.
+  BuildMI(MBB, DebugLoc(), TII->get(AArch64::Bimm))
+    .addMBB(MI->getOperand(CondBrMBBOperand).getMBB());
+  MI->getOperand(CondBrMBBOperand).setMBB(NextBB);
+
+  BBInfo[MBB->getNumber()].Size += TII->getInstSizeInBytes(MBB->back());
+
+  // 26 bits written down in Bimm, specifying a multiple of 4.
+  unsigned OffsetBits = 26 + 2;
+  ImmBranches.push_back(ImmBranch(&MBB->back(), OffsetBits, false));
+
+  adjustBBOffsetsAfter(MBB);
+  return true;
+}
diff --git a/lib/Target/AArch64/AArch64CallingConv.td b/lib/Target/AArch64/AArch64CallingConv.td
new file mode 100644
index 000000000000..b880d8373deb
--- /dev/null
+++ b/lib/Target/AArch64/AArch64CallingConv.td
@@ -0,0 +1,196 @@
+//==-- AArch64CallingConv.td - Calling Conventions for ARM ----*- tblgen -*-==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+// This describes the calling conventions for AArch64 architecture.
+//===----------------------------------------------------------------------===//
+
+
+// The AArch64 Procedure Call Standard is unfortunately specified at a slightly
+// higher level of abstraction than LLVM's target interface presents. In
+// particular, it refers (like other ABIs, in fact) directly to
+// structs. However, generic LLVM code takes the liberty of lowering structure
+// arguments to the component fields before we see them.
+//
+// As a result, the obvious direct map from LLVM IR to PCS concepts can't be
+// implemented, so the goals of this calling convention are, in decreasing
+// priority order:
+//     1. Expose *some* way to express the concepts required to implement the
+//        generic PCS from a front-end.
+//     2. Provide a sane ABI for pure LLVM.
+//     3. Follow the generic PCS as closely as is naturally possible.
+//
+// The suggested front-end implementation of PCS features is:
+//     * Integer, float and vector arguments of all sizes which end up in
+//       registers are passed and returned via the natural LLVM type.
+//     * Structure arguments with size <= 16 bytes are passed and returned in
+//       registers as similar integer or composite types. For example:
+//       [1 x i64], [2 x i64] or [1 x i128] (if alignment 16 needed).
+//     * HFAs in registers follow rules similar to small structs: appropriate
+//       composite types.
+//     * Structure arguments with size > 16 bytes are passed via a pointer,
+//       handled completely by the front-end.
+//     * Structure return values > 16 bytes via an sret pointer argument.
+//     * Other stack-based arguments (not large structs) are passed using byval
+//       pointers. Padding arguments are added beforehand to guarantee a large
+//       struct doesn't later use integer registers.
+//
+// N.b. this means that it is the front-end's responsibility (if it cares about
+// PCS compliance) to check whether enough registers are available for an
+// argument when deciding how to pass it.
+
+class CCIfAlign<int Align, CCAction A>:
+  CCIf<"ArgFlags.getOrigAlign() == " # Align, A>;
+
+def CC_A64_APCS : CallingConv<[
+  // SRet is an LLVM-specific concept, so it takes precedence over general ABI
+  // concerns. However, this rule will be used by C/C++ frontends to implement
+  // structure return.
+  CCIfSRet<CCAssignToReg<[X8]>>,
+
+  // Put ByVal arguments directly on the stack. Minimum size and alignment of a
+  // slot is 64-bit.
+  CCIfByVal<CCPassByVal<8, 8>>,
+
+  // Canonicalise the various types that live in different floating-point
+  // registers. This makes sense because the PCS does not distinguish Short
+  // Vectors and Floating-point types.
+  CCIfType<[v2i8], CCBitConvertToType<f16>>,
+  CCIfType<[v4i8, v2i16], CCBitConvertToType<f32>>,
+  CCIfType<[v8i8, v4i16, v2i32, v2f32], CCBitConvertToType<f64>>,
+  CCIfType<[v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
+           CCBitConvertToType<f128>>,
+
+  // PCS: "C.1: If the argument is a Half-, Single-, Double- or Quad- precision
+  // Floating-point or Short Vector Type and the NSRN is less than 8, then the
+  // argument is allocated to the least significant bits of register
+  // v[NSRN]. The NSRN is incremented by one. The argument has now been
+  // allocated."
+  CCIfType<[f16],  CCAssignToReg<[B0, B1, B2, B3, B4, B5, B6, B7]>>,
+  CCIfType<[f32],  CCAssignToReg<[S0, S1, S2, S3, S4, S5, S6, S7]>>,
+  CCIfType<[f64],  CCAssignToReg<[D0, D1, D2, D3, D4, D5, D6, D7]>>,
+  CCIfType<[f128], CCAssignToReg<[Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]>>,
+
+  // PCS: "C.2: If the argument is an HFA and there are sufficient unallocated
+  // SIMD and Floating-point registers (NSRN - number of elements < 8), then the
+  // argument is allocated to SIMD and Floating-point registers (with one
+  // register per element of the HFA). The NSRN is incremented by the number of
+  // registers used. The argument has now been allocated."
+  //
+  // N.b. As above, this rule is the responsibility of the front-end.
+
+  // "C.3: If the argument is an HFA then the NSRN is set to 8 and the size of
+  // the argument is rounded up to the nearest multiple of 8 bytes."
+  //
+  // "C.4: If the argument is an HFA, a Quad-precision Floating-point or Short
+  // Vector Type then the NSAA is rounded up to the larger of 8 or the Natural
+  // Alignment of the Argument's type."
+  //
+  // It is expected that these will be satisfied by adding dummy arguments to
+  // the prototype.
+
+  // PCS: "C.5: If the argument is a Half- or Single- precision Floating-point
+  // type then the size of the argument is set to 8 bytes. The effect is as if
+  // the argument had been copied to the least significant bits of a 64-bit
+  // register and the remaining bits filled with unspecified values."
+  CCIfType<[f16, f32], CCPromoteToType<f64>>,
+
+  // PCS: "C.6: If the argument is an HFA, a Half-, Single-, Double- or Quad-
+  // precision Floating-point or Short Vector Type, then the argument is copied
+  // to memory at the adjusted NSAA. The NSAA is incremented by the size of the
+  // argument. The argument has now been allocated."
+  CCIfType<[f64], CCAssignToStack<8, 8>>,
+  CCIfType<[f128], CCAssignToStack<16, 16>>,
+
+  // PCS: "C.7: If the argument is an Integral Type, the size of the argument is
+  // less than or equal to 8 bytes and the NGRN is less than 8, the argument is
+  // copied to the least significant bits of x[NGRN]. The NGRN is incremented by
+  // one. The argument has now been allocated."
+
+  // First we implement C.8 and C.9 (128-bit types get even registers). i128 is
+  // represented as two i64s, the first one being split. If we delayed this
+  // operation C.8 would never be reached.
+  CCIfType<[i64],
+        CCIfSplit<CCAssignToRegWithShadow<[X0, X2, X4, X6], [X0, X1, X3, X5]>>>,
+
+  // Note: the promotion also implements C.14.
+  CCIfType<[i8, i16, i32], CCPromoteToType<i64>>,
+
+  // And now the real implementation of C.7
+  CCIfType<[i64], CCAssignToReg<[X0, X1, X2, X3, X4, X5, X6, X7]>>,
+
+  // PCS: "C.8: If the argument has an alignment of 16 then the NGRN is rounded
+  // up to the next even number."
+  //
+  // "C.9: If the argument is an Integral Type, the size of the argument is
+  // equal to 16 and the NGRN is less than 7, the argument is copied to x[NGRN]
+  // and x[NGRN+1], x[NGRN] shall contain the lower addressed double-word of the
+  // memory representation of the argument. The NGRN is incremented by two. The
+  // argument has now been allocated."
+  //
+  // Subtlety here: what if alignment is 16 but it is not an integral type? All
+  // floating-point types have been allocated already, which leaves composite
+  // types: this is why a front-end may need to produce i128 for a struct <= 16
+  // bytes.
+
+  // PCS: "C.10 If the argument is a Composite Type and the size in double-words
+  // of the argument is not more than 8 minus NGRN, then the argument is copied
+  // into consecutive general-purpose registers, starting at x[NGRN]. The
+  // argument is passed as though it had been loaded into the registers from a
+  // double-word aligned address with an appropriate sequence of LDR
+  // instructions loading consecutive registers from memory (the contents of any
+  // unused parts of the registers are unspecified by this standard). The NGRN
+  // is incremented by the number of registers used. The argument has now been
+  // allocated."
+  //
+  // Another one that's the responsibility of the front-end (sigh).
+
+  // PCS: "C.11: The NGRN is set to 8."
+  CCCustom<"CC_AArch64NoMoreRegs">,
+
+  // PCS: "C.12: The NSAA is rounded up to the larger of 8 or the Natural
+  // Alignment of the argument's type."
+  //
+  // PCS: "C.13: If the argument is a composite type then the argument is copied
+  // to memory at the adjusted NSAA. The NSAA is by the size of the
+  // argument. The argument has now been allocated."
+  //
+  // Note that the effect of this corresponds to a memcpy rather than register
+  // stores so that the struct ends up correctly addressable at the adjusted
+  // NSAA.
+
+  // PCS: "C.14: If the size of the argument is less than 8 bytes then the size
+  // of the argument is set to 8 bytes. The effect is as if the argument was
+  // copied to the least significant bits of a 64-bit register and the remaining
+  // bits filled with unspecified values."
+  //
+  // Integer types were widened above. Floating-point and composite types have
+  // already been allocated completely. Nothing to do.
+
+  // PCS: "C.15: The argument is copied to memory at the adjusted NSAA. The NSAA
+  // is incremented by the size of the argument. The argument has now been
+  // allocated."
+  CCIfType<[i64], CCIfSplit<CCAssignToStack<8, 16>>>,
+  CCIfType<[i64], CCAssignToStack<8, 8>>
+
+]>;
+
+// According to the PCS, X19-X30 are callee-saved, however only the low 64-bits
+// of vector registers (8-15) are callee-saved. The order here is is picked up
+// by PrologEpilogInserter.cpp to allocate stack slots, starting from top of
+// stack upon entry. This gives the customary layout of x30 at [sp-8], x29 at
+// [sp-16], ...
+def CSR_PCS : CalleeSavedRegs<(add (sequence "X%u", 30, 19),
+                                   (sequence "D%u", 15, 8))>;
+
+
+// TLS descriptor calls are extremely restricted in their changes, to allow
+// optimisations in the (hopefully) more common fast path where no real action
+// is needed. They actually have to preserve all registers, except for the
+// unavoidable X30 and the return register X0.
+def TLSDesc : CalleeSavedRegs<(add (sequence "X%u", 29, 1),
+                                   (sequence "Q%u", 31, 0))>;
diff --git a/lib/Target/AArch64/AArch64FrameLowering.cpp b/lib/Target/AArch64/AArch64FrameLowering.cpp
new file mode 100644
index 000000000000..dc41f2f60525
--- /dev/null
+++ b/lib/Target/AArch64/AArch64FrameLowering.cpp
@@ -0,0 +1,633 @@
+//===- AArch64FrameLowering.cpp - AArch64 Frame Information ---------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the AArch64 implementation of TargetFrameLowering class.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AArch64.h"
+#include "AArch64FrameLowering.h"
+#include "AArch64MachineFunctionInfo.h"
+#include "AArch64InstrInfo.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineMemOperand.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/IR/Function.h"
+#include "llvm/MC/MachineLocation.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+
+using namespace llvm;
+
+void AArch64FrameLowering::splitSPAdjustments(uint64_t Total,
+                                              uint64_t &Initial,
+                                              uint64_t &Residual) const {
+  // 0x1f0 here is a pessimistic (i.e. realistic) boundary: x-register LDP
+  // instructions have a 7-bit signed immediate scaled by 8, giving a reach of
+  // 0x1f8, but stack adjustment should always be a multiple of 16.
+  if (Total <= 0x1f0) {
+    Initial = Total;
+    Residual = 0;
+  } else {
+    Initial = 0x1f0;
+    Residual = Total - Initial;
+  }
+}
+
+void AArch64FrameLowering::emitPrologue(MachineFunction &MF) const {
+  AArch64MachineFunctionInfo *FuncInfo =
+    MF.getInfo<AArch64MachineFunctionInfo>();
+  MachineBasicBlock &MBB = MF.front();
+  MachineBasicBlock::iterator MBBI = MBB.begin();
+  MachineFrameInfo *MFI = MF.getFrameInfo();
+  const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
+  DebugLoc DL = MBBI != MBB.end() ? MBBI->getDebugLoc() : DebugLoc();
+
+  MachineModuleInfo &MMI = MF.getMMI();
+  std::vector<MachineMove> &Moves = MMI.getFrameMoves();
+  bool NeedsFrameMoves = MMI.hasDebugInfo()
+    || MF.getFunction()->needsUnwindTableEntry();
+
+  uint64_t NumInitialBytes, NumResidualBytes;
+
+  // Currently we expect the stack to be laid out by
+  //     sub sp, sp, #initial
+  //     stp x29, x30, [sp, #offset]
+  //     ...
+  //     str xxx, [sp, #offset]
+  //     sub sp, sp, #rest (possibly via extra instructions).
+  if (MFI->getCalleeSavedInfo().size()) {
+    // If there are callee-saved registers, we want to store them efficiently as
+    // a block, and virtual base assignment happens too early to do it for us so
+    // we adjust the stack in two phases: first just for callee-saved fiddling,
+    // then to allocate the rest of the frame.
+    splitSPAdjustments(MFI->getStackSize(), NumInitialBytes, NumResidualBytes);
+  } else {
+    // If there aren't any callee-saved registers, two-phase adjustment is
+    // inefficient. It's more efficient to adjust with NumInitialBytes too
+    // because when we're in a "callee pops argument space" situation, that pop
+    // must be tacked onto Initial for correctness.
+    NumInitialBytes = MFI->getStackSize();
+    NumResidualBytes = 0;
+  }
+
+  // Tell everyone else how much adjustment we're expecting them to use. In
+  // particular if an adjustment is required for a tail call the epilogue could
+  // have a different view of things.
+  FuncInfo->setInitialStackAdjust(NumInitialBytes);
+
+  emitSPUpdate(MBB, MBBI, DL, TII, AArch64::X16, -NumInitialBytes,
+               MachineInstr::FrameSetup);
+
+  if (NeedsFrameMoves && NumInitialBytes) {
+    // We emit this update even if the CFA is set from a frame pointer later so
+    // that the CFA is valid in the interim.
+    MCSymbol *SPLabel = MMI.getContext().CreateTempSymbol();
+    BuildMI(MBB, MBBI, DL, TII.get(TargetOpcode::PROLOG_LABEL))
+      .addSym(SPLabel);
+
+    MachineLocation Dst(MachineLocation::VirtualFP);
+    MachineLocation Src(AArch64::XSP, NumInitialBytes);
+    Moves.push_back(MachineMove(SPLabel, Dst, Src));
+  }
+
+  // Otherwise we need to set the frame pointer and/or add a second stack
+  // adjustment.
+
+  bool FPNeedsSetting = hasFP(MF);
+  for (; MBBI != MBB.end(); ++MBBI) {
+    // Note that this search makes strong assumptions about the operation used
+    // to store the frame-pointer: it must be "STP x29, x30, ...". This could
+    // change in future, but until then there's no point in implementing
+    // untestable more generic cases.
+    if (FPNeedsSetting && MBBI->getOpcode() == AArch64::LSPair64_STR
+                       && MBBI->getOperand(0).getReg() == AArch64::X29) {
+      int64_t X29FrameIdx = MBBI->getOperand(2).getIndex();
+      FuncInfo->setFramePointerOffset(MFI->getObjectOffset(X29FrameIdx));
+
+      ++MBBI;
+      emitRegUpdate(MBB, MBBI, DL, TII, AArch64::X29, AArch64::XSP,
+                    AArch64::X29,
+                    NumInitialBytes + MFI->getObjectOffset(X29FrameIdx),
+                    MachineInstr::FrameSetup);
+
+      // The offset adjustment used when emitting debugging locations relative
+      // to whatever frame base is set. AArch64 uses the default frame base (FP
+      // or SP) and this adjusts the calculations to be correct.
+      MFI->setOffsetAdjustment(- MFI->getObjectOffset(X29FrameIdx)
+                               - MFI->getStackSize());
+
+      if (NeedsFrameMoves) {
+        MCSymbol *FPLabel = MMI.getContext().CreateTempSymbol();
+        BuildMI(MBB, MBBI, DL, TII.get(TargetOpcode::PROLOG_LABEL))
+          .addSym(FPLabel);
+        MachineLocation Dst(MachineLocation::VirtualFP);
+        MachineLocation Src(AArch64::X29, -MFI->getObjectOffset(X29FrameIdx));
+        Moves.push_back(MachineMove(FPLabel, Dst, Src));
+      }
+
+      FPNeedsSetting = false;
+    }
+
+    if (!MBBI->getFlag(MachineInstr::FrameSetup))
+      break;
+  }
+
+  assert(!FPNeedsSetting && "Frame pointer couldn't be set");
+
+  emitSPUpdate(MBB, MBBI, DL, TII, AArch64::X16, -NumResidualBytes,
+               MachineInstr::FrameSetup);
+
+  // Now we emit the rest of the frame setup information, if necessary: we've
+  // already noted the FP and initial SP moves so we're left with the prologue's
+  // final SP update and callee-saved register locations.
+  if (!NeedsFrameMoves)
+    return;
+
+  // Reuse the label if appropriate, so create it in this outer scope.
+  MCSymbol *CSLabel = 0;
+
+  // The rest of the stack adjustment
+  if (!hasFP(MF) && NumResidualBytes) {
+    CSLabel = MMI.getContext().CreateTempSymbol();
+    BuildMI(MBB, MBBI, DL, TII.get(TargetOpcode::PROLOG_LABEL))
+      .addSym(CSLabel);
+
+    MachineLocation Dst(MachineLocation::VirtualFP);
+    MachineLocation Src(AArch64::XSP, NumResidualBytes + NumInitialBytes);
+    Moves.push_back(MachineMove(CSLabel, Dst, Src));
+  }
+
+  // And any callee-saved registers (it's fine to leave them to the end here,
+  // because the old values are still valid at this point.
+  const std::vector<CalleeSavedInfo> &CSI = MFI->getCalleeSavedInfo();
+  if (CSI.size()) {
+    if (!CSLabel) {
+      CSLabel = MMI.getContext().CreateTempSymbol();
+      BuildMI(MBB, MBBI, DL, TII.get(TargetOpcode::PROLOG_LABEL))
+        .addSym(CSLabel);
+    }
+
+    for (std::vector<CalleeSavedInfo>::const_iterator I = CSI.begin(),
+           E = CSI.end(); I != E; ++I) {
+      MachineLocation Dst(MachineLocation::VirtualFP,
+                          MFI->getObjectOffset(I->getFrameIdx()));
+      MachineLocation Src(I->getReg());
+      Moves.push_back(MachineMove(CSLabel, Dst, Src));
+    }
+  }
+}
+
+void
+AArch64FrameLowering::emitEpilogue(MachineFunction &MF,
+                                   MachineBasicBlock &MBB) const {
+  AArch64MachineFunctionInfo *FuncInfo =
+    MF.getInfo<AArch64MachineFunctionInfo>();
+
+  MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
+  DebugLoc DL = MBBI->getDebugLoc();
+  const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
+  MachineFrameInfo &MFI = *MF.getFrameInfo();
+  unsigned RetOpcode = MBBI->getOpcode();
+
+  // Initial and residual are named for consitency with the prologue. Note that
+  // in the epilogue, the residual adjustment is executed first.
+  uint64_t NumInitialBytes = FuncInfo->getInitialStackAdjust();
+  uint64_t NumResidualBytes = MFI.getStackSize() - NumInitialBytes;
+  uint64_t ArgumentPopSize = 0;
+  if (RetOpcode == AArch64::TC_RETURNdi ||
+      RetOpcode == AArch64::TC_RETURNxi) {
+    MachineOperand &JumpTarget = MBBI->getOperand(0);
+    MachineOperand &StackAdjust = MBBI->getOperand(1);
+
+    MachineInstrBuilder MIB;
+    if (RetOpcode == AArch64::TC_RETURNdi) {
+      MIB = BuildMI(MBB, MBBI, DL, TII.get(AArch64::TAIL_Bimm));
+      if (JumpTarget.isGlobal()) {
+        MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(),
+                             JumpTarget.getTargetFlags());
+      } else {
+        assert(JumpTarget.isSymbol() && "unexpected tail call destination");
+        MIB.addExternalSymbol(JumpTarget.getSymbolName(),
+                              JumpTarget.getTargetFlags());
+      }
+    } else {
+      assert(RetOpcode == AArch64::TC_RETURNxi && JumpTarget.isReg()
+             && "Unexpected tail call");
+
+      MIB = BuildMI(MBB, MBBI, DL, TII.get(AArch64::TAIL_BRx));
+      MIB.addReg(JumpTarget.getReg(), RegState::Kill);
+    }
+
+    // Add the extra operands onto the new tail call instruction even though
+    // they're not used directly (so that liveness is tracked properly etc).
+    for (unsigned i = 2, e = MBBI->getNumOperands(); i != e; ++i)
+        MIB->addOperand(MBBI->getOperand(i));
+
+
+    // Delete the pseudo instruction TC_RETURN.
+    MachineInstr *NewMI = prior(MBBI);
+    MBB.erase(MBBI);
+    MBBI = NewMI;
+
+    // For a tail-call in a callee-pops-arguments environment, some or all of
+    // the stack may actually be in use for the call's arguments, this is
+    // calculated during LowerCall and consumed here...
+    ArgumentPopSize = StackAdjust.getImm();
+  } else {
+    // ... otherwise the amount to pop is *all* of the argument space,
+    // conveniently stored in the MachineFunctionInfo by
+    // LowerFormalArguments. This will, of course, be zero for the C calling
+    // convention.
+    ArgumentPopSize = FuncInfo->getArgumentStackToRestore();
+  }
+
+  assert(NumInitialBytes % 16 == 0 && NumResidualBytes % 16 == 0
+         && "refusing to adjust stack by misaligned amt");
+
+  // We may need to address callee-saved registers differently, so find out the
+  // bound on the frame indices.
+  const std::vector<CalleeSavedInfo> &CSI = MFI.getCalleeSavedInfo();
+  int MinCSFI = 0;
+  int MaxCSFI = -1;
+
+  if (CSI.size()) {
+    MinCSFI = CSI[0].getFrameIdx();
+    MaxCSFI = CSI[CSI.size() - 1].getFrameIdx();
+  }
+
+  // The "residual" stack update comes first from this direction and guarantees
+  // that SP is NumInitialBytes below its value on function entry, either by a
+  // direct update or restoring it from the frame pointer.
+  if (NumInitialBytes + ArgumentPopSize != 0) {
+    emitSPUpdate(MBB, MBBI, DL, TII, AArch64::X16,
+                 NumInitialBytes + ArgumentPopSize);
+    --MBBI;
+  }
+
+
+  // MBBI now points to the instruction just past the last callee-saved
+  // restoration (either RET/B if NumInitialBytes == 0, or the "ADD sp, sp"
+  // otherwise).
+
+  // Now we need to find out where to put the bulk of the stack adjustment
+  MachineBasicBlock::iterator FirstEpilogue = MBBI;
+  while (MBBI != MBB.begin()) {
+    --MBBI;
+
+    unsigned FrameOp;
+    for (FrameOp = 0; FrameOp < MBBI->getNumOperands(); ++FrameOp) {
+      if (MBBI->getOperand(FrameOp).isFI())
+        break;
+    }
+
+    // If this instruction doesn't have a frame index we've reached the end of
+    // the callee-save restoration.
+    if (FrameOp == MBBI->getNumOperands())
+      break;
+
+    // Likewise if it *is* a local reference, but not to a callee-saved object.
+    int FrameIdx = MBBI->getOperand(FrameOp).getIndex();
+    if (FrameIdx < MinCSFI || FrameIdx > MaxCSFI)
+      break;
+
+    FirstEpilogue = MBBI;
+  }
+
+  if (MF.getFrameInfo()->hasVarSizedObjects()) {
+    int64_t StaticFrameBase;
+    StaticFrameBase = -(NumInitialBytes + FuncInfo->getFramePointerOffset());
+    emitRegUpdate(MBB, FirstEpilogue, DL, TII,
+                  AArch64::XSP, AArch64::X29, AArch64::NoRegister,
+                  StaticFrameBase);
+  } else {
+    emitSPUpdate(MBB, FirstEpilogue, DL,TII, AArch64::X16, NumResidualBytes);
+  }
+}
+
+int64_t
+AArch64FrameLowering::resolveFrameIndexReference(MachineFunction &MF,
+                                                 int FrameIndex,
+                                                 unsigned &FrameReg,
+                                                 int SPAdj,
+                                                 bool IsCalleeSaveOp) const {
+  AArch64MachineFunctionInfo *FuncInfo =
+    MF.getInfo<AArch64MachineFunctionInfo>();
+  MachineFrameInfo *MFI = MF.getFrameInfo();
+
+  int64_t TopOfFrameOffset = MFI->getObjectOffset(FrameIndex);
+
+  assert(!(IsCalleeSaveOp && FuncInfo->getInitialStackAdjust() == 0)
+         && "callee-saved register in unexpected place");
+
+  // If the frame for this function is particularly large, we adjust the stack
+  // in two phases which means the callee-save related operations see a
+  // different (intermediate) stack size.
+  int64_t FrameRegPos;
+  if (IsCalleeSaveOp) {
+    FrameReg = AArch64::XSP;
+    FrameRegPos = -static_cast<int64_t>(FuncInfo->getInitialStackAdjust());
+  } else if (useFPForAddressing(MF)) {
+    // Have to use the frame pointer since we have no idea where SP is.
+    FrameReg = AArch64::X29;
+    FrameRegPos = FuncInfo->getFramePointerOffset();
+  } else {
+    FrameReg = AArch64::XSP;
+    FrameRegPos = -static_cast<int64_t>(MFI->getStackSize()) + SPAdj;
+  }
+
+  return TopOfFrameOffset - FrameRegPos;
+}
+
+void
+AArch64FrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
+                                                       RegScavenger *RS) const {
+  const AArch64RegisterInfo *RegInfo =
+    static_cast<const AArch64RegisterInfo *>(MF.getTarget().getRegisterInfo());
+  MachineFrameInfo *MFI = MF.getFrameInfo();
+  const AArch64InstrInfo &TII =
+    *static_cast<const AArch64InstrInfo *>(MF.getTarget().getInstrInfo());
+
+  if (hasFP(MF)) {
+    MF.getRegInfo().setPhysRegUsed(AArch64::X29);
+    MF.getRegInfo().setPhysRegUsed(AArch64::X30);
+  }
+
+  // If addressing of local variables is going to be more complicated than
+  // shoving a base register and an offset into the instruction then we may well
+  // need to scavenge registers. We should either specifically add an
+  // callee-save register for this purpose or allocate an extra spill slot.
+
+  bool BigStack =
+    (RS && MFI->estimateStackSize(MF) >= TII.estimateRSStackLimit(MF))
+    || MFI->hasVarSizedObjects() // Access will be from X29: messes things up
+    || (MFI->adjustsStack() && !hasReservedCallFrame(MF));
+
+  if (!BigStack)
+    return;
+
+  // We certainly need some slack space for the scavenger, preferably an extra
+  // register.
+  const uint16_t *CSRegs = RegInfo->getCalleeSavedRegs();
+  uint16_t ExtraReg = AArch64::NoRegister;
+
+  for (unsigned i = 0; CSRegs[i]; ++i) {
+    if (AArch64::GPR64RegClass.contains(CSRegs[i]) &&
+        !MF.getRegInfo().isPhysRegUsed(CSRegs[i])) {
+      ExtraReg = CSRegs[i];
+      break;
+    }
+  }
+
+  if (ExtraReg != 0) {
+    MF.getRegInfo().setPhysRegUsed(ExtraReg);
+  } else {
+    // Create a stack slot for scavenging purposes. PrologEpilogInserter
+    // helpfully places it near either SP or FP for us to avoid
+    // infinitely-regression during scavenging.
+    const TargetRegisterClass *RC = &AArch64::GPR64RegClass;
+    RS->addScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
+                                                       RC->getAlignment(),
+                                                       false));
+  }
+}
+
+bool AArch64FrameLowering::determinePrologueDeath(MachineBasicBlock &MBB,
+                                                  unsigned Reg) const {
+  // If @llvm.returnaddress is called then it will refer to X30 by some means;
+  // the prologue store does not kill the register.
+  if (Reg == AArch64::X30) {
+    if (MBB.getParent()->getFrameInfo()->isReturnAddressTaken()
+        && MBB.getParent()->getRegInfo().isLiveIn(Reg))
+    return false;
+  }
+
+  // In all other cases, physical registers are dead after they've been saved
+  // but live at the beginning of the prologue block.
+  MBB.addLiveIn(Reg);
+  return true;
+}
+
+void
+AArch64FrameLowering::emitFrameMemOps(bool isPrologue, MachineBasicBlock &MBB,
+                                      MachineBasicBlock::iterator MBBI,
+                                      const std::vector<CalleeSavedInfo> &CSI,
+                                      const TargetRegisterInfo *TRI,
+                                      LoadStoreMethod PossClasses[],
+                                      unsigned NumClasses) const {
+  DebugLoc DL = MBB.findDebugLoc(MBBI);
+  MachineFunction &MF = *MBB.getParent();
+  MachineFrameInfo &MFI = *MF.getFrameInfo();
+  const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
+
+  // A certain amount of implicit contract is present here. The actual stack
+  // offsets haven't been allocated officially yet, so for strictly correct code
+  // we rely on the fact that the elements of CSI are allocated in order
+  // starting at SP, purely as dictated by size and alignment. In practice since
+  // this function handles the only accesses to those slots it's not quite so
+  // important.
+  //
+  // We have also ordered the Callee-saved register list in AArch64CallingConv
+  // so that the above scheme puts registers in order: in particular we want
+  // &X30 to be &X29+8 for an ABI-correct frame record (PCS 5.2.2)
+  for (unsigned i = 0, e = CSI.size(); i < e; ++i) {
+    unsigned Reg = CSI[i].getReg();
+
+    // First we need to find out which register class the register belongs to so
+    // that we can use the correct load/store instrucitons.
+    unsigned ClassIdx;
+    for (ClassIdx = 0; ClassIdx < NumClasses; ++ClassIdx) {
+      if (PossClasses[ClassIdx].RegClass->contains(Reg))
+        break;
+    }
+    assert(ClassIdx != NumClasses
+           && "Asked to store register in unexpected class");
+    const TargetRegisterClass &TheClass = *PossClasses[ClassIdx].RegClass;
+
+    // Now we need to decide whether it's possible to emit a paired instruction:
+    // for this we want the next register to be in the same class.
+    MachineInstrBuilder NewMI;
+    bool Pair = false;
+    if (i + 1 < CSI.size() && TheClass.contains(CSI[i+1].getReg())) {
+      Pair = true;
+      unsigned StLow = 0, StHigh = 0;
+      if (isPrologue) {
+        // Most of these registers will be live-in to the MBB and killed by our
+        // store, though there are exceptions (see determinePrologueDeath).
+        StLow = getKillRegState(determinePrologueDeath(MBB, CSI[i+1].getReg()));
+        StHigh = getKillRegState(determinePrologueDeath(MBB, CSI[i].getReg()));
+      } else {
+        StLow = RegState::Define;
+        StHigh = RegState::Define;
+      }
+
+      NewMI = BuildMI(MBB, MBBI, DL, TII.get(PossClasses[ClassIdx].PairOpcode))
+                .addReg(CSI[i+1].getReg(), StLow)
+                .addReg(CSI[i].getReg(), StHigh);
+
+      // If it's a paired op, we've consumed two registers
+      ++i;
+    } else {
+      unsigned State;
+      if (isPrologue) {
+        State = getKillRegState(determinePrologueDeath(MBB, CSI[i].getReg()));
+      } else {
+        State = RegState::Define;
+      }
+
+      NewMI = BuildMI(MBB, MBBI, DL,
+                      TII.get(PossClasses[ClassIdx].SingleOpcode))
+                .addReg(CSI[i].getReg(), State);
+    }
+
+    // Note that the FrameIdx refers to the second register in a pair: it will
+    // be allocated the smaller numeric address and so is the one an LDP/STP
+    // address must use.
+    int FrameIdx = CSI[i].getFrameIdx();
+    MachineMemOperand::MemOperandFlags Flags;
+    Flags = isPrologue ? MachineMemOperand::MOStore : MachineMemOperand::MOLoad;
+    MachineMemOperand *MMO =
+      MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FrameIdx),
+                             Flags,
+                             Pair ? TheClass.getSize() * 2 : TheClass.getSize(),
+                             MFI.getObjectAlignment(FrameIdx));
+
+    NewMI.addFrameIndex(FrameIdx)
+      .addImm(0)                  // address-register offset
+      .addMemOperand(MMO);
+
+    if (isPrologue)
+      NewMI.setMIFlags(MachineInstr::FrameSetup);
+
+    // For aesthetic reasons, during an epilogue we want to emit complementary
+    // operations to the prologue, but in the opposite order. So we still
+    // iterate through the CalleeSavedInfo list in order, but we put the
+    // instructions successively earlier in the MBB.
+    if (!isPrologue)
+      --MBBI;
+  }
+}
+
+bool
+AArch64FrameLowering::spillCalleeSavedRegisters(MachineBasicBlock &MBB,
+                                        MachineBasicBlock::iterator MBBI,
+                                        const std::vector<CalleeSavedInfo> &CSI,
+                                        const TargetRegisterInfo *TRI) const {
+  if (CSI.empty())
+    return false;
+
+  static LoadStoreMethod PossibleClasses[] = {
+    {&AArch64::GPR64RegClass, AArch64::LSPair64_STR, AArch64::LS64_STR},
+    {&AArch64::FPR64RegClass, AArch64::LSFPPair64_STR, AArch64::LSFP64_STR},
+  };
+  unsigned NumClasses = llvm::array_lengthof(PossibleClasses);
+
+  emitFrameMemOps(/* isPrologue = */ true, MBB, MBBI, CSI, TRI,
+                  PossibleClasses, NumClasses);
+
+  return true;
+}
+
+bool
+AArch64FrameLowering::restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
+                                        MachineBasicBlock::iterator MBBI,
+                                        const std::vector<CalleeSavedInfo> &CSI,
+                                        const TargetRegisterInfo *TRI) const {
+
+  if (CSI.empty())
+    return false;
+
+  static LoadStoreMethod PossibleClasses[] = {
+    {&AArch64::GPR64RegClass, AArch64::LSPair64_LDR, AArch64::LS64_LDR},
+    {&AArch64::FPR64RegClass, AArch64::LSFPPair64_LDR, AArch64::LSFP64_LDR},
+  };
+  unsigned NumClasses = llvm::array_lengthof(PossibleClasses);
+
+  emitFrameMemOps(/* isPrologue = */ false, MBB, MBBI, CSI, TRI,
+                  PossibleClasses, NumClasses);
+
+  return true;
+}
+
+bool
+AArch64FrameLowering::hasFP(const MachineFunction &MF) const {
+  const MachineFrameInfo *MFI = MF.getFrameInfo();
+  const TargetRegisterInfo *RI = MF.getTarget().getRegisterInfo();
+
+  // This is a decision of ABI compliance. The AArch64 PCS gives various options
+  // for conformance, and even at the most stringent level more or less permits
+  // elimination for leaf functions because there's no loss of functionality
+  // (for debugging etc)..
+  if (MF.getTarget().Options.DisableFramePointerElim(MF) && MFI->hasCalls())
+    return true;
+
+  // The following are hard-limits: incorrect code will be generated if we try
+  // to omit the frame.
+  return (RI->needsStackRealignment(MF) ||
+          MFI->hasVarSizedObjects() ||
+          MFI->isFrameAddressTaken());
+}
+
+bool
+AArch64FrameLowering::useFPForAddressing(const MachineFunction &MF) const {
+  return MF.getFrameInfo()->hasVarSizedObjects();
+}
+
+bool
+AArch64FrameLowering::hasReservedCallFrame(const MachineFunction &MF) const {
+  const MachineFrameInfo *MFI = MF.getFrameInfo();
+
+  // Of the various reasons for having a frame pointer, it's actually only
+  // variable-sized objects that prevent reservation of a call frame.
+  return !(hasFP(MF) && MFI->hasVarSizedObjects());
+}
+
+void
+AArch64FrameLowering::eliminateCallFramePseudoInstr(
+                                MachineFunction &MF,
+                                MachineBasicBlock &MBB,
+                                MachineBasicBlock::iterator MI) const {
+  const AArch64InstrInfo &TII =
+    *static_cast<const AArch64InstrInfo *>(MF.getTarget().getInstrInfo());
+  DebugLoc dl = MI->getDebugLoc();
+  int Opcode = MI->getOpcode();
+  bool IsDestroy = Opcode == TII.getCallFrameDestroyOpcode();
+  uint64_t CalleePopAmount = IsDestroy ? MI->getOperand(1).getImm() : 0;
+
+  if (!hasReservedCallFrame(MF)) {
+    unsigned Align = getStackAlignment();
+
+    int64_t Amount = MI->getOperand(0).getImm();
+    Amount = RoundUpToAlignment(Amount, Align);
+    if (!IsDestroy) Amount = -Amount;
+
+    // N.b. if CalleePopAmount is valid but zero (i.e. callee would pop, but it
+    // doesn't have to pop anything), then the first operand will be zero too so
+    // this adjustment is a no-op.
+    if (CalleePopAmount == 0) {
+      // FIXME: in-function stack adjustment for calls is limited to 12-bits
+      // because there's no guaranteed temporary register available. Mostly call
+      // frames will be allocated at the start of a function so this is OK, but
+      // it is a limitation that needs dealing with.
+      assert(Amount > -0xfff && Amount < 0xfff && "call frame too large");
+      emitSPUpdate(MBB, MI, dl, TII, AArch64::NoRegister, Amount);
+    }
+  } else if (CalleePopAmount != 0) {
+    // If the calling convention demands that the callee pops arguments from the
+    // stack, we want to add it back if we have a reserved call frame.
+    assert(CalleePopAmount < 0xfff && "call frame too large");
+    emitSPUpdate(MBB, MI, dl, TII, AArch64::NoRegister, -CalleePopAmount);
+  }
+
+  MBB.erase(MI);
+}
diff --git a/lib/Target/AArch64/AArch64FrameLowering.h b/lib/Target/AArch64/AArch64FrameLowering.h
new file mode 100644
index 000000000000..45ea0ec8e071
--- /dev/null
+++ b/lib/Target/AArch64/AArch64FrameLowering.h
@@ -0,0 +1,108 @@
+//==- AArch64FrameLowering.h - Define frame lowering for AArch64 -*- C++ -*--=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class implements the AArch64-specific parts of the TargetFrameLowering
+// class.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_AARCH64_FRAMEINFO_H
+#define LLVM_AARCH64_FRAMEINFO_H
+
+#include "AArch64Subtarget.h"
+#include "llvm/Target/TargetFrameLowering.h"
+
+namespace llvm {
+class AArch64Subtarget;
+
+class AArch64FrameLowering : public TargetFrameLowering {
+private:
+  // In order to unify the spilling and restoring of callee-saved registers into
+  // emitFrameMemOps, we need to be able to specify which instructions to use
+  // for the relevant memory operations on each register class. An array of the
+  // following struct is populated and passed in to achieve this.
+  struct LoadStoreMethod {
+    const TargetRegisterClass *RegClass; // E.g. GPR64RegClass
+
+    // The preferred instruction.
+    unsigned PairOpcode; // E.g. LSPair64_STR
+
+    // Sometimes only a single register can be handled at once.
+    unsigned SingleOpcode; // E.g. LS64_STR
+  };
+protected:
+  const AArch64Subtarget &STI;
+
+public:
+  explicit AArch64FrameLowering(const AArch64Subtarget &sti)
+    : TargetFrameLowering(TargetFrameLowering::StackGrowsDown, 16, 0, 16),
+      STI(sti) {
+  }
+
+  /// emitProlog/emitEpilog - These methods insert prolog and epilog code into
+  /// the function.
+  virtual void emitPrologue(MachineFunction &MF) const;
+  virtual void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
+
+  /// Decides how much stack adjustment to perform in each phase of the prologue
+  /// and epilogue.
+  void splitSPAdjustments(uint64_t Total, uint64_t &Initial,
+                          uint64_t &Residual) const;
+
+  int64_t resolveFrameIndexReference(MachineFunction &MF, int FrameIndex,
+                                     unsigned &FrameReg, int SPAdj,
+                                     bool IsCalleeSaveOp) const;
+
+  virtual void processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
+                                                    RegScavenger *RS) const;
+
+  virtual bool spillCalleeSavedRegisters(MachineBasicBlock &MBB,
+                                        MachineBasicBlock::iterator MI,
+                                        const std::vector<CalleeSavedInfo> &CSI,
+                                        const TargetRegisterInfo *TRI) const;
+  virtual bool restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
+                                        MachineBasicBlock::iterator MI,
+                                        const std::vector<CalleeSavedInfo> &CSI,
+                                        const TargetRegisterInfo *TRI) const;
+
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator MI) const;
+
+  /// If the register is X30 (i.e. LR) and the return address is used in the
+  /// function then the callee-save store doesn't actually kill the register,
+  /// otherwise it does.
+  bool determinePrologueDeath(MachineBasicBlock &MBB, unsigned Reg) const;
+
+  /// This function emits the loads or stores required during prologue and
+  /// epilogue as efficiently as possible.
+  ///
+  /// The operations involved in setting up and tearing down the frame are
+  /// similar enough to warrant a shared function, particularly as discrepancies
+  /// between the two would be disastrous.
+  void emitFrameMemOps(bool isStore, MachineBasicBlock &MBB,
+                       MachineBasicBlock::iterator MI,
+                       const std::vector<CalleeSavedInfo> &CSI,
+                       const TargetRegisterInfo *TRI,
+                       LoadStoreMethod PossibleClasses[],
+                       unsigned NumClasses) const;
+
+
+  virtual bool hasFP(const MachineFunction &MF) const;
+
+  virtual bool useFPForAddressing(const MachineFunction &MF) const;
+
+  /// On AA
+  virtual bool hasReservedCallFrame(const MachineFunction &MF) const;
+
+};
+
+} // End llvm namespace
+
+#endif
diff --git a/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp b/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
new file mode 100644
index 000000000000..46b822152a00
--- /dev/null
+++ b/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
@@ -0,0 +1,415 @@
+//===-- AArch64ISelDAGToDAG.cpp - A dag to dag inst selector for AArch64 --===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines an instruction selector for the AArch64 target.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "aarch64-isel"
+#include "AArch64.h"
+#include "AArch64InstrInfo.h"
+#include "AArch64Subtarget.h"
+#include "AArch64TargetMachine.h"
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/ADT/APSInt.h"
+#include "llvm/CodeGen/SelectionDAGISel.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+//===--------------------------------------------------------------------===//
+/// AArch64 specific code to select AArch64 machine instructions for
+/// SelectionDAG operations.
+///
+namespace {
+
+class AArch64DAGToDAGISel : public SelectionDAGISel {
+  AArch64TargetMachine &TM;
+  const AArch64InstrInfo *TII;
+
+  /// Keep a pointer to the AArch64Subtarget around so that we can
+  /// make the right decision when generating code for different targets.
+  const AArch64Subtarget *Subtarget;
+
+public:
+  explicit AArch64DAGToDAGISel(AArch64TargetMachine &tm,
+                               CodeGenOpt::Level OptLevel)
+    : SelectionDAGISel(tm, OptLevel), TM(tm),
+      TII(static_cast<const AArch64InstrInfo*>(TM.getInstrInfo())),
+      Subtarget(&TM.getSubtarget<AArch64Subtarget>()) {
+  }
+
+  virtual const char *getPassName() const {
+    return "AArch64 Instruction Selection";
+  }
+
+  // Include the pieces autogenerated from the target description.
+#include "AArch64GenDAGISel.inc"
+
+  template<unsigned MemSize>
+  bool SelectOffsetUImm12(SDValue N, SDValue &UImm12) {
+    const ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N);
+    if (!CN || CN->getZExtValue() % MemSize != 0
+        || CN->getZExtValue() / MemSize > 0xfff)
+      return false;
+
+    UImm12 =  CurDAG->getTargetConstant(CN->getZExtValue() / MemSize, MVT::i64);
+    return true;
+  }
+
+  template<unsigned RegWidth>
+  bool SelectCVTFixedPosOperand(SDValue N, SDValue &FixedPos) {
+    return SelectCVTFixedPosOperand(N, FixedPos, RegWidth);
+  }
+
+  bool SelectFPZeroOperand(SDValue N, SDValue &Dummy);
+
+  bool SelectCVTFixedPosOperand(SDValue N, SDValue &FixedPos,
+                                unsigned RegWidth);
+
+  bool SelectInlineAsmMemoryOperand(const SDValue &Op,
+                                    char ConstraintCode,
+                                    std::vector<SDValue> &OutOps);
+
+  bool SelectLogicalImm(SDValue N, SDValue &Imm);
+
+  template<unsigned RegWidth>
+  bool SelectTSTBOperand(SDValue N, SDValue &FixedPos) {
+    return SelectTSTBOperand(N, FixedPos, RegWidth);
+  }
+
+  bool SelectTSTBOperand(SDValue N, SDValue &FixedPos, unsigned RegWidth);
+
+  SDNode *TrySelectToMoveImm(SDNode *N);
+  SDNode *LowerToFPLitPool(SDNode *Node);
+  SDNode *SelectToLitPool(SDNode *N);
+
+  SDNode* Select(SDNode*);
+private:
+};
+}
+
+bool
+AArch64DAGToDAGISel::SelectCVTFixedPosOperand(SDValue N, SDValue &FixedPos,
+                                              unsigned RegWidth) {
+  const ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(N);
+  if (!CN) return false;
+
+  // An FCVT[SU] instruction performs: convertToInt(Val * 2^fbits) where fbits
+  // is between 1 and 32 for a destination w-register, or 1 and 64 for an
+  // x-register.
+  //
+  // By this stage, we've detected (fp_to_[su]int (fmul Val, THIS_NODE)) so we
+  // want THIS_NODE to be 2^fbits. This is much easier to deal with using
+  // integers.
+  bool IsExact;
+
+  // fbits is between 1 and 64 in the worst-case, which means the fmul
+  // could have 2^64 as an actual operand. Need 65 bits of precision.
+  APSInt IntVal(65, true);
+  CN->getValueAPF().convertToInteger(IntVal, APFloat::rmTowardZero, &IsExact);
+
+  // N.b. isPowerOf2 also checks for > 0.
+  if (!IsExact || !IntVal.isPowerOf2()) return false;
+  unsigned FBits = IntVal.logBase2();
+
+  // Checks above should have guaranteed that we haven't lost information in
+  // finding FBits, but it must still be in range.
+  if (FBits == 0 || FBits > RegWidth) return false;
+
+  FixedPos = CurDAG->getTargetConstant(64 - FBits, MVT::i32);
+  return true;
+}
+
+bool
+AArch64DAGToDAGISel::SelectInlineAsmMemoryOperand(const SDValue &Op,
+                                                 char ConstraintCode,
+                                                 std::vector<SDValue> &OutOps) {
+  switch (ConstraintCode) {
+  default: llvm_unreachable("Unrecognised AArch64 memory constraint");
+  case 'm':
+    // FIXME: more freedom is actually permitted for 'm'. We can go
+    // hunting for a base and an offset if we want. Of course, since
+    // we don't really know how the operand is going to be used we're
+    // probably restricted to the load/store pair's simm7 as an offset
+    // range anyway.
+  case 'Q':
+    OutOps.push_back(Op);
+  }
+
+  return false;
+}
+
+bool
+AArch64DAGToDAGISel::SelectFPZeroOperand(SDValue N, SDValue &Dummy) {
+  ConstantFPSDNode *Imm = dyn_cast<ConstantFPSDNode>(N);
+  if (!Imm || !Imm->getValueAPF().isPosZero())
+    return false;
+
+  // Doesn't actually carry any information, but keeps TableGen quiet.
+  Dummy = CurDAG->getTargetConstant(0, MVT::i32);
+  return true;
+}
+
+bool AArch64DAGToDAGISel::SelectLogicalImm(SDValue N, SDValue &Imm) {
+  uint32_t Bits;
+  uint32_t RegWidth = N.getValueType().getSizeInBits();
+
+  ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N);
+  if (!CN) return false;
+
+  if (!A64Imms::isLogicalImm(RegWidth, CN->getZExtValue(), Bits))
+    return false;
+
+  Imm = CurDAG->getTargetConstant(Bits, MVT::i32);
+  return true;
+}
+
+SDNode *AArch64DAGToDAGISel::TrySelectToMoveImm(SDNode *Node) {
+  SDNode *ResNode;
+  DebugLoc dl = Node->getDebugLoc();
+  EVT DestType = Node->getValueType(0);
+  unsigned DestWidth = DestType.getSizeInBits();
+
+  unsigned MOVOpcode;
+  EVT MOVType;
+  int UImm16, Shift;
+  uint32_t LogicalBits;
+
+  uint64_t BitPat = cast<ConstantSDNode>(Node)->getZExtValue();
+  if (A64Imms::isMOVZImm(DestWidth, BitPat, UImm16, Shift)) {
+    MOVType = DestType;
+    MOVOpcode = DestWidth == 64 ? AArch64::MOVZxii : AArch64::MOVZwii;
+  } else if (A64Imms::isMOVNImm(DestWidth, BitPat, UImm16, Shift)) {
+    MOVType = DestType;
+    MOVOpcode = DestWidth == 64 ? AArch64::MOVNxii : AArch64::MOVNwii;
+  } else if (DestWidth == 64 && A64Imms::isMOVNImm(32, BitPat, UImm16, Shift)) {
+    // To get something like 0x0000_0000_ffff_1234 into a 64-bit register we can
+    // use a 32-bit instruction: "movn w0, 0xedbc".
+    MOVType = MVT::i32;
+    MOVOpcode = AArch64::MOVNwii;
+  } else if (A64Imms::isLogicalImm(DestWidth, BitPat, LogicalBits))  {
+    MOVOpcode = DestWidth == 64 ? AArch64::ORRxxi : AArch64::ORRwwi;
+    uint16_t ZR = DestWidth == 64 ? AArch64::XZR : AArch64::WZR;
+
+    return CurDAG->getMachineNode(MOVOpcode, dl, DestType,
+                              CurDAG->getRegister(ZR, DestType),
+                              CurDAG->getTargetConstant(LogicalBits, MVT::i32));
+  } else {
+    // Can't handle it in one instruction. There's scope for permitting two (or
+    // more) instructions, but that'll need more thought.
+    return NULL;
+  }
+
+  ResNode = CurDAG->getMachineNode(MOVOpcode, dl, MOVType,
+                                   CurDAG->getTargetConstant(UImm16, MVT::i32),
+                                   CurDAG->getTargetConstant(Shift, MVT::i32));
+
+  if (MOVType != DestType) {
+    ResNode = CurDAG->getMachineNode(TargetOpcode::SUBREG_TO_REG, dl,
+                          MVT::i64, MVT::i32, MVT::Other,
+                          CurDAG->getTargetConstant(0, MVT::i64),
+                          SDValue(ResNode, 0),
+                          CurDAG->getTargetConstant(AArch64::sub_32, MVT::i32));
+  }
+
+  return ResNode;
+}
+
+SDNode *AArch64DAGToDAGISel::SelectToLitPool(SDNode *Node) {
+  DebugLoc DL = Node->getDebugLoc();
+  uint64_t UnsignedVal = cast<ConstantSDNode>(Node)->getZExtValue();
+  int64_t SignedVal = cast<ConstantSDNode>(Node)->getSExtValue();
+  EVT DestType = Node->getValueType(0);
+  EVT PtrVT = TLI.getPointerTy();
+
+  // Since we may end up loading a 64-bit constant from a 32-bit entry the
+  // constant in the pool may have a different type to the eventual node.
+  ISD::LoadExtType Extension;
+  EVT MemType;
+
+  assert((DestType == MVT::i64 || DestType == MVT::i32)
+         && "Only expect integer constants at the moment");
+
+  if (DestType == MVT::i32) {
+    Extension = ISD::NON_EXTLOAD;
+    MemType = MVT::i32;
+  } else if (UnsignedVal <= UINT32_MAX) {
+    Extension = ISD::ZEXTLOAD;
+    MemType = MVT::i32;
+  } else if (SignedVal >= INT32_MIN && SignedVal <= INT32_MAX) {
+    Extension = ISD::SEXTLOAD;
+    MemType = MVT::i32;
+  } else {
+    Extension = ISD::NON_EXTLOAD;
+    MemType = MVT::i64;
+  }
+
+  Constant *CV = ConstantInt::get(Type::getIntNTy(*CurDAG->getContext(),
+                                                  MemType.getSizeInBits()),
+                                  UnsignedVal);
+  SDValue PoolAddr;
+  unsigned Alignment = TLI.getDataLayout()->getABITypeAlignment(CV->getType());
+  PoolAddr = CurDAG->getNode(AArch64ISD::WrapperSmall, DL, PtrVT,
+                             CurDAG->getTargetConstantPool(CV, PtrVT, 0, 0,
+                                                         AArch64II::MO_NO_FLAG),
+                             CurDAG->getTargetConstantPool(CV, PtrVT, 0, 0,
+                                                           AArch64II::MO_LO12),
+                             CurDAG->getConstant(Alignment, MVT::i32));
+
+  return CurDAG->getExtLoad(Extension, DL, DestType, CurDAG->getEntryNode(),
+                            PoolAddr,
+                            MachinePointerInfo::getConstantPool(), MemType,
+                            /* isVolatile = */ false,
+                            /* isNonTemporal = */ false,
+                            Alignment).getNode();
+}
+
+SDNode *AArch64DAGToDAGISel::LowerToFPLitPool(SDNode *Node) {
+  DebugLoc DL = Node->getDebugLoc();
+  const ConstantFP *FV = cast<ConstantFPSDNode>(Node)->getConstantFPValue();
+  EVT PtrVT = TLI.getPointerTy();
+  EVT DestType = Node->getValueType(0);
+
+  unsigned Alignment = TLI.getDataLayout()->getABITypeAlignment(FV->getType());
+  SDValue PoolAddr;
+
+  assert(TM.getCodeModel() == CodeModel::Small &&
+         "Only small code model supported");
+  PoolAddr = CurDAG->getNode(AArch64ISD::WrapperSmall, DL, PtrVT,
+                             CurDAG->getTargetConstantPool(FV, PtrVT, 0, 0,
+                                                         AArch64II::MO_NO_FLAG),
+                             CurDAG->getTargetConstantPool(FV, PtrVT, 0, 0,
+                                                           AArch64II::MO_LO12),
+                             CurDAG->getConstant(Alignment, MVT::i32));
+
+  return CurDAG->getLoad(DestType, DL, CurDAG->getEntryNode(), PoolAddr,
+                         MachinePointerInfo::getConstantPool(),
+                         /* isVolatile = */ false,
+                         /* isNonTemporal = */ false,
+                         /* isInvariant = */ true,
+                         Alignment).getNode();
+}
+
+bool
+AArch64DAGToDAGISel::SelectTSTBOperand(SDValue N, SDValue &FixedPos,
+                                       unsigned RegWidth) {
+  const ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N);
+  if (!CN) return false;
+
+  uint64_t Val = CN->getZExtValue();
+
+  if (!isPowerOf2_64(Val)) return false;
+
+  unsigned TestedBit = Log2_64(Val);
+  // Checks above should have guaranteed that we haven't lost information in
+  // finding TestedBit, but it must still be in range.
+  if (TestedBit >= RegWidth) return false;
+
+  FixedPos = CurDAG->getTargetConstant(TestedBit, MVT::i64);
+  return true;
+}
+
+SDNode *AArch64DAGToDAGISel::Select(SDNode *Node) {
+  // Dump information about the Node being selected
+  DEBUG(dbgs() << "Selecting: "; Node->dump(CurDAG); dbgs() << "\n");
+
+  if (Node->isMachineOpcode()) {
+    DEBUG(dbgs() << "== "; Node->dump(CurDAG); dbgs() << "\n");
+    return NULL;
+  }
+
+  switch (Node->getOpcode()) {
+  case ISD::FrameIndex: {
+    int FI = cast<FrameIndexSDNode>(Node)->getIndex();
+    EVT PtrTy = TLI.getPointerTy();
+    SDValue TFI = CurDAG->getTargetFrameIndex(FI, PtrTy);
+    return CurDAG->SelectNodeTo(Node, AArch64::ADDxxi_lsl0_s, PtrTy,
+                                TFI, CurDAG->getTargetConstant(0, PtrTy));
+  }
+  case ISD::ConstantPool: {
+    // Constant pools are fine, just create a Target entry.
+    ConstantPoolSDNode *CN = cast<ConstantPoolSDNode>(Node);
+    const Constant *C = CN->getConstVal();
+    SDValue CP = CurDAG->getTargetConstantPool(C, CN->getValueType(0));
+
+    ReplaceUses(SDValue(Node, 0), CP);
+    return NULL;
+  }
+  case ISD::Constant: {
+    SDNode *ResNode = 0;
+    if (cast<ConstantSDNode>(Node)->getZExtValue() == 0) {
+      // XZR and WZR are probably even better than an actual move: most of the
+      // time they can be folded into another instruction with *no* cost.
+
+      EVT Ty = Node->getValueType(0);
+      assert((Ty == MVT::i32 || Ty == MVT::i64) && "unexpected type");
+      uint16_t Register = Ty == MVT::i32 ? AArch64::WZR : AArch64::XZR;
+      ResNode = CurDAG->getCopyFromReg(CurDAG->getEntryNode(),
+                                       Node->getDebugLoc(),
+                                       Register, Ty).getNode();
+    }
+
+    // Next best option is a move-immediate, see if we can do that.
+    if (!ResNode) {
+      ResNode = TrySelectToMoveImm(Node);
+    }
+
+    if (ResNode)
+      return ResNode;
+
+    // If even that fails we fall back to a lit-pool entry at the moment. Future
+    // tuning may change this to a sequence of MOVZ/MOVN/MOVK instructions.
+    ResNode = SelectToLitPool(Node);
+    assert(ResNode && "We need *some* way to materialise a constant");
+
+    // We want to continue selection at this point since the litpool access
+    // generated used generic nodes for simplicity.
+    ReplaceUses(SDValue(Node, 0), SDValue(ResNode, 0));
+    Node = ResNode;
+    break;
+  }
+  case ISD::ConstantFP: {
+    if (A64Imms::isFPImm(cast<ConstantFPSDNode>(Node)->getValueAPF())) {
+      // FMOV will take care of it from TableGen
+      break;
+    }
+
+    SDNode *ResNode = LowerToFPLitPool(Node);
+    ReplaceUses(SDValue(Node, 0), SDValue(ResNode, 0));
+
+    // We want to continue selection at this point since the litpool access
+    // generated used generic nodes for simplicity.
+    Node = ResNode;
+    break;
+  }
+  default:
+    break; // Let generic code handle it
+  }
+
+  SDNode *ResNode = SelectCode(Node);
+
+  DEBUG(dbgs() << "=> ";
+        if (ResNode == NULL || ResNode == Node)
+          Node->dump(CurDAG);
+        else
+          ResNode->dump(CurDAG);
+        dbgs() << "\n");
+
+  return ResNode;
+}
+
+/// This pass converts a legalized DAG into a AArch64-specific DAG, ready for
+/// instruction scheduling.
+FunctionPass *llvm::createAArch64ISelDAG(AArch64TargetMachine &TM,
+                                         CodeGenOpt::Level OptLevel) {
+  return new AArch64DAGToDAGISel(TM, OptLevel);
+}
diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
new file mode 100644
index 000000000000..e9f449709c40
--- /dev/null
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -0,0 +1,2975 @@
+//===-- AArch64ISelLowering.cpp - AArch64 DAG Lowering Implementation -----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the interfaces that AArch64 uses to lower LLVM code into a
+// selection DAG.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "aarch64-isel"
+#include "AArch64.h"
+#include "AArch64ISelLowering.h"
+#include "AArch64MachineFunctionInfo.h"
+#include "AArch64TargetMachine.h"
+#include "AArch64TargetObjectFile.h"
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/CodeGen/Analysis.h"
+#include "llvm/CodeGen/CallingConvLower.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/IR/CallingConv.h"
+
+using namespace llvm;
+
+static TargetLoweringObjectFile *createTLOF(AArch64TargetMachine &TM) {
+  const AArch64Subtarget *Subtarget = &TM.getSubtarget<AArch64Subtarget>();
+
+  if (Subtarget->isTargetLinux())
+    return new AArch64LinuxTargetObjectFile();
+  if (Subtarget->isTargetELF())
+    return new TargetLoweringObjectFileELF();
+  llvm_unreachable("unknown subtarget type");
+}
+
+
+AArch64TargetLowering::AArch64TargetLowering(AArch64TargetMachine &TM)
+  : TargetLowering(TM, createTLOF(TM)),
+    Subtarget(&TM.getSubtarget<AArch64Subtarget>()),
+    RegInfo(TM.getRegisterInfo()),
+    Itins(TM.getInstrItineraryData()) {
+
+  // SIMD compares set the entire lane's bits to 1
+  setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
+
+  // Scalar register <-> type mapping
+  addRegisterClass(MVT::i32, &AArch64::GPR32RegClass);
+  addRegisterClass(MVT::i64, &AArch64::GPR64RegClass);
+  addRegisterClass(MVT::f16, &AArch64::FPR16RegClass);
+  addRegisterClass(MVT::f32, &AArch64::FPR32RegClass);
+  addRegisterClass(MVT::f64, &AArch64::FPR64RegClass);
+  addRegisterClass(MVT::f128, &AArch64::FPR128RegClass);
+
+  computeRegisterProperties();
+
+  // Some atomic operations can be folded into load-acquire or store-release
+  // instructions on AArch64. It's marginally simpler to let LLVM expand
+  // everything out to a barrier and then recombine the (few) barriers we can.
+  setInsertFencesForAtomic(true);
+  setTargetDAGCombine(ISD::ATOMIC_FENCE);
+  setTargetDAGCombine(ISD::ATOMIC_STORE);
+
+  // We combine OR nodes for bitfield and NEON BSL operations.
+  setTargetDAGCombine(ISD::OR);
+
+  setTargetDAGCombine(ISD::AND);
+  setTargetDAGCombine(ISD::SRA);
+
+  // AArch64 does not have i1 loads, or much of anything for i1 really.
+  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
+  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
+  setLoadExtAction(ISD::EXTLOAD, MVT::i1, Promote);
+
+  setStackPointerRegisterToSaveRestore(AArch64::XSP);
+  setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64, Expand);
+  setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
+  setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
+
+  // We'll lower globals to wrappers for selection.
+  setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
+  setOperationAction(ISD::GlobalTLSAddress, MVT::i64, Custom);
+
+  // A64 instructions have the comparison predicate attached to the user of the
+  // result, but having a separate comparison is valuable for matching.
+  setOperationAction(ISD::BR_CC, MVT::i32, Custom);
+  setOperationAction(ISD::BR_CC, MVT::i64, Custom);
+  setOperationAction(ISD::BR_CC, MVT::f32, Custom);
+  setOperationAction(ISD::BR_CC, MVT::f64, Custom);
+
+  setOperationAction(ISD::SELECT, MVT::i32, Custom);
+  setOperationAction(ISD::SELECT, MVT::i64, Custom);
+  setOperationAction(ISD::SELECT, MVT::f32, Custom);
+  setOperationAction(ISD::SELECT, MVT::f64, Custom);
+
+  setOperationAction(ISD::SELECT_CC, MVT::i32, Custom);
+  setOperationAction(ISD::SELECT_CC, MVT::i64, Custom);
+  setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
+  setOperationAction(ISD::SELECT_CC, MVT::f64, Custom);
+
+  setOperationAction(ISD::BRCOND, MVT::Other, Custom);
+
+  setOperationAction(ISD::SETCC, MVT::i32, Custom);
+  setOperationAction(ISD::SETCC, MVT::i64, Custom);
+  setOperationAction(ISD::SETCC, MVT::f32, Custom);
+  setOperationAction(ISD::SETCC, MVT::f64, Custom);
+
+  setOperationAction(ISD::BR_JT, MVT::Other, Expand);
+  setOperationAction(ISD::JumpTable, MVT::i32, Custom);
+  setOperationAction(ISD::JumpTable, MVT::i64, Custom);
+
+  setOperationAction(ISD::VASTART, MVT::Other, Custom);
+  setOperationAction(ISD::VACOPY, MVT::Other, Custom);
+  setOperationAction(ISD::VAEND, MVT::Other, Expand);
+  setOperationAction(ISD::VAARG, MVT::Other, Expand);
+
+  setOperationAction(ISD::BlockAddress, MVT::i64, Custom);
+
+  setOperationAction(ISD::ROTL, MVT::i32, Expand);
+  setOperationAction(ISD::ROTL, MVT::i64, Expand);
+
+  setOperationAction(ISD::UREM, MVT::i32, Expand);
+  setOperationAction(ISD::UREM, MVT::i64, Expand);
+  setOperationAction(ISD::UDIVREM, MVT::i32, Expand);
+  setOperationAction(ISD::UDIVREM, MVT::i64, Expand);
+
+  setOperationAction(ISD::SREM, MVT::i32, Expand);
+  setOperationAction(ISD::SREM, MVT::i64, Expand);
+  setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
+  setOperationAction(ISD::SDIVREM, MVT::i64, Expand);
+
+  setOperationAction(ISD::CTPOP, MVT::i32, Expand);
+  setOperationAction(ISD::CTPOP, MVT::i64, Expand);
+
+  // Legal floating-point operations.
+  setOperationAction(ISD::FABS, MVT::f32, Legal);
+  setOperationAction(ISD::FABS, MVT::f64, Legal);
+
+  setOperationAction(ISD::FCEIL, MVT::f32, Legal);
+  setOperationAction(ISD::FCEIL, MVT::f64, Legal);
+
+  setOperationAction(ISD::FFLOOR, MVT::f32, Legal);
+  setOperationAction(ISD::FFLOOR, MVT::f64, Legal);
+
+  setOperationAction(ISD::FNEARBYINT, MVT::f32, Legal);
+  setOperationAction(ISD::FNEARBYINT, MVT::f64, Legal);
+
+  setOperationAction(ISD::FNEG, MVT::f32, Legal);
+  setOperationAction(ISD::FNEG, MVT::f64, Legal);
+
+  setOperationAction(ISD::FRINT, MVT::f32, Legal);
+  setOperationAction(ISD::FRINT, MVT::f64, Legal);
+
+  setOperationAction(ISD::FSQRT, MVT::f32, Legal);
+  setOperationAction(ISD::FSQRT, MVT::f64, Legal);
+
+  setOperationAction(ISD::FTRUNC, MVT::f32, Legal);
+  setOperationAction(ISD::FTRUNC, MVT::f64, Legal);
+
+  setOperationAction(ISD::ConstantFP, MVT::f32, Legal);
+  setOperationAction(ISD::ConstantFP, MVT::f64, Legal);
+  setOperationAction(ISD::ConstantFP, MVT::f128, Legal);
+
+  // Illegal floating-point operations.
+  setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
+  setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
+
+  setOperationAction(ISD::FCOS, MVT::f32, Expand);
+  setOperationAction(ISD::FCOS, MVT::f64, Expand);
+
+  setOperationAction(ISD::FEXP, MVT::f32, Expand);
+  setOperationAction(ISD::FEXP, MVT::f64, Expand);
+
+  setOperationAction(ISD::FEXP2, MVT::f32, Expand);
+  setOperationAction(ISD::FEXP2, MVT::f64, Expand);
+
+  setOperationAction(ISD::FLOG, MVT::f32, Expand);
+  setOperationAction(ISD::FLOG, MVT::f64, Expand);
+
+  setOperationAction(ISD::FLOG2, MVT::f32, Expand);
+  setOperationAction(ISD::FLOG2, MVT::f64, Expand);
+
+  setOperationAction(ISD::FLOG10, MVT::f32, Expand);
+  setOperationAction(ISD::FLOG10, MVT::f64, Expand);
+
+  setOperationAction(ISD::FPOW, MVT::f32, Expand);
+  setOperationAction(ISD::FPOW, MVT::f64, Expand);
+
+  setOperationAction(ISD::FPOWI, MVT::f32, Expand);
+  setOperationAction(ISD::FPOWI, MVT::f64, Expand);
+
+  setOperationAction(ISD::FREM, MVT::f32, Expand);
+  setOperationAction(ISD::FREM, MVT::f64, Expand);
+
+  setOperationAction(ISD::FSIN, MVT::f32, Expand);
+  setOperationAction(ISD::FSIN, MVT::f64, Expand);
+
+  setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
+  setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
+
+  // Virtually no operation on f128 is legal, but LLVM can't expand them when
+  // there's a valid register class, so we need custom operations in most cases.
+  setOperationAction(ISD::FABS,       MVT::f128, Expand);
+  setOperationAction(ISD::FADD,       MVT::f128, Custom);
+  setOperationAction(ISD::FCOPYSIGN,  MVT::f128, Expand);
+  setOperationAction(ISD::FCOS,       MVT::f128, Expand);
+  setOperationAction(ISD::FDIV,       MVT::f128, Custom);
+  setOperationAction(ISD::FMA,        MVT::f128, Expand);
+  setOperationAction(ISD::FMUL,       MVT::f128, Custom);
+  setOperationAction(ISD::FNEG,       MVT::f128, Expand);
+  setOperationAction(ISD::FP_EXTEND,  MVT::f128, Expand);
+  setOperationAction(ISD::FP_ROUND,   MVT::f128, Expand);
+  setOperationAction(ISD::FPOW,       MVT::f128, Expand);
+  setOperationAction(ISD::FREM,       MVT::f128, Expand);
+  setOperationAction(ISD::FRINT,      MVT::f128, Expand);
+  setOperationAction(ISD::FSIN,       MVT::f128, Expand);
+  setOperationAction(ISD::FSINCOS,    MVT::f128, Expand);
+  setOperationAction(ISD::FSQRT,      MVT::f128, Expand);
+  setOperationAction(ISD::FSUB,       MVT::f128, Custom);
+  setOperationAction(ISD::FTRUNC,     MVT::f128, Expand);
+  setOperationAction(ISD::SETCC,      MVT::f128, Custom);
+  setOperationAction(ISD::BR_CC,      MVT::f128, Custom);
+  setOperationAction(ISD::SELECT,     MVT::f128, Expand);
+  setOperationAction(ISD::SELECT_CC,  MVT::f128, Custom);
+  setOperationAction(ISD::FP_EXTEND,  MVT::f128, Custom);
+
+  // Lowering for many of the conversions is actually specified by the non-f128
+  // type. The LowerXXX function will be trivial when f128 isn't involved.
+  setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
+  setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
+  setOperationAction(ISD::FP_TO_SINT, MVT::i128, Custom);
+  setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
+  setOperationAction(ISD::FP_TO_UINT, MVT::i64, Custom);
+  setOperationAction(ISD::FP_TO_UINT, MVT::i128, Custom);
+  setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
+  setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
+  setOperationAction(ISD::SINT_TO_FP, MVT::i128, Custom);
+  setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
+  setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
+  setOperationAction(ISD::UINT_TO_FP, MVT::i128, Custom);
+  setOperationAction(ISD::FP_ROUND,  MVT::f32, Custom);
+  setOperationAction(ISD::FP_ROUND,  MVT::f64, Custom);
+
+  // This prevents LLVM trying to compress double constants into a floating
+  // constant-pool entry and trying to load from there. It's of doubtful benefit
+  // for A64: we'd need LDR followed by FCVT, I believe.
+  setLoadExtAction(ISD::EXTLOAD, MVT::f64, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
+
+  setTruncStoreAction(MVT::f128, MVT::f64, Expand);
+  setTruncStoreAction(MVT::f128, MVT::f32, Expand);
+  setTruncStoreAction(MVT::f128, MVT::f16, Expand);
+  setTruncStoreAction(MVT::f64, MVT::f32, Expand);
+  setTruncStoreAction(MVT::f64, MVT::f16, Expand);
+  setTruncStoreAction(MVT::f32, MVT::f16, Expand);
+
+  setOperationAction(ISD::EXCEPTIONADDR, MVT::i64, Expand);
+  setOperationAction(ISD::EHSELECTION, MVT::i64, Expand);
+
+  setExceptionPointerRegister(AArch64::X0);
+  setExceptionSelectorRegister(AArch64::X1);
+}
+
+EVT AArch64TargetLowering::getSetCCResultType(EVT VT) const {
+  // It's reasonably important that this value matches the "natural" legal
+  // promotion from i1 for scalar types. Otherwise LegalizeTypes can get itself
+  // in a twist (e.g. inserting an any_extend which then becomes i64 -> i64).
+  if (!VT.isVector()) return MVT::i32;
+  return VT.changeVectorElementTypeToInteger();
+}
+
+static void getExclusiveOperation(unsigned Size, unsigned &ldrOpc,
+                                  unsigned &strOpc) {
+  switch (Size) {
+  default: llvm_unreachable("unsupported size for atomic binary op!");
+  case 1:
+    ldrOpc = AArch64::LDXR_byte;
+    strOpc = AArch64::STXR_byte;
+    break;
+  case 2:
+    ldrOpc = AArch64::LDXR_hword;
+    strOpc = AArch64::STXR_hword;
+    break;
+  case 4:
+    ldrOpc = AArch64::LDXR_word;
+    strOpc = AArch64::STXR_word;
+    break;
+  case 8:
+    ldrOpc = AArch64::LDXR_dword;
+    strOpc = AArch64::STXR_dword;
+    break;
+  }
+}
+
+MachineBasicBlock *
+AArch64TargetLowering::emitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB,
+                                        unsigned Size,
+                                        unsigned BinOpcode) const {
+  // This also handles ATOMIC_SWAP, indicated by BinOpcode==0.
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+
+  const BasicBlock *LLVM_BB = BB->getBasicBlock();
+  MachineFunction *MF = BB->getParent();
+  MachineFunction::iterator It = BB;
+  ++It;
+
+  unsigned dest = MI->getOperand(0).getReg();
+  unsigned ptr = MI->getOperand(1).getReg();
+  unsigned incr = MI->getOperand(2).getReg();
+  DebugLoc dl = MI->getDebugLoc();
+
+  MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+
+  unsigned ldrOpc, strOpc;
+  getExclusiveOperation(Size, ldrOpc, strOpc);
+
+  MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  MF->insert(It, loopMBB);
+  MF->insert(It, exitMBB);
+
+  // Transfer the remainder of BB and its successor edges to exitMBB.
+  exitMBB->splice(exitMBB->begin(), BB,
+                  llvm::next(MachineBasicBlock::iterator(MI)),
+                  BB->end());
+  exitMBB->transferSuccessorsAndUpdatePHIs(BB);
+
+  const TargetRegisterClass *TRC
+    = Size == 8 ? &AArch64::GPR64RegClass : &AArch64::GPR32RegClass;
+  unsigned scratch = (!BinOpcode) ? incr : MRI.createVirtualRegister(TRC);
+
+  //  thisMBB:
+  //   ...
+  //   fallthrough --> loopMBB
+  BB->addSuccessor(loopMBB);
+
+  //  loopMBB:
+  //   ldxr dest, ptr
+  //   <binop> scratch, dest, incr
+  //   stxr stxr_status, scratch, ptr
+  //   cbnz stxr_status, loopMBB
+  //   fallthrough --> exitMBB
+  BB = loopMBB;
+  BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr);
+  if (BinOpcode) {
+    // All arithmetic operations we'll be creating are designed to take an extra
+    // shift or extend operand, which we can conveniently set to zero.
+
+    // Operand order needs to go the other way for NAND.
+    if (BinOpcode == AArch64::BICwww_lsl || BinOpcode == AArch64::BICxxx_lsl)
+      BuildMI(BB, dl, TII->get(BinOpcode), scratch)
+        .addReg(incr).addReg(dest).addImm(0);
+    else
+      BuildMI(BB, dl, TII->get(BinOpcode), scratch)
+        .addReg(dest).addReg(incr).addImm(0);
+  }
+
+  // From the stxr, the register is GPR32; from the cmp it's GPR32wsp
+  unsigned stxr_status = MRI.createVirtualRegister(&AArch64::GPR32RegClass);
+  MRI.constrainRegClass(stxr_status, &AArch64::GPR32wspRegClass);
+
+  BuildMI(BB, dl, TII->get(strOpc), stxr_status).addReg(scratch).addReg(ptr);
+  BuildMI(BB, dl, TII->get(AArch64::CBNZw))
+    .addReg(stxr_status).addMBB(loopMBB);
+
+  BB->addSuccessor(loopMBB);
+  BB->addSuccessor(exitMBB);
+
+  //  exitMBB:
+  //   ...
+  BB = exitMBB;
+
+  MI->eraseFromParent();   // The instruction is gone now.
+
+  return BB;
+}
+
+MachineBasicBlock *
+AArch64TargetLowering::emitAtomicBinaryMinMax(MachineInstr *MI,
+                                              MachineBasicBlock *BB,
+                                              unsigned Size,
+                                              unsigned CmpOp,
+                                              A64CC::CondCodes Cond) const {
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+
+  const BasicBlock *LLVM_BB = BB->getBasicBlock();
+  MachineFunction *MF = BB->getParent();
+  MachineFunction::iterator It = BB;
+  ++It;
+
+  unsigned dest = MI->getOperand(0).getReg();
+  unsigned ptr = MI->getOperand(1).getReg();
+  unsigned incr = MI->getOperand(2).getReg();
+  unsigned oldval = dest;
+  DebugLoc dl = MI->getDebugLoc();
+
+  MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+  const TargetRegisterClass *TRC, *TRCsp;
+  if (Size == 8) {
+    TRC = &AArch64::GPR64RegClass;
+    TRCsp = &AArch64::GPR64xspRegClass;
+  } else {
+    TRC = &AArch64::GPR32RegClass;
+    TRCsp = &AArch64::GPR32wspRegClass;
+  }
+
+  unsigned ldrOpc, strOpc;
+  getExclusiveOperation(Size, ldrOpc, strOpc);
+
+  MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  MF->insert(It, loopMBB);
+  MF->insert(It, exitMBB);
+
+  // Transfer the remainder of BB and its successor edges to exitMBB.
+  exitMBB->splice(exitMBB->begin(), BB,
+                  llvm::next(MachineBasicBlock::iterator(MI)),
+                  BB->end());
+  exitMBB->transferSuccessorsAndUpdatePHIs(BB);
+
+  unsigned scratch = MRI.createVirtualRegister(TRC);
+  MRI.constrainRegClass(scratch, TRCsp);
+
+  //  thisMBB:
+  //   ...
+  //   fallthrough --> loopMBB
+  BB->addSuccessor(loopMBB);
+
+  //  loopMBB:
+  //   ldxr dest, ptr
+  //   cmp incr, dest (, sign extend if necessary)
+  //   csel scratch, dest, incr, cond
+  //   stxr stxr_status, scratch, ptr
+  //   cbnz stxr_status, loopMBB
+  //   fallthrough --> exitMBB
+  BB = loopMBB;
+  BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr);
+
+  // Build compare and cmov instructions.
+  MRI.constrainRegClass(incr, TRCsp);
+  BuildMI(BB, dl, TII->get(CmpOp))
+    .addReg(incr).addReg(oldval).addImm(0);
+
+  BuildMI(BB, dl, TII->get(Size == 8 ? AArch64::CSELxxxc : AArch64::CSELwwwc),
+          scratch)
+    .addReg(oldval).addReg(incr).addImm(Cond);
+
+  unsigned stxr_status = MRI.createVirtualRegister(&AArch64::GPR32RegClass);
+  MRI.constrainRegClass(stxr_status, &AArch64::GPR32wspRegClass);
+
+  BuildMI(BB, dl, TII->get(strOpc), stxr_status)
+    .addReg(scratch).addReg(ptr);
+  BuildMI(BB, dl, TII->get(AArch64::CBNZw))
+    .addReg(stxr_status).addMBB(loopMBB);
+
+  BB->addSuccessor(loopMBB);
+  BB->addSuccessor(exitMBB);
+
+  //  exitMBB:
+  //   ...
+  BB = exitMBB;
+
+  MI->eraseFromParent();   // The instruction is gone now.
+
+  return BB;
+}
+
+MachineBasicBlock *
+AArch64TargetLowering::emitAtomicCmpSwap(MachineInstr *MI,
+                                         MachineBasicBlock *BB,
+                                         unsigned Size) const {
+  unsigned dest    = MI->getOperand(0).getReg();
+  unsigned ptr     = MI->getOperand(1).getReg();
+  unsigned oldval  = MI->getOperand(2).getReg();
+  unsigned newval  = MI->getOperand(3).getReg();
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  DebugLoc dl = MI->getDebugLoc();
+
+  MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+  const TargetRegisterClass *TRCsp;
+  TRCsp = Size == 8 ? &AArch64::GPR64xspRegClass : &AArch64::GPR32wspRegClass;
+
+  unsigned ldrOpc, strOpc;
+  getExclusiveOperation(Size, ldrOpc, strOpc);
+
+  MachineFunction *MF = BB->getParent();
+  const BasicBlock *LLVM_BB = BB->getBasicBlock();
+  MachineFunction::iterator It = BB;
+  ++It; // insert the new blocks after the current block
+
+  MachineBasicBlock *loop1MBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *loop2MBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  MF->insert(It, loop1MBB);
+  MF->insert(It, loop2MBB);
+  MF->insert(It, exitMBB);
+
+  // Transfer the remainder of BB and its successor edges to exitMBB.
+  exitMBB->splice(exitMBB->begin(), BB,
+                  llvm::next(MachineBasicBlock::iterator(MI)),
+                  BB->end());
+  exitMBB->transferSuccessorsAndUpdatePHIs(BB);
+
+  //  thisMBB:
+  //   ...
+  //   fallthrough --> loop1MBB
+  BB->addSuccessor(loop1MBB);
+
+  // loop1MBB:
+  //   ldxr dest, [ptr]
+  //   cmp dest, oldval
+  //   b.ne exitMBB
+  BB = loop1MBB;
+  BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr);
+
+  unsigned CmpOp = Size == 8 ? AArch64::CMPxx_lsl : AArch64::CMPww_lsl;
+  MRI.constrainRegClass(dest, TRCsp);
+  BuildMI(BB, dl, TII->get(CmpOp))
+    .addReg(dest).addReg(oldval).addImm(0);
+  BuildMI(BB, dl, TII->get(AArch64::Bcc))
+    .addImm(A64CC::NE).addMBB(exitMBB);
+  BB->addSuccessor(loop2MBB);
+  BB->addSuccessor(exitMBB);
+
+  // loop2MBB:
+  //   strex stxr_status, newval, [ptr]
+  //   cbnz stxr_status, loop1MBB
+  BB = loop2MBB;
+  unsigned stxr_status = MRI.createVirtualRegister(&AArch64::GPR32RegClass);
+  MRI.constrainRegClass(stxr_status, &AArch64::GPR32wspRegClass);
+
+  BuildMI(BB, dl, TII->get(strOpc), stxr_status).addReg(newval).addReg(ptr);
+  BuildMI(BB, dl, TII->get(AArch64::CBNZw))
+    .addReg(stxr_status).addMBB(loop1MBB);
+  BB->addSuccessor(loop1MBB);
+  BB->addSuccessor(exitMBB);
+
+  //  exitMBB:
+  //   ...
+  BB = exitMBB;
+
+  MI->eraseFromParent();   // The instruction is gone now.
+
+  return BB;
+}
+
+MachineBasicBlock *
+AArch64TargetLowering::EmitF128CSEL(MachineInstr *MI,
+                                    MachineBasicBlock *MBB) const {
+  // We materialise the F128CSEL pseudo-instruction using conditional branches
+  // and loads, giving an instruciton sequence like:
+  //     str q0, [sp]
+  //     b.ne IfTrue
+  //     b Finish
+  // IfTrue:
+  //     str q1, [sp]
+  // Finish:
+  //     ldr q0, [sp]
+  //
+  // Using virtual registers would probably not be beneficial since COPY
+  // instructions are expensive for f128 (there's no actual instruction to
+  // implement them).
+  //
+  // An alternative would be to do an integer-CSEL on some address. E.g.:
+  //     mov x0, sp
+  //     add x1, sp, #16
+  //     str q0, [x0]
+  //     str q1, [x1]
+  //     csel x0, x0, x1, ne
+  //     ldr q0, [x0]
+  //
+  // It's unclear which approach is actually optimal.
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  MachineFunction *MF = MBB->getParent();
+  const BasicBlock *LLVM_BB = MBB->getBasicBlock();
+  DebugLoc DL = MI->getDebugLoc();
+  MachineFunction::iterator It = MBB;
+  ++It;
+
+  unsigned DestReg = MI->getOperand(0).getReg();
+  unsigned IfTrueReg = MI->getOperand(1).getReg();
+  unsigned IfFalseReg = MI->getOperand(2).getReg();
+  unsigned CondCode = MI->getOperand(3).getImm();
+  bool NZCVKilled = MI->getOperand(4).isKill();
+
+  MachineBasicBlock *TrueBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *EndBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  MF->insert(It, TrueBB);
+  MF->insert(It, EndBB);
+
+  // Transfer rest of current basic-block to EndBB
+  EndBB->splice(EndBB->begin(), MBB,
+                llvm::next(MachineBasicBlock::iterator(MI)),
+                MBB->end());
+  EndBB->transferSuccessorsAndUpdatePHIs(MBB);
+
+  // We need somewhere to store the f128 value needed.
+  int ScratchFI = MF->getFrameInfo()->CreateSpillStackObject(16, 16);
+
+  //     [... start of incoming MBB ...]
+  //     str qIFFALSE, [sp]
+  //     b.cc IfTrue
+  //     b Done
+  BuildMI(MBB, DL, TII->get(AArch64::LSFP128_STR))
+    .addReg(IfFalseReg)
+    .addFrameIndex(ScratchFI)
+    .addImm(0);
+  BuildMI(MBB, DL, TII->get(AArch64::Bcc))
+    .addImm(CondCode)
+    .addMBB(TrueBB);
+  BuildMI(MBB, DL, TII->get(AArch64::Bimm))
+    .addMBB(EndBB);
+  MBB->addSuccessor(TrueBB);
+  MBB->addSuccessor(EndBB);
+
+  // IfTrue:
+  //     str qIFTRUE, [sp]
+  BuildMI(TrueBB, DL, TII->get(AArch64::LSFP128_STR))
+    .addReg(IfTrueReg)
+    .addFrameIndex(ScratchFI)
+    .addImm(0);
+
+  // Note: fallthrough. We can rely on LLVM adding a branch if it reorders the
+  // blocks.
+  TrueBB->addSuccessor(EndBB);
+
+  // Done:
+  //     ldr qDEST, [sp]
+  //     [... rest of incoming MBB ...]
+  if (!NZCVKilled)
+    EndBB->addLiveIn(AArch64::NZCV);
+  MachineInstr *StartOfEnd = EndBB->begin();
+  BuildMI(*EndBB, StartOfEnd, DL, TII->get(AArch64::LSFP128_LDR), DestReg)
+    .addFrameIndex(ScratchFI)
+    .addImm(0);
+
+  MI->eraseFromParent();
+  return EndBB;
+}
+
+MachineBasicBlock *
+AArch64TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
+                                                 MachineBasicBlock *MBB) const {
+  switch (MI->getOpcode()) {
+  default: llvm_unreachable("Unhandled instruction with custom inserter");
+  case AArch64::F128CSEL:
+    return EmitF128CSEL(MI, MBB);
+  case AArch64::ATOMIC_LOAD_ADD_I8:
+    return emitAtomicBinary(MI, MBB, 1, AArch64::ADDwww_lsl);
+  case AArch64::ATOMIC_LOAD_ADD_I16:
+    return emitAtomicBinary(MI, MBB, 2, AArch64::ADDwww_lsl);
+  case AArch64::ATOMIC_LOAD_ADD_I32:
+    return emitAtomicBinary(MI, MBB, 4, AArch64::ADDwww_lsl);
+  case AArch64::ATOMIC_LOAD_ADD_I64:
+    return emitAtomicBinary(MI, MBB, 8, AArch64::ADDxxx_lsl);
+
+  case AArch64::ATOMIC_LOAD_SUB_I8:
+    return emitAtomicBinary(MI, MBB, 1, AArch64::SUBwww_lsl);
+  case AArch64::ATOMIC_LOAD_SUB_I16:
+    return emitAtomicBinary(MI, MBB, 2, AArch64::SUBwww_lsl);
+  case AArch64::ATOMIC_LOAD_SUB_I32:
+    return emitAtomicBinary(MI, MBB, 4, AArch64::SUBwww_lsl);
+  case AArch64::ATOMIC_LOAD_SUB_I64:
+    return emitAtomicBinary(MI, MBB, 8, AArch64::SUBxxx_lsl);
+
+  case AArch64::ATOMIC_LOAD_AND_I8:
+    return emitAtomicBinary(MI, MBB, 1, AArch64::ANDwww_lsl);
+  case AArch64::ATOMIC_LOAD_AND_I16:
+    return emitAtomicBinary(MI, MBB, 2, AArch64::ANDwww_lsl);
+  case AArch64::ATOMIC_LOAD_AND_I32:
+    return emitAtomicBinary(MI, MBB, 4, AArch64::ANDwww_lsl);
+  case AArch64::ATOMIC_LOAD_AND_I64:
+    return emitAtomicBinary(MI, MBB, 8, AArch64::ANDxxx_lsl);
+
+  case AArch64::ATOMIC_LOAD_OR_I8:
+    return emitAtomicBinary(MI, MBB, 1, AArch64::ORRwww_lsl);
+  case AArch64::ATOMIC_LOAD_OR_I16:
+    return emitAtomicBinary(MI, MBB, 2, AArch64::ORRwww_lsl);
+  case AArch64::ATOMIC_LOAD_OR_I32:
+    return emitAtomicBinary(MI, MBB, 4, AArch64::ORRwww_lsl);
+  case AArch64::ATOMIC_LOAD_OR_I64:
+    return emitAtomicBinary(MI, MBB, 8, AArch64::ORRxxx_lsl);
+
+  case AArch64::ATOMIC_LOAD_XOR_I8:
+    return emitAtomicBinary(MI, MBB, 1, AArch64::EORwww_lsl);
+  case AArch64::ATOMIC_LOAD_XOR_I16:
+    return emitAtomicBinary(MI, MBB, 2, AArch64::EORwww_lsl);
+  case AArch64::ATOMIC_LOAD_XOR_I32:
+    return emitAtomicBinary(MI, MBB, 4, AArch64::EORwww_lsl);
+  case AArch64::ATOMIC_LOAD_XOR_I64:
+    return emitAtomicBinary(MI, MBB, 8, AArch64::EORxxx_lsl);
+
+  case AArch64::ATOMIC_LOAD_NAND_I8:
+    return emitAtomicBinary(MI, MBB, 1, AArch64::BICwww_lsl);
+  case AArch64::ATOMIC_LOAD_NAND_I16:
+    return emitAtomicBinary(MI, MBB, 2, AArch64::BICwww_lsl);
+  case AArch64::ATOMIC_LOAD_NAND_I32:
+    return emitAtomicBinary(MI, MBB, 4, AArch64::BICwww_lsl);
+  case AArch64::ATOMIC_LOAD_NAND_I64:
+    return emitAtomicBinary(MI, MBB, 8, AArch64::BICxxx_lsl);
+
+  case AArch64::ATOMIC_LOAD_MIN_I8:
+    return emitAtomicBinaryMinMax(MI, MBB, 1, AArch64::CMPww_sxtb, A64CC::GT);
+  case AArch64::ATOMIC_LOAD_MIN_I16:
+    return emitAtomicBinaryMinMax(MI, MBB, 2, AArch64::CMPww_sxth, A64CC::GT);
+  case AArch64::ATOMIC_LOAD_MIN_I32:
+    return emitAtomicBinaryMinMax(MI, MBB, 4, AArch64::CMPww_lsl, A64CC::GT);
+  case AArch64::ATOMIC_LOAD_MIN_I64:
+    return emitAtomicBinaryMinMax(MI, MBB, 8, AArch64::CMPxx_lsl, A64CC::GT);
+
+  case AArch64::ATOMIC_LOAD_MAX_I8:
+    return emitAtomicBinaryMinMax(MI, MBB, 1, AArch64::CMPww_sxtb, A64CC::LT);
+  case AArch64::ATOMIC_LOAD_MAX_I16:
+    return emitAtomicBinaryMinMax(MI, MBB, 2, AArch64::CMPww_sxth, A64CC::LT);
+  case AArch64::ATOMIC_LOAD_MAX_I32:
+    return emitAtomicBinaryMinMax(MI, MBB, 4, AArch64::CMPww_lsl, A64CC::LT);
+  case AArch64::ATOMIC_LOAD_MAX_I64:
+    return emitAtomicBinaryMinMax(MI, MBB, 8, AArch64::CMPxx_lsl, A64CC::LT);
+
+  case AArch64::ATOMIC_LOAD_UMIN_I8:
+    return emitAtomicBinaryMinMax(MI, MBB, 1, AArch64::CMPww_uxtb, A64CC::HI);
+  case AArch64::ATOMIC_LOAD_UMIN_I16:
+    return emitAtomicBinaryMinMax(MI, MBB, 2, AArch64::CMPww_uxth, A64CC::HI);
+  case AArch64::ATOMIC_LOAD_UMIN_I32:
+    return emitAtomicBinaryMinMax(MI, MBB, 4, AArch64::CMPww_lsl, A64CC::HI);
+  case AArch64::ATOMIC_LOAD_UMIN_I64:
+    return emitAtomicBinaryMinMax(MI, MBB, 8, AArch64::CMPxx_lsl, A64CC::HI);
+
+  case AArch64::ATOMIC_LOAD_UMAX_I8:
+    return emitAtomicBinaryMinMax(MI, MBB, 1, AArch64::CMPww_uxtb, A64CC::LO);
+  case AArch64::ATOMIC_LOAD_UMAX_I16:
+    return emitAtomicBinaryMinMax(MI, MBB, 2, AArch64::CMPww_uxth, A64CC::LO);
+  case AArch64::ATOMIC_LOAD_UMAX_I32:
+    return emitAtomicBinaryMinMax(MI, MBB, 4, AArch64::CMPww_lsl, A64CC::LO);
+  case AArch64::ATOMIC_LOAD_UMAX_I64:
+    return emitAtomicBinaryMinMax(MI, MBB, 8, AArch64::CMPxx_lsl, A64CC::LO);
+
+  case AArch64::ATOMIC_SWAP_I8:
+    return emitAtomicBinary(MI, MBB, 1, 0);
+  case AArch64::ATOMIC_SWAP_I16:
+    return emitAtomicBinary(MI, MBB, 2, 0);
+  case AArch64::ATOMIC_SWAP_I32:
+    return emitAtomicBinary(MI, MBB, 4, 0);
+  case AArch64::ATOMIC_SWAP_I64:
+    return emitAtomicBinary(MI, MBB, 8, 0);
+
+  case AArch64::ATOMIC_CMP_SWAP_I8:
+    return emitAtomicCmpSwap(MI, MBB, 1);
+  case AArch64::ATOMIC_CMP_SWAP_I16:
+    return emitAtomicCmpSwap(MI, MBB, 2);
+  case AArch64::ATOMIC_CMP_SWAP_I32:
+    return emitAtomicCmpSwap(MI, MBB, 4);
+  case AArch64::ATOMIC_CMP_SWAP_I64:
+    return emitAtomicCmpSwap(MI, MBB, 8);
+  }
+}
+
+
+const char *AArch64TargetLowering::getTargetNodeName(unsigned Opcode) const {
+  switch (Opcode) {
+  case AArch64ISD::BR_CC:          return "AArch64ISD::BR_CC";
+  case AArch64ISD::Call:           return "AArch64ISD::Call";
+  case AArch64ISD::FPMOV:          return "AArch64ISD::FPMOV";
+  case AArch64ISD::GOTLoad:        return "AArch64ISD::GOTLoad";
+  case AArch64ISD::BFI:            return "AArch64ISD::BFI";
+  case AArch64ISD::EXTR:           return "AArch64ISD::EXTR";
+  case AArch64ISD::Ret:            return "AArch64ISD::Ret";
+  case AArch64ISD::SBFX:           return "AArch64ISD::SBFX";
+  case AArch64ISD::SELECT_CC:      return "AArch64ISD::SELECT_CC";
+  case AArch64ISD::SETCC:          return "AArch64ISD::SETCC";
+  case AArch64ISD::TC_RETURN:      return "AArch64ISD::TC_RETURN";
+  case AArch64ISD::THREAD_POINTER: return "AArch64ISD::THREAD_POINTER";
+  case AArch64ISD::TLSDESCCALL:    return "AArch64ISD::TLSDESCCALL";
+  case AArch64ISD::WrapperSmall:   return "AArch64ISD::WrapperSmall";
+
+  default:                       return NULL;
+  }
+}
+
+static const uint16_t AArch64FPRArgRegs[] = {
+  AArch64::Q0, AArch64::Q1, AArch64::Q2, AArch64::Q3,
+  AArch64::Q4, AArch64::Q5, AArch64::Q6, AArch64::Q7
+};
+static const unsigned NumFPRArgRegs = llvm::array_lengthof(AArch64FPRArgRegs);
+
+static const uint16_t AArch64ArgRegs[] = {
+  AArch64::X0, AArch64::X1, AArch64::X2, AArch64::X3,
+  AArch64::X4, AArch64::X5, AArch64::X6, AArch64::X7
+};
+static const unsigned NumArgRegs = llvm::array_lengthof(AArch64ArgRegs);
+
+static bool CC_AArch64NoMoreRegs(unsigned ValNo, MVT ValVT, MVT LocVT,
+                                 CCValAssign::LocInfo LocInfo,
+                                 ISD::ArgFlagsTy ArgFlags, CCState &State) {
+  // Mark all remaining general purpose registers as allocated. We don't
+  // backtrack: if (for example) an i128 gets put on the stack, no subsequent
+  // i64 will go in registers (C.11).
+  for (unsigned i = 0; i < NumArgRegs; ++i)
+    State.AllocateReg(AArch64ArgRegs[i]);
+
+  return false;
+}
+
+#include "AArch64GenCallingConv.inc"
+
+CCAssignFn *AArch64TargetLowering::CCAssignFnForNode(CallingConv::ID CC) const {
+
+  switch(CC) {
+  default: llvm_unreachable("Unsupported calling convention");
+  case CallingConv::Fast:
+  case CallingConv::C:
+    return CC_A64_APCS;
+  }
+}
+
+void
+AArch64TargetLowering::SaveVarArgRegisters(CCState &CCInfo, SelectionDAG &DAG,
+                                           DebugLoc DL, SDValue &Chain) const {
+  MachineFunction &MF = DAG.getMachineFunction();
+  MachineFrameInfo *MFI = MF.getFrameInfo();
+  AArch64MachineFunctionInfo *FuncInfo
+    = MF.getInfo<AArch64MachineFunctionInfo>();
+
+  SmallVector<SDValue, 8> MemOps;
+
+  unsigned FirstVariadicGPR = CCInfo.getFirstUnallocated(AArch64ArgRegs,
+                                                         NumArgRegs);
+  unsigned FirstVariadicFPR = CCInfo.getFirstUnallocated(AArch64FPRArgRegs,
+                                                         NumFPRArgRegs);
+
+  unsigned GPRSaveSize = 8 * (NumArgRegs - FirstVariadicGPR);
+  int GPRIdx = 0;
+  if (GPRSaveSize != 0) {
+    GPRIdx = MFI->CreateStackObject(GPRSaveSize, 8, false);
+
+    SDValue FIN = DAG.getFrameIndex(GPRIdx, getPointerTy());
+
+    for (unsigned i = FirstVariadicGPR; i < NumArgRegs; ++i) {
+      unsigned VReg = MF.addLiveIn(AArch64ArgRegs[i], &AArch64::GPR64RegClass);
+      SDValue Val = DAG.getCopyFromReg(Chain, DL, VReg, MVT::i64);
+      SDValue Store = DAG.getStore(Val.getValue(1), DL, Val, FIN,
+                                   MachinePointerInfo::getStack(i * 8),
+                                   false, false, 0);
+      MemOps.push_back(Store);
+      FIN = DAG.getNode(ISD::ADD, DL, getPointerTy(), FIN,
+                        DAG.getConstant(8, getPointerTy()));
+    }
+  }
+
+  unsigned FPRSaveSize = 16 * (NumFPRArgRegs - FirstVariadicFPR);
+  int FPRIdx = 0;
+  if (FPRSaveSize != 0) {
+    FPRIdx = MFI->CreateStackObject(FPRSaveSize, 16, false);
+
+    SDValue FIN = DAG.getFrameIndex(FPRIdx, getPointerTy());
+
+    for (unsigned i = FirstVariadicFPR; i < NumFPRArgRegs; ++i) {
+      unsigned VReg = MF.addLiveIn(AArch64FPRArgRegs[i],
+                                   &AArch64::FPR128RegClass);
+      SDValue Val = DAG.getCopyFromReg(Chain, DL, VReg, MVT::f128);
+      SDValue Store = DAG.getStore(Val.getValue(1), DL, Val, FIN,
+                                   MachinePointerInfo::getStack(i * 16),
+                                   false, false, 0);
+      MemOps.push_back(Store);
+      FIN = DAG.getNode(ISD::ADD, DL, getPointerTy(), FIN,
+                        DAG.getConstant(16, getPointerTy()));
+    }
+  }
+
+  int StackIdx = MFI->CreateFixedObject(8, CCInfo.getNextStackOffset(), true);
+
+  FuncInfo->setVariadicStackIdx(StackIdx);
+  FuncInfo->setVariadicGPRIdx(GPRIdx);
+  FuncInfo->setVariadicGPRSize(GPRSaveSize);
+  FuncInfo->setVariadicFPRIdx(FPRIdx);
+  FuncInfo->setVariadicFPRSize(FPRSaveSize);
+
+  if (!MemOps.empty()) {
+    Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, &MemOps[0],
+                        MemOps.size());
+  }
+}
+
+
+SDValue
+AArch64TargetLowering::LowerFormalArguments(SDValue Chain,
+                                      CallingConv::ID CallConv, bool isVarArg,
+                                      const SmallVectorImpl<ISD::InputArg> &Ins,
+                                      DebugLoc dl, SelectionDAG &DAG,
+                                      SmallVectorImpl<SDValue> &InVals) const {
+  MachineFunction &MF = DAG.getMachineFunction();
+  AArch64MachineFunctionInfo *FuncInfo
+    = MF.getInfo<AArch64MachineFunctionInfo>();
+  MachineFrameInfo *MFI = MF.getFrameInfo();
+  bool TailCallOpt = MF.getTarget().Options.GuaranteedTailCallOpt;
+
+  SmallVector<CCValAssign, 16> ArgLocs;
+  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
+                 getTargetMachine(), ArgLocs, *DAG.getContext());
+  CCInfo.AnalyzeFormalArguments(Ins, CCAssignFnForNode(CallConv));
+
+  SmallVector<SDValue, 16> ArgValues;
+
+  SDValue ArgValue;
+  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
+    CCValAssign &VA = ArgLocs[i];
+    ISD::ArgFlagsTy Flags = Ins[i].Flags;
+
+    if (Flags.isByVal()) {
+      // Byval is used for small structs and HFAs in the PCS, but the system
+      // should work in a non-compliant manner for larger structs.
+      EVT PtrTy = getPointerTy();
+      int Size = Flags.getByValSize();
+      unsigned NumRegs = (Size + 7) / 8;
+
+      unsigned FrameIdx = MFI->CreateFixedObject(8 * NumRegs,
+                                                 VA.getLocMemOffset(),
+                                                 false);
+      SDValue FrameIdxN = DAG.getFrameIndex(FrameIdx, PtrTy);
+      InVals.push_back(FrameIdxN);
+
+      continue;
+    } else if (VA.isRegLoc()) {
+      MVT RegVT = VA.getLocVT();
+      const TargetRegisterClass *RC = getRegClassFor(RegVT);
+      unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC);
+
+      ArgValue = DAG.getCopyFromReg(Chain, dl, Reg, RegVT);
+    } else { // VA.isRegLoc()
+      assert(VA.isMemLoc());
+
+      int FI = MFI->CreateFixedObject(VA.getLocVT().getSizeInBits()/8,
+                                      VA.getLocMemOffset(), true);
+
+      SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
+      ArgValue = DAG.getLoad(VA.getLocVT(), dl, Chain, FIN,
+                             MachinePointerInfo::getFixedStack(FI),
+                             false, false, false, 0);
+
+
+    }
+
+    switch (VA.getLocInfo()) {
+    default: llvm_unreachable("Unknown loc info!");
+    case CCValAssign::Full: break;
+    case CCValAssign::BCvt:
+      ArgValue = DAG.getNode(ISD::BITCAST,dl, VA.getValVT(), ArgValue);
+      break;
+    case CCValAssign::SExt:
+    case CCValAssign::ZExt:
+    case CCValAssign::AExt: {
+      unsigned DestSize = VA.getValVT().getSizeInBits();
+      unsigned DestSubReg;
+
+      switch (DestSize) {
+      case 8: DestSubReg = AArch64::sub_8; break;
+      case 16: DestSubReg = AArch64::sub_16; break;
+      case 32: DestSubReg = AArch64::sub_32; break;
+      case 64: DestSubReg = AArch64::sub_64; break;
+      default: llvm_unreachable("Unexpected argument promotion");
+      }
+
+      ArgValue = SDValue(DAG.getMachineNode(TargetOpcode::EXTRACT_SUBREG, dl,
+                                   VA.getValVT(), ArgValue,
+                                   DAG.getTargetConstant(DestSubReg, MVT::i32)),
+                         0);
+      break;
+    }
+    }
+
+    InVals.push_back(ArgValue);
+  }
+
+  if (isVarArg)
+    SaveVarArgRegisters(CCInfo, DAG, dl, Chain);
+
+  unsigned StackArgSize = CCInfo.getNextStackOffset();
+  if (DoesCalleeRestoreStack(CallConv, TailCallOpt)) {
+    // This is a non-standard ABI so by fiat I say we're allowed to make full
+    // use of the stack area to be popped, which must be aligned to 16 bytes in
+    // any case:
+    StackArgSize = RoundUpToAlignment(StackArgSize, 16);
+
+    // If we're expected to restore the stack (e.g. fastcc) then we'll be adding
+    // a multiple of 16.
+    FuncInfo->setArgumentStackToRestore(StackArgSize);
+
+    // This realignment carries over to the available bytes below. Our own
+    // callers will guarantee the space is free by giving an aligned value to
+    // CALLSEQ_START.
+  }
+  // Even if we're not expected to free up the space, it's useful to know how
+  // much is there while considering tail calls (because we can reuse it).
+  FuncInfo->setBytesInStackArgArea(StackArgSize);
+
+  return Chain;
+}
+
+SDValue
+AArch64TargetLowering::LowerReturn(SDValue Chain,
+                                   CallingConv::ID CallConv, bool isVarArg,
+                                   const SmallVectorImpl<ISD::OutputArg> &Outs,
+                                   const SmallVectorImpl<SDValue> &OutVals,
+                                   DebugLoc dl, SelectionDAG &DAG) const {
+  // CCValAssign - represent the assignment of the return value to a location.
+  SmallVector<CCValAssign, 16> RVLocs;
+
+  // CCState - Info about the registers and stack slots.
+  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
+                 getTargetMachine(), RVLocs, *DAG.getContext());
+
+  // Analyze outgoing return values.
+  CCInfo.AnalyzeReturn(Outs, CCAssignFnForNode(CallConv));
+
+  SDValue Flag;
+  SmallVector<SDValue, 4> RetOps(1, Chain);
+
+  for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
+    // PCS: "If the type, T, of the result of a function is such that
+    // void func(T arg) would require that arg be passed as a value in a
+    // register (or set of registers) according to the rules in 5.4, then the
+    // result is returned in the same registers as would be used for such an
+    // argument.
+    //
+    // Otherwise, the caller shall reserve a block of memory of sufficient
+    // size and alignment to hold the result. The address of the memory block
+    // shall be passed as an additional argument to the function in x8."
+    //
+    // This is implemented in two places. The register-return values are dealt
+    // with here, more complex returns are passed as an sret parameter, which
+    // means we don't have to worry about it during actual return.
+    CCValAssign &VA = RVLocs[i];
+    assert(VA.isRegLoc() && "Only register-returns should be created by PCS");
+
+
+    SDValue Arg = OutVals[i];
+
+    // There's no convenient note in the ABI about this as there is for normal
+    // arguments, but it says return values are passed in the same registers as
+    // an argument would be. I believe that includes the comments about
+    // unspecified higher bits, putting the burden of widening on the *caller*
+    // for return values.
+    switch (VA.getLocInfo()) {
+    default: llvm_unreachable("Unknown loc info");
+    case CCValAssign::Full: break;
+    case CCValAssign::SExt:
+    case CCValAssign::ZExt:
+    case CCValAssign::AExt:
+      // Floating-point values should only be extended when they're going into
+      // memory, which can't happen here so an integer extend is acceptable.
+      Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
+      break;
+    case CCValAssign::BCvt:
+      Arg = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), Arg);
+      break;
+    }
+
+    Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), Arg, Flag);
+    Flag = Chain.getValue(1);
+    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
+  }
+
+  RetOps[0] = Chain;  // Update chain.
+
+  // Add the flag if we have it.
+  if (Flag.getNode())
+    RetOps.push_back(Flag);
+
+  return DAG.getNode(AArch64ISD::Ret, dl, MVT::Other,
+                     &RetOps[0], RetOps.size());
+}
+
+SDValue
+AArch64TargetLowering::LowerCall(CallLoweringInfo &CLI,
+                                 SmallVectorImpl<SDValue> &InVals) const {
+  SelectionDAG &DAG                     = CLI.DAG;
+  DebugLoc &dl                          = CLI.DL;
+  SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
+  SmallVector<SDValue, 32> &OutVals     = CLI.OutVals;
+  SmallVector<ISD::InputArg, 32> &Ins   = CLI.Ins;
+  SDValue Chain                         = CLI.Chain;
+  SDValue Callee                        = CLI.Callee;
+  bool &IsTailCall                      = CLI.IsTailCall;
+  CallingConv::ID CallConv              = CLI.CallConv;
+  bool IsVarArg                         = CLI.IsVarArg;
+
+  MachineFunction &MF = DAG.getMachineFunction();
+  AArch64MachineFunctionInfo *FuncInfo
+    = MF.getInfo<AArch64MachineFunctionInfo>();
+  bool TailCallOpt = MF.getTarget().Options.GuaranteedTailCallOpt;
+  bool IsStructRet = !Outs.empty() && Outs[0].Flags.isSRet();
+  bool IsSibCall = false;
+
+  if (IsTailCall) {
+    IsTailCall = IsEligibleForTailCallOptimization(Callee, CallConv,
+                    IsVarArg, IsStructRet, MF.getFunction()->hasStructRetAttr(),
+                                                   Outs, OutVals, Ins, DAG);
+
+    // A sibling call is one where we're under the usual C ABI and not planning
+    // to change that but can still do a tail call:
+    if (!TailCallOpt && IsTailCall)
+      IsSibCall = true;
+  }
+
+  SmallVector<CCValAssign, 16> ArgLocs;
+  CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(),
+                 getTargetMachine(), ArgLocs, *DAG.getContext());
+  CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForNode(CallConv));
+
+  // On AArch64 (and all other architectures I'm aware of) the most this has to
+  // do is adjust the stack pointer.
+  unsigned NumBytes = RoundUpToAlignment(CCInfo.getNextStackOffset(), 16);
+  if (IsSibCall) {
+    // Since we're not changing the ABI to make this a tail call, the memory
+    // operands are already available in the caller's incoming argument space.
+    NumBytes = 0;
+  }
+
+  // FPDiff is the byte offset of the call's argument area from the callee's.
+  // Stores to callee stack arguments will be placed in FixedStackSlots offset
+  // by this amount for a tail call. In a sibling call it must be 0 because the
+  // caller will deallocate the entire stack and the callee still expects its
+  // arguments to begin at SP+0. Completely unused for non-tail calls.
+  int FPDiff = 0;
+
+  if (IsTailCall && !IsSibCall) {
+    unsigned NumReusableBytes = FuncInfo->getBytesInStackArgArea();
+
+    // FPDiff will be negative if this tail call requires more space than we
+    // would automatically have in our incoming argument space. Positive if we
+    // can actually shrink the stack.
+    FPDiff = NumReusableBytes - NumBytes;
+
+    // The stack pointer must be 16-byte aligned at all times it's used for a
+    // memory operation, which in practice means at *all* times and in
+    // particular across call boundaries. Therefore our own arguments started at
+    // a 16-byte aligned SP and the delta applied for the tail call should
+    // satisfy the same constraint.
+    assert(FPDiff % 16 == 0 && "unaligned stack on tail call");
+  }
+
+  if (!IsSibCall)
+    Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true));
+
+  SDValue StackPtr = DAG.getCopyFromReg(Chain, dl, AArch64::XSP,
+                                        getPointerTy());
+
+  SmallVector<SDValue, 8> MemOpChains;
+  SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
+
+  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
+    CCValAssign &VA = ArgLocs[i];
+    ISD::ArgFlagsTy Flags = Outs[i].Flags;
+    SDValue Arg = OutVals[i];
+
+    // Callee does the actual widening, so all extensions just use an implicit
+    // definition of the rest of the Loc. Aesthetically, this would be nicer as
+    // an ANY_EXTEND, but that isn't valid for floating-point types and this
+    // alternative works on integer types too.
+    switch (VA.getLocInfo()) {
+    default: llvm_unreachable("Unknown loc info!");
+    case CCValAssign::Full: break;
+    case CCValAssign::SExt:
+    case CCValAssign::ZExt:
+    case CCValAssign::AExt: {
+      unsigned SrcSize = VA.getValVT().getSizeInBits();
+      unsigned SrcSubReg;
+
+      switch (SrcSize) {
+      case 8: SrcSubReg = AArch64::sub_8; break;
+      case 16: SrcSubReg = AArch64::sub_16; break;
+      case 32: SrcSubReg = AArch64::sub_32; break;
+      case 64: SrcSubReg = AArch64::sub_64; break;
+      default: llvm_unreachable("Unexpected argument promotion");
+      }
+
+      Arg = SDValue(DAG.getMachineNode(TargetOpcode::INSERT_SUBREG, dl,
+                                    VA.getLocVT(),
+                                    DAG.getUNDEF(VA.getLocVT()),
+                                    Arg,
+                                    DAG.getTargetConstant(SrcSubReg, MVT::i32)),
+                    0);
+
+      break;
+    }
+    case CCValAssign::BCvt:
+      Arg = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), Arg);
+      break;
+    }
+
+    if (VA.isRegLoc()) {
+      // A normal register (sub-) argument. For now we just note it down because
+      // we want to copy things into registers as late as possible to avoid
+      // register-pressure (and possibly worse).
+      RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
+      continue;
+    }
+
+    assert(VA.isMemLoc() && "unexpected argument location");
+
+    SDValue DstAddr;
+    MachinePointerInfo DstInfo;
+    if (IsTailCall) {
+      uint32_t OpSize = Flags.isByVal() ? Flags.getByValSize() :
+                                          VA.getLocVT().getSizeInBits();
+      OpSize = (OpSize + 7) / 8;
+      int32_t Offset = VA.getLocMemOffset() + FPDiff;
+      int FI = MF.getFrameInfo()->CreateFixedObject(OpSize, Offset, true);
+
+      DstAddr = DAG.getFrameIndex(FI, getPointerTy());
+      DstInfo = MachinePointerInfo::getFixedStack(FI);
+
+      // Make sure any stack arguments overlapping with where we're storing are
+      // loaded before this eventual operation. Otherwise they'll be clobbered.
+      Chain = addTokenForArgument(Chain, DAG, MF.getFrameInfo(), FI);
+    } else {
+      SDValue PtrOff = DAG.getIntPtrConstant(VA.getLocMemOffset());
+
+      DstAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, PtrOff);
+      DstInfo = MachinePointerInfo::getStack(VA.getLocMemOffset());
+    }
+
+    if (Flags.isByVal()) {
+      SDValue SizeNode = DAG.getConstant(Flags.getByValSize(), MVT::i64);
+      SDValue Cpy = DAG.getMemcpy(Chain, dl, DstAddr, Arg, SizeNode,
+                                  Flags.getByValAlign(),
+                                  /*isVolatile = */ false,
+                                  /*alwaysInline = */ false,
+                                  DstInfo, MachinePointerInfo(0));
+      MemOpChains.push_back(Cpy);
+    } else {
+      // Normal stack argument, put it where it's needed.
+      SDValue Store = DAG.getStore(Chain, dl, Arg, DstAddr, DstInfo,
+                                   false, false, 0);
+      MemOpChains.push_back(Store);
+    }
+  }
+
+  // The loads and stores generated above shouldn't clash with each
+  // other. Combining them with this TokenFactor notes that fact for the rest of
+  // the backend.
+  if (!MemOpChains.empty())
+    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
+                        &MemOpChains[0], MemOpChains.size());
+
+  // Most of the rest of the instructions need to be glued together; we don't
+  // want assignments to actual registers used by a call to be rearranged by a
+  // well-meaning scheduler.
+  SDValue InFlag;
+
+  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
+    Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
+                             RegsToPass[i].second, InFlag);
+    InFlag = Chain.getValue(1);
+  }
+
+  // The linker is responsible for inserting veneers when necessary to put a
+  // function call destination in range, so we don't need to bother with a
+  // wrapper here.
+  if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
+    const GlobalValue *GV = G->getGlobal();
+    Callee = DAG.getTargetGlobalAddress(GV, dl, getPointerTy());
+  } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
+    const char *Sym = S->getSymbol();
+    Callee = DAG.getTargetExternalSymbol(Sym, getPointerTy());
+  }
+
+  // We don't usually want to end the call-sequence here because we would tidy
+  // the frame up *after* the call, however in the ABI-changing tail-call case
+  // we've carefully laid out the parameters so that when sp is reset they'll be
+  // in the correct location.
+  if (IsTailCall && !IsSibCall) {
+    Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
+                               DAG.getIntPtrConstant(0, true), InFlag);
+    InFlag = Chain.getValue(1);
+  }
+
+  // We produce the following DAG scheme for the actual call instruction:
+  //     (AArch64Call Chain, Callee, reg1, ..., regn, preserveMask, inflag?
+  //
+  // Most arguments aren't going to be used and just keep the values live as
+  // far as LLVM is concerned. It's expected to be selected as simply "bl
+  // callee" (for a direct, non-tail call).
+  std::vector<SDValue> Ops;
+  Ops.push_back(Chain);
+  Ops.push_back(Callee);
+
+  if (IsTailCall) {
+    // Each tail call may have to adjust the stack by a different amount, so
+    // this information must travel along with the operation for eventual
+    // consumption by emitEpilogue.
+    Ops.push_back(DAG.getTargetConstant(FPDiff, MVT::i32));
+  }
+
+  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
+    Ops.push_back(DAG.getRegister(RegsToPass[i].first,
+                                  RegsToPass[i].second.getValueType()));
+
+
+  // Add a register mask operand representing the call-preserved registers. This
+  // is used later in codegen to constrain register-allocation.
+  const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
+  const uint32_t *Mask = TRI->getCallPreservedMask(CallConv);
+  assert(Mask && "Missing call preserved mask for calling convention");
+  Ops.push_back(DAG.getRegisterMask(Mask));
+
+  // If we needed glue, put it in as the last argument.
+  if (InFlag.getNode())
+    Ops.push_back(InFlag);
+
+  SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
+
+  if (IsTailCall) {
+    return DAG.getNode(AArch64ISD::TC_RETURN, dl, NodeTys, &Ops[0], Ops.size());
+  }
+
+  Chain = DAG.getNode(AArch64ISD::Call, dl, NodeTys, &Ops[0], Ops.size());
+  InFlag = Chain.getValue(1);
+
+  // Now we can reclaim the stack, just as well do it before working out where
+  // our return value is.
+  if (!IsSibCall) {
+    uint64_t CalleePopBytes
+      = DoesCalleeRestoreStack(CallConv, TailCallOpt) ? NumBytes : 0;
+
+    Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
+                               DAG.getIntPtrConstant(CalleePopBytes, true),
+                               InFlag);
+    InFlag = Chain.getValue(1);
+  }
+
+  return LowerCallResult(Chain, InFlag, CallConv,
+                         IsVarArg, Ins, dl, DAG, InVals);
+}
+
+SDValue
+AArch64TargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
+                                      CallingConv::ID CallConv, bool IsVarArg,
+                                      const SmallVectorImpl<ISD::InputArg> &Ins,
+                                      DebugLoc dl, SelectionDAG &DAG,
+                                      SmallVectorImpl<SDValue> &InVals) const {
+  // Assign locations to each value returned by this call.
+  SmallVector<CCValAssign, 16> RVLocs;
+  CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(),
+                 getTargetMachine(), RVLocs, *DAG.getContext());
+  CCInfo.AnalyzeCallResult(Ins, CCAssignFnForNode(CallConv));
+
+  for (unsigned i = 0; i != RVLocs.size(); ++i) {
+    CCValAssign VA = RVLocs[i];
+
+    // Return values that are too big to fit into registers should use an sret
+    // pointer, so this can be a lot simpler than the main argument code.
+    assert(VA.isRegLoc() && "Memory locations not expected for call return");
+
+    SDValue Val = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), VA.getLocVT(),
+                                     InFlag);
+    Chain = Val.getValue(1);
+    InFlag = Val.getValue(2);
+
+    switch (VA.getLocInfo()) {
+    default: llvm_unreachable("Unknown loc info!");
+    case CCValAssign::Full: break;
+    case CCValAssign::BCvt:
+      Val = DAG.getNode(ISD::BITCAST, dl, VA.getValVT(), Val);
+      break;
+    case CCValAssign::ZExt:
+    case CCValAssign::SExt:
+    case CCValAssign::AExt:
+      // Floating-point arguments only get extended/truncated if they're going
+      // in memory, so using the integer operation is acceptable here.
+      Val = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), Val);
+      break;
+    }
+
+    InVals.push_back(Val);
+  }
+
+  return Chain;
+}
+
+bool
+AArch64TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
+                                    CallingConv::ID CalleeCC,
+                                    bool IsVarArg,
+                                    bool IsCalleeStructRet,
+                                    bool IsCallerStructRet,
+                                    const SmallVectorImpl<ISD::OutputArg> &Outs,
+                                    const SmallVectorImpl<SDValue> &OutVals,
+                                    const SmallVectorImpl<ISD::InputArg> &Ins,
+                                    SelectionDAG& DAG) const {
+
+  // For CallingConv::C this function knows whether the ABI needs
+  // changing. That's not true for other conventions so they will have to opt in
+  // manually.
+  if (!IsTailCallConvention(CalleeCC) && CalleeCC != CallingConv::C)
+    return false;
+
+  const MachineFunction &MF = DAG.getMachineFunction();
+  const Function *CallerF = MF.getFunction();
+  CallingConv::ID CallerCC = CallerF->getCallingConv();
+  bool CCMatch = CallerCC == CalleeCC;
+
+  // Byval parameters hand the function a pointer directly into the stack area
+  // we want to reuse during a tail call. Working around this *is* possible (see
+  // X86) but less efficient and uglier in LowerCall.
+  for (Function::const_arg_iterator i = CallerF->arg_begin(),
+         e = CallerF->arg_end(); i != e; ++i)
+    if (i->hasByValAttr())
+      return false;
+
+  if (getTargetMachine().Options.GuaranteedTailCallOpt) {
+    if (IsTailCallConvention(CalleeCC) && CCMatch)
+      return true;
+    return false;
+  }
+
+  // Now we search for cases where we can use a tail call without changing the
+  // ABI. Sibcall is used in some places (particularly gcc) to refer to this
+  // concept.
+
+  // I want anyone implementing a new calling convention to think long and hard
+  // about this assert.
+  assert((!IsVarArg || CalleeCC == CallingConv::C)
+         && "Unexpected variadic calling convention");
+
+  if (IsVarArg && !Outs.empty()) {
+    // At least two cases here: if caller is fastcc then we can't have any
+    // memory arguments (we'd be expected to clean up the stack afterwards). If
+    // caller is C then we could potentially use its argument area.
+
+    // FIXME: for now we take the most conservative of these in both cases:
+    // disallow all variadic memory operands.
+    SmallVector<CCValAssign, 16> ArgLocs;
+    CCState CCInfo(CalleeCC, IsVarArg, DAG.getMachineFunction(),
+                   getTargetMachine(), ArgLocs, *DAG.getContext());
+
+    CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForNode(CalleeCC));
+    for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i)
+      if (!ArgLocs[i].isRegLoc())
+        return false;
+  }
+
+  // If the calling conventions do not match, then we'd better make sure the
+  // results are returned in the same way as what the caller expects.
+  if (!CCMatch) {
+    SmallVector<CCValAssign, 16> RVLocs1;
+    CCState CCInfo1(CalleeCC, false, DAG.getMachineFunction(),
+                    getTargetMachine(), RVLocs1, *DAG.getContext());
+    CCInfo1.AnalyzeCallResult(Ins, CCAssignFnForNode(CalleeCC));
+
+    SmallVector<CCValAssign, 16> RVLocs2;
+    CCState CCInfo2(CallerCC, false, DAG.getMachineFunction(),
+                    getTargetMachine(), RVLocs2, *DAG.getContext());
+    CCInfo2.AnalyzeCallResult(Ins, CCAssignFnForNode(CallerCC));
+
+    if (RVLocs1.size() != RVLocs2.size())
+      return false;
+    for (unsigned i = 0, e = RVLocs1.size(); i != e; ++i) {
+      if (RVLocs1[i].isRegLoc() != RVLocs2[i].isRegLoc())
+        return false;
+      if (RVLocs1[i].getLocInfo() != RVLocs2[i].getLocInfo())
+        return false;
+      if (RVLocs1[i].isRegLoc()) {
+        if (RVLocs1[i].getLocReg() != RVLocs2[i].getLocReg())
+          return false;
+      } else {
+        if (RVLocs1[i].getLocMemOffset() != RVLocs2[i].getLocMemOffset())
+          return false;
+      }
+    }
+  }
+
+  // Nothing more to check if the callee is taking no arguments
+  if (Outs.empty())
+    return true;
+
+  SmallVector<CCValAssign, 16> ArgLocs;
+  CCState CCInfo(CalleeCC, IsVarArg, DAG.getMachineFunction(),
+                 getTargetMachine(), ArgLocs, *DAG.getContext());
+
+  CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForNode(CalleeCC));
+
+  const AArch64MachineFunctionInfo *FuncInfo
+    = MF.getInfo<AArch64MachineFunctionInfo>();
+
+  // If the stack arguments for this call would fit into our own save area then
+  // the call can be made tail.
+  return CCInfo.getNextStackOffset() <= FuncInfo->getBytesInStackArgArea();
+}
+
+bool AArch64TargetLowering::DoesCalleeRestoreStack(CallingConv::ID CallCC,
+                                                   bool TailCallOpt) const {
+  return CallCC == CallingConv::Fast && TailCallOpt;
+}
+
+bool AArch64TargetLowering::IsTailCallConvention(CallingConv::ID CallCC) const {
+  return CallCC == CallingConv::Fast;
+}
+
+SDValue AArch64TargetLowering::addTokenForArgument(SDValue Chain,
+                                                   SelectionDAG &DAG,
+                                                   MachineFrameInfo *MFI,
+                                                   int ClobberedFI) const {
+  SmallVector<SDValue, 8> ArgChains;
+  int64_t FirstByte = MFI->getObjectOffset(ClobberedFI);
+  int64_t LastByte = FirstByte + MFI->getObjectSize(ClobberedFI) - 1;
+
+  // Include the original chain at the beginning of the list. When this is
+  // used by target LowerCall hooks, this helps legalize find the
+  // CALLSEQ_BEGIN node.
+  ArgChains.push_back(Chain);
+
+  // Add a chain value for each stack argument corresponding
+  for (SDNode::use_iterator U = DAG.getEntryNode().getNode()->use_begin(),
+         UE = DAG.getEntryNode().getNode()->use_end(); U != UE; ++U)
+    if (LoadSDNode *L = dyn_cast<LoadSDNode>(*U))
+      if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(L->getBasePtr()))
+        if (FI->getIndex() < 0) {
+          int64_t InFirstByte = MFI->getObjectOffset(FI->getIndex());
+          int64_t InLastByte = InFirstByte;
+          InLastByte += MFI->getObjectSize(FI->getIndex()) - 1;
+
+          if ((InFirstByte <= FirstByte && FirstByte <= InLastByte) ||
+              (FirstByte <= InFirstByte && InFirstByte <= LastByte))
+            ArgChains.push_back(SDValue(L, 1));
+        }
+
+   // Build a tokenfactor for all the chains.
+   return DAG.getNode(ISD::TokenFactor, Chain.getDebugLoc(), MVT::Other,
+                      &ArgChains[0], ArgChains.size());
+}
+
+static A64CC::CondCodes IntCCToA64CC(ISD::CondCode CC) {
+  switch (CC) {
+  case ISD::SETEQ:  return A64CC::EQ;
+  case ISD::SETGT:  return A64CC::GT;
+  case ISD::SETGE:  return A64CC::GE;
+  case ISD::SETLT:  return A64CC::LT;
+  case ISD::SETLE:  return A64CC::LE;
+  case ISD::SETNE:  return A64CC::NE;
+  case ISD::SETUGT: return A64CC::HI;
+  case ISD::SETUGE: return A64CC::HS;
+  case ISD::SETULT: return A64CC::LO;
+  case ISD::SETULE: return A64CC::LS;
+  default: llvm_unreachable("Unexpected condition code");
+  }
+}
+
+bool AArch64TargetLowering::isLegalICmpImmediate(int64_t Val) const {
+  // icmp is implemented using adds/subs immediate, which take an unsigned
+  // 12-bit immediate, optionally shifted left by 12 bits.
+
+  // Symmetric by using adds/subs
+  if (Val < 0)
+    Val = -Val;
+
+  return (Val & ~0xfff) == 0 || (Val & ~0xfff000) == 0;
+}
+
+SDValue AArch64TargetLowering::getSelectableIntSetCC(SDValue LHS, SDValue RHS,
+                                        ISD::CondCode CC, SDValue &A64cc,
+                                        SelectionDAG &DAG, DebugLoc &dl) const {
+  if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(RHS.getNode())) {
+    int64_t C = 0;
+    EVT VT = RHSC->getValueType(0);
+    bool knownInvalid = false;
+
+    // I'm not convinced the rest of LLVM handles these edge cases properly, but
+    // we can at least get it right.
+    if (isSignedIntSetCC(CC)) {
+      C = RHSC->getSExtValue();
+    } else if (RHSC->getZExtValue() > INT64_MAX) {
+      // A 64-bit constant not representable by a signed 64-bit integer is far
+      // too big to fit into a SUBS immediate anyway.
+      knownInvalid = true;
+    } else {
+      C = RHSC->getZExtValue();
+    }
+
+    if (!knownInvalid && !isLegalICmpImmediate(C)) {
+      // Constant does not fit, try adjusting it by one?
+      switch (CC) {
+      default: break;
+      case ISD::SETLT:
+      case ISD::SETGE:
+        if (isLegalICmpImmediate(C-1)) {
+          CC = (CC == ISD::SETLT) ? ISD::SETLE : ISD::SETGT;
+          RHS = DAG.getConstant(C-1, VT);
+        }
+        break;
+      case ISD::SETULT:
+      case ISD::SETUGE:
+        if (isLegalICmpImmediate(C-1)) {
+          CC = (CC == ISD::SETULT) ? ISD::SETULE : ISD::SETUGT;
+          RHS = DAG.getConstant(C-1, VT);
+        }
+        break;
+      case ISD::SETLE:
+      case ISD::SETGT:
+        if (isLegalICmpImmediate(C+1)) {
+          CC = (CC == ISD::SETLE) ? ISD::SETLT : ISD::SETGE;
+          RHS = DAG.getConstant(C+1, VT);
+        }
+        break;
+      case ISD::SETULE:
+      case ISD::SETUGT:
+        if (isLegalICmpImmediate(C+1)) {
+          CC = (CC == ISD::SETULE) ? ISD::SETULT : ISD::SETUGE;
+          RHS = DAG.getConstant(C+1, VT);
+        }
+        break;
+      }
+    }
+  }
+
+  A64CC::CondCodes CondCode = IntCCToA64CC(CC);
+  A64cc = DAG.getConstant(CondCode, MVT::i32);
+  return DAG.getNode(AArch64ISD::SETCC, dl, MVT::i32, LHS, RHS,
+                     DAG.getCondCode(CC));
+}
+
+static A64CC::CondCodes FPCCToA64CC(ISD::CondCode CC,
+                                    A64CC::CondCodes &Alternative) {
+  A64CC::CondCodes CondCode = A64CC::Invalid;
+  Alternative = A64CC::Invalid;
+
+  switch (CC) {
+  default: llvm_unreachable("Unknown FP condition!");
+  case ISD::SETEQ:
+  case ISD::SETOEQ: CondCode = A64CC::EQ; break;
+  case ISD::SETGT:
+  case ISD::SETOGT: CondCode = A64CC::GT; break;
+  case ISD::SETGE:
+  case ISD::SETOGE: CondCode = A64CC::GE; break;
+  case ISD::SETOLT: CondCode = A64CC::MI; break;
+  case ISD::SETOLE: CondCode = A64CC::LS; break;
+  case ISD::SETONE: CondCode = A64CC::MI; Alternative = A64CC::GT; break;
+  case ISD::SETO:   CondCode = A64CC::VC; break;
+  case ISD::SETUO:  CondCode = A64CC::VS; break;
+  case ISD::SETUEQ: CondCode = A64CC::EQ; Alternative = A64CC::VS; break;
+  case ISD::SETUGT: CondCode = A64CC::HI; break;
+  case ISD::SETUGE: CondCode = A64CC::PL; break;
+  case ISD::SETLT:
+  case ISD::SETULT: CondCode = A64CC::LT; break;
+  case ISD::SETLE:
+  case ISD::SETULE: CondCode = A64CC::LE; break;
+  case ISD::SETNE:
+  case ISD::SETUNE: CondCode = A64CC::NE; break;
+  }
+  return CondCode;
+}
+
+SDValue
+AArch64TargetLowering::LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT PtrVT = getPointerTy();
+  const BlockAddress *BA = cast<BlockAddressSDNode>(Op)->getBlockAddress();
+
+  assert(getTargetMachine().getCodeModel() == CodeModel::Small
+         && "Only small code model supported at the moment");
+
+  // The most efficient code is PC-relative anyway for the small memory model,
+  // so we don't need to worry about relocation model.
+  return DAG.getNode(AArch64ISD::WrapperSmall, DL, PtrVT,
+                     DAG.getTargetBlockAddress(BA, PtrVT, 0,
+                                               AArch64II::MO_NO_FLAG),
+                     DAG.getTargetBlockAddress(BA, PtrVT, 0,
+                                               AArch64II::MO_LO12),
+                     DAG.getConstant(/*Alignment=*/ 4, MVT::i32));
+}
+
+
+// (BRCOND chain, val, dest)
+SDValue
+AArch64TargetLowering::LowerBRCOND(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc dl = Op.getDebugLoc();
+  SDValue Chain = Op.getOperand(0);
+  SDValue TheBit = Op.getOperand(1);
+  SDValue DestBB = Op.getOperand(2);
+
+  // AArch64 BooleanContents is the default UndefinedBooleanContent, which means
+  // that as the consumer we are responsible for ignoring rubbish in higher
+  // bits.
+  TheBit = DAG.getNode(ISD::AND, dl, MVT::i32, TheBit,
+                       DAG.getConstant(1, MVT::i32));
+
+  SDValue A64CMP = DAG.getNode(AArch64ISD::SETCC, dl, MVT::i32, TheBit,
+                               DAG.getConstant(0, TheBit.getValueType()),
+                               DAG.getCondCode(ISD::SETNE));
+
+  return DAG.getNode(AArch64ISD::BR_CC, dl, MVT::Other, Chain,
+                     A64CMP, DAG.getConstant(A64CC::NE, MVT::i32),
+                     DestBB);
+}
+
+// (BR_CC chain, condcode, lhs, rhs, dest)
+SDValue
+AArch64TargetLowering::LowerBR_CC(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc dl = Op.getDebugLoc();
+  SDValue Chain = Op.getOperand(0);
+  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(1))->get();
+  SDValue LHS = Op.getOperand(2);
+  SDValue RHS = Op.getOperand(3);
+  SDValue DestBB = Op.getOperand(4);
+
+  if (LHS.getValueType() == MVT::f128) {
+    // f128 comparisons are lowered to runtime calls by a routine which sets
+    // LHS, RHS and CC appropriately for the rest of this function to continue.
+    softenSetCCOperands(DAG, MVT::f128, LHS, RHS, CC, dl);
+
+    // If softenSetCCOperands returned a scalar, we need to compare the result
+    // against zero to select between true and false values.
+    if (RHS.getNode() == 0) {
+      RHS = DAG.getConstant(0, LHS.getValueType());
+      CC = ISD::SETNE;
+    }
+  }
+
+  if (LHS.getValueType().isInteger()) {
+    SDValue A64cc;
+
+    // Integers are handled in a separate function because the combinations of
+    // immediates and tests can get hairy and we may want to fiddle things.
+    SDValue CmpOp = getSelectableIntSetCC(LHS, RHS, CC, A64cc, DAG, dl);
+
+    return DAG.getNode(AArch64ISD::BR_CC, dl, MVT::Other,
+                       Chain, CmpOp, A64cc, DestBB);
+  }
+
+  // Note that some LLVM floating-point CondCodes can't be lowered to a single
+  // conditional branch, hence FPCCToA64CC can set a second test, where either
+  // passing is sufficient.
+  A64CC::CondCodes CondCode, Alternative = A64CC::Invalid;
+  CondCode = FPCCToA64CC(CC, Alternative);
+  SDValue A64cc = DAG.getConstant(CondCode, MVT::i32);
+  SDValue SetCC = DAG.getNode(AArch64ISD::SETCC, dl, MVT::i32, LHS, RHS,
+                              DAG.getCondCode(CC));
+  SDValue A64BR_CC = DAG.getNode(AArch64ISD::BR_CC, dl, MVT::Other,
+                                 Chain, SetCC, A64cc, DestBB);
+
+  if (Alternative != A64CC::Invalid) {
+    A64cc = DAG.getConstant(Alternative, MVT::i32);
+    A64BR_CC = DAG.getNode(AArch64ISD::BR_CC, dl, MVT::Other,
+                           A64BR_CC, SetCC, A64cc, DestBB);
+
+  }
+
+  return A64BR_CC;
+}
+
+SDValue
+AArch64TargetLowering::LowerF128ToCall(SDValue Op, SelectionDAG &DAG,
+                                       RTLIB::Libcall Call) const {
+  ArgListTy Args;
+  ArgListEntry Entry;
+  for (unsigned i = 0, e = Op->getNumOperands(); i != e; ++i) {
+    EVT ArgVT = Op.getOperand(i).getValueType();
+    Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
+    Entry.Node = Op.getOperand(i); Entry.Ty = ArgTy;
+    Entry.isSExt = false;
+    Entry.isZExt = false;
+    Args.push_back(Entry);
+  }
+  SDValue Callee = DAG.getExternalSymbol(getLibcallName(Call), getPointerTy());
+
+  Type *RetTy = Op.getValueType().getTypeForEVT(*DAG.getContext());
+
+  // By default, the input chain to this libcall is the entry node of the
+  // function. If the libcall is going to be emitted as a tail call then
+  // isUsedByReturnOnly will change it to the right chain if the return
+  // node which is being folded has a non-entry input chain.
+  SDValue InChain = DAG.getEntryNode();
+
+  // isTailCall may be true since the callee does not reference caller stack
+  // frame. Check if it's in the right position.
+  SDValue TCChain = InChain;
+  bool isTailCall = isInTailCallPosition(DAG, Op.getNode(), TCChain);
+  if (isTailCall)
+    InChain = TCChain;
+
+  TargetLowering::
+  CallLoweringInfo CLI(InChain, RetTy, false, false, false, false,
+                    0, getLibcallCallingConv(Call), isTailCall,
+                    /*doesNotReturn=*/false, /*isReturnValueUsed=*/true,
+                    Callee, Args, DAG, Op->getDebugLoc());
+  std::pair<SDValue, SDValue> CallInfo = LowerCallTo(CLI);
+
+  if (!CallInfo.second.getNode())
+    // It's a tailcall, return the chain (which is the DAG root).
+    return DAG.getRoot();
+
+  return CallInfo.first;
+}
+
+SDValue
+AArch64TargetLowering::LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const {
+  if (Op.getOperand(0).getValueType() != MVT::f128) {
+    // It's legal except when f128 is involved
+    return Op;
+  }
+
+  RTLIB::Libcall LC;
+  LC  = RTLIB::getFPROUND(Op.getOperand(0).getValueType(), Op.getValueType());
+
+  SDValue SrcVal = Op.getOperand(0);
+  return makeLibCall(DAG, LC, Op.getValueType(), &SrcVal, 1,
+                     /*isSigned*/ false, Op.getDebugLoc());
+}
+
+SDValue
+AArch64TargetLowering::LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const {
+  assert(Op.getValueType() == MVT::f128 && "Unexpected lowering");
+
+  RTLIB::Libcall LC;
+  LC  = RTLIB::getFPEXT(Op.getOperand(0).getValueType(), Op.getValueType());
+
+  return LowerF128ToCall(Op, DAG, LC);
+}
+
+SDValue
+AArch64TargetLowering::LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG,
+                                      bool IsSigned) const {
+  if (Op.getOperand(0).getValueType() != MVT::f128) {
+    // It's legal except when f128 is involved
+    return Op;
+  }
+
+  RTLIB::Libcall LC;
+  if (IsSigned)
+    LC = RTLIB::getFPTOSINT(Op.getOperand(0).getValueType(), Op.getValueType());
+  else
+    LC = RTLIB::getFPTOUINT(Op.getOperand(0).getValueType(), Op.getValueType());
+
+  return LowerF128ToCall(Op, DAG, LC);
+}
+
+SDValue
+AArch64TargetLowering::LowerGlobalAddressELF(SDValue Op,
+                                             SelectionDAG &DAG) const {
+  // TableGen doesn't have easy access to the CodeModel or RelocationModel, so
+  // we make that distinction here.
+
+  // We support the small memory model for now.
+  assert(getTargetMachine().getCodeModel() == CodeModel::Small);
+
+  EVT PtrVT = getPointerTy();
+  DebugLoc dl = Op.getDebugLoc();
+  const GlobalAddressSDNode *GN = cast<GlobalAddressSDNode>(Op);
+  const GlobalValue *GV = GN->getGlobal();
+  unsigned Alignment = GV->getAlignment();
+  Reloc::Model RelocM = getTargetMachine().getRelocationModel();
+  if (GV->isWeakForLinker() && GV->isDeclaration() && RelocM == Reloc::Static) {
+    // Weak undefined symbols can't use ADRP/ADD pair since they should evaluate
+    // to zero when they remain undefined. In PIC mode the GOT can take care of
+    // this, but in absolute mode we use a constant pool load.
+    SDValue PoolAddr;
+    PoolAddr = DAG.getNode(AArch64ISD::WrapperSmall, dl, PtrVT,
+                           DAG.getTargetConstantPool(GV, PtrVT, 0, 0,
+                                                     AArch64II::MO_NO_FLAG),
+                           DAG.getTargetConstantPool(GV, PtrVT, 0, 0,
+                                                     AArch64II::MO_LO12),
+                           DAG.getConstant(8, MVT::i32));
+    SDValue GlobalAddr = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), PoolAddr,
+                                     MachinePointerInfo::getConstantPool(),
+                                     /*isVolatile=*/ false,
+                                     /*isNonTemporal=*/ true,
+                                     /*isInvariant=*/ true, 8);
+    if (GN->getOffset() != 0)
+      return DAG.getNode(ISD::ADD, dl, PtrVT, GlobalAddr,
+                         DAG.getConstant(GN->getOffset(), PtrVT));
+
+    return GlobalAddr;
+  }
+
+  if (Alignment == 0) {
+    const PointerType *GVPtrTy = cast<PointerType>(GV->getType());
+    if (GVPtrTy->getElementType()->isSized()) {
+      Alignment
+        = getDataLayout()->getABITypeAlignment(GVPtrTy->getElementType());
+    } else {
+      // Be conservative if we can't guess, not that it really matters:
+      // functions and labels aren't valid for loads, and the methods used to
+      // actually calculate an address work with any alignment.
+      Alignment = 1;
+    }
+  }
+
+  unsigned char HiFixup, LoFixup;
+  bool UseGOT = Subtarget->GVIsIndirectSymbol(GV, RelocM);
+
+  if (UseGOT) {
+    HiFixup = AArch64II::MO_GOT;
+    LoFixup = AArch64II::MO_GOT_LO12;
+    Alignment = 8;
+  } else {
+    HiFixup = AArch64II::MO_NO_FLAG;
+    LoFixup = AArch64II::MO_LO12;
+  }
+
+  // AArch64's small model demands the following sequence:
+  // ADRP x0, somewhere
+  // ADD x0, x0, #:lo12:somewhere ; (or LDR directly).
+  SDValue GlobalRef = DAG.getNode(AArch64ISD::WrapperSmall, dl, PtrVT,
+                                  DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+                                                             HiFixup),
+                                  DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+                                                             LoFixup),
+                                  DAG.getConstant(Alignment, MVT::i32));
+
+  if (UseGOT) {
+    GlobalRef = DAG.getNode(AArch64ISD::GOTLoad, dl, PtrVT, DAG.getEntryNode(),
+                            GlobalRef);
+  }
+
+  if (GN->getOffset() != 0)
+    return DAG.getNode(ISD::ADD, dl, PtrVT, GlobalRef,
+                       DAG.getConstant(GN->getOffset(), PtrVT));
+
+  return GlobalRef;
+}
+
+SDValue AArch64TargetLowering::LowerTLSDescCall(SDValue SymAddr,
+                                                SDValue DescAddr,
+                                                DebugLoc DL,
+                                                SelectionDAG &DAG) const {
+  EVT PtrVT = getPointerTy();
+
+  // The function we need to call is simply the first entry in the GOT for this
+  // descriptor, load it in preparation.
+  SDValue Func, Chain;
+  Func = DAG.getNode(AArch64ISD::GOTLoad, DL, PtrVT, DAG.getEntryNode(),
+                     DescAddr);
+
+  // The function takes only one argument: the address of the descriptor itself
+  // in X0.
+  SDValue Glue;
+  Chain = DAG.getCopyToReg(DAG.getEntryNode(), DL, AArch64::X0, DescAddr, Glue);
+  Glue = Chain.getValue(1);
+
+  // Finally, there's a special calling-convention which means that the lookup
+  // must preserve all registers (except X0, obviously).
+  const TargetRegisterInfo *TRI  = getTargetMachine().getRegisterInfo();
+  const AArch64RegisterInfo *A64RI
+    = static_cast<const AArch64RegisterInfo *>(TRI);
+  const uint32_t *Mask = A64RI->getTLSDescCallPreservedMask();
+
+  // We're now ready to populate the argument list, as with a normal call:
+  std::vector<SDValue> Ops;
+  Ops.push_back(Chain);
+  Ops.push_back(Func);
+  Ops.push_back(SymAddr);
+  Ops.push_back(DAG.getRegister(AArch64::X0, PtrVT));
+  Ops.push_back(DAG.getRegisterMask(Mask));
+  Ops.push_back(Glue);
+
+  SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
+  Chain = DAG.getNode(AArch64ISD::TLSDESCCALL, DL, NodeTys, &Ops[0],
+                      Ops.size());
+  Glue = Chain.getValue(1);
+
+  // After the call, the offset from TPIDR_EL0 is in X0, copy it out and pass it
+  // back to the generic handling code.
+  return DAG.getCopyFromReg(Chain, DL, AArch64::X0, PtrVT, Glue);
+}
+
+SDValue
+AArch64TargetLowering::LowerGlobalTLSAddress(SDValue Op,
+                                             SelectionDAG &DAG) const {
+  assert(Subtarget->isTargetELF() &&
+         "TLS not implemented for non-ELF targets");
+  const GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);
+
+  TLSModel::Model Model = getTargetMachine().getTLSModel(GA->getGlobal());
+
+  SDValue TPOff;
+  EVT PtrVT = getPointerTy();
+  DebugLoc DL = Op.getDebugLoc();
+  const GlobalValue *GV = GA->getGlobal();
+
+  SDValue ThreadBase = DAG.getNode(AArch64ISD::THREAD_POINTER, DL, PtrVT);
+
+  if (Model == TLSModel::InitialExec) {
+    TPOff = DAG.getNode(AArch64ISD::WrapperSmall, DL, PtrVT,
+                        DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
+                                                   AArch64II::MO_GOTTPREL),
+                        DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
+                                                   AArch64II::MO_GOTTPREL_LO12),
+                        DAG.getConstant(8, MVT::i32));
+    TPOff = DAG.getNode(AArch64ISD::GOTLoad, DL, PtrVT, DAG.getEntryNode(),
+                        TPOff);
+  } else if (Model == TLSModel::LocalExec) {
+    SDValue HiVar = DAG.getTargetGlobalAddress(GV, DL, MVT::i64, 0,
+                                               AArch64II::MO_TPREL_G1);
+    SDValue LoVar = DAG.getTargetGlobalAddress(GV, DL, MVT::i64, 0,
+                                               AArch64II::MO_TPREL_G0_NC);
+
+    TPOff = SDValue(DAG.getMachineNode(AArch64::MOVZxii, DL, PtrVT, HiVar,
+                                       DAG.getTargetConstant(0, MVT::i32)), 0);
+    TPOff = SDValue(DAG.getMachineNode(AArch64::MOVKxii, DL, PtrVT,
+                                       TPOff, LoVar,
+                                       DAG.getTargetConstant(0, MVT::i32)), 0);
+  } else if (Model == TLSModel::GeneralDynamic) {
+    // Accesses used in this sequence go via the TLS descriptor which lives in
+    // the GOT. Prepare an address we can use to handle this.
+    SDValue HiDesc = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
+                                                AArch64II::MO_TLSDESC);
+    SDValue LoDesc = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
+                                                AArch64II::MO_TLSDESC_LO12);
+    SDValue DescAddr = DAG.getNode(AArch64ISD::WrapperSmall, DL, PtrVT,
+                                   HiDesc, LoDesc,
+                                   DAG.getConstant(8, MVT::i32));
+    SDValue SymAddr = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0);
+
+    TPOff = LowerTLSDescCall(SymAddr, DescAddr, DL, DAG);
+  } else if (Model == TLSModel::LocalDynamic) {
+    // Local-dynamic accesses proceed in two phases. A general-dynamic TLS
+    // descriptor call against the special symbol _TLS_MODULE_BASE_ to calculate
+    // the beginning of the module's TLS region, followed by a DTPREL offset
+    // calculation.
+
+    // These accesses will need deduplicating if there's more than one.
+    AArch64MachineFunctionInfo* MFI = DAG.getMachineFunction()
+      .getInfo<AArch64MachineFunctionInfo>();
+    MFI->incNumLocalDynamicTLSAccesses();
+
+
+    // Get the location of _TLS_MODULE_BASE_:
+    SDValue HiDesc = DAG.getTargetExternalSymbol("_TLS_MODULE_BASE_", PtrVT,
+                                                AArch64II::MO_TLSDESC);
+    SDValue LoDesc = DAG.getTargetExternalSymbol("_TLS_MODULE_BASE_", PtrVT,
+                                                AArch64II::MO_TLSDESC_LO12);
+    SDValue DescAddr = DAG.getNode(AArch64ISD::WrapperSmall, DL, PtrVT,
+                                   HiDesc, LoDesc,
+                                   DAG.getConstant(8, MVT::i32));
+    SDValue SymAddr = DAG.getTargetExternalSymbol("_TLS_MODULE_BASE_", PtrVT);
+
+    ThreadBase = LowerTLSDescCall(SymAddr, DescAddr, DL, DAG);
+
+    // Get the variable's offset from _TLS_MODULE_BASE_
+    SDValue HiVar = DAG.getTargetGlobalAddress(GV, DL, MVT::i64, 0,
+                                               AArch64II::MO_DTPREL_G1);
+    SDValue LoVar = DAG.getTargetGlobalAddress(GV, DL, MVT::i64, 0,
+                                               AArch64II::MO_DTPREL_G0_NC);
+
+    TPOff = SDValue(DAG.getMachineNode(AArch64::MOVZxii, DL, PtrVT, HiVar,
+                                       DAG.getTargetConstant(0, MVT::i32)), 0);
+    TPOff = SDValue(DAG.getMachineNode(AArch64::MOVKxii, DL, PtrVT,
+                                       TPOff, LoVar,
+                                       DAG.getTargetConstant(0, MVT::i32)), 0);
+  } else
+      llvm_unreachable("Unsupported TLS access model");
+
+
+  return DAG.getNode(ISD::ADD, DL, PtrVT, ThreadBase, TPOff);
+}
+
+SDValue
+AArch64TargetLowering::LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG,
+                                      bool IsSigned) const {
+  if (Op.getValueType() != MVT::f128) {
+    // Legal for everything except f128.
+    return Op;
+  }
+
+  RTLIB::Libcall LC;
+  if (IsSigned)
+    LC = RTLIB::getSINTTOFP(Op.getOperand(0).getValueType(), Op.getValueType());
+  else
+    LC = RTLIB::getUINTTOFP(Op.getOperand(0).getValueType(), Op.getValueType());
+
+  return LowerF128ToCall(Op, DAG, LC);
+}
+
+
+SDValue
+AArch64TargetLowering::LowerJumpTable(SDValue Op, SelectionDAG &DAG) const {
+  JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
+  DebugLoc dl = JT->getDebugLoc();
+
+  // When compiling PIC, jump tables get put in the code section so a static
+  // relocation-style is acceptable for both cases.
+  return DAG.getNode(AArch64ISD::WrapperSmall, dl, getPointerTy(),
+                     DAG.getTargetJumpTable(JT->getIndex(), getPointerTy()),
+                     DAG.getTargetJumpTable(JT->getIndex(), getPointerTy(),
+                                            AArch64II::MO_LO12),
+                     DAG.getConstant(1, MVT::i32));
+}
+
+// (SELECT_CC lhs, rhs, iftrue, iffalse, condcode)
+SDValue
+AArch64TargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc dl = Op.getDebugLoc();
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  SDValue IfTrue = Op.getOperand(2);
+  SDValue IfFalse = Op.getOperand(3);
+  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
+
+  if (LHS.getValueType() == MVT::f128) {
+    // f128 comparisons are lowered to libcalls, but slot in nicely here
+    // afterwards.
+    softenSetCCOperands(DAG, MVT::f128, LHS, RHS, CC, dl);
+
+    // If softenSetCCOperands returned a scalar, we need to compare the result
+    // against zero to select between true and false values.
+    if (RHS.getNode() == 0) {
+      RHS = DAG.getConstant(0, LHS.getValueType());
+      CC = ISD::SETNE;
+    }
+  }
+
+  if (LHS.getValueType().isInteger()) {
+    SDValue A64cc;
+
+    // Integers are handled in a separate function because the combinations of
+    // immediates and tests can get hairy and we may want to fiddle things.
+    SDValue CmpOp = getSelectableIntSetCC(LHS, RHS, CC, A64cc, DAG, dl);
+
+    return DAG.getNode(AArch64ISD::SELECT_CC, dl, Op.getValueType(),
+                       CmpOp, IfTrue, IfFalse, A64cc);
+  }
+
+  // Note that some LLVM floating-point CondCodes can't be lowered to a single
+  // conditional branch, hence FPCCToA64CC can set a second test, where either
+  // passing is sufficient.
+  A64CC::CondCodes CondCode, Alternative = A64CC::Invalid;
+  CondCode = FPCCToA64CC(CC, Alternative);
+  SDValue A64cc = DAG.getConstant(CondCode, MVT::i32);
+  SDValue SetCC = DAG.getNode(AArch64ISD::SETCC, dl, MVT::i32, LHS, RHS,
+                              DAG.getCondCode(CC));
+  SDValue A64SELECT_CC = DAG.getNode(AArch64ISD::SELECT_CC, dl,
+                                     Op.getValueType(),
+                                     SetCC, IfTrue, IfFalse, A64cc);
+
+  if (Alternative != A64CC::Invalid) {
+    A64cc = DAG.getConstant(Alternative, MVT::i32);
+    A64SELECT_CC = DAG.getNode(AArch64ISD::SELECT_CC, dl, Op.getValueType(),
+                               SetCC, IfTrue, A64SELECT_CC, A64cc);
+
+  }
+
+  return A64SELECT_CC;
+}
+
+// (SELECT testbit, iftrue, iffalse)
+SDValue
+AArch64TargetLowering::LowerSELECT(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc dl = Op.getDebugLoc();
+  SDValue TheBit = Op.getOperand(0);
+  SDValue IfTrue = Op.getOperand(1);
+  SDValue IfFalse = Op.getOperand(2);
+
+  // AArch64 BooleanContents is the default UndefinedBooleanContent, which means
+  // that as the consumer we are responsible for ignoring rubbish in higher
+  // bits.
+  TheBit = DAG.getNode(ISD::AND, dl, MVT::i32, TheBit,
+                       DAG.getConstant(1, MVT::i32));
+  SDValue A64CMP = DAG.getNode(AArch64ISD::SETCC, dl, MVT::i32, TheBit,
+                               DAG.getConstant(0, TheBit.getValueType()),
+                               DAG.getCondCode(ISD::SETNE));
+
+  return DAG.getNode(AArch64ISD::SELECT_CC, dl, Op.getValueType(),
+                     A64CMP, IfTrue, IfFalse,
+                     DAG.getConstant(A64CC::NE, MVT::i32));
+}
+
+// (SETCC lhs, rhs, condcode)
+SDValue
+AArch64TargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc dl = Op.getDebugLoc();
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
+  EVT VT = Op.getValueType();
+
+  if (LHS.getValueType() == MVT::f128) {
+    // f128 comparisons will be lowered to libcalls giving a valid LHS and RHS
+    // for the rest of the function (some i32 or i64 values).
+    softenSetCCOperands(DAG, MVT::f128, LHS, RHS, CC, dl);
+
+    // If softenSetCCOperands returned a scalar, use it.
+    if (RHS.getNode() == 0) {
+      assert(LHS.getValueType() == Op.getValueType() &&
+             "Unexpected setcc expansion!");
+      return LHS;
+    }
+  }
+
+  if (LHS.getValueType().isInteger()) {
+    SDValue A64cc;
+
+    // Integers are handled in a separate function because the combinations of
+    // immediates and tests can get hairy and we may want to fiddle things.
+    SDValue CmpOp = getSelectableIntSetCC(LHS, RHS, CC, A64cc, DAG, dl);
+
+    return DAG.getNode(AArch64ISD::SELECT_CC, dl, VT,
+                       CmpOp, DAG.getConstant(1, VT), DAG.getConstant(0, VT),
+                       A64cc);
+  }
+
+  // Note that some LLVM floating-point CondCodes can't be lowered to a single
+  // conditional branch, hence FPCCToA64CC can set a second test, where either
+  // passing is sufficient.
+  A64CC::CondCodes CondCode, Alternative = A64CC::Invalid;
+  CondCode = FPCCToA64CC(CC, Alternative);
+  SDValue A64cc = DAG.getConstant(CondCode, MVT::i32);
+  SDValue CmpOp = DAG.getNode(AArch64ISD::SETCC, dl, MVT::i32, LHS, RHS,
+                              DAG.getCondCode(CC));
+  SDValue A64SELECT_CC = DAG.getNode(AArch64ISD::SELECT_CC, dl, VT,
+                                     CmpOp, DAG.getConstant(1, VT),
+                                     DAG.getConstant(0, VT), A64cc);
+
+  if (Alternative != A64CC::Invalid) {
+    A64cc = DAG.getConstant(Alternative, MVT::i32);
+    A64SELECT_CC = DAG.getNode(AArch64ISD::SELECT_CC, dl, VT, CmpOp,
+                               DAG.getConstant(1, VT), A64SELECT_CC, A64cc);
+  }
+
+  return A64SELECT_CC;
+}
+
+SDValue
+AArch64TargetLowering::LowerVACOPY(SDValue Op, SelectionDAG &DAG) const {
+  const Value *DestSV = cast<SrcValueSDNode>(Op.getOperand(3))->getValue();
+  const Value *SrcSV = cast<SrcValueSDNode>(Op.getOperand(3))->getValue();
+
+  // We have to make sure we copy the entire structure: 8+8+8+4+4 = 32 bytes
+  // rather than just 8.
+  return DAG.getMemcpy(Op.getOperand(0), Op.getDebugLoc(),
+                       Op.getOperand(1), Op.getOperand(2),
+                       DAG.getConstant(32, MVT::i32), 8, false, false,
+                       MachinePointerInfo(DestSV), MachinePointerInfo(SrcSV));
+}
+
+SDValue
+AArch64TargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG) const {
+  // The layout of the va_list struct is specified in the AArch64 Procedure Call
+  // Standard, section B.3.
+  MachineFunction &MF = DAG.getMachineFunction();
+  AArch64MachineFunctionInfo *FuncInfo
+    = MF.getInfo<AArch64MachineFunctionInfo>();
+  DebugLoc DL = Op.getDebugLoc();
+
+  SDValue Chain = Op.getOperand(0);
+  SDValue VAList = Op.getOperand(1);
+  const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
+  SmallVector<SDValue, 4> MemOps;
+
+  // void *__stack at offset 0
+  SDValue Stack = DAG.getFrameIndex(FuncInfo->getVariadicStackIdx(),
+                                    getPointerTy());
+  MemOps.push_back(DAG.getStore(Chain, DL, Stack, VAList,
+                                MachinePointerInfo(SV), false, false, 0));
+
+  // void *__gr_top at offset 8
+  int GPRSize = FuncInfo->getVariadicGPRSize();
+  if (GPRSize > 0) {
+    SDValue GRTop, GRTopAddr;
+
+    GRTopAddr = DAG.getNode(ISD::ADD, DL, getPointerTy(), VAList,
+                            DAG.getConstant(8, getPointerTy()));
+
+    GRTop = DAG.getFrameIndex(FuncInfo->getVariadicGPRIdx(), getPointerTy());
+    GRTop = DAG.getNode(ISD::ADD, DL, getPointerTy(), GRTop,
+                        DAG.getConstant(GPRSize, getPointerTy()));
+
+    MemOps.push_back(DAG.getStore(Chain, DL, GRTop, GRTopAddr,
+                                  MachinePointerInfo(SV, 8),
+                                  false, false, 0));
+  }
+
+  // void *__vr_top at offset 16
+  int FPRSize = FuncInfo->getVariadicFPRSize();
+  if (FPRSize > 0) {
+    SDValue VRTop, VRTopAddr;
+    VRTopAddr = DAG.getNode(ISD::ADD, DL, getPointerTy(), VAList,
+                            DAG.getConstant(16, getPointerTy()));
+
+    VRTop = DAG.getFrameIndex(FuncInfo->getVariadicFPRIdx(), getPointerTy());
+    VRTop = DAG.getNode(ISD::ADD, DL, getPointerTy(), VRTop,
+                        DAG.getConstant(FPRSize, getPointerTy()));
+
+    MemOps.push_back(DAG.getStore(Chain, DL, VRTop, VRTopAddr,
+                                  MachinePointerInfo(SV, 16),
+                                  false, false, 0));
+  }
+
+  // int __gr_offs at offset 24
+  SDValue GROffsAddr = DAG.getNode(ISD::ADD, DL, getPointerTy(), VAList,
+                                   DAG.getConstant(24, getPointerTy()));
+  MemOps.push_back(DAG.getStore(Chain, DL, DAG.getConstant(-GPRSize, MVT::i32),
+                                GROffsAddr, MachinePointerInfo(SV, 24),
+                                false, false, 0));
+
+  // int __vr_offs at offset 28
+  SDValue VROffsAddr = DAG.getNode(ISD::ADD, DL, getPointerTy(), VAList,
+                                   DAG.getConstant(28, getPointerTy()));
+  MemOps.push_back(DAG.getStore(Chain, DL, DAG.getConstant(-FPRSize, MVT::i32),
+                                VROffsAddr, MachinePointerInfo(SV, 28),
+                                false, false, 0));
+
+  return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, &MemOps[0],
+                     MemOps.size());
+}
+
+SDValue
+AArch64TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
+  switch (Op.getOpcode()) {
+  default: llvm_unreachable("Don't know how to custom lower this!");
+  case ISD::FADD: return LowerF128ToCall(Op, DAG, RTLIB::ADD_F128);
+  case ISD::FSUB: return LowerF128ToCall(Op, DAG, RTLIB::SUB_F128);
+  case ISD::FMUL: return LowerF128ToCall(Op, DAG, RTLIB::MUL_F128);
+  case ISD::FDIV: return LowerF128ToCall(Op, DAG, RTLIB::DIV_F128);
+  case ISD::FP_TO_SINT: return LowerFP_TO_INT(Op, DAG, true);
+  case ISD::FP_TO_UINT: return LowerFP_TO_INT(Op, DAG, false);
+  case ISD::SINT_TO_FP: return LowerINT_TO_FP(Op, DAG, true);
+  case ISD::UINT_TO_FP: return LowerINT_TO_FP(Op, DAG, false);
+  case ISD::FP_ROUND: return LowerFP_ROUND(Op, DAG);
+  case ISD::FP_EXTEND: return LowerFP_EXTEND(Op, DAG);
+
+  case ISD::BlockAddress: return LowerBlockAddress(Op, DAG);
+  case ISD::BRCOND: return LowerBRCOND(Op, DAG);
+  case ISD::BR_CC: return LowerBR_CC(Op, DAG);
+  case ISD::GlobalAddress: return LowerGlobalAddressELF(Op, DAG);
+  case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
+  case ISD::JumpTable: return LowerJumpTable(Op, DAG);
+  case ISD::SELECT: return LowerSELECT(Op, DAG);
+  case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
+  case ISD::SETCC: return LowerSETCC(Op, DAG);
+  case ISD::VACOPY: return LowerVACOPY(Op, DAG);
+  case ISD::VASTART: return LowerVASTART(Op, DAG);
+  }
+
+  return SDValue();
+}
+
+static SDValue PerformANDCombine(SDNode *N,
+                                 TargetLowering::DAGCombinerInfo &DCI) {
+
+  SelectionDAG &DAG = DCI.DAG;
+  DebugLoc DL = N->getDebugLoc();
+  EVT VT = N->getValueType(0);
+
+  // We're looking for an SRA/SHL pair which form an SBFX.
+
+  if (VT != MVT::i32 && VT != MVT::i64)
+    return SDValue();
+
+  if (!isa<ConstantSDNode>(N->getOperand(1)))
+    return SDValue();
+
+  uint64_t TruncMask = N->getConstantOperandVal(1);
+  if (!isMask_64(TruncMask))
+    return SDValue();
+
+  uint64_t Width = CountPopulation_64(TruncMask);
+  SDValue Shift = N->getOperand(0);
+
+  if (Shift.getOpcode() != ISD::SRL)
+    return SDValue();
+
+  if (!isa<ConstantSDNode>(Shift->getOperand(1)))
+    return SDValue();
+  uint64_t LSB = Shift->getConstantOperandVal(1);
+
+  if (LSB > VT.getSizeInBits() || Width > VT.getSizeInBits())
+    return SDValue();
+
+  return DAG.getNode(AArch64ISD::UBFX, DL, VT, Shift.getOperand(0),
+                     DAG.getConstant(LSB, MVT::i64),
+                     DAG.getConstant(LSB + Width - 1, MVT::i64));
+}
+
+static SDValue PerformATOMIC_FENCECombine(SDNode *FenceNode,
+                                         TargetLowering::DAGCombinerInfo &DCI) {
+  // An atomic operation followed by an acquiring atomic fence can be reduced to
+  // an acquiring load. The atomic operation provides a convenient pointer to
+  // load from. If the original operation was a load anyway we can actually
+  // combine the two operations into an acquiring load.
+  SelectionDAG &DAG = DCI.DAG;
+  SDValue AtomicOp = FenceNode->getOperand(0);
+  AtomicSDNode *AtomicNode = dyn_cast<AtomicSDNode>(AtomicOp);
+
+  // A fence on its own can't be optimised
+  if (!AtomicNode)
+    return SDValue();
+
+  AtomicOrdering FenceOrder
+    = static_cast<AtomicOrdering>(FenceNode->getConstantOperandVal(1));
+  SynchronizationScope FenceScope
+    = static_cast<SynchronizationScope>(FenceNode->getConstantOperandVal(2));
+
+  if (FenceOrder != Acquire || FenceScope != AtomicNode->getSynchScope())
+    return SDValue();
+
+  // If the original operation was an ATOMIC_LOAD then we'll be replacing it, so
+  // the chain we use should be its input, otherwise we'll put our store after
+  // it so we use its output chain.
+  SDValue Chain = AtomicNode->getOpcode() == ISD::ATOMIC_LOAD ?
+    AtomicNode->getChain() : AtomicOp;
+
+  // We have an acquire fence with a handy atomic operation nearby, we can
+  // convert the fence into a load-acquire, discarding the result.
+  DebugLoc DL = FenceNode->getDebugLoc();
+  SDValue Op = DAG.getAtomic(ISD::ATOMIC_LOAD, DL, AtomicNode->getMemoryVT(),
+                             AtomicNode->getValueType(0),
+                             Chain,                  // Chain
+                             AtomicOp.getOperand(1), // Pointer
+                             AtomicNode->getMemOperand(), Acquire,
+                             FenceScope);
+
+  if (AtomicNode->getOpcode() == ISD::ATOMIC_LOAD)
+    DAG.ReplaceAllUsesWith(AtomicNode, Op.getNode());
+
+  return Op.getValue(1);
+}
+
+static SDValue PerformATOMIC_STORECombine(SDNode *N,
+                                         TargetLowering::DAGCombinerInfo &DCI) {
+  // A releasing atomic fence followed by an atomic store can be combined into a
+  // single store operation.
+  SelectionDAG &DAG = DCI.DAG;
+  AtomicSDNode *AtomicNode = cast<AtomicSDNode>(N);
+  SDValue FenceOp = AtomicNode->getOperand(0);
+
+  if (FenceOp.getOpcode() != ISD::ATOMIC_FENCE)
+    return SDValue();
+
+  AtomicOrdering FenceOrder
+    = static_cast<AtomicOrdering>(FenceOp->getConstantOperandVal(1));
+  SynchronizationScope FenceScope
+    = static_cast<SynchronizationScope>(FenceOp->getConstantOperandVal(2));
+
+  if (FenceOrder != Release || FenceScope != AtomicNode->getSynchScope())
+    return SDValue();
+
+  DebugLoc DL = AtomicNode->getDebugLoc();
+  return DAG.getAtomic(ISD::ATOMIC_STORE, DL, AtomicNode->getMemoryVT(),
+                       FenceOp.getOperand(0),  // Chain
+                       AtomicNode->getOperand(1),       // Pointer
+                       AtomicNode->getOperand(2),       // Value
+                       AtomicNode->getMemOperand(), Release,
+                       FenceScope);
+}
+
+/// For a true bitfield insert, the bits getting into that contiguous mask
+/// should come from the low part of an existing value: they must be formed from
+/// a compatible SHL operation (unless they're already low). This function
+/// checks that condition and returns the least-significant bit that's
+/// intended. If the operation not a field preparation, -1 is returned.
+static int32_t getLSBForBFI(SelectionDAG &DAG, DebugLoc DL, EVT VT,
+                            SDValue &MaskedVal, uint64_t Mask) {
+  if (!isShiftedMask_64(Mask))
+    return -1;
+
+  // Now we need to alter MaskedVal so that it is an appropriate input for a BFI
+  // instruction. BFI will do a left-shift by LSB before applying the mask we've
+  // spotted, so in general we should pre-emptively "undo" that by making sure
+  // the incoming bits have had a right-shift applied to them.
+  //
+  // This right shift, however, will combine with existing left/right shifts. In
+  // the simplest case of a completely straight bitfield operation, it will be
+  // expected to completely cancel out with an existing SHL. More complicated
+  // cases (e.g. bitfield to bitfield copy) may still need a real shift before
+  // the BFI.
+
+  uint64_t LSB = CountTrailingZeros_64(Mask);
+  int64_t ShiftRightRequired = LSB;
+  if (MaskedVal.getOpcode() == ISD::SHL &&
+      isa<ConstantSDNode>(MaskedVal.getOperand(1))) {
+    ShiftRightRequired -= MaskedVal.getConstantOperandVal(1);
+    MaskedVal = MaskedVal.getOperand(0);
+  } else if (MaskedVal.getOpcode() == ISD::SRL &&
+             isa<ConstantSDNode>(MaskedVal.getOperand(1))) {
+    ShiftRightRequired += MaskedVal.getConstantOperandVal(1);
+    MaskedVal = MaskedVal.getOperand(0);
+  }
+
+  if (ShiftRightRequired > 0)
+    MaskedVal = DAG.getNode(ISD::SRL, DL, VT, MaskedVal,
+                            DAG.getConstant(ShiftRightRequired, MVT::i64));
+  else if (ShiftRightRequired < 0) {
+    // We could actually end up with a residual left shift, for example with
+    // "struc.bitfield = val << 1".
+    MaskedVal = DAG.getNode(ISD::SHL, DL, VT, MaskedVal,
+                            DAG.getConstant(-ShiftRightRequired, MVT::i64));
+  }
+
+  return LSB;
+}
+
+/// Searches from N for an existing AArch64ISD::BFI node, possibly surrounded by
+/// a mask and an extension. Returns true if a BFI was found and provides
+/// information on its surroundings.
+static bool findMaskedBFI(SDValue N, SDValue &BFI, uint64_t &Mask,
+                          bool &Extended) {
+  Extended = false;
+  if (N.getOpcode() == ISD::ZERO_EXTEND) {
+    Extended = true;
+    N = N.getOperand(0);
+  }
+
+  if (N.getOpcode() == ISD::AND && isa<ConstantSDNode>(N.getOperand(1))) {
+    Mask = N->getConstantOperandVal(1);
+    N = N.getOperand(0);
+  } else {
+    // Mask is the whole width.
+    Mask = -1ULL >> (64 - N.getValueType().getSizeInBits());
+  }
+
+  if (N.getOpcode() == AArch64ISD::BFI) {
+    BFI = N;
+    return true;
+  }
+
+  return false;
+}
+
+/// Try to combine a subtree (rooted at an OR) into a "masked BFI" node, which
+/// is roughly equivalent to (and (BFI ...), mask). This form is used because it
+/// can often be further combined with a larger mask. Ultimately, we want mask
+/// to be 2^32-1 or 2^64-1 so the AND can be skipped.
+static SDValue tryCombineToBFI(SDNode *N,
+                               TargetLowering::DAGCombinerInfo &DCI,
+                               const AArch64Subtarget *Subtarget) {
+  SelectionDAG &DAG = DCI.DAG;
+  DebugLoc DL = N->getDebugLoc();
+  EVT VT = N->getValueType(0);
+
+  assert(N->getOpcode() == ISD::OR && "Unexpected root");
+
+  // We need the LHS to be (and SOMETHING, MASK). Find out what that mask is or
+  // abandon the effort.
+  SDValue LHS = N->getOperand(0);
+  if (LHS.getOpcode() != ISD::AND)
+    return SDValue();
+
+  uint64_t LHSMask;
+  if (isa<ConstantSDNode>(LHS.getOperand(1)))
+    LHSMask = LHS->getConstantOperandVal(1);
+  else
+    return SDValue();
+
+  // We also need the RHS to be (and SOMETHING, MASK). Find out what that mask
+  // is or abandon the effort.
+  SDValue RHS = N->getOperand(1);
+  if (RHS.getOpcode() != ISD::AND)
+    return SDValue();
+
+  uint64_t RHSMask;
+  if (isa<ConstantSDNode>(RHS.getOperand(1)))
+    RHSMask = RHS->getConstantOperandVal(1);
+  else
+    return SDValue();
+
+  // Can't do anything if the masks are incompatible.
+  if (LHSMask & RHSMask)
+    return SDValue();
+
+  // Now we need one of the masks to be a contiguous field. Without loss of
+  // generality that should be the RHS one.
+  SDValue Bitfield = LHS.getOperand(0);
+  if (getLSBForBFI(DAG, DL, VT, Bitfield, LHSMask) != -1) {
+    // We know that LHS is a candidate new value, and RHS isn't already a better
+    // one.
+    std::swap(LHS, RHS);
+    std::swap(LHSMask, RHSMask);
+  }
+
+  // We've done our best to put the right operands in the right places, all we
+  // can do now is check whether a BFI exists.
+  Bitfield = RHS.getOperand(0);
+  int32_t LSB = getLSBForBFI(DAG, DL, VT, Bitfield, RHSMask);
+  if (LSB == -1)
+    return SDValue();
+
+  uint32_t Width = CountPopulation_64(RHSMask);
+  assert(Width && "Expected non-zero bitfield width");
+
+  SDValue BFI = DAG.getNode(AArch64ISD::BFI, DL, VT,
+                            LHS.getOperand(0), Bitfield,
+                            DAG.getConstant(LSB, MVT::i64),
+                            DAG.getConstant(Width, MVT::i64));
+
+  // Mask is trivial
+  if ((LHSMask | RHSMask) == (-1ULL >> (64 - VT.getSizeInBits())))
+    return BFI;
+
+  return DAG.getNode(ISD::AND, DL, VT, BFI,
+                     DAG.getConstant(LHSMask | RHSMask, VT));
+}
+
+/// Search for the bitwise combining (with careful masks) of a MaskedBFI and its
+/// original input. This is surprisingly common because SROA splits things up
+/// into i8 chunks, so the originally detected MaskedBFI may actually only act
+/// on the low (say) byte of a word. This is then orred into the rest of the
+/// word afterwards.
+///
+/// Basic input: (or (and OLDFIELD, MASK1), (MaskedBFI MASK2, OLDFIELD, ...)).
+///
+/// If MASK1 and MASK2 are compatible, we can fold the whole thing into the
+/// MaskedBFI. We can also deal with a certain amount of extend/truncate being
+/// involved.
+static SDValue tryCombineToLargerBFI(SDNode *N,
+                                     TargetLowering::DAGCombinerInfo &DCI,
+                                     const AArch64Subtarget *Subtarget) {
+  SelectionDAG &DAG = DCI.DAG;
+  DebugLoc DL = N->getDebugLoc();
+  EVT VT = N->getValueType(0);
+
+  // First job is to hunt for a MaskedBFI on either the left or right. Swap
+  // operands if it's actually on the right.
+  SDValue BFI;
+  SDValue PossExtraMask;
+  uint64_t ExistingMask = 0;
+  bool Extended = false;
+  if (findMaskedBFI(N->getOperand(0), BFI, ExistingMask, Extended))
+    PossExtraMask = N->getOperand(1);
+  else if (findMaskedBFI(N->getOperand(1), BFI, ExistingMask, Extended))
+    PossExtraMask = N->getOperand(0);
+  else
+    return SDValue();
+
+  // We can only combine a BFI with another compatible mask.
+  if (PossExtraMask.getOpcode() != ISD::AND ||
+      !isa<ConstantSDNode>(PossExtraMask.getOperand(1)))
+    return SDValue();
+
+  uint64_t ExtraMask = PossExtraMask->getConstantOperandVal(1);
+
+  // Masks must be compatible.
+  if (ExtraMask & ExistingMask)
+    return SDValue();
+
+  SDValue OldBFIVal = BFI.getOperand(0);
+  SDValue NewBFIVal = BFI.getOperand(1);
+  if (Extended) {
+    // We skipped a ZERO_EXTEND above, so the input to the MaskedBFIs should be
+    // 32-bit and we'll be forming a 64-bit MaskedBFI. The MaskedBFI arguments
+    // need to be made compatible.
+    assert(VT == MVT::i64 && BFI.getValueType() == MVT::i32
+           && "Invalid types for BFI");
+    OldBFIVal = DAG.getNode(ISD::ANY_EXTEND, DL, VT, OldBFIVal);
+    NewBFIVal = DAG.getNode(ISD::ANY_EXTEND, DL, VT, NewBFIVal);
+  }
+
+  // We need the MaskedBFI to be combined with a mask of the *same* value.
+  if (PossExtraMask.getOperand(0) != OldBFIVal)
+    return SDValue();
+
+  BFI = DAG.getNode(AArch64ISD::BFI, DL, VT,
+                    OldBFIVal, NewBFIVal,
+                    BFI.getOperand(2), BFI.getOperand(3));
+
+  // If the masking is trivial, we don't need to create it.
+  if ((ExtraMask | ExistingMask) == (-1ULL >> (64 - VT.getSizeInBits())))
+    return BFI;
+
+  return DAG.getNode(ISD::AND, DL, VT, BFI,
+                     DAG.getConstant(ExtraMask | ExistingMask, VT));
+}
+
+/// An EXTR instruction is made up of two shifts, ORed together. This helper
+/// searches for and classifies those shifts.
+static bool findEXTRHalf(SDValue N, SDValue &Src, uint32_t &ShiftAmount,
+                         bool &FromHi) {
+  if (N.getOpcode() == ISD::SHL)
+    FromHi = false;
+  else if (N.getOpcode() == ISD::SRL)
+    FromHi = true;
+  else
+    return false;
+
+  if (!isa<ConstantSDNode>(N.getOperand(1)))
+    return false;
+
+  ShiftAmount = N->getConstantOperandVal(1);
+  Src = N->getOperand(0);
+  return true;
+}
+
+/// EXTR instruction extracts a contiguous chunk of bits from two existing
+/// registers viewed as a high/low pair. This function looks for the pattern:
+/// (or (shl VAL1, #N), (srl VAL2, #RegWidth-N)) and replaces it with an
+/// EXTR. Can't quite be done in TableGen because the two immediates aren't
+/// independent.
+static SDValue tryCombineToEXTR(SDNode *N,
+                                TargetLowering::DAGCombinerInfo &DCI) {
+  SelectionDAG &DAG = DCI.DAG;
+  DebugLoc DL = N->getDebugLoc();
+  EVT VT = N->getValueType(0);
+
+  assert(N->getOpcode() == ISD::OR && "Unexpected root");
+
+  if (VT != MVT::i32 && VT != MVT::i64)
+    return SDValue();
+
+  SDValue LHS;
+  uint32_t ShiftLHS = 0;
+  bool LHSFromHi = 0;
+  if (!findEXTRHalf(N->getOperand(0), LHS, ShiftLHS, LHSFromHi))
+    return SDValue();
+
+  SDValue RHS;
+  uint32_t ShiftRHS = 0;
+  bool RHSFromHi = 0;
+  if (!findEXTRHalf(N->getOperand(1), RHS, ShiftRHS, RHSFromHi))
+    return SDValue();
+
+  // If they're both trying to come from the high part of the register, they're
+  // not really an EXTR.
+  if (LHSFromHi == RHSFromHi)
+    return SDValue();
+
+  if (ShiftLHS + ShiftRHS != VT.getSizeInBits())
+    return SDValue();
+
+  if (LHSFromHi) {
+    std::swap(LHS, RHS);
+    std::swap(ShiftLHS, ShiftRHS);
+  }
+
+  return DAG.getNode(AArch64ISD::EXTR, DL, VT,
+                     LHS, RHS,
+                     DAG.getConstant(ShiftRHS, MVT::i64));
+}
+
+/// Target-specific dag combine xforms for ISD::OR
+static SDValue PerformORCombine(SDNode *N,
+                                TargetLowering::DAGCombinerInfo &DCI,
+                                const AArch64Subtarget *Subtarget) {
+
+  SelectionDAG &DAG = DCI.DAG;
+  EVT VT = N->getValueType(0);
+
+  if(!DAG.getTargetLoweringInfo().isTypeLegal(VT))
+    return SDValue();
+
+  // Attempt to recognise bitfield-insert operations.
+  SDValue Res = tryCombineToBFI(N, DCI, Subtarget);
+  if (Res.getNode())
+    return Res;
+
+  // Attempt to combine an existing MaskedBFI operation into one with a larger
+  // mask.
+  Res = tryCombineToLargerBFI(N, DCI, Subtarget);
+  if (Res.getNode())
+    return Res;
+
+  Res = tryCombineToEXTR(N, DCI);
+  if (Res.getNode())
+    return Res;
+
+  return SDValue();
+}
+
+/// Target-specific dag combine xforms for ISD::SRA
+static SDValue PerformSRACombine(SDNode *N,
+                                 TargetLowering::DAGCombinerInfo &DCI) {
+
+  SelectionDAG &DAG = DCI.DAG;
+  DebugLoc DL = N->getDebugLoc();
+  EVT VT = N->getValueType(0);
+
+  // We're looking for an SRA/SHL pair which form an SBFX.
+
+  if (VT != MVT::i32 && VT != MVT::i64)
+    return SDValue();
+
+  if (!isa<ConstantSDNode>(N->getOperand(1)))
+    return SDValue();
+
+  uint64_t ExtraSignBits = N->getConstantOperandVal(1);
+  SDValue Shift = N->getOperand(0);
+
+  if (Shift.getOpcode() != ISD::SHL)
+    return SDValue();
+
+  if (!isa<ConstantSDNode>(Shift->getOperand(1)))
+    return SDValue();
+
+  uint64_t BitsOnLeft = Shift->getConstantOperandVal(1);
+  uint64_t Width = VT.getSizeInBits() - ExtraSignBits;
+  uint64_t LSB = VT.getSizeInBits() - Width - BitsOnLeft;
+
+  if (LSB > VT.getSizeInBits() || Width > VT.getSizeInBits())
+    return SDValue();
+
+  return DAG.getNode(AArch64ISD::SBFX, DL, VT, Shift.getOperand(0),
+                     DAG.getConstant(LSB, MVT::i64),
+                     DAG.getConstant(LSB + Width - 1, MVT::i64));
+}
+
+
+SDValue
+AArch64TargetLowering::PerformDAGCombine(SDNode *N,
+                                         DAGCombinerInfo &DCI) const {
+  switch (N->getOpcode()) {
+  default: break;
+  case ISD::AND: return PerformANDCombine(N, DCI);
+  case ISD::ATOMIC_FENCE: return PerformATOMIC_FENCECombine(N, DCI);
+  case ISD::ATOMIC_STORE: return PerformATOMIC_STORECombine(N, DCI);
+  case ISD::OR: return PerformORCombine(N, DCI, Subtarget);
+  case ISD::SRA: return PerformSRACombine(N, DCI);
+  }
+  return SDValue();
+}
+
+AArch64TargetLowering::ConstraintType
+AArch64TargetLowering::getConstraintType(const std::string &Constraint) const {
+  if (Constraint.size() == 1) {
+    switch (Constraint[0]) {
+    default: break;
+    case 'w': // An FP/SIMD vector register
+      return C_RegisterClass;
+    case 'I': // Constant that can be used with an ADD instruction
+    case 'J': // Constant that can be used with a SUB instruction
+    case 'K': // Constant that can be used with a 32-bit logical instruction
+    case 'L': // Constant that can be used with a 64-bit logical instruction
+    case 'M': // Constant that can be used as a 32-bit MOV immediate
+    case 'N': // Constant that can be used as a 64-bit MOV immediate
+    case 'Y': // Floating point constant zero
+    case 'Z': // Integer constant zero
+      return C_Other;
+    case 'Q': // A memory reference with base register and no offset
+      return C_Memory;
+    case 'S': // A symbolic address
+      return C_Other;
+    }
+  }
+
+  // FIXME: Ump, Utf, Usa, Ush
+  // Ump: A memory address suitable for ldp/stp in SI, DI, SF and DF modes,
+  //      whatever they may be
+  // Utf: A memory address suitable for ldp/stp in TF mode, whatever it may be
+  // Usa: An absolute symbolic address
+  // Ush: The high part (bits 32:12) of a pc-relative symbolic address
+  assert(Constraint != "Ump" && Constraint != "Utf" && Constraint != "Usa"
+         && Constraint != "Ush" && "Unimplemented constraints");
+
+  return TargetLowering::getConstraintType(Constraint);
+}
+
+TargetLowering::ConstraintWeight
+AArch64TargetLowering::getSingleConstraintMatchWeight(AsmOperandInfo &Info,
+                                                const char *Constraint) const {
+
+  llvm_unreachable("Constraint weight unimplemented");
+}
+
+void
+AArch64TargetLowering::LowerAsmOperandForConstraint(SDValue Op,
+                                                    std::string &Constraint,
+                                                    std::vector<SDValue> &Ops,
+                                                    SelectionDAG &DAG) const {
+  SDValue Result(0, 0);
+
+  // Only length 1 constraints are C_Other.
+  if (Constraint.size() != 1) return;
+
+  // Only C_Other constraints get lowered like this. That means constants for us
+  // so return early if there's no hope the constraint can be lowered.
+
+  switch(Constraint[0]) {
+  default: break;
+  case 'I': case 'J': case 'K': case 'L':
+  case 'M': case 'N': case 'Z': {
+    ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
+    if (!C)
+      return;
+
+    uint64_t CVal = C->getZExtValue();
+    uint32_t Bits;
+
+    switch (Constraint[0]) {
+    default:
+      // FIXME: 'M' and 'N' are MOV pseudo-insts -- unsupported in assembly. 'J'
+      // is a peculiarly useless SUB constraint.
+      llvm_unreachable("Unimplemented C_Other constraint");
+    case 'I':
+      if (CVal <= 0xfff)
+        break;
+      return;
+    case 'K':
+      if (A64Imms::isLogicalImm(32, CVal, Bits))
+        break;
+      return;
+    case 'L':
+      if (A64Imms::isLogicalImm(64, CVal, Bits))
+        break;
+      return;
+    case 'Z':
+      if (CVal == 0)
+        break;
+      return;
+    }
+
+    Result = DAG.getTargetConstant(CVal, Op.getValueType());
+    break;
+  }
+  case 'S': {
+    // An absolute symbolic address or label reference.
+    if (const GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(Op)) {
+      Result = DAG.getTargetGlobalAddress(GA->getGlobal(), Op.getDebugLoc(),
+                                          GA->getValueType(0));
+    } else if (const BlockAddressSDNode *BA
+                 = dyn_cast<BlockAddressSDNode>(Op)) {
+      Result = DAG.getTargetBlockAddress(BA->getBlockAddress(),
+                                         BA->getValueType(0));
+    } else if (const ExternalSymbolSDNode *ES
+                 = dyn_cast<ExternalSymbolSDNode>(Op)) {
+      Result = DAG.getTargetExternalSymbol(ES->getSymbol(),
+                                           ES->getValueType(0));
+    } else
+      return;
+    break;
+  }
+  case 'Y':
+    if (const ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(Op)) {
+      if (CFP->isExactlyValue(0.0)) {
+        Result = DAG.getTargetConstantFP(0.0, CFP->getValueType(0));
+        break;
+      }
+    }
+    return;
+  }
+
+  if (Result.getNode()) {
+    Ops.push_back(Result);
+    return;
+  }
+
+  // It's an unknown constraint for us. Let generic code have a go.
+  TargetLowering::LowerAsmOperandForConstraint(Op, Constraint, Ops, DAG);
+}
+
+std::pair<unsigned, const TargetRegisterClass*>
+AArch64TargetLowering::getRegForInlineAsmConstraint(
+                                                  const std::string &Constraint,
+                                                  EVT VT) const {
+  if (Constraint.size() == 1) {
+    switch (Constraint[0]) {
+    case 'r':
+      if (VT.getSizeInBits() <= 32)
+        return std::make_pair(0U, &AArch64::GPR32RegClass);
+      else if (VT == MVT::i64)
+        return std::make_pair(0U, &AArch64::GPR64RegClass);
+      break;
+    case 'w':
+      if (VT == MVT::f16)
+        return std::make_pair(0U, &AArch64::FPR16RegClass);
+      else if (VT == MVT::f32)
+        return std::make_pair(0U, &AArch64::FPR32RegClass);
+      else if (VT == MVT::f64)
+        return std::make_pair(0U, &AArch64::FPR64RegClass);
+      else if (VT.getSizeInBits() == 64)
+        return std::make_pair(0U, &AArch64::VPR64RegClass);
+      else if (VT == MVT::f128)
+        return std::make_pair(0U, &AArch64::FPR128RegClass);
+      else if (VT.getSizeInBits() == 128)
+        return std::make_pair(0U, &AArch64::VPR128RegClass);
+      break;
+    }
+  }
+
+  // Use the default implementation in TargetLowering to convert the register
+  // constraint into a member of a register class.
+  return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
+}
diff --git a/lib/Target/AArch64/AArch64ISelLowering.h b/lib/Target/AArch64/AArch64ISelLowering.h
new file mode 100644
index 000000000000..4960d286e9de
--- /dev/null
+++ b/lib/Target/AArch64/AArch64ISelLowering.h
@@ -0,0 +1,247 @@
+//==-- AArch64ISelLowering.h - AArch64 DAG Lowering Interface ----*- C++ -*-==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the interfaces that AArch64 uses to lower LLVM code into a
+// selection DAG.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_TARGET_AARCH64_ISELLOWERING_H
+#define LLVM_TARGET_AARCH64_ISELLOWERING_H
+
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/CodeGen/CallingConvLower.h"
+#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/Target/TargetLowering.h"
+
+
+namespace llvm {
+namespace AArch64ISD {
+  enum NodeType {
+    // Start the numbering from where ISD NodeType finishes.
+    FIRST_NUMBER = ISD::BUILTIN_OP_END,
+
+    // This is a conditional branch which also notes the flag needed
+    // (eq/sgt/...). A64 puts this information on the branches rather than
+    // compares as LLVM does.
+    BR_CC,
+
+    // A node to be selected to an actual call operation: either BL or BLR in
+    // the absence of tail calls.
+    Call,
+
+    // Indicates a floating-point immediate which fits into the format required
+    // by the FMOV instructions. First (and only) operand is the 8-bit encoded
+    // value of that immediate.
+    FPMOV,
+
+    // Corresponds directly to an EXTR instruction. Operands are an LHS an RHS
+    // and an LSB.
+    EXTR,
+
+    // Wraps a load from the GOT, which should always be performed with a 64-bit
+    // load instruction. This prevents the DAG combiner folding a truncate to
+    // form a smaller memory access.
+    GOTLoad,
+
+    // Performs a bitfield insert. Arguments are: the value being inserted into;
+    // the value being inserted; least significant bit changed; width of the
+    // field.
+    BFI,
+
+    // Simply a convenient node inserted during ISelLowering to represent
+    // procedure return. Will almost certainly be selected to "RET".
+    Ret,
+
+    /// Extracts a field of contiguous bits from the source and sign extends
+    /// them into a single register. Arguments are: source; immr; imms. Note
+    /// these are pre-encoded since DAG matching can't cope with combining LSB
+    /// and Width into these values itself.
+    SBFX,
+
+    /// This is an A64-ification of the standard LLVM SELECT_CC operation. The
+    /// main difference is that it only has the values and an A64 condition,
+    /// which will be produced by a setcc instruction.
+    SELECT_CC,
+
+    /// This serves most of the functions of the LLVM SETCC instruction, for two
+    /// purposes. First, it prevents optimisations from fiddling with the
+    /// compare after we've moved the CondCode information onto the SELECT_CC or
+    /// BR_CC instructions. Second, it gives a legal instruction for the actual
+    /// comparison.
+    ///
+    /// It keeps a record of the condition flags asked for because certain
+    /// instructions are only valid for a subset of condition codes.
+    SETCC,
+
+    // Designates a node which is a tail call: both a call and a return
+    // instruction as far as selction is concerned. It should be selected to an
+    // unconditional branch. Has the usual plethora of call operands, but: 1st
+    // is callee, 2nd is stack adjustment required immediately before branch.
+    TC_RETURN,
+
+    // Designates a call used to support the TLS descriptor ABI. The call itself
+    // will be indirect ("BLR xN") but a relocation-specifier (".tlsdesccall
+    // var") must be attached somehow during code generation. It takes two
+    // operands: the callee and the symbol to be relocated against.
+    TLSDESCCALL,
+
+    // Leaf node which will be lowered to an appropriate MRS to obtain the
+    // thread pointer: TPIDR_EL0.
+    THREAD_POINTER,
+
+    /// Extracts a field of contiguous bits from the source and zero extends
+    /// them into a single register. Arguments are: source; immr; imms. Note
+    /// these are pre-encoded since DAG matching can't cope with combining LSB
+    /// and Width into these values itself.
+    UBFX,
+
+    // Wraps an address which the ISelLowering phase has decided should be
+    // created using the small absolute memory model: i.e. adrp/add or
+    // adrp/mem-op. This exists to prevent bare TargetAddresses which may never
+    // get selected.
+    WrapperSmall
+  };
+}
+
+
+class AArch64Subtarget;
+class AArch64TargetMachine;
+
+class AArch64TargetLowering : public TargetLowering {
+public:
+  explicit AArch64TargetLowering(AArch64TargetMachine &TM);
+
+  const char *getTargetNodeName(unsigned Opcode) const;
+
+  CCAssignFn *CCAssignFnForNode(CallingConv::ID CC) const;
+
+  SDValue LowerFormalArguments(SDValue Chain,
+                               CallingConv::ID CallConv, bool isVarArg,
+                               const SmallVectorImpl<ISD::InputArg> &Ins,
+                               DebugLoc dl, SelectionDAG &DAG,
+                               SmallVectorImpl<SDValue> &InVals) const;
+
+  SDValue LowerReturn(SDValue Chain,
+                      CallingConv::ID CallConv, bool isVarArg,
+                      const SmallVectorImpl<ISD::OutputArg> &Outs,
+                      const SmallVectorImpl<SDValue> &OutVals,
+                      DebugLoc dl, SelectionDAG &DAG) const;
+
+  SDValue LowerCall(CallLoweringInfo &CLI,
+                    SmallVectorImpl<SDValue> &InVals) const;
+
+  SDValue LowerCallResult(SDValue Chain, SDValue InFlag,
+                          CallingConv::ID CallConv, bool IsVarArg,
+                          const SmallVectorImpl<ISD::InputArg> &Ins,
+                          DebugLoc dl, SelectionDAG &DAG,
+                          SmallVectorImpl<SDValue> &InVals) const;
+
+  void SaveVarArgRegisters(CCState &CCInfo, SelectionDAG &DAG,
+                           DebugLoc DL, SDValue &Chain) const;
+
+
+  /// IsEligibleForTailCallOptimization - Check whether the call is eligible
+  /// for tail call optimization. Targets which want to do tail call
+  /// optimization should implement this function.
+  bool IsEligibleForTailCallOptimization(SDValue Callee,
+                                    CallingConv::ID CalleeCC,
+                                    bool IsVarArg,
+                                    bool IsCalleeStructRet,
+                                    bool IsCallerStructRet,
+                                    const SmallVectorImpl<ISD::OutputArg> &Outs,
+                                    const SmallVectorImpl<SDValue> &OutVals,
+                                    const SmallVectorImpl<ISD::InputArg> &Ins,
+                                    SelectionDAG& DAG) const;
+
+  /// Finds the incoming stack arguments which overlap the given fixed stack
+  /// object and incorporates their load into the current chain. This prevents
+  /// an upcoming store from clobbering the stack argument before it's used.
+  SDValue addTokenForArgument(SDValue Chain, SelectionDAG &DAG,
+                              MachineFrameInfo *MFI, int ClobberedFI) const;
+
+  EVT getSetCCResultType(EVT VT) const;
+
+  bool DoesCalleeRestoreStack(CallingConv::ID CallCC, bool TailCallOpt) const;
+
+  bool IsTailCallConvention(CallingConv::ID CallCC) const;
+
+  SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
+
+  bool isLegalICmpImmediate(int64_t Val) const;
+  SDValue getSelectableIntSetCC(SDValue LHS, SDValue RHS, ISD::CondCode CC,
+                         SDValue &A64cc, SelectionDAG &DAG, DebugLoc &dl) const;
+
+  virtual MachineBasicBlock *
+  EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *MBB) const;
+
+  MachineBasicBlock *
+  emitAtomicBinary(MachineInstr *MI, MachineBasicBlock *MBB,
+                   unsigned Size, unsigned Opcode) const;
+
+  MachineBasicBlock *
+  emitAtomicBinaryMinMax(MachineInstr *MI, MachineBasicBlock *BB,
+                         unsigned Size, unsigned CmpOp,
+                         A64CC::CondCodes Cond) const;
+  MachineBasicBlock *
+  emitAtomicCmpSwap(MachineInstr *MI, MachineBasicBlock *BB,
+                    unsigned Size) const;
+
+  MachineBasicBlock *
+  EmitF128CSEL(MachineInstr *MI, MachineBasicBlock *MBB) const;
+
+  SDValue LowerATOMIC_FENCE(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerATOMIC_STORE(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerBR_CC(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerF128ToCall(SDValue Op, SelectionDAG &DAG,
+                          RTLIB::Libcall Call) const;
+  SDValue LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG, bool IsSigned) const;
+  SDValue LowerGlobalAddressELF(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerTLSDescCall(SDValue SymAddr, SDValue DescAddr, DebugLoc DL,
+                           SelectionDAG &DAG) const;
+  SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG, bool IsSigned) const;
+  SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerVACOPY(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) const;
+
+  virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
+
+  /// isFMAFasterThanMulAndAdd - Return true if an FMA operation is faster than
+  /// a pair of mul and add instructions. fmuladd intrinsics will be expanded to
+  /// FMAs when this method returns true (and FMAs are legal), otherwise fmuladd
+  /// is expanded to mul + add.
+  virtual bool isFMAFasterThanMulAndAdd(EVT) const { return true; }
+
+  ConstraintType getConstraintType(const std::string &Constraint) const;
+
+  ConstraintWeight getSingleConstraintMatchWeight(AsmOperandInfo &Info,
+                                                  const char *Constraint) const;
+  void LowerAsmOperandForConstraint(SDValue Op,
+                                    std::string &Constraint,
+                                    std::vector<SDValue> &Ops,
+                                    SelectionDAG &DAG) const;
+
+  std::pair<unsigned, const TargetRegisterClass*>
+  getRegForInlineAsmConstraint(const std::string &Constraint, EVT VT) const;
+private:
+  const AArch64Subtarget *Subtarget;
+  const TargetRegisterInfo *RegInfo;
+  const InstrItineraryData *Itins;
+};
+} // namespace llvm
+
+#endif // LLVM_TARGET_AARCH64_ISELLOWERING_H
diff --git a/lib/Target/AArch64/AArch64InstrFormats.td b/lib/Target/AArch64/AArch64InstrFormats.td
new file mode 100644
index 000000000000..cb93471058df
--- /dev/null
+++ b/lib/Target/AArch64/AArch64InstrFormats.td
@@ -0,0 +1,961 @@
+//===- AArch64InstrFormats.td - AArch64 Instruction Formats --*- tablegen -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+// This file describes AArch64 instruction formats, down to the level of the
+// instruction's overall class.
+// ===----------------------------------------------------------------------===//
+
+
+//===----------------------------------------------------------------------===//
+// A64 Instruction Format Definitions.
+//===----------------------------------------------------------------------===//
+
+// A64 is currently the only instruction set supported by the AArch64
+// architecture.
+class A64Inst<dag outs, dag ins, string asmstr, list<dag> patterns,
+              InstrItinClass itin>
+    : Instruction {
+  // All A64 instructions are 32-bit. This field will be filled in
+  // gradually going down the hierarchy.
+  field bits<32> Inst;
+
+  field bits<32> Unpredictable = 0;
+  // SoftFail is the generic name for this field, but we alias it so
+  // as to make it more obvious what it means in ARM-land.
+  field bits<32> SoftFail = Unpredictable;
+
+  // LLVM-level model of the AArch64/A64 distinction.
+  let Namespace = "AArch64";
+  let DecoderNamespace = "A64";
+  let Size = 4;
+
+  // Set the templated fields
+  let OutOperandList = outs;
+  let InOperandList = ins;
+  let AsmString = asmstr;
+  let Pattern = patterns;
+  let Itinerary = itin;
+}
+
+class PseudoInst<dag outs, dag ins, list<dag> patterns> : Instruction {
+  let Namespace = "AArch64";
+
+  let OutOperandList = outs;
+  let InOperandList= ins;
+  let Pattern = patterns;
+  let isCodeGenOnly = 1;
+  let isPseudo = 1;
+}
+
+// Represents a pseudo-instruction that represents a single A64 instruction for
+// whatever reason, the eventual result will be a 32-bit real instruction.
+class A64PseudoInst<dag outs, dag ins, list<dag> patterns>
+  : PseudoInst<outs, ins, patterns> {
+  let Size = 4;
+}
+
+// As above, this will be a single A64 instruction, but we can actually give the
+// expansion in TableGen.
+class A64PseudoExpand<dag outs, dag ins, list<dag> patterns, dag Result>
+  : A64PseudoInst<outs, ins, patterns>,
+    PseudoInstExpansion<Result>;
+
+
+// First, some common cross-hierarchy register formats.
+
+class A64InstRd<dag outs, dag ins, string asmstr,
+                list<dag> patterns, InstrItinClass itin>
+  : A64Inst<outs, ins, asmstr, patterns, itin> {
+  bits<5> Rd;
+
+  let Inst{4-0} = Rd;
+}
+
+class A64InstRt<dag outs, dag ins, string asmstr,
+                list<dag> patterns, InstrItinClass itin>
+  : A64Inst<outs, ins, asmstr, patterns, itin> {
+  bits<5> Rt;
+
+  let Inst{4-0} = Rt;
+}
+
+
+class A64InstRdn<dag outs, dag ins, string asmstr,
+                 list<dag> patterns, InstrItinClass itin>
+    : A64InstRd<outs, ins, asmstr, patterns, itin> {
+  // Inherit rdt
+  bits<5> Rn;
+
+  let Inst{9-5} = Rn;
+}
+
+class A64InstRtn<dag outs, dag ins, string asmstr,
+                list<dag> patterns, InstrItinClass itin>
+    : A64InstRt<outs, ins, asmstr, patterns, itin> {
+  // Inherit rdt
+  bits<5> Rn;
+
+  let Inst{9-5} = Rn;
+}
+
+// Instructions taking Rt,Rt2,Rn
+class A64InstRtt2n<dag outs, dag ins, string asmstr,
+                   list<dag> patterns, InstrItinClass itin>
+  : A64InstRtn<outs, ins, asmstr, patterns, itin> {
+  bits<5> Rt2;
+
+  let Inst{14-10} = Rt2;
+}
+
+class A64InstRdnm<dag outs, dag ins, string asmstr,
+                  list<dag> patterns, InstrItinClass itin>
+  : A64InstRdn<outs, ins, asmstr, patterns, itin> {
+  bits<5> Rm;
+
+  let Inst{20-16} = Rm;
+}
+
+//===----------------------------------------------------------------------===//
+//
+// Actual A64 Instruction Formats
+//
+
+// Format for Add-subtract (extended register) instructions.
+class A64I_addsubext<bit sf, bit op, bit S, bits<2> opt, bits<3> option,
+                     dag outs, dag ins, string asmstr, list<dag> patterns,
+                     InstrItinClass itin>
+    : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+    bits<3> Imm3;
+
+    let Inst{31} = sf;
+    let Inst{30} = op;
+    let Inst{29} = S;
+    let Inst{28-24} = 0b01011;
+    let Inst{23-22} = opt;
+    let Inst{21} = 0b1;
+    // Rm inherited in 20-16
+    let Inst{15-13} = option;
+    let Inst{12-10} = Imm3;
+    // Rn inherited in 9-5
+    // Rd inherited in 4-0
+}
+
+// Format for Add-subtract (immediate) instructions.
+class A64I_addsubimm<bit sf, bit op, bit S, bits<2> shift,
+                     dag outs, dag ins, string asmstr,
+                     list<dag> patterns, InstrItinClass itin>
+  : A64InstRdn<outs, ins, asmstr, patterns, itin> {
+  bits<12> Imm12;
+
+  let Inst{31} = sf;
+  let Inst{30} = op;
+  let Inst{29} = S;
+  let Inst{28-24} = 0b10001;
+  let Inst{23-22} = shift;
+  let Inst{21-10} = Imm12;
+}
+
+// Format for Add-subtract (shifted register) instructions.
+class A64I_addsubshift<bit sf, bit op, bit S, bits<2> shift,
+                       dag outs, dag ins, string asmstr, list<dag> patterns,
+                       InstrItinClass itin>
+    : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+    bits<6> Imm6;
+
+    let Inst{31} = sf;
+    let Inst{30} = op;
+    let Inst{29} = S;
+    let Inst{28-24} = 0b01011;
+    let Inst{23-22} = shift;
+    let Inst{21} = 0b0;
+    // Rm inherited in 20-16
+    let Inst{15-10} = Imm6;
+    // Rn inherited in 9-5
+    // Rd inherited in 4-0
+}
+
+// Format for Add-subtract (with carry) instructions.
+class A64I_addsubcarry<bit sf, bit op, bit S, bits<6> opcode2,
+                       dag outs, dag ins, string asmstr, list<dag> patterns,
+                       InstrItinClass itin>
+    : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+    let Inst{31} = sf;
+    let Inst{30} = op;
+    let Inst{29} = S;
+    let Inst{28-21} = 0b11010000;
+    // Rm inherited in 20-16
+    let Inst{15-10} = opcode2;
+    // Rn inherited in 9-5
+    // Rd inherited in 4-0
+}
+
+
+// Format for Bitfield instructions
+class A64I_bitfield<bit sf, bits<2> opc, bit n,
+                    dag outs, dag ins, string asmstr,
+                    list<dag> patterns, InstrItinClass itin>
+  : A64InstRdn<outs, ins, asmstr, patterns, itin> {
+  bits<6> ImmR;
+  bits<6> ImmS;
+
+  let Inst{31} = sf;
+  let Inst{30-29} = opc;
+  let Inst{28-23} = 0b100110;
+  let Inst{22} = n;
+  let Inst{21-16} = ImmR;
+  let Inst{15-10} = ImmS;
+  // Inherit Rn in 9-5
+  // Inherit Rd in 4-0
+}
+
+// Format for compare and branch (immediate) instructions.
+class A64I_cmpbr<bit sf, bit op,
+                  dag outs, dag ins, string asmstr,
+                  list<dag> patterns, InstrItinClass itin>
+  : A64InstRt<outs, ins, asmstr, patterns, itin> {
+  bits<19> Label;
+
+  let Inst{31} = sf;
+  let Inst{30-25} = 0b011010;
+  let Inst{24} = op;
+  let Inst{23-5} = Label;
+  // Inherit Rt in 4-0
+}
+
+// Format for conditional branch (immediate) instructions.
+class A64I_condbr<bit o1, bit o0,
+                  dag outs, dag ins, string asmstr,
+                  list<dag> patterns, InstrItinClass itin>
+  : A64Inst<outs, ins, asmstr, patterns, itin> {
+  bits<19> Label;
+  bits<4> Cond;
+
+  let Inst{31-25} = 0b0101010;
+  let Inst{24} = o1;
+  let Inst{23-5} = Label;
+  let Inst{4} = o0;
+  let Inst{3-0} = Cond;
+}
+
+// Format for conditional compare (immediate) instructions.
+class A64I_condcmpimm<bit sf, bit op, bit o2, bit o3, bit s,
+                      dag outs, dag ins, string asmstr,
+                      list<dag> patterns, InstrItinClass itin>
+  : A64Inst<outs, ins, asmstr, patterns, itin> {
+  bits<5> Rn;
+  bits<5> UImm5;
+  bits<4> NZCVImm;
+  bits<4> Cond;
+
+  let Inst{31} = sf;
+  let Inst{30} = op;
+  let Inst{29} = s;
+  let Inst{28-21} = 0b11010010;
+  let Inst{20-16} = UImm5;
+  let Inst{15-12} = Cond;
+  let Inst{11} = 0b1;
+  let Inst{10} = o2;
+  let Inst{9-5} = Rn;
+  let Inst{4} = o3;
+  let Inst{3-0} = NZCVImm;
+}
+
+// Format for conditional compare (register) instructions.
+class A64I_condcmpreg<bit sf, bit op, bit o2, bit o3, bit s,
+                      dag outs, dag ins, string asmstr,
+                      list<dag> patterns, InstrItinClass itin>
+  : A64Inst<outs, ins, asmstr, patterns, itin> {
+  bits<5> Rn;
+  bits<5> Rm;
+  bits<4> NZCVImm;
+  bits<4> Cond;
+
+
+  let Inst{31} = sf;
+  let Inst{30} = op;
+  let Inst{29} = s;
+  let Inst{28-21} = 0b11010010;
+  let Inst{20-16} = Rm;
+  let Inst{15-12} = Cond;
+  let Inst{11} = 0b0;
+  let Inst{10} = o2;
+  let Inst{9-5} = Rn;
+  let Inst{4} = o3;
+  let Inst{3-0} = NZCVImm;
+}
+
+// Format for conditional select instructions.
+class A64I_condsel<bit sf, bit op, bit s, bits<2> op2,
+                   dag outs, dag ins, string asmstr,
+                   list<dag> patterns, InstrItinClass itin>
+  : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+  bits<4> Cond;
+
+  let Inst{31} = sf;
+  let Inst{30} = op;
+  let Inst{29} = s;
+  let Inst{28-21} = 0b11010100;
+  // Inherit Rm in 20-16
+  let Inst{15-12} = Cond;
+  let Inst{11-10} = op2;
+  // Inherit Rn in 9-5
+  // Inherit Rd in 4-0
+}
+
+// Format for data processing (1 source) instructions
+class A64I_dp_1src<bit sf, bit S, bits<5> opcode2, bits<6> opcode,
+                string asmstr, dag outs, dag ins,
+                list<dag> patterns, InstrItinClass itin>
+  : A64InstRdn<outs, ins, asmstr, patterns, itin> {
+  let Inst{31} = sf;
+  let Inst{30} = 0b1;
+  let Inst{29} = S;
+  let Inst{28-21} = 0b11010110;
+  let Inst{20-16} = opcode2;
+  let Inst{15-10} = opcode;
+}
+
+// Format for data processing (2 source) instructions
+class A64I_dp_2src<bit sf, bits<6> opcode, bit S,
+                string asmstr, dag outs, dag ins,
+                list<dag> patterns, InstrItinClass itin>
+  : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+  let Inst{31} = sf;
+  let Inst{30} = 0b0;
+  let Inst{29} = S;
+  let Inst{28-21} = 0b11010110;
+  let Inst{15-10} = opcode;
+}
+
+// Format for data-processing (3 source) instructions
+
+class A64I_dp3<bit sf, bits<6> opcode,
+               dag outs, dag ins, string asmstr,
+               list<dag> patterns, InstrItinClass itin>
+  : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+  bits<5> Ra;
+
+  let Inst{31} = sf;
+  let Inst{30-29} = opcode{5-4};
+  let Inst{28-24} = 0b11011;
+  let Inst{23-21} = opcode{3-1};
+  // Inherits Rm in 20-16
+  let Inst{15} = opcode{0};
+  let Inst{14-10} = Ra;
+  // Inherits Rn in 9-5
+  // Inherits Rd in 4-0
+}
+
+// Format for exception generation instructions
+class A64I_exception<bits<3> opc, bits<3> op2, bits<2> ll,
+                     dag outs, dag ins, string asmstr,
+                     list<dag> patterns, InstrItinClass itin>
+  : A64Inst<outs, ins, asmstr, patterns, itin> {
+  bits<16> UImm16;
+
+  let Inst{31-24} = 0b11010100;
+  let Inst{23-21} = opc;
+  let Inst{20-5} = UImm16;
+  let Inst{4-2} = op2;
+  let Inst{1-0} = ll;
+}
+
+// Format for extract (immediate) instructions
+class A64I_extract<bit sf, bits<3> op, bit n,
+                   dag outs, dag ins, string asmstr,
+                   list<dag> patterns, InstrItinClass itin>
+  : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+  bits<6> LSB;
+
+  let Inst{31} = sf;
+  let Inst{30-29} = op{2-1};
+  let Inst{28-23} = 0b100111;
+  let Inst{22} = n;
+  let Inst{21} = op{0};
+  // Inherits Rm in bits 20-16
+  let Inst{15-10} = LSB;
+  // Inherits Rn in 9-5
+  // Inherits Rd in 4-0
+}
+
+// Format for floating-point compare instructions.
+class A64I_fpcmp<bit m, bit s, bits<2> type, bits<2> op, bits<5> opcode2,
+                dag outs, dag ins, string asmstr,
+                list<dag> patterns, InstrItinClass itin>
+  : A64Inst<outs, ins, asmstr, patterns, itin> {
+  bits<5> Rn;
+  bits<5> Rm;
+
+  let Inst{31} = m;
+  let Inst{30} = 0b0;
+  let Inst{29} = s;
+  let Inst{28-24} = 0b11110;
+  let Inst{23-22} = type;
+  let Inst{21} = 0b1;
+  let Inst{20-16} = Rm;
+  let Inst{15-14} = op;
+  let Inst{13-10} = 0b1000;
+  let Inst{9-5} = Rn;
+  let Inst{4-0} = opcode2;
+}
+
+// Format for floating-point conditional compare instructions.
+class A64I_fpccmp<bit m, bit s, bits<2> type, bit op,
+                 dag outs, dag ins, string asmstr,
+                 list<dag> patterns, InstrItinClass itin>
+  : A64InstRdn<outs, ins, asmstr, patterns, itin> {
+  bits<5> Rn;
+  bits<5> Rm;
+  bits<4> NZCVImm;
+  bits<4> Cond;
+
+  let Inst{31} = m;
+  let Inst{30} = 0b0;
+  let Inst{29} = s;
+  let Inst{28-24} = 0b11110;
+  let Inst{23-22} = type;
+  let Inst{21} = 0b1;
+  let Inst{20-16} = Rm;
+  let Inst{15-12} = Cond;
+  let Inst{11-10} = 0b01;
+  let Inst{9-5} = Rn;
+  let Inst{4} = op;
+  let Inst{3-0} = NZCVImm;
+}
+
+// Format for floating-point conditional select instructions.
+class A64I_fpcondsel<bit m, bit s, bits<2> type,
+                     dag outs, dag ins, string asmstr,
+                     list<dag> patterns, InstrItinClass itin>
+  : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+  bits<4> Cond;
+
+  let Inst{31} = m;
+  let Inst{30} = 0b0;
+  let Inst{29} = s;
+  let Inst{28-24} = 0b11110;
+  let Inst{23-22} = type;
+  let Inst{21} = 0b1;
+  // Inherit Rm in 20-16
+  let Inst{15-12} = Cond;
+  let Inst{11-10} = 0b11;
+  // Inherit Rn in 9-5
+  // Inherit Rd in 4-0
+}
+
+
+// Format for floating-point data-processing (1 source) instructions.
+class A64I_fpdp1<bit m, bit s, bits<2> type, bits<6> opcode,
+                 dag outs, dag ins, string asmstr,
+                 list<dag> patterns, InstrItinClass itin>
+  : A64InstRdn<outs, ins, asmstr, patterns, itin> {
+  let Inst{31} = m;
+  let Inst{30} = 0b0;
+  let Inst{29} = s;
+  let Inst{28-24} = 0b11110;
+  let Inst{23-22} = type;
+  let Inst{21} = 0b1;
+  let Inst{20-15} = opcode;
+  let Inst{14-10} = 0b10000;
+  // Inherit Rn in 9-5
+  // Inherit Rd in 4-0
+}
+
+// Format for floating-point data-processing (2 sources) instructions.
+class A64I_fpdp2<bit m, bit s, bits<2> type, bits<4> opcode,
+                 dag outs, dag ins, string asmstr,
+                 list<dag> patterns, InstrItinClass itin>
+  : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+  let Inst{31} = m;
+  let Inst{30} = 0b0;
+  let Inst{29} = s;
+  let Inst{28-24} = 0b11110;
+  let Inst{23-22} = type;
+  let Inst{21} = 0b1;
+  // Inherit Rm in 20-16
+  let Inst{15-12} = opcode;
+  let Inst{11-10} = 0b10;
+  // Inherit Rn in 9-5
+  // Inherit Rd in 4-0
+}
+
+// Format for floating-point data-processing (3 sources) instructions.
+class A64I_fpdp3<bit m, bit s, bits<2> type, bit o1, bit o0,
+                 dag outs, dag ins, string asmstr,
+                 list<dag> patterns, InstrItinClass itin>
+  : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+  bits<5> Ra;
+
+  let Inst{31} = m;
+  let Inst{30} = 0b0;
+  let Inst{29} = s;
+  let Inst{28-24} = 0b11111;
+  let Inst{23-22} = type;
+  let Inst{21} = o1;
+  // Inherit Rm in 20-16
+  let Inst{15} = o0;
+  let Inst{14-10} = Ra;
+  // Inherit Rn in 9-5
+  // Inherit Rd in 4-0
+}
+
+// Format for floating-point <-> fixed-point conversion instructions.
+class A64I_fpfixed<bit sf, bit s, bits<2> type, bits<2> mode, bits<3> opcode,
+                 dag outs, dag ins, string asmstr,
+                 list<dag> patterns, InstrItinClass itin>
+  : A64InstRdn<outs, ins, asmstr, patterns, itin> {
+  bits<6> Scale;
+
+  let Inst{31} = sf;
+  let Inst{30} = 0b0;
+  let Inst{29} = s;
+  let Inst{28-24} = 0b11110;
+  let Inst{23-22} = type;
+  let Inst{21} = 0b0;
+  let Inst{20-19} = mode;
+  let Inst{18-16} = opcode;
+  let Inst{15-10} = Scale;
+  // Inherit Rn in 9-5
+  // Inherit Rd in 4-0
+}
+
+// Format for floating-point <-> integer conversion instructions.
+class A64I_fpint<bit sf, bit s, bits<2> type, bits<2> rmode, bits<3> opcode,
+                 dag outs, dag ins, string asmstr,
+                 list<dag> patterns, InstrItinClass itin>
+  : A64InstRdn<outs, ins, asmstr, patterns, itin> {
+  let Inst{31} = sf;
+  let Inst{30} = 0b0;
+  let Inst{29} = s;
+  let Inst{28-24} = 0b11110;
+  let Inst{23-22} = type;
+  let Inst{21} = 0b1;
+  let Inst{20-19} = rmode;
+  let Inst{18-16} = opcode;
+  let Inst{15-10} = 0b000000;
+  // Inherit Rn in 9-5
+  // Inherit Rd in 4-0
+}
+
+
+// Format for floating-point immediate instructions.
+class A64I_fpimm<bit m, bit s, bits<2> type, bits<5> imm5,
+                 dag outs, dag ins, string asmstr,
+                 list<dag> patterns, InstrItinClass itin>
+  : A64InstRd<outs, ins, asmstr, patterns, itin> {
+  bits<8> Imm8;
+
+  let Inst{31} = m;
+  let Inst{30} = 0b0;
+  let Inst{29} = s;
+  let Inst{28-24} = 0b11110;
+  let Inst{23-22} = type;
+  let Inst{21} = 0b1;
+  let Inst{20-13} = Imm8;
+  let Inst{12-10} = 0b100;
+  let Inst{9-5} = imm5;
+  // Inherit Rd in 4-0
+}
+
+// Format for load-register (literal) instructions.
+class A64I_LDRlit<bits<2> opc, bit v,
+                  dag outs, dag ins, string asmstr,
+                  list<dag> patterns, InstrItinClass itin>
+  : A64InstRt<outs, ins, asmstr, patterns, itin> {
+  bits<19> Imm19;
+
+  let Inst{31-30} = opc;
+  let Inst{29-27} = 0b011;
+  let Inst{26} = v;
+  let Inst{25-24} = 0b00;
+  let Inst{23-5} = Imm19;
+  // Inherit Rt in 4-0
+}
+
+// Format for load-store exclusive instructions.
+class A64I_LDSTex_tn<bits<2> size, bit o2, bit L, bit o1, bit o0,
+                 dag outs, dag ins, string asmstr,
+                 list <dag> patterns, InstrItinClass itin>
+  : A64InstRtn<outs, ins, asmstr, patterns, itin> {
+  let Inst{31-30} = size;
+  let Inst{29-24} = 0b001000;
+  let Inst{23} = o2;
+  let Inst{22} = L;
+  let Inst{21} = o1;
+  let Inst{15} = o0;
+}
+
+class A64I_LDSTex_tt2n<bits<2> size, bit o2, bit L, bit o1, bit o0,
+                     dag outs, dag ins, string asmstr,
+                     list <dag> patterns, InstrItinClass itin>:
+      A64I_LDSTex_tn<size, o2, L, o1, o0, outs, ins, asmstr, patterns, itin>{
+   bits<5> Rt2;
+   let Inst{14-10} = Rt2;
+}
+
+class A64I_LDSTex_stn<bits<2> size, bit o2, bit L, bit o1, bit o0,
+                     dag outs, dag ins, string asmstr,
+                     list <dag> patterns, InstrItinClass itin>:
+      A64I_LDSTex_tn<size, o2, L, o1, o0, outs, ins, asmstr, patterns, itin>{
+   bits<5> Rs;
+   let Inst{20-16} = Rs;
+}
+
+class A64I_LDSTex_stt2n<bits<2> size, bit o2, bit L, bit o1, bit o0,
+                     dag outs, dag ins, string asmstr,
+                     list <dag> patterns, InstrItinClass itin>:
+      A64I_LDSTex_stn<size, o2, L, o1, o0, outs, ins, asmstr, patterns, itin>{
+   bits<5> Rt2;
+   let Inst{14-10} = Rt2;
+}
+
+// Format for load-store register (immediate post-indexed) instructions
+class A64I_LSpostind<bits<2> size, bit v, bits<2> opc,
+                     dag outs, dag ins, string asmstr,
+                     list<dag> patterns, InstrItinClass itin>
+  : A64InstRtn<outs, ins, asmstr, patterns, itin> {
+  bits<9> SImm9;
+
+  let Inst{31-30} = size;
+  let Inst{29-27} = 0b111;
+  let Inst{26} = v;
+  let Inst{25-24} = 0b00;
+  let Inst{23-22} = opc;
+  let Inst{21} = 0b0;
+  let Inst{20-12} = SImm9;
+  let Inst{11-10} = 0b01;
+  // Inherit Rn in 9-5
+  // Inherit Rt in 4-0
+}
+
+// Format for load-store register (immediate pre-indexed) instructions
+class A64I_LSpreind<bits<2> size, bit v, bits<2> opc,
+                    dag outs, dag ins, string asmstr,
+                    list<dag> patterns, InstrItinClass itin>
+  : A64InstRtn<outs, ins, asmstr, patterns, itin> {
+  bits<9> SImm9;
+
+
+  let Inst{31-30} = size;
+  let Inst{29-27} = 0b111;
+  let Inst{26} = v;
+  let Inst{25-24} = 0b00;
+  let Inst{23-22} = opc;
+  let Inst{21} = 0b0;
+  let Inst{20-12} = SImm9;
+  let Inst{11-10} = 0b11;
+  // Inherit Rn in 9-5
+  // Inherit Rt in 4-0
+}
+
+// Format for load-store register (unprivileged) instructions
+class A64I_LSunpriv<bits<2> size, bit v, bits<2> opc,
+                    dag outs, dag ins, string asmstr,
+                    list<dag> patterns, InstrItinClass itin>
+  : A64InstRtn<outs, ins, asmstr, patterns, itin> {
+  bits<9> SImm9;
+
+
+  let Inst{31-30} = size;
+  let Inst{29-27} = 0b111;
+  let Inst{26} = v;
+  let Inst{25-24} = 0b00;
+  let Inst{23-22} = opc;
+  let Inst{21} = 0b0;
+  let Inst{20-12} = SImm9;
+  let Inst{11-10} = 0b10;
+  // Inherit Rn in 9-5
+  // Inherit Rt in 4-0
+}
+
+// Format for load-store (unscaled immediate) instructions.
+class A64I_LSunalimm<bits<2> size, bit v, bits<2> opc,
+                     dag outs, dag ins, string asmstr,
+                     list<dag> patterns, InstrItinClass itin>
+  : A64InstRtn<outs, ins, asmstr, patterns, itin> {
+  bits<9> SImm9;
+
+  let Inst{31-30} = size;
+  let Inst{29-27} = 0b111;
+  let Inst{26} = v;
+  let Inst{25-24} = 0b00;
+  let Inst{23-22} = opc;
+  let Inst{21} = 0b0;
+  let Inst{20-12} = SImm9;
+  let Inst{11-10} = 0b00;
+  // Inherit Rn in 9-5
+  // Inherit Rt in 4-0
+}
+
+
+// Format for load-store (unsigned immediate) instructions.
+class A64I_LSunsigimm<bits<2> size, bit v, bits<2> opc,
+                      dag outs, dag ins, string asmstr,
+                      list<dag> patterns, InstrItinClass itin>
+  : A64InstRtn<outs, ins, asmstr, patterns, itin> {
+  bits<12> UImm12;
+
+  let Inst{31-30} = size;
+  let Inst{29-27} = 0b111;
+  let Inst{26} = v;
+  let Inst{25-24} = 0b01;
+  let Inst{23-22} = opc;
+  let Inst{21-10} = UImm12;
+}
+
+// Format for load-store register (register offset) instructions.
+class A64I_LSregoff<bits<2> size, bit v, bits<2> opc, bit optionlo,
+                    dag outs, dag ins, string asmstr,
+                    list<dag> patterns, InstrItinClass itin>
+  : A64InstRtn<outs, ins, asmstr, patterns, itin> {
+  bits<5> Rm;
+
+  // Complex operand selection needed for these instructions, so they
+  // need an "addr" field for encoding/decoding to be generated.
+  bits<3> Ext;
+  // OptionHi = Ext{2-1}
+  // S = Ext{0}
+
+  let Inst{31-30} = size;
+  let Inst{29-27} = 0b111;
+  let Inst{26} = v;
+  let Inst{25-24} = 0b00;
+  let Inst{23-22} = opc;
+  let Inst{21} = 0b1;
+  let Inst{20-16} = Rm;
+  let Inst{15-14} = Ext{2-1};
+  let Inst{13} = optionlo;
+  let Inst{12} = Ext{0};
+  let Inst{11-10} = 0b10;
+  // Inherits Rn in 9-5
+  // Inherits Rt in 4-0
+
+  let AddedComplexity = 50;
+}
+
+// Format for Load-store register pair (offset) instructions
+class A64I_LSPoffset<bits<2> opc, bit v, bit l,
+                      dag outs, dag ins, string asmstr,
+                      list<dag> patterns, InstrItinClass itin>
+  : A64InstRtt2n<outs, ins, asmstr, patterns, itin> {
+  bits<7> SImm7;
+
+  let Inst{31-30} = opc;
+  let Inst{29-27} = 0b101;
+  let Inst{26} = v;
+  let Inst{25-23} = 0b010;
+  let Inst{22} = l;
+  let Inst{21-15} = SImm7;
+  // Inherit Rt2 in 14-10
+  // Inherit Rn in 9-5
+  // Inherit Rt in 4-0
+}
+
+// Format for Load-store register pair (post-indexed) instructions
+class A64I_LSPpostind<bits<2> opc, bit v, bit l,
+                      dag outs, dag ins, string asmstr,
+                      list<dag> patterns, InstrItinClass itin>
+  : A64InstRtt2n<outs, ins, asmstr, patterns, itin> {
+  bits<7> SImm7;
+
+  let Inst{31-30} = opc;
+  let Inst{29-27} = 0b101;
+  let Inst{26} = v;
+  let Inst{25-23} = 0b001;
+  let Inst{22} = l;
+  let Inst{21-15} = SImm7;
+  // Inherit Rt2 in 14-10
+  // Inherit Rn in 9-5
+  // Inherit Rt in 4-0
+}
+
+// Format for Load-store register pair (pre-indexed) instructions
+class A64I_LSPpreind<bits<2> opc, bit v, bit l,
+                      dag outs, dag ins, string asmstr,
+                      list<dag> patterns, InstrItinClass itin>
+  : A64InstRtt2n<outs, ins, asmstr, patterns, itin> {
+  bits<7> SImm7;
+
+  let Inst{31-30} = opc;
+  let Inst{29-27} = 0b101;
+  let Inst{26} = v;
+  let Inst{25-23} = 0b011;
+  let Inst{22} = l;
+  let Inst{21-15} = SImm7;
+  // Inherit Rt2 in 14-10
+  // Inherit Rn in 9-5
+  // Inherit Rt in 4-0
+}
+
+// Format for Load-store non-temporal register pair (offset) instructions
+class A64I_LSPnontemp<bits<2> opc, bit v, bit l,
+                      dag outs, dag ins, string asmstr,
+                      list<dag> patterns, InstrItinClass itin>
+  : A64InstRtt2n<outs, ins, asmstr, patterns, itin> {
+  bits<7> SImm7;
+
+  let Inst{31-30} = opc;
+  let Inst{29-27} = 0b101;
+  let Inst{26} = v;
+  let Inst{25-23} = 0b000;
+  let Inst{22} = l;
+  let Inst{21-15} = SImm7;
+  // Inherit Rt2 in 14-10
+  // Inherit Rn in 9-5
+  // Inherit Rt in 4-0
+}
+
+// Format for Logical (immediate) instructions
+class A64I_logicalimm<bit sf, bits<2> opc,
+                      dag outs, dag ins, string asmstr,
+                      list<dag> patterns, InstrItinClass itin>
+  : A64InstRdn<outs, ins, asmstr, patterns, itin> {
+  bit N;
+  bits<6> ImmR;
+  bits<6> ImmS;
+
+  // N, ImmR and ImmS have no separate existence in any assembly syntax (or for
+  // selection), so we'll combine them into a single field here.
+  bits<13> Imm;
+  // N = Imm{12};
+  // ImmR = Imm{11-6};
+  // ImmS = Imm{5-0};
+
+  let Inst{31} = sf;
+  let Inst{30-29} = opc;
+  let Inst{28-23} = 0b100100;
+  let Inst{22} = Imm{12};
+  let Inst{21-16} = Imm{11-6};
+  let Inst{15-10} = Imm{5-0};
+  // Rn inherited in 9-5
+  // Rd inherited in 4-0
+}
+
+// Format for Logical (shifted register) instructions
+class A64I_logicalshift<bit sf, bits<2> opc, bits<2> shift, bit N,
+                        dag outs, dag ins, string asmstr,
+                        list<dag> patterns, InstrItinClass itin>
+  : A64InstRdnm<outs, ins, asmstr, patterns, itin> {
+  bits<6> Imm6;
+
+  let Inst{31} = sf;
+  let Inst{30-29} = opc;
+  let Inst{28-24} = 0b01010;
+  let Inst{23-22} = shift;
+  let Inst{21} = N;
+  // Rm inherited
+  let Inst{15-10} = Imm6;
+  // Rn inherited
+  // Rd inherited
+}
+
+// Format for Move wide (immediate)
+class A64I_movw<bit sf, bits<2> opc,
+                dag outs, dag ins, string asmstr,
+                list<dag> patterns, InstrItinClass itin>
+  : A64InstRd<outs, ins, asmstr, patterns, itin> {
+  bits<16> UImm16;
+  bits<2> Shift; // Called "hw" officially
+
+  let Inst{31} = sf;
+  let Inst{30-29} = opc;
+  let Inst{28-23} = 0b100101;
+  let Inst{22-21} = Shift;
+  let Inst{20-5} = UImm16;
+  // Inherits Rd in 4-0
+}
+
+// Format for PC-relative addressing instructions, ADR and ADRP.
+class A64I_PCADR<bit op,
+                 dag outs, dag ins, string asmstr,
+                 list<dag> patterns, InstrItinClass itin>
+  : A64InstRd<outs, ins, asmstr, patterns, itin> {
+  bits<21> Label;
+
+  let Inst{31} = op;
+  let Inst{30-29} = Label{1-0};
+  let Inst{28-24} = 0b10000;
+  let Inst{23-5} = Label{20-2};
+}
+
+// Format for system instructions
+class A64I_system<bit l,
+                  dag outs, dag ins, string asmstr,
+                  list<dag> patterns, InstrItinClass itin>
+  : A64Inst<outs, ins, asmstr, patterns, itin> {
+  bits<2> Op0;
+  bits<3> Op1;
+  bits<4> CRn;
+  bits<4> CRm;
+  bits<3> Op2;
+  bits<5> Rt;
+
+  let Inst{31-22} = 0b1101010100;
+  let Inst{21} = l;
+  let Inst{20-19} = Op0;
+  let Inst{18-16} = Op1;
+  let Inst{15-12} = CRn;
+  let Inst{11-8} = CRm;
+  let Inst{7-5} = Op2;
+  let Inst{4-0} = Rt;
+
+  // These instructions can do horrible things.
+  let hasSideEffects = 1;
+}
+
+// Format for unconditional branch (immediate) instructions
+class A64I_Bimm<bit op,
+                dag outs, dag ins, string asmstr,
+                list<dag> patterns, InstrItinClass itin>
+  : A64Inst<outs, ins, asmstr, patterns, itin> {
+  // Doubly special in not even sharing register fields with other
+  // instructions, so we create our own Rn here.
+  bits<26> Label;
+
+  let Inst{31} = op;
+  let Inst{30-26} = 0b00101;
+  let Inst{25-0} = Label;
+}
+
+// Format for Test & branch (immediate) instructions
+class A64I_TBimm<bit op,
+                dag outs, dag ins, string asmstr,
+                list<dag> patterns, InstrItinClass itin>
+  : A64InstRt<outs, ins, asmstr, patterns, itin> {
+  // Doubly special in not even sharing register fields with other
+  // instructions, so we create our own Rn here.
+  bits<6> Imm;
+  bits<14> Label;
+
+  let Inst{31} = Imm{5};
+  let Inst{30-25} = 0b011011;
+  let Inst{24} = op;
+  let Inst{23-19} = Imm{4-0};
+  let Inst{18-5} = Label;
+  // Inherit Rt in 4-0
+}
+
+// Format for Unconditional branch (register) instructions, including
+// RET.  Shares no fields with instructions further up the hierarchy
+// so top-level.
+class A64I_Breg<bits<4> opc, bits<5> op2, bits<6> op3, bits<5> op4,
+                dag outs, dag ins, string asmstr,
+                list<dag> patterns, InstrItinClass itin>
+  : A64Inst<outs, ins, asmstr, patterns, itin> {
+  // Doubly special in not even sharing register fields with other
+  // instructions, so we create our own Rn here.
+  bits<5> Rn;
+
+  let Inst{31-25} = 0b1101011;
+  let Inst{24-21} = opc;
+  let Inst{20-16} = op2;
+  let Inst{15-10} = op3;
+  let Inst{9-5}   = Rn;
+  let Inst{4-0}   = op4;
+}
+
diff --git a/lib/Target/AArch64/AArch64InstrInfo.cpp b/lib/Target/AArch64/AArch64InstrInfo.cpp
new file mode 100644
index 000000000000..cf3a2c3707d9
--- /dev/null
+++ b/lib/Target/AArch64/AArch64InstrInfo.cpp
@@ -0,0 +1,822 @@
+//===- AArch64InstrInfo.cpp - AArch64 Instruction Information -------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the AArch64 implementation of the TargetInstrInfo class.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AArch64.h"
+#include "AArch64InstrInfo.h"
+#include "AArch64MachineFunctionInfo.h"
+#include "AArch64TargetMachine.h"
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/CodeGen/MachineConstantPool.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/Function.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/TargetRegistry.h"
+
+#include <algorithm>
+
+#define GET_INSTRINFO_CTOR
+#include "AArch64GenInstrInfo.inc"
+
+using namespace llvm;
+
+AArch64InstrInfo::AArch64InstrInfo(const AArch64Subtarget &STI)
+  : AArch64GenInstrInfo(AArch64::ADJCALLSTACKDOWN, AArch64::ADJCALLSTACKUP),
+    RI(*this, STI), Subtarget(STI) {}
+
+void AArch64InstrInfo::copyPhysReg(MachineBasicBlock &MBB,
+                                   MachineBasicBlock::iterator I, DebugLoc DL,
+                                   unsigned DestReg, unsigned SrcReg,
+                                   bool KillSrc) const {
+  unsigned Opc = 0;
+  unsigned ZeroReg = 0;
+  if (DestReg == AArch64::XSP || SrcReg == AArch64::XSP) {
+    // E.g. ADD xDst, xsp, #0 (, lsl #0)
+    BuildMI(MBB, I, DL, get(AArch64::ADDxxi_lsl0_s), DestReg)
+      .addReg(SrcReg)
+      .addImm(0);
+    return;
+  } else if (DestReg == AArch64::WSP || SrcReg == AArch64::WSP) {
+    // E.g. ADD wDST, wsp, #0 (, lsl #0)
+    BuildMI(MBB, I, DL, get(AArch64::ADDwwi_lsl0_s), DestReg)
+      .addReg(SrcReg)
+      .addImm(0);
+    return;
+  } else if (DestReg == AArch64::NZCV) {
+    assert(AArch64::GPR64RegClass.contains(SrcReg));
+    // E.g. MSR NZCV, xDST
+    BuildMI(MBB, I, DL, get(AArch64::MSRix))
+      .addImm(A64SysReg::NZCV)
+      .addReg(SrcReg);
+  } else if (SrcReg == AArch64::NZCV) {
+    assert(AArch64::GPR64RegClass.contains(DestReg));
+    // E.g. MRS xDST, NZCV
+    BuildMI(MBB, I, DL, get(AArch64::MRSxi), DestReg)
+      .addImm(A64SysReg::NZCV);
+  } else if (AArch64::GPR64RegClass.contains(DestReg)) {
+    assert(AArch64::GPR64RegClass.contains(SrcReg));
+    Opc = AArch64::ORRxxx_lsl;
+    ZeroReg = AArch64::XZR;
+  } else if (AArch64::GPR32RegClass.contains(DestReg)) {
+    assert(AArch64::GPR32RegClass.contains(SrcReg));
+    Opc = AArch64::ORRwww_lsl;
+    ZeroReg = AArch64::WZR;
+  } else if (AArch64::FPR32RegClass.contains(DestReg)) {
+    assert(AArch64::FPR32RegClass.contains(SrcReg));
+    BuildMI(MBB, I, DL, get(AArch64::FMOVss), DestReg)
+      .addReg(SrcReg);
+    return;
+  } else if (AArch64::FPR64RegClass.contains(DestReg)) {
+    assert(AArch64::FPR64RegClass.contains(SrcReg));
+    BuildMI(MBB, I, DL, get(AArch64::FMOVdd), DestReg)
+      .addReg(SrcReg);
+    return;
+  } else if (AArch64::FPR128RegClass.contains(DestReg)) {
+    assert(AArch64::FPR128RegClass.contains(SrcReg));
+
+    // FIXME: there's no good way to do this, at least without NEON:
+    //   + There's no single move instruction for q-registers
+    //   + We can't create a spill slot and use normal STR/LDR because stack
+    //     allocation has already happened
+    //   + We can't go via X-registers with FMOV because register allocation has
+    //     already happened.
+    // This may not be efficient, but at least it works.
+    BuildMI(MBB, I, DL, get(AArch64::LSFP128_PreInd_STR), AArch64::XSP)
+      .addReg(SrcReg)
+      .addReg(AArch64::XSP)
+      .addImm(0x1ff & -16);
+
+    BuildMI(MBB, I, DL, get(AArch64::LSFP128_PostInd_LDR), DestReg)
+      .addReg(AArch64::XSP, RegState::Define)
+      .addReg(AArch64::XSP)
+      .addImm(16);
+    return;
+  } else {
+    llvm_unreachable("Unknown register class in copyPhysReg");
+  }
+
+  // E.g. ORR xDst, xzr, xSrc, lsl #0
+  BuildMI(MBB, I, DL, get(Opc), DestReg)
+    .addReg(ZeroReg)
+    .addReg(SrcReg)
+    .addImm(0);
+}
+
+MachineInstr *
+AArch64InstrInfo::emitFrameIndexDebugValue(MachineFunction &MF, int FrameIx,
+                                           uint64_t Offset, const MDNode *MDPtr,
+                                           DebugLoc DL) const {
+  MachineInstrBuilder MIB = BuildMI(MF, DL, get(AArch64::DBG_VALUE))
+    .addFrameIndex(FrameIx).addImm(0)
+    .addImm(Offset)
+    .addMetadata(MDPtr);
+  return &*MIB;
+}
+
+/// Does the Opcode represent a conditional branch that we can remove and re-add
+/// at the end of a basic block?
+static bool isCondBranch(unsigned Opc) {
+  return Opc == AArch64::Bcc || Opc == AArch64::CBZw || Opc == AArch64::CBZx ||
+         Opc == AArch64::CBNZw || Opc == AArch64::CBNZx ||
+         Opc == AArch64::TBZwii || Opc == AArch64::TBZxii ||
+         Opc == AArch64::TBNZwii || Opc == AArch64::TBNZxii;
+}
+
+/// Takes apart a given conditional branch MachineInstr (see isCondBranch),
+/// setting TBB to the destination basic block and populating the Cond vector
+/// with data necessary to recreate the conditional branch at a later
+/// date. First element will be the opcode, and subsequent ones define the
+/// conditions being branched on in an instruction-specific manner.
+static void classifyCondBranch(MachineInstr *I, MachineBasicBlock *&TBB,
+                               SmallVectorImpl<MachineOperand> &Cond) {
+  switch(I->getOpcode()) {
+  case AArch64::Bcc:
+  case AArch64::CBZw:
+  case AArch64::CBZx:
+  case AArch64::CBNZw:
+  case AArch64::CBNZx:
+    // These instructions just have one predicate operand in position 0 (either
+    // a condition code or a register being compared).
+    Cond.push_back(MachineOperand::CreateImm(I->getOpcode()));
+    Cond.push_back(I->getOperand(0));
+    TBB = I->getOperand(1).getMBB();
+    return;
+  case AArch64::TBZwii:
+  case AArch64::TBZxii:
+  case AArch64::TBNZwii:
+  case AArch64::TBNZxii:
+    // These have two predicate operands: a register and a bit position.
+    Cond.push_back(MachineOperand::CreateImm(I->getOpcode()));
+    Cond.push_back(I->getOperand(0));
+    Cond.push_back(I->getOperand(1));
+    TBB = I->getOperand(2).getMBB();
+    return;
+  default:
+    llvm_unreachable("Unknown conditional branch to classify");
+  }
+}
+
+
+bool
+AArch64InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,MachineBasicBlock *&TBB,
+                                MachineBasicBlock *&FBB,
+                                SmallVectorImpl<MachineOperand> &Cond,
+                                bool AllowModify) const {
+  // If the block has no terminators, it just falls into the block after it.
+  MachineBasicBlock::iterator I = MBB.end();
+  if (I == MBB.begin())
+    return false;
+  --I;
+  while (I->isDebugValue()) {
+    if (I == MBB.begin())
+      return false;
+    --I;
+  }
+  if (!isUnpredicatedTerminator(I))
+    return false;
+
+  // Get the last instruction in the block.
+  MachineInstr *LastInst = I;
+
+  // If there is only one terminator instruction, process it.
+  unsigned LastOpc = LastInst->getOpcode();
+  if (I == MBB.begin() || !isUnpredicatedTerminator(--I)) {
+    if (LastOpc == AArch64::Bimm) {
+      TBB = LastInst->getOperand(0).getMBB();
+      return false;
+    }
+    if (isCondBranch(LastOpc)) {
+      classifyCondBranch(LastInst, TBB, Cond);
+      return false;
+    }
+    return true;  // Can't handle indirect branch.
+  }
+
+  // Get the instruction before it if it is a terminator.
+  MachineInstr *SecondLastInst = I;
+  unsigned SecondLastOpc = SecondLastInst->getOpcode();
+
+  // If AllowModify is true and the block ends with two or more unconditional
+  // branches, delete all but the first unconditional branch.
+  if (AllowModify && LastOpc == AArch64::Bimm) {
+    while (SecondLastOpc == AArch64::Bimm) {
+      LastInst->eraseFromParent();
+      LastInst = SecondLastInst;
+      LastOpc = LastInst->getOpcode();
+      if (I == MBB.begin() || !isUnpredicatedTerminator(--I)) {
+        // Return now the only terminator is an unconditional branch.
+        TBB = LastInst->getOperand(0).getMBB();
+        return false;
+      } else {
+        SecondLastInst = I;
+        SecondLastOpc = SecondLastInst->getOpcode();
+      }
+    }
+  }
+
+  // If there are three terminators, we don't know what sort of block this is.
+  if (SecondLastInst && I != MBB.begin() && isUnpredicatedTerminator(--I))
+    return true;
+
+  // If the block ends with a B and a Bcc, handle it.
+  if (LastOpc == AArch64::Bimm) {
+    if (SecondLastOpc == AArch64::Bcc) {
+      TBB =  SecondLastInst->getOperand(1).getMBB();
+      Cond.push_back(MachineOperand::CreateImm(AArch64::Bcc));
+      Cond.push_back(SecondLastInst->getOperand(0));
+      FBB = LastInst->getOperand(0).getMBB();
+      return false;
+    } else if (isCondBranch(SecondLastOpc)) {
+      classifyCondBranch(SecondLastInst, TBB, Cond);
+      FBB = LastInst->getOperand(0).getMBB();
+      return false;
+    }
+  }
+
+  // If the block ends with two unconditional branches, handle it.  The second
+  // one is not executed, so remove it.
+  if (SecondLastOpc == AArch64::Bimm && LastOpc == AArch64::Bimm) {
+    TBB = SecondLastInst->getOperand(0).getMBB();
+    I = LastInst;
+    if (AllowModify)
+      I->eraseFromParent();
+    return false;
+  }
+
+  // Otherwise, can't handle this.
+  return true;
+}
+
+bool AArch64InstrInfo::ReverseBranchCondition(
+                                  SmallVectorImpl<MachineOperand> &Cond) const {
+  switch (Cond[0].getImm()) {
+  case AArch64::Bcc: {
+    A64CC::CondCodes CC = static_cast<A64CC::CondCodes>(Cond[1].getImm());
+    CC = A64InvertCondCode(CC);
+    Cond[1].setImm(CC);
+    return false;
+  }
+  case AArch64::CBZw:
+    Cond[0].setImm(AArch64::CBNZw);
+    return false;
+  case AArch64::CBZx:
+    Cond[0].setImm(AArch64::CBNZx);
+    return false;
+  case AArch64::CBNZw:
+    Cond[0].setImm(AArch64::CBZw);
+    return false;
+  case AArch64::CBNZx:
+    Cond[0].setImm(AArch64::CBZx);
+    return false;
+  case AArch64::TBZwii:
+    Cond[0].setImm(AArch64::TBNZwii);
+    return false;
+  case AArch64::TBZxii:
+    Cond[0].setImm(AArch64::TBNZxii);
+    return false;
+  case AArch64::TBNZwii:
+    Cond[0].setImm(AArch64::TBZwii);
+    return false;
+  case AArch64::TBNZxii:
+    Cond[0].setImm(AArch64::TBZxii);
+    return false;
+  default:
+    llvm_unreachable("Unknown branch type");
+  }
+}
+
+
+unsigned
+AArch64InstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
+                               MachineBasicBlock *FBB,
+                               const SmallVectorImpl<MachineOperand> &Cond,
+                               DebugLoc DL) const {
+  if (FBB == 0 && Cond.empty()) {
+    BuildMI(&MBB, DL, get(AArch64::Bimm)).addMBB(TBB);
+    return 1;
+  } else if (FBB == 0) {
+    MachineInstrBuilder MIB = BuildMI(&MBB, DL, get(Cond[0].getImm()));
+    for (int i = 1, e = Cond.size(); i != e; ++i)
+      MIB.addOperand(Cond[i]);
+    MIB.addMBB(TBB);
+    return 1;
+  }
+
+  MachineInstrBuilder MIB = BuildMI(&MBB, DL, get(Cond[0].getImm()));
+  for (int i = 1, e = Cond.size(); i != e; ++i)
+    MIB.addOperand(Cond[i]);
+  MIB.addMBB(TBB);
+
+  BuildMI(&MBB, DL, get(AArch64::Bimm)).addMBB(FBB);
+  return 2;
+}
+
+unsigned AArch64InstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
+  MachineBasicBlock::iterator I = MBB.end();
+  if (I == MBB.begin()) return 0;
+  --I;
+  while (I->isDebugValue()) {
+    if (I == MBB.begin())
+      return 0;
+    --I;
+  }
+  if (I->getOpcode() != AArch64::Bimm && !isCondBranch(I->getOpcode()))
+    return 0;
+
+  // Remove the branch.
+  I->eraseFromParent();
+
+  I = MBB.end();
+
+  if (I == MBB.begin()) return 1;
+  --I;
+  if (!isCondBranch(I->getOpcode()))
+    return 1;
+
+  // Remove the branch.
+  I->eraseFromParent();
+  return 2;
+}
+
+bool
+AArch64InstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MBBI) const {
+  MachineInstr &MI = *MBBI;
+  MachineBasicBlock &MBB = *MI.getParent();
+
+  unsigned Opcode = MI.getOpcode();
+  switch (Opcode) {
+  case AArch64::TLSDESC_BLRx: {
+    MachineInstr *NewMI =
+      BuildMI(MBB, MBBI, MI.getDebugLoc(), get(AArch64::TLSDESCCALL))
+        .addOperand(MI.getOperand(1));
+    MI.setDesc(get(AArch64::BLRx));
+
+    llvm::finalizeBundle(MBB, NewMI, *++MBBI);
+    return true;
+    }
+  default:
+    return false;
+  }
+
+  return false;
+}
+
+void
+AArch64InstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
+                                      MachineBasicBlock::iterator MBBI,
+                                      unsigned SrcReg, bool isKill,
+                                      int FrameIdx,
+                                      const TargetRegisterClass *RC,
+                                      const TargetRegisterInfo *TRI) const {
+  DebugLoc DL = MBB.findDebugLoc(MBBI);
+  MachineFunction &MF = *MBB.getParent();
+  MachineFrameInfo &MFI = *MF.getFrameInfo();
+  unsigned Align = MFI.getObjectAlignment(FrameIdx);
+
+  MachineMemOperand *MMO
+    = MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FrameIdx),
+                              MachineMemOperand::MOStore,
+                              MFI.getObjectSize(FrameIdx),
+                              Align);
+
+  unsigned StoreOp = 0;
+  if (RC->hasType(MVT::i64) || RC->hasType(MVT::i32)) {
+    switch(RC->getSize()) {
+    case 4: StoreOp = AArch64::LS32_STR; break;
+    case 8: StoreOp = AArch64::LS64_STR; break;
+    default:
+      llvm_unreachable("Unknown size for regclass");
+    }
+  } else {
+    assert((RC->hasType(MVT::f32) || RC->hasType(MVT::f64) ||
+            RC->hasType(MVT::f128))
+           && "Expected integer or floating type for store");
+    switch (RC->getSize()) {
+    case 4: StoreOp = AArch64::LSFP32_STR; break;
+    case 8: StoreOp = AArch64::LSFP64_STR; break;
+    case 16: StoreOp = AArch64::LSFP128_STR; break;
+    default:
+      llvm_unreachable("Unknown size for regclass");
+    }
+  }
+
+  MachineInstrBuilder NewMI = BuildMI(MBB, MBBI, DL, get(StoreOp));
+  NewMI.addReg(SrcReg, getKillRegState(isKill))
+    .addFrameIndex(FrameIdx)
+    .addImm(0)
+    .addMemOperand(MMO);
+
+}
+
+void
+AArch64InstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
+                                       MachineBasicBlock::iterator MBBI,
+                                       unsigned DestReg, int FrameIdx,
+                                       const TargetRegisterClass *RC,
+                                       const TargetRegisterInfo *TRI) const {
+  DebugLoc DL = MBB.findDebugLoc(MBBI);
+  MachineFunction &MF = *MBB.getParent();
+  MachineFrameInfo &MFI = *MF.getFrameInfo();
+  unsigned Align = MFI.getObjectAlignment(FrameIdx);
+
+  MachineMemOperand *MMO
+    = MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FrameIdx),
+                              MachineMemOperand::MOLoad,
+                              MFI.getObjectSize(FrameIdx),
+                              Align);
+
+  unsigned LoadOp = 0;
+  if (RC->hasType(MVT::i64) || RC->hasType(MVT::i32)) {
+    switch(RC->getSize()) {
+    case 4: LoadOp = AArch64::LS32_LDR; break;
+    case 8: LoadOp = AArch64::LS64_LDR; break;
+    default:
+      llvm_unreachable("Unknown size for regclass");
+    }
+  } else {
+    assert((RC->hasType(MVT::f32) || RC->hasType(MVT::f64)
+            || RC->hasType(MVT::f128))
+           && "Expected integer or floating type for store");
+    switch (RC->getSize()) {
+    case 4: LoadOp = AArch64::LSFP32_LDR; break;
+    case 8: LoadOp = AArch64::LSFP64_LDR; break;
+    case 16: LoadOp = AArch64::LSFP128_LDR; break;
+    default:
+      llvm_unreachable("Unknown size for regclass");
+    }
+  }
+
+  MachineInstrBuilder NewMI = BuildMI(MBB, MBBI, DL, get(LoadOp), DestReg);
+  NewMI.addFrameIndex(FrameIdx)
+       .addImm(0)
+       .addMemOperand(MMO);
+}
+
+unsigned AArch64InstrInfo::estimateRSStackLimit(MachineFunction &MF) const {
+  unsigned Limit = (1 << 16) - 1;
+  for (MachineFunction::iterator BB = MF.begin(),E = MF.end(); BB != E; ++BB) {
+    for (MachineBasicBlock::iterator I = BB->begin(), E = BB->end();
+         I != E; ++I) {
+      for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
+        if (!I->getOperand(i).isFI()) continue;
+
+        // When using ADDxxi_lsl0_s to get the address of a stack object, 0xfff
+        // is the largest offset guaranteed to fit in the immediate offset.
+        if (I->getOpcode() == AArch64::ADDxxi_lsl0_s) {
+          Limit = std::min(Limit, 0xfffu);
+          break;
+        }
+
+        int AccessScale, MinOffset, MaxOffset;
+        getAddressConstraints(*I, AccessScale, MinOffset, MaxOffset);
+        Limit = std::min(Limit, static_cast<unsigned>(MaxOffset));
+
+        break; // At most one FI per instruction
+      }
+    }
+  }
+
+  return Limit;
+}
+void AArch64InstrInfo::getAddressConstraints(const MachineInstr &MI,
+                                             int &AccessScale, int &MinOffset,
+                                             int &MaxOffset) const {
+  switch (MI.getOpcode()) {
+  default: llvm_unreachable("Unkown load/store kind");
+  case TargetOpcode::DBG_VALUE:
+    AccessScale = 1;
+    MinOffset = INT_MIN;
+    MaxOffset = INT_MAX;
+    return;
+  case AArch64::LS8_LDR: case AArch64::LS8_STR:
+  case AArch64::LSFP8_LDR: case AArch64::LSFP8_STR:
+  case AArch64::LDRSBw:
+  case AArch64::LDRSBx:
+    AccessScale = 1;
+    MinOffset = 0;
+    MaxOffset = 0xfff;
+    return;
+  case AArch64::LS16_LDR: case AArch64::LS16_STR:
+  case AArch64::LSFP16_LDR: case AArch64::LSFP16_STR:
+  case AArch64::LDRSHw:
+  case AArch64::LDRSHx:
+    AccessScale = 2;
+    MinOffset = 0;
+    MaxOffset = 0xfff * AccessScale;
+    return;
+  case AArch64::LS32_LDR:  case AArch64::LS32_STR:
+  case AArch64::LSFP32_LDR: case AArch64::LSFP32_STR:
+  case AArch64::LDRSWx:
+  case AArch64::LDPSWx:
+    AccessScale = 4;
+    MinOffset = 0;
+    MaxOffset = 0xfff * AccessScale;
+    return;
+  case AArch64::LS64_LDR: case AArch64::LS64_STR:
+  case AArch64::LSFP64_LDR: case AArch64::LSFP64_STR:
+  case AArch64::PRFM:
+    AccessScale = 8;
+    MinOffset = 0;
+    MaxOffset = 0xfff * AccessScale;
+    return;
+  case AArch64::LSFP128_LDR: case AArch64::LSFP128_STR:
+    AccessScale = 16;
+    MinOffset = 0;
+    MaxOffset = 0xfff * AccessScale;
+    return;
+  case AArch64::LSPair32_LDR: case AArch64::LSPair32_STR:
+  case AArch64::LSFPPair32_LDR: case AArch64::LSFPPair32_STR:
+    AccessScale = 4;
+    MinOffset = -0x40 * AccessScale;
+    MaxOffset = 0x3f * AccessScale;
+    return;
+  case AArch64::LSPair64_LDR: case AArch64::LSPair64_STR:
+  case AArch64::LSFPPair64_LDR: case AArch64::LSFPPair64_STR:
+    AccessScale = 8;
+    MinOffset = -0x40 * AccessScale;
+    MaxOffset = 0x3f * AccessScale;
+    return;
+  case AArch64::LSFPPair128_LDR: case AArch64::LSFPPair128_STR:
+    AccessScale = 16;
+    MinOffset = -0x40 * AccessScale;
+    MaxOffset = 0x3f * AccessScale;
+    return;
+  }
+}
+
+unsigned AArch64InstrInfo::getInstSizeInBytes(const MachineInstr &MI) const {
+  const MCInstrDesc &MCID = MI.getDesc();
+  const MachineBasicBlock &MBB = *MI.getParent();
+  const MachineFunction &MF = *MBB.getParent();
+  const MCAsmInfo &MAI = *MF.getTarget().getMCAsmInfo();
+
+  if (MCID.getSize())
+    return MCID.getSize();
+
+  if (MI.getOpcode() == AArch64::INLINEASM)
+    return getInlineAsmLength(MI.getOperand(0).getSymbolName(), MAI);
+
+  if (MI.isLabel())
+    return 0;
+
+  switch (MI.getOpcode()) {
+  case TargetOpcode::BUNDLE:
+    return getInstBundleLength(MI);
+  case TargetOpcode::IMPLICIT_DEF:
+  case TargetOpcode::KILL:
+  case TargetOpcode::PROLOG_LABEL:
+  case TargetOpcode::EH_LABEL:
+  case TargetOpcode::DBG_VALUE:
+    return 0;
+  case AArch64::TLSDESCCALL:
+    return 0;
+  default:
+    llvm_unreachable("Unknown instruction class");
+  }
+}
+
+unsigned AArch64InstrInfo::getInstBundleLength(const MachineInstr &MI) const {
+  unsigned Size = 0;
+  MachineBasicBlock::const_instr_iterator I = MI;
+  MachineBasicBlock::const_instr_iterator E = MI.getParent()->instr_end();
+  while (++I != E && I->isInsideBundle()) {
+    assert(!I->isBundle() && "No nested bundle!");
+    Size += getInstSizeInBytes(*I);
+  }
+  return Size;
+}
+
+bool llvm::rewriteA64FrameIndex(MachineInstr &MI, unsigned FrameRegIdx,
+                                unsigned FrameReg, int &Offset,
+                                const AArch64InstrInfo &TII) {
+  MachineBasicBlock &MBB = *MI.getParent();
+  MachineFunction &MF = *MBB.getParent();
+  MachineFrameInfo &MFI = *MF.getFrameInfo();
+
+  MFI.getObjectOffset(FrameRegIdx);
+  llvm_unreachable("Unimplemented rewriteFrameIndex");
+}
+
+void llvm::emitRegUpdate(MachineBasicBlock &MBB,
+                         MachineBasicBlock::iterator MBBI,
+                         DebugLoc dl, const TargetInstrInfo &TII,
+                         unsigned DstReg, unsigned SrcReg, unsigned ScratchReg,
+                         int64_t NumBytes, MachineInstr::MIFlag MIFlags) {
+  if (NumBytes == 0 && DstReg == SrcReg)
+    return;
+  else if (abs64(NumBytes) & ~0xffffff) {
+    // Generically, we have to materialize the offset into a temporary register
+    // and subtract it. There are a couple of ways this could be done, for now
+    // we'll use a movz/movk or movn/movk sequence.
+    uint64_t Bits = static_cast<uint64_t>(abs64(NumBytes));
+    BuildMI(MBB, MBBI, dl, TII.get(AArch64::MOVZxii), ScratchReg)
+      .addImm(0xffff & Bits).addImm(0)
+      .setMIFlags(MIFlags);
+
+    Bits >>= 16;
+    if (Bits & 0xffff) {
+      BuildMI(MBB, MBBI, dl, TII.get(AArch64::MOVKxii), ScratchReg)
+        .addReg(ScratchReg)
+        .addImm(0xffff & Bits).addImm(1)
+        .setMIFlags(MIFlags);
+    }
+
+    Bits >>= 16;
+    if (Bits & 0xffff) {
+      BuildMI(MBB, MBBI, dl, TII.get(AArch64::MOVKxii), ScratchReg)
+        .addReg(ScratchReg)
+        .addImm(0xffff & Bits).addImm(2)
+        .setMIFlags(MIFlags);
+    }
+
+    Bits >>= 16;
+    if (Bits & 0xffff) {
+      BuildMI(MBB, MBBI, dl, TII.get(AArch64::MOVKxii), ScratchReg)
+        .addReg(ScratchReg)
+        .addImm(0xffff & Bits).addImm(3)
+        .setMIFlags(MIFlags);
+    }
+
+    // ADD DST, SRC, xTMP (, lsl #0)
+    unsigned AddOp = NumBytes > 0 ? AArch64::ADDxxx_uxtx : AArch64::SUBxxx_uxtx;
+    BuildMI(MBB, MBBI, dl, TII.get(AddOp), DstReg)
+      .addReg(SrcReg, RegState::Kill)
+      .addReg(ScratchReg, RegState::Kill)
+      .addImm(0)
+      .setMIFlag(MIFlags);
+    return;
+  }
+
+  // Now we know that the adjustment can be done in at most two add/sub
+  // (immediate) instructions, which is always more efficient than a
+  // literal-pool load, or even a hypothetical movz/movk/add sequence
+
+  // Decide whether we're doing addition or subtraction
+  unsigned LowOp, HighOp;
+  if (NumBytes >= 0) {
+    LowOp = AArch64::ADDxxi_lsl0_s;
+    HighOp = AArch64::ADDxxi_lsl12_s;
+  } else {
+    LowOp = AArch64::SUBxxi_lsl0_s;
+    HighOp = AArch64::SUBxxi_lsl12_s;
+    NumBytes = abs64(NumBytes);
+  }
+
+  // If we're here, at the very least a move needs to be produced, which just
+  // happens to be materializable by an ADD.
+  if ((NumBytes & 0xfff) || NumBytes == 0) {
+    BuildMI(MBB, MBBI, dl, TII.get(LowOp), DstReg)
+      .addReg(SrcReg, RegState::Kill)
+      .addImm(NumBytes & 0xfff)
+      .setMIFlag(MIFlags);
+
+    // Next update should use the register we've just defined.
+    SrcReg = DstReg;
+  }
+
+  if (NumBytes & 0xfff000) {
+    BuildMI(MBB, MBBI, dl, TII.get(HighOp), DstReg)
+      .addReg(SrcReg, RegState::Kill)
+      .addImm(NumBytes >> 12)
+      .setMIFlag(MIFlags);
+  }
+}
+
+void llvm::emitSPUpdate(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
+                        DebugLoc dl, const TargetInstrInfo &TII,
+                        unsigned ScratchReg, int64_t NumBytes,
+                        MachineInstr::MIFlag MIFlags) {
+  emitRegUpdate(MBB, MI, dl, TII, AArch64::XSP, AArch64::XSP, AArch64::X16,
+                NumBytes, MIFlags);
+}
+
+
+namespace {
+  struct LDTLSCleanup : public MachineFunctionPass {
+    static char ID;
+    LDTLSCleanup() : MachineFunctionPass(ID) {}
+
+    virtual bool runOnMachineFunction(MachineFunction &MF) {
+      AArch64MachineFunctionInfo* MFI
+        = MF.getInfo<AArch64MachineFunctionInfo>();
+      if (MFI->getNumLocalDynamicTLSAccesses() < 2) {
+        // No point folding accesses if there isn't at least two.
+        return false;
+      }
+
+      MachineDominatorTree *DT = &getAnalysis<MachineDominatorTree>();
+      return VisitNode(DT->getRootNode(), 0);
+    }
+
+    // Visit the dominator subtree rooted at Node in pre-order.
+    // If TLSBaseAddrReg is non-null, then use that to replace any
+    // TLS_base_addr instructions. Otherwise, create the register
+    // when the first such instruction is seen, and then use it
+    // as we encounter more instructions.
+    bool VisitNode(MachineDomTreeNode *Node, unsigned TLSBaseAddrReg) {
+      MachineBasicBlock *BB = Node->getBlock();
+      bool Changed = false;
+
+      // Traverse the current block.
+      for (MachineBasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;
+           ++I) {
+        switch (I->getOpcode()) {
+        case AArch64::TLSDESC_BLRx:
+          // Make sure it's a local dynamic access.
+          if (!I->getOperand(1).isSymbol() ||
+              strcmp(I->getOperand(1).getSymbolName(), "_TLS_MODULE_BASE_"))
+            break;
+
+          if (TLSBaseAddrReg)
+            I = ReplaceTLSBaseAddrCall(I, TLSBaseAddrReg);
+          else
+            I = SetRegister(I, &TLSBaseAddrReg);
+          Changed = true;
+          break;
+        default:
+          break;
+        }
+      }
+
+      // Visit the children of this block in the dominator tree.
+      for (MachineDomTreeNode::iterator I = Node->begin(), E = Node->end();
+           I != E; ++I) {
+        Changed |= VisitNode(*I, TLSBaseAddrReg);
+      }
+
+      return Changed;
+    }
+
+    // Replace the TLS_base_addr instruction I with a copy from
+    // TLSBaseAddrReg, returning the new instruction.
+    MachineInstr *ReplaceTLSBaseAddrCall(MachineInstr *I,
+                                         unsigned TLSBaseAddrReg) {
+      MachineFunction *MF = I->getParent()->getParent();
+      const AArch64TargetMachine *TM =
+          static_cast<const AArch64TargetMachine *>(&MF->getTarget());
+      const AArch64InstrInfo *TII = TM->getInstrInfo();
+
+      // Insert a Copy from TLSBaseAddrReg to x0, which is where the rest of the
+      // code sequence assumes the address will be.
+      MachineInstr *Copy = BuildMI(*I->getParent(), I, I->getDebugLoc(),
+                                   TII->get(TargetOpcode::COPY),
+                                   AArch64::X0)
+        .addReg(TLSBaseAddrReg);
+
+      // Erase the TLS_base_addr instruction.
+      I->eraseFromParent();
+
+      return Copy;
+    }
+
+    // Create a virtal register in *TLSBaseAddrReg, and populate it by
+    // inserting a copy instruction after I. Returns the new instruction.
+    MachineInstr *SetRegister(MachineInstr *I, unsigned *TLSBaseAddrReg) {
+      MachineFunction *MF = I->getParent()->getParent();
+      const AArch64TargetMachine *TM =
+          static_cast<const AArch64TargetMachine *>(&MF->getTarget());
+      const AArch64InstrInfo *TII = TM->getInstrInfo();
+
+      // Create a virtual register for the TLS base address.
+      MachineRegisterInfo &RegInfo = MF->getRegInfo();
+      *TLSBaseAddrReg = RegInfo.createVirtualRegister(&AArch64::GPR64RegClass);
+
+      // Insert a copy from X0 to TLSBaseAddrReg for later.
+      MachineInstr *Next = I->getNextNode();
+      MachineInstr *Copy = BuildMI(*I->getParent(), Next, I->getDebugLoc(),
+                                   TII->get(TargetOpcode::COPY),
+                                   *TLSBaseAddrReg)
+        .addReg(AArch64::X0);
+
+      return Copy;
+    }
+
+    virtual const char *getPassName() const {
+      return "Local Dynamic TLS Access Clean-up";
+    }
+
+    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+      AU.setPreservesCFG();
+      AU.addRequired<MachineDominatorTree>();
+      MachineFunctionPass::getAnalysisUsage(AU);
+    }
+  };
+}
+
+char LDTLSCleanup::ID = 0;
+FunctionPass*
+llvm::createAArch64CleanupLocalDynamicTLSPass() { return new LDTLSCleanup(); }
diff --git a/lib/Target/AArch64/AArch64InstrInfo.h b/lib/Target/AArch64/AArch64InstrInfo.h
new file mode 100644
index 000000000000..22a2ab4cf60a
--- /dev/null
+++ b/lib/Target/AArch64/AArch64InstrInfo.h
@@ -0,0 +1,112 @@
+//===- AArch64InstrInfo.h - AArch64 Instruction Information -----*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the AArch64 implementation of the TargetInstrInfo class.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_TARGET_AARCH64INSTRINFO_H
+#define LLVM_TARGET_AARCH64INSTRINFO_H
+
+#include "llvm/Target/TargetInstrInfo.h"
+#include "AArch64RegisterInfo.h"
+
+#define GET_INSTRINFO_HEADER
+#include "AArch64GenInstrInfo.inc"
+
+namespace llvm {
+
+class AArch64Subtarget;
+
+class AArch64InstrInfo : public AArch64GenInstrInfo {
+  const AArch64RegisterInfo RI;
+  const AArch64Subtarget &Subtarget;
+public:
+  explicit AArch64InstrInfo(const AArch64Subtarget &TM);
+
+  /// getRegisterInfo - TargetInstrInfo is a superset of MRegister info.  As
+  /// such, whenever a client has an instance of instruction info, it should
+  /// always be able to get register info as well (through this method).
+  ///
+  const TargetRegisterInfo &getRegisterInfo() const { return RI; }
+
+  const AArch64Subtarget &getSubTarget() const { return Subtarget; }
+
+  void copyPhysReg(MachineBasicBlock &MBB,
+                   MachineBasicBlock::iterator I, DebugLoc DL,
+                   unsigned DestReg, unsigned SrcReg,
+                   bool KillSrc) const;
+
+  MachineInstr *emitFrameIndexDebugValue(MachineFunction &MF, int FrameIx,
+                                         uint64_t Offset, const MDNode *MDPtr,
+                                         DebugLoc DL) const;
+
+  void storeRegToStackSlot(MachineBasicBlock &MBB,
+                           MachineBasicBlock::iterator MI,
+                           unsigned SrcReg, bool isKill, int FrameIndex,
+                           const TargetRegisterClass *RC,
+                           const TargetRegisterInfo *TRI) const;
+  void loadRegFromStackSlot(MachineBasicBlock &MBB,
+                            MachineBasicBlock::iterator MBBI,
+                            unsigned DestReg, int FrameIdx,
+                            const TargetRegisterClass *RC,
+                            const TargetRegisterInfo *TRI) const;
+
+  bool AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
+                     MachineBasicBlock *&FBB,
+                     SmallVectorImpl<MachineOperand> &Cond,
+                     bool AllowModify = false) const;
+  unsigned InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
+                        MachineBasicBlock *FBB,
+                        const SmallVectorImpl<MachineOperand> &Cond,
+                        DebugLoc DL) const;
+  unsigned RemoveBranch(MachineBasicBlock &MBB) const;
+  bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const;
+
+  bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const;
+
+  /// Look through the instructions in this function and work out the largest
+  /// the stack frame can be while maintaining the ability to address local
+  /// slots with no complexities.
+  unsigned estimateRSStackLimit(MachineFunction &MF) const;
+
+  /// getAddressConstraints - For loads and stores (and PRFMs) taking an
+  /// immediate offset, this function determines the constraints required for
+  /// the immediate. It must satisfy:
+  ///    + MinOffset <= imm <= MaxOffset
+  ///    + imm % OffsetScale == 0
+  void getAddressConstraints(const MachineInstr &MI, int &AccessScale,
+                             int &MinOffset, int &MaxOffset) const;
+
+
+  unsigned getInstSizeInBytes(const MachineInstr &MI) const;
+
+  unsigned getInstBundleLength(const MachineInstr &MI) const;
+
+};
+
+bool rewriteA64FrameIndex(MachineInstr &MI, unsigned FrameRegIdx,
+                          unsigned FrameReg, int &Offset,
+                          const AArch64InstrInfo &TII);
+
+
+void emitRegUpdate(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
+                   DebugLoc dl, const TargetInstrInfo &TII,
+                   unsigned DstReg, unsigned SrcReg, unsigned ScratchReg,
+                   int64_t NumBytes,
+                   MachineInstr::MIFlag MIFlags = MachineInstr::NoFlags);
+
+void emitSPUpdate(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
+                  DebugLoc dl, const TargetInstrInfo &TII,
+                  unsigned ScratchReg, int64_t NumBytes,
+                  MachineInstr::MIFlag MIFlags = MachineInstr::NoFlags);
+
+}
+
+#endif
diff --git a/lib/Target/AArch64/AArch64InstrInfo.td b/lib/Target/AArch64/AArch64InstrInfo.td
new file mode 100644
index 000000000000..37be5e4892e4
--- /dev/null
+++ b/lib/Target/AArch64/AArch64InstrInfo.td
@@ -0,0 +1,5099 @@
+//===----- AArch64InstrInfo.td - AArch64 Instruction Info ----*- tablegen -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file describes the AArch64 scalar instructions in TableGen format.
+//
+//===----------------------------------------------------------------------===//
+
+include "AArch64InstrFormats.td"
+
+//===----------------------------------------------------------------------===//
+// Target-specific ISD nodes and profiles
+//===----------------------------------------------------------------------===//
+
+def SDT_A64ret : SDTypeProfile<0, 0, []>;
+def A64ret : SDNode<"AArch64ISD::Ret", SDT_A64ret, [SDNPHasChain,
+                                                    SDNPOptInGlue,
+                                                    SDNPVariadic]>;
+
+// (ins NZCV, Condition, Dest)
+def SDT_A64br_cc : SDTypeProfile<0, 3, [SDTCisVT<0, i32>]>;
+def A64br_cc : SDNode<"AArch64ISD::BR_CC", SDT_A64br_cc, [SDNPHasChain]>;
+
+// (outs Result), (ins NZCV, IfTrue, IfFalse, Condition)
+def SDT_A64select_cc : SDTypeProfile<1, 4, [SDTCisVT<1, i32>,
+                                            SDTCisSameAs<0, 2>,
+                                            SDTCisSameAs<2, 3>]>;
+def A64select_cc : SDNode<"AArch64ISD::SELECT_CC", SDT_A64select_cc>;
+
+// (outs NZCV), (ins LHS, RHS, Condition)
+def SDT_A64setcc : SDTypeProfile<1, 3, [SDTCisVT<0, i32>,
+                                        SDTCisSameAs<1, 2>]>;
+def A64setcc : SDNode<"AArch64ISD::SETCC", SDT_A64setcc>;
+
+
+// (outs GPR64), (ins)
+def A64threadpointer : SDNode<"AArch64ISD::THREAD_POINTER", SDTPtrLeaf>;
+
+// A64 compares don't care about the cond really (they set all flags) so a
+// simple binary operator is useful.
+def A64cmp : PatFrag<(ops node:$lhs, node:$rhs),
+                     (A64setcc node:$lhs, node:$rhs, cond)>;
+
+
+// When matching a notional (CMP op1, (sub 0, op2)), we'd like to use a CMN
+// instruction on the grounds that "op1 - (-op2) == op1 + op2". However, the C
+// and V flags can be set differently by this operation. It comes down to
+// whether "SInt(~op2)+1 == SInt(~op2+1)" (and the same for UInt). If they are
+// then everything is fine. If not then the optimization is wrong. Thus general
+// comparisons are only valid if op2 != 0.
+
+// So, finally, the only LLVM-native comparisons that don't mention C and V are
+// SETEQ and SETNE. They're the only ones we can safely use CMN for in the
+// absence of information about op2.
+def equality_cond : PatLeaf<(cond), [{
+  return N->get() == ISD::SETEQ || N->get() == ISD::SETNE;
+}]>;
+
+def A64cmn : PatFrag<(ops node:$lhs, node:$rhs),
+                     (A64setcc node:$lhs, (sub 0, node:$rhs), equality_cond)>;
+
+// There are two layers of indirection here, driven by the following
+// considerations.
+//     + TableGen does not know CodeModel or Reloc so that decision should be
+//       made for a variable/address at ISelLowering.
+//     + The output of ISelLowering should be selectable (hence the Wrapper,
+//       rather than a bare target opcode)
+def SDTAArch64Wrapper : SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>,
+                                             SDTCisSameAs<1, 2>,
+                                             SDTCisVT<3, i32>,
+                                             SDTCisPtrTy<0>]>;
+
+def A64WrapperSmall : SDNode<"AArch64ISD::WrapperSmall", SDTAArch64Wrapper>;
+
+
+def SDTAArch64GOTLoad : SDTypeProfile<1, 1, [SDTCisPtrTy<0>, SDTCisPtrTy<1>]>;
+def A64GOTLoad : SDNode<"AArch64ISD::GOTLoad", SDTAArch64GOTLoad,
+                        [SDNPHasChain]>;
+
+
+// (A64BFI LHS, RHS, LSB, Width)
+def SDTA64BFI : SDTypeProfile<1, 4, [SDTCisSameAs<0, 1>,
+                                     SDTCisSameAs<1, 2>,
+                                     SDTCisVT<3, i64>,
+                                     SDTCisVT<4, i64>]>;
+
+def A64Bfi : SDNode<"AArch64ISD::BFI", SDTA64BFI>;
+
+// (A64EXTR HiReg, LoReg, LSB)
+def SDTA64EXTR : SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>,
+                                      SDTCisVT<3, i64>]>;
+def A64Extr : SDNode<"AArch64ISD::EXTR", SDTA64EXTR>;
+
+// (A64[SU]BFX Field, ImmR, ImmS).
+//
+// Note that ImmR and ImmS are already encoded for the actual instructions. The
+// more natural LSB and Width mix together to form ImmR and ImmS, something
+// which TableGen can't handle.
+def SDTA64BFX : SDTypeProfile<1, 3, [SDTCisVT<2, i64>, SDTCisVT<3, i64>]>;
+def A64Sbfx : SDNode<"AArch64ISD::SBFX", SDTA64BFX>;
+
+def A64Ubfx : SDNode<"AArch64ISD::UBFX", SDTA64BFX>;
+
+//===----------------------------------------------------------------------===//
+// Call sequence pseudo-instructions
+//===----------------------------------------------------------------------===//
+
+
+def SDT_AArch64Call : SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>;
+def AArch64Call : SDNode<"AArch64ISD::Call", SDT_AArch64Call,
+                     [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue, SDNPVariadic]>;
+
+def AArch64tcret : SDNode<"AArch64ISD::TC_RETURN", SDT_AArch64Call,
+                          [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
+
+// The TLSDESCCALL node is a variant call which goes to an indirectly calculated
+// destination but needs a relocation against a fixed symbol. As such it has two
+// certain operands: the callee and the relocated variable.
+//
+// The TLS ABI only allows it to be selected to a BLR instructin (with
+// appropriate relocation).
+def SDTTLSDescCall : SDTypeProfile<0, -2, [SDTCisPtrTy<0>, SDTCisPtrTy<1>]>;
+
+def A64tlsdesc_blr : SDNode<"AArch64ISD::TLSDESCCALL", SDTTLSDescCall,
+                            [SDNPInGlue, SDNPOutGlue, SDNPHasChain,
+                             SDNPVariadic]>;
+
+
+def SDT_AArch64CallSeqStart : SDCallSeqStart<[ SDTCisPtrTy<0> ]>;
+def AArch64callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_AArch64CallSeqStart,
+                                  [SDNPHasChain, SDNPOutGlue]>;
+
+def SDT_AArch64CallSeqEnd   : SDCallSeqEnd<[ SDTCisPtrTy<0>, SDTCisPtrTy<1> ]>;
+def AArch64callseq_end : SDNode<"ISD::CALLSEQ_END",   SDT_AArch64CallSeqEnd,
+                                [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
+
+
+
+// These pseudo-instructions have special semantics by virtue of being passed to
+// the InstrInfo constructor. CALLSEQ_START/CALLSEQ_END are produced by
+// LowerCall to (in our case) tell the back-end about stack adjustments for
+// arguments passed on the stack. Here we select those markers to
+// pseudo-instructions which explicitly set the stack, and finally in the
+// RegisterInfo we convert them to a true stack adjustment.
+let Defs = [XSP], Uses = [XSP] in {
+  def ADJCALLSTACKDOWN : PseudoInst<(outs), (ins i64imm:$amt),
+                                    [(AArch64callseq_start timm:$amt)]>;
+
+  def ADJCALLSTACKUP : PseudoInst<(outs), (ins i64imm:$amt1, i64imm:$amt2),
+                                 [(AArch64callseq_end timm:$amt1, timm:$amt2)]>;
+}
+
+//===----------------------------------------------------------------------===//
+// Atomic operation pseudo-instructions
+//===----------------------------------------------------------------------===//
+
+let usesCustomInserter = 1 in {
+multiclass AtomicSizes<string opname> {
+  def _I8 : PseudoInst<(outs GPR32:$dst), (ins GPR64:$ptr, GPR32:$incr),
+          [(set i32:$dst, (!cast<SDNode>(opname # "_8") i64:$ptr, i32:$incr))]>;
+  def _I16 : PseudoInst<(outs GPR32:$dst), (ins GPR64:$ptr, GPR32:$incr),
+         [(set i32:$dst, (!cast<SDNode>(opname # "_16") i64:$ptr, i32:$incr))]>;
+  def _I32 : PseudoInst<(outs GPR32:$dst), (ins GPR64:$ptr, GPR32:$incr),
+         [(set i32:$dst, (!cast<SDNode>(opname # "_32") i64:$ptr, i32:$incr))]>;
+  def _I64 : PseudoInst<(outs GPR64:$dst), (ins GPR64:$ptr, GPR64:$incr),
+         [(set i64:$dst, (!cast<SDNode>(opname # "_64") i64:$ptr, i64:$incr))]>;
+}
+}
+
+defm ATOMIC_LOAD_ADD  : AtomicSizes<"atomic_load_add">;
+defm ATOMIC_LOAD_SUB  : AtomicSizes<"atomic_load_sub">;
+defm ATOMIC_LOAD_AND  : AtomicSizes<"atomic_load_and">;
+defm ATOMIC_LOAD_OR   : AtomicSizes<"atomic_load_or">;
+defm ATOMIC_LOAD_XOR  : AtomicSizes<"atomic_load_xor">;
+defm ATOMIC_LOAD_NAND : AtomicSizes<"atomic_load_nand">;
+defm ATOMIC_SWAP      : AtomicSizes<"atomic_swap">;
+let Defs = [NZCV] in {
+  // These operations need a CMP to calculate the correct value
+  defm ATOMIC_LOAD_MIN  : AtomicSizes<"atomic_load_min">;
+  defm ATOMIC_LOAD_MAX  : AtomicSizes<"atomic_load_max">;
+  defm ATOMIC_LOAD_UMIN : AtomicSizes<"atomic_load_umin">;
+  defm ATOMIC_LOAD_UMAX : AtomicSizes<"atomic_load_umax">;
+}
+
+let usesCustomInserter = 1, Defs = [NZCV] in {
+def ATOMIC_CMP_SWAP_I8
+  : PseudoInst<(outs GPR32:$dst), (ins GPR64:$ptr, GPR32:$old, GPR32:$new),
+            [(set i32:$dst, (atomic_cmp_swap_8 i64:$ptr, i32:$old, i32:$new))]>;
+def ATOMIC_CMP_SWAP_I16
+  : PseudoInst<(outs GPR32:$dst), (ins GPR64:$ptr, GPR32:$old, GPR32:$new),
+           [(set i32:$dst, (atomic_cmp_swap_16 i64:$ptr, i32:$old, i32:$new))]>;
+def ATOMIC_CMP_SWAP_I32
+  : PseudoInst<(outs GPR32:$dst), (ins GPR64:$ptr, GPR32:$old, GPR32:$new),
+           [(set i32:$dst, (atomic_cmp_swap_32 i64:$ptr, i32:$old, i32:$new))]>;
+def ATOMIC_CMP_SWAP_I64
+  : PseudoInst<(outs GPR64:$dst), (ins GPR64:$ptr, GPR64:$old, GPR64:$new),
+           [(set i64:$dst, (atomic_cmp_swap_64 i64:$ptr, i64:$old, i64:$new))]>;
+}
+
+//===----------------------------------------------------------------------===//
+// Add-subtract (extended register) instructions
+//===----------------------------------------------------------------------===//
+// Contains: ADD, ADDS, SUB, SUBS + aliases CMN, CMP
+
+// The RHS of these operations is conceptually a sign/zero-extended
+// register, optionally shifted left by 1-4. The extension can be a
+// NOP (e.g. "sxtx" sign-extending a 64-bit register to 64-bits) but
+// must be specified with one exception:
+
+// If one of the registers is sp/wsp then LSL is an alias for UXTW in
+// 32-bit instructions and UXTX in 64-bit versions, the shift amount
+// is not optional in that case (but can explicitly be 0), and the
+// entire suffix can be skipped (e.g. "add sp, x3, x2").
+
+multiclass extend_operands<string PREFIX, string Diag> {
+     def _asmoperand : AsmOperandClass {
+         let Name = PREFIX;
+         let RenderMethod = "addRegExtendOperands";
+         let PredicateMethod = "isRegExtend<A64SE::" # PREFIX # ">";
+         let DiagnosticType = "AddSubRegExtend" # Diag;
+     }
+
+     def _operand : Operand<i64>,
+                    ImmLeaf<i64, [{ return Imm >= 0 && Imm <= 4; }]> {
+         let PrintMethod = "printRegExtendOperand<A64SE::" # PREFIX # ">";
+         let DecoderMethod = "DecodeRegExtendOperand";
+         let ParserMatchClass = !cast<AsmOperandClass>(PREFIX # "_asmoperand");
+     }
+}
+
+defm UXTB : extend_operands<"UXTB", "Small">;
+defm UXTH : extend_operands<"UXTH", "Small">;
+defm UXTW : extend_operands<"UXTW", "Small">;
+defm UXTX : extend_operands<"UXTX", "Large">;
+defm SXTB : extend_operands<"SXTB", "Small">;
+defm SXTH : extend_operands<"SXTH", "Small">;
+defm SXTW : extend_operands<"SXTW", "Small">;
+defm SXTX : extend_operands<"SXTX", "Large">;
+
+def LSL_extasmoperand : AsmOperandClass {
+    let Name = "RegExtendLSL";
+    let RenderMethod = "addRegExtendOperands";
+    let DiagnosticType = "AddSubRegExtendLarge";
+}
+
+def LSL_extoperand : Operand<i64> {
+    let ParserMatchClass = LSL_extasmoperand;
+}
+
+
+// The patterns for various sign-extensions are a little ugly and
+// non-uniform because everything has already been promoted to the
+// legal i64 and i32 types. We'll wrap the various variants up in a
+// class for use later.
+class extend_types {
+    dag uxtb; dag uxth; dag uxtw; dag uxtx;
+    dag sxtb; dag sxth; dag sxtw; dag sxtx;
+    ValueType ty;
+    RegisterClass GPR;
+}
+
+def extends_to_i64 : extend_types {
+    let uxtb = (and (anyext i32:$Rm), 255);
+    let uxth = (and (anyext i32:$Rm), 65535);
+    let uxtw = (zext i32:$Rm);
+    let uxtx = (i64 $Rm);
+
+    let sxtb = (sext_inreg (anyext i32:$Rm), i8);
+    let sxth = (sext_inreg (anyext i32:$Rm), i16);
+    let sxtw = (sext i32:$Rm);
+    let sxtx = (i64 $Rm);
+
+    let ty = i64;
+    let GPR = GPR64xsp;
+}
+
+
+def extends_to_i32 : extend_types {
+    let uxtb = (and i32:$Rm, 255);
+    let uxth = (and i32:$Rm, 65535);
+    let uxtw = (i32 i32:$Rm);
+    let uxtx = (i32 i32:$Rm);
+
+    let sxtb = (sext_inreg i32:$Rm, i8);
+    let sxth = (sext_inreg i32:$Rm, i16);
+    let sxtw = (i32 i32:$Rm);
+    let sxtx = (i32 i32:$Rm);
+
+    let ty = i32;
+    let GPR = GPR32wsp;
+}
+
+// Now, six of the extensions supported are easy and uniform: if the source size
+// is 32-bits or less, then Rm is always a 32-bit register. We'll instantiate
+// those instructions in one block.
+
+// The uxtx/sxtx could potentially be merged in, but three facts dissuaded me:
+//     + It would break the naming scheme: either ADDxx_uxtx or ADDww_uxtx would
+//       be impossible.
+//     + Patterns are very different as well.
+//     + Passing different registers would be ugly (more fields in extend_types
+//       would probably be the best option).
+multiclass addsub_exts<bit sf, bit op, bit S, string asmop,
+                       SDPatternOperator opfrag,
+                       dag outs, extend_types exts> {
+    def w_uxtb : A64I_addsubext<sf, op, S, 0b00, 0b000,
+                    outs, (ins exts.GPR:$Rn, GPR32:$Rm, UXTB_operand:$Imm3),
+                    !strconcat(asmop, "$Rn, $Rm, $Imm3"),
+                    [(opfrag exts.ty:$Rn, (shl exts.uxtb, UXTB_operand:$Imm3))],
+                    NoItinerary>;
+    def w_uxth : A64I_addsubext<sf, op, S, 0b00, 0b001,
+                    outs, (ins exts.GPR:$Rn, GPR32:$Rm, UXTH_operand:$Imm3),
+                    !strconcat(asmop, "$Rn, $Rm, $Imm3"),
+                    [(opfrag exts.ty:$Rn, (shl exts.uxth, UXTH_operand:$Imm3))],
+                    NoItinerary>;
+    def w_uxtw : A64I_addsubext<sf, op, S, 0b00, 0b010,
+                    outs, (ins exts.GPR:$Rn, GPR32:$Rm, UXTW_operand:$Imm3),
+                    !strconcat(asmop, "$Rn, $Rm, $Imm3"),
+                    [(opfrag exts.ty:$Rn, (shl exts.uxtw, UXTW_operand:$Imm3))],
+                    NoItinerary>;
+
+    def w_sxtb : A64I_addsubext<sf, op, S, 0b00, 0b100,
+                    outs, (ins exts.GPR:$Rn, GPR32:$Rm, SXTB_operand:$Imm3),
+                    !strconcat(asmop, "$Rn, $Rm, $Imm3"),
+                    [(opfrag exts.ty:$Rn, (shl exts.sxtb, SXTB_operand:$Imm3))],
+                    NoItinerary>;
+    def w_sxth : A64I_addsubext<sf, op, S, 0b00, 0b101,
+                    outs, (ins exts.GPR:$Rn, GPR32:$Rm, SXTH_operand:$Imm3),
+                    !strconcat(asmop, "$Rn, $Rm, $Imm3"),
+                    [(opfrag exts.ty:$Rn, (shl exts.sxth, SXTH_operand:$Imm3))],
+                    NoItinerary>;
+    def w_sxtw : A64I_addsubext<sf, op, S, 0b00, 0b110,
+                    outs, (ins exts.GPR:$Rn, GPR32:$Rm, SXTW_operand:$Imm3),
+                    !strconcat(asmop, "$Rn, $Rm, $Imm3"),
+                    [(opfrag exts.ty:$Rn, (shl exts.sxtw, SXTW_operand:$Imm3))],
+                    NoItinerary>;
+}
+
+// These two could be merge in with the above, but their patterns aren't really
+// necessary and the naming-scheme would necessarily break:
+multiclass addsub_xxtx<bit op, bit S, string asmop, SDPatternOperator opfrag,
+                       dag outs> {
+    def x_uxtx : A64I_addsubext<0b1, op, S, 0b00, 0b011,
+                   outs,
+                   (ins GPR64xsp:$Rn, GPR64:$Rm, UXTX_operand:$Imm3),
+                   !strconcat(asmop, "$Rn, $Rm, $Imm3"),
+                   [(opfrag i64:$Rn, (shl i64:$Rm, UXTX_operand:$Imm3))],
+                   NoItinerary>;
+
+    def x_sxtx : A64I_addsubext<0b1, op, S, 0b00, 0b111,
+                   outs,
+                   (ins GPR64xsp:$Rn, GPR64:$Rm, SXTX_operand:$Imm3),
+                   !strconcat(asmop, "$Rn, $Rm, $Imm3"),
+                   [/* No Pattern: same as uxtx */],
+                   NoItinerary>;
+}
+
+multiclass addsub_wxtx<bit op, bit S, string asmop, dag outs> {
+    def w_uxtx : A64I_addsubext<0b0, op, S, 0b00, 0b011,
+                              outs,
+                              (ins GPR32wsp:$Rn, GPR32:$Rm, UXTX_operand:$Imm3),
+                              !strconcat(asmop, "$Rn, $Rm, $Imm3"),
+                              [/* No pattern: probably same as uxtw */],
+                              NoItinerary>;
+
+    def w_sxtx : A64I_addsubext<0b0, op, S, 0b00, 0b111,
+                              outs,
+                              (ins GPR32wsp:$Rn, GPR32:$Rm, SXTX_operand:$Imm3),
+                              !strconcat(asmop, "$Rn, $Rm, $Imm3"),
+                              [/* No Pattern: probably same as uxtw */],
+                              NoItinerary>;
+}
+
+class SetRD<RegisterClass RC, SDPatternOperator op>
+ : PatFrag<(ops node:$lhs, node:$rhs), (set RC:$Rd, (op node:$lhs, node:$rhs))>;
+class SetNZCV<SDPatternOperator op>
+  : PatFrag<(ops node:$lhs, node:$rhs), (set NZCV, (op node:$lhs, node:$rhs))>;
+
+defm ADDxx :addsub_exts<0b1, 0b0, 0b0, "add\t$Rd, ", SetRD<GPR64xsp, add>,
+                        (outs GPR64xsp:$Rd), extends_to_i64>,
+            addsub_xxtx<     0b0, 0b0, "add\t$Rd, ", SetRD<GPR64xsp, add>,
+                        (outs GPR64xsp:$Rd)>;
+defm ADDww :addsub_exts<0b0, 0b0, 0b0, "add\t$Rd, ", SetRD<GPR32wsp, add>,
+                        (outs GPR32wsp:$Rd), extends_to_i32>,
+            addsub_wxtx<     0b0, 0b0, "add\t$Rd, ",
+                        (outs GPR32wsp:$Rd)>;
+defm SUBxx :addsub_exts<0b1, 0b1, 0b0, "sub\t$Rd, ", SetRD<GPR64xsp, sub>,
+                        (outs GPR64xsp:$Rd), extends_to_i64>,
+            addsub_xxtx<     0b1, 0b0, "sub\t$Rd, ", SetRD<GPR64xsp, sub>,
+                        (outs GPR64xsp:$Rd)>;
+defm SUBww :addsub_exts<0b0, 0b1, 0b0, "sub\t$Rd, ", SetRD<GPR32wsp, sub>,
+                        (outs GPR32wsp:$Rd), extends_to_i32>,
+            addsub_wxtx<     0b1, 0b0, "sub\t$Rd, ",
+                        (outs GPR32wsp:$Rd)>;
+
+let Defs = [NZCV] in {
+defm ADDSxx :addsub_exts<0b1, 0b0, 0b1, "adds\t$Rd, ", SetRD<GPR64, addc>,
+                         (outs GPR64:$Rd), extends_to_i64>,
+             addsub_xxtx<     0b0, 0b1, "adds\t$Rd, ", SetRD<GPR64, addc>,
+                         (outs GPR64:$Rd)>;
+defm ADDSww :addsub_exts<0b0, 0b0, 0b1, "adds\t$Rd, ", SetRD<GPR32, addc>,
+                         (outs GPR32:$Rd), extends_to_i32>,
+             addsub_wxtx<     0b0, 0b1, "adds\t$Rd, ",
+                         (outs GPR32:$Rd)>;
+defm SUBSxx :addsub_exts<0b1, 0b1, 0b1, "subs\t$Rd, ", SetRD<GPR64, subc>,
+                         (outs GPR64:$Rd), extends_to_i64>,
+             addsub_xxtx<     0b1, 0b1, "subs\t$Rd, ", SetRD<GPR64, subc>,
+                         (outs GPR64:$Rd)>;
+defm SUBSww :addsub_exts<0b0, 0b1, 0b1, "subs\t$Rd, ", SetRD<GPR32, subc>,
+                         (outs GPR32:$Rd), extends_to_i32>,
+             addsub_wxtx<     0b1, 0b1, "subs\t$Rd, ",
+                         (outs GPR32:$Rd)>;
+
+
+let Rd = 0b11111, isCompare = 1 in {
+defm CMNx : addsub_exts<0b1, 0b0, 0b1, "cmn\t", SetNZCV<A64cmn>,
+                        (outs), extends_to_i64>,
+            addsub_xxtx<     0b0, 0b1, "cmn\t", SetNZCV<A64cmn>, (outs)>;
+defm CMNw : addsub_exts<0b0, 0b0, 0b1, "cmn\t", SetNZCV<A64cmn>,
+                        (outs), extends_to_i32>,
+            addsub_wxtx<     0b0, 0b1, "cmn\t", (outs)>;
+defm CMPx : addsub_exts<0b1, 0b1, 0b1, "cmp\t", SetNZCV<A64cmp>,
+                        (outs), extends_to_i64>,
+            addsub_xxtx<     0b1, 0b1, "cmp\t", SetNZCV<A64cmp>, (outs)>;
+defm CMPw : addsub_exts<0b0, 0b1, 0b1, "cmp\t", SetNZCV<A64cmp>,
+                        (outs), extends_to_i32>,
+            addsub_wxtx<     0b1, 0b1, "cmp\t", (outs)>;
+}
+}
+
+// Now patterns for the operation without a shift being needed. No patterns are
+// created for uxtx/sxtx since they're non-uniform and it's expected that
+// add/sub (shifted register) will handle those cases anyway.
+multiclass addsubext_noshift_patterns<string prefix, SDPatternOperator nodeop,
+                                      extend_types exts> {
+    def : Pat<(nodeop exts.ty:$Rn, exts.uxtb),
+              (!cast<Instruction>(prefix # "w_uxtb") $Rn, $Rm, 0)>;
+    def : Pat<(nodeop exts.ty:$Rn, exts.uxth),
+              (!cast<Instruction>(prefix # "w_uxth") $Rn, $Rm, 0)>;
+    def : Pat<(nodeop exts.ty:$Rn, exts.uxtw),
+              (!cast<Instruction>(prefix # "w_uxtw") $Rn, $Rm, 0)>;
+
+    def : Pat<(nodeop exts.ty:$Rn, exts.sxtb),
+              (!cast<Instruction>(prefix # "w_sxtb") $Rn, $Rm, 0)>;
+    def : Pat<(nodeop exts.ty:$Rn, exts.sxth),
+              (!cast<Instruction>(prefix # "w_sxth") $Rn, $Rm, 0)>;
+    def : Pat<(nodeop exts.ty:$Rn, exts.sxtw),
+              (!cast<Instruction>(prefix # "w_sxtw") $Rn, $Rm, 0)>;
+}
+
+defm : addsubext_noshift_patterns<"ADDxx", add, extends_to_i64>;
+defm : addsubext_noshift_patterns<"ADDww", add, extends_to_i32>;
+defm : addsubext_noshift_patterns<"SUBxx", sub, extends_to_i64>;
+defm : addsubext_noshift_patterns<"SUBww", sub, extends_to_i32>;
+
+defm : addsubext_noshift_patterns<"CMNx", A64cmn, extends_to_i64>;
+defm : addsubext_noshift_patterns<"CMNw", A64cmn, extends_to_i32>;
+defm : addsubext_noshift_patterns<"CMPx", A64cmp, extends_to_i64>;
+defm : addsubext_noshift_patterns<"CMPw", A64cmp, extends_to_i32>;
+
+// An extend of "lsl #imm" is valid if and only if one of Rn and Rd is
+// sp/wsp. It is synonymous with uxtx/uxtw depending on the size of the
+// operation. Also permitted in this case is complete omission of the argument,
+// which implies "lsl #0".
+multiclass lsl_aliases<string asmop, Instruction inst, RegisterClass GPR_Rd,
+                       RegisterClass GPR_Rn, RegisterClass GPR_Rm> {
+    def : InstAlias<!strconcat(asmop, " $Rd, $Rn, $Rm"),
+                    (inst GPR_Rd:$Rd, GPR_Rn:$Rn, GPR_Rm:$Rm, 0)>;
+
+    def : InstAlias<!strconcat(asmop, " $Rd, $Rn, $Rm, $LSL"),
+                (inst GPR_Rd:$Rd, GPR_Rn:$Rn, GPR_Rm:$Rm, LSL_extoperand:$LSL)>;
+
+}
+
+defm : lsl_aliases<"add",  ADDxxx_uxtx,  Rxsp, GPR64xsp, GPR64>;
+defm : lsl_aliases<"add",  ADDxxx_uxtx,  GPR64xsp, Rxsp, GPR64>;
+defm : lsl_aliases<"add",  ADDwww_uxtw,  Rwsp, GPR32wsp, GPR32>;
+defm : lsl_aliases<"add",  ADDwww_uxtw,  GPR32wsp, Rwsp, GPR32>;
+defm : lsl_aliases<"sub",  SUBxxx_uxtx,  Rxsp, GPR64xsp, GPR64>;
+defm : lsl_aliases<"sub",  SUBxxx_uxtx,  GPR64xsp, Rxsp, GPR64>;
+defm : lsl_aliases<"sub",  SUBwww_uxtw,  Rwsp, GPR32wsp, GPR32>;
+defm : lsl_aliases<"sub",  SUBwww_uxtw,  GPR32wsp, Rwsp, GPR32>;
+
+// Rd cannot be sp for flag-setting variants so only half of the aliases are
+// needed.
+defm : lsl_aliases<"adds", ADDSxxx_uxtx, GPR64, Rxsp, GPR64>;
+defm : lsl_aliases<"adds", ADDSwww_uxtw, GPR32, Rwsp, GPR32>;
+defm : lsl_aliases<"subs", SUBSxxx_uxtx, GPR64, Rxsp, GPR64>;
+defm : lsl_aliases<"subs", SUBSwww_uxtw, GPR32, Rwsp, GPR32>;
+
+// CMP unfortunately has to be different because the instruction doesn't have a
+// dest register.
+multiclass cmp_lsl_aliases<string asmop, Instruction inst,
+                       RegisterClass GPR_Rn, RegisterClass GPR_Rm> {
+    def : InstAlias<!strconcat(asmop, " $Rn, $Rm"),
+                    (inst GPR_Rn:$Rn, GPR_Rm:$Rm, 0)>;
+
+    def : InstAlias<!strconcat(asmop, " $Rn, $Rm, $LSL"),
+                    (inst GPR_Rn:$Rn, GPR_Rm:$Rm, LSL_extoperand:$LSL)>;
+}
+
+defm : cmp_lsl_aliases<"cmp", CMPxx_uxtx, Rxsp, GPR64>;
+defm : cmp_lsl_aliases<"cmp", CMPww_uxtw, Rwsp, GPR32>;
+defm : cmp_lsl_aliases<"cmn", CMNxx_uxtx, Rxsp, GPR64>;
+defm : cmp_lsl_aliases<"cmn", CMNww_uxtw, Rwsp, GPR32>;
+
+//===----------------------------------------------------------------------===//
+// Add-subtract (immediate) instructions
+//===----------------------------------------------------------------------===//
+// Contains: ADD, ADDS, SUB, SUBS + aliases CMN, CMP, MOV
+
+// These instructions accept a 12-bit unsigned immediate, optionally shifted
+// left by 12 bits. Official assembly format specifies a 12 bit immediate with
+// one of "", "LSL #0", "LSL #12" supplementary operands.
+
+// There are surprisingly few ways to make this work with TableGen, so this
+// implementation has separate instructions for the "LSL #0" and "LSL #12"
+// variants.
+
+// If the MCInst retained a single combined immediate (which could be 0x123000,
+// for example) then both components (imm & shift) would have to be delegated to
+// a single assembly operand. This would entail a separate operand parser
+// (because the LSL would have to live in the same AArch64Operand as the
+// immediate to be accessible); assembly parsing is rather complex and
+// error-prone C++ code.
+//
+// By splitting the immediate, we can delegate handling this optional operand to
+// an InstAlias. Supporting functions to generate the correct MCInst are still
+// required, but these are essentially trivial and parsing can remain generic.
+//
+// Rejected plans with rationale:
+// ------------------------------
+//
+// In an ideal world you'de have two first class immediate operands (in
+// InOperandList, specifying imm12 and shift). Unfortunately this is not
+// selectable by any means I could discover.
+//
+// An Instruction with two MCOperands hidden behind a single entry in
+// InOperandList (expanded by ComplexPatterns and MIOperandInfo) was functional,
+// but required more C++ code to handle encoding/decoding. Parsing (the intended
+// main beneficiary) ended up equally complex because of the optional nature of
+// "LSL #0".
+//
+// Attempting to circumvent the need for a custom OperandParser above by giving
+// InstAliases without the "lsl #0" failed. add/sub could be accommodated but
+// the cmp/cmn aliases didn't use the MIOperandInfo to determine how operands
+// should be parsed: there was no way to accommodate an "lsl #12".
+
+let ParserMethod = "ParseImmWithLSLOperand",
+    RenderMethod = "addImmWithLSLOperands" in {
+  // Derived PredicateMethod fields are different for each
+  def addsubimm_lsl0_asmoperand : AsmOperandClass {
+    let Name = "AddSubImmLSL0";
+    // If an error is reported against this operand, instruction could also be a
+    // register variant.
+    let DiagnosticType = "AddSubSecondSource";
+  }
+
+  def addsubimm_lsl12_asmoperand : AsmOperandClass {
+    let Name = "AddSubImmLSL12";
+    let DiagnosticType = "AddSubSecondSource";
+  }
+}
+
+def shr_12_XFORM : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(N->getSExtValue() >> 12, MVT::i32);
+}]>;
+
+def shr_12_neg_XFORM : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant((-N->getSExtValue()) >> 12, MVT::i32);
+}]>;
+
+def neg_XFORM : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(-N->getSExtValue(), MVT::i32);
+}]>;
+
+
+multiclass addsub_imm_operands<ValueType ty> {
+ let PrintMethod = "printAddSubImmLSL0Operand",
+      EncoderMethod = "getAddSubImmOpValue",
+      ParserMatchClass = addsubimm_lsl0_asmoperand in {
+    def _posimm_lsl0 : Operand<ty>,
+        ImmLeaf<ty, [{ return Imm >= 0 && (Imm & ~0xfff) == 0; }]>;
+    def _negimm_lsl0 : Operand<ty>,
+        ImmLeaf<ty, [{ return Imm < 0 && (-Imm & ~0xfff) == 0; }],
+                neg_XFORM>;
+  }
+
+  let PrintMethod = "printAddSubImmLSL12Operand",
+      EncoderMethod = "getAddSubImmOpValue",
+      ParserMatchClass = addsubimm_lsl12_asmoperand in {
+    def _posimm_lsl12 : Operand<ty>,
+        ImmLeaf<ty, [{ return Imm >= 0 && (Imm & ~0xfff000) == 0; }],
+                shr_12_XFORM>;
+
+    def _negimm_lsl12 : Operand<ty>,
+        ImmLeaf<ty, [{ return Imm < 0 && (-Imm & ~0xfff000) == 0; }],
+                shr_12_neg_XFORM>;
+  }
+}
+
+// The add operands don't need any transformation
+defm addsubimm_operand_i32 : addsub_imm_operands<i32>;
+defm addsubimm_operand_i64 : addsub_imm_operands<i64>;
+
+multiclass addsubimm_varieties<string prefix, bit sf, bit op, bits<2> shift,
+                               string asmop, string cmpasmop,
+                               Operand imm_operand, Operand cmp_imm_operand,
+                               RegisterClass GPR, RegisterClass GPRsp,
+                               AArch64Reg ZR, ValueType Ty> {
+    // All registers for non-S variants allow SP
+  def _s : A64I_addsubimm<sf, op, 0b0, shift,
+                         (outs GPRsp:$Rd),
+                         (ins GPRsp:$Rn, imm_operand:$Imm12),
+                         !strconcat(asmop, "\t$Rd, $Rn, $Imm12"),
+                         [(set Ty:$Rd, (add Ty:$Rn, imm_operand:$Imm12))],
+                         NoItinerary>;
+
+
+  // S variants can read SP but would write to ZR
+  def _S : A64I_addsubimm<sf, op, 0b1, shift,
+                         (outs GPR:$Rd),
+                         (ins GPRsp:$Rn, imm_operand:$Imm12),
+                         !strconcat(asmop, "s\t$Rd, $Rn, $Imm12"),
+                         [(set Ty:$Rd, (addc Ty:$Rn, imm_operand:$Imm12))],
+                         NoItinerary> {
+    let Defs = [NZCV];
+  }
+
+  // Note that the pattern here for ADDS is subtle. Canonically CMP
+  // a, b becomes SUBS a, b. If b < 0 then this is equivalent to
+  // ADDS a, (-b). This is not true in general.
+  def _cmp : A64I_addsubimm<sf, op, 0b1, shift,
+                            (outs), (ins GPRsp:$Rn, imm_operand:$Imm12),
+                            !strconcat(cmpasmop, " $Rn, $Imm12"),
+                            [(set NZCV,
+                                  (A64cmp Ty:$Rn, cmp_imm_operand:$Imm12))],
+                            NoItinerary> {
+    let Rd = 0b11111;
+    let Defs = [NZCV];
+    let isCompare = 1;
+  }
+}
+
+
+multiclass addsubimm_shifts<string prefix, bit sf, bit op,
+           string asmop, string cmpasmop, string operand, string cmpoperand,
+           RegisterClass GPR, RegisterClass GPRsp, AArch64Reg ZR,
+           ValueType Ty> {
+  defm _lsl0 : addsubimm_varieties<prefix # "_lsl0", sf, op, 0b00,
+                                   asmop, cmpasmop,
+                                   !cast<Operand>(operand # "_lsl0"),
+                                   !cast<Operand>(cmpoperand # "_lsl0"),
+                                   GPR, GPRsp, ZR, Ty>;
+
+  defm _lsl12 : addsubimm_varieties<prefix # "_lsl12", sf, op, 0b01,
+                                    asmop, cmpasmop,
+                                    !cast<Operand>(operand # "_lsl12"),
+                                    !cast<Operand>(cmpoperand # "_lsl12"),
+                                    GPR, GPRsp, ZR, Ty>;
+}
+
+defm ADDwwi : addsubimm_shifts<"ADDwi", 0b0, 0b0, "add", "cmn",
+                              "addsubimm_operand_i32_posimm",
+                              "addsubimm_operand_i32_negimm",
+                              GPR32, GPR32wsp, WZR, i32>;
+defm ADDxxi : addsubimm_shifts<"ADDxi", 0b1, 0b0, "add", "cmn",
+                              "addsubimm_operand_i64_posimm",
+                              "addsubimm_operand_i64_negimm",
+                              GPR64, GPR64xsp, XZR, i64>;
+defm SUBwwi : addsubimm_shifts<"SUBwi", 0b0, 0b1, "sub", "cmp",
+                              "addsubimm_operand_i32_negimm",
+                              "addsubimm_operand_i32_posimm",
+                              GPR32, GPR32wsp, WZR, i32>;
+defm SUBxxi : addsubimm_shifts<"SUBxi", 0b1, 0b1, "sub", "cmp",
+                              "addsubimm_operand_i64_negimm",
+                              "addsubimm_operand_i64_posimm",
+                              GPR64, GPR64xsp, XZR, i64>;
+
+multiclass MOVsp<RegisterClass GPRsp, RegisterClass SP, Instruction addop> {
+  def _fromsp : InstAlias<"mov $Rd, $Rn",
+                          (addop GPRsp:$Rd, SP:$Rn, 0),
+                          0b1>;
+
+  def _tosp : InstAlias<"mov $Rd, $Rn",
+                        (addop SP:$Rd, GPRsp:$Rn, 0),
+                        0b1>;
+}
+
+// Recall Rxsp is a RegisterClass containing *just* xsp.
+defm MOVxx : MOVsp<GPR64xsp, Rxsp, ADDxxi_lsl0_s>;
+defm MOVww : MOVsp<GPR32wsp, Rwsp, ADDwwi_lsl0_s>;
+
+//===----------------------------------------------------------------------===//
+// Add-subtract (shifted register) instructions
+//===----------------------------------------------------------------------===//
+// Contains: ADD, ADDS, SUB, SUBS + aliases CMN, CMP, NEG, NEGS
+
+//===-------------------------------
+// 1. The "shifed register" operands. Shared with logical insts.
+//===-------------------------------
+
+multiclass shift_operands<string prefix, string form> {
+  def _asmoperand_i32 : AsmOperandClass {
+    let Name = "Shift" # form # "i32";
+    let RenderMethod = "addShiftOperands";
+    let PredicateMethod = "isShift<A64SE::" # form # ", false>";
+    let DiagnosticType = "AddSubRegShift32";
+  }
+
+  // Note that the operand type is intentionally i64 because the DAGCombiner
+  // puts these into a canonical form.
+  def _i32 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= 0 && Imm <= 31; }]> {
+    let ParserMatchClass
+          = !cast<AsmOperandClass>(prefix # "_asmoperand_i32");
+    let PrintMethod = "printShiftOperand<A64SE::" # form # ">";
+    let DecoderMethod = "Decode32BitShiftOperand";
+  }
+
+  def _asmoperand_i64 : AsmOperandClass {
+      let Name = "Shift" # form # "i64";
+      let RenderMethod = "addShiftOperands";
+      let PredicateMethod = "isShift<A64SE::" # form # ", true>";
+      let DiagnosticType = "AddSubRegShift64";
+  }
+
+  def _i64 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= 0 && Imm <= 63; }]> {
+    let ParserMatchClass
+          = !cast<AsmOperandClass>(prefix # "_asmoperand_i64");
+    let PrintMethod = "printShiftOperand<A64SE::" # form # ">";
+  }
+}
+
+defm lsl_operand : shift_operands<"lsl_operand", "LSL">;
+defm lsr_operand : shift_operands<"lsr_operand", "LSR">;
+defm asr_operand : shift_operands<"asr_operand", "ASR">;
+
+// Not used for add/sub, but defined here for completeness. The "logical
+// (shifted register)" instructions *do* have an ROR variant.
+defm ror_operand : shift_operands<"ror_operand", "ROR">;
+
+//===-------------------------------
+// 2. The basic 3.5-operand ADD/SUB/ADDS/SUBS instructions.
+//===-------------------------------
+
+// N.b. the commutable parameter is just !N. It will be first against the wall
+// when the revolution comes.
+multiclass addsub_shifts<string prefix, bit sf, bit op, bit s, bit commutable,
+                         string asmop, SDPatternOperator opfrag, ValueType ty,
+                         RegisterClass GPR, list<Register> defs> {
+  let isCommutable = commutable, Defs = defs in {
+  def _lsl : A64I_addsubshift<sf, op, s, 0b00,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("lsl_operand_" # ty):$Imm6),
+                       !strconcat(asmop, "\t$Rd, $Rn, $Rm, $Imm6"),
+                       [(set GPR:$Rd, (opfrag ty:$Rn, (shl ty:$Rm,
+                            !cast<Operand>("lsl_operand_" # ty):$Imm6))
+                       )],
+                       NoItinerary>;
+
+  def _lsr : A64I_addsubshift<sf, op, s, 0b01,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("lsr_operand_" # ty):$Imm6),
+                       !strconcat(asmop, "\t$Rd, $Rn, $Rm, $Imm6"),
+                       [(set ty:$Rd, (opfrag ty:$Rn, (srl ty:$Rm,
+                            !cast<Operand>("lsr_operand_" # ty):$Imm6))
+                       )],
+                       NoItinerary>;
+
+  def _asr : A64I_addsubshift<sf, op, s, 0b10,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("asr_operand_" # ty):$Imm6),
+                       !strconcat(asmop, "\t$Rd, $Rn, $Rm, $Imm6"),
+                       [(set ty:$Rd, (opfrag ty:$Rn, (sra ty:$Rm,
+                            !cast<Operand>("asr_operand_" # ty):$Imm6))
+                       )],
+                       NoItinerary>;
+  }
+
+  def _noshift
+      : InstAlias<!strconcat(asmop, " $Rd, $Rn, $Rm"),
+                 (!cast<Instruction>(prefix # "_lsl") GPR:$Rd, GPR:$Rn,
+                                                      GPR:$Rm, 0)>;
+
+  def : Pat<(opfrag ty:$Rn, ty:$Rm),
+            (!cast<Instruction>(prefix # "_lsl") $Rn, $Rm, 0)>;
+}
+
+multiclass addsub_sizes<string prefix, bit op, bit s, bit commutable,
+                         string asmop, SDPatternOperator opfrag,
+                         list<Register> defs> {
+  defm xxx : addsub_shifts<prefix # "xxx", 0b1, op, s,
+                           commutable, asmop, opfrag, i64, GPR64, defs>;
+  defm www : addsub_shifts<prefix # "www", 0b0, op, s,
+                           commutable, asmop, opfrag, i32, GPR32, defs>;
+}
+
+
+defm ADD : addsub_sizes<"ADD", 0b0, 0b0, 0b1, "add", add, []>;
+defm SUB : addsub_sizes<"SUB", 0b1, 0b0, 0b0, "sub", sub, []>;
+
+defm ADDS : addsub_sizes<"ADDS", 0b0, 0b1, 0b1, "adds", addc, [NZCV]>;
+defm SUBS : addsub_sizes<"SUBS", 0b1, 0b1, 0b0, "subs", subc, [NZCV]>;
+
+//===-------------------------------
+// 1. The NEG/NEGS aliases
+//===-------------------------------
+
+multiclass neg_alias<Instruction INST, RegisterClass GPR, Register ZR,
+                     ValueType ty, Operand shift_operand, SDNode shiftop> {
+   def : InstAlias<"neg $Rd, $Rm, $Imm6",
+                   (INST GPR:$Rd, ZR, GPR:$Rm, shift_operand:$Imm6)>;
+
+   def : Pat<(sub 0, (shiftop ty:$Rm, shift_operand:$Imm6)),
+             (INST ZR, $Rm, shift_operand:$Imm6)>;
+}
+
+defm : neg_alias<SUBwww_lsl, GPR32, WZR, i32, lsl_operand_i32, shl>;
+defm : neg_alias<SUBwww_lsr, GPR32, WZR, i32, lsr_operand_i32, srl>;
+defm : neg_alias<SUBwww_asr, GPR32, WZR, i32, asr_operand_i32, sra>;
+def : InstAlias<"neg $Rd, $Rm", (SUBwww_lsl GPR32:$Rd, WZR, GPR32:$Rm, 0)>;
+def : Pat<(sub 0, i32:$Rm), (SUBwww_lsl WZR, $Rm, 0)>;
+
+defm : neg_alias<SUBxxx_lsl, GPR64, XZR, i64, lsl_operand_i64, shl>;
+defm : neg_alias<SUBxxx_lsr, GPR64, XZR, i64, lsr_operand_i64, srl>;
+defm : neg_alias<SUBxxx_asr, GPR64, XZR, i64, asr_operand_i64, sra>;
+def : InstAlias<"neg $Rd, $Rm", (SUBxxx_lsl GPR64:$Rd, XZR, GPR64:$Rm, 0)>;
+def : Pat<(sub 0, i64:$Rm), (SUBxxx_lsl XZR, $Rm, 0)>;
+
+// NEGS doesn't get any patterns yet: defining multiple outputs means C++ has to
+// be involved.
+class negs_alias<Instruction INST, RegisterClass GPR,
+                 Register ZR, Operand shift_operand, SDNode shiftop>
+  : InstAlias<"negs $Rd, $Rm, $Imm6",
+              (INST GPR:$Rd, ZR, GPR:$Rm, shift_operand:$Imm6)>;
+
+def : negs_alias<SUBSwww_lsl, GPR32, WZR, lsl_operand_i32, shl>;
+def : negs_alias<SUBSwww_lsr, GPR32, WZR, lsr_operand_i32, srl>;
+def : negs_alias<SUBSwww_asr, GPR32, WZR, asr_operand_i32, sra>;
+def : InstAlias<"negs $Rd, $Rm", (SUBSwww_lsl GPR32:$Rd, WZR, GPR32:$Rm, 0)>;
+
+def : negs_alias<SUBSxxx_lsl, GPR64, XZR, lsl_operand_i64, shl>;
+def : negs_alias<SUBSxxx_lsr, GPR64, XZR, lsr_operand_i64, srl>;
+def : negs_alias<SUBSxxx_asr, GPR64, XZR, asr_operand_i64, sra>;
+def : InstAlias<"negs $Rd, $Rm", (SUBSxxx_lsl GPR64:$Rd, XZR, GPR64:$Rm, 0)>;
+
+//===-------------------------------
+// 1. The CMP/CMN aliases
+//===-------------------------------
+
+multiclass cmp_shifts<string prefix, bit sf, bit op, bit commutable,
+                      string asmop, SDPatternOperator opfrag, ValueType ty,
+                      RegisterClass GPR> {
+  let isCommutable = commutable, Rd = 0b11111, Defs = [NZCV] in {
+  def _lsl : A64I_addsubshift<sf, op, 0b1, 0b00,
+                       (outs),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("lsl_operand_" # ty):$Imm6),
+                       !strconcat(asmop, "\t$Rn, $Rm, $Imm6"),
+                       [(set NZCV, (opfrag ty:$Rn, (shl ty:$Rm,
+                            !cast<Operand>("lsl_operand_" # ty):$Imm6))
+                       )],
+                       NoItinerary>;
+
+  def _lsr : A64I_addsubshift<sf, op, 0b1, 0b01,
+                       (outs),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("lsr_operand_" # ty):$Imm6),
+                       !strconcat(asmop, "\t$Rn, $Rm, $Imm6"),
+                       [(set NZCV, (opfrag ty:$Rn, (srl ty:$Rm,
+                            !cast<Operand>("lsr_operand_" # ty):$Imm6))
+                       )],
+                       NoItinerary>;
+
+  def _asr : A64I_addsubshift<sf, op, 0b1, 0b10,
+                       (outs),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("asr_operand_" # ty):$Imm6),
+                       !strconcat(asmop, "\t$Rn, $Rm, $Imm6"),
+                       [(set NZCV, (opfrag ty:$Rn, (sra ty:$Rm,
+                            !cast<Operand>("asr_operand_" # ty):$Imm6))
+                       )],
+                       NoItinerary>;
+  }
+
+  def _noshift
+      : InstAlias<!strconcat(asmop, " $Rn, $Rm"),
+                 (!cast<Instruction>(prefix # "_lsl") GPR:$Rn, GPR:$Rm, 0)>;
+
+  def : Pat<(opfrag ty:$Rn, ty:$Rm),
+            (!cast<Instruction>(prefix # "_lsl") $Rn, $Rm, 0)>;
+}
+
+defm CMPww : cmp_shifts<"CMPww", 0b0, 0b1, 0b0, "cmp", A64cmp, i32, GPR32>;
+defm CMPxx : cmp_shifts<"CMPxx", 0b1, 0b1, 0b0, "cmp", A64cmp, i64, GPR64>;
+
+defm CMNww : cmp_shifts<"CMNww", 0b0, 0b0, 0b1, "cmn", A64cmn, i32, GPR32>;
+defm CMNxx : cmp_shifts<"CMNxx", 0b1, 0b0, 0b1, "cmn", A64cmn, i64, GPR64>;
+
+//===----------------------------------------------------------------------===//
+// Add-subtract (with carry) instructions
+//===----------------------------------------------------------------------===//
+// Contains: ADC, ADCS, SBC, SBCS + aliases NGC, NGCS
+
+multiclass A64I_addsubcarrySizes<bit op, bit s, string asmop> {
+  let Uses = [NZCV] in {
+    def www : A64I_addsubcarry<0b0, op, s, 0b000000,
+                               (outs GPR32:$Rd), (ins GPR32:$Rn, GPR32:$Rm),
+                               !strconcat(asmop, "\t$Rd, $Rn, $Rm"),
+                               [], NoItinerary>;
+
+    def xxx : A64I_addsubcarry<0b1, op, s, 0b000000,
+                               (outs GPR64:$Rd), (ins GPR64:$Rn, GPR64:$Rm),
+                               !strconcat(asmop, "\t$Rd, $Rn, $Rm"),
+                               [], NoItinerary>;
+  }
+}
+
+let isCommutable = 1 in {
+  defm ADC : A64I_addsubcarrySizes<0b0, 0b0, "adc">;
+}
+
+defm SBC : A64I_addsubcarrySizes<0b1, 0b0, "sbc">;
+
+let Defs = [NZCV] in {
+  let isCommutable = 1 in {
+    defm ADCS : A64I_addsubcarrySizes<0b0, 0b1, "adcs">;
+  }
+
+  defm SBCS : A64I_addsubcarrySizes<0b1, 0b1, "sbcs">;
+}
+
+def : InstAlias<"ngc $Rd, $Rm", (SBCwww GPR32:$Rd, WZR, GPR32:$Rm)>;
+def : InstAlias<"ngc $Rd, $Rm", (SBCxxx GPR64:$Rd, XZR, GPR64:$Rm)>;
+def : InstAlias<"ngcs $Rd, $Rm", (SBCSwww GPR32:$Rd, WZR, GPR32:$Rm)>;
+def : InstAlias<"ngcs $Rd, $Rm", (SBCSxxx GPR64:$Rd, XZR, GPR64:$Rm)>;
+
+// Note that adde and sube can form a chain longer than two (e.g. for 256-bit
+// addition). So the flag-setting instructions are appropriate.
+def : Pat<(adde i32:$Rn, i32:$Rm), (ADCSwww $Rn, $Rm)>;
+def : Pat<(adde i64:$Rn, i64:$Rm), (ADCSxxx $Rn, $Rm)>;
+def : Pat<(sube i32:$Rn, i32:$Rm), (SBCSwww $Rn, $Rm)>;
+def : Pat<(sube i64:$Rn, i64:$Rm), (SBCSxxx $Rn, $Rm)>;
+
+//===----------------------------------------------------------------------===//
+// Bitfield
+//===----------------------------------------------------------------------===//
+// Contains: SBFM, BFM, UBFM, [SU]XT[BHW], ASR, LSR, LSL, SBFI[ZX], BFI, BFXIL,
+//     UBFIZ, UBFX
+
+// Because of the rather complicated nearly-overlapping aliases, the decoding of
+// this range of instructions is handled manually. The architectural
+// instructions are BFM, SBFM and UBFM but a disassembler should never produce
+// these.
+//
+// In the end, the best option was to use BFM instructions for decoding under
+// almost all circumstances, but to create aliasing *Instructions* for each of
+// the canonical forms and specify a completely custom decoder which would
+// substitute the correct MCInst as needed.
+//
+// This also simplifies instruction selection, parsing etc because the MCInsts
+// have a shape that's closer to their use in code.
+
+//===-------------------------------
+// 1. The architectural BFM instructions
+//===-------------------------------
+
+def uimm5_asmoperand : AsmOperandClass {
+  let Name = "UImm5";
+  let PredicateMethod = "isUImm<5>";
+  let RenderMethod = "addImmOperands";
+  let DiagnosticType = "UImm5";
+}
+
+def uimm6_asmoperand : AsmOperandClass {
+  let Name = "UImm6";
+  let PredicateMethod = "isUImm<6>";
+  let RenderMethod = "addImmOperands";
+  let DiagnosticType = "UImm6";
+}
+
+def bitfield32_imm : Operand<i64>,
+                     ImmLeaf<i64, [{ return Imm >= 0 && Imm < 32; }]> {
+  let ParserMatchClass = uimm5_asmoperand;
+
+  let DecoderMethod = "DecodeBitfield32ImmOperand";
+}
+
+
+def bitfield64_imm : Operand<i64>,
+                     ImmLeaf<i64, [{ return Imm >= 0 && Imm < 64; }]> {
+  let ParserMatchClass = uimm6_asmoperand;
+
+  // Default decoder works in 64-bit case: the 6-bit field can take any value.
+}
+
+multiclass A64I_bitfieldSizes<bits<2> opc, string asmop> {
+  def wwii : A64I_bitfield<0b0, opc, 0b0, (outs GPR32:$Rd),
+                    (ins GPR32:$Rn, bitfield32_imm:$ImmR, bitfield32_imm:$ImmS),
+                    !strconcat(asmop, "\t$Rd, $Rn, $ImmR, $ImmS"),
+                    [], NoItinerary> {
+    let DecoderMethod = "DecodeBitfieldInstruction";
+  }
+
+  def xxii : A64I_bitfield<0b1, opc, 0b1, (outs GPR64:$Rd),
+                    (ins GPR64:$Rn, bitfield64_imm:$ImmR, bitfield64_imm:$ImmS),
+                    !strconcat(asmop, "\t$Rd, $Rn, $ImmR, $ImmS"),
+                    [], NoItinerary> {
+    let DecoderMethod = "DecodeBitfieldInstruction";
+  }
+}
+
+defm SBFM : A64I_bitfieldSizes<0b00, "sbfm">;
+defm UBFM : A64I_bitfieldSizes<0b10, "ubfm">;
+
+// BFM instructions modify the destination register rather than defining it
+// completely.
+def BFMwwii :
+  A64I_bitfield<0b0, 0b01, 0b0, (outs GPR32:$Rd),
+        (ins GPR32:$src, GPR32:$Rn, bitfield32_imm:$ImmR, bitfield32_imm:$ImmS),
+        "bfm\t$Rd, $Rn, $ImmR, $ImmS", [], NoItinerary> {
+  let DecoderMethod = "DecodeBitfieldInstruction";
+  let Constraints = "$src = $Rd";
+}
+
+def BFMxxii :
+  A64I_bitfield<0b1, 0b01, 0b1, (outs GPR64:$Rd),
+        (ins GPR64:$src, GPR64:$Rn, bitfield64_imm:$ImmR, bitfield64_imm:$ImmS),
+        "bfm\t$Rd, $Rn, $ImmR, $ImmS", [], NoItinerary> {
+  let DecoderMethod = "DecodeBitfieldInstruction";
+  let Constraints = "$src = $Rd";
+}
+
+
+//===-------------------------------
+// 2. Extend aliases to 64-bit dest
+//===-------------------------------
+
+// Unfortunately the extensions that end up as 64-bits cannot be handled by an
+// instruction alias: their syntax is (for example) "SXTB x0, w0", which needs
+// to be mapped to "SBFM x0, x0, #0, 7" (changing the class of Rn). InstAlias is
+// not capable of such a map as far as I'm aware
+
+// Note that these instructions are strictly more specific than the
+// BFM ones (in ImmR) so they can handle their own decoding.
+class A64I_bf_ext<bit sf, bits<2> opc, RegisterClass GPRDest, ValueType dty,
+                    string asmop, bits<6> imms, dag pattern>
+  : A64I_bitfield<sf, opc, sf,
+                  (outs GPRDest:$Rd), (ins GPR32:$Rn),
+                  !strconcat(asmop, "\t$Rd, $Rn"),
+                  [(set dty:$Rd, pattern)], NoItinerary> {
+  let ImmR = 0b000000;
+  let ImmS = imms;
+}
+
+// Signed extensions
+def SXTBxw : A64I_bf_ext<0b1, 0b00, GPR64, i64, "sxtb", 7,
+                         (sext_inreg (anyext i32:$Rn), i8)>;
+def SXTBww : A64I_bf_ext<0b0, 0b00, GPR32, i32, "sxtb", 7,
+                         (sext_inreg i32:$Rn, i8)>;
+def SXTHxw : A64I_bf_ext<0b1, 0b00, GPR64, i64, "sxth", 15,
+                         (sext_inreg (anyext i32:$Rn), i16)>;
+def SXTHww : A64I_bf_ext<0b0, 0b00, GPR32, i32, "sxth", 15,
+                         (sext_inreg i32:$Rn, i16)>;
+def SXTWxw : A64I_bf_ext<0b1, 0b00, GPR64, i64, "sxtw", 31, (sext i32:$Rn)>;
+
+// Unsigned extensions
+def UXTBww : A64I_bf_ext<0b0, 0b10, GPR32, i32, "uxtb", 7,
+                         (and i32:$Rn, 255)>;
+def UXTHww : A64I_bf_ext<0b0, 0b10, GPR32, i32, "uxth", 15,
+                         (and i32:$Rn, 65535)>;
+
+// The 64-bit unsigned variants are not strictly architectural but recommended
+// for consistency.
+let isAsmParserOnly = 1 in {
+  def UXTBxw : A64I_bf_ext<0b0, 0b10, GPR64, i64, "uxtb", 7,
+                           (and (anyext i32:$Rn), 255)>;
+  def UXTHxw : A64I_bf_ext<0b0, 0b10, GPR64, i64, "uxth", 15,
+                           (and (anyext i32:$Rn), 65535)>;
+}
+
+// Extra patterns for when the source register is actually 64-bits
+// too. There's no architectural difference here, it's just LLVM
+// shinanigans. There's no need for equivalent zero-extension patterns
+// because they'll already be caught by logical (immediate) matching.
+def : Pat<(sext_inreg i64:$Rn, i8),
+          (SXTBxw (EXTRACT_SUBREG $Rn, sub_32))>;
+def : Pat<(sext_inreg i64:$Rn, i16),
+          (SXTHxw (EXTRACT_SUBREG $Rn, sub_32))>;
+def : Pat<(sext_inreg i64:$Rn, i32),
+          (SXTWxw (EXTRACT_SUBREG $Rn, sub_32))>;
+
+
+//===-------------------------------
+// 3. Aliases for ASR and LSR (the simple shifts)
+//===-------------------------------
+
+// These also handle their own decoding because ImmS being set makes
+// them take precedence over BFM.
+multiclass A64I_shift<bits<2> opc, string asmop, SDNode opnode> {
+  def wwi : A64I_bitfield<0b0, opc, 0b0,
+                    (outs GPR32:$Rd), (ins GPR32:$Rn, bitfield32_imm:$ImmR),
+                    !strconcat(asmop, "\t$Rd, $Rn, $ImmR"),
+                    [(set i32:$Rd, (opnode i32:$Rn, bitfield32_imm:$ImmR))],
+                    NoItinerary> {
+    let ImmS = 31;
+  }
+
+  def xxi : A64I_bitfield<0b1, opc, 0b1,
+                    (outs GPR64:$Rd), (ins GPR64:$Rn, bitfield64_imm:$ImmR),
+                    !strconcat(asmop, "\t$Rd, $Rn, $ImmR"),
+                    [(set i64:$Rd, (opnode i64:$Rn, bitfield64_imm:$ImmR))],
+                    NoItinerary> {
+    let ImmS = 63;
+  }
+
+}
+
+defm ASR : A64I_shift<0b00, "asr", sra>;
+defm LSR : A64I_shift<0b10, "lsr", srl>;
+
+//===-------------------------------
+// 4. Aliases for LSL
+//===-------------------------------
+
+// Unfortunately LSL and subsequent aliases are much more complicated. We need
+// to be able to say certain output instruction fields depend in a complex
+// manner on combinations of input assembly fields).
+//
+// MIOperandInfo *might* have been able to do it, but at the cost of
+// significantly more C++ code.
+
+// N.b. contrary to usual practice these operands store the shift rather than
+// the machine bits in an MCInst. The complexity overhead of consistency
+// outweighed the benefits in this case (custom asmparser, printer and selection
+// vs custom encoder).
+def bitfield32_lsl_imm : Operand<i64>,
+                         ImmLeaf<i64, [{ return Imm >= 0 && Imm <= 31; }]> {
+  let ParserMatchClass = uimm5_asmoperand;
+  let EncoderMethod = "getBitfield32LSLOpValue";
+}
+
+def bitfield64_lsl_imm : Operand<i64>,
+                         ImmLeaf<i64, [{ return Imm >= 0 && Imm <= 63; }]> {
+  let ParserMatchClass = uimm6_asmoperand;
+  let EncoderMethod = "getBitfield64LSLOpValue";
+}
+
+class A64I_bitfield_lsl<bit sf, RegisterClass GPR, ValueType ty,
+                        Operand operand>
+  : A64I_bitfield<sf, 0b10, sf, (outs GPR:$Rd), (ins GPR:$Rn, operand:$FullImm),
+                  "lsl\t$Rd, $Rn, $FullImm",
+                  [(set ty:$Rd, (shl ty:$Rn, operand:$FullImm))],
+                  NoItinerary> {
+  bits<12> FullImm;
+  let ImmR = FullImm{5-0};
+  let ImmS = FullImm{11-6};
+
+  // No disassembler allowed because it would overlap with BFM which does the
+  // actual work.
+  let isAsmParserOnly = 1;
+}
+
+def LSLwwi : A64I_bitfield_lsl<0b0, GPR32, i32, bitfield32_lsl_imm>;
+def LSLxxi : A64I_bitfield_lsl<0b1, GPR64, i64, bitfield64_lsl_imm>;
+
+//===-------------------------------
+// 5. Aliases for bitfield extract instructions
+//===-------------------------------
+
+def bfx32_width_asmoperand : AsmOperandClass {
+  let Name = "BFX32Width";
+  let PredicateMethod = "isBitfieldWidth<32>";
+  let RenderMethod = "addBFXWidthOperands";
+  let DiagnosticType = "Width32";
+}
+
+def bfx32_width : Operand<i64>, ImmLeaf<i64, [{ return true; }]> {
+  let PrintMethod = "printBFXWidthOperand";
+  let ParserMatchClass = bfx32_width_asmoperand;
+}
+
+def bfx64_width_asmoperand : AsmOperandClass {
+  let Name = "BFX64Width";
+  let PredicateMethod = "isBitfieldWidth<64>";
+  let RenderMethod = "addBFXWidthOperands";
+  let DiagnosticType = "Width64";
+}
+
+def bfx64_width : Operand<i64> {
+  let PrintMethod = "printBFXWidthOperand";
+  let ParserMatchClass = bfx64_width_asmoperand;
+}
+
+
+multiclass A64I_bitfield_extract<bits<2> opc, string asmop, SDNode op> {
+  def wwii : A64I_bitfield<0b0, opc, 0b0, (outs GPR32:$Rd),
+                       (ins GPR32:$Rn, bitfield32_imm:$ImmR, bfx32_width:$ImmS),
+                       !strconcat(asmop, "\t$Rd, $Rn, $ImmR, $ImmS"),
+                       [(set i32:$Rd, (op i32:$Rn, imm:$ImmR, imm:$ImmS))],
+                       NoItinerary> {
+    // As above, no disassembler allowed.
+    let isAsmParserOnly = 1;
+  }
+
+  def xxii : A64I_bitfield<0b1, opc, 0b1, (outs GPR64:$Rd),
+                       (ins GPR64:$Rn, bitfield64_imm:$ImmR, bfx64_width:$ImmS),
+                       !strconcat(asmop, "\t$Rd, $Rn, $ImmR, $ImmS"),
+                       [(set i64:$Rd, (op i64:$Rn, imm:$ImmR, imm:$ImmS))],
+                       NoItinerary> {
+    // As above, no disassembler allowed.
+    let isAsmParserOnly = 1;
+  }
+}
+
+defm SBFX :  A64I_bitfield_extract<0b00, "sbfx", A64Sbfx>;
+defm UBFX :  A64I_bitfield_extract<0b10, "ubfx", A64Ubfx>;
+
+// Again, variants based on BFM modify Rd so need it as an input too.
+def BFXILwwii : A64I_bitfield<0b0, 0b01, 0b0, (outs GPR32:$Rd),
+           (ins GPR32:$src, GPR32:$Rn, bitfield32_imm:$ImmR, bfx32_width:$ImmS),
+           "bfxil\t$Rd, $Rn, $ImmR, $ImmS", [], NoItinerary> {
+  // As above, no disassembler allowed.
+  let isAsmParserOnly = 1;
+  let Constraints = "$src = $Rd";
+}
+
+def BFXILxxii : A64I_bitfield<0b1, 0b01, 0b1, (outs GPR64:$Rd),
+           (ins GPR64:$src, GPR64:$Rn, bitfield64_imm:$ImmR, bfx64_width:$ImmS),
+           "bfxil\t$Rd, $Rn, $ImmR, $ImmS", [], NoItinerary> {
+  // As above, no disassembler allowed.
+  let isAsmParserOnly = 1;
+  let Constraints = "$src = $Rd";
+}
+
+// SBFX instructions can do a 1-instruction sign-extension of boolean values.
+def : Pat<(sext_inreg i64:$Rn, i1), (SBFXxxii $Rn, 0, 0)>;
+def : Pat<(sext_inreg i32:$Rn, i1), (SBFXwwii $Rn, 0, 0)>;
+def : Pat<(i64 (sext_inreg (anyext i32:$Rn), i1)),
+          (SBFXxxii (SUBREG_TO_REG (i64 0), $Rn, sub_32), 0, 0)>;
+
+// UBFX makes sense as an implementation of a 64-bit zero-extension too. Could
+// use either 64-bit or 32-bit variant, but 32-bit might be more efficient.
+def : Pat<(zext i32:$Rn), (SUBREG_TO_REG (i64 0), (UBFXwwii $Rn, 0, 31),
+                                         sub_32)>;
+
+//===-------------------------------
+// 6. Aliases for bitfield insert instructions
+//===-------------------------------
+
+def bfi32_lsb_asmoperand : AsmOperandClass {
+  let Name = "BFI32LSB";
+  let PredicateMethod = "isUImm<5>";
+  let RenderMethod = "addBFILSBOperands<32>";
+  let DiagnosticType = "UImm5";
+}
+
+def bfi32_lsb : Operand<i64>,
+                ImmLeaf<i64, [{ return Imm >= 0 && Imm <= 31; }]> {
+  let PrintMethod = "printBFILSBOperand<32>";
+  let ParserMatchClass = bfi32_lsb_asmoperand;
+}
+
+def bfi64_lsb_asmoperand : AsmOperandClass {
+  let Name = "BFI64LSB";
+  let PredicateMethod = "isUImm<6>";
+  let RenderMethod = "addBFILSBOperands<64>";
+  let DiagnosticType = "UImm6";
+}
+
+def bfi64_lsb : Operand<i64>,
+                ImmLeaf<i64, [{ return Imm >= 0 && Imm <= 63; }]> {
+  let PrintMethod = "printBFILSBOperand<64>";
+  let ParserMatchClass = bfi64_lsb_asmoperand;
+}
+
+// Width verification is performed during conversion so width operand can be
+// shared between 32/64-bit cases. Still needed for the print method though
+// because ImmR encodes "width - 1".
+def bfi32_width_asmoperand : AsmOperandClass {
+  let Name = "BFI32Width";
+  let PredicateMethod = "isBitfieldWidth<32>";
+  let RenderMethod = "addBFIWidthOperands";
+  let DiagnosticType = "Width32";
+}
+
+def bfi32_width : Operand<i64>,
+                  ImmLeaf<i64, [{ return Imm >= 1 && Imm <= 32; }]> {
+  let PrintMethod = "printBFIWidthOperand";
+  let ParserMatchClass = bfi32_width_asmoperand;
+}
+
+def bfi64_width_asmoperand : AsmOperandClass {
+  let Name = "BFI64Width";
+  let PredicateMethod = "isBitfieldWidth<64>";
+  let RenderMethod = "addBFIWidthOperands";
+  let DiagnosticType = "Width64";
+}
+
+def bfi64_width : Operand<i64>,
+                  ImmLeaf<i64, [{ return Imm >= 1 && Imm <= 64; }]> {
+  let PrintMethod = "printBFIWidthOperand";
+  let ParserMatchClass = bfi64_width_asmoperand;
+}
+
+multiclass A64I_bitfield_insert<bits<2> opc, string asmop> {
+  def wwii : A64I_bitfield<0b0, opc, 0b0, (outs GPR32:$Rd),
+                           (ins GPR32:$Rn, bfi32_lsb:$ImmR, bfi32_width:$ImmS),
+                           !strconcat(asmop, "\t$Rd, $Rn, $ImmR, $ImmS"),
+                           [], NoItinerary> {
+    // As above, no disassembler allowed.
+    let isAsmParserOnly = 1;
+  }
+
+  def xxii : A64I_bitfield<0b1, opc, 0b1, (outs GPR64:$Rd),
+                           (ins GPR64:$Rn, bfi64_lsb:$ImmR, bfi64_width:$ImmS),
+                           !strconcat(asmop, "\t$Rd, $Rn, $ImmR, $ImmS"),
+                           [], NoItinerary> {
+    // As above, no disassembler allowed.
+    let isAsmParserOnly = 1;
+  }
+}
+
+defm SBFIZ :  A64I_bitfield_insert<0b00, "sbfiz">;
+defm UBFIZ :  A64I_bitfield_insert<0b10, "ubfiz">;
+
+
+def BFIwwii : A64I_bitfield<0b0, 0b01, 0b0, (outs GPR32:$Rd),
+                (ins GPR32:$src, GPR32:$Rn, bfi32_lsb:$ImmR, bfi32_width:$ImmS),
+                "bfi\t$Rd, $Rn, $ImmR, $ImmS", [], NoItinerary> {
+  // As above, no disassembler allowed.
+  let isAsmParserOnly = 1;
+  let Constraints = "$src = $Rd";
+}
+
+def BFIxxii : A64I_bitfield<0b1, 0b01, 0b1, (outs GPR64:$Rd),
+                (ins GPR64:$src, GPR64:$Rn, bfi64_lsb:$ImmR, bfi64_width:$ImmS),
+                "bfi\t$Rd, $Rn, $ImmR, $ImmS", [], NoItinerary> {
+  // As above, no disassembler allowed.
+  let isAsmParserOnly = 1;
+  let Constraints = "$src = $Rd";
+}
+
+//===----------------------------------------------------------------------===//
+// Compare and branch (immediate)
+//===----------------------------------------------------------------------===//
+// Contains: CBZ, CBNZ
+
+class label_asmoperand<int width, int scale> : AsmOperandClass {
+  let Name = "Label" # width # "_" # scale;
+  let PredicateMethod = "isLabel<" # width # "," # scale # ">";
+  let RenderMethod = "addLabelOperands<" # width # ", " # scale # ">";
+  let DiagnosticType = "Label";
+}
+
+def label_wid19_scal4_asmoperand : label_asmoperand<19, 4>;
+
+// All conditional immediate branches are the same really: 19 signed bits scaled
+// by the instruction-size (4).
+def bcc_target : Operand<OtherVT> {
+  // This label is a 19-bit offset from PC, scaled by the instruction-width: 4.
+  let ParserMatchClass = label_wid19_scal4_asmoperand;
+  let PrintMethod = "printLabelOperand<19, 4>";
+  let EncoderMethod = "getLabelOpValue<AArch64::fixup_a64_condbr>";
+  let OperandType = "OPERAND_PCREL";
+}
+
+multiclass cmpbr_sizes<bit op, string asmop, ImmLeaf SETOP> {
+  let isBranch = 1, isTerminator = 1 in {
+  def x : A64I_cmpbr<0b1, op,
+                     (outs),
+                     (ins GPR64:$Rt, bcc_target:$Label),
+                     !strconcat(asmop,"\t$Rt, $Label"),
+                     [(A64br_cc (A64cmp i64:$Rt, 0), SETOP, bb:$Label)],
+                     NoItinerary>;
+
+  def w : A64I_cmpbr<0b0, op,
+                     (outs),
+                     (ins GPR32:$Rt, bcc_target:$Label),
+                     !strconcat(asmop,"\t$Rt, $Label"),
+                     [(A64br_cc (A64cmp i32:$Rt, 0), SETOP, bb:$Label)],
+                     NoItinerary>;
+  }
+}
+
+defm CBZ  : cmpbr_sizes<0b0, "cbz",  ImmLeaf<i32, [{
+  return Imm == A64CC::EQ;
+}]> >;
+defm CBNZ : cmpbr_sizes<0b1, "cbnz", ImmLeaf<i32, [{
+  return Imm == A64CC::NE;
+}]> >;
+
+//===----------------------------------------------------------------------===//
+// Conditional branch (immediate) instructions
+//===----------------------------------------------------------------------===//
+// Contains: B.cc
+
+def cond_code_asmoperand : AsmOperandClass {
+  let Name = "CondCode";
+  let DiagnosticType = "CondCode";
+}
+
+def cond_code : Operand<i32>, ImmLeaf<i32, [{
+  return Imm >= 0 && Imm <= 15;
+}]> {
+  let PrintMethod = "printCondCodeOperand";
+  let ParserMatchClass = cond_code_asmoperand;
+}
+
+def Bcc : A64I_condbr<0b0, 0b0, (outs),
+                (ins cond_code:$Cond, bcc_target:$Label),
+                "b.$Cond $Label", [(A64br_cc NZCV, (i32 imm:$Cond), bb:$Label)],
+                NoItinerary> {
+  let Uses = [NZCV];
+  let isBranch = 1;
+  let isTerminator = 1;
+}
+
+//===----------------------------------------------------------------------===//
+// Conditional compare (immediate) instructions
+//===----------------------------------------------------------------------===//
+// Contains: CCMN, CCMP
+
+def uimm4_asmoperand : AsmOperandClass {
+  let Name = "UImm4";
+  let PredicateMethod = "isUImm<4>";
+  let RenderMethod = "addImmOperands";
+  let DiagnosticType = "UImm4";
+}
+
+def uimm4 : Operand<i32> {
+  let ParserMatchClass = uimm4_asmoperand;
+}
+
+def uimm5 : Operand<i32> {
+  let ParserMatchClass = uimm5_asmoperand;
+}
+
+// The only difference between this operand and the one for instructions like
+// B.cc is that it's parsed manually. The other get parsed implicitly as part of
+// the mnemonic handling.
+def cond_code_op_asmoperand : AsmOperandClass {
+  let Name = "CondCodeOp";
+  let RenderMethod = "addCondCodeOperands";
+  let PredicateMethod = "isCondCode";
+  let ParserMethod = "ParseCondCodeOperand";
+  let DiagnosticType = "CondCode";
+}
+
+def cond_code_op : Operand<i32> {
+  let PrintMethod = "printCondCodeOperand";
+  let ParserMatchClass = cond_code_op_asmoperand;
+}
+
+class A64I_condcmpimmImpl<bit sf, bit op, RegisterClass GPR, string asmop>
+  : A64I_condcmpimm<sf, op, 0b0, 0b0, 0b1, (outs),
+                (ins GPR:$Rn, uimm5:$UImm5, uimm4:$NZCVImm, cond_code_op:$Cond),
+                !strconcat(asmop, "\t$Rn, $UImm5, $NZCVImm, $Cond"),
+                [], NoItinerary> {
+  let Defs = [NZCV];
+}
+
+def CCMNwi : A64I_condcmpimmImpl<0b0, 0b0, GPR32, "ccmn">;
+def CCMNxi : A64I_condcmpimmImpl<0b1, 0b0, GPR64, "ccmn">;
+def CCMPwi : A64I_condcmpimmImpl<0b0, 0b1, GPR32, "ccmp">;
+def CCMPxi : A64I_condcmpimmImpl<0b1, 0b1, GPR64, "ccmp">;
+
+//===----------------------------------------------------------------------===//
+// Conditional compare (register) instructions
+//===----------------------------------------------------------------------===//
+// Contains: CCMN, CCMP
+
+class A64I_condcmpregImpl<bit sf, bit op, RegisterClass GPR, string asmop>
+  : A64I_condcmpreg<sf, op, 0b0, 0b0, 0b1,
+                    (outs),
+                    (ins GPR:$Rn, GPR:$Rm, uimm4:$NZCVImm, cond_code_op:$Cond),
+                    !strconcat(asmop, "\t$Rn, $Rm, $NZCVImm, $Cond"),
+                    [], NoItinerary> {
+  let Defs = [NZCV];
+}
+
+def CCMNww : A64I_condcmpregImpl<0b0, 0b0, GPR32, "ccmn">;
+def CCMNxx : A64I_condcmpregImpl<0b1, 0b0, GPR64, "ccmn">;
+def CCMPww : A64I_condcmpregImpl<0b0, 0b1, GPR32, "ccmp">;
+def CCMPxx : A64I_condcmpregImpl<0b1, 0b1, GPR64, "ccmp">;
+
+//===----------------------------------------------------------------------===//
+// Conditional select instructions
+//===----------------------------------------------------------------------===//
+// Contains: CSEL, CSINC, CSINV, CSNEG + aliases CSET, CSETM, CINC, CINV, CNEG
+
+// Condition code which is encoded as the inversion (semantically rather than
+// bitwise) in the instruction.
+def inv_cond_code_op_asmoperand : AsmOperandClass {
+  let Name = "InvCondCodeOp";
+  let RenderMethod = "addInvCondCodeOperands";
+  let PredicateMethod = "isCondCode";
+  let ParserMethod = "ParseCondCodeOperand";
+  let DiagnosticType = "CondCode";
+}
+
+def inv_cond_code_op : Operand<i32> {
+  let ParserMatchClass = inv_cond_code_op_asmoperand;
+}
+
+// Having a separate operand for the selectable use-case is debatable, but gives
+// consistency with cond_code.
+def inv_cond_XFORM : SDNodeXForm<imm, [{
+  A64CC::CondCodes CC = static_cast<A64CC::CondCodes>(N->getZExtValue());
+  return CurDAG->getTargetConstant(A64InvertCondCode(CC), MVT::i32);
+}]>;
+
+def inv_cond_code
+  : ImmLeaf<i32, [{ return Imm >= 0 && Imm <= 15; }], inv_cond_XFORM>;
+
+
+multiclass A64I_condselSizes<bit op, bits<2> op2, string asmop,
+                             SDPatternOperator select> {
+  let Uses = [NZCV] in {
+    def wwwc : A64I_condsel<0b0, op, 0b0, op2,
+                            (outs GPR32:$Rd),
+                            (ins GPR32:$Rn, GPR32:$Rm, cond_code_op:$Cond),
+                            !strconcat(asmop, "\t$Rd, $Rn, $Rm, $Cond"),
+                            [(set i32:$Rd, (select i32:$Rn, i32:$Rm))],
+                            NoItinerary>;
+
+
+    def xxxc : A64I_condsel<0b1, op, 0b0, op2,
+                            (outs GPR64:$Rd),
+                            (ins GPR64:$Rn, GPR64:$Rm, cond_code_op:$Cond),
+                            !strconcat(asmop, "\t$Rd, $Rn, $Rm, $Cond"),
+                            [(set i64:$Rd, (select i64:$Rn, i64:$Rm))],
+                            NoItinerary>;
+  }
+}
+
+def simple_select
+  : PatFrag<(ops node:$lhs, node:$rhs),
+            (A64select_cc NZCV, node:$lhs, node:$rhs, (i32 imm:$Cond))>;
+
+class complex_select<SDPatternOperator opnode>
+  : PatFrag<(ops node:$lhs, node:$rhs),
+        (A64select_cc NZCV, node:$lhs, (opnode node:$rhs), (i32 imm:$Cond))>;
+
+
+defm CSEL : A64I_condselSizes<0b0, 0b00, "csel", simple_select>;
+defm CSINC : A64I_condselSizes<0b0, 0b01, "csinc",
+                               complex_select<PatFrag<(ops node:$val),
+                                                      (add node:$val, 1)>>>;
+defm CSINV : A64I_condselSizes<0b1, 0b00, "csinv", complex_select<not>>;
+defm CSNEG : A64I_condselSizes<0b1, 0b01, "csneg", complex_select<ineg>>;
+
+// Now the instruction aliases, which fit nicely into LLVM's model:
+
+def : InstAlias<"cset $Rd, $Cond",
+                (CSINCwwwc GPR32:$Rd, WZR, WZR, inv_cond_code_op:$Cond)>;
+def : InstAlias<"cset $Rd, $Cond",
+                (CSINCxxxc GPR64:$Rd, XZR, XZR, inv_cond_code_op:$Cond)>;
+def : InstAlias<"csetm $Rd, $Cond",
+                (CSINVwwwc GPR32:$Rd, WZR, WZR, inv_cond_code_op:$Cond)>;
+def : InstAlias<"csetm $Rd, $Cond",
+                (CSINVxxxc GPR64:$Rd, XZR, XZR, inv_cond_code_op:$Cond)>;
+def : InstAlias<"cinc $Rd, $Rn, $Cond",
+           (CSINCwwwc GPR32:$Rd, GPR32:$Rn, GPR32:$Rn, inv_cond_code_op:$Cond)>;
+def : InstAlias<"cinc $Rd, $Rn, $Cond",
+           (CSINCxxxc GPR64:$Rd, GPR64:$Rn, GPR64:$Rn, inv_cond_code_op:$Cond)>;
+def : InstAlias<"cinv $Rd, $Rn, $Cond",
+           (CSINVwwwc GPR32:$Rd, GPR32:$Rn, GPR32:$Rn, inv_cond_code_op:$Cond)>;
+def : InstAlias<"cinv $Rd, $Rn, $Cond",
+           (CSINVxxxc GPR64:$Rd, GPR64:$Rn, GPR64:$Rn, inv_cond_code_op:$Cond)>;
+def : InstAlias<"cneg $Rd, $Rn, $Cond",
+           (CSNEGwwwc GPR32:$Rd, GPR32:$Rn, GPR32:$Rn, inv_cond_code_op:$Cond)>;
+def : InstAlias<"cneg $Rd, $Rn, $Cond",
+           (CSNEGxxxc GPR64:$Rd, GPR64:$Rn, GPR64:$Rn, inv_cond_code_op:$Cond)>;
+
+// Finally some helper patterns.
+
+// For CSET (a.k.a. zero-extension of icmp)
+def : Pat<(A64select_cc NZCV, 0, 1, cond_code:$Cond),
+          (CSINCwwwc WZR, WZR, cond_code:$Cond)>;
+def : Pat<(A64select_cc NZCV, 1, 0, inv_cond_code:$Cond),
+          (CSINCwwwc WZR, WZR, inv_cond_code:$Cond)>;
+
+def : Pat<(A64select_cc NZCV, 0, 1, cond_code:$Cond),
+          (CSINCxxxc XZR, XZR, cond_code:$Cond)>;
+def : Pat<(A64select_cc NZCV, 1, 0, inv_cond_code:$Cond),
+          (CSINCxxxc XZR, XZR, inv_cond_code:$Cond)>;
+
+// For CSETM (a.k.a. sign-extension of icmp)
+def : Pat<(A64select_cc NZCV, 0, -1, cond_code:$Cond),
+          (CSINVwwwc WZR, WZR, cond_code:$Cond)>;
+def : Pat<(A64select_cc NZCV, -1, 0, inv_cond_code:$Cond),
+          (CSINVwwwc WZR, WZR, inv_cond_code:$Cond)>;
+
+def : Pat<(A64select_cc NZCV, 0, -1, cond_code:$Cond),
+          (CSINVxxxc XZR, XZR, cond_code:$Cond)>;
+def : Pat<(A64select_cc NZCV, -1, 0, inv_cond_code:$Cond),
+          (CSINVxxxc XZR, XZR, inv_cond_code:$Cond)>;
+
+// CINC, CINV and CNEG get dealt with automatically, which leaves the issue of
+// commutativity. The instructions are to complex for isCommutable to be used,
+// so we have to create the patterns manually:
+
+// No commutable pattern for CSEL since the commuted version is isomorphic.
+
+// CSINC
+def :Pat<(A64select_cc NZCV, (add i32:$Rm, 1), i32:$Rn, inv_cond_code:$Cond),
+         (CSINCwwwc $Rn, $Rm, inv_cond_code:$Cond)>;
+def :Pat<(A64select_cc NZCV, (add i64:$Rm, 1), i64:$Rn, inv_cond_code:$Cond),
+         (CSINCxxxc $Rn, $Rm, inv_cond_code:$Cond)>;
+
+// CSINV
+def :Pat<(A64select_cc NZCV, (not i32:$Rm), i32:$Rn, inv_cond_code:$Cond),
+         (CSINVwwwc $Rn, $Rm, inv_cond_code:$Cond)>;
+def :Pat<(A64select_cc NZCV, (not i64:$Rm), i64:$Rn, inv_cond_code:$Cond),
+         (CSINVxxxc $Rn, $Rm, inv_cond_code:$Cond)>;
+
+// CSNEG
+def :Pat<(A64select_cc NZCV, (ineg i32:$Rm), i32:$Rn, inv_cond_code:$Cond),
+         (CSNEGwwwc $Rn, $Rm, inv_cond_code:$Cond)>;
+def :Pat<(A64select_cc NZCV, (ineg i64:$Rm), i64:$Rn, inv_cond_code:$Cond),
+         (CSNEGxxxc $Rn, $Rm, inv_cond_code:$Cond)>;
+
+//===----------------------------------------------------------------------===//
+// Data Processing (1 source) instructions
+//===----------------------------------------------------------------------===//
+// Contains: RBIT, REV16, REV, REV32, CLZ, CLS.
+
+// We define an unary operator which always fails. We will use this to
+// define unary operators that cannot be matched.
+
+class A64I_dp_1src_impl<bit sf, bits<6> opcode, string asmop,
+                   list<dag> patterns, RegisterClass GPRrc,
+                   InstrItinClass itin>:
+      A64I_dp_1src<sf,
+                   0,
+                   0b00000,
+                   opcode,
+                   !strconcat(asmop, "\t$Rd, $Rn"),
+                   (outs GPRrc:$Rd),
+                   (ins GPRrc:$Rn),
+                   patterns,
+                   itin>;
+
+multiclass A64I_dp_1src <bits<6> opcode, string asmop> {
+  let hasSideEffects = 0 in {
+    def ww : A64I_dp_1src_impl<0b0, opcode, asmop, [], GPR32, NoItinerary>;
+    def xx : A64I_dp_1src_impl<0b1, opcode, asmop, [], GPR64, NoItinerary>;
+  }
+}
+
+defm RBIT  : A64I_dp_1src<0b000000, "rbit">;
+defm CLS   : A64I_dp_1src<0b000101, "cls">;
+defm CLZ   : A64I_dp_1src<0b000100, "clz">;
+
+def : Pat<(ctlz i32:$Rn), (CLZww $Rn)>;
+def : Pat<(ctlz i64:$Rn), (CLZxx $Rn)>;
+def : Pat<(ctlz_zero_undef i32:$Rn), (CLZww $Rn)>;
+def : Pat<(ctlz_zero_undef i64:$Rn), (CLZxx $Rn)>;
+
+def : Pat<(cttz i32:$Rn), (CLZww (RBITww $Rn))>;
+def : Pat<(cttz i64:$Rn), (CLZxx (RBITxx $Rn))>;
+def : Pat<(cttz_zero_undef i32:$Rn), (CLZww (RBITww $Rn))>;
+def : Pat<(cttz_zero_undef i64:$Rn), (CLZxx (RBITxx $Rn))>;
+
+
+def REVww : A64I_dp_1src_impl<0b0, 0b000010, "rev",
+                              [(set i32:$Rd, (bswap i32:$Rn))],
+                              GPR32, NoItinerary>;
+def REVxx : A64I_dp_1src_impl<0b1, 0b000011, "rev",
+                              [(set i64:$Rd, (bswap i64:$Rn))],
+                              GPR64, NoItinerary>;
+def REV32xx : A64I_dp_1src_impl<0b1, 0b000010, "rev32",
+                          [(set i64:$Rd, (bswap (rotr i64:$Rn, (i64 32))))],
+                          GPR64, NoItinerary>;
+def REV16ww : A64I_dp_1src_impl<0b0, 0b000001, "rev16",
+                          [(set i32:$Rd, (bswap (rotr i32:$Rn, (i64 16))))],
+                          GPR32,
+                          NoItinerary>;
+def REV16xx : A64I_dp_1src_impl<0b1, 0b000001, "rev16", [], GPR64, NoItinerary>;
+
+//===----------------------------------------------------------------------===//
+// Data Processing (2 sources) instructions
+//===----------------------------------------------------------------------===//
+// Contains: CRC32C?[BHWX], UDIV, SDIV, LSLV, LSRV, ASRV, RORV + aliases LSL,
+//           LSR, ASR, ROR
+
+
+class dp_2src_impl<bit sf, bits<6> opcode, string asmop, list<dag> patterns,
+                   RegisterClass GPRsp,
+                   InstrItinClass itin>:
+      A64I_dp_2src<sf,
+                   opcode,
+                   0,
+                   !strconcat(asmop, "\t$Rd, $Rn, $Rm"),
+                   (outs GPRsp:$Rd),
+                   (ins GPRsp:$Rn, GPRsp:$Rm),
+                   patterns,
+                   itin>;
+
+multiclass dp_2src_crc<bit c, string asmop> {
+  def B_www : dp_2src_impl<0b0, {0, 1, 0, c, 0, 0},
+                           !strconcat(asmop, "b"), [], GPR32, NoItinerary>;
+  def H_www : dp_2src_impl<0b0, {0, 1, 0, c, 0, 1},
+                           !strconcat(asmop, "h"), [], GPR32, NoItinerary>;
+  def W_www : dp_2src_impl<0b0, {0, 1, 0, c, 1, 0},
+                           !strconcat(asmop, "w"), [], GPR32, NoItinerary>;
+  def X_wwx : A64I_dp_2src<0b1, {0, 1, 0, c, 1, 1}, 0b0,
+                           !strconcat(asmop, "x\t$Rd, $Rn, $Rm"),
+                           (outs GPR32:$Rd), (ins GPR32:$Rn, GPR64:$Rm), [],
+                           NoItinerary>;
+}
+
+multiclass dp_2src_zext <bits<6> opcode, string asmop, SDPatternOperator op> {
+   def www : dp_2src_impl<0b0,
+                         opcode,
+                         asmop,
+                         [(set i32:$Rd,
+                               (op i32:$Rn, (i64 (zext i32:$Rm))))],
+                         GPR32,
+                         NoItinerary>;
+   def xxx : dp_2src_impl<0b1,
+                         opcode,
+                         asmop,
+                         [(set i64:$Rd, (op i64:$Rn, i64:$Rm))],
+                         GPR64,
+                         NoItinerary>;
+}
+
+
+multiclass dp_2src <bits<6> opcode, string asmop, SDPatternOperator op> {
+    def www : dp_2src_impl<0b0,
+                         opcode,
+                         asmop,
+                         [(set i32:$Rd, (op i32:$Rn, i32:$Rm))],
+                         GPR32,
+                         NoItinerary>;
+   def xxx : dp_2src_impl<0b1,
+                         opcode,
+                         asmop,
+                         [(set i64:$Rd, (op i64:$Rn, i64:$Rm))],
+                         GPR64,
+                         NoItinerary>;
+}
+
+// Here we define the data processing 2 source instructions.
+defm CRC32  : dp_2src_crc<0b0, "crc32">;
+defm CRC32C : dp_2src_crc<0b1, "crc32c">;
+
+defm UDIV : dp_2src<0b000010, "udiv", udiv>;
+defm SDIV : dp_2src<0b000011, "sdiv", sdiv>;
+
+defm LSLV : dp_2src_zext<0b001000, "lsl", shl>;
+defm LSRV : dp_2src_zext<0b001001, "lsr", srl>;
+defm ASRV : dp_2src_zext<0b001010, "asr", sra>;
+defm RORV : dp_2src_zext<0b001011, "ror", rotr>;
+
+// Extra patterns for an incoming 64-bit value for a 32-bit
+// operation. Since the LLVM operations are undefined (as in C) if the
+// RHS is out of range, it's perfectly permissible to discard the high
+// bits of the GPR64.
+def : Pat<(shl i32:$Rn, i64:$Rm),
+          (LSLVwww $Rn, (EXTRACT_SUBREG $Rm, sub_32))>;
+def : Pat<(srl i32:$Rn, i64:$Rm),
+          (LSRVwww $Rn, (EXTRACT_SUBREG $Rm, sub_32))>;
+def : Pat<(sra i32:$Rn, i64:$Rm),
+          (ASRVwww $Rn, (EXTRACT_SUBREG $Rm, sub_32))>;
+def : Pat<(rotr i32:$Rn, i64:$Rm),
+          (RORVwww $Rn, (EXTRACT_SUBREG $Rm, sub_32))>;
+
+// Here we define the aliases for the data processing 2 source instructions.
+def LSL_mnemonic : MnemonicAlias<"lslv", "lsl">;
+def LSR_mnemonic : MnemonicAlias<"lsrv", "lsr">;
+def ASR_menmonic : MnemonicAlias<"asrv", "asr">;
+def ROR_menmonic : MnemonicAlias<"rorv", "ror">;
+
+//===----------------------------------------------------------------------===//
+// Data Processing (3 sources) instructions
+//===----------------------------------------------------------------------===//
+// Contains: MADD, MSUB, SMADDL, SMSUBL, SMULH, UMADDL, UMSUBL, UMULH
+//    + aliases MUL, MNEG, SMULL, SMNEGL, UMULL, UMNEGL
+
+class A64I_dp3_4operand<bit sf, bits<6> opcode, RegisterClass AccReg,
+                        ValueType AccTy, RegisterClass SrcReg,
+                        string asmop, dag pattern>
+  : A64I_dp3<sf, opcode,
+             (outs AccReg:$Rd), (ins SrcReg:$Rn, SrcReg:$Rm, AccReg:$Ra),
+             !strconcat(asmop, "\t$Rd, $Rn, $Rm, $Ra"),
+             [(set AccTy:$Rd, pattern)], NoItinerary> {
+  RegisterClass AccGPR = AccReg;
+  RegisterClass SrcGPR = SrcReg;
+}
+
+def MADDwwww : A64I_dp3_4operand<0b0, 0b000000, GPR32, i32, GPR32, "madd",
+                                 (add i32:$Ra, (mul i32:$Rn, i32:$Rm))>;
+def MADDxxxx : A64I_dp3_4operand<0b1, 0b000000, GPR64, i64, GPR64, "madd",
+                                 (add i64:$Ra, (mul i64:$Rn, i64:$Rm))>;
+
+def MSUBwwww : A64I_dp3_4operand<0b0, 0b000001, GPR32, i32, GPR32, "msub",
+                                 (sub i32:$Ra, (mul i32:$Rn, i32:$Rm))>;
+def MSUBxxxx : A64I_dp3_4operand<0b1, 0b000001, GPR64, i64, GPR64, "msub",
+                                 (sub i64:$Ra, (mul i64:$Rn, i64:$Rm))>;
+
+def SMADDLxwwx : A64I_dp3_4operand<0b1, 0b000010, GPR64, i64, GPR32, "smaddl",
+                     (add i64:$Ra, (mul (i64 (sext i32:$Rn)), (sext i32:$Rm)))>;
+def SMSUBLxwwx : A64I_dp3_4operand<0b1, 0b000011, GPR64, i64, GPR32, "smsubl",
+                     (sub i64:$Ra, (mul (i64 (sext i32:$Rn)), (sext i32:$Rm)))>;
+
+def UMADDLxwwx : A64I_dp3_4operand<0b1, 0b001010, GPR64, i64, GPR32, "umaddl",
+                     (add i64:$Ra, (mul (i64 (zext i32:$Rn)), (zext i32:$Rm)))>;
+def UMSUBLxwwx : A64I_dp3_4operand<0b1, 0b001011, GPR64, i64, GPR32, "umsubl",
+                     (sub i64:$Ra, (mul (i64 (zext i32:$Rn)), (zext i32:$Rm)))>;
+
+let isCommutable = 1, PostEncoderMethod = "fixMulHigh" in {
+  def UMULHxxx : A64I_dp3<0b1, 0b001100, (outs GPR64:$Rd),
+                          (ins GPR64:$Rn, GPR64:$Rm),
+                          "umulh\t$Rd, $Rn, $Rm",
+                          [(set i64:$Rd, (mulhu i64:$Rn, i64:$Rm))],
+                          NoItinerary>;
+
+  def SMULHxxx : A64I_dp3<0b1, 0b000100, (outs GPR64:$Rd),
+                          (ins GPR64:$Rn, GPR64:$Rm),
+                          "smulh\t$Rd, $Rn, $Rm",
+                          [(set i64:$Rd, (mulhs i64:$Rn, i64:$Rm))],
+                          NoItinerary>;
+}
+
+multiclass A64I_dp3_3operand<string asmop, A64I_dp3_4operand INST,
+                             Register ZR, dag pattern> {
+  def : InstAlias<asmop # " $Rd, $Rn, $Rm",
+                  (INST INST.AccGPR:$Rd, INST.SrcGPR:$Rn, INST.SrcGPR:$Rm, ZR)>;
+
+  def : Pat<pattern, (INST $Rn, $Rm, ZR)>;
+}
+
+defm : A64I_dp3_3operand<"mul", MADDwwww, WZR, (mul i32:$Rn, i32:$Rm)>;
+defm : A64I_dp3_3operand<"mul", MADDxxxx, XZR, (mul i64:$Rn, i64:$Rm)>;
+
+defm : A64I_dp3_3operand<"mneg", MSUBwwww, WZR,
+                         (sub 0, (mul i32:$Rn, i32:$Rm))>;
+defm : A64I_dp3_3operand<"mneg", MSUBxxxx, XZR,
+                         (sub 0, (mul i64:$Rn, i64:$Rm))>;
+
+defm : A64I_dp3_3operand<"smull", SMADDLxwwx, XZR,
+                         (mul (i64 (sext i32:$Rn)), (sext i32:$Rm))>;
+defm : A64I_dp3_3operand<"smnegl", SMSUBLxwwx, XZR,
+                       (sub 0, (mul (i64 (sext i32:$Rn)), (sext i32:$Rm)))>;
+
+defm : A64I_dp3_3operand<"umull", UMADDLxwwx, XZR,
+                         (mul (i64 (zext i32:$Rn)), (zext i32:$Rm))>;
+defm : A64I_dp3_3operand<"umnegl", UMSUBLxwwx, XZR,
+                       (sub 0, (mul (i64 (zext i32:$Rn)), (zext i32:$Rm)))>;
+
+
+//===----------------------------------------------------------------------===//
+// Exception generation
+//===----------------------------------------------------------------------===//
+// Contains: SVC, HVC, SMC, BRK, HLT, DCPS1, DCPS2, DCPS3
+
+def uimm16_asmoperand : AsmOperandClass {
+  let Name = "UImm16";
+  let PredicateMethod = "isUImm<16>";
+  let RenderMethod = "addImmOperands";
+  let DiagnosticType = "UImm16";
+}
+
+def uimm16 : Operand<i32> {
+  let ParserMatchClass = uimm16_asmoperand;
+}
+
+class A64I_exceptImpl<bits<3> opc, bits<2> ll, string asmop>
+  : A64I_exception<opc, 0b000, ll, (outs), (ins uimm16:$UImm16),
+                   !strconcat(asmop, "\t$UImm16"), [], NoItinerary> {
+  let isBranch = 1;
+  let isTerminator = 1;
+}
+
+def SVCi : A64I_exceptImpl<0b000, 0b01, "svc">;
+def HVCi : A64I_exceptImpl<0b000, 0b10, "hvc">;
+def SMCi : A64I_exceptImpl<0b000, 0b11, "smc">;
+def BRKi : A64I_exceptImpl<0b001, 0b00, "brk">;
+def HLTi : A64I_exceptImpl<0b010, 0b00, "hlt">;
+
+def DCPS1i : A64I_exceptImpl<0b101, 0b01, "dcps1">;
+def DCPS2i : A64I_exceptImpl<0b101, 0b10, "dcps2">;
+def DCPS3i : A64I_exceptImpl<0b101, 0b11, "dcps3">;
+
+// The immediate is optional for the DCPS instructions, defaulting to 0.
+def : InstAlias<"dcps1", (DCPS1i 0)>;
+def : InstAlias<"dcps2", (DCPS2i 0)>;
+def : InstAlias<"dcps3", (DCPS3i 0)>;
+
+//===----------------------------------------------------------------------===//
+// Extract (immediate)
+//===----------------------------------------------------------------------===//
+// Contains: EXTR + alias ROR
+
+def EXTRwwwi : A64I_extract<0b0, 0b000, 0b0,
+                            (outs GPR32:$Rd),
+                            (ins GPR32:$Rn, GPR32:$Rm, bitfield32_imm:$LSB),
+                            "extr\t$Rd, $Rn, $Rm, $LSB",
+                            [(set i32:$Rd,
+                                  (A64Extr i32:$Rn, i32:$Rm, imm:$LSB))],
+                            NoItinerary>;
+def EXTRxxxi : A64I_extract<0b1, 0b000, 0b1,
+                            (outs GPR64:$Rd),
+                            (ins GPR64:$Rn, GPR64:$Rm, bitfield64_imm:$LSB),
+                            "extr\t$Rd, $Rn, $Rm, $LSB",
+                            [(set i64:$Rd,
+                                  (A64Extr i64:$Rn, i64:$Rm, imm:$LSB))],
+                            NoItinerary>;
+
+def : InstAlias<"ror $Rd, $Rs, $LSB",
+               (EXTRwwwi GPR32:$Rd, GPR32:$Rs, GPR32:$Rs, bitfield32_imm:$LSB)>;
+def : InstAlias<"ror $Rd, $Rs, $LSB",
+               (EXTRxxxi GPR64:$Rd, GPR64:$Rs, GPR64:$Rs, bitfield64_imm:$LSB)>;
+
+def : Pat<(rotr i32:$Rn, bitfield32_imm:$LSB),
+          (EXTRwwwi $Rn, $Rn, bitfield32_imm:$LSB)>;
+def : Pat<(rotr i64:$Rn, bitfield64_imm:$LSB),
+          (EXTRxxxi $Rn, $Rn, bitfield64_imm:$LSB)>;
+
+//===----------------------------------------------------------------------===//
+// Floating-point compare instructions
+//===----------------------------------------------------------------------===//
+// Contains: FCMP, FCMPE
+
+def fpzero_asmoperand : AsmOperandClass {
+  let Name = "FPZero";
+  let ParserMethod = "ParseFPImmOperand";
+  let DiagnosticType = "FPZero";
+}
+
+def fpz32 : Operand<f32>,
+            ComplexPattern<f32, 1, "SelectFPZeroOperand", [fpimm]> {
+  let ParserMatchClass = fpzero_asmoperand;
+  let PrintMethod = "printFPZeroOperand";
+  let DecoderMethod = "DecodeFPZeroOperand";
+}
+
+def fpz64 : Operand<f64>,
+            ComplexPattern<f64, 1, "SelectFPZeroOperand", [fpimm]> {
+  let ParserMatchClass = fpzero_asmoperand;
+  let PrintMethod = "printFPZeroOperand";
+  let DecoderMethod = "DecodeFPZeroOperand";
+}
+
+multiclass A64I_fpcmpSignal<bits<2> type, bit imm, dag ins, dag pattern> {
+  def _quiet : A64I_fpcmp<0b0, 0b0, type, 0b00, {0b0, imm, 0b0, 0b0, 0b0},
+                          (outs), ins, "fcmp\t$Rn, $Rm", [pattern],
+                          NoItinerary> {
+    let Defs = [NZCV];
+  }
+
+  def _sig : A64I_fpcmp<0b0, 0b0, type, 0b00, {0b1, imm, 0b0, 0b0, 0b0},
+                        (outs), ins, "fcmpe\t$Rn, $Rm", [], NoItinerary> {
+    let Defs = [NZCV];
+  }
+}
+
+defm FCMPss : A64I_fpcmpSignal<0b00, 0b0, (ins FPR32:$Rn, FPR32:$Rm),
+                               (set NZCV, (A64cmp f32:$Rn, f32:$Rm))>;
+defm FCMPdd : A64I_fpcmpSignal<0b01, 0b0, (ins FPR64:$Rn, FPR64:$Rm),
+                               (set NZCV, (A64cmp f64:$Rn, f64:$Rm))>;
+
+// What would be Rm should be written as 0; note that even though it's called
+// "$Rm" here to fit in with the InstrFormats, it's actually an immediate.
+defm FCMPsi : A64I_fpcmpSignal<0b00, 0b1, (ins FPR32:$Rn, fpz32:$Rm),
+                               (set NZCV, (A64cmp f32:$Rn, fpz32:$Rm))>;
+
+defm FCMPdi : A64I_fpcmpSignal<0b01, 0b1, (ins FPR64:$Rn, fpz64:$Rm),
+                               (set NZCV, (A64cmp f64:$Rn, fpz64:$Rm))>;
+
+
+//===----------------------------------------------------------------------===//
+// Floating-point conditional compare instructions
+//===----------------------------------------------------------------------===//
+// Contains: FCCMP, FCCMPE
+
+class A64I_fpccmpImpl<bits<2> type, bit op, RegisterClass FPR, string asmop>
+  : A64I_fpccmp<0b0, 0b0, type, op,
+                (outs),
+                (ins FPR:$Rn, FPR:$Rm, uimm4:$NZCVImm, cond_code_op:$Cond),
+                !strconcat(asmop, "\t$Rn, $Rm, $NZCVImm, $Cond"),
+                [], NoItinerary> {
+  let Defs = [NZCV];
+}
+
+def FCCMPss : A64I_fpccmpImpl<0b00, 0b0, FPR32, "fccmp">;
+def FCCMPEss : A64I_fpccmpImpl<0b00, 0b1, FPR32, "fccmpe">;
+def FCCMPdd : A64I_fpccmpImpl<0b01, 0b0, FPR64, "fccmp">;
+def FCCMPEdd : A64I_fpccmpImpl<0b01, 0b1, FPR64, "fccmpe">;
+
+//===----------------------------------------------------------------------===//
+// Floating-point conditional select instructions
+//===----------------------------------------------------------------------===//
+// Contains: FCSEL
+
+let Uses = [NZCV] in {
+  def FCSELsssc : A64I_fpcondsel<0b0, 0b0, 0b00, (outs FPR32:$Rd),
+                                 (ins FPR32:$Rn, FPR32:$Rm, cond_code_op:$Cond),
+                                 "fcsel\t$Rd, $Rn, $Rm, $Cond",
+                                 [(set f32:$Rd, 
+                                       (simple_select f32:$Rn, f32:$Rm))],
+                                 NoItinerary>;
+
+
+  def FCSELdddc : A64I_fpcondsel<0b0, 0b0, 0b01, (outs FPR64:$Rd),
+                                 (ins FPR64:$Rn, FPR64:$Rm, cond_code_op:$Cond),
+                                 "fcsel\t$Rd, $Rn, $Rm, $Cond",
+                                 [(set f64:$Rd,
+                                       (simple_select f64:$Rn, f64:$Rm))],
+                                 NoItinerary>;
+}
+
+//===----------------------------------------------------------------------===//
+// Floating-point data-processing (1 source)
+//===----------------------------------------------------------------------===//
+// Contains: FMOV, FABS, FNEG, FSQRT, FCVT, FRINT[NPMZAXI].
+
+def FPNoUnop : PatFrag<(ops node:$val), (fneg node:$val),
+                       [{ (void)N; return false; }]>;
+
+// First we do the fairly trivial bunch with uniform "OP s, s" and "OP d, d"
+// syntax. Default to no pattern because most are odd enough not to have one.
+multiclass A64I_fpdp1sizes<bits<6> opcode, string asmstr,
+                           SDPatternOperator opnode = FPNoUnop> {
+  def ss : A64I_fpdp1<0b0, 0b0, 0b00, opcode, (outs FPR32:$Rd), (ins FPR32:$Rn),
+                     !strconcat(asmstr, "\t$Rd, $Rn"),
+                     [(set f32:$Rd, (opnode f32:$Rn))],
+                     NoItinerary>;
+
+  def dd : A64I_fpdp1<0b0, 0b0, 0b01, opcode, (outs FPR64:$Rd), (ins FPR64:$Rn),
+                     !strconcat(asmstr, "\t$Rd, $Rn"),
+                     [(set f64:$Rd, (opnode f64:$Rn))],
+                     NoItinerary>;
+}
+
+defm FMOV   : A64I_fpdp1sizes<0b000000, "fmov">;
+defm FABS   : A64I_fpdp1sizes<0b000001, "fabs", fabs>;
+defm FNEG   : A64I_fpdp1sizes<0b000010, "fneg", fneg>;
+defm FSQRT  : A64I_fpdp1sizes<0b000011, "fsqrt", fsqrt>;
+
+defm FRINTN : A64I_fpdp1sizes<0b001000, "frintn">;
+defm FRINTP : A64I_fpdp1sizes<0b001001, "frintp", fceil>;
+defm FRINTM : A64I_fpdp1sizes<0b001010, "frintm", ffloor>;
+defm FRINTZ : A64I_fpdp1sizes<0b001011, "frintz", ftrunc>;
+defm FRINTA : A64I_fpdp1sizes<0b001100, "frinta">;
+defm FRINTX : A64I_fpdp1sizes<0b001110, "frintx", frint>;
+defm FRINTI : A64I_fpdp1sizes<0b001111, "frinti", fnearbyint>;
+
+// The FCVT instrucitons have different source and destination register-types,
+// but the fields are uniform everywhere a D-register (say) crops up. Package
+// this information in a Record.
+class FCVTRegType<RegisterClass rc, bits<2> fld, ValueType vt> {
+    RegisterClass Class = rc;
+    ValueType VT = vt;
+    bit t1 = fld{1};
+    bit t0 = fld{0};
+}
+
+def FCVT16 : FCVTRegType<FPR16, 0b11, f16>;
+def FCVT32 : FCVTRegType<FPR32, 0b00, f32>;
+def FCVT64 : FCVTRegType<FPR64, 0b01, f64>;
+
+class A64I_fpdp1_fcvt<FCVTRegType DestReg, FCVTRegType SrcReg, SDNode opnode>
+  : A64I_fpdp1<0b0, 0b0, {SrcReg.t1, SrcReg.t0},
+               {0,0,0,1, DestReg.t1, DestReg.t0},
+               (outs DestReg.Class:$Rd), (ins SrcReg.Class:$Rn),
+               "fcvt\t$Rd, $Rn",
+               [(set DestReg.VT:$Rd, (opnode SrcReg.VT:$Rn))], NoItinerary>;
+
+def FCVTds : A64I_fpdp1_fcvt<FCVT64, FCVT32, fextend>;
+def FCVThs : A64I_fpdp1_fcvt<FCVT16, FCVT32, fround>;
+def FCVTsd : A64I_fpdp1_fcvt<FCVT32, FCVT64, fround>;
+def FCVThd : A64I_fpdp1_fcvt<FCVT16, FCVT64, fround>;
+def FCVTsh : A64I_fpdp1_fcvt<FCVT32, FCVT16, fextend>;
+def FCVTdh : A64I_fpdp1_fcvt<FCVT64, FCVT16, fextend>;
+
+
+//===----------------------------------------------------------------------===//
+// Floating-point data-processing (2 sources) instructions
+//===----------------------------------------------------------------------===//
+// Contains: FMUL, FDIV, FADD, FSUB, FMAX, FMIN, FMAXNM, FMINNM, FNMUL
+
+def FPNoBinop : PatFrag<(ops node:$lhs, node:$rhs), (fadd node:$lhs, node:$rhs),
+                      [{ (void)N; return false; }]>;
+
+multiclass A64I_fpdp2sizes<bits<4> opcode, string asmstr,
+                           SDPatternOperator opnode> {
+  def sss : A64I_fpdp2<0b0, 0b0, 0b00, opcode,
+                      (outs FPR32:$Rd),
+                      (ins FPR32:$Rn, FPR32:$Rm),
+                      !strconcat(asmstr, "\t$Rd, $Rn, $Rm"),
+                      [(set f32:$Rd, (opnode f32:$Rn, f32:$Rm))],
+                      NoItinerary>;
+
+  def ddd : A64I_fpdp2<0b0, 0b0, 0b01, opcode,
+                      (outs FPR64:$Rd),
+                      (ins FPR64:$Rn, FPR64:$Rm),
+                      !strconcat(asmstr, "\t$Rd, $Rn, $Rm"),
+                      [(set f64:$Rd, (opnode f64:$Rn, f64:$Rm))],
+                      NoItinerary>;
+}
+
+let isCommutable = 1 in {
+  defm FMUL   : A64I_fpdp2sizes<0b0000, "fmul", fmul>;
+  defm FADD   : A64I_fpdp2sizes<0b0010, "fadd", fadd>;
+
+  // No patterns for these.
+  defm FMAX   : A64I_fpdp2sizes<0b0100, "fmax", FPNoBinop>;
+  defm FMIN   : A64I_fpdp2sizes<0b0101, "fmin", FPNoBinop>;
+  defm FMAXNM : A64I_fpdp2sizes<0b0110, "fmaxnm", FPNoBinop>;
+  defm FMINNM : A64I_fpdp2sizes<0b0111, "fminnm", FPNoBinop>;
+
+  defm FNMUL  : A64I_fpdp2sizes<0b1000, "fnmul",
+                                PatFrag<(ops node:$lhs, node:$rhs),
+                                        (fneg (fmul node:$lhs, node:$rhs))> >;
+}
+
+defm FDIV : A64I_fpdp2sizes<0b0001, "fdiv", fdiv>;
+defm FSUB : A64I_fpdp2sizes<0b0011, "fsub", fsub>;
+
+//===----------------------------------------------------------------------===//
+// Floating-point data-processing (3 sources) instructions
+//===----------------------------------------------------------------------===//
+// Contains: FMADD, FMSUB, FNMADD, FNMSUB
+
+def fmsub : PatFrag<(ops node:$Rn, node:$Rm, node:$Ra),
+                    (fma (fneg node:$Rn),  node:$Rm, node:$Ra)>;
+def fnmadd : PatFrag<(ops node:$Rn, node:$Rm, node:$Ra),
+                     (fma node:$Rn,  node:$Rm, (fneg node:$Ra))>;
+def fnmsub : PatFrag<(ops node:$Rn, node:$Rm, node:$Ra),
+                     (fma (fneg node:$Rn),  node:$Rm, (fneg node:$Ra))>;
+
+class A64I_fpdp3Impl<string asmop, RegisterClass FPR, ValueType VT,
+                     bits<2> type, bit o1, bit o0, SDPatternOperator fmakind>
+  : A64I_fpdp3<0b0, 0b0, type, o1, o0, (outs FPR:$Rd),
+               (ins FPR:$Rn, FPR:$Rm, FPR:$Ra),
+               !strconcat(asmop,"\t$Rd, $Rn, $Rm, $Ra"),
+               [(set VT:$Rd, (fmakind VT:$Rn, VT:$Rm, VT:$Ra))],
+               NoItinerary>;
+
+def FMADDssss  : A64I_fpdp3Impl<"fmadd",  FPR32, f32, 0b00, 0b0, 0b0, fma>;
+def FMSUBssss  : A64I_fpdp3Impl<"fmsub",  FPR32, f32, 0b00, 0b0, 0b1, fmsub>;
+def FNMADDssss : A64I_fpdp3Impl<"fnmadd", FPR32, f32, 0b00, 0b1, 0b0, fnmadd>;
+def FNMSUBssss : A64I_fpdp3Impl<"fnmsub", FPR32, f32, 0b00, 0b1, 0b1, fnmsub>;
+
+def FMADDdddd  : A64I_fpdp3Impl<"fmadd",  FPR64, f64, 0b01, 0b0, 0b0, fma>;
+def FMSUBdddd  : A64I_fpdp3Impl<"fmsub",  FPR64, f64, 0b01, 0b0, 0b1, fmsub>;
+def FNMADDdddd : A64I_fpdp3Impl<"fnmadd", FPR64, f64, 0b01, 0b1, 0b0, fnmadd>;
+def FNMSUBdddd : A64I_fpdp3Impl<"fnmsub", FPR64, f64, 0b01, 0b1, 0b1, fnmsub>;
+
+//===----------------------------------------------------------------------===//
+// Floating-point <-> fixed-point conversion instructions
+//===----------------------------------------------------------------------===//
+// Contains: FCVTZS, FCVTZU, SCVTF, UCVTF
+
+// #1-#32 allowed, encoded as "64 - <specified imm>
+def fixedpos_asmoperand_i32 : AsmOperandClass {
+  let Name = "CVTFixedPos32";
+  let RenderMethod = "addCVTFixedPosOperands";
+  let PredicateMethod = "isCVTFixedPos<32>";
+  let DiagnosticType = "CVTFixedPos32";
+}
+
+// Also encoded as "64 - <specified imm>" but #1-#64 allowed.
+def fixedpos_asmoperand_i64 : AsmOperandClass {
+  let Name = "CVTFixedPos64";
+  let RenderMethod = "addCVTFixedPosOperands";
+  let PredicateMethod = "isCVTFixedPos<64>";
+  let DiagnosticType = "CVTFixedPos64";
+}
+
+// We need the cartesian product of f32/f64 i32/i64 operands for
+// conversions:
+//   + Selection needs to use operands of correct floating type
+//   + Assembly parsing and decoding depend on integer width
+class cvtfix_i32_op<ValueType FloatVT>
+  : Operand<FloatVT>,
+    ComplexPattern<FloatVT, 1, "SelectCVTFixedPosOperand<32>", [fpimm]> {
+  let ParserMatchClass = fixedpos_asmoperand_i32;
+  let DecoderMethod = "DecodeCVT32FixedPosOperand";
+  let PrintMethod = "printCVTFixedPosOperand";
+}
+
+class cvtfix_i64_op<ValueType FloatVT>
+  : Operand<FloatVT>,
+    ComplexPattern<FloatVT, 1, "SelectCVTFixedPosOperand<64>", [fpimm]> {
+  let ParserMatchClass = fixedpos_asmoperand_i64;
+  let PrintMethod = "printCVTFixedPosOperand";
+}
+
+// Because of the proliferation of weird operands, it's not really
+// worth going for a multiclass here. Oh well.
+
+class A64I_fptofix<bit sf, bits<2> type, bits<3> opcode,
+                   RegisterClass GPR, RegisterClass FPR, 
+                   ValueType DstTy, ValueType SrcTy, 
+                   Operand scale_op, string asmop, SDNode cvtop>
+  : A64I_fpfixed<sf, 0b0, type, 0b11, opcode,
+                 (outs GPR:$Rd), (ins FPR:$Rn, scale_op:$Scale),
+                 !strconcat(asmop, "\t$Rd, $Rn, $Scale"),
+                 [(set DstTy:$Rd, (cvtop (fmul SrcTy:$Rn, scale_op:$Scale)))],
+                 NoItinerary>;
+
+def FCVTZSwsi : A64I_fptofix<0b0, 0b00, 0b000, GPR32, FPR32, i32, f32,
+                             cvtfix_i32_op<f32>, "fcvtzs", fp_to_sint>;
+def FCVTZSxsi : A64I_fptofix<0b1, 0b00, 0b000, GPR64, FPR32, i64, f32,
+                             cvtfix_i64_op<f32>, "fcvtzs", fp_to_sint>;
+def FCVTZUwsi : A64I_fptofix<0b0, 0b00, 0b001, GPR32, FPR32, i32, f32,
+                             cvtfix_i32_op<f32>, "fcvtzu", fp_to_uint>;
+def FCVTZUxsi : A64I_fptofix<0b1, 0b00, 0b001, GPR64, FPR32, i64, f32,
+                             cvtfix_i64_op<f32>, "fcvtzu", fp_to_uint>;
+
+def FCVTZSwdi : A64I_fptofix<0b0, 0b01, 0b000, GPR32, FPR64, i32, f64,
+                             cvtfix_i32_op<f64>, "fcvtzs", fp_to_sint>;
+def FCVTZSxdi : A64I_fptofix<0b1, 0b01, 0b000, GPR64, FPR64, i64, f64,
+                             cvtfix_i64_op<f64>, "fcvtzs", fp_to_sint>;
+def FCVTZUwdi : A64I_fptofix<0b0, 0b01, 0b001, GPR32, FPR64, i32, f64,
+                             cvtfix_i32_op<f64>, "fcvtzu", fp_to_uint>;
+def FCVTZUxdi : A64I_fptofix<0b1, 0b01, 0b001, GPR64, FPR64, i64, f64,
+                             cvtfix_i64_op<f64>, "fcvtzu", fp_to_uint>;
+
+
+class A64I_fixtofp<bit sf, bits<2> type, bits<3> opcode,
+                   RegisterClass FPR, RegisterClass GPR,
+                   ValueType DstTy, ValueType SrcTy,
+                   Operand scale_op, string asmop, SDNode cvtop>
+  : A64I_fpfixed<sf, 0b0, type, 0b00, opcode,
+                 (outs FPR:$Rd), (ins GPR:$Rn, scale_op:$Scale),
+                 !strconcat(asmop, "\t$Rd, $Rn, $Scale"),
+                 [(set DstTy:$Rd, (fdiv (cvtop SrcTy:$Rn), scale_op:$Scale))],
+                 NoItinerary>;
+
+def SCVTFswi : A64I_fixtofp<0b0, 0b00, 0b010, FPR32, GPR32, f32, i32,
+                            cvtfix_i32_op<f32>, "scvtf", sint_to_fp>;
+def SCVTFsxi : A64I_fixtofp<0b1, 0b00, 0b010, FPR32, GPR64, f32, i64,
+                            cvtfix_i64_op<f32>, "scvtf", sint_to_fp>;
+def UCVTFswi : A64I_fixtofp<0b0, 0b00, 0b011, FPR32, GPR32, f32, i32,
+                            cvtfix_i32_op<f32>, "ucvtf", uint_to_fp>;
+def UCVTFsxi : A64I_fixtofp<0b1, 0b00, 0b011, FPR32, GPR64, f32, i64,
+                            cvtfix_i64_op<f32>, "ucvtf", uint_to_fp>;
+def SCVTFdwi : A64I_fixtofp<0b0, 0b01, 0b010, FPR64, GPR32, f64, i32,
+                            cvtfix_i32_op<f64>, "scvtf", sint_to_fp>;
+def SCVTFdxi : A64I_fixtofp<0b1, 0b01, 0b010, FPR64, GPR64, f64, i64,
+                            cvtfix_i64_op<f64>, "scvtf", sint_to_fp>;
+def UCVTFdwi : A64I_fixtofp<0b0, 0b01, 0b011, FPR64, GPR32, f64, i32,
+                            cvtfix_i32_op<f64>, "ucvtf", uint_to_fp>;
+def UCVTFdxi : A64I_fixtofp<0b1, 0b01, 0b011, FPR64, GPR64, f64, i64,
+                            cvtfix_i64_op<f64>, "ucvtf", uint_to_fp>;
+
+//===----------------------------------------------------------------------===//
+// Floating-point <-> integer conversion instructions
+//===----------------------------------------------------------------------===//
+// Contains: FCVTZS, FCVTZU, SCVTF, UCVTF
+
+class A64I_fpintI<bit sf, bits<2> type, bits<2> rmode, bits<3> opcode,
+                   RegisterClass DestPR, RegisterClass SrcPR, string asmop>
+  : A64I_fpint<sf, 0b0, type, rmode, opcode, (outs DestPR:$Rd), (ins SrcPR:$Rn),
+               !strconcat(asmop, "\t$Rd, $Rn"), [], NoItinerary>;
+
+multiclass A64I_fptointRM<bits<2> rmode, bit o2, string asmop> {
+  def Sws : A64I_fpintI<0b0, 0b00, rmode, {o2, 0, 0},
+                        GPR32, FPR32, asmop # "s">;
+  def Sxs : A64I_fpintI<0b1, 0b00, rmode, {o2, 0, 0},
+                        GPR64, FPR32, asmop # "s">;
+  def Uws : A64I_fpintI<0b0, 0b00, rmode, {o2, 0, 1},
+                        GPR32, FPR32, asmop # "u">;
+  def Uxs : A64I_fpintI<0b1, 0b00, rmode, {o2, 0, 1},
+                        GPR64, FPR32, asmop # "u">;
+
+  def Swd : A64I_fpintI<0b0, 0b01, rmode, {o2, 0, 0},
+                        GPR32, FPR64, asmop # "s">;
+  def Sxd : A64I_fpintI<0b1, 0b01, rmode, {o2, 0, 0},
+                        GPR64, FPR64, asmop # "s">;
+  def Uwd : A64I_fpintI<0b0, 0b01, rmode, {o2, 0, 1},
+                        GPR32, FPR64, asmop # "u">;
+  def Uxd : A64I_fpintI<0b1, 0b01, rmode, {o2, 0, 1},
+                        GPR64, FPR64, asmop # "u">;
+}
+
+defm FCVTN : A64I_fptointRM<0b00, 0b0, "fcvtn">;
+defm FCVTP : A64I_fptointRM<0b01, 0b0, "fcvtp">;
+defm FCVTM : A64I_fptointRM<0b10, 0b0, "fcvtm">;
+defm FCVTZ : A64I_fptointRM<0b11, 0b0, "fcvtz">;
+defm FCVTA : A64I_fptointRM<0b00, 0b1, "fcvta">;
+
+def : Pat<(i32 (fp_to_sint f32:$Rn)), (FCVTZSws $Rn)>;
+def : Pat<(i64 (fp_to_sint f32:$Rn)), (FCVTZSxs $Rn)>;
+def : Pat<(i32 (fp_to_uint f32:$Rn)), (FCVTZUws $Rn)>;
+def : Pat<(i64 (fp_to_uint f32:$Rn)), (FCVTZUxs $Rn)>;
+def : Pat<(i32 (fp_to_sint f64:$Rn)), (FCVTZSwd $Rn)>;
+def : Pat<(i64 (fp_to_sint f64:$Rn)), (FCVTZSxd $Rn)>;
+def : Pat<(i32 (fp_to_uint f64:$Rn)), (FCVTZUwd $Rn)>;
+def : Pat<(i64 (fp_to_uint f64:$Rn)), (FCVTZUxd $Rn)>;
+
+multiclass A64I_inttofp<bit o0, string asmop> {
+  def CVTFsw : A64I_fpintI<0b0, 0b00, 0b00, {0, 1, o0}, FPR32, GPR32, asmop>;
+  def CVTFsx : A64I_fpintI<0b1, 0b00, 0b00, {0, 1, o0}, FPR32, GPR64, asmop>;
+  def CVTFdw : A64I_fpintI<0b0, 0b01, 0b00, {0, 1, o0}, FPR64, GPR32, asmop>;
+  def CVTFdx : A64I_fpintI<0b1, 0b01, 0b00, {0, 1, o0}, FPR64, GPR64, asmop>;
+}
+
+defm S : A64I_inttofp<0b0, "scvtf">;
+defm U : A64I_inttofp<0b1, "ucvtf">;
+
+def : Pat<(f32 (sint_to_fp i32:$Rn)), (SCVTFsw $Rn)>;
+def : Pat<(f32 (sint_to_fp i64:$Rn)), (SCVTFsx $Rn)>;
+def : Pat<(f64 (sint_to_fp i32:$Rn)), (SCVTFdw $Rn)>;
+def : Pat<(f64 (sint_to_fp i64:$Rn)), (SCVTFdx $Rn)>;
+def : Pat<(f32 (uint_to_fp i32:$Rn)), (UCVTFsw $Rn)>;
+def : Pat<(f32 (uint_to_fp i64:$Rn)), (UCVTFsx $Rn)>;
+def : Pat<(f64 (uint_to_fp i32:$Rn)), (UCVTFdw $Rn)>;
+def : Pat<(f64 (uint_to_fp i64:$Rn)), (UCVTFdx $Rn)>;
+
+def FMOVws : A64I_fpintI<0b0, 0b00, 0b00, 0b110, GPR32, FPR32, "fmov">;
+def FMOVsw : A64I_fpintI<0b0, 0b00, 0b00, 0b111, FPR32, GPR32, "fmov">;
+def FMOVxd : A64I_fpintI<0b1, 0b01, 0b00, 0b110, GPR64, FPR64, "fmov">;
+def FMOVdx : A64I_fpintI<0b1, 0b01, 0b00, 0b111, FPR64, GPR64, "fmov">;
+
+def : Pat<(i32 (bitconvert f32:$Rn)), (FMOVws $Rn)>;
+def : Pat<(f32 (bitconvert i32:$Rn)), (FMOVsw $Rn)>;
+def : Pat<(i64 (bitconvert f64:$Rn)), (FMOVxd $Rn)>;
+def : Pat<(f64 (bitconvert i64:$Rn)), (FMOVdx $Rn)>;
+
+def lane1_asmoperand : AsmOperandClass {
+  let Name = "Lane1";
+  let RenderMethod = "addImmOperands";
+  let DiagnosticType = "Lane1";
+}
+
+def lane1 : Operand<i32> {
+  let ParserMatchClass = lane1_asmoperand;
+  let PrintMethod = "printBareImmOperand";
+}
+
+let DecoderMethod =  "DecodeFMOVLaneInstruction" in {
+  def FMOVxv : A64I_fpint<0b1, 0b0, 0b10, 0b01, 0b110,
+                          (outs GPR64:$Rd), (ins VPR128:$Rn, lane1:$Lane),
+                          "fmov\t$Rd, $Rn.d[$Lane]", [], NoItinerary>;
+
+  def FMOVvx : A64I_fpint<0b1, 0b0, 0b10, 0b01, 0b111,
+                          (outs VPR128:$Rd), (ins GPR64:$Rn, lane1:$Lane),
+                          "fmov\t$Rd.d[$Lane], $Rn", [], NoItinerary>;
+}
+
+def : InstAlias<"fmov $Rd, $Rn.2d[$Lane]",
+                (FMOVxv GPR64:$Rd, VPR128:$Rn, lane1:$Lane), 0b0>;
+
+def : InstAlias<"fmov $Rd.2d[$Lane], $Rn",
+                (FMOVvx VPR128:$Rd, GPR64:$Rn, lane1:$Lane), 0b0>;
+
+//===----------------------------------------------------------------------===//
+// Floating-point immediate instructions
+//===----------------------------------------------------------------------===//
+// Contains: FMOV
+
+def fpimm_asmoperand : AsmOperandClass {
+  let Name = "FMOVImm";
+  let ParserMethod = "ParseFPImmOperand";
+  let DiagnosticType = "FPImm";
+}
+
+// The MCOperand for these instructions are the encoded 8-bit values.
+def SDXF_fpimm : SDNodeXForm<fpimm, [{
+  uint32_t Imm8;
+  A64Imms::isFPImm(N->getValueAPF(), Imm8);
+  return CurDAG->getTargetConstant(Imm8, MVT::i32);
+}]>;
+
+class fmov_operand<ValueType FT>
+  : Operand<i32>,
+    PatLeaf<(FT fpimm), [{ return A64Imms::isFPImm(N->getValueAPF()); }],
+            SDXF_fpimm> {
+  let PrintMethod = "printFPImmOperand";
+  let ParserMatchClass = fpimm_asmoperand;
+}
+
+def fmov32_operand : fmov_operand<f32>;
+def fmov64_operand : fmov_operand<f64>;
+
+class A64I_fpimm_impl<bits<2> type, RegisterClass Reg, ValueType VT,
+                      Operand fmov_operand>
+  : A64I_fpimm<0b0, 0b0, type, 0b00000,
+               (outs Reg:$Rd),
+               (ins fmov_operand:$Imm8),
+               "fmov\t$Rd, $Imm8",
+               [(set VT:$Rd, fmov_operand:$Imm8)],
+               NoItinerary>;
+
+def FMOVsi : A64I_fpimm_impl<0b00, FPR32, f32, fmov32_operand>;
+def FMOVdi : A64I_fpimm_impl<0b01, FPR64, f64, fmov64_operand>;
+
+//===----------------------------------------------------------------------===//
+// Load-register (literal) instructions
+//===----------------------------------------------------------------------===//
+// Contains: LDR, LDRSW, PRFM
+
+def ldrlit_label_asmoperand : AsmOperandClass {
+  let Name = "LoadLitLabel";
+  let RenderMethod = "addLabelOperands<19, 4>";
+  let DiagnosticType = "Label";
+}
+
+def ldrlit_label : Operand<i64> {
+  let EncoderMethod = "getLoadLitLabelOpValue";
+
+  // This label is a 19-bit offset from PC, scaled by the instruction-width: 4.
+  let PrintMethod = "printLabelOperand<19, 4>";
+  let ParserMatchClass = ldrlit_label_asmoperand;
+  let OperandType = "OPERAND_PCREL";
+}
+
+// Various instructions take an immediate value (which can always be used),
+// where some numbers have a symbolic name to make things easier. These operands
+// and the associated functions abstract away the differences.
+multiclass namedimm<string prefix, string mapper> {
+  def _asmoperand : AsmOperandClass {
+    let Name = "NamedImm" # prefix;
+    let PredicateMethod = "isUImm";
+    let RenderMethod = "addImmOperands";
+    let ParserMethod = "ParseNamedImmOperand<" # mapper # ">";
+    let DiagnosticType = "NamedImm_" # prefix;
+  }
+
+  def _op : Operand<i32> {
+    let ParserMatchClass = !cast<AsmOperandClass>(prefix # "_asmoperand");
+    let PrintMethod = "printNamedImmOperand<" # mapper # ">";
+    let DecoderMethod = "DecodeNamedImmOperand<" # mapper # ">";
+  }
+}
+
+defm prefetch : namedimm<"prefetch", "A64PRFM::PRFMMapper">;
+
+class A64I_LDRlitSimple<bits<2> opc, bit v, RegisterClass OutReg,
+                      list<dag> patterns = []>
+   : A64I_LDRlit<opc, v, (outs OutReg:$Rt), (ins ldrlit_label:$Imm19),
+                 "ldr\t$Rt, $Imm19", patterns, NoItinerary>;
+
+let mayLoad = 1 in {
+  def LDRw_lit : A64I_LDRlitSimple<0b00, 0b0, GPR32>;
+  def LDRx_lit : A64I_LDRlitSimple<0b01, 0b0, GPR64>;
+}
+
+def LDRs_lit  : A64I_LDRlitSimple<0b00, 0b1, FPR32>;
+def LDRd_lit  : A64I_LDRlitSimple<0b01, 0b1, FPR64>;
+
+let mayLoad = 1 in {
+  def LDRq_lit : A64I_LDRlitSimple<0b10, 0b1, FPR128>;
+
+
+  def LDRSWx_lit : A64I_LDRlit<0b10, 0b0,
+                               (outs GPR64:$Rt),
+                               (ins ldrlit_label:$Imm19),
+                               "ldrsw\t$Rt, $Imm19",
+                               [], NoItinerary>;
+
+  def PRFM_lit : A64I_LDRlit<0b11, 0b0,
+                             (outs), (ins prefetch_op:$Rt, ldrlit_label:$Imm19),
+                             "prfm\t$Rt, $Imm19",
+                             [], NoItinerary>;
+}
+
+//===----------------------------------------------------------------------===//
+// Load-store exclusive instructions
+//===----------------------------------------------------------------------===//
+// Contains: STXRB, STXRH, STXR, LDXRB, LDXRH, LDXR. STXP, LDXP, STLXRB,
+//           STLXRH, STLXR, LDAXRB, LDAXRH, LDAXR, STLXP, LDAXP, STLRB,
+//           STLRH, STLR, LDARB, LDARH, LDAR
+
+// Since these instructions have the undefined register bits set to 1 in
+// their canonical form, we need a post encoder method to set those bits
+// to 1 when encoding these instructions. We do this using the
+// fixLoadStoreExclusive function. This function has template parameters:
+//
+// fixLoadStoreExclusive<int hasRs, int hasRt2>
+//
+// hasRs indicates that the instruction uses the Rs field, so we won't set
+// it to 1 (and the same for Rt2). We don't need template parameters for
+// the other register fiels since Rt and Rn are always used.
+
+// This operand parses a GPR64xsp register, followed by an optional immediate
+// #0.
+def GPR64xsp0_asmoperand : AsmOperandClass {
+  let Name = "GPR64xsp0";
+  let PredicateMethod = "isWrappedReg";
+  let RenderMethod = "addRegOperands";
+  let ParserMethod = "ParseLSXAddressOperand";
+  // Diagnostics are provided by ParserMethod
+}
+
+def GPR64xsp0 : RegisterOperand<GPR64xsp> {
+  let ParserMatchClass = GPR64xsp0_asmoperand;
+}
+
+//===----------------------------------
+// Store-exclusive (releasing & normal)
+//===----------------------------------
+
+class A64I_SRexs_impl<bits<2> size, bits<3> opcode, string asm, dag outs,
+                        dag ins, list<dag> pat,
+                        InstrItinClass itin> :
+       A64I_LDSTex_stn <size,
+                        opcode{2}, 0, opcode{1}, opcode{0},
+                        outs, ins,
+                        !strconcat(asm, "\t$Rs, $Rt, [$Rn]"),
+                        pat, itin> {
+  let mayStore = 1;
+  let PostEncoderMethod = "fixLoadStoreExclusive<1,0>";
+}
+
+multiclass A64I_SRex<string asmstr, bits<3> opcode, string prefix> {
+  def _byte:  A64I_SRexs_impl<0b00, opcode, !strconcat(asmstr, "b"),
+                              (outs GPR32:$Rs), (ins GPR32:$Rt, GPR64xsp0:$Rn),
+                              [], NoItinerary>;
+
+  def _hword:  A64I_SRexs_impl<0b01, opcode, !strconcat(asmstr, "h"),
+                               (outs GPR32:$Rs), (ins GPR32:$Rt, GPR64xsp0:$Rn),
+                               [],NoItinerary>;
+
+  def _word:  A64I_SRexs_impl<0b10, opcode, asmstr,
+                              (outs GPR32:$Rs), (ins GPR32:$Rt, GPR64xsp0:$Rn),
+                              [], NoItinerary>;
+
+  def _dword: A64I_SRexs_impl<0b11, opcode, asmstr,
+                              (outs GPR32:$Rs), (ins GPR64:$Rt, GPR64xsp0:$Rn),
+                              [], NoItinerary>;
+}
+
+defm STXR  : A64I_SRex<"stxr",  0b000, "STXR">;
+defm STLXR : A64I_SRex<"stlxr", 0b001, "STLXR">;
+
+//===----------------------------------
+// Loads
+//===----------------------------------
+
+class A64I_LRexs_impl<bits<2> size, bits<3> opcode, string asm, dag outs,
+                        dag ins, list<dag> pat,
+                        InstrItinClass itin> :
+        A64I_LDSTex_tn <size,
+                        opcode{2}, 1, opcode{1}, opcode{0},
+                        outs, ins,
+                        !strconcat(asm, "\t$Rt, [$Rn]"),
+                        pat, itin> {
+  let mayLoad = 1;
+  let PostEncoderMethod = "fixLoadStoreExclusive<0,0>";
+}
+
+multiclass A64I_LRex<string asmstr, bits<3> opcode> {
+  def _byte:  A64I_LRexs_impl<0b00, opcode, !strconcat(asmstr, "b"),
+                            (outs GPR32:$Rt), (ins GPR64xsp0:$Rn),
+                            [], NoItinerary>;
+
+  def _hword:  A64I_LRexs_impl<0b01, opcode, !strconcat(asmstr, "h"),
+                            (outs GPR32:$Rt), (ins GPR64xsp0:$Rn),
+                            [], NoItinerary>;
+
+  def _word:  A64I_LRexs_impl<0b10, opcode, asmstr,
+                            (outs GPR32:$Rt), (ins GPR64xsp0:$Rn),
+                            [], NoItinerary>;
+
+  def _dword: A64I_LRexs_impl<0b11, opcode, asmstr,
+                            (outs GPR64:$Rt), (ins GPR64xsp0:$Rn),
+                            [], NoItinerary>;
+}
+
+defm LDXR  : A64I_LRex<"ldxr",  0b000>;
+defm LDAXR : A64I_LRex<"ldaxr", 0b001>;
+defm LDAR  : A64I_LRex<"ldar",  0b101>;
+
+class acquiring_load<PatFrag base>
+  : PatFrag<(ops node:$ptr), (base node:$ptr), [{
+  return cast<AtomicSDNode>(N)->getOrdering() == Acquire;
+}]>;
+
+def atomic_load_acquire_8  : acquiring_load<atomic_load_8>;
+def atomic_load_acquire_16 : acquiring_load<atomic_load_16>;
+def atomic_load_acquire_32 : acquiring_load<atomic_load_32>;
+def atomic_load_acquire_64 : acquiring_load<atomic_load_64>;
+
+def : Pat<(atomic_load_acquire_8  i64:$Rn), (LDAR_byte  $Rn)>;
+def : Pat<(atomic_load_acquire_16 i64:$Rn), (LDAR_hword $Rn)>;
+def : Pat<(atomic_load_acquire_32 i64:$Rn), (LDAR_word  $Rn)>;
+def : Pat<(atomic_load_acquire_64 i64:$Rn), (LDAR_dword $Rn)>;
+
+//===----------------------------------
+// Store-release (no exclusivity)
+//===----------------------------------
+
+class A64I_SLexs_impl<bits<2> size, bits<3> opcode, string asm, dag outs,
+                        dag ins, list<dag> pat,
+                        InstrItinClass itin> :
+        A64I_LDSTex_tn <size,
+                        opcode{2}, 0, opcode{1}, opcode{0},
+                        outs, ins,
+                        !strconcat(asm, "\t$Rt, [$Rn]"),
+                        pat, itin> {
+  let mayStore = 1;
+  let PostEncoderMethod = "fixLoadStoreExclusive<0,0>";
+}
+
+class releasing_store<PatFrag base>
+  : PatFrag<(ops node:$ptr, node:$val), (base node:$ptr, node:$val), [{
+  return cast<AtomicSDNode>(N)->getOrdering() == Release;
+}]>;
+
+def atomic_store_release_8  : releasing_store<atomic_store_8>;
+def atomic_store_release_16 : releasing_store<atomic_store_16>;
+def atomic_store_release_32 : releasing_store<atomic_store_32>;
+def atomic_store_release_64 : releasing_store<atomic_store_64>;
+
+multiclass A64I_SLex<string asmstr, bits<3> opcode, string prefix> {
+  def _byte:  A64I_SLexs_impl<0b00, opcode, !strconcat(asmstr, "b"),
+                            (outs), (ins GPR32:$Rt, GPR64xsp0:$Rn),
+                            [(atomic_store_release_8 i64:$Rn, i32:$Rt)],
+                            NoItinerary>;
+
+  def _hword:  A64I_SLexs_impl<0b01, opcode, !strconcat(asmstr, "h"),
+                           (outs), (ins GPR32:$Rt, GPR64xsp0:$Rn),
+                           [(atomic_store_release_16 i64:$Rn, i32:$Rt)],
+                           NoItinerary>;
+
+  def _word:  A64I_SLexs_impl<0b10, opcode, asmstr,
+                           (outs), (ins GPR32:$Rt, GPR64xsp0:$Rn),
+                           [(atomic_store_release_32 i64:$Rn, i32:$Rt)],
+                           NoItinerary>;
+
+  def _dword: A64I_SLexs_impl<0b11, opcode, asmstr,
+                           (outs), (ins GPR64:$Rt, GPR64xsp0:$Rn),
+                           [(atomic_store_release_64 i64:$Rn, i64:$Rt)],
+                           NoItinerary>;
+}
+
+defm STLR  : A64I_SLex<"stlr", 0b101, "STLR">;
+
+//===----------------------------------
+// Store-exclusive pair (releasing & normal)
+//===----------------------------------
+
+class A64I_SPexs_impl<bits<2> size, bits<3> opcode, string asm, dag outs,
+                        dag ins, list<dag> pat,
+                        InstrItinClass itin> :
+     A64I_LDSTex_stt2n <size,
+                        opcode{2}, 0, opcode{1}, opcode{0},
+                        outs, ins,
+                        !strconcat(asm, "\t$Rs, $Rt, $Rt2, [$Rn]"),
+                        pat, itin> {
+  let mayStore = 1;
+}
+
+
+multiclass A64I_SPex<string asmstr, bits<3> opcode> {
+  def _word:  A64I_SPexs_impl<0b10, opcode, asmstr, (outs),
+                            (ins GPR32:$Rs, GPR32:$Rt, GPR32:$Rt2,
+                                 GPR64xsp0:$Rn),
+                            [], NoItinerary>;
+
+  def _dword: A64I_SPexs_impl<0b11, opcode, asmstr, (outs),
+                            (ins GPR32:$Rs, GPR64:$Rt, GPR64:$Rt2,
+                                            GPR64xsp0:$Rn),
+                            [], NoItinerary>;
+}
+
+defm STXP  : A64I_SPex<"stxp", 0b010>;
+defm STLXP : A64I_SPex<"stlxp", 0b011>;
+
+//===----------------------------------
+// Load-exclusive pair (acquiring & normal)
+//===----------------------------------
+
+class A64I_LPexs_impl<bits<2> size, bits<3> opcode, string asm, dag outs,
+                        dag ins, list<dag> pat,
+                        InstrItinClass itin> :
+      A64I_LDSTex_tt2n <size,
+                        opcode{2}, 1, opcode{1}, opcode{0},
+                        outs, ins,
+                        !strconcat(asm, "\t$Rt, $Rt2, [$Rn]"),
+                        pat, itin>{
+  let mayLoad = 1;
+  let DecoderMethod = "DecodeLoadPairExclusiveInstruction";
+  let PostEncoderMethod = "fixLoadStoreExclusive<0,1>";
+}
+
+multiclass A64I_LPex<string asmstr, bits<3> opcode> {
+  def _word:  A64I_LPexs_impl<0b10, opcode, asmstr,
+                            (outs GPR32:$Rt, GPR32:$Rt2),
+                            (ins GPR64xsp0:$Rn),
+                            [], NoItinerary>;
+
+  def _dword: A64I_LPexs_impl<0b11, opcode, asmstr,
+                            (outs GPR64:$Rt, GPR64:$Rt2),
+                            (ins GPR64xsp0:$Rn),
+                            [], NoItinerary>;
+}
+
+defm LDXP  : A64I_LPex<"ldxp", 0b010>;
+defm LDAXP : A64I_LPex<"ldaxp", 0b011>;
+
+//===----------------------------------------------------------------------===//
+// Load-store register (unscaled immediate) instructions
+//===----------------------------------------------------------------------===//
+// Contains: LDURB, LDURH, LDRUSB, LDRUSH, LDRUSW, STUR, STURB, STURH and PRFUM
+//
+// and
+//
+//===----------------------------------------------------------------------===//
+// Load-store register (register offset) instructions
+//===----------------------------------------------------------------------===//
+// Contains: LDRB, LDRH, LDRSB, LDRSH, LDRSW, STR, STRB, STRH and PRFM
+//
+// and
+//
+//===----------------------------------------------------------------------===//
+// Load-store register (unsigned immediate) instructions
+//===----------------------------------------------------------------------===//
+// Contains: LDRB, LDRH, LDRSB, LDRSH, LDRSW, STR, STRB, STRH and PRFM
+//
+// and
+//
+//===----------------------------------------------------------------------===//
+// Load-store register (immediate post-indexed) instructions
+//===----------------------------------------------------------------------===//
+// Contains: STRB, STRH, STR, LDRB, LDRH, LDR, LDRSB, LDRSH, LDRSW
+//
+// and
+//
+//===----------------------------------------------------------------------===//
+// Load-store register (immediate pre-indexed) instructions
+//===----------------------------------------------------------------------===//
+// Contains: STRB, STRH, STR, LDRB, LDRH, LDR, LDRSB, LDRSH, LDRSW
+
+// Note that patterns are much later on in a completely separate section (they
+// need ADRPxi to be defined).
+
+//===-------------------------------
+// 1. Various operands needed
+//===-------------------------------
+
+//===-------------------------------
+// 1.1 Unsigned 12-bit immediate operands
+//===-------------------------------
+// The addressing mode for these instructions consists of an unsigned 12-bit
+// immediate which is scaled by the size of the memory access.
+//
+// We represent this in the MC layer by two operands:
+//     1. A base register.
+//     2. A 12-bit immediate: not multiplied by access size, so "LDR x0,[x0,#8]"
+//        would have '1' in this field.
+// This means that separate functions are needed for converting representations
+// which *are* aware of the intended access size.
+
+// Anything that creates an MCInst (Decoding, selection and AsmParsing) has to
+// know the access size via some means. An isolated operand does not have this
+// information unless told from here, which means we need separate tablegen
+// Operands for each access size. This multiclass takes care of instantiating
+// the correct template functions in the rest of the backend.
+
+//===-------------------------------
+// 1.1 Unsigned 12-bit immediate operands
+//===-------------------------------
+
+multiclass offsets_uimm12<int MemSize, string prefix> {
+  def uimm12_asmoperand : AsmOperandClass {
+    let Name = "OffsetUImm12_" # MemSize;
+    let PredicateMethod = "isOffsetUImm12<" # MemSize # ">";
+    let RenderMethod = "addOffsetUImm12Operands<" # MemSize # ">";
+    let DiagnosticType = "LoadStoreUImm12_" # MemSize;
+  }
+
+  // Pattern is really no more than an ImmLeaf, but predicated on MemSize which
+  // complicates things beyond TableGen's ken.
+  def uimm12 : Operand<i64>,
+               ComplexPattern<i64, 1, "SelectOffsetUImm12<" # MemSize # ">"> {
+    let ParserMatchClass
+      = !cast<AsmOperandClass>(prefix # uimm12_asmoperand);
+
+    let PrintMethod = "printOffsetUImm12Operand<" # MemSize # ">";
+    let EncoderMethod = "getOffsetUImm12OpValue<" # MemSize # ">";
+  }
+}
+
+defm byte_  : offsets_uimm12<1, "byte_">;
+defm hword_ : offsets_uimm12<2, "hword_">;
+defm word_  : offsets_uimm12<4, "word_">;
+defm dword_ : offsets_uimm12<8, "dword_">;
+defm qword_ : offsets_uimm12<16, "qword_">;
+
+//===-------------------------------
+// 1.1 Signed 9-bit immediate operands
+//===-------------------------------
+
+// The MCInst is expected to store the bit-wise encoding of the value,
+// which amounts to lopping off the extended sign bits.
+def SDXF_simm9 : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(N->getZExtValue() & 0x1ff, MVT::i32);
+}]>;
+
+def simm9_asmoperand : AsmOperandClass {
+  let Name = "SImm9";
+  let PredicateMethod = "isSImm<9>";
+  let RenderMethod = "addSImmOperands<9>";
+  let DiagnosticType = "LoadStoreSImm9";
+}
+
+def simm9 : Operand<i64>,
+            ImmLeaf<i64, [{ return Imm >= -0x100 && Imm <= 0xff; }],
+            SDXF_simm9> {
+  let PrintMethod = "printOffsetSImm9Operand";
+  let ParserMatchClass = simm9_asmoperand;
+}
+
+
+//===-------------------------------
+// 1.3 Register offset extensions
+//===-------------------------------
+
+// The assembly-syntax for these addressing-modes is:
+//    [<Xn|SP>, <R><m> {, <extend> {<amount>}}]
+//
+// The essential semantics are:
+//     + <amount> is a shift: #<log(transfer size)> or #0
+//     + <R> can be W or X.
+//     + If <R> is W, <extend> can be UXTW or SXTW
+//     + If <R> is X, <extend> can be LSL or SXTX
+//
+// The trickiest of those constraints is that Rm can be either GPR32 or GPR64,
+// which will need separate instructions for LLVM type-consistency. We'll also
+// need separate operands, of course.
+multiclass regexts<int MemSize, int RmSize, RegisterClass GPR,
+                   string Rm, string prefix> {
+  def regext_asmoperand : AsmOperandClass {
+    let Name = "AddrRegExtend_" # MemSize # "_" #  Rm;
+    let PredicateMethod = "isAddrRegExtend<" # MemSize # "," # RmSize # ">";
+    let RenderMethod = "addAddrRegExtendOperands<" # MemSize # ">";
+    let DiagnosticType = "LoadStoreExtend" # RmSize # "_" # MemSize;
+  }
+
+  def regext : Operand<i64> {
+    let PrintMethod
+      = "printAddrRegExtendOperand<" # MemSize # ", " # RmSize # ">";
+
+    let DecoderMethod = "DecodeAddrRegExtendOperand";
+    let ParserMatchClass
+      = !cast<AsmOperandClass>(prefix # regext_asmoperand);
+  }
+}
+
+multiclass regexts_wx<int MemSize, string prefix> {
+  // Rm is an X-register if LSL or SXTX are specified as the shift.
+  defm Xm_ : regexts<MemSize, 64, GPR64, "Xm", prefix # "Xm_">;
+
+  // Rm is a W-register if UXTW or SXTW are specified as the shift.
+  defm Wm_ : regexts<MemSize, 32, GPR32, "Wm", prefix # "Wm_">;
+}
+
+defm byte_  : regexts_wx<1, "byte_">;
+defm hword_ : regexts_wx<2, "hword_">;
+defm word_  : regexts_wx<4, "word_">;
+defm dword_ : regexts_wx<8, "dword_">;
+defm qword_ : regexts_wx<16, "qword_">;
+
+
+//===------------------------------
+// 2. The instructions themselves.
+//===------------------------------
+
+// We have the following instructions to implement:
+// |                 | B     | H     | W     | X      |
+// |-----------------+-------+-------+-------+--------|
+// | unsigned str    | STRB  | STRH  | STR   | STR    |
+// | unsigned ldr    | LDRB  | LDRH  | LDR   | LDR    |
+// | signed ldr to W | LDRSB | LDRSH | -     | -      |
+// | signed ldr to X | LDRSB | LDRSH | LDRSW | (PRFM) |
+
+// This will instantiate the LDR/STR instructions you'd expect to use for an
+// unsigned datatype (first two rows above) or floating-point register, which is
+// reasonably uniform across all access sizes.
+
+
+//===------------------------------
+// 2.1 Regular instructions
+//===------------------------------
+
+// This class covers the basic unsigned or irrelevantly-signed loads and stores,
+// to general-purpose and floating-point registers.
+
+class AddrParams<string prefix> {
+  Operand uimm12 = !cast<Operand>(prefix # "_uimm12");
+
+  Operand regextWm = !cast<Operand>(prefix # "_Wm_regext");
+  Operand regextXm = !cast<Operand>(prefix # "_Xm_regext");
+}
+
+def byte_addrparams : AddrParams<"byte">;
+def hword_addrparams : AddrParams<"hword">;
+def word_addrparams : AddrParams<"word">;
+def dword_addrparams : AddrParams<"dword">;
+def qword_addrparams : AddrParams<"qword">;
+
+multiclass A64I_LDRSTR_unsigned<string prefix, bits<2> size, bit v,
+                                bit high_opc, string asmsuffix,
+                                RegisterClass GPR, AddrParams params> {
+  // Unsigned immediate
+  def _STR : A64I_LSunsigimm<size, v, {high_opc, 0b0},
+                     (outs), (ins GPR:$Rt, GPR64xsp:$Rn, params.uimm12:$UImm12),
+                     "str" # asmsuffix # "\t$Rt, [$Rn, $UImm12]",
+                     [], NoItinerary> {
+    let mayStore = 1;
+  }
+  def : InstAlias<"str" # asmsuffix # " $Rt, [$Rn]",
+                (!cast<Instruction>(prefix # "_STR") GPR:$Rt, GPR64xsp:$Rn, 0)>;
+
+  def _LDR : A64I_LSunsigimm<size, v, {high_opc, 0b1},
+                      (outs GPR:$Rt), (ins GPR64xsp:$Rn, params.uimm12:$UImm12),
+                      "ldr" #  asmsuffix # "\t$Rt, [$Rn, $UImm12]",
+                      [], NoItinerary> {
+    let mayLoad = 1;
+  }
+  def : InstAlias<"ldr" # asmsuffix # " $Rt, [$Rn]",
+                (!cast<Instruction>(prefix # "_LDR") GPR:$Rt, GPR64xsp:$Rn, 0)>;
+
+  // Register offset (four of these: load/store and Wm/Xm).
+  let mayLoad = 1 in {
+    def _Wm_RegOffset_LDR : A64I_LSregoff<size, v, {high_opc, 0b1}, 0b0,
+                            (outs GPR:$Rt),
+                            (ins GPR64xsp:$Rn, GPR32:$Rm, params.regextWm:$Ext),
+                            "ldr" # asmsuffix # "\t$Rt, [$Rn, $Rm, $Ext]",
+                            [], NoItinerary>;
+
+    def _Xm_RegOffset_LDR : A64I_LSregoff<size, v, {high_opc, 0b1}, 0b1,
+                            (outs GPR:$Rt),
+                            (ins GPR64xsp:$Rn, GPR64:$Rm, params.regextXm:$Ext),
+                            "ldr" # asmsuffix # "\t$Rt, [$Rn, $Rm, $Ext]",
+                            [], NoItinerary>;
+  }
+  def : InstAlias<"ldr" # asmsuffix # " $Rt, [$Rn, $Rm]",
+        (!cast<Instruction>(prefix # "_Xm_RegOffset_LDR") GPR:$Rt, GPR64xsp:$Rn,
+                                                          GPR64:$Rm, 2)>;
+
+  let mayStore = 1 in {
+    def _Wm_RegOffset_STR : A64I_LSregoff<size, v, {high_opc, 0b0}, 0b0,
+                                  (outs), (ins GPR:$Rt, GPR64xsp:$Rn, GPR32:$Rm,
+                                               params.regextWm:$Ext),
+                                  "str" # asmsuffix # "\t$Rt, [$Rn, $Rm, $Ext]",
+                                  [], NoItinerary>;
+
+    def _Xm_RegOffset_STR : A64I_LSregoff<size, v, {high_opc, 0b0}, 0b1,
+                                  (outs), (ins GPR:$Rt, GPR64xsp:$Rn, GPR64:$Rm,
+                                               params.regextXm:$Ext),
+                                  "str" # asmsuffix # "\t$Rt, [$Rn, $Rm, $Ext]",
+                                  [], NoItinerary>;
+  }
+  def : InstAlias<"str" # asmsuffix # " $Rt, [$Rn, $Rm]",
+      (!cast<Instruction>(prefix # "_Xm_RegOffset_STR") GPR:$Rt, GPR64xsp:$Rn,
+                                                        GPR64:$Rm, 2)>;
+
+  // Unaligned immediate
+  def _STUR : A64I_LSunalimm<size, v, {high_opc, 0b0},
+                             (outs), (ins GPR:$Rt, GPR64xsp:$Rn, simm9:$SImm9),
+                             "stur" # asmsuffix # "\t$Rt, [$Rn, $SImm9]",
+                             [], NoItinerary> {
+    let mayStore = 1;
+  }
+  def : InstAlias<"stur" # asmsuffix # " $Rt, [$Rn]",
+               (!cast<Instruction>(prefix # "_STUR") GPR:$Rt, GPR64xsp:$Rn, 0)>;
+
+  def _LDUR : A64I_LSunalimm<size, v, {high_opc, 0b1},
+                             (outs GPR:$Rt), (ins GPR64xsp:$Rn, simm9:$SImm9),
+                             "ldur" # asmsuffix # "\t$Rt, [$Rn, $SImm9]",
+                             [], NoItinerary> {
+    let mayLoad = 1;
+  }
+  def : InstAlias<"ldur" # asmsuffix # " $Rt, [$Rn]",
+               (!cast<Instruction>(prefix # "_LDUR") GPR:$Rt, GPR64xsp:$Rn, 0)>;
+
+  // Post-indexed
+  def _PostInd_STR : A64I_LSpostind<size, v, {high_opc, 0b0},
+                               (outs GPR64xsp:$Rn_wb),
+                               (ins GPR:$Rt, GPR64xsp:$Rn, simm9:$SImm9),
+                               "str" # asmsuffix # "\t$Rt, [$Rn], $SImm9",
+                               [], NoItinerary> {
+    let Constraints = "$Rn = $Rn_wb";
+    let mayStore = 1;
+
+    // Decoder only needed for unpredictability checking (FIXME).
+    let DecoderMethod = "DecodeSingleIndexedInstruction";
+  }
+
+  def _PostInd_LDR : A64I_LSpostind<size, v, {high_opc, 0b1},
+                                    (outs GPR:$Rt, GPR64xsp:$Rn_wb),
+                                    (ins GPR64xsp:$Rn, simm9:$SImm9),
+                                    "ldr" # asmsuffix # "\t$Rt, [$Rn], $SImm9",
+                                    [], NoItinerary> {
+    let mayLoad = 1;
+    let Constraints = "$Rn = $Rn_wb";
+    let DecoderMethod = "DecodeSingleIndexedInstruction";
+  }
+
+  // Pre-indexed
+  def _PreInd_STR : A64I_LSpreind<size, v, {high_opc, 0b0},
+                               (outs GPR64xsp:$Rn_wb),
+                               (ins GPR:$Rt, GPR64xsp:$Rn, simm9:$SImm9),
+                               "str" # asmsuffix # "\t$Rt, [$Rn, $SImm9]!",
+                               [], NoItinerary> {
+    let Constraints = "$Rn = $Rn_wb";
+    let mayStore = 1;
+
+    // Decoder only needed for unpredictability checking (FIXME).
+    let DecoderMethod = "DecodeSingleIndexedInstruction";
+  }
+
+  def _PreInd_LDR : A64I_LSpreind<size, v, {high_opc, 0b1},
+                                    (outs GPR:$Rt, GPR64xsp:$Rn_wb),
+                                    (ins GPR64xsp:$Rn, simm9:$SImm9),
+                                    "ldr" # asmsuffix # "\t$Rt, [$Rn, $SImm9]!",
+                                    [], NoItinerary> {
+    let mayLoad = 1;
+    let Constraints = "$Rn = $Rn_wb";
+    let DecoderMethod = "DecodeSingleIndexedInstruction";
+  }
+
+}
+
+// STRB/LDRB: First define the instructions
+defm LS8
+  : A64I_LDRSTR_unsigned<"LS8", 0b00, 0b0, 0b0, "b", GPR32, byte_addrparams>;
+
+// STRH/LDRH
+defm LS16
+  : A64I_LDRSTR_unsigned<"LS16", 0b01, 0b0, 0b0, "h", GPR32, hword_addrparams>;
+
+
+// STR/LDR to/from a W register
+defm LS32
+  : A64I_LDRSTR_unsigned<"LS32", 0b10, 0b0, 0b0, "", GPR32, word_addrparams>;
+
+// STR/LDR to/from an X register
+defm LS64
+  : A64I_LDRSTR_unsigned<"LS64", 0b11, 0b0, 0b0, "", GPR64, dword_addrparams>;
+
+// STR/LDR to/from a B register
+defm LSFP8
+  : A64I_LDRSTR_unsigned<"LSFP8", 0b00, 0b1, 0b0, "", FPR8, byte_addrparams>;
+
+// STR/LDR to/from an H register
+defm LSFP16
+  : A64I_LDRSTR_unsigned<"LSFP16", 0b01, 0b1, 0b0, "", FPR16, hword_addrparams>;
+
+// STR/LDR to/from an S register
+defm LSFP32
+  : A64I_LDRSTR_unsigned<"LSFP32", 0b10, 0b1, 0b0, "", FPR32, word_addrparams>;
+// STR/LDR to/from a D register
+defm LSFP64
+  : A64I_LDRSTR_unsigned<"LSFP64", 0b11, 0b1, 0b0, "", FPR64, dword_addrparams>;
+// STR/LDR to/from a Q register
+defm LSFP128
+  : A64I_LDRSTR_unsigned<"LSFP128", 0b00, 0b1, 0b1, "", FPR128,
+                         qword_addrparams>;
+
+//===------------------------------
+// 2.3 Signed loads
+//===------------------------------
+
+// Byte and half-word signed loads can both go into either an X or a W register,
+// so it's worth factoring out. Signed word loads don't fit because there is no
+// W version.
+multiclass A64I_LDR_signed<bits<2> size, string asmopcode, AddrParams params,
+                           string prefix> {
+  // Unsigned offset
+  def w : A64I_LSunsigimm<size, 0b0, 0b11,
+                          (outs GPR32:$Rt),
+                          (ins GPR64xsp:$Rn, params.uimm12:$UImm12),
+                          "ldrs" # asmopcode # "\t$Rt, [$Rn, $UImm12]",
+                          [], NoItinerary> {
+    let mayLoad = 1;
+  }
+  def : InstAlias<"ldrs" # asmopcode # " $Rt, [$Rn]",
+                  (!cast<Instruction>(prefix # w) GPR32:$Rt, GPR64xsp:$Rn, 0)>;
+
+  def x : A64I_LSunsigimm<size, 0b0, 0b10,
+                          (outs GPR64:$Rt),
+                          (ins GPR64xsp:$Rn, params.uimm12:$UImm12),
+                          "ldrs" # asmopcode # "\t$Rt, [$Rn, $UImm12]",
+                          [], NoItinerary> {
+    let mayLoad = 1;
+  }
+  def : InstAlias<"ldrs" # asmopcode # " $Rt, [$Rn]",
+                  (!cast<Instruction>(prefix # x) GPR64:$Rt, GPR64xsp:$Rn, 0)>;
+
+  // Register offset
+  let mayLoad = 1 in {
+    def w_Wm_RegOffset : A64I_LSregoff<size, 0b0, 0b11, 0b0,
+                            (outs GPR32:$Rt),
+                            (ins GPR64xsp:$Rn, GPR32:$Rm, params.regextWm:$Ext),
+                            "ldrs" # asmopcode # "\t$Rt, [$Rn, $Rm, $Ext]",
+                            [], NoItinerary>;
+
+    def w_Xm_RegOffset : A64I_LSregoff<size, 0b0, 0b11, 0b1,
+                            (outs GPR32:$Rt),
+                            (ins GPR64xsp:$Rn, GPR64:$Rm, params.regextXm:$Ext),
+                            "ldrs" # asmopcode # "\t$Rt, [$Rn, $Rm, $Ext]",
+                            [], NoItinerary>;
+
+    def x_Wm_RegOffset : A64I_LSregoff<size, 0b0, 0b10, 0b0,
+                            (outs GPR64:$Rt),
+                            (ins GPR64xsp:$Rn, GPR32:$Rm, params.regextWm:$Ext),
+                            "ldrs" # asmopcode # "\t$Rt, [$Rn, $Rm, $Ext]",
+                            [], NoItinerary>;
+
+    def x_Xm_RegOffset : A64I_LSregoff<size, 0b0, 0b10, 0b1,
+                            (outs GPR64:$Rt),
+                            (ins GPR64xsp:$Rn, GPR64:$Rm, params.regextXm:$Ext),
+                            "ldrs" # asmopcode # "\t$Rt, [$Rn, $Rm, $Ext]",
+                            [], NoItinerary>;
+  }
+  def : InstAlias<"ldrs" # asmopcode # " $Rt, [$Rn, $Rm]",
+        (!cast<Instruction>(prefix # "w_Xm_RegOffset") GPR32:$Rt, GPR64xsp:$Rn,
+                                                       GPR64:$Rm, 2)>;
+
+  def : InstAlias<"ldrs" # asmopcode # " $Rt, [$Rn, $Rm]",
+        (!cast<Instruction>(prefix # "x_Xm_RegOffset") GPR64:$Rt, GPR64xsp:$Rn,
+                                                       GPR64:$Rm, 2)>;
+
+
+  let mayLoad = 1 in {
+    // Unaligned offset
+    def w_U : A64I_LSunalimm<size, 0b0, 0b11,
+                             (outs GPR32:$Rt),
+                             (ins GPR64xsp:$Rn, simm9:$SImm9),
+                             "ldurs" # asmopcode # "\t$Rt, [$Rn, $SImm9]",
+                             [], NoItinerary>;
+
+    def x_U : A64I_LSunalimm<size, 0b0, 0b10,
+                             (outs GPR64:$Rt),
+                             (ins GPR64xsp:$Rn, simm9:$SImm9),
+                             "ldurs" # asmopcode # "\t$Rt, [$Rn, $SImm9]",
+                             [], NoItinerary>;
+
+
+    // Post-indexed
+    def w_PostInd : A64I_LSpostind<size, 0b0, 0b11,
+                                 (outs GPR32:$Rt, GPR64xsp:$Rn_wb),
+                                 (ins GPR64xsp:$Rn, simm9:$SImm9),
+                                 "ldrs" # asmopcode # "\t$Rt, [$Rn], $SImm9",
+                                 [], NoItinerary> {
+      let Constraints = "$Rn = $Rn_wb";
+      let DecoderMethod = "DecodeSingleIndexedInstruction";
+    }
+
+    def x_PostInd : A64I_LSpostind<size, 0b0, 0b10,
+                                   (outs GPR64:$Rt, GPR64xsp:$Rn_wb),
+                                   (ins GPR64xsp:$Rn, simm9:$SImm9),
+                                   "ldrs" # asmopcode # "\t$Rt, [$Rn], $SImm9",
+                                   [], NoItinerary> {
+      let Constraints = "$Rn = $Rn_wb";
+      let DecoderMethod = "DecodeSingleIndexedInstruction";
+    }
+
+    // Pre-indexed
+    def w_PreInd : A64I_LSpreind<size, 0b0, 0b11,
+                                 (outs GPR32:$Rt, GPR64xsp:$Rn_wb),
+                                 (ins GPR64xsp:$Rn, simm9:$SImm9),
+                                 "ldrs" # asmopcode # "\t$Rt, [$Rn, $SImm9]!",
+                                 [], NoItinerary> {
+      let Constraints = "$Rn = $Rn_wb";
+      let DecoderMethod = "DecodeSingleIndexedInstruction";
+    }
+
+    def x_PreInd : A64I_LSpreind<size, 0b0, 0b10,
+                                 (outs GPR64:$Rt, GPR64xsp:$Rn_wb),
+                                 (ins GPR64xsp:$Rn, simm9:$SImm9),
+                                 "ldrs" # asmopcode # "\t$Rt, [$Rn, $SImm9]!",
+                                 [], NoItinerary> {
+      let Constraints = "$Rn = $Rn_wb";
+      let DecoderMethod = "DecodeSingleIndexedInstruction";
+    }
+  } // let mayLoad = 1
+}
+
+// LDRSB
+defm LDRSB : A64I_LDR_signed<0b00, "b", byte_addrparams, "LDRSB">;
+// LDRSH
+defm LDRSH : A64I_LDR_signed<0b01, "h", hword_addrparams, "LDRSH">;
+
+// LDRSW: load a 32-bit register, sign-extending to 64-bits.
+def LDRSWx
+    : A64I_LSunsigimm<0b10, 0b0, 0b10,
+                    (outs GPR64:$Rt),
+                    (ins GPR64xsp:$Rn, word_uimm12:$UImm12),
+                    "ldrsw\t$Rt, [$Rn, $UImm12]",
+                    [], NoItinerary> {
+  let mayLoad = 1;
+}
+def : InstAlias<"ldrsw $Rt, [$Rn]", (LDRSWx GPR64:$Rt, GPR64xsp:$Rn, 0)>;
+
+let mayLoad = 1 in {
+  def LDRSWx_Wm_RegOffset : A64I_LSregoff<0b10, 0b0, 0b10, 0b0,
+                             (outs GPR64:$Rt),
+                             (ins GPR64xsp:$Rn, GPR32:$Rm, word_Wm_regext:$Ext),
+                             "ldrsw\t$Rt, [$Rn, $Rm, $Ext]",
+                             [], NoItinerary>;
+
+  def LDRSWx_Xm_RegOffset : A64I_LSregoff<0b10, 0b0, 0b10, 0b1,
+                             (outs GPR64:$Rt),
+                             (ins GPR64xsp:$Rn, GPR64:$Rm, word_Xm_regext:$Ext),
+                             "ldrsw\t$Rt, [$Rn, $Rm, $Ext]",
+                             [], NoItinerary>;
+}
+def : InstAlias<"ldrsw $Rt, [$Rn, $Rm]",
+                (LDRSWx_Xm_RegOffset GPR64:$Rt, GPR64xsp:$Rn, GPR64:$Rm, 2)>;
+
+
+def LDURSWx
+    : A64I_LSunalimm<0b10, 0b0, 0b10,
+                    (outs GPR64:$Rt),
+                    (ins GPR64xsp:$Rn, simm9:$SImm9),
+                    "ldursw\t$Rt, [$Rn, $SImm9]",
+                    [], NoItinerary> {
+  let mayLoad = 1;
+}
+def : InstAlias<"ldursw $Rt, [$Rn]", (LDURSWx GPR64:$Rt, GPR64xsp:$Rn, 0)>;
+
+def LDRSWx_PostInd
+    : A64I_LSpostind<0b10, 0b0, 0b10,
+                    (outs GPR64:$Rt, GPR64xsp:$Rn_wb),
+                    (ins GPR64xsp:$Rn, simm9:$SImm9),
+                    "ldrsw\t$Rt, [$Rn], $SImm9",
+                    [], NoItinerary> {
+  let mayLoad = 1;
+  let Constraints = "$Rn = $Rn_wb";
+  let DecoderMethod = "DecodeSingleIndexedInstruction";
+}
+
+def LDRSWx_PreInd : A64I_LSpreind<0b10, 0b0, 0b10,
+                                 (outs GPR64:$Rt, GPR64xsp:$Rn_wb),
+                                 (ins GPR64xsp:$Rn, simm9:$SImm9),
+                                 "ldrsw\t$Rt, [$Rn, $SImm9]!",
+                                 [], NoItinerary> {
+  let mayLoad = 1;
+  let Constraints = "$Rn = $Rn_wb";
+  let DecoderMethod = "DecodeSingleIndexedInstruction";
+}
+
+//===------------------------------
+// 2.4 Prefetch operations
+//===------------------------------
+
+def PRFM : A64I_LSunsigimm<0b11, 0b0, 0b10, (outs),
+                 (ins prefetch_op:$Rt, GPR64xsp:$Rn, dword_uimm12:$UImm12),
+                 "prfm\t$Rt, [$Rn, $UImm12]",
+                 [], NoItinerary> {
+  let mayLoad = 1;
+}
+def : InstAlias<"prfm $Rt, [$Rn]",
+                (PRFM prefetch_op:$Rt, GPR64xsp:$Rn, 0)>;
+
+let mayLoad = 1 in {
+  def PRFM_Wm_RegOffset : A64I_LSregoff<0b11, 0b0, 0b10, 0b0, (outs),
+                                        (ins prefetch_op:$Rt, GPR64xsp:$Rn,
+                                             GPR32:$Rm, dword_Wm_regext:$Ext),
+                                        "prfm\t$Rt, [$Rn, $Rm, $Ext]",
+                                        [], NoItinerary>;
+  def PRFM_Xm_RegOffset : A64I_LSregoff<0b11, 0b0, 0b10, 0b1, (outs),
+                                        (ins prefetch_op:$Rt, GPR64xsp:$Rn,
+                                             GPR64:$Rm, dword_Xm_regext:$Ext),
+                                        "prfm\t$Rt, [$Rn, $Rm, $Ext]",
+                                        [], NoItinerary>;
+}
+
+def : InstAlias<"prfm $Rt, [$Rn, $Rm]",
+                (PRFM_Xm_RegOffset prefetch_op:$Rt, GPR64xsp:$Rn,
+                                   GPR64:$Rm, 2)>;
+
+
+def PRFUM : A64I_LSunalimm<0b11, 0b0, 0b10, (outs),
+                         (ins prefetch_op:$Rt, GPR64xsp:$Rn, simm9:$SImm9),
+                         "prfum\t$Rt, [$Rn, $SImm9]",
+                         [], NoItinerary> {
+  let mayLoad = 1;
+}
+def : InstAlias<"prfum $Rt, [$Rn]",
+                (PRFUM prefetch_op:$Rt, GPR64xsp:$Rn, 0)>;
+
+//===----------------------------------------------------------------------===//
+// Load-store register (unprivileged) instructions
+//===----------------------------------------------------------------------===//
+// Contains: LDTRB, LDTRH, LDTRSB, LDTRSH, LDTRSW, STTR, STTRB and STTRH
+
+// These instructions very much mirror the "unscaled immediate" loads, but since
+// there are no floating-point variants we need to split them out into their own
+// section to avoid instantiation of "ldtr d0, [sp]" etc.
+
+multiclass A64I_LDTRSTTR<bits<2> size, string asmsuffix, RegisterClass GPR,
+                         string prefix> {
+  def _UnPriv_STR : A64I_LSunpriv<size, 0b0, 0b00,
+                              (outs), (ins GPR:$Rt, GPR64xsp:$Rn, simm9:$SImm9),
+                              "sttr" # asmsuffix # "\t$Rt, [$Rn, $SImm9]",
+                              [], NoItinerary> {
+    let mayStore = 1;
+  }
+
+  def : InstAlias<"sttr" # asmsuffix # " $Rt, [$Rn]",
+         (!cast<Instruction>(prefix # "_UnPriv_STR") GPR:$Rt, GPR64xsp:$Rn, 0)>;
+
+  def _UnPriv_LDR : A64I_LSunpriv<size, 0b0, 0b01,
+                               (outs GPR:$Rt), (ins GPR64xsp:$Rn, simm9:$SImm9),
+                               "ldtr" # asmsuffix # "\t$Rt, [$Rn, $SImm9]",
+                               [], NoItinerary> {
+    let mayLoad = 1;
+  }
+
+  def : InstAlias<"ldtr" # asmsuffix # " $Rt, [$Rn]",
+         (!cast<Instruction>(prefix # "_UnPriv_LDR") GPR:$Rt, GPR64xsp:$Rn, 0)>;
+
+}
+
+// STTRB/LDTRB: First define the instructions
+defm LS8 : A64I_LDTRSTTR<0b00, "b", GPR32, "LS8">;
+
+// STTRH/LDTRH
+defm LS16 : A64I_LDTRSTTR<0b01, "h", GPR32, "LS16">;
+
+// STTR/LDTR to/from a W register
+defm LS32 : A64I_LDTRSTTR<0b10, "", GPR32, "LS32">;
+
+// STTR/LDTR to/from an X register
+defm LS64 : A64I_LDTRSTTR<0b11, "", GPR64, "LS64">;
+
+// Now a class for the signed instructions that can go to either 32 or 64
+// bits...
+multiclass A64I_LDTR_signed<bits<2> size, string asmopcode, string prefix> {
+  let mayLoad = 1 in {
+    def w : A64I_LSunpriv<size, 0b0, 0b11,
+                          (outs GPR32:$Rt),
+                          (ins GPR64xsp:$Rn, simm9:$SImm9),
+                          "ldtrs" # asmopcode # "\t$Rt, [$Rn, $SImm9]",
+                          [], NoItinerary>;
+
+    def x : A64I_LSunpriv<size, 0b0, 0b10,
+                          (outs GPR64:$Rt),
+                          (ins GPR64xsp:$Rn, simm9:$SImm9),
+                          "ldtrs" # asmopcode # "\t$Rt, [$Rn, $SImm9]",
+                          [], NoItinerary>;
+  }
+
+  def : InstAlias<"ldtrs" # asmopcode # " $Rt, [$Rn]",
+                 (!cast<Instruction>(prefix # "w") GPR32:$Rt, GPR64xsp:$Rn, 0)>;
+
+  def : InstAlias<"ldtrs" # asmopcode # " $Rt, [$Rn]",
+                 (!cast<Instruction>(prefix # "x") GPR64:$Rt, GPR64xsp:$Rn, 0)>;
+
+}
+
+// LDTRSB
+defm LDTRSB : A64I_LDTR_signed<0b00, "b", "LDTRSB">;
+// LDTRSH
+defm LDTRSH : A64I_LDTR_signed<0b01, "h", "LDTRSH">;
+
+// And finally LDTRSW which only goes to 64 bits.
+def LDTRSWx : A64I_LSunpriv<0b10, 0b0, 0b10,
+                            (outs GPR64:$Rt),
+                            (ins GPR64xsp:$Rn, simm9:$SImm9),
+                            "ldtrsw\t$Rt, [$Rn, $SImm9]",
+                            [], NoItinerary> {
+  let mayLoad = 1;
+}
+def : InstAlias<"ldtrsw $Rt, [$Rn]", (LDTRSWx GPR64:$Rt, GPR64xsp:$Rn, 0)>;
+
+//===----------------------------------------------------------------------===//
+// Load-store register pair (offset) instructions
+//===----------------------------------------------------------------------===//
+//
+// and
+//
+//===----------------------------------------------------------------------===//
+// Load-store register pair (post-indexed) instructions
+//===----------------------------------------------------------------------===//
+// Contains: STP, LDP, LDPSW
+//
+// and
+//
+//===----------------------------------------------------------------------===//
+// Load-store register pair (pre-indexed) instructions
+//===----------------------------------------------------------------------===//
+// Contains: STP, LDP, LDPSW
+//
+// and
+//
+//===----------------------------------------------------------------------===//
+// Load-store non-temporal register pair (offset) instructions
+//===----------------------------------------------------------------------===//
+// Contains: STNP, LDNP
+
+
+// Anything that creates an MCInst (Decoding, selection and AsmParsing) has to
+// know the access size via some means. An isolated operand does not have this
+// information unless told from here, which means we need separate tablegen
+// Operands for each access size. This multiclass takes care of instantiating
+// the correct template functions in the rest of the backend.
+
+multiclass offsets_simm7<string MemSize, string prefix> {
+  // The bare signed 7-bit immediate is used in post-indexed instructions, but
+  // because of the scaling performed a generic "simm7" operand isn't
+  // appropriate here either.
+  def simm7_asmoperand : AsmOperandClass {
+    let Name = "SImm7_Scaled" # MemSize;
+    let PredicateMethod = "isSImm7Scaled<" # MemSize # ">";
+    let RenderMethod = "addSImm7ScaledOperands<" # MemSize # ">";
+    let DiagnosticType = "LoadStoreSImm7_" # MemSize;
+  }
+
+  def simm7 : Operand<i64> {
+    let PrintMethod = "printSImm7ScaledOperand<" # MemSize # ">";
+    let ParserMatchClass = !cast<AsmOperandClass>(prefix # "simm7_asmoperand");
+  }
+}
+
+defm word_  : offsets_simm7<"4", "word_">;
+defm dword_ : offsets_simm7<"8", "dword_">;
+defm qword_ : offsets_simm7<"16", "qword_">;
+
+multiclass A64I_LSPsimple<bits<2> opc, bit v, RegisterClass SomeReg,
+                          Operand simm7, string prefix> {
+  def _STR : A64I_LSPoffset<opc, v, 0b0, (outs),
+                    (ins SomeReg:$Rt, SomeReg:$Rt2, GPR64xsp:$Rn, simm7:$SImm7),
+                    "stp\t$Rt, $Rt2, [$Rn, $SImm7]", [], NoItinerary> {
+    let mayStore = 1;
+    let DecoderMethod = "DecodeLDSTPairInstruction";
+  }
+  def : InstAlias<"stp $Rt, $Rt2, [$Rn]",
+                  (!cast<Instruction>(prefix # "_STR") SomeReg:$Rt,
+                                                SomeReg:$Rt2, GPR64xsp:$Rn, 0)>;
+
+  def _LDR : A64I_LSPoffset<opc, v, 0b1,
+                            (outs SomeReg:$Rt, SomeReg:$Rt2),
+                            (ins GPR64xsp:$Rn, simm7:$SImm7),
+                            "ldp\t$Rt, $Rt2, [$Rn, $SImm7]", [], NoItinerary> {
+    let mayLoad = 1;
+    let DecoderMethod = "DecodeLDSTPairInstruction";
+  }
+  def : InstAlias<"ldp $Rt, $Rt2, [$Rn]",
+                  (!cast<Instruction>(prefix # "_LDR") SomeReg:$Rt,
+                                                SomeReg:$Rt2, GPR64xsp:$Rn, 0)>;
+
+  def _PostInd_STR : A64I_LSPpostind<opc, v, 0b0,
+                               (outs GPR64xsp:$Rn_wb),
+                               (ins SomeReg:$Rt, SomeReg:$Rt2,
+                                    GPR64xsp:$Rn,
+                                    simm7:$SImm7),
+                               "stp\t$Rt, $Rt2, [$Rn], $SImm7",
+                               [], NoItinerary> {
+    let mayStore = 1;
+    let Constraints = "$Rn = $Rn_wb";
+
+    // Decoder only needed for unpredictability checking (FIXME).
+    let DecoderMethod = "DecodeLDSTPairInstruction";
+  }
+
+  def _PostInd_LDR : A64I_LSPpostind<opc, v, 0b1,
+                        (outs SomeReg:$Rt, SomeReg:$Rt2, GPR64xsp:$Rn_wb),
+                        (ins GPR64xsp:$Rn, simm7:$SImm7),
+                        "ldp\t$Rt, $Rt2, [$Rn], $SImm7",
+                        [], NoItinerary> {
+    let mayLoad = 1;
+    let Constraints = "$Rn = $Rn_wb";
+    let DecoderMethod = "DecodeLDSTPairInstruction";
+  }
+
+  def _PreInd_STR : A64I_LSPpreind<opc, v, 0b0, (outs GPR64xsp:$Rn_wb),
+                    (ins SomeReg:$Rt, SomeReg:$Rt2, GPR64xsp:$Rn, simm7:$SImm7),
+                    "stp\t$Rt, $Rt2, [$Rn, $SImm7]!",
+                    [], NoItinerary> {
+    let mayStore = 1;
+    let Constraints = "$Rn = $Rn_wb";
+    let DecoderMethod = "DecodeLDSTPairInstruction";
+  }
+
+  def _PreInd_LDR : A64I_LSPpreind<opc, v, 0b1,
+                              (outs SomeReg:$Rt, SomeReg:$Rt2, GPR64xsp:$Rn_wb),
+                              (ins GPR64xsp:$Rn, simm7:$SImm7),
+                              "ldp\t$Rt, $Rt2, [$Rn, $SImm7]!",
+                              [], NoItinerary> {
+    let mayLoad = 1;
+    let Constraints = "$Rn = $Rn_wb";
+    let DecoderMethod = "DecodeLDSTPairInstruction";
+  }
+
+  def _NonTemp_STR : A64I_LSPnontemp<opc, v, 0b0, (outs),
+                    (ins SomeReg:$Rt, SomeReg:$Rt2, GPR64xsp:$Rn, simm7:$SImm7),
+                    "stnp\t$Rt, $Rt2, [$Rn, $SImm7]", [], NoItinerary> {
+    let mayStore = 1;
+    let DecoderMethod = "DecodeLDSTPairInstruction";
+  }
+  def : InstAlias<"stnp $Rt, $Rt2, [$Rn]",
+                  (!cast<Instruction>(prefix # "_NonTemp_STR") SomeReg:$Rt,
+                                                SomeReg:$Rt2, GPR64xsp:$Rn, 0)>;
+
+  def _NonTemp_LDR : A64I_LSPnontemp<opc, v, 0b1,
+                            (outs SomeReg:$Rt, SomeReg:$Rt2),
+                            (ins GPR64xsp:$Rn, simm7:$SImm7),
+                            "ldnp\t$Rt, $Rt2, [$Rn, $SImm7]", [], NoItinerary> {
+    let mayLoad = 1;
+    let DecoderMethod = "DecodeLDSTPairInstruction";
+  }
+  def : InstAlias<"ldnp $Rt, $Rt2, [$Rn]",
+                  (!cast<Instruction>(prefix # "_NonTemp_LDR") SomeReg:$Rt,
+                                                SomeReg:$Rt2, GPR64xsp:$Rn, 0)>;
+
+}
+
+
+defm LSPair32 : A64I_LSPsimple<0b00, 0b0, GPR32, word_simm7, "LSPair32">;
+defm LSPair64 : A64I_LSPsimple<0b10, 0b0, GPR64, dword_simm7, "LSPair64">;
+defm LSFPPair32 : A64I_LSPsimple<0b00, 0b1, FPR32, word_simm7, "LSFPPair32">;
+defm LSFPPair64 : A64I_LSPsimple<0b01, 0b1, FPR64,  dword_simm7, "LSFPPair64">;
+defm LSFPPair128 : A64I_LSPsimple<0b10, 0b1, FPR128, qword_simm7,
+                                  "LSFPPair128">;
+
+
+def LDPSWx : A64I_LSPoffset<0b01, 0b0, 0b1,
+                           (outs GPR64:$Rt, GPR64:$Rt2),
+                           (ins GPR64xsp:$Rn, word_simm7:$SImm7),
+                           "ldpsw\t$Rt, $Rt2, [$Rn, $SImm7]", [], NoItinerary> {
+  let mayLoad = 1;
+  let DecoderMethod = "DecodeLDSTPairInstruction";
+}
+def : InstAlias<"ldpsw $Rt, $Rt2, [$Rn]",
+                (LDPSWx GPR64:$Rt, GPR64:$Rt2, GPR64xsp:$Rn, 0)>;
+
+def LDPSWx_PostInd : A64I_LSPpostind<0b01, 0b0, 0b1,
+                                  (outs GPR64:$Rt, GPR64:$Rt2, GPR64:$Rn_wb),
+                                  (ins GPR64xsp:$Rn, word_simm7:$SImm7),
+                                  "ldpsw\t$Rt, $Rt2, [$Rn], $SImm7",
+                                  [], NoItinerary> {
+  let mayLoad = 1;
+  let Constraints = "$Rn = $Rn_wb";
+  let DecoderMethod = "DecodeLDSTPairInstruction";
+}
+
+def LDPSWx_PreInd : A64I_LSPpreind<0b01, 0b0, 0b1,
+                                   (outs GPR64:$Rt, GPR64:$Rt2, GPR64:$Rn_wb),
+                                   (ins GPR64xsp:$Rn, word_simm7:$SImm7),
+                                   "ldpsw\t$Rt, $Rt2, [$Rn, $SImm7]!",
+                                   [], NoItinerary> {
+  let mayLoad = 1;
+  let Constraints = "$Rn = $Rn_wb";
+  let DecoderMethod = "DecodeLDSTPairInstruction";
+}
+
+//===----------------------------------------------------------------------===//
+// Logical (immediate) instructions
+//===----------------------------------------------------------------------===//
+// Contains: AND, ORR, EOR, ANDS, + aliases TST, MOV
+
+multiclass logical_imm_operands<string prefix, string note,
+                                int size, ValueType VT> {
+  def _asmoperand : AsmOperandClass {
+    let Name = "LogicalImm" # note # size;
+    let PredicateMethod = "isLogicalImm" # note # "<" # size # ">";
+    let RenderMethod = "addLogicalImmOperands<" # size # ">";
+    let DiagnosticType = "LogicalSecondSource";
+  }
+
+  def _operand
+        : Operand<VT>, ComplexPattern<VT, 1, "SelectLogicalImm", [imm]> {
+    let ParserMatchClass = !cast<AsmOperandClass>(prefix # "_asmoperand");
+    let PrintMethod = "printLogicalImmOperand<" # size # ">";
+    let DecoderMethod = "DecodeLogicalImmOperand<" # size # ">";
+  }
+}
+
+defm logical_imm32 : logical_imm_operands<"logical_imm32", "", 32, i32>;
+defm logical_imm64 : logical_imm_operands<"logical_imm64", "", 64, i64>;
+
+// The mov versions only differ in assembly parsing, where they
+// exclude values representable with either MOVZ or MOVN.
+defm logical_imm32_mov
+  : logical_imm_operands<"logical_imm32_mov", "MOV", 32, i32>;
+defm logical_imm64_mov
+  : logical_imm_operands<"logical_imm64_mov", "MOV", 64, i64>;
+
+
+multiclass A64I_logimmSizes<bits<2> opc, string asmop, SDNode opnode> {
+  def wwi : A64I_logicalimm<0b0, opc, (outs GPR32wsp:$Rd),
+                         (ins GPR32:$Rn, logical_imm32_operand:$Imm),
+                         !strconcat(asmop, "\t$Rd, $Rn, $Imm"),
+                         [(set i32:$Rd,
+                               (opnode i32:$Rn, logical_imm32_operand:$Imm))],
+                         NoItinerary>;
+
+  def xxi : A64I_logicalimm<0b1, opc, (outs GPR64xsp:$Rd),
+                         (ins GPR64:$Rn, logical_imm64_operand:$Imm),
+                         !strconcat(asmop, "\t$Rd, $Rn, $Imm"),
+                         [(set i64:$Rd,
+                               (opnode i64:$Rn, logical_imm64_operand:$Imm))],
+                         NoItinerary>;
+}
+
+defm AND : A64I_logimmSizes<0b00, "and", and>;
+defm ORR : A64I_logimmSizes<0b01, "orr", or>;
+defm EOR : A64I_logimmSizes<0b10, "eor", xor>;
+
+let Defs = [NZCV] in {
+  def ANDSwwi : A64I_logicalimm<0b0, 0b11, (outs GPR32:$Rd),
+                                (ins GPR32:$Rn, logical_imm32_operand:$Imm),
+                                "ands\t$Rd, $Rn, $Imm",
+                                [], NoItinerary>;
+
+  def ANDSxxi : A64I_logicalimm<0b1, 0b11, (outs GPR64:$Rd),
+                                (ins GPR64:$Rn, logical_imm64_operand:$Imm),
+                                "ands\t$Rd, $Rn, $Imm",
+                                [], NoItinerary>;
+}
+
+
+def : InstAlias<"tst $Rn, $Imm",
+                (ANDSwwi WZR, GPR32:$Rn, logical_imm32_operand:$Imm)>;
+def : InstAlias<"tst $Rn, $Imm",
+                (ANDSxxi XZR, GPR64:$Rn, logical_imm64_operand:$Imm)>;
+def : InstAlias<"mov $Rd, $Imm",
+                (ORRwwi GPR32wsp:$Rd, WZR, logical_imm32_mov_operand:$Imm)>;
+def : InstAlias<"mov $Rd, $Imm",
+                (ORRxxi GPR64xsp:$Rd, XZR, logical_imm64_mov_operand:$Imm)>;
+
+//===----------------------------------------------------------------------===//
+// Logical (shifted register) instructions
+//===----------------------------------------------------------------------===//
+// Contains: AND, BIC, ORR, ORN, EOR, EON, ANDS, BICS + aliases TST, MVN, MOV
+
+// Operand for optimizing (icmp (and LHS, RHS), 0, SomeCode). In theory "ANDS"
+// behaves differently for unsigned comparisons, so we defensively only allow
+// signed or n/a as the operand. In practice "unsigned greater than 0" is "not
+// equal to 0" and LLVM gives us this.
+def signed_cond : PatLeaf<(cond), [{
+  return !isUnsignedIntSetCC(N->get());
+}]>;
+
+
+// These instructions share their "shift" operands with add/sub (shifted
+// register instructions). They are defined there.
+
+// N.b. the commutable parameter is just !N. It will be first against the wall
+// when the revolution comes.
+multiclass logical_shifts<string prefix, bit sf, bits<2> opc,
+                          bit N, bit commutable,
+                          string asmop, SDPatternOperator opfrag, ValueType ty,
+                          RegisterClass GPR, list<Register> defs> {
+  let isCommutable = commutable, Defs = defs in {
+  def _lsl : A64I_logicalshift<sf, opc, 0b00, N,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("lsl_operand_" # ty):$Imm6),
+                       !strconcat(asmop, "\t$Rd, $Rn, $Rm, $Imm6"),
+                       [(set ty:$Rd, (opfrag ty:$Rn, (shl ty:$Rm,
+                            !cast<Operand>("lsl_operand_" # ty):$Imm6))
+                       )],
+                       NoItinerary>;
+
+  def _lsr : A64I_logicalshift<sf, opc, 0b01, N,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("lsr_operand_" # ty):$Imm6),
+                       !strconcat(asmop, "\t$Rd, $Rn, $Rm, $Imm6"),
+                       [(set ty:$Rd, (opfrag ty:$Rn, (srl ty:$Rm,
+                            !cast<Operand>("lsr_operand_" # ty):$Imm6))
+                       )],
+                       NoItinerary>;
+
+  def _asr : A64I_logicalshift<sf, opc, 0b10, N,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("asr_operand_" # ty):$Imm6),
+                       !strconcat(asmop, "\t$Rd, $Rn, $Rm, $Imm6"),
+                       [(set ty:$Rd, (opfrag ty:$Rn, (sra ty:$Rm,
+                            !cast<Operand>("asr_operand_" # ty):$Imm6))
+                       )],
+                       NoItinerary>;
+
+  def _ror : A64I_logicalshift<sf, opc, 0b11, N,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("ror_operand_" # ty):$Imm6),
+                       !strconcat(asmop, "\t$Rd, $Rn, $Rm, $Imm6"),
+                       [(set ty:$Rd, (opfrag ty:$Rn, (rotr ty:$Rm,
+                            !cast<Operand>("ror_operand_" # ty):$Imm6))
+                       )],
+                       NoItinerary>;
+  }
+
+  def _noshift
+      : InstAlias<!strconcat(asmop, " $Rd, $Rn, $Rm"),
+                 (!cast<Instruction>(prefix # "_lsl") GPR:$Rd, GPR:$Rn,
+                                                      GPR:$Rm, 0)>;
+
+  def : Pat<(opfrag ty:$Rn, ty:$Rm),
+            (!cast<Instruction>(prefix # "_lsl") $Rn, $Rm, 0)>;
+}
+
+multiclass logical_sizes<string prefix, bits<2> opc, bit N, bit commutable,
+                         string asmop, SDPatternOperator opfrag,
+                         list<Register> defs> {
+  defm xxx : logical_shifts<prefix # "xxx", 0b1, opc, N,
+                            commutable, asmop, opfrag, i64, GPR64, defs>;
+  defm www : logical_shifts<prefix # "www", 0b0, opc, N,
+                            commutable, asmop, opfrag, i32, GPR32, defs>;
+}
+
+
+defm AND : logical_sizes<"AND", 0b00, 0b0, 0b1, "and", and, []>;
+defm ORR : logical_sizes<"ORR", 0b01, 0b0, 0b1, "orr", or, []>;
+defm EOR : logical_sizes<"EOR", 0b10, 0b0, 0b1, "eor", xor, []>;
+defm ANDS : logical_sizes<"ANDS", 0b11, 0b0, 0b1, "ands",
+             PatFrag<(ops node:$lhs, node:$rhs), (and node:$lhs, node:$rhs),
+                     [{ (void)N; return false; }]>,
+             [NZCV]>;
+
+defm BIC : logical_sizes<"BIC", 0b00, 0b1, 0b0, "bic",
+                         PatFrag<(ops node:$lhs, node:$rhs),
+                                 (and node:$lhs, (not node:$rhs))>, []>;
+defm ORN : logical_sizes<"ORN", 0b01, 0b1, 0b0, "orn",
+                         PatFrag<(ops node:$lhs, node:$rhs),
+                                 (or node:$lhs, (not node:$rhs))>, []>;
+defm EON : logical_sizes<"EON", 0b10, 0b1, 0b0, "eon",
+                         PatFrag<(ops node:$lhs, node:$rhs),
+                                 (xor node:$lhs, (not node:$rhs))>, []>;
+defm BICS : logical_sizes<"BICS", 0b11, 0b1, 0b0, "bics",
+                          PatFrag<(ops node:$lhs, node:$rhs),
+                                  (and node:$lhs, (not node:$rhs)),
+                                  [{ (void)N; return false; }]>,
+                          [NZCV]>;
+
+multiclass tst_shifts<string prefix, bit sf, ValueType ty, RegisterClass GPR> {
+  let isCommutable = 1, Rd = 0b11111, Defs = [NZCV] in {
+  def _lsl : A64I_logicalshift<sf, 0b11, 0b00, 0b0,
+                       (outs),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("lsl_operand_" # ty):$Imm6),
+                       "tst\t$Rn, $Rm, $Imm6",
+                       [(set NZCV, (A64setcc (and ty:$Rn, (shl ty:$Rm,
+                           !cast<Operand>("lsl_operand_" # ty):$Imm6)),
+                                          0, signed_cond))],
+                       NoItinerary>;
+
+
+  def _lsr : A64I_logicalshift<sf, 0b11, 0b01, 0b0,
+                       (outs),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("lsr_operand_" # ty):$Imm6),
+                       "tst\t$Rn, $Rm, $Imm6",
+                       [(set NZCV, (A64setcc (and ty:$Rn, (srl ty:$Rm,
+                           !cast<Operand>("lsr_operand_" # ty):$Imm6)),
+                                          0, signed_cond))],
+                       NoItinerary>;
+
+  def _asr : A64I_logicalshift<sf, 0b11, 0b10, 0b0,
+                       (outs),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("asr_operand_" # ty):$Imm6),
+                       "tst\t$Rn, $Rm, $Imm6",
+                       [(set NZCV, (A64setcc (and ty:$Rn, (sra ty:$Rm,
+                           !cast<Operand>("asr_operand_" # ty):$Imm6)),
+                                          0, signed_cond))],
+                       NoItinerary>;
+
+  def _ror : A64I_logicalshift<sf, 0b11, 0b11, 0b0,
+                       (outs),
+                       (ins GPR:$Rn, GPR:$Rm,
+                            !cast<Operand>("ror_operand_" # ty):$Imm6),
+                       "tst\t$Rn, $Rm, $Imm6",
+                       [(set NZCV, (A64setcc (and ty:$Rn, (rotr ty:$Rm,
+                           !cast<Operand>("ror_operand_" # ty):$Imm6)),
+                                          0, signed_cond))],
+                       NoItinerary>;
+  }
+
+  def _noshift : InstAlias<"tst $Rn, $Rm",
+                     (!cast<Instruction>(prefix # "_lsl") GPR:$Rn, GPR:$Rm, 0)>;
+
+  def : Pat<(A64setcc (and ty:$Rn, ty:$Rm), 0, signed_cond),
+            (!cast<Instruction>(prefix # "_lsl") $Rn, $Rm, 0)>;
+}
+
+defm TSTxx : tst_shifts<"TSTxx", 0b1, i64, GPR64>;
+defm TSTww : tst_shifts<"TSTww", 0b0, i32, GPR32>;
+
+
+multiclass mvn_shifts<string prefix, bit sf, ValueType ty, RegisterClass GPR> {
+  let isCommutable = 0, Rn = 0b11111 in {
+  def _lsl : A64I_logicalshift<sf, 0b01, 0b00, 0b1,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rm,
+                            !cast<Operand>("lsl_operand_" # ty):$Imm6),
+                       "mvn\t$Rd, $Rm, $Imm6",
+                       [(set ty:$Rd, (not (shl ty:$Rm,
+                         !cast<Operand>("lsl_operand_" # ty):$Imm6)))],
+                       NoItinerary>;
+
+
+  def _lsr : A64I_logicalshift<sf, 0b01, 0b01, 0b1,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rm,
+                            !cast<Operand>("lsr_operand_" # ty):$Imm6),
+                       "mvn\t$Rd, $Rm, $Imm6",
+                       [(set ty:$Rd, (not (srl ty:$Rm,
+                         !cast<Operand>("lsr_operand_" # ty):$Imm6)))],
+                       NoItinerary>;
+
+  def _asr : A64I_logicalshift<sf, 0b01, 0b10, 0b1,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rm,
+                            !cast<Operand>("asr_operand_" # ty):$Imm6),
+                       "mvn\t$Rd, $Rm, $Imm6",
+                       [(set ty:$Rd, (not (sra ty:$Rm,
+                         !cast<Operand>("asr_operand_" # ty):$Imm6)))],
+                       NoItinerary>;
+
+  def _ror : A64I_logicalshift<sf, 0b01, 0b11, 0b1,
+                       (outs GPR:$Rd),
+                       (ins GPR:$Rm,
+                            !cast<Operand>("ror_operand_" # ty):$Imm6),
+                       "mvn\t$Rd, $Rm, $Imm6",
+                       [(set ty:$Rd, (not (rotr ty:$Rm,
+                         !cast<Operand>("lsl_operand_" # ty):$Imm6)))],
+                       NoItinerary>;
+  }
+
+  def _noshift : InstAlias<"mvn $Rn, $Rm",
+                     (!cast<Instruction>(prefix # "_lsl") GPR:$Rn, GPR:$Rm, 0)>;
+
+  def : Pat<(not ty:$Rm),
+            (!cast<Instruction>(prefix # "_lsl") $Rm, 0)>;
+}
+
+defm MVNxx : mvn_shifts<"MVNxx", 0b1, i64, GPR64>;
+defm MVNww : mvn_shifts<"MVNww", 0b0, i32, GPR32>;
+
+def MOVxx :InstAlias<"mov $Rd, $Rm", (ORRxxx_lsl GPR64:$Rd, XZR, GPR64:$Rm, 0)>;
+def MOVww :InstAlias<"mov $Rd, $Rm", (ORRwww_lsl GPR32:$Rd, WZR, GPR32:$Rm, 0)>;
+
+//===----------------------------------------------------------------------===//
+// Move wide (immediate) instructions
+//===----------------------------------------------------------------------===//
+// Contains: MOVN, MOVZ, MOVK + MOV aliases
+
+// A wide variety of different relocations are needed for variants of these
+// instructions, so it turns out that we need a different operand for all of
+// them.
+multiclass movw_operands<string prefix, string instname, int width> {
+  def _imm_asmoperand : AsmOperandClass {
+    let Name = instname # width # "Shifted" # shift;
+    let PredicateMethod = "is" # instname # width # "Imm";
+    let RenderMethod = "addMoveWideImmOperands";
+    let ParserMethod = "ParseImmWithLSLOperand";
+    let DiagnosticType = "MOVWUImm16";
+  }
+
+  def _imm : Operand<i32> {
+    let ParserMatchClass = !cast<AsmOperandClass>(prefix # "_imm_asmoperand");
+    let PrintMethod = "printMoveWideImmOperand";
+    let EncoderMethod = "getMoveWideImmOpValue";
+    let DecoderMethod = "DecodeMoveWideImmOperand<" # width # ">";
+
+    let MIOperandInfo = (ops uimm16:$UImm16, imm:$Shift);
+  }
+}
+
+defm movn32 : movw_operands<"movn32", "MOVN", 32>;
+defm movn64 : movw_operands<"movn64", "MOVN", 64>;
+defm movz32 : movw_operands<"movz32", "MOVZ", 32>;
+defm movz64 : movw_operands<"movz64", "MOVZ", 64>;
+defm movk32 : movw_operands<"movk32", "MOVK", 32>;
+defm movk64 : movw_operands<"movk64", "MOVK", 64>;
+
+multiclass A64I_movwSizes<bits<2> opc, string asmop, dag ins32bit,
+                          dag ins64bit> {
+
+  def wii : A64I_movw<0b0, opc, (outs GPR32:$Rd), ins32bit,
+                      !strconcat(asmop, "\t$Rd, $FullImm"),
+                      [], NoItinerary> {
+    bits<18> FullImm;
+    let UImm16 = FullImm{15-0};
+    let Shift = FullImm{17-16};
+  }
+
+  def xii : A64I_movw<0b1, opc, (outs GPR64:$Rd), ins64bit,
+                      !strconcat(asmop, "\t$Rd, $FullImm"),
+                      [], NoItinerary> {
+    bits<18> FullImm;
+    let UImm16 = FullImm{15-0};
+    let Shift = FullImm{17-16};
+  }
+}
+
+let isMoveImm = 1, isReMaterializable = 1,
+    isAsCheapAsAMove = 1, hasSideEffects = 0 in {
+  defm MOVN : A64I_movwSizes<0b00, "movn",
+                             (ins movn32_imm:$FullImm),
+                             (ins movn64_imm:$FullImm)>;
+
+  // Some relocations are able to convert between a MOVZ and a MOVN. If these
+  // are applied the instruction must be emitted with the corresponding bits as
+  // 0, which means a MOVZ needs to override that bit from the default.
+  let PostEncoderMethod = "fixMOVZ" in
+  defm MOVZ : A64I_movwSizes<0b10, "movz",
+                             (ins movz32_imm:$FullImm),
+                             (ins movz64_imm:$FullImm)>;
+}
+
+let Constraints = "$src = $Rd" in
+defm MOVK : A64I_movwSizes<0b11, "movk",
+                           (ins GPR32:$src, movk32_imm:$FullImm),
+                           (ins GPR64:$src, movk64_imm:$FullImm)>;
+
+
+// And now the "MOV" aliases. These also need their own operands because what
+// they accept is completely different to what the base instructions accept.
+multiclass movalias_operand<string prefix, string basename,
+                            string immpredicate, int width> {
+  def _asmoperand : AsmOperandClass {
+    let Name = basename # width # "MovAlias";
+    let PredicateMethod
+          = "isMoveWideMovAlias<" # width # ", A64Imms::" # immpredicate # ">";
+    let RenderMethod
+      = "addMoveWideMovAliasOperands<" # width # ", "
+                                       # "A64Imms::" # immpredicate # ">";
+  }
+
+  def _movimm : Operand<i32> {
+    let ParserMatchClass = !cast<AsmOperandClass>(prefix # "_asmoperand");
+
+    let MIOperandInfo = (ops uimm16:$UImm16, imm:$Shift);
+  }
+}
+
+defm movz32 : movalias_operand<"movz32", "MOVZ", "isMOVZImm", 32>;
+defm movz64 : movalias_operand<"movz64", "MOVZ", "isMOVZImm", 64>;
+defm movn32 : movalias_operand<"movn32", "MOVN", "isOnlyMOVNImm", 32>;
+defm movn64 : movalias_operand<"movn64", "MOVN", "isOnlyMOVNImm", 64>;
+
+// FIXME: these are officially canonical aliases, but TableGen is too limited to
+// print them at the moment. I believe in this case an "AliasPredicate" method
+// will need to be implemented. to allow it, as well as the more generally
+// useful handling of non-register, non-constant operands.
+class movalias<Instruction INST, RegisterClass GPR, Operand operand>
+  : InstAlias<"mov $Rd, $FullImm", (INST GPR:$Rd, operand:$FullImm)>;
+
+def : movalias<MOVZwii, GPR32, movz32_movimm>;
+def : movalias<MOVZxii, GPR64, movz64_movimm>;
+def : movalias<MOVNwii, GPR32, movn32_movimm>;
+def : movalias<MOVNxii, GPR64, movn64_movimm>;
+
+//===----------------------------------------------------------------------===//
+// PC-relative addressing instructions
+//===----------------------------------------------------------------------===//
+// Contains: ADR, ADRP
+
+def adr_label : Operand<i64> {
+  let EncoderMethod = "getLabelOpValue<AArch64::fixup_a64_adr_prel>";
+
+  // This label is a 21-bit offset from PC, unscaled
+  let PrintMethod = "printLabelOperand<21, 1>";
+  let ParserMatchClass = label_asmoperand<21, 1>;
+  let OperandType = "OPERAND_PCREL";
+}
+
+def adrp_label_asmoperand : AsmOperandClass {
+  let Name = "AdrpLabel";
+  let RenderMethod = "addLabelOperands<21, 4096>";
+  let DiagnosticType = "Label";
+}
+
+def adrp_label : Operand<i64> {
+  let EncoderMethod = "getAdrpLabelOpValue";
+
+  // This label is a 21-bit offset from PC, scaled by the page-size: 4096.
+  let PrintMethod = "printLabelOperand<21, 4096>";
+  let ParserMatchClass = adrp_label_asmoperand;
+  let OperandType = "OPERAND_PCREL";
+}
+
+let hasSideEffects = 0 in {
+  def ADRxi : A64I_PCADR<0b0, (outs GPR64:$Rd), (ins adr_label:$Label),
+                         "adr\t$Rd, $Label", [], NoItinerary>;
+
+  def ADRPxi : A64I_PCADR<0b1, (outs GPR64:$Rd), (ins adrp_label:$Label),
+                          "adrp\t$Rd, $Label", [], NoItinerary>;
+}
+
+//===----------------------------------------------------------------------===//
+// System instructions
+//===----------------------------------------------------------------------===//
+// Contains: HINT, CLREX, DSB, DMB, ISB, MSR, SYS, SYSL, MRS
+//    + aliases IC, DC, AT, TLBI, NOP, YIELD, WFE, WFI, SEV, SEVL
+
+// Op1 and Op2 fields are sometimes simple 3-bit unsigned immediate values.
+def uimm3_asmoperand : AsmOperandClass {
+  let Name = "UImm3";
+  let PredicateMethod = "isUImm<3>";
+  let RenderMethod = "addImmOperands";
+  let DiagnosticType = "UImm3";
+}
+
+def uimm3 : Operand<i32> {
+  let ParserMatchClass = uimm3_asmoperand;
+}
+
+// The HINT alias can accept a simple unsigned 7-bit immediate.
+def uimm7_asmoperand : AsmOperandClass {
+  let Name = "UImm7";
+  let PredicateMethod = "isUImm<7>";
+  let RenderMethod = "addImmOperands";
+  let DiagnosticType = "UImm7";
+}
+
+def uimm7 : Operand<i32> {
+  let ParserMatchClass = uimm7_asmoperand;
+}
+
+// Multiclass namedimm is defined with the prefetch operands. Most of these fit
+// into the NamedImmMapper scheme well: they either accept a named operand or
+// any immediate under a particular value (which may be 0, implying no immediate
+// is allowed).
+defm dbarrier : namedimm<"dbarrier", "A64DB::DBarrierMapper">;
+defm isb : namedimm<"isb", "A64ISB::ISBMapper">;
+defm ic : namedimm<"ic", "A64IC::ICMapper">;
+defm dc : namedimm<"dc", "A64DC::DCMapper">;
+defm at : namedimm<"at", "A64AT::ATMapper">;
+defm tlbi : namedimm<"tlbi", "A64TLBI::TLBIMapper">;
+
+// However, MRS and MSR are more complicated for a few reasons:
+//   * There are ~1000 generic names S3_<op1>_<CRn>_<CRm>_<Op2> which have an
+//     implementation-defined effect
+//   * Most registers are shared, but some are read-only or write-only.
+//   * There is a variant of MSR which accepts the same register name (SPSel),
+//     but which would have a different encoding.
+
+// In principle these could be resolved in with more complicated subclasses of
+// NamedImmMapper, however that imposes an overhead on other "named
+// immediates". Both in concrete terms with virtual tables and in unnecessary
+// abstraction.
+
+// The solution adopted here is to take the MRS/MSR Mappers out of the usual
+// hierarchy (they're not derived from NamedImmMapper) and to add logic for
+// their special situation.
+def mrs_asmoperand : AsmOperandClass {
+  let Name = "MRS";
+  let ParserMethod = "ParseSysRegOperand";
+  let DiagnosticType = "MRS";
+}
+
+def mrs_op : Operand<i32> {
+  let ParserMatchClass = mrs_asmoperand;
+  let PrintMethod = "printMRSOperand";
+  let DecoderMethod = "DecodeMRSOperand";
+}
+
+def msr_asmoperand : AsmOperandClass {
+  let Name = "MSRWithReg";
+
+  // Note that SPSel is valid for both this and the pstate operands, but with
+  // different immediate encodings. This is why these operands provide a string
+  // AArch64Operand rather than an immediate. The overlap is small enough that
+  // it could be resolved with hackery now, but who can say in future?
+  let ParserMethod = "ParseSysRegOperand";
+  let DiagnosticType = "MSR";
+}
+
+def msr_op : Operand<i32> {
+  let ParserMatchClass = msr_asmoperand;
+  let PrintMethod = "printMSROperand";
+  let DecoderMethod = "DecodeMSROperand";
+}
+
+def pstate_asmoperand : AsmOperandClass {
+  let Name = "MSRPState";
+  // See comment above about parser.
+  let ParserMethod = "ParseSysRegOperand";
+  let DiagnosticType = "MSR";
+}
+
+def pstate_op : Operand<i32> {
+  let ParserMatchClass = pstate_asmoperand;
+  let PrintMethod = "printNamedImmOperand<A64PState::PStateMapper>";
+  let DecoderMethod = "DecodeNamedImmOperand<A64PState::PStateMapper>";
+}
+
+// When <CRn> is specified, an assembler should accept something like "C4", not
+// the usual "#4" immediate.
+def CRx_asmoperand : AsmOperandClass {
+  let Name = "CRx";
+  let PredicateMethod = "isUImm<4>";
+  let RenderMethod = "addImmOperands";
+  let ParserMethod = "ParseCRxOperand";
+  // Diagnostics are handled in all cases by ParseCRxOperand.
+}
+
+def CRx : Operand<i32> {
+  let ParserMatchClass = CRx_asmoperand;
+  let PrintMethod = "printCRxOperand";
+}
+
+
+// Finally, we can start defining the instructions.
+
+// HINT is straightforward, with a few aliases.
+def HINTi : A64I_system<0b0, (outs), (ins uimm7:$UImm7), "hint\t$UImm7",
+                        [], NoItinerary> {
+  bits<7> UImm7;
+  let CRm = UImm7{6-3};
+  let Op2 = UImm7{2-0};
+
+  let Op0 = 0b00;
+  let Op1 = 0b011;
+  let CRn = 0b0010;
+  let Rt = 0b11111;
+}
+
+def : InstAlias<"nop", (HINTi 0)>;
+def : InstAlias<"yield", (HINTi 1)>;
+def : InstAlias<"wfe", (HINTi 2)>;
+def : InstAlias<"wfi", (HINTi 3)>;
+def : InstAlias<"sev", (HINTi 4)>;
+def : InstAlias<"sevl", (HINTi 5)>;
+
+// Quite a few instructions then follow a similar pattern of fixing common
+// fields in the bitpattern, we'll define a helper-class for them.
+class simple_sys<bits<2> op0, bits<3> op1, bits<4> crn, bits<3> op2,
+                 Operand operand, string asmop>
+  : A64I_system<0b0, (outs), (ins operand:$CRm), !strconcat(asmop, "\t$CRm"),
+                [], NoItinerary> {
+  let Op0 = op0;
+  let Op1 = op1;
+  let CRn = crn;
+  let Op2 = op2;
+  let Rt = 0b11111;
+}
+
+
+def CLREXi : simple_sys<0b00, 0b011, 0b0011, 0b010, uimm4, "clrex">;
+def DSBi : simple_sys<0b00, 0b011, 0b0011, 0b100, dbarrier_op, "dsb">;
+def DMBi : simple_sys<0b00, 0b011, 0b0011, 0b101, dbarrier_op, "dmb">;
+def ISBi : simple_sys<0b00, 0b011, 0b0011, 0b110, isb_op, "isb">;
+
+def : InstAlias<"clrex", (CLREXi 0b1111)>;
+def : InstAlias<"isb", (ISBi 0b1111)>;
+
+// (DMBi 0xb) is a "DMB ISH" instruciton, appropriate for Linux SMP
+// configurations at least.
+def : Pat<(atomic_fence imm, imm), (DMBi 0xb)>;
+
+// Any SYS bitpattern can be represented with a complex and opaque "SYS"
+// instruction.
+def SYSiccix : A64I_system<0b0, (outs),
+                           (ins uimm3:$Op1, CRx:$CRn, CRx:$CRm,
+                                uimm3:$Op2, GPR64:$Rt),
+                           "sys\t$Op1, $CRn, $CRm, $Op2, $Rt",
+                           [], NoItinerary> {
+  let Op0 = 0b01;
+}
+
+// You can skip the Xt argument whether it makes sense or not for the generic
+// SYS instruction.
+def : InstAlias<"sys $Op1, $CRn, $CRm, $Op2",
+                (SYSiccix uimm3:$Op1, CRx:$CRn, CRx:$CRm, uimm3:$Op2, XZR)>;
+
+
+// But many have aliases, which obviously don't fit into
+class SYSalias<dag ins, string asmstring>
+  : A64I_system<0b0, (outs), ins, asmstring, [], NoItinerary> {
+  let isAsmParserOnly = 1;
+
+  bits<14> SysOp;
+  let Op0 = 0b01;
+  let Op1 = SysOp{13-11};
+  let CRn = SysOp{10-7};
+  let CRm = SysOp{6-3};
+  let Op2 = SysOp{2-0};
+}
+
+def ICix : SYSalias<(ins ic_op:$SysOp, GPR64:$Rt), "ic\t$SysOp, $Rt">;
+
+def ICi : SYSalias<(ins ic_op:$SysOp), "ic\t$SysOp"> {
+  let Rt = 0b11111;
+}
+
+def DCix : SYSalias<(ins dc_op:$SysOp, GPR64:$Rt), "dc\t$SysOp, $Rt">;
+def ATix : SYSalias<(ins at_op:$SysOp, GPR64:$Rt), "at\t$SysOp, $Rt">;
+
+def TLBIix : SYSalias<(ins tlbi_op:$SysOp, GPR64:$Rt), "tlbi\t$SysOp, $Rt">;
+
+def TLBIi : SYSalias<(ins tlbi_op:$SysOp), "tlbi\t$SysOp"> {
+  let Rt = 0b11111;
+}
+
+
+def SYSLxicci : A64I_system<0b1, (outs GPR64:$Rt),
+                            (ins uimm3:$Op1, CRx:$CRn, CRx:$CRm, uimm3:$Op2),
+                            "sysl\t$Rt, $Op1, $CRn, $CRm, $Op2",
+                            [], NoItinerary> {
+  let Op0 = 0b01;
+}
+
+// The instructions themselves are rather simple for MSR and MRS.
+def MSRix : A64I_system<0b0, (outs), (ins msr_op:$SysReg, GPR64:$Rt),
+                        "msr\t$SysReg, $Rt", [], NoItinerary> {
+  bits<16> SysReg;
+  let Op0 = SysReg{15-14};
+  let Op1 = SysReg{13-11};
+  let CRn = SysReg{10-7};
+  let CRm = SysReg{6-3};
+  let Op2 = SysReg{2-0};
+}
+
+def MRSxi : A64I_system<0b1, (outs GPR64:$Rt), (ins mrs_op:$SysReg),
+                        "mrs\t$Rt, $SysReg", [], NoItinerary> {
+  bits<16> SysReg;
+  let Op0 = SysReg{15-14};
+  let Op1 = SysReg{13-11};
+  let CRn = SysReg{10-7};
+  let CRm = SysReg{6-3};
+  let Op2 = SysReg{2-0};
+}
+
+def MSRii : A64I_system<0b0, (outs), (ins pstate_op:$PState, uimm4:$CRm),
+                        "msr\t$PState, $CRm", [], NoItinerary> {
+  bits<6> PState;
+
+  let Op0 = 0b00;
+  let Op1 = PState{5-3};
+  let CRn = 0b0100;
+  let Op2 = PState{2-0};
+  let Rt = 0b11111;
+}
+
+//===----------------------------------------------------------------------===//
+// Test & branch (immediate) instructions
+//===----------------------------------------------------------------------===//
+// Contains: TBZ, TBNZ
+
+// The bit to test is a simple unsigned 6-bit immediate in the X-register
+// versions.
+def uimm6 : Operand<i64> {
+  let ParserMatchClass = uimm6_asmoperand;
+}
+
+def label_wid14_scal4_asmoperand : label_asmoperand<14, 4>;
+
+def tbimm_target : Operand<OtherVT> {
+  let EncoderMethod = "getLabelOpValue<AArch64::fixup_a64_tstbr>";
+
+  // This label is a 14-bit offset from PC, scaled by the instruction-width: 4.
+  let PrintMethod = "printLabelOperand<14, 4>";
+  let ParserMatchClass = label_wid14_scal4_asmoperand;
+
+  let OperandType = "OPERAND_PCREL";
+}
+
+def A64eq : ImmLeaf<i32, [{ return Imm == A64CC::EQ; }]>;
+def A64ne : ImmLeaf<i32, [{ return Imm == A64CC::NE; }]>;
+
+// These instructions correspond to patterns involving "and" with a power of
+// two, which we need to be able to select.
+def tstb64_pat : ComplexPattern<i64, 1, "SelectTSTBOperand<64>">;
+def tstb32_pat : ComplexPattern<i32, 1, "SelectTSTBOperand<32>">;
+
+let isBranch = 1, isTerminator = 1 in {
+  def TBZxii : A64I_TBimm<0b0, (outs),
+                        (ins GPR64:$Rt, uimm6:$Imm, tbimm_target:$Label),
+                        "tbz\t$Rt, $Imm, $Label",
+                        [(A64br_cc (A64cmp (and i64:$Rt, tstb64_pat:$Imm), 0),
+                                   A64eq, bb:$Label)],
+                        NoItinerary>;
+
+  def TBNZxii : A64I_TBimm<0b1, (outs),
+                        (ins GPR64:$Rt, uimm6:$Imm, tbimm_target:$Label),
+                        "tbnz\t$Rt, $Imm, $Label",
+                        [(A64br_cc (A64cmp (and i64:$Rt, tstb64_pat:$Imm), 0),
+                                   A64ne, bb:$Label)],
+                        NoItinerary>;
+
+
+  // Note, these instructions overlap with the above 64-bit patterns. This is
+  // intentional, "tbz x3, #1, somewhere" and "tbz w3, #1, somewhere" would both
+  // do the same thing and are both permitted assembly. They also both have
+  // sensible DAG patterns.
+  def TBZwii : A64I_TBimm<0b0, (outs),
+                        (ins GPR32:$Rt, uimm5:$Imm, tbimm_target:$Label),
+                        "tbz\t$Rt, $Imm, $Label",
+                        [(A64br_cc (A64cmp (and i32:$Rt, tstb32_pat:$Imm), 0),
+                                   A64eq, bb:$Label)],
+                        NoItinerary> {
+    let Imm{5} = 0b0;
+  }
+
+  def TBNZwii : A64I_TBimm<0b1, (outs),
+                        (ins GPR32:$Rt, uimm5:$Imm, tbimm_target:$Label),
+                        "tbnz\t$Rt, $Imm, $Label",
+                        [(A64br_cc (A64cmp (and i32:$Rt, tstb32_pat:$Imm), 0),
+                                   A64ne, bb:$Label)],
+                        NoItinerary> {
+    let Imm{5} = 0b0;
+  }
+}
+
+//===----------------------------------------------------------------------===//
+// Unconditional branch (immediate) instructions
+//===----------------------------------------------------------------------===//
+// Contains: B, BL
+
+def label_wid26_scal4_asmoperand : label_asmoperand<26, 4>;
+
+def bimm_target : Operand<OtherVT> {
+  let EncoderMethod = "getLabelOpValue<AArch64::fixup_a64_uncondbr>";
+
+  // This label is a 26-bit offset from PC, scaled by the instruction-width: 4.
+  let PrintMethod = "printLabelOperand<26, 4>";
+  let ParserMatchClass = label_wid26_scal4_asmoperand;
+
+  let OperandType = "OPERAND_PCREL";
+}
+
+def blimm_target : Operand<i64> {
+  let EncoderMethod = "getLabelOpValue<AArch64::fixup_a64_call>";
+
+  // This label is a 26-bit offset from PC, scaled by the instruction-width: 4.
+  let PrintMethod = "printLabelOperand<26, 4>";
+  let ParserMatchClass = label_wid26_scal4_asmoperand;
+
+  let OperandType = "OPERAND_PCREL";
+}
+
+class A64I_BimmImpl<bit op, string asmop, list<dag> patterns, Operand lbl_type>
+  : A64I_Bimm<op, (outs), (ins lbl_type:$Label),
+              !strconcat(asmop, "\t$Label"), patterns,
+              NoItinerary>;
+
+let isBranch = 1 in {
+  def Bimm : A64I_BimmImpl<0b0, "b", [(br bb:$Label)], bimm_target> {
+    let isTerminator = 1;
+    let isBarrier = 1;
+  }
+
+  def BLimm : A64I_BimmImpl<0b1, "bl",
+                            [(AArch64Call tglobaladdr:$Label)], blimm_target> {
+    let isCall = 1;
+    let Defs = [X30];
+  }
+}
+
+def : Pat<(AArch64Call texternalsym:$Label), (BLimm texternalsym:$Label)>;
+
+//===----------------------------------------------------------------------===//
+// Unconditional branch (register) instructions
+//===----------------------------------------------------------------------===//
+// Contains: BR, BLR, RET, ERET, DRP.
+
+// Most of the notional opcode fields in the A64I_Breg format are fixed in A64
+// at the moment.
+class A64I_BregImpl<bits<4> opc,
+                    dag outs, dag ins, string asmstr, list<dag> patterns,
+                    InstrItinClass itin = NoItinerary>
+  : A64I_Breg<opc, 0b11111, 0b000000, 0b00000,
+              outs, ins, asmstr, patterns, itin> {
+  let isBranch         = 1;
+  let isIndirectBranch = 1;
+}
+
+// Note that these are not marked isCall or isReturn because as far as LLVM is
+// concerned they're not. "ret" is just another jump unless it has been selected
+// by LLVM as the function's return.
+
+let isBranch = 1 in {
+  def BRx : A64I_BregImpl<0b0000,(outs), (ins GPR64:$Rn),
+                          "br\t$Rn", [(brind i64:$Rn)]> {
+    let isBarrier = 1;
+    let isTerminator = 1;
+  }
+
+  def BLRx : A64I_BregImpl<0b0001, (outs), (ins GPR64:$Rn),
+                           "blr\t$Rn", [(AArch64Call i64:$Rn)]> {
+    let isBarrier = 0;
+    let isCall = 1;
+    let Defs = [X30];
+  }
+
+  def RETx : A64I_BregImpl<0b0010, (outs), (ins GPR64:$Rn),
+                           "ret\t$Rn", []> {
+    let isBarrier = 1;
+    let isTerminator = 1;
+    let isReturn = 1;
+  }
+
+  // Create a separate pseudo-instruction for codegen to use so that we don't
+  // flag x30 as used in every function. It'll be restored before the RET by the
+  // epilogue if it's legitimately used.
+  def RET : A64PseudoExpand<(outs), (ins), [(A64ret)], (RETx (ops X30))> {
+    let isTerminator = 1;
+    let isBarrier = 1;
+    let isReturn = 1;
+  }
+
+  def ERET : A64I_BregImpl<0b0100, (outs), (ins), "eret", []> {
+    let Rn = 0b11111;
+    let isBarrier = 1;
+    let isTerminator = 1;
+    let isReturn = 1;
+  }
+
+  def DRPS : A64I_BregImpl<0b0101, (outs), (ins), "drps", []> {
+    let Rn = 0b11111;
+    let isBarrier = 1;
+  }
+}
+
+def RETAlias : InstAlias<"ret", (RETx X30)>;
+
+
+//===----------------------------------------------------------------------===//
+// Address generation patterns
+//===----------------------------------------------------------------------===//
+
+// Primary method of address generation for the small/absolute memory model is
+// an ADRP/ADR pair:
+//     ADRP x0, some_variable
+//     ADD x0, x0, #:lo12:some_variable
+//
+// The load/store elision of the ADD is accomplished when selecting
+// addressing-modes. This just mops up the cases where that doesn't work and we
+// really need an address in some register.
+
+// This wrapper applies a LO12 modifier to the address. Otherwise we could just
+// use the same address.
+
+class ADRP_ADD<SDNode Wrapper, SDNode addrop>
+ : Pat<(Wrapper addrop:$Hi, addrop:$Lo12, (i32 imm)),
+       (ADDxxi_lsl0_s (ADRPxi addrop:$Hi), addrop:$Lo12)>;
+
+def : ADRP_ADD<A64WrapperSmall, tblockaddress>;
+def : ADRP_ADD<A64WrapperSmall, texternalsym>;
+def : ADRP_ADD<A64WrapperSmall, tglobaladdr>;
+def : ADRP_ADD<A64WrapperSmall, tglobaltlsaddr>;
+def : ADRP_ADD<A64WrapperSmall, tjumptable>;
+
+//===----------------------------------------------------------------------===//
+// GOT access patterns
+//===----------------------------------------------------------------------===//
+
+// FIXME: Wibble
+
+class GOTLoadSmall<SDNode addrfrag>
+  : Pat<(A64GOTLoad (A64WrapperSmall addrfrag:$Hi, addrfrag:$Lo12, 8)),
+        (LS64_LDR (ADRPxi addrfrag:$Hi), addrfrag:$Lo12)>;
+
+def : GOTLoadSmall<texternalsym>;
+def : GOTLoadSmall<tglobaladdr>;
+def : GOTLoadSmall<tglobaltlsaddr>;
+
+//===----------------------------------------------------------------------===//
+// Tail call handling
+//===----------------------------------------------------------------------===//
+
+let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [XSP] in {
+  def TC_RETURNdi
+    : PseudoInst<(outs), (ins i64imm:$dst, i32imm:$FPDiff),
+                 [(AArch64tcret tglobaladdr:$dst, (i32 timm:$FPDiff))]>;
+
+  def TC_RETURNxi
+    : PseudoInst<(outs), (ins tcGPR64:$dst, i32imm:$FPDiff),
+                 [(AArch64tcret i64:$dst, (i32 timm:$FPDiff))]>;
+}
+
+let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1,
+    Uses = [XSP] in {
+  def TAIL_Bimm : A64PseudoExpand<(outs), (ins bimm_target:$Label), [],
+                                  (Bimm bimm_target:$Label)>;
+
+  def TAIL_BRx : A64PseudoExpand<(outs), (ins tcGPR64:$Rd), [],
+                                 (BRx GPR64:$Rd)>;
+}
+
+
+def : Pat<(AArch64tcret texternalsym:$dst, (i32 timm:$FPDiff)),
+          (TC_RETURNdi texternalsym:$dst, imm:$FPDiff)>;
+
+//===----------------------------------------------------------------------===//
+// Thread local storage
+//===----------------------------------------------------------------------===//
+
+// This is a pseudo-instruction representing the ".tlsdesccall" directive in
+// assembly. Its effect is to insert an R_AARCH64_TLSDESC_CALL relocation at the
+// current location. It should always be immediately followed by a BLR
+// instruction, and is intended solely for relaxation by the linker.
+
+def : Pat<(A64threadpointer), (MRSxi 0xde82)>;
+
+def TLSDESCCALL : PseudoInst<(outs), (ins i64imm:$Lbl), []> {
+  let hasSideEffects = 1;
+}
+
+def TLSDESC_BLRx : PseudoInst<(outs), (ins GPR64:$Rn, i64imm:$Var),
+                            [(A64tlsdesc_blr i64:$Rn, tglobaltlsaddr:$Var)]> {
+  let isCall = 1;
+  let Defs = [X30];
+}
+
+def : Pat<(A64tlsdesc_blr i64:$Rn, texternalsym:$Var),
+          (TLSDESC_BLRx $Rn, texternalsym:$Var)>;
+
+//===----------------------------------------------------------------------===//
+// Bitfield patterns
+//===----------------------------------------------------------------------===//
+
+def bfi32_lsb_to_immr : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant((32 - N->getZExtValue()) % 32, MVT::i64);
+}]>;
+
+def bfi64_lsb_to_immr : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant((64 - N->getZExtValue()) % 64, MVT::i64);
+}]>;
+
+def bfi_width_to_imms : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(N->getZExtValue() - 1, MVT::i64);
+}]>;
+
+
+// The simpler patterns deal with cases where no AND mask is actually needed
+// (either all bits are used or the low 32 bits are used).
+let AddedComplexity = 10 in {
+
+def : Pat<(A64Bfi i64:$src, i64:$Rn, imm:$ImmR, imm:$ImmS),
+           (BFIxxii $src, $Rn,
+                    (bfi64_lsb_to_immr (i64 imm:$ImmR)),
+                    (bfi_width_to_imms (i64 imm:$ImmS)))>;
+
+def : Pat<(A64Bfi i32:$src, i32:$Rn, imm:$ImmR, imm:$ImmS),
+          (BFIwwii $src, $Rn,
+                   (bfi32_lsb_to_immr (i64 imm:$ImmR)),
+                   (bfi_width_to_imms (i64 imm:$ImmS)))>;
+
+
+def : Pat<(and (A64Bfi i64:$src, i64:$Rn, imm:$ImmR, imm:$ImmS),
+               (i64 4294967295)),
+          (SUBREG_TO_REG (i64 0),
+                         (BFIwwii (EXTRACT_SUBREG $src, sub_32),
+                                  (EXTRACT_SUBREG $Rn, sub_32),
+                                  (bfi32_lsb_to_immr (i64 imm:$ImmR)),
+                                  (bfi_width_to_imms (i64 imm:$ImmS))),
+                         sub_32)>;
+
+}
+
+//===----------------------------------------------------------------------===//
+// Miscellaneous patterns
+//===----------------------------------------------------------------------===//
+
+// Truncation from 64 to 32-bits just involves renaming your register.
+def : Pat<(i32 (trunc i64:$val)), (EXTRACT_SUBREG $val, sub_32)>;
+
+// Similarly, extension where we don't care about the high bits is
+// just a rename.
+def : Pat<(i64 (anyext i32:$val)),
+          (INSERT_SUBREG (IMPLICIT_DEF), $val, sub_32)>;
+
+// SELECT instructions providing f128 types need to be handled by a
+// pseudo-instruction since the eventual code will need to introduce basic
+// blocks and control flow.
+def F128CSEL : PseudoInst<(outs FPR128:$Rd),
+                         (ins FPR128:$Rn, FPR128:$Rm, cond_code_op:$Cond),
+                         [(set f128:$Rd, (simple_select f128:$Rn, f128:$Rm))]> {
+  let Uses = [NZCV];
+  let usesCustomInserter = 1;
+}
+
+//===----------------------------------------------------------------------===//
+// Load/store patterns
+//===----------------------------------------------------------------------===//
+
+// There are lots of patterns here, because we need to allow at least three
+// parameters to vary independently.
+//   1. Instruction: "ldrb w9, [sp]", "ldrh w9, [sp]", ...
+//   2. LLVM source: zextloadi8, anyextloadi8, ...
+//   3. Address-generation: A64Wrapper, (add BASE, OFFSET), ...
+//
+// The biggest problem turns out to be the address-generation variable. At the
+// point of instantiation we need to produce two DAGs, one for the pattern and
+// one for the instruction. Doing this at the lowest level of classes doesn't
+// work.
+//
+// Consider the simple uimm12 addressing mode, and the desire to match both (add
+// GPR64xsp:$Rn, uimm12:$Offset) and GPR64xsp:$Rn, particularly on the
+// instruction side. We'd need to insert either "GPR64xsp" and "uimm12" or
+// "GPR64xsp" and "0" into an unknown dag. !subst is not capable of this
+// operation, and PatFrags are for selection not output.
+//
+// As a result, the address-generation patterns are the final
+// instantiations. However, we do still need to vary the operand for the address
+// further down (At the point we're deciding A64WrapperSmall, we don't know
+// the memory width of the operation).
+
+//===------------------------------
+// 1. Basic infrastructural defs
+//===------------------------------
+
+// First, some simple classes for !foreach and !subst to use:
+class Decls {
+  dag pattern;
+}
+
+def decls : Decls;
+def ALIGN;
+def INST;
+def OFFSET;
+def SHIFT;
+
+// You can't use !subst on an actual immediate, but you *can* use it on an
+// operand record that happens to match a single immediate. So we do.
+def imm_eq0 : ImmLeaf<i64, [{ return Imm == 0; }]>;
+def imm_eq1 : ImmLeaf<i64, [{ return Imm == 1; }]>;
+def imm_eq2 : ImmLeaf<i64, [{ return Imm == 2; }]>;
+def imm_eq3 : ImmLeaf<i64, [{ return Imm == 3; }]>;
+def imm_eq4 : ImmLeaf<i64, [{ return Imm == 4; }]>;
+
+// If the low bits of a pointer are known to be 0 then an "or" is just as good
+// as addition for computing an offset. This fragment forwards that check for
+// TableGen's use.
+def add_like_or : PatFrag<(ops node:$lhs, node:$rhs), (or node:$lhs, node:$rhs),
+[{
+  return CurDAG->isBaseWithConstantOffset(SDValue(N, 0));
+}]>;
+
+// Load/store (unsigned immediate) operations with relocations against global
+// symbols (for lo12) are only valid if those symbols have correct alignment
+// (since the immediate offset is divided by the access scale, it can't have a
+// remainder).
+//
+// The guaranteed alignment is provided as part of the WrapperSmall
+// operation, and checked against one of these.
+def any_align   : ImmLeaf<i32, [{ (void)Imm; return true; }]>;
+def min_align2  : ImmLeaf<i32, [{ return Imm >= 2; }]>;
+def min_align4  : ImmLeaf<i32, [{ return Imm >= 4; }]>;
+def min_align8  : ImmLeaf<i32, [{ return Imm >= 8; }]>;
+def min_align16 : ImmLeaf<i32, [{ return Imm >= 16; }]>;
+
+// "Normal" load/store instructions can be used on atomic operations, provided
+// the ordering parameter is at most "monotonic". Anything above that needs
+// special handling with acquire/release instructions.
+class simple_load<PatFrag base>
+  : PatFrag<(ops node:$ptr), (base node:$ptr), [{
+  return cast<AtomicSDNode>(N)->getOrdering() <= Monotonic;
+}]>;
+
+def atomic_load_simple_i8  : simple_load<atomic_load_8>;
+def atomic_load_simple_i16 : simple_load<atomic_load_16>;
+def atomic_load_simple_i32 : simple_load<atomic_load_32>;
+def atomic_load_simple_i64 : simple_load<atomic_load_64>;
+
+class simple_store<PatFrag base>
+  : PatFrag<(ops node:$ptr, node:$val), (base node:$ptr, node:$val), [{
+  return cast<AtomicSDNode>(N)->getOrdering() <= Monotonic;
+}]>;
+
+def atomic_store_simple_i8  : simple_store<atomic_store_8>;
+def atomic_store_simple_i16 : simple_store<atomic_store_16>;
+def atomic_store_simple_i32 : simple_store<atomic_store_32>;
+def atomic_store_simple_i64 : simple_store<atomic_store_64>;
+
+//===------------------------------
+// 2. UImm12 and SImm9
+//===------------------------------
+
+// These instructions have two operands providing the address so they can be
+// treated similarly for most purposes.
+
+//===------------------------------
+// 2.1 Base patterns covering extend/truncate semantics
+//===------------------------------
+
+// Atomic patterns can be shared between integer operations of all sizes, a
+// quick multiclass here allows reuse.
+multiclass ls_atomic_pats<Instruction LOAD, Instruction STORE, dag Base,
+                          dag Offset, dag address, ValueType transty,
+                          ValueType sty> {
+  def : Pat<(!cast<PatFrag>("atomic_load_simple_" # sty) address),
+            (LOAD Base, Offset)>;
+
+  def : Pat<(!cast<PatFrag>("atomic_store_simple_" # sty) address, transty:$Rt),
+            (STORE $Rt, Base, Offset)>;
+}
+
+// Instructions accessing a memory chunk smaller than a register (or, in a
+// pinch, the same size) have a characteristic set of patterns they want to
+// match: extending loads and truncating stores. This class deals with the
+// sign-neutral version of those patterns.
+//
+// It will be instantiated across multiple addressing-modes.
+multiclass ls_small_pats<Instruction LOAD, Instruction STORE,
+                         dag Base, dag Offset,
+                         dag address, ValueType sty>
+  : ls_atomic_pats<LOAD, STORE, Base, Offset, address, i32, sty> {
+  def : Pat<(!cast<SDNode>(zextload # sty) address), (LOAD Base, Offset)>;
+
+  def : Pat<(!cast<SDNode>(extload # sty) address), (LOAD Base, Offset)>;
+
+  // For zero-extension to 64-bits we have to tell LLVM that the whole 64-bit
+  // register was actually set.
+  def : Pat<(i64 (!cast<SDNode>(zextload # sty) address)),
+            (SUBREG_TO_REG (i64 0), (LOAD Base, Offset), sub_32)>;
+
+  def : Pat<(i64 (!cast<SDNode>(extload # sty) address)),
+            (SUBREG_TO_REG (i64 0), (LOAD Base, Offset), sub_32)>;
+
+  def : Pat<(!cast<SDNode>(truncstore # sty) i32:$Rt, address),
+            (STORE $Rt, Base, Offset)>;
+
+  // For truncating store from 64-bits, we have to manually tell LLVM to
+  // ignore the high bits of the x register.
+  def : Pat<(!cast<SDNode>(truncstore # sty) i64:$Rt, address),
+            (STORE (EXTRACT_SUBREG $Rt, sub_32), Base, Offset)>;
+}
+
+// Next come patterns for sign-extending loads.
+multiclass load_signed_pats<string T, string U, dag Base, dag Offset,
+                            dag address, ValueType sty> {
+  def : Pat<(i32 (!cast<SDNode>("sextload" # sty) address)),
+            (!cast<Instruction>("LDRS" # T # "w" # U) Base, Offset)>;
+
+  def : Pat<(i64 (!cast<SDNode>("sextload" # sty) address)),
+            (!cast<Instruction>("LDRS" # T # "x" # U) Base, Offset)>;
+
+}
+
+// and finally "natural-width" loads and stores come next.
+multiclass ls_neutral_pats<Instruction LOAD, Instruction STORE, dag Base,
+                           dag Offset, dag address, ValueType sty> {
+  def : Pat<(sty (load address)), (LOAD Base, Offset)>;
+  def : Pat<(store sty:$Rt, address), (STORE $Rt, Base, Offset)>;
+}
+
+// Integer operations also get atomic instructions to select for.
+multiclass ls_int_neutral_pats<Instruction LOAD, Instruction STORE, dag Base,
+                           dag Offset, dag address, ValueType sty>
+  : ls_neutral_pats<LOAD, STORE, Base, Offset, address, sty>,
+    ls_atomic_pats<LOAD, STORE, Base, Offset, address, sty, sty>;
+
+//===------------------------------
+// 2.2. Addressing-mode instantiations
+//===------------------------------
+
+multiclass uimm12_pats<dag address, dag Base, dag Offset> {
+  defm : ls_small_pats<LS8_LDR, LS8_STR, Base,
+                       !foreach(decls.pattern, Offset,
+                                !subst(OFFSET, byte_uimm12, decls.pattern)),
+                       !foreach(decls.pattern, address,
+                                !subst(OFFSET, byte_uimm12,
+                                !subst(ALIGN, any_align, decls.pattern))),
+                       i8>;
+  defm : ls_small_pats<LS16_LDR, LS16_STR, Base,
+                       !foreach(decls.pattern, Offset,
+                                !subst(OFFSET, hword_uimm12, decls.pattern)),
+                       !foreach(decls.pattern, address,
+                                !subst(OFFSET, hword_uimm12,
+                                !subst(ALIGN, min_align2, decls.pattern))),
+                       i16>;
+  defm : ls_small_pats<LS32_LDR, LS32_STR, Base,
+                       !foreach(decls.pattern, Offset,
+                                !subst(OFFSET, word_uimm12, decls.pattern)),
+                       !foreach(decls.pattern, address,
+                                !subst(OFFSET, word_uimm12,
+                                !subst(ALIGN, min_align4, decls.pattern))),
+                       i32>;
+
+  defm : ls_int_neutral_pats<LS32_LDR, LS32_STR, Base,
+                          !foreach(decls.pattern, Offset,
+                                   !subst(OFFSET, word_uimm12, decls.pattern)),
+                          !foreach(decls.pattern, address,
+                                   !subst(OFFSET, word_uimm12,
+                                   !subst(ALIGN, min_align4, decls.pattern))),
+                          i32>;
+
+  defm : ls_int_neutral_pats<LS64_LDR, LS64_STR, Base,
+                          !foreach(decls.pattern, Offset,
+                                   !subst(OFFSET, dword_uimm12, decls.pattern)),
+                          !foreach(decls.pattern, address,
+                                   !subst(OFFSET, dword_uimm12,
+                                   !subst(ALIGN, min_align8, decls.pattern))),
+                          i64>;
+
+  defm : ls_neutral_pats<LSFP16_LDR, LSFP16_STR, Base,
+                          !foreach(decls.pattern, Offset,
+                                   !subst(OFFSET, hword_uimm12, decls.pattern)),
+                          !foreach(decls.pattern, address,
+                                   !subst(OFFSET, hword_uimm12,
+                                   !subst(ALIGN, min_align2, decls.pattern))),
+                          f16>;
+
+  defm : ls_neutral_pats<LSFP32_LDR, LSFP32_STR, Base,
+                          !foreach(decls.pattern, Offset,
+                                   !subst(OFFSET, word_uimm12, decls.pattern)),
+                          !foreach(decls.pattern, address,
+                                   !subst(OFFSET, word_uimm12,
+                                   !subst(ALIGN, min_align4, decls.pattern))),
+                          f32>;
+
+  defm : ls_neutral_pats<LSFP64_LDR, LSFP64_STR, Base,
+                          !foreach(decls.pattern, Offset,
+                                   !subst(OFFSET, dword_uimm12, decls.pattern)),
+                          !foreach(decls.pattern, address,
+                                   !subst(OFFSET, dword_uimm12,
+                                   !subst(ALIGN, min_align8, decls.pattern))),
+                          f64>;
+
+  defm : ls_neutral_pats<LSFP128_LDR, LSFP128_STR, Base,
+                          !foreach(decls.pattern, Offset,
+                                   !subst(OFFSET, qword_uimm12, decls.pattern)),
+                          !foreach(decls.pattern, address,
+                                   !subst(OFFSET, qword_uimm12,
+                                   !subst(ALIGN, min_align16, decls.pattern))),
+                          f128>;
+
+  defm : load_signed_pats<"B", "", Base,
+                          !foreach(decls.pattern, Offset,
+                                   !subst(OFFSET, byte_uimm12, decls.pattern)),
+                          !foreach(decls.pattern, address,
+                                   !subst(OFFSET, byte_uimm12,
+                                   !subst(ALIGN, any_align, decls.pattern))),
+                          i8>;
+
+  defm : load_signed_pats<"H", "", Base,
+                          !foreach(decls.pattern, Offset,
+                                   !subst(OFFSET, hword_uimm12, decls.pattern)),
+                          !foreach(decls.pattern, address,
+                                   !subst(OFFSET, hword_uimm12,
+                                   !subst(ALIGN, min_align2, decls.pattern))),
+                          i16>;
+
+  def : Pat<(sextloadi32 !foreach(decls.pattern, address,
+                                  !subst(OFFSET, word_uimm12,
+                                  !subst(ALIGN, min_align4, decls.pattern)))),
+            (LDRSWx Base, !foreach(decls.pattern, Offset,
+                                  !subst(OFFSET, word_uimm12, decls.pattern)))>;
+}
+
+// Straightforward patterns of last resort: a pointer with or without an
+// appropriate offset.
+defm : uimm12_pats<(i64 i64:$Rn), (i64 i64:$Rn), (i64 0)>;
+defm : uimm12_pats<(add i64:$Rn, OFFSET:$UImm12),
+                   (i64 i64:$Rn), (i64 OFFSET:$UImm12)>;
+
+// The offset could be hidden behind an "or", of course:
+defm : uimm12_pats<(add_like_or i64:$Rn, OFFSET:$UImm12),
+                   (i64 i64:$Rn), (i64 OFFSET:$UImm12)>;
+
+// Global addresses under the small-absolute model should use these
+// instructions. There are ELF relocations specifically for it.
+defm : uimm12_pats<(A64WrapperSmall tglobaladdr:$Hi, tglobaladdr:$Lo12, ALIGN),
+                   (ADRPxi tglobaladdr:$Hi), (i64 tglobaladdr:$Lo12)>;
+
+defm : uimm12_pats<(A64WrapperSmall tglobaltlsaddr:$Hi, tglobaltlsaddr:$Lo12,
+                                    ALIGN),
+                   (ADRPxi tglobaltlsaddr:$Hi), (i64 tglobaltlsaddr:$Lo12)>;
+
+// External symbols that make it this far should also get standard relocations.
+defm : uimm12_pats<(A64WrapperSmall texternalsym:$Hi, texternalsym:$Lo12,
+                                    ALIGN),
+                   (ADRPxi texternalsym:$Hi), (i64 texternalsym:$Lo12)>;
+
+defm : uimm12_pats<(A64WrapperSmall tconstpool:$Hi, tconstpool:$Lo12, ALIGN),
+                   (ADRPxi tconstpool:$Hi), (i64 tconstpool:$Lo12)>;
+
+// We also want to use uimm12 instructions for local variables at the moment.
+def tframeindex_XFORM : SDNodeXForm<frameindex, [{
+  int FI = cast<FrameIndexSDNode>(N)->getIndex();
+  return CurDAG->getTargetFrameIndex(FI, MVT::i64);
+}]>;
+
+defm : uimm12_pats<(i64 frameindex:$Rn),
+                   (tframeindex_XFORM tframeindex:$Rn), (i64 0)>;
+
+// These can be much simpler than uimm12 because we don't to change the operand
+// type (e.g. LDURB and LDURH take the same operands).
+multiclass simm9_pats<dag address, dag Base, dag Offset> {
+  defm : ls_small_pats<LS8_LDUR, LS8_STUR, Base, Offset, address, i8>;
+  defm : ls_small_pats<LS16_LDUR, LS16_STUR, Base, Offset, address, i16>;
+
+  defm : ls_int_neutral_pats<LS32_LDUR, LS32_STUR, Base, Offset, address, i32>;
+  defm : ls_int_neutral_pats<LS64_LDUR, LS64_STUR, Base, Offset, address, i64>;
+
+  defm : ls_neutral_pats<LSFP16_LDUR, LSFP16_STUR, Base, Offset, address, f16>;
+  defm : ls_neutral_pats<LSFP32_LDUR, LSFP32_STUR, Base, Offset, address, f32>;
+  defm : ls_neutral_pats<LSFP64_LDUR, LSFP64_STUR, Base, Offset, address, f64>;
+  defm : ls_neutral_pats<LSFP128_LDUR, LSFP128_STUR, Base, Offset, address,
+                         f128>;
+
+  def : Pat<(i64 (zextloadi32 address)),
+            (SUBREG_TO_REG (i64 0), (LS32_LDUR Base, Offset), sub_32)>;
+
+  def : Pat<(truncstorei32 i64:$Rt, address),
+            (LS32_STUR (EXTRACT_SUBREG $Rt, sub_32), Base, Offset)>;
+
+  defm : load_signed_pats<"B", "_U", Base, Offset, address, i8>;
+  defm : load_signed_pats<"H", "_U", Base, Offset, address, i16>;
+  def : Pat<(sextloadi32 address), (LDURSWx Base, Offset)>;
+}
+
+defm : simm9_pats<(add i64:$Rn, simm9:$SImm9),
+                  (i64 $Rn), (SDXF_simm9 simm9:$SImm9)>;
+
+defm : simm9_pats<(add_like_or i64:$Rn, simm9:$SImm9),
+                  (i64 $Rn), (SDXF_simm9 simm9:$SImm9)>;
+
+
+//===------------------------------
+// 3. Register offset patterns
+//===------------------------------
+
+// Atomic patterns can be shared between integer operations of all sizes, a
+// quick multiclass here allows reuse.
+multiclass ro_atomic_pats<Instruction LOAD, Instruction STORE, dag Base,
+                          dag Offset, dag Extend, dag address,
+                          ValueType transty, ValueType sty> {
+  def : Pat<(!cast<PatFrag>("atomic_load_simple_" # sty) address),
+            (LOAD Base, Offset, Extend)>;
+
+  def : Pat<(!cast<PatFrag>("atomic_store_simple_" # sty) address, transty:$Rt),
+            (STORE $Rt, Base, Offset, Extend)>;
+}
+
+// The register offset instructions take three operands giving the instruction,
+// and have an annoying split between instructions where Rm is 32-bit and
+// 64-bit. So we need a special hierarchy to describe them. Other than that the
+// same operations should be supported as for simm9 and uimm12 addressing.
+
+multiclass ro_small_pats<Instruction LOAD, Instruction STORE,
+                         dag Base, dag Offset, dag Extend,
+                         dag address, ValueType sty>
+  : ro_atomic_pats<LOAD, STORE, Base, Offset, Extend, address, i32, sty> {
+  def : Pat<(!cast<SDNode>(zextload # sty) address),
+            (LOAD Base, Offset, Extend)>;
+
+  def : Pat<(!cast<SDNode>(extload # sty) address),
+            (LOAD Base, Offset, Extend)>;
+
+  // For zero-extension to 64-bits we have to tell LLVM that the whole 64-bit
+  // register was actually set.
+  def : Pat<(i64 (!cast<SDNode>(zextload # sty) address)),
+            (SUBREG_TO_REG (i64 0), (LOAD Base, Offset, Extend), sub_32)>;
+
+  def : Pat<(i64 (!cast<SDNode>(extload # sty) address)),
+            (SUBREG_TO_REG (i64 0), (LOAD Base, Offset, Extend), sub_32)>;
+
+  def : Pat<(!cast<SDNode>(truncstore # sty) i32:$Rt, address),
+            (STORE $Rt, Base, Offset, Extend)>;
+
+  // For truncating store from 64-bits, we have to manually tell LLVM to
+  // ignore the high bits of the x register.
+  def : Pat<(!cast<SDNode>(truncstore # sty) i64:$Rt, address),
+            (STORE (EXTRACT_SUBREG $Rt, sub_32), Base, Offset, Extend)>;
+
+}
+
+// Next come patterns for sign-extending loads.
+multiclass ro_signed_pats<string T, string Rm, dag Base, dag Offset, dag Extend,
+                          dag address, ValueType sty> {
+  def : Pat<(i32 (!cast<SDNode>("sextload" # sty) address)),
+            (!cast<Instruction>("LDRS" # T # "w_" # Rm # "_RegOffset")
+              Base, Offset, Extend)>;
+
+  def : Pat<(i64 (!cast<SDNode>("sextload" # sty) address)),
+            (!cast<Instruction>("LDRS" # T # "x_" # Rm # "_RegOffset")
+              Base, Offset, Extend)>;
+}
+
+// and finally "natural-width" loads and stores come next.
+multiclass ro_neutral_pats<Instruction LOAD, Instruction STORE,
+                           dag Base, dag Offset, dag Extend, dag address,
+                           ValueType sty> {
+  def : Pat<(sty (load address)), (LOAD Base, Offset, Extend)>;
+  def : Pat<(store sty:$Rt, address),
+            (STORE $Rt, Base, Offset, Extend)>;
+}
+
+multiclass ro_int_neutral_pats<Instruction LOAD, Instruction STORE,
+                               dag Base, dag Offset, dag Extend, dag address,
+                               ValueType sty>
+  : ro_neutral_pats<LOAD, STORE, Base, Offset, Extend, address, sty>,
+    ro_atomic_pats<LOAD, STORE, Base, Offset, Extend, address, sty, sty>;
+
+multiclass regoff_pats<string Rm, dag address, dag Base, dag Offset,
+                       dag Extend> {
+  defm : ro_small_pats<!cast<Instruction>("LS8_" # Rm # "_RegOffset_LDR"),
+                       !cast<Instruction>("LS8_" # Rm # "_RegOffset_STR"),
+                       Base, Offset, Extend,
+                       !foreach(decls.pattern, address,
+                                !subst(SHIFT, imm_eq0, decls.pattern)),
+                       i8>;
+  defm : ro_small_pats<!cast<Instruction>("LS16_" # Rm # "_RegOffset_LDR"),
+                       !cast<Instruction>("LS16_" # Rm # "_RegOffset_STR"),
+                       Base, Offset, Extend,
+                       !foreach(decls.pattern, address,
+                                !subst(SHIFT, imm_eq1, decls.pattern)),
+                       i16>;
+  defm : ro_small_pats<!cast<Instruction>("LS32_" # Rm # "_RegOffset_LDR"),
+                       !cast<Instruction>("LS32_" # Rm # "_RegOffset_STR"),
+                       Base, Offset, Extend,
+                       !foreach(decls.pattern, address,
+                                !subst(SHIFT, imm_eq2, decls.pattern)),
+                       i32>;
+
+  defm : ro_int_neutral_pats<
+                            !cast<Instruction>("LS32_" # Rm # "_RegOffset_LDR"),
+                            !cast<Instruction>("LS32_" # Rm # "_RegOffset_STR"),
+                            Base, Offset, Extend,
+                            !foreach(decls.pattern, address,
+                                     !subst(SHIFT, imm_eq2, decls.pattern)),
+                            i32>;
+
+  defm : ro_int_neutral_pats<
+                            !cast<Instruction>("LS64_" # Rm # "_RegOffset_LDR"),
+                            !cast<Instruction>("LS64_" # Rm # "_RegOffset_STR"),
+                            Base, Offset, Extend,
+                            !foreach(decls.pattern, address,
+                                     !subst(SHIFT, imm_eq3, decls.pattern)),
+                            i64>;
+
+  defm : ro_neutral_pats<!cast<Instruction>("LSFP16_" # Rm # "_RegOffset_LDR"),
+                         !cast<Instruction>("LSFP16_" # Rm # "_RegOffset_STR"),
+                         Base, Offset, Extend,
+                         !foreach(decls.pattern, address,
+                                  !subst(SHIFT, imm_eq1, decls.pattern)),
+                         f16>;
+
+  defm : ro_neutral_pats<!cast<Instruction>("LSFP32_" # Rm # "_RegOffset_LDR"),
+                         !cast<Instruction>("LSFP32_" # Rm # "_RegOffset_STR"),
+                         Base, Offset, Extend,
+                         !foreach(decls.pattern, address,
+                                  !subst(SHIFT, imm_eq2, decls.pattern)),
+                         f32>;
+
+  defm : ro_neutral_pats<!cast<Instruction>("LSFP64_" # Rm # "_RegOffset_LDR"),
+                         !cast<Instruction>("LSFP64_" # Rm # "_RegOffset_STR"),
+                         Base, Offset, Extend,
+                         !foreach(decls.pattern, address,
+                                  !subst(SHIFT, imm_eq3, decls.pattern)),
+                         f64>;
+
+  defm : ro_neutral_pats<!cast<Instruction>("LSFP128_" # Rm # "_RegOffset_LDR"),
+                         !cast<Instruction>("LSFP128_" # Rm # "_RegOffset_STR"),
+                         Base, Offset, Extend,
+                         !foreach(decls.pattern, address,
+                                  !subst(SHIFT, imm_eq4, decls.pattern)),
+                         f128>;
+
+  defm : ro_signed_pats<"B", Rm, Base, Offset, Extend,
+                        !foreach(decls.pattern, address,
+                                 !subst(SHIFT, imm_eq0, decls.pattern)),
+                        i8>;
+
+  defm : ro_signed_pats<"H", Rm, Base, Offset, Extend,
+                        !foreach(decls.pattern, address,
+                                 !subst(SHIFT, imm_eq1, decls.pattern)),
+                        i16>;
+
+  def : Pat<(sextloadi32 !foreach(decls.pattern, address,
+                                  !subst(SHIFT, imm_eq2, decls.pattern))),
+            (!cast<Instruction>("LDRSWx_" # Rm # "_RegOffset")
+              Base, Offset, Extend)>;
+}
+
+
+// Finally we're in a position to tell LLVM exactly what addresses are reachable
+// using register-offset instructions. Essentially a base plus a possibly
+// extended, possibly shifted (by access size) offset.
+
+defm : regoff_pats<"Wm", (add i64:$Rn, (sext i32:$Rm)),
+                   (i64 i64:$Rn), (i32 i32:$Rm), (i64 6)>;
+
+defm : regoff_pats<"Wm", (add i64:$Rn, (shl (sext i32:$Rm), SHIFT)),
+                   (i64 i64:$Rn), (i32 i32:$Rm), (i64 7)>;
+
+defm : regoff_pats<"Wm", (add i64:$Rn, (zext i32:$Rm)),
+                   (i64 i64:$Rn), (i32 i32:$Rm), (i64 2)>;
+
+defm : regoff_pats<"Wm", (add i64:$Rn, (shl (zext i32:$Rm), SHIFT)),
+                   (i64 i64:$Rn), (i32 i32:$Rm), (i64 3)>;
+
+defm : regoff_pats<"Xm", (add i64:$Rn, i64:$Rm),
+                   (i64 i64:$Rn), (i64 i64:$Rm), (i64 2)>;
+
+defm : regoff_pats<"Xm", (add i64:$Rn, (shl i64:$Rm, SHIFT)),
+                   (i64 i64:$Rn), (i64 i64:$Rm), (i64 3)>;
diff --git a/lib/Target/AArch64/AArch64MCInstLower.cpp b/lib/Target/AArch64/AArch64MCInstLower.cpp
new file mode 100644
index 000000000000..c96bf85a716c
--- /dev/null
+++ b/lib/Target/AArch64/AArch64MCInstLower.cpp
@@ -0,0 +1,140 @@
+//===-- AArch64MCInstLower.cpp - Convert AArch64 MachineInstr to an MCInst -==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains code to lower AArch64 MachineInstrs to their corresponding
+// MCInst records.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AArch64AsmPrinter.h"
+#include "AArch64TargetMachine.h"
+#include "MCTargetDesc/AArch64MCExpr.h"
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/CodeGen/AsmPrinter.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/Target/Mangler.h"
+
+using namespace llvm;
+
+MCOperand
+AArch64AsmPrinter::lowerSymbolOperand(const MachineOperand &MO,
+                                      const MCSymbol *Sym) const {
+  const MCExpr *Expr = 0;
+
+  Expr = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, OutContext);
+
+  switch (MO.getTargetFlags()) {
+  case AArch64II::MO_GOT:
+    Expr = AArch64MCExpr::CreateGOT(Expr, OutContext);
+    break;
+  case AArch64II::MO_GOT_LO12:
+    Expr = AArch64MCExpr::CreateGOTLo12(Expr, OutContext);
+    break;
+  case AArch64II::MO_LO12:
+    Expr = AArch64MCExpr::CreateLo12(Expr, OutContext);
+    break;
+  case AArch64II::MO_DTPREL_G1:
+    Expr = AArch64MCExpr::CreateDTPREL_G1(Expr, OutContext);
+    break;
+  case AArch64II::MO_DTPREL_G0_NC:
+    Expr = AArch64MCExpr::CreateDTPREL_G0_NC(Expr, OutContext);
+    break;
+  case AArch64II::MO_GOTTPREL:
+    Expr = AArch64MCExpr::CreateGOTTPREL(Expr, OutContext);
+    break;
+  case AArch64II::MO_GOTTPREL_LO12:
+    Expr = AArch64MCExpr::CreateGOTTPRELLo12(Expr, OutContext);
+    break;
+  case AArch64II::MO_TLSDESC:
+    Expr = AArch64MCExpr::CreateTLSDesc(Expr, OutContext);
+    break;
+  case AArch64II::MO_TLSDESC_LO12:
+    Expr = AArch64MCExpr::CreateTLSDescLo12(Expr, OutContext);
+    break;
+  case AArch64II::MO_TPREL_G1:
+    Expr = AArch64MCExpr::CreateTPREL_G1(Expr, OutContext);
+    break;
+  case AArch64II::MO_TPREL_G0_NC:
+    Expr = AArch64MCExpr::CreateTPREL_G0_NC(Expr, OutContext);
+    break;
+  case AArch64II::MO_NO_FLAG:
+    // Expr is already correct
+    break;
+  default:
+    llvm_unreachable("Unexpected MachineOperand flag");
+  }
+
+  if (!MO.isJTI() && MO.getOffset())
+    Expr = MCBinaryExpr::CreateAdd(Expr,
+                                   MCConstantExpr::Create(MO.getOffset(),
+                                                          OutContext),
+                                   OutContext);
+
+  return MCOperand::CreateExpr(Expr);
+}
+
+bool AArch64AsmPrinter::lowerOperand(const MachineOperand &MO,
+                                     MCOperand &MCOp) const {
+  switch (MO.getType()) {
+  default: llvm_unreachable("unknown operand type");
+  case MachineOperand::MO_Register:
+    if (MO.isImplicit())
+      return false;
+    assert(!MO.getSubReg() && "Subregs should be eliminated!");
+    MCOp = MCOperand::CreateReg(MO.getReg());
+    break;
+  case MachineOperand::MO_Immediate:
+    MCOp = MCOperand::CreateImm(MO.getImm());
+    break;
+  case MachineOperand::MO_BlockAddress:
+    MCOp = lowerSymbolOperand(MO, GetBlockAddressSymbol(MO.getBlockAddress()));
+    break;
+  case MachineOperand::MO_ExternalSymbol:
+    MCOp = lowerSymbolOperand(MO, GetExternalSymbolSymbol(MO.getSymbolName()));
+    break;
+  case MachineOperand::MO_GlobalAddress:
+    MCOp = lowerSymbolOperand(MO, Mang->getSymbol(MO.getGlobal()));
+    break;
+  case MachineOperand::MO_MachineBasicBlock:
+    MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
+                                   MO.getMBB()->getSymbol(), OutContext));
+    break;
+  case MachineOperand::MO_JumpTableIndex:
+    MCOp = lowerSymbolOperand(MO, GetJTISymbol(MO.getIndex()));
+    break;
+  case MachineOperand::MO_ConstantPoolIndex:
+    MCOp = lowerSymbolOperand(MO, GetCPISymbol(MO.getIndex()));
+    break;
+  case MachineOperand::MO_RegisterMask:
+    // Ignore call clobbers
+    return false;
+
+  }
+
+  return true;
+}
+
+void llvm::LowerAArch64MachineInstrToMCInst(const MachineInstr *MI,
+                                            MCInst &OutMI,
+                                            AArch64AsmPrinter &AP) {
+  OutMI.setOpcode(MI->getOpcode());
+
+  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+    const MachineOperand &MO = MI->getOperand(i);
+
+    MCOperand MCOp;
+    if (AP.lowerOperand(MO, MCOp))
+      OutMI.addOperand(MCOp);
+  }
+}
diff --git a/lib/Target/CellSPU/SPUSelectionDAGInfo.cpp b/lib/Target/AArch64/AArch64MachineFunctionInfo.cpp
index 5732fd43cdc2..f45d8f784f42 100644
--- a/lib/Target/CellSPU/SPUSelectionDAGInfo.cpp
+++ b/lib/Target/AArch64/AArch64MachineFunctionInfo.cpp
@@ -1,4 +1,4 @@
-//===-- SPUSelectionDAGInfo.cpp - CellSPU SelectionDAG Info ---------------===//
+//===-- AArch64MachineFuctionInfo.cpp - AArch64 machine function info -----===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,17 +7,12 @@
 //
 //===----------------------------------------------------------------------===//
 //
-// This file implements the SPUSelectionDAGInfo class.
+// This file just contains the anchor for the AArch64MachineFunctionInfo to
+// force vtable emission.
 //
 //===----------------------------------------------------------------------===//
+#include "AArch64MachineFunctionInfo.h"
 
-#define DEBUG_TYPE "cellspu-selectiondag-info"
-#include "SPUTargetMachine.h"
 using namespace llvm;
 
-SPUSelectionDAGInfo::SPUSelectionDAGInfo(const SPUTargetMachine &TM)
-  : TargetSelectionDAGInfo(TM) {
-}
-
-SPUSelectionDAGInfo::~SPUSelectionDAGInfo() {
-}
+void AArch64MachineFunctionInfo::anchor() { }
diff --git a/lib/Target/AArch64/AArch64MachineFunctionInfo.h b/lib/Target/AArch64/AArch64MachineFunctionInfo.h
new file mode 100644
index 000000000000..33da54f97fda
--- /dev/null
+++ b/lib/Target/AArch64/AArch64MachineFunctionInfo.h
@@ -0,0 +1,149 @@
+//=- AArch64MachineFuctionInfo.h - AArch64 machine function info -*- C++ -*-==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file declares AArch64-specific per-machine-function information.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef AARCH64MACHINEFUNCTIONINFO_H
+#define AARCH64MACHINEFUNCTIONINFO_H
+
+#include "llvm/CodeGen/MachineFunction.h"
+
+namespace llvm {
+
+/// This class is derived from MachineFunctionInfo and contains private AArch64
+/// target-specific information for each MachineFunction.
+class AArch64MachineFunctionInfo : public MachineFunctionInfo {
+  virtual void anchor();
+
+  /// Number of bytes of arguments this function has on the stack. If the callee
+  /// is expected to restore the argument stack this should be a multiple of 16,
+  /// all usable during a tail call.
+  ///
+  /// The alternative would forbid tail call optimisation in some cases: if we
+  /// want to transfer control from a function with 8-bytes of stack-argument
+  /// space to a function with 16-bytes then misalignment of this value would
+  /// make a stack adjustment necessary, which could not be undone by the
+  /// callee.
+  unsigned BytesInStackArgArea;
+
+  /// The number of bytes to restore to deallocate space for incoming
+  /// arguments. Canonically 0 in the C calling convention, but non-zero when
+  /// callee is expected to pop the args.
+  unsigned ArgumentStackToRestore;
+
+  /// If the stack needs to be adjusted on frame entry in two stages, this
+  /// records the size of the first adjustment just prior to storing
+  /// callee-saved registers. The callee-saved slots are addressed assuming
+  /// SP == <incoming-SP> - InitialStackAdjust.
+  unsigned InitialStackAdjust;
+
+  /// Number of local-dynamic TLS accesses.
+  unsigned NumLocalDynamics;
+
+  /// @see AArch64 Procedure Call Standard, B.3
+  ///
+  /// The Frame index of the area where LowerFormalArguments puts the
+  /// general-purpose registers that might contain variadic parameters.
+  int VariadicGPRIdx;
+
+  /// @see AArch64 Procedure Call Standard, B.3
+  ///
+  /// The size of the frame object used to store the general-purpose registers
+  /// which might contain variadic arguments. This is the offset from
+  /// VariadicGPRIdx to what's stored in __gr_top.
+  unsigned VariadicGPRSize;
+
+  /// @see AArch64 Procedure Call Standard, B.3
+  ///
+  /// The Frame index of the area where LowerFormalArguments puts the
+  /// floating-point registers that might contain variadic parameters.
+  int VariadicFPRIdx;
+
+  /// @see AArch64 Procedure Call Standard, B.3
+  ///
+  /// The size of the frame object used to store the floating-point registers
+  /// which might contain variadic arguments. This is the offset from
+  /// VariadicFPRIdx to what's stored in __vr_top.
+  unsigned VariadicFPRSize;
+
+  /// @see AArch64 Procedure Call Standard, B.3
+  ///
+  /// The Frame index of an object pointing just past the last known stacked
+  /// argument on entry to a variadic function. This goes into the __stack field
+  /// of the va_list type.
+  int VariadicStackIdx;
+
+  /// The offset of the frame pointer from the stack pointer on function
+  /// entry. This is expected to be negative.
+  int FramePointerOffset;
+
+public:
+  AArch64MachineFunctionInfo()
+    : BytesInStackArgArea(0),
+      ArgumentStackToRestore(0),
+      InitialStackAdjust(0),
+      NumLocalDynamics(0),
+      VariadicGPRIdx(0),
+      VariadicGPRSize(0),
+      VariadicFPRIdx(0),
+      VariadicFPRSize(0),
+      VariadicStackIdx(0),
+      FramePointerOffset(0) {}
+
+  explicit AArch64MachineFunctionInfo(MachineFunction &MF)
+    : BytesInStackArgArea(0),
+      ArgumentStackToRestore(0),
+      InitialStackAdjust(0),
+      NumLocalDynamics(0),
+      VariadicGPRIdx(0),
+      VariadicGPRSize(0),
+      VariadicFPRIdx(0),
+      VariadicFPRSize(0),
+      VariadicStackIdx(0),
+      FramePointerOffset(0) {}
+
+  unsigned getBytesInStackArgArea() const { return BytesInStackArgArea; }
+  void setBytesInStackArgArea (unsigned bytes) { BytesInStackArgArea = bytes;}
+
+  unsigned getArgumentStackToRestore() const { return ArgumentStackToRestore; }
+  void setArgumentStackToRestore(unsigned bytes) {
+    ArgumentStackToRestore = bytes;
+  }
+
+  unsigned getInitialStackAdjust() const { return InitialStackAdjust; }
+  void setInitialStackAdjust(unsigned bytes) { InitialStackAdjust = bytes; }
+
+  unsigned getNumLocalDynamicTLSAccesses() const { return NumLocalDynamics; }
+  void incNumLocalDynamicTLSAccesses() { ++NumLocalDynamics; }
+
+  int getVariadicGPRIdx() const { return VariadicGPRIdx; }
+  void setVariadicGPRIdx(int Idx) { VariadicGPRIdx = Idx; }
+
+  unsigned getVariadicGPRSize() const { return VariadicGPRSize; }
+  void setVariadicGPRSize(unsigned Size) { VariadicGPRSize = Size; }
+
+  int getVariadicFPRIdx() const { return VariadicFPRIdx; }
+  void setVariadicFPRIdx(int Idx) { VariadicFPRIdx = Idx; }
+
+  unsigned getVariadicFPRSize() const { return VariadicFPRSize; }
+  void setVariadicFPRSize(unsigned Size) { VariadicFPRSize = Size; }
+
+  int getVariadicStackIdx() const { return VariadicStackIdx; }
+  void setVariadicStackIdx(int Idx) { VariadicStackIdx = Idx; }
+
+  int getFramePointerOffset() const { return FramePointerOffset; }
+  void setFramePointerOffset(int Idx) { FramePointerOffset = Idx; }
+
+};
+
+} // End llvm namespace
+
+#endif
diff --git a/lib/Target/AArch64/AArch64RegisterInfo.cpp b/lib/Target/AArch64/AArch64RegisterInfo.cpp
new file mode 100644
index 000000000000..20b0dcf86f46
--- /dev/null
+++ b/lib/Target/AArch64/AArch64RegisterInfo.cpp
@@ -0,0 +1,171 @@
+//===- AArch64RegisterInfo.cpp - AArch64 Register Information -------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the AArch64 implementation of the TargetRegisterInfo
+// class.
+//
+//===----------------------------------------------------------------------===//
+
+
+#include "AArch64RegisterInfo.h"
+#include "AArch64FrameLowering.h"
+#include "AArch64MachineFunctionInfo.h"
+#include "AArch64TargetMachine.h"
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/ADT/BitVector.h"
+
+#define GET_REGINFO_TARGET_DESC
+#include "AArch64GenRegisterInfo.inc"
+
+using namespace llvm;
+
+AArch64RegisterInfo::AArch64RegisterInfo(const AArch64InstrInfo &tii,
+                                         const AArch64Subtarget &sti)
+  : AArch64GenRegisterInfo(AArch64::X30), TII(tii) {
+}
+
+const uint16_t *
+AArch64RegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
+  return CSR_PCS_SaveList;
+}
+
+const uint32_t*
+AArch64RegisterInfo::getCallPreservedMask(CallingConv::ID) const {
+  return CSR_PCS_RegMask;
+}
+
+const uint32_t *AArch64RegisterInfo::getTLSDescCallPreservedMask() const {
+  return TLSDesc_RegMask;
+}
+
+const TargetRegisterClass *
+AArch64RegisterInfo::getCrossCopyRegClass(const TargetRegisterClass *RC) const {
+  if (RC == &AArch64::FlagClassRegClass)
+    return &AArch64::GPR64RegClass;
+
+  return RC;
+}
+
+
+
+BitVector
+AArch64RegisterInfo::getReservedRegs(const MachineFunction &MF) const {
+  BitVector Reserved(getNumRegs());
+  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
+
+  Reserved.set(AArch64::XSP);
+  Reserved.set(AArch64::WSP);
+
+  Reserved.set(AArch64::XZR);
+  Reserved.set(AArch64::WZR);
+
+  if (TFI->hasFP(MF)) {
+    Reserved.set(AArch64::X29);
+    Reserved.set(AArch64::W29);
+  }
+
+  return Reserved;
+}
+
+void
+AArch64RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MBBI,
+                                         int SPAdj,
+                                         unsigned FIOperandNum,
+                                         RegScavenger *RS) const {
+  assert(SPAdj == 0 && "Cannot deal with nonzero SPAdj yet");
+  MachineInstr &MI = *MBBI;
+  MachineBasicBlock &MBB = *MI.getParent();
+  MachineFunction &MF = *MBB.getParent();
+  MachineFrameInfo *MFI = MF.getFrameInfo();
+  const AArch64FrameLowering *TFI =
+   static_cast<const AArch64FrameLowering *>(MF.getTarget().getFrameLowering());
+
+  // In order to work out the base and offset for addressing, the FrameLowering
+  // code needs to know (sometimes) whether the instruction is storing/loading a
+  // callee-saved register, or whether it's a more generic
+  // operation. Fortunately the frame indices are used *only* for that purpose
+  // and are contiguous, so we can check here.
+  const std::vector<CalleeSavedInfo> &CSI = MFI->getCalleeSavedInfo();
+  int MinCSFI = 0;
+  int MaxCSFI = -1;
+
+  if (CSI.size()) {
+    MinCSFI = CSI[0].getFrameIdx();
+    MaxCSFI = CSI[CSI.size() - 1].getFrameIdx();
+  }
+
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
+  bool IsCalleeSaveOp = FrameIndex >= MinCSFI && FrameIndex <= MaxCSFI;
+
+  unsigned FrameReg;
+  int64_t Offset;
+  Offset = TFI->resolveFrameIndexReference(MF, FrameIndex, FrameReg, SPAdj,
+                                           IsCalleeSaveOp);
+
+  Offset += MI.getOperand(FIOperandNum + 1).getImm();
+
+  // DBG_VALUE instructions have no real restrictions so they can be handled
+  // easily.
+  if (MI.isDebugValue()) {
+    MI.getOperand(FIOperandNum).ChangeToRegister(FrameReg, /*isDef=*/ false);
+    MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset);
+    return;
+  }
+
+  int MinOffset, MaxOffset, OffsetScale;
+  if (MI.getOpcode() == AArch64::ADDxxi_lsl0_s) {
+    MinOffset = 0;
+    MaxOffset = 0xfff;
+    OffsetScale = 1;
+  } else {
+    // Load/store of a stack object
+    TII.getAddressConstraints(MI, OffsetScale, MinOffset, MaxOffset);
+  }
+
+  // The frame lowering has told us a base and offset it thinks we should use to
+  // access this variable, but it's still up to us to make sure the values are
+  // legal for the instruction in question.
+  if (Offset % OffsetScale != 0 || Offset < MinOffset || Offset > MaxOffset) {
+    unsigned BaseReg =
+      MF.getRegInfo().createVirtualRegister(&AArch64::GPR64RegClass);
+    emitRegUpdate(MBB, MBBI, MBBI->getDebugLoc(), TII,
+                  BaseReg, FrameReg, BaseReg, Offset);
+    FrameReg = BaseReg;
+    Offset = 0;
+  }
+
+  // Negative offsets are expected if we address from FP, but for
+  // now this checks nothing has gone horribly wrong.
+  assert(Offset >= 0 && "Unexpected negative offset from SP");
+
+  MI.getOperand(FIOperandNum).ChangeToRegister(FrameReg, false, false, true);
+  MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset / OffsetScale);
+}
+
+unsigned
+AArch64RegisterInfo::getFrameRegister(const MachineFunction &MF) const {
+  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
+
+  if (TFI->hasFP(MF))
+    return AArch64::X29;
+  else
+    return AArch64::XSP;
+}
+
+bool
+AArch64RegisterInfo::useFPForScavengingIndex(const MachineFunction &MF) const {
+  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
+  const AArch64FrameLowering *AFI
+    = static_cast<const AArch64FrameLowering*>(TFI);
+  return AFI->useFPForAddressing(MF);
+}
diff --git a/lib/Target/AArch64/AArch64RegisterInfo.h b/lib/Target/AArch64/AArch64RegisterInfo.h
new file mode 100644
index 000000000000..bb64fd55b2c3
--- /dev/null
+++ b/lib/Target/AArch64/AArch64RegisterInfo.h
@@ -0,0 +1,76 @@
+//==- AArch64RegisterInfo.h - AArch64 Register Information Impl -*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the AArch64 implementation of the MCRegisterInfo class.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_TARGET_AARCH64REGISTERINFO_H
+#define LLVM_TARGET_AARCH64REGISTERINFO_H
+
+#include "llvm/Target/TargetRegisterInfo.h"
+
+#define GET_REGINFO_HEADER
+#include "AArch64GenRegisterInfo.inc"
+
+namespace llvm {
+
+class AArch64InstrInfo;
+class AArch64Subtarget;
+
+struct AArch64RegisterInfo : public AArch64GenRegisterInfo {
+private:
+  const AArch64InstrInfo &TII;
+
+public:
+  AArch64RegisterInfo(const AArch64InstrInfo &tii,
+                      const AArch64Subtarget &sti);
+
+  const uint16_t *getCalleeSavedRegs(const MachineFunction *MF = 0) const;
+  const uint32_t *getCallPreservedMask(CallingConv::ID) const;
+
+  const uint32_t *getTLSDescCallPreservedMask() const;
+
+  BitVector getReservedRegs(const MachineFunction &MF) const;
+  unsigned getFrameRegister(const MachineFunction &MF) const;
+
+  void eliminateFrameIndex(MachineBasicBlock::iterator II, int SPAdj,
+                           unsigned FIOperandNum,
+                           RegScavenger *Rs = NULL) const;
+
+  /// getCrossCopyRegClass - Returns a legal register class to copy a register
+  /// in the specified class to or from. Returns original class if it is
+  /// possible to copy between a two registers of the specified class.
+  const TargetRegisterClass *
+  getCrossCopyRegClass(const TargetRegisterClass *RC) const;
+
+  /// getLargestLegalSuperClass - Returns the largest super class of RC that is
+  /// legal to use in the current sub-target and has the same spill size.
+  const TargetRegisterClass*
+  getLargestLegalSuperClass(const TargetRegisterClass *RC) const {
+    if (RC == &AArch64::tcGPR64RegClass)
+      return &AArch64::GPR64RegClass;
+
+    return RC;
+  }
+
+  bool requiresRegisterScavenging(const MachineFunction &MF) const {
+    return true;
+  }
+
+  bool requiresFrameIndexScavenging(const MachineFunction &MF) const {
+    return true;
+  }
+
+  bool useFPForScavengingIndex(const MachineFunction &MF) const;
+};
+
+} // end namespace llvm
+
+#endif // LLVM_TARGET_AARCH64REGISTERINFO_H
diff --git a/lib/Target/AArch64/AArch64RegisterInfo.td b/lib/Target/AArch64/AArch64RegisterInfo.td
new file mode 100644
index 000000000000..bd79546371c5
--- /dev/null
+++ b/lib/Target/AArch64/AArch64RegisterInfo.td
@@ -0,0 +1,203 @@
+//===- AArch64RegisterInfo.td - ARM Register defs ----------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+//  This file contains declarations that describe the AArch64 register file
+//
+//===----------------------------------------------------------------------===//
+
+let Namespace = "AArch64" in {
+def sub_128 : SubRegIndex;
+def sub_64 : SubRegIndex;
+def sub_32 : SubRegIndex;
+def sub_16 : SubRegIndex;
+def sub_8  : SubRegIndex;
+
+// The VPR registers are handled as sub-registers of FPR equivalents, but
+// they're really the same thing. We give this concept a special index.
+def sub_alias : SubRegIndex;
+}
+
+// Registers are identified with 5-bit ID numbers.
+class AArch64Reg<bits<16> enc, string n> : Register<n> {
+  let HWEncoding = enc;
+  let Namespace = "AArch64";
+}
+
+class AArch64RegWithSubs<bits<16> enc, string n, list<Register> subregs = [],
+                         list<SubRegIndex> inds = []>
+      : AArch64Reg<enc, n> {
+  let SubRegs = subregs;
+  let SubRegIndices = inds;
+}
+
+//===----------------------------------------------------------------------===//
+//  Integer registers: w0-w30, wzr, wsp, x0-x30, xzr, sp
+//===----------------------------------------------------------------------===//
+
+foreach Index = 0-30 in {
+  def W#Index : AArch64Reg< Index, "w"#Index>, DwarfRegNum<[Index]>;
+}
+
+def WSP : AArch64Reg<31, "wsp">, DwarfRegNum<[31]>;
+def WZR : AArch64Reg<31, "wzr">;
+
+// Could be combined with previous loop, but this way leaves w and x registers
+// consecutive as LLVM register numbers, which makes for easier debugging.
+foreach Index = 0-30 in {
+  def X#Index : AArch64RegWithSubs<Index, "x"#Index,
+                                   [!cast<Register>("W"#Index)], [sub_32]>,
+                DwarfRegNum<[Index]>;
+}
+
+def XSP : AArch64RegWithSubs<31, "sp", [WSP], [sub_32]>, DwarfRegNum<[31]>;
+def XZR : AArch64RegWithSubs<31, "xzr", [WZR], [sub_32]>;
+
+// Most instructions treat register 31 as zero for reads and a black-hole for
+// writes.
+
+// Note that the order of registers is important for the Disassembler here:
+// tablegen uses it to form MCRegisterClass::getRegister, which we assume can
+// take an encoding value.
+def GPR32 : RegisterClass<"AArch64", [i32], 32,
+                          (add (sequence "W%u", 0, 30), WZR)> {
+}
+
+def GPR64 : RegisterClass<"AArch64", [i64], 64,
+                          (add (sequence "X%u", 0, 30), XZR)> {
+}
+
+def GPR32nowzr : RegisterClass<"AArch64", [i32], 32,
+                               (sequence "W%u", 0, 30)> {
+}
+
+def GPR64noxzr : RegisterClass<"AArch64", [i64], 64,
+                               (sequence "X%u", 0, 30)> {
+}
+
+// For tail calls, we can't use callee-saved registers or the structure-return
+// register, as they are supposed to be live across function calls and may be
+// clobbered by the epilogue.
+def tcGPR64 : RegisterClass<"AArch64", [i64], 64,
+                            (add (sequence "X%u", 0, 7),
+                                 (sequence "X%u", 9, 18))> {
+}
+
+
+// Certain addressing-useful instructions accept sp directly. Again the order of
+// registers is important to the Disassembler.
+def GPR32wsp : RegisterClass<"AArch64", [i32], 32,
+                             (add (sequence "W%u", 0, 30), WSP)> {
+}
+
+def GPR64xsp : RegisterClass<"AArch64", [i64], 64,
+                             (add (sequence "X%u", 0, 30), XSP)> {
+}
+
+// Some aliases *only* apply to SP (e.g. MOV uses different encoding for SP and
+// non-SP variants). We can't use a bare register in those patterns because
+// TableGen doesn't like it, so we need a class containing just stack registers
+def Rxsp : RegisterClass<"AArch64", [i64], 64,
+                         (add XSP)> {
+}
+
+def Rwsp : RegisterClass<"AArch64", [i32], 32,
+                         (add WSP)> {
+}
+
+//===----------------------------------------------------------------------===//
+//  Scalar registers in the vector unit:
+//  b0-b31, h0-h31, s0-s31, d0-d31, q0-q31
+//===----------------------------------------------------------------------===//
+
+foreach Index = 0-31 in {
+  def B # Index : AArch64Reg< Index, "b" # Index>,
+                  DwarfRegNum<[!add(Index, 64)]>;
+
+  def H # Index : AArch64RegWithSubs<Index, "h" # Index,
+                                     [!cast<Register>("B" # Index)], [sub_8]>,
+                  DwarfRegNum<[!add(Index, 64)]>;
+
+  def S # Index : AArch64RegWithSubs<Index, "s" # Index,
+                                     [!cast<Register>("H" # Index)], [sub_16]>,
+                  DwarfRegNum<[!add(Index, 64)]>;
+
+  def D # Index : AArch64RegWithSubs<Index, "d" # Index,
+                                     [!cast<Register>("S" # Index)], [sub_32]>,
+                  DwarfRegNum<[!add(Index, 64)]>;
+
+  def Q # Index : AArch64RegWithSubs<Index, "q" # Index,
+                                     [!cast<Register>("D" # Index)], [sub_64]>,
+                  DwarfRegNum<[!add(Index, 64)]>;
+}
+
+
+def FPR8 : RegisterClass<"AArch64", [i8], 8,
+                          (sequence "B%u", 0, 31)> {
+}
+
+def FPR16 : RegisterClass<"AArch64", [f16], 16,
+                          (sequence "H%u", 0, 31)> {
+}
+
+def FPR32 : RegisterClass<"AArch64", [f32], 32,
+                          (sequence "S%u", 0, 31)> {
+}
+
+def FPR64 : RegisterClass<"AArch64", [f64], 64,
+                          (sequence "D%u", 0, 31)> {
+}
+
+def FPR128 : RegisterClass<"AArch64", [f128], 128,
+                          (sequence "Q%u", 0, 31)> {
+}
+
+
+//===----------------------------------------------------------------------===//
+//  Vector registers:
+//===----------------------------------------------------------------------===//
+
+// NEON registers simply specify the overall vector, and it's expected that
+// Instructions will individually specify the acceptable data layout. In
+// principle this leaves two approaches open:
+//   + An operand, giving a single ADDvvv instruction (for example). This turns
+//     out to be unworkable in the assembly parser (without every Instruction
+//     having a "cvt" function, at least) because the constraints can't be
+//     properly enforced. It also complicates specifying patterns since each
+//     instruction will accept many types.
+//  + A bare token (e.g. ".2d"). This means the AsmParser has to know specific
+//    details about NEON registers, but simplifies most other details.
+//
+// The second approach was taken.
+
+foreach Index = 0-31 in {
+  def V # Index  : AArch64RegWithSubs<Index, "v" # Index,
+                                      [!cast<Register>("Q" # Index)],
+                                      [sub_alias]>,
+            DwarfRegNum<[!add(Index, 64)]>;
+}
+
+// These two classes contain the same registers, which should be reasonably
+// sensible for MC and allocation purposes, but allows them to be treated
+// separately for things like stack spilling.
+def VPR64 : RegisterClass<"AArch64", [v2f32, v2i32, v4i16, v8i8], 64,
+                          (sequence "V%u", 0, 31)>;
+
+def VPR128 : RegisterClass<"AArch64",
+                           [v2f64, v2i64, v4f32, v4i32, v8i16, v16i8], 128,
+                           (sequence "V%u", 0, 31)>;
+
+// Flags register
+def NZCV : Register<"nzcv"> {
+  let Namespace = "AArch64";
+}
+
+def FlagClass : RegisterClass<"AArch64", [i32], 32, (add NZCV)> {
+  let CopyCost = -1;
+  let isAllocatable = 0;
+}
diff --git a/lib/Target/CellSPU/SPUMachineFunction.cpp b/lib/Target/AArch64/AArch64Schedule.td
index 3e948d071d63..e17cdaa1f6d2 100644
--- a/lib/Target/CellSPU/SPUMachineFunction.cpp
+++ b/lib/Target/AArch64/AArch64Schedule.td
@@ -1,4 +1,4 @@
-//==-- SPUMachineFunctionInfo.cpp - Private data used for CellSPU ---------===//
+//===- AArch64Schedule.td - AArch64 Scheduling Definitions -*- tablegen -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,8 +7,4 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "SPUMachineFunction.h"
-
-using namespace llvm;
-
-void SPUFunctionInfo::anchor() { }
+def GenericItineraries : ProcessorItineraries<[], [], []>;
diff --git a/lib/Target/AArch64/AArch64SelectionDAGInfo.cpp b/lib/Target/AArch64/AArch64SelectionDAGInfo.cpp
new file mode 100644
index 000000000000..6bbe075a1b61
--- /dev/null
+++ b/lib/Target/AArch64/AArch64SelectionDAGInfo.cpp
@@ -0,0 +1,25 @@
+//===-- AArch64SelectionDAGInfo.cpp - AArch64 SelectionDAG Info -----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the AArch64SelectionDAGInfo class.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "arm-selectiondag-info"
+#include "AArch64TargetMachine.h"
+#include "llvm/CodeGen/SelectionDAG.h"
+using namespace llvm;
+
+AArch64SelectionDAGInfo::AArch64SelectionDAGInfo(const AArch64TargetMachine &TM)
+  : TargetSelectionDAGInfo(TM),
+    Subtarget(&TM.getSubtarget<AArch64Subtarget>()) {
+}
+
+AArch64SelectionDAGInfo::~AArch64SelectionDAGInfo() {
+}
diff --git a/lib/Target/CellSPU/SPUSelectionDAGInfo.h b/lib/Target/AArch64/AArch64SelectionDAGInfo.h
index 39257d92c400..d412ed2be180 100644
--- a/lib/Target/CellSPU/SPUSelectionDAGInfo.h
+++ b/lib/Target/AArch64/AArch64SelectionDAGInfo.h
@@ -1,4 +1,4 @@
-//===-- SPUSelectionDAGInfo.h - CellSPU SelectionDAG Info -------*- C++ -*-===//
+//===-- AArch64SelectionDAGInfo.h - AArch64 SelectionDAG Info ---*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,23 +7,24 @@
 //
 //===----------------------------------------------------------------------===//
 //
-// This file defines the CellSPU subclass for TargetSelectionDAGInfo.
+// This file defines the AArch64 subclass for TargetSelectionDAGInfo.
 //
 //===----------------------------------------------------------------------===//
 
-#ifndef CELLSPUSELECTIONDAGINFO_H
-#define CELLSPUSELECTIONDAGINFO_H
+#ifndef LLVM_AARCH64SELECTIONDAGINFO_H
+#define LLVM_AARCH64SELECTIONDAGINFO_H
 
 #include "llvm/Target/TargetSelectionDAGInfo.h"
 
 namespace llvm {
 
-class SPUTargetMachine;
+class AArch64TargetMachine;
 
-class SPUSelectionDAGInfo : public TargetSelectionDAGInfo {
+class AArch64SelectionDAGInfo : public TargetSelectionDAGInfo {
+  const AArch64Subtarget *Subtarget;
 public:
-  explicit SPUSelectionDAGInfo(const SPUTargetMachine &TM);
-  ~SPUSelectionDAGInfo();
+  explicit AArch64SelectionDAGInfo(const AArch64TargetMachine &TM);
+  ~AArch64SelectionDAGInfo();
 };
 
 }
diff --git a/lib/Target/AArch64/AArch64Subtarget.cpp b/lib/Target/AArch64/AArch64Subtarget.cpp
new file mode 100644
index 000000000000..d17b73820994
--- /dev/null
+++ b/lib/Target/AArch64/AArch64Subtarget.cpp
@@ -0,0 +1,43 @@
+//===-- AArch64Subtarget.cpp - AArch64 Subtarget Information --------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the AArch64 specific subclass of TargetSubtargetInfo.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AArch64Subtarget.h"
+#include "AArch64RegisterInfo.h"
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/ADT/SmallVector.h"
+
+#define GET_SUBTARGETINFO_TARGET_DESC
+#define GET_SUBTARGETINFO_CTOR
+#include "AArch64GenSubtargetInfo.inc"
+
+using namespace llvm;
+
+AArch64Subtarget::AArch64Subtarget(StringRef TT, StringRef CPU, StringRef FS)
+  : AArch64GenSubtargetInfo(TT, CPU, FS)
+  , HasNEON(true)
+  , HasCrypto(true)
+  , TargetTriple(TT) {
+
+  ParseSubtargetFeatures(CPU, FS);
+}
+
+bool AArch64Subtarget::GVIsIndirectSymbol(const GlobalValue *GV,
+                                          Reloc::Model RelocM) const {
+  if (RelocM == Reloc::Static)
+    return false;
+
+  return !GV->hasLocalLinkage() && !GV->hasHiddenVisibility();
+}
diff --git a/lib/Target/AArch64/AArch64Subtarget.h b/lib/Target/AArch64/AArch64Subtarget.h
new file mode 100644
index 000000000000..2e9205fc9924
--- /dev/null
+++ b/lib/Target/AArch64/AArch64Subtarget.h
@@ -0,0 +1,54 @@
+//==-- AArch64Subtarget.h - Define Subtarget for the AArch64 ---*- C++ -*--===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file declares the AArch64 specific subclass of TargetSubtargetInfo.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_TARGET_AARCH64_SUBTARGET_H
+#define LLVM_TARGET_AARCH64_SUBTARGET_H
+
+#include "llvm/ADT/Triple.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
+
+#define GET_SUBTARGETINFO_HEADER
+#include "AArch64GenSubtargetInfo.inc"
+
+#include <string>
+
+namespace llvm {
+class StringRef;
+class GlobalValue;
+
+class AArch64Subtarget : public AArch64GenSubtargetInfo {
+protected:
+  bool HasNEON;
+  bool HasCrypto;
+
+  /// TargetTriple - What processor and OS we're targeting.
+  Triple TargetTriple;
+public:
+  /// This constructor initializes the data members to match that
+  /// of the specified triple.
+  ///
+  AArch64Subtarget(StringRef TT, StringRef CPU, StringRef FS);
+
+  /// ParseSubtargetFeatures - Parses features string setting specified
+  /// subtarget options.  Definition of function is auto generated by tblgen.
+  void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
+
+  bool GVIsIndirectSymbol(const GlobalValue *GV, Reloc::Model RelocM) const;
+
+  bool isTargetELF() const { return TargetTriple.isOSBinFormatELF(); }
+  bool isTargetLinux() const { return TargetTriple.getOS() == Triple::Linux; }
+
+};
+} // End llvm namespace
+
+#endif  // LLVM_TARGET_AARCH64_SUBTARGET_H
diff --git a/lib/Target/AArch64/AArch64TargetMachine.cpp b/lib/Target/AArch64/AArch64TargetMachine.cpp
new file mode 100644
index 000000000000..df599d599dd6
--- /dev/null
+++ b/lib/Target/AArch64/AArch64TargetMachine.cpp
@@ -0,0 +1,81 @@
+//===-- AArch64TargetMachine.cpp - Define TargetMachine for AArch64 -------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the implementation of the AArch64TargetMachine
+// methods. Principally just setting up the passes needed to generate correct
+// code on this architecture.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AArch64.h"
+#include "AArch64TargetMachine.h"
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "llvm/PassManager.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/Support/TargetRegistry.h"
+
+using namespace llvm;
+
+extern "C" void LLVMInitializeAArch64Target() {
+  RegisterTargetMachine<AArch64TargetMachine> X(TheAArch64Target);
+}
+
+AArch64TargetMachine::AArch64TargetMachine(const Target &T, StringRef TT,
+                                           StringRef CPU, StringRef FS,
+                                           const TargetOptions &Options,
+                                           Reloc::Model RM, CodeModel::Model CM,
+                                           CodeGenOpt::Level OL)
+  : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+    Subtarget(TT, CPU, FS),
+    InstrInfo(Subtarget),
+    DL("e-p:64:64-i64:64:64-i128:128:128-s0:32:32-f128:128:128-n32:64-S128"),
+    TLInfo(*this),
+    TSInfo(*this),
+    FrameLowering(Subtarget) {
+}
+
+namespace {
+/// AArch64 Code Generator Pass Configuration Options.
+class AArch64PassConfig : public TargetPassConfig {
+public:
+  AArch64PassConfig(AArch64TargetMachine *TM, PassManagerBase &PM)
+    : TargetPassConfig(TM, PM) {}
+
+  AArch64TargetMachine &getAArch64TargetMachine() const {
+    return getTM<AArch64TargetMachine>();
+  }
+
+  const AArch64Subtarget &getAArch64Subtarget() const {
+    return *getAArch64TargetMachine().getSubtargetImpl();
+  }
+
+  virtual bool addInstSelector();
+  virtual bool addPreEmitPass();
+};
+} // namespace
+
+TargetPassConfig *AArch64TargetMachine::createPassConfig(PassManagerBase &PM) {
+  return new AArch64PassConfig(this, PM);
+}
+
+bool AArch64PassConfig::addPreEmitPass() {
+  addPass(&UnpackMachineBundlesID);
+  addPass(createAArch64BranchFixupPass());
+  return true;
+}
+
+bool AArch64PassConfig::addInstSelector() {
+  addPass(createAArch64ISelDAG(getAArch64TargetMachine(), getOptLevel()));
+
+  // For ELF, cleanup any local-dynamic TLS accesses.
+  if (getAArch64Subtarget().isTargetELF() && getOptLevel() != CodeGenOpt::None)
+    addPass(createAArch64CleanupLocalDynamicTLSPass());
+
+  return false;
+}
diff --git a/lib/Target/AArch64/AArch64TargetMachine.h b/lib/Target/AArch64/AArch64TargetMachine.h
new file mode 100644
index 000000000000..c1f47c2e5372
--- /dev/null
+++ b/lib/Target/AArch64/AArch64TargetMachine.h
@@ -0,0 +1,69 @@
+//=== AArch64TargetMachine.h - Define TargetMachine for AArch64 -*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file declares the AArch64 specific subclass of TargetMachine.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_AARCH64TARGETMACHINE_H
+#define LLVM_AARCH64TARGETMACHINE_H
+
+#include "AArch64FrameLowering.h"
+#include "AArch64ISelLowering.h"
+#include "AArch64InstrInfo.h"
+#include "AArch64SelectionDAGInfo.h"
+#include "AArch64Subtarget.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/Target/TargetMachine.h"
+
+namespace llvm {
+
+class AArch64TargetMachine : public LLVMTargetMachine {
+  AArch64Subtarget          Subtarget;
+  AArch64InstrInfo          InstrInfo;
+  const DataLayout          DL;
+  AArch64TargetLowering     TLInfo;
+  AArch64SelectionDAGInfo   TSInfo;
+  AArch64FrameLowering      FrameLowering;
+
+public:
+  AArch64TargetMachine(const Target &T, StringRef TT, StringRef CPU,
+                       StringRef FS, const TargetOptions &Options,
+                       Reloc::Model RM, CodeModel::Model CM,
+                       CodeGenOpt::Level OL);
+
+  const AArch64InstrInfo *getInstrInfo() const {
+    return &InstrInfo;
+  }
+
+  const AArch64FrameLowering *getFrameLowering() const {
+    return &FrameLowering;
+  }
+
+  const AArch64TargetLowering *getTargetLowering() const {
+    return &TLInfo;
+  }
+
+  const AArch64SelectionDAGInfo *getSelectionDAGInfo() const {
+    return &TSInfo;
+  }
+
+  const AArch64Subtarget *getSubtargetImpl() const { return &Subtarget; }
+
+  const DataLayout *getDataLayout() const { return &DL; }
+
+  const TargetRegisterInfo *getRegisterInfo() const {
+    return &InstrInfo.getRegisterInfo();
+  }
+  TargetPassConfig *createPassConfig(PassManagerBase &PM);
+};
+
+}
+
+#endif
diff --git a/lib/Target/AArch64/AArch64TargetObjectFile.cpp b/lib/Target/AArch64/AArch64TargetObjectFile.cpp
new file mode 100644
index 000000000000..b4452f514590
--- /dev/null
+++ b/lib/Target/AArch64/AArch64TargetObjectFile.cpp
@@ -0,0 +1,24 @@
+//===-- AArch64TargetObjectFile.cpp - AArch64 Object Info -----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file deals with any AArch64 specific requirements on object files.
+//
+//===----------------------------------------------------------------------===//
+
+
+#include "AArch64TargetObjectFile.h"
+
+using namespace llvm;
+
+void
+AArch64LinuxTargetObjectFile::Initialize(MCContext &Ctx,
+                                         const TargetMachine &TM) {
+  TargetLoweringObjectFileELF::Initialize(Ctx, TM);
+  InitializeELF(TM.Options.UseInitArray);
+}
diff --git a/lib/Target/AArch64/AArch64TargetObjectFile.h b/lib/Target/AArch64/AArch64TargetObjectFile.h
new file mode 100644
index 000000000000..bf0565a79ec8
--- /dev/null
+++ b/lib/Target/AArch64/AArch64TargetObjectFile.h
@@ -0,0 +1,31 @@
+//===-- AArch64TargetObjectFile.h - AArch64 Object Info ---------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file deals with any AArch64 specific requirements on object files.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_TARGET_AARCH64_TARGETOBJECTFILE_H
+#define LLVM_TARGET_AARCH64_TARGETOBJECTFILE_H
+
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetLoweringObjectFile.h"
+
+namespace llvm {
+
+  /// AArch64LinuxTargetObjectFile - This implementation is used for linux
+  /// AArch64.
+  class AArch64LinuxTargetObjectFile : public TargetLoweringObjectFileELF {
+    virtual void Initialize(MCContext &Ctx, const TargetMachine &TM);
+  };
+
+} // end namespace llvm
+
+#endif
diff --git a/lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp b/lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp
new file mode 100644
index 000000000000..69bb80a48537
--- /dev/null
+++ b/lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp
@@ -0,0 +1,2197 @@
+//==- AArch64AsmParser.cpp - Parse AArch64 assembly to MCInst instructions -==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the (GNU-style) assembly parser for the AArch64
+// architecture.
+//
+//===----------------------------------------------------------------------===//
+
+
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "MCTargetDesc/AArch64MCExpr.h"
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/ADT/APFloat.h"
+#include "llvm/ADT/APInt.h"
+#include "llvm/ADT/StringSwitch.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/MC/MCTargetAsmParser.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/MC/MCParser/MCAsmLexer.h"
+#include "llvm/MC/MCParser/MCAsmParser.h"
+#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/TargetRegistry.h"
+
+using namespace llvm;
+
+namespace {
+
+class AArch64Operand;
+
+class AArch64AsmParser : public MCTargetAsmParser {
+  MCSubtargetInfo &STI;
+  MCAsmParser &Parser;
+
+#define GET_ASSEMBLER_HEADER
+#include "AArch64GenAsmMatcher.inc"
+
+public:
+  enum AArch64MatchResultTy {
+    Match_FirstAArch64 = FIRST_TARGET_MATCH_RESULT_TY,
+#define GET_OPERAND_DIAGNOSTIC_TYPES
+#include "AArch64GenAsmMatcher.inc"
+  };
+
+  AArch64AsmParser(MCSubtargetInfo &_STI, MCAsmParser &_Parser)
+    : MCTargetAsmParser(), STI(_STI), Parser(_Parser) {
+    MCAsmParserExtension::Initialize(_Parser);
+
+    // Initialize the set of available features.
+    setAvailableFeatures(ComputeAvailableFeatures(STI.getFeatureBits()));
+  }
+
+  // These are the public interface of the MCTargetAsmParser
+  bool ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &EndLoc);
+  bool ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
+                        SMLoc NameLoc,
+                        SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  bool ParseDirective(AsmToken DirectiveID);
+  bool ParseDirectiveTLSDescCall(SMLoc L);
+  bool ParseDirectiveWord(unsigned Size, SMLoc L);
+
+  bool MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
+                               SmallVectorImpl<MCParsedAsmOperand*> &Operands,
+                               MCStreamer&Out, unsigned &ErrorInfo,
+                               bool MatchingInlineAsm);
+
+  // The rest of the sub-parsers have more freedom over interface: they return
+  // an OperandMatchResultTy because it's less ambiguous than true/false or
+  // -1/0/1 even if it is more verbose
+  OperandMatchResultTy
+  ParseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
+               StringRef Mnemonic);
+
+  OperandMatchResultTy ParseImmediate(const MCExpr *&ExprVal);
+
+  OperandMatchResultTy ParseRelocPrefix(AArch64MCExpr::VariantKind &RefKind);
+
+  OperandMatchResultTy
+  ParseNEONLane(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
+                uint32_t NumLanes);
+
+  OperandMatchResultTy
+  ParseRegister(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
+                uint32_t &NumLanes);
+
+  OperandMatchResultTy
+  ParseImmWithLSLOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  OperandMatchResultTy
+  ParseCondCodeOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  OperandMatchResultTy
+  ParseCRxOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  OperandMatchResultTy
+  ParseFPImmOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  template<typename SomeNamedImmMapper> OperandMatchResultTy
+  ParseNamedImmOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+    return ParseNamedImmOperand(SomeNamedImmMapper(), Operands);
+  }
+
+  OperandMatchResultTy
+  ParseNamedImmOperand(const NamedImmMapper &Mapper,
+                       SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  OperandMatchResultTy
+  ParseLSXAddressOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  OperandMatchResultTy
+  ParseShiftExtend(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  OperandMatchResultTy
+  ParseSysRegOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  bool validateInstruction(MCInst &Inst,
+                          const SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  /// Scan the next token (which had better be an identifier) and determine
+  /// whether it represents a general-purpose or vector register. It returns
+  /// true if an identifier was found and populates its reference arguments. It
+  /// does not consume the token.
+  bool
+  IdentifyRegister(unsigned &RegNum, SMLoc &RegEndLoc, StringRef &LayoutSpec,
+                   SMLoc &LayoutLoc) const;
+
+};
+
+}
+
+namespace {
+
+/// Instances of this class represent a parsed AArch64 machine instruction.
+class AArch64Operand : public MCParsedAsmOperand {
+private:
+  enum KindTy {
+    k_ImmWithLSL,     // #uimm {, LSL #amt }
+    k_CondCode,       // eq/ne/...
+    k_FPImmediate,    // Limited-precision floating-point imm
+    k_Immediate,      // Including expressions referencing symbols
+    k_Register,
+    k_ShiftExtend,
+    k_SysReg,         // The register operand of MRS and MSR instructions
+    k_Token,          // The mnemonic; other raw tokens the auto-generated
+    k_WrappedRegister // Load/store exclusive permit a wrapped register.
+  } Kind;
+
+  SMLoc StartLoc, EndLoc;
+
+  struct ImmWithLSLOp {
+    const MCExpr *Val;
+    unsigned ShiftAmount;
+    bool ImplicitAmount;
+  };
+
+  struct CondCodeOp {
+    A64CC::CondCodes Code;
+  };
+
+  struct FPImmOp {
+    double Val;
+  };
+
+  struct ImmOp {
+    const MCExpr *Val;
+  };
+
+  struct RegOp {
+    unsigned RegNum;
+  };
+
+  struct ShiftExtendOp {
+    A64SE::ShiftExtSpecifiers ShiftType;
+    unsigned Amount;
+    bool ImplicitAmount;
+  };
+
+  struct SysRegOp {
+    const char *Data;
+    unsigned Length;
+  };
+
+  struct TokOp {
+    const char *Data;
+    unsigned Length;
+  };
+
+  union {
+    struct ImmWithLSLOp ImmWithLSL;
+    struct CondCodeOp CondCode;
+    struct FPImmOp FPImm;
+    struct ImmOp Imm;
+    struct RegOp Reg;
+    struct ShiftExtendOp ShiftExtend;
+    struct SysRegOp SysReg;
+    struct TokOp Tok;
+  };
+
+  AArch64Operand(KindTy K, SMLoc S, SMLoc E)
+    : MCParsedAsmOperand(), Kind(K), StartLoc(S), EndLoc(E) {}
+
+public:
+  AArch64Operand(const AArch64Operand &o) : MCParsedAsmOperand() {
+  }
+
+  SMLoc getStartLoc() const { return StartLoc; }
+  SMLoc getEndLoc() const { return EndLoc; }
+  void print(raw_ostream&) const;
+  void dump() const;
+
+  StringRef getToken() const {
+    assert(Kind == k_Token && "Invalid access!");
+    return StringRef(Tok.Data, Tok.Length);
+  }
+
+  unsigned getReg() const {
+    assert((Kind == k_Register || Kind == k_WrappedRegister)
+           && "Invalid access!");
+    return Reg.RegNum;
+  }
+
+  const MCExpr *getImm() const {
+    assert(Kind == k_Immediate && "Invalid access!");
+    return Imm.Val;
+  }
+
+  A64CC::CondCodes getCondCode() const {
+    assert(Kind == k_CondCode && "Invalid access!");
+    return CondCode.Code;
+  }
+
+  static bool isNonConstantExpr(const MCExpr *E,
+                                AArch64MCExpr::VariantKind &Variant) {
+    if (const AArch64MCExpr *A64E = dyn_cast<AArch64MCExpr>(E)) {
+      Variant = A64E->getKind();
+      return true;
+    } else if (!isa<MCConstantExpr>(E)) {
+      Variant = AArch64MCExpr::VK_AARCH64_None;
+      return true;
+    }
+
+    return false;
+  }
+
+  bool isCondCode() const { return Kind == k_CondCode; }
+  bool isToken() const { return Kind == k_Token; }
+  bool isReg() const { return Kind == k_Register; }
+  bool isImm() const { return Kind == k_Immediate; }
+  bool isMem() const { return false; }
+  bool isFPImm() const { return Kind == k_FPImmediate; }
+  bool isShiftOrExtend() const { return Kind == k_ShiftExtend; }
+  bool isSysReg() const { return Kind == k_SysReg; }
+  bool isImmWithLSL() const { return Kind == k_ImmWithLSL; }
+  bool isWrappedReg() const { return Kind == k_WrappedRegister; }
+
+  bool isAddSubImmLSL0() const {
+    if (!isImmWithLSL()) return false;
+    if (ImmWithLSL.ShiftAmount != 0) return false;
+
+    AArch64MCExpr::VariantKind Variant;
+    if (isNonConstantExpr(ImmWithLSL.Val, Variant)) {
+      return Variant == AArch64MCExpr::VK_AARCH64_LO12
+          || Variant == AArch64MCExpr::VK_AARCH64_DTPREL_LO12
+          || Variant == AArch64MCExpr::VK_AARCH64_DTPREL_LO12_NC
+          || Variant == AArch64MCExpr::VK_AARCH64_TPREL_LO12
+          || Variant == AArch64MCExpr::VK_AARCH64_TPREL_LO12_NC
+          || Variant == AArch64MCExpr::VK_AARCH64_TLSDESC_LO12;
+    }
+
+    // Otherwise it should be a real immediate in range:
+    const MCConstantExpr *CE = cast<MCConstantExpr>(ImmWithLSL.Val);
+    return CE->getValue() >= 0 && CE->getValue() <= 0xfff;
+  }
+
+  bool isAddSubImmLSL12() const {
+    if (!isImmWithLSL()) return false;
+    if (ImmWithLSL.ShiftAmount != 12) return false;
+
+    AArch64MCExpr::VariantKind Variant;
+    if (isNonConstantExpr(ImmWithLSL.Val, Variant)) {
+      return Variant == AArch64MCExpr::VK_AARCH64_DTPREL_HI12
+          || Variant == AArch64MCExpr::VK_AARCH64_TPREL_HI12;
+    }
+
+    // Otherwise it should be a real immediate in range:
+    const MCConstantExpr *CE = cast<MCConstantExpr>(ImmWithLSL.Val);
+    return CE->getValue() >= 0 && CE->getValue() <= 0xfff;
+  }
+
+  template<unsigned MemSize, unsigned RmSize> bool isAddrRegExtend() const {
+    if (!isShiftOrExtend()) return false;
+
+    A64SE::ShiftExtSpecifiers Ext = ShiftExtend.ShiftType;
+    if (RmSize == 32 && !(Ext == A64SE::UXTW || Ext == A64SE::SXTW))
+      return false;
+
+    if (RmSize == 64 && !(Ext == A64SE::LSL || Ext == A64SE::SXTX))
+      return false;
+
+    return ShiftExtend.Amount == Log2_32(MemSize) || ShiftExtend.Amount == 0;
+  }
+
+  bool isAdrpLabel() const {
+    if (!isImm()) return false;
+
+    AArch64MCExpr::VariantKind Variant;
+    if (isNonConstantExpr(getImm(), Variant)) {
+      return Variant == AArch64MCExpr::VK_AARCH64_None
+        || Variant == AArch64MCExpr::VK_AARCH64_GOT
+        || Variant == AArch64MCExpr::VK_AARCH64_GOTTPREL
+        || Variant == AArch64MCExpr::VK_AARCH64_TLSDESC;
+    }
+
+    return isLabel<21, 4096>();
+  }
+
+  template<unsigned RegWidth>  bool isBitfieldWidth() const {
+    if (!isImm()) return false;
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+    if (!CE) return false;
+
+    return CE->getValue() >= 1 && CE->getValue() <= RegWidth;
+  }
+
+  template<int RegWidth>
+  bool isCVTFixedPos() const {
+    if (!isImm()) return false;
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+    if (!CE) return false;
+
+    return CE->getValue() >= 1 && CE->getValue() <= RegWidth;
+  }
+
+  bool isFMOVImm() const {
+    if (!isFPImm()) return false;
+
+    APFloat RealVal(FPImm.Val);
+    uint32_t ImmVal;
+    return A64Imms::isFPImm(RealVal, ImmVal);
+  }
+
+  bool isFPZero() const {
+    if (!isFPImm()) return false;
+
+    APFloat RealVal(FPImm.Val);
+    return RealVal.isPosZero();
+  }
+
+  template<unsigned field_width, unsigned scale>
+  bool isLabel() const {
+    if (!isImm()) return false;
+
+    if (dyn_cast<MCSymbolRefExpr>(Imm.Val)) {
+      return true;
+    } else if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Imm.Val)) {
+      int64_t Val = CE->getValue();
+      int64_t Min = - (scale * (1LL << (field_width - 1)));
+      int64_t Max = scale * ((1LL << (field_width - 1)) - 1);
+      return (Val % scale) == 0 && Val >= Min && Val <= Max;
+    }
+
+    // N.b. this disallows explicit relocation specifications via an
+    // AArch64MCExpr. Users needing that behaviour
+    return false;
+  }
+
+  bool isLane1() const {
+    if (!isImm()) return false;
+
+    // Because it's come through custom assembly parsing, it must always be a
+    // constant expression.
+    return cast<MCConstantExpr>(getImm())->getValue() == 1;
+  }
+
+  bool isLoadLitLabel() const {
+    if (!isImm()) return false;
+
+    AArch64MCExpr::VariantKind Variant;
+    if (isNonConstantExpr(getImm(), Variant)) {
+      return Variant == AArch64MCExpr::VK_AARCH64_None
+          || Variant == AArch64MCExpr::VK_AARCH64_GOTTPREL;
+    }
+
+    return isLabel<19, 4>();
+  }
+
+  template<unsigned RegWidth> bool isLogicalImm() const {
+    if (!isImm()) return false;
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Imm.Val);
+    if (!CE) return false;
+
+    uint32_t Bits;
+    return A64Imms::isLogicalImm(RegWidth, CE->getValue(), Bits);
+  }
+
+  template<unsigned RegWidth> bool isLogicalImmMOV() const {
+    if (!isLogicalImm<RegWidth>()) return false;
+
+    const MCConstantExpr *CE = cast<MCConstantExpr>(Imm.Val);
+
+    // The move alias for ORR is only valid if the immediate cannot be
+    // represented with a move (immediate) instruction; they take priority.
+    int UImm16, Shift;
+    return !A64Imms::isMOVZImm(RegWidth, CE->getValue(), UImm16, Shift)
+      && !A64Imms::isMOVNImm(RegWidth, CE->getValue(), UImm16, Shift);
+  }
+
+  template<int MemSize>
+  bool isOffsetUImm12() const {
+    if (!isImm()) return false;
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+
+    // Assume they know what they're doing for now if they've given us a
+    // non-constant expression. In principle we could check for ridiculous
+    // things that can't possibly work or relocations that would almost
+    // certainly break resulting code.
+    if (!CE)
+      return true;
+
+    int64_t Val = CE->getValue();
+
+    // Must be a multiple of the access size in bytes.
+    if ((Val & (MemSize - 1)) != 0) return false;
+
+    // Must be 12-bit unsigned
+    return Val >= 0 && Val <= 0xfff * MemSize;
+  }
+
+  template<A64SE::ShiftExtSpecifiers SHKind, bool is64Bit>
+  bool isShift() const {
+    if (!isShiftOrExtend()) return false;
+
+    if (ShiftExtend.ShiftType != SHKind)
+      return false;
+
+    return is64Bit ? ShiftExtend.Amount <= 63 : ShiftExtend.Amount <= 31;
+  }
+
+  bool isMOVN32Imm() const {
+    static AArch64MCExpr::VariantKind PermittedModifiers[] = {
+      AArch64MCExpr::VK_AARCH64_SABS_G0,
+      AArch64MCExpr::VK_AARCH64_SABS_G1,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G1,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G0,
+      AArch64MCExpr::VK_AARCH64_GOTTPREL_G1,
+      AArch64MCExpr::VK_AARCH64_TPREL_G1,
+      AArch64MCExpr::VK_AARCH64_TPREL_G0,
+    };
+    unsigned NumModifiers = llvm::array_lengthof(PermittedModifiers);
+
+    return isMoveWideImm(32, PermittedModifiers, NumModifiers);
+  }
+
+  bool isMOVN64Imm() const {
+    static AArch64MCExpr::VariantKind PermittedModifiers[] = {
+      AArch64MCExpr::VK_AARCH64_SABS_G0,
+      AArch64MCExpr::VK_AARCH64_SABS_G1,
+      AArch64MCExpr::VK_AARCH64_SABS_G2,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G2,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G1,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G0,
+      AArch64MCExpr::VK_AARCH64_GOTTPREL_G1,
+      AArch64MCExpr::VK_AARCH64_TPREL_G2,
+      AArch64MCExpr::VK_AARCH64_TPREL_G1,
+      AArch64MCExpr::VK_AARCH64_TPREL_G0,
+    };
+    unsigned NumModifiers = llvm::array_lengthof(PermittedModifiers);
+
+    return isMoveWideImm(64, PermittedModifiers, NumModifiers);
+  }
+
+
+  bool isMOVZ32Imm() const {
+    static AArch64MCExpr::VariantKind PermittedModifiers[] = {
+      AArch64MCExpr::VK_AARCH64_ABS_G0,
+      AArch64MCExpr::VK_AARCH64_ABS_G1,
+      AArch64MCExpr::VK_AARCH64_SABS_G0,
+      AArch64MCExpr::VK_AARCH64_SABS_G1,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G1,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G0,
+      AArch64MCExpr::VK_AARCH64_GOTTPREL_G1,
+      AArch64MCExpr::VK_AARCH64_TPREL_G1,
+      AArch64MCExpr::VK_AARCH64_TPREL_G0,
+    };
+    unsigned NumModifiers = llvm::array_lengthof(PermittedModifiers);
+
+    return isMoveWideImm(32, PermittedModifiers, NumModifiers);
+  }
+
+  bool isMOVZ64Imm() const {
+    static AArch64MCExpr::VariantKind PermittedModifiers[] = {
+      AArch64MCExpr::VK_AARCH64_ABS_G0,
+      AArch64MCExpr::VK_AARCH64_ABS_G1,
+      AArch64MCExpr::VK_AARCH64_ABS_G2,
+      AArch64MCExpr::VK_AARCH64_ABS_G3,
+      AArch64MCExpr::VK_AARCH64_SABS_G0,
+      AArch64MCExpr::VK_AARCH64_SABS_G1,
+      AArch64MCExpr::VK_AARCH64_SABS_G2,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G2,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G1,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G0,
+      AArch64MCExpr::VK_AARCH64_GOTTPREL_G1,
+      AArch64MCExpr::VK_AARCH64_TPREL_G2,
+      AArch64MCExpr::VK_AARCH64_TPREL_G1,
+      AArch64MCExpr::VK_AARCH64_TPREL_G0,
+    };
+    unsigned NumModifiers = llvm::array_lengthof(PermittedModifiers);
+
+    return isMoveWideImm(64, PermittedModifiers, NumModifiers);
+  }
+
+  bool isMOVK32Imm() const {
+    static AArch64MCExpr::VariantKind PermittedModifiers[] = {
+      AArch64MCExpr::VK_AARCH64_ABS_G0_NC,
+      AArch64MCExpr::VK_AARCH64_ABS_G1_NC,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G1_NC,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G0_NC,
+      AArch64MCExpr::VK_AARCH64_GOTTPREL_G0_NC,
+      AArch64MCExpr::VK_AARCH64_TPREL_G1_NC,
+      AArch64MCExpr::VK_AARCH64_TPREL_G0_NC,
+    };
+    unsigned NumModifiers = llvm::array_lengthof(PermittedModifiers);
+
+    return isMoveWideImm(32, PermittedModifiers, NumModifiers);
+  }
+
+  bool isMOVK64Imm() const {
+    static AArch64MCExpr::VariantKind PermittedModifiers[] = {
+      AArch64MCExpr::VK_AARCH64_ABS_G0_NC,
+      AArch64MCExpr::VK_AARCH64_ABS_G1_NC,
+      AArch64MCExpr::VK_AARCH64_ABS_G2_NC,
+      AArch64MCExpr::VK_AARCH64_ABS_G3,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G1_NC,
+      AArch64MCExpr::VK_AARCH64_DTPREL_G0_NC,
+      AArch64MCExpr::VK_AARCH64_GOTTPREL_G0_NC,
+      AArch64MCExpr::VK_AARCH64_TPREL_G1_NC,
+      AArch64MCExpr::VK_AARCH64_TPREL_G0_NC,
+    };
+    unsigned NumModifiers = llvm::array_lengthof(PermittedModifiers);
+
+    return isMoveWideImm(64, PermittedModifiers, NumModifiers);
+  }
+
+  bool isMoveWideImm(unsigned RegWidth,
+                     AArch64MCExpr::VariantKind *PermittedModifiers,
+                     unsigned NumModifiers) const {
+    if (!isImmWithLSL()) return false;
+
+    if (ImmWithLSL.ShiftAmount % 16 != 0) return false;
+    if (ImmWithLSL.ShiftAmount >= RegWidth) return false;
+
+    AArch64MCExpr::VariantKind Modifier;
+    if (isNonConstantExpr(ImmWithLSL.Val, Modifier)) {
+      // E.g. "#:abs_g0:sym, lsl #16" makes no sense.
+      if (!ImmWithLSL.ImplicitAmount) return false;
+
+      for (unsigned i = 0; i < NumModifiers; ++i)
+        if (PermittedModifiers[i] == Modifier) return true;
+
+      return false;
+    }
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(ImmWithLSL.Val);
+    return CE && CE->getValue() >= 0  && CE->getValue() <= 0xffff;
+  }
+
+  template<int RegWidth, bool (*isValidImm)(int, uint64_t, int&, int&)>
+  bool isMoveWideMovAlias() const {
+    if (!isImm()) return false;
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+    if (!CE) return false;
+
+    int UImm16, Shift;
+    uint64_t Value = CE->getValue();
+
+    // If this is a 32-bit instruction then all bits above 32 should be the
+    // same: either of these is fine because signed/unsigned values should be
+    // permitted.
+    if (RegWidth == 32) {
+      if ((Value >> 32) != 0 && (Value >> 32) != 0xffffffff)
+        return false;
+
+      Value &= 0xffffffffULL;
+    }
+
+    return isValidImm(RegWidth, Value, UImm16, Shift);
+  }
+
+  bool isMSRWithReg() const {
+    if (!isSysReg()) return false;
+
+    bool IsKnownRegister;
+    StringRef Name(SysReg.Data, SysReg.Length);
+    A64SysReg::MSRMapper().fromString(Name, IsKnownRegister);
+
+    return IsKnownRegister;
+  }
+
+  bool isMSRPState() const {
+    if (!isSysReg()) return false;
+
+    bool IsKnownRegister;
+    StringRef Name(SysReg.Data, SysReg.Length);
+    A64PState::PStateMapper().fromString(Name, IsKnownRegister);
+
+    return IsKnownRegister;
+  }
+
+  bool isMRS() const {
+    if (!isSysReg()) return false;
+
+    // First check against specific MSR-only (write-only) registers
+    bool IsKnownRegister;
+    StringRef Name(SysReg.Data, SysReg.Length);
+    A64SysReg::MRSMapper().fromString(Name, IsKnownRegister);
+
+    return IsKnownRegister;
+  }
+
+  bool isPRFM() const {
+    if (!isImm()) return false;
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+
+    if (!CE)
+      return false;
+
+    return CE->getValue() >= 0 && CE->getValue() <= 31;
+  }
+
+  template<A64SE::ShiftExtSpecifiers SHKind> bool isRegExtend() const {
+    if (!isShiftOrExtend()) return false;
+
+    if (ShiftExtend.ShiftType != SHKind)
+      return false;
+
+    return ShiftExtend.Amount <= 4;
+  }
+
+  bool isRegExtendLSL() const {
+    if (!isShiftOrExtend()) return false;
+
+    if (ShiftExtend.ShiftType != A64SE::LSL)
+      return false;
+
+    return !ShiftExtend.ImplicitAmount && ShiftExtend.Amount <= 4;
+  }
+
+  template<int MemSize>  bool isSImm7Scaled() const {
+    if (!isImm()) return false;
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+    if (!CE) return false;
+
+    int64_t Val = CE->getValue();
+    if (Val % MemSize != 0) return false;
+
+    Val /= MemSize;
+
+    return Val >= -64 && Val < 64;
+  }
+
+  template<int BitWidth>
+  bool isSImm() const {
+    if (!isImm()) return false;
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+    if (!CE) return false;
+
+    return CE->getValue() >= -(1LL << (BitWidth - 1))
+      && CE->getValue() < (1LL << (BitWidth - 1));
+  }
+
+  template<int bitWidth>
+  bool isUImm() const {
+    if (!isImm()) return false;
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
+    if (!CE) return false;
+
+    return CE->getValue() >= 0 && CE->getValue() < (1LL << bitWidth);
+  }
+
+  bool isUImm() const {
+    if (!isImm()) return false;
+
+    return isa<MCConstantExpr>(getImm());
+  }
+
+  static AArch64Operand *CreateImmWithLSL(const MCExpr *Val,
+                                          unsigned ShiftAmount,
+                                          bool ImplicitAmount,
+                                          SMLoc S, SMLoc E) {
+    AArch64Operand *Op = new AArch64Operand(k_ImmWithLSL, S, E);
+    Op->ImmWithLSL.Val = Val;
+    Op->ImmWithLSL.ShiftAmount = ShiftAmount;
+    Op->ImmWithLSL.ImplicitAmount = ImplicitAmount;
+    return Op;
+  }
+
+  static AArch64Operand *CreateCondCode(A64CC::CondCodes Code,
+                                        SMLoc S, SMLoc E) {
+    AArch64Operand *Op = new AArch64Operand(k_CondCode, S, E);
+    Op->CondCode.Code = Code;
+    return Op;
+  }
+
+  static AArch64Operand *CreateFPImm(double Val,
+                                     SMLoc S, SMLoc E) {
+    AArch64Operand *Op = new AArch64Operand(k_FPImmediate, S, E);
+    Op->FPImm.Val = Val;
+    return Op;
+  }
+
+  static AArch64Operand *CreateImm(const MCExpr *Val, SMLoc S, SMLoc E) {
+    AArch64Operand *Op = new AArch64Operand(k_Immediate, S, E);
+    Op->Imm.Val = Val;
+    return Op;
+  }
+
+  static AArch64Operand *CreateReg(unsigned RegNum, SMLoc S, SMLoc E) {
+    AArch64Operand *Op = new AArch64Operand(k_Register, S, E);
+    Op->Reg.RegNum = RegNum;
+    return Op;
+  }
+
+  static AArch64Operand *CreateWrappedReg(unsigned RegNum, SMLoc S, SMLoc E) {
+    AArch64Operand *Op = new AArch64Operand(k_WrappedRegister, S, E);
+    Op->Reg.RegNum = RegNum;
+    return Op;
+  }
+
+  static AArch64Operand *CreateShiftExtend(A64SE::ShiftExtSpecifiers ShiftTyp,
+                                           unsigned Amount,
+                                           bool ImplicitAmount,
+                                           SMLoc S, SMLoc E) {
+    AArch64Operand *Op = new AArch64Operand(k_ShiftExtend, S, E);
+    Op->ShiftExtend.ShiftType = ShiftTyp;
+    Op->ShiftExtend.Amount = Amount;
+    Op->ShiftExtend.ImplicitAmount = ImplicitAmount;
+    return Op;
+  }
+
+  static AArch64Operand *CreateSysReg(StringRef Str, SMLoc S) {
+    AArch64Operand *Op = new AArch64Operand(k_SysReg, S, S);
+    Op->Tok.Data = Str.data();
+    Op->Tok.Length = Str.size();
+    return Op;
+  }
+
+  static AArch64Operand *CreateToken(StringRef Str, SMLoc S) {
+    AArch64Operand *Op = new AArch64Operand(k_Token, S, S);
+    Op->Tok.Data = Str.data();
+    Op->Tok.Length = Str.size();
+    return Op;
+  }
+
+
+  void addExpr(MCInst &Inst, const MCExpr *Expr) const {
+    // Add as immediates when possible.
+    if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Expr))
+      Inst.addOperand(MCOperand::CreateImm(CE->getValue()));
+    else
+      Inst.addOperand(MCOperand::CreateExpr(Expr));
+  }
+
+  template<unsigned RegWidth>
+  void addBFILSBOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    const MCConstantExpr *CE = cast<MCConstantExpr>(getImm());
+    unsigned EncodedVal = (RegWidth - CE->getValue()) % RegWidth;
+    Inst.addOperand(MCOperand::CreateImm(EncodedVal));
+  }
+
+  void addBFIWidthOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    const MCConstantExpr *CE = cast<MCConstantExpr>(getImm());
+    Inst.addOperand(MCOperand::CreateImm(CE->getValue() - 1));
+  }
+
+  void addBFXWidthOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    uint64_t LSB = Inst.getOperand(Inst.getNumOperands()-1).getImm();
+    const MCConstantExpr *CE = cast<MCConstantExpr>(getImm());
+
+    Inst.addOperand(MCOperand::CreateImm(LSB + CE->getValue() - 1));
+  }
+
+  void addCondCodeOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    Inst.addOperand(MCOperand::CreateImm(getCondCode()));
+  }
+
+  void addCVTFixedPosOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    const MCConstantExpr *CE = cast<MCConstantExpr>(getImm());
+    Inst.addOperand(MCOperand::CreateImm(64 - CE->getValue()));
+  }
+
+  void addFMOVImmOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    APFloat RealVal(FPImm.Val);
+    uint32_t ImmVal;
+    A64Imms::isFPImm(RealVal, ImmVal);
+
+    Inst.addOperand(MCOperand::CreateImm(ImmVal));
+  }
+
+  void addFPZeroOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands");
+    Inst.addOperand(MCOperand::CreateImm(0));
+  }
+
+  void addInvCondCodeOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    unsigned Encoded = A64InvertCondCode(getCondCode());
+    Inst.addOperand(MCOperand::CreateImm(Encoded));
+  }
+
+  void addRegOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    Inst.addOperand(MCOperand::CreateReg(getReg()));
+  }
+
+  void addImmOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    addExpr(Inst, getImm());
+  }
+
+  template<int MemSize>
+  void addSImm7ScaledOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    const MCConstantExpr *CE = cast<MCConstantExpr>(getImm());
+    uint64_t Val = CE->getValue() / MemSize;
+    Inst.addOperand(MCOperand::CreateImm(Val  & 0x7f));
+  }
+
+  template<int BitWidth>
+  void addSImmOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    const MCConstantExpr *CE = cast<MCConstantExpr>(getImm());
+    uint64_t Val = CE->getValue();
+    Inst.addOperand(MCOperand::CreateImm(Val  & ((1ULL << BitWidth) - 1)));
+  }
+
+  void addImmWithLSLOperands(MCInst &Inst, unsigned N) const {
+    assert (N == 1 && "Invalid number of operands!");
+
+    addExpr(Inst, ImmWithLSL.Val);
+  }
+
+  template<unsigned field_width, unsigned scale>
+  void addLabelOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Imm.Val);
+
+    if (!CE) {
+      addExpr(Inst, Imm.Val);
+      return;
+    }
+
+    int64_t Val = CE->getValue();
+    assert(Val % scale == 0 && "Unaligned immediate in instruction");
+    Val /= scale;
+
+    Inst.addOperand(MCOperand::CreateImm(Val & ((1LL << field_width) - 1)));
+  }
+
+  template<int MemSize>
+  void addOffsetUImm12Operands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm())) {
+      Inst.addOperand(MCOperand::CreateImm(CE->getValue() / MemSize));
+    } else {
+      Inst.addOperand(MCOperand::CreateExpr(getImm()));
+    }
+  }
+
+  template<unsigned RegWidth>
+  void addLogicalImmOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands");
+    const MCConstantExpr *CE = cast<MCConstantExpr>(Imm.Val);
+
+    uint32_t Bits;
+    A64Imms::isLogicalImm(RegWidth, CE->getValue(), Bits);
+
+    Inst.addOperand(MCOperand::CreateImm(Bits));
+  }
+
+  void addMRSOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    bool Valid;
+    StringRef Name(SysReg.Data, SysReg.Length);
+    uint32_t Bits = A64SysReg::MRSMapper().fromString(Name, Valid);
+
+    Inst.addOperand(MCOperand::CreateImm(Bits));
+  }
+
+  void addMSRWithRegOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    bool Valid;
+    StringRef Name(SysReg.Data, SysReg.Length);
+    uint32_t Bits = A64SysReg::MSRMapper().fromString(Name, Valid);
+
+    Inst.addOperand(MCOperand::CreateImm(Bits));
+  }
+
+  void addMSRPStateOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    bool Valid;
+    StringRef Name(SysReg.Data, SysReg.Length);
+    uint32_t Bits = A64PState::PStateMapper().fromString(Name, Valid);
+
+    Inst.addOperand(MCOperand::CreateImm(Bits));
+  }
+
+  void addMoveWideImmOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 2 && "Invalid number of operands!");
+
+    addExpr(Inst, ImmWithLSL.Val);
+
+    AArch64MCExpr::VariantKind Variant;
+    if (!isNonConstantExpr(ImmWithLSL.Val, Variant)) {
+      Inst.addOperand(MCOperand::CreateImm(ImmWithLSL.ShiftAmount / 16));
+      return;
+    }
+
+    // We know it's relocated
+    switch (Variant) {
+    case AArch64MCExpr::VK_AARCH64_ABS_G0:
+    case AArch64MCExpr::VK_AARCH64_ABS_G0_NC:
+    case AArch64MCExpr::VK_AARCH64_SABS_G0:
+    case AArch64MCExpr::VK_AARCH64_DTPREL_G0:
+    case AArch64MCExpr::VK_AARCH64_DTPREL_G0_NC:
+    case AArch64MCExpr::VK_AARCH64_GOTTPREL_G0_NC:
+    case AArch64MCExpr::VK_AARCH64_TPREL_G0:
+    case AArch64MCExpr::VK_AARCH64_TPREL_G0_NC:
+      Inst.addOperand(MCOperand::CreateImm(0));
+      break;
+    case AArch64MCExpr::VK_AARCH64_ABS_G1:
+    case AArch64MCExpr::VK_AARCH64_ABS_G1_NC:
+    case AArch64MCExpr::VK_AARCH64_SABS_G1:
+    case AArch64MCExpr::VK_AARCH64_DTPREL_G1:
+    case AArch64MCExpr::VK_AARCH64_DTPREL_G1_NC:
+    case AArch64MCExpr::VK_AARCH64_GOTTPREL_G1:
+    case AArch64MCExpr::VK_AARCH64_TPREL_G1:
+    case AArch64MCExpr::VK_AARCH64_TPREL_G1_NC:
+      Inst.addOperand(MCOperand::CreateImm(1));
+      break;
+    case AArch64MCExpr::VK_AARCH64_ABS_G2:
+    case AArch64MCExpr::VK_AARCH64_ABS_G2_NC:
+    case AArch64MCExpr::VK_AARCH64_SABS_G2:
+    case AArch64MCExpr::VK_AARCH64_DTPREL_G2:
+    case AArch64MCExpr::VK_AARCH64_TPREL_G2:
+      Inst.addOperand(MCOperand::CreateImm(2));
+      break;
+    case AArch64MCExpr::VK_AARCH64_ABS_G3:
+      Inst.addOperand(MCOperand::CreateImm(3));
+      break;
+    default: llvm_unreachable("Inappropriate move wide relocation");
+    }
+  }
+
+  template<int RegWidth, bool isValidImm(int, uint64_t, int&, int&)>
+  void addMoveWideMovAliasOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 2 && "Invalid number of operands!");
+    int UImm16, Shift;
+
+    const MCConstantExpr *CE = cast<MCConstantExpr>(getImm());
+    uint64_t Value = CE->getValue();
+
+    if (RegWidth == 32) {
+      Value &= 0xffffffffULL;
+    }
+
+    bool Valid = isValidImm(RegWidth, Value, UImm16, Shift);
+    (void)Valid;
+    assert(Valid && "Invalid immediates should have been weeded out by now");
+
+    Inst.addOperand(MCOperand::CreateImm(UImm16));
+    Inst.addOperand(MCOperand::CreateImm(Shift));
+  }
+
+  void addPRFMOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    const MCConstantExpr *CE = cast<MCConstantExpr>(getImm());
+    assert(CE->getValue() >= 0 && CE->getValue() <= 31
+           && "PRFM operand should be 5-bits");
+
+    Inst.addOperand(MCOperand::CreateImm(CE->getValue()));
+  }
+
+  // For Add-sub (extended register) operands.
+  void addRegExtendOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    Inst.addOperand(MCOperand::CreateImm(ShiftExtend.Amount));
+  }
+
+  // For the extend in load-store (register offset) instructions.
+  template<unsigned MemSize>
+  void addAddrRegExtendOperands(MCInst &Inst, unsigned N) const {
+    addAddrRegExtendOperands(Inst, N, MemSize);
+  }
+
+  void addAddrRegExtendOperands(MCInst &Inst, unsigned N,
+                                unsigned MemSize) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    // First bit of Option is set in instruction classes, the high two bits are
+    // as follows:
+    unsigned OptionHi = 0;
+    switch (ShiftExtend.ShiftType) {
+    case A64SE::UXTW:
+    case A64SE::LSL:
+      OptionHi = 1;
+      break;
+    case A64SE::SXTW:
+    case A64SE::SXTX:
+      OptionHi = 3;
+      break;
+    default:
+      llvm_unreachable("Invalid extend type for register offset");
+    }
+
+    unsigned S = 0;
+    if (MemSize == 1 && !ShiftExtend.ImplicitAmount)
+      S = 1;
+    else if (MemSize != 1 && ShiftExtend.Amount != 0)
+      S = 1;
+
+    Inst.addOperand(MCOperand::CreateImm((OptionHi << 1) | S));
+  }
+  void addShiftOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+
+    Inst.addOperand(MCOperand::CreateImm(ShiftExtend.Amount));
+  }
+};
+
+} // end anonymous namespace.
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
+                               StringRef Mnemonic) {
+
+  // See if the operand has a custom parser
+  OperandMatchResultTy ResTy = MatchOperandParserImpl(Operands, Mnemonic);
+
+  // It could either succeed, fail or just not care.
+  if (ResTy != MatchOperand_NoMatch)
+    return ResTy;
+
+  switch (getLexer().getKind()) {
+  default:
+    Error(Parser.getTok().getLoc(), "unexpected token in operand");
+    return MatchOperand_ParseFail;
+  case AsmToken::Identifier: {
+    // It might be in the LSL/UXTB family ...
+    OperandMatchResultTy GotShift = ParseShiftExtend(Operands);
+
+    // We can only continue if no tokens were eaten.
+    if (GotShift != MatchOperand_NoMatch)
+      return GotShift;
+
+    // ... or it might be a register ...
+    uint32_t NumLanes = 0;
+    OperandMatchResultTy GotReg = ParseRegister(Operands, NumLanes);
+    assert(GotReg != MatchOperand_ParseFail
+           && "register parsing shouldn't partially succeed");
+
+    if (GotReg == MatchOperand_Success) {
+      if (Parser.getTok().is(AsmToken::LBrac))
+        return ParseNEONLane(Operands, NumLanes);
+      else
+        return MatchOperand_Success;
+    }
+
+    // ... or it might be a symbolish thing
+  }
+    // Fall through
+  case AsmToken::LParen:  // E.g. (strcmp-4)
+  case AsmToken::Integer: // 1f, 2b labels
+  case AsmToken::String:  // quoted labels
+  case AsmToken::Dot:     // . is Current location
+  case AsmToken::Dollar:  // $ is PC
+  case AsmToken::Colon: {
+    SMLoc StartLoc  = Parser.getTok().getLoc();
+    SMLoc EndLoc;
+    const MCExpr *ImmVal = 0;
+
+    if (ParseImmediate(ImmVal) != MatchOperand_Success)
+      return MatchOperand_ParseFail;
+
+    EndLoc = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1);
+    Operands.push_back(AArch64Operand::CreateImm(ImmVal, StartLoc, EndLoc));
+    return MatchOperand_Success;
+  }
+  case AsmToken::Hash: {   // Immediates
+    SMLoc StartLoc = Parser.getTok().getLoc();
+    SMLoc EndLoc;
+    const MCExpr *ImmVal = 0;
+    Parser.Lex();
+
+    if (ParseImmediate(ImmVal) != MatchOperand_Success)
+      return MatchOperand_ParseFail;
+
+    EndLoc = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1);
+    Operands.push_back(AArch64Operand::CreateImm(ImmVal, StartLoc, EndLoc));
+    return MatchOperand_Success;
+  }
+  case AsmToken::LBrac: {
+    SMLoc Loc = Parser.getTok().getLoc();
+    Operands.push_back(AArch64Operand::CreateToken("[", Loc));
+    Parser.Lex(); // Eat '['
+
+    // There's no comma after a '[', so we can parse the next operand
+    // immediately.
+    return ParseOperand(Operands, Mnemonic);
+  }
+  // The following will likely be useful later, but not in very early cases
+  case AsmToken::LCurly:  // Weird SIMD lists
+    llvm_unreachable("Don't know how to deal with '{' in operand");
+    return MatchOperand_ParseFail;
+  }
+}
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseImmediate(const MCExpr *&ExprVal) {
+  if (getLexer().is(AsmToken::Colon)) {
+    AArch64MCExpr::VariantKind RefKind;
+
+    OperandMatchResultTy ResTy = ParseRelocPrefix(RefKind);
+    if (ResTy != MatchOperand_Success)
+      return ResTy;
+
+    const MCExpr *SubExprVal;
+    if (getParser().parseExpression(SubExprVal))
+      return MatchOperand_ParseFail;
+
+    ExprVal = AArch64MCExpr::Create(RefKind, SubExprVal, getContext());
+    return MatchOperand_Success;
+  }
+
+  // No weird AArch64MCExpr prefix
+  return getParser().parseExpression(ExprVal)
+    ? MatchOperand_ParseFail : MatchOperand_Success;
+}
+
+// A lane attached to a NEON register. "[N]", which should yield three tokens:
+// '[', N, ']'. A hash is not allowed to precede the immediate here.
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseNEONLane(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
+                                uint32_t NumLanes) {
+  SMLoc Loc = Parser.getTok().getLoc();
+
+  assert(Parser.getTok().is(AsmToken::LBrac) && "inappropriate operand");
+  Operands.push_back(AArch64Operand::CreateToken("[", Loc));
+  Parser.Lex(); // Eat '['
+
+  if (Parser.getTok().isNot(AsmToken::Integer)) {
+    Error(Parser.getTok().getLoc(), "expected lane number");
+    return MatchOperand_ParseFail;
+  }
+
+  if (Parser.getTok().getIntVal() >= NumLanes) {
+    Error(Parser.getTok().getLoc(), "lane number incompatible with layout");
+    return MatchOperand_ParseFail;
+  }
+
+  const MCExpr *Lane = MCConstantExpr::Create(Parser.getTok().getIntVal(),
+                                              getContext());
+  SMLoc S = Parser.getTok().getLoc();
+  Parser.Lex(); // Eat actual lane
+  SMLoc E = Parser.getTok().getLoc();
+  Operands.push_back(AArch64Operand::CreateImm(Lane, S, E));
+
+
+  if (Parser.getTok().isNot(AsmToken::RBrac)) {
+    Error(Parser.getTok().getLoc(), "expected ']' after lane");
+    return MatchOperand_ParseFail;
+  }
+
+  Operands.push_back(AArch64Operand::CreateToken("]", Loc));
+  Parser.Lex(); // Eat ']'
+
+  return MatchOperand_Success;
+}
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseRelocPrefix(AArch64MCExpr::VariantKind &RefKind) {
+  assert(getLexer().is(AsmToken::Colon) && "expected a ':'");
+  Parser.Lex();
+
+  if (getLexer().isNot(AsmToken::Identifier)) {
+    Error(Parser.getTok().getLoc(),
+          "expected relocation specifier in operand after ':'");
+    return MatchOperand_ParseFail;
+  }
+
+  std::string LowerCase = Parser.getTok().getIdentifier().lower();
+  RefKind = StringSwitch<AArch64MCExpr::VariantKind>(LowerCase)
+    .Case("got",              AArch64MCExpr::VK_AARCH64_GOT)
+    .Case("got_lo12",         AArch64MCExpr::VK_AARCH64_GOT_LO12)
+    .Case("lo12",             AArch64MCExpr::VK_AARCH64_LO12)
+    .Case("abs_g0",           AArch64MCExpr::VK_AARCH64_ABS_G0)
+    .Case("abs_g0_nc",        AArch64MCExpr::VK_AARCH64_ABS_G0_NC)
+    .Case("abs_g1",           AArch64MCExpr::VK_AARCH64_ABS_G1)
+    .Case("abs_g1_nc",        AArch64MCExpr::VK_AARCH64_ABS_G1_NC)
+    .Case("abs_g2",           AArch64MCExpr::VK_AARCH64_ABS_G2)
+    .Case("abs_g2_nc",        AArch64MCExpr::VK_AARCH64_ABS_G2_NC)
+    .Case("abs_g3",           AArch64MCExpr::VK_AARCH64_ABS_G3)
+    .Case("abs_g0_s",         AArch64MCExpr::VK_AARCH64_SABS_G0)
+    .Case("abs_g1_s",         AArch64MCExpr::VK_AARCH64_SABS_G1)
+    .Case("abs_g2_s",         AArch64MCExpr::VK_AARCH64_SABS_G2)
+    .Case("dtprel_g2",        AArch64MCExpr::VK_AARCH64_DTPREL_G2)
+    .Case("dtprel_g1",        AArch64MCExpr::VK_AARCH64_DTPREL_G1)
+    .Case("dtprel_g1_nc",     AArch64MCExpr::VK_AARCH64_DTPREL_G1_NC)
+    .Case("dtprel_g0",        AArch64MCExpr::VK_AARCH64_DTPREL_G0)
+    .Case("dtprel_g0_nc",     AArch64MCExpr::VK_AARCH64_DTPREL_G0_NC)
+    .Case("dtprel_hi12",      AArch64MCExpr::VK_AARCH64_DTPREL_HI12)
+    .Case("dtprel_lo12",      AArch64MCExpr::VK_AARCH64_DTPREL_LO12)
+    .Case("dtprel_lo12_nc",   AArch64MCExpr::VK_AARCH64_DTPREL_LO12_NC)
+    .Case("gottprel_g1",      AArch64MCExpr::VK_AARCH64_GOTTPREL_G1)
+    .Case("gottprel_g0_nc",   AArch64MCExpr::VK_AARCH64_GOTTPREL_G0_NC)
+    .Case("gottprel",         AArch64MCExpr::VK_AARCH64_GOTTPREL)
+    .Case("gottprel_lo12",    AArch64MCExpr::VK_AARCH64_GOTTPREL_LO12)
+    .Case("tprel_g2",         AArch64MCExpr::VK_AARCH64_TPREL_G2)
+    .Case("tprel_g1",         AArch64MCExpr::VK_AARCH64_TPREL_G1)
+    .Case("tprel_g1_nc",      AArch64MCExpr::VK_AARCH64_TPREL_G1_NC)
+    .Case("tprel_g0",         AArch64MCExpr::VK_AARCH64_TPREL_G0)
+    .Case("tprel_g0_nc",      AArch64MCExpr::VK_AARCH64_TPREL_G0_NC)
+    .Case("tprel_hi12",       AArch64MCExpr::VK_AARCH64_TPREL_HI12)
+    .Case("tprel_lo12",       AArch64MCExpr::VK_AARCH64_TPREL_LO12)
+    .Case("tprel_lo12_nc",    AArch64MCExpr::VK_AARCH64_TPREL_LO12_NC)
+    .Case("tlsdesc",          AArch64MCExpr::VK_AARCH64_TLSDESC)
+    .Case("tlsdesc_lo12",     AArch64MCExpr::VK_AARCH64_TLSDESC_LO12)
+    .Default(AArch64MCExpr::VK_AARCH64_None);
+
+  if (RefKind == AArch64MCExpr::VK_AARCH64_None) {
+    Error(Parser.getTok().getLoc(),
+          "expected relocation specifier in operand after ':'");
+    return MatchOperand_ParseFail;
+  }
+  Parser.Lex(); // Eat identifier
+
+  if (getLexer().isNot(AsmToken::Colon)) {
+    Error(Parser.getTok().getLoc(),
+          "expected ':' after relocation specifier");
+    return MatchOperand_ParseFail;
+  }
+  Parser.Lex();
+  return MatchOperand_Success;
+}
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseImmWithLSLOperand(
+                               SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  // FIXME?: I want to live in a world where immediates must start with
+  // #. Please don't dash my hopes (well, do if you have a good reason).
+  if (Parser.getTok().isNot(AsmToken::Hash)) return MatchOperand_NoMatch;
+
+  SMLoc S = Parser.getTok().getLoc();
+  Parser.Lex(); // Eat '#'
+
+  const MCExpr *Imm;
+  if (ParseImmediate(Imm) != MatchOperand_Success)
+    return MatchOperand_ParseFail;
+  else if (Parser.getTok().isNot(AsmToken::Comma)) {
+    SMLoc E = Parser.getTok().getLoc();
+    Operands.push_back(AArch64Operand::CreateImmWithLSL(Imm, 0, true, S, E));
+    return MatchOperand_Success;
+  }
+
+  // Eat ','
+  Parser.Lex();
+
+  // The optional operand must be "lsl #N" where N is non-negative.
+  if (Parser.getTok().is(AsmToken::Identifier)
+      && Parser.getTok().getIdentifier().lower() == "lsl") {
+    Parser.Lex();
+
+    if (Parser.getTok().is(AsmToken::Hash)) {
+      Parser.Lex();
+
+      if (Parser.getTok().isNot(AsmToken::Integer)) {
+        Error(Parser.getTok().getLoc(), "only 'lsl #+N' valid after immediate");
+        return MatchOperand_ParseFail;
+      }
+    }
+  }
+
+  int64_t ShiftAmount = Parser.getTok().getIntVal();
+
+  if (ShiftAmount < 0) {
+    Error(Parser.getTok().getLoc(), "positive shift amount required");
+    return MatchOperand_ParseFail;
+  }
+  Parser.Lex(); // Eat the number
+
+  SMLoc E = Parser.getTok().getLoc();
+  Operands.push_back(AArch64Operand::CreateImmWithLSL(Imm, ShiftAmount,
+                                                      false, S, E));
+  return MatchOperand_Success;
+}
+
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseCondCodeOperand(
+                               SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  if (Parser.getTok().isNot(AsmToken::Identifier))
+    return MatchOperand_NoMatch;
+
+  StringRef Tok = Parser.getTok().getIdentifier();
+  A64CC::CondCodes CondCode = A64StringToCondCode(Tok);
+
+  if (CondCode == A64CC::Invalid)
+    return MatchOperand_NoMatch;
+
+  SMLoc S = Parser.getTok().getLoc();
+  Parser.Lex(); // Eat condition code
+  SMLoc E = Parser.getTok().getLoc();
+
+  Operands.push_back(AArch64Operand::CreateCondCode(CondCode, S, E));
+  return MatchOperand_Success;
+}
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseCRxOperand(
+                               SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  SMLoc S = Parser.getTok().getLoc();
+  if (Parser.getTok().isNot(AsmToken::Identifier)) {
+    Error(S, "Expected cN operand where 0 <= N <= 15");
+    return MatchOperand_ParseFail;
+  }
+
+  std::string LowerTok = Parser.getTok().getIdentifier().lower();
+  StringRef Tok(LowerTok);
+  if (Tok[0] != 'c') {
+    Error(S, "Expected cN operand where 0 <= N <= 15");
+    return MatchOperand_ParseFail;
+  }
+
+  uint32_t CRNum;
+  bool BadNum = Tok.drop_front().getAsInteger(10, CRNum);
+  if (BadNum || CRNum > 15) {
+    Error(S, "Expected cN operand where 0 <= N <= 15");
+    return MatchOperand_ParseFail;
+  }
+
+  const MCExpr *CRImm = MCConstantExpr::Create(CRNum, getContext());
+
+  Parser.Lex();
+  SMLoc E = Parser.getTok().getLoc();
+
+  Operands.push_back(AArch64Operand::CreateImm(CRImm, S, E));
+  return MatchOperand_Success;
+}
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseFPImmOperand(
+                               SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+
+  // FIXME?: I want to live in a world where immediates must start with
+  // #. Please don't dash my hopes (well, do if you have a good reason).
+  if (Parser.getTok().isNot(AsmToken::Hash)) return MatchOperand_NoMatch;
+
+  SMLoc S = Parser.getTok().getLoc();
+  Parser.Lex(); // Eat '#'
+
+  bool Negative = false;
+  if (Parser.getTok().is(AsmToken::Minus)) {
+    Negative = true;
+    Parser.Lex(); // Eat '-'
+  } else if (Parser.getTok().is(AsmToken::Plus)) {
+    Parser.Lex(); // Eat '+'
+  }
+
+  if (Parser.getTok().isNot(AsmToken::Real)) {
+    Error(S, "Expected floating-point immediate");
+    return MatchOperand_ParseFail;
+  }
+
+  APFloat RealVal(APFloat::IEEEdouble, Parser.getTok().getString());
+  if (Negative) RealVal.changeSign();
+  double DblVal = RealVal.convertToDouble();
+
+  Parser.Lex(); // Eat real number
+  SMLoc E = Parser.getTok().getLoc();
+
+  Operands.push_back(AArch64Operand::CreateFPImm(DblVal, S, E));
+  return MatchOperand_Success;
+}
+
+
+// Automatically generated
+static unsigned MatchRegisterName(StringRef Name);
+
+bool
+AArch64AsmParser::IdentifyRegister(unsigned &RegNum, SMLoc &RegEndLoc,
+                                   StringRef &Layout,
+                                   SMLoc &LayoutLoc) const {
+  const AsmToken &Tok = Parser.getTok();
+
+  if (Tok.isNot(AsmToken::Identifier))
+    return false;
+
+  std::string LowerReg = Tok.getString().lower();
+  size_t DotPos = LowerReg.find('.');
+
+  RegNum = MatchRegisterName(LowerReg.substr(0, DotPos));
+  if (RegNum == AArch64::NoRegister) {
+    RegNum = StringSwitch<unsigned>(LowerReg.substr(0, DotPos))
+      .Case("ip0", AArch64::X16)
+      .Case("ip1", AArch64::X17)
+      .Case("fp", AArch64::X29)
+      .Case("lr", AArch64::X30)
+      .Default(AArch64::NoRegister);
+  }
+  if (RegNum == AArch64::NoRegister)
+    return false;
+
+  SMLoc S = Tok.getLoc();
+  RegEndLoc = SMLoc::getFromPointer(S.getPointer() + DotPos);
+
+  if (DotPos == StringRef::npos) {
+    Layout = StringRef();
+  } else {
+    // Everything afterwards needs to be a literal token, expected to be
+    // '.2d','.b' etc for vector registers.
+
+    // This StringSwitch validates the input and (perhaps more importantly)
+    // gives us a permanent string to use in the token (a pointer into LowerReg
+    // would go out of scope when we return).
+    LayoutLoc = SMLoc::getFromPointer(S.getPointer() + DotPos + 1);
+    std::string LayoutText = LowerReg.substr(DotPos, StringRef::npos);
+    Layout = StringSwitch<const char *>(LayoutText)
+      .Case(".d", ".d").Case(".1d", ".1d").Case(".2d", ".2d")
+      .Case(".s", ".s").Case(".2s", ".2s").Case(".4s", ".4s")
+      .Case(".h", ".h").Case(".4h", ".4h").Case(".8h", ".8h")
+      .Case(".b", ".b").Case(".8b", ".8b").Case(".16b", ".16b")
+      .Default("");
+
+    if (Layout.size() == 0) {
+      // Malformed register
+      return false;
+    }
+  }
+
+  return true;
+}
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseRegister(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
+                                uint32_t &NumLanes) {
+  unsigned RegNum;
+  StringRef Layout;
+  SMLoc RegEndLoc, LayoutLoc;
+  SMLoc S = Parser.getTok().getLoc();
+
+  if (!IdentifyRegister(RegNum, RegEndLoc, Layout, LayoutLoc))
+    return MatchOperand_NoMatch;
+
+  Operands.push_back(AArch64Operand::CreateReg(RegNum, S, RegEndLoc));
+
+  if (Layout.size() != 0) {
+    unsigned long long TmpLanes = 0;
+    llvm::getAsUnsignedInteger(Layout.substr(1), 10, TmpLanes);
+    if (TmpLanes != 0) {
+      NumLanes = TmpLanes;
+    } else {
+      // If the number of lanes isn't specified explicitly, a valid instruction
+      // will have an element specifier and be capable of acting on the entire
+      // vector register.
+      switch (Layout.back()) {
+      default: llvm_unreachable("Invalid layout specifier");
+      case 'b': NumLanes = 16; break;
+      case 'h': NumLanes = 8; break;
+      case 's': NumLanes = 4; break;
+      case 'd': NumLanes = 2; break;
+      }
+    }
+
+    Operands.push_back(AArch64Operand::CreateToken(Layout, LayoutLoc));
+  }
+
+  Parser.Lex();
+  return MatchOperand_Success;
+}
+
+bool
+AArch64AsmParser::ParseRegister(unsigned &RegNo, SMLoc &StartLoc,
+                                SMLoc &EndLoc) {
+  // This callback is used for things like DWARF frame directives in
+  // assembly. They don't care about things like NEON layouts or lanes, they
+  // just want to be able to produce the DWARF register number.
+  StringRef LayoutSpec;
+  SMLoc RegEndLoc, LayoutLoc;
+  StartLoc = Parser.getTok().getLoc();
+
+  if (!IdentifyRegister(RegNo, RegEndLoc, LayoutSpec, LayoutLoc))
+    return true;
+
+  Parser.Lex();
+  EndLoc = Parser.getTok().getLoc();
+
+  return false;
+}
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseNamedImmOperand(const NamedImmMapper &Mapper,
+                               SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  // Since these operands occur in very limited circumstances, without
+  // alternatives, we actually signal an error if there is no match. If relaxing
+  // this, beware of unintended consequences: an immediate will be accepted
+  // during matching, no matter how it gets into the AArch64Operand.
+  const AsmToken &Tok = Parser.getTok();
+  SMLoc S = Tok.getLoc();
+
+  if (Tok.is(AsmToken::Identifier)) {
+    bool ValidName;
+    uint32_t Code = Mapper.fromString(Tok.getString().lower(), ValidName);
+
+    if (!ValidName) {
+      Error(S, "operand specifier not recognised");
+      return MatchOperand_ParseFail;
+    }
+
+    Parser.Lex(); // We're done with the identifier. Eat it
+
+    SMLoc E = Parser.getTok().getLoc();
+    const MCExpr *Imm = MCConstantExpr::Create(Code, getContext());
+    Operands.push_back(AArch64Operand::CreateImm(Imm, S, E));
+    return MatchOperand_Success;
+  } else if (Tok.is(AsmToken::Hash)) {
+    Parser.Lex();
+
+    const MCExpr *ImmVal;
+    if (ParseImmediate(ImmVal) != MatchOperand_Success)
+      return MatchOperand_ParseFail;
+
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(ImmVal);
+    if (!CE || CE->getValue() < 0 || !Mapper.validImm(CE->getValue())) {
+      Error(S, "Invalid immediate for instruction");
+      return MatchOperand_ParseFail;
+    }
+
+    SMLoc E = Parser.getTok().getLoc();
+    Operands.push_back(AArch64Operand::CreateImm(ImmVal, S, E));
+    return MatchOperand_Success;
+  }
+
+  Error(S, "unexpected operand for instruction");
+  return MatchOperand_ParseFail;
+}
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseSysRegOperand(
+                               SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  const AsmToken &Tok = Parser.getTok();
+
+  // Any MSR/MRS operand will be an identifier, and we want to store it as some
+  // kind of string: SPSel is valid for two different forms of MSR with two
+  // different encodings. There's no collision at the moment, but the potential
+  // is there.
+  if (!Tok.is(AsmToken::Identifier)) {
+    return MatchOperand_NoMatch;
+  }
+
+  SMLoc S = Tok.getLoc();
+  Operands.push_back(AArch64Operand::CreateSysReg(Tok.getString(), S));
+  Parser.Lex(); // Eat identifier
+
+  return MatchOperand_Success;
+}
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseLSXAddressOperand(
+                               SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  SMLoc S = Parser.getTok().getLoc();
+
+  unsigned RegNum;
+  SMLoc RegEndLoc, LayoutLoc;
+  StringRef Layout;
+  if(!IdentifyRegister(RegNum, RegEndLoc, Layout, LayoutLoc)
+     || !AArch64MCRegisterClasses[AArch64::GPR64xspRegClassID].contains(RegNum)
+     || Layout.size() != 0) {
+    // Check Layout.size because we don't want to let "x3.4s" or similar
+    // through.
+    return MatchOperand_NoMatch;
+  }
+  Parser.Lex(); // Eat register
+
+  if (Parser.getTok().is(AsmToken::RBrac)) {
+    // We're done
+    SMLoc E = Parser.getTok().getLoc();
+    Operands.push_back(AArch64Operand::CreateWrappedReg(RegNum, S, E));
+    return MatchOperand_Success;
+  }
+
+  // Otherwise, only ", #0" is valid
+
+  if (Parser.getTok().isNot(AsmToken::Comma)) {
+    Error(Parser.getTok().getLoc(), "expected ',' or ']' after register");
+    return MatchOperand_ParseFail;
+  }
+  Parser.Lex(); // Eat ','
+
+  if (Parser.getTok().isNot(AsmToken::Hash)) {
+    Error(Parser.getTok().getLoc(), "expected '#0'");
+    return MatchOperand_ParseFail;
+  }
+  Parser.Lex(); // Eat '#'
+
+  if (Parser.getTok().isNot(AsmToken::Integer)
+      || Parser.getTok().getIntVal() != 0 ) {
+    Error(Parser.getTok().getLoc(), "expected '#0'");
+    return MatchOperand_ParseFail;
+  }
+  Parser.Lex(); // Eat '0'
+
+  SMLoc E = Parser.getTok().getLoc();
+  Operands.push_back(AArch64Operand::CreateWrappedReg(RegNum, S, E));
+  return MatchOperand_Success;
+}
+
+AArch64AsmParser::OperandMatchResultTy
+AArch64AsmParser::ParseShiftExtend(
+                               SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  StringRef IDVal = Parser.getTok().getIdentifier();
+  std::string LowerID = IDVal.lower();
+
+  A64SE::ShiftExtSpecifiers Spec =
+    StringSwitch<A64SE::ShiftExtSpecifiers>(LowerID)
+      .Case("lsl", A64SE::LSL)
+      .Case("lsr", A64SE::LSR)
+      .Case("asr", A64SE::ASR)
+      .Case("ror", A64SE::ROR)
+      .Case("uxtb", A64SE::UXTB)
+      .Case("uxth", A64SE::UXTH)
+      .Case("uxtw", A64SE::UXTW)
+      .Case("uxtx", A64SE::UXTX)
+      .Case("sxtb", A64SE::SXTB)
+      .Case("sxth", A64SE::SXTH)
+      .Case("sxtw", A64SE::SXTW)
+      .Case("sxtx", A64SE::SXTX)
+      .Default(A64SE::Invalid);
+
+  if (Spec == A64SE::Invalid)
+    return MatchOperand_NoMatch;
+
+  // Eat the shift
+  SMLoc S, E;
+  S = Parser.getTok().getLoc();
+  Parser.Lex();
+
+  if (Spec != A64SE::LSL && Spec != A64SE::LSR &&
+      Spec != A64SE::ASR && Spec != A64SE::ROR) {
+    // The shift amount can be omitted for the extending versions, but not real
+    // shifts:
+    //     add x0, x0, x0, uxtb
+    // is valid, and equivalent to
+    //     add x0, x0, x0, uxtb #0
+
+    if (Parser.getTok().is(AsmToken::Comma) ||
+        Parser.getTok().is(AsmToken::EndOfStatement) ||
+        Parser.getTok().is(AsmToken::RBrac)) {
+      Operands.push_back(AArch64Operand::CreateShiftExtend(Spec, 0, true,
+                                                           S, E));
+      return MatchOperand_Success;
+    }
+  }
+
+  // Eat # at beginning of immediate
+  if (!Parser.getTok().is(AsmToken::Hash)) {
+    Error(Parser.getTok().getLoc(),
+          "expected #imm after shift specifier");
+    return MatchOperand_ParseFail;
+  }
+  Parser.Lex();
+
+  // Make sure we do actually have a number
+  if (!Parser.getTok().is(AsmToken::Integer)) {
+    Error(Parser.getTok().getLoc(),
+          "expected integer shift amount");
+    return MatchOperand_ParseFail;
+  }
+  unsigned Amount = Parser.getTok().getIntVal();
+  Parser.Lex();
+  E = Parser.getTok().getLoc();
+
+  Operands.push_back(AArch64Operand::CreateShiftExtend(Spec, Amount, false,
+                                                       S, E));
+
+  return MatchOperand_Success;
+}
+
+// FIXME: We would really like to be able to tablegen'erate this.
+bool AArch64AsmParser::
+validateInstruction(MCInst &Inst,
+                    const SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  switch (Inst.getOpcode()) {
+  case AArch64::BFIwwii:
+  case AArch64::BFIxxii:
+  case AArch64::SBFIZwwii:
+  case AArch64::SBFIZxxii:
+  case AArch64::UBFIZwwii:
+  case AArch64::UBFIZxxii:  {
+    unsigned ImmOps = Inst.getNumOperands() - 2;
+    int64_t ImmR = Inst.getOperand(ImmOps).getImm();
+    int64_t ImmS = Inst.getOperand(ImmOps+1).getImm();
+
+    if (ImmR != 0 && ImmS >= ImmR) {
+      return Error(Operands[4]->getStartLoc(),
+                   "requested insert overflows register");
+    }
+    return false;
+  }
+  case AArch64::BFXILwwii:
+  case AArch64::BFXILxxii:
+  case AArch64::SBFXwwii:
+  case AArch64::SBFXxxii:
+  case AArch64::UBFXwwii:
+  case AArch64::UBFXxxii: {
+    unsigned ImmOps = Inst.getNumOperands() - 2;
+    int64_t ImmR = Inst.getOperand(ImmOps).getImm();
+    int64_t ImmS = Inst.getOperand(ImmOps+1).getImm();
+    int64_t RegWidth = 0;
+    switch (Inst.getOpcode()) {
+    case AArch64::SBFXxxii: case AArch64::UBFXxxii: case AArch64::BFXILxxii:
+      RegWidth = 64;
+      break;
+    case AArch64::SBFXwwii: case AArch64::UBFXwwii: case AArch64::BFXILwwii:
+      RegWidth = 32;
+      break;
+    }
+
+    if (ImmS >= RegWidth || ImmS < ImmR) {
+      return Error(Operands[4]->getStartLoc(),
+                   "requested extract overflows register");
+    }
+    return false;
+  }
+  case AArch64::ICix: {
+    int64_t ImmVal = Inst.getOperand(0).getImm();
+    A64IC::ICValues ICOp = static_cast<A64IC::ICValues>(ImmVal);
+    if (!A64IC::NeedsRegister(ICOp)) {
+      return Error(Operands[1]->getStartLoc(),
+                   "specified IC op does not use a register");
+    }
+    return false;
+  }
+  case AArch64::ICi: {
+    int64_t ImmVal = Inst.getOperand(0).getImm();
+    A64IC::ICValues ICOp = static_cast<A64IC::ICValues>(ImmVal);
+    if (A64IC::NeedsRegister(ICOp)) {
+      return Error(Operands[1]->getStartLoc(),
+                   "specified IC op requires a register");
+    }
+    return false;
+  }
+  case AArch64::TLBIix: {
+    int64_t ImmVal = Inst.getOperand(0).getImm();
+    A64TLBI::TLBIValues TLBIOp = static_cast<A64TLBI::TLBIValues>(ImmVal);
+    if (!A64TLBI::NeedsRegister(TLBIOp)) {
+      return Error(Operands[1]->getStartLoc(),
+                   "specified TLBI op does not use a register");
+    }
+    return false;
+  }
+  case AArch64::TLBIi: {
+    int64_t ImmVal = Inst.getOperand(0).getImm();
+    A64TLBI::TLBIValues TLBIOp = static_cast<A64TLBI::TLBIValues>(ImmVal);
+    if (A64TLBI::NeedsRegister(TLBIOp)) {
+      return Error(Operands[1]->getStartLoc(),
+                   "specified TLBI op requires a register");
+    }
+    return false;
+  }
+  }
+
+  return false;
+}
+
+
+// Parses the instruction *together with* all operands, appending each parsed
+// operand to the "Operands" list
+bool AArch64AsmParser::ParseInstruction(ParseInstructionInfo &Info,
+                                        StringRef Name, SMLoc NameLoc,
+                               SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  size_t CondCodePos = Name.find('.');
+
+  StringRef Mnemonic = Name.substr(0, CondCodePos);
+  Operands.push_back(AArch64Operand::CreateToken(Mnemonic, NameLoc));
+
+  if (CondCodePos != StringRef::npos) {
+    // We have a condition code
+    SMLoc S = SMLoc::getFromPointer(NameLoc.getPointer() + CondCodePos + 1);
+    StringRef CondStr = Name.substr(CondCodePos + 1, StringRef::npos);
+    A64CC::CondCodes Code;
+
+    Code = A64StringToCondCode(CondStr);
+
+    if (Code == A64CC::Invalid) {
+      Error(S, "invalid condition code");
+      Parser.eatToEndOfStatement();
+      return true;
+    }
+
+    SMLoc DotL = SMLoc::getFromPointer(NameLoc.getPointer() + CondCodePos);
+
+    Operands.push_back(AArch64Operand::CreateToken(".",  DotL));
+    SMLoc E = SMLoc::getFromPointer(NameLoc.getPointer() + CondCodePos + 3);
+    Operands.push_back(AArch64Operand::CreateCondCode(Code, S, E));
+  }
+
+  // Now we parse the operands of this instruction
+  if (getLexer().isNot(AsmToken::EndOfStatement)) {
+    // Read the first operand.
+    if (ParseOperand(Operands, Mnemonic)) {
+      Parser.eatToEndOfStatement();
+      return true;
+    }
+
+    while (getLexer().is(AsmToken::Comma)) {
+      Parser.Lex();  // Eat the comma.
+
+      // Parse and remember the operand.
+      if (ParseOperand(Operands, Mnemonic)) {
+        Parser.eatToEndOfStatement();
+        return true;
+      }
+
+
+      // After successfully parsing some operands there are two special cases to
+      // consider (i.e. notional operands not separated by commas). Both are due
+      // to memory specifiers:
+      //  + An RBrac will end an address for load/store/prefetch
+      //  + An '!' will indicate a pre-indexed operation.
+      //
+      // It's someone else's responsibility to make sure these tokens are sane
+      // in the given context!
+      if (Parser.getTok().is(AsmToken::RBrac)) {
+        SMLoc Loc = Parser.getTok().getLoc();
+        Operands.push_back(AArch64Operand::CreateToken("]", Loc));
+        Parser.Lex();
+      }
+
+      if (Parser.getTok().is(AsmToken::Exclaim)) {
+        SMLoc Loc = Parser.getTok().getLoc();
+        Operands.push_back(AArch64Operand::CreateToken("!", Loc));
+        Parser.Lex();
+      }
+    }
+  }
+
+  if (getLexer().isNot(AsmToken::EndOfStatement)) {
+    SMLoc Loc = getLexer().getLoc();
+    Parser.eatToEndOfStatement();
+    return Error(Loc, "expected comma before next operand");
+  }
+
+  // Eat the EndOfStatement
+  Parser.Lex();
+
+  return false;
+}
+
+bool AArch64AsmParser::ParseDirective(AsmToken DirectiveID) {
+  StringRef IDVal = DirectiveID.getIdentifier();
+  if (IDVal == ".hword")
+    return ParseDirectiveWord(2, DirectiveID.getLoc());
+  else if (IDVal == ".word")
+    return ParseDirectiveWord(4, DirectiveID.getLoc());
+  else if (IDVal == ".xword")
+    return ParseDirectiveWord(8, DirectiveID.getLoc());
+  else if (IDVal == ".tlsdesccall")
+    return ParseDirectiveTLSDescCall(DirectiveID.getLoc());
+
+  return true;
+}
+
+/// parseDirectiveWord
+///  ::= .word [ expression (, expression)* ]
+bool AArch64AsmParser::ParseDirectiveWord(unsigned Size, SMLoc L) {
+  if (getLexer().isNot(AsmToken::EndOfStatement)) {
+    for (;;) {
+      const MCExpr *Value;
+      if (getParser().parseExpression(Value))
+        return true;
+
+      getParser().getStreamer().EmitValue(Value, Size, 0/*addrspace*/);
+
+      if (getLexer().is(AsmToken::EndOfStatement))
+        break;
+
+      // FIXME: Improve diagnostic.
+      if (getLexer().isNot(AsmToken::Comma))
+        return Error(L, "unexpected token in directive");
+      Parser.Lex();
+    }
+  }
+
+  Parser.Lex();
+  return false;
+}
+
+// parseDirectiveTLSDescCall:
+//   ::= .tlsdesccall symbol
+bool AArch64AsmParser::ParseDirectiveTLSDescCall(SMLoc L) {
+  StringRef Name;
+  if (getParser().parseIdentifier(Name))
+    return Error(L, "expected symbol after directive");
+
+  MCSymbol *Sym = getContext().GetOrCreateSymbol(Name);
+  const MCSymbolRefExpr *Expr = MCSymbolRefExpr::Create(Sym, getContext());
+
+  MCInst Inst;
+  Inst.setOpcode(AArch64::TLSDESCCALL);
+  Inst.addOperand(MCOperand::CreateExpr(Expr));
+
+  getParser().getStreamer().EmitInstruction(Inst);
+  return false;
+}
+
+
+bool AArch64AsmParser::MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
+                                 SmallVectorImpl<MCParsedAsmOperand*> &Operands,
+                                 MCStreamer &Out, unsigned &ErrorInfo,
+                                 bool MatchingInlineAsm) {
+  MCInst Inst;
+  unsigned MatchResult;
+  MatchResult = MatchInstructionImpl(Operands, Inst, ErrorInfo,
+                                     MatchingInlineAsm);
+
+  if (ErrorInfo != ~0U && ErrorInfo >= Operands.size())
+    return Error(IDLoc, "too few operands for instruction");
+
+  switch (MatchResult) {
+  default: break;
+  case Match_Success:
+    if (validateInstruction(Inst, Operands))
+      return true;
+
+    Out.EmitInstruction(Inst);
+    return false;
+  case Match_MissingFeature:
+    Error(IDLoc, "instruction requires a CPU feature not currently enabled");
+    return true;
+  case Match_InvalidOperand: {
+    SMLoc ErrorLoc = IDLoc;
+    if (ErrorInfo != ~0U) {
+      ErrorLoc = ((AArch64Operand*)Operands[ErrorInfo])->getStartLoc();
+      if (ErrorLoc == SMLoc()) ErrorLoc = IDLoc;
+    }
+
+    return Error(ErrorLoc, "invalid operand for instruction");
+  }
+  case Match_MnemonicFail:
+    return Error(IDLoc, "invalid instruction");
+
+  case Match_AddSubRegExtendSmall:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+      "expected '[su]xt[bhw]' or 'lsl' with optional integer in range [0, 4]");
+  case Match_AddSubRegExtendLarge:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+      "expected 'sxtx' 'uxtx' or 'lsl' with optional integer in range [0, 4]");
+  case Match_AddSubRegShift32:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+       "expected 'lsl', 'lsr' or 'asr' with optional integer in range [0, 31]");
+  case Match_AddSubRegShift64:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+       "expected 'lsl', 'lsr' or 'asr' with optional integer in range [0, 63]");
+  case Match_AddSubSecondSource:
+      return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+          "expected compatible register, symbol or integer in range [0, 4095]");
+  case Match_CVTFixedPos32:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [1, 32]");
+  case Match_CVTFixedPos64:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [1, 64]");
+  case Match_CondCode:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected AArch64 condition code");
+  case Match_FPImm:
+    // Any situation which allows a nontrivial floating-point constant also
+    // allows a register.
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected compatible register or floating-point constant");
+  case Match_FPZero:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected floating-point constant #0.0");
+  case Match_Label:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected label or encodable integer pc offset");
+  case Match_Lane1:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected lane specifier '[1]'");
+  case Match_LoadStoreExtend32_1:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'uxtw' or 'sxtw' with optional shift of #0");
+  case Match_LoadStoreExtend32_2:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'uxtw' or 'sxtw' with optional shift of #0 or #1");
+  case Match_LoadStoreExtend32_4:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'uxtw' or 'sxtw' with optional shift of #0 or #2");
+  case Match_LoadStoreExtend32_8:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'uxtw' or 'sxtw' with optional shift of #0 or #3");
+  case Match_LoadStoreExtend32_16:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'lsl' or 'sxtw' with optional shift of #0 or #4");
+  case Match_LoadStoreExtend64_1:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'lsl' or 'sxtx' with optional shift of #0");
+  case Match_LoadStoreExtend64_2:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'lsl' or 'sxtx' with optional shift of #0 or #1");
+  case Match_LoadStoreExtend64_4:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'lsl' or 'sxtx' with optional shift of #0 or #2");
+  case Match_LoadStoreExtend64_8:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'lsl' or 'sxtx' with optional shift of #0 or #3");
+  case Match_LoadStoreExtend64_16:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'lsl' or 'sxtx' with optional shift of #0 or #4");
+  case Match_LoadStoreSImm7_4:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer multiple of 4 in range [-256, 252]");
+  case Match_LoadStoreSImm7_8:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer multiple of 8 in range [-512, 508]");
+  case Match_LoadStoreSImm7_16:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer multiple of 16 in range [-1024, 1016]");
+  case Match_LoadStoreSImm9:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [-256, 255]");
+  case Match_LoadStoreUImm12_1:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected symbolic reference or integer in range [0, 4095]");
+  case Match_LoadStoreUImm12_2:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected symbolic reference or integer in range [0, 8190]");
+  case Match_LoadStoreUImm12_4:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected symbolic reference or integer in range [0, 16380]");
+  case Match_LoadStoreUImm12_8:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected symbolic reference or integer in range [0, 32760]");
+  case Match_LoadStoreUImm12_16:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected symbolic reference or integer in range [0, 65520]");
+  case Match_LogicalSecondSource:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected compatible register or logical immediate");
+  case Match_MOVWUImm16:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected relocated symbol or integer in range [0, 65535]");
+  case Match_MRS:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected readable system register");
+  case Match_MSR:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected writable system register or pstate");
+  case Match_NamedImm_at:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                "expected symbolic 'at' operand: s1e[0-3][rw] or s12e[01][rw]");
+  case Match_NamedImm_dbarrier:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+             "expected integer in range [0, 15] or symbolic barrier operand");
+  case Match_NamedImm_dc:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected symbolic 'dc' operand");
+  case Match_NamedImm_ic:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected 'ic' operand: 'ialluis', 'iallu' or 'ivau'");
+  case Match_NamedImm_isb:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [0, 15] or 'sy'");
+  case Match_NamedImm_prefetch:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected prefetch hint: p(ld|st|i)l[123](strm|keep)");
+  case Match_NamedImm_tlbi:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected translation buffer invalidation operand");
+  case Match_UImm16:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [0, 65535]");
+  case Match_UImm3:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [0, 7]");
+  case Match_UImm4:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [0, 15]");
+  case Match_UImm5:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [0, 31]");
+  case Match_UImm6:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [0, 63]");
+  case Match_UImm7:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [0, 127]");
+  case Match_Width32:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [<lsb>, 31]");
+  case Match_Width64:
+    return Error(((AArch64Operand*)Operands[ErrorInfo])->getStartLoc(),
+                 "expected integer in range [<lsb>, 63]");
+  }
+
+  llvm_unreachable("Implement any new match types added!");
+  return true;
+}
+
+void AArch64Operand::print(raw_ostream &OS) const {
+  switch (Kind) {
+  case k_CondCode:
+    OS << "<CondCode: " << CondCode.Code << ">";
+    break;
+  case k_FPImmediate:
+    OS << "<fpimm: " << FPImm.Val << ">";
+    break;
+  case k_ImmWithLSL:
+    OS << "<immwithlsl: imm=" << ImmWithLSL.Val
+       << ", shift=" << ImmWithLSL.ShiftAmount << ">";
+    break;
+  case k_Immediate:
+    getImm()->print(OS);
+    break;
+  case k_Register:
+    OS << "<register " << getReg() << '>';
+    break;
+  case k_Token:
+    OS << '\'' << getToken() << '\'';
+    break;
+  case k_ShiftExtend:
+    OS << "<shift: type=" << ShiftExtend.ShiftType
+       << ", amount=" << ShiftExtend.Amount << ">";
+    break;
+  case k_SysReg: {
+    StringRef Name(SysReg.Data, SysReg.Length);
+    OS << "<sysreg: " << Name << '>';
+    break;
+  }
+  default:
+    llvm_unreachable("No idea how to print this kind of operand");
+    break;
+  }
+}
+
+void AArch64Operand::dump() const {
+  print(errs());
+}
+
+
+/// Force static initialization.
+extern "C" void LLVMInitializeAArch64AsmParser() {
+  RegisterMCAsmParser<AArch64AsmParser> X(TheAArch64Target);
+}
+
+#define GET_REGISTER_MATCHER
+#define GET_MATCHER_IMPLEMENTATION
+#include "AArch64GenAsmMatcher.inc"
diff --git a/lib/Target/AArch64/AsmParser/CMakeLists.txt b/lib/Target/AArch64/AsmParser/CMakeLists.txt
new file mode 100644
index 000000000000..a018a0aa7b36
--- /dev/null
+++ b/lib/Target/AArch64/AsmParser/CMakeLists.txt
@@ -0,0 +1,7 @@
+include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
+
+add_llvm_library(LLVMAArch64AsmParser
+  AArch64AsmParser.cpp
+  )
+
+add_dependencies(LLVMAArch64AsmParser AArch64CommonTableGen)
diff --git a/lib/Target/AArch64/AsmParser/LLVMBuild.txt b/lib/Target/AArch64/AsmParser/LLVMBuild.txt
new file mode 100644
index 000000000000..bd1fcaf1ffe8
--- /dev/null
+++ b/lib/Target/AArch64/AsmParser/LLVMBuild.txt
@@ -0,0 +1,24 @@
+;===- ./lib/Target/AArch64/AsmParser/LLVMBuild.txt -------------*- Conf -*--===;
+;
+;                     The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+;   http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[component_0]
+type = Library
+name = AArch64AsmParser
+parent = AArch64
+required_libraries = AArch64Desc AArch64Info MC MCParser Support
+add_to_library_groups = AArch64
+
diff --git a/lib/Target/AArch64/AsmParser/Makefile b/lib/Target/AArch64/AsmParser/Makefile
new file mode 100644
index 000000000000..56c9ef52ea58
--- /dev/null
+++ b/lib/Target/AArch64/AsmParser/Makefile
@@ -0,0 +1,15 @@
+##===- lib/Target/AArch64/AsmParser/Makefile ---------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+LEVEL = ../../../..
+LIBRARYNAME = LLVMAArch64AsmParser
+
+# Hack: we need to include 'main' target directory to grab private headers
+CPP.Flags += -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/Target/AArch64/CMakeLists.txt b/lib/Target/AArch64/CMakeLists.txt
new file mode 100644
index 000000000000..8164d6f73c97
--- /dev/null
+++ b/lib/Target/AArch64/CMakeLists.txt
@@ -0,0 +1,36 @@
+set(LLVM_TARGET_DEFINITIONS AArch64.td)
+
+tablegen(LLVM AArch64GenAsmMatcher.inc -gen-asm-matcher)
+tablegen(LLVM AArch64GenAsmWriter.inc -gen-asm-writer)
+tablegen(LLVM AArch64GenCallingConv.inc -gen-callingconv)
+tablegen(LLVM AArch64GenDisassemblerTables.inc -gen-disassembler)
+tablegen(LLVM AArch64GenInstrInfo.inc -gen-instr-info)
+tablegen(LLVM AArch64GenMCCodeEmitter.inc -gen-emitter -mc-emitter)
+tablegen(LLVM AArch64GenMCPseudoLowering.inc -gen-pseudo-lowering)
+tablegen(LLVM AArch64GenRegisterInfo.inc -gen-register-info)
+tablegen(LLVM AArch64GenDAGISel.inc -gen-dag-isel)
+tablegen(LLVM AArch64GenSubtargetInfo.inc -gen-subtarget)
+add_public_tablegen_target(AArch64CommonTableGen)
+
+add_llvm_target(AArch64CodeGen
+  AArch64AsmPrinter.cpp
+  AArch64BranchFixupPass.cpp
+  AArch64FrameLowering.cpp
+  AArch64ISelDAGToDAG.cpp
+  AArch64ISelLowering.cpp
+  AArch64InstrInfo.cpp
+  AArch64MachineFunctionInfo.cpp
+  AArch64MCInstLower.cpp
+  AArch64RegisterInfo.cpp
+  AArch64SelectionDAGInfo.cpp
+  AArch64Subtarget.cpp
+  AArch64TargetMachine.cpp
+  AArch64TargetObjectFile.cpp
+  )
+
+add_subdirectory(AsmParser)
+add_subdirectory(Disassembler)
+add_subdirectory(InstPrinter)
+add_subdirectory(MCTargetDesc)
+add_subdirectory(TargetInfo)
+add_subdirectory(Utils)
\ No newline at end of file
diff --git a/lib/Target/AArch64/Disassembler/AArch64Disassembler.cpp b/lib/Target/AArch64/Disassembler/AArch64Disassembler.cpp
new file mode 100644
index 000000000000..12c1b8f4c81a
--- /dev/null
+++ b/lib/Target/AArch64/Disassembler/AArch64Disassembler.cpp
@@ -0,0 +1,803 @@
+//===- AArch64Disassembler.cpp - Disassembler for AArch64 ISA -------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the functions necessary to decode AArch64 instruction
+// bitpatterns into MCInsts (with the help of TableGenerated information from
+// the instruction definitions).
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "arm-disassembler"
+
+#include "AArch64.h"
+#include "AArch64RegisterInfo.h"
+#include "AArch64Subtarget.h"
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCInstrDesc.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCDisassembler.h"
+#include "llvm/MC/MCFixedLenDisassembler.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/MemoryObject.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+typedef MCDisassembler::DecodeStatus DecodeStatus;
+
+namespace {
+/// AArch64 disassembler for all AArch64 platforms.
+class AArch64Disassembler : public MCDisassembler {
+  const MCRegisterInfo *RegInfo;
+public:
+  /// Initializes the disassembler.
+  ///
+  AArch64Disassembler(const MCSubtargetInfo &STI, const MCRegisterInfo *Info)
+    : MCDisassembler(STI), RegInfo(Info) {
+  }
+
+  ~AArch64Disassembler() {
+  }
+
+  /// See MCDisassembler.
+  DecodeStatus getInstruction(MCInst &instr,
+                              uint64_t &size,
+                              const MemoryObject &region,
+                              uint64_t address,
+                              raw_ostream &vStream,
+                              raw_ostream &cStream) const;
+
+  const MCRegisterInfo *getRegInfo() const { return RegInfo; }
+};
+
+}
+
+// Forward-declarations used in the auto-generated files.
+static DecodeStatus DecodeGPR64RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                                         uint64_t Address, const void *Decoder);
+static DecodeStatus
+DecodeGPR64xspRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                            uint64_t Address, const void *Decoder);
+
+static DecodeStatus DecodeGPR32RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                                         uint64_t Address, const void *Decoder);
+static DecodeStatus
+DecodeGPR32wspRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                            uint64_t Address, const void *Decoder);
+
+static DecodeStatus DecodeFPR8RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                                         uint64_t Address, const void *Decoder);
+static DecodeStatus DecodeFPR16RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                                         uint64_t Address, const void *Decoder);
+static DecodeStatus DecodeFPR32RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                                         uint64_t Address, const void *Decoder);
+static DecodeStatus DecodeFPR64RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                                         uint64_t Address, const void *Decoder);
+static DecodeStatus DecodeFPR128RegisterClass(llvm::MCInst &Inst,
+                                              unsigned RegNo, uint64_t Address,
+                                              const void *Decoder);
+static DecodeStatus DecodeVPR128RegisterClass(llvm::MCInst &Inst,
+                                              unsigned RegNo, uint64_t Address,
+                                              const void *Decoder);
+
+static DecodeStatus DecodeAddrRegExtendOperand(llvm::MCInst &Inst,
+                                               unsigned OptionHiS,
+                                               uint64_t Address,
+                                               const void *Decoder);
+
+
+static DecodeStatus DecodeBitfield32ImmOperand(llvm::MCInst &Inst,
+                                               unsigned Imm6Bits,
+                                               uint64_t Address,
+                                               const void *Decoder);
+
+static DecodeStatus DecodeCVT32FixedPosOperand(llvm::MCInst &Inst,
+                                               unsigned Imm6Bits,
+                                               uint64_t Address,
+                                               const void *Decoder);
+
+static DecodeStatus DecodeFPZeroOperand(llvm::MCInst &Inst,
+                                        unsigned RmBits,
+                                        uint64_t Address,
+                                        const void *Decoder);
+
+template<int RegWidth>
+static DecodeStatus DecodeMoveWideImmOperand(llvm::MCInst &Inst,
+                                             unsigned FullImm,
+                                             uint64_t Address,
+                                             const void *Decoder);
+
+template<int RegWidth>
+static DecodeStatus DecodeLogicalImmOperand(llvm::MCInst &Inst,
+                                            unsigned Bits,
+                                            uint64_t Address,
+                                            const void *Decoder);
+
+static DecodeStatus DecodeRegExtendOperand(llvm::MCInst &Inst,
+                                           unsigned ShiftAmount,
+                                           uint64_t Address,
+                                           const void *Decoder);
+
+static DecodeStatus Decode32BitShiftOperand(llvm::MCInst &Inst,
+                                            unsigned ShiftAmount,
+                                            uint64_t Address,
+                                            const void *Decoder);
+static DecodeStatus DecodeBitfieldInstruction(llvm::MCInst &Inst, unsigned Insn,
+                                              uint64_t Address,
+                                              const void *Decoder);
+
+static DecodeStatus DecodeFMOVLaneInstruction(llvm::MCInst &Inst, unsigned Insn,
+                                              uint64_t Address,
+                                              const void *Decoder);
+
+static DecodeStatus DecodeLDSTPairInstruction(llvm::MCInst &Inst,
+                                              unsigned Insn,
+                                              uint64_t Address,
+                                              const void *Decoder);
+
+static DecodeStatus DecodeLoadPairExclusiveInstruction(llvm::MCInst &Inst,
+                                                       unsigned Val,
+                                                       uint64_t Address,
+                                                       const void *Decoder);
+
+template<typename SomeNamedImmMapper>
+static DecodeStatus DecodeNamedImmOperand(llvm::MCInst &Inst,
+                                          unsigned Val,
+                                          uint64_t Address,
+                                          const void *Decoder);
+
+static DecodeStatus
+DecodeSysRegOperand(const A64SysReg::SysRegMapper &InstMapper,
+                    llvm::MCInst &Inst, unsigned Val,
+                    uint64_t Address, const void *Decoder);
+
+static DecodeStatus DecodeMRSOperand(llvm::MCInst &Inst,
+                                     unsigned Val,
+                                     uint64_t Address,
+                                     const void *Decoder);
+
+static DecodeStatus DecodeMSROperand(llvm::MCInst &Inst,
+                                     unsigned Val,
+                                     uint64_t Address,
+                                     const void *Decoder);
+
+
+static DecodeStatus DecodeSingleIndexedInstruction(llvm::MCInst &Inst,
+                                                   unsigned Val,
+                                                   uint64_t Address,
+                                                   const void *Decoder);
+
+
+static bool Check(DecodeStatus &Out, DecodeStatus In);
+
+#include "AArch64GenDisassemblerTables.inc"
+#include "AArch64GenInstrInfo.inc"
+
+static bool Check(DecodeStatus &Out, DecodeStatus In) {
+  switch (In) {
+    case MCDisassembler::Success:
+      // Out stays the same.
+      return true;
+    case MCDisassembler::SoftFail:
+      Out = In;
+      return true;
+    case MCDisassembler::Fail:
+      Out = In;
+      return false;
+  }
+  llvm_unreachable("Invalid DecodeStatus!");
+}
+
+DecodeStatus AArch64Disassembler::getInstruction(MCInst &MI, uint64_t &Size,
+                                                 const MemoryObject &Region,
+                                                 uint64_t Address,
+                                                 raw_ostream &os,
+                                                 raw_ostream &cs) const {
+  CommentStream = &cs;
+
+  uint8_t bytes[4];
+
+  // We want to read exactly 4 bytes of data.
+  if (Region.readBytes(Address, 4, (uint8_t*)bytes, NULL) == -1) {
+    Size = 0;
+    return MCDisassembler::Fail;
+  }
+
+  // Encoded as a small-endian 32-bit word in the stream.
+  uint32_t insn = (bytes[3] << 24) |
+    (bytes[2] << 16) |
+    (bytes[1] <<  8) |
+    (bytes[0] <<  0);
+
+  // Calling the auto-generated decoder function.
+  DecodeStatus result = decodeInstruction(DecoderTableA6432, MI, insn, Address,
+                                          this, STI);
+  if (result != MCDisassembler::Fail) {
+    Size = 4;
+    return result;
+  }
+
+  MI.clear();
+  Size = 0;
+  return MCDisassembler::Fail;
+}
+
+static unsigned getReg(const void *D, unsigned RC, unsigned RegNo) {
+  const AArch64Disassembler *Dis = static_cast<const AArch64Disassembler*>(D);
+  return Dis->getRegInfo()->getRegClass(RC).getRegister(RegNo);
+}
+
+static DecodeStatus DecodeGPR64RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                                        uint64_t Address, const void *Decoder) {
+  if (RegNo > 31)
+    return MCDisassembler::Fail;
+
+  uint16_t Register = getReg(Decoder, AArch64::GPR64RegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Register));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus
+DecodeGPR64xspRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                            uint64_t Address, const void *Decoder) {
+  if (RegNo > 31)
+    return MCDisassembler::Fail;
+
+  uint16_t Register = getReg(Decoder, AArch64::GPR64xspRegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Register));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus DecodeGPR32RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                                             uint64_t Address,
+                                             const void *Decoder) {
+  if (RegNo > 31)
+    return MCDisassembler::Fail;
+
+  uint16_t Register = getReg(Decoder, AArch64::GPR32RegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Register));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus
+DecodeGPR32wspRegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                            uint64_t Address, const void *Decoder) {
+  if (RegNo > 31)
+    return MCDisassembler::Fail;
+
+  uint16_t Register = getReg(Decoder, AArch64::GPR32wspRegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Register));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus
+DecodeFPR8RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                            uint64_t Address, const void *Decoder) {
+  if (RegNo > 31)
+    return MCDisassembler::Fail;
+
+  uint16_t Register = getReg(Decoder, AArch64::FPR8RegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Register));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus
+DecodeFPR16RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                            uint64_t Address, const void *Decoder) {
+  if (RegNo > 31)
+    return MCDisassembler::Fail;
+
+  uint16_t Register = getReg(Decoder, AArch64::FPR16RegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Register));
+  return MCDisassembler::Success;
+}
+
+
+static DecodeStatus
+DecodeFPR32RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                            uint64_t Address, const void *Decoder) {
+  if (RegNo > 31)
+    return MCDisassembler::Fail;
+
+  uint16_t Register = getReg(Decoder, AArch64::FPR32RegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Register));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus
+DecodeFPR64RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                            uint64_t Address, const void *Decoder) {
+  if (RegNo > 31)
+    return MCDisassembler::Fail;
+
+  uint16_t Register = getReg(Decoder, AArch64::FPR64RegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Register));
+  return MCDisassembler::Success;
+}
+
+
+static DecodeStatus
+DecodeFPR128RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                            uint64_t Address, const void *Decoder) {
+  if (RegNo > 31)
+    return MCDisassembler::Fail;
+
+  uint16_t Register = getReg(Decoder, AArch64::FPR128RegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Register));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus
+DecodeVPR128RegisterClass(llvm::MCInst &Inst, unsigned RegNo,
+                         uint64_t Address, const void *Decoder) {
+  if (RegNo > 31)
+    return MCDisassembler::Fail;
+
+  uint16_t Register = getReg(Decoder, AArch64::VPR128RegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Register));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus DecodeAddrRegExtendOperand(llvm::MCInst &Inst,
+                                               unsigned OptionHiS,
+                                               uint64_t Address,
+                                               const void *Decoder) {
+  // Option{1} must be 1. OptionHiS is made up of {Option{2}, Option{1},
+  // S}. Hence we want to check bit 1.
+  if (!(OptionHiS & 2))
+    return MCDisassembler::Fail;
+
+  Inst.addOperand(MCOperand::CreateImm(OptionHiS));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus DecodeBitfield32ImmOperand(llvm::MCInst &Inst,
+                                               unsigned Imm6Bits,
+                                               uint64_t Address,
+                                               const void *Decoder) {
+  // In the 32-bit variant, bit 6 must be zero. I.e. the immediate must be
+  // between 0 and 31.
+  if (Imm6Bits > 31)
+    return MCDisassembler::Fail;
+
+  Inst.addOperand(MCOperand::CreateImm(Imm6Bits));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus DecodeCVT32FixedPosOperand(llvm::MCInst &Inst,
+                                               unsigned Imm6Bits,
+                                               uint64_t Address,
+                                               const void *Decoder) {
+  // 1 <= Imm <= 32. Encoded as 64 - Imm so: 63 >= Encoded >= 32.
+  if (Imm6Bits < 32)
+    return MCDisassembler::Fail;
+
+  Inst.addOperand(MCOperand::CreateImm(Imm6Bits));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus DecodeFPZeroOperand(llvm::MCInst &Inst,
+                                        unsigned RmBits,
+                                        uint64_t Address,
+                                        const void *Decoder) {
+  // Any bits are valid in the instruction (they're architecturally ignored),
+  // but a code generator should insert 0.
+  Inst.addOperand(MCOperand::CreateImm(0));
+  return MCDisassembler::Success;
+}
+
+
+
+template<int RegWidth>
+static DecodeStatus DecodeMoveWideImmOperand(llvm::MCInst &Inst,
+                                             unsigned FullImm,
+                                             uint64_t Address,
+                                             const void *Decoder) {
+  unsigned Imm16 = FullImm & 0xffff;
+  unsigned Shift = FullImm >> 16;
+
+  if (RegWidth == 32 && Shift > 1) return MCDisassembler::Fail;
+
+  Inst.addOperand(MCOperand::CreateImm(Imm16));
+  Inst.addOperand(MCOperand::CreateImm(Shift));
+  return MCDisassembler::Success;
+}
+
+template<int RegWidth>
+static DecodeStatus DecodeLogicalImmOperand(llvm::MCInst &Inst,
+                                            unsigned Bits,
+                                            uint64_t Address,
+                                            const void *Decoder) {
+  uint64_t Imm;
+  if (!A64Imms::isLogicalImmBits(RegWidth, Bits, Imm))
+    return MCDisassembler::Fail;
+
+  Inst.addOperand(MCOperand::CreateImm(Bits));
+  return MCDisassembler::Success;
+}
+
+
+static DecodeStatus DecodeRegExtendOperand(llvm::MCInst &Inst,
+                                           unsigned ShiftAmount,
+                                           uint64_t Address,
+                                           const void *Decoder) {
+  // Only values 0-4 are valid for this 3-bit field
+  if (ShiftAmount > 4)
+    return MCDisassembler::Fail;
+
+  Inst.addOperand(MCOperand::CreateImm(ShiftAmount));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus Decode32BitShiftOperand(llvm::MCInst &Inst,
+                                            unsigned ShiftAmount,
+                                            uint64_t Address,
+                                            const void *Decoder) {
+  // Only values below 32 are valid for a 32-bit register
+  if (ShiftAmount > 31)
+    return MCDisassembler::Fail;
+
+  Inst.addOperand(MCOperand::CreateImm(ShiftAmount));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus DecodeBitfieldInstruction(llvm::MCInst &Inst, unsigned Insn,
+                                              uint64_t Address,
+                                              const void *Decoder) {
+  unsigned Rd = fieldFromInstruction(Insn, 0, 5);
+  unsigned Rn = fieldFromInstruction(Insn, 5, 5);
+  unsigned ImmS = fieldFromInstruction(Insn, 10, 6);
+  unsigned ImmR = fieldFromInstruction(Insn, 16, 6);
+  unsigned SF = fieldFromInstruction(Insn, 31, 1);
+
+  // Undef for 0b11 just in case it occurs. Don't want the compiler to optimise
+  // out assertions that it thinks should never be hit.
+  enum OpcTypes { SBFM = 0, BFM, UBFM, Undef } Opc;
+  Opc = (OpcTypes)fieldFromInstruction(Insn, 29, 2);
+
+  if (!SF) {
+    // ImmR and ImmS must be between 0 and 31 for 32-bit instructions.
+    if (ImmR > 31 || ImmS > 31)
+      return MCDisassembler::Fail;
+  }
+
+  if (SF) {
+    DecodeGPR64RegisterClass(Inst, Rd, Address, Decoder);
+    // BFM MCInsts use Rd as a source too.
+    if (Opc == BFM) DecodeGPR64RegisterClass(Inst, Rd, Address, Decoder);
+    DecodeGPR64RegisterClass(Inst, Rn, Address, Decoder);
+  } else {
+    DecodeGPR32RegisterClass(Inst, Rd, Address, Decoder);
+    // BFM MCInsts use Rd as a source too.
+    if (Opc == BFM) DecodeGPR32RegisterClass(Inst, Rd, Address, Decoder);
+    DecodeGPR32RegisterClass(Inst, Rn, Address, Decoder);
+  }
+
+  // ASR and LSR have more specific patterns so they won't get here:
+  assert(!(ImmS == 31 && !SF && Opc != BFM)
+         && "shift should have used auto decode");
+  assert(!(ImmS == 63 && SF && Opc != BFM)
+         && "shift should have used auto decode");
+
+  // Extension instructions similarly:
+  if (Opc == SBFM && ImmR == 0) {
+    assert((ImmS != 7 && ImmS != 15) && "extension got here");
+    assert((ImmS != 31 || SF == 0) && "extension got here");
+  } else if (Opc == UBFM && ImmR == 0) {
+    assert((SF != 0 || (ImmS != 7 && ImmS != 15)) && "extension got here");
+  }
+
+  if (Opc == UBFM) {
+    // It might be a LSL instruction, which actually takes the shift amount
+    // itself as an MCInst operand.
+    if (SF && (ImmS + 1) % 64 == ImmR) {
+      Inst.setOpcode(AArch64::LSLxxi);
+      Inst.addOperand(MCOperand::CreateImm(63 - ImmS));
+      return MCDisassembler::Success;
+    } else if (!SF && (ImmS + 1) % 32 == ImmR) {
+      Inst.setOpcode(AArch64::LSLwwi);
+      Inst.addOperand(MCOperand::CreateImm(31 - ImmS));
+      return MCDisassembler::Success;
+    }
+  }
+
+  // Otherwise it's definitely either an extract or an insert depending on which
+  // of ImmR or ImmS is larger.
+  unsigned ExtractOp, InsertOp;
+  switch (Opc) {
+  default: llvm_unreachable("unexpected instruction trying to decode bitfield");
+  case SBFM:
+    ExtractOp = SF ? AArch64::SBFXxxii : AArch64::SBFXwwii;
+    InsertOp = SF ? AArch64::SBFIZxxii : AArch64::SBFIZwwii;
+    break;
+  case BFM:
+    ExtractOp = SF ? AArch64::BFXILxxii : AArch64::BFXILwwii;
+    InsertOp = SF ? AArch64::BFIxxii : AArch64::BFIwwii;
+    break;
+  case UBFM:
+    ExtractOp = SF ? AArch64::UBFXxxii : AArch64::UBFXwwii;
+    InsertOp = SF ? AArch64::UBFIZxxii : AArch64::UBFIZwwii;
+    break;
+  }
+
+  // Otherwise it's a boring insert or extract
+  Inst.addOperand(MCOperand::CreateImm(ImmR));
+  Inst.addOperand(MCOperand::CreateImm(ImmS));
+
+
+  if (ImmS < ImmR)
+    Inst.setOpcode(InsertOp);
+  else
+    Inst.setOpcode(ExtractOp);
+
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus DecodeFMOVLaneInstruction(llvm::MCInst &Inst, unsigned Insn,
+                                              uint64_t Address,
+                                              const void *Decoder) {
+  // This decoder exists to add the dummy Lane operand to the MCInst, which must
+  // be 1 in assembly but has no other real manifestation.
+  unsigned Rd = fieldFromInstruction(Insn, 0, 5);
+  unsigned Rn = fieldFromInstruction(Insn, 5, 5);
+  unsigned IsToVec = fieldFromInstruction(Insn, 16, 1);
+
+  if (IsToVec) {
+    DecodeVPR128RegisterClass(Inst, Rd, Address, Decoder);
+    DecodeGPR64RegisterClass(Inst, Rn, Address, Decoder);
+  } else {
+    DecodeGPR64RegisterClass(Inst, Rd, Address, Decoder);
+    DecodeVPR128RegisterClass(Inst, Rn, Address, Decoder);
+  }
+
+  // Add the lane
+  Inst.addOperand(MCOperand::CreateImm(1));
+
+  return MCDisassembler::Success;
+}
+
+
+static DecodeStatus DecodeLDSTPairInstruction(llvm::MCInst &Inst,
+                                              unsigned Insn,
+                                              uint64_t Address,
+                                              const void *Decoder) {
+  DecodeStatus Result = MCDisassembler::Success;
+  unsigned Rt = fieldFromInstruction(Insn, 0, 5);
+  unsigned Rn = fieldFromInstruction(Insn, 5, 5);
+  unsigned Rt2 = fieldFromInstruction(Insn, 10, 5);
+  unsigned SImm7 = fieldFromInstruction(Insn, 15, 7);
+  unsigned L = fieldFromInstruction(Insn, 22, 1);
+  unsigned V = fieldFromInstruction(Insn, 26, 1);
+  unsigned Opc = fieldFromInstruction(Insn, 30, 2);
+
+  // Not an official name, but it turns out that bit 23 distinguishes indexed
+  // from non-indexed operations.
+  unsigned Indexed = fieldFromInstruction(Insn, 23, 1);
+
+  if (Indexed && L == 0) {
+    // The MCInst for an indexed store has an out operand and 4 ins:
+    //    Rn_wb, Rt, Rt2, Rn, Imm
+    DecodeGPR64xspRegisterClass(Inst, Rn, Address, Decoder);
+  }
+
+  // You shouldn't load to the same register twice in an instruction...
+  if (L && Rt == Rt2)
+    Result = MCDisassembler::SoftFail;
+
+  // ... or do any operation that writes-back to a transfer register. But note
+  // that "stp xzr, xzr, [sp], #4" is fine because xzr and sp are different.
+  if (Indexed && V == 0 && Rn != 31 && (Rt == Rn || Rt2 == Rn))
+    Result = MCDisassembler::SoftFail;
+
+  // Exactly how we decode the MCInst's registers depends on the Opc and V
+  // fields of the instruction. These also obviously determine the size of the
+  // operation so we can fill in that information while we're at it.
+  if (V) {
+    // The instruction operates on the FP/SIMD registers
+    switch (Opc) {
+    default: return MCDisassembler::Fail;
+    case 0:
+      DecodeFPR32RegisterClass(Inst, Rt, Address, Decoder);
+      DecodeFPR32RegisterClass(Inst, Rt2, Address, Decoder);
+      break;
+    case 1:
+      DecodeFPR64RegisterClass(Inst, Rt, Address, Decoder);
+      DecodeFPR64RegisterClass(Inst, Rt2, Address, Decoder);
+      break;
+    case 2:
+      DecodeFPR128RegisterClass(Inst, Rt, Address, Decoder);
+      DecodeFPR128RegisterClass(Inst, Rt2, Address, Decoder);
+      break;
+    }
+  } else {
+    switch (Opc) {
+    default: return MCDisassembler::Fail;
+    case 0:
+      DecodeGPR32RegisterClass(Inst, Rt, Address, Decoder);
+      DecodeGPR32RegisterClass(Inst, Rt2, Address, Decoder);
+      break;
+    case 1:
+      assert(L && "unexpected \"store signed\" attempt");
+      DecodeGPR64RegisterClass(Inst, Rt, Address, Decoder);
+      DecodeGPR64RegisterClass(Inst, Rt2, Address, Decoder);
+      break;
+    case 2:
+      DecodeGPR64RegisterClass(Inst, Rt, Address, Decoder);
+      DecodeGPR64RegisterClass(Inst, Rt2, Address, Decoder);
+      break;
+    }
+  }
+
+  if (Indexed && L == 1) {
+    // The MCInst for an indexed load has 3 out operands and an 3 ins:
+    //    Rt, Rt2, Rn_wb, Rt2, Rn, Imm
+    DecodeGPR64xspRegisterClass(Inst, Rn, Address, Decoder);
+  }
+
+
+  DecodeGPR64xspRegisterClass(Inst, Rn, Address, Decoder);
+  Inst.addOperand(MCOperand::CreateImm(SImm7));
+
+  return Result;
+}
+
+static DecodeStatus DecodeLoadPairExclusiveInstruction(llvm::MCInst &Inst,
+                                                       uint32_t Val,
+                                                       uint64_t Address,
+                                                       const void *Decoder) {
+  unsigned Rt = fieldFromInstruction(Val, 0, 5);
+  unsigned Rn = fieldFromInstruction(Val, 5, 5);
+  unsigned Rt2 = fieldFromInstruction(Val, 10, 5);
+  unsigned MemSize = fieldFromInstruction(Val, 30, 2);
+
+  DecodeStatus S = MCDisassembler::Success;
+  if (Rt == Rt2) S = MCDisassembler::SoftFail;
+
+  switch (MemSize) {
+    case 2:
+      if (!Check(S, DecodeGPR32RegisterClass(Inst, Rt, Address, Decoder)))
+        return MCDisassembler::Fail;
+      if (!Check(S, DecodeGPR32RegisterClass(Inst, Rt2, Address, Decoder)))
+        return MCDisassembler::Fail;
+      break;
+    case 3:
+      if (!Check(S, DecodeGPR64RegisterClass(Inst, Rt, Address, Decoder)))
+        return MCDisassembler::Fail;
+      if (!Check(S, DecodeGPR64RegisterClass(Inst, Rt2, Address, Decoder)))
+        return MCDisassembler::Fail;
+      break;
+    default:
+      llvm_unreachable("Invalid MemSize in DecodeLoadPairExclusiveInstruction");
+  }
+
+  if (!Check(S, DecodeGPR64xspRegisterClass(Inst, Rn, Address, Decoder)))
+    return MCDisassembler::Fail;
+
+  return S;
+}
+
+template<typename SomeNamedImmMapper>
+static DecodeStatus DecodeNamedImmOperand(llvm::MCInst &Inst,
+                                          unsigned Val,
+                                          uint64_t Address,
+                                          const void *Decoder) {
+  SomeNamedImmMapper Mapper;
+  bool ValidNamed;
+  Mapper.toString(Val, ValidNamed);
+  if (ValidNamed || Mapper.validImm(Val)) {
+    Inst.addOperand(MCOperand::CreateImm(Val));
+    return MCDisassembler::Success;
+  }
+
+  return MCDisassembler::Fail;
+}
+
+static DecodeStatus DecodeSysRegOperand(const A64SysReg::SysRegMapper &Mapper,
+                                        llvm::MCInst &Inst,
+                                        unsigned Val,
+                                        uint64_t Address,
+                                        const void *Decoder) {
+  bool ValidNamed;
+  Mapper.toString(Val, ValidNamed);
+
+  Inst.addOperand(MCOperand::CreateImm(Val));
+
+  return ValidNamed ? MCDisassembler::Success : MCDisassembler::Fail;
+}
+
+static DecodeStatus DecodeMRSOperand(llvm::MCInst &Inst,
+                                     unsigned Val,
+                                     uint64_t Address,
+                                     const void *Decoder) {
+  return DecodeSysRegOperand(A64SysReg::MRSMapper(), Inst, Val, Address,
+                             Decoder);
+}
+
+static DecodeStatus DecodeMSROperand(llvm::MCInst &Inst,
+                                     unsigned Val,
+                                     uint64_t Address,
+                                     const void *Decoder) {
+  return DecodeSysRegOperand(A64SysReg::MSRMapper(), Inst, Val, Address,
+                             Decoder);
+}
+
+static DecodeStatus DecodeSingleIndexedInstruction(llvm::MCInst &Inst,
+                                                   unsigned Insn,
+                                                   uint64_t Address,
+                                                   const void *Decoder) {
+  unsigned Rt = fieldFromInstruction(Insn, 0, 5);
+  unsigned Rn = fieldFromInstruction(Insn, 5, 5);
+  unsigned Imm9 = fieldFromInstruction(Insn, 12, 9);
+
+  unsigned Opc = fieldFromInstruction(Insn, 22, 2);
+  unsigned V = fieldFromInstruction(Insn, 26, 1);
+  unsigned Size = fieldFromInstruction(Insn, 30, 2);
+
+  if (Opc == 0 || (V == 1 && Opc == 2)) {
+    // It's a store, the MCInst gets: Rn_wb, Rt, Rn, Imm
+    DecodeGPR64xspRegisterClass(Inst, Rn, Address, Decoder);
+  }
+
+  if (V == 0 && (Opc == 2 || Size == 3)) {
+    DecodeGPR64RegisterClass(Inst, Rt, Address, Decoder);
+  } else if (V == 0) {
+    DecodeGPR32RegisterClass(Inst, Rt, Address, Decoder);
+  } else if (V == 1 && (Opc & 2)) {
+    DecodeFPR128RegisterClass(Inst, Rt, Address, Decoder);
+  } else {
+    switch (Size) {
+    case 0:
+      DecodeFPR8RegisterClass(Inst, Rt, Address, Decoder);
+      break;
+    case 1:
+      DecodeFPR16RegisterClass(Inst, Rt, Address, Decoder);
+      break;
+    case 2:
+      DecodeFPR32RegisterClass(Inst, Rt, Address, Decoder);
+      break;
+    case 3:
+      DecodeFPR64RegisterClass(Inst, Rt, Address, Decoder);
+      break;
+    }
+  }
+
+  if (Opc != 0 && (V != 1 || Opc != 2)) {
+    // It's a load, the MCInst gets: Rt, Rn_wb, Rn, Imm
+    DecodeGPR64xspRegisterClass(Inst, Rn, Address, Decoder);
+  }
+
+  DecodeGPR64xspRegisterClass(Inst, Rn, Address, Decoder);
+
+  Inst.addOperand(MCOperand::CreateImm(Imm9));
+
+  // N.b. The official documentation says undpredictable if Rt == Rn, but this
+  // takes place at the architectural rather than encoding level:
+  //
+  // "STR xzr, [sp], #4" is perfectly valid.
+  if (V == 0 && Rt == Rn && Rn != 31)
+    return MCDisassembler::SoftFail;
+  else
+    return MCDisassembler::Success;
+}
+
+static MCDisassembler *createAArch64Disassembler(const Target &T,
+                                                 const MCSubtargetInfo &STI) {
+  return new AArch64Disassembler(STI, T.createMCRegInfo(""));
+}
+
+extern "C" void LLVMInitializeAArch64Disassembler() {
+  TargetRegistry::RegisterMCDisassembler(TheAArch64Target,
+                                         createAArch64Disassembler);
+}
+
+
diff --git a/lib/Target/AArch64/Disassembler/CMakeLists.txt b/lib/Target/AArch64/Disassembler/CMakeLists.txt
new file mode 100644
index 000000000000..d4bd163dad60
--- /dev/null
+++ b/lib/Target/AArch64/Disassembler/CMakeLists.txt
@@ -0,0 +1,7 @@
+include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
+
+add_llvm_library(LLVMAArch64Disassembler
+  AArch64Disassembler.cpp
+  )
+
+add_dependencies(LLVMAArch64Disassembler AArch64CommonTableGen)
diff --git a/lib/Target/AArch64/Disassembler/LLVMBuild.txt b/lib/Target/AArch64/Disassembler/LLVMBuild.txt
new file mode 100644
index 000000000000..a93e343886d0
--- /dev/null
+++ b/lib/Target/AArch64/Disassembler/LLVMBuild.txt
@@ -0,0 +1,24 @@
+;===- ./lib/Target/AArch64/Disassembler/LLVMBuild.txt ----------*- Conf -*--===;
+;
+;                     The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+;   http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[component_0]
+type = Library
+name = AArch64Disassembler
+parent = AArch64
+required_libraries = AArch64CodeGen AArch64Desc AArch64Info AArch64Utils MC Support
+add_to_library_groups = AArch64
+
diff --git a/lib/Target/AArch64/Disassembler/Makefile b/lib/Target/AArch64/Disassembler/Makefile
new file mode 100644
index 000000000000..5c861207f836
--- /dev/null
+++ b/lib/Target/AArch64/Disassembler/Makefile
@@ -0,0 +1,16 @@
+##===- lib/Target/AArch64/Disassembler/Makefile ------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+
+LEVEL = ../../../..
+LIBRARYNAME = LLVMAArch64Disassembler
+
+# Hack: we need to include 'main' target directory to grab private headers
+CPPFLAGS = -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/Target/AArch64/InstPrinter/AArch64InstPrinter.cpp b/lib/Target/AArch64/InstPrinter/AArch64InstPrinter.cpp
new file mode 100644
index 000000000000..82ce80c8b1a1
--- /dev/null
+++ b/lib/Target/AArch64/InstPrinter/AArch64InstPrinter.cpp
@@ -0,0 +1,408 @@
+//==-- AArch64InstPrinter.cpp - Convert AArch64 MCInst to assembly syntax --==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class prints an AArch64 MCInst to a .s file.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "asm-printer"
+#include "AArch64InstPrinter.h"
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/Format.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+#define GET_INSTRUCTION_NAME
+#define PRINT_ALIAS_INSTR
+#include "AArch64GenAsmWriter.inc"
+
+static int64_t unpackSignedImm(int BitWidth, uint64_t Value) {
+  assert(!(Value & ~((1ULL << BitWidth)-1)) && "immediate not n-bit");
+  if (Value & (1ULL <<  (BitWidth - 1)))
+    return static_cast<int64_t>(Value) - (1LL << BitWidth);
+  else
+    return Value;
+}
+
+AArch64InstPrinter::AArch64InstPrinter(const MCAsmInfo &MAI,
+                                       const MCInstrInfo &MII,
+                                       const MCRegisterInfo &MRI,
+                                       const MCSubtargetInfo &STI) :
+  MCInstPrinter(MAI, MII, MRI) {
+  // Initialize the set of available features.
+  setAvailableFeatures(STI.getFeatureBits());
+}
+
+void AArch64InstPrinter::printRegName(raw_ostream &OS, unsigned RegNo) const {
+  OS << getRegisterName(RegNo);
+}
+
+void
+AArch64InstPrinter::printOffsetSImm9Operand(const MCInst *MI,
+                                              unsigned OpNum, raw_ostream &O) {
+  const MCOperand &MOImm = MI->getOperand(OpNum);
+  int32_t Imm = unpackSignedImm(9, MOImm.getImm());
+
+  O << '#' << Imm;
+}
+
+void
+AArch64InstPrinter::printAddrRegExtendOperand(const MCInst *MI, unsigned OpNum,
+                                          raw_ostream &O, unsigned MemSize,
+                                          unsigned RmSize) {
+  unsigned ExtImm = MI->getOperand(OpNum).getImm();
+  unsigned OptionHi = ExtImm >> 1;
+  unsigned S = ExtImm & 1;
+  bool IsLSL = OptionHi == 1 && RmSize == 64;
+
+  const char *Ext;
+  switch (OptionHi) {
+  case 1:
+    Ext = (RmSize == 32) ? "uxtw" : "lsl";
+    break;
+  case 3:
+    Ext = (RmSize == 32) ? "sxtw" : "sxtx";
+    break;
+  default:
+    llvm_unreachable("Incorrect Option on load/store (reg offset)");
+  }
+  O << Ext;
+
+  if (S) {
+    unsigned ShiftAmt = Log2_32(MemSize);
+    O << " #" << ShiftAmt;
+  } else if (IsLSL) {
+    O << " #0";
+  }
+}
+
+void
+AArch64InstPrinter::printAddSubImmLSL0Operand(const MCInst *MI,
+                                              unsigned OpNum, raw_ostream &O) {
+  const MCOperand &Imm12Op = MI->getOperand(OpNum);
+
+  if (Imm12Op.isImm()) {
+    int64_t Imm12 = Imm12Op.getImm();
+    assert(Imm12 >= 0 && "Invalid immediate for add/sub imm");
+    O << "#" << Imm12;
+  } else {
+    assert(Imm12Op.isExpr() && "Unexpected shift operand type");
+    O << "#" << *Imm12Op.getExpr();
+  }
+}
+
+void
+AArch64InstPrinter::printAddSubImmLSL12Operand(const MCInst *MI, unsigned OpNum,
+                                               raw_ostream &O) {
+
+  printAddSubImmLSL0Operand(MI, OpNum, O);
+
+  O << ", lsl #12";
+}
+
+void
+AArch64InstPrinter::printBareImmOperand(const MCInst *MI, unsigned OpNum,
+                                        raw_ostream &O) {
+  const MCOperand &MO = MI->getOperand(OpNum);
+  O << MO.getImm();
+}
+
+template<unsigned RegWidth> void
+AArch64InstPrinter::printBFILSBOperand(const MCInst *MI, unsigned OpNum,
+                                       raw_ostream &O) {
+  const MCOperand &ImmROp = MI->getOperand(OpNum);
+  unsigned LSB = ImmROp.getImm() == 0 ? 0 : RegWidth - ImmROp.getImm();
+
+  O << '#' << LSB;
+}
+
+void AArch64InstPrinter::printBFIWidthOperand(const MCInst *MI, unsigned OpNum,
+                                              raw_ostream &O) {
+  const MCOperand &ImmSOp = MI->getOperand(OpNum);
+  unsigned Width = ImmSOp.getImm() + 1;
+
+  O << '#' << Width;
+}
+
+void
+AArch64InstPrinter::printBFXWidthOperand(const MCInst *MI, unsigned OpNum,
+                                         raw_ostream &O) {
+  const MCOperand &ImmSOp = MI->getOperand(OpNum);
+  const MCOperand &ImmROp = MI->getOperand(OpNum - 1);
+
+  unsigned ImmR = ImmROp.getImm();
+  unsigned ImmS = ImmSOp.getImm();
+
+  assert(ImmS >= ImmR && "Invalid ImmR, ImmS combination for bitfield extract");
+
+  O << '#' << (ImmS - ImmR + 1);
+}
+
+void
+AArch64InstPrinter::printCRxOperand(const MCInst *MI, unsigned OpNum,
+                                    raw_ostream &O) {
+    const MCOperand &CRx = MI->getOperand(OpNum);
+
+    O << 'c' << CRx.getImm();
+}
+
+
+void
+AArch64InstPrinter::printCVTFixedPosOperand(const MCInst *MI, unsigned OpNum,
+                                            raw_ostream &O) {
+    const MCOperand &ScaleOp = MI->getOperand(OpNum);
+
+    O << '#' << (64 - ScaleOp.getImm());
+}
+
+
+void AArch64InstPrinter::printFPImmOperand(const MCInst *MI, unsigned OpNum,
+                                           raw_ostream &o) {
+  const MCOperand &MOImm8 = MI->getOperand(OpNum);
+
+  assert(MOImm8.isImm()
+         && "Immediate operand required for floating-point immediate inst");
+
+  uint32_t Imm8 = MOImm8.getImm();
+  uint32_t Fraction = Imm8 & 0xf;
+  uint32_t Exponent = (Imm8 >> 4) & 0x7;
+  uint32_t Negative = (Imm8 >> 7) & 0x1;
+
+  float Val = 1.0f + Fraction / 16.0f;
+
+  // That is:
+  // 000 -> 2^1,  001 -> 2^2,  010 -> 2^3,  011 -> 2^4,
+  // 100 -> 2^-3, 101 -> 2^-2, 110 -> 2^-1, 111 -> 2^0
+  if (Exponent & 0x4) {
+    Val /= 1 << (7 - Exponent);
+  } else {
+    Val *= 1 << (Exponent + 1);
+  }
+
+  Val = Negative ? -Val : Val;
+
+  o << '#' << format("%.8f", Val);
+}
+
+void AArch64InstPrinter::printFPZeroOperand(const MCInst *MI, unsigned OpNum,
+                                            raw_ostream &o) {
+  o << "#0.0";
+}
+
+void
+AArch64InstPrinter::printCondCodeOperand(const MCInst *MI, unsigned OpNum,
+                                         raw_ostream &O) {
+  const MCOperand &MO = MI->getOperand(OpNum);
+
+  O << A64CondCodeToString(static_cast<A64CC::CondCodes>(MO.getImm()));
+}
+
+template <unsigned field_width, unsigned scale> void
+AArch64InstPrinter::printLabelOperand(const MCInst *MI, unsigned OpNum,
+                                            raw_ostream &O) {
+  const MCOperand &MO = MI->getOperand(OpNum);
+
+  if (!MO.isImm()) {
+    printOperand(MI, OpNum, O);
+    return;
+  }
+
+  // The immediate of LDR (lit) instructions is a signed 19-bit immediate, which
+  // is multiplied by 4 (because all A64 instructions are 32-bits wide).
+  uint64_t UImm = MO.getImm();
+  uint64_t Sign = UImm & (1LL << (field_width - 1));
+  int64_t SImm = scale * ((UImm & ~Sign) - Sign);
+
+  O << "#" << SImm;
+}
+
+template<unsigned RegWidth> void
+AArch64InstPrinter::printLogicalImmOperand(const MCInst *MI, unsigned OpNum,
+                                           raw_ostream &O) {
+  const MCOperand &MO = MI->getOperand(OpNum);
+  uint64_t Val;
+  A64Imms::isLogicalImmBits(RegWidth, MO.getImm(), Val);
+  O << "#0x";
+  O.write_hex(Val);
+}
+
+void
+AArch64InstPrinter::printOffsetUImm12Operand(const MCInst *MI, unsigned OpNum,
+                                               raw_ostream &O, int MemSize) {
+  const MCOperand &MOImm = MI->getOperand(OpNum);
+
+  if (MOImm.isImm()) {
+    uint32_t Imm = MOImm.getImm() * MemSize;
+
+    O << "#" << Imm;
+  } else {
+    O << "#" << *MOImm.getExpr();
+  }
+}
+
+void
+AArch64InstPrinter::printShiftOperand(const MCInst *MI,  unsigned OpNum,
+                                      raw_ostream &O,
+                                      A64SE::ShiftExtSpecifiers Shift) {
+    const MCOperand &MO = MI->getOperand(OpNum);
+
+    // LSL #0 is not printed
+    if (Shift == A64SE::LSL && MO.isImm() && MO.getImm() == 0)
+        return;
+
+    switch (Shift) {
+    case A64SE::LSL: O << "lsl"; break;
+    case A64SE::LSR: O << "lsr"; break;
+    case A64SE::ASR: O << "asr"; break;
+    case A64SE::ROR: O << "ror"; break;
+    default: llvm_unreachable("Invalid shift specifier in logical instruction");
+    }
+
+  O << " #" << MO.getImm();
+}
+
+void
+AArch64InstPrinter::printMoveWideImmOperand(const MCInst *MI,  unsigned OpNum,
+                                            raw_ostream &O) {
+  const MCOperand &UImm16MO = MI->getOperand(OpNum);
+  const MCOperand &ShiftMO = MI->getOperand(OpNum + 1);
+
+  if (UImm16MO.isImm()) {
+    O << '#' << UImm16MO.getImm();
+
+    if (ShiftMO.getImm() != 0)
+      O << ", lsl #" << (ShiftMO.getImm() * 16);
+
+    return;
+  }
+
+  O << "#" << *UImm16MO.getExpr();
+}
+
+void AArch64InstPrinter::printNamedImmOperand(const NamedImmMapper &Mapper,
+                                              const MCInst *MI, unsigned OpNum,
+                                              raw_ostream &O) {
+  bool ValidName;
+  const MCOperand &MO = MI->getOperand(OpNum);
+  StringRef Name = Mapper.toString(MO.getImm(), ValidName);
+
+  if (ValidName)
+    O << Name;
+  else
+    O << '#' << MO.getImm();
+}
+
+void
+AArch64InstPrinter::printSysRegOperand(const A64SysReg::SysRegMapper &Mapper,
+                                       const MCInst *MI, unsigned OpNum,
+                                       raw_ostream &O) {
+  const MCOperand &MO = MI->getOperand(OpNum);
+
+  bool ValidName;
+  std::string Name = Mapper.toString(MO.getImm(), ValidName);
+  if (ValidName) {
+    O << Name;
+    return;
+  }
+}
+
+
+void AArch64InstPrinter::printRegExtendOperand(const MCInst *MI,
+                                               unsigned OpNum,
+                                               raw_ostream &O,
+                                               A64SE::ShiftExtSpecifiers Ext) {
+  // FIXME: In principle TableGen should be able to detect this itself far more
+  // easily. We will only accumulate more of these hacks.
+  unsigned Reg0 = MI->getOperand(0).getReg();
+  unsigned Reg1 = MI->getOperand(1).getReg();
+
+  if (isStackReg(Reg0) || isStackReg(Reg1)) {
+    A64SE::ShiftExtSpecifiers LSLEquiv;
+
+    if (Reg0 == AArch64::XSP || Reg1 == AArch64::XSP)
+      LSLEquiv = A64SE::UXTX;
+    else
+      LSLEquiv = A64SE::UXTW;
+
+    if (Ext == LSLEquiv) {
+      O << "lsl #" << MI->getOperand(OpNum).getImm();
+      return;
+    }
+  }
+
+  switch (Ext) {
+  case A64SE::UXTB: O << "uxtb"; break;
+  case A64SE::UXTH: O << "uxth"; break;
+  case A64SE::UXTW: O << "uxtw"; break;
+  case A64SE::UXTX: O << "uxtx"; break;
+  case A64SE::SXTB: O << "sxtb"; break;
+  case A64SE::SXTH: O << "sxth"; break;
+  case A64SE::SXTW: O << "sxtw"; break;
+  case A64SE::SXTX: O << "sxtx"; break;
+  default: llvm_unreachable("Unexpected shift type for printing");
+  }
+
+  const MCOperand &MO = MI->getOperand(OpNum);
+  if (MO.getImm() != 0)
+    O << " #" << MO.getImm();
+}
+
+template<int MemScale> void
+AArch64InstPrinter::printSImm7ScaledOperand(const MCInst *MI, unsigned OpNum,
+                                      raw_ostream &O) {
+  const MCOperand &MOImm = MI->getOperand(OpNum);
+  int32_t Imm = unpackSignedImm(7, MOImm.getImm());
+
+  O << "#" << (Imm * MemScale);
+}
+
+void AArch64InstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
+                                      raw_ostream &O) {
+  const MCOperand &Op = MI->getOperand(OpNo);
+  if (Op.isReg()) {
+    unsigned Reg = Op.getReg();
+    O << getRegisterName(Reg);
+  } else if (Op.isImm()) {
+    O << '#' << Op.getImm();
+  } else {
+    assert(Op.isExpr() && "unknown operand kind in printOperand");
+    // If a symbolic branch target was added as a constant expression then print
+    // that address in hex.
+    const MCConstantExpr *BranchTarget = dyn_cast<MCConstantExpr>(Op.getExpr());
+    int64_t Address;
+    if (BranchTarget && BranchTarget->EvaluateAsAbsolute(Address)) {
+      O << "0x";
+      O.write_hex(Address);
+    }
+    else {
+      // Otherwise, just print the expression.
+      O << *Op.getExpr();
+    }
+  }
+}
+
+
+void AArch64InstPrinter::printInst(const MCInst *MI, raw_ostream &O,
+                                   StringRef Annot) {
+  if (MI->getOpcode() == AArch64::TLSDESCCALL) {
+    // This is a special assembler directive which applies an
+    // R_AARCH64_TLSDESC_CALL to the following (BLR) instruction. It has a fixed
+    // form outside the normal TableGenerated scheme.
+    O << "\t.tlsdesccall " << *MI->getOperand(0).getExpr();
+  } else if (!printAliasInstr(MI, O))
+    printInstruction(MI, O);
+
+  printAnnotation(O, Annot);
+}
diff --git a/lib/Target/AArch64/InstPrinter/AArch64InstPrinter.h b/lib/Target/AArch64/InstPrinter/AArch64InstPrinter.h
new file mode 100644
index 000000000000..639fa869c016
--- /dev/null
+++ b/lib/Target/AArch64/InstPrinter/AArch64InstPrinter.h
@@ -0,0 +1,172 @@
+//===-- AArch64InstPrinter.h - Convert AArch64 MCInst to assembly syntax --===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class prints an AArch64 MCInst to a .s file.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_AARCH64INSTPRINTER_H
+#define LLVM_AARCH64INSTPRINTER_H
+
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/MC/MCInstPrinter.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+
+namespace llvm {
+
+class MCOperand;
+
+class AArch64InstPrinter : public MCInstPrinter {
+public:
+  AArch64InstPrinter(const MCAsmInfo &MAI, const MCInstrInfo &MII,
+                     const MCRegisterInfo &MRI, const MCSubtargetInfo &STI);
+
+  // Autogenerated by tblgen
+  void printInstruction(const MCInst *MI, raw_ostream &O);
+  bool printAliasInstr(const MCInst *MI, raw_ostream &O);
+  static const char *getRegisterName(unsigned RegNo);
+  static const char *getInstructionName(unsigned Opcode);
+
+  void printRegName(raw_ostream &O, unsigned RegNum) const;
+
+  template<unsigned MemSize, unsigned RmSize>
+  void printAddrRegExtendOperand(const MCInst *MI, unsigned OpNum,
+                                 raw_ostream &O) {
+    printAddrRegExtendOperand(MI, OpNum, O, MemSize, RmSize);
+  }
+
+
+  void printAddrRegExtendOperand(const MCInst *MI, unsigned OpNum,
+                                 raw_ostream &O, unsigned MemSize,
+                                 unsigned RmSize);
+
+  void printAddSubImmLSL0Operand(const MCInst *MI,
+                                 unsigned OpNum, raw_ostream &O);
+  void printAddSubImmLSL12Operand(const MCInst *MI,
+                                  unsigned OpNum, raw_ostream &O);
+
+  void printBareImmOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+
+  template<unsigned RegWidth>
+  void printBFILSBOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+  void printBFIWidthOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+  void printBFXWidthOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+
+
+  void printCondCodeOperand(const MCInst *MI, unsigned OpNum,
+                            raw_ostream &O);
+
+  void printCRxOperand(const MCInst *MI, unsigned OpNum,
+                       raw_ostream &O);
+
+  void printCVTFixedPosOperand(const MCInst *MI, unsigned OpNum,
+                               raw_ostream &O);
+
+  void printFPImmOperand(const MCInst *MI, unsigned OpNum, raw_ostream &o);
+
+  void printFPZeroOperand(const MCInst *MI, unsigned OpNum, raw_ostream &o);
+
+  template<int MemScale>
+  void printOffsetUImm12Operand(const MCInst *MI,
+                                  unsigned OpNum, raw_ostream &o) {
+    printOffsetUImm12Operand(MI, OpNum, o, MemScale);
+  }
+
+  void printOffsetUImm12Operand(const MCInst *MI, unsigned OpNum,
+                                  raw_ostream &o, int MemScale);
+
+  template<unsigned field_width, unsigned scale>
+  void printLabelOperand(const MCInst *MI, unsigned OpNum,
+                         raw_ostream &O);
+
+  template<unsigned RegWidth>
+  void printLogicalImmOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+
+  template<typename SomeNamedImmMapper>
+  void printNamedImmOperand(const MCInst *MI, unsigned OpNum,
+                            raw_ostream &O) {
+    printNamedImmOperand(SomeNamedImmMapper(), MI, OpNum, O);
+  }
+
+  void printNamedImmOperand(const NamedImmMapper &Mapper,
+                            const MCInst *MI, unsigned OpNum,
+                            raw_ostream &O);
+
+  void printSysRegOperand(const A64SysReg::SysRegMapper &Mapper,
+                          const MCInst *MI, unsigned OpNum,
+                          raw_ostream &O);
+
+  void printMRSOperand(const MCInst *MI, unsigned OpNum,
+                       raw_ostream &O) {
+    printSysRegOperand(A64SysReg::MRSMapper(), MI, OpNum, O);
+  }
+
+  void printMSROperand(const MCInst *MI, unsigned OpNum,
+                       raw_ostream &O) {
+    printSysRegOperand(A64SysReg::MSRMapper(), MI, OpNum, O);
+  }
+
+  void printShiftOperand(const char *name, const MCInst *MI,
+                         unsigned OpIdx, raw_ostream &O);
+
+  void printLSLOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+
+  void printLSROperand(const MCInst *MI, unsigned OpNum, raw_ostream &O) {
+    printShiftOperand("lsr", MI, OpNum, O);
+  }
+  void printASROperand(const MCInst *MI, unsigned OpNum, raw_ostream &O) {
+    printShiftOperand("asr", MI, OpNum, O);
+  }
+  void printROROperand(const MCInst *MI, unsigned OpNum, raw_ostream &O) {
+    printShiftOperand("ror", MI, OpNum, O);
+  }
+
+  template<A64SE::ShiftExtSpecifiers Shift>
+  void printShiftOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O) {
+    printShiftOperand(MI, OpNum, O, Shift);
+  }
+
+  void printShiftOperand(const MCInst *MI, unsigned OpNum,
+                         raw_ostream &O, A64SE::ShiftExtSpecifiers Sh);
+
+
+  void printMoveWideImmOperand(const  MCInst *MI, unsigned OpNum,
+                               raw_ostream &O);
+
+  template<int MemSize> void
+  printSImm7ScaledOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+
+  void printOffsetSImm9Operand(const MCInst *MI, unsigned OpNum,
+                               raw_ostream &O);
+
+  void printPRFMOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+
+  template<A64SE::ShiftExtSpecifiers EXT>
+  void printRegExtendOperand(const MCInst *MI, unsigned OpNum,
+                             raw_ostream &O) {
+    printRegExtendOperand(MI, OpNum, O, EXT);
+  }
+
+  void printRegExtendOperand(const MCInst *MI, unsigned OpNum,
+                             raw_ostream &O, A64SE::ShiftExtSpecifiers Ext);
+
+  void printOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  virtual void printInst(const MCInst *MI, raw_ostream &O, StringRef Annot);
+
+  bool isStackReg(unsigned RegNo) {
+    return RegNo == AArch64::XSP || RegNo == AArch64::WSP;
+  }
+
+
+};
+
+}
+
+#endif
diff --git a/lib/Target/AArch64/InstPrinter/CMakeLists.txt b/lib/Target/AArch64/InstPrinter/CMakeLists.txt
new file mode 100644
index 000000000000..d4b980a94d9b
--- /dev/null
+++ b/lib/Target/AArch64/InstPrinter/CMakeLists.txt
@@ -0,0 +1,8 @@
+include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
+
+add_llvm_library(LLVMAArch64AsmPrinter
+  AArch64InstPrinter.cpp
+  )
+
+add_dependencies(LLVMAArch64AsmPrinter AArch64CommonTableGen)
+
diff --git a/lib/Target/AArch64/InstPrinter/LLVMBuild.txt b/lib/Target/AArch64/InstPrinter/LLVMBuild.txt
new file mode 100644
index 000000000000..4836c7c45d44
--- /dev/null
+++ b/lib/Target/AArch64/InstPrinter/LLVMBuild.txt
@@ -0,0 +1,24 @@
+;===- ./lib/Target/AArch64/InstPrinter/LLVMBuild.txt -----------*- Conf -*--===;
+;
+;                     The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+;   http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[component_0]
+type = Library
+name = AArch64AsmPrinter
+parent = AArch64
+required_libraries = AArch64Utils MC Support
+add_to_library_groups = AArch64
+
diff --git a/lib/Target/AArch64/InstPrinter/Makefile b/lib/Target/AArch64/InstPrinter/Makefile
new file mode 100644
index 000000000000..1c36a8dea798
--- /dev/null
+++ b/lib/Target/AArch64/InstPrinter/Makefile
@@ -0,0 +1,15 @@
+##===- lib/Target/AArch64/AsmPrinter/Makefile --------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+LEVEL = ../../../..
+LIBRARYNAME = LLVMAArch64AsmPrinter
+
+# Hack: we need to include 'main' target directory to grab private headers
+CPP.Flags += -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/Target/AArch64/LLVMBuild.txt b/lib/Target/AArch64/LLVMBuild.txt
new file mode 100644
index 000000000000..3b296fdddc04
--- /dev/null
+++ b/lib/Target/AArch64/LLVMBuild.txt
@@ -0,0 +1,36 @@
+;===- ./lib/Target/AArch64/LLVMBuild.txt -----------------------*- Conf -*--===;
+;
+;                     The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+;   http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[common]
+subdirectories = AsmParser Disassembler InstPrinter MCTargetDesc TargetInfo Utils
+
+[component_0]
+type = TargetGroup
+name = AArch64
+parent = Target
+has_asmparser = 1
+has_asmprinter = 1
+has_disassembler = 1
+;has_jit = 1
+
+[component_1]
+type = Library
+name = AArch64CodeGen
+parent = AArch64
+required_libraries = AArch64AsmPrinter AArch64Desc AArch64Info AsmPrinter CodeGen Core MC SelectionDAG Support Target
+add_to_library_groups = AArch64
+
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64AsmBackend.cpp b/lib/Target/AArch64/MCTargetDesc/AArch64AsmBackend.cpp
new file mode 100644
index 000000000000..a3373b1087bb
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64AsmBackend.cpp
@@ -0,0 +1,585 @@
+//===-- AArch64AsmBackend.cpp - AArch64 Assembler Backend -----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the AArch64 implementation of the MCAsmBackend class,
+// which is principally concerned with relaxation of the various fixup kinds.
+//
+//===----------------------------------------------------------------------===//
+
+#include "MCTargetDesc/AArch64FixupKinds.h"
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "llvm/MC/MCAsmBackend.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/MC/MCELFObjectWriter.h"
+#include "llvm/MC/MCFixupKindInfo.h"
+#include "llvm/MC/MCObjectWriter.h"
+#include "llvm/Support/ELF.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+using namespace llvm;
+
+namespace {
+class AArch64AsmBackend : public MCAsmBackend {
+  const MCSubtargetInfo* STI;
+public:
+  AArch64AsmBackend(const Target &T, const StringRef TT)
+    : MCAsmBackend(),
+      STI(AArch64_MC::createAArch64MCSubtargetInfo(TT, "", ""))
+    {}
+
+
+  ~AArch64AsmBackend() {
+    delete STI;
+  }
+
+  bool writeNopData(uint64_t Count, MCObjectWriter *OW) const;
+
+  virtual void processFixupValue(const MCAssembler &Asm,
+                                 const MCAsmLayout &Layout,
+                                 const MCFixup &Fixup, const MCFragment *DF,
+                                 MCValue &Target, uint64_t &Value,
+                                 bool &IsResolved);
+};
+} // end anonymous namespace
+
+void AArch64AsmBackend::processFixupValue(const MCAssembler &Asm,
+                                          const MCAsmLayout &Layout,
+                                          const MCFixup &Fixup,
+                                          const MCFragment *DF,
+                                          MCValue &Target, uint64_t &Value,
+                                          bool &IsResolved) {
+  // The ADRP instruction adds some multiple of 0x1000 to the current PC &
+  // ~0xfff. This means that the required offset to reach a symbol can vary by
+  // up to one step depending on where the ADRP is in memory. For example:
+  //
+  //     ADRP x0, there
+  //  there:
+  //
+  // If the ADRP occurs at address 0xffc then "there" will be at 0x1000 and
+  // we'll need that as an offset. At any other address "there" will be in the
+  // same page as the ADRP and the instruction should encode 0x0. Assuming the
+  // section isn't 0x1000-aligned, we therefore need to delegate this decision
+  // to the linker -- a relocation!
+  if ((uint32_t)Fixup.getKind() == AArch64::fixup_a64_adr_prel_page ||
+      (uint32_t)Fixup.getKind() == AArch64::fixup_a64_adr_prel_got_page ||
+      (uint32_t)Fixup.getKind() == AArch64::fixup_a64_adr_gottprel_page ||
+      (uint32_t)Fixup.getKind() == AArch64::fixup_a64_tlsdesc_adr_page)
+    IsResolved = false;
+}
+
+
+static uint64_t adjustFixupValue(unsigned Kind, uint64_t Value);
+
+namespace {
+
+class ELFAArch64AsmBackend : public AArch64AsmBackend {
+public:
+  uint8_t OSABI;
+  ELFAArch64AsmBackend(const Target &T, const StringRef TT,
+                       uint8_t _OSABI)
+    : AArch64AsmBackend(T, TT), OSABI(_OSABI) { }
+
+  bool fixupNeedsRelaxation(const MCFixup &Fixup,
+                            uint64_t Value,
+                            const MCRelaxableFragment *DF,
+                            const MCAsmLayout &Layout) const;
+
+  unsigned int getNumFixupKinds() const {
+    return AArch64::NumTargetFixupKinds;
+  }
+
+  const MCFixupKindInfo &getFixupKindInfo(MCFixupKind Kind) const {
+    const static MCFixupKindInfo Infos[AArch64::NumTargetFixupKinds] = {
+// This table *must* be in the order that the fixup_* kinds are defined in
+// AArch64FixupKinds.h.
+//
+// Name                   Offset (bits)    Size (bits)    Flags
+{ "fixup_a64_ld_prel",               0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_adr_prel",              0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_adr_prel_page",         0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_add_lo12",              0,    32,             0 },
+{ "fixup_a64_ldst8_lo12",            0,    32,             0 },
+{ "fixup_a64_ldst16_lo12",           0,    32,             0 },
+{ "fixup_a64_ldst32_lo12",           0,    32,             0 },
+{ "fixup_a64_ldst64_lo12",           0,    32,             0 },
+{ "fixup_a64_ldst128_lo12",          0,    32,             0 },
+{ "fixup_a64_tstbr",                 0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_condbr",                0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_uncondbr",              0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_call",                  0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_movw_uabs_g0",          0,    32,             0 },
+{ "fixup_a64_movw_uabs_g0_nc",       0,    32,             0 },
+{ "fixup_a64_movw_uabs_g1",          0,    32,             0 },
+{ "fixup_a64_movw_uabs_g1_nc",       0,    32,             0 },
+{ "fixup_a64_movw_uabs_g2",          0,    32,             0 },
+{ "fixup_a64_movw_uabs_g2_nc",       0,    32,             0 },
+{ "fixup_a64_movw_uabs_g3",          0,    32,             0 },
+{ "fixup_a64_movw_sabs_g0",          0,    32,             0 },
+{ "fixup_a64_movw_sabs_g1",          0,    32,             0 },
+{ "fixup_a64_movw_sabs_g2",          0,    32,             0 },
+{ "fixup_a64_adr_prel_got_page",     0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_ld64_got_lo12_nc",      0,    32,             0 },
+{ "fixup_a64_movw_dtprel_g2",        0,    32,             0 },
+{ "fixup_a64_movw_dtprel_g1",        0,    32,             0 },
+{ "fixup_a64_movw_dtprel_g1_nc",     0,    32,             0 },
+{ "fixup_a64_movw_dtprel_g0",        0,    32,             0 },
+{ "fixup_a64_movw_dtprel_g0_nc",     0,    32,             0 },
+{ "fixup_a64_add_dtprel_hi12",       0,    32,             0 },
+{ "fixup_a64_add_dtprel_lo12",       0,    32,             0 },
+{ "fixup_a64_add_dtprel_lo12_nc",    0,    32,             0 },
+{ "fixup_a64_ldst8_dtprel_lo12",     0,    32,             0 },
+{ "fixup_a64_ldst8_dtprel_lo12_nc",  0,    32,             0 },
+{ "fixup_a64_ldst16_dtprel_lo12",    0,    32,             0 },
+{ "fixup_a64_ldst16_dtprel_lo12_nc", 0,    32,             0 },
+{ "fixup_a64_ldst32_dtprel_lo12",    0,    32,             0 },
+{ "fixup_a64_ldst32_dtprel_lo12_nc", 0,    32,             0 },
+{ "fixup_a64_ldst64_dtprel_lo12",    0,    32,             0 },
+{ "fixup_a64_ldst64_dtprel_lo12_nc", 0,    32,             0 },
+{ "fixup_a64_movw_gottprel_g1",      0,    32,             0 },
+{ "fixup_a64_movw_gottprel_g0_nc",   0,    32,             0 },
+{ "fixup_a64_adr_gottprel_page",     0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_ld64_gottprel_lo12_nc", 0,    32,             0 },
+{ "fixup_a64_ld_gottprel_prel19",    0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_movw_tprel_g2",         0,    32,             0 },
+{ "fixup_a64_movw_tprel_g1",         0,    32,             0 },
+{ "fixup_a64_movw_tprel_g1_nc",      0,    32,             0 },
+{ "fixup_a64_movw_tprel_g0",         0,    32,             0 },
+{ "fixup_a64_movw_tprel_g0_nc",      0,    32,             0 },
+{ "fixup_a64_add_tprel_hi12",        0,    32,             0 },
+{ "fixup_a64_add_tprel_lo12",        0,    32,             0 },
+{ "fixup_a64_add_tprel_lo12_nc",     0,    32,             0 },
+{ "fixup_a64_ldst8_tprel_lo12",      0,    32,             0 },
+{ "fixup_a64_ldst8_tprel_lo12_nc",   0,    32,             0 },
+{ "fixup_a64_ldst16_tprel_lo12",     0,    32,             0 },
+{ "fixup_a64_ldst16_tprel_lo12_nc",  0,    32,             0 },
+{ "fixup_a64_ldst32_tprel_lo12",     0,    32,             0 },
+{ "fixup_a64_ldst32_tprel_lo12_nc",  0,    32,             0 },
+{ "fixup_a64_ldst64_tprel_lo12",     0,    32,             0 },
+{ "fixup_a64_ldst64_tprel_lo12_nc",  0,    32,             0 },
+{ "fixup_a64_tlsdesc_adr_page",      0,    32, MCFixupKindInfo::FKF_IsPCRel },
+{ "fixup_a64_tlsdesc_ld64_lo12_nc",  0,    32,             0 },
+{ "fixup_a64_tlsdesc_add_lo12_nc",   0,    32,             0 },
+{ "fixup_a64_tlsdesc_call",          0,     0,             0 }
+    };
+    if (Kind < FirstTargetFixupKind)
+      return MCAsmBackend::getFixupKindInfo(Kind);
+
+    assert(unsigned(Kind - FirstTargetFixupKind) < getNumFixupKinds() &&
+           "Invalid kind!");
+    return Infos[Kind - FirstTargetFixupKind];
+  }
+
+  void applyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize,
+                  uint64_t Value) const {
+    unsigned NumBytes = getFixupKindInfo(Fixup.getKind()).TargetSize / 8;
+    Value = adjustFixupValue(Fixup.getKind(), Value);
+    if (!Value) return;           // Doesn't change encoding.
+
+    unsigned Offset = Fixup.getOffset();
+    assert(Offset + NumBytes <= DataSize && "Invalid fixup offset!");
+
+    // For each byte of the fragment that the fixup touches, mask in the bits
+    // from the fixup value.
+    for (unsigned i = 0; i != NumBytes; ++i) {
+      Data[Offset + i] |= uint8_t((Value >> (i * 8)) & 0xff);
+    }
+  }
+
+  bool mayNeedRelaxation(const MCInst&) const {
+    return false;
+  }
+
+  void relaxInstruction(const MCInst&, llvm::MCInst&) const {
+    llvm_unreachable("Cannot relax instructions");
+  }
+
+  MCObjectWriter *createObjectWriter(raw_ostream &OS) const {
+    return createAArch64ELFObjectWriter(OS, OSABI);
+  }
+};
+
+} // end anonymous namespace
+
+bool
+ELFAArch64AsmBackend::fixupNeedsRelaxation(const MCFixup &Fixup,
+                                           uint64_t Value,
+                                           const MCRelaxableFragment *DF,
+                                           const MCAsmLayout &Layout) const {
+  // Correct for now. With all instructions 32-bit only very low-level
+  // considerations could make you select something which may fail.
+  return false;
+}
+
+
+bool AArch64AsmBackend::writeNopData(uint64_t Count, MCObjectWriter *OW) const {
+  // Can't emit NOP with size not multiple of 32-bits
+  if (Count % 4 != 0)
+    return false;
+
+  uint64_t NumNops = Count / 4;
+  for (uint64_t i = 0; i != NumNops; ++i)
+    OW->Write32(0xd503201f);
+
+  return true;
+}
+
+static unsigned ADRImmBits(unsigned Value) {
+  unsigned lo2 = Value & 0x3;
+  unsigned hi19 = (Value & 0x1fffff) >> 2;
+
+  return (hi19 << 5) | (lo2 << 29);
+}
+
+static uint64_t adjustFixupValue(unsigned Kind, uint64_t Value) {
+  switch (Kind) {
+  default:
+    llvm_unreachable("Unknown fixup kind!");
+  case FK_Data_2:
+    assert((int64_t)Value >= -32768 &&
+           (int64_t)Value <= 65536 &&
+           "Out of range ABS16 fixup");
+    return Value;
+  case FK_Data_4:
+    assert((int64_t)Value >= -(1LL << 31) &&
+           (int64_t)Value <= (1LL << 32) - 1 &&
+           "Out of range ABS32 fixup");
+    return Value;
+  case FK_Data_8:
+    return Value;
+
+  case AArch64::fixup_a64_ld_gottprel_prel19:
+    // R_AARCH64_LD_GOTTPREL_PREL19: Set a load-literal immediate to bits 1F
+    // FFFC of G(TPREL(S+A)) - P; check -2^20 <= X < 2^20.
+  case AArch64::fixup_a64_ld_prel:
+    // R_AARCH64_LD_PREL_LO19: Sets a load-literal (immediate) value to bits
+    // 1F FFFC of S+A-P, checking that -2^20 <= S+A-P < 2^20.
+    assert((int64_t)Value >= -(1LL << 20) &&
+           (int64_t)Value < (1LL << 20) && "Out of range LDR (lit) fixup");
+    return (Value & 0x1ffffc) << 3;
+
+  case AArch64::fixup_a64_adr_prel:
+    // R_AARCH64_ADR_PREL_LO21: Sets an ADR immediate value to bits 1F FFFF of
+    // the result of S+A-P, checking that -2^20 <= S+A-P < 2^20.
+    assert((int64_t)Value >= -(1LL << 20) &&
+           (int64_t)Value < (1LL << 20) && "Out of range ADR fixup");
+    return ADRImmBits(Value & 0x1fffff);
+
+  case AArch64::fixup_a64_adr_prel_page:
+    // R_AARCH64_ADR_PREL_PG_HI21: Sets an ADRP immediate value to bits 1 FFFF
+    // F000 of the result of the operation, checking that -2^32 <= result <
+    // 2^32.
+    assert((int64_t)Value >= -(1LL << 32) &&
+           (int64_t)Value < (1LL << 32) && "Out of range ADRP fixup");
+    return ADRImmBits((Value & 0x1fffff000ULL) >> 12);
+
+  case AArch64::fixup_a64_add_dtprel_hi12:
+    // R_AARCH64_TLSLD_ADD_DTPREL_LO12: Set an ADD immediate field to bits
+    // FF F000 of DTPREL(S+A), check 0 <= X < 2^24.
+  case AArch64::fixup_a64_add_tprel_hi12:
+    // R_AARCH64_TLSLD_ADD_TPREL_LO12: Set an ADD immediate field to bits
+    // FF F000 of TPREL(S+A), check 0 <= X < 2^24.
+    assert((int64_t)Value >= 0 &&
+           (int64_t)Value < (1LL << 24) && "Out of range ADD fixup");
+    return (Value & 0xfff000) >> 2;
+
+  case AArch64::fixup_a64_add_dtprel_lo12:
+    // R_AARCH64_TLSLD_ADD_DTPREL_LO12: Set an ADD immediate field to bits
+    // FFF of DTPREL(S+A), check 0 <= X < 2^12.
+  case AArch64::fixup_a64_add_tprel_lo12:
+    // R_AARCH64_TLSLD_ADD_TPREL_LO12: Set an ADD immediate field to bits
+    // FFF of TPREL(S+A), check 0 <= X < 2^12.
+    assert((int64_t)Value >= 0 &&
+           (int64_t)Value < (1LL << 12) && "Out of range ADD fixup");
+    // ... fallthrough to no-checking versions ...
+  case AArch64::fixup_a64_add_dtprel_lo12_nc:
+    // R_AARCH64_TLSLD_ADD_DTPREL_LO12_NC: Set an ADD immediate field to bits
+    // FFF of DTPREL(S+A) with no overflow check.
+  case AArch64::fixup_a64_add_tprel_lo12_nc:
+    // R_AARCH64_TLSLD_ADD_TPREL_LO12_NC: Set an ADD immediate field to bits
+    // FFF of TPREL(S+A) with no overflow check.
+  case AArch64::fixup_a64_tlsdesc_add_lo12_nc:
+    // R_AARCH64_TLSDESC_ADD_LO12_NC: Set an ADD immediate field to bits
+    // FFF of G(TLSDESC(S+A)), with no overflow check.
+  case AArch64::fixup_a64_add_lo12:
+    // R_AARCH64_ADD_ABS_LO12_NC: Sets an ADD immediate value to bits FFF of
+    // S+A, with no overflow check.
+    return (Value & 0xfff) << 10;
+
+  case AArch64::fixup_a64_ldst8_dtprel_lo12:
+    // R_AARCH64_TLSLD_LDST8_DTPREL_LO12: Set an LD/ST offset field to bits FFF
+    // of DTPREL(S+A), check 0 <= X < 2^12.
+  case AArch64::fixup_a64_ldst8_tprel_lo12:
+    // R_AARCH64_TLSLE_LDST8_TPREL_LO12: Set an LD/ST offset field to bits FFF
+    // of DTPREL(S+A), check 0 <= X < 2^12.
+    assert((int64_t) Value >= 0 &&
+           (int64_t) Value < (1LL << 12) && "Out of range LD/ST fixup");
+    // ... fallthrough to no-checking versions ...
+  case AArch64::fixup_a64_ldst8_dtprel_lo12_nc:
+    // R_AARCH64_TLSLD_LDST8_DTPREL_LO12: Set an LD/ST offset field to bits FFF
+    // of DTPREL(S+A), with no overflow check.
+  case AArch64::fixup_a64_ldst8_tprel_lo12_nc:
+    // R_AARCH64_TLSLD_LDST8_TPREL_LO12: Set an LD/ST offset field to bits FFF
+    // of TPREL(S+A), with no overflow check.
+  case AArch64::fixup_a64_ldst8_lo12:
+    // R_AARCH64_LDST8_ABS_LO12_NC: Sets an LD/ST immediate value to bits FFF
+    // of S+A, with no overflow check.
+    return (Value & 0xfff) << 10;
+
+  case AArch64::fixup_a64_ldst16_dtprel_lo12:
+    // R_AARCH64_TLSLD_LDST16_DTPREL_LO12: Set an LD/ST offset field to bits FFE
+    // of DTPREL(S+A), check 0 <= X < 2^12.
+  case AArch64::fixup_a64_ldst16_tprel_lo12:
+    // R_AARCH64_TLSLE_LDST16_TPREL_LO12: Set an LD/ST offset field to bits FFE
+    // of DTPREL(S+A), check 0 <= X < 2^12.
+    assert((int64_t) Value >= 0 &&
+           (int64_t) Value < (1LL << 12) && "Out of range LD/ST fixup");
+    // ... fallthrough to no-checking versions ...
+  case AArch64::fixup_a64_ldst16_dtprel_lo12_nc:
+    // R_AARCH64_TLSLD_LDST16_DTPREL_LO12: Set an LD/ST offset field to bits FFE
+    // of DTPREL(S+A), with no overflow check.
+  case AArch64::fixup_a64_ldst16_tprel_lo12_nc:
+    // R_AARCH64_TLSLD_LDST16_TPREL_LO12: Set an LD/ST offset field to bits FFE
+    // of TPREL(S+A), with no overflow check.
+  case AArch64::fixup_a64_ldst16_lo12:
+    // R_AARCH64_LDST16_ABS_LO12_NC: Sets an LD/ST immediate value to bits FFE
+    // of S+A, with no overflow check.
+    return (Value & 0xffe) << 9;
+
+  case AArch64::fixup_a64_ldst32_dtprel_lo12:
+    // R_AARCH64_TLSLD_LDST32_DTPREL_LO12: Set an LD/ST offset field to bits FFC
+    // of DTPREL(S+A), check 0 <= X < 2^12.
+  case AArch64::fixup_a64_ldst32_tprel_lo12:
+    // R_AARCH64_TLSLE_LDST32_TPREL_LO12: Set an LD/ST offset field to bits FFC
+    // of DTPREL(S+A), check 0 <= X < 2^12.
+    assert((int64_t) Value >= 0 &&
+           (int64_t) Value < (1LL << 12) && "Out of range LD/ST fixup");
+    // ... fallthrough to no-checking versions ...
+  case AArch64::fixup_a64_ldst32_dtprel_lo12_nc:
+    // R_AARCH64_TLSLD_LDST32_DTPREL_LO12: Set an LD/ST offset field to bits FFC
+    // of DTPREL(S+A), with no overflow check.
+  case AArch64::fixup_a64_ldst32_tprel_lo12_nc:
+    // R_AARCH64_TLSLD_LDST32_TPREL_LO12: Set an LD/ST offset field to bits FFC
+    // of TPREL(S+A), with no overflow check.
+  case AArch64::fixup_a64_ldst32_lo12:
+    // R_AARCH64_LDST32_ABS_LO12_NC: Sets an LD/ST immediate value to bits FFC
+    // of S+A, with no overflow check.
+    return (Value & 0xffc) << 8;
+
+  case AArch64::fixup_a64_ldst64_dtprel_lo12:
+    // R_AARCH64_TLSLD_LDST64_DTPREL_LO12: Set an LD/ST offset field to bits FF8
+    // of DTPREL(S+A), check 0 <= X < 2^12.
+  case AArch64::fixup_a64_ldst64_tprel_lo12:
+    // R_AARCH64_TLSLE_LDST64_TPREL_LO12: Set an LD/ST offset field to bits FF8
+    // of DTPREL(S+A), check 0 <= X < 2^12.
+    assert((int64_t) Value >= 0 &&
+           (int64_t) Value < (1LL << 12) && "Out of range LD/ST fixup");
+    // ... fallthrough to no-checking versions ...
+  case AArch64::fixup_a64_ldst64_dtprel_lo12_nc:
+    // R_AARCH64_TLSLD_LDST64_DTPREL_LO12: Set an LD/ST offset field to bits FF8
+    // of DTPREL(S+A), with no overflow check.
+  case AArch64::fixup_a64_ldst64_tprel_lo12_nc:
+    // R_AARCH64_TLSLD_LDST64_TPREL_LO12: Set an LD/ST offset field to bits FF8
+    // of TPREL(S+A), with no overflow check.
+  case AArch64::fixup_a64_ldst64_lo12:
+    // R_AARCH64_LDST64_ABS_LO12_NC: Sets an LD/ST immediate value to bits FF8
+    // of S+A, with no overflow check.
+    return (Value & 0xff8) << 7;
+
+  case AArch64::fixup_a64_ldst128_lo12:
+    // R_AARCH64_LDST128_ABS_LO12_NC: Sets an LD/ST immediate value to bits FF0
+    // of S+A, with no overflow check.
+    return (Value & 0xff0) << 6;
+
+  case AArch64::fixup_a64_movw_uabs_g0:
+    // R_AARCH64_MOVW_UABS_G0: Sets a MOVZ immediate field to bits FFFF of S+A
+    // with a check that S+A < 2^16
+    assert(Value <= 0xffff && "Out of range move wide fixup");
+    return (Value & 0xffff) << 5;
+
+  case AArch64::fixup_a64_movw_dtprel_g0_nc:
+    // R_AARCH64_TLSLD_MOVW_DTPREL_G0_NC: Sets a MOVK immediate field to bits
+    // FFFF of DTPREL(S+A) with no overflow check.
+  case AArch64::fixup_a64_movw_gottprel_g0_nc:
+    // R_AARCH64_TLSIE_MOVW_GOTTPREL_G0_NC: Sets a MOVK immediate field to bits
+    // FFFF of G(TPREL(S+A)) - GOT with no overflow check.
+  case AArch64::fixup_a64_movw_tprel_g0_nc:
+    // R_AARCH64_TLSLE_MOVW_TPREL_G0_NC: Sets a MOVK immediate field to bits
+    // FFFF of TPREL(S+A) with no overflow check.
+  case AArch64::fixup_a64_movw_uabs_g0_nc:
+    // R_AARCH64_MOVW_UABS_G0_NC: Sets a MOVK immediate field to bits FFFF of
+    // S+A with no overflow check.
+    return (Value & 0xffff) << 5;
+
+  case AArch64::fixup_a64_movw_uabs_g1:
+    // R_AARCH64_MOVW_UABS_G1: Sets a MOVZ immediate field to bits FFFF0000 of
+    // S+A with a check that S+A < 2^32
+    assert(Value <= 0xffffffffull && "Out of range move wide fixup");
+    return ((Value >> 16) & 0xffff) << 5;
+
+  case AArch64::fixup_a64_movw_dtprel_g1_nc:
+    // R_AARCH64_TLSLD_MOVW_DTPREL_G1_NC: Set a MOVK immediate field
+    // to bits FFFF0000 of DTPREL(S+A), with no overflow check.
+  case AArch64::fixup_a64_movw_tprel_g1_nc:
+    // R_AARCH64_TLSLD_MOVW_TPREL_G1_NC: Set a MOVK immediate field
+    // to bits FFFF0000 of TPREL(S+A), with no overflow check.
+  case AArch64::fixup_a64_movw_uabs_g1_nc:
+    // R_AARCH64_MOVW_UABS_G1_NC: Sets a MOVK immediate field to bits
+    // FFFF0000 of S+A with no overflow check.
+    return ((Value >> 16) & 0xffff) << 5;
+
+  case AArch64::fixup_a64_movw_uabs_g2:
+    // R_AARCH64_MOVW_UABS_G2: Sets a MOVZ immediate field to bits FFFF 0000
+    // 0000 of S+A with a check that S+A < 2^48
+    assert(Value <= 0xffffffffffffull && "Out of range move wide fixup");
+    return ((Value >> 32) & 0xffff) << 5;
+
+  case AArch64::fixup_a64_movw_uabs_g2_nc:
+    // R_AARCH64_MOVW_UABS_G2: Sets a MOVK immediate field to bits FFFF 0000
+    // 0000 of S+A with no overflow check.
+    return ((Value >> 32) & 0xffff) << 5;
+
+  case AArch64::fixup_a64_movw_uabs_g3:
+    // R_AARCH64_MOVW_UABS_G3: Sets a MOVZ immediate field to bits FFFF 0000
+    // 0000 0000 of S+A (no overflow check needed)
+    return ((Value >> 48) & 0xffff) << 5;
+
+  case AArch64::fixup_a64_movw_dtprel_g0:
+    // R_AARCH64_TLSLD_MOVW_DTPREL_G0: Set a MOV[NZ] immediate field
+    // to bits FFFF of DTPREL(S+A).
+  case AArch64::fixup_a64_movw_tprel_g0:
+    // R_AARCH64_TLSLE_MOVW_TPREL_G0: Set a MOV[NZ] immediate field to
+    // bits FFFF of TPREL(S+A).
+  case AArch64::fixup_a64_movw_sabs_g0: {
+    // R_AARCH64_MOVW_SABS_G0: Sets MOV[NZ] immediate field using bits FFFF of
+    // S+A (see notes below); check -2^16 <= S+A < 2^16. (notes say that we
+    // should convert between MOVN and MOVZ to achieve our goals).
+    int64_t Signed = Value;
+    assert(Signed >= -(1LL << 16) && Signed < (1LL << 16)
+           && "Out of range move wide fixup");
+    if (Signed >= 0) {
+      Value = (Value & 0xffff) << 5;
+      // Bit 30 converts the MOVN encoding into a MOVZ
+      Value |= 1 << 30;
+    } else {
+      // MCCodeEmitter should have encoded a MOVN, which is fine.
+      Value = (~Value & 0xffff) << 5;
+    }
+    return Value;
+  }
+
+  case AArch64::fixup_a64_movw_dtprel_g1:
+    // R_AARCH64_TLSLD_MOVW_DTPREL_G1: Set a MOV[NZ] immediate field
+    // to bits FFFF0000 of DTPREL(S+A).
+  case AArch64::fixup_a64_movw_gottprel_g1:
+    // R_AARCH64_TLSIE_MOVW_GOTTPREL_G1: Set a MOV[NZ] immediate field
+    // to bits FFFF0000 of G(TPREL(S+A)) - GOT.
+  case AArch64::fixup_a64_movw_tprel_g1:
+    // R_AARCH64_TLSLE_MOVW_TPREL_G1: Set a MOV[NZ] immediate field to
+    // bits FFFF0000 of TPREL(S+A).
+  case AArch64::fixup_a64_movw_sabs_g1: {
+    // R_AARCH64_MOVW_SABS_G1: Sets MOV[NZ] immediate field using bits FFFF 0000
+    // of S+A (see notes below); check -2^32 <= S+A < 2^32. (notes say that we
+    // should convert between MOVN and MOVZ to achieve our goals).
+    int64_t Signed = Value;
+    assert(Signed >= -(1LL << 32) && Signed < (1LL << 32)
+           && "Out of range move wide fixup");
+    if (Signed >= 0) {
+      Value = ((Value >> 16) & 0xffff) << 5;
+      // Bit 30 converts the MOVN encoding into a MOVZ
+      Value |= 1 << 30;
+    } else {
+      Value = ((~Value >> 16) & 0xffff) << 5;
+    }
+    return Value;
+  }
+
+  case AArch64::fixup_a64_movw_dtprel_g2:
+    // R_AARCH64_TLSLD_MOVW_DTPREL_G2: Set a MOV[NZ] immediate field
+    // to bits FFFF 0000 0000 of DTPREL(S+A).
+  case AArch64::fixup_a64_movw_tprel_g2:
+    // R_AARCH64_TLSLE_MOVW_TPREL_G2: Set a MOV[NZ] immediate field to
+    // bits FFFF 0000 0000 of TPREL(S+A).
+  case AArch64::fixup_a64_movw_sabs_g2: {
+    // R_AARCH64_MOVW_SABS_G2: Sets MOV[NZ] immediate field using bits FFFF 0000
+    // 0000 of S+A (see notes below); check -2^48 <= S+A < 2^48. (notes say that
+    // we should convert between MOVN and MOVZ to achieve our goals).
+    int64_t Signed = Value;
+    assert(Signed >= -(1LL << 48) && Signed < (1LL << 48)
+           && "Out of range move wide fixup");
+    if (Signed >= 0) {
+      Value = ((Value >> 32) & 0xffff) << 5;
+      // Bit 30 converts the MOVN encoding into a MOVZ
+      Value |= 1 << 30;
+    } else {
+      Value = ((~Value >> 32) & 0xffff) << 5;
+    }
+    return Value;
+  }
+
+  case AArch64::fixup_a64_tstbr:
+    // R_AARCH64_TSTBR14: Sets the immediate field of a TBZ/TBNZ instruction to
+    // bits FFFC of S+A-P, checking -2^15 <= S+A-P < 2^15.
+    assert((int64_t)Value >= -(1LL << 15) &&
+           (int64_t)Value < (1LL << 15) && "Out of range TBZ/TBNZ fixup");
+    return (Value & 0xfffc) << (5 - 2);
+
+  case AArch64::fixup_a64_condbr:
+    // R_AARCH64_CONDBR19: Sets the immediate field of a conditional branch
+    // instruction to bits 1FFFFC of S+A-P, checking -2^20 <= S+A-P < 2^20.
+    assert((int64_t)Value >= -(1LL << 20) &&
+           (int64_t)Value < (1LL << 20) && "Out of range B.cond fixup");
+    return (Value & 0x1ffffc) << (5 - 2);
+
+  case AArch64::fixup_a64_uncondbr:
+    // R_AARCH64_JUMP26 same as below (except to a linker, possibly).
+  case AArch64::fixup_a64_call:
+    // R_AARCH64_CALL26: Sets a CALL immediate field to bits FFFFFFC of S+A-P,
+    // checking that -2^27 <= S+A-P < 2^27.
+    assert((int64_t)Value >= -(1LL << 27) &&
+           (int64_t)Value < (1LL << 27) && "Out of range branch fixup");
+    return (Value & 0xffffffc) >> 2;
+
+  case AArch64::fixup_a64_adr_gottprel_page:
+    // R_AARCH64_TLSIE_ADR_GOTTPREL_PAGE21: Set an ADRP immediate field to bits
+    // 1FFFFF000 of Page(G(TPREL(S+A))) - Page(P); check -2^32 <= X < 2^32.
+  case AArch64::fixup_a64_tlsdesc_adr_page:
+    // R_AARCH64_TLSDESC_ADR_PAGE: Set an ADRP immediate field to bits 1FFFFF000
+    // of Page(G(TLSDESC(S+A))) - Page(P); check -2^32 <= X < 2^32.
+  case AArch64::fixup_a64_adr_prel_got_page:
+    // R_AARCH64_ADR_GOT_PAGE: Sets the immediate value of an ADRP to bits
+    // 1FFFFF000 of the operation, checking that -2^32 < Page(G(S))-Page(GOT) <
+    // 2^32.
+    assert((int64_t)Value >= -(1LL << 32) &&
+           (int64_t)Value < (1LL << 32) && "Out of range ADRP fixup");
+    return ADRImmBits((Value & 0x1fffff000ULL) >> 12);
+
+  case AArch64::fixup_a64_ld64_gottprel_lo12_nc:
+    // R_AARCH64_TLSIE_LD64_GOTTPREL_LO12_NC: Set an LD offset field to bits FF8
+    // of X, with no overflow check. Check that X & 7 == 0.
+  case AArch64::fixup_a64_tlsdesc_ld64_lo12_nc:
+    // R_AARCH64_TLSDESC_LD64_LO12_NC: Set an LD offset field to bits FF8 of
+    // G(TLSDESC(S+A)), with no overflow check. Check that X & 7 == 0.
+  case AArch64::fixup_a64_ld64_got_lo12_nc:
+    // R_AARCH64_LD64_GOT_LO12_NC: Sets the LD/ST immediate field to bits FF8 of
+    // G(S) with no overflow check. Check X & 7 == 0
+    assert(((int64_t)Value & 7) == 0 && "Misaligned fixup");
+    return (Value & 0xff8) << 7;
+
+  case AArch64::fixup_a64_tlsdesc_call:
+    // R_AARCH64_TLSDESC_CALL: For relaxation only.
+    return 0;
+  }
+}
+
+MCAsmBackend *
+llvm::createAArch64AsmBackend(const Target &T, StringRef TT, StringRef CPU) {
+  Triple TheTriple(TT);
+
+  return new ELFAArch64AsmBackend(T, TT, TheTriple.getOS());
+}
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64ELFObjectWriter.cpp b/lib/Target/AArch64/MCTargetDesc/AArch64ELFObjectWriter.cpp
new file mode 100644
index 000000000000..4bcc65dfca27
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64ELFObjectWriter.cpp
@@ -0,0 +1,292 @@
+//===-- AArch64ELFObjectWriter.cpp - AArch64 ELF Writer -------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file handles ELF-specific object emission, converting LLVM's internal
+// fixups into the appropriate relocations.
+//
+//===----------------------------------------------------------------------===//
+
+#include "MCTargetDesc/AArch64FixupKinds.h"
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "llvm/MC/MCELFObjectWriter.h"
+#include "llvm/MC/MCValue.h"
+#include "llvm/Support/ErrorHandling.h"
+
+using namespace llvm;
+
+namespace {
+class AArch64ELFObjectWriter : public MCELFObjectTargetWriter {
+public:
+  AArch64ELFObjectWriter(uint8_t OSABI);
+
+  virtual ~AArch64ELFObjectWriter();
+
+protected:
+  virtual unsigned GetRelocType(const MCValue &Target, const MCFixup &Fixup,
+                                bool IsPCRel, bool IsRelocWithSymbol,
+                                int64_t Addend) const;
+private:
+};
+}
+
+AArch64ELFObjectWriter::AArch64ELFObjectWriter(uint8_t OSABI)
+  : MCELFObjectTargetWriter(/*Is64Bit*/ true, OSABI, ELF::EM_AARCH64,
+                            /*HasRelocationAddend*/ true)
+{}
+
+AArch64ELFObjectWriter::~AArch64ELFObjectWriter()
+{}
+
+unsigned AArch64ELFObjectWriter::GetRelocType(const MCValue &Target,
+                                              const MCFixup &Fixup,
+                                              bool IsPCRel,
+                                              bool IsRelocWithSymbol,
+                                              int64_t Addend) const {
+  unsigned Type;
+  if (IsPCRel) {
+    switch ((unsigned)Fixup.getKind()) {
+    default:
+      llvm_unreachable("Unimplemented fixup -> relocation");
+    case FK_Data_8:
+      return ELF::R_AARCH64_PREL64;
+    case FK_Data_4:
+      return ELF::R_AARCH64_PREL32;
+    case FK_Data_2:
+      return ELF::R_AARCH64_PREL16;
+    case AArch64::fixup_a64_ld_prel:
+      Type = ELF::R_AARCH64_LD_PREL_LO19;
+      break;
+    case AArch64::fixup_a64_adr_prel:
+      Type = ELF::R_AARCH64_ADR_PREL_LO21;
+      break;
+    case AArch64::fixup_a64_adr_prel_page:
+      Type = ELF::R_AARCH64_ADR_PREL_PG_HI21;
+      break;
+    case AArch64::fixup_a64_adr_prel_got_page:
+      Type = ELF::R_AARCH64_ADR_GOT_PAGE;
+      break;
+    case AArch64::fixup_a64_tstbr:
+      Type = ELF::R_AARCH64_TSTBR14;
+      break;
+    case AArch64::fixup_a64_condbr:
+      Type = ELF::R_AARCH64_CONDBR19;
+      break;
+    case AArch64::fixup_a64_uncondbr:
+      Type = ELF::R_AARCH64_JUMP26;
+      break;
+    case AArch64::fixup_a64_call:
+      Type = ELF::R_AARCH64_CALL26;
+      break;
+    case AArch64::fixup_a64_adr_gottprel_page:
+      Type = ELF::R_AARCH64_TLSIE_ADR_GOTTPREL_PAGE21;
+      break;
+    case AArch64::fixup_a64_ld_gottprel_prel19:
+      Type =  ELF::R_AARCH64_TLSIE_LD_GOTTPREL_PREL19;
+      break;
+    case AArch64::fixup_a64_tlsdesc_adr_page:
+      Type = ELF::R_AARCH64_TLSDESC_ADR_PAGE;
+      break;
+    }
+  } else {
+    switch ((unsigned)Fixup.getKind()) {
+    default:
+      llvm_unreachable("Unimplemented fixup -> relocation");
+    case FK_Data_8:
+      return ELF::R_AARCH64_ABS64;
+    case FK_Data_4:
+      return ELF::R_AARCH64_ABS32;
+    case FK_Data_2:
+      return ELF::R_AARCH64_ABS16;
+    case AArch64::fixup_a64_add_lo12:
+      Type = ELF::R_AARCH64_ADD_ABS_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ld64_got_lo12_nc:
+      Type = ELF::R_AARCH64_LD64_GOT_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst8_lo12:
+      Type = ELF::R_AARCH64_LDST8_ABS_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst16_lo12:
+      Type = ELF::R_AARCH64_LDST16_ABS_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst32_lo12:
+      Type = ELF::R_AARCH64_LDST32_ABS_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst64_lo12:
+      Type = ELF::R_AARCH64_LDST64_ABS_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst128_lo12:
+      Type = ELF::R_AARCH64_LDST128_ABS_LO12_NC;
+      break;
+    case AArch64::fixup_a64_movw_uabs_g0:
+      Type = ELF::R_AARCH64_MOVW_UABS_G0;
+      break;
+    case AArch64::fixup_a64_movw_uabs_g0_nc:
+      Type = ELF::R_AARCH64_MOVW_UABS_G0_NC;
+      break;
+    case AArch64::fixup_a64_movw_uabs_g1:
+      Type = ELF::R_AARCH64_MOVW_UABS_G1;
+      break;
+    case AArch64::fixup_a64_movw_uabs_g1_nc:
+      Type = ELF::R_AARCH64_MOVW_UABS_G1_NC;
+      break;
+    case AArch64::fixup_a64_movw_uabs_g2:
+      Type = ELF::R_AARCH64_MOVW_UABS_G2;
+      break;
+    case AArch64::fixup_a64_movw_uabs_g2_nc:
+      Type = ELF::R_AARCH64_MOVW_UABS_G2_NC;
+      break;
+    case AArch64::fixup_a64_movw_uabs_g3:
+      Type = ELF::R_AARCH64_MOVW_UABS_G3;
+      break;
+    case AArch64::fixup_a64_movw_sabs_g0:
+      Type = ELF::R_AARCH64_MOVW_SABS_G0;
+      break;
+    case AArch64::fixup_a64_movw_sabs_g1:
+      Type = ELF::R_AARCH64_MOVW_SABS_G1;
+      break;
+    case AArch64::fixup_a64_movw_sabs_g2:
+      Type = ELF::R_AARCH64_MOVW_SABS_G2;
+      break;
+
+    // TLS Local-dynamic block
+    case AArch64::fixup_a64_movw_dtprel_g2:
+      Type = ELF::R_AARCH64_TLSLD_MOVW_DTPREL_G2;
+      break;
+    case AArch64::fixup_a64_movw_dtprel_g1:
+      Type = ELF::R_AARCH64_TLSLD_MOVW_DTPREL_G1;
+      break;
+    case AArch64::fixup_a64_movw_dtprel_g1_nc:
+      Type = ELF::R_AARCH64_TLSLD_MOVW_DTPREL_G1_NC;
+      break;
+    case AArch64::fixup_a64_movw_dtprel_g0:
+      Type = ELF::R_AARCH64_TLSLD_MOVW_DTPREL_G0;
+      break;
+    case AArch64::fixup_a64_movw_dtprel_g0_nc:
+      Type = ELF::R_AARCH64_TLSLD_MOVW_DTPREL_G0_NC;
+      break;
+    case AArch64::fixup_a64_add_dtprel_hi12:
+      Type = ELF::R_AARCH64_TLSLD_ADD_DTPREL_HI12;
+      break;
+    case AArch64::fixup_a64_add_dtprel_lo12:
+      Type = ELF::R_AARCH64_TLSLD_ADD_DTPREL_LO12;
+      break;
+    case AArch64::fixup_a64_add_dtprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSLD_ADD_DTPREL_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst8_dtprel_lo12:
+      Type = ELF::R_AARCH64_TLSLD_LDST8_DTPREL_LO12;
+      break;
+    case AArch64::fixup_a64_ldst8_dtprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSLD_LDST8_DTPREL_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst16_dtprel_lo12:
+      Type = ELF::R_AARCH64_TLSLD_LDST16_DTPREL_LO12;
+      break;
+    case AArch64::fixup_a64_ldst16_dtprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSLD_LDST16_DTPREL_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst32_dtprel_lo12:
+      Type = ELF::R_AARCH64_TLSLD_LDST32_DTPREL_LO12;
+      break;
+    case AArch64::fixup_a64_ldst32_dtprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSLD_LDST32_DTPREL_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst64_dtprel_lo12:
+      Type = ELF::R_AARCH64_TLSLD_LDST64_DTPREL_LO12;
+      break;
+    case AArch64::fixup_a64_ldst64_dtprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSLD_LDST64_DTPREL_LO12_NC;
+      break;
+
+    // TLS initial-exec block
+    case AArch64::fixup_a64_movw_gottprel_g1:
+      Type = ELF::R_AARCH64_TLSIE_MOVW_GOTTPREL_G1;
+      break;
+    case AArch64::fixup_a64_movw_gottprel_g0_nc:
+      Type = ELF::R_AARCH64_TLSIE_MOVW_GOTTPREL_G0_NC;
+      break;
+    case AArch64::fixup_a64_ld64_gottprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSIE_LD64_GOTTPREL_LO12_NC;
+      break;
+
+    // TLS local-exec block
+    case AArch64::fixup_a64_movw_tprel_g2:
+      Type = ELF::R_AARCH64_TLSLE_MOVW_TPREL_G2;
+      break;
+    case AArch64::fixup_a64_movw_tprel_g1:
+      Type = ELF::R_AARCH64_TLSLE_MOVW_TPREL_G1;
+      break;
+    case AArch64::fixup_a64_movw_tprel_g1_nc:
+      Type = ELF::R_AARCH64_TLSLE_MOVW_TPREL_G1_NC;
+      break;
+    case AArch64::fixup_a64_movw_tprel_g0:
+      Type = ELF::R_AARCH64_TLSLE_MOVW_TPREL_G0;
+      break;
+    case AArch64::fixup_a64_movw_tprel_g0_nc:
+      Type = ELF::R_AARCH64_TLSLE_MOVW_TPREL_G0_NC;
+      break;
+    case AArch64::fixup_a64_add_tprel_hi12:
+      Type = ELF::R_AARCH64_TLSLE_ADD_TPREL_HI12;
+      break;
+    case AArch64::fixup_a64_add_tprel_lo12:
+      Type = ELF::R_AARCH64_TLSLE_ADD_TPREL_LO12;
+      break;
+    case AArch64::fixup_a64_add_tprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSLE_ADD_TPREL_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst8_tprel_lo12:
+      Type = ELF::R_AARCH64_TLSLE_LDST8_TPREL_LO12;
+      break;
+    case AArch64::fixup_a64_ldst8_tprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSLE_LDST8_TPREL_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst16_tprel_lo12:
+      Type = ELF::R_AARCH64_TLSLE_LDST16_TPREL_LO12;
+      break;
+    case AArch64::fixup_a64_ldst16_tprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSLE_LDST16_TPREL_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst32_tprel_lo12:
+      Type = ELF::R_AARCH64_TLSLE_LDST32_TPREL_LO12;
+      break;
+    case AArch64::fixup_a64_ldst32_tprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSLE_LDST32_TPREL_LO12_NC;
+      break;
+    case AArch64::fixup_a64_ldst64_tprel_lo12:
+      Type = ELF::R_AARCH64_TLSLE_LDST64_TPREL_LO12;
+      break;
+    case AArch64::fixup_a64_ldst64_tprel_lo12_nc:
+      Type = ELF::R_AARCH64_TLSLE_LDST64_TPREL_LO12_NC;
+      break;
+
+    // TLS general-dynamic block
+    case AArch64::fixup_a64_tlsdesc_adr_page:
+      Type = ELF::R_AARCH64_TLSDESC_ADR_PAGE;
+      break;
+    case AArch64::fixup_a64_tlsdesc_ld64_lo12_nc:
+      Type = ELF::R_AARCH64_TLSDESC_LD64_LO12_NC;
+      break;
+    case AArch64::fixup_a64_tlsdesc_add_lo12_nc:
+      Type = ELF::R_AARCH64_TLSDESC_ADD_LO12_NC;
+      break;
+    case AArch64::fixup_a64_tlsdesc_call:
+      Type = ELF::R_AARCH64_TLSDESC_CALL;
+      break;
+    }
+  }
+
+  return Type;
+}
+
+MCObjectWriter *llvm::createAArch64ELFObjectWriter(raw_ostream &OS,
+                                                   uint8_t OSABI) {
+  MCELFObjectTargetWriter *MOTW = new AArch64ELFObjectWriter(OSABI);
+  return createELFObjectWriter(MOTW, OS,  /*IsLittleEndian=*/true);
+}
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64ELFStreamer.cpp b/lib/Target/AArch64/MCTargetDesc/AArch64ELFStreamer.cpp
new file mode 100644
index 000000000000..b83577af45c6
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64ELFStreamer.cpp
@@ -0,0 +1,160 @@
+//===- lib/MC/AArch64ELFStreamer.cpp - ELF Object Output for AArch64 ------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file assembles .s files and emits AArch64 ELF .o object files. Different
+// from generic ELF streamer in emitting mapping symbols ($x and $d) to delimit
+// regions of data and code.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/MC/MCELFStreamer.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/Twine.h"
+#include "llvm/MC/MCAsmBackend.h"
+#include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCELF.h"
+#include "llvm/MC/MCELFStreamer.h"
+#include "llvm/MC/MCELFSymbolFlags.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCObjectStreamer.h"
+#include "llvm/MC/MCSection.h"
+#include "llvm/MC/MCSectionELF.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/MC/MCSymbol.h"
+#include "llvm/MC/MCValue.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ELF.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+namespace {
+
+/// Extend the generic ELFStreamer class so that it can emit mapping symbols at
+/// the appropriate points in the object files. These symbols are defined in the
+/// AArch64 ELF ABI:
+///    infocenter.arm.com/help/topic/com.arm.doc.ihi0056a/IHI0056A_aaelf64.pdf
+///
+/// In brief: $x or $d should be emitted at the start of each contiguous region
+/// of A64 code or data in a section. In practice, this emission does not rely
+/// on explicit assembler directives but on inherent properties of the
+/// directives doing the emission (e.g. ".byte" is data, "add x0, x0, x0" an
+/// instruction).
+///
+/// As a result this system is orthogonal to the DataRegion infrastructure used
+/// by MachO. Beware!
+class AArch64ELFStreamer : public MCELFStreamer {
+public:
+  AArch64ELFStreamer(MCContext &Context, MCAsmBackend &TAB,
+                 raw_ostream &OS, MCCodeEmitter *Emitter)
+    : MCELFStreamer(Context, TAB, OS, Emitter),
+      MappingSymbolCounter(0), LastEMS(EMS_None) {
+  }
+
+  ~AArch64ELFStreamer() {}
+
+  virtual void ChangeSection(const MCSection *Section) {
+    // We have to keep track of the mapping symbol state of any sections we
+    // use. Each one should start off as EMS_None, which is provided as the
+    // default constructor by DenseMap::lookup.
+    LastMappingSymbols[getPreviousSection()] = LastEMS;
+    LastEMS = LastMappingSymbols.lookup(Section);
+
+    MCELFStreamer::ChangeSection(Section);
+  }
+
+  /// This function is the one used to emit instruction data into the ELF
+  /// streamer. We override it to add the appropriate mapping symbol if
+  /// necessary.
+  virtual void EmitInstruction(const MCInst& Inst) {
+    EmitA64MappingSymbol();
+    MCELFStreamer::EmitInstruction(Inst);
+  }
+
+  /// This is one of the functions used to emit data into an ELF section, so the
+  /// AArch64 streamer overrides it to add the appropriate mapping symbol ($d)
+  /// if necessary.
+  virtual void EmitBytes(StringRef Data, unsigned AddrSpace) {
+    EmitDataMappingSymbol();
+    MCELFStreamer::EmitBytes(Data, AddrSpace);
+  }
+
+  /// This is one of the functions used to emit data into an ELF section, so the
+  /// AArch64 streamer overrides it to add the appropriate mapping symbol ($d)
+  /// if necessary.
+  virtual void EmitValueImpl(const MCExpr *Value, unsigned Size,
+                             unsigned AddrSpace) {
+    EmitDataMappingSymbol();
+    MCELFStreamer::EmitValueImpl(Value, Size, AddrSpace);
+  }
+
+private:
+  enum ElfMappingSymbol {
+    EMS_None,
+    EMS_A64,
+    EMS_Data
+  };
+
+  void EmitDataMappingSymbol() {
+    if (LastEMS == EMS_Data) return;
+    EmitMappingSymbol("$d");
+    LastEMS = EMS_Data;
+  }
+
+  void EmitA64MappingSymbol() {
+    if (LastEMS == EMS_A64) return;
+    EmitMappingSymbol("$x");
+    LastEMS = EMS_A64;
+  }
+
+  void EmitMappingSymbol(StringRef Name) {
+    MCSymbol *Start = getContext().CreateTempSymbol();
+    EmitLabel(Start);
+
+    MCSymbol *Symbol =
+      getContext().GetOrCreateSymbol(Name + "." +
+                                     Twine(MappingSymbolCounter++));
+
+    MCSymbolData &SD = getAssembler().getOrCreateSymbolData(*Symbol);
+    MCELF::SetType(SD, ELF::STT_NOTYPE);
+    MCELF::SetBinding(SD, ELF::STB_LOCAL);
+    SD.setExternal(false);
+    Symbol->setSection(*getCurrentSection());
+
+    const MCExpr *Value = MCSymbolRefExpr::Create(Start, getContext());
+    Symbol->setVariableValue(Value);
+  }
+
+  int64_t MappingSymbolCounter;
+
+  DenseMap<const MCSection *, ElfMappingSymbol> LastMappingSymbols;
+  ElfMappingSymbol LastEMS;
+
+  /// @}
+};
+}
+
+namespace llvm {
+  MCELFStreamer* createAArch64ELFStreamer(MCContext &Context, MCAsmBackend &TAB,
+                                      raw_ostream &OS, MCCodeEmitter *Emitter,
+                                      bool RelaxAll, bool NoExecStack) {
+    AArch64ELFStreamer *S = new AArch64ELFStreamer(Context, TAB, OS, Emitter);
+    if (RelaxAll)
+      S->getAssembler().setRelaxAll(true);
+    if (NoExecStack)
+      S->getAssembler().setNoExecStack(true);
+    return S;
+  }
+}
+
+
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64ELFStreamer.h b/lib/Target/AArch64/MCTargetDesc/AArch64ELFStreamer.h
new file mode 100644
index 000000000000..5a89ca50cee8
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64ELFStreamer.h
@@ -0,0 +1,27 @@
+//===-- AArch64ELFStreamer.h - ELF Streamer for AArch64 ---------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements ELF streamer information for the AArch64 backend.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_AARCH64_ELF_STREAMER_H
+#define LLVM_AARCH64_ELF_STREAMER_H
+
+#include "llvm/MC/MCELFStreamer.h"
+
+namespace llvm {
+
+  MCELFStreamer* createAArch64ELFStreamer(MCContext &Context, MCAsmBackend &TAB,
+                                          raw_ostream &OS,
+                                          MCCodeEmitter *Emitter,
+                                          bool RelaxAll, bool NoExecStack);
+}
+
+#endif // AArch64_ELF_STREAMER_H
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64FixupKinds.h b/lib/Target/AArch64/MCTargetDesc/AArch64FixupKinds.h
new file mode 100644
index 000000000000..eeb122d38494
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64FixupKinds.h
@@ -0,0 +1,113 @@
+//=- AArch64/AArch64FixupKinds.h - AArch64 Specific Fixup Entries -*- C++ -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file describes the LLVM fixups applied to MCInsts in the AArch64
+// backend.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_AARCH64_AARCH64FIXUPKINDS_H
+#define LLVM_AARCH64_AARCH64FIXUPKINDS_H
+
+#include "llvm/MC/MCFixup.h"
+
+namespace llvm {
+  namespace AArch64 {
+    enum Fixups {
+      fixup_a64_ld_prel = FirstTargetFixupKind,
+      fixup_a64_adr_prel,
+      fixup_a64_adr_prel_page,
+
+      fixup_a64_add_lo12,
+
+      fixup_a64_ldst8_lo12,
+      fixup_a64_ldst16_lo12,
+      fixup_a64_ldst32_lo12,
+      fixup_a64_ldst64_lo12,
+      fixup_a64_ldst128_lo12,
+
+      fixup_a64_tstbr,
+      fixup_a64_condbr,
+      fixup_a64_uncondbr,
+      fixup_a64_call,
+
+      fixup_a64_movw_uabs_g0,
+      fixup_a64_movw_uabs_g0_nc,
+      fixup_a64_movw_uabs_g1,
+      fixup_a64_movw_uabs_g1_nc,
+      fixup_a64_movw_uabs_g2,
+      fixup_a64_movw_uabs_g2_nc,
+      fixup_a64_movw_uabs_g3,
+
+      fixup_a64_movw_sabs_g0,
+      fixup_a64_movw_sabs_g1,
+      fixup_a64_movw_sabs_g2,
+
+      fixup_a64_adr_prel_got_page,
+      fixup_a64_ld64_got_lo12_nc,
+
+      // Produce offsets relative to the module's dynamic TLS area.
+      fixup_a64_movw_dtprel_g2,
+      fixup_a64_movw_dtprel_g1,
+      fixup_a64_movw_dtprel_g1_nc,
+      fixup_a64_movw_dtprel_g0,
+      fixup_a64_movw_dtprel_g0_nc,
+      fixup_a64_add_dtprel_hi12,
+      fixup_a64_add_dtprel_lo12,
+      fixup_a64_add_dtprel_lo12_nc,
+      fixup_a64_ldst8_dtprel_lo12,
+      fixup_a64_ldst8_dtprel_lo12_nc,
+      fixup_a64_ldst16_dtprel_lo12,
+      fixup_a64_ldst16_dtprel_lo12_nc,
+      fixup_a64_ldst32_dtprel_lo12,
+      fixup_a64_ldst32_dtprel_lo12_nc,
+      fixup_a64_ldst64_dtprel_lo12,
+      fixup_a64_ldst64_dtprel_lo12_nc,
+
+      // Produce the GOT entry containing a variable's address in TLS's
+      // initial-exec mode.
+      fixup_a64_movw_gottprel_g1,
+      fixup_a64_movw_gottprel_g0_nc,
+      fixup_a64_adr_gottprel_page,
+      fixup_a64_ld64_gottprel_lo12_nc,
+      fixup_a64_ld_gottprel_prel19,
+
+      // Produce offsets relative to the thread pointer: TPIDR_EL0.
+      fixup_a64_movw_tprel_g2,
+      fixup_a64_movw_tprel_g1,
+      fixup_a64_movw_tprel_g1_nc,
+      fixup_a64_movw_tprel_g0,
+      fixup_a64_movw_tprel_g0_nc,
+      fixup_a64_add_tprel_hi12,
+      fixup_a64_add_tprel_lo12,
+      fixup_a64_add_tprel_lo12_nc,
+      fixup_a64_ldst8_tprel_lo12,
+      fixup_a64_ldst8_tprel_lo12_nc,
+      fixup_a64_ldst16_tprel_lo12,
+      fixup_a64_ldst16_tprel_lo12_nc,
+      fixup_a64_ldst32_tprel_lo12,
+      fixup_a64_ldst32_tprel_lo12_nc,
+      fixup_a64_ldst64_tprel_lo12,
+      fixup_a64_ldst64_tprel_lo12_nc,
+
+      // Produce the special fixups used by the general-dynamic TLS model.
+      fixup_a64_tlsdesc_adr_page,
+      fixup_a64_tlsdesc_ld64_lo12_nc,
+      fixup_a64_tlsdesc_add_lo12_nc,
+      fixup_a64_tlsdesc_call,
+
+
+      // Marker
+      LastTargetFixupKind,
+      NumTargetFixupKinds = LastTargetFixupKind - FirstTargetFixupKind
+    };
+  }
+}
+
+#endif
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64MCAsmInfo.cpp b/lib/Target/AArch64/MCTargetDesc/AArch64MCAsmInfo.cpp
new file mode 100644
index 000000000000..8ec8cbf1c525
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64MCAsmInfo.cpp
@@ -0,0 +1,41 @@
+//===-- AArch64MCAsmInfo.cpp - AArch64 asm properties ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the declarations of the AArch64MCAsmInfo properties.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AArch64MCAsmInfo.h"
+
+using namespace llvm;
+
+AArch64ELFMCAsmInfo::AArch64ELFMCAsmInfo() {
+  PointerSize = 8;
+
+  // ".comm align is in bytes but .align is pow-2."
+  AlignmentIsInBytes = false;
+
+  CommentString = "//";
+  PrivateGlobalPrefix = ".L";
+  Code32Directive = ".code\t32";
+
+  Data16bitsDirective = "\t.hword\t";
+  Data32bitsDirective = "\t.word\t";
+  Data64bitsDirective = "\t.xword\t";
+
+  UseDataRegionDirectives = true;
+
+  WeakRefDirective = "\t.weak\t";
+
+  HasLEB128 = true;
+  SupportsDebugInformation = true;
+
+  // Exceptions handling
+  ExceptionsType = ExceptionHandling::DwarfCFI;
+}
diff --git a/lib/Target/CellSPU/MCTargetDesc/SPUMCAsmInfo.h b/lib/Target/AArch64/MCTargetDesc/AArch64MCAsmInfo.h
index f786147b9267..a20bc471c20d 100644
--- a/lib/Target/CellSPU/MCTargetDesc/SPUMCAsmInfo.h
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64MCAsmInfo.h
@@ -1,4 +1,4 @@
-//===-- SPUMCAsmInfo.h - Cell SPU asm properties ---------------*- C++ -*--===//
+//==-- AArch64MCAsmInfo.h - AArch64 asm properties -------------*- C++ -*--===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,24 +7,21 @@
 //
 //===----------------------------------------------------------------------===//
 //
-// This file contains the declaration of the SPUMCAsmInfo class.
+// This file contains the declaration of the AArch64MCAsmInfo class.
 //
 //===----------------------------------------------------------------------===//
 
-#ifndef SPUTARGETASMINFO_H
-#define SPUTARGETASMINFO_H
+#ifndef LLVM_AARCH64TARGETASMINFO_H
+#define LLVM_AARCH64TARGETASMINFO_H
 
-#include "llvm/ADT/StringRef.h"
 #include "llvm/MC/MCAsmInfo.h"
 
 namespace llvm {
-  class Target;
-  
-  class SPULinuxMCAsmInfo : public MCAsmInfo {
-    virtual void anchor();
-  public:
-    explicit SPULinuxMCAsmInfo(const Target &T, StringRef TT);
+
+  struct AArch64ELFMCAsmInfo : public MCAsmInfo {
+    explicit AArch64ELFMCAsmInfo();
   };
+
 } // namespace llvm
 
-#endif /* SPUTARGETASMINFO_H */
+#endif
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64MCCodeEmitter.cpp b/lib/Target/AArch64/MCTargetDesc/AArch64MCCodeEmitter.cpp
new file mode 100644
index 000000000000..a5c591eee800
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64MCCodeEmitter.cpp
@@ -0,0 +1,502 @@
+//=- AArch64/AArch64MCCodeEmitter.cpp - Convert AArch64 code to machine code =//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the AArch64MCCodeEmitter class.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "mccodeemitter"
+#include "MCTargetDesc/AArch64FixupKinds.h"
+#include "MCTargetDesc/AArch64MCExpr.h"
+#include "MCTargetDesc/AArch64MCTargetDesc.h"
+#include "Utils/AArch64BaseInfo.h"
+#include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+namespace {
+class AArch64MCCodeEmitter : public MCCodeEmitter {
+  AArch64MCCodeEmitter(const AArch64MCCodeEmitter &) LLVM_DELETED_FUNCTION;
+  void operator=(const AArch64MCCodeEmitter &) LLVM_DELETED_FUNCTION;
+  MCContext &Ctx;
+
+public:
+  AArch64MCCodeEmitter(MCContext &ctx) : Ctx(ctx) {}
+
+  ~AArch64MCCodeEmitter() {}
+
+  unsigned getAddSubImmOpValue(const MCInst &MI, unsigned OpIdx,
+                               SmallVectorImpl<MCFixup> &Fixups) const;
+
+  unsigned getAdrpLabelOpValue(const MCInst &MI, unsigned OpIdx,
+                               SmallVectorImpl<MCFixup> &Fixups) const;
+
+  template<int MemSize>
+  unsigned getOffsetUImm12OpValue(const MCInst &MI, unsigned OpIdx,
+                                    SmallVectorImpl<MCFixup> &Fixups) const {
+    return getOffsetUImm12OpValue(MI, OpIdx, Fixups, MemSize);
+  }
+
+  unsigned getOffsetUImm12OpValue(const MCInst &MI, unsigned OpIdx,
+                                    SmallVectorImpl<MCFixup> &Fixups,
+                                    int MemSize) const;
+
+  unsigned getBitfield32LSLOpValue(const MCInst &MI, unsigned OpIdx,
+                                   SmallVectorImpl<MCFixup> &Fixups) const;
+  unsigned getBitfield64LSLOpValue(const MCInst &MI, unsigned OpIdx,
+                                   SmallVectorImpl<MCFixup> &Fixups) const;
+
+
+  // Labels are handled mostly the same way: a symbol is needed, and
+  // just gets some fixup attached.
+  template<AArch64::Fixups fixupDesired>
+  unsigned getLabelOpValue(const MCInst &MI, unsigned OpIdx,
+                           SmallVectorImpl<MCFixup> &Fixups) const;
+
+  unsigned  getLoadLitLabelOpValue(const MCInst &MI, unsigned OpIdx,
+                                   SmallVectorImpl<MCFixup> &Fixups) const;
+
+
+  unsigned getMoveWideImmOpValue(const MCInst &MI, unsigned OpIdx,
+                                 SmallVectorImpl<MCFixup> &Fixups) const;
+
+
+  unsigned getAddressWithFixup(const MCOperand &MO,
+                               unsigned FixupKind,
+                               SmallVectorImpl<MCFixup> &Fixups) const;
+
+
+  // getBinaryCodeForInstr - TableGen'erated function for getting the
+  // binary encoding for an instruction.
+  uint64_t getBinaryCodeForInstr(const MCInst &MI,
+                                 SmallVectorImpl<MCFixup> &Fixups) const;
+
+  /// getMachineOpValue - Return binary encoding of operand. If the machine
+  /// operand requires relocation, record the relocation and return zero.
+  unsigned getMachineOpValue(const MCInst &MI,const MCOperand &MO,
+                             SmallVectorImpl<MCFixup> &Fixups) const;
+
+
+  void EmitByte(unsigned char C, raw_ostream &OS) const {
+    OS << (char)C;
+  }
+
+  void EmitInstruction(uint32_t Val, raw_ostream &OS) const {
+    // Output the constant in little endian byte order.
+    for (unsigned i = 0; i != 4; ++i) {
+      EmitByte(Val & 0xff, OS);
+      Val >>= 8;
+    }
+  }
+
+
+  void EncodeInstruction(const MCInst &MI, raw_ostream &OS,
+                         SmallVectorImpl<MCFixup> &Fixups) const;
+
+  template<int hasRs, int hasRt2> unsigned
+  fixLoadStoreExclusive(const MCInst &MI, unsigned EncodedValue) const;
+
+  unsigned fixMOVZ(const MCInst &MI, unsigned EncodedValue) const;
+
+  unsigned fixMulHigh(const MCInst &MI, unsigned EncodedValue) const;
+
+
+};
+
+} // end anonymous namespace
+
+unsigned AArch64MCCodeEmitter::getAddressWithFixup(const MCOperand &MO,
+                                       unsigned FixupKind,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+  if (!MO.isExpr()) {
+    // This can occur for manually decoded or constructed MCInsts, but neither
+    // the assembly-parser nor instruction selection will currently produce an
+    // MCInst that's not a symbol reference.
+    assert(MO.isImm() && "Unexpected address requested");
+    return MO.getImm();
+  }
+
+  const MCExpr *Expr = MO.getExpr();
+  MCFixupKind Kind = MCFixupKind(FixupKind);
+  Fixups.push_back(MCFixup::Create(0, Expr, Kind));
+
+  return 0;
+}
+
+unsigned AArch64MCCodeEmitter::
+getOffsetUImm12OpValue(const MCInst &MI, unsigned OpIdx,
+                       SmallVectorImpl<MCFixup> &Fixups,
+                       int MemSize) const {
+  const MCOperand &ImmOp = MI.getOperand(OpIdx);
+  if (ImmOp.isImm())
+    return ImmOp.getImm();
+
+  assert(ImmOp.isExpr() && "Unexpected operand type");
+  const AArch64MCExpr *Expr = cast<AArch64MCExpr>(ImmOp.getExpr());
+  unsigned FixupKind;
+
+
+  switch (Expr->getKind()) {
+  default: llvm_unreachable("Unexpected operand modifier");
+  case AArch64MCExpr::VK_AARCH64_LO12: {
+    unsigned FixupsBySize[] = { AArch64::fixup_a64_ldst8_lo12,
+                                AArch64::fixup_a64_ldst16_lo12,
+                                AArch64::fixup_a64_ldst32_lo12,
+                                AArch64::fixup_a64_ldst64_lo12,
+                                AArch64::fixup_a64_ldst128_lo12 };
+    assert(MemSize <= 16 && "Invalid fixup for operation");
+    FixupKind = FixupsBySize[Log2_32(MemSize)];
+    break;
+  }
+  case AArch64MCExpr::VK_AARCH64_GOT_LO12:
+    assert(MemSize == 8 && "Invalid fixup for operation");
+    FixupKind = AArch64::fixup_a64_ld64_got_lo12_nc;
+    break;
+  case AArch64MCExpr::VK_AARCH64_DTPREL_LO12:  {
+    unsigned FixupsBySize[] = { AArch64::fixup_a64_ldst8_dtprel_lo12,
+                                AArch64::fixup_a64_ldst16_dtprel_lo12,
+                                AArch64::fixup_a64_ldst32_dtprel_lo12,
+                                AArch64::fixup_a64_ldst64_dtprel_lo12 };
+    assert(MemSize <= 8 && "Invalid fixup for operation");
+    FixupKind = FixupsBySize[Log2_32(MemSize)];
+    break;
+  }
+  case AArch64MCExpr::VK_AARCH64_DTPREL_LO12_NC: {
+    unsigned FixupsBySize[] = { AArch64::fixup_a64_ldst8_dtprel_lo12_nc,
+                                AArch64::fixup_a64_ldst16_dtprel_lo12_nc,
+                                AArch64::fixup_a64_ldst32_dtprel_lo12_nc,
+                                AArch64::fixup_a64_ldst64_dtprel_lo12_nc };
+    assert(MemSize <= 8 && "Invalid fixup for operation");
+    FixupKind = FixupsBySize[Log2_32(MemSize)];
+    break;
+  }
+  case AArch64MCExpr::VK_AARCH64_GOTTPREL_LO12:
+    assert(MemSize == 8 && "Invalid fixup for operation");
+    FixupKind = AArch64::fixup_a64_ld64_gottprel_lo12_nc;
+    break;
+  case AArch64MCExpr::VK_AARCH64_TPREL_LO12:{
+    unsigned FixupsBySize[] = { AArch64::fixup_a64_ldst8_tprel_lo12,
+                                AArch64::fixup_a64_ldst16_tprel_lo12,
+                                AArch64::fixup_a64_ldst32_tprel_lo12,
+                                AArch64::fixup_a64_ldst64_tprel_lo12 };
+    assert(MemSize <= 8 && "Invalid fixup for operation");
+    FixupKind = FixupsBySize[Log2_32(MemSize)];
+    break;
+  }
+  case AArch64MCExpr::VK_AARCH64_TPREL_LO12_NC: {
+    unsigned FixupsBySize[] = { AArch64::fixup_a64_ldst8_tprel_lo12_nc,
+                                AArch64::fixup_a64_ldst16_tprel_lo12_nc,
+                                AArch64::fixup_a64_ldst32_tprel_lo12_nc,
+                                AArch64::fixup_a64_ldst64_tprel_lo12_nc };
+    assert(MemSize <= 8 && "Invalid fixup for operation");
+    FixupKind = FixupsBySize[Log2_32(MemSize)];
+    break;
+  }
+  case AArch64MCExpr::VK_AARCH64_TLSDESC_LO12:
+    assert(MemSize == 8 && "Invalid fixup for operation");
+    FixupKind = AArch64::fixup_a64_tlsdesc_ld64_lo12_nc;
+    break;
+  }
+
+  return getAddressWithFixup(ImmOp, FixupKind, Fixups);
+}
+
+unsigned
+AArch64MCCodeEmitter::getAddSubImmOpValue(const MCInst &MI, unsigned OpIdx,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+  const MCOperand &MO = MI.getOperand(OpIdx);
+  if (MO.isImm())
+    return static_cast<unsigned>(MO.getImm());
+
+  assert(MO.isExpr());
+
+  unsigned FixupKind = 0;
+  switch(cast<AArch64MCExpr>(MO.getExpr())->getKind()) {
+  default: llvm_unreachable("Invalid expression modifier");
+  case AArch64MCExpr::VK_AARCH64_LO12:
+    FixupKind = AArch64::fixup_a64_add_lo12; break;
+  case AArch64MCExpr::VK_AARCH64_DTPREL_HI12:
+    FixupKind = AArch64::fixup_a64_add_dtprel_hi12; break;
+  case AArch64MCExpr::VK_AARCH64_DTPREL_LO12:
+    FixupKind = AArch64::fixup_a64_add_dtprel_lo12; break;
+  case AArch64MCExpr::VK_AARCH64_DTPREL_LO12_NC:
+    FixupKind = AArch64::fixup_a64_add_dtprel_lo12_nc; break;
+  case AArch64MCExpr::VK_AARCH64_TPREL_HI12:
+    FixupKind = AArch64::fixup_a64_add_tprel_hi12; break;
+  case AArch64MCExpr::VK_AARCH64_TPREL_LO12:
+    FixupKind = AArch64::fixup_a64_add_tprel_lo12; break;
+  case AArch64MCExpr::VK_AARCH64_TPREL_LO12_NC:
+    FixupKind = AArch64::fixup_a64_add_tprel_lo12_nc; break;
+  case AArch64MCExpr::VK_AARCH64_TLSDESC_LO12:
+    FixupKind = AArch64::fixup_a64_tlsdesc_add_lo12_nc; break;
+  }
+
+  return getAddressWithFixup(MO, FixupKind, Fixups);
+}
+
+unsigned
+AArch64MCCodeEmitter::getAdrpLabelOpValue(const MCInst &MI, unsigned OpIdx,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+
+  const MCOperand &MO = MI.getOperand(OpIdx);
+  if (MO.isImm())
+    return static_cast<unsigned>(MO.getImm());
+
+  assert(MO.isExpr());
+
+  unsigned Modifier = AArch64MCExpr::VK_AARCH64_None;
+  if (const AArch64MCExpr *Expr = dyn_cast<AArch64MCExpr>(MO.getExpr()))
+    Modifier = Expr->getKind();
+
+  unsigned FixupKind = 0;
+  switch(Modifier) {
+  case AArch64MCExpr::VK_AARCH64_None:
+    FixupKind = AArch64::fixup_a64_adr_prel_page;
+    break;
+  case AArch64MCExpr::VK_AARCH64_GOT:
+    FixupKind = AArch64::fixup_a64_adr_prel_got_page;
+    break;
+  case AArch64MCExpr::VK_AARCH64_GOTTPREL:
+    FixupKind = AArch64::fixup_a64_adr_gottprel_page;
+    break;
+  case AArch64MCExpr::VK_AARCH64_TLSDESC:
+    FixupKind = AArch64::fixup_a64_tlsdesc_adr_page;
+    break;
+  default:
+    llvm_unreachable("Unknown symbol reference kind for ADRP instruction");
+  }
+
+  return getAddressWithFixup(MO, FixupKind, Fixups);
+}
+
+unsigned
+AArch64MCCodeEmitter::getBitfield32LSLOpValue(const MCInst &MI, unsigned OpIdx,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+
+  const MCOperand &MO = MI.getOperand(OpIdx);
+  assert(MO.isImm() && "Only immediate expected for shift");
+
+  return ((32 - MO.getImm()) & 0x1f) | (31 - MO.getImm()) << 6;
+}
+
+unsigned
+AArch64MCCodeEmitter::getBitfield64LSLOpValue(const MCInst &MI, unsigned OpIdx,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+
+  const MCOperand &MO = MI.getOperand(OpIdx);
+  assert(MO.isImm() && "Only immediate expected for shift");
+
+  return ((64 - MO.getImm()) & 0x3f) | (63 - MO.getImm()) << 6;
+}
+
+
+template<AArch64::Fixups fixupDesired> unsigned
+AArch64MCCodeEmitter::getLabelOpValue(const MCInst &MI,
+                                      unsigned OpIdx,
+                                      SmallVectorImpl<MCFixup> &Fixups) const {
+  const MCOperand &MO = MI.getOperand(OpIdx);
+
+  if (MO.isExpr())
+    return getAddressWithFixup(MO, fixupDesired, Fixups);
+
+  assert(MO.isImm());
+  return MO.getImm();
+}
+
+unsigned
+AArch64MCCodeEmitter::getLoadLitLabelOpValue(const MCInst &MI,
+                                       unsigned OpIdx,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+  const MCOperand &MO = MI.getOperand(OpIdx);
+
+  if (MO.isImm())
+    return MO.getImm();
+
+  assert(MO.isExpr());
+
+  unsigned FixupKind;
+  if (isa<AArch64MCExpr>(MO.getExpr())) {
+    assert(dyn_cast<AArch64MCExpr>(MO.getExpr())->getKind()
+           == AArch64MCExpr::VK_AARCH64_GOTTPREL
+           && "Invalid symbol modifier for literal load");
+    FixupKind = AArch64::fixup_a64_ld_gottprel_prel19;
+  } else {
+    FixupKind = AArch64::fixup_a64_ld_prel;
+  }
+
+  return getAddressWithFixup(MO, FixupKind, Fixups);
+}
+
+
+unsigned
+AArch64MCCodeEmitter::getMachineOpValue(const MCInst &MI,
+                                       const MCOperand &MO,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+  if (MO.isReg()) {
+    return Ctx.getRegisterInfo().getEncodingValue(MO.getReg());
+  } else if (MO.isImm()) {
+    return static_cast<unsigned>(MO.getImm());
+  }
+
+  llvm_unreachable("Unable to encode MCOperand!");
+  return 0;
+}
+
+unsigned
+AArch64MCCodeEmitter::getMoveWideImmOpValue(const MCInst &MI, unsigned OpIdx,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+  const MCOperand &UImm16MO = MI.getOperand(OpIdx);
+  const MCOperand &ShiftMO = MI.getOperand(OpIdx + 1);
+
+  unsigned Result = static_cast<unsigned>(ShiftMO.getImm()) << 16;
+
+  if (UImm16MO.isImm()) {
+    Result |= UImm16MO.getImm();
+    return Result;
+  }
+
+  const AArch64MCExpr *A64E = cast<AArch64MCExpr>(UImm16MO.getExpr());
+  AArch64::Fixups requestedFixup;
+  switch (A64E->getKind()) {
+  default: llvm_unreachable("unexpected expression modifier");
+  case AArch64MCExpr::VK_AARCH64_ABS_G0:
+    requestedFixup = AArch64::fixup_a64_movw_uabs_g0; break;
+  case AArch64MCExpr::VK_AARCH64_ABS_G0_NC:
+    requestedFixup = AArch64::fixup_a64_movw_uabs_g0_nc; break;
+  case AArch64MCExpr::VK_AARCH64_ABS_G1:
+    requestedFixup = AArch64::fixup_a64_movw_uabs_g1; break;
+  case AArch64MCExpr::VK_AARCH64_ABS_G1_NC:
+    requestedFixup = AArch64::fixup_a64_movw_uabs_g1_nc; break;
+  case AArch64MCExpr::VK_AARCH64_ABS_G2:
+    requestedFixup = AArch64::fixup_a64_movw_uabs_g2; break;
+  case AArch64MCExpr::VK_AARCH64_ABS_G2_NC:
+    requestedFixup = AArch64::fixup_a64_movw_uabs_g2_nc; break;
+  case AArch64MCExpr::VK_AARCH64_ABS_G3:
+    requestedFixup = AArch64::fixup_a64_movw_uabs_g3; break;
+  case AArch64MCExpr::VK_AARCH64_SABS_G0:
+    requestedFixup = AArch64::fixup_a64_movw_sabs_g0; break;
+  case AArch64MCExpr::VK_AARCH64_SABS_G1:
+    requestedFixup = AArch64::fixup_a64_movw_sabs_g1; break;
+  case AArch64MCExpr::VK_AARCH64_SABS_G2:
+    requestedFixup = AArch64::fixup_a64_movw_sabs_g2; break;
+  case AArch64MCExpr::VK_AARCH64_DTPREL_G2:
+    requestedFixup = AArch64::fixup_a64_movw_dtprel_g2; break;
+  case AArch64MCExpr::VK_AARCH64_DTPREL_G1:
+    requestedFixup = AArch64::fixup_a64_movw_dtprel_g1; break;
+  case AArch64MCExpr::VK_AARCH64_DTPREL_G1_NC:
+    requestedFixup = AArch64::fixup_a64_movw_dtprel_g1_nc; break;
+  case AArch64MCExpr::VK_AARCH64_DTPREL_G0:
+    requestedFixup = AArch64::fixup_a64_movw_dtprel_g0; break;
+  case AArch64MCExpr::VK_AARCH64_DTPREL_G0_NC:
+    requestedFixup = AArch64::fixup_a64_movw_dtprel_g0_nc; break;
+  case AArch64MCExpr::VK_AARCH64_GOTTPREL_G1:
+    requestedFixup = AArch64::fixup_a64_movw_gottprel_g1; break;
+  case AArch64MCExpr::VK_AARCH64_GOTTPREL_G0_NC:
+    requestedFixup = AArch64::fixup_a64_movw_gottprel_g0_nc; break;
+  case AArch64MCExpr::VK_AARCH64_TPREL_G2:
+    requestedFixup = AArch64::fixup_a64_movw_tprel_g2; break;
+  case AArch64MCExpr::VK_AARCH64_TPREL_G1:
+    requestedFixup = AArch64::fixup_a64_movw_tprel_g1; break;
+  case AArch64MCExpr::VK_AARCH64_TPREL_G1_NC:
+    requestedFixup = AArch64::fixup_a64_movw_tprel_g1_nc; break;
+  case AArch64MCExpr::VK_AARCH64_TPREL_G0:
+    requestedFixup = AArch64::fixup_a64_movw_tprel_g0; break;
+  case AArch64MCExpr::VK_AARCH64_TPREL_G0_NC:
+    requestedFixup = AArch64::fixup_a64_movw_tprel_g0_nc; break;
+  }
+
+  return Result | getAddressWithFixup(UImm16MO, requestedFixup, Fixups);
+}
+
+template<int hasRs, int hasRt2> unsigned
+AArch64MCCodeEmitter::fixLoadStoreExclusive(const MCInst &MI,
+                                            unsigned EncodedValue) const {
+  if (!hasRs) EncodedValue |= 0x001F0000;
+  if (!hasRt2) EncodedValue |= 0x00007C00;
+
+  return EncodedValue;
+}
+
+unsigned
+AArch64MCCodeEmitter::fixMOVZ(const MCInst &MI, unsigned EncodedValue) const {
+  // If one of the signed fixup kinds is applied to a MOVZ instruction, the
+  // eventual result could be either a MOVZ or a MOVN. It's the MCCodeEmitter's
+  // job to ensure that any bits possibly affected by this are 0. This means we
+  // must zero out bit 30 (essentially emitting a MOVN).
+  MCOperand UImm16MO = MI.getOperand(1);
+
+  // Nothing to do if there's no fixup.
+  if (UImm16MO.isImm())
+    return EncodedValue;
+
+  const AArch64MCExpr *A64E = cast<AArch64MCExpr>(UImm16MO.getExpr());
+  switch (A64E->getKind()) {
+  case AArch64MCExpr::VK_AARCH64_SABS_G0:
+  case AArch64MCExpr::VK_AARCH64_SABS_G1:
+  case AArch64MCExpr::VK_AARCH64_SABS_G2:
+  case AArch64MCExpr::VK_AARCH64_DTPREL_G2:
+  case AArch64MCExpr::VK_AARCH64_DTPREL_G1:
+  case AArch64MCExpr::VK_AARCH64_DTPREL_G0:
+  case AArch64MCExpr::VK_AARCH64_GOTTPREL_G1:
+  case AArch64MCExpr::VK_AARCH64_TPREL_G2:
+  case AArch64MCExpr::VK_AARCH64_TPREL_G1:
+  case AArch64MCExpr::VK_AARCH64_TPREL_G0:
+    return EncodedValue & ~(1u << 30);
+  default:
+    // Nothing to do for an unsigned fixup.
+    return EncodedValue;
+  }
+
+  llvm_unreachable("Should have returned by now");
+}
+
+unsigned
+AArch64MCCodeEmitter::fixMulHigh(const MCInst &MI,
+                                 unsigned EncodedValue) const {
+  // The Ra field of SMULH and UMULH is unused: it should be assembled as 31
+  // (i.e. all bits 1) but is ignored by the processor.
+  EncodedValue |= 0x1f << 10;
+  return EncodedValue;
+}
+
+MCCodeEmitter *llvm::createAArch64MCCodeEmitter(const MCInstrInfo &MCII,
+                                                const MCRegisterInfo &MRI,
+                                                const MCSubtargetInfo &STI,
+                                                MCContext &Ctx) {
+  return new AArch64MCCodeEmitter(Ctx);
+}
+
+void AArch64MCCodeEmitter::
+EncodeInstruction(const MCInst &MI, raw_ostream &OS,
+                  SmallVectorImpl<MCFixup> &Fixups) const {
+  if (MI.getOpcode() == AArch64::TLSDESCCALL) {
+    // This is a directive which applies an R_AARCH64_TLSDESC_CALL to the
+    // following (BLR) instruction. It doesn't emit any code itself so it
+    // doesn't go through the normal TableGenerated channels.
+    MCFixupKind Fixup = MCFixupKind(AArch64::fixup_a64_tlsdesc_call);
+    const MCExpr *Expr;
+    Expr = AArch64MCExpr::CreateTLSDesc(MI.getOperand(0).getExpr(), Ctx);
+    Fixups.push_back(MCFixup::Create(0, Expr, Fixup));
+    return;
+  }
+
+  uint32_t Binary = getBinaryCodeForInstr(MI, Fixups);
+
+  EmitInstruction(Binary, OS);
+}
+
+
+#include "AArch64GenMCCodeEmitter.inc"
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64MCExpr.cpp b/lib/Target/AArch64/MCTargetDesc/AArch64MCExpr.cpp
new file mode 100644
index 000000000000..c1abfe74dfdd
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64MCExpr.cpp
@@ -0,0 +1,178 @@
+//===-- AArch64MCExpr.cpp - AArch64 specific MC expression classes --------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the implementation of the assembly expression modifiers
+// accepted by the AArch64 architecture (e.g. ":lo12:", ":gottprel_g1:", ...).
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "aarch64mcexpr"
+#include "AArch64MCExpr.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCELF.h"
+#include "llvm/Object/ELF.h"
+
+using namespace llvm;
+
+const AArch64MCExpr*
+AArch64MCExpr::Create(VariantKind Kind, const MCExpr *Expr,
+                      MCContext &Ctx) {
+  return new (Ctx) AArch64MCExpr(Kind, Expr);
+}
+
+void AArch64MCExpr::PrintImpl(raw_ostream &OS) const {
+  switch (Kind) {
+  default: llvm_unreachable("Invalid kind!");
+  case VK_AARCH64_GOT:              OS << ":got:"; break;
+  case VK_AARCH64_GOT_LO12:         OS << ":got_lo12:"; break;
+  case VK_AARCH64_LO12:             OS << ":lo12:"; break;
+  case VK_AARCH64_ABS_G0:           OS << ":abs_g0:"; break;
+  case VK_AARCH64_ABS_G0_NC:        OS << ":abs_g0_nc:"; break;
+  case VK_AARCH64_ABS_G1:           OS << ":abs_g1:"; break;
+  case VK_AARCH64_ABS_G1_NC:        OS << ":abs_g1_nc:"; break;
+  case VK_AARCH64_ABS_G2:           OS << ":abs_g2:"; break;
+  case VK_AARCH64_ABS_G2_NC:        OS << ":abs_g2_nc:"; break;
+  case VK_AARCH64_ABS_G3:           OS << ":abs_g3:"; break;
+  case VK_AARCH64_SABS_G0:          OS << ":abs_g0_s:"; break;
+  case VK_AARCH64_SABS_G1:          OS << ":abs_g1_s:"; break;
+  case VK_AARCH64_SABS_G2:          OS << ":abs_g2_s:"; break;
+  case VK_AARCH64_DTPREL_G2:        OS << ":dtprel_g2:"; break;
+  case VK_AARCH64_DTPREL_G1:        OS << ":dtprel_g1:"; break;
+  case VK_AARCH64_DTPREL_G1_NC:     OS << ":dtprel_g1_nc:"; break;
+  case VK_AARCH64_DTPREL_G0:        OS << ":dtprel_g0:"; break;
+  case VK_AARCH64_DTPREL_G0_NC:     OS << ":dtprel_g0_nc:"; break;
+  case VK_AARCH64_DTPREL_HI12:      OS << ":dtprel_hi12:"; break;
+  case VK_AARCH64_DTPREL_LO12:      OS << ":dtprel_lo12:"; break;
+  case VK_AARCH64_DTPREL_LO12_NC:   OS << ":dtprel_lo12_nc:"; break;
+  case VK_AARCH64_GOTTPREL_G1:      OS << ":gottprel_g1:"; break;
+  case VK_AARCH64_GOTTPREL_G0_NC:   OS << ":gottprel_g0_nc:"; break;
+  case VK_AARCH64_GOTTPREL:         OS << ":gottprel:"; break;
+  case VK_AARCH64_GOTTPREL_LO12:    OS << ":gottprel_lo12:"; break;
+  case VK_AARCH64_TPREL_G2:         OS << ":tprel_g2:"; break;
+  case VK_AARCH64_TPREL_G1:         OS << ":tprel_g1:"; break;
+  case VK_AARCH64_TPREL_G1_NC:      OS << ":tprel_g1_nc:"; break;
+  case VK_AARCH64_TPREL_G0:         OS << ":tprel_g0:"; break;
+  case VK_AARCH64_TPREL_G0_NC:      OS << ":tprel_g0_nc:"; break;
+  case VK_AARCH64_TPREL_HI12:       OS << ":tprel_hi12:"; break;
+  case VK_AARCH64_TPREL_LO12:       OS << ":tprel_lo12:"; break;
+  case VK_AARCH64_TPREL_LO12_NC:    OS << ":tprel_lo12_nc:"; break;
+  case VK_AARCH64_TLSDESC:          OS << ":tlsdesc:"; break;
+  case VK_AARCH64_TLSDESC_LO12:     OS << ":tlsdesc_lo12:"; break;
+
+  }
+
+  const MCExpr *Expr = getSubExpr();
+  if (Expr->getKind() != MCExpr::SymbolRef)
+    OS << '(';
+  Expr->print(OS);
+  if (Expr->getKind() != MCExpr::SymbolRef)
+    OS << ')';
+}
+
+bool
+AArch64MCExpr::EvaluateAsRelocatableImpl(MCValue &Res,
+                                         const MCAsmLayout *Layout) const {
+  return getSubExpr()->EvaluateAsRelocatable(Res, *Layout);
+}
+
+static void fixELFSymbolsInTLSFixupsImpl(const MCExpr *Expr, MCAssembler &Asm) {
+  switch (Expr->getKind()) {
+  case MCExpr::Target:
+    llvm_unreachable("Can't handle nested target expression");
+    break;
+  case MCExpr::Constant:
+    break;
+
+  case MCExpr::Binary: {
+    const MCBinaryExpr *BE = cast<MCBinaryExpr>(Expr);
+    fixELFSymbolsInTLSFixupsImpl(BE->getLHS(), Asm);
+    fixELFSymbolsInTLSFixupsImpl(BE->getRHS(), Asm);
+    break;
+  }
+
+  case MCExpr::SymbolRef: {
+    // We're known to be under a TLS fixup, so any symbol should be
+    // modified. There should be only one.
+    const MCSymbolRefExpr &SymRef = *cast<MCSymbolRefExpr>(Expr);
+    MCSymbolData &SD = Asm.getOrCreateSymbolData(SymRef.getSymbol());
+    MCELF::SetType(SD, ELF::STT_TLS);
+    break;
+  }
+
+  case MCExpr::Unary:
+    fixELFSymbolsInTLSFixupsImpl(cast<MCUnaryExpr>(Expr)->getSubExpr(), Asm);
+    break;
+  }
+}
+
+void AArch64MCExpr::fixELFSymbolsInTLSFixups(MCAssembler &Asm) const {
+  switch (getKind()) {
+  default:
+    return;
+  case VK_AARCH64_DTPREL_G2:
+  case VK_AARCH64_DTPREL_G1:
+  case VK_AARCH64_DTPREL_G1_NC:
+  case VK_AARCH64_DTPREL_G0:
+  case VK_AARCH64_DTPREL_G0_NC:
+  case VK_AARCH64_DTPREL_HI12:
+  case VK_AARCH64_DTPREL_LO12:
+  case VK_AARCH64_DTPREL_LO12_NC:
+  case VK_AARCH64_GOTTPREL_G1:
+  case VK_AARCH64_GOTTPREL_G0_NC:
+  case VK_AARCH64_GOTTPREL:
+  case VK_AARCH64_GOTTPREL_LO12:
+  case VK_AARCH64_TPREL_G2:
+  case VK_AARCH64_TPREL_G1:
+  case VK_AARCH64_TPREL_G1_NC:
+  case VK_AARCH64_TPREL_G0:
+  case VK_AARCH64_TPREL_G0_NC:
+  case VK_AARCH64_TPREL_HI12:
+  case VK_AARCH64_TPREL_LO12:
+  case VK_AARCH64_TPREL_LO12_NC:
+  case VK_AARCH64_TLSDESC:
+  case VK_AARCH64_TLSDESC_LO12:
+    break;
+  }
+
+  fixELFSymbolsInTLSFixupsImpl(getSubExpr(), Asm);
+}
+
+// FIXME: This basically copies MCObjectStreamer::AddValueSymbols. Perhaps
+// that method should be made public?
+// FIXME: really do above: now that two backends are using it.
+static void AddValueSymbolsImpl(const MCExpr *Value, MCAssembler *Asm) {
+  switch (Value->getKind()) {
+  case MCExpr::Target:
+    llvm_unreachable("Can't handle nested target expr!");
+    break;
+
+  case MCExpr::Constant:
+    break;
+
+  case MCExpr::Binary: {
+    const MCBinaryExpr *BE = cast<MCBinaryExpr>(Value);
+    AddValueSymbolsImpl(BE->getLHS(), Asm);
+    AddValueSymbolsImpl(BE->getRHS(), Asm);
+    break;
+  }
+
+  case MCExpr::SymbolRef:
+    Asm->getOrCreateSymbolData(cast<MCSymbolRefExpr>(Value)->getSymbol());
+    break;
+
+  case MCExpr::Unary:
+    AddValueSymbolsImpl(cast<MCUnaryExpr>(Value)->getSubExpr(), Asm);
+    break;
+  }
+}
+
+void AArch64MCExpr::AddValueSymbols(MCAssembler *Asm) const {
+  AddValueSymbolsImpl(getSubExpr(), Asm);
+}
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64MCExpr.h b/lib/Target/AArch64/MCTargetDesc/AArch64MCExpr.h
new file mode 100644
index 000000000000..c0e3b29474d1
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64MCExpr.h
@@ -0,0 +1,167 @@
+//==- AArch64MCExpr.h - AArch64 specific MC expression classes --*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file describes AArch64-specific MCExprs, used for modifiers like
+// ":lo12:" or ":gottprel_g1:".
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_AARCH64MCEXPR_H
+#define LLVM_AARCH64MCEXPR_H
+
+#include "llvm/MC/MCExpr.h"
+
+namespace llvm {
+
+class AArch64MCExpr : public MCTargetExpr {
+public:
+  enum VariantKind {
+    VK_AARCH64_None,
+    VK_AARCH64_GOT,      // :got: modifier in assembly
+    VK_AARCH64_GOT_LO12, // :got_lo12:
+    VK_AARCH64_LO12,     // :lo12:
+
+    VK_AARCH64_ABS_G0, // :abs_g0:
+    VK_AARCH64_ABS_G0_NC, // :abs_g0_nc:
+    VK_AARCH64_ABS_G1,
+    VK_AARCH64_ABS_G1_NC,
+    VK_AARCH64_ABS_G2,
+    VK_AARCH64_ABS_G2_NC,
+    VK_AARCH64_ABS_G3,
+
+    VK_AARCH64_SABS_G0, // :abs_g0_s:
+    VK_AARCH64_SABS_G1,
+    VK_AARCH64_SABS_G2,
+
+    VK_AARCH64_DTPREL_G2, // :dtprel_g2:
+    VK_AARCH64_DTPREL_G1,
+    VK_AARCH64_DTPREL_G1_NC,
+    VK_AARCH64_DTPREL_G0,
+    VK_AARCH64_DTPREL_G0_NC,
+    VK_AARCH64_DTPREL_HI12,
+    VK_AARCH64_DTPREL_LO12,
+    VK_AARCH64_DTPREL_LO12_NC,
+
+    VK_AARCH64_GOTTPREL_G1, // :gottprel:
+    VK_AARCH64_GOTTPREL_G0_NC,
+    VK_AARCH64_GOTTPREL,
+    VK_AARCH64_GOTTPREL_LO12,
+
+    VK_AARCH64_TPREL_G2, // :tprel:
+    VK_AARCH64_TPREL_G1,
+    VK_AARCH64_TPREL_G1_NC,
+    VK_AARCH64_TPREL_G0,
+    VK_AARCH64_TPREL_G0_NC,
+    VK_AARCH64_TPREL_HI12,
+    VK_AARCH64_TPREL_LO12,
+    VK_AARCH64_TPREL_LO12_NC,
+
+    VK_AARCH64_TLSDESC, // :tlsdesc:
+    VK_AARCH64_TLSDESC_LO12
+  };
+
+private:
+  const VariantKind Kind;
+  const MCExpr *Expr;
+
+  explicit AArch64MCExpr(VariantKind _Kind, const MCExpr *_Expr)
+    : Kind(_Kind), Expr(_Expr) {}
+
+public:
+  /// @name Construction
+  /// @{
+
+  static const AArch64MCExpr *Create(VariantKind Kind, const MCExpr *Expr,
+                                     MCContext &Ctx);
+
+  static const AArch64MCExpr *CreateLo12(const MCExpr *Expr, MCContext &Ctx) {
+    return Create(VK_AARCH64_LO12, Expr, Ctx);
+  }
+
+  static const AArch64MCExpr *CreateGOT(const MCExpr *Expr, MCContext &Ctx) {
+    return Create(VK_AARCH64_GOT, Expr, Ctx);
+  }
+
+  static const AArch64MCExpr *CreateGOTLo12(const MCExpr *Expr,
+                                            MCContext &Ctx) {
+    return Create(VK_AARCH64_GOT_LO12, Expr, Ctx);
+  }
+
+  static const AArch64MCExpr *CreateDTPREL_G1(const MCExpr *Expr,
+                                             MCContext &Ctx) {
+    return Create(VK_AARCH64_DTPREL_G1, Expr, Ctx);
+  }
+
+  static const AArch64MCExpr *CreateDTPREL_G0_NC(const MCExpr *Expr,
+                                                MCContext &Ctx) {
+    return Create(VK_AARCH64_DTPREL_G0_NC, Expr, Ctx);
+  }
+
+  static const AArch64MCExpr *CreateGOTTPREL(const MCExpr *Expr,
+                                             MCContext &Ctx) {
+    return Create(VK_AARCH64_GOTTPREL, Expr, Ctx);
+  }
+
+  static const AArch64MCExpr *CreateGOTTPRELLo12(const MCExpr *Expr,
+                                                 MCContext &Ctx) {
+    return Create(VK_AARCH64_GOTTPREL_LO12, Expr, Ctx);
+  }
+
+  static const AArch64MCExpr *CreateTLSDesc(const MCExpr *Expr,
+                                            MCContext &Ctx) {
+    return Create(VK_AARCH64_TLSDESC, Expr, Ctx);
+  }
+
+  static const AArch64MCExpr *CreateTLSDescLo12(const MCExpr *Expr,
+                                                MCContext &Ctx) {
+    return Create(VK_AARCH64_TLSDESC_LO12, Expr, Ctx);
+  }
+
+  static const AArch64MCExpr *CreateTPREL_G1(const MCExpr *Expr,
+                                             MCContext &Ctx) {
+    return Create(VK_AARCH64_TPREL_G1, Expr, Ctx);
+  }
+
+  static const AArch64MCExpr *CreateTPREL_G0_NC(const MCExpr *Expr,
+                                                MCContext &Ctx) {
+    return Create(VK_AARCH64_TPREL_G0_NC, Expr, Ctx);
+  }
+
+  /// @}
+  /// @name Accessors
+  /// @{
+
+  /// getOpcode - Get the kind of this expression.
+  VariantKind getKind() const { return Kind; }
+
+  /// getSubExpr - Get the child of this expression.
+  const MCExpr *getSubExpr() const { return Expr; }
+
+  /// @}
+
+  void PrintImpl(raw_ostream &OS) const;
+  bool EvaluateAsRelocatableImpl(MCValue &Res,
+                                 const MCAsmLayout *Layout) const;
+  void AddValueSymbols(MCAssembler *) const;
+  const MCSection *FindAssociatedSection() const {
+    return getSubExpr()->FindAssociatedSection();
+  }
+
+  void fixELFSymbolsInTLSFixups(MCAssembler &Asm) const;
+
+  static bool classof(const MCExpr *E) {
+    return E->getKind() == MCExpr::Target;
+  }
+
+  static bool classof(const AArch64MCExpr *) { return true; }
+
+};
+} // end namespace llvm
+
+#endif
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64MCTargetDesc.cpp b/lib/Target/AArch64/MCTargetDesc/AArch64MCTargetDesc.cpp
new file mode 100644
index 000000000000..7960db08c8d6
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64MCTargetDesc.cpp
@@ -0,0 +1,194 @@
+//===-- AArch64MCTargetDesc.cpp - AArch64 Target Descriptions -------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file provides AArch64 specific target descriptions.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AArch64MCTargetDesc.h"
+#include "AArch64ELFStreamer.h"
+#include "AArch64MCAsmInfo.h"
+#include "InstPrinter/AArch64InstPrinter.h"
+#include "llvm/ADT/APInt.h"
+#include "llvm/MC/MCCodeGenInfo.h"
+#include "llvm/MC/MCInstrAnalysis.h"
+#include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/ErrorHandling.h"
+
+#define GET_REGINFO_MC_DESC
+#include "AArch64GenRegisterInfo.inc"
+
+#define GET_INSTRINFO_MC_DESC
+#include "AArch64GenInstrInfo.inc"
+
+#define GET_SUBTARGETINFO_MC_DESC
+#include "AArch64GenSubtargetInfo.inc"
+
+using namespace llvm;
+
+MCSubtargetInfo *AArch64_MC::createAArch64MCSubtargetInfo(StringRef TT,
+                                                          StringRef CPU,
+                                                          StringRef FS) {
+  MCSubtargetInfo *X = new MCSubtargetInfo();
+  InitAArch64MCSubtargetInfo(X, TT, CPU, "");
+  return X;
+}
+
+
+static MCInstrInfo *createAArch64MCInstrInfo() {
+  MCInstrInfo *X = new MCInstrInfo();
+  InitAArch64MCInstrInfo(X);
+  return X;
+}
+
+static MCRegisterInfo *createAArch64MCRegisterInfo(StringRef Triple) {
+  MCRegisterInfo *X = new MCRegisterInfo();
+  InitAArch64MCRegisterInfo(X, AArch64::X30);
+  return X;
+}
+
+static MCAsmInfo *createAArch64MCAsmInfo(const Target &T, StringRef TT) {
+  Triple TheTriple(TT);
+
+  MCAsmInfo *MAI = new AArch64ELFMCAsmInfo();
+  MachineLocation Dst(MachineLocation::VirtualFP);
+  MachineLocation Src(AArch64::XSP, 0);
+  MAI->addInitialFrameState(0, Dst, Src);
+
+  return MAI;
+}
+
+static MCCodeGenInfo *createAArch64MCCodeGenInfo(StringRef TT, Reloc::Model RM,
+                                                 CodeModel::Model CM,
+                                                 CodeGenOpt::Level OL) {
+  MCCodeGenInfo *X = new MCCodeGenInfo();
+  if (RM == Reloc::Default || RM == Reloc::DynamicNoPIC) {
+    // On ELF platforms the default static relocation model has a smart enough
+    // linker to cope with referencing external symbols defined in a shared
+    // library. Hence DynamicNoPIC doesn't need to be promoted to PIC.
+    RM = Reloc::Static;
+  }
+
+  if (CM == CodeModel::Default)
+    CM = CodeModel::Small;
+
+  X->InitMCCodeGenInfo(RM, CM, OL);
+  return X;
+}
+
+static MCStreamer *createMCStreamer(const Target &T, StringRef TT,
+                                    MCContext &Ctx, MCAsmBackend &MAB,
+                                    raw_ostream &OS,
+                                    MCCodeEmitter *Emitter,
+                                    bool RelaxAll,
+                                    bool NoExecStack) {
+  Triple TheTriple(TT);
+
+  return createAArch64ELFStreamer(Ctx, MAB, OS, Emitter, RelaxAll, NoExecStack);
+}
+
+
+static MCInstPrinter *createAArch64MCInstPrinter(const Target &T,
+                                                 unsigned SyntaxVariant,
+                                                 const MCAsmInfo &MAI,
+                                                 const MCInstrInfo &MII,
+                                                 const MCRegisterInfo &MRI,
+                                                 const MCSubtargetInfo &STI) {
+  if (SyntaxVariant == 0)
+    return new AArch64InstPrinter(MAI, MII, MRI, STI);
+  return 0;
+}
+
+namespace {
+
+class AArch64MCInstrAnalysis : public MCInstrAnalysis {
+public:
+  AArch64MCInstrAnalysis(const MCInstrInfo *Info) : MCInstrAnalysis(Info) {}
+
+  virtual bool isUnconditionalBranch(const MCInst &Inst) const {
+    if (Inst.getOpcode() == AArch64::Bcc
+        && Inst.getOperand(0).getImm() == A64CC::AL)
+      return true;
+    return MCInstrAnalysis::isUnconditionalBranch(Inst);
+  }
+
+  virtual bool isConditionalBranch(const MCInst &Inst) const {
+    if (Inst.getOpcode() == AArch64::Bcc
+        && Inst.getOperand(0).getImm() == A64CC::AL)
+      return false;
+    return MCInstrAnalysis::isConditionalBranch(Inst);
+  }
+
+  uint64_t evaluateBranch(const MCInst &Inst, uint64_t Addr,
+                          uint64_t Size) const {
+    unsigned LblOperand = Inst.getOpcode() == AArch64::Bcc ? 1 : 0;
+    // FIXME: We only handle PCRel branches for now.
+    if (Info->get(Inst.getOpcode()).OpInfo[LblOperand].OperandType
+        != MCOI::OPERAND_PCREL)
+      return -1ULL;
+
+    int64_t Imm = Inst.getOperand(LblOperand).getImm();
+
+    return Addr + Imm;
+  }
+};
+
+}
+
+static MCInstrAnalysis *createAArch64MCInstrAnalysis(const MCInstrInfo *Info) {
+  return new AArch64MCInstrAnalysis(Info);
+}
+
+
+
+extern "C" void LLVMInitializeAArch64TargetMC() {
+  // Register the MC asm info.
+  RegisterMCAsmInfoFn A(TheAArch64Target, createAArch64MCAsmInfo);
+
+  // Register the MC codegen info.
+  TargetRegistry::RegisterMCCodeGenInfo(TheAArch64Target,
+                                        createAArch64MCCodeGenInfo);
+
+  // Register the MC instruction info.
+  TargetRegistry::RegisterMCInstrInfo(TheAArch64Target,
+                                      createAArch64MCInstrInfo);
+
+  // Register the MC register info.
+  TargetRegistry::RegisterMCRegInfo(TheAArch64Target,
+                                    createAArch64MCRegisterInfo);
+
+  // Register the MC subtarget info.
+  using AArch64_MC::createAArch64MCSubtargetInfo;
+  TargetRegistry::RegisterMCSubtargetInfo(TheAArch64Target,
+                                          createAArch64MCSubtargetInfo);
+
+  // Register the MC instruction analyzer.
+  TargetRegistry::RegisterMCInstrAnalysis(TheAArch64Target,
+                                          createAArch64MCInstrAnalysis);
+
+  // Register the MC Code Emitter
+  TargetRegistry::RegisterMCCodeEmitter(TheAArch64Target,
+                                        createAArch64MCCodeEmitter);
+
+  // Register the asm backend.
+  TargetRegistry::RegisterMCAsmBackend(TheAArch64Target,
+                                       createAArch64AsmBackend);
+
+  // Register the object streamer.
+  TargetRegistry::RegisterMCObjectStreamer(TheAArch64Target,
+                                           createMCStreamer);
+
+  // Register the MCInstPrinter.
+  TargetRegistry::RegisterMCInstPrinter(TheAArch64Target,
+                                        createAArch64MCInstPrinter);
+}
diff --git a/lib/Target/AArch64/MCTargetDesc/AArch64MCTargetDesc.h b/lib/Target/AArch64/MCTargetDesc/AArch64MCTargetDesc.h
new file mode 100644
index 000000000000..3849fe379513
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/AArch64MCTargetDesc.h
@@ -0,0 +1,65 @@
+//===-- AArch64MCTargetDesc.h - AArch64 Target Descriptions -----*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file provides AArch64 specific target descriptions.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_AARCH64MCTARGETDESC_H
+#define LLVM_AARCH64MCTARGETDESC_H
+
+#include "llvm/Support/DataTypes.h"
+
+namespace llvm {
+class MCAsmBackend;
+class MCCodeEmitter;
+class MCContext;
+class MCInstrInfo;
+class MCObjectWriter;
+class MCRegisterInfo;
+class MCSubtargetInfo;
+class StringRef;
+class Target;
+class raw_ostream;
+
+extern Target TheAArch64Target;
+
+namespace AArch64_MC {
+  MCSubtargetInfo *createAArch64MCSubtargetInfo(StringRef TT, StringRef CPU,
+                                                StringRef FS);
+}
+
+MCCodeEmitter *createAArch64MCCodeEmitter(const MCInstrInfo &MCII,
+                                          const MCRegisterInfo &MRI,
+                                          const MCSubtargetInfo &STI,
+                                          MCContext &Ctx);
+
+MCObjectWriter *createAArch64ELFObjectWriter(raw_ostream &OS,
+                                             uint8_t OSABI);
+
+MCAsmBackend *createAArch64AsmBackend(const Target &T, StringRef TT,
+                                      StringRef CPU);
+
+} // End llvm namespace
+
+// Defines symbolic names for AArch64 registers.  This defines a mapping from
+// register name to register number.
+//
+#define GET_REGINFO_ENUM
+#include "AArch64GenRegisterInfo.inc"
+
+// Defines symbolic names for the AArch64 instructions.
+//
+#define GET_INSTRINFO_ENUM
+#include "AArch64GenInstrInfo.inc"
+
+#define GET_SUBTARGETINFO_ENUM
+#include "AArch64GenSubtargetInfo.inc"
+
+#endif
diff --git a/lib/Target/AArch64/MCTargetDesc/CMakeLists.txt b/lib/Target/AArch64/MCTargetDesc/CMakeLists.txt
new file mode 100644
index 000000000000..44c66a224e30
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/CMakeLists.txt
@@ -0,0 +1,13 @@
+add_llvm_library(LLVMAArch64Desc
+  AArch64AsmBackend.cpp
+  AArch64ELFObjectWriter.cpp
+  AArch64ELFStreamer.cpp
+  AArch64MCAsmInfo.cpp
+  AArch64MCCodeEmitter.cpp
+  AArch64MCExpr.cpp
+  AArch64MCTargetDesc.cpp
+  )
+add_dependencies(LLVMAArch64Desc AArch64CommonTableGen)
+
+# Hack: we need to include 'main' target directory to grab private headers
+include_directories(${CMAKE_CURRENT_SOURCE_DIR}/.. ${CMAKE_CURRENT_BINARY_DIR}/..)
diff --git a/lib/Target/AArch64/MCTargetDesc/LLVMBuild.txt b/lib/Target/AArch64/MCTargetDesc/LLVMBuild.txt
new file mode 100644
index 000000000000..37c8035a49f9
--- /dev/null
+++ b/lib/Target/AArch64/MCTargetDesc/LLVMBuild.txt
@@ -0,0 +1,24 @@
+;===- ./lib/Target/AArch64/MCTargetDesc/LLVMBuild.txt ----------*- Conf -*--===;
+;
+;                     The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+;   http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[component_0]
+type = Library
+name = AArch64Desc
+parent = AArch64
+required_libraries = AArch64AsmPrinter AArch64Info MC Support
+add_to_library_groups = AArch64
+
diff --git a/lib/Target/CellSPU/MCTargetDesc/Makefile b/lib/Target/AArch64/MCTargetDesc/Makefile
index 10d9a42239ad..5779ac5ac60a 100644
--- a/lib/Target/CellSPU/MCTargetDesc/Makefile
+++ b/lib/Target/AArch64/MCTargetDesc/Makefile
@@ -1,4 +1,4 @@
-##===- lib/Target/CellSPU/TargetDesc/Makefile --------------*- Makefile -*-===##
+##===- lib/Target/AArch64/TargetDesc/Makefile --------------*- Makefile -*-===##
 #
 #                     The LLVM Compiler Infrastructure
 #
@@ -8,7 +8,7 @@
 ##===----------------------------------------------------------------------===##
 
 LEVEL = ../../../..
-LIBRARYNAME = LLVMCellSPUDesc
+LIBRARYNAME = LLVMAArch64Desc
 
 # Hack: we need to include 'main' target directory to grab private headers
 CPP.Flags += -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
diff --git a/lib/Target/AArch64/Makefile b/lib/Target/AArch64/Makefile
new file mode 100644
index 000000000000..641bb83c4775
--- /dev/null
+++ b/lib/Target/AArch64/Makefile
@@ -0,0 +1,30 @@
+##===- lib/Target/AArch64/Makefile -------------------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+
+LEVEL = ../../..
+LIBRARYNAME = LLVMAArch64CodeGen
+TARGET = AArch64
+
+# Make sure that tblgen is run, first thing.
+BUILT_SOURCES = AArch64GenAsmMatcher.inc \
+   AArch64GenAsmWriter.inc \
+   AArch64GenCallingConv.inc \
+   AArch64GenDAGISel.inc \
+   AArch64GenDisassemblerTables.inc \
+   AArch64GenInstrInfo.inc \
+   AArch64GenMCCodeEmitter.inc \
+   AArch64GenMCPseudoLowering.inc \
+   AArch64GenRegisterInfo.inc \
+   AArch64GenSubtargetInfo.inc
+
+DIRS = InstPrinter AsmParser Disassembler TargetInfo MCTargetDesc Utils
+
+include $(LEVEL)/Makefile.common
+
+
diff --git a/lib/Target/AArch64/README.txt b/lib/Target/AArch64/README.txt
new file mode 100644
index 000000000000..601990f17dee
--- /dev/null
+++ b/lib/Target/AArch64/README.txt
@@ -0,0 +1,2 @@
+This file will contain changes that need to be made before AArch64 can become an
+officially supported target. Currently a placeholder.
diff --git a/lib/Target/AArch64/TargetInfo/AArch64TargetInfo.cpp b/lib/Target/AArch64/TargetInfo/AArch64TargetInfo.cpp
new file mode 100644
index 000000000000..b8099cb26b0f
--- /dev/null
+++ b/lib/Target/AArch64/TargetInfo/AArch64TargetInfo.cpp
@@ -0,0 +1,24 @@
+//===-- AArch64TargetInfo.cpp - AArch64 Target Implementation -------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the key registration step for the architecture.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AArch64.h"
+#include "llvm/IR/Module.h"
+#include "llvm/Support/TargetRegistry.h"
+using namespace llvm;
+
+Target llvm::TheAArch64Target;
+
+extern "C" void LLVMInitializeAArch64TargetInfo() {
+  RegisterTarget<Triple::aarch64>
+    X(TheAArch64Target, "aarch64", "AArch64");
+}
diff --git a/lib/Target/AArch64/TargetInfo/CMakeLists.txt b/lib/Target/AArch64/TargetInfo/CMakeLists.txt
new file mode 100644
index 000000000000..e236eed00be1
--- /dev/null
+++ b/lib/Target/AArch64/TargetInfo/CMakeLists.txt
@@ -0,0 +1,7 @@
+include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
+
+add_llvm_library(LLVMAArch64Info
+  AArch64TargetInfo.cpp
+  )
+
+add_dependencies(LLVMAArch64Info AArch64CommonTableGen)
diff --git a/lib/Target/CellSPU/TargetInfo/LLVMBuild.txt b/lib/Target/AArch64/TargetInfo/LLVMBuild.txt
index 6937e705ff7f..5b003f012218 100644
--- a/lib/Target/CellSPU/TargetInfo/LLVMBuild.txt
+++ b/lib/Target/AArch64/TargetInfo/LLVMBuild.txt
@@ -1,4 +1,4 @@
-;===- ./lib/Target/CellSPU/TargetInfo/LLVMBuild.txt ------------*- Conf -*--===;
+;===- ./lib/Target/AArch64/TargetInfo/LLVMBuild.txt ------------*- Conf -*--===;
 ;
 ;                     The LLVM Compiler Infrastructure
 ;
@@ -17,7 +17,8 @@
 
 [component_0]
 type = Library
-name = CellSPUInfo
-parent = CellSPU
+name = AArch64Info
+parent = AArch64
 required_libraries = MC Support Target
-add_to_library_groups = CellSPU
+add_to_library_groups = AArch64
+
diff --git a/lib/Target/CellSPU/TargetInfo/Makefile b/lib/Target/AArch64/TargetInfo/Makefile
index 9cb6827b4323..9dc9aa4bccf7 100644
--- a/lib/Target/CellSPU/TargetInfo/Makefile
+++ b/lib/Target/AArch64/TargetInfo/Makefile
@@ -1,4 +1,4 @@
-##===- lib/Target/CellSPU/TargetInfo/Makefile --------------*- Makefile -*-===##
+##===- lib/Target/AArch64/TargetInfo/Makefile --------------*- Makefile -*-===##
 #
 #                     The LLVM Compiler Infrastructure
 #
@@ -7,7 +7,7 @@
 #
 ##===----------------------------------------------------------------------===##
 LEVEL = ../../../..
-LIBRARYNAME = LLVMCellSPUInfo
+LIBRARYNAME = LLVMAArch64Info
 
 # Hack: we need to include 'main' target directory to grab private headers
 CPPFLAGS = -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
diff --git a/lib/Target/AArch64/Utils/AArch64BaseInfo.cpp b/lib/Target/AArch64/Utils/AArch64BaseInfo.cpp
new file mode 100644
index 000000000000..1678559aa084
--- /dev/null
+++ b/lib/Target/AArch64/Utils/AArch64BaseInfo.cpp
@@ -0,0 +1,1103 @@
+//===-- AArch64BaseInfo.cpp - AArch64 Base encoding information------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file provides basic encoding and assembly information for AArch64.
+//
+//===----------------------------------------------------------------------===//
+#include "AArch64BaseInfo.h"
+#include "llvm/ADT/APFloat.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/StringExtras.h"
+#include "llvm/Support/Regex.h"
+
+using namespace llvm;
+
+StringRef NamedImmMapper::toString(uint32_t Value, bool &Valid) const {
+  for (unsigned i = 0; i < NumPairs; ++i) {
+    if (Pairs[i].Value == Value) {
+      Valid = true;
+      return Pairs[i].Name;
+    }
+  }
+
+  Valid = false;
+  return StringRef();
+}
+
+uint32_t NamedImmMapper::fromString(StringRef Name, bool &Valid) const {
+  std::string LowerCaseName = Name.lower();
+  for (unsigned i = 0; i < NumPairs; ++i) {
+    if (Pairs[i].Name == LowerCaseName) {
+      Valid = true;
+      return Pairs[i].Value;
+    }
+  }
+
+  Valid = false;
+  return -1;
+}
+
+bool NamedImmMapper::validImm(uint32_t Value) const {
+  return Value < TooBigImm;
+}
+
+const NamedImmMapper::Mapping A64AT::ATMapper::ATPairs[] = {
+  {"s1e1r", S1E1R},
+  {"s1e2r", S1E2R},
+  {"s1e3r", S1E3R},
+  {"s1e1w", S1E1W},
+  {"s1e2w", S1E2W},
+  {"s1e3w", S1E3W},
+  {"s1e0r", S1E0R},
+  {"s1e0w", S1E0W},
+  {"s12e1r", S12E1R},
+  {"s12e1w", S12E1W},
+  {"s12e0r", S12E0R},
+  {"s12e0w", S12E0W},
+};
+
+A64AT::ATMapper::ATMapper()
+  : NamedImmMapper(ATPairs, 0) {}
+
+const NamedImmMapper::Mapping A64DB::DBarrierMapper::DBarrierPairs[] = {
+  {"oshld", OSHLD},
+  {"oshst", OSHST},
+  {"osh", OSH},
+  {"nshld", NSHLD},
+  {"nshst", NSHST},
+  {"nsh", NSH},
+  {"ishld", ISHLD},
+  {"ishst", ISHST},
+  {"ish", ISH},
+  {"ld", LD},
+  {"st", ST},
+  {"sy", SY}
+};
+
+A64DB::DBarrierMapper::DBarrierMapper()
+  : NamedImmMapper(DBarrierPairs, 16u) {}
+
+const NamedImmMapper::Mapping A64DC::DCMapper::DCPairs[] = {
+  {"zva", ZVA},
+  {"ivac", IVAC},
+  {"isw", ISW},
+  {"cvac", CVAC},
+  {"csw", CSW},
+  {"cvau", CVAU},
+  {"civac", CIVAC},
+  {"cisw", CISW}
+};
+
+A64DC::DCMapper::DCMapper()
+  : NamedImmMapper(DCPairs, 0) {}
+
+const NamedImmMapper::Mapping A64IC::ICMapper::ICPairs[] = {
+  {"ialluis",  IALLUIS},
+  {"iallu", IALLU},
+  {"ivau", IVAU}
+};
+
+A64IC::ICMapper::ICMapper()
+  : NamedImmMapper(ICPairs, 0) {}
+
+const NamedImmMapper::Mapping A64ISB::ISBMapper::ISBPairs[] = {
+  {"sy",  SY},
+};
+
+A64ISB::ISBMapper::ISBMapper()
+  : NamedImmMapper(ISBPairs, 16) {}
+
+const NamedImmMapper::Mapping A64PRFM::PRFMMapper::PRFMPairs[] = {
+  {"pldl1keep", PLDL1KEEP},
+  {"pldl1strm", PLDL1STRM},
+  {"pldl2keep", PLDL2KEEP},
+  {"pldl2strm", PLDL2STRM},
+  {"pldl3keep", PLDL3KEEP},
+  {"pldl3strm", PLDL3STRM},
+  {"plil1keep", PLIL1KEEP},
+  {"plil1strm", PLIL1STRM},
+  {"plil2keep", PLIL2KEEP},
+  {"plil2strm", PLIL2STRM},
+  {"plil3keep", PLIL3KEEP},
+  {"plil3strm", PLIL3STRM},
+  {"pstl1keep", PSTL1KEEP},
+  {"pstl1strm", PSTL1STRM},
+  {"pstl2keep", PSTL2KEEP},
+  {"pstl2strm", PSTL2STRM},
+  {"pstl3keep", PSTL3KEEP},
+  {"pstl3strm", PSTL3STRM}
+};
+
+A64PRFM::PRFMMapper::PRFMMapper()
+  : NamedImmMapper(PRFMPairs, 32) {}
+
+const NamedImmMapper::Mapping A64PState::PStateMapper::PStatePairs[] = {
+  {"spsel", SPSel},
+  {"daifset", DAIFSet},
+  {"daifclr", DAIFClr}
+};
+
+A64PState::PStateMapper::PStateMapper()
+  : NamedImmMapper(PStatePairs, 0) {}
+
+const NamedImmMapper::Mapping A64SysReg::MRSMapper::MRSPairs[] = {
+  {"mdccsr_el0", MDCCSR_EL0},
+  {"dbgdtrrx_el0", DBGDTRRX_EL0},
+  {"mdrar_el1", MDRAR_EL1},
+  {"oslsr_el1", OSLSR_EL1},
+  {"dbgauthstatus_el1", DBGAUTHSTATUS_EL1},
+  {"pmceid0_el0", PMCEID0_EL0},
+  {"pmceid1_el0", PMCEID1_EL0},
+  {"midr_el1", MIDR_EL1},
+  {"ccsidr_el1", CCSIDR_EL1},
+  {"clidr_el1", CLIDR_EL1},
+  {"ctr_el0", CTR_EL0},
+  {"mpidr_el1", MPIDR_EL1},
+  {"revidr_el1", REVIDR_EL1},
+  {"aidr_el1", AIDR_EL1},
+  {"dczid_el0", DCZID_EL0},
+  {"id_pfr0_el1", ID_PFR0_EL1},
+  {"id_pfr1_el1", ID_PFR1_EL1},
+  {"id_dfr0_el1", ID_DFR0_EL1},
+  {"id_afr0_el1", ID_AFR0_EL1},
+  {"id_mmfr0_el1", ID_MMFR0_EL1},
+  {"id_mmfr1_el1", ID_MMFR1_EL1},
+  {"id_mmfr2_el1", ID_MMFR2_EL1},
+  {"id_mmfr3_el1", ID_MMFR3_EL1},
+  {"id_isar0_el1", ID_ISAR0_EL1},
+  {"id_isar1_el1", ID_ISAR1_EL1},
+  {"id_isar2_el1", ID_ISAR2_EL1},
+  {"id_isar3_el1", ID_ISAR3_EL1},
+  {"id_isar4_el1", ID_ISAR4_EL1},
+  {"id_isar5_el1", ID_ISAR5_EL1},
+  {"id_aa64pfr0_el1", ID_AA64PFR0_EL1},
+  {"id_aa64pfr1_el1", ID_AA64PFR1_EL1},
+  {"id_aa64dfr0_el1", ID_AA64DFR0_EL1},
+  {"id_aa64dfr1_el1", ID_AA64DFR1_EL1},
+  {"id_aa64afr0_el1", ID_AA64AFR0_EL1},
+  {"id_aa64afr1_el1", ID_AA64AFR1_EL1},
+  {"id_aa64isar0_el1", ID_AA64ISAR0_EL1},
+  {"id_aa64isar1_el1", ID_AA64ISAR1_EL1},
+  {"id_aa64mmfr0_el1", ID_AA64MMFR0_EL1},
+  {"id_aa64mmfr1_el1", ID_AA64MMFR1_EL1},
+  {"mvfr0_el1", MVFR0_EL1},
+  {"mvfr1_el1", MVFR1_EL1},
+  {"mvfr2_el1", MVFR2_EL1},
+  {"rvbar_el1", RVBAR_EL1},
+  {"rvbar_el2", RVBAR_EL2},
+  {"rvbar_el3", RVBAR_EL3},
+  {"isr_el1", ISR_EL1},
+  {"cntpct_el0", CNTPCT_EL0},
+  {"cntvct_el0", CNTVCT_EL0},
+
+  // Trace registers
+  {"trcstatr", TRCSTATR},
+  {"trcidr8", TRCIDR8},
+  {"trcidr9", TRCIDR9},
+  {"trcidr10", TRCIDR10},
+  {"trcidr11", TRCIDR11},
+  {"trcidr12", TRCIDR12},
+  {"trcidr13", TRCIDR13},
+  {"trcidr0", TRCIDR0},
+  {"trcidr1", TRCIDR1},
+  {"trcidr2", TRCIDR2},
+  {"trcidr3", TRCIDR3},
+  {"trcidr4", TRCIDR4},
+  {"trcidr5", TRCIDR5},
+  {"trcidr6", TRCIDR6},
+  {"trcidr7", TRCIDR7},
+  {"trcoslsr", TRCOSLSR},
+  {"trcpdsr", TRCPDSR},
+  {"trcdevaff0", TRCDEVAFF0},
+  {"trcdevaff1", TRCDEVAFF1},
+  {"trclsr", TRCLSR},
+  {"trcauthstatus", TRCAUTHSTATUS},
+  {"trcdevarch", TRCDEVARCH},
+  {"trcdevid", TRCDEVID},
+  {"trcdevtype", TRCDEVTYPE},
+  {"trcpidr4", TRCPIDR4},
+  {"trcpidr5", TRCPIDR5},
+  {"trcpidr6", TRCPIDR6},
+  {"trcpidr7", TRCPIDR7},
+  {"trcpidr0", TRCPIDR0},
+  {"trcpidr1", TRCPIDR1},
+  {"trcpidr2", TRCPIDR2},
+  {"trcpidr3", TRCPIDR3},
+  {"trccidr0", TRCCIDR0},
+  {"trccidr1", TRCCIDR1},
+  {"trccidr2", TRCCIDR2},
+  {"trccidr3", TRCCIDR3},
+
+  // GICv3 registers
+  {"icc_iar1_el1", ICC_IAR1_EL1},
+  {"icc_iar0_el1", ICC_IAR0_EL1},
+  {"icc_hppir1_el1", ICC_HPPIR1_EL1},
+  {"icc_hppir0_el1", ICC_HPPIR0_EL1},
+  {"icc_rpr_el1", ICC_RPR_EL1},
+  {"ich_vtr_el2", ICH_VTR_EL2},
+  {"ich_eisr_el2", ICH_EISR_EL2},
+  {"ich_elsr_el2", ICH_ELSR_EL2}
+};
+
+A64SysReg::MRSMapper::MRSMapper() {
+    InstPairs = &MRSPairs[0];
+    NumInstPairs = llvm::array_lengthof(MRSPairs);
+}
+
+const NamedImmMapper::Mapping A64SysReg::MSRMapper::MSRPairs[] = {
+  {"dbgdtrtx_el0", DBGDTRTX_EL0},
+  {"oslar_el1", OSLAR_EL1},
+  {"pmswinc_el0", PMSWINC_EL0},
+
+  // Trace registers
+  {"trcoslar", TRCOSLAR},
+  {"trclar", TRCLAR},
+
+  // GICv3 registers
+  {"icc_eoir1_el1", ICC_EOIR1_EL1},
+  {"icc_eoir0_el1", ICC_EOIR0_EL1},
+  {"icc_dir_el1", ICC_DIR_EL1},
+  {"icc_sgi1r_el1", ICC_SGI1R_EL1},
+  {"icc_asgi1r_el1", ICC_ASGI1R_EL1},
+  {"icc_sgi0r_el1", ICC_SGI0R_EL1}
+};
+
+A64SysReg::MSRMapper::MSRMapper() {
+    InstPairs = &MSRPairs[0];
+    NumInstPairs = llvm::array_lengthof(MSRPairs);
+}
+
+
+const NamedImmMapper::Mapping A64SysReg::SysRegMapper::SysRegPairs[] = {
+  {"osdtrrx_el1", OSDTRRX_EL1},
+  {"osdtrtx_el1",  OSDTRTX_EL1},
+  {"teecr32_el1", TEECR32_EL1},
+  {"mdccint_el1", MDCCINT_EL1},
+  {"mdscr_el1", MDSCR_EL1},
+  {"dbgdtr_el0", DBGDTR_EL0},
+  {"oseccr_el1", OSECCR_EL1},
+  {"dbgvcr32_el2", DBGVCR32_EL2},
+  {"dbgbvr0_el1", DBGBVR0_EL1},
+  {"dbgbvr1_el1", DBGBVR1_EL1},
+  {"dbgbvr2_el1", DBGBVR2_EL1},
+  {"dbgbvr3_el1", DBGBVR3_EL1},
+  {"dbgbvr4_el1", DBGBVR4_EL1},
+  {"dbgbvr5_el1", DBGBVR5_EL1},
+  {"dbgbvr6_el1", DBGBVR6_EL1},
+  {"dbgbvr7_el1", DBGBVR7_EL1},
+  {"dbgbvr8_el1", DBGBVR8_EL1},
+  {"dbgbvr9_el1", DBGBVR9_EL1},
+  {"dbgbvr10_el1", DBGBVR10_EL1},
+  {"dbgbvr11_el1", DBGBVR11_EL1},
+  {"dbgbvr12_el1", DBGBVR12_EL1},
+  {"dbgbvr13_el1", DBGBVR13_EL1},
+  {"dbgbvr14_el1", DBGBVR14_EL1},
+  {"dbgbvr15_el1", DBGBVR15_EL1},
+  {"dbgbcr0_el1", DBGBCR0_EL1},
+  {"dbgbcr1_el1", DBGBCR1_EL1},
+  {"dbgbcr2_el1", DBGBCR2_EL1},
+  {"dbgbcr3_el1", DBGBCR3_EL1},
+  {"dbgbcr4_el1", DBGBCR4_EL1},
+  {"dbgbcr5_el1", DBGBCR5_EL1},
+  {"dbgbcr6_el1", DBGBCR6_EL1},
+  {"dbgbcr7_el1", DBGBCR7_EL1},
+  {"dbgbcr8_el1", DBGBCR8_EL1},
+  {"dbgbcr9_el1", DBGBCR9_EL1},
+  {"dbgbcr10_el1", DBGBCR10_EL1},
+  {"dbgbcr11_el1", DBGBCR11_EL1},
+  {"dbgbcr12_el1", DBGBCR12_EL1},
+  {"dbgbcr13_el1", DBGBCR13_EL1},
+  {"dbgbcr14_el1", DBGBCR14_EL1},
+  {"dbgbcr15_el1", DBGBCR15_EL1},
+  {"dbgwvr0_el1", DBGWVR0_EL1},
+  {"dbgwvr1_el1", DBGWVR1_EL1},
+  {"dbgwvr2_el1", DBGWVR2_EL1},
+  {"dbgwvr3_el1", DBGWVR3_EL1},
+  {"dbgwvr4_el1", DBGWVR4_EL1},
+  {"dbgwvr5_el1", DBGWVR5_EL1},
+  {"dbgwvr6_el1", DBGWVR6_EL1},
+  {"dbgwvr7_el1", DBGWVR7_EL1},
+  {"dbgwvr8_el1", DBGWVR8_EL1},
+  {"dbgwvr9_el1", DBGWVR9_EL1},
+  {"dbgwvr10_el1", DBGWVR10_EL1},
+  {"dbgwvr11_el1", DBGWVR11_EL1},
+  {"dbgwvr12_el1", DBGWVR12_EL1},
+  {"dbgwvr13_el1", DBGWVR13_EL1},
+  {"dbgwvr14_el1", DBGWVR14_EL1},
+  {"dbgwvr15_el1", DBGWVR15_EL1},
+  {"dbgwcr0_el1", DBGWCR0_EL1},
+  {"dbgwcr1_el1", DBGWCR1_EL1},
+  {"dbgwcr2_el1", DBGWCR2_EL1},
+  {"dbgwcr3_el1", DBGWCR3_EL1},
+  {"dbgwcr4_el1", DBGWCR4_EL1},
+  {"dbgwcr5_el1", DBGWCR5_EL1},
+  {"dbgwcr6_el1", DBGWCR6_EL1},
+  {"dbgwcr7_el1", DBGWCR7_EL1},
+  {"dbgwcr8_el1", DBGWCR8_EL1},
+  {"dbgwcr9_el1", DBGWCR9_EL1},
+  {"dbgwcr10_el1", DBGWCR10_EL1},
+  {"dbgwcr11_el1", DBGWCR11_EL1},
+  {"dbgwcr12_el1", DBGWCR12_EL1},
+  {"dbgwcr13_el1", DBGWCR13_EL1},
+  {"dbgwcr14_el1", DBGWCR14_EL1},
+  {"dbgwcr15_el1", DBGWCR15_EL1},
+  {"teehbr32_el1", TEEHBR32_EL1},
+  {"osdlr_el1", OSDLR_EL1},
+  {"dbgprcr_el1", DBGPRCR_EL1},
+  {"dbgclaimset_el1", DBGCLAIMSET_EL1},
+  {"dbgclaimclr_el1", DBGCLAIMCLR_EL1},
+  {"csselr_el1", CSSELR_EL1},
+  {"vpidr_el2", VPIDR_EL2},
+  {"vmpidr_el2", VMPIDR_EL2},
+  {"sctlr_el1", SCTLR_EL1},
+  {"sctlr_el2", SCTLR_EL2},
+  {"sctlr_el3", SCTLR_EL3},
+  {"actlr_el1", ACTLR_EL1},
+  {"actlr_el2", ACTLR_EL2},
+  {"actlr_el3", ACTLR_EL3},
+  {"cpacr_el1", CPACR_EL1},
+  {"hcr_el2", HCR_EL2},
+  {"scr_el3", SCR_EL3},
+  {"mdcr_el2", MDCR_EL2},
+  {"sder32_el3", SDER32_EL3},
+  {"cptr_el2", CPTR_EL2},
+  {"cptr_el3", CPTR_EL3},
+  {"hstr_el2", HSTR_EL2},
+  {"hacr_el2", HACR_EL2},
+  {"mdcr_el3", MDCR_EL3},
+  {"ttbr0_el1", TTBR0_EL1},
+  {"ttbr0_el2", TTBR0_EL2},
+  {"ttbr0_el3", TTBR0_EL3},
+  {"ttbr1_el1", TTBR1_EL1},
+  {"tcr_el1", TCR_EL1},
+  {"tcr_el2", TCR_EL2},
+  {"tcr_el3", TCR_EL3},
+  {"vttbr_el2", VTTBR_EL2},
+  {"vtcr_el2", VTCR_EL2},
+  {"dacr32_el2", DACR32_EL2},
+  {"spsr_el1", SPSR_EL1},
+  {"spsr_el2", SPSR_EL2},
+  {"spsr_el3", SPSR_EL3},
+  {"elr_el1", ELR_EL1},
+  {"elr_el2", ELR_EL2},
+  {"elr_el3", ELR_EL3},
+  {"sp_el0", SP_EL0},
+  {"sp_el1", SP_EL1},
+  {"sp_el2", SP_EL2},
+  {"spsel", SPSel},
+  {"nzcv", NZCV},
+  {"daif", DAIF},
+  {"currentel", CurrentEL},
+  {"spsr_irq", SPSR_irq},
+  {"spsr_abt", SPSR_abt},
+  {"spsr_und", SPSR_und},
+  {"spsr_fiq", SPSR_fiq},
+  {"fpcr", FPCR},
+  {"fpsr", FPSR},
+  {"dspsr_el0", DSPSR_EL0},
+  {"dlr_el0", DLR_EL0},
+  {"ifsr32_el2", IFSR32_EL2},
+  {"afsr0_el1", AFSR0_EL1},
+  {"afsr0_el2", AFSR0_EL2},
+  {"afsr0_el3", AFSR0_EL3},
+  {"afsr1_el1", AFSR1_EL1},
+  {"afsr1_el2", AFSR1_EL2},
+  {"afsr1_el3", AFSR1_EL3},
+  {"esr_el1", ESR_EL1},
+  {"esr_el2", ESR_EL2},
+  {"esr_el3", ESR_EL3},
+  {"fpexc32_el2", FPEXC32_EL2},
+  {"far_el1", FAR_EL1},
+  {"far_el2", FAR_EL2},
+  {"far_el3", FAR_EL3},
+  {"hpfar_el2", HPFAR_EL2},
+  {"par_el1", PAR_EL1},
+  {"pmcr_el0", PMCR_EL0},
+  {"pmcntenset_el0", PMCNTENSET_EL0},
+  {"pmcntenclr_el0", PMCNTENCLR_EL0},
+  {"pmovsclr_el0", PMOVSCLR_EL0},
+  {"pmselr_el0", PMSELR_EL0},
+  {"pmccntr_el0", PMCCNTR_EL0},
+  {"pmxevtyper_el0", PMXEVTYPER_EL0},
+  {"pmxevcntr_el0", PMXEVCNTR_EL0},
+  {"pmuserenr_el0", PMUSERENR_EL0},
+  {"pmintenset_el1", PMINTENSET_EL1},
+  {"pmintenclr_el1", PMINTENCLR_EL1},
+  {"pmovsset_el0", PMOVSSET_EL0},
+  {"mair_el1", MAIR_EL1},
+  {"mair_el2", MAIR_EL2},
+  {"mair_el3", MAIR_EL3},
+  {"amair_el1", AMAIR_EL1},
+  {"amair_el2", AMAIR_EL2},
+  {"amair_el3", AMAIR_EL3},
+  {"vbar_el1", VBAR_EL1},
+  {"vbar_el2", VBAR_EL2},
+  {"vbar_el3", VBAR_EL3},
+  {"rmr_el1", RMR_EL1},
+  {"rmr_el2", RMR_EL2},
+  {"rmr_el3", RMR_EL3},
+  {"contextidr_el1", CONTEXTIDR_EL1},
+  {"tpidr_el0", TPIDR_EL0},
+  {"tpidr_el2", TPIDR_EL2},
+  {"tpidr_el3", TPIDR_EL3},
+  {"tpidrro_el0", TPIDRRO_EL0},
+  {"tpidr_el1", TPIDR_EL1},
+  {"cntfrq_el0", CNTFRQ_EL0},
+  {"cntvoff_el2", CNTVOFF_EL2},
+  {"cntkctl_el1", CNTKCTL_EL1},
+  {"cnthctl_el2", CNTHCTL_EL2},
+  {"cntp_tval_el0", CNTP_TVAL_EL0},
+  {"cnthp_tval_el2", CNTHP_TVAL_EL2},
+  {"cntps_tval_el1", CNTPS_TVAL_EL1},
+  {"cntp_ctl_el0", CNTP_CTL_EL0},
+  {"cnthp_ctl_el2", CNTHP_CTL_EL2},
+  {"cntps_ctl_el1", CNTPS_CTL_EL1},
+  {"cntp_cval_el0", CNTP_CVAL_EL0},
+  {"cnthp_cval_el2", CNTHP_CVAL_EL2},
+  {"cntps_cval_el1", CNTPS_CVAL_EL1},
+  {"cntv_tval_el0", CNTV_TVAL_EL0},
+  {"cntv_ctl_el0", CNTV_CTL_EL0},
+  {"cntv_cval_el0", CNTV_CVAL_EL0},
+  {"pmevcntr0_el0", PMEVCNTR0_EL0},
+  {"pmevcntr1_el0", PMEVCNTR1_EL0},
+  {"pmevcntr2_el0", PMEVCNTR2_EL0},
+  {"pmevcntr3_el0", PMEVCNTR3_EL0},
+  {"pmevcntr4_el0", PMEVCNTR4_EL0},
+  {"pmevcntr5_el0", PMEVCNTR5_EL0},
+  {"pmevcntr6_el0", PMEVCNTR6_EL0},
+  {"pmevcntr7_el0", PMEVCNTR7_EL0},
+  {"pmevcntr8_el0", PMEVCNTR8_EL0},
+  {"pmevcntr9_el0", PMEVCNTR9_EL0},
+  {"pmevcntr10_el0", PMEVCNTR10_EL0},
+  {"pmevcntr11_el0", PMEVCNTR11_EL0},
+  {"pmevcntr12_el0", PMEVCNTR12_EL0},
+  {"pmevcntr13_el0", PMEVCNTR13_EL0},
+  {"pmevcntr14_el0", PMEVCNTR14_EL0},
+  {"pmevcntr15_el0", PMEVCNTR15_EL0},
+  {"pmevcntr16_el0", PMEVCNTR16_EL0},
+  {"pmevcntr17_el0", PMEVCNTR17_EL0},
+  {"pmevcntr18_el0", PMEVCNTR18_EL0},
+  {"pmevcntr19_el0", PMEVCNTR19_EL0},
+  {"pmevcntr20_el0", PMEVCNTR20_EL0},
+  {"pmevcntr21_el0", PMEVCNTR21_EL0},
+  {"pmevcntr22_el0", PMEVCNTR22_EL0},
+  {"pmevcntr23_el0", PMEVCNTR23_EL0},
+  {"pmevcntr24_el0", PMEVCNTR24_EL0},
+  {"pmevcntr25_el0", PMEVCNTR25_EL0},
+  {"pmevcntr26_el0", PMEVCNTR26_EL0},
+  {"pmevcntr27_el0", PMEVCNTR27_EL0},
+  {"pmevcntr28_el0", PMEVCNTR28_EL0},
+  {"pmevcntr29_el0", PMEVCNTR29_EL0},
+  {"pmevcntr30_el0", PMEVCNTR30_EL0},
+  {"pmccfiltr_el0", PMCCFILTR_EL0},
+  {"pmevtyper0_el0", PMEVTYPER0_EL0},
+  {"pmevtyper1_el0", PMEVTYPER1_EL0},
+  {"pmevtyper2_el0", PMEVTYPER2_EL0},
+  {"pmevtyper3_el0", PMEVTYPER3_EL0},
+  {"pmevtyper4_el0", PMEVTYPER4_EL0},
+  {"pmevtyper5_el0", PMEVTYPER5_EL0},
+  {"pmevtyper6_el0", PMEVTYPER6_EL0},
+  {"pmevtyper7_el0", PMEVTYPER7_EL0},
+  {"pmevtyper8_el0", PMEVTYPER8_EL0},
+  {"pmevtyper9_el0", PMEVTYPER9_EL0},
+  {"pmevtyper10_el0", PMEVTYPER10_EL0},
+  {"pmevtyper11_el0", PMEVTYPER11_EL0},
+  {"pmevtyper12_el0", PMEVTYPER12_EL0},
+  {"pmevtyper13_el0", PMEVTYPER13_EL0},
+  {"pmevtyper14_el0", PMEVTYPER14_EL0},
+  {"pmevtyper15_el0", PMEVTYPER15_EL0},
+  {"pmevtyper16_el0", PMEVTYPER16_EL0},
+  {"pmevtyper17_el0", PMEVTYPER17_EL0},
+  {"pmevtyper18_el0", PMEVTYPER18_EL0},
+  {"pmevtyper19_el0", PMEVTYPER19_EL0},
+  {"pmevtyper20_el0", PMEVTYPER20_EL0},
+  {"pmevtyper21_el0", PMEVTYPER21_EL0},
+  {"pmevtyper22_el0", PMEVTYPER22_EL0},
+  {"pmevtyper23_el0", PMEVTYPER23_EL0},
+  {"pmevtyper24_el0", PMEVTYPER24_EL0},
+  {"pmevtyper25_el0", PMEVTYPER25_EL0},
+  {"pmevtyper26_el0", PMEVTYPER26_EL0},
+  {"pmevtyper27_el0", PMEVTYPER27_EL0},
+  {"pmevtyper28_el0", PMEVTYPER28_EL0},
+  {"pmevtyper29_el0", PMEVTYPER29_EL0},
+  {"pmevtyper30_el0", PMEVTYPER30_EL0},
+
+  // Trace registers
+  {"trcprgctlr", TRCPRGCTLR},
+  {"trcprocselr", TRCPROCSELR},
+  {"trcconfigr", TRCCONFIGR},
+  {"trcauxctlr", TRCAUXCTLR},
+  {"trceventctl0r", TRCEVENTCTL0R},
+  {"trceventctl1r", TRCEVENTCTL1R},
+  {"trcstallctlr", TRCSTALLCTLR},
+  {"trctsctlr", TRCTSCTLR},
+  {"trcsyncpr", TRCSYNCPR},
+  {"trcccctlr", TRCCCCTLR},
+  {"trcbbctlr", TRCBBCTLR},
+  {"trctraceidr", TRCTRACEIDR},
+  {"trcqctlr", TRCQCTLR},
+  {"trcvictlr", TRCVICTLR},
+  {"trcviiectlr", TRCVIIECTLR},
+  {"trcvissctlr", TRCVISSCTLR},
+  {"trcvipcssctlr", TRCVIPCSSCTLR},
+  {"trcvdctlr", TRCVDCTLR},
+  {"trcvdsacctlr", TRCVDSACCTLR},
+  {"trcvdarcctlr", TRCVDARCCTLR},
+  {"trcseqevr0", TRCSEQEVR0},
+  {"trcseqevr1", TRCSEQEVR1},
+  {"trcseqevr2", TRCSEQEVR2},
+  {"trcseqrstevr", TRCSEQRSTEVR},
+  {"trcseqstr", TRCSEQSTR},
+  {"trcextinselr", TRCEXTINSELR},
+  {"trccntrldvr0", TRCCNTRLDVR0},
+  {"trccntrldvr1", TRCCNTRLDVR1},
+  {"trccntrldvr2", TRCCNTRLDVR2},
+  {"trccntrldvr3", TRCCNTRLDVR3},
+  {"trccntctlr0", TRCCNTCTLR0},
+  {"trccntctlr1", TRCCNTCTLR1},
+  {"trccntctlr2", TRCCNTCTLR2},
+  {"trccntctlr3", TRCCNTCTLR3},
+  {"trccntvr0", TRCCNTVR0},
+  {"trccntvr1", TRCCNTVR1},
+  {"trccntvr2", TRCCNTVR2},
+  {"trccntvr3", TRCCNTVR3},
+  {"trcimspec0", TRCIMSPEC0},
+  {"trcimspec1", TRCIMSPEC1},
+  {"trcimspec2", TRCIMSPEC2},
+  {"trcimspec3", TRCIMSPEC3},
+  {"trcimspec4", TRCIMSPEC4},
+  {"trcimspec5", TRCIMSPEC5},
+  {"trcimspec6", TRCIMSPEC6},
+  {"trcimspec7", TRCIMSPEC7},
+  {"trcrsctlr2", TRCRSCTLR2},
+  {"trcrsctlr3", TRCRSCTLR3},
+  {"trcrsctlr4", TRCRSCTLR4},
+  {"trcrsctlr5", TRCRSCTLR5},
+  {"trcrsctlr6", TRCRSCTLR6},
+  {"trcrsctlr7", TRCRSCTLR7},
+  {"trcrsctlr8", TRCRSCTLR8},
+  {"trcrsctlr9", TRCRSCTLR9},
+  {"trcrsctlr10", TRCRSCTLR10},
+  {"trcrsctlr11", TRCRSCTLR11},
+  {"trcrsctlr12", TRCRSCTLR12},
+  {"trcrsctlr13", TRCRSCTLR13},
+  {"trcrsctlr14", TRCRSCTLR14},
+  {"trcrsctlr15", TRCRSCTLR15},
+  {"trcrsctlr16", TRCRSCTLR16},
+  {"trcrsctlr17", TRCRSCTLR17},
+  {"trcrsctlr18", TRCRSCTLR18},
+  {"trcrsctlr19", TRCRSCTLR19},
+  {"trcrsctlr20", TRCRSCTLR20},
+  {"trcrsctlr21", TRCRSCTLR21},
+  {"trcrsctlr22", TRCRSCTLR22},
+  {"trcrsctlr23", TRCRSCTLR23},
+  {"trcrsctlr24", TRCRSCTLR24},
+  {"trcrsctlr25", TRCRSCTLR25},
+  {"trcrsctlr26", TRCRSCTLR26},
+  {"trcrsctlr27", TRCRSCTLR27},
+  {"trcrsctlr28", TRCRSCTLR28},
+  {"trcrsctlr29", TRCRSCTLR29},
+  {"trcrsctlr30", TRCRSCTLR30},
+  {"trcrsctlr31", TRCRSCTLR31},
+  {"trcssccr0", TRCSSCCR0},
+  {"trcssccr1", TRCSSCCR1},
+  {"trcssccr2", TRCSSCCR2},
+  {"trcssccr3", TRCSSCCR3},
+  {"trcssccr4", TRCSSCCR4},
+  {"trcssccr5", TRCSSCCR5},
+  {"trcssccr6", TRCSSCCR6},
+  {"trcssccr7", TRCSSCCR7},
+  {"trcsscsr0", TRCSSCSR0},
+  {"trcsscsr1", TRCSSCSR1},
+  {"trcsscsr2", TRCSSCSR2},
+  {"trcsscsr3", TRCSSCSR3},
+  {"trcsscsr4", TRCSSCSR4},
+  {"trcsscsr5", TRCSSCSR5},
+  {"trcsscsr6", TRCSSCSR6},
+  {"trcsscsr7", TRCSSCSR7},
+  {"trcsspcicr0", TRCSSPCICR0},
+  {"trcsspcicr1", TRCSSPCICR1},
+  {"trcsspcicr2", TRCSSPCICR2},
+  {"trcsspcicr3", TRCSSPCICR3},
+  {"trcsspcicr4", TRCSSPCICR4},
+  {"trcsspcicr5", TRCSSPCICR5},
+  {"trcsspcicr6", TRCSSPCICR6},
+  {"trcsspcicr7", TRCSSPCICR7},
+  {"trcpdcr", TRCPDCR},
+  {"trcacvr0", TRCACVR0},
+  {"trcacvr1", TRCACVR1},
+  {"trcacvr2", TRCACVR2},
+  {"trcacvr3", TRCACVR3},
+  {"trcacvr4", TRCACVR4},
+  {"trcacvr5", TRCACVR5},
+  {"trcacvr6", TRCACVR6},
+  {"trcacvr7", TRCACVR7},
+  {"trcacvr8", TRCACVR8},
+  {"trcacvr9", TRCACVR9},
+  {"trcacvr10", TRCACVR10},
+  {"trcacvr11", TRCACVR11},
+  {"trcacvr12", TRCACVR12},
+  {"trcacvr13", TRCACVR13},
+  {"trcacvr14", TRCACVR14},
+  {"trcacvr15", TRCACVR15},
+  {"trcacatr0", TRCACATR0},
+  {"trcacatr1", TRCACATR1},
+  {"trcacatr2", TRCACATR2},
+  {"trcacatr3", TRCACATR3},
+  {"trcacatr4", TRCACATR4},
+  {"trcacatr5", TRCACATR5},
+  {"trcacatr6", TRCACATR6},
+  {"trcacatr7", TRCACATR7},
+  {"trcacatr8", TRCACATR8},
+  {"trcacatr9", TRCACATR9},
+  {"trcacatr10", TRCACATR10},
+  {"trcacatr11", TRCACATR11},
+  {"trcacatr12", TRCACATR12},
+  {"trcacatr13", TRCACATR13},
+  {"trcacatr14", TRCACATR14},
+  {"trcacatr15", TRCACATR15},
+  {"trcdvcvr0", TRCDVCVR0},
+  {"trcdvcvr1", TRCDVCVR1},
+  {"trcdvcvr2", TRCDVCVR2},
+  {"trcdvcvr3", TRCDVCVR3},
+  {"trcdvcvr4", TRCDVCVR4},
+  {"trcdvcvr5", TRCDVCVR5},
+  {"trcdvcvr6", TRCDVCVR6},
+  {"trcdvcvr7", TRCDVCVR7},
+  {"trcdvcmr0", TRCDVCMR0},
+  {"trcdvcmr1", TRCDVCMR1},
+  {"trcdvcmr2", TRCDVCMR2},
+  {"trcdvcmr3", TRCDVCMR3},
+  {"trcdvcmr4", TRCDVCMR4},
+  {"trcdvcmr5", TRCDVCMR5},
+  {"trcdvcmr6", TRCDVCMR6},
+  {"trcdvcmr7", TRCDVCMR7},
+  {"trccidcvr0", TRCCIDCVR0},
+  {"trccidcvr1", TRCCIDCVR1},
+  {"trccidcvr2", TRCCIDCVR2},
+  {"trccidcvr3", TRCCIDCVR3},
+  {"trccidcvr4", TRCCIDCVR4},
+  {"trccidcvr5", TRCCIDCVR5},
+  {"trccidcvr6", TRCCIDCVR6},
+  {"trccidcvr7", TRCCIDCVR7},
+  {"trcvmidcvr0", TRCVMIDCVR0},
+  {"trcvmidcvr1", TRCVMIDCVR1},
+  {"trcvmidcvr2", TRCVMIDCVR2},
+  {"trcvmidcvr3", TRCVMIDCVR3},
+  {"trcvmidcvr4", TRCVMIDCVR4},
+  {"trcvmidcvr5", TRCVMIDCVR5},
+  {"trcvmidcvr6", TRCVMIDCVR6},
+  {"trcvmidcvr7", TRCVMIDCVR7},
+  {"trccidcctlr0", TRCCIDCCTLR0},
+  {"trccidcctlr1", TRCCIDCCTLR1},
+  {"trcvmidcctlr0", TRCVMIDCCTLR0},
+  {"trcvmidcctlr1", TRCVMIDCCTLR1},
+  {"trcitctrl", TRCITCTRL},
+  {"trcclaimset", TRCCLAIMSET},
+  {"trcclaimclr", TRCCLAIMCLR},
+
+  // GICv3 registers
+  {"icc_bpr1_el1", ICC_BPR1_EL1},
+  {"icc_bpr0_el1", ICC_BPR0_EL1},
+  {"icc_pmr_el1", ICC_PMR_EL1},
+  {"icc_ctlr_el1", ICC_CTLR_EL1},
+  {"icc_ctlr_el3", ICC_CTLR_EL3},
+  {"icc_sre_el1", ICC_SRE_EL1},
+  {"icc_sre_el2", ICC_SRE_EL2},
+  {"icc_sre_el3", ICC_SRE_EL3},
+  {"icc_igrpen0_el1", ICC_IGRPEN0_EL1},
+  {"icc_igrpen1_el1", ICC_IGRPEN1_EL1},
+  {"icc_igrpen1_el3", ICC_IGRPEN1_EL3},
+  {"icc_seien_el1", ICC_SEIEN_EL1},
+  {"icc_ap0r0_el1", ICC_AP0R0_EL1},
+  {"icc_ap0r1_el1", ICC_AP0R1_EL1},
+  {"icc_ap0r2_el1", ICC_AP0R2_EL1},
+  {"icc_ap0r3_el1", ICC_AP0R3_EL1},
+  {"icc_ap1r0_el1", ICC_AP1R0_EL1},
+  {"icc_ap1r1_el1", ICC_AP1R1_EL1},
+  {"icc_ap1r2_el1", ICC_AP1R2_EL1},
+  {"icc_ap1r3_el1", ICC_AP1R3_EL1},
+  {"ich_ap0r0_el2", ICH_AP0R0_EL2},
+  {"ich_ap0r1_el2", ICH_AP0R1_EL2},
+  {"ich_ap0r2_el2", ICH_AP0R2_EL2},
+  {"ich_ap0r3_el2", ICH_AP0R3_EL2},
+  {"ich_ap1r0_el2", ICH_AP1R0_EL2},
+  {"ich_ap1r1_el2", ICH_AP1R1_EL2},
+  {"ich_ap1r2_el2", ICH_AP1R2_EL2},
+  {"ich_ap1r3_el2", ICH_AP1R3_EL2},
+  {"ich_hcr_el2", ICH_HCR_EL2},
+  {"ich_misr_el2", ICH_MISR_EL2},
+  {"ich_vmcr_el2", ICH_VMCR_EL2},
+  {"ich_vseir_el2", ICH_VSEIR_EL2},
+  {"ich_lr0_el2", ICH_LR0_EL2},
+  {"ich_lr1_el2", ICH_LR1_EL2},
+  {"ich_lr2_el2", ICH_LR2_EL2},
+  {"ich_lr3_el2", ICH_LR3_EL2},
+  {"ich_lr4_el2", ICH_LR4_EL2},
+  {"ich_lr5_el2", ICH_LR5_EL2},
+  {"ich_lr6_el2", ICH_LR6_EL2},
+  {"ich_lr7_el2", ICH_LR7_EL2},
+  {"ich_lr8_el2", ICH_LR8_EL2},
+  {"ich_lr9_el2", ICH_LR9_EL2},
+  {"ich_lr10_el2", ICH_LR10_EL2},
+  {"ich_lr11_el2", ICH_LR11_EL2},
+  {"ich_lr12_el2", ICH_LR12_EL2},
+  {"ich_lr13_el2", ICH_LR13_EL2},
+  {"ich_lr14_el2", ICH_LR14_EL2},
+  {"ich_lr15_el2", ICH_LR15_EL2}
+};
+
+uint32_t
+A64SysReg::SysRegMapper::fromString(StringRef Name, bool &Valid) const {
+  // First search the registers shared by all
+  std::string NameLower = Name.lower();
+  for (unsigned i = 0; i < array_lengthof(SysRegPairs); ++i) {
+    if (SysRegPairs[i].Name == NameLower) {
+      Valid = true;
+      return SysRegPairs[i].Value;
+    }
+  }
+
+  // Now try the instruction-specific registers (either read-only or
+  // write-only).
+  for (unsigned i = 0; i < NumInstPairs; ++i) {
+    if (InstPairs[i].Name == NameLower) {
+      Valid = true;
+      return InstPairs[i].Value;
+    }
+  }
+
+  // Try to parse an S<op0>_<op1>_<Cn>_<Cm>_<op2> register name, where the bits
+  // are: 11 xxx 1x11 xxxx xxx
+  Regex GenericRegPattern("^s3_([0-7])_c(1[15])_c([0-9]|1[0-5])_([0-7])$");
+
+  SmallVector<StringRef, 4> Ops;
+  if (!GenericRegPattern.match(NameLower, &Ops)) {
+    Valid = false;
+    return -1;
+  }
+
+  uint32_t Op0 = 3, Op1 = 0, CRn = 0, CRm = 0, Op2 = 0;
+  uint32_t Bits;
+  Ops[1].getAsInteger(10, Op1);
+  Ops[2].getAsInteger(10, CRn);
+  Ops[3].getAsInteger(10, CRm);
+  Ops[4].getAsInteger(10, Op2);
+  Bits = (Op0 << 14) | (Op1 << 11) | (CRn << 7) | (CRm << 3) | Op2;
+
+  Valid = true;
+  return Bits;
+}
+
+std::string
+A64SysReg::SysRegMapper::toString(uint32_t Bits, bool &Valid) const {
+  for (unsigned i = 0; i < array_lengthof(SysRegPairs); ++i) {
+    if (SysRegPairs[i].Value == Bits) {
+      Valid = true;
+      return SysRegPairs[i].Name;
+    }
+  }
+
+  for (unsigned i = 0; i < NumInstPairs; ++i) {
+    if (InstPairs[i].Value == Bits) {
+      Valid = true;
+      return InstPairs[i].Name;
+    }
+  }
+
+  uint32_t Op0 = (Bits >> 14) & 0x3;
+  uint32_t Op1 = (Bits >> 11) & 0x7;
+  uint32_t CRn = (Bits >> 7) & 0xf;
+  uint32_t CRm = (Bits >> 3) & 0xf;
+  uint32_t Op2 = Bits & 0x7;
+
+  // Only combinations matching: 11 xxx 1x11 xxxx xxx are valid for a generic
+  // name.
+  if (Op0 != 3 || (CRn != 11 && CRn != 15)) {
+      Valid = false;
+      return "";
+  }
+
+  assert(Op0 == 3 && (CRn == 11 || CRn == 15) && "Invalid generic sysreg");
+
+  Valid = true;
+  return "s3_" + utostr(Op1) + "_c" + utostr(CRn)
+               + "_c" + utostr(CRm) + "_" + utostr(Op2);
+}
+
+const NamedImmMapper::Mapping A64TLBI::TLBIMapper::TLBIPairs[] = {
+  {"ipas2e1is", IPAS2E1IS},
+  {"ipas2le1is", IPAS2LE1IS},
+  {"vmalle1is", VMALLE1IS},
+  {"alle2is", ALLE2IS},
+  {"alle3is", ALLE3IS},
+  {"vae1is", VAE1IS},
+  {"vae2is", VAE2IS},
+  {"vae3is", VAE3IS},
+  {"aside1is", ASIDE1IS},
+  {"vaae1is", VAAE1IS},
+  {"alle1is", ALLE1IS},
+  {"vale1is", VALE1IS},
+  {"vale2is", VALE2IS},
+  {"vale3is", VALE3IS},
+  {"vmalls12e1is", VMALLS12E1IS},
+  {"vaale1is", VAALE1IS},
+  {"ipas2e1", IPAS2E1},
+  {"ipas2le1", IPAS2LE1},
+  {"vmalle1", VMALLE1},
+  {"alle2", ALLE2},
+  {"alle3", ALLE3},
+  {"vae1", VAE1},
+  {"vae2", VAE2},
+  {"vae3", VAE3},
+  {"aside1", ASIDE1},
+  {"vaae1", VAAE1},
+  {"alle1", ALLE1},
+  {"vale1", VALE1},
+  {"vale2", VALE2},
+  {"vale3", VALE3},
+  {"vmalls12e1", VMALLS12E1},
+  {"vaale1", VAALE1}
+};
+
+A64TLBI::TLBIMapper::TLBIMapper()
+  : NamedImmMapper(TLBIPairs, 0) {}
+
+bool A64Imms::isFPImm(const APFloat &Val, uint32_t &Imm8Bits) {
+  const fltSemantics &Sem = Val.getSemantics();
+  unsigned FracBits = APFloat::semanticsPrecision(Sem) - 1;
+
+  uint32_t ExpMask;
+  switch (FracBits) {
+  case 10: // IEEE half-precision
+    ExpMask = 0x1f;
+    break;
+  case 23: // IEEE single-precision
+    ExpMask = 0xff;
+    break;
+  case 52: // IEEE double-precision
+    ExpMask = 0x7ff;
+    break;
+  case 112: // IEEE quad-precision
+    // No immediates are valid for double precision.
+    return false;
+  default:
+    llvm_unreachable("Only half, single and double precision supported");
+  }
+
+  uint32_t ExpStart = FracBits;
+  uint64_t FracMask = (1ULL << FracBits) - 1;
+
+  uint32_t Sign = Val.isNegative();
+
+  uint64_t Bits= Val.bitcastToAPInt().getLimitedValue();
+  uint64_t Fraction = Bits & FracMask;
+  int32_t Exponent = ((Bits >> ExpStart) & ExpMask);
+  Exponent -= ExpMask >> 1;
+
+  // S[d] = imm8<7>:NOT(imm8<6>):Replicate(imm8<6>, 5):imm8<5:0>:Zeros(19)
+  // D[d] = imm8<7>:NOT(imm8<6>):Replicate(imm8<6>, 8):imm8<5:0>:Zeros(48)
+  // This translates to: only 4 bits of fraction; -3 <= exp <= 4.
+  uint64_t A64FracStart = FracBits - 4;
+  uint64_t A64FracMask = 0xf;
+
+  // Are there too many fraction bits?
+  if (Fraction & ~(A64FracMask << A64FracStart))
+    return false;
+
+  if (Exponent < -3 || Exponent > 4)
+    return false;
+
+  uint32_t PackedFraction = (Fraction >> A64FracStart) & A64FracMask;
+  uint32_t PackedExp = (Exponent + 7) & 0x7;
+
+  Imm8Bits = (Sign << 7) | (PackedExp << 4) | PackedFraction;
+  return true;
+}
+
+// Encoding of the immediate for logical (immediate) instructions:
+//
+// | N | imms   | immr   | size | R            | S            |
+// |---+--------+--------+------+--------------+--------------|
+// | 1 | ssssss | rrrrrr |   64 | UInt(rrrrrr) | UInt(ssssss) |
+// | 0 | 0sssss | xrrrrr |   32 | UInt(rrrrr)  | UInt(sssss)  |
+// | 0 | 10ssss | xxrrrr |   16 | UInt(rrrr)   | UInt(ssss)   |
+// | 0 | 110sss | xxxrrr |    8 | UInt(rrr)    | UInt(sss)    |
+// | 0 | 1110ss | xxxxrr |    4 | UInt(rr)     | UInt(ss)     |
+// | 0 | 11110s | xxxxxr |    2 | UInt(r)      | UInt(s)      |
+// | 0 | 11111x | -      |      | UNALLOCATED  |              |
+//
+// Columns 'R', 'S' and 'size' specify a "bitmask immediate" of size bits in
+// which the lower S+1 bits are ones and the remaining bits are zero, then
+// rotated right by R bits, which is then replicated across the datapath.
+//
+// + Values of 'N', 'imms' and 'immr' which do not match the above table are
+//   RESERVED.
+// + If all 's' bits in the imms field are set then the instruction is
+//   RESERVED.
+// + The 'x' bits in the 'immr' field are IGNORED.
+
+bool A64Imms::isLogicalImm(unsigned RegWidth, uint64_t Imm, uint32_t &Bits) {
+  int RepeatWidth;
+  int Rotation = 0;
+  int Num1s = 0;
+
+  // Because there are S+1 ones in the replicated mask, an immediate of all
+  // zeros is not allowed. Filtering it here is probably more efficient.
+  if (Imm == 0) return false;
+
+  for (RepeatWidth = RegWidth; RepeatWidth > 1; RepeatWidth /= 2) {
+    uint64_t RepeatMask = RepeatWidth == 64 ? -1 : (1ULL << RepeatWidth) - 1;
+    uint64_t ReplicatedMask = Imm & RepeatMask;
+
+    if (ReplicatedMask == 0) continue;
+
+    // First we have to make sure the mask is actually repeated in each slot for
+    // this width-specifier.
+    bool IsReplicatedMask = true;
+    for (unsigned i = RepeatWidth; i < RegWidth; i += RepeatWidth) {
+      if (((Imm >> i) & RepeatMask) != ReplicatedMask) {
+        IsReplicatedMask = false;
+        break;
+      }
+    }
+    if (!IsReplicatedMask) continue;
+
+    // Now we have to work out the amount of rotation needed. The first part of
+    // this calculation is actually independent of RepeatWidth, but the complex
+    // case will depend on it.
+    Rotation = CountTrailingZeros_64(Imm);
+    if (Rotation == 0) {
+      // There were no leading zeros, which means it's either in place or there
+      // are 1s at each end (e.g. 0x8003 needs rotating).
+      Rotation = RegWidth == 64 ? CountLeadingOnes_64(Imm)
+                                : CountLeadingOnes_32(Imm);
+      Rotation = RepeatWidth - Rotation;
+    }
+
+    uint64_t ReplicatedOnes = (ReplicatedMask >> Rotation)
+      | ((ReplicatedMask << (RepeatWidth - Rotation)) & RepeatMask);
+    // Of course, they may not actually be ones, so we have to check that:
+    if (!isMask_64(ReplicatedOnes))
+      continue;
+
+    Num1s = CountTrailingOnes_64(ReplicatedOnes);
+
+    // We know we've got an almost valid encoding (certainly, if this is invalid
+    // no other parameters would work).
+    break;
+  }
+
+  // The encodings which would produce all 1s are RESERVED.
+  if (RepeatWidth == 1 || Num1s == RepeatWidth) return false;
+
+  uint32_t N = RepeatWidth == 64;
+  uint32_t ImmR = RepeatWidth - Rotation;
+  uint32_t ImmS = Num1s - 1;
+
+  switch (RepeatWidth) {
+  default: break; // No action required for other valid rotations.
+  case 16: ImmS |= 0x20; break; // 10ssss
+  case 8: ImmS |= 0x30; break;  // 110sss
+  case 4: ImmS |= 0x38; break;  // 1110ss
+  case 2: ImmS |= 0x3c; break;  // 11110s
+  }
+
+  Bits = ImmS | (ImmR << 6) | (N << 12);
+
+  return true;
+}
+
+
+bool A64Imms::isLogicalImmBits(unsigned RegWidth, uint32_t Bits,
+                               uint64_t &Imm) {
+  uint32_t N = Bits >> 12;
+  uint32_t ImmR = (Bits >> 6) & 0x3f;
+  uint32_t ImmS = Bits & 0x3f;
+
+  // N=1 encodes a 64-bit replication and is invalid for the 32-bit
+  // instructions.
+  if (RegWidth == 32 && N != 0) return false;
+
+  int Width = 0;
+  if (N == 1)
+    Width = 64;
+  else if ((ImmS & 0x20) == 0)
+    Width = 32;
+  else if ((ImmS & 0x10) == 0)
+    Width = 16;
+  else if ((ImmS & 0x08) == 0)
+    Width = 8;
+  else if ((ImmS & 0x04) == 0)
+    Width = 4;
+  else if ((ImmS & 0x02) == 0)
+    Width = 2;
+  else {
+    // ImmS  is 0b11111x: UNALLOCATED
+    return false;
+  }
+
+  int Num1s = (ImmS & (Width - 1)) + 1;
+
+  // All encodings which would map to -1 (signed) are RESERVED.
+  if (Num1s == Width) return false;
+
+  int Rotation = (ImmR & (Width - 1));
+  uint64_t Mask = (1ULL << Num1s) - 1;
+  uint64_t WidthMask = Width == 64 ? -1 : (1ULL << Width) - 1;
+  Mask = (Mask >> Rotation)
+    | ((Mask << (Width - Rotation)) & WidthMask);
+
+  Imm = 0;
+  for (unsigned i = 0; i < RegWidth / Width; ++i) {
+    Imm |= Mask;
+    Mask <<= Width;
+  }
+
+  return true;
+}
+
+bool A64Imms::isMOVZImm(int RegWidth, uint64_t Value, int &UImm16, int &Shift) {
+  // If high bits are set then a 32-bit MOVZ can't possibly work.
+  if (RegWidth == 32 && (Value & ~0xffffffffULL))
+    return false;
+
+  for (int i = 0; i < RegWidth; i += 16) {
+    // If the value is 0 when we mask out all the bits that could be set with
+    // the current LSL value then it's representable.
+    if ((Value & ~(0xffffULL << i)) == 0) {
+      Shift = i / 16;
+      UImm16 = (Value >> i) & 0xffff;
+      return true;
+    }
+  }
+  return false;
+}
+
+bool A64Imms::isMOVNImm(int RegWidth, uint64_t Value, int &UImm16, int &Shift) {
+  // MOVN is defined to set its register to NOT(LSL(imm16, shift)).
+
+  // We have to be a little careful about a 32-bit register: 0xffff_1234 *is*
+  // representable, but ~0xffff_1234 == 0xffff_ffff_0000_edcb which is not
+  // a valid input for isMOVZImm.
+  if (RegWidth == 32 && (Value & ~0xffffffffULL))
+    return false;
+
+  uint64_t MOVZEquivalent = RegWidth == 32 ? ~Value & 0xffffffff : ~Value;
+
+  return isMOVZImm(RegWidth, MOVZEquivalent, UImm16, Shift);
+}
+
+bool A64Imms::isOnlyMOVNImm(int RegWidth, uint64_t Value,
+                            int &UImm16, int &Shift) {
+  if (isMOVZImm(RegWidth, Value, UImm16, Shift))
+    return false;
+
+  return isMOVNImm(RegWidth, Value, UImm16, Shift);
+}
diff --git a/lib/Target/AArch64/Utils/AArch64BaseInfo.h b/lib/Target/AArch64/Utils/AArch64BaseInfo.h
new file mode 100644
index 000000000000..1b773d632ebe
--- /dev/null
+++ b/lib/Target/AArch64/Utils/AArch64BaseInfo.h
@@ -0,0 +1,1068 @@
+//===-- AArch64BaseInfo.h - Top level definitions for AArch64- --*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains small standalone helper functions and enum definitions for
+// the AArch64 target useful for the compiler back-end and the MC libraries.
+// As such, it deliberately does not include references to LLVM core
+// code gen types, passes, etc..
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_AARCH64_BASEINFO_H
+#define LLVM_AARCH64_BASEINFO_H
+
+#include "llvm/ADT/StringSwitch.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/Support/ErrorHandling.h"
+
+namespace llvm {
+
+// // Enums corresponding to AArch64 condition codes
+namespace A64CC {
+  // The CondCodes constants map directly to the 4-bit encoding of the
+  // condition field for predicated instructions.
+  enum CondCodes {   // Meaning (integer)          Meaning (floating-point)
+    EQ = 0,        // Equal                      Equal
+    NE,            // Not equal                  Not equal, or unordered
+    HS,            // Unsigned higher or same    >, ==, or unordered
+    LO,            // Unsigned lower or same     Less than
+    MI,            // Minus, negative            Less than
+    PL,            // Plus, positive or zero     >, ==, or unordered
+    VS,            // Overflow                   Unordered
+    VC,            // No overflow                Ordered
+    HI,            // Unsigned higher            Greater than, or unordered
+    LS,            // Unsigned lower or same     Less than or equal
+    GE,            // Greater than or equal      Greater than or equal
+    LT,            // Less than                  Less than, or unordered
+    GT,            // Signed greater than        Greater than
+    LE,            // Signed less than or equal  <, ==, or unordered
+    AL,            // Always (unconditional)     Always (unconditional)
+    NV,             // Always (unconditional)     Always (unconditional)
+    // Note the NV exists purely to disassemble 0b1111. Execution
+    // is "always".
+    Invalid
+  };
+
+} // namespace A64CC
+
+inline static const char *A64CondCodeToString(A64CC::CondCodes CC) {
+  switch (CC) {
+  default: llvm_unreachable("Unknown condition code");
+  case A64CC::EQ:  return "eq";
+  case A64CC::NE:  return "ne";
+  case A64CC::HS:  return "hs";
+  case A64CC::LO:  return "lo";
+  case A64CC::MI:  return "mi";
+  case A64CC::PL:  return "pl";
+  case A64CC::VS:  return "vs";
+  case A64CC::VC:  return "vc";
+  case A64CC::HI:  return "hi";
+  case A64CC::LS:  return "ls";
+  case A64CC::GE:  return "ge";
+  case A64CC::LT:  return "lt";
+  case A64CC::GT:  return "gt";
+  case A64CC::LE:  return "le";
+  case A64CC::AL:  return "al";
+  case A64CC::NV:  return "nv";
+  }
+}
+
+inline static A64CC::CondCodes A64StringToCondCode(StringRef CondStr) {
+  return StringSwitch<A64CC::CondCodes>(CondStr.lower())
+             .Case("eq", A64CC::EQ)
+             .Case("ne", A64CC::NE)
+             .Case("ne", A64CC::NE)
+             .Case("hs", A64CC::HS)
+             .Case("cs", A64CC::HS)
+             .Case("lo", A64CC::LO)
+             .Case("cc", A64CC::LO)
+             .Case("mi", A64CC::MI)
+             .Case("pl", A64CC::PL)
+             .Case("vs", A64CC::VS)
+             .Case("vc", A64CC::VC)
+             .Case("hi", A64CC::HI)
+             .Case("ls", A64CC::LS)
+             .Case("ge", A64CC::GE)
+             .Case("lt", A64CC::LT)
+             .Case("gt", A64CC::GT)
+             .Case("le", A64CC::LE)
+             .Case("al", A64CC::AL)
+             .Case("nv", A64CC::NV)
+             .Default(A64CC::Invalid);
+}
+
+inline static A64CC::CondCodes A64InvertCondCode(A64CC::CondCodes CC) {
+  // It turns out that the condition codes have been designed so that in order
+  // to reverse the intent of the condition you only have to invert the low bit:
+
+  return static_cast<A64CC::CondCodes>(static_cast<unsigned>(CC) ^ 0x1);
+}
+
+/// Instances of this class can perform bidirectional mapping from random
+/// identifier strings to operand encodings. For example "MSR" takes a named
+/// system-register which must be encoded somehow and decoded for printing. This
+/// central location means that the information for those transformations is not
+/// duplicated and remains in sync.
+///
+/// FIXME: currently the algorithm is a completely unoptimised linear
+/// search. Obviously this could be improved, but we would probably want to work
+/// out just how often these instructions are emitted before working on it. It
+/// might even be optimal to just reorder the tables for the common instructions
+/// rather than changing the algorithm.
+struct NamedImmMapper {
+  struct Mapping {
+    const char *Name;
+    uint32_t Value;
+  };
+
+  template<int N>
+  NamedImmMapper(const Mapping (&Pairs)[N], uint32_t TooBigImm)
+    : Pairs(&Pairs[0]), NumPairs(N), TooBigImm(TooBigImm) {}
+
+  StringRef toString(uint32_t Value, bool &Valid) const;
+  uint32_t fromString(StringRef Name, bool &Valid) const;
+
+  /// Many of the instructions allow an alternative assembly form consisting of
+  /// a simple immediate. Currently the only valid forms are ranges [0, N) where
+  /// N being 0 indicates no immediate syntax-form is allowed.
+  bool validImm(uint32_t Value) const;
+protected:
+  const Mapping *Pairs;
+  size_t NumPairs;
+  uint32_t TooBigImm;
+};
+
+namespace A64AT {
+  enum ATValues {
+    Invalid = -1,    // Op0 Op1  CRn   CRm   Op2
+    S1E1R = 0x43c0,  // 01  000  0111  1000  000
+    S1E2R = 0x63c0,  // 01  100  0111  1000  000
+    S1E3R = 0x73c0,  // 01  110  0111  1000  000
+    S1E1W = 0x43c1,  // 01  000  0111  1000  001
+    S1E2W = 0x63c1,  // 01  100  0111  1000  001
+    S1E3W = 0x73c1,  // 01  110  0111  1000  001
+    S1E0R = 0x43c2,  // 01  000  0111  1000  010
+    S1E0W = 0x43c3,  // 01  000  0111  1000  011
+    S12E1R = 0x63c4, // 01  100  0111  1000  100
+    S12E1W = 0x63c5, // 01  100  0111  1000  101
+    S12E0R = 0x63c6, // 01  100  0111  1000  110
+    S12E0W = 0x63c7  // 01  100  0111  1000  111
+  };
+
+  struct ATMapper : NamedImmMapper {
+    const static Mapping ATPairs[];
+
+    ATMapper();
+  };
+
+}
+namespace A64DB {
+  enum DBValues {
+    Invalid = -1,
+    OSHLD = 0x1,
+    OSHST = 0x2,
+    OSH =   0x3,
+    NSHLD = 0x5,
+    NSHST = 0x6,
+    NSH =   0x7,
+    ISHLD = 0x9,
+    ISHST = 0xa,
+    ISH =   0xb,
+    LD =    0xd,
+    ST =    0xe,
+    SY =    0xf
+  };
+
+  struct DBarrierMapper : NamedImmMapper {
+    const static Mapping DBarrierPairs[];
+
+    DBarrierMapper();
+  };
+}
+
+namespace  A64DC {
+  enum DCValues {
+    Invalid = -1,   // Op1  CRn   CRm   Op2
+    ZVA   = 0x5ba1, // 01  011  0111  0100  001
+    IVAC  = 0x43b1, // 01  000  0111  0110  001
+    ISW   = 0x43b2, // 01  000  0111  0110  010
+    CVAC  = 0x5bd1, // 01  011  0111  1010  001
+    CSW   = 0x43d2, // 01  000  0111  1010  010
+    CVAU  = 0x5bd9, // 01  011  0111  1011  001
+    CIVAC = 0x5bf1, // 01  011  0111  1110  001
+    CISW  = 0x43f2  // 01  000  0111  1110  010
+  };
+
+  struct DCMapper : NamedImmMapper {
+    const static Mapping DCPairs[];
+
+    DCMapper();
+  };
+
+}
+
+namespace  A64IC {
+  enum ICValues {
+    Invalid = -1,     // Op1  CRn   CRm   Op2
+    IALLUIS = 0x0388, // 000  0111  0001  000
+    IALLU = 0x03a8,   // 000  0111  0101  000
+    IVAU = 0x1ba9     // 011  0111  0101  001
+  };
+
+
+  struct ICMapper : NamedImmMapper {
+    const static Mapping ICPairs[];
+
+    ICMapper();
+  };
+
+  static inline bool NeedsRegister(ICValues Val) {
+    return Val == IVAU;
+  }
+}
+
+namespace  A64ISB {
+  enum ISBValues {
+    Invalid = -1,
+    SY = 0xf
+  };
+  struct ISBMapper : NamedImmMapper {
+    const static Mapping ISBPairs[];
+
+    ISBMapper();
+  };
+}
+
+namespace A64PRFM {
+  enum PRFMValues {
+    Invalid = -1,
+    PLDL1KEEP = 0x00,
+    PLDL1STRM = 0x01,
+    PLDL2KEEP = 0x02,
+    PLDL2STRM = 0x03,
+    PLDL3KEEP = 0x04,
+    PLDL3STRM = 0x05,
+    PLIL1KEEP = 0x08,
+    PLIL1STRM = 0x09,
+    PLIL2KEEP = 0x0a,
+    PLIL2STRM = 0x0b,
+    PLIL3KEEP = 0x0c,
+    PLIL3STRM = 0x0d,
+    PSTL1KEEP = 0x10,
+    PSTL1STRM = 0x11,
+    PSTL2KEEP = 0x12,
+    PSTL2STRM = 0x13,
+    PSTL3KEEP = 0x14,
+    PSTL3STRM = 0x15
+  };
+
+  struct PRFMMapper : NamedImmMapper {
+    const static Mapping PRFMPairs[];
+
+    PRFMMapper();
+  };
+}
+
+namespace A64PState {
+  enum PStateValues {
+    Invalid = -1,
+    SPSel = 0x05,
+    DAIFSet = 0x1e,
+    DAIFClr = 0x1f
+  };
+
+  struct PStateMapper : NamedImmMapper {
+    const static Mapping PStatePairs[];
+
+    PStateMapper();
+  };
+
+}
+
+namespace A64SE {
+    enum ShiftExtSpecifiers {
+        Invalid = -1,
+        LSL,
+        LSR,
+        ASR,
+        ROR,
+
+        UXTB,
+        UXTH,
+        UXTW,
+        UXTX,
+
+        SXTB,
+        SXTH,
+        SXTW,
+        SXTX
+    };
+}
+
+namespace A64SysReg {
+  enum SysRegROValues {
+    MDCCSR_EL0        = 0x9808, // 10  011  0000  0001  000
+    DBGDTRRX_EL0      = 0x9828, // 10  011  0000  0101  000
+    MDRAR_EL1         = 0x8080, // 10  000  0001  0000  000
+    OSLSR_EL1         = 0x808c, // 10  000  0001  0001  100
+    DBGAUTHSTATUS_EL1 = 0x83f6, // 10  000  0111  1110  110
+    PMCEID0_EL0       = 0xdce6, // 11  011  1001  1100  110
+    PMCEID1_EL0       = 0xdce7, // 11  011  1001  1100  111
+    MIDR_EL1          = 0xc000, // 11  000  0000  0000  000
+    CCSIDR_EL1        = 0xc800, // 11  001  0000  0000  000
+    CLIDR_EL1         = 0xc801, // 11  001  0000  0000  001
+    CTR_EL0           = 0xd801, // 11  011  0000  0000  001
+    MPIDR_EL1         = 0xc005, // 11  000  0000  0000  101
+    REVIDR_EL1        = 0xc006, // 11  000  0000  0000  110
+    AIDR_EL1          = 0xc807, // 11  001  0000  0000  111
+    DCZID_EL0         = 0xd807, // 11  011  0000  0000  111
+    ID_PFR0_EL1       = 0xc008, // 11  000  0000  0001  000
+    ID_PFR1_EL1       = 0xc009, // 11  000  0000  0001  001
+    ID_DFR0_EL1       = 0xc00a, // 11  000  0000  0001  010
+    ID_AFR0_EL1       = 0xc00b, // 11  000  0000  0001  011
+    ID_MMFR0_EL1      = 0xc00c, // 11  000  0000  0001  100
+    ID_MMFR1_EL1      = 0xc00d, // 11  000  0000  0001  101
+    ID_MMFR2_EL1      = 0xc00e, // 11  000  0000  0001  110
+    ID_MMFR3_EL1      = 0xc00f, // 11  000  0000  0001  111
+    ID_ISAR0_EL1      = 0xc010, // 11  000  0000  0010  000
+    ID_ISAR1_EL1      = 0xc011, // 11  000  0000  0010  001
+    ID_ISAR2_EL1      = 0xc012, // 11  000  0000  0010  010
+    ID_ISAR3_EL1      = 0xc013, // 11  000  0000  0010  011
+    ID_ISAR4_EL1      = 0xc014, // 11  000  0000  0010  100
+    ID_ISAR5_EL1      = 0xc015, // 11  000  0000  0010  101
+    ID_AA64PFR0_EL1   = 0xc020, // 11  000  0000  0100  000
+    ID_AA64PFR1_EL1   = 0xc021, // 11  000  0000  0100  001
+    ID_AA64DFR0_EL1   = 0xc028, // 11  000  0000  0101  000
+    ID_AA64DFR1_EL1   = 0xc029, // 11  000  0000  0101  001
+    ID_AA64AFR0_EL1   = 0xc02c, // 11  000  0000  0101  100
+    ID_AA64AFR1_EL1   = 0xc02d, // 11  000  0000  0101  101
+    ID_AA64ISAR0_EL1  = 0xc030, // 11  000  0000  0110  000
+    ID_AA64ISAR1_EL1  = 0xc031, // 11  000  0000  0110  001
+    ID_AA64MMFR0_EL1  = 0xc038, // 11  000  0000  0111  000
+    ID_AA64MMFR1_EL1  = 0xc039, // 11  000  0000  0111  001
+    MVFR0_EL1         = 0xc018, // 11  000  0000  0011  000
+    MVFR1_EL1         = 0xc019, // 11  000  0000  0011  001
+    MVFR2_EL1         = 0xc01a, // 11  000  0000  0011  010
+    RVBAR_EL1         = 0xc601, // 11  000  1100  0000  001
+    RVBAR_EL2         = 0xe601, // 11  100  1100  0000  001
+    RVBAR_EL3         = 0xf601, // 11  110  1100  0000  001
+    ISR_EL1           = 0xc608, // 11  000  1100  0001  000
+    CNTPCT_EL0        = 0xdf01, // 11  011  1110  0000  001
+    CNTVCT_EL0        = 0xdf02,  // 11  011  1110  0000  010
+
+    // Trace registers
+    TRCSTATR          = 0x8818, // 10  001  0000  0011  000
+    TRCIDR8           = 0x8806, // 10  001  0000  0000  110
+    TRCIDR9           = 0x880e, // 10  001  0000  0001  110
+    TRCIDR10          = 0x8816, // 10  001  0000  0010  110
+    TRCIDR11          = 0x881e, // 10  001  0000  0011  110
+    TRCIDR12          = 0x8826, // 10  001  0000  0100  110
+    TRCIDR13          = 0x882e, // 10  001  0000  0101  110
+    TRCIDR0           = 0x8847, // 10  001  0000  1000  111
+    TRCIDR1           = 0x884f, // 10  001  0000  1001  111
+    TRCIDR2           = 0x8857, // 10  001  0000  1010  111
+    TRCIDR3           = 0x885f, // 10  001  0000  1011  111
+    TRCIDR4           = 0x8867, // 10  001  0000  1100  111
+    TRCIDR5           = 0x886f, // 10  001  0000  1101  111
+    TRCIDR6           = 0x8877, // 10  001  0000  1110  111
+    TRCIDR7           = 0x887f, // 10  001  0000  1111  111
+    TRCOSLSR          = 0x888c, // 10  001  0001  0001  100
+    TRCPDSR           = 0x88ac, // 10  001  0001  0101  100
+    TRCDEVAFF0        = 0x8bd6, // 10  001  0111  1010  110
+    TRCDEVAFF1        = 0x8bde, // 10  001  0111  1011  110
+    TRCLSR            = 0x8bee, // 10  001  0111  1101  110
+    TRCAUTHSTATUS     = 0x8bf6, // 10  001  0111  1110  110
+    TRCDEVARCH        = 0x8bfe, // 10  001  0111  1111  110
+    TRCDEVID          = 0x8b97, // 10  001  0111  0010  111
+    TRCDEVTYPE        = 0x8b9f, // 10  001  0111  0011  111
+    TRCPIDR4          = 0x8ba7, // 10  001  0111  0100  111
+    TRCPIDR5          = 0x8baf, // 10  001  0111  0101  111
+    TRCPIDR6          = 0x8bb7, // 10  001  0111  0110  111
+    TRCPIDR7          = 0x8bbf, // 10  001  0111  0111  111
+    TRCPIDR0          = 0x8bc7, // 10  001  0111  1000  111
+    TRCPIDR1          = 0x8bcf, // 10  001  0111  1001  111
+    TRCPIDR2          = 0x8bd7, // 10  001  0111  1010  111
+    TRCPIDR3          = 0x8bdf, // 10  001  0111  1011  111
+    TRCCIDR0          = 0x8be7, // 10  001  0111  1100  111
+    TRCCIDR1          = 0x8bef, // 10  001  0111  1101  111
+    TRCCIDR2          = 0x8bf7, // 10  001  0111  1110  111
+    TRCCIDR3          = 0x8bff, // 10  001  0111  1111  111
+
+    // GICv3 registers
+    ICC_IAR1_EL1      = 0xc660, // 11  000  1100  1100  000
+    ICC_IAR0_EL1      = 0xc640, // 11  000  1100  1000  000
+    ICC_HPPIR1_EL1    = 0xc662, // 11  000  1100  1100  010
+    ICC_HPPIR0_EL1    = 0xc642, // 11  000  1100  1000  010
+    ICC_RPR_EL1       = 0xc65b, // 11  000  1100  1011  011
+    ICH_VTR_EL2       = 0xe659, // 11  100  1100  1011  001
+    ICH_EISR_EL2      = 0xe65b, // 11  100  1100  1011  011
+    ICH_ELSR_EL2      = 0xe65d  // 11  100  1100  1011  101
+  };
+
+  enum SysRegWOValues {
+    DBGDTRTX_EL0      = 0x9828, // 10  011  0000  0101  000
+    OSLAR_EL1         = 0x8084, // 10  000  0001  0000  100
+    PMSWINC_EL0       = 0xdce4,  // 11  011  1001  1100  100
+
+    // Trace Registers
+    TRCOSLAR          = 0x8884, // 10  001  0001  0000  100
+    TRCLAR            = 0x8be6, // 10  001  0111  1100  110
+
+    // GICv3 registers
+    ICC_EOIR1_EL1     = 0xc661, // 11  000  1100  1100  001
+    ICC_EOIR0_EL1     = 0xc641, // 11  000  1100  1000  001
+    ICC_DIR_EL1       = 0xc659, // 11  000  1100  1011  001
+    ICC_SGI1R_EL1     = 0xc65d, // 11  000  1100  1011  101
+    ICC_ASGI1R_EL1    = 0xc65e, // 11  000  1100  1011  110
+    ICC_SGI0R_EL1     = 0xc65f  // 11  000  1100  1011  111
+  };
+
+  enum SysRegValues {
+    Invalid = -1,               // Op0 Op1  CRn   CRm   Op2
+    OSDTRRX_EL1       = 0x8002, // 10  000  0000  0000  010
+    OSDTRTX_EL1       = 0x801a, // 10  000  0000  0011  010
+    TEECR32_EL1       = 0x9000, // 10  010  0000  0000  000
+    MDCCINT_EL1       = 0x8010, // 10  000  0000  0010  000
+    MDSCR_EL1         = 0x8012, // 10  000  0000  0010  010
+    DBGDTR_EL0        = 0x9820, // 10  011  0000  0100  000
+    OSECCR_EL1        = 0x8032, // 10  000  0000  0110  010
+    DBGVCR32_EL2      = 0xa038, // 10  100  0000  0111  000
+    DBGBVR0_EL1       = 0x8004, // 10  000  0000  0000  100
+    DBGBVR1_EL1       = 0x800c, // 10  000  0000  0001  100
+    DBGBVR2_EL1       = 0x8014, // 10  000  0000  0010  100
+    DBGBVR3_EL1       = 0x801c, // 10  000  0000  0011  100
+    DBGBVR4_EL1       = 0x8024, // 10  000  0000  0100  100
+    DBGBVR5_EL1       = 0x802c, // 10  000  0000  0101  100
+    DBGBVR6_EL1       = 0x8034, // 10  000  0000  0110  100
+    DBGBVR7_EL1       = 0x803c, // 10  000  0000  0111  100
+    DBGBVR8_EL1       = 0x8044, // 10  000  0000  1000  100
+    DBGBVR9_EL1       = 0x804c, // 10  000  0000  1001  100
+    DBGBVR10_EL1      = 0x8054, // 10  000  0000  1010  100
+    DBGBVR11_EL1      = 0x805c, // 10  000  0000  1011  100
+    DBGBVR12_EL1      = 0x8064, // 10  000  0000  1100  100
+    DBGBVR13_EL1      = 0x806c, // 10  000  0000  1101  100
+    DBGBVR14_EL1      = 0x8074, // 10  000  0000  1110  100
+    DBGBVR15_EL1      = 0x807c, // 10  000  0000  1111  100
+    DBGBCR0_EL1       = 0x8005, // 10  000  0000  0000  101
+    DBGBCR1_EL1       = 0x800d, // 10  000  0000  0001  101
+    DBGBCR2_EL1       = 0x8015, // 10  000  0000  0010  101
+    DBGBCR3_EL1       = 0x801d, // 10  000  0000  0011  101
+    DBGBCR4_EL1       = 0x8025, // 10  000  0000  0100  101
+    DBGBCR5_EL1       = 0x802d, // 10  000  0000  0101  101
+    DBGBCR6_EL1       = 0x8035, // 10  000  0000  0110  101
+    DBGBCR7_EL1       = 0x803d, // 10  000  0000  0111  101
+    DBGBCR8_EL1       = 0x8045, // 10  000  0000  1000  101
+    DBGBCR9_EL1       = 0x804d, // 10  000  0000  1001  101
+    DBGBCR10_EL1      = 0x8055, // 10  000  0000  1010  101
+    DBGBCR11_EL1      = 0x805d, // 10  000  0000  1011  101
+    DBGBCR12_EL1      = 0x8065, // 10  000  0000  1100  101
+    DBGBCR13_EL1      = 0x806d, // 10  000  0000  1101  101
+    DBGBCR14_EL1      = 0x8075, // 10  000  0000  1110  101
+    DBGBCR15_EL1      = 0x807d, // 10  000  0000  1111  101
+    DBGWVR0_EL1       = 0x8006, // 10  000  0000  0000  110
+    DBGWVR1_EL1       = 0x800e, // 10  000  0000  0001  110
+    DBGWVR2_EL1       = 0x8016, // 10  000  0000  0010  110
+    DBGWVR3_EL1       = 0x801e, // 10  000  0000  0011  110
+    DBGWVR4_EL1       = 0x8026, // 10  000  0000  0100  110
+    DBGWVR5_EL1       = 0x802e, // 10  000  0000  0101  110
+    DBGWVR6_EL1       = 0x8036, // 10  000  0000  0110  110
+    DBGWVR7_EL1       = 0x803e, // 10  000  0000  0111  110
+    DBGWVR8_EL1       = 0x8046, // 10  000  0000  1000  110
+    DBGWVR9_EL1       = 0x804e, // 10  000  0000  1001  110
+    DBGWVR10_EL1      = 0x8056, // 10  000  0000  1010  110
+    DBGWVR11_EL1      = 0x805e, // 10  000  0000  1011  110
+    DBGWVR12_EL1      = 0x8066, // 10  000  0000  1100  110
+    DBGWVR13_EL1      = 0x806e, // 10  000  0000  1101  110
+    DBGWVR14_EL1      = 0x8076, // 10  000  0000  1110  110
+    DBGWVR15_EL1      = 0x807e, // 10  000  0000  1111  110
+    DBGWCR0_EL1       = 0x8007, // 10  000  0000  0000  111
+    DBGWCR1_EL1       = 0x800f, // 10  000  0000  0001  111
+    DBGWCR2_EL1       = 0x8017, // 10  000  0000  0010  111
+    DBGWCR3_EL1       = 0x801f, // 10  000  0000  0011  111
+    DBGWCR4_EL1       = 0x8027, // 10  000  0000  0100  111
+    DBGWCR5_EL1       = 0x802f, // 10  000  0000  0101  111
+    DBGWCR6_EL1       = 0x8037, // 10  000  0000  0110  111
+    DBGWCR7_EL1       = 0x803f, // 10  000  0000  0111  111
+    DBGWCR8_EL1       = 0x8047, // 10  000  0000  1000  111
+    DBGWCR9_EL1       = 0x804f, // 10  000  0000  1001  111
+    DBGWCR10_EL1      = 0x8057, // 10  000  0000  1010  111
+    DBGWCR11_EL1      = 0x805f, // 10  000  0000  1011  111
+    DBGWCR12_EL1      = 0x8067, // 10  000  0000  1100  111
+    DBGWCR13_EL1      = 0x806f, // 10  000  0000  1101  111
+    DBGWCR14_EL1      = 0x8077, // 10  000  0000  1110  111
+    DBGWCR15_EL1      = 0x807f, // 10  000  0000  1111  111
+    TEEHBR32_EL1      = 0x9080, // 10  010  0001  0000  000
+    OSDLR_EL1         = 0x809c, // 10  000  0001  0011  100
+    DBGPRCR_EL1       = 0x80a4, // 10  000  0001  0100  100
+    DBGCLAIMSET_EL1   = 0x83c6, // 10  000  0111  1000  110
+    DBGCLAIMCLR_EL1   = 0x83ce, // 10  000  0111  1001  110
+    CSSELR_EL1        = 0xd000, // 11  010  0000  0000  000
+    VPIDR_EL2         = 0xe000, // 11  100  0000  0000  000
+    VMPIDR_EL2        = 0xe005, // 11  100  0000  0000  101
+    CPACR_EL1         = 0xc082, // 11  000  0001  0000  010
+    SCTLR_EL1         = 0xc080, // 11  000  0001  0000  000
+    SCTLR_EL2         = 0xe080, // 11  100  0001  0000  000
+    SCTLR_EL3         = 0xf080, // 11  110  0001  0000  000
+    ACTLR_EL1         = 0xc081, // 11  000  0001  0000  001
+    ACTLR_EL2         = 0xe081, // 11  100  0001  0000  001
+    ACTLR_EL3         = 0xf081, // 11  110  0001  0000  001
+    HCR_EL2           = 0xe088, // 11  100  0001  0001  000
+    SCR_EL3           = 0xf088, // 11  110  0001  0001  000
+    MDCR_EL2          = 0xe089, // 11  100  0001  0001  001
+    SDER32_EL3        = 0xf089, // 11  110  0001  0001  001
+    CPTR_EL2          = 0xe08a, // 11  100  0001  0001  010
+    CPTR_EL3          = 0xf08a, // 11  110  0001  0001  010
+    HSTR_EL2          = 0xe08b, // 11  100  0001  0001  011
+    HACR_EL2          = 0xe08f, // 11  100  0001  0001  111
+    MDCR_EL3          = 0xf099, // 11  110  0001  0011  001
+    TTBR0_EL1         = 0xc100, // 11  000  0010  0000  000
+    TTBR0_EL2         = 0xe100, // 11  100  0010  0000  000
+    TTBR0_EL3         = 0xf100, // 11  110  0010  0000  000
+    TTBR1_EL1         = 0xc101, // 11  000  0010  0000  001
+    TCR_EL1           = 0xc102, // 11  000  0010  0000  010
+    TCR_EL2           = 0xe102, // 11  100  0010  0000  010
+    TCR_EL3           = 0xf102, // 11  110  0010  0000  010
+    VTTBR_EL2         = 0xe108, // 11  100  0010  0001  000
+    VTCR_EL2          = 0xe10a, // 11  100  0010  0001  010
+    DACR32_EL2        = 0xe180, // 11  100  0011  0000  000
+    SPSR_EL1          = 0xc200, // 11  000  0100  0000  000
+    SPSR_EL2          = 0xe200, // 11  100  0100  0000  000
+    SPSR_EL3          = 0xf200, // 11  110  0100  0000  000
+    ELR_EL1           = 0xc201, // 11  000  0100  0000  001
+    ELR_EL2           = 0xe201, // 11  100  0100  0000  001
+    ELR_EL3           = 0xf201, // 11  110  0100  0000  001
+    SP_EL0            = 0xc208, // 11  000  0100  0001  000
+    SP_EL1            = 0xe208, // 11  100  0100  0001  000
+    SP_EL2            = 0xf208, // 11  110  0100  0001  000
+    SPSel             = 0xc210, // 11  000  0100  0010  000
+    NZCV              = 0xda10, // 11  011  0100  0010  000
+    DAIF              = 0xda11, // 11  011  0100  0010  001
+    CurrentEL         = 0xc212, // 11  000  0100  0010  010
+    SPSR_irq          = 0xe218, // 11  100  0100  0011  000
+    SPSR_abt          = 0xe219, // 11  100  0100  0011  001
+    SPSR_und          = 0xe21a, // 11  100  0100  0011  010
+    SPSR_fiq          = 0xe21b, // 11  100  0100  0011  011
+    FPCR              = 0xda20, // 11  011  0100  0100  000
+    FPSR              = 0xda21, // 11  011  0100  0100  001
+    DSPSR_EL0         = 0xda28, // 11  011  0100  0101  000
+    DLR_EL0           = 0xda29, // 11  011  0100  0101  001
+    IFSR32_EL2        = 0xe281, // 11  100  0101  0000  001
+    AFSR0_EL1         = 0xc288, // 11  000  0101  0001  000
+    AFSR0_EL2         = 0xe288, // 11  100  0101  0001  000
+    AFSR0_EL3         = 0xf288, // 11  110  0101  0001  000
+    AFSR1_EL1         = 0xc289, // 11  000  0101  0001  001
+    AFSR1_EL2         = 0xe289, // 11  100  0101  0001  001
+    AFSR1_EL3         = 0xf289, // 11  110  0101  0001  001
+    ESR_EL1           = 0xc290, // 11  000  0101  0010  000
+    ESR_EL2           = 0xe290, // 11  100  0101  0010  000
+    ESR_EL3           = 0xf290, // 11  110  0101  0010  000
+    FPEXC32_EL2       = 0xe298, // 11  100  0101  0011  000
+    FAR_EL1           = 0xc300, // 11  000  0110  0000  000
+    FAR_EL2           = 0xe300, // 11  100  0110  0000  000
+    FAR_EL3           = 0xf300, // 11  110  0110  0000  000
+    HPFAR_EL2         = 0xe304, // 11  100  0110  0000  100
+    PAR_EL1           = 0xc3a0, // 11  000  0111  0100  000
+    PMCR_EL0          = 0xdce0, // 11  011  1001  1100  000
+    PMCNTENSET_EL0    = 0xdce1, // 11  011  1001  1100  001
+    PMCNTENCLR_EL0    = 0xdce2, // 11  011  1001  1100  010
+    PMOVSCLR_EL0      = 0xdce3, // 11  011  1001  1100  011
+    PMSELR_EL0        = 0xdce5, // 11  011  1001  1100  101
+    PMCCNTR_EL0       = 0xdce8, // 11  011  1001  1101  000
+    PMXEVTYPER_EL0    = 0xdce9, // 11  011  1001  1101  001
+    PMXEVCNTR_EL0     = 0xdcea, // 11  011  1001  1101  010
+    PMUSERENR_EL0     = 0xdcf0, // 11  011  1001  1110  000
+    PMINTENSET_EL1    = 0xc4f1, // 11  000  1001  1110  001
+    PMINTENCLR_EL1    = 0xc4f2, // 11  000  1001  1110  010
+    PMOVSSET_EL0      = 0xdcf3, // 11  011  1001  1110  011
+    MAIR_EL1          = 0xc510, // 11  000  1010  0010  000
+    MAIR_EL2          = 0xe510, // 11  100  1010  0010  000
+    MAIR_EL3          = 0xf510, // 11  110  1010  0010  000
+    AMAIR_EL1         = 0xc518, // 11  000  1010  0011  000
+    AMAIR_EL2         = 0xe518, // 11  100  1010  0011  000
+    AMAIR_EL3         = 0xf518, // 11  110  1010  0011  000
+    VBAR_EL1          = 0xc600, // 11  000  1100  0000  000
+    VBAR_EL2          = 0xe600, // 11  100  1100  0000  000
+    VBAR_EL3          = 0xf600, // 11  110  1100  0000  000
+    RMR_EL1           = 0xc602, // 11  000  1100  0000  010
+    RMR_EL2           = 0xe602, // 11  100  1100  0000  010
+    RMR_EL3           = 0xf602, // 11  110  1100  0000  010
+    CONTEXTIDR_EL1    = 0xc681, // 11  000  1101  0000  001
+    TPIDR_EL0         = 0xde82, // 11  011  1101  0000  010
+    TPIDR_EL2         = 0xe682, // 11  100  1101  0000  010
+    TPIDR_EL3         = 0xf682, // 11  110  1101  0000  010
+    TPIDRRO_EL0       = 0xde83, // 11  011  1101  0000  011
+    TPIDR_EL1         = 0xc684, // 11  000  1101  0000  100
+    CNTFRQ_EL0        = 0xdf00, // 11  011  1110  0000  000
+    CNTVOFF_EL2       = 0xe703, // 11  100  1110  0000  011
+    CNTKCTL_EL1       = 0xc708, // 11  000  1110  0001  000
+    CNTHCTL_EL2       = 0xe708, // 11  100  1110  0001  000
+    CNTP_TVAL_EL0     = 0xdf10, // 11  011  1110  0010  000
+    CNTHP_TVAL_EL2    = 0xe710, // 11  100  1110  0010  000
+    CNTPS_TVAL_EL1    = 0xff10, // 11  111  1110  0010  000
+    CNTP_CTL_EL0      = 0xdf11, // 11  011  1110  0010  001
+    CNTHP_CTL_EL2     = 0xe711, // 11  100  1110  0010  001
+    CNTPS_CTL_EL1     = 0xff11, // 11  111  1110  0010  001
+    CNTP_CVAL_EL0     = 0xdf12, // 11  011  1110  0010  010
+    CNTHP_CVAL_EL2    = 0xe712, // 11  100  1110  0010  010
+    CNTPS_CVAL_EL1    = 0xff12, // 11  111  1110  0010  010
+    CNTV_TVAL_EL0     = 0xdf18, // 11  011  1110  0011  000
+    CNTV_CTL_EL0      = 0xdf19, // 11  011  1110  0011  001
+    CNTV_CVAL_EL0     = 0xdf1a, // 11  011  1110  0011  010
+    PMEVCNTR0_EL0     = 0xdf40, // 11  011  1110  1000  000
+    PMEVCNTR1_EL0     = 0xdf41, // 11  011  1110  1000  001
+    PMEVCNTR2_EL0     = 0xdf42, // 11  011  1110  1000  010
+    PMEVCNTR3_EL0     = 0xdf43, // 11  011  1110  1000  011
+    PMEVCNTR4_EL0     = 0xdf44, // 11  011  1110  1000  100
+    PMEVCNTR5_EL0     = 0xdf45, // 11  011  1110  1000  101
+    PMEVCNTR6_EL0     = 0xdf46, // 11  011  1110  1000  110
+    PMEVCNTR7_EL0     = 0xdf47, // 11  011  1110  1000  111
+    PMEVCNTR8_EL0     = 0xdf48, // 11  011  1110  1001  000
+    PMEVCNTR9_EL0     = 0xdf49, // 11  011  1110  1001  001
+    PMEVCNTR10_EL0    = 0xdf4a, // 11  011  1110  1001  010
+    PMEVCNTR11_EL0    = 0xdf4b, // 11  011  1110  1001  011
+    PMEVCNTR12_EL0    = 0xdf4c, // 11  011  1110  1001  100
+    PMEVCNTR13_EL0    = 0xdf4d, // 11  011  1110  1001  101
+    PMEVCNTR14_EL0    = 0xdf4e, // 11  011  1110  1001  110
+    PMEVCNTR15_EL0    = 0xdf4f, // 11  011  1110  1001  111
+    PMEVCNTR16_EL0    = 0xdf50, // 11  011  1110  1010  000
+    PMEVCNTR17_EL0    = 0xdf51, // 11  011  1110  1010  001
+    PMEVCNTR18_EL0    = 0xdf52, // 11  011  1110  1010  010
+    PMEVCNTR19_EL0    = 0xdf53, // 11  011  1110  1010  011
+    PMEVCNTR20_EL0    = 0xdf54, // 11  011  1110  1010  100
+    PMEVCNTR21_EL0    = 0xdf55, // 11  011  1110  1010  101
+    PMEVCNTR22_EL0    = 0xdf56, // 11  011  1110  1010  110
+    PMEVCNTR23_EL0    = 0xdf57, // 11  011  1110  1010  111
+    PMEVCNTR24_EL0    = 0xdf58, // 11  011  1110  1011  000
+    PMEVCNTR25_EL0    = 0xdf59, // 11  011  1110  1011  001
+    PMEVCNTR26_EL0    = 0xdf5a, // 11  011  1110  1011  010
+    PMEVCNTR27_EL0    = 0xdf5b, // 11  011  1110  1011  011
+    PMEVCNTR28_EL0    = 0xdf5c, // 11  011  1110  1011  100
+    PMEVCNTR29_EL0    = 0xdf5d, // 11  011  1110  1011  101
+    PMEVCNTR30_EL0    = 0xdf5e, // 11  011  1110  1011  110
+    PMCCFILTR_EL0     = 0xdf7f, // 11  011  1110  1111  111
+    PMEVTYPER0_EL0    = 0xdf60, // 11  011  1110  1100  000
+    PMEVTYPER1_EL0    = 0xdf61, // 11  011  1110  1100  001
+    PMEVTYPER2_EL0    = 0xdf62, // 11  011  1110  1100  010
+    PMEVTYPER3_EL0    = 0xdf63, // 11  011  1110  1100  011
+    PMEVTYPER4_EL0    = 0xdf64, // 11  011  1110  1100  100
+    PMEVTYPER5_EL0    = 0xdf65, // 11  011  1110  1100  101
+    PMEVTYPER6_EL0    = 0xdf66, // 11  011  1110  1100  110
+    PMEVTYPER7_EL0    = 0xdf67, // 11  011  1110  1100  111
+    PMEVTYPER8_EL0    = 0xdf68, // 11  011  1110  1101  000
+    PMEVTYPER9_EL0    = 0xdf69, // 11  011  1110  1101  001
+    PMEVTYPER10_EL0   = 0xdf6a, // 11  011  1110  1101  010
+    PMEVTYPER11_EL0   = 0xdf6b, // 11  011  1110  1101  011
+    PMEVTYPER12_EL0   = 0xdf6c, // 11  011  1110  1101  100
+    PMEVTYPER13_EL0   = 0xdf6d, // 11  011  1110  1101  101
+    PMEVTYPER14_EL0   = 0xdf6e, // 11  011  1110  1101  110
+    PMEVTYPER15_EL0   = 0xdf6f, // 11  011  1110  1101  111
+    PMEVTYPER16_EL0   = 0xdf70, // 11  011  1110  1110  000
+    PMEVTYPER17_EL0   = 0xdf71, // 11  011  1110  1110  001
+    PMEVTYPER18_EL0   = 0xdf72, // 11  011  1110  1110  010
+    PMEVTYPER19_EL0   = 0xdf73, // 11  011  1110  1110  011
+    PMEVTYPER20_EL0   = 0xdf74, // 11  011  1110  1110  100
+    PMEVTYPER21_EL0   = 0xdf75, // 11  011  1110  1110  101
+    PMEVTYPER22_EL0   = 0xdf76, // 11  011  1110  1110  110
+    PMEVTYPER23_EL0   = 0xdf77, // 11  011  1110  1110  111
+    PMEVTYPER24_EL0   = 0xdf78, // 11  011  1110  1111  000
+    PMEVTYPER25_EL0   = 0xdf79, // 11  011  1110  1111  001
+    PMEVTYPER26_EL0   = 0xdf7a, // 11  011  1110  1111  010
+    PMEVTYPER27_EL0   = 0xdf7b, // 11  011  1110  1111  011
+    PMEVTYPER28_EL0   = 0xdf7c, // 11  011  1110  1111  100
+    PMEVTYPER29_EL0   = 0xdf7d, // 11  011  1110  1111  101
+    PMEVTYPER30_EL0   = 0xdf7e, // 11  011  1110  1111  110
+
+    // Trace registers
+    TRCPRGCTLR        = 0x8808, // 10  001  0000  0001  000
+    TRCPROCSELR       = 0x8810, // 10  001  0000  0010  000
+    TRCCONFIGR        = 0x8820, // 10  001  0000  0100  000
+    TRCAUXCTLR        = 0x8830, // 10  001  0000  0110  000
+    TRCEVENTCTL0R     = 0x8840, // 10  001  0000  1000  000
+    TRCEVENTCTL1R     = 0x8848, // 10  001  0000  1001  000
+    TRCSTALLCTLR      = 0x8858, // 10  001  0000  1011  000
+    TRCTSCTLR         = 0x8860, // 10  001  0000  1100  000
+    TRCSYNCPR         = 0x8868, // 10  001  0000  1101  000
+    TRCCCCTLR         = 0x8870, // 10  001  0000  1110  000
+    TRCBBCTLR         = 0x8878, // 10  001  0000  1111  000
+    TRCTRACEIDR       = 0x8801, // 10  001  0000  0000  001
+    TRCQCTLR          = 0x8809, // 10  001  0000  0001  001
+    TRCVICTLR         = 0x8802, // 10  001  0000  0000  010
+    TRCVIIECTLR       = 0x880a, // 10  001  0000  0001  010
+    TRCVISSCTLR       = 0x8812, // 10  001  0000  0010  010
+    TRCVIPCSSCTLR     = 0x881a, // 10  001  0000  0011  010
+    TRCVDCTLR         = 0x8842, // 10  001  0000  1000  010
+    TRCVDSACCTLR      = 0x884a, // 10  001  0000  1001  010
+    TRCVDARCCTLR      = 0x8852, // 10  001  0000  1010  010
+    TRCSEQEVR0        = 0x8804, // 10  001  0000  0000  100
+    TRCSEQEVR1        = 0x880c, // 10  001  0000  0001  100
+    TRCSEQEVR2        = 0x8814, // 10  001  0000  0010  100
+    TRCSEQRSTEVR      = 0x8834, // 10  001  0000  0110  100
+    TRCSEQSTR         = 0x883c, // 10  001  0000  0111  100
+    TRCEXTINSELR      = 0x8844, // 10  001  0000  1000  100
+    TRCCNTRLDVR0      = 0x8805, // 10  001  0000  0000  101
+    TRCCNTRLDVR1      = 0x880d, // 10  001  0000  0001  101
+    TRCCNTRLDVR2      = 0x8815, // 10  001  0000  0010  101
+    TRCCNTRLDVR3      = 0x881d, // 10  001  0000  0011  101
+    TRCCNTCTLR0       = 0x8825, // 10  001  0000  0100  101
+    TRCCNTCTLR1       = 0x882d, // 10  001  0000  0101  101
+    TRCCNTCTLR2       = 0x8835, // 10  001  0000  0110  101
+    TRCCNTCTLR3       = 0x883d, // 10  001  0000  0111  101
+    TRCCNTVR0         = 0x8845, // 10  001  0000  1000  101
+    TRCCNTVR1         = 0x884d, // 10  001  0000  1001  101
+    TRCCNTVR2         = 0x8855, // 10  001  0000  1010  101
+    TRCCNTVR3         = 0x885d, // 10  001  0000  1011  101
+    TRCIMSPEC0        = 0x8807, // 10  001  0000  0000  111
+    TRCIMSPEC1        = 0x880f, // 10  001  0000  0001  111
+    TRCIMSPEC2        = 0x8817, // 10  001  0000  0010  111
+    TRCIMSPEC3        = 0x881f, // 10  001  0000  0011  111
+    TRCIMSPEC4        = 0x8827, // 10  001  0000  0100  111
+    TRCIMSPEC5        = 0x882f, // 10  001  0000  0101  111
+    TRCIMSPEC6        = 0x8837, // 10  001  0000  0110  111
+    TRCIMSPEC7        = 0x883f, // 10  001  0000  0111  111
+    TRCRSCTLR2        = 0x8890, // 10  001  0001  0010  000
+    TRCRSCTLR3        = 0x8898, // 10  001  0001  0011  000
+    TRCRSCTLR4        = 0x88a0, // 10  001  0001  0100  000
+    TRCRSCTLR5        = 0x88a8, // 10  001  0001  0101  000
+    TRCRSCTLR6        = 0x88b0, // 10  001  0001  0110  000
+    TRCRSCTLR7        = 0x88b8, // 10  001  0001  0111  000
+    TRCRSCTLR8        = 0x88c0, // 10  001  0001  1000  000
+    TRCRSCTLR9        = 0x88c8, // 10  001  0001  1001  000
+    TRCRSCTLR10       = 0x88d0, // 10  001  0001  1010  000
+    TRCRSCTLR11       = 0x88d8, // 10  001  0001  1011  000
+    TRCRSCTLR12       = 0x88e0, // 10  001  0001  1100  000
+    TRCRSCTLR13       = 0x88e8, // 10  001  0001  1101  000
+    TRCRSCTLR14       = 0x88f0, // 10  001  0001  1110  000
+    TRCRSCTLR15       = 0x88f8, // 10  001  0001  1111  000
+    TRCRSCTLR16       = 0x8881, // 10  001  0001  0000  001
+    TRCRSCTLR17       = 0x8889, // 10  001  0001  0001  001
+    TRCRSCTLR18       = 0x8891, // 10  001  0001  0010  001
+    TRCRSCTLR19       = 0x8899, // 10  001  0001  0011  001
+    TRCRSCTLR20       = 0x88a1, // 10  001  0001  0100  001
+    TRCRSCTLR21       = 0x88a9, // 10  001  0001  0101  001
+    TRCRSCTLR22       = 0x88b1, // 10  001  0001  0110  001
+    TRCRSCTLR23       = 0x88b9, // 10  001  0001  0111  001
+    TRCRSCTLR24       = 0x88c1, // 10  001  0001  1000  001
+    TRCRSCTLR25       = 0x88c9, // 10  001  0001  1001  001
+    TRCRSCTLR26       = 0x88d1, // 10  001  0001  1010  001
+    TRCRSCTLR27       = 0x88d9, // 10  001  0001  1011  001
+    TRCRSCTLR28       = 0x88e1, // 10  001  0001  1100  001
+    TRCRSCTLR29       = 0x88e9, // 10  001  0001  1101  001
+    TRCRSCTLR30       = 0x88f1, // 10  001  0001  1110  001
+    TRCRSCTLR31       = 0x88f9, // 10  001  0001  1111  001
+    TRCSSCCR0         = 0x8882, // 10  001  0001  0000  010
+    TRCSSCCR1         = 0x888a, // 10  001  0001  0001  010
+    TRCSSCCR2         = 0x8892, // 10  001  0001  0010  010
+    TRCSSCCR3         = 0x889a, // 10  001  0001  0011  010
+    TRCSSCCR4         = 0x88a2, // 10  001  0001  0100  010
+    TRCSSCCR5         = 0x88aa, // 10  001  0001  0101  010
+    TRCSSCCR6         = 0x88b2, // 10  001  0001  0110  010
+    TRCSSCCR7         = 0x88ba, // 10  001  0001  0111  010
+    TRCSSCSR0         = 0x88c2, // 10  001  0001  1000  010
+    TRCSSCSR1         = 0x88ca, // 10  001  0001  1001  010
+    TRCSSCSR2         = 0x88d2, // 10  001  0001  1010  010
+    TRCSSCSR3         = 0x88da, // 10  001  0001  1011  010
+    TRCSSCSR4         = 0x88e2, // 10  001  0001  1100  010
+    TRCSSCSR5         = 0x88ea, // 10  001  0001  1101  010
+    TRCSSCSR6         = 0x88f2, // 10  001  0001  1110  010
+    TRCSSCSR7         = 0x88fa, // 10  001  0001  1111  010
+    TRCSSPCICR0       = 0x8883, // 10  001  0001  0000  011
+    TRCSSPCICR1       = 0x888b, // 10  001  0001  0001  011
+    TRCSSPCICR2       = 0x8893, // 10  001  0001  0010  011
+    TRCSSPCICR3       = 0x889b, // 10  001  0001  0011  011
+    TRCSSPCICR4       = 0x88a3, // 10  001  0001  0100  011
+    TRCSSPCICR5       = 0x88ab, // 10  001  0001  0101  011
+    TRCSSPCICR6       = 0x88b3, // 10  001  0001  0110  011
+    TRCSSPCICR7       = 0x88bb, // 10  001  0001  0111  011
+    TRCPDCR           = 0x88a4, // 10  001  0001  0100  100
+    TRCACVR0          = 0x8900, // 10  001  0010  0000  000
+    TRCACVR1          = 0x8910, // 10  001  0010  0010  000
+    TRCACVR2          = 0x8920, // 10  001  0010  0100  000
+    TRCACVR3          = 0x8930, // 10  001  0010  0110  000
+    TRCACVR4          = 0x8940, // 10  001  0010  1000  000
+    TRCACVR5          = 0x8950, // 10  001  0010  1010  000
+    TRCACVR6          = 0x8960, // 10  001  0010  1100  000
+    TRCACVR7          = 0x8970, // 10  001  0010  1110  000
+    TRCACVR8          = 0x8901, // 10  001  0010  0000  001
+    TRCACVR9          = 0x8911, // 10  001  0010  0010  001
+    TRCACVR10         = 0x8921, // 10  001  0010  0100  001
+    TRCACVR11         = 0x8931, // 10  001  0010  0110  001
+    TRCACVR12         = 0x8941, // 10  001  0010  1000  001
+    TRCACVR13         = 0x8951, // 10  001  0010  1010  001
+    TRCACVR14         = 0x8961, // 10  001  0010  1100  001
+    TRCACVR15         = 0x8971, // 10  001  0010  1110  001
+    TRCACATR0         = 0x8902, // 10  001  0010  0000  010
+    TRCACATR1         = 0x8912, // 10  001  0010  0010  010
+    TRCACATR2         = 0x8922, // 10  001  0010  0100  010
+    TRCACATR3         = 0x8932, // 10  001  0010  0110  010
+    TRCACATR4         = 0x8942, // 10  001  0010  1000  010
+    TRCACATR5         = 0x8952, // 10  001  0010  1010  010
+    TRCACATR6         = 0x8962, // 10  001  0010  1100  010
+    TRCACATR7         = 0x8972, // 10  001  0010  1110  010
+    TRCACATR8         = 0x8903, // 10  001  0010  0000  011
+    TRCACATR9         = 0x8913, // 10  001  0010  0010  011
+    TRCACATR10        = 0x8923, // 10  001  0010  0100  011
+    TRCACATR11        = 0x8933, // 10  001  0010  0110  011
+    TRCACATR12        = 0x8943, // 10  001  0010  1000  011
+    TRCACATR13        = 0x8953, // 10  001  0010  1010  011
+    TRCACATR14        = 0x8963, // 10  001  0010  1100  011
+    TRCACATR15        = 0x8973, // 10  001  0010  1110  011
+    TRCDVCVR0         = 0x8904, // 10  001  0010  0000  100
+    TRCDVCVR1         = 0x8924, // 10  001  0010  0100  100
+    TRCDVCVR2         = 0x8944, // 10  001  0010  1000  100
+    TRCDVCVR3         = 0x8964, // 10  001  0010  1100  100
+    TRCDVCVR4         = 0x8905, // 10  001  0010  0000  101
+    TRCDVCVR5         = 0x8925, // 10  001  0010  0100  101
+    TRCDVCVR6         = 0x8945, // 10  001  0010  1000  101
+    TRCDVCVR7         = 0x8965, // 10  001  0010  1100  101
+    TRCDVCMR0         = 0x8906, // 10  001  0010  0000  110
+    TRCDVCMR1         = 0x8926, // 10  001  0010  0100  110
+    TRCDVCMR2         = 0x8946, // 10  001  0010  1000  110
+    TRCDVCMR3         = 0x8966, // 10  001  0010  1100  110
+    TRCDVCMR4         = 0x8907, // 10  001  0010  0000  111
+    TRCDVCMR5         = 0x8927, // 10  001  0010  0100  111
+    TRCDVCMR6         = 0x8947, // 10  001  0010  1000  111
+    TRCDVCMR7         = 0x8967, // 10  001  0010  1100  111
+    TRCCIDCVR0        = 0x8980, // 10  001  0011  0000  000
+    TRCCIDCVR1        = 0x8990, // 10  001  0011  0010  000
+    TRCCIDCVR2        = 0x89a0, // 10  001  0011  0100  000
+    TRCCIDCVR3        = 0x89b0, // 10  001  0011  0110  000
+    TRCCIDCVR4        = 0x89c0, // 10  001  0011  1000  000
+    TRCCIDCVR5        = 0x89d0, // 10  001  0011  1010  000
+    TRCCIDCVR6        = 0x89e0, // 10  001  0011  1100  000
+    TRCCIDCVR7        = 0x89f0, // 10  001  0011  1110  000
+    TRCVMIDCVR0       = 0x8981, // 10  001  0011  0000  001
+    TRCVMIDCVR1       = 0x8991, // 10  001  0011  0010  001
+    TRCVMIDCVR2       = 0x89a1, // 10  001  0011  0100  001
+    TRCVMIDCVR3       = 0x89b1, // 10  001  0011  0110  001
+    TRCVMIDCVR4       = 0x89c1, // 10  001  0011  1000  001
+    TRCVMIDCVR5       = 0x89d1, // 10  001  0011  1010  001
+    TRCVMIDCVR6       = 0x89e1, // 10  001  0011  1100  001
+    TRCVMIDCVR7       = 0x89f1, // 10  001  0011  1110  001
+    TRCCIDCCTLR0      = 0x8982, // 10  001  0011  0000  010
+    TRCCIDCCTLR1      = 0x898a, // 10  001  0011  0001  010
+    TRCVMIDCCTLR0     = 0x8992, // 10  001  0011  0010  010
+    TRCVMIDCCTLR1     = 0x899a, // 10  001  0011  0011  010
+    TRCITCTRL         = 0x8b84, // 10  001  0111  0000  100
+    TRCCLAIMSET       = 0x8bc6, // 10  001  0111  1000  110
+    TRCCLAIMCLR       = 0x8bce, // 10  001  0111  1001  110
+
+    // GICv3 registers
+    ICC_BPR1_EL1      = 0xc663, // 11  000  1100  1100  011
+    ICC_BPR0_EL1      = 0xc643, // 11  000  1100  1000  011
+    ICC_PMR_EL1       = 0xc230, // 11  000  0100  0110  000
+    ICC_CTLR_EL1      = 0xc664, // 11  000  1100  1100  100
+    ICC_CTLR_EL3      = 0xf664, // 11  110  1100  1100  100
+    ICC_SRE_EL1       = 0xc665, // 11  000  1100  1100  101
+    ICC_SRE_EL2       = 0xe64d, // 11  100  1100  1001  101
+    ICC_SRE_EL3       = 0xf665, // 11  110  1100  1100  101
+    ICC_IGRPEN0_EL1   = 0xc666, // 11  000  1100  1100  110
+    ICC_IGRPEN1_EL1   = 0xc667, // 11  000  1100  1100  111
+    ICC_IGRPEN1_EL3   = 0xf667, // 11  110  1100  1100  111
+    ICC_SEIEN_EL1     = 0xc668, // 11  000  1100  1101  000
+    ICC_AP0R0_EL1     = 0xc644, // 11  000  1100  1000  100
+    ICC_AP0R1_EL1     = 0xc645, // 11  000  1100  1000  101
+    ICC_AP0R2_EL1     = 0xc646, // 11  000  1100  1000  110
+    ICC_AP0R3_EL1     = 0xc647, // 11  000  1100  1000  111
+    ICC_AP1R0_EL1     = 0xc648, // 11  000  1100  1001  000
+    ICC_AP1R1_EL1     = 0xc649, // 11  000  1100  1001  001
+    ICC_AP1R2_EL1     = 0xc64a, // 11  000  1100  1001  010
+    ICC_AP1R3_EL1     = 0xc64b, // 11  000  1100  1001  011
+    ICH_AP0R0_EL2     = 0xe640, // 11  100  1100  1000  000
+    ICH_AP0R1_EL2     = 0xe641, // 11  100  1100  1000  001
+    ICH_AP0R2_EL2     = 0xe642, // 11  100  1100  1000  010
+    ICH_AP0R3_EL2     = 0xe643, // 11  100  1100  1000  011
+    ICH_AP1R0_EL2     = 0xe648, // 11  100  1100  1001  000
+    ICH_AP1R1_EL2     = 0xe649, // 11  100  1100  1001  001
+    ICH_AP1R2_EL2     = 0xe64a, // 11  100  1100  1001  010
+    ICH_AP1R3_EL2     = 0xe64b, // 11  100  1100  1001  011
+    ICH_HCR_EL2       = 0xe658, // 11  100  1100  1011  000
+    ICH_MISR_EL2      = 0xe65a, // 11  100  1100  1011  010
+    ICH_VMCR_EL2      = 0xe65f, // 11  100  1100  1011  111
+    ICH_VSEIR_EL2     = 0xe64c, // 11  100  1100  1001  100
+    ICH_LR0_EL2       = 0xe660, // 11  100  1100  1100  000
+    ICH_LR1_EL2       = 0xe661, // 11  100  1100  1100  001
+    ICH_LR2_EL2       = 0xe662, // 11  100  1100  1100  010
+    ICH_LR3_EL2       = 0xe663, // 11  100  1100  1100  011
+    ICH_LR4_EL2       = 0xe664, // 11  100  1100  1100  100
+    ICH_LR5_EL2       = 0xe665, // 11  100  1100  1100  101
+    ICH_LR6_EL2       = 0xe666, // 11  100  1100  1100  110
+    ICH_LR7_EL2       = 0xe667, // 11  100  1100  1100  111
+    ICH_LR8_EL2       = 0xe668, // 11  100  1100  1101  000
+    ICH_LR9_EL2       = 0xe669, // 11  100  1100  1101  001
+    ICH_LR10_EL2      = 0xe66a, // 11  100  1100  1101  010
+    ICH_LR11_EL2      = 0xe66b, // 11  100  1100  1101  011
+    ICH_LR12_EL2      = 0xe66c, // 11  100  1100  1101  100
+    ICH_LR13_EL2      = 0xe66d, // 11  100  1100  1101  101
+    ICH_LR14_EL2      = 0xe66e, // 11  100  1100  1101  110
+    ICH_LR15_EL2      = 0xe66f  // 11  100  1100  1101  111
+  };
+
+  // Note that these do not inherit from NamedImmMapper. This class is
+  // sufficiently different in its behaviour that I don't believe it's worth
+  // burdening the common NamedImmMapper with abstractions only needed in
+  // this one case.
+  struct SysRegMapper {
+    static const NamedImmMapper::Mapping SysRegPairs[];
+
+    const NamedImmMapper::Mapping *InstPairs;
+    size_t NumInstPairs;
+
+    SysRegMapper() {}
+    uint32_t fromString(StringRef Name, bool &Valid) const;
+    std::string toString(uint32_t Bits, bool &Valid) const;
+  };
+
+  struct MSRMapper : SysRegMapper {
+    static const NamedImmMapper::Mapping MSRPairs[];
+    MSRMapper();
+  };
+
+  struct MRSMapper : SysRegMapper {
+    static const NamedImmMapper::Mapping MRSPairs[];
+    MRSMapper();
+  };
+
+  uint32_t ParseGenericRegister(StringRef Name, bool &Valid);
+}
+
+namespace A64TLBI {
+  enum TLBIValues {
+    Invalid = -1,          // Op0 Op1  CRn   CRm   Op2
+    IPAS2E1IS    = 0x6401, // 01  100  1000  0000  001
+    IPAS2LE1IS   = 0x6405, // 01  100  1000  0000  101
+    VMALLE1IS    = 0x4418, // 01  000  1000  0011  000
+    ALLE2IS      = 0x6418, // 01  100  1000  0011  000
+    ALLE3IS      = 0x7418, // 01  110  1000  0011  000
+    VAE1IS       = 0x4419, // 01  000  1000  0011  001
+    VAE2IS       = 0x6419, // 01  100  1000  0011  001
+    VAE3IS       = 0x7419, // 01  110  1000  0011  001
+    ASIDE1IS     = 0x441a, // 01  000  1000  0011  010
+    VAAE1IS      = 0x441b, // 01  000  1000  0011  011
+    ALLE1IS      = 0x641c, // 01  100  1000  0011  100
+    VALE1IS      = 0x441d, // 01  000  1000  0011  101
+    VALE2IS      = 0x641d, // 01  100  1000  0011  101
+    VALE3IS      = 0x741d, // 01  110  1000  0011  101
+    VMALLS12E1IS = 0x641e, // 01  100  1000  0011  110
+    VAALE1IS     = 0x441f, // 01  000  1000  0011  111
+    IPAS2E1      = 0x6421, // 01  100  1000  0100  001
+    IPAS2LE1     = 0x6425, // 01  100  1000  0100  101
+    VMALLE1      = 0x4438, // 01  000  1000  0111  000
+    ALLE2        = 0x6438, // 01  100  1000  0111  000
+    ALLE3        = 0x7438, // 01  110  1000  0111  000
+    VAE1         = 0x4439, // 01  000  1000  0111  001
+    VAE2         = 0x6439, // 01  100  1000  0111  001
+    VAE3         = 0x7439, // 01  110  1000  0111  001
+    ASIDE1       = 0x443a, // 01  000  1000  0111  010
+    VAAE1        = 0x443b, // 01  000  1000  0111  011
+    ALLE1        = 0x643c, // 01  100  1000  0111  100
+    VALE1        = 0x443d, // 01  000  1000  0111  101
+    VALE2        = 0x643d, // 01  100  1000  0111  101
+    VALE3        = 0x743d, // 01  110  1000  0111  101
+    VMALLS12E1   = 0x643e, // 01  100  1000  0111  110
+    VAALE1       = 0x443f  // 01  000  1000  0111  111
+  };
+
+  struct TLBIMapper : NamedImmMapper {
+    const static Mapping TLBIPairs[];
+
+    TLBIMapper();
+  };
+
+  static inline bool NeedsRegister(TLBIValues Val) {
+    switch (Val) {
+    case VMALLE1IS:
+    case ALLE2IS:
+    case ALLE3IS:
+    case ALLE1IS:
+    case VMALLS12E1IS:
+    case VMALLE1:
+    case ALLE2:
+    case ALLE3:
+    case ALLE1:
+    case VMALLS12E1:
+      return false;
+    default:
+      return true;
+    }
+  }
+}
+
+namespace AArch64II {
+
+  enum TOF {
+    //===--------------------------------------------------------------===//
+    // AArch64 Specific MachineOperand flags.
+
+    MO_NO_FLAG,
+
+    // MO_GOT - Represents a relocation referring to the GOT entry of a given
+    // symbol. Used in adrp.
+    MO_GOT,
+
+    // MO_GOT_LO12 - Represents a relocation referring to the low 12 bits of the
+    // GOT entry of a given symbol. Used in ldr only.
+    MO_GOT_LO12,
+
+    // MO_DTPREL_* - Represents a relocation referring to the offset from a
+    // module's dynamic thread pointer. Used in the local-dynamic TLS access
+    // model.
+    MO_DTPREL_G1,
+    MO_DTPREL_G0_NC,
+
+    // MO_GOTTPREL_* - Represents a relocation referring to a GOT entry
+    // providing the offset of a variable from the thread-pointer. Used in
+    // initial-exec TLS model where this offset is assigned in the static thread
+    // block and thus known by the dynamic linker.
+    MO_GOTTPREL,
+    MO_GOTTPREL_LO12,
+
+    // MO_TLSDESC_* - Represents a relocation referring to a GOT entry providing
+    // a TLS descriptor chosen by the dynamic linker. Used for the
+    // general-dynamic and local-dynamic TLS access models where very littls is
+    // known at link-time.
+    MO_TLSDESC,
+    MO_TLSDESC_LO12,
+
+    // MO_TPREL_* - Represents a relocation referring to the offset of a
+    // variable from the thread pointer itself. Used in the local-exec TLS
+    // access model.
+    MO_TPREL_G1,
+    MO_TPREL_G0_NC,
+
+    // MO_LO12 - On a symbol operand, this represents a relocation containing
+    // lower 12 bits of the address. Used in add/sub/ldr/str.
+    MO_LO12
+  };
+}
+
+class APFloat;
+
+namespace A64Imms {
+  bool isFPImm(const APFloat &Val, uint32_t &Imm8Bits);
+
+  inline bool isFPImm(const APFloat &Val) {
+    uint32_t Imm8;
+    return isFPImm(Val, Imm8);
+  }
+
+  bool isLogicalImm(unsigned RegWidth, uint64_t Imm, uint32_t &Bits);
+  bool isLogicalImmBits(unsigned RegWidth, uint32_t Bits, uint64_t &Imm);
+
+  bool isMOVZImm(int RegWidth, uint64_t Value, int &UImm16, int &Shift);
+  bool isMOVNImm(int RegWidth, uint64_t Value, int &UImm16, int &Shift);
+
+  // We sometimes want to know whether the immediate is representable with a
+  // MOVN but *not* with a MOVZ (because that would take priority).
+  bool isOnlyMOVNImm(int RegWidth, uint64_t Value, int &UImm16, int &Shift);
+
+}
+
+} // end namespace llvm;
+
+#endif
diff --git a/lib/Target/AArch64/Utils/CMakeLists.txt b/lib/Target/AArch64/Utils/CMakeLists.txt
new file mode 100644
index 000000000000..2c28348d7d81
--- /dev/null
+++ b/lib/Target/AArch64/Utils/CMakeLists.txt
@@ -0,0 +1,5 @@
+include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
+
+add_llvm_library(LLVMAArch64Utils
+  AArch64BaseInfo.cpp
+  )
diff --git a/lib/Target/CellSPU/MCTargetDesc/LLVMBuild.txt b/lib/Target/AArch64/Utils/LLVMBuild.txt
index 71e5bbc629ca..1be537598ae5 100644
--- a/lib/Target/CellSPU/MCTargetDesc/LLVMBuild.txt
+++ b/lib/Target/AArch64/Utils/LLVMBuild.txt
@@ -1,4 +1,4 @@
-;===- ./lib/Target/CellSPU/MCTargetDesc/LLVMBuild.txt ----------*- Conf -*--===;
+;===- ./lib/Target/AArch646/Utils/LLVMBuild.txt ----------------*- Conf -*--===;
 ;
 ;                     The LLVM Compiler Infrastructure
 ;
@@ -17,7 +17,7 @@
 
 [component_0]
 type = Library
-name = CellSPUDesc
-parent = CellSPU
-required_libraries = CellSPUInfo MC
-add_to_library_groups = CellSPU
+name = AArch64Utils
+parent = AArch64
+required_libraries = Core Support
+add_to_library_groups = AArch64
diff --git a/lib/Target/AArch64/Utils/Makefile b/lib/Target/AArch64/Utils/Makefile
new file mode 100644
index 000000000000..0f4a64527123
--- /dev/null
+++ b/lib/Target/AArch64/Utils/Makefile
@@ -0,0 +1,15 @@
+##===- lib/Target/AArch64/Utils/Makefile -------------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+LEVEL = ../../../..
+LIBRARYNAME = LLVMAArch64Utils
+
+# Hack: we need to include 'main' AArch64 target directory to grab private headers
+#CPP.Flags += -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/Target/ARM/A15SDOptimizer.cpp b/lib/Target/ARM/A15SDOptimizer.cpp
new file mode 100644
index 000000000000..f0d4dbe2bfb3
--- /dev/null
+++ b/lib/Target/ARM/A15SDOptimizer.cpp
@@ -0,0 +1,704 @@
+//=== A15SDOptimizerPass.cpp - Optimize DPR and SPR register accesses on A15==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// The Cortex-A15 processor employs a tracking scheme in its register renaming
+// in order to process each instruction's micro-ops speculatively and
+// out-of-order with appropriate forwarding. The ARM architecture allows VFP
+// instructions to read and write 32-bit S-registers.  Each S-register
+// corresponds to one half (upper or lower) of an overlaid 64-bit D-register.
+//
+// There are several instruction patterns which can be used to provide this
+// capability which can provide higher performance than other, potentially more
+// direct patterns, specifically around when one micro-op reads a D-register
+// operand that has recently been written as one or more S-register results.
+//
+// This file defines a pre-regalloc pass which looks for SPR producers which
+// are going to be used by a DPR (or QPR) consumers and creates the more
+// optimized access pattern.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "a15-sd-optimizer"
+#include "ARM.h"
+#include "ARMBaseInstrInfo.h"
+#include "ARMSubtarget.h"
+#include "ARMISelLowering.h"
+#include "ARMTargetMachine.h"
+
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetRegisterInfo.h"
+
+#include <set>
+
+using namespace llvm;
+
+namespace {
+  struct A15SDOptimizer : public MachineFunctionPass {
+    static char ID;
+    A15SDOptimizer() : MachineFunctionPass(ID) {}
+
+    virtual bool runOnMachineFunction(MachineFunction &Fn);
+
+    virtual const char *getPassName() const {
+      return "ARM A15 S->D optimizer";
+    }
+
+  private:
+    const ARMBaseInstrInfo *TII;
+    const TargetRegisterInfo *TRI;
+    MachineRegisterInfo *MRI;
+
+    bool runOnInstruction(MachineInstr *MI);
+
+    //
+    // Instruction builder helpers
+    //
+    unsigned createDupLane(MachineBasicBlock &MBB,
+                           MachineBasicBlock::iterator InsertBefore,
+                           DebugLoc DL,
+                           unsigned Reg, unsigned Lane,
+                           bool QPR=false);
+
+    unsigned createExtractSubreg(MachineBasicBlock &MBB,
+                                 MachineBasicBlock::iterator InsertBefore,
+                                 DebugLoc DL,
+                                 unsigned DReg, unsigned Lane,
+                                 const TargetRegisterClass *TRC);
+
+    unsigned createVExt(MachineBasicBlock &MBB,
+                        MachineBasicBlock::iterator InsertBefore,
+                        DebugLoc DL,
+                        unsigned Ssub0, unsigned Ssub1);
+
+    unsigned createRegSequence(MachineBasicBlock &MBB,
+                               MachineBasicBlock::iterator InsertBefore,
+                               DebugLoc DL,
+                               unsigned Reg1, unsigned Reg2);
+
+    unsigned createInsertSubreg(MachineBasicBlock &MBB,
+                                MachineBasicBlock::iterator InsertBefore,
+                                DebugLoc DL, unsigned DReg, unsigned Lane,
+                                unsigned ToInsert);
+
+    unsigned createImplicitDef(MachineBasicBlock &MBB,
+                               MachineBasicBlock::iterator InsertBefore,
+                               DebugLoc DL);
+    
+    //
+    // Various property checkers
+    //
+    bool usesRegClass(MachineOperand &MO, const TargetRegisterClass *TRC);
+    bool hasPartialWrite(MachineInstr *MI);
+    SmallVector<unsigned, 8> getReadDPRs(MachineInstr *MI);
+    unsigned getDPRLaneFromSPR(unsigned SReg);
+
+    //
+    // Methods used for getting the definitions of partial registers
+    //
+
+    MachineInstr *elideCopies(MachineInstr *MI);
+    void elideCopiesAndPHIs(MachineInstr *MI,
+                            SmallVectorImpl<MachineInstr*> &Outs);
+
+    //
+    // Pattern optimization methods
+    //
+    unsigned optimizeAllLanesPattern(MachineInstr *MI, unsigned Reg);
+    unsigned optimizeSDPattern(MachineInstr *MI);
+    unsigned getPrefSPRLane(unsigned SReg);
+
+    //
+    // Sanitizing method - used to make sure if don't leave dead code around.
+    //
+    void eraseInstrWithNoUses(MachineInstr *MI);
+
+    //
+    // A map used to track the changes done by this pass.
+    //
+    std::map<MachineInstr*, unsigned> Replacements;
+    std::set<MachineInstr *> DeadInstr;
+  };
+  char A15SDOptimizer::ID = 0;
+} // end anonymous namespace
+
+// Returns true if this is a use of a SPR register.
+bool A15SDOptimizer::usesRegClass(MachineOperand &MO,
+                                  const TargetRegisterClass *TRC) {
+  if (!MO.isReg())
+    return false;
+  unsigned Reg = MO.getReg();
+
+  if (TargetRegisterInfo::isVirtualRegister(Reg))
+    return MRI->getRegClass(Reg)->hasSuperClassEq(TRC);
+  else
+    return TRC->contains(Reg);
+}
+
+unsigned A15SDOptimizer::getDPRLaneFromSPR(unsigned SReg) {
+  unsigned DReg = TRI->getMatchingSuperReg(SReg, ARM::ssub_1,
+                                           &ARM::DPRRegClass);
+  if (DReg != ARM::NoRegister) return ARM::ssub_1;
+  return ARM::ssub_0;
+}
+
+// Get the subreg type that is most likely to be coalesced
+// for an SPR register that will be used in VDUP32d pseudo.
+unsigned A15SDOptimizer::getPrefSPRLane(unsigned SReg) {
+  if (!TRI->isVirtualRegister(SReg))
+    return getDPRLaneFromSPR(SReg);
+
+  MachineInstr *MI = MRI->getVRegDef(SReg);
+  if (!MI) return ARM::ssub_0;
+  MachineOperand *MO = MI->findRegisterDefOperand(SReg);
+
+  assert(MO->isReg() && "Non register operand found!");
+  if (!MO) return ARM::ssub_0;
+
+  if (MI->isCopy() && usesRegClass(MI->getOperand(1),
+                                    &ARM::SPRRegClass)) {
+    SReg = MI->getOperand(1).getReg();
+  }
+
+  if (TargetRegisterInfo::isVirtualRegister(SReg)) {
+    if (MO->getSubReg() == ARM::ssub_1) return ARM::ssub_1;
+    return ARM::ssub_0;
+  }
+  return getDPRLaneFromSPR(SReg);
+}
+
+// MI is known to be dead. Figure out what instructions
+// are also made dead by this and mark them for removal.
+void A15SDOptimizer::eraseInstrWithNoUses(MachineInstr *MI) {
+  SmallVector<MachineInstr *, 8> Front;
+  DeadInstr.insert(MI);
+
+  DEBUG(dbgs() << "Deleting base instruction " << *MI << "\n");
+  Front.push_back(MI);
+
+  while (Front.size() != 0) {
+    MI = Front.back();
+    Front.pop_back();
+
+    // MI is already known to be dead. We need to see
+    // if other instructions can also be removed.
+    for (unsigned int i = 0; i < MI->getNumOperands(); ++i) {
+      MachineOperand &MO = MI->getOperand(i);
+      if ((!MO.isReg()) || (!MO.isUse()))
+        continue;
+      unsigned Reg = MO.getReg();
+      if (!TRI->isVirtualRegister(Reg))
+        continue;
+      MachineOperand *Op = MI->findRegisterDefOperand(Reg);
+
+      if (!Op)
+        continue;
+
+      MachineInstr *Def = Op->getParent();
+
+      // We don't need to do anything if we have already marked
+      // this instruction as being dead.
+      if (DeadInstr.find(Def) != DeadInstr.end())
+        continue;
+
+      // Check if all the uses of this instruction are marked as
+      // dead. If so, we can also mark this instruction as being
+      // dead.
+      bool IsDead = true;
+      for (unsigned int j = 0; j < Def->getNumOperands(); ++j) {
+        MachineOperand &MODef = Def->getOperand(j);
+        if ((!MODef.isReg()) || (!MODef.isDef()))
+          continue;
+        unsigned DefReg = MODef.getReg();
+        if (!TRI->isVirtualRegister(DefReg)) {
+          IsDead = false;
+          break;
+        }
+        for (MachineRegisterInfo::use_iterator II = MRI->use_begin(Reg),
+                            EE = MRI->use_end();
+                            II != EE; ++II) {
+          // We don't care about self references.
+          if (&*II == Def)
+            continue;
+          if (DeadInstr.find(&*II) == DeadInstr.end()) {
+            IsDead = false;
+            break;
+          }
+        }
+      }
+
+      if (!IsDead) continue;
+
+      DEBUG(dbgs() << "Deleting instruction " << *Def << "\n");
+      DeadInstr.insert(Def);
+    }
+  }
+}
+
+// Creates the more optimized patterns and generally does all the code
+// transformations in this pass.
+unsigned A15SDOptimizer::optimizeSDPattern(MachineInstr *MI) {
+  if (MI->isCopy()) {
+    return optimizeAllLanesPattern(MI, MI->getOperand(1).getReg());
+  }
+
+  if (MI->isInsertSubreg()) {
+    unsigned DPRReg = MI->getOperand(1).getReg();
+    unsigned SPRReg = MI->getOperand(2).getReg();
+
+    if (TRI->isVirtualRegister(DPRReg) && TRI->isVirtualRegister(SPRReg)) {
+      MachineInstr *DPRMI = MRI->getVRegDef(MI->getOperand(1).getReg());
+      MachineInstr *SPRMI = MRI->getVRegDef(MI->getOperand(2).getReg());
+
+      if (DPRMI && SPRMI) {
+        // See if the first operand of this insert_subreg is IMPLICIT_DEF
+        MachineInstr *ECDef = elideCopies(DPRMI);
+        if (ECDef != 0 && ECDef->isImplicitDef()) {
+          // Another corner case - if we're inserting something that is purely
+          // a subreg copy of a DPR, just use that DPR.
+
+          MachineInstr *EC = elideCopies(SPRMI);
+          // Is it a subreg copy of ssub_0?
+          if (EC && EC->isCopy() &&
+              EC->getOperand(1).getSubReg() == ARM::ssub_0) {
+            DEBUG(dbgs() << "Found a subreg copy: " << *SPRMI);
+
+            // Find the thing we're subreg copying out of - is it of the same
+            // regclass as DPRMI? (i.e. a DPR or QPR).
+            unsigned FullReg = SPRMI->getOperand(1).getReg();
+            const TargetRegisterClass *TRC =
+              MRI->getRegClass(MI->getOperand(1).getReg());
+            if (TRC->hasSuperClassEq(MRI->getRegClass(FullReg))) {
+              DEBUG(dbgs() << "Subreg copy is compatible - returning ");
+              DEBUG(dbgs() << PrintReg(FullReg) << "\n");
+              eraseInstrWithNoUses(MI);
+              return FullReg;
+            }
+          }
+
+          return optimizeAllLanesPattern(MI, MI->getOperand(2).getReg());
+        }
+      }
+    }
+    return optimizeAllLanesPattern(MI, MI->getOperand(0).getReg());
+  }
+
+  if (MI->isRegSequence() && usesRegClass(MI->getOperand(1),
+                                          &ARM::SPRRegClass)) {
+    // See if all bar one of the operands are IMPLICIT_DEF and insert the
+    // optimizer pattern accordingly.
+    unsigned NumImplicit = 0, NumTotal = 0;
+    unsigned NonImplicitReg = ~0U;
+
+    for (unsigned I = 1; I < MI->getNumExplicitOperands(); ++I) {
+      if (!MI->getOperand(I).isReg())
+        continue;
+      ++NumTotal;
+      unsigned OpReg = MI->getOperand(I).getReg();
+
+      if (!TRI->isVirtualRegister(OpReg))
+        break;
+
+      MachineInstr *Def = MRI->getVRegDef(OpReg);
+      if (!Def)
+        break;
+      if (Def->isImplicitDef())
+        ++NumImplicit;
+      else
+        NonImplicitReg = MI->getOperand(I).getReg();
+    }
+
+    if (NumImplicit == NumTotal - 1)
+      return optimizeAllLanesPattern(MI, NonImplicitReg);
+    else
+      return optimizeAllLanesPattern(MI, MI->getOperand(0).getReg());
+  }
+
+  assert(0 && "Unhandled update pattern!");
+  return 0;
+}
+
+// Return true if this MachineInstr inserts a scalar (SPR) value into
+// a D or Q register.
+bool A15SDOptimizer::hasPartialWrite(MachineInstr *MI) {
+  // The only way we can do a partial register update is through a COPY,
+  // INSERT_SUBREG or REG_SEQUENCE.
+  if (MI->isCopy() && usesRegClass(MI->getOperand(1), &ARM::SPRRegClass))
+    return true;
+
+  if (MI->isInsertSubreg() && usesRegClass(MI->getOperand(2),
+                                           &ARM::SPRRegClass))
+    return true;
+
+  if (MI->isRegSequence() && usesRegClass(MI->getOperand(1), &ARM::SPRRegClass))
+    return true;
+
+  return false;
+}
+
+// Looks through full copies to get the instruction that defines the input
+// operand for MI.
+MachineInstr *A15SDOptimizer::elideCopies(MachineInstr *MI) {
+  if (!MI->isFullCopy())
+    return MI;
+  if (!TRI->isVirtualRegister(MI->getOperand(1).getReg()))
+    return NULL;
+  MachineInstr *Def = MRI->getVRegDef(MI->getOperand(1).getReg());
+  if (!Def)
+    return NULL;
+  return elideCopies(Def);
+}
+
+// Look through full copies and PHIs to get the set of non-copy MachineInstrs
+// that can produce MI.
+void A15SDOptimizer::elideCopiesAndPHIs(MachineInstr *MI,
+                                        SmallVectorImpl<MachineInstr*> &Outs) {
+   // Looking through PHIs may create loops so we need to track what
+   // instructions we have visited before.
+   std::set<MachineInstr *> Reached;
+   SmallVector<MachineInstr *, 8> Front;
+   Front.push_back(MI);
+   while (Front.size() != 0) {
+     MI = Front.back();
+     Front.pop_back();
+
+     // If we have already explored this MachineInstr, ignore it.
+     if (Reached.find(MI) != Reached.end())
+       continue;
+     Reached.insert(MI);
+     if (MI->isPHI()) {
+       for (unsigned I = 1, E = MI->getNumOperands(); I != E; I += 2) {
+         unsigned Reg = MI->getOperand(I).getReg();
+         if (!TRI->isVirtualRegister(Reg)) {
+           continue;
+         }
+         MachineInstr *NewMI = MRI->getVRegDef(Reg);
+         if (!NewMI)
+           continue;
+         Front.push_back(NewMI);
+       }
+     } else if (MI->isFullCopy()) {
+       if (!TRI->isVirtualRegister(MI->getOperand(1).getReg()))
+         continue;
+       MachineInstr *NewMI = MRI->getVRegDef(MI->getOperand(1).getReg());
+       if (!NewMI)
+         continue;
+       Front.push_back(NewMI);
+     } else {
+       DEBUG(dbgs() << "Found partial copy" << *MI <<"\n");
+       Outs.push_back(MI);
+     }
+   }
+}
+
+// Return the DPR virtual registers that are read by this machine instruction
+// (if any).
+SmallVector<unsigned, 8> A15SDOptimizer::getReadDPRs(MachineInstr *MI) {
+  if (MI->isCopyLike() || MI->isInsertSubreg() || MI->isRegSequence() ||
+      MI->isKill())
+    return SmallVector<unsigned, 8>();
+
+  SmallVector<unsigned, 8> Defs;
+  for (unsigned i = 0; i < MI->getNumOperands(); ++i) {
+    MachineOperand &MO = MI->getOperand(i);
+
+    if (!MO.isReg() || !MO.isUse())
+      continue;
+    if (!usesRegClass(MO, &ARM::DPRRegClass) &&
+        !usesRegClass(MO, &ARM::QPRRegClass))
+      continue;
+
+    Defs.push_back(MO.getReg());
+  }
+  return Defs;
+}
+
+// Creates a DPR register from an SPR one by using a VDUP.
+unsigned
+A15SDOptimizer::createDupLane(MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator InsertBefore,
+                              DebugLoc DL,
+                              unsigned Reg, unsigned Lane, bool QPR) {
+  unsigned Out = MRI->createVirtualRegister(QPR ? &ARM::QPRRegClass :
+                                                  &ARM::DPRRegClass);
+  AddDefaultPred(BuildMI(MBB,
+                         InsertBefore,
+                         DL,
+                         TII->get(QPR ? ARM::VDUPLN32q : ARM::VDUPLN32d),
+                         Out)
+                   .addReg(Reg)
+                   .addImm(Lane));
+ 
+  return Out;
+}
+
+// Creates a SPR register from a DPR by copying the value in lane 0.
+unsigned
+A15SDOptimizer::createExtractSubreg(MachineBasicBlock &MBB,
+                                    MachineBasicBlock::iterator InsertBefore,
+                                    DebugLoc DL,
+                                    unsigned DReg, unsigned Lane,
+                                    const TargetRegisterClass *TRC) {
+  unsigned Out = MRI->createVirtualRegister(TRC);
+  BuildMI(MBB,
+          InsertBefore,
+          DL,
+          TII->get(TargetOpcode::COPY), Out)
+    .addReg(DReg, 0, Lane);
+
+  return Out;
+}
+
+// Takes two SPR registers and creates a DPR by using a REG_SEQUENCE.
+unsigned
+A15SDOptimizer::createRegSequence(MachineBasicBlock &MBB,
+                                  MachineBasicBlock::iterator InsertBefore,
+                                  DebugLoc DL,
+                                  unsigned Reg1, unsigned Reg2) {
+  unsigned Out = MRI->createVirtualRegister(&ARM::QPRRegClass);
+  BuildMI(MBB,
+          InsertBefore,
+          DL,
+          TII->get(TargetOpcode::REG_SEQUENCE), Out)
+    .addReg(Reg1)
+    .addImm(ARM::dsub_0)
+    .addReg(Reg2)
+    .addImm(ARM::dsub_1);
+  return Out;
+}
+
+// Takes two DPR registers that have previously been VDUPed (Ssub0 and Ssub1)
+// and merges them into one DPR register.
+unsigned
+A15SDOptimizer::createVExt(MachineBasicBlock &MBB,
+                           MachineBasicBlock::iterator InsertBefore,
+                           DebugLoc DL,
+                           unsigned Ssub0, unsigned Ssub1) {
+  unsigned Out = MRI->createVirtualRegister(&ARM::DPRRegClass);
+  AddDefaultPred(BuildMI(MBB,
+                         InsertBefore,
+                         DL,
+                         TII->get(ARM::VEXTd32), Out)
+                   .addReg(Ssub0)
+                   .addReg(Ssub1)
+                   .addImm(1));
+  return Out;
+}
+
+unsigned
+A15SDOptimizer::createInsertSubreg(MachineBasicBlock &MBB,
+                                   MachineBasicBlock::iterator InsertBefore,
+                                   DebugLoc DL, unsigned DReg, unsigned Lane,
+                                   unsigned ToInsert) {
+  unsigned Out = MRI->createVirtualRegister(&ARM::DPR_VFP2RegClass);
+  BuildMI(MBB,
+          InsertBefore,
+          DL,
+          TII->get(TargetOpcode::INSERT_SUBREG), Out)
+    .addReg(DReg)
+    .addReg(ToInsert)
+    .addImm(Lane);
+
+  return Out;
+}
+
+unsigned
+A15SDOptimizer::createImplicitDef(MachineBasicBlock &MBB,
+                                  MachineBasicBlock::iterator InsertBefore,
+                                  DebugLoc DL) {
+  unsigned Out = MRI->createVirtualRegister(&ARM::DPRRegClass);
+  BuildMI(MBB,
+          InsertBefore,
+          DL,
+          TII->get(TargetOpcode::IMPLICIT_DEF), Out);
+  return Out;
+}
+
+// This function inserts instructions in order to optimize interactions between
+// SPR registers and DPR/QPR registers. It does so by performing VDUPs on all
+// lanes, and the using VEXT instructions to recompose the result.
+unsigned
+A15SDOptimizer::optimizeAllLanesPattern(MachineInstr *MI, unsigned Reg) {
+  MachineBasicBlock::iterator InsertPt(MI);
+  DebugLoc DL = MI->getDebugLoc();
+  MachineBasicBlock &MBB = *MI->getParent();
+  InsertPt++;
+  unsigned Out;
+
+  if (MRI->getRegClass(Reg)->hasSuperClassEq(&ARM::QPRRegClass)) {
+    unsigned DSub0 = createExtractSubreg(MBB, InsertPt, DL, Reg,
+                                         ARM::dsub_0, &ARM::DPRRegClass);
+    unsigned DSub1 = createExtractSubreg(MBB, InsertPt, DL, Reg,
+                                         ARM::dsub_1, &ARM::DPRRegClass);
+
+    unsigned Out1 = createDupLane(MBB, InsertPt, DL, DSub0, 0);
+    unsigned Out2 = createDupLane(MBB, InsertPt, DL, DSub0, 1);
+    Out = createVExt(MBB, InsertPt, DL, Out1, Out2);
+
+    unsigned Out3 = createDupLane(MBB, InsertPt, DL, DSub1, 0);
+    unsigned Out4 = createDupLane(MBB, InsertPt, DL, DSub1, 1);
+    Out2 = createVExt(MBB, InsertPt, DL, Out3, Out4);
+
+    Out = createRegSequence(MBB, InsertPt, DL, Out, Out2);
+
+  } else if (MRI->getRegClass(Reg)->hasSuperClassEq(&ARM::DPRRegClass)) {
+    unsigned Out1 = createDupLane(MBB, InsertPt, DL, Reg, 0);
+    unsigned Out2 = createDupLane(MBB, InsertPt, DL, Reg, 1);
+    Out = createVExt(MBB, InsertPt, DL, Out1, Out2);
+
+  } else {
+    assert(MRI->getRegClass(Reg)->hasSuperClassEq(&ARM::SPRRegClass) &&
+           "Found unexpected regclass!");
+
+    unsigned PrefLane = getPrefSPRLane(Reg);
+    unsigned Lane;
+    switch (PrefLane) {
+      case ARM::ssub_0: Lane = 0; break;
+      case ARM::ssub_1: Lane = 1; break;
+      default: llvm_unreachable("Unknown preferred lane!");
+    }
+
+    bool UsesQPR = usesRegClass(MI->getOperand(0), &ARM::QPRRegClass);
+
+    Out = createImplicitDef(MBB, InsertPt, DL);
+    Out = createInsertSubreg(MBB, InsertPt, DL, Out, PrefLane, Reg);
+    Out = createDupLane(MBB, InsertPt, DL, Out, Lane, UsesQPR);
+    eraseInstrWithNoUses(MI);
+  }
+  return Out;
+}
+
+bool A15SDOptimizer::runOnInstruction(MachineInstr *MI) {
+  // We look for instructions that write S registers that are then read as
+  // D/Q registers. These can only be caused by COPY, INSERT_SUBREG and
+  // REG_SEQUENCE pseudos that insert an SPR value into a DPR register or
+  // merge two SPR values to form a DPR register.  In order avoid false
+  // positives we make sure that there is an SPR producer so we look past
+  // COPY and PHI nodes to find it.
+  //
+  // The best code pattern for when an SPR producer is going to be used by a
+  // DPR or QPR consumer depends on whether the other lanes of the
+  // corresponding DPR/QPR are currently defined.
+  //
+  // We can handle these efficiently, depending on the type of
+  // pseudo-instruction that is producing the pattern
+  //
+  //   * COPY:          * VDUP all lanes and merge the results together
+  //                      using VEXTs.
+  //
+  //   * INSERT_SUBREG: * If the SPR value was originally in another DPR/QPR
+  //                      lane, and the other lane(s) of the DPR/QPR register
+  //                      that we are inserting in are undefined, use the
+  //                      original DPR/QPR value. 
+  //                    * Otherwise, fall back on the same stategy as COPY.
+  //
+  //   * REG_SEQUENCE:  * If all except one of the input operands are
+  //                      IMPLICIT_DEFs, insert the VDUP pattern for just the
+  //                      defined input operand
+  //                    * Otherwise, fall back on the same stategy as COPY.
+  //
+
+  // First, get all the reads of D-registers done by this instruction.
+  SmallVector<unsigned, 8> Defs = getReadDPRs(MI);
+  bool Modified = false;
+
+  for (SmallVector<unsigned, 8>::iterator I = Defs.begin(), E = Defs.end();
+     I != E; ++I) {
+    // Follow the def-use chain for this DPR through COPYs, and also through
+    // PHIs (which are essentially multi-way COPYs). It is because of PHIs that
+    // we can end up with multiple defs of this DPR.
+
+    SmallVector<MachineInstr *, 8> DefSrcs;
+    if (!TRI->isVirtualRegister(*I))
+      continue;
+    MachineInstr *Def = MRI->getVRegDef(*I);
+    if (!Def)
+      continue;
+
+    elideCopiesAndPHIs(Def, DefSrcs);
+
+    for (SmallVector<MachineInstr*, 8>::iterator II = DefSrcs.begin(),
+      EE = DefSrcs.end(); II != EE; ++II) {
+      MachineInstr *MI = *II;
+
+      // If we've already analyzed and replaced this operand, don't do
+      // anything.
+      if (Replacements.find(MI) != Replacements.end())
+        continue;
+
+      // Now, work out if the instruction causes a SPR->DPR dependency.
+      if (!hasPartialWrite(MI))
+        continue;
+
+      // Collect all the uses of this MI's DPR def for updating later.
+      SmallVector<MachineOperand*, 8> Uses;
+      unsigned DPRDefReg = MI->getOperand(0).getReg();
+      for (MachineRegisterInfo::use_iterator I = MRI->use_begin(DPRDefReg),
+             E = MRI->use_end(); I != E; ++I)
+        Uses.push_back(&I.getOperand());
+
+      // We can optimize this.
+      unsigned NewReg = optimizeSDPattern(MI);
+
+      if (NewReg != 0) {
+        Modified = true;
+        for (SmallVector<MachineOperand*, 8>::const_iterator I = Uses.begin(),
+               E = Uses.end(); I != E; ++I) {
+          DEBUG(dbgs() << "Replacing operand "
+                       << **I << " with "
+                       << PrintReg(NewReg) << "\n");
+          (*I)->substVirtReg(NewReg, 0, *TRI);
+        }
+      }
+      Replacements[MI] = NewReg;
+    }
+  }
+  return Modified;
+}
+
+bool A15SDOptimizer::runOnMachineFunction(MachineFunction &Fn) {
+  TII = static_cast<const ARMBaseInstrInfo*>(Fn.getTarget().getInstrInfo());
+  TRI = Fn.getTarget().getRegisterInfo();
+  MRI = &Fn.getRegInfo();
+  bool Modified = false;
+
+  DEBUG(dbgs() << "Running on function " << Fn.getName()<< "\n");
+
+  DeadInstr.clear();
+  Replacements.clear();
+
+  for (MachineFunction::iterator MFI = Fn.begin(), E = Fn.end(); MFI != E;
+       ++MFI) {
+
+    for (MachineBasicBlock::iterator MI = MFI->begin(), ME = MFI->end();
+      MI != ME;) {
+      Modified |= runOnInstruction(MI++);
+    }
+ 
+  }
+
+  for (std::set<MachineInstr *>::iterator I = DeadInstr.begin(),
+                                            E = DeadInstr.end();
+                                            I != E; ++I) {
+    (*I)->eraseFromParent();
+  }
+
+  return Modified;
+}
+
+FunctionPass *llvm::createA15SDOptimizerPass() {
+  return new A15SDOptimizer();
+}
diff --git a/lib/Target/ARM/ARM.h b/lib/Target/ARM/ARM.h
index 1446bbbb8e7c..80e5f37eb086 100644
--- a/lib/Target/ARM/ARM.h
+++ b/lib/Target/ARM/ARM.h
@@ -35,6 +35,7 @@ FunctionPass *createARMISelDag(ARMBaseTargetMachine &TM,
 FunctionPass *createARMJITCodeEmitterPass(ARMBaseTargetMachine &TM,
                                           JITCodeEmitter &JCE);
 
+FunctionPass *createA15SDOptimizerPass();
 FunctionPass *createARMLoadStoreOptimizationPass(bool PreAlloc = false);
 FunctionPass *createARMExpandPseudoPass();
 FunctionPass *createARMGlobalBaseRegPass();
@@ -44,6 +45,9 @@ FunctionPass *createMLxExpansionPass();
 FunctionPass *createThumb2ITBlockPass();
 FunctionPass *createThumb2SizeReductionPass();
 
+/// \brief Creates an ARM-specific Target Transformation Info pass.
+ImmutablePass *createARMTargetTransformInfoPass(const ARMBaseTargetMachine *TM);
+
 void LowerARMMachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI,
                                   ARMAsmPrinter &AP);
 
diff --git a/lib/Target/ARM/ARM.td b/lib/Target/ARM/ARM.td
index 23974ad9052c..68380847a022 100644
--- a/lib/Target/ARM/ARM.td
+++ b/lib/Target/ARM/ARM.td
@@ -89,6 +89,10 @@ def FeatureAvoidPartialCPSR : SubtargetFeature<"avoid-partial-cpsr",
                                                "AvoidCPSRPartialUpdate", "true",
                                  "Avoid CPSR partial update for OOO execution">;
 
+def FeatureAvoidMOVsShOp : SubtargetFeature<"avoid-movs-shop",
+                                            "AvoidMOVsShifterOperand", "true",
+                                "Avoid movs instructions with shifter operand">;
+
 // Some processors perform return stack prediction. CodeGen should avoid issue
 // "normal" call instructions to callees which do not return.
 def FeatureHasRAS : SubtargetFeature<"ras", "HasRAS", "true",
@@ -106,6 +110,11 @@ def FeatureMP : SubtargetFeature<"mp", "HasMPExtension", "true",
 def FeatureMClass : SubtargetFeature<"mclass", "IsMClass", "true",
                                      "Is microcontroller profile ('M' series)">;
 
+// Special TRAP encoding for NaCl, which looks like a TRAP in Thumb too.
+// See ARMInstrInfo.td for details.
+def FeatureNaClTrap : SubtargetFeature<"nacl-trap", "UseNaClTrap", "true",
+                                       "NaCl trap">;
+
 // ARM ISAs.
 def HasV4TOps   : SubtargetFeature<"v4t", "HasV4TOps", "true",
                                    "Support ARM v4T instructions">;
@@ -132,11 +141,14 @@ def HasV7Ops    : SubtargetFeature<"v7", "HasV7Ops", "true",
 include "ARMSchedule.td"
 
 // ARM processor families.
+def ProcA5      : SubtargetFeature<"a5", "ARMProcFamily", "CortexA5",
+                                   "Cortex-A5 ARM processors",
+                                   [FeatureSlowFPBrcc, FeatureHasSlowFPVMLx,
+                                    FeatureVMLxForwarding, FeatureT2XtPk]>;
 def ProcA8      : SubtargetFeature<"a8", "ARMProcFamily", "CortexA8",
                                    "Cortex-A8 ARM processors",
-                                   [FeatureSlowFPBrcc, FeatureNEONForFP,
-                                    FeatureHasSlowFPVMLx, FeatureVMLxForwarding,
-                                    FeatureT2XtPk]>;
+                                   [FeatureSlowFPBrcc, FeatureHasSlowFPVMLx,
+                                    FeatureVMLxForwarding, FeatureT2XtPk]>;
 def ProcA9      : SubtargetFeature<"a9", "ARMProcFamily", "CortexA9",
                                    "Cortex-A9 ARM processors",
                                    [FeatureVMLxForwarding,
@@ -147,6 +159,7 @@ def ProcSwift   : SubtargetFeature<"swift", "ARMProcFamily", "Swift",
                                    [FeatureNEONForFP, FeatureT2XtPk,
                                     FeatureVFP4, FeatureMP, FeatureHWDiv,
                                     FeatureHWDivARM, FeatureAvoidPartialCPSR,
+                                    FeatureAvoidMOVsShOp,
                                     FeatureHasSlowFPVMLx]>;
 
 // FIXME: It has not been determined if A15 has these features.
@@ -154,6 +167,12 @@ def ProcA15      : SubtargetFeature<"a15", "ARMProcFamily", "CortexA15",
                                    "Cortex-A15 ARM processors",
                                    [FeatureT2XtPk, FeatureFP16,
                                     FeatureAvoidPartialCPSR]>;
+def ProcR5      : SubtargetFeature<"r5", "ARMProcFamily", "CortexR5",
+                                   "Cortex-R5 ARM processors",
+                                   [FeatureSlowFPBrcc, FeatureHWDivARM,
+                                    FeatureHasSlowFPVMLx,
+                                    FeatureAvoidPartialCPSR,
+                                    FeatureT2XtPk]>;
 
 class ProcNoItin<string Name, list<SubtargetFeature> Features>
  : Processor<Name, NoItineraries, Features>;
@@ -219,6 +238,11 @@ def : Processor<"arm1156t2f-s",     ARMV6Itineraries, [HasV6T2Ops, FeatureVFP2,
                                                        FeatureDSPThumb2]>;
 
 // V7a Processors.
+// FIXME: A5 has currently the same Schedule model as A8
+def : ProcessorModel<"cortex-a5",   CortexA8Model,
+                                    [ProcA5, HasV7Ops, FeatureNEON, FeatureDB,
+                                     FeatureVFP4, FeatureDSPThumb2,
+                                     FeatureHasRAS]>;
 def : ProcessorModel<"cortex-a8",   CortexA8Model,
                                     [ProcA8, HasV7Ops, FeatureNEON, FeatureDB,
                                      FeatureDSPThumb2, FeatureHasRAS]>;
@@ -233,6 +257,11 @@ def : ProcessorModel<"cortex-a9-mp", CortexA9Model,
 def : ProcessorModel<"cortex-a15",   CortexA9Model,
                                     [ProcA15, HasV7Ops, FeatureNEON, FeatureDB,
                                      FeatureDSPThumb2, FeatureHasRAS]>;
+// FIXME: R5 has currently the same ProcessorModel as A8.
+def : ProcessorModel<"cortex-r5",   CortexA8Model,
+                                    [ProcR5, HasV7Ops, FeatureDB,
+                                     FeatureVFP3, FeatureDSPThumb2,
+                                     FeatureHasRAS]>;
 
 // V7M Processors.
 def : ProcNoItin<"cortex-m3",       [HasV7Ops,
diff --git a/lib/Target/ARM/ARMAsmPrinter.cpp b/lib/Target/ARM/ARMAsmPrinter.cpp
index d439d1d7cb7e..13ec2087938a 100644
--- a/lib/Target/ARM/ARMAsmPrinter.cpp
+++ b/lib/Target/ARM/ARMAsmPrinter.cpp
@@ -25,30 +25,33 @@
 #include "MCTargetDesc/ARMMCExpr.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/SmallString.h"
-#include "llvm/Constants.h"
-#include "llvm/DebugInfo.h"
-#include "llvm/Module.h"
-#include "llvm/Type.h"
 #include "llvm/Assembly/Writer.h"
-#include "llvm/CodeGen/MachineModuleInfoImpls.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
+#include "llvm/CodeGen/MachineModuleInfoImpls.h"
+#include "llvm/DebugInfo.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Type.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCAssembler.h"
 #include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCELFStreamer.h"
 #include "llvm/MC/MCInst.h"
-#include "llvm/MC/MCSectionMachO.h"
+#include "llvm/MC/MCInstBuilder.h"
 #include "llvm/MC/MCObjectStreamer.h"
+#include "llvm/MC/MCSectionMachO.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/Target/Mangler.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/ELF.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/Mangler.h"
+#include "llvm/Target/TargetMachine.h"
 #include <cctype>
 using namespace llvm;
 
@@ -181,7 +184,7 @@ namespace {
       const size_t TagHeaderSize = 1 + 4;
 
       Streamer.EmitIntValue(VendorHeaderSize + TagHeaderSize + ContentsSize, 4);
-      Streamer.EmitBytes(CurrentVendor, 0);
+      Streamer.EmitBytes(CurrentVendor);
       Streamer.EmitIntValue(0, 1); // '\0'
 
       Streamer.EmitIntValue(ARMBuildAttrs::File, 1);
@@ -191,14 +194,14 @@ namespace {
       // emit each field as its type (ULEB or String)
       for (unsigned int i=0; i<Contents.size(); ++i) {
         AttributeItemType item = Contents[i];
-        Streamer.EmitULEB128IntValue(item.Tag, 0);
+        Streamer.EmitULEB128IntValue(item.Tag);
         switch (item.Type) {
         default: llvm_unreachable("Invalid attribute type");
         case AttributeItemType::NumericAttribute:
-          Streamer.EmitULEB128IntValue(item.IntValue, 0);
+          Streamer.EmitULEB128IntValue(item.IntValue);
           break;
         case AttributeItemType::TextAttribute:
-          Streamer.EmitBytes(item.StringValue.upper(), 0);
+          Streamer.EmitBytes(item.StringValue.upper());
           Streamer.EmitIntValue(0, 1); // '\0'
           break;
         }
@@ -339,6 +342,11 @@ void ARMAsmPrinter::printOperand(const MachineInstr *MI, int OpNum,
     unsigned Reg = MO.getReg();
     assert(TargetRegisterInfo::isPhysicalRegister(Reg));
     assert(!MO.getSubReg() && "Subregs should be eliminated!");
+    if(ARM::GPRPairRegClass.contains(Reg)) {
+      const MachineFunction &MF = *MI->getParent()->getParent();
+      const TargetRegisterInfo *TRI = MF.getTarget().getRegisterInfo();
+      Reg = TRI->getSubReg(Reg, ARM::gsub_0);
+    }
     O << ARMInstPrinter::getRegisterName(Reg);
     break;
   }
@@ -398,7 +406,7 @@ GetARMJTIPICJumpTableLabel2(unsigned uid, unsigned uid2) const {
 }
 
 
-MCSymbol *ARMAsmPrinter::GetARMSJLJEHLabel(void) const {
+MCSymbol *ARMAsmPrinter::GetARMSJLJEHLabel() const {
   SmallString<60> Name;
   raw_svector_ostream(Name) << MAI->getPrivateGlobalPrefix() << "SJLJEH"
     << getFunctionNumber();
@@ -527,14 +535,12 @@ bool ARMAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNum,
       const MachineOperand &MO = MI->getOperand(OpNum);
       if (!MO.isReg())
         return true;
-      const TargetRegisterClass &RC = ARM::GPRRegClass;
       const MachineFunction &MF = *MI->getParent()->getParent();
       const TargetRegisterInfo *TRI = MF.getTarget().getRegisterInfo();
-
-      unsigned RegIdx = TRI->getEncodingValue(MO.getReg());
-      RegIdx |= 1; //The odd register is also the higher-numbered one of a pair.
-
-      unsigned Reg = RC.getRegister(RegIdx);
+      unsigned Reg = MO.getReg();
+      if(!ARM::GPRPairRegClass.contains(Reg))
+        return false;
+      Reg = TRI->getSubReg(Reg, ARM::gsub_1);
       O << ARMInstPrinter::getRegisterName(Reg);
       return false;
     }
@@ -656,7 +662,7 @@ void ARMAsmPrinter::EmitEndOfAsmFile(Module &M) {
 
         if (MCSym.getInt())
           // External to current translation unit.
-          OutStreamer.EmitIntValue(0, 4/*size*/, 0/*addrspace*/);
+          OutStreamer.EmitIntValue(0, 4/*size*/);
         else
           // Internal to current translation unit.
           //
@@ -666,7 +672,7 @@ void ARMAsmPrinter::EmitEndOfAsmFile(Module &M) {
           // We need to fill in the value for the NLP in those cases.
           OutStreamer.EmitValue(MCSymbolRefExpr::Create(MCSym.getPointer(),
                                                         OutContext),
-                                4/*size*/, 0/*addrspace*/);
+                                4/*size*/);
       }
 
       Stubs.clear();
@@ -684,7 +690,7 @@ void ARMAsmPrinter::EmitEndOfAsmFile(Module &M) {
         OutStreamer.EmitValue(MCSymbolRefExpr::
                               Create(Stubs[i].second.getPointer(),
                                      OutContext),
-                              4/*size*/, 0/*addrspace*/);
+                              4/*size*/);
       }
 
       Stubs.clear();
@@ -698,6 +704,11 @@ void ARMAsmPrinter::EmitEndOfAsmFile(Module &M) {
     // generates code that does this, it is always safe to set.
     OutStreamer.EmitAssemblerFlag(MCAF_SubsectionsViaSymbols);
   }
+  // FIXME: This should eventually end up somewhere else where more
+  // intelligent flag decisions can be made. For now we are just maintaining
+  // the status quo for ARM and setting EF_ARM_EABI_VER5 as the default.
+  if (MCELFStreamer *MES = dyn_cast<MCELFStreamer>(&OutStreamer))
+    MES->getAssembler().setELFHeaderEFlags(ELF::EF_ARM_EABI_VER5);
 }
 
 //===----------------------------------------------------------------------===//
@@ -1051,12 +1062,10 @@ void ARMAsmPrinter::EmitJump2Table(const MachineInstr *MI) {
                                                       OutContext);
     // If this isn't a TBB or TBH, the entries are direct branch instructions.
     if (OffsetWidth == 4) {
-      MCInst BrInst;
-      BrInst.setOpcode(ARM::t2B);
-      BrInst.addOperand(MCOperand::CreateExpr(MBBSymbolExpr));
-      BrInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      BrInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(BrInst);
+      OutStreamer.EmitInstruction(MCInstBuilder(ARM::t2B)
+        .addExpr(MBBSymbolExpr)
+        .addImm(ARMCC::AL)
+        .addReg(0));
       continue;
     }
     // Otherwise it's an offset from the dispatch instruction. Construct an
@@ -1100,18 +1109,6 @@ void ARMAsmPrinter::PrintDebugValueComment(const MachineInstr *MI,
   printOperand(MI, NOps-2, OS);
 }
 
-static void populateADROperands(MCInst &Inst, unsigned Dest,
-                                const MCSymbol *Label,
-                                unsigned pred, unsigned ccreg,
-                                MCContext &Ctx) {
-  const MCExpr *SymbolExpr = MCSymbolRefExpr::Create(Label, Ctx);
-  Inst.addOperand(MCOperand::CreateReg(Dest));
-  Inst.addOperand(MCOperand::CreateExpr(SymbolExpr));
-  // Add predicate operands.
-  Inst.addOperand(MCOperand::CreateImm(pred));
-  Inst.addOperand(MCOperand::CreateReg(ccreg));
-}
-
 void ARMAsmPrinter::EmitUnwindingInstruction(const MachineInstr *MI) {
   assert(MI->getFlag(MachineInstr::FrameSetup) &&
       "Only instruction which are involved into frame setup code are allowed");
@@ -1288,129 +1285,104 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
   case ARM::tLEApcrel:
   case ARM::t2LEApcrel: {
     // FIXME: Need to also handle globals and externals
-    MCInst TmpInst;
-    TmpInst.setOpcode(MI->getOpcode() == ARM::t2LEApcrel ? ARM::t2ADR
-                      : (MI->getOpcode() == ARM::tLEApcrel ? ARM::tADR
-                         : ARM::ADR));
-    populateADROperands(TmpInst, MI->getOperand(0).getReg(),
-                        GetCPISymbol(MI->getOperand(1).getIndex()),
-                        MI->getOperand(2).getImm(), MI->getOperand(3).getReg(),
-                        OutContext);
-    OutStreamer.EmitInstruction(TmpInst);
+    MCSymbol *CPISymbol = GetCPISymbol(MI->getOperand(1).getIndex());
+    OutStreamer.EmitInstruction(MCInstBuilder(MI->getOpcode() ==
+                                              ARM::t2LEApcrel ? ARM::t2ADR
+                  : (MI->getOpcode() == ARM::tLEApcrel ? ARM::tADR
+                     : ARM::ADR))
+      .addReg(MI->getOperand(0).getReg())
+      .addExpr(MCSymbolRefExpr::Create(CPISymbol, OutContext))
+      // Add predicate operands.
+      .addImm(MI->getOperand(2).getImm())
+      .addReg(MI->getOperand(3).getReg()));
     return;
   }
   case ARM::LEApcrelJT:
   case ARM::tLEApcrelJT:
   case ARM::t2LEApcrelJT: {
-    MCInst TmpInst;
-    TmpInst.setOpcode(MI->getOpcode() == ARM::t2LEApcrelJT ? ARM::t2ADR
-                      : (MI->getOpcode() == ARM::tLEApcrelJT ? ARM::tADR
-                         : ARM::ADR));
-    populateADROperands(TmpInst, MI->getOperand(0).getReg(),
-                      GetARMJTIPICJumpTableLabel2(MI->getOperand(1).getIndex(),
-                                                  MI->getOperand(2).getImm()),
-                      MI->getOperand(3).getImm(), MI->getOperand(4).getReg(),
-                      OutContext);
-    OutStreamer.EmitInstruction(TmpInst);
+    MCSymbol *JTIPICSymbol =
+      GetARMJTIPICJumpTableLabel2(MI->getOperand(1).getIndex(),
+                                  MI->getOperand(2).getImm());
+    OutStreamer.EmitInstruction(MCInstBuilder(MI->getOpcode() ==
+                                              ARM::t2LEApcrelJT ? ARM::t2ADR
+                  : (MI->getOpcode() == ARM::tLEApcrelJT ? ARM::tADR
+                     : ARM::ADR))
+      .addReg(MI->getOperand(0).getReg())
+      .addExpr(MCSymbolRefExpr::Create(JTIPICSymbol, OutContext))
+      // Add predicate operands.
+      .addImm(MI->getOperand(3).getImm())
+      .addReg(MI->getOperand(4).getReg()));
     return;
   }
   // Darwin call instructions are just normal call instructions with different
   // clobber semantics (they clobber R9).
   case ARM::BX_CALL: {
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::MOVr);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::LR));
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::MOVr)
+      .addReg(ARM::LR)
+      .addReg(ARM::PC)
       // Add predicate operands.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
+      .addImm(ARMCC::AL)
+      .addReg(0)
       // Add 's' bit operand (always reg0 for this)
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::BX);
-      TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::BX)
+      .addReg(MI->getOperand(0).getReg()));
     return;
   }
   case ARM::tBX_CALL: {
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tMOVr);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::LR));
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tMOVr)
+      .addReg(ARM::LR)
+      .addReg(ARM::PC)
       // Add predicate operands.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tBX);
-      TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tBX)
+      .addReg(MI->getOperand(0).getReg())
       // Add predicate operands.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
+      .addImm(ARMCC::AL)
+      .addReg(0));
     return;
   }
   case ARM::BMOVPCRX_CALL: {
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::MOVr);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::LR));
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::MOVr)
+      .addReg(ARM::LR)
+      .addReg(ARM::PC)
       // Add predicate operands.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
+      .addImm(ARMCC::AL)
+      .addReg(0)
       // Add 's' bit operand (always reg0 for this)
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::MOVr);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
-      TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::MOVr)
+      .addReg(ARM::PC)
+      .addReg(MI->getOperand(0).getReg())
       // Add predicate operands.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
+      .addImm(ARMCC::AL)
+      .addReg(0)
       // Add 's' bit operand (always reg0 for this)
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
+      .addReg(0));
     return;
   }
   case ARM::BMOVPCB_CALL: {
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::MOVr);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::LR));
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::MOVr)
+      .addReg(ARM::LR)
+      .addReg(ARM::PC)
       // Add predicate operands.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
+      .addImm(ARMCC::AL)
+      .addReg(0)
       // Add 's' bit operand (always reg0 for this)
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::Bcc);
-      const GlobalValue *GV = MI->getOperand(0).getGlobal();
-      MCSymbol *GVSym = Mang->getSymbol(GV);
-      const MCExpr *GVSymExpr = MCSymbolRefExpr::Create(GVSym, OutContext);
-      TmpInst.addOperand(MCOperand::CreateExpr(GVSymExpr));
+      .addReg(0));
+
+    const GlobalValue *GV = MI->getOperand(0).getGlobal();
+    MCSymbol *GVSym = Mang->getSymbol(GV);
+    const MCExpr *GVSymExpr = MCSymbolRefExpr::Create(GVSym, OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::Bcc)
+      .addExpr(GVSymExpr)
       // Add predicate operands.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
+      .addImm(ARMCC::AL)
+      .addReg(0));
     return;
   }
   case ARM::MOVi16_ga_pcrel:
@@ -1498,15 +1470,13 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
                           OutContext));
 
     // Form and emit the add.
-    MCInst AddInst;
-    AddInst.setOpcode(ARM::tADDhirr);
-    AddInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-    AddInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-    AddInst.addOperand(MCOperand::CreateReg(ARM::PC));
-    // Add predicate operands.
-    AddInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-    AddInst.addOperand(MCOperand::CreateReg(0));
-    OutStreamer.EmitInstruction(AddInst);
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tADDhirr)
+      .addReg(MI->getOperand(0).getReg())
+      .addReg(MI->getOperand(0).getReg())
+      .addReg(ARM::PC)
+      // Add predicate operands.
+      .addImm(ARMCC::AL)
+      .addReg(0));
     return;
   }
   case ARM::PICADD: {
@@ -1521,17 +1491,15 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
                           OutContext));
 
     // Form and emit the add.
-    MCInst AddInst;
-    AddInst.setOpcode(ARM::ADDrr);
-    AddInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-    AddInst.addOperand(MCOperand::CreateReg(ARM::PC));
-    AddInst.addOperand(MCOperand::CreateReg(MI->getOperand(1).getReg()));
-    // Add predicate operands.
-    AddInst.addOperand(MCOperand::CreateImm(MI->getOperand(3).getImm()));
-    AddInst.addOperand(MCOperand::CreateReg(MI->getOperand(4).getReg()));
-    // Add 's' bit operand (always reg0 for this)
-    AddInst.addOperand(MCOperand::CreateReg(0));
-    OutStreamer.EmitInstruction(AddInst);
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::ADDrr)
+      .addReg(MI->getOperand(0).getReg())
+      .addReg(ARM::PC)
+      .addReg(MI->getOperand(1).getReg())
+      // Add predicate operands.
+      .addImm(MI->getOperand(3).getImm())
+      .addReg(MI->getOperand(4).getReg())
+      // Add 's' bit operand (always reg0 for this)
+      .addReg(0));
     return;
   }
   case ARM::PICSTR:
@@ -1567,16 +1535,14 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     case ARM::PICLDRSB: Opcode = ARM::LDRSB; break;
     case ARM::PICLDRSH: Opcode = ARM::LDRSH; break;
     }
-    MCInst LdStInst;
-    LdStInst.setOpcode(Opcode);
-    LdStInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-    LdStInst.addOperand(MCOperand::CreateReg(ARM::PC));
-    LdStInst.addOperand(MCOperand::CreateReg(MI->getOperand(1).getReg()));
-    LdStInst.addOperand(MCOperand::CreateImm(0));
-    // Add predicate operands.
-    LdStInst.addOperand(MCOperand::CreateImm(MI->getOperand(3).getImm()));
-    LdStInst.addOperand(MCOperand::CreateReg(MI->getOperand(4).getReg()));
-    OutStreamer.EmitInstruction(LdStInst);
+    OutStreamer.EmitInstruction(MCInstBuilder(Opcode)
+      .addReg(MI->getOperand(0).getReg())
+      .addReg(ARM::PC)
+      .addReg(MI->getOperand(1).getReg())
+      .addImm(0)
+      // Add predicate operands.
+      .addImm(MI->getOperand(3).getImm())
+      .addReg(MI->getOperand(4).getReg()));
 
     return;
   }
@@ -1606,29 +1572,26 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
   }
   case ARM::t2BR_JT: {
     // Lower and emit the instruction itself, then the jump table following it.
-    MCInst TmpInst;
-    TmpInst.setOpcode(ARM::tMOVr);
-    TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
-    TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-    // Add predicate operands.
-    TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-    TmpInst.addOperand(MCOperand::CreateReg(0));
-    OutStreamer.EmitInstruction(TmpInst);
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tMOVr)
+      .addReg(ARM::PC)
+      .addReg(MI->getOperand(0).getReg())
+      // Add predicate operands.
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
     // Output the data for the jump table itself
     EmitJump2Table(MI);
     return;
   }
   case ARM::t2TBB_JT: {
     // Lower and emit the instruction itself, then the jump table following it.
-    MCInst TmpInst;
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::t2TBB)
+      .addReg(ARM::PC)
+      .addReg(MI->getOperand(0).getReg())
+      // Add predicate operands.
+      .addImm(ARMCC::AL)
+      .addReg(0));
 
-    TmpInst.setOpcode(ARM::t2TBB);
-    TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
-    TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-    // Add predicate operands.
-    TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-    TmpInst.addOperand(MCOperand::CreateReg(0));
-    OutStreamer.EmitInstruction(TmpInst);
     // Output the data for the jump table itself
     EmitJump2Table(MI);
     // Make sure the next instruction is 2-byte aligned.
@@ -1637,15 +1600,13 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
   }
   case ARM::t2TBH_JT: {
     // Lower and emit the instruction itself, then the jump table following it.
-    MCInst TmpInst;
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::t2TBH)
+      .addReg(ARM::PC)
+      .addReg(MI->getOperand(0).getReg())
+      // Add predicate operands.
+      .addImm(ARMCC::AL)
+      .addReg(0));
 
-    TmpInst.setOpcode(ARM::t2TBH);
-    TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
-    TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-    // Add predicate operands.
-    TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-    TmpInst.addOperand(MCOperand::CreateReg(0));
-    OutStreamer.EmitInstruction(TmpInst);
     // Output the data for the jump table itself
     EmitJump2Table(MI);
     return;
@@ -1705,17 +1666,15 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
   case ARM::BR_JTadd: {
     // Lower and emit the instruction itself, then the jump table following it.
     // add pc, target, idx
-    MCInst TmpInst;
-    TmpInst.setOpcode(ARM::ADDrr);
-    TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
-    TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-    TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(1).getReg()));
-    // Add predicate operands.
-    TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-    TmpInst.addOperand(MCOperand::CreateReg(0));
-    // Add 's' bit operand (always reg0 for this)
-    TmpInst.addOperand(MCOperand::CreateReg(0));
-    OutStreamer.EmitInstruction(TmpInst);
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::ADDrr)
+      .addReg(ARM::PC)
+      .addReg(MI->getOperand(0).getReg())
+      .addReg(MI->getOperand(1).getReg())
+      // Add predicate operands.
+      .addImm(ARMCC::AL)
+      .addReg(0)
+      // Add 's' bit operand (always reg0 for this)
+      .addReg(0));
 
     // Output the data for the jump table itself
     EmitJumpTable(MI);
@@ -1733,6 +1692,13 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     }
     break;
   }
+  case ARM::TRAPNaCl: {
+    //.long 0xe7fedef0 @ trap
+    uint32_t Val = 0xe7fedef0UL;
+    OutStreamer.AddComment("trap");
+    OutStreamer.EmitIntValue(Val, 4);
+    return;
+  }
   case ARM::tTRAP: {
     // Non-Darwin binutils don't yet support the "trap" mnemonic.
     // FIXME: Remove this special case when they do.
@@ -1759,75 +1725,57 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     unsigned SrcReg = MI->getOperand(0).getReg();
     unsigned ValReg = MI->getOperand(1).getReg();
     MCSymbol *Label = GetARMSJLJEHLabel();
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tMOVr);
-      TmpInst.addOperand(MCOperand::CreateReg(ValReg));
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
+    OutStreamer.AddComment("eh_setjmp begin");
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tMOVr)
+      .addReg(ValReg)
+      .addReg(ARM::PC)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.AddComment("eh_setjmp begin");
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tADDi3);
-      TmpInst.addOperand(MCOperand::CreateReg(ValReg));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tADDi3)
+      .addReg(ValReg)
       // 's' bit operand
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::CPSR));
-      TmpInst.addOperand(MCOperand::CreateReg(ValReg));
-      TmpInst.addOperand(MCOperand::CreateImm(7));
+      .addReg(ARM::CPSR)
+      .addReg(ValReg)
+      .addImm(7)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tSTRi);
-      TmpInst.addOperand(MCOperand::CreateReg(ValReg));
-      TmpInst.addOperand(MCOperand::CreateReg(SrcReg));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tSTRi)
+      .addReg(ValReg)
+      .addReg(SrcReg)
       // The offset immediate is #4. The operand value is scaled by 4 for the
       // tSTR instruction.
-      TmpInst.addOperand(MCOperand::CreateImm(1));
+      .addImm(1)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tMOVi8);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::R0));
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::CPSR));
-      TmpInst.addOperand(MCOperand::CreateImm(0));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tMOVi8)
+      .addReg(ARM::R0)
+      .addReg(ARM::CPSR)
+      .addImm(0)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      const MCExpr *SymbolExpr = MCSymbolRefExpr::Create(Label, OutContext);
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tB);
-      TmpInst.addOperand(MCOperand::CreateExpr(SymbolExpr));
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tMOVi8);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::R0));
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::CPSR));
-      TmpInst.addOperand(MCOperand::CreateImm(1));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    const MCExpr *SymbolExpr = MCSymbolRefExpr::Create(Label, OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tB)
+      .addExpr(SymbolExpr)
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.AddComment("eh_setjmp end");
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tMOVi8)
+      .addReg(ARM::R0)
+      .addReg(ARM::CPSR)
+      .addImm(1)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.AddComment("eh_setjmp end");
-      OutStreamer.EmitInstruction(TmpInst);
-    }
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
     OutStreamer.EmitLabel(Label);
     return;
   }
@@ -1843,69 +1791,53 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     unsigned SrcReg = MI->getOperand(0).getReg();
     unsigned ValReg = MI->getOperand(1).getReg();
 
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::ADDri);
-      TmpInst.addOperand(MCOperand::CreateReg(ValReg));
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
-      TmpInst.addOperand(MCOperand::CreateImm(8));
+    OutStreamer.AddComment("eh_setjmp begin");
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::ADDri)
+      .addReg(ValReg)
+      .addReg(ARM::PC)
+      .addImm(8)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
+      .addImm(ARMCC::AL)
+      .addReg(0)
       // 's' bit operand (always reg0 for this).
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.AddComment("eh_setjmp begin");
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::STRi12);
-      TmpInst.addOperand(MCOperand::CreateReg(ValReg));
-      TmpInst.addOperand(MCOperand::CreateReg(SrcReg));
-      TmpInst.addOperand(MCOperand::CreateImm(4));
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::STRi12)
+      .addReg(ValReg)
+      .addReg(SrcReg)
+      .addImm(4)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::MOVi);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::R0));
-      TmpInst.addOperand(MCOperand::CreateImm(0));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::MOVi)
+      .addReg(ARM::R0)
+      .addImm(0)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
+      .addImm(ARMCC::AL)
+      .addReg(0)
       // 's' bit operand (always reg0 for this).
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::ADDri);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::PC));
-      TmpInst.addOperand(MCOperand::CreateImm(0));
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::ADDri)
+      .addReg(ARM::PC)
+      .addReg(ARM::PC)
+      .addImm(0)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
+      .addImm(ARMCC::AL)
+      .addReg(0)
       // 's' bit operand (always reg0 for this).
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::MOVi);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::R0));
-      TmpInst.addOperand(MCOperand::CreateImm(1));
+      .addReg(0));
+
+    OutStreamer.AddComment("eh_setjmp end");
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::MOVi)
+      .addReg(ARM::R0)
+      .addImm(1)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
+      .addImm(ARMCC::AL)
+      .addReg(0)
       // 's' bit operand (always reg0 for this).
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.AddComment("eh_setjmp end");
-      OutStreamer.EmitInstruction(TmpInst);
-    }
+      .addReg(0));
     return;
   }
   case ARM::Int_eh_sjlj_longjmp: {
@@ -1915,48 +1847,35 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     // bx $scratch
     unsigned SrcReg = MI->getOperand(0).getReg();
     unsigned ScratchReg = MI->getOperand(1).getReg();
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::LDRi12);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::SP));
-      TmpInst.addOperand(MCOperand::CreateReg(SrcReg));
-      TmpInst.addOperand(MCOperand::CreateImm(8));
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::LDRi12)
+      .addReg(ARM::SP)
+      .addReg(SrcReg)
+      .addImm(8)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::LDRi12);
-      TmpInst.addOperand(MCOperand::CreateReg(ScratchReg));
-      TmpInst.addOperand(MCOperand::CreateReg(SrcReg));
-      TmpInst.addOperand(MCOperand::CreateImm(4));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::LDRi12)
+      .addReg(ScratchReg)
+      .addReg(SrcReg)
+      .addImm(4)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::LDRi12);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::R7));
-      TmpInst.addOperand(MCOperand::CreateReg(SrcReg));
-      TmpInst.addOperand(MCOperand::CreateImm(0));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::LDRi12)
+      .addReg(ARM::R7)
+      .addReg(SrcReg)
+      .addImm(0)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::BX);
-      TmpInst.addOperand(MCOperand::CreateReg(ScratchReg));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::BX)
+      .addReg(ScratchReg)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
+      .addImm(ARMCC::AL)
+      .addReg(0));
     return;
   }
   case ARM::tInt_eh_sjlj_longjmp: {
@@ -1967,60 +1886,44 @@ void ARMAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     // bx $scratch
     unsigned SrcReg = MI->getOperand(0).getReg();
     unsigned ScratchReg = MI->getOperand(1).getReg();
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tLDRi);
-      TmpInst.addOperand(MCOperand::CreateReg(ScratchReg));
-      TmpInst.addOperand(MCOperand::CreateReg(SrcReg));
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tLDRi)
+      .addReg(ScratchReg)
+      .addReg(SrcReg)
       // The offset immediate is #8. The operand value is scaled by 4 for the
       // tLDR instruction.
-      TmpInst.addOperand(MCOperand::CreateImm(2));
+      .addImm(2)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tMOVr);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::SP));
-      TmpInst.addOperand(MCOperand::CreateReg(ScratchReg));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tMOVr)
+      .addReg(ARM::SP)
+      .addReg(ScratchReg)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tLDRi);
-      TmpInst.addOperand(MCOperand::CreateReg(ScratchReg));
-      TmpInst.addOperand(MCOperand::CreateReg(SrcReg));
-      TmpInst.addOperand(MCOperand::CreateImm(1));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tLDRi)
+      .addReg(ScratchReg)
+      .addReg(SrcReg)
+      .addImm(1)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tLDRi);
-      TmpInst.addOperand(MCOperand::CreateReg(ARM::R7));
-      TmpInst.addOperand(MCOperand::CreateReg(SrcReg));
-      TmpInst.addOperand(MCOperand::CreateImm(0));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tLDRi)
+      .addReg(ARM::R7)
+      .addReg(SrcReg)
+      .addImm(0)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
-    {
-      MCInst TmpInst;
-      TmpInst.setOpcode(ARM::tBX);
-      TmpInst.addOperand(MCOperand::CreateReg(ScratchReg));
+      .addImm(ARMCC::AL)
+      .addReg(0));
+
+    OutStreamer.EmitInstruction(MCInstBuilder(ARM::tBX)
+      .addReg(ScratchReg)
       // Predicate.
-      TmpInst.addOperand(MCOperand::CreateImm(ARMCC::AL));
-      TmpInst.addOperand(MCOperand::CreateReg(0));
-      OutStreamer.EmitInstruction(TmpInst);
-    }
+      .addImm(ARMCC::AL)
+      .addReg(0));
     return;
   }
   }
diff --git a/lib/Target/ARM/ARMAsmPrinter.h b/lib/Target/ARM/ARMAsmPrinter.h
index c875b2cbdffe..c945e4f28699 100644
--- a/lib/Target/ARM/ARMAsmPrinter.h
+++ b/lib/Target/ARM/ARMAsmPrinter.h
@@ -1,4 +1,4 @@
-//===-- ARMAsmPrinter.h - Print machine code to an ARM .s file --*- C++ -*-===//
+//===-- ARMAsmPrinter.h - ARM implementation of AsmPrinter ------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -6,10 +6,6 @@
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
-//
-// ARM Assembly printer class.
-//
-//===----------------------------------------------------------------------===//
 
 #ifndef ARMASMPRINTER_H
 #define ARMASMPRINTER_H
@@ -54,7 +50,7 @@ public:
     }
 
   virtual const char *getPassName() const LLVM_OVERRIDE {
-    return "ARM Assembly Printer";
+    return "ARM Assembly / Object Emitter";
   }
 
   void printOperand(const MachineInstr *MI, int OpNum, raw_ostream &O,
@@ -121,7 +117,7 @@ private:
   MCOperand GetSymbolRef(const MachineOperand &MO, const MCSymbol *Symbol);
   MCSymbol *GetARMJTIPICJumpTableLabel2(unsigned uid, unsigned uid2) const;
 
-  MCSymbol *GetARMSJLJEHLabel(void) const;
+  MCSymbol *GetARMSJLJEHLabel() const;
 
   MCSymbol *GetARMGVSymbol(const GlobalValue *GV);
 
diff --git a/lib/Target/ARM/ARMBaseInstrInfo.cpp b/lib/Target/ARM/ARMBaseInstrInfo.cpp
index 3c7bb24f42f8..9e68ff44890e 100644
--- a/lib/Target/ARM/ARMBaseInstrInfo.cpp
+++ b/lib/Target/ARM/ARMBaseInstrInfo.cpp
@@ -18,9 +18,7 @@
 #include "ARMHazardRecognizer.h"
 #include "ARMMachineFunctionInfo.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/Constants.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalValue.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/LiveVariables.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
@@ -29,12 +27,14 @@
 #include "llvm/CodeGen/MachineMemOperand.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAGNodes.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalValue.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/Support/BranchProbability.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
-#include "llvm/ADT/STLExtras.h"
 
 #define GET_INSTRINFO_CTOR
 #include "ARMGenInstrInfo.inc"
@@ -106,7 +106,7 @@ CreateTargetHazardRecognizer(const TargetMachine *TM,
     const InstrItineraryData *II = TM->getInstrItineraryData();
     return new ScoreboardHazardRecognizer(II, DAG, "pre-RA-sched");
   }
-  return TargetInstrInfoImpl::CreateTargetHazardRecognizer(TM, DAG);
+  return TargetInstrInfo::CreateTargetHazardRecognizer(TM, DAG);
 }
 
 ScheduleHazardRecognizer *ARMBaseInstrInfo::
@@ -115,7 +115,7 @@ CreateTargetPostRAHazardRecognizer(const InstrItineraryData *II,
   if (Subtarget.isThumb2() || Subtarget.hasVFP2())
     return (ScheduleHazardRecognizer *)
       new ARMHazardRecognizer(II, *this, getRegisterInfo(), Subtarget, DAG);
-  return TargetInstrInfoImpl::CreateTargetPostRAHazardRecognizer(II, DAG);
+  return TargetInstrInfo::CreateTargetPostRAHazardRecognizer(II, DAG);
 }
 
 MachineInstr *
@@ -464,8 +464,9 @@ PredicateInstruction(MachineInstr *MI,
   unsigned Opc = MI->getOpcode();
   if (isUncondBranchOpcode(Opc)) {
     MI->setDesc(get(getMatchingCondBranchOpcode(Opc)));
-    MI->addOperand(MachineOperand::CreateImm(Pred[0].getImm()));
-    MI->addOperand(MachineOperand::CreateReg(Pred[1].getReg(), false));
+    MachineInstrBuilder(*MI->getParent()->getParent(), MI)
+      .addImm(Pred[0].getImm())
+      .addReg(Pred[1].getReg());
     return true;
   }
 
@@ -1124,7 +1125,7 @@ bool ARMBaseInstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const{
   // copyPhysReg() calls.  Look for VMOVS instructions that can legally be
   // widened to VMOVD.  We prefer the VMOVD when possible because it may be
   // changed into a VORR that can go down the NEON pipeline.
-  if (!WidenVMOVS || !MI->isCopy())
+  if (!WidenVMOVS || !MI->isCopy() || Subtarget.isCortexA15())
     return false;
 
   // Look for a copy between even S-registers.  That is where we keep floats
@@ -1154,6 +1155,7 @@ bool ARMBaseInstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const{
 
   // All clear, widen the COPY.
   DEBUG(dbgs() << "widening:    " << *MI);
+  MachineInstrBuilder MIB(*MI->getParent()->getParent(), MI);
 
   // Get rid of the old <imp-def> of DstRegD.  Leave it if it defines a Q-reg
   // or some other super-register.
@@ -1165,14 +1167,14 @@ bool ARMBaseInstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const{
   MI->setDesc(get(ARM::VMOVD));
   MI->getOperand(0).setReg(DstRegD);
   MI->getOperand(1).setReg(SrcRegD);
-  AddDefaultPred(MachineInstrBuilder(MI));
+  AddDefaultPred(MIB);
 
   // We are now reading SrcRegD instead of SrcRegS.  This may upset the
   // register scavenger and machine verifier, so we need to indicate that we
   // are reading an undefined value from SrcRegD, but a proper value from
   // SrcRegS.
   MI->getOperand(1).setIsUndef();
-  MachineInstrBuilder(MI).addReg(SrcRegS, RegState::Implicit);
+  MIB.addReg(SrcRegS, RegState::Implicit);
 
   // SrcRegD may actually contain an unrelated value in the ssub_1
   // sub-register.  Don't kill it.  Only kill the ssub_0 sub-register.
@@ -1269,7 +1271,7 @@ reMaterialize(MachineBasicBlock &MBB,
 
 MachineInstr *
 ARMBaseInstrInfo::duplicate(MachineInstr *Orig, MachineFunction &MF) const {
-  MachineInstr *MI = TargetInstrInfoImpl::duplicate(Orig, MF);
+  MachineInstr *MI = TargetInstrInfo::duplicate(Orig, MF);
   switch(Orig->getOpcode()) {
   case ARM::tLDRpci_pic:
   case ARM::t2LDRpci_pic: {
@@ -1373,6 +1375,9 @@ bool ARMBaseInstrInfo::produceSameValue(const MachineInstr *MI0,
 /// only return true if the base pointers are the same and the only differences
 /// between the two addresses is the offset. It also returns the offsets by
 /// reference.
+///
+/// FIXME: remove this in favor of the MachineInstr interface once pre-RA-sched
+/// is permanently disabled.
 bool ARMBaseInstrInfo::areLoadsFromSameBasePtr(SDNode *Load1, SDNode *Load2,
                                                int64_t &Offset1,
                                                int64_t &Offset2) const {
@@ -1447,6 +1452,9 @@ bool ARMBaseInstrInfo::areLoadsFromSameBasePtr(SDNode *Load1, SDNode *Load2,
 /// from the common base address. It returns true if it decides it's desirable
 /// to schedule the two loads together. "NumLoads" is the number of loads that
 /// have already been scheduled after Load1.
+///
+/// FIXME: remove this in favor of the MachineInstr interface once pre-RA-sched
+/// is permanently disabled.
 bool ARMBaseInstrInfo::shouldScheduleLoadsNear(SDNode *Load1, SDNode *Load2,
                                                int64_t Offset1, int64_t Offset2,
                                                unsigned NumLoads) const {
@@ -1598,7 +1606,7 @@ ARMBaseInstrInfo::commuteInstruction(MachineInstr *MI, bool NewMI) const {
     // MOVCC AL can't be inverted. Shouldn't happen.
     if (CC == ARMCC::AL || PredReg != ARM::CPSR)
       return NULL;
-    MI = TargetInstrInfoImpl::commuteInstruction(MI, NewMI);
+    MI = TargetInstrInfo::commuteInstruction(MI, NewMI);
     if (!MI)
       return NULL;
     // After swapping the MOVCC operands, also invert the condition.
@@ -1607,7 +1615,7 @@ ARMBaseInstrInfo::commuteInstruction(MachineInstr *MI, bool NewMI) const {
     return MI;
   }
   }
-  return TargetInstrInfoImpl::commuteInstruction(MI, NewMI);
+  return TargetInstrInfo::commuteInstruction(MI, NewMI);
 }
 
 /// Identify instructions that can be folded into a MOVCC instruction, and
@@ -1710,7 +1718,7 @@ MachineInstr *ARMBaseInstrInfo::optimizeSelect(MachineInstr *MI,
   // same register as operand 0.
   MachineOperand FalseReg = MI->getOperand(Invert ? 2 : 1);
   FalseReg.setImplicit();
-  NewMI->addOperand(FalseReg);
+  NewMI.addOperand(FalseReg);
   NewMI->tieOperands(0, NewMI->getNumOperands() - 1);
 
   // The caller will erase MI, but not DefMI.
@@ -2711,7 +2719,6 @@ ARMBaseInstrInfo::getNumMicroOps(const InstrItineraryData *ItinData,
   case ARM::t2STMDB_UPD: {
     unsigned NumRegs = MI->getNumOperands() - Desc.getNumOperands() + 1;
     if (Subtarget.isSwift()) {
-      // rdar://8402126
       int UOps = 1 + NumRegs;  // One for address computation, one for each ld / st.
       switch (Opc) {
       default: break;
@@ -3321,8 +3328,9 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData,
     // instructions).
     if (Latency > 0 && Subtarget.isThumb2()) {
       const MachineFunction *MF = DefMI->getParent()->getParent();
-      if (MF->getFunction()->getFnAttributes().
-            hasAttribute(Attributes::OptimizeForSize))
+      if (MF->getFunction()->getAttributes().
+            hasAttribute(AttributeSet::FunctionIndex,
+                         Attribute::OptimizeForSize))
         --Latency;
     }
     return Latency;
@@ -3726,9 +3734,9 @@ ARMBaseInstrInfo::getExecutionDomain(const MachineInstr *MI) const {
   if (MI->getOpcode() == ARM::VMOVD && !isPredicated(MI))
     return std::make_pair(ExeVFP, (1<<ExeVFP) | (1<<ExeNEON));
 
-  // A9-like cores are particularly picky about mixing the two and want these
+  // CortexA9 is particularly picky about mixing the two and wants these
   // converted.
-  if (Subtarget.isLikeA9() && !isPredicated(MI) &&
+  if (Subtarget.isCortexA9() && !isPredicated(MI) &&
       (MI->getOpcode() == ARM::VMOVRS ||
        MI->getOpcode() == ARM::VMOVSR ||
        MI->getOpcode() == ARM::VMOVS))
@@ -3813,7 +3821,7 @@ void
 ARMBaseInstrInfo::setExecutionDomain(MachineInstr *MI, unsigned Domain) const {
   unsigned DstReg, SrcReg, DReg;
   unsigned Lane;
-  MachineInstrBuilder MIB(MI);
+  MachineInstrBuilder MIB(*MI->getParent()->getParent(), MI);
   const TargetRegisterInfo *TRI = &getRegisterInfo();
   switch (MI->getOpcode()) {
     default:
@@ -4015,14 +4023,12 @@ ARMBaseInstrInfo::setExecutionDomain(MachineInstr *MI, unsigned Domain) const {
 // VLD1DUPd32 - Writes all D-regs, no partial reg update, 2 uops.
 //
 // FCONSTD can be used as a dependency-breaking instruction.
-
-
 unsigned ARMBaseInstrInfo::
 getPartialRegUpdateClearance(const MachineInstr *MI,
                              unsigned OpNum,
                              const TargetRegisterInfo *TRI) const {
-  // Only Swift has partial register update problems.
-  if (!SwiftPartialUpdateClearance || !Subtarget.isSwift())
+  if (!SwiftPartialUpdateClearance ||
+      !(Subtarget.isSwift() || Subtarget.isCortexA15()))
     return 0;
 
   assert(TRI && "Need TRI instance");
@@ -4038,7 +4044,6 @@ getPartialRegUpdateClearance(const MachineInstr *MI,
   case ARM::VLDRS:
   case ARM::FCONSTS:
   case ARM::VMOVSR:
-    // rdar://problem/8791586
   case ARM::VMOVv8i8:
   case ARM::VMOVv4i16:
   case ARM::VMOVv2i32:
@@ -4049,7 +4054,7 @@ getPartialRegUpdateClearance(const MachineInstr *MI,
 
     // Explicitly reads the dependency.
   case ARM::VLD1LNd32:
-    UseOp = 1;
+    UseOp = 3;
     break;
   default:
     return 0;
@@ -4118,3 +4123,15 @@ breakPartialRegDependency(MachineBasicBlock::iterator MI,
 bool ARMBaseInstrInfo::hasNOP() const {
   return (Subtarget.getFeatureBits() & ARM::HasV6T2Ops) != 0;
 }
+
+bool ARMBaseInstrInfo::isSwiftFastImmShift(const MachineInstr *MI) const {
+  unsigned ShOpVal = MI->getOperand(3).getImm();
+  unsigned ShImm = ARM_AM::getSORegOffset(ShOpVal);
+  // Swift supports faster shifts for: lsl 2, lsl 1, and lsr 1.
+  if ((ShImm == 1 && ARM_AM::getSORegShOp(ShOpVal) == ARM_AM::lsr) ||
+      ((ShImm == 1 || ShImm == 2) &&
+       ARM_AM::getSORegShOp(ShOpVal) == ARM_AM::lsl))
+    return true;
+
+  return false;
+}
diff --git a/lib/Target/ARM/ARMBaseInstrInfo.h b/lib/Target/ARM/ARMBaseInstrInfo.h
index 6f38e35124eb..7c107bb41951 100644
--- a/lib/Target/ARM/ARMBaseInstrInfo.h
+++ b/lib/Target/ARM/ARMBaseInstrInfo.h
@@ -15,10 +15,10 @@
 #define ARMBASEINSTRUCTIONINFO_H
 
 #include "ARM.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallSet.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/Target/TargetInstrInfo.h"
 
 #define GET_INSTRINFO_HEADER
 #include "ARMGenInstrInfo.inc"
@@ -314,6 +314,10 @@ public:
   bool canCauseFpMLxStall(unsigned Opcode) const {
     return MLxHazardOpcodes.count(Opcode);
   }
+
+  /// Returns true if the instruction has a shift by immediate that can be
+  /// executed in one cycle less.
+  bool isSwiftFastImmShift(const MachineInstr *MI) const;
 };
 
 static inline
diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/lib/Target/ARM/ARMBaseRegisterInfo.cpp
index e5b300fc7792..b6b27f849a23 100644
--- a/lib/Target/ARM/ARMBaseRegisterInfo.cpp
+++ b/lib/Target/ARM/ARMBaseRegisterInfo.cpp
@@ -18,44 +18,34 @@
 #include "ARMMachineFunctionInfo.h"
 #include "ARMSubtarget.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/LLVMContext.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/SmallVector.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/CodeGen/VirtRegMap.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/LLVMContext.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetFrameLowering.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetOptions.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/Support/CommandLine.h"
 
 #define GET_REGINFO_TARGET_DESC
 #include "ARMGenRegisterInfo.inc"
 
 using namespace llvm;
 
-static cl::opt<bool>
-ForceAllBaseRegAlloc("arm-force-base-reg-alloc", cl::Hidden, cl::init(false),
-          cl::desc("Force use of virtual base registers for stack load/store"));
-static cl::opt<bool>
-EnableLocalStackAlloc("enable-local-stack-alloc", cl::init(true), cl::Hidden,
-          cl::desc("Enable pre-regalloc stack frame index allocation"));
-static cl::opt<bool>
-EnableBasePointer("arm-use-base-pointer", cl::Hidden, cl::init(true),
-          cl::desc("Enable use of a base pointer for complex stack frames"));
-
 ARMBaseRegisterInfo::ARMBaseRegisterInfo(const ARMBaseInstrInfo &tii,
                                          const ARMSubtarget &sti)
-  : ARMGenRegisterInfo(ARM::LR), TII(tii), STI(sti),
+  : ARMGenRegisterInfo(ARM::LR, 0, 0, ARM::PC), TII(tii), STI(sti),
     FramePtr((STI.isTargetDarwin() || STI.isThumb()) ? ARM::R7 : ARM::R11),
     BasePtr(ARM::R6) {
 }
@@ -173,154 +163,63 @@ ARMBaseRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC,
   }
 }
 
-/// getRawAllocationOrder - Returns the register allocation order for a
-/// specified register class with a target-dependent hint.
-ArrayRef<uint16_t>
-ARMBaseRegisterInfo::getRawAllocationOrder(const TargetRegisterClass *RC,
-                                           unsigned HintType, unsigned HintReg,
-                                           const MachineFunction &MF) const {
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-  // Alternative register allocation orders when favoring even / odd registers
-  // of register pairs.
-
-  // No FP, R9 is available.
-  static const uint16_t GPREven1[] = {
-    ARM::R0, ARM::R2, ARM::R4, ARM::R6, ARM::R8, ARM::R10,
-    ARM::R1, ARM::R3, ARM::R12,ARM::LR, ARM::R5, ARM::R7,
-    ARM::R9, ARM::R11
-  };
-  static const uint16_t GPROdd1[] = {
-    ARM::R1, ARM::R3, ARM::R5, ARM::R7, ARM::R9, ARM::R11,
-    ARM::R0, ARM::R2, ARM::R12,ARM::LR, ARM::R4, ARM::R6,
-    ARM::R8, ARM::R10
-  };
-
-  // FP is R7, R9 is available.
-  static const uint16_t GPREven2[] = {
-    ARM::R0, ARM::R2, ARM::R4,          ARM::R8, ARM::R10,
-    ARM::R1, ARM::R3, ARM::R12,ARM::LR, ARM::R5, ARM::R6,
-    ARM::R9, ARM::R11
-  };
-  static const uint16_t GPROdd2[] = {
-    ARM::R1, ARM::R3, ARM::R5,          ARM::R9, ARM::R11,
-    ARM::R0, ARM::R2, ARM::R12,ARM::LR, ARM::R4, ARM::R6,
-    ARM::R8, ARM::R10
-  };
-
-  // FP is R11, R9 is available.
-  static const uint16_t GPREven3[] = {
-    ARM::R0, ARM::R2, ARM::R4, ARM::R6, ARM::R8,
-    ARM::R1, ARM::R3, ARM::R10,ARM::R12,ARM::LR, ARM::R5, ARM::R7,
-    ARM::R9
-  };
-  static const uint16_t GPROdd3[] = {
-    ARM::R1, ARM::R3, ARM::R5, ARM::R6, ARM::R9,
-    ARM::R0, ARM::R2, ARM::R10,ARM::R12,ARM::LR, ARM::R4, ARM::R7,
-    ARM::R8
-  };
-
-  // No FP, R9 is not available.
-  static const uint16_t GPREven4[] = {
-    ARM::R0, ARM::R2, ARM::R4, ARM::R6,          ARM::R10,
-    ARM::R1, ARM::R3, ARM::R12,ARM::LR, ARM::R5, ARM::R7, ARM::R8,
-    ARM::R11
-  };
-  static const uint16_t GPROdd4[] = {
-    ARM::R1, ARM::R3, ARM::R5, ARM::R7,          ARM::R11,
-    ARM::R0, ARM::R2, ARM::R12,ARM::LR, ARM::R4, ARM::R6, ARM::R8,
-    ARM::R10
-  };
-
-  // FP is R7, R9 is not available.
-  static const uint16_t GPREven5[] = {
-    ARM::R0, ARM::R2, ARM::R4,                   ARM::R10,
-    ARM::R1, ARM::R3, ARM::R12,ARM::LR, ARM::R5, ARM::R6, ARM::R8,
-    ARM::R11
-  };
-  static const uint16_t GPROdd5[] = {
-    ARM::R1, ARM::R3, ARM::R5,                   ARM::R11,
-    ARM::R0, ARM::R2, ARM::R12,ARM::LR, ARM::R4, ARM::R6, ARM::R8,
-    ARM::R10
-  };
-
-  // FP is R11, R9 is not available.
-  static const uint16_t GPREven6[] = {
-    ARM::R0, ARM::R2, ARM::R4, ARM::R6,
-    ARM::R1, ARM::R3, ARM::R10,ARM::R12,ARM::LR, ARM::R5, ARM::R7, ARM::R8
-  };
-  static const uint16_t GPROdd6[] = {
-    ARM::R1, ARM::R3, ARM::R5, ARM::R7,
-    ARM::R0, ARM::R2, ARM::R10,ARM::R12,ARM::LR, ARM::R4, ARM::R6, ARM::R8
-  };
-
-  // We only support even/odd hints for GPR and rGPR.
-  if (RC != &ARM::GPRRegClass && RC != &ARM::rGPRRegClass)
-    return RC->getRawAllocationOrder(MF);
-
-  if (HintType == ARMRI::RegPairEven) {
-    if (isPhysicalRegister(HintReg) && getRegisterPairEven(HintReg, MF) == 0)
-      // It's no longer possible to fulfill this hint. Return the default
-      // allocation order.
-      return RC->getRawAllocationOrder(MF);
-
-    if (!TFI->hasFP(MF)) {
-      if (!STI.isR9Reserved())
-        return makeArrayRef(GPREven1);
-      else
-        return makeArrayRef(GPREven4);
-    } else if (FramePtr == ARM::R7) {
-      if (!STI.isR9Reserved())
-        return makeArrayRef(GPREven2);
-      else
-        return makeArrayRef(GPREven5);
-    } else { // FramePtr == ARM::R11
-      if (!STI.isR9Reserved())
-        return makeArrayRef(GPREven3);
-      else
-        return makeArrayRef(GPREven6);
-    }
-  } else if (HintType == ARMRI::RegPairOdd) {
-    if (isPhysicalRegister(HintReg) && getRegisterPairOdd(HintReg, MF) == 0)
-      // It's no longer possible to fulfill this hint. Return the default
-      // allocation order.
-      return RC->getRawAllocationOrder(MF);
-
-    if (!TFI->hasFP(MF)) {
-      if (!STI.isR9Reserved())
-        return makeArrayRef(GPROdd1);
-      else
-        return makeArrayRef(GPROdd4);
-    } else if (FramePtr == ARM::R7) {
-      if (!STI.isR9Reserved())
-        return makeArrayRef(GPROdd2);
-      else
-        return makeArrayRef(GPROdd5);
-    } else { // FramePtr == ARM::R11
-      if (!STI.isR9Reserved())
-        return makeArrayRef(GPROdd3);
-      else
-        return makeArrayRef(GPROdd6);
-    }
-  }
-  return RC->getRawAllocationOrder(MF);
+// Get the other register in a GPRPair.
+static unsigned getPairedGPR(unsigned Reg, bool Odd, const MCRegisterInfo *RI) {
+  for (MCSuperRegIterator Supers(Reg, RI); Supers.isValid(); ++Supers)
+    if (ARM::GPRPairRegClass.contains(*Supers))
+      return RI->getSubReg(*Supers, Odd ? ARM::gsub_1 : ARM::gsub_0);
+  return 0;
 }
 
-/// ResolveRegAllocHint - Resolves the specified register allocation hint
-/// to a physical register. Returns the physical register if it is successful.
-unsigned
-ARMBaseRegisterInfo::ResolveRegAllocHint(unsigned Type, unsigned Reg,
-                                         const MachineFunction &MF) const {
-  if (Reg == 0 || !isPhysicalRegister(Reg))
-    return 0;
-  if (Type == 0)
-    return Reg;
-  else if (Type == (unsigned)ARMRI::RegPairOdd)
-    // Odd register.
-    return getRegisterPairOdd(Reg, MF);
-  else if (Type == (unsigned)ARMRI::RegPairEven)
-    // Even register.
-    return getRegisterPairEven(Reg, MF);
-  return 0;
+// Resolve the RegPairEven / RegPairOdd register allocator hints.
+void
+ARMBaseRegisterInfo::getRegAllocationHints(unsigned VirtReg,
+                                           ArrayRef<MCPhysReg> Order,
+                                           SmallVectorImpl<MCPhysReg> &Hints,
+                                           const MachineFunction &MF,
+                                           const VirtRegMap *VRM) const {
+  const MachineRegisterInfo &MRI = MF.getRegInfo();
+  std::pair<unsigned, unsigned> Hint = MRI.getRegAllocationHint(VirtReg);
+
+  unsigned Odd;
+  switch (Hint.first) {
+  case ARMRI::RegPairEven:
+    Odd = 0;
+    break;
+  case ARMRI::RegPairOdd:
+    Odd = 1;
+    break;
+  default:
+    TargetRegisterInfo::getRegAllocationHints(VirtReg, Order, Hints, MF, VRM);
+    return;
+  }
+
+  // This register should preferably be even (Odd == 0) or odd (Odd == 1).
+  // Check if the other part of the pair has already been assigned, and provide
+  // the paired register as the first hint.
+  unsigned PairedPhys = 0;
+  if (VRM && VRM->hasPhys(Hint.second)) {
+    PairedPhys = getPairedGPR(VRM->getPhys(Hint.second), Odd, this);
+    if (PairedPhys && MRI.isReserved(PairedPhys))
+      PairedPhys = 0;
+  }
+
+  // First prefer the paired physreg.
+  if (PairedPhys &&
+      std::find(Order.begin(), Order.end(), PairedPhys) != Order.end())
+    Hints.push_back(PairedPhys);
+
+  // Then prefer even or odd registers.
+  for (unsigned I = 0, E = Order.size(); I != E; ++I) {
+    unsigned Reg = Order[I];
+    if (Reg == PairedPhys || (getEncodingValue(Reg) & 1) != Odd)
+      continue;
+    // Don't provide hints that are paired to a reserved register.
+    unsigned Paired = getPairedGPR(Reg, !Odd, this);
+    if (!Paired || MRI.isReserved(Paired))
+      continue;
+    Hints.push_back(Reg);
+  }
 }
 
 void
@@ -371,9 +270,6 @@ bool ARMBaseRegisterInfo::hasBasePointer(const MachineFunction &MF) const {
   const ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
   const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
 
-  if (!EnableBasePointer)
-    return false;
-
   // When outgoing call frames are so large that we adjust the stack pointer
   // around the call, we can no longer use the stack pointer to reach the
   // emergency spill slot.
@@ -419,8 +315,6 @@ bool ARMBaseRegisterInfo::canRealignStack(const MachineFunction &MF) const {
   // pointer adjustments around calls.
   if (MF.getTarget().getFrameLowering()->hasReservedCallFrame(MF))
     return true;
-  if (!EnableBasePointer)
-    return false;
   // A base pointer is required and allowed.  Check that it isn't too late to
   // reserve it.
   return MRI->canReserveReg(BasePtr);
@@ -433,7 +327,8 @@ needsStackRealignment(const MachineFunction &MF) const {
   unsigned StackAlign = MF.getTarget().getFrameLowering()->getStackAlignment();
   bool requiresRealignment =
     ((MFI->getMaxAlignment() > StackAlign) ||
-     F->getFnAttributes().hasAttribute(Attributes::StackAlignment));
+     F->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
+                                     Attribute::StackAlignment));
 
   return requiresRealignment && canRealignStack(MF);
 }
@@ -464,114 +359,6 @@ unsigned ARMBaseRegisterInfo::getEHHandlerRegister() const {
   llvm_unreachable("What is the exception handler register");
 }
 
-unsigned ARMBaseRegisterInfo::getRegisterPairEven(unsigned Reg,
-                                              const MachineFunction &MF) const {
-  const MachineRegisterInfo &MRI = MF.getRegInfo();
-  switch (Reg) {
-  default: break;
-  // Return 0 if either register of the pair is a special register.
-  // So no R12, etc.
-  case ARM::R1: return ARM::R0;
-  case ARM::R3: return ARM::R2;
-  case ARM::R5: return ARM::R4;
-  case ARM::R7:
-    return (MRI.isReserved(ARM::R7) || MRI.isReserved(ARM::R6))
-      ? 0 : ARM::R6;
-  case ARM::R9: return MRI.isReserved(ARM::R9)  ? 0 :ARM::R8;
-  case ARM::R11: return MRI.isReserved(ARM::R11) ? 0 : ARM::R10;
-
-  case ARM::S1: return ARM::S0;
-  case ARM::S3: return ARM::S2;
-  case ARM::S5: return ARM::S4;
-  case ARM::S7: return ARM::S6;
-  case ARM::S9: return ARM::S8;
-  case ARM::S11: return ARM::S10;
-  case ARM::S13: return ARM::S12;
-  case ARM::S15: return ARM::S14;
-  case ARM::S17: return ARM::S16;
-  case ARM::S19: return ARM::S18;
-  case ARM::S21: return ARM::S20;
-  case ARM::S23: return ARM::S22;
-  case ARM::S25: return ARM::S24;
-  case ARM::S27: return ARM::S26;
-  case ARM::S29: return ARM::S28;
-  case ARM::S31: return ARM::S30;
-
-  case ARM::D1: return ARM::D0;
-  case ARM::D3: return ARM::D2;
-  case ARM::D5: return ARM::D4;
-  case ARM::D7: return ARM::D6;
-  case ARM::D9: return ARM::D8;
-  case ARM::D11: return ARM::D10;
-  case ARM::D13: return ARM::D12;
-  case ARM::D15: return ARM::D14;
-  case ARM::D17: return ARM::D16;
-  case ARM::D19: return ARM::D18;
-  case ARM::D21: return ARM::D20;
-  case ARM::D23: return ARM::D22;
-  case ARM::D25: return ARM::D24;
-  case ARM::D27: return ARM::D26;
-  case ARM::D29: return ARM::D28;
-  case ARM::D31: return ARM::D30;
-  }
-
-  return 0;
-}
-
-unsigned ARMBaseRegisterInfo::getRegisterPairOdd(unsigned Reg,
-                                             const MachineFunction &MF) const {
-  const MachineRegisterInfo &MRI = MF.getRegInfo();
-  switch (Reg) {
-  default: break;
-  // Return 0 if either register of the pair is a special register.
-  // So no R12, etc.
-  case ARM::R0: return ARM::R1;
-  case ARM::R2: return ARM::R3;
-  case ARM::R4: return ARM::R5;
-  case ARM::R6:
-    return (MRI.isReserved(ARM::R7) || MRI.isReserved(ARM::R6))
-      ? 0 : ARM::R7;
-  case ARM::R8: return MRI.isReserved(ARM::R9)  ? 0 :ARM::R9;
-  case ARM::R10: return MRI.isReserved(ARM::R11) ? 0 : ARM::R11;
-
-  case ARM::S0: return ARM::S1;
-  case ARM::S2: return ARM::S3;
-  case ARM::S4: return ARM::S5;
-  case ARM::S6: return ARM::S7;
-  case ARM::S8: return ARM::S9;
-  case ARM::S10: return ARM::S11;
-  case ARM::S12: return ARM::S13;
-  case ARM::S14: return ARM::S15;
-  case ARM::S16: return ARM::S17;
-  case ARM::S18: return ARM::S19;
-  case ARM::S20: return ARM::S21;
-  case ARM::S22: return ARM::S23;
-  case ARM::S24: return ARM::S25;
-  case ARM::S26: return ARM::S27;
-  case ARM::S28: return ARM::S29;
-  case ARM::S30: return ARM::S31;
-
-  case ARM::D0: return ARM::D1;
-  case ARM::D2: return ARM::D3;
-  case ARM::D4: return ARM::D5;
-  case ARM::D6: return ARM::D7;
-  case ARM::D8: return ARM::D9;
-  case ARM::D10: return ARM::D11;
-  case ARM::D12: return ARM::D13;
-  case ARM::D14: return ARM::D15;
-  case ARM::D16: return ARM::D17;
-  case ARM::D18: return ARM::D19;
-  case ARM::D20: return ARM::D21;
-  case ARM::D22: return ARM::D23;
-  case ARM::D24: return ARM::D25;
-  case ARM::D26: return ARM::D27;
-  case ARM::D28: return ARM::D29;
-  case ARM::D30: return ARM::D31;
-  }
-
-  return 0;
-}
-
 /// emitLoadConstPool - Emits a load from constpool to materialize the
 /// specified immediate.
 void ARMBaseRegisterInfo::
@@ -611,65 +398,7 @@ requiresFrameIndexScavenging(const MachineFunction &MF) const {
 
 bool ARMBaseRegisterInfo::
 requiresVirtualBaseRegisters(const MachineFunction &MF) const {
-  return EnableLocalStackAlloc;
-}
-
-static void
-emitSPUpdate(bool isARM,
-             MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
-             DebugLoc dl, const ARMBaseInstrInfo &TII,
-             int NumBytes,
-             ARMCC::CondCodes Pred = ARMCC::AL, unsigned PredReg = 0) {
-  if (isARM)
-    emitARMRegPlusImmediate(MBB, MBBI, dl, ARM::SP, ARM::SP, NumBytes,
-                            Pred, PredReg, TII);
-  else
-    emitT2RegPlusImmediate(MBB, MBBI, dl, ARM::SP, ARM::SP, NumBytes,
-                           Pred, PredReg, TII);
-}
-
-
-void ARMBaseRegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-  if (!TFI->hasReservedCallFrame(MF)) {
-    // If we have alloca, convert as follows:
-    // ADJCALLSTACKDOWN -> sub, sp, sp, amount
-    // ADJCALLSTACKUP   -> add, sp, sp, amount
-    MachineInstr *Old = I;
-    DebugLoc dl = Old->getDebugLoc();
-    unsigned Amount = Old->getOperand(0).getImm();
-    if (Amount != 0) {
-      // We need to keep the stack aligned properly.  To do this, we round the
-      // amount of space needed for the outgoing arguments up to the next
-      // alignment boundary.
-      unsigned Align = TFI->getStackAlignment();
-      Amount = (Amount+Align-1)/Align*Align;
-
-      ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
-      assert(!AFI->isThumb1OnlyFunction() &&
-             "This eliminateCallFramePseudoInstr does not support Thumb1!");
-      bool isARM = !AFI->isThumbFunction();
-
-      // Replace the pseudo instruction with a new instruction...
-      unsigned Opc = Old->getOpcode();
-      int PIdx = Old->findFirstPredOperandIdx();
-      ARMCC::CondCodes Pred = (PIdx == -1)
-        ? ARMCC::AL : (ARMCC::CondCodes)Old->getOperand(PIdx).getImm();
-      if (Opc == ARM::ADJCALLSTACKDOWN || Opc == ARM::tADJCALLSTACKDOWN) {
-        // Note: PredReg is operand 2 for ADJCALLSTACKDOWN.
-        unsigned PredReg = Old->getOperand(2).getReg();
-        emitSPUpdate(isARM, MBB, I, dl, TII, -Amount, Pred, PredReg);
-      } else {
-        // Note: PredReg is operand 3 for ADJCALLSTACKUP.
-        unsigned PredReg = Old->getOperand(3).getReg();
-        assert(Opc == ARM::ADJCALLSTACKUP || Opc == ARM::tADJCALLSTACKUP);
-        emitSPUpdate(isARM, MBB, I, dl, TII, Amount, Pred, PredReg);
-      }
-    }
-  }
-  MBB.erase(I);
+  return true;
 }
 
 int64_t ARMBaseRegisterInfo::
@@ -750,8 +479,6 @@ needsFrameBaseReg(MachineInstr *MI, int64_t Offset) const {
   case ARM::VLDRS: case ARM::VLDRD:
   case ARM::VSTRS: case ARM::VSTRD:
   case ARM::tSTRspi: case ARM::tLDRspi:
-    if (ForceAllBaseRegAlloc)
-      return true;
     break;
   default:
     return false;
@@ -933,8 +660,8 @@ bool ARMBaseRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI,
 
 void
 ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
-                                         int SPAdj, RegScavenger *RS) const {
-  unsigned i = 0;
+                                         int SPAdj, unsigned FIOperandNum,
+                                         RegScavenger *RS) const {
   MachineInstr &MI = *II;
   MachineBasicBlock &MBB = *MI.getParent();
   MachineFunction &MF = *MBB.getParent();
@@ -943,13 +670,7 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
   assert(!AFI->isThumb1OnlyFunction() &&
          "This eliminateFrameIndex does not support Thumb1!");
-
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
-  }
-
-  int FrameIndex = MI.getOperand(i).getIndex();
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
   unsigned FrameReg;
 
   int Offset = TFI->ResolveFrameIndexReference(MF, FrameIndex, FrameReg, SPAdj);
@@ -959,7 +680,7 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   // means the stack pointer cannot be used to access the emergency spill slot
   // when !hasReservedCallFrame().
 #ifndef NDEBUG
-  if (RS && FrameReg == ARM::SP && FrameIndex == RS->getScavengingFrameIndex()){
+  if (RS && FrameReg == ARM::SP && RS->isScavengingFrameIndex(FrameIndex)){
     assert(TFI->hasReservedCallFrame(MF) &&
            "Cannot use SP to access the emergency spill slot in "
            "functions without a reserved call frame");
@@ -971,18 +692,18 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
 
   // Special handling of dbg_value instructions.
   if (MI.isDebugValue()) {
-    MI.getOperand(i).  ChangeToRegister(FrameReg, false /*isDef*/);
-    MI.getOperand(i+1).ChangeToImmediate(Offset);
+    MI.getOperand(FIOperandNum).  ChangeToRegister(FrameReg, false /*isDef*/);
+    MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset);
     return;
   }
 
   // Modify MI as necessary to handle as much of 'Offset' as possible
   bool Done = false;
   if (!AFI->isThumbFunction())
-    Done = rewriteARMFrameIndex(MI, i, FrameReg, Offset, TII);
+    Done = rewriteARMFrameIndex(MI, FIOperandNum, FrameReg, Offset, TII);
   else {
     assert(AFI->isThumb2Function());
-    Done = rewriteT2FrameIndex(MI, i, FrameReg, Offset, TII);
+    Done = rewriteT2FrameIndex(MI, FIOperandNum, FrameReg, Offset, TII);
   }
   if (Done)
     return;
@@ -1002,7 +723,7 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   unsigned PredReg = (PIdx == -1) ? 0 : MI.getOperand(PIdx+1).getReg();
   if (Offset == 0)
     // Must be addrmode4/6.
-    MI.getOperand(i).ChangeToRegister(FrameReg, false, false, false);
+    MI.getOperand(FIOperandNum).ChangeToRegister(FrameReg, false, false, false);
   else {
     ScratchReg = MF.getRegInfo().createVirtualRegister(&ARM::GPRRegClass);
     if (!AFI->isThumbFunction())
@@ -1014,6 +735,6 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
                              Offset, Pred, PredReg, TII);
     }
     // Update the original instruction to use the scratch register.
-    MI.getOperand(i).ChangeToRegister(ScratchReg, false, false, true);
+    MI.getOperand(FIOperandNum).ChangeToRegister(ScratchReg, false, false,true);
   }
 }
diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.h b/lib/Target/ARM/ARMBaseRegisterInfo.h
index e2bdd046db57..725033b7e573 100644
--- a/lib/Target/ARM/ARMBaseRegisterInfo.h
+++ b/lib/Target/ARM/ARMBaseRegisterInfo.h
@@ -111,12 +111,11 @@ public:
   unsigned getRegPressureLimit(const TargetRegisterClass *RC,
                                MachineFunction &MF) const;
 
-  ArrayRef<uint16_t> getRawAllocationOrder(const TargetRegisterClass *RC,
-                                           unsigned HintType, unsigned HintReg,
-                                           const MachineFunction &MF) const;
-
-  unsigned ResolveRegAllocHint(unsigned Type, unsigned Reg,
-                               const MachineFunction &MF) const;
+  void getRegAllocationHints(unsigned VirtReg,
+                             ArrayRef<MCPhysReg> Order,
+                             SmallVectorImpl<MCPhysReg> &Hints,
+                             const MachineFunction &MF,
+                             const VirtRegMap *VRM) const;
 
   void UpdateRegAllocHint(unsigned Reg, unsigned NewReg,
                           MachineFunction &MF) const;
@@ -169,17 +168,9 @@ public:
 
   virtual bool requiresVirtualBaseRegisters(const MachineFunction &MF) const;
 
-  virtual void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                           MachineBasicBlock &MBB,
-                                           MachineBasicBlock::iterator I) const;
-
   virtual void eliminateFrameIndex(MachineBasicBlock::iterator II,
-                                   int SPAdj, RegScavenger *RS = NULL) const;
-
-private:
-  unsigned getRegisterPairEven(unsigned Reg, const MachineFunction &MF) const;
-
-  unsigned getRegisterPairOdd(unsigned Reg, const MachineFunction &MF) const;
+                                   int SPAdj, unsigned FIOperandNum,
+                                   RegScavenger *RS = NULL) const;
 };
 
 } // end namespace llvm
diff --git a/lib/Target/ARM/ARMCallingConv.h b/lib/Target/ARM/ARMCallingConv.h
index 0bd1c3ee2feb..e6e8c3d5fac6 100644
--- a/lib/Target/ARM/ARMCallingConv.h
+++ b/lib/Target/ARM/ARMCallingConv.h
@@ -18,8 +18,8 @@
 #include "ARM.h"
 #include "ARMBaseInstrInfo.h"
 #include "ARMSubtarget.h"
-#include "llvm/CallingConv.h"
 #include "llvm/CodeGen/CallingConvLower.h"
+#include "llvm/IR/CallingConv.h"
 #include "llvm/Target/TargetInstrInfo.h"
 
 namespace llvm {
diff --git a/lib/Target/ARM/ARMCodeEmitter.cpp b/lib/Target/ARM/ARMCodeEmitter.cpp
index 6adbf4f27e6e..5e8e1739a984 100644
--- a/lib/Target/ARM/ARMCodeEmitter.cpp
+++ b/lib/Target/ARM/ARMCodeEmitter.cpp
@@ -14,16 +14,13 @@
 
 #define DEBUG_TYPE "jit"
 #include "ARM.h"
-#include "ARMConstantPoolValue.h"
 #include "ARMBaseInstrInfo.h"
+#include "ARMConstantPoolValue.h"
 #include "ARMRelocations.h"
 #include "ARMSubtarget.h"
 #include "ARMTargetMachine.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/PassManager.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/JITCodeEmitter.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
@@ -31,7 +28,10 @@
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/Passes.h"
-#include "llvm/ADT/Statistic.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
@@ -371,12 +371,16 @@ FunctionPass *llvm::createARMJITCodeEmitterPass(ARMBaseTargetMachine &TM,
 }
 
 bool ARMCodeEmitter::runOnMachineFunction(MachineFunction &MF) {
-  assert((MF.getTarget().getRelocationModel() != Reloc::Default ||
-          MF.getTarget().getRelocationModel() != Reloc::Static) &&
+  TargetMachine &Target = const_cast<TargetMachine&>(MF.getTarget());
+
+  assert((Target.getRelocationModel() != Reloc::Default ||
+          Target.getRelocationModel() != Reloc::Static) &&
          "JIT relocation model must be set to static or default!");
-  JTI = ((ARMBaseTargetMachine &)MF.getTarget()).getJITInfo();
-  II = (const ARMBaseInstrInfo *)MF.getTarget().getInstrInfo();
-  TD = MF.getTarget().getDataLayout();
+
+  JTI = static_cast<ARMJITInfo*>(Target.getJITInfo());
+  II = static_cast<const ARMBaseInstrInfo*>(Target.getInstrInfo());
+  TD = Target.getDataLayout();
+
   Subtarget = &TM.getSubtarget<ARMSubtarget>();
   MCPEs = &MF.getConstantPool()->getConstants();
   MJTEs = 0;
diff --git a/lib/Target/ARM/ARMConstantIslandPass.cpp b/lib/Target/ARM/ARMConstantIslandPass.cpp
index a57368fdb5d8..4891609b336f 100644
--- a/lib/Target/ARM/ARMConstantIslandPass.cpp
+++ b/lib/Target/ARM/ARMConstantIslandPass.cpp
@@ -16,23 +16,23 @@
 #define DEBUG_TYPE "arm-cp-islands"
 #include "ARM.h"
 #include "ARMMachineFunctionInfo.h"
-#include "Thumb2InstrInfo.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
+#include "Thumb2InstrInfo.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetMachine.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/Format.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/SmallSet.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/STLExtras.h"
-#include "llvm/ADT/Statistic.h"
-#include "llvm/Support/CommandLine.h"
+#include "llvm/Target/TargetMachine.h"
 #include <algorithm>
 using namespace llvm;
 
@@ -1468,7 +1468,7 @@ void ARMConstantIslands::removeDeadCPEMI(MachineInstr *CPEMI) {
   if (CPEBB->empty()) {
     BBInfo[CPEBB->getNumber()].Size = 0;
 
-    // This block no longer needs to be aligned. <rdar://problem/10534709>.
+    // This block no longer needs to be aligned.
     CPEBB->setAlignment(0);
   } else
     // Entries are sorted by descending alignment, so realign from the front.
diff --git a/lib/Target/ARM/ARMConstantPoolValue.cpp b/lib/Target/ARM/ARMConstantPoolValue.cpp
index fa3226e37eb9..4e703ec3c1a8 100644
--- a/lib/Target/ARM/ARMConstantPoolValue.cpp
+++ b/lib/Target/ARM/ARMConstantPoolValue.cpp
@@ -13,11 +13,11 @@
 
 #include "ARMConstantPoolValue.h"
 #include "llvm/ADT/FoldingSet.h"
-#include "llvm/Constant.h"
-#include "llvm/Constants.h"
-#include "llvm/GlobalValue.h"
-#include "llvm/Type.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/IR/Constant.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Support/raw_ostream.h"
 #include <cstdlib>
 using namespace llvm;
@@ -206,11 +206,7 @@ ARMConstantPoolSymbol::ARMConstantPoolSymbol(LLVMContext &C, const char *s,
                                              bool AddCurrentAddress)
   : ARMConstantPoolValue(C, id, ARMCP::CPExtSymbol, PCAdj, Modifier,
                          AddCurrentAddress),
-    S(strdup(s)) {}
-
-ARMConstantPoolSymbol::~ARMConstantPoolSymbol() {
-  free((void*)S);
-}
+    S(s) {}
 
 ARMConstantPoolSymbol *
 ARMConstantPoolSymbol::Create(LLVMContext &C, const char *s,
@@ -218,14 +214,6 @@ ARMConstantPoolSymbol::Create(LLVMContext &C, const char *s,
   return new ARMConstantPoolSymbol(C, s, ID, PCAdj, ARMCP::no_modifier, false);
 }
 
-static bool CPV_streq(const char *S1, const char *S2) {
-  if (S1 == S2)
-    return true;
-  if (S1 && S2 && strcmp(S1, S2) == 0)
-    return true;
-  return false;
-}
-
 int ARMConstantPoolSymbol::getExistingMachineCPValue(MachineConstantPool *CP,
                                                      unsigned Alignment) {
   unsigned AlignMask = Alignment - 1;
@@ -238,7 +226,7 @@ int ARMConstantPoolSymbol::getExistingMachineCPValue(MachineConstantPool *CP,
       ARMConstantPoolSymbol *APS = dyn_cast<ARMConstantPoolSymbol>(CPV);
       if (!APS) continue;
 
-      if (CPV_streq(APS->S, S) && equals(APS))
+      if (APS->S == S && equals(APS))
         return i;
     }
   }
@@ -248,12 +236,11 @@ int ARMConstantPoolSymbol::getExistingMachineCPValue(MachineConstantPool *CP,
 
 bool ARMConstantPoolSymbol::hasSameValue(ARMConstantPoolValue *ACPV) {
   const ARMConstantPoolSymbol *ACPS = dyn_cast<ARMConstantPoolSymbol>(ACPV);
-  return ACPS && CPV_streq(ACPS->S, S) &&
-    ARMConstantPoolValue::hasSameValue(ACPV);
+  return ACPS && ACPS->S == S && ARMConstantPoolValue::hasSameValue(ACPV);
 }
 
 void ARMConstantPoolSymbol::addSelectionDAGCSEId(FoldingSetNodeID &ID) {
-  ID.AddPointer(S);
+  ID.AddString(S);
   ARMConstantPoolValue::addSelectionDAGCSEId(ID);
 }
 
diff --git a/lib/Target/ARM/ARMConstantPoolValue.h b/lib/Target/ARM/ARMConstantPoolValue.h
index ae531c4ea888..93812fe6bb37 100644
--- a/lib/Target/ARM/ARMConstantPoolValue.h
+++ b/lib/Target/ARM/ARMConstantPoolValue.h
@@ -161,19 +161,17 @@ public:
 /// ARMConstantPoolSymbol - ARM-specific constantpool values for external
 /// symbols.
 class ARMConstantPoolSymbol : public ARMConstantPoolValue {
-  const char *S;                // ExtSymbol being loaded.
+  const std::string S;          // ExtSymbol being loaded.
 
   ARMConstantPoolSymbol(LLVMContext &C, const char *s, unsigned id,
                         unsigned char PCAdj, ARMCP::ARMCPModifier Modifier,
                         bool AddCurrentAddress);
 
 public:
-  ~ARMConstantPoolSymbol();
-
   static ARMConstantPoolSymbol *Create(LLVMContext &C, const char *s,
                                        unsigned ID, unsigned char PCAdj);
 
-  const char *getSymbol() const { return S; }
+  const char *getSymbol() const { return S.c_str(); }
 
   virtual int getExistingMachineCPValue(MachineConstantPool *CP,
                                         unsigned Alignment);
diff --git a/lib/Target/ARM/ARMExpandPseudoInsts.cpp b/lib/Target/ARM/ARMExpandPseudoInsts.cpp
index 8c45e0b98d8e..beb843ca9aa8 100644
--- a/lib/Target/ARM/ARMExpandPseudoInsts.cpp
+++ b/lib/Target/ARM/ARMExpandPseudoInsts.cpp
@@ -23,10 +23,10 @@
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/raw_ostream.h" // FIXME: for debug only. remove!
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 using namespace llvm;
 
 static cl::opt<bool>
diff --git a/lib/Target/ARM/ARMFastISel.cpp b/lib/Target/ARM/ARMFastISel.cpp
index 6611862ca071..29fcd4009af3 100644
--- a/lib/Target/ARM/ARMFastISel.cpp
+++ b/lib/Target/ARM/ARMFastISel.cpp
@@ -16,31 +16,31 @@
 #include "ARM.h"
 #include "ARMBaseInstrInfo.h"
 #include "ARMCallingConv.h"
-#include "ARMTargetMachine.h"
-#include "ARMSubtarget.h"
 #include "ARMConstantPoolValue.h"
+#include "ARMSubtarget.h"
+#include "ARMTargetMachine.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/CallingConv.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/Instructions.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/Module.h"
-#include "llvm/Operator.h"
 #include "llvm/CodeGen/Analysis.h"
 #include "llvm/CodeGen/FastISel.h"
 #include "llvm/CodeGen/FunctionLoweringInfo.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineMemOperand.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Operator.h"
 #include "llvm/Support/CallSite.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/GetElementPtrTypeIterator.h"
-#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetLowering.h"
 #include "llvm/Target/TargetMachine.h"
@@ -146,6 +146,7 @@ class ARMFastISel : public FastISel {
     virtual unsigned TargetMaterializeAlloca(const AllocaInst *AI);
     virtual bool TryToFoldLoad(MachineInstr *MI, unsigned OpNo,
                                const LoadInst *LI);
+    virtual bool FastLowerArguments();
   private:
   #include "ARMGenFastISel.inc"
 
@@ -178,23 +179,24 @@ class ARMFastISel : public FastISel {
     bool isLoadTypeLegal(Type *Ty, MVT &VT);
     bool ARMEmitCmp(const Value *Src1Value, const Value *Src2Value,
                     bool isZExt);
-    bool ARMEmitLoad(EVT VT, unsigned &ResultReg, Address &Addr,
+    bool ARMEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr,
                      unsigned Alignment = 0, bool isZExt = true,
                      bool allocReg = true);
-    bool ARMEmitStore(EVT VT, unsigned SrcReg, Address &Addr,
+    bool ARMEmitStore(MVT VT, unsigned SrcReg, Address &Addr,
                       unsigned Alignment = 0);
     bool ARMComputeAddress(const Value *Obj, Address &Addr);
-    void ARMSimplifyAddress(Address &Addr, EVT VT, bool useAM3);
+    void ARMSimplifyAddress(Address &Addr, MVT VT, bool useAM3);
     bool ARMIsMemCpySmall(uint64_t Len);
-    bool ARMTryEmitSmallMemCpy(Address Dest, Address Src, uint64_t Len);
-    unsigned ARMEmitIntExt(EVT SrcVT, unsigned SrcReg, EVT DestVT, bool isZExt);
-    unsigned ARMMaterializeFP(const ConstantFP *CFP, EVT VT);
-    unsigned ARMMaterializeInt(const Constant *C, EVT VT);
-    unsigned ARMMaterializeGV(const GlobalValue *GV, EVT VT);
-    unsigned ARMMoveToFPReg(EVT VT, unsigned SrcReg);
-    unsigned ARMMoveToIntReg(EVT VT, unsigned SrcReg);
+    bool ARMTryEmitSmallMemCpy(Address Dest, Address Src, uint64_t Len,
+                               unsigned Alignment);
+    unsigned ARMEmitIntExt(MVT SrcVT, unsigned SrcReg, MVT DestVT, bool isZExt);
+    unsigned ARMMaterializeFP(const ConstantFP *CFP, MVT VT);
+    unsigned ARMMaterializeInt(const Constant *C, MVT VT);
+    unsigned ARMMaterializeGV(const GlobalValue *GV, MVT VT);
+    unsigned ARMMoveToFPReg(MVT VT, unsigned SrcReg);
+    unsigned ARMMoveToIntReg(MVT VT, unsigned SrcReg);
     unsigned ARMSelectCallOp(bool UseReg);
-    unsigned ARMLowerPICELF(const GlobalValue *GV, unsigned Align, EVT VT);
+    unsigned ARMLowerPICELF(const GlobalValue *GV, unsigned Align, MVT VT);
 
     // Call handling routines.
   private:
@@ -220,7 +222,7 @@ class ARMFastISel : public FastISel {
     bool isARMNEONPred(const MachineInstr *MI);
     bool DefinesOptionalPredicate(MachineInstr *MI, bool *CPSR);
     const MachineInstrBuilder &AddOptionalDefs(const MachineInstrBuilder &MIB);
-    void AddLoadStoreOperands(EVT VT, Address &Addr,
+    void AddLoadStoreOperands(MVT VT, Address &Addr,
                               const MachineInstrBuilder &MIB,
                               unsigned Flags, bool useAM3);
 };
@@ -486,7 +488,7 @@ unsigned ARMFastISel::FastEmitInst_extractsubreg(MVT RetVT,
 
 // TODO: Don't worry about 64-bit now, but when this is fixed remove the
 // checks from the various callers.
-unsigned ARMFastISel::ARMMoveToFPReg(EVT VT, unsigned SrcReg) {
+unsigned ARMFastISel::ARMMoveToFPReg(MVT VT, unsigned SrcReg) {
   if (VT == MVT::f64) return 0;
 
   unsigned MoveReg = createResultReg(TLI.getRegClassFor(VT));
@@ -496,7 +498,7 @@ unsigned ARMFastISel::ARMMoveToFPReg(EVT VT, unsigned SrcReg) {
   return MoveReg;
 }
 
-unsigned ARMFastISel::ARMMoveToIntReg(EVT VT, unsigned SrcReg) {
+unsigned ARMFastISel::ARMMoveToIntReg(MVT VT, unsigned SrcReg) {
   if (VT == MVT::i64) return 0;
 
   unsigned MoveReg = createResultReg(TLI.getRegClassFor(VT));
@@ -509,7 +511,7 @@ unsigned ARMFastISel::ARMMoveToIntReg(EVT VT, unsigned SrcReg) {
 // For double width floating point we need to materialize two constants
 // (the high and the low) into integer registers then use a move to get
 // the combined constant into an FP reg.
-unsigned ARMFastISel::ARMMaterializeFP(const ConstantFP *CFP, EVT VT) {
+unsigned ARMFastISel::ARMMaterializeFP(const ConstantFP *CFP, MVT VT) {
   const APFloat Val = CFP->getValueAPF();
   bool is64bit = VT == MVT::f64;
 
@@ -553,7 +555,7 @@ unsigned ARMFastISel::ARMMaterializeFP(const ConstantFP *CFP, EVT VT) {
   return DestReg;
 }
 
-unsigned ARMFastISel::ARMMaterializeInt(const Constant *C, EVT VT) {
+unsigned ARMFastISel::ARMMaterializeInt(const Constant *C, MVT VT) {
 
   if (VT != MVT::i32 && VT != MVT::i16 && VT != MVT::i8 && VT != MVT::i1)
     return false;
@@ -563,7 +565,9 @@ unsigned ARMFastISel::ARMMaterializeInt(const Constant *C, EVT VT) {
   const ConstantInt *CI = cast<ConstantInt>(C);
   if (Subtarget->hasV6T2Ops() && isUInt<16>(CI->getZExtValue())) {
     unsigned Opc = isThumb2 ? ARM::t2MOVi16 : ARM::MOVi16;
-    unsigned ImmReg = createResultReg(TLI.getRegClassFor(MVT::i32));
+    const TargetRegisterClass *RC = isThumb2 ? &ARM::rGPRRegClass :
+      &ARM::GPRRegClass;
+    unsigned ImmReg = createResultReg(RC);
     AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                             TII.get(Opc), ImmReg)
                     .addImm(CI->getZExtValue()));
@@ -613,7 +617,7 @@ unsigned ARMFastISel::ARMMaterializeInt(const Constant *C, EVT VT) {
   return DestReg;
 }
 
-unsigned ARMFastISel::ARMMaterializeGV(const GlobalValue *GV, EVT VT) {
+unsigned ARMFastISel::ARMMaterializeGV(const GlobalValue *GV, MVT VT) {
   // For now 32-bit only.
   if (VT != MVT::i32) return 0;
 
@@ -716,10 +720,11 @@ unsigned ARMFastISel::ARMMaterializeGV(const GlobalValue *GV, EVT VT) {
 }
 
 unsigned ARMFastISel::TargetMaterializeConstant(const Constant *C) {
-  EVT VT = TLI.getValueType(C->getType(), true);
+  EVT CEVT = TLI.getValueType(C->getType(), true);
 
   // Only handle simple types.
-  if (!VT.isSimple()) return 0;
+  if (!CEVT.isSimple()) return 0;
+  MVT VT = CEVT.getSimpleVT();
 
   if (const ConstantFP *CFP = dyn_cast<ConstantFP>(C))
     return ARMMaterializeFP(CFP, VT);
@@ -895,12 +900,9 @@ bool ARMFastISel::ARMComputeAddress(const Value *Obj, Address &Addr) {
   return Addr.Base.Reg != 0;
 }
 
-void ARMFastISel::ARMSimplifyAddress(Address &Addr, EVT VT, bool useAM3) {
-
-  assert(VT.isSimple() && "Non-simple types are invalid here!");
-
+void ARMFastISel::ARMSimplifyAddress(Address &Addr, MVT VT, bool useAM3) {
   bool needsLowering = false;
-  switch (VT.getSimpleVT().SimpleTy) {
+  switch (VT.SimpleTy) {
     default: llvm_unreachable("Unhandled load/store type!");
     case MVT::i1:
     case MVT::i8:
@@ -951,13 +953,12 @@ void ARMFastISel::ARMSimplifyAddress(Address &Addr, EVT VT, bool useAM3) {
   }
 }
 
-void ARMFastISel::AddLoadStoreOperands(EVT VT, Address &Addr,
+void ARMFastISel::AddLoadStoreOperands(MVT VT, Address &Addr,
                                        const MachineInstrBuilder &MIB,
                                        unsigned Flags, bool useAM3) {
   // addrmode5 output depends on the selection dag addressing dividing the
   // offset by 4 that it then later multiplies. Do this here as well.
-  if (VT.getSimpleVT().SimpleTy == MVT::f32 ||
-      VT.getSimpleVT().SimpleTy == MVT::f64)
+  if (VT.SimpleTy == MVT::f32 || VT.SimpleTy == MVT::f64)
     Addr.Offset /= 4;
 
   // Frame base works a bit differently. Handle it separately.
@@ -1000,14 +1001,13 @@ void ARMFastISel::AddLoadStoreOperands(EVT VT, Address &Addr,
   AddOptionalDefs(MIB);
 }
 
-bool ARMFastISel::ARMEmitLoad(EVT VT, unsigned &ResultReg, Address &Addr,
+bool ARMFastISel::ARMEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr,
                               unsigned Alignment, bool isZExt, bool allocReg) {
-  assert(VT.isSimple() && "Non-simple types are invalid here!");
   unsigned Opc;
   bool useAM3 = false;
   bool needVMOV = false;
   const TargetRegisterClass *RC;
-  switch (VT.getSimpleVT().SimpleTy) {
+  switch (VT.SimpleTy) {
     // This is mostly going to be Neon/vector support.
     default: return false;
     case MVT::i1:
@@ -1124,11 +1124,11 @@ bool ARMFastISel::SelectLoad(const Instruction *I) {
   return true;
 }
 
-bool ARMFastISel::ARMEmitStore(EVT VT, unsigned SrcReg, Address &Addr,
+bool ARMFastISel::ARMEmitStore(MVT VT, unsigned SrcReg, Address &Addr,
                                unsigned Alignment) {
   unsigned StrOpc;
   bool useAM3 = false;
-  switch (VT.getSimpleVT().SimpleTy) {
+  switch (VT.SimpleTy) {
     // This is mostly going to be Neon/vector support.
     default: return false;
     case MVT::i1: {
@@ -1402,8 +1402,9 @@ bool ARMFastISel::SelectIndirectBr(const Instruction *I) {
 bool ARMFastISel::ARMEmitCmp(const Value *Src1Value, const Value *Src2Value,
                              bool isZExt) {
   Type *Ty = Src1Value->getType();
-  EVT SrcVT = TLI.getValueType(Ty, true);
-  if (!SrcVT.isSimple()) return false;
+  EVT SrcEVT = TLI.getValueType(Ty, true);
+  if (!SrcEVT.isSimple()) return false;
+  MVT SrcVT = SrcEVT.getSimpleVT();
 
   bool isFloat = (Ty->isFloatTy() || Ty->isDoubleTy());
   if (isFloat && !Subtarget->hasVFP2())
@@ -1440,7 +1441,7 @@ bool ARMFastISel::ARMEmitCmp(const Value *Src1Value, const Value *Src2Value,
   unsigned CmpOpc;
   bool isICmp = true;
   bool needsExt = false;
-  switch (SrcVT.getSimpleVT().SimpleTy) {
+  switch (SrcVT.SimpleTy) {
     default: return false;
     // TODO: Verify compares.
     case MVT::f32:
@@ -1592,7 +1593,10 @@ bool ARMFastISel::SelectIToFP(const Instruction *I, bool isSigned) {
     return false;
 
   Value *Src = I->getOperand(0);
-  EVT SrcVT = TLI.getValueType(Src->getType(), true);
+  EVT SrcEVT = TLI.getValueType(Src->getType(), true);
+  if (!SrcEVT.isSimple())
+    return false;
+  MVT SrcVT = SrcEVT.getSimpleVT();
   if (SrcVT != MVT::i32 && SrcVT != MVT::i16 && SrcVT != MVT::i8)
     return false;
 
@@ -1601,8 +1605,7 @@ bool ARMFastISel::SelectIToFP(const Instruction *I, bool isSigned) {
 
   // Handle sign-extension.
   if (SrcVT == MVT::i16 || SrcVT == MVT::i8) {
-    EVT DestVT = MVT::i32;
-    SrcReg = ARMEmitIntExt(SrcVT, SrcReg, DestVT,
+    SrcReg = ARMEmitIntExt(SrcVT, SrcReg, MVT::i32,
                                        /*isZExt*/!isSigned);
     if (SrcReg == 0) return false;
   }
@@ -1665,7 +1668,6 @@ bool ARMFastISel::SelectSelect(const Instruction *I) {
 
   // Things need to be register sized for register moves.
   if (VT != MVT::i32) return false;
-  const TargetRegisterClass *RC = TLI.getRegClassFor(VT);
 
   unsigned CondReg = getRegForValue(I->getOperand(0));
   if (CondReg == 0) return false;
@@ -1698,14 +1700,16 @@ bool ARMFastISel::SelectSelect(const Instruction *I) {
                   .addReg(CondReg).addImm(0));
 
   unsigned MovCCOpc;
+  const TargetRegisterClass *RC;
   if (!UseImm) {
+    RC = isThumb2 ? &ARM::tGPRRegClass : &ARM::GPRRegClass;
     MovCCOpc = isThumb2 ? ARM::t2MOVCCr : ARM::MOVCCr;
   } else {
-    if (!isNegativeImm) {
+    RC = isThumb2 ? &ARM::rGPRRegClass : &ARM::GPRRegClass;
+    if (!isNegativeImm)
       MovCCOpc = isThumb2 ? ARM::t2MOVCCi : ARM::MOVCCi;
-    } else {
+    else
       MovCCOpc = isThumb2 ? ARM::t2MVNCCi : ARM::MVNCCi;
-    }
   }
   unsigned ResultReg = createResultReg(RC);
   if (!UseImm)
@@ -1807,7 +1811,9 @@ bool ARMFastISel::SelectBinaryIntOp(const Instruction *I, unsigned ISDOpcode) {
 }
 
 bool ARMFastISel::SelectBinaryFPOp(const Instruction *I, unsigned ISDOpcode) {
-  EVT VT  = TLI.getValueType(I->getType(), true);
+  EVT FPVT = TLI.getValueType(I->getType(), true);
+  if (!FPVT.isSimple()) return false;
+  MVT VT = FPVT.getSimpleVT();
 
   // We can get here in the case when we want to use NEON for our fp
   // operations, but can't figure out how to. Just use the vfp instructions
@@ -1838,7 +1844,7 @@ bool ARMFastISel::SelectBinaryFPOp(const Instruction *I, unsigned ISDOpcode) {
   unsigned Op2 = getRegForValue(I->getOperand(1));
   if (Op2 == 0) return false;
 
-  unsigned ResultReg = createResultReg(TLI.getRegClassFor(VT));
+  unsigned ResultReg = createResultReg(TLI.getRegClassFor(VT.SimpleTy));
   AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                           TII.get(Opc), ResultReg)
                   .addReg(Op1).addReg(Op2));
@@ -2051,7 +2057,7 @@ bool ARMFastISel::FinishCall(MVT RetVT, SmallVectorImpl<unsigned> &UsedRegs,
     if (RVLocs.size() == 2 && RetVT == MVT::f64) {
       // For this move we copy into two registers and then move into the
       // double fp reg we want.
-      EVT DestVT = RVLocs[0].getValVT();
+      MVT DestVT = RVLocs[0].getValVT();
       const TargetRegisterClass* DstRC = TLI.getRegClassFor(DestVT);
       unsigned ResultReg = createResultReg(DstRC);
       AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
@@ -2066,7 +2072,7 @@ bool ARMFastISel::FinishCall(MVT RetVT, SmallVectorImpl<unsigned> &UsedRegs,
       UpdateValueMap(I, ResultReg);
     } else {
       assert(RVLocs.size() == 1 &&"Can't handle non-double multi-reg retvals!");
-      EVT CopyVT = RVLocs[0].getValVT();
+      MVT CopyVT = RVLocs[0].getValVT();
 
       // Special handling for extended integers.
       if (RetVT == MVT::i1 || RetVT == MVT::i8 || RetVT == MVT::i16)
@@ -2094,11 +2100,13 @@ bool ARMFastISel::SelectRet(const Instruction *I) {
   if (!FuncInfo.CanLowerReturn)
     return false;
 
+  // Build a list of return value registers.
+  SmallVector<unsigned, 4> RetRegs;
+
   CallingConv::ID CC = F.getCallingConv();
   if (Ret->getNumOperands() > 0) {
     SmallVector<ISD::OutputArg, 4> Outs;
-    GetReturnInfo(F.getReturnType(), F.getAttributes().getRetAttributes(),
-                  Outs, TLI);
+    GetReturnInfo(F.getReturnType(), F.getAttributes(), Outs, TLI);
 
     // Analyze operands of the call, assigning locations to each operand.
     SmallVector<CCValAssign, 16> ValLocs;
@@ -2125,8 +2133,10 @@ bool ARMFastISel::SelectRet(const Instruction *I) {
       return false;
 
     unsigned SrcReg = Reg + VA.getValNo();
-    EVT RVVT = TLI.getValueType(RV->getType());
-    EVT DestVT = VA.getValVT();
+    EVT RVEVT = TLI.getValueType(RV->getType());
+    if (!RVEVT.isSimple()) return false;
+    MVT RVVT = RVEVT.getSimpleVT();
+    MVT DestVT = VA.getValVT();
     // Special handling for extended integers.
     if (RVVT != DestVT) {
       if (RVVT != MVT::i1 && RVVT != MVT::i8 && RVVT != MVT::i16)
@@ -2151,13 +2161,16 @@ bool ARMFastISel::SelectRet(const Instruction *I) {
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
             DstReg).addReg(SrcReg);
 
-    // Mark the register as live out of the function.
-    MRI.addLiveOut(VA.getLocReg());
+    // Add register to return instruction.
+    RetRegs.push_back(VA.getLocReg());
   }
 
   unsigned RetOpc = isThumb2 ? ARM::tBX_RET : ARM::BX_RET;
-  AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
-                          TII.get(RetOpc)));
+  MachineInstrBuilder MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+                                    TII.get(RetOpc));
+  AddOptionalDefs(MIB);
+  for (unsigned i = 0, e = RetRegs.size(); i != e; ++i)
+    MIB.addReg(RetRegs[i], RegState::Implicit);
   return true;
 }
 
@@ -2171,7 +2184,9 @@ unsigned ARMFastISel::ARMSelectCallOp(bool UseReg) {
 unsigned ARMFastISel::getLibcallReg(const Twine &Name) {
   GlobalValue *GV = new GlobalVariable(Type::getInt32Ty(*Context), false,
                                        GlobalValue::ExternalLinkage, 0, Name);
-  return ARMMaterializeGV(GV, TLI.getValueType(GV->getType()));
+  EVT LCREVT = TLI.getValueType(GV->getType());
+  if (!LCREVT.isSimple()) return 0;
+  return ARMMaterializeGV(GV, LCREVT.getSimpleVT());
 }
 
 // A quick function that will emit a call for a named libcall in F with the
@@ -2280,6 +2295,9 @@ bool ARMFastISel::SelectCall(const Instruction *I,
   // Can't handle inline asm.
   if (isa<InlineAsm>(Callee)) return false;
 
+  // Allow SelectionDAG isel to handle tail calls.
+  if (CI->isTailCall()) return false;
+
   // Check the calling convention.
   ImmutableCallSite CS(CI);
   CallingConv::ID CC = CS.getCallingConv();
@@ -2328,16 +2346,16 @@ bool ARMFastISel::SelectCall(const Instruction *I,
 
     ISD::ArgFlagsTy Flags;
     unsigned AttrInd = i - CS.arg_begin() + 1;
-    if (CS.paramHasAttr(AttrInd, Attributes::SExt))
+    if (CS.paramHasAttr(AttrInd, Attribute::SExt))
       Flags.setSExt();
-    if (CS.paramHasAttr(AttrInd, Attributes::ZExt))
+    if (CS.paramHasAttr(AttrInd, Attribute::ZExt))
       Flags.setZExt();
 
     // FIXME: Only handle *easy* calls for now.
-    if (CS.paramHasAttr(AttrInd, Attributes::InReg) ||
-        CS.paramHasAttr(AttrInd, Attributes::StructRet) ||
-        CS.paramHasAttr(AttrInd, Attributes::Nest) ||
-        CS.paramHasAttr(AttrInd, Attributes::ByVal))
+    if (CS.paramHasAttr(AttrInd, Attribute::InReg) ||
+        CS.paramHasAttr(AttrInd, Attribute::StructRet) ||
+        CS.paramHasAttr(AttrInd, Attribute::Nest) ||
+        CS.paramHasAttr(AttrInd, Attribute::ByVal))
       return false;
 
     Type *ArgTy = (*i)->getType();
@@ -2419,21 +2437,29 @@ bool ARMFastISel::ARMIsMemCpySmall(uint64_t Len) {
 }
 
 bool ARMFastISel::ARMTryEmitSmallMemCpy(Address Dest, Address Src,
-                                        uint64_t Len) {
+                                        uint64_t Len, unsigned Alignment) {
   // Make sure we don't bloat code by inlining very large memcpy's.
   if (!ARMIsMemCpySmall(Len))
     return false;
 
-  // We don't care about alignment here since we just emit integer accesses.
   while (Len) {
     MVT VT;
-    if (Len >= 4)
-      VT = MVT::i32;
-    else if (Len >= 2)
-      VT = MVT::i16;
-    else {
-      assert(Len == 1);
-      VT = MVT::i8;
+    if (!Alignment || Alignment >= 4) {
+      if (Len >= 4)
+        VT = MVT::i32;
+      else if (Len >= 2)
+        VT = MVT::i16;
+      else {
+        assert (Len == 1 && "Expected a length of 1!");
+        VT = MVT::i8;
+      }
+    } else {
+      // Bound based on alignment.
+      if (Len >= 2 && Alignment == 2)
+        VT = MVT::i16;
+      else {
+        VT = MVT::i8;
+      }
     }
 
     bool RV;
@@ -2512,7 +2538,8 @@ bool ARMFastISel::SelectIntrinsicCall(const IntrinsicInst &I) {
         if (!ARMComputeAddress(MTI.getRawDest(), Dest) ||
             !ARMComputeAddress(MTI.getRawSource(), Src))
           return false;
-        if (ARMTryEmitSmallMemCpy(Dest, Src, Len))
+        unsigned Alignment = MTI.getAlignment();
+        if (ARMTryEmitSmallMemCpy(Dest, Src, Len, Alignment))
           return true;
       }
     }
@@ -2541,7 +2568,8 @@ bool ARMFastISel::SelectIntrinsicCall(const IntrinsicInst &I) {
     return SelectCall(&I, "memset");
   }
   case Intrinsic::trap: {
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(ARM::TRAP));
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(
+      Subtarget->useNaClTrap() ? ARM::TRAPNaCl : ARM::TRAP));
     return true;
   }
   }
@@ -2570,18 +2598,19 @@ bool ARMFastISel::SelectTrunc(const Instruction *I) {
   return true;
 }
 
-unsigned ARMFastISel::ARMEmitIntExt(EVT SrcVT, unsigned SrcReg, EVT DestVT,
+unsigned ARMFastISel::ARMEmitIntExt(MVT SrcVT, unsigned SrcReg, MVT DestVT,
                                     bool isZExt) {
   if (DestVT != MVT::i32 && DestVT != MVT::i16 && DestVT != MVT::i8)
     return 0;
 
   unsigned Opc;
   bool isBoolZext = false;
-  if (!SrcVT.isSimple()) return 0;
-  switch (SrcVT.getSimpleVT().SimpleTy) {
+  const TargetRegisterClass *RC = TLI.getRegClassFor(MVT::i32);
+  switch (SrcVT.SimpleTy) {
   default: return 0;
   case MVT::i16:
     if (!Subtarget->hasV6Ops()) return 0;
+    RC = isThumb2 ? &ARM::rGPRRegClass : &ARM::GPRnopcRegClass;
     if (isZExt)
       Opc = isThumb2 ? ARM::t2UXTH : ARM::UXTH;
     else
@@ -2589,6 +2618,7 @@ unsigned ARMFastISel::ARMEmitIntExt(EVT SrcVT, unsigned SrcReg, EVT DestVT,
     break;
   case MVT::i8:
     if (!Subtarget->hasV6Ops()) return 0;
+    RC = isThumb2 ? &ARM::rGPRRegClass : &ARM::GPRnopcRegClass;
     if (isZExt)
       Opc = isThumb2 ? ARM::t2UXTB : ARM::UXTB;
     else
@@ -2596,6 +2626,7 @@ unsigned ARMFastISel::ARMEmitIntExt(EVT SrcVT, unsigned SrcReg, EVT DestVT,
     break;
   case MVT::i1:
     if (isZExt) {
+      RC = isThumb2 ? &ARM::rGPRRegClass : &ARM::GPRRegClass;
       Opc = isThumb2 ? ARM::t2ANDri : ARM::ANDri;
       isBoolZext = true;
       break;
@@ -2603,7 +2634,7 @@ unsigned ARMFastISel::ARMEmitIntExt(EVT SrcVT, unsigned SrcReg, EVT DestVT,
     return 0;
   }
 
-  unsigned ResultReg = createResultReg(TLI.getRegClassFor(MVT::i32));
+  unsigned ResultReg = createResultReg(RC);
   MachineInstrBuilder MIB;
   MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), ResultReg)
         .addReg(SrcReg);
@@ -2622,14 +2653,18 @@ bool ARMFastISel::SelectIntExt(const Instruction *I) {
   Value *Src = I->getOperand(0);
   Type *SrcTy = Src->getType();
 
-  EVT SrcVT, DestVT;
-  SrcVT = TLI.getValueType(SrcTy, true);
-  DestVT = TLI.getValueType(DestTy, true);
-
   bool isZExt = isa<ZExtInst>(I);
   unsigned SrcReg = getRegForValue(Src);
   if (!SrcReg) return false;
 
+  EVT SrcEVT, DestEVT;
+  SrcEVT = TLI.getValueType(SrcTy, true);
+  DestEVT = TLI.getValueType(DestTy, true);
+  if (!SrcEVT.isSimple()) return false;
+  if (!DestEVT.isSimple()) return false;
+
+  MVT SrcVT = SrcEVT.getSimpleVT();
+  MVT DestVT = DestEVT.getSimpleVT();
   unsigned ResultReg = ARMEmitIntExt(SrcVT, SrcReg, DestVT, isZExt);
   if (ResultReg == 0) return false;
   UpdateValueMap(I, ResultReg);
@@ -2809,7 +2844,7 @@ bool ARMFastISel::TryToFoldLoad(MachineInstr *MI, unsigned OpNo,
 }
 
 unsigned ARMFastISel::ARMLowerPICELF(const GlobalValue *GV,
-                                     unsigned Align, EVT VT) {
+                                     unsigned Align, MVT VT) {
   bool UseGOTOFF = GV->hasLocalLinkage() || GV->hasHiddenVisibility();
   ARMConstantPoolConstant *CPV =
     ARMConstantPoolConstant::Create(GV, UseGOTOFF ? ARMCP::GOTOFF : ARMCP::GOT);
@@ -2849,6 +2884,80 @@ unsigned ARMFastISel::ARMLowerPICELF(const GlobalValue *GV,
   return DestReg2;
 }
 
+bool ARMFastISel::FastLowerArguments() {
+  if (!FuncInfo.CanLowerReturn)
+    return false;
+
+  const Function *F = FuncInfo.Fn;
+  if (F->isVarArg())
+    return false;
+
+  CallingConv::ID CC = F->getCallingConv();
+  switch (CC) {
+  default:
+    return false;
+  case CallingConv::Fast:
+  case CallingConv::C:
+  case CallingConv::ARM_AAPCS_VFP:
+  case CallingConv::ARM_AAPCS:
+  case CallingConv::ARM_APCS:
+    break;
+  }
+
+  // Only handle simple cases. i.e. Up to 4 i8/i16/i32 scalar arguments
+  // which are passed in r0 - r3.
+  unsigned Idx = 1;
+  for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
+       I != E; ++I, ++Idx) {
+    if (Idx > 4)
+      return false;
+
+    if (F->getAttributes().hasAttribute(Idx, Attribute::InReg) ||
+        F->getAttributes().hasAttribute(Idx, Attribute::StructRet) ||
+        F->getAttributes().hasAttribute(Idx, Attribute::ByVal))
+      return false;
+
+    Type *ArgTy = I->getType();
+    if (ArgTy->isStructTy() || ArgTy->isArrayTy() || ArgTy->isVectorTy())
+      return false;
+
+    EVT ArgVT = TLI.getValueType(ArgTy);
+    if (!ArgVT.isSimple()) return false;
+    switch (ArgVT.getSimpleVT().SimpleTy) {
+    case MVT::i8:
+    case MVT::i16:
+    case MVT::i32:
+      break;
+    default:
+      return false;
+    }
+  }
+
+
+  static const uint16_t GPRArgRegs[] = {
+    ARM::R0, ARM::R1, ARM::R2, ARM::R3
+  };
+
+  const TargetRegisterClass *RC = TLI.getRegClassFor(MVT::i32);
+  Idx = 0;
+  for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
+       I != E; ++I, ++Idx) {
+    if (I->use_empty())
+      continue;
+    unsigned SrcReg = GPRArgRegs[Idx];
+    unsigned DstReg = FuncInfo.MF->addLiveIn(SrcReg, RC);
+    // FIXME: Unfortunately it's necessary to emit a copy from the livein copy.
+    // Without this, EmitLiveInCopies may eliminate the livein if its only
+    // use is a bitcast (which isn't turned into an instruction).
+    unsigned ResultReg = createResultReg(RC);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
+            ResultReg).addReg(DstReg, getKillRegState(true));
+    UpdateValueMap(I, ResultReg);
+  }
+
+  return true;
+}
+
 namespace llvm {
   FastISel *ARM::createFastISel(FunctionLoweringInfo &funcInfo,
                                 const TargetLibraryInfo *libInfo) {
diff --git a/lib/Target/ARM/ARMFrameLowering.cpp b/lib/Target/ARM/ARMFrameLowering.cpp
index 9392497fd07d..7a02adf24633 100644
--- a/lib/Target/ARM/ARMFrameLowering.cpp
+++ b/lib/Target/ARM/ARMFrameLowering.cpp
@@ -15,17 +15,16 @@
 #include "ARMBaseInstrInfo.h"
 #include "ARMBaseRegisterInfo.h"
 #include "ARMMachineFunctionInfo.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Function.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
@@ -120,13 +119,14 @@ static void
 emitSPUpdate(bool isARM,
              MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
              DebugLoc dl, const ARMBaseInstrInfo &TII,
-             int NumBytes, unsigned MIFlags = MachineInstr::NoFlags) {
+             int NumBytes, unsigned MIFlags = MachineInstr::NoFlags,
+             ARMCC::CondCodes Pred = ARMCC::AL, unsigned PredReg = 0) {
   if (isARM)
     emitARMRegPlusImmediate(MBB, MBBI, dl, ARM::SP, ARM::SP, NumBytes,
-                            ARMCC::AL, 0, TII, MIFlags);
+                            Pred, PredReg, TII, MIFlags);
   else
     emitT2RegPlusImmediate(MBB, MBBI, dl, ARM::SP, ARM::SP, NumBytes,
-                           ARMCC::AL, 0, TII, MIFlags);
+                           Pred, PredReg, TII, MIFlags);
 }
 
 void ARMFrameLowering::emitPrologue(MachineFunction &MF) const {
@@ -696,7 +696,7 @@ void ARMFrameLowering::emitPopInst(MachineBasicBlock &MBB,
       for (unsigned i = 0, e = Regs.size(); i < e; ++i)
         MIB.addReg(Regs[i], getDefRegState(true));
       if (DeleteRet) {
-        MIB->copyImplicitOps(&*MI);
+        MIB.copyImplicitOps(&*MI);
         MI->eraseFromParent();
       }
       MI = MIB;
@@ -1038,58 +1038,6 @@ static unsigned GetFunctionSizeInBytes(const MachineFunction &MF,
   return FnSize;
 }
 
-/// estimateStackSize - Estimate and return the size of the frame.
-/// FIXME: Make generic?
-static unsigned estimateStackSize(MachineFunction &MF) {
-  const MachineFrameInfo *MFI = MF.getFrameInfo();
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-  const TargetRegisterInfo *RegInfo = MF.getTarget().getRegisterInfo();
-  unsigned MaxAlign = MFI->getMaxAlignment();
-  int Offset = 0;
-
-  // This code is very, very similar to PEI::calculateFrameObjectOffsets().
-  // It really should be refactored to share code. Until then, changes
-  // should keep in mind that there's tight coupling between the two.
-
-  for (int i = MFI->getObjectIndexBegin(); i != 0; ++i) {
-    int FixedOff = -MFI->getObjectOffset(i);
-    if (FixedOff > Offset) Offset = FixedOff;
-  }
-  for (unsigned i = 0, e = MFI->getObjectIndexEnd(); i != e; ++i) {
-    if (MFI->isDeadObjectIndex(i))
-      continue;
-    Offset += MFI->getObjectSize(i);
-    unsigned Align = MFI->getObjectAlignment(i);
-    // Adjust to alignment boundary
-    Offset = (Offset+Align-1)/Align*Align;
-
-    MaxAlign = std::max(Align, MaxAlign);
-  }
-
-  if (MFI->adjustsStack() && TFI->hasReservedCallFrame(MF))
-    Offset += MFI->getMaxCallFrameSize();
-
-  // Round up the size to a multiple of the alignment.  If the function has
-  // any calls or alloca's, align to the target's StackAlignment value to
-  // ensure that the callee's frame or the alloca data is suitably aligned;
-  // otherwise, for leaf functions, align to the TransientStackAlignment
-  // value.
-  unsigned StackAlign;
-  if (MFI->adjustsStack() || MFI->hasVarSizedObjects() ||
-      (RegInfo->needsStackRealignment(MF) && MFI->getObjectIndexEnd() != 0))
-    StackAlign = TFI->getStackAlignment();
-  else
-    StackAlign = TFI->getTransientStackAlignment();
-
-  // If the frame pointer is eliminated, all frame offsets will be relative to
-  // SP not FP. Align to MaxAlign so this works.
-  StackAlign = std::max(StackAlign, MaxAlign);
-  unsigned AlignMask = StackAlign - 1;
-  Offset = (Offset + AlignMask) & ~uint64_t(AlignMask);
-
-  return (unsigned)Offset;
-}
-
 /// estimateRSStackSizeLimit - Look at each instruction that references stack
 /// frames and return the stack size limit beyond which some of these
 /// instructions will require a scratch register during their expansion later.
@@ -1153,7 +1101,8 @@ static void checkNumAlignedDPRCS2Regs(MachineFunction &MF) {
     return;
 
   // Naked functions don't spill callee-saved registers.
-  if (MF.getFunction()->getFnAttributes().hasAttribute(Attributes::Naked))
+  if (MF.getFunction()->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
+                                                     Attribute::Naked))
     return;
 
   // We are planning to use NEON instructions vst1 / vld1.
@@ -1234,7 +1183,7 @@ ARMFrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
     // we've used all the registers and so R4 is already used, so not marking
     // it here will be OK.
     // FIXME: It will be better just to find spare register here.
-    unsigned StackSize = estimateStackSize(MF);
+    unsigned StackSize = MFI->estimateStackSize(MF);
     if (MFI->hasVarSizedObjects() || StackSize > 508)
       MRI.setPhysRegUsed(ARM::R4);
   }
@@ -1329,7 +1278,8 @@ ARMFrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
   //        worth the effort and added fragility?
   bool BigStack =
     (RS &&
-     (estimateStackSize(MF) + ((hasFP(MF) && AFI->hasStackFrame()) ? 4:0) >=
+     (MFI->estimateStackSize(MF) +
+      ((hasFP(MF) && AFI->hasStackFrame()) ? 4:0) >=
       estimateRSStackSizeLimit(MF, this)))
     || MFI->hasVarSizedObjects()
     || (MFI->adjustsStack() && !canSimplifyCallFramePseudos(MF));
@@ -1418,7 +1368,7 @@ ARMFrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
         // note: Thumb1 functions spill to R12, not the stack.  Reserve a slot
         // closest to SP or frame pointer.
         const TargetRegisterClass *RC = &ARM::GPRRegClass;
-        RS->setScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
+        RS->addScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
                                                            RC->getAlignment(),
                                                            false));
       }
@@ -1430,3 +1380,51 @@ ARMFrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
     AFI->setLRIsSpilledForFarJump(true);
   }
 }
+
+
+void ARMFrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  const ARMBaseInstrInfo &TII =
+    *static_cast<const ARMBaseInstrInfo*>(MF.getTarget().getInstrInfo());
+  if (!hasReservedCallFrame(MF)) {
+    // If we have alloca, convert as follows:
+    // ADJCALLSTACKDOWN -> sub, sp, sp, amount
+    // ADJCALLSTACKUP   -> add, sp, sp, amount
+    MachineInstr *Old = I;
+    DebugLoc dl = Old->getDebugLoc();
+    unsigned Amount = Old->getOperand(0).getImm();
+    if (Amount != 0) {
+      // We need to keep the stack aligned properly.  To do this, we round the
+      // amount of space needed for the outgoing arguments up to the next
+      // alignment boundary.
+      unsigned Align = getStackAlignment();
+      Amount = (Amount+Align-1)/Align*Align;
+
+      ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+      assert(!AFI->isThumb1OnlyFunction() &&
+             "This eliminateCallFramePseudoInstr does not support Thumb1!");
+      bool isARM = !AFI->isThumbFunction();
+
+      // Replace the pseudo instruction with a new instruction...
+      unsigned Opc = Old->getOpcode();
+      int PIdx = Old->findFirstPredOperandIdx();
+      ARMCC::CondCodes Pred = (PIdx == -1)
+        ? ARMCC::AL : (ARMCC::CondCodes)Old->getOperand(PIdx).getImm();
+      if (Opc == ARM::ADJCALLSTACKDOWN || Opc == ARM::tADJCALLSTACKDOWN) {
+        // Note: PredReg is operand 2 for ADJCALLSTACKDOWN.
+        unsigned PredReg = Old->getOperand(2).getReg();
+        emitSPUpdate(isARM, MBB, I, dl, TII, -Amount, MachineInstr::NoFlags,
+                     Pred, PredReg);
+      } else {
+        // Note: PredReg is operand 3 for ADJCALLSTACKUP.
+        unsigned PredReg = Old->getOperand(3).getReg();
+        assert(Opc == ARM::ADJCALLSTACKUP || Opc == ARM::tADJCALLSTACKUP);
+        emitSPUpdate(isARM, MBB, I, dl, TII, Amount, MachineInstr::NoFlags,
+                     Pred, PredReg);
+      }
+    }
+  }
+  MBB.erase(I);
+}
+
diff --git a/lib/Target/ARM/ARMFrameLowering.h b/lib/Target/ARM/ARMFrameLowering.h
index a1c2b93562c9..efa255a5574a 100644
--- a/lib/Target/ARM/ARMFrameLowering.h
+++ b/lib/Target/ARM/ARMFrameLowering.h
@@ -70,6 +70,11 @@ public:
                    unsigned LdrOpc, bool isVarArg, bool NoGap,
                    bool(*Func)(unsigned, bool),
                    unsigned NumAlignedDPRCS2Regs) const;
+
+  virtual void eliminateCallFramePseudoInstr(
+                                    MachineFunction &MF,
+                                    MachineBasicBlock &MBB,
+                                    MachineBasicBlock::iterator MI) const;
 };
 
 } // End llvm namespace
diff --git a/lib/Target/ARM/ARMISelDAGToDAG.cpp b/lib/Target/ARM/ARMISelDAGToDAG.cpp
index efd6d2b8399e..2c51de23f7dc 100644
--- a/lib/Target/ARM/ARMISelDAGToDAG.cpp
+++ b/lib/Target/ARM/ARMISelDAGToDAG.cpp
@@ -16,24 +16,25 @@
 #include "ARMBaseInstrInfo.h"
 #include "ARMTargetMachine.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/LLVMContext.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
-#include "llvm/Target/TargetLowering.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/LLVMContext.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
@@ -78,6 +79,8 @@ public:
     return "ARM Instruction Selection";
   }
 
+  virtual void PreprocessISelDAG();
+
   /// getI32Imm - Return a target constant of type i32 with the specified
   /// value.
   inline SDValue getI32Imm(unsigned Imm) {
@@ -255,6 +258,8 @@ private:
   // Select special operations if node forms integer ABS pattern
   SDNode *SelectABSOp(SDNode *N);
 
+  SDNode *SelectInlineAsm(SDNode *N);
+
   SDNode *SelectConcatVector(SDNode *N);
 
   SDNode *SelectAtomic64(SDNode *Node, unsigned Opc);
@@ -265,15 +270,16 @@ private:
                                             char ConstraintCode,
                                             std::vector<SDValue> &OutOps);
 
-  // Form pairs of consecutive S, D, or Q registers.
-  SDNode *PairSRegs(EVT VT, SDValue V0, SDValue V1);
-  SDNode *PairDRegs(EVT VT, SDValue V0, SDValue V1);
-  SDNode *PairQRegs(EVT VT, SDValue V0, SDValue V1);
+  // Form pairs of consecutive R, S, D, or Q registers.
+  SDNode *createGPRPairNode(EVT VT, SDValue V0, SDValue V1);
+  SDNode *createSRegPairNode(EVT VT, SDValue V0, SDValue V1);
+  SDNode *createDRegPairNode(EVT VT, SDValue V0, SDValue V1);
+  SDNode *createQRegPairNode(EVT VT, SDValue V0, SDValue V1);
 
   // Form sequences of 4 consecutive S, D, or Q registers.
-  SDNode *QuadSRegs(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3);
-  SDNode *QuadDRegs(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3);
-  SDNode *QuadQRegs(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3);
+  SDNode *createQuadSRegsNode(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3);
+  SDNode *createQuadDRegsNode(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3);
+  SDNode *createQuadQRegsNode(EVT VT, SDValue V0, SDValue V1, SDValue V2, SDValue V3);
 
   // Get the alignment operand for a NEON VLD or VST instruction.
   SDValue GetVLDSTAlign(SDValue Align, unsigned NumVecs, bool is64BitVector);
@@ -326,6 +332,87 @@ static bool isScaledConstantInRange(SDValue Node, int Scale,
   return ScaledConstant >= RangeMin && ScaledConstant < RangeMax;
 }
 
+void ARMDAGToDAGISel::PreprocessISelDAG() {
+  if (!Subtarget->hasV6T2Ops())
+    return;
+
+  bool isThumb2 = Subtarget->isThumb();
+  for (SelectionDAG::allnodes_iterator I = CurDAG->allnodes_begin(),
+       E = CurDAG->allnodes_end(); I != E; ) {
+    SDNode *N = I++;  // Preincrement iterator to avoid invalidation issues.
+
+    if (N->getOpcode() != ISD::ADD)
+      continue;
+
+    // Look for (add X1, (and (srl X2, c1), c2)) where c2 is constant with
+    // leading zeros, followed by consecutive set bits, followed by 1 or 2
+    // trailing zeros, e.g. 1020.
+    // Transform the expression to
+    // (add X1, (shl (and (srl X2, c1), (c2>>tz)), tz)) where tz is the number
+    // of trailing zeros of c2. The left shift would be folded as an shifter
+    // operand of 'add' and the 'and' and 'srl' would become a bits extraction
+    // node (UBFX).
+
+    SDValue N0 = N->getOperand(0);
+    SDValue N1 = N->getOperand(1);
+    unsigned And_imm = 0;
+    if (!isOpcWithIntImmediate(N1.getNode(), ISD::AND, And_imm)) {
+      if (isOpcWithIntImmediate(N0.getNode(), ISD::AND, And_imm))
+        std::swap(N0, N1);
+    }
+    if (!And_imm)
+      continue;
+
+    // Check if the AND mask is an immediate of the form: 000.....1111111100
+    unsigned TZ = CountTrailingZeros_32(And_imm);
+    if (TZ != 1 && TZ != 2)
+      // Be conservative here. Shifter operands aren't always free. e.g. On
+      // Swift, left shifter operand of 1 / 2 for free but others are not.
+      // e.g.
+      //  ubfx   r3, r1, #16, #8
+      //  ldr.w  r3, [r0, r3, lsl #2]
+      // vs.
+      //  mov.w  r9, #1020
+      //  and.w  r2, r9, r1, lsr #14
+      //  ldr    r2, [r0, r2]
+      continue;
+    And_imm >>= TZ;
+    if (And_imm & (And_imm + 1))
+      continue;
+
+    // Look for (and (srl X, c1), c2).
+    SDValue Srl = N1.getOperand(0);
+    unsigned Srl_imm = 0;
+    if (!isOpcWithIntImmediate(Srl.getNode(), ISD::SRL, Srl_imm) ||
+        (Srl_imm <= 2))
+      continue;
+
+    // Make sure first operand is not a shifter operand which would prevent
+    // folding of the left shift.
+    SDValue CPTmp0;
+    SDValue CPTmp1;
+    SDValue CPTmp2;
+    if (isThumb2) {
+      if (SelectT2ShifterOperandReg(N0, CPTmp0, CPTmp1))
+        continue;
+    } else {
+      if (SelectImmShifterOperand(N0, CPTmp0, CPTmp1) ||
+          SelectRegShifterOperand(N0, CPTmp0, CPTmp1, CPTmp2))
+        continue;
+    }
+
+    // Now make the transformation.
+    Srl = CurDAG->getNode(ISD::SRL, Srl.getDebugLoc(), MVT::i32,
+                          Srl.getOperand(0),
+                          CurDAG->getConstant(Srl_imm+TZ, MVT::i32));
+    N1 = CurDAG->getNode(ISD::AND, N1.getDebugLoc(), MVT::i32,
+                         Srl, CurDAG->getConstant(And_imm, MVT::i32));
+    N1 = CurDAG->getNode(ISD::SHL, N1.getDebugLoc(), MVT::i32,
+                         N1, CurDAG->getConstant(TZ, MVT::i32));
+    CurDAG->UpdateNodeOperands(N, N0, N1);
+  }  
+}
+
 /// hasNoVMLxHazardUse - Return true if it's desirable to select a FP MLA / MLS
 /// node. VFP / NEON fp VMLA / VMLS instructions have special RAW hazards (at
 /// least on current ARM implementations) which should be avoidded.
@@ -1444,9 +1531,19 @@ SDNode *ARMDAGToDAGISel::SelectT2IndexedLoad(SDNode *N) {
   return NULL;
 }
 
-/// PairSRegs - Form a D register from a pair of S registers.
-///
-SDNode *ARMDAGToDAGISel::PairSRegs(EVT VT, SDValue V0, SDValue V1) {
+/// \brief Form a GPRPair pseudo register from a pair of GPR regs.
+SDNode *ARMDAGToDAGISel::createGPRPairNode(EVT VT, SDValue V0, SDValue V1) {
+  DebugLoc dl = V0.getNode()->getDebugLoc();
+  SDValue RegClass =
+    CurDAG->getTargetConstant(ARM::GPRPairRegClassID, MVT::i32);
+  SDValue SubReg0 = CurDAG->getTargetConstant(ARM::gsub_0, MVT::i32);
+  SDValue SubReg1 = CurDAG->getTargetConstant(ARM::gsub_1, MVT::i32);
+  const SDValue Ops[] = { RegClass, V0, SubReg0, V1, SubReg1 };
+  return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
+}
+
+/// \brief Form a D register from a pair of S registers.
+SDNode *ARMDAGToDAGISel::createSRegPairNode(EVT VT, SDValue V0, SDValue V1) {
   DebugLoc dl = V0.getNode()->getDebugLoc();
   SDValue RegClass =
     CurDAG->getTargetConstant(ARM::DPR_VFP2RegClassID, MVT::i32);
@@ -1456,9 +1553,8 @@ SDNode *ARMDAGToDAGISel::PairSRegs(EVT VT, SDValue V0, SDValue V1) {
   return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
 }
 
-/// PairDRegs - Form a quad register from a pair of D registers.
-///
-SDNode *ARMDAGToDAGISel::PairDRegs(EVT VT, SDValue V0, SDValue V1) {
+/// \brief Form a quad register from a pair of D registers.
+SDNode *ARMDAGToDAGISel::createDRegPairNode(EVT VT, SDValue V0, SDValue V1) {
   DebugLoc dl = V0.getNode()->getDebugLoc();
   SDValue RegClass = CurDAG->getTargetConstant(ARM::QPRRegClassID, MVT::i32);
   SDValue SubReg0 = CurDAG->getTargetConstant(ARM::dsub_0, MVT::i32);
@@ -1467,9 +1563,8 @@ SDNode *ARMDAGToDAGISel::PairDRegs(EVT VT, SDValue V0, SDValue V1) {
   return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
 }
 
-/// PairQRegs - Form 4 consecutive D registers from a pair of Q registers.
-///
-SDNode *ARMDAGToDAGISel::PairQRegs(EVT VT, SDValue V0, SDValue V1) {
+/// \brief Form 4 consecutive D registers from a pair of Q registers.
+SDNode *ARMDAGToDAGISel::createQRegPairNode(EVT VT, SDValue V0, SDValue V1) {
   DebugLoc dl = V0.getNode()->getDebugLoc();
   SDValue RegClass = CurDAG->getTargetConstant(ARM::QQPRRegClassID, MVT::i32);
   SDValue SubReg0 = CurDAG->getTargetConstant(ARM::qsub_0, MVT::i32);
@@ -1478,9 +1573,8 @@ SDNode *ARMDAGToDAGISel::PairQRegs(EVT VT, SDValue V0, SDValue V1) {
   return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 5);
 }
 
-/// QuadSRegs - Form 4 consecutive S registers.
-///
-SDNode *ARMDAGToDAGISel::QuadSRegs(EVT VT, SDValue V0, SDValue V1,
+/// \brief Form 4 consecutive S registers.
+SDNode *ARMDAGToDAGISel::createQuadSRegsNode(EVT VT, SDValue V0, SDValue V1,
                                    SDValue V2, SDValue V3) {
   DebugLoc dl = V0.getNode()->getDebugLoc();
   SDValue RegClass =
@@ -1494,9 +1588,8 @@ SDNode *ARMDAGToDAGISel::QuadSRegs(EVT VT, SDValue V0, SDValue V1,
   return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9);
 }
 
-/// QuadDRegs - Form 4 consecutive D registers.
-///
-SDNode *ARMDAGToDAGISel::QuadDRegs(EVT VT, SDValue V0, SDValue V1,
+/// \brief Form 4 consecutive D registers.
+SDNode *ARMDAGToDAGISel::createQuadDRegsNode(EVT VT, SDValue V0, SDValue V1,
                                    SDValue V2, SDValue V3) {
   DebugLoc dl = V0.getNode()->getDebugLoc();
   SDValue RegClass = CurDAG->getTargetConstant(ARM::QQPRRegClassID, MVT::i32);
@@ -1509,9 +1602,8 @@ SDNode *ARMDAGToDAGISel::QuadDRegs(EVT VT, SDValue V0, SDValue V1,
   return CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, dl, VT, Ops, 9);
 }
 
-/// QuadQRegs - Form 4 consecutive Q registers.
-///
-SDNode *ARMDAGToDAGISel::QuadQRegs(EVT VT, SDValue V0, SDValue V1,
+/// \brief Form 4 consecutive Q registers.
+SDNode *ARMDAGToDAGISel::createQuadQRegsNode(EVT VT, SDValue V0, SDValue V1,
                                    SDValue V2, SDValue V3) {
   DebugLoc dl = V0.getNode()->getDebugLoc();
   SDValue RegClass = CurDAG->getTargetConstant(ARM::QQQQPRRegClassID, MVT::i32);
@@ -1784,7 +1876,7 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
       SDValue V0 = N->getOperand(Vec0Idx + 0);
       SDValue V1 = N->getOperand(Vec0Idx + 1);
       if (NumVecs == 2)
-        SrcReg = SDValue(PairDRegs(MVT::v2i64, V0, V1), 0);
+        SrcReg = SDValue(createDRegPairNode(MVT::v2i64, V0, V1), 0);
       else {
         SDValue V2 = N->getOperand(Vec0Idx + 2);
         // If it's a vst3, form a quad D-register and leave the last part as
@@ -1792,13 +1884,13 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
         SDValue V3 = (NumVecs == 3)
           ? SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF,dl,VT), 0)
           : N->getOperand(Vec0Idx + 3);
-        SrcReg = SDValue(QuadDRegs(MVT::v4i64, V0, V1, V2, V3), 0);
+        SrcReg = SDValue(createQuadDRegsNode(MVT::v4i64, V0, V1, V2, V3), 0);
       }
     } else {
       // Form a QQ register.
       SDValue Q0 = N->getOperand(Vec0Idx);
       SDValue Q1 = N->getOperand(Vec0Idx + 1);
-      SrcReg = SDValue(PairQRegs(MVT::v4i64, Q0, Q1), 0);
+      SrcReg = SDValue(createQRegPairNode(MVT::v4i64, Q0, Q1), 0);
     }
 
     unsigned Opc = (is64BitVector ? DOpcodes[OpcodeIndex] :
@@ -1840,7 +1932,7 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
   SDValue V3 = (NumVecs == 3)
     ? SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, dl, VT), 0)
     : N->getOperand(Vec0Idx + 3);
-  SDValue RegSeq = SDValue(QuadQRegs(MVT::v8i64, V0, V1, V2, V3), 0);
+  SDValue RegSeq = SDValue(createQuadQRegsNode(MVT::v8i64, V0, V1, V2, V3), 0);
 
   // Store the even D registers.  This is always an updating store, so that it
   // provides the address to the second store for the odd subregs.
@@ -1950,18 +2042,18 @@ SDNode *ARMDAGToDAGISel::SelectVLDSTLane(SDNode *N, bool IsLoad,
   SDValue V1 = N->getOperand(Vec0Idx + 1);
   if (NumVecs == 2) {
     if (is64BitVector)
-      SuperReg = SDValue(PairDRegs(MVT::v2i64, V0, V1), 0);
+      SuperReg = SDValue(createDRegPairNode(MVT::v2i64, V0, V1), 0);
     else
-      SuperReg = SDValue(PairQRegs(MVT::v4i64, V0, V1), 0);
+      SuperReg = SDValue(createQRegPairNode(MVT::v4i64, V0, V1), 0);
   } else {
     SDValue V2 = N->getOperand(Vec0Idx + 2);
     SDValue V3 = (NumVecs == 3)
       ? SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, dl, VT), 0)
       : N->getOperand(Vec0Idx + 3);
     if (is64BitVector)
-      SuperReg = SDValue(QuadDRegs(MVT::v4i64, V0, V1, V2, V3), 0);
+      SuperReg = SDValue(createQuadDRegsNode(MVT::v4i64, V0, V1, V2, V3), 0);
     else
-      SuperReg = SDValue(QuadQRegs(MVT::v8i64, V0, V1, V2, V3), 0);
+      SuperReg = SDValue(createQuadQRegsNode(MVT::v8i64, V0, V1, V2, V3), 0);
   }
   Ops.push_back(SuperReg);
   Ops.push_back(getI32Imm(Lane));
@@ -2087,7 +2179,7 @@ SDNode *ARMDAGToDAGISel::SelectVTBL(SDNode *N, bool IsExt, unsigned NumVecs,
   SDValue V0 = N->getOperand(FirstTblReg + 0);
   SDValue V1 = N->getOperand(FirstTblReg + 1);
   if (NumVecs == 2)
-    RegSeq = SDValue(PairDRegs(MVT::v16i8, V0, V1), 0);
+    RegSeq = SDValue(createDRegPairNode(MVT::v16i8, V0, V1), 0);
   else {
     SDValue V2 = N->getOperand(FirstTblReg + 2);
     // If it's a vtbl3, form a quad D-register and leave the last part as
@@ -2095,7 +2187,7 @@ SDNode *ARMDAGToDAGISel::SelectVTBL(SDNode *N, bool IsExt, unsigned NumVecs,
     SDValue V3 = (NumVecs == 3)
       ? SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, dl, VT), 0)
       : N->getOperand(FirstTblReg + 3);
-    RegSeq = SDValue(QuadDRegs(MVT::v4i64, V0, V1, V2, V3), 0);
+    RegSeq = SDValue(createQuadDRegsNode(MVT::v4i64, V0, V1, V2, V3), 0);
   }
 
   SmallVector<SDValue, 6> Ops;
@@ -2113,10 +2205,10 @@ SDNode *ARMDAGToDAGISel::SelectV6T2BitfieldExtractOp(SDNode *N,
   if (!Subtarget->hasV6T2Ops())
     return NULL;
 
-  unsigned Opc = isSigned ? (Subtarget->isThumb() ? ARM::t2SBFX : ARM::SBFX)
+  unsigned Opc = isSigned
+    ? (Subtarget->isThumb() ? ARM::t2SBFX : ARM::SBFX)
     : (Subtarget->isThumb() ? ARM::t2UBFX : ARM::UBFX);
 
-
   // For unsigned extracts, check for a shift right and mask
   unsigned And_imm = 0;
   if (N->getOpcode() == ISD::AND) {
@@ -2134,7 +2226,29 @@ SDNode *ARMDAGToDAGISel::SelectV6T2BitfieldExtractOp(SDNode *N,
         // Note: The width operand is encoded as width-1.
         unsigned Width = CountTrailingOnes_32(And_imm) - 1;
         unsigned LSB = Srl_imm;
+
         SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
+
+        if ((LSB + Width + 1) == N->getValueType(0).getSizeInBits()) {
+          // It's cheaper to use a right shift to extract the top bits.
+          if (Subtarget->isThumb()) {
+            Opc = isSigned ? ARM::t2ASRri : ARM::t2LSRri;
+            SDValue Ops[] = { N->getOperand(0).getOperand(0),
+                              CurDAG->getTargetConstant(LSB, MVT::i32),
+                              getAL(CurDAG), Reg0, Reg0 };
+            return CurDAG->SelectNodeTo(N, Opc, MVT::i32, Ops, 5);
+          }
+
+          // ARM models shift instructions as MOVsi with shifter operand.
+          ARM_AM::ShiftOpc ShOpcVal = ARM_AM::getShiftOpcForNode(ISD::SRL);
+          SDValue ShOpc =
+            CurDAG->getTargetConstant(ARM_AM::getSORegOpc(ShOpcVal, LSB),
+                                      MVT::i32);
+          SDValue Ops[] = { N->getOperand(0).getOperand(0), ShOpc,
+                            getAL(CurDAG), Reg0, Reg0 };
+          return CurDAG->SelectNodeTo(N, ARM::MOVsi, MVT::i32, Ops, 5);
+        }
+
         SDValue Ops[] = { N->getOperand(0).getOperand(0),
                           CurDAG->getTargetConstant(LSB, MVT::i32),
                           CurDAG->getTargetConstant(Width, MVT::i32),
@@ -2411,7 +2525,7 @@ SDNode *ARMDAGToDAGISel::SelectConcatVector(SDNode *N) {
   EVT VT = N->getValueType(0);
   if (!VT.is128BitVector() || N->getNumOperands() != 2)
     llvm_unreachable("unexpected CONCAT_VECTORS");
-  return PairDRegs(VT, N->getOperand(0), N->getOperand(1));
+  return createDRegPairNode(VT, N->getOperand(0), N->getOperand(1));
 }
 
 SDNode *ARMDAGToDAGISel::SelectAtomic64(SDNode *Node, unsigned Opc) {
@@ -2441,6 +2555,12 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
 
   switch (N->getOpcode()) {
   default: break;
+  case ISD::INLINEASM: {
+    SDNode *ResNode = SelectInlineAsm(N);
+    if (ResNode)
+      return ResNode;
+    break;
+  }
   case ISD::XOR: {
     // Select special operations if XOR node forms integer ABS pattern
     SDNode *ResNode = SelectABSOp(N);
@@ -2790,13 +2910,13 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     unsigned NumElts = VecVT.getVectorNumElements();
     if (EltVT == MVT::f64) {
       assert(NumElts == 2 && "unexpected type for BUILD_VECTOR");
-      return PairDRegs(VecVT, N->getOperand(0), N->getOperand(1));
+      return createDRegPairNode(VecVT, N->getOperand(0), N->getOperand(1));
     }
     assert(EltVT == MVT::f32 && "unexpected type for BUILD_VECTOR");
     if (NumElts == 2)
-      return PairSRegs(VecVT, N->getOperand(0), N->getOperand(1));
+      return createSRegPairNode(VecVT, N->getOperand(0), N->getOperand(1));
     assert(NumElts == 4 && "unexpected type for BUILD_VECTOR");
-    return QuadSRegs(VecVT, N->getOperand(0), N->getOperand(1),
+    return createQuadSRegsNode(VecVT, N->getOperand(0), N->getOperand(1),
                      N->getOperand(2), N->getOperand(3));
   }
 
@@ -3009,17 +3129,19 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
       DebugLoc dl = N->getDebugLoc();
       SDValue Chain = N->getOperand(0);
 
-      unsigned NewOpc = ARM::LDREXD;
-      if (Subtarget->isThumb() && Subtarget->hasThumb2())
-        NewOpc = ARM::t2LDREXD;
+      bool isThumb = Subtarget->isThumb() && Subtarget->hasThumb2();
+      unsigned NewOpc = isThumb ? ARM::t2LDREXD :ARM::LDREXD;
 
       // arm_ldrexd returns a i64 value in {i32, i32}
       std::vector<EVT> ResTys;
-      ResTys.push_back(MVT::i32);
-      ResTys.push_back(MVT::i32);
+      if (isThumb) {
+        ResTys.push_back(MVT::i32);
+        ResTys.push_back(MVT::i32);
+      } else
+        ResTys.push_back(MVT::Untyped);
       ResTys.push_back(MVT::Other);
 
-      // place arguments in the right order
+      // Place arguments in the right order.
       SmallVector<SDValue, 7> Ops;
       Ops.push_back(MemAddr);
       Ops.push_back(getAL(CurDAG));
@@ -3032,30 +3154,33 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
       MemOp[0] = cast<MemIntrinsicSDNode>(N)->getMemOperand();
       cast<MachineSDNode>(Ld)->setMemRefs(MemOp, MemOp + 1);
 
-      // Until there's support for specifing explicit register constraints
-      // like the use of even/odd register pair, hardcode ldrexd to always
-      // use the pair [R0, R1] to hold the load result.
-      Chain = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, ARM::R0,
-                                   SDValue(Ld, 0), SDValue(0,0));
-      Chain = CurDAG->getCopyToReg(Chain, dl, ARM::R1,
-                                   SDValue(Ld, 1), Chain.getValue(1));
-
       // Remap uses.
-      SDValue Glue = Chain.getValue(1);
+      SDValue OutChain = isThumb ? SDValue(Ld, 2) : SDValue(Ld, 1);
       if (!SDValue(N, 0).use_empty()) {
-        SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
-                                                ARM::R0, MVT::i32, Glue);
-        Glue = Result.getValue(2);
+        SDValue Result;
+        if (isThumb)
+          Result = SDValue(Ld, 0);
+        else {
+          SDValue SubRegIdx = CurDAG->getTargetConstant(ARM::gsub_0, MVT::i32);
+          SDNode *ResNode = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG,
+              dl, MVT::i32, SDValue(Ld, 0), SubRegIdx);
+          Result = SDValue(ResNode,0);
+        }
         ReplaceUses(SDValue(N, 0), Result);
       }
       if (!SDValue(N, 1).use_empty()) {
-        SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
-                                                ARM::R1, MVT::i32, Glue);
-        Glue = Result.getValue(2);
+        SDValue Result;
+        if (isThumb)
+          Result = SDValue(Ld, 1);
+        else {
+          SDValue SubRegIdx = CurDAG->getTargetConstant(ARM::gsub_1, MVT::i32);
+          SDNode *ResNode = CurDAG->getMachineNode(TargetOpcode::EXTRACT_SUBREG,
+              dl, MVT::i32, SDValue(Ld, 0), SubRegIdx);
+          Result = SDValue(ResNode,0);
+        }
         ReplaceUses(SDValue(N, 1), Result);
       }
-
-      ReplaceUses(SDValue(N, 2), SDValue(Ld, 2));
+      ReplaceUses(SDValue(N, 2), OutChain);
       return NULL;
     }
 
@@ -3066,38 +3191,25 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
       SDValue Val1 = N->getOperand(3);
       SDValue MemAddr = N->getOperand(4);
 
-      // Until there's support for specifing explicit register constraints
-      // like the use of even/odd register pair, hardcode strexd to always
-      // use the pair [R2, R3] to hold the i64 (i32, i32) value to be stored.
-      Chain = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, ARM::R2, Val0,
-                                   SDValue(0, 0));
-      Chain = CurDAG->getCopyToReg(Chain, dl, ARM::R3, Val1, Chain.getValue(1));
-
-      SDValue Glue = Chain.getValue(1);
-      Val0 = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
-                                    ARM::R2, MVT::i32, Glue);
-      Glue = Val0.getValue(1);
-      Val1 = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
-                                    ARM::R3, MVT::i32, Glue);
-
       // Store exclusive double return a i32 value which is the return status
       // of the issued store.
-      std::vector<EVT> ResTys;
-      ResTys.push_back(MVT::i32);
-      ResTys.push_back(MVT::Other);
+      EVT ResTys[] = { MVT::i32, MVT::Other };
 
-      // place arguments in the right order
+      bool isThumb = Subtarget->isThumb() && Subtarget->hasThumb2();
+      // Place arguments in the right order.
       SmallVector<SDValue, 7> Ops;
-      Ops.push_back(Val0);
-      Ops.push_back(Val1);
+      if (isThumb) {
+        Ops.push_back(Val0);
+        Ops.push_back(Val1);
+      } else
+        // arm_strexd uses GPRPair.
+        Ops.push_back(SDValue(createGPRPairNode(MVT::Untyped, Val0, Val1), 0));
       Ops.push_back(MemAddr);
       Ops.push_back(getAL(CurDAG));
       Ops.push_back(CurDAG->getRegister(0, MVT::i32));
       Ops.push_back(Chain);
 
-      unsigned NewOpc = ARM::STREXD;
-      if (Subtarget->isThumb() && Subtarget->hasThumb2())
-        NewOpc = ARM::t2STREXD;
+      unsigned NewOpc = isThumb ? ARM::t2STREXD : ARM::STREXD;
 
       SDNode *St = CurDAG->getMachineNode(NewOpc, dl, ResTys, Ops.data(),
                                           Ops.size());
@@ -3295,7 +3407,7 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     // Form a REG_SEQUENCE to force register allocation.
     SDValue V0 = N->getOperand(0);
     SDValue V1 = N->getOperand(1);
-    SDValue RegSeq = SDValue(PairDRegs(MVT::v16i8, V0, V1), 0);
+    SDValue RegSeq = SDValue(createDRegPairNode(MVT::v16i8, V0, V1), 0);
 
     SmallVector<SDValue, 6> Ops;
     Ops.push_back(RegSeq);
@@ -3325,11 +3437,152 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
     return SelectAtomic64(N, ARM::ATOMSWAP6432);
   case ARMISD::ATOMCMPXCHG64_DAG:
     return SelectAtomic64(N, ARM::ATOMCMPXCHG6432);
+
+  case ARMISD::ATOMMIN64_DAG:
+    return SelectAtomic64(N, ARM::ATOMMIN6432);
+  case ARMISD::ATOMUMIN64_DAG:
+    return SelectAtomic64(N, ARM::ATOMUMIN6432);
+  case ARMISD::ATOMMAX64_DAG:
+    return SelectAtomic64(N, ARM::ATOMMAX6432);
+  case ARMISD::ATOMUMAX64_DAG:
+    return SelectAtomic64(N, ARM::ATOMUMAX6432);
   }
 
   return SelectCode(N);
 }
 
+SDNode *ARMDAGToDAGISel::SelectInlineAsm(SDNode *N){
+  std::vector<SDValue> AsmNodeOperands;
+  unsigned Flag, Kind;
+  bool Changed = false;
+  unsigned NumOps = N->getNumOperands();
+
+  ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(
+      N->getOperand(InlineAsm::Op_AsmString));
+  StringRef AsmString = StringRef(S->getSymbol());
+
+  // Normally, i64 data is bounded to two arbitrary GRPs for "%r" constraint.
+  // However, some instrstions (e.g. ldrexd/strexd in ARM mode) require
+  // (even/even+1) GPRs and use %n and %Hn to refer to the individual regs
+  // respectively. Since there is no constraint to explicitly specify a
+  // reg pair, we search %H operand inside the asm string. If it is found, the
+  // transformation below enforces a GPRPair reg class for "%r" for 64-bit data.
+  if (AsmString.find(":H}") == StringRef::npos)
+    return NULL;
+
+  DebugLoc dl = N->getDebugLoc();
+  SDValue Glue = N->getOperand(NumOps-1);
+
+  // Glue node will be appended late.
+  for(unsigned i = 0; i < NumOps -1; ++i) {
+    SDValue op = N->getOperand(i);
+    AsmNodeOperands.push_back(op);
+
+    if (i < InlineAsm::Op_FirstOperand)
+      continue;
+
+    if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(i))) {
+      Flag = C->getZExtValue();
+      Kind = InlineAsm::getKind(Flag);
+    }
+    else
+      continue;
+
+    if (Kind != InlineAsm::Kind_RegUse && Kind != InlineAsm::Kind_RegDef
+        && Kind != InlineAsm::Kind_RegDefEarlyClobber)
+      continue;
+
+    unsigned RegNum = InlineAsm::getNumOperandRegisters(Flag);
+    unsigned RC;
+    bool HasRC = InlineAsm::hasRegClassConstraint(Flag, RC);
+    if (!HasRC || RC != ARM::GPRRegClassID || RegNum != 2)
+      continue;
+
+    assert((i+2 < NumOps-1) && "Invalid number of operands in inline asm");
+    SDValue V0 = N->getOperand(i+1);
+    SDValue V1 = N->getOperand(i+2);
+    unsigned Reg0 = cast<RegisterSDNode>(V0)->getReg();
+    unsigned Reg1 = cast<RegisterSDNode>(V1)->getReg();
+    SDValue PairedReg;
+    MachineRegisterInfo &MRI = MF->getRegInfo();
+
+    if (Kind == InlineAsm::Kind_RegDef ||
+        Kind == InlineAsm::Kind_RegDefEarlyClobber) {
+      // Replace the two GPRs with 1 GPRPair and copy values from GPRPair to
+      // the original GPRs.
+
+      unsigned GPVR = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+      PairedReg = CurDAG->getRegister(GPVR, MVT::Untyped);
+      SDValue Chain = SDValue(N,0);
+
+      SDNode *GU = N->getGluedUser();
+      SDValue RegCopy = CurDAG->getCopyFromReg(Chain, dl, GPVR, MVT::Untyped,
+                                               Chain.getValue(1));
+
+      // Extract values from a GPRPair reg and copy to the original GPR reg.
+      SDValue Sub0 = CurDAG->getTargetExtractSubreg(ARM::gsub_0, dl, MVT::i32,
+                                                    RegCopy);
+      SDValue Sub1 = CurDAG->getTargetExtractSubreg(ARM::gsub_1, dl, MVT::i32,
+                                                    RegCopy);
+      SDValue T0 = CurDAG->getCopyToReg(Sub0, dl, Reg0, Sub0,
+                                        RegCopy.getValue(1));
+      SDValue T1 = CurDAG->getCopyToReg(Sub1, dl, Reg1, Sub1, T0.getValue(1));
+
+      // Update the original glue user.
+      std::vector<SDValue> Ops(GU->op_begin(), GU->op_end()-1);
+      Ops.push_back(T1.getValue(1));
+      CurDAG->UpdateNodeOperands(GU, &Ops[0], Ops.size());
+      GU = T1.getNode();
+    }
+    else {
+      // For Kind  == InlineAsm::Kind_RegUse, we first copy two GPRs into a
+      // GPRPair and then pass the GPRPair to the inline asm.
+      SDValue Chain = AsmNodeOperands[InlineAsm::Op_InputChain];
+
+      // As REG_SEQ doesn't take RegisterSDNode, we copy them first.
+      SDValue T0 = CurDAG->getCopyFromReg(Chain, dl, Reg0, MVT::i32,
+                                          Chain.getValue(1));
+      SDValue T1 = CurDAG->getCopyFromReg(Chain, dl, Reg1, MVT::i32,
+                                          T0.getValue(1));
+      SDValue Pair = SDValue(createGPRPairNode(MVT::Untyped, T0, T1), 0);
+
+      // Copy REG_SEQ into a GPRPair-typed VR and replace the original two
+      // i32 VRs of inline asm with it.
+      unsigned GPVR = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+      PairedReg = CurDAG->getRegister(GPVR, MVT::Untyped);
+      Chain = CurDAG->getCopyToReg(T1, dl, GPVR, Pair, T1.getValue(1));
+
+      AsmNodeOperands[InlineAsm::Op_InputChain] = Chain;
+      Glue = Chain.getValue(1);
+    }
+
+    Changed = true;
+
+    if(PairedReg.getNode()) {
+      Flag = InlineAsm::getFlagWord(Kind, 1 /* RegNum*/);
+      Flag = InlineAsm::getFlagWordForRegClass(Flag, ARM::GPRPairRegClassID);
+      // Replace the current flag.
+      AsmNodeOperands[AsmNodeOperands.size() -1] = CurDAG->getTargetConstant(
+          Flag, MVT::i32);
+      // Add the new register node and skip the original two GPRs.
+      AsmNodeOperands.push_back(PairedReg);
+      // Skip the next two GPRs.
+      i += 2;
+    }
+  }
+
+  AsmNodeOperands.push_back(Glue);
+  if (!Changed)
+    return NULL;
+
+  SDValue New = CurDAG->getNode(ISD::INLINEASM, N->getDebugLoc(),
+      CurDAG->getVTList(MVT::Other, MVT::Glue), &AsmNodeOperands[0],
+                        AsmNodeOperands.size());
+  New->setNodeId(-1);
+  return New.getNode();
+}
+
+
 bool ARMDAGToDAGISel::
 SelectInlineAsmMemoryOperand(const SDValue &Op, char ConstraintCode,
                              std::vector<SDValue> &OutOps) {
diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index ff99b04078e8..bb26090d2d8d 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -23,14 +23,8 @@
 #include "ARMTargetMachine.h"
 #include "ARMTargetObjectFile.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalValue.h"
-#include "llvm/Instruction.h"
-#include "llvm/Instructions.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/Type.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/IntrinsicLowering.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
@@ -40,14 +34,20 @@
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/Instruction.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Type.h"
 #include "llvm/MC/MCSectionMachO.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/ADT/StringExtras.h"
-#include "llvm/ADT/Statistic.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetOptions.h"
 using namespace llvm;
 
 STATISTIC(NumTailCalls, "Number of tail calls");
@@ -504,6 +504,7 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
     setOperationAction(ISD::FRINT, MVT::v2f64, Expand);
     setOperationAction(ISD::FNEARBYINT, MVT::v2f64, Expand);
     setOperationAction(ISD::FFLOOR, MVT::v2f64, Expand);
+    setOperationAction(ISD::FMA, MVT::v2f64, Expand);
 
     setOperationAction(ISD::FSQRT, MVT::v4f32, Expand);
     setOperationAction(ISD::FSIN, MVT::v4f32, Expand);
@@ -515,8 +516,29 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
     setOperationAction(ISD::FLOG10, MVT::v4f32, Expand);
     setOperationAction(ISD::FEXP, MVT::v4f32, Expand);
     setOperationAction(ISD::FEXP2, MVT::v4f32, Expand);
+    setOperationAction(ISD::FCEIL, MVT::v4f32, Expand);
+    setOperationAction(ISD::FTRUNC, MVT::v4f32, Expand);
+    setOperationAction(ISD::FRINT, MVT::v4f32, Expand);
+    setOperationAction(ISD::FNEARBYINT, MVT::v4f32, Expand);
     setOperationAction(ISD::FFLOOR, MVT::v4f32, Expand);
 
+    // Mark v2f32 intrinsics.
+    setOperationAction(ISD::FSQRT, MVT::v2f32, Expand);
+    setOperationAction(ISD::FSIN, MVT::v2f32, Expand);
+    setOperationAction(ISD::FCOS, MVT::v2f32, Expand);
+    setOperationAction(ISD::FPOWI, MVT::v2f32, Expand);
+    setOperationAction(ISD::FPOW, MVT::v2f32, Expand);
+    setOperationAction(ISD::FLOG, MVT::v2f32, Expand);
+    setOperationAction(ISD::FLOG2, MVT::v2f32, Expand);
+    setOperationAction(ISD::FLOG10, MVT::v2f32, Expand);
+    setOperationAction(ISD::FEXP, MVT::v2f32, Expand);
+    setOperationAction(ISD::FEXP2, MVT::v2f32, Expand);
+    setOperationAction(ISD::FCEIL, MVT::v2f32, Expand);
+    setOperationAction(ISD::FTRUNC, MVT::v2f32, Expand);
+    setOperationAction(ISD::FRINT, MVT::v2f32, Expand);
+    setOperationAction(ISD::FNEARBYINT, MVT::v2f32, Expand);
+    setOperationAction(ISD::FFLOOR, MVT::v2f32, Expand);
+
     // Neon does not support some operations on v1i64 and v2i64 types.
     setOperationAction(ISD::MUL, MVT::v1i64, Expand);
     // Custom handling for some quad-vector types to detect VMULL.
@@ -539,6 +561,33 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
     setOperationAction(ISD::FP_TO_UINT, MVT::v4i16, Custom);
     setOperationAction(ISD::FP_TO_SINT, MVT::v4i16, Custom);
 
+    setOperationAction(ISD::FP_ROUND,   MVT::v2f32, Expand);
+    setOperationAction(ISD::FP_EXTEND,  MVT::v2f64, Expand);
+
+    // Custom expand long extensions to vectors.
+    setOperationAction(ISD::SIGN_EXTEND, MVT::v8i32,  Custom);
+    setOperationAction(ISD::ZERO_EXTEND, MVT::v8i32,  Custom);
+    setOperationAction(ISD::SIGN_EXTEND, MVT::v4i64,  Custom);
+    setOperationAction(ISD::ZERO_EXTEND, MVT::v4i64,  Custom);
+    setOperationAction(ISD::SIGN_EXTEND, MVT::v16i32, Custom);
+    setOperationAction(ISD::ZERO_EXTEND, MVT::v16i32, Custom);
+    setOperationAction(ISD::SIGN_EXTEND, MVT::v8i64,  Custom);
+    setOperationAction(ISD::ZERO_EXTEND, MVT::v8i64,  Custom);
+
+    // NEON does not have single instruction CTPOP for vectors with element
+    // types wider than 8-bits.  However, custom lowering can leverage the
+    // v8i8/v16i8 vcnt instruction.
+    setOperationAction(ISD::CTPOP,      MVT::v2i32, Custom);
+    setOperationAction(ISD::CTPOP,      MVT::v4i32, Custom);
+    setOperationAction(ISD::CTPOP,      MVT::v4i16, Custom);
+    setOperationAction(ISD::CTPOP,      MVT::v8i16, Custom);
+
+    // NEON only has FMA instructions as of VFP4.
+    if (!Subtarget->hasVFP4()) {
+      setOperationAction(ISD::FMA, MVT::v2f32, Expand);
+      setOperationAction(ISD::FMA, MVT::v4f32, Expand);
+    }
+
     setTargetDAGCombine(ISD::INTRINSIC_VOID);
     setTargetDAGCombine(ISD::INTRINSIC_W_CHAIN);
     setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
@@ -688,7 +737,11 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
     setOperationAction(ISD::ATOMIC_LOAD_AND,  MVT::i64, Custom);
     setOperationAction(ISD::ATOMIC_LOAD_OR,   MVT::i64, Custom);
     setOperationAction(ISD::ATOMIC_LOAD_XOR,  MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_SWAP,  MVT::i64, Custom);
+    setOperationAction(ISD::ATOMIC_SWAP,      MVT::i64, Custom);
+    setOperationAction(ISD::ATOMIC_LOAD_MIN,  MVT::i64, Custom);
+    setOperationAction(ISD::ATOMIC_LOAD_MAX,  MVT::i64, Custom);
+    setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i64, Custom);
+    setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i64, Custom);
     setOperationAction(ISD::ATOMIC_CMP_SWAP,  MVT::i64, Custom);
     // Automatically insert fences (dmb ist) around ATOMIC_SWAP etc.
     setInsertFencesForAtomic(true);
@@ -762,6 +815,8 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
   setOperationAction(ISD::FSIN,      MVT::f32, Expand);
   setOperationAction(ISD::FCOS,      MVT::f32, Expand);
   setOperationAction(ISD::FCOS,      MVT::f64, Expand);
+  setOperationAction(ISD::FSINCOS,   MVT::f64, Expand);
+  setOperationAction(ISD::FSINCOS,   MVT::f32, Expand);
   setOperationAction(ISD::FREM,      MVT::f64, Expand);
   setOperationAction(ISD::FREM,      MVT::f32, Expand);
   if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
@@ -814,18 +869,19 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
     setSchedulingPreference(Sched::Hybrid);
 
   //// temporary - rewrite interface to use type
-  maxStoresPerMemcpy = maxStoresPerMemcpyOptSize = 1;
-  maxStoresPerMemset = 16;
-  maxStoresPerMemsetOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
+  MaxStoresPerMemset = 8;
+  MaxStoresPerMemsetOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
+  MaxStoresPerMemcpy = 4; // For @llvm.memcpy -> sequence of stores
+  MaxStoresPerMemcpyOptSize = Subtarget->isTargetDarwin() ? 4 : 2;
+  MaxStoresPerMemmove = 4; // For @llvm.memmove -> sequence of stores
+  MaxStoresPerMemmoveOptSize = Subtarget->isTargetDarwin() ? 4 : 2;
 
   // On ARM arguments smaller than 4 bytes are extended, so all arguments
   // are at least 4 bytes aligned.
   setMinStackArgumentAlignment(4);
 
-  benefitFromCodePlacementOpt = true;
-
   // Prefer likely predicted branches to selects on out-of-order cores.
-  predictableSelectIsExpensive = Subtarget->isLikeA9();
+  PredictableSelectIsExpensive = Subtarget->isLikeA9();
 
   setMinFunctionAlignment(Subtarget->isThumb() ? 1 : 2);
 }
@@ -841,10 +897,10 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
 // due to the common occurrence of cross class copies and subregister insertions
 // and extractions.
 std::pair<const TargetRegisterClass*, uint8_t>
-ARMTargetLowering::findRepresentativeClass(EVT VT) const{
+ARMTargetLowering::findRepresentativeClass(MVT VT) const{
   const TargetRegisterClass *RRC = 0;
   uint8_t Cost = 1;
-  switch (VT.getSimpleVT().SimpleTy) {
+  switch (VT.SimpleTy) {
   default:
     return TargetLowering::findRepresentativeClass(VT);
   // Use DPR as representative register class for all floating point
@@ -1024,7 +1080,7 @@ EVT ARMTargetLowering::getSetCCResultType(EVT VT) const {
 
 /// getRegClassFor - Return the register class that should be used for the
 /// specified value type.
-const TargetRegisterClass *ARMTargetLowering::getRegClassFor(EVT VT) const {
+const TargetRegisterClass *ARMTargetLowering::getRegClassFor(MVT VT) const {
   // Map v4i64 to QQ registers but do not make the type legal. Similarly map
   // v8i64 to QQQQ registers. v4i64 and v8i64 are only used for REG_SEQUENCE to
   // load / store 4 to 8 consecutive D registers.
@@ -1557,7 +1613,7 @@ ARMTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       // On ELF targets for PIC code, direct calls should go through the PLT
       unsigned OpFlags = 0;
       if (Subtarget->isTargetELF() &&
-                  getTargetMachine().getRelocationModel() == Reloc::PIC_)
+          getTargetMachine().getRelocationModel() == Reloc::PIC_)
         OpFlags = ARMII::MO_PLT;
       Callee = DAG.getTargetGlobalAddress(GV, dl, getPointerTy(), 0, OpFlags);
     }
@@ -1594,8 +1650,8 @@ ARMTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
 
   // FIXME: handle tail calls differently.
   unsigned CallOpc;
-  bool HasMinSizeAttr = MF.getFunction()->getFnAttributes().
-    hasAttribute(Attributes::MinSize);
+  bool HasMinSizeAttr = MF.getFunction()->getAttributes().
+    hasAttribute(AttributeSet::FunctionIndex, Attribute::MinSize);
   if (Subtarget->isThumb()) {
     if ((!isDirect || isARMFunc) && !Subtarget->hasV5TOps())
       CallOpc = ARMISD::CALL_NOLINK;
@@ -1875,6 +1931,17 @@ ARMTargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
   return true;
 }
 
+bool
+ARMTargetLowering::CanLowerReturn(CallingConv::ID CallConv,
+                                  MachineFunction &MF, bool isVarArg,
+                                  const SmallVectorImpl<ISD::OutputArg> &Outs,
+                                  LLVMContext &Context) const {
+  SmallVector<CCValAssign, 16> RVLocs;
+  CCState CCInfo(CallConv, isVarArg, MF, getTargetMachine(), RVLocs, Context);
+  return CCInfo.CheckReturn(Outs, CCAssignFnForNode(CallConv, /*Return=*/true,
+                                                    isVarArg));
+}
+
 SDValue
 ARMTargetLowering::LowerReturn(SDValue Chain,
                                CallingConv::ID CallConv, bool isVarArg,
@@ -1893,15 +1960,9 @@ ARMTargetLowering::LowerReturn(SDValue Chain,
   CCInfo.AnalyzeReturn(Outs, CCAssignFnForNode(CallConv, /* Return */ true,
                                                isVarArg));
 
-  // If this is the first return lowered for this function, add
-  // the regs to the liveout set for the function.
-  if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
-    for (unsigned i = 0; i != RVLocs.size(); ++i)
-      if (RVLocs[i].isRegLoc())
-        DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
-  }
-
   SDValue Flag;
+  SmallVector<SDValue, 4> RetOps;
+  RetOps.push_back(Chain); // Operand #0 = Chain (updated below)
 
   // Copy the result values into the output registers.
   for (unsigned i = 0, realRVLocIdx = 0;
@@ -1930,10 +1991,12 @@ ARMTargetLowering::LowerReturn(SDValue Chain,
 
         Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), HalfGPRs, Flag);
         Flag = Chain.getValue(1);
+        RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
         VA = RVLocs[++i]; // skip ahead to next loc
         Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
                                  HalfGPRs.getValue(1), Flag);
         Flag = Chain.getValue(1);
+        RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
         VA = RVLocs[++i]; // skip ahead to next loc
 
         // Extract the 2nd half and fall through to handle it as an f64 value.
@@ -1946,6 +2009,7 @@ ARMTargetLowering::LowerReturn(SDValue Chain,
                                   DAG.getVTList(MVT::i32, MVT::i32), &Arg, 1);
       Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), fmrrd, Flag);
       Flag = Chain.getValue(1);
+      RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
       VA = RVLocs[++i]; // skip ahead to next loc
       Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), fmrrd.getValue(1),
                                Flag);
@@ -1955,15 +2019,16 @@ ARMTargetLowering::LowerReturn(SDValue Chain,
     // Guarantee that all emitted copies are
     // stuck together, avoiding something bad.
     Flag = Chain.getValue(1);
+    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
   }
 
-  SDValue result;
+  // Update chain and glue.
+  RetOps[0] = Chain;
   if (Flag.getNode())
-    result = DAG.getNode(ARMISD::RET_FLAG, dl, MVT::Other, Chain, Flag);
-  else // Return Void
-    result = DAG.getNode(ARMISD::RET_FLAG, dl, MVT::Other, Chain);
+    RetOps.push_back(Flag);
 
-  return result;
+  return DAG.getNode(ARMISD::RET_FLAG, dl, MVT::Other,
+                     RetOps.data(), RetOps.size());
 }
 
 bool ARMTargetLowering::isUsedByReturnOnly(SDNode *N, SDValue &Chain) const {
@@ -2214,8 +2279,7 @@ SDValue ARMTargetLowering::LowerGlobalAddressELF(SDValue Op,
   EVT PtrVT = getPointerTy();
   DebugLoc dl = Op.getDebugLoc();
   const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
-  Reloc::Model RelocM = getTargetMachine().getRelocationModel();
-  if (RelocM == Reloc::PIC_) {
+  if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
     bool UseGOTOFF = GV->hasLocalLinkage() || GV->hasHiddenVisibility();
     ARMConstantPoolValue *CPV =
       ARMConstantPoolConstant::Create(GV,
@@ -2259,8 +2323,6 @@ SDValue ARMTargetLowering::LowerGlobalAddressDarwin(SDValue Op,
   DebugLoc dl = Op.getDebugLoc();
   const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
   Reloc::Model RelocM = getTargetMachine().getRelocationModel();
-  MachineFunction &MF = DAG.getMachineFunction();
-  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
 
   // FIXME: Enable this for static codegen when tool issues are fixed.  Also
   // update ARMFastISel::ARMMaterializeGV.
@@ -2288,6 +2350,7 @@ SDValue ARMTargetLowering::LowerGlobalAddressDarwin(SDValue Op,
   if (RelocM == Reloc::Static) {
     CPAddr = DAG.getTargetConstantPool(GV, PtrVT, 4);
   } else {
+    ARMFunctionInfo *AFI = DAG.getMachineFunction().getInfo<ARMFunctionInfo>();
     ARMPCLabelIndex = AFI->createPICLabelUId();
     unsigned PCAdj = (RelocM != Reloc::PIC_) ? 0 : (Subtarget->isThumb()?4:8);
     ARMConstantPoolValue *CPV =
@@ -2368,7 +2431,6 @@ ARMTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG,
     ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
     unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
     EVT PtrVT = getPointerTy();
-    DebugLoc dl = Op.getDebugLoc();
     Reloc::Model RelocM = getTargetMachine().getRelocationModel();
     SDValue CPAddr;
     unsigned PCAdj = (RelocM != Reloc::PIC_)
@@ -2543,7 +2605,7 @@ ARMTargetLowering::computeRegArea(CCState &CCInfo, MachineFunction &MF,
 }
 
 // The remaining GPRs hold either the beginning of variable-argument
-// data, or the beginning of an aggregate passed by value (usuall
+// data, or the beginning of an aggregate passed by value (usually
 // byval).  Either way, we allocate stack slots adjacent to the data
 // provided by our caller, and store the unallocated registers there.
 // If this is a variadic function, the va_list pointer will begin with
@@ -2628,7 +2690,7 @@ ARMTargetLowering::LowerFormalArguments(SDValue Chain,
   CCInfo.AnalyzeFormalArguments(Ins,
                                 CCAssignFnForNode(CallConv, /* Return*/ false,
                                                   isVarArg));
-  
+
   SmallVector<SDValue, 16> ArgValues;
   int lastInsIndex = -1;
   SDValue ArgValue;
@@ -2743,7 +2805,7 @@ ARMTargetLowering::LowerFormalArguments(SDValue Chain,
             } else {
               int FI = MFI->CreateFixedObject(Flags.getByValSize(),
                                               VA.getLocMemOffset(), false);
-              InVals.push_back(DAG.getFrameIndex(FI, getPointerTy()));              
+              InVals.push_back(DAG.getFrameIndex(FI, getPointerTy()));
             }
           } else {
             int FI = MFI->CreateFixedObject(VA.getLocVT().getSizeInBits()/8,
@@ -3379,6 +3441,47 @@ SDValue ARMTargetLowering::LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const {
   return FrameAddr;
 }
 
+/// Custom Expand long vector extensions, where size(DestVec) > 2*size(SrcVec),
+/// and size(DestVec) > 128-bits.
+/// This is achieved by doing the one extension from the SrcVec, splitting the
+/// result, extending these parts, and then concatenating these into the
+/// destination.
+static SDValue ExpandVectorExtension(SDNode *N, SelectionDAG &DAG) {
+  SDValue Op = N->getOperand(0);
+  EVT SrcVT = Op.getValueType();
+  EVT DestVT = N->getValueType(0);
+
+  assert(DestVT.getSizeInBits() > 128 &&
+         "Custom sext/zext expansion needs >128-bit vector.");
+  // If this is a normal length extension, use the default expansion.
+  if (SrcVT.getSizeInBits()*4 != DestVT.getSizeInBits() &&
+      SrcVT.getSizeInBits()*8 != DestVT.getSizeInBits())
+    return SDValue();
+
+  DebugLoc dl = N->getDebugLoc();
+  unsigned SrcEltSize = SrcVT.getVectorElementType().getSizeInBits();
+  unsigned DestEltSize = DestVT.getVectorElementType().getSizeInBits();
+  unsigned NumElts = SrcVT.getVectorNumElements();
+  LLVMContext &Ctx = *DAG.getContext();
+  SDValue Mid, SplitLo, SplitHi, ExtLo, ExtHi;
+
+  EVT MidVT = EVT::getVectorVT(Ctx, EVT::getIntegerVT(Ctx, SrcEltSize*2),
+                               NumElts);
+  EVT SplitVT = EVT::getVectorVT(Ctx, EVT::getIntegerVT(Ctx, SrcEltSize*2),
+                                 NumElts/2);
+  EVT ExtVT = EVT::getVectorVT(Ctx, EVT::getIntegerVT(Ctx, DestEltSize),
+                               NumElts/2);
+
+  Mid = DAG.getNode(N->getOpcode(), dl, MidVT, Op);
+  SplitLo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, SplitVT, Mid,
+                        DAG.getIntPtrConstant(0));
+  SplitHi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, SplitVT, Mid,
+                        DAG.getIntPtrConstant(NumElts/2));
+  ExtLo = DAG.getNode(N->getOpcode(), dl, ExtVT, SplitLo);
+  ExtHi = DAG.getNode(N->getOpcode(), dl, ExtVT, SplitHi);
+  return DAG.getNode(ISD::CONCAT_VECTORS, dl, DestVT, ExtLo, ExtHi);
+}
+
 /// ExpandBITCAST - If the target supports VFP, this function is called to
 /// expand a bit convert where either the source or destination type is i64 to
 /// use a VMOVDRR or VMOVRRD node.  This should not be done when the non-i64
@@ -3532,6 +3635,114 @@ static SDValue LowerCTTZ(SDNode *N, SelectionDAG &DAG,
   return DAG.getNode(ISD::CTLZ, dl, VT, rbit);
 }
 
+/// getCTPOP16BitCounts - Returns a v8i8/v16i8 vector containing the bit-count
+/// for each 16-bit element from operand, repeated.  The basic idea is to
+/// leverage vcnt to get the 8-bit counts, gather and add the results.
+///
+/// Trace for v4i16:
+/// input    = [v0    v1    v2    v3   ] (vi 16-bit element)
+/// cast: N0 = [w0 w1 w2 w3 w4 w5 w6 w7] (v0 = [w0 w1], wi 8-bit element)
+/// vcnt: N1 = [b0 b1 b2 b3 b4 b5 b6 b7] (bi = bit-count of 8-bit element wi)
+/// vrev: N2 = [b1 b0 b3 b2 b5 b4 b7 b6]
+///            [b0 b1 b2 b3 b4 b5 b6 b7]
+///           +[b1 b0 b3 b2 b5 b4 b7 b6]
+/// N3=N1+N2 = [k0 k0 k1 k1 k2 k2 k3 k3] (k0 = b0+b1 = bit-count of 16-bit v0,
+/// vuzp:    = [k0 k1 k2 k3 k0 k1 k2 k3]  each ki is 8-bits)
+static SDValue getCTPOP16BitCounts(SDNode *N, SelectionDAG &DAG) {
+  EVT VT = N->getValueType(0);
+  DebugLoc DL = N->getDebugLoc();
+
+  EVT VT8Bit = VT.is64BitVector() ? MVT::v8i8 : MVT::v16i8;
+  SDValue N0 = DAG.getNode(ISD::BITCAST, DL, VT8Bit, N->getOperand(0));
+  SDValue N1 = DAG.getNode(ISD::CTPOP, DL, VT8Bit, N0);
+  SDValue N2 = DAG.getNode(ARMISD::VREV16, DL, VT8Bit, N1);
+  SDValue N3 = DAG.getNode(ISD::ADD, DL, VT8Bit, N1, N2);
+  return DAG.getNode(ARMISD::VUZP, DL, VT8Bit, N3, N3);
+}
+
+/// lowerCTPOP16BitElements - Returns a v4i16/v8i16 vector containing the
+/// bit-count for each 16-bit element from the operand.  We need slightly
+/// different sequencing for v4i16 and v8i16 to stay within NEON's available
+/// 64/128-bit registers.
+///
+/// Trace for v4i16:
+/// input           = [v0    v1    v2    v3    ] (vi 16-bit element)
+/// v8i8: BitCounts = [k0 k1 k2 k3 k0 k1 k2 k3 ] (ki is the bit-count of vi)
+/// v8i16:Extended  = [k0    k1    k2    k3    k0    k1    k2    k3    ]
+/// v4i16:Extracted = [k0    k1    k2    k3    ]
+static SDValue lowerCTPOP16BitElements(SDNode *N, SelectionDAG &DAG) {
+  EVT VT = N->getValueType(0);
+  DebugLoc DL = N->getDebugLoc();
+
+  SDValue BitCounts = getCTPOP16BitCounts(N, DAG);
+  if (VT.is64BitVector()) {
+    SDValue Extended = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v8i16, BitCounts);
+    return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v4i16, Extended,
+                       DAG.getIntPtrConstant(0));
+  } else {
+    SDValue Extracted = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v8i8,
+                                    BitCounts, DAG.getIntPtrConstant(0));
+    return DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v8i16, Extracted);
+  }
+}
+
+/// lowerCTPOP32BitElements - Returns a v2i32/v4i32 vector containing the
+/// bit-count for each 32-bit element from the operand.  The idea here is
+/// to split the vector into 16-bit elements, leverage the 16-bit count
+/// routine, and then combine the results.
+///
+/// Trace for v2i32 (v4i32 similar with Extracted/Extended exchanged):
+/// input    = [v0    v1    ] (vi: 32-bit elements)
+/// Bitcast  = [w0 w1 w2 w3 ] (wi: 16-bit elements, v0 = [w0 w1])
+/// Counts16 = [k0 k1 k2 k3 ] (ki: 16-bit elements, bit-count of wi)
+/// vrev: N0 = [k1 k0 k3 k2 ]
+///            [k0 k1 k2 k3 ]
+///       N1 =+[k1 k0 k3 k2 ]
+///            [k0 k2 k1 k3 ]
+///       N2 =+[k1 k3 k0 k2 ]
+///            [k0    k2    k1    k3    ]
+/// Extended =+[k1    k3    k0    k2    ]
+///            [k0    k2    ]
+/// Extracted=+[k1    k3    ]
+///
+static SDValue lowerCTPOP32BitElements(SDNode *N, SelectionDAG &DAG) {
+  EVT VT = N->getValueType(0);
+  DebugLoc DL = N->getDebugLoc();
+
+  EVT VT16Bit = VT.is64BitVector() ? MVT::v4i16 : MVT::v8i16;
+
+  SDValue Bitcast = DAG.getNode(ISD::BITCAST, DL, VT16Bit, N->getOperand(0));
+  SDValue Counts16 = lowerCTPOP16BitElements(Bitcast.getNode(), DAG);
+  SDValue N0 = DAG.getNode(ARMISD::VREV32, DL, VT16Bit, Counts16);
+  SDValue N1 = DAG.getNode(ISD::ADD, DL, VT16Bit, Counts16, N0);
+  SDValue N2 = DAG.getNode(ARMISD::VUZP, DL, VT16Bit, N1, N1);
+
+  if (VT.is64BitVector()) {
+    SDValue Extended = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v4i32, N2);
+    return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v2i32, Extended,
+                       DAG.getIntPtrConstant(0));
+  } else {
+    SDValue Extracted = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v4i16, N2,
+                                    DAG.getIntPtrConstant(0));
+    return DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v4i32, Extracted);
+  }
+}
+
+static SDValue LowerCTPOP(SDNode *N, SelectionDAG &DAG,
+                          const ARMSubtarget *ST) {
+  EVT VT = N->getValueType(0);
+
+  assert(ST->hasNEON() && "Custom ctpop lowering requires NEON.");
+  assert((VT == MVT::v2i32 || VT == MVT::v4i32 ||
+          VT == MVT::v4i16 || VT == MVT::v8i16) &&
+         "Unexpected type for custom ctpop lowering");
+
+  if (VT.getVectorElementType() == MVT::i32)
+    return lowerCTPOP32BitElements(N, DAG);
+  else
+    return lowerCTPOP16BitElements(N, DAG);
+}
+
 static SDValue LowerShift(SDNode *N, SelectionDAG &DAG,
                           const ARMSubtarget *ST) {
   EVT VT = N->getValueType(0);
@@ -4153,6 +4364,21 @@ static bool isVZIP_v_undef_Mask(ArrayRef<int> M, EVT VT, unsigned &WhichResult){
   return true;
 }
 
+/// \return true if this is a reverse operation on an vector.
+static bool isReverseMask(ArrayRef<int> M, EVT VT) {
+  unsigned NumElts = VT.getVectorNumElements();
+  // Make sure the mask has the right size.
+  if (NumElts != M.size())
+      return false;
+
+  // Look for <15, ..., 3, -1, 1, 0>.
+  for (unsigned i = 0; i != NumElts; ++i)
+    if (M[i] >= 0 && M[i] != (int) (NumElts - 1 - i))
+      return false;
+
+  return true;
+}
+
 // If N is an integer constant that can be moved into a register in one
 // instruction, return an SDValue of such a constant (will become a MOV
 // instruction).  Otherwise return null.
@@ -4247,7 +4473,7 @@ SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
 
     ValueCounts.insert(std::make_pair(V, 0));
     unsigned &Count = ValueCounts[V];
-    
+
     // Is this value dominant? (takes up more than half of the lanes)
     if (++Count > (NumElts / 2)) {
       hasDominantValue = true;
@@ -4275,8 +4501,11 @@ SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
 
       // If we are VDUPing a value that comes directly from a vector, that will
       // cause an unnecessary move to and from a GPR, where instead we could
-      // just use VDUPLANE.
-      if (Value->getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
+      // just use VDUPLANE. We can only do this if the lane being extracted
+      // is at a constant index, as the VDUP from lane instructions only have
+      // constant-index forms.
+      if (Value->getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
+          isa<ConstantSDNode>(Value->getOperand(1))) {
         // We need to create a new undef vector to use for the VDUPLANE if the
         // size of the vector from which we get the value is different than the
         // size of the vector that we need to create. We will insert the element
@@ -4291,12 +4520,10 @@ SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
                  DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, DAG.getUNDEF(VT),
                         Value, DAG.getConstant(index, MVT::i32)),
                            DAG.getConstant(index, MVT::i32));
-        } else {
+        } else
           N = DAG.getNode(ARMISD::VDUPLANE, dl, VT,
                         Value->getOperand(0), Value->getOperand(1));
-        }
-      }
-      else
+      } else
         N = DAG.getNode(ARMISD::VDUP, dl, VT, Value);
 
       if (!usesOnlyOneValue) {
@@ -4328,7 +4555,7 @@ SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
     if (usesOnlyOneValue) {
       SDValue Val = IsSingleInstrConstant(Value, DAG, ST, dl);
       if (isConstant && Val.getNode())
-        return DAG.getNode(ARMISD::VDUP, dl, VT, Val); 
+        return DAG.getNode(ARMISD::VDUP, dl, VT, Val);
     }
   }
 
@@ -4548,7 +4775,8 @@ ARMTargetLowering::isShuffleMaskLegal(const SmallVectorImpl<int> &M,
           isVZIPMask(M, VT, WhichResult) ||
           isVTRN_v_undef_Mask(M, VT, WhichResult) ||
           isVUZP_v_undef_Mask(M, VT, WhichResult) ||
-          isVZIP_v_undef_Mask(M, VT, WhichResult));
+          isVZIP_v_undef_Mask(M, VT, WhichResult) ||
+          ((VT == MVT::v8i16 || VT == MVT::v16i8) && isReverseMask(M, VT)));
 }
 
 /// GeneratePerfectShuffle - Given an entry in the perfect-shuffle table, emit
@@ -4652,6 +4880,23 @@ static SDValue LowerVECTOR_SHUFFLEv8i8(SDValue Op,
                                  &VTBLMask[0], 8));
 }
 
+static SDValue LowerReverse_VECTOR_SHUFFLEv16i8_v8i16(SDValue Op,
+                                                      SelectionDAG &DAG) {
+  DebugLoc DL = Op.getDebugLoc();
+  SDValue OpLHS = Op.getOperand(0);
+  EVT VT = OpLHS.getValueType();
+
+  assert((VT == MVT::v8i16 || VT == MVT::v16i8) &&
+         "Expect an v8i16/v16i8 type");
+  OpLHS = DAG.getNode(ARMISD::VREV64, DL, VT, OpLHS);
+  // For a v16i8 type: After the VREV, we have got <8, ...15, 8, ..., 0>. Now,
+  // extract the first 8 bytes into the top double word and the last 8 bytes
+  // into the bottom double word. The v8i16 case is similar.
+  unsigned ExtractNum = (VT == MVT::v16i8) ? 8 : 4;
+  return DAG.getNode(ARMISD::VEXT, DL, VT, OpLHS, OpLHS,
+                     DAG.getConstant(ExtractNum, MVT::i32));
+}
+
 static SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) {
   SDValue V1 = Op.getOperand(0);
   SDValue V2 = Op.getOperand(1);
@@ -4789,6 +5034,9 @@ static SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) {
     return DAG.getNode(ISD::BITCAST, dl, VT, Val);
   }
 
+  if ((VT == MVT::v8i16 || VT == MVT::v16i8) && isReverseMask(ShuffleMask, VT))
+    return LowerReverse_VECTOR_SHUFFLEv16i8_v8i16(Op, DAG);
+
   if (VT == MVT::v8i8) {
     SDValue NewOp = LowerVECTOR_SHUFFLEv8i8(Op, ShuffleMask, DAG);
     if (NewOp.getNode())
@@ -4917,16 +5165,76 @@ static bool isZeroExtended(SDNode *N, SelectionDAG &DAG) {
   return false;
 }
 
-/// SkipExtension - For a node that is a SIGN_EXTEND, ZERO_EXTEND, extending
-/// load, or BUILD_VECTOR with extended elements, return the unextended value.
-static SDValue SkipExtension(SDNode *N, SelectionDAG &DAG) {
+/// AddRequiredExtensionForVMULL - Add a sign/zero extension to extend the total
+/// value size to 64 bits. We need a 64-bit D register as an operand to VMULL.
+/// We insert the required extension here to get the vector to fill a D register.
+static SDValue AddRequiredExtensionForVMULL(SDValue N, SelectionDAG &DAG,
+                                            const EVT &OrigTy,
+                                            const EVT &ExtTy,
+                                            unsigned ExtOpcode) {
+  // The vector originally had a size of OrigTy. It was then extended to ExtTy.
+  // We expect the ExtTy to be 128-bits total. If the OrigTy is less than
+  // 64-bits we need to insert a new extension so that it will be 64-bits.
+  assert(ExtTy.is128BitVector() && "Unexpected extension size");
+  if (OrigTy.getSizeInBits() >= 64)
+    return N;
+
+  // Must extend size to at least 64 bits to be used as an operand for VMULL.
+  MVT::SimpleValueType OrigSimpleTy = OrigTy.getSimpleVT().SimpleTy;
+  EVT NewVT;
+  switch (OrigSimpleTy) {
+  default: llvm_unreachable("Unexpected Orig Vector Type");
+  case MVT::v2i8:
+  case MVT::v2i16:
+    NewVT = MVT::v2i32;
+    break;
+  case MVT::v4i8:
+    NewVT = MVT::v4i16;
+    break;
+  }
+  return DAG.getNode(ExtOpcode, N->getDebugLoc(), NewVT, N);
+}
+
+/// SkipLoadExtensionForVMULL - return a load of the original vector size that
+/// does not do any sign/zero extension. If the original vector is less
+/// than 64 bits, an appropriate extension will be added after the load to
+/// reach a total size of 64 bits. We have to add the extension separately
+/// because ARM does not have a sign/zero extending load for vectors.
+static SDValue SkipLoadExtensionForVMULL(LoadSDNode *LD, SelectionDAG& DAG) {
+  SDValue NonExtendingLoad =
+    DAG.getLoad(LD->getMemoryVT(), LD->getDebugLoc(), LD->getChain(),
+                LD->getBasePtr(), LD->getPointerInfo(), LD->isVolatile(),
+                LD->isNonTemporal(), LD->isInvariant(),
+                LD->getAlignment());
+  unsigned ExtOp = 0;
+  switch (LD->getExtensionType()) {
+  default: llvm_unreachable("Unexpected LoadExtType");
+  case ISD::EXTLOAD:
+  case ISD::SEXTLOAD: ExtOp = ISD::SIGN_EXTEND; break;
+  case ISD::ZEXTLOAD: ExtOp = ISD::ZERO_EXTEND; break;
+  }
+  MVT::SimpleValueType MemType = LD->getMemoryVT().getSimpleVT().SimpleTy;
+  MVT::SimpleValueType ExtType = LD->getValueType(0).getSimpleVT().SimpleTy;
+  return AddRequiredExtensionForVMULL(NonExtendingLoad, DAG,
+                                      MemType, ExtType, ExtOp);
+}
+
+/// SkipExtensionForVMULL - For a node that is a SIGN_EXTEND, ZERO_EXTEND,
+/// extending load, or BUILD_VECTOR with extended elements, return the
+/// unextended value. The unextended vector should be 64 bits so that it can
+/// be used as an operand to a VMULL instruction. If the original vector size
+/// before extension is less than 64 bits we add a an extension to resize
+/// the vector to 64 bits.
+static SDValue SkipExtensionForVMULL(SDNode *N, SelectionDAG &DAG) {
   if (N->getOpcode() == ISD::SIGN_EXTEND || N->getOpcode() == ISD::ZERO_EXTEND)
-    return N->getOperand(0);
+    return AddRequiredExtensionForVMULL(N->getOperand(0), DAG,
+                                        N->getOperand(0)->getValueType(0),
+                                        N->getValueType(0),
+                                        N->getOpcode());
+
   if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N))
-    return DAG.getLoad(LD->getMemoryVT(), N->getDebugLoc(), LD->getChain(),
-                       LD->getBasePtr(), LD->getPointerInfo(), LD->isVolatile(),
-                       LD->isNonTemporal(), LD->isInvariant(),
-                       LD->getAlignment());
+    return SkipLoadExtensionForVMULL(LD, DAG);
+
   // Otherwise, the value must be a BUILD_VECTOR.  For v2i64, it will
   // have been legalized as a BITCAST from v4i32.
   if (N->getOpcode() == ISD::BITCAST) {
@@ -4981,7 +5289,8 @@ static SDValue LowerMUL(SDValue Op, SelectionDAG &DAG) {
   // Multiplications are only custom-lowered for 128-bit vectors so that
   // VMULL can be detected.  Otherwise v2i64 multiplications are not legal.
   EVT VT = Op.getValueType();
-  assert(VT.is128BitVector() && "unexpected type for custom-lowering ISD::MUL");
+  assert(VT.is128BitVector() && VT.isInteger() &&
+         "unexpected type for custom-lowering ISD::MUL");
   SDNode *N0 = Op.getOperand(0).getNode();
   SDNode *N1 = Op.getOperand(1).getNode();
   unsigned NewOpc = 0;
@@ -5024,9 +5333,9 @@ static SDValue LowerMUL(SDValue Op, SelectionDAG &DAG) {
   // Legalize to a VMULL instruction.
   DebugLoc DL = Op.getDebugLoc();
   SDValue Op0;
-  SDValue Op1 = SkipExtension(N1, DAG);
+  SDValue Op1 = SkipExtensionForVMULL(N1, DAG);
   if (!isMLA) {
-    Op0 = SkipExtension(N0, DAG);
+    Op0 = SkipExtensionForVMULL(N0, DAG);
     assert(Op0.getValueType().is64BitVector() &&
            Op1.getValueType().is64BitVector() &&
            "unexpected types for extended operands to VMULL");
@@ -5041,8 +5350,8 @@ static SDValue LowerMUL(SDValue Op, SelectionDAG &DAG) {
   //   vaddl q0, d4, d5
   //   vmovl q1, d6
   //   vmul  q0, q0, q1
-  SDValue N00 = SkipExtension(N0->getOperand(0).getNode(), DAG);
-  SDValue N01 = SkipExtension(N0->getOperand(1).getNode(), DAG);
+  SDValue N00 = SkipExtensionForVMULL(N0->getOperand(0).getNode(), DAG);
+  SDValue N01 = SkipExtensionForVMULL(N0->getOperand(1).getNode(), DAG);
   EVT Op1VT = Op1.getValueType();
   return DAG.getNode(N0->getOpcode(), DL, VT,
                      DAG.getNode(NewOpc, DL, VT,
@@ -5328,6 +5637,7 @@ SDValue ARMTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::SRL_PARTS:
   case ISD::SRA_PARTS:     return LowerShiftRightParts(Op, DAG);
   case ISD::CTTZ:          return LowerCTTZ(Op.getNode(), DAG, Subtarget);
+  case ISD::CTPOP:         return LowerCTPOP(Op.getNode(), DAG, Subtarget);
   case ISD::SETCC:         return LowerVSETCC(Op, DAG);
   case ISD::ConstantFP:    return LowerConstantFP(Op, DAG, Subtarget);
   case ISD::BUILD_VECTOR:  return LowerBUILD_VECTOR(Op, DAG, Subtarget);
@@ -5360,6 +5670,10 @@ void ARMTargetLowering::ReplaceNodeResults(SDNode *N,
   case ISD::BITCAST:
     Res = ExpandBITCAST(N, DAG);
     break;
+  case ISD::SIGN_EXTEND:
+  case ISD::ZERO_EXTEND:
+    Res = ExpandVectorExtension(N, DAG);
+    break;
   case ISD::SRL:
   case ISD::SRA:
     Res = Expand64BitShift(N, DAG, Subtarget);
@@ -5388,6 +5702,18 @@ void ARMTargetLowering::ReplaceNodeResults(SDNode *N,
   case ISD::ATOMIC_CMP_SWAP:
     ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMCMPXCHG64_DAG);
     return;
+  case ISD::ATOMIC_LOAD_MIN:
+    ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMMIN64_DAG);
+    return;
+  case ISD::ATOMIC_LOAD_UMIN:
+    ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMUMIN64_DAG);
+    return;
+  case ISD::ATOMIC_LOAD_MAX:
+    ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMMAX64_DAG);
+    return;
+  case ISD::ATOMIC_LOAD_UMAX:
+    ReplaceATOMIC_OP_64(N, Results, DAG, ARMISD::ATOMUMAX64_DAG);
+    return;
   }
   if (Res.getNode())
     Results.push_back(Res);
@@ -5727,7 +6053,8 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI,
 MachineBasicBlock *
 ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
                                       unsigned Op1, unsigned Op2,
-                                      bool NeedsCarry, bool IsCmpxchg) const {
+                                      bool NeedsCarry, bool IsCmpxchg,
+                                      bool IsMinMax, ARMCC::CondCodes CC) const {
   // This also handles ATOMIC_SWAP, indicated by Op1==0.
   const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
 
@@ -5751,21 +6078,17 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
     MRI.constrainRegClass(ptr, &ARM::rGPRRegClass);
   }
 
-  unsigned ldrOpc = isThumb2 ? ARM::t2LDREXD : ARM::LDREXD;
-  unsigned strOpc = isThumb2 ? ARM::t2STREXD : ARM::STREXD;
-
   MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
   MachineBasicBlock *contBB = 0, *cont2BB = 0;
-  if (IsCmpxchg) {
+  if (IsCmpxchg || IsMinMax)
     contBB = MF->CreateMachineBasicBlock(LLVM_BB);
+  if (IsCmpxchg)
     cont2BB = MF->CreateMachineBasicBlock(LLVM_BB);
-  }
   MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+
   MF->insert(It, loopMBB);
-  if (IsCmpxchg) {
-    MF->insert(It, contBB);
-    MF->insert(It, cont2BB);
-  }
+  if (IsCmpxchg || IsMinMax) MF->insert(It, contBB);
+  if (IsCmpxchg) MF->insert(It, cont2BB);
   MF->insert(It, exitMBB);
 
   // Transfer the remainder of BB and its successor edges to exitMBB.
@@ -5792,22 +6115,26 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
   //   cmp storesuccess, #0
   //   bne- loopMBB
   //   fallthrough --> exitMBB
-  //
-  // Note that the registers are explicitly specified because there is not any
-  // way to force the register allocator to allocate a register pair.
-  //
-  // FIXME: The hardcoded registers are not necessary for Thumb2, but we
-  // need to properly enforce the restriction that the two output registers
-  // for ldrexd must be different.
   BB = loopMBB;
+
   // Load
-  AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc))
-                 .addReg(ARM::R2, RegState::Define)
-                 .addReg(ARM::R3, RegState::Define).addReg(ptr));
-  // Copy r2/r3 into dest.  (This copy will normally be coalesced.)
-  BuildMI(BB, dl, TII->get(TargetOpcode::COPY), destlo).addReg(ARM::R2);
-  BuildMI(BB, dl, TII->get(TargetOpcode::COPY), desthi).addReg(ARM::R3);
+  if (isThumb2) {
+    AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2LDREXD))
+                   .addReg(destlo, RegState::Define)
+                   .addReg(desthi, RegState::Define)
+                   .addReg(ptr));
+  } else {
+    unsigned GPRPair0 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::LDREXD))
+                   .addReg(GPRPair0, RegState::Define).addReg(ptr));
+    // Copy r2/r3 into dest.  (This copy will normally be coalesced.)
+    BuildMI(BB, dl, TII->get(TargetOpcode::COPY), destlo)
+      .addReg(GPRPair0, 0, ARM::gsub_0);
+    BuildMI(BB, dl, TII->get(TargetOpcode::COPY), desthi)
+      .addReg(GPRPair0, 0, ARM::gsub_1);
+  }
 
+  unsigned StoreLo, StoreHi;
   if (IsCmpxchg) {
     // Add early exit
     for (unsigned i = 0; i < 2; i++) {
@@ -5823,26 +6150,60 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB,
     }
 
     // Copy to physregs for strexd
-    unsigned setlo = MI->getOperand(5).getReg();
-    unsigned sethi = MI->getOperand(6).getReg();
-    BuildMI(BB, dl, TII->get(TargetOpcode::COPY), ARM::R0).addReg(setlo);
-    BuildMI(BB, dl, TII->get(TargetOpcode::COPY), ARM::R1).addReg(sethi);
+    StoreLo = MI->getOperand(5).getReg();
+    StoreHi = MI->getOperand(6).getReg();
   } else if (Op1) {
     // Perform binary operation
-    AddDefaultPred(BuildMI(BB, dl, TII->get(Op1), ARM::R0)
+    unsigned tmpRegLo = MRI.createVirtualRegister(TRC);
+    AddDefaultPred(BuildMI(BB, dl, TII->get(Op1), tmpRegLo)
                    .addReg(destlo).addReg(vallo))
         .addReg(NeedsCarry ? ARM::CPSR : 0, getDefRegState(NeedsCarry));
-    AddDefaultPred(BuildMI(BB, dl, TII->get(Op2), ARM::R1)
-                   .addReg(desthi).addReg(valhi)).addReg(0);
+    unsigned tmpRegHi = MRI.createVirtualRegister(TRC);
+    AddDefaultPred(BuildMI(BB, dl, TII->get(Op2), tmpRegHi)
+                   .addReg(desthi).addReg(valhi))
+        .addReg(IsMinMax ? ARM::CPSR : 0, getDefRegState(IsMinMax));
+
+    StoreLo = tmpRegLo;
+    StoreHi = tmpRegHi;
   } else {
     // Copy to physregs for strexd
-    BuildMI(BB, dl, TII->get(TargetOpcode::COPY), ARM::R0).addReg(vallo);
-    BuildMI(BB, dl, TII->get(TargetOpcode::COPY), ARM::R1).addReg(valhi);
+    StoreLo = vallo;
+    StoreHi = valhi;
+  }
+  if (IsMinMax) {
+    // Compare and branch to exit block.
+    BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
+      .addMBB(exitMBB).addImm(CC).addReg(ARM::CPSR);
+    BB->addSuccessor(exitMBB);
+    BB->addSuccessor(contBB);
+    BB = contBB;
+    StoreLo = vallo;
+    StoreHi = valhi;
   }
 
   // Store
-  AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), storesuccess)
-                 .addReg(ARM::R0).addReg(ARM::R1).addReg(ptr));
+  if (isThumb2) {
+    AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2STREXD), storesuccess)
+                   .addReg(StoreLo).addReg(StoreHi).addReg(ptr));
+  } else {
+    // Marshal a pair...
+    unsigned StorePair = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    unsigned UndefPair = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    unsigned r1 = MRI.createVirtualRegister(&ARM::GPRPairRegClass);
+    BuildMI(BB, dl, TII->get(TargetOpcode::IMPLICIT_DEF), UndefPair);
+    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), r1)
+      .addReg(UndefPair)
+      .addReg(StoreLo)
+      .addImm(ARM::gsub_0);
+    BuildMI(BB, dl, TII->get(TargetOpcode::INSERT_SUBREG), StorePair)
+      .addReg(r1)
+      .addReg(StoreHi)
+      .addImm(ARM::gsub_1);
+
+    // ...and store it
+    AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::STREXD), storesuccess)
+                   .addReg(StorePair).addReg(ptr));
+  }
   // Cmp+jump
   AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
                  .addReg(storesuccess).addImm(0));
@@ -6043,6 +6404,7 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
     MF->getOrCreateJumpTableInfo(MachineJumpTableInfo::EK_Inline);
   unsigned MJTI = JTI->createJumpTableIndex(LPadList);
   unsigned UId = AFI->createJumpTableUId();
+  Reloc::Model RelocM = getTargetMachine().getRelocationModel();
 
   // Create the MBBs for the dispatch code.
 
@@ -6051,7 +6413,13 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
   DispatchBB->setIsLandingPad();
 
   MachineBasicBlock *TrapBB = MF->CreateMachineBasicBlock();
-  BuildMI(TrapBB, dl, TII->get(Subtarget->isThumb() ? ARM::tTRAP : ARM::TRAP));
+  unsigned trap_opcode;
+  if (Subtarget->isThumb())
+    trap_opcode = ARM::tTRAP;
+  else
+    trap_opcode = Subtarget->useNaClTrap() ? ARM::TRAPNaCl : ARM::TRAP;
+
+  BuildMI(TrapBB, dl, TII->get(trap_opcode));
   DispatchBB->addSuccessor(TrapBB);
 
   MachineBasicBlock *DispContBB = MF->CreateMachineBasicBlock();
@@ -6197,11 +6565,14 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
                    .addImm(0)
                    .addMemOperand(JTMMOLd));
 
-    unsigned NewVReg6 = MRI->createVirtualRegister(TRC);
-    AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg6)
-                   .addReg(ARM::CPSR, RegState::Define)
-                   .addReg(NewVReg5, RegState::Kill)
-                   .addReg(NewVReg3));
+    unsigned NewVReg6 = NewVReg5;
+    if (RelocM == Reloc::PIC_) {
+      NewVReg6 = MRI->createVirtualRegister(TRC);
+      AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg6)
+                     .addReg(ARM::CPSR, RegState::Define)
+                     .addReg(NewVReg5, RegState::Kill)
+                     .addReg(NewVReg3));
+    }
 
     BuildMI(DispContBB, dl, TII->get(ARM::tBR_JTr))
       .addReg(NewVReg6, RegState::Kill)
@@ -6281,11 +6652,18 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
       .addImm(0)
       .addMemOperand(JTMMOLd));
 
-    BuildMI(DispContBB, dl, TII->get(ARM::BR_JTadd))
-      .addReg(NewVReg5, RegState::Kill)
-      .addReg(NewVReg4)
-      .addJumpTableIndex(MJTI)
-      .addImm(UId);
+    if (RelocM == Reloc::PIC_) {
+      BuildMI(DispContBB, dl, TII->get(ARM::BR_JTadd))
+        .addReg(NewVReg5, RegState::Kill)
+        .addReg(NewVReg4)
+        .addJumpTableIndex(MJTI)
+        .addImm(UId);
+    } else {
+      BuildMI(DispContBB, dl, TII->get(ARM::BR_JTr))
+        .addReg(NewVReg5, RegState::Kill)
+        .addJumpTableIndex(MJTI)
+        .addImm(UId);
+    }
   }
 
   // Add the jump table entries as successors to the MBB.
@@ -6334,7 +6712,7 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
         DefRegs[OI->getReg()] = true;
       }
 
-      MachineInstrBuilder MIB(&*II);
+      MachineInstrBuilder MIB(*MF, &*II);
 
       for (unsigned i = 0; SavedRegs[i] != 0; ++i) {
         unsigned Reg = SavedRegs[i];
@@ -6411,8 +6789,9 @@ EmitStructByval(MachineInstr *MI, MachineBasicBlock *BB) const {
     UnitSize = 2;
   } else {
     // Check whether we can use NEON instructions.
-    if (!MF->getFunction()->getFnAttributes().
-          hasAttribute(Attributes::NoImplicitFloat) &&
+    if (!MF->getFunction()->getAttributes().
+          hasAttribute(AttributeSet::FunctionIndex,
+                       Attribute::NoImplicitFloat) &&
         Subtarget->hasNEON()) {
       if ((Align % 16 == 0) && SizeVal >= 16) {
         ldrOpc = ARM::VLD1q32wb_fixed;
@@ -6840,6 +7219,26 @@ ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
     return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
                               isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
                               /*NeedsCarry*/ false, /*IsCmpxchg*/true);
+  case ARM::ATOMMIN6432:
+    return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
+                              isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
+                              /*NeedsCarry*/ true, /*IsCmpxchg*/false,
+                              /*IsMinMax*/ true, ARMCC::LT);
+  case ARM::ATOMMAX6432:
+    return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
+                              isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
+                              /*NeedsCarry*/ true, /*IsCmpxchg*/false,
+                              /*IsMinMax*/ true, ARMCC::GE);
+  case ARM::ATOMUMIN6432:
+    return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
+                              isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
+                              /*NeedsCarry*/ true, /*IsCmpxchg*/false,
+                              /*IsMinMax*/ true, ARMCC::LO);
+  case ARM::ATOMUMAX6432:
+    return EmitAtomicBinary64(MI, BB, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr,
+                              isThumb2 ? ARM::t2SBCrr : ARM::SBCrr,
+                              /*NeedsCarry*/ true, /*IsCmpxchg*/false,
+                              /*IsMinMax*/ true, ARMCC::HS);
 
   case ARM::tMOVCCr_pseudo: {
     // To "insert" a SELECT_CC instruction, we actually have to insert the
@@ -9111,7 +9510,7 @@ bool ARMTargetLowering::isDesirableToTransformToIntegerOp(unsigned Opc,
   return (VT == MVT::f32) && (Opc == ISD::LOAD || Opc == ISD::STORE);
 }
 
-bool ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT) const {
+bool ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const {
   // The AllowsUnaliged flag models the SCTLR.A setting in ARM cpus
   bool AllowsUnaligned = Subtarget->allowsUnalignedMem();
 
@@ -9120,15 +9519,27 @@ bool ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT) const {
     return false;
   case MVT::i8:
   case MVT::i16:
-  case MVT::i32:
+  case MVT::i32: {
     // Unaligned access can use (for example) LRDB, LRDH, LDR
-    return AllowsUnaligned;
+    if (AllowsUnaligned) {
+      if (Fast)
+        *Fast = Subtarget->hasV7Ops();
+      return true;
+    }
+    return false;
+  }
   case MVT::f64:
-  case MVT::v2f64:
+  case MVT::v2f64: {
     // For any little-endian targets with neon, we can support unaligned ld/st
     // of D and Q (e.g. {D0,D1}) registers by using vld1.i8/vst1.i8.
     // A big-endian target may also explictly support unaligned accesses
-    return Subtarget->hasNEON() && (AllowsUnaligned || isLittleEndian());
+    if (Subtarget->hasNEON() && (AllowsUnaligned || isLittleEndian())) {
+      if (Fast)
+        *Fast = true;
+      return true;
+    }
+    return false;
+  }
   }
 }
 
@@ -9140,33 +9551,59 @@ static bool memOpAlign(unsigned DstAlign, unsigned SrcAlign,
 
 EVT ARMTargetLowering::getOptimalMemOpType(uint64_t Size,
                                            unsigned DstAlign, unsigned SrcAlign,
-                                           bool IsZeroVal,
+                                           bool IsMemset, bool ZeroMemset,
                                            bool MemcpyStrSrc,
                                            MachineFunction &MF) const {
   const Function *F = MF.getFunction();
 
   // See if we can use NEON instructions for this...
-  if (IsZeroVal &&
-      !F->getFnAttributes().hasAttribute(Attributes::NoImplicitFloat) &&
-      Subtarget->hasNEON()) {
-    if (memOpAlign(SrcAlign, DstAlign, 16) && Size >= 16) {
-      return MVT::v4i32;
-    } else if (memOpAlign(SrcAlign, DstAlign, 8) && Size >= 8) {
-      return MVT::v2i32;
+  if ((!IsMemset || ZeroMemset) &&
+      Subtarget->hasNEON() &&
+      !F->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
+                                       Attribute::NoImplicitFloat)) {
+    bool Fast;
+    if (Size >= 16 &&
+        (memOpAlign(SrcAlign, DstAlign, 16) ||
+         (allowsUnalignedMemoryAccesses(MVT::v2f64, &Fast) && Fast))) {
+      return MVT::v2f64;
+    } else if (Size >= 8 &&
+               (memOpAlign(SrcAlign, DstAlign, 8) ||
+                (allowsUnalignedMemoryAccesses(MVT::f64, &Fast) && Fast))) {
+      return MVT::f64;
     }
   }
 
   // Lowering to i32/i16 if the size permits.
-  if (Size >= 4) {
+  if (Size >= 4)
     return MVT::i32;
-  } else if (Size >= 2) {
+  else if (Size >= 2)
     return MVT::i16;
-  }
 
   // Let the target-independent logic figure it out.
   return MVT::Other;
 }
 
+bool ARMTargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
+  if (Val.getOpcode() != ISD::LOAD)
+    return false;
+
+  EVT VT1 = Val.getValueType();
+  if (!VT1.isSimple() || !VT1.isInteger() ||
+      !VT2.isSimple() || !VT2.isInteger())
+    return false;
+
+  switch (VT1.getSimpleVT().SimpleTy) {
+  default: break;
+  case MVT::i1:
+  case MVT::i8:
+  case MVT::i16:
+    // 8-bit and 16-bit loads implicitly zero-extend to 32-bits.
+    return true;
+  }
+
+  return false;
+}
+
 static bool isLegalT1AddressImmediate(int64_t V, EVT VT) {
   if (V < 0)
     return false;
diff --git a/lib/Target/ARM/ARMISelLowering.h b/lib/Target/ARM/ARMISelLowering.h
index 4eb3b2cb5150..9ee17f0781b9 100644
--- a/lib/Target/ARM/ARMISelLowering.h
+++ b/lib/Target/ARM/ARMISelLowering.h
@@ -17,11 +17,11 @@
 
 #include "ARM.h"
 #include "ARMSubtarget.h"
-#include "llvm/Target/TargetLowering.h"
-#include "llvm/Target/TargetRegisterInfo.h"
+#include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/FastISel.h"
 #include "llvm/CodeGen/SelectionDAG.h"
-#include "llvm/CodeGen/CallingConvLower.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 #include <vector>
 
 namespace llvm {
@@ -232,7 +232,11 @@ namespace llvm {
       ATOMAND64_DAG,
       ATOMNAND64_DAG,
       ATOMSWAP64_DAG,
-      ATOMCMPXCHG64_DAG
+      ATOMCMPXCHG64_DAG,
+      ATOMMIN64_DAG,
+      ATOMUMIN64_DAG,
+      ATOMMAX64_DAG,
+      ATOMUMAX64_DAG
     };
   }
 
@@ -248,7 +252,7 @@ namespace llvm {
   public:
     explicit ARMTargetLowering(TargetMachine &TM);
 
-    virtual unsigned getJumpTableEncoding(void) const;
+    virtual unsigned getJumpTableEncoding() const;
 
     virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
 
@@ -281,15 +285,19 @@ namespace llvm {
     bool isDesirableToTransformToIntegerOp(unsigned Opc, EVT VT) const;
 
     /// allowsUnalignedMemoryAccesses - Returns true if the target allows
-    /// unaligned memory accesses. of the specified type.
-    virtual bool allowsUnalignedMemoryAccesses(EVT VT) const;
+    /// unaligned memory accesses of the specified type. Returns whether it
+    /// is "fast" by reference in the second argument.
+    virtual bool allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const;
 
     virtual EVT getOptimalMemOpType(uint64_t Size,
                                     unsigned DstAlign, unsigned SrcAlign,
-                                    bool IsZeroVal,
+                                    bool IsMemset, bool ZeroMemset,
                                     bool MemcpyStrSrc,
                                     MachineFunction &MF) const;
 
+    using TargetLowering::isZExtFree;
+    virtual bool isZExtFree(SDValue Val, EVT VT2) const;
+
     /// isLegalAddressingMode - Return true if the addressing mode represented
     /// by AM is legal for this target, for a load/store of the specified type.
     virtual bool isLegalAddressingMode(const AddrMode &AM, Type *Ty)const;
@@ -358,7 +366,7 @@ namespace llvm {
 
     /// getRegClassFor - Return the register class that should be used for the
     /// specified value type.
-    virtual const TargetRegisterClass *getRegClassFor(EVT VT) const;
+    virtual const TargetRegisterClass *getRegClassFor(MVT VT) const;
 
     /// getMaximalGlobalOffset - Returns the maximal possible offset which can
     /// be used for loads / stores from the global.
@@ -384,7 +392,7 @@ namespace llvm {
                                     unsigned Intrinsic) const;
   protected:
     std::pair<const TargetRegisterClass*, uint8_t>
-    findRepresentativeClass(EVT VT) const;
+    findRepresentativeClass(MVT VT) const;
 
   private:
     /// Subtarget - Keep a pointer to the ARMSubtarget around so that we can
@@ -495,6 +503,12 @@ namespace llvm {
                                     const SmallVectorImpl<SDValue> &OutVals,
                                     const SmallVectorImpl<ISD::InputArg> &Ins,
                                            SelectionDAG& DAG) const;
+
+    virtual bool CanLowerReturn(CallingConv::ID CallConv,
+                                MachineFunction &MF, bool isVarArg,
+                                const SmallVectorImpl<ISD::OutputArg> &Outs,
+                                LLVMContext &Context) const;
+
     virtual SDValue
       LowerReturn(SDValue Chain,
                   CallingConv::ID CallConv, bool isVarArg,
@@ -526,7 +540,9 @@ namespace llvm {
                                           unsigned Op1,
                                           unsigned Op2,
                                           bool NeedsCarry = false,
-                                          bool IsCmpxchg = false) const;
+                                          bool IsCmpxchg = false,
+                                          bool IsMinMax = false,
+                                          ARMCC::CondCodes CC = ARMCC::AL) const;
     MachineBasicBlock * EmitAtomicBinaryMinMax(MachineInstr *MI,
                                                MachineBasicBlock *BB,
                                                unsigned Size,
diff --git a/lib/Target/ARM/ARMInstrInfo.cpp b/lib/Target/ARM/ARMInstrInfo.cpp
index a0b6f249a286..80f0ec74376a 100644
--- a/lib/Target/ARM/ARMInstrInfo.cpp
+++ b/lib/Target/ARM/ARMInstrInfo.cpp
@@ -22,8 +22,8 @@
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalVariable.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCInst.h"
 using namespace llvm;
diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td
index df2e55ed5c0e..11550c5ae678 100644
--- a/lib/Target/ARM/ARMInstrInfo.td
+++ b/lib/Target/ARM/ARMInstrInfo.td
@@ -117,7 +117,7 @@ def ARMcall_nolink   : SDNode<"ARMISD::CALL_NOLINK", SDT_ARMcall,
                                SDNPVariadic]>;
 
 def ARMretflag       : SDNode<"ARMISD::RET_FLAG", SDTNone,
-                              [SDNPHasChain, SDNPOptInGlue]>;
+                              [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
 
 def ARMcmov          : SDNode<"ARMISD::CMOV", SDT_ARMCMov,
                               [SDNPInGlue]>;
@@ -239,6 +239,9 @@ def IsARM            : Predicate<"!Subtarget->isThumb()">,
 def IsIOS            : Predicate<"Subtarget->isTargetIOS()">;
 def IsNotIOS         : Predicate<"!Subtarget->isTargetIOS()">;
 def IsNaCl           : Predicate<"Subtarget->isTargetNaCl()">;
+def UseNaClTrap      : Predicate<"Subtarget->useNaClTrap()">,
+                                 AssemblerPredicate<"FeatureNaClTrap", "NaCl">;
+def DontUseNaClTrap  : Predicate<"!Subtarget->useNaClTrap()">;
 
 // FIXME: Eventually this will be just "hasV6T2Ops".
 def UseMovt          : Predicate<"Subtarget->useMovt()">;
@@ -417,6 +420,8 @@ def reglist : Operand<i32> {
   let DecoderMethod = "DecodeRegListOperand";
 }
 
+def GPRPairOp : RegisterOperand<GPRPair, "printGPRPairOperand">;
+
 def DPRRegListAsmOperand : AsmOperandClass { let Name = "DPRRegList"; }
 def dpr_reglist : Operand<i32> {
   let EncoderMethod = "getRegisterListOpValue";
@@ -1005,7 +1010,8 @@ multiclass AsI1_bin_irs<bits<4> opcod, string opc,
   let isReMaterializable = 1 in {
   def ri : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), DPFrm,
                iii, opc, "\t$Rd, $Rn, $imm",
-               [(set GPR:$Rd, (opnode GPR:$Rn, so_imm:$imm))]> {
+               [(set GPR:$Rd, (opnode GPR:$Rn, so_imm:$imm))]>,
+           Sched<[WriteALU, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> imm;
@@ -1017,7 +1023,8 @@ multiclass AsI1_bin_irs<bits<4> opcod, string opc,
   }
   def rr : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), DPFrm,
                iir, opc, "\t$Rd, $Rn, $Rm",
-               [(set GPR:$Rd, (opnode GPR:$Rn, GPR:$Rm))]> {
+               [(set GPR:$Rd, (opnode GPR:$Rn, GPR:$Rm))]>,
+           Sched<[WriteALU, ReadALU, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<4> Rm;
@@ -1032,7 +1039,8 @@ multiclass AsI1_bin_irs<bits<4> opcod, string opc,
   def rsi : AsI1<opcod, (outs GPR:$Rd),
                (ins GPR:$Rn, so_reg_imm:$shift), DPSoRegImmFrm,
                iis, opc, "\t$Rd, $Rn, $shift",
-               [(set GPR:$Rd, (opnode GPR:$Rn, so_reg_imm:$shift))]> {
+               [(set GPR:$Rd, (opnode GPR:$Rn, so_reg_imm:$shift))]>,
+            Sched<[WriteALUsi, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> shift;
@@ -1047,7 +1055,8 @@ multiclass AsI1_bin_irs<bits<4> opcod, string opc,
   def rsr : AsI1<opcod, (outs GPR:$Rd),
                (ins GPR:$Rn, so_reg_reg:$shift), DPSoRegRegFrm,
                iis, opc, "\t$Rd, $Rn, $shift",
-               [(set GPR:$Rd, (opnode GPR:$Rn, so_reg_reg:$shift))]> {
+               [(set GPR:$Rd, (opnode GPR:$Rn, so_reg_reg:$shift))]>,
+            Sched<[WriteALUsr, ReadALUsr]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> shift;
@@ -1074,7 +1083,8 @@ multiclass AsI1_rbin_irs<bits<4> opcod, string opc,
   let isReMaterializable = 1 in {
   def ri : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), DPFrm,
                iii, opc, "\t$Rd, $Rn, $imm",
-               [(set GPR:$Rd, (opnode so_imm:$imm, GPR:$Rn))]> {
+               [(set GPR:$Rd, (opnode so_imm:$imm, GPR:$Rn))]>,
+           Sched<[WriteALU, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> imm;
@@ -1086,7 +1096,8 @@ multiclass AsI1_rbin_irs<bits<4> opcod, string opc,
   }
   def rr : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), DPFrm,
                iir, opc, "\t$Rd, $Rn, $Rm",
-               [/* pattern left blank */]> {
+               [/* pattern left blank */]>,
+           Sched<[WriteALU, ReadALU, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<4> Rm;
@@ -1100,7 +1111,8 @@ multiclass AsI1_rbin_irs<bits<4> opcod, string opc,
   def rsi : AsI1<opcod, (outs GPR:$Rd),
                (ins GPR:$Rn, so_reg_imm:$shift), DPSoRegImmFrm,
                iis, opc, "\t$Rd, $Rn, $shift",
-               [(set GPR:$Rd, (opnode so_reg_imm:$shift, GPR:$Rn))]> {
+               [(set GPR:$Rd, (opnode so_reg_imm:$shift, GPR:$Rn))]>,
+            Sched<[WriteALUsi, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> shift;
@@ -1115,7 +1127,8 @@ multiclass AsI1_rbin_irs<bits<4> opcod, string opc,
   def rsr : AsI1<opcod, (outs GPR:$Rd),
                (ins GPR:$Rn, so_reg_reg:$shift), DPSoRegRegFrm,
                iis, opc, "\t$Rd, $Rn, $shift",
-               [(set GPR:$Rd, (opnode so_reg_reg:$shift, GPR:$Rn))]> {
+               [(set GPR:$Rd, (opnode so_reg_reg:$shift, GPR:$Rn))]>,
+            Sched<[WriteALUsr, ReadALUsr]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> shift;
@@ -1140,24 +1153,28 @@ multiclass AsI1_bin_s_irs<InstrItinClass iii, InstrItinClass iir,
                           bit Commutable = 0> {
   def ri : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm, pred:$p),
                          4, iii,
-                         [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, so_imm:$imm))]>;
+                         [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, so_imm:$imm))]>,
+                         Sched<[WriteALU, ReadALU]>;
 
   def rr : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, pred:$p),
                          4, iir,
-                         [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, GPR:$Rm))]> {
+                         [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, GPR:$Rm))]>,
+                         Sched<[WriteALU, ReadALU, ReadALU]> {
     let isCommutable = Commutable;
   }
   def rsi : ARMPseudoInst<(outs GPR:$Rd),
                           (ins GPR:$Rn, so_reg_imm:$shift, pred:$p),
                           4, iis,
                           [(set GPR:$Rd, CPSR, (opnode GPR:$Rn,
-                                                so_reg_imm:$shift))]>;
+                                                so_reg_imm:$shift))]>,
+                          Sched<[WriteALUsi, ReadALU]>;
 
   def rsr : ARMPseudoInst<(outs GPR:$Rd),
                           (ins GPR:$Rn, so_reg_reg:$shift, pred:$p),
                           4, iis,
                           [(set GPR:$Rd, CPSR, (opnode GPR:$Rn,
-                                                so_reg_reg:$shift))]>;
+                                                so_reg_reg:$shift))]>,
+                          Sched<[WriteALUSsr, ReadALUsr]>;
 }
 }
 
@@ -1169,19 +1186,22 @@ multiclass AsI1_rbin_s_is<InstrItinClass iii, InstrItinClass iir,
                           bit Commutable = 0> {
   def ri : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm, pred:$p),
                          4, iii,
-                         [(set GPR:$Rd, CPSR, (opnode so_imm:$imm, GPR:$Rn))]>;
+                         [(set GPR:$Rd, CPSR, (opnode so_imm:$imm, GPR:$Rn))]>,
+           Sched<[WriteALU, ReadALU]>;
 
   def rsi : ARMPseudoInst<(outs GPR:$Rd),
                           (ins GPR:$Rn, so_reg_imm:$shift, pred:$p),
                           4, iis,
                           [(set GPR:$Rd, CPSR, (opnode so_reg_imm:$shift,
-                                             GPR:$Rn))]>;
+                                             GPR:$Rn))]>,
+            Sched<[WriteALUsi, ReadALU]>;
 
   def rsr : ARMPseudoInst<(outs GPR:$Rd),
                           (ins GPR:$Rn, so_reg_reg:$shift, pred:$p),
                           4, iis,
                           [(set GPR:$Rd, CPSR, (opnode so_reg_reg:$shift,
-                                             GPR:$Rn))]>;
+                                             GPR:$Rn))]>,
+            Sched<[WriteALUSsr, ReadALUsr]>;
 }
 }
 
@@ -1194,7 +1214,8 @@ multiclass AI1_cmp_irs<bits<4> opcod, string opc,
                        PatFrag opnode, bit Commutable = 0> {
   def ri : AI1<opcod, (outs), (ins GPR:$Rn, so_imm:$imm), DPFrm, iii,
                opc, "\t$Rn, $imm",
-               [(opnode GPR:$Rn, so_imm:$imm)]> {
+               [(opnode GPR:$Rn, so_imm:$imm)]>,
+           Sched<[WriteCMP, ReadALU]> {
     bits<4> Rn;
     bits<12> imm;
     let Inst{25} = 1;
@@ -1207,7 +1228,8 @@ multiclass AI1_cmp_irs<bits<4> opcod, string opc,
   }
   def rr : AI1<opcod, (outs), (ins GPR:$Rn, GPR:$Rm), DPFrm, iir,
                opc, "\t$Rn, $Rm",
-               [(opnode GPR:$Rn, GPR:$Rm)]> {
+               [(opnode GPR:$Rn, GPR:$Rm)]>,
+           Sched<[WriteCMP, ReadALU, ReadALU]> {
     bits<4> Rn;
     bits<4> Rm;
     let isCommutable = Commutable;
@@ -1223,7 +1245,8 @@ multiclass AI1_cmp_irs<bits<4> opcod, string opc,
   def rsi : AI1<opcod, (outs),
                (ins GPR:$Rn, so_reg_imm:$shift), DPSoRegImmFrm, iis,
                opc, "\t$Rn, $shift",
-               [(opnode GPR:$Rn, so_reg_imm:$shift)]> {
+               [(opnode GPR:$Rn, so_reg_imm:$shift)]>,
+            Sched<[WriteCMPsi, ReadALU]> {
     bits<4> Rn;
     bits<12> shift;
     let Inst{25} = 0;
@@ -1239,7 +1262,8 @@ multiclass AI1_cmp_irs<bits<4> opcod, string opc,
   def rsr : AI1<opcod, (outs),
                (ins GPRnopc:$Rn, so_reg_reg:$shift), DPSoRegRegFrm, iis,
                opc, "\t$Rn, $shift",
-               [(opnode GPRnopc:$Rn, so_reg_reg:$shift)]> {
+               [(opnode GPRnopc:$Rn, so_reg_reg:$shift)]>,
+            Sched<[WriteCMPsr, ReadALU]> {
     bits<4> Rn;
     bits<12> shift;
     let Inst{25} = 0;
@@ -1321,7 +1345,8 @@ multiclass AI1_adde_sube_irs<bits<4> opcod, string opc, PatFrag opnode,
   def ri : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm),
                 DPFrm, IIC_iALUi, opc, "\t$Rd, $Rn, $imm",
                [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, so_imm:$imm, CPSR))]>,
-               Requires<[IsARM]> {
+               Requires<[IsARM]>,
+           Sched<[WriteALU, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> imm;
@@ -1333,7 +1358,8 @@ multiclass AI1_adde_sube_irs<bits<4> opcod, string opc, PatFrag opnode,
   def rr : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm),
                 DPFrm, IIC_iALUr, opc, "\t$Rd, $Rn, $Rm",
                [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, GPR:$Rm, CPSR))]>,
-               Requires<[IsARM]> {
+               Requires<[IsARM]>,
+           Sched<[WriteALU, ReadALU, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<4> Rm;
@@ -1348,7 +1374,8 @@ multiclass AI1_adde_sube_irs<bits<4> opcod, string opc, PatFrag opnode,
                 (ins GPR:$Rn, so_reg_imm:$shift),
                 DPSoRegImmFrm, IIC_iALUsr, opc, "\t$Rd, $Rn, $shift",
               [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, so_reg_imm:$shift, CPSR))]>,
-               Requires<[IsARM]> {
+               Requires<[IsARM]>,
+            Sched<[WriteALUsi, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> shift;
@@ -1364,7 +1391,8 @@ multiclass AI1_adde_sube_irs<bits<4> opcod, string opc, PatFrag opnode,
                 DPSoRegRegFrm, IIC_iALUsr, opc, "\t$Rd, $Rn, $shift",
               [(set GPRnopc:$Rd, CPSR,
                     (opnode GPRnopc:$Rn, so_reg_reg:$shift, CPSR))]>,
-               Requires<[IsARM]> {
+               Requires<[IsARM]>,
+            Sched<[WriteALUsr, ReadALUsr]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> shift;
@@ -1387,7 +1415,8 @@ multiclass AI1_rsc_irs<bits<4> opcod, string opc, PatFrag opnode> {
   def ri : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm),
                 DPFrm, IIC_iALUi, opc, "\t$Rd, $Rn, $imm",
                [(set GPR:$Rd, CPSR, (opnode so_imm:$imm, GPR:$Rn, CPSR))]>,
-               Requires<[IsARM]> {
+               Requires<[IsARM]>,
+           Sched<[WriteALU, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> imm;
@@ -1398,7 +1427,8 @@ multiclass AI1_rsc_irs<bits<4> opcod, string opc, PatFrag opnode> {
   }
   def rr : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm),
                 DPFrm, IIC_iALUr, opc, "\t$Rd, $Rn, $Rm",
-               [/* pattern left blank */]> {
+               [/* pattern left blank */]>,
+           Sched<[WriteALU, ReadALU, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<4> Rm;
@@ -1411,7 +1441,8 @@ multiclass AI1_rsc_irs<bits<4> opcod, string opc, PatFrag opnode> {
   def rsi : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_reg_imm:$shift),
                 DPSoRegImmFrm, IIC_iALUsr, opc, "\t$Rd, $Rn, $shift",
               [(set GPR:$Rd, CPSR, (opnode so_reg_imm:$shift, GPR:$Rn, CPSR))]>,
-               Requires<[IsARM]> {
+               Requires<[IsARM]>,
+            Sched<[WriteALUsi, ReadALU]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> shift;
@@ -1425,7 +1456,8 @@ multiclass AI1_rsc_irs<bits<4> opcod, string opc, PatFrag opnode> {
   def rsr : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_reg_reg:$shift),
                 DPSoRegRegFrm, IIC_iALUsr, opc, "\t$Rd, $Rn, $shift",
               [(set GPR:$Rd, CPSR, (opnode so_reg_reg:$shift, GPR:$Rn, CPSR))]>,
-               Requires<[IsARM]> {
+               Requires<[IsARM]>,
+            Sched<[WriteALUsr, ReadALUsr]> {
     bits<4> Rd;
     bits<4> Rn;
     bits<12> shift;
@@ -1622,6 +1654,18 @@ def ATOMCMPXCHG6432 : PseudoInst<(outs GPR:$dst1, GPR:$dst2),
                                  (ins GPR:$addr, GPR:$cmp1, GPR:$cmp2,
                                       GPR:$set1, GPR:$set2),
                                  NoItinerary, []>;
+def ATOMMIN6432  : PseudoInst<(outs GPR:$dst1, GPR:$dst2),
+                              (ins GPR:$addr, GPR:$src1, GPR:$src2),
+                              NoItinerary, []>;
+def ATOMUMIN6432  : PseudoInst<(outs GPR:$dst1, GPR:$dst2),
+                              (ins GPR:$addr, GPR:$src1, GPR:$src2),
+                              NoItinerary, []>;
+def ATOMMAX6432  : PseudoInst<(outs GPR:$dst1, GPR:$dst2),
+                              (ins GPR:$addr, GPR:$src1, GPR:$src2),
+                              NoItinerary, []>;
+def ATOMUMAX6432  : PseudoInst<(outs GPR:$dst1, GPR:$dst2),
+                              (ins GPR:$addr, GPR:$src1, GPR:$src2),
+                              NoItinerary, []>;
 }
 
 def HINT : AI<(outs), (ins imm0_255:$imm), MiscFrm, NoItinerary,
@@ -1748,11 +1792,32 @@ def DBG : AI<(outs), (ins imm0_15:$opt), MiscFrm, NoItinerary, "dbg", "\t$opt",
   let Inst{3-0} = opt;
 }
 
-// A5.4 Permanently UNDEFINED instructions.
+/*
+ * A5.4 Permanently UNDEFINED instructions.
+ *
+ * For most targets use UDF #65006, for which the OS will generate SIGTRAP.
+ * Other UDF encodings generate SIGILL.
+ *
+ * NaCl's OS instead chooses an ARM UDF encoding that's also a UDF in Thumb.
+ * Encoding A1:
+ *  1110 0111 1111 iiii iiii iiii 1111 iiii
+ * Encoding T1:
+ *  1101 1110 iiii iiii
+ * It uses the following encoding:
+ *  1110 0111 1111 1110 1101 1110 1111 0000
+ *  - In ARM: UDF #60896;
+ *  - In Thumb: UDF #254 followed by a branch-to-self.
+ */
+let isBarrier = 1, isTerminator = 1 in
+def TRAPNaCl : AXI<(outs), (ins), MiscFrm, NoItinerary,
+               "trap", [(trap)]>,
+           Requires<[IsARM,UseNaClTrap]> {
+  let Inst = 0xe7fedef0;
+}
 let isBarrier = 1, isTerminator = 1 in
 def TRAP : AXI<(outs), (ins), MiscFrm, NoItinerary,
                "trap", [(trap)]>,
-           Requires<[IsARM]> {
+           Requires<[IsARM,DontUseNaClTrap]> {
   let Inst = 0xe7ffdefe;
 }
 
@@ -1804,7 +1869,8 @@ let neverHasSideEffects = 1, isReMaterializable = 1 in
 // the instruction. The {24-21} opcode bits are set by the fixup, as we don't
 // know until then which form of the instruction will be used.
 def ADR : AI1<{0,?,?,0}, (outs GPR:$Rd), (ins adrlabel:$label),
-                 MiscFrm, IIC_iALUi, "adr", "\t$Rd, $label", []> {
+                 MiscFrm, IIC_iALUi, "adr", "\t$Rd, $label", []>,
+                 Sched<[WriteALU, ReadALU]> {
   bits<4> Rd;
   bits<14> label;
   let Inst{27-25} = 0b001;
@@ -2065,6 +2131,18 @@ def SRSIB_UPD : SRSI<1, "srsib\tsp!, $mode"> {
   let Inst{24-23} = 0b11;
 }
 
+def : ARMInstAlias<"srsda $mode", (SRSDA imm0_31:$mode)>;
+def : ARMInstAlias<"srsda $mode!", (SRSDA_UPD imm0_31:$mode)>;
+
+def : ARMInstAlias<"srsdb $mode", (SRSDB imm0_31:$mode)>;
+def : ARMInstAlias<"srsdb $mode!", (SRSDB_UPD imm0_31:$mode)>;
+
+def : ARMInstAlias<"srsia $mode", (SRSIA imm0_31:$mode)>;
+def : ARMInstAlias<"srsia $mode!", (SRSIA_UPD imm0_31:$mode)>;
+
+def : ARMInstAlias<"srsib $mode", (SRSIB imm0_31:$mode)>;
+def : ARMInstAlias<"srsib $mode!", (SRSIB_UPD imm0_31:$mode)>;
+
 // Return From Exception
 class RFEI<bit wb, string asm>
   : XI<(outs), (ins GPR:$Rn), AddrModeNone, 4, IndexModeNone, BrFrm,
@@ -3816,28 +3894,33 @@ def UDIV : ADivA1I<0b011, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), IIC_iDIV,
 
 def CLZ  : AMiscA1I<0b000010110, 0b0001, (outs GPR:$Rd), (ins GPR:$Rm),
               IIC_iUNAr, "clz", "\t$Rd, $Rm",
-              [(set GPR:$Rd, (ctlz GPR:$Rm))]>, Requires<[IsARM, HasV5T]>;
+              [(set GPR:$Rd, (ctlz GPR:$Rm))]>, Requires<[IsARM, HasV5T]>,
+           Sched<[WriteALU]>;
 
 def RBIT : AMiscA1I<0b01101111, 0b0011, (outs GPR:$Rd), (ins GPR:$Rm),
               IIC_iUNAr, "rbit", "\t$Rd, $Rm",
               [(set GPR:$Rd, (ARMrbit GPR:$Rm))]>,
-           Requires<[IsARM, HasV6T2]>;
+           Requires<[IsARM, HasV6T2]>,
+           Sched<[WriteALU]>;
 
 def REV  : AMiscA1I<0b01101011, 0b0011, (outs GPR:$Rd), (ins GPR:$Rm),
               IIC_iUNAr, "rev", "\t$Rd, $Rm",
-              [(set GPR:$Rd, (bswap GPR:$Rm))]>, Requires<[IsARM, HasV6]>;
+              [(set GPR:$Rd, (bswap GPR:$Rm))]>, Requires<[IsARM, HasV6]>,
+           Sched<[WriteALU]>;
 
 let AddedComplexity = 5 in
 def REV16 : AMiscA1I<0b01101011, 0b1011, (outs GPR:$Rd), (ins GPR:$Rm),
                IIC_iUNAr, "rev16", "\t$Rd, $Rm",
                [(set GPR:$Rd, (rotr (bswap GPR:$Rm), (i32 16)))]>,
-               Requires<[IsARM, HasV6]>;
+               Requires<[IsARM, HasV6]>,
+           Sched<[WriteALU]>;
 
 let AddedComplexity = 5 in
 def REVSH : AMiscA1I<0b01101111, 0b1011, (outs GPR:$Rd), (ins GPR:$Rm),
                IIC_iUNAr, "revsh", "\t$Rd, $Rm",
                [(set GPR:$Rd, (sra (bswap GPR:$Rm), (i32 16)))]>,
-               Requires<[IsARM, HasV6]>;
+               Requires<[IsARM, HasV6]>,
+           Sched<[WriteALU]>;
 
 def : ARMV6Pat<(or (sra (shl GPR:$Rm, (i32 24)), (i32 16)),
                    (and (srl GPR:$Rm, (i32 8)), 0xFF)),
@@ -3849,7 +3932,8 @@ def PKHBT : APKHI<0b01101000, 0, (outs GPRnopc:$Rd),
                [(set GPRnopc:$Rd, (or (and GPRnopc:$Rn, 0xFFFF),
                                       (and (shl GPRnopc:$Rm, pkh_lsl_amt:$sh),
                                            0xFFFF0000)))]>,
-               Requires<[IsARM, HasV6]>;
+               Requires<[IsARM, HasV6]>,
+           Sched<[WriteALUsi, ReadALU]>;
 
 // Alternate cases for PKHBT where identities eliminate some nodes.
 def : ARMV6Pat<(or (and GPRnopc:$Rn, 0xFFFF), (and GPRnopc:$Rm, 0xFFFF0000)),
@@ -3865,7 +3949,8 @@ def PKHTB : APKHI<0b01101000, 1, (outs GPRnopc:$Rd),
                [(set GPRnopc:$Rd, (or (and GPRnopc:$Rn, 0xFFFF0000),
                                       (and (sra GPRnopc:$Rm, pkh_asr_amt:$sh),
                                            0xFFFF)))]>,
-               Requires<[IsARM, HasV6]>;
+               Requires<[IsARM, HasV6]>,
+           Sched<[WriteALUsi, ReadALU]>;
 
 // Alternate cases for PKHTB where identities eliminate some nodes.  Note that
 // a shift amount of 0 is *not legal* here, it is PKHBT instead.
@@ -4229,8 +4314,8 @@ def LDREXH : AIldrex<0b11, (outs GPR:$Rt), (ins addr_offset_none:$addr),
 def LDREX  : AIldrex<0b00, (outs GPR:$Rt), (ins addr_offset_none:$addr),
                      NoItinerary, "ldrex", "\t$Rt, $addr", []>;
 let hasExtraDefRegAllocReq = 1 in
-def LDREXD: AIldrex<0b01, (outs GPR:$Rt, GPR:$Rt2),(ins addr_offset_none:$addr),
-                      NoItinerary, "ldrexd", "\t$Rt, $Rt2, $addr", []> {
+def LDREXD: AIldrex<0b01, (outs GPRPairOp:$Rt),(ins addr_offset_none:$addr),
+                      NoItinerary, "ldrexd", "\t$Rt, $addr", []> {
   let DecoderMethod = "DecodeDoubleRegLoad";
 }
 }
@@ -4244,8 +4329,8 @@ def STREX : AIstrex<0b00, (outs GPR:$Rd), (ins GPR:$Rt, addr_offset_none:$addr),
                     NoItinerary, "strex", "\t$Rd, $Rt, $addr", []>;
 let hasExtraSrcRegAllocReq = 1 in
 def STREXD : AIstrex<0b01, (outs GPR:$Rd),
-                    (ins GPR:$Rt, GPR:$Rt2, addr_offset_none:$addr),
-                    NoItinerary, "strexd", "\t$Rd, $Rt, $Rt2, $addr", []> {
+                    (ins GPRPairOp:$Rt, addr_offset_none:$addr),
+                    NoItinerary, "strexd", "\t$Rd, $Rt, $addr", []> {
   let DecoderMethod = "DecodeDoubleRegStore";
 }
 }
diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td
index 3cf213cbffee..0411ac4e282a 100644
--- a/lib/Target/ARM/ARMInstrNEON.td
+++ b/lib/Target/ARM/ARMInstrNEON.td
@@ -4264,6 +4264,7 @@ def  VCEQfd   : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
 def  VCEQfq   : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
                      NEONvceq, 1>;
 
+let TwoOperandAliasConstraint = "$Vm = $Vd" in
 defm VCEQz    : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
                             "$Vd, $Vm, #0", NEONvceqz>;
 
@@ -4277,10 +4278,12 @@ def  VCGEfd   : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
 def  VCGEfq   : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
                      NEONvcge, 0>;
 
+let TwoOperandAliasConstraint = "$Vm = $Vd" in {
 defm VCGEz    : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
                             "$Vd, $Vm, #0", NEONvcgez>;
 defm VCLEz    : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
                             "$Vd, $Vm, #0", NEONvclez>;
+}
 
 //   VCGT     : Vector Compare Greater Than
 defm VCGTs    : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
@@ -4292,10 +4295,12 @@ def  VCGTfd   : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
 def  VCGTfq   : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
                      NEONvcgt, 0>;
 
+let TwoOperandAliasConstraint = "$Vm = $Vd" in {
 defm VCGTz    : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
                             "$Vd, $Vm, #0", NEONvcgtz>;
 defm VCLTz    : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
                             "$Vd, $Vm, #0", NEONvcltz>;
+}
 
 //   VACGE    : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
 def  VACGEd   : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
@@ -4877,12 +4882,15 @@ defm VSRI     : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
 defm VABS     : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
                            IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
                            int_arm_neon_vabs>;
-def  VABSfd   : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
-                        IIC_VUNAD, "vabs", "f32",
-                        v2f32, v2f32, int_arm_neon_vabs>;
-def  VABSfq   : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
-                        IIC_VUNAQ, "vabs", "f32",
-                        v4f32, v4f32, int_arm_neon_vabs>;
+def  VABSfd   : N2VD<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
+                     "vabs", "f32",
+                     v2f32, v2f32, fabs>;
+def  VABSfq   : N2VQ<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
+                     "vabs", "f32",
+                      v4f32, v4f32, fabs>;
+
+def : Pat<(v2f32 (int_arm_neon_vabs (v2f32 DPR:$src))), (VABSfd DPR:$src)>;
+def : Pat<(v4f32 (int_arm_neon_vabs (v4f32 QPR:$src))), (VABSfq QPR:$src)>;
 
 //   VQABS    : Vector Saturating Absolute Value
 defm VQABS    : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
@@ -5737,6 +5745,10 @@ def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
 
+// Fold extracting an element out of a v2i32 into a vfp register.
+def : Pat<(f32 (bitconvert (i32 (extractelt (v2i32 DPR:$src), imm:$lane)))),
+          (f32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
+
 // Vector lengthening move with load, matching extending loads.
 
 // extload, zextload and sextload for a standard lengthening load. Example:
diff --git a/lib/Target/ARM/ARMInstrThumb2.td b/lib/Target/ARM/ARMInstrThumb2.td
index 002d64a2d039..c9d709eb5222 100644
--- a/lib/Target/ARM/ARMInstrThumb2.td
+++ b/lib/Target/ARM/ARMInstrThumb2.td
@@ -130,8 +130,9 @@ def imm0_4095_neg : Operand<i32>, PatLeaf<(i32 imm), [{
   let ParserMatchClass = imm0_4095_neg_asmoperand;
 }
 
-def imm0_255_neg : PatLeaf<(i32 imm), [{
-  return (uint32_t)(-N->getZExtValue()) < 255;
+def imm1_255_neg : PatLeaf<(i32 imm), [{
+  uint32_t Val = -N->getZExtValue();
+  return (Val > 0 && Val < 255);
 }], imm_neg_XFORM>;
 
 def imm0_255_not : PatLeaf<(i32 imm), [{
@@ -1928,8 +1929,8 @@ defm t2RSBS : T2I_rbin_s_is <BinOpFrag<(ARMsubc node:$LHS, node:$RHS)>>;
 // The AddedComplexity preferences the first variant over the others since
 // it can be shrunk to a 16-bit wide encoding, while the others cannot.
 let AddedComplexity = 1 in
-def : T2Pat<(add        GPR:$src, imm0_255_neg:$imm),
-            (t2SUBri    GPR:$src, imm0_255_neg:$imm)>;
+def : T2Pat<(add        GPR:$src, imm1_255_neg:$imm),
+            (t2SUBri    GPR:$src, imm1_255_neg:$imm)>;
 def : T2Pat<(add        GPR:$src, t2_so_imm_neg:$imm),
             (t2SUBri    GPR:$src, t2_so_imm_neg:$imm)>;
 def : T2Pat<(add        GPR:$src, imm0_4095_neg:$imm),
@@ -1938,8 +1939,8 @@ def : T2Pat<(add        GPR:$src, imm0_65535_neg:$imm),
             (t2SUBrr    GPR:$src, (t2MOVi16 (imm_neg_XFORM imm:$imm)))>;
 
 let AddedComplexity = 1 in
-def : T2Pat<(ARMaddc    rGPR:$src, imm0_255_neg:$imm),
-            (t2SUBSri   rGPR:$src, imm0_255_neg:$imm)>;
+def : T2Pat<(ARMaddc    rGPR:$src, imm1_255_neg:$imm),
+            (t2SUBSri   rGPR:$src, imm1_255_neg:$imm)>;
 def : T2Pat<(ARMaddc    rGPR:$src, t2_so_imm_neg:$imm),
             (t2SUBSri   rGPR:$src, t2_so_imm_neg:$imm)>;
 def : T2Pat<(ARMaddc    rGPR:$src, imm0_65535_neg:$imm),
@@ -2314,13 +2315,15 @@ defm t2ORN  : T2I_bin_irs<0b0011, "orn",
 /// changed to modify CPSR.
 multiclass T2I_un_irs<bits<4> opcod, string opc,
                      InstrItinClass iii, InstrItinClass iir, InstrItinClass iis,
-                      PatFrag opnode, bit Cheap = 0, bit ReMat = 0> {
+                      PatFrag opnode,
+                      bit Cheap = 0, bit ReMat = 0, bit MoveImm = 0> {
    // shifted imm
    def i : T2sOneRegImm<(outs rGPR:$Rd), (ins t2_so_imm:$imm), iii,
                 opc, "\t$Rd, $imm",
                 [(set rGPR:$Rd, (opnode t2_so_imm:$imm))]> {
      let isAsCheapAsAMove = Cheap;
      let isReMaterializable = ReMat;
+     let isMoveImm = MoveImm;
      let Inst{31-27} = 0b11110;
      let Inst{25} = 0;
      let Inst{24-21} = opcod;
@@ -2354,7 +2357,7 @@ multiclass T2I_un_irs<bits<4> opcod, string opc,
 let AddedComplexity = 1 in
 defm t2MVN  : T2I_un_irs <0b0011, "mvn",
                           IIC_iMVNi, IIC_iMVNr, IIC_iMVNsi,
-                          UnOpFrag<(not node:$Src)>, 1, 1>;
+                          UnOpFrag<(not node:$Src)>, 1, 1, 1>;
 
 let AddedComplexity = 1 in
 def : T2Pat<(and     rGPR:$src, t2_so_imm_not:$imm),
@@ -3478,6 +3481,13 @@ def t2SRSIA_UPD : T2SRS<0b11, 1, (outs), (ins imm0_31:$mode), NoItinerary,
 def t2SRSIA  : T2SRS<0b11, 0, (outs), (ins imm0_31:$mode), NoItinerary,
                      "srsia","\tsp, $mode", []>;
 
+
+def : t2InstAlias<"srsdb${p} $mode", (t2SRSDB imm0_31:$mode, pred:$p)>;
+def : t2InstAlias<"srsdb${p} $mode!", (t2SRSDB_UPD imm0_31:$mode, pred:$p)>;
+
+def : t2InstAlias<"srsia${p} $mode", (t2SRSIA imm0_31:$mode, pred:$p)>;
+def : t2InstAlias<"srsia${p} $mode!", (t2SRSIA_UPD imm0_31:$mode, pred:$p)>;
+
 // Return From Exception is a system instruction.
 class T2RFE<bits<12> op31_20, dag oops, dag iops, InstrItinClass itin,
           string opc, string asm, list<dag> pattern>
diff --git a/lib/Target/ARM/ARMJITInfo.cpp b/lib/Target/ARM/ARMJITInfo.cpp
index 254d8f6b7c7a..351a290e2aa0 100644
--- a/lib/Target/ARM/ARMJITInfo.cpp
+++ b/lib/Target/ARM/ARMJITInfo.cpp
@@ -17,12 +17,12 @@
 #include "ARMConstantPoolValue.h"
 #include "ARMRelocations.h"
 #include "ARMSubtarget.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/JITCodeEmitter.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/raw_ostream.h"
 #include "llvm/Support/Memory.h"
+#include "llvm/Support/raw_ostream.h"
 #include <cstdlib>
 using namespace llvm;
 
diff --git a/lib/Target/ARM/ARMJITInfo.h b/lib/Target/ARM/ARMJITInfo.h
index 792818442724..23a6a9b512f4 100644
--- a/lib/Target/ARM/ARMJITInfo.h
+++ b/lib/Target/ARM/ARMJITInfo.h
@@ -15,12 +15,12 @@
 #define ARMJITINFO_H
 
 #include "ARMMachineFunctionInfo.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallVector.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
 #include "llvm/Target/TargetJITInfo.h"
-#include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/SmallVector.h"
 
 namespace llvm {
   class ARMTargetMachine;
diff --git a/lib/Target/ARM/ARMLoadStoreOptimizer.cpp b/lib/Target/ARM/ARMLoadStoreOptimizer.cpp
index 0185289f3bd8..b7ac5d57c362 100644
--- a/lib/Target/ARM/ARMLoadStoreOptimizer.cpp
+++ b/lib/Target/ARM/ARMLoadStoreOptimizer.cpp
@@ -18,8 +18,12 @@
 #include "ARMBaseRegisterInfo.h"
 #include "ARMMachineFunctionInfo.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
@@ -27,19 +31,15 @@
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
 #include "llvm/CodeGen/SelectionDAGNodes.h"
-#include "llvm/DataLayout.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/STLExtras.h"
-#include "llvm/ADT/SmallPtrSet.h"
-#include "llvm/ADT/SmallSet.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/Statistic.h"
 using namespace llvm;
 
 STATISTIC(NumLDMGened , "Number of ldm instructions generated");
@@ -87,6 +87,53 @@ namespace {
                       MachineBasicBlock::iterator i)
         : Offset(o), Reg(r), isKill(k), Position(p), MBBI(i), Merged(false) {}
     };
+    class UnitRegsMap {
+    public:
+      UnitRegsMap(const TargetRegisterInfo* _TRI) : TRI(_TRI) {}
+      const SmallVector<unsigned, 4>& operator[](unsigned Reg) {
+        DenseMap<unsigned, SmallVector<unsigned, 4> >::iterator found =
+            Cache.find(Reg);
+        if (found != Cache.end())
+          return found->second;
+        else
+          return Cache.insert(std::make_pair(Reg, this->getUnitRegs(Reg)))
+                      .first->second;
+      }
+    private:
+      SmallVector<unsigned, 4> getUnitRegs(unsigned Reg) {
+        SmallVector<unsigned, 4> Res;
+
+        const TargetRegisterClass* TRC = TRI->getMinimalPhysRegClass(Reg);
+        if (TRC == &ARM::QPRRegClass) {
+          if (Reg > ARM::Q7) {
+            Res.push_back(TRI->getSubReg(Reg, ARM::dsub_0));
+            Res.push_back(TRI->getSubReg(Reg, ARM::dsub_1));
+            return Res;
+          }
+
+          Res.push_back(TRI->getSubReg(Reg, ARM::ssub_0));
+          Res.push_back(TRI->getSubReg(Reg, ARM::ssub_1));
+          Res.push_back(TRI->getSubReg(Reg, ARM::ssub_2));
+          Res.push_back(TRI->getSubReg(Reg, ARM::ssub_3));
+
+          return Res;
+        }
+
+        if (TRC == &ARM::DPRRegClass && Reg < ARM::D15) {
+          Res.push_back(TRI->getSubReg(Reg, ARM::ssub_0));
+          Res.push_back(TRI->getSubReg(Reg, ARM::ssub_1));
+
+          return Res;
+        }
+
+        Res.push_back(Reg);
+
+        return Res;
+
+      }
+      const TargetRegisterInfo* TRI;
+      DenseMap<unsigned, SmallVector<unsigned, 4> > Cache;
+    };
     typedef SmallVector<MemOpQueueEntry,8> MemOpQueue;
     typedef MemOpQueue::iterator MemOpQueueIter;
 
@@ -128,6 +175,11 @@ namespace {
                                    MachineBasicBlock::iterator MBBI,
                                    bool &Advance,
                                    MachineBasicBlock::iterator &I);
+    unsigned AddMemOp(MemOpQueue& MemOps,
+                      const MemOpQueueEntry newEntry,
+                      UnitRegsMap& UnitRegsInfo,
+                      SmallSet<unsigned, 4>& UsedUnitRegs,
+                      unsigned At = -1U);
     bool LoadStoreMultipleOpti(MachineBasicBlock &MBB);
     bool MergeReturnIntoLDM(MachineBasicBlock &MBB);
   };
@@ -865,7 +917,7 @@ bool ARMLoadStoreOpt::MergeBaseUpdateLoadStore(MachineBasicBlock &MBB,
   bool isLd = isi32Load(Opcode) || Opcode == ARM::VLDRS || Opcode == ARM::VLDRD;
   // Can't do the merge if the destination register is the same as the would-be
   // writeback register.
-  if (isLd && MI->getOperand(0).getReg() == Base)
+  if (MI->getOperand(0).getReg() == Base)
     return false;
 
   unsigned PredReg = 0;
@@ -1188,7 +1240,6 @@ bool ARMLoadStoreOpt::FixInvalidRegPairOp(MachineBasicBlock &MBB,
           OddDeadKill = true;
         }
         // Never kill the base register in the first instruction.
-        // <rdar://problem/11101911>
         if (EvenReg == BaseReg)
           EvenDeadKill = false;
         InsertLDR_STR(MBB, MBBI, OffImm, isLd, dl, NewOpc,
@@ -1214,12 +1265,103 @@ bool ARMLoadStoreOpt::FixInvalidRegPairOp(MachineBasicBlock &MBB,
   return false;
 }
 
+/// AddMemOp - helper for ARMLoadStoreOpt::LoadStoreMultipleOpti.
+/// It adds store mem ops with simple push_back/insert method,
+/// without any additional logic.
+/// For load operation it does the next:
+/// 1. Adds new load operation into MemOp collection at "At" position.
+/// 2. Removes any "load" operations from MemOps, that changes "Reg" register
+/// contents, prior to "At".
+/// UnitRegsInfo - Map of type Map< Register, UnitRegisters-vector >
+/// UsedUnitRegs - set of unit-registers currently in use.
+/// At - position at which it would added, and prior which the clean-up
+/// should be made (for load operation).
+/// FIXME: The clean-up also should be made for store operations,
+/// but the memory address should be analyzed instead of unit registers.
+unsigned ARMLoadStoreOpt::AddMemOp(MemOpQueue& MemOps,
+                                   const MemOpQueueEntry NewEntry,
+                                   UnitRegsMap& UnitRegsInfo,
+                                   SmallSet<unsigned, 4>& UsedUnitRegs,
+                                   unsigned At) {
+  unsigned Cleaned = 0;
+
+  if (At == -1U) {
+    At = MemOps.size();
+    MemOps.push_back(NewEntry);
+  } else
+    MemOps.insert(&MemOps[At], NewEntry);
+
+  // FIXME:
+  // If operation is not load, leave it as is by now,
+  // So 0 overridden ops would cleaned in this case.
+  if (!NewEntry.MBBI->mayLoad())
+    return 0;
+
+  const SmallVector<unsigned, 4>& NewEntryUnitRegs = UnitRegsInfo[NewEntry.Reg];
+
+  bool FoundOverriddenLoads = false;
+
+  for (unsigned i = 0, e = NewEntryUnitRegs.size(); i != e; ++i)
+    if (UsedUnitRegs.count(NewEntryUnitRegs[i])) {
+      FoundOverriddenLoads = true;
+      break;
+    }
+
+  // If we detect that this register is used by load operations that are
+  // predecessors for the new one, remove them from MemOps then.
+  if (FoundOverriddenLoads) {
+    MemOpQueue UpdatedMemOps;
+
+    // Scan through MemOps entries.
+    for (unsigned i = 0; i != At; ++i) {
+      MemOpQueueEntry& MemOpEntry = MemOps[i];
+
+      // FIXME: Skip non-load operations by now.
+      if (!MemOpEntry.MBBI->mayLoad())
+        continue;
+
+      const SmallVector<unsigned, 4>& MemOpUnitRegs =
+          UnitRegsInfo[MemOpEntry.Reg];
+
+      // Lookup entry that loads contents into register used by new entry.
+      bool ReleaseThisEntry = false;
+      for (unsigned m = 0, em = MemOpUnitRegs.size(); m != em; ++m) {
+        if (std::find(NewEntryUnitRegs.begin(), NewEntryUnitRegs.end(),
+                      MemOpUnitRegs[m]) != NewEntryUnitRegs.end()) {
+          ReleaseThisEntry = true;
+          ++Cleaned;
+          break;
+        }
+      }
+
+      if (ReleaseThisEntry) {
+        const SmallVector<unsigned, 4>& RelesedRegs = UnitRegsInfo[MemOpEntry.Reg];
+        for (unsigned r = 0, er = RelesedRegs.size(); r != er; ++r)
+          UsedUnitRegs.erase(RelesedRegs[r]);
+      } else
+        UpdatedMemOps.push_back(MemOpEntry);
+    }
+
+    // Keep anything without changes after At position.
+    for (unsigned i = At, e = MemOps.size(); i != e; ++i)
+      UpdatedMemOps.push_back(MemOps[i]);
+
+    MemOps.swap(UpdatedMemOps);
+  }
+
+  UsedUnitRegs.insert(NewEntryUnitRegs.begin(), NewEntryUnitRegs.end());
+
+  return Cleaned;
+}
+
 /// LoadStoreMultipleOpti - An optimization pass to turn multiple LDR / STR
 /// ops of the same base and incrementing offset into LDM / STM ops.
 bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) {
   unsigned NumMerges = 0;
   unsigned NumMemOps = 0;
   MemOpQueue MemOps;
+  UnitRegsMap UnitRegsInfo(TRI);
+  SmallSet<unsigned, 4> UsedRegUnits;
   unsigned CurrBase = 0;
   int CurrOpc = -1;
   unsigned CurrSize = 0;
@@ -1266,8 +1408,11 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) {
         CurrSize = Size;
         CurrPred = Pred;
         CurrPredReg = PredReg;
+
         MemOps.push_back(MemOpQueueEntry(Offset, Reg, isKill, Position, MBBI));
         ++NumMemOps;
+        const SmallVector<unsigned, 4>& EntryUnitRegs = UnitRegsInfo[Reg];
+        UsedRegUnits.insert(EntryUnitRegs.begin(), EntryUnitRegs.end());
         Advance = true;
       } else {
         if (Clobber) {
@@ -1279,20 +1424,24 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) {
           // No need to match PredReg.
           // Continue adding to the queue.
           if (Offset > MemOps.back().Offset) {
-            MemOps.push_back(MemOpQueueEntry(Offset, Reg, isKill,
-                                             Position, MBBI));
-            ++NumMemOps;
+            unsigned OverridesCleaned =
+              AddMemOp(MemOps,
+                           MemOpQueueEntry(Offset, Reg, isKill, Position, MBBI),
+                           UnitRegsInfo, UsedRegUnits) != 0;
+            NumMemOps += 1 - OverridesCleaned;
             Advance = true;
           } else {
-            for (MemOpQueueIter I = MemOps.begin(), E = MemOps.end();
-                 I != E; ++I) {
-              if (Offset < I->Offset) {
-                MemOps.insert(I, MemOpQueueEntry(Offset, Reg, isKill,
-                                                 Position, MBBI));
-                ++NumMemOps;
+            for (unsigned I = 0; I != NumMemOps; ++I) {
+              if (Offset < MemOps[I].Offset) {
+                MemOpQueueEntry entry(Offset, Reg, isKill, Position, MBBI);
+                unsigned OverridesCleaned =
+                    AddMemOp(MemOps, entry, UnitRegsInfo,
+                                 UsedRegUnits, I) != 0;
+                NumMemOps += 1 - OverridesCleaned;
+
                 Advance = true;
                 break;
-              } else if (Offset == I->Offset) {
+              } else if (Offset == MemOps[I].Offset) {
                 // Collision! This can't be merged!
                 break;
               }
@@ -1363,6 +1512,7 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) {
       CurrPredReg = 0;
       if (NumMemOps) {
         MemOps.clear();
+        UsedRegUnits.clear();
         NumMemOps = 0;
       }
 
@@ -1408,7 +1558,7 @@ bool ARMLoadStoreOpt::MergeReturnIntoLDM(MachineBasicBlock &MBB) {
               Opcode == ARM::LDMIA_UPD) && "Unsupported multiple load-return!");
       PrevMI->setDesc(TII->get(NewOpc));
       MO.setReg(ARM::PC);
-      PrevMI->copyImplicitOps(&*MBBI);
+      PrevMI->copyImplicitOps(*MBB.getParent(), &*MBBI);
       MBB.erase(MBBI);
       return true;
     }
diff --git a/lib/Target/ARM/ARMMCInstLower.cpp b/lib/Target/ARM/ARMMCInstLower.cpp
index e2ac9a466ed8..b6414832003d 100644
--- a/lib/Target/ARM/ARMMCInstLower.cpp
+++ b/lib/Target/ARM/ARMMCInstLower.cpp
@@ -15,8 +15,8 @@
 #include "ARM.h"
 #include "ARMAsmPrinter.h"
 #include "MCTargetDesc/ARMMCExpr.h"
-#include "llvm/Constants.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/IR/Constants.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/Target/Mangler.h"
diff --git a/lib/Target/ARM/ARMMachineFunctionInfo.h b/lib/Target/ARM/ARMMachineFunctionInfo.h
index c0ac04b6003c..88d96c0be8a7 100644
--- a/lib/Target/ARM/ARMMachineFunctionInfo.h
+++ b/lib/Target/ARM/ARMMachineFunctionInfo.h
@@ -15,10 +15,10 @@
 #define ARMMACHINEFUNCTIONINFO_H
 
 #include "ARMSubtarget.h"
+#include "llvm/ADT/BitVector.h"
 #include "llvm/CodeGen/MachineFunction.h"
-#include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Target/TargetMachine.h"
-#include "llvm/ADT/BitVector.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 
 namespace llvm {
 
diff --git a/lib/Target/ARM/ARMSchedule.td b/lib/Target/ARM/ARMSchedule.td
index 02196d06bfd3..2d088de96e27 100644
--- a/lib/Target/ARM/ARMSchedule.td
+++ b/lib/Target/ARM/ARMSchedule.td
@@ -6,6 +6,77 @@
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+// Instruction scheduling annotations for out-of-order CPUs.
+// These annotations are independent of the itinerary class defined below.
+// Here we define the subtarget independent read/write per-operand resources.
+// The subtarget schedule definitions will then map these to the subtarget's
+// resource usages.
+// For example:
+// The instruction cycle timings table might contain an entry for an operation
+// like the following:
+// Rd <- ADD Rn, Rm, <shift> Rs
+//  Uops | Latency from register | Uops - resource requirements - latency
+//  2    | Rn: 1 Rm: 4 Rs: 4     | uop T0, Rm, Rs - P01 - 3
+//       |                       | uopc Rd, Rn, T0 -  P01 - 1
+// This is telling us that the result will be available in destination register
+// Rd after a minimum of three cycles after the result in Rm and Rs is available
+// and one cycle after the result in Rn is available. The micro-ops can execute
+// on resource P01.
+// To model this, we need to express that we need to dispatch two micro-ops,
+// that the resource P01 is needed and that the latency to Rn is different than
+// the latency to Rm and Rs. The scheduler can decrease Rn's producer latency by
+// two.
+// We will do this by assigning (abstract) resources to register defs/uses.
+// ARMSchedule.td:
+//   def WriteALUsr : SchedWrite;
+//   def ReadAdvanceALUsr : ScheRead;
+//
+// ARMInstrInfo.td:
+//   def ADDrs : I<>, Sched<[WriteALUsr, ReadAdvanceALUsr, ReadDefault,
+//                           ReadDefault]> { ...}
+// ReadAdvance read resources allow us to define "pipeline by-passes" or
+// shorter latencies to certain registers as needed in the example above.
+// The "ReadDefault" can be omitted.
+// Next, the subtarget td file assigns resources to the abstract resources
+// defined here.
+// ARMScheduleSubtarget.td:
+//  // Resources.
+//  def P01 : ProcResource<3>; // ALU unit (3 of it).
+//  ...
+//  // Resource usages.
+//  def : WriteRes<WriteALUsr, [P01, P01]> {
+//    Latency = 4; // Latency of 4.
+//    NumMicroOps = 2; // Dispatch 2 micro-ops.
+//    // The two instances of resource P01 are occupied for one cycle. It is one
+//    // cycle because these resources happen to be pipelined.
+//    ResourceCycles = [1, 1];
+//  }
+//  def : ReadAdvance<ReadAdvanceALUsr, 3>;
+
+// Basic ALU operation.
+def WriteALU : SchedWrite;
+def ReadALU : SchedRead;
+
+// Basic ALU with shifts.
+def WriteALUsi : SchedWrite; // Shift by immediate.
+def WriteALUsr : SchedWrite; // Shift by register.
+def WriteALUSsr : SchedWrite; // Shift by register (flag setting).
+def ReadALUsr : SchedRead; // Some operands are read later.
+
+// Compares.
+def WriteCMP : SchedWrite;
+def WriteCMPsi : SchedWrite;
+def WriteCMPsr : SchedWrite;
+
+// Define TII for use in SchedVariant Predicates.
+def : PredicateProlog<[{
+  const ARMBaseInstrInfo *TII =
+    static_cast<const ARMBaseInstrInfo*>(SchedModel->getInstrInfo());
+  (void)TII;
+}]>;
+
+def IsPredicatedPred : SchedPredicate<[{TII->isPredicated(MI)}]>;
 
 //===----------------------------------------------------------------------===//
 // Instruction Itinerary classes used for ARM
diff --git a/lib/Target/ARM/ARMScheduleA9.td b/lib/Target/ARM/ARMScheduleA9.td
index 404634fee989..9739ed20ce2e 100644
--- a/lib/Target/ARM/ARMScheduleA9.td
+++ b/lib/Target/ARM/ARMScheduleA9.td
@@ -1887,6 +1887,9 @@ def CortexA9Model : SchedMachineModel {
   let LoadLatency = 2; // Optimistic load latency assuming bypass.
                        // This is overriden by OperandCycles if the
                        // Itineraries are queried instead.
+  let ILPWindow = 10; // Don't reschedule small blocks to hide
+                      // latency. Minimum latency requirements are already
+                      // modeled strictly by reserving resources.
   let MispredictPenalty = 8; // Based on estimate of pipeline depth.
 
   let Itineraries = CortexA9Itineraries;
@@ -1895,6 +1898,8 @@ def CortexA9Model : SchedMachineModel {
 //===----------------------------------------------------------------------===//
 // Define each kind of processor resource and number available.
 
+let SchedModel = CortexA9Model in {
+
 def A9UnitALU : ProcResource<2>;
 def A9UnitMul : ProcResource<1> { let Super = A9UnitALU; }
 def A9UnitAGU : ProcResource<1>;
@@ -1915,11 +1920,11 @@ def A9WriteI : SchedWriteRes<[A9UnitALU]>;
 def A9WriteIsr : SchedWriteRes<[A9UnitALU]> { let Latency = 2; }
 
 // Basic ALU.
-def A9WriteA : SchedWriteRes<[A9UnitALU]>;
+def A9WriteALU : SchedWriteRes<[A9UnitALU]>;
 // ALU with operand shifted by immediate.
-def A9WriteAsi : SchedWriteRes<[A9UnitALU]> { let Latency = 2; }
+def : WriteRes<WriteALUsi, [A9UnitALU]> { let Latency = 2; }
 // ALU with operand shifted by register.
-def A9WriteAsr : SchedWriteRes<[A9UnitALU]> { let Latency = 3; }
+def A9WriteALUsr : SchedWriteRes<[A9UnitALU]> { let Latency = 3; }
 
 // Multiplication
 def A9WriteM   : SchedWriteRes<[A9UnitMul, A9UnitMul]> { let Latency = 4; }
@@ -2000,13 +2005,6 @@ foreach NumCycles = 2-8 in {
 def A9WriteCycle#NumCycles : WriteSequence<[A9WriteCycle1], NumCycles>;
 } // foreach NumCycles
 
-// Define TII for use in SchedVariant Predicates.
-def : PredicateProlog<[{
-  const ARMBaseInstrInfo *TII =
-    static_cast<const ARMBaseInstrInfo*>(SchedModel->getInstrInfo());
-  (void)TII;
-}]>;
-
 // Define address generation sequences and predicates for 8 flavors of LDMs.
 foreach NumAddr = 1-8 in {
 
@@ -2251,11 +2249,11 @@ def A9WriteLMfp : SchedWriteVariant<[
 // These mov immediate writers are unconditionally expanded with
 // additive latency.
 def A9WriteI2 : WriteSequence<[A9WriteI, A9WriteI]>;
-def A9WriteI2pc : WriteSequence<[A9WriteI, A9WriteI, A9WriteA]>;
+def A9WriteI2pc : WriteSequence<[A9WriteI, A9WriteI, WriteALU]>;
 def A9WriteI2ld  : WriteSequence<[A9WriteI, A9WriteI, A9WriteL]>;
 
 // Some ALU operations can read loaded integer values one cycle early.
-def A9ReadA : SchedReadAdvance<1,
+def A9ReadALU : SchedReadAdvance<1,
   [A9WriteL, A9WriteLHi, A9WriteLsi, A9WriteLb, A9WriteLbsi,
    A9WriteL1, A9WriteL2, A9WriteL3, A9WriteL4,
    A9WriteL5, A9WriteL6, A9WriteL7, A9WriteL8,
@@ -2276,26 +2274,25 @@ def A9Read4 : SchedReadAdvance<3>;
 
 // This table follows the ARM Cortex-A9 Technical Reference Manuals,
 // mostly in order.
-let SchedModel = CortexA9Model in {
 
 def :ItinRW<[A9WriteI], [IIC_iMOVi,IIC_iMOVr,IIC_iMOVsi,
                          IIC_iMVNi,IIC_iMVNsi,
                          IIC_iCMOVi,IIC_iCMOVr,IIC_iCMOVsi]>;
-def :ItinRW<[A9WriteI,A9ReadA],[IIC_iMVNr]>;
+def :ItinRW<[A9WriteI,A9ReadALU],[IIC_iMVNr]>;
 def :ItinRW<[A9WriteIsr], [IIC_iMOVsr,IIC_iMVNsr,IIC_iCMOVsr]>;
 
 def :ItinRW<[A9WriteI2],   [IIC_iMOVix2,IIC_iCMOVix2]>;
 def :ItinRW<[A9WriteI2pc], [IIC_iMOVix2addpc]>;
 def :ItinRW<[A9WriteI2ld], [IIC_iMOVix2ld]>;
 
-def :ItinRW<[A9WriteA], [IIC_iBITi,IIC_iBITr,IIC_iUNAr,IIC_iTSTi,IIC_iTSTr]>;
-def :ItinRW<[A9WriteA, A9ReadA], [IIC_iALUi, IIC_iCMPi, IIC_iCMPsi]>;
-def :ItinRW<[A9WriteA, A9ReadA, A9ReadA],[IIC_iALUr,IIC_iCMPr]>;
-def :ItinRW<[A9WriteAsi], [IIC_iBITsi,IIC_iUNAsi,IIC_iEXTr,IIC_iTSTsi]>;
-def :ItinRW<[A9WriteAsi, A9ReadA], [IIC_iALUsi]>;
-def :ItinRW<[A9WriteAsi, ReadDefault, A9ReadA], [IIC_iALUsir]>; // RSB
-def :ItinRW<[A9WriteAsr], [IIC_iBITsr,IIC_iTSTsr,IIC_iEXTAr,IIC_iEXTAsr]>;
-def :ItinRW<[A9WriteAsr, A9ReadA], [IIC_iALUsr,IIC_iCMPsr]>;
+def :ItinRW<[WriteALU], [IIC_iBITi,IIC_iBITr,IIC_iUNAr,IIC_iTSTi,IIC_iTSTr]>;
+def :ItinRW<[WriteALU, A9ReadALU], [IIC_iALUi, IIC_iCMPi, IIC_iCMPsi]>;
+def :ItinRW<[WriteALU, A9ReadALU, A9ReadALU],[IIC_iALUr,IIC_iCMPr]>;
+def :ItinRW<[WriteALUsi], [IIC_iBITsi,IIC_iUNAsi,IIC_iEXTr,IIC_iTSTsi]>;
+def :ItinRW<[WriteALUsi, A9ReadALU], [IIC_iALUsi]>;
+def :ItinRW<[WriteALUsi, ReadDefault, A9ReadALU], [IIC_iALUsir]>; // RSB
+def :ItinRW<[A9WriteALUsr], [IIC_iBITsr,IIC_iTSTsr,IIC_iEXTAr,IIC_iEXTAsr]>;
+def :ItinRW<[A9WriteALUsr, A9ReadALU], [IIC_iALUsr,IIC_iCMPsr]>;
 
 // A9WriteHi ignored for MUL32.
 def :ItinRW<[A9WriteM, A9WriteMHi], [IIC_iMUL32,IIC_iMAC32,
@@ -2368,7 +2365,7 @@ def :ItinRW<[A9WriteLMAdr, A9WriteLM, A9WriteIssue], [IIC_iLoad_mu,
                                                       IIC_iStore_m,
                                                       IIC_iStore_mu]>;
 def :ItinRW<[A9WriteLM, A9WriteLMAdr, A9WriteB], [IIC_iLoad_mBr, IIC_iPop_Br]>;
-def :ItinRW<[A9WriteL, A9WriteAdr, A9WriteA], [IIC_iLoadiALU]>;
+def :ItinRW<[A9WriteL, A9WriteAdr, WriteALU], [IIC_iLoadiALU]>;
 
 def :ItinRW<[A9WriteLSfp, A9WriteAdr], [IIC_fpLoad32, IIC_fpLoad64]>;
 
@@ -2483,4 +2480,17 @@ def :ItinRW<[A9WriteV9, A9Read3, A9Read2], [IIC_VMACD, IIC_VFMACD]>;
 def :ItinRW<[A9WriteV10, A9Read3, A9Read2], [IIC_VMACQ, IIC_VFMACQ]>;
 def :ItinRW<[A9WriteV9, A9Read2, A9Read2], [IIC_VRECSD]>;
 def :ItinRW<[A9WriteV10, A9Read2, A9Read2], [IIC_VRECSQ]>;
+
+// Map SchedRWs that are identical for cortexa9 to existing resources.
+def : SchedAlias<WriteALU, A9WriteALU>;
+def : SchedAlias<WriteALUsr, A9WriteALUsr>;
+def : SchedAlias<WriteALUSsr, A9WriteALUsr>;
+def : SchedAlias<ReadALU, A9ReadALU>;
+def : SchedAlias<ReadALUsr, A9ReadALU>;
+// FIXME: need to special case AND, ORR, EOR, BIC because they don't read
+// advance. But our instrinfo claims it does.
+
+def : SchedAlias<WriteCMP, A9WriteALU>;
+def : SchedAlias<WriteCMPsi, A9WriteALU>;
+def : SchedAlias<WriteCMPsr, A9WriteALU>;
 } // SchedModel = CortexA9Model
diff --git a/lib/Target/ARM/ARMScheduleSwift.td b/lib/Target/ARM/ARMScheduleSwift.td
index e9bc3e0f3955..7c6df410706e 100644
--- a/lib/Target/ARM/ARMScheduleSwift.td
+++ b/lib/Target/ARM/ARMScheduleSwift.td
@@ -1078,8 +1078,67 @@ def SwiftModel : SchedMachineModel {
   let IssueWidth = 3; // 3 micro-ops are dispatched per cycle.
   let MinLatency = 0; // Data dependencies are allowed within dispatch groups.
   let LoadLatency = 3;
+  let MispredictPenalty = 14; // A branch direction mispredict.
 
   let Itineraries = SwiftItineraries;
 }
 
-// TODO: Add Swift processor and scheduler resources.
+// Swift predicates.
+def IsFastImmShiftSwiftPred : SchedPredicate<[{TII->isSwiftFastImmShift(MI)}]>;
+
+// Swift resource mapping.
+let SchedModel = SwiftModel in {
+  // Processor resources.
+  def SwiftUnitP01 : ProcResource<2>; // ALU unit.
+  def SwiftUnitP0 : ProcResource<1> { let Super = SwiftUnitP01; } // Mul unit.
+  def SwiftUnitP1 : ProcResource<1> { let Super = SwiftUnitP01; } // Br unit.
+  def SwiftUnitP2 : ProcResource<1>; // LS unit.
+  def SwiftUnitDiv : ProcResource<1>;
+
+  // Generic resource requirements.
+  def SwiftWriteP01TwoCycle : SchedWriteRes<[SwiftUnitP01]> { let Latency = 2; }
+  def SwiftWriteP01ThreeCycleTwoUops :
+    SchedWriteRes<[SwiftUnitP01, SwiftUnitP01]> {
+    let Latency = 3;
+    let NumMicroOps = 2;
+  }
+  def SwiftWriteP0ThreeCycleThreeUops : SchedWriteRes<[SwiftUnitP0]> {
+    let Latency = 3;
+    let NumMicroOps = 3;
+    let ResourceCycles = [3];
+  }
+
+  // 4.2.4 Arithmetic and Logical.
+  // ALU operation register shifted by immediate variant.
+  def SwiftWriteALUsi : SchedWriteVariant<[
+    // lsl #2, lsl #1, or lsr #1.
+    SchedVar<IsFastImmShiftSwiftPred, [SwiftWriteP01TwoCycle]>,
+    SchedVar<NoSchedPred,             [WriteALU]>
+  ]>;
+  def SwiftWriteALUsr : SchedWriteVariant<[
+    SchedVar<IsPredicatedPred, [SwiftWriteP01ThreeCycleTwoUops]>,
+    SchedVar<NoSchedPred,      [SwiftWriteP01TwoCycle]>
+  ]>;
+  def SwiftWriteALUSsr : SchedWriteVariant<[
+    SchedVar<IsPredicatedPred, [SwiftWriteP0ThreeCycleThreeUops]>,
+    SchedVar<NoSchedPred,      [SwiftWriteP01TwoCycle]>
+  ]>;
+  def SwiftReadAdvanceALUsr : SchedReadVariant<[
+    SchedVar<IsPredicatedPred, [SchedReadAdvance<2>]>,
+    SchedVar<NoSchedPred,      [NoReadAdvance]>
+  ]>;
+  // ADC,ADD,NEG,RSB,RSC,SBC,SUB,ADR
+  // AND,BIC,EOR,ORN,ORR
+  // CLZ,RBIT,REV,REV16,REVSH,PKH
+  def : WriteRes<WriteALU, [SwiftUnitP01]>;
+  def : SchedAlias<WriteALUsi, SwiftWriteALUsi>;
+  def : SchedAlias<WriteALUsr, SwiftWriteALUsr>;
+  def : SchedAlias<WriteALUSsr, SwiftWriteALUSsr>;
+  def : ReadAdvance<ReadALU, 0>;
+  def : SchedAlias<ReadALUsr, SwiftReadAdvanceALUsr>;
+
+  // 4.2.5 Integer comparison
+  def : WriteRes<WriteCMP, [SwiftUnitP01]>;
+  def : WriteRes<WriteCMPsi, [SwiftUnitP01]>;
+  def : WriteRes<WriteCMPsr, [SwiftUnitP01]>;
+}
diff --git a/lib/Target/ARM/ARMSelectionDAGInfo.cpp b/lib/Target/ARM/ARMSelectionDAGInfo.cpp
index b33b3c915a6e..41a7e0c2c8a5 100644
--- a/lib/Target/ARM/ARMSelectionDAGInfo.cpp
+++ b/lib/Target/ARM/ARMSelectionDAGInfo.cpp
@@ -13,8 +13,8 @@
 
 #define DEBUG_TYPE "arm-selectiondag-info"
 #include "ARMTargetMachine.h"
-#include "llvm/DerivedTypes.h"
 #include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/IR/DerivedTypes.h"
 using namespace llvm;
 
 ARMSelectionDAGInfo::ARMSelectionDAGInfo(const TargetMachine &TM)
diff --git a/lib/Target/ARM/ARMSubtarget.cpp b/lib/Target/ARM/ARMSubtarget.cpp
index bcc9db4ae3e3..739300e4eff9 100644
--- a/lib/Target/ARM/ARMSubtarget.cpp
+++ b/lib/Target/ARM/ARMSubtarget.cpp
@@ -12,11 +12,14 @@
 //===----------------------------------------------------------------------===//
 
 #include "ARMSubtarget.h"
-#include "ARMBaseRegisterInfo.h"
 #include "ARMBaseInstrInfo.h"
-#include "llvm/GlobalValue.h"
-#include "llvm/Target/TargetInstrInfo.h"
+#include "ARMBaseRegisterInfo.h"
+#include "llvm/IR/Attributes.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetOptions.h"
 
 #define GET_SUBTARGETINFO_TARGET_DESC
 #define GET_SUBTARGETINFO_CTOR
@@ -40,60 +43,88 @@ StrictAlign("arm-strict-align", cl::Hidden,
             cl::desc("Disallow all unaligned memory accesses"));
 
 ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &CPU,
-                           const std::string &FS)
+                           const std::string &FS, const TargetOptions &Options)
   : ARMGenSubtargetInfo(TT, CPU, FS)
   , ARMProcFamily(Others)
-  , HasV4TOps(false)
-  , HasV5TOps(false)
-  , HasV5TEOps(false)
-  , HasV6Ops(false)
-  , HasV6T2Ops(false)
-  , HasV7Ops(false)
-  , HasVFPv2(false)
-  , HasVFPv3(false)
-  , HasVFPv4(false)
-  , HasNEON(false)
-  , UseNEONForSinglePrecisionFP(false)
-  , UseMulOps(UseFusedMulOps)
-  , SlowFPVMLx(false)
-  , HasVMLxForwarding(false)
-  , SlowFPBrcc(false)
-  , InThumbMode(false)
-  , HasThumb2(false)
-  , IsMClass(false)
-  , NoARM(false)
-  , PostRAScheduler(false)
-  , IsR9Reserved(ReserveR9)
-  , UseMovt(false)
-  , SupportsTailCall(false)
-  , HasFP16(false)
-  , HasD16(false)
-  , HasHardwareDivide(false)
-  , HasHardwareDivideInARM(false)
-  , HasT2ExtractPack(false)
-  , HasDataBarrier(false)
-  , Pref32BitThumb(false)
-  , AvoidCPSRPartialUpdate(false)
-  , HasRAS(false)
-  , HasMPExtension(false)
-  , FPOnlySP(false)
-  , AllowsUnalignedMem(false)
-  , Thumb2DSP(false)
   , stackAlignment(4)
   , CPUString(CPU)
   , TargetTriple(TT)
+  , Options(Options)
   , TargetABI(ARM_ABI_APCS) {
-  // Determine default and user specified characteristics
+  initializeEnvironment();
+  resetSubtargetFeatures(CPU, FS);
+}
+
+void ARMSubtarget::initializeEnvironment() {
+  HasV4TOps = false;
+  HasV5TOps = false;
+  HasV5TEOps = false;
+  HasV6Ops = false;
+  HasV6T2Ops = false;
+  HasV7Ops = false;
+  HasVFPv2 = false;
+  HasVFPv3 = false;
+  HasVFPv4 = false;
+  HasNEON = false;
+  UseNEONForSinglePrecisionFP = false;
+  UseMulOps = UseFusedMulOps;
+  SlowFPVMLx = false;
+  HasVMLxForwarding = false;
+  SlowFPBrcc = false;
+  InThumbMode = false;
+  HasThumb2 = false;
+  IsMClass = false;
+  NoARM = false;
+  PostRAScheduler = false;
+  IsR9Reserved = ReserveR9;
+  UseMovt = false;
+  SupportsTailCall = false;
+  HasFP16 = false;
+  HasD16 = false;
+  HasHardwareDivide = false;
+  HasHardwareDivideInARM = false;
+  HasT2ExtractPack = false;
+  HasDataBarrier = false;
+  Pref32BitThumb = false;
+  AvoidCPSRPartialUpdate = false;
+  AvoidMOVsShifterOperand = false;
+  HasRAS = false;
+  HasMPExtension = false;
+  FPOnlySP = false;
+  AllowsUnalignedMem = false;
+  Thumb2DSP = false;
+  UseNaClTrap = false;
+  UnsafeFPMath = false;
+}
+
+void ARMSubtarget::resetSubtargetFeatures(const MachineFunction *MF) {
+  AttributeSet FnAttrs = MF->getFunction()->getAttributes();
+  Attribute CPUAttr = FnAttrs.getAttribute(AttributeSet::FunctionIndex,
+                                           "target-cpu");
+  Attribute FSAttr = FnAttrs.getAttribute(AttributeSet::FunctionIndex,
+                                          "target-features");
+  std::string CPU =
+    !CPUAttr.hasAttribute(Attribute::None) ?CPUAttr.getValueAsString() : "";
+  std::string FS =
+    !FSAttr.hasAttribute(Attribute::None) ? FSAttr.getValueAsString() : "";
+  if (!FS.empty()) {
+    initializeEnvironment();
+    resetSubtargetFeatures(CPU, FS);
+  }
+}
+
+void ARMSubtarget::resetSubtargetFeatures(StringRef CPU, StringRef FS) {
   if (CPUString.empty())
     CPUString = "generic";
 
   // Insert the architecture feature derived from the target triple into the
   // feature string. This is important for setting features that are implied
   // based on the architecture version.
-  std::string ArchFS = ARM_MC::ParseARMTriple(TT, CPUString);
+  std::string ArchFS = ARM_MC::ParseARMTriple(TargetTriple.getTriple(),
+                                              CPUString);
   if (!FS.empty()) {
     if (!ArchFS.empty())
-      ArchFS = ArchFS + "," + FS;
+      ArchFS = ArchFS + "," + FS.str();
     else
       ArchFS = FS;
   }
@@ -110,7 +141,8 @@ ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &CPU,
   // Initialize scheduling itinerary for the specified CPU.
   InstrItins = getInstrItineraryForCPU(CPUString);
 
-  if ((TT.find("eabi") != std::string::npos) || (isTargetIOS() && isMClass()))
+  if ((TargetTriple.getTriple().find("eabi") != std::string::npos) ||
+      (isTargetIOS() && isMClass()))
     // FIXME: We might want to separate AAPCS and EABI. Some systems, e.g.
     // Darwin-EABI conforms to AACPS but not the rest of EABI.
     TargetABI = ARM_ABI_AAPCS;
@@ -133,6 +165,12 @@ ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &CPU,
   // configuration.
   if (!StrictAlign && hasV6Ops() && isTargetDarwin())
     AllowsUnalignedMem = true;
+
+  // NEON f32 ops are non-IEEE 754 compliant. Darwin is ok with it by default.
+  uint64_t Bits = getFeatureBits();
+  if ((Bits & ARM::ProcA5 || Bits & ARM::ProcA8) && // Where this matters
+      (Options.UnsafeFPMath || isTargetDarwin()))
+    UseNEONForSinglePrecisionFP = true;
 }
 
 /// GVIsIndirectSymbol - true if the GV will be accessed via an indirect symbol.
diff --git a/lib/Target/ARM/ARMSubtarget.h b/lib/Target/ARM/ARMSubtarget.h
index 8e6b6506022d..5b5ee6aeb865 100644
--- a/lib/Target/ARM/ARMSubtarget.h
+++ b/lib/Target/ARM/ARMSubtarget.h
@@ -15,9 +15,9 @@
 #define ARMSUBTARGET_H
 
 #include "MCTargetDesc/ARMMCTargetDesc.h"
-#include "llvm/Target/TargetSubtargetInfo.h"
-#include "llvm/MC/MCInstrItineraries.h"
 #include "llvm/ADT/Triple.h"
+#include "llvm/MC/MCInstrItineraries.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
 #include <string>
 
 #define GET_SUBTARGETINFO_HEADER
@@ -26,11 +26,12 @@
 namespace llvm {
 class GlobalValue;
 class StringRef;
+class TargetOptions;
 
 class ARMSubtarget : public ARMGenSubtargetInfo {
 protected:
   enum ARMProcFamilyEnum {
-    Others, CortexA8, CortexA9, CortexA15, Swift
+    Others, CortexA5, CortexA8, CortexA9, CortexA15, CortexR5, Swift
   };
 
   /// ARMProcFamily - ARM processor family: Cortex-A8, Cortex-A9, and others.
@@ -131,6 +132,10 @@ protected:
   /// CPSR setting instruction.
   bool AvoidCPSRPartialUpdate;
 
+  /// AvoidMOVsShifterOperand - If true, codegen should avoid using flag setting
+  /// movs with shifter operand (i.e. asr, lsl, lsr).
+  bool AvoidMOVsShifterOperand;
+
   /// HasRAS - Some processors perform return stack prediction. CodeGen should
   /// avoid issue "normal" call instructions to callees which do not return.
   bool HasRAS;
@@ -152,6 +157,12 @@ protected:
   /// and such) instructions in Thumb2 code.
   bool Thumb2DSP;
 
+  /// NaCl TRAP instruction is generated instead of the regular TRAP.
+  bool UseNaClTrap;
+
+  /// Target machine allowed unsafe FP math (such as use of NEON fp)
+  bool UnsafeFPMath;
+
   /// stackAlignment - The minimum alignment known to hold of the stack frame on
   /// entry to the function and which must be maintained by every function.
   unsigned stackAlignment;
@@ -168,6 +179,9 @@ protected:
   /// Selected instruction itineraries (one entry per itinerary class.)
   InstrItineraryData InstrItins;
 
+  /// Options passed via command line that could influence the target
+  const TargetOptions &Options;
+
  public:
   enum {
     isELF, isDarwin
@@ -182,7 +196,7 @@ protected:
   /// of the specified triple.
   ///
   ARMSubtarget(const std::string &TT, const std::string &CPU,
-               const std::string &FS);
+               const std::string &FS, const TargetOptions &Options);
 
   /// getMaxInlineSizeThreshold - Returns the maximum memset / memcpy size
   /// that still makes it profitable to inline the call.
@@ -195,6 +209,12 @@ protected:
   /// subtarget options.  Definition of function is auto generated by tblgen.
   void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
 
+  /// \brief Reset the features for the ARM target.
+  virtual void resetSubtargetFeatures(const MachineFunction *MF);
+private:
+  void initializeEnvironment();
+  void resetSubtargetFeatures(StringRef CPU, StringRef FS);
+public:
   void computeIssueWidth();
 
   bool hasV4TOps()  const { return HasV4TOps;  }
@@ -204,12 +224,14 @@ protected:
   bool hasV6T2Ops() const { return HasV6T2Ops; }
   bool hasV7Ops()   const { return HasV7Ops;  }
 
+  bool isCortexA5() const { return ARMProcFamily == CortexA5; }
   bool isCortexA8() const { return ARMProcFamily == CortexA8; }
   bool isCortexA9() const { return ARMProcFamily == CortexA9; }
   bool isCortexA15() const { return ARMProcFamily == CortexA15; }
   bool isSwift()    const { return ARMProcFamily == Swift; }
   bool isCortexM3() const { return CPUString == "cortex-m3"; }
   bool isLikeA9() const { return isCortexA9() || isCortexA15(); }
+  bool isCortexR5() const { return ARMProcFamily == CortexR5; }
 
   bool hasARMOps() const { return !NoARM; }
 
@@ -231,9 +253,11 @@ protected:
   bool isFPOnlySP() const { return FPOnlySP; }
   bool prefers32BitThumb() const { return Pref32BitThumb; }
   bool avoidCPSRPartialUpdate() const { return AvoidCPSRPartialUpdate; }
+  bool avoidMOVsShifterOperand() const { return AvoidMOVsShifterOperand; }
   bool hasRAS() const { return HasRAS; }
   bool hasMPExtension() const { return HasMPExtension; }
   bool hasThumb2DSP() const { return Thumb2DSP; }
+  bool useNaClTrap() const { return UseNaClTrap; }
 
   bool hasFP16() const { return HasFP16; }
   bool hasD16() const { return HasD16; }
@@ -243,7 +267,7 @@ protected:
   bool isTargetIOS() const { return TargetTriple.getOS() == Triple::IOS; }
   bool isTargetDarwin() const { return TargetTriple.isOSDarwin(); }
   bool isTargetNaCl() const {
-    return TargetTriple.getOS() == Triple::NativeClient;
+    return TargetTriple.getOS() == Triple::NaCl;
   }
   bool isTargetELF() const { return !isTargetDarwin(); }
 
diff --git a/lib/Target/ARM/ARMTargetMachine.cpp b/lib/Target/ARM/ARMTargetMachine.cpp
index b486d4fe2ef9..42c7d2c437e0 100644
--- a/lib/Target/ARM/ARMTargetMachine.cpp
+++ b/lib/Target/ARM/ARMTargetMachine.cpp
@@ -11,11 +11,11 @@
 //===----------------------------------------------------------------------===//
 
 #include "ARMTargetMachine.h"
-#include "ARMFrameLowering.h"
 #include "ARM.h"
-#include "llvm/PassManager.h"
+#include "ARMFrameLowering.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/MC/MCAsmInfo.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/FormattedStream.h"
 #include "llvm/Support/TargetRegistry.h"
@@ -28,6 +28,11 @@ EnableGlobalMerge("global-merge", cl::Hidden,
                   cl::desc("Enable global merge pass"),
                   cl::init(true));
 
+static cl::opt<bool>
+DisableA15SDOptimization("disable-a15-sd-optimization", cl::Hidden,
+                   cl::desc("Inhibit optimization of S->D register accesses on A15"),
+                   cl::init(false));
+
 extern "C" void LLVMInitializeARMTarget() {
   // Register the target.
   RegisterTargetMachine<ARMTargetMachine> X(TheARMTarget);
@@ -43,7 +48,7 @@ ARMBaseTargetMachine::ARMBaseTargetMachine(const Target &T, StringRef TT,
                                            Reloc::Model RM, CodeModel::Model CM,
                                            CodeGenOpt::Level OL)
   : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
-    Subtarget(TT, CPU, FS),
+    Subtarget(TT, CPU, FS, Options),
     JITInfo(),
     InstrItins(Subtarget.getInstrItineraryData()) {
   // Default to soft float ABI
@@ -51,6 +56,15 @@ ARMBaseTargetMachine::ARMBaseTargetMachine(const Target &T, StringRef TT,
     this->Options.FloatABIType = FloatABI::Soft;
 }
 
+void ARMBaseTargetMachine::addAnalysisPasses(PassManagerBase &PM) {
+  // Add first the target-independent BasicTTI pass, then our ARM pass. This
+  // allows the ARM pass to delegate to the target independent layer when
+  // appropriate.
+  PM.add(createBasicTargetTransformInfoPass(getTargetLowering()));
+  PM.add(createARMTargetTransformInfoPass(this));
+}
+
+
 void ARMTargetMachine::anchor() { }
 
 ARMTargetMachine::ARMTargetMachine(const Target &T, StringRef TT,
@@ -70,8 +84,7 @@ ARMTargetMachine::ARMTargetMachine(const Target &T, StringRef TT,
                            "v128:64:128-v64:64:64-n32-S32")),
     TLInfo(*this),
     TSInfo(*this),
-    FrameLowering(Subtarget),
-    STTI(&TLInfo), VTTI(&TLInfo) {
+    FrameLowering(Subtarget) {
   if (!Subtarget.hasARMOps())
     report_fatal_error("CPU: '" + Subtarget.getCPUString() + "' does not "
                        "support ARM mode execution!");
@@ -103,8 +116,7 @@ ThumbTargetMachine::ThumbTargetMachine(const Target &T, StringRef TT,
     TSInfo(*this),
     FrameLowering(Subtarget.hasThumb2()
               ? new ARMFrameLowering(Subtarget)
-              : (ARMFrameLowering*)new Thumb1FrameLowering(Subtarget)),
-    STTI(&TLInfo), VTTI(&TLInfo) {
+              : (ARMFrameLowering*)new Thumb1FrameLowering(Subtarget)) {
 }
 
 namespace {
@@ -157,6 +169,12 @@ bool ARMPassConfig::addPreRegAlloc() {
     addPass(createARMLoadStoreOptimizationPass(true));
   if (getOptLevel() != CodeGenOpt::None && getARMSubtarget().isLikeA9())
     addPass(createMLxExpansionPass());
+  // Since the A15SDOptimizer pass can insert VDUP instructions, it can only be
+  // enabled when NEON is available.
+  if (getOptLevel() != CodeGenOpt::None && getARMSubtarget().isCortexA15() &&
+    getARMSubtarget().hasNEON() && !DisableA15SDOptimization) {
+    addPass(createA15SDOptimizerPass());
+  }
   return true;
 }
 
diff --git a/lib/Target/ARM/ARMTargetMachine.h b/lib/Target/ARM/ARMTargetMachine.h
index ebdd5b4d64c9..d4caf5ca6e19 100644
--- a/lib/Target/ARM/ARMTargetMachine.h
+++ b/lib/Target/ARM/ARMTargetMachine.h
@@ -14,20 +14,19 @@
 #ifndef ARMTARGETMACHINE_H
 #define ARMTARGETMACHINE_H
 
-#include "ARMInstrInfo.h"
 #include "ARMFrameLowering.h"
-#include "ARMJITInfo.h"
-#include "ARMSubtarget.h"
 #include "ARMISelLowering.h"
+#include "ARMInstrInfo.h"
+#include "ARMJITInfo.h"
 #include "ARMSelectionDAGInfo.h"
-#include "Thumb1InstrInfo.h"
+#include "ARMSubtarget.h"
 #include "Thumb1FrameLowering.h"
+#include "Thumb1InstrInfo.h"
 #include "Thumb2InstrInfo.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetTransformImpl.h"
-#include "llvm/DataLayout.h"
-#include "llvm/MC/MCStreamer.h"
 #include "llvm/ADT/OwningPtr.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/Target/TargetMachine.h"
 
 namespace llvm {
 
@@ -47,10 +46,17 @@ public:
 
   virtual       ARMJITInfo       *getJITInfo()         { return &JITInfo; }
   virtual const ARMSubtarget  *getSubtargetImpl() const { return &Subtarget; }
+  virtual const ARMTargetLowering *getTargetLowering() const {
+    // Implemented by derived classes
+    llvm_unreachable("getTargetLowering not implemented");
+  }
   virtual const InstrItineraryData *getInstrItineraryData() const {
     return &InstrItins;
   }
 
+  /// \brief Register ARM analysis passes with a pass manager.
+  virtual void addAnalysisPasses(PassManagerBase &PM);
+
   // Pass Pipeline Configuration
   virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
 
@@ -66,8 +72,6 @@ class ARMTargetMachine : public ARMBaseTargetMachine {
   ARMTargetLowering   TLInfo;
   ARMSelectionDAGInfo TSInfo;
   ARMFrameLowering    FrameLowering;
-  ScalarTargetTransformImpl STTI;
-  VectorTargetTransformImpl VTTI;
  public:
   ARMTargetMachine(const Target &T, StringRef TT,
                    StringRef CPU, StringRef FS,
@@ -89,12 +93,6 @@ class ARMTargetMachine : public ARMBaseTargetMachine {
   virtual const ARMFrameLowering *getFrameLowering() const {
     return &FrameLowering;
   }
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
   virtual const ARMInstrInfo     *getInstrInfo() const { return &InstrInfo; }
   virtual const DataLayout       *getDataLayout() const { return &DL; }
 };
@@ -112,8 +110,6 @@ class ThumbTargetMachine : public ARMBaseTargetMachine {
   ARMSelectionDAGInfo TSInfo;
   // Either Thumb1FrameLowering or ARMFrameLowering.
   OwningPtr<ARMFrameLowering> FrameLowering;
-  ScalarTargetTransformImpl STTI;
-  VectorTargetTransformImpl VTTI;
 public:
   ThumbTargetMachine(const Target &T, StringRef TT,
                      StringRef CPU, StringRef FS,
@@ -142,12 +138,6 @@ public:
   virtual const ARMFrameLowering *getFrameLowering() const {
     return FrameLowering.get();
   }
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
   virtual const DataLayout       *getDataLayout() const { return &DL; }
 };
 
diff --git a/lib/Target/ARM/ARMTargetObjectFile.cpp b/lib/Target/ARM/ARMTargetObjectFile.cpp
index 3d85ca7d6995..dfdf6ab356a3 100644
--- a/lib/Target/ARM/ARMTargetObjectFile.cpp
+++ b/lib/Target/ARM/ARMTargetObjectFile.cpp
@@ -9,12 +9,14 @@
 
 #include "ARMTargetObjectFile.h"
 #include "ARMSubtarget.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCSectionELF.h"
 #include "llvm/Support/Dwarf.h"
 #include "llvm/Support/ELF.h"
+#include "llvm/Target/Mangler.h"
 #include "llvm/Target/TargetMachine.h"
-#include "llvm/ADT/StringExtras.h"
 using namespace llvm;
 using namespace dwarf;
 
@@ -38,3 +40,14 @@ void ARMElfTargetObjectFile::Initialize(MCContext &Ctx,
                                0,
                                SectionKind::getMetadata());
 }
+
+const MCExpr *ARMElfTargetObjectFile::
+getTTypeGlobalReference(const GlobalValue *GV, Mangler *Mang,
+                        MachineModuleInfo *MMI, unsigned Encoding,
+                        MCStreamer &Streamer) const {
+  assert(Encoding == DW_EH_PE_absptr && "Can handle absptr encoding only");
+
+  return MCSymbolRefExpr::Create(Mang->getSymbol(GV),
+                                 MCSymbolRefExpr::VK_ARM_TARGET2,
+                                 getContext());
+}
diff --git a/lib/Target/ARM/ARMTargetObjectFile.h b/lib/Target/ARM/ARMTargetObjectFile.h
index c6a7261439d7..7f60727e5305 100644
--- a/lib/Target/ARM/ARMTargetObjectFile.h
+++ b/lib/Target/ARM/ARMTargetObjectFile.h
@@ -28,6 +28,11 @@ public:
 
   virtual void Initialize(MCContext &Ctx, const TargetMachine &TM);
 
+  const MCExpr *
+  getTTypeGlobalReference(const GlobalValue *GV, Mangler *Mang,
+                          MachineModuleInfo *MMI, unsigned Encoding,
+                          MCStreamer &Streamer) const;
+  
   virtual const MCSection *getAttributesSection() const {
     return AttributesSection;
   }
diff --git a/lib/Target/ARM/ARMTargetTransformInfo.cpp b/lib/Target/ARM/ARMTargetTransformInfo.cpp
new file mode 100644
index 000000000000..1019b972e957
--- /dev/null
+++ b/lib/Target/ARM/ARMTargetTransformInfo.cpp
@@ -0,0 +1,458 @@
+//===-- ARMTargetTransformInfo.cpp - ARM specific TTI pass ----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+/// \file
+/// This file implements a TargetTransformInfo analysis pass specific to the
+/// ARM target machine. It uses the target's detailed information to provide
+/// more precise answers to certain TTI queries, while letting the target
+/// independent and default TTI implementations handle the rest.
+///
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "armtti"
+#include "ARM.h"
+#include "ARMTargetMachine.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/CostTable.h"
+using namespace llvm;
+
+// Declare the pass initialization routine locally as target-specific passes
+// don't havve a target-wide initialization entry point, and so we rely on the
+// pass constructor initialization.
+namespace llvm {
+void initializeARMTTIPass(PassRegistry &);
+}
+
+namespace {
+
+class ARMTTI : public ImmutablePass, public TargetTransformInfo {
+  const ARMBaseTargetMachine *TM;
+  const ARMSubtarget *ST;
+  const ARMTargetLowering *TLI;
+
+  /// Estimate the overhead of scalarizing an instruction. Insert and Extract
+  /// are set if the result needs to be inserted and/or extracted from vectors.
+  unsigned getScalarizationOverhead(Type *Ty, bool Insert, bool Extract) const;
+
+public:
+  ARMTTI() : ImmutablePass(ID), TM(0), ST(0), TLI(0) {
+    llvm_unreachable("This pass cannot be directly constructed");
+  }
+
+  ARMTTI(const ARMBaseTargetMachine *TM)
+      : ImmutablePass(ID), TM(TM), ST(TM->getSubtargetImpl()),
+        TLI(TM->getTargetLowering()) {
+    initializeARMTTIPass(*PassRegistry::getPassRegistry());
+  }
+
+  virtual void initializePass() {
+    pushTTIStack(this);
+  }
+
+  virtual void finalizePass() {
+    popTTIStack();
+  }
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+    TargetTransformInfo::getAnalysisUsage(AU);
+  }
+
+  /// Pass identification.
+  static char ID;
+
+  /// Provide necessary pointer adjustments for the two base classes.
+  virtual void *getAdjustedAnalysisPointer(const void *ID) {
+    if (ID == &TargetTransformInfo::ID)
+      return (TargetTransformInfo*)this;
+    return this;
+  }
+
+  /// \name Scalar TTI Implementations
+  /// @{
+
+  virtual unsigned getIntImmCost(const APInt &Imm, Type *Ty) const;
+
+  /// @}
+
+
+  /// \name Vector TTI Implementations
+  /// @{
+
+  unsigned getNumberOfRegisters(bool Vector) const {
+    if (Vector) {
+      if (ST->hasNEON())
+        return 16;
+      return 0;
+    }
+
+    if (ST->isThumb1Only())
+      return 8;
+    return 16;
+  }
+
+  unsigned getRegisterBitWidth(bool Vector) const {
+    if (Vector) {
+      if (ST->hasNEON())
+        return 128;
+      return 0;
+    }
+
+    return 32;
+  }
+
+  unsigned getMaximumUnrollFactor() const {
+    // These are out of order CPUs:
+    if (ST->isCortexA15() || ST->isSwift())
+      return 2;
+    return 1;
+  }
+
+  unsigned getShuffleCost(ShuffleKind Kind, Type *Tp,
+                          int Index, Type *SubTp) const;
+
+  unsigned getCastInstrCost(unsigned Opcode, Type *Dst,
+                                      Type *Src) const;
+
+  unsigned getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy) const;
+
+  unsigned getVectorInstrCost(unsigned Opcode, Type *Val, unsigned Index) const;
+
+  unsigned getAddressComputationCost(Type *Val) const;
+  /// @}
+};
+
+} // end anonymous namespace
+
+INITIALIZE_AG_PASS(ARMTTI, TargetTransformInfo, "armtti",
+                   "ARM Target Transform Info", true, true, false)
+char ARMTTI::ID = 0;
+
+ImmutablePass *
+llvm::createARMTargetTransformInfoPass(const ARMBaseTargetMachine *TM) {
+  return new ARMTTI(TM);
+}
+
+
+unsigned ARMTTI::getIntImmCost(const APInt &Imm, Type *Ty) const {
+  assert(Ty->isIntegerTy());
+
+  unsigned Bits = Ty->getPrimitiveSizeInBits();
+  if (Bits == 0 || Bits > 32)
+    return 4;
+
+  int32_t SImmVal = Imm.getSExtValue();
+  uint32_t ZImmVal = Imm.getZExtValue();
+  if (!ST->isThumb()) {
+    if ((SImmVal >= 0 && SImmVal < 65536) ||
+        (ARM_AM::getSOImmVal(ZImmVal) != -1) ||
+        (ARM_AM::getSOImmVal(~ZImmVal) != -1))
+      return 1;
+    return ST->hasV6T2Ops() ? 2 : 3;
+  } else if (ST->isThumb2()) {
+    if ((SImmVal >= 0 && SImmVal < 65536) ||
+        (ARM_AM::getT2SOImmVal(ZImmVal) != -1) ||
+        (ARM_AM::getT2SOImmVal(~ZImmVal) != -1))
+      return 1;
+    return ST->hasV6T2Ops() ? 2 : 3;
+  } else /*Thumb1*/ {
+    if (SImmVal >= 0 && SImmVal < 256)
+      return 1;
+    if ((~ZImmVal < 256) || ARM_AM::isThumbImmShiftedVal(ZImmVal))
+      return 2;
+    // Load from constantpool.
+    return 3;
+  }
+  return 2;
+}
+
+unsigned ARMTTI::getCastInstrCost(unsigned Opcode, Type *Dst,
+                                    Type *Src) const {
+  int ISD = TLI->InstructionOpcodeToISD(Opcode);
+  assert(ISD && "Invalid opcode");
+
+  // Single to/from double precision conversions.
+  static const CostTblEntry<MVT> NEONFltDblTbl[] = {
+    // Vector fptrunc/fpext conversions.
+    { ISD::FP_ROUND,   MVT::v2f64, 2 },
+    { ISD::FP_EXTEND,  MVT::v2f32, 2 },
+    { ISD::FP_EXTEND,  MVT::v4f32, 4 }
+  };
+
+  if (Src->isVectorTy() && ST->hasNEON() && (ISD == ISD::FP_ROUND ||
+                                          ISD == ISD::FP_EXTEND)) {
+    std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Src);
+    int Idx = CostTableLookup<MVT>(NEONFltDblTbl, array_lengthof(NEONFltDblTbl),
+                                ISD, LT.second);
+    if (Idx != -1)
+      return LT.first * NEONFltDblTbl[Idx].Cost;
+  }
+
+  EVT SrcTy = TLI->getValueType(Src);
+  EVT DstTy = TLI->getValueType(Dst);
+
+  if (!SrcTy.isSimple() || !DstTy.isSimple())
+    return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
+
+  // Some arithmetic, load and store operations have specific instructions
+  // to cast up/down their types automatically at no extra cost.
+  // TODO: Get these tables to know at least what the related operations are.
+  static const TypeConversionCostTblEntry<MVT> NEONVectorConversionTbl[] = {
+    { ISD::SIGN_EXTEND, MVT::v4i32, MVT::v4i16, 0 },
+    { ISD::ZERO_EXTEND, MVT::v4i32, MVT::v4i16, 0 },
+    { ISD::SIGN_EXTEND, MVT::v2i64, MVT::v2i32, 1 },
+    { ISD::ZERO_EXTEND, MVT::v2i64, MVT::v2i32, 1 },
+    { ISD::TRUNCATE,    MVT::v4i32, MVT::v4i64, 0 },
+    { ISD::TRUNCATE,    MVT::v4i16, MVT::v4i32, 1 },
+
+    // The number of vmovl instructions for the extension.
+    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i16, 3 },
+    { ISD::ZERO_EXTEND, MVT::v4i64, MVT::v4i16, 3 },
+    { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i8, 3 },
+    { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i8, 3 },
+    { ISD::SIGN_EXTEND, MVT::v8i64, MVT::v8i8, 7 },
+    { ISD::ZERO_EXTEND, MVT::v8i64, MVT::v8i8, 7 },
+    { ISD::SIGN_EXTEND, MVT::v8i64, MVT::v8i16, 6 },
+    { ISD::ZERO_EXTEND, MVT::v8i64, MVT::v8i16, 6 },
+    { ISD::SIGN_EXTEND, MVT::v16i32, MVT::v16i8, 6 },
+    { ISD::ZERO_EXTEND, MVT::v16i32, MVT::v16i8, 6 },
+
+    // Operations that we legalize using load/stores to the stack.
+    { ISD::TRUNCATE,    MVT::v16i8, MVT::v16i32, 4*1 + 16*2 + 2*1 },
+    { ISD::TRUNCATE,    MVT::v8i8, MVT::v8i32, 2*1 + 8*2 + 1 },
+
+    // Vector float <-> i32 conversions.
+    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i32, 1 },
+    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i32, 1 },
+
+    { ISD::SINT_TO_FP,  MVT::v2f32, MVT::v2i8, 3 },
+    { ISD::UINT_TO_FP,  MVT::v2f32, MVT::v2i8, 3 },
+    { ISD::SINT_TO_FP,  MVT::v2f32, MVT::v2i16, 2 },
+    { ISD::UINT_TO_FP,  MVT::v2f32, MVT::v2i16, 2 },
+    { ISD::SINT_TO_FP,  MVT::v2f32, MVT::v2i32, 1 },
+    { ISD::UINT_TO_FP,  MVT::v2f32, MVT::v2i32, 1 },
+    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i1, 3 },
+    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i1, 3 },
+    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i8, 3 },
+    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i8, 3 },
+    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i16, 2 },
+    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i16, 2 },
+    { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i16, 4 },
+    { ISD::UINT_TO_FP,  MVT::v8f32, MVT::v8i16, 4 },
+    { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i32, 2 },
+    { ISD::UINT_TO_FP,  MVT::v8f32, MVT::v8i32, 2 },
+    { ISD::SINT_TO_FP,  MVT::v16f32, MVT::v16i16, 8 },
+    { ISD::UINT_TO_FP,  MVT::v16f32, MVT::v16i16, 8 },
+    { ISD::SINT_TO_FP,  MVT::v16f32, MVT::v16i32, 4 },
+    { ISD::UINT_TO_FP,  MVT::v16f32, MVT::v16i32, 4 },
+
+    { ISD::FP_TO_SINT,  MVT::v4i32, MVT::v4f32, 1 },
+    { ISD::FP_TO_UINT,  MVT::v4i32, MVT::v4f32, 1 },
+    { ISD::FP_TO_SINT,  MVT::v4i8, MVT::v4f32, 3 },
+    { ISD::FP_TO_UINT,  MVT::v4i8, MVT::v4f32, 3 },
+    { ISD::FP_TO_SINT,  MVT::v4i16, MVT::v4f32, 2 },
+    { ISD::FP_TO_UINT,  MVT::v4i16, MVT::v4f32, 2 },
+
+    // Vector double <-> i32 conversions.
+    { ISD::SINT_TO_FP,  MVT::v2f64, MVT::v2i32, 2 },
+    { ISD::UINT_TO_FP,  MVT::v2f64, MVT::v2i32, 2 },
+
+    { ISD::SINT_TO_FP,  MVT::v2f64, MVT::v2i8, 4 },
+    { ISD::UINT_TO_FP,  MVT::v2f64, MVT::v2i8, 4 },
+    { ISD::SINT_TO_FP,  MVT::v2f64, MVT::v2i16, 3 },
+    { ISD::UINT_TO_FP,  MVT::v2f64, MVT::v2i16, 3 },
+    { ISD::SINT_TO_FP,  MVT::v2f64, MVT::v2i32, 2 },
+    { ISD::UINT_TO_FP,  MVT::v2f64, MVT::v2i32, 2 },
+
+    { ISD::FP_TO_SINT,  MVT::v2i32, MVT::v2f64, 2 },
+    { ISD::FP_TO_UINT,  MVT::v2i32, MVT::v2f64, 2 },
+    { ISD::FP_TO_SINT,  MVT::v8i16, MVT::v8f32, 4 },
+    { ISD::FP_TO_UINT,  MVT::v8i16, MVT::v8f32, 4 },
+    { ISD::FP_TO_SINT,  MVT::v16i16, MVT::v16f32, 8 },
+    { ISD::FP_TO_UINT,  MVT::v16i16, MVT::v16f32, 8 }
+  };
+
+  if (SrcTy.isVector() && ST->hasNEON()) {
+    int Idx = ConvertCostTableLookup<MVT>(NEONVectorConversionTbl,
+                                array_lengthof(NEONVectorConversionTbl),
+                                ISD, DstTy.getSimpleVT(), SrcTy.getSimpleVT());
+    if (Idx != -1)
+      return NEONVectorConversionTbl[Idx].Cost;
+  }
+
+  // Scalar float to integer conversions.
+  static const TypeConversionCostTblEntry<MVT> NEONFloatConversionTbl[] = {
+    { ISD::FP_TO_SINT,  MVT::i1, MVT::f32, 2 },
+    { ISD::FP_TO_UINT,  MVT::i1, MVT::f32, 2 },
+    { ISD::FP_TO_SINT,  MVT::i1, MVT::f64, 2 },
+    { ISD::FP_TO_UINT,  MVT::i1, MVT::f64, 2 },
+    { ISD::FP_TO_SINT,  MVT::i8, MVT::f32, 2 },
+    { ISD::FP_TO_UINT,  MVT::i8, MVT::f32, 2 },
+    { ISD::FP_TO_SINT,  MVT::i8, MVT::f64, 2 },
+    { ISD::FP_TO_UINT,  MVT::i8, MVT::f64, 2 },
+    { ISD::FP_TO_SINT,  MVT::i16, MVT::f32, 2 },
+    { ISD::FP_TO_UINT,  MVT::i16, MVT::f32, 2 },
+    { ISD::FP_TO_SINT,  MVT::i16, MVT::f64, 2 },
+    { ISD::FP_TO_UINT,  MVT::i16, MVT::f64, 2 },
+    { ISD::FP_TO_SINT,  MVT::i32, MVT::f32, 2 },
+    { ISD::FP_TO_UINT,  MVT::i32, MVT::f32, 2 },
+    { ISD::FP_TO_SINT,  MVT::i32, MVT::f64, 2 },
+    { ISD::FP_TO_UINT,  MVT::i32, MVT::f64, 2 },
+    { ISD::FP_TO_SINT,  MVT::i64, MVT::f32, 10 },
+    { ISD::FP_TO_UINT,  MVT::i64, MVT::f32, 10 },
+    { ISD::FP_TO_SINT,  MVT::i64, MVT::f64, 10 },
+    { ISD::FP_TO_UINT,  MVT::i64, MVT::f64, 10 }
+  };
+  if (SrcTy.isFloatingPoint() && ST->hasNEON()) {
+    int Idx = ConvertCostTableLookup<MVT>(NEONFloatConversionTbl,
+                                        array_lengthof(NEONFloatConversionTbl),
+                                        ISD, DstTy.getSimpleVT(),
+                                        SrcTy.getSimpleVT());
+    if (Idx != -1)
+        return NEONFloatConversionTbl[Idx].Cost;
+  }
+
+  // Scalar integer to float conversions.
+  static const TypeConversionCostTblEntry<MVT> NEONIntegerConversionTbl[] = {
+    { ISD::SINT_TO_FP,  MVT::f32, MVT::i1, 2 },
+    { ISD::UINT_TO_FP,  MVT::f32, MVT::i1, 2 },
+    { ISD::SINT_TO_FP,  MVT::f64, MVT::i1, 2 },
+    { ISD::UINT_TO_FP,  MVT::f64, MVT::i1, 2 },
+    { ISD::SINT_TO_FP,  MVT::f32, MVT::i8, 2 },
+    { ISD::UINT_TO_FP,  MVT::f32, MVT::i8, 2 },
+    { ISD::SINT_TO_FP,  MVT::f64, MVT::i8, 2 },
+    { ISD::UINT_TO_FP,  MVT::f64, MVT::i8, 2 },
+    { ISD::SINT_TO_FP,  MVT::f32, MVT::i16, 2 },
+    { ISD::UINT_TO_FP,  MVT::f32, MVT::i16, 2 },
+    { ISD::SINT_TO_FP,  MVT::f64, MVT::i16, 2 },
+    { ISD::UINT_TO_FP,  MVT::f64, MVT::i16, 2 },
+    { ISD::SINT_TO_FP,  MVT::f32, MVT::i32, 2 },
+    { ISD::UINT_TO_FP,  MVT::f32, MVT::i32, 2 },
+    { ISD::SINT_TO_FP,  MVT::f64, MVT::i32, 2 },
+    { ISD::UINT_TO_FP,  MVT::f64, MVT::i32, 2 },
+    { ISD::SINT_TO_FP,  MVT::f32, MVT::i64, 10 },
+    { ISD::UINT_TO_FP,  MVT::f32, MVT::i64, 10 },
+    { ISD::SINT_TO_FP,  MVT::f64, MVT::i64, 10 },
+    { ISD::UINT_TO_FP,  MVT::f64, MVT::i64, 10 }
+  };
+
+  if (SrcTy.isInteger() && ST->hasNEON()) {
+    int Idx = ConvertCostTableLookup<MVT>(NEONIntegerConversionTbl,
+                                       array_lengthof(NEONIntegerConversionTbl),
+                                       ISD, DstTy.getSimpleVT(),
+                                       SrcTy.getSimpleVT());
+    if (Idx != -1)
+      return NEONIntegerConversionTbl[Idx].Cost;
+  }
+
+  // Scalar integer conversion costs.
+  static const TypeConversionCostTblEntry<MVT> ARMIntegerConversionTbl[] = {
+    // i16 -> i64 requires two dependent operations.
+    { ISD::SIGN_EXTEND, MVT::i64, MVT::i16, 2 },
+
+    // Truncates on i64 are assumed to be free.
+    { ISD::TRUNCATE,    MVT::i32, MVT::i64, 0 },
+    { ISD::TRUNCATE,    MVT::i16, MVT::i64, 0 },
+    { ISD::TRUNCATE,    MVT::i8,  MVT::i64, 0 },
+    { ISD::TRUNCATE,    MVT::i1,  MVT::i64, 0 }
+  };
+
+  if (SrcTy.isInteger()) {
+    int Idx =
+      ConvertCostTableLookup<MVT>(ARMIntegerConversionTbl,
+                                  array_lengthof(ARMIntegerConversionTbl),
+                                  ISD, DstTy.getSimpleVT(),
+                                  SrcTy.getSimpleVT());
+    if (Idx != -1)
+      return ARMIntegerConversionTbl[Idx].Cost;
+  }
+
+  return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
+}
+
+unsigned ARMTTI::getVectorInstrCost(unsigned Opcode, Type *ValTy,
+                                    unsigned Index) const {
+  // Penalize inserting into an D-subregister. We end up with a three times
+  // lower estimated throughput on swift.
+  if (ST->isSwift() &&
+      Opcode == Instruction::InsertElement &&
+      ValTy->isVectorTy() &&
+      ValTy->getScalarSizeInBits() <= 32)
+    return 3;
+
+  return TargetTransformInfo::getVectorInstrCost(Opcode, ValTy, Index);
+}
+
+unsigned ARMTTI::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
+                                    Type *CondTy) const {
+
+  int ISD = TLI->InstructionOpcodeToISD(Opcode);
+  // On NEON a a vector select gets lowered to vbsl.
+  if (ST->hasNEON() && ValTy->isVectorTy() && ISD == ISD::SELECT) {
+    // Lowering of some vector selects is currently far from perfect.
+    static const TypeConversionCostTblEntry<MVT> NEONVectorSelectTbl[] = {
+      { ISD::SELECT, MVT::v16i1, MVT::v16i16, 2*16 + 1 + 3*1 + 4*1 },
+      { ISD::SELECT, MVT::v8i1, MVT::v8i32, 4*8 + 1*3 + 1*4 + 1*2 },
+      { ISD::SELECT, MVT::v16i1, MVT::v16i32, 4*16 + 1*6 + 1*8 + 1*4 },
+      { ISD::SELECT, MVT::v4i1, MVT::v4i64, 4*4 + 1*2 + 1 },
+      { ISD::SELECT, MVT::v8i1, MVT::v8i64, 50 },
+      { ISD::SELECT, MVT::v16i1, MVT::v16i64, 100 }
+    };
+
+    EVT SelCondTy = TLI->getValueType(CondTy);
+    EVT SelValTy = TLI->getValueType(ValTy);
+    int Idx = ConvertCostTableLookup<MVT>(NEONVectorSelectTbl,
+                                          array_lengthof(NEONVectorSelectTbl),
+                                          ISD, SelCondTy.getSimpleVT(),
+                                          SelValTy.getSimpleVT());
+    if (Idx != -1)
+      return NEONVectorSelectTbl[Idx].Cost;
+
+    std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(ValTy);
+    return LT.first;
+  }
+
+  return TargetTransformInfo::getCmpSelInstrCost(Opcode, ValTy, CondTy);
+}
+
+unsigned ARMTTI::getAddressComputationCost(Type *Ty) const {
+  // In many cases the address computation is not merged into the instruction
+  // addressing mode.
+  return 1;
+}
+
+unsigned ARMTTI::getShuffleCost(ShuffleKind Kind, Type *Tp, int Index,
+                                Type *SubTp) const {
+  // We only handle costs of reverse shuffles for now.
+  if (Kind != SK_Reverse)
+    return TargetTransformInfo::getShuffleCost(Kind, Tp, Index, SubTp);
+
+  static const CostTblEntry<MVT> NEONShuffleTbl[] = {
+    // Reverse shuffle cost one instruction if we are shuffling within a double
+    // word (vrev) or two if we shuffle a quad word (vrev, vext).
+    { ISD::VECTOR_SHUFFLE, MVT::v2i32, 1 },
+    { ISD::VECTOR_SHUFFLE, MVT::v2f32, 1 },
+    { ISD::VECTOR_SHUFFLE, MVT::v2i64, 1 },
+    { ISD::VECTOR_SHUFFLE, MVT::v2f64, 1 },
+
+    { ISD::VECTOR_SHUFFLE, MVT::v4i32, 2 },
+    { ISD::VECTOR_SHUFFLE, MVT::v4f32, 2 },
+    { ISD::VECTOR_SHUFFLE, MVT::v8i16, 2 },
+    { ISD::VECTOR_SHUFFLE, MVT::v16i8, 2 }
+  };
+
+  std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Tp);
+
+  int Idx = CostTableLookup<MVT>(NEONShuffleTbl, array_lengthof(NEONShuffleTbl),
+                                 ISD::VECTOR_SHUFFLE, LT.second);
+  if (Idx == -1)
+    return TargetTransformInfo::getShuffleCost(Kind, Tp, Index, SubTp);
+
+  return LT.first * NEONShuffleTbl[Idx].Cost;
+}
diff --git a/lib/Target/ARM/AsmParser/ARMAsmLexer.cpp b/lib/Target/ARM/AsmParser/ARMAsmLexer.cpp
deleted file mode 100644
index fda8536fcf6b..000000000000
--- a/lib/Target/ARM/AsmParser/ARMAsmLexer.cpp
+++ /dev/null
@@ -1,138 +0,0 @@
-//===-- ARMAsmLexer.cpp - Tokenize ARM assembly to AsmTokens --------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#include "MCTargetDesc/ARMBaseInfo.h"
-
-#include "llvm/MC/MCAsmInfo.h"
-#include "llvm/MC/MCParser/MCAsmLexer.h"
-#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
-#include "llvm/MC/MCRegisterInfo.h"
-#include "llvm/MC/MCTargetAsmLexer.h"
-
-#include "llvm/Support/TargetRegistry.h"
-
-#include "llvm/ADT/StringSwitch.h"
-
-#include <string>
-#include <map>
-
-using namespace llvm;
-
-namespace {
-
-class ARMBaseAsmLexer : public MCTargetAsmLexer {
-  const MCAsmInfo &AsmInfo;
-
-  const AsmToken &lexDefinite() {
-    return getLexer()->Lex();
-  }
-
-  AsmToken LexTokenUAL();
-protected:
-  typedef std::map <std::string, unsigned> rmap_ty;
-
-  rmap_ty RegisterMap;
-
-  void InitRegisterMap(const MCRegisterInfo *info) {
-    unsigned numRegs = info->getNumRegs();
-
-    for (unsigned i = 0; i < numRegs; ++i) {
-      const char *regName = info->getName(i);
-      if (regName)
-        RegisterMap[regName] = i;
-    }
-  }
-
-  unsigned MatchRegisterName(StringRef Name) {
-    rmap_ty::iterator iter = RegisterMap.find(Name.str());
-    if (iter != RegisterMap.end())
-      return iter->second;
-    else
-      return 0;
-  }
-
-  AsmToken LexToken() {
-    if (!Lexer) {
-      SetError(SMLoc(), "No MCAsmLexer installed");
-      return AsmToken(AsmToken::Error, "", 0);
-    }
-
-    switch (AsmInfo.getAssemblerDialect()) {
-    default:
-      SetError(SMLoc(), "Unhandled dialect");
-      return AsmToken(AsmToken::Error, "", 0);
-    case 0:
-      return LexTokenUAL();
-    }
-  }
-public:
-  ARMBaseAsmLexer(const Target &T, const MCAsmInfo &MAI)
-    : MCTargetAsmLexer(T), AsmInfo(MAI) {
-  }
-};
-
-class ARMAsmLexer : public ARMBaseAsmLexer {
-public:
-  ARMAsmLexer(const Target &T, const MCRegisterInfo &MRI, const MCAsmInfo &MAI)
-    : ARMBaseAsmLexer(T, MAI) {
-    InitRegisterMap(&MRI);
-  }
-};
-
-class ThumbAsmLexer : public ARMBaseAsmLexer {
-public:
-  ThumbAsmLexer(const Target &T, const MCRegisterInfo &MRI,const MCAsmInfo &MAI)
-    : ARMBaseAsmLexer(T, MAI) {
-    InitRegisterMap(&MRI);
-  }
-};
-
-} // end anonymous namespace
-
-AsmToken ARMBaseAsmLexer::LexTokenUAL() {
-  const AsmToken &lexedToken = lexDefinite();
-
-  switch (lexedToken.getKind()) {
-  default: break;
-  case AsmToken::Error:
-    SetError(Lexer->getErrLoc(), Lexer->getErr());
-    break;
-  case AsmToken::Identifier: {
-    std::string lowerCase = lexedToken.getString().lower();
-
-    unsigned regID = MatchRegisterName(lowerCase);
-    // Check for register aliases.
-    //   r13 -> sp
-    //   r14 -> lr
-    //   r15 -> pc
-    //   ip  -> r12
-    //   FIXME: Some assemblers support lots of others. Do we want them all?
-    if (!regID) {
-      regID = StringSwitch<unsigned>(lowerCase)
-        .Case("r13", ARM::SP)
-        .Case("r14", ARM::LR)
-        .Case("r15", ARM::PC)
-        .Case("ip", ARM::R12)
-        .Default(0);
-    }
-
-    if (regID)
-      return AsmToken(AsmToken::Register,
-                      lexedToken.getString(),
-                      static_cast<int64_t>(regID));
-  }
-  }
-
-  return AsmToken(lexedToken);
-}
-
-extern "C" void LLVMInitializeARMAsmLexer() {
-  RegisterMCAsmLexer<ARMAsmLexer> X(TheARMTarget);
-  RegisterMCAsmLexer<ThumbAsmLexer> Y(TheThumbTarget);
-}
diff --git a/lib/Target/ARM/AsmParser/ARMAsmParser.cpp b/lib/Target/ARM/AsmParser/ARMAsmParser.cpp
index c61e3bd99d77..ed7b7ec9d2cd 100644
--- a/lib/Target/ARM/AsmParser/ARMAsmParser.cpp
+++ b/lib/Target/ARM/AsmParser/ARMAsmParser.cpp
@@ -7,31 +7,34 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "MCTargetDesc/ARMBaseInfo.h"
+#include "llvm/MC/MCTargetAsmParser.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
+#include "MCTargetDesc/ARMBaseInfo.h"
 #include "MCTargetDesc/ARMMCExpr.h"
-#include "llvm/MC/MCParser/MCAsmLexer.h"
-#include "llvm/MC/MCParser/MCAsmParser.h"
-#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/OwningPtr.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/StringSwitch.h"
+#include "llvm/ADT/Twine.h"
 #include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCAssembler.h"
 #include "llvm/MC/MCContext.h"
-#include "llvm/MC/MCStreamer.h"
+#include "llvm/MC/MCELFStreamer.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrDesc.h"
+#include "llvm/MC/MCParser/MCAsmLexer.h"
+#include "llvm/MC/MCParser/MCAsmParser.h"
+#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
 #include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSubtargetInfo.h"
-#include "llvm/MC/MCTargetAsmParser.h"
+#include "llvm/Support/ELF.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/SourceMgr.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/OwningPtr.h"
-#include "llvm/ADT/STLExtras.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/StringSwitch.h"
-#include "llvm/ADT/Twine.h"
 
 using namespace llvm;
 
@@ -178,7 +181,8 @@ class ARMAsmParser : public MCTargetAsmParser {
   OperandMatchResultTy parseAM3Offset(SmallVectorImpl<MCParsedAsmOperand*>&);
   OperandMatchResultTy parseFPImm(SmallVectorImpl<MCParsedAsmOperand*>&);
   OperandMatchResultTy parseVectorList(SmallVectorImpl<MCParsedAsmOperand*>&);
-  OperandMatchResultTy parseVectorLane(VectorLaneTy &LaneKind, unsigned &Index);
+  OperandMatchResultTy parseVectorLane(VectorLaneTy &LaneKind, unsigned &Index,
+                                       SMLoc &EndLoc);
 
   // Asm Match Converter Methods
   void cvtT2LdrdPre(MCInst &Inst, const SmallVectorImpl<MCParsedAsmOperand*> &);
@@ -249,6 +253,13 @@ public:
 
     // Not in an ITBlock to start with.
     ITState.CurPosition = ~0U;
+
+    // Set ELF header flags.
+    // FIXME: This should eventually end up somewhere else where more
+    // intelligent flag decisions can be made. For now we are just maintaining
+    // the statu/parseDirects quo for ARM and setting EF_ARM_EABI_VER5 as the default.
+    if (MCELFStreamer *MES = dyn_cast<MCELFStreamer>(&Parser.getStreamer()))
+      MES->getAssembler().setELFHeaderEFlags(ELF::EF_ARM_EABI_VER5);
   }
 
   // Implementation of the MCTargetAsmParser interface:
@@ -258,6 +269,7 @@ public:
                         SmallVectorImpl<MCParsedAsmOperand*> &Operands);
   bool ParseDirective(AsmToken DirectiveID);
 
+  unsigned validateTargetOperandClass(MCParsedAsmOperand *Op, unsigned Kind);
   unsigned checkTargetMatchPredicate(MCInst &Inst);
 
   bool MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
@@ -270,7 +282,7 @@ public:
 namespace {
 
 /// ARMOperand - Instances of this class represent a parsed ARM machine
-/// instruction.
+/// operand.
 class ARMOperand : public MCParsedAsmOperand {
   enum KindTy {
     k_CondCode,
@@ -304,103 +316,127 @@ class ARMOperand : public MCParsedAsmOperand {
   SMLoc StartLoc, EndLoc;
   SmallVector<unsigned, 8> Registers;
 
+  struct CCOp {
+    ARMCC::CondCodes Val;
+  };
+
+  struct CopOp {
+    unsigned Val;
+  };
+
+  struct CoprocOptionOp {
+    unsigned Val;
+  };
+
+  struct ITMaskOp {
+    unsigned Mask:4;
+  };
+
+  struct MBOptOp {
+    ARM_MB::MemBOpt Val;
+  };
+
+  struct IFlagsOp {
+    ARM_PROC::IFlags Val;
+  };
+
+  struct MMaskOp {
+    unsigned Val;
+  };
+
+  struct TokOp {
+    const char *Data;
+    unsigned Length;
+  };
+
+  struct RegOp {
+    unsigned RegNum;
+  };
+
+  // A vector register list is a sequential list of 1 to 4 registers.
+  struct VectorListOp {
+    unsigned RegNum;
+    unsigned Count;
+    unsigned LaneIndex;
+    bool isDoubleSpaced;
+  };
+
+  struct VectorIndexOp {
+    unsigned Val;
+  };
+
+  struct ImmOp {
+    const MCExpr *Val;
+  };
+
+  /// Combined record for all forms of ARM address expressions.
+  struct MemoryOp {
+    unsigned BaseRegNum;
+    // Offset is in OffsetReg or OffsetImm. If both are zero, no offset
+    // was specified.
+    const MCConstantExpr *OffsetImm;  // Offset immediate value
+    unsigned OffsetRegNum;    // Offset register num, when OffsetImm == NULL
+    ARM_AM::ShiftOpc ShiftType; // Shift type for OffsetReg
+    unsigned ShiftImm;        // shift for OffsetReg.
+    unsigned Alignment;       // 0 = no alignment specified
+    // n = alignment in bytes (2, 4, 8, 16, or 32)
+    unsigned isNegative : 1;  // Negated OffsetReg? (~'U' bit)
+  };
+
+  struct PostIdxRegOp {
+    unsigned RegNum;
+    bool isAdd;
+    ARM_AM::ShiftOpc ShiftTy;
+    unsigned ShiftImm;
+  };
+
+  struct ShifterImmOp {
+    bool isASR;
+    unsigned Imm;
+  };
+
+  struct RegShiftedRegOp {
+    ARM_AM::ShiftOpc ShiftTy;
+    unsigned SrcReg;
+    unsigned ShiftReg;
+    unsigned ShiftImm;
+  };
+
+  struct RegShiftedImmOp {
+    ARM_AM::ShiftOpc ShiftTy;
+    unsigned SrcReg;
+    unsigned ShiftImm;
+  };
+
+  struct RotImmOp {
+    unsigned Imm;
+  };
+
+  struct BitfieldOp {
+    unsigned LSB;
+    unsigned Width;
+  };
+
   union {
-    struct {
-      ARMCC::CondCodes Val;
-    } CC;
-
-    struct {
-      unsigned Val;
-    } Cop;
-
-    struct {
-      unsigned Val;
-    } CoprocOption;
-
-    struct {
-      unsigned Mask:4;
-    } ITMask;
-
-    struct {
-      ARM_MB::MemBOpt Val;
-    } MBOpt;
-
-    struct {
-      ARM_PROC::IFlags Val;
-    } IFlags;
-
-    struct {
-      unsigned Val;
-    } MMask;
-
-    struct {
-      const char *Data;
-      unsigned Length;
-    } Tok;
-
-    struct {
-      unsigned RegNum;
-    } Reg;
-
-    // A vector register list is a sequential list of 1 to 4 registers.
-    struct {
-      unsigned RegNum;
-      unsigned Count;
-      unsigned LaneIndex;
-      bool isDoubleSpaced;
-    } VectorList;
-
-    struct {
-      unsigned Val;
-    } VectorIndex;
-
-    struct {
-      const MCExpr *Val;
-    } Imm;
-
-    /// Combined record for all forms of ARM address expressions.
-    struct {
-      unsigned BaseRegNum;
-      // Offset is in OffsetReg or OffsetImm. If both are zero, no offset
-      // was specified.
-      const MCConstantExpr *OffsetImm;  // Offset immediate value
-      unsigned OffsetRegNum;    // Offset register num, when OffsetImm == NULL
-      ARM_AM::ShiftOpc ShiftType; // Shift type for OffsetReg
-      unsigned ShiftImm;        // shift for OffsetReg.
-      unsigned Alignment;       // 0 = no alignment specified
-                                // n = alignment in bytes (2, 4, 8, 16, or 32)
-      unsigned isNegative : 1;  // Negated OffsetReg? (~'U' bit)
-    } Memory;
-
-    struct {
-      unsigned RegNum;
-      bool isAdd;
-      ARM_AM::ShiftOpc ShiftTy;
-      unsigned ShiftImm;
-    } PostIdxReg;
-
-    struct {
-      bool isASR;
-      unsigned Imm;
-    } ShifterImm;
-    struct {
-      ARM_AM::ShiftOpc ShiftTy;
-      unsigned SrcReg;
-      unsigned ShiftReg;
-      unsigned ShiftImm;
-    } RegShiftedReg;
-    struct {
-      ARM_AM::ShiftOpc ShiftTy;
-      unsigned SrcReg;
-      unsigned ShiftImm;
-    } RegShiftedImm;
-    struct {
-      unsigned Imm;
-    } RotImm;
-    struct {
-      unsigned LSB;
-      unsigned Width;
-    } Bitfield;
+    struct CCOp CC;
+    struct CopOp Cop;
+    struct CoprocOptionOp CoprocOption;
+    struct MBOptOp MBOpt;
+    struct ITMaskOp ITMask;
+    struct IFlagsOp IFlags;
+    struct MMaskOp MMask;
+    struct TokOp Tok;
+    struct RegOp Reg;
+    struct VectorListOp VectorList;
+    struct VectorIndexOp VectorIndex;
+    struct ImmOp Imm;
+    struct MemoryOp Memory;
+    struct PostIdxRegOp PostIdxReg;
+    struct ShifterImmOp ShifterImm;
+    struct RegShiftedRegOp RegShiftedReg;
+    struct RegShiftedImmOp RegShiftedImm;
+    struct RotImmOp RotImm;
+    struct BitfieldOp Bitfield;
   };
 
   ARMOperand(KindTy K) : MCParsedAsmOperand(), Kind(K) {}
@@ -2450,8 +2486,8 @@ static unsigned MatchRegisterName(StringRef Name);
 bool ARMAsmParser::ParseRegister(unsigned &RegNo,
                                  SMLoc &StartLoc, SMLoc &EndLoc) {
   StartLoc = Parser.getTok().getLoc();
+  EndLoc = Parser.getTok().getEndLoc();
   RegNo = tryParseRegister();
-  EndLoc = Parser.getTok().getLoc();
 
   return (RegNo == (unsigned)-1);
 }
@@ -2540,6 +2576,8 @@ int ARMAsmParser::tryParseShiftRegister(
   if (!PrevOp->isReg())
     return Error(PrevOp->getStartLoc(), "shift must be of a register");
   int SrcReg = PrevOp->getReg();
+
+  SMLoc EndLoc;
   int64_t Imm = 0;
   int ShiftReg = 0;
   if (ShiftTy == ARM_AM::rrx) {
@@ -2554,7 +2592,7 @@ int ARMAsmParser::tryParseShiftRegister(
       Parser.Lex(); // Eat hash.
       SMLoc ImmLoc = Parser.getTok().getLoc();
       const MCExpr *ShiftExpr = 0;
-      if (getParser().ParseExpression(ShiftExpr)) {
+      if (getParser().parseExpression(ShiftExpr, EndLoc)) {
         Error(ImmLoc, "invalid immediate shift value");
         return -1;
       }
@@ -2579,8 +2617,9 @@ int ARMAsmParser::tryParseShiftRegister(
       if (Imm == 0)
         ShiftTy = ARM_AM::lsl;
     } else if (Parser.getTok().is(AsmToken::Identifier)) {
-      ShiftReg = tryParseRegister();
       SMLoc L = Parser.getTok().getLoc();
+      EndLoc = Parser.getTok().getEndLoc();
+      ShiftReg = tryParseRegister();
       if (ShiftReg == -1) {
         Error (L, "expected immediate or register in shift operand");
         return -1;
@@ -2595,10 +2634,10 @@ int ARMAsmParser::tryParseShiftRegister(
   if (ShiftReg && ShiftTy != ARM_AM::rrx)
     Operands.push_back(ARMOperand::CreateShiftedRegister(ShiftTy, SrcReg,
                                                          ShiftReg, Imm,
-                                               S, Parser.getTok().getLoc()));
+                                                         S, EndLoc));
   else
     Operands.push_back(ARMOperand::CreateShiftedImmediate(ShiftTy, SrcReg, Imm,
-                                               S, Parser.getTok().getLoc()));
+                                                          S, EndLoc));
 
   return 0;
 }
@@ -2612,12 +2651,13 @@ int ARMAsmParser::tryParseShiftRegister(
 /// parse for a specific register type.
 bool ARMAsmParser::
 tryParseRegisterWithWriteBack(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
-  SMLoc S = Parser.getTok().getLoc();
+  const AsmToken &RegTok = Parser.getTok();
   int RegNo = tryParseRegister();
   if (RegNo == -1)
     return true;
 
-  Operands.push_back(ARMOperand::CreateReg(RegNo, S, Parser.getTok().getLoc()));
+  Operands.push_back(ARMOperand::CreateReg(RegNo, RegTok.getLoc(),
+                                           RegTok.getEndLoc()));
 
   const AsmToken &ExclaimTok = Parser.getTok();
   if (ExclaimTok.is(AsmToken::Exclaim)) {
@@ -2635,16 +2675,16 @@ tryParseRegisterWithWriteBack(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     Parser.Lex(); // Eat left bracket token.
 
     const MCExpr *ImmVal;
-    if (getParser().ParseExpression(ImmVal))
+    if (getParser().parseExpression(ImmVal))
       return true;
     const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(ImmVal);
     if (!MCE)
       return TokError("immediate value expected for vector index");
 
-    SMLoc E = Parser.getTok().getLoc();
     if (Parser.getTok().isNot(AsmToken::RBrac))
-      return Error(E, "']' expected");
+      return Error(Parser.getTok().getLoc(), "']' expected");
 
+    SMLoc E = Parser.getTok().getEndLoc();
     Parser.Lex(); // Eat right bracket token.
 
     Operands.push_back(ARMOperand::CreateVectorIndex(MCE->getValue(),
@@ -2780,7 +2820,7 @@ parseCoprocOptionOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
 
   const MCExpr *Expr;
   SMLoc Loc = Parser.getTok().getLoc();
-  if (getParser().ParseExpression(Expr)) {
+  if (getParser().parseExpression(Expr)) {
     Error(Loc, "illegal expression");
     return MatchOperand_ParseFail;
   }
@@ -2794,7 +2834,7 @@ parseCoprocOptionOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   // Check for and consume the closing '}'
   if (Parser.getTok().isNot(AsmToken::RCurly))
     return MatchOperand_ParseFail;
-  SMLoc E = Parser.getTok().getLoc();
+  SMLoc E = Parser.getTok().getEndLoc();
   Parser.Lex(); // Eat the '}'
 
   Operands.push_back(ARMOperand::CreateCoprocOption(Val, S, E));
@@ -2891,10 +2931,10 @@ parseRegisterList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
          Parser.getTok().is(AsmToken::Minus)) {
     if (Parser.getTok().is(AsmToken::Minus)) {
       Parser.Lex(); // Eat the minus.
-      SMLoc EndLoc = Parser.getTok().getLoc();
+      SMLoc AfterMinusLoc = Parser.getTok().getLoc();
       int EndReg = tryParseRegister();
       if (EndReg == -1)
-        return Error(EndLoc, "register expected");
+        return Error(AfterMinusLoc, "register expected");
       // Allow Q regs and just interpret them as the two D sub-registers.
       if (ARMMCRegisterClasses[ARM::QPRRegClassID].contains(EndReg))
         EndReg = getDRegFromQReg(EndReg) + 1;
@@ -2904,10 +2944,10 @@ parseRegisterList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
         continue;
       // The register must be in the same register class as the first.
       if (!RC->contains(EndReg))
-        return Error(EndLoc, "invalid register in register list");
+        return Error(AfterMinusLoc, "invalid register in register list");
       // Ranges must go from low to high.
       if (MRI->getEncodingValue(Reg) > MRI->getEncodingValue(EndReg))
-        return Error(EndLoc, "bad range in register list");
+        return Error(AfterMinusLoc, "bad range in register list");
 
       // Add all the registers in the range to the register list.
       while (Reg != EndReg) {
@@ -2955,9 +2995,9 @@ parseRegisterList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
       Registers.push_back(std::pair<unsigned, SMLoc>(++Reg, RegLoc));
   }
 
-  SMLoc E = Parser.getTok().getLoc();
   if (Parser.getTok().isNot(AsmToken::RCurly))
-    return Error(E, "'}' expected");
+    return Error(Parser.getTok().getLoc(), "'}' expected");
+  SMLoc E = Parser.getTok().getEndLoc();
   Parser.Lex(); // Eat '}' token.
 
   // Push the register list operand.
@@ -2974,13 +3014,14 @@ parseRegisterList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
 
 // Helper function to parse the lane index for vector lists.
 ARMAsmParser::OperandMatchResultTy ARMAsmParser::
-parseVectorLane(VectorLaneTy &LaneKind, unsigned &Index) {
+parseVectorLane(VectorLaneTy &LaneKind, unsigned &Index, SMLoc &EndLoc) {
   Index = 0; // Always return a defined index value.
   if (Parser.getTok().is(AsmToken::LBrac)) {
     Parser.Lex(); // Eat the '['.
     if (Parser.getTok().is(AsmToken::RBrac)) {
       // "Dn[]" is the 'all lanes' syntax.
       LaneKind = AllLanes;
+      EndLoc = Parser.getTok().getEndLoc();
       Parser.Lex(); // Eat the ']'.
       return MatchOperand_Success;
     }
@@ -2992,7 +3033,7 @@ parseVectorLane(VectorLaneTy &LaneKind, unsigned &Index) {
 
     const MCExpr *LaneIndex;
     SMLoc Loc = Parser.getTok().getLoc();
-    if (getParser().ParseExpression(LaneIndex)) {
+    if (getParser().parseExpression(LaneIndex)) {
       Error(Loc, "illegal expression");
       return MatchOperand_ParseFail;
     }
@@ -3005,6 +3046,7 @@ parseVectorLane(VectorLaneTy &LaneKind, unsigned &Index) {
       Error(Parser.getTok().getLoc(), "']' expected");
       return MatchOperand_ParseFail;
     }
+    EndLoc = Parser.getTok().getEndLoc();
     Parser.Lex(); // Eat the ']'.
     int64_t Val = CE->getValue();
 
@@ -3031,21 +3073,19 @@ parseVectorList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   // (without encosing curly braces) as a single or double entry list,
   // respectively.
   if (Parser.getTok().is(AsmToken::Identifier)) {
+    SMLoc E = Parser.getTok().getEndLoc();
     int Reg = tryParseRegister();
     if (Reg == -1)
       return MatchOperand_NoMatch;
-    SMLoc E = Parser.getTok().getLoc();
     if (ARMMCRegisterClasses[ARM::DPRRegClassID].contains(Reg)) {
-      OperandMatchResultTy Res = parseVectorLane(LaneKind, LaneIndex);
+      OperandMatchResultTy Res = parseVectorLane(LaneKind, LaneIndex, E);
       if (Res != MatchOperand_Success)
         return Res;
       switch (LaneKind) {
       case NoLanes:
-        E = Parser.getTok().getLoc();
         Operands.push_back(ARMOperand::CreateVectorList(Reg, 1, false, S, E));
         break;
       case AllLanes:
-        E = Parser.getTok().getLoc();
         Operands.push_back(ARMOperand::CreateVectorListAllLanes(Reg, 1, false,
                                                                 S, E));
         break;
@@ -3059,18 +3099,16 @@ parseVectorList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     }
     if (ARMMCRegisterClasses[ARM::QPRRegClassID].contains(Reg)) {
       Reg = getDRegFromQReg(Reg);
-      OperandMatchResultTy Res = parseVectorLane(LaneKind, LaneIndex);
+      OperandMatchResultTy Res = parseVectorLane(LaneKind, LaneIndex, E);
       if (Res != MatchOperand_Success)
         return Res;
       switch (LaneKind) {
       case NoLanes:
-        E = Parser.getTok().getLoc();
         Reg = MRI->getMatchingSuperReg(Reg, ARM::dsub_0,
                                    &ARMMCRegisterClasses[ARM::DPairRegClassID]);
         Operands.push_back(ARMOperand::CreateVectorList(Reg, 2, false, S, E));
         break;
       case AllLanes:
-        E = Parser.getTok().getLoc();
         Reg = MRI->getMatchingSuperReg(Reg, ARM::dsub_0,
                                    &ARMMCRegisterClasses[ARM::DPairRegClassID]);
         Operands.push_back(ARMOperand::CreateVectorListAllLanes(Reg, 2, false,
@@ -3111,7 +3149,9 @@ parseVectorList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     ++Reg;
     ++Count;
   }
-  if (parseVectorLane(LaneKind, LaneIndex) != MatchOperand_Success)
+
+  SMLoc E;
+  if (parseVectorLane(LaneKind, LaneIndex, E) != MatchOperand_Success)
     return MatchOperand_ParseFail;
 
   while (Parser.getTok().is(AsmToken::Comma) ||
@@ -3125,10 +3165,10 @@ parseVectorList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
         return MatchOperand_ParseFail;
       }
       Parser.Lex(); // Eat the minus.
-      SMLoc EndLoc = Parser.getTok().getLoc();
+      SMLoc AfterMinusLoc = Parser.getTok().getLoc();
       int EndReg = tryParseRegister();
       if (EndReg == -1) {
-        Error(EndLoc, "register expected");
+        Error(AfterMinusLoc, "register expected");
         return MatchOperand_ParseFail;
       }
       // Allow Q regs and just interpret them as the two D sub-registers.
@@ -3140,24 +3180,24 @@ parseVectorList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
         continue;
       // The register must be in the same register class as the first.
       if (!ARMMCRegisterClasses[ARM::DPRRegClassID].contains(EndReg)) {
-        Error(EndLoc, "invalid register in register list");
+        Error(AfterMinusLoc, "invalid register in register list");
         return MatchOperand_ParseFail;
       }
       // Ranges must go from low to high.
       if (Reg > EndReg) {
-        Error(EndLoc, "bad range in register list");
+        Error(AfterMinusLoc, "bad range in register list");
         return MatchOperand_ParseFail;
       }
       // Parse the lane specifier if present.
       VectorLaneTy NextLaneKind;
       unsigned NextLaneIndex;
-      if (parseVectorLane(NextLaneKind, NextLaneIndex) != MatchOperand_Success)
+      if (parseVectorLane(NextLaneKind, NextLaneIndex, E) !=
+          MatchOperand_Success)
         return MatchOperand_ParseFail;
       if (NextLaneKind != LaneKind || LaneIndex != NextLaneIndex) {
-        Error(EndLoc, "mismatched lane index in register list");
+        Error(AfterMinusLoc, "mismatched lane index in register list");
         return MatchOperand_ParseFail;
       }
-      EndLoc = Parser.getTok().getLoc();
 
       // Add all the registers in the range to the register list.
       Count += EndReg - Reg;
@@ -3196,11 +3236,12 @@ parseVectorList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
       // Parse the lane specifier if present.
       VectorLaneTy NextLaneKind;
       unsigned NextLaneIndex;
-      SMLoc EndLoc = Parser.getTok().getLoc();
-      if (parseVectorLane(NextLaneKind, NextLaneIndex) != MatchOperand_Success)
+      SMLoc LaneLoc = Parser.getTok().getLoc();
+      if (parseVectorLane(NextLaneKind, NextLaneIndex, E) !=
+          MatchOperand_Success)
         return MatchOperand_ParseFail;
       if (NextLaneKind != LaneKind || LaneIndex != NextLaneIndex) {
-        Error(EndLoc, "mismatched lane index in register list");
+        Error(LaneLoc, "mismatched lane index in register list");
         return MatchOperand_ParseFail;
       }
       continue;
@@ -3221,7 +3262,7 @@ parseVectorList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     VectorLaneTy NextLaneKind;
     unsigned NextLaneIndex;
     SMLoc EndLoc = Parser.getTok().getLoc();
-    if (parseVectorLane(NextLaneKind, NextLaneIndex) != MatchOperand_Success)
+    if (parseVectorLane(NextLaneKind, NextLaneIndex, E) != MatchOperand_Success)
       return MatchOperand_ParseFail;
     if (NextLaneKind != LaneKind || LaneIndex != NextLaneIndex) {
       Error(EndLoc, "mismatched lane index in register list");
@@ -3229,11 +3270,11 @@ parseVectorList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     }
   }
 
-  SMLoc E = Parser.getTok().getLoc();
   if (Parser.getTok().isNot(AsmToken::RCurly)) {
-    Error(E, "'}' expected");
+    Error(Parser.getTok().getLoc(), "'}' expected");
     return MatchOperand_ParseFail;
   }
+  E = Parser.getTok().getEndLoc();
   Parser.Lex(); // Eat '}' token.
 
   switch (LaneKind) {
@@ -3310,7 +3351,7 @@ parseMemBarrierOptOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     SMLoc Loc = Parser.getTok().getLoc();
 
     const MCExpr *MemBarrierID;
-    if (getParser().ParseExpression(MemBarrierID)) {
+    if (getParser().parseExpression(MemBarrierID)) {
       Error(Loc, "illegal expression");
       return MatchOperand_ParseFail;
     }
@@ -3525,7 +3566,8 @@ parsePKHImm(SmallVectorImpl<MCParsedAsmOperand*> &Operands, StringRef Op,
 
   const MCExpr *ShiftAmount;
   SMLoc Loc = Parser.getTok().getLoc();
-  if (getParser().ParseExpression(ShiftAmount)) {
+  SMLoc EndLoc;
+  if (getParser().parseExpression(ShiftAmount, EndLoc)) {
     Error(Loc, "illegal expression");
     return MatchOperand_ParseFail;
   }
@@ -3540,7 +3582,7 @@ parsePKHImm(SmallVectorImpl<MCParsedAsmOperand*> &Operands, StringRef Op,
     return MatchOperand_ParseFail;
   }
 
-  Operands.push_back(ARMOperand::CreateImm(CE, Loc, Parser.getTok().getLoc()));
+  Operands.push_back(ARMOperand::CreateImm(CE, Loc, EndLoc));
 
   return MatchOperand_Success;
 }
@@ -3550,7 +3592,7 @@ parseSetEndImm(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   const AsmToken &Tok = Parser.getTok();
   SMLoc S = Tok.getLoc();
   if (Tok.isNot(AsmToken::Identifier)) {
-    Error(Tok.getLoc(), "'be' or 'le' operand expected");
+    Error(S, "'be' or 'le' operand expected");
     return MatchOperand_ParseFail;
   }
   int Val = StringSwitch<int>(Tok.getString())
@@ -3560,12 +3602,12 @@ parseSetEndImm(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   Parser.Lex(); // Eat the token.
 
   if (Val == -1) {
-    Error(Tok.getLoc(), "'be' or 'le' operand expected");
+    Error(S, "'be' or 'le' operand expected");
     return MatchOperand_ParseFail;
   }
   Operands.push_back(ARMOperand::CreateImm(MCConstantExpr::Create(Val,
                                                                   getContext()),
-                                           S, Parser.getTok().getLoc()));
+                                           S, Tok.getEndLoc()));
   return MatchOperand_Success;
 }
 
@@ -3601,16 +3643,17 @@ parseShifterImm(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     return MatchOperand_ParseFail;
   }
   Parser.Lex(); // Eat hash token.
+  SMLoc ExLoc = Parser.getTok().getLoc();
 
   const MCExpr *ShiftAmount;
-  SMLoc E = Parser.getTok().getLoc();
-  if (getParser().ParseExpression(ShiftAmount)) {
-    Error(E, "malformed shift expression");
+  SMLoc EndLoc;
+  if (getParser().parseExpression(ShiftAmount, EndLoc)) {
+    Error(ExLoc, "malformed shift expression");
     return MatchOperand_ParseFail;
   }
   const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(ShiftAmount);
   if (!CE) {
-    Error(E, "shift amount must be an immediate");
+    Error(ExLoc, "shift amount must be an immediate");
     return MatchOperand_ParseFail;
   }
 
@@ -3618,25 +3661,24 @@ parseShifterImm(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   if (isASR) {
     // Shift amount must be in [1,32]
     if (Val < 1 || Val > 32) {
-      Error(E, "'asr' shift amount must be in range [1,32]");
+      Error(ExLoc, "'asr' shift amount must be in range [1,32]");
       return MatchOperand_ParseFail;
     }
     // asr #32 encoded as asr #0, but is not allowed in Thumb2 mode.
     if (isThumb() && Val == 32) {
-      Error(E, "'asr #32' shift amount not allowed in Thumb mode");
+      Error(ExLoc, "'asr #32' shift amount not allowed in Thumb mode");
       return MatchOperand_ParseFail;
     }
     if (Val == 32) Val = 0;
   } else {
     // Shift amount must be in [1,32]
     if (Val < 0 || Val > 31) {
-      Error(E, "'lsr' shift amount must be in range [0,31]");
+      Error(ExLoc, "'lsr' shift amount must be in range [0,31]");
       return MatchOperand_ParseFail;
     }
   }
 
-  E = Parser.getTok().getLoc();
-  Operands.push_back(ARMOperand::CreateShifterImm(isASR, Val, S, E));
+  Operands.push_back(ARMOperand::CreateShifterImm(isASR, Val, S, EndLoc));
 
   return MatchOperand_Success;
 }
@@ -3662,16 +3704,17 @@ parseRotImm(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     return MatchOperand_ParseFail;
   }
   Parser.Lex(); // Eat hash token.
+  SMLoc ExLoc = Parser.getTok().getLoc();
 
   const MCExpr *ShiftAmount;
-  SMLoc E = Parser.getTok().getLoc();
-  if (getParser().ParseExpression(ShiftAmount)) {
-    Error(E, "malformed rotate expression");
+  SMLoc EndLoc;
+  if (getParser().parseExpression(ShiftAmount, EndLoc)) {
+    Error(ExLoc, "malformed rotate expression");
     return MatchOperand_ParseFail;
   }
   const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(ShiftAmount);
   if (!CE) {
-    Error(E, "rotate amount must be an immediate");
+    Error(ExLoc, "rotate amount must be an immediate");
     return MatchOperand_ParseFail;
   }
 
@@ -3680,12 +3723,11 @@ parseRotImm(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   // normally, zero is represented in asm by omitting the rotate operand
   // entirely.
   if (Val != 8 && Val != 16 && Val != 24 && Val != 0) {
-    Error(E, "'ror' rotate amount must be 8, 16, or 24");
+    Error(ExLoc, "'ror' rotate amount must be 8, 16, or 24");
     return MatchOperand_ParseFail;
   }
 
-  E = Parser.getTok().getLoc();
-  Operands.push_back(ARMOperand::CreateRotImm(Val, S, E));
+  Operands.push_back(ARMOperand::CreateRotImm(Val, S, EndLoc));
 
   return MatchOperand_Success;
 }
@@ -3703,7 +3745,7 @@ parseBitfield(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
 
   const MCExpr *LSBExpr;
   SMLoc E = Parser.getTok().getLoc();
-  if (getParser().ParseExpression(LSBExpr)) {
+  if (getParser().parseExpression(LSBExpr)) {
     Error(E, "malformed immediate expression");
     return MatchOperand_ParseFail;
   }
@@ -3735,7 +3777,8 @@ parseBitfield(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   Parser.Lex(); // Eat hash token.
 
   const MCExpr *WidthExpr;
-  if (getParser().ParseExpression(WidthExpr)) {
+  SMLoc EndLoc;
+  if (getParser().parseExpression(WidthExpr, EndLoc)) {
     Error(E, "malformed immediate expression");
     return MatchOperand_ParseFail;
   }
@@ -3751,9 +3794,8 @@ parseBitfield(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     Error(E, "'width' operand must be in the range [1,32-lsb]");
     return MatchOperand_ParseFail;
   }
-  E = Parser.getTok().getLoc();
 
-  Operands.push_back(ARMOperand::CreateBitfield(LSB, Width, S, E));
+  Operands.push_back(ARMOperand::CreateBitfield(LSB, Width, S, EndLoc));
 
   return MatchOperand_Success;
 }
@@ -3772,7 +3814,6 @@ parsePostIdxReg(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   SMLoc S = Tok.getLoc();
   bool haveEaten = false;
   bool isAdd = true;
-  int Reg = -1;
   if (Tok.is(AsmToken::Plus)) {
     Parser.Lex(); // Eat the '+' token.
     haveEaten = true;
@@ -3781,15 +3822,15 @@ parsePostIdxReg(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     isAdd = false;
     haveEaten = true;
   }
-  if (Parser.getTok().is(AsmToken::Identifier))
-    Reg = tryParseRegister();
+
+  SMLoc E = Parser.getTok().getEndLoc();
+  int Reg = tryParseRegister();
   if (Reg == -1) {
     if (!haveEaten)
       return MatchOperand_NoMatch;
     Error(Parser.getTok().getLoc(), "register expected");
     return MatchOperand_ParseFail;
   }
-  SMLoc E = Parser.getTok().getLoc();
 
   ARM_AM::ShiftOpc ShiftTy = ARM_AM::no_shift;
   unsigned ShiftImm = 0;
@@ -3797,6 +3838,9 @@ parsePostIdxReg(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     Parser.Lex(); // Eat the ','.
     if (parseMemRegOffsetShift(ShiftTy, ShiftImm))
       return MatchOperand_ParseFail;
+
+    // FIXME: Only approximates end...may include intervening whitespace.
+    E = Parser.getTok().getLoc();
   }
 
   Operands.push_back(ARMOperand::CreatePostIdxReg(Reg, isAdd, ShiftTy,
@@ -3829,14 +3873,14 @@ parseAM3Offset(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     // differently.
     bool isNegative = Parser.getTok().is(AsmToken::Minus);
     const MCExpr *Offset;
-    if (getParser().ParseExpression(Offset))
+    SMLoc E;
+    if (getParser().parseExpression(Offset, E))
       return MatchOperand_ParseFail;
     const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Offset);
     if (!CE) {
       Error(S, "constant expression expected");
       return MatchOperand_ParseFail;
     }
-    SMLoc E = Tok.getLoc();
     // Negative zero is encoded as the flag value INT32_MIN.
     int32_t Val = CE->getValue();
     if (isNegative && Val == 0)
@@ -3851,7 +3895,6 @@ parseAM3Offset(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
 
   bool haveEaten = false;
   bool isAdd = true;
-  int Reg = -1;
   if (Tok.is(AsmToken::Plus)) {
     Parser.Lex(); // Eat the '+' token.
     haveEaten = true;
@@ -3860,18 +3903,18 @@ parseAM3Offset(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     isAdd = false;
     haveEaten = true;
   }
-  if (Parser.getTok().is(AsmToken::Identifier))
-    Reg = tryParseRegister();
+  
+  Tok = Parser.getTok();
+  int Reg = tryParseRegister();
   if (Reg == -1) {
     if (!haveEaten)
       return MatchOperand_NoMatch;
-    Error(Parser.getTok().getLoc(), "register expected");
+    Error(Tok.getLoc(), "register expected");
     return MatchOperand_ParseFail;
   }
-  SMLoc E = Parser.getTok().getLoc();
 
   Operands.push_back(ARMOperand::CreatePostIdxReg(Reg, isAdd, ARM_AM::no_shift,
-                                                  0, S, E));
+                                                  0, S, Tok.getEndLoc()));
 
   return MatchOperand_Success;
 }
@@ -4218,13 +4261,14 @@ parseMemory(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   if (BaseRegNum == -1)
     return Error(BaseRegTok.getLoc(), "register expected");
 
-  // The next token must either be a comma or a closing bracket.
+  // The next token must either be a comma, a colon or a closing bracket.
   const AsmToken &Tok = Parser.getTok();
-  if (!Tok.is(AsmToken::Comma) && !Tok.is(AsmToken::RBrac))
+  if (!Tok.is(AsmToken::Colon) && !Tok.is(AsmToken::Comma) &&
+      !Tok.is(AsmToken::RBrac))
     return Error(Tok.getLoc(), "malformed memory operand");
 
   if (Tok.is(AsmToken::RBrac)) {
-    E = Tok.getLoc();
+    E = Tok.getEndLoc();
     Parser.Lex(); // Eat right bracket token.
 
     Operands.push_back(ARMOperand::CreateMem(BaseRegNum, 0, 0, ARM_AM::no_shift,
@@ -4240,8 +4284,11 @@ parseMemory(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     return false;
   }
 
-  assert(Tok.is(AsmToken::Comma) && "Lost comma in memory operand?!");
-  Parser.Lex(); // Eat the comma.
+  assert((Tok.is(AsmToken::Colon) || Tok.is(AsmToken::Comma)) &&
+         "Lost colon or comma in memory operand?!");
+  if (Tok.is(AsmToken::Comma)) {
+    Parser.Lex(); // Eat the comma.
+  }
 
   // If we have a ':', it's an alignment specifier.
   if (Parser.getTok().is(AsmToken::Colon)) {
@@ -4249,7 +4296,7 @@ parseMemory(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     E = Parser.getTok().getLoc();
 
     const MCExpr *Expr;
-    if (getParser().ParseExpression(Expr))
+    if (getParser().parseExpression(Expr))
      return true;
 
     // The expression has to be a constant. Memory references with relocations
@@ -4272,9 +4319,9 @@ parseMemory(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
     }
 
     // Now we should have the closing ']'
-    E = Parser.getTok().getLoc();
     if (Parser.getTok().isNot(AsmToken::RBrac))
-      return Error(E, "']' expected");
+      return Error(Parser.getTok().getLoc(), "']' expected");
+    E = Parser.getTok().getEndLoc();
     Parser.Lex(); // Eat right bracket token.
 
     // Don't worry about range checking the value here. That's handled by
@@ -4305,7 +4352,7 @@ parseMemory(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
 
     bool isNegative = getParser().getTok().is(AsmToken::Minus);
     const MCExpr *Offset;
-    if (getParser().ParseExpression(Offset))
+    if (getParser().parseExpression(Offset))
      return true;
 
     // The expression has to be a constant. Memory references with relocations
@@ -4321,9 +4368,9 @@ parseMemory(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
       CE = MCConstantExpr::Create(INT32_MIN, getContext());
 
     // Now we should have the closing ']'
-    E = Parser.getTok().getLoc();
     if (Parser.getTok().isNot(AsmToken::RBrac))
-      return Error(E, "']' expected");
+      return Error(Parser.getTok().getLoc(), "']' expected");
+    E = Parser.getTok().getEndLoc();
     Parser.Lex(); // Eat right bracket token.
 
     // Don't worry about range checking the value here. That's handled by
@@ -4367,9 +4414,9 @@ parseMemory(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   }
 
   // Now we should have the closing ']'
-  E = Parser.getTok().getLoc();
   if (Parser.getTok().isNot(AsmToken::RBrac))
-    return Error(E, "']' expected");
+    return Error(Parser.getTok().getLoc(), "']' expected");
+  E = Parser.getTok().getEndLoc();
   Parser.Lex(); // Eat right bracket token.
 
   Operands.push_back(ARMOperand::CreateMem(BaseRegNum, 0, OffsetRegNum,
@@ -4424,7 +4471,7 @@ bool ARMAsmParser::parseMemRegOffsetShift(ARM_AM::ShiftOpc &St,
     Parser.Lex(); // Eat hash token.
 
     const MCExpr *Expr;
-    if (getParser().ParseExpression(Expr))
+    if (getParser().parseExpression(Expr))
       return true;
     // Range check the immediate.
     // lsl, ror: 0 <= imm <= 31
@@ -4453,7 +4500,7 @@ bool ARMAsmParser::parseMemRegOffsetShift(ARM_AM::ShiftOpc &St,
 ARMAsmParser::OperandMatchResultTy ARMAsmParser::
 parseFPImm(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   // Anything that can accept a floating point constant as an operand
-  // needs to go through here, as the regular ParseExpression is
+  // needs to go through here, as the regular parseExpression is
   // integer only.
   //
   // This routine still creates a generic Immediate operand, containing
@@ -4546,20 +4593,26 @@ bool ARMAsmParser::parseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
     Error(Parser.getTok().getLoc(), "unexpected token in operand");
     return true;
   case AsmToken::Identifier: {
-    if (!tryParseRegisterWithWriteBack(Operands))
-      return false;
-    int Res = tryParseShiftRegister(Operands);
-    if (Res == 0) // success
-      return false;
-    else if (Res == -1) // irrecoverable error
-      return true;
-    // If this is VMRS, check for the apsr_nzcv operand.
-    if (Mnemonic == "vmrs" &&
-        Parser.getTok().getString().equals_lower("apsr_nzcv")) {
-      S = Parser.getTok().getLoc();
-      Parser.Lex();
-      Operands.push_back(ARMOperand::CreateToken("APSR_nzcv", S));
-      return false;
+    // If we've seen a branch mnemonic, the next operand must be a label.  This
+    // is true even if the label is a register name.  So "br r1" means branch to
+    // label "r1".
+    bool ExpectLabel = Mnemonic == "b" || Mnemonic == "bl";
+    if (!ExpectLabel) {
+      if (!tryParseRegisterWithWriteBack(Operands))
+        return false;
+      int Res = tryParseShiftRegister(Operands);
+      if (Res == 0) // success
+        return false;
+      else if (Res == -1) // irrecoverable error
+        return true;
+      // If this is VMRS, check for the apsr_nzcv operand.
+      if (Mnemonic == "vmrs" &&
+          Parser.getTok().getString().equals_lower("apsr_nzcv")) {
+        S = Parser.getTok().getLoc();
+        Parser.Lex();
+        Operands.push_back(ARMOperand::CreateToken("APSR_nzcv", S));
+        return false;
+      }
     }
 
     // Fall though for the Identifier case that is not a register or a
@@ -4573,7 +4626,7 @@ bool ARMAsmParser::parseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
     // identifier (like labels) as expressions and create them as immediates.
     const MCExpr *IdVal;
     S = Parser.getTok().getLoc();
-    if (getParser().ParseExpression(IdVal))
+    if (getParser().parseExpression(IdVal))
       return true;
     E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1);
     Operands.push_back(ARMOperand::CreateImm(IdVal, S, E));
@@ -4592,7 +4645,7 @@ bool ARMAsmParser::parseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
     if (Parser.getTok().isNot(AsmToken::Colon)) {
       bool isNegative = Parser.getTok().is(AsmToken::Minus);
       const MCExpr *ImmVal;
-      if (getParser().ParseExpression(ImmVal))
+      if (getParser().parseExpression(ImmVal))
         return true;
       const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(ImmVal);
       if (CE) {
@@ -4602,6 +4655,15 @@ bool ARMAsmParser::parseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
       }
       E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1);
       Operands.push_back(ARMOperand::CreateImm(ImmVal, S, E));
+
+      // There can be a trailing '!' on operands that we want as a separate
+      // '!' Token operand. Handle that here. For example, the compatibilty
+      // alias for 'srsdb sp!, #imm' is 'srsdb #imm!'.
+      if (Parser.getTok().is(AsmToken::Exclaim)) {
+        Operands.push_back(ARMOperand::CreateToken(Parser.getTok().getString(),
+                                                   Parser.getTok().getLoc()));
+        Parser.Lex(); // Eat exclaim token
+      }
       return false;
     }
     // w/ a ':' after the '#', it's just like a plain ':'.
@@ -4616,7 +4678,7 @@ bool ARMAsmParser::parseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
       return true;
 
     const MCExpr *SubExprVal;
-    if (getParser().ParseExpression(SubExprVal))
+    if (getParser().parseExpression(SubExprVal))
       return true;
 
     const MCExpr *ExprVal = ARMMCExpr::Create(RefKind, SubExprVal,
@@ -4989,7 +5051,7 @@ bool ARMAsmParser::ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
 
   // In Thumb1, only the branch (B) instruction can be predicated.
   if (isThumbOne() && PredicationCode != ARMCC::AL && Mnemonic != "b") {
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return Error(NameLoc, "conditional execution not supported in Thumb1");
   }
 
@@ -5003,14 +5065,14 @@ bool ARMAsmParser::ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
   if (Mnemonic == "it") {
     SMLoc Loc = SMLoc::getFromPointer(NameLoc.getPointer() + 2);
     if (ITMask.size() > 3) {
-      Parser.EatToEndOfStatement();
+      Parser.eatToEndOfStatement();
       return Error(Loc, "too many conditions on IT instruction");
     }
     unsigned Mask = 8;
     for (unsigned i = ITMask.size(); i != 0; --i) {
       char pos = ITMask[i - 1];
       if (pos != 't' && pos != 'e') {
-        Parser.EatToEndOfStatement();
+        Parser.eatToEndOfStatement();
         return Error(Loc, "illegal IT block condition mask '" + ITMask + "'");
       }
       Mask >>= 1;
@@ -5036,14 +5098,14 @@ bool ARMAsmParser::ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
   // If we had a carry-set on an instruction that can't do that, issue an
   // error.
   if (!CanAcceptCarrySet && CarrySetting) {
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return Error(NameLoc, "instruction '" + Mnemonic +
                  "' can not set flags, but 's' suffix specified");
   }
   // If we had a predication code on an instruction that can't do that, issue an
   // error.
   if (!CanAcceptPredicationCode && PredicationCode != ARMCC::AL) {
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return Error(NameLoc, "instruction '" + Mnemonic +
                  "' is not predicable, but condition code specified");
   }
@@ -5092,7 +5154,7 @@ bool ARMAsmParser::ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     // Read the first operand.
     if (parseOperand(Operands, Mnemonic)) {
-      Parser.EatToEndOfStatement();
+      Parser.eatToEndOfStatement();
       return true;
     }
 
@@ -5101,7 +5163,7 @@ bool ARMAsmParser::ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
 
       // Parse and remember the operand.
       if (parseOperand(Operands, Mnemonic)) {
-        Parser.EatToEndOfStatement();
+        Parser.eatToEndOfStatement();
         return true;
       }
     }
@@ -5109,7 +5171,7 @@ bool ARMAsmParser::ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
 
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     SMLoc Loc = getLexer().getLoc();
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return Error(Loc, "unexpected token in argument list");
   }
 
@@ -5140,50 +5202,42 @@ bool ARMAsmParser::ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
     delete Op;
   }
 
-  // The vector-compare-to-zero instructions have a literal token "#0" at
-  // the end that comes to here as an immediate operand. Convert it to a
-  // token to play nicely with the matcher.
-  if ((Mnemonic == "vceq" || Mnemonic == "vcge" || Mnemonic == "vcgt" ||
-      Mnemonic == "vcle" || Mnemonic == "vclt") && Operands.size() == 6 &&
-      static_cast<ARMOperand*>(Operands[5])->isImm()) {
-    ARMOperand *Op = static_cast<ARMOperand*>(Operands[5]);
-    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Op->getImm());
-    if (CE && CE->getValue() == 0) {
-      Operands.erase(Operands.begin() + 5);
-      Operands.push_back(ARMOperand::CreateToken("#0", Op->getStartLoc()));
-      delete Op;
-    }
-  }
-  // VCMP{E} does the same thing, but with a different operand count.
-  if ((Mnemonic == "vcmp" || Mnemonic == "vcmpe") && Operands.size() == 5 &&
-      static_cast<ARMOperand*>(Operands[4])->isImm()) {
-    ARMOperand *Op = static_cast<ARMOperand*>(Operands[4]);
-    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Op->getImm());
-    if (CE && CE->getValue() == 0) {
-      Operands.erase(Operands.begin() + 4);
-      Operands.push_back(ARMOperand::CreateToken("#0", Op->getStartLoc()));
-      delete Op;
-    }
-  }
-  // Similarly, the Thumb1 "RSB" instruction has a literal "#0" on the
-  // end. Convert it to a token here. Take care not to convert those
-  // that should hit the Thumb2 encoding.
-  if (Mnemonic == "rsb" && isThumb() && Operands.size() == 6 &&
-      static_cast<ARMOperand*>(Operands[3])->isReg() &&
-      static_cast<ARMOperand*>(Operands[4])->isReg() &&
-      static_cast<ARMOperand*>(Operands[5])->isImm()) {
-    ARMOperand *Op = static_cast<ARMOperand*>(Operands[5]);
-    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Op->getImm());
-    if (CE && CE->getValue() == 0 &&
-        (isThumbOne() ||
-         // The cc_out operand matches the IT block.
-         ((inITBlock() != CarrySetting) &&
-         // Neither register operand is a high register.
-         (isARMLowRegister(static_cast<ARMOperand*>(Operands[3])->getReg()) &&
-          isARMLowRegister(static_cast<ARMOperand*>(Operands[4])->getReg()))))){
-      Operands.erase(Operands.begin() + 5);
-      Operands.push_back(ARMOperand::CreateToken("#0", Op->getStartLoc()));
-      delete Op;
+  // Adjust operands of ldrexd/strexd to MCK_GPRPair.
+  // ldrexd/strexd require even/odd GPR pair. To enforce this constraint,
+  // a single GPRPair reg operand is used in the .td file to replace the two
+  // GPRs. However, when parsing from asm, the two GRPs cannot be automatically
+  // expressed as a GPRPair, so we have to manually merge them.
+  // FIXME: We would really like to be able to tablegen'erate this.
+  if (!isThumb() && Operands.size() > 4 &&
+      (Mnemonic == "ldrexd" || Mnemonic == "strexd")) {
+    bool isLoad = (Mnemonic == "ldrexd");
+    unsigned Idx = isLoad ? 2 : 3;
+    ARMOperand* Op1 = static_cast<ARMOperand*>(Operands[Idx]);
+    ARMOperand* Op2 = static_cast<ARMOperand*>(Operands[Idx+1]);
+
+    const MCRegisterClass& MRC = MRI->getRegClass(ARM::GPRRegClassID);
+    // Adjust only if Op1 and Op2 are GPRs.
+    if (Op1->isReg() && Op2->isReg() && MRC.contains(Op1->getReg()) &&
+        MRC.contains(Op2->getReg())) {
+      unsigned Reg1 = Op1->getReg();
+      unsigned Reg2 = Op2->getReg();
+      unsigned Rt = MRI->getEncodingValue(Reg1);
+      unsigned Rt2 = MRI->getEncodingValue(Reg2);
+
+      // Rt2 must be Rt + 1 and Rt must be even.
+      if (Rt + 1 != Rt2 || (Rt & 1)) {
+        Error(Op2->getStartLoc(), isLoad ?
+            "destination operands must be sequential" :
+            "source operands must be sequential");
+        return true;
+      }
+      unsigned NewReg = MRI->getMatchingSuperReg(Reg1, ARM::gsub_0,
+          &(MRI->getRegClass(ARM::GPRPairRegClassID)));
+      Operands.erase(Operands.begin() + Idx, Operands.begin() + Idx + 2);
+      Operands.insert(Operands.begin() + Idx, ARMOperand::CreateReg(
+            NewReg, Op1->getStartLoc(), Op2->getEndLoc()));
+      delete Op1;
+      delete Op2;
     }
   }
 
@@ -5274,8 +5328,7 @@ validateInstruction(MCInst &Inst,
   switch (Inst.getOpcode()) {
   case ARM::LDRD:
   case ARM::LDRD_PRE:
-  case ARM::LDRD_POST:
-  case ARM::LDREXD: {
+  case ARM::LDRD_POST: {
     // Rt2 must be Rt + 1.
     unsigned Rt = MRI->getEncodingValue(Inst.getOperand(0).getReg());
     unsigned Rt2 = MRI->getEncodingValue(Inst.getOperand(1).getReg());
@@ -5294,8 +5347,7 @@ validateInstruction(MCInst &Inst,
     return false;
   }
   case ARM::STRD_PRE:
-  case ARM::STRD_POST:
-  case ARM::STREXD: {
+  case ARM::STRD_POST: {
     // Rt2 must be Rt + 1.
     unsigned Rt = MRI->getEncodingValue(Inst.getOperand(1).getReg());
     unsigned Rt2 = MRI->getEncodingValue(Inst.getOperand(2).getReg());
@@ -5686,7 +5738,12 @@ processInstruction(MCInst &Inst,
   }
   // Aliases for alternate PC+imm syntax of LDR instructions.
   case ARM::t2LDRpcrel:
-    Inst.setOpcode(ARM::t2LDRpci);
+    // Select the narrow version if the immediate will fit.
+    if (Inst.getOperand(1).getImm() > 0 &&
+        Inst.getOperand(1).getImm() <= 0xff)
+      Inst.setOpcode(ARM::tLDRpci);
+    else
+      Inst.setOpcode(ARM::t2LDRpci);
     return true;
   case ARM::t2LDRBpcrel:
     Inst.setOpcode(ARM::t2LDRBpci);
@@ -7483,6 +7540,7 @@ MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
                         bool MatchingInlineAsm) {
   MCInst Inst;
   unsigned MatchResult;
+
   MatchResult = MatchInstructionImpl(Operands, Inst, ErrorInfo,
                                      MatchingInlineAsm);
   switch (MatchResult) {
@@ -7595,10 +7653,10 @@ bool ARMAsmParser::parseDirectiveWord(unsigned Size, SMLoc L) {
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     for (;;) {
       const MCExpr *Value;
-      if (getParser().ParseExpression(Value))
+      if (getParser().parseExpression(Value))
         return true;
 
-      getParser().getStreamer().EmitValue(Value, Size, 0/*addrspace*/);
+      getParser().getStreamer().EmitValue(Value, Size);
 
       if (getLexer().is(AsmToken::EndOfStatement))
         break;
@@ -7742,13 +7800,13 @@ bool ARMAsmParser::parseDirectiveReq(StringRef Name, SMLoc L) {
   unsigned Reg;
   SMLoc SRegLoc, ERegLoc;
   if (ParseRegister(Reg, SRegLoc, ERegLoc)) {
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return Error(SRegLoc, "register name expected");
   }
 
   // Shouldn't be anything else.
   if (Parser.getTok().isNot(AsmToken::EndOfStatement)) {
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return Error(Parser.getTok().getLoc(),
                  "unexpected input in .req directive.");
   }
@@ -7766,7 +7824,7 @@ bool ARMAsmParser::parseDirectiveReq(StringRef Name, SMLoc L) {
 ///  ::= .unreq registername
 bool ARMAsmParser::parseDirectiveUnreq(SMLoc L) {
   if (Parser.getTok().isNot(AsmToken::Identifier)) {
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return Error(L, "unexpected input in .unreq directive.");
   }
   RegisterReqs.erase(Parser.getTok().getIdentifier());
@@ -7786,16 +7844,31 @@ bool ARMAsmParser::parseDirectiveEabiAttr(SMLoc L) {
   return true;
 }
 
-extern "C" void LLVMInitializeARMAsmLexer();
-
 /// Force static initialization.
 extern "C" void LLVMInitializeARMAsmParser() {
   RegisterMCAsmParser<ARMAsmParser> X(TheARMTarget);
   RegisterMCAsmParser<ARMAsmParser> Y(TheThumbTarget);
-  LLVMInitializeARMAsmLexer();
 }
 
 #define GET_REGISTER_MATCHER
 #define GET_SUBTARGET_FEATURE_NAME
 #define GET_MATCHER_IMPLEMENTATION
 #include "ARMGenAsmMatcher.inc"
+
+// Define this matcher function after the auto-generated include so we
+// have the match class enum definitions.
+unsigned ARMAsmParser::validateTargetOperandClass(MCParsedAsmOperand *AsmOp,
+                                                  unsigned Kind) {
+  ARMOperand *Op = static_cast<ARMOperand*>(AsmOp);
+  // If the kind is a token for a literal immediate, check if our asm
+  // operand matches. This is for InstAliases which have a fixed-value
+  // immediate in the syntax.
+  if (Kind == MCK__35_0 && Op->isImm()) {
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Op->getImm());
+    if (!CE)
+      return Match_InvalidOperand;
+    if (CE->getValue() == 0)
+      return Match_Success;
+  }
+  return Match_InvalidOperand;
+}
diff --git a/lib/Target/ARM/AsmParser/CMakeLists.txt b/lib/Target/ARM/AsmParser/CMakeLists.txt
index e24a1b17867a..d2012c387cda 100644
--- a/lib/Target/ARM/AsmParser/CMakeLists.txt
+++ b/lib/Target/ARM/AsmParser/CMakeLists.txt
@@ -1,7 +1,6 @@
 include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
 
 add_llvm_library(LLVMARMAsmParser
-  ARMAsmLexer.cpp
   ARMAsmParser.cpp
   )
 
diff --git a/lib/Target/ARM/CMakeLists.txt b/lib/Target/ARM/CMakeLists.txt
index 377bd9243c2e..b832508a086c 100644
--- a/lib/Target/ARM/CMakeLists.txt
+++ b/lib/Target/ARM/CMakeLists.txt
@@ -11,11 +11,11 @@ tablegen(LLVM ARMGenDAGISel.inc -gen-dag-isel)
 tablegen(LLVM ARMGenFastISel.inc -gen-fast-isel)
 tablegen(LLVM ARMGenCallingConv.inc -gen-callingconv)
 tablegen(LLVM ARMGenSubtargetInfo.inc -gen-subtarget)
-tablegen(LLVM ARMGenEDInfo.inc -gen-enhanced-disassembly-info)
 tablegen(LLVM ARMGenDisassemblerTables.inc -gen-disassembler)
 add_public_tablegen_target(ARMCommonTableGen)
 
 add_llvm_target(ARMCodeGen
+  A15SDOptimizer.cpp
   ARMAsmPrinter.cpp
   ARMBaseInstrInfo.cpp
   ARMBaseRegisterInfo.cpp
@@ -38,6 +38,7 @@ add_llvm_target(ARMCodeGen
   ARMSubtarget.cpp
   ARMTargetMachine.cpp
   ARMTargetObjectFile.cpp
+  ARMTargetTransformInfo.cpp
   MLxExpansionPass.cpp
   Thumb1FrameLowering.cpp
   Thumb1InstrInfo.cpp
diff --git a/lib/Target/ARM/Disassembler/ARMDisassembler.cpp b/lib/Target/ARM/Disassembler/ARMDisassembler.cpp
index f00142de50dc..2e009e55e3b0 100644
--- a/lib/Target/ARM/Disassembler/ARMDisassembler.cpp
+++ b/lib/Target/ARM/Disassembler/ARMDisassembler.cpp
@@ -9,21 +9,20 @@
 
 #define DEBUG_TYPE "arm-disassembler"
 
+#include "llvm/MC/MCDisassembler.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "MCTargetDesc/ARMMCExpr.h"
 #include "MCTargetDesc/ARMBaseInfo.h"
-#include "llvm/MC/EDInstInfo.h"
-#include "llvm/MC/MCInst.h"
-#include "llvm/MC/MCInstrDesc.h"
-#include "llvm/MC/MCExpr.h"
+#include "MCTargetDesc/ARMMCExpr.h"
 #include "llvm/MC/MCContext.h"
-#include "llvm/MC/MCDisassembler.h"
+#include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCFixedLenDisassembler.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCInstrDesc.h"
 #include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/Support/Debug.h"
-#include "llvm/Support/MemoryObject.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/LEB128.h"
+#include "llvm/Support/MemoryObject.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
 #include <vector>
@@ -105,10 +104,6 @@ public:
                               uint64_t address,
                               raw_ostream &vStream,
                               raw_ostream &cStream) const;
-
-  /// getEDInfo - See MCDisassembler.
-  const EDInstInfo *getEDInfo() const;
-private:
 };
 
 /// ThumbDisassembler - Thumb disassembler for all Thumb platforms.
@@ -131,8 +126,6 @@ public:
                               raw_ostream &vStream,
                               raw_ostream &cStream) const;
 
-  /// getEDInfo - See MCDisassembler.
-  const EDInstInfo *getEDInfo() const;
 private:
   mutable ITStatus ITBlock;
   DecodeStatus AddThumbPredicate(MCInst&) const;
@@ -385,7 +378,6 @@ static DecodeStatus DecodeLDR(MCInst &Inst, unsigned Val,
 static DecodeStatus DecodeMRRC2(llvm::MCInst &Inst, unsigned Val,
                                 uint64_t Address, const void *Decoder);
 #include "ARMGenDisassemblerTables.inc"
-#include "ARMGenEDInfo.inc"
 
 static MCDisassembler *createARMDisassembler(const Target &T, const MCSubtargetInfo &STI) {
   return new ARMDisassembler(STI);
@@ -395,14 +387,6 @@ static MCDisassembler *createThumbDisassembler(const Target &T, const MCSubtarge
   return new ThumbDisassembler(STI);
 }
 
-const EDInstInfo *ARMDisassembler::getEDInfo() const {
-  return instInfoARM;
-}
-
-const EDInstInfo *ThumbDisassembler::getEDInfo() const {
-  return instInfoARM;
-}
-
 DecodeStatus ARMDisassembler::getInstruction(MCInst &MI, uint64_t &Size,
                                              const MemoryObject &Region,
                                              uint64_t Address,
@@ -1281,7 +1265,13 @@ static DecodeStatus DecodeBitfieldMaskOperand(MCInst &Inst, unsigned Val,
   unsigned lsb = fieldFromInstruction(Val, 0, 5);
 
   DecodeStatus S = MCDisassembler::Success;
-  if (lsb > msb) Check(S, MCDisassembler::SoftFail);
+  if (lsb > msb) {
+    Check(S, MCDisassembler::SoftFail);
+    // The check above will cause the warning for the "potentially undefined
+    // instruction encoding" but we can't build a bad MCOperand value here
+    // with a lsb > msb or else printing the MCInst will cause a crash.
+    lsb = msb;
+  }
 
   uint32_t msb_mask = 0xFFFFFFFF;
   if (msb != 31) msb_mask = (1U << (msb+1)) - 1;
@@ -3059,9 +3049,9 @@ static DecodeStatus DecodeT2BROperand(MCInst &Inst, unsigned Val,
 
 static DecodeStatus DecodeThumbCmpBROperand(MCInst &Inst, unsigned Val,
                                  uint64_t Address, const void *Decoder) {
-  if (!tryAddingSymbolicOperand(Address, Address + SignExtend32<7>(Val<<1) + 4,
+  if (!tryAddingSymbolicOperand(Address, Address + (Val<<1) + 4,
                                 true, 2, Inst, Decoder))
-    Inst.addOperand(MCOperand::CreateImm(SignExtend32<7>(Val << 1)));
+    Inst.addOperand(MCOperand::CreateImm(Val << 1));
   return MCDisassembler::Success;
 }
 
@@ -3288,7 +3278,7 @@ static DecodeStatus DecodeT2LdStPre(MCInst &Inst, unsigned Insn,
       return MCDisassembler::Fail;
   }
 
-  if (!Check(S, DecoderGPRRegisterClass(Inst, Rt, Address, Decoder)))
+  if (!Check(S, DecodeGPRRegisterClass(Inst, Rt, Address, Decoder)))
     return MCDisassembler::Fail;
 
   if (load) {
diff --git a/lib/Target/ARM/InstPrinter/ARMInstPrinter.cpp b/lib/Target/ARM/InstPrinter/ARMInstPrinter.cpp
index dcc41d93f5ce..2afb20d6686a 100644
--- a/lib/Target/ARM/InstPrinter/ARMInstPrinter.cpp
+++ b/lib/Target/ARM/InstPrinter/ARMInstPrinter.cpp
@@ -13,11 +13,11 @@
 
 #define DEBUG_TYPE "asm-printer"
 #include "ARMInstPrinter.h"
-#include "MCTargetDesc/ARMBaseInfo.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/MC/MCInst.h"
+#include "MCTargetDesc/ARMBaseInfo.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
 #include "llvm/Support/raw_ostream.h"
@@ -252,6 +252,35 @@ void ARMInstPrinter::printInst(const MCInst *MI, raw_ostream &O,
     return;
   }
 
+  // Combine 2 GPRs from disassember into a GPRPair to match with instr def.
+  // ldrexd/strexd require even/odd GPR pair. To enforce this constraint,
+  // a single GPRPair reg operand is used in the .td file to replace the two
+  // GPRs. However, when decoding them, the two GRPs cannot be automatically
+  // expressed as a GPRPair, so we have to manually merge them.
+  // FIXME: We would really like to be able to tablegen'erate this.
+  if (Opcode == ARM::LDREXD || Opcode == ARM::STREXD) {
+    const MCRegisterClass& MRC = MRI.getRegClass(ARM::GPRRegClassID);
+    bool isStore = Opcode == ARM::STREXD;
+    unsigned Reg = MI->getOperand(isStore ? 1 : 0).getReg();
+    if (MRC.contains(Reg)) {
+      MCInst NewMI;
+      MCOperand NewReg;
+      NewMI.setOpcode(Opcode);
+
+      if (isStore)
+        NewMI.addOperand(MI->getOperand(0));
+      NewReg = MCOperand::CreateReg(MRI.getMatchingSuperReg(Reg, ARM::gsub_0,
+        &MRI.getRegClass(ARM::GPRPairRegClassID)));
+      NewMI.addOperand(NewReg);
+
+      // Copy the rest operands into NewMI.
+      for(unsigned i= isStore ? 3 : 2; i < MI->getNumOperands(); ++i)
+        NewMI.addOperand(MI->getOperand(i));
+      printInstruction(&NewMI, O);
+      return;
+    }
+  }
+
   printInstruction(MI, O);
   printAnnotation(O, Annot);
 }
@@ -264,7 +293,7 @@ void ARMInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
     printRegName(O, Reg);
   } else if (Op.isImm()) {
     O << markup("<imm:")
-      << '#' << Op.getImm()
+      << '#' << formatImm(Op.getImm())
       << markup(">");
   } else {
     assert(Op.isExpr() && "unknown operand kind in printOperand");
@@ -290,7 +319,7 @@ void ARMInstPrinter::printThumbLdrLabelOperand(const MCInst *MI, unsigned OpNum,
     O << *MO1.getExpr();
   else if (MO1.isImm()) {
     O << markup("<mem:") << "[pc, "
-      << markup("<imm:") << "#" << MO1.getImm()
+      << markup("<imm:") << "#" << formatImm(MO1.getImm())
       << markup(">]>", "]");
   }
   else
@@ -598,8 +627,7 @@ void ARMInstPrinter::printAddrMode6Operand(const MCInst *MI, unsigned OpNum,
   O << markup("<mem:") << "[";
   printRegName(O, MO1.getReg());
   if (MO2.getImm()) {
-    // FIXME: Both darwin as and GNU as violate ARM docs here.
-    O << ", :" << (MO2.getImm() << 3);
+    O << ":" << (MO2.getImm() << 3);
   }
   O << "]" << markup(">");
 }
@@ -691,6 +719,15 @@ void ARMInstPrinter::printRegisterList(const MCInst *MI, unsigned OpNum,
   O << "}";
 }
 
+void ARMInstPrinter::printGPRPairOperand(const MCInst *MI, unsigned OpNum,
+                                         raw_ostream &O) {
+  unsigned Reg = MI->getOperand(OpNum).getReg();
+  printRegName(O, MRI.getSubReg(Reg, ARM::gsub_0));
+  O << ", ";
+  printRegName(O, MRI.getSubReg(Reg, ARM::gsub_1));
+}
+
+
 void ARMInstPrinter::printSetendOperand(const MCInst *MI, unsigned OpNum,
                                         raw_ostream &O) {
   const MCOperand &Op = MI->getOperand(OpNum);
@@ -873,7 +910,7 @@ void ARMInstPrinter::printAdrLabelOperand(const MCInst *MI, unsigned OpNum,
 void ARMInstPrinter::printThumbS4ImmOperand(const MCInst *MI, unsigned OpNum,
                                             raw_ostream &O) {
   O << markup("<imm:")
-    << "#" << MI->getOperand(OpNum).getImm() * 4
+    << "#" << formatImm(MI->getOperand(OpNum).getImm() * 4)
     << markup(">");
 }
 
@@ -881,7 +918,7 @@ void ARMInstPrinter::printThumbSRImm(const MCInst *MI, unsigned OpNum,
                                      raw_ostream &O) {
   unsigned Imm = MI->getOperand(OpNum).getImm();
   O << markup("<imm:")
-    << "#" << (Imm == 0 ? 32 : Imm)
+    << "#" << formatImm((Imm == 0 ? 32 : Imm))
     << markup(">");
 }
 
@@ -938,7 +975,7 @@ void ARMInstPrinter::printThumbAddrModeImm5SOperand(const MCInst *MI,
   if (unsigned ImmOffs = MO2.getImm()) {
     O << ", "
       << markup("<imm:")
-      << "#" << ImmOffs * Scale
+      << "#" << formatImm(ImmOffs * Scale)
       << markup(">");
   }
   O << "]" << markup(">");
@@ -1089,7 +1126,7 @@ void ARMInstPrinter::printT2AddrModeImm0_1020s4Operand(const MCInst *MI,
   if (MO2.getImm()) {
     O << ", "
       << markup("<imm:")
-      << "#" << MO2.getImm() * 4
+      << "#" << formatImm(MO2.getImm() * 4)
       << markup(">");
   }
   O << "]" << markup(">");
@@ -1179,7 +1216,7 @@ void ARMInstPrinter::printImmPlusOneOperand(const MCInst *MI, unsigned OpNum,
                                             raw_ostream &O) {
   unsigned Imm = MI->getOperand(OpNum).getImm();
   O << markup("<imm:")
-    << "#" << Imm + 1
+    << "#" << formatImm(Imm + 1)
     << markup(">");
 }
 
diff --git a/lib/Target/ARM/InstPrinter/ARMInstPrinter.h b/lib/Target/ARM/InstPrinter/ARMInstPrinter.h
index b7bab5fdcd8e..edff75d886e9 100644
--- a/lib/Target/ARM/InstPrinter/ARMInstPrinter.h
+++ b/lib/Target/ARM/InstPrinter/ARMInstPrinter.h
@@ -124,6 +124,7 @@ public:
   void printNEONModImmOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
   void printImmPlusOneOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
   void printRotImmOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+  void printGPRPairOperand(const MCInst *MI, unsigned OpNum, raw_ostream &O);
 
   void printPCLabel(const MCInst *MI, unsigned OpNum, raw_ostream &O);
   void printThumbLdrLabelOperand(const MCInst *MI, unsigned OpNum,
diff --git a/lib/Target/ARM/LICENSE.TXT b/lib/Target/ARM/LICENSE.TXT
new file mode 100755
index 000000000000..68afea12ed44
--- /dev/null
+++ b/lib/Target/ARM/LICENSE.TXT
@@ -0,0 +1,47 @@
+ARM Limited
+
+Software Grant License Agreement ("Agreement")
+
+Except for the license granted herein to you, ARM Limited ("ARM") reserves all
+right, title, and interest in and to the Software (defined below).
+
+Definition
+
+"Software" means the code and documentation as well as any original work of
+authorship, including any modifications or additions to an existing work, that
+is intentionally submitted by ARM to llvm.org (http://llvm.org) ("LLVM") for
+inclusion in, or documentation of, any of the products owned or managed by LLVM
+(the "Work"). For the purposes of this definition, "submitted" means any form of
+electronic, verbal, or written communication sent to LLVM or its
+representatives, including but not limited to communication on electronic
+mailing lists, source code control systems, and issue tracking systems that are
+managed by, or on behalf of, LLVM for the purpose of discussing and improving
+the Work, but excluding communication that is conspicuously marked otherwise.
+
+1. Grant of Copyright License. Subject to the terms and conditions of this
+   Agreement, ARM hereby grants to you and to recipients of the Software
+   distributed by LLVM a perpetual, worldwide, non-exclusive, no-charge,
+   royalty-free, irrevocable copyright license to reproduce, prepare derivative
+   works of, publicly display, publicly perform, sublicense, and distribute the
+   Software and such derivative works.
+
+2. Grant of Patent License. Subject to the terms and conditions of this
+   Agreement, ARM hereby grants you and to recipients of the Software
+   distributed by LLVM a perpetual, worldwide, non-exclusive, no-charge,
+   royalty-free, irrevocable (except as stated in this section) patent license
+   to make, have made, use, offer to sell, sell, import, and otherwise transfer
+   the Work, where such license applies only to those patent claims licensable
+   by ARM that are necessarily infringed by ARM's Software alone or by
+   combination of the Software with the Work to which such Software was
+   submitted. If any entity institutes patent litigation against ARM or any
+   other entity (including a cross-claim or counterclaim in a lawsuit) alleging
+   that ARM's Software, or the Work to which ARM has contributed constitutes
+   direct or contributory patent infringement, then any patent licenses granted
+   to that entity under this Agreement for the Software or Work shall terminate
+   as of the date such litigation is filed.
+
+Unless required by applicable law or agreed to in writing, the software is
+provided on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,
+either express or implied, including, without limitation, any warranties or
+conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+PARTICULAR PURPOSE.
diff --git a/lib/Target/ARM/MCTargetDesc/ARMAsmBackend.cpp b/lib/Target/ARM/MCTargetDesc/ARMAsmBackend.cpp
index 1ba6ab039f20..e66e98567873 100644
--- a/lib/Target/ARM/MCTargetDesc/ARMAsmBackend.cpp
+++ b/lib/Target/ARM/MCTargetDesc/ARMAsmBackend.cpp
@@ -8,9 +8,11 @@
 //===----------------------------------------------------------------------===//
 
 #include "MCTargetDesc/ARMMCTargetDesc.h"
+#include "MCTargetDesc/ARMAddressingModes.h"
 #include "MCTargetDesc/ARMBaseInfo.h"
 #include "MCTargetDesc/ARMFixupKinds.h"
-#include "MCTargetDesc/ARMAddressingModes.h"
+#include "llvm/ADT/StringSwitch.h"
+#include "llvm/MC/MCAsmBackend.h"
 #include "llvm/MC/MCAssembler.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCDirectives.h"
@@ -21,7 +23,6 @@
 #include "llvm/MC/MCObjectWriter.h"
 #include "llvm/MC/MCSectionELF.h"
 #include "llvm/MC/MCSectionMachO.h"
-#include "llvm/MC/MCAsmBackend.h"
 #include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/MC/MCValue.h"
 #include "llvm/Object/MachOFormat.h"
@@ -114,11 +115,15 @@ public:
                          MCValue &Target, uint64_t &Value,
                          bool &IsResolved);
 
+
+  void applyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize,
+                  uint64_t Value) const;
+
   bool mayNeedRelaxation(const MCInst &Inst) const;
 
   bool fixupNeedsRelaxation(const MCFixup &Fixup,
                             uint64_t Value,
-                            const MCInstFragment *DF,
+                            const MCRelaxableFragment *DF,
                             const MCAsmLayout &Layout) const;
 
   void relaxInstruction(const MCInst &Inst, MCInst &Res) const;
@@ -161,7 +166,7 @@ bool ARMAsmBackend::mayNeedRelaxation(const MCInst &Inst) const {
 
 bool ARMAsmBackend::fixupNeedsRelaxation(const MCFixup &Fixup,
                                          uint64_t Value,
-                                         const MCInstFragment *DF,
+                                         const MCRelaxableFragment *DF,
                                          const MCAsmLayout &Layout) const {
   switch ((unsigned)Fixup.getKind()) {
   case ARM::fixup_arm_thumb_br: {
@@ -216,7 +221,7 @@ void ARMAsmBackend::relaxInstruction(const MCInst &Inst, MCInst &Res) const {
 bool ARMAsmBackend::writeNopData(uint64_t Count, MCObjectWriter *OW) const {
   const uint16_t Thumb1_16bitNopEncoding = 0x46c0; // using MOV r8,r8
   const uint16_t Thumb2_16bitNopEncoding = 0xbf00; // NOP
-  const uint32_t ARMv4_NopEncoding = 0xe1a0000; // using MOV r0,r0
+  const uint32_t ARMv4_NopEncoding = 0xe1a00000; // using MOV r0,r0
   const uint32_t ARMv6T2_NopEncoding = 0xe320f000; // NOP
   if (isThumb()) {
     const uint16_t nopEncoding = hasNOP() ? Thumb2_16bitNopEncoding
@@ -552,65 +557,6 @@ void ARMAsmBackend::processFixupValue(const MCAssembler &Asm,
   (void)adjustFixupValue(Fixup, Value, &Asm.getContext());
 }
 
-namespace {
-
-// FIXME: This should be in a separate file.
-// ELF is an ELF of course...
-class ELFARMAsmBackend : public ARMAsmBackend {
-public:
-  uint8_t OSABI;
-  ELFARMAsmBackend(const Target &T, const StringRef TT,
-                   uint8_t _OSABI)
-    : ARMAsmBackend(T, TT), OSABI(_OSABI) { }
-
-  void applyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize,
-                  uint64_t Value) const;
-
-  MCObjectWriter *createObjectWriter(raw_ostream &OS) const {
-    return createARMELFObjectWriter(OS, OSABI);
-  }
-};
-
-// FIXME: Raise this to share code between Darwin and ELF.
-void ELFARMAsmBackend::applyFixup(const MCFixup &Fixup, char *Data,
-                                  unsigned DataSize, uint64_t Value) const {
-  unsigned NumBytes = 4;        // FIXME: 2 for Thumb
-  Value = adjustFixupValue(Fixup, Value);
-  if (!Value) return;           // Doesn't change encoding.
-
-  unsigned Offset = Fixup.getOffset();
-
-  // For each byte of the fragment that the fixup touches, mask in the bits from
-  // the fixup value. The Value has been "split up" into the appropriate
-  // bitfields above.
-  for (unsigned i = 0; i != NumBytes; ++i)
-    Data[Offset + i] |= uint8_t((Value >> (i * 8)) & 0xff);
-}
-
-// FIXME: This should be in a separate file.
-class DarwinARMAsmBackend : public ARMAsmBackend {
-public:
-  const object::mach::CPUSubtypeARM Subtype;
-  DarwinARMAsmBackend(const Target &T, const StringRef TT,
-                      object::mach::CPUSubtypeARM st)
-    : ARMAsmBackend(T, TT), Subtype(st) {
-      HasDataInCodeSupport = true;
-    }
-
-  MCObjectWriter *createObjectWriter(raw_ostream &OS) const {
-    return createARMMachObjectWriter(OS, /*Is64Bit=*/false,
-                                     object::mach::CTM_ARM,
-                                     Subtype);
-  }
-
-  void applyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize,
-                  uint64_t Value) const;
-
-  virtual bool doesSectionRequireSymbols(const MCSection &Section) const {
-    return false;
-  }
-};
-
 /// getFixupKindNumBytes - The number of bytes the fixup may change.
 static unsigned getFixupKindNumBytes(unsigned Kind) {
   switch (Kind) {
@@ -659,8 +605,8 @@ static unsigned getFixupKindNumBytes(unsigned Kind) {
   }
 }
 
-void DarwinARMAsmBackend::applyFixup(const MCFixup &Fixup, char *Data,
-                                     unsigned DataSize, uint64_t Value) const {
+void ARMAsmBackend::applyFixup(const MCFixup &Fixup, char *Data,
+                               unsigned DataSize, uint64_t Value) const {
   unsigned NumBytes = getFixupKindNumBytes(Fixup.getKind());
   Value = adjustFixupValue(Fixup, Value);
   if (!Value) return;           // Doesn't change encoding.
@@ -668,37 +614,70 @@ void DarwinARMAsmBackend::applyFixup(const MCFixup &Fixup, char *Data,
   unsigned Offset = Fixup.getOffset();
   assert(Offset + NumBytes <= DataSize && "Invalid fixup offset!");
 
-  // For each byte of the fragment that the fixup touches, mask in the
-  // bits from the fixup value.
+  // For each byte of the fragment that the fixup touches, mask in the bits from
+  // the fixup value. The Value has been "split up" into the appropriate
+  // bitfields above.
   for (unsigned i = 0; i != NumBytes; ++i)
     Data[Offset + i] |= uint8_t((Value >> (i * 8)) & 0xff);
 }
 
+namespace {
+
+// FIXME: This should be in a separate file.
+// ELF is an ELF of course...
+class ELFARMAsmBackend : public ARMAsmBackend {
+public:
+  uint8_t OSABI;
+  ELFARMAsmBackend(const Target &T, const StringRef TT,
+                   uint8_t _OSABI)
+    : ARMAsmBackend(T, TT), OSABI(_OSABI) { }
+
+  MCObjectWriter *createObjectWriter(raw_ostream &OS) const {
+    return createARMELFObjectWriter(OS, OSABI);
+  }
+};
+
+// FIXME: This should be in a separate file.
+class DarwinARMAsmBackend : public ARMAsmBackend {
+public:
+  const object::mach::CPUSubtypeARM Subtype;
+  DarwinARMAsmBackend(const Target &T, const StringRef TT,
+                      object::mach::CPUSubtypeARM st)
+    : ARMAsmBackend(T, TT), Subtype(st) {
+      HasDataInCodeSupport = true;
+    }
+
+  MCObjectWriter *createObjectWriter(raw_ostream &OS) const {
+    return createARMMachObjectWriter(OS, /*Is64Bit=*/false,
+                                     object::mach::CTM_ARM,
+                                     Subtype);
+  }
+
+  virtual bool doesSectionRequireSymbols(const MCSection &Section) const {
+    return false;
+  }
+};
+
 } // end anonymous namespace
 
 MCAsmBackend *llvm::createARMAsmBackend(const Target &T, StringRef TT, StringRef CPU) {
   Triple TheTriple(TT);
 
   if (TheTriple.isOSDarwin()) {
-    if (TheTriple.getArchName() == "armv4t" ||
-        TheTriple.getArchName() == "thumbv4t")
-      return new DarwinARMAsmBackend(T, TT, object::mach::CSARM_V4T);
-    else if (TheTriple.getArchName() == "armv5e" ||
-        TheTriple.getArchName() == "thumbv5e")
-      return new DarwinARMAsmBackend(T, TT, object::mach::CSARM_V5TEJ);
-    else if (TheTriple.getArchName() == "armv6" ||
-        TheTriple.getArchName() == "thumbv6")
-      return new DarwinARMAsmBackend(T, TT, object::mach::CSARM_V6);
-    else if (TheTriple.getArchName() == "armv7f" ||
-        TheTriple.getArchName() == "thumbv7f")
-      return new DarwinARMAsmBackend(T, TT, object::mach::CSARM_V7F);
-    else if (TheTriple.getArchName() == "armv7k" ||
-        TheTriple.getArchName() == "thumbv7k")
-      return new DarwinARMAsmBackend(T, TT, object::mach::CSARM_V7K);
-    else if (TheTriple.getArchName() == "armv7s" ||
-        TheTriple.getArchName() == "thumbv7s")
-      return new DarwinARMAsmBackend(T, TT, object::mach::CSARM_V7S);
-    return new DarwinARMAsmBackend(T, TT, object::mach::CSARM_V7);
+    object::mach::CPUSubtypeARM CS =
+      StringSwitch<object::mach::CPUSubtypeARM>(TheTriple.getArchName())
+      .Cases("armv4t", "thumbv4t", object::mach::CSARM_V4T)
+      .Cases("armv5e", "thumbv5e",object::mach::CSARM_V5TEJ)
+      .Cases("armv6", "thumbv6", object::mach::CSARM_V6)
+      .Cases("armv6m", "thumbv6m", object::mach::CSARM_V6M)
+      .Cases("armv7em", "thumbv7em", object::mach::CSARM_V7EM)
+      .Cases("armv7f", "thumbv7f", object::mach::CSARM_V7F)
+      .Cases("armv7k", "thumbv7k", object::mach::CSARM_V7K)
+      .Cases("armv7m", "thumbv7m", object::mach::CSARM_V7M)
+      .Cases("armv7s", "thumbv7s", object::mach::CSARM_V7S)
+      .Default(object::mach::CSARM_V7);
+
+    return new DarwinARMAsmBackend(T, TT, CS);
   }
 
   if (TheTriple.isOSWindows())
diff --git a/lib/Target/ARM/MCTargetDesc/ARMELFObjectWriter.cpp b/lib/Target/ARM/MCTargetDesc/ARMELFObjectWriter.cpp
index 99e4f713f690..f98bbd204c7a 100644
--- a/lib/Target/ARM/MCTargetDesc/ARMELFObjectWriter.cpp
+++ b/lib/Target/ARM/MCTargetDesc/ARMELFObjectWriter.cpp
@@ -7,17 +7,17 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "MCTargetDesc/ARMFixupKinds.h"
 #include "MCTargetDesc/ARMMCTargetDesc.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/raw_ostream.h"
+#include "MCTargetDesc/ARMFixupKinds.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/MC/MCELFObjectWriter.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCSectionELF.h"
 #include "llvm/MC/MCValue.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
 
 using namespace llvm;
 
@@ -37,7 +37,6 @@ namespace {
     virtual unsigned GetRelocType(const MCValue &Target, const MCFixup &Fixup,
                                   bool IsPCRel, bool IsRelocWithSymbol,
                                   int64_t Addend) const;
-    virtual unsigned getEFlags() const;
     virtual const MCSymbol *ExplicitRelSym(const MCAssembler &Asm,
                                    const MCValue &Target,
                                    const MCFragment &F,
@@ -53,11 +52,6 @@ ARMELFObjectWriter::ARMELFObjectWriter(uint8_t OSABI)
 
 ARMELFObjectWriter::~ARMELFObjectWriter() {}
 
-// FIXME: get the real EABI Version from the Triple.
-unsigned ARMELFObjectWriter::getEFlags() const {
-  return ELF::EF_ARM_EABIMASK & DefaultEABIVersion;
-}
-
 // In ARM, _MergedGlobals and other most symbols get emitted directly.
 // I.e. not as an offset to a section symbol.
 // This code is an approximation of what ARM/gcc does.
@@ -133,6 +127,7 @@ const MCSymbol *ARMELFObjectWriter::ExplicitRelSym(const MCAssembler &Asm,
     switch (RelocType) {
     default: EmitThisSym = true; break;
     case ELF::R_ARM_ABS32: EmitThisSym = false; break;
+    case ELF::R_ARM_PREL31: EmitThisSym = false; break;
     }
   }
 
@@ -225,6 +220,9 @@ unsigned ARMELFObjectWriter::GetRelocTypeInner(const MCValue &Target,
     case FK_Data_4:
       switch (Modifier) {
       default: llvm_unreachable("Unsupported Modifier");
+      case MCSymbolRefExpr::VK_ARM_NONE:
+        Type = ELF::R_ARM_NONE;
+        break;
       case MCSymbolRefExpr::VK_ARM_GOT:
         Type = ELF::R_ARM_GOT_BREL;
         break;
@@ -249,7 +247,10 @@ unsigned ARMELFObjectWriter::GetRelocTypeInner(const MCValue &Target,
       case MCSymbolRefExpr::VK_ARM_TARGET2:
         Type = ELF::R_ARM_TARGET2;
         break;
-      } 
+      case MCSymbolRefExpr::VK_ARM_PREL31:
+        Type = ELF::R_ARM_PREL31;
+        break;
+      }
       break;
     case ARM::fixup_arm_ldst_pcrel_12:
     case ARM::fixup_arm_pcrel_10:
diff --git a/lib/Target/ARM/MCTargetDesc/ARMELFStreamer.cpp b/lib/Target/ARM/MCTargetDesc/ARMELFStreamer.cpp
new file mode 100644
index 000000000000..418971df3292
--- /dev/null
+++ b/lib/Target/ARM/MCTargetDesc/ARMELFStreamer.cpp
@@ -0,0 +1,418 @@
+//===- lib/MC/ARMELFStreamer.cpp - ELF Object Output for ARM --------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file assembles .s files and emits ARM ELF .o object files. Different
+// from generic ELF streamer in emitting mapping symbols ($a, $t and $d) to
+// delimit regions of data and code.
+//
+//===----------------------------------------------------------------------===//
+
+#include "ARMUnwindOp.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/Twine.h"
+#include "llvm/MC/MCAsmBackend.h"
+#include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCELF.h"
+#include "llvm/MC/MCELFStreamer.h"
+#include "llvm/MC/MCELFSymbolFlags.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCObjectStreamer.h"
+#include "llvm/MC/MCSection.h"
+#include "llvm/MC/MCSectionELF.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/MC/MCSymbol.h"
+#include "llvm/MC/MCValue.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ELF.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+namespace {
+
+/// Extend the generic ELFStreamer class so that it can emit mapping symbols at
+/// the appropriate points in the object files. These symbols are defined in the
+/// ARM ELF ABI: infocenter.arm.com/help/topic/com.arm.../IHI0044D_aaelf.pdf.
+///
+/// In brief: $a, $t or $d should be emitted at the start of each contiguous
+/// region of ARM code, Thumb code or data in a section. In practice, this
+/// emission does not rely on explicit assembler directives but on inherent
+/// properties of the directives doing the emission (e.g. ".byte" is data, "add
+/// r0, r0, r0" an instruction).
+///
+/// As a result this system is orthogonal to the DataRegion infrastructure used
+/// by MachO. Beware!
+class ARMELFStreamer : public MCELFStreamer {
+public:
+  ARMELFStreamer(MCContext &Context, MCAsmBackend &TAB, raw_ostream &OS,
+                 MCCodeEmitter *Emitter, bool IsThumb)
+      : MCELFStreamer(SK_ARMELFStreamer, Context, TAB, OS, Emitter),
+        IsThumb(IsThumb), MappingSymbolCounter(0), LastEMS(EMS_None), ExTab(0),
+        FnStart(0), Personality(0), CantUnwind(false) {}
+
+  ~ARMELFStreamer() {}
+
+  // ARM exception handling directives
+  virtual void EmitFnStart();
+  virtual void EmitFnEnd();
+  virtual void EmitCantUnwind();
+  virtual void EmitPersonality(const MCSymbol *Per);
+  virtual void EmitHandlerData();
+  virtual void EmitSetFP(unsigned NewFpReg,
+                         unsigned NewSpReg,
+                         int64_t Offset = 0);
+  virtual void EmitPad(int64_t Offset);
+  virtual void EmitRegSave(const SmallVectorImpl<unsigned> &RegList,
+                           bool isVector);
+
+  virtual void ChangeSection(const MCSection *Section) {
+    // We have to keep track of the mapping symbol state of any sections we
+    // use. Each one should start off as EMS_None, which is provided as the
+    // default constructor by DenseMap::lookup.
+    LastMappingSymbols[getPreviousSection()] = LastEMS;
+    LastEMS = LastMappingSymbols.lookup(Section);
+
+    MCELFStreamer::ChangeSection(Section);
+  }
+
+  /// This function is the one used to emit instruction data into the ELF
+  /// streamer. We override it to add the appropriate mapping symbol if
+  /// necessary.
+  virtual void EmitInstruction(const MCInst& Inst) {
+    if (IsThumb)
+      EmitThumbMappingSymbol();
+    else
+      EmitARMMappingSymbol();
+
+    MCELFStreamer::EmitInstruction(Inst);
+  }
+
+  /// This is one of the functions used to emit data into an ELF section, so the
+  /// ARM streamer overrides it to add the appropriate mapping symbol ($d) if
+  /// necessary.
+  virtual void EmitBytes(StringRef Data, unsigned AddrSpace) {
+    EmitDataMappingSymbol();
+    MCELFStreamer::EmitBytes(Data, AddrSpace);
+  }
+
+  /// This is one of the functions used to emit data into an ELF section, so the
+  /// ARM streamer overrides it to add the appropriate mapping symbol ($d) if
+  /// necessary.
+  virtual void EmitValueImpl(const MCExpr *Value, unsigned Size,
+                             unsigned AddrSpace) {
+    EmitDataMappingSymbol();
+    MCELFStreamer::EmitValueImpl(Value, Size, AddrSpace);
+  }
+
+  virtual void EmitAssemblerFlag(MCAssemblerFlag Flag) {
+    MCELFStreamer::EmitAssemblerFlag(Flag);
+
+    switch (Flag) {
+    case MCAF_SyntaxUnified:
+      return; // no-op here.
+    case MCAF_Code16:
+      IsThumb = true;
+      return; // Change to Thumb mode
+    case MCAF_Code32:
+      IsThumb = false;
+      return; // Change to ARM mode
+    case MCAF_Code64:
+      return;
+    case MCAF_SubsectionsViaSymbols:
+      return;
+    }
+  }
+
+  static bool classof(const MCStreamer *S) {
+    return S->getKind() == SK_ARMELFStreamer;
+  }
+
+private:
+  enum ElfMappingSymbol {
+    EMS_None,
+    EMS_ARM,
+    EMS_Thumb,
+    EMS_Data
+  };
+
+  void EmitDataMappingSymbol() {
+    if (LastEMS == EMS_Data) return;
+    EmitMappingSymbol("$d");
+    LastEMS = EMS_Data;
+  }
+
+  void EmitThumbMappingSymbol() {
+    if (LastEMS == EMS_Thumb) return;
+    EmitMappingSymbol("$t");
+    LastEMS = EMS_Thumb;
+  }
+
+  void EmitARMMappingSymbol() {
+    if (LastEMS == EMS_ARM) return;
+    EmitMappingSymbol("$a");
+    LastEMS = EMS_ARM;
+  }
+
+  void EmitMappingSymbol(StringRef Name) {
+    MCSymbol *Start = getContext().CreateTempSymbol();
+    EmitLabel(Start);
+
+    MCSymbol *Symbol =
+      getContext().GetOrCreateSymbol(Name + "." +
+                                     Twine(MappingSymbolCounter++));
+
+    MCSymbolData &SD = getAssembler().getOrCreateSymbolData(*Symbol);
+    MCELF::SetType(SD, ELF::STT_NOTYPE);
+    MCELF::SetBinding(SD, ELF::STB_LOCAL);
+    SD.setExternal(false);
+    Symbol->setSection(*getCurrentSection());
+
+    const MCExpr *Value = MCSymbolRefExpr::Create(Start, getContext());
+    Symbol->setVariableValue(Value);
+  }
+
+  void EmitThumbFunc(MCSymbol *Func) {
+    // FIXME: Anything needed here to flag the function as thumb?
+
+    getAssembler().setIsThumbFunc(Func);
+
+    MCSymbolData &SD = getAssembler().getOrCreateSymbolData(*Func);
+    SD.setFlags(SD.getFlags() | ELF_Other_ThumbFunc);
+  }
+
+  // Helper functions for ARM exception handling directives
+  void Reset();
+
+  void EmitPersonalityFixup(StringRef Name);
+
+  void SwitchToEHSection(const char *Prefix, unsigned Type, unsigned Flags,
+                         SectionKind Kind, const MCSymbol &Fn);
+  void SwitchToExTabSection(const MCSymbol &FnStart);
+  void SwitchToExIdxSection(const MCSymbol &FnStart);
+
+  bool IsThumb;
+  int64_t MappingSymbolCounter;
+
+  DenseMap<const MCSection *, ElfMappingSymbol> LastMappingSymbols;
+  ElfMappingSymbol LastEMS;
+
+  // ARM Exception Handling Frame Information
+  MCSymbol *ExTab;
+  MCSymbol *FnStart;
+  const MCSymbol *Personality;
+  bool CantUnwind;
+};
+}
+
+inline void ARMELFStreamer::SwitchToEHSection(const char *Prefix,
+                                              unsigned Type,
+                                              unsigned Flags,
+                                              SectionKind Kind,
+                                              const MCSymbol &Fn) {
+  const MCSectionELF &FnSection =
+    static_cast<const MCSectionELF &>(Fn.getSection());
+
+  // Create the name for new section
+  StringRef FnSecName(FnSection.getSectionName());
+  SmallString<128> EHSecName(Prefix);
+  if (FnSecName != ".text") {
+    EHSecName += FnSecName;
+  }
+
+  // Get .ARM.extab or .ARM.exidx section
+  const MCSectionELF *EHSection = NULL;
+  if (const MCSymbol *Group = FnSection.getGroup()) {
+    EHSection = getContext().getELFSection(
+      EHSecName, Type, Flags | ELF::SHF_GROUP, Kind,
+      FnSection.getEntrySize(), Group->getName());
+  } else {
+    EHSection = getContext().getELFSection(EHSecName, Type, Flags, Kind);
+  }
+  assert(EHSection);
+
+  // Switch to .ARM.extab or .ARM.exidx section
+  SwitchSection(EHSection);
+  EmitCodeAlignment(4, 0);
+}
+
+inline void ARMELFStreamer::SwitchToExTabSection(const MCSymbol &FnStart) {
+  SwitchToEHSection(".ARM.extab",
+                    ELF::SHT_PROGBITS,
+                    ELF::SHF_ALLOC,
+                    SectionKind::getDataRel(),
+                    FnStart);
+}
+
+inline void ARMELFStreamer::SwitchToExIdxSection(const MCSymbol &FnStart) {
+  SwitchToEHSection(".ARM.exidx",
+                    ELF::SHT_ARM_EXIDX,
+                    ELF::SHF_ALLOC | ELF::SHF_LINK_ORDER,
+                    SectionKind::getDataRel(),
+                    FnStart);
+}
+
+void ARMELFStreamer::Reset() {
+  ExTab = NULL;
+  FnStart = NULL;
+  Personality = NULL;
+  CantUnwind = false;
+}
+
+// Add the R_ARM_NONE fixup at the same position
+void ARMELFStreamer::EmitPersonalityFixup(StringRef Name) {
+  const MCSymbol *PersonalitySym = getContext().GetOrCreateSymbol(Name);
+
+  const MCSymbolRefExpr *PersonalityRef =
+    MCSymbolRefExpr::Create(PersonalitySym,
+                            MCSymbolRefExpr::VK_ARM_NONE,
+                            getContext());
+
+  AddValueSymbols(PersonalityRef);
+  MCDataFragment *DF = getOrCreateDataFragment();
+  DF->getFixups().push_back(
+    MCFixup::Create(DF->getContents().size(), PersonalityRef,
+                    MCFixup::getKindForSize(4, false)));
+}
+
+void ARMELFStreamer::EmitFnStart() {
+  assert(FnStart == 0);
+  FnStart = getContext().CreateTempSymbol();
+  EmitLabel(FnStart);
+}
+
+void ARMELFStreamer::EmitFnEnd() {
+  assert(FnStart && ".fnstart must preceeds .fnend");
+
+  // Emit unwind opcodes if there is no .handlerdata directive
+  int PersonalityIndex = -1;
+  if (!ExTab && !CantUnwind) {
+    // For __aeabi_unwind_cpp_pr1, we have to emit opcodes in .ARM.extab.
+    SwitchToExTabSection(*FnStart);
+
+    // Create .ARM.extab label for offset in .ARM.exidx
+    ExTab = getContext().CreateTempSymbol();
+    EmitLabel(ExTab);
+
+    PersonalityIndex = 1;
+
+    uint32_t Entry = 0;
+    uint32_t NumExtraEntryWords = 0;
+    Entry |= NumExtraEntryWords << 24;
+    Entry |= (EHT_COMPACT | PersonalityIndex) << 16;
+
+    // TODO: This should be generated according to .save, .vsave, .setfp
+    // directives.  Currently, we are simply generating FINISH opcode.
+    Entry |= UNWIND_OPCODE_FINISH << 8;
+    Entry |= UNWIND_OPCODE_FINISH;
+
+    EmitIntValue(Entry, 4, 0);
+  }
+
+  // Emit the exception index table entry
+  SwitchToExIdxSection(*FnStart);
+
+  if (PersonalityIndex == 1)
+    EmitPersonalityFixup("__aeabi_unwind_cpp_pr1");
+
+  const MCSymbolRefExpr *FnStartRef =
+    MCSymbolRefExpr::Create(FnStart,
+                            MCSymbolRefExpr::VK_ARM_PREL31,
+                            getContext());
+
+  EmitValue(FnStartRef, 4, 0);
+
+  if (CantUnwind) {
+    EmitIntValue(EXIDX_CANTUNWIND, 4, 0);
+  } else {
+    const MCSymbolRefExpr *ExTabEntryRef =
+      MCSymbolRefExpr::Create(ExTab,
+                              MCSymbolRefExpr::VK_ARM_PREL31,
+                              getContext());
+    EmitValue(ExTabEntryRef, 4, 0);
+  }
+
+  // Clean exception handling frame information
+  Reset();
+}
+
+void ARMELFStreamer::EmitCantUnwind() {
+  CantUnwind = true;
+}
+
+void ARMELFStreamer::EmitHandlerData() {
+  SwitchToExTabSection(*FnStart);
+
+  // Create .ARM.extab label for offset in .ARM.exidx
+  assert(!ExTab);
+  ExTab = getContext().CreateTempSymbol();
+  EmitLabel(ExTab);
+
+  // Emit Personality
+  assert(Personality && ".personality directive must preceed .handlerdata");
+
+  const MCSymbolRefExpr *PersonalityRef =
+    MCSymbolRefExpr::Create(Personality,
+                            MCSymbolRefExpr::VK_ARM_PREL31,
+                            getContext());
+
+  EmitValue(PersonalityRef, 4, 0);
+
+  // Emit unwind opcodes
+  uint32_t Entry = 0;
+  uint32_t NumExtraEntryWords = 0;
+
+  // TODO: This should be generated according to .save, .vsave, .setfp
+  // directives.  Currently, we are simply generating FINISH opcode.
+  Entry |= NumExtraEntryWords << 24;
+  Entry |= UNWIND_OPCODE_FINISH << 16;
+  Entry |= UNWIND_OPCODE_FINISH << 8;
+  Entry |= UNWIND_OPCODE_FINISH;
+
+  EmitIntValue(Entry, 4, 0);
+}
+
+void ARMELFStreamer::EmitPersonality(const MCSymbol *Per) {
+  Personality = Per;
+}
+
+void ARMELFStreamer::EmitSetFP(unsigned NewFpReg,
+                               unsigned NewSpReg,
+                               int64_t Offset) {
+  // TODO: Not implemented
+}
+
+void ARMELFStreamer::EmitPad(int64_t Offset) {
+  // TODO: Not implemented
+}
+
+void ARMELFStreamer::EmitRegSave(const SmallVectorImpl<unsigned> &RegList,
+                                 bool IsVector) {
+  // TODO: Not implemented
+}
+
+namespace llvm {
+  MCELFStreamer* createARMELFStreamer(MCContext &Context, MCAsmBackend &TAB,
+                                      raw_ostream &OS, MCCodeEmitter *Emitter,
+                                      bool RelaxAll, bool NoExecStack,
+                                      bool IsThumb) {
+    ARMELFStreamer *S = new ARMELFStreamer(Context, TAB, OS, Emitter, IsThumb);
+    if (RelaxAll)
+      S->getAssembler().setRelaxAll(true);
+    if (NoExecStack)
+      S->getAssembler().setNoExecStack(true);
+    return S;
+  }
+
+}
+
+
diff --git a/lib/Target/ARM/MCTargetDesc/ARMELFStreamer.h b/lib/Target/ARM/MCTargetDesc/ARMELFStreamer.h
new file mode 100644
index 000000000000..77ae5d23628e
--- /dev/null
+++ b/lib/Target/ARM/MCTargetDesc/ARMELFStreamer.h
@@ -0,0 +1,27 @@
+//===-- ARMELFStreamer.h - ELF Streamer for ARM ------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements ELF streamer information for the ARM backend.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef ARM_ELF_STREAMER_H
+#define ARM_ELF_STREAMER_H
+
+#include "llvm/MC/MCELFStreamer.h"
+
+namespace llvm {
+
+  MCELFStreamer* createARMELFStreamer(MCContext &Context, MCAsmBackend &TAB,
+                                      raw_ostream &OS, MCCodeEmitter *Emitter,
+                                      bool RelaxAll, bool NoExecStack,
+                                      bool IsThumb);
+}
+
+#endif // ARM_ELF_STREAMER_H
diff --git a/lib/Target/ARM/MCTargetDesc/ARMMCCodeEmitter.cpp b/lib/Target/ARM/MCTargetDesc/ARMMCCodeEmitter.cpp
index d0e127a8f335..7a59a7dd5055 100644
--- a/lib/Target/ARM/MCTargetDesc/ARMMCCodeEmitter.cpp
+++ b/lib/Target/ARM/MCTargetDesc/ARMMCCodeEmitter.cpp
@@ -12,11 +12,13 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "mccodeemitter"
+#include "MCTargetDesc/ARMMCTargetDesc.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
 #include "MCTargetDesc/ARMBaseInfo.h"
 #include "MCTargetDesc/ARMFixupKinds.h"
 #include "MCTargetDesc/ARMMCExpr.h"
-#include "MCTargetDesc/ARMMCTargetDesc.h"
+#include "llvm/ADT/APFloat.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/MC/MCCodeEmitter.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
@@ -24,8 +26,6 @@
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
 #include "llvm/MC/MCSubtargetInfo.h"
-#include "llvm/ADT/APFloat.h"
-#include "llvm/ADT/Statistic.h"
 #include "llvm/Support/raw_ostream.h"
 
 using namespace llvm;
@@ -655,15 +655,28 @@ getAdrLabelOpValue(const MCInst &MI, unsigned OpIdx,
   int32_t offset = MO.getImm();
   uint32_t Val = 0x2000;
 
+  int SoImmVal;
   if (offset == INT32_MIN) {
     Val = 0x1000;
-    offset = 0;
+    SoImmVal = 0;
   } else if (offset < 0) {
     Val = 0x1000;
     offset *= -1;
+    SoImmVal = ARM_AM::getSOImmVal(offset);
+    if(SoImmVal == -1) {
+      Val = 0x2000;
+      offset *= -1;
+      SoImmVal = ARM_AM::getSOImmVal(offset);
+    }
+  } else {
+    SoImmVal = ARM_AM::getSOImmVal(offset);
+    if(SoImmVal == -1) {
+      Val = 0x1000;
+      offset *= -1;
+      SoImmVal = ARM_AM::getSOImmVal(offset);
+    }
   }
 
-  int SoImmVal = ARM_AM::getSOImmVal(offset);
   assert(SoImmVal != -1 && "Not a valid so_imm value!");
 
   Val |= SoImmVal;
diff --git a/lib/Target/ARM/MCTargetDesc/ARMMCExpr.cpp b/lib/Target/ARM/MCTargetDesc/ARMMCExpr.cpp
index 22e14a2281de..fc8505b052bd 100644
--- a/lib/Target/ARM/MCTargetDesc/ARMMCExpr.cpp
+++ b/lib/Target/ARM/MCTargetDesc/ARMMCExpr.cpp
@@ -9,8 +9,8 @@
 
 #define DEBUG_TYPE "armmcexpr"
 #include "ARMMCExpr.h"
-#include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCContext.h"
 using namespace llvm;
 
 const ARMMCExpr*
diff --git a/lib/Target/ARM/MCTargetDesc/ARMMCExpr.h b/lib/Target/ARM/MCTargetDesc/ARMMCExpr.h
index b404e6c6e014..cd4067a52955 100644
--- a/lib/Target/ARM/MCTargetDesc/ARMMCExpr.h
+++ b/lib/Target/ARM/MCTargetDesc/ARMMCExpr.h
@@ -64,6 +64,9 @@ public:
     return getSubExpr()->FindAssociatedSection();
   }
 
+  // There are no TLS ARMMCExprs at the moment.
+  void fixELFSymbolsInTLSFixups(MCAssembler &Asm) const {}
+
   static bool classof(const MCExpr *E) {
     return E->getKind() == MCExpr::Target;
   }
diff --git a/lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.cpp b/lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.cpp
index 00ffc94ac7d1..f09fb5a94fd8 100644
--- a/lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.cpp
+++ b/lib/Target/ARM/MCTargetDesc/ARMMCTargetDesc.cpp
@@ -11,10 +11,12 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "ARMMCTargetDesc.h"
-#include "ARMMCAsmInfo.h"
 #include "ARMBaseInfo.h"
+#include "ARMELFStreamer.h"
+#include "ARMMCAsmInfo.h"
+#include "ARMMCTargetDesc.h"
 #include "InstPrinter/ARMInstPrinter.h"
+#include "llvm/ADT/Triple.h"
 #include "llvm/MC/MCCodeGenInfo.h"
 #include "llvm/MC/MCInstrAnalysis.h"
 #include "llvm/MC/MCInstrInfo.h"
@@ -36,6 +38,8 @@
 using namespace llvm;
 
 std::string ARM_MC::ParseARMTriple(StringRef TT, StringRef CPU) {
+  Triple triple(TT);
+
   // Set the boolean corresponding to the current target triple, or the default
   // if one cannot be determined, to true.
   unsigned Len = TT.size();
@@ -118,6 +122,13 @@ std::string ARM_MC::ParseARMTriple(StringRef TT, StringRef CPU) {
       ARMArchFeature += ",+thumb-mode";
   }
 
+  if (triple.isOSNaCl()) {
+    if (ARMArchFeature.empty())
+      ARMArchFeature = "+nacl-trap";
+    else
+      ARMArchFeature += ",+nacl-trap";
+  }
+
   return ARMArchFeature;
 }
 
@@ -144,7 +155,7 @@ static MCInstrInfo *createARMMCInstrInfo() {
 
 static MCRegisterInfo *createARMMCRegisterInfo(StringRef Triple) {
   MCRegisterInfo *X = new MCRegisterInfo();
-  InitARMMCRegisterInfo(X, ARM::LR);
+  InitARMMCRegisterInfo(X, ARM::LR, 0, 0, ARM::PC);
   return X;
 }
 
@@ -186,7 +197,8 @@ static MCStreamer *createMCStreamer(const Target &T, StringRef TT,
     llvm_unreachable("ARM does not support Windows COFF format");
   }
 
-  return createELFStreamer(Ctx, MAB, OS, Emitter, false, NoExecStack);
+  return createARMELFStreamer(Ctx, MAB, OS, Emitter, false, NoExecStack,
+                              TheTriple.getArch() == Triple::thumb);
 }
 
 static MCInstPrinter *createARMMCInstPrinter(const Target &T,
diff --git a/lib/Target/ARM/MCTargetDesc/ARMMachObjectWriter.cpp b/lib/Target/ARM/MCTargetDesc/ARMMachObjectWriter.cpp
index 2154c931769a..b9efe74b41e5 100644
--- a/lib/Target/ARM/MCTargetDesc/ARMMachObjectWriter.cpp
+++ b/lib/Target/ARM/MCTargetDesc/ARMMachObjectWriter.cpp
@@ -7,17 +7,18 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "MCTargetDesc/ARMMCTargetDesc.h"
 #include "MCTargetDesc/ARMBaseInfo.h"
 #include "MCTargetDesc/ARMFixupKinds.h"
 #include "llvm/ADT/Twine.h"
-#include "llvm/MC/MCAssembler.h"
 #include "llvm/MC/MCAsmLayout.h"
-#include "llvm/MC/MCMachObjectWriter.h"
+#include "llvm/MC/MCAssembler.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCFixup.h"
 #include "llvm/MC/MCFixupKindInfo.h"
 #include "llvm/MC/MCMachOSymbolFlags.h"
+#include "llvm/MC/MCMachObjectWriter.h"
 #include "llvm/MC/MCValue.h"
 #include "llvm/Object/MachOFormat.h"
 #include "llvm/Support/ErrorHandling.h"
diff --git a/lib/Target/ARM/MCTargetDesc/ARMUnwindOp.h b/lib/Target/ARM/MCTargetDesc/ARMUnwindOp.h
new file mode 100644
index 000000000000..dad5576df4cd
--- /dev/null
+++ b/lib/Target/ARM/MCTargetDesc/ARMUnwindOp.h
@@ -0,0 +1,112 @@
+//===-- ARMUnwindOp.h - ARM Unwind Opcodes ----------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the constants for the ARM unwind opcodes and exception
+// handling table entry kinds.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef ARM_UNWIND_OP_H
+#define ARM_UNWIND_OP_H
+
+namespace llvm {
+
+  /// ARM exception handling table entry kinds
+  enum ARMEHTEntryKind {
+    EHT_GENERIC = 0x00,
+    EHT_COMPACT = 0x80
+  };
+
+  enum {
+    /// Special entry for the function never unwind
+    EXIDX_CANTUNWIND = 0x1
+  };
+
+  /// ARM-defined frame unwinding opcodes
+  enum ARMUnwindOpcodes {
+    // Format: 00xxxxxx
+    // Purpose: vsp = vsp + ((x << 2) + 4)
+    UNWIND_OPCODE_INC_VSP = 0x00,
+
+    // Format: 01xxxxxx
+    // Purpose: vsp = vsp - ((x << 2) + 4)
+    UNWIND_OPCODE_DEC_VSP = 0x40,
+
+    // Format: 10000000 00000000
+    // Purpose: refuse to unwind
+    UNWIND_OPCODE_REFUSE = 0x8000,
+
+    // Format: 1000xxxx xxxxxxxx
+    // Purpose: pop r[15:12], r[11:4]
+    // Constraint: x != 0
+    UNWIND_OPCODE_POP_REG_MASK_R4 = 0x8000,
+
+    // Format: 1001xxxx
+    // Purpose: vsp = r[x]
+    // Constraint: x != 13 && x != 15
+    UNWIND_OPCODE_SET_VSP = 0x90,
+
+    // Format: 10100xxx
+    // Purpose: pop r[(4+x):4]
+    UNWIND_OPCODE_POP_REG_RANGE_R4 = 0xa0,
+
+    // Format: 10101xxx
+    // Purpose: pop r14, r[(4+x):4]
+    UNWIND_OPCODE_POP_REG_RANGE_R4_R14 = 0xa8,
+
+    // Format: 10110000
+    // Purpose: finish
+    UNWIND_OPCODE_FINISH = 0xb0,
+
+    // Format: 10110001 0000xxxx
+    // Purpose: pop r[3:0]
+    // Constraint: x != 0
+    UNWIND_OPCODE_POP_REG_MASK = 0xb100,
+
+    // Format: 10110010 x(uleb128)
+    // Purpose: vsp = vsp + ((x << 2) + 0x204)
+    UNWIND_OPCODE_INC_VSP_ULEB128 = 0xb2,
+
+    // Format: 10110011 xxxxyyyy
+    // Purpose: pop d[(x+y):x]
+    UNWIND_OPCODE_POP_VFP_REG_RANGE_FSTMFDX = 0xb300,
+
+    // Format: 10111xxx
+    // Purpose: pop d[(8+x):8]
+    UNWIND_OPCODE_POP_VFP_REG_RANGE_FSTMFDX_D8 = 0xb8,
+
+    // Format: 11000xxx
+    // Purpose: pop wR[(10+x):10]
+    UNWIND_OPCODE_POP_WIRELESS_MMX_REG_RANGE_WR10 = 0xc0,
+
+    // Format: 11000110 xxxxyyyy
+    // Purpose: pop wR[(x+y):x]
+    UNWIND_OPCODE_POP_WIRELESS_MMX_REG_RANGE = 0xc600,
+
+    // Format: 11000111 0000xxxx
+    // Purpose: pop wCGR[3:0]
+    // Constraint: x != 0
+    UNWIND_OPCODE_POP_WIRELESS_MMX_REG_MASK = 0xc700,
+
+    // Format: 11001000 xxxxyyyy
+    // Purpose: pop d[(16+x+y):(16+x)]
+    UNWIND_OPCODE_POP_VFP_REG_RANGE_FSTMFDD_D16 = 0xc800,
+
+    // Format: 11001001 xxxxyyyy
+    // Purpose: pop d[(x+y):x]
+    UNWIND_OPCODE_POP_VFP_REG_RANGE_FSTMFDD = 0xc900,
+
+    // Format: 11010xxx
+    // Purpose: pop d[(8+x):8]
+    UNWIND_OPCODE_POP_VFP_REG_RANGE_FSTMFDD_D8 = 0xd0
+  };
+
+}
+
+#endif // ARM_UNWIND_OP_H
diff --git a/lib/Target/ARM/MCTargetDesc/CMakeLists.txt b/lib/Target/ARM/MCTargetDesc/CMakeLists.txt
index 256599412e8b..e17eb4d5e987 100644
--- a/lib/Target/ARM/MCTargetDesc/CMakeLists.txt
+++ b/lib/Target/ARM/MCTargetDesc/CMakeLists.txt
@@ -1,6 +1,7 @@
 add_llvm_library(LLVMARMDesc
   ARMAsmBackend.cpp
   ARMELFObjectWriter.cpp
+  ARMELFStreamer.cpp
   ARMMCAsmInfo.cpp
   ARMMCCodeEmitter.cpp
   ARMMCExpr.cpp
diff --git a/lib/Target/ARM/MLxExpansionPass.cpp b/lib/Target/ARM/MLxExpansionPass.cpp
index 70643bcda3ac..2e266c2e9624 100644
--- a/lib/Target/ARM/MLxExpansionPass.cpp
+++ b/lib/Target/ARM/MLxExpansionPass.cpp
@@ -16,16 +16,16 @@
 #include "ARM.h"
 #include "ARMBaseInstrInfo.h"
 #include "ARMSubtarget.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/ADT/SmallPtrSet.h"
-#include "llvm/ADT/Statistic.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 using namespace llvm;
 
 static cl::opt<bool>
diff --git a/lib/Target/ARM/Makefile b/lib/Target/ARM/Makefile
index 3e48ed1189cc..f069535ff3c0 100644
--- a/lib/Target/ARM/Makefile
+++ b/lib/Target/ARM/Makefile
@@ -16,7 +16,7 @@ BUILT_SOURCES = ARMGenRegisterInfo.inc ARMGenInstrInfo.inc \
 		ARMGenAsmWriter.inc ARMGenAsmMatcher.inc \
                 ARMGenDAGISel.inc ARMGenSubtargetInfo.inc \
                 ARMGenCodeEmitter.inc ARMGenCallingConv.inc \
-                ARMGenEDInfo.inc ARMGenFastISel.inc ARMGenMCCodeEmitter.inc \
+                ARMGenFastISel.inc ARMGenMCCodeEmitter.inc \
                 ARMGenMCPseudoLowering.inc ARMGenDisassemblerTables.inc
 
 DIRS = InstPrinter AsmParser Disassembler TargetInfo MCTargetDesc
diff --git a/lib/Target/ARM/README-Thumb.txt b/lib/Target/ARM/README-Thumb.txt
index 463c440852f5..a64707e6f34f 100644
--- a/lib/Target/ARM/README-Thumb.txt
+++ b/lib/Target/ARM/README-Thumb.txt
@@ -173,7 +173,6 @@ GCC is doing a couple of clever things here:
         mov r1, #1
         lsl r1, r1, #8
         tst r2, r1
-  
 
 //===---------------------------------------------------------------------===//
 
@@ -196,7 +195,6 @@ This is especially bad when dynamic alloca is used. The all fixed size stack
 objects are referenced off the frame pointer with negative offsets. See
 oggenc for an example.
 
-
 //===---------------------------------------------------------------------===//
 
 Poor codegen test/CodeGen/ARM/select.ll f7:
diff --git a/lib/Target/ARM/TargetInfo/ARMTargetInfo.cpp b/lib/Target/ARM/TargetInfo/ARMTargetInfo.cpp
index 500e3de82db3..fa5681fb12bf 100644
--- a/lib/Target/ARM/TargetInfo/ARMTargetInfo.cpp
+++ b/lib/Target/ARM/TargetInfo/ARMTargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "ARM.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
diff --git a/lib/Target/ARM/Thumb1FrameLowering.cpp b/lib/Target/ARM/Thumb1FrameLowering.cpp
index edd73c20c0be..2c3388cc452c 100644
--- a/lib/Target/ARM/Thumb1FrameLowering.cpp
+++ b/lib/Target/ARM/Thumb1FrameLowering.cpp
@@ -43,6 +43,41 @@ emitSPUpdate(MachineBasicBlock &MBB,
                             MRI, MIFlags);
 }
 
+
+void Thumb1FrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  const Thumb1InstrInfo &TII =
+    *static_cast<const Thumb1InstrInfo*>(MF.getTarget().getInstrInfo());
+  const Thumb1RegisterInfo *RegInfo =
+    static_cast<const Thumb1RegisterInfo*>(MF.getTarget().getRegisterInfo());
+  if (!hasReservedCallFrame(MF)) {
+    // If we have alloca, convert as follows:
+    // ADJCALLSTACKDOWN -> sub, sp, sp, amount
+    // ADJCALLSTACKUP   -> add, sp, sp, amount
+    MachineInstr *Old = I;
+    DebugLoc dl = Old->getDebugLoc();
+    unsigned Amount = Old->getOperand(0).getImm();
+    if (Amount != 0) {
+      // We need to keep the stack aligned properly.  To do this, we round the
+      // amount of space needed for the outgoing arguments up to the next
+      // alignment boundary.
+      unsigned Align = getStackAlignment();
+      Amount = (Amount+Align-1)/Align*Align;
+
+      // Replace the pseudo instruction with a new instruction...
+      unsigned Opc = Old->getOpcode();
+      if (Opc == ARM::ADJCALLSTACKDOWN || Opc == ARM::tADJCALLSTACKDOWN) {
+        emitSPUpdate(MBB, I, TII, dl, *RegInfo, -Amount);
+      } else {
+        assert(Opc == ARM::ADJCALLSTACKUP || Opc == ARM::tADJCALLSTACKUP);
+        emitSPUpdate(MBB, I, TII, dl, *RegInfo, Amount);
+      }
+    }
+  }
+  MBB.erase(I);
+}
+
 void Thumb1FrameLowering::emitPrologue(MachineFunction &MF) const {
   MachineBasicBlock &MBB = MF.front();
   MachineBasicBlock::iterator MBBI = MBB.begin();
@@ -124,14 +159,17 @@ void Thumb1FrameLowering::emitPrologue(MachineFunction &MF) const {
   unsigned DPRCSOffset  = NumBytes - (GPRCS1Size + GPRCS2Size + DPRCSSize);
   unsigned GPRCS2Offset = DPRCSOffset + DPRCSSize;
   unsigned GPRCS1Offset = GPRCS2Offset + GPRCS2Size;
-  AFI->setFramePtrSpillOffset(MFI->getObjectOffset(FramePtrSpillFI) + NumBytes);
+  bool HasFP = hasFP(MF);
+  if (HasFP)
+    AFI->setFramePtrSpillOffset(MFI->getObjectOffset(FramePtrSpillFI) +
+                                NumBytes);
   AFI->setGPRCalleeSavedArea1Offset(GPRCS1Offset);
   AFI->setGPRCalleeSavedArea2Offset(GPRCS2Offset);
   AFI->setDPRCalleeSavedAreaOffset(DPRCSOffset);
   NumBytes = DPRCSOffset;
 
   // Adjust FP so it point to the stack slot that contains the previous FP.
-  if (hasFP(MF)) {
+  if (HasFP) {
     AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tADDrSPi), FramePtr)
       .addFrameIndex(FramePtrSpillFI).addImm(0)
       .setMIFlags(MachineInstr::FrameSetup));
@@ -146,7 +184,7 @@ void Thumb1FrameLowering::emitPrologue(MachineFunction &MF) const {
     emitSPUpdate(MBB, MBBI, TII, dl, *RegInfo, -NumBytes,
                  MachineInstr::FrameSetup);
 
-  if (STI.isTargetELF() && hasFP(MF))
+  if (STI.isTargetELF() && HasFP)
     MFI->setOffsetAdjustment(MFI->getOffsetAdjustment() -
                              AFI->getFramePtrSpillOffset());
 
@@ -281,7 +319,7 @@ void Thumb1FrameLowering::emitEpilogue(MachineFunction &MF,
       BuildMI(MBB, MBBI, dl, TII.get(ARM::tBX_RET_vararg))
       .addReg(ARM::R3, RegState::Kill);
     AddDefaultPred(MIB);
-    MIB->copyImplicitOps(&*MBBI);
+    MIB.copyImplicitOps(&*MBBI);
     // erase the old tBX_RET instruction
     MBB.erase(MBBI);
   }
@@ -352,7 +390,7 @@ restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
         continue;
       Reg = ARM::PC;
       (*MIB).setDesc(TII.get(ARM::tPOP_RET));
-      MIB->copyImplicitOps(&*MI);
+      MIB.copyImplicitOps(&*MI);
       MI = MBB.erase(MI);
     }
     MIB.addReg(Reg, getDefRegState(true));
diff --git a/lib/Target/ARM/Thumb1FrameLowering.h b/lib/Target/ARM/Thumb1FrameLowering.h
index bcfc5165fad0..5a300afd5d36 100644
--- a/lib/Target/ARM/Thumb1FrameLowering.h
+++ b/lib/Target/ARM/Thumb1FrameLowering.h
@@ -45,6 +45,10 @@ public:
                                    const TargetRegisterInfo *TRI) const;
 
   bool hasReservedCallFrame(const MachineFunction &MF) const;
+
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator MI) const;
 };
 
 } // End llvm namespace
diff --git a/lib/Target/ARM/Thumb1InstrInfo.cpp b/lib/Target/ARM/Thumb1InstrInfo.cpp
index 735b255759b7..095736d52a88 100644
--- a/lib/Target/ARM/Thumb1InstrInfo.cpp
+++ b/lib/Target/ARM/Thumb1InstrInfo.cpp
@@ -15,8 +15,8 @@
 #include "ARM.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/MachineMemOperand.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/MC/MCInst.h"
 
 using namespace llvm;
diff --git a/lib/Target/ARM/Thumb1RegisterInfo.cpp b/lib/Target/ARM/Thumb1RegisterInfo.cpp
index a39b722caef5..7452fb776ebd 100644
--- a/lib/Target/ARM/Thumb1RegisterInfo.cpp
+++ b/lib/Target/ARM/Thumb1RegisterInfo.cpp
@@ -18,21 +18,21 @@
 #include "ARMMachineFunctionInfo.h"
 #include "ARMSubtarget.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/LLVMContext.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetMachine.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/LLVMContext.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetMachine.h"
 
 namespace llvm {
 extern cl::opt<bool> ReuseFrameIndexVals;
@@ -296,47 +296,6 @@ void llvm::emitThumbRegPlusImmediate(MachineBasicBlock &MBB,
   }
 }
 
-static void emitSPUpdate(MachineBasicBlock &MBB,
-                         MachineBasicBlock::iterator &MBBI,
-                         const TargetInstrInfo &TII, DebugLoc dl,
-                         const Thumb1RegisterInfo &MRI,
-                         int NumBytes) {
-  emitThumbRegPlusImmediate(MBB, MBBI, dl, ARM::SP, ARM::SP, NumBytes, TII,
-                            MRI);
-}
-
-void Thumb1RegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-
-  if (!TFI->hasReservedCallFrame(MF)) {
-    // If we have alloca, convert as follows:
-    // ADJCALLSTACKDOWN -> sub, sp, sp, amount
-    // ADJCALLSTACKUP   -> add, sp, sp, amount
-    MachineInstr *Old = I;
-    DebugLoc dl = Old->getDebugLoc();
-    unsigned Amount = Old->getOperand(0).getImm();
-    if (Amount != 0) {
-      // We need to keep the stack aligned properly.  To do this, we round the
-      // amount of space needed for the outgoing arguments up to the next
-      // alignment boundary.
-      unsigned Align = TFI->getStackAlignment();
-      Amount = (Amount+Align-1)/Align*Align;
-
-      // Replace the pseudo instruction with a new instruction...
-      unsigned Opc = Old->getOpcode();
-      if (Opc == ARM::ADJCALLSTACKDOWN || Opc == ARM::tADJCALLSTACKDOWN) {
-        emitSPUpdate(MBB, I, TII, dl, *this, -Amount);
-      } else {
-        assert(Opc == ARM::ADJCALLSTACKUP || Opc == ARM::tADJCALLSTACKUP);
-        emitSPUpdate(MBB, I, TII, dl, *this, Amount);
-      }
-    }
-  }
-  MBB.erase(I);
-}
-
 /// emitThumbConstant - Emit a series of instructions to materialize a
 /// constant.
 static void emitThumbConstant(MachineBasicBlock &MBB,
@@ -390,6 +349,7 @@ rewriteFrameIndex(MachineBasicBlock::iterator II, unsigned FrameRegIdx,
   MachineInstr &MI = *II;
   MachineBasicBlock &MBB = *MI.getParent();
   DebugLoc dl = MI.getDebugLoc();
+  MachineInstrBuilder MIB(*MBB.getParent(), &MI);
   unsigned Opcode = MI.getOpcode();
   const MCInstrDesc &Desc = MI.getDesc();
   unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask);
@@ -417,7 +377,6 @@ rewriteFrameIndex(MachineBasicBlock::iterator II, unsigned FrameRegIdx,
       MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false);
       // Remove offset
       MI.RemoveOperand(FrameRegIdx+1);
-      MachineInstrBuilder MIB(&MI);
       return true;
     }
 
@@ -428,7 +387,6 @@ rewriteFrameIndex(MachineBasicBlock::iterator II, unsigned FrameRegIdx,
       if (Opcode == ARM::tADDi3) {
         MI.setDesc(TII.get(Opcode));
         removeOperands(MI, FrameRegIdx);
-        MachineInstrBuilder MIB(&MI);
         AddDefaultPred(AddDefaultT1CC(MIB).addReg(FrameReg)
                        .addImm(Offset / Scale));
       } else {
@@ -457,7 +415,6 @@ rewriteFrameIndex(MachineBasicBlock::iterator II, unsigned FrameRegIdx,
       if (Opcode == ARM::tADDi3) {
         MI.setDesc(TII.get(Opcode));
         removeOperands(MI, FrameRegIdx);
-        MachineInstrBuilder MIB(&MI);
         AddDefaultPred(AddDefaultT1CC(MIB).addReg(FrameReg).addImm(Mask));
       } else {
         MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false);
@@ -595,22 +552,18 @@ Thumb1RegisterInfo::saveScavengerRegister(MachineBasicBlock &MBB,
 
 void
 Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
-                                        int SPAdj, RegScavenger *RS) const {
+                                        int SPAdj, unsigned FIOperandNum,
+                                        RegScavenger *RS) const {
   unsigned VReg = 0;
-  unsigned i = 0;
   MachineInstr &MI = *II;
   MachineBasicBlock &MBB = *MI.getParent();
   MachineFunction &MF = *MBB.getParent();
   ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
   DebugLoc dl = MI.getDebugLoc();
-
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
-  }
+  MachineInstrBuilder MIB(*MBB.getParent(), &MI);
 
   unsigned FrameReg = ARM::SP;
-  int FrameIndex = MI.getOperand(i).getIndex();
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
   int Offset = MF.getFrameInfo()->getObjectOffset(FrameIndex) +
                MF.getFrameInfo()->getStackSize() + SPAdj;
 
@@ -635,7 +588,7 @@ Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   // means the stack pointer cannot be used to access the emergency spill slot
   // when !hasReservedCallFrame().
 #ifndef NDEBUG
-  if (RS && FrameReg == ARM::SP && FrameIndex == RS->getScavengingFrameIndex()){
+  if (RS && FrameReg == ARM::SP && RS->isScavengingFrameIndex(FrameIndex)){
     assert(MF.getTarget().getFrameLowering()->hasReservedCallFrame(MF) &&
            "Cannot use SP to access the emergency spill slot in "
            "functions without a reserved call frame");
@@ -647,15 +600,15 @@ Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
 
   // Special handling of dbg_value instructions.
   if (MI.isDebugValue()) {
-    MI.getOperand(i).  ChangeToRegister(FrameReg, false /*isDef*/);
-    MI.getOperand(i+1).ChangeToImmediate(Offset);
+    MI.getOperand(FIOperandNum).  ChangeToRegister(FrameReg, false /*isDef*/);
+    MI.getOperand(FIOperandNum+1).ChangeToImmediate(Offset);
     return;
   }
 
   // Modify MI as necessary to handle as much of 'Offset' as possible
   assert(AFI->isThumbFunction() &&
          "This eliminateFrameIndex only supports Thumb1!");
-  if (rewriteFrameIndex(MI, i, FrameReg, Offset, TII))
+  if (rewriteFrameIndex(MI, FIOperandNum, FrameReg, Offset, TII))
     return;
 
   // If we get here, the immediate doesn't fit into the instruction.  We folded
@@ -688,11 +641,12 @@ Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
     }
 
     MI.setDesc(TII.get(UseRR ? ARM::tLDRr : ARM::tLDRi));
-    MI.getOperand(i).ChangeToRegister(TmpReg, false, false, true);
+    MI.getOperand(FIOperandNum).ChangeToRegister(TmpReg, false, false, true);
     if (UseRR)
       // Use [reg, reg] addrmode. Replace the immediate operand w/ the frame
       // register. The offset is already handled in the vreg value.
-      MI.getOperand(i+1).ChangeToRegister(FrameReg, false, false, false);
+      MI.getOperand(FIOperandNum+1).ChangeToRegister(FrameReg, false, false,
+                                                     false);
   } else if (MI.mayStore()) {
       VReg = MF.getRegInfo().createVirtualRegister(&ARM::tGPRRegClass);
       bool UseRR = false;
@@ -709,18 +663,17 @@ Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
         emitThumbRegPlusImmediate(MBB, II, dl, VReg, FrameReg, Offset, TII,
                                   *this);
       MI.setDesc(TII.get(UseRR ? ARM::tSTRr : ARM::tSTRi));
-      MI.getOperand(i).ChangeToRegister(VReg, false, false, true);
+      MI.getOperand(FIOperandNum).ChangeToRegister(VReg, false, false, true);
       if (UseRR)
         // Use [reg, reg] addrmode. Replace the immediate operand w/ the frame
         // register. The offset is already handled in the vreg value.
-        MI.getOperand(i+1).ChangeToRegister(FrameReg, false, false, false);
+        MI.getOperand(FIOperandNum+1).ChangeToRegister(FrameReg, false, false,
+                                                       false);
   } else {
     llvm_unreachable("Unexpected opcode!");
   }
 
   // Add predicate back if it's needed.
-  if (MI.isPredicable()) {
-    MachineInstrBuilder MIB(&MI);
+  if (MI.isPredicable())
     AddDefaultPred(MIB);
-  }
 }
diff --git a/lib/Target/ARM/Thumb1RegisterInfo.h b/lib/Target/ARM/Thumb1RegisterInfo.h
index f2e4b08f798e..ebbab36dd7b8 100644
--- a/lib/Target/ARM/Thumb1RegisterInfo.h
+++ b/lib/Target/ARM/Thumb1RegisterInfo.h
@@ -43,11 +43,6 @@ public:
                         unsigned PredReg = 0,
                         unsigned MIFlags = MachineInstr::NoFlags) const;
 
-  /// Code Generation virtual methods...
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
-
   // rewrite MI to access 'Offset' bytes from the FP. Update Offset to be
   // however much remains to be handled. Return 'true' if no further
   // work is required.
@@ -62,7 +57,8 @@ public:
                              const TargetRegisterClass *RC,
                              unsigned Reg) const;
   void eliminateFrameIndex(MachineBasicBlock::iterator II,
-                           int SPAdj, RegScavenger *RS = NULL) const;
+                           int SPAdj, unsigned FIOperandNum,
+                           RegScavenger *RS = NULL) const;
 };
 }
 
diff --git a/lib/Target/ARM/Thumb2ITBlockPass.cpp b/lib/Target/ARM/Thumb2ITBlockPass.cpp
index d54aa935325c..97c254ce75a5 100644
--- a/lib/Target/ARM/Thumb2ITBlockPass.cpp
+++ b/lib/Target/ARM/Thumb2ITBlockPass.cpp
@@ -11,12 +11,12 @@
 #include "ARM.h"
 #include "ARMMachineFunctionInfo.h"
 #include "Thumb2InstrInfo.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineInstrBundle.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/ADT/SmallSet.h"
-#include "llvm/ADT/Statistic.h"
 using namespace llvm;
 
 STATISTIC(NumITs,        "Number of IT blocks inserted");
diff --git a/lib/Target/ARM/Thumb2InstrInfo.cpp b/lib/Target/ARM/Thumb2InstrInfo.cpp
index e9e20ddd8783..67e8ec7c5ff2 100644
--- a/lib/Target/ARM/Thumb2InstrInfo.cpp
+++ b/lib/Target/ARM/Thumb2InstrInfo.cpp
@@ -51,7 +51,7 @@ Thumb2InstrInfo::ReplaceTailWithBranchTo(MachineBasicBlock::iterator Tail,
   MachineBasicBlock *MBB = Tail->getParent();
   ARMFunctionInfo *AFI = MBB->getParent()->getInfo<ARMFunctionInfo>();
   if (!AFI->hasITBlocks()) {
-    TargetInstrInfoImpl::ReplaceTailWithBranchTo(Tail, NewDest);
+    TargetInstrInfo::ReplaceTailWithBranchTo(Tail, NewDest);
     return;
   }
 
@@ -65,7 +65,7 @@ Thumb2InstrInfo::ReplaceTailWithBranchTo(MachineBasicBlock::iterator Tail,
     --MBBI;
 
   // Actually replace the tail.
-  TargetInstrInfoImpl::ReplaceTailWithBranchTo(Tail, NewDest);
+  TargetInstrInfo::ReplaceTailWithBranchTo(Tail, NewDest);
 
   // Fix up IT.
   if (CC != ARMCC::AL) {
@@ -408,7 +408,7 @@ bool llvm::rewriteT2FrameIndex(MachineInstr &MI, unsigned FrameRegIdx,
       // Remove offset and remaining explicit predicate operands.
       do MI.RemoveOperand(FrameRegIdx+1);
       while (MI.getNumOperands() > FrameRegIdx+1);
-      MachineInstrBuilder MIB(&MI);
+      MachineInstrBuilder MIB(*MI.getParent()->getParent(), &MI);
       AddDefaultPred(MIB);
       return true;
     }
diff --git a/lib/Target/ARM/Thumb2RegisterInfo.cpp b/lib/Target/ARM/Thumb2RegisterInfo.cpp
index 29a87d016227..1a7a4d450cfe 100644
--- a/lib/Target/ARM/Thumb2RegisterInfo.cpp
+++ b/lib/Target/ARM/Thumb2RegisterInfo.cpp
@@ -16,12 +16,12 @@
 #include "ARM.h"
 #include "ARMBaseInstrInfo.h"
 #include "ARMSubtarget.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
 using namespace llvm;
 
 Thumb2RegisterInfo::Thumb2RegisterInfo(const ARMBaseInstrInfo &tii,
diff --git a/lib/Target/ARM/Thumb2SizeReduction.cpp b/lib/Target/ARM/Thumb2SizeReduction.cpp
index f18f491f4995..d50f5d972232 100644
--- a/lib/Target/ARM/Thumb2SizeReduction.cpp
+++ b/lib/Target/ARM/Thumb2SizeReduction.cpp
@@ -9,19 +9,21 @@
 
 #define DEBUG_TYPE "t2-reduce-size"
 #include "ARM.h"
-#include "ARMBaseRegisterInfo.h"
 #include "ARMBaseInstrInfo.h"
+#include "ARMBaseRegisterInfo.h"
 #include "ARMSubtarget.h"
-#include "Thumb2InstrInfo.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
+#include "Thumb2InstrInfo.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/PostOrderIterator.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/IR/Function.h"        // To access Function attributes
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/Statistic.h"
 using namespace llvm;
 
 STATISTIC(NumNarrows,  "Number of 32-bit instrs reduced to 16-bit ones");
@@ -52,81 +54,79 @@ namespace {
     unsigned PredCC2  : 2;
     unsigned PartFlag : 1; // 16-bit instruction does partial flag update
     unsigned Special  : 1; // Needs to be dealt with specially
+    unsigned AvoidMovs: 1; // Avoid movs with shifter operand (for Swift)
   };
 
   static const ReduceEntry ReduceTable[] = {
-    // Wide,        Narrow1,      Narrow2,     imm1,imm2,  lo1, lo2, P/C, PF, S
-    { ARM::t2ADCrr, 0,            ARM::tADC,     0,   0,    0,   1,  0,0, 0,0 },
-    { ARM::t2ADDri, ARM::tADDi3,  ARM::tADDi8,   3,   8,    1,   1,  0,0, 0,1 },
-    { ARM::t2ADDrr, ARM::tADDrr,  ARM::tADDhirr, 0,   0,    1,   0,  0,1, 0,0 },
-    { ARM::t2ADDSri,ARM::tADDi3,  ARM::tADDi8,   3,   8,    1,   1,  2,2, 0,1 },
-    { ARM::t2ADDSrr,ARM::tADDrr,  0,             0,   0,    1,   0,  2,0, 0,1 },
-    { ARM::t2ANDrr, 0,            ARM::tAND,     0,   0,    0,   1,  0,0, 1,0 },
-    { ARM::t2ASRri, ARM::tASRri,  0,             5,   0,    1,   0,  0,0, 1,0 },
-    { ARM::t2ASRrr, 0,            ARM::tASRrr,   0,   0,    0,   1,  0,0, 1,0 },
-    { ARM::t2BICrr, 0,            ARM::tBIC,     0,   0,    0,   1,  0,0, 1,0 },
-    //FIXME: Disable CMN, as CCodes are backwards from compare expectations
-    //{ ARM::t2CMNrr, ARM::tCMN,  0,             0,   0,    1,   0,  2,0, 0,0 },
-    { ARM::t2CMNzrr, ARM::tCMNz,  0,             0,   0,    1,   0,  2,0, 0,0 },
-    { ARM::t2CMPri, ARM::tCMPi8,  0,             8,   0,    1,   0,  2,0, 0,0 },
-    { ARM::t2CMPrr, ARM::tCMPhir, 0,             0,   0,    0,   0,  2,0, 0,1 },
-    { ARM::t2EORrr, 0,            ARM::tEOR,     0,   0,    0,   1,  0,0, 1,0 },
-    // FIXME: adr.n immediate offset must be multiple of 4.
-    //{ ARM::t2LEApcrelJT,ARM::tLEApcrelJT, 0,   0,   0,    1,   0,  1,0, 0,0 },
-    { ARM::t2LSLri, ARM::tLSLri,  0,             5,   0,    1,   0,  0,0, 1,0 },
-    { ARM::t2LSLrr, 0,            ARM::tLSLrr,   0,   0,    0,   1,  0,0, 1,0 },
-    { ARM::t2LSRri, ARM::tLSRri,  0,             5,   0,    1,   0,  0,0, 1,0 },
-    { ARM::t2LSRrr, 0,            ARM::tLSRrr,   0,   0,    0,   1,  0,0, 1,0 },
-    // FIXME: tMOVi8 and tMVN also partially update CPSR but they are less
-    // likely to cause issue in the loop. As a size / performance workaround,
-    // they are not marked as such.
-    { ARM::t2MOVi,  ARM::tMOVi8,  0,             8,   0,    1,   0,  0,0, 0,0 },
-    { ARM::t2MOVi16,ARM::tMOVi8,  0,             8,   0,    1,   0,  0,0, 0,1 },
-    // FIXME: Do we need the 16-bit 'S' variant?
-    { ARM::t2MOVr,ARM::tMOVr,     0,             0,   0,    0,   0,  1,0, 0,0 },
-    { ARM::t2MUL,   0,            ARM::tMUL,     0,   0,    0,   1,  0,0, 1,0 },
-    { ARM::t2MVNr,  ARM::tMVN,    0,             0,   0,    1,   0,  0,0, 0,0 },
-    { ARM::t2ORRrr, 0,            ARM::tORR,     0,   0,    0,   1,  0,0, 1,0 },
-    { ARM::t2REV,   ARM::tREV,    0,             0,   0,    1,   0,  1,0, 0,0 },
-    { ARM::t2REV16, ARM::tREV16,  0,             0,   0,    1,   0,  1,0, 0,0 },
-    { ARM::t2REVSH, ARM::tREVSH,  0,             0,   0,    1,   0,  1,0, 0,0 },
-    { ARM::t2RORrr, 0,            ARM::tROR,     0,   0,    0,   1,  0,0, 1,0 },
-    { ARM::t2RSBri, ARM::tRSB,    0,             0,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2RSBSri,ARM::tRSB,    0,             0,   0,    1,   0,  2,0, 0,1 },
-    { ARM::t2SBCrr, 0,            ARM::tSBC,     0,   0,    0,   1,  0,0, 0,0 },
-    { ARM::t2SUBri, ARM::tSUBi3,  ARM::tSUBi8,   3,   8,    1,   1,  0,0, 0,0 },
-    { ARM::t2SUBrr, ARM::tSUBrr,  0,             0,   0,    1,   0,  0,0, 0,0 },
-    { ARM::t2SUBSri,ARM::tSUBi3,  ARM::tSUBi8,   3,   8,    1,   1,  2,2, 0,0 },
-    { ARM::t2SUBSrr,ARM::tSUBrr,  0,             0,   0,    1,   0,  2,0, 0,0 },
-    { ARM::t2SXTB,  ARM::tSXTB,   0,             0,   0,    1,   0,  1,0, 0,1 },
-    { ARM::t2SXTH,  ARM::tSXTH,   0,             0,   0,    1,   0,  1,0, 0,1 },
-    { ARM::t2TSTrr, ARM::tTST,    0,             0,   0,    1,   0,  2,0, 0,0 },
-    { ARM::t2UXTB,  ARM::tUXTB,   0,             0,   0,    1,   0,  1,0, 0,1 },
-    { ARM::t2UXTH,  ARM::tUXTH,   0,             0,   0,    1,   0,  1,0, 0,1 },
-
-    // FIXME: Clean this up after splitting each Thumb load / store opcode
-    // into multiple ones.
-    { ARM::t2LDRi12,ARM::tLDRi,   ARM::tLDRspi,  5,   8,    1,   0,  0,0, 0,1 },
-    { ARM::t2LDRs,  ARM::tLDRr,   0,             0,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2LDRBi12,ARM::tLDRBi, 0,             5,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2LDRBs, ARM::tLDRBr,  0,             0,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2LDRHi12,ARM::tLDRHi, 0,             5,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2LDRHs, ARM::tLDRHr,  0,             0,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2LDRSBs,ARM::tLDRSB,  0,             0,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2LDRSHs,ARM::tLDRSH,  0,             0,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2STRi12,ARM::tSTRi,   ARM::tSTRspi,  5,   8,    1,   0,  0,0, 0,1 },
-    { ARM::t2STRs,  ARM::tSTRr,   0,             0,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2STRBi12,ARM::tSTRBi, 0,             5,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2STRBs, ARM::tSTRBr,  0,             0,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2STRHi12,ARM::tSTRHi, 0,             5,   0,    1,   0,  0,0, 0,1 },
-    { ARM::t2STRHs, ARM::tSTRHr,  0,             0,   0,    1,   0,  0,0, 0,1 },
-
-    { ARM::t2LDMIA, ARM::tLDMIA,  0,             0,   0,    1,   1,  1,1, 0,1 },
-    { ARM::t2LDMIA_RET,0,         ARM::tPOP_RET, 0,   0,    1,   1,  1,1, 0,1 },
-    { ARM::t2LDMIA_UPD,ARM::tLDMIA_UPD,ARM::tPOP,0,   0,    1,   1,  1,1, 0,1 },
-    // ARM::t2STM (with no basereg writeback) has no Thumb1 equivalent
-    { ARM::t2STMIA_UPD,ARM::tSTMIA_UPD, 0,       0,   0,    1,   1,  1,1, 0,1 },
-    { ARM::t2STMDB_UPD, 0,        ARM::tPUSH,    0,   0,    1,   1,  1,1, 0,1 },
+  // Wide,        Narrow1,      Narrow2,     imm1,imm2, lo1, lo2, P/C,PF,S,AM
+  { ARM::t2ADCrr, 0,            ARM::tADC,     0,   0,   0,   1,  0,0, 0,0,0 },
+  { ARM::t2ADDri, ARM::tADDi3,  ARM::tADDi8,   3,   8,   1,   1,  0,0, 0,1,0 },
+  { ARM::t2ADDrr, ARM::tADDrr,  ARM::tADDhirr, 0,   0,   1,   0,  0,1, 0,0,0 },
+  { ARM::t2ADDSri,ARM::tADDi3,  ARM::tADDi8,   3,   8,   1,   1,  2,2, 0,1,0 },
+  { ARM::t2ADDSrr,ARM::tADDrr,  0,             0,   0,   1,   0,  2,0, 0,1,0 },
+  { ARM::t2ANDrr, 0,            ARM::tAND,     0,   0,   0,   1,  0,0, 1,0,0 },
+  { ARM::t2ASRri, ARM::tASRri,  0,             5,   0,   1,   0,  0,0, 1,0,1 },
+  { ARM::t2ASRrr, 0,            ARM::tASRrr,   0,   0,   0,   1,  0,0, 1,0,1 },
+  { ARM::t2BICrr, 0,            ARM::tBIC,     0,   0,   0,   1,  0,0, 1,0,0 },
+  //FIXME: Disable CMN, as CCodes are backwards from compare expectations
+  //{ ARM::t2CMNrr, ARM::tCMN,  0,             0,   0,   1,   0,  2,0, 0,0,0 },
+  { ARM::t2CMNzrr, ARM::tCMNz,  0,             0,   0,   1,   0,  2,0, 0,0,0 },
+  { ARM::t2CMPri, ARM::tCMPi8,  0,             8,   0,   1,   0,  2,0, 0,0,0 },
+  { ARM::t2CMPrr, ARM::tCMPhir, 0,             0,   0,   0,   0,  2,0, 0,1,0 },
+  { ARM::t2EORrr, 0,            ARM::tEOR,     0,   0,   0,   1,  0,0, 1,0,0 },
+  // FIXME: adr.n immediate offset must be multiple of 4.
+  //{ ARM::t2LEApcrelJT,ARM::tLEApcrelJT, 0,   0,   0,   1,   0,  1,0, 0,0,0 },
+  { ARM::t2LSLri, ARM::tLSLri,  0,             5,   0,   1,   0,  0,0, 1,0,1 },
+  { ARM::t2LSLrr, 0,            ARM::tLSLrr,   0,   0,   0,   1,  0,0, 1,0,1 },
+  { ARM::t2LSRri, ARM::tLSRri,  0,             5,   0,   1,   0,  0,0, 1,0,1 },
+  { ARM::t2LSRrr, 0,            ARM::tLSRrr,   0,   0,   0,   1,  0,0, 1,0,1 },
+  { ARM::t2MOVi,  ARM::tMOVi8,  0,             8,   0,   1,   0,  0,0, 1,0,0 },
+  { ARM::t2MOVi16,ARM::tMOVi8,  0,             8,   0,   1,   0,  0,0, 1,1,0 },
+  // FIXME: Do we need the 16-bit 'S' variant?
+  { ARM::t2MOVr,ARM::tMOVr,     0,             0,   0,   0,   0,  1,0, 0,0,0 },
+  { ARM::t2MUL,   0,            ARM::tMUL,     0,   0,   0,   1,  0,0, 1,0,0 },
+  { ARM::t2MVNr,  ARM::tMVN,    0,             0,   0,   1,   0,  0,0, 0,0,0 },
+  { ARM::t2ORRrr, 0,            ARM::tORR,     0,   0,   0,   1,  0,0, 1,0,0 },
+  { ARM::t2REV,   ARM::tREV,    0,             0,   0,   1,   0,  1,0, 0,0,0 },
+  { ARM::t2REV16, ARM::tREV16,  0,             0,   0,   1,   0,  1,0, 0,0,0 },
+  { ARM::t2REVSH, ARM::tREVSH,  0,             0,   0,   1,   0,  1,0, 0,0,0 },
+  { ARM::t2RORrr, 0,            ARM::tROR,     0,   0,   0,   1,  0,0, 1,0,0 },
+  { ARM::t2RSBri, ARM::tRSB,    0,             0,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2RSBSri,ARM::tRSB,    0,             0,   0,   1,   0,  2,0, 0,1,0 },
+  { ARM::t2SBCrr, 0,            ARM::tSBC,     0,   0,   0,   1,  0,0, 0,0,0 },
+  { ARM::t2SUBri, ARM::tSUBi3,  ARM::tSUBi8,   3,   8,   1,   1,  0,0, 0,0,0 },
+  { ARM::t2SUBrr, ARM::tSUBrr,  0,             0,   0,   1,   0,  0,0, 0,0,0 },
+  { ARM::t2SUBSri,ARM::tSUBi3,  ARM::tSUBi8,   3,   8,   1,   1,  2,2, 0,0,0 },
+  { ARM::t2SUBSrr,ARM::tSUBrr,  0,             0,   0,   1,   0,  2,0, 0,0,0 },
+  { ARM::t2SXTB,  ARM::tSXTB,   0,             0,   0,   1,   0,  1,0, 0,1,0 },
+  { ARM::t2SXTH,  ARM::tSXTH,   0,             0,   0,   1,   0,  1,0, 0,1,0 },
+  { ARM::t2TSTrr, ARM::tTST,    0,             0,   0,   1,   0,  2,0, 0,0,0 },
+  { ARM::t2UXTB,  ARM::tUXTB,   0,             0,   0,   1,   0,  1,0, 0,1,0 },
+  { ARM::t2UXTH,  ARM::tUXTH,   0,             0,   0,   1,   0,  1,0, 0,1,0 },
+
+  // FIXME: Clean this up after splitting each Thumb load / store opcode
+  // into multiple ones.
+  { ARM::t2LDRi12,ARM::tLDRi,   ARM::tLDRspi,  5,   8,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2LDRs,  ARM::tLDRr,   0,             0,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2LDRBi12,ARM::tLDRBi, 0,             5,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2LDRBs, ARM::tLDRBr,  0,             0,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2LDRHi12,ARM::tLDRHi, 0,             5,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2LDRHs, ARM::tLDRHr,  0,             0,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2LDRSBs,ARM::tLDRSB,  0,             0,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2LDRSHs,ARM::tLDRSH,  0,             0,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2STRi12,ARM::tSTRi,   ARM::tSTRspi,  5,   8,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2STRs,  ARM::tSTRr,   0,             0,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2STRBi12,ARM::tSTRBi, 0,             5,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2STRBs, ARM::tSTRBr,  0,             0,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2STRHi12,ARM::tSTRHi, 0,             5,   0,   1,   0,  0,0, 0,1,0 },
+  { ARM::t2STRHs, ARM::tSTRHr,  0,             0,   0,   1,   0,  0,0, 0,1,0 },
+
+  { ARM::t2LDMIA, ARM::tLDMIA,  0,             0,   0,   1,   1,  1,1, 0,1,0 },
+  { ARM::t2LDMIA_RET,0,         ARM::tPOP_RET, 0,   0,   1,   1,  1,1, 0,1,0 },
+  { ARM::t2LDMIA_UPD,ARM::tLDMIA_UPD,ARM::tPOP,0,   0,   1,   1,  1,1, 0,1,0 },
+  // ARM::t2STM (with no basereg writeback) has no Thumb1 equivalent
+  { ARM::t2STMIA_UPD,ARM::tSTMIA_UPD, 0,       0,   0,   1,   1,  1,1, 0,1,0 },
+  { ARM::t2STMDB_UPD, 0,        ARM::tPUSH,    0,   0,   1,   1,  1,1, 0,1,0 }
   };
 
   class Thumb2SizeReduce : public MachineFunctionPass {
@@ -147,8 +147,7 @@ namespace {
     /// ReduceOpcodeMap - Maps wide opcode to index of entry in ReduceTable.
     DenseMap<unsigned, unsigned> ReduceOpcodeMap;
 
-    bool canAddPseudoFlagDep(MachineInstr *Def, MachineInstr *Use,
-                             bool IsSelfLoop);
+    bool canAddPseudoFlagDep(MachineInstr *Use, bool IsSelfLoop);
 
     bool VerifyPredAndCC(MachineInstr *MI, const ReduceEntry &Entry,
                          bool is2Addr, ARMCC::CondCodes Pred,
@@ -158,30 +157,52 @@ namespace {
                          const ReduceEntry &Entry);
 
     bool ReduceSpecial(MachineBasicBlock &MBB, MachineInstr *MI,
-                       const ReduceEntry &Entry, bool LiveCPSR,
-                       MachineInstr *CPSRDef, bool IsSelfLoop);
+                       const ReduceEntry &Entry, bool LiveCPSR, bool IsSelfLoop);
 
     /// ReduceTo2Addr - Reduce a 32-bit instruction to a 16-bit two-address
     /// instruction.
     bool ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI,
-                       const ReduceEntry &Entry,
-                       bool LiveCPSR, MachineInstr *CPSRDef,
+                       const ReduceEntry &Entry, bool LiveCPSR,
                        bool IsSelfLoop);
 
     /// ReduceToNarrow - Reduce a 32-bit instruction to a 16-bit
     /// non-two-address instruction.
     bool ReduceToNarrow(MachineBasicBlock &MBB, MachineInstr *MI,
-                        const ReduceEntry &Entry,
-                        bool LiveCPSR, MachineInstr *CPSRDef,
+                        const ReduceEntry &Entry, bool LiveCPSR,
                         bool IsSelfLoop);
 
+    /// ReduceMI - Attempt to reduce MI, return true on success.
+    bool ReduceMI(MachineBasicBlock &MBB, MachineInstr *MI,
+                  bool LiveCPSR, bool IsSelfLoop);
+
     /// ReduceMBB - Reduce width of instructions in the specified basic block.
     bool ReduceMBB(MachineBasicBlock &MBB);
+
+    bool OptimizeSize;
+    bool MinimizeSize;
+
+    // Last instruction to define CPSR in the current block.
+    MachineInstr *CPSRDef;
+    // Was CPSR last defined by a high latency instruction?
+    // When CPSRDef is null, this refers to CPSR defs in predecessors.
+    bool HighLatencyCPSR;
+
+    struct MBBInfo {
+      // The flags leaving this block have high latency.
+      bool HighLatencyCPSR;
+      // Has this block been visited yet?
+      bool Visited;
+
+      MBBInfo() : HighLatencyCPSR(false), Visited(false) {}
+    };
+
+    SmallVector<MBBInfo, 8> BlockInfo;
   };
   char Thumb2SizeReduce::ID = 0;
 }
 
 Thumb2SizeReduce::Thumb2SizeReduce() : MachineFunctionPass(ID) {
+  OptimizeSize = MinimizeSize = false;
   for (unsigned i = 0, e = array_lengthof(ReduceTable); i != e; ++i) {
     unsigned FromOpc = ReduceTable[i].WideOpc;
     if (!ReduceOpcodeMap.insert(std::make_pair(FromOpc, i)).second)
@@ -196,6 +217,16 @@ static bool HasImplicitCPSRDef(const MCInstrDesc &MCID) {
   return false;
 }
 
+// Check for a likely high-latency flag def.
+static bool isHighLatencyCPSR(MachineInstr *Def) {
+  switch(Def->getOpcode()) {
+  case ARM::FMSTAT:
+  case ARM::tMUL:
+    return true;
+  }
+  return false;
+}
+
 /// canAddPseudoFlagDep - For A9 (and other out-of-order) implementations,
 /// the 's' 16-bit instruction partially update CPSR. Abort the
 /// transformation to avoid adding false dependency on last CPSR setting
@@ -214,20 +245,19 @@ static bool HasImplicitCPSRDef(const MCInstrDesc &MCID) {
 /// In this case it would have been ok to narrow the mul.w to muls since there
 /// are indirect RAW dependency between the muls and the mul.w
 bool
-Thumb2SizeReduce::canAddPseudoFlagDep(MachineInstr *Def, MachineInstr *Use,
-                                      bool FirstInSelfLoop) {
-  // FIXME: Disable check for -Oz (aka OptimizeForSizeHarder).
-  if (!STI->avoidCPSRPartialUpdate())
+Thumb2SizeReduce::canAddPseudoFlagDep(MachineInstr *Use, bool FirstInSelfLoop) {
+  // Disable the check for -Oz (aka OptimizeForSizeHarder).
+  if (MinimizeSize || !STI->avoidCPSRPartialUpdate())
     return false;
 
-  if (!Def)
+  if (!CPSRDef)
     // If this BB loops back to itself, conservatively avoid narrowing the
     // first instruction that does partial flag update.
-    return FirstInSelfLoop;
+    return HighLatencyCPSR || FirstInSelfLoop;
 
   SmallSet<unsigned, 2> Defs;
-  for (unsigned i = 0, e = Def->getNumOperands(); i != e; ++i) {
-    const MachineOperand &MO = Def->getOperand(i);
+  for (unsigned i = 0, e = CPSRDef->getNumOperands(); i != e; ++i) {
+    const MachineOperand &MO = CPSRDef->getOperand(i);
     if (!MO.isReg() || MO.isUndef() || MO.isUse())
       continue;
     unsigned Reg = MO.getReg();
@@ -245,6 +275,16 @@ Thumb2SizeReduce::canAddPseudoFlagDep(MachineInstr *Def, MachineInstr *Use,
       return false;
   }
 
+  // If the current CPSR has high latency, try to avoid the false dependency.
+  if (HighLatencyCPSR)
+    return true;
+
+  // tMOVi8 usually doesn't start long dependency chains, and there are a lot
+  // of them, so always shrink them when CPSR doesn't have high latency.
+  if (Use->getOpcode() == ARM::t2MOVi ||
+      Use->getOpcode() == ARM::t2MOVi16)
+    return false;
+
   // No read-after-write dependency. The narrowing will add false dependency.
   return true;
 }
@@ -487,16 +527,15 @@ Thumb2SizeReduce::ReduceLoadStore(MachineBasicBlock &MBB, MachineInstr *MI,
 bool
 Thumb2SizeReduce::ReduceSpecial(MachineBasicBlock &MBB, MachineInstr *MI,
                                 const ReduceEntry &Entry,
-                                bool LiveCPSR, MachineInstr *CPSRDef,
-                                bool IsSelfLoop) {
+                                bool LiveCPSR, bool IsSelfLoop) {
   unsigned Opc = MI->getOpcode();
   if (Opc == ARM::t2ADDri) {
     // If the source register is SP, try to reduce to tADDrSPi, otherwise
     // it's a normal reduce.
     if (MI->getOperand(1).getReg() != ARM::SP) {
-      if (ReduceTo2Addr(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop))
+      if (ReduceTo2Addr(MBB, MI, Entry, LiveCPSR, IsSelfLoop))
         return true;
-      return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop);
+      return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, IsSelfLoop);
     }
     // Try to reduce to tADDrSPi.
     unsigned Imm = MI->getOperand(2).getImm();
@@ -546,12 +585,12 @@ Thumb2SizeReduce::ReduceSpecial(MachineBasicBlock &MBB, MachineInstr *MI,
       switch (Opc) {
       default: break;
       case ARM::t2ADDSri: {
-        if (ReduceTo2Addr(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop))
+        if (ReduceTo2Addr(MBB, MI, Entry, LiveCPSR, IsSelfLoop))
           return true;
         // fallthrough
       }
       case ARM::t2ADDSrr:
-        return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop);
+        return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, IsSelfLoop);
       }
     }
     break;
@@ -563,13 +602,13 @@ Thumb2SizeReduce::ReduceSpecial(MachineBasicBlock &MBB, MachineInstr *MI,
   case ARM::t2UXTB:
   case ARM::t2UXTH:
     if (MI->getOperand(2).getImm() == 0)
-      return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop);
+      return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, IsSelfLoop);
     break;
   case ARM::t2MOVi16:
     // Can convert only 'pure' immediate operands, not immediates obtained as
     // globals' addresses.
     if (MI->getOperand(1).isImm())
-      return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop);
+      return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, IsSelfLoop);
     break;
   case ARM::t2CMPrr: {
     // Try to reduce to the lo-reg only version first. Why there are two
@@ -578,10 +617,10 @@ Thumb2SizeReduce::ReduceSpecial(MachineBasicBlock &MBB, MachineInstr *MI,
     // are prioritized, but the table assumes a unique entry for each
     // source insn opcode. So for now, we hack a local entry record to use.
     static const ReduceEntry NarrowEntry =
-      { ARM::t2CMPrr,ARM::tCMPr, 0, 0, 0, 1, 1,2, 0, 0,1 };
-    if (ReduceToNarrow(MBB, MI, NarrowEntry, LiveCPSR, CPSRDef, IsSelfLoop))
+      { ARM::t2CMPrr,ARM::tCMPr, 0, 0, 0, 1, 1,2, 0, 0,1,0 };
+    if (ReduceToNarrow(MBB, MI, NarrowEntry, LiveCPSR, IsSelfLoop))
       return true;
-    return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop);
+    return ReduceToNarrow(MBB, MI, Entry, LiveCPSR, IsSelfLoop);
   }
   }
   return false;
@@ -590,12 +629,17 @@ Thumb2SizeReduce::ReduceSpecial(MachineBasicBlock &MBB, MachineInstr *MI,
 bool
 Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI,
                                 const ReduceEntry &Entry,
-                                bool LiveCPSR, MachineInstr *CPSRDef,
-                                bool IsSelfLoop) {
+                                bool LiveCPSR, bool IsSelfLoop) {
 
   if (ReduceLimit2Addr != -1 && ((int)Num2Addrs >= ReduceLimit2Addr))
     return false;
 
+  if (!MinimizeSize && !OptimizeSize && Entry.AvoidMovs &&
+      STI->avoidMOVsShifterOperand())
+    // Don't issue movs with shifter operand for some CPUs unless we
+    // are optimizing / minimizing for size.
+    return false;
+
   unsigned Reg0 = MI->getOperand(0).getReg();
   unsigned Reg1 = MI->getOperand(1).getReg();
   // t2MUL is "special". The tied source operand is second, not first.
@@ -666,7 +710,7 @@ Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI,
   // Avoid adding a false dependency on partial flag update by some 16-bit
   // instructions which has the 's' bit set.
   if (Entry.PartFlag && NewMCID.hasOptionalDef() && HasCC &&
-      canAddPseudoFlagDep(CPSRDef, MI, IsSelfLoop))
+      canAddPseudoFlagDep(MI, IsSelfLoop))
     return false;
 
   // Add the 16-bit instruction.
@@ -703,11 +747,16 @@ Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI,
 bool
 Thumb2SizeReduce::ReduceToNarrow(MachineBasicBlock &MBB, MachineInstr *MI,
                                  const ReduceEntry &Entry,
-                                 bool LiveCPSR, MachineInstr *CPSRDef,
-                                 bool IsSelfLoop) {
+                                 bool LiveCPSR, bool IsSelfLoop) {
   if (ReduceLimit != -1 && ((int)NumNarrows >= ReduceLimit))
     return false;
 
+  if (!MinimizeSize && !OptimizeSize && Entry.AvoidMovs &&
+      STI->avoidMOVsShifterOperand())
+    // Don't issue movs with shifter operand for some CPUs unless we
+    // are optimizing / minimizing for size.
+    return false;
+
   unsigned Limit = ~0U;
   if (Entry.Imm1Limit)
     Limit = (1 << Entry.Imm1Limit) - 1;
@@ -757,7 +806,7 @@ Thumb2SizeReduce::ReduceToNarrow(MachineBasicBlock &MBB, MachineInstr *MI,
   // Avoid adding a false dependency on partial flag update by some 16-bit
   // instructions which has the 's' bit set.
   if (Entry.PartFlag && NewMCID.hasOptionalDef() && HasCC &&
-      canAddPseudoFlagDep(CPSRDef, MI, IsSelfLoop))
+      canAddPseudoFlagDep(MI, IsSelfLoop))
     return false;
 
   // Add the 16-bit instruction.
@@ -841,14 +890,57 @@ static bool UpdateCPSRUse(MachineInstr &MI, bool LiveCPSR) {
   return LiveCPSR;
 }
 
+bool Thumb2SizeReduce::ReduceMI(MachineBasicBlock &MBB, MachineInstr *MI,
+                                bool LiveCPSR, bool IsSelfLoop) {
+  unsigned Opcode = MI->getOpcode();
+  DenseMap<unsigned, unsigned>::iterator OPI = ReduceOpcodeMap.find(Opcode);
+  if (OPI == ReduceOpcodeMap.end())
+    return false;
+  const ReduceEntry &Entry = ReduceTable[OPI->second];
+
+  // Don't attempt normal reductions on "special" cases for now.
+  if (Entry.Special)
+    return ReduceSpecial(MBB, MI, Entry, LiveCPSR, IsSelfLoop);
+
+  // Try to transform to a 16-bit two-address instruction.
+  if (Entry.NarrowOpc2 &&
+      ReduceTo2Addr(MBB, MI, Entry, LiveCPSR, IsSelfLoop))
+    return true;
+
+  // Try to transform to a 16-bit non-two-address instruction.
+  if (Entry.NarrowOpc1 &&
+      ReduceToNarrow(MBB, MI, Entry, LiveCPSR, IsSelfLoop))
+    return true;
+
+  return false;
+}
+
 bool Thumb2SizeReduce::ReduceMBB(MachineBasicBlock &MBB) {
   bool Modified = false;
 
   // Yes, CPSR could be livein.
   bool LiveCPSR = MBB.isLiveIn(ARM::CPSR);
-  MachineInstr *CPSRDef = 0;
   MachineInstr *BundleMI = 0;
 
+  CPSRDef = 0;
+  HighLatencyCPSR = false;
+
+  // Check predecessors for the latest CPSRDef.
+  bool HasBackEdges = false;
+  for (MachineBasicBlock::pred_iterator
+       I = MBB.pred_begin(), E = MBB.pred_end(); I != E; ++I) {
+    const MBBInfo &PInfo = BlockInfo[(*I)->getNumber()];
+    if (!PInfo.Visited) {
+      // Since blocks are visited in RPO, this must be a back-edge.
+      HasBackEdges = true;
+      continue;
+    }
+    if (PInfo.HighLatencyCPSR) {
+      HighLatencyCPSR = true;
+      break;
+    }
+  }
+
   // If this BB loops back to itself, conservatively avoid narrowing the
   // first instruction that does partial flag update.
   bool IsSelfLoop = MBB.isSuccessor(&MBB);
@@ -862,43 +954,25 @@ bool Thumb2SizeReduce::ReduceMBB(MachineBasicBlock &MBB) {
       BundleMI = MI;
       continue;
     }
+    if (MI->isDebugValue())
+      continue;
 
     LiveCPSR = UpdateCPSRUse(*MI, LiveCPSR);
 
-    unsigned Opcode = MI->getOpcode();
-    DenseMap<unsigned, unsigned>::iterator OPI = ReduceOpcodeMap.find(Opcode);
-    if (OPI != ReduceOpcodeMap.end()) {
-      const ReduceEntry &Entry = ReduceTable[OPI->second];
-      // Ignore "special" cases for now.
-      if (Entry.Special) {
-        if (ReduceSpecial(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop)) {
-          Modified = true;
-          MachineBasicBlock::instr_iterator I = prior(NextMII);
-          MI = &*I;
-        }
-        goto ProcessNext;
-      }
-
-      // Try to transform to a 16-bit two-address instruction.
-      if (Entry.NarrowOpc2 &&
-          ReduceTo2Addr(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop)) {
-        Modified = true;
-        MachineBasicBlock::instr_iterator I = prior(NextMII);
-        MI = &*I;
-        goto ProcessNext;
-      }
-
-      // Try to transform to a 16-bit non-two-address instruction.
-      if (Entry.NarrowOpc1 &&
-          ReduceToNarrow(MBB, MI, Entry, LiveCPSR, CPSRDef, IsSelfLoop)) {
-        Modified = true;
-        MachineBasicBlock::instr_iterator I = prior(NextMII);
-        MI = &*I;
-      }
+    // Does NextMII belong to the same bundle as MI?
+    bool NextInSameBundle = NextMII != E && NextMII->isBundledWithPred();
+
+    if (ReduceMI(MBB, MI, LiveCPSR, IsSelfLoop)) {
+      Modified = true;
+      MachineBasicBlock::instr_iterator I = prior(NextMII);
+      MI = &*I;
+      // Removing and reinserting the first instruction in a bundle will break
+      // up the bundle. Fix the bundling if it was broken.
+      if (NextInSameBundle && !NextMII->isBundledWithPred())
+        NextMII->bundleWithPred();
     }
 
-  ProcessNext:
-    if (NextMII != E && MI->isInsideBundle() && !NextMII->isInsideBundle()) {
+    if (!NextInSameBundle && MI->isInsideBundle()) {
       // FIXME: Since post-ra scheduler operates on bundles, the CPSR kill
       // marker is only on the BUNDLE instruction. Process the BUNDLE
       // instruction as we finish with the bundled instruction to work around
@@ -915,14 +989,19 @@ bool Thumb2SizeReduce::ReduceMBB(MachineBasicBlock &MBB) {
     if (MI->isCall()) {
       // Calls don't really set CPSR.
       CPSRDef = 0;
+      HighLatencyCPSR = false;
       IsSelfLoop = false;
     } else if (DefCPSR) {
       // This is the last CPSR defining instruction.
       CPSRDef = MI;
+      HighLatencyCPSR = isHighLatencyCPSR(CPSRDef);
       IsSelfLoop = false;
     }
   }
 
+  MBBInfo &Info = BlockInfo[MBB.getNumber()];
+  Info.HighLatencyCPSR = HighLatencyCPSR;
+  Info.Visited = true;
   return Modified;
 }
 
@@ -931,9 +1010,23 @@ bool Thumb2SizeReduce::runOnMachineFunction(MachineFunction &MF) {
   TII = static_cast<const Thumb2InstrInfo*>(TM.getInstrInfo());
   STI = &TM.getSubtarget<ARMSubtarget>();
 
+  // Optimizing / minimizing size?
+  AttributeSet FnAttrs = MF.getFunction()->getAttributes();
+  OptimizeSize = FnAttrs.hasAttribute(AttributeSet::FunctionIndex,
+                                      Attribute::OptimizeForSize);
+  MinimizeSize = FnAttrs.hasAttribute(AttributeSet::FunctionIndex,
+                                      Attribute::MinSize);
+
+  BlockInfo.clear();
+  BlockInfo.resize(MF.getNumBlockIDs());
+
+  // Visit blocks in reverse post-order so LastCPSRDef is known for all
+  // predecessors.
+  ReversePostOrderTraversal<MachineFunction*> RPOT(&MF);
   bool Modified = false;
-  for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I)
-    Modified |= ReduceMBB(*I);
+  for (ReversePostOrderTraversal<MachineFunction*>::rpo_iterator
+       I = RPOT.begin(), E = RPOT.end(); I != E; ++I)
+    Modified |= ReduceMBB(**I);
   return Modified;
 }
 
diff --git a/lib/Target/CMakeLists.txt b/lib/Target/CMakeLists.txt
index a85acaaa1494..02ac493b4215 100644
--- a/lib/Target/CMakeLists.txt
+++ b/lib/Target/CMakeLists.txt
@@ -1,16 +1,13 @@
 add_llvm_library(LLVMTarget
   Mangler.cpp
   Target.cpp
-  TargetInstrInfo.cpp
   TargetIntrinsicInfo.cpp
   TargetJITInfo.cpp
   TargetLibraryInfo.cpp
   TargetLoweringObjectFile.cpp
   TargetMachine.cpp
   TargetMachineC.cpp
-  TargetRegisterInfo.cpp
   TargetSubtargetInfo.cpp
-  TargetTransformImpl.cpp
   )
 
 foreach(t ${LLVM_TARGETS_TO_BUILD})
diff --git a/lib/Target/CellSPU/CMakeLists.txt b/lib/Target/CellSPU/CMakeLists.txt
deleted file mode 100644
index 1f8ca8681c09..000000000000
--- a/lib/Target/CellSPU/CMakeLists.txt
+++ /dev/null
@@ -1,30 +0,0 @@
-set(LLVM_TARGET_DEFINITIONS SPU.td)
-
-tablegen(LLVM SPUGenAsmWriter.inc -gen-asm-writer)
-tablegen(LLVM SPUGenCodeEmitter.inc -gen-emitter)
-tablegen(LLVM SPUGenRegisterInfo.inc -gen-register-info)
-tablegen(LLVM SPUGenInstrInfo.inc -gen-instr-info)
-tablegen(LLVM SPUGenDAGISel.inc -gen-dag-isel)
-tablegen(LLVM SPUGenSubtargetInfo.inc -gen-subtarget)
-tablegen(LLVM SPUGenCallingConv.inc -gen-callingconv)
-add_public_tablegen_target(CellSPUCommonTableGen)
-
-add_llvm_target(CellSPUCodeGen
-  SPUAsmPrinter.cpp
-  SPUHazardRecognizers.cpp
-  SPUInstrInfo.cpp
-  SPUISelDAGToDAG.cpp
-  SPUISelLowering.cpp
-  SPUFrameLowering.cpp
-  SPUMachineFunction.cpp
-  SPURegisterInfo.cpp
-  SPUSubtarget.cpp
-  SPUTargetMachine.cpp
-  SPUSelectionDAGInfo.cpp
-  SPUNopFiller.cpp
-  )
-
-add_dependencies(LLVMCellSPUCodeGen intrinsics_gen)
-
-add_subdirectory(TargetInfo)
-add_subdirectory(MCTargetDesc)
diff --git a/lib/Target/CellSPU/CellSDKIntrinsics.td b/lib/Target/CellSPU/CellSDKIntrinsics.td
deleted file mode 100644
index cdb4099ffbca..000000000000
--- a/lib/Target/CellSPU/CellSDKIntrinsics.td
+++ /dev/null
@@ -1,449 +0,0 @@
-//===-- CellSDKIntrinsics.td - Cell SDK Intrinsics ---------*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-///--==-- Arithmetic ops intrinsics --==--
-def CellSDKah:
-    RR_Int_v8i16<0b00010011000, "ah", IntegerOp, int_spu_si_ah>;
-def CellSDKahi:
-    RI10_Int_v8i16<0b00010011000, "ahi", IntegerOp, int_spu_si_ahi>;
-def CellSDKa:
-    RR_Int_v4i32<0b00000011000, "a", IntegerOp, int_spu_si_a>;
-def CellSDKai:
-    RI10_Int_v4i32<0b00111000, "ai", IntegerOp, int_spu_si_ai>;
-def CellSDKsfh:
-    RR_Int_v8i16<0b00010010000, "sfh", IntegerOp, int_spu_si_sfh>;
-def CellSDKsfhi:
-    RI10_Int_v8i16<0b10110000, "sfhi", IntegerOp, int_spu_si_sfhi>;
-def CellSDKsf:
-    RR_Int_v4i32<0b00000010000, "sf", IntegerOp, int_spu_si_sf>;
-def CellSDKsfi:
-    RI10_Int_v4i32<0b00110000, "sfi", IntegerOp, int_spu_si_sfi>;
-def CellSDKaddx:
-    RR_Int_v4i32<0b00000010110, "addx", IntegerOp, int_spu_si_addx>;
-def CellSDKcg:
-    RR_Int_v4i32<0b0100001100, "cg", IntegerOp, int_spu_si_cg>;
-def CellSDKcgx:
-    RR_Int_v4i32<0b01000010110, "cgx", IntegerOp, int_spu_si_cgx>;
-def CellSDKsfx:
-    RR_Int_v4i32<0b10000010110, "sfx", IntegerOp, int_spu_si_sfx>;
-def CellSDKbg:
-    RR_Int_v4i32<0b01000010000, "bg", IntegerOp, int_spu_si_bg>;
-def CellSDKbgx:
-    RR_Int_v4i32<0b11000010110, "bgx", IntegerOp, int_spu_si_bgx>;
-
-def CellSDKmpy:
-    RRForm<0b00100011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "mpy $rT, $rA, $rB", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpy (v8i16 VECREG:$rA),
-                                                (v8i16 VECREG:$rB)))]>;
-
-def CellSDKmpyu:
-    RRForm<0b00110011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "mpyu $rT, $rA, $rB", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpyu (v8i16 VECREG:$rA),
-                                                 (v8i16 VECREG:$rB)))] >;
-
-def CellSDKmpyi:
-    RI10Form<0b00101110, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-      "mpyi $rT, $rA, $val", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpyi (v8i16 VECREG:$rA),
-                                                 i16ImmSExt10:$val))]>;
-
-def CellSDKmpyui:
-    RI10Form<0b10101110, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-      "mpyui $rT, $rA, $val", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpyui (v8i16 VECREG:$rA),
-                                                  i16ImmSExt10:$val))]>;
-
-def CellSDKmpya:
-    RRRForm<0b0011, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-      "mpya $rT, $rA, $rB, $rC", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpya (v8i16 VECREG:$rA),
-                                                 (v8i16 VECREG:$rB),
-                                                 (v8i16 VECREG:$rC)))]>;
-
-def CellSDKmpyh:
-    RRForm<0b10100011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "mpyh $rT, $rA, $rB", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpyh (v4i32 VECREG:$rA),
-                                                 (v8i16 VECREG:$rB)))]>;
-
-def CellSDKmpys:
-    RRForm<0b11100011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "mpys $rT, $rA, $rB", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpys (v8i16 VECREG:$rA),
-                                                 (v8i16 VECREG:$rB)))]>;
-
-def CellSDKmpyhh:
-    RRForm<0b01100011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "mpyhh $rT, $rA, $rB", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpyhh (v8i16 VECREG:$rA),
-                                                  (v8i16 VECREG:$rB)))]>;
-
-def CellSDKmpyhha:
-    RRForm<0b01100010110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "mpyhha $rT, $rA, $rB", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpyhha (v8i16 VECREG:$rA),
-                                                   (v8i16 VECREG:$rB)))]>;
-
-// Not sure how to match a (set $rT, (add $rT (mpyhh $rA, $rB)))... so leave
-// as an intrinsic for the time being
-def CellSDKmpyhhu:
-    RRForm<0b01110011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "mpyhhu $rT, $rA, $rB", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpyhhu (v8i16 VECREG:$rA),
-                                                   (v8i16 VECREG:$rB)))]>;
-
-def CellSDKmpyhhau:
-    RRForm<0b01110010110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "mpyhhau $rT, $rA, $rB", IntegerMulDiv,
-      [(set (v4i32 VECREG:$rT), (int_spu_si_mpyhhau (v8i16 VECREG:$rA),
-                                                    (v8i16 VECREG:$rB)))]>;
-
-def CellSDKand:
-        RRForm<0b1000011000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-          "and\t $rT, $rA, $rB", IntegerOp,
-          [(set (v4i32 VECREG:$rT),
-                (int_spu_si_and (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-def CellSDKandc:
-        RRForm<0b10000011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-          "andc\t $rT, $rA, $rB", IntegerOp,
-          [(set (v4i32 VECREG:$rT),
-                (int_spu_si_andc (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-def CellSDKandbi:
-     RI10Form<0b01101000, (outs VECREG:$rT), (ins VECREG:$rA, u10imm_i8:$val),
-       "andbi\t $rT, $rA, $val", BranchResolv,
-       [(set (v16i8 VECREG:$rT),
-             (int_spu_si_andbi (v16i8 VECREG:$rA), immU8:$val))]>;
-
-def CellSDKandhi:
-     RI10Form<0b10101000, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-           "andhi\t $rT, $rA, $val", BranchResolv,
-       [(set (v8i16 VECREG:$rT),
-             (int_spu_si_andhi (v8i16 VECREG:$rA), i16ImmSExt10:$val))]>;
-
-def CellSDKandi:
-     RI10Form<0b00101000, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-           "andi\t $rT, $rA, $val", BranchResolv,
-       [(set (v4i32 VECREG:$rT),
-             (int_spu_si_andi (v4i32 VECREG:$rA), i32ImmSExt10:$val))]>;
-
-def CellSDKor:
-        RRForm<0b10000010000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-          "or\t $rT, $rA, $rB", IntegerOp,
-          [(set (v4i32 VECREG:$rT),
-                (int_spu_si_or (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-def CellSDKorc:
-        RRForm<0b10010011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-          "addc\t $rT, $rA, $rB", IntegerOp,
-          [(set (v4i32 VECREG:$rT),
-                (int_spu_si_orc (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-def CellSDKorbi:
-     RI10Form<0b01100000, (outs VECREG:$rT), (ins VECREG:$rA, u10imm_i8:$val),
-       "orbi\t $rT, $rA, $val", BranchResolv,
-       [(set (v16i8 VECREG:$rT),
-             (int_spu_si_orbi (v16i8 VECREG:$rA), immU8:$val))]>;
-
-def CellSDKorhi:
-     RI10Form<0b10100000, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-           "orhi\t $rT, $rA, $val", BranchResolv,
-       [(set (v8i16 VECREG:$rT),
-             (int_spu_si_orhi (v8i16 VECREG:$rA), i16ImmSExt10:$val))]>;
-
-def CellSDKori:
-     RI10Form<0b00100000, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-           "ori\t $rT, $rA, $val", BranchResolv,
-       [(set (v4i32 VECREG:$rT),
-             (int_spu_si_ori (v4i32 VECREG:$rA), i32ImmSExt10:$val))]>;
-
-def CellSDKxor:
-        RRForm<0b10000010000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-          "xor\t $rT, $rA, $rB", IntegerOp,
-          [(set (v4i32 VECREG:$rT), 
-                (int_spu_si_xor (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-def CellSDKxorbi:
-     RI10Form<0b01100000, (outs VECREG:$rT), (ins VECREG:$rA, u10imm_i8:$val),
-       "xorbi\t $rT, $rA, $val", BranchResolv,
-       [(set (v16i8 VECREG:$rT), (int_spu_si_xorbi (v16i8 VECREG:$rA), immU8:$val))]>;
-
-def CellSDKxorhi:
-     RI10Form<0b10100000, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-       "xorhi\t $rT, $rA, $val", BranchResolv,
-       [(set (v8i16 VECREG:$rT), 
-             (int_spu_si_xorhi (v8i16 VECREG:$rA), i16ImmSExt10:$val))]>;
-
-def CellSDKxori:
-     RI10Form<0b00100000, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-           "xori\t $rT, $rA, $val", BranchResolv,
-       [(set (v4i32 VECREG:$rT), 
-             (int_spu_si_xori (v4i32 VECREG:$rA), i32ImmSExt10:$val))]>;
-
-def CellSDKnor:
-        RRForm<0b10000010000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-          "nor\t $rT, $rA, $rB", IntegerOp,
-          [(set (v4i32 VECREG:$rT), 
-                (int_spu_si_nor (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-def CellSDKnand:
-        RRForm<0b10000010000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-          "nand\t $rT, $rA, $rB", IntegerOp,
-          [(set (v4i32 VECREG:$rT), 
-                (int_spu_si_nand (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-//===----------------------------------------------------------------------===//
-// Shift/rotate intrinsics:
-//===----------------------------------------------------------------------===//
-
-def CellSDKshli:
-  Pat<(int_spu_si_shli (v4i32 VECREG:$rA), uimm7:$val),
-      (SHLIv4i32 VECREG:$rA, (TO_IMM32 imm:$val))>;
-
-def CellSDKshlqbi:
-  Pat<(int_spu_si_shlqbi VECREG:$rA, R32C:$rB),
-      (SHLQBIv16i8 VECREG:$rA, R32C:$rB)>;
-
-def CellSDKshlqii:
-  Pat<(int_spu_si_shlqbii VECREG:$rA, uimm7:$val),
-      (SHLQBIIv16i8 VECREG:$rA, (TO_IMM32 imm:$val))>;
-
-def CellSDKshlqby:
-  Pat<(int_spu_si_shlqby VECREG:$rA, R32C:$rB),
-      (SHLQBYv16i8 VECREG:$rA, R32C:$rB)>;
-
-def CellSDKshlqbyi:
-  Pat<(int_spu_si_shlqbyi VECREG:$rA, uimm7:$val),
-      (SHLQBYIv16i8 VECREG:$rA, (TO_IMM32 imm:$val))>;
-
-          
-//===----------------------------------------------------------------------===//
-// Branch/compare intrinsics:
-//===----------------------------------------------------------------------===//
-
-def CellSDKceq:
-  RRForm<0b00000011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-        "ceq\t $rT, $rA, $rB", BranchResolv,
-        [(set (v4i32 VECREG:$rT), 
-              (int_spu_si_ceq (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-def CellSDKceqi:
-  RI10Form<0b00111110, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-        "ceqi\t $rT, $rA, $val", BranchResolv,
-    [(set (v4i32 VECREG:$rT), 
-          (int_spu_si_ceqi (v4i32 VECREG:$rA), i32ImmSExt10:$val))]>;
-
-def CellSDKceqb:
-  RRForm<0b00001011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-        "ceqb\t $rT, $rA, $rB", BranchResolv,
-        [(set (v16i8 VECREG:$rT), 
-              (int_spu_si_ceqb (v16i8 VECREG:$rA), (v16i8 VECREG:$rB)))]>;
-
-def CellSDKceqbi:
-  RI10Form<0b01111110, (outs VECREG:$rT), (ins VECREG:$rA, u10imm_i8:$val),
-        "ceqbi\t $rT, $rA, $val", BranchResolv,
-    [(set (v16i8 VECREG:$rT), (int_spu_si_ceqbi (v16i8 VECREG:$rA), immU8:$val))]>;
-
-def CellSDKceqh:
-  RRForm<0b00010011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-        "ceqh\t $rT, $rA, $rB", BranchResolv,
-        [(set (v8i16 VECREG:$rT), 
-              (int_spu_si_ceqh (v8i16 VECREG:$rA), (v8i16 VECREG:$rB)))]>;
-
-def CellSDKceqhi:
-  RI10Form<0b10111110, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-        "ceqhi\t $rT, $rA, $val", BranchResolv,
-    [(set (v8i16 VECREG:$rT), 
-          (int_spu_si_ceqhi (v8i16 VECREG:$rA), i16ImmSExt10:$val))]>;
-def CellSDKcgth:
-  RRForm<0b00010011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "cgth\t $rT, $rA, $rB", BranchResolv,
-        [(set (v8i16 VECREG:$rT),
-              (int_spu_si_cgth (v8i16 VECREG:$rA), (v8i16 VECREG:$rB)))]>;
-
-def CellSDKcgthi:
-  RI10Form<0b10111110, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-    "cgthi\t $rT, $rA, $val", BranchResolv,
-        [(set (v8i16 VECREG:$rT), 
-              (int_spu_si_cgthi (v8i16 VECREG:$rA), i16ImmSExt10:$val))]>;
-
-def CellSDKcgt:
-  RRForm<0b00000010010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "cgt\t $rT, $rA, $rB", BranchResolv,
-        [(set (v4i32 VECREG:$rT), 
-              (int_spu_si_cgt (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-def CellSDKcgti:
-  RI10Form<0b00110010, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-    "cgti\t $rT, $rA, $val", BranchResolv,
-        [(set (v4i32 VECREG:$rT), 
-              (int_spu_si_cgti (v4i32 VECREG:$rA), i32ImmSExt10:$val))]>;
-
-def CellSDKcgtb:
-  RRForm<0b00001010010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "cgtb\t $rT, $rA, $rB", BranchResolv,
-        [(set (v16i8 VECREG:$rT), 
-              (int_spu_si_cgtb (v16i8 VECREG:$rA), (v16i8 VECREG:$rB)))]>;
-
-def CellSDKcgtbi:
-  RI10Form<0b01110010, (outs VECREG:$rT), (ins VECREG:$rA, u10imm_i8:$val),
-    "cgtbi\t $rT, $rA, $val", BranchResolv,
-        [(set (v16i8 VECREG:$rT), (int_spu_si_cgtbi (v16i8 VECREG:$rA), immU8:$val))]>;
-
-def CellSDKclgth:
-  RRForm<0b00010011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "clgth\t $rT, $rA, $rB", BranchResolv,
-        [(set (v8i16 VECREG:$rT), 
-              (int_spu_si_clgth (v8i16 VECREG:$rA), (v8i16 VECREG:$rB)))]>;
-
-def CellSDKclgthi:
-  RI10Form<0b10111010, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-    "clgthi\t $rT, $rA, $val", BranchResolv,
-        [(set (v8i16 VECREG:$rT), 
-              (int_spu_si_clgthi (v8i16 VECREG:$rA), i16ImmSExt10:$val))]>;
-
-def CellSDKclgt:
-  RRForm<0b00000011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "clgt\t $rT, $rA, $rB", BranchResolv,
-        [(set (v4i32 VECREG:$rT), 
-              (int_spu_si_clgt (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-def CellSDKclgti:
-  RI10Form<0b00111010, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-    "clgti\t $rT, $rA, $val", BranchResolv,
-        [(set (v4i32 VECREG:$rT), 
-              (int_spu_si_clgti (v4i32 VECREG:$rA), i32ImmSExt10:$val))]>;
-
-def CellSDKclgtb:
-  RRForm<0b00001011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "clgtb\t $rT, $rA, $rB", BranchResolv,
-    [(set (v16i8 VECREG:$rT),
-          (int_spu_si_clgtb (v16i8 VECREG:$rA), (v16i8 VECREG:$rB)))]>;
-
-def CellSDKclgtbi:
-  RI10Form<0b01111010, (outs VECREG:$rT), (ins VECREG:$rA, u10imm_i8:$val),
-    "clgtbi\t $rT, $rA, $val", BranchResolv,
-    [(set (v16i8 VECREG:$rT),
-          (int_spu_si_clgtbi (v16i8 VECREG:$rA), immU8:$val))]>;
-
-//===----------------------------------------------------------------------===//
-// Floating-point intrinsics:
-//===----------------------------------------------------------------------===//
-
-def CellSDKfa:
-  RRForm<0b00100011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "fa\t $rT, $rA, $rB", SPrecFP,
-        [(set (v4f32 VECREG:$rT), (int_spu_si_fa (v4f32 VECREG:$rA),
-                                                 (v4f32 VECREG:$rB)))]>;
-
-def CellSDKfs:
-  RRForm<0b10100011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "fs\t $rT, $rA, $rB", SPrecFP,
-        [(set (v4f32 VECREG:$rT), (int_spu_si_fs (v4f32 VECREG:$rA),
-                                                 (v4f32 VECREG:$rB)))]>;
-
-def CellSDKfm:
-  RRForm<0b01100011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "fm\t $rT, $rA, $rB", SPrecFP,
-        [(set (v4f32 VECREG:$rT), (int_spu_si_fm (v4f32 VECREG:$rA),
-                                                 (v4f32 VECREG:$rB)))]>;
-
-def CellSDKfceq:
-  RRForm<0b01000011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "fceq\t $rT, $rA, $rB", SPrecFP,
-        [(set (v4f32 VECREG:$rT), (int_spu_si_fceq (v4f32 VECREG:$rA),
-                                                   (v4f32 VECREG:$rB)))]>;
-
-def CellSDKfcgt:
-  RRForm<0b01000011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "fcgt\t $rT, $rA, $rB", SPrecFP,
-        [(set (v4f32 VECREG:$rT), (int_spu_si_fcgt (v4f32 VECREG:$rA),
-                                                   (v4f32 VECREG:$rB)))]>;
-
-def CellSDKfcmeq:
-  RRForm<0b01010011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "fcmeq\t $rT, $rA, $rB", SPrecFP,
-        [(set (v4f32 VECREG:$rT), (int_spu_si_fcmeq (v4f32 VECREG:$rA),
-                                                    (v4f32 VECREG:$rB)))]>;
-
-def CellSDKfcmgt:
-  RRForm<0b01010011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "fcmgt\t $rT, $rA, $rB", SPrecFP,
-        [(set (v4f32 VECREG:$rT), (int_spu_si_fcmgt (v4f32 VECREG:$rA),
-                                                    (v4f32 VECREG:$rB)))]>;
-
-def CellSDKfma:
-  RRRForm<0b0111, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-    "fma\t $rT, $rA, $rB, $rC", SPrecFP,
-        [(set (v4f32 VECREG:$rT), (int_spu_si_fma (v4f32 VECREG:$rA),
-                                                  (v4f32 VECREG:$rB),
-                                                  (v4f32 VECREG:$rC)))]>;
-
-def CellSDKfnms:
-  RRRForm<0b1011, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-    "fnms\t $rT, $rA, $rB, $rC", SPrecFP,
-        [(set (v4f32 VECREG:$rT), (int_spu_si_fnms (v4f32 VECREG:$rA),
-                                                   (v4f32 VECREG:$rB),
-                                                   (v4f32 VECREG:$rC)))]>;
-
-def CellSDKfms:
-  RRRForm<0b1111, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-    "fms\t $rT, $rA, $rB, $rC", SPrecFP,
-        [(set (v4f32 VECREG:$rT), (int_spu_si_fms (v4f32 VECREG:$rA),
-                                                  (v4f32 VECREG:$rB),
-                                                  (v4f32 VECREG:$rC)))]>;
-
-//===----------------------------------------------------------------------===//
-// Double precision floating-point intrinsics:
-//===----------------------------------------------------------------------===//
-
-def CellSDKdfa:
-  RRForm<0b00110011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "dfa\t $rT, $rA, $rB", DPrecFP,
-        [(set (v2f64 VECREG:$rT), (int_spu_si_dfa (v2f64 VECREG:$rA),
-                                                  (v2f64 VECREG:$rB)))]>;
-
-def CellSDKdfs:
-  RRForm<0b10110011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "dfs\t $rT, $rA, $rB", DPrecFP,
-        [(set (v2f64 VECREG:$rT), (int_spu_si_dfs (v2f64 VECREG:$rA),
-                                                  (v2f64 VECREG:$rB)))]>;
-
-def CellSDKdfm:
-  RRForm<0b01110011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "dfm\t $rT, $rA, $rB", DPrecFP,
-        [(set (v2f64 VECREG:$rT), (int_spu_si_dfm (v2f64 VECREG:$rA),
-                                                  (v2f64 VECREG:$rB)))]>;
-
-def CellSDKdfma:
-  RRForm<0b00111010110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "dfma\t $rT, $rA, $rB", DPrecFP,
-        [(set (v2f64 VECREG:$rT), (int_spu_si_dfma (v2f64 VECREG:$rA),
-                                                   (v2f64 VECREG:$rB)))]>;
-
-def CellSDKdfnma:
-  RRForm<0b11111010110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "dfnma\t $rT, $rA, $rB", DPrecFP,
-        [(set (v2f64 VECREG:$rT), (int_spu_si_dfnma (v2f64 VECREG:$rA),
-                                                    (v2f64 VECREG:$rB)))]>;
-
-def CellSDKdfnms:
-  RRForm<0b01111010110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "dfnms\t $rT, $rA, $rB", DPrecFP,
-        [(set (v2f64 VECREG:$rT), (int_spu_si_dfnms (v2f64 VECREG:$rA),
-                                                    (v2f64 VECREG:$rB)))]>;
-
-def CellSDKdfms:
-  RRForm<0b10111010110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "dfms\t $rT, $rA, $rB", DPrecFP,
-        [(set (v2f64 VECREG:$rT), (int_spu_si_dfms (v2f64 VECREG:$rA),
-                                                   (v2f64 VECREG:$rB)))]>;
diff --git a/lib/Target/CellSPU/MCTargetDesc/CMakeLists.txt b/lib/Target/CellSPU/MCTargetDesc/CMakeLists.txt
deleted file mode 100644
index 0027bdbf6ca1..000000000000
--- a/lib/Target/CellSPU/MCTargetDesc/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-add_llvm_library(LLVMCellSPUDesc
-  SPUMCTargetDesc.cpp
-  SPUMCAsmInfo.cpp
-  )
-
-add_dependencies(LLVMCellSPUDesc CellSPUCommonTableGen)
diff --git a/lib/Target/CellSPU/MCTargetDesc/SPUMCAsmInfo.cpp b/lib/Target/CellSPU/MCTargetDesc/SPUMCAsmInfo.cpp
deleted file mode 100644
index 4bad37eacaf7..000000000000
--- a/lib/Target/CellSPU/MCTargetDesc/SPUMCAsmInfo.cpp
+++ /dev/null
@@ -1,43 +0,0 @@
-//===-- SPUMCAsmInfo.cpp - Cell SPU asm properties ------------------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file contains the declarations of the SPUMCAsmInfo properties.
-//
-//===----------------------------------------------------------------------===//
-
-#include "SPUMCAsmInfo.h"
-using namespace llvm;
-
-void SPULinuxMCAsmInfo::anchor() { }
-
-SPULinuxMCAsmInfo::SPULinuxMCAsmInfo(const Target &T, StringRef TT) {
-  IsLittleEndian = false;
-
-  ZeroDirective = "\t.space\t";
-  Data64bitsDirective = "\t.quad\t";
-  AlignmentIsInBytes = false;
-      
-  PCSymbol = ".";
-  CommentString = "#";
-  GlobalPrefix = "";
-  PrivateGlobalPrefix = ".L";
-
-  // Has leb128
-  HasLEB128 = true;
-
-  SupportsDebugInformation = true;
-
-  // Exception handling is not supported on CellSPU (think about it: you only
-  // have 256K for code+data. Would you support exception handling?)
-  ExceptionsType = ExceptionHandling::None;
-
-  // SPU assembly requires ".section" before ".bss" 
-  UsesELFSectionDirectiveForBSS = true;  
-}
-
diff --git a/lib/Target/CellSPU/MCTargetDesc/SPUMCTargetDesc.cpp b/lib/Target/CellSPU/MCTargetDesc/SPUMCTargetDesc.cpp
deleted file mode 100644
index 8450e2c6634c..000000000000
--- a/lib/Target/CellSPU/MCTargetDesc/SPUMCTargetDesc.cpp
+++ /dev/null
@@ -1,94 +0,0 @@
-//===-- SPUMCTargetDesc.cpp - Cell SPU Target Descriptions ----------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file provides Cell SPU specific target descriptions.
-//
-//===----------------------------------------------------------------------===//
-
-#include "SPUMCTargetDesc.h"
-#include "SPUMCAsmInfo.h"
-#include "llvm/MC/MachineLocation.h"
-#include "llvm/MC/MCCodeGenInfo.h"
-#include "llvm/MC/MCInstrInfo.h"
-#include "llvm/MC/MCRegisterInfo.h"
-#include "llvm/MC/MCSubtargetInfo.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/TargetRegistry.h"
-
-#define GET_INSTRINFO_MC_DESC
-#include "SPUGenInstrInfo.inc"
-
-#define GET_SUBTARGETINFO_MC_DESC
-#include "SPUGenSubtargetInfo.inc"
-
-#define GET_REGINFO_MC_DESC
-#include "SPUGenRegisterInfo.inc"
-
-using namespace llvm;
-
-static MCInstrInfo *createSPUMCInstrInfo() {
-  MCInstrInfo *X = new MCInstrInfo();
-  InitSPUMCInstrInfo(X);
-  return X;
-}
-
-static MCRegisterInfo *createCellSPUMCRegisterInfo(StringRef TT) {
-  MCRegisterInfo *X = new MCRegisterInfo();
-  InitSPUMCRegisterInfo(X, SPU::R0);
-  return X;
-}
-
-static MCSubtargetInfo *createSPUMCSubtargetInfo(StringRef TT, StringRef CPU,
-                                                 StringRef FS) {
-  MCSubtargetInfo *X = new MCSubtargetInfo();
-  InitSPUMCSubtargetInfo(X, TT, CPU, FS);
-  return X;
-}
-
-static MCAsmInfo *createSPUMCAsmInfo(const Target &T, StringRef TT) {
-  MCAsmInfo *MAI = new SPULinuxMCAsmInfo(T, TT);
-
-  // Initial state of the frame pointer is R1.
-  MachineLocation Dst(MachineLocation::VirtualFP);
-  MachineLocation Src(SPU::R1, 0);
-  MAI->addInitialFrameState(0, Dst, Src);
-
-  return MAI;
-}
-
-static MCCodeGenInfo *createSPUMCCodeGenInfo(StringRef TT, Reloc::Model RM,
-                                             CodeModel::Model CM,
-                                             CodeGenOpt::Level OL) {
-  MCCodeGenInfo *X = new MCCodeGenInfo();
-  // For the time being, use static relocations, since there's really no
-  // support for PIC yet.
-  X->InitMCCodeGenInfo(Reloc::Static, CM, OL);
-  return X;
-}
-
-// Force static initialization.
-extern "C" void LLVMInitializeCellSPUTargetMC() {
-  // Register the MC asm info.
-  RegisterMCAsmInfoFn X(TheCellSPUTarget, createSPUMCAsmInfo);
-
-  // Register the MC codegen info.
-  TargetRegistry::RegisterMCCodeGenInfo(TheCellSPUTarget,
-                                        createSPUMCCodeGenInfo);
-
-  // Register the MC instruction info.
-  TargetRegistry::RegisterMCInstrInfo(TheCellSPUTarget, createSPUMCInstrInfo);
-
-  // Register the MC register info.
-  TargetRegistry::RegisterMCRegInfo(TheCellSPUTarget,
-                                    createCellSPUMCRegisterInfo);
-
-  // Register the MC subtarget info.
-  TargetRegistry::RegisterMCSubtargetInfo(TheCellSPUTarget,
-                                          createSPUMCSubtargetInfo);
-}
diff --git a/lib/Target/CellSPU/MCTargetDesc/SPUMCTargetDesc.h b/lib/Target/CellSPU/MCTargetDesc/SPUMCTargetDesc.h
deleted file mode 100644
index d26449e8908f..000000000000
--- a/lib/Target/CellSPU/MCTargetDesc/SPUMCTargetDesc.h
+++ /dev/null
@@ -1,38 +0,0 @@
-//===-- SPUMCTargetDesc.h - CellSPU Target Descriptions ---------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file provides CellSPU specific target descriptions.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef SPUMCTARGETDESC_H
-#define SPUMCTARGETDESC_H
-
-namespace llvm {
-class Target;
-
-extern Target TheCellSPUTarget;
-
-} // End llvm namespace
-
-// Define symbolic names for Cell registers.  This defines a mapping from
-// register name to register number.
-//
-#define GET_REGINFO_ENUM
-#include "SPUGenRegisterInfo.inc"
-
-// Defines symbolic names for the SPU instructions.
-//
-#define GET_INSTRINFO_ENUM
-#include "SPUGenInstrInfo.inc"
-
-#define GET_SUBTARGETINFO_ENUM
-#include "SPUGenSubtargetInfo.inc"
-
-#endif
diff --git a/lib/Target/CellSPU/Makefile b/lib/Target/CellSPU/Makefile
deleted file mode 100644
index d7a8247f5702..000000000000
--- a/lib/Target/CellSPU/Makefile
+++ /dev/null
@@ -1,20 +0,0 @@
-##===- lib/Target/CellSPU/Makefile -------------------------*- Makefile -*-===##
-#
-#                     The LLVM Compiler Infrastructure
-#
-# This file is distributed under the University of Illinois Open Source
-# License. See LICENSE.TXT for details.
-#
-##===----------------------------------------------------------------------===##
-
-LEVEL = ../../..
-LIBRARYNAME = LLVMCellSPUCodeGen
-TARGET = SPU
-BUILT_SOURCES = SPUGenInstrInfo.inc SPUGenRegisterInfo.inc \
-		SPUGenAsmWriter.inc SPUGenCodeEmitter.inc \
-		SPUGenDAGISel.inc \
-		SPUGenSubtargetInfo.inc SPUGenCallingConv.inc
-
-DIRS = TargetInfo MCTargetDesc
-
-include $(LEVEL)/Makefile.common
diff --git a/lib/Target/CellSPU/README.txt b/lib/Target/CellSPU/README.txt
deleted file mode 100644
index 3bce9609bfef..000000000000
--- a/lib/Target/CellSPU/README.txt
+++ /dev/null
@@ -1,106 +0,0 @@
-//===- README.txt - Notes for improving CellSPU-specific code gen ---------===//
-
-This code was contributed by a team from the Computer Systems Research
-Department in The Aerospace Corporation:
-
-- Scott Michel (head bottle washer and much of the non-floating point
-  instructions)
-- Mark Thomas (floating point instructions)
-- Michael AuYeung (intrinsics)
-- Chandler Carruth (LLVM expertise)
-- Nehal Desai (debugging, i32 operations, RoadRunner SPU expertise)
-
-Some minor fixes added by Kalle Raiskila.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, NONINFRINGEMENT, OR
-OTHERWISE.  IN NO EVENT SHALL THE AEROSPACE CORPORATION BE LIABLE FOR DAMAGES
-OF ANY KIND OR NATURE WHETHER BASED IN CONTRACT, TORT, OR OTHERWISE ARISING
-OUT OF OR IN CONNECTION WITH THE USE OF THE SOFTWARE INCLUDING, WITHOUT
-LIMITATION, DAMAGES RESULTING FROM LOST OR CONTAMINATED DATA, LOST PROFITS OR
-REVENUE, COMPUTER MALFUNCTION, OR FOR ANY SPECIAL, INCIDENTAL, CONSEQUENTIAL,
-OR PUNITIVE  DAMAGES, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGES OR
-SUCH DAMAGES ARE FORESEEABLE.
-
----------------------------------------------------------------------------
---WARNING--:
---WARNING--: The CellSPU work is work-in-progress and "alpha" quality code.
---WARNING--:
-
-If you are brave enough to try this code or help to hack on it, be sure
-to add 'spu' to configure's --enable-targets option, e.g.:
-
-        ./configure <your_configure_flags_here> \
-           --enable-targets=x86,x86_64,powerpc,spu
-
----------------------------------------------------------------------------
-
-TODO:
-* In commit r142152 vector legalization was set to element promotion per
-  default. This breaks half vectors (e.g. v2i32) badly as they get element
-  promoted to much slower types (v2i64).
-
-* Many CellSPU specific codegen tests only grep & count the number of 
-  instructions, not checking their place with FileCheck. There have also
-  been some commits that change the CellSPU checks, some of which might
-  have not been thoroughly scrutinized w.r.t. to the changes they cause in SPU
-  assembly. (especially since about the time of r142152)  
-
-* Some of the i64 math have huge tablegen rules, which sometime cause
-  tablegen to run out of memory. See e.g. bug 8850. i64 arithmetics 
-  should probably be done with libraries.
-
-* Create a machine pass for performing dual-pipeline scheduling specifically
-  for CellSPU, and insert branch prediction instructions as needed.
-
-* i32 instructions:
-
-  * i32 division (work-in-progress)
-
-* i64 support (see i64operations.c test harness):
-
-  * shifts and comparison operators: done
-  * sign and zero extension: done
-  * addition: done
-  * subtraction: needed
-  * multiplication: done
-
-* i128 support:
-
-  * zero extension, any extension: done
-  * sign extension: done
-  * arithmetic operators (add, sub, mul, div): needed
-  * logical operations (and, or, shl, srl, sra, xor, nor, nand): needed
-
-    * or: done
-
-* f64 support
-
-  * Comparison operators:
-    SETOEQ              unimplemented
-    SETOGT              unimplemented
-    SETOGE              unimplemented
-    SETOLT              unimplemented
-    SETOLE              unimplemented
-    SETONE              unimplemented
-    SETO                done (lowered)
-    SETUO               done (lowered)
-    SETUEQ              unimplemented
-    SETUGT              unimplemented
-    SETUGE              unimplemented
-    SETULT              unimplemented
-    SETULE              unimplemented
-    SETUNE              unimplemented
-
-* LLVM vector suport
-
-  * VSETCC needs to be implemented. It's pretty straightforward to code, but
-    needs implementation.
-
-* Intrinsics
-
-  * spu.h instrinsics added but not tested. Need to have an operational
-    llvm-spu-gcc in order to write a unit test harness.
-
-===-------------------------------------------------------------------------===
diff --git a/lib/Target/CellSPU/SPU.h b/lib/Target/CellSPU/SPU.h
deleted file mode 100644
index c660131706cb..000000000000
--- a/lib/Target/CellSPU/SPU.h
+++ /dev/null
@@ -1,31 +0,0 @@
-//===-- SPU.h - Top-level interface for Cell SPU Target ---------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file contains the entry points for global functions defined in the LLVM
-// Cell SPU back-end.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef LLVM_TARGET_IBMCELLSPU_H
-#define LLVM_TARGET_IBMCELLSPU_H
-
-#include "MCTargetDesc/SPUMCTargetDesc.h"
-#include "llvm/Target/TargetMachine.h"
-
-namespace llvm {
-  class SPUTargetMachine;
-  class FunctionPass;
-  class formatted_raw_ostream;
-
-  FunctionPass *createSPUISelDag(SPUTargetMachine &TM);
-  FunctionPass *createSPUNopFillerPass(SPUTargetMachine &tm);
-
-}
-
-#endif /* LLVM_TARGET_IBMCELLSPU_H */
diff --git a/lib/Target/CellSPU/SPU.td b/lib/Target/CellSPU/SPU.td
deleted file mode 100644
index e835b9cac8e1..000000000000
--- a/lib/Target/CellSPU/SPU.td
+++ /dev/null
@@ -1,66 +0,0 @@
-//===-- SPU.td - Describe the STI Cell SPU Target Machine --*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This is the top level entry point for the STI Cell SPU target machine.
-//
-//===----------------------------------------------------------------------===//
-
-// Get the target-independent interfaces which we are implementing.
-//
-include "llvm/Target/Target.td"
-
-// Holder of code fragments (you'd think this'd already be in
-// a td file somewhere... :-)
-
-class CodeFrag<dag frag> {
-  dag Fragment = frag;
-}
-
-//===----------------------------------------------------------------------===//
-// Register File Description
-//===----------------------------------------------------------------------===//
-
-include "SPURegisterInfo.td"
-
-//===----------------------------------------------------------------------===//
-// Instruction formats, instructions
-//===----------------------------------------------------------------------===//
-
-include "SPUNodes.td"
-include "SPUOperands.td"
-include "SPUSchedule.td"
-include "SPUInstrFormats.td"
-include "SPUInstrInfo.td"
-
-//===----------------------------------------------------------------------===//
-// Subtarget features:
-//===----------------------------------------------------------------------===//
-
-def DefaultProc: SubtargetFeature<"", "ProcDirective", "SPU::DEFAULT_PROC", "">;
-def LargeMemFeature:
-  SubtargetFeature<"large_mem","UseLargeMem", "true",
-                   "Use large (>256) LSA memory addressing [default = false]">;
-
-def SPURev0 : Processor<"v0", SPUItineraries, [DefaultProc]>;
-
-//===----------------------------------------------------------------------===//
-// Calling convention:
-//===----------------------------------------------------------------------===//
-
-include "SPUCallingConv.td"
-
-// Target:
-
-def SPUInstrInfo : InstrInfo {
-  let isLittleEndianEncoding = 1;
-}
-
-def SPU : Target {
-  let InstructionSet = SPUInstrInfo;
-}
diff --git a/lib/Target/CellSPU/SPU128InstrInfo.td b/lib/Target/CellSPU/SPU128InstrInfo.td
deleted file mode 100644
index e051e047333a..000000000000
--- a/lib/Target/CellSPU/SPU128InstrInfo.td
+++ /dev/null
@@ -1,41 +0,0 @@
-//===-- SPU128InstrInfo.td - Cell SPU 128-bit operations --*- tablegen -*--===//
-//
-//                     Cell SPU 128-bit operations
-//
-//===----------------------------------------------------------------------===//
-
-// zext 32->128: Zero extend 32-bit to 128-bit
-def : Pat<(i128 (zext R32C:$rSrc)),
-          (ROTQMBYIr128_zext_r32 R32C:$rSrc, 12)>;
-
-// zext 64->128: Zero extend 64-bit to 128-bit
-def : Pat<(i128 (zext R64C:$rSrc)),
-          (ROTQMBYIr128_zext_r64 R64C:$rSrc, 8)>;
-
-// zext 16->128: Zero extend 16-bit to 128-bit
-def : Pat<(i128 (zext R16C:$rSrc)),
-          (ROTQMBYIr128_zext_r32 (ANDi16i32 R16C:$rSrc, (ILAr32 0xffff)), 12)>;
-
-// zext 8->128: Zero extend 8-bit to 128-bit
-def : Pat<(i128 (zext R8C:$rSrc)),
-          (ROTQMBYIr128_zext_r32 (ANDIi8i32 R8C:$rSrc, 0xf), 12)>;
-
-// anyext 32->128: Zero extend 32-bit to 128-bit
-def : Pat<(i128 (anyext R32C:$rSrc)),
-          (ROTQMBYIr128_zext_r32 R32C:$rSrc, 12)>;
-
-// anyext 64->128: Zero extend 64-bit to 128-bit
-def : Pat<(i128 (anyext R64C:$rSrc)),
-          (ROTQMBYIr128_zext_r64 R64C:$rSrc, 8)>;
-
-// anyext 16->128: Zero extend 16-bit to 128-bit
-def : Pat<(i128 (anyext R16C:$rSrc)),
-          (ROTQMBYIr128_zext_r32 (ANDi16i32 R16C:$rSrc, (ILAr32 0xffff)), 12)>;
-
-// anyext 8->128: Zero extend 8-bit to 128-bit
-def : Pat<(i128 (anyext R8C:$rSrc)),
-          (ROTQMBYIr128_zext_r32 (ANDIi8i32 R8C:$rSrc, 0xf), 12)>;
-
-// Shift left
-def : Pat<(shl GPRC:$rA, R32C:$rB),
-          (SHLQBYBIr128 (SHLQBIr128 GPRC:$rA, R32C:$rB), R32C:$rB)>;
diff --git a/lib/Target/CellSPU/SPU64InstrInfo.td b/lib/Target/CellSPU/SPU64InstrInfo.td
deleted file mode 100644
index bea33b5362d2..000000000000
--- a/lib/Target/CellSPU/SPU64InstrInfo.td
+++ /dev/null
@@ -1,408 +0,0 @@
-//====-- SPU64InstrInfo.td - Cell SPU 64-bit operations ---*- tablegen -*--===//
-//
-//                     Cell SPU 64-bit operations
-//
-//===----------------------------------------------------------------------===//
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// 64-bit comparisons:
-//
-// 1. The instruction sequences for vector vice scalar differ by a
-//    constant. In the scalar case, we're only interested in the
-//    top two 32-bit slots, whereas we're interested in an exact
-//    all-four-slot match in the vector case.
-//
-// 2. There are no "immediate" forms, since loading 64-bit constants
-//    could be a constant pool load.
-//
-// 3. i64 setcc results are i32, which are subsequently converted to a FSM
-//    mask when used in a select pattern.
-//
-// 4. v2i64 setcc results are v4i32, which can be converted to a FSM mask (TODO)
-//    [Note: this may be moot, since gb produces v4i32 or r32.]
-//
-// 5. The code sequences for r64 and v2i64 are probably overly conservative,
-//    compared to the code that gcc produces.
-//
-// M00$E B!tes Kan be Pretty N@sTi!!!!! (apologies to Monty!)
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-// selb instruction definition for i64. Note that the selection mask is
-// a vector, produced by various forms of FSM:
-def SELBr64_cond:
-  SELBInst<(outs R64C:$rT), (ins R64C:$rA, R64C:$rB, VECREG:$rC),
-           [/* no pattern */]>;
-
-// The generic i64 select pattern, which assumes that the comparison result
-// is in a 32-bit register that contains a select mask pattern (i.e., gather
-// bits result):
-
-def : Pat<(select R32C:$rCond, R64C:$rFalse, R64C:$rTrue),
-          (SELBr64_cond R64C:$rTrue, R64C:$rFalse, (FSMr32 R32C:$rCond))>;
-
-// select the negative condition:
-class I64SELECTNegCond<PatFrag cond, CodeFrag compare>:
-  Pat<(select (i32 (cond R64C:$rA, R64C:$rB)), R64C:$rTrue, R64C:$rFalse),
-      (SELBr64_cond R64C:$rTrue, R64C:$rFalse, (FSMr32 compare.Fragment))>;
-
-// setcc the negative condition:
-class I64SETCCNegCond<PatFrag cond, CodeFrag compare>:
-  Pat<(cond R64C:$rA, R64C:$rB),
-      (XORIr32 compare.Fragment, -1)>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// The i64 seteq fragment that does the scalar->vector conversion and
-// comparison:
-def CEQr64compare:
-    CodeFrag<(CGTIv4i32 (GBv4i32 (CEQv4i32 (COPY_TO_REGCLASS R64C:$rA, VECREG),
-                                           (COPY_TO_REGCLASS R64C:$rB, VECREG))), 0xb)>;
-
-// The i64 seteq fragment that does the vector comparison
-def CEQv2i64compare:
-    CodeFrag<(CEQIv4i32 (GBv4i32 (CEQv4i32 VECREG:$rA, VECREG:$rB)), 0xf)>;
-
-// i64 seteq (equality): the setcc result is i32, which is converted to a
-// vector FSM mask when used in a select pattern.
-//
-// v2i64 seteq (equality): the setcc result is v4i32
-multiclass CompareEqual64 {
-  // Plain old comparison, converts back to i32 scalar
-  def r64: CodeFrag<(i32 (COPY_TO_REGCLASS CEQr64compare.Fragment, R32C))>;
-  def v2i64: CodeFrag<(i32 (COPY_TO_REGCLASS CEQv2i64compare.Fragment, R32C))>;
-
-  // SELB mask from FSM:
-  def r64mask: CodeFrag<(i32 (COPY_TO_REGCLASS 
-                               (FSMv4i32 CEQr64compare.Fragment), R32C))>;
-  def v2i64mask: CodeFrag<(i32 (COPY_TO_REGCLASS 
-                               (FSMv4i32 CEQv2i64compare.Fragment), R32C))>;
-}
-
-defm I64EQ: CompareEqual64;
-
-def : Pat<(seteq R64C:$rA, R64C:$rB), I64EQr64.Fragment>;
-def : Pat<(seteq (v2i64 VECREG:$rA), (v2i64 VECREG:$rB)), I64EQv2i64.Fragment>;
-
-// i64 setne:
-def : I64SETCCNegCond<setne, I64EQr64>;
-def : I64SELECTNegCond<setne, I64EQr64>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// i64 setugt/setule:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-def CLGTr64ugt:
-    CodeFrag<(CLGTv4i32 (COPY_TO_REGCLASS R64C:$rA, VECREG), 
-                        (COPY_TO_REGCLASS R64C:$rB, VECREG))>;
-
-def CLGTr64eq:
-    CodeFrag<(CEQv4i32 (COPY_TO_REGCLASS R64C:$rA, VECREG), 
-                       (COPY_TO_REGCLASS R64C:$rB, VECREG))>;
-    
-def CLGTr64compare:
-    CodeFrag<(SELBv2i64 CLGTr64ugt.Fragment,
-                        (XSWDv2i64 CLGTr64ugt.Fragment),
-                        CLGTr64eq.Fragment)>;
-
-def CLGTv2i64ugt:
-    CodeFrag<(CLGTv4i32 VECREG:$rA, VECREG:$rB)>;
-
-def CLGTv2i64eq:
-    CodeFrag<(CEQv4i32 VECREG:$rA, VECREG:$rB)>;
-    
-def CLGTv2i64compare:
-    CodeFrag<(SELBv2i64 CLGTv2i64ugt.Fragment,
-                        (XSWDv2i64 CLGTr64ugt.Fragment),
-                        CLGTv2i64eq.Fragment)>;
-
-multiclass CompareLogicalGreaterThan64 {
-  // Plain old comparison, converts back to i32 scalar
-  def r64: CodeFrag<(i32 (COPY_TO_REGCLASS CLGTr64compare.Fragment, R32C))>;
-  def v2i64: CodeFrag<CLGTv2i64compare.Fragment>;
-
-  // SELB mask from FSM:
-  def r64mask: CodeFrag<(i32 (COPY_TO_REGCLASS 
-                               (FSMv4i32 CLGTr64compare.Fragment), R32C))>;
-  def v2i64mask: CodeFrag<(i32 (COPY_TO_REGCLASS 
-                               (FSMv4i32 CLGTv2i64compare.Fragment), R32C))>;
-}
-
-defm I64LGT: CompareLogicalGreaterThan64;
-
-def : Pat<(setugt R64C:$rA, R64C:$rB), I64LGTr64.Fragment>;
-//def : Pat<(setugt (v2i64 VECREG:$rA), (v2i64 VECREG:$rB)),
-//          I64LGTv2i64.Fragment>;
-
-// i64 setult:
-def : I64SETCCNegCond<setule, I64LGTr64>;
-def : I64SELECTNegCond<setule, I64LGTr64>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// i64 setuge/setult:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-def CLGEr64compare:
-    CodeFrag<(CGTIv4i32 (GBv4i32 (ORv4i32 CLGTr64ugt.Fragment,
-                                          CLGTr64eq.Fragment)), 0xb)>;
-
-def CLGEv2i64compare:
-    CodeFrag<(CEQIv4i32 (GBv4i32 (ORv4i32 CLGTv2i64ugt.Fragment,
-                                          CLGTv2i64eq.Fragment)), 0xf)>;
-
-multiclass CompareLogicalGreaterEqual64 {
-  // Plain old comparison, converts back to i32 scalar
-  def r64: CodeFrag<(i32 (COPY_TO_REGCLASS CLGEr64compare.Fragment, R32C))>;
-  def v2i64: CodeFrag<CLGEv2i64compare.Fragment>;
-
-  // SELB mask from FSM:
-  def r64mask: CodeFrag<(i32 (COPY_TO_REGCLASS 
-                           (FSMv4i32 CLGEr64compare.Fragment), R32C))>;
-  def v2i64mask: CodeFrag<(i32 (COPY_TO_REGCLASS 
-                           (FSMv4i32 CLGEv2i64compare.Fragment),R32C))>;
-}
-
-defm I64LGE: CompareLogicalGreaterEqual64;
-
-def : Pat<(setuge R64C:$rA, R64C:$rB), I64LGEr64.Fragment>;
-def : Pat<(v2i64 (setuge (v2i64 VECREG:$rA), (v2i64 VECREG:$rB))),
-          I64LGEv2i64.Fragment>;
-                  
-
-// i64 setult:
-def : I64SETCCNegCond<setult, I64LGEr64>;
-def : I64SELECTNegCond<setult, I64LGEr64>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// i64 setgt/setle:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-def CGTr64sgt:
-    CodeFrag<(CGTv4i32 (COPY_TO_REGCLASS R64C:$rA, VECREG), 
-                       (COPY_TO_REGCLASS R64C:$rB, VECREG))>;
-
-def CGTr64eq:
-    CodeFrag<(CEQv4i32 (COPY_TO_REGCLASS R64C:$rA, VECREG), 
-                       (COPY_TO_REGCLASS R64C:$rB, VECREG))>;
-    
-def CGTr64compare:
-    CodeFrag<(SELBv2i64 CGTr64sgt.Fragment,
-                        (XSWDv2i64 CGTr64sgt.Fragment),
-                        CGTr64eq.Fragment)>;
-
-def CGTv2i64sgt:
-    CodeFrag<(CGTv4i32 VECREG:$rA, VECREG:$rB)>;
-
-def CGTv2i64eq:
-    CodeFrag<(CEQv4i32 VECREG:$rA, VECREG:$rB)>;
-    
-def CGTv2i64compare:
-    CodeFrag<(SELBv2i64 CGTv2i64sgt.Fragment,
-                        (XSWDv2i64 CGTr64sgt.Fragment),
-                        CGTv2i64eq.Fragment)>;
-
-multiclass CompareGreaterThan64 {
-  // Plain old comparison, converts back to i32 scalar
-  def r64: CodeFrag<(i32 (COPY_TO_REGCLASS CGTr64compare.Fragment, R32C))>;
-  def v2i64: CodeFrag<CGTv2i64compare.Fragment>;
-
-  // SELB mask from FSM:
-  def r64mask: CodeFrag<(i32 (COPY_TO_REGCLASS 
-                             (FSMv4i32 CGTr64compare.Fragment), R32C))>;
-  def v2i64mask: CodeFrag<(i32 (COPY_TO_REGCLASS 
-                               (FSMv4i32 CGTv2i64compare.Fragment), R32C))>;
-}
-
-defm I64GT: CompareLogicalGreaterThan64;
-
-def : Pat<(setgt R64C:$rA, R64C:$rB), I64GTr64.Fragment>;
-//def : Pat<(setgt (v2i64 VECREG:$rA), (v2i64 VECREG:$rB)),
-//                  I64GTv2i64.Fragment>;
-
-// i64 setult:
-def : I64SETCCNegCond<setle, I64GTr64>;
-def : I64SELECTNegCond<setle, I64GTr64>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// i64 setge/setlt:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-    
-def CGEr64compare:
-    CodeFrag<(CGTIv4i32 (GBv4i32 (ORv4i32 CGTr64sgt.Fragment,
-                                          CGTr64eq.Fragment)), 0xb)>;
-
-def CGEv2i64compare:
-    CodeFrag<(CEQIv4i32 (GBv4i32 (ORv4i32 CGTv2i64sgt.Fragment,
-                                          CGTv2i64eq.Fragment)), 0xf)>;
-
-multiclass CompareGreaterEqual64 {
-  // Plain old comparison, converts back to i32 scalar
-  def r64: CodeFrag<(i32 (COPY_TO_REGCLASS CGEr64compare.Fragment, R32C))>;
-  def v2i64: CodeFrag<CGEv2i64compare.Fragment>;
-
-  // SELB mask from FSM:
-  def r64mask: CodeFrag<(i32 (COPY_TO_REGCLASS (FSMv4i32 CGEr64compare.Fragment),R32C))>;
-  def v2i64mask: CodeFrag<(i32 (COPY_TO_REGCLASS (FSMv4i32 CGEv2i64compare.Fragment),R32C))>;
-}
-
-defm I64GE: CompareGreaterEqual64;
-
-def : Pat<(setge R64C:$rA, R64C:$rB), I64GEr64.Fragment>;
-def : Pat<(v2i64 (setge (v2i64 VECREG:$rA), (v2i64 VECREG:$rB))),
-          I64GEv2i64.Fragment>;
-
-// i64 setult:
-def : I64SETCCNegCond<setlt, I64GEr64>;
-def : I64SELECTNegCond<setlt, I64GEr64>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// v2i64, i64 add
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class v2i64_add_cg<dag lhs, dag rhs>:
-    CodeFrag<(CGv4i32 lhs, rhs)>;
-
-class v2i64_add_1<dag lhs, dag rhs, dag cg, dag cg_mask>:
-    CodeFrag<(ADDXv4i32 lhs, rhs, (SHUFBv4i32 cg, cg, cg_mask))>;
-
-class v2i64_add<dag lhs, dag rhs, dag cg_mask>:
-    v2i64_add_1<lhs, rhs, v2i64_add_cg<lhs, rhs>.Fragment, cg_mask>;
-
-def : Pat<(SPUadd64 R64C:$rA, R64C:$rB, (v4i32 VECREG:$rCGmask)),
-           (COPY_TO_REGCLASS v2i64_add<(COPY_TO_REGCLASS R64C:$rA, VECREG),
-                                  (COPY_TO_REGCLASS R64C:$rB, VECREG),
-                                  (v4i32 VECREG:$rCGmask)>.Fragment, R64C)>;
-
-def : Pat<(SPUadd64 (v2i64 VECREG:$rA), (v2i64 VECREG:$rB),
-                    (v4i32 VECREG:$rCGmask)),
-           v2i64_add<(v2i64 VECREG:$rA),
-                     (v2i64 VECREG:$rB),
-                     (v4i32 VECREG:$rCGmask)>.Fragment>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// v2i64, i64 subtraction
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class v2i64_sub_bg<dag lhs, dag rhs>: CodeFrag<(BGv4i32 lhs, rhs)>;
-
-class v2i64_sub<dag lhs, dag rhs, dag bg, dag bg_mask>:
-    CodeFrag<(SFXv4i32 lhs, rhs, (SHUFBv4i32 bg, bg, bg_mask))>;
-
-def : Pat<(SPUsub64 R64C:$rA, R64C:$rB, (v4i32 VECREG:$rCGmask)),
-           (COPY_TO_REGCLASS 
-               v2i64_sub<(COPY_TO_REGCLASS R64C:$rA, VECREG),
-                         (COPY_TO_REGCLASS R64C:$rB, VECREG),
-                         v2i64_sub_bg<(COPY_TO_REGCLASS R64C:$rA, VECREG),
-                                      (COPY_TO_REGCLASS R64C:$rB, VECREG)>.Fragment,
-                                  (v4i32 VECREG:$rCGmask)>.Fragment, R64C)>;
-
-def : Pat<(SPUsub64 (v2i64 VECREG:$rA), (v2i64 VECREG:$rB),
-                    (v4i32 VECREG:$rCGmask)),
-           v2i64_sub<(v2i64 VECREG:$rA),
-                     (v2i64 VECREG:$rB),
-                     v2i64_sub_bg<(v2i64 VECREG:$rA),
-                                  (v2i64 VECREG:$rB)>.Fragment,
-                     (v4i32 VECREG:$rCGmask)>.Fragment>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// v2i64, i64 multiply
-//
-// Note: i64 multiply is simply the vector->scalar conversion of the
-// full-on v2i64 multiply, since the entire vector has to be manipulated
-// anyway.
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class v2i64_mul_ahi64<dag rA> :
-    CodeFrag<(SELBv4i32 rA, (ILv4i32 0), (FSMBIv4i32 0x0f0f))>;
-
-class v2i64_mul_bhi64<dag rB> :
-    CodeFrag<(SELBv4i32 rB, (ILv4i32 0), (FSMBIv4i32 0x0f0f))>;
-
-class v2i64_mul_alo64<dag rB> :
-    CodeFrag<(SELBv4i32 rB, (ILv4i32 0), (FSMBIv4i32 0xf0f0))>;
-
-class v2i64_mul_blo64<dag rB> :
-    CodeFrag<(SELBv4i32 rB, (ILv4i32 0), (FSMBIv4i32 0xf0f0))>;
-
-class v2i64_mul_ashlq2<dag rA>:
-    CodeFrag<(SHLQBYIv4i32 rA, 0x2)>;
-
-class v2i64_mul_ashlq4<dag rA>:
-    CodeFrag<(SHLQBYIv4i32 rA, 0x4)>;
-
-class v2i64_mul_bshlq2<dag rB> :
-    CodeFrag<(SHLQBYIv4i32 rB, 0x2)>;
-
-class v2i64_mul_bshlq4<dag rB> :
-    CodeFrag<(SHLQBYIv4i32 rB, 0x4)>;
-
-class v2i64_highprod<dag rA, dag rB>:
-    CodeFrag<(Av4i32
-                (Av4i32
-                  (MPYUv4i32 v2i64_mul_bshlq4<rB>.Fragment,     // a1 x b3
-                             v2i64_mul_ahi64<rA>.Fragment),
-                  (MPYHv4i32 v2i64_mul_ahi64<rA>.Fragment,      // a0 x b3
-                             v2i64_mul_bshlq4<rB>.Fragment)),
-                (Av4i32
-                  (MPYHv4i32 v2i64_mul_bhi64<rB>.Fragment,
-                             v2i64_mul_ashlq4<rA>.Fragment),
-                  (Av4i32
-                      (MPYHv4i32 v2i64_mul_ashlq4<rA>.Fragment,
-                                 v2i64_mul_bhi64<rB>.Fragment),
-                    (Av4i32
-                      (MPYUv4i32 v2i64_mul_ashlq4<rA>.Fragment,
-                                 v2i64_mul_bhi64<rB>.Fragment),
-                      (Av4i32
-                        (MPYHv4i32 v2i64_mul_ashlq2<rA>.Fragment,
-                                   v2i64_mul_bshlq2<rB>.Fragment),
-                        (MPYUv4i32 v2i64_mul_ashlq2<rA>.Fragment,
-                                   v2i64_mul_bshlq2<rB>.Fragment))))))>;
-
-class v2i64_mul_a3_b3<dag rA, dag rB>:
-    CodeFrag<(MPYUv4i32 v2i64_mul_alo64<rA>.Fragment,
-                        v2i64_mul_blo64<rB>.Fragment)>;
-
-class v2i64_mul_a2_b3<dag rA, dag rB>:
-    CodeFrag<(SELBv4i32 (SHLQBYIv4i32
-                          (MPYHHUv4i32 v2i64_mul_alo64<rA>.Fragment,
-                                       v2i64_mul_bshlq2<rB>.Fragment), 0x2),
-                        (ILv4i32 0),
-                        (FSMBIv4i32 0xc3c3))>;
-
-class v2i64_mul_a3_b2<dag rA, dag rB>:
-    CodeFrag<(SELBv4i32 (SHLQBYIv4i32
-                          (MPYHHUv4i32 v2i64_mul_blo64<rB>.Fragment,
-                                       v2i64_mul_ashlq2<rA>.Fragment), 0x2),
-                        (ILv4i32 0),
-                        (FSMBIv4i32 0xc3c3))>;
-
-class v2i64_lowsum<dag rA, dag rB, dag rCGmask>:
-    v2i64_add<v2i64_add<v2i64_mul_a3_b3<rA, rB>.Fragment,
-                        v2i64_mul_a2_b3<rA, rB>.Fragment, rCGmask>.Fragment,
-              v2i64_mul_a3_b2<rA, rB>.Fragment, rCGmask>;
-
-class v2i64_mul<dag rA, dag rB, dag rCGmask>:
-    v2i64_add<v2i64_lowsum<rA, rB, rCGmask>.Fragment,
-              (SELBv4i32 v2i64_highprod<rA, rB>.Fragment,
-                         (ILv4i32 0),
-                         (FSMBIv4i32 0x0f0f)),
-              rCGmask>;
-
-def : Pat<(SPUmul64 R64C:$rA, R64C:$rB, (v4i32 VECREG:$rCGmask)),
-          (COPY_TO_REGCLASS v2i64_mul<(COPY_TO_REGCLASS R64C:$rA, VECREG),
-                                 (COPY_TO_REGCLASS R64C:$rB, VECREG),
-                                 (v4i32 VECREG:$rCGmask)>.Fragment, R64C)>;
-
-def : Pat<(SPUmul64 (v2i64 VECREG:$rA), (v2i64 VECREG:$rB),
-                    (v4i32 VECREG:$rCGmask)),
-          v2i64_mul<(v2i64 VECREG:$rA), (v2i64 VECREG:$rB),
-                    (v4i32 VECREG:$rCGmask)>.Fragment>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// f64 comparisons
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-// selb instruction definition for i64. Note that the selection mask is
-// a vector, produced by various forms of FSM:
-def SELBf64_cond:
-   SELBInst<(outs R64FP:$rT), (ins R64FP:$rA, R64FP:$rB, R32C:$rC),
-            [(set R64FP:$rT,
-                  (select R32C:$rC, R64FP:$rB, R64FP:$rA))]>;
diff --git a/lib/Target/CellSPU/SPUAsmPrinter.cpp b/lib/Target/CellSPU/SPUAsmPrinter.cpp
deleted file mode 100644
index 3396e8b1ef39..000000000000
--- a/lib/Target/CellSPU/SPUAsmPrinter.cpp
+++ /dev/null
@@ -1,333 +0,0 @@
-//===-- SPUAsmPrinter.cpp - Print machine instrs to Cell SPU assembly -----===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file contains a printer that converts from our internal representation
-// of machine-dependent LLVM code to Cell SPU assembly language. This printer
-// is the output mechanism used by `llc'.
-//
-//===----------------------------------------------------------------------===//
-
-#define DEBUG_TYPE "asmprinter"
-#include "SPU.h"
-#include "SPUTargetMachine.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Module.h"
-#include "llvm/CodeGen/AsmPrinter.h"
-#include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/MC/MCStreamer.h"
-#include "llvm/MC/MCAsmInfo.h"
-#include "llvm/MC/MCSymbol.h"
-#include "llvm/Target/Mangler.h"
-#include "llvm/Target/TargetLoweringObjectFile.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/ADT/SmallString.h"
-#include "llvm/ADT/StringExtras.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/TargetRegistry.h"
-#include "llvm/Support/raw_ostream.h"
-using namespace llvm;
-
-namespace {
-  class SPUAsmPrinter : public AsmPrinter {
-  public:
-    explicit SPUAsmPrinter(TargetMachine &TM, MCStreamer &Streamer) :
-      AsmPrinter(TM, Streamer) {}
-
-    virtual const char *getPassName() const {
-      return "STI CBEA SPU Assembly Printer";
-    }
-
-    /// printInstruction - This method is automatically generated by tablegen
-    /// from the instruction set description.
-    void printInstruction(const MachineInstr *MI, raw_ostream &OS);
-    static const char *getRegisterName(unsigned RegNo);
-
-
-    void EmitInstruction(const MachineInstr *MI) {
-      SmallString<128> Str;
-      raw_svector_ostream OS(Str);
-      printInstruction(MI, OS);
-      OutStreamer.EmitRawText(OS.str());
-    }
-    void printOp(const MachineOperand &MO, raw_ostream &OS);
-
-    void printOperand(const MachineInstr *MI, unsigned OpNo, raw_ostream &O) {
-      const MachineOperand &MO = MI->getOperand(OpNo);
-      if (MO.isReg()) {
-        O << getRegisterName(MO.getReg());
-      } else if (MO.isImm()) {
-        O << MO.getImm();
-      } else {
-        printOp(MO, O);
-      }
-    }
-
-    bool PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
-                         unsigned AsmVariant, const char *ExtraCode,
-                         raw_ostream &O);
-    bool PrintAsmMemoryOperand(const MachineInstr *MI, unsigned OpNo,
-                               unsigned AsmVariant, const char *ExtraCode,
-                               raw_ostream &O);
-
-
-    void
-    printU7ImmOperand(const MachineInstr *MI, unsigned OpNo, raw_ostream &O)
-    {
-      unsigned int value = MI->getOperand(OpNo).getImm();
-      assert(value < (1 << 8) && "Invalid u7 argument");
-      O << value;
-    }
-
-    void
-    printShufAddr(const MachineInstr *MI, unsigned OpNo, raw_ostream &O)
-    {
-      char value = MI->getOperand(OpNo).getImm();
-      O << (int) value;
-      O << "(";
-      printOperand(MI, OpNo+1, O);
-      O << ")";
-    }
-
-    void
-    printS16ImmOperand(const MachineInstr *MI, unsigned OpNo, raw_ostream &O)
-    {
-      O << (short) MI->getOperand(OpNo).getImm();
-    }
-
-    void
-    printU16ImmOperand(const MachineInstr *MI, unsigned OpNo, raw_ostream &O)
-    {
-      O << (unsigned short)MI->getOperand(OpNo).getImm();
-    }
-
-    void
-    printMemRegReg(const MachineInstr *MI, unsigned OpNo, raw_ostream &O) {
-      // When used as the base register, r0 reads constant zero rather than
-      // the value contained in the register.  For this reason, the darwin
-      // assembler requires that we print r0 as 0 (no r) when used as the base.
-      const MachineOperand &MO = MI->getOperand(OpNo);
-      O << getRegisterName(MO.getReg()) << ", ";
-      printOperand(MI, OpNo+1, O);
-    }
-
-    void
-    printU18ImmOperand(const MachineInstr *MI, unsigned OpNo, raw_ostream &O)
-    {
-      unsigned int value = MI->getOperand(OpNo).getImm();
-      assert(value <= (1 << 19) - 1 && "Invalid u18 argument");
-      O << value;
-    }
-
-    void
-    printS10ImmOperand(const MachineInstr *MI, unsigned OpNo, raw_ostream &O)
-    {
-      short value = MI->getOperand(OpNo).getImm();
-      assert((value >= -(1 << 9) && value <= (1 << 9) - 1)
-             && "Invalid s10 argument");
-      O << value;
-    }
-
-    void
-    printU10ImmOperand(const MachineInstr *MI, unsigned OpNo, raw_ostream &O)
-    {
-      short value = MI->getOperand(OpNo).getImm();
-      assert((value <= (1 << 10) - 1) && "Invalid u10 argument");
-      O << value;
-    }
-
-    void
-    printDFormAddr(const MachineInstr *MI, unsigned OpNo, raw_ostream &O)
-    {
-      assert(MI->getOperand(OpNo).isImm() &&
-             "printDFormAddr first operand is not immediate");
-      int64_t value = int64_t(MI->getOperand(OpNo).getImm());
-      int16_t value16 = int16_t(value);
-      assert((value16 >= -(1 << (9+4)) && value16 <= (1 << (9+4)) - 1)
-             && "Invalid dform s10 offset argument");
-      O << (value16 & ~0xf) << "(";
-      printOperand(MI, OpNo+1, O);
-      O << ")";
-    }
-
-    void
-    printAddr256K(const MachineInstr *MI, unsigned OpNo, raw_ostream &O)
-    {
-      /* Note: operand 1 is an offset or symbol name. */
-      if (MI->getOperand(OpNo).isImm()) {
-        printS16ImmOperand(MI, OpNo, O);
-      } else {
-        printOp(MI->getOperand(OpNo), O);
-        if (MI->getOperand(OpNo+1).isImm()) {
-          int displ = int(MI->getOperand(OpNo+1).getImm());
-          if (displ > 0)
-            O << "+" << displ;
-          else if (displ < 0)
-            O << displ;
-        }
-      }
-    }
-
-    void printCallOperand(const MachineInstr *MI, unsigned OpNo, raw_ostream &O) {
-      printOp(MI->getOperand(OpNo), O);
-    }
-
-    void printHBROperand(const MachineInstr *MI, unsigned OpNo, raw_ostream &O) {
-      printOp(MI->getOperand(OpNo), O);
-    }
-
-    void printPCRelativeOperand(const MachineInstr *MI, unsigned OpNo, raw_ostream &O) {
-      // Used to generate a ".-<target>", but it turns out that the assembler
-      // really wants the target.
-      //
-      // N.B.: This operand is used for call targets. Branch hints are another
-      // animal entirely.
-      printOp(MI->getOperand(OpNo), O);
-    }
-
-    void printSymbolHi(const MachineInstr *MI, unsigned OpNo, raw_ostream &O) {
-      if (MI->getOperand(OpNo).isImm()) {
-        printS16ImmOperand(MI, OpNo, O);
-      } else {
-        printOp(MI->getOperand(OpNo), O);
-        O << "@h";
-      }
-    }
-
-    void printSymbolLo(const MachineInstr *MI, unsigned OpNo, raw_ostream &O) {
-      if (MI->getOperand(OpNo).isImm()) {
-        printS16ImmOperand(MI, OpNo, O);
-      } else {
-        printOp(MI->getOperand(OpNo), O);
-        O << "@l";
-      }
-    }
-
-    /// Print local store address
-    void printSymbolLSA(const MachineInstr *MI, unsigned OpNo, raw_ostream &O) {
-      printOp(MI->getOperand(OpNo), O);
-    }
-
-    void printROTHNeg7Imm(const MachineInstr *MI, unsigned OpNo,
-                          raw_ostream &O) {
-      if (MI->getOperand(OpNo).isImm()) {
-        int value = (int) MI->getOperand(OpNo).getImm();
-        assert((value >= 0 && value < 16)
-               && "Invalid negated immediate rotate 7-bit argument");
-        O << -value;
-      } else {
-        llvm_unreachable("Invalid/non-immediate rotate amount in printRotateNeg7Imm");
-      }
-    }
-
-    void printROTNeg7Imm(const MachineInstr *MI, unsigned OpNo, raw_ostream &O){
-      assert(MI->getOperand(OpNo).isImm() &&
-             "Invalid/non-immediate rotate amount in printRotateNeg7Imm");
-      int value = (int) MI->getOperand(OpNo).getImm();
-      assert((value >= 0 && value <= 32)
-             && "Invalid negated immediate rotate 7-bit argument");
-      O << -value;
-    }
-  };
-} // end of anonymous namespace
-
-// Include the auto-generated portion of the assembly writer
-#include "SPUGenAsmWriter.inc"
-
-void SPUAsmPrinter::printOp(const MachineOperand &MO, raw_ostream &O) {
-  switch (MO.getType()) {
-  case MachineOperand::MO_Immediate:
-    report_fatal_error("printOp() does not handle immediate values");
-
-  case MachineOperand::MO_MachineBasicBlock:
-    O << *MO.getMBB()->getSymbol();
-    return;
-  case MachineOperand::MO_JumpTableIndex:
-    O << MAI->getPrivateGlobalPrefix() << "JTI" << getFunctionNumber()
-      << '_' << MO.getIndex();
-    return;
-  case MachineOperand::MO_ConstantPoolIndex:
-    O << MAI->getPrivateGlobalPrefix() << "CPI" << getFunctionNumber()
-      << '_' << MO.getIndex();
-    return;
-  case MachineOperand::MO_ExternalSymbol:
-    // Computing the address of an external symbol, not calling it.
-    if (TM.getRelocationModel() != Reloc::Static) {
-      O << "L" << MAI->getGlobalPrefix() << MO.getSymbolName()
-        << "$non_lazy_ptr";
-      return;
-    }
-    O << *GetExternalSymbolSymbol(MO.getSymbolName());
-    return;
-  case MachineOperand::MO_GlobalAddress:
-    // External or weakly linked global variables need non-lazily-resolved
-    // stubs
-    if (TM.getRelocationModel() != Reloc::Static) {
-      const GlobalValue *GV = MO.getGlobal();
-      if (((GV->isDeclaration() || GV->hasWeakLinkage() ||
-            GV->hasLinkOnceLinkage() || GV->hasCommonLinkage()))) {
-        O << *GetSymbolWithGlobalValueBase(GV, "$non_lazy_ptr");
-        return;
-      }
-    }
-    O << *Mang->getSymbol(MO.getGlobal());
-    return;
-  case MachineOperand::MO_MCSymbol:
-    O << *(MO.getMCSymbol());
-    return;
-  default:
-    O << "<unknown operand type: " << MO.getType() << ">";
-    return;
-  }
-}
-
-/// PrintAsmOperand - Print out an operand for an inline asm expression.
-///
-bool SPUAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
-                                    unsigned AsmVariant,
-                                    const char *ExtraCode, raw_ostream &O) {
-  // Does this asm operand have a single letter operand modifier?
-  if (ExtraCode && ExtraCode[0]) {
-    if (ExtraCode[1] != 0) return true; // Unknown modifier.
-
-    switch (ExtraCode[0]) {
-    default:
-      // See if this is a generic print operand
-      return AsmPrinter::PrintAsmOperand(MI, OpNo, AsmVariant, ExtraCode, O);
-    case 'L': // Write second word of DImode reference.
-      // Verify that this operand has two consecutive registers.
-      if (!MI->getOperand(OpNo).isReg() ||
-          OpNo+1 == MI->getNumOperands() ||
-          !MI->getOperand(OpNo+1).isReg())
-        return true;
-      ++OpNo;   // Return the high-part.
-      break;
-    }
-  }
-
-  printOperand(MI, OpNo, O);
-  return false;
-}
-
-bool SPUAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI,
-                                          unsigned OpNo, unsigned AsmVariant,
-                                          const char *ExtraCode,
-                                          raw_ostream &O) {
-  if (ExtraCode && ExtraCode[0])
-    return true; // Unknown modifier.
-  printMemRegReg(MI, OpNo, O);
-  return false;
-}
-
-// Force static initialization.
-extern "C" void LLVMInitializeCellSPUAsmPrinter() { 
-  RegisterAsmPrinter<SPUAsmPrinter> X(TheCellSPUTarget);
-}
diff --git a/lib/Target/CellSPU/SPUCallingConv.td b/lib/Target/CellSPU/SPUCallingConv.td
deleted file mode 100644
index 9bc6be79860b..000000000000
--- a/lib/Target/CellSPU/SPUCallingConv.td
+++ /dev/null
@@ -1,53 +0,0 @@
-//===- SPUCallingConv.td - Calling Conventions for CellSPU -*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This describes the calling conventions for the STI Cell SPU architecture.
-//
-//===----------------------------------------------------------------------===//
-
-//===----------------------------------------------------------------------===//
-// Return Value Calling Convention
-//===----------------------------------------------------------------------===//
-
-// Return-value convention for Cell SPU: return value to be passed in reg 3-74
-def RetCC_SPU : CallingConv<[
-  CCIfType<[i8,i16,i32,i64,i128,f32,f64,v16i8,v8i16,v4i32,v2i64,v4f32,v2f64],
-  CCAssignToReg<[R3,   R4,  R5,  R6,  R7,  R8,  R9, R10, R11,
-                 R12, R13, R14, R15, R16, R17, R18, R19, R20,
-                 R21, R22, R23, R24, R25, R26, R27, R28, R29,
-                 R30, R31, R32, R33, R34, R35, R36, R37, R38,
-                 R39, R40, R41, R42, R43, R44, R45, R46, R47,
-                 R48, R49, R50, R51, R52, R53, R54, R55, R56,
-                 R57, R58, R59, R60, R61, R62, R63, R64, R65,
-                 R66, R67, R68, R69, R70, R71, R72, R73, R74]>>
-]>;
-
-
-//===----------------------------------------------------------------------===//
-// CellSPU Argument Calling Conventions
-//===----------------------------------------------------------------------===//
-def CCC_SPU : CallingConv<[
-  CCIfType<[i8, i16, i32, i64, i128, f32, f64, 
-            v16i8, v8i16, v4i32, v4f32, v2i64, v2f64],
-            CCAssignToReg<[R3,   R4,  R5,  R6,  R7,  R8,  R9, R10, R11,
-                           R12, R13, R14, R15, R16, R17, R18, R19, R20,
-                           R21, R22, R23, R24, R25, R26, R27, R28, R29,
-                           R30, R31, R32, R33, R34, R35, R36, R37, R38,
-                           R39, R40, R41, R42, R43, R44, R45, R46, R47,
-                           R48, R49, R50, R51, R52, R53, R54, R55, R56,
-                           R57, R58, R59, R60, R61, R62, R63, R64, R65,
-                           R66, R67, R68, R69, R70, R71, R72, R73, R74]>>,
-  // Integer/FP values get stored in stack slots that are 8 bytes in size and
-  // 8-byte aligned if there are no more registers to hold them.
-  CCIfType<[i32, i64, f32, f64], CCAssignToStack<8, 8>>,
-  
-  // Vectors get 16-byte stack slots that are 16-byte aligned.
-  CCIfType<[v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
-              CCAssignToStack<16, 16>>
-]>;
diff --git a/lib/Target/CellSPU/SPUFrameLowering.cpp b/lib/Target/CellSPU/SPUFrameLowering.cpp
deleted file mode 100644
index f01199515a11..000000000000
--- a/lib/Target/CellSPU/SPUFrameLowering.cpp
+++ /dev/null
@@ -1,256 +0,0 @@
-//===-- SPUTargetMachine.cpp - Define TargetMachine for Cell SPU ----------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// Top-level implementation for the Cell SPU target.
-//
-//===----------------------------------------------------------------------===//
-
-#include "SPUFrameLowering.h"
-#include "SPU.h"
-#include "SPUInstrBuilder.h"
-#include "SPUInstrInfo.h"
-#include "llvm/Function.h"
-#include "llvm/CodeGen/MachineFrameInfo.h"
-#include "llvm/CodeGen/MachineFunction.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Support/CommandLine.h"
-using namespace llvm;
-
-//===----------------------------------------------------------------------===//
-// SPUFrameLowering:
-//===----------------------------------------------------------------------===//
-
-SPUFrameLowering::SPUFrameLowering(const SPUSubtarget &sti)
-  : TargetFrameLowering(TargetFrameLowering::StackGrowsDown, 16, 0),
-    Subtarget(sti) {
-  LR[0].first = SPU::R0;
-  LR[0].second = 16;
-}
-
-
-//--------------------------------------------------------------------------
-// hasFP - Return true if the specified function actually has a dedicated frame
-// pointer register.  This is true if the function needs a frame pointer and has
-// a non-zero stack size.
-bool SPUFrameLowering::hasFP(const MachineFunction &MF) const {
-  const MachineFrameInfo *MFI = MF.getFrameInfo();
-
-  return MFI->getStackSize() &&
-    (MF.getTarget().Options.DisableFramePointerElim(MF) ||
-     MFI->hasVarSizedObjects());
-}
-
-
-/// determineFrameLayout - Determine the size of the frame and maximum call
-/// frame size.
-void SPUFrameLowering::determineFrameLayout(MachineFunction &MF) const {
-  MachineFrameInfo *MFI = MF.getFrameInfo();
-
-  // Get the number of bytes to allocate from the FrameInfo
-  unsigned FrameSize = MFI->getStackSize();
-
-  // Get the alignments provided by the target, and the maximum alignment
-  // (if any) of the fixed frame objects.
-  unsigned TargetAlign = getStackAlignment();
-  unsigned Align = std::max(TargetAlign, MFI->getMaxAlignment());
-  assert(isPowerOf2_32(Align) && "Alignment is not power of 2");
-  unsigned AlignMask = Align - 1;
-
-  // Get the maximum call frame size of all the calls.
-  unsigned maxCallFrameSize = MFI->getMaxCallFrameSize();
-
-  // If we have dynamic alloca then maxCallFrameSize needs to be aligned so
-  // that allocations will be aligned.
-  if (MFI->hasVarSizedObjects())
-    maxCallFrameSize = (maxCallFrameSize + AlignMask) & ~AlignMask;
-
-  // Update maximum call frame size.
-  MFI->setMaxCallFrameSize(maxCallFrameSize);
-
-  // Include call frame size in total.
-  FrameSize += maxCallFrameSize;
-
-  // Make sure the frame is aligned.
-  FrameSize = (FrameSize + AlignMask) & ~AlignMask;
-
-  // Update frame info.
-  MFI->setStackSize(FrameSize);
-}
-
-void SPUFrameLowering::emitPrologue(MachineFunction &MF) const {
-  MachineBasicBlock &MBB = MF.front();   // Prolog goes in entry BB
-  MachineBasicBlock::iterator MBBI = MBB.begin();
-  MachineFrameInfo *MFI = MF.getFrameInfo();
-  const SPUInstrInfo &TII =
-    *static_cast<const SPUInstrInfo*>(MF.getTarget().getInstrInfo());
-  MachineModuleInfo &MMI = MF.getMMI();
-  DebugLoc dl = MBBI != MBB.end() ? MBBI->getDebugLoc() : DebugLoc();
-
-  // Prepare for debug frame info.
-  bool hasDebugInfo = MMI.hasDebugInfo();
-  MCSymbol *FrameLabel = 0;
-
-  // Move MBBI back to the beginning of the function.
-  MBBI = MBB.begin();
-
-  // Work out frame sizes.
-  determineFrameLayout(MF);
-  int FrameSize = MFI->getStackSize();
-
-  assert((FrameSize & 0xf) == 0
-         && "SPURegisterInfo::emitPrologue: FrameSize not aligned");
-
-  // the "empty" frame size is 16 - just the register scavenger spill slot
-  if (FrameSize > 16 || MFI->adjustsStack()) {
-    FrameSize = -(FrameSize + SPUFrameLowering::minStackSize());
-    if (hasDebugInfo) {
-      // Mark effective beginning of when frame pointer becomes valid.
-      FrameLabel = MMI.getContext().CreateTempSymbol();
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::PROLOG_LABEL)).addSym(FrameLabel);
-    }
-
-    // Adjust stack pointer, spilling $lr -> 16($sp) and $sp -> -FrameSize($sp)
-    // for the ABI
-    BuildMI(MBB, MBBI, dl, TII.get(SPU::STQDr32), SPU::R0).addImm(16)
-      .addReg(SPU::R1);
-    if (isInt<10>(FrameSize)) {
-      // Spill $sp to adjusted $sp
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::STQDr32), SPU::R1).addImm(FrameSize)
-        .addReg(SPU::R1);
-      // Adjust $sp by required amout
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::AIr32), SPU::R1).addReg(SPU::R1)
-        .addImm(FrameSize);
-    } else if (isInt<16>(FrameSize)) {
-      // Frame size can be loaded into ILr32n, so temporarily spill $r2 and use
-      // $r2 to adjust $sp:
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::STQDr128), SPU::R2)
-        .addImm(-16)
-        .addReg(SPU::R1);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::ILr32), SPU::R2)
-        .addImm(FrameSize);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::STQXr32), SPU::R1)
-        .addReg(SPU::R2)
-        .addReg(SPU::R1);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::Ar32), SPU::R1)
-        .addReg(SPU::R1)
-        .addReg(SPU::R2);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::SFIr32), SPU::R2)
-        .addReg(SPU::R2)
-        .addImm(16);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::LQXr128), SPU::R2)
-        .addReg(SPU::R2)
-        .addReg(SPU::R1);
-    } else {
-      report_fatal_error("Unhandled frame size: " + Twine(FrameSize));
-    }
-
-    if (hasDebugInfo) {
-      std::vector<MachineMove> &Moves = MMI.getFrameMoves();
-
-      // Show update of SP.
-      MachineLocation SPDst(MachineLocation::VirtualFP);
-      MachineLocation SPSrc(MachineLocation::VirtualFP, -FrameSize);
-      Moves.push_back(MachineMove(FrameLabel, SPDst, SPSrc));
-
-      // Add callee saved registers to move list.
-      const std::vector<CalleeSavedInfo> &CSI = MFI->getCalleeSavedInfo();
-      for (unsigned I = 0, E = CSI.size(); I != E; ++I) {
-        int Offset = MFI->getObjectOffset(CSI[I].getFrameIdx());
-        unsigned Reg = CSI[I].getReg();
-        if (Reg == SPU::R0) continue;
-        MachineLocation CSDst(MachineLocation::VirtualFP, Offset);
-        MachineLocation CSSrc(Reg);
-        Moves.push_back(MachineMove(FrameLabel, CSDst, CSSrc));
-      }
-
-      // Mark effective beginning of when frame pointer is ready.
-      MCSymbol *ReadyLabel = MMI.getContext().CreateTempSymbol();
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::PROLOG_LABEL)).addSym(ReadyLabel);
-
-      MachineLocation FPDst(SPU::R1);
-      MachineLocation FPSrc(MachineLocation::VirtualFP);
-      Moves.push_back(MachineMove(ReadyLabel, FPDst, FPSrc));
-    }
-  }
-}
-
-void SPUFrameLowering::emitEpilogue(MachineFunction &MF,
-                                MachineBasicBlock &MBB) const {
-  MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
-  const SPUInstrInfo &TII =
-    *static_cast<const SPUInstrInfo*>(MF.getTarget().getInstrInfo());
-  const MachineFrameInfo *MFI = MF.getFrameInfo();
-  int FrameSize = MFI->getStackSize();
-  int LinkSlotOffset = SPUFrameLowering::stackSlotSize();
-  DebugLoc dl = MBBI->getDebugLoc();
-
-  assert(MBBI->getOpcode() == SPU::RET &&
-         "Can only insert epilog into returning blocks");
-  assert((FrameSize & 0xf) == 0 && "FrameSize not aligned");
-
-  // the "empty" frame size is 16 - just the register scavenger spill slot
-  if (FrameSize > 16 || MFI->adjustsStack()) {
-    FrameSize = FrameSize + SPUFrameLowering::minStackSize();
-    if (isInt<10>(FrameSize + LinkSlotOffset)) {
-      // Reload $lr, adjust $sp by required amount
-      // Note: We do this to slightly improve dual issue -- not by much, but it
-      // is an opportunity for dual issue.
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::LQDr128), SPU::R0)
-        .addImm(FrameSize + LinkSlotOffset)
-        .addReg(SPU::R1);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::AIr32), SPU::R1)
-        .addReg(SPU::R1)
-        .addImm(FrameSize);
-    } else if (FrameSize <= (1 << 16) - 1 && FrameSize >= -(1 << 16)) {
-      // Frame size can be loaded into ILr32n, so temporarily spill $r2 and use
-      // $r2 to adjust $sp:
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::STQDr128), SPU::R2)
-        .addImm(16)
-        .addReg(SPU::R1);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::ILr32), SPU::R2)
-        .addImm(FrameSize);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::Ar32), SPU::R1)
-        .addReg(SPU::R1)
-        .addReg(SPU::R2);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::LQDr128), SPU::R0)
-        .addImm(16)
-        .addReg(SPU::R1);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::SFIr32), SPU::R2).
-        addReg(SPU::R2)
-        .addImm(16);
-      BuildMI(MBB, MBBI, dl, TII.get(SPU::LQXr128), SPU::R2)
-        .addReg(SPU::R2)
-        .addReg(SPU::R1);
-    } else {
-      report_fatal_error("Unhandled frame size: " + Twine(FrameSize));
-    }
-  }
-}
-
-void SPUFrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
-                                                        RegScavenger *RS) const{
-  // Mark LR and SP unused, since the prolog spills them to stack and
-  // we don't want anyone else to spill them for us.
-  //
-  // Also, unless R2 is really used someday, don't spill it automatically.
-  MF.getRegInfo().setPhysRegUnused(SPU::R0);
-  MF.getRegInfo().setPhysRegUnused(SPU::R1);
-  MF.getRegInfo().setPhysRegUnused(SPU::R2);
-
-  MachineFrameInfo *MFI = MF.getFrameInfo();
-  const TargetRegisterClass *RC = &SPU::R32CRegClass;
-  RS->setScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
-                                                     RC->getAlignment(),
-                                                     false));
-}
diff --git a/lib/Target/CellSPU/SPUFrameLowering.h b/lib/Target/CellSPU/SPUFrameLowering.h
deleted file mode 100644
index 11c52818dd9c..000000000000
--- a/lib/Target/CellSPU/SPUFrameLowering.h
+++ /dev/null
@@ -1,80 +0,0 @@
-//===-- SPUFrameLowering.h - SPU Frame Lowering stuff ----------*- C++ -*--===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file contains CellSPU frame information that doesn't fit anywhere else
-// cleanly...
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef SPU_FRAMEINFO_H
-#define SPU_FRAMEINFO_H
-
-#include "SPURegisterInfo.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetMachine.h"
-
-namespace llvm {
-  class SPUSubtarget;
-
-  class SPUFrameLowering: public TargetFrameLowering {
-    const SPUSubtarget &Subtarget;
-    std::pair<unsigned, int> LR[1];
-
-  public:
-    SPUFrameLowering(const SPUSubtarget &sti);
-
-    //! Determine the frame's layour
-    void determineFrameLayout(MachineFunction &MF) const;
-
-    /// emitProlog/emitEpilog - These methods insert prolog and epilog code into
-    /// the function.
-    void emitPrologue(MachineFunction &MF) const;
-    void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
-
-    //! Prediate: Target has dedicated frame pointer
-    bool hasFP(const MachineFunction &MF) const;
-
-    void processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
-                                              RegScavenger *RS = NULL) const;
-
-    //! Return a function's saved spill slots
-    /*!
-      For CellSPU, a function's saved spill slots is just the link register.
-     */
-    const std::pair<unsigned, int> *
-    getCalleeSaveSpillSlots(unsigned &NumEntries) const;
-
-    //! Stack slot size (16 bytes)
-    static int stackSlotSize() {
-      return 16;
-    }
-    //! Maximum frame offset representable by a signed 10-bit integer
-    /*!
-      This is the maximum frame offset that can be expressed as a 10-bit
-      integer, used in D-form addresses.
-     */
-    static int maxFrameOffset() {
-      return ((1 << 9) - 1) * stackSlotSize();
-    }
-    //! Minimum frame offset representable by a signed 10-bit integer
-    static int minFrameOffset() {
-      return -(1 << 9) * stackSlotSize();
-    }
-    //! Minimum frame size (enough to spill LR + SP)
-    static int minStackSize() {
-      return (2 * stackSlotSize());
-    }
-    //! Convert frame index to stack offset
-    static int FItoStackOffset(int frame_index) {
-      return frame_index * stackSlotSize();
-    }
-  };
-}
-
-#endif
diff --git a/lib/Target/CellSPU/SPUHazardRecognizers.cpp b/lib/Target/CellSPU/SPUHazardRecognizers.cpp
deleted file mode 100644
index 67a83f16a649..000000000000
--- a/lib/Target/CellSPU/SPUHazardRecognizers.cpp
+++ /dev/null
@@ -1,135 +0,0 @@
-//===-- SPUHazardRecognizers.cpp - Cell Hazard Recognizer Impls -----------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file implements hazard recognizers for scheduling on Cell SPU
-// processors.
-//
-//===----------------------------------------------------------------------===//
-
-#define DEBUG_TYPE "sched"
-
-#include "SPUHazardRecognizers.h"
-#include "SPU.h"
-#include "SPUInstrInfo.h"
-#include "llvm/CodeGen/ScheduleDAG.h"
-#include "llvm/CodeGen/SelectionDAGNodes.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/raw_ostream.h"
-using namespace llvm;
-
-//===----------------------------------------------------------------------===//
-// Cell SPU hazard recognizer
-//
-// This is the pipeline hazard recognizer for the Cell SPU processor. It does
-// very little right now.
-//===----------------------------------------------------------------------===//
-
-/// Return the pipeline hazard type encountered or generated by this
-/// instruction. Currently returns NoHazard.
-///
-/// \return NoHazard
-ScheduleHazardRecognizer::HazardType
-SPUHazardRecognizer::getHazardType(SUnit *SU, int Stalls)
-{
-  // Initial thoughts on how to do this, but this code cannot work unless the
-  // function's prolog and epilog code are also being scheduled so that we can
-  // accurately determine which pipeline is being scheduled.
-#if 0
-  assert(Stalls == 0 && "SPU hazards don't yet support scoreboard lookahead");
-
-  const SDNode *Node = SU->getNode()->getFlaggedMachineNode();
-  ScheduleHazardRecognizer::HazardType retval = NoHazard;
-  bool mustBeOdd = false;
-
-  switch (Node->getOpcode()) {
-  case SPU::LQDv16i8:
-  case SPU::LQDv8i16:
-  case SPU::LQDv4i32:
-  case SPU::LQDv4f32:
-  case SPU::LQDv2f64:
-  case SPU::LQDr128:
-  case SPU::LQDr64:
-  case SPU::LQDr32:
-  case SPU::LQDr16:
-  case SPU::LQAv16i8:
-  case SPU::LQAv8i16:
-  case SPU::LQAv4i32:
-  case SPU::LQAv4f32:
-  case SPU::LQAv2f64:
-  case SPU::LQAr128:
-  case SPU::LQAr64:
-  case SPU::LQAr32:
-  case SPU::LQXv4i32:
-  case SPU::LQXr128:
-  case SPU::LQXr64:
-  case SPU::LQXr32:
-  case SPU::LQXr16:
-  case SPU::STQDv16i8:
-  case SPU::STQDv8i16:
-  case SPU::STQDv4i32:
-  case SPU::STQDv4f32:
-  case SPU::STQDv2f64:
-  case SPU::STQDr128:
-  case SPU::STQDr64:
-  case SPU::STQDr32:
-  case SPU::STQDr16:
-  case SPU::STQDr8:
-  case SPU::STQAv16i8:
-  case SPU::STQAv8i16:
-  case SPU::STQAv4i32:
-  case SPU::STQAv4f32:
-  case SPU::STQAv2f64:
-  case SPU::STQAr128:
-  case SPU::STQAr64:
-  case SPU::STQAr32:
-  case SPU::STQAr16:
-  case SPU::STQAr8:
-  case SPU::STQXv16i8:
-  case SPU::STQXv8i16:
-  case SPU::STQXv4i32:
-  case SPU::STQXv4f32:
-  case SPU::STQXv2f64:
-  case SPU::STQXr128:
-  case SPU::STQXr64:
-  case SPU::STQXr32:
-  case SPU::STQXr16:
-  case SPU::STQXr8:
-  case SPU::RET:
-    mustBeOdd = true;
-    break;
-  default:
-    // Assume that this instruction can be on the even pipe
-    break;
-  }
-
-  if (mustBeOdd && !EvenOdd)
-    retval = Hazard;
-
-  DEBUG(errs() << "SPUHazardRecognizer EvenOdd " << EvenOdd << " Hazard "
-               << retval << "\n");
-  EvenOdd ^= 1;
-  return retval;
-#else
-  return NoHazard;
-#endif
-}
-
-void SPUHazardRecognizer::EmitInstruction(SUnit *SU)
-{
-}
-
-void SPUHazardRecognizer::AdvanceCycle()
-{
-  DEBUG(errs() << "SPUHazardRecognizer::AdvanceCycle\n");
-}
-
-void SPUHazardRecognizer::EmitNoop()
-{
-  AdvanceCycle();
-}
diff --git a/lib/Target/CellSPU/SPUHazardRecognizers.h b/lib/Target/CellSPU/SPUHazardRecognizers.h
deleted file mode 100644
index 30acaeaa36fb..000000000000
--- a/lib/Target/CellSPU/SPUHazardRecognizers.h
+++ /dev/null
@@ -1,37 +0,0 @@
-//===-- SPUHazardRecognizers.h - Cell SPU Hazard Recognizer -----*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file defines hazard recognizers for scheduling on the Cell SPU
-// processor.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef SPUHAZRECS_H
-#define SPUHAZRECS_H
-
-#include "llvm/CodeGen/ScheduleHazardRecognizer.h"
-
-namespace llvm {
-
-class TargetInstrInfo;
-
-/// SPUHazardRecognizer
-class SPUHazardRecognizer : public ScheduleHazardRecognizer
-{
-public:
-  SPUHazardRecognizer(const TargetInstrInfo &/*TII*/) {}
-  virtual HazardType getHazardType(SUnit *SU, int Stalls);
-  virtual void EmitInstruction(SUnit *SU);
-  virtual void AdvanceCycle();
-  virtual void EmitNoop();
-};
-
-} // end namespace llvm
-
-#endif
diff --git a/lib/Target/CellSPU/SPUISelDAGToDAG.cpp b/lib/Target/CellSPU/SPUISelDAGToDAG.cpp
deleted file mode 100644
index 5d5061054b08..000000000000
--- a/lib/Target/CellSPU/SPUISelDAGToDAG.cpp
+++ /dev/null
@@ -1,1192 +0,0 @@
-//===-- SPUISelDAGToDAG.cpp - CellSPU pattern matching inst selector ------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file defines a pattern matching instruction selector for the Cell SPU,
-// converting from a legalized dag to a SPU-target dag.
-//
-//===----------------------------------------------------------------------===//
-
-#include "SPU.h"
-#include "SPUTargetMachine.h"
-#include "SPUHazardRecognizers.h"
-#include "SPUFrameLowering.h"
-#include "SPUTargetMachine.h"
-#include "llvm/CodeGen/MachineConstantPool.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunction.h"
-#include "llvm/CodeGen/SelectionDAG.h"
-#include "llvm/CodeGen/SelectionDAGISel.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/ADT/Statistic.h"
-#include "llvm/Constants.h"
-#include "llvm/GlobalValue.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/LLVMContext.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/MathExtras.h"
-#include "llvm/Support/Compiler.h"
-#include "llvm/Support/raw_ostream.h"
-
-using namespace llvm;
-
-namespace {
-  //! ConstantSDNode predicate for i32 sign-extended, 10-bit immediates
-  bool
-  isI32IntS10Immediate(ConstantSDNode *CN)
-  {
-    return isInt<10>(CN->getSExtValue());
-  }
-
-  //! ConstantSDNode predicate for i32 unsigned 10-bit immediate values
-  bool
-  isI32IntU10Immediate(ConstantSDNode *CN)
-  {
-    return isUInt<10>(CN->getSExtValue());
-  }
-
-  //! ConstantSDNode predicate for i16 sign-extended, 10-bit immediate values
-  bool
-  isI16IntS10Immediate(ConstantSDNode *CN)
-  {
-    return isInt<10>(CN->getSExtValue());
-  }
-
-  //! ConstantSDNode predicate for i16 unsigned 10-bit immediate values
-  bool
-  isI16IntU10Immediate(ConstantSDNode *CN)
-  {
-    return isUInt<10>((short) CN->getZExtValue());
-  }
-
-  //! ConstantSDNode predicate for signed 16-bit values
-  /*!
-    \param CN The constant SelectionDAG node holding the value
-    \param Imm The returned 16-bit value, if returning true
-
-    This predicate tests the value in \a CN to see whether it can be
-    represented as a 16-bit, sign-extended quantity. Returns true if
-    this is the case.
-   */
-  bool
-  isIntS16Immediate(ConstantSDNode *CN, short &Imm)
-  {
-    EVT vt = CN->getValueType(0);
-    Imm = (short) CN->getZExtValue();
-    if (vt.getSimpleVT() >= MVT::i1 && vt.getSimpleVT() <= MVT::i16) {
-      return true;
-    } else if (vt == MVT::i32) {
-      int32_t i_val = (int32_t) CN->getZExtValue();
-      return i_val == SignExtend32<16>(i_val);
-    } else {
-      int64_t i_val = (int64_t) CN->getZExtValue();
-      return i_val == SignExtend64<16>(i_val);
-    }
-  }
-
-  //! ConstantFPSDNode predicate for representing floats as 16-bit sign ext.
-  static bool
-  isFPS16Immediate(ConstantFPSDNode *FPN, short &Imm)
-  {
-    EVT vt = FPN->getValueType(0);
-    if (vt == MVT::f32) {
-      int val = FloatToBits(FPN->getValueAPF().convertToFloat());
-      if (val == SignExtend32<16>(val)) {
-        Imm = (short) val;
-        return true;
-      }
-    }
-
-    return false;
-  }
-
-  //! Generate the carry-generate shuffle mask.
-  SDValue getCarryGenerateShufMask(SelectionDAG &DAG, DebugLoc dl) {
-    SmallVector<SDValue, 16 > ShufBytes;
-
-    // Create the shuffle mask for "rotating" the borrow up one register slot
-    // once the borrow is generated.
-    ShufBytes.push_back(DAG.getConstant(0x04050607, MVT::i32));
-    ShufBytes.push_back(DAG.getConstant(0x80808080, MVT::i32));
-    ShufBytes.push_back(DAG.getConstant(0x0c0d0e0f, MVT::i32));
-    ShufBytes.push_back(DAG.getConstant(0x80808080, MVT::i32));
-
-    return DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                       &ShufBytes[0], ShufBytes.size());
-  }
-
-  //! Generate the borrow-generate shuffle mask
-  SDValue getBorrowGenerateShufMask(SelectionDAG &DAG, DebugLoc dl) {
-    SmallVector<SDValue, 16 > ShufBytes;
-
-    // Create the shuffle mask for "rotating" the borrow up one register slot
-    // once the borrow is generated.
-    ShufBytes.push_back(DAG.getConstant(0x04050607, MVT::i32));
-    ShufBytes.push_back(DAG.getConstant(0xc0c0c0c0, MVT::i32));
-    ShufBytes.push_back(DAG.getConstant(0x0c0d0e0f, MVT::i32));
-    ShufBytes.push_back(DAG.getConstant(0xc0c0c0c0, MVT::i32));
-
-    return DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                       &ShufBytes[0], ShufBytes.size());
-  }
-
-  //===------------------------------------------------------------------===//
-  /// SPUDAGToDAGISel - Cell SPU-specific code to select SPU machine
-  /// instructions for SelectionDAG operations.
-  ///
-  class SPUDAGToDAGISel :
-    public SelectionDAGISel
-  {
-    const SPUTargetMachine &TM;
-    const SPUTargetLowering &SPUtli;
-    unsigned GlobalBaseReg;
-
-  public:
-    explicit SPUDAGToDAGISel(SPUTargetMachine &tm) :
-      SelectionDAGISel(tm),
-      TM(tm),
-      SPUtli(*tm.getTargetLowering())
-    { }
-
-    virtual bool runOnMachineFunction(MachineFunction &MF) {
-      // Make sure we re-emit a set of the global base reg if necessary
-      GlobalBaseReg = 0;
-      SelectionDAGISel::runOnMachineFunction(MF);
-      return true;
-    }
-
-    /// getI32Imm - Return a target constant with the specified value, of type
-    /// i32.
-    inline SDValue getI32Imm(uint32_t Imm) {
-      return CurDAG->getTargetConstant(Imm, MVT::i32);
-    }
-
-    /// getSmallIPtrImm - Return a target constant of pointer type.
-    inline SDValue getSmallIPtrImm(unsigned Imm) {
-      return CurDAG->getTargetConstant(Imm, SPUtli.getPointerTy());
-    }
-
-    SDNode *emitBuildVector(SDNode *bvNode) {
-      EVT vecVT = bvNode->getValueType(0);
-      DebugLoc dl = bvNode->getDebugLoc();
-
-      // Check to see if this vector can be represented as a CellSPU immediate
-      // constant by invoking all of the instruction selection predicates:
-      if (((vecVT == MVT::v8i16) &&
-           (SPU::get_vec_i16imm(bvNode, *CurDAG, MVT::i16).getNode() != 0)) ||
-          ((vecVT == MVT::v4i32) &&
-           ((SPU::get_vec_i16imm(bvNode, *CurDAG, MVT::i32).getNode() != 0) ||
-            (SPU::get_ILHUvec_imm(bvNode, *CurDAG, MVT::i32).getNode() != 0) ||
-            (SPU::get_vec_u18imm(bvNode, *CurDAG, MVT::i32).getNode() != 0) ||
-            (SPU::get_v4i32_imm(bvNode, *CurDAG).getNode() != 0))) ||
-          ((vecVT == MVT::v2i64) &&
-           ((SPU::get_vec_i16imm(bvNode, *CurDAG, MVT::i64).getNode() != 0) ||
-            (SPU::get_ILHUvec_imm(bvNode, *CurDAG, MVT::i64).getNode() != 0) ||
-            (SPU::get_vec_u18imm(bvNode, *CurDAG, MVT::i64).getNode() != 0)))) {
-        HandleSDNode Dummy(SDValue(bvNode, 0));
-        if (SDNode *N = Select(bvNode))
-          return N;
-        return Dummy.getValue().getNode();
-      }
-
-      // No, need to emit a constant pool spill:
-      std::vector<Constant*> CV;
-
-      for (size_t i = 0; i < bvNode->getNumOperands(); ++i) {
-        ConstantSDNode *V = cast<ConstantSDNode > (bvNode->getOperand(i));
-        CV.push_back(const_cast<ConstantInt *>(V->getConstantIntValue()));
-      }
-
-      const Constant *CP = ConstantVector::get(CV);
-      SDValue CPIdx = CurDAG->getConstantPool(CP, SPUtli.getPointerTy());
-      unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
-      SDValue CGPoolOffset =
-              SPU::LowerConstantPool(CPIdx, *CurDAG, TM);
-
-      HandleSDNode Dummy(CurDAG->getLoad(vecVT, dl,
-                                         CurDAG->getEntryNode(), CGPoolOffset,
-                                         MachinePointerInfo::getConstantPool(),
-                                         false, false, false, Alignment));
-      CurDAG->ReplaceAllUsesWith(SDValue(bvNode, 0), Dummy.getValue());
-      if (SDNode *N = SelectCode(Dummy.getValue().getNode()))
-        return N;
-      return Dummy.getValue().getNode();
-    }
-
-    /// Select - Convert the specified operand from a target-independent to a
-    /// target-specific node if it hasn't already been changed.
-    SDNode *Select(SDNode *N);
-
-    //! Emit the instruction sequence for i64 shl
-    SDNode *SelectSHLi64(SDNode *N, EVT OpVT);
-
-    //! Emit the instruction sequence for i64 srl
-    SDNode *SelectSRLi64(SDNode *N, EVT OpVT);
-
-    //! Emit the instruction sequence for i64 sra
-    SDNode *SelectSRAi64(SDNode *N, EVT OpVT);
-
-    //! Emit the necessary sequence for loading i64 constants:
-    SDNode *SelectI64Constant(SDNode *N, EVT OpVT, DebugLoc dl);
-
-    //! Alternate instruction emit sequence for loading i64 constants
-    SDNode *SelectI64Constant(uint64_t i64const, EVT OpVT, DebugLoc dl);
-
-    //! Returns true if the address N is an A-form (local store) address
-    bool SelectAFormAddr(SDNode *Op, SDValue N, SDValue &Base,
-                         SDValue &Index);
-
-    //! D-form address predicate
-    bool SelectDFormAddr(SDNode *Op, SDValue N, SDValue &Base,
-                         SDValue &Index);
-
-    /// Alternate D-form address using i7 offset predicate
-    bool SelectDForm2Addr(SDNode *Op, SDValue N, SDValue &Disp,
-                          SDValue &Base);
-
-    /// D-form address selection workhorse
-    bool DFormAddressPredicate(SDNode *Op, SDValue N, SDValue &Disp,
-                               SDValue &Base, int minOffset, int maxOffset);
-
-    //! Address predicate if N can be expressed as an indexed [r+r] operation.
-    bool SelectXFormAddr(SDNode *Op, SDValue N, SDValue &Base,
-                         SDValue &Index);
-
-    /// SelectInlineAsmMemoryOperand - Implement addressing mode selection for
-    /// inline asm expressions.
-    virtual bool SelectInlineAsmMemoryOperand(const SDValue &Op,
-                                              char ConstraintCode,
-                                              std::vector<SDValue> &OutOps) {
-      SDValue Op0, Op1;
-      switch (ConstraintCode) {
-      default: return true;
-      case 'm':   // memory
-        if (!SelectDFormAddr(Op.getNode(), Op, Op0, Op1)
-            && !SelectAFormAddr(Op.getNode(), Op, Op0, Op1))
-          SelectXFormAddr(Op.getNode(), Op, Op0, Op1);
-        break;
-      case 'o':   // offsetable
-        if (!SelectDFormAddr(Op.getNode(), Op, Op0, Op1)
-            && !SelectAFormAddr(Op.getNode(), Op, Op0, Op1)) {
-          Op0 = Op;
-          Op1 = getSmallIPtrImm(0);
-        }
-        break;
-      case 'v':   // not offsetable
-#if 1
-        llvm_unreachable("InlineAsmMemoryOperand 'v' constraint not handled.");
-#else
-        SelectAddrIdxOnly(Op, Op, Op0, Op1);
-        break;
-#endif
-      }
-
-      OutOps.push_back(Op0);
-      OutOps.push_back(Op1);
-      return false;
-    }
-
-    virtual const char *getPassName() const {
-      return "Cell SPU DAG->DAG Pattern Instruction Selection";
-    }
-
-  private:
-    SDValue getRC( MVT );
-
-    // Include the pieces autogenerated from the target description.
-#include "SPUGenDAGISel.inc"
-  };
-}
-
-/*!
- \param Op The ISD instruction operand
- \param N The address to be tested
- \param Base The base address
- \param Index The base address index
- */
-bool
-SPUDAGToDAGISel::SelectAFormAddr(SDNode *Op, SDValue N, SDValue &Base,
-                    SDValue &Index) {
-  // These match the addr256k operand type:
-  EVT OffsVT = MVT::i16;
-  SDValue Zero = CurDAG->getTargetConstant(0, OffsVT);
-  int64_t val;
-
-  switch (N.getOpcode()) {
-  case ISD::Constant:
-    val = dyn_cast<ConstantSDNode>(N.getNode())->getSExtValue();
-    Base = CurDAG->getTargetConstant( val , MVT::i32);
-    Index = Zero;
-    return true;
-  case ISD::ConstantPool:
-  case ISD::GlobalAddress:
-    report_fatal_error("SPU SelectAFormAddr: Pool/Global not lowered.");
-    /*NOTREACHED*/
-
-  case ISD::TargetConstant:
-  case ISD::TargetGlobalAddress:
-  case ISD::TargetJumpTable:
-    report_fatal_error("SPUSelectAFormAddr: Target Constant/Pool/Global "
-                      "not wrapped as A-form address.");
-    /*NOTREACHED*/
-
-  case SPUISD::AFormAddr:
-    // Just load from memory if there's only a single use of the location,
-    // otherwise, this will get handled below with D-form offset addresses
-    if (N.hasOneUse()) {
-      SDValue Op0 = N.getOperand(0);
-      switch (Op0.getOpcode()) {
-      case ISD::TargetConstantPool:
-      case ISD::TargetJumpTable:
-        Base = Op0;
-        Index = Zero;
-        return true;
-
-      case ISD::TargetGlobalAddress: {
-        GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op0);
-        const GlobalValue *GV = GSDN->getGlobal();
-        if (GV->getAlignment() == 16) {
-          Base = Op0;
-          Index = Zero;
-          return true;
-        }
-        break;
-      }
-      }
-    }
-    break;
-  }
-  return false;
-}
-
-bool
-SPUDAGToDAGISel::SelectDForm2Addr(SDNode *Op, SDValue N, SDValue &Disp,
-                                  SDValue &Base) {
-  const int minDForm2Offset = -(1 << 7);
-  const int maxDForm2Offset = (1 << 7) - 1;
-  return DFormAddressPredicate(Op, N, Disp, Base, minDForm2Offset,
-                               maxDForm2Offset);
-}
-
-/*!
-  \param Op The ISD instruction (ignored)
-  \param N The address to be tested
-  \param Base Base address register/pointer
-  \param Index Base address index
-
-  Examine the input address by a base register plus a signed 10-bit
-  displacement, [r+I10] (D-form address).
-
-  \return true if \a N is a D-form address with \a Base and \a Index set
-  to non-empty SDValue instances.
-*/
-bool
-SPUDAGToDAGISel::SelectDFormAddr(SDNode *Op, SDValue N, SDValue &Base,
-                                 SDValue &Index) {
-  return DFormAddressPredicate(Op, N, Base, Index,
-                               SPUFrameLowering::minFrameOffset(),
-                               SPUFrameLowering::maxFrameOffset());
-}
-
-bool
-SPUDAGToDAGISel::DFormAddressPredicate(SDNode *Op, SDValue N, SDValue &Base,
-                                      SDValue &Index, int minOffset,
-                                      int maxOffset) {
-  unsigned Opc = N.getOpcode();
-  EVT PtrTy = SPUtli.getPointerTy();
-
-  if (Opc == ISD::FrameIndex) {
-    // Stack frame index must be less than 512 (divided by 16):
-    FrameIndexSDNode *FIN = cast<FrameIndexSDNode>(N);
-    int FI = int(FIN->getIndex());
-    DEBUG(errs() << "SelectDFormAddr: ISD::FrameIndex = "
-               << FI << "\n");
-    if (SPUFrameLowering::FItoStackOffset(FI) < maxOffset) {
-      Base = CurDAG->getTargetConstant(0, PtrTy);
-      Index = CurDAG->getTargetFrameIndex(FI, PtrTy);
-      return true;
-    }
-  } else if (Opc == ISD::ADD) {
-    // Generated by getelementptr
-    const SDValue Op0 = N.getOperand(0);
-    const SDValue Op1 = N.getOperand(1);
-
-    if ((Op0.getOpcode() == SPUISD::Hi && Op1.getOpcode() == SPUISD::Lo)
-        || (Op1.getOpcode() == SPUISD::Hi && Op0.getOpcode() == SPUISD::Lo)) {
-      Base = CurDAG->getTargetConstant(0, PtrTy);
-      Index = N;
-      return true;
-    } else if (Op1.getOpcode() == ISD::Constant
-               || Op1.getOpcode() == ISD::TargetConstant) {
-      ConstantSDNode *CN = cast<ConstantSDNode>(Op1);
-      int32_t offset = int32_t(CN->getSExtValue());
-
-      if (Op0.getOpcode() == ISD::FrameIndex) {
-        FrameIndexSDNode *FIN = cast<FrameIndexSDNode>(Op0);
-        int FI = int(FIN->getIndex());
-        DEBUG(errs() << "SelectDFormAddr: ISD::ADD offset = " << offset
-                   << " frame index = " << FI << "\n");
-
-        if (SPUFrameLowering::FItoStackOffset(FI) < maxOffset) {
-          Base = CurDAG->getTargetConstant(offset, PtrTy);
-          Index = CurDAG->getTargetFrameIndex(FI, PtrTy);
-          return true;
-        }
-      } else if (offset > minOffset && offset < maxOffset) {
-        Base = CurDAG->getTargetConstant(offset, PtrTy);
-        Index = Op0;
-        return true;
-      }
-    } else if (Op0.getOpcode() == ISD::Constant
-               || Op0.getOpcode() == ISD::TargetConstant) {
-      ConstantSDNode *CN = cast<ConstantSDNode>(Op0);
-      int32_t offset = int32_t(CN->getSExtValue());
-
-      if (Op1.getOpcode() == ISD::FrameIndex) {
-        FrameIndexSDNode *FIN = cast<FrameIndexSDNode>(Op1);
-        int FI = int(FIN->getIndex());
-        DEBUG(errs() << "SelectDFormAddr: ISD::ADD offset = " << offset
-                   << " frame index = " << FI << "\n");
-
-        if (SPUFrameLowering::FItoStackOffset(FI) < maxOffset) {
-          Base = CurDAG->getTargetConstant(offset, PtrTy);
-          Index = CurDAG->getTargetFrameIndex(FI, PtrTy);
-          return true;
-        }
-      } else if (offset > minOffset && offset < maxOffset) {
-        Base = CurDAG->getTargetConstant(offset, PtrTy);
-        Index = Op1;
-        return true;
-      }
-    }
-  } else if (Opc == SPUISD::IndirectAddr) {
-    // Indirect with constant offset -> D-Form address
-    const SDValue Op0 = N.getOperand(0);
-    const SDValue Op1 = N.getOperand(1);
-
-    if (Op0.getOpcode() == SPUISD::Hi
-        && Op1.getOpcode() == SPUISD::Lo) {
-      // (SPUindirect (SPUhi <arg>, 0), (SPUlo <arg>, 0))
-      Base = CurDAG->getTargetConstant(0, PtrTy);
-      Index = N;
-      return true;
-    } else if (isa<ConstantSDNode>(Op0) || isa<ConstantSDNode>(Op1)) {
-      int32_t offset = 0;
-      SDValue idxOp;
-
-      if (isa<ConstantSDNode>(Op1)) {
-        ConstantSDNode *CN = cast<ConstantSDNode>(Op1);
-        offset = int32_t(CN->getSExtValue());
-        idxOp = Op0;
-      } else if (isa<ConstantSDNode>(Op0)) {
-        ConstantSDNode *CN = cast<ConstantSDNode>(Op0);
-        offset = int32_t(CN->getSExtValue());
-        idxOp = Op1;
-      }
-
-      if (offset >= minOffset && offset <= maxOffset) {
-        Base = CurDAG->getTargetConstant(offset, PtrTy);
-        Index = idxOp;
-        return true;
-      }
-    }
-  } else if (Opc == SPUISD::AFormAddr) {
-    Base = CurDAG->getTargetConstant(0, N.getValueType());
-    Index = N;
-    return true;
-  } else if (Opc == SPUISD::LDRESULT) {
-    Base = CurDAG->getTargetConstant(0, N.getValueType());
-    Index = N;
-    return true;
-  } else if (Opc == ISD::Register
-           ||Opc == ISD::CopyFromReg
-           ||Opc == ISD::UNDEF
-           ||Opc == ISD::Constant) {
-    unsigned OpOpc = Op->getOpcode();
-
-    if (OpOpc == ISD::STORE || OpOpc == ISD::LOAD) {
-      // Direct load/store without getelementptr
-      SDValue Offs;
-
-      Offs = ((OpOpc == ISD::STORE) ? Op->getOperand(3) : Op->getOperand(2));
-
-      if (Offs.getOpcode() == ISD::Constant || Offs.getOpcode() == ISD::UNDEF) {
-        if (Offs.getOpcode() == ISD::UNDEF)
-          Offs = CurDAG->getTargetConstant(0, Offs.getValueType());
-
-        Base = Offs;
-        Index = N;
-        return true;
-      }
-    } else {
-      /* If otherwise unadorned, default to D-form address with 0 offset: */
-      if (Opc == ISD::CopyFromReg) {
-        Index = N.getOperand(1);
-      } else {
-        Index = N;
-      }
-
-      Base = CurDAG->getTargetConstant(0, Index.getValueType());
-      return true;
-    }
-  }
-
-  return false;
-}
-
-/*!
-  \param Op The ISD instruction operand
-  \param N The address operand
-  \param Base The base pointer operand
-  \param Index The offset/index operand
-
-  If the address \a N can be expressed as an A-form or D-form address, returns
-  false.  Otherwise, creates two operands, Base and Index that will become the
-  (r)(r) X-form address.
-*/
-bool
-SPUDAGToDAGISel::SelectXFormAddr(SDNode *Op, SDValue N, SDValue &Base,
-                                 SDValue &Index) {
-  if (!SelectAFormAddr(Op, N, Base, Index)
-      && !SelectDFormAddr(Op, N, Base, Index)) {
-    // If the address is neither A-form or D-form, punt and use an X-form
-    // address:
-    Base = N.getOperand(1);
-    Index = N.getOperand(0);
-    return true;
-  }
-
-  return false;
-}
-
-/*!
- Utility function to use with COPY_TO_REGCLASS instructions. Returns a SDValue
- to be used as the last parameter of a
-CurDAG->getMachineNode(COPY_TO_REGCLASS,..., ) function call
- \param VT the value type for which we want a register class
-*/
-SDValue SPUDAGToDAGISel::getRC( MVT VT ) {
-  switch( VT.SimpleTy ) {
-  case MVT::i8:
-    return CurDAG->getTargetConstant(SPU::R8CRegClass.getID(), MVT::i32);
-  case MVT::i16:
-    return CurDAG->getTargetConstant(SPU::R16CRegClass.getID(), MVT::i32);
-  case MVT::i32:
-    return CurDAG->getTargetConstant(SPU::R32CRegClass.getID(), MVT::i32);
-  case MVT::f32:
-    return CurDAG->getTargetConstant(SPU::R32FPRegClass.getID(), MVT::i32);
-  case MVT::i64:
-    return CurDAG->getTargetConstant(SPU::R64CRegClass.getID(), MVT::i32);
-  case MVT::i128:
-    return CurDAG->getTargetConstant(SPU::GPRCRegClass.getID(), MVT::i32);
-  case MVT::v16i8:
-  case MVT::v8i16:
-  case MVT::v4i32:
-  case MVT::v4f32:
-  case MVT::v2i64:
-  case MVT::v2f64:
-    return CurDAG->getTargetConstant(SPU::VECREGRegClass.getID(), MVT::i32);
-  default:
-    assert( false && "add a new case here" );
-    return SDValue();
-  }
-}
-
-//! Convert the operand from a target-independent to a target-specific node
-/*!
- */
-SDNode *
-SPUDAGToDAGISel::Select(SDNode *N) {
-  unsigned Opc = N->getOpcode();
-  int n_ops = -1;
-  unsigned NewOpc = 0;
-  EVT OpVT = N->getValueType(0);
-  SDValue Ops[8];
-  DebugLoc dl = N->getDebugLoc();
-
-  if (N->isMachineOpcode())
-    return NULL;   // Already selected.
-
-  if (Opc == ISD::FrameIndex) {
-    int FI = cast<FrameIndexSDNode>(N)->getIndex();
-    SDValue TFI = CurDAG->getTargetFrameIndex(FI, N->getValueType(0));
-    SDValue Imm0 = CurDAG->getTargetConstant(0, N->getValueType(0));
-
-    if (FI < 128) {
-      NewOpc = SPU::AIr32;
-      Ops[0] = TFI;
-      Ops[1] = Imm0;
-      n_ops = 2;
-    } else {
-      NewOpc = SPU::Ar32;
-      Ops[0] = CurDAG->getRegister(SPU::R1, N->getValueType(0));
-      Ops[1] = SDValue(CurDAG->getMachineNode(SPU::ILAr32, dl,
-                                              N->getValueType(0), TFI),
-                       0);
-      n_ops = 2;
-    }
-  } else if (Opc == ISD::Constant && OpVT == MVT::i64) {
-    // Catch the i64 constants that end up here. Note: The backend doesn't
-    // attempt to legalize the constant (it's useless because DAGCombiner
-    // will insert 64-bit constants and we can't stop it).
-    return SelectI64Constant(N, OpVT, N->getDebugLoc());
-  } else if ((Opc == ISD::ZERO_EXTEND || Opc == ISD::ANY_EXTEND)
-             && OpVT == MVT::i64) {
-    SDValue Op0 = N->getOperand(0);
-    EVT Op0VT = Op0.getValueType();
-    EVT Op0VecVT = EVT::getVectorVT(*CurDAG->getContext(),
-                                    Op0VT, (128 / Op0VT.getSizeInBits()));
-    EVT OpVecVT = EVT::getVectorVT(*CurDAG->getContext(),
-                                   OpVT, (128 / OpVT.getSizeInBits()));
-    SDValue shufMask;
-
-    switch (Op0VT.getSimpleVT().SimpleTy) {
-    default:
-      report_fatal_error("CellSPU Select: Unhandled zero/any extend EVT");
-      /*NOTREACHED*/
-    case MVT::i32:
-      shufMask = CurDAG->getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                                 CurDAG->getConstant(0x80808080, MVT::i32),
-                                 CurDAG->getConstant(0x00010203, MVT::i32),
-                                 CurDAG->getConstant(0x80808080, MVT::i32),
-                                 CurDAG->getConstant(0x08090a0b, MVT::i32));
-      break;
-
-    case MVT::i16:
-      shufMask = CurDAG->getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                                 CurDAG->getConstant(0x80808080, MVT::i32),
-                                 CurDAG->getConstant(0x80800203, MVT::i32),
-                                 CurDAG->getConstant(0x80808080, MVT::i32),
-                                 CurDAG->getConstant(0x80800a0b, MVT::i32));
-      break;
-
-    case MVT::i8:
-      shufMask = CurDAG->getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                                 CurDAG->getConstant(0x80808080, MVT::i32),
-                                 CurDAG->getConstant(0x80808003, MVT::i32),
-                                 CurDAG->getConstant(0x80808080, MVT::i32),
-                                 CurDAG->getConstant(0x8080800b, MVT::i32));
-      break;
-    }
-
-    SDNode *shufMaskLoad = emitBuildVector(shufMask.getNode());
-
-    HandleSDNode PromoteScalar(CurDAG->getNode(SPUISD::PREFSLOT2VEC, dl,
-                                               Op0VecVT, Op0));
-
-    SDValue PromScalar;
-    if (SDNode *N = SelectCode(PromoteScalar.getValue().getNode()))
-      PromScalar = SDValue(N, 0);
-    else
-      PromScalar = PromoteScalar.getValue();
-
-    SDValue zextShuffle =
-            CurDAG->getNode(SPUISD::SHUFB, dl, OpVecVT,
-                            PromScalar, PromScalar,
-                            SDValue(shufMaskLoad, 0));
-
-    HandleSDNode Dummy2(zextShuffle);
-    if (SDNode *N = SelectCode(Dummy2.getValue().getNode()))
-      zextShuffle = SDValue(N, 0);
-    else
-      zextShuffle = Dummy2.getValue();
-    HandleSDNode Dummy(CurDAG->getNode(SPUISD::VEC2PREFSLOT, dl, OpVT,
-                                       zextShuffle));
-
-    CurDAG->ReplaceAllUsesWith(N, Dummy.getValue().getNode());
-    SelectCode(Dummy.getValue().getNode());
-    return Dummy.getValue().getNode();
-  } else if (Opc == ISD::ADD && (OpVT == MVT::i64 || OpVT == MVT::v2i64)) {
-    SDNode *CGLoad =
-            emitBuildVector(getCarryGenerateShufMask(*CurDAG, dl).getNode());
-
-    HandleSDNode Dummy(CurDAG->getNode(SPUISD::ADD64_MARKER, dl, OpVT,
-                                       N->getOperand(0), N->getOperand(1),
-                                       SDValue(CGLoad, 0)));
-
-    CurDAG->ReplaceAllUsesWith(N, Dummy.getValue().getNode());
-    if (SDNode *N = SelectCode(Dummy.getValue().getNode()))
-      return N;
-    return Dummy.getValue().getNode();
-  } else if (Opc == ISD::SUB && (OpVT == MVT::i64 || OpVT == MVT::v2i64)) {
-    SDNode *CGLoad =
-            emitBuildVector(getBorrowGenerateShufMask(*CurDAG, dl).getNode());
-
-    HandleSDNode Dummy(CurDAG->getNode(SPUISD::SUB64_MARKER, dl, OpVT,
-                                       N->getOperand(0), N->getOperand(1),
-                                       SDValue(CGLoad, 0)));
-
-    CurDAG->ReplaceAllUsesWith(N, Dummy.getValue().getNode());
-    if (SDNode *N = SelectCode(Dummy.getValue().getNode()))
-      return N;
-    return Dummy.getValue().getNode();
-  } else if (Opc == ISD::MUL && (OpVT == MVT::i64 || OpVT == MVT::v2i64)) {
-    SDNode *CGLoad =
-            emitBuildVector(getCarryGenerateShufMask(*CurDAG, dl).getNode());
-
-    HandleSDNode Dummy(CurDAG->getNode(SPUISD::MUL64_MARKER, dl, OpVT,
-                                       N->getOperand(0), N->getOperand(1),
-                                       SDValue(CGLoad, 0)));
-    CurDAG->ReplaceAllUsesWith(N, Dummy.getValue().getNode());
-    if (SDNode *N = SelectCode(Dummy.getValue().getNode()))
-      return N;
-    return Dummy.getValue().getNode();
-  } else if (Opc == ISD::TRUNCATE) {
-    SDValue Op0 = N->getOperand(0);
-    if ((Op0.getOpcode() == ISD::SRA || Op0.getOpcode() == ISD::SRL)
-        && OpVT == MVT::i32
-        && Op0.getValueType() == MVT::i64) {
-      // Catch (truncate:i32 ([sra|srl]:i64 arg, c), where c >= 32
-      //
-      // Take advantage of the fact that the upper 32 bits are in the
-      // i32 preferred slot and avoid shuffle gymnastics:
-      ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Op0.getOperand(1));
-      if (CN != 0) {
-        unsigned shift_amt = unsigned(CN->getZExtValue());
-
-        if (shift_amt >= 32) {
-          SDNode *hi32 =
-                  CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl, OpVT,
-                                         Op0.getOperand(0), getRC(MVT::i32));
-
-          shift_amt -= 32;
-          if (shift_amt > 0) {
-            // Take care of the additional shift, if present:
-            SDValue shift = CurDAG->getTargetConstant(shift_amt, MVT::i32);
-            unsigned Opc = SPU::ROTMAIr32_i32;
-
-            if (Op0.getOpcode() == ISD::SRL)
-              Opc = SPU::ROTMr32;
-
-            hi32 = CurDAG->getMachineNode(Opc, dl, OpVT, SDValue(hi32, 0),
-                                          shift);
-          }
-
-          return hi32;
-        }
-      }
-    }
-  } else if (Opc == ISD::SHL) {
-    if (OpVT == MVT::i64)
-      return SelectSHLi64(N, OpVT);
-  } else if (Opc == ISD::SRL) {
-    if (OpVT == MVT::i64)
-      return SelectSRLi64(N, OpVT);
-  } else if (Opc == ISD::SRA) {
-    if (OpVT == MVT::i64)
-      return SelectSRAi64(N, OpVT);
-  } else if (Opc == ISD::FNEG
-             && (OpVT == MVT::f64 || OpVT == MVT::v2f64)) {
-    DebugLoc dl = N->getDebugLoc();
-    // Check if the pattern is a special form of DFNMS:
-    // (fneg (fsub (fmul R64FP:$rA, R64FP:$rB), R64FP:$rC))
-    SDValue Op0 = N->getOperand(0);
-    if (Op0.getOpcode() == ISD::FSUB) {
-      SDValue Op00 = Op0.getOperand(0);
-      if (Op00.getOpcode() == ISD::FMUL) {
-        unsigned Opc = SPU::DFNMSf64;
-        if (OpVT == MVT::v2f64)
-          Opc = SPU::DFNMSv2f64;
-
-        return CurDAG->getMachineNode(Opc, dl, OpVT,
-                                      Op00.getOperand(0),
-                                      Op00.getOperand(1),
-                                      Op0.getOperand(1));
-      }
-    }
-
-    SDValue negConst = CurDAG->getConstant(0x8000000000000000ULL, MVT::i64);
-    SDNode *signMask = 0;
-    unsigned Opc = SPU::XORfneg64;
-
-    if (OpVT == MVT::f64) {
-      signMask = SelectI64Constant(negConst.getNode(), MVT::i64, dl);
-    } else if (OpVT == MVT::v2f64) {
-      Opc = SPU::XORfnegvec;
-      signMask = emitBuildVector(CurDAG->getNode(ISD::BUILD_VECTOR, dl,
-                                                 MVT::v2i64,
-                                                 negConst, negConst).getNode());
-    }
-
-    return CurDAG->getMachineNode(Opc, dl, OpVT,
-                                  N->getOperand(0), SDValue(signMask, 0));
-  } else if (Opc == ISD::FABS) {
-    if (OpVT == MVT::f64) {
-      SDNode *signMask = SelectI64Constant(0x7fffffffffffffffULL, MVT::i64, dl);
-      return CurDAG->getMachineNode(SPU::ANDfabs64, dl, OpVT,
-                                    N->getOperand(0), SDValue(signMask, 0));
-    } else if (OpVT == MVT::v2f64) {
-      SDValue absConst = CurDAG->getConstant(0x7fffffffffffffffULL, MVT::i64);
-      SDValue absVec = CurDAG->getNode(ISD::BUILD_VECTOR, dl, MVT::v2i64,
-                                       absConst, absConst);
-      SDNode *signMask = emitBuildVector(absVec.getNode());
-      return CurDAG->getMachineNode(SPU::ANDfabsvec, dl, OpVT,
-                                    N->getOperand(0), SDValue(signMask, 0));
-    }
-  } else if (Opc == SPUISD::LDRESULT) {
-    // Custom select instructions for LDRESULT
-    EVT VT = N->getValueType(0);
-    SDValue Arg = N->getOperand(0);
-    SDValue Chain = N->getOperand(1);
-    SDNode *Result;
-
-    Result = CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl, VT,
-                                    MVT::Other, Arg,
-                                    getRC( VT.getSimpleVT()), Chain);
-    return Result;
-
-  } else if (Opc == SPUISD::IndirectAddr) {
-    // Look at the operands: SelectCode() will catch the cases that aren't
-    // specifically handled here.
-    //
-    // SPUInstrInfo catches the following patterns:
-    // (SPUindirect (SPUhi ...), (SPUlo ...))
-    // (SPUindirect $sp, imm)
-    EVT VT = N->getValueType(0);
-    SDValue Op0 = N->getOperand(0);
-    SDValue Op1 = N->getOperand(1);
-    RegisterSDNode *RN;
-
-    if ((Op0.getOpcode() != SPUISD::Hi && Op1.getOpcode() != SPUISD::Lo)
-        || (Op0.getOpcode() == ISD::Register
-            && ((RN = dyn_cast<RegisterSDNode>(Op0.getNode())) != 0
-                && RN->getReg() != SPU::R1))) {
-      NewOpc = SPU::Ar32;
-      Ops[1] = Op1;
-      if (Op1.getOpcode() == ISD::Constant) {
-        ConstantSDNode *CN = cast<ConstantSDNode>(Op1);
-        Op1 = CurDAG->getTargetConstant(CN->getSExtValue(), VT);
-        if (isInt<10>(CN->getSExtValue())) {
-          NewOpc = SPU::AIr32;
-          Ops[1] = Op1;
-        } else {
-          Ops[1] = SDValue(CurDAG->getMachineNode(SPU::ILr32, dl,
-                                                  N->getValueType(0),
-                                                  Op1),
-                           0);
-        }
-      }
-      Ops[0] = Op0;
-      n_ops = 2;
-    }
-  }
-
-  if (n_ops > 0) {
-    if (N->hasOneUse())
-      return CurDAG->SelectNodeTo(N, NewOpc, OpVT, Ops, n_ops);
-    else
-      return CurDAG->getMachineNode(NewOpc, dl, OpVT, Ops, n_ops);
-  } else
-    return SelectCode(N);
-}
-
-/*!
- * Emit the instruction sequence for i64 left shifts. The basic algorithm
- * is to fill the bottom two word slots with zeros so that zeros are shifted
- * in as the entire quadword is shifted left.
- *
- * \note This code could also be used to implement v2i64 shl.
- *
- * @param Op The shl operand
- * @param OpVT Op's machine value value type (doesn't need to be passed, but
- * makes life easier.)
- * @return The SDNode with the entire instruction sequence
- */
-SDNode *
-SPUDAGToDAGISel::SelectSHLi64(SDNode *N, EVT OpVT) {
-  SDValue Op0 = N->getOperand(0);
-  EVT VecVT = EVT::getVectorVT(*CurDAG->getContext(),
-                               OpVT, (128 / OpVT.getSizeInBits()));
-  SDValue ShiftAmt = N->getOperand(1);
-  EVT ShiftAmtVT = ShiftAmt.getValueType();
-  SDNode *VecOp0, *SelMask, *ZeroFill, *Shift = 0;
-  SDValue SelMaskVal;
-  DebugLoc dl = N->getDebugLoc();
-
-  VecOp0 = CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl, VecVT,
-                                  Op0, getRC(MVT::v2i64) );
-  SelMaskVal = CurDAG->getTargetConstant(0xff00ULL, MVT::i16);
-  SelMask = CurDAG->getMachineNode(SPU::FSMBIv2i64, dl, VecVT, SelMaskVal);
-  ZeroFill = CurDAG->getMachineNode(SPU::ILv2i64, dl, VecVT,
-                                    CurDAG->getTargetConstant(0, OpVT));
-  VecOp0 = CurDAG->getMachineNode(SPU::SELBv2i64, dl, VecVT,
-                                  SDValue(ZeroFill, 0),
-                                  SDValue(VecOp0, 0),
-                                  SDValue(SelMask, 0));
-
-  if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(ShiftAmt)) {
-    unsigned bytes = unsigned(CN->getZExtValue()) >> 3;
-    unsigned bits = unsigned(CN->getZExtValue()) & 7;
-
-    if (bytes > 0) {
-      Shift =
-        CurDAG->getMachineNode(SPU::SHLQBYIv2i64, dl, VecVT,
-                               SDValue(VecOp0, 0),
-                               CurDAG->getTargetConstant(bytes, ShiftAmtVT));
-    }
-
-    if (bits > 0) {
-      Shift =
-        CurDAG->getMachineNode(SPU::SHLQBIIv2i64, dl, VecVT,
-                               SDValue((Shift != 0 ? Shift : VecOp0), 0),
-                               CurDAG->getTargetConstant(bits, ShiftAmtVT));
-    }
-  } else {
-    SDNode *Bytes =
-      CurDAG->getMachineNode(SPU::ROTMIr32, dl, ShiftAmtVT,
-                             ShiftAmt,
-                             CurDAG->getTargetConstant(3, ShiftAmtVT));
-    SDNode *Bits =
-      CurDAG->getMachineNode(SPU::ANDIr32, dl, ShiftAmtVT,
-                             ShiftAmt,
-                             CurDAG->getTargetConstant(7, ShiftAmtVT));
-    Shift =
-      CurDAG->getMachineNode(SPU::SHLQBYv2i64, dl, VecVT,
-                             SDValue(VecOp0, 0), SDValue(Bytes, 0));
-    Shift =
-      CurDAG->getMachineNode(SPU::SHLQBIv2i64, dl, VecVT,
-                             SDValue(Shift, 0), SDValue(Bits, 0));
-  }
-
-  return CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl,
-                                OpVT, SDValue(Shift, 0), getRC(MVT::i64));
-}
-
-/*!
- * Emit the instruction sequence for i64 logical right shifts.
- *
- * @param Op The shl operand
- * @param OpVT Op's machine value value type (doesn't need to be passed, but
- * makes life easier.)
- * @return The SDNode with the entire instruction sequence
- */
-SDNode *
-SPUDAGToDAGISel::SelectSRLi64(SDNode *N, EVT OpVT) {
-  SDValue Op0 = N->getOperand(0);
-  EVT VecVT = EVT::getVectorVT(*CurDAG->getContext(),
-                               OpVT, (128 / OpVT.getSizeInBits()));
-  SDValue ShiftAmt = N->getOperand(1);
-  EVT ShiftAmtVT = ShiftAmt.getValueType();
-  SDNode *VecOp0, *Shift = 0;
-  DebugLoc dl = N->getDebugLoc();
-
-  VecOp0 = CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl, VecVT,
-                                  Op0, getRC(MVT::v2i64) );
-
-  if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(ShiftAmt)) {
-    unsigned bytes = unsigned(CN->getZExtValue()) >> 3;
-    unsigned bits = unsigned(CN->getZExtValue()) & 7;
-
-    if (bytes > 0) {
-      Shift =
-        CurDAG->getMachineNode(SPU::ROTQMBYIv2i64, dl, VecVT,
-                               SDValue(VecOp0, 0),
-                               CurDAG->getTargetConstant(bytes, ShiftAmtVT));
-    }
-
-    if (bits > 0) {
-      Shift =
-        CurDAG->getMachineNode(SPU::ROTQMBIIv2i64, dl, VecVT,
-                               SDValue((Shift != 0 ? Shift : VecOp0), 0),
-                               CurDAG->getTargetConstant(bits, ShiftAmtVT));
-    }
-  } else {
-    SDNode *Bytes =
-      CurDAG->getMachineNode(SPU::ROTMIr32, dl, ShiftAmtVT,
-                             ShiftAmt,
-                             CurDAG->getTargetConstant(3, ShiftAmtVT));
-    SDNode *Bits =
-      CurDAG->getMachineNode(SPU::ANDIr32, dl, ShiftAmtVT,
-                             ShiftAmt,
-                             CurDAG->getTargetConstant(7, ShiftAmtVT));
-
-    // Ensure that the shift amounts are negated!
-    Bytes = CurDAG->getMachineNode(SPU::SFIr32, dl, ShiftAmtVT,
-                                   SDValue(Bytes, 0),
-                                   CurDAG->getTargetConstant(0, ShiftAmtVT));
-
-    Bits = CurDAG->getMachineNode(SPU::SFIr32, dl, ShiftAmtVT,
-                                  SDValue(Bits, 0),
-                                  CurDAG->getTargetConstant(0, ShiftAmtVT));
-
-    Shift =
-      CurDAG->getMachineNode(SPU::ROTQMBYv2i64, dl, VecVT,
-                             SDValue(VecOp0, 0), SDValue(Bytes, 0));
-    Shift =
-      CurDAG->getMachineNode(SPU::ROTQMBIv2i64, dl, VecVT,
-                             SDValue(Shift, 0), SDValue(Bits, 0));
-  }
-
-  return CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl,
-                                OpVT, SDValue(Shift, 0), getRC(MVT::i64));
-}
-
-/*!
- * Emit the instruction sequence for i64 arithmetic right shifts.
- *
- * @param Op The shl operand
- * @param OpVT Op's machine value value type (doesn't need to be passed, but
- * makes life easier.)
- * @return The SDNode with the entire instruction sequence
- */
-SDNode *
-SPUDAGToDAGISel::SelectSRAi64(SDNode *N, EVT OpVT) {
-  // Promote Op0 to vector
-  EVT VecVT = EVT::getVectorVT(*CurDAG->getContext(),
-                               OpVT, (128 / OpVT.getSizeInBits()));
-  SDValue ShiftAmt = N->getOperand(1);
-  EVT ShiftAmtVT = ShiftAmt.getValueType();
-  DebugLoc dl = N->getDebugLoc();
-
-  SDNode *VecOp0 =
-    CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl,
-                           VecVT, N->getOperand(0), getRC(MVT::v2i64));
-
-  SDValue SignRotAmt = CurDAG->getTargetConstant(31, ShiftAmtVT);
-  SDNode *SignRot =
-    CurDAG->getMachineNode(SPU::ROTMAIv2i64_i32, dl, MVT::v2i64,
-                           SDValue(VecOp0, 0), SignRotAmt);
-  SDNode *UpperHalfSign =
-    CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl,
-                           MVT::i32, SDValue(SignRot, 0), getRC(MVT::i32));
-
-  SDNode *UpperHalfSignMask =
-    CurDAG->getMachineNode(SPU::FSM64r32, dl, VecVT, SDValue(UpperHalfSign, 0));
-  SDNode *UpperLowerMask =
-    CurDAG->getMachineNode(SPU::FSMBIv2i64, dl, VecVT,
-                           CurDAG->getTargetConstant(0xff00ULL, MVT::i16));
-  SDNode *UpperLowerSelect =
-    CurDAG->getMachineNode(SPU::SELBv2i64, dl, VecVT,
-                           SDValue(UpperHalfSignMask, 0),
-                           SDValue(VecOp0, 0),
-                           SDValue(UpperLowerMask, 0));
-
-  SDNode *Shift = 0;
-
-  if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(ShiftAmt)) {
-    unsigned bytes = unsigned(CN->getZExtValue()) >> 3;
-    unsigned bits = unsigned(CN->getZExtValue()) & 7;
-
-    if (bytes > 0) {
-      bytes = 31 - bytes;
-      Shift =
-        CurDAG->getMachineNode(SPU::ROTQBYIv2i64, dl, VecVT,
-                               SDValue(UpperLowerSelect, 0),
-                               CurDAG->getTargetConstant(bytes, ShiftAmtVT));
-    }
-
-    if (bits > 0) {
-      bits = 8 - bits;
-      Shift =
-        CurDAG->getMachineNode(SPU::ROTQBIIv2i64, dl, VecVT,
-                               SDValue((Shift != 0 ? Shift : UpperLowerSelect), 0),
-                               CurDAG->getTargetConstant(bits, ShiftAmtVT));
-    }
-  } else {
-    SDNode *NegShift =
-      CurDAG->getMachineNode(SPU::SFIr32, dl, ShiftAmtVT,
-                             ShiftAmt, CurDAG->getTargetConstant(0, ShiftAmtVT));
-
-    Shift =
-      CurDAG->getMachineNode(SPU::ROTQBYBIv2i64_r32, dl, VecVT,
-                             SDValue(UpperLowerSelect, 0), SDValue(NegShift, 0));
-    Shift =
-      CurDAG->getMachineNode(SPU::ROTQBIv2i64, dl, VecVT,
-                             SDValue(Shift, 0), SDValue(NegShift, 0));
-  }
-
-  return CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl,
-                                OpVT, SDValue(Shift, 0), getRC(MVT::i64));
-}
-
-/*!
- Do the necessary magic necessary to load a i64 constant
- */
-SDNode *SPUDAGToDAGISel::SelectI64Constant(SDNode *N, EVT OpVT,
-                                           DebugLoc dl) {
-  ConstantSDNode *CN = cast<ConstantSDNode>(N);
-  return SelectI64Constant(CN->getZExtValue(), OpVT, dl);
-}
-
-SDNode *SPUDAGToDAGISel::SelectI64Constant(uint64_t Value64, EVT OpVT,
-                                           DebugLoc dl) {
-  EVT OpVecVT = EVT::getVectorVT(*CurDAG->getContext(), OpVT, 2);
-  SDValue i64vec =
-          SPU::LowerV2I64Splat(OpVecVT, *CurDAG, Value64, dl);
-
-  // Here's where it gets interesting, because we have to parse out the
-  // subtree handed back in i64vec:
-
-  if (i64vec.getOpcode() == ISD::BITCAST) {
-    // The degenerate case where the upper and lower bits in the splat are
-    // identical:
-    SDValue Op0 = i64vec.getOperand(0);
-
-    ReplaceUses(i64vec, Op0);
-    return CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl, OpVT,
-                                  SDValue(emitBuildVector(Op0.getNode()), 0),
-                                  getRC(MVT::i64));
-  } else if (i64vec.getOpcode() == SPUISD::SHUFB) {
-    SDValue lhs = i64vec.getOperand(0);
-    SDValue rhs = i64vec.getOperand(1);
-    SDValue shufmask = i64vec.getOperand(2);
-
-    if (lhs.getOpcode() == ISD::BITCAST) {
-      ReplaceUses(lhs, lhs.getOperand(0));
-      lhs = lhs.getOperand(0);
-    }
-
-    SDNode *lhsNode = (lhs.getNode()->isMachineOpcode()
-                       ? lhs.getNode()
-                       : emitBuildVector(lhs.getNode()));
-
-    if (rhs.getOpcode() == ISD::BITCAST) {
-      ReplaceUses(rhs, rhs.getOperand(0));
-      rhs = rhs.getOperand(0);
-    }
-
-    SDNode *rhsNode = (rhs.getNode()->isMachineOpcode()
-                       ? rhs.getNode()
-                       : emitBuildVector(rhs.getNode()));
-
-    if (shufmask.getOpcode() == ISD::BITCAST) {
-      ReplaceUses(shufmask, shufmask.getOperand(0));
-      shufmask = shufmask.getOperand(0);
-    }
-
-    SDNode *shufMaskNode = (shufmask.getNode()->isMachineOpcode()
-                            ? shufmask.getNode()
-                            : emitBuildVector(shufmask.getNode()));
-
-   SDValue shufNode =
-            CurDAG->getNode(SPUISD::SHUFB, dl, OpVecVT,
-                                   SDValue(lhsNode, 0), SDValue(rhsNode, 0),
-                                   SDValue(shufMaskNode, 0));
-    HandleSDNode Dummy(shufNode);
-    SDNode *SN = SelectCode(Dummy.getValue().getNode());
-    if (SN == 0) SN = Dummy.getValue().getNode();
-
-    return CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl,
-                                  OpVT, SDValue(SN, 0), getRC(MVT::i64));
-  } else if (i64vec.getOpcode() == ISD::BUILD_VECTOR) {
-    return CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, dl, OpVT,
-                                  SDValue(emitBuildVector(i64vec.getNode()), 0),
-                                  getRC(MVT::i64));
-  } else {
-    report_fatal_error("SPUDAGToDAGISel::SelectI64Constant: Unhandled i64vec"
-                      "condition");
-  }
-}
-
-/// createSPUISelDag - This pass converts a legalized DAG into a
-/// SPU-specific DAG, ready for instruction scheduling.
-///
-FunctionPass *llvm::createSPUISelDag(SPUTargetMachine &TM) {
-  return new SPUDAGToDAGISel(TM);
-}
diff --git a/lib/Target/CellSPU/SPUISelLowering.cpp b/lib/Target/CellSPU/SPUISelLowering.cpp
deleted file mode 100644
index 4e9fcd1bc765..000000000000
--- a/lib/Target/CellSPU/SPUISelLowering.cpp
+++ /dev/null
@@ -1,3266 +0,0 @@
-//===-- SPUISelLowering.cpp - Cell SPU DAG Lowering Implementation --------===//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file implements the SPUTargetLowering class.
-//
-//===----------------------------------------------------------------------===//
-
-#include "SPUISelLowering.h"
-#include "SPUTargetMachine.h"
-#include "SPUFrameLowering.h"
-#include "SPUMachineFunction.h"
-#include "llvm/Constants.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Type.h"
-#include "llvm/CodeGen/CallingConvLower.h"
-#include "llvm/CodeGen/MachineFrameInfo.h"
-#include "llvm/CodeGen/MachineFunction.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/SelectionDAG.h"
-#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/MathExtras.h"
-#include "llvm/Support/raw_ostream.h"
-
-using namespace llvm;
-
-namespace {
-  // Byte offset of the preferred slot (counted from the MSB)
-  int prefslotOffset(EVT VT) {
-    int retval=0;
-    if (VT==MVT::i1) retval=3;
-    if (VT==MVT::i8) retval=3;
-    if (VT==MVT::i16) retval=2;
-
-    return retval;
-  }
-
-  //! Expand a library call into an actual call DAG node
-  /*!
-   \note
-   This code is taken from SelectionDAGLegalize, since it is not exposed as
-   part of the LLVM SelectionDAG API.
-   */
-
-  SDValue
-  ExpandLibCall(RTLIB::Libcall LC, SDValue Op, SelectionDAG &DAG,
-                bool isSigned, SDValue &Hi, const SPUTargetLowering &TLI) {
-    // The input chain to this libcall is the entry node of the function.
-    // Legalizing the call will automatically add the previous call to the
-    // dependence.
-    SDValue InChain = DAG.getEntryNode();
-
-    TargetLowering::ArgListTy Args;
-    TargetLowering::ArgListEntry Entry;
-    for (unsigned i = 0, e = Op.getNumOperands(); i != e; ++i) {
-      EVT ArgVT = Op.getOperand(i).getValueType();
-      Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
-      Entry.Node = Op.getOperand(i);
-      Entry.Ty = ArgTy;
-      Entry.isSExt = isSigned;
-      Entry.isZExt = !isSigned;
-      Args.push_back(Entry);
-    }
-    SDValue Callee = DAG.getExternalSymbol(TLI.getLibcallName(LC),
-                                           TLI.getPointerTy());
-
-    // Splice the libcall in wherever FindInputOutputChains tells us to.
-    Type *RetTy =
-                Op.getNode()->getValueType(0).getTypeForEVT(*DAG.getContext());
-    TargetLowering::CallLoweringInfo CLI(InChain, RetTy, isSigned, !isSigned,
-                                         false, false,
-                            0, TLI.getLibcallCallingConv(LC),
-                            /*isTailCall=*/false,
-                                         /*doesNotRet=*/false,
-                                         /*isReturnValueUsed=*/true,
-                            Callee, Args, DAG, Op.getDebugLoc());
-    std::pair<SDValue, SDValue> CallInfo = TLI.LowerCallTo(CLI);
-
-    return CallInfo.first;
-  }
-}
-
-SPUTargetLowering::SPUTargetLowering(SPUTargetMachine &TM)
-  : TargetLowering(TM, new TargetLoweringObjectFileELF()),
-    SPUTM(TM) {
-
-  // Use _setjmp/_longjmp instead of setjmp/longjmp.
-  setUseUnderscoreSetJmp(true);
-  setUseUnderscoreLongJmp(true);
-
-  // Set RTLIB libcall names as used by SPU:
-  setLibcallName(RTLIB::DIV_F64, "__fast_divdf3");
-
-  // Set up the SPU's register classes:
-  addRegisterClass(MVT::i8,   &SPU::R8CRegClass);
-  addRegisterClass(MVT::i16,  &SPU::R16CRegClass);
-  addRegisterClass(MVT::i32,  &SPU::R32CRegClass);
-  addRegisterClass(MVT::i64,  &SPU::R64CRegClass);
-  addRegisterClass(MVT::f32,  &SPU::R32FPRegClass);
-  addRegisterClass(MVT::f64,  &SPU::R64FPRegClass);
-  addRegisterClass(MVT::i128, &SPU::GPRCRegClass);
-
-  // SPU has no sign or zero extended loads for i1, i8, i16:
-  setLoadExtAction(ISD::EXTLOAD,  MVT::i1, Promote);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
-
-  setLoadExtAction(ISD::EXTLOAD,  MVT::f32, Expand);
-  setLoadExtAction(ISD::EXTLOAD,  MVT::f64, Expand);
-
-  setTruncStoreAction(MVT::i128, MVT::i64, Expand);
-  setTruncStoreAction(MVT::i128, MVT::i32, Expand);
-  setTruncStoreAction(MVT::i128, MVT::i16, Expand);
-  setTruncStoreAction(MVT::i128, MVT::i8, Expand);
-
-  setTruncStoreAction(MVT::f64, MVT::f32, Expand);
-
-  // SPU constant load actions are custom lowered:
-  setOperationAction(ISD::ConstantFP, MVT::f32, Legal);
-  setOperationAction(ISD::ConstantFP, MVT::f64, Custom);
-
-  // SPU's loads and stores have to be custom lowered:
-  for (unsigned sctype = (unsigned) MVT::i8; sctype < (unsigned) MVT::i128;
-       ++sctype) {
-    MVT::SimpleValueType VT = (MVT::SimpleValueType)sctype;
-
-    setOperationAction(ISD::LOAD,   VT, Custom);
-    setOperationAction(ISD::STORE,  VT, Custom);
-    setLoadExtAction(ISD::EXTLOAD,  VT, Custom);
-    setLoadExtAction(ISD::ZEXTLOAD, VT, Custom);
-    setLoadExtAction(ISD::SEXTLOAD, VT, Custom);
-
-    for (unsigned stype = sctype - 1; stype >= (unsigned) MVT::i8; --stype) {
-      MVT::SimpleValueType StoreVT = (MVT::SimpleValueType) stype;
-      setTruncStoreAction(VT, StoreVT, Expand);
-    }
-  }
-
-  for (unsigned sctype = (unsigned) MVT::f32; sctype < (unsigned) MVT::f64;
-       ++sctype) {
-    MVT::SimpleValueType VT = (MVT::SimpleValueType) sctype;
-
-    setOperationAction(ISD::LOAD,   VT, Custom);
-    setOperationAction(ISD::STORE,  VT, Custom);
-
-    for (unsigned stype = sctype - 1; stype >= (unsigned) MVT::f32; --stype) {
-      MVT::SimpleValueType StoreVT = (MVT::SimpleValueType) stype;
-      setTruncStoreAction(VT, StoreVT, Expand);
-    }
-  }
-
-  // Expand the jumptable branches
-  setOperationAction(ISD::BR_JT,        MVT::Other, Expand);
-  setOperationAction(ISD::BR_CC,        MVT::Other, Expand);
-
-  // Custom lower SELECT_CC for most cases, but expand by default
-  setOperationAction(ISD::SELECT_CC,    MVT::Other, Expand);
-  setOperationAction(ISD::SELECT_CC,    MVT::i8,    Custom);
-  setOperationAction(ISD::SELECT_CC,    MVT::i16,   Custom);
-  setOperationAction(ISD::SELECT_CC,    MVT::i32,   Custom);
-  setOperationAction(ISD::SELECT_CC,    MVT::i64,   Custom);
-
-  // SPU has no intrinsics for these particular operations:
-  setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand);
-  setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Expand);
-
-  // SPU has no division/remainder instructions
-  setOperationAction(ISD::SREM,    MVT::i8,   Expand);
-  setOperationAction(ISD::UREM,    MVT::i8,   Expand);
-  setOperationAction(ISD::SDIV,    MVT::i8,   Expand);
-  setOperationAction(ISD::UDIV,    MVT::i8,   Expand);
-  setOperationAction(ISD::SDIVREM, MVT::i8,   Expand);
-  setOperationAction(ISD::UDIVREM, MVT::i8,   Expand);
-  setOperationAction(ISD::SREM,    MVT::i16,  Expand);
-  setOperationAction(ISD::UREM,    MVT::i16,  Expand);
-  setOperationAction(ISD::SDIV,    MVT::i16,  Expand);
-  setOperationAction(ISD::UDIV,    MVT::i16,  Expand);
-  setOperationAction(ISD::SDIVREM, MVT::i16,  Expand);
-  setOperationAction(ISD::UDIVREM, MVT::i16,  Expand);
-  setOperationAction(ISD::SREM,    MVT::i32,  Expand);
-  setOperationAction(ISD::UREM,    MVT::i32,  Expand);
-  setOperationAction(ISD::SDIV,    MVT::i32,  Expand);
-  setOperationAction(ISD::UDIV,    MVT::i32,  Expand);
-  setOperationAction(ISD::SDIVREM, MVT::i32,  Expand);
-  setOperationAction(ISD::UDIVREM, MVT::i32,  Expand);
-  setOperationAction(ISD::SREM,    MVT::i64,  Expand);
-  setOperationAction(ISD::UREM,    MVT::i64,  Expand);
-  setOperationAction(ISD::SDIV,    MVT::i64,  Expand);
-  setOperationAction(ISD::UDIV,    MVT::i64,  Expand);
-  setOperationAction(ISD::SDIVREM, MVT::i64,  Expand);
-  setOperationAction(ISD::UDIVREM, MVT::i64,  Expand);
-  setOperationAction(ISD::SREM,    MVT::i128, Expand);
-  setOperationAction(ISD::UREM,    MVT::i128, Expand);
-  setOperationAction(ISD::SDIV,    MVT::i128, Expand);
-  setOperationAction(ISD::UDIV,    MVT::i128, Expand);
-  setOperationAction(ISD::SDIVREM, MVT::i128, Expand);
-  setOperationAction(ISD::UDIVREM, MVT::i128, Expand);
-
-  // We don't support sin/cos/sqrt/fmod
-  setOperationAction(ISD::FSIN , MVT::f64, Expand);
-  setOperationAction(ISD::FCOS , MVT::f64, Expand);
-  setOperationAction(ISD::FREM , MVT::f64, Expand);
-  setOperationAction(ISD::FSIN , MVT::f32, Expand);
-  setOperationAction(ISD::FCOS , MVT::f32, Expand);
-  setOperationAction(ISD::FREM , MVT::f32, Expand);
-
-  // Expand fsqrt to the appropriate libcall (NOTE: should use h/w fsqrt
-  // for f32!)
-  setOperationAction(ISD::FSQRT, MVT::f64, Expand);
-  setOperationAction(ISD::FSQRT, MVT::f32, Expand);
-
-  setOperationAction(ISD::FMA, MVT::f64, Expand);
-  setOperationAction(ISD::FMA, MVT::f32, Expand);
-
-  setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
-  setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
-
-  // SPU can do rotate right and left, so legalize it... but customize for i8
-  // because instructions don't exist.
-
-  // FIXME: Change from "expand" to appropriate type once ROTR is supported in
-  //        .td files.
-  setOperationAction(ISD::ROTR, MVT::i32,    Expand /*Legal*/);
-  setOperationAction(ISD::ROTR, MVT::i16,    Expand /*Legal*/);
-  setOperationAction(ISD::ROTR, MVT::i8,     Expand /*Custom*/);
-
-  setOperationAction(ISD::ROTL, MVT::i32,    Legal);
-  setOperationAction(ISD::ROTL, MVT::i16,    Legal);
-  setOperationAction(ISD::ROTL, MVT::i8,     Custom);
-
-  // SPU has no native version of shift left/right for i8
-  setOperationAction(ISD::SHL,  MVT::i8,     Custom);
-  setOperationAction(ISD::SRL,  MVT::i8,     Custom);
-  setOperationAction(ISD::SRA,  MVT::i8,     Custom);
-
-  // Make these operations legal and handle them during instruction selection:
-  setOperationAction(ISD::SHL,  MVT::i64,    Legal);
-  setOperationAction(ISD::SRL,  MVT::i64,    Legal);
-  setOperationAction(ISD::SRA,  MVT::i64,    Legal);
-
-  // Custom lower i8, i32 and i64 multiplications
-  setOperationAction(ISD::MUL,  MVT::i8,     Custom);
-  setOperationAction(ISD::MUL,  MVT::i32,    Legal);
-  setOperationAction(ISD::MUL,  MVT::i64,    Legal);
-
-  // Expand double-width multiplication
-  // FIXME: It would probably be reasonable to support some of these operations
-  setOperationAction(ISD::UMUL_LOHI, MVT::i8,  Expand);
-  setOperationAction(ISD::SMUL_LOHI, MVT::i8,  Expand);
-  setOperationAction(ISD::MULHU,     MVT::i8,  Expand);
-  setOperationAction(ISD::MULHS,     MVT::i8,  Expand);
-  setOperationAction(ISD::UMUL_LOHI, MVT::i16, Expand);
-  setOperationAction(ISD::SMUL_LOHI, MVT::i16, Expand);
-  setOperationAction(ISD::MULHU,     MVT::i16, Expand);
-  setOperationAction(ISD::MULHS,     MVT::i16, Expand);
-  setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
-  setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
-  setOperationAction(ISD::MULHU,     MVT::i32, Expand);
-  setOperationAction(ISD::MULHS,     MVT::i32, Expand);
-  setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
-  setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);
-  setOperationAction(ISD::MULHU,     MVT::i64, Expand);
-  setOperationAction(ISD::MULHS,     MVT::i64, Expand);
-
-  // Need to custom handle (some) common i8, i64 math ops
-  setOperationAction(ISD::ADD,  MVT::i8,     Custom);
-  setOperationAction(ISD::ADD,  MVT::i64,    Legal);
-  setOperationAction(ISD::SUB,  MVT::i8,     Custom);
-  setOperationAction(ISD::SUB,  MVT::i64,    Legal);
-
-  // SPU does not have BSWAP. It does have i32 support CTLZ.
-  // CTPOP has to be custom lowered.
-  setOperationAction(ISD::BSWAP, MVT::i32,   Expand);
-  setOperationAction(ISD::BSWAP, MVT::i64,   Expand);
-
-  setOperationAction(ISD::CTPOP, MVT::i8,    Custom);
-  setOperationAction(ISD::CTPOP, MVT::i16,   Custom);
-  setOperationAction(ISD::CTPOP, MVT::i32,   Custom);
-  setOperationAction(ISD::CTPOP, MVT::i64,   Custom);
-  setOperationAction(ISD::CTPOP, MVT::i128,  Expand);
-
-  setOperationAction(ISD::CTTZ , MVT::i8,    Expand);
-  setOperationAction(ISD::CTTZ , MVT::i16,   Expand);
-  setOperationAction(ISD::CTTZ , MVT::i32,   Expand);
-  setOperationAction(ISD::CTTZ , MVT::i64,   Expand);
-  setOperationAction(ISD::CTTZ , MVT::i128,  Expand);
-  setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i8,    Expand);
-  setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i16,   Expand);
-  setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i32,   Expand);
-  setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i64,   Expand);
-  setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i128,  Expand);
-
-  setOperationAction(ISD::CTLZ , MVT::i8,    Promote);
-  setOperationAction(ISD::CTLZ , MVT::i16,   Promote);
-  setOperationAction(ISD::CTLZ , MVT::i32,   Legal);
-  setOperationAction(ISD::CTLZ , MVT::i64,   Expand);
-  setOperationAction(ISD::CTLZ , MVT::i128,  Expand);
-  setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i8,    Expand);
-  setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i16,   Expand);
-  setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i32,   Expand);
-  setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i64,   Expand);
-  setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i128,  Expand);
-
-  // SPU has a version of select that implements (a&~c)|(b&c), just like
-  // select ought to work:
-  setOperationAction(ISD::SELECT, MVT::i8,   Legal);
-  setOperationAction(ISD::SELECT, MVT::i16,  Legal);
-  setOperationAction(ISD::SELECT, MVT::i32,  Legal);
-  setOperationAction(ISD::SELECT, MVT::i64,  Legal);
-
-  setOperationAction(ISD::SETCC, MVT::i8,    Legal);
-  setOperationAction(ISD::SETCC, MVT::i16,   Legal);
-  setOperationAction(ISD::SETCC, MVT::i32,   Legal);
-  setOperationAction(ISD::SETCC, MVT::i64,   Legal);
-  setOperationAction(ISD::SETCC, MVT::f64,   Custom);
-
-  // Custom lower i128 -> i64 truncates
-  setOperationAction(ISD::TRUNCATE, MVT::i64, Custom);
-
-  // Custom lower i32/i64 -> i128 sign extend
-  setOperationAction(ISD::SIGN_EXTEND, MVT::i128, Custom);
-
-  setOperationAction(ISD::FP_TO_SINT, MVT::i8, Promote);
-  setOperationAction(ISD::FP_TO_UINT, MVT::i8, Promote);
-  setOperationAction(ISD::FP_TO_SINT, MVT::i16, Promote);
-  setOperationAction(ISD::FP_TO_UINT, MVT::i16, Promote);
-  // SPU has a legal FP -> signed INT instruction for f32, but for f64, need
-  // to expand to a libcall, hence the custom lowering:
-  setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
-  setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
-  setOperationAction(ISD::FP_TO_SINT, MVT::i64, Expand);
-  setOperationAction(ISD::FP_TO_UINT, MVT::i64, Expand);
-  setOperationAction(ISD::FP_TO_SINT, MVT::i128, Expand);
-  setOperationAction(ISD::FP_TO_UINT, MVT::i128, Expand);
-
-  // FDIV on SPU requires custom lowering
-  setOperationAction(ISD::FDIV, MVT::f64, Expand);      // to libcall
-
-  // SPU has [U|S]INT_TO_FP for f32->i32, but not for f64->i32, f64->i64:
-  setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
-  setOperationAction(ISD::SINT_TO_FP, MVT::i16, Promote);
-  setOperationAction(ISD::SINT_TO_FP, MVT::i8,  Promote);
-  setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
-  setOperationAction(ISD::UINT_TO_FP, MVT::i16, Promote);
-  setOperationAction(ISD::UINT_TO_FP, MVT::i8,  Promote);
-  setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
-  setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
-
-  setOperationAction(ISD::BITCAST, MVT::i32, Legal);
-  setOperationAction(ISD::BITCAST, MVT::f32, Legal);
-  setOperationAction(ISD::BITCAST, MVT::i64, Legal);
-  setOperationAction(ISD::BITCAST, MVT::f64, Legal);
-
-  // We cannot sextinreg(i1).  Expand to shifts.
-  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
-
-  // We want to legalize GlobalAddress and ConstantPool nodes into the
-  // appropriate instructions to materialize the address.
-  for (unsigned sctype = (unsigned) MVT::i8; sctype < (unsigned) MVT::f128;
-       ++sctype) {
-    MVT::SimpleValueType VT = (MVT::SimpleValueType)sctype;
-
-    setOperationAction(ISD::GlobalAddress,  VT, Custom);
-    setOperationAction(ISD::ConstantPool,   VT, Custom);
-    setOperationAction(ISD::JumpTable,      VT, Custom);
-  }
-
-  // VASTART needs to be custom lowered to use the VarArgsFrameIndex
-  setOperationAction(ISD::VASTART           , MVT::Other, Custom);
-
-  // Use the default implementation.
-  setOperationAction(ISD::VAARG             , MVT::Other, Expand);
-  setOperationAction(ISD::VACOPY            , MVT::Other, Expand);
-  setOperationAction(ISD::VAEND             , MVT::Other, Expand);
-  setOperationAction(ISD::STACKSAVE         , MVT::Other, Expand);
-  setOperationAction(ISD::STACKRESTORE      , MVT::Other, Expand);
-  setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32  , Expand);
-  setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64  , Expand);
-
-  // Cell SPU has instructions for converting between i64 and fp.
-  setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
-  setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
-
-  // To take advantage of the above i64 FP_TO_SINT, promote i32 FP_TO_UINT
-  setOperationAction(ISD::FP_TO_UINT, MVT::i32, Promote);
-
-  // BUILD_PAIR can't be handled natively, and should be expanded to shl/or
-  setOperationAction(ISD::BUILD_PAIR, MVT::i64, Expand);
-
-  // First set operation action for all vector types to expand. Then we
-  // will selectively turn on ones that can be effectively codegen'd.
-  addRegisterClass(MVT::v16i8, &SPU::VECREGRegClass);
-  addRegisterClass(MVT::v8i16, &SPU::VECREGRegClass);
-  addRegisterClass(MVT::v4i32, &SPU::VECREGRegClass);
-  addRegisterClass(MVT::v2i64, &SPU::VECREGRegClass);
-  addRegisterClass(MVT::v4f32, &SPU::VECREGRegClass);
-  addRegisterClass(MVT::v2f64, &SPU::VECREGRegClass);
-
-  for (unsigned i = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
-       i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) {
-    MVT::SimpleValueType VT = (MVT::SimpleValueType)i;
-
-    // Set operation actions to legal types only.
-    if (!isTypeLegal(VT)) continue;
-
-    // add/sub are legal for all supported vector VT's.
-    setOperationAction(ISD::ADD,     VT, Legal);
-    setOperationAction(ISD::SUB,     VT, Legal);
-    // mul has to be custom lowered.
-    setOperationAction(ISD::MUL,     VT, Legal);
-
-    setOperationAction(ISD::AND,     VT, Legal);
-    setOperationAction(ISD::OR,      VT, Legal);
-    setOperationAction(ISD::XOR,     VT, Legal);
-    setOperationAction(ISD::LOAD,    VT, Custom);
-    setOperationAction(ISD::SELECT,  VT, Legal);
-    setOperationAction(ISD::STORE,   VT, Custom);
-
-    // These operations need to be expanded:
-    setOperationAction(ISD::SDIV,    VT, Expand);
-    setOperationAction(ISD::SREM,    VT, Expand);
-    setOperationAction(ISD::UDIV,    VT, Expand);
-    setOperationAction(ISD::UREM,    VT, Expand);
-
-    // Expand all trunc stores
-    for (unsigned j = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
-         j <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++j) {
-      MVT::SimpleValueType TargetVT = (MVT::SimpleValueType)j;
-    setTruncStoreAction(VT, TargetVT, Expand);
-    }
-
-    // Custom lower build_vector, constant pool spills, insert and
-    // extract vector elements:
-    setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
-    setOperationAction(ISD::ConstantPool, VT, Custom);
-    setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Custom);
-    setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
-    setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
-    setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
-  }
-
-  setOperationAction(ISD::SHL, MVT::v2i64, Expand);
-
-  setOperationAction(ISD::AND, MVT::v16i8, Custom);
-  setOperationAction(ISD::OR,  MVT::v16i8, Custom);
-  setOperationAction(ISD::XOR, MVT::v16i8, Custom);
-  setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v4f32, Custom);
-
-  setOperationAction(ISD::FDIV, MVT::v4f32, Legal);
-
-  setBooleanContents(ZeroOrNegativeOneBooleanContent);
-  setBooleanVectorContents(ZeroOrNegativeOneBooleanContent); // FIXME: Is this correct?
-
-  setStackPointerRegisterToSaveRestore(SPU::R1);
-
-  // We have target-specific dag combine patterns for the following nodes:
-  setTargetDAGCombine(ISD::ADD);
-  setTargetDAGCombine(ISD::ZERO_EXTEND);
-  setTargetDAGCombine(ISD::SIGN_EXTEND);
-  setTargetDAGCombine(ISD::ANY_EXTEND);
-
-  setMinFunctionAlignment(3);
-
-  computeRegisterProperties();
-
-  // Set pre-RA register scheduler default to BURR, which produces slightly
-  // better code than the default (could also be TDRR, but TargetLowering.h
-  // needs a mod to support that model):
-  setSchedulingPreference(Sched::RegPressure);
-}
-
-const char *SPUTargetLowering::getTargetNodeName(unsigned Opcode) const {
-  switch (Opcode) {
-  default: return 0;
-  case SPUISD::RET_FLAG: return "SPUISD::RET_FLAG";
-  case SPUISD::Hi: return "SPUISD::Hi";
-  case SPUISD::Lo: return "SPUISD::Lo";
-  case SPUISD::PCRelAddr: return "SPUISD::PCRelAddr";
-  case SPUISD::AFormAddr: return "SPUISD::AFormAddr";
-  case SPUISD::IndirectAddr: return "SPUISD::IndirectAddr";
-  case SPUISD::LDRESULT: return "SPUISD::LDRESULT";
-  case SPUISD::CALL: return "SPUISD::CALL";
-  case SPUISD::SHUFB: return "SPUISD::SHUFB";
-  case SPUISD::SHUFFLE_MASK: return "SPUISD::SHUFFLE_MASK";
-  case SPUISD::CNTB: return "SPUISD::CNTB";
-  case SPUISD::PREFSLOT2VEC: return "SPUISD::PREFSLOT2VEC";
-  case SPUISD::VEC2PREFSLOT: return "SPUISD::VEC2PREFSLOT";
-  case SPUISD::SHL_BITS: return "SPUISD::SHL_BITS";
-  case SPUISD::SHL_BYTES: return "SPUISD::SHL_BYTES";
-  case SPUISD::VEC_ROTL: return "SPUISD::VEC_ROTL";
-  case SPUISD::VEC_ROTR: return "SPUISD::VEC_ROTR";
-  case SPUISD::ROTBYTES_LEFT: return "SPUISD::ROTBYTES_LEFT";
-  case SPUISD::ROTBYTES_LEFT_BITS: return "SPUISD::ROTBYTES_LEFT_BITS";
-  case SPUISD::SELECT_MASK: return "SPUISD::SELECT_MASK";
-  case SPUISD::SELB: return "SPUISD::SELB";
-  case SPUISD::ADD64_MARKER: return "SPUISD::ADD64_MARKER";
-  case SPUISD::SUB64_MARKER: return "SPUISD::SUB64_MARKER";
-  case SPUISD::MUL64_MARKER: return "SPUISD::MUL64_MARKER";
-  }
-}
-
-//===----------------------------------------------------------------------===//
-// Return the Cell SPU's SETCC result type
-//===----------------------------------------------------------------------===//
-
-EVT SPUTargetLowering::getSetCCResultType(EVT VT) const {
-  // i8, i16 and i32 are valid SETCC result types
-  MVT::SimpleValueType retval;
-
-  switch(VT.getSimpleVT().SimpleTy){
-    case MVT::i1:
-    case MVT::i8:
-      retval = MVT::i8; break;
-    case MVT::i16:
-      retval = MVT::i16; break;
-    case MVT::i32:
-    default:
-      retval = MVT::i32;
-  }
-  return retval;
-}
-
-//===----------------------------------------------------------------------===//
-// Calling convention code:
-//===----------------------------------------------------------------------===//
-
-#include "SPUGenCallingConv.inc"
-
-//===----------------------------------------------------------------------===//
-//  LowerOperation implementation
-//===----------------------------------------------------------------------===//
-
-/// Custom lower loads for CellSPU
-/*!
- All CellSPU loads and stores are aligned to 16-byte boundaries, so for elements
- within a 16-byte block, we have to rotate to extract the requested element.
-
- For extending loads, we also want to ensure that the following sequence is
- emitted, e.g. for MVT::f32 extending load to MVT::f64:
-
-\verbatim
-%1  v16i8,ch = load
-%2  v16i8,ch = rotate %1
-%3  v4f8, ch = bitconvert %2
-%4  f32      = vec2perfslot %3
-%5  f64      = fp_extend %4
-\endverbatim
-*/
-static SDValue
-LowerLOAD(SDValue Op, SelectionDAG &DAG, const SPUSubtarget *ST) {
-  LoadSDNode *LN = cast<LoadSDNode>(Op);
-  SDValue the_chain = LN->getChain();
-  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
-  EVT InVT = LN->getMemoryVT();
-  EVT OutVT = Op.getValueType();
-  ISD::LoadExtType ExtType = LN->getExtensionType();
-  unsigned alignment = LN->getAlignment();
-  int pso = prefslotOffset(InVT);
-  DebugLoc dl = Op.getDebugLoc();
-  EVT vecVT = InVT.isVector()? InVT: EVT::getVectorVT(*DAG.getContext(), InVT,
-                                                  (128 / InVT.getSizeInBits()));
-
-  // two sanity checks
-  assert( LN->getAddressingMode() == ISD::UNINDEXED
-          && "we should get only UNINDEXED adresses");
-  // clean aligned loads can be selected as-is
-  if (InVT.getSizeInBits() == 128 && (alignment%16) == 0)
-    return SDValue();
-
-  // Get pointerinfos to the memory chunk(s) that contain the data to load
-  uint64_t mpi_offset = LN->getPointerInfo().Offset;
-  mpi_offset -= mpi_offset%16;
-  MachinePointerInfo lowMemPtr(LN->getPointerInfo().V, mpi_offset);
-  MachinePointerInfo highMemPtr(LN->getPointerInfo().V, mpi_offset+16);
-
-  SDValue result;
-  SDValue basePtr = LN->getBasePtr();
-  SDValue rotate;
-
-  if ((alignment%16) == 0) {
-    ConstantSDNode *CN;
-
-    // Special cases for a known aligned load to simplify the base pointer
-    // and the rotation amount:
-    if (basePtr.getOpcode() == ISD::ADD
-        && (CN = dyn_cast<ConstantSDNode > (basePtr.getOperand(1))) != 0) {
-      // Known offset into basePtr
-      int64_t offset = CN->getSExtValue();
-      int64_t rotamt = int64_t((offset & 0xf) - pso);
-
-      if (rotamt < 0)
-        rotamt += 16;
-
-      rotate = DAG.getConstant(rotamt, MVT::i16);
-
-      // Simplify the base pointer for this case:
-      basePtr = basePtr.getOperand(0);
-      if ((offset & ~0xf) > 0) {
-        basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
-                              basePtr,
-                              DAG.getConstant((offset & ~0xf), PtrVT));
-      }
-    } else if ((basePtr.getOpcode() == SPUISD::AFormAddr)
-               || (basePtr.getOpcode() == SPUISD::IndirectAddr
-                   && basePtr.getOperand(0).getOpcode() == SPUISD::Hi
-                   && basePtr.getOperand(1).getOpcode() == SPUISD::Lo)) {
-      // Plain aligned a-form address: rotate into preferred slot
-      // Same for (SPUindirect (SPUhi ...), (SPUlo ...))
-      int64_t rotamt = -pso;
-      if (rotamt < 0)
-        rotamt += 16;
-      rotate = DAG.getConstant(rotamt, MVT::i16);
-    } else {
-      // Offset the rotate amount by the basePtr and the preferred slot
-      // byte offset
-      int64_t rotamt = -pso;
-      if (rotamt < 0)
-        rotamt += 16;
-      rotate = DAG.getNode(ISD::ADD, dl, PtrVT,
-                           basePtr,
-                           DAG.getConstant(rotamt, PtrVT));
-    }
-  } else {
-    // Unaligned load: must be more pessimistic about addressing modes:
-    if (basePtr.getOpcode() == ISD::ADD) {
-      MachineFunction &MF = DAG.getMachineFunction();
-      MachineRegisterInfo &RegInfo = MF.getRegInfo();
-      unsigned VReg = RegInfo.createVirtualRegister(&SPU::R32CRegClass);
-      SDValue Flag;
-
-      SDValue Op0 = basePtr.getOperand(0);
-      SDValue Op1 = basePtr.getOperand(1);
-
-      if (isa<ConstantSDNode>(Op1)) {
-        // Convert the (add <ptr>, <const>) to an indirect address contained
-        // in a register. Note that this is done because we need to avoid
-        // creating a 0(reg) d-form address due to the SPU's block loads.
-        basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Op0, Op1);
-        the_chain = DAG.getCopyToReg(the_chain, dl, VReg, basePtr, Flag);
-        basePtr = DAG.getCopyFromReg(the_chain, dl, VReg, PtrVT);
-      } else {
-        // Convert the (add <arg1>, <arg2>) to an indirect address, which
-        // will likely be lowered as a reg(reg) x-form address.
-        basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Op0, Op1);
-      }
-    } else {
-      basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
-                            basePtr,
-                            DAG.getConstant(0, PtrVT));
-   }
-
-    // Offset the rotate amount by the basePtr and the preferred slot
-    // byte offset
-    rotate = DAG.getNode(ISD::ADD, dl, PtrVT,
-                         basePtr,
-                         DAG.getConstant(-pso, PtrVT));
-  }
-
-  // Do the load as a i128 to allow possible shifting
-  SDValue low = DAG.getLoad(MVT::i128, dl, the_chain, basePtr,
-                       lowMemPtr,
-                       LN->isVolatile(), LN->isNonTemporal(), false, 16);
-
-  // When the size is not greater than alignment we get all data with just
-  // one load
-  if (alignment >= InVT.getSizeInBits()/8) {
-    // Update the chain
-    the_chain = low.getValue(1);
-
-    // Rotate into the preferred slot:
-    result = DAG.getNode(SPUISD::ROTBYTES_LEFT, dl, MVT::i128,
-                         low.getValue(0), rotate);
-
-    // Convert the loaded v16i8 vector to the appropriate vector type
-    // specified by the operand:
-    EVT vecVT = EVT::getVectorVT(*DAG.getContext(),
-                                 InVT, (128 / InVT.getSizeInBits()));
-    result = DAG.getNode(SPUISD::VEC2PREFSLOT, dl, InVT,
-                         DAG.getNode(ISD::BITCAST, dl, vecVT, result));
-  }
-  // When alignment is less than the size, we might need (known only at
-  // run-time) two loads
-  // TODO: if the memory address is composed only from constants, we have
-  // extra kowledge, and might avoid the second load
-  else {
-    // storage position offset from lower 16 byte aligned memory chunk
-    SDValue offset = DAG.getNode(ISD::AND, dl, MVT::i32,
-                                  basePtr, DAG.getConstant( 0xf, MVT::i32 ) );
-    // get a registerfull of ones. (this implementation is a workaround: LLVM
-    // cannot handle 128 bit signed int constants)
-    SDValue ones = DAG.getConstant(-1, MVT::v4i32 );
-    ones = DAG.getNode(ISD::BITCAST, dl, MVT::i128, ones);
-
-    SDValue high = DAG.getLoad(MVT::i128, dl, the_chain,
-                               DAG.getNode(ISD::ADD, dl, PtrVT,
-                                           basePtr,
-                                           DAG.getConstant(16, PtrVT)),
-                               highMemPtr,
-                               LN->isVolatile(), LN->isNonTemporal(), false, 
-                               16);
-
-    the_chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, low.getValue(1),
-                                                              high.getValue(1));
-
-    // Shift the (possible) high part right to compensate the misalignemnt.
-    // if there is no highpart (i.e. value is i64 and offset is 4), this
-    // will zero out the high value.
-    high = DAG.getNode(SPUISD::SRL_BYTES, dl, MVT::i128, high,
-                                     DAG.getNode(ISD::SUB, dl, MVT::i32,
-                                                 DAG.getConstant( 16, MVT::i32),
-                                                 offset
-                                                ));
-
-    // Shift the low similarly
-    // TODO: add SPUISD::SHL_BYTES
-    low = DAG.getNode(SPUISD::SHL_BYTES, dl, MVT::i128, low, offset );
-
-    // Merge the two parts
-    result = DAG.getNode(ISD::BITCAST, dl, vecVT,
-                          DAG.getNode(ISD::OR, dl, MVT::i128, low, high));
-
-    if (!InVT.isVector()) {
-      result = DAG.getNode(SPUISD::VEC2PREFSLOT, dl, InVT, result );
-     }
-
-  }
-    // Handle extending loads by extending the scalar result:
-    if (ExtType == ISD::SEXTLOAD) {
-      result = DAG.getNode(ISD::SIGN_EXTEND, dl, OutVT, result);
-    } else if (ExtType == ISD::ZEXTLOAD) {
-      result = DAG.getNode(ISD::ZERO_EXTEND, dl, OutVT, result);
-    } else if (ExtType == ISD::EXTLOAD) {
-      unsigned NewOpc = ISD::ANY_EXTEND;
-
-      if (OutVT.isFloatingPoint())
-        NewOpc = ISD::FP_EXTEND;
-
-      result = DAG.getNode(NewOpc, dl, OutVT, result);
-    }
-
-    SDVTList retvts = DAG.getVTList(OutVT, MVT::Other);
-    SDValue retops[2] = {
-      result,
-      the_chain
-    };
-
-    result = DAG.getNode(SPUISD::LDRESULT, dl, retvts,
-                         retops, sizeof(retops) / sizeof(retops[0]));
-    return result;
-}
-
-/// Custom lower stores for CellSPU
-/*!
- All CellSPU stores are aligned to 16-byte boundaries, so for elements
- within a 16-byte block, we have to generate a shuffle to insert the
- requested element into its place, then store the resulting block.
- */
-static SDValue
-LowerSTORE(SDValue Op, SelectionDAG &DAG, const SPUSubtarget *ST) {
-  StoreSDNode *SN = cast<StoreSDNode>(Op);
-  SDValue Value = SN->getValue();
-  EVT VT = Value.getValueType();
-  EVT StVT = (!SN->isTruncatingStore() ? VT : SN->getMemoryVT());
-  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
-  DebugLoc dl = Op.getDebugLoc();
-  unsigned alignment = SN->getAlignment();
-  SDValue result;
-  EVT vecVT = StVT.isVector()? StVT: EVT::getVectorVT(*DAG.getContext(), StVT,
-                                                 (128 / StVT.getSizeInBits()));
-  // Get pointerinfos to the memory chunk(s) that contain the data to load
-  uint64_t mpi_offset = SN->getPointerInfo().Offset;
-  mpi_offset -= mpi_offset%16;
-  MachinePointerInfo lowMemPtr(SN->getPointerInfo().V, mpi_offset);
-  MachinePointerInfo highMemPtr(SN->getPointerInfo().V, mpi_offset+16);
-
-
-  // two sanity checks
-  assert( SN->getAddressingMode() == ISD::UNINDEXED
-          && "we should get only UNINDEXED adresses");
-  // clean aligned loads can be selected as-is
-  if (StVT.getSizeInBits() == 128 && (alignment%16) == 0)
-    return SDValue();
-
-  SDValue alignLoadVec;
-  SDValue basePtr = SN->getBasePtr();
-  SDValue the_chain = SN->getChain();
-  SDValue insertEltOffs;
-
-  if ((alignment%16) == 0) {
-    ConstantSDNode *CN;
-    // Special cases for a known aligned load to simplify the base pointer
-    // and insertion byte:
-    if (basePtr.getOpcode() == ISD::ADD
-        && (CN = dyn_cast<ConstantSDNode>(basePtr.getOperand(1))) != 0) {
-      // Known offset into basePtr
-      int64_t offset = CN->getSExtValue();
-
-      // Simplify the base pointer for this case:
-      basePtr = basePtr.getOperand(0);
-      insertEltOffs = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
-                                  basePtr,
-                                  DAG.getConstant((offset & 0xf), PtrVT));
-
-      if ((offset & ~0xf) > 0) {
-        basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
-                              basePtr,
-                              DAG.getConstant((offset & ~0xf), PtrVT));
-      }
-    } else {
-      // Otherwise, assume it's at byte 0 of basePtr
-      insertEltOffs = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
-                                  basePtr,
-                                  DAG.getConstant(0, PtrVT));
-      basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
-                                  basePtr,
-                                  DAG.getConstant(0, PtrVT));
-    }
-  } else {
-    // Unaligned load: must be more pessimistic about addressing modes:
-    if (basePtr.getOpcode() == ISD::ADD) {
-      MachineFunction &MF = DAG.getMachineFunction();
-      MachineRegisterInfo &RegInfo = MF.getRegInfo();
-      unsigned VReg = RegInfo.createVirtualRegister(&SPU::R32CRegClass);
-      SDValue Flag;
-
-      SDValue Op0 = basePtr.getOperand(0);
-      SDValue Op1 = basePtr.getOperand(1);
-
-      if (isa<ConstantSDNode>(Op1)) {
-        // Convert the (add <ptr>, <const>) to an indirect address contained
-        // in a register. Note that this is done because we need to avoid
-        // creating a 0(reg) d-form address due to the SPU's block loads.
-        basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Op0, Op1);
-        the_chain = DAG.getCopyToReg(the_chain, dl, VReg, basePtr, Flag);
-        basePtr = DAG.getCopyFromReg(the_chain, dl, VReg, PtrVT);
-      } else {
-        // Convert the (add <arg1>, <arg2>) to an indirect address, which
-        // will likely be lowered as a reg(reg) x-form address.
-        basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Op0, Op1);
-      }
-    } else {
-      basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
-                            basePtr,
-                            DAG.getConstant(0, PtrVT));
-    }
-
-    // Insertion point is solely determined by basePtr's contents
-    insertEltOffs = DAG.getNode(ISD::ADD, dl, PtrVT,
-                                basePtr,
-                                DAG.getConstant(0, PtrVT));
-  }
-
-  // Load the lower part of the memory to which to store.
-  SDValue low = DAG.getLoad(vecVT, dl, the_chain, basePtr,
-                          lowMemPtr, SN->isVolatile(), SN->isNonTemporal(),
-                            false, 16);
-
-  // if we don't need to store over the 16 byte boundary, one store suffices
-  if (alignment >= StVT.getSizeInBits()/8) {
-    // Update the chain
-    the_chain = low.getValue(1);
-
-    LoadSDNode *LN = cast<LoadSDNode>(low);
-    SDValue theValue = SN->getValue();
-
-    if (StVT != VT
-        && (theValue.getOpcode() == ISD::AssertZext
-            || theValue.getOpcode() == ISD::AssertSext)) {
-      // Drill down and get the value for zero- and sign-extended
-      // quantities
-      theValue = theValue.getOperand(0);
-    }
-
-    // If the base pointer is already a D-form address, then just create
-    // a new D-form address with a slot offset and the orignal base pointer.
-    // Otherwise generate a D-form address with the slot offset relative
-    // to the stack pointer, which is always aligned.
-#if !defined(NDEBUG)
-      if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
-        errs() << "CellSPU LowerSTORE: basePtr = ";
-        basePtr.getNode()->dump(&DAG);
-        errs() << "\n";
-      }
-#endif
-
-    SDValue insertEltOp = DAG.getNode(SPUISD::SHUFFLE_MASK, dl, vecVT,
-                                      insertEltOffs);
-    SDValue vectorizeOp = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, vecVT,
-                                      theValue);
-
-    result = DAG.getNode(SPUISD::SHUFB, dl, vecVT,
-                         vectorizeOp, low,
-                         DAG.getNode(ISD::BITCAST, dl,
-                                     MVT::v4i32, insertEltOp));
-
-    result = DAG.getStore(the_chain, dl, result, basePtr,
-                          lowMemPtr,
-                          LN->isVolatile(), LN->isNonTemporal(),
-                          16);
-
-  }
-  // do the store when it might cross the 16 byte memory access boundary.
-  else {
-    // TODO issue a warning if SN->isVolatile()== true? This is likely not
-    // what the user wanted.
-
-    // address offset from nearest lower 16byte alinged address
-    SDValue offset = DAG.getNode(ISD::AND, dl, MVT::i32,
-                                    SN->getBasePtr(),
-                                    DAG.getConstant(0xf, MVT::i32));
-    // 16 - offset
-    SDValue offset_compl = DAG.getNode(ISD::SUB, dl, MVT::i32,
-                                           DAG.getConstant( 16, MVT::i32),
-                                           offset);
-    // 16 - sizeof(Value)
-    SDValue surplus = DAG.getNode(ISD::SUB, dl, MVT::i32,
-                                     DAG.getConstant( 16, MVT::i32),
-                                     DAG.getConstant( VT.getSizeInBits()/8,
-                                                      MVT::i32));
-    // get a registerfull of ones
-    SDValue ones = DAG.getConstant(-1, MVT::v4i32);
-    ones = DAG.getNode(ISD::BITCAST, dl, MVT::i128, ones);
-
-    // Create the 128 bit masks that have ones where the data to store is
-    // located.
-    SDValue lowmask, himask;
-    // if the value to store don't fill up the an entire 128 bits, zero
-    // out the last bits of the mask so that only the value we want to store
-    // is masked.
-    // this is e.g. in the case of store i32, align 2
-    if (!VT.isVector()){
-      Value = DAG.getNode(SPUISD::PREFSLOT2VEC, dl, vecVT, Value);
-      lowmask = DAG.getNode(SPUISD::SRL_BYTES, dl, MVT::i128, ones, surplus);
-      lowmask = DAG.getNode(SPUISD::SHL_BYTES, dl, MVT::i128, lowmask,
-                                                               surplus);
-      Value = DAG.getNode(ISD::BITCAST, dl, MVT::i128, Value);
-      Value = DAG.getNode(ISD::AND, dl, MVT::i128, Value, lowmask);
-
-    }
-    else {
-      lowmask = ones;
-      Value = DAG.getNode(ISD::BITCAST, dl, MVT::i128, Value);
-    }
-    // this will zero, if there are no data that goes to the high quad
-    himask = DAG.getNode(SPUISD::SHL_BYTES, dl, MVT::i128, lowmask,
-                                                            offset_compl);
-    lowmask = DAG.getNode(SPUISD::SRL_BYTES, dl, MVT::i128, lowmask,
-                                                             offset);
-
-    // Load in the old data and zero out the parts that will be overwritten with
-    // the new data to store.
-    SDValue hi = DAG.getLoad(MVT::i128, dl, the_chain,
-                               DAG.getNode(ISD::ADD, dl, PtrVT, basePtr,
-                                           DAG.getConstant( 16, PtrVT)),
-                               highMemPtr,
-                               SN->isVolatile(), SN->isNonTemporal(), 
-                               false, 16);
-    the_chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, low.getValue(1),
-                                                              hi.getValue(1));
-
-    low = DAG.getNode(ISD::AND, dl, MVT::i128,
-                        DAG.getNode( ISD::BITCAST, dl, MVT::i128, low),
-                        DAG.getNode( ISD::XOR, dl, MVT::i128, lowmask, ones));
-    hi = DAG.getNode(ISD::AND, dl, MVT::i128,
-                        DAG.getNode( ISD::BITCAST, dl, MVT::i128, hi),
-                        DAG.getNode( ISD::XOR, dl, MVT::i128, himask, ones));
-
-    // Shift the Value to store into place. rlow contains the parts that go to
-    // the lower memory chunk, rhi has the parts that go to the upper one.
-    SDValue rlow = DAG.getNode(SPUISD::SRL_BYTES, dl, MVT::i128, Value, offset);
-    rlow = DAG.getNode(ISD::AND, dl, MVT::i128, rlow, lowmask);
-    SDValue rhi = DAG.getNode(SPUISD::SHL_BYTES, dl, MVT::i128, Value,
-                                                            offset_compl);
-
-    // Merge the old data and the new data and store the results
-    // Need to convert vectors here to integer as 'OR'ing floats assert
-    rlow = DAG.getNode(ISD::OR, dl, MVT::i128,
-                          DAG.getNode(ISD::BITCAST, dl, MVT::i128, low),
-                          DAG.getNode(ISD::BITCAST, dl, MVT::i128, rlow));
-    rhi = DAG.getNode(ISD::OR, dl, MVT::i128,
-                         DAG.getNode(ISD::BITCAST, dl, MVT::i128, hi),
-                         DAG.getNode(ISD::BITCAST, dl, MVT::i128, rhi));
-
-    low = DAG.getStore(the_chain, dl, rlow, basePtr,
-                          lowMemPtr,
-                          SN->isVolatile(), SN->isNonTemporal(), 16);
-    hi  = DAG.getStore(the_chain, dl, rhi,
-                            DAG.getNode(ISD::ADD, dl, PtrVT, basePtr,
-                                        DAG.getConstant( 16, PtrVT)),
-                            highMemPtr,
-                            SN->isVolatile(), SN->isNonTemporal(), 16);
-    result = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, low.getValue(0),
-                                                           hi.getValue(0));
-  }
-
-  return result;
-}
-
-//! Generate the address of a constant pool entry.
-static SDValue
-LowerConstantPool(SDValue Op, SelectionDAG &DAG, const SPUSubtarget *ST) {
-  EVT PtrVT = Op.getValueType();
-  ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
-  const Constant *C = CP->getConstVal();
-  SDValue CPI = DAG.getTargetConstantPool(C, PtrVT, CP->getAlignment());
-  SDValue Zero = DAG.getConstant(0, PtrVT);
-  const TargetMachine &TM = DAG.getTarget();
-  // FIXME there is no actual debug info here
-  DebugLoc dl = Op.getDebugLoc();
-
-  if (TM.getRelocationModel() == Reloc::Static) {
-    if (!ST->usingLargeMem()) {
-      // Just return the SDValue with the constant pool address in it.
-      return DAG.getNode(SPUISD::AFormAddr, dl, PtrVT, CPI, Zero);
-    } else {
-      SDValue Hi = DAG.getNode(SPUISD::Hi, dl, PtrVT, CPI, Zero);
-      SDValue Lo = DAG.getNode(SPUISD::Lo, dl, PtrVT, CPI, Zero);
-      return DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Hi, Lo);
-    }
-  }
-
-  llvm_unreachable("LowerConstantPool: Relocation model other than static"
-                   " not supported.");
-}
-
-//! Alternate entry point for generating the address of a constant pool entry
-SDValue
-SPU::LowerConstantPool(SDValue Op, SelectionDAG &DAG, const SPUTargetMachine &TM) {
-  return ::LowerConstantPool(Op, DAG, TM.getSubtargetImpl());
-}
-
-static SDValue
-LowerJumpTable(SDValue Op, SelectionDAG &DAG, const SPUSubtarget *ST) {
-  EVT PtrVT = Op.getValueType();
-  JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
-  SDValue JTI = DAG.getTargetJumpTable(JT->getIndex(), PtrVT);
-  SDValue Zero = DAG.getConstant(0, PtrVT);
-  const TargetMachine &TM = DAG.getTarget();
-  // FIXME there is no actual debug info here
-  DebugLoc dl = Op.getDebugLoc();
-
-  if (TM.getRelocationModel() == Reloc::Static) {
-    if (!ST->usingLargeMem()) {
-      return DAG.getNode(SPUISD::AFormAddr, dl, PtrVT, JTI, Zero);
-    } else {
-      SDValue Hi = DAG.getNode(SPUISD::Hi, dl, PtrVT, JTI, Zero);
-      SDValue Lo = DAG.getNode(SPUISD::Lo, dl, PtrVT, JTI, Zero);
-      return DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Hi, Lo);
-    }
-  }
-
-  llvm_unreachable("LowerJumpTable: Relocation model other than static"
-                   " not supported.");
-}
-
-static SDValue
-LowerGlobalAddress(SDValue Op, SelectionDAG &DAG, const SPUSubtarget *ST) {
-  EVT PtrVT = Op.getValueType();
-  GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
-  const GlobalValue *GV = GSDN->getGlobal();
-  SDValue GA = DAG.getTargetGlobalAddress(GV, Op.getDebugLoc(),
-                                          PtrVT, GSDN->getOffset());
-  const TargetMachine &TM = DAG.getTarget();
-  SDValue Zero = DAG.getConstant(0, PtrVT);
-  // FIXME there is no actual debug info here
-  DebugLoc dl = Op.getDebugLoc();
-
-  if (TM.getRelocationModel() == Reloc::Static) {
-    if (!ST->usingLargeMem()) {
-      return DAG.getNode(SPUISD::AFormAddr, dl, PtrVT, GA, Zero);
-    } else {
-      SDValue Hi = DAG.getNode(SPUISD::Hi, dl, PtrVT, GA, Zero);
-      SDValue Lo = DAG.getNode(SPUISD::Lo, dl, PtrVT, GA, Zero);
-      return DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Hi, Lo);
-    }
-  } else {
-    report_fatal_error("LowerGlobalAddress: Relocation model other than static"
-                      "not supported.");
-    /*NOTREACHED*/
-  }
-}
-
-//! Custom lower double precision floating point constants
-static SDValue
-LowerConstantFP(SDValue Op, SelectionDAG &DAG) {
-  EVT VT = Op.getValueType();
-  // FIXME there is no actual debug info here
-  DebugLoc dl = Op.getDebugLoc();
-
-  if (VT == MVT::f64) {
-    ConstantFPSDNode *FP = cast<ConstantFPSDNode>(Op.getNode());
-
-    assert((FP != 0) &&
-           "LowerConstantFP: Node is not ConstantFPSDNode");
-
-    uint64_t dbits = DoubleToBits(FP->getValueAPF().convertToDouble());
-    SDValue T = DAG.getConstant(dbits, MVT::i64);
-    SDValue Tvec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v2i64, T, T);
-    return DAG.getNode(SPUISD::VEC2PREFSLOT, dl, VT,
-                       DAG.getNode(ISD::BITCAST, dl, MVT::v2f64, Tvec));
-  }
-
-  return SDValue();
-}
-
-SDValue
-SPUTargetLowering::LowerFormalArguments(SDValue Chain,
-                                        CallingConv::ID CallConv, bool isVarArg,
-                                        const SmallVectorImpl<ISD::InputArg>
-                                          &Ins,
-                                        DebugLoc dl, SelectionDAG &DAG,
-                                        SmallVectorImpl<SDValue> &InVals)
-                                          const {
-
-  MachineFunction &MF = DAG.getMachineFunction();
-  MachineFrameInfo *MFI = MF.getFrameInfo();
-  MachineRegisterInfo &RegInfo = MF.getRegInfo();
-  SPUFunctionInfo *FuncInfo = MF.getInfo<SPUFunctionInfo>();
-
-  unsigned ArgOffset = SPUFrameLowering::minStackSize();
-  unsigned ArgRegIdx = 0;
-  unsigned StackSlotSize = SPUFrameLowering::stackSlotSize();
-
-  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
-
-  SmallVector<CCValAssign, 16> ArgLocs;
-  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
-                 getTargetMachine(), ArgLocs, *DAG.getContext());
-  // FIXME: allow for other calling conventions
-  CCInfo.AnalyzeFormalArguments(Ins, CCC_SPU);
-
-  // Add DAG nodes to load the arguments or copy them out of registers.
-  for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo) {
-    EVT ObjectVT = Ins[ArgNo].VT;
-    unsigned ObjSize = ObjectVT.getSizeInBits()/8;
-    SDValue ArgVal;
-    CCValAssign &VA = ArgLocs[ArgNo];
-
-    if (VA.isRegLoc()) {
-      const TargetRegisterClass *ArgRegClass;
-
-      switch (ObjectVT.getSimpleVT().SimpleTy) {
-      default:
-        report_fatal_error("LowerFormalArguments Unhandled argument type: " +
-                           Twine(ObjectVT.getEVTString()));
-      case MVT::i8:
-        ArgRegClass = &SPU::R8CRegClass;
-        break;
-      case MVT::i16:
-        ArgRegClass = &SPU::R16CRegClass;
-        break;
-      case MVT::i32:
-        ArgRegClass = &SPU::R32CRegClass;
-        break;
-      case MVT::i64:
-        ArgRegClass = &SPU::R64CRegClass;
-        break;
-      case MVT::i128:
-        ArgRegClass = &SPU::GPRCRegClass;
-        break;
-      case MVT::f32:
-        ArgRegClass = &SPU::R32FPRegClass;
-        break;
-      case MVT::f64:
-        ArgRegClass = &SPU::R64FPRegClass;
-        break;
-      case MVT::v2f64:
-      case MVT::v4f32:
-      case MVT::v2i64:
-      case MVT::v4i32:
-      case MVT::v8i16:
-      case MVT::v16i8:
-        ArgRegClass = &SPU::VECREGRegClass;
-        break;
-      }
-
-      unsigned VReg = RegInfo.createVirtualRegister(ArgRegClass);
-      RegInfo.addLiveIn(VA.getLocReg(), VReg);
-      ArgVal = DAG.getCopyFromReg(Chain, dl, VReg, ObjectVT);
-      ++ArgRegIdx;
-    } else {
-      // We need to load the argument to a virtual register if we determined
-      // above that we ran out of physical registers of the appropriate type
-      // or we're forced to do vararg
-      int FI = MFI->CreateFixedObject(ObjSize, ArgOffset, true);
-      SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
-      ArgVal = DAG.getLoad(ObjectVT, dl, Chain, FIN, MachinePointerInfo(),
-                           false, false, false, 0);
-      ArgOffset += StackSlotSize;
-    }
-
-    InVals.push_back(ArgVal);
-    // Update the chain
-    Chain = ArgVal.getOperand(0);
-  }
-
-  // vararg handling:
-  if (isVarArg) {
-    // FIXME: we should be able to query the argument registers from
-    //        tablegen generated code.
-    static const uint16_t ArgRegs[] = {
-      SPU::R3,  SPU::R4,  SPU::R5,  SPU::R6,  SPU::R7,  SPU::R8,  SPU::R9,
-      SPU::R10, SPU::R11, SPU::R12, SPU::R13, SPU::R14, SPU::R15, SPU::R16,
-      SPU::R17, SPU::R18, SPU::R19, SPU::R20, SPU::R21, SPU::R22, SPU::R23,
-      SPU::R24, SPU::R25, SPU::R26, SPU::R27, SPU::R28, SPU::R29, SPU::R30,
-      SPU::R31, SPU::R32, SPU::R33, SPU::R34, SPU::R35, SPU::R36, SPU::R37,
-      SPU::R38, SPU::R39, SPU::R40, SPU::R41, SPU::R42, SPU::R43, SPU::R44,
-      SPU::R45, SPU::R46, SPU::R47, SPU::R48, SPU::R49, SPU::R50, SPU::R51,
-      SPU::R52, SPU::R53, SPU::R54, SPU::R55, SPU::R56, SPU::R57, SPU::R58,
-      SPU::R59, SPU::R60, SPU::R61, SPU::R62, SPU::R63, SPU::R64, SPU::R65,
-      SPU::R66, SPU::R67, SPU::R68, SPU::R69, SPU::R70, SPU::R71, SPU::R72,
-      SPU::R73, SPU::R74, SPU::R75, SPU::R76, SPU::R77, SPU::R78, SPU::R79
-    };
-    // size of ArgRegs array
-    const unsigned NumArgRegs = 77;
-
-    // We will spill (79-3)+1 registers to the stack
-    SmallVector<SDValue, 79-3+1> MemOps;
-
-    // Create the frame slot
-    for (; ArgRegIdx != NumArgRegs; ++ArgRegIdx) {
-      FuncInfo->setVarArgsFrameIndex(
-        MFI->CreateFixedObject(StackSlotSize, ArgOffset, true));
-      SDValue FIN = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
-      unsigned VReg = MF.addLiveIn(ArgRegs[ArgRegIdx], &SPU::VECREGRegClass);
-      SDValue ArgVal = DAG.getRegister(VReg, MVT::v16i8);
-      SDValue Store = DAG.getStore(Chain, dl, ArgVal, FIN, MachinePointerInfo(),
-                                   false, false, 0);
-      Chain = Store.getOperand(0);
-      MemOps.push_back(Store);
-
-      // Increment address by stack slot size for the next stored argument
-      ArgOffset += StackSlotSize;
-    }
-    if (!MemOps.empty())
-      Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
-                          &MemOps[0], MemOps.size());
-  }
-
-  return Chain;
-}
-
-/// isLSAAddress - Return the immediate to use if the specified
-/// value is representable as a LSA address.
-static SDNode *isLSAAddress(SDValue Op, SelectionDAG &DAG) {
-  ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
-  if (!C) return 0;
-
-  int Addr = C->getZExtValue();
-  if ((Addr & 3) != 0 ||  // Low 2 bits are implicitly zero.
-      (Addr << 14 >> 14) != Addr)
-    return 0;  // Top 14 bits have to be sext of immediate.
-
-  return DAG.getConstant((int)C->getZExtValue() >> 2, MVT::i32).getNode();
-}
-
-SDValue
-SPUTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
-                             SmallVectorImpl<SDValue> &InVals) const {
-  SelectionDAG &DAG                     = CLI.DAG;
-  DebugLoc &dl                          = CLI.DL;
-  SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
-  SmallVector<SDValue, 32> &OutVals     = CLI.OutVals;
-  SmallVector<ISD::InputArg, 32> &Ins   = CLI.Ins;
-  SDValue Chain                         = CLI.Chain;
-  SDValue Callee                        = CLI.Callee;
-  bool &isTailCall                      = CLI.IsTailCall;
-  CallingConv::ID CallConv              = CLI.CallConv;
-  bool isVarArg                         = CLI.IsVarArg;
-
-  // CellSPU target does not yet support tail call optimization.
-  isTailCall = false;
-
-  const SPUSubtarget *ST = SPUTM.getSubtargetImpl();
-  unsigned NumOps     = Outs.size();
-  unsigned StackSlotSize = SPUFrameLowering::stackSlotSize();
-
-  SmallVector<CCValAssign, 16> ArgLocs;
-  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
-                 getTargetMachine(), ArgLocs, *DAG.getContext());
-  // FIXME: allow for other calling conventions
-  CCInfo.AnalyzeCallOperands(Outs, CCC_SPU);
-
-  const unsigned NumArgRegs = ArgLocs.size();
-
-
-  // Handy pointer type
-  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
-
-  // Set up a copy of the stack pointer for use loading and storing any
-  // arguments that may not fit in the registers available for argument
-  // passing.
-  SDValue StackPtr = DAG.getRegister(SPU::R1, MVT::i32);
-
-  // Figure out which arguments are going to go in registers, and which in
-  // memory.
-  unsigned ArgOffset = SPUFrameLowering::minStackSize(); // Just below [LR]
-  unsigned ArgRegIdx = 0;
-
-  // Keep track of registers passing arguments
-  std::vector<std::pair<unsigned, SDValue> > RegsToPass;
-  // And the arguments passed on the stack
-  SmallVector<SDValue, 8> MemOpChains;
-
-  for (; ArgRegIdx != NumOps; ++ArgRegIdx) {
-    SDValue Arg = OutVals[ArgRegIdx];
-    CCValAssign &VA = ArgLocs[ArgRegIdx];
-
-    // PtrOff will be used to store the current argument to the stack if a
-    // register cannot be found for it.
-    SDValue PtrOff = DAG.getConstant(ArgOffset, StackPtr.getValueType());
-    PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr, PtrOff);
-
-    switch (Arg.getValueType().getSimpleVT().SimpleTy) {
-    default: llvm_unreachable("Unexpected ValueType for argument!");
-    case MVT::i8:
-    case MVT::i16:
-    case MVT::i32:
-    case MVT::i64:
-    case MVT::i128:
-    case MVT::f32:
-    case MVT::f64:
-    case MVT::v2i64:
-    case MVT::v2f64:
-    case MVT::v4f32:
-    case MVT::v4i32:
-    case MVT::v8i16:
-    case MVT::v16i8:
-      if (ArgRegIdx != NumArgRegs) {
-        RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
-      } else {
-        MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
-                                           MachinePointerInfo(),
-                                           false, false, 0));
-        ArgOffset += StackSlotSize;
-      }
-      break;
-    }
-  }
-
-  // Accumulate how many bytes are to be pushed on the stack, including the
-  // linkage area, and parameter passing area.  According to the SPU ABI,
-  // we minimally need space for [LR] and [SP].
-  unsigned NumStackBytes = ArgOffset - SPUFrameLowering::minStackSize();
-
-  // Insert a call sequence start
-  Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumStackBytes,
-                                                            true));
-
-  if (!MemOpChains.empty()) {
-    // Adjust the stack pointer for the stack arguments.
-    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
-                        &MemOpChains[0], MemOpChains.size());
-  }
-
-  // Build a sequence of copy-to-reg nodes chained together with token chain
-  // and flag operands which copy the outgoing args into the appropriate regs.
-  SDValue InFlag;
-  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
-    Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
-                             RegsToPass[i].second, InFlag);
-    InFlag = Chain.getValue(1);
-  }
-
-  SmallVector<SDValue, 8> Ops;
-  unsigned CallOpc = SPUISD::CALL;
-
-  // If the callee is a GlobalAddress/ExternalSymbol node (quite common, every
-  // direct call is) turn it into a TargetGlobalAddress/TargetExternalSymbol
-  // node so that legalize doesn't hack it.
-  if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
-    const GlobalValue *GV = G->getGlobal();
-    EVT CalleeVT = Callee.getValueType();
-    SDValue Zero = DAG.getConstant(0, PtrVT);
-    SDValue GA = DAG.getTargetGlobalAddress(GV, dl, CalleeVT);
-
-    if (!ST->usingLargeMem()) {
-      // Turn calls to targets that are defined (i.e., have bodies) into BRSL
-      // style calls, otherwise, external symbols are BRASL calls. This assumes
-      // that declared/defined symbols are in the same compilation unit and can
-      // be reached through PC-relative jumps.
-      //
-      // NOTE:
-      // This may be an unsafe assumption for JIT and really large compilation
-      // units.
-      if (GV->isDeclaration()) {
-        Callee = DAG.getNode(SPUISD::AFormAddr, dl, CalleeVT, GA, Zero);
-      } else {
-        Callee = DAG.getNode(SPUISD::PCRelAddr, dl, CalleeVT, GA, Zero);
-      }
-    } else {
-      // "Large memory" mode: Turn all calls into indirect calls with a X-form
-      // address pairs:
-      Callee = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, GA, Zero);
-    }
-  } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
-    EVT CalleeVT = Callee.getValueType();
-    SDValue Zero = DAG.getConstant(0, PtrVT);
-    SDValue ExtSym = DAG.getTargetExternalSymbol(S->getSymbol(),
-        Callee.getValueType());
-
-    if (!ST->usingLargeMem()) {
-      Callee = DAG.getNode(SPUISD::AFormAddr, dl, CalleeVT, ExtSym, Zero);
-    } else {
-      Callee = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, ExtSym, Zero);
-    }
-  } else if (SDNode *Dest = isLSAAddress(Callee, DAG)) {
-    // If this is an absolute destination address that appears to be a legal
-    // local store address, use the munged value.
-    Callee = SDValue(Dest, 0);
-  }
-
-  Ops.push_back(Chain);
-  Ops.push_back(Callee);
-
-  // Add argument registers to the end of the list so that they are known live
-  // into the call.
-  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
-    Ops.push_back(DAG.getRegister(RegsToPass[i].first,
-                                  RegsToPass[i].second.getValueType()));
-
-  if (InFlag.getNode())
-    Ops.push_back(InFlag);
-  // Returns a chain and a flag for retval copy to use.
-  Chain = DAG.getNode(CallOpc, dl, DAG.getVTList(MVT::Other, MVT::Glue),
-                      &Ops[0], Ops.size());
-  InFlag = Chain.getValue(1);
-
-  Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumStackBytes, true),
-                             DAG.getIntPtrConstant(0, true), InFlag);
-  if (!Ins.empty())
-    InFlag = Chain.getValue(1);
-
-  // If the function returns void, just return the chain.
-  if (Ins.empty())
-    return Chain;
-
-  // Now handle the return value(s)
-  SmallVector<CCValAssign, 16> RVLocs;
-  CCState CCRetInfo(CallConv, isVarArg, DAG.getMachineFunction(),
-                    getTargetMachine(), RVLocs, *DAG.getContext());
-  CCRetInfo.AnalyzeCallResult(Ins, CCC_SPU);
-
-
-  // If the call has results, copy the values out of the ret val registers.
-  for (unsigned i = 0; i != RVLocs.size(); ++i) {
-    CCValAssign VA = RVLocs[i];
-
-    SDValue Val = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), VA.getLocVT(),
-                                     InFlag);
-    Chain = Val.getValue(1);
-    InFlag = Val.getValue(2);
-    InVals.push_back(Val);
-   }
-
-  return Chain;
-}
-
-SDValue
-SPUTargetLowering::LowerReturn(SDValue Chain,
-                               CallingConv::ID CallConv, bool isVarArg,
-                               const SmallVectorImpl<ISD::OutputArg> &Outs,
-                               const SmallVectorImpl<SDValue> &OutVals,
-                               DebugLoc dl, SelectionDAG &DAG) const {
-
-  SmallVector<CCValAssign, 16> RVLocs;
-  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
-                 getTargetMachine(), RVLocs, *DAG.getContext());
-  CCInfo.AnalyzeReturn(Outs, RetCC_SPU);
-
-  // If this is the first return lowered for this function, add the regs to the
-  // liveout set for the function.
-  if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
-    for (unsigned i = 0; i != RVLocs.size(); ++i)
-      DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
-  }
-
-  SDValue Flag;
-
-  // Copy the result values into the output registers.
-  for (unsigned i = 0; i != RVLocs.size(); ++i) {
-    CCValAssign &VA = RVLocs[i];
-    assert(VA.isRegLoc() && "Can only return in registers!");
-    Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
-                             OutVals[i], Flag);
-    Flag = Chain.getValue(1);
-  }
-
-  if (Flag.getNode())
-    return DAG.getNode(SPUISD::RET_FLAG, dl, MVT::Other, Chain, Flag);
-  else
-    return DAG.getNode(SPUISD::RET_FLAG, dl, MVT::Other, Chain);
-}
-
-
-//===----------------------------------------------------------------------===//
-// Vector related lowering:
-//===----------------------------------------------------------------------===//
-
-static ConstantSDNode *
-getVecImm(SDNode *N) {
-  SDValue OpVal(0, 0);
-
-  // Check to see if this buildvec has a single non-undef value in its elements.
-  for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
-    if (N->getOperand(i).getOpcode() == ISD::UNDEF) continue;
-    if (OpVal.getNode() == 0)
-      OpVal = N->getOperand(i);
-    else if (OpVal != N->getOperand(i))
-      return 0;
-  }
-
-  if (OpVal.getNode() != 0) {
-    if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(OpVal)) {
-      return CN;
-    }
-  }
-
-  return 0;
-}
-
-/// get_vec_i18imm - Test if this vector is a vector filled with the same value
-/// and the value fits into an unsigned 18-bit constant, and if so, return the
-/// constant
-SDValue SPU::get_vec_u18imm(SDNode *N, SelectionDAG &DAG,
-                              EVT ValueType) {
-  if (ConstantSDNode *CN = getVecImm(N)) {
-    uint64_t Value = CN->getZExtValue();
-    if (ValueType == MVT::i64) {
-      uint64_t UValue = CN->getZExtValue();
-      uint32_t upper = uint32_t(UValue >> 32);
-      uint32_t lower = uint32_t(UValue);
-      if (upper != lower)
-        return SDValue();
-      Value = Value >> 32;
-    }
-    if (Value <= 0x3ffff)
-      return DAG.getTargetConstant(Value, ValueType);
-  }
-
-  return SDValue();
-}
-
-/// get_vec_i16imm - Test if this vector is a vector filled with the same value
-/// and the value fits into a signed 16-bit constant, and if so, return the
-/// constant
-SDValue SPU::get_vec_i16imm(SDNode *N, SelectionDAG &DAG,
-                              EVT ValueType) {
-  if (ConstantSDNode *CN = getVecImm(N)) {
-    int64_t Value = CN->getSExtValue();
-    if (ValueType == MVT::i64) {
-      uint64_t UValue = CN->getZExtValue();
-      uint32_t upper = uint32_t(UValue >> 32);
-      uint32_t lower = uint32_t(UValue);
-      if (upper != lower)
-        return SDValue();
-      Value = Value >> 32;
-    }
-    if (Value >= -(1 << 15) && Value <= ((1 << 15) - 1)) {
-      return DAG.getTargetConstant(Value, ValueType);
-    }
-  }
-
-  return SDValue();
-}
-
-/// get_vec_i10imm - Test if this vector is a vector filled with the same value
-/// and the value fits into a signed 10-bit constant, and if so, return the
-/// constant
-SDValue SPU::get_vec_i10imm(SDNode *N, SelectionDAG &DAG,
-                              EVT ValueType) {
-  if (ConstantSDNode *CN = getVecImm(N)) {
-    int64_t Value = CN->getSExtValue();
-    if (ValueType == MVT::i64) {
-      uint64_t UValue = CN->getZExtValue();
-      uint32_t upper = uint32_t(UValue >> 32);
-      uint32_t lower = uint32_t(UValue);
-      if (upper != lower)
-        return SDValue();
-      Value = Value >> 32;
-    }
-    if (isInt<10>(Value))
-      return DAG.getTargetConstant(Value, ValueType);
-  }
-
-  return SDValue();
-}
-
-/// get_vec_i8imm - Test if this vector is a vector filled with the same value
-/// and the value fits into a signed 8-bit constant, and if so, return the
-/// constant.
-///
-/// @note: The incoming vector is v16i8 because that's the only way we can load
-/// constant vectors. Thus, we test to see if the upper and lower bytes are the
-/// same value.
-SDValue SPU::get_vec_i8imm(SDNode *N, SelectionDAG &DAG,
-                             EVT ValueType) {
-  if (ConstantSDNode *CN = getVecImm(N)) {
-    int Value = (int) CN->getZExtValue();
-    if (ValueType == MVT::i16
-        && Value <= 0xffff                 /* truncated from uint64_t */
-        && ((short) Value >> 8) == ((short) Value & 0xff))
-      return DAG.getTargetConstant(Value & 0xff, ValueType);
-    else if (ValueType == MVT::i8
-             && (Value & 0xff) == Value)
-      return DAG.getTargetConstant(Value, ValueType);
-  }
-
-  return SDValue();
-}
-
-/// get_ILHUvec_imm - Test if this vector is a vector filled with the same value
-/// and the value fits into a signed 16-bit constant, and if so, return the
-/// constant
-SDValue SPU::get_ILHUvec_imm(SDNode *N, SelectionDAG &DAG,
-                               EVT ValueType) {
-  if (ConstantSDNode *CN = getVecImm(N)) {
-    uint64_t Value = CN->getZExtValue();
-    if ((ValueType == MVT::i32
-          && ((unsigned) Value & 0xffff0000) == (unsigned) Value)
-        || (ValueType == MVT::i64 && (Value & 0xffff0000) == Value))
-      return DAG.getTargetConstant(Value >> 16, ValueType);
-  }
-
-  return SDValue();
-}
-
-/// get_v4i32_imm - Catch-all for general 32-bit constant vectors
-SDValue SPU::get_v4i32_imm(SDNode *N, SelectionDAG &DAG) {
-  if (ConstantSDNode *CN = getVecImm(N)) {
-    return DAG.getTargetConstant((unsigned) CN->getZExtValue(), MVT::i32);
-  }
-
-  return SDValue();
-}
-
-/// get_v4i32_imm - Catch-all for general 64-bit constant vectors
-SDValue SPU::get_v2i64_imm(SDNode *N, SelectionDAG &DAG) {
-  if (ConstantSDNode *CN = getVecImm(N)) {
-    return DAG.getTargetConstant((unsigned) CN->getZExtValue(), MVT::i64);
-  }
-
-  return SDValue();
-}
-
-//! Lower a BUILD_VECTOR instruction creatively:
-static SDValue
-LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) {
-  EVT VT = Op.getValueType();
-  EVT EltVT = VT.getVectorElementType();
-  DebugLoc dl = Op.getDebugLoc();
-  BuildVectorSDNode *BCN = dyn_cast<BuildVectorSDNode>(Op.getNode());
-  assert(BCN != 0 && "Expected BuildVectorSDNode in SPU LowerBUILD_VECTOR");
-  unsigned minSplatBits = EltVT.getSizeInBits();
-
-  if (minSplatBits < 16)
-    minSplatBits = 16;
-
-  APInt APSplatBits, APSplatUndef;
-  unsigned SplatBitSize;
-  bool HasAnyUndefs;
-
-  if (!BCN->isConstantSplat(APSplatBits, APSplatUndef, SplatBitSize,
-                            HasAnyUndefs, minSplatBits)
-      || minSplatBits < SplatBitSize)
-    return SDValue();   // Wasn't a constant vector or splat exceeded min
-
-  uint64_t SplatBits = APSplatBits.getZExtValue();
-
-  switch (VT.getSimpleVT().SimpleTy) {
-  default:
-    report_fatal_error("CellSPU: Unhandled VT in LowerBUILD_VECTOR, VT = " +
-                       Twine(VT.getEVTString()));
-    /*NOTREACHED*/
-  case MVT::v4f32: {
-    uint32_t Value32 = uint32_t(SplatBits);
-    assert(SplatBitSize == 32
-           && "LowerBUILD_VECTOR: Unexpected floating point vector element.");
-    // NOTE: pretend the constant is an integer. LLVM won't load FP constants
-    SDValue T = DAG.getConstant(Value32, MVT::i32);
-    return DAG.getNode(ISD::BITCAST, dl, MVT::v4f32,
-                       DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, T,T,T,T));
-  }
-  case MVT::v2f64: {
-    uint64_t f64val = uint64_t(SplatBits);
-    assert(SplatBitSize == 64
-           && "LowerBUILD_VECTOR: 64-bit float vector size > 8 bytes.");
-    // NOTE: pretend the constant is an integer. LLVM won't load FP constants
-    SDValue T = DAG.getConstant(f64val, MVT::i64);
-    return DAG.getNode(ISD::BITCAST, dl, MVT::v2f64,
-                       DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v2i64, T, T));
-  }
-  case MVT::v16i8: {
-   // 8-bit constants have to be expanded to 16-bits
-   unsigned short Value16 = SplatBits /* | (SplatBits << 8) */;
-   SmallVector<SDValue, 8> Ops;
-
-   Ops.assign(8, DAG.getConstant(Value16, MVT::i16));
-   return DAG.getNode(ISD::BITCAST, dl, VT,
-                      DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i16, &Ops[0], Ops.size()));
-  }
-  case MVT::v8i16: {
-    unsigned short Value16 = SplatBits;
-    SDValue T = DAG.getConstant(Value16, EltVT);
-    SmallVector<SDValue, 8> Ops;
-
-    Ops.assign(8, T);
-    return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Ops[0], Ops.size());
-  }
-  case MVT::v4i32: {
-    SDValue T = DAG.getConstant(unsigned(SplatBits), VT.getVectorElementType());
-    return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, T, T, T, T);
-  }
-  case MVT::v2i64: {
-    return SPU::LowerV2I64Splat(VT, DAG, SplatBits, dl);
-  }
-  }
-}
-
-/*!
- */
-SDValue
-SPU::LowerV2I64Splat(EVT OpVT, SelectionDAG& DAG, uint64_t SplatVal,
-                     DebugLoc dl) {
-  uint32_t upper = uint32_t(SplatVal >> 32);
-  uint32_t lower = uint32_t(SplatVal);
-
-  if (upper == lower) {
-    // Magic constant that can be matched by IL, ILA, et. al.
-    SDValue Val = DAG.getTargetConstant(upper, MVT::i32);
-    return DAG.getNode(ISD::BITCAST, dl, OpVT,
-                       DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                                   Val, Val, Val, Val));
-  } else {
-    bool upper_special, lower_special;
-
-    // NOTE: This code creates common-case shuffle masks that can be easily
-    // detected as common expressions. It is not attempting to create highly
-    // specialized masks to replace any and all 0's, 0xff's and 0x80's.
-
-    // Detect if the upper or lower half is a special shuffle mask pattern:
-    upper_special = (upper == 0 || upper == 0xffffffff || upper == 0x80000000);
-    lower_special = (lower == 0 || lower == 0xffffffff || lower == 0x80000000);
-
-    // Both upper and lower are special, lower to a constant pool load:
-    if (lower_special && upper_special) {
-      SDValue UpperVal = DAG.getConstant(upper, MVT::i32);
-      SDValue LowerVal = DAG.getConstant(lower, MVT::i32);
-      SDValue BV = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                         UpperVal, LowerVal, UpperVal, LowerVal);
-      return DAG.getNode(ISD::BITCAST, dl, OpVT, BV);
-    }
-
-    SDValue LO32;
-    SDValue HI32;
-    SmallVector<SDValue, 16> ShufBytes;
-    SDValue Result;
-
-    // Create lower vector if not a special pattern
-    if (!lower_special) {
-      SDValue LO32C = DAG.getConstant(lower, MVT::i32);
-      LO32 = DAG.getNode(ISD::BITCAST, dl, OpVT,
-                         DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                                     LO32C, LO32C, LO32C, LO32C));
-    }
-
-    // Create upper vector if not a special pattern
-    if (!upper_special) {
-      SDValue HI32C = DAG.getConstant(upper, MVT::i32);
-      HI32 = DAG.getNode(ISD::BITCAST, dl, OpVT,
-                         DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                                     HI32C, HI32C, HI32C, HI32C));
-    }
-
-    // If either upper or lower are special, then the two input operands are
-    // the same (basically, one of them is a "don't care")
-    if (lower_special)
-      LO32 = HI32;
-    if (upper_special)
-      HI32 = LO32;
-
-    for (int i = 0; i < 4; ++i) {
-      uint64_t val = 0;
-      for (int j = 0; j < 4; ++j) {
-        SDValue V;
-        bool process_upper, process_lower;
-        val <<= 8;
-        process_upper = (upper_special && (i & 1) == 0);
-        process_lower = (lower_special && (i & 1) == 1);
-
-        if (process_upper || process_lower) {
-          if ((process_upper && upper == 0)
-                  || (process_lower && lower == 0))
-            val |= 0x80;
-          else if ((process_upper && upper == 0xffffffff)
-                  || (process_lower && lower == 0xffffffff))
-            val |= 0xc0;
-          else if ((process_upper && upper == 0x80000000)
-                  || (process_lower && lower == 0x80000000))
-            val |= (j == 0 ? 0xe0 : 0x80);
-        } else
-          val |= i * 4 + j + ((i & 1) * 16);
-      }
-
-      ShufBytes.push_back(DAG.getConstant(val, MVT::i32));
-    }
-
-    return DAG.getNode(SPUISD::SHUFB, dl, OpVT, HI32, LO32,
-                       DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                                   &ShufBytes[0], ShufBytes.size()));
-  }
-}
-
-/// LowerVECTOR_SHUFFLE - Lower a vector shuffle (V1, V2, V3) to something on
-/// which the Cell can operate. The code inspects V3 to ascertain whether the
-/// permutation vector, V3, is monotonically increasing with one "exception"
-/// element, e.g., (0, 1, _, 3). If this is the case, then generate a
-/// SHUFFLE_MASK synthetic instruction. Otherwise, spill V3 to the constant pool.
-/// In either case, the net result is going to eventually invoke SHUFB to
-/// permute/shuffle the bytes from V1 and V2.
-/// \note
-/// SHUFFLE_MASK is eventually selected as one of the C*D instructions, generate
-/// control word for byte/halfword/word insertion. This takes care of a single
-/// element move from V2 into V1.
-/// \note
-/// SPUISD::SHUFB is eventually selected as Cell's <i>shufb</i> instructions.
-static SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) {
-  const ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op);
-  SDValue V1 = Op.getOperand(0);
-  SDValue V2 = Op.getOperand(1);
-  DebugLoc dl = Op.getDebugLoc();
-
-  if (V2.getOpcode() == ISD::UNDEF) V2 = V1;
-
-  // If we have a single element being moved from V1 to V2, this can be handled
-  // using the C*[DX] compute mask instructions, but the vector elements have
-  // to be monotonically increasing with one exception element, and the source
-  // slot of the element to move must be the same as the destination.
-  EVT VecVT = V1.getValueType();
-  EVT EltVT = VecVT.getVectorElementType();
-  unsigned EltsFromV2 = 0;
-  unsigned V2EltOffset = 0;
-  unsigned V2EltIdx0 = 0;
-  unsigned CurrElt = 0;
-  unsigned MaxElts = VecVT.getVectorNumElements();
-  unsigned PrevElt = 0;
-  bool monotonic = true;
-  bool rotate = true;
-  int rotamt=0;
-  EVT maskVT;             // which of the c?d instructions to use
-
-  if (EltVT == MVT::i8) {
-    V2EltIdx0 = 16;
-    maskVT = MVT::v16i8;
-  } else if (EltVT == MVT::i16) {
-    V2EltIdx0 = 8;
-    maskVT = MVT::v8i16;
-  } else if (EltVT == MVT::i32 || EltVT == MVT::f32) {
-    V2EltIdx0 = 4;
-    maskVT = MVT::v4i32;
-  } else if (EltVT == MVT::i64 || EltVT == MVT::f64) {
-    V2EltIdx0 = 2;
-    maskVT = MVT::v2i64;
-  } else
-    llvm_unreachable("Unhandled vector type in LowerVECTOR_SHUFFLE");
-
-  for (unsigned i = 0; i != MaxElts; ++i) {
-    if (SVN->getMaskElt(i) < 0)
-      continue;
-
-    unsigned SrcElt = SVN->getMaskElt(i);
-
-    if (monotonic) {
-      if (SrcElt >= V2EltIdx0) {
-        // TODO: optimize for the monotonic case when several consecutive
-        // elements are taken form V2. Do we ever get such a case?
-        if (EltsFromV2 == 0 && CurrElt == (SrcElt - V2EltIdx0))
-          V2EltOffset = (SrcElt - V2EltIdx0) * (EltVT.getSizeInBits()/8);
-        else
-          monotonic = false;
-        ++EltsFromV2;
-      } else if (CurrElt != SrcElt) {
-        monotonic = false;
-      }
-
-      ++CurrElt;
-    }
-
-    if (rotate) {
-      if (PrevElt > 0 && SrcElt < MaxElts) {
-        if ((PrevElt == SrcElt - 1)
-            || (PrevElt == MaxElts - 1 && SrcElt == 0)) {
-          PrevElt = SrcElt;
-        } else {
-          rotate = false;
-        }
-      } else if (i == 0 || (PrevElt==0 && SrcElt==1)) {
-        // First time or after a "wrap around"
-        rotamt = SrcElt-i;
-        PrevElt = SrcElt;
-      } else {
-        // This isn't a rotation, takes elements from vector 2
-        rotate = false;
-      }
-    }
-  }
-
-  if (EltsFromV2 == 1 && monotonic) {
-    // Compute mask and shuffle
-    EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
-
-    // As SHUFFLE_MASK becomes a c?d instruction, feed it an address
-    // R1 ($sp) is used here only as it is guaranteed to have last bits zero
-    SDValue Pointer = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
-                                DAG.getRegister(SPU::R1, PtrVT),
-                                DAG.getConstant(V2EltOffset, MVT::i32));
-    SDValue ShufMaskOp = DAG.getNode(SPUISD::SHUFFLE_MASK, dl,
-                                     maskVT, Pointer);
-
-    // Use shuffle mask in SHUFB synthetic instruction:
-    return DAG.getNode(SPUISD::SHUFB, dl, V1.getValueType(), V2, V1,
-                       ShufMaskOp);
-  } else if (rotate) {
-    if (rotamt < 0)
-      rotamt +=MaxElts;
-    rotamt *= EltVT.getSizeInBits()/8;
-    return DAG.getNode(SPUISD::ROTBYTES_LEFT, dl, V1.getValueType(),
-                       V1, DAG.getConstant(rotamt, MVT::i16));
-  } else {
-   // Convert the SHUFFLE_VECTOR mask's input element units to the
-   // actual bytes.
-    unsigned BytesPerElement = EltVT.getSizeInBits()/8;
-
-    SmallVector<SDValue, 16> ResultMask;
-    for (unsigned i = 0, e = MaxElts; i != e; ++i) {
-      unsigned SrcElt = SVN->getMaskElt(i) < 0 ? 0 : SVN->getMaskElt(i);
-
-      for (unsigned j = 0; j < BytesPerElement; ++j)
-        ResultMask.push_back(DAG.getConstant(SrcElt*BytesPerElement+j,MVT::i8));
-    }
-    SDValue VPermMask = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v16i8,
-                                    &ResultMask[0], ResultMask.size());
-    return DAG.getNode(SPUISD::SHUFB, dl, V1.getValueType(), V1, V2, VPermMask);
-  }
-}
-
-static SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) {
-  SDValue Op0 = Op.getOperand(0);                     // Op0 = the scalar
-  DebugLoc dl = Op.getDebugLoc();
-
-  if (Op0.getNode()->getOpcode() == ISD::Constant) {
-    // For a constant, build the appropriate constant vector, which will
-    // eventually simplify to a vector register load.
-
-    ConstantSDNode *CN = cast<ConstantSDNode>(Op0.getNode());
-    SmallVector<SDValue, 16> ConstVecValues;
-    EVT VT;
-    size_t n_copies;
-
-    // Create a constant vector:
-    switch (Op.getValueType().getSimpleVT().SimpleTy) {
-    default: llvm_unreachable("Unexpected constant value type in "
-                              "LowerSCALAR_TO_VECTOR");
-    case MVT::v16i8: n_copies = 16; VT = MVT::i8; break;
-    case MVT::v8i16: n_copies = 8; VT = MVT::i16; break;
-    case MVT::v4i32: n_copies = 4; VT = MVT::i32; break;
-    case MVT::v4f32: n_copies = 4; VT = MVT::f32; break;
-    case MVT::v2i64: n_copies = 2; VT = MVT::i64; break;
-    case MVT::v2f64: n_copies = 2; VT = MVT::f64; break;
-    }
-
-    SDValue CValue = DAG.getConstant(CN->getZExtValue(), VT);
-    for (size_t j = 0; j < n_copies; ++j)
-      ConstVecValues.push_back(CValue);
-
-    return DAG.getNode(ISD::BUILD_VECTOR, dl, Op.getValueType(),
-                       &ConstVecValues[0], ConstVecValues.size());
-  } else {
-    // Otherwise, copy the value from one register to another:
-    switch (Op0.getValueType().getSimpleVT().SimpleTy) {
-    default: llvm_unreachable("Unexpected value type in LowerSCALAR_TO_VECTOR");
-    case MVT::i8:
-    case MVT::i16:
-    case MVT::i32:
-    case MVT::i64:
-    case MVT::f32:
-    case MVT::f64:
-      return DAG.getNode(SPUISD::PREFSLOT2VEC, dl, Op.getValueType(), Op0, Op0);
-    }
-  }
-}
-
-static SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
-  EVT VT = Op.getValueType();
-  SDValue N = Op.getOperand(0);
-  SDValue Elt = Op.getOperand(1);
-  DebugLoc dl = Op.getDebugLoc();
-  SDValue retval;
-
-  if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Elt)) {
-    // Constant argument:
-    int EltNo = (int) C->getZExtValue();
-
-    // sanity checks:
-    if (VT == MVT::i8 && EltNo >= 16)
-      llvm_unreachable("SPU LowerEXTRACT_VECTOR_ELT: i8 extraction slot > 15");
-    else if (VT == MVT::i16 && EltNo >= 8)
-      llvm_unreachable("SPU LowerEXTRACT_VECTOR_ELT: i16 extraction slot > 7");
-    else if (VT == MVT::i32 && EltNo >= 4)
-      llvm_unreachable("SPU LowerEXTRACT_VECTOR_ELT: i32 extraction slot > 4");
-    else if (VT == MVT::i64 && EltNo >= 2)
-      llvm_unreachable("SPU LowerEXTRACT_VECTOR_ELT: i64 extraction slot > 2");
-
-    if (EltNo == 0 && (VT == MVT::i32 || VT == MVT::i64)) {
-      // i32 and i64: Element 0 is the preferred slot
-      return DAG.getNode(SPUISD::VEC2PREFSLOT, dl, VT, N);
-    }
-
-    // Need to generate shuffle mask and extract:
-    int prefslot_begin = -1, prefslot_end = -1;
-    int elt_byte = EltNo * VT.getSizeInBits() / 8;
-
-    switch (VT.getSimpleVT().SimpleTy) {
-    default: llvm_unreachable("Invalid value type!");
-    case MVT::i8: {
-      prefslot_begin = prefslot_end = 3;
-      break;
-    }
-    case MVT::i16: {
-      prefslot_begin = 2; prefslot_end = 3;
-      break;
-    }
-    case MVT::i32:
-    case MVT::f32: {
-      prefslot_begin = 0; prefslot_end = 3;
-      break;
-    }
-    case MVT::i64:
-    case MVT::f64: {
-      prefslot_begin = 0; prefslot_end = 7;
-      break;
-    }
-    }
-
-    assert(prefslot_begin != -1 && prefslot_end != -1 &&
-           "LowerEXTRACT_VECTOR_ELT: preferred slots uninitialized");
-
-    unsigned int ShufBytes[16] = {
-      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
-    };
-    for (int i = 0; i < 16; ++i) {
-      // zero fill uppper part of preferred slot, don't care about the
-      // other slots:
-      unsigned int mask_val;
-      if (i <= prefslot_end) {
-        mask_val =
-          ((i < prefslot_begin)
-           ? 0x80
-           : elt_byte + (i - prefslot_begin));
-
-        ShufBytes[i] = mask_val;
-      } else
-        ShufBytes[i] = ShufBytes[i % (prefslot_end + 1)];
-    }
-
-    SDValue ShufMask[4];
-    for (unsigned i = 0; i < sizeof(ShufMask)/sizeof(ShufMask[0]); ++i) {
-      unsigned bidx = i * 4;
-      unsigned int bits = ((ShufBytes[bidx] << 24) |
-                           (ShufBytes[bidx+1] << 16) |
-                           (ShufBytes[bidx+2] << 8) |
-                           ShufBytes[bidx+3]);
-      ShufMask[i] = DAG.getConstant(bits, MVT::i32);
-    }
-
-    SDValue ShufMaskVec =
-      DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                  &ShufMask[0], sizeof(ShufMask)/sizeof(ShufMask[0]));
-
-    retval = DAG.getNode(SPUISD::VEC2PREFSLOT, dl, VT,
-                         DAG.getNode(SPUISD::SHUFB, dl, N.getValueType(),
-                                     N, N, ShufMaskVec));
-  } else {
-    // Variable index: Rotate the requested element into slot 0, then replicate
-    // slot 0 across the vector
-    EVT VecVT = N.getValueType();
-    if (!VecVT.isSimple() || !VecVT.isVector()) {
-      report_fatal_error("LowerEXTRACT_VECTOR_ELT: Must have a simple, 128-bit"
-                        "vector type!");
-    }
-
-    // Make life easier by making sure the index is zero-extended to i32
-    if (Elt.getValueType() != MVT::i32)
-      Elt = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Elt);
-
-    // Scale the index to a bit/byte shift quantity
-    APInt scaleFactor =
-            APInt(32, uint64_t(16 / N.getValueType().getVectorNumElements()), false);
-    unsigned scaleShift = scaleFactor.logBase2();
-    SDValue vecShift;
-
-    if (scaleShift > 0) {
-      // Scale the shift factor:
-      Elt = DAG.getNode(ISD::SHL, dl, MVT::i32, Elt,
-                        DAG.getConstant(scaleShift, MVT::i32));
-    }
-
-    vecShift = DAG.getNode(SPUISD::SHL_BYTES, dl, VecVT, N, Elt);
-
-    // Replicate the bytes starting at byte 0 across the entire vector (for
-    // consistency with the notion of a unified register set)
-    SDValue replicate;
-
-    switch (VT.getSimpleVT().SimpleTy) {
-    default:
-      report_fatal_error("LowerEXTRACT_VECTOR_ELT(varable): Unhandled vector"
-                        "type");
-      /*NOTREACHED*/
-    case MVT::i8: {
-      SDValue factor = DAG.getConstant(0x00000000, MVT::i32);
-      replicate = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                              factor, factor, factor, factor);
-      break;
-    }
-    case MVT::i16: {
-      SDValue factor = DAG.getConstant(0x00010001, MVT::i32);
-      replicate = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                              factor, factor, factor, factor);
-      break;
-    }
-    case MVT::i32:
-    case MVT::f32: {
-      SDValue factor = DAG.getConstant(0x00010203, MVT::i32);
-      replicate = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                              factor, factor, factor, factor);
-      break;
-    }
-    case MVT::i64:
-    case MVT::f64: {
-      SDValue loFactor = DAG.getConstant(0x00010203, MVT::i32);
-      SDValue hiFactor = DAG.getConstant(0x04050607, MVT::i32);
-      replicate = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                              loFactor, hiFactor, loFactor, hiFactor);
-      break;
-    }
-    }
-
-    retval = DAG.getNode(SPUISD::VEC2PREFSLOT, dl, VT,
-                         DAG.getNode(SPUISD::SHUFB, dl, VecVT,
-                                     vecShift, vecShift, replicate));
-  }
-
-  return retval;
-}
-
-static SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
-  SDValue VecOp = Op.getOperand(0);
-  SDValue ValOp = Op.getOperand(1);
-  SDValue IdxOp = Op.getOperand(2);
-  DebugLoc dl = Op.getDebugLoc();
-  EVT VT = Op.getValueType();
-  EVT eltVT = ValOp.getValueType();
-
-  // use 0 when the lane to insert to is 'undef'
-  int64_t Offset=0;
-  if (IdxOp.getOpcode() != ISD::UNDEF) {
-    ConstantSDNode *CN = cast<ConstantSDNode>(IdxOp);
-    assert(CN != 0 && "LowerINSERT_VECTOR_ELT: Index is not constant!");
-    Offset = (CN->getSExtValue()) * eltVT.getSizeInBits()/8;
-  }
-
-  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
-  // Use $sp ($1) because it's always 16-byte aligned and it's available:
-  SDValue Pointer = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
-                                DAG.getRegister(SPU::R1, PtrVT),
-                                DAG.getConstant(Offset, PtrVT));
-  // widen the mask when dealing with half vectors
-  EVT maskVT = EVT::getVectorVT(*(DAG.getContext()), VT.getVectorElementType(),
-                                128/ VT.getVectorElementType().getSizeInBits());
-  SDValue ShufMask = DAG.getNode(SPUISD::SHUFFLE_MASK, dl, maskVT, Pointer);
-
-  SDValue result =
-    DAG.getNode(SPUISD::SHUFB, dl, VT,
-                DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, ValOp),
-                VecOp,
-                DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, ShufMask));
-
-  return result;
-}
-
-static SDValue LowerI8Math(SDValue Op, SelectionDAG &DAG, unsigned Opc,
-                           const TargetLowering &TLI)
-{
-  SDValue N0 = Op.getOperand(0);      // Everything has at least one operand
-  DebugLoc dl = Op.getDebugLoc();
-  EVT ShiftVT = TLI.getShiftAmountTy(N0.getValueType());
-
-  assert(Op.getValueType() == MVT::i8);
-  switch (Opc) {
-  default:
-    llvm_unreachable("Unhandled i8 math operator");
-  case ISD::ADD: {
-    // 8-bit addition: Promote the arguments up to 16-bits and truncate
-    // the result:
-    SDValue N1 = Op.getOperand(1);
-    N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N0);
-    N1 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N1);
-    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
-                       DAG.getNode(Opc, dl, MVT::i16, N0, N1));
-
-  }
-
-  case ISD::SUB: {
-    // 8-bit subtraction: Promote the arguments up to 16-bits and truncate
-    // the result:
-    SDValue N1 = Op.getOperand(1);
-    N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N0);
-    N1 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N1);
-    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
-                       DAG.getNode(Opc, dl, MVT::i16, N0, N1));
-  }
-  case ISD::ROTR:
-  case ISD::ROTL: {
-    SDValue N1 = Op.getOperand(1);
-    EVT N1VT = N1.getValueType();
-
-    N0 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i16, N0);
-    if (!N1VT.bitsEq(ShiftVT)) {
-      unsigned N1Opc = N1.getValueType().bitsLT(ShiftVT)
-                       ? ISD::ZERO_EXTEND
-                       : ISD::TRUNCATE;
-      N1 = DAG.getNode(N1Opc, dl, ShiftVT, N1);
-    }
-
-    // Replicate lower 8-bits into upper 8:
-    SDValue ExpandArg =
-      DAG.getNode(ISD::OR, dl, MVT::i16, N0,
-                  DAG.getNode(ISD::SHL, dl, MVT::i16,
-                              N0, DAG.getConstant(8, MVT::i32)));
-
-    // Truncate back down to i8
-    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
-                       DAG.getNode(Opc, dl, MVT::i16, ExpandArg, N1));
-  }
-  case ISD::SRL:
-  case ISD::SHL: {
-    SDValue N1 = Op.getOperand(1);
-    EVT N1VT = N1.getValueType();
-
-    N0 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i16, N0);
-    if (!N1VT.bitsEq(ShiftVT)) {
-      unsigned N1Opc = ISD::ZERO_EXTEND;
-
-      if (N1.getValueType().bitsGT(ShiftVT))
-        N1Opc = ISD::TRUNCATE;
-
-      N1 = DAG.getNode(N1Opc, dl, ShiftVT, N1);
-    }
-
-    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
-                       DAG.getNode(Opc, dl, MVT::i16, N0, N1));
-  }
-  case ISD::SRA: {
-    SDValue N1 = Op.getOperand(1);
-    EVT N1VT = N1.getValueType();
-
-    N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N0);
-    if (!N1VT.bitsEq(ShiftVT)) {
-      unsigned N1Opc = ISD::SIGN_EXTEND;
-
-      if (N1VT.bitsGT(ShiftVT))
-        N1Opc = ISD::TRUNCATE;
-      N1 = DAG.getNode(N1Opc, dl, ShiftVT, N1);
-    }
-
-    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
-                       DAG.getNode(Opc, dl, MVT::i16, N0, N1));
-  }
-  case ISD::MUL: {
-    SDValue N1 = Op.getOperand(1);
-
-    N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N0);
-    N1 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N1);
-    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
-                       DAG.getNode(Opc, dl, MVT::i16, N0, N1));
-  }
-  }
-}
-
-//! Lower byte immediate operations for v16i8 vectors:
-static SDValue
-LowerByteImmed(SDValue Op, SelectionDAG &DAG) {
-  SDValue ConstVec;
-  SDValue Arg;
-  EVT VT = Op.getValueType();
-  DebugLoc dl = Op.getDebugLoc();
-
-  ConstVec = Op.getOperand(0);
-  Arg = Op.getOperand(1);
-  if (ConstVec.getNode()->getOpcode() != ISD::BUILD_VECTOR) {
-    if (ConstVec.getNode()->getOpcode() == ISD::BITCAST) {
-      ConstVec = ConstVec.getOperand(0);
-    } else {
-      ConstVec = Op.getOperand(1);
-      Arg = Op.getOperand(0);
-      if (ConstVec.getNode()->getOpcode() == ISD::BITCAST) {
-        ConstVec = ConstVec.getOperand(0);
-      }
-    }
-  }
-
-  if (ConstVec.getNode()->getOpcode() == ISD::BUILD_VECTOR) {
-    BuildVectorSDNode *BCN = dyn_cast<BuildVectorSDNode>(ConstVec.getNode());
-    assert(BCN != 0 && "Expected BuildVectorSDNode in SPU LowerByteImmed");
-
-    APInt APSplatBits, APSplatUndef;
-    unsigned SplatBitSize;
-    bool HasAnyUndefs;
-    unsigned minSplatBits = VT.getVectorElementType().getSizeInBits();
-
-    if (BCN->isConstantSplat(APSplatBits, APSplatUndef, SplatBitSize,
-                              HasAnyUndefs, minSplatBits)
-        && minSplatBits <= SplatBitSize) {
-      uint64_t SplatBits = APSplatBits.getZExtValue();
-      SDValue tc = DAG.getTargetConstant(SplatBits & 0xff, MVT::i8);
-
-      SmallVector<SDValue, 16> tcVec;
-      tcVec.assign(16, tc);
-      return DAG.getNode(Op.getNode()->getOpcode(), dl, VT, Arg,
-                         DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &tcVec[0], tcVec.size()));
-    }
-  }
-
-  // These operations (AND, OR, XOR) are legal, they just couldn't be custom
-  // lowered.  Return the operation, rather than a null SDValue.
-  return Op;
-}
-
-//! Custom lowering for CTPOP (count population)
-/*!
-  Custom lowering code that counts the number ones in the input
-  operand. SPU has such an instruction, but it counts the number of
-  ones per byte, which then have to be accumulated.
-*/
-static SDValue LowerCTPOP(SDValue Op, SelectionDAG &DAG) {
-  EVT VT = Op.getValueType();
-  EVT vecVT = EVT::getVectorVT(*DAG.getContext(),
-                               VT, (128 / VT.getSizeInBits()));
-  DebugLoc dl = Op.getDebugLoc();
-
-  switch (VT.getSimpleVT().SimpleTy) {
-  default: llvm_unreachable("Invalid value type!");
-  case MVT::i8: {
-    SDValue N = Op.getOperand(0);
-    SDValue Elt0 = DAG.getConstant(0, MVT::i32);
-
-    SDValue Promote = DAG.getNode(SPUISD::PREFSLOT2VEC, dl, vecVT, N, N);
-    SDValue CNTB = DAG.getNode(SPUISD::CNTB, dl, vecVT, Promote);
-
-    return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i8, CNTB, Elt0);
-  }
-
-  case MVT::i16: {
-    MachineFunction &MF = DAG.getMachineFunction();
-    MachineRegisterInfo &RegInfo = MF.getRegInfo();
-
-    unsigned CNTB_reg = RegInfo.createVirtualRegister(&SPU::R16CRegClass);
-
-    SDValue N = Op.getOperand(0);
-    SDValue Elt0 = DAG.getConstant(0, MVT::i16);
-    SDValue Mask0 = DAG.getConstant(0x0f, MVT::i16);
-    SDValue Shift1 = DAG.getConstant(8, MVT::i32);
-
-    SDValue Promote = DAG.getNode(SPUISD::PREFSLOT2VEC, dl, vecVT, N, N);
-    SDValue CNTB = DAG.getNode(SPUISD::CNTB, dl, vecVT, Promote);
-
-    // CNTB_result becomes the chain to which all of the virtual registers
-    // CNTB_reg, SUM1_reg become associated:
-    SDValue CNTB_result =
-      DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i16, CNTB, Elt0);
-
-    SDValue CNTB_rescopy =
-      DAG.getCopyToReg(CNTB_result, dl, CNTB_reg, CNTB_result);
-
-    SDValue Tmp1 = DAG.getCopyFromReg(CNTB_rescopy, dl, CNTB_reg, MVT::i16);
-
-    return DAG.getNode(ISD::AND, dl, MVT::i16,
-                       DAG.getNode(ISD::ADD, dl, MVT::i16,
-                                   DAG.getNode(ISD::SRL, dl, MVT::i16,
-                                               Tmp1, Shift1),
-                                   Tmp1),
-                       Mask0);
-  }
-
-  case MVT::i32: {
-    MachineFunction &MF = DAG.getMachineFunction();
-    MachineRegisterInfo &RegInfo = MF.getRegInfo();
-
-    unsigned CNTB_reg = RegInfo.createVirtualRegister(&SPU::R32CRegClass);
-    unsigned SUM1_reg = RegInfo.createVirtualRegister(&SPU::R32CRegClass);
-
-    SDValue N = Op.getOperand(0);
-    SDValue Elt0 = DAG.getConstant(0, MVT::i32);
-    SDValue Mask0 = DAG.getConstant(0xff, MVT::i32);
-    SDValue Shift1 = DAG.getConstant(16, MVT::i32);
-    SDValue Shift2 = DAG.getConstant(8, MVT::i32);
-
-    SDValue Promote = DAG.getNode(SPUISD::PREFSLOT2VEC, dl, vecVT, N, N);
-    SDValue CNTB = DAG.getNode(SPUISD::CNTB, dl, vecVT, Promote);
-
-    // CNTB_result becomes the chain to which all of the virtual registers
-    // CNTB_reg, SUM1_reg become associated:
-    SDValue CNTB_result =
-      DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i32, CNTB, Elt0);
-
-    SDValue CNTB_rescopy =
-      DAG.getCopyToReg(CNTB_result, dl, CNTB_reg, CNTB_result);
-
-    SDValue Comp1 =
-      DAG.getNode(ISD::SRL, dl, MVT::i32,
-                  DAG.getCopyFromReg(CNTB_rescopy, dl, CNTB_reg, MVT::i32),
-                  Shift1);
-
-    SDValue Sum1 =
-      DAG.getNode(ISD::ADD, dl, MVT::i32, Comp1,
-                  DAG.getCopyFromReg(CNTB_rescopy, dl, CNTB_reg, MVT::i32));
-
-    SDValue Sum1_rescopy =
-      DAG.getCopyToReg(CNTB_result, dl, SUM1_reg, Sum1);
-
-    SDValue Comp2 =
-      DAG.getNode(ISD::SRL, dl, MVT::i32,
-                  DAG.getCopyFromReg(Sum1_rescopy, dl, SUM1_reg, MVT::i32),
-                  Shift2);
-    SDValue Sum2 =
-      DAG.getNode(ISD::ADD, dl, MVT::i32, Comp2,
-                  DAG.getCopyFromReg(Sum1_rescopy, dl, SUM1_reg, MVT::i32));
-
-    return DAG.getNode(ISD::AND, dl, MVT::i32, Sum2, Mask0);
-  }
-
-  case MVT::i64:
-    break;
-  }
-
-  return SDValue();
-}
-
-//! Lower ISD::FP_TO_SINT, ISD::FP_TO_UINT for i32
-/*!
- f32->i32 passes through unchanged, whereas f64->i32 expands to a libcall.
- All conversions to i64 are expanded to a libcall.
- */
-static SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG,
-                              const SPUTargetLowering &TLI) {
-  EVT OpVT = Op.getValueType();
-  SDValue Op0 = Op.getOperand(0);
-  EVT Op0VT = Op0.getValueType();
-
-  if ((OpVT == MVT::i32 && Op0VT == MVT::f64)
-      || OpVT == MVT::i64) {
-    // Convert f32 / f64 to i32 / i64 via libcall.
-    RTLIB::Libcall LC =
-            (Op.getOpcode() == ISD::FP_TO_SINT)
-             ? RTLIB::getFPTOSINT(Op0VT, OpVT)
-             : RTLIB::getFPTOUINT(Op0VT, OpVT);
-    assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unexpectd fp-to-int conversion!");
-    SDValue Dummy;
-    return ExpandLibCall(LC, Op, DAG, false, Dummy, TLI);
-  }
-
-  return Op;
-}
-
-//! Lower ISD::SINT_TO_FP, ISD::UINT_TO_FP for i32
-/*!
- i32->f32 passes through unchanged, whereas i32->f64 is expanded to a libcall.
- All conversions from i64 are expanded to a libcall.
- */
-static SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG,
-                              const SPUTargetLowering &TLI) {
-  EVT OpVT = Op.getValueType();
-  SDValue Op0 = Op.getOperand(0);
-  EVT Op0VT = Op0.getValueType();
-
-  if ((OpVT == MVT::f64 && Op0VT == MVT::i32)
-      || Op0VT == MVT::i64) {
-    // Convert i32, i64 to f64 via libcall:
-    RTLIB::Libcall LC =
-            (Op.getOpcode() == ISD::SINT_TO_FP)
-             ? RTLIB::getSINTTOFP(Op0VT, OpVT)
-             : RTLIB::getUINTTOFP(Op0VT, OpVT);
-    assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unexpectd int-to-fp conversion!");
-    SDValue Dummy;
-    return ExpandLibCall(LC, Op, DAG, false, Dummy, TLI);
-  }
-
-  return Op;
-}
-
-//! Lower ISD::SETCC
-/*!
- This handles MVT::f64 (double floating point) condition lowering
- */
-static SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG,
-                          const TargetLowering &TLI) {
-  CondCodeSDNode *CC = dyn_cast<CondCodeSDNode>(Op.getOperand(2));
-  DebugLoc dl = Op.getDebugLoc();
-  assert(CC != 0 && "LowerSETCC: CondCodeSDNode should not be null here!\n");
-
-  SDValue lhs = Op.getOperand(0);
-  SDValue rhs = Op.getOperand(1);
-  EVT lhsVT = lhs.getValueType();
-  assert(lhsVT == MVT::f64 && "LowerSETCC: type other than MVT::64\n");
-
-  EVT ccResultVT = TLI.getSetCCResultType(lhs.getValueType());
-  APInt ccResultOnes = APInt::getAllOnesValue(ccResultVT.getSizeInBits());
-  EVT IntVT(MVT::i64);
-
-  // Take advantage of the fact that (truncate (sra arg, 32)) is efficiently
-  // selected to a NOP:
-  SDValue i64lhs = DAG.getNode(ISD::BITCAST, dl, IntVT, lhs);
-  SDValue lhsHi32 =
-          DAG.getNode(ISD::TRUNCATE, dl, MVT::i32,
-                      DAG.getNode(ISD::SRL, dl, IntVT,
-                                  i64lhs, DAG.getConstant(32, MVT::i32)));
-  SDValue lhsHi32abs =
-          DAG.getNode(ISD::AND, dl, MVT::i32,
-                      lhsHi32, DAG.getConstant(0x7fffffff, MVT::i32));
-  SDValue lhsLo32 =
-          DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, i64lhs);
-
-  // SETO and SETUO only use the lhs operand:
-  if (CC->get() == ISD::SETO) {
-    // Evaluates to true if Op0 is not [SQ]NaN - lowers to the inverse of
-    // SETUO
-    APInt ccResultAllOnes = APInt::getAllOnesValue(ccResultVT.getSizeInBits());
-    return DAG.getNode(ISD::XOR, dl, ccResultVT,
-                       DAG.getSetCC(dl, ccResultVT,
-                                    lhs, DAG.getConstantFP(0.0, lhsVT),
-                                    ISD::SETUO),
-                       DAG.getConstant(ccResultAllOnes, ccResultVT));
-  } else if (CC->get() == ISD::SETUO) {
-    // Evaluates to true if Op0 is [SQ]NaN
-    return DAG.getNode(ISD::AND, dl, ccResultVT,
-                       DAG.getSetCC(dl, ccResultVT,
-                                    lhsHi32abs,
-                                    DAG.getConstant(0x7ff00000, MVT::i32),
-                                    ISD::SETGE),
-                       DAG.getSetCC(dl, ccResultVT,
-                                    lhsLo32,
-                                    DAG.getConstant(0, MVT::i32),
-                                    ISD::SETGT));
-  }
-
-  SDValue i64rhs = DAG.getNode(ISD::BITCAST, dl, IntVT, rhs);
-  SDValue rhsHi32 =
-          DAG.getNode(ISD::TRUNCATE, dl, MVT::i32,
-                      DAG.getNode(ISD::SRL, dl, IntVT,
-                                  i64rhs, DAG.getConstant(32, MVT::i32)));
-
-  // If a value is negative, subtract from the sign magnitude constant:
-  SDValue signMag2TC = DAG.getConstant(0x8000000000000000ULL, IntVT);
-
-  // Convert the sign-magnitude representation into 2's complement:
-  SDValue lhsSelectMask = DAG.getNode(ISD::SRA, dl, ccResultVT,
-                                      lhsHi32, DAG.getConstant(31, MVT::i32));
-  SDValue lhsSignMag2TC = DAG.getNode(ISD::SUB, dl, IntVT, signMag2TC, i64lhs);
-  SDValue lhsSelect =
-          DAG.getNode(ISD::SELECT, dl, IntVT,
-                      lhsSelectMask, lhsSignMag2TC, i64lhs);
-
-  SDValue rhsSelectMask = DAG.getNode(ISD::SRA, dl, ccResultVT,
-                                      rhsHi32, DAG.getConstant(31, MVT::i32));
-  SDValue rhsSignMag2TC = DAG.getNode(ISD::SUB, dl, IntVT, signMag2TC, i64rhs);
-  SDValue rhsSelect =
-          DAG.getNode(ISD::SELECT, dl, IntVT,
-                      rhsSelectMask, rhsSignMag2TC, i64rhs);
-
-  unsigned compareOp;
-
-  switch (CC->get()) {
-  case ISD::SETOEQ:
-  case ISD::SETUEQ:
-    compareOp = ISD::SETEQ; break;
-  case ISD::SETOGT:
-  case ISD::SETUGT:
-    compareOp = ISD::SETGT; break;
-  case ISD::SETOGE:
-  case ISD::SETUGE:
-    compareOp = ISD::SETGE; break;
-  case ISD::SETOLT:
-  case ISD::SETULT:
-    compareOp = ISD::SETLT; break;
-  case ISD::SETOLE:
-  case ISD::SETULE:
-    compareOp = ISD::SETLE; break;
-  case ISD::SETUNE:
-  case ISD::SETONE:
-    compareOp = ISD::SETNE; break;
-  default:
-    report_fatal_error("CellSPU ISel Select: unimplemented f64 condition");
-  }
-
-  SDValue result =
-          DAG.getSetCC(dl, ccResultVT, lhsSelect, rhsSelect,
-                       (ISD::CondCode) compareOp);
-
-  if ((CC->get() & 0x8) == 0) {
-    // Ordered comparison:
-    SDValue lhsNaN = DAG.getSetCC(dl, ccResultVT,
-                                  lhs, DAG.getConstantFP(0.0, MVT::f64),
-                                  ISD::SETO);
-    SDValue rhsNaN = DAG.getSetCC(dl, ccResultVT,
-                                  rhs, DAG.getConstantFP(0.0, MVT::f64),
-                                  ISD::SETO);
-    SDValue ordered = DAG.getNode(ISD::AND, dl, ccResultVT, lhsNaN, rhsNaN);
-
-    result = DAG.getNode(ISD::AND, dl, ccResultVT, ordered, result);
-  }
-
-  return result;
-}
-
-//! Lower ISD::SELECT_CC
-/*!
-  ISD::SELECT_CC can (generally) be implemented directly on the SPU using the
-  SELB instruction.
-
-  \note Need to revisit this in the future: if the code path through the true
-  and false value computations is longer than the latency of a branch (6
-  cycles), then it would be more advantageous to branch and insert a new basic
-  block and branch on the condition. However, this code does not make that
-  assumption, given the simplisitc uses so far.
- */
-
-static SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG,
-                              const TargetLowering &TLI) {
-  EVT VT = Op.getValueType();
-  SDValue lhs = Op.getOperand(0);
-  SDValue rhs = Op.getOperand(1);
-  SDValue trueval = Op.getOperand(2);
-  SDValue falseval = Op.getOperand(3);
-  SDValue condition = Op.getOperand(4);
-  DebugLoc dl = Op.getDebugLoc();
-
-  // NOTE: SELB's arguments: $rA, $rB, $mask
-  //
-  // SELB selects bits from $rA where bits in $mask are 0, bits from $rB
-  // where bits in $mask are 1. CCond will be inverted, having 1s where the
-  // condition was true and 0s where the condition was false. Hence, the
-  // arguments to SELB get reversed.
-
-  // Note: Really should be ISD::SELECT instead of SPUISD::SELB, but LLVM's
-  // legalizer insists on combining SETCC/SELECT into SELECT_CC, so we end up
-  // with another "cannot select select_cc" assert:
-
-  SDValue compare = DAG.getNode(ISD::SETCC, dl,
-                                TLI.getSetCCResultType(Op.getValueType()),
-                                lhs, rhs, condition);
-  return DAG.getNode(SPUISD::SELB, dl, VT, falseval, trueval, compare);
-}
-
-//! Custom lower ISD::TRUNCATE
-static SDValue LowerTRUNCATE(SDValue Op, SelectionDAG &DAG)
-{
-  // Type to truncate to
-  EVT VT = Op.getValueType();
-  MVT simpleVT = VT.getSimpleVT();
-  EVT VecVT = EVT::getVectorVT(*DAG.getContext(),
-                               VT, (128 / VT.getSizeInBits()));
-  DebugLoc dl = Op.getDebugLoc();
-
-  // Type to truncate from
-  SDValue Op0 = Op.getOperand(0);
-  EVT Op0VT = Op0.getValueType();
-
-  if (Op0VT == MVT::i128 && simpleVT == MVT::i64) {
-    // Create shuffle mask, least significant doubleword of quadword
-    unsigned maskHigh = 0x08090a0b;
-    unsigned maskLow = 0x0c0d0e0f;
-    // Use a shuffle to perform the truncation
-    SDValue shufMask = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                                   DAG.getConstant(maskHigh, MVT::i32),
-                                   DAG.getConstant(maskLow, MVT::i32),
-                                   DAG.getConstant(maskHigh, MVT::i32),
-                                   DAG.getConstant(maskLow, MVT::i32));
-
-    SDValue truncShuffle = DAG.getNode(SPUISD::SHUFB, dl, VecVT,
-                                       Op0, Op0, shufMask);
-
-    return DAG.getNode(SPUISD::VEC2PREFSLOT, dl, VT, truncShuffle);
-  }
-
-  return SDValue();             // Leave the truncate unmolested
-}
-
-/*!
- * Emit the instruction sequence for i64/i32 -> i128 sign extend. The basic
- * algorithm is to duplicate the sign bit using rotmai to generate at
- * least one byte full of sign bits. Then propagate the "sign-byte" into
- * the leftmost words and the i64/i32 into the rightmost words using shufb.
- *
- * @param Op The sext operand
- * @param DAG The current DAG
- * @return The SDValue with the entire instruction sequence
- */
-static SDValue LowerSIGN_EXTEND(SDValue Op, SelectionDAG &DAG)
-{
-  DebugLoc dl = Op.getDebugLoc();
-
-  // Type to extend to
-  MVT OpVT = Op.getValueType().getSimpleVT();
-
-  // Type to extend from
-  SDValue Op0 = Op.getOperand(0);
-  MVT Op0VT = Op0.getValueType().getSimpleVT();
-
-  // extend i8 & i16 via i32
-  if (Op0VT == MVT::i8 || Op0VT == MVT::i16) {
-    Op0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i32, Op0);
-    Op0VT = MVT::i32;
-  }
-
-  // The type to extend to needs to be a i128 and
-  // the type to extend from needs to be i64 or i32.
-  assert((OpVT == MVT::i128 && (Op0VT == MVT::i64 || Op0VT == MVT::i32)) &&
-          "LowerSIGN_EXTEND: input and/or output operand have wrong size");
-  (void)OpVT;
-
-  // Create shuffle mask
-  unsigned mask1 = 0x10101010; // byte 0 - 3 and 4 - 7
-  unsigned mask2 = Op0VT == MVT::i64 ? 0x00010203 : 0x10101010; // byte  8 - 11
-  unsigned mask3 = Op0VT == MVT::i64 ? 0x04050607 : 0x00010203; // byte 12 - 15
-  SDValue shufMask = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
-                                 DAG.getConstant(mask1, MVT::i32),
-                                 DAG.getConstant(mask1, MVT::i32),
-                                 DAG.getConstant(mask2, MVT::i32),
-                                 DAG.getConstant(mask3, MVT::i32));
-
-  // Word wise arithmetic right shift to generate at least one byte
-  // that contains sign bits.
-  MVT mvt = Op0VT == MVT::i64 ? MVT::v2i64 : MVT::v4i32;
-  SDValue sraVal = DAG.getNode(ISD::SRA,
-                 dl,
-                 mvt,
-                 DAG.getNode(SPUISD::PREFSLOT2VEC, dl, mvt, Op0, Op0),
-                 DAG.getConstant(31, MVT::i32));
-
-  // reinterpret as a i128 (SHUFB requires it). This gets lowered away.
-  SDValue extended = SDValue(DAG.getMachineNode(TargetOpcode::COPY_TO_REGCLASS,
-                                        dl, Op0VT, Op0,
-                                        DAG.getTargetConstant(
-                                                  SPU::GPRCRegClass.getID(),
-                                                  MVT::i32)), 0);
-  // Shuffle bytes - Copy the sign bits into the upper 64 bits
-  // and the input value into the lower 64 bits.
-  SDValue extShuffle = DAG.getNode(SPUISD::SHUFB, dl, mvt,
-        extended, sraVal, shufMask);
-  return DAG.getNode(ISD::BITCAST, dl, MVT::i128, extShuffle);
-}
-
-//! Custom (target-specific) lowering entry point
-/*!
-  This is where LLVM's DAG selection process calls to do target-specific
-  lowering of nodes.
- */
-SDValue
-SPUTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const
-{
-  unsigned Opc = (unsigned) Op.getOpcode();
-  EVT VT = Op.getValueType();
-
-  switch (Opc) {
-  default: {
-#ifndef NDEBUG
-    errs() << "SPUTargetLowering::LowerOperation(): need to lower this!\n";
-    errs() << "Op.getOpcode() = " << Opc << "\n";
-    errs() << "*Op.getNode():\n";
-    Op.getNode()->dump();
-#endif
-    llvm_unreachable(0);
-  }
-  case ISD::LOAD:
-  case ISD::EXTLOAD:
-  case ISD::SEXTLOAD:
-  case ISD::ZEXTLOAD:
-    return LowerLOAD(Op, DAG, SPUTM.getSubtargetImpl());
-  case ISD::STORE:
-    return LowerSTORE(Op, DAG, SPUTM.getSubtargetImpl());
-  case ISD::ConstantPool:
-    return LowerConstantPool(Op, DAG, SPUTM.getSubtargetImpl());
-  case ISD::GlobalAddress:
-    return LowerGlobalAddress(Op, DAG, SPUTM.getSubtargetImpl());
-  case ISD::JumpTable:
-    return LowerJumpTable(Op, DAG, SPUTM.getSubtargetImpl());
-  case ISD::ConstantFP:
-    return LowerConstantFP(Op, DAG);
-
-  // i8, i64 math ops:
-  case ISD::ADD:
-  case ISD::SUB:
-  case ISD::ROTR:
-  case ISD::ROTL:
-  case ISD::SRL:
-  case ISD::SHL:
-  case ISD::SRA: {
-    if (VT == MVT::i8)
-      return LowerI8Math(Op, DAG, Opc, *this);
-    break;
-  }
-
-  case ISD::FP_TO_SINT:
-  case ISD::FP_TO_UINT:
-    return LowerFP_TO_INT(Op, DAG, *this);
-
-  case ISD::SINT_TO_FP:
-  case ISD::UINT_TO_FP:
-    return LowerINT_TO_FP(Op, DAG, *this);
-
-  // Vector-related lowering.
-  case ISD::BUILD_VECTOR:
-    return LowerBUILD_VECTOR(Op, DAG);
-  case ISD::SCALAR_TO_VECTOR:
-    return LowerSCALAR_TO_VECTOR(Op, DAG);
-  case ISD::VECTOR_SHUFFLE:
-    return LowerVECTOR_SHUFFLE(Op, DAG);
-  case ISD::EXTRACT_VECTOR_ELT:
-    return LowerEXTRACT_VECTOR_ELT(Op, DAG);
-  case ISD::INSERT_VECTOR_ELT:
-    return LowerINSERT_VECTOR_ELT(Op, DAG);
-
-  // Look for ANDBI, ORBI and XORBI opportunities and lower appropriately:
-  case ISD::AND:
-  case ISD::OR:
-  case ISD::XOR:
-    return LowerByteImmed(Op, DAG);
-
-  // Vector and i8 multiply:
-  case ISD::MUL:
-    if (VT == MVT::i8)
-      return LowerI8Math(Op, DAG, Opc, *this);
-
-  case ISD::CTPOP:
-    return LowerCTPOP(Op, DAG);
-
-  case ISD::SELECT_CC:
-    return LowerSELECT_CC(Op, DAG, *this);
-
-  case ISD::SETCC:
-    return LowerSETCC(Op, DAG, *this);
-
-  case ISD::TRUNCATE:
-    return LowerTRUNCATE(Op, DAG);
-
-  case ISD::SIGN_EXTEND:
-    return LowerSIGN_EXTEND(Op, DAG);
-  }
-
-  return SDValue();
-}
-
-void SPUTargetLowering::ReplaceNodeResults(SDNode *N,
-                                           SmallVectorImpl<SDValue>&Results,
-                                           SelectionDAG &DAG) const
-{
-#if 0
-  unsigned Opc = (unsigned) N->getOpcode();
-  EVT OpVT = N->getValueType(0);
-
-  switch (Opc) {
-  default: {
-    errs() << "SPUTargetLowering::ReplaceNodeResults(): need to fix this!\n";
-    errs() << "Op.getOpcode() = " << Opc << "\n";
-    errs() << "*Op.getNode():\n";
-    N->dump();
-    abort();
-    /*NOTREACHED*/
-  }
-  }
-#endif
-
-  /* Otherwise, return unchanged */
-}
-
-//===----------------------------------------------------------------------===//
-// Target Optimization Hooks
-//===----------------------------------------------------------------------===//
-
-SDValue
-SPUTargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const
-{
-#if 0
-  TargetMachine &TM = getTargetMachine();
-#endif
-  const SPUSubtarget *ST = SPUTM.getSubtargetImpl();
-  SelectionDAG &DAG = DCI.DAG;
-  SDValue Op0 = N->getOperand(0);       // everything has at least one operand
-  EVT NodeVT = N->getValueType(0);      // The node's value type
-  EVT Op0VT = Op0.getValueType();       // The first operand's result
-  SDValue Result;                       // Initially, empty result
-  DebugLoc dl = N->getDebugLoc();
-
-  switch (N->getOpcode()) {
-  default: break;
-  case ISD::ADD: {
-    SDValue Op1 = N->getOperand(1);
-
-    if (Op0.getOpcode() == SPUISD::IndirectAddr
-        || Op1.getOpcode() == SPUISD::IndirectAddr) {
-      // Normalize the operands to reduce repeated code
-      SDValue IndirectArg = Op0, AddArg = Op1;
-
-      if (Op1.getOpcode() == SPUISD::IndirectAddr) {
-        IndirectArg = Op1;
-        AddArg = Op0;
-      }
-
-      if (isa<ConstantSDNode>(AddArg)) {
-        ConstantSDNode *CN0 = cast<ConstantSDNode > (AddArg);
-        SDValue IndOp1 = IndirectArg.getOperand(1);
-
-        if (CN0->isNullValue()) {
-          // (add (SPUindirect <arg>, <arg>), 0) ->
-          // (SPUindirect <arg>, <arg>)
-
-#if !defined(NDEBUG)
-          if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
-            errs() << "\n"
-                 << "Replace: (add (SPUindirect <arg>, <arg>), 0)\n"
-                 << "With:    (SPUindirect <arg>, <arg>)\n";
-          }
-#endif
-
-          return IndirectArg;
-        } else if (isa<ConstantSDNode>(IndOp1)) {
-          // (add (SPUindirect <arg>, <const>), <const>) ->
-          // (SPUindirect <arg>, <const + const>)
-          ConstantSDNode *CN1 = cast<ConstantSDNode > (IndOp1);
-          int64_t combinedConst = CN0->getSExtValue() + CN1->getSExtValue();
-          SDValue combinedValue = DAG.getConstant(combinedConst, Op0VT);
-
-#if !defined(NDEBUG)
-          if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
-            errs() << "\n"
-                 << "Replace: (add (SPUindirect <arg>, " << CN1->getSExtValue()
-                 << "), " << CN0->getSExtValue() << ")\n"
-                 << "With:    (SPUindirect <arg>, "
-                 << combinedConst << ")\n";
-          }
-#endif
-
-          return DAG.getNode(SPUISD::IndirectAddr, dl, Op0VT,
-                             IndirectArg, combinedValue);
-        }
-      }
-    }
-    break;
-  }
-  case ISD::SIGN_EXTEND:
-  case ISD::ZERO_EXTEND:
-  case ISD::ANY_EXTEND: {
-    if (Op0.getOpcode() == SPUISD::VEC2PREFSLOT && NodeVT == Op0VT) {
-      // (any_extend (SPUextract_elt0 <arg>)) ->
-      // (SPUextract_elt0 <arg>)
-      // Types must match, however...
-#if !defined(NDEBUG)
-      if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
-        errs() << "\nReplace: ";
-        N->dump(&DAG);
-        errs() << "\nWith:    ";
-        Op0.getNode()->dump(&DAG);
-        errs() << "\n";
-      }
-#endif
-
-      return Op0;
-    }
-    break;
-  }
-  case SPUISD::IndirectAddr: {
-    if (!ST->usingLargeMem() && Op0.getOpcode() == SPUISD::AFormAddr) {
-      ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N->getOperand(1));
-      if (CN != 0 && CN->isNullValue()) {
-        // (SPUindirect (SPUaform <addr>, 0), 0) ->
-        // (SPUaform <addr>, 0)
-
-        DEBUG(errs() << "Replace: ");
-        DEBUG(N->dump(&DAG));
-        DEBUG(errs() << "\nWith:    ");
-        DEBUG(Op0.getNode()->dump(&DAG));
-        DEBUG(errs() << "\n");
-
-        return Op0;
-      }
-    } else if (Op0.getOpcode() == ISD::ADD) {
-      SDValue Op1 = N->getOperand(1);
-      if (ConstantSDNode *CN1 = dyn_cast<ConstantSDNode>(Op1)) {
-        // (SPUindirect (add <arg>, <arg>), 0) ->
-        // (SPUindirect <arg>, <arg>)
-        if (CN1->isNullValue()) {
-
-#if !defined(NDEBUG)
-          if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
-            errs() << "\n"
-                 << "Replace: (SPUindirect (add <arg>, <arg>), 0)\n"
-                 << "With:    (SPUindirect <arg>, <arg>)\n";
-          }
-#endif
-
-          return DAG.getNode(SPUISD::IndirectAddr, dl, Op0VT,
-                             Op0.getOperand(0), Op0.getOperand(1));
-        }
-      }
-    }
-    break;
-  }
-  case SPUISD::SHL_BITS:
-  case SPUISD::SHL_BYTES:
-  case SPUISD::ROTBYTES_LEFT: {
-    SDValue Op1 = N->getOperand(1);
-
-    // Kill degenerate vector shifts:
-    if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Op1)) {
-      if (CN->isNullValue()) {
-        Result = Op0;
-      }
-    }
-    break;
-  }
-  case SPUISD::PREFSLOT2VEC: {
-    switch (Op0.getOpcode()) {
-    default:
-      break;
-    case ISD::ANY_EXTEND:
-    case ISD::ZERO_EXTEND:
-    case ISD::SIGN_EXTEND: {
-      // (SPUprefslot2vec (any|zero|sign_extend (SPUvec2prefslot <arg>))) ->
-      // <arg>
-      // but only if the SPUprefslot2vec and <arg> types match.
-      SDValue Op00 = Op0.getOperand(0);
-      if (Op00.getOpcode() == SPUISD::VEC2PREFSLOT) {
-        SDValue Op000 = Op00.getOperand(0);
-        if (Op000.getValueType() == NodeVT) {
-          Result = Op000;
-        }
-      }
-      break;
-    }
-    case SPUISD::VEC2PREFSLOT: {
-      // (SPUprefslot2vec (SPUvec2prefslot <arg>)) ->
-      // <arg>
-      Result = Op0.getOperand(0);
-      break;
-    }
-    }
-    break;
-  }
-  }
-
-  // Otherwise, return unchanged.
-#ifndef NDEBUG
-  if (Result.getNode()) {
-    DEBUG(errs() << "\nReplace.SPU: ");
-    DEBUG(N->dump(&DAG));
-    DEBUG(errs() << "\nWith:        ");
-    DEBUG(Result.getNode()->dump(&DAG));
-    DEBUG(errs() << "\n");
-  }
-#endif
-
-  return Result;
-}
-
-//===----------------------------------------------------------------------===//
-// Inline Assembly Support
-//===----------------------------------------------------------------------===//
-
-/// getConstraintType - Given a constraint letter, return the type of
-/// constraint it is for this target.
-SPUTargetLowering::ConstraintType
-SPUTargetLowering::getConstraintType(const std::string &ConstraintLetter) const {
-  if (ConstraintLetter.size() == 1) {
-    switch (ConstraintLetter[0]) {
-    default: break;
-    case 'b':
-    case 'r':
-    case 'f':
-    case 'v':
-    case 'y':
-      return C_RegisterClass;
-    }
-  }
-  return TargetLowering::getConstraintType(ConstraintLetter);
-}
-
-/// Examine constraint type and operand type and determine a weight value.
-/// This object must already have been set up with the operand type
-/// and the current alternative constraint selected.
-TargetLowering::ConstraintWeight
-SPUTargetLowering::getSingleConstraintMatchWeight(
-    AsmOperandInfo &info, const char *constraint) const {
-  ConstraintWeight weight = CW_Invalid;
-  Value *CallOperandVal = info.CallOperandVal;
-    // If we don't have a value, we can't do a match,
-    // but allow it at the lowest weight.
-  if (CallOperandVal == NULL)
-    return CW_Default;
-  // Look at the constraint type.
-  switch (*constraint) {
-  default:
-    weight = TargetLowering::getSingleConstraintMatchWeight(info, constraint);
-    break;
-    //FIXME: Seems like the supported constraint letters were just copied
-    // from PPC, as the following doesn't correspond to the GCC docs.
-    // I'm leaving it so until someone adds the corresponding lowering support.
-  case 'b':
-  case 'r':
-  case 'f':
-  case 'd':
-  case 'v':
-  case 'y':
-    weight = CW_Register;
-    break;
-  }
-  return weight;
-}
-
-std::pair<unsigned, const TargetRegisterClass*>
-SPUTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
-                                                EVT VT) const
-{
-  if (Constraint.size() == 1) {
-    // GCC RS6000 Constraint Letters
-    switch (Constraint[0]) {
-    case 'b':   // R1-R31
-    case 'r':   // R0-R31
-      if (VT == MVT::i64)
-        return std::make_pair(0U, &SPU::R64CRegClass);
-      return std::make_pair(0U, &SPU::R32CRegClass);
-    case 'f':
-      if (VT == MVT::f32)
-        return std::make_pair(0U, &SPU::R32FPRegClass);
-      if (VT == MVT::f64)
-        return std::make_pair(0U, &SPU::R64FPRegClass);
-      break;
-    case 'v':
-      return std::make_pair(0U, &SPU::GPRCRegClass);
-    }
-  }
-
-  return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
-}
-
-//! Compute used/known bits for a SPU operand
-void
-SPUTargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
-                                                  APInt &KnownZero,
-                                                  APInt &KnownOne,
-                                                  const SelectionDAG &DAG,
-                                                  unsigned Depth ) const {
-#if 0
-  const uint64_t uint64_sizebits = sizeof(uint64_t) * CHAR_BIT;
-
-  switch (Op.getOpcode()) {
-  default:
-    // KnownZero = KnownOne = APInt(Mask.getBitWidth(), 0);
-    break;
-  case CALL:
-  case SHUFB:
-  case SHUFFLE_MASK:
-  case CNTB:
-  case SPUISD::PREFSLOT2VEC:
-  case SPUISD::LDRESULT:
-  case SPUISD::VEC2PREFSLOT:
-  case SPUISD::SHLQUAD_L_BITS:
-  case SPUISD::SHLQUAD_L_BYTES:
-  case SPUISD::VEC_ROTL:
-  case SPUISD::VEC_ROTR:
-  case SPUISD::ROTBYTES_LEFT:
-  case SPUISD::SELECT_MASK:
-  case SPUISD::SELB:
-  }
-#endif
-}
-
-unsigned
-SPUTargetLowering::ComputeNumSignBitsForTargetNode(SDValue Op,
-                                                   unsigned Depth) const {
-  switch (Op.getOpcode()) {
-  default:
-    return 1;
-
-  case ISD::SETCC: {
-    EVT VT = Op.getValueType();
-
-    if (VT != MVT::i8 && VT != MVT::i16 && VT != MVT::i32) {
-      VT = MVT::i32;
-    }
-    return VT.getSizeInBits();
-  }
-  }
-}
-
-// LowerAsmOperandForConstraint
-void
-SPUTargetLowering::LowerAsmOperandForConstraint(SDValue Op,
-                                                std::string &Constraint,
-                                                std::vector<SDValue> &Ops,
-                                                SelectionDAG &DAG) const {
-  // Default, for the time being, to the base class handler
-  TargetLowering::LowerAsmOperandForConstraint(Op, Constraint, Ops, DAG);
-}
-
-/// isLegalAddressImmediate - Return true if the integer value can be used
-/// as the offset of the target addressing mode.
-bool SPUTargetLowering::isLegalAddressImmediate(int64_t V,
-                                                Type *Ty) const {
-  // SPU's addresses are 256K:
-  return (V > -(1 << 18) && V < (1 << 18) - 1);
-}
-
-bool SPUTargetLowering::isLegalAddressImmediate(GlobalValue* GV) const {
-  return false;
-}
-
-bool
-SPUTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
-  // The SPU target isn't yet aware of offsets.
-  return false;
-}
-
-// can we compare to Imm without writing it into a register?
-bool SPUTargetLowering::isLegalICmpImmediate(int64_t Imm) const {
-  //ceqi, cgti, etc. all take s10 operand
-  return isInt<10>(Imm);
-}
-
-bool
-SPUTargetLowering::isLegalAddressingMode(const AddrMode &AM,
-                                         Type * ) const{
-
-  // A-form: 18bit absolute address.
-  if (AM.BaseGV && !AM.HasBaseReg && AM.Scale == 0 && AM.BaseOffs == 0)
-    return true;
-
-  // D-form: reg + 14bit offset
-  if (AM.BaseGV ==0 && AM.HasBaseReg && AM.Scale == 0 && isInt<14>(AM.BaseOffs))
-    return true;
-
-  // X-form: reg+reg
-  if (AM.BaseGV == 0 && AM.HasBaseReg && AM.Scale == 1 && AM.BaseOffs ==0)
-    return true;
-
-  return false;
-}
diff --git a/lib/Target/CellSPU/SPUISelLowering.h b/lib/Target/CellSPU/SPUISelLowering.h
deleted file mode 100644
index 9f1599fa6fed..000000000000
--- a/lib/Target/CellSPU/SPUISelLowering.h
+++ /dev/null
@@ -1,178 +0,0 @@
-//===-- SPUISelLowering.h - Cell SPU DAG Lowering Interface -----*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file defines the interfaces that Cell SPU uses to lower LLVM code into
-// a selection DAG.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef SPU_ISELLOWERING_H
-#define SPU_ISELLOWERING_H
-
-#include "SPU.h"
-#include "llvm/Target/TargetLowering.h"
-#include "llvm/CodeGen/SelectionDAG.h"
-
-namespace llvm {
-  namespace SPUISD {
-    enum NodeType {
-      // Start the numbering where the builting ops and target ops leave off.
-      FIRST_NUMBER = ISD::BUILTIN_OP_END,
-
-      // Pseudo instructions:
-      RET_FLAG,                 ///< Return with flag, matched by bi instruction
-
-      Hi,                       ///< High address component (upper 16)
-      Lo,                       ///< Low address component (lower 16)
-      PCRelAddr,                ///< Program counter relative address
-      AFormAddr,                ///< A-form address (local store)
-      IndirectAddr,             ///< D-Form "imm($r)" and X-form "$r($r)"
-
-      LDRESULT,                 ///< Load result (value, chain)
-      CALL,                     ///< CALL instruction
-      SHUFB,                    ///< Vector shuffle (permute)
-      SHUFFLE_MASK,             ///< Shuffle mask
-      CNTB,                     ///< Count leading ones in bytes
-      PREFSLOT2VEC,             ///< Promote scalar->vector
-      VEC2PREFSLOT,             ///< Extract element 0
-      SHL_BITS,                 ///< Shift quad left, by bits
-      SHL_BYTES,                ///< Shift quad left, by bytes
-      SRL_BYTES,                ///< Shift quad right, by bytes. Insert zeros.
-      VEC_ROTL,                 ///< Vector rotate left
-      VEC_ROTR,                 ///< Vector rotate right
-      ROTBYTES_LEFT,            ///< Rotate bytes (loads -> ROTQBYI)
-      ROTBYTES_LEFT_BITS,       ///< Rotate bytes left by bit shift count
-      SELECT_MASK,              ///< Select Mask (FSM, FSMB, FSMH, FSMBI)
-      SELB,                     ///< Select bits -> (b & mask) | (a & ~mask)
-      // Markers: These aren't used to generate target-dependent nodes, but
-      // are used during instruction selection.
-      ADD64_MARKER,             ///< i64 addition marker
-      SUB64_MARKER,             ///< i64 subtraction marker
-      MUL64_MARKER,             ///< i64 multiply marker
-      LAST_SPUISD               ///< Last user-defined instruction
-    };
-  }
-
-  //! Utility functions specific to CellSPU:
-  namespace SPU {
-    SDValue get_vec_u18imm(SDNode *N, SelectionDAG &DAG,
-                             EVT ValueType);
-    SDValue get_vec_i16imm(SDNode *N, SelectionDAG &DAG,
-                             EVT ValueType);
-    SDValue get_vec_i10imm(SDNode *N, SelectionDAG &DAG,
-                             EVT ValueType);
-    SDValue get_vec_i8imm(SDNode *N, SelectionDAG &DAG,
-                            EVT ValueType);
-    SDValue get_ILHUvec_imm(SDNode *N, SelectionDAG &DAG,
-                              EVT ValueType);
-    SDValue get_v4i32_imm(SDNode *N, SelectionDAG &DAG);
-    SDValue get_v2i64_imm(SDNode *N, SelectionDAG &DAG);
-
-    SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG,
-                              const SPUTargetMachine &TM);
-    //! Simplify a EVT::v2i64 constant splat to CellSPU-ready form
-    SDValue LowerV2I64Splat(EVT OpVT, SelectionDAG &DAG, uint64_t splat,
-                             DebugLoc dl);
-  }
-
-  class SPUTargetMachine;            // forward dec'l.
-
-  class SPUTargetLowering :
-    public TargetLowering
-  {
-    SPUTargetMachine &SPUTM;
-
-  public:
-    //! The venerable constructor
-    /*!
-     This is where the CellSPU backend sets operation handling (i.e., legal,
-     custom, expand or promote.)
-     */
-    SPUTargetLowering(SPUTargetMachine &TM);
-
-    //! Get the target machine
-    SPUTargetMachine &getSPUTargetMachine() {
-      return SPUTM;
-    }
-
-    /// getTargetNodeName() - This method returns the name of a target specific
-    /// DAG node.
-    virtual const char *getTargetNodeName(unsigned Opcode) const;
-
-    /// getSetCCResultType - Return the ValueType for ISD::SETCC
-    virtual EVT getSetCCResultType(EVT VT) const;
-
-    virtual MVT getShiftAmountTy(EVT LHSTy) const { return MVT::i32; }
-
-    //! Custom lowering hooks
-    virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
-
-    //! Custom lowering hook for nodes with illegal result types.
-    virtual void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue>&Results,
-                                    SelectionDAG &DAG) const;
-
-    virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
-
-    virtual void computeMaskedBitsForTargetNode(const SDValue Op,
-                                                APInt &KnownZero,
-                                                APInt &KnownOne,
-                                                const SelectionDAG &DAG,
-                                                unsigned Depth = 0) const;
-
-    virtual unsigned ComputeNumSignBitsForTargetNode(SDValue Op,
-                                                   unsigned Depth = 0) const;
-
-    ConstraintType getConstraintType(const std::string &ConstraintLetter) const;
-
-    /// Examine constraint string and operand type and determine a weight value.
-    /// The operand object must already have been set up with the operand type.
-    ConstraintWeight getSingleConstraintMatchWeight(
-      AsmOperandInfo &info, const char *constraint) const;
-
-    std::pair<unsigned, const TargetRegisterClass*>
-      getRegForInlineAsmConstraint(const std::string &Constraint,
-                                   EVT VT) const;
-
-    void LowerAsmOperandForConstraint(SDValue Op, std::string &Constraint,
-                                      std::vector<SDValue> &Ops,
-                                      SelectionDAG &DAG) const;
-
-    /// isLegalAddressImmediate - Return true if the integer value can be used
-    /// as the offset of the target addressing mode.
-    virtual bool isLegalAddressImmediate(int64_t V, Type *Ty) const;
-    virtual bool isLegalAddressImmediate(GlobalValue *) const;
-
-    virtual bool isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const;
-
-    virtual SDValue
-      LowerFormalArguments(SDValue Chain,
-                           CallingConv::ID CallConv, bool isVarArg,
-                           const SmallVectorImpl<ISD::InputArg> &Ins,
-                           DebugLoc dl, SelectionDAG &DAG,
-                           SmallVectorImpl<SDValue> &InVals) const;
-
-    virtual SDValue
-      LowerCall(TargetLowering::CallLoweringInfo &CLI,
-                SmallVectorImpl<SDValue> &InVals) const;
-
-    virtual SDValue
-      LowerReturn(SDValue Chain,
-                  CallingConv::ID CallConv, bool isVarArg,
-                  const SmallVectorImpl<ISD::OutputArg> &Outs,
-                  const SmallVectorImpl<SDValue> &OutVals,
-                  DebugLoc dl, SelectionDAG &DAG) const;
-
-    virtual bool isLegalICmpImmediate(int64_t Imm) const;
-
-    virtual bool isLegalAddressingMode(const AddrMode &AM,
-                                       Type *Ty) const;
-  };
-}
-
-#endif
diff --git a/lib/Target/CellSPU/SPUInstrBuilder.h b/lib/Target/CellSPU/SPUInstrBuilder.h
deleted file mode 100644
index b495537fc2c8..000000000000
--- a/lib/Target/CellSPU/SPUInstrBuilder.h
+++ /dev/null
@@ -1,43 +0,0 @@
-//===-- SPUInstrBuilder.h - Aides for building Cell SPU insts ---*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file exposes functions that may be used with BuildMI from the
-// MachineInstrBuilder.h file to simplify generating frame and constant pool
-// references.
-//
-// For reference, the order of operands for memory references is:
-// (Operand), Dest Reg, Base Reg, and either Reg Index or Immediate
-// Displacement.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef SPU_INSTRBUILDER_H
-#define SPU_INSTRBUILDER_H
-
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-
-namespace llvm {
-
-/// addFrameReference - This function is used to add a reference to the base of
-/// an abstract object on the stack frame of the current function.  This
-/// reference has base register as the FrameIndex offset until it is resolved.
-/// This allows a constant offset to be specified as well...
-///
-inline const MachineInstrBuilder&
-addFrameReference(const MachineInstrBuilder &MIB, int FI, int Offset = 0,
-                  bool mem = true) {
-  if (mem)
-    return MIB.addImm(Offset).addFrameIndex(FI);
-  else
-    return MIB.addFrameIndex(FI).addImm(Offset);
-}
-
-} // End llvm namespace
-
-#endif
diff --git a/lib/Target/CellSPU/SPUInstrFormats.td b/lib/Target/CellSPU/SPUInstrFormats.td
deleted file mode 100644
index cd3f42214345..000000000000
--- a/lib/Target/CellSPU/SPUInstrFormats.td
+++ /dev/null
@@ -1,320 +0,0 @@
-//===-- SPUInstrFormats.td - Cell SPU Instruction Formats --*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-//===----------------------------------------------------------------------===//
-//
-// Cell SPU instruction formats. Note that these are notationally similar to
-// PowerPC, like "A-Form". But the sizes of operands and fields differ.
-
-// This was kiped from the PPC instruction formats (seemed like a good idea...)
-
-class SPUInstr<dag OOL, dag IOL, string asmstr, InstrItinClass itin>
-        : Instruction {
-  field bits<32> Inst;
-
-  let Namespace = "SPU";
-  let OutOperandList = OOL;
-  let InOperandList = IOL;
-  let AsmString = asmstr;
-  let Itinerary = itin;
-}
-
-// RR Format
-class RRForm<bits<11> opcode, dag OOL, dag IOL, string asmstr, 
-              InstrItinClass itin, list<dag> pattern>
-         : SPUInstr<OOL, IOL, asmstr, itin> {
-  bits<7> RA;
-  bits<7> RB;
-  bits<7> RT;
-
-  let Pattern = pattern;
-
-  let Inst{0-10} = opcode;
-  let Inst{11-17} = RB;
-  let Inst{18-24} = RA;
-  let Inst{25-31} = RT;
-}
-
-let RB = 0 in {
-  // RR Format, where RB is zeroed (dont care):
-  class RRForm_1<bits<11> opcode, dag OOL, dag IOL, string asmstr, 
-                 InstrItinClass itin, list<dag> pattern>
-           : RRForm<opcode, OOL, IOL, asmstr, itin, pattern>
-  { }
-
-  let RA = 0 in {
-    // RR Format, where RA and RB are zeroed (dont care):
-    // Used for reads from status control registers (see FPSCRRr32)
-    class RRForm_2<bits<11> opcode, dag OOL, dag IOL, string asmstr,
-                   InstrItinClass itin, list<dag> pattern>
-             : RRForm<opcode, OOL, IOL, asmstr, itin, pattern>
-    { }
-  }
-}
-
-let RT = 0 in {
-  // RR Format, where RT is zeroed (don't care), or as the instruction handbook
-  // says, "RT is a false target." Used in "Halt if" instructions
-  class RRForm_3<bits<11> opcode, dag OOL, dag IOL, string asmstr,
-                 InstrItinClass itin, list<dag> pattern>
-      : RRForm<opcode, OOL, IOL, asmstr, itin, pattern>
-  { }
-}
-
-// RRR Format
-class RRRForm<bits<4> opcode, dag OOL, dag IOL, string asmstr,
-              InstrItinClass itin, list<dag> pattern>
-        : SPUInstr<OOL, IOL, asmstr, itin>
-{
-  bits<7> RA;
-  bits<7> RB;
-  bits<7> RC;
-  bits<7> RT;
-
-  let Pattern = pattern;
-
-  let Inst{0-3} = opcode;
-  let Inst{4-10} = RT;
-  let Inst{11-17} = RB;
-  let Inst{18-24} = RA;
-  let Inst{25-31} = RC;
-}
-
-// RI7 Format
-class RI7Form<bits<11> opcode, dag OOL, dag IOL, string asmstr,
-              InstrItinClass itin, list<dag> pattern>
-        : SPUInstr<OOL, IOL, asmstr, itin>
-{
-  bits<7> i7;
-  bits<7> RA;
-  bits<7> RT;
-
-  let Pattern = pattern;
-
-  let Inst{0-10} = opcode;
-  let Inst{11-17} = i7;
-  let Inst{18-24} = RA;
-  let Inst{25-31} = RT;
-}
-
-// CVTIntFp Format
-class CVTIntFPForm<bits<10> opcode, dag OOL, dag IOL, string asmstr,
-                   InstrItinClass itin, list<dag> pattern>
-        : SPUInstr<OOL, IOL, asmstr, itin>
-{
-  bits<7> RA;
-  bits<7> RT;
-
-  let Pattern = pattern;
-
-  let Inst{0-9} = opcode;
-  let Inst{10-17} = 0;
-  let Inst{18-24} = RA;
-  let Inst{25-31} = RT;
-}
-
-let RA = 0 in {
-  class BICondForm<bits<11> opcode, dag OOL, dag IOL, string asmstr, list<dag> pattern>
-           : RRForm<opcode, OOL, IOL, asmstr, BranchResolv, pattern>
-  { }
-
-  let RT = 0 in {
-    // Branch instruction format (without D/E flag settings)
-    class BRForm<bits<11> opcode, dag OOL, dag IOL, string asmstr,
-               InstrItinClass itin, list<dag> pattern>
-          : RRForm<opcode, OOL, IOL, asmstr, itin, pattern>
-    { }
-
-    class BIForm<bits<11> opcode, string asmstr, list<dag> pattern>
-             : RRForm<opcode, (outs), (ins R32C:$func), asmstr, BranchResolv,
-                      pattern>
-    { }
-
-    let RB = 0 in {
-      // Return instruction (bi, branch indirect), RA is zero (LR):
-      class RETForm<string asmstr, list<dag> pattern>
-             : BRForm<0b00010101100, (outs), (ins), asmstr, BranchResolv,
-                      pattern>
-      { }
-    }
-  }
-}
-
-// Branch indirect external data forms:
-class BISLEDForm<bits<2> DE_flag, string asmstr, list<dag> pattern>
-         : SPUInstr<(outs), (ins indcalltarget:$func), asmstr, BranchResolv>
-{
-  bits<7> Rcalldest;
-
-  let Pattern = pattern;
-
-  let Inst{0-10} = 0b11010101100;
-  let Inst{11} = 0;
-  let Inst{12-13} = DE_flag;
-  let Inst{14-17} = 0b0000;
-  let Inst{18-24} = Rcalldest;
-  let Inst{25-31} = 0b0000000;
-}
-
-// RI10 Format
-class RI10Form<bits<8> opcode, dag OOL, dag IOL, string asmstr,
-              InstrItinClass itin, list<dag> pattern>
-        : SPUInstr<OOL, IOL, asmstr, itin>
-{
-  bits<10> i10;
-  bits<7> RA;
-  bits<7> RT;
-
-  let Pattern = pattern;
-
-  let Inst{0-7} = opcode;
-  let Inst{8-17} = i10;
-  let Inst{18-24} = RA;
-  let Inst{25-31} = RT;
-}
-
-// RI10 Format, where the constant is zero (or effectively ignored by the
-// SPU)
-let i10 = 0 in {
-  class RI10Form_1<bits<8> opcode, dag OOL, dag IOL, string asmstr,
-                   InstrItinClass itin, list<dag> pattern>
-          : RI10Form<opcode, OOL, IOL, asmstr, itin, pattern>
-  { }
-}
-
-// RI10 Format, where RT is ignored.
-// This format is used primarily by the Halt If ... Immediate set of
-// instructions
-let RT = 0 in {
-  class RI10Form_2<bits<8> opcode, dag OOL, dag IOL, string asmstr,
-                   InstrItinClass itin, list<dag> pattern>
-        : RI10Form<opcode, OOL, IOL, asmstr, itin, pattern>
-  { }
-}
-
-// RI16 Format
-class RI16Form<bits<9> opcode, dag OOL, dag IOL, string asmstr,
-              InstrItinClass itin, list<dag> pattern>
-        : SPUInstr<OOL, IOL, asmstr, itin>
-{
-  bits<16> i16;
-  bits<7> RT;
-
-  let Pattern = pattern;
-
-  let Inst{0-8} = opcode;
-  let Inst{9-24} = i16;
-  let Inst{25-31} = RT;
-}
-
-// Specialized version of the RI16 Format for unconditional branch relative and
-// branch absolute, branch and set link. Note that for branch and set link, the
-// link register doesn't have to be $lr, but this is actually hard coded into
-// the instruction pattern.
-
-let RT = 0 in {
-  class UncondBranch<bits<9> opcode, dag OOL, dag IOL, string asmstr,
-                     list<dag> pattern>
-    : RI16Form<opcode, OOL, IOL, asmstr, BranchResolv, pattern>
-  { }
-
-  class BranchSetLink<bits<9> opcode, dag OOL, dag IOL, string asmstr,
-                      list<dag> pattern>
-        : RI16Form<opcode, OOL, IOL, asmstr, BranchResolv, pattern>
-  { }
-}
-
-//===----------------------------------------------------------------------===//
-// Specialized versions of RI16:
-//===----------------------------------------------------------------------===//
-
-// RI18 Format
-class RI18Form<bits<7> opcode, dag OOL, dag IOL, string asmstr,
-              InstrItinClass itin, list<dag> pattern>
-        : SPUInstr<OOL, IOL, asmstr, itin>
-{
-  bits<18> i18;
-  bits<7> RT;
-
-  let Pattern = pattern;
-
-  let Inst{0-6} = opcode;
-  let Inst{7-24} = i18;
-  let Inst{25-31} = RT;
-}
-
-//===----------------------------------------------------------------------===//
-// Instruction formats for intrinsics:
-//===----------------------------------------------------------------------===//
-
-// RI10 Format for v8i16 intrinsics
-class RI10_Int_v8i16<bits<8> opcode, string opc, InstrItinClass itin,
-                     Intrinsic IntID> :
-  RI10Form<opcode, (outs VECREG:$rT), (ins s10imm:$val, VECREG:$rA),
-           !strconcat(opc, " $rT, $rA, $val"), itin,
-           [(set (v8i16 VECREG:$rT), (IntID (v8i16 VECREG:$rA),
-                                            i16ImmSExt10:$val))] >;
-
-class RI10_Int_v4i32<bits<8> opcode, string opc, InstrItinClass itin,
-                     Intrinsic IntID> :
-  RI10Form<opcode, (outs VECREG:$rT), (ins s10imm:$val, VECREG:$rA),
-           !strconcat(opc, " $rT, $rA, $val"), itin,
-           [(set (v4i32 VECREG:$rT), (IntID (v4i32 VECREG:$rA),
-                                            i32ImmSExt10:$val))] >;
-
-// RR Format for v8i16 intrinsics
-class RR_Int_v8i16<bits<11> opcode, string opc, InstrItinClass itin,
-                   Intrinsic IntID> :
-  RRForm<opcode, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-         !strconcat(opc, " $rT, $rA, $rB"), itin,
-         [(set (v8i16 VECREG:$rT), (IntID (v8i16 VECREG:$rA),
-                                          (v8i16 VECREG:$rB)))] >;
-
-// RR Format for v4i32 intrinsics
-class RR_Int_v4i32<bits<11> opcode, string opc, InstrItinClass itin,
-                   Intrinsic IntID> :
-  RRForm<opcode, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-         !strconcat(opc, " $rT, $rA, $rB"), itin,
-         [(set (v4i32 VECREG:$rT), (IntID (v4i32 VECREG:$rA),
-                                          (v4i32 VECREG:$rB)))] >;
-
-//===----------------------------------------------------------------------===//
-// Pseudo instructions, like call frames:
-//===----------------------------------------------------------------------===//
-
-class Pseudo<dag OOL, dag IOL, string asmstr, list<dag> pattern>
-    : SPUInstr<OOL, IOL, asmstr, NoItinerary> {
-  let OutOperandList = OOL;
-  let InOperandList = IOL;
-  let AsmString   = asmstr;
-  let Pattern = pattern;
-  let Inst{31-0} = 0;
-}
-
-//===----------------------------------------------------------------------===//
-// Branch hint formats
-//===----------------------------------------------------------------------===//
-// For hbrr and hbra
-class HBI16Form<bits<7> opcode, dag IOL, string asmstr>
-        : Instruction {
-  field bits<32> Inst;
-  bits<16>i16;
-  bits<9>RO;
-
-  let Namespace = "SPU";
-  let InOperandList = IOL;
-  let OutOperandList = (outs); //no output
-  let AsmString = asmstr;
-  let Itinerary = BranchHints;
-
-  let Inst{0-6} = opcode;
-  let Inst{7-8} = RO{8-7};
-  let Inst{9-24} = i16;
-  let Inst{25-31} = RO{6-0};
-}
diff --git a/lib/Target/CellSPU/SPUInstrInfo.cpp b/lib/Target/CellSPU/SPUInstrInfo.cpp
deleted file mode 100644
index b25a6397ec3a..000000000000
--- a/lib/Target/CellSPU/SPUInstrInfo.cpp
+++ /dev/null
@@ -1,449 +0,0 @@
-//===-- SPUInstrInfo.cpp - Cell SPU Instruction Information ---------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file contains the Cell SPU implementation of the TargetInstrInfo class.
-//
-//===----------------------------------------------------------------------===//
-
-#include "SPUInstrInfo.h"
-#include "SPUInstrBuilder.h"
-#include "SPUTargetMachine.h"
-#include "SPUHazardRecognizers.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/MC/MCContext.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/TargetRegistry.h"
-#include "llvm/Support/raw_ostream.h"
-
-#define GET_INSTRINFO_CTOR
-#include "SPUGenInstrInfo.inc"
-
-using namespace llvm;
-
-namespace {
-  //! Predicate for an unconditional branch instruction
-  inline bool isUncondBranch(const MachineInstr *I) {
-    unsigned opc = I->getOpcode();
-
-    return (opc == SPU::BR
-            || opc == SPU::BRA
-            || opc == SPU::BI);
-  }
-
-  //! Predicate for a conditional branch instruction
-  inline bool isCondBranch(const MachineInstr *I) {
-    unsigned opc = I->getOpcode();
-
-    return (opc == SPU::BRNZr32
-            || opc == SPU::BRNZv4i32
-            || opc == SPU::BRZr32
-            || opc == SPU::BRZv4i32
-            || opc == SPU::BRHNZr16
-            || opc == SPU::BRHNZv8i16
-            || opc == SPU::BRHZr16
-            || opc == SPU::BRHZv8i16);
-  }
-}
-
-SPUInstrInfo::SPUInstrInfo(SPUTargetMachine &tm)
-  : SPUGenInstrInfo(SPU::ADJCALLSTACKDOWN, SPU::ADJCALLSTACKUP),
-    TM(tm),
-    RI(*TM.getSubtargetImpl(), *this)
-{ /* NOP */ }
-
-/// CreateTargetHazardRecognizer - Return the hazard recognizer to use for
-/// this target when scheduling the DAG.
-ScheduleHazardRecognizer *SPUInstrInfo::CreateTargetHazardRecognizer(
-  const TargetMachine *TM,
-  const ScheduleDAG *DAG) const {
-  const TargetInstrInfo *TII = TM->getInstrInfo();
-  assert(TII && "No InstrInfo?");
-  return new SPUHazardRecognizer(*TII);
-}
-
-unsigned
-SPUInstrInfo::isLoadFromStackSlot(const MachineInstr *MI,
-                                  int &FrameIndex) const {
-  switch (MI->getOpcode()) {
-  default: break;
-  case SPU::LQDv16i8:
-  case SPU::LQDv8i16:
-  case SPU::LQDv4i32:
-  case SPU::LQDv4f32:
-  case SPU::LQDv2f64:
-  case SPU::LQDr128:
-  case SPU::LQDr64:
-  case SPU::LQDr32:
-  case SPU::LQDr16: {
-    const MachineOperand MOp1 = MI->getOperand(1);
-    const MachineOperand MOp2 = MI->getOperand(2);
-    if (MOp1.isImm() && MOp2.isFI()) {
-      FrameIndex = MOp2.getIndex();
-      return MI->getOperand(0).getReg();
-    }
-    break;
-  }
-  }
-  return 0;
-}
-
-unsigned
-SPUInstrInfo::isStoreToStackSlot(const MachineInstr *MI,
-                                 int &FrameIndex) const {
-  switch (MI->getOpcode()) {
-  default: break;
-  case SPU::STQDv16i8:
-  case SPU::STQDv8i16:
-  case SPU::STQDv4i32:
-  case SPU::STQDv4f32:
-  case SPU::STQDv2f64:
-  case SPU::STQDr128:
-  case SPU::STQDr64:
-  case SPU::STQDr32:
-  case SPU::STQDr16:
-  case SPU::STQDr8: {
-    const MachineOperand MOp1 = MI->getOperand(1);
-    const MachineOperand MOp2 = MI->getOperand(2);
-    if (MOp1.isImm() && MOp2.isFI()) {
-      FrameIndex = MOp2.getIndex();
-      return MI->getOperand(0).getReg();
-    }
-    break;
-  }
-  }
-  return 0;
-}
-
-void SPUInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
-                               MachineBasicBlock::iterator I, DebugLoc DL,
-                               unsigned DestReg, unsigned SrcReg,
-                               bool KillSrc) const
-{
-  // We support cross register class moves for our aliases, such as R3 in any
-  // reg class to any other reg class containing R3.  This is required because
-  // we instruction select bitconvert i64 -> f64 as a noop for example, so our
-  // types have no specific meaning.
-
-  BuildMI(MBB, I, DL, get(SPU::LRr128), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-}
-
-void
-SPUInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
-                                  MachineBasicBlock::iterator MI,
-                                  unsigned SrcReg, bool isKill, int FrameIdx,
-                                  const TargetRegisterClass *RC,
-                                  const TargetRegisterInfo *TRI) const {
-  unsigned opc;
-  bool isValidFrameIdx = (FrameIdx < SPUFrameLowering::maxFrameOffset());
-  if (RC == &SPU::GPRCRegClass)
-    opc = isValidFrameIdx ? SPU::STQDr128 : SPU::STQXr128;
-  else if (RC == &SPU::R64CRegClass)
-    opc = isValidFrameIdx ? SPU::STQDr64 : SPU::STQXr64;
-  else if (RC == &SPU::R64FPRegClass)
-    opc = isValidFrameIdx ? SPU::STQDr64 : SPU::STQXr64;
-  else if (RC == &SPU::R32CRegClass)
-    opc = isValidFrameIdx ? SPU::STQDr32 : SPU::STQXr32;
-  else if (RC == &SPU::R32FPRegClass)
-    opc = isValidFrameIdx ? SPU::STQDr32 : SPU::STQXr32;
-  else if (RC == &SPU::R16CRegClass)
-    opc = isValidFrameIdx ? SPU::STQDr16 : SPU::STQXr16;
-  else if (RC == &SPU::R8CRegClass)
-    opc = isValidFrameIdx ? SPU::STQDr8 : SPU::STQXr8;
-  else if (RC == &SPU::VECREGRegClass)
-    opc = isValidFrameIdx ? SPU::STQDv16i8 : SPU::STQXv16i8;
-  else
-    llvm_unreachable("Unknown regclass!");
-
-  DebugLoc DL;
-  if (MI != MBB.end()) DL = MI->getDebugLoc();
-  addFrameReference(BuildMI(MBB, MI, DL, get(opc))
-                    .addReg(SrcReg, getKillRegState(isKill)), FrameIdx);
-}
-
-void
-SPUInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
-                                   MachineBasicBlock::iterator MI,
-                                   unsigned DestReg, int FrameIdx,
-                                   const TargetRegisterClass *RC,
-                                   const TargetRegisterInfo *TRI) const {
-  unsigned opc;
-  bool isValidFrameIdx = (FrameIdx < SPUFrameLowering::maxFrameOffset());
-  if (RC == &SPU::GPRCRegClass)
-    opc = isValidFrameIdx ? SPU::LQDr128 : SPU::LQXr128;
-  else if (RC == &SPU::R64CRegClass)
-    opc = isValidFrameIdx ? SPU::LQDr64 : SPU::LQXr64;
-  else if (RC == &SPU::R64FPRegClass)
-    opc = isValidFrameIdx ? SPU::LQDr64 : SPU::LQXr64;
-  else if (RC == &SPU::R32CRegClass)
-    opc = isValidFrameIdx ? SPU::LQDr32 : SPU::LQXr32;
-  else if (RC == &SPU::R32FPRegClass)
-    opc = isValidFrameIdx ? SPU::LQDr32 : SPU::LQXr32;
-  else if (RC == &SPU::R16CRegClass)
-    opc = isValidFrameIdx ? SPU::LQDr16 : SPU::LQXr16;
-  else if (RC == &SPU::R8CRegClass)
-    opc = isValidFrameIdx ? SPU::LQDr8 : SPU::LQXr8;
-  else if (RC == &SPU::VECREGRegClass)
-    opc = isValidFrameIdx ? SPU::LQDv16i8 : SPU::LQXv16i8;
-  else
-    llvm_unreachable("Unknown regclass in loadRegFromStackSlot!");
-
-  DebugLoc DL;
-  if (MI != MBB.end()) DL = MI->getDebugLoc();
-  addFrameReference(BuildMI(MBB, MI, DL, get(opc), DestReg), FrameIdx);
-}
-
-//! Branch analysis
-/*!
-  \note This code was kiped from PPC. There may be more branch analysis for
-  CellSPU than what's currently done here.
- */
-bool
-SPUInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
-                            MachineBasicBlock *&FBB,
-                            SmallVectorImpl<MachineOperand> &Cond,
-                            bool AllowModify) const {
-  // If the block has no terminators, it just falls into the block after it.
-  MachineBasicBlock::iterator I = MBB.end();
-  if (I == MBB.begin())
-    return false;
-  --I;
-  while (I->isDebugValue()) {
-    if (I == MBB.begin())
-      return false;
-    --I;
-  }
-  if (!isUnpredicatedTerminator(I))
-    return false;
-
-  // Get the last instruction in the block.
-  MachineInstr *LastInst = I;
-
-  // If there is only one terminator instruction, process it.
-  if (I == MBB.begin() || !isUnpredicatedTerminator(--I)) {
-    if (isUncondBranch(LastInst)) {
-      // Check for jump tables
-      if (!LastInst->getOperand(0).isMBB())
-        return true;
-      TBB = LastInst->getOperand(0).getMBB();
-      return false;
-    } else if (isCondBranch(LastInst)) {
-      // Block ends with fall-through condbranch.
-      TBB = LastInst->getOperand(1).getMBB();
-      DEBUG(errs() << "Pushing LastInst:               ");
-      DEBUG(LastInst->dump());
-      Cond.push_back(MachineOperand::CreateImm(LastInst->getOpcode()));
-      Cond.push_back(LastInst->getOperand(0));
-      return false;
-    }
-    // Otherwise, don't know what this is.
-    return true;
-  }
-
-  // Get the instruction before it if it's a terminator.
-  MachineInstr *SecondLastInst = I;
-
-  // If there are three terminators, we don't know what sort of block this is.
-  if (SecondLastInst && I != MBB.begin() &&
-      isUnpredicatedTerminator(--I))
-    return true;
-
-  // If the block ends with a conditional and unconditional branch, handle it.
-  if (isCondBranch(SecondLastInst) && isUncondBranch(LastInst)) {
-    TBB =  SecondLastInst->getOperand(1).getMBB();
-    DEBUG(errs() << "Pushing SecondLastInst:         ");
-    DEBUG(SecondLastInst->dump());
-    Cond.push_back(MachineOperand::CreateImm(SecondLastInst->getOpcode()));
-    Cond.push_back(SecondLastInst->getOperand(0));
-    FBB = LastInst->getOperand(0).getMBB();
-    return false;
-  }
-
-  // If the block ends with two unconditional branches, handle it.  The second
-  // one is not executed, so remove it.
-  if (isUncondBranch(SecondLastInst) && isUncondBranch(LastInst)) {
-    TBB = SecondLastInst->getOperand(0).getMBB();
-    I = LastInst;
-    if (AllowModify)
-      I->eraseFromParent();
-    return false;
-  }
-
-  // Otherwise, can't handle this.
-  return true;
-}
-
-// search MBB for branch hint labels and branch hit ops
-static void removeHBR( MachineBasicBlock &MBB) {
-  for (MachineBasicBlock::iterator I = MBB.begin(); I != MBB.end(); ++I){
-    if (I->getOpcode() == SPU::HBRA ||
-        I->getOpcode() == SPU::HBR_LABEL){
-      I=MBB.erase(I);
-      if (I == MBB.end())
-        break;
-    }
-  }
-}
-
-unsigned
-SPUInstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
-  MachineBasicBlock::iterator I = MBB.end();
-  removeHBR(MBB);
-  if (I == MBB.begin())
-    return 0;
-  --I;
-  while (I->isDebugValue()) {
-    if (I == MBB.begin())
-      return 0;
-    --I;
-  }
-  if (!isCondBranch(I) && !isUncondBranch(I))
-    return 0;
-
-  // Remove the first branch.
-  DEBUG(errs() << "Removing branch:                ");
-  DEBUG(I->dump());
-  I->eraseFromParent();
-  I = MBB.end();
-  if (I == MBB.begin())
-    return 1;
-
-  --I;
-  if (!(isCondBranch(I) || isUncondBranch(I)))
-    return 1;
-
-  // Remove the second branch.
-  DEBUG(errs() << "Removing second branch:         ");
-  DEBUG(I->dump());
-  I->eraseFromParent();
-  return 2;
-}
-
-/** Find the optimal position for a hint branch instruction in a basic block.
- * This should take into account:
- *   -the branch hint delays
- *   -congestion of the memory bus
- *   -dual-issue scheduling (i.e. avoid insertion of nops)
- * Current implementation is rather simplistic.
- */
-static MachineBasicBlock::iterator findHBRPosition(MachineBasicBlock &MBB)
-{
-   MachineBasicBlock::iterator J = MBB.end();
-   for( int i=0; i<8; i++) {
-     if( J == MBB.begin() ) return J;
-     J--;
-   }
-   return J;
-}
-
-unsigned
-SPUInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
-                           MachineBasicBlock *FBB,
-                           const SmallVectorImpl<MachineOperand> &Cond,
-                           DebugLoc DL) const {
-  // Shouldn't be a fall through.
-  assert(TBB && "InsertBranch must not be told to insert a fallthrough");
-  assert((Cond.size() == 2 || Cond.size() == 0) &&
-         "SPU branch conditions have two components!");
-
-  MachineInstrBuilder MIB;
-  //TODO: make a more accurate algorithm.
-  bool haveHBR = MBB.size()>8;
-
-  removeHBR(MBB);
-  MCSymbol *branchLabel = MBB.getParent()->getContext().CreateTempSymbol();
-  // Add a label just before the branch
-  if (haveHBR)
-    MIB = BuildMI(&MBB, DL, get(SPU::HBR_LABEL)).addSym(branchLabel);
-
-  // One-way branch.
-  if (FBB == 0) {
-    if (Cond.empty()) {
-      // Unconditional branch
-      MIB = BuildMI(&MBB, DL, get(SPU::BR));
-      MIB.addMBB(TBB);
-
-      DEBUG(errs() << "Inserted one-way uncond branch: ");
-      DEBUG((*MIB).dump());
-
-      // basic blocks have just one branch so it is safe to add the hint a its
-      if (haveHBR) {
-        MIB = BuildMI( MBB, findHBRPosition(MBB), DL, get(SPU::HBRA));
-        MIB.addSym(branchLabel);
-        MIB.addMBB(TBB);
-      }
-    } else {
-      // Conditional branch
-      MIB = BuildMI(&MBB, DL, get(Cond[0].getImm()));
-      MIB.addReg(Cond[1].getReg()).addMBB(TBB);
-
-      if (haveHBR) {
-        MIB = BuildMI(MBB, findHBRPosition(MBB), DL, get(SPU::HBRA));
-        MIB.addSym(branchLabel);
-        MIB.addMBB(TBB);
-      }
-
-      DEBUG(errs() << "Inserted one-way cond branch:   ");
-      DEBUG((*MIB).dump());
-    }
-    return 1;
-  } else {
-    MIB = BuildMI(&MBB, DL, get(Cond[0].getImm()));
-    MachineInstrBuilder MIB2 = BuildMI(&MBB, DL, get(SPU::BR));
-
-    // Two-way Conditional Branch.
-    MIB.addReg(Cond[1].getReg()).addMBB(TBB);
-    MIB2.addMBB(FBB);
-
-    if (haveHBR) {
-      MIB = BuildMI( MBB, findHBRPosition(MBB), DL, get(SPU::HBRA));
-      MIB.addSym(branchLabel);
-      MIB.addMBB(FBB);
-    }
-
-    DEBUG(errs() << "Inserted conditional branch:    ");
-    DEBUG((*MIB).dump());
-    DEBUG(errs() << "part 2: ");
-    DEBUG((*MIB2).dump());
-   return 2;
-  }
-}
-
-//! Reverses a branch's condition, returning false on success.
-bool
-SPUInstrInfo::ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond)
-  const {
-  // Pretty brainless way of inverting the condition, but it works, considering
-  // there are only two conditions...
-  static struct {
-    unsigned Opc;               //! The incoming opcode
-    unsigned RevCondOpc;        //! The reversed condition opcode
-  } revconds[] = {
-    { SPU::BRNZr32, SPU::BRZr32 },
-    { SPU::BRNZv4i32, SPU::BRZv4i32 },
-    { SPU::BRZr32, SPU::BRNZr32 },
-    { SPU::BRZv4i32, SPU::BRNZv4i32 },
-    { SPU::BRHNZr16, SPU::BRHZr16 },
-    { SPU::BRHNZv8i16, SPU::BRHZv8i16 },
-    { SPU::BRHZr16, SPU::BRHNZr16 },
-    { SPU::BRHZv8i16, SPU::BRHNZv8i16 }
-  };
-
-  unsigned Opc = unsigned(Cond[0].getImm());
-  // Pretty dull mapping between the two conditions that SPU can generate:
-  for (int i = sizeof(revconds)/sizeof(revconds[0]) - 1; i >= 0; --i) {
-    if (revconds[i].Opc == Opc) {
-      Cond[0].setImm(revconds[i].RevCondOpc);
-      return false;
-    }
-  }
-
-  return true;
-}
diff --git a/lib/Target/CellSPU/SPUInstrInfo.h b/lib/Target/CellSPU/SPUInstrInfo.h
deleted file mode 100644
index 85e5821aefa1..000000000000
--- a/lib/Target/CellSPU/SPUInstrInfo.h
+++ /dev/null
@@ -1,84 +0,0 @@
-//===-- SPUInstrInfo.h - Cell SPU Instruction Information -------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file contains the CellSPU implementation of the TargetInstrInfo class.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef SPU_INSTRUCTIONINFO_H
-#define SPU_INSTRUCTIONINFO_H
-
-#include "SPU.h"
-#include "SPURegisterInfo.h"
-#include "llvm/Target/TargetInstrInfo.h"
-
-#define GET_INSTRINFO_HEADER
-#include "SPUGenInstrInfo.inc"
-
-namespace llvm {
-  //! Cell SPU instruction information class
-  class SPUInstrInfo : public SPUGenInstrInfo {
-    SPUTargetMachine &TM;
-    const SPURegisterInfo RI;
-  public:
-    explicit SPUInstrInfo(SPUTargetMachine &tm);
-
-    /// getRegisterInfo - TargetInstrInfo is a superset of MRegister info.  As
-    /// such, whenever a client has an instance of instruction info, it should
-    /// always be able to get register info as well (through this method).
-    ///
-    virtual const SPURegisterInfo &getRegisterInfo() const { return RI; }
-
-    ScheduleHazardRecognizer *
-    CreateTargetHazardRecognizer(const TargetMachine *TM,
-                                 const ScheduleDAG *DAG) const;
-
-    unsigned isLoadFromStackSlot(const MachineInstr *MI,
-                                 int &FrameIndex) const;
-    unsigned isStoreToStackSlot(const MachineInstr *MI,
-                                int &FrameIndex) const;
-
-    virtual void copyPhysReg(MachineBasicBlock &MBB,
-                             MachineBasicBlock::iterator I, DebugLoc DL,
-                             unsigned DestReg, unsigned SrcReg,
-                             bool KillSrc) const;
-
-    //! Store a register to a stack slot, based on its register class.
-    virtual void storeRegToStackSlot(MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator MBBI,
-                                     unsigned SrcReg, bool isKill, int FrameIndex,
-                                     const TargetRegisterClass *RC,
-                                     const TargetRegisterInfo *TRI) const;
-
-    //! Load a register from a stack slot, based on its register class.
-    virtual void loadRegFromStackSlot(MachineBasicBlock &MBB,
-                                      MachineBasicBlock::iterator MBBI,
-                                      unsigned DestReg, int FrameIndex,
-                                      const TargetRegisterClass *RC,
-                                      const TargetRegisterInfo *TRI) const;
-
-    //! Reverses a branch's condition, returning false on success.
-    virtual
-    bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const;
-
-    virtual bool AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
-                               MachineBasicBlock *&FBB,
-                               SmallVectorImpl<MachineOperand> &Cond,
-                               bool AllowModify) const;
-
-    virtual unsigned RemoveBranch(MachineBasicBlock &MBB) const;
-
-    virtual unsigned InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
-                                  MachineBasicBlock *FBB,
-                                  const SmallVectorImpl<MachineOperand> &Cond,
-                                  DebugLoc DL) const;
-   };
-}
-
-#endif
diff --git a/lib/Target/CellSPU/SPUInstrInfo.td b/lib/Target/CellSPU/SPUInstrInfo.td
deleted file mode 100644
index 117acd736aaa..000000000000
--- a/lib/Target/CellSPU/SPUInstrInfo.td
+++ /dev/null
@@ -1,4484 +0,0 @@
-//==- SPUInstrInfo.td - Describe the Cell SPU Instructions -*- tablegen -*-==//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-// Cell SPU Instructions:
-//===----------------------------------------------------------------------===//
-
-//===----------------------------------------------------------------------===//
-// TODO Items (not urgent today, but would be nice, low priority)
-//
-// ANDBI, ORBI: SPU constructs a 4-byte constant for these instructions by
-// concatenating the byte argument b as "bbbb". Could recognize this bit pattern
-// in 16-bit and 32-bit constants and reduce instruction count.
-//===----------------------------------------------------------------------===//
-
-//===----------------------------------------------------------------------===//
-// Pseudo instructions:
-//===----------------------------------------------------------------------===//
-
-let hasCtrlDep = 1, Defs = [R1], Uses = [R1] in {
-  def ADJCALLSTACKDOWN : Pseudo<(outs), (ins u16imm_i32:$amt),
-                                "${:comment} ADJCALLSTACKDOWN",
-                                [(callseq_start timm:$amt)]>;
-  def ADJCALLSTACKUP   : Pseudo<(outs), (ins u16imm_i32:$amt),
-                                "${:comment} ADJCALLSTACKUP",
-                                [(callseq_end timm:$amt)]>;
-  def HBR_LABEL        : Pseudo<(outs), (ins hbrtarget:$targ), 
-                                "$targ:\t${:comment}branch hint target",[ ]>;
-}
-
-//===----------------------------------------------------------------------===//
-// Loads:
-// NB: The ordering is actually important, since the instruction selection
-// will try each of the instructions in sequence, i.e., the D-form first with
-// the 10-bit displacement, then the A-form with the 16 bit displacement, and
-// finally the X-form with the register-register.
-//===----------------------------------------------------------------------===//
-
-let canFoldAsLoad = 1 in {
-  class LoadDFormVec<ValueType vectype>
-    : RI10Form<0b00101100, (outs VECREG:$rT), (ins dformaddr:$src),
-               "lqd\t$rT, $src",
-               LoadStore,
-               [(set (vectype VECREG:$rT), (load dform_addr:$src))]>
-  { }
-
-  class LoadDForm<RegisterClass rclass>
-    : RI10Form<0b00101100, (outs rclass:$rT), (ins dformaddr:$src),
-               "lqd\t$rT, $src",
-               LoadStore,
-               [(set rclass:$rT, (load dform_addr:$src))]>
-  { }
-
-  multiclass LoadDForms
-  {
-    def v16i8: LoadDFormVec<v16i8>;
-    def v8i16: LoadDFormVec<v8i16>;
-    def v4i32: LoadDFormVec<v4i32>;
-    def v2i64: LoadDFormVec<v2i64>;
-    def v4f32: LoadDFormVec<v4f32>;
-    def v2f64: LoadDFormVec<v2f64>;
-
-    def r128:  LoadDForm<GPRC>;
-    def r64:   LoadDForm<R64C>;
-    def r32:   LoadDForm<R32C>;
-    def f32:   LoadDForm<R32FP>;
-    def f64:   LoadDForm<R64FP>;
-    def r16:   LoadDForm<R16C>;
-    def r8:    LoadDForm<R8C>;
-  }
-
-  class LoadAFormVec<ValueType vectype>
-    : RI16Form<0b100001100, (outs VECREG:$rT), (ins addr256k:$src),
-               "lqa\t$rT, $src",
-               LoadStore,
-               [(set (vectype VECREG:$rT), (load aform_addr:$src))]>
-  { }
-
-  class LoadAForm<RegisterClass rclass>
-    : RI16Form<0b100001100, (outs rclass:$rT), (ins addr256k:$src),
-               "lqa\t$rT, $src",
-               LoadStore,
-               [(set rclass:$rT, (load aform_addr:$src))]>
-  { }
-
-  multiclass LoadAForms
-  {
-    def v16i8: LoadAFormVec<v16i8>;
-    def v8i16: LoadAFormVec<v8i16>;
-    def v4i32: LoadAFormVec<v4i32>;
-    def v2i64: LoadAFormVec<v2i64>;
-    def v4f32: LoadAFormVec<v4f32>;
-    def v2f64: LoadAFormVec<v2f64>;
-
-    def r128:  LoadAForm<GPRC>;
-    def r64:   LoadAForm<R64C>;
-    def r32:   LoadAForm<R32C>;
-    def f32:   LoadAForm<R32FP>;
-    def f64:   LoadAForm<R64FP>;
-    def r16:   LoadAForm<R16C>;
-    def r8:    LoadAForm<R8C>;
-  }
-
-  class LoadXFormVec<ValueType vectype>
-    : RRForm<0b00100011100, (outs VECREG:$rT), (ins memrr:$src),
-             "lqx\t$rT, $src",
-             LoadStore,
-             [(set (vectype VECREG:$rT), (load xform_addr:$src))]>
-  { }
-
-  class LoadXForm<RegisterClass rclass>
-    : RRForm<0b00100011100, (outs rclass:$rT), (ins memrr:$src),
-             "lqx\t$rT, $src",
-             LoadStore,
-             [(set rclass:$rT, (load xform_addr:$src))]>
-  { }
-
-  multiclass LoadXForms
-  {
-    def v16i8: LoadXFormVec<v16i8>;
-    def v8i16: LoadXFormVec<v8i16>;
-    def v4i32: LoadXFormVec<v4i32>;
-    def v2i64: LoadXFormVec<v2i64>;
-    def v4f32: LoadXFormVec<v4f32>;
-    def v2f64: LoadXFormVec<v2f64>;
-
-    def r128:  LoadXForm<GPRC>;
-    def r64:   LoadXForm<R64C>;
-    def r32:   LoadXForm<R32C>;
-    def f32:   LoadXForm<R32FP>;
-    def f64:   LoadXForm<R64FP>;
-    def r16:   LoadXForm<R16C>;
-    def r8:    LoadXForm<R8C>;
-  }
-
-  defm LQA : LoadAForms;
-  defm LQD : LoadDForms;
-  defm LQX : LoadXForms;
-
-/* Load quadword, PC relative: Not much use at this point in time.
-   Might be of use later for relocatable code. It's effectively the
-   same as LQA, but uses PC-relative addressing.
-  def LQR : RI16Form<0b111001100, (outs VECREG:$rT), (ins s16imm:$disp),
-                     "lqr\t$rT, $disp", LoadStore,
-                     [(set VECREG:$rT, (load iaddr:$disp))]>;
- */
-}
-
-//===----------------------------------------------------------------------===//
-// Stores:
-//===----------------------------------------------------------------------===//
-class StoreDFormVec<ValueType vectype>
-  : RI10Form<0b00100100, (outs), (ins VECREG:$rT, dformaddr:$src),
-             "stqd\t$rT, $src",
-             LoadStore,
-             [(store (vectype VECREG:$rT), dform_addr:$src)]>
-{ }
-
-class StoreDForm<RegisterClass rclass>
-  : RI10Form<0b00100100, (outs), (ins rclass:$rT, dformaddr:$src),
-             "stqd\t$rT, $src",
-             LoadStore,
-             [(store rclass:$rT, dform_addr:$src)]>
-{ }
-
-multiclass StoreDForms
-{
-  def v16i8: StoreDFormVec<v16i8>;
-  def v8i16: StoreDFormVec<v8i16>;
-  def v4i32: StoreDFormVec<v4i32>;
-  def v2i64: StoreDFormVec<v2i64>;
-  def v4f32: StoreDFormVec<v4f32>;
-  def v2f64: StoreDFormVec<v2f64>;
-
-  def r128:  StoreDForm<GPRC>;
-  def r64:   StoreDForm<R64C>;
-  def r32:   StoreDForm<R32C>;
-  def f32:   StoreDForm<R32FP>;
-  def f64:   StoreDForm<R64FP>;
-  def r16:   StoreDForm<R16C>;
-  def r8:    StoreDForm<R8C>;
-}
-
-class StoreAFormVec<ValueType vectype>
-  : RI16Form<0b0010010, (outs), (ins VECREG:$rT, addr256k:$src),
-             "stqa\t$rT, $src",
-             LoadStore,
-             [(store (vectype VECREG:$rT), aform_addr:$src)]>;
-
-class StoreAForm<RegisterClass rclass>
-  : RI16Form<0b001001, (outs), (ins rclass:$rT, addr256k:$src),
-             "stqa\t$rT, $src",
-             LoadStore,
-             [(store rclass:$rT, aform_addr:$src)]>;
-
-multiclass StoreAForms
-{
-  def v16i8: StoreAFormVec<v16i8>;
-  def v8i16: StoreAFormVec<v8i16>;
-  def v4i32: StoreAFormVec<v4i32>;
-  def v2i64: StoreAFormVec<v2i64>;
-  def v4f32: StoreAFormVec<v4f32>;
-  def v2f64: StoreAFormVec<v2f64>;
-
-  def r128:  StoreAForm<GPRC>;
-  def r64:   StoreAForm<R64C>;
-  def r32:   StoreAForm<R32C>;
-  def f32:   StoreAForm<R32FP>;
-  def f64:   StoreAForm<R64FP>;
-  def r16:   StoreAForm<R16C>;
-  def r8:    StoreAForm<R8C>;
-}
-
-class StoreXFormVec<ValueType vectype>
-  : RRForm<0b00100100, (outs), (ins VECREG:$rT, memrr:$src),
-           "stqx\t$rT, $src",
-           LoadStore,
-           [(store (vectype VECREG:$rT), xform_addr:$src)]>
-{ }
-
-class StoreXForm<RegisterClass rclass>
-  : RRForm<0b00100100, (outs), (ins rclass:$rT, memrr:$src),
-           "stqx\t$rT, $src",
-           LoadStore,
-           [(store rclass:$rT, xform_addr:$src)]>
-{ }
-
-multiclass StoreXForms
-{
-  def v16i8: StoreXFormVec<v16i8>;
-  def v8i16: StoreXFormVec<v8i16>;
-  def v4i32: StoreXFormVec<v4i32>;
-  def v2i64: StoreXFormVec<v2i64>;
-  def v4f32: StoreXFormVec<v4f32>;
-  def v2f64: StoreXFormVec<v2f64>;
-
-  def r128:  StoreXForm<GPRC>;
-  def r64:   StoreXForm<R64C>;
-  def r32:   StoreXForm<R32C>;
-  def f32:   StoreXForm<R32FP>;
-  def f64:   StoreXForm<R64FP>;
-  def r16:   StoreXForm<R16C>;
-  def r8:    StoreXForm<R8C>;
-}
-
-defm STQD : StoreDForms;
-defm STQA : StoreAForms;
-defm STQX : StoreXForms;
-
-/* Store quadword, PC relative: Not much use at this point in time. Might
-   be useful for relocatable code.
-def STQR : RI16Form<0b111000100, (outs), (ins VECREG:$rT, s16imm:$disp),
-                   "stqr\t$rT, $disp", LoadStore,
-                   [(store VECREG:$rT, iaddr:$disp)]>;
-*/
-
-//===----------------------------------------------------------------------===//
-// Generate Controls for Insertion:
-//===----------------------------------------------------------------------===//
-
-def CBD: RI7Form<0b10101111100, (outs VECREG:$rT), (ins shufaddr:$src),
-    "cbd\t$rT, $src", ShuffleOp,
-    [(set (v16i8 VECREG:$rT), (SPUshufmask dform2_addr:$src))]>;
-
-def CBX: RRForm<0b00101011100, (outs VECREG:$rT), (ins memrr:$src),
-    "cbx\t$rT, $src", ShuffleOp,
-    [(set (v16i8 VECREG:$rT), (SPUshufmask xform_addr:$src))]>;
-
-def CHD: RI7Form<0b10101111100, (outs VECREG:$rT), (ins shufaddr:$src),
-    "chd\t$rT, $src", ShuffleOp,
-    [(set (v8i16 VECREG:$rT), (SPUshufmask dform2_addr:$src))]>;
-
-def CHX: RRForm<0b10101011100, (outs VECREG:$rT), (ins memrr:$src),
-    "chx\t$rT, $src", ShuffleOp,
-    [(set (v8i16 VECREG:$rT), (SPUshufmask xform_addr:$src))]>;
-
-def CWD: RI7Form<0b01101111100, (outs VECREG:$rT), (ins shufaddr:$src),
-    "cwd\t$rT, $src", ShuffleOp,
-    [(set (v4i32 VECREG:$rT), (SPUshufmask dform2_addr:$src))]>;
-
-def CWX: RRForm<0b01101011100, (outs VECREG:$rT), (ins memrr:$src),
-    "cwx\t$rT, $src", ShuffleOp,
-    [(set (v4i32 VECREG:$rT), (SPUshufmask xform_addr:$src))]>;
-
-def CWDf32: RI7Form<0b01101111100, (outs VECREG:$rT), (ins shufaddr:$src),
-    "cwd\t$rT, $src", ShuffleOp,
-    [(set (v4f32 VECREG:$rT), (SPUshufmask dform2_addr:$src))]>;
-
-def CWXf32: RRForm<0b01101011100, (outs VECREG:$rT), (ins memrr:$src),
-    "cwx\t$rT, $src", ShuffleOp,
-    [(set (v4f32 VECREG:$rT), (SPUshufmask xform_addr:$src))]>;
-
-def CDD: RI7Form<0b11101111100, (outs VECREG:$rT), (ins shufaddr:$src),
-    "cdd\t$rT, $src", ShuffleOp,
-    [(set (v2i64 VECREG:$rT), (SPUshufmask dform2_addr:$src))]>;
-
-def CDX: RRForm<0b11101011100, (outs VECREG:$rT), (ins memrr:$src),
-    "cdx\t$rT, $src", ShuffleOp,
-    [(set (v2i64 VECREG:$rT), (SPUshufmask xform_addr:$src))]>;
-
-def CDDf64: RI7Form<0b11101111100, (outs VECREG:$rT), (ins shufaddr:$src),
-    "cdd\t$rT, $src", ShuffleOp,
-    [(set (v2f64 VECREG:$rT), (SPUshufmask dform2_addr:$src))]>;
-
-def CDXf64: RRForm<0b11101011100, (outs VECREG:$rT), (ins memrr:$src),
-    "cdx\t$rT, $src", ShuffleOp,
-    [(set (v2f64 VECREG:$rT), (SPUshufmask xform_addr:$src))]>;
-
-//===----------------------------------------------------------------------===//
-// Constant formation:
-//===----------------------------------------------------------------------===//
-
-def ILHv8i16:
-  RI16Form<0b110000010, (outs VECREG:$rT), (ins s16imm:$val),
-    "ilh\t$rT, $val", ImmLoad,
-    [(set (v8i16 VECREG:$rT), (v8i16 v8i16SExt16Imm:$val))]>;
-
-def ILHr16:
-  RI16Form<0b110000010, (outs R16C:$rT), (ins s16imm:$val),
-    "ilh\t$rT, $val", ImmLoad,
-    [(set R16C:$rT, immSExt16:$val)]>;
-
-// Cell SPU doesn't have a native 8-bit immediate load, but ILH works ("with
-// the right constant")
-def ILHr8:
-  RI16Form<0b110000010, (outs R8C:$rT), (ins s16imm_i8:$val),
-    "ilh\t$rT, $val", ImmLoad,
-    [(set R8C:$rT, immSExt8:$val)]>;
-
-// IL does sign extension!
-
-class ILInst<dag OOL, dag IOL, list<dag> pattern>:
-  RI16Form<0b100000010, OOL, IOL, "il\t$rT, $val",
-           ImmLoad, pattern>;
-
-class ILVecInst<ValueType vectype, Operand immtype, PatLeaf xform>:
-  ILInst<(outs VECREG:$rT), (ins immtype:$val),
-         [(set (vectype VECREG:$rT), (vectype xform:$val))]>;
-
-class ILRegInst<RegisterClass rclass, Operand immtype, PatLeaf xform>:
-  ILInst<(outs rclass:$rT), (ins immtype:$val),
-         [(set rclass:$rT, xform:$val)]>;
-
-multiclass ImmediateLoad
-{
-  def v2i64: ILVecInst<v2i64, s16imm_i64, v2i64SExt16Imm>;
-  def v4i32: ILVecInst<v4i32, s16imm_i32, v4i32SExt16Imm>;
-
-  // TODO: Need v2f64, v4f32
-
-  def r64: ILRegInst<R64C, s16imm_i64, immSExt16>;
-  def r32: ILRegInst<R32C, s16imm_i32, immSExt16>;
-  def f32: ILRegInst<R32FP, s16imm_f32, fpimmSExt16>;
-  def f64: ILRegInst<R64FP, s16imm_f64, fpimmSExt16>;
-}
-
-defm IL : ImmediateLoad;
-
-class ILHUInst<dag OOL, dag IOL, list<dag> pattern>:
-  RI16Form<0b010000010, OOL, IOL, "ilhu\t$rT, $val",
-           ImmLoad, pattern>;
-
-class ILHUVecInst<ValueType vectype, Operand immtype, PatLeaf xform>:
-  ILHUInst<(outs VECREG:$rT), (ins immtype:$val),
-           [(set (vectype VECREG:$rT), (vectype xform:$val))]>;
-
-class ILHURegInst<RegisterClass rclass, Operand immtype, PatLeaf xform>:
-  ILHUInst<(outs rclass:$rT), (ins immtype:$val),
-           [(set rclass:$rT, xform:$val)]>;
-
-multiclass ImmLoadHalfwordUpper
-{
-  def v2i64: ILHUVecInst<v2i64, u16imm_i64, immILHUvec_i64>;
-  def v4i32: ILHUVecInst<v4i32, u16imm_i32, immILHUvec>;
-
-  def r64: ILHURegInst<R64C, u16imm_i64, hi16>;
-  def r32: ILHURegInst<R32C, u16imm_i32, hi16>;
-
-  // Loads the high portion of an address
-  def hi: ILHURegInst<R32C, symbolHi, hi16>;
-
-  // Used in custom lowering constant SFP loads:
-  def f32: ILHURegInst<R32FP, f16imm, hi16_f32>;
-}
-
-defm ILHU : ImmLoadHalfwordUpper;
-
-// Immediate load address (can also be used to load 18-bit unsigned constants,
-// see the zext 16->32 pattern)
-
-class ILAInst<dag OOL, dag IOL, list<dag> pattern>:
-  RI18Form<0b1000010, OOL, IOL, "ila\t$rT, $val",
-           LoadNOP, pattern>;
-
-class ILAVecInst<ValueType vectype, Operand immtype, PatLeaf xform>:
-  ILAInst<(outs VECREG:$rT), (ins immtype:$val),
-          [(set (vectype VECREG:$rT), (vectype xform:$val))]>;
-
-class ILARegInst<RegisterClass rclass, Operand immtype, PatLeaf xform>:
-  ILAInst<(outs rclass:$rT), (ins immtype:$val),
-          [(set rclass:$rT, xform:$val)]>;
-
-multiclass ImmLoadAddress
-{
-  def v2i64: ILAVecInst<v2i64, u18imm, v2i64Uns18Imm>;
-  def v4i32: ILAVecInst<v4i32, u18imm, v4i32Uns18Imm>;
-
-  def r64: ILARegInst<R64C, u18imm_i64, imm18>;
-  def r32: ILARegInst<R32C, u18imm, imm18>;
-  def f32: ILARegInst<R32FP, f18imm, fpimm18>;
-  def f64: ILARegInst<R64FP, f18imm_f64, fpimm18>;
-
-  def hi: ILARegInst<R32C, symbolHi, imm18>;
-  def lo: ILARegInst<R32C, symbolLo, imm18>;
-
-  def lsa: ILAInst<(outs R32C:$rT), (ins symbolLSA:$val),
-                   [(set R32C:$rT, imm18:$val)]>;
-}
-
-defm ILA : ImmLoadAddress;
-
-// Immediate OR, Halfword Lower: The "other" part of loading large constants
-// into 32-bit registers. See the anonymous pattern Pat<(i32 imm:$imm), ...>
-// Note that these are really two operand instructions, but they're encoded
-// as three operands with the first two arguments tied-to each other.
-
-class IOHLInst<dag OOL, dag IOL, list<dag> pattern>:
-  RI16Form<0b100000110, OOL, IOL, "iohl\t$rT, $val",
-           ImmLoad, pattern>,
-  RegConstraint<"$rS = $rT">,
-  NoEncode<"$rS">;
-
-class IOHLVecInst<ValueType vectype, Operand immtype /* , PatLeaf xform */>:
-  IOHLInst<(outs VECREG:$rT), (ins VECREG:$rS, immtype:$val),
-           [/* no pattern */]>;
-
-class IOHLRegInst<RegisterClass rclass, Operand immtype /* , PatLeaf xform */>:
-  IOHLInst<(outs rclass:$rT), (ins rclass:$rS, immtype:$val),
-           [/* no pattern */]>;
-
-multiclass ImmOrHalfwordLower
-{
-  def v2i64: IOHLVecInst<v2i64, u16imm_i64>;
-  def v4i32: IOHLVecInst<v4i32, u16imm_i32>;
-
-  def r32: IOHLRegInst<R32C, i32imm>;
-  def f32: IOHLRegInst<R32FP, f32imm>;
-
-  def lo: IOHLRegInst<R32C, symbolLo>;
-}
-
-defm IOHL: ImmOrHalfwordLower;
-
-// Form select mask for bytes using immediate, used in conjunction with the
-// SELB instruction:
-
-class FSMBIVec<ValueType vectype>:
-  RI16Form<0b101001100, (outs VECREG:$rT), (ins u16imm:$val),
-          "fsmbi\t$rT, $val",
-          SelectOp,
-          [(set (vectype VECREG:$rT), (SPUselmask (i16 immU16:$val)))]>;
-
-multiclass FormSelectMaskBytesImm
-{
-  def v16i8: FSMBIVec<v16i8>;
-  def v8i16: FSMBIVec<v8i16>;
-  def v4i32: FSMBIVec<v4i32>;
-  def v2i64: FSMBIVec<v2i64>;
-}
-
-defm FSMBI : FormSelectMaskBytesImm;
-
-// fsmb: Form select mask for bytes. N.B. Input operand, $rA, is 16-bits
-class FSMBInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm_1<0b01101101100, OOL, IOL, "fsmb\t$rT, $rA", SelectOp,
-             pattern>;
-
-class FSMBRegInst<RegisterClass rclass, ValueType vectype>:
-    FSMBInst<(outs VECREG:$rT), (ins rclass:$rA),
-             [(set (vectype VECREG:$rT), (SPUselmask rclass:$rA))]>;
-
-class FSMBVecInst<ValueType vectype>:
-    FSMBInst<(outs VECREG:$rT), (ins VECREG:$rA),
-             [(set (vectype VECREG:$rT),
-                   (SPUselmask (vectype VECREG:$rA)))]>;
-
-multiclass FormSelectMaskBits {
-  def v16i8_r16: FSMBRegInst<R16C, v16i8>;
-  def v16i8:     FSMBVecInst<v16i8>;
-}
-
-defm FSMB: FormSelectMaskBits;
-
-// fsmh: Form select mask for halfwords. N.B., Input operand, $rA, is
-// only 8-bits wide (even though it's input as 16-bits here)
-
-class FSMHInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm_1<0b10101101100, OOL, IOL, "fsmh\t$rT, $rA", SelectOp,
-             pattern>;
-
-class FSMHRegInst<RegisterClass rclass, ValueType vectype>:
-    FSMHInst<(outs VECREG:$rT), (ins rclass:$rA),
-             [(set (vectype VECREG:$rT), (SPUselmask rclass:$rA))]>;
-
-class FSMHVecInst<ValueType vectype>:
-    FSMHInst<(outs VECREG:$rT), (ins VECREG:$rA),
-             [(set (vectype VECREG:$rT),
-                   (SPUselmask (vectype VECREG:$rA)))]>;
-
-multiclass FormSelectMaskHalfword {
-  def v8i16_r16: FSMHRegInst<R16C, v8i16>;
-  def v8i16:     FSMHVecInst<v8i16>;
-}
-
-defm FSMH: FormSelectMaskHalfword;
-
-// fsm: Form select mask for words. Like the other fsm* instructions,
-// only the lower 4 bits of $rA are significant.
-
-class FSMInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm_1<0b00101101100, OOL, IOL, "fsm\t$rT, $rA", SelectOp,
-             pattern>;
-
-class FSMRegInst<ValueType vectype, RegisterClass rclass>:
-    FSMInst<(outs VECREG:$rT), (ins rclass:$rA),
-            [(set (vectype VECREG:$rT), (SPUselmask rclass:$rA))]>;
-
-class FSMVecInst<ValueType vectype>:
-    FSMInst<(outs VECREG:$rT), (ins VECREG:$rA),
-            [(set (vectype VECREG:$rT), (SPUselmask (vectype VECREG:$rA)))]>;
-
-multiclass FormSelectMaskWord {
-  def v4i32: FSMVecInst<v4i32>;
-
-  def r32 :  FSMRegInst<v4i32, R32C>;
-  def r16 :  FSMRegInst<v4i32, R16C>;
-}
-
-defm FSM : FormSelectMaskWord;
-
-// Special case when used for i64 math operations
-multiclass FormSelectMaskWord64 {
-  def r32 : FSMRegInst<v2i64, R32C>;
-  def r16 : FSMRegInst<v2i64, R16C>;
-}
-
-defm FSM64 : FormSelectMaskWord64;
-
-//===----------------------------------------------------------------------===//
-// Integer and Logical Operations:
-//===----------------------------------------------------------------------===//
-
-def AHv8i16:
-  RRForm<0b00010011000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "ah\t$rT, $rA, $rB", IntegerOp,
-    [(set (v8i16 VECREG:$rT), (int_spu_si_ah VECREG:$rA, VECREG:$rB))]>;
-
-def : Pat<(add (v8i16 VECREG:$rA), (v8i16 VECREG:$rB)),
-          (AHv8i16 VECREG:$rA, VECREG:$rB)>;
-
-def AHr16:
-  RRForm<0b00010011000, (outs R16C:$rT), (ins R16C:$rA, R16C:$rB),
-    "ah\t$rT, $rA, $rB", IntegerOp,
-    [(set R16C:$rT, (add R16C:$rA, R16C:$rB))]>;
-
-def AHIvec:
-    RI10Form<0b10111000, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-      "ahi\t$rT, $rA, $val", IntegerOp,
-      [(set (v8i16 VECREG:$rT), (add (v8i16 VECREG:$rA),
-                                     v8i16SExt10Imm:$val))]>;
-
-def AHIr16:
-  RI10Form<0b10111000, (outs R16C:$rT), (ins R16C:$rA, s10imm:$val),
-    "ahi\t$rT, $rA, $val", IntegerOp,
-    [(set R16C:$rT, (add R16C:$rA, i16ImmSExt10:$val))]>;
-
-// v4i32, i32 add instruction:
-
-class AInst<dag OOL, dag IOL, list<dag> pattern>:
-  RRForm<0b00000011000, OOL, IOL,
-         "a\t$rT, $rA, $rB", IntegerOp,
-         pattern>;
-
-class AVecInst<ValueType vectype>:
-  AInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-        [(set (vectype VECREG:$rT), (add (vectype VECREG:$rA),
-                                         (vectype VECREG:$rB)))]>;
-
-class ARegInst<RegisterClass rclass>:
-  AInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-        [(set rclass:$rT, (add rclass:$rA, rclass:$rB))]>;
-        
-multiclass AddInstruction {
-  def v4i32: AVecInst<v4i32>;
-  def v16i8: AVecInst<v16i8>;
-  def r32:   ARegInst<R32C>;
-}
-
-defm A : AddInstruction;
-
-class AIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI10Form<0b00111000, OOL, IOL,
-             "ai\t$rT, $rA, $val", IntegerOp,
-             pattern>;
-
-class AIVecInst<ValueType vectype, PatLeaf immpred>:
-    AIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-            [(set (vectype VECREG:$rT), (add (vectype VECREG:$rA), immpred:$val))]>;
-
-class AIFPVecInst<ValueType vectype, PatLeaf immpred>:
-    AIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-            [/* no pattern */]>;
-
-class AIRegInst<RegisterClass rclass, PatLeaf immpred>:
-    AIInst<(outs rclass:$rT), (ins rclass:$rA, s10imm_i32:$val),
-           [(set rclass:$rT, (add rclass:$rA, immpred:$val))]>;
-
-// This is used to add epsilons to floating point numbers in the f32 fdiv code:
-class AIFPInst<RegisterClass rclass, PatLeaf immpred>:
-    AIInst<(outs rclass:$rT), (ins rclass:$rA, s10imm_i32:$val),
-           [/* no pattern */]>;
-
-multiclass AddImmediate {
-  def v4i32: AIVecInst<v4i32, v4i32SExt10Imm>;
-
-  def r32: AIRegInst<R32C, i32ImmSExt10>;
-
-  def v4f32: AIFPVecInst<v4f32, v4i32SExt10Imm>;
-  def f32: AIFPInst<R32FP, i32ImmSExt10>;
-}
-
-defm AI : AddImmediate;
-
-def SFHvec:
-    RRForm<0b00010010000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "sfh\t$rT, $rA, $rB", IntegerOp,
-      [(set (v8i16 VECREG:$rT), (sub (v8i16 VECREG:$rA),
-                                     (v8i16 VECREG:$rB)))]>;
-
-def SFHr16:
-    RRForm<0b00010010000, (outs R16C:$rT), (ins R16C:$rA, R16C:$rB),
-      "sfh\t$rT, $rA, $rB", IntegerOp,
-      [(set R16C:$rT, (sub R16C:$rB, R16C:$rA))]>;
-
-def SFHIvec:
-    RI10Form<0b10110000, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-      "sfhi\t$rT, $rA, $val", IntegerOp,
-      [(set (v8i16 VECREG:$rT), (sub v8i16SExt10Imm:$val,
-                                     (v8i16 VECREG:$rA)))]>;
-
-def SFHIr16 : RI10Form<0b10110000, (outs R16C:$rT), (ins R16C:$rA, s10imm:$val),
-  "sfhi\t$rT, $rA, $val", IntegerOp,
-  [(set R16C:$rT, (sub i16ImmSExt10:$val, R16C:$rA))]>;
-
-def SFvec : RRForm<0b00000010000, (outs VECREG:$rT),
-                                  (ins VECREG:$rA, VECREG:$rB),
-  "sf\t$rT, $rA, $rB", IntegerOp,
-  [(set (v4i32 VECREG:$rT), (sub (v4i32 VECREG:$rB), (v4i32 VECREG:$rA)))]>;
-
-
-def SFr32 : RRForm<0b00000010000, (outs R32C:$rT), (ins R32C:$rA, R32C:$rB),
-  "sf\t$rT, $rA, $rB", IntegerOp,
-  [(set R32C:$rT, (sub R32C:$rB, R32C:$rA))]>;
-
-def SFIvec:
-    RI10Form<0b00110000, (outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-      "sfi\t$rT, $rA, $val", IntegerOp,
-      [(set (v4i32 VECREG:$rT), (sub v4i32SExt10Imm:$val,
-                                     (v4i32 VECREG:$rA)))]>;
-
-def SFIr32 : RI10Form<0b00110000, (outs R32C:$rT),
-                                  (ins R32C:$rA, s10imm_i32:$val),
-  "sfi\t$rT, $rA, $val", IntegerOp,
-  [(set R32C:$rT, (sub i32ImmSExt10:$val, R32C:$rA))]>;
-
-// ADDX: only available in vector form, doesn't match a pattern.
-class ADDXInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b00000010110, OOL, IOL,
-      "addx\t$rT, $rA, $rB",
-      IntegerOp, pattern>;
-
-class ADDXVecInst<ValueType vectype>:
-    ADDXInst<(outs VECREG:$rT),
-             (ins VECREG:$rA, VECREG:$rB, VECREG:$rCarry),
-             [/* no pattern */]>,
-    RegConstraint<"$rCarry = $rT">,
-    NoEncode<"$rCarry">;
-
-class ADDXRegInst<RegisterClass rclass>:
-    ADDXInst<(outs rclass:$rT),
-             (ins rclass:$rA, rclass:$rB, rclass:$rCarry),
-             [/* no pattern */]>,
-    RegConstraint<"$rCarry = $rT">,
-    NoEncode<"$rCarry">;
-
-multiclass AddExtended {
-  def v2i64 : ADDXVecInst<v2i64>;
-  def v4i32 : ADDXVecInst<v4i32>;
-  def r64 : ADDXRegInst<R64C>;
-  def r32 : ADDXRegInst<R32C>;
-}
-
-defm ADDX : AddExtended;
-
-// CG: Generate carry for add
-class CGInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b01000011000, OOL, IOL,
-      "cg\t$rT, $rA, $rB",
-      IntegerOp, pattern>;
-
-class CGVecInst<ValueType vectype>:
-    CGInst<(outs VECREG:$rT),
-           (ins VECREG:$rA, VECREG:$rB),
-           [/* no pattern */]>;
-
-class CGRegInst<RegisterClass rclass>:
-    CGInst<(outs rclass:$rT),
-           (ins rclass:$rA, rclass:$rB),
-           [/* no pattern */]>;
-
-multiclass CarryGenerate {
-  def v2i64 : CGVecInst<v2i64>;
-  def v4i32 : CGVecInst<v4i32>;
-  def r64 : CGRegInst<R64C>;
-  def r32 : CGRegInst<R32C>;
-}
-
-defm CG : CarryGenerate;
-
-// SFX: Subract from, extended. This is used in conjunction with BG to subtract
-// with carry (borrow, in this case)
-class SFXInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10000010110, OOL, IOL,
-      "sfx\t$rT, $rA, $rB",
-      IntegerOp, pattern>;
-
-class SFXVecInst<ValueType vectype>:
-    SFXInst<(outs VECREG:$rT),
-            (ins VECREG:$rA, VECREG:$rB, VECREG:$rCarry),
-             [/* no pattern */]>,
-    RegConstraint<"$rCarry = $rT">,
-    NoEncode<"$rCarry">;
-
-class SFXRegInst<RegisterClass rclass>:
-    SFXInst<(outs rclass:$rT),
-            (ins rclass:$rA, rclass:$rB, rclass:$rCarry),
-             [/* no pattern */]>,
-    RegConstraint<"$rCarry = $rT">,
-    NoEncode<"$rCarry">;
-
-multiclass SubtractExtended {
-  def v2i64 : SFXVecInst<v2i64>;
-  def v4i32 : SFXVecInst<v4i32>;
-  def r64 : SFXRegInst<R64C>;
-  def r32 : SFXRegInst<R32C>;
-}
-
-defm SFX : SubtractExtended;
-
-// BG: only available in vector form, doesn't match a pattern.
-class BGInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b01000010000, OOL, IOL,
-      "bg\t$rT, $rA, $rB",
-      IntegerOp, pattern>;
-
-class BGVecInst<ValueType vectype>:
-    BGInst<(outs VECREG:$rT),
-           (ins VECREG:$rA, VECREG:$rB),
-           [/* no pattern */]>;
-
-class BGRegInst<RegisterClass rclass>:
-    BGInst<(outs rclass:$rT),
-           (ins rclass:$rA, rclass:$rB),
-           [/* no pattern */]>;
-
-multiclass BorrowGenerate {
-  def v4i32 : BGVecInst<v4i32>;
-  def v2i64 : BGVecInst<v2i64>;
-  def r64 : BGRegInst<R64C>;
-  def r32 : BGRegInst<R32C>;
-}
-
-defm BG : BorrowGenerate;
-
-// BGX: Borrow generate, extended.
-def BGXvec:
-    RRForm<0b11000010110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB,
-                                VECREG:$rCarry),
-      "bgx\t$rT, $rA, $rB", IntegerOp,
-      []>,
-    RegConstraint<"$rCarry = $rT">,
-    NoEncode<"$rCarry">;
-
-// Halfword multiply variants:
-// N.B: These can be used to build up larger quantities (16x16 -> 32)
-
-def MPYv8i16:
-  RRForm<0b00100011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-    "mpy\t$rT, $rA, $rB", IntegerMulDiv,
-    [/* no pattern */]>;
-
-def MPYr16:
-  RRForm<0b00100011110, (outs R16C:$rT), (ins R16C:$rA, R16C:$rB),
-    "mpy\t$rT, $rA, $rB", IntegerMulDiv,
-    [(set R16C:$rT, (mul R16C:$rA, R16C:$rB))]>;
-
-// Unsigned 16-bit multiply:
-
-class MPYUInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b00110011110, OOL, IOL,
-      "mpyu\t$rT, $rA, $rB", IntegerMulDiv,
-      pattern>;
-
-def MPYUv4i32:
-  MPYUInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-           [/* no pattern */]>;
-
-def MPYUr16:
-  MPYUInst<(outs R32C:$rT), (ins R16C:$rA, R16C:$rB),
-           [(set R32C:$rT, (mul (zext R16C:$rA), (zext R16C:$rB)))]>;
-
-def MPYUr32:
-  MPYUInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB),
-           [/* no pattern */]>;
-
-// mpyi: multiply 16 x s10imm -> 32 result.
-
-class MPYIInst<dag OOL, dag IOL, list<dag> pattern>:
-  RI10Form<0b00101110, OOL, IOL,
-    "mpyi\t$rT, $rA, $val", IntegerMulDiv,
-    pattern>;
-
-def MPYIvec:
-  MPYIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-           [(set (v8i16 VECREG:$rT),
-                 (mul (v8i16 VECREG:$rA), v8i16SExt10Imm:$val))]>;
-
-def MPYIr16:
-  MPYIInst<(outs R16C:$rT), (ins R16C:$rA, s10imm:$val),
-           [(set R16C:$rT, (mul R16C:$rA, i16ImmSExt10:$val))]>;
-
-// mpyui: same issues as other multiplies, plus, this doesn't match a
-// pattern... but may be used during target DAG selection or lowering
-
-class MPYUIInst<dag OOL, dag IOL, list<dag> pattern>:
-  RI10Form<0b10101110, OOL, IOL,
-           "mpyui\t$rT, $rA, $val", IntegerMulDiv,
-           pattern>;
-    
-def MPYUIvec:
-  MPYUIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-            []>;
-
-def MPYUIr16:
-  MPYUIInst<(outs R16C:$rT), (ins R16C:$rA, s10imm:$val),
-            []>;
-
-// mpya: 16 x 16 + 16 -> 32 bit result
-class MPYAInst<dag OOL, dag IOL, list<dag> pattern>:
-  RRRForm<0b0011, OOL, IOL,
-          "mpya\t$rT, $rA, $rB, $rC", IntegerMulDiv,
-          pattern>;
-          
-def MPYAv4i32:
-  MPYAInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-           [(set (v4i32 VECREG:$rT),
-                 (add (v4i32 (bitconvert (mul (v8i16 VECREG:$rA),
-                                              (v8i16 VECREG:$rB)))),
-                      (v4i32 VECREG:$rC)))]>;
-
-def MPYAr32:
-  MPYAInst<(outs R32C:$rT), (ins R16C:$rA, R16C:$rB, R32C:$rC),
-           [(set R32C:$rT, (add (sext (mul R16C:$rA, R16C:$rB)),
-                                R32C:$rC))]>;
-                                
-def MPYAr32_sext:
-  MPYAInst<(outs R32C:$rT), (ins R16C:$rA, R16C:$rB, R32C:$rC),
-           [(set R32C:$rT, (add (mul (sext R16C:$rA), (sext R16C:$rB)),
-                                R32C:$rC))]>;
-
-def MPYAr32_sextinreg:
-  MPYAInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB, R32C:$rC),
-           [(set R32C:$rT, (add (mul (sext_inreg R32C:$rA, i16),
-                                     (sext_inreg R32C:$rB, i16)),
-                                R32C:$rC))]>;
-
-// mpyh: multiply high, used to synthesize 32-bit multiplies
-class MPYHInst<dag OOL, dag IOL, list<dag> pattern>:
-  RRForm<0b10100011110, OOL, IOL,
-         "mpyh\t$rT, $rA, $rB", IntegerMulDiv,
-         pattern>;
-         
-def MPYHv4i32:
-    MPYHInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-             [/* no pattern */]>;
-
-def MPYHr32:
-    MPYHInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB),
-             [/* no pattern */]>;
-
-// mpys: multiply high and shift right (returns the top half of
-// a 16-bit multiply, sign extended to 32 bits.)
-
-class MPYSInst<dag OOL, dag IOL>:
-    RRForm<0b11100011110, OOL, IOL, 
-      "mpys\t$rT, $rA, $rB", IntegerMulDiv,
-      [/* no pattern */]>;
-
-def MPYSv4i32:
-    MPYSInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB)>;
-    
-def MPYSr16:
-    MPYSInst<(outs R32C:$rT), (ins R16C:$rA, R16C:$rB)>;
-
-// mpyhh: multiply high-high (returns the 32-bit result from multiplying
-// the top 16 bits of the $rA, $rB)
-
-class MPYHHInst<dag OOL, dag IOL>:
-  RRForm<0b01100011110, OOL, IOL,
-        "mpyhh\t$rT, $rA, $rB", IntegerMulDiv,
-        [/* no pattern */]>;
-        
-def MPYHHv8i16:
-    MPYHHInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB)>;
-
-def MPYHHr32:
-    MPYHHInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB)>;
-
-// mpyhha: Multiply high-high, add to $rT:
-
-class MPYHHAInst<dag OOL, dag IOL>:
-    RRForm<0b01100010110, OOL, IOL,
-      "mpyhha\t$rT, $rA, $rB", IntegerMulDiv,
-      [/* no pattern */]>;
-
-def MPYHHAvec:
-    MPYHHAInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB)>;
-    
-def MPYHHAr32:
-    MPYHHAInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB)>;
-
-// mpyhhu: Multiply high-high, unsigned, e.g.:
-//
-// +-------+-------+   +-------+-------+   +---------+
-// |  a0   .  a1   | x |  b0   .  b1   | = | a0 x b0 |
-// +-------+-------+   +-------+-------+   +---------+
-//
-// where a0, b0 are the upper 16 bits of the 32-bit word
-
-class MPYHHUInst<dag OOL, dag IOL>:
-    RRForm<0b01110011110, OOL, IOL,
-      "mpyhhu\t$rT, $rA, $rB", IntegerMulDiv,
-      [/* no pattern */]>;
-
-def MPYHHUv4i32:
-    MPYHHUInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB)>;
-    
-def MPYHHUr32:
-    MPYHHUInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB)>;
-
-// mpyhhau: Multiply high-high, unsigned
-
-class MPYHHAUInst<dag OOL, dag IOL>:
-    RRForm<0b01110010110, OOL, IOL,
-      "mpyhhau\t$rT, $rA, $rB", IntegerMulDiv,
-      [/* no pattern */]>;
-
-def MPYHHAUvec:
-    MPYHHAUInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB)>;
-    
-def MPYHHAUr32:
-    MPYHHAUInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB)>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// clz: Count leading zeroes
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-class CLZInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm_1<0b10100101010, OOL, IOL, "clz\t$rT, $rA",
-             IntegerOp, pattern>;
-
-class CLZRegInst<RegisterClass rclass>:
-    CLZInst<(outs rclass:$rT), (ins rclass:$rA),
-            [(set rclass:$rT, (ctlz rclass:$rA))]>;
-
-class CLZVecInst<ValueType vectype>:
-    CLZInst<(outs VECREG:$rT), (ins VECREG:$rA),
-            [(set (vectype VECREG:$rT), (ctlz (vectype VECREG:$rA)))]>;
-
-multiclass CountLeadingZeroes {
-  def v4i32 : CLZVecInst<v4i32>;
-  def r32   : CLZRegInst<R32C>;
-}
-
-defm CLZ : CountLeadingZeroes;
-
-// cntb: Count ones in bytes (aka "population count")
-//
-// NOTE: This instruction is really a vector instruction, but the custom
-// lowering code uses it in unorthodox ways to support CTPOP for other
-// data types!
-
-def CNTBv16i8:
-    RRForm_1<0b00101101010, (outs VECREG:$rT), (ins VECREG:$rA),
-      "cntb\t$rT, $rA", IntegerOp,
-      [(set (v16i8 VECREG:$rT), (SPUcntb (v16i8 VECREG:$rA)))]>;
-
-def CNTBv8i16 :
-    RRForm_1<0b00101101010, (outs VECREG:$rT), (ins VECREG:$rA),
-      "cntb\t$rT, $rA", IntegerOp,
-      [(set (v8i16 VECREG:$rT), (SPUcntb (v8i16 VECREG:$rA)))]>;
-
-def CNTBv4i32 :
-    RRForm_1<0b00101101010, (outs VECREG:$rT), (ins VECREG:$rA),
-      "cntb\t$rT, $rA", IntegerOp,
-      [(set (v4i32 VECREG:$rT), (SPUcntb (v4i32 VECREG:$rA)))]>;
-
-// gbb: Gather the low order bits from each byte in $rA into a single 16-bit
-// quantity stored into $rT's slot 0, upper 16 bits are zeroed, as are
-// slots 1-3.
-//
-// Note: This instruction "pairs" with the fsmb instruction for all of the
-// various types defined here.
-//
-// Note 2: The "VecInst" and "RegInst" forms refer to the result being either
-// a vector or register.
-
-class GBBInst<dag OOL, dag IOL, list<dag> pattern>:
-  RRForm_1<0b01001101100, OOL, IOL, "gbb\t$rT, $rA", GatherOp, pattern>;
-
-class GBBRegInst<RegisterClass rclass, ValueType vectype>:
-  GBBInst<(outs rclass:$rT), (ins VECREG:$rA),
-          [/* no pattern */]>;
-
-class GBBVecInst<ValueType vectype>:
-  GBBInst<(outs VECREG:$rT), (ins VECREG:$rA),
-          [/* no pattern */]>;
-
-multiclass GatherBitsFromBytes {
-  def v16i8_r32: GBBRegInst<R32C, v16i8>;
-  def v16i8_r16: GBBRegInst<R16C, v16i8>;
-  def v16i8:     GBBVecInst<v16i8>;
-}
-
-defm GBB: GatherBitsFromBytes;
-
-// gbh: Gather all low order bits from each halfword in $rA into a single
-// 8-bit quantity stored in $rT's slot 0, with the upper bits of $rT set to 0
-// and slots 1-3 also set to 0.
-//
-// See notes for GBBInst, above.
-
-class GBHInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm_1<0b10001101100, OOL, IOL, "gbh\t$rT, $rA", GatherOp,
-             pattern>;
-
-class GBHRegInst<RegisterClass rclass, ValueType vectype>:
-    GBHInst<(outs rclass:$rT), (ins VECREG:$rA),
-            [/* no pattern */]>;
-
-class GBHVecInst<ValueType vectype>:
-    GBHInst<(outs VECREG:$rT), (ins VECREG:$rA),
-            [/* no pattern */]>;
-
-multiclass GatherBitsHalfword {
-  def v8i16_r32: GBHRegInst<R32C, v8i16>;
-  def v8i16_r16: GBHRegInst<R16C, v8i16>;
-  def v8i16:     GBHVecInst<v8i16>;
-}
-
-defm GBH: GatherBitsHalfword;
-
-// gb: Gather all low order bits from each word in $rA into a single
-// 4-bit quantity stored in $rT's slot 0, upper bits in $rT set to 0,
-// as well as slots 1-3.
-//
-// See notes for gbb, above.
-
-class GBInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm_1<0b00001101100, OOL, IOL, "gb\t$rT, $rA", GatherOp,
-             pattern>;
-
-class GBRegInst<RegisterClass rclass, ValueType vectype>:
-    GBInst<(outs rclass:$rT), (ins VECREG:$rA),
-           [/* no pattern */]>;
-
-class GBVecInst<ValueType vectype>:
-    GBInst<(outs VECREG:$rT), (ins VECREG:$rA),
-           [/* no pattern */]>;
-
-multiclass GatherBitsWord {
-  def v4i32_r32: GBRegInst<R32C, v4i32>;
-  def v4i32_r16: GBRegInst<R16C, v4i32>;
-  def v4i32:     GBVecInst<v4i32>;
-}
-
-defm GB: GatherBitsWord;
-
-// avgb: average bytes
-def AVGB:
-    RRForm<0b11001011000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "avgb\t$rT, $rA, $rB", ByteOp,
-      []>;
-
-// absdb: absolute difference of bytes
-def ABSDB:
-    RRForm<0b11001010000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "absdb\t$rT, $rA, $rB", ByteOp,
-      []>;
-
-// sumb: sum bytes into halfwords
-def SUMB:
-    RRForm<0b11001010010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "sumb\t$rT, $rA, $rB", ByteOp,
-      []>;
-
-// Sign extension operations:
-class XSBHInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm_1<0b01101101010, OOL, IOL,
-      "xsbh\t$rDst, $rSrc",
-      IntegerOp, pattern>;
-
-class XSBHInRegInst<RegisterClass rclass, list<dag> pattern>:
-    XSBHInst<(outs rclass:$rDst), (ins rclass:$rSrc),
-             pattern>;
-
-multiclass ExtendByteHalfword {
-  def v16i8:     XSBHInst<(outs VECREG:$rDst), (ins VECREG:$rSrc),
-                          [
-                  /*(set (v8i16 VECREG:$rDst), (sext (v8i16 VECREG:$rSrc)))*/]>;
-  def r8:        XSBHInst<(outs R16C:$rDst), (ins R8C:$rSrc),
-                          [(set R16C:$rDst, (sext R8C:$rSrc))]>;
-  def r16:       XSBHInRegInst<R16C,
-                               [(set R16C:$rDst, (sext_inreg R16C:$rSrc, i8))]>;
-
-  // 32-bit form for XSBH: used to sign extend 8-bit quantities to 16-bit
-  // quantities to 32-bit quantities via a 32-bit register (see the sext 8->32
-  // pattern below). Intentionally doesn't match a pattern because we want the
-  // sext 8->32 pattern to do the work for us, namely because we need the extra
-  // XSHWr32.
-  def r32:   XSBHInRegInst<R32C, [/* no pattern */]>;
-  
-  // Same as the 32-bit version, but for i64
-  def r64:   XSBHInRegInst<R64C, [/* no pattern */]>;
-}
-
-defm XSBH : ExtendByteHalfword;
-
-// Sign extend halfwords to words:
-
-class XSHWInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm_1<0b01101101010, OOL, IOL, "xshw\t$rDest, $rSrc",
-            IntegerOp, pattern>;
-
-class XSHWVecInst<ValueType in_vectype, ValueType out_vectype>:
-    XSHWInst<(outs VECREG:$rDest), (ins VECREG:$rSrc),
-             [(set (out_vectype VECREG:$rDest),
-                   (sext (in_vectype VECREG:$rSrc)))]>;
-
-class XSHWInRegInst<RegisterClass rclass, list<dag> pattern>:
-    XSHWInst<(outs rclass:$rDest), (ins rclass:$rSrc),
-             pattern>;
-             
-class XSHWRegInst<RegisterClass rclass>:
-    XSHWInst<(outs rclass:$rDest), (ins R16C:$rSrc),
-             [(set rclass:$rDest, (sext R16C:$rSrc))]>;
-
-multiclass ExtendHalfwordWord {
-  def v4i32: XSHWVecInst<v8i16, v4i32>;
-
-  def r16:   XSHWRegInst<R32C>;
-
-  def r32:   XSHWInRegInst<R32C,
-                          [(set R32C:$rDest, (sext_inreg R32C:$rSrc, i16))]>;
-  def r64:   XSHWInRegInst<R64C, [/* no pattern */]>;
-}
-
-defm XSHW : ExtendHalfwordWord;
-
-// Sign-extend words to doublewords (32->64 bits)
-
-class XSWDInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm_1<0b01100101010, OOL, IOL, "xswd\t$rDst, $rSrc",
-              IntegerOp, pattern>;
-      
-class XSWDVecInst<ValueType in_vectype, ValueType out_vectype>:
-    XSWDInst<(outs VECREG:$rDst), (ins VECREG:$rSrc),
-             [/*(set (out_vectype VECREG:$rDst),
-                   (sext (out_vectype VECREG:$rSrc)))*/]>;
-      
-class XSWDRegInst<RegisterClass in_rclass, RegisterClass out_rclass>:
-    XSWDInst<(outs out_rclass:$rDst), (ins in_rclass:$rSrc),
-             [(set out_rclass:$rDst, (sext in_rclass:$rSrc))]>;
-             
-multiclass ExtendWordToDoubleWord {
-  def v2i64: XSWDVecInst<v4i32, v2i64>;
-  def r64:   XSWDRegInst<R32C, R64C>;
-  
-  def r64_inreg: XSWDInst<(outs R64C:$rDst), (ins R64C:$rSrc),
-                          [(set R64C:$rDst, (sext_inreg R64C:$rSrc, i32))]>;
-}
-
-defm XSWD : ExtendWordToDoubleWord;
-
-// AND operations
-
-class ANDInst<dag OOL, dag IOL, list<dag> pattern> :
-    RRForm<0b10000011000, OOL, IOL, "and\t$rT, $rA, $rB",
-           IntegerOp, pattern>;
-
-class ANDVecInst<ValueType vectype>:
-    ANDInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-             [(set (vectype VECREG:$rT), (and (vectype VECREG:$rA),
-                                              (vectype VECREG:$rB)))]>;
-
-class ANDRegInst<RegisterClass rclass>:
-    ANDInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-             [(set rclass:$rT, (and rclass:$rA, rclass:$rB))]>;
-
-multiclass BitwiseAnd
-{
-  def v16i8: ANDVecInst<v16i8>;
-  def v8i16: ANDVecInst<v8i16>;
-  def v4i32: ANDVecInst<v4i32>;
-  def v2i64: ANDVecInst<v2i64>;
-
-  def r128:  ANDRegInst<GPRC>;
-  def r64:   ANDRegInst<R64C>;
-  def r32:   ANDRegInst<R32C>;
-  def r16:   ANDRegInst<R16C>;
-  def r8:    ANDRegInst<R8C>;
-
-  //===---------------------------------------------
-  // Special instructions to perform the fabs instruction
-  def fabs32: ANDInst<(outs R32FP:$rT), (ins R32FP:$rA, R32C:$rB),
-                      [/* Intentionally does not match a pattern */]>;
-
-  def fabs64: ANDInst<(outs R64FP:$rT), (ins R64FP:$rA, R64C:$rB),
-                      [/* Intentionally does not match a pattern */]>;
-
-  def fabsvec: ANDInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-                       [/* Intentionally does not match a pattern */]>;
-
-  //===---------------------------------------------
-
-  // Hacked form of AND to zero-extend 16-bit quantities to 32-bit
-  // quantities -- see 16->32 zext pattern.
-  //
-  // This pattern is somewhat artificial, since it might match some
-  // compiler generated pattern but it is unlikely to do so.
-
-  def i16i32: ANDInst<(outs R32C:$rT), (ins R16C:$rA, R32C:$rB),
-                      [(set R32C:$rT, (and (zext R16C:$rA), R32C:$rB))]>;
-}
-
-defm AND : BitwiseAnd;
-
-
-def vnot_cell_conv : PatFrag<(ops node:$in),
-                             (xor node:$in, (bitconvert (v4i32 immAllOnesV)))>;
-
-// N.B.: vnot_cell_conv is one of those special target selection pattern
-// fragments,
-// in which we expect there to be a bit_convert on the constant. Bear in mind
-// that llvm translates "not <reg>" to "xor <reg>, -1" (or in this case, a
-// constant -1 vector.)
-
-class ANDCInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10000011010, OOL, IOL, "andc\t$rT, $rA, $rB",
-           IntegerOp, pattern>;
-
-class ANDCVecInst<ValueType vectype, PatFrag vnot_frag = vnot>:
-    ANDCInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-             [(set (vectype VECREG:$rT),
-                   (and (vectype VECREG:$rA),
-                        (vnot_frag (vectype VECREG:$rB))))]>;
-
-class ANDCRegInst<RegisterClass rclass>:
-    ANDCInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-             [(set rclass:$rT, (and rclass:$rA, (not rclass:$rB)))]>;
-
-multiclass AndComplement
-{
-  def v16i8: ANDCVecInst<v16i8>;
-  def v8i16: ANDCVecInst<v8i16>;
-  def v4i32: ANDCVecInst<v4i32>;
-  def v2i64: ANDCVecInst<v2i64>;
-
-  def r128: ANDCRegInst<GPRC>;
-  def r64:  ANDCRegInst<R64C>;
-  def r32:  ANDCRegInst<R32C>;
-  def r16:  ANDCRegInst<R16C>;
-  def r8:   ANDCRegInst<R8C>;
-
-  // Sometimes, the xor pattern has a bitcast constant:
-  def v16i8_conv: ANDCVecInst<v16i8, vnot_cell_conv>;
-}
-
-defm ANDC : AndComplement;
-
-class ANDBIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI10Form<0b01101000, OOL, IOL, "andbi\t$rT, $rA, $val",
-             ByteOp, pattern>;
-
-multiclass AndByteImm
-{
-  def v16i8: ANDBIInst<(outs VECREG:$rT), (ins VECREG:$rA, u10imm:$val),
-                       [(set (v16i8 VECREG:$rT),
-                             (and (v16i8 VECREG:$rA),
-                                  (v16i8 v16i8U8Imm:$val)))]>;
-
-  def r8: ANDBIInst<(outs R8C:$rT), (ins R8C:$rA, u10imm_i8:$val),
-                    [(set R8C:$rT, (and R8C:$rA, immU8:$val))]>;
-}
-
-defm ANDBI : AndByteImm;
-
-class ANDHIInst<dag OOL, dag IOL, list<dag> pattern> :
-    RI10Form<0b10101000, OOL, IOL, "andhi\t$rT, $rA, $val",
-             ByteOp, pattern>;
-
-multiclass AndHalfwordImm
-{
-  def v8i16: ANDHIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-                       [(set (v8i16 VECREG:$rT),
-                             (and (v8i16 VECREG:$rA), v8i16SExt10Imm:$val))]>;
-
-  def r16: ANDHIInst<(outs R16C:$rT), (ins R16C:$rA, u10imm:$val),
-                     [(set R16C:$rT, (and R16C:$rA, i16ImmUns10:$val))]>;
-
-  // Zero-extend i8 to i16:
-  def i8i16: ANDHIInst<(outs R16C:$rT), (ins R8C:$rA, u10imm:$val),
-                      [(set R16C:$rT, (and (zext R8C:$rA), i16ImmUns10:$val))]>;
-}
-
-defm ANDHI : AndHalfwordImm;
-
-class ANDIInst<dag OOL, dag IOL, list<dag> pattern> :
-    RI10Form<0b00101000, OOL, IOL, "andi\t$rT, $rA, $val",
-             IntegerOp, pattern>;
-
-multiclass AndWordImm
-{
-  def v4i32: ANDIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-                      [(set (v4i32 VECREG:$rT),
-                            (and (v4i32 VECREG:$rA), v4i32SExt10Imm:$val))]>;
-
-  def r32: ANDIInst<(outs R32C:$rT), (ins R32C:$rA, s10imm_i32:$val),
-                    [(set R32C:$rT, (and R32C:$rA, i32ImmSExt10:$val))]>;
-
-  // Hacked form of ANDI to zero-extend i8 quantities to i32. See the zext 8->32
-  // pattern below.
-  def i8i32: ANDIInst<(outs R32C:$rT), (ins R8C:$rA, s10imm_i32:$val),
-                      [(set R32C:$rT,
-                            (and (zext R8C:$rA), i32ImmSExt10:$val))]>;
-
-  // Hacked form of ANDI to zero-extend i16 quantities to i32. See the
-  // zext 16->32 pattern below.
-  //
-  // Note that this pattern is somewhat artificial, since it might match
-  // something the compiler generates but is unlikely to occur in practice.
-  def i16i32: ANDIInst<(outs R32C:$rT), (ins R16C:$rA, s10imm_i32:$val),
-                       [(set R32C:$rT,
-                             (and (zext R16C:$rA), i32ImmSExt10:$val))]>;
-}
-
-defm ANDI : AndWordImm;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Bitwise OR group:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-// Bitwise "or" (N.B.: These are also register-register copy instructions...)
-class ORInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10000010000, OOL, IOL, "or\t$rT, $rA, $rB",
-           IntegerOp, pattern>;
-
-class ORVecInst<ValueType vectype>:
-    ORInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-           [(set (vectype VECREG:$rT), (or (vectype VECREG:$rA),
-                                           (vectype VECREG:$rB)))]>;
-
-class ORRegInst<RegisterClass rclass>:
-    ORInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-           [(set rclass:$rT, (or rclass:$rA, rclass:$rB))]>;
-
-
-multiclass BitwiseOr
-{
-  def v16i8: ORVecInst<v16i8>;
-  def v8i16: ORVecInst<v8i16>;
-  def v4i32: ORVecInst<v4i32>;
-  def v2i64: ORVecInst<v2i64>;
-
-  def v4f32: ORInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-                    [(set (v4f32 VECREG:$rT),
-                          (v4f32 (bitconvert (or (v4i32 VECREG:$rA),
-                                                 (v4i32 VECREG:$rB)))))]>;
-
-  def v2f64: ORInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-                    [(set (v2f64 VECREG:$rT),
-                          (v2f64 (bitconvert (or (v2i64 VECREG:$rA),
-                                                 (v2i64 VECREG:$rB)))))]>;
-
-  def r128: ORRegInst<GPRC>;
-  def r64:  ORRegInst<R64C>;
-  def r32:  ORRegInst<R32C>;
-  def r16:  ORRegInst<R16C>;
-  def r8:   ORRegInst<R8C>;
-
-  // OR instructions used to copy f32 and f64 registers.
-  def f32: ORInst<(outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB),
-                  [/* no pattern */]>;
-
-  def f64: ORInst<(outs R64FP:$rT), (ins R64FP:$rA, R64FP:$rB),
-                  [/* no pattern */]>;
-}
-
-defm OR : BitwiseOr;
-
-//===----------------------------------------------------------------------===//
-// SPU::PREFSLOT2VEC and VEC2PREFSLOT re-interpretations of registers
-//===----------------------------------------------------------------------===//
-def : Pat<(v16i8 (SPUprefslot2vec R8C:$rA)),
-          (COPY_TO_REGCLASS R8C:$rA, VECREG)>;
-
-def : Pat<(v8i16 (SPUprefslot2vec R16C:$rA)),
-          (COPY_TO_REGCLASS R16C:$rA, VECREG)>;
-
-def : Pat<(v4i32 (SPUprefslot2vec R32C:$rA)),
-          (COPY_TO_REGCLASS R32C:$rA, VECREG)>;
-
-def : Pat<(v2i64 (SPUprefslot2vec R64C:$rA)),
-          (COPY_TO_REGCLASS R64C:$rA, VECREG)>;
-
-def : Pat<(v4f32 (SPUprefslot2vec R32FP:$rA)),
-          (COPY_TO_REGCLASS R32FP:$rA, VECREG)>;
-
-def : Pat<(v2f64 (SPUprefslot2vec R64FP:$rA)),
-          (COPY_TO_REGCLASS R64FP:$rA, VECREG)>;
- 
-def : Pat<(i8 (SPUvec2prefslot (v16i8 VECREG:$rA))),
-          (COPY_TO_REGCLASS (v16i8 VECREG:$rA), R8C)>;
-
-def : Pat<(i16 (SPUvec2prefslot (v8i16 VECREG:$rA))),
-          (COPY_TO_REGCLASS (v8i16 VECREG:$rA), R16C)>;
-
-def : Pat<(i32 (SPUvec2prefslot (v4i32 VECREG:$rA))),
-          (COPY_TO_REGCLASS (v4i32 VECREG:$rA), R32C)>;
-
-def : Pat<(i64 (SPUvec2prefslot (v2i64 VECREG:$rA))),
-          (COPY_TO_REGCLASS (v2i64 VECREG:$rA), R64C)>;
-
-def : Pat<(f32 (SPUvec2prefslot (v4f32 VECREG:$rA))),
-          (COPY_TO_REGCLASS (v4f32 VECREG:$rA), R32FP)>;
-
-def : Pat<(f64 (SPUvec2prefslot (v2f64 VECREG:$rA))),
-          (COPY_TO_REGCLASS (v2f64 VECREG:$rA), R64FP)>;
-
-// Load Register: This is an assembler alias for a bitwise OR of a register
-// against itself. It's here because it brings some clarity to assembly
-// language output.
-
-let hasCtrlDep = 1 in {
-    class LRInst<dag OOL, dag IOL>
-              : SPUInstr<OOL, IOL, "lr\t$rT, $rA", IntegerOp> {
-      bits<7> RA;
-      bits<7> RT;
-
-      let Pattern = [/*no pattern*/];
-
-      let Inst{0-10} = 0b10000010000;   /* It's an OR operation */
-      let Inst{11-17} = RA;
-      let Inst{18-24} = RA;
-      let Inst{25-31} = RT;
-    }
-
-    class LRVecInst<ValueType vectype>:
-        LRInst<(outs VECREG:$rT), (ins VECREG:$rA)>;
-
-    class LRRegInst<RegisterClass rclass>:
-        LRInst<(outs rclass:$rT), (ins rclass:$rA)>;
-
-    multiclass LoadRegister {
-      def v2i64: LRVecInst<v2i64>;
-      def v2f64: LRVecInst<v2f64>;
-      def v4i32: LRVecInst<v4i32>;
-      def v4f32: LRVecInst<v4f32>;
-      def v8i16: LRVecInst<v8i16>;
-      def v16i8: LRVecInst<v16i8>;
-
-      def r128:  LRRegInst<GPRC>;
-      def r64:   LRRegInst<R64C>;
-      def f64:   LRRegInst<R64FP>;
-      def r32:   LRRegInst<R32C>;
-      def f32:   LRRegInst<R32FP>;
-      def r16:   LRRegInst<R16C>;
-      def r8:    LRRegInst<R8C>;
-    }
-
-    defm LR: LoadRegister;
-}
-
-// ORC: Bitwise "or" with complement (c = a | ~b)
-
-class ORCInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10010010000, OOL, IOL, "orc\t$rT, $rA, $rB",
-           IntegerOp, pattern>;
-
-class ORCVecInst<ValueType vectype>:
-    ORCInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-            [(set (vectype VECREG:$rT), (or (vectype VECREG:$rA),
-                                            (vnot (vectype VECREG:$rB))))]>;
-
-class ORCRegInst<RegisterClass rclass>:
-  ORCInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-          [(set rclass:$rT, (or rclass:$rA, (not rclass:$rB)))]>;
-
-multiclass BitwiseOrComplement
-{
-  def v16i8: ORCVecInst<v16i8>;
-  def v8i16: ORCVecInst<v8i16>;
-  def v4i32: ORCVecInst<v4i32>;
-  def v2i64: ORCVecInst<v2i64>;
-
-  def r128:  ORCRegInst<GPRC>;
-  def r64:   ORCRegInst<R64C>;
-  def r32:   ORCRegInst<R32C>;
-  def r16:   ORCRegInst<R16C>;
-  def r8:    ORCRegInst<R8C>;
-}
-
-defm ORC : BitwiseOrComplement;
-
-// OR byte immediate
-class ORBIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI10Form<0b01100000, OOL, IOL, "orbi\t$rT, $rA, $val",
-             IntegerOp, pattern>;
-
-class ORBIVecInst<ValueType vectype, PatLeaf immpred>:
-    ORBIInst<(outs VECREG:$rT), (ins VECREG:$rA, u10imm:$val),
-             [(set (v16i8 VECREG:$rT), (or (vectype VECREG:$rA),
-                                           (vectype immpred:$val)))]>;
-
-multiclass BitwiseOrByteImm
-{
-  def v16i8: ORBIVecInst<v16i8, v16i8U8Imm>;
-
-  def r8: ORBIInst<(outs R8C:$rT), (ins R8C:$rA, u10imm_i8:$val),
-                   [(set R8C:$rT, (or R8C:$rA, immU8:$val))]>;
-}
-
-defm ORBI : BitwiseOrByteImm;
-
-// OR halfword immediate
-class ORHIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI10Form<0b10100000, OOL, IOL, "orhi\t$rT, $rA, $val",
-             IntegerOp, pattern>;
-
-class ORHIVecInst<ValueType vectype, PatLeaf immpred>:
-    ORHIInst<(outs VECREG:$rT), (ins VECREG:$rA, u10imm:$val),
-              [(set (vectype VECREG:$rT), (or (vectype VECREG:$rA),
-                                              immpred:$val))]>;
-
-multiclass BitwiseOrHalfwordImm
-{
-  def v8i16: ORHIVecInst<v8i16, v8i16Uns10Imm>;
-
-  def r16: ORHIInst<(outs R16C:$rT), (ins R16C:$rA, u10imm:$val),
-                    [(set R16C:$rT, (or R16C:$rA, i16ImmUns10:$val))]>;
-
-  // Specialized ORHI form used to promote 8-bit registers to 16-bit
-  def i8i16: ORHIInst<(outs R16C:$rT), (ins R8C:$rA, s10imm:$val),
-                      [(set R16C:$rT, (or (anyext R8C:$rA),
-                                          i16ImmSExt10:$val))]>;
-}
-
-defm ORHI : BitwiseOrHalfwordImm;
-
-class ORIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI10Form<0b00100000, OOL, IOL, "ori\t$rT, $rA, $val",
-             IntegerOp, pattern>;
-
-class ORIVecInst<ValueType vectype, PatLeaf immpred>:
-    ORIInst<(outs VECREG:$rT), (ins VECREG:$rA, u10imm:$val),
-            [(set (vectype VECREG:$rT), (or (vectype VECREG:$rA),
-                                            immpred:$val))]>;
-
-// Bitwise "or" with immediate
-multiclass BitwiseOrImm
-{
-  def v4i32: ORIVecInst<v4i32, v4i32Uns10Imm>;
-
-  def r32: ORIInst<(outs R32C:$rT), (ins R32C:$rA, s10imm_i32:$val),
-                   [(set R32C:$rT, (or R32C:$rA, i32ImmSExt10:$val))]>;
-
-  // i16i32: hacked version of the ori instruction to extend 16-bit quantities
-  // to 32-bit quantities. used exclusively to match "anyext" conversions (vide
-  // infra "anyext 16->32" pattern.)
-  def i16i32: ORIInst<(outs R32C:$rT), (ins R16C:$rA, s10imm_i32:$val),
-                      [(set R32C:$rT, (or (anyext R16C:$rA),
-                                          i32ImmSExt10:$val))]>;
-
-  // i8i32: Hacked version of the ORI instruction to extend 16-bit quantities
-  // to 32-bit quantities. Used exclusively to match "anyext" conversions (vide
-  // infra "anyext 16->32" pattern.)
-  def i8i32: ORIInst<(outs R32C:$rT), (ins R8C:$rA, s10imm_i32:$val),
-                     [(set R32C:$rT, (or (anyext R8C:$rA),
-                                         i32ImmSExt10:$val))]>;
-}
-
-defm ORI : BitwiseOrImm;
-
-// ORX: "or" across the vector: or's $rA's word slots leaving the result in
-// $rT[0], slots 1-3 are zeroed.
-//
-// FIXME: Needs to match an intrinsic pattern.
-def ORXv4i32:
-    RRForm<0b10010010000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "orx\t$rT, $rA, $rB", IntegerOp,
-      []>;
-
-// XOR:
-
-class XORInst<dag OOL, dag IOL, list<dag> pattern> :
-    RRForm<0b10010010000, OOL, IOL, "xor\t$rT, $rA, $rB",
-           IntegerOp, pattern>;
-
-class XORVecInst<ValueType vectype>:
-    XORInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-             [(set (vectype VECREG:$rT), (xor (vectype VECREG:$rA),
-                                              (vectype VECREG:$rB)))]>;
-
-class XORRegInst<RegisterClass rclass>:
-    XORInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-             [(set rclass:$rT, (xor rclass:$rA, rclass:$rB))]>;
-
-multiclass BitwiseExclusiveOr
-{
-  def v16i8: XORVecInst<v16i8>;
-  def v8i16: XORVecInst<v8i16>;
-  def v4i32: XORVecInst<v4i32>;
-  def v2i64: XORVecInst<v2i64>;
-
-  def r128:  XORRegInst<GPRC>;
-  def r64:   XORRegInst<R64C>;
-  def r32:   XORRegInst<R32C>;
-  def r16:   XORRegInst<R16C>;
-  def r8:    XORRegInst<R8C>;
-
-  // XOR instructions used to negate f32 and f64 quantities.
-
-  def fneg32: XORInst<(outs R32FP:$rT), (ins R32FP:$rA, R32C:$rB),
-                     [/* no pattern */]>;
-
-  def fneg64: XORInst<(outs R64FP:$rT), (ins R64FP:$rA, R64C:$rB),
-                     [/* no pattern */]>;
-
-  def fnegvec: XORInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-                      [/* no pattern, see fneg{32,64} */]>;
-}
-
-defm XOR : BitwiseExclusiveOr;
-
-//==----------------------------------------------------------
-
-class XORBIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI10Form<0b01100000, OOL, IOL, "xorbi\t$rT, $rA, $val",
-             IntegerOp, pattern>;
-
-multiclass XorByteImm
-{
-  def v16i8:
-    XORBIInst<(outs VECREG:$rT), (ins VECREG:$rA, u10imm:$val),
-              [(set (v16i8 VECREG:$rT), (xor (v16i8 VECREG:$rA), v16i8U8Imm:$val))]>;
-
-  def r8:
-    XORBIInst<(outs R8C:$rT), (ins R8C:$rA, u10imm_i8:$val),
-              [(set R8C:$rT, (xor R8C:$rA, immU8:$val))]>;
-}
-
-defm XORBI : XorByteImm;
-
-def XORHIv8i16:
-    RI10Form<0b10100000, (outs VECREG:$rT), (ins VECREG:$rA, u10imm:$val),
-      "xorhi\t$rT, $rA, $val", IntegerOp,
-      [(set (v8i16 VECREG:$rT), (xor (v8i16 VECREG:$rA),
-                                      v8i16SExt10Imm:$val))]>;
-
-def XORHIr16:
-    RI10Form<0b10100000, (outs R16C:$rT), (ins R16C:$rA, s10imm:$val),
-      "xorhi\t$rT, $rA, $val", IntegerOp,
-      [(set R16C:$rT, (xor R16C:$rA, i16ImmSExt10:$val))]>;
-
-def XORIv4i32:
-    RI10Form<0b00100000, (outs VECREG:$rT), (ins VECREG:$rA, s10imm_i32:$val),
-      "xori\t$rT, $rA, $val", IntegerOp,
-      [(set (v4i32 VECREG:$rT), (xor (v4i32 VECREG:$rA),
-                                     v4i32SExt10Imm:$val))]>;
-
-def XORIr32:
-    RI10Form<0b00100000, (outs R32C:$rT), (ins R32C:$rA, s10imm_i32:$val),
-      "xori\t$rT, $rA, $val", IntegerOp,
-      [(set R32C:$rT, (xor R32C:$rA, i32ImmSExt10:$val))]>;
-
-// NAND:
-
-class NANDInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10010011000, OOL, IOL, "nand\t$rT, $rA, $rB",
-           IntegerOp, pattern>;
-
-class NANDVecInst<ValueType vectype>:
-    NANDInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-             [(set (vectype VECREG:$rT), (vnot (and (vectype VECREG:$rA),
-                                                    (vectype VECREG:$rB))))]>;
-class NANDRegInst<RegisterClass rclass>:
-    NANDInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-             [(set rclass:$rT, (not (and rclass:$rA, rclass:$rB)))]>;
-
-multiclass BitwiseNand
-{
-  def v16i8: NANDVecInst<v16i8>;
-  def v8i16: NANDVecInst<v8i16>;
-  def v4i32: NANDVecInst<v4i32>;
-  def v2i64: NANDVecInst<v2i64>;
-
-  def r128:  NANDRegInst<GPRC>;
-  def r64:   NANDRegInst<R64C>;
-  def r32:   NANDRegInst<R32C>;
-  def r16:   NANDRegInst<R16C>;
-  def r8:    NANDRegInst<R8C>;
-}
-
-defm NAND : BitwiseNand;
-
-// NOR:
-
-class NORInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10010010000, OOL, IOL, "nor\t$rT, $rA, $rB",
-           IntegerOp, pattern>;
-
-class NORVecInst<ValueType vectype>:
-    NORInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-            [(set (vectype VECREG:$rT), (vnot (or (vectype VECREG:$rA),
-                                                  (vectype VECREG:$rB))))]>;
-class NORRegInst<RegisterClass rclass>:
-    NORInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-            [(set rclass:$rT, (not (or rclass:$rA, rclass:$rB)))]>;
-
-multiclass BitwiseNor
-{
-  def v16i8: NORVecInst<v16i8>;
-  def v8i16: NORVecInst<v8i16>;
-  def v4i32: NORVecInst<v4i32>;
-  def v2i64: NORVecInst<v2i64>;
-
-  def r128:  NORRegInst<GPRC>;
-  def r64:   NORRegInst<R64C>;
-  def r32:   NORRegInst<R32C>;
-  def r16:   NORRegInst<R16C>;
-  def r8:    NORRegInst<R8C>;
-}
-
-defm NOR : BitwiseNor;
-
-// Select bits:
-class SELBInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRRForm<0b1000, OOL, IOL, "selb\t$rT, $rA, $rB, $rC",
-            IntegerOp, pattern>;
-
-class SELBVecInst<ValueType vectype, PatFrag vnot_frag = vnot>:
-  SELBInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-           [(set (vectype VECREG:$rT),
-                 (or (and (vectype VECREG:$rC), (vectype VECREG:$rB)),
-                     (and (vnot_frag (vectype VECREG:$rC)),
-                          (vectype VECREG:$rA))))]>;
-
-class SELBVecVCondInst<ValueType vectype>:
-  SELBInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-           [(set (vectype VECREG:$rT),
-                 (select (vectype VECREG:$rC),
-                         (vectype VECREG:$rB),
-                         (vectype VECREG:$rA)))]>;
-
-class SELBVecCondInst<ValueType vectype>:
-  SELBInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, R32C:$rC),
-           [(set (vectype VECREG:$rT),
-                 (select R32C:$rC,
-                         (vectype VECREG:$rB),
-                         (vectype VECREG:$rA)))]>;
-
-class SELBRegInst<RegisterClass rclass>:
-  SELBInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB, rclass:$rC),
-           [(set rclass:$rT,
-                 (or (and rclass:$rB, rclass:$rC),
-                     (and rclass:$rA, (not rclass:$rC))))]>;
-
-class SELBRegCondInst<RegisterClass rcond, RegisterClass rclass>:
-  SELBInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB, rcond:$rC),
-           [(set rclass:$rT,
-                 (select rcond:$rC, rclass:$rB, rclass:$rA))]>;
-
-multiclass SelectBits
-{
-  def v16i8: SELBVecInst<v16i8>;
-  def v8i16: SELBVecInst<v8i16>;
-  def v4i32: SELBVecInst<v4i32>;
-  def v2i64: SELBVecInst<v2i64, vnot_cell_conv>;
-
-  def r128:  SELBRegInst<GPRC>;
-  def r64:   SELBRegInst<R64C>;
-  def r32:   SELBRegInst<R32C>;
-  def r16:   SELBRegInst<R16C>;
-  def r8:    SELBRegInst<R8C>;
-
-  def v16i8_cond: SELBVecCondInst<v16i8>;
-  def v8i16_cond: SELBVecCondInst<v8i16>;
-  def v4i32_cond: SELBVecCondInst<v4i32>;
-  def v2i64_cond: SELBVecCondInst<v2i64>;
-
-  def v16i8_vcond: SELBVecCondInst<v16i8>;
-  def v8i16_vcond: SELBVecCondInst<v8i16>;
-  def v4i32_vcond: SELBVecCondInst<v4i32>;
-  def v2i64_vcond: SELBVecCondInst<v2i64>;
-
-  def v4f32_cond:
-        SELBInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-                 [(set (v4f32 VECREG:$rT),
-                       (select (v4i32 VECREG:$rC),
-                               (v4f32 VECREG:$rB),
-                               (v4f32 VECREG:$rA)))]>;
-
-  // SELBr64_cond is defined in SPU64InstrInfo.td
-  def r32_cond:   SELBRegCondInst<R32C, R32C>;
-  def f32_cond:   SELBRegCondInst<R32C, R32FP>;
-  def r16_cond:   SELBRegCondInst<R16C, R16C>;
-  def r8_cond:    SELBRegCondInst<R8C,  R8C>;
-}
-
-defm SELB : SelectBits;
-
-class SPUselbPatVec<ValueType vectype, SPUInstr inst>:
-   Pat<(SPUselb (vectype VECREG:$rA), (vectype VECREG:$rB), (vectype VECREG:$rC)),
-       (inst VECREG:$rA, VECREG:$rB, VECREG:$rC)>;
-
-def : SPUselbPatVec<v16i8, SELBv16i8>;
-def : SPUselbPatVec<v8i16, SELBv8i16>;
-def : SPUselbPatVec<v4i32, SELBv4i32>;
-def : SPUselbPatVec<v2i64, SELBv2i64>;
-
-class SPUselbPatReg<RegisterClass rclass, SPUInstr inst>:
-   Pat<(SPUselb rclass:$rA, rclass:$rB, rclass:$rC),
-       (inst rclass:$rA, rclass:$rB, rclass:$rC)>;
-
-def : SPUselbPatReg<R8C,   SELBr8>;
-def : SPUselbPatReg<R16C,  SELBr16>;
-def : SPUselbPatReg<R32C,  SELBr32>;
-def : SPUselbPatReg<R64C,  SELBr64>;
-
-// EQV: Equivalence (1 for each same bit, otherwise 0)
-//
-// Note: There are a lot of ways to match this bit operator and these patterns
-// attempt to be as exhaustive as possible.
-
-class EQVInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10010010000, OOL, IOL, "eqv\t$rT, $rA, $rB",
-           IntegerOp, pattern>;
-
-class EQVVecInst<ValueType vectype>:
-    EQVInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-            [(set (vectype VECREG:$rT),
-                  (or (and (vectype VECREG:$rA), (vectype VECREG:$rB)),
-                      (and (vnot (vectype VECREG:$rA)),
-                           (vnot (vectype VECREG:$rB)))))]>;
-
-class EQVRegInst<RegisterClass rclass>:
-    EQVInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-            [(set rclass:$rT, (or (and rclass:$rA, rclass:$rB),
-                                  (and (not rclass:$rA), (not rclass:$rB))))]>;
-
-class EQVVecPattern1<ValueType vectype>:
-  EQVInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-          [(set (vectype VECREG:$rT),
-                (xor (vectype VECREG:$rA), (vnot (vectype VECREG:$rB))))]>;
-
-class EQVRegPattern1<RegisterClass rclass>:
-  EQVInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-          [(set rclass:$rT, (xor rclass:$rA, (not rclass:$rB)))]>;
-
-class EQVVecPattern2<ValueType vectype>:
-  EQVInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-          [(set (vectype VECREG:$rT),
-                (or (and (vectype VECREG:$rA), (vectype VECREG:$rB)),
-                    (vnot (or (vectype VECREG:$rA), (vectype VECREG:$rB)))))]>;
-
-class EQVRegPattern2<RegisterClass rclass>:
-  EQVInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-          [(set rclass:$rT,
-                (or (and rclass:$rA, rclass:$rB),
-                    (not (or rclass:$rA, rclass:$rB))))]>;
-
-class EQVVecPattern3<ValueType vectype>:
-  EQVInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-          [(set (vectype VECREG:$rT),
-                (not (xor (vectype VECREG:$rA), (vectype VECREG:$rB))))]>;
-
-class EQVRegPattern3<RegisterClass rclass>:
-  EQVInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-          [(set rclass:$rT, (not (xor rclass:$rA, rclass:$rB)))]>;
-
-multiclass BitEquivalence
-{
-  def v16i8: EQVVecInst<v16i8>;
-  def v8i16: EQVVecInst<v8i16>;
-  def v4i32: EQVVecInst<v4i32>;
-  def v2i64: EQVVecInst<v2i64>;
-
-  def v16i8_1: EQVVecPattern1<v16i8>;
-  def v8i16_1: EQVVecPattern1<v8i16>;
-  def v4i32_1: EQVVecPattern1<v4i32>;
-  def v2i64_1: EQVVecPattern1<v2i64>;
-
-  def v16i8_2: EQVVecPattern2<v16i8>;
-  def v8i16_2: EQVVecPattern2<v8i16>;
-  def v4i32_2: EQVVecPattern2<v4i32>;
-  def v2i64_2: EQVVecPattern2<v2i64>;
-
-  def v16i8_3: EQVVecPattern3<v16i8>;
-  def v8i16_3: EQVVecPattern3<v8i16>;
-  def v4i32_3: EQVVecPattern3<v4i32>;
-  def v2i64_3: EQVVecPattern3<v2i64>;
-
-  def r128:  EQVRegInst<GPRC>;
-  def r64:   EQVRegInst<R64C>;
-  def r32:   EQVRegInst<R32C>;
-  def r16:   EQVRegInst<R16C>;
-  def r8:    EQVRegInst<R8C>;
-
-  def r128_1: EQVRegPattern1<GPRC>;
-  def r64_1:  EQVRegPattern1<R64C>;
-  def r32_1:  EQVRegPattern1<R32C>;
-  def r16_1:  EQVRegPattern1<R16C>;
-  def r8_1:   EQVRegPattern1<R8C>;
-
-  def r128_2: EQVRegPattern2<GPRC>;
-  def r64_2:  EQVRegPattern2<R64C>;
-  def r32_2:  EQVRegPattern2<R32C>;
-  def r16_2:  EQVRegPattern2<R16C>;
-  def r8_2:   EQVRegPattern2<R8C>;
-
-  def r128_3: EQVRegPattern3<GPRC>;
-  def r64_3:  EQVRegPattern3<R64C>;
-  def r32_3:  EQVRegPattern3<R32C>;
-  def r16_3:  EQVRegPattern3<R16C>;
-  def r8_3:   EQVRegPattern3<R8C>;
-}
-
-defm EQV: BitEquivalence;
-
-//===----------------------------------------------------------------------===//
-// Vector shuffle...
-//===----------------------------------------------------------------------===//
-// SPUshuffle is generated in LowerVECTOR_SHUFFLE and gets replaced with SHUFB.
-// See the SPUshuffle SDNode operand above, which sets up the DAG pattern
-// matcher to emit something when the LowerVECTOR_SHUFFLE generates a node with
-// the SPUISD::SHUFB opcode.
-//===----------------------------------------------------------------------===//
-
-class SHUFBInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRRForm<0b1000, OOL, IOL, "shufb\t$rT, $rA, $rB, $rC",
-            ShuffleOp, pattern>;
-
-class SHUFBVecInst<ValueType resultvec, ValueType maskvec>:
-    SHUFBInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-              [(set (resultvec VECREG:$rT),
-                    (SPUshuffle (resultvec VECREG:$rA),
-                                (resultvec VECREG:$rB),
-                                (maskvec VECREG:$rC)))]>;
-
-class SHUFBGPRCInst:
-    SHUFBInst<(outs VECREG:$rT), (ins GPRC:$rA, GPRC:$rB, VECREG:$rC),
-              [/* no pattern */]>;
-
-multiclass ShuffleBytes
-{
-  def v16i8     : SHUFBVecInst<v16i8, v16i8>;
-  def v16i8_m32 : SHUFBVecInst<v16i8, v4i32>;
-  def v8i16     : SHUFBVecInst<v8i16, v16i8>;
-  def v8i16_m32 : SHUFBVecInst<v8i16, v4i32>;
-  def v4i32     : SHUFBVecInst<v4i32, v16i8>;
-  def v4i32_m32 : SHUFBVecInst<v4i32, v4i32>;
-  def v2i64     : SHUFBVecInst<v2i64, v16i8>;
-  def v2i64_m32 : SHUFBVecInst<v2i64, v4i32>;
-
-  def v4f32     : SHUFBVecInst<v4f32, v16i8>;
-  def v4f32_m32 : SHUFBVecInst<v4f32, v4i32>;
-
-  def v2f64     : SHUFBVecInst<v2f64, v16i8>;
-  def v2f64_m32 : SHUFBVecInst<v2f64, v4i32>;
-
-  def gprc      : SHUFBGPRCInst;
-}
-
-defm SHUFB : ShuffleBytes;
-
-//===----------------------------------------------------------------------===//
-// Shift and rotate group:
-//===----------------------------------------------------------------------===//
-
-class SHLHInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b11111010000, OOL, IOL, "shlh\t$rT, $rA, $rB",
-           RotShiftVec, pattern>;
-
-class SHLHVecInst<ValueType vectype>:
-    SHLHInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-             [(set (vectype VECREG:$rT),
-                   (SPUvec_shl (vectype VECREG:$rA), (vectype VECREG:$rB)))]>;
-
-multiclass ShiftLeftHalfword
-{
-  def v8i16: SHLHVecInst<v8i16>;
-  def r16:   SHLHInst<(outs R16C:$rT), (ins R16C:$rA, R16C:$rB),
-                      [(set R16C:$rT, (shl R16C:$rA, R16C:$rB))]>;
-  def r16_r32: SHLHInst<(outs R16C:$rT), (ins R16C:$rA, R32C:$rB),
-                        [(set R16C:$rT, (shl R16C:$rA, R32C:$rB))]>;
-}
-
-defm SHLH : ShiftLeftHalfword;
-
-//===----------------------------------------------------------------------===//
-
-class SHLHIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b11111010000, OOL, IOL, "shlhi\t$rT, $rA, $val",
-            RotShiftVec, pattern>;
-
-class SHLHIVecInst<ValueType vectype>:
-    SHLHIInst<(outs VECREG:$rT), (ins VECREG:$rA, u7imm:$val),
-              [(set (vectype VECREG:$rT),
-                    (SPUvec_shl (vectype VECREG:$rA), (i16 uimm7:$val)))]>;
-
-multiclass ShiftLeftHalfwordImm
-{
-  def v8i16: SHLHIVecInst<v8i16>;
-  def r16: SHLHIInst<(outs R16C:$rT), (ins R16C:$rA, u7imm:$val),
-                     [(set R16C:$rT, (shl R16C:$rA, (i16 uimm7:$val)))]>;
-}
-
-defm SHLHI : ShiftLeftHalfwordImm;
-
-def : Pat<(SPUvec_shl (v8i16 VECREG:$rA), (i32 uimm7:$val)),
-          (SHLHIv8i16 VECREG:$rA, (TO_IMM16 uimm7:$val))>;
-
-def : Pat<(shl R16C:$rA, (i32 uimm7:$val)),
-          (SHLHIr16 R16C:$rA, (TO_IMM16 uimm7:$val))>;
-
-//===----------------------------------------------------------------------===//
-
-class SHLInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b11111010000, OOL, IOL, "shl\t$rT, $rA, $rB",
-           RotShiftVec, pattern>;
-
-multiclass ShiftLeftWord
-{
-  def v4i32:
-      SHLInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-              [(set (v4i32 VECREG:$rT),
-                    (SPUvec_shl (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-  def r32:
-      SHLInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB),
-              [(set R32C:$rT, (shl R32C:$rA, R32C:$rB))]>;
-}
-
-defm SHL: ShiftLeftWord;
-
-//===----------------------------------------------------------------------===//
-
-class SHLIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b11111010000, OOL, IOL, "shli\t$rT, $rA, $val",
-            RotShiftVec, pattern>;
-
-multiclass ShiftLeftWordImm
-{
-  def v4i32:
-    SHLIInst<(outs VECREG:$rT), (ins VECREG:$rA, u7imm_i32:$val),
-             [(set (v4i32 VECREG:$rT),
-                   (SPUvec_shl (v4i32 VECREG:$rA), (i32 uimm7:$val)))]>;
-
-  def r32:
-    SHLIInst<(outs R32C:$rT), (ins R32C:$rA, u7imm_i32:$val),
-             [(set R32C:$rT, (shl R32C:$rA, (i32 uimm7:$val)))]>;
-}
-
-defm SHLI : ShiftLeftWordImm;
-
-//===----------------------------------------------------------------------===//
-// SHLQBI vec form: Note that this will shift the entire vector (the 128-bit
-// register) to the left. Vector form is here to ensure type correctness.
-//
-// The shift count is in the lowest 3 bits (29-31) of $rB, so only a bit shift
-// of 7 bits is actually possible.
-//
-// Note also that SHLQBI/SHLQBII are used in conjunction with SHLQBY/SHLQBYI
-// to shift i64 and i128. SHLQBI is the residual left over after shifting by
-// bytes with SHLQBY.
-
-class SHLQBIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b11011011100, OOL, IOL, "shlqbi\t$rT, $rA, $rB",
-           RotShiftQuad, pattern>;
-
-class SHLQBIVecInst<ValueType vectype>:
-    SHLQBIInst<(outs VECREG:$rT), (ins VECREG:$rA, R32C:$rB),
-               [(set (vectype VECREG:$rT),
-                     (SPUshlquad_l_bits (vectype VECREG:$rA), R32C:$rB))]>;
-
-class SHLQBIRegInst<RegisterClass rclass>:
-    SHLQBIInst<(outs rclass:$rT), (ins rclass:$rA, R32C:$rB),
-               [/* no pattern */]>;
-
-multiclass ShiftLeftQuadByBits
-{
-  def v16i8: SHLQBIVecInst<v16i8>;
-  def v8i16: SHLQBIVecInst<v8i16>;
-  def v4i32: SHLQBIVecInst<v4i32>;
-  def v4f32: SHLQBIVecInst<v4f32>;
-  def v2i64: SHLQBIVecInst<v2i64>;
-  def v2f64: SHLQBIVecInst<v2f64>;
-
-  def r128:  SHLQBIRegInst<GPRC>;
-}
-
-defm SHLQBI : ShiftLeftQuadByBits;
-
-// See note above on SHLQBI. In this case, the predicate actually does then
-// enforcement, whereas with SHLQBI, we have to "take it on faith."
-class SHLQBIIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b11011111100, OOL, IOL, "shlqbii\t$rT, $rA, $val",
-            RotShiftQuad, pattern>;
-
-class SHLQBIIVecInst<ValueType vectype>:
-    SHLQBIIInst<(outs VECREG:$rT), (ins VECREG:$rA, u7imm_i32:$val),
-                [(set (vectype VECREG:$rT),
-                      (SPUshlquad_l_bits (vectype VECREG:$rA), (i32 bitshift:$val)))]>;
-
-multiclass ShiftLeftQuadByBitsImm
-{
-  def v16i8 : SHLQBIIVecInst<v16i8>;
-  def v8i16 : SHLQBIIVecInst<v8i16>;
-  def v4i32 : SHLQBIIVecInst<v4i32>;
-  def v4f32 : SHLQBIIVecInst<v4f32>;
-  def v2i64 : SHLQBIIVecInst<v2i64>;
-  def v2f64 : SHLQBIIVecInst<v2f64>;
-}
-
-defm SHLQBII : ShiftLeftQuadByBitsImm;
-
-// SHLQBY, SHLQBYI vector forms: Shift the entire vector to the left by bytes,
-// not by bits. See notes above on SHLQBI.
-
-class SHLQBYInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b11111011100, OOL, IOL, "shlqby\t$rT, $rA, $rB",
-            RotShiftQuad, pattern>;
-
-class SHLQBYVecInst<ValueType vectype>:
-    SHLQBYInst<(outs VECREG:$rT), (ins VECREG:$rA, R32C:$rB),
-               [(set (vectype VECREG:$rT),
-                     (SPUshlquad_l_bytes (vectype VECREG:$rA), R32C:$rB))]>;
-
-multiclass ShiftLeftQuadBytes
-{
-  def v16i8: SHLQBYVecInst<v16i8>;
-  def v8i16: SHLQBYVecInst<v8i16>;
-  def v4i32: SHLQBYVecInst<v4i32>;
-  def v4f32: SHLQBYVecInst<v4f32>;
-  def v2i64: SHLQBYVecInst<v2i64>;
-  def v2f64: SHLQBYVecInst<v2f64>;
-  def r128: SHLQBYInst<(outs GPRC:$rT), (ins GPRC:$rA, R32C:$rB),
-                       [(set GPRC:$rT, (SPUshlquad_l_bytes GPRC:$rA, R32C:$rB))]>;
-}
-
-defm SHLQBY: ShiftLeftQuadBytes;
-
-class SHLQBYIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b11111111100, OOL, IOL, "shlqbyi\t$rT, $rA, $val",
-            RotShiftQuad, pattern>;
-
-class SHLQBYIVecInst<ValueType vectype>:
-    SHLQBYIInst<(outs VECREG:$rT), (ins VECREG:$rA, u7imm_i32:$val),
-                [(set (vectype VECREG:$rT),
-                      (SPUshlquad_l_bytes (vectype VECREG:$rA), (i32 uimm7:$val)))]>;
-
-multiclass ShiftLeftQuadBytesImm
-{
-  def v16i8: SHLQBYIVecInst<v16i8>;
-  def v8i16: SHLQBYIVecInst<v8i16>;
-  def v4i32: SHLQBYIVecInst<v4i32>;
-  def v4f32: SHLQBYIVecInst<v4f32>;
-  def v2i64: SHLQBYIVecInst<v2i64>;
-  def v2f64: SHLQBYIVecInst<v2f64>;
-  def r128:  SHLQBYIInst<(outs GPRC:$rT), (ins GPRC:$rA, u7imm_i32:$val),
-                         [(set GPRC:$rT,
-                               (SPUshlquad_l_bytes GPRC:$rA, (i32 uimm7:$val)))]>;
-}
-
-defm SHLQBYI : ShiftLeftQuadBytesImm;
-
-class SHLQBYBIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b00111001111, OOL, IOL, "shlqbybi\t$rT, $rA, $rB",
-           RotShiftQuad, pattern>;
-
-class SHLQBYBIVecInst<ValueType vectype>:
-    SHLQBYBIInst<(outs VECREG:$rT), (ins VECREG:$rA, R32C:$rB),
-                [/* no pattern */]>;
-
-class SHLQBYBIRegInst<RegisterClass rclass>:
-    SHLQBYBIInst<(outs rclass:$rT), (ins rclass:$rA, R32C:$rB),
-                 [/* no pattern */]>;
-
-multiclass ShiftLeftQuadBytesBitCount
-{
-  def v16i8: SHLQBYBIVecInst<v16i8>;
-  def v8i16: SHLQBYBIVecInst<v8i16>;
-  def v4i32: SHLQBYBIVecInst<v4i32>;
-  def v4f32: SHLQBYBIVecInst<v4f32>;
-  def v2i64: SHLQBYBIVecInst<v2i64>;
-  def v2f64: SHLQBYBIVecInst<v2f64>;
-
-  def r128:  SHLQBYBIRegInst<GPRC>;
-}
-
-defm SHLQBYBI : ShiftLeftQuadBytesBitCount;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Rotate halfword:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-class ROTHInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b00111010000, OOL, IOL, "roth\t$rT, $rA, $rB",
-           RotShiftVec, pattern>;
-
-class ROTHVecInst<ValueType vectype>:
-    ROTHInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-             [(set (vectype VECREG:$rT),
-                   (SPUvec_rotl VECREG:$rA, (v8i16 VECREG:$rB)))]>;
-
-class ROTHRegInst<RegisterClass rclass>:
-    ROTHInst<(outs rclass:$rT), (ins rclass:$rA, rclass:$rB),
-             [(set rclass:$rT, (rotl rclass:$rA, rclass:$rB))]>;
-
-multiclass RotateLeftHalfword
-{
-  def v8i16: ROTHVecInst<v8i16>;
-  def r16: ROTHRegInst<R16C>;
-}
-
-defm ROTH: RotateLeftHalfword;
-
-def ROTHr16_r32: ROTHInst<(outs R16C:$rT), (ins R16C:$rA, R32C:$rB),
-                          [(set R16C:$rT, (rotl R16C:$rA, R32C:$rB))]>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Rotate halfword, immediate:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-class ROTHIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b00111110000, OOL, IOL, "rothi\t$rT, $rA, $val",
-            RotShiftVec, pattern>;
-
-class ROTHIVecInst<ValueType vectype>:
-    ROTHIInst<(outs VECREG:$rT), (ins VECREG:$rA, u7imm:$val),
-              [(set (vectype VECREG:$rT),
-                    (SPUvec_rotl VECREG:$rA, (i16 uimm7:$val)))]>;
-
-multiclass RotateLeftHalfwordImm
-{
-  def v8i16: ROTHIVecInst<v8i16>;
-  def r16: ROTHIInst<(outs R16C:$rT), (ins R16C:$rA, u7imm:$val),
-                     [(set R16C:$rT, (rotl R16C:$rA, (i16 uimm7:$val)))]>;
-  def r16_r32: ROTHIInst<(outs R16C:$rT), (ins R16C:$rA, u7imm_i32:$val),
-                         [(set R16C:$rT, (rotl R16C:$rA, (i32 uimm7:$val)))]>;
-}
-
-defm ROTHI: RotateLeftHalfwordImm;
-
-def : Pat<(SPUvec_rotl (v8i16 VECREG:$rA), (i32 uimm7:$val)),
-          (ROTHIv8i16 VECREG:$rA, (TO_IMM16 imm:$val))>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Rotate word:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class ROTInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b00011010000, OOL, IOL, "rot\t$rT, $rA, $rB",
-           RotShiftVec, pattern>;
-
-class ROTVecInst<ValueType vectype>:
-    ROTInst<(outs VECREG:$rT), (ins VECREG:$rA, R32C:$rB),
-            [(set (vectype VECREG:$rT),
-                  (SPUvec_rotl (vectype VECREG:$rA), R32C:$rB))]>;
-
-class ROTRegInst<RegisterClass rclass>:
-    ROTInst<(outs rclass:$rT), (ins rclass:$rA, R32C:$rB),
-            [(set rclass:$rT,
-                  (rotl rclass:$rA, R32C:$rB))]>;
-
-multiclass RotateLeftWord
-{
-  def v4i32: ROTVecInst<v4i32>;
-  def r32:   ROTRegInst<R32C>;
-}
-
-defm ROT: RotateLeftWord;
-
-// The rotate amount is in the same bits whether we've got an 8-bit, 16-bit or
-// 32-bit register
-def ROTr32_r16_anyext:
-    ROTInst<(outs R32C:$rT), (ins R32C:$rA, R16C:$rB),
-            [(set R32C:$rT, (rotl R32C:$rA, (i32 (anyext R16C:$rB))))]>;
-
-def : Pat<(rotl R32C:$rA, (i32 (zext R16C:$rB))),
-          (ROTr32_r16_anyext R32C:$rA, R16C:$rB)>;
-
-def : Pat<(rotl R32C:$rA, (i32 (sext R16C:$rB))),
-          (ROTr32_r16_anyext R32C:$rA, R16C:$rB)>;
-
-def ROTr32_r8_anyext:
-    ROTInst<(outs R32C:$rT), (ins R32C:$rA, R8C:$rB),
-            [(set R32C:$rT, (rotl R32C:$rA, (i32 (anyext R8C:$rB))))]>;
-
-def : Pat<(rotl R32C:$rA, (i32 (zext R8C:$rB))),
-          (ROTr32_r8_anyext R32C:$rA, R8C:$rB)>;
-
-def : Pat<(rotl R32C:$rA, (i32 (sext R8C:$rB))),
-          (ROTr32_r8_anyext R32C:$rA, R8C:$rB)>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Rotate word, immediate
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class ROTIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b00011110000, OOL, IOL, "roti\t$rT, $rA, $val",
-            RotShiftVec, pattern>;
-
-class ROTIVecInst<ValueType vectype, Operand optype, ValueType inttype, PatLeaf pred>:
-    ROTIInst<(outs VECREG:$rT), (ins VECREG:$rA, optype:$val),
-             [(set (vectype VECREG:$rT),
-                   (SPUvec_rotl (vectype VECREG:$rA), (inttype pred:$val)))]>;
-
-class ROTIRegInst<RegisterClass rclass, Operand optype, ValueType inttype, PatLeaf pred>:
-    ROTIInst<(outs rclass:$rT), (ins rclass:$rA, optype:$val),
-             [(set rclass:$rT, (rotl rclass:$rA, (inttype pred:$val)))]>;
-
-multiclass RotateLeftWordImm
-{
-  def v4i32: ROTIVecInst<v4i32, u7imm_i32, i32, uimm7>;
-  def v4i32_i16: ROTIVecInst<v4i32, u7imm, i16, uimm7>;
-  def v4i32_i8:  ROTIVecInst<v4i32, u7imm_i8, i8, uimm7>;
-
-  def r32:       ROTIRegInst<R32C, u7imm_i32, i32, uimm7>;
-  def r32_i16:   ROTIRegInst<R32C, u7imm, i16, uimm7>;
-  def r32_i8:    ROTIRegInst<R32C, u7imm_i8, i8, uimm7>;
-}
-
-defm ROTI : RotateLeftWordImm;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Rotate quad by byte (count)
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class ROTQBYInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b00111011100, OOL, IOL, "rotqby\t$rT, $rA, $rB",
-           RotShiftQuad, pattern>;
-
-class ROTQBYGenInst<ValueType type, RegisterClass rc>:
-    ROTQBYInst<(outs rc:$rT), (ins rc:$rA, R32C:$rB),
-               [(set (type rc:$rT),
-                     (SPUrotbytes_left (type rc:$rA), R32C:$rB))]>;
-
-class ROTQBYVecInst<ValueType type>:
-    ROTQBYGenInst<type, VECREG>;
-
-multiclass RotateQuadLeftByBytes
-{
-  def v16i8: ROTQBYVecInst<v16i8>;
-  def v8i16: ROTQBYVecInst<v8i16>;
-  def v4i32: ROTQBYVecInst<v4i32>;
-  def v4f32: ROTQBYVecInst<v4f32>;
-  def v2i64: ROTQBYVecInst<v2i64>;
-  def v2f64: ROTQBYVecInst<v2f64>;
-  def i128:  ROTQBYGenInst<i128, GPRC>;
-}
-
-defm ROTQBY: RotateQuadLeftByBytes;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Rotate quad by byte (count), immediate
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class ROTQBYIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b00111111100, OOL, IOL, "rotqbyi\t$rT, $rA, $val",
-            RotShiftQuad, pattern>;
-
-class ROTQBYIGenInst<ValueType type, RegisterClass rclass>:
-    ROTQBYIInst<(outs rclass:$rT), (ins rclass:$rA, u7imm:$val),
-                [(set (type rclass:$rT),
-                      (SPUrotbytes_left (type rclass:$rA), (i16 uimm7:$val)))]>;
-
-class ROTQBYIVecInst<ValueType vectype>:
-    ROTQBYIGenInst<vectype, VECREG>;
-
-multiclass RotateQuadByBytesImm
-{
-  def v16i8: ROTQBYIVecInst<v16i8>;
-  def v8i16: ROTQBYIVecInst<v8i16>;
-  def v4i32: ROTQBYIVecInst<v4i32>;
-  def v4f32: ROTQBYIVecInst<v4f32>;
-  def v2i64: ROTQBYIVecInst<v2i64>;
-  def vfi64: ROTQBYIVecInst<v2f64>;
-  def i128:  ROTQBYIGenInst<i128, GPRC>;
-}
-
-defm ROTQBYI: RotateQuadByBytesImm;
-
-// See ROTQBY note above.
-class ROTQBYBIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b00110011100, OOL, IOL,
-      "rotqbybi\t$rT, $rA, $shift",
-      RotShiftQuad, pattern>;
-
-class ROTQBYBIVecInst<ValueType vectype, RegisterClass rclass>:
-    ROTQBYBIInst<(outs VECREG:$rT), (ins VECREG:$rA, rclass:$shift),
-      [(set (vectype VECREG:$rT),
-            (SPUrotbytes_left_bits (vectype VECREG:$rA), rclass:$shift))]>;
-
-multiclass RotateQuadByBytesByBitshift {
-  def v16i8_r32: ROTQBYBIVecInst<v16i8, R32C>;
-  def v8i16_r32: ROTQBYBIVecInst<v8i16, R32C>;
-  def v4i32_r32: ROTQBYBIVecInst<v4i32, R32C>;
-  def v2i64_r32: ROTQBYBIVecInst<v2i64, R32C>;
-}
-
-defm ROTQBYBI : RotateQuadByBytesByBitshift;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// See ROTQBY note above.
-//
-// Assume that the user of this instruction knows to shift the rotate count
-// into bit 29
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class ROTQBIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b00011011100, OOL, IOL, "rotqbi\t$rT, $rA, $rB",
-           RotShiftQuad, pattern>;
-
-class ROTQBIVecInst<ValueType vectype>:
-    ROTQBIInst<(outs VECREG:$rT), (ins VECREG:$rA, R32C:$rB),
-               [/* no pattern yet */]>;
-
-class ROTQBIRegInst<RegisterClass rclass>:
-    ROTQBIInst<(outs rclass:$rT), (ins rclass:$rA, R32C:$rB),
-               [/* no pattern yet */]>;
-
-multiclass RotateQuadByBitCount
-{
-  def v16i8: ROTQBIVecInst<v16i8>;
-  def v8i16: ROTQBIVecInst<v8i16>;
-  def v4i32: ROTQBIVecInst<v4i32>;
-  def v2i64: ROTQBIVecInst<v2i64>;
-
-  def r128:  ROTQBIRegInst<GPRC>;
-  def r64:   ROTQBIRegInst<R64C>;
-}
-
-defm ROTQBI: RotateQuadByBitCount;
-
-class ROTQBIIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b00011111100, OOL, IOL, "rotqbii\t$rT, $rA, $val",
-            RotShiftQuad, pattern>;
-
-class ROTQBIIVecInst<ValueType vectype, Operand optype, ValueType inttype,
-                     PatLeaf pred>:
-    ROTQBIIInst<(outs VECREG:$rT), (ins VECREG:$rA, optype:$val),
-                [/* no pattern yet */]>;
-
-class ROTQBIIRegInst<RegisterClass rclass, Operand optype, ValueType inttype,
-                     PatLeaf pred>:
-    ROTQBIIInst<(outs rclass:$rT), (ins rclass:$rA, optype:$val),
-                [/* no pattern yet */]>;
-
-multiclass RotateQuadByBitCountImm
-{
-  def v16i8: ROTQBIIVecInst<v16i8, u7imm_i32, i32, uimm7>;
-  def v8i16: ROTQBIIVecInst<v8i16, u7imm_i32, i32, uimm7>;
-  def v4i32: ROTQBIIVecInst<v4i32, u7imm_i32, i32, uimm7>;
-  def v2i64: ROTQBIIVecInst<v2i64, u7imm_i32, i32, uimm7>;
-
-  def r128:  ROTQBIIRegInst<GPRC, u7imm_i32, i32, uimm7>;
-  def r64:   ROTQBIIRegInst<R64C, u7imm_i32, i32, uimm7>;
-}
-
-defm ROTQBII : RotateQuadByBitCountImm;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// ROTHM v8i16 form:
-// NOTE(1): No vector rotate is generated by the C/C++ frontend (today),
-//          so this only matches a synthetically generated/lowered code
-//          fragment.
-// NOTE(2): $rB must be negated before the right rotate!
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class ROTHMInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10111010000, OOL, IOL, "rothm\t$rT, $rA, $rB",
-           RotShiftVec, pattern>;
-
-def ROTHMv8i16:
-    ROTHMInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-              [/* see patterns below - $rB must be negated */]>;
-
-def : Pat<(SPUvec_srl (v8i16 VECREG:$rA), (v8i16 VECREG:$rB)),
-          (ROTHMv8i16 VECREG:$rA, (SFHIvec VECREG:$rB, 0))>;
-
-// ROTHM r16 form: Rotate 16-bit quantity to right, zero fill at the left
-// Note: This instruction doesn't match a pattern because rB must be negated
-// for the instruction to work. Thus, the pattern below the instruction!
-
-def ROTHMr16:
-    ROTHMInst<(outs R16C:$rT), (ins R16C:$rA, R32C:$rB),
-              [/* see patterns below - $rB must be negated! */]>;
-
-def : Pat<(srl R16C:$rA, R32C:$rB),
-          (ROTHMr16 R16C:$rA, (SFIr32 R32C:$rB, 0))>;
-
-def : Pat<(srl R16C:$rA, R16C:$rB),
-          (ROTHMr16 R16C:$rA,
-                    (SFIr32 (XSHWr16 R16C:$rB), 0))>;
-
-def : Pat<(srl R16C:$rA, R8C:$rB),
-          (ROTHMr16 R16C:$rA,
-                    (SFIr32 (XSHWr16 (XSBHr8 R8C:$rB) ), 0))>;
-
-// ROTHMI v8i16 form: See the comment for ROTHM v8i16. The difference here is
-// that the immediate can be complemented, so that the user doesn't have to
-// worry about it.
-
-class ROTHMIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b10111110000, OOL, IOL, "rothmi\t$rT, $rA, $val",
-            RotShiftVec, pattern>;
-
-def ROTHMIv8i16:
-    ROTHMIInst<(outs VECREG:$rT), (ins VECREG:$rA, rothNeg7imm:$val),
-               [/* no pattern */]>;
-
-def : Pat<(SPUvec_srl (v8i16 VECREG:$rA), (i32 imm:$val)),
-          (ROTHMIv8i16 VECREG:$rA, imm:$val)>;
-
-def: Pat<(SPUvec_srl (v8i16 VECREG:$rA), (i16 imm:$val)),
-         (ROTHMIv8i16 VECREG:$rA, (TO_IMM32 imm:$val))>;
-
-def: Pat<(SPUvec_srl (v8i16 VECREG:$rA), (i8 imm:$val)),
-         (ROTHMIv8i16 VECREG:$rA, (TO_IMM32 imm:$val))>;
-
-def ROTHMIr16:
-    ROTHMIInst<(outs R16C:$rT), (ins R16C:$rA, rothNeg7imm:$val),
-               [/* no pattern */]>;
-
-def: Pat<(srl R16C:$rA, (i32 uimm7:$val)),
-         (ROTHMIr16 R16C:$rA, uimm7:$val)>;
-
-def: Pat<(srl R16C:$rA, (i16 uimm7:$val)),
-         (ROTHMIr16 R16C:$rA, (TO_IMM32 uimm7:$val))>;
-
-def: Pat<(srl R16C:$rA, (i8 uimm7:$val)),
-         (ROTHMIr16 R16C:$rA, (TO_IMM32 uimm7:$val))>;
-
-// ROTM v4i32 form: See the ROTHM v8i16 comments.
-class ROTMInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10011010000, OOL, IOL, "rotm\t$rT, $rA, $rB",
-           RotShiftVec, pattern>;
-
-def ROTMv4i32:
-    ROTMInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-             [/* see patterns below - $rB must be negated */]>;
-
-def : Pat<(SPUvec_srl (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)),
-          (ROTMv4i32 VECREG:$rA, (SFIvec VECREG:$rB, 0))>;
-
-def ROTMr32:
-    ROTMInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB),
-             [/* see patterns below - $rB must be negated */]>;
-
-def : Pat<(srl R32C:$rA, R32C:$rB),
-          (ROTMr32 R32C:$rA, (SFIr32 R32C:$rB, 0))>;
-
-def : Pat<(srl R32C:$rA, R16C:$rB),
-          (ROTMr32 R32C:$rA,
-                   (SFIr32 (XSHWr16 R16C:$rB), 0))>;
-
-def : Pat<(srl R32C:$rA, R8C:$rB),
-          (ROTMr32 R32C:$rA,
-                   (SFIr32 (XSHWr16 (XSBHr8 R8C:$rB)), 0))>;
-
-// ROTMI v4i32 form: See the comment for ROTHM v8i16.
-def ROTMIv4i32:
-    RI7Form<0b10011110000, (outs VECREG:$rT), (ins VECREG:$rA, rotNeg7imm:$val),
-      "rotmi\t$rT, $rA, $val", RotShiftVec,
-      [(set (v4i32 VECREG:$rT),
-            (SPUvec_srl VECREG:$rA, (i32 uimm7:$val)))]>;
-
-def : Pat<(SPUvec_srl (v4i32 VECREG:$rA), (i16 uimm7:$val)),
-          (ROTMIv4i32 VECREG:$rA, (TO_IMM32 uimm7:$val))>;
-
-def : Pat<(SPUvec_srl (v4i32 VECREG:$rA), (i8 uimm7:$val)),
-          (ROTMIv4i32 VECREG:$rA, (TO_IMM32 uimm7:$val))>;
-
-// ROTMI r32 form: know how to complement the immediate value.
-def ROTMIr32:
-    RI7Form<0b10011110000, (outs R32C:$rT), (ins R32C:$rA, rotNeg7imm:$val),
-      "rotmi\t$rT, $rA, $val", RotShiftVec,
-      [(set R32C:$rT, (srl R32C:$rA, (i32 uimm7:$val)))]>;
-
-def : Pat<(srl R32C:$rA, (i16 imm:$val)),
-          (ROTMIr32 R32C:$rA, (TO_IMM32 uimm7:$val))>;
-
-def : Pat<(srl R32C:$rA, (i8 imm:$val)),
-          (ROTMIr32 R32C:$rA, (TO_IMM32 uimm7:$val))>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// ROTQMBY: This is a vector form merely so that when used in an
-// instruction pattern, type checking will succeed. This instruction assumes
-// that the user knew to negate $rB.
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class ROTQMBYInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10111011100, OOL, IOL, "rotqmby\t$rT, $rA, $rB",
-           RotShiftQuad, pattern>;
-
-class ROTQMBYVecInst<ValueType vectype>:
-    ROTQMBYInst<(outs VECREG:$rT), (ins VECREG:$rA, R32C:$rB),
-                [/* no pattern, $rB must be negated */]>;
-
-class ROTQMBYRegInst<RegisterClass rclass>:
-    ROTQMBYInst<(outs rclass:$rT), (ins rclass:$rA, R32C:$rB),
-                [/* no pattern */]>;
-
-multiclass RotateQuadBytes
-{
-  def v16i8: ROTQMBYVecInst<v16i8>;
-  def v8i16: ROTQMBYVecInst<v8i16>;
-  def v4i32: ROTQMBYVecInst<v4i32>;
-  def v2i64: ROTQMBYVecInst<v2i64>;
-
-  def r128: ROTQMBYRegInst<GPRC>;
-  def r64:  ROTQMBYRegInst<R64C>;
-}
-
-defm ROTQMBY : RotateQuadBytes;
-
-def : Pat<(SPUsrl_bytes GPRC:$rA, R32C:$rB),
-          (ROTQMBYr128  GPRC:$rA, 
-                        (SFIr32 R32C:$rB, 0))>;
-
-class ROTQMBYIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b10111111100, OOL, IOL, "rotqmbyi\t$rT, $rA, $val",
-            RotShiftQuad, pattern>;
-
-class ROTQMBYIVecInst<ValueType vectype>:
-    ROTQMBYIInst<(outs VECREG:$rT), (ins VECREG:$rA, rotNeg7imm:$val),
-                 [/* no pattern */]>;
-
-class ROTQMBYIRegInst<RegisterClass rclass, Operand optype, ValueType inttype,
-                      PatLeaf pred>:
-    ROTQMBYIInst<(outs rclass:$rT), (ins rclass:$rA, optype:$val),
-                 [/* no pattern */]>;
-
-// 128-bit zero extension form:
-class ROTQMBYIZExtInst<RegisterClass rclass, Operand optype, PatLeaf pred>:
-    ROTQMBYIInst<(outs GPRC:$rT), (ins rclass:$rA, optype:$val),
-                 [/* no pattern */]>;
-
-multiclass RotateQuadBytesImm
-{
-  def v16i8: ROTQMBYIVecInst<v16i8>;
-  def v8i16: ROTQMBYIVecInst<v8i16>;
-  def v4i32: ROTQMBYIVecInst<v4i32>;
-  def v2i64: ROTQMBYIVecInst<v2i64>;
-
-  def r128:  ROTQMBYIRegInst<GPRC, rotNeg7imm, i32, uimm7>;
-  def r64:   ROTQMBYIRegInst<R64C, rotNeg7imm, i32, uimm7>;
-  
-  def r128_zext_r8:  ROTQMBYIZExtInst<R8C, rotNeg7imm, uimm7>;
-  def r128_zext_r16: ROTQMBYIZExtInst<R16C, rotNeg7imm, uimm7>;
-  def r128_zext_r32: ROTQMBYIZExtInst<R32C, rotNeg7imm, uimm7>;
-  def r128_zext_r64: ROTQMBYIZExtInst<R64C, rotNeg7imm, uimm7>;
-}
-
-defm ROTQMBYI : RotateQuadBytesImm;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Rotate right and mask by bit count
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class ROTQMBYBIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10110011100, OOL, IOL, "rotqmbybi\t$rT, $rA, $rB",
-           RotShiftQuad, pattern>;
-
-class ROTQMBYBIVecInst<ValueType vectype>:
-    ROTQMBYBIInst<(outs VECREG:$rT), (ins VECREG:$rA, R32C:$rB),
-                  [/* no pattern, */]>;
-
-multiclass RotateMaskQuadByBitCount
-{
-  def v16i8: ROTQMBYBIVecInst<v16i8>;
-  def v8i16: ROTQMBYBIVecInst<v8i16>;
-  def v4i32: ROTQMBYBIVecInst<v4i32>;
-  def v2i64: ROTQMBYBIVecInst<v2i64>;
-  def r128: ROTQMBYBIInst<(outs GPRC:$rT), (ins GPRC:$rA, R32C:$rB),
-                           [/*no pattern*/]>;
-}
-
-defm ROTQMBYBI: RotateMaskQuadByBitCount;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Rotate quad and mask by bits
-// Note that the rotate amount has to be negated
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class ROTQMBIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b10011011100, OOL, IOL, "rotqmbi\t$rT, $rA, $rB",
-           RotShiftQuad, pattern>;
-
-class ROTQMBIVecInst<ValueType vectype>:
-    ROTQMBIInst<(outs VECREG:$rT), (ins VECREG:$rA, R32C:$rB),
-                [/* no pattern */]>;
-
-class ROTQMBIRegInst<RegisterClass rclass>:
-    ROTQMBIInst<(outs rclass:$rT), (ins rclass:$rA, R32C:$rB),
-                [/* no pattern */]>;
-
-multiclass RotateMaskQuadByBits
-{
-  def v16i8: ROTQMBIVecInst<v16i8>;
-  def v8i16: ROTQMBIVecInst<v8i16>;
-  def v4i32: ROTQMBIVecInst<v4i32>;
-  def v2i64: ROTQMBIVecInst<v2i64>;
-
-  def r128:  ROTQMBIRegInst<GPRC>;
-  def r64:   ROTQMBIRegInst<R64C>;
-}
-
-defm ROTQMBI: RotateMaskQuadByBits;
-
-def : Pat<(srl GPRC:$rA, R32C:$rB),
-          (ROTQMBYBIr128 (ROTQMBIr128  GPRC:$rA, 
-                                       (SFIr32 R32C:$rB, 0)),
-                         (SFIr32 R32C:$rB, 0))>;
-
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Rotate quad and mask by bits, immediate
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class ROTQMBIIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RI7Form<0b10011111100, OOL, IOL, "rotqmbii\t$rT, $rA, $val",
-            RotShiftQuad, pattern>;
-
-class ROTQMBIIVecInst<ValueType vectype>:
-   ROTQMBIIInst<(outs VECREG:$rT), (ins VECREG:$rA, rotNeg7imm:$val),
-                 [/* no pattern */]>;
-
-class ROTQMBIIRegInst<RegisterClass rclass>:
-   ROTQMBIIInst<(outs rclass:$rT), (ins rclass:$rA, rotNeg7imm:$val),
-                 [/* no pattern */]>;
-
-multiclass RotateMaskQuadByBitsImm
-{
-  def v16i8: ROTQMBIIVecInst<v16i8>;
-  def v8i16: ROTQMBIIVecInst<v8i16>;
-  def v4i32: ROTQMBIIVecInst<v4i32>;
-  def v2i64: ROTQMBIIVecInst<v2i64>;
-
-  def r128:  ROTQMBIIRegInst<GPRC>;
-  def r64:   ROTQMBIIRegInst<R64C>;
-}
-
-defm ROTQMBII: RotateMaskQuadByBitsImm;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-def ROTMAHv8i16:
-    RRForm<0b01111010000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "rotmah\t$rT, $rA, $rB", RotShiftVec,
-      [/* see patterns below - $rB must be negated */]>;
-
-def : Pat<(SPUvec_sra (v8i16 VECREG:$rA), (v8i16 VECREG:$rB)),
-          (ROTMAHv8i16 VECREG:$rA, (SFHIvec VECREG:$rB, 0))>;
-
-def ROTMAHr16:
-    RRForm<0b01111010000, (outs R16C:$rT), (ins R16C:$rA, R32C:$rB),
-      "rotmah\t$rT, $rA, $rB", RotShiftVec,
-      [/* see patterns below - $rB must be negated */]>;
-
-def : Pat<(sra R16C:$rA, R32C:$rB),
-          (ROTMAHr16 R16C:$rA, (SFIr32 R32C:$rB, 0))>;
-
-def : Pat<(sra R16C:$rA, R16C:$rB),
-          (ROTMAHr16 R16C:$rA,
-                     (SFIr32 (XSHWr16 R16C:$rB), 0))>;
-
-def : Pat<(sra R16C:$rA, R8C:$rB),
-          (ROTMAHr16 R16C:$rA,
-                     (SFIr32 (XSHWr16 (XSBHr8 R8C:$rB)), 0))>;
-
-def ROTMAHIv8i16:
-    RRForm<0b01111110000, (outs VECREG:$rT), (ins VECREG:$rA, rothNeg7imm:$val),
-      "rotmahi\t$rT, $rA, $val", RotShiftVec,
-      [(set (v8i16 VECREG:$rT),
-            (SPUvec_sra (v8i16 VECREG:$rA), (i32 uimm7:$val)))]>;
-
-def : Pat<(SPUvec_sra (v8i16 VECREG:$rA), (i16 uimm7:$val)),
-          (ROTMAHIv8i16 (v8i16 VECREG:$rA), (TO_IMM32 uimm7:$val))>;
-
-def : Pat<(SPUvec_sra (v8i16 VECREG:$rA), (i8 uimm7:$val)),
-          (ROTMAHIv8i16 (v8i16 VECREG:$rA), (TO_IMM32 uimm7:$val))>;
-
-def ROTMAHIr16:
-    RRForm<0b01111110000, (outs R16C:$rT), (ins R16C:$rA, rothNeg7imm_i16:$val),
-      "rotmahi\t$rT, $rA, $val", RotShiftVec,
-      [(set R16C:$rT, (sra R16C:$rA, (i16 uimm7:$val)))]>;
-
-def : Pat<(sra R16C:$rA, (i32 imm:$val)),
-          (ROTMAHIr16 R16C:$rA, (TO_IMM32 uimm7:$val))>;
-
-def : Pat<(sra R16C:$rA, (i8 imm:$val)),
-          (ROTMAHIr16 R16C:$rA, (TO_IMM32 uimm7:$val))>;
-
-def ROTMAv4i32:
-    RRForm<0b01011010000, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "rotma\t$rT, $rA, $rB", RotShiftVec,
-      [/* see patterns below - $rB must be negated */]>;
-
-def : Pat<(SPUvec_sra (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)),
-          (ROTMAv4i32 VECREG:$rA, (SFIvec (v4i32 VECREG:$rB), 0))>;
-
-def ROTMAr32:
-    RRForm<0b01011010000, (outs R32C:$rT), (ins R32C:$rA, R32C:$rB),
-      "rotma\t$rT, $rA, $rB", RotShiftVec,
-      [/* see patterns below - $rB must be negated */]>;
-
-def : Pat<(sra R32C:$rA, R32C:$rB),
-          (ROTMAr32 R32C:$rA, (SFIr32 R32C:$rB, 0))>;
-
-def : Pat<(sra R32C:$rA, R16C:$rB),
-          (ROTMAr32 R32C:$rA,
-                    (SFIr32 (XSHWr16 R16C:$rB), 0))>;
-
-def : Pat<(sra R32C:$rA, R8C:$rB),
-          (ROTMAr32 R32C:$rA,
-                    (SFIr32 (XSHWr16 (XSBHr8 R8C:$rB)), 0))>;
-
-class ROTMAIInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b01011110000, OOL, IOL,
-      "rotmai\t$rT, $rA, $val",
-      RotShiftVec, pattern>;
-
-class ROTMAIVecInst<ValueType vectype, Operand intop, ValueType inttype>:
-    ROTMAIInst<(outs VECREG:$rT), (ins VECREG:$rA, intop:$val),
-      [(set (vectype VECREG:$rT),
-            (SPUvec_sra VECREG:$rA, (inttype uimm7:$val)))]>;
-
-class ROTMAIRegInst<RegisterClass rclass, Operand intop, ValueType inttype>:
-    ROTMAIInst<(outs rclass:$rT), (ins rclass:$rA, intop:$val),
-      [(set rclass:$rT, (sra rclass:$rA, (inttype uimm7:$val)))]>;
-
-multiclass RotateMaskAlgebraicImm {
-  def v2i64_i32 : ROTMAIVecInst<v2i64, rotNeg7imm, i32>;
-  def v4i32_i32 : ROTMAIVecInst<v4i32, rotNeg7imm, i32>;
-  def r64_i32 : ROTMAIRegInst<R64C, rotNeg7imm, i32>;
-  def r32_i32 : ROTMAIRegInst<R32C, rotNeg7imm, i32>;
-}
-
-defm ROTMAI : RotateMaskAlgebraicImm;
-
-//===----------------------------------------------------------------------===//
-// Branch and conditionals:
-//===----------------------------------------------------------------------===//
-
-let isTerminator = 1, isBarrier = 1 in {
-  // Halt If Equal (r32 preferred slot only, no vector form)
-  def HEQr32:
-    RRForm_3<0b00011011110, (outs), (ins R32C:$rA, R32C:$rB),
-      "heq\t$rA, $rB", BranchResolv,
-      [/* no pattern to match */]>;
-
-  def HEQIr32 :
-    RI10Form_2<0b11111110, (outs), (ins R32C:$rA, s10imm:$val),
-      "heqi\t$rA, $val", BranchResolv,
-      [/* no pattern to match */]>;
-
-  // HGT/HGTI: These instructions use signed arithmetic for the comparison,
-  // contrasting with HLGT/HLGTI, which use unsigned comparison:
-  def HGTr32:
-    RRForm_3<0b00011010010, (outs), (ins R32C:$rA, R32C:$rB),
-      "hgt\t$rA, $rB", BranchResolv,
-      [/* no pattern to match */]>;
-
-  def HGTIr32:
-    RI10Form_2<0b11110010, (outs), (ins R32C:$rA, s10imm:$val),
-      "hgti\t$rA, $val", BranchResolv,
-      [/* no pattern to match */]>;
-
-  def HLGTr32:
-    RRForm_3<0b00011011010, (outs), (ins R32C:$rA, R32C:$rB),
-      "hlgt\t$rA, $rB", BranchResolv,
-      [/* no pattern to match */]>;
-
-  def HLGTIr32:
-    RI10Form_2<0b11111010, (outs), (ins R32C:$rA, s10imm:$val),
-      "hlgti\t$rA, $val", BranchResolv,
-      [/* no pattern to match */]>;
-}
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// Comparison operators for i8, i16 and i32:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class CEQBInst<dag OOL, dag IOL, list<dag> pattern> :
-  RRForm<0b00001011110, OOL, IOL, "ceqb\t$rT, $rA, $rB",
-         ByteOp, pattern>;
-
-multiclass CmpEqualByte
-{
-  def v16i8 :
-    CEQBInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      [(set (v16i8 VECREG:$rT), (seteq (v8i16 VECREG:$rA),
-                                       (v8i16 VECREG:$rB)))]>;
-
-  def r8 :
-    CEQBInst<(outs R8C:$rT), (ins R8C:$rA, R8C:$rB),
-             [(set R8C:$rT, (seteq R8C:$rA, R8C:$rB))]>;
-}
-
-class CEQBIInst<dag OOL, dag IOL, list<dag> pattern> :
-  RI10Form<0b01111110, OOL, IOL, "ceqbi\t$rT, $rA, $val",
-           ByteOp, pattern>;
-
-multiclass CmpEqualByteImm
-{
-  def v16i8 :
-    CEQBIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm_i8:$val),
-              [(set (v16i8 VECREG:$rT), (seteq (v16i8 VECREG:$rA),
-                                               v16i8SExt8Imm:$val))]>;
-  def r8:
-    CEQBIInst<(outs R8C:$rT), (ins R8C:$rA, s10imm_i8:$val),
-             [(set R8C:$rT, (seteq R8C:$rA, immSExt8:$val))]>;
-}
-
-class CEQHInst<dag OOL, dag IOL, list<dag> pattern> :
-  RRForm<0b00010011110, OOL, IOL, "ceqh\t$rT, $rA, $rB",
-         ByteOp, pattern>;
-
-multiclass CmpEqualHalfword
-{
-  def v8i16 : CEQHInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-                       [(set (v8i16 VECREG:$rT), (seteq (v8i16 VECREG:$rA),
-                                                        (v8i16 VECREG:$rB)))]>;
-
-  def r16 : CEQHInst<(outs R16C:$rT), (ins R16C:$rA, R16C:$rB),
-                     [(set R16C:$rT, (seteq R16C:$rA, R16C:$rB))]>;
-}
-
-class CEQHIInst<dag OOL, dag IOL, list<dag> pattern> :
-  RI10Form<0b10111110, OOL, IOL, "ceqhi\t$rT, $rA, $val",
-           ByteOp, pattern>;
-
-multiclass CmpEqualHalfwordImm
-{
-  def v8i16 : CEQHIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-                        [(set (v8i16 VECREG:$rT),
-                              (seteq (v8i16 VECREG:$rA),
-                                     (v8i16 v8i16SExt10Imm:$val)))]>;
-  def r16 : CEQHIInst<(outs R16C:$rT), (ins R16C:$rA, s10imm:$val),
-                      [(set R16C:$rT, (seteq R16C:$rA, i16ImmSExt10:$val))]>;
-}
-
-class CEQInst<dag OOL, dag IOL, list<dag> pattern> :
-  RRForm<0b00000011110, OOL, IOL, "ceq\t$rT, $rA, $rB",
-         ByteOp, pattern>;
-
-multiclass CmpEqualWord
-{
-  def v4i32 : CEQInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-                      [(set (v4i32 VECREG:$rT),
-                            (seteq (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-  def r32 : CEQInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB),
-                    [(set R32C:$rT, (seteq R32C:$rA, R32C:$rB))]>;
-}
-
-class CEQIInst<dag OOL, dag IOL, list<dag> pattern> :
-  RI10Form<0b00111110, OOL, IOL, "ceqi\t$rT, $rA, $val",
-           ByteOp, pattern>;
-
-multiclass CmpEqualWordImm
-{
-  def v4i32 : CEQIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-                       [(set (v4i32 VECREG:$rT),
-                             (seteq (v4i32 VECREG:$rA),
-                                    (v4i32 v4i32SExt16Imm:$val)))]>;
-
-  def r32: CEQIInst<(outs R32C:$rT), (ins R32C:$rA, s10imm_i32:$val),
-                    [(set R32C:$rT, (seteq R32C:$rA, i32ImmSExt10:$val))]>;
-}
-
-class CGTBInst<dag OOL, dag IOL, list<dag> pattern> :
-  RRForm<0b00001010010, OOL, IOL, "cgtb\t$rT, $rA, $rB",
-         ByteOp, pattern>;
-
-multiclass CmpGtrByte
-{
-  def v16i8 :
-    CGTBInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      [(set (v16i8 VECREG:$rT), (setgt (v8i16 VECREG:$rA),
-                                       (v8i16 VECREG:$rB)))]>;
-
-  def r8 :
-    CGTBInst<(outs R8C:$rT), (ins R8C:$rA, R8C:$rB),
-             [(set R8C:$rT, (setgt R8C:$rA, R8C:$rB))]>;
-}
-
-class CGTBIInst<dag OOL, dag IOL, list<dag> pattern> :
-  RI10Form<0b01110010, OOL, IOL, "cgtbi\t$rT, $rA, $val",
-           ByteOp, pattern>;
-
-multiclass CmpGtrByteImm
-{
-  def v16i8 :
-    CGTBIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm_i8:$val),
-              [(set (v16i8 VECREG:$rT), (setgt (v16i8 VECREG:$rA),
-                                               v16i8SExt8Imm:$val))]>;
-  def r8:
-    CGTBIInst<(outs R8C:$rT), (ins R8C:$rA, s10imm_i8:$val),
-              [(set R8C:$rT, (setgt R8C:$rA, immSExt8:$val))]>;
-}
-
-class CGTHInst<dag OOL, dag IOL, list<dag> pattern> :
-  RRForm<0b00010010010, OOL, IOL, "cgth\t$rT, $rA, $rB",
-         ByteOp, pattern>;
-
-multiclass CmpGtrHalfword
-{
-  def v8i16 : CGTHInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-                       [(set (v8i16 VECREG:$rT), (setgt (v8i16 VECREG:$rA),
-                                                        (v8i16 VECREG:$rB)))]>;
-
-  def r16 : CGTHInst<(outs R16C:$rT), (ins R16C:$rA, R16C:$rB),
-                     [(set R16C:$rT, (setgt R16C:$rA, R16C:$rB))]>;
-}
-
-class CGTHIInst<dag OOL, dag IOL, list<dag> pattern> :
-  RI10Form<0b10110010, OOL, IOL, "cgthi\t$rT, $rA, $val",
-           ByteOp, pattern>;
-
-multiclass CmpGtrHalfwordImm
-{
-  def v8i16 : CGTHIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-                        [(set (v8i16 VECREG:$rT),
-                              (setgt (v8i16 VECREG:$rA),
-                                     (v8i16 v8i16SExt10Imm:$val)))]>;
-  def r16 : CGTHIInst<(outs R16C:$rT), (ins R16C:$rA, s10imm:$val),
-                      [(set R16C:$rT, (setgt R16C:$rA, i16ImmSExt10:$val))]>;
-}
-
-class CGTInst<dag OOL, dag IOL, list<dag> pattern> :
-  RRForm<0b00000010010, OOL, IOL, "cgt\t$rT, $rA, $rB",
-         ByteOp, pattern>;
-
-multiclass CmpGtrWord
-{
-  def v4i32 : CGTInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-                      [(set (v4i32 VECREG:$rT),
-                            (setgt (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-  def r32 : CGTInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB),
-                    [(set R32C:$rT, (setgt R32C:$rA, R32C:$rB))]>;
-}
-
-class CGTIInst<dag OOL, dag IOL, list<dag> pattern> :
-  RI10Form<0b00110010, OOL, IOL, "cgti\t$rT, $rA, $val",
-           ByteOp, pattern>;
-
-multiclass CmpGtrWordImm
-{
-  def v4i32 : CGTIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-                       [(set (v4i32 VECREG:$rT),
-                             (setgt (v4i32 VECREG:$rA),
-                                    (v4i32 v4i32SExt16Imm:$val)))]>;
-
-  def r32: CGTIInst<(outs R32C:$rT), (ins R32C:$rA, s10imm_i32:$val),
-                    [(set R32C:$rT, (setgt R32C:$rA, i32ImmSExt10:$val))]>;
-
-  // CGTIv4f32, CGTIf32: These are used in the f32 fdiv instruction sequence:
-  def v4f32: CGTIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-                       [(set (v4i32 VECREG:$rT),
-                             (setgt (v4i32 (bitconvert (v4f32 VECREG:$rA))),
-                                    (v4i32 v4i32SExt16Imm:$val)))]>;
-
-  def f32:   CGTIInst<(outs R32C:$rT), (ins R32FP:$rA, s10imm_i32:$val),
-                      [/* no pattern */]>;
-}
-
-class CLGTBInst<dag OOL, dag IOL, list<dag> pattern> :
-  RRForm<0b00001011010, OOL, IOL, "clgtb\t$rT, $rA, $rB",
-         ByteOp, pattern>;
-
-multiclass CmpLGtrByte
-{
-  def v16i8 :
-    CLGTBInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      [(set (v16i8 VECREG:$rT), (setugt (v8i16 VECREG:$rA),
-                                       (v8i16 VECREG:$rB)))]>;
-
-  def r8 :
-    CLGTBInst<(outs R8C:$rT), (ins R8C:$rA, R8C:$rB),
-             [(set R8C:$rT, (setugt R8C:$rA, R8C:$rB))]>;
-}
-
-class CLGTBIInst<dag OOL, dag IOL, list<dag> pattern> :
-  RI10Form<0b01111010, OOL, IOL, "clgtbi\t$rT, $rA, $val",
-           ByteOp, pattern>;
-
-multiclass CmpLGtrByteImm
-{
-  def v16i8 :
-    CLGTBIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm_i8:$val),
-              [(set (v16i8 VECREG:$rT), (setugt (v16i8 VECREG:$rA),
-                                               v16i8SExt8Imm:$val))]>;
-  def r8:
-    CLGTBIInst<(outs R8C:$rT), (ins R8C:$rA, s10imm_i8:$val),
-             [(set R8C:$rT, (setugt R8C:$rA, immSExt8:$val))]>;
-}
-
-class CLGTHInst<dag OOL, dag IOL, list<dag> pattern> :
-  RRForm<0b00010011010, OOL, IOL, "clgth\t$rT, $rA, $rB",
-         ByteOp, pattern>;
-
-multiclass CmpLGtrHalfword
-{
-  def v8i16 : CLGTHInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-                       [(set (v8i16 VECREG:$rT), (setugt (v8i16 VECREG:$rA),
-                                                        (v8i16 VECREG:$rB)))]>;
-
-  def r16 : CLGTHInst<(outs R16C:$rT), (ins R16C:$rA, R16C:$rB),
-                     [(set R16C:$rT, (setugt R16C:$rA, R16C:$rB))]>;
-}
-
-class CLGTHIInst<dag OOL, dag IOL, list<dag> pattern> :
-  RI10Form<0b10111010, OOL, IOL, "clgthi\t$rT, $rA, $val",
-           ByteOp, pattern>;
-
-multiclass CmpLGtrHalfwordImm
-{
-  def v8i16 : CLGTHIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-                         [(set (v8i16 VECREG:$rT),
-                               (setugt (v8i16 VECREG:$rA),
-                                       (v8i16 v8i16SExt10Imm:$val)))]>;
-  def r16 : CLGTHIInst<(outs R16C:$rT), (ins R16C:$rA, s10imm:$val),
-                       [(set R16C:$rT, (setugt R16C:$rA, i16ImmSExt10:$val))]>;
-}
-
-class CLGTInst<dag OOL, dag IOL, list<dag> pattern> :
-  RRForm<0b00000011010, OOL, IOL, "clgt\t$rT, $rA, $rB",
-         ByteOp, pattern>;
-
-multiclass CmpLGtrWord
-{
-  def v4i32 : CLGTInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-                      [(set (v4i32 VECREG:$rT),
-                            (setugt (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)))]>;
-
-  def r32 : CLGTInst<(outs R32C:$rT), (ins R32C:$rA, R32C:$rB),
-                     [(set R32C:$rT, (setugt R32C:$rA, R32C:$rB))]>;
-}
-
-class CLGTIInst<dag OOL, dag IOL, list<dag> pattern> :
-  RI10Form<0b00111010, OOL, IOL, "clgti\t$rT, $rA, $val",
-           ByteOp, pattern>;
-
-multiclass CmpLGtrWordImm
-{
-  def v4i32 : CLGTIInst<(outs VECREG:$rT), (ins VECREG:$rA, s10imm:$val),
-                       [(set (v4i32 VECREG:$rT),
-                             (setugt (v4i32 VECREG:$rA),
-                                    (v4i32 v4i32SExt16Imm:$val)))]>;
-
-  def r32: CLGTIInst<(outs R32C:$rT), (ins R32C:$rA, s10imm_i32:$val),
-                     [(set R32C:$rT, (setugt R32C:$rA, i32ImmSExt10:$val))]>;
-}
-
-defm CEQB   : CmpEqualByte;
-defm CEQBI  : CmpEqualByteImm;
-defm CEQH   : CmpEqualHalfword;
-defm CEQHI  : CmpEqualHalfwordImm;
-defm CEQ    : CmpEqualWord;
-defm CEQI   : CmpEqualWordImm;
-defm CGTB   : CmpGtrByte;
-defm CGTBI  : CmpGtrByteImm;
-defm CGTH   : CmpGtrHalfword;
-defm CGTHI  : CmpGtrHalfwordImm;
-defm CGT    : CmpGtrWord;
-defm CGTI   : CmpGtrWordImm;
-defm CLGTB  : CmpLGtrByte;
-defm CLGTBI : CmpLGtrByteImm;
-defm CLGTH  : CmpLGtrHalfword;
-defm CLGTHI : CmpLGtrHalfwordImm;
-defm CLGT   : CmpLGtrWord;
-defm CLGTI  : CmpLGtrWordImm;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// For SETCC primitives not supported above (setlt, setle, setge, etc.)
-// define a pattern to generate the right code, as a binary operator
-// (in a manner of speaking.)
-//
-// Notes:
-// 1. This only matches the setcc set of conditionals. Special pattern
-//    matching is used for select conditionals.
-//
-// 2. The "DAG" versions of these classes is almost exclusively used for
-//    i64 comparisons. See the tblgen fundamentals documentation for what
-//    ".ResultInstrs[0]" means; see TargetSelectionDAG.td and the Pattern
-//    class for where ResultInstrs originates.
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class SETCCNegCondReg<PatFrag cond, RegisterClass rclass, ValueType inttype,
-                      SPUInstr xorinst, SPUInstr cmpare>:
-  Pat<(cond rclass:$rA, rclass:$rB),
-      (xorinst (cmpare rclass:$rA, rclass:$rB), (inttype -1))>;
-
-class SETCCNegCondImm<PatFrag cond, RegisterClass rclass, ValueType inttype,
-                      PatLeaf immpred, SPUInstr xorinst, SPUInstr cmpare>:
-  Pat<(cond rclass:$rA, (inttype immpred:$imm)),
-      (xorinst (cmpare rclass:$rA, (inttype immpred:$imm)), (inttype -1))>;
-
-def : SETCCNegCondReg<setne, R8C, i8, XORBIr8,  CEQBr8>;
-def : SETCCNegCondImm<setne, R8C, i8, immSExt8, XORBIr8, CEQBIr8>;
-
-def : SETCCNegCondReg<setne, R16C, i16, XORHIr16,     CEQHr16>;
-def : SETCCNegCondImm<setne, R16C, i16, i16ImmSExt10, XORHIr16, CEQHIr16>;
-
-def : SETCCNegCondReg<setne, R32C, i32, XORIr32, CEQr32>;
-def : SETCCNegCondImm<setne, R32C, i32, i32ImmSExt10, XORIr32, CEQIr32>;
-
-class SETCCBinOpReg<PatFrag cond, RegisterClass rclass,
-                    SPUInstr binop, SPUInstr cmpOp1, SPUInstr cmpOp2>:
-    Pat<(cond rclass:$rA, rclass:$rB),
-        (binop (cmpOp1 rclass:$rA, rclass:$rB),
-               (cmpOp2 rclass:$rA, rclass:$rB))>;
-
-class SETCCBinOpImm<PatFrag cond, RegisterClass rclass, PatLeaf immpred,
-                    ValueType immtype,
-                    SPUInstr binop, SPUInstr cmpOp1, SPUInstr cmpOp2>:
-    Pat<(cond rclass:$rA, (immtype immpred:$imm)),
-        (binop (cmpOp1 rclass:$rA, (immtype immpred:$imm)),
-               (cmpOp2 rclass:$rA, (immtype immpred:$imm)))>;
-
-def : SETCCBinOpReg<setge, R8C, ORr8, CGTBr8, CEQBr8>;
-def : SETCCBinOpImm<setge, R8C, immSExt8, i8, ORr8, CGTBIr8, CEQBIr8>;
-def : SETCCBinOpReg<setlt, R8C, NORr8, CGTBr8, CEQBr8>;
-def : SETCCBinOpImm<setlt, R8C, immSExt8, i8, NORr8, CGTBIr8, CEQBIr8>;
-def : Pat<(setle R8C:$rA, R8C:$rB),
-          (XORBIr8 (CGTBr8 R8C:$rA, R8C:$rB), 0xff)>;
-def :  Pat<(setle R8C:$rA, immU8:$imm),
-           (XORBIr8 (CGTBIr8 R8C:$rA, immU8:$imm), 0xff)>;
-
-def : SETCCBinOpReg<setge, R16C, ORr16, CGTHr16, CEQHr16>;
-def : SETCCBinOpImm<setge, R16C, i16ImmSExt10, i16,
-                    ORr16, CGTHIr16, CEQHIr16>;
-def : SETCCBinOpReg<setlt, R16C, NORr16, CGTHr16, CEQHr16>;
-def : SETCCBinOpImm<setlt, R16C, i16ImmSExt10, i16, NORr16, CGTHIr16, CEQHIr16>;
-def : Pat<(setle R16C:$rA, R16C:$rB),
-          (XORHIr16 (CGTHr16 R16C:$rA, R16C:$rB), 0xffff)>;
-def : Pat<(setle R16C:$rA, i16ImmSExt10:$imm),
-          (XORHIr16 (CGTHIr16 R16C:$rA, i16ImmSExt10:$imm), 0xffff)>;
-
-def : SETCCBinOpReg<setge, R32C, ORr32, CGTr32, CEQr32>;
-def : SETCCBinOpImm<setge, R32C, i32ImmSExt10, i32,
-                    ORr32, CGTIr32, CEQIr32>;
-def : SETCCBinOpReg<setlt, R32C, NORr32, CGTr32, CEQr32>;
-def : SETCCBinOpImm<setlt, R32C, i32ImmSExt10, i32, NORr32, CGTIr32, CEQIr32>;
-def : Pat<(setle R32C:$rA, R32C:$rB),
-          (XORIr32 (CGTr32 R32C:$rA, R32C:$rB), 0xffffffff)>;
-def : Pat<(setle R32C:$rA, i32ImmSExt10:$imm),
-          (XORIr32 (CGTIr32 R32C:$rA, i32ImmSExt10:$imm), 0xffffffff)>;
-
-def : SETCCBinOpReg<setuge, R8C, ORr8, CLGTBr8, CEQBr8>;
-def : SETCCBinOpImm<setuge, R8C, immSExt8, i8, ORr8, CLGTBIr8, CEQBIr8>;
-def : SETCCBinOpReg<setult, R8C, NORr8, CLGTBr8, CEQBr8>;
-def : SETCCBinOpImm<setult, R8C, immSExt8, i8, NORr8, CLGTBIr8, CEQBIr8>;
-def : Pat<(setule R8C:$rA, R8C:$rB),
-          (XORBIr8 (CLGTBr8 R8C:$rA, R8C:$rB), 0xff)>;
-def :  Pat<(setule R8C:$rA, immU8:$imm),
-           (XORBIr8 (CLGTBIr8 R8C:$rA, immU8:$imm), 0xff)>;
-
-def : SETCCBinOpReg<setuge, R16C, ORr16, CLGTHr16, CEQHr16>;
-def : SETCCBinOpImm<setuge, R16C, i16ImmSExt10, i16,
-                    ORr16, CLGTHIr16, CEQHIr16>;
-def : SETCCBinOpReg<setult, R16C, NORr16, CLGTHr16, CEQHr16>;
-def : SETCCBinOpImm<setult, R16C, i16ImmSExt10, i16, NORr16,
-                    CLGTHIr16, CEQHIr16>;
-def : Pat<(setule R16C:$rA, R16C:$rB),
-          (XORHIr16 (CLGTHr16 R16C:$rA, R16C:$rB), 0xffff)>;
-def :  Pat<(setule R16C:$rA, i16ImmSExt10:$imm),
-           (XORHIr16 (CLGTHIr16 R16C:$rA, i16ImmSExt10:$imm), 0xffff)>;
-
-def : SETCCBinOpReg<setuge, R32C, ORr32, CLGTr32, CEQr32>;
-def : SETCCBinOpImm<setuge, R32C, i32ImmSExt10, i32,
-                    ORr32, CLGTIr32, CEQIr32>;
-def : SETCCBinOpReg<setult, R32C, NORr32, CLGTr32, CEQr32>;
-def : SETCCBinOpImm<setult, R32C, i32ImmSExt10, i32, NORr32, CLGTIr32, CEQIr32>;
-def : Pat<(setule R32C:$rA, R32C:$rB),
-          (XORIr32 (CLGTr32 R32C:$rA, R32C:$rB), 0xffffffff)>;
-def : Pat<(setule R32C:$rA, i32ImmSExt10:$imm),
-          (XORIr32 (CLGTIr32 R32C:$rA, i32ImmSExt10:$imm), 0xffffffff)>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// select conditional patterns:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-class SELECTNegCondReg<PatFrag cond, RegisterClass rclass, ValueType inttype,
-                       SPUInstr selinstr, SPUInstr cmpare>:
-  Pat<(select (inttype (cond rclass:$rA, rclass:$rB)),
-              rclass:$rTrue, rclass:$rFalse),
-      (selinstr rclass:$rTrue, rclass:$rFalse,
-                (cmpare rclass:$rA, rclass:$rB))>;
-
-class SELECTNegCondImm<PatFrag cond, RegisterClass rclass, ValueType inttype,
-                       PatLeaf immpred, SPUInstr selinstr, SPUInstr cmpare>:
-  Pat<(select (inttype (cond rclass:$rA, immpred:$imm)),
-              rclass:$rTrue, rclass:$rFalse),
-      (selinstr rclass:$rTrue, rclass:$rFalse,
-                (cmpare rclass:$rA, immpred:$imm))>;
-
-def : SELECTNegCondReg<setne, R8C, i8, SELBr8, CEQBr8>;
-def : SELECTNegCondImm<setne, R8C, i8, immSExt8, SELBr8, CEQBIr8>;
-def : SELECTNegCondReg<setle, R8C, i8, SELBr8, CGTBr8>;
-def : SELECTNegCondImm<setle, R8C, i8, immSExt8, SELBr8, CGTBr8>;
-def : SELECTNegCondReg<setule, R8C, i8, SELBr8, CLGTBr8>;
-def : SELECTNegCondImm<setule, R8C, i8, immU8, SELBr8, CLGTBIr8>;
-
-def : SELECTNegCondReg<setne, R16C, i16, SELBr16, CEQHr16>;
-def : SELECTNegCondImm<setne, R16C, i16, i16ImmSExt10, SELBr16, CEQHIr16>;
-def : SELECTNegCondReg<setle, R16C, i16, SELBr16, CGTHr16>;
-def : SELECTNegCondImm<setle, R16C, i16, i16ImmSExt10, SELBr16, CGTHIr16>;
-def : SELECTNegCondReg<setule, R16C, i16, SELBr16, CLGTHr16>;
-def : SELECTNegCondImm<setule, R16C, i16, i16ImmSExt10, SELBr16, CLGTHIr16>;
-
-def : SELECTNegCondReg<setne, R32C, i32, SELBr32, CEQr32>;
-def : SELECTNegCondImm<setne, R32C, i32, i32ImmSExt10, SELBr32, CEQIr32>;
-def : SELECTNegCondReg<setle, R32C, i32, SELBr32, CGTr32>;
-def : SELECTNegCondImm<setle, R32C, i32, i32ImmSExt10, SELBr32, CGTIr32>;
-def : SELECTNegCondReg<setule, R32C, i32, SELBr32, CLGTr32>;
-def : SELECTNegCondImm<setule, R32C, i32, i32ImmSExt10, SELBr32, CLGTIr32>;
-
-class SELECTBinOpReg<PatFrag cond, RegisterClass rclass, ValueType inttype,
-                     SPUInstr selinstr, SPUInstr binop, SPUInstr cmpOp1,
-                     SPUInstr cmpOp2>:
-  Pat<(select (inttype (cond rclass:$rA, rclass:$rB)),
-              rclass:$rTrue, rclass:$rFalse),
-      (selinstr rclass:$rFalse, rclass:$rTrue,
-                (binop (cmpOp1 rclass:$rA, rclass:$rB),
-                       (cmpOp2 rclass:$rA, rclass:$rB)))>;
-
-class SELECTBinOpImm<PatFrag cond, RegisterClass rclass, PatLeaf immpred,
-                     ValueType inttype,
-                     SPUInstr selinstr, SPUInstr binop, SPUInstr cmpOp1,
-                     SPUInstr cmpOp2>:
-    Pat<(select (inttype (cond rclass:$rA, (inttype immpred:$imm))),
-                rclass:$rTrue, rclass:$rFalse),
-        (selinstr rclass:$rFalse, rclass:$rTrue,
-                  (binop (cmpOp1 rclass:$rA, (inttype immpred:$imm)),
-                         (cmpOp2 rclass:$rA, (inttype immpred:$imm))))>;
-
-def : SELECTBinOpReg<setge, R8C, i8, SELBr8, ORr8, CGTBr8, CEQBr8>;
-def : SELECTBinOpImm<setge, R8C, immSExt8, i8,
-                     SELBr8, ORr8, CGTBIr8, CEQBIr8>;
-
-def : SELECTBinOpReg<setge, R16C, i16, SELBr16, ORr16, CGTHr16, CEQHr16>;
-def : SELECTBinOpImm<setge, R16C, i16ImmSExt10, i16,
-                     SELBr16, ORr16, CGTHIr16, CEQHIr16>;
-
-def : SELECTBinOpReg<setge, R32C, i32, SELBr32, ORr32, CGTr32, CEQr32>;
-def : SELECTBinOpImm<setge, R32C, i32ImmSExt10, i32,
-                     SELBr32, ORr32, CGTIr32, CEQIr32>;
-
-def : SELECTBinOpReg<setuge, R8C, i8, SELBr8, ORr8, CLGTBr8, CEQBr8>;
-def : SELECTBinOpImm<setuge, R8C, immSExt8, i8,
-                     SELBr8, ORr8, CLGTBIr8, CEQBIr8>;
-
-def : SELECTBinOpReg<setuge, R16C, i16, SELBr16, ORr16, CLGTHr16, CEQHr16>;
-def : SELECTBinOpImm<setuge, R16C, i16ImmUns10, i16,
-                     SELBr16, ORr16, CLGTHIr16, CEQHIr16>;
-
-def : SELECTBinOpReg<setuge, R32C, i32, SELBr32, ORr32, CLGTr32, CEQr32>;
-def : SELECTBinOpImm<setuge, R32C, i32ImmUns10, i32,
-                     SELBr32, ORr32, CLGTIr32, CEQIr32>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-let isCall = 1,
-  // All calls clobber the non-callee-saved registers:
-  Defs = [R0, R1, R2, R3, R4, R5, R6, R7, R8, R9,
-          R10,R11,R12,R13,R14,R15,R16,R17,R18,R19,
-          R20,R21,R22,R23,R24,R25,R26,R27,R28,R29,
-          R30,R31,R32,R33,R34,R35,R36,R37,R38,R39,
-          R40,R41,R42,R43,R44,R45,R46,R47,R48,R49,
-          R50,R51,R52,R53,R54,R55,R56,R57,R58,R59,
-          R60,R61,R62,R63,R64,R65,R66,R67,R68,R69,
-          R70,R71,R72,R73,R74,R75,R76,R77,R78,R79],
-  // All of these instructions use $lr (aka $0)
-  Uses = [R0]  in {
-  // Branch relative and set link: Used if we actually know that the target
-  // is within [-32768, 32767] bytes of the target
-  def BRSL:
-    BranchSetLink<0b011001100, (outs), (ins relcalltarget:$func),
-      "brsl\t$$lr, $func",
-      [(SPUcall (SPUpcrel tglobaladdr:$func, 0))]>;
-
-  // Branch absolute and set link: Used if we actually know that the target
-  // is an absolute address
-  def BRASL:
-    BranchSetLink<0b011001100, (outs), (ins calltarget:$func),
-      "brasl\t$$lr, $func",
-      [(SPUcall (SPUaform tglobaladdr:$func, 0))]>;
-
-  // Branch indirect and set link if external data. These instructions are not
-  // actually generated, matched by an intrinsic:
-  def BISLED_00: BISLEDForm<0b11, "bisled\t$$lr, $func", [/* empty pattern */]>;
-  def BISLED_E0: BISLEDForm<0b10, "bisled\t$$lr, $func", [/* empty pattern */]>;
-  def BISLED_0D: BISLEDForm<0b01, "bisled\t$$lr, $func", [/* empty pattern */]>;
-  def BISLED_ED: BISLEDForm<0b00, "bisled\t$$lr, $func", [/* empty pattern */]>;
-
-  // Branch indirect and set link. This is the "X-form" address version of a
-  // function call
-  def BISL:
-    BIForm<0b10010101100, "bisl\t$$lr, $func", [(SPUcall R32C:$func)]>;
-}
-
-// Support calls to external symbols:      
-def : Pat<(SPUcall (SPUpcrel texternalsym:$func, 0)),
-          (BRSL texternalsym:$func)>;
-      
-def : Pat<(SPUcall (SPUaform texternalsym:$func, 0)),
-          (BRASL texternalsym:$func)>;
-
-// Unconditional branches:
-let isBranch = 1, isTerminator = 1, hasCtrlDep = 1 in {
-  let isBarrier = 1 in {
-    def BR :
-      UncondBranch<0b001001100, (outs), (ins brtarget:$dest),
-        "br\t$dest",
-        [(br bb:$dest)]>;
-
-    // Unconditional, absolute address branch
-    def BRA:
-      UncondBranch<0b001100000, (outs), (ins brtarget:$dest),
-        "bra\t$dest",
-        [/* no pattern */]>;
-
-    // Indirect branch
-    let isIndirectBranch = 1 in {
-      def BI:
-        BIForm<0b00010101100, "bi\t$func", [(brind R32C:$func)]>;
-    }
-  }
-
-  // Conditional branches:
-  class BRNZInst<dag IOL, list<dag> pattern>:
-    RI16Form<0b010000100, (outs), IOL, "brnz\t$rCond,$dest",
-             BranchResolv, pattern>;
-
-  class BRNZRegInst<RegisterClass rclass>:
-    BRNZInst<(ins rclass:$rCond, brtarget:$dest),
-             [(brcond rclass:$rCond, bb:$dest)]>;
-
-  class BRNZVecInst<ValueType vectype>:
-    BRNZInst<(ins VECREG:$rCond, brtarget:$dest),
-             [(brcond (vectype VECREG:$rCond), bb:$dest)]>;
-
-  multiclass BranchNotZero {
-    def v4i32 : BRNZVecInst<v4i32>;
-    def r32   : BRNZRegInst<R32C>;
-  }
-
-  defm BRNZ : BranchNotZero;
-
-  class BRZInst<dag IOL, list<dag> pattern>:
-    RI16Form<0b000000100, (outs), IOL, "brz\t$rT,$dest",
-             BranchResolv, pattern>;
-
-  class BRZRegInst<RegisterClass rclass>:
-    BRZInst<(ins rclass:$rT, brtarget:$dest), [/* no pattern */]>;
-
-  class BRZVecInst<ValueType vectype>:
-    BRZInst<(ins VECREG:$rT, brtarget:$dest), [/* no pattern */]>;
-
-  multiclass BranchZero {
-    def v4i32: BRZVecInst<v4i32>;
-    def r32:   BRZRegInst<R32C>;
-  }
-
-  defm BRZ: BranchZero;
-
-  // Note: LLVM doesn't do branch conditional, indirect. Otherwise these would
-  // be useful:
-  /*
-  class BINZInst<dag IOL, list<dag> pattern>:
-   BICondForm<0b10010100100, (outs), IOL, "binz\t$rA, $dest", pattern>;
-
-  class BINZRegInst<RegisterClass rclass>:
-    BINZInst<(ins rclass:$rA, brtarget:$dest),
-             [(brcond rclass:$rA, R32C:$dest)]>;
-
-  class BINZVecInst<ValueType vectype>:
-    BINZInst<(ins VECREG:$rA, R32C:$dest),
-             [(brcond (vectype VECREG:$rA), R32C:$dest)]>;
-
-  multiclass BranchNotZeroIndirect {
-    def v4i32: BINZVecInst<v4i32>;
-    def r32:   BINZRegInst<R32C>;
-  }
-
-  defm BINZ: BranchNotZeroIndirect;
-
-  class BIZInst<dag IOL, list<dag> pattern>:
-    BICondForm<0b00010100100, (outs), IOL, "biz\t$rA, $func", pattern>;
-
-  class BIZRegInst<RegisterClass rclass>:
-    BIZInst<(ins rclass:$rA, R32C:$func), [/* no pattern */]>;
-
-  class BIZVecInst<ValueType vectype>:
-    BIZInst<(ins VECREG:$rA, R32C:$func), [/* no pattern */]>;
-
-  multiclass BranchZeroIndirect {
-    def v4i32: BIZVecInst<v4i32>;
-    def r32:   BIZRegInst<R32C>;
-  }
-
-  defm BIZ: BranchZeroIndirect;
-  */
-
-  class BRHNZInst<dag IOL, list<dag> pattern>:
-    RI16Form<0b011000100, (outs), IOL, "brhnz\t$rCond,$dest", BranchResolv,
-             pattern>;
-
-  class BRHNZRegInst<RegisterClass rclass>:
-    BRHNZInst<(ins rclass:$rCond, brtarget:$dest),
-              [(brcond rclass:$rCond, bb:$dest)]>;
-
-  class BRHNZVecInst<ValueType vectype>:
-    BRHNZInst<(ins VECREG:$rCond, brtarget:$dest), [/* no pattern */]>;
-
-  multiclass BranchNotZeroHalfword {
-    def v8i16: BRHNZVecInst<v8i16>;
-    def r16:   BRHNZRegInst<R16C>;
-  }
-
-  defm BRHNZ: BranchNotZeroHalfword;
-
-  class BRHZInst<dag IOL, list<dag> pattern>:
-    RI16Form<0b001000100, (outs), IOL, "brhz\t$rT,$dest", BranchResolv,
-             pattern>;
-
-  class BRHZRegInst<RegisterClass rclass>:
-    BRHZInst<(ins rclass:$rT, brtarget:$dest), [/* no pattern */]>;
-
-  class BRHZVecInst<ValueType vectype>:
-    BRHZInst<(ins VECREG:$rT, brtarget:$dest), [/* no pattern */]>;
-
-  multiclass BranchZeroHalfword {
-    def v8i16: BRHZVecInst<v8i16>;
-    def r16:   BRHZRegInst<R16C>;
-  }
-
-  defm BRHZ: BranchZeroHalfword;
-}
-
-//===----------------------------------------------------------------------===//
-// setcc and brcond patterns:
-//===----------------------------------------------------------------------===//
-
-def : Pat<(brcond (i16 (seteq R16C:$rA, 0)), bb:$dest),
-          (BRHZr16 R16C:$rA, bb:$dest)>;
-def : Pat<(brcond (i16 (setne R16C:$rA, 0)), bb:$dest),
-          (BRHNZr16 R16C:$rA, bb:$dest)>;
-
-def : Pat<(brcond (i32 (seteq R32C:$rA, 0)), bb:$dest),
-          (BRZr32 R32C:$rA, bb:$dest)>;
-def : Pat<(brcond (i32 (setne R32C:$rA, 0)), bb:$dest),
-          (BRNZr32 R32C:$rA, bb:$dest)>;
-
-multiclass BranchCondEQ<PatFrag cond, SPUInstr brinst16, SPUInstr brinst32>
-{
-  def r16imm: Pat<(brcond (i16 (cond R16C:$rA, i16ImmSExt10:$val)), bb:$dest),
-                  (brinst16 (CEQHIr16 R16C:$rA, i16ImmSExt10:$val), bb:$dest)>;
-
-  def r16 : Pat<(brcond (i16 (cond R16C:$rA, R16C:$rB)), bb:$dest),
-                (brinst16 (CEQHr16 R16C:$rA, R16:$rB), bb:$dest)>;
-
-  def r32imm : Pat<(brcond (i32 (cond R32C:$rA, i32ImmSExt10:$val)), bb:$dest),
-                   (brinst32 (CEQIr32 R32C:$rA, i32ImmSExt10:$val), bb:$dest)>;
-
-  def r32 : Pat<(brcond (i32 (cond R32C:$rA, R32C:$rB)), bb:$dest),
-                (brinst32 (CEQr32 R32C:$rA, R32C:$rB), bb:$dest)>;
-}
-
-defm BRCONDeq : BranchCondEQ<seteq, BRHNZr16, BRNZr32>;
-defm BRCONDne : BranchCondEQ<setne, BRHZr16, BRZr32>;
-
-multiclass BranchCondLGT<PatFrag cond, SPUInstr brinst16, SPUInstr brinst32>
-{
-  def r16imm : Pat<(brcond (i16 (cond R16C:$rA, i16ImmSExt10:$val)), bb:$dest),
-                   (brinst16 (CLGTHIr16 R16C:$rA, i16ImmSExt10:$val), bb:$dest)>;
-
-  def r16 : Pat<(brcond (i16 (cond R16C:$rA, R16C:$rB)), bb:$dest),
-                (brinst16 (CLGTHr16 R16C:$rA, R16:$rB), bb:$dest)>;
-
-  def r32imm : Pat<(brcond (i32 (cond R32C:$rA, i32ImmSExt10:$val)), bb:$dest),
-                   (brinst32 (CLGTIr32 R32C:$rA, i32ImmSExt10:$val), bb:$dest)>;
-
-  def r32 : Pat<(brcond (i32 (cond R32C:$rA, R32C:$rB)), bb:$dest),
-                (brinst32 (CLGTr32 R32C:$rA, R32C:$rB), bb:$dest)>;
-}
-
-defm BRCONDugt : BranchCondLGT<setugt, BRHNZr16, BRNZr32>;
-defm BRCONDule : BranchCondLGT<setule, BRHZr16, BRZr32>;
-
-multiclass BranchCondLGTEQ<PatFrag cond, SPUInstr orinst16, SPUInstr brinst16,
-                           SPUInstr orinst32, SPUInstr brinst32>
-{
-  def r16imm: Pat<(brcond (i16 (cond R16C:$rA, i16ImmSExt10:$val)), bb:$dest),
-                  (brinst16 (orinst16 (CLGTHIr16 R16C:$rA, i16ImmSExt10:$val),
-                                      (CEQHIr16 R16C:$rA, i16ImmSExt10:$val)),
-                            bb:$dest)>;
-
-  def r16: Pat<(brcond (i16 (cond R16C:$rA, R16C:$rB)), bb:$dest),
-               (brinst16 (orinst16 (CLGTHr16 R16C:$rA, R16:$rB),
-                                   (CEQHr16 R16C:$rA, R16:$rB)),
-                         bb:$dest)>;
-
-  def r32imm : Pat<(brcond (i32 (cond R32C:$rA, i32ImmSExt10:$val)), bb:$dest),
-                   (brinst32 (orinst32 (CLGTIr32 R32C:$rA, i32ImmSExt10:$val),
-                                       (CEQIr32 R32C:$rA, i32ImmSExt10:$val)),
-                             bb:$dest)>;
-
-  def r32 : Pat<(brcond (i32 (cond R32C:$rA, R32C:$rB)), bb:$dest),
-                (brinst32 (orinst32 (CLGTr32 R32C:$rA, R32C:$rB),
-                                    (CEQr32 R32C:$rA, R32C:$rB)),
-                          bb:$dest)>;
-}
-
-defm BRCONDuge : BranchCondLGTEQ<setuge, ORr16, BRHNZr16, ORr32, BRNZr32>;
-defm BRCONDult : BranchCondLGTEQ<setult, ORr16, BRHZr16, ORr32, BRZr32>;
-
-multiclass BranchCondGT<PatFrag cond, SPUInstr brinst16, SPUInstr brinst32>
-{
-  def r16imm : Pat<(brcond (i16 (cond R16C:$rA, i16ImmSExt10:$val)), bb:$dest),
-                   (brinst16 (CGTHIr16 R16C:$rA, i16ImmSExt10:$val), bb:$dest)>;
-
-  def r16 : Pat<(brcond (i16 (cond R16C:$rA, R16C:$rB)), bb:$dest),
-                (brinst16 (CGTHr16 R16C:$rA, R16:$rB), bb:$dest)>;
-
-  def r32imm : Pat<(brcond (i32 (cond R32C:$rA, i32ImmSExt10:$val)), bb:$dest),
-                   (brinst32 (CGTIr32 R32C:$rA, i32ImmSExt10:$val), bb:$dest)>;
-
-  def r32 : Pat<(brcond (i32 (cond R32C:$rA, R32C:$rB)), bb:$dest),
-                (brinst32 (CGTr32 R32C:$rA, R32C:$rB), bb:$dest)>;
-}
-
-defm BRCONDgt : BranchCondGT<setgt, BRHNZr16, BRNZr32>;
-defm BRCONDle : BranchCondGT<setle, BRHZr16, BRZr32>;
-
-multiclass BranchCondGTEQ<PatFrag cond, SPUInstr orinst16, SPUInstr brinst16,
-                          SPUInstr orinst32, SPUInstr brinst32>
-{
-  def r16imm: Pat<(brcond (i16 (cond R16C:$rA, i16ImmSExt10:$val)), bb:$dest),
-                  (brinst16 (orinst16 (CGTHIr16 R16C:$rA, i16ImmSExt10:$val),
-                                      (CEQHIr16 R16C:$rA, i16ImmSExt10:$val)),
-                            bb:$dest)>;
-
-  def r16: Pat<(brcond (i16 (cond R16C:$rA, R16C:$rB)), bb:$dest),
-               (brinst16 (orinst16 (CGTHr16 R16C:$rA, R16:$rB),
-                                   (CEQHr16 R16C:$rA, R16:$rB)),
-                         bb:$dest)>;
-
-  def r32imm : Pat<(brcond (i32 (cond R32C:$rA, i32ImmSExt10:$val)), bb:$dest),
-                   (brinst32 (orinst32 (CGTIr32 R32C:$rA, i32ImmSExt10:$val),
-                                       (CEQIr32 R32C:$rA, i32ImmSExt10:$val)),
-                             bb:$dest)>;
-
-  def r32 : Pat<(brcond (i32 (cond R32C:$rA, R32C:$rB)), bb:$dest),
-                (brinst32 (orinst32 (CGTr32 R32C:$rA, R32C:$rB),
-                                    (CEQr32 R32C:$rA, R32C:$rB)),
-                          bb:$dest)>;
-}
-
-defm BRCONDge : BranchCondGTEQ<setge, ORr16, BRHNZr16, ORr32, BRNZr32>;
-defm BRCONDlt : BranchCondGTEQ<setlt, ORr16, BRHZr16, ORr32, BRZr32>;
-
-let isTerminator = 1, isBarrier = 1 in {
-  let isReturn = 1 in {
-    def RET:
-        RETForm<"bi\t$$lr", [(retflag)]>;
-  }
-}
-
-//===----------------------------------------------------------------------===//
-// Single precision floating point instructions
-//===----------------------------------------------------------------------===//
-
-class FAInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b01011000100, OOL, IOL, "fa\t$rT, $rA, $rB",
-           SPrecFP, pattern>;
-
-class FAVecInst<ValueType vectype>:
-    FAInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-             [(set (vectype VECREG:$rT),
-                   (fadd (vectype VECREG:$rA), (vectype VECREG:$rB)))]>;
-
-multiclass SFPAdd
-{
-  def v4f32: FAVecInst<v4f32>;
-  def f32:   FAInst<(outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB),
-                    [(set R32FP:$rT, (fadd R32FP:$rA, R32FP:$rB))]>;
-}
-
-defm FA : SFPAdd;
-
-class FSInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b01011000100, OOL, IOL, "fs\t$rT, $rA, $rB",
-           SPrecFP, pattern>;
-
-class FSVecInst<ValueType vectype>:
-    FSInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-           [(set (vectype VECREG:$rT),
-                 (fsub (vectype VECREG:$rA), (vectype VECREG:$rB)))]>;
-
-multiclass SFPSub
-{
-  def v4f32: FSVecInst<v4f32>;
-  def f32:   FSInst<(outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB),
-                    [(set R32FP:$rT, (fsub R32FP:$rA, R32FP:$rB))]>;
-}
-
-defm FS : SFPSub;
-
-class FMInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b01100011010, OOL, IOL,
-      "fm\t$rT, $rA, $rB", SPrecFP,
-      pattern>;
-
-class FMVecInst<ValueType type>:
-    FMInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-           [(set (type VECREG:$rT),
-                 (fmul (type VECREG:$rA), (type VECREG:$rB)))]>;
-
-multiclass SFPMul
-{
-  def v4f32: FMVecInst<v4f32>;
-  def f32:   FMInst<(outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB),
-                     [(set R32FP:$rT, (fmul R32FP:$rA, R32FP:$rB))]>; 
-}
-
-defm FM : SFPMul;
-
-// Floating point multiply and add
-// e.g. d = c + (a * b)
-def FMAv4f32:
-    RRRForm<0b0111, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-      "fma\t$rT, $rA, $rB, $rC", SPrecFP,
-      [(set (v4f32 VECREG:$rT),
-            (fadd (v4f32 VECREG:$rC),
-                  (fmul (v4f32 VECREG:$rA), (v4f32 VECREG:$rB))))]>;
-
-def FMAf32:
-    RRRForm<0b0111, (outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB, R32FP:$rC),
-      "fma\t$rT, $rA, $rB, $rC", SPrecFP,
-      [(set R32FP:$rT, (fadd R32FP:$rC, (fmul R32FP:$rA, R32FP:$rB)))]>;
-
-// FP multiply and subtract
-// Subtracts value in rC from product
-// res = a * b - c
-def FMSv4f32 :
-    RRRForm<0b0111, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-      "fms\t$rT, $rA, $rB, $rC", SPrecFP,
-      [(set (v4f32 VECREG:$rT),
-            (fsub (fmul (v4f32 VECREG:$rA), (v4f32 VECREG:$rB)),
-                  (v4f32 VECREG:$rC)))]>;
-
-def FMSf32 :
-    RRRForm<0b0111, (outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB, R32FP:$rC),
-      "fms\t$rT, $rA, $rB, $rC", SPrecFP,
-      [(set R32FP:$rT,
-            (fsub (fmul R32FP:$rA, R32FP:$rB), R32FP:$rC))]>;
-
-// Floating Negative Mulitply and Subtract
-// Subtracts product from value in rC
-// res = fneg(fms a b c)
-//     = - (a * b - c)
-//     = c - a * b
-// NOTE: subtraction order
-// fsub a b = a - b
-// fs a b = b - a?
-def FNMSf32 :
-    RRRForm<0b1101, (outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB, R32FP:$rC),
-      "fnms\t$rT, $rA, $rB, $rC", SPrecFP,
-      [(set R32FP:$rT, (fsub R32FP:$rC, (fmul R32FP:$rA, R32FP:$rB)))]>;
-
-def FNMSv4f32 :
-    RRRForm<0b1101, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-      "fnms\t$rT, $rA, $rB, $rC", SPrecFP,
-      [(set (v4f32 VECREG:$rT),
-            (fsub (v4f32 VECREG:$rC),
-                  (fmul (v4f32 VECREG:$rA),
-                        (v4f32 VECREG:$rB))))]>;
-
-
-
-
-// Floating point reciprocal estimate
-
-class FRESTInst<dag OOL, dag IOL>:
-  RRForm_1<0b00110111000, OOL, IOL,
-           "frest\t$rT, $rA", SPrecFP,
-           [/* no pattern */]>;
-
-def FRESTv4f32 :
-    FRESTInst<(outs VECREG:$rT), (ins VECREG:$rA)>;
-
-def FRESTf32 :
-    FRESTInst<(outs R32FP:$rT), (ins R32FP:$rA)>;
-
-// Floating point interpolate (used in conjunction with reciprocal estimate)
-def FIv4f32 :
-    RRForm<0b00101011110, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "fi\t$rT, $rA, $rB", SPrecFP,
-      [/* no pattern */]>;
-
-def FIf32 :
-    RRForm<0b00101011110, (outs R32FP:$rT), (ins R32FP:$rA, R32FP:$rB),
-      "fi\t$rT, $rA, $rB", SPrecFP,
-      [/* no pattern */]>;
-
-//--------------------------------------------------------------------------
-// Basic single precision floating point comparisons:
-//
-// Note: There is no support on SPU for single precision NaN. Consequently,
-// ordered and unordered comparisons are the same.
-//--------------------------------------------------------------------------
-
-def FCEQf32 :
-    RRForm<0b01000011110, (outs R32C:$rT), (ins R32FP:$rA, R32FP:$rB),
-      "fceq\t$rT, $rA, $rB", SPrecFP,
-      [(set R32C:$rT, (setueq R32FP:$rA, R32FP:$rB))]>;
-
-def : Pat<(setoeq R32FP:$rA, R32FP:$rB),
-          (FCEQf32 R32FP:$rA, R32FP:$rB)>;
-
-def FCMEQf32 :
-    RRForm<0b01010011110, (outs R32C:$rT), (ins R32FP:$rA, R32FP:$rB),
-      "fcmeq\t$rT, $rA, $rB", SPrecFP,
-      [(set R32C:$rT, (setueq (fabs R32FP:$rA), (fabs R32FP:$rB)))]>;
-
-def : Pat<(setoeq (fabs R32FP:$rA), (fabs R32FP:$rB)),
-          (FCMEQf32 R32FP:$rA, R32FP:$rB)>;
-
-def FCGTf32 :
-    RRForm<0b01000011010, (outs R32C:$rT), (ins R32FP:$rA, R32FP:$rB),
-      "fcgt\t$rT, $rA, $rB", SPrecFP,
-      [(set R32C:$rT, (setugt R32FP:$rA, R32FP:$rB))]>;
-
-def : Pat<(setogt R32FP:$rA, R32FP:$rB),
-          (FCGTf32 R32FP:$rA, R32FP:$rB)>;
-
-def FCMGTf32 :
-    RRForm<0b01010011010, (outs R32C:$rT), (ins R32FP:$rA, R32FP:$rB),
-      "fcmgt\t$rT, $rA, $rB", SPrecFP,
-      [(set R32C:$rT, (setugt (fabs R32FP:$rA), (fabs R32FP:$rB)))]>;
-
-def : Pat<(setogt (fabs R32FP:$rA), (fabs R32FP:$rB)),
-          (FCMGTf32 R32FP:$rA, R32FP:$rB)>;
-
-//--------------------------------------------------------------------------
-// Single precision floating point comparisons and SETCC equivalents:
-//--------------------------------------------------------------------------
-
-def : SETCCNegCondReg<setune, R32FP, i32, XORIr32, FCEQf32>;
-def : SETCCNegCondReg<setone, R32FP, i32, XORIr32, FCEQf32>;
-
-def : SETCCBinOpReg<setuge, R32FP, ORr32, FCGTf32, FCEQf32>;
-def : SETCCBinOpReg<setoge, R32FP, ORr32, FCGTf32, FCEQf32>;
-
-def : SETCCBinOpReg<setult, R32FP, NORr32, FCGTf32, FCEQf32>;
-def : SETCCBinOpReg<setolt, R32FP, NORr32, FCGTf32, FCEQf32>;
-
-def : Pat<(setule R32FP:$rA, R32FP:$rB),
-          (XORIr32 (FCGTf32 R32FP:$rA, R32FP:$rB), 0xffffffff)>;
-def : Pat<(setole R32FP:$rA, R32FP:$rB),
-          (XORIr32 (FCGTf32 R32FP:$rA, R32FP:$rB), 0xffffffff)>;
-
-// FP Status and Control Register Write
-// Why isn't rT a don't care in the ISA?
-// Should we create a special RRForm_3 for this guy and zero out the rT?
-def FSCRWf32 :
-    RRForm_1<0b01011101110, (outs R32FP:$rT), (ins R32FP:$rA),
-      "fscrwr\t$rA", SPrecFP,
-      [/* This instruction requires an intrinsic. Note: rT is unused. */]>;
-
-// FP Status and Control Register Read
-def FSCRRf32 :
-    RRForm_2<0b01011101110, (outs R32FP:$rT), (ins),
-      "fscrrd\t$rT", SPrecFP,
-      [/* This instruction requires an intrinsic */]>;
-
-// llvm instruction space
-// How do these map onto cell instructions?
-// fdiv rA rB
-//   frest rC rB        # c = 1/b (both lines)
-//   fi rC rB rC
-//   fm rD rA rC        # d = a * 1/b
-//   fnms rB rD rB rA # b = - (d * b - a) --should == 0 in a perfect world
-//   fma rB rB rC rD            # b = b * c + d
-//                              = -(d *b -a) * c + d
-//                              = a * c - c ( a *b *c - a)
-
-// fcopysign (???)
-
-// Library calls:
-// These llvm instructions will actually map to library calls.
-// All that's needed, then, is to check that the appropriate library is
-// imported and do a brsl to the proper function name.
-// frem # fmod(x, y): x - (x/y) * y
-// (Note: fmod(double, double), fmodf(float,float)
-// fsqrt?
-// fsin?
-// fcos?
-// Unimplemented SPU instruction space
-// floating reciprocal absolute square root estimate (frsqest)
-
-// The following are probably just intrinsics
-// status and control register write
-// status and control register read
-
-//--------------------------------------
-// Floating Point Conversions
-// Signed conversions:
-def CSiFv4f32:
-    CVTIntFPForm<0b0101101110, (outs VECREG:$rT), (ins VECREG:$rA),
-      "csflt\t$rT, $rA, 0", SPrecFP,
-      [(set (v4f32 VECREG:$rT), (sint_to_fp (v4i32 VECREG:$rA)))]>;
-
-// Convert signed integer to floating point
-def CSiFf32 :
-    CVTIntFPForm<0b0101101110, (outs R32FP:$rT), (ins R32C:$rA),
-      "csflt\t$rT, $rA, 0", SPrecFP,
-      [(set R32FP:$rT, (sint_to_fp R32C:$rA))]>;
-
-// Convert unsigned into to float
-def CUiFv4f32 :
-    CVTIntFPForm<0b1101101110, (outs VECREG:$rT), (ins VECREG:$rA),
-      "cuflt\t$rT, $rA, 0", SPrecFP,
-      [(set (v4f32 VECREG:$rT), (uint_to_fp (v4i32 VECREG:$rA)))]>;
-
-def CUiFf32 :
-    CVTIntFPForm<0b1101101110, (outs R32FP:$rT), (ins R32C:$rA),
-      "cuflt\t$rT, $rA, 0", SPrecFP,
-      [(set R32FP:$rT, (uint_to_fp R32C:$rA))]>;
-
-// Convert float to unsigned int
-// Assume that scale = 0
-
-def CFUiv4f32 :
-    CVTIntFPForm<0b1101101110, (outs VECREG:$rT), (ins VECREG:$rA),
-      "cfltu\t$rT, $rA, 0", SPrecFP,
-      [(set (v4i32 VECREG:$rT), (fp_to_uint (v4f32 VECREG:$rA)))]>;
-
-def CFUif32 :
-    CVTIntFPForm<0b1101101110, (outs R32C:$rT), (ins R32FP:$rA),
-      "cfltu\t$rT, $rA, 0", SPrecFP,
-      [(set R32C:$rT, (fp_to_uint R32FP:$rA))]>;
-
-// Convert float to signed int
-// Assume that scale = 0
-
-def CFSiv4f32 :
-    CVTIntFPForm<0b1101101110, (outs VECREG:$rT), (ins VECREG:$rA),
-      "cflts\t$rT, $rA, 0", SPrecFP,
-      [(set (v4i32 VECREG:$rT), (fp_to_sint (v4f32 VECREG:$rA)))]>;
-
-def CFSif32 :
-    CVTIntFPForm<0b1101101110, (outs R32C:$rT), (ins R32FP:$rA),
-      "cflts\t$rT, $rA, 0", SPrecFP,
-      [(set R32C:$rT, (fp_to_sint R32FP:$rA))]>;
-
-//===----------------------------------------------------------------------==//
-// Single<->Double precision conversions
-//===----------------------------------------------------------------------==//
-
-// NOTE: We use "vec" name suffix here to avoid confusion (e.g. input is a
-// v4f32, output is v2f64--which goes in the name?)
-
-// Floating point extend single to double
-// NOTE: Not sure if passing in v4f32 to FESDvec is correct since it
-// operates on two double-word slots (i.e. 1st and 3rd fp numbers
-// are ignored).
-def FESDvec :
-    RRForm_1<0b00011101110, (outs VECREG:$rT), (ins VECREG:$rA),
-      "fesd\t$rT, $rA", SPrecFP,
-      [/*(set (v2f64 VECREG:$rT), (fextend (v4f32 VECREG:$rA)))*/]>;
-
-def FESDf32 :
-    RRForm_1<0b00011101110, (outs R64FP:$rT), (ins R32FP:$rA),
-      "fesd\t$rT, $rA", SPrecFP,
-      [(set R64FP:$rT, (fextend R32FP:$rA))]>;
-
-// Floating point round double to single
-//def FRDSvec :
-//    RRForm_1<0b10011101110, (outs VECREG:$rT), (ins VECREG:$rA),
-//      "frds\t$rT, $rA,", SPrecFP,
-//      [(set (v4f32 R32FP:$rT), (fround (v2f64 R64FP:$rA)))]>;
-
-def FRDSf64 :
-    RRForm_1<0b10011101110, (outs R32FP:$rT), (ins R64FP:$rA),
-      "frds\t$rT, $rA", SPrecFP,
-      [(set R32FP:$rT, (fround R64FP:$rA))]>;
-
-//ToDo include anyextend?
-
-//===----------------------------------------------------------------------==//
-// Double precision floating point instructions
-//===----------------------------------------------------------------------==//
-def FAf64 :
-    RRForm<0b00110011010, (outs R64FP:$rT), (ins R64FP:$rA, R64FP:$rB),
-      "dfa\t$rT, $rA, $rB", DPrecFP,
-      [(set R64FP:$rT, (fadd R64FP:$rA, R64FP:$rB))]>;
-
-def FAv2f64 :
-    RRForm<0b00110011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "dfa\t$rT, $rA, $rB", DPrecFP,
-      [(set (v2f64 VECREG:$rT), (fadd (v2f64 VECREG:$rA), (v2f64 VECREG:$rB)))]>;
-
-def FSf64 :
-    RRForm<0b10100011010, (outs R64FP:$rT), (ins R64FP:$rA, R64FP:$rB),
-      "dfs\t$rT, $rA, $rB", DPrecFP,
-      [(set R64FP:$rT, (fsub R64FP:$rA, R64FP:$rB))]>;
-
-def FSv2f64 :
-    RRForm<0b10100011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "dfs\t$rT, $rA, $rB", DPrecFP,
-      [(set (v2f64 VECREG:$rT),
-            (fsub (v2f64 VECREG:$rA), (v2f64 VECREG:$rB)))]>;
-
-def FMf64 :
-    RRForm<0b01100011010, (outs R64FP:$rT), (ins R64FP:$rA, R64FP:$rB),
-      "dfm\t$rT, $rA, $rB", DPrecFP,
-      [(set R64FP:$rT, (fmul R64FP:$rA, R64FP:$rB))]>;
-
-def FMv2f64:
-    RRForm<0b00100011010, (outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB),
-      "dfm\t$rT, $rA, $rB", DPrecFP,
-      [(set (v2f64 VECREG:$rT),
-            (fmul (v2f64 VECREG:$rA), (v2f64 VECREG:$rB)))]>;
-
-def FMAf64:
-    RRForm<0b00111010110, (outs R64FP:$rT),
-                          (ins R64FP:$rA, R64FP:$rB, R64FP:$rC),
-      "dfma\t$rT, $rA, $rB", DPrecFP,
-      [(set R64FP:$rT, (fadd R64FP:$rC, (fmul R64FP:$rA, R64FP:$rB)))]>,
-    RegConstraint<"$rC = $rT">,
-    NoEncode<"$rC">;
-
-def FMAv2f64:
-    RRForm<0b00111010110, (outs VECREG:$rT),
-                          (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-      "dfma\t$rT, $rA, $rB", DPrecFP,
-      [(set (v2f64 VECREG:$rT),
-            (fadd (v2f64 VECREG:$rC),
-                  (fmul (v2f64 VECREG:$rA), (v2f64 VECREG:$rB))))]>,
-    RegConstraint<"$rC = $rT">,
-    NoEncode<"$rC">;
-
-def FMSf64 :
-    RRForm<0b10111010110, (outs R64FP:$rT),
-                          (ins R64FP:$rA, R64FP:$rB, R64FP:$rC),
-      "dfms\t$rT, $rA, $rB", DPrecFP,
-      [(set R64FP:$rT, (fsub (fmul R64FP:$rA, R64FP:$rB), R64FP:$rC))]>,
-    RegConstraint<"$rC = $rT">,
-    NoEncode<"$rC">;
-
-def FMSv2f64 :
-    RRForm<0b10111010110, (outs VECREG:$rT),
-                          (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-      "dfms\t$rT, $rA, $rB", DPrecFP,
-      [(set (v2f64 VECREG:$rT),
-            (fsub (fmul (v2f64 VECREG:$rA), (v2f64 VECREG:$rB)),
-                  (v2f64 VECREG:$rC)))]>;
-
-// DFNMS: - (a * b - c)
-// - (a * b) + c => c - (a * b)
-
-class DFNMSInst<dag OOL, dag IOL, list<dag> pattern>:
-    RRForm<0b01111010110, OOL, IOL, "dfnms\t$rT, $rA, $rB",
-           DPrecFP, pattern>,
-    RegConstraint<"$rC = $rT">,
-    NoEncode<"$rC">;
-
-class DFNMSVecInst<list<dag> pattern>:
-    DFNMSInst<(outs VECREG:$rT), (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-              pattern>;
-
-class DFNMSRegInst<list<dag> pattern>:
-    DFNMSInst<(outs R64FP:$rT), (ins R64FP:$rA, R64FP:$rB, R64FP:$rC),
-             pattern>;
-
-multiclass DFMultiplySubtract
-{
-  def v2f64 : DFNMSVecInst<[(set (v2f64 VECREG:$rT), 
-                                 (fsub (v2f64 VECREG:$rC),
-                                       (fmul (v2f64 VECREG:$rA),
-                                             (v2f64 VECREG:$rB))))]>;
-
-  def f64 : DFNMSRegInst<[(set R64FP:$rT,
-                               (fsub R64FP:$rC,
-                                     (fmul R64FP:$rA, R64FP:$rB)))]>;
-}
-
-defm DFNMS : DFMultiplySubtract;
-
-// - (a * b + c)
-// - (a * b) - c
-def FNMAf64 :
-    RRForm<0b11111010110, (outs R64FP:$rT),
-                          (ins R64FP:$rA, R64FP:$rB, R64FP:$rC),
-      "dfnma\t$rT, $rA, $rB", DPrecFP,
-      [(set R64FP:$rT, (fneg (fadd R64FP:$rC, (fmul R64FP:$rA, R64FP:$rB))))]>,
-    RegConstraint<"$rC = $rT">,
-    NoEncode<"$rC">;
-
-def FNMAv2f64 :
-    RRForm<0b11111010110, (outs VECREG:$rT),
-                          (ins VECREG:$rA, VECREG:$rB, VECREG:$rC),
-      "dfnma\t$rT, $rA, $rB", DPrecFP,
-      [(set (v2f64 VECREG:$rT),
-            (fneg (fadd (v2f64 VECREG:$rC),
-                        (fmul (v2f64 VECREG:$rA),
-                              (v2f64 VECREG:$rB)))))]>,
-    RegConstraint<"$rC = $rT">,
-    NoEncode<"$rC">;
-
-//===----------------------------------------------------------------------==//
-// Floating point negation and absolute value
-//===----------------------------------------------------------------------==//
-
-def : Pat<(fneg (v4f32 VECREG:$rA)),
-          (XORfnegvec (v4f32 VECREG:$rA),
-                      (v4f32 (ILHUv4i32 0x8000)))>;
-
-def : Pat<(fneg R32FP:$rA),
-          (XORfneg32 R32FP:$rA, (ILHUr32 0x8000))>;
-
-// Floating point absolute value
-// Note: f64 fabs is custom-selected.
-
-def : Pat<(fabs R32FP:$rA),
-          (ANDfabs32 R32FP:$rA, (IOHLr32 (ILHUr32 0x7fff), 0xffff))>;
-
-def : Pat<(fabs (v4f32 VECREG:$rA)),
-          (ANDfabsvec (v4f32 VECREG:$rA),
-                      (IOHLv4i32 (ILHUv4i32 0x7fff), 0xffff))>;
-
-//===----------------------------------------------------------------------===//
-// Hint for branch instructions:
-//===----------------------------------------------------------------------===//
-def HBRA :
-    HBI16Form<0b0001001,(ins hbrtarget:$brinst, brtarget:$btarg), "hbra\t$brinst, $btarg">;
-
-//===----------------------------------------------------------------------===//
-// Execution, Load NOP (execute NOPs belong in even pipeline, load NOPs belong
-// in the odd pipeline)
-//===----------------------------------------------------------------------===//
-
-def ENOP : SPUInstr<(outs), (ins), "nop", ExecNOP> {
-  let Pattern = [];
-
-  let Inst{0-10} = 0b10000000010;
-  let Inst{11-17} = 0;
-  let Inst{18-24} = 0;
-  let Inst{25-31} = 0;
-}
-
-def LNOP : SPUInstr<(outs), (ins), "lnop", LoadNOP> {
-  let Pattern = [];
-
-  let Inst{0-10} = 0b10000000000;
-  let Inst{11-17} = 0;
-  let Inst{18-24} = 0;
-  let Inst{25-31} = 0;
-}
-
-//===----------------------------------------------------------------------===//
-// Bit conversions (type conversions between vector/packed types)
-// NOTE: Promotions are handled using the XS* instructions.
-//===----------------------------------------------------------------------===//
-def : Pat<(v16i8 (bitconvert (v8i16 VECREG:$src))), (v16i8 VECREG:$src)>;
-def : Pat<(v16i8 (bitconvert (v4i32 VECREG:$src))), (v16i8 VECREG:$src)>;
-def : Pat<(v16i8 (bitconvert (v2i64 VECREG:$src))), (v16i8 VECREG:$src)>;
-def : Pat<(v16i8 (bitconvert (v4f32 VECREG:$src))), (v16i8 VECREG:$src)>;
-def : Pat<(v16i8 (bitconvert (v2f64 VECREG:$src))), (v16i8 VECREG:$src)>;
-
-def : Pat<(v8i16 (bitconvert (v16i8 VECREG:$src))), (v8i16 VECREG:$src)>;
-def : Pat<(v8i16 (bitconvert (v4i32 VECREG:$src))), (v8i16 VECREG:$src)>;
-def : Pat<(v8i16 (bitconvert (v2i64 VECREG:$src))), (v8i16 VECREG:$src)>;
-def : Pat<(v8i16 (bitconvert (v4f32 VECREG:$src))), (v8i16 VECREG:$src)>;
-def : Pat<(v8i16 (bitconvert (v2f64 VECREG:$src))), (v8i16 VECREG:$src)>;
-
-def : Pat<(v4i32 (bitconvert (v16i8 VECREG:$src))), (v4i32 VECREG:$src)>;
-def : Pat<(v4i32 (bitconvert (v8i16 VECREG:$src))), (v4i32 VECREG:$src)>;
-def : Pat<(v4i32 (bitconvert (v2i64 VECREG:$src))), (v4i32 VECREG:$src)>;
-def : Pat<(v4i32 (bitconvert (v4f32 VECREG:$src))), (v4i32 VECREG:$src)>;
-def : Pat<(v4i32 (bitconvert (v2f64 VECREG:$src))), (v4i32 VECREG:$src)>;
-
-def : Pat<(v2i64 (bitconvert (v16i8 VECREG:$src))), (v2i64 VECREG:$src)>;
-def : Pat<(v2i64 (bitconvert (v8i16 VECREG:$src))), (v2i64 VECREG:$src)>;
-def : Pat<(v2i64 (bitconvert (v4i32 VECREG:$src))), (v2i64 VECREG:$src)>;
-def : Pat<(v2i64 (bitconvert (v4f32 VECREG:$src))), (v2i64 VECREG:$src)>;
-def : Pat<(v2i64 (bitconvert (v2f64 VECREG:$src))), (v2i64 VECREG:$src)>;
-
-def : Pat<(v4f32 (bitconvert (v16i8 VECREG:$src))), (v4f32 VECREG:$src)>;
-def : Pat<(v4f32 (bitconvert (v8i16 VECREG:$src))), (v4f32 VECREG:$src)>;
-def : Pat<(v4f32 (bitconvert (v2i64 VECREG:$src))), (v4f32 VECREG:$src)>;
-def : Pat<(v4f32 (bitconvert (v4i32 VECREG:$src))), (v4f32 VECREG:$src)>;
-def : Pat<(v4f32 (bitconvert (v2f64 VECREG:$src))), (v4f32 VECREG:$src)>;
-
-def : Pat<(v2f64 (bitconvert (v16i8 VECREG:$src))), (v2f64 VECREG:$src)>;
-def : Pat<(v2f64 (bitconvert (v8i16 VECREG:$src))), (v2f64 VECREG:$src)>;
-def : Pat<(v2f64 (bitconvert (v4i32 VECREG:$src))), (v2f64 VECREG:$src)>;
-def : Pat<(v2f64 (bitconvert (v2i64 VECREG:$src))), (v2f64 VECREG:$src)>;
-def : Pat<(v2f64 (bitconvert (v4f32 VECREG:$src))), (v2f64 VECREG:$src)>;
-
-def : Pat<(i128 (bitconvert (v16i8 VECREG:$src))),
-          (COPY_TO_REGCLASS VECREG:$src, GPRC)>;
-def : Pat<(i128 (bitconvert (v8i16 VECREG:$src))),
-          (COPY_TO_REGCLASS VECREG:$src, GPRC)>;
-def : Pat<(i128 (bitconvert (v4i32 VECREG:$src))),
-          (COPY_TO_REGCLASS VECREG:$src, GPRC)>;
-def : Pat<(i128 (bitconvert (v2i64 VECREG:$src))),
-          (COPY_TO_REGCLASS VECREG:$src, GPRC)>;
-def : Pat<(i128 (bitconvert (v4f32 VECREG:$src))),
-          (COPY_TO_REGCLASS VECREG:$src, GPRC)>;
-def : Pat<(i128 (bitconvert (v2f64 VECREG:$src))),
-          (COPY_TO_REGCLASS VECREG:$src, GPRC)>;
-
-def : Pat<(v16i8 (bitconvert (i128 GPRC:$src))),
-          (v16i8 (COPY_TO_REGCLASS GPRC:$src, VECREG))>;
-def : Pat<(v8i16 (bitconvert (i128 GPRC:$src))),
-          (v8i16 (COPY_TO_REGCLASS GPRC:$src, VECREG))>;
-def : Pat<(v4i32 (bitconvert (i128 GPRC:$src))),
-          (v4i32 (COPY_TO_REGCLASS GPRC:$src, VECREG))>;
-def : Pat<(v2i64 (bitconvert (i128 GPRC:$src))),
-          (v2i64 (COPY_TO_REGCLASS GPRC:$src, VECREG))>;
-def : Pat<(v4f32 (bitconvert (i128 GPRC:$src))),
-          (v4f32 (COPY_TO_REGCLASS GPRC:$src, VECREG))>;
-def : Pat<(v2f64 (bitconvert (i128 GPRC:$src))),
-          (v2f64 (COPY_TO_REGCLASS GPRC:$src, VECREG))>;
-
-def : Pat<(i32 (bitconvert R32FP:$rA)),
-          (COPY_TO_REGCLASS R32FP:$rA, R32C)>;
-
-def : Pat<(f32 (bitconvert R32C:$rA)),
-          (COPY_TO_REGCLASS R32C:$rA, R32FP)>;
-
-def : Pat<(i64 (bitconvert R64FP:$rA)),
-          (COPY_TO_REGCLASS R64FP:$rA, R64C)>;
-
-def : Pat<(f64 (bitconvert R64C:$rA)),
-          (COPY_TO_REGCLASS R64C:$rA, R64FP)>;
-
-
-//===----------------------------------------------------------------------===//
-// Instruction patterns:
-//===----------------------------------------------------------------------===//
-
-// General 32-bit constants:
-def : Pat<(i32 imm:$imm),
-          (IOHLr32 (ILHUr32 (HI16 imm:$imm)), (LO16 imm:$imm))>;
-
-// Single precision float constants:
-def : Pat<(f32 fpimm:$imm),
-          (IOHLf32 (ILHUf32 (HI16_f32 fpimm:$imm)), (LO16_f32 fpimm:$imm))>;
-
-// General constant 32-bit vectors
-def : Pat<(v4i32 v4i32Imm:$imm),
-          (IOHLv4i32 (v4i32 (ILHUv4i32 (HI16_vec v4i32Imm:$imm))),
-                     (LO16_vec v4i32Imm:$imm))>;
-
-// 8-bit constants
-def : Pat<(i8 imm:$imm),
-          (ILHr8 imm:$imm)>;
-
-//===----------------------------------------------------------------------===//
-// Zero/Any/Sign extensions
-//===----------------------------------------------------------------------===//
-
-// sext 8->32: Sign extend bytes to words
-def : Pat<(sext_inreg R32C:$rSrc, i8),
-          (XSHWr32 (XSBHr32 R32C:$rSrc))>;
-
-def : Pat<(i32 (sext R8C:$rSrc)),
-          (XSHWr16 (XSBHr8 R8C:$rSrc))>;
-
-// sext 8->64: Sign extend bytes to double word
-def : Pat<(sext_inreg R64C:$rSrc, i8),
-          (XSWDr64_inreg (XSHWr64 (XSBHr64 R64C:$rSrc)))>;
-          
-def : Pat<(i64 (sext R8C:$rSrc)),
-          (XSWDr64 (XSHWr16 (XSBHr8 R8C:$rSrc)))>;
-
-// zext 8->16: Zero extend bytes to halfwords
-def : Pat<(i16 (zext R8C:$rSrc)),
-          (ANDHIi8i16 R8C:$rSrc, 0xff)>;
-
-// zext 8->32: Zero extend bytes to words
-def : Pat<(i32 (zext R8C:$rSrc)),
-          (ANDIi8i32 R8C:$rSrc, 0xff)>;
-
-// zext 8->64: Zero extend bytes to double words
-def : Pat<(i64 (zext R8C:$rSrc)),
-          (COPY_TO_REGCLASS (SELBv4i32 (ROTQMBYv4i32
-                                    (COPY_TO_REGCLASS 
-                                       (ANDIi8i32 R8C:$rSrc,0xff), VECREG),
-                                    0x4),
-                                  (ILv4i32 0x0),
-                                  (FSMBIv4i32 0x0f0f)), R64C)>;
-
-// anyext 8->16: Extend 8->16 bits, irrespective of sign, preserves high bits
-def : Pat<(i16 (anyext R8C:$rSrc)),
-          (ORHIi8i16 R8C:$rSrc, 0)>;
-
-// anyext 8->32: Extend 8->32 bits, irrespective of sign, preserves high bits
-def : Pat<(i32 (anyext R8C:$rSrc)),
-          (COPY_TO_REGCLASS R8C:$rSrc, R32C)>;
-
-// sext 16->64: Sign extend halfword to double word
-def : Pat<(sext_inreg R64C:$rSrc, i16),
-          (XSWDr64_inreg (XSHWr64 R64C:$rSrc))>;
-          
-def : Pat<(sext R16C:$rSrc),
-          (XSWDr64 (XSHWr16 R16C:$rSrc))>;
-
-// zext 16->32: Zero extend halfwords to words
-def : Pat<(i32 (zext R16C:$rSrc)),
-          (ANDi16i32 R16C:$rSrc, (ILAr32 0xffff))>;
-
-def : Pat<(i32 (zext (and R16C:$rSrc, 0xf))),
-          (ANDIi16i32 R16C:$rSrc, 0xf)>;
-
-def : Pat<(i32 (zext (and R16C:$rSrc, 0xff))),
-          (ANDIi16i32 R16C:$rSrc, 0xff)>;
-
-def : Pat<(i32 (zext (and R16C:$rSrc, 0xfff))),
-          (ANDIi16i32 R16C:$rSrc, 0xfff)>;
-
-// anyext 16->32: Extend 16->32 bits, irrespective of sign
-def : Pat<(i32 (anyext R16C:$rSrc)),
-          (COPY_TO_REGCLASS R16C:$rSrc, R32C)>;
-
-//===----------------------------------------------------------------------===//
-// Truncates:
-// These truncates are for the SPU's supported types (i8, i16, i32). i64 and
-// above are custom lowered.
-//===----------------------------------------------------------------------===//
-
-def : Pat<(i8 (trunc GPRC:$src)),
-          (COPY_TO_REGCLASS
-            (SHUFBgprc GPRC:$src, GPRC:$src,
-                       (IOHLv4i32 (ILHUv4i32 0x0f0f), 0x0f0f)), R8C)>;
-
-def : Pat<(i8 (trunc R64C:$src)),
-          (COPY_TO_REGCLASS
-            (SHUFBv2i64_m32
-              (COPY_TO_REGCLASS R64C:$src, VECREG),
-              (COPY_TO_REGCLASS R64C:$src, VECREG),
-              (IOHLv4i32 (ILHUv4i32 0x0707), 0x0707)), R8C)>;
-
-def : Pat<(i8 (trunc R32C:$src)),
-          (COPY_TO_REGCLASS
-            (SHUFBv4i32_m32
-               (COPY_TO_REGCLASS R32C:$src, VECREG),
-               (COPY_TO_REGCLASS R32C:$src, VECREG),
-               (IOHLv4i32 (ILHUv4i32 0x0303), 0x0303)), R8C)>;
-
-def : Pat<(i8 (trunc R16C:$src)),
-          (COPY_TO_REGCLASS
-            (SHUFBv4i32_m32
-               (COPY_TO_REGCLASS R16C:$src, VECREG),
-               (COPY_TO_REGCLASS R16C:$src, VECREG),
-               (IOHLv4i32 (ILHUv4i32 0x0303), 0x0303)), R8C)>;
-
-def : Pat<(i16 (trunc GPRC:$src)),
-          (COPY_TO_REGCLASS
-            (SHUFBgprc GPRC:$src, GPRC:$src,
-                       (IOHLv4i32 (ILHUv4i32 0x0e0f), 0x0e0f)), R16C)>;
-
-def : Pat<(i16 (trunc R64C:$src)),
-          (COPY_TO_REGCLASS
-            (SHUFBv2i64_m32
-              (COPY_TO_REGCLASS R64C:$src, VECREG),
-              (COPY_TO_REGCLASS R64C:$src, VECREG),
-              (IOHLv4i32 (ILHUv4i32 0x0607), 0x0607)), R16C)>;
-
-def : Pat<(i16 (trunc R32C:$src)),
-          (COPY_TO_REGCLASS
-            (SHUFBv4i32_m32
-               (COPY_TO_REGCLASS R32C:$src, VECREG),
-               (COPY_TO_REGCLASS R32C:$src, VECREG),
-               (IOHLv4i32 (ILHUv4i32 0x0203), 0x0203)), R16C)>;
-
-def : Pat<(i32 (trunc GPRC:$src)),
-          (COPY_TO_REGCLASS
-            (SHUFBgprc GPRC:$src, GPRC:$src,
-                       (IOHLv4i32 (ILHUv4i32 0x0c0d), 0x0e0f)), R32C)>;
-
-def : Pat<(i32 (trunc R64C:$src)),
-          (COPY_TO_REGCLASS
-            (SHUFBv2i64_m32
-              (COPY_TO_REGCLASS R64C:$src, VECREG),
-              (COPY_TO_REGCLASS R64C:$src, VECREG),
-              (IOHLv4i32 (ILHUv4i32 0x0405), 0x0607)), R32C)>;
-
-//===----------------------------------------------------------------------===//
-// Address generation: SPU, like PPC, has to split addresses into high and
-// low parts in order to load them into a register.
-//===----------------------------------------------------------------------===//
-
-def : Pat<(SPUaform tglobaladdr:$in, 0),  (ILAlsa tglobaladdr:$in)>;
-def : Pat<(SPUaform texternalsym:$in, 0), (ILAlsa texternalsym:$in)>;
-def : Pat<(SPUaform tjumptable:$in, 0),   (ILAlsa tjumptable:$in)>;
-def : Pat<(SPUaform tconstpool:$in, 0),   (ILAlsa  tconstpool:$in)>;
-
-def : Pat<(SPUindirect (SPUhi tglobaladdr:$in, 0),
-                       (SPUlo tglobaladdr:$in, 0)),
-          (IOHLlo (ILHUhi tglobaladdr:$in), tglobaladdr:$in)>;
-
-def : Pat<(SPUindirect (SPUhi texternalsym:$in, 0),
-                       (SPUlo texternalsym:$in, 0)),
-          (IOHLlo (ILHUhi texternalsym:$in), texternalsym:$in)>;
-
-def : Pat<(SPUindirect (SPUhi tjumptable:$in, 0),
-                       (SPUlo tjumptable:$in, 0)),
-          (IOHLlo (ILHUhi tjumptable:$in), tjumptable:$in)>;
-
-def : Pat<(SPUindirect (SPUhi tconstpool:$in, 0),
-                       (SPUlo tconstpool:$in, 0)),
-          (IOHLlo (ILHUhi tconstpool:$in), tconstpool:$in)>;
-
-def : Pat<(add (SPUhi tglobaladdr:$in, 0), (SPUlo tglobaladdr:$in, 0)),
-          (IOHLlo (ILHUhi tglobaladdr:$in), tglobaladdr:$in)>;
-
-def : Pat<(add (SPUhi texternalsym:$in, 0), (SPUlo texternalsym:$in, 0)),
-          (IOHLlo (ILHUhi texternalsym:$in), texternalsym:$in)>;
-
-def : Pat<(add (SPUhi tjumptable:$in, 0), (SPUlo tjumptable:$in, 0)),
-          (IOHLlo (ILHUhi tjumptable:$in), tjumptable:$in)>;
-
-def : Pat<(add (SPUhi tconstpool:$in, 0), (SPUlo tconstpool:$in, 0)),
-          (IOHLlo (ILHUhi tconstpool:$in), tconstpool:$in)>;
-
-// Intrinsics:
-include "CellSDKIntrinsics.td"
-// Various math operator instruction sequences
-include "SPUMathInstr.td"
-// 64-bit "instructions"/support
-include "SPU64InstrInfo.td"
-// 128-bit "instructions"/support
-include "SPU128InstrInfo.td"
diff --git a/lib/Target/CellSPU/SPUMachineFunction.h b/lib/Target/CellSPU/SPUMachineFunction.h
deleted file mode 100644
index 399684bb0887..000000000000
--- a/lib/Target/CellSPU/SPUMachineFunction.h
+++ /dev/null
@@ -1,50 +0,0 @@
-//===-- SPUMachineFunctionInfo.h - Private data used for CellSPU --*- C++ -*-=//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file declares the IBM Cell SPU specific subclass of MachineFunctionInfo.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef SPU_MACHINE_FUNCTION_INFO_H
-#define SPU_MACHINE_FUNCTION_INFO_H
-
-#include "llvm/CodeGen/MachineFunction.h"
-
-namespace llvm {
-
-/// SPUFunctionInfo - Cell SPU target-specific information for each
-/// MachineFunction
-class SPUFunctionInfo : public MachineFunctionInfo {
-  virtual void anchor();
-
-  /// UsesLR - Indicates whether LR is used in the current function.
-  ///
-  bool UsesLR;
-
-  // VarArgsFrameIndex - FrameIndex for start of varargs area.
-  int VarArgsFrameIndex;
-
-public:
-  SPUFunctionInfo(MachineFunction& MF) 
-  : UsesLR(false),
-    VarArgsFrameIndex(0)
-  {}
-
-  void setUsesLR(bool U) { UsesLR = U; }
-  bool usesLR()          { return UsesLR; }
-
-  int getVarArgsFrameIndex() const { return VarArgsFrameIndex; }
-  void setVarArgsFrameIndex(int Index) { VarArgsFrameIndex = Index; }
-};
-
-} // end of namespace llvm
-
-
-#endif
-
diff --git a/lib/Target/CellSPU/SPUMathInstr.td b/lib/Target/CellSPU/SPUMathInstr.td
deleted file mode 100644
index 9a5c3976afbe..000000000000
--- a/lib/Target/CellSPU/SPUMathInstr.td
+++ /dev/null
@@ -1,97 +0,0 @@
-//===-- SPUMathInst.td - Cell SPU math operations ---------*- tablegen -*--===//
-//
-//                     Cell SPU math operations
-//
-// This target description file contains instruction sequences for various
-// math operations, such as vector multiplies, i32 multiply, etc., for the
-// SPU's i32, i16 i8 and corresponding vector types.
-//
-// Any resemblance to libsimdmath or the Cell SDK simdmath library is
-// purely and completely coincidental.
-//===----------------------------------------------------------------------===//
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// v16i8 multiply instruction sequence:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-def : Pat<(mul (v16i8 VECREG:$rA), (v16i8 VECREG:$rB)),
-          (ORv4i32
-           (ANDv4i32
-            (SELBv4i32 (MPYv8i16 VECREG:$rA, VECREG:$rB),
-                       (SHLHIv8i16 (MPYv8i16 (ROTMAHIv8i16 VECREG:$rA, 8),
-                                             (ROTMAHIv8i16 VECREG:$rB, 8)), 8),
-                       (FSMBIv8i16 0x2222)),
-            (ILAv4i32 0x0000ffff)),
-           (SHLIv4i32
-            (SELBv4i32 (MPYv8i16 (ROTMAIv4i32_i32 VECREG:$rA, 16),
-                                 (ROTMAIv4i32_i32 VECREG:$rB, 16)),
-                       (SHLHIv8i16 (MPYv8i16 (ROTMAIv4i32_i32 VECREG:$rA, 8),
-                                             (ROTMAIv4i32_i32 VECREG:$rB, 8)), 8),
-                       (FSMBIv8i16 0x2222)), 16))>;
-                        
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// v8i16 multiply instruction sequence:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-def : Pat<(mul (v8i16 VECREG:$rA), (v8i16 VECREG:$rB)),
-          (SELBv8i16 (MPYv8i16 VECREG:$rA, VECREG:$rB),
-                     (SHLIv4i32 (MPYHHv8i16 VECREG:$rA, VECREG:$rB), 16),
-                     (FSMBIv8i16 0xcccc))>;
-                 
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// v4i32, i32 multiply instruction sequence:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-def MPYv4i32:
-  Pat<(mul (v4i32 VECREG:$rA), (v4i32 VECREG:$rB)),
-      (Av4i32
-        (v4i32 (Av4i32 (v4i32 (MPYHv4i32 VECREG:$rA, VECREG:$rB)),
-                       (v4i32 (MPYHv4i32 VECREG:$rB, VECREG:$rA)))),
-        (v4i32 (MPYUv4i32 VECREG:$rA, VECREG:$rB)))>;
-
-def MPYi32:
-  Pat<(mul R32C:$rA, R32C:$rB),
-      (Ar32
-        (Ar32 (MPYHr32 R32C:$rA, R32C:$rB),
-              (MPYHr32 R32C:$rB, R32C:$rA)),
-        (MPYUr32 R32C:$rA, R32C:$rB))>;
-
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-// f32, v4f32 divide instruction sequence:
-//-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
-
-// Reciprocal estimate and interpolation
-def Interpf32: CodeFrag<(FIf32 R32FP:$rB, (FRESTf32 R32FP:$rB))>;
-// Division estimate
-def DivEstf32: CodeFrag<(FMf32 R32FP:$rA, Interpf32.Fragment)>;
-// Newton-Raphson iteration
-def NRaphf32: CodeFrag<(FMAf32 (FNMSf32 DivEstf32.Fragment, R32FP:$rB, R32FP:$rA),
-                               Interpf32.Fragment,
-                               DivEstf32.Fragment)>;
-// Epsilon addition
-def Epsilonf32: CodeFrag<(AIf32 NRaphf32.Fragment, 1)>;
-
-def : Pat<(fdiv R32FP:$rA, R32FP:$rB),
-          (SELBf32_cond NRaphf32.Fragment,
-                        Epsilonf32.Fragment,
-                        (CGTIf32 (FNMSf32 R32FP:$rB, Epsilonf32.Fragment, R32FP:$rA), -1))>;
-
-// Reciprocal estimate and interpolation
-def Interpv4f32: CodeFrag<(FIv4f32 (v4f32 VECREG:$rB), (FRESTv4f32 (v4f32 VECREG:$rB)))>;
-// Division estimate
-def DivEstv4f32: CodeFrag<(FMv4f32 (v4f32 VECREG:$rA), Interpv4f32.Fragment)>;
-// Newton-Raphson iteration
-def NRaphv4f32: CodeFrag<(FMAv4f32 (FNMSv4f32 DivEstv4f32.Fragment,
-                                              (v4f32 VECREG:$rB),
-                                              (v4f32 VECREG:$rA)),
-                                   Interpv4f32.Fragment,
-                                   DivEstv4f32.Fragment)>;
-// Epsilon addition
-def Epsilonv4f32: CodeFrag<(AIv4f32 NRaphv4f32.Fragment, 1)>;
-
-def : Pat<(fdiv (v4f32 VECREG:$rA), (v4f32 VECREG:$rB)),
-          (SELBv4f32_cond NRaphv4f32.Fragment,
-                        Epsilonv4f32.Fragment,
-                        (CGTIv4f32 (FNMSv4f32 (v4f32 VECREG:$rB),
-                                              Epsilonv4f32.Fragment,
-                                              (v4f32 VECREG:$rA)), -1))>;
diff --git a/lib/Target/CellSPU/SPUNodes.td b/lib/Target/CellSPU/SPUNodes.td
deleted file mode 100644
index a47e9ef0167c..000000000000
--- a/lib/Target/CellSPU/SPUNodes.td
+++ /dev/null
@@ -1,159 +0,0 @@
-//=== SPUNodes.td - Specialized SelectionDAG nodes by CellSPU -*- tablegen -*-//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// Type profiles and SelectionDAG nodes used by CellSPU
-//
-//===----------------------------------------------------------------------===//
-
-// Type profile for a call sequence
-def SDT_SPUCallSeq : SDTypeProfile<0, 1, [ SDTCisVT<0, i32> ]>;
-
-// SPU_GenControl: Type profile for generating control words for insertions
-def SPU_GenControl : SDTypeProfile<1, 1, []>;
-def SPUshufmask    : SDNode<"SPUISD::SHUFFLE_MASK", SPU_GenControl, []>;
-
-def callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_SPUCallSeq,
-                           [SDNPHasChain, SDNPOutGlue]>;
-def callseq_end   : SDNode<"ISD::CALLSEQ_END",   SDT_SPUCallSeq,
-                           [SDNPHasChain, SDNPInGlue, SDNPOutGlue]>;
-//===----------------------------------------------------------------------===//
-// Operand constraints:
-//===----------------------------------------------------------------------===//
-
-def SDT_SPUCall   : SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>;
-def SPUcall       : SDNode<"SPUISD::CALL", SDT_SPUCall,
-                           [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
-                            SDNPVariadic]>;
-
-// Operand type constraints for vector shuffle/permute operations
-def SDT_SPUshuffle   : SDTypeProfile<1, 3, [
-  SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>
-]>;
-
-// Vector binary operator type constraints (needs a further constraint to
-// ensure that operand 0 is a vector...):
-
-def SPUVecBinop: SDTypeProfile<1, 2, [
-  SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>
-]>;
-
-// Trinary operators, e.g., addx, carry generate
-def SPUIntTrinaryOp : SDTypeProfile<1, 3, [
-  SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisSameAs<0, 3>, SDTCisInt<0>
-]>;
-
-// SELECT_MASK type constraints: There are several variations for the various
-// vector types (this avoids having to bit_convert all over the place.)
-def SPUselmask_type: SDTypeProfile<1, 1, [
-  SDTCisInt<1>
-]>;
-
-// SELB type constraints:
-def SPUselb_type: SDTypeProfile<1, 3, [
-  SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>, SDTCisSameAs<0, 3> ]>;
-
-// SPU Vector shift pseudo-instruction type constraints
-def SPUvecshift_type: SDTypeProfile<1, 2, [
-  SDTCisSameAs<0, 1>, SDTCisInt<2>]>;
-
-// "marker" type for i64 operators that need a shuffle mask
-// (i.e., uses cg or bg or another instruction that needs to
-// use shufb to get things in the right place.)
-// Op0: The result
-// Op1, 2: LHS, RHS
-// Op3: Carry-generate shuffle mask
-
-def SPUmarker_type : SDTypeProfile<1, 3, [
-  SDTCisInt<0>, SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2> ]>;
-
-//===----------------------------------------------------------------------===//
-// Synthetic/pseudo-instructions
-//===----------------------------------------------------------------------===//
-
-// SPU CNTB:
-def SPUcntb : SDNode<"SPUISD::CNTB", SDTIntUnaryOp>;
-
-// SPU vector shuffle node, matched by the SPUISD::SHUFB enum (see
-// SPUISelLowering.h):
-def SPUshuffle: SDNode<"SPUISD::SHUFB", SDT_SPUshuffle, []>;
-
-// Vector shifts (ISD::SHL,SRL,SRA are for _integers_ only):
-def SPUvec_shl: SDNode<"ISD::SHL", SPUvecshift_type, []>;
-def SPUvec_srl: SDNode<"ISD::SRL", SPUvecshift_type, []>;
-def SPUvec_sra: SDNode<"ISD::SRA", SPUvecshift_type, []>;
-
-def SPUvec_rotl: SDNode<"SPUISD::VEC_ROTL", SPUvecshift_type, []>;
-def SPUvec_rotr: SDNode<"SPUISD::VEC_ROTR", SPUvecshift_type, []>;
-
-// Vector rotate left, bits shifted out of the left are rotated in on the right
-def SPUrotbytes_left: SDNode<"SPUISD::ROTBYTES_LEFT",
-                             SPUvecshift_type, []>;
-
-// Vector rotate left by bytes, but the count is given in bits and the SPU
-// internally converts it to bytes (saves an instruction to mask off lower
-// three bits)
-def SPUrotbytes_left_bits : SDNode<"SPUISD::ROTBYTES_LEFT_BITS",
-                                   SPUvecshift_type>;
-
-// Shift entire quad left by bytes/bits. Zeros are shifted in on the right
-// SHL_BITS the same as SHL for i128, but ISD::SHL is not implemented for i128
-def SPUshlquad_l_bytes: SDNode<"SPUISD::SHL_BYTES", SPUvecshift_type, []>;
-def SPUshlquad_l_bits: SDNode<"SPUISD::SHL_BITS", SPUvecshift_type, []>;
-def SPUsrl_bytes: SDNode<"SPUISD::SRL_BYTES", SPUvecshift_type, []>;
-
-// SPU form select mask for bytes, immediate
-def SPUselmask: SDNode<"SPUISD::SELECT_MASK", SPUselmask_type, []>;
-
-// SPU select bits instruction
-def SPUselb: SDNode<"SPUISD::SELB", SPUselb_type, []>;
-
-def SDTprefslot2vec: SDTypeProfile<1, 1, []>;
-def SPUprefslot2vec: SDNode<"SPUISD::PREFSLOT2VEC", SDTprefslot2vec, []>;
-
-def SPU_vec_demote   : SDTypeProfile<1, 1, []>;
-def SPUvec2prefslot: SDNode<"SPUISD::VEC2PREFSLOT", SPU_vec_demote, []>;
-
-// Address high and low components, used for [r+r] type addressing
-def SPUhi : SDNode<"SPUISD::Hi", SDTIntBinOp, []>;
-def SPUlo : SDNode<"SPUISD::Lo", SDTIntBinOp, []>;
-
-// PC-relative address
-def SPUpcrel : SDNode<"SPUISD::PCRelAddr", SDTIntBinOp, []>;
-
-// A-Form local store addresses
-def SPUaform : SDNode<"SPUISD::AFormAddr", SDTIntBinOp, []>;
-
-// Indirect [D-Form "imm($reg)" and X-Form "$reg($reg)"] addresses
-def SPUindirect : SDNode<"SPUISD::IndirectAddr", SDTIntBinOp, []>;
-
-// i64 markers: supplies extra operands used to generate the i64 operator
-// instruction sequences
-def SPUadd64 : SDNode<"SPUISD::ADD64_MARKER", SPUmarker_type, []>;
-def SPUsub64 : SDNode<"SPUISD::SUB64_MARKER", SPUmarker_type, []>;
-def SPUmul64 : SDNode<"SPUISD::MUL64_MARKER", SPUmarker_type, []>;
-
-//===----------------------------------------------------------------------===//
-// Constraints: (taken from PPCInstrInfo.td)
-//===----------------------------------------------------------------------===//
-
-class RegConstraint<string C> {
-  string Constraints = C;
-}
-
-class NoEncode<string E> {
-  string DisableEncoding = E;
-}
-
-//===----------------------------------------------------------------------===//
-// Return (flag isn't quite what it means: the operations are flagged so that
-// instruction scheduling doesn't disassociate them.)
-//===----------------------------------------------------------------------===//
-
-def retflag     : SDNode<"SPUISD::RET_FLAG", SDTNone,
-                         [SDNPHasChain, SDNPOptInGlue]>;
diff --git a/lib/Target/CellSPU/SPUNopFiller.cpp b/lib/Target/CellSPU/SPUNopFiller.cpp
deleted file mode 100644
index 7c58041e3b84..000000000000
--- a/lib/Target/CellSPU/SPUNopFiller.cpp
+++ /dev/null
@@ -1,153 +0,0 @@
-//===-- SPUNopFiller.cpp - Add nops/lnops to align the pipelines ----------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// The final pass just before assembly printing. This pass is the last
-// checkpoint where nops and lnops are added to the instruction stream to 
-// satisfy the dual issue requirements. The actual dual issue scheduling is 
-// done (TODO: nowhere, currently)
-//
-//===----------------------------------------------------------------------===//
-
-#include "SPU.h"
-#include "SPUTargetMachine.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/raw_ostream.h"
-
-using namespace llvm;
-
-namespace {
-  struct SPUNopFiller : public MachineFunctionPass {
-
-    TargetMachine &TM;
-    const TargetInstrInfo *TII;
-    const InstrItineraryData *IID;
-    bool isEvenPlace;  // the instruction slot (mem address) at hand is even/odd
-
-    static char ID;
-    SPUNopFiller(TargetMachine &tm) 
-      : MachineFunctionPass(ID), TM(tm), TII(tm.getInstrInfo()), 
-        IID(tm.getInstrItineraryData()) 
-    {
-      DEBUG( dbgs() << "********** SPU Nop filler **********\n" ; );
-    }
-
-    virtual const char *getPassName() const {
-      return "SPU nop/lnop Filler";
-    }
-
-    void runOnMachineBasicBlock(MachineBasicBlock &MBB);
-
-    bool runOnMachineFunction(MachineFunction &F) {
-      isEvenPlace = true; //all functions get an .align 3 directive at start 
-      for (MachineFunction::iterator FI = F.begin(), FE = F.end();
-           FI != FE; ++FI)
-        runOnMachineBasicBlock(*FI);
-      return true; //never-ever do any more modifications, just print it!
-    }
-
-    typedef enum { none   = 0, // no more instructions in this function / BB
-                   pseudo = 1, // this does not get executed
-                   even   = 2, 
-                   odd    = 3 } SPUOpPlace;
-    SPUOpPlace getOpPlacement( MachineInstr &instr );
-
-  };
-  char SPUNopFiller::ID = 0;
-
-} 
-
-// Fill a BasicBlock to alignment. 
-// In the assebly we align the functions to 'even' adresses, but
-// basic blocks have an implicit alignmnet. We hereby define 
-// basic blocks to have the same, even, alignment.
-void SPUNopFiller::
-runOnMachineBasicBlock(MachineBasicBlock &MBB) 
-{
-  assert( isEvenPlace && "basic block start from odd address");
-  for (MachineBasicBlock::iterator I = MBB.begin(); I != MBB.end(); ++I)
-  {
-    SPUOpPlace this_optype, next_optype;
-    MachineBasicBlock::iterator J = I;
-    J++;
-
-    this_optype = getOpPlacement( *I );
-    next_optype = none;
-    while (J!=MBB.end()){
-      next_optype = getOpPlacement( *J );
-      ++J;
-      if (next_optype != pseudo ) 
-        break;
-    }
-
-    // padd: odd(wrong), even(wrong), ...
-    // to:   nop(corr), odd(corr), even(corr)...
-    if( isEvenPlace && this_optype == odd && next_optype == even ) {
-      DEBUG( dbgs() <<"Adding NOP before: "; );
-      DEBUG( I->dump(); );
-      BuildMI(MBB, I, I->getDebugLoc(), TII->get(SPU::ENOP));
-      isEvenPlace=false;
-    }
-    
-    // padd: even(wrong), odd(wrong), ...
-    // to:   lnop(corr), even(corr), odd(corr)...
-    else if ( !isEvenPlace && this_optype == even && next_optype == odd){
-      DEBUG( dbgs() <<"Adding LNOP before: "; );
-      DEBUG( I->dump(); );
-      BuildMI(MBB, I, I->getDebugLoc(), TII->get(SPU::LNOP));
-      isEvenPlace=true;
-    }
-      
-    // now go to next mem slot
-    if( this_optype != pseudo )
-      isEvenPlace = !isEvenPlace;    
-
-  }
-
-  // padd basicblock end
-  if( !isEvenPlace ){
-    MachineBasicBlock::iterator J = MBB.end();
-    J--;
-    if (getOpPlacement( *J ) == odd) {
-      DEBUG( dbgs() <<"Padding basic block with NOP\n"; );
-      BuildMI(MBB, J, J->getDebugLoc(), TII->get(SPU::ENOP));
-    }  
-    else {
-      J++;
-      DEBUG( dbgs() <<"Padding basic block with LNOP\n"; );
-      BuildMI(MBB, J, DebugLoc(), TII->get(SPU::LNOP));
-    }
-    isEvenPlace=true;
-  }
-}
-
-FunctionPass *llvm::createSPUNopFillerPass(SPUTargetMachine &tm) {
-  return new SPUNopFiller(tm);
-}
-
-// Figure out if 'instr' is executed in the even or odd pipeline
-SPUNopFiller::SPUOpPlace 
-SPUNopFiller::getOpPlacement( MachineInstr &instr ) {
-  int sc = instr.getDesc().getSchedClass();
-  const InstrStage *stage = IID->beginStage(sc);
-  unsigned FUs = stage->getUnits();
-  SPUOpPlace retval;
-
-  switch( FUs ) {
-    case 0: retval = pseudo; break;
-    case 1: retval = odd;    break;
-    case 2: retval = even;   break;
-    default: retval= pseudo; 
-             assert( false && "got unknown FuncUnit\n");
-             break;
-  };
-  return retval;
-}
diff --git a/lib/Target/CellSPU/SPUOperands.td b/lib/Target/CellSPU/SPUOperands.td
deleted file mode 100644
index 6f8deef5530f..000000000000
--- a/lib/Target/CellSPU/SPUOperands.td
+++ /dev/null
@@ -1,664 +0,0 @@
-//===-- SPUOperands.td - Cell SPU Instruction Operands -----*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-// Cell SPU Instruction Operands:
-//===----------------------------------------------------------------------===//
-
-// TO_IMM32 - Convert an i8/i16 to i32.
-def TO_IMM32 : SDNodeXForm<imm, [{
-  return getI32Imm(N->getZExtValue());
-}]>;
-
-// TO_IMM16 - Convert an i8/i32 to i16.
-def TO_IMM16 : SDNodeXForm<imm, [{
-  return CurDAG->getTargetConstant(N->getZExtValue(), MVT::i16);
-}]>;
-
-
-def LO16 : SDNodeXForm<imm, [{
-  unsigned val = N->getZExtValue();
-  // Transformation function: get the low 16 bits.
-  return getI32Imm(val & 0xffff);
-}]>;
-
-def LO16_vec : SDNodeXForm<scalar_to_vector, [{
-  SDValue OpVal(0, 0);
-
-  // Transformation function: get the low 16 bit immediate from a build_vector
-  // node.
-  assert(N->getOpcode() == ISD::BUILD_VECTOR
-         && "LO16_vec got something other than a BUILD_VECTOR");
-
-  // Get first constant operand...
-  for (unsigned i = 0, e = N->getNumOperands();
-       OpVal.getNode() == 0 && i != e; ++i) {
-    if (N->getOperand(i).getOpcode() == ISD::UNDEF) continue;
-    if (OpVal.getNode() == 0)
-      OpVal = N->getOperand(i);
-  }
-  
-  assert(OpVal.getNode() != 0 && "LO16_vec did not locate a <defined> node");
-  ConstantSDNode *CN = cast<ConstantSDNode>(OpVal);
-  return getI32Imm((unsigned)CN->getZExtValue() & 0xffff);
-}]>;
-
-// Transform an immediate, returning the high 16 bits shifted down:
-def HI16 : SDNodeXForm<imm, [{
-  return getI32Imm((unsigned)N->getZExtValue() >> 16);
-}]>;
-
-// Transformation function: shift the high 16 bit immediate from a build_vector
-// node into the low 16 bits, and return a 16-bit constant.
-def HI16_vec : SDNodeXForm<scalar_to_vector, [{
-  SDValue OpVal(0, 0);
-
-  assert(N->getOpcode() == ISD::BUILD_VECTOR
-         && "HI16_vec got something other than a BUILD_VECTOR");
-  
-  // Get first constant operand...
-  for (unsigned i = 0, e = N->getNumOperands();
-       OpVal.getNode() == 0 && i != e; ++i) {
-    if (N->getOperand(i).getOpcode() == ISD::UNDEF) continue;
-    if (OpVal.getNode() == 0)
-      OpVal = N->getOperand(i);
-  }
-  
-  assert(OpVal.getNode() != 0 && "HI16_vec did not locate a <defined> node");
-  ConstantSDNode *CN = cast<ConstantSDNode>(OpVal);
-  return getI32Imm((unsigned)CN->getZExtValue() >> 16);
-}]>;
-
-// simm7 predicate - True if the immediate fits in an 7-bit signed
-// field.
-def simm7: PatLeaf<(imm), [{
-  int sextVal = int(N->getSExtValue());
-  return (sextVal >= -64 && sextVal <= 63);
-}]>;
-
-// uimm7 predicate - True if the immediate fits in an 7-bit unsigned
-// field.
-def uimm7: PatLeaf<(imm), [{
-  return (N->getZExtValue() <= 0x7f);
-}]>;
-
-// immSExt8 predicate - True if the immediate fits in an 8-bit sign extended
-// field.
-def immSExt8  : PatLeaf<(imm), [{
-  int Value = int(N->getSExtValue());
-  return (Value >= -(1 << 8) && Value <= (1 << 8) - 1);
-}]>;
-
-// immU8: immediate, unsigned 8-bit quantity
-def immU8 : PatLeaf<(imm), [{
-  return (N->getZExtValue() <= 0xff);
-}]>;
-
-// i32ImmSExt10 predicate - True if the i32 immediate fits in a 10-bit sign
-// extended field.  Used by RI10Form instructions like 'ldq'.
-def i32ImmSExt10  : PatLeaf<(imm), [{
-  return isI32IntS10Immediate(N);
-}]>;
-
-// i32ImmUns10 predicate - True if the i32 immediate fits in a 10-bit unsigned
-// field.  Used by RI10Form instructions like 'ldq'.
-def i32ImmUns10  : PatLeaf<(imm), [{
-  return isI32IntU10Immediate(N);
-}]>;
-
-// i16ImmSExt10 predicate - True if the i16 immediate fits in a 10-bit sign
-// extended field.  Used by RI10Form instructions like 'ldq'.
-def i16ImmSExt10  : PatLeaf<(imm), [{
-  return isI16IntS10Immediate(N);
-}]>;
-
-// i16ImmUns10 predicate - True if the i16 immediate fits into a 10-bit unsigned
-// value. Used by RI10Form instructions.
-def i16ImmUns10 : PatLeaf<(imm), [{
-  return isI16IntU10Immediate(N);
-}]>;
-
-def immSExt16  : PatLeaf<(imm), [{
-  // immSExt16 predicate - True if the immediate fits in a 16-bit sign extended
-  // field.
-  short Ignored;
-  return isIntS16Immediate(N, Ignored);
-}]>;
-
-def immZExt16  : PatLeaf<(imm), [{
-  // immZExt16 predicate - True if the immediate fits in a 16-bit zero extended
-  // field.
-  return (uint64_t)N->getZExtValue() == (unsigned short)N->getZExtValue();
-}], LO16>;
-
-def immU16 : PatLeaf<(imm), [{
-  // immU16 predicate- True if the immediate fits into a 16-bit unsigned field.
-  return (uint64_t)N->getZExtValue() == (N->getZExtValue() & 0xffff);
-}]>;
-
-def imm18  : PatLeaf<(imm), [{
-  // imm18 predicate: True if the immediate fits into an 18-bit unsigned field.
-  int Value = (int) N->getZExtValue();
-  return isUInt<18>(Value); 
-}]>;
-
-def lo16 : PatLeaf<(imm), [{
-  // lo16 predicate - returns true if the immediate has all zeros in the
-  // low order bits and is a 32-bit constant:
-  if (N->getValueType(0) == MVT::i32) {
-    uint32_t val = N->getZExtValue();
-    return ((val & 0x0000ffff) == val);
-  }
-
-  return false;
-}], LO16>;
-
-def hi16 : PatLeaf<(imm), [{
-  // hi16 predicate - returns true if the immediate has all zeros in the
-  // low order bits and is a 32-bit constant:
-  if (N->getValueType(0) == MVT::i32) {
-    uint32_t val = uint32_t(N->getZExtValue());
-    return ((val & 0xffff0000) == val);
-  } else if (N->getValueType(0) == MVT::i64) {
-    uint64_t val = N->getZExtValue();
-    return ((val & 0xffff0000ULL) == val);
-  }
-
-  return false;
-}], HI16>;
-
-def bitshift : PatLeaf<(imm), [{
-  // bitshift predicate - returns true if 0 < imm <= 7 for SHLQBII
-  // (shift left quadword by bits immediate)
-  int64_t Val = N->getZExtValue();
-  return (Val > 0 && Val <= 7);
-}]>;
-
-//===----------------------------------------------------------------------===//
-// Floating point operands:
-//===----------------------------------------------------------------------===//
-
-// Transform a float, returning the high 16 bits shifted down, as if
-// the float was really an unsigned integer:
-def HI16_f32 : SDNodeXForm<fpimm, [{
-  float fval = N->getValueAPF().convertToFloat();
-  return getI32Imm(FloatToBits(fval) >> 16);
-}]>;
-
-// Transformation function on floats: get the low 16 bits as if the float was
-// an unsigned integer.
-def LO16_f32 : SDNodeXForm<fpimm, [{
-  float fval = N->getValueAPF().convertToFloat();
-  return getI32Imm(FloatToBits(fval) & 0xffff);
-}]>;
-
-def FPimm_sext16 : SDNodeXForm<fpimm, [{
-  float fval = N->getValueAPF().convertToFloat();
-  return getI32Imm((int) ((FloatToBits(fval) << 16) >> 16));
-}]>;
-
-def FPimm_u18 : SDNodeXForm<fpimm, [{
-  float fval = N->getValueAPF().convertToFloat();
-  return getI32Imm(FloatToBits(fval) & ((1 << 18) - 1));
-}]>;
-
-def fpimmSExt16 : PatLeaf<(fpimm), [{
-  short Ignored;
-  return isFPS16Immediate(N, Ignored);  
-}], FPimm_sext16>;
-
-// Does the SFP constant only have upp 16 bits set?
-def hi16_f32 : PatLeaf<(fpimm), [{
-  if (N->getValueType(0) == MVT::f32) {
-    uint32_t val = FloatToBits(N->getValueAPF().convertToFloat());
-    return ((val & 0xffff0000) == val);
-  }
-
-  return false;
-}], HI16_f32>;
-
-// Does the SFP constant fit into 18 bits?
-def fpimm18  : PatLeaf<(fpimm), [{
-  if (N->getValueType(0) == MVT::f32) {
-    uint32_t Value = FloatToBits(N->getValueAPF().convertToFloat());
-    return isUInt<18>(Value);
-  }
-
-  return false;
-}], FPimm_u18>;
-
-//===----------------------------------------------------------------------===//
-// 64-bit operands (TODO):
-//===----------------------------------------------------------------------===//
-
-//===----------------------------------------------------------------------===//
-// build_vector operands:
-//===----------------------------------------------------------------------===//
-
-// v16i8SExt8Imm_xform function: convert build_vector to 8-bit sign extended
-// immediate constant load for v16i8 vectors. N.B.: The incoming constant has
-// to be a 16-bit quantity with the upper and lower bytes equal (e.g., 0x2a2a).
-def v16i8SExt8Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i8imm(N, *CurDAG, MVT::i8);
-}]>;
-
-// v16i8SExt8Imm: Predicate test for 8-bit sign extended immediate constant
-// load, works in conjunction with its transform function. N.B.: This relies the
-// incoming constant being a 16-bit quantity, where the upper and lower bytes
-// are EXACTLY the same (e.g., 0x2a2a)
-def v16i8SExt8Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i8imm(N, *CurDAG, MVT::i8).getNode() != 0;
-}], v16i8SExt8Imm_xform>;
-
-// v16i8U8Imm_xform function: convert build_vector to unsigned 8-bit
-// immediate constant load for v16i8 vectors. N.B.: The incoming constant has
-// to be a 16-bit quantity with the upper and lower bytes equal (e.g., 0x2a2a).
-def v16i8U8Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i8imm(N, *CurDAG, MVT::i8);
-}]>;
-
-// v16i8U8Imm: Predicate test for unsigned 8-bit immediate constant
-// load, works in conjunction with its transform function. N.B.: This relies the
-// incoming constant being a 16-bit quantity, where the upper and lower bytes
-// are EXACTLY the same (e.g., 0x2a2a)
-def v16i8U8Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i8imm(N, *CurDAG, MVT::i8).getNode() != 0;
-}], v16i8U8Imm_xform>;
-
-// v8i16SExt8Imm_xform function: convert build_vector to 8-bit sign extended
-// immediate constant load for v8i16 vectors.
-def v8i16SExt8Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i8imm(N, *CurDAG, MVT::i16);
-}]>;
-
-// v8i16SExt8Imm: Predicate test for 8-bit sign extended immediate constant
-// load, works in conjunction with its transform function.
-def v8i16SExt8Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i8imm(N, *CurDAG, MVT::i16).getNode() != 0;
-}], v8i16SExt8Imm_xform>;
-
-// v8i16SExt10Imm_xform function: convert build_vector to 16-bit sign extended
-// immediate constant load for v8i16 vectors.
-def v8i16SExt10Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i10imm(N, *CurDAG, MVT::i16);
-}]>;
-
-// v8i16SExt10Imm: Predicate test for 16-bit sign extended immediate constant
-// load, works in conjunction with its transform function.
-def v8i16SExt10Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i10imm(N, *CurDAG, MVT::i16).getNode() != 0;
-}], v8i16SExt10Imm_xform>;
-
-// v8i16Uns10Imm_xform function: convert build_vector to 16-bit unsigned
-// immediate constant load for v8i16 vectors.
-def v8i16Uns10Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i10imm(N, *CurDAG, MVT::i16);
-}]>;
-
-// v8i16Uns10Imm: Predicate test for 16-bit unsigned immediate constant
-// load, works in conjunction with its transform function.
-def v8i16Uns10Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i10imm(N, *CurDAG, MVT::i16).getNode() != 0;
-}], v8i16Uns10Imm_xform>;
-
-// v8i16SExt16Imm_xform function: convert build_vector to 16-bit sign extended
-// immediate constant load for v8i16 vectors.
-def v8i16Uns16Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i16imm(N, *CurDAG, MVT::i16);
-}]>;
-
-// v8i16SExt16Imm: Predicate test for 16-bit sign extended immediate constant
-// load, works in conjunction with its transform function.
-def v8i16SExt16Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i16imm(N, *CurDAG, MVT::i16).getNode() != 0;
-}], v8i16Uns16Imm_xform>;
-
-// v4i32SExt10Imm_xform function: convert build_vector to 10-bit sign extended
-// immediate constant load for v4i32 vectors.
-def v4i32SExt10Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i10imm(N, *CurDAG, MVT::i32);
-}]>;
-
-// v4i32SExt10Imm: Predicate test for 10-bit sign extended immediate constant
-// load, works in conjunction with its transform function.
-def v4i32SExt10Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i10imm(N, *CurDAG, MVT::i32).getNode() != 0;
-}], v4i32SExt10Imm_xform>;
-
-// v4i32Uns10Imm_xform function: convert build_vector to 10-bit unsigned
-// immediate constant load for v4i32 vectors.
-def v4i32Uns10Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i10imm(N, *CurDAG, MVT::i32);
-}]>;
-
-// v4i32Uns10Imm: Predicate test for 10-bit unsigned immediate constant
-// load, works in conjunction with its transform function.
-def v4i32Uns10Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i10imm(N, *CurDAG, MVT::i32).getNode() != 0;
-}], v4i32Uns10Imm_xform>;
-
-// v4i32SExt16Imm_xform function: convert build_vector to 16-bit sign extended
-// immediate constant load for v4i32 vectors.
-def v4i32SExt16Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i16imm(N, *CurDAG, MVT::i32);
-}]>;
-
-// v4i32SExt16Imm: Predicate test for 16-bit sign extended immediate constant
-// load, works in conjunction with its transform function.
-def v4i32SExt16Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i16imm(N, *CurDAG, MVT::i32).getNode() != 0;
-}], v4i32SExt16Imm_xform>;
-
-// v4i32Uns18Imm_xform function: convert build_vector to 18-bit unsigned
-// immediate constant load for v4i32 vectors.
-def v4i32Uns18Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_u18imm(N, *CurDAG, MVT::i32);
-}]>;
-
-// v4i32Uns18Imm: Predicate test for 18-bit unsigned immediate constant load,
-// works in conjunction with its transform function.
-def v4i32Uns18Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_u18imm(N, *CurDAG, MVT::i32).getNode() != 0;
-}], v4i32Uns18Imm_xform>;
-
-// ILHUvec_get_imm xform function: convert build_vector to ILHUvec imm constant
-// load.
-def ILHUvec_get_imm: SDNodeXForm<build_vector, [{
-  return SPU::get_ILHUvec_imm(N, *CurDAG, MVT::i32);
-}]>;
-
-/// immILHUvec: Predicate test for a ILHU constant vector.
-def immILHUvec: PatLeaf<(build_vector), [{
-  return SPU::get_ILHUvec_imm(N, *CurDAG, MVT::i32).getNode() != 0;
-}], ILHUvec_get_imm>;
-
-// Catch-all for any other i32 vector constants
-def v4i32_get_imm: SDNodeXForm<build_vector, [{
-  return SPU::get_v4i32_imm(N, *CurDAG);
-}]>;
-
-def v4i32Imm: PatLeaf<(build_vector), [{
-  return SPU::get_v4i32_imm(N, *CurDAG).getNode() != 0;
-}], v4i32_get_imm>;
-
-// v2i64SExt10Imm_xform function: convert build_vector to 10-bit sign extended
-// immediate constant load for v2i64 vectors.
-def v2i64SExt10Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i10imm(N, *CurDAG, MVT::i64);
-}]>;
-
-// v2i64SExt10Imm: Predicate test for 10-bit sign extended immediate constant
-// load, works in conjunction with its transform function.
-def v2i64SExt10Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i10imm(N, *CurDAG, MVT::i64).getNode() != 0;
-}], v2i64SExt10Imm_xform>;
-
-// v2i64SExt16Imm_xform function: convert build_vector to 16-bit sign extended
-// immediate constant load for v2i64 vectors.
-def v2i64SExt16Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_i16imm(N, *CurDAG, MVT::i64);
-}]>;
-
-// v2i64SExt16Imm: Predicate test for 16-bit sign extended immediate constant
-// load, works in conjunction with its transform function.
-def v2i64SExt16Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_i16imm(N, *CurDAG, MVT::i64).getNode() != 0;
-}], v2i64SExt16Imm_xform>;
-
-// v2i64Uns18Imm_xform function: convert build_vector to 18-bit unsigned
-// immediate constant load for v2i64 vectors.
-def v2i64Uns18Imm_xform: SDNodeXForm<build_vector, [{
-  return SPU::get_vec_u18imm(N, *CurDAG, MVT::i64);
-}]>;
-
-// v2i64Uns18Imm: Predicate test for 18-bit unsigned immediate constant load,
-// works in conjunction with its transform function.
-def v2i64Uns18Imm: PatLeaf<(build_vector), [{
-  return SPU::get_vec_u18imm(N, *CurDAG, MVT::i64).getNode() != 0;
-}], v2i64Uns18Imm_xform>;
-
-/// immILHUvec: Predicate test for a ILHU constant vector.
-def immILHUvec_i64: PatLeaf<(build_vector), [{
-  return SPU::get_ILHUvec_imm(N, *CurDAG, MVT::i64).getNode() != 0;
-}], ILHUvec_get_imm>;
-
-// Catch-all for any other i32 vector constants
-def v2i64_get_imm: SDNodeXForm<build_vector, [{
-  return SPU::get_v2i64_imm(N, *CurDAG);
-}]>;
-
-def v2i64Imm: PatLeaf<(build_vector), [{
-  return SPU::get_v2i64_imm(N, *CurDAG).getNode() != 0;
-}], v2i64_get_imm>;
-
-//===----------------------------------------------------------------------===//
-// Operand Definitions.
-
-def s7imm: Operand<i8> {
-  let PrintMethod = "printS7ImmOperand";
-}
-
-def s7imm_i8: Operand<i8> {
-  let PrintMethod = "printS7ImmOperand";
-}
-
-def u7imm: Operand<i16> {
-  let PrintMethod = "printU7ImmOperand";
-}
-
-def u7imm_i8: Operand<i8> {
-  let PrintMethod = "printU7ImmOperand";
-}
-
-def u7imm_i32: Operand<i32> {
-  let PrintMethod = "printU7ImmOperand";
-}
-
-// Halfword, signed 10-bit constant
-def s10imm : Operand<i16> {
-  let PrintMethod = "printS10ImmOperand";
-}
-
-def s10imm_i8: Operand<i8> {
-  let PrintMethod = "printS10ImmOperand";
-}
-
-def s10imm_i32: Operand<i32> {
-  let PrintMethod = "printS10ImmOperand";
-}
-
-def s10imm_i64: Operand<i64> {
-  let PrintMethod = "printS10ImmOperand";
-}
-
-// Unsigned 10-bit integers:
-def u10imm: Operand<i16> {
-  let PrintMethod = "printU10ImmOperand";
-}
-
-def u10imm_i8: Operand<i8> {
-  let PrintMethod = "printU10ImmOperand";
-}
-
-def u10imm_i32: Operand<i32> {
-  let PrintMethod = "printU10ImmOperand";
-}
-
-def s16imm  : Operand<i16> {
-  let PrintMethod = "printS16ImmOperand";
-}
-
-def s16imm_i8: Operand<i8> {
-  let PrintMethod = "printS16ImmOperand";
-}
-
-def s16imm_i32: Operand<i32> {
-  let PrintMethod = "printS16ImmOperand";
-}
-
-def s16imm_i64: Operand<i64> {
-  let PrintMethod = "printS16ImmOperand";
-}
-
-def s16imm_f32: Operand<f32> {
-  let PrintMethod = "printS16ImmOperand";
-}
-
-def s16imm_f64: Operand<f64> {
-  let PrintMethod = "printS16ImmOperand";
-}
-
-def u16imm_i64 : Operand<i64> {
-  let PrintMethod = "printU16ImmOperand";
-}
-
-def u16imm_i32 : Operand<i32> {
-  let PrintMethod = "printU16ImmOperand";
-}
-
-def u16imm : Operand<i16> {
-  let PrintMethod = "printU16ImmOperand";
-}
-
-def f16imm : Operand<f32> {
-  let PrintMethod = "printU16ImmOperand";
-}
-
-def s18imm  : Operand<i32> {
-  let PrintMethod = "printS18ImmOperand";
-}
-
-def u18imm : Operand<i32> {
-  let PrintMethod = "printU18ImmOperand";
-}
-
-def u18imm_i64 : Operand<i64> {
-  let PrintMethod = "printU18ImmOperand";
-}
-
-def f18imm : Operand<f32> {
-  let PrintMethod = "printU18ImmOperand";
-}
-
-def f18imm_f64 : Operand<f64> {
-  let PrintMethod = "printU18ImmOperand";
-}
-
-// Negated 7-bit halfword rotate immediate operands
-def rothNeg7imm : Operand<i32> {
-  let PrintMethod = "printROTHNeg7Imm";
-}
-
-def rothNeg7imm_i16 : Operand<i16> {
-  let PrintMethod = "printROTHNeg7Imm";
-}
-
-// Negated 7-bit word rotate immediate operands
-def rotNeg7imm : Operand<i32> {
-  let PrintMethod = "printROTNeg7Imm";
-}
-
-def rotNeg7imm_i16 : Operand<i16> {
-  let PrintMethod = "printROTNeg7Imm";
-}
-
-def rotNeg7imm_i8 : Operand<i8> {
-  let PrintMethod = "printROTNeg7Imm";
-}
-
-def target : Operand<OtherVT> {
-  let PrintMethod = "printBranchOperand";
-}
-
-// Absolute address call target
-def calltarget : Operand<iPTR> {
-  let PrintMethod = "printCallOperand";
-  let MIOperandInfo = (ops u18imm:$calldest);
-}
-
-// PC relative call target
-def relcalltarget : Operand<iPTR> {
-  let PrintMethod = "printPCRelativeOperand";
-  let MIOperandInfo = (ops s16imm:$calldest);
-}
-
-// Branch targets:
-def brtarget : Operand<OtherVT> {
-  let PrintMethod = "printPCRelativeOperand";
-}
-
-// Hint for branch target
-def hbrtarget : Operand<OtherVT> {
-  let PrintMethod = "printHBROperand";
-}
-
-// Indirect call target
-def indcalltarget : Operand<iPTR> {
-  let PrintMethod = "printCallOperand";
-  let MIOperandInfo = (ops ptr_rc:$calldest);
-}
-
-def symbolHi: Operand<i32> {
-  let PrintMethod = "printSymbolHi";
-}
-
-def symbolLo: Operand<i32> {
-  let PrintMethod = "printSymbolLo";
-}
-
-def symbolLSA: Operand<i32> {
-  let PrintMethod = "printSymbolLSA";
-}
-
-// Shuffle address memory operaand [s7imm(reg) d-format]
-def shufaddr : Operand<iPTR> {
-  let PrintMethod = "printShufAddr";
-  let MIOperandInfo = (ops s7imm:$imm, ptr_rc:$reg);
-}
-
-// memory s10imm(reg) operand
-def dformaddr : Operand<iPTR> {
-  let PrintMethod = "printDFormAddr";
-  let MIOperandInfo = (ops s10imm:$imm, ptr_rc:$reg);
-}
-
-// 256K local store address
-// N.B.: The tblgen code generator expects to have two operands, an offset
-// and a pointer. Of these, only the immediate is actually used.
-def addr256k : Operand<iPTR> {
-  let PrintMethod = "printAddr256K";
-  let MIOperandInfo = (ops s16imm:$imm, ptr_rc:$reg);
-}
-
-// memory s18imm(reg) operand
-def memri18 : Operand<iPTR> {
-  let PrintMethod = "printMemRegImmS18";
-  let MIOperandInfo = (ops s18imm:$imm, ptr_rc:$reg);
-}
-
-// memory register + register operand
-def memrr : Operand<iPTR> {
-  let PrintMethod = "printMemRegReg";
-  let MIOperandInfo = (ops ptr_rc:$reg_a, ptr_rc:$reg_b);
-}
-
-// Define SPU-specific addressing modes: These come in three basic
-// flavors:
-//
-// D-form   : [r+I10] (10-bit signed offset + reg)
-// X-form   : [r+r]   (reg+reg)
-// A-form   : abs     (256K LSA offset)
-// D-form(2): [r+I7]  (7-bit signed offset + reg)
-
-def dform_addr   : ComplexPattern<iPTR, 2, "SelectDFormAddr",
-                                  [], [SDNPWantRoot]>;
-def xform_addr   : ComplexPattern<iPTR, 2, "SelectXFormAddr",
-                                  [], [SDNPWantRoot]>;
-def aform_addr   : ComplexPattern<iPTR, 2, "SelectAFormAddr",
-                                  [], [SDNPWantRoot]>;
-def dform2_addr  : ComplexPattern<iPTR, 2, "SelectDForm2Addr",
-                                  [], [SDNPWantRoot]>;
diff --git a/lib/Target/CellSPU/SPURegisterInfo.cpp b/lib/Target/CellSPU/SPURegisterInfo.cpp
deleted file mode 100644
index e6c872d0bbb7..000000000000
--- a/lib/Target/CellSPU/SPURegisterInfo.cpp
+++ /dev/null
@@ -1,357 +0,0 @@
-//===-- SPURegisterInfo.cpp - Cell SPU Register Information ---------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file contains the Cell implementation of the TargetRegisterInfo class.
-//
-//===----------------------------------------------------------------------===//
-
-#define DEBUG_TYPE "reginfo"
-#include "SPURegisterInfo.h"
-#include "SPU.h"
-#include "SPUInstrBuilder.h"
-#include "SPUSubtarget.h"
-#include "SPUMachineFunction.h"
-#include "SPUFrameLowering.h"
-#include "llvm/Constants.h"
-#include "llvm/Type.h"
-#include "llvm/CodeGen/ValueTypes.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/CodeGen/MachineFunction.h"
-#include "llvm/CodeGen/MachineFrameInfo.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/CodeGen/ValueTypes.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Support/CommandLine.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/MathExtras.h"
-#include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/STLExtras.h"
-#include <cstdlib>
-
-#define GET_REGINFO_TARGET_DESC
-#include "SPUGenRegisterInfo.inc"
-
-using namespace llvm;
-
-/// getRegisterNumbering - Given the enum value for some register, e.g.
-/// PPC::F14, return the number that it corresponds to (e.g. 14).
-unsigned SPURegisterInfo::getRegisterNumbering(unsigned RegEnum) {
-  using namespace SPU;
-  switch (RegEnum) {
-  case SPU::R0: return 0;
-  case SPU::R1: return 1;
-  case SPU::R2: return 2;
-  case SPU::R3: return 3;
-  case SPU::R4: return 4;
-  case SPU::R5: return 5;
-  case SPU::R6: return 6;
-  case SPU::R7: return 7;
-  case SPU::R8: return 8;
-  case SPU::R9: return 9;
-  case SPU::R10: return 10;
-  case SPU::R11: return 11;
-  case SPU::R12: return 12;
-  case SPU::R13: return 13;
-  case SPU::R14: return 14;
-  case SPU::R15: return 15;
-  case SPU::R16: return 16;
-  case SPU::R17: return 17;
-  case SPU::R18: return 18;
-  case SPU::R19: return 19;
-  case SPU::R20: return 20;
-  case SPU::R21: return 21;
-  case SPU::R22: return 22;
-  case SPU::R23: return 23;
-  case SPU::R24: return 24;
-  case SPU::R25: return 25;
-  case SPU::R26: return 26;
-  case SPU::R27: return 27;
-  case SPU::R28: return 28;
-  case SPU::R29: return 29;
-  case SPU::R30: return 30;
-  case SPU::R31: return 31;
-  case SPU::R32: return 32;
-  case SPU::R33: return 33;
-  case SPU::R34: return 34;
-  case SPU::R35: return 35;
-  case SPU::R36: return 36;
-  case SPU::R37: return 37;
-  case SPU::R38: return 38;
-  case SPU::R39: return 39;
-  case SPU::R40: return 40;
-  case SPU::R41: return 41;
-  case SPU::R42: return 42;
-  case SPU::R43: return 43;
-  case SPU::R44: return 44;
-  case SPU::R45: return 45;
-  case SPU::R46: return 46;
-  case SPU::R47: return 47;
-  case SPU::R48: return 48;
-  case SPU::R49: return 49;
-  case SPU::R50: return 50;
-  case SPU::R51: return 51;
-  case SPU::R52: return 52;
-  case SPU::R53: return 53;
-  case SPU::R54: return 54;
-  case SPU::R55: return 55;
-  case SPU::R56: return 56;
-  case SPU::R57: return 57;
-  case SPU::R58: return 58;
-  case SPU::R59: return 59;
-  case SPU::R60: return 60;
-  case SPU::R61: return 61;
-  case SPU::R62: return 62;
-  case SPU::R63: return 63;
-  case SPU::R64: return 64;
-  case SPU::R65: return 65;
-  case SPU::R66: return 66;
-  case SPU::R67: return 67;
-  case SPU::R68: return 68;
-  case SPU::R69: return 69;
-  case SPU::R70: return 70;
-  case SPU::R71: return 71;
-  case SPU::R72: return 72;
-  case SPU::R73: return 73;
-  case SPU::R74: return 74;
-  case SPU::R75: return 75;
-  case SPU::R76: return 76;
-  case SPU::R77: return 77;
-  case SPU::R78: return 78;
-  case SPU::R79: return 79;
-  case SPU::R80: return 80;
-  case SPU::R81: return 81;
-  case SPU::R82: return 82;
-  case SPU::R83: return 83;
-  case SPU::R84: return 84;
-  case SPU::R85: return 85;
-  case SPU::R86: return 86;
-  case SPU::R87: return 87;
-  case SPU::R88: return 88;
-  case SPU::R89: return 89;
-  case SPU::R90: return 90;
-  case SPU::R91: return 91;
-  case SPU::R92: return 92;
-  case SPU::R93: return 93;
-  case SPU::R94: return 94;
-  case SPU::R95: return 95;
-  case SPU::R96: return 96;
-  case SPU::R97: return 97;
-  case SPU::R98: return 98;
-  case SPU::R99: return 99;
-  case SPU::R100: return 100;
-  case SPU::R101: return 101;
-  case SPU::R102: return 102;
-  case SPU::R103: return 103;
-  case SPU::R104: return 104;
-  case SPU::R105: return 105;
-  case SPU::R106: return 106;
-  case SPU::R107: return 107;
-  case SPU::R108: return 108;
-  case SPU::R109: return 109;
-  case SPU::R110: return 110;
-  case SPU::R111: return 111;
-  case SPU::R112: return 112;
-  case SPU::R113: return 113;
-  case SPU::R114: return 114;
-  case SPU::R115: return 115;
-  case SPU::R116: return 116;
-  case SPU::R117: return 117;
-  case SPU::R118: return 118;
-  case SPU::R119: return 119;
-  case SPU::R120: return 120;
-  case SPU::R121: return 121;
-  case SPU::R122: return 122;
-  case SPU::R123: return 123;
-  case SPU::R124: return 124;
-  case SPU::R125: return 125;
-  case SPU::R126: return 126;
-  case SPU::R127: return 127;
-  default:
-    report_fatal_error("Unhandled reg in SPURegisterInfo::getRegisterNumbering");
-  }
-}
-
-SPURegisterInfo::SPURegisterInfo(const SPUSubtarget &subtarget,
-                                 const TargetInstrInfo &tii) :
-  SPUGenRegisterInfo(SPU::R0), Subtarget(subtarget), TII(tii)
-{
-}
-
-/// getPointerRegClass - Return the register class to use to hold pointers.
-/// This is used for addressing modes.
-const TargetRegisterClass *
-SPURegisterInfo::getPointerRegClass(const MachineFunction &MF, unsigned Kind)
-                                                                        const {
-  return &SPU::R32CRegClass;
-}
-
-const uint16_t *
-SPURegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const
-{
-  // Cell ABI calling convention
-  static const uint16_t SPU_CalleeSaveRegs[] = {
-    SPU::R80, SPU::R81, SPU::R82, SPU::R83,
-    SPU::R84, SPU::R85, SPU::R86, SPU::R87,
-    SPU::R88, SPU::R89, SPU::R90, SPU::R91,
-    SPU::R92, SPU::R93, SPU::R94, SPU::R95,
-    SPU::R96, SPU::R97, SPU::R98, SPU::R99,
-    SPU::R100, SPU::R101, SPU::R102, SPU::R103,
-    SPU::R104, SPU::R105, SPU::R106, SPU::R107,
-    SPU::R108, SPU::R109, SPU::R110, SPU::R111,
-    SPU::R112, SPU::R113, SPU::R114, SPU::R115,
-    SPU::R116, SPU::R117, SPU::R118, SPU::R119,
-    SPU::R120, SPU::R121, SPU::R122, SPU::R123,
-    SPU::R124, SPU::R125, SPU::R126, SPU::R127,
-    SPU::R2,    /* environment pointer */
-    SPU::R1,    /* stack pointer */
-    SPU::R0,    /* link register */
-    0 /* end */
-  };
-
-  return SPU_CalleeSaveRegs;
-}
-
-/*!
- R0 (link register), R1 (stack pointer) and R2 (environment pointer -- this is
- generally unused) are the Cell's reserved registers
- */
-BitVector SPURegisterInfo::getReservedRegs(const MachineFunction &MF) const {
-  BitVector Reserved(getNumRegs());
-  Reserved.set(SPU::R0);                // LR
-  Reserved.set(SPU::R1);                // SP
-  Reserved.set(SPU::R2);                // environment pointer
-  return Reserved;
-}
-
-//===----------------------------------------------------------------------===//
-// Stack Frame Processing methods
-//===----------------------------------------------------------------------===//
-
-//--------------------------------------------------------------------------
-void
-SPURegisterInfo::eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                               MachineBasicBlock &MBB,
-                                               MachineBasicBlock::iterator I)
-  const
-{
-  // Simply discard ADJCALLSTACKDOWN, ADJCALLSTACKUP instructions.
-  MBB.erase(I);
-}
-
-void
-SPURegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, int SPAdj,
-                                     RegScavenger *RS) const
-{
-  unsigned i = 0;
-  MachineInstr &MI = *II;
-  MachineBasicBlock &MBB = *MI.getParent();
-  MachineFunction &MF = *MBB.getParent();
-  MachineFrameInfo *MFI = MF.getFrameInfo();
-  DebugLoc dl = II->getDebugLoc();
-
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
-  }
-
-  MachineOperand &SPOp = MI.getOperand(i);
-  int FrameIndex = SPOp.getIndex();
-
-  // Now add the frame object offset to the offset from r1.
-  int Offset = MFI->getObjectOffset(FrameIndex);
-
-  // Most instructions, except for generated FrameIndex additions using AIr32
-  // and ILAr32, have the immediate in operand 1. AIr32 and ILAr32 have the
-  // immediate in operand 2.
-  unsigned OpNo = 1;
-  if (MI.getOpcode() == SPU::AIr32 || MI.getOpcode() == SPU::ILAr32)
-    OpNo = 2;
-
-  MachineOperand &MO = MI.getOperand(OpNo);
-
-  // Offset is biased by $lr's slot at the bottom.
-  Offset += MO.getImm() + MFI->getStackSize() + SPUFrameLowering::minStackSize();
-  assert((Offset & 0xf) == 0
-         && "16-byte alignment violated in eliminateFrameIndex");
-
-  // Replace the FrameIndex with base register with $sp (aka $r1)
-  SPOp.ChangeToRegister(SPU::R1, false);
-
-  // if 'Offset' doesn't fit to the D-form instruction's
-  // immediate, convert the instruction to X-form
-  // if the instruction is not an AI (which takes a s10 immediate), assume
-  // it is a load/store that can take a s14 immediate
-  if ((MI.getOpcode() == SPU::AIr32 && !isInt<10>(Offset))
-      || !isInt<14>(Offset)) {
-    int newOpcode = convertDFormToXForm(MI.getOpcode());
-    unsigned tmpReg = findScratchRegister(II, RS, &SPU::R32CRegClass, SPAdj);
-    BuildMI(MBB, II, dl, TII.get(SPU::ILr32), tmpReg )
-        .addImm(Offset);
-    BuildMI(MBB, II, dl, TII.get(newOpcode), MI.getOperand(0).getReg())
-        .addReg(tmpReg, RegState::Kill)
-        .addReg(SPU::R1);
-    // remove the replaced D-form instruction
-    MBB.erase(II);
-  } else {
-    MO.ChangeToImmediate(Offset);
-  }
-}
-
-unsigned
-SPURegisterInfo::getFrameRegister(const MachineFunction &MF) const
-{
-  return SPU::R1;
-}
-
-int
-SPURegisterInfo::convertDFormToXForm(int dFormOpcode) const
-{
-  switch(dFormOpcode)
-  {
-    case SPU::AIr32:     return SPU::Ar32;
-    case SPU::LQDr32:    return SPU::LQXr32;
-    case SPU::LQDr128:   return SPU::LQXr128;
-    case SPU::LQDv16i8:  return SPU::LQXv16i8;
-    case SPU::LQDv4i32:  return SPU::LQXv4i32;
-    case SPU::LQDv4f32:  return SPU::LQXv4f32;
-    case SPU::STQDr32:   return SPU::STQXr32;
-    case SPU::STQDr128:  return SPU::STQXr128;
-    case SPU::STQDv16i8: return SPU::STQXv16i8;
-    case SPU::STQDv4i32: return SPU::STQXv4i32;
-    case SPU::STQDv4f32: return SPU::STQXv4f32;
-
-    default: assert( false && "Unhandled D to X-form conversion");
-  }
-  // default will assert, but need to return something to keep the
-  // compiler happy.
-  return dFormOpcode;
-}
-
-// TODO this is already copied from PPC. Could this convenience function
-// be moved to the RegScavenger class?
-unsigned
-SPURegisterInfo::findScratchRegister(MachineBasicBlock::iterator II,
-                                     RegScavenger *RS,
-                                     const TargetRegisterClass *RC,
-                                     int SPAdj) const
-{
-  assert(RS && "Register scavenging must be on");
-  unsigned Reg = RS->FindUnusedReg(RC);
-  if (Reg == 0)
-    Reg = RS->scavengeRegister(RC, II, SPAdj);
-  assert( Reg && "Register scavenger failed");
-  return Reg;
-}
diff --git a/lib/Target/CellSPU/SPURegisterInfo.h b/lib/Target/CellSPU/SPURegisterInfo.h
deleted file mode 100644
index e9f9aba63a48..000000000000
--- a/lib/Target/CellSPU/SPURegisterInfo.h
+++ /dev/null
@@ -1,106 +0,0 @@
-//===-- SPURegisterInfo.h - Cell SPU Register Information Impl --*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file contains the Cell SPU implementation of the TargetRegisterInfo
-// class.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef SPU_REGISTERINFO_H
-#define SPU_REGISTERINFO_H
-
-#include "SPU.h"
-
-#define GET_REGINFO_HEADER
-#include "SPUGenRegisterInfo.inc"
-
-namespace llvm {
-  class SPUSubtarget;
-  class TargetInstrInfo;
-  class Type;
-
-  class SPURegisterInfo : public SPUGenRegisterInfo {
-  private:
-    const SPUSubtarget &Subtarget;
-    const TargetInstrInfo &TII;
-
-    //! Predicate: Does the machine function use the link register?
-    bool usesLR(MachineFunction &MF) const;
-
-  public:
-    SPURegisterInfo(const SPUSubtarget &subtarget, const TargetInstrInfo &tii);
- 
-    //! Translate a register's enum value to a register number
-    /*!
-      This method translates a register's enum value to it's regiser number,
-      e.g. SPU::R14 -> 14.
-     */
-    static unsigned getRegisterNumbering(unsigned RegEnum);
-
-    /// getPointerRegClass - Return the register class to use to hold pointers.
-    /// This is used for addressing modes.
-    virtual const TargetRegisterClass *
-    getPointerRegClass(const MachineFunction &MF, unsigned Kind = 0) const;
-
-    /// After allocating this many registers, the allocator should feel
-    /// register pressure. The value is a somewhat random guess, based on the
-    /// number of non callee saved registers in the C calling convention.
-    virtual unsigned getRegPressureLimit( const TargetRegisterClass *RC,
-                                          MachineFunction &MF) const{
-      return 50;
-    }
-
-    //! Return the array of callee-saved registers
-    virtual const uint16_t* getCalleeSavedRegs(const MachineFunction *MF) const;
-
-    //! Allow for scavenging, so we can get scratch registers when needed.
-    virtual bool requiresRegisterScavenging(const MachineFunction &MF) const
-    { return true; }
-
-    //! Enable tracking of liveness after register allocation, since register
-    // scavenging is enabled.
-    virtual bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const
-    { return true; }
-
-    //! Return the reserved registers
-    BitVector getReservedRegs(const MachineFunction &MF) const;
-
-    //! Eliminate the call frame setup pseudo-instructions
-    void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                       MachineBasicBlock &MBB,
-                                       MachineBasicBlock::iterator I) const;
-    //! Convert frame indicies into machine operands
-    void eliminateFrameIndex(MachineBasicBlock::iterator II, int SPAdj,
-                             RegScavenger *RS = NULL) const;
-
-    //! Get the stack frame register (SP, aka R1)
-    unsigned getFrameRegister(const MachineFunction &MF) const;
-
-    //------------------------------------------------------------------------
-    // New methods added:
-    //------------------------------------------------------------------------
-
-    //! Convert D-form load/store to X-form load/store
-    /*!
-      Converts a regiser displacement load/store into a register-indexed
-      load/store for large stack frames, when the stack frame exceeds the
-      range of a s10 displacement.
-     */
-    int convertDFormToXForm(int dFormOpcode) const;
-
-    //! Acquire an unused register in an emergency.
-    unsigned findScratchRegister(MachineBasicBlock::iterator II,
-                                 RegScavenger *RS,
-                                 const TargetRegisterClass *RC, 
-                                 int SPAdj) const;
-    
-  };
-} // end namespace llvm
-
-#endif
diff --git a/lib/Target/CellSPU/SPURegisterInfo.td b/lib/Target/CellSPU/SPURegisterInfo.td
deleted file mode 100644
index f27b042edd63..000000000000
--- a/lib/Target/CellSPU/SPURegisterInfo.td
+++ /dev/null
@@ -1,183 +0,0 @@
-//===-- SPURegisterInfo.td - The Cell SPU Register File ----*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-//
-//===----------------------------------------------------------------------===//
-
-class SPUReg<string n> : Register<n> {
-  let Namespace = "SPU";
-}
-
-// The SPU's register are all 128-bits wide, which makes specifying the
-// registers relatively easy, if relatively mundane:
-
-class SPUVecReg<bits<7> num, string n> : SPUReg<n> {
-  field bits<7> Num = num;
-}
-
-def R0 : SPUVecReg<0, "$lr">, DwarfRegNum<[0]>;
-def R1 : SPUVecReg<1, "$sp">, DwarfRegNum<[1]>;
-def R2 : SPUVecReg<2, "$2">, DwarfRegNum<[2]>;
-def R3 : SPUVecReg<3, "$3">, DwarfRegNum<[3]>;
-def R4 : SPUVecReg<4, "$4">, DwarfRegNum<[4]>;
-def R5 : SPUVecReg<5, "$5">, DwarfRegNum<[5]>;
-def R6 : SPUVecReg<6, "$6">, DwarfRegNum<[6]>;
-def R7 : SPUVecReg<7, "$7">, DwarfRegNum<[7]>;
-def R8 : SPUVecReg<8, "$8">, DwarfRegNum<[8]>;
-def R9 : SPUVecReg<9, "$9">, DwarfRegNum<[9]>;
-def R10 : SPUVecReg<10, "$10">, DwarfRegNum<[10]>;
-def R11 : SPUVecReg<11, "$11">, DwarfRegNum<[11]>;
-def R12 : SPUVecReg<12, "$12">, DwarfRegNum<[12]>;
-def R13 : SPUVecReg<13, "$13">, DwarfRegNum<[13]>;
-def R14 : SPUVecReg<14, "$14">, DwarfRegNum<[14]>;
-def R15 : SPUVecReg<15, "$15">, DwarfRegNum<[15]>;
-def R16 : SPUVecReg<16, "$16">, DwarfRegNum<[16]>;
-def R17 : SPUVecReg<17, "$17">, DwarfRegNum<[17]>;
-def R18 : SPUVecReg<18, "$18">, DwarfRegNum<[18]>;
-def R19 : SPUVecReg<19, "$19">, DwarfRegNum<[19]>;
-def R20 : SPUVecReg<20, "$20">, DwarfRegNum<[20]>;
-def R21 : SPUVecReg<21, "$21">, DwarfRegNum<[21]>;
-def R22 : SPUVecReg<22, "$22">, DwarfRegNum<[22]>;
-def R23 : SPUVecReg<23, "$23">, DwarfRegNum<[23]>;
-def R24 : SPUVecReg<24, "$24">, DwarfRegNum<[24]>;
-def R25 : SPUVecReg<25, "$25">, DwarfRegNum<[25]>;
-def R26 : SPUVecReg<26, "$26">, DwarfRegNum<[26]>;
-def R27 : SPUVecReg<27, "$27">, DwarfRegNum<[27]>;
-def R28 : SPUVecReg<28, "$28">, DwarfRegNum<[28]>;
-def R29 : SPUVecReg<29, "$29">, DwarfRegNum<[29]>;
-def R30 : SPUVecReg<30, "$30">, DwarfRegNum<[30]>;
-def R31 : SPUVecReg<31, "$31">, DwarfRegNum<[31]>;
-def R32 : SPUVecReg<32, "$32">, DwarfRegNum<[32]>;
-def R33 : SPUVecReg<33, "$33">, DwarfRegNum<[33]>;
-def R34 : SPUVecReg<34, "$34">, DwarfRegNum<[34]>;
-def R35 : SPUVecReg<35, "$35">, DwarfRegNum<[35]>;
-def R36 : SPUVecReg<36, "$36">, DwarfRegNum<[36]>;
-def R37 : SPUVecReg<37, "$37">, DwarfRegNum<[37]>;
-def R38 : SPUVecReg<38, "$38">, DwarfRegNum<[38]>;
-def R39 : SPUVecReg<39, "$39">, DwarfRegNum<[39]>;
-def R40 : SPUVecReg<40, "$40">, DwarfRegNum<[40]>;
-def R41 : SPUVecReg<41, "$41">, DwarfRegNum<[41]>;
-def R42 : SPUVecReg<42, "$42">, DwarfRegNum<[42]>;
-def R43 : SPUVecReg<43, "$43">, DwarfRegNum<[43]>;
-def R44 : SPUVecReg<44, "$44">, DwarfRegNum<[44]>;
-def R45 : SPUVecReg<45, "$45">, DwarfRegNum<[45]>;
-def R46 : SPUVecReg<46, "$46">, DwarfRegNum<[46]>;
-def R47 : SPUVecReg<47, "$47">, DwarfRegNum<[47]>;
-def R48 : SPUVecReg<48, "$48">, DwarfRegNum<[48]>;
-def R49 : SPUVecReg<49, "$49">, DwarfRegNum<[49]>;
-def R50 : SPUVecReg<50, "$50">, DwarfRegNum<[50]>;
-def R51 : SPUVecReg<51, "$51">, DwarfRegNum<[51]>;
-def R52 : SPUVecReg<52, "$52">, DwarfRegNum<[52]>;
-def R53 : SPUVecReg<53, "$53">, DwarfRegNum<[53]>;
-def R54 : SPUVecReg<54, "$54">, DwarfRegNum<[54]>;
-def R55 : SPUVecReg<55, "$55">, DwarfRegNum<[55]>;
-def R56 : SPUVecReg<56, "$56">, DwarfRegNum<[56]>;
-def R57 : SPUVecReg<57, "$57">, DwarfRegNum<[57]>;
-def R58 : SPUVecReg<58, "$58">, DwarfRegNum<[58]>;
-def R59 : SPUVecReg<59, "$59">, DwarfRegNum<[59]>;
-def R60 : SPUVecReg<60, "$60">, DwarfRegNum<[60]>;
-def R61 : SPUVecReg<61, "$61">, DwarfRegNum<[61]>;
-def R62 : SPUVecReg<62, "$62">, DwarfRegNum<[62]>;
-def R63 : SPUVecReg<63, "$63">, DwarfRegNum<[63]>;
-def R64 : SPUVecReg<64, "$64">, DwarfRegNum<[64]>;
-def R65 : SPUVecReg<65, "$65">, DwarfRegNum<[65]>;
-def R66 : SPUVecReg<66, "$66">, DwarfRegNum<[66]>;
-def R67 : SPUVecReg<67, "$67">, DwarfRegNum<[67]>;
-def R68 : SPUVecReg<68, "$68">, DwarfRegNum<[68]>;
-def R69 : SPUVecReg<69, "$69">, DwarfRegNum<[69]>;
-def R70 : SPUVecReg<70, "$70">, DwarfRegNum<[70]>;
-def R71 : SPUVecReg<71, "$71">, DwarfRegNum<[71]>;
-def R72 : SPUVecReg<72, "$72">, DwarfRegNum<[72]>;
-def R73 : SPUVecReg<73, "$73">, DwarfRegNum<[73]>;
-def R74 : SPUVecReg<74, "$74">, DwarfRegNum<[74]>;
-def R75 : SPUVecReg<75, "$75">, DwarfRegNum<[75]>;
-def R76 : SPUVecReg<76, "$76">, DwarfRegNum<[76]>;
-def R77 : SPUVecReg<77, "$77">, DwarfRegNum<[77]>;
-def R78 : SPUVecReg<78, "$78">, DwarfRegNum<[78]>;
-def R79 : SPUVecReg<79, "$79">, DwarfRegNum<[79]>;
-def R80 : SPUVecReg<80, "$80">, DwarfRegNum<[80]>;
-def R81 : SPUVecReg<81, "$81">, DwarfRegNum<[81]>;
-def R82 : SPUVecReg<82, "$82">, DwarfRegNum<[82]>;
-def R83 : SPUVecReg<83, "$83">, DwarfRegNum<[83]>;
-def R84 : SPUVecReg<84, "$84">, DwarfRegNum<[84]>;
-def R85 : SPUVecReg<85, "$85">, DwarfRegNum<[85]>;
-def R86 : SPUVecReg<86, "$86">, DwarfRegNum<[86]>;
-def R87 : SPUVecReg<87, "$87">, DwarfRegNum<[87]>;
-def R88 : SPUVecReg<88, "$88">, DwarfRegNum<[88]>;
-def R89 : SPUVecReg<89, "$89">, DwarfRegNum<[89]>;
-def R90 : SPUVecReg<90, "$90">, DwarfRegNum<[90]>;
-def R91 : SPUVecReg<91, "$91">, DwarfRegNum<[91]>;
-def R92 : SPUVecReg<92, "$92">, DwarfRegNum<[92]>;
-def R93 : SPUVecReg<93, "$93">, DwarfRegNum<[93]>;
-def R94 : SPUVecReg<94, "$94">, DwarfRegNum<[94]>;
-def R95 : SPUVecReg<95, "$95">, DwarfRegNum<[95]>;
-def R96 : SPUVecReg<96, "$96">, DwarfRegNum<[96]>;
-def R97 : SPUVecReg<97, "$97">, DwarfRegNum<[97]>;
-def R98 : SPUVecReg<98, "$98">, DwarfRegNum<[98]>;
-def R99 : SPUVecReg<99, "$99">, DwarfRegNum<[99]>;
-def R100 : SPUVecReg<100, "$100">, DwarfRegNum<[100]>;
-def R101 : SPUVecReg<101, "$101">, DwarfRegNum<[101]>;
-def R102 : SPUVecReg<102, "$102">, DwarfRegNum<[102]>;
-def R103 : SPUVecReg<103, "$103">, DwarfRegNum<[103]>;
-def R104 : SPUVecReg<104, "$104">, DwarfRegNum<[104]>;
-def R105 : SPUVecReg<105, "$105">, DwarfRegNum<[105]>;
-def R106 : SPUVecReg<106, "$106">, DwarfRegNum<[106]>;
-def R107 : SPUVecReg<107, "$107">, DwarfRegNum<[107]>;
-def R108 : SPUVecReg<108, "$108">, DwarfRegNum<[108]>;
-def R109 : SPUVecReg<109, "$109">, DwarfRegNum<[109]>;
-def R110 : SPUVecReg<110, "$110">, DwarfRegNum<[110]>;
-def R111 : SPUVecReg<111, "$111">, DwarfRegNum<[111]>;
-def R112 : SPUVecReg<112, "$112">, DwarfRegNum<[112]>;
-def R113 : SPUVecReg<113, "$113">, DwarfRegNum<[113]>;
-def R114 : SPUVecReg<114, "$114">, DwarfRegNum<[114]>;
-def R115 : SPUVecReg<115, "$115">, DwarfRegNum<[115]>;
-def R116 : SPUVecReg<116, "$116">, DwarfRegNum<[116]>;
-def R117 : SPUVecReg<117, "$117">, DwarfRegNum<[117]>;
-def R118 : SPUVecReg<118, "$118">, DwarfRegNum<[118]>;
-def R119 : SPUVecReg<119, "$119">, DwarfRegNum<[119]>;
-def R120 : SPUVecReg<120, "$120">, DwarfRegNum<[120]>;
-def R121 : SPUVecReg<121, "$121">, DwarfRegNum<[121]>;
-def R122 : SPUVecReg<122, "$122">, DwarfRegNum<[122]>;
-def R123 : SPUVecReg<123, "$123">, DwarfRegNum<[123]>;
-def R124 : SPUVecReg<124, "$124">, DwarfRegNum<[124]>;
-def R125 : SPUVecReg<125, "$125">, DwarfRegNum<[125]>;
-def R126 : SPUVecReg<126, "$126">, DwarfRegNum<[126]>;
-def R127 : SPUVecReg<127, "$127">, DwarfRegNum<[127]>;
-
-/* Need floating point status register here: */
-/* def FPCSR : ... */
-
-// The SPU's registers as 128-bit wide entities, and can function as general
-// purpose registers, where the operands are in the "preferred slot":
-// The non-volatile registers are allocated in reverse order, like PPC does it.
-def GPRC : RegisterClass<"SPU", [i128], 128,
-                         (add (sequence "R%u", 0, 79),
-                              (sequence "R%u", 127, 80))>;
-
-// The SPU's registers as 64-bit wide (double word integer) "preferred slot":
-def R64C : RegisterClass<"SPU", [i64], 128, (add GPRC)>;
-
-// The SPU's registers as 64-bit wide (double word) FP "preferred slot":
-def R64FP : RegisterClass<"SPU", [f64], 128, (add GPRC)>;
-
-// The SPU's registers as 32-bit wide (word) "preferred slot":
-def R32C : RegisterClass<"SPU", [i32], 128, (add GPRC)>;
-
-// The SPU's registers as single precision floating point "preferred slot":
-def R32FP : RegisterClass<"SPU", [f32], 128, (add GPRC)>;
-
-// The SPU's registers as 16-bit wide (halfword) "preferred slot":
-def R16C : RegisterClass<"SPU", [i16], 128, (add GPRC)>;
-
-// The SPU's registers as 8-bit wide (byte) "preferred slot":
-def R8C : RegisterClass<"SPU", [i8], 128, (add GPRC)>;
-
-// The SPU's registers as vector registers:
-def VECREG : RegisterClass<"SPU", [v16i8,v8i16,v4i32,v4f32,v2i64,v2f64], 128,
-                           (add GPRC)>;
diff --git a/lib/Target/CellSPU/SPURegisterNames.h b/lib/Target/CellSPU/SPURegisterNames.h
deleted file mode 100644
index e557ed340a28..000000000000
--- a/lib/Target/CellSPU/SPURegisterNames.h
+++ /dev/null
@@ -1,19 +0,0 @@
-//===- SPURegisterNames.h - Wrapper header for SPU register names -*- C++ -*-=//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef SPU_REGISTER_NAMES_H
-#define SPU_REGISTER_NAMES_H
-
-// Define symbolic names for Cell registers.  This defines a mapping from
-// register name to register number.
-//
-#define GET_REGINFO_ENUM
-#include "SPUGenRegisterInfo.inc"
-
-#endif
diff --git a/lib/Target/CellSPU/SPUSchedule.td b/lib/Target/CellSPU/SPUSchedule.td
deleted file mode 100644
index 9ccd0844e48e..000000000000
--- a/lib/Target/CellSPU/SPUSchedule.td
+++ /dev/null
@@ -1,59 +0,0 @@
-//===-- SPUSchedule.td - Cell Scheduling Definitions -------*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-//===----------------------------------------------------------------------===//
-// Even pipeline:
-
-def EVEN_UNIT : FuncUnit;       // Even execution unit: (PC & 0x7 == 000)
-def ODD_UNIT  : FuncUnit;       // Odd execution unit:  (PC & 0x7 == 100)
-
-//===----------------------------------------------------------------------===//
-// Instruction Itinerary classes used for Cell SPU
-//===----------------------------------------------------------------------===//
-
-def LoadStore    : InstrItinClass;              // ODD_UNIT
-def BranchHints  : InstrItinClass;              // ODD_UNIT
-def BranchResolv : InstrItinClass;              // ODD_UNIT
-def ChanOpSPR    : InstrItinClass;              // ODD_UNIT
-def ShuffleOp    : InstrItinClass;              // ODD_UNIT
-def SelectOp     : InstrItinClass;              // ODD_UNIT
-def GatherOp     : InstrItinClass;              // ODD_UNIT
-def LoadNOP      : InstrItinClass;              // ODD_UNIT
-def ExecNOP      : InstrItinClass;              // EVEN_UNIT
-def SPrecFP      : InstrItinClass;              // EVEN_UNIT
-def DPrecFP      : InstrItinClass;              // EVEN_UNIT
-def FPInt        : InstrItinClass;              // EVEN_UNIT (FP<->integer)
-def ByteOp       : InstrItinClass;              // EVEN_UNIT
-def IntegerOp    : InstrItinClass;              // EVEN_UNIT
-def IntegerMulDiv: InstrItinClass;              // EVEN_UNIT
-def RotShiftVec  : InstrItinClass;              // EVEN_UNIT Inter vector
-def RotShiftQuad : InstrItinClass;              // ODD_UNIT Entire quad
-def ImmLoad      : InstrItinClass;              // EVEN_UNIT
-
-/* Note: The itinerary for the Cell SPU is somewhat contrived... */
-def SPUItineraries : ProcessorItineraries<[ODD_UNIT, EVEN_UNIT], [], [
-  InstrItinData<LoadStore   , [InstrStage<6,  [ODD_UNIT]>]>,
-  InstrItinData<BranchHints , [InstrStage<6,  [ODD_UNIT]>]>,
-  InstrItinData<BranchResolv, [InstrStage<4,  [ODD_UNIT]>]>,
-  InstrItinData<ChanOpSPR   , [InstrStage<6,  [ODD_UNIT]>]>,
-  InstrItinData<ShuffleOp   , [InstrStage<4,  [ODD_UNIT]>]>,
-  InstrItinData<SelectOp    , [InstrStage<4,  [ODD_UNIT]>]>,
-  InstrItinData<GatherOp    , [InstrStage<4,  [ODD_UNIT]>]>,
-  InstrItinData<LoadNOP     , [InstrStage<1,  [ODD_UNIT]>]>,
-  InstrItinData<ExecNOP     , [InstrStage<1,  [EVEN_UNIT]>]>,
-  InstrItinData<SPrecFP     , [InstrStage<6,  [EVEN_UNIT]>]>,
-  InstrItinData<DPrecFP     , [InstrStage<13, [EVEN_UNIT]>]>,
-  InstrItinData<FPInt       , [InstrStage<2,  [EVEN_UNIT]>]>,
-  InstrItinData<ByteOp      , [InstrStage<4,  [EVEN_UNIT]>]>,
-  InstrItinData<IntegerOp   , [InstrStage<2,  [EVEN_UNIT]>]>,
-  InstrItinData<RotShiftVec , [InstrStage<4,  [EVEN_UNIT]>]>, 
-  InstrItinData<RotShiftQuad, [InstrStage<4,  [ODD_UNIT]>]>,
-  InstrItinData<IntegerMulDiv,[InstrStage<7,  [EVEN_UNIT]>]>,
-  InstrItinData<ImmLoad     , [InstrStage<2,  [EVEN_UNIT]>]>
-  ]>;
diff --git a/lib/Target/CellSPU/SPUSubtarget.cpp b/lib/Target/CellSPU/SPUSubtarget.cpp
deleted file mode 100644
index eec2d250be7f..000000000000
--- a/lib/Target/CellSPU/SPUSubtarget.cpp
+++ /dev/null
@@ -1,65 +0,0 @@
-//===-- SPUSubtarget.cpp - STI Cell SPU Subtarget Information -------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file implements the CellSPU-specific subclass of TargetSubtargetInfo.
-//
-//===----------------------------------------------------------------------===//
-
-#include "SPUSubtarget.h"
-#include "SPU.h"
-#include "SPURegisterInfo.h"
-#include "llvm/Support/TargetRegistry.h"
-
-#define GET_SUBTARGETINFO_TARGET_DESC
-#define GET_SUBTARGETINFO_CTOR
-#include "SPUGenSubtargetInfo.inc"
-
-using namespace llvm;
-
-SPUSubtarget::SPUSubtarget(const std::string &TT, const std::string &CPU,
-                           const std::string &FS) :
-  SPUGenSubtargetInfo(TT, CPU, FS),
-  StackAlignment(16),
-  ProcDirective(SPU::DEFAULT_PROC),
-  UseLargeMem(false)
-{
-  // Should be the target SPU processor type. For now, since there's only
-  // one, simply default to the current "v0" default:
-  std::string default_cpu("v0");
-
-  // Parse features string.
-  ParseSubtargetFeatures(default_cpu, FS);
-
-  // Initialize scheduling itinerary for the specified CPU.
-  InstrItins = getInstrItineraryForCPU(default_cpu);
-}
-
-/// SetJITMode - This is called to inform the subtarget info that we are
-/// producing code for the JIT.
-void SPUSubtarget::SetJITMode() {
-}
-
-/// Enable PostRA scheduling for optimization levels -O2 and -O3.
-bool SPUSubtarget::enablePostRAScheduler(
-                       CodeGenOpt::Level OptLevel,
-                       TargetSubtargetInfo::AntiDepBreakMode& Mode,
-                       RegClassVector& CriticalPathRCs) const {
-  Mode = TargetSubtargetInfo::ANTIDEP_CRITICAL;
-  // CriticalPathsRCs seems to be the set of
-  // RegisterClasses that antidep breakings are performed for.
-  // Do it for all register classes 
-  CriticalPathRCs.clear();
-  CriticalPathRCs.push_back(&SPU::R8CRegClass);
-  CriticalPathRCs.push_back(&SPU::R16CRegClass);
-  CriticalPathRCs.push_back(&SPU::R32CRegClass);
-  CriticalPathRCs.push_back(&SPU::R32FPRegClass);
-  CriticalPathRCs.push_back(&SPU::R64CRegClass);
-  CriticalPathRCs.push_back(&SPU::VECREGRegClass);
-  return OptLevel >= CodeGenOpt::Default;
-}
diff --git a/lib/Target/CellSPU/SPUSubtarget.h b/lib/Target/CellSPU/SPUSubtarget.h
deleted file mode 100644
index 27d28b22dd04..000000000000
--- a/lib/Target/CellSPU/SPUSubtarget.h
+++ /dev/null
@@ -1,97 +0,0 @@
-//===-- SPUSubtarget.h - Define Subtarget for the Cell SPU ------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file declares the Cell SPU-specific subclass of TargetSubtargetInfo.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef CELLSUBTARGET_H
-#define CELLSUBTARGET_H
-
-#include "llvm/Target/TargetSubtargetInfo.h"
-#include "llvm/MC/MCInstrItineraries.h"
-#include <string>
-
-#define GET_SUBTARGETINFO_HEADER
-#include "SPUGenSubtargetInfo.inc"
-
-namespace llvm {
-  class GlobalValue;
-  class StringRef;
-
-  namespace SPU {
-    enum {
-      PROC_NONE,
-      DEFAULT_PROC
-    };
-  }
-    
-  class SPUSubtarget : public SPUGenSubtargetInfo {
-  protected:
-    /// stackAlignment - The minimum alignment known to hold of the stack frame
-    /// on entry to the function and which must be maintained by every function.
-    unsigned StackAlignment;
-    
-    /// Selected instruction itineraries (one entry per itinerary class.)
-    InstrItineraryData InstrItins;
-
-    /// Which SPU processor (this isn't really used, but it's there to keep
-    /// the C compiler happy)
-    unsigned ProcDirective;
-
-    /// Use (assume) large memory -- effectively disables the LQA/STQA
-    /// instructions that assume 259K local store.
-    bool UseLargeMem;
-    
-  public:
-    /// This constructor initializes the data members to match that
-    /// of the specified triple.
-    ///
-    SPUSubtarget(const std::string &TT, const std::string &CPU,
-                 const std::string &FS);
-    
-    /// ParseSubtargetFeatures - Parses features string setting specified 
-    /// subtarget options.  Definition of function is auto generated by tblgen.
-    void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
-
-    /// SetJITMode - This is called to inform the subtarget info that we are
-    /// producing code for the JIT.
-    void SetJITMode();
-
-    /// getStackAlignment - Returns the minimum alignment known to hold of the
-    /// stack frame on entry to the function and which must be maintained by
-    /// every function for this subtarget.
-    unsigned getStackAlignment() const { return StackAlignment; }
-    
-    /// getInstrItins - Return the instruction itineraies based on subtarget 
-    /// selection.
-    const InstrItineraryData &getInstrItineraryData() const {
-      return InstrItins;
-    }
-
-    /// Use large memory addressing predicate
-    bool usingLargeMem() const {
-      return UseLargeMem;
-    }
-
-    /// getDataLayoutString - Return the pointer size and type alignment
-    /// properties of this subtarget.
-    const char *getDataLayoutString() const {
-      return "E-p:32:32:128-f64:64:128-f32:32:128-i64:32:128-i32:32:128"
-             "-i16:16:128-i8:8:128-i1:8:128-a:0:128-v64:64:128-v128:128:128"
-             "-s:128:128-n32:64";
-    }
-
-    bool enablePostRAScheduler(CodeGenOpt::Level OptLevel,
-                               TargetSubtargetInfo::AntiDepBreakMode& Mode,
-                               RegClassVector& CriticalPathRCs) const;
-  };
-} // End llvm namespace
-
-#endif
diff --git a/lib/Target/CellSPU/SPUTargetMachine.cpp b/lib/Target/CellSPU/SPUTargetMachine.cpp
deleted file mode 100644
index 918316572a2e..000000000000
--- a/lib/Target/CellSPU/SPUTargetMachine.cpp
+++ /dev/null
@@ -1,94 +0,0 @@
-//===-- SPUTargetMachine.cpp - Define TargetMachine for Cell SPU ----------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// Top-level implementation for the Cell SPU target.
-//
-//===----------------------------------------------------------------------===//
-
-#include "SPUTargetMachine.h"
-#include "SPU.h"
-#include "llvm/PassManager.h"
-#include "llvm/CodeGen/SchedulerRegistry.h"
-#include "llvm/Support/DynamicLibrary.h"
-#include "llvm/Support/TargetRegistry.h"
-
-using namespace llvm;
-
-extern "C" void LLVMInitializeCellSPUTarget() {
-  // Register the target.
-  RegisterTargetMachine<SPUTargetMachine> X(TheCellSPUTarget);
-}
-
-const std::pair<unsigned, int> *
-SPUFrameLowering::getCalleeSaveSpillSlots(unsigned &NumEntries) const {
-  NumEntries = 1;
-  return &LR[0];
-}
-
-SPUTargetMachine::SPUTargetMachine(const Target &T, StringRef TT,
-                                   StringRef CPU, StringRef FS,
-                                   const TargetOptions &Options,
-                                   Reloc::Model RM, CodeModel::Model CM,
-                                   CodeGenOpt::Level OL)
-  : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
-    Subtarget(TT, CPU, FS),
-    DL(Subtarget.getDataLayoutString()),
-    InstrInfo(*this),
-    FrameLowering(Subtarget),
-    TLInfo(*this),
-    TSInfo(*this),
-    InstrItins(Subtarget.getInstrItineraryData()),
-    STTI(&TLInfo), VTTI(&TLInfo) {
-}
-
-//===----------------------------------------------------------------------===//
-// Pass Pipeline Configuration
-//===----------------------------------------------------------------------===//
-
-namespace {
-/// SPU Code Generator Pass Configuration Options.
-class SPUPassConfig : public TargetPassConfig {
-public:
-  SPUPassConfig(SPUTargetMachine *TM, PassManagerBase &PM)
-    : TargetPassConfig(TM, PM) {}
-
-  SPUTargetMachine &getSPUTargetMachine() const {
-    return getTM<SPUTargetMachine>();
-  }
-
-  virtual bool addInstSelector();
-  virtual bool addPreEmitPass();
-};
-} // namespace
-
-TargetPassConfig *SPUTargetMachine::createPassConfig(PassManagerBase &PM) {
-  return new SPUPassConfig(this, PM);
-}
-
-bool SPUPassConfig::addInstSelector() {
-  // Install an instruction selector.
-  addPass(createSPUISelDag(getSPUTargetMachine()));
-  return false;
-}
-
-// passes to run just before printing the assembly
-bool SPUPassConfig::addPreEmitPass() {
-  // load the TCE instruction scheduler, if available via
-  // loaded plugins
-  typedef llvm::FunctionPass* (*BuilderFunc)(const char*);
-  BuilderFunc schedulerCreator =
-    (BuilderFunc)(intptr_t)sys::DynamicLibrary::SearchForAddressOfSymbol(
-          "createTCESchedulerPass");
-  if (schedulerCreator != NULL)
-      addPass(schedulerCreator("cellspu"));
-
-  //align instructions with nops/lnops for dual issue
-  addPass(createSPUNopFillerPass(getSPUTargetMachine()));
-  return true;
-}
diff --git a/lib/Target/CellSPU/SPUTargetMachine.h b/lib/Target/CellSPU/SPUTargetMachine.h
deleted file mode 100644
index 7f53ea6fbeb2..000000000000
--- a/lib/Target/CellSPU/SPUTargetMachine.h
+++ /dev/null
@@ -1,96 +0,0 @@
-//===-- SPUTargetMachine.h - Define TargetMachine for Cell SPU --*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file declares the CellSPU-specific subclass of TargetMachine.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef SPU_TARGETMACHINE_H
-#define SPU_TARGETMACHINE_H
-
-#include "SPUSubtarget.h"
-#include "SPUInstrInfo.h"
-#include "SPUISelLowering.h"
-#include "SPUSelectionDAGInfo.h"
-#include "SPUFrameLowering.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetTransformImpl.h"
-#include "llvm/DataLayout.h"
-
-namespace llvm {
-
-/// SPUTargetMachine
-///
-class SPUTargetMachine : public LLVMTargetMachine {
-  SPUSubtarget        Subtarget;
-  const DataLayout    DL;
-  SPUInstrInfo        InstrInfo;
-  SPUFrameLowering    FrameLowering;
-  SPUTargetLowering   TLInfo;
-  SPUSelectionDAGInfo TSInfo;
-  InstrItineraryData  InstrItins;
-  ScalarTargetTransformImpl STTI;
-  VectorTargetTransformImpl VTTI;
-public:
-  SPUTargetMachine(const Target &T, StringRef TT,
-                   StringRef CPU, StringRef FS, const TargetOptions &Options,
-                   Reloc::Model RM, CodeModel::Model CM,
-                   CodeGenOpt::Level OL);
-
-  /// Return the subtarget implementation object
-  virtual const SPUSubtarget     *getSubtargetImpl() const {
-    return &Subtarget;
-  }
-  virtual const SPUInstrInfo     *getInstrInfo() const {
-    return &InstrInfo;
-  }
-  virtual const SPUFrameLowering *getFrameLowering() const {
-    return &FrameLowering;
-  }
-  /*!
-    \note Cell SPU does not support JIT today. It could support JIT at some
-    point.
-   */
-  virtual       TargetJITInfo    *getJITInfo() {
-    return NULL;
-  }
-
-  virtual const SPUTargetLowering *getTargetLowering() const {
-   return &TLInfo;
-  }
-
-  virtual const SPUSelectionDAGInfo* getSelectionDAGInfo() const {
-    return &TSInfo;
-  }
-
-  virtual const SPURegisterInfo *getRegisterInfo() const {
-    return &InstrInfo.getRegisterInfo();
-  }
-
-  virtual const DataLayout *getDataLayout() const {
-    return &DL;
-  }
-
-  virtual const InstrItineraryData *getInstrItineraryData() const {
-    return &InstrItins;
-  }
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
-
-  // Pass Pipeline Configuration
-  virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
-};
-
-} // end namespace llvm
-
-#endif
diff --git a/lib/Target/CellSPU/TargetInfo/CMakeLists.txt b/lib/Target/CellSPU/TargetInfo/CMakeLists.txt
deleted file mode 100644
index 6a98f95db664..000000000000
--- a/lib/Target/CellSPU/TargetInfo/CMakeLists.txt
+++ /dev/null
@@ -1,7 +0,0 @@
-include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
-
-add_llvm_library(LLVMCellSPUInfo
-  CellSPUTargetInfo.cpp
-  )
-
-add_dependencies(LLVMCellSPUInfo CellSPUCommonTableGen)
diff --git a/lib/Target/CellSPU/TargetInfo/CellSPUTargetInfo.cpp b/lib/Target/CellSPU/TargetInfo/CellSPUTargetInfo.cpp
deleted file mode 100644
index 84aadfad6f8d..000000000000
--- a/lib/Target/CellSPU/TargetInfo/CellSPUTargetInfo.cpp
+++ /dev/null
@@ -1,20 +0,0 @@
-//===-- CellSPUTargetInfo.cpp - CellSPU Target Implementation -------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#include "SPU.h"
-#include "llvm/Module.h"
-#include "llvm/Support/TargetRegistry.h"
-using namespace llvm;
-
-Target llvm::TheCellSPUTarget;
-
-extern "C" void LLVMInitializeCellSPUTargetInfo() { 
-  RegisterTarget<Triple::cellspu> 
-    X(TheCellSPUTarget, "cellspu", "STI CBEA Cell SPU [experimental]");
-}
diff --git a/lib/Target/CppBackend/CPPBackend.cpp b/lib/Target/CppBackend/CPPBackend.cpp
index 5c909903f94b..3e69098edcc3 100644
--- a/lib/Target/CppBackend/CPPBackend.cpp
+++ b/lib/Target/CppBackend/CPPBackend.cpp
@@ -13,25 +13,25 @@
 //===----------------------------------------------------------------------===//
 
 #include "CPPTargetMachine.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/InlineAsm.h"
-#include "llvm/Instruction.h"
-#include "llvm/Instructions.h"
-#include "llvm/Module.h"
-#include "llvm/Pass.h"
-#include "llvm/PassManager.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/StringExtras.h"
+#include "llvm/Config/config.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/InlineAsm.h"
+#include "llvm/IR/Instruction.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Module.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCSubtargetInfo.h"
-#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/Pass.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormattedStream.h"
 #include "llvm/Support/TargetRegistry.h"
-#include "llvm/ADT/StringExtras.h"
-#include "llvm/Config/config.h"
 #include <algorithm>
 #include <cstdio>
 #include <map>
@@ -141,7 +141,7 @@ namespace {
     std::string getCppName(const Value* val);
     inline void printCppName(const Value* val);
 
-    void printAttributes(const AttrListPtr &PAL, const std::string &name);
+    void printAttributes(const AttributeSet &PAL, const std::string &name);
     void printType(Type* Ty);
     void printTypes(const Module* M);
 
@@ -464,24 +464,25 @@ void CppWriter::printCppName(const Value* val) {
   printEscapedString(getCppName(val));
 }
 
-void CppWriter::printAttributes(const AttrListPtr &PAL,
+void CppWriter::printAttributes(const AttributeSet &PAL,
                                 const std::string &name) {
-  Out << "AttrListPtr " << name << "_PAL;";
+  Out << "AttributeSet " << name << "_PAL;";
   nl(Out);
   if (!PAL.isEmpty()) {
     Out << '{'; in(); nl(Out);
-    Out << "SmallVector<AttributeWithIndex, 4> Attrs;"; nl(Out);
-    Out << "AttributeWithIndex PAWI;"; nl(Out);
+    Out << "SmallVector<AttributeSet, 4> Attrs;"; nl(Out);
+    Out << "AttributeSet PAS;"; in(); nl(Out);
     for (unsigned i = 0; i < PAL.getNumSlots(); ++i) {
-      unsigned index = PAL.getSlot(i).Index;
-      AttrBuilder attrs(PAL.getSlot(i).Attrs);
-      Out << "PAWI.Index = " << index << "U;\n";
-      Out << " {\n    AttrBuilder B;\n";
-
-#define HANDLE_ATTR(X)                                     \
-      if (attrs.hasAttribute(Attributes::X))               \
-        Out << "    B.addAttribute(Attributes::" #X ");\n"; \
-      attrs.removeAttribute(Attributes::X);
+      unsigned index = PAL.getSlotIndex(i);
+      AttrBuilder attrs(PAL.getSlotAttributes(i), index);
+      Out << "{"; in(); nl(Out);
+      Out << "AttrBuilder B;"; nl(Out);
+
+#define HANDLE_ATTR(X)                                                  \
+      if (attrs.contains(Attribute::X)) {                               \
+        Out << "B.addAttribute(Attribute::" #X ");"; nl(Out);           \
+        attrs.removeAttribute(Attribute::X);                            \
+      }
 
       HANDLE_ATTR(SExt);
       HANDLE_ATTR(ZExt);
@@ -499,6 +500,7 @@ void CppWriter::printAttributes(const AttrListPtr &PAL,
       HANDLE_ATTR(OptimizeForSize);
       HANDLE_ATTR(StackProtect);
       HANDLE_ATTR(StackProtectReq);
+      HANDLE_ATTR(StackProtectStrong);
       HANDLE_ATTR(NoCapture);
       HANDLE_ATTR(NoRedZone);
       HANDLE_ATTR(NoImplicitFloat);
@@ -509,16 +511,24 @@ void CppWriter::printAttributes(const AttrListPtr &PAL,
       HANDLE_ATTR(NonLazyBind);
       HANDLE_ATTR(MinSize);
 #undef HANDLE_ATTR
-      if (attrs.hasAttribute(Attributes::StackAlignment))
-        Out << "    B.addStackAlignmentAttr(" << attrs.getStackAlignment() << ")\n";
-      attrs.removeAttribute(Attributes::StackAlignment);
-      assert(!attrs.hasAttributes() && "Unhandled attribute!");
-      Out << "    PAWI.Attrs = Attributes::get(mod->getContext(), B);\n }";
-      nl(Out);
-      Out << "Attrs.push_back(PAWI);";
+
+      if (attrs.contains(Attribute::StackAlignment)) {
+        Out << "B.addStackAlignmentAttr(" << attrs.getStackAlignment()<<')';
+        nl(Out);
+        attrs.removeAttribute(Attribute::StackAlignment);
+      }
+
+      Out << "PAS = AttributeSet::get(mod->getContext(), ";
+      if (index == ~0U)
+        Out << "~0U,";
+      else
+        Out << index << "U,";
+      Out << " B);"; out(); nl(Out);
+      Out << "}"; out(); nl(Out);
       nl(Out);
+      Out << "Attrs.push_back(PAS);"; nl(Out);
     }
-    Out << name << "_PAL = AttrListPtr::get(mod->getContext(), Attrs);";
+    Out << name << "_PAL = AttributeSet::get(mod->getContext(), Attrs);";
     nl(Out);
     out(); nl(Out);
     Out << '}'; nl(Out);
@@ -1888,23 +1898,24 @@ void CppWriter::printModuleBody() {
 
 void CppWriter::printProgram(const std::string& fname,
                              const std::string& mName) {
-  Out << "#include <llvm/LLVMContext.h>\n";
-  Out << "#include <llvm/Module.h>\n";
-  Out << "#include <llvm/DerivedTypes.h>\n";
-  Out << "#include <llvm/Constants.h>\n";
-  Out << "#include <llvm/GlobalVariable.h>\n";
-  Out << "#include <llvm/Function.h>\n";
-  Out << "#include <llvm/CallingConv.h>\n";
-  Out << "#include <llvm/BasicBlock.h>\n";
-  Out << "#include <llvm/Instructions.h>\n";
-  Out << "#include <llvm/InlineAsm.h>\n";
-  Out << "#include <llvm/Support/FormattedStream.h>\n";
-  Out << "#include <llvm/Support/MathExtras.h>\n";
   Out << "#include <llvm/Pass.h>\n";
   Out << "#include <llvm/PassManager.h>\n";
+
   Out << "#include <llvm/ADT/SmallVector.h>\n";
   Out << "#include <llvm/Analysis/Verifier.h>\n";
   Out << "#include <llvm/Assembly/PrintModulePass.h>\n";
+  Out << "#include <llvm/IR/BasicBlock.h>\n";
+  Out << "#include <llvm/IR/CallingConv.h>\n";
+  Out << "#include <llvm/IR/Constants.h>\n";
+  Out << "#include <llvm/IR/DerivedTypes.h>\n";
+  Out << "#include <llvm/IR/Function.h>\n";
+  Out << "#include <llvm/IR/GlobalVariable.h>\n";
+  Out << "#include <llvm/IR/InlineAsm.h>\n";
+  Out << "#include <llvm/IR/Instructions.h>\n";
+  Out << "#include <llvm/IR/LLVMContext.h>\n";
+  Out << "#include <llvm/IR/Module.h>\n";
+  Out << "#include <llvm/Support/FormattedStream.h>\n";
+  Out << "#include <llvm/Support/MathExtras.h>\n";
   Out << "#include <algorithm>\n";
   Out << "using namespace llvm;\n\n";
   Out << "Module* " << fname << "();\n\n";
@@ -1941,14 +1952,6 @@ void CppWriter::printModule(const std::string& fname,
   }
   nl(Out);
 
-  // Loop over the dependent libraries and emit them.
-  Module::lib_iterator LI = TheModule->lib_begin();
-  Module::lib_iterator LE = TheModule->lib_end();
-  while (LI != LE) {
-    Out << "mod->addLibrary(\"" << *LI << "\");";
-    nl(Out);
-    ++LI;
-  }
   printModuleBody();
   nl(Out) << "return mod;";
   nl(Out,-1) << "}";
diff --git a/lib/Target/CppBackend/CPPTargetMachine.h b/lib/Target/CppBackend/CPPTargetMachine.h
index 30d765d6c9ce..477e788ee2fd 100644
--- a/lib/Target/CppBackend/CPPTargetMachine.h
+++ b/lib/Target/CppBackend/CPPTargetMachine.h
@@ -14,8 +14,8 @@
 #ifndef CPPTARGETMACHINE_H
 #define CPPTARGETMACHINE_H
 
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Target/TargetMachine.h"
-#include "llvm/DataLayout.h"
 
 namespace llvm {
 
diff --git a/lib/Target/CppBackend/TargetInfo/CppBackendTargetInfo.cpp b/lib/Target/CppBackend/TargetInfo/CppBackendTargetInfo.cpp
index a8ac0a282cd1..1ca74a4895c4 100644
--- a/lib/Target/CppBackend/TargetInfo/CppBackendTargetInfo.cpp
+++ b/lib/Target/CppBackend/TargetInfo/CppBackendTargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "CPPTargetMachine.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
diff --git a/lib/Target/Hexagon/CMakeLists.txt b/lib/Target/Hexagon/CMakeLists.txt
index 306084bb8c52..b5b887e7c7c8 100644
--- a/lib/Target/Hexagon/CMakeLists.txt
+++ b/lib/Target/Hexagon/CMakeLists.txt
@@ -9,6 +9,8 @@ tablegen(LLVM HexagonGenSubtargetInfo.inc -gen-subtarget)
 tablegen(LLVM HexagonGenDFAPacketizer.inc -gen-dfa-packetizer)
 add_public_tablegen_target(HexagonCommonTableGen)
 
+set(LLVM_COMMON_DEPENDS intrinsics_gen)
+
 add_llvm_target(HexagonCodeGen
   HexagonAsmPrinter.cpp
   HexagonCallingConvLower.cpp
@@ -16,6 +18,7 @@ add_llvm_target(HexagonCodeGen
   HexagonExpandPredSpillCode.cpp
   HexagonFrameLowering.cpp
   HexagonHardwareLoops.cpp
+  HexagonFixupHwLoops.cpp
   HexagonMachineScheduler.cpp
   HexagonMCInstLower.cpp
   HexagonInstrInfo.cpp
@@ -33,8 +36,6 @@ add_llvm_target(HexagonCodeGen
   HexagonNewValueJump.cpp
 )
 
-add_dependencies(LLVMHexagonCodeGen intrinsics_gen)
-
 add_subdirectory(TargetInfo)
 add_subdirectory(InstPrinter)
 add_subdirectory(MCTargetDesc)
diff --git a/lib/Target/Hexagon/Hexagon.h b/lib/Target/Hexagon/Hexagon.h
index 45f857bab8c6..dfbefc864283 100644
--- a/lib/Target/Hexagon/Hexagon.h
+++ b/lib/Target/Hexagon/Hexagon.h
@@ -21,14 +21,16 @@
 
 namespace llvm {
   class FunctionPass;
+  class ModulePass;
   class TargetMachine;
   class MachineInstr;
-  class MCInst;
+  class HexagonMCInst;
   class HexagonAsmPrinter;
   class HexagonTargetMachine;
   class raw_ostream;
 
-  FunctionPass *createHexagonISelDag(HexagonTargetMachine &TM);
+  FunctionPass *createHexagonISelDag(HexagonTargetMachine &TM,
+                                     CodeGenOpt::Level OptLevel);
   FunctionPass *createHexagonDelaySlotFillerPass(TargetMachine &TM);
   FunctionPass *createHexagonFPMoverPass(TargetMachine &TM);
   FunctionPass *createHexagonRemoveExtendOps(HexagonTargetMachine &TM);
@@ -53,7 +55,7 @@ namespace llvm {
   TargetAsmBackend *createHexagonAsmBackend(const Target &,
                                                   const std::string &);
 */
-  void HexagonLowerToMC(const MachineInstr *MI, MCInst &MCI,
+  void HexagonLowerToMC(const MachineInstr *MI, HexagonMCInst &MCI,
                         HexagonAsmPrinter &AP);
 } // end namespace llvm;
 
diff --git a/lib/Target/Hexagon/Hexagon.td b/lib/Target/Hexagon/Hexagon.td
index 451e56206e60..8a5ee40590bb 100644
--- a/lib/Target/Hexagon/Hexagon.td
+++ b/lib/Target/Hexagon/Hexagon.td
@@ -32,6 +32,107 @@ def ArchV5       : SubtargetFeature<"v5", "HexagonArchVersion", "V5",
                                     "Hexagon v5">;
 
 //===----------------------------------------------------------------------===//
+// Hexagon Instruction Predicate Definitions.
+//===----------------------------------------------------------------------===//
+def HasV2T                      : Predicate<"Subtarget.hasV2TOps()">;
+def HasV2TOnly                  : Predicate<"Subtarget.hasV2TOpsOnly()">;
+def NoV2T                       : Predicate<"!Subtarget.hasV2TOps()">;
+def HasV3T                      : Predicate<"Subtarget.hasV3TOps()">;
+def HasV3TOnly                  : Predicate<"Subtarget.hasV3TOpsOnly()">;
+def NoV3T                       : Predicate<"!Subtarget.hasV3TOps()">;
+def HasV4T                      : Predicate<"Subtarget.hasV4TOps()">;
+def NoV4T                       : Predicate<"!Subtarget.hasV4TOps()">;
+def HasV5T                      : Predicate<"Subtarget.hasV5TOps()">;
+def NoV5T                       : Predicate<"!Subtarget.hasV5TOps()">;
+def UseMEMOP                    : Predicate<"Subtarget.useMemOps()">;
+def IEEERndNearV5T              : Predicate<"Subtarget.modeIEEERndNear()">;
+
+//===----------------------------------------------------------------------===//
+// Classes used for relation maps.
+//===----------------------------------------------------------------------===//
+// PredRel - Filter class used to relate non-predicated instructions with their
+// predicated forms.
+class PredRel;
+// PredNewRel - Filter class used to relate predicated instructions with their
+// predicate-new forms.
+class PredNewRel: PredRel;
+// ImmRegRel - Filter class used to relate instructions having reg-reg form
+// with their reg-imm counterparts.
+class ImmRegRel;
+// NewValueRel - Filter class used to relate regular store instructions with
+// their new-value store form.
+class NewValueRel: PredNewRel;
+// NewValueRel - Filter class used to relate load/store instructions having
+// different addressing modes with each other.
+class AddrModeRel: NewValueRel;
+
+//===----------------------------------------------------------------------===//
+// Generate mapping table to relate non-predicate instructions with their
+// predicated formats - true and false.
+//
+
+def getPredOpcode : InstrMapping {
+  let FilterClass = "PredRel";
+  // Instructions with the same BaseOpcode and isNVStore values form a row.
+  let RowFields = ["BaseOpcode", "isNVStore", "PNewValue"];
+  // Instructions with the same predicate sense form a column.
+  let ColFields = ["PredSense"];
+  // The key column is the unpredicated instructions.
+  let KeyCol = [""];
+  // Value columns are PredSense=true and PredSense=false
+  let ValueCols = [["true"], ["false"]];
+}
+
+//===----------------------------------------------------------------------===//
+// Generate mapping table to relate predicated instructions with their .new
+// format.
+//
+def getPredNewOpcode : InstrMapping {
+  let FilterClass = "PredNewRel";
+  let RowFields = ["BaseOpcode", "PredSense", "isNVStore"];
+  let ColFields = ["PNewValue"];
+  let KeyCol = [""];
+  let ValueCols = [["new"]];
+}
+
+//===----------------------------------------------------------------------===//
+// Generate mapping table to relate store instructions with their new-value
+// format.
+//
+def getNewValueOpcode : InstrMapping {
+  let FilterClass = "NewValueRel";
+  let RowFields = ["BaseOpcode", "PredSense", "PNewValue"];
+  let ColFields = ["isNVStore"];
+  let KeyCol = ["0"];
+  let ValueCols = [["1"]];
+}
+
+def getBasedWithImmOffset : InstrMapping {
+  let FilterClass = "AddrModeRel";
+  let RowFields = ["CextOpcode", "PredSense", "PNewValue", "isNVStore",
+                   "isMEMri", "isFloat"];
+  let ColFields = ["addrMode"];
+  let KeyCol = ["Absolute"];
+  let ValueCols = [["BaseImmOffset"]];
+}
+
+def getBaseWithRegOffset : InstrMapping {
+  let FilterClass = "AddrModeRel";
+  let RowFields = ["CextOpcode", "PredSense", "PNewValue", "isNVStore"];
+  let ColFields = ["addrMode"];
+  let KeyCol = ["BaseImmOffset"];
+  let ValueCols = [["BaseRegOffset"]];
+}
+
+def getRegForm : InstrMapping {
+  let FilterClass = "ImmRegRel";
+  let RowFields = ["CextOpcode", "PredSense", "PNewValue"];
+  let ColFields = ["InputType"];
+  let KeyCol = ["imm"];
+  let ValueCols = [["reg"]];
+}
+
+//===----------------------------------------------------------------------===//
 // Register File, Calling Conv, Instruction Descriptions
 //===----------------------------------------------------------------------===//
 include "HexagonSchedule.td"
diff --git a/lib/Target/Hexagon/HexagonAsmPrinter.cpp b/lib/Target/Hexagon/HexagonAsmPrinter.cpp
index c15bce608f5e..88cd3fbacea0 100644
--- a/lib/Target/Hexagon/HexagonAsmPrinter.cpp
+++ b/lib/Target/Hexagon/HexagonAsmPrinter.cpp
@@ -17,20 +17,24 @@
 #include "Hexagon.h"
 #include "HexagonAsmPrinter.h"
 #include "HexagonMachineFunctionInfo.h"
-#include "HexagonMCInst.h"
 #include "HexagonTargetMachine.h"
 #include "HexagonSubtarget.h"
+#include "MCTargetDesc/HexagonMCInst.h"
 #include "InstPrinter/HexagonInstPrinter.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Module.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/Analysis/ConstantFolding.h"
 #include "llvm/Assembly/Writer.h"
 #include "llvm/CodeGen/AsmPrinter.h"
-#include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Module.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
@@ -38,22 +42,18 @@
 #include "llvm/MC/MCSection.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/Support/MathExtras.h"
 #include "llvm/Support/CommandLine.h"
-#include "llvm/Support/Debug.h"
 #include "llvm/Support/Compiler.h"
+#include "llvm/Support/Debug.h"
 #include "llvm/Support/Format.h"
-#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/MathExtras.h"
 #include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/Mangler.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetLoweringObjectFile.h"
-#include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetLoweringObjectFile.h"
 #include "llvm/Target/TargetOptions.h"
-#include "llvm/ADT/SmallString.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/StringExtras.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 
 using namespace llvm;
 
@@ -220,8 +220,8 @@ void HexagonAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     assert((Size+IgnoreCount) == MI->getBundleSize() && "Corrupt Bundle!");
     for (unsigned Index = 0; Index < Size; Index++) {
       HexagonMCInst MCI;
-      MCI.setStartPacket(Index == 0);
-      MCI.setEndPacket(Index == (Size-1));
+      MCI.setPacketStart(Index == 0);
+      MCI.setPacketEnd(Index == (Size-1));
 
       HexagonLowerToMC(BundleMIs[Index], MCI, *this);
       OutStreamer.EmitInstruction(MCI);
@@ -230,8 +230,8 @@ void HexagonAsmPrinter::EmitInstruction(const MachineInstr *MI) {
   else {
     HexagonMCInst MCI;
     if (MI->getOpcode() == Hexagon::ENDLOOP0) {
-      MCI.setStartPacket(true);
-      MCI.setEndPacket(true);
+      MCI.setPacketStart(true);
+      MCI.setPacketEnd(true);
     }
     HexagonLowerToMC(MI, MCI, *this);
     OutStreamer.EmitInstruction(MCI);
diff --git a/lib/Target/Hexagon/HexagonCFGOptimizer.cpp b/lib/Target/Hexagon/HexagonCFGOptimizer.cpp
index 9bca9e070709..d4078ad28b60 100644
--- a/lib/Target/Hexagon/HexagonCFGOptimizer.cpp
+++ b/lib/Target/Hexagon/HexagonCFGOptimizer.cpp
@@ -7,21 +7,22 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "hexagon_cfg"
-#include "HexagonTargetMachine.h"
-#include "HexagonSubtarget.h"
+#include "Hexagon.h"
 #include "HexagonMachineFunctionInfo.h"
+#include "HexagonSubtarget.h"
+#include "HexagonTargetMachine.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineLoopInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/Passes.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/MathExtras.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 
 using namespace llvm;
 
diff --git a/lib/Target/Hexagon/HexagonCallingConvLower.cpp b/lib/Target/Hexagon/HexagonCallingConvLower.cpp
index 73f9d9acab26..2c93d04f98e6 100644
--- a/lib/Target/Hexagon/HexagonCallingConvLower.cpp
+++ b/lib/Target/Hexagon/HexagonCallingConvLower.cpp
@@ -15,12 +15,12 @@
 
 #include "HexagonCallingConvLower.h"
 #include "Hexagon.h"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetMachine.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 using namespace llvm;
 
 Hexagon_CCState::Hexagon_CCState(CallingConv::ID CC, bool isVarArg,
diff --git a/lib/Target/Hexagon/HexagonCallingConvLower.h b/lib/Target/Hexagon/HexagonCallingConvLower.h
index 1f601e87ad68..489b3a3e5985 100644
--- a/lib/Target/Hexagon/HexagonCallingConvLower.h
+++ b/lib/Target/Hexagon/HexagonCallingConvLower.h
@@ -17,9 +17,9 @@
 #define LLVM_Hexagon_CODEGEN_CALLINGCONVLOWER_H
 
 #include "llvm/ADT/SmallVector.h"
-#include "llvm/CodeGen/ValueTypes.h"
-#include "llvm/CodeGen/SelectionDAGNodes.h"
 #include "llvm/CodeGen/CallingConvLower.h"
+#include "llvm/CodeGen/SelectionDAGNodes.h"
+#include "llvm/CodeGen/ValueTypes.h"
 
 //
 // Need to handle varargs.
diff --git a/lib/Target/Hexagon/HexagonExpandPredSpillCode.cpp b/lib/Target/Hexagon/HexagonExpandPredSpillCode.cpp
index ae2ca378881d..08144217fd30 100644
--- a/lib/Target/Hexagon/HexagonExpandPredSpillCode.cpp
+++ b/lib/Target/Hexagon/HexagonExpandPredSpillCode.cpp
@@ -17,9 +17,10 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "HexagonTargetMachine.h"
-#include "HexagonSubtarget.h"
+#include "Hexagon.h"
 #include "HexagonMachineFunctionInfo.h"
+#include "HexagonSubtarget.h"
+#include "HexagonTargetMachine.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/LatencyPriorityQueue.h"
 #include "llvm/CodeGen/MachineDominators.h"
@@ -30,12 +31,12 @@
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/ScheduleHazardRecognizer.h"
 #include "llvm/CodeGen/SchedulerRegistry.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/MathExtras.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 
 using namespace llvm;
 
diff --git a/lib/Target/Hexagon/HexagonFixupHwLoops.cpp b/lib/Target/Hexagon/HexagonFixupHwLoops.cpp
new file mode 100644
index 000000000000..240cc9566648
--- /dev/null
+++ b/lib/Target/Hexagon/HexagonFixupHwLoops.cpp
@@ -0,0 +1,183 @@
+//===---- HexagonFixupHwLoops.cpp - Fixup HW loops too far from LOOPn. ----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+// The loop start address in the LOOPn instruction is encoded as a distance
+// from the LOOPn instruction itself.  If the start address is too far from
+// the LOOPn instruction, the loop needs to be set up manually, i.e. via
+// direct transfers to SAn and LCn.
+// This pass will identify and convert such LOOPn instructions to a proper
+// form.
+//===----------------------------------------------------------------------===//
+
+
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/PassSupport.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "Hexagon.h"
+#include "HexagonTargetMachine.h"
+
+using namespace llvm;
+
+namespace llvm {
+  void initializeHexagonFixupHwLoopsPass(PassRegistry&);
+}
+
+namespace {
+  struct HexagonFixupHwLoops : public MachineFunctionPass {
+  public:
+    static char ID;
+
+    HexagonFixupHwLoops() : MachineFunctionPass(ID) {
+      initializeHexagonFixupHwLoopsPass(*PassRegistry::getPassRegistry());
+    }
+
+    virtual bool runOnMachineFunction(MachineFunction &MF);
+
+    const char *getPassName() const { return "Hexagon Hardware Loop Fixup"; }
+
+    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+      AU.setPreservesCFG();
+      MachineFunctionPass::getAnalysisUsage(AU);
+    }
+
+  private:
+    /// \brief Maximum distance between the loop instr and the basic block.
+    /// Just an estimate.
+    static const unsigned MAX_LOOP_DISTANCE = 200;
+
+    /// \brief Check the offset between each loop instruction and
+    /// the loop basic block to determine if we can use the LOOP instruction
+    /// or if we need to set the LC/SA registers explicitly.
+    bool fixupLoopInstrs(MachineFunction &MF);
+
+    /// \brief Add the instruction to set the LC and SA registers explicitly.
+    void convertLoopInstr(MachineFunction &MF,
+                          MachineBasicBlock::iterator &MII,
+                          RegScavenger &RS);
+
+  };
+
+  char HexagonFixupHwLoops::ID = 0;
+}
+
+INITIALIZE_PASS(HexagonFixupHwLoops, "hwloopsfixup",
+                "Hexagon Hardware Loops Fixup", false, false)
+
+FunctionPass *llvm::createHexagonFixupHwLoops() {
+  return new HexagonFixupHwLoops();
+}
+
+
+/// \brief Returns true if the instruction is a hardware loop instruction.
+static bool isHardwareLoop(const MachineInstr *MI) {
+  return MI->getOpcode() == Hexagon::LOOP0_r ||
+         MI->getOpcode() == Hexagon::LOOP0_i;
+}
+
+
+bool HexagonFixupHwLoops::runOnMachineFunction(MachineFunction &MF) {
+  bool Changed = fixupLoopInstrs(MF);
+  return Changed;
+}
+
+
+/// \brief For Hexagon, if the loop label is to far from the
+/// loop instruction then we need to set the LC0 and SA0 registers
+/// explicitly instead of using LOOP(start,count).  This function
+/// checks the distance, and generates register assignments if needed.
+///
+/// This function makes two passes over the basic blocks.  The first
+/// pass computes the offset of the basic block from the start.
+/// The second pass checks all the loop instructions.
+bool HexagonFixupHwLoops::fixupLoopInstrs(MachineFunction &MF) {
+
+  // Offset of the current instruction from the start.
+  unsigned InstOffset = 0;
+  // Map for each basic block to it's first instruction.
+  DenseMap<MachineBasicBlock*, unsigned> BlockToInstOffset;
+
+  // First pass - compute the offset of each basic block.
+  for (MachineFunction::iterator MBB = MF.begin(), MBBe = MF.end();
+       MBB != MBBe; ++MBB) {
+    BlockToInstOffset[MBB] = InstOffset;
+    InstOffset += (MBB->size() * 4);
+  }
+
+  // Second pass - check each loop instruction to see if it needs to
+  // be converted.
+  InstOffset = 0;
+  bool Changed = false;
+  RegScavenger RS;
+
+  // Loop over all the basic blocks.
+  for (MachineFunction::iterator MBB = MF.begin(), MBBe = MF.end();
+       MBB != MBBe; ++MBB) {
+    InstOffset = BlockToInstOffset[MBB];
+    RS.enterBasicBlock(MBB);
+
+    // Loop over all the instructions.
+    MachineBasicBlock::iterator MIE = MBB->end();
+    MachineBasicBlock::iterator MII = MBB->begin();
+    while (MII != MIE) {
+      if (isHardwareLoop(MII)) {
+        RS.forward(MII);
+        assert(MII->getOperand(0).isMBB() &&
+               "Expect a basic block as loop operand");
+        int Sub = InstOffset - BlockToInstOffset[MII->getOperand(0).getMBB()];
+        unsigned Dist = Sub > 0 ? Sub : -Sub;
+        if (Dist > MAX_LOOP_DISTANCE) {
+          // Convert to explicity setting LC0 and SA0.
+          convertLoopInstr(MF, MII, RS);
+          MII = MBB->erase(MII);
+          Changed = true;
+        } else {
+          ++MII;
+        }
+      } else {
+        ++MII;
+      }
+      InstOffset += 4;
+    }
+  }
+
+  return Changed;
+}
+
+
+/// \brief convert a loop instruction to a sequence of instructions that
+/// set the LC0 and SA0 register explicitly.
+void HexagonFixupHwLoops::convertLoopInstr(MachineFunction &MF,
+                                           MachineBasicBlock::iterator &MII,
+                                           RegScavenger &RS) {
+  const TargetInstrInfo *TII = MF.getTarget().getInstrInfo();
+  MachineBasicBlock *MBB = MII->getParent();
+  DebugLoc DL = MII->getDebugLoc();
+  unsigned Scratch = RS.scavengeRegister(&Hexagon::IntRegsRegClass, MII, 0);
+
+  // First, set the LC0 with the trip count.
+  if (MII->getOperand(1).isReg()) {
+    // Trip count is a register
+    BuildMI(*MBB, MII, DL, TII->get(Hexagon::TFCR), Hexagon::LC0)
+      .addReg(MII->getOperand(1).getReg());
+  } else {
+    // Trip count is an immediate.
+    BuildMI(*MBB, MII, DL, TII->get(Hexagon::TFRI), Scratch)
+      .addImm(MII->getOperand(1).getImm());
+    BuildMI(*MBB, MII, DL, TII->get(Hexagon::TFCR), Hexagon::LC0)
+      .addReg(Scratch);
+  }
+  // Then, set the SA0 with the loop start address.
+  BuildMI(*MBB, MII, DL, TII->get(Hexagon::CONST32_Label), Scratch)
+    .addMBB(MII->getOperand(0).getMBB());
+  BuildMI(*MBB, MII, DL, TII->get(Hexagon::TFCR), Hexagon::SA0)
+    .addReg(Scratch);
+}
diff --git a/lib/Target/Hexagon/HexagonFrameLowering.cpp b/lib/Target/Hexagon/HexagonFrameLowering.cpp
index cd682df7a574..d6a9329cd407 100644
--- a/lib/Target/Hexagon/HexagonFrameLowering.cpp
+++ b/lib/Target/Hexagon/HexagonFrameLowering.cpp
@@ -11,28 +11,28 @@
 #include "HexagonFrameLowering.h"
 #include "Hexagon.h"
 #include "HexagonInstrInfo.h"
+#include "HexagonMachineFunctionInfo.h"
 #include "HexagonRegisterInfo.h"
 #include "HexagonSubtarget.h"
 #include "HexagonTargetMachine.h"
-#include "HexagonMachineFunctionInfo.h"
-#include "llvm/Function.h"
-#include "llvm/Type.h"
 #include "llvm/ADT/BitVector.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/AsmPrinter.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/MC/MachineLocation.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Type.h"
 #include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MachineLocation.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetOptions.h"
-#include "llvm/Support/CommandLine.h"
 
 using namespace llvm;
 
@@ -166,7 +166,8 @@ bool HexagonFrameLowering::hasTailCall(MachineBasicBlock &MBB) const {
   MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
   unsigned RetOpcode = MBBI->getOpcode();
 
-  return RetOpcode == Hexagon::TCRETURNtg || RetOpcode == Hexagon::TCRETURNtext;}
+  return RetOpcode == Hexagon::TCRETURNtg || RetOpcode == Hexagon::TCRETURNtext;
+}
 
 void HexagonFrameLowering::emitEpilogue(MachineFunction &MF,
                                      MachineBasicBlock &MBB) const {
@@ -326,6 +327,21 @@ bool HexagonFrameLowering::restoreCalleeSavedRegisters(
   return true;
 }
 
+void HexagonFrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  MachineInstr &MI = *I;
+
+  if (MI.getOpcode() == Hexagon::ADJCALLSTACKDOWN) {
+    // Hexagon_TODO: add code
+  } else if (MI.getOpcode() == Hexagon::ADJCALLSTACKUP) {
+    // Hexagon_TODO: add code
+  } else {
+    llvm_unreachable("Cannot handle this call frame pseudo instruction");
+  }
+  MBB.erase(I);
+}
+
 int HexagonFrameLowering::getFrameIndexOffset(const MachineFunction &MF,
                                               int FI) const {
   return MF.getFrameInfo()->getObjectOffset(FI);
diff --git a/lib/Target/Hexagon/HexagonFrameLowering.h b/lib/Target/Hexagon/HexagonFrameLowering.h
index ad87f11e2457..a62c76aaf676 100644
--- a/lib/Target/Hexagon/HexagonFrameLowering.h
+++ b/lib/Target/Hexagon/HexagonFrameLowering.h
@@ -35,6 +35,11 @@ public:
                             MachineBasicBlock::iterator MI,
                             const std::vector<CalleeSavedInfo> &CSI,
                             const TargetRegisterInfo *TRI) const;
+
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I) const;
+
   virtual bool
   restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
                               MachineBasicBlock::iterator MI,
diff --git a/lib/Target/Hexagon/HexagonHardwareLoops.cpp b/lib/Target/Hexagon/HexagonHardwareLoops.cpp
index d756aec9bef9..178662447a7f 100644
--- a/lib/Target/Hexagon/HexagonHardwareLoops.cpp
+++ b/lib/Target/Hexagon/HexagonHardwareLoops.cpp
@@ -27,89 +27,202 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "hwloops"
-#include "Hexagon.h"
-#include "HexagonTargetMachine.h"
-#include "llvm/Constants.h"
-#include "llvm/PassSupport.h"
-#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/Statistic.h"
-#include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineLoopInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/PassSupport.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetInstrInfo.h"
+#include "Hexagon.h"
+#include "HexagonTargetMachine.h"
+
 #include <algorithm>
+#include <vector>
 
 using namespace llvm;
 
+#ifndef NDEBUG
+static cl::opt<int> HWLoopLimit("max-hwloop", cl::Hidden, cl::init(-1));
+#endif
+
 STATISTIC(NumHWLoops, "Number of loops converted to hardware loops");
 
+namespace llvm {
+  void initializeHexagonHardwareLoopsPass(PassRegistry&);
+}
+
 namespace {
   class CountValue;
   struct HexagonHardwareLoops : public MachineFunctionPass {
-    MachineLoopInfo       *MLI;
-    MachineRegisterInfo   *MRI;
-    const TargetInstrInfo *TII;
+    MachineLoopInfo            *MLI;
+    MachineRegisterInfo        *MRI;
+    MachineDominatorTree       *MDT;
+    const HexagonTargetMachine *TM;
+    const HexagonInstrInfo     *TII;
+    const HexagonRegisterInfo  *TRI;
+#ifndef NDEBUG
+    static int Counter;
+#endif
 
   public:
-    static char ID;   // Pass identification, replacement for typeid
+    static char ID;
 
-    HexagonHardwareLoops() : MachineFunctionPass(ID) {}
+    HexagonHardwareLoops() : MachineFunctionPass(ID) {
+      initializeHexagonHardwareLoopsPass(*PassRegistry::getPassRegistry());
+    }
 
     virtual bool runOnMachineFunction(MachineFunction &MF);
 
     const char *getPassName() const { return "Hexagon Hardware Loops"; }
 
     virtual void getAnalysisUsage(AnalysisUsage &AU) const {
-      AU.setPreservesCFG();
       AU.addRequired<MachineDominatorTree>();
-      AU.addPreserved<MachineDominatorTree>();
       AU.addRequired<MachineLoopInfo>();
-      AU.addPreserved<MachineLoopInfo>();
       MachineFunctionPass::getAnalysisUsage(AU);
     }
 
   private:
-    /// getCanonicalInductionVariable - Check to see if the loop has a canonical
-    /// induction variable.
-    /// Should be defined in MachineLoop. Based upon version in class Loop.
-    const MachineInstr *getCanonicalInductionVariable(MachineLoop *L) const;
-
-    /// getTripCount - Return a loop-invariant LLVM register indicating the
-    /// number of times the loop will be executed.  If the trip-count cannot
-    /// be determined, this return null.
-    CountValue *getTripCount(MachineLoop *L) const;
-
-    /// isInductionOperation - Return true if the instruction matches the
-    /// pattern for an opertion that defines an induction variable.
-    bool isInductionOperation(const MachineInstr *MI, unsigned IVReg) const;
+    /// Kinds of comparisons in the compare instructions.
+    struct Comparison {
+      enum Kind {
+        EQ  = 0x01,
+        NE  = 0x02,
+        L   = 0x04, // Less-than property.
+        G   = 0x08, // Greater-than property.
+        U   = 0x40, // Unsigned property.
+        LTs = L,
+        LEs = L | EQ,
+        GTs = G,
+        GEs = G | EQ,
+        LTu = L      | U,
+        LEu = L | EQ | U,
+        GTu = G      | U,
+        GEu = G | EQ | U
+      };
+
+      static Kind getSwappedComparison(Kind Cmp) {
+        assert ((!((Cmp & L) && (Cmp & G))) && "Malformed comparison operator");
+        if ((Cmp & L) || (Cmp & G))
+          return (Kind)(Cmp ^ (L|G));
+        return Cmp;
+      }
+    };
 
-    /// isInvalidOperation - Return true if the instruction is not valid within
-    /// a hardware loop.
+    /// \brief Find the register that contains the loop controlling
+    /// induction variable.
+    /// If successful, it will return true and set the \p Reg, \p IVBump
+    /// and \p IVOp arguments.  Otherwise it will return false.
+    /// The returned induction register is the register R that follows the
+    /// following induction pattern:
+    /// loop:
+    ///   R = phi ..., [ R.next, LatchBlock ]
+    ///   R.next = R + #bump
+    ///   if (R.next < #N) goto loop
+    /// IVBump is the immediate value added to R, and IVOp is the instruction
+    /// "R.next = R + #bump".
+    bool findInductionRegister(MachineLoop *L, unsigned &Reg,
+                               int64_t &IVBump, MachineInstr *&IVOp) const;
+
+    /// \brief Analyze the statements in a loop to determine if the loop
+    /// has a computable trip count and, if so, return a value that represents
+    /// the trip count expression.
+    CountValue *getLoopTripCount(MachineLoop *L,
+                                 SmallVector<MachineInstr*, 2> &OldInsts);
+
+    /// \brief Return the expression that represents the number of times
+    /// a loop iterates.  The function takes the operands that represent the
+    /// loop start value, loop end value, and induction value.  Based upon
+    /// these operands, the function attempts to compute the trip count.
+    /// If the trip count is not directly available (as an immediate value,
+    /// or a register), the function will attempt to insert computation of it
+    /// to the loop's preheader.
+    CountValue *computeCount(MachineLoop *Loop,
+                             const MachineOperand *Start,
+                             const MachineOperand *End,
+                             unsigned IVReg,
+                             int64_t IVBump,
+                             Comparison::Kind Cmp) const;
+
+    /// \brief Return true if the instruction is not valid within a hardware
+    /// loop.
     bool isInvalidLoopOperation(const MachineInstr *MI) const;
 
-    /// containsInavlidInstruction - Return true if the loop contains an
-    /// instruction that inhibits using the hardware loop.
+    /// \brief Return true if the loop contains an instruction that inhibits
+    /// using the hardware loop.
     bool containsInvalidInstruction(MachineLoop *L) const;
 
-    /// converToHardwareLoop - Given a loop, check if we can convert it to a
-    /// hardware loop.  If so, then perform the conversion and return true.
+    /// \brief Given a loop, check if we can convert it to a hardware loop.
+    /// If so, then perform the conversion and return true.
     bool convertToHardwareLoop(MachineLoop *L);
 
+    /// \brief Return true if the instruction is now dead.
+    bool isDead(const MachineInstr *MI,
+                SmallVector<MachineInstr*, 1> &DeadPhis) const;
+
+    /// \brief Remove the instruction if it is now dead.
+    void removeIfDead(MachineInstr *MI);
+
+    /// \brief Make sure that the "bump" instruction executes before the
+    /// compare.  We need that for the IV fixup, so that the compare
+    /// instruction would not use a bumped value that has not yet been
+    /// defined.  If the instructions are out of order, try to reorder them.
+    bool orderBumpCompare(MachineInstr *BumpI, MachineInstr *CmpI);
+
+    /// \brief Get the instruction that loads an immediate value into \p R,
+    /// or 0 if such an instruction does not exist.
+    MachineInstr *defWithImmediate(unsigned R);
+
+    /// \brief Get the immediate value referenced to by \p MO, either for
+    /// immediate operands, or for register operands, where the register
+    /// was defined with an immediate value.
+    int64_t getImmediate(MachineOperand &MO);
+
+    /// \brief Reset the given machine operand to now refer to a new immediate
+    /// value.  Assumes that the operand was already referencing an immediate
+    /// value, either directly, or via a register.
+    void setImmediate(MachineOperand &MO, int64_t Val);
+
+    /// \brief Fix the data flow of the induction varible.
+    /// The desired flow is: phi ---> bump -+-> comparison-in-latch.
+    ///                                     |
+    ///                                     +-> back to phi
+    /// where "bump" is the increment of the induction variable:
+    ///   iv = iv + #const.
+    /// Due to some prior code transformations, the actual flow may look
+    /// like this:
+    ///   phi -+-> bump ---> back to phi
+    ///        |
+    ///        +-> comparison-in-latch (against upper_bound-bump),
+    /// i.e. the comparison that controls the loop execution may be using
+    /// the value of the induction variable from before the increment.
+    ///
+    /// Return true if the loop's flow is the desired one (i.e. it's
+    /// either been fixed, or no fixing was necessary).
+    /// Otherwise, return false.  This can happen if the induction variable
+    /// couldn't be identified, or if the value in the latch's comparison
+    /// cannot be adjusted to reflect the post-bump value.
+    bool fixupInductionVariable(MachineLoop *L);
+
+    /// \brief Given a loop, if it does not have a preheader, create one.
+    /// Return the block that is the preheader.
+    MachineBasicBlock *createPreheaderForLoop(MachineLoop *L);
   };
 
   char HexagonHardwareLoops::ID = 0;
+#ifndef NDEBUG
+  int HexagonHardwareLoops::Counter = 0;
+#endif
 
-
-  // CountValue class - Abstraction for a trip count of a loop. A
-  // smaller vesrsion of the MachineOperand class without the concerns
-  // of changing the operand representation.
+  /// \brief Abstraction for a trip count of a loop. A smaller vesrsion
+  /// of the MachineOperand class without the concerns of changing the
+  /// operand representation.
   class CountValue {
   public:
     enum CountValueType {
@@ -119,101 +232,62 @@ namespace {
   private:
     CountValueType Kind;
     union Values {
-      unsigned RegNum;
-      int64_t ImmVal;
-      Values(unsigned r) : RegNum(r) {}
-      Values(int64_t i) : ImmVal(i) {}
+      struct {
+        unsigned Reg;
+        unsigned Sub;
+      } R;
+      unsigned ImmVal;
     } Contents;
-    bool isNegative;
 
   public:
-    CountValue(unsigned r, bool neg) : Kind(CV_Register), Contents(r),
-                                       isNegative(neg) {}
-    explicit CountValue(int64_t i) : Kind(CV_Immediate), Contents(i),
-                                     isNegative(i < 0) {}
-    CountValueType getType() const { return Kind; }
+    explicit CountValue(CountValueType t, unsigned v, unsigned u = 0) {
+      Kind = t;
+      if (Kind == CV_Register) {
+        Contents.R.Reg = v;
+        Contents.R.Sub = u;
+      } else {
+        Contents.ImmVal = v;
+      }
+    }
     bool isReg() const { return Kind == CV_Register; }
     bool isImm() const { return Kind == CV_Immediate; }
-    bool isNeg() const { return isNegative; }
 
     unsigned getReg() const {
       assert(isReg() && "Wrong CountValue accessor");
-      return Contents.RegNum;
+      return Contents.R.Reg;
     }
-    void setReg(unsigned Val) {
-      Contents.RegNum = Val;
+    unsigned getSubReg() const {
+      assert(isReg() && "Wrong CountValue accessor");
+      return Contents.R.Sub;
     }
-    int64_t getImm() const {
+    unsigned getImm() const {
       assert(isImm() && "Wrong CountValue accessor");
-      if (isNegative) {
-        return -Contents.ImmVal;
-      }
       return Contents.ImmVal;
     }
-    void setImm(int64_t Val) {
-      Contents.ImmVal = Val;
-    }
 
     void print(raw_ostream &OS, const TargetMachine *TM = 0) const {
-      if (isReg()) { OS << PrintReg(getReg()); }
-      if (isImm()) { OS << getImm(); }
-    }
-  };
-
-  struct HexagonFixupHwLoops : public MachineFunctionPass {
-  public:
-    static char ID;     // Pass identification, replacement for typeid.
-
-    HexagonFixupHwLoops() : MachineFunctionPass(ID) {}
-
-    virtual bool runOnMachineFunction(MachineFunction &MF);
-
-    const char *getPassName() const { return "Hexagon Hardware Loop Fixup"; }
-
-    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
-      AU.setPreservesCFG();
-      MachineFunctionPass::getAnalysisUsage(AU);
+      const TargetRegisterInfo *TRI = TM ? TM->getRegisterInfo() : 0;
+      if (isReg()) { OS << PrintReg(Contents.R.Reg, TRI, Contents.R.Sub); }
+      if (isImm()) { OS << Contents.ImmVal; }
     }
-
-  private:
-    /// Maximum distance between the loop instr and the basic block.
-    /// Just an estimate.
-    static const unsigned MAX_LOOP_DISTANCE = 200;
-
-    /// fixupLoopInstrs - Check the offset between each loop instruction and
-    /// the loop basic block to determine if we can use the LOOP instruction
-    /// or if we need to set the LC/SA registers explicitly.
-    bool fixupLoopInstrs(MachineFunction &MF);
-
-    /// convertLoopInstr - Add the instruction to set the LC and SA registers
-    /// explicitly.
-    void convertLoopInstr(MachineFunction &MF,
-                          MachineBasicBlock::iterator &MII,
-                          RegScavenger &RS);
-
   };
+} // end anonymous namespace
 
-  char HexagonFixupHwLoops::ID = 0;
 
-} // end anonymous namespace
+INITIALIZE_PASS_BEGIN(HexagonHardwareLoops, "hwloops",
+                      "Hexagon Hardware Loops", false, false)
+INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
+INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)
+INITIALIZE_PASS_END(HexagonHardwareLoops, "hwloops",
+                    "Hexagon Hardware Loops", false, false)
 
 
-/// isHardwareLoop - Returns true if the instruction is a hardware loop
-/// instruction.
+/// \brief Returns true if the instruction is a hardware loop instruction.
 static bool isHardwareLoop(const MachineInstr *MI) {
   return MI->getOpcode() == Hexagon::LOOP0_r ||
     MI->getOpcode() == Hexagon::LOOP0_i;
 }
 
-/// isCompareEquals - Returns true if the instruction is a compare equals
-/// instruction with an immediate operand.
-static bool isCompareEqualsImm(const MachineInstr *MI) {
-  return MI->getOpcode() == Hexagon::CMPEQri;
-}
-
-
-/// createHexagonHardwareLoops - Factory for creating
-/// the hardware loop phase.
 FunctionPass *llvm::createHexagonHardwareLoops() {
   return new HexagonHardwareLoops();
 }
@@ -224,45 +298,149 @@ bool HexagonHardwareLoops::runOnMachineFunction(MachineFunction &MF) {
 
   bool Changed = false;
 
-  // get the loop information
   MLI = &getAnalysis<MachineLoopInfo>();
-  // get the register information
   MRI = &MF.getRegInfo();
-  // the target specific instructio info.
-  TII = MF.getTarget().getInstrInfo();
+  MDT = &getAnalysis<MachineDominatorTree>();
+  TM  = static_cast<const HexagonTargetMachine*>(&MF.getTarget());
+  TII = static_cast<const HexagonInstrInfo*>(TM->getInstrInfo());
+  TRI = static_cast<const HexagonRegisterInfo*>(TM->getRegisterInfo());
 
   for (MachineLoopInfo::iterator I = MLI->begin(), E = MLI->end();
        I != E; ++I) {
     MachineLoop *L = *I;
-    if (!L->getParentLoop()) {
+    if (!L->getParentLoop())
       Changed |= convertToHardwareLoop(L);
-    }
   }
 
   return Changed;
 }
 
-/// getCanonicalInductionVariable - Check to see if the loop has a canonical
-/// induction variable. We check for a simple recurrence pattern - an
-/// integer recurrence that decrements by one each time through the loop and
-/// ends at zero.  If so, return the phi node that corresponds to it.
-///
-/// Based upon the similar code in LoopInfo except this code is specific to
-/// the machine.
-/// This method assumes that the IndVarSimplify pass has been run by 'opt'.
+
+bool HexagonHardwareLoops::findInductionRegister(MachineLoop *L,
+                                                 unsigned &Reg,
+                                                 int64_t &IVBump,
+                                                 MachineInstr *&IVOp
+                                                 ) const {
+  MachineBasicBlock *Header = L->getHeader();
+  MachineBasicBlock *Preheader = L->getLoopPreheader();
+  MachineBasicBlock *Latch = L->getLoopLatch();
+  if (!Header || !Preheader || !Latch)
+    return false;
+
+  // This pair represents an induction register together with an immediate
+  // value that will be added to it in each loop iteration.
+  typedef std::pair<unsigned,int64_t> RegisterBump;
+
+  // Mapping:  R.next -> (R, bump), where R, R.next and bump are derived
+  // from an induction operation
+  //   R.next = R + bump
+  // where bump is an immediate value.
+  typedef std::map<unsigned,RegisterBump> InductionMap;
+
+  InductionMap IndMap;
+
+  typedef MachineBasicBlock::instr_iterator instr_iterator;
+  for (instr_iterator I = Header->instr_begin(), E = Header->instr_end();
+       I != E && I->isPHI(); ++I) {
+    MachineInstr *Phi = &*I;
+
+    // Have a PHI instruction.  Get the operand that corresponds to the
+    // latch block, and see if is a result of an addition of form "reg+imm",
+    // where the "reg" is defined by the PHI node we are looking at.
+    for (unsigned i = 1, n = Phi->getNumOperands(); i < n; i += 2) {
+      if (Phi->getOperand(i+1).getMBB() != Latch)
+        continue;
+
+      unsigned PhiOpReg = Phi->getOperand(i).getReg();
+      MachineInstr *DI = MRI->getVRegDef(PhiOpReg);
+      unsigned UpdOpc = DI->getOpcode();
+      bool isAdd = (UpdOpc == Hexagon::ADD_ri);
+
+      if (isAdd) {
+        // If the register operand to the add is the PHI we're
+        // looking at, this meets the induction pattern.
+        unsigned IndReg = DI->getOperand(1).getReg();
+        if (MRI->getVRegDef(IndReg) == Phi) {
+          unsigned UpdReg = DI->getOperand(0).getReg();
+          int64_t V = DI->getOperand(2).getImm();
+          IndMap.insert(std::make_pair(UpdReg, std::make_pair(IndReg, V)));
+        }
+      }
+    }  // for (i)
+  }  // for (instr)
+
+  SmallVector<MachineOperand,2> Cond;
+  MachineBasicBlock *TB = 0, *FB = 0;
+  bool NotAnalyzed = TII->AnalyzeBranch(*Latch, TB, FB, Cond, false);
+  if (NotAnalyzed)
+    return false;
+
+  unsigned CSz = Cond.size();
+  assert (CSz == 1 || CSz == 2);
+  unsigned PredR = Cond[CSz-1].getReg();
+
+  MachineInstr *PredI = MRI->getVRegDef(PredR);
+  if (!PredI->isCompare())
+    return false;
+
+  unsigned CmpReg1 = 0, CmpReg2 = 0;
+  int CmpImm = 0, CmpMask = 0;
+  bool CmpAnalyzed = TII->analyzeCompare(PredI, CmpReg1, CmpReg2,
+                                         CmpMask, CmpImm);
+  // Fail if the compare was not analyzed, or it's not comparing a register
+  // with an immediate value.  Not checking the mask here, since we handle
+  // the individual compare opcodes (including CMPb) later on.
+  if (!CmpAnalyzed)
+    return false;
+
+  // Exactly one of the input registers to the comparison should be among
+  // the induction registers.
+  InductionMap::iterator IndMapEnd = IndMap.end();
+  InductionMap::iterator F = IndMapEnd;
+  if (CmpReg1 != 0) {
+    InductionMap::iterator F1 = IndMap.find(CmpReg1);
+    if (F1 != IndMapEnd)
+      F = F1;
+  }
+  if (CmpReg2 != 0) {
+    InductionMap::iterator F2 = IndMap.find(CmpReg2);
+    if (F2 != IndMapEnd) {
+      if (F != IndMapEnd)
+        return false;
+      F = F2;
+    }
+  }
+  if (F == IndMapEnd)
+    return false;
+
+  Reg = F->second.first;
+  IVBump = F->second.second;
+  IVOp = MRI->getVRegDef(F->first);
+  return true;
+}
+
+
+/// \brief Analyze the statements in a loop to determine if the loop has
+/// a computable trip count and, if so, return a value that represents
+/// the trip count expression.
 ///
-const MachineInstr
-*HexagonHardwareLoops::getCanonicalInductionVariable(MachineLoop *L) const {
+/// This function iterates over the phi nodes in the loop to check for
+/// induction variable patterns that are used in the calculation for
+/// the number of time the loop is executed.
+CountValue *HexagonHardwareLoops::getLoopTripCount(MachineLoop *L,
+                                SmallVector<MachineInstr*, 2> &OldInsts) {
   MachineBasicBlock *TopMBB = L->getTopBlock();
   MachineBasicBlock::pred_iterator PI = TopMBB->pred_begin();
   assert(PI != TopMBB->pred_end() &&
          "Loop must have more than one incoming edge!");
   MachineBasicBlock *Backedge = *PI++;
-  if (PI == TopMBB->pred_end()) return 0;  // dead loop
+  if (PI == TopMBB->pred_end())  // dead loop?
+    return 0;
   MachineBasicBlock *Incoming = *PI++;
-  if (PI != TopMBB->pred_end()) return 0;  // multiple backedges?
+  if (PI != TopMBB->pred_end())  // multiple backedges?
+    return 0;
 
-  // make sure there is one incoming and one backedge and determine which
+  // Make sure there is one incoming and one backedge and determine which
   // is which.
   if (L->contains(Incoming)) {
     if (L->contains(Backedge))
@@ -271,139 +449,433 @@ const MachineInstr
   } else if (!L->contains(Backedge))
     return 0;
 
-  // Loop over all of the PHI nodes, looking for a canonical induction variable:
-  //   - The PHI node is "reg1 = PHI reg2, BB1, reg3, BB2".
-  //   - The recurrence comes from the backedge.
-  //   - the definition is an induction operatio.n
-  for (MachineBasicBlock::iterator I = TopMBB->begin(), E = TopMBB->end();
-       I != E && I->isPHI(); ++I) {
-    const MachineInstr *MPhi = &*I;
-    unsigned DefReg = MPhi->getOperand(0).getReg();
-    for (unsigned i = 1; i != MPhi->getNumOperands(); i += 2) {
-      // Check each operand for the value from the backedge.
-      MachineBasicBlock *MBB = MPhi->getOperand(i+1).getMBB();
-      if (L->contains(MBB)) { // operands comes from the backedge
-        // Check if the definition is an induction operation.
-        const MachineInstr *DI = MRI->getVRegDef(MPhi->getOperand(i).getReg());
-        if (isInductionOperation(DI, DefReg)) {
-          return MPhi;
-        }
-      }
+  // Look for the cmp instruction to determine if we can get a useful trip
+  // count.  The trip count can be either a register or an immediate.  The
+  // location of the value depends upon the type (reg or imm).
+  MachineBasicBlock *Latch = L->getLoopLatch();
+  if (!Latch)
+    return 0;
+
+  unsigned IVReg = 0;
+  int64_t IVBump = 0;
+  MachineInstr *IVOp;
+  bool FoundIV = findInductionRegister(L, IVReg, IVBump, IVOp);
+  if (!FoundIV)
+    return 0;
+
+  MachineBasicBlock *Preheader = L->getLoopPreheader();
+
+  MachineOperand *InitialValue = 0;
+  MachineInstr *IV_Phi = MRI->getVRegDef(IVReg);
+  for (unsigned i = 1, n = IV_Phi->getNumOperands(); i < n; i += 2) {
+    MachineBasicBlock *MBB = IV_Phi->getOperand(i+1).getMBB();
+    if (MBB == Preheader)
+      InitialValue = &IV_Phi->getOperand(i);
+    else if (MBB == Latch)
+      IVReg = IV_Phi->getOperand(i).getReg();  // Want IV reg after bump.
+  }
+  if (!InitialValue)
+    return 0;
+
+  SmallVector<MachineOperand,2> Cond;
+  MachineBasicBlock *TB = 0, *FB = 0;
+  bool NotAnalyzed = TII->AnalyzeBranch(*Latch, TB, FB, Cond, false);
+  if (NotAnalyzed)
+    return 0;
+
+  MachineBasicBlock *Header = L->getHeader();
+  // TB must be non-null.  If FB is also non-null, one of them must be
+  // the header.  Otherwise, branch to TB could be exiting the loop, and
+  // the fall through can go to the header.
+  assert (TB && "Latch block without a branch?");
+  assert ((!FB || TB == Header || FB == Header) && "Branches not to header?");
+  if (!TB || (FB && TB != Header && FB != Header))
+    return 0;
+
+  // Branches of form "if (!P) ..." cause HexagonInstrInfo::AnalyzeBranch
+  // to put imm(0), followed by P in the vector Cond.
+  // If TB is not the header, it means that the "not-taken" path must lead
+  // to the header.
+  bool Negated = (Cond.size() > 1) ^ (TB != Header);
+  unsigned PredReg = Cond[Cond.size()-1].getReg();
+  MachineInstr *CondI = MRI->getVRegDef(PredReg);
+  unsigned CondOpc = CondI->getOpcode();
+
+  unsigned CmpReg1 = 0, CmpReg2 = 0;
+  int Mask = 0, ImmValue = 0;
+  bool AnalyzedCmp = TII->analyzeCompare(CondI, CmpReg1, CmpReg2,
+                                         Mask, ImmValue);
+  if (!AnalyzedCmp)
+    return 0;
+
+  // The comparison operator type determines how we compute the loop
+  // trip count.
+  OldInsts.push_back(CondI);
+  OldInsts.push_back(IVOp);
+
+  // Sadly, the following code gets information based on the position
+  // of the operands in the compare instruction.  This has to be done
+  // this way, because the comparisons check for a specific relationship
+  // between the operands (e.g. is-less-than), rather than to find out
+  // what relationship the operands are in (as on PPC).
+  Comparison::Kind Cmp;
+  bool isSwapped = false;
+  const MachineOperand &Op1 = CondI->getOperand(1);
+  const MachineOperand &Op2 = CondI->getOperand(2);
+  const MachineOperand *EndValue = 0;
+
+  if (Op1.isReg()) {
+    if (Op2.isImm() || Op1.getReg() == IVReg)
+      EndValue = &Op2;
+    else {
+      EndValue = &Op1;
+      isSwapped = true;
     }
   }
-  return 0;
-}
 
-/// getTripCount - Return a loop-invariant LLVM value indicating the
-/// number of times the loop will be executed.  The trip count can
-/// be either a register or a constant value.  If the trip-count
-/// cannot be determined, this returns null.
-///
-/// We find the trip count from the phi instruction that defines the
-/// induction variable.  We follow the links to the CMP instruction
-/// to get the trip count.
-///
-/// Based upon getTripCount in LoopInfo.
-///
-CountValue *HexagonHardwareLoops::getTripCount(MachineLoop *L) const {
-  // Check that the loop has a induction variable.
-  const MachineInstr *IV_Inst = getCanonicalInductionVariable(L);
-  if (IV_Inst == 0) return 0;
-
-  // Canonical loops will end with a 'cmpeq_ri IV, Imm',
-  //  if Imm is 0, get the count from the PHI opnd
-  //  if Imm is -M, than M is the count
-  //  Otherwise, Imm is the count
-  const MachineOperand *IV_Opnd;
-  const MachineOperand *InitialValue;
-  if (!L->contains(IV_Inst->getOperand(2).getMBB())) {
-    InitialValue = &IV_Inst->getOperand(1);
-    IV_Opnd = &IV_Inst->getOperand(3);
-  } else {
-    InitialValue = &IV_Inst->getOperand(3);
-    IV_Opnd = &IV_Inst->getOperand(1);
-  }
-
-  // Look for the cmp instruction to determine if we
-  // can get a useful trip count.  The trip count can
-  // be either a register or an immediate.  The location
-  // of the value depends upon the type (reg or imm).
-  for (MachineRegisterInfo::reg_iterator
-       RI = MRI->reg_begin(IV_Opnd->getReg()), RE = MRI->reg_end();
-       RI != RE; ++RI) {
-    IV_Opnd = &RI.getOperand();
-    const MachineInstr *MI = IV_Opnd->getParent();
-    if (L->contains(MI) && isCompareEqualsImm(MI)) {
-      const MachineOperand &MO = MI->getOperand(2);
-      assert(MO.isImm() && "IV Cmp Operand should be 0");
-      int64_t ImmVal = MO.getImm();
-
-      const MachineInstr *IV_DefInstr = MRI->getVRegDef(IV_Opnd->getReg());
-      assert(L->contains(IV_DefInstr->getParent()) &&
-             "IV definition should occurs in loop");
-      int64_t iv_value = IV_DefInstr->getOperand(2).getImm();
-
-      if (ImmVal == 0) {
-        // Make sure the induction variable changes by one on each iteration.
-        if (iv_value != 1 && iv_value != -1) {
+  if (!EndValue)
+    return 0;
+
+  switch (CondOpc) {
+    case Hexagon::CMPEQri:
+    case Hexagon::CMPEQrr:
+      Cmp = !Negated ? Comparison::EQ : Comparison::NE;
+      break;
+    case Hexagon::CMPLTrr:
+      Cmp = !Negated ? Comparison::LTs : Comparison::GEs;
+      break;
+    case Hexagon::CMPLTUrr:
+      Cmp = !Negated ? Comparison::LTu : Comparison::GEu;
+      break;
+    case Hexagon::CMPGTUri:
+    case Hexagon::CMPGTUrr:
+      Cmp = !Negated ? Comparison::GTu : Comparison::LEu;
+      break;
+    case Hexagon::CMPGTri:
+    case Hexagon::CMPGTrr:
+      Cmp = !Negated ? Comparison::GTs : Comparison::LEs;
+      break;
+    // Very limited support for byte/halfword compares.
+    case Hexagon::CMPbEQri_V4:
+    case Hexagon::CMPhEQri_V4: {
+      if (IVBump != 1)
+        return 0;
+
+      int64_t InitV, EndV;
+      // Since the comparisons are "ri", the EndValue should be an
+      // immediate.  Check it just in case.
+      assert(EndValue->isImm() && "Unrecognized latch comparison");
+      EndV = EndValue->getImm();
+      // Allow InitialValue to be a register defined with an immediate.
+      if (InitialValue->isReg()) {
+        if (!defWithImmediate(InitialValue->getReg()))
           return 0;
-        }
-        return new CountValue(InitialValue->getReg(), iv_value > 0);
+        InitV = getImmediate(*InitialValue);
       } else {
-        assert(InitialValue->isReg() && "Expecting register for init value");
-        const MachineInstr *DefInstr = MRI->getVRegDef(InitialValue->getReg());
-        if (DefInstr && DefInstr->getOpcode() == Hexagon::TFRI) {
-          int64_t count = ImmVal - DefInstr->getOperand(1).getImm();
-          if ((count % iv_value) != 0) {
-            return 0;
-          }
-          return new CountValue(count/iv_value);
-        }
+        assert(InitialValue->isImm());
+        InitV = InitialValue->getImm();
+      }
+      if (InitV >= EndV)
+        return 0;
+      if (CondOpc == Hexagon::CMPbEQri_V4) {
+        if (!isInt<8>(InitV) || !isInt<8>(EndV))
+          return 0;
+      } else {  // Hexagon::CMPhEQri_V4
+        if (!isInt<16>(InitV) || !isInt<16>(EndV))
+          return 0;
       }
+      Cmp = !Negated ? Comparison::EQ : Comparison::NE;
+      break;
     }
+    default:
+      return 0;
   }
-  return 0;
+
+  if (isSwapped)
+   Cmp = Comparison::getSwappedComparison(Cmp);
+
+  if (InitialValue->isReg()) {
+    unsigned R = InitialValue->getReg();
+    MachineBasicBlock *DefBB = MRI->getVRegDef(R)->getParent();
+    if (!MDT->properlyDominates(DefBB, Header))
+      return 0;
+    OldInsts.push_back(MRI->getVRegDef(R));
+  }
+  if (EndValue->isReg()) {
+    unsigned R = EndValue->getReg();
+    MachineBasicBlock *DefBB = MRI->getVRegDef(R)->getParent();
+    if (!MDT->properlyDominates(DefBB, Header))
+      return 0;
+  }
+
+  return computeCount(L, InitialValue, EndValue, IVReg, IVBump, Cmp);
 }
 
-/// isInductionOperation - return true if the operation is matches the
-/// pattern that defines an induction variable:
-///    add iv, c
-///
-bool
-HexagonHardwareLoops::isInductionOperation(const MachineInstr *MI,
-                                           unsigned IVReg) const {
-  return (MI->getOpcode() ==
-          Hexagon::ADD_ri && MI->getOperand(1).getReg() == IVReg);
+/// \brief Helper function that returns the expression that represents the
+/// number of times a loop iterates.  The function takes the operands that
+/// represent the loop start value, loop end value, and induction value.
+/// Based upon these operands, the function attempts to compute the trip count.
+CountValue *HexagonHardwareLoops::computeCount(MachineLoop *Loop,
+                                               const MachineOperand *Start,
+                                               const MachineOperand *End,
+                                               unsigned IVReg,
+                                               int64_t IVBump,
+                                               Comparison::Kind Cmp) const {
+  // Cannot handle comparison EQ, i.e. while (A == B).
+  if (Cmp == Comparison::EQ)
+    return 0;
+
+  // Check if either the start or end values are an assignment of an immediate.
+  // If so, use the immediate value rather than the register.
+  if (Start->isReg()) {
+    const MachineInstr *StartValInstr = MRI->getVRegDef(Start->getReg());
+    if (StartValInstr && StartValInstr->getOpcode() == Hexagon::TFRI)
+      Start = &StartValInstr->getOperand(1);
+  }
+  if (End->isReg()) {
+    const MachineInstr *EndValInstr = MRI->getVRegDef(End->getReg());
+    if (EndValInstr && EndValInstr->getOpcode() == Hexagon::TFRI)
+      End = &EndValInstr->getOperand(1);
+  }
+
+  assert (Start->isReg() || Start->isImm());
+  assert (End->isReg() || End->isImm());
+
+  bool CmpLess =     Cmp & Comparison::L;
+  bool CmpGreater =  Cmp & Comparison::G;
+  bool CmpHasEqual = Cmp & Comparison::EQ;
+
+  // Avoid certain wrap-arounds.  This doesn't detect all wrap-arounds.
+  // If loop executes while iv is "less" with the iv value going down, then
+  // the iv must wrap.
+  if (CmpLess && IVBump < 0)
+    return 0;
+  // If loop executes while iv is "greater" with the iv value going up, then
+  // the iv must wrap.
+  if (CmpGreater && IVBump > 0)
+    return 0;
+
+  if (Start->isImm() && End->isImm()) {
+    // Both, start and end are immediates.
+    int64_t StartV = Start->getImm();
+    int64_t EndV = End->getImm();
+    int64_t Dist = EndV - StartV;
+    if (Dist == 0)
+      return 0;
+
+    bool Exact = (Dist % IVBump) == 0;
+
+    if (Cmp == Comparison::NE) {
+      if (!Exact)
+        return 0;
+      if ((Dist < 0) ^ (IVBump < 0))
+        return 0;
+    }
+
+    // For comparisons that include the final value (i.e. include equality
+    // with the final value), we need to increase the distance by 1.
+    if (CmpHasEqual)
+      Dist = Dist > 0 ? Dist+1 : Dist-1;
+
+    // assert (CmpLess => Dist > 0);
+    assert ((!CmpLess || Dist > 0) && "Loop should never iterate!");
+    // assert (CmpGreater => Dist < 0);
+    assert ((!CmpGreater || Dist < 0) && "Loop should never iterate!");
+
+    // "Normalized" distance, i.e. with the bump set to +-1.
+    int64_t Dist1 = (IVBump > 0) ? (Dist +  (IVBump-1)) /   IVBump
+                               :  (-Dist + (-IVBump-1)) / (-IVBump);
+    assert (Dist1 > 0 && "Fishy thing.  Both operands have the same sign.");
+
+    uint64_t Count = Dist1;
+
+    if (Count > 0xFFFFFFFFULL)
+      return 0;
+
+    return new CountValue(CountValue::CV_Immediate, Count);
+  }
+
+  // A general case: Start and End are some values, but the actual
+  // iteration count may not be available.  If it is not, insert
+  // a computation of it into the preheader.
+
+  // If the induction variable bump is not a power of 2, quit.
+  // Othwerise we'd need a general integer division.
+  if (!isPowerOf2_64(abs64(IVBump)))
+    return 0;
+
+  MachineBasicBlock *PH = Loop->getLoopPreheader();
+  assert (PH && "Should have a preheader by now");
+  MachineBasicBlock::iterator InsertPos = PH->getFirstTerminator();
+  DebugLoc DL = (InsertPos != PH->end()) ? InsertPos->getDebugLoc()
+                                         : DebugLoc();
+
+  // If Start is an immediate and End is a register, the trip count
+  // will be "reg - imm".  Hexagon's "subtract immediate" instruction
+  // is actually "reg + -imm".
+
+  // If the loop IV is going downwards, i.e. if the bump is negative,
+  // then the iteration count (computed as End-Start) will need to be
+  // negated.  To avoid the negation, just swap Start and End.
+  if (IVBump < 0) {
+    std::swap(Start, End);
+    IVBump = -IVBump;
+  }
+  // Cmp may now have a wrong direction, e.g.  LEs may now be GEs.
+  // Signedness, and "including equality" are preserved.
+
+  bool RegToImm = Start->isReg() && End->isImm(); // for (reg..imm)
+  bool RegToReg = Start->isReg() && End->isReg(); // for (reg..reg)
+
+  int64_t StartV = 0, EndV = 0;
+  if (Start->isImm())
+    StartV = Start->getImm();
+  if (End->isImm())
+    EndV = End->getImm();
+
+  int64_t AdjV = 0;
+  // To compute the iteration count, we would need this computation:
+  //   Count = (End - Start + (IVBump-1)) / IVBump
+  // or, when CmpHasEqual:
+  //   Count = (End - Start + (IVBump-1)+1) / IVBump
+  // The "IVBump-1" part is the adjustment (AdjV).  We can avoid
+  // generating an instruction specifically to add it if we can adjust
+  // the immediate values for Start or End.
+
+  if (CmpHasEqual) {
+    // Need to add 1 to the total iteration count.
+    if (Start->isImm())
+      StartV--;
+    else if (End->isImm())
+      EndV++;
+    else
+      AdjV += 1;
+  }
+
+  if (Cmp != Comparison::NE) {
+    if (Start->isImm())
+      StartV -= (IVBump-1);
+    else if (End->isImm())
+      EndV += (IVBump-1);
+    else
+      AdjV += (IVBump-1);
+  }
+
+  unsigned R = 0, SR = 0;
+  if (Start->isReg()) {
+    R = Start->getReg();
+    SR = Start->getSubReg();
+  } else {
+    R = End->getReg();
+    SR = End->getSubReg();
+  }
+  const TargetRegisterClass *RC = MRI->getRegClass(R);
+  // Hardware loops cannot handle 64-bit registers.  If it's a double
+  // register, it has to have a subregister.
+  if (!SR && RC == &Hexagon::DoubleRegsRegClass)
+    return 0;
+  const TargetRegisterClass *IntRC = &Hexagon::IntRegsRegClass;
+
+  // Compute DistR (register with the distance between Start and End).
+  unsigned DistR, DistSR;
+
+  // Avoid special case, where the start value is an imm(0).
+  if (Start->isImm() && StartV == 0) {
+    DistR = End->getReg();
+    DistSR = End->getSubReg();
+  } else {
+    const MCInstrDesc &SubD = RegToReg ? TII->get(Hexagon::SUB_rr) :
+                              (RegToImm ? TII->get(Hexagon::SUB_ri) :
+                                          TII->get(Hexagon::ADD_ri));
+    unsigned SubR = MRI->createVirtualRegister(IntRC);
+    MachineInstrBuilder SubIB =
+      BuildMI(*PH, InsertPos, DL, SubD, SubR);
+
+    if (RegToReg) {
+      SubIB.addReg(End->getReg(), 0, End->getSubReg())
+           .addReg(Start->getReg(), 0, Start->getSubReg());
+    } else if (RegToImm) {
+      SubIB.addImm(EndV)
+           .addReg(Start->getReg(), 0, Start->getSubReg());
+    } else { // ImmToReg
+      SubIB.addReg(End->getReg(), 0, End->getSubReg())
+           .addImm(-StartV);
+    }
+    DistR = SubR;
+    DistSR = 0;
+  }
+
+  // From DistR, compute AdjR (register with the adjusted distance).
+  unsigned AdjR, AdjSR;
+
+  if (AdjV == 0) {
+    AdjR = DistR;
+    AdjSR = DistSR;
+  } else {
+    // Generate CountR = ADD DistR, AdjVal
+    unsigned AddR = MRI->createVirtualRegister(IntRC);
+    const MCInstrDesc &AddD = TII->get(Hexagon::ADD_ri);
+    BuildMI(*PH, InsertPos, DL, AddD, AddR)
+      .addReg(DistR, 0, DistSR)
+      .addImm(AdjV);
+
+    AdjR = AddR;
+    AdjSR = 0;
+  }
+
+  // From AdjR, compute CountR (register with the final count).
+  unsigned CountR, CountSR;
+
+  if (IVBump == 1) {
+    CountR = AdjR;
+    CountSR = AdjSR;
+  } else {
+    // The IV bump is a power of two. Log_2(IV bump) is the shift amount.
+    unsigned Shift = Log2_32(IVBump);
+
+    // Generate NormR = LSR DistR, Shift.
+    unsigned LsrR = MRI->createVirtualRegister(IntRC);
+    const MCInstrDesc &LsrD = TII->get(Hexagon::LSR_ri);
+    BuildMI(*PH, InsertPos, DL, LsrD, LsrR)
+      .addReg(AdjR, 0, AdjSR)
+      .addImm(Shift);
+
+    CountR = LsrR;
+    CountSR = 0;
+  }
+
+  return new CountValue(CountValue::CV_Register, CountR, CountSR);
 }
 
-/// isInvalidOperation - Return true if the operation is invalid within
-/// hardware loop.
-bool
-HexagonHardwareLoops::isInvalidLoopOperation(const MachineInstr *MI) const {
+
+/// \brief Return true if the operation is invalid within hardware loop.
+bool HexagonHardwareLoops::isInvalidLoopOperation(
+      const MachineInstr *MI) const {
 
   // call is not allowed because the callee may use a hardware loop
-  if (MI->getDesc().isCall()) {
+  if (MI->getDesc().isCall())
     return true;
-  }
+
   // do not allow nested hardware loops
-  if (isHardwareLoop(MI)) {
+  if (isHardwareLoop(MI))
     return true;
-  }
+
   // check if the instruction defines a hardware loop register
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     const MachineOperand &MO = MI->getOperand(i);
-    if (MO.isReg() && MO.isDef() &&
-        (MO.getReg() == Hexagon::LC0 || MO.getReg() == Hexagon::LC1 ||
-         MO.getReg() == Hexagon::SA0 || MO.getReg() == Hexagon::SA0)) {
+    if (!MO.isReg() || !MO.isDef())
+      continue;
+    unsigned R = MO.getReg();
+    if (R == Hexagon::LC0 || R == Hexagon::LC1 ||
+        R == Hexagon::SA0 || R == Hexagon::SA1)
       return true;
-    }
   }
   return false;
 }
 
-/// containsInvalidInstruction - Return true if the loop contains
-/// an instruction that inhibits the use of the hardware loop function.
-///
+
+/// \brief - Return true if the loop contains an instruction that inhibits
+/// the use of the hardware loop function.
 bool HexagonHardwareLoops::containsInvalidInstruction(MachineLoop *L) const {
   const std::vector<MachineBasicBlock*> Blocks = L->getBlocks();
   for (unsigned i = 0, e = Blocks.size(); i != e; ++i) {
@@ -411,126 +883,258 @@ bool HexagonHardwareLoops::containsInvalidInstruction(MachineLoop *L) const {
     for (MachineBasicBlock::iterator
            MII = MBB->begin(), E = MBB->end(); MII != E; ++MII) {
       const MachineInstr *MI = &*MII;
-      if (isInvalidLoopOperation(MI)) {
+      if (isInvalidLoopOperation(MI))
         return true;
-      }
     }
   }
   return false;
 }
 
-/// converToHardwareLoop - check if the loop is a candidate for
-/// converting to a hardware loop.  If so, then perform the
-/// transformation.
+
+/// \brief Returns true if the instruction is dead.  This was essentially
+/// copied from DeadMachineInstructionElim::isDead, but with special cases
+/// for inline asm, physical registers and instructions with side effects
+/// removed.
+bool HexagonHardwareLoops::isDead(const MachineInstr *MI,
+                             SmallVector<MachineInstr*, 1> &DeadPhis) const {
+  // Examine each operand.
+  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+    const MachineOperand &MO = MI->getOperand(i);
+    if (!MO.isReg() || !MO.isDef())
+      continue;
+
+    unsigned Reg = MO.getReg();
+    if (MRI->use_nodbg_empty(Reg))
+      continue;
+
+    typedef MachineRegisterInfo::use_nodbg_iterator use_nodbg_iterator;
+
+    // This instruction has users, but if the only user is the phi node for the
+    // parent block, and the only use of that phi node is this instruction, then
+    // this instruction is dead: both it (and the phi node) can be removed.
+    use_nodbg_iterator I = MRI->use_nodbg_begin(Reg);
+    use_nodbg_iterator End = MRI->use_nodbg_end();
+    if (llvm::next(I) != End || !I.getOperand().getParent()->isPHI())
+      return false;
+
+    MachineInstr *OnePhi = I.getOperand().getParent();
+    for (unsigned j = 0, f = OnePhi->getNumOperands(); j != f; ++j) {
+      const MachineOperand &OPO = OnePhi->getOperand(j);
+      if (!OPO.isReg() || !OPO.isDef())
+        continue;
+
+      unsigned OPReg = OPO.getReg();
+      use_nodbg_iterator nextJ;
+      for (use_nodbg_iterator J = MRI->use_nodbg_begin(OPReg);
+           J != End; J = nextJ) {
+        nextJ = llvm::next(J);
+        MachineOperand &Use = J.getOperand();
+        MachineInstr *UseMI = Use.getParent();
+
+        // If the phi node has a user that is not MI, bail...
+        if (MI != UseMI)
+          return false;
+      }
+    }
+    DeadPhis.push_back(OnePhi);
+  }
+
+  // If there are no defs with uses, the instruction is dead.
+  return true;
+}
+
+void HexagonHardwareLoops::removeIfDead(MachineInstr *MI) {
+  // This procedure was essentially copied from DeadMachineInstructionElim.
+
+  SmallVector<MachineInstr*, 1> DeadPhis;
+  if (isDead(MI, DeadPhis)) {
+    DEBUG(dbgs() << "HW looping will remove: " << *MI);
+
+    // It is possible that some DBG_VALUE instructions refer to this
+    // instruction.  Examine each def operand for such references;
+    // if found, mark the DBG_VALUE as undef (but don't delete it).
+    for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+      const MachineOperand &MO = MI->getOperand(i);
+      if (!MO.isReg() || !MO.isDef())
+        continue;
+      unsigned Reg = MO.getReg();
+      MachineRegisterInfo::use_iterator nextI;
+      for (MachineRegisterInfo::use_iterator I = MRI->use_begin(Reg),
+           E = MRI->use_end(); I != E; I = nextI) {
+        nextI = llvm::next(I);  // I is invalidated by the setReg
+        MachineOperand &Use = I.getOperand();
+        MachineInstr *UseMI = Use.getParent();
+        if (UseMI == MI)
+          continue;
+        if (Use.isDebug())
+          UseMI->getOperand(0).setReg(0U);
+        // This may also be a "instr -> phi -> instr" case which can
+        // be removed too.
+      }
+    }
+
+    MI->eraseFromParent();
+    for (unsigned i = 0; i < DeadPhis.size(); ++i)
+      DeadPhis[i]->eraseFromParent();
+  }
+}
+
+/// \brief Check if the loop is a candidate for converting to a hardware
+/// loop.  If so, then perform the transformation.
 ///
-/// This function works on innermost loops first.  A loop can
-/// be converted if it is a counting loop; either a register
-/// value or an immediate.
+/// This function works on innermost loops first.  A loop can be converted
+/// if it is a counting loop; either a register value or an immediate.
 ///
-/// The code makes several assumptions about the representation
-/// of the loop in llvm.
+/// The code makes several assumptions about the representation of the loop
+/// in llvm.
 bool HexagonHardwareLoops::convertToHardwareLoop(MachineLoop *L) {
+  // This is just for sanity.
+  assert(L->getHeader() && "Loop without a header?");
+
   bool Changed = false;
   // Process nested loops first.
-  for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I) {
+  for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I)
     Changed |= convertToHardwareLoop(*I);
-  }
+
   // If a nested loop has been converted, then we can't convert this loop.
-  if (Changed) {
+  if (Changed)
     return Changed;
+
+#ifndef NDEBUG
+  // Stop trying after reaching the limit (if any).
+  int Limit = HWLoopLimit;
+  if (Limit >= 0) {
+    if (Counter >= HWLoopLimit)
+      return false;
+    Counter++;
   }
-  // Are we able to determine the trip count for the loop?
-  CountValue *TripCount = getTripCount(L);
-  if (TripCount == 0) {
-    return false;
-  }
+#endif
+
   // Does the loop contain any invalid instructions?
-  if (containsInvalidInstruction(L)) {
+  if (containsInvalidInstruction(L))
     return false;
-  }
-  MachineBasicBlock *Preheader = L->getLoopPreheader();
-  // No preheader means there's not place for the loop instr.
-  if (Preheader == 0) {
+
+  // Is the induction variable bump feeding the latch condition?
+  if (!fixupInductionVariable(L))
     return false;
-  }
-  MachineBasicBlock::iterator InsertPos = Preheader->getFirstTerminator();
 
   MachineBasicBlock *LastMBB = L->getExitingBlock();
   // Don't generate hw loop if the loop has more than one exit.
-  if (LastMBB == 0) {
+  if (LastMBB == 0)
     return false;
-  }
+
   MachineBasicBlock::iterator LastI = LastMBB->getFirstTerminator();
+  if (LastI == LastMBB->end())
+    return false;
+
+  // Ensure the loop has a preheader: the loop instruction will be
+  // placed there.
+  bool NewPreheader = false;
+  MachineBasicBlock *Preheader = L->getLoopPreheader();
+  if (!Preheader) {
+    Preheader = createPreheaderForLoop(L);
+    if (!Preheader)
+      return false;
+    NewPreheader = true;
+  }
+  MachineBasicBlock::iterator InsertPos = Preheader->getFirstTerminator();
+
+  SmallVector<MachineInstr*, 2> OldInsts;
+  // Are we able to determine the trip count for the loop?
+  CountValue *TripCount = getLoopTripCount(L, OldInsts);
+  if (TripCount == 0)
+    return false;
+
+  // Is the trip count available in the preheader?
+  if (TripCount->isReg()) {
+    // There will be a use of the register inserted into the preheader,
+    // so make sure that the register is actually defined at that point.
+    MachineInstr *TCDef = MRI->getVRegDef(TripCount->getReg());
+    MachineBasicBlock *BBDef = TCDef->getParent();
+    if (!NewPreheader) {
+      if (!MDT->dominates(BBDef, Preheader))
+        return false;
+    } else {
+      // If we have just created a preheader, the dominator tree won't be
+      // aware of it.  Check if the definition of the register dominates
+      // the header, but is not the header itself.
+      if (!MDT->properlyDominates(BBDef, L->getHeader()))
+        return false;
+    }
+  }
 
   // Determine the loop start.
   MachineBasicBlock *LoopStart = L->getTopBlock();
   if (L->getLoopLatch() != LastMBB) {
     // When the exit and latch are not the same, use the latch block as the
     // start.
-    // The loop start address is used only after the 1st iteration, and the loop
-    // latch may contains instrs. that need to be executed after the 1st iter.
+    // The loop start address is used only after the 1st iteration, and the
+    // loop latch may contains instrs. that need to be executed after the
+    // first iteration.
     LoopStart = L->getLoopLatch();
     // Make sure the latch is a successor of the exit, otherwise it won't work.
-    if (!LastMBB->isSuccessor(LoopStart)) {
+    if (!LastMBB->isSuccessor(LoopStart))
       return false;
-    }
   }
 
-  // Convert the loop to a hardware loop
+  // Convert the loop to a hardware loop.
   DEBUG(dbgs() << "Change to hardware loop at "; L->dump());
+  DebugLoc DL;
+  if (InsertPos != Preheader->end())
+    DL = InsertPos->getDebugLoc();
 
   if (TripCount->isReg()) {
     // Create a copy of the loop count register.
-    MachineFunction *MF = LastMBB->getParent();
-    const TargetRegisterClass *RC =
-      MF->getRegInfo().getRegClass(TripCount->getReg());
-    unsigned CountReg = MF->getRegInfo().createVirtualRegister(RC);
-    BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
-            TII->get(TargetOpcode::COPY), CountReg).addReg(TripCount->getReg());
-    if (TripCount->isNeg()) {
-      unsigned CountReg1 = CountReg;
-      CountReg = MF->getRegInfo().createVirtualRegister(RC);
-      BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
-              TII->get(Hexagon::NEG), CountReg).addReg(CountReg1);
-    }
-
+    unsigned CountReg = MRI->createVirtualRegister(&Hexagon::IntRegsRegClass);
+    BuildMI(*Preheader, InsertPos, DL, TII->get(TargetOpcode::COPY), CountReg)
+      .addReg(TripCount->getReg(), 0, TripCount->getSubReg());
     // Add the Loop instruction to the beginning of the loop.
-    BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
-            TII->get(Hexagon::LOOP0_r)).addMBB(LoopStart).addReg(CountReg);
+    BuildMI(*Preheader, InsertPos, DL, TII->get(Hexagon::LOOP0_r))
+      .addMBB(LoopStart)
+      .addReg(CountReg);
   } else {
-    assert(TripCount->isImm() && "Expecting immedate vaule for trip count");
-    // Add the Loop immediate instruction to the beginning of the loop.
+    assert(TripCount->isImm() && "Expecting immediate value for trip count");
+    // Add the Loop immediate instruction to the beginning of the loop,
+    // if the immediate fits in the instructions.  Otherwise, we need to
+    // create a new virtual register.
     int64_t CountImm = TripCount->getImm();
-    BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
-            TII->get(Hexagon::LOOP0_i)).addMBB(LoopStart).addImm(CountImm);
+    if (!TII->isValidOffset(Hexagon::LOOP0_i, CountImm)) {
+      unsigned CountReg = MRI->createVirtualRegister(&Hexagon::IntRegsRegClass);
+      BuildMI(*Preheader, InsertPos, DL, TII->get(Hexagon::TFRI), CountReg)
+        .addImm(CountImm);
+      BuildMI(*Preheader, InsertPos, DL, TII->get(Hexagon::LOOP0_r))
+        .addMBB(LoopStart).addReg(CountReg);
+    } else
+      BuildMI(*Preheader, InsertPos, DL, TII->get(Hexagon::LOOP0_i))
+        .addMBB(LoopStart).addImm(CountImm);
   }
 
-  // Make sure the loop start always has a reference in the CFG.  We need to
-  // create a BlockAddress operand to get this mechanism to work both the
+  // Make sure the loop start always has a reference in the CFG.  We need
+  // to create a BlockAddress operand to get this mechanism to work both the
   // MachineBasicBlock and BasicBlock objects need the flag set.
   LoopStart->setHasAddressTaken();
   // This line is needed to set the hasAddressTaken flag on the BasicBlock
-  // object
+  // object.
   BlockAddress::get(const_cast<BasicBlock *>(LoopStart->getBasicBlock()));
 
   // Replace the loop branch with an endloop instruction.
-  DebugLoc dl = LastI->getDebugLoc();
-  BuildMI(*LastMBB, LastI, dl, TII->get(Hexagon::ENDLOOP0)).addMBB(LoopStart);
+  DebugLoc LastIDL = LastI->getDebugLoc();
+  BuildMI(*LastMBB, LastI, LastIDL,
+          TII->get(Hexagon::ENDLOOP0)).addMBB(LoopStart);
 
   // The loop ends with either:
   //  - a conditional branch followed by an unconditional branch, or
   //  - a conditional branch to the loop start.
   if (LastI->getOpcode() == Hexagon::JMP_c ||
       LastI->getOpcode() == Hexagon::JMP_cNot) {
-    // delete one and change/add an uncond. branch to out of the loop
+    // Delete one and change/add an uncond. branch to out of the loop.
     MachineBasicBlock *BranchTarget = LastI->getOperand(1).getMBB();
     LastI = LastMBB->erase(LastI);
     if (!L->contains(BranchTarget)) {
-      if (LastI != LastMBB->end()) {
-        TII->RemoveBranch(*LastMBB);
-      }
+      if (LastI != LastMBB->end())
+        LastI = LastMBB->erase(LastI);
       SmallVector<MachineOperand, 0> Cond;
-      TII->InsertBranch(*LastMBB, BranchTarget, 0, Cond, dl);
+      TII->InsertBranch(*LastMBB, BranchTarget, 0, Cond, LastIDL);
     }
   } else {
     // Conditional branch to loop start; just delete it.
@@ -538,110 +1142,413 @@ bool HexagonHardwareLoops::convertToHardwareLoop(MachineLoop *L) {
   }
   delete TripCount;
 
+  // The induction operation and the comparison may now be
+  // unneeded. If these are unneeded, then remove them.
+  for (unsigned i = 0; i < OldInsts.size(); ++i)
+    removeIfDead(OldInsts[i]);
+
   ++NumHWLoops;
   return true;
 }
 
-/// createHexagonFixupHwLoops - Factory for creating the hardware loop
-/// phase.
-FunctionPass *llvm::createHexagonFixupHwLoops() {
-  return new HexagonFixupHwLoops();
+
+bool HexagonHardwareLoops::orderBumpCompare(MachineInstr *BumpI,
+                                            MachineInstr *CmpI) {
+  assert (BumpI != CmpI && "Bump and compare in the same instruction?");
+
+  MachineBasicBlock *BB = BumpI->getParent();
+  if (CmpI->getParent() != BB)
+    return false;
+
+  typedef MachineBasicBlock::instr_iterator instr_iterator;
+  // Check if things are in order to begin with.
+  for (instr_iterator I = BumpI, E = BB->instr_end(); I != E; ++I)
+    if (&*I == CmpI)
+      return true;
+
+  // Out of order.
+  unsigned PredR = CmpI->getOperand(0).getReg();
+  bool FoundBump = false;
+  instr_iterator CmpIt = CmpI, NextIt = llvm::next(CmpIt);
+  for (instr_iterator I = NextIt, E = BB->instr_end(); I != E; ++I) {
+    MachineInstr *In = &*I;
+    for (unsigned i = 0, n = In->getNumOperands(); i < n; ++i) {
+      MachineOperand &MO = In->getOperand(i);
+      if (MO.isReg() && MO.isUse()) {
+        if (MO.getReg() == PredR)  // Found an intervening use of PredR.
+          return false;
+      }
+    }
+
+    if (In == BumpI) {
+      instr_iterator After = BumpI;
+      instr_iterator From = CmpI;
+      BB->splice(llvm::next(After), BB, From);
+      FoundBump = true;
+      break;
+    }
+  }
+  assert (FoundBump && "Cannot determine instruction order");
+  return FoundBump;
 }
 
-bool HexagonFixupHwLoops::runOnMachineFunction(MachineFunction &MF) {
-  DEBUG(dbgs() << "****** Hexagon Hardware Loop Fixup ******\n");
 
-  bool Changed = fixupLoopInstrs(MF);
-  return Changed;
+MachineInstr *HexagonHardwareLoops::defWithImmediate(unsigned R) {
+  MachineInstr *DI = MRI->getVRegDef(R);
+  unsigned DOpc = DI->getOpcode();
+  switch (DOpc) {
+    case Hexagon::TFRI:
+    case Hexagon::TFRI64:
+    case Hexagon::CONST32_Int_Real:
+    case Hexagon::CONST64_Int_Real:
+      return DI;
+  }
+  return 0;
 }
 
-/// fixupLoopInsts - For Hexagon, if the loop label is to far from the
-/// loop instruction then we need to set the LC0 and SA0 registers
-/// explicitly instead of using LOOP(start,count).  This function
-/// checks the distance, and generates register assignments if needed.
-///
-/// This function makes two passes over the basic blocks.  The first
-/// pass computes the offset of the basic block from the start.
-/// The second pass checks all the loop instructions.
-bool HexagonFixupHwLoops::fixupLoopInstrs(MachineFunction &MF) {
-
-  // Offset of the current instruction from the start.
-  unsigned InstOffset = 0;
-  // Map for each basic block to it's first instruction.
-  DenseMap<MachineBasicBlock*, unsigned> BlockToInstOffset;
-
-  // First pass - compute the offset of each basic block.
-  for (MachineFunction::iterator MBB = MF.begin(), MBBe = MF.end();
-       MBB != MBBe; ++MBB) {
-    BlockToInstOffset[MBB] = InstOffset;
-    InstOffset += (MBB->size() * 4);
-  }
-
-  // Second pass - check each loop instruction to see if it needs to
-  // be converted.
-  InstOffset = 0;
-  bool Changed = false;
-  RegScavenger RS;
-
-  // Loop over all the basic blocks.
-  for (MachineFunction::iterator MBB = MF.begin(), MBBe = MF.end();
-       MBB != MBBe; ++MBB) {
-    InstOffset = BlockToInstOffset[MBB];
-    RS.enterBasicBlock(MBB);
-
-    // Loop over all the instructions.
-    MachineBasicBlock::iterator MIE = MBB->end();
-    MachineBasicBlock::iterator MII = MBB->begin();
-    while (MII != MIE) {
-      if (isHardwareLoop(MII)) {
-        RS.forward(MII);
-        assert(MII->getOperand(0).isMBB() &&
-               "Expect a basic block as loop operand");
-        int diff = InstOffset - BlockToInstOffset[MII->getOperand(0).getMBB()];
-        diff = (diff > 0 ? diff : -diff);
-        if ((unsigned)diff > MAX_LOOP_DISTANCE) {
-          // Convert to explicity setting LC0 and SA0.
-          convertLoopInstr(MF, MII, RS);
-          MII = MBB->erase(MII);
-          Changed = true;
-        } else {
-          ++MII;
+
+int64_t HexagonHardwareLoops::getImmediate(MachineOperand &MO) {
+  if (MO.isImm())
+    return MO.getImm();
+  assert(MO.isReg());
+  unsigned R = MO.getReg();
+  MachineInstr *DI = defWithImmediate(R);
+  assert(DI && "Need an immediate operand");
+  // All currently supported "define-with-immediate" instructions have the
+  // actual immediate value in the operand(1).
+  int64_t v = DI->getOperand(1).getImm();
+  return v;
+}
+
+
+void HexagonHardwareLoops::setImmediate(MachineOperand &MO, int64_t Val) {
+  if (MO.isImm()) {
+    MO.setImm(Val);
+    return;
+  }
+
+  assert(MO.isReg());
+  unsigned R = MO.getReg();
+  MachineInstr *DI = defWithImmediate(R);
+  if (MRI->hasOneNonDBGUse(R)) {
+    // If R has only one use, then just change its defining instruction to
+    // the new immediate value.
+    DI->getOperand(1).setImm(Val);
+    return;
+  }
+
+  const TargetRegisterClass *RC = MRI->getRegClass(R);
+  unsigned NewR = MRI->createVirtualRegister(RC);
+  MachineBasicBlock &B = *DI->getParent();
+  DebugLoc DL = DI->getDebugLoc();
+  BuildMI(B, DI, DL, TII->get(DI->getOpcode()), NewR)
+    .addImm(Val);
+  MO.setReg(NewR);
+}
+
+
+bool HexagonHardwareLoops::fixupInductionVariable(MachineLoop *L) {
+  MachineBasicBlock *Header = L->getHeader();
+  MachineBasicBlock *Preheader = L->getLoopPreheader();
+  MachineBasicBlock *Latch = L->getLoopLatch();
+
+  if (!Header || !Preheader || !Latch)
+    return false;
+
+  // These data structures follow the same concept as the corresponding
+  // ones in findInductionRegister (where some comments are).
+  typedef std::pair<unsigned,int64_t> RegisterBump;
+  typedef std::pair<unsigned,RegisterBump> RegisterInduction;
+  typedef std::set<RegisterInduction> RegisterInductionSet;
+
+  // Register candidates for induction variables, with their associated bumps.
+  RegisterInductionSet IndRegs;
+
+  // Look for induction patterns:
+  //   vreg1 = PHI ..., [ latch, vreg2 ]
+  //   vreg2 = ADD vreg1, imm
+  typedef MachineBasicBlock::instr_iterator instr_iterator;
+  for (instr_iterator I = Header->instr_begin(), E = Header->instr_end();
+       I != E && I->isPHI(); ++I) {
+    MachineInstr *Phi = &*I;
+
+    // Have a PHI instruction.
+    for (unsigned i = 1, n = Phi->getNumOperands(); i < n; i += 2) {
+      if (Phi->getOperand(i+1).getMBB() != Latch)
+        continue;
+
+      unsigned PhiReg = Phi->getOperand(i).getReg();
+      MachineInstr *DI = MRI->getVRegDef(PhiReg);
+      unsigned UpdOpc = DI->getOpcode();
+      bool isAdd = (UpdOpc == Hexagon::ADD_ri);
+
+      if (isAdd) {
+        // If the register operand to the add/sub is the PHI we are looking
+        // at, this meets the induction pattern.
+        unsigned IndReg = DI->getOperand(1).getReg();
+        if (MRI->getVRegDef(IndReg) == Phi) {
+          unsigned UpdReg = DI->getOperand(0).getReg();
+          int64_t V = DI->getOperand(2).getImm();
+          IndRegs.insert(std::make_pair(UpdReg, std::make_pair(IndReg, V)));
         }
-      } else {
-        ++MII;
       }
-      InstOffset += 4;
+    }  // for (i)
+  }  // for (instr)
+
+  if (IndRegs.empty())
+    return false;
+
+  MachineBasicBlock *TB = 0, *FB = 0;
+  SmallVector<MachineOperand,2> Cond;
+  // AnalyzeBranch returns true if it fails to analyze branch.
+  bool NotAnalyzed = TII->AnalyzeBranch(*Latch, TB, FB, Cond, false);
+  if (NotAnalyzed)
+    return false;
+
+  // Check if the latch branch is unconditional.
+  if (Cond.empty())
+    return false;
+
+  if (TB != Header && FB != Header)
+    // The latch does not go back to the header.  Not a latch we know and love.
+    return false;
+
+  // Expecting a predicate register as a condition.  It won't be a hardware
+  // predicate register at this point yet, just a vreg.
+  // HexagonInstrInfo::AnalyzeBranch for negated branches inserts imm(0)
+  // into Cond, followed by the predicate register.  For non-negated branches
+  // it's just the register.
+  unsigned CSz = Cond.size();
+  if (CSz != 1 && CSz != 2)
+    return false;
+
+  unsigned P = Cond[CSz-1].getReg();
+  MachineInstr *PredDef = MRI->getVRegDef(P);
+
+  if (!PredDef->isCompare())
+    return false;
+
+  SmallSet<unsigned,2> CmpRegs;
+  MachineOperand *CmpImmOp = 0;
+
+  // Go over all operands to the compare and look for immediate and register
+  // operands.  Assume that if the compare has a single register use and a
+  // single immediate operand, then the register is being compared with the
+  // immediate value.
+  for (unsigned i = 0, n = PredDef->getNumOperands(); i < n; ++i) {
+    MachineOperand &MO = PredDef->getOperand(i);
+    if (MO.isReg()) {
+      // Skip all implicit references.  In one case there was:
+      //   %vreg140<def> = FCMPUGT32_rr %vreg138, %vreg139, %USR<imp-use>
+      if (MO.isImplicit())
+        continue;
+      if (MO.isUse()) {
+        unsigned R = MO.getReg();
+        if (!defWithImmediate(R)) {
+          CmpRegs.insert(MO.getReg());
+          continue;
+        }
+        // Consider the register to be the "immediate" operand.
+        if (CmpImmOp)
+          return false;
+        CmpImmOp = &MO;
+      }
+    } else if (MO.isImm()) {
+      if (CmpImmOp)    // A second immediate argument?  Confusing.  Bail out.
+        return false;
+      CmpImmOp = &MO;
     }
   }
 
-  return Changed;
+  if (CmpRegs.empty())
+    return false;
+
+  // Check if the compared register follows the order we want.  Fix if needed.
+  for (RegisterInductionSet::iterator I = IndRegs.begin(), E = IndRegs.end();
+       I != E; ++I) {
+    // This is a success.  If the register used in the comparison is one that
+    // we have identified as a bumped (updated) induction register, there is
+    // nothing to do.
+    if (CmpRegs.count(I->first))
+      return true;
+
+    // Otherwise, if the register being compared comes out of a PHI node,
+    // and has been recognized as following the induction pattern, and is
+    // compared against an immediate, we can fix it.
+    const RegisterBump &RB = I->second;
+    if (CmpRegs.count(RB.first)) {
+      if (!CmpImmOp)
+        return false;
+
+      int64_t CmpImm = getImmediate(*CmpImmOp);
+      int64_t V = RB.second;
+      if (V > 0 && CmpImm+V < CmpImm)  // Overflow (64-bit).
+        return false;
+      if (V < 0 && CmpImm+V > CmpImm)  // Overflow (64-bit).
+        return false;
+      CmpImm += V;
+      // Some forms of cmp-immediate allow u9 and s10.  Assume the worst case
+      // scenario, i.e. an 8-bit value.
+      if (CmpImmOp->isImm() && !isInt<8>(CmpImm))
+        return false;
+
+      // Make sure that the compare happens after the bump.  Otherwise,
+      // after the fixup, the compare would use a yet-undefined register.
+      MachineInstr *BumpI = MRI->getVRegDef(I->first);
+      bool Order = orderBumpCompare(BumpI, PredDef);
+      if (!Order)
+        return false;
+
+      // Finally, fix the compare instruction.
+      setImmediate(*CmpImmOp, CmpImm);
+      for (unsigned i = 0, n = PredDef->getNumOperands(); i < n; ++i) {
+        MachineOperand &MO = PredDef->getOperand(i);
+        if (MO.isReg() && MO.getReg() == RB.first) {
+          MO.setReg(I->first);
+          return true;
+        }
+      }
+    }
+  }
 
+  return false;
 }
 
-/// convertLoopInstr - convert a loop instruction to a sequence of instructions
-/// that set the lc and sa register explicitly.
-void HexagonFixupHwLoops::convertLoopInstr(MachineFunction &MF,
-                                           MachineBasicBlock::iterator &MII,
-                                           RegScavenger &RS) {
-  const TargetInstrInfo *TII = MF.getTarget().getInstrInfo();
-  MachineBasicBlock *MBB = MII->getParent();
-  DebugLoc DL = MII->getDebugLoc();
-  unsigned Scratch = RS.scavengeRegister(&Hexagon::IntRegsRegClass, MII, 0);
-
-  // First, set the LC0 with the trip count.
-  if (MII->getOperand(1).isReg()) {
-    // Trip count is a register
-    BuildMI(*MBB, MII, DL, TII->get(Hexagon::TFCR), Hexagon::LC0)
-      .addReg(MII->getOperand(1).getReg());
+
+/// \brief Create a preheader for a given loop.
+MachineBasicBlock *HexagonHardwareLoops::createPreheaderForLoop(
+      MachineLoop *L) {
+  if (MachineBasicBlock *TmpPH = L->getLoopPreheader())
+    return TmpPH;
+
+  MachineBasicBlock *Header = L->getHeader();
+  MachineBasicBlock *Latch = L->getLoopLatch();
+  MachineFunction *MF = Header->getParent();
+  DebugLoc DL;
+
+  if (!Latch || Header->hasAddressTaken())
+    return 0;
+
+  typedef MachineBasicBlock::instr_iterator instr_iterator;
+
+  // Verify that all existing predecessors have analyzable branches
+  // (or no branches at all).
+  typedef std::vector<MachineBasicBlock*> MBBVector;
+  MBBVector Preds(Header->pred_begin(), Header->pred_end());
+  SmallVector<MachineOperand,2> Tmp1;
+  MachineBasicBlock *TB = 0, *FB = 0;
+
+  if (TII->AnalyzeBranch(*Latch, TB, FB, Tmp1, false))
+    return 0;
+
+  for (MBBVector::iterator I = Preds.begin(), E = Preds.end(); I != E; ++I) {
+    MachineBasicBlock *PB = *I;
+    if (PB != Latch) {
+      bool NotAnalyzed = TII->AnalyzeBranch(*PB, TB, FB, Tmp1, false);
+      if (NotAnalyzed)
+        return 0;
+    }
+  }
+
+  MachineBasicBlock *NewPH = MF->CreateMachineBasicBlock();
+  MF->insert(Header, NewPH);
+
+  if (Header->pred_size() > 2) {
+    // Ensure that the header has only two predecessors: the preheader and
+    // the loop latch.  Any additional predecessors of the header should
+    // join at the newly created preheader.  Inspect all PHI nodes from the
+    // header and create appropriate corresponding PHI nodes in the preheader.
+
+    for (instr_iterator I = Header->instr_begin(), E = Header->instr_end();
+         I != E && I->isPHI(); ++I) {
+      MachineInstr *PN = &*I;
+
+      const MCInstrDesc &PD = TII->get(TargetOpcode::PHI);
+      MachineInstr *NewPN = MF->CreateMachineInstr(PD, DL);
+      NewPH->insert(NewPH->end(), NewPN);
+
+      unsigned PR = PN->getOperand(0).getReg();
+      const TargetRegisterClass *RC = MRI->getRegClass(PR);
+      unsigned NewPR = MRI->createVirtualRegister(RC);
+      NewPN->addOperand(MachineOperand::CreateReg(NewPR, true));
+
+      // Copy all non-latch operands of a header's PHI node to the newly
+      // created PHI node in the preheader.
+      for (unsigned i = 1, n = PN->getNumOperands(); i < n; i += 2) {
+        unsigned PredR = PN->getOperand(i).getReg();
+        MachineBasicBlock *PredB = PN->getOperand(i+1).getMBB();
+        if (PredB == Latch)
+          continue;
+
+        NewPN->addOperand(MachineOperand::CreateReg(PredR, false));
+        NewPN->addOperand(MachineOperand::CreateMBB(PredB));
+      }
+
+      // Remove copied operands from the old PHI node and add the value
+      // coming from the preheader's PHI.
+      for (int i = PN->getNumOperands()-2; i > 0; i -= 2) {
+        MachineBasicBlock *PredB = PN->getOperand(i+1).getMBB();
+        if (PredB != Latch) {
+          PN->RemoveOperand(i+1);
+          PN->RemoveOperand(i);
+        }
+      }
+      PN->addOperand(MachineOperand::CreateReg(NewPR, false));
+      PN->addOperand(MachineOperand::CreateMBB(NewPH));
+    }
+
   } else {
-    // Trip count is an immediate.
-    BuildMI(*MBB, MII, DL, TII->get(Hexagon::TFRI), Scratch)
-      .addImm(MII->getOperand(1).getImm());
-    BuildMI(*MBB, MII, DL, TII->get(Hexagon::TFCR), Hexagon::LC0)
-      .addReg(Scratch);
-  }
-  // Then, set the SA0 with the loop start address.
-  BuildMI(*MBB, MII, DL, TII->get(Hexagon::CONST32_Label), Scratch)
-    .addMBB(MII->getOperand(0).getMBB());
-  BuildMI(*MBB, MII, DL, TII->get(Hexagon::TFCR), Hexagon::SA0).addReg(Scratch);
+    assert(Header->pred_size() == 2);
+
+    // The header has only two predecessors, but the non-latch predecessor
+    // is not a preheader (e.g. it has other successors, etc.)
+    // In such a case we don't need any extra PHI nodes in the new preheader,
+    // all we need is to adjust existing PHIs in the header to now refer to
+    // the new preheader.
+    for (instr_iterator I = Header->instr_begin(), E = Header->instr_end();
+         I != E && I->isPHI(); ++I) {
+      MachineInstr *PN = &*I;
+      for (unsigned i = 1, n = PN->getNumOperands(); i < n; i += 2) {
+        MachineOperand &MO = PN->getOperand(i+1);
+        if (MO.getMBB() != Latch)
+          MO.setMBB(NewPH);
+      }
+    }
+  }
+
+  // "Reroute" the CFG edges to link in the new preheader.
+  // If any of the predecessors falls through to the header, insert a branch
+  // to the new preheader in that place.
+  SmallVector<MachineOperand,1> Tmp2;
+  SmallVector<MachineOperand,1> EmptyCond;
+
+  TB = FB = 0;
+
+  for (MBBVector::iterator I = Preds.begin(), E = Preds.end(); I != E; ++I) {
+    MachineBasicBlock *PB = *I;
+    if (PB != Latch) {
+      Tmp2.clear();
+      bool NotAnalyzed = TII->AnalyzeBranch(*PB, TB, FB, Tmp2, false);
+      (void)NotAnalyzed; // supress compiler warning
+      assert (!NotAnalyzed && "Should be analyzable!");
+      if (TB != Header && (Tmp2.empty() || FB != Header))
+        TII->InsertBranch(*PB, NewPH, 0, EmptyCond, DL);
+      PB->ReplaceUsesOfBlockWith(Header, NewPH);
+    }
+  }
+
+  // It can happen that the latch block will fall through into the header.
+  // Insert an unconditional branch to the header.
+  TB = FB = 0;
+  bool LatchNotAnalyzed = TII->AnalyzeBranch(*Latch, TB, FB, Tmp2, false);
+  (void)LatchNotAnalyzed; // supress compiler warning
+  assert (!LatchNotAnalyzed && "Should be analyzable!");
+  if (!TB && !FB)
+    TII->InsertBranch(*Latch, Header, 0, EmptyCond, DL);
+
+  // Finally, the branch from the preheader to the header.
+  TII->InsertBranch(*NewPH, Header, 0, EmptyCond, DL);
+  NewPH->addSuccessor(Header);
+
+  return NewPH;
 }
diff --git a/lib/Target/Hexagon/HexagonISelDAGToDAG.cpp b/lib/Target/Hexagon/HexagonISelDAGToDAG.cpp
index 5499134eb98b..8fc9ba1ee8cf 100644
--- a/lib/Target/Hexagon/HexagonISelDAGToDAG.cpp
+++ b/lib/Target/Hexagon/HexagonISelDAGToDAG.cpp
@@ -12,20 +12,32 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "hexagon-isel"
+#include "Hexagon.h"
 #include "HexagonISelLowering.h"
 #include "HexagonTargetMachine.h"
-#include "llvm/Intrinsics.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
-
 using namespace llvm;
 
+static
+cl::opt<unsigned>
+MaxNumOfUsesForConstExtenders("ga-max-num-uses-for-constant-extenders",
+  cl::Hidden, cl::init(2),
+  cl::desc("Maximum number of uses of a global address such that we still us a"
+           "constant extended instruction"));
 
 //===----------------------------------------------------------------------===//
 // Instruction Selector Implementation
 //===----------------------------------------------------------------------===//
 
+namespace llvm {
+  void initializeHexagonDAGToDAGISelPass(PassRegistry&);
+}
+
 //===--------------------------------------------------------------------===//
 /// HexagonDAGToDAGISel - Hexagon specific code to select Hexagon machine
 /// instructions for SelectionDAG operations.
@@ -39,19 +51,24 @@ class HexagonDAGToDAGISel : public SelectionDAGISel {
   // Keep a reference to HexagonTargetMachine.
   HexagonTargetMachine& TM;
   const HexagonInstrInfo *TII;
-
+  DenseMap<const GlobalValue *, unsigned> GlobalAddressUseCountMap;
 public:
-  explicit HexagonDAGToDAGISel(HexagonTargetMachine &targetmachine)
-    : SelectionDAGISel(targetmachine),
+  explicit HexagonDAGToDAGISel(HexagonTargetMachine &targetmachine,
+                               CodeGenOpt::Level OptLevel)
+    : SelectionDAGISel(targetmachine, OptLevel),
       Subtarget(targetmachine.getSubtarget<HexagonSubtarget>()),
       TM(targetmachine),
       TII(static_cast<const HexagonInstrInfo*>(TM.getInstrInfo())) {
-
+    initializeHexagonDAGToDAGISelPass(*PassRegistry::getPassRegistry());
   }
+  bool hasNumUsesBelowThresGA(SDNode *N) const;
 
   SDNode *Select(SDNode *N);
 
   // Complex Pattern Selectors.
+  inline bool foldGlobalAddress(SDValue &N, SDValue &R);
+  inline bool foldGlobalAddressGP(SDValue &N, SDValue &R);
+  bool foldGlobalAddressImpl(SDValue &N, SDValue &R, bool ShouldLookForGP);
   bool SelectADDRri(SDValue& N, SDValue &R1, SDValue &R2);
   bool SelectADDRriS11_0(SDValue& N, SDValue &R1, SDValue &R2);
   bool SelectADDRriS11_1(SDValue& N, SDValue &R1, SDValue &R2);
@@ -94,8 +111,56 @@ public:
   SDNode *SelectConstant(SDNode *N);
   SDNode *SelectConstantFP(SDNode *N);
   SDNode *SelectAdd(SDNode *N);
+  bool isConstExtProfitable(SDNode *N) const;
+
+// XformMskToBitPosU5Imm - Returns the bit position which
+// the single bit 32 bit mask represents.
+// Used in Clr and Set bit immediate memops.
+SDValue XformMskToBitPosU5Imm(uint32_t Imm) {
+  int32_t bitPos;
+  bitPos = Log2_32(Imm);
+  assert(bitPos >= 0 && bitPos < 32 &&
+         "Constant out of range for 32 BitPos Memops");
+  return CurDAG->getTargetConstant(bitPos, MVT::i32);
+}
+
+// XformMskToBitPosU4Imm - Returns the bit position which the single bit 16 bit
+// mask represents. Used in Clr and Set bit immediate memops.
+SDValue XformMskToBitPosU4Imm(uint16_t Imm) {
+  return XformMskToBitPosU5Imm(Imm);
+}
+
+// XformMskToBitPosU3Imm - Returns the bit position which the single bit 8 bit
+// mask represents. Used in Clr and Set bit immediate memops.
+SDValue XformMskToBitPosU3Imm(uint8_t Imm) {
+  return XformMskToBitPosU5Imm(Imm);
+}
+
+// Return true if there is exactly one bit set in V, i.e., if V is one of the
+// following integers: 2^0, 2^1, ..., 2^31.
+bool ImmIsSingleBit(uint32_t v) const {
+  uint32_t c = CountPopulation_64(v);
+  // Only return true if we counted 1 bit.
+  return c == 1;
+}
+
+// XformM5ToU5Imm - Return a target constant with the specified value, of type
+// i32 where the negative literal is transformed into a positive literal for
+// use in -= memops.
+inline SDValue XformM5ToU5Imm(signed Imm) {
+   assert( (Imm >= -31 && Imm <= -1)  && "Constant out of range for Memops");
+   return CurDAG->getTargetConstant( - Imm, MVT::i32);
+}
+
+
+// XformU7ToU7M1Imm - Return a target constant decremented by 1, in range
+// [1..128], used in cmpb.gtu instructions.
+inline SDValue XformU7ToU7M1Imm(signed Imm) {
+  assert((Imm >= 1 && Imm <= 128) && "Constant out of range for cmpb op");
+  return CurDAG->getTargetConstant(Imm - 1, MVT::i8);
+}
 
-  // Include the pieces autogenerated from the target description.
+// Include the pieces autogenerated from the target description.
 #include "HexagonGenDAGISel.inc"
 };
 }  // end anonymous namespace
@@ -104,10 +169,23 @@ public:
 /// createHexagonISelDag - This pass converts a legalized DAG into a
 /// Hexagon-specific DAG, ready for instruction scheduling.
 ///
-FunctionPass *llvm::createHexagonISelDag(HexagonTargetMachine &TM) {
-  return new HexagonDAGToDAGISel(TM);
+FunctionPass *llvm::createHexagonISelDag(HexagonTargetMachine &TM,
+                                         CodeGenOpt::Level OptLevel) {
+  return new HexagonDAGToDAGISel(TM, OptLevel);
 }
 
+static void initializePassOnce(PassRegistry &Registry) {
+  const char *Name = "Hexagon DAG->DAG Pattern Instruction Selection";
+  PassInfo *PI = new PassInfo(Name, "hexagon-isel",
+                              &SelectionDAGISel::ID, 0, false, false);
+  Registry.registerPass(*PI, true);
+}
+
+void llvm::initializeHexagonDAGToDAGISelPass(PassRegistry &Registry) {
+  CALL_ONCE_INITIALIZATION(initializePassOnce)
+}
+
+
 static bool IsS11_0_Offset(SDNode * S) {
     ConstantSDNode *N = cast<ConstantSDNode>(S);
 
@@ -606,8 +684,8 @@ SDNode *HexagonDAGToDAGISel::SelectIndexedStore(StoreSDNode *ST, DebugLoc dl) {
   // Offset value must be within representable range
   // and must have correct alignment properties.
   if (TII->isValidAutoIncImm(StoredVT, Val)) {
-    SDValue Ops[] = { Value, Base,
-                      CurDAG->getTargetConstant(Val, MVT::i32), Chain};
+    SDValue Ops[] = {Base, CurDAG->getTargetConstant(Val, MVT::i32), Value,
+                     Chain};
     unsigned Opcode = 0;
 
     // Figure out the post inc version of opcode.
@@ -1507,3 +1585,79 @@ SelectInlineAsmMemoryOperand(const SDValue &Op, char ConstraintCode,
   OutOps.push_back(Op1);
   return false;
 }
+
+bool HexagonDAGToDAGISel::isConstExtProfitable(SDNode *N) const {
+  unsigned UseCount = 0;
+  for (SDNode::use_iterator I = N->use_begin(), E = N->use_end(); I != E; ++I) {
+    UseCount++;
+  }
+
+  return (UseCount <= 1);
+
+}
+
+//===--------------------------------------------------------------------===//
+// Return 'true' if use count of the global address is below threshold.
+//===--------------------------------------------------------------------===//
+bool HexagonDAGToDAGISel::hasNumUsesBelowThresGA(SDNode *N) const {
+  assert(N->getOpcode() == ISD::TargetGlobalAddress &&
+         "Expecting a target global address");
+
+  // Always try to fold the address.
+  if (TM.getOptLevel() == CodeGenOpt::Aggressive)
+    return true;
+
+  GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(N);
+  DenseMap<const GlobalValue *, unsigned>::const_iterator GI =
+    GlobalAddressUseCountMap.find(GA->getGlobal());
+
+  if (GI == GlobalAddressUseCountMap.end())
+    return false;
+
+  return GI->second <= MaxNumOfUsesForConstExtenders;
+}
+
+//===--------------------------------------------------------------------===//
+// Return true if the non GP-relative global address can be folded.
+//===--------------------------------------------------------------------===//
+inline bool HexagonDAGToDAGISel::foldGlobalAddress(SDValue &N, SDValue &R) {
+  return foldGlobalAddressImpl(N, R, false);
+}
+
+//===--------------------------------------------------------------------===//
+// Return true if the GP-relative global address can be folded.
+//===--------------------------------------------------------------------===//
+inline bool HexagonDAGToDAGISel::foldGlobalAddressGP(SDValue &N, SDValue &R) {
+  return foldGlobalAddressImpl(N, R, true);
+}
+
+//===--------------------------------------------------------------------===//
+// Fold offset of the global address if number of uses are below threshold.
+//===--------------------------------------------------------------------===//
+bool HexagonDAGToDAGISel::foldGlobalAddressImpl(SDValue &N, SDValue &R,
+                                                bool ShouldLookForGP) {
+  if (N.getOpcode() == ISD::ADD) {
+    SDValue N0 = N.getOperand(0);
+    SDValue N1 = N.getOperand(1);
+    if ((ShouldLookForGP && (N0.getOpcode() == HexagonISD::CONST32_GP)) ||
+        (!ShouldLookForGP && (N0.getOpcode() == HexagonISD::CONST32))) {
+      ConstantSDNode *Const = dyn_cast<ConstantSDNode>(N1);
+      GlobalAddressSDNode *GA =
+        dyn_cast<GlobalAddressSDNode>(N0.getOperand(0));
+
+      if (Const && GA &&
+          (GA->getOpcode() == ISD::TargetGlobalAddress)) {
+        if ((N0.getOpcode() == HexagonISD::CONST32) &&
+                !hasNumUsesBelowThresGA(GA))
+            return false;
+        R = CurDAG->getTargetGlobalAddress(GA->getGlobal(),
+                                          Const->getDebugLoc(),
+                                          N.getValueType(),
+                                          GA->getOffset() +
+                                          (uint64_t)Const->getSExtValue());
+        return true;
+      }
+    }
+  }
+  return false;
+}
diff --git a/lib/Target/Hexagon/HexagonISelLowering.cpp b/lib/Target/Hexagon/HexagonISelLowering.cpp
index 1c891f14d8fe..15858a9368ae 100644
--- a/lib/Target/Hexagon/HexagonISelLowering.cpp
+++ b/lib/Target/Hexagon/HexagonISelLowering.cpp
@@ -13,17 +13,10 @@
 //===----------------------------------------------------------------------===//
 
 #include "HexagonISelLowering.h"
-#include "HexagonTargetMachine.h"
 #include "HexagonMachineFunctionInfo.h"
-#include "HexagonTargetObjectFile.h"
 #include "HexagonSubtarget.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/InlineAsm.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/GlobalAlias.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/CallingConv.h"
+#include "HexagonTargetMachine.h"
+#include "HexagonTargetObjectFile.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -32,6 +25,13 @@
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/InlineAsm.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
@@ -103,6 +103,16 @@ CC_Hexagon_VarArg (unsigned ValNo, MVT ValVT,
     State.addLoc(CCValAssign::getMem(ValNo, ValVT, ofst, LocVT, LocInfo));
     return false;
   }
+  if (LocVT == MVT::i1 || LocVT == MVT::i8 || LocVT == MVT::i16) {
+    LocVT = MVT::i32;
+    ValVT = MVT::i32;
+    if (ArgFlags.isSExt())
+      LocInfo = CCValAssign::SExt;
+    else if (ArgFlags.isZExt())
+      LocInfo = CCValAssign::ZExt;
+    else
+      LocInfo = CCValAssign::AExt;
+  }
   if (LocVT == MVT::i32 || LocVT == MVT::f32) {
     ofst = State.AllocateStack(4, 4);
     State.addLoc(CCValAssign::getMem(ValNo, ValVT, ofst, LocVT, LocInfo));
@@ -304,15 +314,9 @@ HexagonTargetLowering::LowerReturn(SDValue Chain,
   // Analyze return values of ISD::RET
   CCInfo.AnalyzeReturn(Outs, RetCC_Hexagon);
 
-  // If this is the first return lowered for this function, add the regs to the
-  // liveout set for the function.
-  if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
-    for (unsigned i = 0; i != RVLocs.size(); ++i)
-      if (RVLocs[i].isRegLoc())
-        DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
-  }
-
   SDValue Flag;
+  SmallVector<SDValue, 4> RetOps(1, Chain);
+
   // Copy the result values into the output registers.
   for (unsigned i = 0; i != RVLocs.size(); ++i) {
     CCValAssign &VA = RVLocs[i];
@@ -321,12 +325,17 @@ HexagonTargetLowering::LowerReturn(SDValue Chain,
 
     // Guarantee that all emitted copies are stuck together with flags.
     Flag = Chain.getValue(1);
+    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
   }
 
+  RetOps[0] = Chain;  // Update chain.
+
+  // Add the flag if we have it.
   if (Flag.getNode())
-    return DAG.getNode(HexagonISD::RET_FLAG, dl, MVT::Other, Chain, Flag);
+    RetOps.push_back(Flag);
 
-  return DAG.getNode(HexagonISD::RET_FLAG, dl, MVT::Other, Chain);
+  return DAG.getNode(HexagonISD::RET_FLAG, dl, MVT::Other,
+                     &RetOps[0], RetOps.size());
 }
 
 
@@ -608,7 +617,7 @@ static bool getIndexedAddressParts(SDNode *Ptr, EVT VT,
 
 // TODO: Put this function along with the other isS* functions in
 // HexagonISelDAGToDAG.cpp into a common file. Or better still, use the
-// functions defined in HexagonImmediates.td.
+// functions defined in HexagonOperands.td.
 static bool Is_PostInc_S4_Offset(SDNode * S, int ShiftAmount) {
   ConstantSDNode *N = cast<ConstantSDNode>(S);
 
@@ -1016,8 +1025,8 @@ SDValue HexagonTargetLowering::LowerGLOBALADDRESS(SDValue Op,
   DebugLoc dl = Op.getDebugLoc();
   Result = DAG.getTargetGlobalAddress(GV, dl, getPointerTy(), Offset);
 
-  HexagonTargetObjectFile &TLOF =
-    (HexagonTargetObjectFile&)getObjFileLowering();
+  const HexagonTargetObjectFile &TLOF =
+      static_cast<const HexagonTargetObjectFile &>(getObjFileLowering());
   if (TLOF.IsGlobalInSmallSection(GV, getTargetMachine())) {
     return DAG.getNode(HexagonISD::CONST32_GP, dl, getPointerTy(), Result);
   }
@@ -1025,6 +1034,14 @@ SDValue HexagonTargetLowering::LowerGLOBALADDRESS(SDValue Op,
   return DAG.getNode(HexagonISD::CONST32, dl, getPointerTy(), Result);
 }
 
+SDValue
+HexagonTargetLowering::LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const {
+  const BlockAddress *BA = cast<BlockAddressSDNode>(Op)->getBlockAddress();
+  SDValue BA_SD =  DAG.getTargetBlockAddress(BA, MVT::i32);
+  DebugLoc dl = Op.getDebugLoc();
+  return DAG.getNode(HexagonISD::CONST32_GP, dl, getPointerTy(), BA_SD);
+}
+
 //===----------------------------------------------------------------------===//
 // TargetLowering Implementation
 //===----------------------------------------------------------------------===//
@@ -1053,8 +1070,8 @@ HexagonTargetLowering::HexagonTargetLowering(HexagonTargetMachine
     setPrefLoopAlignment(4);
 
     // Limits for inline expansion of memcpy/memmove
-    maxStoresPerMemcpy = 6;
-    maxStoresPerMemmove = 6;
+    MaxStoresPerMemcpy = 6;
+    MaxStoresPerMemmove = 6;
 
     //
     // Library calls for unsupported operations
@@ -1298,6 +1315,7 @@ HexagonTargetLowering::HexagonTargetLowering(HexagonTargetMachine
     // Custom legalize GlobalAddress nodes into CONST32.
     setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
     setOperationAction(ISD::GlobalAddress, MVT::i8, Custom);
+    setOperationAction(ISD::BlockAddress, MVT::i32, Custom);
     // Truncate action?
     setOperationAction(ISD::TRUNCATE, MVT::i64, Expand);
 
@@ -1343,7 +1361,6 @@ HexagonTargetLowering::HexagonTargetLowering(HexagonTargetMachine
 
     }
 
-    setOperationAction(ISD::BR_CC, MVT::Other, Expand);
     setOperationAction(ISD::BRIND, MVT::Other, Expand);
     if (EmitJumpTables) {
       setOperationAction(ISD::BR_JT, MVT::Other, Custom);
@@ -1353,7 +1370,12 @@ HexagonTargetLowering::HexagonTargetLowering(HexagonTargetMachine
     // Increase jump tables cutover to 5, was 4.
     setMinimumJumpTableEntries(5);
 
+    setOperationAction(ISD::BR_CC, MVT::Other, Expand);
+    setOperationAction(ISD::BR_CC, MVT::f32, Expand);
+    setOperationAction(ISD::BR_CC, MVT::f64, Expand);
+    setOperationAction(ISD::BR_CC, MVT::i1,  Expand);
     setOperationAction(ISD::BR_CC, MVT::i32, Expand);
+    setOperationAction(ISD::BR_CC, MVT::i64, Expand);
 
     setOperationAction(ISD::MEMBARRIER, MVT::Other, Custom);
     setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom);
@@ -1364,11 +1386,41 @@ HexagonTargetLowering::HexagonTargetLowering(HexagonTargetMachine
     setOperationAction(ISD::FSIN , MVT::f32, Expand);
     setOperationAction(ISD::FCOS , MVT::f32, Expand);
     setOperationAction(ISD::FREM , MVT::f32, Expand);
+    setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
+    setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
+
+    // In V4, we have double word add/sub with carry. The problem with
+    // modelling this instruction is that it produces 2 results - Rdd and Px.
+    // To model update of Px, we will have to use Defs[p0..p3] which will
+    // cause any predicate live range to spill. So, we pretend we dont't
+    // have these instructions.
+    setOperationAction(ISD::ADDE, MVT::i8, Expand);
+    setOperationAction(ISD::ADDE, MVT::i16, Expand);
+    setOperationAction(ISD::ADDE, MVT::i32, Expand);
+    setOperationAction(ISD::ADDE, MVT::i64, Expand);
+    setOperationAction(ISD::SUBE, MVT::i8, Expand);
+    setOperationAction(ISD::SUBE, MVT::i16, Expand);
+    setOperationAction(ISD::SUBE, MVT::i32, Expand);
+    setOperationAction(ISD::SUBE, MVT::i64, Expand);
+    setOperationAction(ISD::ADDC, MVT::i8, Expand);
+    setOperationAction(ISD::ADDC, MVT::i16, Expand);
+    setOperationAction(ISD::ADDC, MVT::i32, Expand);
+    setOperationAction(ISD::ADDC, MVT::i64, Expand);
+    setOperationAction(ISD::SUBC, MVT::i8, Expand);
+    setOperationAction(ISD::SUBC, MVT::i16, Expand);
+    setOperationAction(ISD::SUBC, MVT::i32, Expand);
+    setOperationAction(ISD::SUBC, MVT::i64, Expand);
+
     setOperationAction(ISD::CTPOP, MVT::i32, Expand);
+    setOperationAction(ISD::CTPOP, MVT::i64, Expand);
     setOperationAction(ISD::CTTZ , MVT::i32, Expand);
+    setOperationAction(ISD::CTTZ , MVT::i64, Expand);
     setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i32, Expand);
+    setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i64, Expand);
     setOperationAction(ISD::CTLZ , MVT::i32, Expand);
+    setOperationAction(ISD::CTLZ , MVT::i64, Expand);
     setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i32, Expand);
+    setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i64, Expand);
     setOperationAction(ISD::ROTL , MVT::i32, Expand);
     setOperationAction(ISD::ROTR , MVT::i32, Expand);
     setOperationAction(ISD::BSWAP, MVT::i32, Expand);
@@ -1430,6 +1482,8 @@ HexagonTargetLowering::getTargetNodeName(unsigned Opcode) const {
   switch (Opcode) {
     default: return 0;
     case HexagonISD::CONST32:     return "HexagonISD::CONST32";
+    case HexagonISD::CONST32_GP: return "HexagonISD::CONST32_GP";
+    case HexagonISD::CONST32_Int_Real: return "HexagonISD::CONST32_Int_Real";
     case HexagonISD::ADJDYNALLOC: return "HexagonISD::ADJDYNALLOC";
     case HexagonISD::CMPICC:      return "HexagonISD::CMPICC";
     case HexagonISD::CMPFCC:      return "HexagonISD::CMPFCC";
@@ -1478,6 +1532,7 @@ HexagonTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
     case ISD::MEMBARRIER:         return LowerMEMBARRIER(Op, DAG);
     case ISD::ATOMIC_FENCE:       return LowerATOMIC_FENCE(Op, DAG);
     case ISD::GlobalAddress:      return LowerGLOBALADDRESS(Op, DAG);
+    case ISD::BlockAddress:       return LowerBlockAddress(Op, DAG);
     case ISD::VASTART:            return LowerVASTART(Op, DAG);
     case ISD::BR_JT:              return LowerBR_JT(Op, DAG);
 
diff --git a/lib/Target/Hexagon/HexagonISelLowering.h b/lib/Target/Hexagon/HexagonISelLowering.h
index fe6c905adfcb..3279cc652434 100644
--- a/lib/Target/Hexagon/HexagonISelLowering.h
+++ b/lib/Target/Hexagon/HexagonISelLowering.h
@@ -16,9 +16,9 @@
 #define Hexagon_ISELLOWERING_H
 
 #include "Hexagon.h"
-#include "llvm/Target/TargetLowering.h"
-#include "llvm/CallingConv.h"
 #include "llvm/CodeGen/CallingConvLower.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/Target/TargetLowering.h"
 
 namespace llvm {
   namespace HexagonISD {
@@ -27,6 +27,7 @@ namespace llvm {
 
       CONST32,
       CONST32_GP,  // For marking data present in GP.
+      CONST32_Int_Real,
       FCONST32,
       SETCC,
       ADJDYNALLOC,
@@ -50,6 +51,17 @@ namespace llvm {
       BARRIER,     // Memory barrier.
       WrapperJT,
       WrapperCP,
+      WrapperCombineII,
+      WrapperCombineRR,
+      WrapperCombineRI_V4,
+      WrapperCombineIR_V4,
+      WrapperPackhl,
+      WrapperSplatB,
+      WrapperSplatH,
+      WrapperShuffEB,
+      WrapperShuffEH,
+      WrapperShuffOB,
+      WrapperShuffOH,
       TC_RETURN
     };
   }
@@ -95,6 +107,7 @@ namespace llvm {
                                  DebugLoc dl, SelectionDAG &DAG,
                                  SmallVectorImpl<SDValue> &InVals) const;
     SDValue LowerGLOBALADDRESS(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
 
     SDValue LowerCall(TargetLowering::CallLoweringInfo &CLI,
                       SmallVectorImpl<SDValue> &InVals) const;
diff --git a/lib/Target/Hexagon/HexagonImmediates.td b/lib/Target/Hexagon/HexagonImmediates.td
deleted file mode 100644
index 18692c4dcc5e..000000000000
--- a/lib/Target/Hexagon/HexagonImmediates.td
+++ /dev/null
@@ -1,508 +0,0 @@
-//===- HexagonImmediates.td - Hexagon immediate processing -*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illnois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-// From IA64's InstrInfo file
-def s32Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s16Imm : Operand<i32> {
-  let PrintMethod = "printImmOperand";
-}
-
-def s12Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s11Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s11_0Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s11_1Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s11_2Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s11_3Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s10Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s9Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s8Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s8Imm64 : Operand<i64> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s6Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s4Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s4_0Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s4_1Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s4_2Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def s4_3Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u64Imm : Operand<i64> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u32Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u16Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u16_0Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u16_1Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u16_2Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u11_3Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u10Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u9Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u8Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u7Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u6Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u6_0Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u6_1Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u6_2Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u6_3Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u5Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u4Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u3Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u2Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def u1Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def n8Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def m6Imm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printImmOperand";
-}
-
-def nOneImm : Operand<i32> {
-  // For now, we use a generic print function for all operands.
-  let PrintMethod = "printNOneImmOperand";
-}
-
-//
-// Immediate predicates
-//
-def s32ImmPred  : PatLeaf<(i32 imm), [{
-  // immS16 predicate - True if the immediate fits in a 16-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<32>(v);
-}]>;
-
-def s32_24ImmPred  : PatLeaf<(i32 imm), [{
-  // s32_24ImmPred predicate - True if the immediate fits in a 32-bit sign
-  // extended field that is a multiple of 0x1000000.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedInt<32,24>(v);
-}]>;
-
-def s32_16s8ImmPred  : PatLeaf<(i32 imm), [{
-  // s32_16s8ImmPred predicate - True if the immediate fits in a 32-bit sign
-  // extended field that is a multiple of 0x10000.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedInt<24,16>(v);
-}]>;
-
-def s16ImmPred  : PatLeaf<(i32 imm), [{
-  // immS16 predicate - True if the immediate fits in a 16-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<16>(v);
-}]>;
-
-
-def s13ImmPred  : PatLeaf<(i32 imm), [{
-  // immS13 predicate - True if the immediate fits in a 13-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<13>(v);
-}]>;
-
-
-def s12ImmPred  : PatLeaf<(i32 imm), [{
-  // immS16 predicate - True if the immediate fits in a 16-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<12>(v);
-}]>;
-
-def s11_0ImmPred  : PatLeaf<(i32 imm), [{
-  // immS16 predicate - True if the immediate fits in a 16-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<11>(v);
-}]>;
-
-
-def s11_1ImmPred  : PatLeaf<(i32 imm), [{
-  // immS16 predicate - True if the immediate fits in a 16-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedInt<11,1>(v);
-}]>;
-
-
-def s11_2ImmPred  : PatLeaf<(i32 imm), [{
-  // immS16 predicate - True if the immediate fits in a 16-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedInt<11,2>(v);
-}]>;
-
-
-def s11_3ImmPred  : PatLeaf<(i32 imm), [{
-  // immS16 predicate - True if the immediate fits in a 16-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedInt<11,3>(v);
-}]>;
-
-
-def s10ImmPred  : PatLeaf<(i32 imm), [{
-  // s10ImmPred predicate - True if the immediate fits in a 10-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<10>(v);
-}]>;
-
-
-def s9ImmPred  : PatLeaf<(i32 imm), [{
-  // s9ImmPred predicate - True if the immediate fits in a 9-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<9>(v);
-}]>;
-
-
-def s8ImmPred  : PatLeaf<(i32 imm), [{
-  // s8ImmPred predicate - True if the immediate fits in a 8-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<8>(v);
-}]>;
-
-
-def s8Imm64Pred  : PatLeaf<(i64 imm), [{
-  // s8ImmPred predicate - True if the immediate fits in a 8-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<8>(v);
-}]>;
-
-
-def s6ImmPred  : PatLeaf<(i32 imm), [{
-  // s6ImmPred predicate - True if the immediate fits in a 6-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<6>(v);
-}]>;
-
-
-def s4_0ImmPred  : PatLeaf<(i32 imm), [{
-  // s4_0ImmPred predicate - True if the immediate fits in a 4-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<4>(v);
-}]>;
-
-
-def s4_1ImmPred  : PatLeaf<(i32 imm), [{
-  // s4_1ImmPred predicate - True if the immediate fits in a 4-bit sign extended
-  // field of 2.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedInt<4,1>(v);
-}]>;
-
-
-def s4_2ImmPred  : PatLeaf<(i32 imm), [{
-  // s4_2ImmPred predicate - True if the immediate fits in a 4-bit sign extended
-  // field that is a multiple of 4.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedInt<4,2>(v);
-}]>;
-
-
-def s4_3ImmPred  : PatLeaf<(i32 imm), [{
-  // s4_3ImmPred predicate - True if the immediate fits in a 4-bit sign extended
-  // field that is a multiple of 8.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedInt<4,3>(v);
-}]>;
-
-
-def u64ImmPred  : PatLeaf<(i64 imm), [{
-  // immS16 predicate - True if the immediate fits in a 16-bit sign extended
-  // field.
-  // Adding "N ||" to suppress gcc unused warning.
-  return (N || true);
-}]>;
-
-def u32ImmPred  : PatLeaf<(i32 imm), [{
-  // immS16 predicate - True if the immediate fits in a 16-bit sign extended
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<32>(v);
-}]>;
-
-def u16ImmPred  : PatLeaf<(i32 imm), [{
-  // u16ImmPred predicate - True if the immediate fits in a 16-bit unsigned
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<16>(v);
-}]>;
-
-def u16_s8ImmPred  : PatLeaf<(i32 imm), [{
-  // u16_s8ImmPred predicate - True if the immediate fits in a 16-bit sign
-  // extended s8 field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedUInt<16,8>(v);
-}]>;
-
-def u9ImmPred  : PatLeaf<(i32 imm), [{
-  // u9ImmPred predicate - True if the immediate fits in a 9-bit unsigned
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<9>(v);
-}]>;
-
-
-def u8ImmPred  : PatLeaf<(i32 imm), [{
-  // u8ImmPred predicate - True if the immediate fits in a 8-bit unsigned
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<8>(v);
-}]>;
-
-def u7ImmPred  : PatLeaf<(i32 imm), [{
-  // u7ImmPred predicate - True if the immediate fits in a 8-bit unsigned
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<7>(v);
-}]>;
-
-
-def u6ImmPred  : PatLeaf<(i32 imm), [{
-  // u6ImmPred predicate - True if the immediate fits in a 6-bit unsigned
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<6>(v);
-}]>;
-
-def u6_0ImmPred  : PatLeaf<(i32 imm), [{
-  // u6_0ImmPred predicate - True if the immediate fits in a 6-bit unsigned
-  // field. Same as u6ImmPred.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<6>(v);
-}]>;
-
-def u6_1ImmPred  : PatLeaf<(i32 imm), [{
-  // u6_1ImmPred predicate - True if the immediate fits in a 6-bit unsigned
-  // field that is 1 bit alinged - multiple of 2.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedUInt<6,1>(v);
-}]>;
-
-def u6_2ImmPred  : PatLeaf<(i32 imm), [{
-  // u6_2ImmPred predicate - True if the immediate fits in a 6-bit unsigned
-  // field that is 2 bits alinged - multiple of 4.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedUInt<6,2>(v);
-}]>;
-
-def u6_3ImmPred  : PatLeaf<(i32 imm), [{
-  // u6_3ImmPred predicate - True if the immediate fits in a 6-bit unsigned
-  // field that is 3 bits alinged - multiple of 8.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isShiftedUInt<6,3>(v);
-}]>;
-
-def u5ImmPred  : PatLeaf<(i32 imm), [{
-  // u5ImmPred predicate - True if the immediate fits in a 5-bit unsigned
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<5>(v);
-}]>;
-
-
-def u3ImmPred  : PatLeaf<(i32 imm), [{
-  // u3ImmPred predicate - True if the immediate fits in a 3-bit unsigned
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<3>(v);
-}]>;
-
-
-def u2ImmPred  : PatLeaf<(i32 imm), [{
-  // u2ImmPred predicate - True if the immediate fits in a 2-bit unsigned
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<2>(v);
-}]>;
-
-
-def u1ImmPred  : PatLeaf<(i1 imm), [{
-  // u1ImmPred predicate - True if the immediate fits in a 1-bit unsigned
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isUInt<1>(v);
-}]>;
-
-def m6ImmPred  : PatLeaf<(i32 imm), [{
-  // m6ImmPred predicate - True if the immediate is negative and fits in
-  // a 6-bit negative number.
-  int64_t v = (int64_t)N->getSExtValue();
-  return isInt<6>(v);
-}]>;
-
-//InN means negative integers in [-(2^N - 1), 0]
-def n8ImmPred  : PatLeaf<(i32 imm), [{
-  // n8ImmPred predicate - True if the immediate fits in a 8-bit signed
-  // field.
-  int64_t v = (int64_t)N->getSExtValue();
-  return (-255 <= v && v <= 0);
-}]>;
-
-def nOneImmPred  : PatLeaf<(i32 imm), [{
-  // nOneImmPred predicate - True if the immediate is -1.
-  int64_t v = (int64_t)N->getSExtValue();
-  return (-1 == v);
-}]>;
-
diff --git a/lib/Target/Hexagon/HexagonInstrFormats.td b/lib/Target/Hexagon/HexagonInstrFormats.td
index a64c7a18164f..587fa7d7f10e 100644
--- a/lib/Target/Hexagon/HexagonInstrFormats.td
+++ b/lib/Target/Hexagon/HexagonInstrFormats.td
@@ -13,28 +13,77 @@
 //                    *** Must match HexagonBaseInfo.h ***
 //===----------------------------------------------------------------------===//
 
-class Type<bits<5> t> {
+class IType<bits<5> t> {
   bits<5> Value = t;
 }
-def TypePSEUDO : Type<0>;
-def TypeALU32  : Type<1>;
-def TypeCR     : Type<2>;
-def TypeJR     : Type<3>;
-def TypeJ      : Type<4>;
-def TypeLD     : Type<5>;
-def TypeST     : Type<6>;
-def TypeSYSTEM : Type<7>;
-def TypeXTYPE  : Type<8>;
-def TypeMARKER : Type<31>;
+def TypePSEUDO : IType<0>;
+def TypeALU32  : IType<1>;
+def TypeCR     : IType<2>;
+def TypeJR     : IType<3>;
+def TypeJ      : IType<4>;
+def TypeLD     : IType<5>;
+def TypeST     : IType<6>;
+def TypeSYSTEM : IType<7>;
+def TypeXTYPE  : IType<8>;
+def TypeENDLOOP: IType<31>;
+
+// Maintain list of valid subtargets for each instruction.
+class SubTarget<bits<4> value> {
+  bits<4> Value = value;
+}
+
+def HasV2SubT     : SubTarget<0xf>;
+def HasV2SubTOnly : SubTarget<0x1>;
+def NoV2SubT      : SubTarget<0x0>;
+def HasV3SubT     : SubTarget<0xe>;
+def HasV3SubTOnly : SubTarget<0x2>;
+def NoV3SubT      : SubTarget<0x1>;
+def HasV4SubT     : SubTarget<0xc>;
+def NoV4SubT      : SubTarget<0x3>;
+def HasV5SubT     : SubTarget<0x8>;
+def NoV5SubT      : SubTarget<0x7>;
+
+// Addressing modes for load/store instructions
+class AddrModeType<bits<3> value> {
+  bits<3> Value = value;
+}
+
+def NoAddrMode     : AddrModeType<0>;  // No addressing mode
+def Absolute       : AddrModeType<1>;  // Absolute addressing mode
+def AbsoluteSet    : AddrModeType<2>;  // Absolute set addressing mode
+def BaseImmOffset  : AddrModeType<3>;  // Indirect with offset
+def BaseLongOffset : AddrModeType<4>;  // Indirect with long offset
+def BaseRegOffset  : AddrModeType<5>;  // Indirect with register offset
+
+class MemAccessSize<bits<3> value> {
+  bits<3> Value = value;
+}
+
+def NoMemAccess      : MemAccessSize<0>;// Not a memory acces instruction.
+def ByteAccess       : MemAccessSize<1>;// Byte access instruction (memb).
+def HalfWordAccess   : MemAccessSize<2>;// Half word access instruction (memh).
+def WordAccess       : MemAccessSize<3>;// Word access instrution (memw).
+def DoubleWordAccess : MemAccessSize<4>;// Double word access instruction (memd)
+
 
 //===----------------------------------------------------------------------===//
 //                         Intruction Class Declaration +
 //===----------------------------------------------------------------------===//
 
-class InstHexagon<dag outs, dag ins, string asmstr, list<dag> pattern,
-                  string cstr, InstrItinClass itin, Type type> : Instruction {
-  field bits<32> Inst;
+class OpcodeHexagon {
+  field bits<32> Inst = ?; // Default to an invalid insn.
+  bits<4> IClass = 0; // ICLASS
+  bits<2> IParse = 0; // Parse bits.
+
+  let Inst{31-28} = IClass;
+  let Inst{15-14} = IParse;
+
+  bits<1> zero = 0;
+}
 
+class InstHexagon<dag outs, dag ins, string asmstr, list<dag> pattern,
+                  string cstr, InstrItinClass itin, IType type>
+  : Instruction, OpcodeHexagon {
   let Namespace = "Hexagon";
 
   dag OutOperandList = outs;
@@ -45,20 +94,63 @@ class InstHexagon<dag outs, dag ins, string asmstr, list<dag> pattern,
   let Itinerary = itin;
   let Size = 4;
 
-  // *** Must match HexagonBaseInfo.h ***
+  // *** Must match MCTargetDesc/HexagonBaseInfo.h ***
+
   // Instruction type according to the ISA.
-  Type HexagonType = type;
-  let TSFlags{4-0} = HexagonType.Value;
+  IType Type = type;
+  let TSFlags{4-0} = Type.Value;
+
   // Solo instructions, i.e., those that cannot be in a packet with others.
-  bits<1> isHexagonSolo = 0;
-  let TSFlags{5} = isHexagonSolo;
+  bits<1> isSolo = 0;
+  let TSFlags{5} = isSolo;
+
   // Predicated instructions.
   bits<1> isPredicated = 0;
   let TSFlags{6} = isPredicated;
-
-  // Dot new value store instructions.
+  bits<1> isPredicatedFalse = 0;
+  let TSFlags{7} = isPredicatedFalse;
+  bits<1> isPredicatedNew = 0;
+  let TSFlags{8} = isPredicatedNew;
+
+  // New-value insn helper fields.
+  bits<1> isNewValue = 0;
+  let TSFlags{9} = isNewValue; // New-value consumer insn.
+  bits<1> hasNewValue = 0;
+  let TSFlags{10} = hasNewValue; // New-value producer insn.
+  bits<3> opNewValue = 0;
+  let TSFlags{13-11} = opNewValue; // New-value produced operand.
+  bits<2> opNewBits = 0;
+  let TSFlags{15-14} = opNewBits; // New-value opcode bits location: 0, 8, 16.
+  bits<1> isNVStorable = 0;
+  let TSFlags{16} = isNVStorable; // Store that can become new-value store.
   bits<1> isNVStore = 0;
-  let TSFlags{8} = isNVStore;
+  let TSFlags{17} = isNVStore; // New-value store insn.
+
+  // Immediate extender helper fields.
+  bits<1> isExtendable = 0;
+  let TSFlags{18} = isExtendable; // Insn may be extended.
+  bits<1> isExtended = 0;
+  let TSFlags{19} = isExtended; // Insn must be extended.
+  bits<3> opExtendable = 0;
+  let TSFlags{22-20} = opExtendable; // Which operand may be extended.
+  bits<1> isExtentSigned = 0;
+  let TSFlags{23} = isExtentSigned; // Signed or unsigned range.
+  bits<5> opExtentBits = 0;
+  let TSFlags{28-24} = opExtentBits; //Number of bits of range before extending.
+
+  // If an instruction is valid on a subtarget (v2-v5), set the corresponding
+  // bit from validSubTargets. v2 is the least significant bit.
+  // By default, instruction is valid on all subtargets.
+  SubTarget validSubTargets = HasV2SubT;
+  let TSFlags{32-29} = validSubTargets.Value;
+
+  // Addressing mode for load/store instructions.
+  AddrModeType addrMode = NoAddrMode;
+  let TSFlags{35-33} = addrMode.Value;
+
+  // Memory access size for mem access instructions (load/store)
+  MemAccessSize accessSize = NoMemAccess;
+  let TSFlags{38-36} = accessSize.Value;
 
   // Fields used for relation models.
   string BaseOpcode = "";
@@ -66,7 +158,15 @@ class InstHexagon<dag outs, dag ins, string asmstr, list<dag> pattern,
   string PredSense = "";
   string PNewValue = "";
   string InputType = "";    // Input is "imm" or "reg" type.
-  // *** The code above must match HexagonBaseInfo.h ***
+  string isMEMri = "false"; // Set to "true" for load/store with MEMri operand.
+  string isFloat = "false"; // Set to "true" for the floating-point load/store.
+  string isBrTaken = ""; // Set to "true"/"false" for jump instructions
+
+  let PredSense = !if(isPredicated, !if(isPredicatedFalse, "false", "true"),
+                                    "");
+  let PNewValue = !if(isPredicatedNew, "new", "");
+
+  // *** Must match MCTargetDesc/HexagonBaseInfo.h ***
 }
 
 //===----------------------------------------------------------------------===//
@@ -75,187 +175,143 @@ class InstHexagon<dag outs, dag ins, string asmstr, list<dag> pattern,
 
 // LD Instruction Class in V2/V3/V4.
 // Definition of the instruction class NOT CHANGED.
-class LDInst<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", LD, TypeLD> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<13> imm13;
-}
+class LDInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+             string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, LD, TypeLD>;
 
-class LDInst2<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", LD, TypeLD> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<13> imm13;
-  let mayLoad = 1;
-}
+let mayLoad = 1 in
+class LDInst2<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+              string cstr = "">
+  : LDInst<outs, ins, asmstr, pattern, cstr>;
+
+class CONSTLDInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                  string cstr = "">
+  : LDInst<outs, ins, asmstr, pattern, cstr>;
 
 // LD Instruction Class in V2/V3/V4.
 // Definition of the instruction class NOT CHANGED.
-class LDInstPost<dag outs, dag ins, string asmstr, list<dag> pattern,
-                 string cstr>
-  : InstHexagon<outs, ins, asmstr, pattern, cstr, LD, TypeLD> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<5> rt;
-  bits<13> imm13;
-}
+class LDInstPost<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                 string cstr = "">
+  : LDInst<outs, ins, asmstr, pattern, cstr>;
+
+let mayLoad = 1 in
+class LD0Inst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+              string cstr = "">
+  : LDInst<outs, ins, asmstr, pattern, cstr>;
 
 // ST Instruction Class in V2/V3 can take SLOT0 only.
 // ST Instruction Class in V4    can take SLOT0 & SLOT1.
 // Definition of the instruction class CHANGED from V2/V3 to V4.
-class STInst<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", ST, TypeST> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<13> imm13;
-}
+let mayStore = 1 in
+class STInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+             string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, ST, TypeST>;
 
-class STInst2<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", ST, TypeST> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<13> imm13;
-  let mayStore = 1;
-}
+class STInst2<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+              string cstr = "">
+  : STInst<outs, ins, asmstr, pattern, cstr>;
 
-// SYSTEM Instruction Class in V4 can take SLOT0 only
-// In V2/V3 we used ST for this but in v4 ST can take SLOT0 or SLOT1.
-class SYSInst<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", SYS, TypeSYSTEM> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<13> imm13;
-}
+let mayStore = 1 in
+class ST0Inst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+              string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, ST0, TypeST>;
 
 // ST Instruction Class in V2/V3 can take SLOT0 only.
 // ST Instruction Class in V4    can take SLOT0 & SLOT1.
 // Definition of the instruction class CHANGED from V2/V3 to V4.
-class STInstPost<dag outs, dag ins, string asmstr, list<dag> pattern,
-                 string cstr>
-  : InstHexagon<outs, ins, asmstr, pattern, cstr, ST, TypeST> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<5> rt;
-  bits<13> imm13;
-}
+class STInstPost<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                 string cstr = "">
+  : STInst<outs, ins, asmstr, pattern, cstr>;
+
+// SYSTEM Instruction Class in V4 can take SLOT0 only
+// In V2/V3 we used ST for this but in v4 ST can take SLOT0 or SLOT1.
+class SYSInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+              string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, SYS, TypeSYSTEM>;
 
 // ALU32 Instruction Class in V2/V3/V4.
 // Definition of the instruction class NOT CHANGED.
-class ALU32Type<dag outs, dag ins, string asmstr, list<dag> pattern>
-   : InstHexagon<outs, ins, asmstr, pattern, "", ALU32, TypeALU32> {
-  bits<5>  rd;
-  bits<5>  rs;
-  bits<5>  rt;
-  bits<16> imm16;
-  bits<16> imm16_2;
-}
+class ALU32Inst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                string cstr = "">
+   : InstHexagon<outs, ins, asmstr, pattern, cstr, ALU32, TypeALU32>;
 
 // ALU64 Instruction Class in V2/V3.
 // XTYPE Instruction Class in V4.
 // Definition of the instruction class NOT CHANGED.
 // Name of the Instruction Class changed from ALU64 to XTYPE from V2/V3 to V4.
-class ALU64Type<dag outs, dag ins, string asmstr, list<dag> pattern>
-   : InstHexagon<outs, ins, asmstr, pattern, "", ALU64, TypeXTYPE> {
-  bits<5>  rd;
-  bits<5>  rs;
-  bits<5>  rt;
-  bits<16> imm16;
-  bits<16> imm16_2;
-}
+class ALU64Inst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                string cstr = "">
+   : InstHexagon<outs, ins, asmstr, pattern, cstr, ALU64, TypeXTYPE>;
+
+class ALU64_acc<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                string cstr = "">
+  : ALU64Inst<outs, ins, asmstr, pattern, cstr>;
 
-class ALU64_acc<dag outs, dag ins, string asmstr, list<dag> pattern,
-   string cstr>
-   : InstHexagon<outs, ins, asmstr, pattern, cstr, ALU64, TypeXTYPE> {
-  bits<5>  rd;
-  bits<5>  rs;
-  bits<5>  rt;
-  bits<16> imm16;
-  bits<16> imm16_2;
-}
 
 // M Instruction Class in V2/V3.
 // XTYPE Instruction Class in V4.
 // Definition of the instruction class NOT CHANGED.
 // Name of the Instruction Class changed from M to XTYPE from V2/V3 to V4.
-class MInst<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", M, TypeXTYPE> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<5> rt;
-}
+class MInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+            string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, M, TypeXTYPE>;
 
 // M Instruction Class in V2/V3.
 // XTYPE Instruction Class in V4.
 // Definition of the instruction class NOT CHANGED.
 // Name of the Instruction Class changed from M to XTYPE from V2/V3 to V4.
-class MInst_acc<dag outs, dag ins, string asmstr, list<dag> pattern,
-    string cstr>
-    : InstHexagon<outs, ins, asmstr, pattern, cstr, M, TypeXTYPE> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<5> rt;
-}
+class MInst_acc<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                string cstr = "">
+    : MInst<outs, ins, asmstr, pattern, cstr>;
 
 // S Instruction Class in V2/V3.
 // XTYPE Instruction Class in V4.
 // Definition of the instruction class NOT CHANGED.
 // Name of the Instruction Class changed from S to XTYPE from V2/V3 to V4.
-class SInst<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", S, TypeXTYPE> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<5> rt;
-}
+class SInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+            string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, S, TypeXTYPE>;
 
 // S Instruction Class in V2/V3.
 // XTYPE Instruction Class in V4.
 // Definition of the instruction class NOT CHANGED.
 // Name of the Instruction Class changed from S to XTYPE from V2/V3 to V4.
-class SInst_acc<dag outs, dag ins, string asmstr, list<dag> pattern,
-   string cstr>
-  : InstHexagon<outs, ins, asmstr, pattern, cstr, S, TypeXTYPE> {
-//  : InstHexagon<outs, ins, asmstr, pattern, cstr,  S> {
-//  : InstHexagon<outs, ins, asmstr, pattern, cstr, !if(V4T, XTYPE_V4, S)> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<5> rt;
-}
+class SInst_acc<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                string cstr = "">
+  : SInst<outs, ins, asmstr, pattern, cstr>;
 
 // J Instruction Class in V2/V3/V4.
 // Definition of the instruction class NOT CHANGED.
-class JType<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", J, TypeJ> {
-  bits<16> imm16;
-}
+class JInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+            string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, J, TypeJ>;
 
 // JR Instruction Class in V2/V3/V4.
 // Definition of the instruction class NOT CHANGED.
-class JRType<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", JR, TypeJR> {
-  bits<5>  rs;
-  bits<5>  pu; // Predicate register
-}
+class JRInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+             string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, JR, TypeJR>;
 
 // CR Instruction Class in V2/V3/V4.
 // Definition of the instruction class NOT CHANGED.
-class CRInst<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", CR, TypeCR> {
-  bits<5> rs;
-  bits<10> imm10;
-}
+class CRInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+             string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, CR, TypeCR>;
 
-class Marker<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", MARKER, TypeMARKER> {
-  let isCodeGenOnly = 1;
-  let isPseudo = 1;
-}
+let isCodeGenOnly = 1, isPseudo = 1 in
+class Endloop<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+              string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, ENDLOOP, TypeENDLOOP>;
 
-class Pseudo<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", PSEUDO, TypePSEUDO> {
-  let isCodeGenOnly = 1;
-  let isPseudo = 1;
-}
+let isCodeGenOnly = 1, isPseudo = 1 in
+class Pseudo<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+             string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, PSEUDO, TypePSEUDO>;
+
+let isCodeGenOnly = 1, isPseudo = 1 in
+class PseudoM<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+              string cstr="">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, PSEUDOM, TypePSEUDO>;
 
 //===----------------------------------------------------------------------===//
 //                         Intruction Classes Definitions -
@@ -265,75 +321,52 @@ class Pseudo<dag outs, dag ins, string asmstr, list<dag> pattern>
 //
 // ALU32 patterns
 //.
-class ALU32_rr<dag outs, dag ins, string asmstr, list<dag> pattern>
-   : ALU32Type<outs, ins, asmstr, pattern> {
-}
+class ALU32_rr<dag outs, dag ins, string asmstr, list<dag> pattern,
+               string cstr = "">
+   : ALU32Inst<outs, ins, asmstr, pattern, cstr>;
 
-class ALU32_ir<dag outs, dag ins, string asmstr, list<dag> pattern>
-   : ALU32Type<outs, ins, asmstr, pattern> {
-   let rt{0-4} = 0;
-}
+class ALU32_ir<dag outs, dag ins, string asmstr, list<dag> pattern,
+               string cstr = "">
+   : ALU32Inst<outs, ins, asmstr, pattern, cstr>;
 
-class ALU32_ri<dag outs, dag ins, string asmstr, list<dag> pattern>
-   : ALU32Type<outs, ins, asmstr, pattern> {
-  let rt{0-4} = 0;
-}
+class ALU32_ri<dag outs, dag ins, string asmstr, list<dag> pattern,
+               string cstr = "">
+   : ALU32Inst<outs, ins, asmstr, pattern, cstr>;
 
-class ALU32_ii<dag outs, dag ins, string asmstr, list<dag> pattern>
-   : ALU32Type<outs, ins, asmstr, pattern> {
-  let rt{0-4} = 0;
-}
+class ALU32_ii<dag outs, dag ins, string asmstr, list<dag> pattern,
+               string cstr = "">
+   : ALU32Inst<outs, ins, asmstr, pattern, cstr>;
 
 //
 // ALU64 patterns.
 //
-class ALU64_rr<dag outs, dag ins, string asmstr, list<dag> pattern>
-   : ALU64Type<outs, ins, asmstr, pattern> {
-}
-
-class ALU64_ri<dag outs, dag ins, string asmstr, list<dag> pattern>
-   : ALU64Type<outs, ins, asmstr, pattern> {
-  let rt{0-4} = 0;
-}
-
-// J Type Instructions.
-class JInst<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : JType<outs, ins, asmstr, pattern> {
-}
-
-// JR type Instructions.
-class JRInst<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : JRType<outs, ins, asmstr, pattern> {
-}
+class ALU64_rr<dag outs, dag ins, string asmstr, list<dag> pattern,
+               string cstr = "">
+   : ALU64Inst<outs, ins, asmstr, pattern, cstr>;
 
+class ALU64_ri<dag outs, dag ins, string asmstr, list<dag> pattern,
+               string cstr = "">
+   : ALU64Inst<outs, ins, asmstr, pattern, cstr>;
 
 // Post increment ST Instruction.
-class STInstPI<dag outs, dag ins, string asmstr, list<dag> pattern,
-               string cstr>
-  : STInstPost<outs, ins, asmstr, pattern, cstr> {
-  let rt{0-4} = 0;
-}
+class STInstPI<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+               string cstr = "">
+  : STInst<outs, ins, asmstr, pattern, cstr>;
 
-class STInst2PI<dag outs, dag ins, string asmstr, list<dag> pattern,
-                string cstr>
-  : STInstPost<outs, ins, asmstr, pattern, cstr> {
-  let rt{0-4} = 0;
-  let mayStore = 1;
-}
+let mayStore = 1 in
+class STInst2PI<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                string cstr = "">
+  : STInst<outs, ins, asmstr, pattern, cstr>;
 
 // Post increment LD Instruction.
-class LDInstPI<dag outs, dag ins, string asmstr, list<dag> pattern,
-               string cstr>
-  : LDInstPost<outs, ins, asmstr, pattern, cstr> {
-  let rt{0-4} = 0;
-}
-
-class LDInst2PI<dag outs, dag ins, string asmstr, list<dag> pattern,
-                string cstr>
-  : LDInstPost<outs, ins, asmstr, pattern, cstr> {
-  let rt{0-4} = 0;
-  let mayLoad = 1;
-}
+class LDInstPI<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+               string cstr = "">
+  : LDInst<outs, ins, asmstr, pattern, cstr>;
+
+let mayLoad = 1 in
+class LDInst2PI<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                string cstr = "">
+  : LDInst<outs, ins, asmstr, pattern, cstr>;
 
 //===----------------------------------------------------------------------===//
 // V4 Instruction Format Definitions +
diff --git a/lib/Target/Hexagon/HexagonInstrFormatsV4.td b/lib/Target/Hexagon/HexagonInstrFormatsV4.td
index 49741a3d1b20..9fda0da91612 100644
--- a/lib/Target/Hexagon/HexagonInstrFormatsV4.td
+++ b/lib/Target/Hexagon/HexagonInstrFormatsV4.td
@@ -17,9 +17,9 @@
 //                        *** Must match BaseInfo.h ***
 //----------------------------------------------------------------------------//
 
-def TypeMEMOP  : Type<9>;
-def TypeNV     : Type<10>;
-def TypePREFIX : Type<30>;
+def TypeMEMOP  : IType<9>;
+def TypeNV     : IType<10>;
+def TypePREFIX : IType<30>;
 
 //----------------------------------------------------------------------------//
 //                         Intruction Classes Definitions +
@@ -28,40 +28,39 @@ def TypePREFIX : Type<30>;
 //
 // NV type instructions.
 //
-class NVInst_V4<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", NV_V4, TypeNV> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<13> imm13;
-}
+class NVInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+             string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, NV_V4, TypeNV>;
+
+class NVInst_V4<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                string cstr = "">
+  : NVInst<outs, ins, asmstr, pattern, cstr>;
 
 // Definition of Post increment new value store.
-class NVInstPost_V4<dag outs, dag ins, string asmstr, list<dag> pattern,
-                    string cstr>
-  : InstHexagon<outs, ins, asmstr, pattern, cstr, NV_V4, TypeNV> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<5> rt;
-  bits<13> imm13;
-}
+class NVInstPost_V4<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+               string cstr = "">
+  : NVInst<outs, ins, asmstr, pattern, cstr>;
 
 // Post increment ST Instruction.
-class NVInstPI_V4<dag outs, dag ins, string asmstr, list<dag> pattern,
-                  string cstr>
-  : NVInstPost_V4<outs, ins, asmstr, pattern, cstr> {
-  let rt{0-4} = 0;
-}
+let mayStore = 1 in
+class NVInstPI_V4<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+               string cstr = "">
+  : NVInst<outs, ins, asmstr, pattern, cstr>;
+
+// New-value conditional branch.
+class NCJInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+              string cstr = "">
+  : NVInst<outs, ins, asmstr, pattern, cstr>;
 
-class MEMInst_V4<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", MEM_V4, TypeMEMOP> {
-  bits<5> rd;
-  bits<5> rs;
-  bits<6> imm6;
-}
+let mayLoad = 1, mayStore = 1 in
+class MEMInst<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+              string cstr = "">
+  : InstHexagon<outs, ins, asmstr, pattern, cstr, MEM_V4, TypeMEMOP>;
 
-class Immext<dag outs, dag ins, string asmstr, list<dag> pattern>
-  : InstHexagon<outs, ins, asmstr, pattern, "", PREFIX, TypePREFIX> {
-  let isCodeGenOnly = 1;
+class MEMInst_V4<dag outs, dag ins, string asmstr, list<dag> pattern = [],
+                 string cstr = "">
+  : MEMInst<outs, ins, asmstr, pattern, cstr>;
 
-  bits<26> imm26;
-}
+let isCodeGenOnly = 1 in
+class EXTENDERInst<dag outs, dag ins, string asmstr, list<dag> pattern = []>
+  : InstHexagon<outs, ins, asmstr, pattern, "", PREFIX, TypePREFIX>;
diff --git a/lib/Target/Hexagon/HexagonInstrInfo.cpp b/lib/Target/Hexagon/HexagonInstrInfo.cpp
index 84354403084d..60b12ac01c9c 100644
--- a/lib/Target/Hexagon/HexagonInstrInfo.cpp
+++ b/lib/Target/Hexagon/HexagonInstrInfo.cpp
@@ -12,16 +12,16 @@
 //===----------------------------------------------------------------------===//
 
 #include "HexagonInstrInfo.h"
+#include "Hexagon.h"
 #include "HexagonRegisterInfo.h"
 #include "HexagonSubtarget.h"
-#include "Hexagon.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/CodeGen/DFAPacketizer.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineMemOperand.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/PseudoSourceValue.h"
 #include "llvm/Support/MathExtras.h"
 #define GET_INSTRINFO_CTOR
@@ -305,6 +305,88 @@ unsigned HexagonInstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
 }
 
 
+/// \brief For a comparison instruction, return the source registers in
+/// \p SrcReg and \p SrcReg2 if having two register operands, and the value it
+/// compares against in CmpValue. Return true if the comparison instruction
+/// can be analyzed.
+bool HexagonInstrInfo::analyzeCompare(const MachineInstr *MI,
+                                      unsigned &SrcReg, unsigned &SrcReg2,
+                                      int &Mask, int &Value) const {
+  unsigned Opc = MI->getOpcode();
+
+  // Set mask and the first source register.
+  switch (Opc) {
+    case Hexagon::CMPEHexagon4rr:
+    case Hexagon::CMPEQri:
+    case Hexagon::CMPEQrr:
+    case Hexagon::CMPGT64rr:
+    case Hexagon::CMPGTU64rr:
+    case Hexagon::CMPGTUri:
+    case Hexagon::CMPGTUrr:
+    case Hexagon::CMPGTri:
+    case Hexagon::CMPGTrr:
+    case Hexagon::CMPLTUrr:
+    case Hexagon::CMPLTrr:
+      SrcReg = MI->getOperand(1).getReg();
+      Mask = ~0;
+      break;
+    case Hexagon::CMPbEQri_V4:
+    case Hexagon::CMPbEQrr_sbsb_V4:
+    case Hexagon::CMPbEQrr_ubub_V4:
+    case Hexagon::CMPbGTUri_V4:
+    case Hexagon::CMPbGTUrr_V4:
+    case Hexagon::CMPbGTrr_V4:
+      SrcReg = MI->getOperand(1).getReg();
+      Mask = 0xFF;
+      break;
+    case Hexagon::CMPhEQri_V4:
+    case Hexagon::CMPhEQrr_shl_V4:
+    case Hexagon::CMPhEQrr_xor_V4:
+    case Hexagon::CMPhGTUri_V4:
+    case Hexagon::CMPhGTUrr_V4:
+    case Hexagon::CMPhGTrr_shl_V4:
+      SrcReg = MI->getOperand(1).getReg();
+      Mask = 0xFFFF;
+      break;
+  }
+
+  // Set the value/second source register.
+  switch (Opc) {
+    case Hexagon::CMPEHexagon4rr:
+    case Hexagon::CMPEQrr:
+    case Hexagon::CMPGT64rr:
+    case Hexagon::CMPGTU64rr:
+    case Hexagon::CMPGTUrr:
+    case Hexagon::CMPGTrr:
+    case Hexagon::CMPbEQrr_sbsb_V4:
+    case Hexagon::CMPbEQrr_ubub_V4:
+    case Hexagon::CMPbGTUrr_V4:
+    case Hexagon::CMPbGTrr_V4:
+    case Hexagon::CMPhEQrr_shl_V4:
+    case Hexagon::CMPhEQrr_xor_V4:
+    case Hexagon::CMPhGTUrr_V4:
+    case Hexagon::CMPhGTrr_shl_V4:
+    case Hexagon::CMPLTUrr:
+    case Hexagon::CMPLTrr:
+      SrcReg2 = MI->getOperand(2).getReg();
+      return true;
+
+    case Hexagon::CMPEQri:
+    case Hexagon::CMPGTUri:
+    case Hexagon::CMPGTri:
+    case Hexagon::CMPbEQri_V4:
+    case Hexagon::CMPbGTUri_V4:
+    case Hexagon::CMPhEQri_V4:
+    case Hexagon::CMPhGTUri_V4:
+      SrcReg2 = 0;
+      Value = MI->getOperand(2).getImm();
+      return true;
+  }
+
+  return false;
+}
+
+
 void HexagonInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
                                  MachineBasicBlock::iterator I, DebugLoc DL,
                                  unsigned DestReg, unsigned SrcReg,
@@ -314,7 +396,7 @@ void HexagonInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
     return;
   }
   if (Hexagon::DoubleRegsRegClass.contains(SrcReg, DestReg)) {
-    BuildMI(MBB, I, DL, get(Hexagon::TFR_64), DestReg).addReg(SrcReg);
+    BuildMI(MBB, I, DL, get(Hexagon::TFR64), DestReg).addReg(SrcReg);
     return;
   }
   if (Hexagon::PredRegsRegClass.contains(SrcReg, DestReg)) {
@@ -344,6 +426,18 @@ void HexagonInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
     BuildMI(MBB, I, DL, get(Hexagon::TFCR), DestReg).addReg(SrcReg);
     return;
   }
+  if (Hexagon::PredRegsRegClass.contains(SrcReg) &&
+      Hexagon::IntRegsRegClass.contains(DestReg)) {
+    BuildMI(MBB, I, DL, get(Hexagon::TFR_RsPd), DestReg).
+      addReg(SrcReg, getKillRegState(KillSrc));
+    return;
+  }
+  if (Hexagon::IntRegsRegClass.contains(SrcReg) &&
+      Hexagon::PredRegsRegClass.contains(DestReg)) {
+    BuildMI(MBB, I, DL, get(Hexagon::TFR_PdRs), DestReg).
+      addReg(SrcReg, getKillRegState(KillSrc));
+    return;
+  }
 
   llvm_unreachable("Unimplemented");
 }
@@ -443,6 +537,15 @@ MachineInstr *HexagonInstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
   return(0);
 }
 
+MachineInstr*
+HexagonInstrInfo::emitFrameIndexDebugValue(MachineFunction &MF,
+                                           int FrameIx, uint64_t Offset,
+                                           const MDNode *MDPtr,
+                                           DebugLoc DL) const {
+  MachineInstrBuilder MIB = BuildMI(MF, DL, get(Hexagon::DBG_VALUE))
+    .addImm(0).addImm(Offset).addMetadata(MDPtr);
+  return &*MIB;
+}
 
 unsigned HexagonInstrInfo::createVR(MachineFunction* MF, MVT VT) const {
 
@@ -463,270 +566,43 @@ unsigned HexagonInstrInfo::createVR(MachineFunction* MF, MVT VT) const {
 }
 
 bool HexagonInstrInfo::isExtendable(const MachineInstr *MI) const {
-  switch(MI->getOpcode()) {
-    default: return false;
-    // JMP_EQri
-    case Hexagon::JMP_EQriPt_nv_V4:
-    case Hexagon::JMP_EQriPnt_nv_V4:
-    case Hexagon::JMP_EQriNotPt_nv_V4:
-    case Hexagon::JMP_EQriNotPnt_nv_V4:
-
-    // JMP_EQri - with -1
-    case Hexagon::JMP_EQriPtneg_nv_V4:
-    case Hexagon::JMP_EQriPntneg_nv_V4:
-    case Hexagon::JMP_EQriNotPtneg_nv_V4:
-    case Hexagon::JMP_EQriNotPntneg_nv_V4:
-
-    // JMP_EQrr
-    case Hexagon::JMP_EQrrPt_nv_V4:
-    case Hexagon::JMP_EQrrPnt_nv_V4:
-    case Hexagon::JMP_EQrrNotPt_nv_V4:
-    case Hexagon::JMP_EQrrNotPnt_nv_V4:
-
-    // JMP_GTri
-    case Hexagon::JMP_GTriPt_nv_V4:
-    case Hexagon::JMP_GTriPnt_nv_V4:
-    case Hexagon::JMP_GTriNotPt_nv_V4:
-    case Hexagon::JMP_GTriNotPnt_nv_V4:
-
-    // JMP_GTri - with -1
-    case Hexagon::JMP_GTriPtneg_nv_V4:
-    case Hexagon::JMP_GTriPntneg_nv_V4:
-    case Hexagon::JMP_GTriNotPtneg_nv_V4:
-    case Hexagon::JMP_GTriNotPntneg_nv_V4:
-
-    // JMP_GTrr
-    case Hexagon::JMP_GTrrPt_nv_V4:
-    case Hexagon::JMP_GTrrPnt_nv_V4:
-    case Hexagon::JMP_GTrrNotPt_nv_V4:
-    case Hexagon::JMP_GTrrNotPnt_nv_V4:
-
-    // JMP_GTrrdn
-    case Hexagon::JMP_GTrrdnPt_nv_V4:
-    case Hexagon::JMP_GTrrdnPnt_nv_V4:
-    case Hexagon::JMP_GTrrdnNotPt_nv_V4:
-    case Hexagon::JMP_GTrrdnNotPnt_nv_V4:
-
-    // JMP_GTUri
-    case Hexagon::JMP_GTUriPt_nv_V4:
-    case Hexagon::JMP_GTUriPnt_nv_V4:
-    case Hexagon::JMP_GTUriNotPt_nv_V4:
-    case Hexagon::JMP_GTUriNotPnt_nv_V4:
-
-    // JMP_GTUrr
-    case Hexagon::JMP_GTUrrPt_nv_V4:
-    case Hexagon::JMP_GTUrrPnt_nv_V4:
-    case Hexagon::JMP_GTUrrNotPt_nv_V4:
-    case Hexagon::JMP_GTUrrNotPnt_nv_V4:
+  // Constant extenders are allowed only for V4 and above.
+  if (!Subtarget.hasV4TOps())
+    return false;
 
-    // JMP_GTUrrdn
-    case Hexagon::JMP_GTUrrdnPt_nv_V4:
-    case Hexagon::JMP_GTUrrdnPnt_nv_V4:
-    case Hexagon::JMP_GTUrrdnNotPt_nv_V4:
-    case Hexagon::JMP_GTUrrdnNotPnt_nv_V4:
+  const MCInstrDesc &MID = MI->getDesc();
+  const uint64_t F = MID.TSFlags;
+  if ((F >> HexagonII::ExtendablePos) & HexagonII::ExtendableMask)
+    return true;
 
-    // TFR_FI
+  // TODO: This is largely obsolete now. Will need to be removed
+  // in consecutive patches.
+  switch(MI->getOpcode()) {
+    // TFR_FI Remains a special case.
     case Hexagon::TFR_FI:
       return true;
+    default:
+      return false;
   }
+  return  false;
 }
 
+// This returns true in two cases:
+// - The OP code itself indicates that this is an extended instruction.
+// - One of MOs has been marked with HMOTF_ConstExtended flag.
 bool HexagonInstrInfo::isExtended(const MachineInstr *MI) const {
-  switch(MI->getOpcode()) {
-    default: return false;
-    // JMP_EQri
-    case Hexagon::JMP_EQriPt_ie_nv_V4:
-    case Hexagon::JMP_EQriPnt_ie_nv_V4:
-    case Hexagon::JMP_EQriNotPt_ie_nv_V4:
-    case Hexagon::JMP_EQriNotPnt_ie_nv_V4:
-
-    // JMP_EQri - with -1
-    case Hexagon::JMP_EQriPtneg_ie_nv_V4:
-    case Hexagon::JMP_EQriPntneg_ie_nv_V4:
-    case Hexagon::JMP_EQriNotPtneg_ie_nv_V4:
-    case Hexagon::JMP_EQriNotPntneg_ie_nv_V4:
-
-    // JMP_EQrr
-    case Hexagon::JMP_EQrrPt_ie_nv_V4:
-    case Hexagon::JMP_EQrrPnt_ie_nv_V4:
-    case Hexagon::JMP_EQrrNotPt_ie_nv_V4:
-    case Hexagon::JMP_EQrrNotPnt_ie_nv_V4:
-
-    // JMP_GTri
-    case Hexagon::JMP_GTriPt_ie_nv_V4:
-    case Hexagon::JMP_GTriPnt_ie_nv_V4:
-    case Hexagon::JMP_GTriNotPt_ie_nv_V4:
-    case Hexagon::JMP_GTriNotPnt_ie_nv_V4:
-
-    // JMP_GTri - with -1
-    case Hexagon::JMP_GTriPtneg_ie_nv_V4:
-    case Hexagon::JMP_GTriPntneg_ie_nv_V4:
-    case Hexagon::JMP_GTriNotPtneg_ie_nv_V4:
-    case Hexagon::JMP_GTriNotPntneg_ie_nv_V4:
-
-    // JMP_GTrr
-    case Hexagon::JMP_GTrrPt_ie_nv_V4:
-    case Hexagon::JMP_GTrrPnt_ie_nv_V4:
-    case Hexagon::JMP_GTrrNotPt_ie_nv_V4:
-    case Hexagon::JMP_GTrrNotPnt_ie_nv_V4:
-
-    // JMP_GTrrdn
-    case Hexagon::JMP_GTrrdnPt_ie_nv_V4:
-    case Hexagon::JMP_GTrrdnPnt_ie_nv_V4:
-    case Hexagon::JMP_GTrrdnNotPt_ie_nv_V4:
-    case Hexagon::JMP_GTrrdnNotPnt_ie_nv_V4:
-
-    // JMP_GTUri
-    case Hexagon::JMP_GTUriPt_ie_nv_V4:
-    case Hexagon::JMP_GTUriPnt_ie_nv_V4:
-    case Hexagon::JMP_GTUriNotPt_ie_nv_V4:
-    case Hexagon::JMP_GTUriNotPnt_ie_nv_V4:
-
-    // JMP_GTUrr
-    case Hexagon::JMP_GTUrrPt_ie_nv_V4:
-    case Hexagon::JMP_GTUrrPnt_ie_nv_V4:
-    case Hexagon::JMP_GTUrrNotPt_ie_nv_V4:
-    case Hexagon::JMP_GTUrrNotPnt_ie_nv_V4:
-
-    // JMP_GTUrrdn
-    case Hexagon::JMP_GTUrrdnPt_ie_nv_V4:
-    case Hexagon::JMP_GTUrrdnPnt_ie_nv_V4:
-    case Hexagon::JMP_GTUrrdnNotPt_ie_nv_V4:
-    case Hexagon::JMP_GTUrrdnNotPnt_ie_nv_V4:
-
-    // V4 absolute set addressing.
-    case Hexagon::LDrid_abs_setimm_V4:
-    case Hexagon::LDriw_abs_setimm_V4:
-    case Hexagon::LDrih_abs_setimm_V4:
-    case Hexagon::LDrib_abs_setimm_V4:
-    case Hexagon::LDriuh_abs_setimm_V4:
-    case Hexagon::LDriub_abs_setimm_V4:
-
-    case Hexagon::STrid_abs_setimm_V4:
-    case Hexagon::STrib_abs_setimm_V4:
-    case Hexagon::STrih_abs_setimm_V4:
-    case Hexagon::STriw_abs_setimm_V4:
-
-    // V4 global address load.
-    case Hexagon::LDrid_GP_cPt_V4 :
-    case Hexagon::LDrid_GP_cNotPt_V4 :
-    case Hexagon::LDrid_GP_cdnPt_V4 :
-    case Hexagon::LDrid_GP_cdnNotPt_V4 :
-    case Hexagon::LDrib_GP_cPt_V4 :
-    case Hexagon::LDrib_GP_cNotPt_V4 :
-    case Hexagon::LDrib_GP_cdnPt_V4 :
-    case Hexagon::LDrib_GP_cdnNotPt_V4 :
-    case Hexagon::LDriub_GP_cPt_V4 :
-    case Hexagon::LDriub_GP_cNotPt_V4 :
-    case Hexagon::LDriub_GP_cdnPt_V4 :
-    case Hexagon::LDriub_GP_cdnNotPt_V4 :
-    case Hexagon::LDrih_GP_cPt_V4 :
-    case Hexagon::LDrih_GP_cNotPt_V4 :
-    case Hexagon::LDrih_GP_cdnPt_V4 :
-    case Hexagon::LDrih_GP_cdnNotPt_V4 :
-    case Hexagon::LDriuh_GP_cPt_V4 :
-    case Hexagon::LDriuh_GP_cNotPt_V4 :
-    case Hexagon::LDriuh_GP_cdnPt_V4 :
-    case Hexagon::LDriuh_GP_cdnNotPt_V4 :
-    case Hexagon::LDriw_GP_cPt_V4 :
-    case Hexagon::LDriw_GP_cNotPt_V4 :
-    case Hexagon::LDriw_GP_cdnPt_V4 :
-    case Hexagon::LDriw_GP_cdnNotPt_V4 :
-    case Hexagon::LDd_GP_cPt_V4 :
-    case Hexagon::LDd_GP_cNotPt_V4 :
-    case Hexagon::LDd_GP_cdnPt_V4 :
-    case Hexagon::LDd_GP_cdnNotPt_V4 :
-    case Hexagon::LDb_GP_cPt_V4 :
-    case Hexagon::LDb_GP_cNotPt_V4 :
-    case Hexagon::LDb_GP_cdnPt_V4 :
-    case Hexagon::LDb_GP_cdnNotPt_V4 :
-    case Hexagon::LDub_GP_cPt_V4 :
-    case Hexagon::LDub_GP_cNotPt_V4 :
-    case Hexagon::LDub_GP_cdnPt_V4 :
-    case Hexagon::LDub_GP_cdnNotPt_V4 :
-    case Hexagon::LDh_GP_cPt_V4 :
-    case Hexagon::LDh_GP_cNotPt_V4 :
-    case Hexagon::LDh_GP_cdnPt_V4 :
-    case Hexagon::LDh_GP_cdnNotPt_V4 :
-    case Hexagon::LDuh_GP_cPt_V4 :
-    case Hexagon::LDuh_GP_cNotPt_V4 :
-    case Hexagon::LDuh_GP_cdnPt_V4 :
-    case Hexagon::LDuh_GP_cdnNotPt_V4 :
-    case Hexagon::LDw_GP_cPt_V4 :
-    case Hexagon::LDw_GP_cNotPt_V4 :
-    case Hexagon::LDw_GP_cdnPt_V4 :
-    case Hexagon::LDw_GP_cdnNotPt_V4 :
-
-    // V4 global address store.
-    case Hexagon::STrid_GP_cPt_V4 :
-    case Hexagon::STrid_GP_cNotPt_V4 :
-    case Hexagon::STrid_GP_cdnPt_V4 :
-    case Hexagon::STrid_GP_cdnNotPt_V4 :
-    case Hexagon::STrib_GP_cPt_V4 :
-    case Hexagon::STrib_GP_cNotPt_V4 :
-    case Hexagon::STrib_GP_cdnPt_V4 :
-    case Hexagon::STrib_GP_cdnNotPt_V4 :
-    case Hexagon::STrih_GP_cPt_V4 :
-    case Hexagon::STrih_GP_cNotPt_V4 :
-    case Hexagon::STrih_GP_cdnPt_V4 :
-    case Hexagon::STrih_GP_cdnNotPt_V4 :
-    case Hexagon::STriw_GP_cPt_V4 :
-    case Hexagon::STriw_GP_cNotPt_V4 :
-    case Hexagon::STriw_GP_cdnPt_V4 :
-    case Hexagon::STriw_GP_cdnNotPt_V4 :
-    case Hexagon::STd_GP_cPt_V4 :
-    case Hexagon::STd_GP_cNotPt_V4 :
-    case Hexagon::STd_GP_cdnPt_V4 :
-    case Hexagon::STd_GP_cdnNotPt_V4 :
-    case Hexagon::STb_GP_cPt_V4 :
-    case Hexagon::STb_GP_cNotPt_V4 :
-    case Hexagon::STb_GP_cdnPt_V4 :
-    case Hexagon::STb_GP_cdnNotPt_V4 :
-    case Hexagon::STh_GP_cPt_V4 :
-    case Hexagon::STh_GP_cNotPt_V4 :
-    case Hexagon::STh_GP_cdnPt_V4 :
-    case Hexagon::STh_GP_cdnNotPt_V4 :
-    case Hexagon::STw_GP_cPt_V4 :
-    case Hexagon::STw_GP_cNotPt_V4 :
-    case Hexagon::STw_GP_cdnPt_V4 :
-    case Hexagon::STw_GP_cdnNotPt_V4 :
-
-    // V4 predicated global address new value store.
-    case Hexagon::STrib_GP_cPt_nv_V4 :
-    case Hexagon::STrib_GP_cNotPt_nv_V4 :
-    case Hexagon::STrib_GP_cdnPt_nv_V4 :
-    case Hexagon::STrib_GP_cdnNotPt_nv_V4 :
-    case Hexagon::STrih_GP_cPt_nv_V4 :
-    case Hexagon::STrih_GP_cNotPt_nv_V4 :
-    case Hexagon::STrih_GP_cdnPt_nv_V4 :
-    case Hexagon::STrih_GP_cdnNotPt_nv_V4 :
-    case Hexagon::STriw_GP_cPt_nv_V4 :
-    case Hexagon::STriw_GP_cNotPt_nv_V4 :
-    case Hexagon::STriw_GP_cdnPt_nv_V4 :
-    case Hexagon::STriw_GP_cdnNotPt_nv_V4 :
-    case Hexagon::STb_GP_cPt_nv_V4 :
-    case Hexagon::STb_GP_cNotPt_nv_V4 :
-    case Hexagon::STb_GP_cdnPt_nv_V4 :
-    case Hexagon::STb_GP_cdnNotPt_nv_V4 :
-    case Hexagon::STh_GP_cPt_nv_V4 :
-    case Hexagon::STh_GP_cNotPt_nv_V4 :
-    case Hexagon::STh_GP_cdnPt_nv_V4 :
-    case Hexagon::STh_GP_cdnNotPt_nv_V4 :
-    case Hexagon::STw_GP_cPt_nv_V4 :
-    case Hexagon::STw_GP_cNotPt_nv_V4 :
-    case Hexagon::STw_GP_cdnPt_nv_V4 :
-    case Hexagon::STw_GP_cdnNotPt_nv_V4 :
-
-    // TFR_FI
-    case Hexagon::TFR_FI_immext_V4:
-
-    // TFRI_F
-    case Hexagon::TFRI_f:
-    case Hexagon::TFRI_cPt_f:
-    case Hexagon::TFRI_cNotPt_f:
-    case Hexagon::CONST64_Float_Real:
+  // First check if this is permanently extended op code.
+  const uint64_t F = MI->getDesc().TSFlags;
+  if ((F >> HexagonII::ExtendedPos) & HexagonII::ExtendedMask)
+    return true;
+  // Use MO operand flags to determine if one of MI's operands
+  // has HMOTF_ConstExtended flag set.
+  for (MachineInstr::const_mop_iterator I = MI->operands_begin(),
+       E = MI->operands_end(); I != E; ++I) {
+    if (I->getTargetFlags() && HexagonII::HMOTF_ConstExtended)
       return true;
   }
+  return  false;
 }
 
 bool HexagonInstrInfo::isNewValueJump(const MachineInstr *MI) const {
@@ -835,264 +711,6 @@ bool HexagonInstrInfo::isNewValueJump(const MachineInstr *MI) const {
   }
 }
 
-unsigned HexagonInstrInfo::getImmExtForm(const MachineInstr* MI) const {
-  switch(MI->getOpcode()) {
-    default: llvm_unreachable("Unknown type of instruction.");
-    // JMP_EQri
-    case Hexagon::JMP_EQriPt_nv_V4:
-      return Hexagon::JMP_EQriPt_ie_nv_V4;
-    case Hexagon::JMP_EQriNotPt_nv_V4:
-      return Hexagon::JMP_EQriNotPt_ie_nv_V4;
-    case Hexagon::JMP_EQriPnt_nv_V4:
-      return Hexagon::JMP_EQriPnt_ie_nv_V4;
-    case Hexagon::JMP_EQriNotPnt_nv_V4:
-      return Hexagon::JMP_EQriNotPnt_ie_nv_V4;
-
-    // JMP_EQri -- with -1
-    case Hexagon::JMP_EQriPtneg_nv_V4:
-      return Hexagon::JMP_EQriPtneg_ie_nv_V4;
-    case Hexagon::JMP_EQriNotPtneg_nv_V4:
-      return Hexagon::JMP_EQriNotPtneg_ie_nv_V4;
-    case Hexagon::JMP_EQriPntneg_nv_V4:
-      return Hexagon::JMP_EQriPntneg_ie_nv_V4;
-    case Hexagon::JMP_EQriNotPntneg_nv_V4:
-      return Hexagon::JMP_EQriNotPntneg_ie_nv_V4;
-
-    // JMP_EQrr
-    case Hexagon::JMP_EQrrPt_nv_V4:
-      return Hexagon::JMP_EQrrPt_ie_nv_V4;
-    case Hexagon::JMP_EQrrNotPt_nv_V4:
-      return Hexagon::JMP_EQrrNotPt_ie_nv_V4;
-    case Hexagon::JMP_EQrrPnt_nv_V4:
-      return Hexagon::JMP_EQrrPnt_ie_nv_V4;
-    case Hexagon::JMP_EQrrNotPnt_nv_V4:
-      return Hexagon::JMP_EQrrNotPnt_ie_nv_V4;
-
-    // JMP_GTri
-    case Hexagon::JMP_GTriPt_nv_V4:
-      return Hexagon::JMP_GTriPt_ie_nv_V4;
-    case Hexagon::JMP_GTriNotPt_nv_V4:
-      return Hexagon::JMP_GTriNotPt_ie_nv_V4;
-    case Hexagon::JMP_GTriPnt_nv_V4:
-      return Hexagon::JMP_GTriPnt_ie_nv_V4;
-    case Hexagon::JMP_GTriNotPnt_nv_V4:
-      return Hexagon::JMP_GTriNotPnt_ie_nv_V4;
-
-    // JMP_GTri -- with -1
-    case Hexagon::JMP_GTriPtneg_nv_V4:
-      return Hexagon::JMP_GTriPtneg_ie_nv_V4;
-    case Hexagon::JMP_GTriNotPtneg_nv_V4:
-      return Hexagon::JMP_GTriNotPtneg_ie_nv_V4;
-    case Hexagon::JMP_GTriPntneg_nv_V4:
-      return Hexagon::JMP_GTriPntneg_ie_nv_V4;
-    case Hexagon::JMP_GTriNotPntneg_nv_V4:
-      return Hexagon::JMP_GTriNotPntneg_ie_nv_V4;
-
-    // JMP_GTrr
-    case Hexagon::JMP_GTrrPt_nv_V4:
-      return Hexagon::JMP_GTrrPt_ie_nv_V4;
-    case Hexagon::JMP_GTrrNotPt_nv_V4:
-      return Hexagon::JMP_GTrrNotPt_ie_nv_V4;
-    case Hexagon::JMP_GTrrPnt_nv_V4:
-      return Hexagon::JMP_GTrrPnt_ie_nv_V4;
-    case Hexagon::JMP_GTrrNotPnt_nv_V4:
-      return Hexagon::JMP_GTrrNotPnt_ie_nv_V4;
-
-    // JMP_GTrrdn
-    case Hexagon::JMP_GTrrdnPt_nv_V4:
-      return Hexagon::JMP_GTrrdnPt_ie_nv_V4;
-    case Hexagon::JMP_GTrrdnNotPt_nv_V4:
-      return Hexagon::JMP_GTrrdnNotPt_ie_nv_V4;
-    case Hexagon::JMP_GTrrdnPnt_nv_V4:
-      return Hexagon::JMP_GTrrdnPnt_ie_nv_V4;
-    case Hexagon::JMP_GTrrdnNotPnt_nv_V4:
-      return Hexagon::JMP_GTrrdnNotPnt_ie_nv_V4;
-
-    // JMP_GTUri
-    case Hexagon::JMP_GTUriPt_nv_V4:
-      return Hexagon::JMP_GTUriPt_ie_nv_V4;
-    case Hexagon::JMP_GTUriNotPt_nv_V4:
-      return Hexagon::JMP_GTUriNotPt_ie_nv_V4;
-    case Hexagon::JMP_GTUriPnt_nv_V4:
-      return Hexagon::JMP_GTUriPnt_ie_nv_V4;
-    case Hexagon::JMP_GTUriNotPnt_nv_V4:
-      return Hexagon::JMP_GTUriNotPnt_ie_nv_V4;
-
-    // JMP_GTUrr
-    case Hexagon::JMP_GTUrrPt_nv_V4:
-      return Hexagon::JMP_GTUrrPt_ie_nv_V4;
-    case Hexagon::JMP_GTUrrNotPt_nv_V4:
-      return Hexagon::JMP_GTUrrNotPt_ie_nv_V4;
-    case Hexagon::JMP_GTUrrPnt_nv_V4:
-      return Hexagon::JMP_GTUrrPnt_ie_nv_V4;
-    case Hexagon::JMP_GTUrrNotPnt_nv_V4:
-      return Hexagon::JMP_GTUrrNotPnt_ie_nv_V4;
-
-    // JMP_GTUrrdn
-    case Hexagon::JMP_GTUrrdnPt_nv_V4:
-      return Hexagon::JMP_GTUrrdnPt_ie_nv_V4;
-    case Hexagon::JMP_GTUrrdnNotPt_nv_V4:
-      return Hexagon::JMP_GTUrrdnNotPt_ie_nv_V4;
-    case Hexagon::JMP_GTUrrdnPnt_nv_V4:
-      return Hexagon::JMP_GTUrrdnPnt_ie_nv_V4;
-    case Hexagon::JMP_GTUrrdnNotPnt_nv_V4:
-      return Hexagon::JMP_GTUrrdnNotPnt_ie_nv_V4;
-
-    case Hexagon::TFR_FI:
-        return Hexagon::TFR_FI_immext_V4;
-
-    case Hexagon::MEMw_ADDSUBi_indexed_MEM_V4 :
-    case Hexagon::MEMw_ADDi_indexed_MEM_V4 :
-    case Hexagon::MEMw_SUBi_indexed_MEM_V4 :
-    case Hexagon::MEMw_ADDr_indexed_MEM_V4 :
-    case Hexagon::MEMw_SUBr_indexed_MEM_V4 :
-    case Hexagon::MEMw_ANDr_indexed_MEM_V4 :
-    case Hexagon::MEMw_ORr_indexed_MEM_V4 :
-    case Hexagon::MEMw_ADDSUBi_MEM_V4 :
-    case Hexagon::MEMw_ADDi_MEM_V4 :
-    case Hexagon::MEMw_SUBi_MEM_V4 :
-    case Hexagon::MEMw_ADDr_MEM_V4 :
-    case Hexagon::MEMw_SUBr_MEM_V4 :
-    case Hexagon::MEMw_ANDr_MEM_V4 :
-    case Hexagon::MEMw_ORr_MEM_V4 :
-    case Hexagon::MEMh_ADDSUBi_indexed_MEM_V4 :
-    case Hexagon::MEMh_ADDi_indexed_MEM_V4 :
-    case Hexagon::MEMh_SUBi_indexed_MEM_V4 :
-    case Hexagon::MEMh_ADDr_indexed_MEM_V4 :
-    case Hexagon::MEMh_SUBr_indexed_MEM_V4 :
-    case Hexagon::MEMh_ANDr_indexed_MEM_V4 :
-    case Hexagon::MEMh_ORr_indexed_MEM_V4 :
-    case Hexagon::MEMh_ADDSUBi_MEM_V4 :
-    case Hexagon::MEMh_ADDi_MEM_V4 :
-    case Hexagon::MEMh_SUBi_MEM_V4 :
-    case Hexagon::MEMh_ADDr_MEM_V4 :
-    case Hexagon::MEMh_SUBr_MEM_V4 :
-    case Hexagon::MEMh_ANDr_MEM_V4 :
-    case Hexagon::MEMh_ORr_MEM_V4 :
-    case Hexagon::MEMb_ADDSUBi_indexed_MEM_V4 :
-    case Hexagon::MEMb_ADDi_indexed_MEM_V4 :
-    case Hexagon::MEMb_SUBi_indexed_MEM_V4 :
-    case Hexagon::MEMb_ADDr_indexed_MEM_V4 :
-    case Hexagon::MEMb_SUBr_indexed_MEM_V4 :
-    case Hexagon::MEMb_ANDr_indexed_MEM_V4 :
-    case Hexagon::MEMb_ORr_indexed_MEM_V4 :
-    case Hexagon::MEMb_ADDSUBi_MEM_V4 :
-    case Hexagon::MEMb_ADDi_MEM_V4 :
-    case Hexagon::MEMb_SUBi_MEM_V4 :
-    case Hexagon::MEMb_ADDr_MEM_V4 :
-    case Hexagon::MEMb_SUBr_MEM_V4 :
-    case Hexagon::MEMb_ANDr_MEM_V4 :
-    case Hexagon::MEMb_ORr_MEM_V4 :
-      llvm_unreachable("Needs implementing.");
-  }
-}
-
-unsigned HexagonInstrInfo::getNormalBranchForm(const MachineInstr* MI) const {
-  switch(MI->getOpcode()) {
-    default: llvm_unreachable("Unknown type of jump instruction.");
-    // JMP_EQri
-    case Hexagon::JMP_EQriPt_ie_nv_V4:
-      return Hexagon::JMP_EQriPt_nv_V4;
-    case Hexagon::JMP_EQriNotPt_ie_nv_V4:
-      return Hexagon::JMP_EQriNotPt_nv_V4;
-    case Hexagon::JMP_EQriPnt_ie_nv_V4:
-      return Hexagon::JMP_EQriPnt_nv_V4;
-    case Hexagon::JMP_EQriNotPnt_ie_nv_V4:
-      return Hexagon::JMP_EQriNotPnt_nv_V4;
-
-    // JMP_EQri -- with -1
-    case Hexagon::JMP_EQriPtneg_ie_nv_V4:
-      return Hexagon::JMP_EQriPtneg_nv_V4;
-    case Hexagon::JMP_EQriNotPtneg_ie_nv_V4:
-      return Hexagon::JMP_EQriNotPtneg_nv_V4;
-    case Hexagon::JMP_EQriPntneg_ie_nv_V4:
-      return Hexagon::JMP_EQriPntneg_nv_V4;
-    case Hexagon::JMP_EQriNotPntneg_ie_nv_V4:
-      return Hexagon::JMP_EQriNotPntneg_nv_V4;
-
-    // JMP_EQrr
-    case Hexagon::JMP_EQrrPt_ie_nv_V4:
-      return Hexagon::JMP_EQrrPt_nv_V4;
-    case Hexagon::JMP_EQrrNotPt_ie_nv_V4:
-      return Hexagon::JMP_EQrrNotPt_nv_V4;
-    case Hexagon::JMP_EQrrPnt_ie_nv_V4:
-      return Hexagon::JMP_EQrrPnt_nv_V4;
-    case Hexagon::JMP_EQrrNotPnt_ie_nv_V4:
-      return Hexagon::JMP_EQrrNotPnt_nv_V4;
-
-    // JMP_GTri
-    case Hexagon::JMP_GTriPt_ie_nv_V4:
-      return Hexagon::JMP_GTriPt_nv_V4;
-    case Hexagon::JMP_GTriNotPt_ie_nv_V4:
-      return Hexagon::JMP_GTriNotPt_nv_V4;
-    case Hexagon::JMP_GTriPnt_ie_nv_V4:
-      return Hexagon::JMP_GTriPnt_nv_V4;
-    case Hexagon::JMP_GTriNotPnt_ie_nv_V4:
-      return Hexagon::JMP_GTriNotPnt_nv_V4;
-
-    // JMP_GTri -- with -1
-    case Hexagon::JMP_GTriPtneg_ie_nv_V4:
-      return Hexagon::JMP_GTriPtneg_nv_V4;
-    case Hexagon::JMP_GTriNotPtneg_ie_nv_V4:
-      return Hexagon::JMP_GTriNotPtneg_nv_V4;
-    case Hexagon::JMP_GTriPntneg_ie_nv_V4:
-      return Hexagon::JMP_GTriPntneg_nv_V4;
-    case Hexagon::JMP_GTriNotPntneg_ie_nv_V4:
-      return Hexagon::JMP_GTriNotPntneg_nv_V4;
-
-    // JMP_GTrr
-    case Hexagon::JMP_GTrrPt_ie_nv_V4:
-      return Hexagon::JMP_GTrrPt_nv_V4;
-    case Hexagon::JMP_GTrrNotPt_ie_nv_V4:
-      return Hexagon::JMP_GTrrNotPt_nv_V4;
-    case Hexagon::JMP_GTrrPnt_ie_nv_V4:
-      return Hexagon::JMP_GTrrPnt_nv_V4;
-    case Hexagon::JMP_GTrrNotPnt_ie_nv_V4:
-      return Hexagon::JMP_GTrrNotPnt_nv_V4;
-
-    // JMP_GTrrdn
-    case Hexagon::JMP_GTrrdnPt_ie_nv_V4:
-      return Hexagon::JMP_GTrrdnPt_nv_V4;
-    case Hexagon::JMP_GTrrdnNotPt_ie_nv_V4:
-      return Hexagon::JMP_GTrrdnNotPt_nv_V4;
-    case Hexagon::JMP_GTrrdnPnt_ie_nv_V4:
-      return Hexagon::JMP_GTrrdnPnt_nv_V4;
-    case Hexagon::JMP_GTrrdnNotPnt_ie_nv_V4:
-      return Hexagon::JMP_GTrrdnNotPnt_nv_V4;
-
-    // JMP_GTUri
-    case Hexagon::JMP_GTUriPt_ie_nv_V4:
-      return Hexagon::JMP_GTUriPt_nv_V4;
-    case Hexagon::JMP_GTUriNotPt_ie_nv_V4:
-      return Hexagon::JMP_GTUriNotPt_nv_V4;
-    case Hexagon::JMP_GTUriPnt_ie_nv_V4:
-      return Hexagon::JMP_GTUriPnt_nv_V4;
-    case Hexagon::JMP_GTUriNotPnt_ie_nv_V4:
-      return Hexagon::JMP_GTUriNotPnt_nv_V4;
-
-    // JMP_GTUrr
-    case Hexagon::JMP_GTUrrPt_ie_nv_V4:
-      return Hexagon::JMP_GTUrrPt_nv_V4;
-    case Hexagon::JMP_GTUrrNotPt_ie_nv_V4:
-      return Hexagon::JMP_GTUrrNotPt_nv_V4;
-    case Hexagon::JMP_GTUrrPnt_ie_nv_V4:
-      return Hexagon::JMP_GTUrrPnt_nv_V4;
-    case Hexagon::JMP_GTUrrNotPnt_ie_nv_V4:
-      return Hexagon::JMP_GTUrrNotPnt_nv_V4;
-
-    // JMP_GTUrrdn
-    case Hexagon::JMP_GTUrrdnPt_ie_nv_V4:
-      return Hexagon::JMP_GTUrrdnPt_nv_V4;
-    case Hexagon::JMP_GTUrrdnNotPt_ie_nv_V4:
-      return Hexagon::JMP_GTUrrdnNotPt_nv_V4;
-    case Hexagon::JMP_GTUrrdnPnt_ie_nv_V4:
-      return Hexagon::JMP_GTUrrdnPnt_nv_V4;
-    case Hexagon::JMP_GTUrrdnNotPnt_ie_nv_V4:
-      return Hexagon::JMP_GTUrrdnNotPnt_nv_V4;
-  }
-}
-
-
 bool HexagonInstrInfo::isNewValueStore(const MachineInstr *MI) const {
   switch (MI->getOpcode()) {
     default: return false;
@@ -1101,7 +719,6 @@ bool HexagonInstrInfo::isNewValueStore(const MachineInstr *MI) const {
     case Hexagon::STrib_indexed_nv_V4:
     case Hexagon::STrib_indexed_shl_nv_V4:
     case Hexagon::STrib_shl_nv_V4:
-    case Hexagon::STrib_GP_nv_V4:
     case Hexagon::STb_GP_nv_V4:
     case Hexagon::POST_STbri_nv_V4:
     case Hexagon::STrib_cPt_nv_V4:
@@ -1124,10 +741,6 @@ bool HexagonInstrInfo::isNewValueStore(const MachineInstr *MI) const {
     case Hexagon::STb_GP_cNotPt_nv_V4:
     case Hexagon::STb_GP_cdnPt_nv_V4:
     case Hexagon::STb_GP_cdnNotPt_nv_V4:
-    case Hexagon::STrib_GP_cPt_nv_V4:
-    case Hexagon::STrib_GP_cNotPt_nv_V4:
-    case Hexagon::STrib_GP_cdnPt_nv_V4:
-    case Hexagon::STrib_GP_cdnNotPt_nv_V4:
     case Hexagon::STrib_abs_nv_V4:
     case Hexagon::STrib_abs_cPt_nv_V4:
     case Hexagon::STrib_abs_cdnPt_nv_V4:
@@ -1144,7 +757,6 @@ bool HexagonInstrInfo::isNewValueStore(const MachineInstr *MI) const {
     case Hexagon::STrih_indexed_nv_V4:
     case Hexagon::STrih_indexed_shl_nv_V4:
     case Hexagon::STrih_shl_nv_V4:
-    case Hexagon::STrih_GP_nv_V4:
     case Hexagon::STh_GP_nv_V4:
     case Hexagon::POST_SThri_nv_V4:
     case Hexagon::STrih_cPt_nv_V4:
@@ -1167,10 +779,6 @@ bool HexagonInstrInfo::isNewValueStore(const MachineInstr *MI) const {
     case Hexagon::STh_GP_cNotPt_nv_V4:
     case Hexagon::STh_GP_cdnPt_nv_V4:
     case Hexagon::STh_GP_cdnNotPt_nv_V4:
-    case Hexagon::STrih_GP_cPt_nv_V4:
-    case Hexagon::STrih_GP_cNotPt_nv_V4:
-    case Hexagon::STrih_GP_cdnPt_nv_V4:
-    case Hexagon::STrih_GP_cdnNotPt_nv_V4:
     case Hexagon::STrih_abs_nv_V4:
     case Hexagon::STrih_abs_cPt_nv_V4:
     case Hexagon::STrih_abs_cdnPt_nv_V4:
@@ -1187,7 +795,6 @@ bool HexagonInstrInfo::isNewValueStore(const MachineInstr *MI) const {
     case Hexagon::STriw_indexed_nv_V4:
     case Hexagon::STriw_indexed_shl_nv_V4:
     case Hexagon::STriw_shl_nv_V4:
-    case Hexagon::STriw_GP_nv_V4:
     case Hexagon::STw_GP_nv_V4:
     case Hexagon::POST_STwri_nv_V4:
     case Hexagon::STriw_cPt_nv_V4:
@@ -1210,10 +817,6 @@ bool HexagonInstrInfo::isNewValueStore(const MachineInstr *MI) const {
     case Hexagon::STw_GP_cNotPt_nv_V4:
     case Hexagon::STw_GP_cdnPt_nv_V4:
     case Hexagon::STw_GP_cdnNotPt_nv_V4:
-    case Hexagon::STriw_GP_cPt_nv_V4:
-    case Hexagon::STriw_GP_cNotPt_nv_V4:
-    case Hexagon::STriw_GP_cdnPt_nv_V4:
-    case Hexagon::STriw_GP_cdnNotPt_nv_V4:
     case Hexagon::STriw_abs_nv_V4:
     case Hexagon::STriw_abs_cPt_nv_V4:
     case Hexagon::STriw_abs_cdnPt_nv_V4:
@@ -1305,6 +908,16 @@ bool HexagonInstrInfo::isPostIncrement (const MachineInstr* MI) const {
   }
 }
 
+bool HexagonInstrInfo::isNewValueInst(const MachineInstr *MI) const {
+  if (isNewValueJump(MI))
+    return true;
+
+  if (isNewValueStore(MI))
+    return true;
+
+  return false;
+}
+
 bool HexagonInstrInfo::isSaveCalleeSavedRegsCall(const MachineInstr *MI) const {
   return MI->getOpcode() == Hexagon::SAVE_REGISTERS_CALL_V4;
 }
@@ -1506,26 +1119,11 @@ unsigned HexagonInstrInfo::getInvertedPredicatedOpcode(const int Opc) const {
       return Hexagon::JMPR_cPt;
 
   // V4 indexed+scaled load.
-    case Hexagon::LDrid_indexed_cPt_V4:
-      return Hexagon::LDrid_indexed_cNotPt_V4;
-    case Hexagon::LDrid_indexed_cNotPt_V4:
-      return Hexagon::LDrid_indexed_cPt_V4;
-
     case Hexagon::LDrid_indexed_shl_cPt_V4:
       return Hexagon::LDrid_indexed_shl_cNotPt_V4;
     case Hexagon::LDrid_indexed_shl_cNotPt_V4:
       return Hexagon::LDrid_indexed_shl_cPt_V4;
 
-    case Hexagon::LDrib_indexed_cPt_V4:
-      return Hexagon::LDrib_indexed_cNotPt_V4;
-    case Hexagon::LDrib_indexed_cNotPt_V4:
-      return Hexagon::LDrib_indexed_cPt_V4;
-
-    case Hexagon::LDriub_indexed_cPt_V4:
-      return Hexagon::LDriub_indexed_cNotPt_V4;
-    case Hexagon::LDriub_indexed_cNotPt_V4:
-      return Hexagon::LDriub_indexed_cPt_V4;
-
     case Hexagon::LDrib_indexed_shl_cPt_V4:
       return Hexagon::LDrib_indexed_shl_cNotPt_V4;
     case Hexagon::LDrib_indexed_shl_cNotPt_V4:
@@ -1536,16 +1134,6 @@ unsigned HexagonInstrInfo::getInvertedPredicatedOpcode(const int Opc) const {
     case Hexagon::LDriub_indexed_shl_cNotPt_V4:
       return Hexagon::LDriub_indexed_shl_cPt_V4;
 
-    case Hexagon::LDrih_indexed_cPt_V4:
-      return Hexagon::LDrih_indexed_cNotPt_V4;
-    case Hexagon::LDrih_indexed_cNotPt_V4:
-      return Hexagon::LDrih_indexed_cPt_V4;
-
-    case Hexagon::LDriuh_indexed_cPt_V4:
-      return Hexagon::LDriuh_indexed_cNotPt_V4;
-    case Hexagon::LDriuh_indexed_cNotPt_V4:
-      return Hexagon::LDriuh_indexed_cPt_V4;
-
     case Hexagon::LDrih_indexed_shl_cPt_V4:
       return Hexagon::LDrih_indexed_shl_cNotPt_V4;
     case Hexagon::LDrih_indexed_shl_cNotPt_V4:
@@ -1556,11 +1144,6 @@ unsigned HexagonInstrInfo::getInvertedPredicatedOpcode(const int Opc) const {
     case Hexagon::LDriuh_indexed_shl_cNotPt_V4:
       return Hexagon::LDriuh_indexed_shl_cPt_V4;
 
-    case Hexagon::LDriw_indexed_cPt_V4:
-      return Hexagon::LDriw_indexed_cNotPt_V4;
-    case Hexagon::LDriw_indexed_cNotPt_V4:
-      return Hexagon::LDriw_indexed_cPt_V4;
-
     case Hexagon::LDriw_indexed_shl_cPt_V4:
       return Hexagon::LDriw_indexed_shl_cNotPt_V4;
     case Hexagon::LDriw_indexed_shl_cNotPt_V4:
@@ -1686,26 +1269,6 @@ unsigned HexagonInstrInfo::getInvertedPredicatedOpcode(const int Opc) const {
     case Hexagon::STw_GP_cNotPt_V4:
       return Hexagon::STw_GP_cPt_V4;
 
-    case Hexagon::STrid_GP_cPt_V4:
-      return Hexagon::STrid_GP_cNotPt_V4;
-    case Hexagon::STrid_GP_cNotPt_V4:
-      return Hexagon::STrid_GP_cPt_V4;
-
-    case Hexagon::STrib_GP_cPt_V4:
-      return Hexagon::STrib_GP_cNotPt_V4;
-    case Hexagon::STrib_GP_cNotPt_V4:
-      return Hexagon::STrib_GP_cPt_V4;
-
-    case Hexagon::STrih_GP_cPt_V4:
-      return Hexagon::STrih_GP_cNotPt_V4;
-    case Hexagon::STrih_GP_cNotPt_V4:
-      return Hexagon::STrih_GP_cPt_V4;
-
-    case Hexagon::STriw_GP_cPt_V4:
-      return Hexagon::STriw_GP_cNotPt_V4;
-    case Hexagon::STriw_GP_cNotPt_V4:
-      return Hexagon::STriw_GP_cPt_V4;
-
   // Load.
     case Hexagon::LDrid_cPt:
       return Hexagon::LDrid_cNotPt;
@@ -1971,75 +1534,26 @@ getMatchingCondBranchOpcode(int Opc, bool invertPredicate) const {
                               Hexagon::JMPR_cNotPt;
 
   // V4 indexed+scaled load.
-  case Hexagon::LDrid_indexed_V4:
-    return !invertPredicate ? Hexagon::LDrid_indexed_cPt_V4 :
-                              Hexagon::LDrid_indexed_cNotPt_V4;
   case Hexagon::LDrid_indexed_shl_V4:
     return !invertPredicate ? Hexagon::LDrid_indexed_shl_cPt_V4 :
                               Hexagon::LDrid_indexed_shl_cNotPt_V4;
-  case Hexagon::LDrib_indexed_V4:
-    return !invertPredicate ? Hexagon::LDrib_indexed_cPt_V4 :
-                              Hexagon::LDrib_indexed_cNotPt_V4;
-  case Hexagon::LDriub_indexed_V4:
-    return !invertPredicate ? Hexagon::LDriub_indexed_cPt_V4 :
-                              Hexagon::LDriub_indexed_cNotPt_V4;
-  case Hexagon::LDriub_ae_indexed_V4:
-    return !invertPredicate ? Hexagon::LDriub_indexed_cPt_V4 :
-                              Hexagon::LDriub_indexed_cNotPt_V4;
   case Hexagon::LDrib_indexed_shl_V4:
     return !invertPredicate ? Hexagon::LDrib_indexed_shl_cPt_V4 :
                               Hexagon::LDrib_indexed_shl_cNotPt_V4;
   case Hexagon::LDriub_indexed_shl_V4:
     return !invertPredicate ? Hexagon::LDriub_indexed_shl_cPt_V4 :
                               Hexagon::LDriub_indexed_shl_cNotPt_V4;
-  case Hexagon::LDriub_ae_indexed_shl_V4:
-    return !invertPredicate ? Hexagon::LDriub_indexed_shl_cPt_V4 :
-                              Hexagon::LDriub_indexed_shl_cNotPt_V4;
-  case Hexagon::LDrih_indexed_V4:
-    return !invertPredicate ? Hexagon::LDrih_indexed_cPt_V4 :
-                              Hexagon::LDrih_indexed_cNotPt_V4;
-  case Hexagon::LDriuh_indexed_V4:
-    return !invertPredicate ? Hexagon::LDriuh_indexed_cPt_V4 :
-                              Hexagon::LDriuh_indexed_cNotPt_V4;
-  case Hexagon::LDriuh_ae_indexed_V4:
-    return !invertPredicate ? Hexagon::LDriuh_indexed_cPt_V4 :
-                              Hexagon::LDriuh_indexed_cNotPt_V4;
   case Hexagon::LDrih_indexed_shl_V4:
     return !invertPredicate ? Hexagon::LDrih_indexed_shl_cPt_V4 :
                               Hexagon::LDrih_indexed_shl_cNotPt_V4;
   case Hexagon::LDriuh_indexed_shl_V4:
     return !invertPredicate ? Hexagon::LDriuh_indexed_shl_cPt_V4 :
                               Hexagon::LDriuh_indexed_shl_cNotPt_V4;
-  case Hexagon::LDriuh_ae_indexed_shl_V4:
-    return !invertPredicate ? Hexagon::LDriuh_indexed_shl_cPt_V4 :
-                              Hexagon::LDriuh_indexed_shl_cNotPt_V4;
-  case Hexagon::LDriw_indexed_V4:
-    return !invertPredicate ? Hexagon::LDriw_indexed_cPt_V4 :
-                              Hexagon::LDriw_indexed_cNotPt_V4;
   case Hexagon::LDriw_indexed_shl_V4:
     return !invertPredicate ? Hexagon::LDriw_indexed_shl_cPt_V4 :
                               Hexagon::LDriw_indexed_shl_cNotPt_V4;
 
   // V4 Load from global address
-  case Hexagon::LDrid_GP_V4:
-    return !invertPredicate ? Hexagon::LDrid_GP_cPt_V4 :
-                              Hexagon::LDrid_GP_cNotPt_V4;
-  case Hexagon::LDrib_GP_V4:
-    return !invertPredicate ? Hexagon::LDrib_GP_cPt_V4 :
-                              Hexagon::LDrib_GP_cNotPt_V4;
-  case Hexagon::LDriub_GP_V4:
-    return !invertPredicate ? Hexagon::LDriub_GP_cPt_V4 :
-                              Hexagon::LDriub_GP_cNotPt_V4;
-  case Hexagon::LDrih_GP_V4:
-    return !invertPredicate ? Hexagon::LDrih_GP_cPt_V4 :
-                              Hexagon::LDrih_GP_cNotPt_V4;
-  case Hexagon::LDriuh_GP_V4:
-    return !invertPredicate ? Hexagon::LDriuh_GP_cPt_V4 :
-                              Hexagon::LDriuh_GP_cNotPt_V4;
-  case Hexagon::LDriw_GP_V4:
-    return !invertPredicate ? Hexagon::LDriw_GP_cPt_V4 :
-                              Hexagon::LDriw_GP_cNotPt_V4;
-
   case Hexagon::LDd_GP_V4:
     return !invertPredicate ? Hexagon::LDd_GP_cPt_V4 :
                               Hexagon::LDd_GP_cNotPt_V4;
@@ -2122,19 +1636,6 @@ getMatchingCondBranchOpcode(int Opc, bool invertPredicate) const {
                               Hexagon::STrid_indexed_shl_cNotPt_V4;
 
   // V4 Store to global address
-  case Hexagon::STrid_GP_V4:
-    return !invertPredicate ? Hexagon::STrid_GP_cPt_V4 :
-                              Hexagon::STrid_GP_cNotPt_V4;
-  case Hexagon::STrib_GP_V4:
-    return !invertPredicate ? Hexagon::STrib_GP_cPt_V4 :
-                              Hexagon::STrib_GP_cNotPt_V4;
-  case Hexagon::STrih_GP_V4:
-    return !invertPredicate ? Hexagon::STrih_GP_cPt_V4 :
-                              Hexagon::STrih_GP_cNotPt_V4;
-  case Hexagon::STriw_GP_V4:
-    return !invertPredicate ? Hexagon::STriw_GP_cPt_V4 :
-                              Hexagon::STriw_GP_cNotPt_V4;
-
   case Hexagon::STd_GP_V4:
     return !invertPredicate ? Hexagon::STd_GP_cPt_V4 :
                               Hexagon::STd_GP_cNotPt_V4;
@@ -2221,38 +1722,141 @@ PredicateInstruction(MachineInstr *MI,
   assert (isPredicable(MI) && "Expected predicable instruction");
   bool invertJump = (!Cond.empty() && Cond[0].isImm() &&
                      (Cond[0].getImm() == 0));
+
+  // This will change MI's opcode to its predicate version.
+  // However, its operand list is still the old one, i.e. the
+  // non-predicate one.
   MI->setDesc(get(getMatchingCondBranchOpcode(Opc, invertJump)));
-  //
-  // This assumes that the predicate is always the first operand
-  // in the set of inputs.
-  //
-  MI->addOperand(MI->getOperand(MI->getNumOperands()-1));
-  int oper;
-  for (oper = MI->getNumOperands() - 3; oper >= 0; --oper) {
-    MachineOperand MO = MI->getOperand(oper);
-    if ((MO.isReg() && !MO.isUse() && !MO.isImplicit())) {
-      break;
-    }
 
-    if (MO.isReg()) {
-      MI->getOperand(oper+1).ChangeToRegister(MO.getReg(), MO.isDef(),
-                                              MO.isImplicit(), MO.isKill(),
-                                              MO.isDead(), MO.isUndef(),
-                                              MO.isDebug());
-    } else if (MO.isImm()) {
-      MI->getOperand(oper+1).ChangeToImmediate(MO.getImm());
-    } else {
-      llvm_unreachable("Unexpected operand type");
+  int oper = -1;
+  unsigned int GAIdx = 0;
+
+  // Indicates whether the current MI has a GlobalAddress operand
+  bool hasGAOpnd = false;
+  std::vector<MachineOperand> tmpOpnds;
+
+  // Indicates whether we need to shift operands to right.
+  bool needShift = true;
+
+  // The predicate is ALWAYS the FIRST input operand !!!
+  if (MI->getNumOperands() == 0) {
+    // The non-predicate version of MI does not take any operands,
+    // i.e. no outs and no ins. In this condition, the predicate
+    // operand will be directly placed at Operands[0]. No operand
+    // shift is needed.
+    // Example: BARRIER
+    needShift = false;
+    oper = -1;
+  }
+  else if (   MI->getOperand(MI->getNumOperands()-1).isReg()
+           && MI->getOperand(MI->getNumOperands()-1).isDef()
+           && !MI->getOperand(MI->getNumOperands()-1).isImplicit()) {
+    // The non-predicate version of MI does not have any input operands.
+    // In this condition, we extend the length of Operands[] by one and
+    // copy the original last operand to the newly allocated slot.
+    // At this moment, it is just a place holder. Later, we will put
+    // predicate operand directly into it. No operand shift is needed.
+    // Example: r0=BARRIER (this is a faked insn used here for illustration)
+    MI->addOperand(MI->getOperand(MI->getNumOperands()-1));
+    needShift = false;
+    oper = MI->getNumOperands() - 2;
+  }
+  else {
+    // We need to right shift all input operands by one. Duplicate the
+    // last operand into the newly allocated slot.
+    MI->addOperand(MI->getOperand(MI->getNumOperands()-1));
+  }
+
+  if (needShift)
+  {
+    // Operands[ MI->getNumOperands() - 2 ] has been copied into
+    // Operands[ MI->getNumOperands() - 1 ], so we start from
+    // Operands[ MI->getNumOperands() - 3 ].
+    // oper is a signed int.
+    // It is ok if "MI->getNumOperands()-3" is -3, -2, or -1.
+    for (oper = MI->getNumOperands() - 3; oper >= 0; --oper)
+    {
+      MachineOperand &MO = MI->getOperand(oper);
+
+      // Opnd[0] Opnd[1] Opnd[2] Opnd[3] Opnd[4]   Opnd[5]   Opnd[6]   Opnd[7]
+      // <Def0>  <Def1>  <Use0>  <Use1>  <ImpDef0> <ImpDef1> <ImpUse0> <ImpUse1>
+      //               /\~
+      //              /||\~
+      //               ||
+      //        Predicate Operand here
+      if (MO.isReg() && !MO.isUse() && !MO.isImplicit()) {
+        break;
+      }
+      if (MO.isReg()) {
+        MI->getOperand(oper+1).ChangeToRegister(MO.getReg(), MO.isDef(),
+                                                MO.isImplicit(), MO.isKill(),
+                                                MO.isDead(), MO.isUndef(),
+                                                MO.isDebug());
+      }
+      else if (MO.isImm()) {
+        MI->getOperand(oper+1).ChangeToImmediate(MO.getImm());
+      }
+      else if (MO.isGlobal()) {
+        // MI can not have more than one GlobalAddress operand.
+        assert(hasGAOpnd == false && "MI can only have one GlobalAddress opnd");
+
+        // There is no member function called "ChangeToGlobalAddress" in the
+        // MachineOperand class (not like "ChangeToRegister" and
+        // "ChangeToImmediate"). So we have to remove them from Operands[] list
+        // first, and then add them back after we have inserted the predicate
+        // operand. tmpOpnds[] is to remember these operands before we remove
+        // them.
+        tmpOpnds.push_back(MO);
+
+        // Operands[oper] is a GlobalAddress operand;
+        // Operands[oper+1] has been copied into Operands[oper+2];
+        hasGAOpnd = true;
+        GAIdx = oper;
+        continue;
+      }
+      else {
+        assert(false && "Unexpected operand type");
+      }
     }
   }
 
   int regPos = invertJump ? 1 : 0;
   MachineOperand PredMO = Cond[regPos];
+
+  // [oper] now points to the last explicit Def. Predicate operand must be
+  // located at [oper+1]. See diagram above.
+  // This assumes that the predicate is always the first operand,
+  // i.e. Operands[0+numResults], in the set of inputs
+  // It is better to have an assert here to check this. But I don't know how
+  // to write this assert because findFirstPredOperandIdx() would return -1
+  if (oper < -1) oper = -1;
   MI->getOperand(oper+1).ChangeToRegister(PredMO.getReg(), PredMO.isDef(),
                                           PredMO.isImplicit(), PredMO.isKill(),
                                           PredMO.isDead(), PredMO.isUndef(),
                                           PredMO.isDebug());
 
+  if (hasGAOpnd)
+  {
+    unsigned int i;
+
+    // Operands[GAIdx] is the original GlobalAddress operand, which is
+    // already copied into tmpOpnds[0].
+    // Operands[GAIdx] now stores a copy of Operands[GAIdx-1]
+    // Operands[GAIdx+1] has already been copied into Operands[GAIdx+2],
+    // so we start from [GAIdx+2]
+    for (i = GAIdx + 2; i < MI->getNumOperands(); ++i)
+      tmpOpnds.push_back(MI->getOperand(i));
+
+    // Remove all operands in range [ (GAIdx+1) ... (MI->getNumOperands()-1) ]
+    // It is very important that we always remove from the end of Operands[]
+    // MI->getNumOperands() is at least 2 if program goes to here.
+    for (i = MI->getNumOperands() - 1; i > GAIdx; --i)
+      MI->RemoveOperand(i);
+
+    for (i = 0; i < tmpOpnds.size(); ++i)
+      MI->addOperand(tmpOpnds[i]);
+  }
+
   return true;
 }
 
@@ -2286,6 +1890,13 @@ bool HexagonInstrInfo::isPredicated(const MachineInstr *MI) const {
   return ((F >> HexagonII::PredicatedPos) & HexagonII::PredicatedMask);
 }
 
+bool HexagonInstrInfo::isPredicatedNew(const MachineInstr *MI) const {
+  const uint64_t F = MI->getDesc().TSFlags;
+
+  assert(isPredicated(MI));
+  return ((F >> HexagonII::PredicatedNewPos) & HexagonII::PredicatedNewMask);
+}
+
 bool
 HexagonInstrInfo::DefinesPredicate(MachineInstr *MI,
                                    std::vector<MachineOperand> &Pred) const {
@@ -2354,29 +1965,34 @@ isValidOffset(const int Opcode, const int Offset) const {
   // the given "Opcode". If "Offset" is not in the correct range, "ADD_ri" is
   // inserted to calculate the final address. Due to this reason, the function
   // assumes that the "Offset" has correct alignment.
+  // We used to assert if the offset was not properly aligned, however,
+  // there are cases where a misaligned pointer recast can cause this
+  // problem, and we need to allow for it. The front end warns of such
+  // misaligns with respect to load size.
 
   switch(Opcode) {
 
   case Hexagon::LDriw:
+  case Hexagon::LDriw_indexed:
   case Hexagon::LDriw_f:
+  case Hexagon::STriw_indexed:
   case Hexagon::STriw:
   case Hexagon::STriw_f:
-    assert((Offset % 4 == 0) && "Offset has incorrect alignment");
     return (Offset >= Hexagon_MEMW_OFFSET_MIN) &&
       (Offset <= Hexagon_MEMW_OFFSET_MAX);
 
   case Hexagon::LDrid:
+  case Hexagon::LDrid_indexed:
   case Hexagon::LDrid_f:
   case Hexagon::STrid:
+  case Hexagon::STrid_indexed:
   case Hexagon::STrid_f:
-    assert((Offset % 8 == 0) && "Offset has incorrect alignment");
     return (Offset >= Hexagon_MEMD_OFFSET_MIN) &&
       (Offset <= Hexagon_MEMD_OFFSET_MAX);
 
   case Hexagon::LDrih:
   case Hexagon::LDriuh:
   case Hexagon::STrih:
-    assert((Offset % 2 == 0) && "Offset has incorrect alignment");
     return (Offset >= Hexagon_MEMH_OFFSET_MIN) &&
       (Offset <= Hexagon_MEMH_OFFSET_MAX);
 
@@ -2391,54 +2007,28 @@ isValidOffset(const int Opcode, const int Offset) const {
     return (Offset >= Hexagon_ADDI_OFFSET_MIN) &&
       (Offset <= Hexagon_ADDI_OFFSET_MAX);
 
-  case Hexagon::MEMw_ADDSUBi_indexed_MEM_V4 :
-  case Hexagon::MEMw_ADDi_indexed_MEM_V4 :
-  case Hexagon::MEMw_SUBi_indexed_MEM_V4 :
-  case Hexagon::MEMw_ADDr_indexed_MEM_V4 :
-  case Hexagon::MEMw_SUBr_indexed_MEM_V4 :
-  case Hexagon::MEMw_ANDr_indexed_MEM_V4 :
-  case Hexagon::MEMw_ORr_indexed_MEM_V4 :
-  case Hexagon::MEMw_ADDSUBi_MEM_V4 :
-  case Hexagon::MEMw_ADDi_MEM_V4 :
-  case Hexagon::MEMw_SUBi_MEM_V4 :
-  case Hexagon::MEMw_ADDr_MEM_V4 :
-  case Hexagon::MEMw_SUBr_MEM_V4 :
-  case Hexagon::MEMw_ANDr_MEM_V4 :
-  case Hexagon::MEMw_ORr_MEM_V4 :
-    assert ((Offset % 4) == 0 && "MEMOPw offset is not aligned correctly." );
+  case Hexagon::MemOPw_ADDi_V4 :
+  case Hexagon::MemOPw_SUBi_V4 :
+  case Hexagon::MemOPw_ADDr_V4 :
+  case Hexagon::MemOPw_SUBr_V4 :
+  case Hexagon::MemOPw_ANDr_V4 :
+  case Hexagon::MemOPw_ORr_V4 :
     return (0 <= Offset && Offset <= 255);
 
-  case Hexagon::MEMh_ADDSUBi_indexed_MEM_V4 :
-  case Hexagon::MEMh_ADDi_indexed_MEM_V4 :
-  case Hexagon::MEMh_SUBi_indexed_MEM_V4 :
-  case Hexagon::MEMh_ADDr_indexed_MEM_V4 :
-  case Hexagon::MEMh_SUBr_indexed_MEM_V4 :
-  case Hexagon::MEMh_ANDr_indexed_MEM_V4 :
-  case Hexagon::MEMh_ORr_indexed_MEM_V4 :
-  case Hexagon::MEMh_ADDSUBi_MEM_V4 :
-  case Hexagon::MEMh_ADDi_MEM_V4 :
-  case Hexagon::MEMh_SUBi_MEM_V4 :
-  case Hexagon::MEMh_ADDr_MEM_V4 :
-  case Hexagon::MEMh_SUBr_MEM_V4 :
-  case Hexagon::MEMh_ANDr_MEM_V4 :
-  case Hexagon::MEMh_ORr_MEM_V4 :
-    assert ((Offset % 2) == 0 && "MEMOPh offset is not aligned correctly." );
+  case Hexagon::MemOPh_ADDi_V4 :
+  case Hexagon::MemOPh_SUBi_V4 :
+  case Hexagon::MemOPh_ADDr_V4 :
+  case Hexagon::MemOPh_SUBr_V4 :
+  case Hexagon::MemOPh_ANDr_V4 :
+  case Hexagon::MemOPh_ORr_V4 :
     return (0 <= Offset && Offset <= 127);
 
-  case Hexagon::MEMb_ADDSUBi_indexed_MEM_V4 :
-  case Hexagon::MEMb_ADDi_indexed_MEM_V4 :
-  case Hexagon::MEMb_SUBi_indexed_MEM_V4 :
-  case Hexagon::MEMb_ADDr_indexed_MEM_V4 :
-  case Hexagon::MEMb_SUBr_indexed_MEM_V4 :
-  case Hexagon::MEMb_ANDr_indexed_MEM_V4 :
-  case Hexagon::MEMb_ORr_indexed_MEM_V4 :
-  case Hexagon::MEMb_ADDSUBi_MEM_V4 :
-  case Hexagon::MEMb_ADDi_MEM_V4 :
-  case Hexagon::MEMb_SUBi_MEM_V4 :
-  case Hexagon::MEMb_ADDr_MEM_V4 :
-  case Hexagon::MEMb_SUBr_MEM_V4 :
-  case Hexagon::MEMb_ANDr_MEM_V4 :
-  case Hexagon::MEMb_ORr_MEM_V4 :
+  case Hexagon::MemOPb_ADDi_V4 :
+  case Hexagon::MemOPb_SUBi_V4 :
+  case Hexagon::MemOPb_ADDr_V4 :
+  case Hexagon::MemOPb_SUBr_V4 :
+  case Hexagon::MemOPb_ANDr_V4 :
+  case Hexagon::MemOPb_ORr_V4 :
     return (0 <= Offset && Offset <= 63);
 
   // LDri_pred and STriw_pred are pseudo operations, so it has to take offset of
@@ -2447,6 +2037,9 @@ isValidOffset(const int Opcode, const int Offset) const {
   case Hexagon::LDriw_pred:
     return true;
 
+  case Hexagon::LOOP0_i:
+    return isUInt<10>(Offset);
+
   // INLINEASM is very special.
   case Hexagon::INLINEASM:
     return true;
@@ -2491,50 +2084,33 @@ isMemOp(const MachineInstr *MI) const {
   switch (MI->getOpcode())
   {
     default: return false;
-    case Hexagon::MEMw_ADDSUBi_indexed_MEM_V4 :
-    case Hexagon::MEMw_ADDi_indexed_MEM_V4 :
-    case Hexagon::MEMw_SUBi_indexed_MEM_V4 :
-    case Hexagon::MEMw_ADDr_indexed_MEM_V4 :
-    case Hexagon::MEMw_SUBr_indexed_MEM_V4 :
-    case Hexagon::MEMw_ANDr_indexed_MEM_V4 :
-    case Hexagon::MEMw_ORr_indexed_MEM_V4 :
-    case Hexagon::MEMw_ADDSUBi_MEM_V4 :
-    case Hexagon::MEMw_ADDi_MEM_V4 :
-    case Hexagon::MEMw_SUBi_MEM_V4 :
-    case Hexagon::MEMw_ADDr_MEM_V4 :
-    case Hexagon::MEMw_SUBr_MEM_V4 :
-    case Hexagon::MEMw_ANDr_MEM_V4 :
-    case Hexagon::MEMw_ORr_MEM_V4 :
-    case Hexagon::MEMh_ADDSUBi_indexed_MEM_V4 :
-    case Hexagon::MEMh_ADDi_indexed_MEM_V4 :
-    case Hexagon::MEMh_SUBi_indexed_MEM_V4 :
-    case Hexagon::MEMh_ADDr_indexed_MEM_V4 :
-    case Hexagon::MEMh_SUBr_indexed_MEM_V4 :
-    case Hexagon::MEMh_ANDr_indexed_MEM_V4 :
-    case Hexagon::MEMh_ORr_indexed_MEM_V4 :
-    case Hexagon::MEMh_ADDSUBi_MEM_V4 :
-    case Hexagon::MEMh_ADDi_MEM_V4 :
-    case Hexagon::MEMh_SUBi_MEM_V4 :
-    case Hexagon::MEMh_ADDr_MEM_V4 :
-    case Hexagon::MEMh_SUBr_MEM_V4 :
-    case Hexagon::MEMh_ANDr_MEM_V4 :
-    case Hexagon::MEMh_ORr_MEM_V4 :
-    case Hexagon::MEMb_ADDSUBi_indexed_MEM_V4 :
-    case Hexagon::MEMb_ADDi_indexed_MEM_V4 :
-    case Hexagon::MEMb_SUBi_indexed_MEM_V4 :
-    case Hexagon::MEMb_ADDr_indexed_MEM_V4 :
-    case Hexagon::MEMb_SUBr_indexed_MEM_V4 :
-    case Hexagon::MEMb_ANDr_indexed_MEM_V4 :
-    case Hexagon::MEMb_ORr_indexed_MEM_V4 :
-    case Hexagon::MEMb_ADDSUBi_MEM_V4 :
-    case Hexagon::MEMb_ADDi_MEM_V4 :
-    case Hexagon::MEMb_SUBi_MEM_V4 :
-    case Hexagon::MEMb_ADDr_MEM_V4 :
-    case Hexagon::MEMb_SUBr_MEM_V4 :
-    case Hexagon::MEMb_ANDr_MEM_V4 :
-    case Hexagon::MEMb_ORr_MEM_V4 :
-      return true;
+    case Hexagon::MemOPw_ADDi_V4 :
+    case Hexagon::MemOPw_SUBi_V4 :
+    case Hexagon::MemOPw_ADDr_V4 :
+    case Hexagon::MemOPw_SUBr_V4 :
+    case Hexagon::MemOPw_ANDr_V4 :
+    case Hexagon::MemOPw_ORr_V4 :
+    case Hexagon::MemOPh_ADDi_V4 :
+    case Hexagon::MemOPh_SUBi_V4 :
+    case Hexagon::MemOPh_ADDr_V4 :
+    case Hexagon::MemOPh_SUBr_V4 :
+    case Hexagon::MemOPh_ANDr_V4 :
+    case Hexagon::MemOPh_ORr_V4 :
+    case Hexagon::MemOPb_ADDi_V4 :
+    case Hexagon::MemOPb_SUBi_V4 :
+    case Hexagon::MemOPb_ADDr_V4 :
+    case Hexagon::MemOPb_SUBr_V4 :
+    case Hexagon::MemOPb_ANDr_V4 :
+    case Hexagon::MemOPb_ORr_V4 :
+    case Hexagon::MemOPb_SETBITi_V4:
+    case Hexagon::MemOPh_SETBITi_V4:
+    case Hexagon::MemOPw_SETBITi_V4:
+    case Hexagon::MemOPb_CLRBITi_V4:
+    case Hexagon::MemOPh_CLRBITi_V4:
+    case Hexagon::MemOPw_CLRBITi_V4:
+    return true;
   }
+  return false;
 }
 
 
@@ -2661,28 +2237,16 @@ isConditionalLoad (const MachineInstr* MI) const {
     case Hexagon::POST_LDriub_cPt :
     case Hexagon::POST_LDriub_cNotPt :
       return QRI.Subtarget.hasV4TOps();
-    case Hexagon::LDrid_indexed_cPt_V4 :
-    case Hexagon::LDrid_indexed_cNotPt_V4 :
     case Hexagon::LDrid_indexed_shl_cPt_V4 :
     case Hexagon::LDrid_indexed_shl_cNotPt_V4 :
-    case Hexagon::LDrib_indexed_cPt_V4 :
-    case Hexagon::LDrib_indexed_cNotPt_V4 :
     case Hexagon::LDrib_indexed_shl_cPt_V4 :
     case Hexagon::LDrib_indexed_shl_cNotPt_V4 :
-    case Hexagon::LDriub_indexed_cPt_V4 :
-    case Hexagon::LDriub_indexed_cNotPt_V4 :
     case Hexagon::LDriub_indexed_shl_cPt_V4 :
     case Hexagon::LDriub_indexed_shl_cNotPt_V4 :
-    case Hexagon::LDrih_indexed_cPt_V4 :
-    case Hexagon::LDrih_indexed_cNotPt_V4 :
     case Hexagon::LDrih_indexed_shl_cPt_V4 :
     case Hexagon::LDrih_indexed_shl_cNotPt_V4 :
-    case Hexagon::LDriuh_indexed_cPt_V4 :
-    case Hexagon::LDriuh_indexed_cNotPt_V4 :
     case Hexagon::LDriuh_indexed_shl_cPt_V4 :
     case Hexagon::LDriuh_indexed_shl_cNotPt_V4 :
-    case Hexagon::LDriw_indexed_cPt_V4 :
-    case Hexagon::LDriw_indexed_cNotPt_V4 :
     case Hexagon::LDriw_indexed_shl_cPt_V4 :
     case Hexagon::LDriw_indexed_shl_cNotPt_V4 :
       return QRI.Subtarget.hasV4TOps();
@@ -2765,14 +2329,6 @@ isConditionalStore (const MachineInstr* MI) const {
       return QRI.Subtarget.hasV4TOps();
 
     // V4 global address store before promoting to dot new.
-    case Hexagon::STrid_GP_cPt_V4 :
-    case Hexagon::STrid_GP_cNotPt_V4 :
-    case Hexagon::STrib_GP_cPt_V4 :
-    case Hexagon::STrib_GP_cNotPt_V4 :
-    case Hexagon::STrih_GP_cPt_V4 :
-    case Hexagon::STrih_GP_cNotPt_V4 :
-    case Hexagon::STriw_GP_cPt_V4 :
-    case Hexagon::STriw_GP_cNotPt_V4 :
     case Hexagon::STd_GP_cPt_V4 :
     case Hexagon::STd_GP_cNotPt_V4 :
     case Hexagon::STb_GP_cPt_V4 :
@@ -2813,7 +2369,33 @@ isConditionalStore (const MachineInstr* MI) const {
   }
 }
 
+// Returns true, if any one of the operands is a dot new
+// insn, whether it is predicated dot new or register dot new.
+bool HexagonInstrInfo::isDotNewInst (const MachineInstr* MI) const {
+  return (isNewValueInst(MI) ||
+     (isPredicated(MI) && isPredicatedNew(MI)));
+}
+
+unsigned HexagonInstrInfo::getAddrMode(const MachineInstr* MI) const {
+  const uint64_t F = MI->getDesc().TSFlags;
+
+  return((F >> HexagonII::AddrModePos) & HexagonII::AddrModeMask);
+}
 
+/// immediateExtend - Changes the instruction in place to one using an immediate
+/// extender.
+void HexagonInstrInfo::immediateExtend(MachineInstr *MI) const {
+  assert((isExtendable(MI)||isConstExtended(MI)) &&
+                               "Instruction must be extendable");
+  // Find which operand is extendable.
+  short ExtOpNum = getCExtOpNum(MI);
+  MachineOperand &MO = MI->getOperand(ExtOpNum);
+  // This needs to be something we understand.
+  assert((MO.isMBB() || MO.isImm()) &&
+         "Branch with unknown extendable field type");
+  // Mark given operand as extended.
+  MO.addTargetFlag(HexagonII::HMOTF_ConstExtended);
+}
 
 DFAPacketizer *HexagonInstrInfo::
 CreateTargetScheduleState(const TargetMachine *TM,
@@ -2840,3 +2422,155 @@ bool HexagonInstrInfo::isSchedulingBoundary(const MachineInstr *MI,
 
   return false;
 }
+
+bool HexagonInstrInfo::isConstExtended(MachineInstr *MI) const {
+
+  // Constant extenders are allowed only for V4 and above.
+  if (!Subtarget.hasV4TOps())
+    return false;
+
+  const uint64_t F = MI->getDesc().TSFlags;
+  unsigned isExtended = (F >> HexagonII::ExtendedPos) & HexagonII::ExtendedMask;
+  if (isExtended) // Instruction must be extended.
+    return true;
+
+  unsigned isExtendable = (F >> HexagonII::ExtendablePos)
+                          & HexagonII::ExtendableMask;
+  if (!isExtendable)
+    return false;
+
+  short ExtOpNum = getCExtOpNum(MI);
+  const MachineOperand &MO = MI->getOperand(ExtOpNum);
+  // Use MO operand flags to determine if MO
+  // has the HMOTF_ConstExtended flag set.
+  if (MO.getTargetFlags() && HexagonII::HMOTF_ConstExtended)
+    return true;
+  // If this is a Machine BB address we are talking about, and it is
+  // not marked as extended, say so.
+  if (MO.isMBB())
+    return false;
+
+  // We could be using an instruction with an extendable immediate and shoehorn
+  // a global address into it. If it is a global address it will be constant
+  // extended. We do this for COMBINE.
+  // We currently only handle isGlobal() because it is the only kind of
+  // object we are going to end up with here for now.
+  // In the future we probably should add isSymbol(), etc.
+  if (MO.isGlobal() || MO.isSymbol())
+    return true;
+
+  // If the extendable operand is not 'Immediate' type, the instruction should
+  // have 'isExtended' flag set.
+  assert(MO.isImm() && "Extendable operand must be Immediate type");
+
+  int MinValue = getMinValue(MI);
+  int MaxValue = getMaxValue(MI);
+  int ImmValue = MO.getImm();
+
+  return (ImmValue < MinValue || ImmValue > MaxValue);
+}
+
+// Returns true if a particular operand is extendable for an instruction.
+bool HexagonInstrInfo::isOperandExtended(const MachineInstr *MI,
+                                         unsigned short OperandNum) const {
+  // Constant extenders are allowed only for V4 and above.
+  if (!Subtarget.hasV4TOps())
+    return false;
+
+  const uint64_t F = MI->getDesc().TSFlags;
+
+  return ((F >> HexagonII::ExtendableOpPos) & HexagonII::ExtendableOpMask)
+          == OperandNum;
+}
+
+// Returns Operand Index for the constant extended instruction.
+unsigned short HexagonInstrInfo::getCExtOpNum(const MachineInstr *MI) const {
+  const uint64_t F = MI->getDesc().TSFlags;
+  return ((F >> HexagonII::ExtendableOpPos) & HexagonII::ExtendableOpMask);
+}
+
+// Returns the min value that doesn't need to be extended.
+int HexagonInstrInfo::getMinValue(const MachineInstr *MI) const {
+  const uint64_t F = MI->getDesc().TSFlags;
+  unsigned isSigned = (F >> HexagonII::ExtentSignedPos)
+                    & HexagonII::ExtentSignedMask;
+  unsigned bits =  (F >> HexagonII::ExtentBitsPos)
+                    & HexagonII::ExtentBitsMask;
+
+  if (isSigned) // if value is signed
+    return -1 << (bits - 1);
+  else
+    return 0;
+}
+
+// Returns the max value that doesn't need to be extended.
+int HexagonInstrInfo::getMaxValue(const MachineInstr *MI) const {
+  const uint64_t F = MI->getDesc().TSFlags;
+  unsigned isSigned = (F >> HexagonII::ExtentSignedPos)
+                    & HexagonII::ExtentSignedMask;
+  unsigned bits =  (F >> HexagonII::ExtentBitsPos)
+                    & HexagonII::ExtentBitsMask;
+
+  if (isSigned) // if value is signed
+    return ~(-1 << (bits - 1));
+  else
+    return ~(-1 << bits);
+}
+
+// Returns true if an instruction can be converted into a non-extended
+// equivalent instruction.
+bool HexagonInstrInfo::NonExtEquivalentExists (const MachineInstr *MI) const {
+
+  short NonExtOpcode;
+  // Check if the instruction has a register form that uses register in place
+  // of the extended operand, if so return that as the non-extended form.
+  if (Hexagon::getRegForm(MI->getOpcode()) >= 0)
+    return true;
+
+  if (MI->getDesc().mayLoad() || MI->getDesc().mayStore()) {
+    // Check addressing mode and retreive non-ext equivalent instruction.
+
+    switch (getAddrMode(MI)) {
+    case HexagonII::Absolute :
+      // Load/store with absolute addressing mode can be converted into
+      // base+offset mode.
+      NonExtOpcode = Hexagon::getBasedWithImmOffset(MI->getOpcode());
+      break;
+    case HexagonII::BaseImmOffset :
+      // Load/store with base+offset addressing mode can be converted into
+      // base+register offset addressing mode. However left shift operand should
+      // be set to 0.
+      NonExtOpcode = Hexagon::getBaseWithRegOffset(MI->getOpcode());
+      break;
+    default:
+      return false;
+    }
+    if (NonExtOpcode < 0)
+      return false;
+    return true;
+  }
+  return false;
+}
+
+// Returns opcode of the non-extended equivalent instruction.
+short HexagonInstrInfo::getNonExtOpcode (const MachineInstr *MI) const {
+
+  // Check if the instruction has a register form that uses register in place
+  // of the extended operand, if so return that as the non-extended form.
+  short NonExtOpcode = Hexagon::getRegForm(MI->getOpcode());
+    if (NonExtOpcode >= 0)
+      return NonExtOpcode;
+
+  if (MI->getDesc().mayLoad() || MI->getDesc().mayStore()) {
+    // Check addressing mode and retreive non-ext equivalent instruction.
+    switch (getAddrMode(MI)) {
+    case HexagonII::Absolute :
+      return Hexagon::getBasedWithImmOffset(MI->getOpcode());
+    case HexagonII::BaseImmOffset :
+      return Hexagon::getBaseWithRegOffset(MI->getOpcode());
+    default:
+      return -1;
+    }
+  }
+  return -1;
+}
diff --git a/lib/Target/Hexagon/HexagonInstrInfo.h b/lib/Target/Hexagon/HexagonInstrInfo.h
index 2bb53f899ce1..5df13a88b5d3 100644
--- a/lib/Target/Hexagon/HexagonInstrInfo.h
+++ b/lib/Target/Hexagon/HexagonInstrInfo.h
@@ -16,8 +16,8 @@
 
 #include "HexagonRegisterInfo.h"
 #include "MCTargetDesc/HexagonBaseInfo.h"
-#include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetInstrInfo.h"
 
 
 #define GET_INSTRINFO_HEADER
@@ -66,6 +66,10 @@ public:
                                 const SmallVectorImpl<MachineOperand> &Cond,
                                 DebugLoc DL) const;
 
+  virtual bool analyzeCompare(const MachineInstr *MI,
+                              unsigned &SrcReg, unsigned &SrcReg2,
+                              int &Mask, int &Value) const;
+
   virtual void copyPhysReg(MachineBasicBlock &MBB,
                            MachineBasicBlock::iterator I, DebugLoc DL,
                            unsigned DestReg, unsigned SrcReg,
@@ -123,6 +127,7 @@ public:
                                    const BranchProbability &Probability) const;
 
   virtual bool isPredicated(const MachineInstr *MI) const;
+  virtual bool isPredicatedNew(const MachineInstr *MI) const;
   virtual bool DefinesPredicate(MachineInstr *MI,
                                 std::vector<MachineOperand> &Pred) const;
   virtual bool
@@ -136,6 +141,11 @@ public:
   isProfitableToDupForIfCvt(MachineBasicBlock &MBB,unsigned NumCycles,
                             const BranchProbability &Probability) const;
 
+  virtual MachineInstr *emitFrameIndexDebugValue(MachineFunction &MF,
+                                                 int FrameIx,
+                                                 uint64_t Offset,
+                                                 const MDNode *MDPtr,
+                                                 DebugLoc DL) const;
   virtual DFAPacketizer*
   CreateTargetScheduleState(const TargetMachine *TM,
                             const ScheduleDAG *DAG) const;
@@ -165,6 +175,8 @@ public:
   bool isConditionalALU32 (const MachineInstr* MI) const;
   bool isConditionalLoad (const MachineInstr* MI) const;
   bool isConditionalStore(const MachineInstr* MI) const;
+  bool isNewValueInst(const MachineInstr* MI) const;
+  bool isDotNewInst(const MachineInstr* MI) const;
   bool isDeallocRet(const MachineInstr *MI) const;
   unsigned getInvertedPredicatedOpcode(const int Opc) const;
   bool isExtendable(const MachineInstr* MI) const;
@@ -173,9 +185,18 @@ public:
   bool isNewValueStore(const MachineInstr* MI) const;
   bool isNewValueJump(const MachineInstr* MI) const;
   bool isNewValueJumpCandidate(const MachineInstr *MI) const;
-  unsigned getImmExtForm(const MachineInstr* MI) const;
-  unsigned getNormalBranchForm(const MachineInstr* MI) const;
 
+
+  void immediateExtend(MachineInstr *MI) const;
+  bool isConstExtended(MachineInstr *MI) const;
+  unsigned getAddrMode(const MachineInstr* MI) const;
+  bool isOperandExtended(const MachineInstr *MI,
+                         unsigned short OperandNum) const;
+  unsigned short getCExtOpNum(const MachineInstr *MI) const;
+  int getMinValue(const MachineInstr *MI) const;
+  int getMaxValue(const MachineInstr *MI) const;
+  bool NonExtEquivalentExists (const MachineInstr *MI) const;
+  short getNonExtOpcode(const MachineInstr *MI) const;
 private:
   int getMatchingCondBranchOpcode(int Opc, bool sense) const;
 
diff --git a/lib/Target/Hexagon/HexagonInstrInfo.td b/lib/Target/Hexagon/HexagonInstrInfo.td
index 1d4a7060adf0..74dc0ca72a04 100644
--- a/lib/Target/Hexagon/HexagonInstrInfo.td
+++ b/lib/Target/Hexagon/HexagonInstrInfo.td
@@ -12,88 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 include "HexagonInstrFormats.td"
-include "HexagonImmediates.td"
-
-//===----------------------------------------------------------------------===//
-// Classes used for relation maps.
-//===----------------------------------------------------------------------===//
-// PredRel - Filter class used to relate non-predicated instructions with their
-// predicated forms.
-class PredRel;
-// PredNewRel - Filter class used to relate predicated instructions with their
-// predicate-new forms.
-class PredNewRel: PredRel;
-// ImmRegRel - Filter class used to relate instructions having reg-reg form
-// with their reg-imm counterparts.
-class ImmRegRel;
-//===----------------------------------------------------------------------===//
-// Hexagon Instruction Predicate Definitions.
-//===----------------------------------------------------------------------===//
-def HasV2T                      : Predicate<"Subtarget.hasV2TOps()">;
-def HasV2TOnly                  : Predicate<"Subtarget.hasV2TOpsOnly()">;
-def NoV2T                       : Predicate<"!Subtarget.hasV2TOps()">;
-def HasV3T                      : Predicate<"Subtarget.hasV3TOps()">;
-def HasV3TOnly                  : Predicate<"Subtarget.hasV3TOpsOnly()">;
-def NoV3T                       : Predicate<"!Subtarget.hasV3TOps()">;
-def HasV4T                      : Predicate<"Subtarget.hasV4TOps()">;
-def NoV4T                       : Predicate<"!Subtarget.hasV4TOps()">;
-def HasV5T                      : Predicate<"Subtarget.hasV5TOps()">;
-def NoV5T                       : Predicate<"!Subtarget.hasV5TOps()">;
-def UseMEMOP                    : Predicate<"Subtarget.useMemOps()">;
-def IEEERndNearV5T              : Predicate<"Subtarget.modeIEEERndNear()">;
-
-// Addressing modes.
-def ADDRrr : ComplexPattern<i32, 2, "SelectADDRrr", [], []>;
-def ADDRri : ComplexPattern<i32, 2, "SelectADDRri", [frameindex], []>;
-def ADDRriS11_0 : ComplexPattern<i32, 2, "SelectADDRriS11_0", [frameindex], []>;
-def ADDRriS11_1 : ComplexPattern<i32, 2, "SelectADDRriS11_1", [frameindex], []>;
-def ADDRriS11_2 : ComplexPattern<i32, 2, "SelectADDRriS11_2", [frameindex], []>;
-def ADDRriS11_3 : ComplexPattern<i32, 2, "SelectADDRriS11_3", [frameindex], []>;
-def ADDRriU6_0 : ComplexPattern<i32, 2, "SelectADDRriU6_0", [frameindex], []>;
-def ADDRriU6_1 : ComplexPattern<i32, 2, "SelectADDRriU6_1", [frameindex], []>;
-def ADDRriU6_2 : ComplexPattern<i32, 2, "SelectADDRriU6_2", [frameindex], []>;
-
-// Address operands.
-def MEMrr : Operand<i32> {
-  let PrintMethod = "printMEMrrOperand";
-  let MIOperandInfo = (ops IntRegs, IntRegs);
-}
-
-// Address operands
-def MEMri : Operand<i32> {
-  let PrintMethod = "printMEMriOperand";
-  let MIOperandInfo = (ops IntRegs, IntRegs);
-}
-
-def MEMri_s11_2 : Operand<i32>,
-  ComplexPattern<i32, 2, "SelectMEMriS11_2", []> {
-  let PrintMethod = "printMEMriOperand";
-  let MIOperandInfo = (ops IntRegs, s11Imm);
-}
-
-def FrameIndex : Operand<i32> {
-  let PrintMethod = "printFrameIndexOperand";
-  let MIOperandInfo = (ops IntRegs, s11Imm);
-}
-
-let PrintMethod = "printGlobalOperand" in
-  def globaladdress : Operand<i32>;
-
-let PrintMethod = "printJumpTable" in
- def jumptablebase : Operand<i32>;
-
-def brtarget : Operand<OtherVT>;
-def calltarget : Operand<i32>;
-
-def bblabel : Operand<i32>;
-def bbl   : SDNode<"ISD::BasicBlock", SDTPtrLeaf   , [], "BasicBlockSDNode">;
-
-def symbolHi32 : Operand<i32> {
-  let PrintMethod = "printSymbolHi";
-}
-def symbolLo32 : Operand<i32> {
-  let PrintMethod = "printSymbolLo";
-}
+include "HexagonOperands.td"
 
 // Multi-class for logical operators.
 multiclass ALU32_rr_ri<string OpcStr, SDNode OpNode> {
@@ -122,40 +41,54 @@ multiclass CMP32_rr<string OpcStr, PatFrag OpNode> {
                        (OpNode (i32 IntRegs:$b), (i32 IntRegs:$c)))]>;
 }
 
-multiclass CMP32_rr_ri_s10<string OpcStr, PatFrag OpNode> {
-  def rr : ALU32_rr<(outs PredRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
-                 !strconcat("$dst = ", !strconcat(OpcStr, "($b, $c)")),
-                 [(set (i1 PredRegs:$dst),
-                       (OpNode (i32 IntRegs:$b), (i32 IntRegs:$c)))]>;
-  def ri : ALU32_ri<(outs PredRegs:$dst), (ins IntRegs:$b, s10Imm:$c),
-                 !strconcat("$dst = ", !strconcat(OpcStr, "($b, #$c)")),
-                 [(set (i1 PredRegs:$dst),
-                       (OpNode (i32 IntRegs:$b), s10ImmPred:$c))]>;
+multiclass CMP32_rr_ri_s10<string OpcStr, string CextOp, PatFrag OpNode> {
+  let CextOpcode = CextOp in {
+    let InputType = "reg" in
+    def rr : ALU32_rr<(outs PredRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
+                   !strconcat("$dst = ", !strconcat(OpcStr, "($b, $c)")),
+                   [(set (i1 PredRegs:$dst),
+                         (OpNode (i32 IntRegs:$b), (i32 IntRegs:$c)))]>;
+
+    let isExtendable = 1, opExtendable = 2, isExtentSigned = 1,
+    opExtentBits = 10, InputType = "imm" in
+    def ri : ALU32_ri<(outs PredRegs:$dst), (ins IntRegs:$b, s10Ext:$c),
+                   !strconcat("$dst = ", !strconcat(OpcStr, "($b, #$c)")),
+                   [(set (i1 PredRegs:$dst),
+                         (OpNode (i32 IntRegs:$b), s10ExtPred:$c))]>;
+  }
 }
 
-multiclass CMP32_rr_ri_u9<string OpcStr, PatFrag OpNode> {
-  def rr : ALU32_rr<(outs PredRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
-                 !strconcat("$dst = ", !strconcat(OpcStr, "($b, $c)")),
-                 [(set (i1 PredRegs:$dst),
-                       (OpNode (i32 IntRegs:$b), (i32 IntRegs:$c)))]>;
-  def ri : ALU32_ri<(outs PredRegs:$dst), (ins IntRegs:$b, u9Imm:$c),
-                 !strconcat("$dst = ", !strconcat(OpcStr, "($b, #$c)")),
-                 [(set (i1 PredRegs:$dst),
-                       (OpNode (i32 IntRegs:$b), u9ImmPred:$c))]>;
+multiclass CMP32_rr_ri_u9<string OpcStr, string CextOp, PatFrag OpNode> {
+  let CextOpcode = CextOp in {
+    let InputType = "reg" in
+    def rr : ALU32_rr<(outs PredRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
+                   !strconcat("$dst = ", !strconcat(OpcStr, "($b, $c)")),
+                   [(set (i1 PredRegs:$dst),
+                         (OpNode (i32 IntRegs:$b), (i32 IntRegs:$c)))]>;
+
+    let isExtendable = 1, opExtendable = 2, isExtentSigned = 0,
+    opExtentBits = 9, InputType = "imm" in
+    def ri : ALU32_ri<(outs PredRegs:$dst), (ins IntRegs:$b, u9Ext:$c),
+                   !strconcat("$dst = ", !strconcat(OpcStr, "($b, #$c)")),
+                   [(set (i1 PredRegs:$dst),
+                         (OpNode (i32 IntRegs:$b), u9ExtPred:$c))]>;
+  }
 }
 
 multiclass CMP32_ri_u8<string OpcStr, PatFrag OpNode> {
-  def ri : ALU32_ri<(outs PredRegs:$dst), (ins IntRegs:$b, u8Imm:$c),
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 8 in
+  def ri : ALU32_ri<(outs PredRegs:$dst), (ins IntRegs:$b, u8Ext:$c),
                  !strconcat("$dst = ", !strconcat(OpcStr, "($b, #$c)")),
                  [(set (i1 PredRegs:$dst), (OpNode (i32 IntRegs:$b),
-                                                   u8ImmPred:$c))]>;
+                                                   u8ExtPred:$c))]>;
 }
 
 multiclass CMP32_ri_s8<string OpcStr, PatFrag OpNode> {
-  def ri : ALU32_ri<(outs PredRegs:$dst), (ins IntRegs:$b, s8Imm:$c),
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8 in
+  def ri : ALU32_ri<(outs PredRegs:$dst), (ins IntRegs:$b, s8Ext:$c),
                  !strconcat("$dst = ", !strconcat(OpcStr, "($b, #$c)")),
                  [(set (i1 PredRegs:$dst), (OpNode (i32 IntRegs:$b),
-                                                   s8ImmPred:$c))]>;
+                                                   s8ExtPred:$c))]>;
 }
 }
 
@@ -165,8 +98,8 @@ multiclass CMP32_ri_s8<string OpcStr, PatFrag OpNode> {
 multiclass ALU32_Pbase<string mnemonic, bit isNot,
                        bit isPredNew> {
 
-  let PNewValue = #!if(isPredNew, "new", "") in
-  def #NAME# : ALU32_rr<(outs IntRegs:$dst),
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : ALU32_rr<(outs IntRegs:$dst),
             (ins PredRegs:$src1, IntRegs:$src2, IntRegs: $src3),
             !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew,".new) $dst = ",
             ") $dst = ")#mnemonic#"($src2, $src3)",
@@ -174,10 +107,10 @@ multiclass ALU32_Pbase<string mnemonic, bit isNot,
 }
 
 multiclass ALU32_Pred<string mnemonic, bit PredNot> {
-  let PredSense = #!if(PredNot, "false", "true") in {
-    defm _c#NAME# : ALU32_Pbase<mnemonic, PredNot, 0>;
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ALU32_Pbase<mnemonic, PredNot, 0>;
     // Predicate new
-    defm _cdn#NAME# : ALU32_Pbase<mnemonic, PredNot, 1>;
+    defm _cdn#NAME : ALU32_Pbase<mnemonic, PredNot, 1>;
   }
 }
 
@@ -185,7 +118,7 @@ let InputType = "reg" in
 multiclass ALU32_base<string mnemonic, string CextOp, SDNode OpNode> {
   let CextOpcode = CextOp, BaseOpcode = CextOp#_rr in {
     let isPredicable = 1 in
-    def #NAME# : ALU32_rr<(outs IntRegs:$dst),
+    def NAME : ALU32_rr<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2),
             "$dst = "#mnemonic#"($src1, $src2)",
             [(set (i32 IntRegs:$dst), (OpNode (i32 IntRegs:$src1),
@@ -211,33 +144,35 @@ defm SUB_rr : ALU32_base<"sub", "SUB", sub>, ImmRegRel, PredNewRel;
 // ALU32/ALU (ADD with register-immediate form)
 //===----------------------------------------------------------------------===//
 multiclass ALU32ri_Pbase<string mnemonic, bit isNot, bit isPredNew> {
-  let PNewValue = #!if(isPredNew, "new", "") in
-  def #NAME# : ALU32_ri<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, s8Imm: $src3),
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : ALU32_ri<(outs IntRegs:$dst),
+            (ins PredRegs:$src1, IntRegs:$src2, s8Ext: $src3),
             !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew,".new) $dst = ",
             ") $dst = ")#mnemonic#"($src2, #$src3)",
             []>;
 }
 
 multiclass ALU32ri_Pred<string mnemonic, bit PredNot> {
-  let PredSense = #!if(PredNot, "false", "true") in {
-    defm _c#NAME# : ALU32ri_Pbase<mnemonic, PredNot, 0>;
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ALU32ri_Pbase<mnemonic, PredNot, 0>;
     // Predicate new
-    defm _cdn#NAME# : ALU32ri_Pbase<mnemonic, PredNot, 1>;
+    defm _cdn#NAME : ALU32ri_Pbase<mnemonic, PredNot, 1>;
   }
 }
 
-let InputType = "imm" in
+let isExtendable = 1, InputType = "imm" in
 multiclass ALU32ri_base<string mnemonic, string CextOp, SDNode OpNode> {
   let CextOpcode = CextOp, BaseOpcode = CextOp#_ri in {
-    let isPredicable = 1 in
-    def #NAME# : ALU32_ri<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, s16Imm:$src2),
+    let opExtendable = 2, isExtentSigned = 1, opExtentBits = 16,
+    isPredicable = 1 in
+    def NAME : ALU32_ri<(outs IntRegs:$dst),
+            (ins IntRegs:$src1, s16Ext:$src2),
             "$dst = "#mnemonic#"($src1, #$src2)",
             [(set (i32 IntRegs:$dst), (OpNode (i32 IntRegs:$src1),
-                                              (s16ImmPred:$src2)))]>;
+                                              (s16ExtPred:$src2)))]>;
 
-    let neverHasSideEffects = 1, isPredicated = 1 in {
+    let opExtendable = 3, isExtentSigned = 1, opExtentBits = 8,
+    neverHasSideEffects = 1, isPredicated = 1 in {
       defm Pt : ALU32ri_Pred<mnemonic, 0>;
       defm NotPt : ALU32ri_Pred<mnemonic, 1>;
     }
@@ -246,23 +181,26 @@ multiclass ALU32ri_base<string mnemonic, string CextOp, SDNode OpNode> {
 
 defm ADD_ri : ALU32ri_base<"add", "ADD", add>, ImmRegRel, PredNewRel;
 
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 10,
+CextOpcode = "OR", InputType = "imm" in
 def OR_ri : ALU32_ri<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, s10Imm:$src2),
+            (ins IntRegs:$src1, s10Ext:$src2),
             "$dst = or($src1, #$src2)",
             [(set (i32 IntRegs:$dst), (or (i32 IntRegs:$src1),
-                                          s10ImmPred:$src2))]>;
+                                          s10ExtPred:$src2))]>, ImmRegRel;
 
 def NOT_rr : ALU32_rr<(outs IntRegs:$dst),
             (ins IntRegs:$src1),
             "$dst = not($src1)",
             [(set (i32 IntRegs:$dst), (not (i32 IntRegs:$src1)))]>;
 
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 10,
+InputType = "imm", CextOpcode = "AND" in
 def AND_ri : ALU32_ri<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, s10Imm:$src2),
+            (ins IntRegs:$src1, s10Ext:$src2),
             "$dst = and($src1, #$src2)",
             [(set (i32 IntRegs:$dst), (and (i32 IntRegs:$src1),
-                                           s10ImmPred:$src2))]>;
-
+                                           s10ExtPred:$src2))]>, ImmRegRel;
 // Negate.
 def NEG : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
           "$dst = neg($src1)",
@@ -274,27 +212,138 @@ def NOP : ALU32_rr<(outs), (ins),
           []>;
 
 // Rd32=sub(#s10,Rs32)
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 10,
+CextOpcode = "SUB", InputType = "imm" in
 def SUB_ri : ALU32_ri<(outs IntRegs:$dst),
-            (ins s10Imm:$src1, IntRegs:$src2),
+            (ins s10Ext:$src1, IntRegs:$src2),
             "$dst = sub(#$src1, $src2)",
-            [(set IntRegs:$dst, (sub s10ImmPred:$src1, IntRegs:$src2))]>;
-
-// Transfer immediate.
-let isMoveImm = 1, isReMaterializable = 1, isPredicable = 1 in
-def TFRI : ALU32_ri<(outs IntRegs:$dst), (ins s16Imm:$src1),
-           "$dst = #$src1",
-           [(set (i32 IntRegs:$dst), s16ImmPred:$src1)]>;
-
-// Transfer register.
-let neverHasSideEffects = 1, isPredicable = 1 in
-def TFR : ALU32_ri<(outs IntRegs:$dst), (ins IntRegs:$src1),
-          "$dst = $src1",
-          []>;
+            [(set IntRegs:$dst, (sub s10ExtPred:$src1, IntRegs:$src2))]>,
+            ImmRegRel;
 
-let neverHasSideEffects = 1, isPredicable = 1 in
-def TFR64 : ALU32_ri<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1),
-          "$dst = $src1",
-          []>;
+
+multiclass TFR_Pred<bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    def _c#NAME : ALU32_rr<(outs IntRegs:$dst),
+                           (ins PredRegs:$src1, IntRegs:$src2),
+            !if(PredNot, "if (!$src1", "if ($src1")#") $dst = $src2",
+            []>;
+    // Predicate new
+    let PNewValue = "new" in
+    def _cdn#NAME : ALU32_rr<(outs IntRegs:$dst),
+                             (ins PredRegs:$src1, IntRegs:$src2),
+            !if(PredNot, "if (!$src1", "if ($src1")#".new) $dst = $src2",
+            []>;
+  }
+}
+
+let InputType = "reg", neverHasSideEffects = 1 in
+multiclass TFR_base<string CextOp> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp in {
+    let isPredicable = 1 in
+    def NAME : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
+            "$dst = $src1",
+            []>;
+
+    let  isPredicated = 1 in {
+      defm Pt : TFR_Pred<0>;
+      defm NotPt : TFR_Pred<1>;
+    }
+  }
+}
+
+class T_TFR64_Pred<bit PredNot, bit isPredNew>
+            : ALU32_rr<(outs DoubleRegs:$dst),
+                       (ins PredRegs:$src1, DoubleRegs:$src2),
+            !if(PredNot, "if (!$src1", "if ($src1")#
+            !if(isPredNew, ".new) ", ") ")#"$dst = $src2", []>
+{
+    bits<5> dst;
+    bits<2> src1;
+    bits<5> src2;
+
+    let IClass = 0b1111;
+    let Inst{27-24} = 0b1101;
+    let Inst{13} = isPredNew;
+    let Inst{7} = PredNot;
+    let Inst{4-0} = dst;
+    let Inst{6-5} = src1;
+    let Inst{20-17} = src2{4-1};
+    let Inst{16} = 0b1;
+    let Inst{12-9} = src2{4-1};
+    let Inst{8} = 0b0;
+}
+
+multiclass TFR64_Pred<bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    def _c#NAME : T_TFR64_Pred<PredNot, 0>;
+
+    let PNewValue = "new" in
+    def _cdn#NAME : T_TFR64_Pred<PredNot, 1>; // Predicate new
+  }
+}
+
+let neverHasSideEffects = 1 in
+multiclass TFR64_base<string BaseName> {
+  let BaseOpcode = BaseName in {
+    let isPredicable = 1 in
+    def NAME : ALU32Inst <(outs DoubleRegs:$dst),
+                          (ins DoubleRegs:$src1),
+                          "$dst = $src1" > {
+        bits<5> dst;
+        bits<5> src1;
+
+        let IClass = 0b1111;
+        let Inst{27-23} = 0b01010;
+        let Inst{4-0} = dst;
+        let Inst{20-17} = src1{4-1};
+        let Inst{16} = 0b1;
+        let Inst{12-9} = src1{4-1};
+        let Inst{8} = 0b0;
+    }
+
+    let  isPredicated = 1 in {
+      defm Pt : TFR64_Pred<0>;
+      defm NotPt : TFR64_Pred<1>;
+    }
+  }
+}
+
+multiclass TFRI_Pred<bit PredNot> {
+  let isMoveImm = 1, PredSense = !if(PredNot, "false", "true") in {
+    def _c#NAME : ALU32_ri<(outs IntRegs:$dst),
+                           (ins PredRegs:$src1, s12Ext:$src2),
+            !if(PredNot, "if (!$src1", "if ($src1")#") $dst = #$src2",
+            []>;
+
+    // Predicate new
+    let PNewValue = "new" in
+    def _cdn#NAME : ALU32_rr<(outs IntRegs:$dst),
+                             (ins PredRegs:$src1, s12Ext:$src2),
+            !if(PredNot, "if (!$src1", "if ($src1")#".new) $dst = #$src2",
+            []>;
+  }
+}
+
+let InputType = "imm", isExtendable = 1, isExtentSigned = 1 in
+multiclass TFRI_base<string CextOp> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp#I in {
+    let isAsCheapAsAMove = 1 , opExtendable = 1, opExtentBits = 16,
+    isMoveImm = 1, isPredicable = 1, isReMaterializable = 1 in
+    def NAME : ALU32_ri<(outs IntRegs:$dst), (ins s16Ext:$src1),
+            "$dst = #$src1",
+            [(set (i32 IntRegs:$dst), s16ExtPred:$src1)]>;
+
+    let opExtendable = 2,  opExtentBits = 12, neverHasSideEffects = 1,
+    isPredicated = 1 in {
+      defm Pt    : TFRI_Pred<0>;
+      defm NotPt : TFRI_Pred<1>;
+    }
+  }
+}
+
+defm TFRI : TFRI_base<"TFR">, ImmRegRel, PredNewRel;
+defm TFR : TFR_base<"TFR">, ImmRegRel, PredNewRel;
+defm TFR64 : TFR64_base<"TFR64">, PredNewRel;
 
 // Transfer control register.
 let neverHasSideEffects = 1 in
@@ -311,17 +360,50 @@ def TFCR : CRInst<(outs CRRegs:$dst), (ins IntRegs:$src1),
 //===----------------------------------------------------------------------===//
 
 // Combine.
-let isPredicable = 1, neverHasSideEffects = 1 in
-def COMBINE_rr : ALU32_rr<(outs DoubleRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2),
-            "$dst = combine($src1, $src2)",
-            []>;
 
-let neverHasSideEffects = 1 in
-def COMBINE_ii : ALU32_ii<(outs DoubleRegs:$dst),
-            (ins s8Imm:$src1, s8Imm:$src2),
-            "$dst = combine(#$src1, #$src2)",
-            []>;
+def SDTHexagonI64I32I32 : SDTypeProfile<1, 2,
+  [SDTCisVT<0, i64>, SDTCisVT<1, i32>, SDTCisSameAs<1, 2>]>;
+
+def HexagonWrapperCombineII :
+  SDNode<"HexagonISD::WrapperCombineII", SDTHexagonI64I32I32>;
+def HexagonWrapperCombineRR :
+  SDNode<"HexagonISD::WrapperCombineRR", SDTHexagonI64I32I32>;
+
+// Combines the two integer registers SRC1 and SRC2 into a double register.
+let isPredicable = 1 in
+def COMBINE_rr : ALU32_rr<(outs DoubleRegs:$dst), (ins IntRegs:$src1,
+                                                       IntRegs:$src2),
+  "$dst = combine($src1, $src2)",
+  [(set (i64 DoubleRegs:$dst),
+        (i64 (HexagonWrapperCombineRR (i32 IntRegs:$src1),
+                                      (i32 IntRegs:$src2))))]>;
+
+// Rd=combine(Rt.[HL], Rs.[HL])
+class COMBINE_halves<string A, string B>: ALU32_rr<(outs IntRegs:$dst),
+                                                   (ins IntRegs:$src1,
+                                                        IntRegs:$src2),
+  "$dst = combine($src1."# A #", $src2."# B #")", []>;
+
+let isPredicable = 1 in {
+  def COMBINE_hh : COMBINE_halves<"H", "H">;
+  def COMBINE_hl : COMBINE_halves<"H", "L">;
+  def COMBINE_lh : COMBINE_halves<"L", "H">;
+  def COMBINE_ll : COMBINE_halves<"L", "L">;
+}
+
+def : Pat<(i32 (trunc (i64 (srl (i64 DoubleRegs:$a), (i32 16))))),
+  (COMBINE_lh (EXTRACT_SUBREG (i64 DoubleRegs:$a), subreg_hireg),
+              (EXTRACT_SUBREG (i64 DoubleRegs:$a), subreg_loreg))>;
+
+// Combines the two immediates SRC1 and SRC2 into a double register.
+class COMBINE_imm<Operand imm1, Operand imm2, PatLeaf pat1, PatLeaf pat2> :
+  ALU32_ii<(outs DoubleRegs:$dst), (ins imm1:$src1, imm2:$src2),
+  "$dst = combine(#$src1, #$src2)",
+  [(set (i64 DoubleRegs:$dst),
+        (i64 (HexagonWrapperCombineII (i32 pat1:$src1), (i32 pat2:$src2))))]>;
+
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 8 in
+def COMBINE_Ii : COMBINE_imm<s8Ext, s8Imm, s8ExtPred, s8ImmPred>;
 
 // Mux.
 def VMUX_prr64 : ALU64_rr<(outs DoubleRegs:$dst), (ins PredRegs:$src1,
@@ -330,66 +412,92 @@ def VMUX_prr64 : ALU64_rr<(outs DoubleRegs:$dst), (ins PredRegs:$src1,
             "$dst = vmux($src1, $src2, $src3)",
             []>;
 
+let CextOpcode = "MUX", InputType = "reg" in
 def MUX_rr : ALU32_rr<(outs IntRegs:$dst), (ins PredRegs:$src1,
                                             IntRegs:$src2, IntRegs:$src3),
              "$dst = mux($src1, $src2, $src3)",
-             [(set (i32 IntRegs:$dst), (i32 (select (i1 PredRegs:$src1),
-                                                    (i32 IntRegs:$src2),
-                                                    (i32 IntRegs:$src3))))]>;
+             [(set (i32 IntRegs:$dst),
+                   (i32 (select (i1 PredRegs:$src1), (i32 IntRegs:$src2),
+                                (i32 IntRegs:$src3))))]>, ImmRegRel;
 
-def MUX_ir : ALU32_ir<(outs IntRegs:$dst), (ins PredRegs:$src1, s8Imm:$src2,
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8,
+CextOpcode = "MUX", InputType = "imm" in
+def MUX_ir : ALU32_ir<(outs IntRegs:$dst), (ins PredRegs:$src1, s8Ext:$src2,
                                                 IntRegs:$src3),
              "$dst = mux($src1, #$src2, $src3)",
-             [(set (i32 IntRegs:$dst), (i32 (select (i1 PredRegs:$src1),
-                                                    s8ImmPred:$src2,
-                                                    (i32 IntRegs:$src3))))]>;
+             [(set (i32 IntRegs:$dst),
+                   (i32 (select (i1 PredRegs:$src1), s8ExtPred:$src2,
+                                (i32 IntRegs:$src3))))]>, ImmRegRel;
 
+let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 8,
+CextOpcode = "MUX", InputType = "imm" in
 def MUX_ri : ALU32_ri<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2,
-                                                s8Imm:$src3),
+                                                s8Ext:$src3),
              "$dst = mux($src1, $src2, #$src3)",
-             [(set (i32 IntRegs:$dst), (i32 (select (i1 PredRegs:$src1),
-                                                    (i32 IntRegs:$src2),
-                                                    s8ImmPred:$src3)))]>;
+             [(set (i32 IntRegs:$dst),
+                   (i32 (select (i1 PredRegs:$src1), (i32 IntRegs:$src2),
+                                 s8ExtPred:$src3)))]>, ImmRegRel;
 
-def MUX_ii : ALU32_ii<(outs IntRegs:$dst), (ins PredRegs:$src1, s8Imm:$src2,
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8 in
+def MUX_ii : ALU32_ii<(outs IntRegs:$dst), (ins PredRegs:$src1, s8Ext:$src2,
                                                 s8Imm:$src3),
              "$dst = mux($src1, #$src2, #$src3)",
              [(set (i32 IntRegs:$dst), (i32 (select (i1 PredRegs:$src1),
-                                                    s8ImmPred:$src2,
+                                                    s8ExtPred:$src2,
                                                     s8ImmPred:$src3)))]>;
 
-// Shift halfword.
-let isPredicable = 1 in
-def ASLH : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
-           "$dst = aslh($src1)",
-           [(set (i32 IntRegs:$dst), (shl 16, (i32 IntRegs:$src1)))]>;
+// ALU32 - aslh, asrh, sxtb, sxth, zxtb, zxth
+multiclass ALU32_2op_Pbase<string mnemonic, bit isNot, bit isPredNew> {
+  let isPredicatedNew = isPredNew in
+  def NAME : ALU32Inst<(outs IntRegs:$dst),
+                       (ins PredRegs:$src1, IntRegs:$src2),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew,".new) $dst = ",
+            ") $dst = ")#mnemonic#"($src2)">,
+            Requires<[HasV4T]>;
+}
 
-let isPredicable = 1 in
-def ASRH : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
-           "$dst = asrh($src1)",
-           [(set (i32 IntRegs:$dst), (sra 16, (i32 IntRegs:$src1)))]>;
+multiclass ALU32_2op_Pred<string mnemonic, bit PredNot> {
+  let isPredicatedFalse = PredNot in {
+    defm _c#NAME : ALU32_2op_Pbase<mnemonic, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : ALU32_2op_Pbase<mnemonic, PredNot, 1>;
+  }
+}
 
-// Sign extend.
-let isPredicable = 1 in
-def SXTB : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
-           "$dst = sxtb($src1)",
-           [(set (i32 IntRegs:$dst), (sext_inreg (i32 IntRegs:$src1), i8))]>;
+multiclass ALU32_2op_base<string mnemonic> {
+  let BaseOpcode = mnemonic in {
+    let isPredicable = 1, neverHasSideEffects = 1 in
+    def NAME : ALU32Inst<(outs IntRegs:$dst),
+                         (ins IntRegs:$src1),
+            "$dst = "#mnemonic#"($src1)">;
+
+    let Predicates = [HasV4T], validSubTargets = HasV4SubT, isPredicated = 1,
+    neverHasSideEffects = 1 in {
+      defm Pt_V4    : ALU32_2op_Pred<mnemonic, 0>;
+      defm NotPt_V4 : ALU32_2op_Pred<mnemonic, 1>;
+    }
+  }
+}
 
-let isPredicable = 1 in
-def SXTH : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
-           "$dst = sxth($src1)",
-           [(set (i32 IntRegs:$dst), (sext_inreg (i32 IntRegs:$src1), i16))]>;
-
-// Zero extend.
-let isPredicable = 1, neverHasSideEffects = 1 in
-def ZXTB : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
-           "$dst = zxtb($src1)",
-           []>;
+defm ASLH : ALU32_2op_base<"aslh">, PredNewRel;
+defm ASRH : ALU32_2op_base<"asrh">, PredNewRel;
+defm SXTB : ALU32_2op_base<"sxtb">, PredNewRel;
+defm SXTH : ALU32_2op_base<"sxth">,  PredNewRel;
+defm ZXTB : ALU32_2op_base<"zxtb">, PredNewRel;
+defm ZXTH : ALU32_2op_base<"zxth">,  PredNewRel;
+
+def : Pat <(shl (i32 IntRegs:$src1), (i32 16)),
+           (ASLH IntRegs:$src1)>;
+
+def : Pat <(sra (i32 IntRegs:$src1), (i32 16)),
+           (ASRH IntRegs:$src1)>;
+
+def : Pat <(sext_inreg (i32 IntRegs:$src1), i8),
+           (SXTB IntRegs:$src1)>;
+
+def : Pat <(sext_inreg (i32 IntRegs:$src1), i16),
+           (SXTH IntRegs:$src1)>;
 
-let isPredicable = 1, neverHasSideEffects = 1 in
-def ZXTH : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
-                    "$dst = zxth($src1)",
-                    []>;
 //===----------------------------------------------------------------------===//
 // ALU32/PERM -
 //===----------------------------------------------------------------------===//
@@ -400,98 +508,66 @@ def ZXTH : ALU32_rr<(outs IntRegs:$dst), (ins IntRegs:$src1),
 //===----------------------------------------------------------------------===//
 
 // Conditional combine.
-
-let neverHasSideEffects = 1, isPredicated = 1 in
+let neverHasSideEffects = 1, isPredicated = 1 in {
 def COMBINE_rr_cPt : ALU32_rr<(outs DoubleRegs:$dst),
             (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
             "if ($src1) $dst = combine($src2, $src3)",
             []>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
+let isPredicatedFalse = 1 in
 def COMBINE_rr_cNotPt : ALU32_rr<(outs DoubleRegs:$dst),
             (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
             "if (!$src1) $dst = combine($src2, $src3)",
             []>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
+let isPredicatedNew = 1 in
 def COMBINE_rr_cdnPt : ALU32_rr<(outs DoubleRegs:$dst),
             (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
             "if ($src1.new) $dst = combine($src2, $src3)",
             []>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
+let isPredicatedNew = 1, isPredicatedFalse = 1 in
 def COMBINE_rr_cdnNotPt : ALU32_rr<(outs DoubleRegs:$dst),
             (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
             "if (!$src1.new) $dst = combine($src2, $src3)",
             []>;
+}
 
-// Conditional transfer.
-let neverHasSideEffects = 1, isPredicated = 1 in
-def TFR_cPt : ALU32_rr<(outs IntRegs:$dst), (ins PredRegs:$src1, IntRegs:$src2),
-              "if ($src1) $dst = $src2",
-              []>;
+// Compare.
+defm CMPGTU : CMP32_rr_ri_u9<"cmp.gtu", "CMPGTU", setugt>, ImmRegRel;
+defm CMPGT : CMP32_rr_ri_s10<"cmp.gt", "CMPGT", setgt>, ImmRegRel;
+defm CMPLT : CMP32_rr<"cmp.lt", setlt>;
+defm CMPLTU : CMP32_rr<"cmp.ltu", setult>;
+defm CMPEQ : CMP32_rr_ri_s10<"cmp.eq", "CMPEQ", seteq>, ImmRegRel;
+defm CMPGE : CMP32_ri_s8<"cmp.ge", setge>;
+defm CMPGEU : CMP32_ri_u8<"cmp.geu", setuge>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def TFR_cNotPt : ALU32_rr<(outs IntRegs:$dst), (ins PredRegs:$src1,
-                                                    IntRegs:$src2),
-                 "if (!$src1) $dst = $src2",
-                 []>;
+def CTLZ_rr : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1),
+    "$dst = cl0($src1)",
+    [(set (i32 IntRegs:$dst), (ctlz (i32 IntRegs:$src1)))]>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def TFR64_cPt : ALU32_rr<(outs DoubleRegs:$dst), (ins PredRegs:$src1,
-                                                   DoubleRegs:$src2),
-              "if ($src1) $dst = $src2",
-              []>;
+def CTTZ_rr : SInst<(outs IntRegs:$dst), (ins IntRegs:$src1),
+    "$dst = ct0($src1)",
+    [(set (i32 IntRegs:$dst), (cttz (i32 IntRegs:$src1)))]>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def TFR64_cNotPt : ALU32_rr<(outs DoubleRegs:$dst), (ins PredRegs:$src1,
-                                                    DoubleRegs:$src2),
-                 "if (!$src1) $dst = $src2",
-                 []>;
+def CTLZ64_rr : SInst<(outs IntRegs:$dst), (ins DoubleRegs:$src1),
+    "$dst = cl0($src1)",
+    [(set (i32 IntRegs:$dst), (i32 (trunc (ctlz (i64 DoubleRegs:$src1)))))]>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def TFRI_cPt : ALU32_ri<(outs IntRegs:$dst), (ins PredRegs:$src1, s12Imm:$src2),
-               "if ($src1) $dst = #$src2",
-               []>;
+def CTTZ64_rr : SInst<(outs IntRegs:$dst), (ins DoubleRegs:$src1),
+    "$dst = ct0($src1)",
+    [(set (i32 IntRegs:$dst), (i32 (trunc (cttz (i64 DoubleRegs:$src1)))))]>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def TFRI_cNotPt : ALU32_ri<(outs IntRegs:$dst), (ins PredRegs:$src1,
-                                                     s12Imm:$src2),
-                  "if (!$src1) $dst = #$src2",
-                  []>;
+def TSTBIT_rr : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
+    "$dst = tstbit($src1, $src2)",
+    [(set (i1 PredRegs:$dst),
+          (setne (and (shl 1, (i32 IntRegs:$src2)), (i32 IntRegs:$src1)), 0))]>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def TFR_cdnPt : ALU32_rr<(outs IntRegs:$dst), (ins PredRegs:$src1,
-                                                   IntRegs:$src2),
-                "if ($src1.new) $dst = $src2",
-                []>;
+def TSTBIT_ri : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
+    "$dst = tstbit($src1, $src2)",
+    [(set (i1 PredRegs:$dst),
+          (setne (and (shl 1, (u5ImmPred:$src2)), (i32 IntRegs:$src1)), 0))]>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def TFR_cdnNotPt : ALU32_rr<(outs IntRegs:$dst), (ins PredRegs:$src1,
-                                                      IntRegs:$src2),
-                   "if (!$src1.new) $dst = $src2",
-                   []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def TFRI_cdnPt : ALU32_ri<(outs IntRegs:$dst), (ins PredRegs:$src1,
-                                                    s12Imm:$src2),
-                 "if ($src1.new) $dst = #$src2",
-                 []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def TFRI_cdnNotPt : ALU32_ri<(outs IntRegs:$dst), (ins PredRegs:$src1,
-                                                       s12Imm:$src2),
-                    "if (!$src1.new) $dst = #$src2",
-                    []>;
-
-// Compare.
-defm CMPGTU : CMP32_rr_ri_u9<"cmp.gtu", setugt>;
-defm CMPGT : CMP32_rr_ri_s10<"cmp.gt", setgt>;
-defm CMPLT : CMP32_rr<"cmp.lt", setlt>;
-defm CMPLTU : CMP32_rr<"cmp.ltu", setult>;
-defm CMPEQ : CMP32_rr_ri_s10<"cmp.eq", seteq>;
-defm CMPGE : CMP32_ri_s8<"cmp.ge", setge>;
-defm CMPGEU : CMP32_ri_u8<"cmp.geu", setuge>;
 //===----------------------------------------------------------------------===//
 // ALU32/PRED -
 //===----------------------------------------------------------------------===//
@@ -608,11 +684,6 @@ def SUB64_rr : ALU64_rr<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1,
 
 // Subtract halfword.
 
-// Transfer register.
-let neverHasSideEffects = 1 in
-def TFR_64 : ALU64_rr<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1),
-             "$dst = $src1",
-             []>;
 //===----------------------------------------------------------------------===//
 // ALU64/ALU -
 //===----------------------------------------------------------------------===//
@@ -784,7 +855,7 @@ let isBranch = 1, isTerminator=1, neverHasSideEffects = 1, Defs = [PC],
 // JR +
 //===----------------------------------------------------------------------===//
 def retflag : SDNode<"HexagonISD::RET_FLAG", SDTNone,
-                               [SDNPHasChain, SDNPOptInGlue]>;
+                               [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
 
 // Jump to address from register.
 let isPredicable =1, isReturn = 1, isTerminator = 1, isBarrier = 1,
@@ -818,241 +889,218 @@ let isReturn = 1, isTerminator = 1, isBarrier = 1, isPredicated = 1,
 // LD +
 //===----------------------------------------------------------------------===//
 ///
-/// Make sure that in post increment load, the first operand is always the post
-/// increment operand.
-///
-// Load doubleword.
-let isPredicable = 1 in
-def LDrid : LDInst<(outs DoubleRegs:$dst),
-            (ins MEMri:$addr),
-            "$dst = memd($addr)",
-            [(set (i64 DoubleRegs:$dst), (i64 (load ADDRriS11_3:$addr)))]>;
-
-let isPredicable = 1, AddedComplexity = 20 in
-def LDrid_indexed : LDInst<(outs DoubleRegs:$dst),
-            (ins IntRegs:$src1, s11_3Imm:$offset),
-            "$dst = memd($src1+#$offset)",
-            [(set (i64 DoubleRegs:$dst),
-                  (i64 (load (add (i32 IntRegs:$src1),
-                                  s11_3ImmPred:$offset))))]>;
+// Load -- MEMri operand
+multiclass LD_MEMri_Pbase<string mnemonic, RegisterClass RC,
+                          bit isNot, bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : LDInst2<(outs RC:$dst),
+                       (ins PredRegs:$src1, MEMri:$addr),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#"$dst = "#mnemonic#"($addr)",
+            []>;
+}
 
-let neverHasSideEffects = 1 in
-def LDrid_GP : LDInst2<(outs DoubleRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst = memd(#$global+$offset)",
-            []>,
-            Requires<[NoV4T]>;
+multiclass LD_MEMri_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : LD_MEMri_Pbase<mnemonic, RC, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : LD_MEMri_Pbase<mnemonic, RC, PredNot, 1>;
+  }
+}
 
-let neverHasSideEffects = 1 in
-def LDd_GP : LDInst2<(outs DoubleRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst = memd(#$global)",
-            []>,
-            Requires<[NoV4T]>;
-
-let isPredicable = 1, hasCtrlDep = 1, neverHasSideEffects = 1 in
-def POST_LDrid : LDInst2PI<(outs DoubleRegs:$dst, IntRegs:$dst2),
-            (ins IntRegs:$src1, s4Imm:$offset),
-            "$dst = memd($src1++#$offset)",
-            [],
-            "$src1 = $dst2">;
+let isExtendable = 1, neverHasSideEffects = 1 in
+multiclass LD_MEMri<string mnemonic, string CextOp, RegisterClass RC,
+                    bits<5> ImmBits, bits<5> PredImmBits> {
 
-// Load doubleword conditionally.
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_cPt : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1) $dst = memd($addr)",
-            []>;
+  let CextOpcode = CextOp, BaseOpcode = CextOp in {
+    let opExtendable = 2, isExtentSigned = 1, opExtentBits = ImmBits,
+        isPredicable = 1 in
+      def NAME : LDInst2<(outs RC:$dst), (ins MEMri:$addr),
+                   "$dst = "#mnemonic#"($addr)",
+                   []>;
 
+    let opExtendable = 3, isExtentSigned = 0, opExtentBits = PredImmBits,
+        isPredicated = 1 in {
+      defm Pt : LD_MEMri_Pred<mnemonic, RC, 0 >;
+      defm NotPt : LD_MEMri_Pred<mnemonic, RC, 1 >;
+    }
+  }
+}
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_cNotPt : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1) $dst = memd($addr)",
-            []>;
+let addrMode = BaseImmOffset, isMEMri = "true" in {
+  defm LDrib: LD_MEMri < "memb", "LDrib", IntRegs, 11, 6>, AddrModeRel;
+  defm LDriub: LD_MEMri < "memub" , "LDriub", IntRegs, 11, 6>, AddrModeRel;
+  defm LDrih: LD_MEMri < "memh", "LDrih", IntRegs, 12, 7>, AddrModeRel;
+  defm LDriuh: LD_MEMri < "memuh", "LDriuh", IntRegs, 12, 7>, AddrModeRel;
+  defm LDriw: LD_MEMri < "memw", "LDriw", IntRegs, 13, 8>, AddrModeRel;
+  defm LDrid: LD_MEMri < "memd", "LDrid", DoubleRegs, 14, 9>, AddrModeRel;
+}
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_indexed_cPt : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_3Imm:$src3),
-            "if ($src1) $dst = memd($src2+#$src3)",
-            []>;
+def : Pat < (i32 (sextloadi8 ADDRriS11_0:$addr)),
+            (LDrib ADDRriS11_0:$addr) >;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_indexed_cNotPt : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_3Imm:$src3),
-            "if (!$src1) $dst = memd($src2+#$src3)",
-            []>;
+def : Pat < (i32 (zextloadi8 ADDRriS11_0:$addr)),
+            (LDriub ADDRriS11_0:$addr) >;
 
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrid_cPt : LDInst2PI<(outs DoubleRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_3Imm:$src3),
-            "if ($src1) $dst1 = memd($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
+def : Pat < (i32 (sextloadi16 ADDRriS11_1:$addr)),
+            (LDrih ADDRriS11_1:$addr) >;
 
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrid_cNotPt : LDInst2PI<(outs DoubleRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_3Imm:$src3),
-            "if (!$src1) $dst1 = memd($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
+def : Pat < (i32 (zextloadi16 ADDRriS11_1:$addr)),
+            (LDriuh ADDRriS11_1:$addr) >;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_cdnPt : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1.new) $dst = memd($addr)",
-            []>;
+def : Pat < (i32 (load ADDRriS11_2:$addr)),
+            (LDriw ADDRriS11_2:$addr) >;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_cdnNotPt : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1.new) $dst = memd($addr)",
-            []>;
+def : Pat < (i64 (load ADDRriS11_3:$addr)),
+            (LDrid ADDRriS11_3:$addr) >;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_indexed_cdnPt : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_3Imm:$src3),
-            "if ($src1.new) $dst = memd($src2+#$src3)",
-            []>;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_indexed_cdnNotPt : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_3Imm:$src3),
-            "if (!$src1.new) $dst = memd($src2+#$src3)",
+// Load - Base with Immediate offset addressing mode
+multiclass LD_Idxd_Pbase<string mnemonic, RegisterClass RC, Operand predImmOp,
+                        bit isNot, bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : LDInst2<(outs RC:$dst),
+                     (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#"$dst = "#mnemonic#"($src2+#$src3)",
             []>;
+}
 
+multiclass LD_Idxd_Pred<string mnemonic, RegisterClass RC, Operand predImmOp,
+                        bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : LD_Idxd_Pbase<mnemonic, RC, predImmOp, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : LD_Idxd_Pbase<mnemonic, RC, predImmOp, PredNot, 1>;
+  }
+}
 
-// Load byte.
-let isPredicable = 1 in
-def LDrib : LDInst<(outs IntRegs:$dst),
-            (ins MEMri:$addr),
-            "$dst = memb($addr)",
-            [(set (i32 IntRegs:$dst), (i32 (sextloadi8 ADDRriS11_0:$addr)))]>;
-
-// Load byte any-extend.
-def : Pat < (i32 (extloadi8 ADDRriS11_0:$addr)),
-            (i32 (LDrib ADDRriS11_0:$addr)) >;
+let isExtendable = 1, neverHasSideEffects = 1 in
+multiclass LD_Idxd<string mnemonic, string CextOp, RegisterClass RC,
+                   Operand ImmOp, Operand predImmOp, bits<5> ImmBits,
+                   bits<5> PredImmBits> {
 
-// Indexed load byte.
-let isPredicable = 1, AddedComplexity = 20 in
-def LDrib_indexed : LDInst<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, s11_0Imm:$offset),
-            "$dst = memb($src1+#$offset)",
-            [(set (i32 IntRegs:$dst),
-                  (i32 (sextloadi8 (add (i32 IntRegs:$src1),
-                                        s11_0ImmPred:$offset))))]>;
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
+    let opExtendable = 2, isExtentSigned = 1, opExtentBits = ImmBits,
+        isPredicable = 1, AddedComplexity = 20 in
+      def NAME : LDInst2<(outs RC:$dst), (ins IntRegs:$src1, ImmOp:$offset),
+                   "$dst = "#mnemonic#"($src1+#$offset)",
+                   []>;
 
-// Indexed load byte any-extend.
-let AddedComplexity = 20 in
-def : Pat < (i32 (extloadi8 (add IntRegs:$src1, s11_0ImmPred:$offset))),
-            (i32 (LDrib_indexed IntRegs:$src1, s11_0ImmPred:$offset)) >;
+    let opExtendable = 3, isExtentSigned = 0, opExtentBits = PredImmBits,
+        isPredicated = 1 in {
+      defm Pt : LD_Idxd_Pred<mnemonic, RC, predImmOp, 0 >;
+      defm NotPt : LD_Idxd_Pred<mnemonic, RC, predImmOp, 1 >;
+    }
+  }
+}
 
-let neverHasSideEffects = 1 in
-def LDrib_GP : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst = memb(#$global+$offset)",
-            []>,
-            Requires<[NoV4T]>;
+let addrMode = BaseImmOffset in {
+  defm LDrib_indexed: LD_Idxd <"memb", "LDrib", IntRegs, s11_0Ext, u6_0Ext,
+                               11, 6>, AddrModeRel;
+  defm LDriub_indexed: LD_Idxd <"memub" , "LDriub", IntRegs, s11_0Ext, u6_0Ext,
+                                11, 6>, AddrModeRel;
+  defm LDrih_indexed: LD_Idxd <"memh", "LDrih", IntRegs, s11_1Ext, u6_1Ext,
+                               12, 7>, AddrModeRel;
+  defm LDriuh_indexed: LD_Idxd <"memuh", "LDriuh", IntRegs, s11_1Ext, u6_1Ext,
+                                12, 7>, AddrModeRel;
+  defm LDriw_indexed: LD_Idxd <"memw", "LDriw", IntRegs, s11_2Ext, u6_2Ext,
+                               13, 8>, AddrModeRel;
+  defm LDrid_indexed: LD_Idxd <"memd", "LDrid", DoubleRegs, s11_3Ext, u6_3Ext,
+                               14, 9>, AddrModeRel;
+}
 
-let neverHasSideEffects = 1 in
-def LDb_GP : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst = memb(#$global)",
-            []>,
-            Requires<[NoV4T]>;
+let AddedComplexity = 20 in {
+def : Pat < (i32 (sextloadi8 (add IntRegs:$src1, s11_0ExtPred:$offset))),
+            (LDrib_indexed IntRegs:$src1, s11_0ExtPred:$offset) >;
 
-let neverHasSideEffects = 1 in
-def LDub_GP : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst = memub(#$global)",
-            []>,
-            Requires<[NoV4T]>;
-
-let isPredicable = 1, hasCtrlDep = 1, neverHasSideEffects = 1 in
-def POST_LDrib : LDInst2PI<(outs IntRegs:$dst, IntRegs:$dst2),
-            (ins IntRegs:$src1, s4Imm:$offset),
-            "$dst = memb($src1++#$offset)",
-            [],
-            "$src1 = $dst2">;
+def : Pat < (i32 (zextloadi8 (add IntRegs:$src1, s11_0ExtPred:$offset))),
+            (LDriub_indexed IntRegs:$src1, s11_0ExtPred:$offset) >;
 
-// Load byte conditionally.
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_cPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1) $dst = memb($addr)",
-            []>;
+def : Pat < (i32 (sextloadi16 (add IntRegs:$src1, s11_1ExtPred:$offset))),
+            (LDrih_indexed IntRegs:$src1, s11_1ExtPred:$offset) >;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_cNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1) $dst = memb($addr)",
-            []>;
+def : Pat < (i32 (zextloadi16 (add IntRegs:$src1, s11_1ExtPred:$offset))),
+            (LDriuh_indexed IntRegs:$src1, s11_1ExtPred:$offset) >;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_indexed_cPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3),
-            "if ($src1) $dst = memb($src2+#$src3)",
-            []>;
+def : Pat < (i32 (load (add IntRegs:$src1, s11_2ExtPred:$offset))),
+            (LDriw_indexed IntRegs:$src1, s11_2ExtPred:$offset) >;
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_indexed_cNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3),
-            "if (!$src1) $dst = memb($src2+#$src3)",
-            []>;
+def : Pat < (i64 (load (add IntRegs:$src1, s11_3ExtPred:$offset))),
+            (LDrid_indexed IntRegs:$src1, s11_3ExtPred:$offset) >;
+}
 
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrib_cPt : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_0Imm:$src3),
-            "if ($src1) $dst1 = memb($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
+//===----------------------------------------------------------------------===//
+// Post increment load
+// Make sure that in post increment load, the first operand is always the post
+// increment operand.
+//===----------------------------------------------------------------------===//
 
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrib_cNotPt : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_0Imm:$src3),
-            "if (!$src1) $dst1 = memb($src2++#$src3)",
+multiclass LD_PostInc_Pbase<string mnemonic, RegisterClass RC, Operand ImmOp,
+                            bit isNot, bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : LDInst2PI<(outs RC:$dst, IntRegs:$dst2),
+                       (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#"$dst = "#mnemonic#"($src2++#$offset)",
             [],
             "$src2 = $dst2">;
+}
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_cdnPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1.new) $dst = memb($addr)",
-            []>;
+multiclass LD_PostInc_Pred<string mnemonic, RegisterClass RC,
+                           Operand ImmOp, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : LD_PostInc_Pbase<mnemonic, RC, ImmOp, PredNot, 0>;
+    // Predicate new
+    let Predicates = [HasV4T], validSubTargets = HasV4SubT in
+    defm _cdn#NAME#_V4 : LD_PostInc_Pbase<mnemonic, RC, ImmOp, PredNot, 1>;
+  }
+}
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_cdnNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1.new) $dst = memb($addr)",
-            []>;
+multiclass LD_PostInc<string mnemonic, string BaseOp, RegisterClass RC,
+                      Operand ImmOp> {
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_indexed_cdnPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3),
-            "if ($src1.new) $dst = memb($src2+#$src3)",
-            []>;
+  let BaseOpcode = "POST_"#BaseOp in {
+    let isPredicable = 1 in
+    def NAME : LDInst2PI<(outs RC:$dst, IntRegs:$dst2),
+                         (ins IntRegs:$src1, ImmOp:$offset),
+                 "$dst = "#mnemonic#"($src1++#$offset)",
+                 [],
+                 "$src1 = $dst2">;
+
+    let isPredicated = 1 in {
+      defm Pt : LD_PostInc_Pred<mnemonic, RC, ImmOp, 0 >;
+      defm NotPt : LD_PostInc_Pred<mnemonic, RC, ImmOp, 1 >;
+    }
+  }
+}
 
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_indexed_cdnNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3),
-            "if (!$src1.new) $dst = memb($src2+#$src3)",
-            []>;
+let hasCtrlDep = 1, neverHasSideEffects = 1 in {
+  defm POST_LDrib : LD_PostInc<"memb", "LDrib", IntRegs, s4_0Imm>,
+                    PredNewRel;
+  defm POST_LDriub : LD_PostInc<"memub", "LDriub", IntRegs, s4_0Imm>,
+                    PredNewRel;
+  defm POST_LDrih : LD_PostInc<"memh", "LDrih", IntRegs, s4_1Imm>,
+                    PredNewRel;
+  defm POST_LDriuh : LD_PostInc<"memuh", "LDriuh", IntRegs, s4_1Imm>,
+                    PredNewRel;
+  defm POST_LDriw : LD_PostInc<"memw", "LDriw", IntRegs, s4_2Imm>,
+                    PredNewRel;
+  defm POST_LDrid : LD_PostInc<"memd", "LDrid", DoubleRegs, s4_3Imm>,
+                    PredNewRel;
+}
 
+def : Pat< (i32 (extloadi1 ADDRriS11_0:$addr)),
+           (i32 (LDrib ADDRriS11_0:$addr)) >;
 
-// Load halfword.
-let isPredicable = 1 in
-def LDrih : LDInst<(outs IntRegs:$dst),
-            (ins MEMri:$addr),
-            "$dst = memh($addr)",
-            [(set (i32 IntRegs:$dst), (i32 (sextloadi16 ADDRriS11_1:$addr)))]>;
+// Load byte any-extend.
+def : Pat < (i32 (extloadi8 ADDRriS11_0:$addr)),
+            (i32 (LDrib ADDRriS11_0:$addr)) >;
 
-let isPredicable = 1, AddedComplexity = 20 in
-def LDrih_indexed : LDInst<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, s11_1Imm:$offset),
-            "$dst = memh($src1+#$offset)",
-            [(set (i32 IntRegs:$dst),
-                  (i32 (sextloadi16 (add (i32 IntRegs:$src1),
-                                         s11_1ImmPred:$offset))))]>;
+// Indexed load byte any-extend.
+let AddedComplexity = 20 in
+def : Pat < (i32 (extloadi8 (add IntRegs:$src1, s11_0ImmPred:$offset))),
+            (i32 (LDrib_indexed IntRegs:$src1, s11_0ImmPred:$offset)) >;
 
 def : Pat < (i32 (extloadi16 ADDRriS11_1:$addr)),
             (i32 (LDrih ADDRriS11_1:$addr))>;
@@ -1061,399 +1109,25 @@ let AddedComplexity = 20 in
 def : Pat < (i32 (extloadi16 (add IntRegs:$src1, s11_1ImmPred:$offset))),
             (i32 (LDrih_indexed IntRegs:$src1, s11_1ImmPred:$offset)) >;
 
-let neverHasSideEffects = 1 in
-def LDrih_GP : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst = memh(#$global+$offset)",
-            []>,
-            Requires<[NoV4T]>;
-
-let neverHasSideEffects = 1 in
-def LDh_GP : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst = memh(#$global)",
-            []>,
-            Requires<[NoV4T]>;
-
-let neverHasSideEffects = 1 in
-def LDuh_GP : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst = memuh(#$global)",
-            []>,
-            Requires<[NoV4T]>;
-
-let isPredicable = 1, hasCtrlDep = 1, neverHasSideEffects = 1 in
-def POST_LDrih : LDInst2PI<(outs IntRegs:$dst, IntRegs:$dst2),
-            (ins IntRegs:$src1, s4Imm:$offset),
-            "$dst = memh($src1++#$offset)",
-            [],
-            "$src1 = $dst2">;
-
-// Load halfword conditionally.
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_cPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1) $dst = memh($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_cNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1) $dst = memh($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_indexed_cPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3),
-            "if ($src1) $dst = memh($src2+#$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_indexed_cNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3),
-            "if (!$src1) $dst = memh($src2+#$src3)",
-            []>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrih_cPt : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_1Imm:$src3),
-            "if ($src1) $dst1 = memh($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrih_cNotPt : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_1Imm:$src3),
-            "if (!$src1) $dst1 = memh($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_cdnPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1.new) $dst = memh($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_cdnNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1.new) $dst = memh($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_indexed_cdnPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3),
-            "if ($src1.new) $dst = memh($src2+#$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_indexed_cdnNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3),
-            "if (!$src1.new) $dst = memh($src2+#$src3)",
-            []>;
-
-// Load unsigned byte.
-let isPredicable = 1 in
-def LDriub : LDInst<(outs IntRegs:$dst),
-            (ins MEMri:$addr),
-            "$dst = memub($addr)",
-            [(set (i32 IntRegs:$dst), (i32 (zextloadi8 ADDRriS11_0:$addr)))]>;
-
+let AddedComplexity = 10 in
 def : Pat < (i32 (zextloadi1 ADDRriS11_0:$addr)),
             (i32 (LDriub ADDRriS11_0:$addr))>;
 
-let isPredicable = 1, AddedComplexity = 20 in
-def LDriub_indexed : LDInst<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, s11_0Imm:$offset),
-            "$dst = memub($src1+#$offset)",
-            [(set (i32 IntRegs:$dst),
-                  (i32 (zextloadi8 (add (i32 IntRegs:$src1),
-                                        s11_0ImmPred:$offset))))]>;
-
 let AddedComplexity = 20 in
 def : Pat < (i32 (zextloadi1 (add IntRegs:$src1, s11_0ImmPred:$offset))),
             (i32 (LDriub_indexed IntRegs:$src1, s11_0ImmPred:$offset))>;
 
-let neverHasSideEffects = 1 in
-def LDriub_GP : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst = memub(#$global+$offset)",
-            []>,
-            Requires<[NoV4T]>;
-
-let isPredicable = 1, hasCtrlDep = 1, neverHasSideEffects = 1 in
-def POST_LDriub : LDInst2PI<(outs IntRegs:$dst, IntRegs:$dst2),
-            (ins IntRegs:$src1, s4Imm:$offset),
-            "$dst = memub($src1++#$offset)",
-            [],
-            "$src1 = $dst2">;
-
-// Load unsigned byte conditionally.
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_cPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1) $dst = memub($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_cNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1) $dst = memub($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_indexed_cPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3),
-            "if ($src1) $dst = memub($src2+#$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_indexed_cNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3),
-            "if (!$src1) $dst = memub($src2+#$src3)",
-            []>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriub_cPt : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_0Imm:$src3),
-            "if ($src1) $dst1 = memub($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriub_cNotPt : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_0Imm:$src3),
-            "if (!$src1) $dst1 = memub($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_cdnPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1.new) $dst = memub($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_cdnNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1.new) $dst = memub($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_indexed_cdnPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3),
-            "if ($src1.new) $dst = memub($src2+#$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_indexed_cdnNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3),
-            "if (!$src1.new) $dst = memub($src2+#$src3)",
-            []>;
-
-// Load unsigned halfword.
-let isPredicable = 1 in
-def LDriuh : LDInst<(outs IntRegs:$dst),
-            (ins MEMri:$addr),
-            "$dst = memuh($addr)",
-            [(set (i32 IntRegs:$dst), (i32 (zextloadi16 ADDRriS11_1:$addr)))]>;
-
-// Indexed load unsigned halfword.
-let isPredicable = 1, AddedComplexity = 20 in
-def LDriuh_indexed : LDInst<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, s11_1Imm:$offset),
-            "$dst = memuh($src1+#$offset)",
-            [(set (i32 IntRegs:$dst),
-                  (i32 (zextloadi16 (add (i32 IntRegs:$src1),
-                                         s11_1ImmPred:$offset))))]>;
-
-let neverHasSideEffects = 1 in
-def LDriuh_GP : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst = memuh(#$global+$offset)",
-            []>,
-            Requires<[NoV4T]>;
-
-let isPredicable = 1, hasCtrlDep = 1, neverHasSideEffects = 1 in
-def POST_LDriuh : LDInst2PI<(outs IntRegs:$dst, IntRegs:$dst2),
-            (ins IntRegs:$src1, s4Imm:$offset),
-            "$dst = memuh($src1++#$offset)",
-            [],
-            "$src1 = $dst2">;
-
-// Load unsigned halfword conditionally.
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_cPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1) $dst = memuh($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_cNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1) $dst = memuh($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_indexed_cPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3),
-            "if ($src1) $dst = memuh($src2+#$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_indexed_cNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3),
-            "if (!$src1) $dst = memuh($src2+#$src3)",
-            []>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriuh_cPt : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_1Imm:$src3),
-            "if ($src1) $dst1 = memuh($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriuh_cNotPt : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_1Imm:$src3),
-            "if (!$src1) $dst1 = memuh($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_cdnPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1.new) $dst = memuh($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_cdnNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1.new) $dst = memuh($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_indexed_cdnPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3),
-            "if ($src1.new) $dst = memuh($src2+#$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_indexed_cdnNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3),
-            "if (!$src1.new) $dst = memuh($src2+#$src3)",
-            []>;
-
-
-// Load word.
-let isPredicable = 1 in
-def LDriw : LDInst<(outs IntRegs:$dst),
-            (ins MEMri:$addr), "$dst = memw($addr)",
-            [(set IntRegs:$dst, (i32 (load ADDRriS11_2:$addr)))]>;
-
 // Load predicate.
-let Defs = [R10,R11,D5], neverHasSideEffects = 1 in
-def LDriw_pred : LDInst<(outs PredRegs:$dst),
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 13,
+isPseudo = 1, Defs = [R10,R11,D5], neverHasSideEffects = 1 in
+def LDriw_pred : LDInst2<(outs PredRegs:$dst),
             (ins MEMri:$addr),
             "Error; should not emit",
             []>;
 
-// Indexed load.
-let isPredicable = 1, AddedComplexity = 20 in
-def LDriw_indexed : LDInst<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, s11_2Imm:$offset),
-            "$dst = memw($src1+#$offset)",
-            [(set IntRegs:$dst, (i32 (load (add IntRegs:$src1,
-                                           s11_2ImmPred:$offset))))]>;
-
-let neverHasSideEffects = 1 in
-def LDriw_GP : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst = memw(#$global+$offset)",
-            []>,
-            Requires<[NoV4T]>;
-
-let neverHasSideEffects = 1 in
-def LDw_GP : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst = memw(#$global)",
-            []>,
-            Requires<[NoV4T]>;
-
-let isPredicable = 1, hasCtrlDep = 1, neverHasSideEffects = 1 in
-def POST_LDriw : LDInst2PI<(outs IntRegs:$dst, IntRegs:$dst2),
-            (ins IntRegs:$src1, s4Imm:$offset),
-            "$dst = memw($src1++#$offset)",
-            [],
-            "$src1 = $dst2">;
-
-// Load word conditionally.
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_cPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1) $dst = memw($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_cNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1) $dst = memw($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_indexed_cPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3),
-            "if ($src1) $dst = memw($src2+#$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_indexed_cNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3),
-            "if (!$src1) $dst = memw($src2+#$src3)",
-            []>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriw_cPt : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_2Imm:$src3),
-            "if ($src1) $dst1 = memw($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriw_cNotPt : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_2Imm:$src3),
-            "if (!$src1) $dst1 = memw($src2++#$src3)",
-            [],
-            "$src2 = $dst2">;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_cdnPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if ($src1.new) $dst = memw($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_cdnNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, MEMri:$addr),
-            "if (!$src1.new) $dst = memw($addr)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_indexed_cdnPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3),
-            "if ($src1.new) $dst = memw($src2+#$src3)",
-            []>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_indexed_cdnNotPt : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3),
-            "if (!$src1.new) $dst = memw($src2+#$src3)",
-            []>;
-
 // Deallocate stack frame.
 let Defs = [R29, R30, R31], Uses = [R29], neverHasSideEffects = 1 in {
-  def DEALLOCFRAME : LDInst2<(outs), (ins i32imm:$amt1),
+  def DEALLOCFRAME : LDInst2<(outs), (ins),
                      "deallocframe",
                      []>;
 }
@@ -1482,57 +1156,65 @@ let Defs = [R29, R30, R31], Uses = [R29], neverHasSideEffects = 1 in {
 //===----------------------------------------------------------------------===//
 // Multiply and use lower result.
 // Rd=+mpyi(Rs,#u8)
-def MPYI_riu : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, u8Imm:$src2),
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 8 in
+def MPYI_riu : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, u8Ext:$src2),
               "$dst =+ mpyi($src1, #$src2)",
               [(set (i32 IntRegs:$dst), (mul (i32 IntRegs:$src1),
-                                             u8ImmPred:$src2))]>;
+                                             u8ExtPred:$src2))]>;
 
 // Rd=-mpyi(Rs,#u8)
-def MPYI_rin : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, n8Imm:$src2),
+def MPYI_rin : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, u8Imm:$src2),
               "$dst =- mpyi($src1, #$src2)",
-              [(set (i32 IntRegs:$dst), (mul (i32 IntRegs:$src1),
-                                             n8ImmPred:$src2))]>;
+              [(set (i32 IntRegs:$dst), (ineg (mul (i32 IntRegs:$src1),
+                                                   u8ImmPred:$src2)))]>;
 
 // Rd=mpyi(Rs,#m9)
 // s9 is NOT the same as m9 - but it works.. so far.
 // Assembler maps to either Rd=+mpyi(Rs,#u8 or Rd=-mpyi(Rs,#u8)
 // depending on the value of m9. See Arch Spec.
-def MPYI_ri : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, s9Imm:$src2),
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 9,
+CextOpcode = "MPYI", InputType = "imm" in
+def MPYI_ri : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, s9Ext:$src2),
               "$dst = mpyi($src1, #$src2)",
               [(set (i32 IntRegs:$dst), (mul (i32 IntRegs:$src1),
-                                             s9ImmPred:$src2))]>;
+                                             s9ExtPred:$src2))]>, ImmRegRel;
 
 // Rd=mpyi(Rs,Rt)
+let CextOpcode = "MPYI", InputType = "reg" in
 def MPYI : MInst<(outs IntRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
            "$dst = mpyi($src1, $src2)",
            [(set (i32 IntRegs:$dst), (mul (i32 IntRegs:$src1),
-                                          (i32 IntRegs:$src2)))]>;
+                                          (i32 IntRegs:$src2)))]>, ImmRegRel;
 
 // Rx+=mpyi(Rs,#u8)
+let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 8,
+CextOpcode = "MPYI_acc", InputType = "imm" in
 def MPYI_acc_ri : MInst_acc<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2, u8Imm:$src3),
+            (ins IntRegs:$src1, IntRegs:$src2, u8Ext:$src3),
             "$dst += mpyi($src2, #$src3)",
             [(set (i32 IntRegs:$dst),
-                  (add (mul (i32 IntRegs:$src2), u8ImmPred:$src3),
+                  (add (mul (i32 IntRegs:$src2), u8ExtPred:$src3),
                        (i32 IntRegs:$src1)))],
-            "$src1 = $dst">;
+            "$src1 = $dst">, ImmRegRel;
 
 // Rx+=mpyi(Rs,Rt)
+let CextOpcode = "MPYI_acc", InputType = "reg" in
 def MPYI_acc_rr : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
             "$dst += mpyi($src2, $src3)",
             [(set (i32 IntRegs:$dst),
                   (add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
                        (i32 IntRegs:$src1)))],
-            "$src1 = $dst">;
+            "$src1 = $dst">, ImmRegRel;
 
 // Rx-=mpyi(Rs,#u8)
+let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 8 in
 def MPYI_sub_ri : MInst_acc<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2, u8Imm:$src3),
+            (ins IntRegs:$src1, IntRegs:$src2, u8Ext:$src3),
             "$dst -= mpyi($src2, #$src3)",
             [(set (i32 IntRegs:$dst),
                   (sub (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
-                                                 u8ImmPred:$src3)))],
+                                                 u8ExtPred:$src3)))],
             "$src1 = $dst">;
 
 // Multiply and use upper result.
@@ -1601,7 +1283,7 @@ def MPYU64_acc : MInst_acc<(outs DoubleRegs:$dst), (ins DoubleRegs:$src1,
 // Rxx-=mpyu(Rs,Rt)
 def MPYU64_sub : MInst_acc<(outs DoubleRegs:$dst),
             (ins DoubleRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-            "$dst += mpyu($src2, $src3)",
+            "$dst -= mpyu($src2, $src3)",
             [(set (i64 DoubleRegs:$dst),
                   (sub (i64 DoubleRegs:$src1),
                        (mul (i64 (anyext (i32 IntRegs:$src2))),
@@ -1609,37 +1291,43 @@ def MPYU64_sub : MInst_acc<(outs DoubleRegs:$dst),
             "$src1 = $dst">;
 
 
+let InputType = "reg", CextOpcode = "ADD_acc" in
 def ADDrr_acc : MInst_acc<(outs IntRegs: $dst), (ins IntRegs:$src1,
                             IntRegs:$src2, IntRegs:$src3),
              "$dst += add($src2, $src3)",
              [(set (i32 IntRegs:$dst), (add (add (i32 IntRegs:$src2),
                                                  (i32 IntRegs:$src3)),
                                             (i32 IntRegs:$src1)))],
-             "$src1 = $dst">;
+             "$src1 = $dst">, ImmRegRel;
 
+let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 8,
+InputType = "imm", CextOpcode = "ADD_acc" in
 def ADDri_acc : MInst_acc<(outs IntRegs: $dst), (ins IntRegs:$src1,
-                            IntRegs:$src2, s8Imm:$src3),
+                            IntRegs:$src2, s8Ext:$src3),
              "$dst += add($src2, #$src3)",
              [(set (i32 IntRegs:$dst), (add (add (i32 IntRegs:$src2),
-                                                 s8ImmPred:$src3),
+                                                 s8_16ExtPred:$src3),
                                             (i32 IntRegs:$src1)))],
-             "$src1 = $dst">;
+             "$src1 = $dst">, ImmRegRel;
 
+let CextOpcode = "SUB_acc", InputType = "reg" in
 def SUBrr_acc : MInst_acc<(outs IntRegs: $dst), (ins IntRegs:$src1,
                             IntRegs:$src2, IntRegs:$src3),
              "$dst -= add($src2, $src3)",
              [(set (i32 IntRegs:$dst),
                    (sub (i32 IntRegs:$src1), (add (i32 IntRegs:$src2),
                                                   (i32 IntRegs:$src3))))],
-             "$src1 = $dst">;
+             "$src1 = $dst">, ImmRegRel;
 
+let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 8,
+CextOpcode = "SUB_acc", InputType = "imm" in
 def SUBri_acc : MInst_acc<(outs IntRegs: $dst), (ins IntRegs:$src1,
-                            IntRegs:$src2, s8Imm:$src3),
+                            IntRegs:$src2, s8Ext:$src3),
              "$dst -= add($src2, #$src3)",
              [(set (i32 IntRegs:$dst), (sub (i32 IntRegs:$src1),
                                             (add (i32 IntRegs:$src2),
-                                                 s8ImmPred:$src3)))],
-             "$src1 = $dst">;
+                                                 s8_16ExtPred:$src3)))],
+             "$src1 = $dst">, ImmRegRel;
 
 //===----------------------------------------------------------------------===//
 // MTYPE/MPYH -
@@ -1670,282 +1358,219 @@ def SUBri_acc : MInst_acc<(outs IntRegs: $dst), (ins IntRegs:$src1,
 // ST +
 //===----------------------------------------------------------------------===//
 ///
-/// Assumptions::: ****** DO NOT IGNORE ********
-/// 1. Make sure that in post increment store, the zero'th operand is always the
-///    post increment operand.
-/// 2. Make sure that the store value operand(Rt/Rtt) in a store is always the
-///    last operand.
-///
 // Store doubleword.
-let isPredicable = 1 in
-def STrid : STInst<(outs),
-            (ins MEMri:$addr, DoubleRegs:$src1),
-            "memd($addr) = $src1",
-            [(store (i64 DoubleRegs:$src1), ADDRriS11_3:$addr)]>;
-
-// Indexed store double word.
-let AddedComplexity = 10, isPredicable = 1 in
-def STrid_indexed : STInst<(outs),
-            (ins IntRegs:$src1, s11_3Imm:$src2,  DoubleRegs:$src3),
-            "memd($src1+#$src2) = $src3",
-            [(store (i64 DoubleRegs:$src3),
-                    (add (i32 IntRegs:$src1), s11_3ImmPred:$src2))]>;
 
-let neverHasSideEffects = 1 in
-def STrid_GP : STInst2<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, DoubleRegs:$src),
-            "memd(#$global+$offset) = $src",
-            []>,
-            Requires<[NoV4T]>;
+//===----------------------------------------------------------------------===//
+// Post increment store
+//===----------------------------------------------------------------------===//
 
-let neverHasSideEffects = 1 in
-def STd_GP : STInst2<(outs),
-            (ins globaladdress:$global, DoubleRegs:$src),
-            "memd(#$global) = $src",
-            []>,
-            Requires<[NoV4T]>;
-
-let hasCtrlDep = 1, isPredicable = 1 in
-def POST_STdri : STInstPI<(outs IntRegs:$dst),
-            (ins DoubleRegs:$src1, IntRegs:$src2, s4Imm:$offset),
-            "memd($src2++#$offset) = $src1",
-            [(set IntRegs:$dst,
-            (post_store (i64 DoubleRegs:$src1), (i32 IntRegs:$src2),
-                        s4_3ImmPred:$offset))],
+multiclass ST_PostInc_Pbase<string mnemonic, RegisterClass RC, Operand ImmOp,
+                            bit isNot, bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : STInst2PI<(outs IntRegs:$dst),
+            (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset, RC:$src3),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"($src2++#$offset) = $src3",
+            [],
             "$src2 = $dst">;
+}
 
-// Store doubleword conditionally.
-// if ([!]Pv) memd(Rs+#u6:3)=Rtt
-// if (Pv) memd(Rs+#u6:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_cPt : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, DoubleRegs:$src2),
-            "if ($src1) memd($addr) = $src2",
-            []>;
+multiclass ST_PostInc_Pred<string mnemonic, RegisterClass RC,
+                           Operand ImmOp, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME# : ST_PostInc_Pbase<mnemonic, RC, ImmOp, PredNot, 0>;
+    // Predicate new
+    let Predicates = [HasV4T], validSubTargets = HasV4SubT in
+    defm _cdn#NAME#_V4 : ST_PostInc_Pbase<mnemonic, RC, ImmOp, PredNot, 1>;
+  }
+}
 
-// if (!Pv) memd(Rs+#u6:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_cNotPt : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, DoubleRegs:$src2),
-            "if (!$src1) memd($addr) = $src2",
-            []>;
+let hasCtrlDep = 1, isNVStorable = 1, neverHasSideEffects = 1 in
+multiclass ST_PostInc<string mnemonic, string BaseOp, RegisterClass RC,
+                      Operand ImmOp> {
 
-// if (Pv) memd(Rs+#u6:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_indexed_cPt : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_3Imm:$src3,
-                 DoubleRegs:$src4),
-            "if ($src1) memd($src2+#$src3) = $src4",
-            []>;
+  let hasCtrlDep = 1, BaseOpcode = "POST_"#BaseOp in {
+    let isPredicable = 1 in
+    def NAME : STInst2PI<(outs IntRegs:$dst),
+                (ins IntRegs:$src1, ImmOp:$offset, RC:$src2),
+                #mnemonic#"($src1++#$offset) = $src2",
+                [],
+                "$src1 = $dst">;
+
+    let isPredicated = 1 in {
+      defm Pt : ST_PostInc_Pred<mnemonic, RC, ImmOp, 0 >;
+      defm NotPt : ST_PostInc_Pred<mnemonic, RC, ImmOp, 1 >;
+    }
+  }
+}
 
-// if (!Pv) memd(Rs+#u6:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_indexed_cNotPt : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_3Imm:$src3,
-                 DoubleRegs:$src4),
-            "if (!$src1) memd($src2+#$src3) = $src4",
-            []>;
+defm POST_STbri: ST_PostInc <"memb", "STrib", IntRegs, s4_0Imm>, AddrModeRel;
+defm POST_SThri: ST_PostInc <"memh", "STrih", IntRegs, s4_1Imm>, AddrModeRel;
+defm POST_STwri: ST_PostInc <"memw", "STriw", IntRegs, s4_2Imm>, AddrModeRel;
 
-// if ([!]Pv) memd(Rx++#s4:3)=Rtt
-// if (Pv) memd(Rx++#s4:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def POST_STdri_cPt : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, DoubleRegs:$src2, IntRegs:$src3,
-                 s4_3Imm:$offset),
-            "if ($src1) memd($src3++#$offset) = $src2",
-            [],
-            "$src3 = $dst">;
-
-// if (!Pv) memd(Rx++#s4:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1, isPredicated = 1,
-    isPredicated = 1 in
-def POST_STdri_cNotPt : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, DoubleRegs:$src2, IntRegs:$src3,
-                 s4_3Imm:$offset),
-            "if (!$src1) memd($src3++#$offset) = $src2",
-            [],
-            "$src3 = $dst">;
+let isNVStorable = 0 in
+defm POST_STdri: ST_PostInc <"memd", "STrid", DoubleRegs, s4_3Imm>, AddrModeRel;
 
+def : Pat<(post_truncsti8 (i32 IntRegs:$src1), IntRegs:$src2,
+                           s4_3ImmPred:$offset),
+          (POST_STbri IntRegs:$src2, s4_0ImmPred:$offset, IntRegs:$src1)>;
 
-// Store byte.
-// memb(Rs+#s11:0)=Rt
-let isPredicable = 1 in
-def STrib : STInst<(outs),
-            (ins MEMri:$addr, IntRegs:$src1),
-            "memb($addr) = $src1",
-            [(truncstorei8 (i32 IntRegs:$src1), ADDRriS11_0:$addr)]>;
-
-let AddedComplexity = 10, isPredicable = 1 in
-def STrib_indexed : STInst<(outs),
-            (ins IntRegs:$src1, s11_0Imm:$src2, IntRegs:$src3),
-            "memb($src1+#$src2) = $src3",
-            [(truncstorei8 (i32 IntRegs:$src3), (add (i32 IntRegs:$src1),
-                                                     s11_0ImmPred:$src2))]>;
-
-// memb(gp+#u16:0)=Rt
-let neverHasSideEffects = 1 in
-def STrib_GP : STInst2<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src),
-            "memb(#$global+$offset) = $src",
-            []>,
-            Requires<[NoV4T]>;
+def : Pat<(post_truncsti16 (i32 IntRegs:$src1), IntRegs:$src2,
+                            s4_3ImmPred:$offset),
+          (POST_SThri IntRegs:$src2, s4_1ImmPred:$offset, IntRegs:$src1)>;
 
-// memb(#global)=Rt
-let neverHasSideEffects = 1 in
-def STb_GP : STInst2<(outs),
-            (ins globaladdress:$global, IntRegs:$src),
-            "memb(#$global) = $src",
-            []>,
-            Requires<[NoV4T]>;
-
-// memb(Rx++#s4:0)=Rt
-let hasCtrlDep = 1, isPredicable = 1 in
-def POST_STbri : STInstPI<(outs IntRegs:$dst), (ins IntRegs:$src1,
-                                                    IntRegs:$src2,
-                                                    s4Imm:$offset),
-            "memb($src2++#$offset) = $src1",
-            [(set IntRegs:$dst,
-            (post_truncsti8 (i32 IntRegs:$src1), (i32 IntRegs:$src2),
-                            s4_0ImmPred:$offset))],
-            "$src2 = $dst">;
+def : Pat<(post_store (i32 IntRegs:$src1), IntRegs:$src2, s4_2ImmPred:$offset),
+          (POST_STwri IntRegs:$src2, s4_1ImmPred:$offset, IntRegs:$src1)>;
 
-// Store byte conditionally.
-// if ([!]Pv) memb(Rs+#u6:0)=Rt
-// if (Pv) memb(Rs+#u6:0)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrib_cPt : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1) memb($addr) = $src2",
-            []>;
+def : Pat<(post_store (i64 DoubleRegs:$src1), IntRegs:$src2,
+                       s4_3ImmPred:$offset),
+          (POST_STdri IntRegs:$src2, s4_3ImmPred:$offset, DoubleRegs:$src1)>;
 
-// if (!Pv) memb(Rs+#u6:0)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrib_cNotPt : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1) memb($addr) = $src2",
+//===----------------------------------------------------------------------===//
+// multiclass for the store instructions with MEMri operand.
+//===----------------------------------------------------------------------===//
+multiclass ST_MEMri_Pbase<string mnemonic, RegisterClass RC, bit isNot,
+                          bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : STInst2<(outs),
+            (ins PredRegs:$src1, MEMri:$addr, RC: $src2),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"($addr) = $src2",
             []>;
+}
 
-// if (Pv) memb(Rs+#u6:0)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrib_indexed_cPt : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4),
-            "if ($src1) memb($src2+#$src3) = $src4",
-            []>;
+multiclass ST_MEMri_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ST_MEMri_Pbase<mnemonic, RC, PredNot, 0>;
 
-// if (!Pv) memb(Rs+#u6:0)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrib_indexed_cNotPt : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4),
-            "if (!$src1) memb($src2+#$src3) = $src4",
-            []>;
+    // Predicate new
+    let validSubTargets = HasV4SubT, Predicates = [HasV4T] in
+    defm _cdn#NAME#_V4 : ST_MEMri_Pbase<mnemonic, RC, PredNot, 1>;
+  }
+}
 
-// if ([!]Pv) memb(Rx++#s4:0)=Rt
-// if (Pv) memb(Rx++#s4:0)=Rt
-let hasCtrlDep = 1, isPredicated = 1 in
-def POST_STbri_cPt : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset),
-            "if ($src1) memb($src3++#$offset) = $src2",
-            [],"$src3 = $dst">;
+let isExtendable = 1, isNVStorable = 1, neverHasSideEffects = 1 in
+multiclass ST_MEMri<string mnemonic, string CextOp, RegisterClass RC,
+                    bits<5> ImmBits, bits<5> PredImmBits> {
 
-// if (!Pv) memb(Rx++#s4:0)=Rt
-let hasCtrlDep = 1, isPredicated = 1 in
-def POST_STbri_cNotPt : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset),
-            "if (!$src1) memb($src3++#$offset) = $src2",
-            [],"$src3 = $dst">;
+  let CextOpcode = CextOp, BaseOpcode = CextOp in {
+    let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits,
+         isPredicable = 1 in
+    def NAME : STInst2<(outs),
+            (ins MEMri:$addr, RC:$src),
+            mnemonic#"($addr) = $src",
+            []>;
 
+    let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits,
+        isPredicated = 1 in {
+      defm Pt : ST_MEMri_Pred<mnemonic, RC, 0>;
+      defm NotPt : ST_MEMri_Pred<mnemonic, RC, 1>;
+    }
+  }
+}
 
-// Store halfword.
-// memh(Rs+#s11:1)=Rt
-let isPredicable = 1 in
-def STrih : STInst<(outs),
-            (ins MEMri:$addr, IntRegs:$src1),
-            "memh($addr) = $src1",
-            [(truncstorei16 (i32 IntRegs:$src1), ADDRriS11_1:$addr)]>;
+let addrMode = BaseImmOffset, isMEMri = "true" in {
+  defm STrib: ST_MEMri < "memb", "STrib", IntRegs, 11, 6>, AddrModeRel;
+  defm STrih: ST_MEMri < "memh", "STrih", IntRegs, 12, 7>, AddrModeRel;
+  defm STriw: ST_MEMri < "memw", "STriw", IntRegs, 13, 8>, AddrModeRel;
 
+  let isNVStorable = 0 in
+  defm STrid: ST_MEMri < "memd", "STrid", DoubleRegs, 14, 9>, AddrModeRel;
+}
 
-let AddedComplexity = 10, isPredicable = 1 in
-def STrih_indexed : STInst<(outs),
-            (ins IntRegs:$src1, s11_1Imm:$src2,  IntRegs:$src3),
-            "memh($src1+#$src2) = $src3",
-            [(truncstorei16 (i32 IntRegs:$src3), (add (i32 IntRegs:$src1),
-                                                      s11_1ImmPred:$src2))]>;
+def : Pat<(truncstorei8 (i32 IntRegs:$src1), ADDRriS11_0:$addr),
+          (STrib ADDRriS11_0:$addr, (i32 IntRegs:$src1))>;
 
-let neverHasSideEffects = 1 in
-def STrih_GP : STInst2<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src),
-            "memh(#$global+$offset) = $src",
-            []>,
-            Requires<[NoV4T]>;
+def : Pat<(truncstorei16 (i32 IntRegs:$src1), ADDRriS11_1:$addr),
+          (STrih ADDRriS11_1:$addr, (i32 IntRegs:$src1))>;
 
-let neverHasSideEffects = 1 in
-def STh_GP   : STInst2<(outs),
-            (ins globaladdress:$global, IntRegs:$src),
-            "memh(#$global) = $src",
-            []>,
-            Requires<[NoV4T]>;
+def : Pat<(store (i32 IntRegs:$src1), ADDRriS11_2:$addr),
+          (STriw ADDRriS11_2:$addr, (i32 IntRegs:$src1))>;
 
-// memh(Rx++#s4:1)=Rt.H
-// memh(Rx++#s4:1)=Rt
-let hasCtrlDep = 1, isPredicable = 1 in
-def POST_SThri : STInstPI<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2, s4Imm:$offset),
-            "memh($src2++#$offset) = $src1",
-            [(set IntRegs:$dst,
-            (post_truncsti16 (i32 IntRegs:$src1), (i32 IntRegs:$src2),
-                             s4_1ImmPred:$offset))],
-            "$src2 = $dst">;
+def : Pat<(store (i64 DoubleRegs:$src1), ADDRriS11_3:$addr),
+          (STrid ADDRriS11_3:$addr, (i64 DoubleRegs:$src1))>;
 
-// Store halfword conditionally.
-// if ([!]Pv) memh(Rs+#u6:1)=Rt
-// if (Pv) memh(Rs+#u6:1)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrih_cPt : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1) memh($addr) = $src2",
-            []>;
 
-// if (!Pv) memh(Rs+#u6:1)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrih_cNotPt : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1) memh($addr) = $src2",
+//===----------------------------------------------------------------------===//
+// multiclass for the store instructions with base+immediate offset
+// addressing mode
+//===----------------------------------------------------------------------===//
+multiclass ST_Idxd_Pbase<string mnemonic, RegisterClass RC, Operand predImmOp,
+                        bit isNot, bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : STInst2<(outs),
+            (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3, RC: $src4),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"($src2+#$src3) = $src4",
             []>;
+}
 
-// if (Pv) memh(Rs+#u6:1)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrih_indexed_cPt : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4),
-            "if ($src1) memh($src2+#$src3) = $src4",
-            []>;
+multiclass ST_Idxd_Pred<string mnemonic, RegisterClass RC, Operand predImmOp,
+                        bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true"), isPredicated = 1 in {
+    defm _c#NAME : ST_Idxd_Pbase<mnemonic, RC, predImmOp, PredNot, 0>;
 
-// if (!Pv) memh(Rs+#u6:1)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrih_indexed_cNotPt : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4),
-            "if (!$src1) memh($src2+#$src3) = $src4",
+    // Predicate new
+    let validSubTargets = HasV4SubT, Predicates = [HasV4T] in
+    defm _cdn#NAME#_V4 : ST_Idxd_Pbase<mnemonic, RC, predImmOp, PredNot, 1>;
+  }
+}
+
+let isExtendable = 1, isNVStorable = 1, neverHasSideEffects = 1 in
+multiclass ST_Idxd<string mnemonic, string CextOp, RegisterClass RC,
+                   Operand ImmOp, Operand predImmOp, bits<5> ImmBits,
+                   bits<5> PredImmBits> {
+
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
+    let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits,
+         isPredicable = 1 in
+    def NAME : STInst2<(outs),
+            (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
+            mnemonic#"($src1+#$src2) = $src3",
             []>;
 
-// if ([!]Pv) memh(Rx++#s4:1)=Rt
-// if (Pv) memh(Rx++#s4:1)=Rt
-let hasCtrlDep = 1, isPredicated = 1 in
-def POST_SThri_cPt : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset),
-            "if ($src1) memh($src3++#$offset) = $src2",
-            [],"$src3 = $dst">;
+    let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits in {
+      defm Pt : ST_Idxd_Pred<mnemonic, RC, predImmOp, 0>;
+      defm NotPt : ST_Idxd_Pred<mnemonic, RC, predImmOp, 1>;
+    }
+  }
+}
+
+let addrMode = BaseImmOffset, InputType = "reg" in {
+  defm STrib_indexed: ST_Idxd < "memb", "STrib", IntRegs, s11_0Ext,
+                                u6_0Ext, 11, 6>, AddrModeRel, ImmRegRel;
+  defm STrih_indexed: ST_Idxd < "memh", "STrih", IntRegs, s11_1Ext,
+                                u6_1Ext, 12, 7>, AddrModeRel, ImmRegRel;
+  defm STriw_indexed: ST_Idxd < "memw", "STriw", IntRegs, s11_2Ext,
+                                u6_2Ext, 13, 8>, AddrModeRel, ImmRegRel;
+  let isNVStorable = 0 in
+  defm STrid_indexed: ST_Idxd < "memd", "STrid", DoubleRegs, s11_3Ext,
+                                u6_3Ext, 14, 9>, AddrModeRel;
+}
+
+let AddedComplexity = 10 in {
+def : Pat<(truncstorei8 (i32 IntRegs:$src1), (add IntRegs:$src2,
+                                                  s11_0ExtPred:$offset)),
+          (STrib_indexed IntRegs:$src2, s11_0ImmPred:$offset,
+                         (i32 IntRegs:$src1))>;
+
+def : Pat<(truncstorei16 (i32 IntRegs:$src1), (add IntRegs:$src2,
+                                                   s11_1ExtPred:$offset)),
+          (STrih_indexed IntRegs:$src2, s11_1ImmPred:$offset,
+                         (i32 IntRegs:$src1))>;
 
-// if (!Pv) memh(Rx++#s4:1)=Rt
-let hasCtrlDep = 1, isPredicated = 1 in
-def POST_SThri_cNotPt : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset),
-            "if (!$src1) memh($src3++#$offset) = $src2",
-            [],"$src3 = $dst">;
+def : Pat<(store (i32 IntRegs:$src1), (add IntRegs:$src2,
+                                           s11_2ExtPred:$offset)),
+          (STriw_indexed IntRegs:$src2, s11_2ImmPred:$offset,
+                         (i32 IntRegs:$src1))>;
 
+def : Pat<(store (i64 DoubleRegs:$src1), (add IntRegs:$src2,
+                                              s11_3ExtPred:$offset)),
+          (STrid_indexed IntRegs:$src2, s11_3ImmPred:$offset,
+                         (i64 DoubleRegs:$src1))>;
+}
+
+// memh(Rx++#s4:1)=Rt.H
 
 // Store word.
 // Store predicate.
@@ -1955,90 +1580,6 @@ def STriw_pred : STInst2<(outs),
             "Error; should not emit",
             []>;
 
-// memw(Rs+#s11:2)=Rt
-let isPredicable = 1 in
-def STriw : STInst<(outs),
-            (ins MEMri:$addr, IntRegs:$src1),
-            "memw($addr) = $src1",
-            [(store (i32 IntRegs:$src1), ADDRriS11_2:$addr)]>;
-
-let AddedComplexity = 10, isPredicable = 1 in
-def STriw_indexed : STInst<(outs),
-            (ins IntRegs:$src1, s11_2Imm:$src2, IntRegs:$src3),
-            "memw($src1+#$src2) = $src3",
-            [(store (i32 IntRegs:$src3),
-                    (add (i32 IntRegs:$src1), s11_2ImmPred:$src2))]>;
-
-let neverHasSideEffects = 1 in
-def STriw_GP : STInst2<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src),
-            "memw(#$global+$offset) = $src",
-            []>,
-            Requires<[NoV4T]>;
-
-let neverHasSideEffects = 1 in
-def STw_GP : STInst2<(outs),
-            (ins globaladdress:$global, IntRegs:$src),
-            "memw(#$global) = $src",
-            []>,
-            Requires<[NoV4T]>;
-
-let hasCtrlDep = 1, isPredicable = 1  in
-def POST_STwri : STInstPI<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2, s4Imm:$offset),
-            "memw($src2++#$offset) = $src1",
-            [(set IntRegs:$dst,
-            (post_store (i32 IntRegs:$src1), (i32 IntRegs:$src2),
-                        s4_2ImmPred:$offset))],
-            "$src2 = $dst">;
-
-// Store word conditionally.
-// if ([!]Pv) memw(Rs+#u6:2)=Rt
-// if (Pv) memw(Rs+#u6:2)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STriw_cPt : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1) memw($addr) = $src2",
-            []>;
-
-// if (!Pv) memw(Rs+#u6:2)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STriw_cNotPt : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1) memw($addr) = $src2",
-            []>;
-
-// if (Pv) memw(Rs+#u6:2)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STriw_indexed_cPt : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4),
-            "if ($src1) memw($src2+#$src3) = $src4",
-            []>;
-
-// if (!Pv) memw(Rs+#u6:2)=Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STriw_indexed_cNotPt : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4),
-            "if (!$src1) memw($src2+#$src3) = $src4",
-            []>;
-
-// if ([!]Pv) memw(Rx++#s4:2)=Rt
-// if (Pv) memw(Rx++#s4:2)=Rt
-let hasCtrlDep = 1, isPredicated = 1 in
-def POST_STwri_cPt : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset),
-            "if ($src1) memw($src3++#$offset) = $src2",
-            [],"$src3 = $dst">;
-
-// if (!Pv) memw(Rx++#s4:2)=Rt
-let hasCtrlDep = 1, isPredicated = 1 in
-def POST_STwri_cNotPt : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset),
-            "if (!$src1) memw($src3++#$offset) = $src2",
-            [],"$src3 = $dst">;
-
-
-
 // Allocate stack frame.
 let Defs = [R29, R30], Uses = [R31, R30], neverHasSideEffects = 1 in {
   def ALLOCFRAME : STInst2<(outs),
@@ -2241,7 +1782,7 @@ def SDHexagonBARRIER: SDTypeProfile<0, 0, []>;
 def HexagonBARRIER: SDNode<"HexagonISD::BARRIER", SDHexagonBARRIER,
                            [SDNPHasChain]>;
 
-let hasSideEffects = 1, isHexagonSolo = 1 in
+let hasSideEffects = 1, isSolo = 1 in
 def BARRIER : SYSInst<(outs), (ins),
                      "barrier",
                      [(HexagonBARRIER)]>;
@@ -2316,9 +1857,9 @@ def LOOP0_r : CRInst<(outs), (ins brtarget:$offset, IntRegs:$src2),
 
 let isBranch = 1, isTerminator = 1, neverHasSideEffects = 1,
     Defs = [PC, LC0], Uses = [SA0, LC0] in {
-def ENDLOOP0 : Marker<(outs), (ins brtarget:$offset),
-                      ":endloop0",
-                      []>;
+def ENDLOOP0 : Endloop<(outs), (ins brtarget:$offset),
+                       ":endloop0",
+                       []>;
 }
 
 // Support for generating global address.
@@ -2406,6 +1947,10 @@ def CONST32_Int_Real : LDInst2<(outs IntRegs:$dst), (ins i32imm:$global),
                        "$dst = CONST32(#$global)",
                        [(set (i32 IntRegs:$dst), imm:$global) ]>;
 
+// Map BlockAddress lowering to CONST32_Int_Real
+def : Pat<(HexagonCONST32_GP tblockaddress:$addr),
+          (CONST32_Int_Real tblockaddress:$addr)>;
+
 let isReMaterializable = 1, isMoveImm = 1 in
 def CONST32_Label : LDInst2<(outs IntRegs:$dst), (ins bblabel:$label),
                     "$dst = CONST32($label)",
@@ -2509,68 +2054,26 @@ def : Pat<(HexagonTCRet (i32 IntRegs:$dst)),
 
 // Atomic load and store support
 // 8 bit atomic load
-def : Pat<(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)),
-          (i32 (LDub_GP tglobaladdr:$global))>,
-            Requires<[NoV4T]>;
-
-def : Pat<(atomic_load_8 (add (HexagonCONST32_GP tglobaladdr:$global),
-                              u16ImmPred:$offset)),
-          (i32 (LDriub_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-            Requires<[NoV4T]>;
-
 def : Pat<(atomic_load_8 ADDRriS11_0:$src1),
           (i32 (LDriub ADDRriS11_0:$src1))>;
 
 def : Pat<(atomic_load_8 (add (i32 IntRegs:$src1), s11_0ImmPred:$offset)),
           (i32 (LDriub_indexed (i32 IntRegs:$src1), s11_0ImmPred:$offset))>;
 
-
-
 // 16 bit atomic load
-def : Pat<(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)),
-          (i32 (LDuh_GP tglobaladdr:$global))>,
-            Requires<[NoV4T]>;
-
-def : Pat<(atomic_load_16 (add (HexagonCONST32_GP tglobaladdr:$global),
-                               u16ImmPred:$offset)),
-          (i32 (LDriuh_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-            Requires<[NoV4T]>;
-
 def : Pat<(atomic_load_16 ADDRriS11_1:$src1),
           (i32 (LDriuh ADDRriS11_1:$src1))>;
 
 def : Pat<(atomic_load_16 (add (i32 IntRegs:$src1), s11_1ImmPred:$offset)),
           (i32 (LDriuh_indexed (i32 IntRegs:$src1), s11_1ImmPred:$offset))>;
 
-
-
-// 32 bit atomic load
-def : Pat<(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)),
-          (i32 (LDw_GP tglobaladdr:$global))>,
-            Requires<[NoV4T]>;
-
-def : Pat<(atomic_load_32 (add (HexagonCONST32_GP tglobaladdr:$global),
-                               u16ImmPred:$offset)),
-          (i32 (LDriw_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-            Requires<[NoV4T]>;
-
 def : Pat<(atomic_load_32 ADDRriS11_2:$src1),
           (i32 (LDriw ADDRriS11_2:$src1))>;
 
 def : Pat<(atomic_load_32 (add (i32 IntRegs:$src1), s11_2ImmPred:$offset)),
           (i32 (LDriw_indexed (i32 IntRegs:$src1), s11_2ImmPred:$offset))>;
 
-
 // 64 bit atomic load
-def : Pat<(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)),
-          (i64 (LDd_GP tglobaladdr:$global))>,
-            Requires<[NoV4T]>;
-
-def : Pat<(atomic_load_64 (add (HexagonCONST32_GP tglobaladdr:$global),
-                               u16ImmPred:$offset)),
-          (i64 (LDrid_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-          Requires<[NoV4T]>;
-
 def : Pat<(atomic_load_64 ADDRriS11_3:$src1),
           (i64 (LDrid ADDRriS11_3:$src1))>;
 
@@ -2578,30 +2081,6 @@ def : Pat<(atomic_load_64 (add (i32 IntRegs:$src1), s11_3ImmPred:$offset)),
           (i64 (LDrid_indexed (i32 IntRegs:$src1), s11_3ImmPred:$offset))>;
 
 
-// 64 bit atomic store
-def : Pat<(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global),
-                           (i64 DoubleRegs:$src1)),
-          (STd_GP tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
-          Requires<[NoV4T]>;
-
-def : Pat<(atomic_store_64 (add (HexagonCONST32_GP tglobaladdr:$global),
-                                u16ImmPred:$offset),
-                           (i64 DoubleRegs:$src1)),
-          (STrid_GP tglobaladdr:$global, u16ImmPred:$offset,
-                    (i64 DoubleRegs:$src1))>, Requires<[NoV4T]>;
-
-// 8 bit atomic store
-def : Pat<(atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global),
-                          (i32 IntRegs:$src1)),
-          (STb_GP tglobaladdr:$global, (i32 IntRegs:$src1))>,
-          Requires<[NoV4T]>;
-
-def : Pat<(atomic_store_8 (add (HexagonCONST32_GP tglobaladdr:$global),
-                               u16ImmPred:$offset),
-                          (i32 IntRegs:$src1)),
-          (STrib_GP tglobaladdr:$global, u16ImmPred:$offset,
-                    (i32 IntRegs:$src1))>, Requires<[NoV4T]>;
-
 def : Pat<(atomic_store_8 ADDRriS11_0:$src2, (i32 IntRegs:$src1)),
           (STrib ADDRriS11_0:$src2, (i32 IntRegs:$src1))>;
 
@@ -2611,18 +2090,6 @@ def : Pat<(atomic_store_8 (add (i32 IntRegs:$src2), s11_0ImmPred:$offset),
                          (i32 IntRegs:$src1))>;
 
 
-// 16 bit atomic store
-def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global),
-                           (i32 IntRegs:$src1)),
-          (STh_GP tglobaladdr:$global, (i32 IntRegs:$src1))>,
-          Requires<[NoV4T]>;
-
-def : Pat<(atomic_store_16 (add (HexagonCONST32_GP tglobaladdr:$global),
-                                u16ImmPred:$offset),
-                           (i32 IntRegs:$src1)),
-          (STrih_GP tglobaladdr:$global, u16ImmPred:$offset,
-                    (i32 IntRegs:$src1))>, Requires<[NoV4T]>;
-
 def : Pat<(atomic_store_16 ADDRriS11_1:$src2, (i32 IntRegs:$src1)),
           (STrih ADDRriS11_1:$src2, (i32 IntRegs:$src1))>;
 
@@ -2631,20 +2098,6 @@ def : Pat<(atomic_store_16 (i32 IntRegs:$src1),
           (STrih_indexed (i32 IntRegs:$src2), s11_1ImmPred:$offset,
                          (i32 IntRegs:$src1))>;
 
-
-// 32 bit atomic store
-def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global),
-                           (i32 IntRegs:$src1)),
-          (STw_GP tglobaladdr:$global, (i32 IntRegs:$src1))>,
-          Requires<[NoV4T]>;
-
-def : Pat<(atomic_store_32 (add (HexagonCONST32_GP tglobaladdr:$global),
-                                u16ImmPred:$offset),
-                           (i32 IntRegs:$src1)),
-          (STriw_GP tglobaladdr:$global, u16ImmPred:$offset,
-                                         (i32 IntRegs:$src1))>,
-            Requires<[NoV4T]>;
-
 def : Pat<(atomic_store_32 ADDRriS11_2:$src2, (i32 IntRegs:$src1)),
           (STriw ADDRriS11_2:$src2, (i32 IntRegs:$src1))>;
 
@@ -2713,198 +2166,8 @@ def : Pat <(brcond (not PredRegs:$src1), bb:$offset),
 def : Pat <(and PredRegs:$src1, (not PredRegs:$src2)),
       (i1 (AND_pnotp (i1 PredRegs:$src1), (i1 PredRegs:$src2)))>;
 
-// Map from store(globaladdress + x) -> memd(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(store (i64 DoubleRegs:$src1),
-                  (add (HexagonCONST32_GP tglobaladdr:$global),
-                       u16ImmPred:$offset)),
-      (STrid_GP tglobaladdr:$global, u16ImmPred:$offset,
-                (i64 DoubleRegs:$src1))>, Requires<[NoV4T]>;
-
-// Map from store(globaladdress) -> memd(#foo).
-let AddedComplexity = 100 in
-def : Pat <(store (i64 DoubleRegs:$src1),
-                  (HexagonCONST32_GP tglobaladdr:$global)),
-      (STd_GP tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
-      Requires<[NoV4T]>;
-
-// Map from store(globaladdress + x) -> memw(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(store (i32 IntRegs:$src1),
-              (add (HexagonCONST32_GP tglobaladdr:$global),
-                                      u16ImmPred:$offset)),
-      (STriw_GP tglobaladdr:$global, u16ImmPred:$offset, (i32 IntRegs:$src1))>,
-      Requires<[NoV4T]>;
-
-// Map from store(globaladdress) -> memw(#foo + 0).
-let AddedComplexity = 100 in
-def : Pat <(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)),
-      (STriw_GP tglobaladdr:$global, 0, (i32 IntRegs:$src1))>;
-
-// Map from store(globaladdress) -> memw(#foo).
-let AddedComplexity = 100 in
-def : Pat <(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)),
-      (STriw_GP tglobaladdr:$global, 0, (i32 IntRegs:$src1))>,
-      Requires<[NoV4T]>;
-
-// Map from store(globaladdress + x) -> memh(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(truncstorei16 (i32 IntRegs:$src1),
-                          (add (HexagonCONST32_GP tglobaladdr:$global),
-                               u16ImmPred:$offset)),
-      (STrih_GP tglobaladdr:$global, u16ImmPred:$offset, (i32 IntRegs:$src1))>,
-      Requires<[NoV4T]>;
-
-// Map from store(globaladdress) -> memh(#foo).
-let AddedComplexity = 100 in
-def : Pat <(truncstorei16 (i32 IntRegs:$src1),
-                          (HexagonCONST32_GP tglobaladdr:$global)),
-      (STh_GP tglobaladdr:$global, (i32 IntRegs:$src1))>,
-      Requires<[NoV4T]>;
-
-// Map from store(globaladdress + x) -> memb(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(truncstorei8 (i32 IntRegs:$src1),
-                         (add (HexagonCONST32_GP tglobaladdr:$global),
-                              u16ImmPred:$offset)),
-      (STrib_GP tglobaladdr:$global, u16ImmPred:$offset, (i32 IntRegs:$src1))>,
-      Requires<[NoV4T]>;
-
-// Map from store(globaladdress) -> memb(#foo).
-let AddedComplexity = 100 in
-def : Pat <(truncstorei8 (i32 IntRegs:$src1),
-                         (HexagonCONST32_GP tglobaladdr:$global)),
-      (STb_GP tglobaladdr:$global, (i32 IntRegs:$src1))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress + x) -> memw(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(i32 (load (add (HexagonCONST32_GP tglobaladdr:$global),
-                      u16ImmPred:$offset))),
-      (i32 (LDriw_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress) -> memw(#foo).
-let AddedComplexity = 100 in
-def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDw_GP tglobaladdr:$global))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress + x) -> memd(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(i64 (load (add (HexagonCONST32_GP tglobaladdr:$global),
-                           u16ImmPred:$offset))),
-      (i64 (LDrid_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress) -> memw(#foo + 0).
-let AddedComplexity = 100 in
-def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))),
-      (i64 (LDd_GP tglobaladdr:$global))>,
-      Requires<[NoV4T]>;
-
-// Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd.
-let AddedComplexity = 100 in
-def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))),
-      (i1 (TFR_PdRs (i32 (LDb_GP tglobaladdr:$global))))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress + x) -> memh(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(i32 (extloadi16 (add (HexagonCONST32_GP tglobaladdr:$global),
-                            u16ImmPred:$offset))),
-      (i32 (LDrih_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress + x) -> memh(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDrih_GP tglobaladdr:$global, 0))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress + x) -> memuh(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi16 (add (HexagonCONST32_GP tglobaladdr:$global),
-                             u16ImmPred:$offset))),
-      (i32 (LDriuh_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress) -> memuh(#foo).
-let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDriuh_GP tglobaladdr:$global, 0))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress) -> memh(#foo).
-let AddedComplexity = 100 in
-def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDh_GP tglobaladdr:$global))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress) -> memuh(#foo).
-let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDuh_GP tglobaladdr:$global))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress + x) -> memb(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(i32 (extloadi8 (add (HexagonCONST32_GP tglobaladdr:$global),
-                           u16ImmPred:$offset))),
-      (i32 (LDrib_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress + x) -> memb(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(i32 (sextloadi8 (add (HexagonCONST32_GP tglobaladdr:$global),
-                            u16ImmPred:$offset))),
-      (i32 (LDrib_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress + x) -> memub(#foo + x).
-let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi8 (add (HexagonCONST32_GP tglobaladdr:$global),
-                            u16ImmPred:$offset))),
-      (i32 (LDriub_GP tglobaladdr:$global, u16ImmPred:$offset))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress) -> memb(#foo).
-let AddedComplexity = 100 in
-def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDb_GP tglobaladdr:$global))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress) -> memb(#foo).
-let AddedComplexity = 100 in
-def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDb_GP tglobaladdr:$global))>,
-      Requires<[NoV4T]>;
-
-// Map from load(globaladdress) -> memub(#foo).
-let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDub_GP tglobaladdr:$global))>,
-      Requires<[NoV4T]>;
-
-// When the Interprocedural Global Variable optimizer realizes that a
-// certain global variable takes only two constant values, it shrinks the
-// global to a boolean. Catch those loads here in the following 3 patterns.
-let AddedComplexity = 100 in
-def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDb_GP tglobaladdr:$global))>,
-      Requires<[NoV4T]>;
-
-let AddedComplexity = 100 in
-def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDb_GP tglobaladdr:$global))>,
-      Requires<[NoV4T]>;
-
-let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
-      (i32 (LDub_GP tglobaladdr:$global))>,
-      Requires<[NoV4T]>;
-
 // Map from i1 loads to 32 bits. This assumes that the i1* is byte aligned.
+let AddedComplexity = 10 in
 def : Pat <(i32 (zextloadi1 ADDRriS11_0:$addr)),
       (i32 (AND_rr (i32 (LDrib ADDRriS11_0:$addr)), (TFRI 0x1)))>;
 
@@ -3020,12 +2283,6 @@ def : Pat<(truncstorei32 (i64 DoubleRegs:$src), ADDRriS11_0:$addr),
 def : Pat<(store (i1 -1), ADDRriS11_2:$addr),
       (STrib ADDRriS11_2:$addr, (TFRI 1))>;
 
-let AddedComplexity = 100 in
-// Map from i1 = constant<-1>; memw(CONST32(#foo)) = i1 -> r0 = 1;
-// memw(#foo) = r0
-def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)),
-      (STb_GP tglobaladdr:$global, (TFRI 1))>,
-      Requires<[NoV4T]>;
 
 // Map from i1 = constant<-1>; store i1 -> r0 = 1; store r0.
 def : Pat<(store (i1 -1), ADDRriS11_2:$addr),
@@ -3181,23 +2438,54 @@ def : Pat <(i32 (zext (i1 PredRegs:$src1))),
 
 // i1 -> i64
 def : Pat <(i64 (zext (i1 PredRegs:$src1))),
-      (i64 (COMBINE_rr (TFRI 0), (MUX_ii (i1 PredRegs:$src1), 1, 0)))>;
+      (i64 (COMBINE_rr (TFRI 0), (MUX_ii (i1 PredRegs:$src1), 1, 0)))>,
+      Requires<[NoV4T]>;
 
 // i32 -> i64
 def : Pat <(i64 (zext (i32 IntRegs:$src1))),
-      (i64 (COMBINE_rr (TFRI 0), (i32 IntRegs:$src1)))>;
+      (i64 (COMBINE_rr (TFRI 0), (i32 IntRegs:$src1)))>,
+      Requires<[NoV4T]>;
 
 // i8 -> i64
 def:  Pat <(i64 (zextloadi8 ADDRriS11_0:$src1)),
-      (i64 (COMBINE_rr (TFRI 0), (LDriub ADDRriS11_0:$src1)))>;
+      (i64 (COMBINE_rr (TFRI 0), (LDriub ADDRriS11_0:$src1)))>,
+      Requires<[NoV4T]>;
+
+let AddedComplexity = 20 in
+def:  Pat <(i64 (zextloadi8 (add (i32 IntRegs:$src1),
+                                s11_0ExtPred:$offset))),
+      (i64 (COMBINE_rr (TFRI 0), (LDriub_indexed IntRegs:$src1,
+                                  s11_0ExtPred:$offset)))>,
+      Requires<[NoV4T]>;
+
+// i1 -> i64
+def:  Pat <(i64 (zextloadi1 ADDRriS11_0:$src1)),
+      (i64 (COMBINE_rr (TFRI 0), (LDriub ADDRriS11_0:$src1)))>,
+      Requires<[NoV4T]>;
+
+let AddedComplexity = 20 in
+def:  Pat <(i64 (zextloadi1 (add (i32 IntRegs:$src1),
+                                s11_0ExtPred:$offset))),
+      (i64 (COMBINE_rr (TFRI 0), (LDriub_indexed IntRegs:$src1,
+                                  s11_0ExtPred:$offset)))>,
+      Requires<[NoV4T]>;
 
 // i16 -> i64
 def:  Pat <(i64 (zextloadi16 ADDRriS11_1:$src1)),
-      (i64 (COMBINE_rr (TFRI 0), (LDriuh ADDRriS11_1:$src1)))>;
+      (i64 (COMBINE_rr (TFRI 0), (LDriuh ADDRriS11_1:$src1)))>,
+      Requires<[NoV4T]>;
+
+let AddedComplexity = 20 in
+def:  Pat <(i64 (zextloadi16 (add (i32 IntRegs:$src1),
+                                  s11_1ExtPred:$offset))),
+      (i64 (COMBINE_rr (TFRI 0), (LDriuh_indexed IntRegs:$src1,
+                                  s11_1ExtPred:$offset)))>,
+      Requires<[NoV4T]>;
 
 // i32 -> i64
 def:  Pat <(i64 (zextloadi32 ADDRriS11_2:$src1)),
-      (i64 (COMBINE_rr (TFRI 0), (LDriw ADDRriS11_2:$src1)))>;
+      (i64 (COMBINE_rr (TFRI 0), (LDriw ADDRriS11_2:$src1)))>,
+      Requires<[NoV4T]>;
 
 def:  Pat <(i32 (zextloadi1 ADDRriS11_0:$src1)),
       (i32 (LDriw ADDRriS11_0:$src1))>;
@@ -3218,15 +2506,41 @@ def : Pat <(i64 (anyext (i1 PredRegs:$src1))),
 // Any extended 64-bit load.
 // anyext i32 -> i64
 def:  Pat <(i64 (extloadi32 ADDRriS11_2:$src1)),
-      (i64 (COMBINE_rr (TFRI 0), (LDriw ADDRriS11_2:$src1)))>;
+      (i64 (COMBINE_rr (TFRI 0), (LDriw ADDRriS11_2:$src1)))>,
+      Requires<[NoV4T]>;
+
+// When there is an offset we should prefer the pattern below over the pattern above.
+// The complexity of the above is 13 (gleaned from HexagonGenDAGIsel.inc)
+// So this complexity below is comfortably higher to allow for choosing the below.
+// If this is not done then we generate addresses such as
+// ********************************************
+//        r1 = add (r0, #4)
+//        r1 = memw(r1 + #0)
+//  instead of
+//        r1 = memw(r0 + #4)
+// ********************************************
+let AddedComplexity = 100 in
+def:  Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
+      (i64 (COMBINE_rr (TFRI 0), (LDriw_indexed IntRegs:$src1,
+                                  s11_2ExtPred:$offset)))>,
+      Requires<[NoV4T]>;
 
 // anyext i16 -> i64.
 def:  Pat <(i64 (extloadi16 ADDRriS11_2:$src1)),
-      (i64 (COMBINE_rr (TFRI 0), (LDrih ADDRriS11_2:$src1)))>;
+      (i64 (COMBINE_rr (TFRI 0), (LDrih ADDRriS11_2:$src1)))>,
+      Requires<[NoV4T]>;
+
+let AddedComplexity = 20 in
+def:  Pat <(i64 (extloadi16 (add (i32 IntRegs:$src1),
+                                  s11_1ExtPred:$offset))),
+      (i64 (COMBINE_rr (TFRI 0), (LDrih_indexed IntRegs:$src1,
+                                  s11_1ExtPred:$offset)))>,
+      Requires<[NoV4T]>;
 
 // Map from Rdd = zxtw(Rs) -> Rdd = combine(0, Rs).
 def : Pat<(i64 (zext (i32 IntRegs:$src1))),
-      (i64 (COMBINE_rr (TFRI 0), (i32 IntRegs:$src1)))>;
+      (i64 (COMBINE_rr (TFRI 0), (i32 IntRegs:$src1)))>,
+      Requires<[NoV4T]>;
 
 // Multiply 64-bit unsigned and use upper result.
 def : Pat <(mulhu (i64 DoubleRegs:$src1), (i64 DoubleRegs:$src2)),
@@ -3331,6 +2645,11 @@ def BR_JT : JRInst<(outs), (ins IntRegs:$src),
                    "jumpr $src",
                    [(HexagonBR_JT (i32 IntRegs:$src))]>;
 
+let isBranch=1, isIndirectBranch=1, isTerminator=1 in
+def BRIND : JRInst<(outs), (ins IntRegs:$src),
+                   "jumpr $src",
+                   [(brind (i32 IntRegs:$src))]>;
+
 def HexagonWrapperJT: SDNode<"HexagonISD::WrapperJT", SDTIntUnaryOp>;
 
 def : Pat<(HexagonWrapperJT tjumptable:$dst),
@@ -3444,32 +2763,3 @@ include "HexagonInstrInfoV5.td"
 //===----------------------------------------------------------------------===//
 // V5 Instructions -
 //===----------------------------------------------------------------------===//
-
-//===----------------------------------------------------------------------===//
-// Generate mapping table to relate non-predicate instructions with their
-// predicated formats - true and false.
-//
-
-def getPredOpcode : InstrMapping {
-  let FilterClass = "PredRel";
-  // Instructions with the same BaseOpcode and isNVStore values form a row.
-  let RowFields = ["BaseOpcode", "isNVStore", "PNewValue"];
-  // Instructions with the same predicate sense form a column.
-  let ColFields = ["PredSense"];
-  // The key column is the unpredicated instructions.
-  let KeyCol = [""];
-  // Value columns are PredSense=true and PredSense=false
-  let ValueCols = [["true"], ["false"]];
-}
-
-//===----------------------------------------------------------------------===//
-// Generate mapping table to relate predicated instructions with their .new
-// format.
-//
-def getPredNewOpcode : InstrMapping {
-  let FilterClass = "PredNewRel";
-  let RowFields = ["BaseOpcode", "PredSense", "isNVStore"];
-  let ColFields = ["PNewValue"];
-  let KeyCol = [""];
-  let ValueCols = [["new"]];
-}
diff --git a/lib/Target/Hexagon/HexagonInstrInfoV4.td b/lib/Target/Hexagon/HexagonInstrInfoV4.td
index 70448fc7af38..cd0e4758968c 100644
--- a/lib/Target/Hexagon/HexagonInstrInfoV4.td
+++ b/lib/Target/Hexagon/HexagonInstrInfoV4.td
@@ -12,10 +12,25 @@
 //===----------------------------------------------------------------------===//
 
 let neverHasSideEffects = 1 in
-def IMMEXT : Immext<(outs), (ins),
-                    "/* immext #... */",
-                    []>,
-             Requires<[HasV4T]>;
+class T_Immext<dag ins> :
+  EXTENDERInst<(outs), ins, "immext(#$imm)", []>,
+  Requires<[HasV4T]>;
+
+def IMMEXT_b : T_Immext<(ins brtarget:$imm)>;
+def IMMEXT_c : T_Immext<(ins calltarget:$imm)>;
+def IMMEXT_g : T_Immext<(ins globaladdress:$imm)>;
+def IMMEXT_i : T_Immext<(ins u26_6Imm:$imm)>;
+
+// Fold (add (CONST32 tglobaladdr:$addr) <offset>) into a global address.
+def FoldGlobalAddr : ComplexPattern<i32, 1, "foldGlobalAddress", [], []>;
+
+// Fold (add (CONST32_GP tglobaladdr:$addr) <offset>) into a global address.
+def FoldGlobalAddrGP : ComplexPattern<i32, 1, "foldGlobalAddressGP", [], []>;
+
+def NumUsesBelowThresCONST32 : PatFrag<(ops node:$addr),
+                                       (HexagonCONST32 node:$addr), [{
+  return hasNumUsesBelowThresGA(N->getOperand(0).getNode());
+}]>;
 
 // Hexagon V4 Architecture spec defines 8 instruction classes:
 // LD ST ALU32 XTYPE J JR MEMOP NV CR SYSTEM(system is not implemented in the
@@ -80,190 +95,63 @@ def IMMEXT : Immext<(outs), (ins),
 //===----------------------------------------------------------------------===//
 // ALU32 +
 //===----------------------------------------------------------------------===//
-
-// Shift halfword.
-
-let isPredicated = 1 in
-def ASLH_cPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1) $dst = aslh($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def ASLH_cNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1) $dst = aslh($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def ASLH_cdnPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1.new) $dst = aslh($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def ASLH_cdnNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1.new) $dst = aslh($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def ASRH_cPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1) $dst = asrh($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def ASRH_cNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1) $dst = asrh($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def ASRH_cdnPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1.new) $dst = asrh($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def ASRH_cdnNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1.new) $dst = asrh($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-// Sign extend.
-
-let isPredicated = 1 in
-def SXTB_cPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1) $dst = sxtb($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def SXTB_cNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1) $dst = sxtb($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def SXTB_cdnPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1.new) $dst = sxtb($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def SXTB_cdnNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1.new) $dst = sxtb($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-let isPredicated = 1 in
-def SXTH_cPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1) $dst = sxth($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def SXTH_cNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1) $dst = sxth($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def SXTH_cdnPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1.new) $dst = sxth($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicated = 1 in
-def SXTH_cdnNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1.new) $dst = sxth($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-// Zero exten.
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ZXTB_cPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1) $dst = zxtb($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ZXTB_cNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1) $dst = zxtb($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ZXTB_cdnPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1.new) $dst = zxtb($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ZXTB_cdnNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1.new) $dst = zxtb($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ZXTH_cPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1) $dst = zxth($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ZXTH_cNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1) $dst = zxth($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ZXTH_cdnPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if ($src1.new) $dst = zxth($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def ZXTH_cdnNotPt_V4 : ALU32_rr<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2),
-            "if (!$src1.new) $dst = zxth($src2)",
-            []>,
-            Requires<[HasV4T]>;
-
 // Generate frame index addresses.
-let neverHasSideEffects = 1, isReMaterializable = 1 in
+let neverHasSideEffects = 1, isReMaterializable = 1,
+isExtended = 1, opExtendable = 2, validSubTargets = HasV4SubT in
 def TFR_FI_immext_V4 : ALU32_ri<(outs IntRegs:$dst),
             (ins IntRegs:$src1, s32Imm:$offset),
             "$dst = add($src1, ##$offset)",
             []>,
             Requires<[HasV4T]>;
 
+// Rd=cmp.eq(Rs,#s8)
+let validSubTargets = HasV4SubT, isExtendable = 1, opExtendable = 2,
+isExtentSigned = 1, opExtentBits = 8 in
+def V4_A4_rcmpeqi : ALU32_ri<(outs IntRegs:$Rd),
+                    (ins IntRegs:$Rs, s8Ext:$s8),
+                    "$Rd = cmp.eq($Rs, #$s8)",
+                    [(set (i32 IntRegs:$Rd),
+                          (i32 (zext (i1 (seteq (i32 IntRegs:$Rs),
+                                                s8ExtPred:$s8)))))]>,
+                    Requires<[HasV4T]>;
+
+// Preserve the TSTBIT generation
+def : Pat <(i32 (zext (i1 (setne (i32 (and (i32 (shl 1, (i32 IntRegs:$src2))),
+                                           (i32 IntRegs:$src1))), 0)))),
+      (i32 (MUX_ii (i1 (TSTBIT_rr (i32 IntRegs:$src1), (i32 IntRegs:$src2))),
+                   1, 0))>;
+
+// Interfered with tstbit generation, above pattern preserves, see : tstbit.ll
+// Rd=cmp.ne(Rs,#s8)
+let validSubTargets = HasV4SubT, isExtendable = 1, opExtendable = 2,
+isExtentSigned = 1, opExtentBits = 8 in
+def V4_A4_rcmpneqi : ALU32_ri<(outs IntRegs:$Rd),
+                     (ins IntRegs:$Rs, s8Ext:$s8),
+                     "$Rd = !cmp.eq($Rs, #$s8)",
+                     [(set (i32 IntRegs:$Rd),
+                           (i32 (zext (i1 (setne (i32 IntRegs:$Rs),
+                                                 s8ExtPred:$s8)))))]>,
+                     Requires<[HasV4T]>;
+
+// Rd=cmp.eq(Rs,Rt)
+let validSubTargets = HasV4SubT in
+def V4_A4_rcmpeq : ALU32_ri<(outs IntRegs:$Rd),
+                   (ins IntRegs:$Rs, IntRegs:$Rt),
+                   "$Rd = cmp.eq($Rs, $Rt)",
+                   [(set (i32 IntRegs:$Rd),
+                         (i32 (zext (i1 (seteq (i32 IntRegs:$Rs),
+                                               IntRegs:$Rt)))))]>,
+                   Requires<[HasV4T]>;
+
+// Rd=cmp.ne(Rs,Rt)
+let validSubTargets = HasV4SubT in
+def V4_A4_rcmpneq : ALU32_ri<(outs IntRegs:$Rd),
+                    (ins IntRegs:$Rs, IntRegs:$Rt),
+                    "$Rd = !cmp.eq($Rs, $Rt)",
+                    [(set (i32 IntRegs:$Rd),
+                          (i32 (zext (i1 (setne (i32 IntRegs:$Rs),
+                                               IntRegs:$Rt)))))]>,
+                    Requires<[HasV4T]>;
 
 //===----------------------------------------------------------------------===//
 // ALU32 -
@@ -276,19 +164,44 @@ def TFR_FI_immext_V4 : ALU32_ri<(outs IntRegs:$dst),
 
 // Combine
 // Rdd=combine(Rs, #s8)
-let neverHasSideEffects = 1 in
-def COMBINE_ri_V4 : ALU32_ri<(outs DoubleRegs:$dst),
-            (ins IntRegs:$src1, s8Imm:$src2),
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8,
+    neverHasSideEffects = 1, validSubTargets = HasV4SubT in
+def COMBINE_rI_V4 : ALU32_ri<(outs DoubleRegs:$dst),
+            (ins IntRegs:$src1, s8Ext:$src2),
             "$dst = combine($src1, #$src2)",
             []>,
             Requires<[HasV4T]>;
+
 // Rdd=combine(#s8, Rs)
-let neverHasSideEffects = 1 in
-def COMBINE_ir_V4 : ALU32_ir<(outs DoubleRegs:$dst),
-            (ins s8Imm:$src1, IntRegs:$src2),
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 1, opExtentBits = 8,
+    neverHasSideEffects = 1, validSubTargets = HasV4SubT in
+def COMBINE_Ir_V4 : ALU32_ir<(outs DoubleRegs:$dst),
+            (ins s8Ext:$src1, IntRegs:$src2),
             "$dst = combine(#$src1, $src2)",
             []>,
             Requires<[HasV4T]>;
+
+def HexagonWrapperCombineRI_V4 :
+  SDNode<"HexagonISD::WrapperCombineRI_V4", SDTHexagonI64I32I32>;
+def HexagonWrapperCombineIR_V4 :
+  SDNode<"HexagonISD::WrapperCombineIR_V4", SDTHexagonI64I32I32>;
+
+def : Pat <(HexagonWrapperCombineRI_V4 IntRegs:$r, s8ExtPred:$i),
+           (COMBINE_rI_V4 IntRegs:$r, s8ExtPred:$i)>,
+          Requires<[HasV4T]>;
+
+def : Pat <(HexagonWrapperCombineIR_V4 s8ExtPred:$i, IntRegs:$r),
+           (COMBINE_Ir_V4 s8ExtPred:$i, IntRegs:$r)>,
+          Requires<[HasV4T]>;
+
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 6,
+    neverHasSideEffects = 1, validSubTargets = HasV4SubT in
+def COMBINE_iI_V4 : ALU32_ii<(outs DoubleRegs:$dst),
+            (ins s8Imm:$src1, u6Ext:$src2),
+            "$dst = combine(#$src1, #$src2)",
+            []>,
+            Requires<[HasV4T]>;
+
 //===----------------------------------------------------------------------===//
 // ALU32/PERM +
 //===----------------------------------------------------------------------===//
@@ -300,1436 +213,310 @@ def COMBINE_ir_V4 : ALU32_ir<(outs DoubleRegs:$dst),
 // These absolute set addressing mode instructions accept immediate as
 // an operand. We have duplicated these patterns to take global address.
 
-let neverHasSideEffects = 1 in
+let isExtended = 1, opExtendable = 2, neverHasSideEffects = 1,
+validSubTargets = HasV4SubT in {
 def LDrid_abs_setimm_V4 : LDInst2<(outs DoubleRegs:$dst1, IntRegs:$dst2),
-            (ins u6Imm:$addr),
-            "$dst1 = memd($dst2=#$addr)",
+            (ins u0AlwaysExt:$addr),
+            "$dst1 = memd($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
 
 // Rd=memb(Re=#U6)
-let neverHasSideEffects = 1 in
 def LDrib_abs_setimm_V4 : LDInst2<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins u6Imm:$addr),
-            "$dst1 = memb($dst2=#$addr)",
+            (ins u0AlwaysExt:$addr),
+            "$dst1 = memb($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
 
 // Rd=memh(Re=#U6)
-let neverHasSideEffects = 1 in
 def LDrih_abs_setimm_V4 : LDInst2<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins u6Imm:$addr),
-            "$dst1 = memh($dst2=#$addr)",
+            (ins u0AlwaysExt:$addr),
+            "$dst1 = memh($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
 
 // Rd=memub(Re=#U6)
-let neverHasSideEffects = 1 in
 def LDriub_abs_setimm_V4 : LDInst2<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins u6Imm:$addr),
-            "$dst1 = memub($dst2=#$addr)",
+            (ins u0AlwaysExt:$addr),
+            "$dst1 = memub($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
 
 // Rd=memuh(Re=#U6)
-let neverHasSideEffects = 1 in
 def LDriuh_abs_setimm_V4 : LDInst2<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins u6Imm:$addr),
-            "$dst1 = memuh($dst2=#$addr)",
+            (ins u0AlwaysExt:$addr),
+            "$dst1 = memuh($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
 
 // Rd=memw(Re=#U6)
-let neverHasSideEffects = 1 in
 def LDriw_abs_setimm_V4 : LDInst2<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins u6Imm:$addr),
-            "$dst1 = memw($dst2=#$addr)",
+            (ins u0AlwaysExt:$addr),
+            "$dst1 = memw($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
+}
 
 // Following patterns are defined for absolute set addressing mode
 // instruction which take global address as operand.
-let neverHasSideEffects = 1 in
+let isExtended = 1, opExtendable = 2, neverHasSideEffects = 1,
+validSubTargets = HasV4SubT in {
 def LDrid_abs_set_V4 : LDInst2<(outs DoubleRegs:$dst1, IntRegs:$dst2),
-            (ins globaladdress:$addr),
+            (ins globaladdressExt:$addr),
             "$dst1 = memd($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
 
 // Rd=memb(Re=#U6)
-let neverHasSideEffects = 1 in
 def LDrib_abs_set_V4 : LDInst2<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins globaladdress:$addr),
+            (ins globaladdressExt:$addr),
             "$dst1 = memb($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
 
 // Rd=memh(Re=#U6)
-let neverHasSideEffects = 1 in
 def LDrih_abs_set_V4 : LDInst2<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins globaladdress:$addr),
+            (ins globaladdressExt:$addr),
             "$dst1 = memh($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
 
 // Rd=memub(Re=#U6)
-let neverHasSideEffects = 1 in
 def LDriub_abs_set_V4 : LDInst2<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins globaladdress:$addr),
+            (ins globaladdressExt:$addr),
             "$dst1 = memub($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
 
 // Rd=memuh(Re=#U6)
-let neverHasSideEffects = 1 in
 def LDriuh_abs_set_V4 : LDInst2<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins globaladdress:$addr),
+            (ins globaladdressExt:$addr),
             "$dst1 = memuh($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
 
 // Rd=memw(Re=#U6)
-let neverHasSideEffects = 1 in
 def LDriw_abs_set_V4 : LDInst2<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins globaladdress:$addr),
+            (ins globaladdressExt:$addr),
             "$dst1 = memw($dst2=##$addr)",
             []>,
             Requires<[HasV4T]>;
+}
 
-// Load doubleword.
-//
-// Make sure that in post increment load, the first operand is always the post
-// increment operand.
-//
-// Rdd=memd(Rs+Rt<<#u2)
-// Special case pattern for indexed load without offset which is easier to
-// match. AddedComplexity of this pattern should be lower than base+offset load
-// and lower yet than the more generic version with offset/shift below
-// Similar approach is taken for all other base+index loads.
-let AddedComplexity = 10, isPredicable = 1 in
-def LDrid_indexed_V4 : LDInst<(outs DoubleRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2),
-                    "$dst=memd($src1+$src2<<#0)",
-                    [(set (i64 DoubleRegs:$dst),
-                          (i64 (load (add (i32 IntRegs:$src1),
-                                          (i32 IntRegs:$src2)))))]>,
-                    Requires<[HasV4T]>;
-
-let AddedComplexity = 40, isPredicable = 1 in
-def LDrid_indexed_shl_V4 : LDInst<(outs DoubleRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset),
-                    "$dst=memd($src1+$src2<<#$offset)",
-                    [(set (i64 DoubleRegs:$dst),
-                          (i64 (load (add (i32 IntRegs:$src1),
-                                          (shl (i32 IntRegs:$src2),
-                                               u2ImmPred:$offset)))))]>,
-                    Requires<[HasV4T]>;
-
-//// Load doubleword conditionally.
-// if ([!]Pv[.new]) Rd=memd(Rs+Rt<<#u2)
-// if (Pv) Rd=memd(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrid_indexed_cPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1) $dst=memd($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memd(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrid_indexed_cdnPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1.new) $dst=memd($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memd(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrid_indexed_cNotPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1) $dst=memd($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memd(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrid_indexed_cdnNotPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1.new) $dst=memd($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv) Rd=memd(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrid_indexed_shl_cPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1) $dst=memd($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memd(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrid_indexed_shl_cdnPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1.new) $dst=memd($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memd(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrid_indexed_shl_cNotPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1) $dst=memd($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memd(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrid_indexed_shl_cdnNotPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1.new) $dst=memd($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// Rdd=memd(Rt<<#u2+#U6)
-
-//// Load byte.
-// Rd=memb(Rs+Rt<<#u2)
-let AddedComplexity = 10, isPredicable = 1 in
-def LDrib_indexed_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2),
-                    "$dst=memb($src1+$src2<<#0)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (sextloadi8 (add (i32 IntRegs:$src1),
-                                                (i32 IntRegs:$src2)))))]>,
-                    Requires<[HasV4T]>;
-
-let AddedComplexity = 10, isPredicable = 1 in
-def LDriub_indexed_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2),
-                    "$dst=memub($src1+$src2<<#0)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (zextloadi8 (add (i32 IntRegs:$src1),
-                                                (i32 IntRegs:$src2)))))]>,
-                    Requires<[HasV4T]>;
-
-let AddedComplexity = 10, isPredicable = 1 in
-def LDriub_ae_indexed_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2),
-                    "$dst=memub($src1+$src2<<#0)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (extloadi8 (add (i32 IntRegs:$src1),
-                                               (i32 IntRegs:$src2)))))]>,
-                    Requires<[HasV4T]>;
-
-let AddedComplexity = 40, isPredicable = 1 in
-def LDrib_indexed_shl_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset),
-                    "$dst=memb($src1+$src2<<#$offset)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (sextloadi8 (add (i32 IntRegs:$src1),
-                                                (shl (i32 IntRegs:$src2),
-                                                     u2ImmPred:$offset)))))]>,
-                    Requires<[HasV4T]>;
-
-let AddedComplexity = 40, isPredicable = 1 in
-def LDriub_indexed_shl_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset),
-                    "$dst=memub($src1+$src2<<#$offset)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (zextloadi8 (add (i32 IntRegs:$src1),
-                                                (shl (i32 IntRegs:$src2),
-                                                     u2ImmPred:$offset)))))]>,
-                    Requires<[HasV4T]>;
-
-let AddedComplexity = 40, isPredicable = 1 in
-def LDriub_ae_indexed_shl_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset),
-                    "$dst=memub($src1+$src2<<#$offset)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (extloadi8 (add (i32 IntRegs:$src1),
-                                               (shl (i32 IntRegs:$src2),
-                                                    u2ImmPred:$offset)))))]>,
-                    Requires<[HasV4T]>;
-
-//// Load byte conditionally.
-// if ([!]Pv[.new]) Rd=memb(Rs+Rt<<#u2)
-// if (Pv) Rd=memb(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrib_indexed_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1) $dst=memb($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memb(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrib_indexed_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1.new) $dst=memb($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memb(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrib_indexed_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1) $dst=memb($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memb(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrib_indexed_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1.new) $dst=memb($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv) Rd=memb(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrib_indexed_shl_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1) $dst=memb($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memb(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrib_indexed_shl_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1.new) $dst=memb($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memb(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrib_indexed_shl_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1) $dst=memb($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memb(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrib_indexed_shl_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1.new) $dst=memb($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-//// Load unsigned byte conditionally.
-// if ([!]Pv[.new]) Rd=memub(Rs+Rt<<#u2)
-// if (Pv) Rd=memub(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriub_indexed_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1) $dst=memub($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memub(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriub_indexed_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1.new) $dst=memub($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memub(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriub_indexed_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1) $dst=memub($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memub(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriub_indexed_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1.new) $dst=memub($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv) Rd=memub(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriub_indexed_shl_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1) $dst=memub($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memub(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriub_indexed_shl_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1.new) $dst=memub($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memub(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriub_indexed_shl_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1) $dst=memub($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memub(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriub_indexed_shl_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1.new) $dst=memub($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// Rd=memb(Rt<<#u2+#U6)
-
-//// Load halfword
-// Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 10, isPredicable = 1 in
-def LDrih_indexed_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2),
-                    "$dst=memh($src1+$src2<<#0)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (sextloadi16 (add (i32 IntRegs:$src1),
-                                                 (i32 IntRegs:$src2)))))]>,
-                    Requires<[HasV4T]>;
-
-let AddedComplexity = 10, isPredicable = 1 in
-def LDriuh_indexed_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2),
-                    "$dst=memuh($src1+$src2<<#0)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (zextloadi16 (add (i32 IntRegs:$src1),
-                                                 (i32 IntRegs:$src2)))))]>,
-                    Requires<[HasV4T]>;
-
-let AddedComplexity = 10, isPredicable = 1 in
-def LDriuh_ae_indexed_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2),
-                    "$dst=memuh($src1+$src2<<#0)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (extloadi16 (add (i32 IntRegs:$src1),
-                                                (i32 IntRegs:$src2)))))]>,
-                    Requires<[HasV4T]>;
-
-// Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 40, isPredicable = 1 in
-def LDrih_indexed_shl_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset),
-                    "$dst=memh($src1+$src2<<#$offset)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (sextloadi16 (add (i32 IntRegs:$src1),
-                                                 (shl (i32 IntRegs:$src2),
-                                                      u2ImmPred:$offset)))))]>,
-                    Requires<[HasV4T]>;
-
-let AddedComplexity = 40, isPredicable = 1 in
-def LDriuh_indexed_shl_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset),
-                    "$dst=memuh($src1+$src2<<#$offset)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (zextloadi16 (add (i32 IntRegs:$src1),
-                                                 (shl (i32 IntRegs:$src2),
-                                                      u2ImmPred:$offset)))))]>,
-                    Requires<[HasV4T]>;
-
-let AddedComplexity = 40, isPredicable = 1 in
-def LDriuh_ae_indexed_shl_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset),
-                    "$dst=memuh($src1+$src2<<#$offset)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (extloadi16 (add (i32 IntRegs:$src1),
-                                                (shl (i32 IntRegs:$src2),
-                                                     u2ImmPred:$offset)))))]>,
-                    Requires<[HasV4T]>;
-
-//// Load halfword conditionally.
-// if ([!]Pv[.new]) Rd=memh(Rs+Rt<<#u2)
-// if (Pv) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrih_indexed_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1) $dst=memh($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrih_indexed_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1.new) $dst=memh($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrih_indexed_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1) $dst=memh($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDrih_indexed_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1.new) $dst=memh($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrih_indexed_shl_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1) $dst=memh($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrih_indexed_shl_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1.new) $dst=memh($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrih_indexed_shl_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1) $dst=memh($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDrih_indexed_shl_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1.new) $dst=memh($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-//// Load unsigned halfword conditionally.
-// if ([!]Pv[.new]) Rd=memuh(Rs+Rt<<#u2)
-// if (Pv) Rd=memuh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriuh_indexed_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1) $dst=memuh($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memuh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriuh_indexed_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1.new) $dst=memuh($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memuh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriuh_indexed_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1) $dst=memuh($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memuh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriuh_indexed_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1.new) $dst=memuh($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv) Rd=memuh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriuh_indexed_shl_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1) $dst=memuh($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memuh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriuh_indexed_shl_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1.new) $dst=memuh($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memuh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriuh_indexed_shl_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1) $dst=memuh($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memuh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriuh_indexed_shl_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1.new) $dst=memuh($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// Rd=memh(Rt<<#u2+#U6)
-
-//// Load word.
-// Load predicate: Fix for bug 5279.
-let neverHasSideEffects = 1 in
-def LDriw_pred_V4 : LDInst2<(outs PredRegs:$dst),
-            (ins MEMri:$addr),
-            "Error; should not emit",
-            []>,
-            Requires<[HasV4T]>;
-
-// Rd=memw(Re=#U6)
-
-// Rd=memw(Rs+Rt<<#u2)
-let AddedComplexity = 10, isPredicable = 1 in
-def LDriw_indexed_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2),
-                    "$dst=memw($src1+$src2<<#0)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (load (add (i32 IntRegs:$src1),
-                                          (i32 IntRegs:$src2)))))]>,
-                    Requires<[HasV4T]>;
-
-// Rd=memw(Rs+Rt<<#u2)
-let AddedComplexity = 40, isPredicable = 1 in
-def LDriw_indexed_shl_V4 : LDInst<(outs IntRegs:$dst),
-                    (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset),
-                    "$dst=memw($src1+$src2<<#$offset)",
-                    [(set (i32 IntRegs:$dst),
-                          (i32 (load (add (i32 IntRegs:$src1),
-                                          (shl (i32 IntRegs:$src2),
-                                               u2ImmPred:$offset)))))]>,
-                    Requires<[HasV4T]>;
-
-//// Load word conditionally.
-// if ([!]Pv[.new]) Rd=memw(Rs+Rt<<#u2)
-// if (Pv) Rd=memw(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriw_indexed_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1) $dst=memw($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriw_indexed_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if ($src1.new) $dst=memw($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriw_indexed_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1) $dst=memw($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 15, isPredicated = 1 in
-def LDriw_indexed_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3),
-                    "if (!$src1.new) $dst=memw($src2+$src3<<#0)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriw_indexed_shl_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1) $dst=memw($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (Pv.new) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriw_indexed_shl_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if ($src1.new) $dst=memw($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriw_indexed_shl_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1) $dst=memw($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// if (!Pv.new) Rd=memh(Rs+Rt<<#u2)
-let AddedComplexity = 45, isPredicated = 1 in
-def LDriw_indexed_shl_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-                    (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3,
-                         u2Imm:$offset),
-                    "if (!$src1.new) $dst=memw($src2+$src3<<#$offset)",
-                    []>,
-                    Requires<[HasV4T]>;
-
-// Rd=memw(Rt<<#u2+#U6)
-
-
-// Post-inc Load, Predicated, Dot new
-
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrid_cdnPt_V4 : LDInst2PI<(outs DoubleRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_3Imm:$src3),
-            "if ($src1.new) $dst1 = memd($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrid_cdnNotPt_V4 : LDInst2PI<(outs DoubleRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_3Imm:$src3),
-            "if (!$src1.new) $dst1 = memd($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrib_cdnPt_V4 : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_0Imm:$src3),
-            "if ($src1.new) $dst1 = memb($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrib_cdnNotPt_V4 : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_0Imm:$src3),
-            "if (!$src1.new) $dst1 = memb($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrih_cdnPt_V4 : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_1Imm:$src3),
-            "if ($src1.new) $dst1 = memh($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDrih_cdnNotPt_V4 : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_1Imm:$src3),
-            "if (!$src1.new) $dst1 = memh($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriub_cdnPt_V4 : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_0Imm:$src3),
-            "if ($src1.new) $dst1 = memub($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriub_cdnNotPt_V4 : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_0Imm:$src3),
-            "if (!$src1.new) $dst1 = memub($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriuh_cdnPt_V4 : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_1Imm:$src3),
-            "if ($src1.new) $dst1 = memuh($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriuh_cdnNotPt_V4 : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_1Imm:$src3),
-            "if (!$src1.new) $dst1 = memuh($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriw_cdnPt_V4 : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_2Imm:$src3),
-            "if ($src1.new) $dst1 = memw($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-let hasCtrlDep = 1, neverHasSideEffects = 1, isPredicated = 1 in
-def POST_LDriw_cdnNotPt_V4 : LDInst2PI<(outs IntRegs:$dst1, IntRegs:$dst2),
-            (ins PredRegs:$src1, IntRegs:$src2, s4_2Imm:$src3),
-            "if (!$src1.new) $dst1 = memw($src2++#$src3)",
-            [],
-            "$src2 = $dst2">,
-            Requires<[HasV4T]>;
-
-/// Load from global offset
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDrid_GP_V4 : LDInst2<(outs DoubleRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst=memd(#$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_GP_cPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1) $dst=memd(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_GP_cNotPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1) $dst=memd(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_GP_cdnPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1.new) $dst=memd(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrid_GP_cdnNotPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1.new) $dst=memd(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDrib_GP_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst=memb(#$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_GP_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1) $dst=memb(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_GP_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1) $dst=memb(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_GP_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1.new) $dst=memb(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrib_GP_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1.new) $dst=memb(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDriub_GP_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst=memub(#$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_GP_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1) $dst=memub(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_GP_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1) $dst=memub(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_GP_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1.new) $dst=memub(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriub_GP_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1.new) $dst=memub(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDrih_GP_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst=memh(#$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_GP_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1) $dst=memh(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_GP_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1) $dst=memh(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_GP_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1.new) $dst=memh(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDrih_GP_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1.new) $dst=memh(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDriuh_GP_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst=memuh(#$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_GP_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1) $dst=memuh(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_GP_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1) $dst=memuh(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_GP_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1.new) $dst=memuh(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriuh_GP_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1.new) $dst=memuh(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDriw_GP_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global, u16Imm:$offset),
-            "$dst=memw(#$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_GP_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1) $dst=memw(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_GP_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1) $dst=memw(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_GP_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if ($src1.new) $dst=memw(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDriw_GP_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset),
-            "if (!$src1.new) $dst=memw(##$global+$offset)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDd_GP_V4 : LDInst2<(outs DoubleRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst=memd(#$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) Rtt=memd(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDd_GP_cPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1) $dst=memd(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-// if (!Pv) Rtt=memd(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDd_GP_cNotPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1) $dst=memd(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) Rtt=memd(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDd_GP_cdnPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1.new) $dst=memd(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-// if (!Pv) Rtt=memd(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDd_GP_cdnNotPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1.new) $dst=memd(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDb_GP_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst=memb(#$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) Rt=memb(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDb_GP_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1) $dst=memb(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) Rt=memb(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDb_GP_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1) $dst=memb(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) Rt=memb(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDb_GP_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1.new) $dst=memb(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) Rt=memb(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDb_GP_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1.new) $dst=memb(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDub_GP_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst=memub(#$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) Rt=memub(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDub_GP_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1) $dst=memub(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-// if (!Pv) Rt=memub(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDub_GP_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1) $dst=memub(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) Rt=memub(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDub_GP_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1.new) $dst=memub(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-
-// if (!Pv) Rt=memub(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDub_GP_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1.new) $dst=memub(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDh_GP_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst=memh(#$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) Rt=memh(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDh_GP_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1) $dst=memh(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) Rt=memh(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDh_GP_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1) $dst=memh(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) Rt=memh(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDh_GP_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1.new) $dst=memh(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) Rt=memh(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDh_GP_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1.new) $dst=memh(##$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDuh_GP_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst=memuh(#$global)",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) Rt=memuh(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDuh_GP_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1) $dst=memuh(##$global)",
-            []>,
-            Requires<[HasV4T]>;
+// multiclass for load instructions with base + register offset
+// addressing mode
+multiclass ld_idxd_shl_pbase<string mnemonic, RegisterClass RC, bit isNot,
+                             bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : LDInst2<(outs RC:$dst),
+            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$offset),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#"$dst = "#mnemonic#"($src2+$src3<<#$offset)",
+            []>, Requires<[HasV4T]>;
+}
 
-// if (!Pv) Rt=memuh(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDuh_GP_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1) $dst=memuh(##$global)",
-            []>,
-            Requires<[HasV4T]>;
+multiclass ld_idxd_shl_pred<string mnemonic, RegisterClass RC, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ld_idxd_shl_pbase<mnemonic, RC, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : ld_idxd_shl_pbase<mnemonic, RC, PredNot, 1>;
+  }
+}
 
-// if (Pv) Rt=memuh(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDuh_GP_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1.new) $dst=memuh(##$global)",
-            []>,
-            Requires<[HasV4T]>;
+let neverHasSideEffects  = 1 in
+multiclass ld_idxd_shl<string mnemonic, string CextOp, RegisterClass RC> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
+    let isPredicable = 1 in
+    def NAME#_V4 : LDInst2<(outs RC:$dst),
+            (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$offset),
+            "$dst = "#mnemonic#"($src1+$src2<<#$offset)",
+            []>, Requires<[HasV4T]>;
+
+    let isPredicated = 1 in {
+      defm Pt_V4 : ld_idxd_shl_pred<mnemonic, RC, 0 >;
+      defm NotPt_V4 : ld_idxd_shl_pred<mnemonic, RC, 1>;
+    }
+  }
+}
 
-// if (!Pv) Rt=memuh(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDuh_GP_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1.new) $dst=memuh(##$global)",
-            []>,
-            Requires<[HasV4T]>;
+let addrMode = BaseRegOffset in {
+  defm LDrib_indexed_shl: ld_idxd_shl<"memb", "LDrib", IntRegs>, AddrModeRel;
+  defm LDriub_indexed_shl: ld_idxd_shl<"memub", "LDriub", IntRegs>, AddrModeRel;
+  defm LDrih_indexed_shl: ld_idxd_shl<"memh", "LDrih", IntRegs>, AddrModeRel;
+  defm LDriuh_indexed_shl: ld_idxd_shl<"memuh", "LDriuh", IntRegs>, AddrModeRel;
+  defm LDriw_indexed_shl: ld_idxd_shl<"memw", "LDriw", IntRegs>, AddrModeRel;
+  defm LDrid_indexed_shl: ld_idxd_shl<"memd", "LDrid", DoubleRegs>, AddrModeRel;
+}
 
-let isPredicable = 1, neverHasSideEffects = 1 in
-def LDw_GP_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$global),
-            "$dst=memw(#$global)",
-            []>,
+// 'def pats' for load instructions with base + register offset and non-zero
+// immediate value. Immediate value is used to left-shift the second
+// register operand.
+let AddedComplexity = 40 in {
+def : Pat <(i32 (sextloadi8 (add IntRegs:$src1,
+                                 (shl IntRegs:$src2, u2ImmPred:$offset)))),
+           (LDrib_indexed_shl_V4 IntRegs:$src1,
+            IntRegs:$src2, u2ImmPred:$offset)>,
             Requires<[HasV4T]>;
 
-// if (Pv) Rt=memw(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDw_GP_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1) $dst=memw(##$global)",
-            []>,
+def : Pat <(i32 (zextloadi8 (add IntRegs:$src1,
+                                 (shl IntRegs:$src2, u2ImmPred:$offset)))),
+           (LDriub_indexed_shl_V4 IntRegs:$src1,
+            IntRegs:$src2, u2ImmPred:$offset)>,
             Requires<[HasV4T]>;
 
-
-// if (!Pv) Rt=memw(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDw_GP_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1) $dst=memw(##$global)",
-            []>,
+def : Pat <(i32 (extloadi8 (add IntRegs:$src1,
+                                (shl IntRegs:$src2, u2ImmPred:$offset)))),
+           (LDriub_indexed_shl_V4 IntRegs:$src1,
+            IntRegs:$src2, u2ImmPred:$offset)>,
             Requires<[HasV4T]>;
 
-// if (Pv) Rt=memw(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDw_GP_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if ($src1.new) $dst=memw(##$global)",
-            []>,
+def : Pat <(i32 (sextloadi16 (add IntRegs:$src1,
+                                  (shl IntRegs:$src2, u2ImmPred:$offset)))),
+           (LDrih_indexed_shl_V4 IntRegs:$src1,
+            IntRegs:$src2, u2ImmPred:$offset)>,
             Requires<[HasV4T]>;
 
-
-// if (!Pv) Rt=memw(##global)
-let neverHasSideEffects = 1, isPredicated = 1 in
-def LDw_GP_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$global),
-            "if (!$src1.new) $dst=memw(##$global)",
-            []>,
+def : Pat <(i32 (zextloadi16 (add IntRegs:$src1,
+                                  (shl IntRegs:$src2, u2ImmPred:$offset)))),
+           (LDriuh_indexed_shl_V4 IntRegs:$src1,
+            IntRegs:$src2, u2ImmPred:$offset)>,
             Requires<[HasV4T]>;
 
-
-
-def : Pat <(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)),
-           (i64 (LDd_GP_V4 tglobaladdr:$global))>,
+def : Pat <(i32 (extloadi16 (add IntRegs:$src1,
+                                 (shl IntRegs:$src2, u2ImmPred:$offset)))),
+           (LDriuh_indexed_shl_V4 IntRegs:$src1,
+            IntRegs:$src2, u2ImmPred:$offset)>,
             Requires<[HasV4T]>;
 
-def : Pat <(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)),
-           (i32 (LDw_GP_V4 tglobaladdr:$global))>,
+def : Pat <(i32 (load (add IntRegs:$src1,
+                           (shl IntRegs:$src2, u2ImmPred:$offset)))),
+           (LDriw_indexed_shl_V4 IntRegs:$src1,
+            IntRegs:$src2, u2ImmPred:$offset)>,
             Requires<[HasV4T]>;
 
-def : Pat <(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)),
-           (i32 (LDuh_GP_V4 tglobaladdr:$global))>,
-            Requires<[HasV4T]>;
-
-def : Pat <(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)),
-           (i32 (LDub_GP_V4 tglobaladdr:$global))>,
-            Requires<[HasV4T]>;
-
-// Map from load(globaladdress) -> memw(#foo + 0)
-let AddedComplexity = 100 in
-def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))),
-           (i64 (LDd_GP_V4 tglobaladdr:$global))>,
-            Requires<[HasV4T]>;
-
-// Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd
-let AddedComplexity = 100 in
-def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))),
-           (i1 (TFR_PdRs (i32 (LDb_GP_V4 tglobaladdr:$global))))>,
-           Requires<[HasV4T]>;
-
-// When the Interprocedural Global Variable optimizer realizes that a certain
-// global variable takes only two constant values, it shrinks the global to
-// a boolean. Catch those loads here in the following 3 patterns.
-let AddedComplexity = 100 in
-def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
-           (i32 (LDb_GP_V4 tglobaladdr:$global))>,
-            Requires<[HasV4T]>;
-
-let AddedComplexity = 100 in
-def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
-           (i32 (LDb_GP_V4 tglobaladdr:$global))>,
+def : Pat <(i64 (load (add IntRegs:$src1,
+                           (shl IntRegs:$src2, u2ImmPred:$offset)))),
+           (LDrid_indexed_shl_V4 IntRegs:$src1,
+            IntRegs:$src2, u2ImmPred:$offset)>,
             Requires<[HasV4T]>;
+}
 
-// Map from load(globaladdress) -> memb(#foo)
-let AddedComplexity = 100 in
-def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
-           (i32 (LDb_GP_V4 tglobaladdr:$global))>,
-            Requires<[HasV4T]>;
 
-// Map from load(globaladdress) -> memb(#foo)
-let AddedComplexity = 100 in
-def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
-           (i32 (LDb_GP_V4 tglobaladdr:$global))>,
+// 'def pats' for load instruction base + register offset and
+// zero immediate value.
+let AddedComplexity = 10 in {
+def : Pat <(i64 (load (add IntRegs:$src1, IntRegs:$src2))),
+           (LDrid_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
             Requires<[HasV4T]>;
 
-let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
-           (i32 (LDub_GP_V4 tglobaladdr:$global))>,
+def : Pat <(i32 (sextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
+           (LDrib_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
             Requires<[HasV4T]>;
 
-// Map from load(globaladdress) -> memub(#foo)
-let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
-           (i32 (LDub_GP_V4 tglobaladdr:$global))>,
+def : Pat <(i32 (zextloadi8 (add IntRegs:$src1, IntRegs:$src2))),
+           (LDriub_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
             Requires<[HasV4T]>;
 
-// Map from load(globaladdress) -> memh(#foo)
-let AddedComplexity = 100 in
-def : Pat <(i32 (extloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
-           (i32 (LDh_GP_V4 tglobaladdr:$global))>,
+def : Pat <(i32 (extloadi8 (add IntRegs:$src1, IntRegs:$src2))),
+           (LDriub_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
             Requires<[HasV4T]>;
 
-// Map from load(globaladdress) -> memh(#foo)
-let AddedComplexity = 100 in
-def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
-           (i32 (LDh_GP_V4 tglobaladdr:$global))>,
+def : Pat <(i32 (sextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
+           (LDrih_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
             Requires<[HasV4T]>;
 
-// Map from load(globaladdress) -> memuh(#foo)
-let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
-           (i32 (LDuh_GP_V4 tglobaladdr:$global))>,
+def : Pat <(i32 (zextloadi16 (add IntRegs:$src1, IntRegs:$src2))),
+           (LDriuh_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
             Requires<[HasV4T]>;
 
-// Map from load(globaladdress) -> memw(#foo)
-let AddedComplexity = 100 in
-def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))),
-           (i32 (LDw_GP_V4 tglobaladdr:$global))>,
+def : Pat <(i32 (extloadi16 (add IntRegs:$src1, IntRegs:$src2))),
+           (LDriuh_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
             Requires<[HasV4T]>;
 
-def : Pat <(atomic_load_64 (add (HexagonCONST32_GP tglobaladdr:$global),
-                                u16ImmPred:$offset)),
-           (i64 (LDrid_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-           Requires<[HasV4T]>;
-
-def : Pat <(atomic_load_32 (add (HexagonCONST32_GP tglobaladdr:$global),
-                                u16ImmPred:$offset)),
-           (i32 (LDriw_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-            Requires<[HasV4T]>;
-
-def : Pat <(atomic_load_16 (add (HexagonCONST32_GP tglobaladdr:$global),
-                                u16ImmPred:$offset)),
-           (i32 (LDriuh_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
+def : Pat <(i32 (load (add IntRegs:$src1, IntRegs:$src2))),
+           (LDriw_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2, 0)>,
             Requires<[HasV4T]>;
+}
 
-def : Pat <(atomic_load_8 (add (HexagonCONST32_GP tglobaladdr:$global),
-                               u16ImmPred:$offset)),
-           (i32 (LDriub_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-           Requires<[HasV4T]>;
-
-// Map from load(globaladdress + x) -> memd(#foo + x)
-let AddedComplexity = 100 in
-def : Pat <(i64 (load (add (HexagonCONST32_GP tglobaladdr:$global),
-                           u16ImmPred:$offset))),
-           (i64 (LDrid_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-           Requires<[HasV4T]>;
-
-// Map from load(globaladdress + x) -> memb(#foo + x)
-let AddedComplexity = 100 in
-def : Pat <(i32 (extloadi8 (add (HexagonCONST32_GP tglobaladdr:$global),
-                           u16ImmPred:$offset))),
-           (i32 (LDrib_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-           Requires<[HasV4T]>;
-
-// Map from load(globaladdress + x) -> memb(#foo + x)
-let AddedComplexity = 100 in
-def : Pat <(i32 (sextloadi8 (add (HexagonCONST32_GP tglobaladdr:$global),
-                            u16ImmPred:$offset))),
-           (i32 (LDrib_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-           Requires<[HasV4T]>;
-
-// Map from load(globaladdress + x) -> memub(#foo + x)
-let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi8 (add (HexagonCONST32_GP tglobaladdr:$global),
-                            u16ImmPred:$offset))),
-           (i32 (LDriub_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-           Requires<[HasV4T]>;
-
-// Map from load(globaladdress + x) -> memuh(#foo + x)
-let AddedComplexity = 100 in
-def : Pat <(i32 (extloadi16 (add (HexagonCONST32_GP tglobaladdr:$global),
-                            u16ImmPred:$offset))),
-           (i32 (LDrih_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-            Requires<[HasV4T]>;
+// zext i1->i64
+def : Pat <(i64 (zext (i1 PredRegs:$src1))),
+      (i64 (COMBINE_Ir_V4 0, (MUX_ii (i1 PredRegs:$src1), 1, 0)))>,
+      Requires<[HasV4T]>;
+
+// zext i32->i64
+def : Pat <(i64 (zext (i32 IntRegs:$src1))),
+      (i64 (COMBINE_Ir_V4 0, (i32 IntRegs:$src1)))>,
+      Requires<[HasV4T]>;
+// zext i8->i64
+def:  Pat <(i64 (zextloadi8 ADDRriS11_0:$src1)),
+      (i64 (COMBINE_Ir_V4 0, (LDriub ADDRriS11_0:$src1)))>,
+      Requires<[HasV4T]>;
+
+let AddedComplexity = 20 in
+def:  Pat <(i64 (zextloadi8 (add (i32 IntRegs:$src1),
+                                s11_0ExtPred:$offset))),
+      (i64 (COMBINE_Ir_V4 0, (LDriub_indexed IntRegs:$src1,
+                                  s11_0ExtPred:$offset)))>,
+      Requires<[HasV4T]>;
+
+// zext i1->i64
+def:  Pat <(i64 (zextloadi1 ADDRriS11_0:$src1)),
+      (i64 (COMBINE_Ir_V4 0, (LDriub ADDRriS11_0:$src1)))>,
+      Requires<[HasV4T]>;
+
+let AddedComplexity = 20 in
+def:  Pat <(i64 (zextloadi1 (add (i32 IntRegs:$src1),
+                                s11_0ExtPred:$offset))),
+      (i64 (COMBINE_Ir_V4 0, (LDriub_indexed IntRegs:$src1,
+                                  s11_0ExtPred:$offset)))>,
+      Requires<[HasV4T]>;
+
+// zext i16->i64
+def:  Pat <(i64 (zextloadi16 ADDRriS11_1:$src1)),
+      (i64 (COMBINE_Ir_V4 0, (LDriuh ADDRriS11_1:$src1)))>,
+      Requires<[HasV4T]>;
+
+let AddedComplexity = 20 in
+def:  Pat <(i64 (zextloadi16 (add (i32 IntRegs:$src1),
+                                  s11_1ExtPred:$offset))),
+      (i64 (COMBINE_Ir_V4 0, (LDriuh_indexed IntRegs:$src1,
+                                  s11_1ExtPred:$offset)))>,
+      Requires<[HasV4T]>;
+
+// anyext i16->i64
+def:  Pat <(i64 (extloadi16 ADDRriS11_2:$src1)),
+      (i64 (COMBINE_Ir_V4 0, (LDrih ADDRriS11_2:$src1)))>,
+      Requires<[HasV4T]>;
+
+let AddedComplexity = 20 in
+def:  Pat <(i64 (extloadi16 (add (i32 IntRegs:$src1),
+                                  s11_1ExtPred:$offset))),
+      (i64 (COMBINE_Ir_V4 0, (LDrih_indexed IntRegs:$src1,
+                                  s11_1ExtPred:$offset)))>,
+      Requires<[HasV4T]>;
+
+// zext i32->i64
+def:  Pat <(i64 (zextloadi32 ADDRriS11_2:$src1)),
+      (i64 (COMBINE_Ir_V4 0, (LDriw ADDRriS11_2:$src1)))>,
+      Requires<[HasV4T]>;
 
-// Map from load(globaladdress + x) -> memh(#foo + x)
 let AddedComplexity = 100 in
-def : Pat <(i32 (sextloadi16 (add (HexagonCONST32_GP tglobaladdr:$global),
-                             u16ImmPred:$offset))),
-           (i32 (LDrih_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-           Requires<[HasV4T]>;
+def:  Pat <(i64 (zextloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
+      (i64 (COMBINE_Ir_V4 0, (LDriw_indexed IntRegs:$src1,
+                                  s11_2ExtPred:$offset)))>,
+      Requires<[HasV4T]>;
 
+// anyext i32->i64
+def:  Pat <(i64 (extloadi32 ADDRriS11_2:$src1)),
+      (i64 (COMBINE_Ir_V4 0, (LDriw ADDRriS11_2:$src1)))>,
+      Requires<[HasV4T]>;
 
-// Map from load(globaladdress + x) -> memuh(#foo + x)
 let AddedComplexity = 100 in
-def : Pat <(i32 (zextloadi16 (add (HexagonCONST32_GP tglobaladdr:$global),
-                             u16ImmPred:$offset))),
-           (i32 (LDriuh_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-            Requires<[HasV4T]>;
+def:  Pat <(i64 (extloadi32 (i32 (add IntRegs:$src1, s11_2ExtPred:$offset)))),
+      (i64 (COMBINE_Ir_V4 0, (LDriw_indexed IntRegs:$src1,
+                                  s11_2ExtPred:$offset)))>,
+      Requires<[HasV4T]>;
 
-// Map from load(globaladdress + x) -> memw(#foo + x)
-let AddedComplexity = 100 in
-def : Pat <(i32 (load (add (HexagonCONST32_GP tglobaladdr:$global),
-                      u16ImmPred:$offset))),
-           (i32 (LDriw_GP_V4 tglobaladdr:$global, u16ImmPred:$offset))>,
-            Requires<[HasV4T]>;
 
 
 //===----------------------------------------------------------------------===//
@@ -1747,80 +534,192 @@ def : Pat <(i32 (load (add (HexagonCONST32_GP tglobaladdr:$global),
 ///    last operand.
 ///
 
-// memd(Re=#U6)=Rtt
+// memd(Re=#U)=Rtt
+let isExtended = 1, opExtendable = 2, validSubTargets = HasV4SubT in {
 def STrid_abs_setimm_V4 : STInst2<(outs IntRegs:$dst1),
-            (ins DoubleRegs:$src1, u6Imm:$src2),
-            "memd($dst1=#$src2) = $src1",
+            (ins DoubleRegs:$src1, u0AlwaysExt:$src2),
+            "memd($dst1=##$src2) = $src1",
             []>,
             Requires<[HasV4T]>;
 
-// memb(Re=#U6)=Rs
+// memb(Re=#U)=Rs
 def STrib_abs_setimm_V4 : STInst2<(outs IntRegs:$dst1),
-            (ins IntRegs:$src1, u6Imm:$src2),
-            "memb($dst1=#$src2) = $src1",
+            (ins IntRegs:$src1, u0AlwaysExt:$src2),
+            "memb($dst1=##$src2) = $src1",
             []>,
             Requires<[HasV4T]>;
 
-// memh(Re=#U6)=Rs
+// memh(Re=#U)=Rs
 def STrih_abs_setimm_V4 : STInst2<(outs IntRegs:$dst1),
-            (ins IntRegs:$src1, u6Imm:$src2),
-            "memh($dst1=#$src2) = $src1",
+            (ins IntRegs:$src1, u0AlwaysExt:$src2),
+            "memh($dst1=##$src2) = $src1",
             []>,
             Requires<[HasV4T]>;
 
-// memw(Re=#U6)=Rs
+// memw(Re=#U)=Rs
 def STriw_abs_setimm_V4 : STInst2<(outs IntRegs:$dst1),
-            (ins IntRegs:$src1, u6Imm:$src2),
-            "memw($dst1=#$src2) = $src1",
+            (ins IntRegs:$src1, u0AlwaysExt:$src2),
+            "memw($dst1=##$src2) = $src1",
             []>,
             Requires<[HasV4T]>;
+}
 
-// memd(Re=#U6)=Rtt
+// memd(Re=#U)=Rtt
+let isExtended = 1, opExtendable = 2, validSubTargets = HasV4SubT in {
 def STrid_abs_set_V4 : STInst2<(outs IntRegs:$dst1),
-            (ins DoubleRegs:$src1, globaladdress:$src2),
+            (ins DoubleRegs:$src1, globaladdressExt:$src2),
             "memd($dst1=##$src2) = $src1",
             []>,
             Requires<[HasV4T]>;
 
-// memb(Re=#U6)=Rs
+// memb(Re=#U)=Rs
 def STrib_abs_set_V4 : STInst2<(outs IntRegs:$dst1),
-            (ins IntRegs:$src1, globaladdress:$src2),
+            (ins IntRegs:$src1, globaladdressExt:$src2),
             "memb($dst1=##$src2) = $src1",
             []>,
             Requires<[HasV4T]>;
 
-// memh(Re=#U6)=Rs
+// memh(Re=#U)=Rs
 def STrih_abs_set_V4 : STInst2<(outs IntRegs:$dst1),
-            (ins IntRegs:$src1, globaladdress:$src2),
+            (ins IntRegs:$src1, globaladdressExt:$src2),
             "memh($dst1=##$src2) = $src1",
             []>,
             Requires<[HasV4T]>;
 
-// memw(Re=#U6)=Rs
+// memw(Re=#U)=Rs
 def STriw_abs_set_V4 : STInst2<(outs IntRegs:$dst1),
-            (ins IntRegs:$src1, globaladdress:$src2),
+            (ins IntRegs:$src1, globaladdressExt:$src2),
             "memw($dst1=##$src2) = $src1",
             []>,
             Requires<[HasV4T]>;
+}
 
-// memd(Rs+Ru<<#u2)=Rtt
-let AddedComplexity = 10, isPredicable = 1 in
-def STrid_indexed_shl_V4 : STInst<(outs),
-            (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, DoubleRegs:$src4),
-            "memd($src1+$src2<<#$src3) = $src4",
-            [(store (i64 DoubleRegs:$src4),
-                    (add (i32 IntRegs:$src1),
-                         (shl (i32 IntRegs:$src2), u2ImmPred:$src3)))]>,
+// multiclass for store instructions with base + register offset addressing
+// mode
+multiclass ST_Idxd_shl_Pbase<string mnemonic, RegisterClass RC, bit isNot,
+                             bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : STInst2<(outs),
+            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
+                 RC:$src5),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"($src2+$src3<<#$src4) = $src5",
+            []>,
             Requires<[HasV4T]>;
+}
+
+multiclass ST_Idxd_shl_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ST_Idxd_shl_Pbase<mnemonic, RC, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : ST_Idxd_shl_Pbase<mnemonic, RC, PredNot, 1>;
+  }
+}
+
+let isNVStorable = 1 in
+multiclass ST_Idxd_shl<string mnemonic, string CextOp, RegisterClass RC> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
+    let isPredicable = 1 in
+    def NAME#_V4 : STInst2<(outs),
+            (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, RC:$src4),
+            mnemonic#"($src1+$src2<<#$src3) = $src4",
+            []>,
+            Requires<[HasV4T]>;
+
+    let isPredicated = 1 in {
+      defm Pt_V4 : ST_Idxd_shl_Pred<mnemonic, RC, 0 >;
+      defm NotPt_V4 : ST_Idxd_shl_Pred<mnemonic, RC, 1>;
+    }
+  }
+}
+
+// multiclass for new-value store instructions with base + register offset
+// addressing mode.
+multiclass ST_Idxd_shl_Pbase_nv<string mnemonic, RegisterClass RC, bit isNot,
+                             bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME#_nv_V4 : NVInst_V4<(outs),
+            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
+                 RC:$src5),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"($src2+$src3<<#$src4) = $src5.new",
+            []>,
+            Requires<[HasV4T]>;
+}
+
+multiclass ST_Idxd_shl_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ST_Idxd_shl_Pbase_nv<mnemonic, RC, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : ST_Idxd_shl_Pbase_nv<mnemonic, RC, PredNot, 1>;
+  }
+}
+
+let mayStore = 1, isNVStore = 1 in
+multiclass ST_Idxd_shl_nv<string mnemonic, string CextOp, RegisterClass RC> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed_shl in {
+    let isPredicable = 1 in
+    def NAME#_nv_V4 : NVInst_V4<(outs),
+            (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, RC:$src4),
+            mnemonic#"($src1+$src2<<#$src3) = $src4.new",
+            []>,
+            Requires<[HasV4T]>;
+
+    let isPredicated = 1 in {
+      defm Pt : ST_Idxd_shl_Pred_nv<mnemonic, RC, 0 >;
+      defm NotPt : ST_Idxd_shl_Pred_nv<mnemonic, RC, 1>;
+    }
+  }
+}
+
+let addrMode = BaseRegOffset, neverHasSideEffects = 1,
+validSubTargets = HasV4SubT in {
+  defm STrib_indexed_shl: ST_Idxd_shl<"memb", "STrib", IntRegs>,
+                          ST_Idxd_shl_nv<"memb", "STrib", IntRegs>, AddrModeRel;
+
+  defm STrih_indexed_shl: ST_Idxd_shl<"memh", "STrih", IntRegs>,
+                          ST_Idxd_shl_nv<"memh", "STrih", IntRegs>, AddrModeRel;
+
+  defm STriw_indexed_shl: ST_Idxd_shl<"memw", "STriw", IntRegs>,
+                          ST_Idxd_shl_nv<"memw", "STriw", IntRegs>, AddrModeRel;
+
+  let isNVStorable = 0 in
+  defm STrid_indexed_shl: ST_Idxd_shl<"memd", "STrid", DoubleRegs>, AddrModeRel;
+}
+
+let Predicates = [HasV4T], AddedComplexity = 10 in {
+def : Pat<(truncstorei8 (i32 IntRegs:$src4),
+                       (add IntRegs:$src1, (shl IntRegs:$src2,
+                                                u2ImmPred:$src3))),
+          (STrib_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2,
+                                u2ImmPred:$src3, IntRegs:$src4)>;
+
+def : Pat<(truncstorei16 (i32 IntRegs:$src4),
+                        (add IntRegs:$src1, (shl IntRegs:$src2,
+                                                 u2ImmPred:$src3))),
+          (STrih_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2,
+                                u2ImmPred:$src3, IntRegs:$src4)>;
+
+def : Pat<(store (i32 IntRegs:$src4),
+                 (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
+          (STriw_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2,
+                                u2ImmPred:$src3, IntRegs:$src4)>;
+
+def : Pat<(store (i64 DoubleRegs:$src4),
+                (add IntRegs:$src1, (shl IntRegs:$src2, u2ImmPred:$src3))),
+          (STrid_indexed_shl_V4 IntRegs:$src1, IntRegs:$src2,
+                                u2ImmPred:$src3, DoubleRegs:$src4)>;
+}
 
 // memd(Ru<<#u2+#U6)=Rtt
-let AddedComplexity = 10 in
+let isExtended = 1, opExtendable = 2, AddedComplexity = 10,
+validSubTargets = HasV4SubT in
 def STrid_shl_V4 : STInst<(outs),
-            (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, DoubleRegs:$src4),
+            (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, DoubleRegs:$src4),
             "memd($src1<<#$src2+#$src3) = $src4",
             [(store (i64 DoubleRegs:$src4),
                     (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2),
-                         u6ImmPred:$src3))]>,
+                         u0AlwaysExtPred:$src3))]>,
             Requires<[HasV4T]>;
 
 // memd(Rx++#s4:3)=Rtt
@@ -1834,143 +733,81 @@ def STrid_shl_V4 : STInst<(outs),
 // if ([!]Pv[.new]) memd(#u6)=Rtt
 // TODO: needs to be implemented.
 
-// if ([!]Pv[.new]) memd(Rs+#u6:3)=Rtt
-// if (Pv) memd(Rs+#u6:3)=Rtt
-// if (Pv.new) memd(Rs+#u6:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, DoubleRegs:$src2),
-            "if ($src1.new) memd($addr) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memd(Rs+#u6:3)=Rtt
-// if (!Pv.new) memd(Rs+#u6:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, DoubleRegs:$src2),
-            "if (!$src1.new) memd($addr) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memd(Rs+#u6:3)=Rtt
-// if (Pv.new) memd(Rs+#u6:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_indexed_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_3Imm:$src3,
-                 DoubleRegs:$src4),
-            "if ($src1.new) memd($src2+#$src3) = $src4",
+//===----------------------------------------------------------------------===//
+// multiclass for store instructions with base + immediate offset
+// addressing mode and immediate stored value.
+// mem[bhw](Rx++#s4:3)=#s8
+// if ([!]Pv[.new]) mem[bhw](Rx++#s4:3)=#s6
+//===----------------------------------------------------------------------===//
+multiclass ST_Imm_Pbase<string mnemonic, Operand OffsetOp, bit isNot,
+                        bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : STInst2<(outs),
+            (ins PredRegs:$src1, IntRegs:$src2, OffsetOp:$src3, s6Ext:$src4),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"($src2+#$src3) = #$src4",
             []>,
             Requires<[HasV4T]>;
+}
 
-// if (!Pv) memd(Rs+#u6:3)=Rtt
-// if (!Pv.new) memd(Rs+#u6:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_indexed_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_3Imm:$src3,
-                 DoubleRegs:$src4),
-            "if (!$src1.new) memd($src2+#$src3) = $src4",
-            []>,
-            Requires<[HasV4T]>;
+multiclass ST_Imm_Pred<string mnemonic, Operand OffsetOp, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ST_Imm_Pbase<mnemonic, OffsetOp, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : ST_Imm_Pbase<mnemonic, OffsetOp, PredNot, 1>;
+  }
+}
 
-// if ([!]Pv[.new]) memd(Rs+Ru<<#u2)=Rtt
-// if (Pv) memd(Rs+Ru<<#u2)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_indexed_shl_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 DoubleRegs:$src5),
-            "if ($src1) memd($src2+$src3<<#$src4) = $src5",
+let isExtendable = 1, isExtentSigned = 1, neverHasSideEffects = 1 in
+multiclass ST_Imm<string mnemonic, string CextOp, Operand OffsetOp> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_imm in {
+    let opExtendable = 2, opExtentBits = 8, isPredicable = 1 in
+    def NAME#_V4 : STInst2<(outs),
+            (ins IntRegs:$src1, OffsetOp:$src2, s8Ext:$src3),
+            mnemonic#"($src1+#$src2) = #$src3",
             []>,
             Requires<[HasV4T]>;
 
-// if (Pv.new) memd(Rs+Ru<<#u2)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_indexed_shl_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 DoubleRegs:$src5),
-            "if ($src1.new) memd($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-// if (!Pv) memd(Rs+Ru<<#u2)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_indexed_shl_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 DoubleRegs:$src5),
-            "if (!$src1) memd($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-// if (!Pv.new) memd(Rs+Ru<<#u2)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrid_indexed_shl_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 DoubleRegs:$src5),
-            "if (!$src1.new) memd($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
+    let opExtendable = 3, opExtentBits = 6, isPredicated = 1 in {
+      defm Pt_V4 : ST_Imm_Pred<mnemonic, OffsetOp, 0>;
+      defm NotPt_V4 : ST_Imm_Pred<mnemonic, OffsetOp, 1 >;
+    }
+  }
+}
 
-// if ([!]Pv[.new]) memd(Rx++#s4:3)=Rtt
-// if (Pv) memd(Rx++#s4:3)=Rtt
-// if (Pv.new) memd(Rx++#s4:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def POST_STdri_cdnPt_V4 : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, DoubleRegs:$src2, IntRegs:$src3,
-                 s4_3Imm:$offset),
-            "if ($src1.new) memd($src3++#$offset) = $src2",
-            [],
-            "$src3 = $dst">,
-            Requires<[HasV4T]>;
+let addrMode = BaseImmOffset, InputType = "imm",
+    validSubTargets = HasV4SubT in {
+  defm STrib_imm : ST_Imm<"memb", "STrib", u6_0Imm>, ImmRegRel, PredNewRel;
+  defm STrih_imm : ST_Imm<"memh", "STrih", u6_1Imm>, ImmRegRel, PredNewRel;
+  defm STriw_imm : ST_Imm<"memw", "STriw", u6_2Imm>, ImmRegRel, PredNewRel;
+}
 
-// if (!Pv) memd(Rx++#s4:3)=Rtt
-// if (!Pv.new) memd(Rx++#s4:3)=Rtt
-let AddedComplexity = 10, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def POST_STdri_cdnNotPt_V4 : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, DoubleRegs:$src2, IntRegs:$src3,
-                 s4_3Imm:$offset),
-            "if (!$src1.new) memd($src3++#$offset) = $src2",
-            [],
-            "$src3 = $dst">,
-            Requires<[HasV4T]>;
+let Predicates = [HasV4T], AddedComplexity = 10 in {
+def: Pat<(truncstorei8 s8ExtPred:$src3, (add IntRegs:$src1, u6_0ImmPred:$src2)),
+            (STrib_imm_V4 IntRegs:$src1, u6_0ImmPred:$src2, s8ExtPred:$src3)>;
 
+def: Pat<(truncstorei16 s8ExtPred:$src3, (add IntRegs:$src1,
+                                              u6_1ImmPred:$src2)),
+            (STrih_imm_V4 IntRegs:$src1, u6_1ImmPred:$src2, s8ExtPred:$src3)>;
 
-// Store byte.
-// memb(Rs+#u6:0)=#S8
-let AddedComplexity = 10, isPredicable = 1 in
-def STrib_imm_V4 : STInst<(outs),
-            (ins IntRegs:$src1, u6_0Imm:$src2, s8Imm:$src3),
-            "memb($src1+#$src2) = #$src3",
-            [(truncstorei8 s8ImmPred:$src3, (add (i32 IntRegs:$src1),
-                                                 u6_0ImmPred:$src2))]>,
-            Requires<[HasV4T]>;
+def: Pat<(store s8ExtPred:$src3, (add IntRegs:$src1, u6_2ImmPred:$src2)),
+            (STriw_imm_V4 IntRegs:$src1, u6_2ImmPred:$src2, s8ExtPred:$src3)>;
+}
 
-// memb(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10, isPredicable = 1 in
-def STrib_indexed_shl_V4 : STInst<(outs),
-            (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, IntRegs:$src4),
-            "memb($src1+$src2<<#$src3) = $src4",
-            [(truncstorei8 (i32 IntRegs:$src4),
-                           (add (i32 IntRegs:$src1),
-                                (shl (i32 IntRegs:$src2),
-                                          u2ImmPred:$src3)))]>,
-            Requires<[HasV4T]>;
+let AddedComplexity = 6 in
+def : Pat <(truncstorei8 s8ExtPred:$src2, (i32 IntRegs:$src1)),
+           (STrib_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>,
+           Requires<[HasV4T]>;
 
 // memb(Ru<<#u2+#U6)=Rt
-let AddedComplexity = 10 in
+let isExtended = 1, opExtendable = 2, AddedComplexity = 10, isNVStorable = 1,
+validSubTargets = HasV4SubT in
 def STrib_shl_V4 : STInst<(outs),
-            (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, IntRegs:$src4),
+            (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
             "memb($src1<<#$src2+#$src3) = $src4",
             [(truncstorei8 (i32 IntRegs:$src4),
                            (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2),
-                                u6ImmPred:$src3))]>,
+                                u0AlwaysExtPred:$src3))]>,
             Requires<[HasV4T]>;
 
 // memb(Rx++#s4:0:circ(Mu))=Rt
@@ -1980,185 +817,28 @@ def STrib_shl_V4 : STInst<(outs),
 // memb(gp+#u16:0)=Rt
 
 
-// Store byte conditionally.
-// if ([!]Pv[.new]) memb(#u6)=Rt
-// if ([!]Pv[.new]) memb(Rs+#u6:0)=#S6
-// if (Pv) memb(Rs+#u6:0)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_imm_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, s6Imm:$src4),
-            "if ($src1) memb($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memb(Rs+#u6:0)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_imm_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, s6Imm:$src4),
-            "if ($src1.new) memb($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memb(Rs+#u6:0)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_imm_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, s6Imm:$src4),
-            "if (!$src1) memb($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memb(Rs+#u6:0)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_imm_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, s6Imm:$src4),
-            "if (!$src1.new) memb($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if ([!]Pv[.new]) memb(Rs+#u6:0)=Rt
-// if (Pv) memb(Rs+#u6:0)=Rt
-// if (Pv.new) memb(Rs+#u6:0)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1.new) memb($addr) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memb(Rs+#u6:0)=Rt
-// if (!Pv.new) memb(Rs+#u6:0)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1.new) memb($addr) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memb(Rs+#u6:0)=Rt
-// if (!Pv) memb(Rs+#u6:0)=Rt
-// if (Pv.new) memb(Rs+#u6:0)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_indexed_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4),
-            "if ($src1.new) memb($src2+#$src3) = $src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memb(Rs+#u6:0)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_indexed_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4),
-            "if (!$src1.new) memb($src2+#$src3) = $src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if ([!]Pv[.new]) memb(Rs+Ru<<#u2)=Rt
-// if (Pv) memb(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STrib_indexed_shl_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1) memb($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memb(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STrib_indexed_shl_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1.new) memb($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memb(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STrib_indexed_shl_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1) memb($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memb(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STrib_indexed_shl_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1.new) memb($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if ([!]Pv[.new]) memb(Rx++#s4:0)=Rt
-// if (Pv) memb(Rx++#s4:0)=Rt
-// if (Pv.new) memb(Rx++#s4:0)=Rt
-let hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STbri_cdnPt_V4 : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset),
-            "if ($src1.new) memb($src3++#$offset) = $src2",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memb(Rx++#s4:0)=Rt
-// if (!Pv.new) memb(Rx++#s4:0)=Rt
-let hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STbri_cdnNotPt_V4 : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset),
-            "if (!$src1.new) memb($src3++#$offset) = $src2",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-
 // Store halfword.
 // TODO: needs to be implemented
 // memh(Re=#U6)=Rt.H
 // memh(Rs+#s11:1)=Rt.H
-// memh(Rs+#u6:1)=#S8
-let AddedComplexity = 10, isPredicable = 1 in
-def STrih_imm_V4 : STInst<(outs),
-            (ins IntRegs:$src1, u6_1Imm:$src2, s8Imm:$src3),
-            "memh($src1+#$src2) = #$src3",
-            [(truncstorei16 s8ImmPred:$src3, (add (i32 IntRegs:$src1),
-                                                  u6_1ImmPred:$src2))]>,
-            Requires<[HasV4T]>;
+let AddedComplexity = 6 in
+def : Pat <(truncstorei16 s8ExtPred:$src2, (i32 IntRegs:$src1)),
+           (STrih_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>,
+           Requires<[HasV4T]>;
 
 // memh(Rs+Ru<<#u2)=Rt.H
 // TODO: needs to be implemented.
 
-// memh(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10, isPredicable = 1 in
-def STrih_indexed_shl_V4 : STInst<(outs),
-            (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, IntRegs:$src4),
-            "memh($src1+$src2<<#$src3) = $src4",
-            [(truncstorei16 (i32 IntRegs:$src4),
-                            (add (i32 IntRegs:$src1),
-                                 (shl (i32 IntRegs:$src2),
-                                      u2ImmPred:$src3)))]>,
-            Requires<[HasV4T]>;
-
 // memh(Ru<<#u2+#U6)=Rt.H
 // memh(Ru<<#u2+#U6)=Rt
-let AddedComplexity = 10 in
+let isExtended = 1, opExtendable = 2, AddedComplexity = 10, isNVStorable = 1,
+validSubTargets = HasV4SubT in
 def STrih_shl_V4 : STInst<(outs),
-            (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, IntRegs:$src4),
+            (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
             "memh($src1<<#$src2+#$src3) = $src4",
             [(truncstorei16 (i32 IntRegs:$src4),
                             (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2),
-                                 u6ImmPred:$src3))]>,
+                                 u0AlwaysExtPred:$src3))]>,
             Requires<[HasV4T]>;
 
 // memh(Rx++#s4:1:circ(Mu))=Rt.H
@@ -2173,152 +853,13 @@ def STrih_shl_V4 : STInst<(outs),
 // if ([!]Pv[.new]) memh(#u6)=Rt.H
 // if ([!]Pv[.new]) memh(#u6)=Rt
 
-// if ([!]Pv[.new]) memh(Rs+#u6:1)=#S6
-// if (Pv) memh(Rs+#u6:1)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_imm_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, s6Imm:$src4),
-            "if ($src1) memh($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memh(Rs+#u6:1)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_imm_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, s6Imm:$src4),
-            "if ($src1.new) memh($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memh(Rs+#u6:1)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_imm_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, s6Imm:$src4),
-            "if (!$src1) memh($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memh(Rs+#u6:1)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_imm_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, s6Imm:$src4),
-            "if (!$src1.new) memh($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
 
 // if ([!]Pv[.new]) memh(Rs+#u6:1)=Rt.H
 // TODO: needs to be implemented.
 
-// if ([!]Pv[.new]) memh(Rs+#u6:1)=Rt
-// if (Pv) memh(Rs+#u6:1)=Rt
-// if (Pv.new) memh(Rs+#u6:1)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1.new) memh($addr) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memh(Rs+#u6:1)=Rt
-// if (!Pv.new) memh(Rs+#u6:1)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1.new) memh($addr) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memh(Rs+#u6:1)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_indexed_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4),
-            "if ($src1.new) memh($src2+#$src3) = $src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memh(Rs+#u6:1)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_indexed_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4),
-            "if (!$src1.new) memh($src2+#$src3) = $src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if ([!]Pv[.new]) memh(Rs+Ru<<#u2)=Rt.H
-// if ([!]Pv[.new]) memh(Rs+Ru<<#u2)=Rt
-// if (Pv) memh(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STrih_indexed_shl_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1) memh($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memh(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STrih_indexed_shl_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1.new) memh($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memh(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STrih_indexed_shl_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1) memh($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memh(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STrih_indexed_shl_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1.new) memh($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
 // if ([!]Pv[.new]) memh(Rx++#s4:1)=Rt.H
 // TODO: Needs to be implemented.
 
-// if ([!]Pv[.new]) memh(Rx++#s4:1)=Rt
-// if (Pv) memh(Rx++#s4:1)=Rt
-// if (Pv.new) memh(Rx++#s4:1)=Rt
-let hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_SThri_cdnPt_V4 : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset),
-            "if ($src1.new) memh($src3++#$offset) = $src2",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memh(Rx++#s4:1)=Rt
-// if (!Pv.new) memh(Rx++#s4:1)=Rt
-let hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_SThri_cdnNotPt_V4 : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset),
-            "if (!$src1.new) memh($src3++#$offset) = $src2",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-
 // Store word.
 // memw(Re=#U6)=Rt
 // TODO: Needs to be implemented.
@@ -2331,34 +872,20 @@ def STriw_pred_V4 : STInst2<(outs),
             []>,
             Requires<[HasV4T]>;
 
-
-// memw(Rs+#u6:2)=#S8
-let AddedComplexity = 10, isPredicable = 1 in
-def STriw_imm_V4 : STInst<(outs),
-            (ins IntRegs:$src1, u6_2Imm:$src2, s8Imm:$src3),
-            "memw($src1+#$src2) = #$src3",
-            [(store s8ImmPred:$src3, (add (i32 IntRegs:$src1),
-                                          u6_2ImmPred:$src2))]>,
-            Requires<[HasV4T]>;
-
-// memw(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10, isPredicable = 1 in
-def STriw_indexed_shl_V4 : STInst<(outs),
-            (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, IntRegs:$src4),
-            "memw($src1+$src2<<#$src3) = $src4",
-            [(store (i32 IntRegs:$src4), (add (i32 IntRegs:$src1),
-                                    (shl (i32 IntRegs:$src2),
-                                         u2ImmPred:$src3)))]>,
-            Requires<[HasV4T]>;
+let AddedComplexity = 6 in
+def : Pat <(store s8ExtPred:$src2, (i32 IntRegs:$src1)),
+           (STriw_imm_V4 IntRegs:$src1, 0, s8ExtPred:$src2)>,
+           Requires<[HasV4T]>;
 
 // memw(Ru<<#u2+#U6)=Rt
-let AddedComplexity = 10 in
+let isExtended = 1, opExtendable = 2, AddedComplexity = 10, isNVStorable = 1,
+validSubTargets = HasV4SubT in
 def STriw_shl_V4 : STInst<(outs),
-            (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, IntRegs:$src4),
+            (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
             "memw($src1<<#$src2+#$src3) = $src4",
             [(store (i32 IntRegs:$src4),
                     (add (shl (i32 IntRegs:$src1), u2ImmPred:$src2),
-                              u6ImmPred:$src3))]>,
+                              u0AlwaysExtPred:$src3))]>,
             Requires<[HasV4T]>;
 
 // memw(Rx++#s4:2)=Rt
@@ -2366,592 +893,6 @@ def STriw_shl_V4 : STInst<(outs),
 // memw(Rx++I:circ(Mu))=Rt
 // memw(Rx++Mu)=Rt
 // memw(Rx++Mu:brev)=Rt
-// memw(gp+#u16:2)=Rt
-
-
-// Store word conditionally.
-
-// if ([!]Pv[.new]) memw(Rs+#u6:2)=#S6
-// if (Pv) memw(Rs+#u6:2)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_imm_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, s6Imm:$src4),
-            "if ($src1) memw($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memw(Rs+#u6:2)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_imm_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, s6Imm:$src4),
-            "if ($src1.new) memw($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memw(Rs+#u6:2)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_imm_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, s6Imm:$src4),
-            "if (!$src1) memw($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memw(Rs+#u6:2)=#S6
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_imm_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, s6Imm:$src4),
-            "if (!$src1.new) memw($src2+#$src3) = #$src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if ([!]Pv[.new]) memw(Rs+#u6:2)=Rt
-// if (Pv) memw(Rs+#u6:2)=Rt
-// if (Pv.new) memw(Rs+#u6:2)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1.new) memw($addr) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memw(Rs+#u6:2)=Rt
-// if (!Pv.new) memw(Rs+#u6:2)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1.new) memw($addr) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memw(Rs+#u6:2)=Rt
-// if (!Pv) memw(Rs+#u6:2)=Rt
-// if (Pv.new) memw(Rs+#u6:2)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_indexed_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4),
-            "if ($src1.new) memw($src2+#$src3) = $src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memw(Rs+#u6:2)=Rt
-let neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_indexed_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4),
-            "if (!$src1.new) memw($src2+#$src3) = $src4",
-            []>,
-            Requires<[HasV4T]>;
-
-// if ([!]Pv[.new]) memw(Rs+Ru<<#u2)=Rt
-// if (Pv) memw(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STriw_indexed_shl_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1) memw($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memw(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STriw_indexed_shl_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1.new) memw($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memw(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STriw_indexed_shl_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1) memw($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memw(Rs+Ru<<#u2)=Rt
-let AddedComplexity = 10,
-    isPredicated = 1 in
-def STriw_indexed_shl_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1.new) memw($src2+$src3<<#$src4) = $src5",
-            []>,
-            Requires<[HasV4T]>;
-
-// if ([!]Pv[.new]) memw(Rx++#s4:2)=Rt
-// if (Pv) memw(Rx++#s4:2)=Rt
-// if (Pv.new) memw(Rx++#s4:2)=Rt
-let hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STwri_cdnPt_V4 : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset),
-            "if ($src1.new) memw($src3++#$offset) = $src2",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memw(Rx++#s4:2)=Rt
-// if (!Pv.new) memw(Rx++#s4:2)=Rt
-let hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STwri_cdnNotPt_V4 : STInst2PI<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset),
-            "if (!$src1.new) memw($src3++#$offset) = $src2",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-
-/// store to global address
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def STrid_GP_V4 : STInst2<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, DoubleRegs:$src),
-            "memd(#$global+$offset) = $src",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrid_GP_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        DoubleRegs:$src2),
-            "if ($src1) memd(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrid_GP_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        DoubleRegs:$src2),
-            "if (!$src1) memd(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrid_GP_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        DoubleRegs:$src2),
-            "if ($src1.new) memd(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrid_GP_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        DoubleRegs:$src2),
-            "if (!$src1.new) memd(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def STrib_GP_V4 : STInst2<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src),
-            "memb(#$global+$offset) = $src",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrib_GP_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1) memb(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrib_GP_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1) memb(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrib_GP_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1.new) memb(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrib_GP_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1.new) memb(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def STrih_GP_V4 : STInst2<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src),
-            "memh(#$global+$offset) = $src",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrih_GP_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1) memh(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrih_GP_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1) memh(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrih_GP_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1.new) memh(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STrih_GP_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1.new) memh(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let isPredicable = 1, neverHasSideEffects = 1 in
-def STriw_GP_V4 : STInst2<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src),
-            "memw(#$global+$offset) = $src",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STriw_GP_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1) memw(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STriw_GP_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1) memw(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STriw_GP_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1.new) memw(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STriw_GP_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1.new) memw(##$global+$offset) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// memd(#global)=Rtt
-let isPredicable = 1, neverHasSideEffects = 1 in
-def STd_GP_V4 : STInst2<(outs),
-            (ins globaladdress:$global, DoubleRegs:$src),
-            "memd(#$global) = $src",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memd(##global) = Rtt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STd_GP_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, DoubleRegs:$src2),
-            "if ($src1) memd(##$global) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memd(##global) = Rtt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STd_GP_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, DoubleRegs:$src2),
-            "if (!$src1) memd(##$global) = $src2",
-            []>,
-              Requires<[HasV4T]>;
-
-// if (Pv) memd(##global) = Rtt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STd_GP_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, DoubleRegs:$src2),
-            "if ($src1.new) memd(##$global) = $src2",
-            []>,
-              Requires<[HasV4T]>;
-
-// if (!Pv) memd(##global) = Rtt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STd_GP_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, DoubleRegs:$src2),
-            "if (!$src1.new) memd(##$global) = $src2",
-            []>,
-            Requires<[HasV4T]>;
-
-// memb(#global)=Rt
-let isPredicable = 1, neverHasSideEffects = 1 in
-def STb_GP_V4 : STInst2<(outs),
-            (ins globaladdress:$global, IntRegs:$src),
-            "memb(#$global) = $src",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memb(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STb_GP_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1) memb(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// if (!Pv) memb(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STb_GP_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1) memb(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// if (Pv) memb(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STb_GP_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1.new) memb(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// if (!Pv) memb(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STb_GP_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1.new) memb(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// memh(#global)=Rt
-let isPredicable = 1, neverHasSideEffects = 1 in
-def STh_GP_V4 : STInst2<(outs),
-            (ins globaladdress:$global, IntRegs:$src),
-            "memh(#$global) = $src",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memh(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STh_GP_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1) memh(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// if (!Pv) memh(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STh_GP_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1) memh(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// if (Pv) memh(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STh_GP_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1.new) memh(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// if (!Pv) memh(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STh_GP_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1.new) memh(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// memw(#global)=Rt
-let isPredicable = 1, neverHasSideEffects = 1 in
-def STw_GP_V4 : STInst2<(outs),
-            (ins globaladdress:$global, IntRegs:$src),
-            "memw(#$global) = $src",
-              []>,
-              Requires<[HasV4T]>;
-
-// if (Pv) memw(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STw_GP_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1) memw(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// if (!Pv) memw(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STw_GP_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1) memw(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// if (Pv) memw(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STw_GP_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1.new) memw(##$global) = $src2",
-              []>,
-              Requires<[HasV4T]>;
-
-// if (!Pv) memw(##global) = Rt
-let neverHasSideEffects = 1, isPredicated = 1 in
-def STw_GP_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1.new) memw(##$global) = $src2",
-            []>,
-              Requires<[HasV4T]>;
-
-// 64 bit atomic store
-def : Pat <(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global),
-                            (i64 DoubleRegs:$src1)),
-           (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
-           Requires<[HasV4T]>;
-
-// Map from store(globaladdress) -> memd(#foo)
-let AddedComplexity = 100 in
-def : Pat <(store (i64 DoubleRegs:$src1),
-                  (HexagonCONST32_GP tglobaladdr:$global)),
-           (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
-           Requires<[HasV4T]>;
-
-// 8 bit atomic store
-def : Pat < (atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global),
-                            (i32 IntRegs:$src1)),
-            (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>,
-              Requires<[HasV4T]>;
-
-// Map from store(globaladdress) -> memb(#foo)
-let AddedComplexity = 100 in
-def : Pat<(truncstorei8 (i32 IntRegs:$src1),
-          (HexagonCONST32_GP tglobaladdr:$global)),
-          (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-// Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1"
-//       to "r0 = 1; memw(#foo) = r0"
-let AddedComplexity = 100 in
-def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)),
-          (STb_GP_V4 tglobaladdr:$global, (TFRI 1))>,
-          Requires<[HasV4T]>;
-
-def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global),
-                           (i32 IntRegs:$src1)),
-          (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-// Map from store(globaladdress) -> memh(#foo)
-let AddedComplexity = 100 in
-def : Pat<(truncstorei16 (i32 IntRegs:$src1),
-                         (HexagonCONST32_GP tglobaladdr:$global)),
-          (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-// 32 bit atomic store
-def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global),
-                           (i32 IntRegs:$src1)),
-          (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-// Map from store(globaladdress) -> memw(#foo)
-let AddedComplexity = 100 in
-def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)),
-          (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-def : Pat<(atomic_store_64 (add (HexagonCONST32_GP tglobaladdr:$global),
-                                u16ImmPred:$offset),
-                           (i64 DoubleRegs:$src1)),
-          (STrid_GP_V4 tglobaladdr:$global, u16ImmPred:$offset,
-                                            (i64 DoubleRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-def : Pat<(atomic_store_32 (add (HexagonCONST32_GP tglobaladdr:$global),
-                                u16ImmPred:$offset),
-                           (i32 IntRegs:$src1)),
-          (STriw_GP_V4 tglobaladdr:$global, u16ImmPred:$offset,
-                                            (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-def : Pat<(atomic_store_16 (add (HexagonCONST32_GP tglobaladdr:$global),
-                                u16ImmPred:$offset),
-                           (i32 IntRegs:$src1)),
-          (STrih_GP_V4 tglobaladdr:$global, u16ImmPred:$offset,
-                                            (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-def : Pat<(atomic_store_8 (add (HexagonCONST32_GP tglobaladdr:$global),
-                               u16ImmPred:$offset),
-                          (i32 IntRegs:$src1)),
-          (STrib_GP_V4 tglobaladdr:$global, u16ImmPred:$offset,
-                                            (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-// Map from store(globaladdress + x) -> memd(#foo + x)
-let AddedComplexity = 100 in
-def : Pat<(store (i64 DoubleRegs:$src1),
-                    (add (HexagonCONST32_GP tglobaladdr:$global),
-                                        u16ImmPred:$offset)),
-          (STrid_GP_V4 tglobaladdr:$global, u16ImmPred:$offset,
-                                            (i64 DoubleRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-// Map from store(globaladdress + x) -> memb(#foo + x)
-let AddedComplexity = 100 in
-def : Pat<(truncstorei8 (i32 IntRegs:$src1),
-                        (add (HexagonCONST32_GP tglobaladdr:$global),
-                             u16ImmPred:$offset)),
-          (STrib_GP_V4 tglobaladdr:$global, u16ImmPred:$offset,
-                                            (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-// Map from store(globaladdress + x) -> memh(#foo + x)
-let AddedComplexity = 100 in
-def : Pat<(truncstorei16 (i32 IntRegs:$src1),
-                         (add (HexagonCONST32_GP tglobaladdr:$global),
-                              u16ImmPred:$offset)),
-          (STrih_GP_V4 tglobaladdr:$global, u16ImmPred:$offset,
-                                            (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-// Map from store(globaladdress + x) -> memw(#foo + x)
-let AddedComplexity = 100 in
-def : Pat<(store (i32 IntRegs:$src1),
-                 (add (HexagonCONST32_GP tglobaladdr:$global),
-                                u16ImmPred:$offset)),
-          (STriw_GP_V4 tglobaladdr:$global, u16ImmPred:$offset,
-                                            (i32 IntRegs:$src1))>,
-          Requires<[HasV4T]>;
-
-
 
 //===----------------------------------------------------------------------===
 // ST -
@@ -2962,853 +903,202 @@ def : Pat<(store (i32 IntRegs:$src1),
 // NV/ST +
 //===----------------------------------------------------------------------===//
 
-// Store new-value byte.
-
-// memb(Re=#U6)=Nt.new
-// memb(Rs+#s11:0)=Nt.new
-let mayStore = 1, isPredicable = 1 in
-def STrib_nv_V4 : NVInst_V4<(outs), (ins MEMri:$addr, IntRegs:$src1),
-            "memb($addr) = $src1.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, isPredicable = 1 in
-def STrib_indexed_nv_V4 : NVInst_V4<(outs),
-            (ins IntRegs:$src1, s11_0Imm:$src2, IntRegs:$src3),
-            "memb($src1+#$src2) = $src3.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// memb(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10, isPredicable = 1 in
-def STrib_indexed_shl_nv_V4 : NVInst_V4<(outs),
-            (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, IntRegs:$src4),
-            "memb($src1+$src2<<#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// memb(Ru<<#u2+#U6)=Nt.new
-let mayStore = 1, AddedComplexity = 10 in
-def STrib_shl_nv_V4 : NVInst_V4<(outs),
-            (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, IntRegs:$src4),
-            "memb($src1<<#$src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// memb(Rx++#s4:0)=Nt.new
-let mayStore = 1, hasCtrlDep = 1, isPredicable = 1  in
-def POST_STbri_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2, s4_0Imm:$offset),
-            "memb($src2++#$offset) = $src1.new",
-            [],
-            "$src2 = $dst">,
-            Requires<[HasV4T]>;
-
-// memb(Rx++#s4:0:circ(Mu))=Nt.new
-// memb(Rx++I:circ(Mu))=Nt.new
-// memb(Rx++Mu)=Nt.new
-// memb(Rx++Mu:brev)=Nt.new
-
-// memb(gp+#u16:0)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1 in
-def STrib_GP_nv_V4 : NVInst_V4<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src),
-            "memb(#$global+$offset) = $src.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// memb(#global)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1 in
-def STb_GP_nv_V4 : NVInst_V4<(outs),
-            (ins globaladdress:$global, IntRegs:$src),
-            "memb(#$global) = $src.new",
+// multiclass for new-value store instructions with base + immediate offset.
+//
+multiclass ST_Idxd_Pbase_nv<string mnemonic, RegisterClass RC,
+                            Operand predImmOp, bit isNot, bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME#_nv_V4 : NVInst_V4<(outs),
+            (ins PredRegs:$src1, IntRegs:$src2, predImmOp:$src3, RC: $src4),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"($src2+#$src3) = $src4.new",
             []>,
             Requires<[HasV4T]>;
+}
 
-// Store new-value byte conditionally.
-// if ([!]Pv[.new]) memb(#u6)=Nt.new
-// if (Pv) memb(Rs+#u6:0)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1) memb($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
+multiclass ST_Idxd_Pred_nv<string mnemonic, RegisterClass RC, Operand predImmOp,
+                           bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ST_Idxd_Pbase_nv<mnemonic, RC, predImmOp, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : ST_Idxd_Pbase_nv<mnemonic, RC, predImmOp, PredNot, 1>;
+  }
+}
 
-// if (Pv.new) memb(Rs+#u6:0)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1.new) memb($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
+let mayStore = 1, isNVStore = 1, neverHasSideEffects = 1, isExtendable = 1 in
+multiclass ST_Idxd_nv<string mnemonic, string CextOp, RegisterClass RC,
+                   Operand ImmOp, Operand predImmOp, bits<5> ImmBits,
+                   bits<5> PredImmBits> {
 
-// if (!Pv) memb(Rs+#u6:0)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1) memb($addr) = $src2.new",
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_indexed in {
+    let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits,
+    isPredicable = 1 in
+    def NAME#_nv_V4 : NVInst_V4<(outs),
+            (ins IntRegs:$src1, ImmOp:$src2, RC:$src3),
+            mnemonic#"($src1+#$src2) = $src3.new",
             []>,
             Requires<[HasV4T]>;
 
-// if (!Pv.new) memb(Rs+#u6:0)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1.new) memb($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
+    let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits,
+    isPredicated = 1 in {
+      defm Pt : ST_Idxd_Pred_nv<mnemonic, RC, predImmOp, 0>;
+      defm NotPt : ST_Idxd_Pred_nv<mnemonic, RC, predImmOp, 1>;
+    }
+  }
+}
 
-// if (Pv) memb(Rs+#u6:0)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_indexed_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4),
-            "if ($src1) memb($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
+let addrMode = BaseImmOffset, validSubTargets = HasV4SubT in {
+  defm STrib_indexed: ST_Idxd_nv<"memb", "STrib", IntRegs, s11_0Ext,
+                                 u6_0Ext, 11, 6>, AddrModeRel;
+  defm STrih_indexed: ST_Idxd_nv<"memh", "STrih", IntRegs, s11_1Ext,
+                                 u6_1Ext, 12, 7>, AddrModeRel;
+  defm STriw_indexed: ST_Idxd_nv<"memw", "STriw", IntRegs, s11_2Ext,
+                                 u6_2Ext, 13, 8>, AddrModeRel;
+}
 
-// if (Pv.new) memb(Rs+#u6:0)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_indexed_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4),
-            "if ($src1.new) memb($src2+#$src3) = $src4.new",
+// multiclass for new-value store instructions with base + immediate offset.
+// and MEMri operand.
+multiclass ST_MEMri_Pbase_nv<string mnemonic, RegisterClass RC, bit isNot,
+                          bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME#_nv_V4 : NVInst_V4<(outs),
+            (ins PredRegs:$src1, MEMri:$addr, RC: $src2),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"($addr) = $src2.new",
             []>,
             Requires<[HasV4T]>;
+}
 
-// if (!Pv) memb(Rs+#u6:0)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_indexed_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4),
-            "if (!$src1) memb($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
+multiclass ST_MEMri_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ST_MEMri_Pbase_nv<mnemonic, RC, PredNot, 0>;
 
-// if (!Pv.new) memb(Rs+#u6:0)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrib_indexed_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_0Imm:$src3, IntRegs:$src4),
-            "if (!$src1.new) memb($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
+    // Predicate new
+    defm _cdn#NAME : ST_MEMri_Pbase_nv<mnemonic, RC, PredNot, 1>;
+  }
+}
 
+let mayStore = 1, isNVStore = 1, isExtendable = 1, neverHasSideEffects = 1 in
+multiclass ST_MEMri_nv<string mnemonic, string CextOp, RegisterClass RC,
+                    bits<5> ImmBits, bits<5> PredImmBits> {
 
-// if ([!]Pv[.new]) memb(Rs+Ru<<#u2)=Nt.new
-// if (Pv) memb(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STrib_indexed_shl_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1) memb($src2+$src3<<#$src4) = $src5.new",
+  let CextOpcode = CextOp, BaseOpcode = CextOp in {
+    let opExtendable = 1, isExtentSigned = 1, opExtentBits = ImmBits,
+         isPredicable = 1 in
+    def NAME#_nv_V4 : NVInst_V4<(outs),
+            (ins MEMri:$addr, RC:$src),
+            mnemonic#"($addr) = $src.new",
             []>,
             Requires<[HasV4T]>;
 
-// if (Pv.new) memb(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STrib_indexed_shl_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1.new) memb($src2+$src3<<#$src4) = $src5.new",
-            []>,
-            Requires<[HasV4T]>;
+    let opExtendable = 2, isExtentSigned = 0, opExtentBits = PredImmBits,
+        neverHasSideEffects = 1, isPredicated = 1 in {
+      defm Pt : ST_MEMri_Pred_nv<mnemonic, RC, 0>;
+      defm NotPt : ST_MEMri_Pred_nv<mnemonic, RC, 1>;
+    }
+  }
+}
 
-// if (!Pv) memb(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STrib_indexed_shl_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1) memb($src2+$src3<<#$src4) = $src5.new",
-            []>,
-            Requires<[HasV4T]>;
+let addrMode = BaseImmOffset, isMEMri = "true", validSubTargets = HasV4SubT,
+mayStore = 1 in {
+  defm STrib: ST_MEMri_nv<"memb", "STrib", IntRegs, 11, 6>, AddrModeRel;
+  defm STrih: ST_MEMri_nv<"memh", "STrih", IntRegs, 12, 7>, AddrModeRel;
+  defm STriw: ST_MEMri_nv<"memw", "STriw", IntRegs, 13, 8>, AddrModeRel;
+}
 
-// if (!Pv.new) memb(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STrib_indexed_shl_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1.new) memb($src2+$src3<<#$src4) = $src5.new",
+// memb(Ru<<#u2+#U6)=Nt.new
+let isExtended = 1, opExtendable = 2, mayStore = 1, AddedComplexity = 10,
+isNVStore = 1, validSubTargets = HasV4SubT in
+def STrib_shl_nv_V4 : NVInst_V4<(outs),
+            (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
+            "memb($src1<<#$src2+#$src3) = $src4.new",
             []>,
             Requires<[HasV4T]>;
 
-// if ([!]Pv[.new]) memb(Rx++#s4:0)=Nt.new
-// if (Pv) memb(Rx++#s4:0)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STbri_cPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset),
-            "if ($src1) memb($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memb(Rx++#s4:0)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STbri_cdnPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset),
-            "if ($src1.new) memb($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memb(Rx++#s4:0)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STbri_cNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset),
-            "if (!$src1) memb($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
+//===----------------------------------------------------------------------===//
+// Post increment store
+// mem[bhwd](Rx++#s4:[0123])=Nt.new
+//===----------------------------------------------------------------------===//
 
-// if (!Pv.new) memb(Rx++#s4:0)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STbri_cdnNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_0Imm:$offset),
-            "if (!$src1.new) memb($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
+multiclass ST_PostInc_Pbase_nv<string mnemonic, RegisterClass RC, Operand ImmOp,
+                            bit isNot, bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME#_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
+            (ins PredRegs:$src1, IntRegs:$src2, ImmOp:$offset, RC:$src3),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"($src2++#$offset) = $src3.new",
+            [],
+            "$src2 = $dst">,
             Requires<[HasV4T]>;
+}
 
+multiclass ST_PostInc_Pred_nv<string mnemonic, RegisterClass RC,
+                           Operand ImmOp, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ST_PostInc_Pbase_nv<mnemonic, RC, ImmOp, PredNot, 0>;
+    // Predicate new
+    let Predicates = [HasV4T], validSubTargets = HasV4SubT in
+    defm _cdn#NAME : ST_PostInc_Pbase_nv<mnemonic, RC, ImmOp, PredNot, 1>;
+  }
+}
 
-// Store new-value halfword.
-// memh(Re=#U6)=Nt.new
-// memh(Rs+#s11:1)=Nt.new
-let mayStore = 1, isPredicable = 1 in
-def STrih_nv_V4 : NVInst_V4<(outs), (ins MEMri:$addr, IntRegs:$src1),
-            "memh($addr) = $src1.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, isPredicable = 1 in
-def STrih_indexed_nv_V4 : NVInst_V4<(outs),
-            (ins IntRegs:$src1, s11_1Imm:$src2, IntRegs:$src3),
-            "memh($src1+#$src2) = $src3.new",
-            []>,
-            Requires<[HasV4T]>;
+let hasCtrlDep = 1, isNVStore = 1, neverHasSideEffects = 1 in
+multiclass ST_PostInc_nv<string mnemonic, string BaseOp, RegisterClass RC,
+                      Operand ImmOp> {
+
+  let BaseOpcode = "POST_"#BaseOp in {
+    let isPredicable = 1 in
+    def NAME#_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
+                (ins IntRegs:$src1, ImmOp:$offset, RC:$src2),
+                mnemonic#"($src1++#$offset) = $src2.new",
+                [],
+                "$src1 = $dst">,
+                Requires<[HasV4T]>;
+
+    let isPredicated = 1 in {
+      defm Pt : ST_PostInc_Pred_nv<mnemonic, RC, ImmOp, 0 >;
+      defm NotPt : ST_PostInc_Pred_nv<mnemonic, RC, ImmOp, 1 >;
+    }
+  }
+}
 
-// memh(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10, isPredicable = 1 in
-def STrih_indexed_shl_nv_V4 : NVInst_V4<(outs),
-            (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, IntRegs:$src4),
-            "memh($src1+$src2<<#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
+let validSubTargets = HasV4SubT in {
+defm POST_STbri: ST_PostInc_nv <"memb", "STrib", IntRegs, s4_0Imm>, AddrModeRel;
+defm POST_SThri: ST_PostInc_nv <"memh", "STrih", IntRegs, s4_1Imm>, AddrModeRel;
+defm POST_STwri: ST_PostInc_nv <"memw", "STriw", IntRegs, s4_2Imm>, AddrModeRel;
+}
 
+// memb(Rx++#s4:0:circ(Mu))=Nt.new
+// memb(Rx++I:circ(Mu))=Nt.new
+// memb(Rx++Mu)=Nt.new
+// memb(Rx++Mu:brev)=Nt.new
 // memh(Ru<<#u2+#U6)=Nt.new
-let mayStore = 1, AddedComplexity = 10 in
+let isExtended = 1, opExtendable = 2, mayStore = 1, AddedComplexity = 10,
+isNVStore = 1, validSubTargets = HasV4SubT in
 def STrih_shl_nv_V4 : NVInst_V4<(outs),
-            (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, IntRegs:$src4),
+            (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
             "memh($src1<<#$src2+#$src3) = $src4.new",
             []>,
             Requires<[HasV4T]>;
 
-// memh(Rx++#s4:1)=Nt.new
-let mayStore = 1, hasCtrlDep = 1, isPredicable = 1  in
-def POST_SThri_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2, s4_1Imm:$offset),
-            "memh($src2++#$offset) = $src1.new",
-            [],
-            "$src2 = $dst">,
-            Requires<[HasV4T]>;
-
 // memh(Rx++#s4:1:circ(Mu))=Nt.new
 // memh(Rx++I:circ(Mu))=Nt.new
 // memh(Rx++Mu)=Nt.new
 // memh(Rx++Mu:brev)=Nt.new
 
-// memh(gp+#u16:1)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1 in
-def STrih_GP_nv_V4 : NVInst_V4<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src),
-            "memh(#$global+$offset) = $src.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// memh(#global)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1 in
-def STh_GP_nv_V4 : NVInst_V4<(outs),
-            (ins globaladdress:$global, IntRegs:$src),
-            "memh(#$global) = $src.new",
-            []>,
-            Requires<[HasV4T]>;
-
-
-// Store new-value halfword conditionally.
-
-// if ([!]Pv[.new]) memh(#u6)=Nt.new
-
-// if ([!]Pv[.new]) memh(Rs+#u6:1)=Nt.new
-// if (Pv) memh(Rs+#u6:1)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1) memh($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memh(Rs+#u6:1)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1.new) memh($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memh(Rs+#u6:1)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1) memh($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memh(Rs+#u6:1)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1.new) memh($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memh(Rs+#u6:1)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_indexed_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4),
-            "if ($src1) memh($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memh(Rs+#u6:1)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_indexed_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4),
-            "if ($src1.new) memh($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memh(Rs+#u6:1)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_indexed_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4),
-            "if (!$src1) memh($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memh(Rs+#u6:1)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STrih_indexed_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_1Imm:$src3, IntRegs:$src4),
-            "if (!$src1.new) memh($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if ([!]Pv[.new]) memh(Rs+Ru<<#u2)=Nt.new
-// if (Pv) memh(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STrih_indexed_shl_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1) memh($src2+$src3<<#$src4) = $src5.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memh(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STrih_indexed_shl_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1.new) memh($src2+$src3<<#$src4) = $src5.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memh(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STrih_indexed_shl_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1) memh($src2+$src3<<#$src4) = $src5.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memh(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STrih_indexed_shl_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1.new) memh($src2+$src3<<#$src4) = $src5.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if ([!]Pv[]) memh(Rx++#s4:1)=Nt.new
-// if (Pv) memh(Rx++#s4:1)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_SThri_cPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset),
-            "if ($src1) memh($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memh(Rx++#s4:1)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_SThri_cdnPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset),
-            "if ($src1.new) memh($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memh(Rx++#s4:1)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_SThri_cNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset),
-            "if (!$src1) memh($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memh(Rx++#s4:1)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_SThri_cdnNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_1Imm:$offset),
-            "if (!$src1.new) memh($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-
-// Store new-value word.
-
-// memw(Re=#U6)=Nt.new
-// memw(Rs+#s11:2)=Nt.new
-let mayStore = 1, isPredicable = 1 in
-def STriw_nv_V4 : NVInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$src1),
-            "memw($addr) = $src1.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, isPredicable = 1 in
-def STriw_indexed_nv_V4 : NVInst_V4<(outs),
-            (ins IntRegs:$src1, s11_2Imm:$src2, IntRegs:$src3),
-            "memw($src1+#$src2) = $src3.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// memw(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10, isPredicable = 1 in
-def STriw_indexed_shl_nv_V4 : NVInst_V4<(outs),
-            (ins IntRegs:$src1, IntRegs:$src2, u2Imm:$src3, IntRegs:$src4),
-            "memw($src1+$src2<<#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
 // memw(Ru<<#u2+#U6)=Nt.new
-let mayStore = 1, AddedComplexity = 10 in
+let isExtended = 1, opExtendable = 2, mayStore = 1, AddedComplexity = 10,
+isNVStore = 1, validSubTargets = HasV4SubT in
 def STriw_shl_nv_V4 : NVInst_V4<(outs),
-            (ins IntRegs:$src1, u2Imm:$src2, u6Imm:$src3, IntRegs:$src4),
+            (ins IntRegs:$src1, u2Imm:$src2, u0AlwaysExt:$src3, IntRegs:$src4),
             "memw($src1<<#$src2+#$src3) = $src4.new",
             []>,
             Requires<[HasV4T]>;
 
-// memw(Rx++#s4:2)=Nt.new
-let mayStore = 1, hasCtrlDep = 1, isPredicable = 1  in
-def POST_STwri_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2, s4_2Imm:$offset),
-            "memw($src2++#$offset) = $src1.new",
-            [],
-            "$src2 = $dst">,
-            Requires<[HasV4T]>;
-
 // memw(Rx++#s4:2:circ(Mu))=Nt.new
 // memw(Rx++I:circ(Mu))=Nt.new
 // memw(Rx++Mu)=Nt.new
 // memw(Rx++Mu:brev)=Nt.new
-// memw(gp+#u16:2)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1 in
-def STriw_GP_nv_V4 : NVInst_V4<(outs),
-            (ins globaladdress:$global, u16Imm:$offset, IntRegs:$src),
-            "memw(#$global+$offset) = $src.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STw_GP_nv_V4 : NVInst_V4<(outs),
-            (ins globaladdress:$global, IntRegs:$src),
-            "memw(#$global) = $src.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// Store new-value word conditionally.
-
-// if ([!]Pv[.new]) memw(#u6)=Nt.new
-
-// if ([!]Pv[.new]) memw(Rs+#u6:2)=Nt.new
-// if (Pv) memw(Rs+#u6:2)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1) memw($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memw(Rs+#u6:2)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if ($src1.new) memw($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memw(Rs+#u6:2)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1) memw($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memw(Rs+#u6:2)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, MEMri:$addr, IntRegs:$src2),
-            "if (!$src1.new) memw($addr) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memw(Rs+#u6:2)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_indexed_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4),
-            "if ($src1) memw($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memw(Rs+#u6:2)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_indexed_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4),
-            "if ($src1.new) memw($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memw(Rs+#u6:2)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_indexed_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4),
-            "if (!$src1) memw($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memw(Rs+#u6:2)=Nt.new
-let mayStore = 1, neverHasSideEffects = 1,
-    isPredicated = 1 in
-def STriw_indexed_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, u6_2Imm:$src3, IntRegs:$src4),
-            "if (!$src1.new) memw($src2+#$src3) = $src4.new",
-            []>,
-            Requires<[HasV4T]>;
-
-
-// if ([!]Pv[.new]) memw(Rs+Ru<<#u2)=Nt.new
-// if (Pv) memw(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STriw_indexed_shl_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1) memw($src2+$src3<<#$src4) = $src5.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memw(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STriw_indexed_shl_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if ($src1.new) memw($src2+$src3<<#$src4) = $src5.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memw(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STriw_indexed_shl_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1) memw($src2+$src3<<#$src4) = $src5.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memw(Rs+Ru<<#u2)=Nt.new
-let mayStore = 1, AddedComplexity = 10,
-    isPredicated = 1 in
-def STriw_indexed_shl_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, u2Imm:$src4,
-                 IntRegs:$src5),
-            "if (!$src1.new) memw($src2+$src3<<#$src4) = $src5.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if ([!]Pv[.new]) memw(Rx++#s4:2)=Nt.new
-// if (Pv) memw(Rx++#s4:2)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STwri_cPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset),
-            "if ($src1) memw($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (Pv.new) memw(Rx++#s4:2)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STwri_cdnPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset),
-            "if ($src1.new) memw($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memw(Rx++#s4:2)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STwri_cNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset),
-            "if (!$src1) memw($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-// if (!Pv.new) memw(Rx++#s4:2)=Nt.new
-let mayStore = 1, hasCtrlDep = 1,
-    isPredicated = 1 in
-def POST_STwri_cdnNotPt_nv_V4 : NVInstPI_V4<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, IntRegs:$src2, IntRegs:$src3, s4_2Imm:$offset),
-            "if (!$src1.new) memw($src3++#$offset) = $src2.new",
-            [],"$src3 = $dst">,
-            Requires<[HasV4T]>;
-
-
-
-// if (Pv) memb(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STb_GP_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1) memb(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memb(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STb_GP_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1) memb(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memb(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STb_GP_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1.new) memb(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memb(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STb_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1.new) memb(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memh(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STh_GP_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1) memh(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memh(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STh_GP_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1) memh(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memh(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STh_GP_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1.new) memh(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memh(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STh_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1.new) memh(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memw(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STw_GP_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1) memw(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memw(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STw_GP_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1) memw(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (Pv) memw(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STw_GP_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if ($src1.new) memw(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-// if (!Pv) memw(##global) = Rt
-let mayStore = 1, neverHasSideEffects = 1 in
-def STw_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, IntRegs:$src2),
-            "if (!$src1.new) memw(##$global) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STrib_GP_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1) memb(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STrib_GP_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1) memb(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STrib_GP_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1.new) memb(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STrib_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1.new) memb(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STrih_GP_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1) memh(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STrih_GP_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1) memh(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STrih_GP_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1.new) memh(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STrih_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1.new) memh(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STriw_GP_cPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1) memw(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STriw_GP_cNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1) memw(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STriw_GP_cdnPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if ($src1.new) memw(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
-
-let mayStore = 1, neverHasSideEffects = 1 in
-def STriw_GP_cdnNotPt_nv_V4 : NVInst_V4<(outs),
-            (ins PredRegs:$src1, globaladdress:$global, u16Imm:$offset,
-                                                        IntRegs:$src2),
-            "if (!$src1.new) memw(##$global+$offset) = $src2.new",
-            []>,
-            Requires<[HasV4T]>;
 
 //===----------------------------------------------------------------------===//
 // NV/ST -
@@ -3998,31 +1288,37 @@ let isBranch = 1, isTerminator=1, neverHasSideEffects = 1, Defs = [PC] in {
 
 //  Add and accumulate.
 //  Rd=add(Rs,add(Ru,#s6))
+let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 6,
+validSubTargets = HasV4SubT in
 def ADDr_ADDri_V4 : MInst<(outs IntRegs:$dst),
-          (ins IntRegs:$src1, IntRegs:$src2, s6Imm:$src3),
+          (ins IntRegs:$src1, IntRegs:$src2, s6Ext:$src3),
           "$dst = add($src1, add($src2, #$src3))",
           [(set (i32 IntRegs:$dst),
            (add (i32 IntRegs:$src1), (add (i32 IntRegs:$src2),
-                                          s6ImmPred:$src3)))]>,
+                                          s6_16ExtPred:$src3)))]>,
           Requires<[HasV4T]>;
 
 //  Rd=add(Rs,sub(#s6,Ru))
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
+validSubTargets = HasV4SubT in
 def ADDr_SUBri_V4 : MInst<(outs IntRegs:$dst),
-          (ins IntRegs:$src1, s6Imm:$src2, IntRegs:$src3),
+          (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
           "$dst = add($src1, sub(#$src2, $src3))",
           [(set (i32 IntRegs:$dst),
-           (add (i32 IntRegs:$src1), (sub s6ImmPred:$src2,
+           (add (i32 IntRegs:$src1), (sub s6_10ExtPred:$src2,
                                           (i32 IntRegs:$src3))))]>,
           Requires<[HasV4T]>;
 
 // Generates the same instruction as ADDr_SUBri_V4 but matches different
 // pattern.
 //  Rd=add(Rs,sub(#s6,Ru))
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 6,
+validSubTargets = HasV4SubT in
 def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst),
-          (ins IntRegs:$src1, s6Imm:$src2, IntRegs:$src3),
+          (ins IntRegs:$src1, s6Ext:$src2, IntRegs:$src3),
           "$dst = add($src1, sub(#$src2, $src3))",
           [(set (i32 IntRegs:$dst),
-                (sub (add (i32 IntRegs:$src1), s6ImmPred:$src2),
+                (sub (add (i32 IntRegs:$src1), s6_10ExtPred:$src2),
                      (i32 IntRegs:$src3)))]>,
           Requires<[HasV4T]>;
 
@@ -4036,6 +1332,7 @@ def ADDri_SUBr_V4 : MInst<(outs IntRegs:$dst),
 
 //  Logical doublewords.
 //  Rdd=and(Rtt,~Rss)
+let validSubTargets = HasV4SubT in
 def ANDd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
           (ins DoubleRegs:$src1, DoubleRegs:$src2),
           "$dst = and($src1, ~$src2)",
@@ -4044,6 +1341,7 @@ def ANDd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
           Requires<[HasV4T]>;
 
 //  Rdd=or(Rtt,~Rss)
+let validSubTargets = HasV4SubT in
 def ORd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
           (ins DoubleRegs:$src1, DoubleRegs:$src2),
           "$dst = or($src1, ~$src2)",
@@ -4054,6 +1352,7 @@ def ORd_NOTd_V4 : MInst<(outs DoubleRegs:$dst),
 
 //  Logical-logical doublewords.
 //  Rxx^=xor(Rss,Rtt)
+let validSubTargets = HasV4SubT in
 def XORd_XORdd: MInst_acc<(outs DoubleRegs:$dst),
           (ins DoubleRegs:$src1, DoubleRegs:$src2, DoubleRegs:$src3),
           "$dst ^= xor($src2, $src3)",
@@ -4066,17 +1365,20 @@ def XORd_XORdd: MInst_acc<(outs DoubleRegs:$dst),
 
 // Logical-logical words.
 // Rx=or(Ru,and(Rx,#s10))
+let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
+validSubTargets = HasV4SubT in
 def ORr_ANDri_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs: $src2, s10Imm:$src3),
+            (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
             "$dst = or($src1, and($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
-                                                s10ImmPred:$src3)))],
+                                                s10ExtPred:$src3)))],
             "$src2 = $dst">,
             Requires<[HasV4T]>;
 
 // Rx[&|^]=and(Rs,Rt)
 // Rx&=and(Rs,Rt)
+let validSubTargets = HasV4SubT in
 def ANDr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst &= and($src2, $src3)",
@@ -4087,6 +1389,7 @@ def ANDr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
             Requires<[HasV4T]>;
 
 // Rx|=and(Rs,Rt)
+let validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "reg" in
 def ORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst |= and($src2, $src3)",
@@ -4094,9 +1397,10 @@ def ORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
                   (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
                                                 (i32 IntRegs:$src3))))],
             "$src1 = $dst">,
-            Requires<[HasV4T]>;
+            Requires<[HasV4T]>, ImmRegRel;
 
 // Rx^=and(Rs,Rt)
+let validSubTargets = HasV4SubT in
 def XORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst ^= and($src2, $src3)",
@@ -4108,6 +1412,7 @@ def XORr_ANDrr_V4 : MInst_acc<(outs IntRegs:$dst),
 
 // Rx[&|^]=and(Rs,~Rt)
 // Rx&=and(Rs,~Rt)
+let validSubTargets = HasV4SubT in
 def ANDr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst &= and($src2, ~$src3)",
@@ -4118,6 +1423,7 @@ def ANDr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
             Requires<[HasV4T]>;
 
 // Rx|=and(Rs,~Rt)
+let validSubTargets = HasV4SubT in
 def ORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst |= and($src2, ~$src3)",
@@ -4128,6 +1434,7 @@ def ORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
             Requires<[HasV4T]>;
 
 // Rx^=and(Rs,~Rt)
+let validSubTargets = HasV4SubT in
 def XORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst ^= and($src2, ~$src3)",
@@ -4139,6 +1446,7 @@ def XORr_ANDr_NOTr_V4 : MInst_acc<(outs IntRegs:$dst),
 
 // Rx[&|^]=or(Rs,Rt)
 // Rx&=or(Rs,Rt)
+let validSubTargets = HasV4SubT in
 def ANDr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst &= or($src2, $src3)",
@@ -4149,6 +1457,7 @@ def ANDr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
             Requires<[HasV4T]>;
 
 // Rx|=or(Rs,Rt)
+let validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "reg" in
 def ORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst |= or($src2, $src3)",
@@ -4156,9 +1465,10 @@ def ORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
                   (or (i32 IntRegs:$src1), (or (i32 IntRegs:$src2),
                                                (i32 IntRegs:$src3))))],
             "$src1 = $dst">,
-            Requires<[HasV4T]>;
+            Requires<[HasV4T]>, ImmRegRel;
 
 // Rx^=or(Rs,Rt)
+let validSubTargets = HasV4SubT in
 def XORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst ^= or($src2, $src3)",
@@ -4170,6 +1480,7 @@ def XORr_ORrr_V4 : MInst_acc<(outs IntRegs:$dst),
 
 // Rx[&|^]=xor(Rs,Rt)
 // Rx&=xor(Rs,Rt)
+let validSubTargets = HasV4SubT in
 def ANDr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst &= xor($src2, $src3)",
@@ -4180,6 +1491,7 @@ def ANDr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
             Requires<[HasV4T]>;
 
 // Rx|=xor(Rs,Rt)
+let validSubTargets = HasV4SubT in
 def ORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst |= xor($src2, $src3)",
@@ -4190,6 +1502,7 @@ def ORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
             Requires<[HasV4T]>;
 
 // Rx^=xor(Rs,Rt)
+let validSubTargets = HasV4SubT in
 def XORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs: $src2, IntRegs:$src3),
             "$dst ^= xor($src2, $src3)",
@@ -4200,24 +1513,28 @@ def XORr_XORrr_V4 : MInst_acc<(outs IntRegs:$dst),
             Requires<[HasV4T]>;
 
 // Rx|=and(Rs,#s10)
+let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
+validSubTargets = HasV4SubT, CextOpcode = "ORr_ANDr", InputType = "imm" in
 def ORr_ANDri2_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs: $src2, s10Imm:$src3),
+            (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
             "$dst |= and($src2, #$src3)",
             [(set (i32 IntRegs:$dst),
                   (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
-                                                s10ImmPred:$src3)))],
+                                                s10ExtPred:$src3)))],
             "$src1 = $dst">,
-            Requires<[HasV4T]>;
+            Requires<[HasV4T]>, ImmRegRel;
 
 // Rx|=or(Rs,#s10)
+let isExtendable = 1, opExtendable = 3, isExtentSigned = 1, opExtentBits = 10,
+validSubTargets = HasV4SubT, CextOpcode = "ORr_ORr", InputType = "imm" in
 def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs: $src2, s10Imm:$src3),
+            (ins IntRegs:$src1, IntRegs: $src2, s10Ext:$src3),
             "$dst |= or($src2, #$src3)",
             [(set (i32 IntRegs:$dst),
                   (or (i32 IntRegs:$src1), (and (i32 IntRegs:$src2),
-                                                s10ImmPred:$src3)))],
+                                                s10ExtPred:$src3)))],
             "$src1 = $dst">,
-            Requires<[HasV4T]>;
+            Requires<[HasV4T]>, ImmRegRel;
 
 
 //    Modulo wrap
@@ -4264,25 +1581,41 @@ def ORr_ORri_V4 : MInst_acc<(outs IntRegs:$dst),
 
 // Multiply and user lower result.
 // Rd=add(#u6,mpyi(Rs,#U6))
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
+validSubTargets = HasV4SubT in
 def ADDi_MPYri_V4 : MInst<(outs IntRegs:$dst),
-            (ins u6Imm:$src1, IntRegs:$src2, u6Imm:$src3),
+            (ins u6Ext:$src1, IntRegs:$src2, u6Imm:$src3),
             "$dst = add(#$src1, mpyi($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
-                       u6ImmPred:$src1))]>,
+                       u6ExtPred:$src1))]>,
             Requires<[HasV4T]>;
 
-// Rd=add(#u6,mpyi(Rs,Rt))
+// Rd=add(##,mpyi(Rs,#U6))
+def : Pat <(add (mul (i32 IntRegs:$src2), u6ImmPred:$src3),
+                     (HexagonCONST32 tglobaladdr:$src1)),
+           (i32 (ADDi_MPYri_V4 tglobaladdr:$src1, IntRegs:$src2,
+                               u6ImmPred:$src3))>;
 
+// Rd=add(#u6,mpyi(Rs,Rt))
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 6,
+validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
 def ADDi_MPYrr_V4 : MInst<(outs IntRegs:$dst),
-            (ins u6Imm:$src1, IntRegs:$src2, IntRegs:$src3),
+            (ins u6Ext:$src1, IntRegs:$src2, IntRegs:$src3),
             "$dst = add(#$src1, mpyi($src2, $src3))",
             [(set (i32 IntRegs:$dst),
                   (add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
-                       u6ImmPred:$src1))]>,
-            Requires<[HasV4T]>;
+                       u6ExtPred:$src1))]>,
+            Requires<[HasV4T]>, ImmRegRel;
+
+// Rd=add(##,mpyi(Rs,Rt))
+def : Pat <(add (mul (i32 IntRegs:$src2), (i32 IntRegs:$src3)),
+                     (HexagonCONST32 tglobaladdr:$src1)),
+           (i32 (ADDi_MPYrr_V4 tglobaladdr:$src1, IntRegs:$src2,
+                               IntRegs:$src3))>;
 
 // Rd=add(Ru,mpyi(#u6:2,Rs))
+let validSubTargets = HasV4SubT in
 def ADDr_MPYir_V4 : MInst<(outs IntRegs:$dst),
             (ins IntRegs:$src1, u6Imm:$src2, IntRegs:$src3),
             "$dst = add($src1, mpyi(#$src2, $src3))",
@@ -4292,15 +1625,18 @@ def ADDr_MPYir_V4 : MInst<(outs IntRegs:$dst),
             Requires<[HasV4T]>;
 
 // Rd=add(Ru,mpyi(Rs,#u6))
+let isExtendable = 1, opExtendable = 3, isExtentSigned = 0, opExtentBits = 6,
+validSubTargets = HasV4SubT, InputType = "imm", CextOpcode = "ADD_MPY" in
 def ADDr_MPYri_V4 : MInst<(outs IntRegs:$dst),
-            (ins IntRegs:$src1, IntRegs:$src2, u6Imm:$src3),
+            (ins IntRegs:$src1, IntRegs:$src2, u6Ext:$src3),
             "$dst = add($src1, mpyi($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
-                                                 u6ImmPred:$src3)))]>,
-            Requires<[HasV4T]>;
+                                                 u6ExtPred:$src3)))]>,
+            Requires<[HasV4T]>, ImmRegRel;
 
 // Rx=add(Ru,mpyi(Rx,Rs))
+let validSubTargets = HasV4SubT, InputType = "reg", CextOpcode = "ADD_MPY" in
 def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2, IntRegs:$src3),
             "$dst = add($src1, mpyi($src2, $src3))",
@@ -4308,7 +1644,7 @@ def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst),
              (add (i32 IntRegs:$src1), (mul (i32 IntRegs:$src2),
                                             (i32 IntRegs:$src3))))],
             "$src2 = $dst">,
-            Requires<[HasV4T]>;
+            Requires<[HasV4T]>, ImmRegRel;
 
 
 // Polynomial multiply words
@@ -4351,92 +1687,107 @@ def ADDr_MPYrr_V4 : MInst_acc<(outs IntRegs:$dst),
 
 // Shift by immediate and accumulate.
 // Rx=add(#u8,asl(Rx,#U5))
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
+validSubTargets = HasV4SubT in
 def ADDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3),
+            (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
             "$dst = add(#$src1, asl($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (add (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
-                       u8ImmPred:$src1))],
+                       u8ExtPred:$src1))],
             "$src2 = $dst">,
             Requires<[HasV4T]>;
 
 // Rx=add(#u8,lsr(Rx,#U5))
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
+validSubTargets = HasV4SubT in
 def ADDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3),
+            (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
             "$dst = add(#$src1, lsr($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (add (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
-                       u8ImmPred:$src1))],
+                       u8ExtPred:$src1))],
             "$src2 = $dst">,
             Requires<[HasV4T]>;
 
 // Rx=sub(#u8,asl(Rx,#U5))
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
+validSubTargets = HasV4SubT in
 def SUBi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3),
+            (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
             "$dst = sub(#$src1, asl($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (sub (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
-                       u8ImmPred:$src1))],
+                       u8ExtPred:$src1))],
             "$src2 = $dst">,
             Requires<[HasV4T]>;
 
 // Rx=sub(#u8,lsr(Rx,#U5))
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
+validSubTargets = HasV4SubT in
 def SUBi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3),
+            (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
             "$dst = sub(#$src1, lsr($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (sub (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
-                       u8ImmPred:$src1))],
+                       u8ExtPred:$src1))],
             "$src2 = $dst">,
             Requires<[HasV4T]>;
 
 
 //Shift by immediate and logical.
 //Rx=and(#u8,asl(Rx,#U5))
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
+validSubTargets = HasV4SubT in
 def ANDi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3),
+            (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
             "$dst = and(#$src1, asl($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (and (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
-                       u8ImmPred:$src1))],
+                       u8ExtPred:$src1))],
             "$src2 = $dst">,
             Requires<[HasV4T]>;
 
 //Rx=and(#u8,lsr(Rx,#U5))
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
+validSubTargets = HasV4SubT in
 def ANDi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3),
+            (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
             "$dst = and(#$src1, lsr($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (and (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
-                       u8ImmPred:$src1))],
+                       u8ExtPred:$src1))],
             "$src2 = $dst">,
             Requires<[HasV4T]>;
 
 //Rx=or(#u8,asl(Rx,#U5))
-let AddedComplexity = 30 in
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
+AddedComplexity = 30, validSubTargets = HasV4SubT in
 def ORi_ASLri_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3),
+            (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
             "$dst = or(#$src1, asl($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (or (shl (i32 IntRegs:$src2), u5ImmPred:$src3),
-                      u8ImmPred:$src1))],
+                      u8ExtPred:$src1))],
             "$src2 = $dst">,
             Requires<[HasV4T]>;
 
 //Rx=or(#u8,lsr(Rx,#U5))
-let AddedComplexity = 30 in
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0, opExtentBits = 8,
+AddedComplexity = 30, validSubTargets = HasV4SubT in
 def ORi_LSRri_V4 : MInst_acc<(outs IntRegs:$dst),
-            (ins u8Imm:$src1, IntRegs:$src2, u5Imm:$src3),
+            (ins u8Ext:$src1, IntRegs:$src2, u5Imm:$src3),
             "$dst = or(#$src1, lsr($src2, #$src3))",
             [(set (i32 IntRegs:$dst),
                   (or (srl (i32 IntRegs:$src2), u5ImmPred:$src3),
-                      u8ImmPred:$src1))],
+                      u8ExtPred:$src1))],
             "$src2 = $dst">,
             Requires<[HasV4T]>;
 
 
 //Shift by register.
 //Rd=lsl(#s6,Rt)
+let validSubTargets = HasV4SubT in {
 def LSLi_V4 : MInst<(outs IntRegs:$dst), (ins s6Imm:$src1, IntRegs:$src2),
             "$dst = lsl(#$src1, $src2)",
             [(set (i32 IntRegs:$dst), (shl s6ImmPred:$src1,
@@ -4484,7 +1835,7 @@ def LSRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
                                                     (i32 IntRegs:$src3))))],
             "$src1 = $dst">,
             Requires<[HasV4T]>;
-
+}
 
 //===----------------------------------------------------------------------===//
 // XTYPE/SHIFT -
@@ -4494,488 +1845,367 @@ def LSRd_rr_xor_V4 : MInst_acc<(outs DoubleRegs:$dst),
 // MEMOP: Word, Half, Byte
 //===----------------------------------------------------------------------===//
 
+def MEMOPIMM : SDNodeXForm<imm, [{
+  // Call the transformation function XformM5ToU5Imm to get the negative
+  // immediate's positive counterpart.
+  int32_t imm = N->getSExtValue();
+  return XformM5ToU5Imm(imm);
+}]>;
+
+def MEMOPIMM_HALF : SDNodeXForm<imm, [{
+  // -1 .. -31 represented as 65535..65515
+  // assigning to a short restores our desired signed value.
+  // Call the transformation function XformM5ToU5Imm to get the negative
+  // immediate's positive counterpart.
+  int16_t imm = N->getSExtValue();
+  return XformM5ToU5Imm(imm);
+}]>;
+
+def MEMOPIMM_BYTE : SDNodeXForm<imm, [{
+  // -1 .. -31 represented as 255..235
+  // assigning to a char restores our desired signed value.
+  // Call the transformation function XformM5ToU5Imm to get the negative
+  // immediate's positive counterpart.
+  int8_t imm = N->getSExtValue();
+  return XformM5ToU5Imm(imm);
+}]>;
+
+def SETMEMIMM : SDNodeXForm<imm, [{
+   // Return the bit position we will set [0-31].
+   // As an SDNode.
+   int32_t imm = N->getSExtValue();
+   return XformMskToBitPosU5Imm(imm);
+}]>;
+
+def CLRMEMIMM : SDNodeXForm<imm, [{
+   // Return the bit position we will clear [0-31].
+   // As an SDNode.
+   // we bit negate the value first
+   int32_t imm = ~(N->getSExtValue());
+   return XformMskToBitPosU5Imm(imm);
+}]>;
+
+def SETMEMIMM_SHORT : SDNodeXForm<imm, [{
+   // Return the bit position we will set [0-15].
+   // As an SDNode.
+   int16_t imm = N->getSExtValue();
+   return XformMskToBitPosU4Imm(imm);
+}]>;
+
+def CLRMEMIMM_SHORT : SDNodeXForm<imm, [{
+   // Return the bit position we will clear [0-15].
+   // As an SDNode.
+   // we bit negate the value first
+   int16_t imm = ~(N->getSExtValue());
+   return XformMskToBitPosU4Imm(imm);
+}]>;
+
+def SETMEMIMM_BYTE : SDNodeXForm<imm, [{
+   // Return the bit position we will set [0-7].
+   // As an SDNode.
+   int8_t imm =  N->getSExtValue();
+   return XformMskToBitPosU3Imm(imm);
+}]>;
+
+def CLRMEMIMM_BYTE : SDNodeXForm<imm, [{
+   // Return the bit position we will clear [0-7].
+   // As an SDNode.
+   // we bit negate the value first
+   int8_t imm = ~(N->getSExtValue());
+   return XformMskToBitPosU3Imm(imm);
+}]>;
+
 //===----------------------------------------------------------------------===//
-// MEMOP: Word
-//
-//  Implemented:
-//     MEMw_ADDi_indexed_V4  : memw(Rs+#u6:2)+=#U5
-//     MEMw_SUBi_indexed_V4  : memw(Rs+#u6:2)-=#U5
-//     MEMw_ADDr_indexed_V4  : memw(Rs+#u6:2)+=Rt
-//     MEMw_SUBr_indexed_V4  : memw(Rs+#u6:2)-=Rt
-//     MEMw_CLRr_indexed_V4  : memw(Rs+#u6:2)&=Rt
-//     MEMw_SETr_indexed_V4  : memw(Rs+#u6:2)|=Rt
-//     MEMw_ADDi_V4          : memw(Rs+#u6:2)+=#U5
-//     MEMw_SUBi_V4          : memw(Rs+#u6:2)-=#U5
-//     MEMw_ADDr_V4          : memw(Rs+#u6:2)+=Rt
-//     MEMw_SUBr_V4          : memw(Rs+#u6:2)-=Rt
-//     MEMw_CLRr_V4          : memw(Rs+#u6:2)&=Rt
-//     MEMw_SETr_V4          : memw(Rs+#u6:2)|=Rt
-//
-//   Not implemented:
-//     MEMw_CLRi_indexed_V4  : memw(Rs+#u6:2)=clrbit(#U5)
-//     MEMw_SETi_indexed_V4  : memw(Rs+#u6:2)=setbit(#U5)
-//     MEMw_CLRi_V4          : memw(Rs+#u6:2)=clrbit(#U5)
-//     MEMw_SETi_V4          : memw(Rs+#u6:2)=setbit(#U5)
+// Template class for MemOp instructions with the register value.
+//===----------------------------------------------------------------------===//
+class MemOp_rr_base <string opc, bits<2> opcBits, Operand ImmOp,
+                     string memOp, bits<2> memOpBits> :
+      MEMInst_V4<(outs),
+                 (ins IntRegs:$base, ImmOp:$offset, IntRegs:$delta),
+                 opc#"($base+#$offset)"#memOp#"$delta",
+                 []>,
+                 Requires<[HasV4T, UseMEMOP]> {
+
+    bits<5> base;
+    bits<5> delta;
+    bits<32> offset;
+    bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
+
+    let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
+                     !if (!eq(opcBits, 0b01), offset{6-1},
+                     !if (!eq(opcBits, 0b10), offset{7-2},0)));
+
+    let IClass = 0b0011;
+    let Inst{27-24} = 0b1110;
+    let Inst{22-21} = opcBits;
+    let Inst{20-16} = base;
+    let Inst{13} = 0b0;
+    let Inst{12-7} = offsetBits;
+    let Inst{6-5} = memOpBits;
+    let Inst{4-0} = delta;
+}
+
+//===----------------------------------------------------------------------===//
+// Template class for MemOp instructions with the immediate value.
 //===----------------------------------------------------------------------===//
+class MemOp_ri_base <string opc, bits<2> opcBits, Operand ImmOp,
+                     string memOp, bits<2> memOpBits> :
+      MEMInst_V4 <(outs),
+                  (ins IntRegs:$base, ImmOp:$offset, u5Imm:$delta),
+                  opc#"($base+#$offset)"#memOp#"#$delta"
+                  #!if(memOpBits{1},")", ""), // clrbit, setbit - include ')'
+                  []>,
+                  Requires<[HasV4T, UseMEMOP]> {
+
+    bits<5> base;
+    bits<5> delta;
+    bits<32> offset;
+    bits<6> offsetBits; // memb - u6:0 , memh - u6:1, memw - u6:2
+
+    let offsetBits = !if (!eq(opcBits, 0b00), offset{5-0},
+                     !if (!eq(opcBits, 0b01), offset{6-1},
+                     !if (!eq(opcBits, 0b10), offset{7-2},0)));
+
+    let IClass = 0b0011;
+    let Inst{27-24} = 0b1111;
+    let Inst{22-21} = opcBits;
+    let Inst{20-16} = base;
+    let Inst{13} = 0b0;
+    let Inst{12-7} = offsetBits;
+    let Inst{6-5} = memOpBits;
+    let Inst{4-0} = delta;
+}
 
+// multiclass to define MemOp instructions with register operand.
+multiclass MemOp_rr<string opc, bits<2> opcBits, Operand ImmOp> {
+  def _ADD#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " += ", 0b00>; // add
+  def _SUB#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " -= ", 0b01>; // sub
+  def _AND#NAME#_V4 : MemOp_rr_base <opc, opcBits, ImmOp, " &= ", 0b10>; // and
+  def _OR#NAME#_V4  : MemOp_rr_base <opc, opcBits, ImmOp, " |= ", 0b11>; // or
+}
 
-// MEMw_ADDSUBi_indexed_V4:
-//   pseudo operation for MEMw_ADDi_indexed_V4 and
-//   MEMw_SUBi_indexed_V4 a later pass will change it
-//   to the corresponding pattern.
-let AddedComplexity = 30 in
-def MEMw_ADDSUBi_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_2Imm:$offset, m6Imm:$addend),
-            "Error; should not emit",
-            [(store (add (load (add (i32 IntRegs:$base), u6_2ImmPred:$offset)),
-                         m6ImmPred:$addend),
-                    (add (i32 IntRegs:$base), u6_2ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memw(Rs+#u6:2) += #U5
-let AddedComplexity = 30 in
-def MEMw_ADDi_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_2Imm:$offset, u5Imm:$addend),
-            "memw($base+#$offset) += #$addend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
+// multiclass to define MemOp instructions with immediate Operand.
+multiclass MemOp_ri<string opc, bits<2> opcBits, Operand ImmOp> {
+  def _ADD#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " += ", 0b00 >;
+  def _SUB#NAME#_V4 : MemOp_ri_base <opc, opcBits, ImmOp, " -= ", 0b01 >;
+  def _CLRBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =clrbit(", 0b10>;
+  def _SETBIT#NAME#_V4 : MemOp_ri_base<opc, opcBits, ImmOp, " =setbit(", 0b11>;
+}
 
-// memw(Rs+#u6:2) -= #U5
-let AddedComplexity = 30 in
-def MEMw_SUBi_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_2Imm:$offset, u5Imm:$subend),
-            "memw($base+#$offset) -= #$subend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memw(Rs+#u6:2) += Rt
-let AddedComplexity = 30 in
-def MEMw_ADDr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_2Imm:$offset, IntRegs:$addend),
-            "memw($base+#$offset) += $addend",
-            [(store (add (load (add (i32 IntRegs:$base), u6_2ImmPred:$offset)),
-                         (i32 IntRegs:$addend)),
-                    (add (i32 IntRegs:$base), u6_2ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memw(Rs+#u6:2) -= Rt
-let AddedComplexity = 30 in
-def MEMw_SUBr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_2Imm:$offset, IntRegs:$subend),
-            "memw($base+#$offset) -= $subend",
-            [(store (sub (load (add (i32 IntRegs:$base), u6_2ImmPred:$offset)),
-                         (i32 IntRegs:$subend)),
-                    (add (i32 IntRegs:$base), u6_2ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memw(Rs+#u6:2) &= Rt
-let AddedComplexity = 30 in
-def MEMw_ANDr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_2Imm:$offset, IntRegs:$andend),
-            "memw($base+#$offset) &= $andend",
-            [(store (and (load (add (i32 IntRegs:$base), u6_2ImmPred:$offset)),
-                         (i32 IntRegs:$andend)),
-                    (add (i32 IntRegs:$base), u6_2ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memw(Rs+#u6:2) |= Rt
-let AddedComplexity = 30 in
-def MEMw_ORr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_2Imm:$offset, IntRegs:$orend),
-            "memw($base+#$offset) |= $orend",
-            [(store (or (load (add (i32 IntRegs:$base), u6_2ImmPred:$offset)),
-                        (i32 IntRegs:$orend)),
-                    (add (i32 IntRegs:$base), u6_2ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// MEMw_ADDSUBi_V4:
-//   Pseudo operation for MEMw_ADDi_V4 and MEMw_SUBi_V4
-//   a later pass will change it to the right pattern.
-let AddedComplexity = 30 in
-def MEMw_ADDSUBi_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, m6Imm:$addend),
-            "Error; should not emit",
-            [(store (add (load ADDRriU6_2:$addr), m6ImmPred:$addend),
-                    ADDRriU6_2:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memw(Rs+#u6:2) += #U5
-let AddedComplexity = 30 in
-def MEMw_ADDi_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, u5Imm:$addend),
-            "memw($addr) += $addend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
+multiclass MemOp_base <string opc, bits<2> opcBits, Operand ImmOp> {
+  defm r : MemOp_rr <opc, opcBits, ImmOp>;
+  defm i : MemOp_ri <opc, opcBits, ImmOp>;
+}
 
-// memw(Rs+#u6:2) -= #U5
-let AddedComplexity = 30 in
-def MEMw_SUBi_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, u5Imm:$subend),
-            "memw($addr) -= $subend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memw(Rs+#u6:2) += Rt
-let AddedComplexity = 30 in
-def MEMw_ADDr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$addend),
-            "memw($addr) += $addend",
-            [(store (add (load ADDRriU6_2:$addr), (i32 IntRegs:$addend)),
-                    ADDRriU6_2:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memw(Rs+#u6:2) -= Rt
-let AddedComplexity = 30 in
-def MEMw_SUBr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$subend),
-            "memw($addr) -= $subend",
-            [(store (sub (load ADDRriU6_2:$addr), (i32 IntRegs:$subend)),
-                    ADDRriU6_2:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memw(Rs+#u6:2) &= Rt
-let AddedComplexity = 30 in
-def MEMw_ANDr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$andend),
-            "memw($addr) &= $andend",
-            [(store (and (load ADDRriU6_2:$addr), (i32 IntRegs:$andend)),
-                    ADDRriU6_2:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memw(Rs+#u6:2) |= Rt
-let AddedComplexity = 30 in
-def MEMw_ORr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$orend),
-            "memw($addr) |= $orend",
-            [(store (or (load ADDRriU6_2:$addr), (i32 IntRegs:$orend)),
-                    ADDRriU6_2:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
+// Define MemOp instructions.
+let isExtendable = 1, opExtendable = 1, isExtentSigned = 0,
+validSubTargets =HasV4SubT in {
+  let opExtentBits = 6, accessSize = ByteAccess in
+  defm MemOPb : MemOp_base <"memb", 0b00, u6_0Ext>;
+
+  let opExtentBits = 7, accessSize = HalfWordAccess in
+  defm MemOPh : MemOp_base <"memh", 0b01, u6_1Ext>;
+
+  let opExtentBits = 8, accessSize = WordAccess in
+  defm MemOPw : MemOp_base <"memw", 0b10, u6_2Ext>;
+}
 
 //===----------------------------------------------------------------------===//
-// MEMOP: Halfword
-//
-//  Implemented:
-//     MEMh_ADDi_indexed_V4  : memw(Rs+#u6:2)+=#U5
-//     MEMh_SUBi_indexed_V4  : memw(Rs+#u6:2)-=#U5
-//     MEMh_ADDr_indexed_V4  : memw(Rs+#u6:2)+=Rt
-//     MEMh_SUBr_indexed_V4  : memw(Rs+#u6:2)-=Rt
-//     MEMh_CLRr_indexed_V4  : memw(Rs+#u6:2)&=Rt
-//     MEMh_SETr_indexed_V4  : memw(Rs+#u6:2)|=Rt
-//     MEMh_ADDi_V4          : memw(Rs+#u6:2)+=#U5
-//     MEMh_SUBi_V4          : memw(Rs+#u6:2)-=#U5
-//     MEMh_ADDr_V4          : memw(Rs+#u6:2)+=Rt
-//     MEMh_SUBr_V4          : memw(Rs+#u6:2)-=Rt
-//     MEMh_CLRr_V4          : memw(Rs+#u6:2)&=Rt
-//     MEMh_SETr_V4          : memw(Rs+#u6:2)|=Rt
-//
-//   Not implemented:
-//     MEMh_CLRi_indexed_V4  : memw(Rs+#u6:2)=clrbit(#U5)
-//     MEMh_SETi_indexed_V4  : memw(Rs+#u6:2)=setbit(#U5)
-//     MEMh_CLRi_V4          : memw(Rs+#u6:2)=clrbit(#U5)
-//     MEMh_SETi_V4          : memw(Rs+#u6:2)=setbit(#U5)
+// Multiclass to define 'Def Pats' for ALU operations on the memory
+// Here value used for the ALU operation is an immediate value.
+// mem[bh](Rs+#0) += #U5
+// mem[bh](Rs+#u6) += #U5
 //===----------------------------------------------------------------------===//
 
+multiclass MemOpi_u5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
+                          InstHexagon MI, SDNode OpNode> {
+  let AddedComplexity = 180 in
+  def : Pat < (stOp (OpNode (ldOp IntRegs:$addr), u5ImmPred:$addend),
+                    IntRegs:$addr),
+              (MI IntRegs:$addr, #0, u5ImmPred:$addend )>;
+
+  let AddedComplexity = 190 in
+  def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, ExtPred:$offset)),
+                     u5ImmPred:$addend),
+             (add IntRegs:$base, ExtPred:$offset)),
+       (MI IntRegs:$base, ExtPred:$offset, u5ImmPred:$addend)>;
+}
 
-// MEMh_ADDSUBi_indexed_V4:
-//   Pseudo operation for MEMh_ADDi_indexed_V4 and
-//   MEMh_SUBi_indexed_V4 a later pass will change it
-//   to the corresponding pattern.
-let AddedComplexity = 30 in
-def MEMh_ADDSUBi_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_1Imm:$offset, m6Imm:$addend),
-            "Error; should not emit",
-            [(truncstorei16 (add (sextloadi16 (add (i32 IntRegs:$base),
-                                                   u6_1ImmPred:$offset)),
-                                 m6ImmPred:$addend),
-                            (add (i32 IntRegs:$base), u6_1ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memh(Rs+#u6:1) += #U5
-let AddedComplexity = 30 in
-def MEMh_ADDi_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_1Imm:$offset, u5Imm:$addend),
-            "memh($base+#$offset) += $addend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
+multiclass MemOpi_u5ALUOp<PatFrag ldOp, PatFrag stOp, PatLeaf ExtPred,
+                          InstHexagon addMI, InstHexagon subMI> {
+  defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, addMI, add>;
+  defm : MemOpi_u5Pats<ldOp, stOp, ExtPred, subMI, sub>;
+}
 
-// memh(Rs+#u6:1) -= #U5
-let AddedComplexity = 30 in
-def MEMh_SUBi_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_1Imm:$offset, u5Imm:$subend),
-            "memh($base+#$offset) -= $subend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memh(Rs+#u6:1) += Rt
-let AddedComplexity = 30 in
-def MEMh_ADDr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_1Imm:$offset, IntRegs:$addend),
-            "memh($base+#$offset) += $addend",
-            [(truncstorei16 (add (sextloadi16 (add (i32 IntRegs:$base),
-                                                   u6_1ImmPred:$offset)),
-                                 (i32 IntRegs:$addend)),
-                            (add (i32 IntRegs:$base), u6_1ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memh(Rs+#u6:1) -= Rt
-let AddedComplexity = 30 in
-def MEMh_SUBr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_1Imm:$offset, IntRegs:$subend),
-            "memh($base+#$offset) -= $subend",
-            [(truncstorei16 (sub (sextloadi16 (add (i32 IntRegs:$base),
-                                                   u6_1ImmPred:$offset)),
-                                 (i32 IntRegs:$subend)),
-                            (add (i32 IntRegs:$base), u6_1ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memh(Rs+#u6:1) &= Rt
-let AddedComplexity = 30 in
-def MEMh_ANDr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_1Imm:$offset, IntRegs:$andend),
-            "memh($base+#$offset) += $andend",
-            [(truncstorei16 (and (sextloadi16 (add (i32 IntRegs:$base),
-                                                   u6_1ImmPred:$offset)),
-                                 (i32 IntRegs:$andend)),
-                            (add (i32 IntRegs:$base), u6_1ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memh(Rs+#u6:1) |= Rt
-let AddedComplexity = 30 in
-def MEMh_ORr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_1Imm:$offset, IntRegs:$orend),
-            "memh($base+#$offset) |= $orend",
-            [(truncstorei16 (or (sextloadi16 (add (i32 IntRegs:$base),
-                                              u6_1ImmPred:$offset)),
-                             (i32 IntRegs:$orend)),
-                            (add (i32 IntRegs:$base), u6_1ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// MEMh_ADDSUBi_V4:
-//   Pseudo operation for MEMh_ADDi_V4 and MEMh_SUBi_V4
-//   a later pass will change it to the right pattern.
-let AddedComplexity = 30 in
-def MEMh_ADDSUBi_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, m6Imm:$addend),
-            "Error; should not emit",
-            [(truncstorei16 (add (sextloadi16 ADDRriU6_1:$addr),
-                                 m6ImmPred:$addend), ADDRriU6_1:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memh(Rs+#u6:1) += #U5
-let AddedComplexity = 30 in
-def MEMh_ADDi_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, u5Imm:$addend),
-            "memh($addr) += $addend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
+multiclass MemOpi_u5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
+  // Half Word
+  defm : MemOpi_u5ALUOp <ldOpHalf, truncstorei16, u6_1ExtPred,
+                         MemOPh_ADDi_V4, MemOPh_SUBi_V4>;
+  // Byte
+  defm : MemOpi_u5ALUOp <ldOpByte, truncstorei8, u6ExtPred,
+                         MemOPb_ADDi_V4, MemOPb_SUBi_V4>;
+}
 
-// memh(Rs+#u6:1) -= #U5
-let AddedComplexity = 30 in
-def MEMh_SUBi_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, u5Imm:$subend),
-            "memh($addr) -= $subend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memh(Rs+#u6:1) += Rt
-let AddedComplexity = 30 in
-def MEMh_ADDr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$addend),
-            "memh($addr) += $addend",
-            [(truncstorei16 (add (sextloadi16 ADDRriU6_1:$addr),
-                                 (i32 IntRegs:$addend)), ADDRriU6_1:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memh(Rs+#u6:1) -= Rt
-let AddedComplexity = 30 in
-def MEMh_SUBr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$subend),
-            "memh($addr) -= $subend",
-            [(truncstorei16 (sub (sextloadi16 ADDRriU6_1:$addr),
-                                 (i32 IntRegs:$subend)), ADDRriU6_1:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memh(Rs+#u6:1) &= Rt
-let AddedComplexity = 30 in
-def MEMh_ANDr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$andend),
-            "memh($addr) &= $andend",
-            [(truncstorei16 (and (sextloadi16 ADDRriU6_1:$addr),
-                                 (i32 IntRegs:$andend)), ADDRriU6_1:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memh(Rs+#u6:1) |= Rt
-let AddedComplexity = 30 in
-def MEMh_ORr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$orend),
-            "memh($addr) |= $orend",
-            [(truncstorei16 (or (sextloadi16 ADDRriU6_1:$addr),
-                                (i32 IntRegs:$orend)), ADDRriU6_1:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
+let Predicates = [HasV4T, UseMEMOP] in {
+  defm : MemOpi_u5ExtType<zextloadi8, zextloadi16>; // zero extend
+  defm : MemOpi_u5ExtType<sextloadi8, sextloadi16>; // sign extend
+  defm : MemOpi_u5ExtType<extloadi8,  extloadi16>;  // any extend
 
+  // Word
+  defm : MemOpi_u5ALUOp <load, store, u6_2ExtPred, MemOPw_ADDi_V4,
+                         MemOPw_SUBi_V4>;
+}
 
 //===----------------------------------------------------------------------===//
-// MEMOP: Byte
-//
-//  Implemented:
-//     MEMb_ADDi_indexed_V4  : memb(Rs+#u6:0)+=#U5
-//     MEMb_SUBi_indexed_V4  : memb(Rs+#u6:0)-=#U5
-//     MEMb_ADDr_indexed_V4  : memb(Rs+#u6:0)+=Rt
-//     MEMb_SUBr_indexed_V4  : memb(Rs+#u6:0)-=Rt
-//     MEMb_CLRr_indexed_V4  : memb(Rs+#u6:0)&=Rt
-//     MEMb_SETr_indexed_V4  : memb(Rs+#u6:0)|=Rt
-//     MEMb_ADDi_V4          : memb(Rs+#u6:0)+=#U5
-//     MEMb_SUBi_V4          : memb(Rs+#u6:0)-=#U5
-//     MEMb_ADDr_V4          : memb(Rs+#u6:0)+=Rt
-//     MEMb_SUBr_V4          : memb(Rs+#u6:0)-=Rt
-//     MEMb_CLRr_V4          : memb(Rs+#u6:0)&=Rt
-//     MEMb_SETr_V4          : memb(Rs+#u6:0)|=Rt
-//
-//   Not implemented:
-//     MEMb_CLRi_indexed_V4  : memb(Rs+#u6:0)=clrbit(#U5)
-//     MEMb_SETi_indexed_V4  : memb(Rs+#u6:0)=setbit(#U5)
-//     MEMb_CLRi_V4          : memb(Rs+#u6:0)=clrbit(#U5)
-//     MEMb_SETi_V4          : memb(Rs+#u6:0)=setbit(#U5)
+// multiclass to define 'Def Pats' for ALU operations on the memory.
+// Here value used for the ALU operation is a negative value.
+// mem[bh](Rs+#0) += #m5
+// mem[bh](Rs+#u6) += #m5
 //===----------------------------------------------------------------------===//
 
+multiclass MemOpi_m5Pats <PatFrag ldOp, PatFrag stOp, PatLeaf extPred,
+                          PatLeaf immPred, ComplexPattern addrPred,
+                          SDNodeXForm xformFunc, InstHexagon MI> {
+  let AddedComplexity = 190 in
+  def : Pat <(stOp (add (ldOp IntRegs:$addr), immPred:$subend),
+                   IntRegs:$addr),
+             (MI IntRegs:$addr, #0, (xformFunc immPred:$subend) )>;
+
+  let AddedComplexity = 195 in
+  def : Pat<(stOp (add (ldOp (add IntRegs:$base, extPred:$offset)),
+                       immPred:$subend),
+                  (add IntRegs:$base, extPred:$offset)),
+            (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$subend))>;
+}
 
-// MEMb_ADDSUBi_indexed_V4:
-//   Pseudo operation for MEMb_ADDi_indexed_V4 and
-//   MEMb_SUBi_indexed_V4 a later pass will change it
-//   to the corresponding pattern.
-let AddedComplexity = 30 in
-def MEMb_ADDSUBi_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_0Imm:$offset, m6Imm:$addend),
-            "Error; should not emit",
-            [(truncstorei8 (add (sextloadi8 (add (i32 IntRegs:$base),
-                                                 u6_0ImmPred:$offset)),
-                                m6ImmPred:$addend),
-                           (add (i32 IntRegs:$base), u6_0ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memb(Rs+#u6:0) += #U5
-let AddedComplexity = 30 in
-def MEMb_ADDi_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_0Imm:$offset, u5Imm:$addend),
-            "memb($base+#$offset) += $addend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
+multiclass MemOpi_m5ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
+  // Half Word
+  defm : MemOpi_m5Pats <ldOpHalf, truncstorei16, u6_1ExtPred, m5HImmPred,
+                        ADDRriU6_1, MEMOPIMM_HALF, MemOPh_SUBi_V4>;
+  // Byte
+  defm : MemOpi_m5Pats <ldOpByte, truncstorei8, u6ExtPred, m5BImmPred,
+                        ADDRriU6_0, MEMOPIMM_BYTE, MemOPb_SUBi_V4>;
+}
 
-// memb(Rs+#u6:0) -= #U5
-let AddedComplexity = 30 in
-def MEMb_SUBi_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_0Imm:$offset, u5Imm:$subend),
-            "memb($base+#$offset) -= $subend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memb(Rs+#u6:0) += Rt
-let AddedComplexity = 30 in
-def MEMb_ADDr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_0Imm:$offset, IntRegs:$addend),
-            "memb($base+#$offset) += $addend",
-            [(truncstorei8 (add (sextloadi8 (add (i32 IntRegs:$base),
-                                                 u6_0ImmPred:$offset)),
-                                (i32 IntRegs:$addend)),
-                           (add (i32 IntRegs:$base), u6_0ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memb(Rs+#u6:0) -= Rt
-let AddedComplexity = 30 in
-def MEMb_SUBr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_0Imm:$offset, IntRegs:$subend),
-            "memb($base+#$offset) -= $subend",
-            [(truncstorei8 (sub (sextloadi8 (add (i32 IntRegs:$base),
-                                                 u6_0ImmPred:$offset)),
-                                (i32 IntRegs:$subend)),
-                           (add (i32 IntRegs:$base), u6_0ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memb(Rs+#u6:0) &= Rt
-let AddedComplexity = 30 in
-def MEMb_ANDr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_0Imm:$offset, IntRegs:$andend),
-            "memb($base+#$offset) += $andend",
-            [(truncstorei8 (and (sextloadi8 (add (i32 IntRegs:$base),
-                                                 u6_0ImmPred:$offset)),
-                                (i32 IntRegs:$andend)),
-                           (add (i32 IntRegs:$base), u6_0ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memb(Rs+#u6:0) |= Rt
-let AddedComplexity = 30 in
-def MEMb_ORr_indexed_MEM_V4 : MEMInst_V4<(outs),
-            (ins IntRegs:$base, u6_0Imm:$offset, IntRegs:$orend),
-            "memb($base+#$offset) |= $orend",
-            [(truncstorei8 (or (sextloadi8 (add (i32 IntRegs:$base),
-                                                u6_0ImmPred:$offset)),
-                               (i32 IntRegs:$orend)),
-                           (add (i32 IntRegs:$base), u6_0ImmPred:$offset))]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// MEMb_ADDSUBi_V4:
-//   Pseudo operation for MEMb_ADDi_V4 and MEMb_SUBi_V4
-//   a later pass will change it to the right pattern.
-let AddedComplexity = 30 in
-def MEMb_ADDSUBi_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, m6Imm:$addend),
-            "Error; should not emit",
-            [(truncstorei8 (add (sextloadi8 ADDRriU6_0:$addr),
-                                m6ImmPred:$addend), ADDRriU6_0:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memb(Rs+#u6:0) += #U5
-let AddedComplexity = 30 in
-def MEMb_ADDi_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, u5Imm:$addend),
-            "memb($addr) += $addend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
+let Predicates = [HasV4T, UseMEMOP] in {
+  defm : MemOpi_m5ExtType<zextloadi8, zextloadi16>; // zero extend
+  defm : MemOpi_m5ExtType<sextloadi8, sextloadi16>; // sign extend
+  defm : MemOpi_m5ExtType<extloadi8,  extloadi16>;  // any extend
 
-// memb(Rs+#u6:0) -= #U5
-let AddedComplexity = 30 in
-def MEMb_SUBi_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, u5Imm:$subend),
-            "memb($addr) -= $subend",
-            []>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memb(Rs+#u6:0) += Rt
-let AddedComplexity = 30 in
-def MEMb_ADDr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$addend),
-            "memb($addr) += $addend",
-            [(truncstorei8 (add (sextloadi8 ADDRriU6_0:$addr),
-                                (i32 IntRegs:$addend)), ADDRriU6_0:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memb(Rs+#u6:0) -= Rt
-let AddedComplexity = 30 in
-def MEMb_SUBr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$subend),
-            "memb($addr) -= $subend",
-            [(truncstorei8 (sub (sextloadi8 ADDRriU6_0:$addr),
-                                (i32 IntRegs:$subend)), ADDRriU6_0:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memb(Rs+#u6:0) &= Rt
-let AddedComplexity = 30 in
-def MEMb_ANDr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$andend),
-            "memb($addr) &= $andend",
-            [(truncstorei8 (and (sextloadi8 ADDRriU6_0:$addr),
-                                (i32 IntRegs:$andend)), ADDRriU6_0:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
-
-// memb(Rs+#u6:0) |= Rt
-let AddedComplexity = 30 in
-def MEMb_ORr_MEM_V4 : MEMInst_V4<(outs),
-            (ins MEMri:$addr, IntRegs:$orend),
-            "memb($addr) |= $orend",
-            [(truncstorei8 (or (sextloadi8 ADDRriU6_0:$addr),
-                               (i32 IntRegs:$orend)), ADDRriU6_0:$addr)]>,
-            Requires<[HasV4T, UseMEMOP]>;
+  // Word
+  defm : MemOpi_m5Pats <load, store, u6_2ExtPred, m5ImmPred,
+                          ADDRriU6_2, MEMOPIMM, MemOPw_SUBi_V4>;
+}
 
+//===----------------------------------------------------------------------===//
+// Multiclass to define 'def Pats' for bit operations on the memory.
+// mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
+// mem[bhw](Rs+#u6) = [clrbit|setbit](#U5)
+//===----------------------------------------------------------------------===//
+
+multiclass MemOpi_bitPats <PatFrag ldOp, PatFrag stOp, PatLeaf immPred,
+                     PatLeaf extPred, ComplexPattern addrPred,
+                     SDNodeXForm xformFunc, InstHexagon MI, SDNode OpNode> {
+
+  // mem[bhw](Rs+#u6:[012]) = [clrbit|setbit](#U5)
+  let AddedComplexity = 250 in
+  def : Pat<(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
+                          immPred:$bitend),
+                  (add IntRegs:$base, extPred:$offset)),
+            (MI IntRegs:$base, extPred:$offset, (xformFunc immPred:$bitend))>;
+
+  // mem[bhw](Rs+#0) = [clrbit|setbit](#U5)
+  let AddedComplexity = 225 in
+  def : Pat <(stOp (OpNode (ldOp addrPred:$addr), immPred:$bitend),
+                   addrPred:$addr),
+             (MI IntRegs:$addr, #0, (xformFunc immPred:$bitend))>;
+}
+
+multiclass MemOpi_bitExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
+  // Byte - clrbit
+  defm : MemOpi_bitPats<ldOpByte, truncstorei8, Clr3ImmPred, u6ExtPred,
+                       ADDRriU6_0, CLRMEMIMM_BYTE, MemOPb_CLRBITi_V4, and>;
+  // Byte - setbit
+  defm : MemOpi_bitPats<ldOpByte, truncstorei8, Set3ImmPred,  u6ExtPred,
+                       ADDRriU6_0, SETMEMIMM_BYTE, MemOPb_SETBITi_V4, or>;
+  // Half Word - clrbit
+  defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Clr4ImmPred, u6_1ExtPred,
+                       ADDRriU6_1, CLRMEMIMM_SHORT, MemOPh_CLRBITi_V4, and>;
+  // Half Word - setbit
+  defm : MemOpi_bitPats<ldOpHalf, truncstorei16, Set4ImmPred, u6_1ExtPred,
+                       ADDRriU6_1, SETMEMIMM_SHORT, MemOPh_SETBITi_V4, or>;
+}
+
+let Predicates = [HasV4T, UseMEMOP] in {
+  // mem[bh](Rs+#0) = [clrbit|setbit](#U5)
+  // mem[bh](Rs+#u6:[01]) = [clrbit|setbit](#U5)
+  defm : MemOpi_bitExtType<zextloadi8, zextloadi16>; // zero extend
+  defm : MemOpi_bitExtType<sextloadi8, sextloadi16>; // sign extend
+  defm : MemOpi_bitExtType<extloadi8,  extloadi16>;  // any extend
+
+  // memw(Rs+#0) = [clrbit|setbit](#U5)
+  // memw(Rs+#u6:2) = [clrbit|setbit](#U5)
+  defm : MemOpi_bitPats<load, store, Clr5ImmPred, u6_2ExtPred, ADDRriU6_2,
+                       CLRMEMIMM, MemOPw_CLRBITi_V4, and>;
+  defm : MemOpi_bitPats<load, store, Set5ImmPred, u6_2ExtPred, ADDRriU6_2,
+                       SETMEMIMM, MemOPw_SETBITi_V4, or>;
+}
+
+//===----------------------------------------------------------------------===//
+// Multiclass to define 'def Pats' for ALU operations on the memory
+// where addend is a register.
+// mem[bhw](Rs+#0) [+-&|]= Rt
+// mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
+//===----------------------------------------------------------------------===//
+
+multiclass MemOpr_Pats <PatFrag ldOp, PatFrag stOp, ComplexPattern addrPred,
+                     PatLeaf extPred, InstHexagon MI, SDNode OpNode> {
+  let AddedComplexity = 141 in
+  // mem[bhw](Rs+#0) [+-&|]= Rt
+  def : Pat <(stOp (OpNode (ldOp addrPred:$addr), (i32 IntRegs:$addend)),
+                   addrPred:$addr),
+             (MI IntRegs:$addr, #0, (i32 IntRegs:$addend) )>;
+
+  // mem[bhw](Rs+#U6:[012]) [+-&|]= Rt
+  let AddedComplexity = 150 in
+  def : Pat <(stOp (OpNode (ldOp (add IntRegs:$base, extPred:$offset)),
+                           (i32 IntRegs:$orend)),
+                   (add IntRegs:$base, extPred:$offset)),
+             (MI IntRegs:$base, extPred:$offset, (i32 IntRegs:$orend) )>;
+}
+
+multiclass MemOPr_ALUOp<PatFrag ldOp, PatFrag stOp,
+                        ComplexPattern addrPred, PatLeaf extPred,
+                        InstHexagon addMI, InstHexagon subMI,
+                        InstHexagon andMI, InstHexagon orMI > {
+
+  defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, addMI, add>;
+  defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, subMI, sub>;
+  defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, andMI, and>;
+  defm : MemOpr_Pats <ldOp, stOp, addrPred, extPred, orMI,  or>;
+}
+
+multiclass MemOPr_ExtType<PatFrag ldOpByte, PatFrag ldOpHalf > {
+  // Half Word
+  defm : MemOPr_ALUOp <ldOpHalf, truncstorei16, ADDRriU6_1, u6_1ExtPred,
+                       MemOPh_ADDr_V4, MemOPh_SUBr_V4,
+                       MemOPh_ANDr_V4, MemOPh_ORr_V4>;
+  // Byte
+  defm : MemOPr_ALUOp <ldOpByte, truncstorei8, ADDRriU6_0, u6ExtPred,
+                       MemOPb_ADDr_V4, MemOPb_SUBr_V4,
+                       MemOPb_ANDr_V4, MemOPb_ORr_V4>;
+}
+
+// Define 'def Pats' for MemOps with register addend.
+let Predicates = [HasV4T, UseMEMOP] in {
+  // Byte, Half Word
+  defm : MemOPr_ExtType<zextloadi8, zextloadi16>; // zero extend
+  defm : MemOPr_ExtType<sextloadi8, sextloadi16>; // sign extend
+  defm : MemOPr_ExtType<extloadi8,  extloadi16>;  // any extend
+  // Word
+  defm : MemOPr_ALUOp <load, store, ADDRriU6_2, u6_2ExtPred, MemOPw_ADDr_V4,
+                       MemOPw_SUBr_V4, MemOPw_ANDr_V4, MemOPw_ORr_V4 >;
+}
 
 //===----------------------------------------------------------------------===//
 // XTYPE/PRED +
@@ -4992,7 +2222,61 @@ def MEMb_ORr_MEM_V4 : MEMInst_V4<(outs),
 // incorrect code for negative numbers.
 // Pd=cmpb.eq(Rs,#u8)
 
-let isCompare = 1 in
+// p=!cmp.eq(r1,r2)
+let isCompare = 1, validSubTargets = HasV4SubT in
+def CMPnotEQ_rr : ALU32_rr<(outs PredRegs:$dst),
+                           (ins IntRegs:$src1, IntRegs:$src2),
+      "$dst = !cmp.eq($src1, $src2)",
+      [(set (i1 PredRegs:$dst),
+            (setne (i32 IntRegs:$src1), (i32 IntRegs:$src2)))]>,
+      Requires<[HasV4T]>;
+
+// p=!cmp.eq(r1,#s10)
+let isCompare = 1, validSubTargets = HasV4SubT in
+def CMPnotEQ_ri : ALU32_ri<(outs PredRegs:$dst),
+                           (ins IntRegs:$src1, s10Ext:$src2),
+      "$dst = !cmp.eq($src1, #$src2)",
+      [(set (i1 PredRegs:$dst),
+            (setne (i32 IntRegs:$src1), s10ImmPred:$src2))]>,
+      Requires<[HasV4T]>;
+
+// p=!cmp.gt(r1,r2)
+let isCompare = 1, validSubTargets = HasV4SubT in
+def CMPnotGT_rr : ALU32_rr<(outs PredRegs:$dst),
+                           (ins IntRegs:$src1, IntRegs:$src2),
+      "$dst = !cmp.gt($src1, $src2)",
+      [(set (i1 PredRegs:$dst),
+            (not (setgt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>,
+      Requires<[HasV4T]>;
+
+// p=!cmp.gt(r1,#s10)
+let isCompare = 1, validSubTargets = HasV4SubT in
+def CMPnotGT_ri : ALU32_ri<(outs PredRegs:$dst),
+                           (ins IntRegs:$src1, s10Ext:$src2),
+      "$dst = !cmp.gt($src1, #$src2)",
+      [(set (i1 PredRegs:$dst),
+            (not (setgt (i32 IntRegs:$src1), s10ImmPred:$src2)))]>,
+      Requires<[HasV4T]>;
+
+// p=!cmp.gtu(r1,r2)
+let isCompare = 1, validSubTargets = HasV4SubT in
+def CMPnotGTU_rr : ALU32_rr<(outs PredRegs:$dst),
+                            (ins IntRegs:$src1, IntRegs:$src2),
+      "$dst = !cmp.gtu($src1, $src2)",
+      [(set (i1 PredRegs:$dst),
+            (not (setugt (i32 IntRegs:$src1), (i32 IntRegs:$src2))))]>,
+      Requires<[HasV4T]>;
+
+// p=!cmp.gtu(r1,#u9)
+let isCompare = 1, validSubTargets = HasV4SubT in
+def CMPnotGTU_ri : ALU32_ri<(outs PredRegs:$dst),
+                            (ins IntRegs:$src1, u9Ext:$src2),
+      "$dst = !cmp.gtu($src1, #$src2)",
+      [(set (i1 PredRegs:$dst),
+            (not (setugt (i32 IntRegs:$src1), u9ImmPred:$src2)))]>,
+      Requires<[HasV4T]>;
+
+let isCompare = 1, validSubTargets = HasV4SubT in
 def CMPbEQri_V4 : MInst<(outs PredRegs:$dst),
             (ins IntRegs:$src1, u8Imm:$src2),
             "$dst = cmpb.eq($src1, #$src2)",
@@ -5000,8 +2284,14 @@ def CMPbEQri_V4 : MInst<(outs PredRegs:$dst),
                   (seteq (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2))]>,
             Requires<[HasV4T]>;
 
+def : Pat <(brcond (i1 (setne (and (i32 IntRegs:$src1), 255), u8ImmPred:$src2)),
+                       bb:$offset),
+      (JMP_cNot (CMPbEQri_V4 (i32 IntRegs:$src1), u8ImmPred:$src2),
+                bb:$offset)>,
+      Requires<[HasV4T]>;
+
 // Pd=cmpb.eq(Rs,Rt)
-let isCompare = 1 in
+let isCompare = 1, validSubTargets = HasV4SubT in
 def CMPbEQrr_ubub_V4 : MInst<(outs PredRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2),
             "$dst = cmpb.eq($src1, $src2)",
@@ -5011,7 +2301,7 @@ def CMPbEQrr_ubub_V4 : MInst<(outs PredRegs:$dst),
             Requires<[HasV4T]>;
 
 // Pd=cmpb.eq(Rs,Rt)
-let isCompare = 1 in
+let isCompare = 1, validSubTargets = HasV4SubT in
 def CMPbEQrr_sbsb_V4 : MInst<(outs PredRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2),
             "$dst = cmpb.eq($src1, $src2)",
@@ -5021,7 +2311,7 @@ def CMPbEQrr_sbsb_V4 : MInst<(outs PredRegs:$dst),
             Requires<[HasV4T]>;
 
 // Pd=cmpb.gt(Rs,Rt)
-let isCompare = 1 in
+let isCompare = 1, validSubTargets = HasV4SubT in
 def CMPbGTrr_V4 : MInst<(outs PredRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2),
             "$dst = cmpb.gt($src1, $src2)",
@@ -5031,29 +2321,237 @@ def CMPbGTrr_V4 : MInst<(outs PredRegs:$dst),
             Requires<[HasV4T]>;
 
 // Pd=cmpb.gtu(Rs,#u7)
-let isCompare = 1 in
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
+isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU", InputType = "imm" in
 def CMPbGTUri_V4 : MInst<(outs PredRegs:$dst),
-            (ins IntRegs:$src1, u7Imm:$src2),
+            (ins IntRegs:$src1, u7Ext:$src2),
             "$dst = cmpb.gtu($src1, #$src2)",
             [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
-                                              u7ImmPred:$src2))]>,
-            Requires<[HasV4T]>;
+                                              u7ExtPred:$src2))]>,
+            Requires<[HasV4T]>, ImmRegRel;
+
+// SDNode for converting immediate C to C-1.
+def DEC_CONST_BYTE : SDNodeXForm<imm, [{
+   // Return the byte immediate const-1 as an SDNode.
+   int32_t imm = N->getSExtValue();
+   return XformU7ToU7M1Imm(imm);
+}]>;
+
+// For the sequence
+//   zext( seteq ( and(Rs, 255), u8))
+// Generate
+//   Pd=cmpb.eq(Rs, #u8)
+//   if (Pd.new) Rd=#1
+//   if (!Pd.new) Rd=#0
+def : Pat <(i32 (zext (i1 (seteq (i32 (and (i32 IntRegs:$Rs), 255)),
+                                           u8ExtPred:$u8)))),
+           (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs),
+                                                 (u8ExtPred:$u8))),
+                                1, 0))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setne ( and(Rs, 255), u8))
+// Generate
+//   Pd=cmpb.eq(Rs, #u8)
+//   if (Pd.new) Rd=#0
+//   if (!Pd.new) Rd=#1
+def : Pat <(i32 (zext (i1 (setne (i32 (and (i32 IntRegs:$Rs), 255)),
+                                           u8ExtPred:$u8)))),
+           (i32 (TFR_condset_ii (i1 (CMPbEQri_V4 (i32 IntRegs:$Rs),
+                                                 (u8ExtPred:$u8))),
+                                0, 1))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( seteq (Rs, and(Rt, 255)))
+// Generate
+//   Pd=cmpb.eq(Rs, Rt)
+//   if (Pd.new) Rd=#1
+//   if (!Pd.new) Rd=#0
+def : Pat <(i32 (zext (i1 (seteq (i32 IntRegs:$Rt),
+                                 (i32 (and (i32 IntRegs:$Rs), 255)))))),
+           (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
+                                                      (i32 IntRegs:$Rt))),
+                                1, 0))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setne (Rs, and(Rt, 255)))
+// Generate
+//   Pd=cmpb.eq(Rs, Rt)
+//   if (Pd.new) Rd=#0
+//   if (!Pd.new) Rd=#1
+def : Pat <(i32 (zext (i1 (setne (i32 IntRegs:$Rt),
+                                 (i32 (and (i32 IntRegs:$Rs), 255)))))),
+           (i32 (TFR_condset_ii (i1 (CMPbEQrr_ubub_V4 (i32 IntRegs:$Rs),
+                                                      (i32 IntRegs:$Rt))),
+                                0, 1))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setugt ( and(Rs, 255), u8))
+// Generate
+//   Pd=cmpb.gtu(Rs, #u8)
+//   if (Pd.new) Rd=#1
+//   if (!Pd.new) Rd=#0
+def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 255)),
+                                            u8ExtPred:$u8)))),
+           (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
+                                                  (u8ExtPred:$u8))),
+                                1, 0))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setugt ( and(Rs, 254), u8))
+// Generate
+//   Pd=cmpb.gtu(Rs, #u8)
+//   if (Pd.new) Rd=#1
+//   if (!Pd.new) Rd=#0
+def : Pat <(i32 (zext (i1 (setugt (i32 (and (i32 IntRegs:$Rs), 254)),
+                                            u8ExtPred:$u8)))),
+           (i32 (TFR_condset_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$Rs),
+                                                  (u8ExtPred:$u8))),
+                                1, 0))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setult ( Rs, Rt))
+// Generate
+//   Pd=cmp.ltu(Rs, Rt)
+//   if (Pd.new) Rd=#1
+//   if (!Pd.new) Rd=#0
+// cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
+def : Pat <(i32 (zext (i1 (setult (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
+           (i32 (TFR_condset_ii (i1 (CMPGTUrr (i32 IntRegs:$Rt),
+                                              (i32 IntRegs:$Rs))),
+                                1, 0))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setlt ( Rs, Rt))
+// Generate
+//   Pd=cmp.lt(Rs, Rt)
+//   if (Pd.new) Rd=#1
+//   if (!Pd.new) Rd=#0
+// cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
+def : Pat <(i32 (zext (i1 (setlt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
+           (i32 (TFR_condset_ii (i1 (CMPGTrr (i32 IntRegs:$Rt),
+                                             (i32 IntRegs:$Rs))),
+                                1, 0))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setugt ( Rs, Rt))
+// Generate
+//   Pd=cmp.gtu(Rs, Rt)
+//   if (Pd.new) Rd=#1
+//   if (!Pd.new) Rd=#0
+def : Pat <(i32 (zext (i1 (setugt (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
+           (i32 (TFR_condset_ii (i1 (CMPGTUrr (i32 IntRegs:$Rs),
+                                              (i32 IntRegs:$Rt))),
+                                1, 0))>,
+           Requires<[HasV4T]>;
+
+// This pattern interefers with coremark performance, not implementing at this
+// time.
+// For the sequence
+//   zext( setgt ( Rs, Rt))
+// Generate
+//   Pd=cmp.gt(Rs, Rt)
+//   if (Pd.new) Rd=#1
+//   if (!Pd.new) Rd=#0
+
+// For the sequence
+//   zext( setuge ( Rs, Rt))
+// Generate
+//   Pd=cmp.ltu(Rs, Rt)
+//   if (Pd.new) Rd=#0
+//   if (!Pd.new) Rd=#1
+// cmp.ltu(Rs, Rt) -> cmp.gtu(Rt, Rs)
+def : Pat <(i32 (zext (i1 (setuge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
+           (i32 (TFR_condset_ii (i1 (CMPGTUrr (i32 IntRegs:$Rt),
+                                              (i32 IntRegs:$Rs))),
+                                0, 1))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setge ( Rs, Rt))
+// Generate
+//   Pd=cmp.lt(Rs, Rt)
+//   if (Pd.new) Rd=#0
+//   if (!Pd.new) Rd=#1
+// cmp.lt(Rs, Rt) -> cmp.gt(Rt, Rs)
+def : Pat <(i32 (zext (i1 (setge (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
+           (i32 (TFR_condset_ii (i1 (CMPGTrr (i32 IntRegs:$Rt),
+                                             (i32 IntRegs:$Rs))),
+                                0, 1))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setule ( Rs, Rt))
+// Generate
+//   Pd=cmp.gtu(Rs, Rt)
+//   if (Pd.new) Rd=#0
+//   if (!Pd.new) Rd=#1
+def : Pat <(i32 (zext (i1 (setule (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
+           (i32 (TFR_condset_ii (i1 (CMPGTUrr (i32 IntRegs:$Rs),
+                                              (i32 IntRegs:$Rt))),
+                                0, 1))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setle ( Rs, Rt))
+// Generate
+//   Pd=cmp.gt(Rs, Rt)
+//   if (Pd.new) Rd=#0
+//   if (!Pd.new) Rd=#1
+def : Pat <(i32 (zext (i1 (setle (i32 IntRegs:$Rs), (i32 IntRegs:$Rt))))),
+           (i32 (TFR_condset_ii (i1 (CMPGTrr (i32 IntRegs:$Rs),
+                                             (i32 IntRegs:$Rt))),
+                                0, 1))>,
+           Requires<[HasV4T]>;
+
+// For the sequence
+//   zext( setult ( and(Rs, 255), u8))
+// Use the isdigit transformation below
+
+// Generate code of the form 'mux_ii(cmpbgtu(Rdd, C-1),0,1)'
+// for C code of the form r = ((c>='0') & (c<='9')) ? 1 : 0;.
+// The isdigit transformation relies on two 'clever' aspects:
+// 1) The data type is unsigned which allows us to eliminate a zero test after
+//    biasing the expression by 48. We are depending on the representation of
+//    the unsigned types, and semantics.
+// 2) The front end has converted <= 9 into < 10 on entry to LLVM
+//
+// For the C code:
+//   retval = ((c>='0') & (c<='9')) ? 1 : 0;
+// The code is transformed upstream of llvm into
+//   retval = (c-48) < 10 ? 1 : 0;
+let AddedComplexity = 139 in
+def : Pat <(i32 (zext (i1 (setult (i32 (and (i32 IntRegs:$src1), 255)),
+                                  u7StrictPosImmPred:$src2)))),
+  (i32 (MUX_ii (i1 (CMPbGTUri_V4 (i32 IntRegs:$src1),
+                                 (DEC_CONST_BYTE u7StrictPosImmPred:$src2))),
+                   0, 1))>,
+                   Requires<[HasV4T]>;
 
 // Pd=cmpb.gtu(Rs,Rt)
-let isCompare = 1 in
+let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPbGTU",
+InputType = "reg" in
 def CMPbGTUrr_V4 : MInst<(outs PredRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2),
             "$dst = cmpb.gtu($src1, $src2)",
             [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 255),
                                              (and (i32 IntRegs:$src2), 255)))]>,
-            Requires<[HasV4T]>;
+            Requires<[HasV4T]>, ImmRegRel;
 
 // Following instruction is not being extended as it results into the incorrect
 // code for negative numbers.
 
 // Signed half compare(.eq) ri.
 // Pd=cmph.eq(Rs,#s8)
-let isCompare = 1 in
+let isCompare = 1, validSubTargets = HasV4SubT in
 def CMPhEQri_V4 : MInst<(outs PredRegs:$dst),
             (ins IntRegs:$src1, s8Imm:$src2),
             "$dst = cmph.eq($src1, #$src2)",
@@ -5067,7 +2565,7 @@ def CMPhEQri_V4 : MInst<(outs PredRegs:$dst),
 //   r0=and(r0,#0xffff)
 //   p0=cmp.eq(r0,#0)
 // Pd=cmph.eq(Rs,Rt)
-let isCompare = 1 in
+let isCompare = 1, validSubTargets = HasV4SubT in
 def CMPhEQrr_xor_V4 : MInst<(outs PredRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2),
             "$dst = cmph.eq($src1, $src2)",
@@ -5082,7 +2580,7 @@ def CMPhEQrr_xor_V4 : MInst<(outs PredRegs:$dst),
 //   r1=asl(r1,16)
 //   p0=cmp.eq(r0,r1)
 // Pd=cmph.eq(Rs,Rt)
-let isCompare = 1 in
+let isCompare = 1, validSubTargets = HasV4SubT in
 def CMPhEQrr_shl_V4 : MInst<(outs PredRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2),
             "$dst = cmph.eq($src1, $src2)",
@@ -5096,19 +2594,20 @@ used in the cmph.gt instruction.
 // Signed half compare(.gt) ri.
 // Pd=cmph.gt(Rs,#s8)
 
-let isCompare = 1 in
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 1, opExtentBits = 8,
+isCompare = 1, validSubTargets = HasV4SubT in
 def CMPhGTri_V4 : MInst<(outs PredRegs:$dst),
-            (ins IntRegs:$src1, s8Imm:$src2),
+            (ins IntRegs:$src1, s8Ext:$src2),
             "$dst = cmph.gt($src1, #$src2)",
             [(set (i1 PredRegs:$dst),
                   (setgt (shl (i32 IntRegs:$src1), (i32 16)),
-                         s8ImmPred:$src2))]>,
+                         s8ExtPred:$src2))]>,
             Requires<[HasV4T]>;
 */
 
 // Signed half compare(.gt) rr.
 // Pd=cmph.gt(Rs,Rt)
-let isCompare = 1 in
+let isCompare = 1, validSubTargets = HasV4SubT in
 def CMPhGTrr_shl_V4 : MInst<(outs PredRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2),
             "$dst = cmph.gt($src1, $src2)",
@@ -5119,24 +2618,41 @@ def CMPhGTrr_shl_V4 : MInst<(outs PredRegs:$dst),
 
 // Unsigned half compare rr (.gtu).
 // Pd=cmph.gtu(Rs,Rt)
-let isCompare = 1 in
+let isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
+InputType = "reg" in
 def CMPhGTUrr_V4 : MInst<(outs PredRegs:$dst),
             (ins IntRegs:$src1, IntRegs:$src2),
             "$dst = cmph.gtu($src1, $src2)",
             [(set (i1 PredRegs:$dst),
                   (setugt (and (i32 IntRegs:$src1), 65535),
                           (and (i32 IntRegs:$src2), 65535)))]>,
-            Requires<[HasV4T]>;
+            Requires<[HasV4T]>, ImmRegRel;
 
 // Unsigned half compare ri (.gtu).
 // Pd=cmph.gtu(Rs,#u7)
-let isCompare = 1 in
+let isExtendable = 1, opExtendable = 2, isExtentSigned = 0, opExtentBits = 7,
+isCompare = 1, validSubTargets = HasV4SubT, CextOpcode = "CMPhGTU",
+InputType = "imm" in
 def CMPhGTUri_V4 : MInst<(outs PredRegs:$dst),
-            (ins IntRegs:$src1, u7Imm:$src2),
+            (ins IntRegs:$src1, u7Ext:$src2),
             "$dst = cmph.gtu($src1, #$src2)",
             [(set (i1 PredRegs:$dst), (setugt (and (i32 IntRegs:$src1), 65535),
-                                              u7ImmPred:$src2))]>,
-            Requires<[HasV4T]>;
+                                              u7ExtPred:$src2))]>,
+            Requires<[HasV4T]>, ImmRegRel;
+
+let validSubTargets = HasV4SubT in
+def NTSTBIT_rr : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, IntRegs:$src2),
+    "$dst = !tstbit($src1, $src2)",
+    [(set (i1 PredRegs:$dst),
+          (seteq (and (shl 1, (i32 IntRegs:$src2)), (i32 IntRegs:$src1)), 0))]>,
+    Requires<[HasV4T]>;
+
+let validSubTargets = HasV4SubT in
+def NTSTBIT_ri : SInst<(outs PredRegs:$dst), (ins IntRegs:$src1, u5Imm:$src2),
+    "$dst = !tstbit($src1, $src2)",
+    [(set (i1 PredRegs:$dst),
+          (seteq (and (shl 1, u5ImmPred:$src2), (i32 IntRegs:$src1)), 0))]>,
+    Requires<[HasV4T]>;
 
 //===----------------------------------------------------------------------===//
 // XTYPE/PRED -
@@ -5248,227 +2764,258 @@ let isReturn = 1, isTerminator = 1,
             Requires<[HasV4T]>;
 }
 
-
 // Load/Store with absolute addressing mode
 // memw(#u6)=Rt
 
-multiclass ST_abs<string OpcStr> {
-  let isPredicable = 1 in
-  def _abs_V4 : STInst2<(outs),
-            (ins globaladdress:$absaddr, IntRegs:$src),
-            !strconcat(OpcStr, "(##$absaddr) = $src"),
-            []>,
-            Requires<[HasV4T]>;
-
-  let isPredicated = 1 in
-  def _abs_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2),
-            !strconcat("if ($src1)",
-            !strconcat(OpcStr, "(##$absaddr) = $src2")),
-            []>,
-            Requires<[HasV4T]>;
-
-  let isPredicated = 1 in
-  def _abs_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2),
-            !strconcat("if (!$src1)",
-            !strconcat(OpcStr, "(##$absaddr) = $src2")),
+multiclass ST_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
+                           bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME#_V4 : STInst2<(outs),
+            (ins PredRegs:$src1, globaladdressExt:$absaddr, RC: $src2),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"(##$absaddr) = $src2",
             []>,
             Requires<[HasV4T]>;
+}
 
-  let isPredicated = 1 in
-  def _abs_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2),
-            !strconcat("if ($src1.new)",
-            !strconcat(OpcStr, "(##$absaddr) = $src2")),
-            []>,
-            Requires<[HasV4T]>;
+multiclass ST_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : ST_Abs_Predbase<mnemonic, RC, PredNot, 1>;
+  }
+}
 
-  let isPredicated = 1 in
-  def _abs_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2),
-            !strconcat("if (!$src1.new)",
-            !strconcat(OpcStr, "(##$absaddr) = $src2")),
+let isNVStorable = 1, isExtended = 1, neverHasSideEffects = 1 in
+multiclass ST_Abs<string mnemonic, string CextOp, RegisterClass RC> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
+    let opExtendable = 0, isPredicable = 1 in
+    def NAME#_V4 : STInst2<(outs),
+            (ins globaladdressExt:$absaddr, RC:$src),
+            mnemonic#"(##$absaddr) = $src",
             []>,
             Requires<[HasV4T]>;
 
-  def _abs_nv_V4 : STInst2<(outs),
-            (ins globaladdress:$absaddr, IntRegs:$src),
-            !strconcat(OpcStr, "(##$absaddr) = $src.new"),
-            []>,
-            Requires<[HasV4T]>;
+    let opExtendable = 1, isPredicated = 1 in {
+      defm Pt : ST_Abs_Pred<mnemonic, RC, 0>;
+      defm NotPt : ST_Abs_Pred<mnemonic, RC, 1>;
+    }
+  }
+}
 
-  let isPredicated = 1 in
-  def _abs_cPt_nv_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2),
-            !strconcat("if ($src1)",
-            !strconcat(OpcStr, "(##$absaddr) = $src2.new")),
+multiclass ST_Abs_Predbase_nv<string mnemonic, RegisterClass RC, bit isNot,
+                           bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME#_nv_V4 : NVInst_V4<(outs),
+            (ins PredRegs:$src1, globaladdressExt:$absaddr, RC: $src2),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#mnemonic#"(##$absaddr) = $src2.new",
             []>,
             Requires<[HasV4T]>;
+}
 
-  let isPredicated = 1 in
-  def _abs_cNotPt_nv_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2),
-            !strconcat("if (!$src1)",
-            !strconcat(OpcStr, "(##$absaddr) = $src2.new")),
-            []>,
-            Requires<[HasV4T]>;
+multiclass ST_Abs_Pred_nv<string mnemonic, RegisterClass RC, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : ST_Abs_Predbase_nv<mnemonic, RC, PredNot, 1>;
+  }
+}
 
-  let isPredicated = 1 in
-  def _abs_cdnPt_nv_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2),
-            !strconcat("if ($src1.new)",
-            !strconcat(OpcStr, "(##$absaddr) = $src2.new")),
+let mayStore = 1, isNVStore = 1, isExtended = 1, neverHasSideEffects = 1 in
+multiclass ST_Abs_nv<string mnemonic, string CextOp, RegisterClass RC> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
+    let opExtendable = 0, isPredicable = 1 in
+    def NAME#_nv_V4 : NVInst_V4<(outs),
+            (ins globaladdressExt:$absaddr, RC:$src),
+            mnemonic#"(##$absaddr) = $src.new",
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
-  def _abs_cdnNotPt_nv_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, globaladdress:$absaddr, IntRegs:$src2),
-            !strconcat("if (!$src1.new)",
-            !strconcat(OpcStr, "(##$absaddr) = $src2.new")),
-            []>,
-            Requires<[HasV4T]>;
+    let opExtendable = 1, isPredicated = 1 in {
+      defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
+      defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
+    }
+  }
 }
 
-let AddedComplexity = 30, isPredicable = 1 in
-def STrid_abs_V4 : STInst<(outs),
-          (ins globaladdress:$absaddr, DoubleRegs:$src),
-           "memd(##$absaddr) = $src",
-          [(store (i64 DoubleRegs:$src),
-                  (HexagonCONST32 tglobaladdr:$absaddr))]>,
-          Requires<[HasV4T]>;
-
-let AddedComplexity = 30, isPredicated = 1 in
-def STrid_abs_cPt_V4 : STInst2<(outs),
-          (ins PredRegs:$src1, globaladdress:$absaddr, DoubleRegs:$src2),
-          "if ($src1) memd(##$absaddr) = $src2",
-          []>,
-          Requires<[HasV4T]>;
-
-let AddedComplexity = 30, isPredicated = 1 in
-def STrid_abs_cNotPt_V4 : STInst2<(outs),
-          (ins PredRegs:$src1, globaladdress:$absaddr, DoubleRegs:$src2),
-          "if (!$src1) memd(##$absaddr) = $src2",
-          []>,
-          Requires<[HasV4T]>;
+let addrMode = Absolute in {
+    defm STrib_abs : ST_Abs<"memb", "STrib", IntRegs>,
+                     ST_Abs_nv<"memb", "STrib", IntRegs>, AddrModeRel;
 
-let AddedComplexity = 30, isPredicated = 1 in
-def STrid_abs_cdnPt_V4 : STInst2<(outs),
-          (ins PredRegs:$src1, globaladdress:$absaddr, DoubleRegs:$src2),
-          "if ($src1.new) memd(##$absaddr) = $src2",
-          []>,
-          Requires<[HasV4T]>;
+    defm STrih_abs : ST_Abs<"memh", "STrih", IntRegs>,
+                     ST_Abs_nv<"memh", "STrih", IntRegs>, AddrModeRel;
 
-let AddedComplexity = 30, isPredicated = 1 in
-def STrid_abs_cdnNotPt_V4 : STInst2<(outs),
-          (ins PredRegs:$src1, globaladdress:$absaddr, DoubleRegs:$src2),
-          "if (!$src1.new) memd(##$absaddr) = $src2",
-          []>,
-          Requires<[HasV4T]>;
+    defm STriw_abs : ST_Abs<"memw", "STriw", IntRegs>,
+                     ST_Abs_nv<"memw", "STriw", IntRegs>, AddrModeRel;
 
-defm STrib : ST_abs<"memb">;
-defm STrih : ST_abs<"memh">;
-defm STriw : ST_abs<"memw">;
+  let isNVStorable = 0 in
+    defm STrid_abs : ST_Abs<"memd", "STrid", DoubleRegs>, AddrModeRel;
+}
 
-let Predicates = [HasV4T], AddedComplexity  = 30 in
+let Predicates = [HasV4T], AddedComplexity = 30 in {
 def : Pat<(truncstorei8 (i32 IntRegs:$src1),
                         (HexagonCONST32 tglobaladdr:$absaddr)),
           (STrib_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
 
-let Predicates = [HasV4T], AddedComplexity  = 30 in
 def : Pat<(truncstorei16 (i32 IntRegs:$src1),
                           (HexagonCONST32 tglobaladdr:$absaddr)),
           (STrih_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
 
-let Predicates = [HasV4T], AddedComplexity  = 30 in
 def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32 tglobaladdr:$absaddr)),
           (STriw_abs_V4 tglobaladdr: $absaddr, IntRegs: $src1)>;
 
+def : Pat<(store (i64 DoubleRegs:$src1),
+                 (HexagonCONST32 tglobaladdr:$absaddr)),
+          (STrid_abs_V4 tglobaladdr: $absaddr, DoubleRegs: $src1)>;
+}
 
-multiclass LD_abs<string OpcStr> {
-  let isPredicable = 1 in
-  def _abs_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins globaladdress:$absaddr),
-            !strconcat("$dst = ", !strconcat(OpcStr, "(##$absaddr)")),
-            []>,
-            Requires<[HasV4T]>;
-
-  let isPredicated = 1 in
-  def _abs_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$absaddr),
-            !strconcat("if ($src1) $dst = ",
-            !strconcat(OpcStr, "(##$absaddr)")),
-            []>,
-            Requires<[HasV4T]>;
+//===----------------------------------------------------------------------===//
+// multiclass for store instructions with GP-relative addressing mode.
+// mem[bhwd](#global)=Rt
+// if ([!]Pv[.new]) mem[bhwd](##global) = Rt
+//===----------------------------------------------------------------------===//
+multiclass ST_GP<string mnemonic, string BaseOp, RegisterClass RC> {
+  let BaseOpcode = BaseOp, isPredicable = 1 in
+  def NAME#_V4 : STInst2<(outs),
+          (ins globaladdress:$global, RC:$src),
+          mnemonic#"(#$global) = $src",
+          []>;
+
+  // When GP-relative instructions are predicated, their addressing mode is
+  // changed to absolute and they are always constant extended.
+  let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
+  isPredicated = 1 in {
+    defm Pt : ST_Abs_Pred <mnemonic, RC, 0>;
+    defm NotPt : ST_Abs_Pred <mnemonic, RC, 1>;
+  }
+}
 
-  let isPredicated = 1 in
-  def _abs_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$absaddr),
-            !strconcat("if (!$src1) $dst = ",
-            !strconcat(OpcStr, "(##$absaddr)")),
-            []>,
-            Requires<[HasV4T]>;
+let mayStore = 1, isNVStore = 1 in
+multiclass ST_GP_nv<string mnemonic, string BaseOp, RegisterClass RC> {
+  let BaseOpcode = BaseOp, isPredicable = 1 in
+  def NAME#_nv_V4 : NVInst_V4<(outs),
+          (ins u0AlwaysExt:$global, RC:$src),
+          mnemonic#"(#$global) = $src.new",
+          []>,
+          Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
-  def _abs_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$absaddr),
-            !strconcat("if ($src1.new) $dst = ",
-            !strconcat(OpcStr, "(##$absaddr)")),
-            []>,
-            Requires<[HasV4T]>;
+  // When GP-relative instructions are predicated, their addressing mode is
+  // changed to absolute and they are always constant extended.
+  let BaseOpcode = BaseOp, isExtended = 1, opExtendable = 1,
+  isPredicated = 1 in {
+    defm Pt : ST_Abs_Pred_nv<mnemonic, RC, 0>;
+    defm NotPt : ST_Abs_Pred_nv<mnemonic, RC, 1>;
+  }
+}
 
-  let isPredicated = 1 in
-  def _abs_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, globaladdress:$absaddr),
-            !strconcat("if (!$src1.new) $dst = ",
-            !strconcat(OpcStr, "(##$absaddr)")),
-            []>,
-            Requires<[HasV4T]>;
+let validSubTargets = HasV4SubT,  validSubTargets = HasV4SubT in {
+defm STd_GP : ST_GP <"memd", "STd_GP", DoubleRegs>,
+              ST_GP_nv<"memd", "STd_GP", DoubleRegs>, NewValueRel ;
+defm STb_GP : ST_GP<"memb",  "STb_GP", IntRegs>,
+              ST_GP_nv<"memb", "STb_GP", IntRegs>, NewValueRel ;
+defm STh_GP : ST_GP<"memh",  "STh_GP", IntRegs>,
+              ST_GP_nv<"memh", "STh_GP", IntRegs>, NewValueRel ;
+defm STw_GP : ST_GP<"memw",  "STw_GP", IntRegs>,
+              ST_GP_nv<"memw", "STw_GP", IntRegs>, NewValueRel ;
 }
 
-let AddedComplexity = 30 in
-def LDrid_abs_V4 : LDInst<(outs DoubleRegs:$dst),
-          (ins globaladdress:$absaddr),
-          "$dst = memd(##$absaddr)",
-          [(set (i64 DoubleRegs:$dst),
-                (load (HexagonCONST32 tglobaladdr:$absaddr)))]>,
-          Requires<[HasV4T]>;
+// 64 bit atomic store
+def : Pat <(atomic_store_64 (HexagonCONST32_GP tglobaladdr:$global),
+                            (i64 DoubleRegs:$src1)),
+           (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>,
+           Requires<[HasV4T]>;
 
-let AddedComplexity = 30, isPredicated = 1 in
-def LDrid_abs_cPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-          (ins PredRegs:$src1, globaladdress:$absaddr),
-          "if ($src1) $dst = memd(##$absaddr)",
-          []>,
-          Requires<[HasV4T]>;
+// Map from store(globaladdress) -> memd(#foo)
+let AddedComplexity = 100 in
+def : Pat <(store (i64 DoubleRegs:$src1),
+                  (HexagonCONST32_GP tglobaladdr:$global)),
+           (STd_GP_V4 tglobaladdr:$global, (i64 DoubleRegs:$src1))>;
 
-let AddedComplexity = 30, isPredicated = 1 in
-def LDrid_abs_cNotPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-          (ins PredRegs:$src1, globaladdress:$absaddr),
-          "if (!$src1) $dst = memd(##$absaddr)",
-          []>,
-          Requires<[HasV4T]>;
+// 8 bit atomic store
+def : Pat < (atomic_store_8 (HexagonCONST32_GP tglobaladdr:$global),
+                            (i32 IntRegs:$src1)),
+            (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
 
-let AddedComplexity = 30, isPredicated = 1 in
-def LDrid_abs_cdnPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-          (ins PredRegs:$src1, globaladdress:$absaddr),
-          "if ($src1.new) $dst = memd(##$absaddr)",
-          []>,
-          Requires<[HasV4T]>;
+// Map from store(globaladdress) -> memb(#foo)
+let AddedComplexity = 100 in
+def : Pat<(truncstorei8 (i32 IntRegs:$src1),
+          (HexagonCONST32_GP tglobaladdr:$global)),
+          (STb_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
 
-let AddedComplexity = 30, isPredicated = 1 in
-def LDrid_abs_cdnNotPt_V4 : LDInst2<(outs DoubleRegs:$dst),
-          (ins PredRegs:$src1, globaladdress:$absaddr),
-          "if (!$src1.new) $dst = memd(##$absaddr)",
-          []>,
-          Requires<[HasV4T]>;
+// Map from "i1 = constant<-1>; memw(CONST32(#foo)) = i1"
+//       to "r0 = 1; memw(#foo) = r0"
+let AddedComplexity = 100 in
+def : Pat<(store (i1 -1), (HexagonCONST32_GP tglobaladdr:$global)),
+          (STb_GP_V4 tglobaladdr:$global, (TFRI 1))>;
+
+def : Pat<(atomic_store_16 (HexagonCONST32_GP tglobaladdr:$global),
+                           (i32 IntRegs:$src1)),
+          (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
+
+// Map from store(globaladdress) -> memh(#foo)
+let AddedComplexity = 100 in
+def : Pat<(truncstorei16 (i32 IntRegs:$src1),
+                         (HexagonCONST32_GP tglobaladdr:$global)),
+          (STh_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
+
+// 32 bit atomic store
+def : Pat<(atomic_store_32 (HexagonCONST32_GP tglobaladdr:$global),
+                           (i32 IntRegs:$src1)),
+          (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
+
+// Map from store(globaladdress) -> memw(#foo)
+let AddedComplexity = 100 in
+def : Pat<(store (i32 IntRegs:$src1), (HexagonCONST32_GP tglobaladdr:$global)),
+          (STw_GP_V4 tglobaladdr:$global, (i32 IntRegs:$src1))>;
+
+//===----------------------------------------------------------------------===//
+// Multiclass for the load instructions with absolute addressing mode.
+//===----------------------------------------------------------------------===//
+multiclass LD_Abs_Predbase<string mnemonic, RegisterClass RC, bit isNot,
+                           bit isPredNew> {
+  let PNewValue = !if(isPredNew, "new", "") in
+  def NAME : LDInst2<(outs RC:$dst),
+            (ins PredRegs:$src1, globaladdressExt:$absaddr),
+            !if(isNot, "if (!$src1", "if ($src1")#!if(isPredNew, ".new) ",
+            ") ")#"$dst = "#mnemonic#"(##$absaddr)",
+            []>,
+            Requires<[HasV4T]>;
+}
 
-defm LDrib : LD_abs<"memb">;
-defm LDriub : LD_abs<"memub">;
-defm LDrih : LD_abs<"memh">;
-defm LDriuh : LD_abs<"memuh">;
-defm LDriw : LD_abs<"memw">;
+multiclass LD_Abs_Pred<string mnemonic, RegisterClass RC, bit PredNot> {
+  let PredSense = !if(PredNot, "false", "true") in {
+    defm _c#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 0>;
+    // Predicate new
+    defm _cdn#NAME : LD_Abs_Predbase<mnemonic, RC, PredNot, 1>;
+  }
+}
+
+let isExtended = 1, neverHasSideEffects = 1 in
+multiclass LD_Abs<string mnemonic, string CextOp, RegisterClass RC> {
+  let CextOpcode = CextOp, BaseOpcode = CextOp#_abs in {
+    let  opExtendable = 1, isPredicable = 1 in
+    def NAME#_V4 : LDInst2<(outs RC:$dst),
+            (ins globaladdressExt:$absaddr),
+            "$dst = "#mnemonic#"(##$absaddr)",
+            []>,
+            Requires<[HasV4T]>;
 
+    let opExtendable = 2, isPredicated = 1 in {
+      defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
+      defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
+    }
+  }
+}
+
+let addrMode = Absolute in {
+    defm LDrib_abs  : LD_Abs<"memb", "LDrib", IntRegs>, AddrModeRel;
+    defm LDriub_abs : LD_Abs<"memub", "LDriub", IntRegs>, AddrModeRel;
+    defm LDrih_abs  : LD_Abs<"memh", "LDrih", IntRegs>, AddrModeRel;
+    defm LDriuh_abs : LD_Abs<"memuh", "LDriuh", IntRegs>, AddrModeRel;
+    defm LDriw_abs  : LD_Abs<"memw", "LDriw", IntRegs>, AddrModeRel;
+    defm LDrid_abs : LD_Abs<"memd",  "LDrid", DoubleRegs>, AddrModeRel;
+}
 
 let Predicates = [HasV4T], AddedComplexity  = 30 in
 def : Pat<(i32 (load (HexagonCONST32 tglobaladdr:$absaddr))),
@@ -5490,6 +3037,107 @@ let Predicates = [HasV4T], AddedComplexity=30 in
 def : Pat<(i32 (zextloadi16 (HexagonCONST32 tglobaladdr:$absaddr))),
           (LDriuh_abs_V4 tglobaladdr:$absaddr)>;
 
+//===----------------------------------------------------------------------===//
+// multiclass for load instructions with GP-relative addressing mode.
+// Rx=mem[bhwd](##global)
+// if ([!]Pv[.new]) Rx=mem[bhwd](##global)
+//===----------------------------------------------------------------------===//
+let neverHasSideEffects = 1, validSubTargets = HasV4SubT in
+multiclass LD_GP<string mnemonic, string BaseOp, RegisterClass RC> {
+  let BaseOpcode = BaseOp in {
+    let isPredicable = 1 in
+    def NAME#_V4 : LDInst2<(outs RC:$dst),
+            (ins globaladdress:$global),
+            "$dst = "#mnemonic#"(#$global)",
+            []>;
+
+    let isExtended = 1, opExtendable = 2, isPredicated = 1 in {
+      defm Pt_V4 : LD_Abs_Pred<mnemonic, RC, 0>;
+      defm NotPt_V4 : LD_Abs_Pred<mnemonic, RC, 1>;
+    }
+  }
+}
+
+defm LDd_GP  : LD_GP<"memd",  "LDd_GP",  DoubleRegs>;
+defm LDb_GP  : LD_GP<"memb",  "LDb_GP",  IntRegs>;
+defm LDub_GP : LD_GP<"memub", "LDub_GP", IntRegs>;
+defm LDh_GP  : LD_GP<"memh",  "LDh_GP",  IntRegs>;
+defm LDuh_GP : LD_GP<"memuh", "LDuh_GP", IntRegs>;
+defm LDw_GP  : LD_GP<"memw",  "LDw_GP",  IntRegs>;
+
+def : Pat <(atomic_load_64 (HexagonCONST32_GP tglobaladdr:$global)),
+           (i64 (LDd_GP_V4 tglobaladdr:$global))>;
+
+def : Pat <(atomic_load_32 (HexagonCONST32_GP tglobaladdr:$global)),
+           (i32 (LDw_GP_V4 tglobaladdr:$global))>;
+
+def : Pat <(atomic_load_16 (HexagonCONST32_GP tglobaladdr:$global)),
+           (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
+
+def : Pat <(atomic_load_8 (HexagonCONST32_GP tglobaladdr:$global)),
+           (i32 (LDub_GP_V4 tglobaladdr:$global))>;
+
+// Map from load(globaladdress) -> memw(#foo + 0)
+let AddedComplexity = 100 in
+def : Pat <(i64 (load (HexagonCONST32_GP tglobaladdr:$global))),
+           (i64 (LDd_GP_V4 tglobaladdr:$global))>;
+
+// Map from Pd = load(globaladdress) -> Rd = memb(globaladdress), Pd = Rd
+let AddedComplexity = 100 in
+def : Pat <(i1 (load (HexagonCONST32_GP tglobaladdr:$global))),
+           (i1 (TFR_PdRs (i32 (LDb_GP_V4 tglobaladdr:$global))))>;
+
+// When the Interprocedural Global Variable optimizer realizes that a certain
+// global variable takes only two constant values, it shrinks the global to
+// a boolean. Catch those loads here in the following 3 patterns.
+let AddedComplexity = 100 in
+def : Pat <(i32 (extloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
+           (i32 (LDb_GP_V4 tglobaladdr:$global))>;
+
+let AddedComplexity = 100 in
+def : Pat <(i32 (sextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
+           (i32 (LDb_GP_V4 tglobaladdr:$global))>;
+
+// Map from load(globaladdress) -> memb(#foo)
+let AddedComplexity = 100 in
+def : Pat <(i32 (extloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
+           (i32 (LDb_GP_V4 tglobaladdr:$global))>;
+
+// Map from load(globaladdress) -> memb(#foo)
+let AddedComplexity = 100 in
+def : Pat <(i32 (sextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
+           (i32 (LDb_GP_V4 tglobaladdr:$global))>;
+
+let AddedComplexity = 100 in
+def : Pat <(i32 (zextloadi1 (HexagonCONST32_GP tglobaladdr:$global))),
+           (i32 (LDub_GP_V4 tglobaladdr:$global))>;
+
+// Map from load(globaladdress) -> memub(#foo)
+let AddedComplexity = 100 in
+def : Pat <(i32 (zextloadi8 (HexagonCONST32_GP tglobaladdr:$global))),
+           (i32 (LDub_GP_V4 tglobaladdr:$global))>;
+
+// Map from load(globaladdress) -> memh(#foo)
+let AddedComplexity = 100 in
+def : Pat <(i32 (extloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
+           (i32 (LDh_GP_V4 tglobaladdr:$global))>;
+
+// Map from load(globaladdress) -> memh(#foo)
+let AddedComplexity = 100 in
+def : Pat <(i32 (sextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
+           (i32 (LDh_GP_V4 tglobaladdr:$global))>;
+
+// Map from load(globaladdress) -> memuh(#foo)
+let AddedComplexity = 100 in
+def : Pat <(i32 (zextloadi16 (HexagonCONST32_GP tglobaladdr:$global))),
+           (i32 (LDuh_GP_V4 tglobaladdr:$global))>;
+
+// Map from load(globaladdress) -> memw(#foo)
+let AddedComplexity = 100 in
+def : Pat <(i32 (load (HexagonCONST32_GP tglobaladdr:$global))),
+           (i32 (LDw_GP_V4 tglobaladdr:$global))>;
+
+
 // Transfer global address into a register
 let AddedComplexity=50, isMoveImm = 1, isReMaterializable = 1 in
 def TFRI_V4 : ALU32_ri<(outs IntRegs:$dst), (ins globaladdress:$src1),
@@ -5497,6 +3145,11 @@ def TFRI_V4 : ALU32_ri<(outs IntRegs:$dst), (ins globaladdress:$src1),
            [(set IntRegs:$dst, (HexagonCONST32 tglobaladdr:$src1))]>,
            Requires<[HasV4T]>;
 
+// Transfer a block address into a register
+def : Pat<(HexagonCONST32_GP tblockaddress:$src1),
+          (TFRI_V4 tblockaddress:$src1)>,
+          Requires<[HasV4T]>;
+
 let AddedComplexity=50, neverHasSideEffects = 1, isPredicated = 1 in
 def TFRI_cPt_V4 : ALU32_ri<(outs IntRegs:$dst),
                            (ins PredRegs:$src1, globaladdress:$src2),
@@ -5588,172 +3241,167 @@ defm STrih_ind : ST_indirect_lo<"memh", truncstorei16>;
 defm STriw_ind : ST_indirect_lo<"memw", store>;
 
 // Store - absolute addressing mode: These instruction take constant
-// value as the extended operand
+// value as the extended operand.
 multiclass ST_absimm<string OpcStr> {
-  let isPredicable = 1 in
+let isExtended = 1, opExtendable = 0, isPredicable = 1,
+validSubTargets = HasV4SubT in
   def _abs_V4 : STInst2<(outs),
-            (ins u6Imm:$src1, IntRegs:$src2),
-            !strconcat(OpcStr, "(#$src1) = $src2"),
+            (ins u0AlwaysExt:$src1, IntRegs:$src2),
+            !strconcat(OpcStr, "(##$src1) = $src2"),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
+let isExtended = 1, opExtendable = 1, isPredicated = 1,
+validSubTargets = HasV4SubT in {
   def _abs_cPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3),
-            !strconcat("if ($src1)", !strconcat(OpcStr, "(#$src2) = $src3")),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2, IntRegs:$src3),
+            !strconcat("if ($src1)", !strconcat(OpcStr, "(##$src2) = $src3")),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
   def _abs_cNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3),
-            !strconcat("if (!$src1)", !strconcat(OpcStr, "(#$src2) = $src3")),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2, IntRegs:$src3),
+            !strconcat("if (!$src1)", !strconcat(OpcStr, "(##$src2) = $src3")),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
   def _abs_cdnPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2, IntRegs:$src3),
             !strconcat("if ($src1.new)",
-            !strconcat(OpcStr, "(#$src2) = $src3")),
+            !strconcat(OpcStr, "(##$src2) = $src3")),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
   def _abs_cdnNotPt_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2, IntRegs:$src3),
             !strconcat("if (!$src1.new)",
-            !strconcat(OpcStr, "(#$src2) = $src3")),
+            !strconcat(OpcStr, "(##$src2) = $src3")),
             []>,
             Requires<[HasV4T]>;
+}
 
-  def _abs_nv_V4 : STInst2<(outs),
-            (ins u6Imm:$src1, IntRegs:$src2),
-            !strconcat(OpcStr, "(#$src1) = $src2.new"),
+let isExtended = 1, opExtendable = 0, mayStore = 1, isNVStore = 1,
+validSubTargets = HasV4SubT in
+  def _abs_nv_V4 : NVInst_V4<(outs),
+            (ins u0AlwaysExt:$src1, IntRegs:$src2),
+            !strconcat(OpcStr, "(##$src1) = $src2.new"),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
-  def _abs_cPt_nv_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3),
+let isExtended = 1, opExtendable = 1, mayStore = 1, isPredicated = 1,
+isNVStore = 1, validSubTargets = HasV4SubT in {
+  def _abs_cPt_nv_V4 : NVInst_V4<(outs),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2, IntRegs:$src3),
             !strconcat("if ($src1)",
-            !strconcat(OpcStr, "(#$src2) = $src3.new")),
+            !strconcat(OpcStr, "(##$src2) = $src3.new")),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
-  def _abs_cNotPt_nv_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3),
+  def _abs_cNotPt_nv_V4 : NVInst_V4<(outs),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2, IntRegs:$src3),
             !strconcat("if (!$src1)",
-            !strconcat(OpcStr, "(#$src2) = $src3.new")),
+            !strconcat(OpcStr, "(##$src2) = $src3.new")),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
-  def _abs_cdnPt_nv_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3),
+  def _abs_cdnPt_nv_V4 : NVInst_V4<(outs),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2, IntRegs:$src3),
             !strconcat("if ($src1.new)",
-            !strconcat(OpcStr, "(#$src2) = $src3.new")),
+            !strconcat(OpcStr, "(##$src2) = $src3.new")),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
-  def _abs_cdnNotPt_nv_V4 : STInst2<(outs),
-            (ins PredRegs:$src1, u6Imm:$src2, IntRegs:$src3),
+  def _abs_cdnNotPt_nv_V4 : NVInst_V4<(outs),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2, IntRegs:$src3),
             !strconcat("if (!$src1.new)",
-            !strconcat(OpcStr, "(#$src2) = $src3.new")),
+            !strconcat(OpcStr, "(##$src2) = $src3.new")),
             []>,
             Requires<[HasV4T]>;
 }
+}
 
 defm STrib_imm : ST_absimm<"memb">;
 defm STrih_imm : ST_absimm<"memh">;
 defm STriw_imm : ST_absimm<"memw">;
 
-let Predicates = [HasV4T], AddedComplexity  = 30 in
-def : Pat<(truncstorei8 (i32 IntRegs:$src1), u6ImmPred:$src2),
-          (STrib_imm_abs_V4 u6ImmPred:$src2, IntRegs: $src1)>;
+let Predicates = [HasV4T], AddedComplexity  = 30 in {
+def : Pat<(truncstorei8 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
+          (STrib_imm_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
 
-let Predicates = [HasV4T], AddedComplexity  = 30 in
-def : Pat<(truncstorei16 (i32 IntRegs:$src1), u6ImmPred:$src2),
-          (STrih_imm_abs_V4 u6ImmPred:$src2, IntRegs: $src1)>;
-
-let Predicates = [HasV4T], AddedComplexity  = 30 in
-def : Pat<(store (i32 IntRegs:$src1), u6ImmPred:$src2),
-          (STriw_imm_abs_V4 u6ImmPred:$src2, IntRegs: $src1)>;
+def : Pat<(truncstorei16 (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
+          (STrih_imm_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
 
+def : Pat<(store (i32 IntRegs:$src1), u0AlwaysExtPred:$src2),
+          (STriw_imm_abs_V4 u0AlwaysExtPred:$src2, IntRegs: $src1)>;
+}
 
 // Load - absolute addressing mode: These instruction take constant
 // value as the extended operand
 
 multiclass LD_absimm<string OpcStr> {
-  let isPredicable = 1 in
+let isExtended = 1, opExtendable = 1, isPredicable = 1,
+validSubTargets = HasV4SubT in
   def _abs_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins u6Imm:$src),
+            (ins u0AlwaysExt:$src),
             !strconcat("$dst = ",
-            !strconcat(OpcStr, "(#$src)")),
+            !strconcat(OpcStr, "(##$src)")),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
+let isExtended = 1, opExtendable = 2, isPredicated = 1,
+validSubTargets = HasV4SubT in {
   def _abs_cPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, u6Imm:$src2),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2),
             !strconcat("if ($src1) $dst = ",
-            !strconcat(OpcStr, "(#$src2)")),
+            !strconcat(OpcStr, "(##$src2)")),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
   def _abs_cNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, u6Imm:$src2),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2),
             !strconcat("if (!$src1) $dst = ",
-            !strconcat(OpcStr, "(#$src2)")),
+            !strconcat(OpcStr, "(##$src2)")),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
   def _abs_cdnPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, u6Imm:$src2),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2),
             !strconcat("if ($src1.new) $dst = ",
-            !strconcat(OpcStr, "(#$src2)")),
+            !strconcat(OpcStr, "(##$src2)")),
             []>,
             Requires<[HasV4T]>;
 
-  let isPredicated = 1 in
   def _abs_cdnNotPt_V4 : LDInst2<(outs IntRegs:$dst),
-            (ins PredRegs:$src1, u6Imm:$src2),
+            (ins PredRegs:$src1, u0AlwaysExt:$src2),
             !strconcat("if (!$src1.new) $dst = ",
-            !strconcat(OpcStr, "(#$src2)")),
+            !strconcat(OpcStr, "(##$src2)")),
             []>,
             Requires<[HasV4T]>;
 }
+}
 
-defm LDrib_imm : LD_absimm<"memb">;
+defm LDrib_imm  : LD_absimm<"memb">;
 defm LDriub_imm : LD_absimm<"memub">;
-defm LDrih_imm : LD_absimm<"memh">;
+defm LDrih_imm  : LD_absimm<"memh">;
 defm LDriuh_imm : LD_absimm<"memuh">;
-defm LDriw_imm : LD_absimm<"memw">;
+defm LDriw_imm  : LD_absimm<"memw">;
 
-let Predicates = [HasV4T], AddedComplexity  = 30 in
-def : Pat<(i32 (load u6ImmPred:$src)),
-          (LDriw_imm_abs_V4 u6ImmPred:$src)>;
+let Predicates = [HasV4T], AddedComplexity  = 30 in {
+def : Pat<(i32 (load u0AlwaysExtPred:$src)),
+          (LDriw_imm_abs_V4 u0AlwaysExtPred:$src)>;
 
-let Predicates = [HasV4T], AddedComplexity=30 in
-def : Pat<(i32 (sextloadi8 u6ImmPred:$src)),
-          (LDrib_imm_abs_V4 u6ImmPred:$src)>;
+def : Pat<(i32 (sextloadi8 u0AlwaysExtPred:$src)),
+          (LDrib_imm_abs_V4 u0AlwaysExtPred:$src)>;
 
-let Predicates = [HasV4T], AddedComplexity=30 in
-def : Pat<(i32 (zextloadi8 u6ImmPred:$src)),
-          (LDriub_imm_abs_V4 u6ImmPred:$src)>;
+def : Pat<(i32 (zextloadi8 u0AlwaysExtPred:$src)),
+          (LDriub_imm_abs_V4 u0AlwaysExtPred:$src)>;
 
-let Predicates = [HasV4T], AddedComplexity=30 in
-def : Pat<(i32 (sextloadi16 u6ImmPred:$src)),
-          (LDrih_imm_abs_V4 u6ImmPred:$src)>;
-
-let Predicates = [HasV4T], AddedComplexity=30 in
-def : Pat<(i32 (zextloadi16 u6ImmPred:$src)),
-          (LDriuh_imm_abs_V4 u6ImmPred:$src)>;
+def : Pat<(i32 (sextloadi16 u0AlwaysExtPred:$src)),
+          (LDrih_imm_abs_V4 u0AlwaysExtPred:$src)>;
 
+def : Pat<(i32 (zextloadi16 u0AlwaysExtPred:$src)),
+          (LDriuh_imm_abs_V4 u0AlwaysExtPred:$src)>;
+}
 
 // Indexed store double word - global address.
 // memw(Rs+#u6:2)=#S8
@@ -5775,3 +3423,109 @@ def STrih_offset_ext_V4 : STInst<(outs),
             [(truncstorei16 (HexagonCONST32 tglobaladdr:$src3),
                     (add IntRegs:$src1, u6_1ImmPred:$src2))]>,
             Requires<[HasV4T]>;
+// Map from store(globaladdress + x) -> memd(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(store (i64 DoubleRegs:$src1),
+                 FoldGlobalAddrGP:$addr),
+          (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
+          Requires<[HasV4T]>;
+
+def : Pat<(atomic_store_64 FoldGlobalAddrGP:$addr,
+                           (i64 DoubleRegs:$src1)),
+          (STrid_abs_V4 FoldGlobalAddrGP:$addr, (i64 DoubleRegs:$src1))>,
+          Requires<[HasV4T]>;
+
+// Map from store(globaladdress + x) -> memb(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(truncstorei8 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
+          (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+            Requires<[HasV4T]>;
+
+def : Pat<(atomic_store_8 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
+          (STrib_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+            Requires<[HasV4T]>;
+
+// Map from store(globaladdress + x) -> memh(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(truncstorei16 (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
+          (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+            Requires<[HasV4T]>;
+
+def : Pat<(atomic_store_16 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
+          (STrih_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+            Requires<[HasV4T]>;
+
+// Map from store(globaladdress + x) -> memw(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(store (i32 IntRegs:$src1), FoldGlobalAddrGP:$addr),
+          (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+           Requires<[HasV4T]>;
+
+def : Pat<(atomic_store_32 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1)),
+          (STriw_abs_V4 FoldGlobalAddrGP:$addr, (i32 IntRegs:$src1))>,
+            Requires<[HasV4T]>;
+
+// Map from load(globaladdress + x) -> memd(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(i64 (load FoldGlobalAddrGP:$addr)),
+          (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+def : Pat<(atomic_load_64 FoldGlobalAddrGP:$addr),
+          (i64 (LDrid_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+// Map from load(globaladdress + x) -> memb(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(i32 (extloadi8 FoldGlobalAddrGP:$addr)),
+          (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+// Map from load(globaladdress + x) -> memb(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(i32 (sextloadi8 FoldGlobalAddrGP:$addr)),
+          (i32 (LDrib_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+//let AddedComplexity = 100 in
+let AddedComplexity = 100 in
+def : Pat<(i32 (extloadi16 FoldGlobalAddrGP:$addr)),
+          (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+// Map from load(globaladdress + x) -> memh(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(i32 (sextloadi16 FoldGlobalAddrGP:$addr)),
+          (i32 (LDrih_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+// Map from load(globaladdress + x) -> memuh(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(i32 (zextloadi16 FoldGlobalAddrGP:$addr)),
+          (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+def : Pat<(atomic_load_16 FoldGlobalAddrGP:$addr),
+          (i32 (LDriuh_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+// Map from load(globaladdress + x) -> memub(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(i32 (zextloadi8 FoldGlobalAddrGP:$addr)),
+          (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+def : Pat<(atomic_load_8 FoldGlobalAddrGP:$addr),
+          (i32 (LDriub_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+// Map from load(globaladdress + x) -> memw(#foo + x)
+let AddedComplexity = 100 in
+def : Pat<(i32 (load FoldGlobalAddrGP:$addr)),
+          (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
+def : Pat<(atomic_load_32 FoldGlobalAddrGP:$addr),
+          (i32 (LDriw_abs_V4 FoldGlobalAddrGP:$addr))>,
+           Requires<[HasV4T]>;
+
diff --git a/lib/Target/Hexagon/HexagonMCInst.h b/lib/Target/Hexagon/HexagonMCInst.h
deleted file mode 100644
index 7a16c241ff8f..000000000000
--- a/lib/Target/Hexagon/HexagonMCInst.h
+++ /dev/null
@@ -1,41 +0,0 @@
-//===- HexagonMCInst.h - Hexagon sub-class of MCInst ----------------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This class extends MCInst to allow some VLIW annotation.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef HEXAGONMCINST_H
-#define HEXAGONMCINST_H
-
-#include "llvm/MC/MCInst.h"
-#include "llvm/CodeGen/MachineInstr.h"
-
-namespace llvm {
-  class HexagonMCInst: public MCInst {
-    // Packet start and end markers
-    unsigned startPacket: 1, endPacket: 1;
-    const MachineInstr *MachineI;
-  public:
-    explicit HexagonMCInst(): MCInst(),
-                              startPacket(0), endPacket(0) {}
-
-    const MachineInstr* getMI() const { return MachineI; }
-
-    void setMI(const MachineInstr *MI) { MachineI = MI; }
-
-    bool isStartPacket() const { return (startPacket); }
-    bool isEndPacket() const { return (endPacket); }
-
-    void setStartPacket(bool yes) { startPacket = yes; }
-    void setEndPacket(bool yes) { endPacket = yes; }
-  };
-}
-
-#endif
diff --git a/lib/Target/Hexagon/HexagonMCInstLower.cpp b/lib/Target/Hexagon/HexagonMCInstLower.cpp
index 70bddcc76a59..f011d51bd61a 100644
--- a/lib/Target/Hexagon/HexagonMCInstLower.cpp
+++ b/lib/Target/Hexagon/HexagonMCInstLower.cpp
@@ -15,8 +15,9 @@
 #include "Hexagon.h"
 #include "HexagonAsmPrinter.h"
 #include "HexagonMachineFunctionInfo.h"
-#include "llvm/Constants.h"
+#include "MCTargetDesc/HexagonMCInst.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/IR/Constants.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/Target/Mangler.h"
@@ -38,9 +39,10 @@ static MCOperand GetSymbolRef(const MachineOperand& MO, const MCSymbol* Symbol,
 }
 
 // Create an MCInst from a MachineInstr
-void llvm::HexagonLowerToMC(const MachineInstr* MI, MCInst& MCI,
+void llvm::HexagonLowerToMC(const MachineInstr* MI, HexagonMCInst& MCI,
                             HexagonAsmPrinter& AP) {
   MCI.setOpcode(MI->getOpcode());
+  MCI.setDesc(MI->getDesc());
 
   for (unsigned i = 0, e = MI->getNumOperands(); i < e; i++) {
     const MachineOperand &MO = MI->getOperand(i);
diff --git a/lib/Target/Hexagon/HexagonMachineScheduler.cpp b/lib/Target/Hexagon/HexagonMachineScheduler.cpp
index 0e9ef4838d8a..1388ad4f167d 100644
--- a/lib/Target/Hexagon/HexagonMachineScheduler.cpp
+++ b/lib/Target/Hexagon/HexagonMachineScheduler.cpp
@@ -15,8 +15,8 @@
 #define DEBUG_TYPE "misched"
 
 #include "HexagonMachineScheduler.h"
-
-#include <queue>
+#include "llvm/CodeGen/MachineLoopInfo.h"
+#include "llvm/IR/Function.h"
 
 using namespace llvm;
 
@@ -153,7 +153,16 @@ void VLIWMachineScheduler::schedule() {
   // Postprocess the DAG to add platform specific artificial dependencies.
   postprocessDAG();
 
+  SmallVector<SUnit*, 8> TopRoots, BotRoots;
+  findRootsAndBiasEdges(TopRoots, BotRoots);
+
+  // Initialize the strategy before modifying the DAG.
+  SchedImpl->initialize(this);
+
   // To view Height/Depth correctly, they should be accessed at least once.
+  //
+  // FIXME: SUnit::dumpAll always recompute depth and height now. The max
+  // depth/height could be computed directly from the roots and leaves.
   DEBUG(unsigned maxH = 0;
         for (unsigned su = 0, e = SUnits.size(); su != e; ++su)
           if (SUnits[su].getHeight() > maxH)
@@ -167,7 +176,7 @@ void VLIWMachineScheduler::schedule() {
   DEBUG(for (unsigned su = 0, e = SUnits.size(); su != e; ++su)
           SUnits[su].dumpAll(this));
 
-  initQueues();
+  initQueues(TopRoots, BotRoots);
 
   bool IsTopNode = false;
   while (SUnit *SU = SchedImpl->pickNode(IsTopNode)) {
@@ -187,6 +196,7 @@ void ConvergingVLIWScheduler::initialize(ScheduleDAGMI *dag) {
   DAG = static_cast<VLIWMachineScheduler*>(dag);
   SchedModel = DAG->getSchedModel();
   TRI = DAG->TRI;
+
   Top.init(DAG, SchedModel);
   Bot.init(DAG, SchedModel);
 
@@ -194,6 +204,8 @@ void ConvergingVLIWScheduler::initialize(ScheduleDAGMI *dag) {
   // are disabled, then these HazardRecs will be disabled.
   const InstrItineraryData *Itin = DAG->getSchedModel()->getInstrItineraries();
   const TargetMachine &TM = DAG->MF.getTarget();
+  delete Top.HazardRec;
+  delete Bot.HazardRec;
   Top.HazardRec = TM.getInstrInfo()->CreateTargetMIHazardRecognizer(Itin, DAG);
   Bot.HazardRec = TM.getInstrInfo()->CreateTargetMIHazardRecognizer(Itin, DAG);
 
@@ -678,4 +690,3 @@ void ConvergingVLIWScheduler::schedNode(SUnit *SU, bool IsTopNode) {
     Bot.bumpNode(SU);
   }
 }
-
diff --git a/lib/Target/Hexagon/HexagonMachineScheduler.h b/lib/Target/Hexagon/HexagonMachineScheduler.h
index fe0242a0f74e..f68dadf29210 100644
--- a/lib/Target/Hexagon/HexagonMachineScheduler.h
+++ b/lib/Target/Hexagon/HexagonMachineScheduler.h
@@ -14,6 +14,9 @@
 #ifndef HEXAGONASMPRINTER_H
 #define HEXAGONASMPRINTER_H
 
+#include "llvm/ADT/OwningPtr.h"
+#include "llvm/ADT/PriorityQueue.h"
+#include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/CodeGen/LiveIntervalAnalysis.h"
 #include "llvm/CodeGen/MachineScheduler.h"
 #include "llvm/CodeGen/Passes.h"
@@ -22,14 +25,11 @@
 #include "llvm/CodeGen/ResourcePriorityQueue.h"
 #include "llvm/CodeGen/ScheduleDAGInstrs.h"
 #include "llvm/CodeGen/ScheduleHazardRecognizer.h"
-#include "llvm/Analysis/AliasAnalysis.h"
-#include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/OwningPtr.h"
-#include "llvm/ADT/PriorityQueue.h"
+#include "llvm/Target/TargetInstrInfo.h"
 
 using namespace llvm;
 
diff --git a/lib/Target/Hexagon/HexagonNewValueJump.cpp b/lib/Target/Hexagon/HexagonNewValueJump.cpp
index 1e91c3948550..5e80e48b01d5 100644
--- a/lib/Target/Hexagon/HexagonNewValueJump.cpp
+++ b/lib/Target/Hexagon/HexagonNewValueJump.cpp
@@ -22,31 +22,29 @@
 //
 //===----------------------------------------------------------------------===//
 #define DEBUG_TYPE "hexagon-nvj"
-#include "llvm/PassSupport.h"
-#include "llvm/Support/Compiler.h"
-#include "llvm/Support/Debug.h"
+#include "Hexagon.h"
+#include "HexagonInstrInfo.h"
+#include "HexagonMachineFunctionInfo.h"
+#include "HexagonRegisterInfo.h"
+#include "HexagonSubtarget.h"
+#include "HexagonTargetMachine.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/Statistic.h"
-#include "llvm/CodeGen/Passes.h"
-#include "llvm/CodeGen/ScheduleDAGInstrs.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/LiveVariables.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
-#include "llvm/Target/TargetMachine.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/CodeGen/ScheduleDAGInstrs.h"
+#include "llvm/PassSupport.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Compiler.h"
+#include "llvm/Support/Debug.h"
 #include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetRegisterInfo.h"
-#include "Hexagon.h"
-#include "HexagonTargetMachine.h"
-#include "HexagonRegisterInfo.h"
-#include "HexagonSubtarget.h"
-#include "HexagonInstrInfo.h"
-#include "HexagonMachineFunctionInfo.h"
-
 #include <map>
-
-#include "llvm/Support/CommandLine.h"
 using namespace llvm;
 
 STATISTIC(NumNVJGenerated, "Number of New Value Jump Instructions created");
@@ -222,7 +220,7 @@ static bool canCompareBeNewValueJump(const HexagonInstrInfo *QII,
       return false;
   }
 
-  unsigned cmpReg1, cmpOp2 = 0; // cmpOp2 assignment silences compiler warning.
+  unsigned cmpReg1, cmpOp2;
   cmpReg1 = MI->getOperand(1).getReg();
 
   if (secondReg) {
diff --git a/lib/Target/Hexagon/HexagonOperands.td b/lib/Target/Hexagon/HexagonOperands.td
new file mode 100644
index 000000000000..c79d78f21080
--- /dev/null
+++ b/lib/Target/Hexagon/HexagonOperands.td
@@ -0,0 +1,858 @@
+//===- HexagonOperands.td - Hexagon immediate processing -*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illnois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+// Immediate operands.
+
+let PrintMethod = "printImmOperand" in {
+  // f32Ext type is used to identify constant extended floating point immediates.
+  def f32Ext : Operand<f32>;
+  def s32Imm : Operand<i32>;
+  def s26_6Imm : Operand<i32>;
+  def s16Imm : Operand<i32>;
+  def s12Imm : Operand<i32>;
+  def s11Imm : Operand<i32>;
+  def s11_0Imm : Operand<i32>;
+  def s11_1Imm : Operand<i32>;
+  def s11_2Imm : Operand<i32>;
+  def s11_3Imm : Operand<i32>;
+  def s10Imm : Operand<i32>;
+  def s9Imm : Operand<i32>;
+  def m9Imm : Operand<i32>;
+  def s8Imm : Operand<i32>;
+  def s8Imm64 : Operand<i64>;
+  def s6Imm : Operand<i32>;
+  def s4Imm : Operand<i32>;
+  def s4_0Imm : Operand<i32>;
+  def s4_1Imm : Operand<i32>;
+  def s4_2Imm : Operand<i32>;
+  def s4_3Imm : Operand<i32>;
+  def u64Imm : Operand<i64>;
+  def u32Imm : Operand<i32>;
+  def u26_6Imm : Operand<i32>;
+  def u16Imm : Operand<i32>;
+  def u16_0Imm : Operand<i32>;
+  def u16_1Imm : Operand<i32>;
+  def u16_2Imm : Operand<i32>;
+  def u11_3Imm : Operand<i32>;
+  def u10Imm : Operand<i32>;
+  def u9Imm : Operand<i32>;
+  def u8Imm : Operand<i32>;
+  def u7Imm : Operand<i32>;
+  def u6Imm : Operand<i32>;
+  def u6_0Imm : Operand<i32>;
+  def u6_1Imm : Operand<i32>;
+  def u6_2Imm : Operand<i32>;
+  def u6_3Imm : Operand<i32>;
+  def u5Imm : Operand<i32>;
+  def u4Imm : Operand<i32>;
+  def u3Imm : Operand<i32>;
+  def u2Imm : Operand<i32>;
+  def u1Imm : Operand<i32>;
+  def n8Imm : Operand<i32>;
+  def m6Imm : Operand<i32>;
+}
+
+let PrintMethod = "printNOneImmOperand" in
+def nOneImm : Operand<i32>;
+
+//
+// Immediate predicates
+//
+def s32ImmPred  : PatLeaf<(i32 imm), [{
+  // s32ImmPred predicate - True if the immediate fits in a 32-bit sign extended
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<32>(v);
+}]>;
+
+def s32_24ImmPred  : PatLeaf<(i32 imm), [{
+  // s32_24ImmPred predicate - True if the immediate fits in a 32-bit sign
+  // extended field that is a multiple of 0x1000000.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedInt<32,24>(v);
+}]>;
+
+def s32_16s8ImmPred  : PatLeaf<(i32 imm), [{
+  // s32_16s8ImmPred predicate - True if the immediate fits in a 32-bit sign
+  // extended field that is a multiple of 0x10000.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedInt<24,16>(v);
+}]>;
+
+def s26_6ImmPred  : PatLeaf<(i32 imm), [{
+  // s26_6ImmPred predicate - True if the immediate fits in a 32-bit
+  // sign extended field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedInt<26,6>(v);
+}]>;
+
+
+def s16ImmPred  : PatLeaf<(i32 imm), [{
+  // s16ImmPred predicate - True if the immediate fits in a 16-bit sign extended
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<16>(v);
+}]>;
+
+
+def s13ImmPred  : PatLeaf<(i32 imm), [{
+  // s13ImmPred predicate - True if the immediate fits in a 13-bit sign extended
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<13>(v);
+}]>;
+
+
+def s12ImmPred  : PatLeaf<(i32 imm), [{
+  // s12ImmPred predicate - True if the immediate fits in a 12-bit
+  // sign extended field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<12>(v);
+}]>;
+
+def s11_0ImmPred  : PatLeaf<(i32 imm), [{
+  // s11_0ImmPred predicate - True if the immediate fits in a 11-bit
+  // sign extended field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<11>(v);
+}]>;
+
+
+def s11_1ImmPred  : PatLeaf<(i32 imm), [{
+  // s11_1ImmPred predicate - True if the immediate fits in a 12-bit
+  // sign extended field and is a multiple of 2.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedInt<11,1>(v);
+}]>;
+
+
+def s11_2ImmPred  : PatLeaf<(i32 imm), [{
+  // s11_2ImmPred predicate - True if the immediate fits in a 13-bit
+  // sign extended field and is a multiple of 4.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedInt<11,2>(v);
+}]>;
+
+
+def s11_3ImmPred  : PatLeaf<(i32 imm), [{
+  // s11_3ImmPred predicate - True if the immediate fits in a 14-bit
+  // sign extended field and is a multiple of 8.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedInt<11,3>(v);
+}]>;
+
+
+def s10ImmPred  : PatLeaf<(i32 imm), [{
+  // s10ImmPred predicate - True if the immediate fits in a 10-bit sign extended
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<10>(v);
+}]>;
+
+
+def s9ImmPred  : PatLeaf<(i32 imm), [{
+  // s9ImmPred predicate - True if the immediate fits in a 9-bit sign extended
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<9>(v);
+}]>;
+
+def m9ImmPred  : PatLeaf<(i32 imm), [{
+  // m9ImmPred predicate - True if the immediate fits in a 9-bit magnitude
+  // field. The range of m9 is -255 to 255.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<9>(v) && (v != -256);
+}]>;
+
+def s8ImmPred  : PatLeaf<(i32 imm), [{
+  // s8ImmPred predicate - True if the immediate fits in a 8-bit sign extended
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<8>(v);
+}]>;
+
+
+def s8Imm64Pred  : PatLeaf<(i64 imm), [{
+  // s8ImmPred predicate - True if the immediate fits in a 8-bit sign extended
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<8>(v);
+}]>;
+
+
+def s6ImmPred  : PatLeaf<(i32 imm), [{
+  // s6ImmPred predicate - True if the immediate fits in a 6-bit sign extended
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<6>(v);
+}]>;
+
+
+def s4_0ImmPred  : PatLeaf<(i32 imm), [{
+  // s4_0ImmPred predicate - True if the immediate fits in a 4-bit sign extended
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isInt<4>(v);
+}]>;
+
+
+def s4_1ImmPred  : PatLeaf<(i32 imm), [{
+  // s4_1ImmPred predicate - True if the immediate fits in a 4-bit sign extended
+  // field of 2.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedInt<4,1>(v);
+}]>;
+
+
+def s4_2ImmPred  : PatLeaf<(i32 imm), [{
+  // s4_2ImmPred predicate - True if the immediate fits in a 4-bit sign extended
+  // field that is a multiple of 4.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedInt<4,2>(v);
+}]>;
+
+
+def s4_3ImmPred  : PatLeaf<(i32 imm), [{
+  // s4_3ImmPred predicate - True if the immediate fits in a 4-bit sign extended
+  // field that is a multiple of 8.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedInt<4,3>(v);
+}]>;
+
+
+def u64ImmPred  : PatLeaf<(i64 imm), [{
+  // Adding "N ||" to suppress gcc unused warning.
+  return (N || true);
+}]>;
+
+def u32ImmPred  : PatLeaf<(i32 imm), [{
+  // u32ImmPred predicate - True if the immediate fits in a 32-bit field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<32>(v);
+}]>;
+
+def u26_6ImmPred  : PatLeaf<(i32 imm), [{
+  // u26_6ImmPred - True if the immediate fits in a 32-bit field and
+  // is a multiple of 64.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedUInt<26,6>(v);
+}]>;
+
+def u16ImmPred  : PatLeaf<(i32 imm), [{
+  // u16ImmPred predicate - True if the immediate fits in a 16-bit unsigned
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<16>(v);
+}]>;
+
+def u16_s8ImmPred  : PatLeaf<(i32 imm), [{
+  // u16_s8ImmPred predicate - True if the immediate fits in a 16-bit sign
+  // extended s8 field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedUInt<16,8>(v);
+}]>;
+
+def u9ImmPred  : PatLeaf<(i32 imm), [{
+  // u9ImmPred predicate - True if the immediate fits in a 9-bit unsigned
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<9>(v);
+}]>;
+
+
+def u8ImmPred  : PatLeaf<(i32 imm), [{
+  // u8ImmPred predicate - True if the immediate fits in a 8-bit unsigned
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<8>(v);
+}]>;
+
+def u7StrictPosImmPred : ImmLeaf<i32, [{
+  // u7StrictPosImmPred predicate - True if the immediate fits in an 7-bit
+  // unsigned field and is strictly greater than 0.
+  return isUInt<7>(Imm) && Imm > 0;
+}]>;
+
+def u7ImmPred  : PatLeaf<(i32 imm), [{
+  // u7ImmPred predicate - True if the immediate fits in a 7-bit unsigned
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<7>(v);
+}]>;
+
+
+def u6ImmPred  : PatLeaf<(i32 imm), [{
+  // u6ImmPred predicate - True if the immediate fits in a 6-bit unsigned
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<6>(v);
+}]>;
+
+def u6_0ImmPred  : PatLeaf<(i32 imm), [{
+  // u6_0ImmPred predicate - True if the immediate fits in a 6-bit unsigned
+  // field. Same as u6ImmPred.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<6>(v);
+}]>;
+
+def u6_1ImmPred  : PatLeaf<(i32 imm), [{
+  // u6_1ImmPred predicate - True if the immediate fits in a 7-bit unsigned
+  // field that is 1 bit alinged - multiple of 2.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedUInt<6,1>(v);
+}]>;
+
+def u6_2ImmPred  : PatLeaf<(i32 imm), [{
+  // u6_2ImmPred predicate - True if the immediate fits in a 8-bit unsigned
+  // field that is 2 bits alinged - multiple of 4.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedUInt<6,2>(v);
+}]>;
+
+def u6_3ImmPred  : PatLeaf<(i32 imm), [{
+  // u6_3ImmPred predicate - True if the immediate fits in a 9-bit unsigned
+  // field that is 3 bits alinged - multiple of 8.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isShiftedUInt<6,3>(v);
+}]>;
+
+def u5ImmPred  : PatLeaf<(i32 imm), [{
+  // u5ImmPred predicate - True if the immediate fits in a 5-bit unsigned
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<5>(v);
+}]>;
+
+
+def u3ImmPred  : PatLeaf<(i32 imm), [{
+  // u3ImmPred predicate - True if the immediate fits in a 3-bit unsigned
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<3>(v);
+}]>;
+
+
+def u2ImmPred  : PatLeaf<(i32 imm), [{
+  // u2ImmPred predicate - True if the immediate fits in a 2-bit unsigned
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<2>(v);
+}]>;
+
+
+def u1ImmPred  : PatLeaf<(i1 imm), [{
+  // u1ImmPred predicate - True if the immediate fits in a 1-bit unsigned
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return isUInt<1>(v);
+}]>;
+
+def m5BImmPred  : PatLeaf<(i32 imm), [{
+  // m5BImmPred predicate - True if the (char) number is in range -1 .. -31
+  // and will fit in a 5 bit field when made positive, for use in memops.
+  // this is specific to the zero extending of a negative by CombineInstr
+  int8_t v = (int8_t)N->getSExtValue();
+  return (-31 <= v && v <= -1);
+}]>;
+
+def m5HImmPred  : PatLeaf<(i32 imm), [{
+  // m5HImmPred predicate - True if the (short) number is in range -1 .. -31
+  // and will fit in a 5 bit field when made positive, for use in memops.
+  // this is specific to the zero extending of a negative by CombineInstr
+  int16_t v = (int16_t)N->getSExtValue();
+  return (-31 <= v && v <= -1);
+}]>;
+
+def m5ImmPred  : PatLeaf<(i32 imm), [{
+  // m5ImmPred predicate - True if the number is in range -1 .. -31
+  // and will fit in a 5 bit field when made positive, for use in memops.
+  int64_t v = (int64_t)N->getSExtValue();
+  return (-31 <= v && v <= -1);
+}]>;
+
+//InN means negative integers in [-(2^N - 1), 0]
+def n8ImmPred  : PatLeaf<(i32 imm), [{
+  // n8ImmPred predicate - True if the immediate fits in a 8-bit signed
+  // field.
+  int64_t v = (int64_t)N->getSExtValue();
+  return (-255 <= v && v <= 0);
+}]>;
+
+def nOneImmPred  : PatLeaf<(i32 imm), [{
+  // nOneImmPred predicate - True if the immediate is -1.
+  int64_t v = (int64_t)N->getSExtValue();
+  return (-1 == v);
+}]>;
+
+def Set5ImmPred : PatLeaf<(i32 imm), [{
+  // Set5ImmPred predicate - True if the number is in the series of values.
+  // [ 2^0, 2^1, ... 2^31 ]
+  // For use in setbit immediate.
+  uint32_t v = (int32_t)N->getSExtValue();
+  // Constrain to 32 bits, and then check for single bit.
+  return ImmIsSingleBit(v);
+}]>;
+
+def Clr5ImmPred : PatLeaf<(i32 imm), [{
+  // Clr5ImmPred predicate - True if the number is in the series of
+  // bit negated values.
+  // [ 2^0, 2^1, ... 2^31 ]
+  // For use in clrbit immediate.
+  // Note: we are bit NOTing the value.
+  uint32_t v = ~ (int32_t)N->getSExtValue();
+  // Constrain to 32 bits, and then check for single bit.
+  return ImmIsSingleBit(v);
+}]>;
+
+def SetClr5ImmPred : PatLeaf<(i32 imm), [{
+  // SetClr5ImmPred predicate - True if the immediate is in range 0..31.
+  int32_t v = (int32_t)N->getSExtValue();
+  return (v >= 0 && v <= 31);
+}]>;
+
+def Set4ImmPred : PatLeaf<(i32 imm), [{
+  // Set4ImmPred predicate - True if the number is in the series of values:
+  // [ 2^0, 2^1, ... 2^15 ].
+  // For use in setbit immediate.
+  uint16_t v = (int16_t)N->getSExtValue();
+  // Constrain to 16 bits, and then check for single bit.
+  return ImmIsSingleBit(v);
+}]>;
+
+def Clr4ImmPred : PatLeaf<(i32 imm), [{
+  // Clr4ImmPred predicate - True if the number is in the series of
+  // bit negated values:
+  // [ 2^0, 2^1, ... 2^15 ].
+  // For use in setbit and clrbit immediate.
+  uint16_t v = ~ (int16_t)N->getSExtValue();
+  // Constrain to 16 bits, and then check for single bit.
+  return ImmIsSingleBit(v);
+}]>;
+
+def SetClr4ImmPred : PatLeaf<(i32 imm), [{
+  // SetClr4ImmPred predicate - True if the immediate is in the range 0..15.
+  int16_t v = (int16_t)N->getSExtValue();
+  return (v >= 0 && v <= 15);
+}]>;
+
+def Set3ImmPred : PatLeaf<(i32 imm), [{
+  // Set3ImmPred predicate - True if the number is in the series of values:
+  // [ 2^0, 2^1, ... 2^7 ].
+  // For use in setbit immediate.
+  uint8_t v = (int8_t)N->getSExtValue();
+  // Constrain to 8 bits, and then check for single bit.
+  return ImmIsSingleBit(v);
+}]>;
+
+def Clr3ImmPred : PatLeaf<(i32 imm), [{
+  // Clr3ImmPred predicate - True if the number is in the series of
+  // bit negated values:
+  // [ 2^0, 2^1, ... 2^7 ].
+  // For use in setbit and clrbit immediate.
+  uint8_t v = ~ (int8_t)N->getSExtValue();
+  // Constrain to 8 bits, and then check for single bit.
+  return ImmIsSingleBit(v);
+}]>;
+
+def SetClr3ImmPred : PatLeaf<(i32 imm), [{
+  // SetClr3ImmPred predicate - True if the immediate is in the range  0..7.
+  int8_t v = (int8_t)N->getSExtValue();
+  return (v >= 0 && v <= 7);
+}]>;
+
+
+// Extendable immediate operands.
+
+let PrintMethod = "printExtOperand" in {
+  def s16Ext : Operand<i32>;
+  def s12Ext : Operand<i32>;
+  def s10Ext : Operand<i32>;
+  def s9Ext : Operand<i32>;
+  def s8Ext : Operand<i32>;
+  def s6Ext : Operand<i32>;
+  def s11_0Ext : Operand<i32>;
+  def s11_1Ext : Operand<i32>;
+  def s11_2Ext : Operand<i32>;
+  def s11_3Ext : Operand<i32>;
+  def u6Ext : Operand<i32>;
+  def u7Ext : Operand<i32>;
+  def u8Ext : Operand<i32>;
+  def u9Ext : Operand<i32>;
+  def u10Ext : Operand<i32>;
+  def u6_0Ext : Operand<i32>;
+  def u6_1Ext : Operand<i32>;
+  def u6_2Ext : Operand<i32>;
+  def u6_3Ext : Operand<i32>;
+}
+
+let PrintMethod = "printImmOperand" in
+def u0AlwaysExt : Operand<i32>;
+
+// Predicates for constant extendable operands
+def s16ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 16-bit sign extended field.
+    return isInt<16>(v);
+  else {
+    if (isInt<16>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit signed field.
+    return isConstExtProfitable(Node) && isInt<32>(v);
+  }
+}]>;
+
+def s10ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 10-bit sign extended field.
+    return isInt<10>(v);
+  else {
+    if (isInt<10>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit signed field.
+    return isConstExtProfitable(Node) && isInt<32>(v);
+  }
+}]>;
+
+def s9ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 9-bit sign extended field.
+    return isInt<9>(v);
+  else {
+    if (isInt<9>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit unsigned field.
+    return isConstExtProfitable(Node) && isInt<32>(v);
+  }
+}]>;
+
+def s8ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 8-bit sign extended field.
+    return isInt<8>(v);
+  else {
+    if (isInt<8>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit signed field.
+    return isConstExtProfitable(Node) && isInt<32>(v);
+  }
+}]>;
+
+def s8_16ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate fits in a 8-bit sign extended field.
+    return isInt<8>(v);
+  else {
+    if (isInt<8>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can't fit in a 16-bit signed field. This is required to avoid
+    // unnecessary constant extenders.
+    return isConstExtProfitable(Node) && !isInt<16>(v);
+  }
+}]>;
+
+def s6ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 6-bit sign extended field.
+    return isInt<6>(v);
+  else {
+    if (isInt<6>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit unsigned field.
+    return isConstExtProfitable(Node) && isInt<32>(v);
+  }
+}]>;
+
+def s6_16ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate fits in a 6-bit sign extended field.
+    return isInt<6>(v);
+  else {
+    if (isInt<6>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can't fit in a 16-bit signed field. This is required to avoid
+    // unnecessary constant extenders.
+    return isConstExtProfitable(Node) && !isInt<16>(v);
+  }
+}]>;
+
+def s6_10ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 6-bit sign extended field.
+    return isInt<6>(v);
+  else {
+    if (isInt<6>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can't fit in a 10-bit signed field. This is required to avoid
+    // unnecessary constant extenders.
+    return isConstExtProfitable(Node) && !isInt<10>(v);
+  }
+}]>;
+
+def s11_0ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 11-bit sign extended field.
+    return isShiftedInt<11,0>(v);
+  else {
+    if (isInt<11>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit signed field.
+    return isConstExtProfitable(Node) && isInt<32>(v);
+  }
+}]>;
+
+def s11_1ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 12-bit sign extended field and
+    // is 2 byte aligned.
+    return isShiftedInt<11,1>(v);
+  else {
+    if (isInt<12>(v))
+      return isShiftedInt<11,1>(v);
+
+    // Return true if extending this immediate is profitable and the low 1 bit
+    // is zero (2-byte aligned).
+    return isConstExtProfitable(Node) && isInt<32>(v) && ((v % 2) == 0);
+  }
+}]>;
+
+def s11_2ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 13-bit sign extended field and
+    // is 4-byte aligned.
+    return isShiftedInt<11,2>(v);
+  else {
+    if (isInt<13>(v))
+      return isShiftedInt<11,2>(v);
+
+    // Return true if extending this immediate is profitable and the low 2-bits
+    // are zero (4-byte aligned).
+    return isConstExtProfitable(Node)  && isInt<32>(v) && ((v % 4) == 0);
+  }
+}]>;
+
+def s11_3ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 14-bit sign extended field and
+    // is 8-byte aligned.
+    return isShiftedInt<11,3>(v);
+  else {
+    if (isInt<14>(v))
+     return isShiftedInt<11,3>(v);
+
+    // Return true if extending this immediate is profitable and the low 3-bits
+    // are zero (8-byte aligned).
+    return isConstExtProfitable(Node)  && isInt<32>(v) && ((v % 8) == 0);
+  }
+}]>;
+
+def u0AlwaysExtPred : PatLeaf<(i32 imm), [{
+  // Predicate for an unsigned 32-bit value that always needs to be extended.
+  if (Subtarget.hasV4TOps()) {
+    if (isConstExtProfitable(Node)) {
+      int64_t v = (int64_t)N->getSExtValue();
+      return isUInt<32>(v);
+    }
+  }
+  return false;
+}]>;
+
+def u6ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 6-bit unsigned field.
+    return isUInt<6>(v);
+  else {
+    if (isUInt<6>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit unsigned field.
+    return isConstExtProfitable(Node) && isUInt<32>(v);
+  }
+}]>;
+
+def u7ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 7-bit unsigned field.
+    return isUInt<7>(v);
+  else {
+    if (isUInt<7>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit unsigned field.
+    return isConstExtProfitable(Node) && isUInt<32>(v);
+  }
+}]>;
+
+def u8ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 8-bit unsigned field.
+    return isUInt<8>(v);
+  else {
+    if (isUInt<8>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit unsigned field.
+    return isConstExtProfitable(Node) && isUInt<32>(v);
+  }
+}]>;
+
+def u9ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 9-bit unsigned field.
+    return isUInt<9>(v);
+  else {
+    if (isUInt<9>(v))
+      return true;
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit unsigned field.
+    return isConstExtProfitable(Node) && isUInt<32>(v);
+  }
+}]>;
+
+def u6_1ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 7-bit unsigned field and
+    // is 2-byte aligned.
+    return isShiftedUInt<6,1>(v);
+  else {
+    if (isUInt<7>(v))
+      return isShiftedUInt<6,1>(v);
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit unsigned field.
+    return isConstExtProfitable(Node) && isUInt<32>(v) && ((v % 2) == 0);
+  }
+}]>;
+
+def u6_2ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 8-bit unsigned field and
+    // is 4-byte aligned.
+    return isShiftedUInt<6,2>(v);
+  else {
+    if (isUInt<8>(v))
+      return isShiftedUInt<6,2>(v);
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit unsigned field.
+    return isConstExtProfitable(Node) && isUInt<32>(v) && ((v % 4) == 0);
+  }
+}]>;
+
+def u6_3ExtPred  : PatLeaf<(i32 imm), [{
+  int64_t v = (int64_t)N->getSExtValue();
+  if (!Subtarget.hasV4TOps())
+    // Return true if the immediate can fit in a 9-bit unsigned field and
+    // is 8-byte aligned.
+    return isShiftedUInt<6,3>(v);
+  else {
+    if (isUInt<9>(v))
+      return isShiftedUInt<6,3>(v);
+
+    // Return true if extending this immediate is profitable and the value
+    // can fit in a 32-bit unsigned field.
+    return isConstExtProfitable(Node) && isUInt<32>(v) && ((v % 8) == 0);
+  }
+}]>;
+
+// Addressing modes.
+
+def ADDRrr : ComplexPattern<i32, 2, "SelectADDRrr", [], []>;
+def ADDRri : ComplexPattern<i32, 2, "SelectADDRri", [frameindex], []>;
+def ADDRriS11_0 : ComplexPattern<i32, 2, "SelectADDRriS11_0", [frameindex], []>;
+def ADDRriS11_1 : ComplexPattern<i32, 2, "SelectADDRriS11_1", [frameindex], []>;
+def ADDRriS11_2 : ComplexPattern<i32, 2, "SelectADDRriS11_2", [frameindex], []>;
+def ADDRriS11_3 : ComplexPattern<i32, 2, "SelectADDRriS11_3", [frameindex], []>;
+def ADDRriU6_0 : ComplexPattern<i32, 2, "SelectADDRriU6_0", [frameindex], []>;
+def ADDRriU6_1 : ComplexPattern<i32, 2, "SelectADDRriU6_1", [frameindex], []>;
+def ADDRriU6_2 : ComplexPattern<i32, 2, "SelectADDRriU6_2", [frameindex], []>;
+
+// Address operands.
+
+def MEMrr : Operand<i32> {
+  let PrintMethod = "printMEMrrOperand";
+  let MIOperandInfo = (ops IntRegs, IntRegs);
+}
+
+def MEMri : Operand<i32> {
+  let PrintMethod = "printMEMriOperand";
+  let MIOperandInfo = (ops IntRegs, IntRegs);
+}
+
+def MEMri_s11_2 : Operand<i32>,
+  ComplexPattern<i32, 2, "SelectMEMriS11_2", []> {
+  let PrintMethod = "printMEMriOperand";
+  let MIOperandInfo = (ops IntRegs, s11Imm);
+}
+
+def FrameIndex : Operand<i32> {
+  let PrintMethod = "printFrameIndexOperand";
+  let MIOperandInfo = (ops IntRegs, s11Imm);
+}
+
+let PrintMethod = "printGlobalOperand" in {
+  def globaladdress : Operand<i32>;
+  def globaladdressExt : Operand<i32>;
+}
+
+let PrintMethod = "printJumpTable" in
+def jumptablebase : Operand<i32>;
+
+def brtarget : Operand<OtherVT>;
+def brtargetExt : Operand<OtherVT>;
+def calltarget : Operand<i32>;
+
+def bblabel : Operand<i32>;
+def bbl   : SDNode<"ISD::BasicBlock", SDTPtrLeaf   , [], "BasicBlockSDNode">;
+
+def symbolHi32 : Operand<i32> {
+  let PrintMethod = "printSymbolHi";
+}
+def symbolLo32 : Operand<i32> {
+  let PrintMethod = "printSymbolLo";
+}
diff --git a/lib/Target/Hexagon/HexagonPeephole.cpp b/lib/Target/Hexagon/HexagonPeephole.cpp
index a295015de561..576f1d7d0790 100644
--- a/lib/Target/Hexagon/HexagonPeephole.cpp
+++ b/lib/Target/Hexagon/HexagonPeephole.cpp
@@ -38,21 +38,21 @@
 #define DEBUG_TYPE "hexagon-peephole"
 #include "Hexagon.h"
 #include "HexagonTargetMachine.h"
-#include "llvm/Constants.h"
-#include "llvm/PassSupport.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/Statistic.h"
-#include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/PassSupport.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/Target/TargetInstrInfo.h"
 #include <algorithm>
 
 using namespace llvm;
diff --git a/lib/Target/Hexagon/HexagonRegisterInfo.cpp b/lib/Target/Hexagon/HexagonRegisterInfo.cpp
index 37424860564f..d8b4e2fcb368 100644
--- a/lib/Target/Hexagon/HexagonRegisterInfo.cpp
+++ b/lib/Target/Hexagon/HexagonRegisterInfo.cpp
@@ -17,8 +17,6 @@
 #include "HexagonSubtarget.h"
 #include "HexagonTargetMachine.h"
 #include "HexagonMachineFunctionInfo.h"
-#include "llvm/Function.h"
-#include "llvm/Type.h"
 #include "llvm/ADT/BitVector.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
@@ -26,7 +24,10 @@
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/PseudoSourceValue.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Type.h"
 #include "llvm/MC/MachineLocation.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetMachine.h"
@@ -117,37 +118,15 @@ HexagonRegisterInfo::getCalleeSavedRegClasses(const MachineFunction *MF) const {
                    "architecture version");
 }
 
-void HexagonRegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  MachineInstr &MI = *I;
-
-  if (MI.getOpcode() == Hexagon::ADJCALLSTACKDOWN) {
-    // Hexagon_TODO: add code
-  } else if (MI.getOpcode() == Hexagon::ADJCALLSTACKUP) {
-    // Hexagon_TODO: add code
-  } else {
-    llvm_unreachable("Cannot handle this call frame pseudo instruction");
-  }
-  MBB.erase(I);
-}
-
 void HexagonRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
-                                            int SPAdj, RegScavenger *RS) const {
-
+                                              int SPAdj, unsigned FIOperandNum,
+                                              RegScavenger *RS) const {
   //
   // Hexagon_TODO: Do we need to enforce this for Hexagon?
   assert(SPAdj == 0 && "Unexpected");
 
-
-  unsigned i = 0;
   MachineInstr &MI = *II;
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
-  }
-
-  int FrameIndex = MI.getOperand(i).getIndex();
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
 
   // Addressable stack objects are accessed using neg. offsets from %fp.
   MachineFunction &MF = *MI.getParent()->getParent();
@@ -167,8 +146,9 @@ void HexagonRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
       TII.isValidOffset(MI.getOpcode(), (FrameSize+Offset)) &&
       !TII.isSpillPredRegOp(&MI)) {
     // Replace frame index with a stack pointer reference.
-    MI.getOperand(i).ChangeToRegister(getStackRegister(), false, false, true);
-    MI.getOperand(i+1).ChangeToImmediate(FrameSize+Offset);
+    MI.getOperand(FIOperandNum).ChangeToRegister(getStackRegister(), false,
+                                                 false, true);
+    MI.getOperand(FIOperandNum + 1).ChangeToImmediate(FrameSize+Offset);
   } else {
     // Replace frame index with a frame pointer reference.
     if (!TII.isValidOffset(MI.getOpcode(), Offset)) {
@@ -205,8 +185,8 @@ void HexagonRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
                   dstReg).addReg(FrameReg).addImm(Offset);
         }
 
-        MI.getOperand(i).ChangeToRegister(dstReg, false, false, true);
-        MI.getOperand(i+1).ChangeToImmediate(0);
+        MI.getOperand(FIOperandNum).ChangeToRegister(dstReg, false, false,true);
+        MI.getOperand(FIOperandNum+1).ChangeToImmediate(0);
       } else if ((MI.getOpcode() == Hexagon::STriw_indexed) ||
                  (MI.getOpcode() == Hexagon::STriw) ||
                  (MI.getOpcode() == Hexagon::STrid) ||
@@ -233,29 +213,44 @@ void HexagonRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
                   TII.get(Hexagon::ADD_ri),
                   resReg).addReg(FrameReg).addImm(Offset);
         }
-        MI.getOperand(i).ChangeToRegister(resReg, false, false, true);
-        MI.getOperand(i+1).ChangeToImmediate(0);
+        MI.getOperand(FIOperandNum).ChangeToRegister(resReg, false, false,true);
+        MI.getOperand(FIOperandNum+1).ChangeToImmediate(0);
       } else if (TII.isMemOp(&MI)) {
-        unsigned resReg = HEXAGON_RESERVED_REG_1;
-        if (!MFI.hasVarSizedObjects() &&
-            TII.isValidOffset(MI.getOpcode(), (FrameSize+Offset))) {
-          MI.getOperand(i).ChangeToRegister(getStackRegister(), false, false,
-                                            true);
-          MI.getOperand(i+1).ChangeToImmediate(FrameSize+Offset);
-        } else if (!TII.isValidOffset(Hexagon::ADD_ri, Offset)) {
-          BuildMI(*MI.getParent(), II, MI.getDebugLoc(),
-                  TII.get(Hexagon::CONST32_Int_Real), resReg).addImm(Offset);
-          BuildMI(*MI.getParent(), II, MI.getDebugLoc(),
-                  TII.get(Hexagon::ADD_rr),
-                  resReg).addReg(FrameReg).addReg(resReg);
-          MI.getOperand(i).ChangeToRegister(resReg, false, false, true);
-          MI.getOperand(i+1).ChangeToImmediate(0);
+        // use the constant extender if the instruction provides it
+        // and we are V4TOps.
+        if (Subtarget.hasV4TOps()) {
+          if (TII.isConstExtended(&MI)) {
+            MI.getOperand(FIOperandNum).ChangeToRegister(FrameReg, false);
+            MI.getOperand(FIOperandNum+1).ChangeToImmediate(Offset);
+            TII.immediateExtend(&MI);
+          } else {
+            llvm_unreachable("Need to implement for memops");
+          }
         } else {
-          BuildMI(*MI.getParent(), II, MI.getDebugLoc(),
-                  TII.get(Hexagon::ADD_ri),
-                  resReg).addReg(FrameReg).addImm(Offset);
-          MI.getOperand(i).ChangeToRegister(resReg, false, false, true);
-          MI.getOperand(i+1).ChangeToImmediate(0);
+          // Only V3 and older instructions here.
+          unsigned ResReg = HEXAGON_RESERVED_REG_1;
+          if (!MFI.hasVarSizedObjects() &&
+              TII.isValidOffset(MI.getOpcode(), (FrameSize+Offset))) {
+            MI.getOperand(FIOperandNum).ChangeToRegister(getStackRegister(),
+                                                         false, false, false);
+            MI.getOperand(FIOperandNum+1).ChangeToImmediate(FrameSize+Offset);
+          } else if (!TII.isValidOffset(Hexagon::ADD_ri, Offset)) {
+            BuildMI(*MI.getParent(), II, MI.getDebugLoc(),
+                    TII.get(Hexagon::CONST32_Int_Real), ResReg).addImm(Offset);
+            BuildMI(*MI.getParent(), II, MI.getDebugLoc(),
+                    TII.get(Hexagon::ADD_rr), ResReg).addReg(FrameReg).
+              addReg(ResReg);
+            MI.getOperand(FIOperandNum).ChangeToRegister(ResReg, false, false,
+                                                         true);
+            MI.getOperand(FIOperandNum+1).ChangeToImmediate(0);
+          } else {
+            BuildMI(*MI.getParent(), II, MI.getDebugLoc(),
+                    TII.get(Hexagon::ADD_ri), ResReg).addReg(FrameReg).
+              addImm(Offset);
+            MI.getOperand(FIOperandNum).ChangeToRegister(ResReg, false, false,
+                                                         true);
+            MI.getOperand(FIOperandNum+1).ChangeToImmediate(0);
+          }
         }
       } else {
         unsigned dstReg = MI.getOperand(0).getReg();
@@ -265,14 +260,14 @@ void HexagonRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
                 TII.get(Hexagon::ADD_rr),
                 dstReg).addReg(FrameReg).addReg(dstReg);
         // Can we delete MI??? r2 = add (r2, #0).
-        MI.getOperand(i).ChangeToRegister(dstReg, false, false, true);
-        MI.getOperand(i+1).ChangeToImmediate(0);
+        MI.getOperand(FIOperandNum).ChangeToRegister(dstReg, false, false,true);
+        MI.getOperand(FIOperandNum+1).ChangeToImmediate(0);
       }
     } else {
       // If the offset is small enough to fit in the immediate field, directly
       // encode it.
-      MI.getOperand(i).ChangeToRegister(FrameReg, false);
-      MI.getOperand(i+1).ChangeToImmediate(Offset);
+      MI.getOperand(FIOperandNum).ChangeToRegister(FrameReg, false);
+      MI.getOperand(FIOperandNum+1).ChangeToImmediate(Offset);
     }
   }
 
@@ -310,58 +305,6 @@ void HexagonRegisterInfo::getInitialFrameState(std::vector<MachineMove>
   Moves.push_back(MachineMove(0, Dst, Src));
 }
 
-// Get the weight in units of pressure for this register class.
-const RegClassWeight &
-HexagonRegisterInfo::getRegClassWeight(const TargetRegisterClass *RC) const {
-  // Each TargetRegisterClass has a per register weight, and weight
-  // limit which must be less than the limits of its pressure sets.
-  static const RegClassWeight RCWeightTable[] = {
-    {1, 32}, // IntRegs
-    {1, 8},  // CRRegs
-    {1, 4},  // PredRegs
-    {2, 16}, // DoubleRegs
-    {0, 0} };
-  return RCWeightTable[RC->getID()];
-}
-
-/// Get the number of dimensions of register pressure.
-unsigned HexagonRegisterInfo::getNumRegPressureSets() const {
-  return 4;
-}
-
-/// Get the name of this register unit pressure set.
-const char *HexagonRegisterInfo::getRegPressureSetName(unsigned Idx) const {
-  static const char *const RegPressureSetName[] = {
-    "IntRegsRegSet",
-    "CRRegsRegSet",
-    "PredRegsRegSet",
-    "DoubleRegsRegSet"
-  };
-  assert((Idx < 4) && "Index out of bounds");
-  return RegPressureSetName[Idx];
-}
-
-/// Get the register unit pressure limit for this dimension.
-/// This limit must be adjusted dynamically for reserved registers.
-unsigned HexagonRegisterInfo::getRegPressureSetLimit(unsigned Idx) const {
-  static const int RegPressureLimit [] = { 16, 4, 2, 8 };
-  assert((Idx < 4) && "Index out of bounds");
-  return RegPressureLimit[Idx];
-}
-
-const int*
-HexagonRegisterInfo::getRegClassPressureSets(const TargetRegisterClass *RC)
-  const {
-  static const int RCSetsTable[] = {
-    0,  -1,  // IntRegs
-    1,  -1,  // CRRegs
-    2,  -1,  // PredRegs
-    0,  -1,  // DoubleRegs
-    -1 };
-  static const unsigned RCSetStartTable[] = { 0, 2, 4, 6, 0 };
-  unsigned SetListStart = RCSetStartTable[RC->getID()];
-  return &RCSetsTable[SetListStart];
-}
 unsigned HexagonRegisterInfo::getEHExceptionRegister() const {
   llvm_unreachable("What is the exception register");
 }
diff --git a/lib/Target/Hexagon/HexagonRegisterInfo.h b/lib/Target/Hexagon/HexagonRegisterInfo.h
index 8820d13e0122..8a3f94a3fd12 100644
--- a/lib/Target/Hexagon/HexagonRegisterInfo.h
+++ b/lib/Target/Hexagon/HexagonRegisterInfo.h
@@ -15,8 +15,8 @@
 #ifndef HexagonREGISTERINFO_H
 #define HexagonREGISTERINFO_H
 
-#include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/MC/MachineLocation.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 
 #define GET_REGINFO_HEADER
 #include "HexagonGenRegisterInfo.inc"
@@ -56,12 +56,9 @@ struct HexagonRegisterInfo : public HexagonGenRegisterInfo {
 
   BitVector getReservedRegs(const MachineFunction &MF) const;
 
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
-
   void eliminateFrameIndex(MachineBasicBlock::iterator II,
-                           int SPAdj, RegScavenger *RS = NULL) const;
+                           int SPAdj, unsigned FIOperandNum,
+                           RegScavenger *RS = NULL) const;
 
   /// determineFrameLayout - Determine the size of the frame and maximum call
   /// frame size.
@@ -87,11 +84,6 @@ struct HexagonRegisterInfo : public HexagonGenRegisterInfo {
   // Exception handling queries.
   unsigned getEHExceptionRegister() const;
   unsigned getEHHandlerRegister() const;
-  const RegClassWeight &getRegClassWeight(const TargetRegisterClass *RC) const;
-  unsigned getNumRegPressureSets() const;
-  const char *getRegPressureSetName(unsigned Idx) const;
-  unsigned getRegPressureSetLimit(unsigned Idx) const;
-  const int* getRegClassPressureSets(const TargetRegisterClass *RC) const;
 };
 
 } // end namespace llvm
diff --git a/lib/Target/Hexagon/HexagonRemoveSZExtArgs.cpp b/lib/Target/Hexagon/HexagonRemoveSZExtArgs.cpp
index 4d93dd18d4e0..34bf4eacfdc0 100644
--- a/lib/Target/Hexagon/HexagonRemoveSZExtArgs.cpp
+++ b/lib/Target/Hexagon/HexagonRemoveSZExtArgs.cpp
@@ -12,11 +12,12 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "Hexagon.h"
 #include "HexagonTargetMachine.h"
-#include "llvm/Function.h"
-#include "llvm/Instructions.h"
-#include "llvm/Pass.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/Pass.h"
 #include "llvm/Transforms/Scalar.h"
 
 using namespace llvm;
@@ -50,7 +51,7 @@ bool HexagonRemoveExtendArgs::runOnFunction(Function &F) {
   unsigned Idx = 1;
   for (Function::arg_iterator AI = F.arg_begin(), AE = F.arg_end(); AI != AE;
        ++AI, ++Idx) {
-    if (F.getParamAttributes(Idx).hasAttribute(Attributes::SExt)) {
+    if (F.getAttributes().hasAttribute(Idx, Attribute::SExt)) {
       Argument* Arg = AI;
       if (!isa<PointerType>(Arg->getType())) {
         for (Instruction::use_iterator UI = Arg->use_begin();
diff --git a/lib/Target/Hexagon/HexagonSchedule.td b/lib/Target/Hexagon/HexagonSchedule.td
index b5ff69a701cd..c2cfbb9710a6 100644
--- a/lib/Target/Hexagon/HexagonSchedule.td
+++ b/lib/Target/Hexagon/HexagonSchedule.td
@@ -8,10 +8,11 @@
 //===----------------------------------------------------------------------===//
 
 // Functional Units
-def LUNIT     : FuncUnit;
-def LSUNIT    : FuncUnit;
-def MUNIT     : FuncUnit;
-def SUNIT     : FuncUnit;
+def LSUNIT    : FuncUnit; // SLOT0
+def LUNIT     : FuncUnit; // SLOT1
+def MUNIT     : FuncUnit; // SLOT2
+def SUNIT     : FuncUnit; // SLOT3
+def LOOPUNIT  : FuncUnit;
 
 // Itinerary classes
 def ALU32     : InstrItinClass;
@@ -20,27 +21,34 @@ def CR        : InstrItinClass;
 def J         : InstrItinClass;
 def JR        : InstrItinClass;
 def LD        : InstrItinClass;
+def LD0       : InstrItinClass;
 def M         : InstrItinClass;
 def ST        : InstrItinClass;
+def ST0       : InstrItinClass;
 def S         : InstrItinClass;
 def SYS       : InstrItinClass;
-def MARKER    : InstrItinClass;
+def ENDLOOP   : InstrItinClass;
 def PSEUDO    : InstrItinClass;
+def PSEUDOM   : InstrItinClass;
 
 def HexagonItineraries :
-      ProcessorItineraries<[LUNIT, LSUNIT, MUNIT, SUNIT], [], [
+      ProcessorItineraries<[LSUNIT, LUNIT, MUNIT, SUNIT, LOOPUNIT], [], [
         InstrItinData<ALU32  , [InstrStage<1, [LUNIT, LSUNIT, MUNIT, SUNIT]>]>,
         InstrItinData<ALU64  , [InstrStage<1, [MUNIT, SUNIT]>]>,
         InstrItinData<CR     , [InstrStage<1, [SUNIT]>]>,
         InstrItinData<J      , [InstrStage<1, [SUNIT, MUNIT]>]>,
         InstrItinData<JR     , [InstrStage<1, [MUNIT]>]>,
         InstrItinData<LD     , [InstrStage<1, [LUNIT, LSUNIT]>]>,
+        InstrItinData<LD0    , [InstrStage<1, [LSUNIT]>]>,
         InstrItinData<M      , [InstrStage<1, [MUNIT, SUNIT]>]>,
         InstrItinData<ST     , [InstrStage<1, [LSUNIT]>]>,
+        InstrItinData<ST0    , [InstrStage<1, [LSUNIT]>]>,
         InstrItinData<S      , [InstrStage<1, [SUNIT, MUNIT]>]>,
         InstrItinData<SYS    , [InstrStage<1, [LSUNIT]>]>,
-        InstrItinData<MARKER , [InstrStage<1, [LUNIT, LSUNIT, MUNIT, SUNIT]>]>,
-        InstrItinData<PSEUDO , [InstrStage<1, [LUNIT, LSUNIT, MUNIT, SUNIT]>]>
+        InstrItinData<ENDLOOP, [InstrStage<1, [LOOPUNIT]>]>,
+        InstrItinData<PSEUDO , [InstrStage<1, [LUNIT, LSUNIT, MUNIT, SUNIT]>]>,
+        InstrItinData<PSEUDOM, [InstrStage<1, [MUNIT, SUNIT], 0>,
+                                InstrStage<1, [MUNIT, SUNIT]>]>
       ]>;
 
 def HexagonModel : SchedMachineModel {
diff --git a/lib/Target/Hexagon/HexagonScheduleV4.td b/lib/Target/Hexagon/HexagonScheduleV4.td
index 5668ae81e82e..ef72cf4068bf 100644
--- a/lib/Target/Hexagon/HexagonScheduleV4.td
+++ b/lib/Target/Hexagon/HexagonScheduleV4.td
@@ -28,6 +28,10 @@ def SLOT0       : FuncUnit;
 def SLOT1       : FuncUnit;
 def SLOT2       : FuncUnit;
 def SLOT3       : FuncUnit;
+// Endloop is a pseudo instruction that is encoded with 2 bits in a packet
+// rather than taking an execution slot. This special unit is needed
+// to schedule an ENDLOOP with 4 other instructions.
+def SLOT_ENDLOOP: FuncUnit;
 
 // Itinerary classes.
 def NV_V4       : InstrItinClass;
@@ -36,22 +40,26 @@ def MEM_V4      : InstrItinClass;
 def PREFIX      : InstrItinClass;
 
 def HexagonItinerariesV4 :
-      ProcessorItineraries<[SLOT0, SLOT1, SLOT2, SLOT3], [], [
+      ProcessorItineraries<[SLOT0, SLOT1, SLOT2, SLOT3, SLOT_ENDLOOP], [], [
         InstrItinData<ALU32  , [InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
         InstrItinData<ALU64  , [InstrStage<1, [SLOT2, SLOT3]>]>,
         InstrItinData<CR     , [InstrStage<1, [SLOT3]>]>,
         InstrItinData<J      , [InstrStage<1, [SLOT2, SLOT3]>]>,
         InstrItinData<JR     , [InstrStage<1, [SLOT2]>]>,
         InstrItinData<LD     , [InstrStage<1, [SLOT0, SLOT1]>]>,
+        InstrItinData<LD0    , [InstrStage<1, [SLOT0]>]>,
         InstrItinData<M      , [InstrStage<1, [SLOT2, SLOT3]>]>,
         InstrItinData<ST     , [InstrStage<1, [SLOT0, SLOT1]>]>,
+        InstrItinData<ST0    , [InstrStage<1, [SLOT0]>]>,
         InstrItinData<S      , [InstrStage<1, [SLOT2, SLOT3]>]>,
         InstrItinData<SYS    , [InstrStage<1, [SLOT0]>]>,
         InstrItinData<NV_V4  , [InstrStage<1, [SLOT0]>]>,
         InstrItinData<MEM_V4 , [InstrStage<1, [SLOT0]>]>,
-        InstrItinData<MARKER , [InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
+        InstrItinData<ENDLOOP, [InstrStage<1, [SLOT_ENDLOOP]>]>,
         InstrItinData<PREFIX , [InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
-        InstrItinData<PSEUDO , [InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>
+        InstrItinData<PSEUDO , [InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
+        InstrItinData<PSEUDOM, [InstrStage<1, [SLOT2, SLOT3], 0>,
+                                InstrStage<1, [SLOT2, SLOT3]>]>
       ]>;
 
 def HexagonModelV4 : SchedMachineModel {
diff --git a/lib/Target/Hexagon/HexagonSplitTFRCondSets.cpp b/lib/Target/Hexagon/HexagonSplitTFRCondSets.cpp
index a81cd913a6ec..814249fa6832 100644
--- a/lib/Target/Hexagon/HexagonSplitTFRCondSets.cpp
+++ b/lib/Target/Hexagon/HexagonSplitTFRCondSets.cpp
@@ -27,24 +27,25 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "xfer"
-#include "HexagonTargetMachine.h"
-#include "HexagonSubtarget.h"
+#include "Hexagon.h"
 #include "HexagonMachineFunctionInfo.h"
-#include "llvm/CodeGen/Passes.h"
+#include "HexagonSubtarget.h"
+#include "HexagonTargetMachine.h"
 #include "llvm/CodeGen/LatencyPriorityQueue.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineLoopInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/ScheduleHazardRecognizer.h"
 #include "llvm/CodeGen/SchedulerRegistry.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/MathExtras.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 
 using namespace llvm;
 
diff --git a/lib/Target/Hexagon/HexagonSubtarget.cpp b/lib/Target/Hexagon/HexagonSubtarget.cpp
index 4bacb8fa670d..07d5ce1d8ab0 100644
--- a/lib/Target/Hexagon/HexagonSubtarget.cpp
+++ b/lib/Target/Hexagon/HexagonSubtarget.cpp
@@ -29,8 +29,16 @@ EnableV3("enable-hexagon-v3", cl::Hidden,
 static cl::opt<bool>
 EnableMemOps(
     "enable-hexagon-memops",
-    cl::Hidden, cl::ZeroOrMore, cl::ValueDisallowed,
-    cl::desc("Generate V4 memop instructions."));
+    cl::Hidden, cl::ZeroOrMore, cl::ValueDisallowed, cl::init(true),
+    cl::desc(
+      "Generate V4 MEMOP in code generation for Hexagon target"));
+
+static cl::opt<bool>
+DisableMemOps(
+    "disable-hexagon-memops",
+    cl::Hidden, cl::ZeroOrMore, cl::ValueDisallowed, cl::init(false),
+    cl::desc(
+      "Do not generate V4 MEMOP in code generation for Hexagon target"));
 
 static cl::opt<bool>
 EnableIEEERndNear(
@@ -64,7 +72,10 @@ HexagonSubtarget::HexagonSubtarget(StringRef TT, StringRef CPU, StringRef FS):
   // Initialize scheduling itinerary for the specified CPU.
   InstrItins = getInstrItineraryForCPU(CPUString);
 
-  if (EnableMemOps)
+  // UseMemOps on by default unless disabled explicitly
+  if (DisableMemOps)
+    UseMemOps = false;
+  else if (EnableMemOps)
     UseMemOps = true;
   else
     UseMemOps = false;
diff --git a/lib/Target/Hexagon/HexagonSubtarget.h b/lib/Target/Hexagon/HexagonSubtarget.h
index 5d9d6d890d98..76a8fba195f3 100644
--- a/lib/Target/Hexagon/HexagonSubtarget.h
+++ b/lib/Target/Hexagon/HexagonSubtarget.h
@@ -14,8 +14,8 @@
 #ifndef Hexagon_SUBTARGET_H
 #define Hexagon_SUBTARGET_H
 
-#include "llvm/Target/TargetSubtargetInfo.h"
 #include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
 #include <string>
 
 #define GET_SUBTARGETINFO_HEADER
diff --git a/lib/Target/Hexagon/HexagonTargetMachine.cpp b/lib/Target/Hexagon/HexagonTargetMachine.cpp
index 30866e9eeba8..ce45c626f799 100644
--- a/lib/Target/Hexagon/HexagonTargetMachine.cpp
+++ b/lib/Target/Hexagon/HexagonTargetMachine.cpp
@@ -15,13 +15,13 @@
 #include "Hexagon.h"
 #include "HexagonISelLowering.h"
 #include "HexagonMachineScheduler.h"
-#include "llvm/Module.h"
 #include "llvm/CodeGen/Passes.h"
+#include "llvm/IR/Module.h"
 #include "llvm/PassManager.h"
-#include "llvm/Transforms/IPO/PassManagerBuilder.h"
-#include "llvm/Transforms/Scalar.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/TargetRegistry.h"
+#include "llvm/Transforms/IPO/PassManagerBuilder.h"
+#include "llvm/Transforms/Scalar.h"
 
 using namespace llvm;
 
@@ -35,6 +35,10 @@ opt<bool> DisableHexagonMISched("disable-hexagon-misched",
                                 cl::Hidden, cl::ZeroOrMore, cl::init(false),
                                 cl::desc("Disable Hexagon MI Scheduling"));
 
+static cl::opt<bool> DisableHexagonCFGOpt("disable-hexagon-cfgopt",
+    cl::Hidden, cl::ZeroOrMore, cl::init(false),
+    cl::desc("Disable Hexagon CFG Optimization"));
+
 /// HexagonTargetMachineModule - Note that this is used on hosts that
 /// cannot link in a library unless there are references into the
 /// library.  In particular, it seems that it is not possible to get
@@ -74,21 +78,21 @@ HexagonTargetMachine::HexagonTargetMachine(const Target &T, StringRef TT,
     Subtarget(TT, CPU, FS), InstrInfo(Subtarget), TLInfo(*this),
     TSInfo(*this),
     FrameLowering(Subtarget),
-    InstrItins(&Subtarget.getInstrItineraryData()),
-    STTI(&TLInfo), VTTI(&TLInfo) {
-  setMCUseCFI(false);
+    InstrItins(&Subtarget.getInstrItineraryData()) {
+    setMCUseCFI(false);
 }
 
 // addPassesForOptimizations - Allow the backend (target) to add Target
 // Independent Optimization passes to the Pass Manager.
 bool HexagonTargetMachine::addPassesForOptimizations(PassManagerBase &PM) {
-
-  PM.add(createConstantPropagationPass());
-  PM.add(createLoopSimplifyPass());
-  PM.add(createDeadCodeEliminationPass());
-  PM.add(createConstantPropagationPass());
-  PM.add(createLoopUnrollPass());
-  PM.add(createLoopStrengthReducePass());
+  if (getOptLevel() != CodeGenOpt::None) {
+    PM.add(createConstantPropagationPass());
+    PM.add(createLoopSimplifyPass());
+    PM.add(createDeadCodeEliminationPass());
+    PM.add(createConstantPropagationPass());
+    PM.add(createLoopUnrollPass());
+    PM.add(createLoopStrengthReducePass());
+  }
   return true;
 }
 
@@ -122,38 +126,45 @@ TargetPassConfig *HexagonTargetMachine::createPassConfig(PassManagerBase &PM) {
 }
 
 bool HexagonPassConfig::addInstSelector() {
-  addPass(createHexagonRemoveExtendOps(getHexagonTargetMachine()));
-  addPass(createHexagonISelDag(getHexagonTargetMachine()));
-  addPass(createHexagonPeephole());
+
+  if (getOptLevel() != CodeGenOpt::None)
+    addPass(createHexagonRemoveExtendOps(getHexagonTargetMachine()));
+
+  addPass(createHexagonISelDag(getHexagonTargetMachine(), getOptLevel()));
+
+  if (getOptLevel() != CodeGenOpt::None)
+    addPass(createHexagonPeephole());
+
   return false;
 }
 
 
 bool HexagonPassConfig::addPreRegAlloc() {
-  if (!DisableHardwareLoops) {
+  if (!DisableHardwareLoops && getOptLevel() != CodeGenOpt::None)
     addPass(createHexagonHardwareLoops());
-  }
   return false;
 }
 
 bool HexagonPassConfig::addPostRegAlloc() {
-  addPass(createHexagonCFGOptimizer(getHexagonTargetMachine()));
+  if (!DisableHexagonCFGOpt && getOptLevel() != CodeGenOpt::None)
+    addPass(createHexagonCFGOptimizer(getHexagonTargetMachine()));
   return true;
 }
 
 
 bool HexagonPassConfig::addPreSched2() {
-  addPass(&IfConverterID);
+  if (getOptLevel() != CodeGenOpt::None)
+    addPass(&IfConverterID);
   return true;
 }
 
 bool HexagonPassConfig::addPreEmitPass() {
 
-  if (!DisableHardwareLoops) {
+  if (!DisableHardwareLoops && getOptLevel() != CodeGenOpt::None)
     addPass(createHexagonFixupHwLoops());
-  }
 
-  addPass(createHexagonNewValueJump());
+  if (getOptLevel() != CodeGenOpt::None)
+    addPass(createHexagonNewValueJump());
 
   // Expand Spill code for predicate registers.
   addPass(createHexagonExpandPredSpillCode(getHexagonTargetMachine()));
@@ -162,7 +173,8 @@ bool HexagonPassConfig::addPreEmitPass() {
   addPass(createHexagonSplitTFRCondSets(getHexagonTargetMachine()));
 
   // Create Packets.
-  addPass(createHexagonPacketizer());
+  if (getOptLevel() != CodeGenOpt::None)
+    addPass(createHexagonPacketizer());
 
   return false;
 }
diff --git a/lib/Target/Hexagon/HexagonTargetMachine.h b/lib/Target/Hexagon/HexagonTargetMachine.h
index 7a4215c119a9..cf8f9aa3612f 100644
--- a/lib/Target/Hexagon/HexagonTargetMachine.h
+++ b/lib/Target/Hexagon/HexagonTargetMachine.h
@@ -14,14 +14,13 @@
 #ifndef HexagonTARGETMACHINE_H
 #define HexagonTARGETMACHINE_H
 
-#include "HexagonInstrInfo.h"
-#include "HexagonSubtarget.h"
+#include "HexagonFrameLowering.h"
 #include "HexagonISelLowering.h"
+#include "HexagonInstrInfo.h"
 #include "HexagonSelectionDAGInfo.h"
-#include "HexagonFrameLowering.h"
+#include "HexagonSubtarget.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Target/TargetMachine.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetTransformImpl.h"
 
 namespace llvm {
 
@@ -35,8 +34,6 @@ class HexagonTargetMachine : public LLVMTargetMachine {
   HexagonSelectionDAGInfo TSInfo;
   HexagonFrameLowering FrameLowering;
   const InstrItineraryData* InstrItins;
-  ScalarTargetTransformImpl STTI;
-  VectorTargetTransformImpl VTTI;
 
 public:
   HexagonTargetMachine(const Target &T, StringRef TT,StringRef CPU,
@@ -71,14 +68,6 @@ public:
     return &TSInfo;
   }
 
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
-
   virtual const DataLayout       *getDataLayout() const { return &DL; }
   static unsigned getModuleMatchQuality(const Module &M);
 
diff --git a/lib/Target/Hexagon/HexagonTargetObjectFile.cpp b/lib/Target/Hexagon/HexagonTargetObjectFile.cpp
index f4d7761ac358..993fcfaed43e 100644
--- a/lib/Target/Hexagon/HexagonTargetObjectFile.cpp
+++ b/lib/Target/Hexagon/HexagonTargetObjectFile.cpp
@@ -14,13 +14,13 @@
 #include "HexagonTargetObjectFile.h"
 #include "HexagonSubtarget.h"
 #include "HexagonTargetMachine.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/DataLayout.h"
-#include "llvm/DerivedTypes.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
 #include "llvm/MC/MCContext.h"
-#include "llvm/Support/ELF.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Support/ELF.h"
 
 using namespace llvm;
 
diff --git a/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp b/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
index 3d5f685028ea..c0d86da1c05e 100644
--- a/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
+++ b/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
@@ -149,7 +149,6 @@ namespace {
     bool canReserveResourcesForConstExt(MachineInstr *MI);
     void reserveResourcesForConstExt(MachineInstr* MI);
     bool isNewValueInst(MachineInstr* MI);
-    bool isDotNewInst(MachineInstr* MI);
   };
 }
 
@@ -242,8 +241,9 @@ static bool IsIndirectCall(MachineInstr* MI) {
 // reservation fail.
 void HexagonPacketizerList::reserveResourcesForConstExt(MachineInstr* MI) {
   const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII;
-  MachineInstr *PseudoMI = MI->getParent()->getParent()->CreateMachineInstr(
-                                  QII->get(Hexagon::IMMEXT), MI->getDebugLoc());
+  MachineFunction *MF = MI->getParent()->getParent();
+  MachineInstr *PseudoMI = MF->CreateMachineInstr(QII->get(Hexagon::IMMEXT_i),
+                                                  MI->getDebugLoc());
 
   if (ResourceTracker->canReserveResources(PseudoMI)) {
     ResourceTracker->reserveResources(PseudoMI);
@@ -257,10 +257,10 @@ void HexagonPacketizerList::reserveResourcesForConstExt(MachineInstr* MI) {
 
 bool HexagonPacketizerList::canReserveResourcesForConstExt(MachineInstr *MI) {
   const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII;
-  assert(QII->isExtended(MI) &&
+  assert((QII->isExtended(MI) || QII->isConstExtended(MI)) &&
          "Should only be called for constant extended instructions");
   MachineFunction *MF = MI->getParent()->getParent();
-  MachineInstr *PseudoMI = MF->CreateMachineInstr(QII->get(Hexagon::IMMEXT),
+  MachineInstr *PseudoMI = MF->CreateMachineInstr(QII->get(Hexagon::IMMEXT_i),
                                                   MI->getDebugLoc());
   bool CanReserve = ResourceTracker->canReserveResources(PseudoMI);
   MF->DeleteMachineInstr(PseudoMI);
@@ -271,8 +271,9 @@ bool HexagonPacketizerList::canReserveResourcesForConstExt(MachineInstr *MI) {
 // true, otherwise, return false.
 bool HexagonPacketizerList::tryAllocateResourcesForConstExt(MachineInstr* MI) {
   const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII;
-  MachineInstr *PseudoMI = MI->getParent()->getParent()->CreateMachineInstr(
-                                  QII->get(Hexagon::IMMEXT), MI->getDebugLoc());
+  MachineFunction *MF = MI->getParent()->getParent();
+  MachineInstr *PseudoMI = MF->CreateMachineInstr(QII->get(Hexagon::IMMEXT_i),
+                                                  MI->getDebugLoc());
 
   if (ResourceTracker->canReserveResources(PseudoMI)) {
     ResourceTracker->reserveResources(PseudoMI);
@@ -349,17 +350,6 @@ static bool IsControlFlow(MachineInstr* MI) {
   return (MI->getDesc().isTerminator() || MI->getDesc().isCall());
 }
 
-bool HexagonPacketizerList::isNewValueInst(MachineInstr* MI) {
-  const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII;
-  if (QII->isNewValueJump(MI))
-    return true;
-
-  if (QII->isNewValueStore(MI))
-    return true;
-
-  return false;
-}
-
 // Function returns true if an instruction can be promoted to the new-value
 // store. It will always return false for v2 and v3.
 // It lists all the conditional and unconditional stores that can be promoted
@@ -375,7 +365,6 @@ bool HexagonPacketizerList::IsNewifyStore (MachineInstr* MI) {
     case Hexagon::STrib_indexed:
     case Hexagon::STrib_indexed_shl_V4:
     case Hexagon::STrib_shl_V4:
-    case Hexagon::STrib_GP_V4:
     case Hexagon::STb_GP_V4:
     case Hexagon::POST_STbri:
     case Hexagon::STrib_cPt:
@@ -398,17 +387,12 @@ bool HexagonPacketizerList::IsNewifyStore (MachineInstr* MI) {
     case Hexagon::STb_GP_cNotPt_V4:
     case Hexagon::STb_GP_cdnPt_V4:
     case Hexagon::STb_GP_cdnNotPt_V4:
-    case Hexagon::STrib_GP_cPt_V4:
-    case Hexagon::STrib_GP_cNotPt_V4:
-    case Hexagon::STrib_GP_cdnPt_V4:
-    case Hexagon::STrib_GP_cdnNotPt_V4:
 
     // store halfword
     case Hexagon::STrih:
     case Hexagon::STrih_indexed:
     case Hexagon::STrih_indexed_shl_V4:
     case Hexagon::STrih_shl_V4:
-    case Hexagon::STrih_GP_V4:
     case Hexagon::STh_GP_V4:
     case Hexagon::POST_SThri:
     case Hexagon::STrih_cPt:
@@ -431,17 +415,12 @@ bool HexagonPacketizerList::IsNewifyStore (MachineInstr* MI) {
     case Hexagon::STh_GP_cNotPt_V4:
     case Hexagon::STh_GP_cdnPt_V4:
     case Hexagon::STh_GP_cdnNotPt_V4:
-    case Hexagon::STrih_GP_cPt_V4:
-    case Hexagon::STrih_GP_cNotPt_V4:
-    case Hexagon::STrih_GP_cdnPt_V4:
-    case Hexagon::STrih_GP_cdnNotPt_V4:
 
     // store word
     case Hexagon::STriw:
     case Hexagon::STriw_indexed:
     case Hexagon::STriw_indexed_shl_V4:
     case Hexagon::STriw_shl_V4:
-    case Hexagon::STriw_GP_V4:
     case Hexagon::STw_GP_V4:
     case Hexagon::POST_STwri:
     case Hexagon::STriw_cPt:
@@ -464,10 +443,6 @@ bool HexagonPacketizerList::IsNewifyStore (MachineInstr* MI) {
     case Hexagon::STw_GP_cNotPt_V4:
     case Hexagon::STw_GP_cdnPt_V4:
     case Hexagon::STw_GP_cdnNotPt_V4:
-    case Hexagon::STriw_GP_cPt_V4:
-    case Hexagon::STriw_GP_cNotPt_V4:
-    case Hexagon::STriw_GP_cdnPt_V4:
-    case Hexagon::STriw_GP_cdnNotPt_V4:
         return QRI->Subtarget.hasV4TOps();
   }
   return false;
@@ -507,9 +482,6 @@ static int GetDotNewOp(const int opc) {
   case Hexagon::STrib_shl_V4:
     return Hexagon::STrib_shl_nv_V4;
 
-  case Hexagon::STrib_GP_V4:
-    return Hexagon::STrib_GP_nv_V4;
-
   case Hexagon::STb_GP_V4:
     return Hexagon::STb_GP_nv_V4;
 
@@ -576,18 +548,6 @@ static int GetDotNewOp(const int opc) {
   case Hexagon::STb_GP_cdnNotPt_V4:
     return Hexagon::STb_GP_cdnNotPt_nv_V4;
 
-  case Hexagon::STrib_GP_cPt_V4:
-    return Hexagon::STrib_GP_cPt_nv_V4;
-
-  case Hexagon::STrib_GP_cNotPt_V4:
-    return Hexagon::STrib_GP_cNotPt_nv_V4;
-
-  case Hexagon::STrib_GP_cdnPt_V4:
-    return Hexagon::STrib_GP_cdnPt_nv_V4;
-
-  case Hexagon::STrib_GP_cdnNotPt_V4:
-    return Hexagon::STrib_GP_cdnNotPt_nv_V4;
-
   // store new value halfword
   case Hexagon::STrih:
     return Hexagon::STrih_nv_V4;
@@ -601,9 +561,6 @@ static int GetDotNewOp(const int opc) {
   case Hexagon::STrih_shl_V4:
     return Hexagon::STrih_shl_nv_V4;
 
-  case Hexagon::STrih_GP_V4:
-    return Hexagon::STrih_GP_nv_V4;
-
   case Hexagon::STh_GP_V4:
     return Hexagon::STh_GP_nv_V4;
 
@@ -670,18 +627,6 @@ static int GetDotNewOp(const int opc) {
   case Hexagon::STh_GP_cdnNotPt_V4:
     return Hexagon::STh_GP_cdnNotPt_nv_V4;
 
-  case Hexagon::STrih_GP_cPt_V4:
-    return Hexagon::STrih_GP_cPt_nv_V4;
-
-  case Hexagon::STrih_GP_cNotPt_V4:
-    return Hexagon::STrih_GP_cNotPt_nv_V4;
-
-  case Hexagon::STrih_GP_cdnPt_V4:
-    return Hexagon::STrih_GP_cdnPt_nv_V4;
-
-  case Hexagon::STrih_GP_cdnNotPt_V4:
-    return Hexagon::STrih_GP_cdnNotPt_nv_V4;
-
   // store new value word
   case Hexagon::STriw:
     return Hexagon::STriw_nv_V4;
@@ -695,9 +640,6 @@ static int GetDotNewOp(const int opc) {
   case Hexagon::STriw_shl_V4:
     return Hexagon::STriw_shl_nv_V4;
 
-  case Hexagon::STriw_GP_V4:
-    return Hexagon::STriw_GP_nv_V4;
-
   case Hexagon::STw_GP_V4:
     return Hexagon::STw_GP_nv_V4;
 
@@ -764,17 +706,6 @@ static int GetDotNewOp(const int opc) {
   case Hexagon::STw_GP_cdnNotPt_V4:
     return Hexagon::STw_GP_cdnNotPt_nv_V4;
 
-  case Hexagon::STriw_GP_cPt_V4:
-    return Hexagon::STriw_GP_cPt_nv_V4;
-
-  case Hexagon::STriw_GP_cNotPt_V4:
-    return Hexagon::STriw_GP_cNotPt_nv_V4;
-
-  case Hexagon::STriw_GP_cdnPt_V4:
-    return Hexagon::STriw_GP_cdnPt_nv_V4;
-
-  case Hexagon::STriw_GP_cdnNotPt_V4:
-    return Hexagon::STriw_GP_cdnNotPt_nv_V4;
   }
 }
 
@@ -820,12 +751,6 @@ static int GetDotNewPredOp(const int opc) {
   case Hexagon::STb_GP_cNotPt_V4 :
     return Hexagon::STb_GP_cdnNotPt_V4;
 
-  case Hexagon::STrib_GP_cPt_V4 :
-    return Hexagon::STrib_GP_cdnPt_V4;
-
-  case Hexagon::STrib_GP_cNotPt_V4 :
-    return Hexagon::STrib_GP_cdnNotPt_V4;
-
   // Store doubleword conditionally
   case Hexagon::STrid_cPt :
     return Hexagon::STrid_cdnPt_V4;
@@ -857,12 +782,6 @@ static int GetDotNewPredOp(const int opc) {
   case Hexagon::STd_GP_cNotPt_V4 :
     return Hexagon::STd_GP_cdnNotPt_V4;
 
-  case Hexagon::STrid_GP_cPt_V4 :
-    return Hexagon::STrid_GP_cdnPt_V4;
-
-  case Hexagon::STrid_GP_cNotPt_V4 :
-    return Hexagon::STrid_GP_cdnNotPt_V4;
-
   // Store halfword conditionally
   case Hexagon::STrih_cPt :
     return Hexagon::STrih_cdnPt_V4;
@@ -900,12 +819,6 @@ static int GetDotNewPredOp(const int opc) {
   case Hexagon::STh_GP_cNotPt_V4 :
     return Hexagon::STh_GP_cdnNotPt_V4;
 
-  case Hexagon::STrih_GP_cPt_V4 :
-    return Hexagon::STrih_GP_cdnPt_V4;
-
-  case Hexagon::STrih_GP_cNotPt_V4 :
-    return Hexagon::STrih_GP_cdnNotPt_V4;
-
   // Store word conditionally
   case Hexagon::STriw_cPt :
     return Hexagon::STriw_cdnPt_V4;
@@ -943,12 +856,6 @@ static int GetDotNewPredOp(const int opc) {
   case Hexagon::STw_GP_cNotPt_V4 :
     return Hexagon::STw_GP_cdnNotPt_V4;
 
-  case Hexagon::STriw_GP_cPt_V4 :
-    return Hexagon::STriw_GP_cdnPt_V4;
-
-  case Hexagon::STriw_GP_cNotPt_V4 :
-    return Hexagon::STriw_GP_cdnNotPt_V4;
-
   // Condtional Jumps
   case Hexagon::JMP_c:
     return Hexagon::JMP_cdnPt;
@@ -1091,72 +998,36 @@ static int GetDotNewPredOp(const int opc) {
 
   // V4 indexed+scaled load
 
-  case Hexagon::LDrid_indexed_cPt_V4 :
-    return Hexagon::LDrid_indexed_cdnPt_V4;
-
-  case Hexagon::LDrid_indexed_cNotPt_V4 :
-    return Hexagon::LDrid_indexed_cdnNotPt_V4;
-
   case Hexagon::LDrid_indexed_shl_cPt_V4 :
     return Hexagon::LDrid_indexed_shl_cdnPt_V4;
 
   case Hexagon::LDrid_indexed_shl_cNotPt_V4 :
     return Hexagon::LDrid_indexed_shl_cdnNotPt_V4;
 
-  case Hexagon::LDrib_indexed_cPt_V4 :
-    return Hexagon::LDrib_indexed_cdnPt_V4;
-
-  case Hexagon::LDrib_indexed_cNotPt_V4 :
-    return Hexagon::LDrib_indexed_cdnNotPt_V4;
-
   case Hexagon::LDrib_indexed_shl_cPt_V4 :
     return Hexagon::LDrib_indexed_shl_cdnPt_V4;
 
   case Hexagon::LDrib_indexed_shl_cNotPt_V4 :
     return Hexagon::LDrib_indexed_shl_cdnNotPt_V4;
 
-  case Hexagon::LDriub_indexed_cPt_V4 :
-    return Hexagon::LDriub_indexed_cdnPt_V4;
-
-  case Hexagon::LDriub_indexed_cNotPt_V4 :
-    return Hexagon::LDriub_indexed_cdnNotPt_V4;
-
   case Hexagon::LDriub_indexed_shl_cPt_V4 :
     return Hexagon::LDriub_indexed_shl_cdnPt_V4;
 
   case Hexagon::LDriub_indexed_shl_cNotPt_V4 :
     return Hexagon::LDriub_indexed_shl_cdnNotPt_V4;
 
-  case Hexagon::LDrih_indexed_cPt_V4 :
-    return Hexagon::LDrih_indexed_cdnPt_V4;
-
-  case Hexagon::LDrih_indexed_cNotPt_V4 :
-    return Hexagon::LDrih_indexed_cdnNotPt_V4;
-
   case Hexagon::LDrih_indexed_shl_cPt_V4 :
     return Hexagon::LDrih_indexed_shl_cdnPt_V4;
 
   case Hexagon::LDrih_indexed_shl_cNotPt_V4 :
     return Hexagon::LDrih_indexed_shl_cdnNotPt_V4;
 
-  case Hexagon::LDriuh_indexed_cPt_V4 :
-    return Hexagon::LDriuh_indexed_cdnPt_V4;
-
-  case Hexagon::LDriuh_indexed_cNotPt_V4 :
-    return Hexagon::LDriuh_indexed_cdnNotPt_V4;
-
   case Hexagon::LDriuh_indexed_shl_cPt_V4 :
     return Hexagon::LDriuh_indexed_shl_cdnPt_V4;
 
   case Hexagon::LDriuh_indexed_shl_cNotPt_V4 :
     return Hexagon::LDriuh_indexed_shl_cdnNotPt_V4;
 
-  case Hexagon::LDriw_indexed_cPt_V4 :
-    return Hexagon::LDriw_indexed_cdnPt_V4;
-
-  case Hexagon::LDriw_indexed_cNotPt_V4 :
-    return Hexagon::LDriw_indexed_cdnNotPt_V4;
-
   case Hexagon::LDriw_indexed_shl_cPt_V4 :
     return Hexagon::LDriw_indexed_shl_cdnPt_V4;
 
@@ -1201,42 +1072,6 @@ static int GetDotNewPredOp(const int opc) {
   case Hexagon::LDw_GP_cNotPt_V4:
     return Hexagon::LDw_GP_cdnNotPt_V4;
 
-  case Hexagon::LDrid_GP_cPt_V4:
-    return Hexagon::LDrid_GP_cdnPt_V4;
-
-  case Hexagon::LDrid_GP_cNotPt_V4:
-    return Hexagon::LDrid_GP_cdnNotPt_V4;
-
-  case Hexagon::LDrib_GP_cPt_V4:
-    return Hexagon::LDrib_GP_cdnPt_V4;
-
-  case Hexagon::LDrib_GP_cNotPt_V4:
-    return Hexagon::LDrib_GP_cdnNotPt_V4;
-
-  case Hexagon::LDriub_GP_cPt_V4:
-    return Hexagon::LDriub_GP_cdnPt_V4;
-
-  case Hexagon::LDriub_GP_cNotPt_V4:
-    return Hexagon::LDriub_GP_cdnNotPt_V4;
-
-  case Hexagon::LDrih_GP_cPt_V4:
-    return Hexagon::LDrih_GP_cdnPt_V4;
-
-  case Hexagon::LDrih_GP_cNotPt_V4:
-    return Hexagon::LDrih_GP_cdnNotPt_V4;
-
-  case Hexagon::LDriuh_GP_cPt_V4:
-    return Hexagon::LDriuh_GP_cdnPt_V4;
-
-  case Hexagon::LDriuh_GP_cNotPt_V4:
-    return Hexagon::LDriuh_GP_cdnNotPt_V4;
-
-  case Hexagon::LDriw_GP_cPt_V4:
-    return Hexagon::LDriw_GP_cdnPt_V4;
-
-  case Hexagon::LDriw_GP_cNotPt_V4:
-    return Hexagon::LDriw_GP_cdnNotPt_V4;
-
   // Conditional store new-value byte
   case Hexagon::STrib_cPt_nv_V4 :
     return Hexagon::STrib_cdnPt_nv_V4;
@@ -1264,12 +1099,6 @@ static int GetDotNewPredOp(const int opc) {
   case Hexagon::STb_GP_cNotPt_nv_V4 :
     return Hexagon::STb_GP_cdnNotPt_nv_V4;
 
-  case Hexagon::STrib_GP_cPt_nv_V4 :
-    return Hexagon::STrib_GP_cdnPt_nv_V4;
-
-  case Hexagon::STrib_GP_cNotPt_nv_V4 :
-    return Hexagon::STrib_GP_cdnNotPt_nv_V4;
-
   // Conditional store new-value halfword
   case Hexagon::STrih_cPt_nv_V4 :
     return Hexagon::STrih_cdnPt_nv_V4;
@@ -1297,12 +1126,6 @@ static int GetDotNewPredOp(const int opc) {
   case Hexagon::STh_GP_cNotPt_nv_V4 :
     return Hexagon::STh_GP_cdnNotPt_nv_V4;
 
-  case Hexagon::STrih_GP_cPt_nv_V4 :
-    return Hexagon::STrih_GP_cdnPt_nv_V4;
-
-  case Hexagon::STrih_GP_cNotPt_nv_V4 :
-    return Hexagon::STrih_GP_cdnNotPt_nv_V4;
-
   // Conditional store new-value word
   case Hexagon::STriw_cPt_nv_V4 :
     return  Hexagon::STriw_cdnPt_nv_V4;
@@ -1330,12 +1153,6 @@ static int GetDotNewPredOp(const int opc) {
   case Hexagon::STw_GP_cNotPt_nv_V4 :
     return Hexagon::STw_GP_cdnNotPt_nv_V4;
 
-  case Hexagon::STriw_GP_cPt_nv_V4 :
-    return Hexagon::STriw_GP_cdnPt_nv_V4;
-
-  case Hexagon::STriw_GP_cNotPt_nv_V4 :
-    return Hexagon::STriw_GP_cdnNotPt_nv_V4;
-
   // Conditional add
   case Hexagon::ADD_ri_cPt :
     return Hexagon::ADD_ri_cdnPt;
@@ -1622,72 +1439,36 @@ static int GetDotOldOp(const int opc) {
 
   // V4 indexed+scaled Load
 
-  case Hexagon::LDrid_indexed_cdnPt_V4 :
-    return Hexagon::LDrid_indexed_cPt_V4;
-
-  case Hexagon::LDrid_indexed_cdnNotPt_V4 :
-    return Hexagon::LDrid_indexed_cNotPt_V4;
-
   case Hexagon::LDrid_indexed_shl_cdnPt_V4 :
     return Hexagon::LDrid_indexed_shl_cPt_V4;
 
   case Hexagon::LDrid_indexed_shl_cdnNotPt_V4 :
     return Hexagon::LDrid_indexed_shl_cNotPt_V4;
 
-  case Hexagon::LDrib_indexed_cdnPt_V4 :
-    return Hexagon::LDrib_indexed_cPt_V4;
-
-  case Hexagon::LDrib_indexed_cdnNotPt_V4 :
-    return Hexagon::LDrib_indexed_cNotPt_V4;
-
   case Hexagon::LDrib_indexed_shl_cdnPt_V4 :
     return Hexagon::LDrib_indexed_shl_cPt_V4;
 
   case Hexagon::LDrib_indexed_shl_cdnNotPt_V4 :
     return Hexagon::LDrib_indexed_shl_cNotPt_V4;
 
-  case Hexagon::LDriub_indexed_cdnPt_V4 :
-    return Hexagon::LDriub_indexed_cPt_V4;
-
-  case Hexagon::LDriub_indexed_cdnNotPt_V4 :
-    return Hexagon::LDriub_indexed_cNotPt_V4;
-
   case Hexagon::LDriub_indexed_shl_cdnPt_V4 :
     return Hexagon::LDriub_indexed_shl_cPt_V4;
 
   case Hexagon::LDriub_indexed_shl_cdnNotPt_V4 :
     return Hexagon::LDriub_indexed_shl_cNotPt_V4;
 
-  case Hexagon::LDrih_indexed_cdnPt_V4 :
-    return Hexagon::LDrih_indexed_cPt_V4;
-
-  case Hexagon::LDrih_indexed_cdnNotPt_V4 :
-    return Hexagon::LDrih_indexed_cNotPt_V4;
-
   case Hexagon::LDrih_indexed_shl_cdnPt_V4 :
     return Hexagon::LDrih_indexed_shl_cPt_V4;
 
   case Hexagon::LDrih_indexed_shl_cdnNotPt_V4 :
     return Hexagon::LDrih_indexed_shl_cNotPt_V4;
 
-  case Hexagon::LDriuh_indexed_cdnPt_V4 :
-    return Hexagon::LDriuh_indexed_cPt_V4;
-
-  case Hexagon::LDriuh_indexed_cdnNotPt_V4 :
-    return Hexagon::LDriuh_indexed_cNotPt_V4;
-
   case Hexagon::LDriuh_indexed_shl_cdnPt_V4 :
     return Hexagon::LDriuh_indexed_shl_cPt_V4;
 
   case Hexagon::LDriuh_indexed_shl_cdnNotPt_V4 :
     return Hexagon::LDriuh_indexed_shl_cNotPt_V4;
 
-  case Hexagon::LDriw_indexed_cdnPt_V4 :
-    return Hexagon::LDriw_indexed_cPt_V4;
-
-  case Hexagon::LDriw_indexed_cdnNotPt_V4 :
-    return Hexagon::LDriw_indexed_cNotPt_V4;
-
   case Hexagon::LDriw_indexed_shl_cdnPt_V4 :
     return Hexagon::LDriw_indexed_shl_cPt_V4;
 
@@ -1732,42 +1513,6 @@ static int GetDotOldOp(const int opc) {
   case Hexagon::LDw_GP_cdnNotPt_V4:
     return Hexagon::LDw_GP_cNotPt_V4;
 
-  case Hexagon::LDrid_GP_cdnPt_V4:
-    return Hexagon::LDrid_GP_cPt_V4;
-
-  case Hexagon::LDrid_GP_cdnNotPt_V4:
-    return Hexagon::LDrid_GP_cNotPt_V4;
-
-  case Hexagon::LDrib_GP_cdnPt_V4:
-    return Hexagon::LDrib_GP_cPt_V4;
-
-  case Hexagon::LDrib_GP_cdnNotPt_V4:
-    return Hexagon::LDrib_GP_cNotPt_V4;
-
-  case Hexagon::LDriub_GP_cdnPt_V4:
-    return Hexagon::LDriub_GP_cPt_V4;
-
-  case Hexagon::LDriub_GP_cdnNotPt_V4:
-    return Hexagon::LDriub_GP_cNotPt_V4;
-
-  case Hexagon::LDrih_GP_cdnPt_V4:
-    return Hexagon::LDrih_GP_cPt_V4;
-
-  case Hexagon::LDrih_GP_cdnNotPt_V4:
-    return Hexagon::LDrih_GP_cNotPt_V4;
-
-  case Hexagon::LDriuh_GP_cdnPt_V4:
-    return Hexagon::LDriuh_GP_cPt_V4;
-
-  case Hexagon::LDriuh_GP_cdnNotPt_V4:
-    return Hexagon::LDriuh_GP_cNotPt_V4;
-
-  case Hexagon::LDriw_GP_cdnPt_V4:
-    return Hexagon::LDriw_GP_cPt_V4;
-
-  case Hexagon::LDriw_GP_cdnNotPt_V4:
-    return Hexagon::LDriw_GP_cNotPt_V4;
-
   // Conditional add
 
   case Hexagon::ADD_ri_cdnPt :
@@ -1901,16 +1646,6 @@ static int GetDotOldOp(const int opc) {
   case Hexagon::STb_GP_cNotPt_nv_V4:
     return Hexagon::STb_GP_cNotPt_V4;
 
-  case Hexagon::STrib_GP_cdnPt_nv_V4:
-  case Hexagon::STrib_GP_cdnPt_V4:
-  case Hexagon::STrib_GP_cPt_nv_V4:
-    return Hexagon::STrib_GP_cPt_V4;
-
-  case Hexagon::STrib_GP_cdnNotPt_nv_V4:
-  case Hexagon::STrib_GP_cdnNotPt_V4:
-  case Hexagon::STrib_GP_cNotPt_nv_V4:
-    return Hexagon::STrib_GP_cNotPt_V4;
-
   // Store new-value byte - unconditional
   case Hexagon::STrib_nv_V4:
     return Hexagon::STrib;
@@ -1924,9 +1659,6 @@ static int GetDotOldOp(const int opc) {
   case Hexagon::STrib_shl_nv_V4:
     return Hexagon::STrib_shl_V4;
 
-  case Hexagon::STrib_GP_nv_V4:
-    return Hexagon::STrib_GP_V4;
-
   case Hexagon::STb_GP_nv_V4:
     return Hexagon::STb_GP_V4;
 
@@ -1990,16 +1722,6 @@ static int GetDotOldOp(const int opc) {
   case Hexagon::STh_GP_cNotPt_nv_V4:
     return Hexagon::STh_GP_cNotPt_V4;
 
-  case Hexagon::STrih_GP_cdnPt_nv_V4:
-  case Hexagon::STrih_GP_cdnPt_V4:
-  case Hexagon::STrih_GP_cPt_nv_V4:
-    return Hexagon::STrih_GP_cPt_V4;
-
-  case Hexagon::STrih_GP_cdnNotPt_nv_V4:
-  case Hexagon::STrih_GP_cdnNotPt_V4:
-  case Hexagon::STrih_GP_cNotPt_nv_V4:
-    return Hexagon::STrih_GP_cNotPt_V4;
-
   // Store new-value halfword - unconditional
 
   case Hexagon::STrih_nv_V4:
@@ -2014,9 +1736,6 @@ static int GetDotOldOp(const int opc) {
   case Hexagon::STrih_shl_nv_V4:
     return Hexagon::STrih_shl_V4;
 
-  case Hexagon::STrih_GP_nv_V4:
-    return Hexagon::STrih_GP_V4;
-
   case Hexagon::STh_GP_nv_V4:
     return Hexagon::STh_GP_V4;
 
@@ -2081,16 +1800,6 @@ static int GetDotOldOp(const int opc) {
   case Hexagon::STw_GP_cNotPt_nv_V4:
     return Hexagon::STw_GP_cNotPt_V4;
 
-  case Hexagon::STriw_GP_cdnPt_nv_V4:
-  case Hexagon::STriw_GP_cdnPt_V4:
-  case Hexagon::STriw_GP_cPt_nv_V4:
-    return Hexagon::STriw_GP_cPt_V4;
-
-  case Hexagon::STriw_GP_cdnNotPt_nv_V4:
-  case Hexagon::STriw_GP_cdnNotPt_V4:
-  case Hexagon::STriw_GP_cNotPt_nv_V4:
-    return Hexagon::STriw_GP_cNotPt_V4;
-
   // Store new-value word - unconditional
 
   case Hexagon::STriw_nv_V4:
@@ -2105,9 +1814,6 @@ static int GetDotOldOp(const int opc) {
   case Hexagon::STriw_shl_nv_V4:
     return Hexagon::STriw_shl_V4;
 
-  case Hexagon::STriw_GP_nv_V4:
-    return Hexagon::STriw_GP_V4;
-
   case Hexagon::STw_GP_nv_V4:
     return Hexagon::STw_GP_V4;
 
@@ -2146,11 +1852,6 @@ static int GetDotOldOp(const int opc) {
   case Hexagon::STd_GP_cdnNotPt_V4 :
     return Hexagon::STd_GP_cNotPt_V4;
 
-  case Hexagon::STrid_GP_cdnPt_V4 :
-    return Hexagon::STrid_GP_cPt_V4;
-
-  case Hexagon::STrid_GP_cdnNotPt_V4 :
-    return Hexagon::STrid_GP_cNotPt_V4;
   }
 }
 
@@ -2248,28 +1949,16 @@ static bool GetPredicateSense(MachineInstr* MI,
   case Hexagon::LDriub_indexed_cdnPt :
   case Hexagon::POST_LDriub_cPt :
   case Hexagon::POST_LDriub_cdnPt_V4 :
-  case Hexagon::LDrid_indexed_cPt_V4 :
-  case Hexagon::LDrid_indexed_cdnPt_V4 :
   case Hexagon::LDrid_indexed_shl_cPt_V4 :
   case Hexagon::LDrid_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDrib_indexed_cPt_V4 :
-  case Hexagon::LDrib_indexed_cdnPt_V4 :
   case Hexagon::LDrib_indexed_shl_cPt_V4 :
   case Hexagon::LDrib_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDriub_indexed_cPt_V4 :
-  case Hexagon::LDriub_indexed_cdnPt_V4 :
   case Hexagon::LDriub_indexed_shl_cPt_V4 :
   case Hexagon::LDriub_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDrih_indexed_cPt_V4 :
-  case Hexagon::LDrih_indexed_cdnPt_V4 :
   case Hexagon::LDrih_indexed_shl_cPt_V4 :
   case Hexagon::LDrih_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDriuh_indexed_cPt_V4 :
-  case Hexagon::LDriuh_indexed_cdnPt_V4 :
   case Hexagon::LDriuh_indexed_shl_cPt_V4 :
   case Hexagon::LDriuh_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDriw_indexed_cPt_V4 :
-  case Hexagon::LDriw_indexed_cdnPt_V4 :
   case Hexagon::LDriw_indexed_shl_cPt_V4 :
   case Hexagon::LDriw_indexed_shl_cdnPt_V4 :
   case Hexagon::ADD_ri_cPt :
@@ -2298,42 +1987,22 @@ static bool GetPredicateSense(MachineInstr* MI,
   case Hexagon::ZXTB_cdnPt_V4 :
   case Hexagon::ZXTH_cPt_V4 :
   case Hexagon::ZXTH_cdnPt_V4 :
-  case Hexagon::LDrid_GP_cPt_V4 :
-  case Hexagon::LDrib_GP_cPt_V4 :
-  case Hexagon::LDriub_GP_cPt_V4 :
-  case Hexagon::LDrih_GP_cPt_V4 :
-  case Hexagon::LDriuh_GP_cPt_V4 :
-  case Hexagon::LDriw_GP_cPt_V4 :
   case Hexagon::LDd_GP_cPt_V4 :
   case Hexagon::LDb_GP_cPt_V4 :
   case Hexagon::LDub_GP_cPt_V4 :
   case Hexagon::LDh_GP_cPt_V4 :
   case Hexagon::LDuh_GP_cPt_V4 :
   case Hexagon::LDw_GP_cPt_V4 :
-  case Hexagon::STrid_GP_cPt_V4 :
-  case Hexagon::STrib_GP_cPt_V4 :
-  case Hexagon::STrih_GP_cPt_V4 :
-  case Hexagon::STriw_GP_cPt_V4 :
   case Hexagon::STd_GP_cPt_V4 :
   case Hexagon::STb_GP_cPt_V4 :
   case Hexagon::STh_GP_cPt_V4 :
   case Hexagon::STw_GP_cPt_V4 :
-  case Hexagon::LDrid_GP_cdnPt_V4 :
-  case Hexagon::LDrib_GP_cdnPt_V4 :
-  case Hexagon::LDriub_GP_cdnPt_V4 :
-  case Hexagon::LDrih_GP_cdnPt_V4 :
-  case Hexagon::LDriuh_GP_cdnPt_V4 :
-  case Hexagon::LDriw_GP_cdnPt_V4 :
   case Hexagon::LDd_GP_cdnPt_V4 :
   case Hexagon::LDb_GP_cdnPt_V4 :
   case Hexagon::LDub_GP_cdnPt_V4 :
   case Hexagon::LDh_GP_cdnPt_V4 :
   case Hexagon::LDuh_GP_cdnPt_V4 :
   case Hexagon::LDw_GP_cdnPt_V4 :
-  case Hexagon::STrid_GP_cdnPt_V4 :
-  case Hexagon::STrib_GP_cdnPt_V4 :
-  case Hexagon::STrih_GP_cdnPt_V4 :
-  case Hexagon::STriw_GP_cdnPt_V4 :
   case Hexagon::STd_GP_cdnPt_V4 :
   case Hexagon::STb_GP_cdnPt_V4 :
   case Hexagon::STh_GP_cdnPt_V4 :
@@ -2419,28 +2088,16 @@ static bool GetPredicateSense(MachineInstr* MI,
   case Hexagon::LDriub_indexed_cdnNotPt :
   case Hexagon::POST_LDriub_cNotPt :
   case Hexagon::POST_LDriub_cdnNotPt_V4 :
-  case Hexagon::LDrid_indexed_cNotPt_V4 :
-  case Hexagon::LDrid_indexed_cdnNotPt_V4 :
   case Hexagon::LDrid_indexed_shl_cNotPt_V4 :
   case Hexagon::LDrid_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::LDrib_indexed_cNotPt_V4 :
-  case Hexagon::LDrib_indexed_cdnNotPt_V4 :
   case Hexagon::LDrib_indexed_shl_cNotPt_V4 :
   case Hexagon::LDrib_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::LDriub_indexed_cNotPt_V4 :
-  case Hexagon::LDriub_indexed_cdnNotPt_V4 :
   case Hexagon::LDriub_indexed_shl_cNotPt_V4 :
   case Hexagon::LDriub_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::LDrih_indexed_cNotPt_V4 :
-  case Hexagon::LDrih_indexed_cdnNotPt_V4 :
   case Hexagon::LDrih_indexed_shl_cNotPt_V4 :
   case Hexagon::LDrih_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::LDriuh_indexed_cNotPt_V4 :
-  case Hexagon::LDriuh_indexed_cdnNotPt_V4 :
   case Hexagon::LDriuh_indexed_shl_cNotPt_V4 :
   case Hexagon::LDriuh_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::LDriw_indexed_cNotPt_V4 :
-  case Hexagon::LDriw_indexed_cdnNotPt_V4 :
   case Hexagon::LDriw_indexed_shl_cNotPt_V4 :
   case Hexagon::LDriw_indexed_shl_cdnNotPt_V4 :
   case Hexagon::ADD_ri_cNotPt :
@@ -2470,42 +2127,22 @@ static bool GetPredicateSense(MachineInstr* MI,
   case Hexagon::ZXTH_cNotPt_V4 :
   case Hexagon::ZXTH_cdnNotPt_V4 :
 
-  case Hexagon::LDrid_GP_cNotPt_V4 :
-  case Hexagon::LDrib_GP_cNotPt_V4 :
-  case Hexagon::LDriub_GP_cNotPt_V4 :
-  case Hexagon::LDrih_GP_cNotPt_V4 :
-  case Hexagon::LDriuh_GP_cNotPt_V4 :
-  case Hexagon::LDriw_GP_cNotPt_V4 :
   case Hexagon::LDd_GP_cNotPt_V4 :
   case Hexagon::LDb_GP_cNotPt_V4 :
   case Hexagon::LDub_GP_cNotPt_V4 :
   case Hexagon::LDh_GP_cNotPt_V4 :
   case Hexagon::LDuh_GP_cNotPt_V4 :
   case Hexagon::LDw_GP_cNotPt_V4 :
-  case Hexagon::STrid_GP_cNotPt_V4 :
-  case Hexagon::STrib_GP_cNotPt_V4 :
-  case Hexagon::STrih_GP_cNotPt_V4 :
-  case Hexagon::STriw_GP_cNotPt_V4 :
   case Hexagon::STd_GP_cNotPt_V4 :
   case Hexagon::STb_GP_cNotPt_V4 :
   case Hexagon::STh_GP_cNotPt_V4 :
   case Hexagon::STw_GP_cNotPt_V4 :
-  case Hexagon::LDrid_GP_cdnNotPt_V4 :
-  case Hexagon::LDrib_GP_cdnNotPt_V4 :
-  case Hexagon::LDriub_GP_cdnNotPt_V4 :
-  case Hexagon::LDrih_GP_cdnNotPt_V4 :
-  case Hexagon::LDriuh_GP_cdnNotPt_V4 :
-  case Hexagon::LDriw_GP_cdnNotPt_V4 :
   case Hexagon::LDd_GP_cdnNotPt_V4 :
   case Hexagon::LDb_GP_cdnNotPt_V4 :
   case Hexagon::LDub_GP_cdnNotPt_V4 :
   case Hexagon::LDh_GP_cdnNotPt_V4 :
   case Hexagon::LDuh_GP_cdnNotPt_V4 :
   case Hexagon::LDw_GP_cdnNotPt_V4 :
-  case Hexagon::STrid_GP_cdnNotPt_V4 :
-  case Hexagon::STrib_GP_cdnNotPt_V4 :
-  case Hexagon::STrih_GP_cdnNotPt_V4 :
-  case Hexagon::STriw_GP_cdnNotPt_V4 :
   case Hexagon::STd_GP_cdnNotPt_V4 :
   case Hexagon::STb_GP_cdnNotPt_V4 :
   case Hexagon::STh_GP_cdnNotPt_V4 :
@@ -2516,203 +2153,6 @@ static bool GetPredicateSense(MachineInstr* MI,
   return false;
 }
 
-bool HexagonPacketizerList::isDotNewInst(MachineInstr* MI) {
-  if (isNewValueInst(MI))
-    return true;
-
-  switch (MI->getOpcode()) {
-  case Hexagon::TFR_cdnNotPt:
-  case Hexagon::TFR_cdnPt:
-  case Hexagon::TFRI_cdnNotPt:
-  case Hexagon::TFRI_cdnPt:
-  case Hexagon::LDrid_cdnPt :
-  case Hexagon::LDrid_cdnNotPt :
-  case Hexagon::LDrid_indexed_cdnPt :
-  case Hexagon::LDrid_indexed_cdnNotPt :
-  case Hexagon::POST_LDrid_cdnPt_V4 :
-  case Hexagon::POST_LDrid_cdnNotPt_V4 :
-  case Hexagon::LDriw_cdnPt :
-  case Hexagon::LDriw_cdnNotPt :
-  case Hexagon::LDriw_indexed_cdnPt :
-  case Hexagon::LDriw_indexed_cdnNotPt :
-  case Hexagon::POST_LDriw_cdnPt_V4 :
-  case Hexagon::POST_LDriw_cdnNotPt_V4 :
-  case Hexagon::LDrih_cdnPt :
-  case Hexagon::LDrih_cdnNotPt :
-  case Hexagon::LDrih_indexed_cdnPt :
-  case Hexagon::LDrih_indexed_cdnNotPt :
-  case Hexagon::POST_LDrih_cdnPt_V4 :
-  case Hexagon::POST_LDrih_cdnNotPt_V4 :
-  case Hexagon::LDrib_cdnPt :
-  case Hexagon::LDrib_cdnNotPt :
-  case Hexagon::LDrib_indexed_cdnPt :
-  case Hexagon::LDrib_indexed_cdnNotPt :
-  case Hexagon::POST_LDrib_cdnPt_V4 :
-  case Hexagon::POST_LDrib_cdnNotPt_V4 :
-  case Hexagon::LDriuh_cdnPt :
-  case Hexagon::LDriuh_cdnNotPt :
-  case Hexagon::LDriuh_indexed_cdnPt :
-  case Hexagon::LDriuh_indexed_cdnNotPt :
-  case Hexagon::POST_LDriuh_cdnPt_V4 :
-  case Hexagon::POST_LDriuh_cdnNotPt_V4 :
-  case Hexagon::LDriub_cdnPt :
-  case Hexagon::LDriub_cdnNotPt :
-  case Hexagon::LDriub_indexed_cdnPt :
-  case Hexagon::LDriub_indexed_cdnNotPt :
-  case Hexagon::POST_LDriub_cdnPt_V4 :
-  case Hexagon::POST_LDriub_cdnNotPt_V4 :
-
-  case Hexagon::LDrid_indexed_cdnPt_V4 :
-  case Hexagon::LDrid_indexed_cdnNotPt_V4 :
-  case Hexagon::LDrid_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDrid_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::LDrib_indexed_cdnPt_V4 :
-  case Hexagon::LDrib_indexed_cdnNotPt_V4 :
-  case Hexagon::LDrib_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDrib_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::LDriub_indexed_cdnPt_V4 :
-  case Hexagon::LDriub_indexed_cdnNotPt_V4 :
-  case Hexagon::LDriub_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDriub_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::LDrih_indexed_cdnPt_V4 :
-  case Hexagon::LDrih_indexed_cdnNotPt_V4 :
-  case Hexagon::LDrih_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDrih_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::LDriuh_indexed_cdnPt_V4 :
-  case Hexagon::LDriuh_indexed_cdnNotPt_V4 :
-  case Hexagon::LDriuh_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDriuh_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::LDriw_indexed_cdnPt_V4 :
-  case Hexagon::LDriw_indexed_cdnNotPt_V4 :
-  case Hexagon::LDriw_indexed_shl_cdnPt_V4 :
-  case Hexagon::LDriw_indexed_shl_cdnNotPt_V4 :
-
-// Coditional add
-  case Hexagon::ADD_ri_cdnPt:
-  case Hexagon::ADD_ri_cdnNotPt:
-  case Hexagon::ADD_rr_cdnPt:
-  case Hexagon::ADD_rr_cdnNotPt:
-
-  // Conditional logical operations
-  case Hexagon::XOR_rr_cdnPt :
-  case Hexagon::XOR_rr_cdnNotPt :
-  case Hexagon::AND_rr_cdnPt :
-  case Hexagon::AND_rr_cdnNotPt :
-  case Hexagon::OR_rr_cdnPt :
-  case Hexagon::OR_rr_cdnNotPt :
-
-  // Conditonal subtract
-  case Hexagon::SUB_rr_cdnPt :
-  case Hexagon::SUB_rr_cdnNotPt :
-
-  // Conditional combine
-  case Hexagon::COMBINE_rr_cdnPt :
-  case Hexagon::COMBINE_rr_cdnNotPt :
-
-  // Conditional shift operations
-  case Hexagon::ASLH_cdnPt_V4:
-  case Hexagon::ASLH_cdnNotPt_V4:
-  case Hexagon::ASRH_cdnPt_V4:
-  case Hexagon::ASRH_cdnNotPt_V4:
-  case Hexagon::SXTB_cdnPt_V4:
-  case Hexagon::SXTB_cdnNotPt_V4:
-  case Hexagon::SXTH_cdnPt_V4:
-  case Hexagon::SXTH_cdnNotPt_V4:
-  case Hexagon::ZXTB_cdnPt_V4:
-  case Hexagon::ZXTB_cdnNotPt_V4:
-  case Hexagon::ZXTH_cdnPt_V4:
-  case Hexagon::ZXTH_cdnNotPt_V4:
-
-  // Conditional stores
-  case Hexagon::STrib_imm_cdnPt_V4 :
-  case Hexagon::STrib_imm_cdnNotPt_V4 :
-  case Hexagon::STrib_cdnPt_V4 :
-  case Hexagon::STrib_cdnNotPt_V4 :
-  case Hexagon::STrib_indexed_cdnPt_V4 :
-  case Hexagon::STrib_indexed_cdnNotPt_V4 :
-  case Hexagon::POST_STbri_cdnPt_V4 :
-  case Hexagon::POST_STbri_cdnNotPt_V4 :
-  case Hexagon::STrib_indexed_shl_cdnPt_V4 :
-  case Hexagon::STrib_indexed_shl_cdnNotPt_V4 :
-
-  // Store doubleword conditionally
-  case Hexagon::STrid_indexed_cdnPt_V4 :
-  case Hexagon::STrid_indexed_cdnNotPt_V4 :
-  case Hexagon::STrid_indexed_shl_cdnPt_V4 :
-  case Hexagon::STrid_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::POST_STdri_cdnPt_V4 :
-  case Hexagon::POST_STdri_cdnNotPt_V4 :
-
-  // Store halfword conditionally
-  case Hexagon::STrih_cdnPt_V4 :
-  case Hexagon::STrih_cdnNotPt_V4 :
-  case Hexagon::STrih_indexed_cdnPt_V4 :
-  case Hexagon::STrih_indexed_cdnNotPt_V4 :
-  case Hexagon::STrih_imm_cdnPt_V4 :
-  case Hexagon::STrih_imm_cdnNotPt_V4 :
-  case Hexagon::STrih_indexed_shl_cdnPt_V4 :
-  case Hexagon::STrih_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::POST_SThri_cdnPt_V4 :
-  case Hexagon::POST_SThri_cdnNotPt_V4 :
-
-  // Store word conditionally
-  case Hexagon::STriw_cdnPt_V4 :
-  case Hexagon::STriw_cdnNotPt_V4 :
-  case Hexagon::STriw_indexed_cdnPt_V4 :
-  case Hexagon::STriw_indexed_cdnNotPt_V4 :
-  case Hexagon::STriw_imm_cdnPt_V4 :
-  case Hexagon::STriw_imm_cdnNotPt_V4 :
-  case Hexagon::STriw_indexed_shl_cdnPt_V4 :
-  case Hexagon::STriw_indexed_shl_cdnNotPt_V4 :
-  case Hexagon::POST_STwri_cdnPt_V4 :
-  case Hexagon::POST_STwri_cdnNotPt_V4 :
-
-  case Hexagon::LDd_GP_cdnPt_V4:
-  case Hexagon::LDd_GP_cdnNotPt_V4:
-  case Hexagon::LDb_GP_cdnPt_V4:
-  case Hexagon::LDb_GP_cdnNotPt_V4:
-  case Hexagon::LDub_GP_cdnPt_V4:
-  case Hexagon::LDub_GP_cdnNotPt_V4:
-  case Hexagon::LDh_GP_cdnPt_V4:
-  case Hexagon::LDh_GP_cdnNotPt_V4:
-  case Hexagon::LDuh_GP_cdnPt_V4:
-  case Hexagon::LDuh_GP_cdnNotPt_V4:
-  case Hexagon::LDw_GP_cdnPt_V4:
-  case Hexagon::LDw_GP_cdnNotPt_V4:
-  case Hexagon::LDrid_GP_cdnPt_V4:
-  case Hexagon::LDrid_GP_cdnNotPt_V4:
-  case Hexagon::LDrib_GP_cdnPt_V4:
-  case Hexagon::LDrib_GP_cdnNotPt_V4:
-  case Hexagon::LDriub_GP_cdnPt_V4:
-  case Hexagon::LDriub_GP_cdnNotPt_V4:
-  case Hexagon::LDrih_GP_cdnPt_V4:
-  case Hexagon::LDrih_GP_cdnNotPt_V4:
-  case Hexagon::LDriuh_GP_cdnPt_V4:
-  case Hexagon::LDriuh_GP_cdnNotPt_V4:
-  case Hexagon::LDriw_GP_cdnPt_V4:
-  case Hexagon::LDriw_GP_cdnNotPt_V4:
-
-  case Hexagon::STrid_GP_cdnPt_V4:
-  case Hexagon::STrid_GP_cdnNotPt_V4:
-  case Hexagon::STrib_GP_cdnPt_V4:
-  case Hexagon::STrib_GP_cdnNotPt_V4:
-  case Hexagon::STrih_GP_cdnPt_V4:
-  case Hexagon::STrih_GP_cdnNotPt_V4:
-  case Hexagon::STriw_GP_cdnPt_V4:
-  case Hexagon::STriw_GP_cdnNotPt_V4:
-  case Hexagon::STd_GP_cdnPt_V4:
-  case Hexagon::STd_GP_cdnNotPt_V4:
-  case Hexagon::STb_GP_cdnPt_V4:
-  case Hexagon::STb_GP_cdnNotPt_V4:
-  case Hexagon::STh_GP_cdnPt_V4:
-  case Hexagon::STh_GP_cdnNotPt_V4:
-  case Hexagon::STw_GP_cdnPt_V4:
-  case Hexagon::STw_GP_cdnNotPt_V4:
-    return true;
-  }
-  return false;
-}
-
 static MachineOperand& GetPostIncrementOperand(MachineInstr *MI,
                                                const HexagonInstrInfo *QII) {
   assert(QII->isPostIncrement(MI) && "Not a post increment operation.");
@@ -2883,7 +2323,7 @@ bool HexagonPacketizerList::CanPromoteToNewValueStore( MachineInstr *MI,
     // sense, i.e, either both should be negated or both should be none negated.
 
     if (( predRegNumDst != predRegNumSrc) ||
-          isDotNewInst(PacketMI) != isDotNewInst(MI)  ||
+          QII->isDotNewInst(PacketMI) != QII->isDotNewInst(MI)  ||
           GetPredicateSense(MI, QII) != GetPredicateSense(PacketMI, QII)) {
       return false;
     }
@@ -2993,8 +2433,9 @@ bool HexagonPacketizerList::CanPromoteToDotNew( MachineInstr *MI,
                               MachineBasicBlock::iterator &MII,
                               const TargetRegisterClass* RC )
 {
-  // already a dot new instruction
-  if (isDotNewInst(MI) && !IsNewifyStore(MI))
+  const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII;
+  // Already a dot new instruction.
+  if (QII->isDotNewInst(MI) && !IsNewifyStore(MI))
     return false;
 
   if (!isNewifiable(MI))
@@ -3009,7 +2450,6 @@ bool HexagonPacketizerList::CanPromoteToDotNew( MachineInstr *MI,
   else {
     // Create a dot new machine instruction to see if resources can be
     // allocated. If not, bail out now.
-    const HexagonInstrInfo *QII = (const HexagonInstrInfo *) TII;
     int NewOpcode = GetDotNewOp(MI->getOpcode());
     const MCInstrDesc &desc = QII->get(NewOpcode);
     DebugLoc dl;
@@ -3152,7 +2592,7 @@ bool HexagonPacketizerList::ArePredicatesComplements (MachineInstr* MI1,
   // !p0 is not complimentary to p0.new
   return ((MI1->getOperand(1).getReg() == MI2->getOperand(1).getReg()) &&
           (GetPredicateSense(MI1, QII) != GetPredicateSense(MI2, QII)) &&
-          (isDotNewInst(MI1) == isDotNewInst(MI2)));
+          (QII->isDotNewInst(MI1) == QII->isDotNewInst(MI2)));
 }
 
 // initPacketizerState - Initialize packetizer flags
@@ -3277,13 +2717,13 @@ bool HexagonPacketizerList::isLegalToPacketizeTogether(SUnit *SUI, SUnit *SUJ) {
   // dealloc_return and memop always take SLOT0.
   // Arch spec 3.4.4.2
   if (QRI->Subtarget.hasV4TOps()) {
-
-    if (MCIDI.mayStore() && MCIDJ.mayStore() && isNewValueInst(J)) {
+    if (MCIDI.mayStore() && MCIDJ.mayStore() &&
+       (QII->isNewValueInst(J) || QII->isMemOp(J) || QII->isMemOp(I))) {
       Dependence = true;
       return false;
     }
 
-    if (   (QII->isMemOp(J) && MCIDI.mayStore())
+    if ((QII->isMemOp(J) && MCIDI.mayStore())
         || (MCIDJ.mayStore() && QII->isMemOp(I))
         || (QII->isMemOp(J) && QII->isMemOp(I))) {
       Dependence = true;
@@ -3580,7 +3020,7 @@ HexagonPacketizerList::addToPacket(MachineInstr *MI) {
       MachineInstr *nvjMI = MII;
       assert(ResourceTracker->canReserveResources(MI));
       ResourceTracker->reserveResources(MI);
-      if (QII->isExtended(MI) &&
+      if ((QII->isExtended(MI) || QII->isConstExtended(MI)) &&
           !tryAllocateResourcesForConstExt(MI)) {
         endPacket(MBB, MI);
         ResourceTracker->reserveResources(MI);
@@ -3600,7 +3040,7 @@ HexagonPacketizerList::addToPacket(MachineInstr *MI) {
             && (!tryAllocateResourcesForConstExt(nvjMI)
                 || !ResourceTracker->canReserveResources(nvjMI)))
         || // For non-extended instruction, no need to allocate extra 4 bytes.
-        (!QII->isExtended(nvjMI) && 
+        (!QII->isExtended(nvjMI) &&
               !ResourceTracker->canReserveResources(nvjMI)))
       {
         endPacket(MBB, MI);
@@ -3616,7 +3056,7 @@ HexagonPacketizerList::addToPacket(MachineInstr *MI) {
       CurrentPacketMIs.push_back(MI);
       CurrentPacketMIs.push_back(nvjMI);
     } else {
-      if (   QII->isExtended(MI)
+      if (   (QII->isExtended(MI) || QII->isConstExtended(MI))
           && (   !tryAllocateResourcesForConstExt(MI)
               || !ResourceTracker->canReserveResources(MI)))
       {
diff --git a/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.cpp b/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.cpp
index 035afe88d5bc..36da6dfcc3d0 100644
--- a/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.cpp
+++ b/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.cpp
@@ -12,14 +12,14 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "asm-printer"
-#include "Hexagon.h"
 #include "HexagonAsmPrinter.h"
+#include "Hexagon.h"
 #include "HexagonInstPrinter.h"
-#include "HexagonMCInst.h"
+#include "MCTargetDesc/HexagonMCInst.h"
 #include "llvm/MC/MCInst.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCExpr.h"
-#include "llvm/ADT/StringExtras.h"
 #include "llvm/Support/raw_ostream.h"
 #include <cstdio>
 
@@ -28,6 +28,8 @@ using namespace llvm;
 #define GET_INSTRUCTION_NAME
 #include "HexagonGenAsmWriter.inc"
 
+const char HexagonInstPrinter::PacketPadding = '\t';
+
 StringRef HexagonInstPrinter::getOpcodeName(unsigned Opcode) const {
   return MII.getName(Opcode);
 }
@@ -43,43 +45,42 @@ void HexagonInstPrinter::printInst(const MCInst *MI, raw_ostream &O,
 
 void HexagonInstPrinter::printInst(const HexagonMCInst *MI, raw_ostream &O,
                                    StringRef Annot) {
-  const char packetPadding[] = "      ";
   const char startPacket = '{',
              endPacket = '}';
   // TODO: add outer HW loop when it's supported too.
   if (MI->getOpcode() == Hexagon::ENDLOOP0) {
     // Ending a harware loop is different from ending an regular packet.
-    assert(MI->isEndPacket() && "Loop end must also end the packet");
+    assert(MI->isPacketEnd() && "Loop-end must also end the packet");
 
-    if (MI->isStartPacket()) {
+    if (MI->isPacketStart()) {
       // There must be a packet to end a loop.
       // FIXME: when shuffling is always run, this shouldn't be needed.
       HexagonMCInst Nop;
       StringRef NoAnnot;
 
       Nop.setOpcode (Hexagon::NOP);
-      Nop.setStartPacket (MI->isStartPacket());
+      Nop.setPacketStart (MI->isPacketStart());
       printInst (&Nop, O, NoAnnot);
     }
 
     // Close the packet.
-    if (MI->isEndPacket())
-      O << packetPadding << endPacket;
+    if (MI->isPacketEnd())
+      O << PacketPadding << endPacket;
 
     printInstruction(MI, O);
   }
   else {
     // Prefix the insn opening the packet.
-    if (MI->isStartPacket())
-      O << packetPadding << startPacket << '\n';
+    if (MI->isPacketStart())
+      O << PacketPadding << startPacket << '\n';
 
     printInstruction(MI, O);
 
     // Suffix the insn closing the packet.
-    if (MI->isEndPacket())
+    if (MI->isPacketEnd())
       // Suffix the packet in a new line always, since the GNU assembler has
       // issues with a closing brace on the same line as CONST{32,64}.
-      O << '\n' << packetPadding << endPacket;
+      O << '\n' << PacketPadding << endPacket;
   }
 
   printAnnotation(O, Annot);
@@ -102,12 +103,23 @@ void HexagonInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
 
 void HexagonInstPrinter::printImmOperand(const MCInst *MI, unsigned OpNo,
                                          raw_ostream &O) const {
-  O << MI->getOperand(OpNo).getImm();
+  const MCOperand& MO = MI->getOperand(OpNo);
+
+  if(MO.isExpr()) {
+    O << *MO.getExpr();
+  } else if(MO.isImm()) {
+    O << MI->getOperand(OpNo).getImm();
+  } else {
+    llvm_unreachable("Unknown operand");
+  }
 }
 
 void HexagonInstPrinter::printExtOperand(const MCInst *MI, unsigned OpNo,
                                          raw_ostream &O) const {
-  O << MI->getOperand(OpNo).getImm();
+  const HexagonMCInst *HMCI = static_cast<const HexagonMCInst*>(MI);
+  if (HMCI->isConstExtended())
+    O << "#";
+  printOperand(MI, OpNo, O);
 }
 
 void HexagonInstPrinter::printUnsignedImmOperand(const MCInst *MI,
diff --git a/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.h b/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.h
index 902a32352f1c..d0cef683da95 100644
--- a/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.h
+++ b/lib/Target/Hexagon/InstPrinter/HexagonInstPrinter.h
@@ -14,16 +14,18 @@
 #ifndef HEXAGONINSTPRINTER_H
 #define HEXAGONINSTPRINTER_H
 
-#include "HexagonMCInst.h"
 #include "llvm/MC/MCInstPrinter.h"
+#include "llvm/MC/MCInstrInfo.h"
 
 namespace llvm {
+  class HexagonMCInst;
+
   class HexagonInstPrinter : public MCInstPrinter {
   public:
     explicit HexagonInstPrinter(const MCAsmInfo &MAI,
                                 const MCInstrInfo &MII,
                                 const MCRegisterInfo &MRI)
-      : MCInstPrinter(MAI, MII, MRI) {}
+      : MCInstPrinter(MAI, MII, MRI), MII(MII) {}
 
     virtual void printInst(const MCInst *MI, raw_ostream &O, StringRef Annot);
     void printInst(const HexagonMCInst *MI, raw_ostream &O, StringRef Annot);
@@ -65,10 +67,19 @@ namespace llvm {
     void printSymbolLo(const MCInst *MI, unsigned OpNo, raw_ostream &O) const
       { printSymbol(MI, OpNo, O, false); }
 
-    bool isConstExtended(const MCInst *MI) const;
+    const MCInstrInfo &getMII() const {
+      return MII;
+    }
+
   protected:
     void printSymbol(const MCInst *MI, unsigned OpNo, raw_ostream &O, bool hi)
            const;
+
+    static const char PacketPadding;
+
+  private:
+    const MCInstrInfo &MII;
+
   };
 
 } // end namespace llvm
diff --git a/lib/Target/Hexagon/InstPrinter/LLVMBuild.txt b/lib/Target/Hexagon/InstPrinter/LLVMBuild.txt
index 8678401feee4..59849aa7e1c7 100644
--- a/lib/Target/Hexagon/InstPrinter/LLVMBuild.txt
+++ b/lib/Target/Hexagon/InstPrinter/LLVMBuild.txt
@@ -19,5 +19,5 @@
 type = Library
 name = HexagonAsmPrinter
 parent = Hexagon
-required_libraries = MC Support
+required_libraries = HexagonDesc MC Support
 add_to_library_groups = Hexagon
diff --git a/lib/Target/Hexagon/MCTargetDesc/CMakeLists.txt b/lib/Target/Hexagon/MCTargetDesc/CMakeLists.txt
index 8e3da99404ee..62b9b600ce8e 100644
--- a/lib/Target/Hexagon/MCTargetDesc/CMakeLists.txt
+++ b/lib/Target/Hexagon/MCTargetDesc/CMakeLists.txt
@@ -1,6 +1,7 @@
 add_llvm_library(LLVMHexagonDesc
-  HexagonMCTargetDesc.cpp
   HexagonMCAsmInfo.cpp
+  HexagonMCInst.cpp
+  HexagonMCTargetDesc.cpp
   )
 
 add_dependencies(LLVMHexagonDesc HexagonCommonTableGen)
diff --git a/lib/Target/Hexagon/MCTargetDesc/HexagonBaseInfo.h b/lib/Target/Hexagon/MCTargetDesc/HexagonBaseInfo.h
index 7221e906342e..d4a93b5c87a4 100644
--- a/lib/Target/Hexagon/MCTargetDesc/HexagonBaseInfo.h
+++ b/lib/Target/Hexagon/MCTargetDesc/HexagonBaseInfo.h
@@ -17,6 +17,9 @@
 #ifndef HEXAGONBASEINFO_H
 #define HEXAGONBASEINFO_H
 
+#include "HexagonMCTargetDesc.h"
+#include "llvm/Support/ErrorHandling.h"
+
 namespace llvm {
 
 /// HexagonII - This namespace holds all of the target specific flags that
@@ -28,22 +31,50 @@ namespace HexagonII {
   // Insn types.
   // *** Must match HexagonInstrFormat*.td ***
   enum Type {
-    TypePSEUDO = 0,
-    TypeALU32  = 1,
-    TypeCR     = 2,
-    TypeJR     = 3,
-    TypeJ      = 4,
-    TypeLD     = 5,
-    TypeST     = 6,
-    TypeSYSTEM = 7,
-    TypeXTYPE  = 8,
-    TypeMEMOP  = 9,
-    TypeNV     = 10,
-    TypePREFIX = 30, // Such as extenders.
-    TypeMARKER = 31  // Such as end of a HW loop.
+    TypePSEUDO  = 0,
+    TypeALU32   = 1,
+    TypeCR      = 2,
+    TypeJR      = 3,
+    TypeJ       = 4,
+    TypeLD      = 5,
+    TypeST      = 6,
+    TypeSYSTEM  = 7,
+    TypeXTYPE   = 8,
+    TypeMEMOP   = 9,
+    TypeNV      = 10,
+    TypePREFIX  = 30, // Such as extenders.
+    TypeENDLOOP = 31  // Such as end of a HW loop.
   };
 
+  enum SubTarget {
+    HasV2SubT     = 0xf,
+    HasV2SubTOnly = 0x1,
+    NoV2SubT      = 0x0,
+    HasV3SubT     = 0xe,
+    HasV3SubTOnly = 0x2,
+    NoV3SubT      = 0x1,
+    HasV4SubT     = 0xc,
+    NoV4SubT      = 0x3,
+    HasV5SubT     = 0x8,
+    NoV5SubT      = 0x7
+  };
 
+  enum AddrMode {
+    NoAddrMode     = 0,  // No addressing mode
+    Absolute       = 1,  // Absolute addressing mode
+    AbsoluteSet    = 2,  // Absolute set addressing mode
+    BaseImmOffset  = 3,  // Indirect with offset
+    BaseLongOffset = 4,  // Indirect with long offset
+    BaseRegOffset  = 5   // Indirect with register offset
+  };
+
+  enum MemAccessSize {
+    NoMemAccess = 0,            // Not a memory acces instruction.
+    ByteAccess = 1,             // Byte access instruction (memb).
+    HalfWordAccess = 2,         // Half word access instruction (memh).
+    WordAccess = 3,             // Word access instrution (memw).
+    DoubleWordAccess = 4        // Double word access instruction (memd)
+  };
 
   // MCInstrDesc TSFlags
   // *** Must match HexagonInstrFormat*.td ***
@@ -58,11 +89,93 @@ namespace HexagonII {
 
     // Predicated instructions.
     PredicatedPos  = 6,
-    PredicatedMask = 0x1
+    PredicatedMask = 0x1,
+    PredicatedFalsePos  = 7,
+    PredicatedFalseMask = 0x1,
+    PredicatedNewPos  = 8,
+    PredicatedNewMask = 0x1,
+
+    // New-Value consumer instructions.
+    NewValuePos  = 9,
+    NewValueMask = 0x1,
+
+    // New-Value producer instructions.
+    hasNewValuePos  = 10,
+    hasNewValueMask = 0x1,
+
+    // Which operand consumes or produces a new value.
+    NewValueOpPos  = 11,
+    NewValueOpMask = 0x7,
+
+    // Which bits encode the new value.
+    NewValueBitsPos  = 14,
+    NewValueBitsMask = 0x3,
+
+    // Stores that can become new-value stores.
+    mayNVStorePos  = 16,
+    mayNVStoreMask = 0x1,
+
+    // New-value store instructions.
+    NVStorePos  = 17,
+    NVStoreMask = 0x1,
+
+    // Extendable insns.
+    ExtendablePos  = 18,
+    ExtendableMask = 0x1,
+
+    // Insns must be extended.
+    ExtendedPos  = 19,
+    ExtendedMask = 0x1,
+
+    // Which operand may be extended.
+    ExtendableOpPos  = 20,
+    ExtendableOpMask = 0x7,
+
+    // Signed or unsigned range.
+    ExtentSignedPos = 23,
+    ExtentSignedMask = 0x1,
+
+    // Number of bits of range before extending operand.
+    ExtentBitsPos  = 24,
+    ExtentBitsMask = 0x1f,
+
+    // Valid subtargets
+    validSubTargetPos = 29,
+    validSubTargetMask = 0xf,
+
+    // Addressing mode for load/store instructions.
+    AddrModePos = 33,
+    AddrModeMask = 0x7,
+
+    // Access size of memory access instructions (load/store).
+    MemAccessSizePos = 36,
+    MemAccesSizeMask = 0x7
   };
 
   // *** The code above must match HexagonInstrFormat*.td *** //
 
+  // Hexagon specific MO operand flag mask.
+  enum HexagonMOTargetFlagVal {
+    //===------------------------------------------------------------------===//
+    // Hexagon Specific MachineOperand flags.
+    MO_NO_FLAG,
+
+    HMOTF_ConstExtended = 1,
+
+    /// MO_PCREL - On a symbol operand, indicates a PC-relative relocation
+    /// Used for computing a global address for PIC compilations
+    MO_PCREL,
+
+    /// MO_GOT - Indicates a GOT-relative relocation
+    MO_GOT,
+
+    // Low or high part of a symbol.
+    MO_LO16, MO_HI16,
+
+    // Offset from the base of the SDA.
+    MO_GPREL
+  };
+
 } // End namespace HexagonII.
 
 } // End namespace llvm.
diff --git a/lib/Target/Hexagon/MCTargetDesc/HexagonMCAsmInfo.cpp b/lib/Target/Hexagon/MCTargetDesc/HexagonMCAsmInfo.cpp
index 86f75d1c2d7a..3deb8d1deb42 100644
--- a/lib/Target/Hexagon/MCTargetDesc/HexagonMCAsmInfo.cpp
+++ b/lib/Target/Hexagon/MCTargetDesc/HexagonMCAsmInfo.cpp
@@ -31,6 +31,7 @@ HexagonMCAsmInfo::HexagonMCAsmInfo(const Target &T, StringRef TT) {
   AscizDirective = "\t.string\t";
   WeakRefDirective = "\t.weak\t";
 
+  SupportsDebugInformation = true;
   UsesELFSectionDirectiveForBSS  = true;
   ExceptionsType = ExceptionHandling::DwarfCFI;
 }
diff --git a/lib/Target/Hexagon/MCTargetDesc/HexagonMCInst.cpp b/lib/Target/Hexagon/MCTargetDesc/HexagonMCInst.cpp
new file mode 100644
index 000000000000..9260b4a27661
--- /dev/null
+++ b/lib/Target/Hexagon/MCTargetDesc/HexagonMCInst.cpp
@@ -0,0 +1,175 @@
+//===- HexagonMCInst.cpp - Hexagon sub-class of MCInst --------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class extends MCInst to allow some Hexagon VLIW annotations.
+//
+//===----------------------------------------------------------------------===//
+
+#include "HexagonInstrInfo.h"
+#include "MCTargetDesc/HexagonBaseInfo.h"
+#include "MCTargetDesc/HexagonMCInst.h"
+#include "MCTargetDesc/HexagonMCTargetDesc.h"
+
+using namespace llvm;
+
+// Return the slots used by the insn.
+unsigned HexagonMCInst::getUnits(const HexagonTargetMachine* TM) const {
+  const HexagonInstrInfo* QII = TM->getInstrInfo();
+  const InstrItineraryData* II = TM->getInstrItineraryData();
+  const InstrStage*
+    IS = II->beginStage(QII->get(this->getOpcode()).getSchedClass());
+
+  return (IS->getUnits());
+}
+
+// Return the Hexagon ISA class for the insn.
+unsigned HexagonMCInst::getType() const {
+  const uint64_t F = MCID->TSFlags;
+
+  return ((F >> HexagonII::TypePos) & HexagonII::TypeMask);
+}
+
+// Return whether the insn is an actual insn.
+bool HexagonMCInst::isCanon() const {
+  return (!MCID->isPseudo() &&
+          !isPrefix() &&
+          getType() != HexagonII::TypeENDLOOP);
+}
+
+// Return whether the insn is a prefix.
+bool HexagonMCInst::isPrefix() const {
+  return (getType() == HexagonII::TypePREFIX);
+}
+
+// Return whether the insn is solo, i.e., cannot be in a packet.
+bool HexagonMCInst::isSolo() const {
+  const uint64_t F = MCID->TSFlags;
+  return ((F >> HexagonII::SoloPos) & HexagonII::SoloMask);
+}
+
+// Return whether the insn is a new-value consumer.
+bool HexagonMCInst::isNewValue() const {
+  const uint64_t F = MCID->TSFlags;
+  return ((F >> HexagonII::NewValuePos) & HexagonII::NewValueMask);
+}
+
+// Return whether the instruction is a legal new-value producer.
+bool HexagonMCInst::hasNewValue() const {
+  const uint64_t F = MCID->TSFlags;
+  return ((F >> HexagonII::hasNewValuePos) & HexagonII::hasNewValueMask);
+}
+
+// Return the operand that consumes or produces a new value.
+const MCOperand& HexagonMCInst::getNewValue() const {
+  const uint64_t F = MCID->TSFlags;
+  const unsigned O = (F >> HexagonII::NewValueOpPos) &
+                     HexagonII::NewValueOpMask;
+  const MCOperand& MCO = getOperand(O);
+
+  assert ((isNewValue() || hasNewValue()) && MCO.isReg());
+  return (MCO);
+}
+
+// Return whether the instruction needs to be constant extended.
+// 1) Always return true if the instruction has 'isExtended' flag set.
+//
+// isExtendable:
+// 2) For immediate extended operands, return true only if the value is
+//    out-of-range.
+// 3) For global address, always return true.
+
+bool HexagonMCInst::isConstExtended(void) const {
+  if (isExtended())
+    return true;
+
+  if (!isExtendable())
+    return false;
+
+  short ExtOpNum = getCExtOpNum();
+  int MinValue   = getMinValue();
+  int MaxValue   = getMaxValue();
+  const MCOperand& MO = getOperand(ExtOpNum);
+
+  // We could be using an instruction with an extendable immediate and shoehorn
+  // a global address into it. If it is a global address it will be constant
+  // extended. We do this for COMBINE.
+  // We currently only handle isGlobal() because it is the only kind of
+  // object we are going to end up with here for now.
+  // In the future we probably should add isSymbol(), etc.
+  if (MO.isExpr())
+    return true;
+
+  // If the extendable operand is not 'Immediate' type, the instruction should
+  // have 'isExtended' flag set.
+  assert(MO.isImm() && "Extendable operand must be Immediate type");
+
+  int ImmValue = MO.getImm();
+  return (ImmValue < MinValue || ImmValue > MaxValue);
+}
+
+// Return whether the instruction must be always extended.
+bool HexagonMCInst::isExtended(void) const {
+  const uint64_t F = MCID->TSFlags;
+  return (F >> HexagonII::ExtendedPos) & HexagonII::ExtendedMask;
+}
+
+// Return true if the instruction may be extended based on the operand value.
+bool HexagonMCInst::isExtendable(void) const {
+  const uint64_t F = MCID->TSFlags;
+  return (F >> HexagonII::ExtendablePos) & HexagonII::ExtendableMask;
+}
+
+// Return number of bits in the constant extended operand.
+unsigned HexagonMCInst::getBitCount(void) const {
+  const uint64_t F = MCID->TSFlags;
+  return ((F >> HexagonII::ExtentBitsPos) & HexagonII::ExtentBitsMask);
+}
+
+// Return constant extended operand number.
+unsigned short HexagonMCInst::getCExtOpNum(void) const {
+  const uint64_t F = MCID->TSFlags;
+  return ((F >> HexagonII::ExtendableOpPos) & HexagonII::ExtendableOpMask);
+}
+
+// Return whether the operand can be constant extended.
+bool HexagonMCInst::isOperandExtended(const unsigned short OperandNum) const {
+  const uint64_t F = MCID->TSFlags;
+  return ((F >> HexagonII::ExtendableOpPos) & HexagonII::ExtendableOpMask)
+          == OperandNum;
+}
+
+// Return the min value that a constant extendable operand can have
+// without being extended.
+int HexagonMCInst::getMinValue(void) const {
+  const uint64_t F = MCID->TSFlags;
+  unsigned isSigned = (F >> HexagonII::ExtentSignedPos)
+                    & HexagonII::ExtentSignedMask;
+  unsigned bits =  (F >> HexagonII::ExtentBitsPos)
+                    & HexagonII::ExtentBitsMask;
+
+  if (isSigned) // if value is signed
+    return -1 << (bits - 1);
+  else
+    return 0;
+}
+
+// Return the max value that a constant extendable operand can have
+// without being extended.
+int HexagonMCInst::getMaxValue(void) const {
+  const uint64_t F = MCID->TSFlags;
+  unsigned isSigned = (F >> HexagonII::ExtentSignedPos)
+                    & HexagonII::ExtentSignedMask;
+  unsigned bits =  (F >> HexagonII::ExtentBitsPos)
+                    & HexagonII::ExtentBitsMask;
+
+  if (isSigned) // if value is signed
+    return ~(-1 << (bits - 1));
+  else
+    return ~(-1 << bits);
+}
diff --git a/lib/Target/Hexagon/MCTargetDesc/HexagonMCInst.h b/lib/Target/Hexagon/MCTargetDesc/HexagonMCInst.h
new file mode 100644
index 000000000000..3ca71f00b241
--- /dev/null
+++ b/lib/Target/Hexagon/MCTargetDesc/HexagonMCInst.h
@@ -0,0 +1,100 @@
+//===- HexagonMCInst.h - Hexagon sub-class of MCInst ----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class extends MCInst to allow some VLIW annotations.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef HEXAGONMCINST_H
+#define HEXAGONMCINST_H
+
+#include "HexagonTargetMachine.h"
+#include "llvm/MC/MCInst.h"
+
+namespace llvm {
+  class MCOperand;
+
+  class HexagonMCInst: public MCInst {
+    // MCID is set during instruction lowering.
+    // It is needed in order to access TSFlags for
+    // use in checking MC instruction properties.
+    const MCInstrDesc *MCID;
+
+    // Packet start and end markers
+    unsigned packetStart: 1, packetEnd: 1;
+
+  public:
+    explicit HexagonMCInst():
+      MCInst(), MCID(0), packetStart(0), packetEnd(0) {};
+    HexagonMCInst(const MCInstrDesc& mcid):
+      MCInst(), MCID(&mcid), packetStart(0), packetEnd(0) {};
+
+    bool isPacketStart() const { return (packetStart); };
+    bool isPacketEnd() const { return (packetEnd); };
+    void setPacketStart(bool Y) { packetStart = Y; };
+    void setPacketEnd(bool Y) { packetEnd = Y; };
+    void resetPacket() { setPacketStart(false); setPacketEnd(false); };
+
+    // Return the slots used by the insn.
+    unsigned getUnits(const HexagonTargetMachine* TM) const;
+
+    // Return the Hexagon ISA class for the insn.
+    unsigned getType() const;
+
+    void setDesc(const MCInstrDesc& mcid) { MCID = &mcid; };
+    const MCInstrDesc& getDesc(void) const { return *MCID; };
+
+    // Return whether the insn is an actual insn.
+    bool isCanon() const;
+
+    // Return whether the insn is a prefix.
+    bool isPrefix() const;
+
+    // Return whether the insn is solo, i.e., cannot be in a packet.
+    bool isSolo() const;
+
+    // Return whether the instruction needs to be constant extended.
+    bool isConstExtended() const;
+
+    // Return constant extended operand number.
+    unsigned short getCExtOpNum(void) const;
+
+    // Return whether the insn is a new-value consumer.
+    bool isNewValue() const;
+
+    // Return whether the instruction is a legal new-value producer.
+    bool hasNewValue() const;
+
+    // Return the operand that consumes or produces a new value.
+    const MCOperand& getNewValue() const;
+
+    // Return number of bits in the constant extended operand.
+    unsigned getBitCount(void) const;
+
+  private:
+    // Return whether the instruction must be always extended.
+    bool isExtended() const;
+
+    // Return true if the insn may be extended based on the operand value.
+    bool isExtendable() const;
+
+    // Return true if the operand can be constant extended.
+    bool isOperandExtended(const unsigned short OperandNum) const;
+
+    // Return the min value that a constant extendable operand can have
+    // without being extended.
+    int getMinValue() const;
+
+    // Return the max value that a constant extendable operand can have
+    // without being extended.
+    int getMaxValue() const;
+  };
+}
+
+#endif
diff --git a/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp b/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp
index 3cfa4fddd87c..6b1d2d161958 100644
--- a/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp
+++ b/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp
@@ -13,10 +13,12 @@
 
 #include "HexagonMCTargetDesc.h"
 #include "HexagonMCAsmInfo.h"
+#include "InstPrinter/HexagonInstPrinter.h"
 #include "llvm/MC/MachineLocation.h"
 #include "llvm/MC/MCCodeGenInfo.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
diff --git a/lib/Target/Hexagon/TargetInfo/HexagonTargetInfo.cpp b/lib/Target/Hexagon/TargetInfo/HexagonTargetInfo.cpp
index 7aa5dd3b8980..40f6c8d23ea8 100644
--- a/lib/Target/Hexagon/TargetInfo/HexagonTargetInfo.cpp
+++ b/lib/Target/Hexagon/TargetInfo/HexagonTargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "Hexagon.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
diff --git a/lib/Target/LLVMBuild.txt b/lib/Target/LLVMBuild.txt
index 8995080974cc..c06e8bc3cdbe 100644
--- a/lib/Target/LLVMBuild.txt
+++ b/lib/Target/LLVMBuild.txt
@@ -16,7 +16,7 @@
 ;===------------------------------------------------------------------------===;
 
 [common]
-subdirectories = ARM CellSPU CppBackend Hexagon MBlaze MSP430 NVPTX Mips PowerPC Sparc X86 XCore
+subdirectories = AArch64 ARM CppBackend Hexagon MBlaze MSP430 NVPTX Mips PowerPC R600 Sparc X86 XCore
 
 ; This is a special group whose required libraries are extended (by llvm-build)
 ; with the best execution engine (the native JIT, if available, or the
diff --git a/lib/Target/MBlaze/AsmParser/CMakeLists.txt b/lib/Target/MBlaze/AsmParser/CMakeLists.txt
index 813767ba6d65..4a7d8e8d8887 100644
--- a/lib/Target/MBlaze/AsmParser/CMakeLists.txt
+++ b/lib/Target/MBlaze/AsmParser/CMakeLists.txt
@@ -2,7 +2,6 @@ include_directories( ${CMAKE_CURRENT_BINARY_DIR}/..
                      ${CMAKE_CURRENT_SOURCE_DIR}/.. )
 
 add_llvm_library(LLVMMBlazeAsmParser
-  MBlazeAsmLexer.cpp
   MBlazeAsmParser.cpp
   )
 
diff --git a/lib/Target/MBlaze/AsmParser/MBlazeAsmLexer.cpp b/lib/Target/MBlaze/AsmParser/MBlazeAsmLexer.cpp
deleted file mode 100644
index 59a1ed97d3d4..000000000000
--- a/lib/Target/MBlaze/AsmParser/MBlazeAsmLexer.cpp
+++ /dev/null
@@ -1,115 +0,0 @@
-//===-- MBlazeAsmLexer.cpp - Tokenize MBlaze assembly to AsmTokens --------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#include "MCTargetDesc/MBlazeBaseInfo.h"
-
-#include "llvm/MC/MCAsmInfo.h"
-#include "llvm/MC/MCParser/MCAsmLexer.h"
-#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
-#include "llvm/MC/MCRegisterInfo.h"
-#include "llvm/MC/MCTargetAsmLexer.h"
-
-#include "llvm/Support/TargetRegistry.h"
-
-#include <string>
-#include <map>
-
-using namespace llvm;
-
-namespace {
-  
-  class MBlazeBaseAsmLexer : public MCTargetAsmLexer {
-    const MCAsmInfo &AsmInfo;
-    
-    const AsmToken &lexDefinite() {
-      return getLexer()->Lex();
-    }
-    
-    AsmToken LexTokenUAL();
-  protected:
-    typedef std::map <std::string, unsigned> rmap_ty;
-    
-    rmap_ty RegisterMap;
-    
-    void InitRegisterMap(const MCRegisterInfo *info) {
-      unsigned numRegs = info->getNumRegs();
-
-      for (unsigned i = 0; i < numRegs; ++i) {
-        const char *regName = info->getName(i);
-        if (regName)
-          RegisterMap[regName] = i;
-      }
-    }
-    
-    unsigned MatchRegisterName(StringRef Name) {
-      rmap_ty::iterator iter = RegisterMap.find(Name.str());
-      if (iter != RegisterMap.end())
-        return iter->second;
-      else
-        return 0;
-    }
-    
-    AsmToken LexToken() {
-      if (!Lexer) {
-        SetError(SMLoc(), "No MCAsmLexer installed");
-        return AsmToken(AsmToken::Error, "", 0);
-      }
-      
-      switch (AsmInfo.getAssemblerDialect()) {
-      default:
-        SetError(SMLoc(), "Unhandled dialect");
-        return AsmToken(AsmToken::Error, "", 0);
-      case 0:
-        return LexTokenUAL();
-      }
-    }
-  public:
-    MBlazeBaseAsmLexer(const Target &T, const MCAsmInfo &MAI)
-      : MCTargetAsmLexer(T), AsmInfo(MAI) {
-    }
-  };
-  
-  class MBlazeAsmLexer : public MBlazeBaseAsmLexer {
-  public:
-    MBlazeAsmLexer(const Target &T, const MCRegisterInfo &MRI,
-                   const MCAsmInfo &MAI)
-      : MBlazeBaseAsmLexer(T, MAI) {
-      InitRegisterMap(&MRI);
-    }
-  };
-}
-
-AsmToken MBlazeBaseAsmLexer::LexTokenUAL() {
-  const AsmToken &lexedToken = lexDefinite();
-  
-  switch (lexedToken.getKind()) {
-  default:
-    return AsmToken(lexedToken);
-  case AsmToken::Error:
-    SetError(Lexer->getErrLoc(), Lexer->getErr());
-    return AsmToken(lexedToken);
-  case AsmToken::Identifier:
-  {
-    unsigned regID = MatchRegisterName(lexedToken.getString().lower());
-    
-    if (regID) {
-      return AsmToken(AsmToken::Register,
-                      lexedToken.getString(),
-                      static_cast<int64_t>(regID));
-    } else {
-      return AsmToken(lexedToken);
-    }
-  }
-  }
-}
-
-extern "C" void LLVMInitializeMBlazeAsmLexer() {
-  RegisterMCAsmLexer<MBlazeAsmLexer> X(TheMBlazeTarget);
-}
-
diff --git a/lib/Target/MBlaze/AsmParser/MBlazeAsmParser.cpp b/lib/Target/MBlaze/AsmParser/MBlazeAsmParser.cpp
index f7809caeb32f..dda6e247ac4f 100644
--- a/lib/Target/MBlaze/AsmParser/MBlazeAsmParser.cpp
+++ b/lib/Target/MBlaze/AsmParser/MBlazeAsmParser.cpp
@@ -8,18 +8,18 @@
 //===----------------------------------------------------------------------===//
 
 #include "MCTargetDesc/MBlazeBaseInfo.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Twine.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCParser/MCAsmLexer.h"
 #include "llvm/MC/MCParser/MCAsmParser.h"
 #include "llvm/MC/MCParser/MCParsedAsmOperand.h"
 #include "llvm/MC/MCStreamer.h"
-#include "llvm/MC/MCExpr.h"
-#include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCTargetAsmParser.h"
 #include "llvm/Support/SourceMgr.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/Twine.h"
 using namespace llvm;
 
 namespace {
@@ -35,7 +35,8 @@ class MBlazeAsmParser : public MCTargetAsmParser {
   bool Error(SMLoc L, const Twine &Msg) { return Parser.Error(L, Msg); }
 
   MBlazeOperand *ParseMemory(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
-  MBlazeOperand *ParseRegister(unsigned &RegNo);
+  MBlazeOperand *ParseRegister();
+  MBlazeOperand *ParseRegister(SMLoc &StartLoc, SMLoc &EndLoc);
   MBlazeOperand *ParseImmediate();
   MBlazeOperand *ParseFsl();
   MBlazeOperand* ParseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
@@ -81,29 +82,35 @@ struct MBlazeOperand : public MCParsedAsmOperand {
 
   SMLoc StartLoc, EndLoc;
 
+  struct TokOp {
+    const char *Data;
+    unsigned Length;
+  };
+
+  struct RegOp {
+    unsigned RegNum;
+  };
+
+  struct ImmOp {
+    const MCExpr *Val;
+  };
+
+  struct MemOp {
+    unsigned Base;
+    unsigned OffReg;
+    const MCExpr *Off;
+  };
+
+  struct FslImmOp {
+    const MCExpr *Val;
+  };
+
   union {
-    struct {
-      const char *Data;
-      unsigned Length;
-    } Tok;
-
-    struct {
-      unsigned RegNum;
-    } Reg;
-
-    struct {
-      const MCExpr *Val;
-    } Imm;
-
-    struct {
-      unsigned Base;
-      unsigned OffReg;
-      const MCExpr *Off;
-    } Mem;
-
-    struct {
-      const MCExpr *Val;
-    } FslImm;
+    struct TokOp Tok;
+    struct RegOp Reg;
+    struct ImmOp Imm;
+    struct MemOp Mem;
+    struct FslImmOp FslImm;
   };
 
   MBlazeOperand(KindTy K) : MCParsedAsmOperand(), Kind(K) {}
@@ -383,23 +390,31 @@ ParseMemory(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
 
 bool MBlazeAsmParser::ParseRegister(unsigned &RegNo,
                                     SMLoc &StartLoc, SMLoc &EndLoc) {
-  return (ParseRegister(RegNo) == 0);
+  MBlazeOperand *Reg = ParseRegister(StartLoc, EndLoc);
+  if (!Reg)
+    return true;
+  RegNo = Reg->getReg();
+  return false;
 }
 
-MBlazeOperand *MBlazeAsmParser::ParseRegister(unsigned &RegNo) {
-  SMLoc S = Parser.getTok().getLoc();
-  SMLoc E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1);
+MBlazeOperand *MBlazeAsmParser::ParseRegister() {
+  SMLoc S, E;
+  return ParseRegister(S, E);
+}
 
-  switch (getLexer().getKind()) {
-  default: return 0;
-  case AsmToken::Identifier:
-    RegNo = MatchRegisterName(getLexer().getTok().getIdentifier());
-    if (RegNo == 0)
-      return 0;
+MBlazeOperand *MBlazeAsmParser::ParseRegister(SMLoc &StartLoc, SMLoc &EndLoc) {
+  StartLoc = Parser.getTok().getLoc();
+  EndLoc = Parser.getTok().getEndLoc();
 
-    getLexer().Lex();
-    return MBlazeOperand::CreateReg(RegNo, S, E);
-  }
+  if (getLexer().getKind() != AsmToken::Identifier)
+    return 0;
+
+  unsigned RegNo = MatchRegisterName(getLexer().getTok().getIdentifier());
+  if (RegNo == 0)
+    return 0;
+
+  getLexer().Lex();
+  return MBlazeOperand::CreateReg(RegNo, StartLoc, EndLoc);
 }
 
 static unsigned MatchFslRegister(StringRef String) {
@@ -415,7 +430,7 @@ static unsigned MatchFslRegister(StringRef String) {
 
 MBlazeOperand *MBlazeAsmParser::ParseFsl() {
   SMLoc S = Parser.getTok().getLoc();
-  SMLoc E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1);
+  SMLoc E = Parser.getTok().getEndLoc();
 
   switch (getLexer().getKind()) {
   default: return 0;
@@ -432,7 +447,7 @@ MBlazeOperand *MBlazeAsmParser::ParseFsl() {
 
 MBlazeOperand *MBlazeAsmParser::ParseImmediate() {
   SMLoc S = Parser.getTok().getLoc();
-  SMLoc E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1);
+  SMLoc E = Parser.getTok().getEndLoc();
 
   const MCExpr *EVal;
   switch (getLexer().getKind()) {
@@ -442,7 +457,7 @@ MBlazeOperand *MBlazeAsmParser::ParseImmediate() {
   case AsmToken::Minus:
   case AsmToken::Integer:
   case AsmToken::Identifier:
-    if (getParser().ParseExpression(EVal))
+    if (getParser().parseExpression(EVal))
       return 0;
 
     return MBlazeOperand::CreateImm(EVal, S, E);
@@ -454,8 +469,7 @@ ParseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
   MBlazeOperand *Op;
 
   // Attempt to parse the next token as a register name
-  unsigned RegNo;
-  Op = ParseRegister(RegNo);
+  Op = ParseRegister();
 
   // Attempt to parse the next token as an FSL immediate
   if (!Op)
@@ -529,10 +543,10 @@ bool MBlazeAsmParser::ParseDirectiveWord(unsigned Size, SMLoc L) {
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     for (;;) {
       const MCExpr *Value;
-      if (getParser().ParseExpression(Value))
+      if (getParser().parseExpression(Value))
         return true;
 
-      getParser().getStreamer().EmitValue(Value, Size, 0 /*addrspace*/);
+      getParser().getStreamer().EmitValue(Value, Size);
 
       if (getLexer().is(AsmToken::EndOfStatement))
         break;
@@ -548,12 +562,9 @@ bool MBlazeAsmParser::ParseDirectiveWord(unsigned Size, SMLoc L) {
   return false;
 }
 
-extern "C" void LLVMInitializeMBlazeAsmLexer();
-
 /// Force static initialization.
 extern "C" void LLVMInitializeMBlazeAsmParser() {
   RegisterMCAsmParser<MBlazeAsmParser> X(TheMBlazeTarget);
-  LLVMInitializeMBlazeAsmLexer();
 }
 
 #define GET_REGISTER_MATCHER
diff --git a/lib/Target/MBlaze/CMakeLists.txt b/lib/Target/MBlaze/CMakeLists.txt
index 0bf93d71dab8..91a41f39b5d8 100644
--- a/lib/Target/MBlaze/CMakeLists.txt
+++ b/lib/Target/MBlaze/CMakeLists.txt
@@ -9,7 +9,6 @@ tablegen(LLVM MBlazeGenDAGISel.inc -gen-dag-isel)
 tablegen(LLVM MBlazeGenCallingConv.inc -gen-callingconv)
 tablegen(LLVM MBlazeGenSubtargetInfo.inc -gen-subtarget)
 tablegen(LLVM MBlazeGenIntrinsics.inc -gen-tgt-intrinsic)
-tablegen(LLVM MBlazeGenEDInfo.inc -gen-enhanced-disassembly-info)
 add_public_tablegen_target(MBlazeCommonTableGen)
 
 add_llvm_target(MBlazeCodeGen
diff --git a/lib/Target/MBlaze/Disassembler/MBlazeDisassembler.cpp b/lib/Target/MBlaze/Disassembler/MBlazeDisassembler.cpp
index 6b958c85eebf..c03ab3803b60 100644
--- a/lib/Target/MBlaze/Disassembler/MBlazeDisassembler.cpp
+++ b/lib/Target/MBlaze/Disassembler/MBlazeDisassembler.cpp
@@ -12,10 +12,8 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "MBlaze.h"
 #include "MBlazeDisassembler.h"
-
-#include "llvm/MC/EDInstInfo.h"
+#include "MBlaze.h"
 #include "llvm/MC/MCDisassembler.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrDesc.h"
@@ -26,7 +24,6 @@
 
 // #include "MBlazeGenDecoderTables.inc"
 // #include "MBlazeGenRegisterNames.inc"
-#include "MBlazeGenEDInfo.inc"
 
 namespace llvm {
 extern const MCInstrDesc MBlazeInsts[];
@@ -492,10 +489,6 @@ static unsigned getOPCODE(uint32_t insn) {
   }
 }
 
-const EDInstInfo *MBlazeDisassembler::getEDInfo() const {
-  return instInfoMBlaze;
-}
-
 //
 // Public interface for the disassembler
 //
diff --git a/lib/Target/MBlaze/Disassembler/MBlazeDisassembler.h b/lib/Target/MBlaze/Disassembler/MBlazeDisassembler.h
index 5c4ae3b1ace8..b8ff8f607265 100644
--- a/lib/Target/MBlaze/Disassembler/MBlazeDisassembler.h
+++ b/lib/Target/MBlaze/Disassembler/MBlazeDisassembler.h
@@ -23,8 +23,6 @@ class MCInst;
 class MemoryObject;
 class raw_ostream;
 
-struct EDInstInfo;
-  
 /// MBlazeDisassembler - Disassembler for all MBlaze platforms.
 class MBlazeDisassembler : public MCDisassembler {
 public:
@@ -44,9 +42,6 @@ public:
                       uint64_t address,
                       raw_ostream &vStream,
                       raw_ostream &cStream) const;
-
-  /// getEDInfo - See MCDisassembler.
-  const EDInstInfo *getEDInfo() const;
 };
 
 } // namespace llvm
diff --git a/lib/Target/MBlaze/InstPrinter/MBlazeInstPrinter.cpp b/lib/Target/MBlaze/InstPrinter/MBlazeInstPrinter.cpp
index a1f1dbc7a23b..fc2b3d51b44c 100644
--- a/lib/Target/MBlaze/InstPrinter/MBlazeInstPrinter.cpp
+++ b/lib/Target/MBlaze/InstPrinter/MBlazeInstPrinter.cpp
@@ -12,11 +12,11 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "asm-printer"
-#include "MBlaze.h"
 #include "MBlazeInstPrinter.h"
-#include "llvm/MC/MCInst.h"
+#include "MBlaze.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormattedStream.h"
 using namespace llvm;
diff --git a/lib/Target/MBlaze/MBlazeAsmPrinter.cpp b/lib/Target/MBlaze/MBlazeAsmPrinter.cpp
index b679a318c3e0..7dafaef0af08 100644
--- a/lib/Target/MBlaze/MBlazeAsmPrinter.cpp
+++ b/lib/Target/MBlaze/MBlazeAsmPrinter.cpp
@@ -15,32 +15,32 @@
 #define DEBUG_TYPE "mblaze-asm-printer"
 
 #include "MBlaze.h"
-#include "MBlazeSubtarget.h"
+#include "InstPrinter/MBlazeInstPrinter.h"
 #include "MBlazeInstrInfo.h"
-#include "MBlazeTargetMachine.h"
-#include "MBlazeMachineFunction.h"
 #include "MBlazeMCInstLower.h"
-#include "InstPrinter/MBlazeInstPrinter.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Module.h"
+#include "MBlazeMachineFunction.h"
+#include "MBlazeSubtarget.h"
+#include "MBlazeTargetMachine.h"
 #include "llvm/CodeGen/AsmPrinter.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Module.h"
+#include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCStreamer.h"
-#include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCSymbol.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/Mangler.h"
-#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLoweringObjectFile.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetOptions.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/TargetRegistry.h"
-#include "llvm/Support/raw_ostream.h"
 #include <cctype>
 
 using namespace llvm;
diff --git a/lib/Target/MBlaze/MBlazeDelaySlotFiller.cpp b/lib/Target/MBlaze/MBlazeDelaySlotFiller.cpp
index 19e787d8622d..3d0d1cecd1f1 100644
--- a/lib/Target/MBlaze/MBlazeDelaySlotFiller.cpp
+++ b/lib/Target/MBlaze/MBlazeDelaySlotFiller.cpp
@@ -16,14 +16,14 @@
 
 #include "MBlaze.h"
 #include "MBlazeTargetMachine.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/ADT/Statistic.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetInstrInfo.h"
 
 using namespace llvm;
 
diff --git a/lib/Target/MBlaze/MBlazeFrameLowering.cpp b/lib/Target/MBlaze/MBlazeFrameLowering.cpp
index 9e467bf337e0..172304bd5b45 100644
--- a/lib/Target/MBlaze/MBlazeFrameLowering.cpp
+++ b/lib/Target/MBlaze/MBlazeFrameLowering.cpp
@@ -14,21 +14,21 @@
 #define DEBUG_TYPE "mblaze-frame-lowering"
 
 #include "MBlazeFrameLowering.h"
+#include "InstPrinter/MBlazeInstPrinter.h"
 #include "MBlazeInstrInfo.h"
 #include "MBlazeMachineFunction.h"
-#include "InstPrinter/MBlazeInstPrinter.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
@@ -426,6 +426,45 @@ void MBlazeFrameLowering::emitEpilogue(MachineFunction &MF,
   }
 }
 
+// Eliminate ADJCALLSTACKDOWN/ADJCALLSTACKUP pseudo instructions
+void MBlazeFrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  const MBlazeInstrInfo &TII =
+    *static_cast<const MBlazeInstrInfo*>(MF.getTarget().getInstrInfo());
+  if (!hasReservedCallFrame(MF)) {
+    // If we have a frame pointer, turn the adjcallstackup instruction into a
+    // 'addi r1, r1, -<amt>' and the adjcallstackdown instruction into
+    // 'addi r1, r1, <amt>'
+    MachineInstr *Old = I;
+    int Amount = Old->getOperand(0).getImm() + 4;
+    if (Amount != 0) {
+      // We need to keep the stack aligned properly.  To do this, we round the
+      // amount of space needed for the outgoing arguments up to the next
+      // alignment boundary.
+      unsigned Align = getStackAlignment();
+      Amount = (Amount+Align-1)/Align*Align;
+
+      MachineInstr *New;
+      if (Old->getOpcode() == MBlaze::ADJCALLSTACKDOWN) {
+        New = BuildMI(MF,Old->getDebugLoc(), TII.get(MBlaze::ADDIK),MBlaze::R1)
+                .addReg(MBlaze::R1).addImm(-Amount);
+      } else {
+        assert(Old->getOpcode() == MBlaze::ADJCALLSTACKUP);
+        New = BuildMI(MF,Old->getDebugLoc(), TII.get(MBlaze::ADDIK),MBlaze::R1)
+                .addReg(MBlaze::R1).addImm(Amount);
+      }
+
+      // Replace the pseudo instruction with a new instruction...
+      MBB.insert(I, New);
+    }
+  }
+
+  // Simply discard ADJCALLSTACKDOWN, ADJCALLSTACKUP instructions.
+  MBB.erase(I);
+}
+
+
 void MBlazeFrameLowering::
 processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
                                      RegScavenger *RS) const {
diff --git a/lib/Target/MBlaze/MBlazeFrameLowering.h b/lib/Target/MBlaze/MBlazeFrameLowering.h
index 01e6578a352f..f4228c5f0890 100644
--- a/lib/Target/MBlaze/MBlazeFrameLowering.h
+++ b/lib/Target/MBlaze/MBlazeFrameLowering.h
@@ -39,6 +39,10 @@ public:
   void emitPrologue(MachineFunction &MF) const;
   void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
 
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I) const;
+
   bool hasFP(const MachineFunction &MF) const;
 
   int getFrameIndexOffset(const MachineFunction &MF, int FI) const;
diff --git a/lib/Target/MBlaze/MBlazeISelDAGToDAG.cpp b/lib/Target/MBlaze/MBlazeISelDAGToDAG.cpp
index 6b4349766f37..78ad24debb1b 100644
--- a/lib/Target/MBlaze/MBlazeISelDAGToDAG.cpp
+++ b/lib/Target/MBlaze/MBlazeISelDAGToDAG.cpp
@@ -17,21 +17,21 @@
 #include "MBlazeRegisterInfo.h"
 #include "MBlazeSubtarget.h"
 #include "MBlazeTargetMachine.h"
-#include "llvm/GlobalValue.h"
-#include "llvm/Instructions.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/Support/CFG.h"
-#include "llvm/Type.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
-#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
-#include "llvm/Target/TargetMachine.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Type.h"
+#include "llvm/Support/CFG.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetMachine.h"
 using namespace llvm;
 
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/MBlaze/MBlazeISelLowering.cpp b/lib/Target/MBlaze/MBlazeISelLowering.cpp
index 310c25e839c3..d4f943297acb 100644
--- a/lib/Target/MBlaze/MBlazeISelLowering.cpp
+++ b/lib/Target/MBlaze/MBlazeISelLowering.cpp
@@ -15,14 +15,9 @@
 #define DEBUG_TYPE "mblaze-lower"
 #include "MBlazeISelLowering.h"
 #include "MBlazeMachineFunction.h"
+#include "MBlazeSubtarget.h"
 #include "MBlazeTargetMachine.h"
 #include "MBlazeTargetObjectFile.h"
-#include "MBlazeSubtarget.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/CallingConv.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -30,6 +25,11 @@
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
@@ -81,6 +81,7 @@ MBlazeTargetLowering::MBlazeTargetLowering(MBlazeTargetMachine &TM)
   setOperationAction(ISD::FCOPYSIGN,  MVT::f64, Expand);
   setOperationAction(ISD::FSIN,       MVT::f32, Expand);
   setOperationAction(ISD::FCOS,       MVT::f32, Expand);
+  setOperationAction(ISD::FSINCOS,    MVT::f32, Expand);
   setOperationAction(ISD::FPOWI,      MVT::f32, Expand);
   setOperationAction(ISD::FPOW,       MVT::f32, Expand);
   setOperationAction(ISD::FLOG,       MVT::f32, Expand);
@@ -159,7 +160,8 @@ MBlazeTargetLowering::MBlazeTargetLowering(MBlazeTargetMachine &TM)
   // Operations not directly supported by MBlaze.
   setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32,   Expand);
   setOperationAction(ISD::BR_JT,              MVT::Other, Expand);
-  setOperationAction(ISD::BR_CC,              MVT::Other, Expand);
+  setOperationAction(ISD::BR_CC,              MVT::f32,   Expand);
+  setOperationAction(ISD::BR_CC,              MVT::i32,   Expand);
   setOperationAction(ISD::SIGN_EXTEND_INREG,  MVT::i1,    Expand);
   setOperationAction(ISD::ROTL,               MVT::i32,   Expand);
   setOperationAction(ISD::ROTR,               MVT::i32,   Expand);
@@ -1027,15 +1029,17 @@ LowerReturn(SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
   // Analize return values.
   CCInfo.AnalyzeReturn(Outs, RetCC_MBlaze);
 
-  // If this is the first return lowered for this function, add
-  // the regs to the liveout set for the function.
-  if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
-    for (unsigned i = 0; i != RVLocs.size(); ++i)
-      if (RVLocs[i].isRegLoc())
-        DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
-  }
-
   SDValue Flag;
+  SmallVector<SDValue, 4> RetOps(1, Chain);
+
+  // If this function is using the interrupt_handler calling convention
+  // then use "rtid r14, 0" otherwise use "rtsd r15, 8"
+  unsigned Ret = (CallConv == CallingConv::MBLAZE_INTR) ? MBlazeISD::IRet
+                                                        : MBlazeISD::Ret;
+  unsigned Reg = (CallConv == CallingConv::MBLAZE_INTR) ? MBlaze::R14
+                                                        : MBlaze::R15;
+  RetOps.push_back(DAG.getRegister(Reg, MVT::i32));
+
 
   // Copy the result values into the output registers.
   for (unsigned i = 0; i != RVLocs.size(); ++i) {
@@ -1048,20 +1052,16 @@ LowerReturn(SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
     // guarantee that all emitted copies are
     // stuck together, avoiding something bad
     Flag = Chain.getValue(1);
+    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
   }
 
-  // If this function is using the interrupt_handler calling convention
-  // then use "rtid r14, 0" otherwise use "rtsd r15, 8"
-  unsigned Ret = (CallConv == CallingConv::MBLAZE_INTR) ? MBlazeISD::IRet
-                                                        : MBlazeISD::Ret;
-  unsigned Reg = (CallConv == CallingConv::MBLAZE_INTR) ? MBlaze::R14
-                                                        : MBlaze::R15;
-  SDValue DReg = DAG.getRegister(Reg, MVT::i32);
+  RetOps[0] = Chain;  // Update chain.
 
+  // Add the flag if we have it.
   if (Flag.getNode())
-    return DAG.getNode(Ret, dl, MVT::Other, Chain, DReg, Flag);
+    RetOps.push_back(Flag);
 
-  return DAG.getNode(Ret, dl, MVT::Other, Chain, DReg);
+  return DAG.getNode(Ret, dl, MVT::Other, &RetOps[0], RetOps.size());
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/MBlaze/MBlazeISelLowering.h b/lib/Target/MBlaze/MBlazeISelLowering.h
index a01fab567c8a..f6b4095a93dc 100644
--- a/lib/Target/MBlaze/MBlazeISelLowering.h
+++ b/lib/Target/MBlaze/MBlazeISelLowering.h
@@ -17,8 +17,8 @@
 
 #include "MBlaze.h"
 #include "MBlazeSubtarget.h"
-#include "llvm/Support/ErrorHandling.h"
 #include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Target/TargetLowering.h"
 
 namespace llvm {
diff --git a/lib/Target/MBlaze/MBlazeInstrInfo.cpp b/lib/Target/MBlaze/MBlazeInstrInfo.cpp
index b5025fc8ee6c..79449f73f74e 100644
--- a/lib/Target/MBlaze/MBlazeInstrInfo.cpp
+++ b/lib/Target/MBlaze/MBlazeInstrInfo.cpp
@@ -12,15 +12,15 @@
 //===----------------------------------------------------------------------===//
 
 #include "MBlazeInstrInfo.h"
-#include "MBlazeTargetMachine.h"
 #include "MBlazeMachineFunction.h"
+#include "MBlazeTargetMachine.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/ScoreboardHazardRecognizer.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
-#include "llvm/ADT/STLExtras.h"
 
 #define GET_INSTRINFO_CTOR
 #include "MBlazeGenInstrInfo.inc"
diff --git a/lib/Target/MBlaze/MBlazeInstrInfo.td b/lib/Target/MBlaze/MBlazeInstrInfo.td
index 139bf7156a69..f86bc0b0b5a4 100644
--- a/lib/Target/MBlaze/MBlazeInstrInfo.td
+++ b/lib/Target/MBlaze/MBlazeInstrInfo.td
@@ -28,9 +28,9 @@ def SDT_MBCallSeqEnd   : SDCallSeqEnd<[SDTCisVT<0, i32>, SDTCisVT<1, i32>]>;
 //===----------------------------------------------------------------------===//
 
 def MBlazeRet     : SDNode<"MBlazeISD::Ret", SDT_MBlazeRet,
-                           [SDNPHasChain, SDNPOptInGlue]>;
+                           [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
 def MBlazeIRet    : SDNode<"MBlazeISD::IRet", SDT_MBlazeIRet,
-                           [SDNPHasChain, SDNPOptInGlue]>;
+                           [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
 
 def MBlazeJmpLink : SDNode<"MBlazeISD::JmpLink",SDT_MBlazeJmpLink,
                            [SDNPHasChain,SDNPOptInGlue,SDNPOutGlue,
diff --git a/lib/Target/MBlaze/MBlazeIntrinsicInfo.cpp b/lib/Target/MBlaze/MBlazeIntrinsicInfo.cpp
index 1c2e3b26613e..8d262a01e706 100644
--- a/lib/Target/MBlaze/MBlazeIntrinsicInfo.cpp
+++ b/lib/Target/MBlaze/MBlazeIntrinsicInfo.cpp
@@ -12,13 +12,13 @@
 //===----------------------------------------------------------------------===//
 
 #include "MBlazeIntrinsicInfo.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/Module.h"
-#include "llvm/Type.h"
-#include "llvm/Support/raw_ostream.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
 #include <cstring>
 
 using namespace llvm;
@@ -104,7 +104,7 @@ Function *MBlazeIntrinsicInfo::getDeclaration(Module *M, unsigned IntrID,
                                                 Type **Tys,
                                                 unsigned numTy) const {
   assert(!isOverloaded(IntrID) && "MBlaze intrinsics are not overloaded");
-  AttrListPtr AList = getAttributes(M->getContext(),
+  AttributeSet AList = getAttributes(M->getContext(),
                                     (mblazeIntrinsic::ID) IntrID);
   return cast<Function>(M->getOrInsertFunction(getName(IntrID),
                                                getType(M->getContext(), IntrID),
diff --git a/lib/Target/MBlaze/MBlazeMCInstLower.cpp b/lib/Target/MBlaze/MBlazeMCInstLower.cpp
index 6b9f42ec91a6..ad414ac40fd7 100644
--- a/lib/Target/MBlaze/MBlazeMCInstLower.cpp
+++ b/lib/Target/MBlaze/MBlazeMCInstLower.cpp
@@ -14,19 +14,19 @@
 
 #include "MBlazeMCInstLower.h"
 #include "MBlazeInstrInfo.h"
-#include "llvm/Constants.h"
+#include "llvm/ADT/SmallString.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/IR/Constants.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
-#include "llvm/Target/Mangler.h"
 #include "llvm/Support/Debug.h"
-#include "llvm/Support/raw_ostream.h"
 #include "llvm/Support/ErrorHandling.h"
-#include "llvm/ADT/SmallString.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/Mangler.h"
 using namespace llvm;
 
 MCSymbol *MBlazeMCInstLower::
diff --git a/lib/Target/MBlaze/MBlazeMachineFunction.h b/lib/Target/MBlaze/MBlazeMachineFunction.h
index 95cc5077cc16..10d507f37bbc 100644
--- a/lib/Target/MBlaze/MBlazeMachineFunction.h
+++ b/lib/Target/MBlaze/MBlazeMachineFunction.h
@@ -16,8 +16,8 @@
 
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallVector.h"
-#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
 
 namespace llvm {
 
diff --git a/lib/Target/MBlaze/MBlazeRegisterInfo.cpp b/lib/Target/MBlaze/MBlazeRegisterInfo.cpp
index daa76e887fca..bd83afc1cc83 100644
--- a/lib/Target/MBlaze/MBlazeRegisterInfo.cpp
+++ b/lib/Target/MBlaze/MBlazeRegisterInfo.cpp
@@ -16,25 +16,25 @@
 
 #include "MBlazeRegisterInfo.h"
 #include "MBlaze.h"
-#include "MBlazeSubtarget.h"
 #include "MBlazeMachineFunction.h"
-#include "llvm/Constants.h"
-#include "llvm/Type.h"
-#include "llvm/Function.h"
-#include "llvm/CodeGen/ValueTypes.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunction.h"
+#include "MBlazeSubtarget.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/STLExtras.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
 
 #define GET_REGINFO_TARGET_DESC
 #include "MBlazeGenRegisterInfo.inc"
@@ -83,67 +83,21 @@ getReservedRegs(const MachineFunction &MF) const {
   return Reserved;
 }
 
-// This function eliminate ADJCALLSTACKDOWN/ADJCALLSTACKUP pseudo instructions
-void MBlazeRegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-
-  if (!TFI->hasReservedCallFrame(MF)) {
-    // If we have a frame pointer, turn the adjcallstackup instruction into a
-    // 'addi r1, r1, -<amt>' and the adjcallstackdown instruction into
-    // 'addi r1, r1, <amt>'
-    MachineInstr *Old = I;
-    int Amount = Old->getOperand(0).getImm() + 4;
-    if (Amount != 0) {
-      // We need to keep the stack aligned properly.  To do this, we round the
-      // amount of space needed for the outgoing arguments up to the next
-      // alignment boundary.
-      unsigned Align = TFI->getStackAlignment();
-      Amount = (Amount+Align-1)/Align*Align;
-
-      MachineInstr *New;
-      if (Old->getOpcode() == MBlaze::ADJCALLSTACKDOWN) {
-        New = BuildMI(MF,Old->getDebugLoc(),TII.get(MBlaze::ADDIK),MBlaze::R1)
-                .addReg(MBlaze::R1).addImm(-Amount);
-      } else {
-        assert(Old->getOpcode() == MBlaze::ADJCALLSTACKUP);
-        New = BuildMI(MF,Old->getDebugLoc(),TII.get(MBlaze::ADDIK),MBlaze::R1)
-                .addReg(MBlaze::R1).addImm(Amount);
-      }
-
-      // Replace the pseudo instruction with a new instruction...
-      MBB.insert(I, New);
-    }
-  }
-
-  // Simply discard ADJCALLSTACKDOWN, ADJCALLSTACKUP instructions.
-  MBB.erase(I);
-}
-
 // FrameIndex represent objects inside a abstract stack.
 // We must replace FrameIndex with an stack/frame pointer
 // direct reference.
 void MBlazeRegisterInfo::
 eliminateFrameIndex(MachineBasicBlock::iterator II, int SPAdj,
-                    RegScavenger *RS) const {
+                    unsigned FIOperandNum, RegScavenger *RS) const {
   MachineInstr &MI = *II;
   MachineFunction &MF = *MI.getParent()->getParent();
   MachineFrameInfo *MFI = MF.getFrameInfo();
-
-  unsigned i = 0;
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() &&
-           "Instr doesn't have FrameIndex operand!");
-  }
-
-  unsigned oi = i == 2 ? 1 : 2;
+  unsigned OFIOperandNum = FIOperandNum == 2 ? 1 : 2;
 
   DEBUG(dbgs() << "\nFunction : " << MF.getName() << "\n";
         dbgs() << "<--------->\n" << MI);
 
-  int FrameIndex = MI.getOperand(i).getIndex();
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
   int stackSize  = MFI->getStackSize();
   int spOffset   = MFI->getObjectOffset(FrameIndex);
 
@@ -159,16 +113,16 @@ eliminateFrameIndex(MachineBasicBlock::iterator II, int SPAdj,
   // as explained on LowerFormalArguments, detect negative offsets
   // and adjust SPOffsets considering the final stack size.
   int Offset = (spOffset < 0) ? (stackSize - spOffset) : spOffset;
-  Offset += MI.getOperand(oi).getImm();
+  Offset += MI.getOperand(OFIOperandNum).getImm();
 
   DEBUG(dbgs() << "Offset     : " << Offset << "\n" << "<--------->\n");
 
-  MI.getOperand(oi).ChangeToImmediate(Offset);
-  MI.getOperand(i).ChangeToRegister(getFrameRegister(MF), false);
+  MI.getOperand(OFIOperandNum).ChangeToImmediate(Offset);
+  MI.getOperand(FIOperandNum).ChangeToRegister(getFrameRegister(MF), false);
 }
 
 void MBlazeRegisterInfo::
-processFunctionBeforeFrameFinalized(MachineFunction &MF) const {
+processFunctionBeforeFrameFinalized(MachineFunction &MF, RegScavenger *) const {
   // Set the stack offset where GP must be saved/loaded from.
   MachineFrameInfo *MFI = MF.getFrameInfo();
   MBlazeFunctionInfo *MBlazeFI = MF.getInfo<MBlazeFunctionInfo>();
diff --git a/lib/Target/MBlaze/MBlazeRegisterInfo.h b/lib/Target/MBlaze/MBlazeRegisterInfo.h
index 1d5116293516..497f3866c9ca 100644
--- a/lib/Target/MBlaze/MBlazeRegisterInfo.h
+++ b/lib/Target/MBlaze/MBlazeRegisterInfo.h
@@ -50,15 +50,13 @@ struct MBlazeRegisterInfo : public MBlazeGenRegisterInfo {
 
   BitVector getReservedRegs(const MachineFunction &MF) const;
 
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
-
   /// Stack Frame Processing Methods
   void eliminateFrameIndex(MachineBasicBlock::iterator II,
-                           int SPAdj, RegScavenger *RS = NULL) const;
+                           int SPAdj, unsigned FIOperandNum,
+                           RegScavenger *RS = NULL) const;
 
-  void processFunctionBeforeFrameFinalized(MachineFunction &MF) const;
+  void processFunctionBeforeFrameFinalized(MachineFunction &MF,
+                                           RegScavenger *RS = NULL) const;
 
   /// Debug information queries.
   unsigned getFrameRegister(const MachineFunction &MF) const;
diff --git a/lib/Target/MBlaze/MBlazeSubtarget.h b/lib/Target/MBlaze/MBlazeSubtarget.h
index eb375046f218..ed43d21f30c5 100644
--- a/lib/Target/MBlaze/MBlazeSubtarget.h
+++ b/lib/Target/MBlaze/MBlazeSubtarget.h
@@ -14,8 +14,8 @@
 #ifndef MBLAZESUBTARGET_H
 #define MBLAZESUBTARGET_H
 
-#include "llvm/Target/TargetSubtargetInfo.h"
 #include "llvm/MC/MCInstrItineraries.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
 #include <string>
 
 #define GET_SUBTARGETINFO_HEADER
diff --git a/lib/Target/MBlaze/MBlazeTargetMachine.cpp b/lib/Target/MBlaze/MBlazeTargetMachine.cpp
index f180652f1127..bcdd32fed947 100644
--- a/lib/Target/MBlaze/MBlazeTargetMachine.cpp
+++ b/lib/Target/MBlaze/MBlazeTargetMachine.cpp
@@ -13,8 +13,8 @@
 
 #include "MBlazeTargetMachine.h"
 #include "MBlaze.h"
-#include "llvm/PassManager.h"
 #include "llvm/CodeGen/Passes.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/FormattedStream.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Target/TargetOptions.h"
@@ -42,8 +42,7 @@ MBlazeTargetMachine(const Target &T, StringRef TT,
     InstrInfo(*this),
     FrameLowering(Subtarget),
     TLInfo(*this), TSInfo(*this),
-    InstrItins(Subtarget.getInstrItineraryData()),
-    STTI(&TLInfo), VTTI(&TLInfo) {
+    InstrItins(Subtarget.getInstrItineraryData()) {
 }
 
 namespace {
diff --git a/lib/Target/MBlaze/MBlazeTargetMachine.h b/lib/Target/MBlaze/MBlazeTargetMachine.h
index a8df4e63e3ee..956794dddaf9 100644
--- a/lib/Target/MBlaze/MBlazeTargetMachine.h
+++ b/lib/Target/MBlaze/MBlazeTargetMachine.h
@@ -14,17 +14,16 @@
 #ifndef MBLAZE_TARGETMACHINE_H
 #define MBLAZE_TARGETMACHINE_H
 
-#include "MBlazeSubtarget.h"
-#include "MBlazeInstrInfo.h"
+#include "MBlazeFrameLowering.h"
 #include "MBlazeISelLowering.h"
-#include "MBlazeSelectionDAGInfo.h"
+#include "MBlazeInstrInfo.h"
 #include "MBlazeIntrinsicInfo.h"
-#include "MBlazeFrameLowering.h"
+#include "MBlazeSelectionDAGInfo.h"
+#include "MBlazeSubtarget.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/MC/MCStreamer.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetTransformImpl.h"
+#include "llvm/Target/TargetMachine.h"
 
 namespace llvm {
   class formatted_raw_ostream;
@@ -38,8 +37,6 @@ namespace llvm {
     MBlazeSelectionDAGInfo TSInfo;
     MBlazeIntrinsicInfo    IntrinsicInfo;
     InstrItineraryData     InstrItins;
-    ScalarTargetTransformImpl STTI;
-    VectorTargetTransformImpl VTTI;
 
   public:
     MBlazeTargetMachine(const Target &T, StringRef TT,
@@ -75,11 +72,6 @@ namespace llvm {
     const TargetIntrinsicInfo *getIntrinsicInfo() const
     { return &IntrinsicInfo; }
 
-    virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const
-    { return &STTI; }
-    virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const
-    { return &VTTI; }
-
     // Pass Pipeline Configuration
     virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
   };
diff --git a/lib/Target/MBlaze/MBlazeTargetObjectFile.cpp b/lib/Target/MBlaze/MBlazeTargetObjectFile.cpp
index 899c74ee8ed7..a7a0a68b1612 100644
--- a/lib/Target/MBlaze/MBlazeTargetObjectFile.cpp
+++ b/lib/Target/MBlaze/MBlazeTargetObjectFile.cpp
@@ -9,14 +9,14 @@
 
 #include "MBlazeTargetObjectFile.h"
 #include "MBlazeSubtarget.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/GlobalVariable.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/GlobalVariable.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCSectionELF.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ELF.h"
+#include "llvm/Target/TargetMachine.h"
 using namespace llvm;
 
 void MBlazeTargetObjectFile::
diff --git a/lib/Target/MBlaze/MCTargetDesc/MBlazeAsmBackend.cpp b/lib/Target/MBlaze/MCTargetDesc/MBlazeAsmBackend.cpp
index 44feeb49e7f1..6f9752c42951 100644
--- a/lib/Target/MBlaze/MCTargetDesc/MBlazeAsmBackend.cpp
+++ b/lib/Target/MBlaze/MCTargetDesc/MBlazeAsmBackend.cpp
@@ -8,9 +8,10 @@
 //===----------------------------------------------------------------------===//
 
 #include "MCTargetDesc/MBlazeMCTargetDesc.h"
+#include "llvm/ADT/Twine.h"
 #include "llvm/MC/MCAsmBackend.h"
-#include "llvm/MC/MCAssembler.h"
 #include "llvm/MC/MCAsmLayout.h"
+#include "llvm/MC/MCAssembler.h"
 #include "llvm/MC/MCELFObjectWriter.h"
 #include "llvm/MC/MCELFSymbolFlags.h"
 #include "llvm/MC/MCExpr.h"
@@ -18,7 +19,6 @@
 #include "llvm/MC/MCSectionELF.h"
 #include "llvm/MC/MCSectionMachO.h"
 #include "llvm/MC/MCValue.h"
-#include "llvm/ADT/Twine.h"
 #include "llvm/Support/ELF.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
@@ -54,7 +54,7 @@ public:
 
   bool fixupNeedsRelaxation(const MCFixup &Fixup,
                             uint64_t Value,
-                            const MCInstFragment *DF,
+                            const MCRelaxableFragment *DF,
                             const MCAsmLayout &Layout) const;
 
   void relaxInstruction(const MCInst &Inst, MCInst &Res) const;
@@ -88,7 +88,7 @@ bool MBlazeAsmBackend::mayNeedRelaxation(const MCInst &Inst) const {
 
 bool MBlazeAsmBackend::fixupNeedsRelaxation(const MCFixup &Fixup,
                                             uint64_t Value,
-                                            const MCInstFragment *DF,
+                                            const MCRelaxableFragment *DF,
                                             const MCAsmLayout &Layout) const {
   // FIXME: Is this right? It's what the "generic" code was doing before,
   // but is X86 specific. Is it actually true for MBlaze also, or was it
diff --git a/lib/Target/MBlaze/MCTargetDesc/MBlazeMCCodeEmitter.cpp b/lib/Target/MBlaze/MCTargetDesc/MBlazeMCCodeEmitter.cpp
index 2b71d9d3c844..8faff6ade441 100644
--- a/lib/Target/MBlaze/MCTargetDesc/MBlazeMCCodeEmitter.cpp
+++ b/lib/Target/MBlaze/MCTargetDesc/MBlazeMCCodeEmitter.cpp
@@ -12,16 +12,16 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "mccodeemitter"
-#include "MCTargetDesc/MBlazeBaseInfo.h"
 #include "MCTargetDesc/MBlazeMCTargetDesc.h"
+#include "MCTargetDesc/MBlazeBaseInfo.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/MC/MCCodeEmitter.h"
 #include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCFixup.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/MC/MCFixup.h"
-#include "llvm/ADT/Statistic.h"
 #include "llvm/Support/raw_ostream.h"
 using namespace llvm;
 
diff --git a/lib/Target/MBlaze/MCTargetDesc/MBlazeMCTargetDesc.cpp b/lib/Target/MBlaze/MCTargetDesc/MBlazeMCTargetDesc.cpp
index 9a7549b0e7cf..380750d50f4c 100644
--- a/lib/Target/MBlaze/MCTargetDesc/MBlazeMCTargetDesc.cpp
+++ b/lib/Target/MBlaze/MCTargetDesc/MBlazeMCTargetDesc.cpp
@@ -12,8 +12,8 @@
 //===----------------------------------------------------------------------===//
 
 #include "MBlazeMCTargetDesc.h"
-#include "MBlazeMCAsmInfo.h"
 #include "InstPrinter/MBlazeInstPrinter.h"
+#include "MBlazeMCAsmInfo.h"
 #include "llvm/MC/MCCodeGenInfo.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
diff --git a/lib/Target/MBlaze/Makefile b/lib/Target/MBlaze/Makefile
index 83c2a7d34da1..512ce9a08103 100644
--- a/lib/Target/MBlaze/Makefile
+++ b/lib/Target/MBlaze/Makefile
@@ -15,8 +15,7 @@ BUILT_SOURCES = MBlazeGenRegisterInfo.inc MBlazeGenInstrInfo.inc \
 		MBlazeGenAsmWriter.inc \
                 MBlazeGenDAGISel.inc MBlazeGenAsmMatcher.inc \
                 MBlazeGenCodeEmitter.inc MBlazeGenCallingConv.inc \
-                MBlazeGenSubtargetInfo.inc MBlazeGenIntrinsics.inc \
-                MBlazeGenEDInfo.inc
+                MBlazeGenSubtargetInfo.inc MBlazeGenIntrinsics.inc
 
 DIRS = InstPrinter AsmParser Disassembler TargetInfo MCTargetDesc
 
diff --git a/lib/Target/MBlaze/TargetInfo/MBlazeTargetInfo.cpp b/lib/Target/MBlaze/TargetInfo/MBlazeTargetInfo.cpp
index 71210d8db466..323a7f647d56 100644
--- a/lib/Target/MBlaze/TargetInfo/MBlazeTargetInfo.cpp
+++ b/lib/Target/MBlaze/TargetInfo/MBlazeTargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "MBlaze.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
diff --git a/lib/Target/MSP430/InstPrinter/MSP430InstPrinter.cpp b/lib/Target/MSP430/InstPrinter/MSP430InstPrinter.cpp
index 0930c453e954..4b12aeadd3e4 100644
--- a/lib/Target/MSP430/InstPrinter/MSP430InstPrinter.cpp
+++ b/lib/Target/MSP430/InstPrinter/MSP430InstPrinter.cpp
@@ -12,11 +12,11 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "asm-printer"
-#include "MSP430.h"
 #include "MSP430InstPrinter.h"
-#include "llvm/MC/MCInst.h"
+#include "MSP430.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormattedStream.h"
 using namespace llvm;
diff --git a/lib/Target/MSP430/MCTargetDesc/MSP430MCAsmInfo.cpp b/lib/Target/MSP430/MCTargetDesc/MSP430MCAsmInfo.cpp
index 2e328cb5d6ac..3c9576056946 100644
--- a/lib/Target/MSP430/MCTargetDesc/MSP430MCAsmInfo.cpp
+++ b/lib/Target/MSP430/MCTargetDesc/MSP430MCAsmInfo.cpp
@@ -18,7 +18,7 @@ using namespace llvm;
 void MSP430MCAsmInfo::anchor() { }
 
 MSP430MCAsmInfo::MSP430MCAsmInfo(const Target &T, StringRef TT) {
-  PointerSize = 2;
+  PointerSize = CalleeSaveStackSlotSize = 2;
 
   PrivateGlobalPrefix = ".L";
   WeakRefDirective ="\t.weak\t";
diff --git a/lib/Target/MSP430/MCTargetDesc/MSP430MCTargetDesc.cpp b/lib/Target/MSP430/MCTargetDesc/MSP430MCTargetDesc.cpp
index c455f6bc24f2..530e6aae92fd 100644
--- a/lib/Target/MSP430/MCTargetDesc/MSP430MCTargetDesc.cpp
+++ b/lib/Target/MSP430/MCTargetDesc/MSP430MCTargetDesc.cpp
@@ -12,8 +12,8 @@
 //===----------------------------------------------------------------------===//
 
 #include "MSP430MCTargetDesc.h"
-#include "MSP430MCAsmInfo.h"
 #include "InstPrinter/MSP430InstPrinter.h"
+#include "MSP430MCAsmInfo.h"
 #include "llvm/MC/MCCodeGenInfo.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
diff --git a/lib/Target/MSP430/MSP430AsmPrinter.cpp b/lib/Target/MSP430/MSP430AsmPrinter.cpp
index 86bc183c1bdf..0a04e5ddb75d 100644
--- a/lib/Target/MSP430/MSP430AsmPrinter.cpp
+++ b/lib/Target/MSP430/MSP430AsmPrinter.cpp
@@ -14,26 +14,26 @@
 
 #define DEBUG_TYPE "asm-printer"
 #include "MSP430.h"
+#include "InstPrinter/MSP430InstPrinter.h"
 #include "MSP430InstrInfo.h"
 #include "MSP430MCInstLower.h"
 #include "MSP430TargetMachine.h"
-#include "InstPrinter/MSP430InstPrinter.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Module.h"
 #include "llvm/Assembly/Writer.h"
 #include "llvm/CodeGen/AsmPrinter.h"
-#include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Module.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/Target/Mangler.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/Mangler.h"
 using namespace llvm;
 
 namespace {
diff --git a/lib/Target/MSP430/MSP430BranchSelector.cpp b/lib/Target/MSP430/MSP430BranchSelector.cpp
index bdeb0c590f2d..f128427f8066 100644
--- a/lib/Target/MSP430/MSP430BranchSelector.cpp
+++ b/lib/Target/MSP430/MSP430BranchSelector.cpp
@@ -18,11 +18,11 @@
 #define DEBUG_TYPE "msp430-branch-select"
 #include "MSP430.h"
 #include "MSP430InstrInfo.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/Support/MathExtras.h"
+#include "llvm/Target/TargetMachine.h"
 using namespace llvm;
 
 STATISTIC(NumExpanded, "Number of branches expanded to long format");
diff --git a/lib/Target/MSP430/MSP430CallingConv.td b/lib/Target/MSP430/MSP430CallingConv.td
index ad27cc9122a8..b448cc4ed9b8 100644
--- a/lib/Target/MSP430/MSP430CallingConv.td
+++ b/lib/Target/MSP430/MSP430CallingConv.td
@@ -24,6 +24,9 @@ def RetCC_MSP430 : CallingConv<[
 // MSP430 Argument Calling Conventions
 //===----------------------------------------------------------------------===//
 def CC_MSP430 : CallingConv<[
+  // Pass by value if the byval attribute is given
+  CCIfByVal<CCPassByVal<2, 2>>,
+
   // Promote i8 arguments to i16.
   CCIfType<[i8], CCPromoteToType<i16>>,
 
diff --git a/lib/Target/MSP430/MSP430FrameLowering.cpp b/lib/Target/MSP430/MSP430FrameLowering.cpp
index 2e170f17bf9d..e504011dfdc8 100644
--- a/lib/Target/MSP430/MSP430FrameLowering.cpp
+++ b/lib/Target/MSP430/MSP430FrameLowering.cpp
@@ -14,15 +14,15 @@
 #include "MSP430FrameLowering.h"
 #include "MSP430InstrInfo.h"
 #include "MSP430MachineFunctionInfo.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
@@ -222,13 +222,73 @@ MSP430FrameLowering::restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
   return true;
 }
 
-void
-MSP430FrameLowering::processFunctionBeforeFrameFinalized(MachineFunction &MF)
-                                                                         const {
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
+void MSP430FrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  const MSP430InstrInfo &TII =
+    *static_cast<const MSP430InstrInfo*>(MF.getTarget().getInstrInfo());
+  unsigned StackAlign = getStackAlignment();
+
+  if (!hasReservedCallFrame(MF)) {
+    // If the stack pointer can be changed after prologue, turn the
+    // adjcallstackup instruction into a 'sub SPW, <amt>' and the
+    // adjcallstackdown instruction into 'add SPW, <amt>'
+    // TODO: consider using push / pop instead of sub + store / add
+    MachineInstr *Old = I;
+    uint64_t Amount = Old->getOperand(0).getImm();
+    if (Amount != 0) {
+      // We need to keep the stack aligned properly.  To do this, we round the
+      // amount of space needed for the outgoing arguments up to the next
+      // alignment boundary.
+      Amount = (Amount+StackAlign-1)/StackAlign*StackAlign;
+
+      MachineInstr *New = 0;
+      if (Old->getOpcode() == TII.getCallFrameSetupOpcode()) {
+        New = BuildMI(MF, Old->getDebugLoc(),
+                      TII.get(MSP430::SUB16ri), MSP430::SPW)
+          .addReg(MSP430::SPW).addImm(Amount);
+      } else {
+        assert(Old->getOpcode() == TII.getCallFrameDestroyOpcode());
+        // factor out the amount the callee already popped.
+        uint64_t CalleeAmt = Old->getOperand(1).getImm();
+        Amount -= CalleeAmt;
+        if (Amount)
+          New = BuildMI(MF, Old->getDebugLoc(),
+                        TII.get(MSP430::ADD16ri), MSP430::SPW)
+            .addReg(MSP430::SPW).addImm(Amount);
+      }
+
+      if (New) {
+        // The SRW implicit def is dead.
+        New->getOperand(3).setIsDead();
+
+        // Replace the pseudo instruction with a new instruction...
+        MBB.insert(I, New);
+      }
+    }
+  } else if (I->getOpcode() == TII.getCallFrameDestroyOpcode()) {
+    // If we are performing frame pointer elimination and if the callee pops
+    // something off the stack pointer, add it back.
+    if (uint64_t CalleeAmt = I->getOperand(1).getImm()) {
+      MachineInstr *Old = I;
+      MachineInstr *New =
+        BuildMI(MF, Old->getDebugLoc(), TII.get(MSP430::SUB16ri),
+                MSP430::SPW).addReg(MSP430::SPW).addImm(CalleeAmt);
+      // The SRW implicit def is dead.
+      New->getOperand(3).setIsDead();
 
+      MBB.insert(I, New);
+    }
+  }
+
+  MBB.erase(I);
+}
+
+void
+MSP430FrameLowering::processFunctionBeforeFrameFinalized(MachineFunction &MF,
+                                                         RegScavenger *) const {
   // Create a frame entry for the FPW register that must be saved.
-  if (TFI->hasFP(MF)) {
+  if (hasFP(MF)) {
     int FrameIdx = MF.getFrameInfo()->CreateFixedObject(2, -4, true);
     (void)FrameIdx;
     assert(FrameIdx == MF.getFrameInfo()->getObjectIndexBegin() &&
diff --git a/lib/Target/MSP430/MSP430FrameLowering.h b/lib/Target/MSP430/MSP430FrameLowering.h
index cb02545852b5..c673f59b5efc 100644
--- a/lib/Target/MSP430/MSP430FrameLowering.h
+++ b/lib/Target/MSP430/MSP430FrameLowering.h
@@ -35,6 +35,10 @@ public:
   void emitPrologue(MachineFunction &MF) const;
   void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
 
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I) const;
+
   bool spillCalleeSavedRegisters(MachineBasicBlock &MBB,
                                  MachineBasicBlock::iterator MI,
                                  const std::vector<CalleeSavedInfo> &CSI,
@@ -46,7 +50,8 @@ public:
 
   bool hasFP(const MachineFunction &MF) const;
   bool hasReservedCallFrame(const MachineFunction &MF) const;
-  void processFunctionBeforeFrameFinalized(MachineFunction &MF) const;
+  void processFunctionBeforeFrameFinalized(MachineFunction &MF,
+                                       RegScavenger *RS = NULL) const;
 };
 
 } // End llvm namespace
diff --git a/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp b/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
index 5efc6a36b894..1566c096037e 100644
--- a/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
+++ b/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
@@ -13,22 +13,22 @@
 
 #include "MSP430.h"
 #include "MSP430TargetMachine.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
-#include "llvm/Target/TargetLowering.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetLowering.h"
 using namespace llvm;
 
 namespace {
diff --git a/lib/Target/MSP430/MSP430ISelLowering.cpp b/lib/Target/MSP430/MSP430ISelLowering.cpp
index fc677aec38ef..09cdf3268553 100644
--- a/lib/Target/MSP430/MSP430ISelLowering.cpp
+++ b/lib/Target/MSP430/MSP430ISelLowering.cpp
@@ -16,14 +16,8 @@
 #include "MSP430ISelLowering.h"
 #include "MSP430.h"
 #include "MSP430MachineFunctionInfo.h"
-#include "MSP430TargetMachine.h"
 #include "MSP430Subtarget.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/CallingConv.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/GlobalAlias.h"
+#include "MSP430TargetMachine.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -32,6 +26,12 @@
 #include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
 #include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
@@ -164,6 +164,12 @@ MSP430TargetLowering::MSP430TargetLowering(MSP430TargetMachine &tm) :
   setOperationAction(ISD::SDIVREM,          MVT::i16,   Expand);
   setOperationAction(ISD::SREM,             MVT::i16,   Expand);
 
+  // varargs support
+  setOperationAction(ISD::VASTART,          MVT::Other, Custom);
+  setOperationAction(ISD::VAARG,            MVT::Other, Expand);
+  setOperationAction(ISD::VAEND,            MVT::Other, Expand);
+  setOperationAction(ISD::VACOPY,           MVT::Other, Expand);
+
   // Libcalls names.
   if (HWMultMode == HWMultIntr) {
     setLibcallName(RTLIB::MUL_I8,  "__mulqi3hw");
@@ -192,6 +198,7 @@ SDValue MSP430TargetLowering::LowerOperation(SDValue Op,
   case ISD::SIGN_EXTEND:      return LowerSIGN_EXTEND(Op, DAG);
   case ISD::RETURNADDR:       return LowerRETURNADDR(Op, DAG);
   case ISD::FRAMEADDR:        return LowerFRAMEADDR(Op, DAG);
+  case ISD::VASTART:          return LowerVASTART(Op, DAG);
   default:
     llvm_unreachable("unimplemented operand");
   }
@@ -297,7 +304,6 @@ MSP430TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
 /// LowerCCCArguments - transform physical registers into virtual registers and
 /// generate load operations for arguments places on the stack.
 // FIXME: struct return stuff
-// FIXME: varargs
 SDValue
 MSP430TargetLowering::LowerCCCArguments(SDValue Chain,
                                         CallingConv::ID CallConv,
@@ -311,6 +317,7 @@ MSP430TargetLowering::LowerCCCArguments(SDValue Chain,
   MachineFunction &MF = DAG.getMachineFunction();
   MachineFrameInfo *MFI = MF.getFrameInfo();
   MachineRegisterInfo &RegInfo = MF.getRegInfo();
+  MSP430MachineFunctionInfo *FuncInfo = MF.getInfo<MSP430MachineFunctionInfo>();
 
   // Assign locations to all of the incoming arguments.
   SmallVector<CCValAssign, 16> ArgLocs;
@@ -318,7 +325,11 @@ MSP430TargetLowering::LowerCCCArguments(SDValue Chain,
                  getTargetMachine(), ArgLocs, *DAG.getContext());
   CCInfo.AnalyzeFormalArguments(Ins, CC_MSP430);
 
-  assert(!isVarArg && "Varargs not supported yet");
+  // Create frame index for the start of the first vararg value
+  if (isVarArg) {
+    unsigned Offset = CCInfo.getNextStackOffset();
+    FuncInfo->setVarArgsFrameIndex(MFI->CreateFixedObject(1, Offset, true));
+  }
 
   for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
     CCValAssign &VA = ArgLocs[i];
@@ -357,22 +368,34 @@ MSP430TargetLowering::LowerCCCArguments(SDValue Chain,
     } else {
       // Sanity check
       assert(VA.isMemLoc());
-      // Load the argument to a virtual register
-      unsigned ObjSize = VA.getLocVT().getSizeInBits()/8;
-      if (ObjSize > 2) {
-        errs() << "LowerFormalArguments Unhandled argument type: "
-             << EVT(VA.getLocVT()).getEVTString()
-             << "\n";
+
+      SDValue InVal;
+      ISD::ArgFlagsTy Flags = Ins[i].Flags;
+
+      if (Flags.isByVal()) {
+        int FI = MFI->CreateFixedObject(Flags.getByValSize(),
+                                        VA.getLocMemOffset(), true);
+        InVal = DAG.getFrameIndex(FI, getPointerTy());
+      } else {
+        // Load the argument to a virtual register
+        unsigned ObjSize = VA.getLocVT().getSizeInBits()/8;
+        if (ObjSize > 2) {
+            errs() << "LowerFormalArguments Unhandled argument type: "
+                << EVT(VA.getLocVT()).getEVTString()
+                << "\n";
+        }
+        // Create the frame index object for this incoming parameter...
+        int FI = MFI->CreateFixedObject(ObjSize, VA.getLocMemOffset(), true);
+
+        // Create the SelectionDAG nodes corresponding to a load
+        //from this parameter
+        SDValue FIN = DAG.getFrameIndex(FI, MVT::i16);
+        InVal = DAG.getLoad(VA.getLocVT(), dl, Chain, FIN,
+                            MachinePointerInfo::getFixedStack(FI),
+                            false, false, false, 0);
       }
-      // Create the frame index object for this incoming parameter...
-      int FI = MFI->CreateFixedObject(ObjSize, VA.getLocMemOffset(), true);
-
-      // Create the SelectionDAG nodes corresponding to a load
-      //from this parameter
-      SDValue FIN = DAG.getFrameIndex(FI, MVT::i16);
-      InVals.push_back(DAG.getLoad(VA.getLocVT(), dl, Chain, FIN,
-                                   MachinePointerInfo::getFixedStack(FI),
-                                   false, false, false, 0));
+
+      InVals.push_back(InVal);
     }
   }
 
@@ -400,15 +423,8 @@ MSP430TargetLowering::LowerReturn(SDValue Chain,
   // Analize return values.
   CCInfo.AnalyzeReturn(Outs, RetCC_MSP430);
 
-  // If this is the first return lowered for this function, add the regs to the
-  // liveout set for the function.
-  if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
-    for (unsigned i = 0; i != RVLocs.size(); ++i)
-      if (RVLocs[i].isRegLoc())
-        DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
-  }
-
   SDValue Flag;
+  SmallVector<SDValue, 4> RetOps(1, Chain);
 
   // Copy the result values into the output registers.
   for (unsigned i = 0; i != RVLocs.size(); ++i) {
@@ -421,16 +437,19 @@ MSP430TargetLowering::LowerReturn(SDValue Chain,
     // Guarantee that all emitted copies are stuck together,
     // avoiding something bad.
     Flag = Chain.getValue(1);
+    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
   }
 
   unsigned Opc = (CallConv == CallingConv::MSP430_INTR ?
                   MSP430ISD::RETI_FLAG : MSP430ISD::RET_FLAG);
 
+  RetOps[0] = Chain;  // Update chain.
+
+  // Add the flag if we have it.
   if (Flag.getNode())
-    return DAG.getNode(Opc, dl, MVT::Other, Chain, Flag);
+    RetOps.push_back(Flag);
 
-  // Return Void
-  return DAG.getNode(Opc, dl, MVT::Other, Chain);
+  return DAG.getNode(Opc, dl, MVT::Other, &RetOps[0], RetOps.size());
 }
 
 /// LowerCCCCallTo - functions arguments are copied from virtual regs to
@@ -498,9 +517,23 @@ MSP430TargetLowering::LowerCCCCallTo(SDValue Chain, SDValue Callee,
                                    StackPtr,
                                    DAG.getIntPtrConstant(VA.getLocMemOffset()));
 
+      SDValue MemOp;
+      ISD::ArgFlagsTy Flags = Outs[i].Flags;
+
+      if (Flags.isByVal()) {
+        SDValue SizeNode = DAG.getConstant(Flags.getByValSize(), MVT::i16);
+        MemOp = DAG.getMemcpy(Chain, dl, PtrOff, Arg, SizeNode,
+                              Flags.getByValAlign(),
+                              /*isVolatile*/false,
+                              /*AlwaysInline=*/true,
+                              MachinePointerInfo(),
+                              MachinePointerInfo());
+      } else {
+        MemOp = DAG.getStore(Chain, dl, Arg, PtrOff, MachinePointerInfo(),
+                             false, false, 0);
+      }
 
-      MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
-                                         MachinePointerInfo(),false, false, 0));
+      MemOpChains.push_back(MemOp);
     }
   }
 
@@ -931,6 +964,22 @@ SDValue MSP430TargetLowering::LowerFRAMEADDR(SDValue Op,
   return FrameAddr;
 }
 
+SDValue MSP430TargetLowering::LowerVASTART(SDValue Op,
+                                           SelectionDAG &DAG) const {
+  MachineFunction &MF = DAG.getMachineFunction();
+  MSP430MachineFunctionInfo *FuncInfo = MF.getInfo<MSP430MachineFunctionInfo>();
+
+  // Frame index of first vararg argument
+  SDValue FrameIndex = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(),
+                                         getPointerTy());
+  const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
+
+  // Create a store of the frame index to the location operand
+  return DAG.getStore(Op.getOperand(0), Op.getDebugLoc(), FrameIndex,
+                      Op.getOperand(1), MachinePointerInfo(SV),
+                      false, false, 0);
+}
+
 /// getPostIndexedAddressParts - returns true by value, base pointer and
 /// offset pointer and addressing mode by reference if this node can be
 /// combined with a load / store to form a post-indexed load / store.
@@ -1010,6 +1059,10 @@ bool MSP430TargetLowering::isZExtFree(EVT VT1, EVT VT2) const {
   return 0 && VT1 == MVT::i8 && VT2 == MVT::i16;
 }
 
+bool MSP430TargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
+  return isZExtFree(Val.getValueType(), VT2);
+}
+
 //===----------------------------------------------------------------------===//
 //  Other Lowering Code
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/MSP430/MSP430ISelLowering.h b/lib/Target/MSP430/MSP430ISelLowering.h
index 991304c23de3..e0ed870f5653 100644
--- a/lib/Target/MSP430/MSP430ISelLowering.h
+++ b/lib/Target/MSP430/MSP430ISelLowering.h
@@ -73,7 +73,7 @@ namespace llvm {
   public:
     explicit MSP430TargetLowering(MSP430TargetMachine &TM);
 
-    virtual MVT getShiftAmountTy(EVT LHSTy) const { return MVT::i8; }
+    virtual MVT getScalarShiftAmountTy(EVT LHSTy) const { return MVT::i8; }
 
     /// LowerOperation - Provide custom lowering hooks for some operations.
     virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
@@ -92,6 +92,7 @@ namespace llvm {
     SDValue LowerSIGN_EXTEND(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) const;
     SDValue getReturnAddressFrameIndex(SelectionDAG &DAG) const;
 
     TargetLowering::ConstraintType
@@ -115,6 +116,7 @@ namespace llvm {
     /// out to 16 bits.
     virtual bool isZExtFree(Type *Ty1, Type *Ty2) const;
     virtual bool isZExtFree(EVT VT1, EVT VT2) const;
+    virtual bool isZExtFree(SDValue Val, EVT VT2) const;
 
     MachineBasicBlock* EmitInstrWithCustomInserter(MachineInstr *MI,
                                                    MachineBasicBlock *BB) const;
diff --git a/lib/Target/MSP430/MSP430InstrInfo.cpp b/lib/Target/MSP430/MSP430InstrInfo.cpp
index be332f05b30b..a6b5f2f6d0bd 100644
--- a/lib/Target/MSP430/MSP430InstrInfo.cpp
+++ b/lib/Target/MSP430/MSP430InstrInfo.cpp
@@ -15,10 +15,10 @@
 #include "MSP430.h"
 #include "MSP430MachineFunctionInfo.h"
 #include "MSP430TargetMachine.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
 
diff --git a/lib/Target/MSP430/MSP430InstrInfo.td b/lib/Target/MSP430/MSP430InstrInfo.td
index f003574eda00..e45780d05803 100644
--- a/lib/Target/MSP430/MSP430InstrInfo.td
+++ b/lib/Target/MSP430/MSP430InstrInfo.td
@@ -40,9 +40,9 @@ def SDT_MSP430Shift        : SDTypeProfile<1, 2, [SDTCisSameAs<0, 1>,
 // MSP430 Specific Node Definitions.
 //===----------------------------------------------------------------------===//
 def MSP430retflag  : SDNode<"MSP430ISD::RET_FLAG", SDTNone,
-                       [SDNPHasChain, SDNPOptInGlue]>;
+                       [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
 def MSP430retiflag : SDNode<"MSP430ISD::RETI_FLAG", SDTNone,
-                       [SDNPHasChain, SDNPOptInGlue]>;
+                       [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
 
 def MSP430rra     : SDNode<"MSP430ISD::RRA", SDTIntUnaryOp, []>;
 def MSP430rla     : SDNode<"MSP430ISD::RLA", SDTIntUnaryOp, []>;
diff --git a/lib/Target/MSP430/MSP430MCInstLower.cpp b/lib/Target/MSP430/MSP430MCInstLower.cpp
index b1773fba7e92..043e5becadbb 100644
--- a/lib/Target/MSP430/MSP430MCInstLower.cpp
+++ b/lib/Target/MSP430/MSP430MCInstLower.cpp
@@ -13,6 +13,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "MSP430MCInstLower.h"
+#include "llvm/ADT/SmallString.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineInstr.h"
@@ -20,10 +21,9 @@
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
-#include "llvm/Target/Mangler.h"
-#include "llvm/Support/raw_ostream.h"
 #include "llvm/Support/ErrorHandling.h"
-#include "llvm/ADT/SmallString.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/Mangler.h"
 using namespace llvm;
 
 MCSymbol *MSP430MCInstLower::
diff --git a/lib/Target/MSP430/MSP430MachineFunctionInfo.h b/lib/Target/MSP430/MSP430MachineFunctionInfo.h
index 632d6dee275f..d1697f478cc2 100644
--- a/lib/Target/MSP430/MSP430MachineFunctionInfo.h
+++ b/lib/Target/MSP430/MSP430MachineFunctionInfo.h
@@ -30,6 +30,9 @@ class MSP430MachineFunctionInfo : public MachineFunctionInfo {
   /// ReturnAddrIndex - FrameIndex for return slot.
   int ReturnAddrIndex;
 
+  /// VarArgsFrameIndex - FrameIndex for start of varargs area.
+  int VarArgsFrameIndex;
+
 public:
   MSP430MachineFunctionInfo() : CalleeSavedFrameSize(0) {}
 
@@ -41,6 +44,9 @@ public:
 
   int getRAIndex() const { return ReturnAddrIndex; }
   void setRAIndex(int Index) { ReturnAddrIndex = Index; }
+
+  int getVarArgsFrameIndex() const { return VarArgsFrameIndex;}
+  void setVarArgsFrameIndex(int Index) { VarArgsFrameIndex = Index; }
 };
 
 } // End llvm namespace
diff --git a/lib/Target/MSP430/MSP430RegisterInfo.cpp b/lib/Target/MSP430/MSP430RegisterInfo.cpp
index 9ae238f66f57..0b3e9e259649 100644
--- a/lib/Target/MSP430/MSP430RegisterInfo.cpp
+++ b/lib/Target/MSP430/MSP430RegisterInfo.cpp
@@ -17,14 +17,14 @@
 #include "MSP430.h"
 #include "MSP430MachineFunctionInfo.h"
 #include "MSP430TargetMachine.h"
-#include "llvm/Function.h"
+#include "llvm/ADT/BitVector.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/IR/Function.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetOptions.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/Support/ErrorHandling.h"
 
 #define GET_REGINFO_TARGET_DESC
 #include "MSP430GenRegisterInfo.inc"
@@ -101,83 +101,18 @@ MSP430RegisterInfo::getPointerRegClass(const MachineFunction &MF, unsigned Kind)
   return &MSP430::GR16RegClass;
 }
 
-void MSP430RegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-
-  if (!TFI->hasReservedCallFrame(MF)) {
-    // If the stack pointer can be changed after prologue, turn the
-    // adjcallstackup instruction into a 'sub SPW, <amt>' and the
-    // adjcallstackdown instruction into 'add SPW, <amt>'
-    // TODO: consider using push / pop instead of sub + store / add
-    MachineInstr *Old = I;
-    uint64_t Amount = Old->getOperand(0).getImm();
-    if (Amount != 0) {
-      // We need to keep the stack aligned properly.  To do this, we round the
-      // amount of space needed for the outgoing arguments up to the next
-      // alignment boundary.
-      Amount = (Amount+StackAlign-1)/StackAlign*StackAlign;
-
-      MachineInstr *New = 0;
-      if (Old->getOpcode() == TII.getCallFrameSetupOpcode()) {
-        New = BuildMI(MF, Old->getDebugLoc(),
-                      TII.get(MSP430::SUB16ri), MSP430::SPW)
-          .addReg(MSP430::SPW).addImm(Amount);
-      } else {
-        assert(Old->getOpcode() == TII.getCallFrameDestroyOpcode());
-        // factor out the amount the callee already popped.
-        uint64_t CalleeAmt = Old->getOperand(1).getImm();
-        Amount -= CalleeAmt;
-        if (Amount)
-          New = BuildMI(MF, Old->getDebugLoc(),
-                        TII.get(MSP430::ADD16ri), MSP430::SPW)
-            .addReg(MSP430::SPW).addImm(Amount);
-      }
-
-      if (New) {
-        // The SRW implicit def is dead.
-        New->getOperand(3).setIsDead();
-
-        // Replace the pseudo instruction with a new instruction...
-        MBB.insert(I, New);
-      }
-    }
-  } else if (I->getOpcode() == TII.getCallFrameDestroyOpcode()) {
-    // If we are performing frame pointer elimination and if the callee pops
-    // something off the stack pointer, add it back.
-    if (uint64_t CalleeAmt = I->getOperand(1).getImm()) {
-      MachineInstr *Old = I;
-      MachineInstr *New =
-        BuildMI(MF, Old->getDebugLoc(), TII.get(MSP430::SUB16ri),
-                MSP430::SPW).addReg(MSP430::SPW).addImm(CalleeAmt);
-      // The SRW implicit def is dead.
-      New->getOperand(3).setIsDead();
-
-      MBB.insert(I, New);
-    }
-  }
-
-  MBB.erase(I);
-}
-
 void
 MSP430RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
-                                        int SPAdj, RegScavenger *RS) const {
+                                        int SPAdj, unsigned FIOperandNum,
+                                        RegScavenger *RS) const {
   assert(SPAdj == 0 && "Unexpected");
 
-  unsigned i = 0;
   MachineInstr &MI = *II;
   MachineBasicBlock &MBB = *MI.getParent();
   MachineFunction &MF = *MBB.getParent();
   const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
   DebugLoc dl = MI.getDebugLoc();
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
-  }
-
-  int FrameIndex = MI.getOperand(i).getIndex();
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
 
   unsigned BasePtr = (TFI->hasFP(MF) ? MSP430::FPW : MSP430::SPW);
   int Offset = MF.getFrameInfo()->getObjectOffset(FrameIndex);
@@ -191,7 +126,7 @@ MSP430RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
     Offset += 2; // Skip the saved FPW
 
   // Fold imm into offset
-  Offset += MI.getOperand(i+1).getImm();
+  Offset += MI.getOperand(FIOperandNum + 1).getImm();
 
   if (MI.getOpcode() == MSP430::ADD16ri) {
     // This is actually "load effective address" of the stack slot
@@ -199,7 +134,7 @@ MSP430RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
     // expand it into mov + add
 
     MI.setDesc(TII.get(MSP430::MOV16rr));
-    MI.getOperand(i).ChangeToRegister(BasePtr, false);
+    MI.getOperand(FIOperandNum).ChangeToRegister(BasePtr, false);
 
     if (Offset == 0)
       return;
@@ -216,8 +151,8 @@ MSP430RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
     return;
   }
 
-  MI.getOperand(i).ChangeToRegister(BasePtr, false);
-  MI.getOperand(i+1).ChangeToImmediate(Offset);
+  MI.getOperand(FIOperandNum).ChangeToRegister(BasePtr, false);
+  MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset);
 }
 
 unsigned MSP430RegisterInfo::getFrameRegister(const MachineFunction &MF) const {
diff --git a/lib/Target/MSP430/MSP430RegisterInfo.h b/lib/Target/MSP430/MSP430RegisterInfo.h
index 64a43bcafbb4..69cccb275259 100644
--- a/lib/Target/MSP430/MSP430RegisterInfo.h
+++ b/lib/Target/MSP430/MSP430RegisterInfo.h
@@ -42,12 +42,9 @@ public:
   const TargetRegisterClass*
   getPointerRegClass(const MachineFunction &MF, unsigned Kind = 0) const;
 
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
-
   void eliminateFrameIndex(MachineBasicBlock::iterator II,
-                           int SPAdj, RegScavenger *RS = NULL) const;
+                           int SPAdj, unsigned FIOperandNum,
+                           RegScavenger *RS = NULL) const;
 
   // Debug information queries.
   unsigned getFrameRegister(const MachineFunction &MF) const;
diff --git a/lib/Target/MSP430/MSP430TargetMachine.cpp b/lib/Target/MSP430/MSP430TargetMachine.cpp
index 13e37b373533..164e351df952 100644
--- a/lib/Target/MSP430/MSP430TargetMachine.cpp
+++ b/lib/Target/MSP430/MSP430TargetMachine.cpp
@@ -13,9 +13,9 @@
 
 #include "MSP430TargetMachine.h"
 #include "MSP430.h"
-#include "llvm/PassManager.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/MC/MCAsmInfo.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
@@ -36,7 +36,7 @@ MSP430TargetMachine::MSP430TargetMachine(const Target &T,
     // FIXME: Check DataLayout string.
     DL("e-p:16:16:16-i8:8:8-i16:16:16-i32:16:32-n8:16"),
     InstrInfo(*this), TLInfo(*this), TSInfo(*this),
-    FrameLowering(Subtarget), STTI(&TLInfo), VTTI(&TLInfo) { }
+    FrameLowering(Subtarget) { }
 
 namespace {
 /// MSP430 Code Generator Pass Configuration Options.
diff --git a/lib/Target/MSP430/MSP430TargetMachine.h b/lib/Target/MSP430/MSP430TargetMachine.h
index 186172ede428..be695a211109 100644
--- a/lib/Target/MSP430/MSP430TargetMachine.h
+++ b/lib/Target/MSP430/MSP430TargetMachine.h
@@ -15,16 +15,15 @@
 #ifndef LLVM_TARGET_MSP430_TARGETMACHINE_H
 #define LLVM_TARGET_MSP430_TARGETMACHINE_H
 
-#include "MSP430InstrInfo.h"
-#include "MSP430ISelLowering.h"
 #include "MSP430FrameLowering.h"
-#include "MSP430SelectionDAGInfo.h"
+#include "MSP430ISelLowering.h"
+#include "MSP430InstrInfo.h"
 #include "MSP430RegisterInfo.h"
+#include "MSP430SelectionDAGInfo.h"
 #include "MSP430Subtarget.h"
-#include "llvm/DataLayout.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Target/TargetFrameLowering.h"
 #include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetTransformImpl.h"
 
 namespace llvm {
 
@@ -37,8 +36,6 @@ class MSP430TargetMachine : public LLVMTargetMachine {
   MSP430TargetLowering   TLInfo;
   MSP430SelectionDAGInfo TSInfo;
   MSP430FrameLowering    FrameLowering;
-  ScalarTargetTransformImpl STTI;
-  VectorTargetTransformImpl VTTI;
 
 public:
   MSP430TargetMachine(const Target &T, StringRef TT,
@@ -64,12 +61,6 @@ public:
   virtual const MSP430SelectionDAGInfo* getSelectionDAGInfo() const {
     return &TSInfo;
   }
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
   virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
 }; // MSP430TargetMachine.
 
diff --git a/lib/Target/MSP430/TargetInfo/MSP430TargetInfo.cpp b/lib/Target/MSP430/TargetInfo/MSP430TargetInfo.cpp
index 8b3e01ecf52c..0d71d04ebe22 100644
--- a/lib/Target/MSP430/TargetInfo/MSP430TargetInfo.cpp
+++ b/lib/Target/MSP430/TargetInfo/MSP430TargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "MSP430.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
diff --git a/lib/Target/Mangler.cpp b/lib/Target/Mangler.cpp
index 539a1f723bdd..edfd421d8532 100644
--- a/lib/Target/Mangler.cpp
+++ b/lib/Target/Mangler.cpp
@@ -12,14 +12,14 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/Target/Mangler.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/DataLayout.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/Twine.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/SmallString.h"
-#include "llvm/ADT/Twine.h"
 using namespace llvm;
 
 static bool isAcceptableChar(char C, bool AllowPeriod, bool AllowUTF8) {
diff --git a/lib/Target/Mips/AsmParser/MipsAsmParser.cpp b/lib/Target/Mips/AsmParser/MipsAsmParser.cpp
index 67b524883cf8..c403f216b0d6 100644
--- a/lib/Target/Mips/AsmParser/MipsAsmParser.cpp
+++ b/lib/Target/Mips/AsmParser/MipsAsmParser.cpp
@@ -13,11 +13,11 @@
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCParser/MCAsmLexer.h"
+#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/MC/MCParser/MCAsmLexer.h"
-#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
 #include "llvm/MC/MCTargetAsmParser.h"
 #include "llvm/Support/TargetRegistry.h"
 
@@ -84,15 +84,33 @@ class MipsAsmParser : public MCTargetAsmParser {
   bool ParseDirective(AsmToken DirectiveID);
 
   MipsAsmParser::OperandMatchResultTy
-  parseMemOperand(SmallVectorImpl<MCParsedAsmOperand*>&);
+  parseMemOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  MipsAsmParser::OperandMatchResultTy
+  parseCPURegs(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  MipsAsmParser::OperandMatchResultTy
+  parseCPU64Regs(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  MipsAsmParser::OperandMatchResultTy
+  parseHWRegs(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  MipsAsmParser::OperandMatchResultTy
+  parseHW64Regs(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  MipsAsmParser::OperandMatchResultTy
+  parseCCRRegs(SmallVectorImpl<MCParsedAsmOperand*> &Operands);
+
+  bool searchSymbolAlias(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
+                         unsigned RegisterClass);
 
   bool ParseOperand(SmallVectorImpl<MCParsedAsmOperand*> &,
                     StringRef Mnemonic);
 
-  int tryParseRegister(StringRef Mnemonic);
+  int tryParseRegister(bool is64BitReg);
 
   bool tryParseRegisterOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
-                               StringRef Mnemonic);
+                               bool is64BitReg);
 
   bool needsExpansion(MCInst &Inst);
 
@@ -104,6 +122,9 @@ class MipsAsmParser : public MCTargetAsmParser {
                             SmallVectorImpl<MCInst> &Instructions);
   void expandLoadAddressReg(MCInst &Inst, SMLoc IDLoc,
                             SmallVectorImpl<MCInst> &Instructions);
+  void expandMemInst(MCInst &Inst, SMLoc IDLoc,
+                     SmallVectorImpl<MCInst> &Instructions,
+                     bool isLoad,bool isImmOpnd);
   bool reportParseError(StringRef ErrorMsg);
 
   bool parseMemOffset(const MCExpr *&Res);
@@ -118,6 +139,10 @@ class MipsAsmParser : public MCTargetAsmParser {
   bool parseSetReorderDirective();
   bool parseSetNoReorderDirective();
 
+  bool parseSetAssignment();
+
+  bool parseDirectiveWord(unsigned Size, SMLoc L);
+
   MCSymbolRefExpr::VariantKind getVariantKind(StringRef Symbol);
 
   bool isMips64() const {
@@ -128,9 +153,11 @@ class MipsAsmParser : public MCTargetAsmParser {
     return (STI.getFeatureBits() & Mips::FeatureFP64Bit) != 0;
   }
 
-  int matchRegisterName(StringRef Symbol);
+  int matchRegisterName(StringRef Symbol, bool is64BitReg);
 
-  int matchRegisterByNumber(unsigned RegNum, StringRef Mnemonic);
+  int matchCPURegisterName(StringRef Symbol);
+
+  int matchRegisterByNumber(unsigned RegNum, unsigned RegClass);
 
   void setFpFormat(FpFormatTy Format) {
     FpFormat = Format;
@@ -146,7 +173,10 @@ class MipsAsmParser : public MCTargetAsmParser {
 
   unsigned getReg(int RC,int RegNo);
 
-  unsigned getATReg();
+  int getATReg();
+
+  bool processInstruction(MCInst &Inst, SMLoc IDLoc,
+                        SmallVectorImpl<MCInst> &Instructions);
 public:
   MipsAsmParser(MCSubtargetInfo &sti, MCAsmParser &parser)
     : MCTargetAsmParser(), STI(sti), Parser(parser) {
@@ -166,6 +196,20 @@ namespace {
 /// instruction.
 class MipsOperand : public MCParsedAsmOperand {
 
+public:
+  enum RegisterKind {
+    Kind_None,
+    Kind_CPURegs,
+    Kind_CPU64Regs,
+    Kind_HWRegs,
+    Kind_HW64Regs,
+    Kind_FGR32Regs,
+    Kind_FGR64Regs,
+    Kind_AFGR64Regs,
+    Kind_CCRRegs
+  };
+
+private:
   enum KindTy {
     k_CondCode,
     k_CoprocNum,
@@ -178,24 +222,30 @@ class MipsOperand : public MCParsedAsmOperand {
 
   MipsOperand(KindTy K) : MCParsedAsmOperand(), Kind(K) {}
 
+  struct Token {
+    const char *Data;
+    unsigned Length;
+  };
+
+  struct RegOp {
+    unsigned RegNum;
+    RegisterKind Kind;
+  };
+
+  struct ImmOp {
+    const MCExpr *Val;
+  };
+
+  struct MemOp {
+    unsigned Base;
+    const MCExpr *Off;
+  };
+
   union {
-    struct {
-      const char *Data;
-      unsigned Length;
-    } Tok;
-
-    struct {
-      unsigned RegNum;
-    } Reg;
-
-    struct {
-      const MCExpr *Val;
-    } Imm;
-
-    struct {
-      unsigned Base;
-      const MCExpr *Off;
-    } Mem;
+    struct Token Tok;
+    struct RegOp Reg;
+    struct ImmOp Imm;
+    struct MemOp Mem;
   };
 
   SMLoc StartLoc, EndLoc;
@@ -246,6 +296,11 @@ public:
     return Reg.RegNum;
   }
 
+  void setRegKind(RegisterKind RegKind) {
+    assert((Kind == k_Register) && "Invalid access!");
+    Reg.Kind = RegKind;
+  }
+
   const MCExpr *getImm() const {
     assert((Kind == k_Immediate) && "Invalid access!");
     return Imm.Val;
@@ -296,6 +351,45 @@ public:
     return Op;
   }
 
+  bool isCPURegsAsm() const {
+    return Kind == k_Register && Reg.Kind == Kind_CPURegs;
+  }
+  void addCPURegsAsmOperands(MCInst &Inst, unsigned N) const {
+    Inst.addOperand(MCOperand::CreateReg(Reg.RegNum));
+  }
+
+  bool isCPU64RegsAsm() const {
+    return Kind == k_Register && Reg.Kind == Kind_CPU64Regs;
+  }
+  void addCPU64RegsAsmOperands(MCInst &Inst, unsigned N) const {
+    Inst.addOperand(MCOperand::CreateReg(Reg.RegNum));
+  }
+
+  bool isHWRegsAsm() const {
+    assert((Kind == k_Register) && "Invalid access!");
+    return Reg.Kind == Kind_HWRegs;
+  }
+  void addHWRegsAsmOperands(MCInst &Inst, unsigned N) const {
+    Inst.addOperand(MCOperand::CreateReg(Reg.RegNum));
+  }
+
+  bool isHW64RegsAsm() const {
+    assert((Kind == k_Register) && "Invalid access!");
+    return Reg.Kind == Kind_HW64Regs;
+  }
+  void addHW64RegsAsmOperands(MCInst &Inst, unsigned N) const {
+    Inst.addOperand(MCOperand::CreateReg(Reg.RegNum));
+  }
+
+  void addCCRAsmOperands(MCInst &Inst, unsigned N) const {
+    Inst.addOperand(MCOperand::CreateReg(Reg.RegNum));
+  }
+
+  bool isCCRAsm() const {
+    assert((Kind == k_Register) && "Invalid access!");
+    return Reg.Kind == Kind_CCRRegs;
+  }
+
   /// getStartLoc - Get the location of the first token of this operand.
   SMLoc getStartLoc() const { return StartLoc; }
   /// getEndLoc - Get the location of the last token of this operand.
@@ -307,6 +401,56 @@ public:
 };
 }
 
+namespace llvm {
+extern const MCInstrDesc MipsInsts[];
+}
+static const MCInstrDesc &getInstDesc(unsigned Opcode) {
+  return MipsInsts[Opcode];
+}
+
+bool MipsAsmParser::processInstruction(MCInst &Inst, SMLoc IDLoc,
+                        SmallVectorImpl<MCInst> &Instructions) {
+  const MCInstrDesc &MCID = getInstDesc(Inst.getOpcode());
+  Inst.setLoc(IDLoc);
+  if (MCID.mayLoad() || MCID.mayStore()) {
+    // Check the offset of memory operand, if it is a symbol
+    // reference or immediate we may have to expand instructions
+    for (unsigned i=0;i<MCID.getNumOperands();i++) {
+      const MCOperandInfo &OpInfo = MCID.OpInfo[i];
+      if ((OpInfo.OperandType == MCOI::OPERAND_MEMORY) ||
+          (OpInfo.OperandType == MCOI::OPERAND_UNKNOWN)) {
+        MCOperand &Op = Inst.getOperand(i);
+        if (Op.isImm()) {
+          int MemOffset = Op.getImm();
+          if (MemOffset < -32768 || MemOffset > 32767) {
+            // Offset can't exceed 16bit value
+            expandMemInst(Inst,IDLoc,Instructions,MCID.mayLoad(),true);
+            return false;
+          }
+        } else if (Op.isExpr()) {
+          const MCExpr *Expr = Op.getExpr();
+          if (Expr->getKind() == MCExpr::SymbolRef){
+            const MCSymbolRefExpr *SR =
+                    static_cast<const MCSymbolRefExpr*>(Expr);
+            if (SR->getKind() == MCSymbolRefExpr::VK_None) {
+              // Expand symbol
+              expandMemInst(Inst,IDLoc,Instructions,MCID.mayLoad(),false);
+              return false;
+            }
+          }
+        }
+      }
+    }
+  }
+
+  if (needsExpansion(Inst))
+    expandInstruction(Inst, IDLoc, Instructions);
+  else
+    Instructions.push_back(Inst);
+
+  return false;
+}
+
 bool MipsAsmParser::needsExpansion(MCInst &Inst) {
 
   switch(Inst.getOpcode()) {
@@ -344,31 +488,31 @@ void MipsAsmParser::expandLoadImm(MCInst &Inst, SMLoc IDLoc,
   if ( 0 <= ImmValue && ImmValue <= 65535) {
     // for 0 <= j <= 65535.
     // li d,j => ori d,$zero,j
-    tmpInst.setOpcode(isMips64() ? Mips::ORi64 : Mips::ORi);
+    tmpInst.setOpcode(Mips::ORi);
     tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg()));
     tmpInst.addOperand(
-              MCOperand::CreateReg(isMips64() ? Mips::ZERO_64 : Mips::ZERO));
+              MCOperand::CreateReg(Mips::ZERO));
     tmpInst.addOperand(MCOperand::CreateImm(ImmValue));
     Instructions.push_back(tmpInst);
   } else if ( ImmValue < 0 && ImmValue >= -32768) {
     // for -32768 <= j < 0.
     // li d,j => addiu d,$zero,j
-    tmpInst.setOpcode(Mips::ADDiu); //TODO:no ADDiu64 in td files?
+    tmpInst.setOpcode(Mips::ADDiu);
     tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg()));
     tmpInst.addOperand(
-              MCOperand::CreateReg(isMips64() ? Mips::ZERO_64 : Mips::ZERO));
+              MCOperand::CreateReg(Mips::ZERO));
     tmpInst.addOperand(MCOperand::CreateImm(ImmValue));
     Instructions.push_back(tmpInst);
   } else {
     // for any other value of j that is representable as a 32-bit integer.
     // li d,j => lui d,hi16(j)
     //           ori d,d,lo16(j)
-    tmpInst.setOpcode(isMips64() ? Mips::LUi64 : Mips::LUi);
+    tmpInst.setOpcode(Mips::LUi);
     tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateImm((ImmValue & 0xffff0000) >> 16));
     Instructions.push_back(tmpInst);
     tmpInst.clear();
-    tmpInst.setOpcode(isMips64() ? Mips::ORi64 : Mips::ORi);
+    tmpInst.setOpcode(Mips::ORi);
     tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateImm(ImmValue & 0xffff));
@@ -390,7 +534,7 @@ void MipsAsmParser::expandLoadAddressReg(MCInst &Inst, SMLoc IDLoc,
   if ( -32768 <= ImmValue && ImmValue <= 65535) {
     //for -32768 <= j <= 65535.
     //la d,j(s) => addiu d,s,j
-    tmpInst.setOpcode(Mips::ADDiu); //TODO:no ADDiu64 in td files?
+    tmpInst.setOpcode(Mips::ADDiu);
     tmpInst.addOperand(MCOperand::CreateReg(DstRegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateReg(SrcRegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateImm(ImmValue));
@@ -400,12 +544,12 @@ void MipsAsmParser::expandLoadAddressReg(MCInst &Inst, SMLoc IDLoc,
     //la d,j(s) => lui d,hi16(j)
     //             ori d,d,lo16(j)
     //             addu d,d,s
-    tmpInst.setOpcode(isMips64()?Mips::LUi64:Mips::LUi);
+    tmpInst.setOpcode(Mips::LUi);
     tmpInst.addOperand(MCOperand::CreateReg(DstRegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateImm((ImmValue & 0xffff0000) >> 16));
     Instructions.push_back(tmpInst);
     tmpInst.clear();
-    tmpInst.setOpcode(isMips64()?Mips::ORi64:Mips::ORi);
+    tmpInst.setOpcode(Mips::ORi);
     tmpInst.addOperand(MCOperand::CreateReg(DstRegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateReg(DstRegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateImm(ImmValue & 0xffff));
@@ -433,19 +577,19 @@ void MipsAsmParser::expandLoadAddressImm(MCInst &Inst, SMLoc IDLoc,
     tmpInst.setOpcode(Mips::ADDiu);
     tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg()));
     tmpInst.addOperand(
-              MCOperand::CreateReg(isMips64()?Mips::ZERO_64:Mips::ZERO));
+              MCOperand::CreateReg(Mips::ZERO));
     tmpInst.addOperand(MCOperand::CreateImm(ImmValue));
     Instructions.push_back(tmpInst);
   } else {
     //for any other value of j that is representable as a 32-bit integer.
     //la d,j => lui d,hi16(j)
     //          ori d,d,lo16(j)
-    tmpInst.setOpcode(isMips64()?Mips::LUi64:Mips::LUi);
+    tmpInst.setOpcode(Mips::LUi);
     tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateImm((ImmValue & 0xffff0000) >> 16));
     Instructions.push_back(tmpInst);
     tmpInst.clear();
-    tmpInst.setOpcode(isMips64()?Mips::ORi64:Mips::ORi);
+    tmpInst.setOpcode(Mips::ORi);
     tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg()));
     tmpInst.addOperand(MCOperand::CreateImm(ImmValue & 0xffff));
@@ -453,28 +597,103 @@ void MipsAsmParser::expandLoadAddressImm(MCInst &Inst, SMLoc IDLoc,
   }
 }
 
+void MipsAsmParser::expandMemInst(MCInst &Inst, SMLoc IDLoc,
+                     SmallVectorImpl<MCInst> &Instructions,
+                     bool isLoad,bool isImmOpnd) {
+  const MCSymbolRefExpr *SR;
+  MCInst TempInst;
+  unsigned ImmOffset,HiOffset,LoOffset;
+  const MCExpr *ExprOffset;
+  unsigned TmpRegNum;
+  unsigned AtRegNum = getReg((isMips64()) ? Mips::CPU64RegsRegClassID:
+                                            Mips::CPURegsRegClassID,
+                                            getATReg());
+  // 1st operand is either source or dst register
+  assert(Inst.getOperand(0).isReg() && "expected register operand kind");
+  unsigned RegOpNum = Inst.getOperand(0).getReg();
+  // 2nd operand is base register
+  assert(Inst.getOperand(1).isReg() && "expected register operand kind");
+  unsigned BaseRegNum = Inst.getOperand(1).getReg();
+  // 3rd operand is either immediate or expression
+  if (isImmOpnd) {
+    assert(Inst.getOperand(2).isImm() && "expected immediate operand kind");
+    ImmOffset = Inst.getOperand(2).getImm();
+    LoOffset = ImmOffset & 0x0000ffff;
+    HiOffset = (ImmOffset & 0xffff0000) >> 16;
+    // If msb of LoOffset is 1(negative number) we must increment HiOffset
+    if (LoOffset & 0x8000)
+      HiOffset++;
+  }
+  else
+    ExprOffset = Inst.getOperand(2).getExpr();
+  // All instructions will have the same location
+  TempInst.setLoc(IDLoc);
+  // 1st instruction in expansion is LUi. For load instruction we can use
+  // the dst register as a temporary if base and dst are different,
+  // but for stores we must use $at
+  TmpRegNum = (isLoad && (BaseRegNum != RegOpNum))?RegOpNum:AtRegNum;
+  TempInst.setOpcode(Mips::LUi);
+  TempInst.addOperand(MCOperand::CreateReg(TmpRegNum));
+  if (isImmOpnd)
+    TempInst.addOperand(MCOperand::CreateImm(HiOffset));
+  else {
+    if (ExprOffset->getKind() == MCExpr::SymbolRef) {
+      SR = static_cast<const MCSymbolRefExpr*>(ExprOffset);
+      const MCSymbolRefExpr *HiExpr = MCSymbolRefExpr::
+                                        Create(SR->getSymbol().getName(),
+                                        MCSymbolRefExpr::VK_Mips_ABS_HI,
+                                        getContext());
+      TempInst.addOperand(MCOperand::CreateExpr(HiExpr));
+    }
+  }
+  // Add the instruction to the list
+  Instructions.push_back(TempInst);
+  // and prepare TempInst for next instruction
+  TempInst.clear();
+  // which is add temp register to base
+  TempInst.setOpcode(Mips::ADDu);
+  TempInst.addOperand(MCOperand::CreateReg(TmpRegNum));
+  TempInst.addOperand(MCOperand::CreateReg(TmpRegNum));
+  TempInst.addOperand(MCOperand::CreateReg(BaseRegNum));
+  Instructions.push_back(TempInst);
+  TempInst.clear();
+  // and finaly, create original instruction with low part
+  // of offset and new base
+  TempInst.setOpcode(Inst.getOpcode());
+  TempInst.addOperand(MCOperand::CreateReg(RegOpNum));
+  TempInst.addOperand(MCOperand::CreateReg(TmpRegNum));
+  if (isImmOpnd)
+    TempInst.addOperand(MCOperand::CreateImm(LoOffset));
+  else {
+    if (ExprOffset->getKind() == MCExpr::SymbolRef) {
+      const MCSymbolRefExpr *LoExpr = MCSymbolRefExpr::
+                                      Create(SR->getSymbol().getName(),
+                                      MCSymbolRefExpr::VK_Mips_ABS_LO,
+                                      getContext());
+      TempInst.addOperand(MCOperand::CreateExpr(LoExpr));
+    }
+  }
+  Instructions.push_back(TempInst);
+  TempInst.clear();
+}
+
 bool MipsAsmParser::
 MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
                         SmallVectorImpl<MCParsedAsmOperand*> &Operands,
                         MCStreamer &Out, unsigned &ErrorInfo,
                         bool MatchingInlineAsm) {
   MCInst Inst;
+  SmallVector<MCInst, 8> Instructions;
   unsigned MatchResult = MatchInstructionImpl(Operands, Inst, ErrorInfo,
                                               MatchingInlineAsm);
 
   switch (MatchResult) {
   default: break;
   case Match_Success: {
-    if (needsExpansion(Inst)) {
-      SmallVector<MCInst, 4> Instructions;
-      expandInstruction(Inst, IDLoc, Instructions);
-      for(unsigned i =0; i < Instructions.size(); i++){
-        Out.EmitInstruction(Instructions[i]);
-      }
-    } else {
-        Inst.setLoc(IDLoc);
-        Out.EmitInstruction(Inst);
-      }
+    if (processInstruction(Inst,IDLoc,Instructions))
+      return true;
+    for(unsigned i =0; i < Instructions.size(); i++)
+      Out.EmitInstruction(Instructions[i]);
     return false;
   }
   case Match_MissingFeature:
@@ -498,84 +717,72 @@ MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
   return true;
 }
 
-int MipsAsmParser::matchRegisterName(StringRef Name) {
-
+int MipsAsmParser::matchCPURegisterName(StringRef Name) {
    int CC;
-   if (!isMips64())
+
+  if (Name == "at")
+    return getATReg();
+
     CC = StringSwitch<unsigned>(Name)
-      .Case("zero",  Mips::ZERO)
-      .Case("a0",  Mips::A0)
-      .Case("a1",  Mips::A1)
-      .Case("a2",  Mips::A2)
-      .Case("a3",  Mips::A3)
-      .Case("v0",  Mips::V0)
-      .Case("v1",  Mips::V1)
-      .Case("s0",  Mips::S0)
-      .Case("s1",  Mips::S1)
-      .Case("s2",  Mips::S2)
-      .Case("s3",  Mips::S3)
-      .Case("s4",  Mips::S4)
-      .Case("s5",  Mips::S5)
-      .Case("s6",  Mips::S6)
-      .Case("s7",  Mips::S7)
-      .Case("k0",  Mips::K0)
-      .Case("k1",  Mips::K1)
-      .Case("sp",  Mips::SP)
-      .Case("fp",  Mips::FP)
-      .Case("gp",  Mips::GP)
-      .Case("ra",  Mips::RA)
-      .Case("t0",  Mips::T0)
-      .Case("t1",  Mips::T1)
-      .Case("t2",  Mips::T2)
-      .Case("t3",  Mips::T3)
-      .Case("t4",  Mips::T4)
-      .Case("t5",  Mips::T5)
-      .Case("t6",  Mips::T6)
-      .Case("t7",  Mips::T7)
-      .Case("t8",  Mips::T8)
-      .Case("t9",  Mips::T9)
-      .Case("at",  Mips::AT)
-      .Case("fcc0",  Mips::FCC0)
-      .Default(-1);
-   else
+    .Case("zero", 0)
+    .Case("a0",   4)
+    .Case("a1",   5)
+    .Case("a2",   6)
+    .Case("a3",   7)
+    .Case("v0",   2)
+    .Case("v1",   3)
+    .Case("s0",  16)
+    .Case("s1",  17)
+    .Case("s2",  18)
+    .Case("s3",  19)
+    .Case("s4",  20)
+    .Case("s5",  21)
+    .Case("s6",  22)
+    .Case("s7",  23)
+    .Case("k0",  26)
+    .Case("k1",  27)
+    .Case("sp",  29)
+    .Case("fp",  30)
+    .Case("gp",  28)
+    .Case("ra",  31)
+    .Case("t0",   8)
+    .Case("t1",   9)
+    .Case("t2",  10)
+    .Case("t3",  11)
+    .Case("t4",  12)
+    .Case("t5",  13)
+    .Case("t6",  14)
+    .Case("t7",  15)
+    .Case("t8",  24)
+    .Case("t9",  25)
+    .Default(-1);
+
+  // Although SGI documentation just cut out t0-t3 for n32/n64,
+  // GNU pushes the values of t0-t3 to override the o32/o64 values for t4-t7
+  // We are supporting both cases, so for t0-t3 we'll just push them to t4-t7.
+  if (isMips64() && 8 <= CC  && CC <= 11)
+    CC += 4;
+
+  if (CC == -1 && isMips64())
     CC = StringSwitch<unsigned>(Name)
-      .Case("zero", Mips::ZERO_64)
-      .Case("at", Mips::AT_64)
-      .Case("v0", Mips::V0_64)
-      .Case("v1", Mips::V1_64)
-      .Case("a0", Mips::A0_64)
-      .Case("a1", Mips::A1_64)
-      .Case("a2", Mips::A2_64)
-      .Case("a3", Mips::A3_64)
-      .Case("a4", Mips::T0_64)
-      .Case("a5", Mips::T1_64)
-      .Case("a6", Mips::T2_64)
-      .Case("a7", Mips::T3_64)
-      .Case("t4", Mips::T4_64)
-      .Case("t5", Mips::T5_64)
-      .Case("t6", Mips::T6_64)
-      .Case("t7", Mips::T7_64)
-      .Case("s0", Mips::S0_64)
-      .Case("s1", Mips::S1_64)
-      .Case("s2", Mips::S2_64)
-      .Case("s3", Mips::S3_64)
-      .Case("s4", Mips::S4_64)
-      .Case("s5", Mips::S5_64)
-      .Case("s6", Mips::S6_64)
-      .Case("s7", Mips::S7_64)
-      .Case("t8", Mips::T8_64)
-      .Case("t9", Mips::T9_64)
-      .Case("kt0", Mips::K0_64)
-      .Case("kt1", Mips::K1_64)
-      .Case("gp", Mips::GP_64)
-      .Case("sp", Mips::SP_64)
-      .Case("fp", Mips::FP_64)
-      .Case("s8", Mips::FP_64)
-      .Case("ra", Mips::RA_64)
+      .Case("a4",   8)
+      .Case("a5",   9)
+      .Case("a6",  10)
+      .Case("a7",  11)
+      .Case("kt0", 26)
+      .Case("kt1", 27)
+      .Case("s8",  30)
       .Default(-1);
 
+  return CC;
+}
+int MipsAsmParser::matchRegisterName(StringRef Name, bool is64BitReg) {
+
+  int CC;
+  CC = matchCPURegisterName(Name);
   if (CC != -1)
-    return CC;
+    return matchRegisterByNumber(CC,is64BitReg?Mips::CPU64RegsRegClassID:
+                               Mips::CPURegsRegClassID);
 
   if (Name[0] == 'f') {
     StringRef NumString = Name.substr(1);
@@ -639,70 +846,44 @@ bool MipsAssemblerOptions::setATReg(unsigned Reg) {
   return true;
 }
 
-unsigned MipsAsmParser::getATReg() {
-  unsigned Reg = Options.getATRegNum();
-  if (isMips64())
-    return getReg(Mips::CPU64RegsRegClassID,Reg);
-  
-  return getReg(Mips::CPURegsRegClassID,Reg);
+int MipsAsmParser::getATReg() {
+  return Options.getATRegNum();
 }
 
 unsigned MipsAsmParser::getReg(int RC,int RegNo) {
   return *(getContext().getRegisterInfo().getRegClass(RC).begin() + RegNo);
 }
 
-int MipsAsmParser::matchRegisterByNumber(unsigned RegNum, StringRef Mnemonic) {
-
-  if (Mnemonic.lower() == "rdhwr") {
-    // at the moment only hwreg29 is supported
-    if (RegNum != 29)
-      return -1;
-    return Mips::HWR29;
-  }
+int MipsAsmParser::matchRegisterByNumber(unsigned RegNum, unsigned RegClass) {
 
   if (RegNum > 31)
     return -1;
 
-  // MIPS64 registers are numbered 1 after the 32-bit equivalents
-  return getReg(Mips::CPURegsRegClassID, RegNum) + isMips64();
+  return getReg(RegClass, RegNum);
 }
 
-int MipsAsmParser::tryParseRegister(StringRef Mnemonic) {
+int MipsAsmParser::tryParseRegister(bool is64BitReg) {
   const AsmToken &Tok = Parser.getTok();
   int RegNum = -1;
 
   if (Tok.is(AsmToken::Identifier)) {
     std::string lowerCase = Tok.getString().lower();
-    RegNum = matchRegisterName(lowerCase);
+    RegNum = matchRegisterName(lowerCase, is64BitReg);
   } else if (Tok.is(AsmToken::Integer))
     RegNum = matchRegisterByNumber(static_cast<unsigned>(Tok.getIntVal()),
-                                   Mnemonic.lower());
-    else
-      return RegNum;  //error
-  // 64 bit div operations require Mips::ZERO instead of MIPS::ZERO_64
-  if (isMips64() && RegNum == Mips::ZERO_64) {
-    if (Mnemonic.find("ddiv") != StringRef::npos)
-      RegNum = Mips::ZERO;
-  }
+                                   is64BitReg ? Mips::CPU64RegsRegClassID
+                                              : Mips::CPURegsRegClassID);
   return RegNum;
 }
 
 bool MipsAsmParser::
   tryParseRegisterOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
-                          StringRef Mnemonic){
+                          bool is64BitReg){
 
   SMLoc S = Parser.getTok().getLoc();
   int RegNo = -1;
 
-  // FIXME: we should make a more generic method for CCR
-  if ((Mnemonic == "cfc1" || Mnemonic == "ctc1")
-      && Operands.size() == 2 && Parser.getTok().is(AsmToken::Integer)){
-    RegNo = Parser.getTok().getIntVal();  // get the int value
-    // at the moment only fcc0 is supported
-    if (RegNo ==  0)
-      RegNo = Mips::FCC0;
-  } else
-    RegNo = tryParseRegister(Mnemonic);
+  RegNo = tryParseRegister(is64BitReg);
   if (RegNo == -1)
     return true;
 
@@ -734,7 +915,7 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands,
     SMLoc S = Parser.getTok().getLoc();
     Parser.Lex(); // Eat dollar token.
     // parse register operand
-    if (!tryParseRegisterOperand(Operands, Mnemonic)) {
+    if (!tryParseRegisterOperand(Operands, isMips64())) {
       if (getLexer().is(AsmToken::LParen)) {
         // check if it is indexed addressing operand
         Operands.push_back(MipsOperand::CreateToken("(", S));
@@ -743,7 +924,7 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands,
           return true;
 
         Parser.Lex(); // eat dollar
-        if (tryParseRegisterOperand(Operands, Mnemonic))
+        if (tryParseRegisterOperand(Operands, isMips64()))
           return true;
 
         if (!getLexer().is(AsmToken::RParen))
@@ -757,7 +938,7 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands,
     }
     // maybe it is a symbol reference
     StringRef Identifier;
-    if (Parser.ParseIdentifier(Identifier))
+    if (Parser.parseIdentifier(Identifier))
       return true;
 
     SMLoc E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1);
@@ -772,6 +953,11 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands,
     return false;
   }
   case AsmToken::Identifier:
+    // Look for the existing symbol, we should check if
+    // we need to assigne the propper RegisterKind
+   if (searchSymbolAlias(Operands,MipsOperand::Kind_None))
+     return false;
+    //else drop to expression parsing
   case AsmToken::LParen:
   case AsmToken::Minus:
   case AsmToken::Plus:
@@ -780,7 +966,7 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands,
      // quoted label names
     const MCExpr *IdVal;
     SMLoc S = Parser.getTok().getLoc();
-    if (getParser().ParseExpression(IdVal))
+    if (getParser().parseExpression(IdVal))
       return true;
     SMLoc E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1);
     Operands.push_back(MipsOperand::CreateImm(IdVal, S, E));
@@ -832,7 +1018,7 @@ bool MipsAsmParser::parseRelocOperand(const MCExpr *&Res) {
       } else
         break;
     }
-    if (getParser().ParseParenExpression(IdVal,EndLoc))
+    if (getParser().parseParenExpression(IdVal,EndLoc))
       return true;
 
     while (getLexer().getKind() == AsmToken::RParen)
@@ -843,19 +1029,25 @@ bool MipsAsmParser::parseRelocOperand(const MCExpr *&Res) {
 
   // Check the type of the expression
   if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(IdVal)) {
-    // it's a constant, evaluate lo or hi value
-    int Val = MCE->getValue();
+    // It's a constant, evaluate lo or hi value
     if (Str == "lo") {
-      Val = Val & 0xffff;
+      short Val = MCE->getValue();
+      Res = MCConstantExpr::Create(Val, getContext());
     } else if (Str == "hi") {
+      int Val = MCE->getValue();
+      int LoSign = Val & 0x8000;
       Val = (Val & 0xffff0000) >> 16;
+      // Lower part is treated as a signed int, so if it is negative
+      // we must add 1 to the hi part to compensate
+      if (LoSign)
+        Val++;
+      Res = MCConstantExpr::Create(Val, getContext());
     }
-    Res = MCConstantExpr::Create(Val, getContext());
     return false;
   }
 
   if (const MCSymbolRefExpr *MSRE = dyn_cast<MCSymbolRefExpr>(IdVal)) {
-    // it's a symbol, create symbolic expression from symbol
+    // It's a symbol, create symbolic expression from symbol
     StringRef Symbol = MSRE->getSymbol().getName();
     MCSymbolRefExpr::VariantKind VK = getVariantKind(Str);
     Res = MCSymbolRefExpr::Create(Symbol,VK,getContext());
@@ -868,7 +1060,7 @@ bool MipsAsmParser::ParseRegister(unsigned &RegNo, SMLoc &StartLoc,
                                   SMLoc &EndLoc) {
 
   StartLoc = Parser.getTok().getLoc();
-  RegNo = tryParseRegister("");
+  RegNo = tryParseRegister(isMips64());
   EndLoc = Parser.getTok().getLoc();
   return (RegNo == (unsigned)-1);
 }
@@ -880,10 +1072,11 @@ bool MipsAsmParser::parseMemOffset(const MCExpr *&Res) {
   switch(getLexer().getKind()) {
   default:
     return true;
+  case AsmToken::Identifier:
   case AsmToken::Integer:
   case AsmToken::Minus:
   case AsmToken::Plus:
-    return (getParser().ParseExpression(Res));
+    return (getParser().parseExpression(Res));
   case AsmToken::Percent:
     return parseRelocOperand(Res);
   case AsmToken::LParen:
@@ -907,7 +1100,7 @@ MipsAsmParser::OperandMatchResultTy MipsAsmParser::parseMemOperand(
   if (Tok.isNot(AsmToken::LParen)) {
     MipsOperand *Mnemonic = static_cast<MipsOperand*>(Operands[0]);
     if (Mnemonic->getToken() == "la") {
-      SMLoc E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer()-1);
+      SMLoc E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() -1);
       Operands.push_back(MipsOperand::CreateImm(IdVal, S, E));
       return MatchOperand_Success;
     }
@@ -920,7 +1113,7 @@ MipsAsmParser::OperandMatchResultTy MipsAsmParser::parseMemOperand(
   const AsmToken &Tok1 = Parser.getTok(); // get next token
   if (Tok1.is(AsmToken::Dollar)) {
     Parser.Lex(); // Eat '$' token.
-    if (tryParseRegisterOperand(Operands,"")) {
+    if (tryParseRegisterOperand(Operands, isMips64())) {
       Error(Parser.getTok().getLoc(), "unexpected token in operand");
       return MatchOperand_ParseFail;
     }
@@ -954,6 +1147,180 @@ MipsAsmParser::OperandMatchResultTy MipsAsmParser::parseMemOperand(
   return MatchOperand_Success;
 }
 
+MipsAsmParser::OperandMatchResultTy
+MipsAsmParser::parseCPU64Regs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+
+  if (!isMips64())
+    return MatchOperand_NoMatch;
+  if (getLexer().getKind() == AsmToken::Identifier) {
+    if (searchSymbolAlias(Operands,MipsOperand::Kind_CPU64Regs))
+      return MatchOperand_Success;
+    return MatchOperand_NoMatch;
+  }
+  // if the first token is not '$' we have an error
+  if (Parser.getTok().isNot(AsmToken::Dollar))
+    return MatchOperand_NoMatch;
+
+  Parser.Lex(); // Eat $
+  if(!tryParseRegisterOperand(Operands, true)) {
+    // set the proper register kind
+    MipsOperand* op = static_cast<MipsOperand*>(Operands.back());
+    op->setRegKind(MipsOperand::Kind_CPU64Regs);
+    return MatchOperand_Success;
+  }
+  return MatchOperand_NoMatch;
+}
+
+bool MipsAsmParser::
+searchSymbolAlias(SmallVectorImpl<MCParsedAsmOperand*> &Operands,
+                  unsigned RegisterKind) {
+
+  MCSymbol *Sym = getContext().LookupSymbol(Parser.getTok().getIdentifier());
+  if (Sym) {
+    SMLoc S = Parser.getTok().getLoc();
+    const MCExpr *Expr;
+    if (Sym->isVariable())
+      Expr = Sym->getVariableValue();
+    else
+      return false;
+    if (Expr->getKind() == MCExpr::SymbolRef) {
+      const MCSymbolRefExpr *Ref = static_cast<const MCSymbolRefExpr*>(Expr);
+      const StringRef DefSymbol = Ref->getSymbol().getName();
+      if (DefSymbol.startswith("$")) {
+        // Lookup for the register with corresponding name
+        int RegNum = matchRegisterName(DefSymbol.substr(1),isMips64());
+        if (RegNum > -1) {
+          Parser.Lex();
+          MipsOperand *op = MipsOperand::CreateReg(RegNum,S,
+                                         Parser.getTok().getLoc());
+          op->setRegKind((MipsOperand::RegisterKind)RegisterKind);
+          Operands.push_back(op);
+          return true;
+        }
+      }
+    } else if (Expr->getKind() == MCExpr::Constant) {
+      Parser.Lex();
+      const MCConstantExpr *Const = static_cast<const MCConstantExpr*>(Expr);
+      MipsOperand *op = MipsOperand::CreateImm(Const,S,
+                                     Parser.getTok().getLoc());
+      Operands.push_back(op);
+      return true;
+    }
+  }
+  return false;
+}
+MipsAsmParser::OperandMatchResultTy
+MipsAsmParser::parseCPURegs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+
+  if (getLexer().getKind() == AsmToken::Identifier) {
+    if (searchSymbolAlias(Operands,MipsOperand::Kind_CPURegs))
+      return MatchOperand_Success;
+    return MatchOperand_NoMatch;
+  }
+  // if the first token is not '$' we have an error
+  if (Parser.getTok().isNot(AsmToken::Dollar))
+    return MatchOperand_NoMatch;
+
+  Parser.Lex(); // Eat $
+  if(!tryParseRegisterOperand(Operands, false)) {
+    // set the propper register kind
+    MipsOperand* op = static_cast<MipsOperand*>(Operands.back());
+    op->setRegKind(MipsOperand::Kind_CPURegs);
+    return MatchOperand_Success;
+  }
+  return MatchOperand_NoMatch;
+}
+
+MipsAsmParser::OperandMatchResultTy
+MipsAsmParser::parseHWRegs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+
+  if (isMips64())
+    return MatchOperand_NoMatch;
+
+  // if the first token is not '$' we have error
+  if (Parser.getTok().isNot(AsmToken::Dollar))
+    return MatchOperand_NoMatch;
+  SMLoc S = Parser.getTok().getLoc();
+  Parser.Lex(); // Eat $
+
+  const AsmToken &Tok = Parser.getTok(); // get next token
+  if (Tok.isNot(AsmToken::Integer))
+    return MatchOperand_NoMatch;
+
+  unsigned RegNum = Tok.getIntVal();
+  // at the moment only hwreg29 is supported
+  if (RegNum != 29)
+    return MatchOperand_ParseFail;
+
+  MipsOperand *op = MipsOperand::CreateReg(Mips::HWR29, S,
+        Parser.getTok().getLoc());
+  op->setRegKind(MipsOperand::Kind_HWRegs);
+  Operands.push_back(op);
+
+  Parser.Lex(); // Eat reg number
+  return MatchOperand_Success;
+}
+
+MipsAsmParser::OperandMatchResultTy
+MipsAsmParser::parseHW64Regs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+
+  if (!isMips64())
+    return MatchOperand_NoMatch;
+    //if the first token is not '$' we have error
+  if (Parser.getTok().isNot(AsmToken::Dollar))
+    return MatchOperand_NoMatch;
+  SMLoc S = Parser.getTok().getLoc();
+  Parser.Lex(); // Eat $
+
+  const AsmToken &Tok = Parser.getTok(); // get next token
+  if (Tok.isNot(AsmToken::Integer))
+    return MatchOperand_NoMatch;
+
+  unsigned RegNum = Tok.getIntVal();
+  // at the moment only hwreg29 is supported
+  if (RegNum != 29)
+    return MatchOperand_ParseFail;
+
+  MipsOperand *op = MipsOperand::CreateReg(Mips::HWR29_64, S,
+        Parser.getTok().getLoc());
+  op->setRegKind(MipsOperand::Kind_HW64Regs);
+  Operands.push_back(op);
+
+  Parser.Lex(); // Eat reg number
+  return MatchOperand_Success;
+}
+
+MipsAsmParser::OperandMatchResultTy
+MipsAsmParser::parseCCRRegs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  unsigned RegNum;
+  //if the first token is not '$' we have error
+  if (Parser.getTok().isNot(AsmToken::Dollar))
+    return MatchOperand_NoMatch;
+  SMLoc S = Parser.getTok().getLoc();
+  Parser.Lex(); // Eat $
+
+  const AsmToken &Tok = Parser.getTok(); // get next token
+  if (Tok.is(AsmToken::Integer)) {
+    RegNum = Tok.getIntVal();
+    // at the moment only fcc0 is supported
+    if (RegNum != 0)
+      return MatchOperand_ParseFail;
+  } else if (Tok.is(AsmToken::Identifier)) {
+    // at the moment only fcc0 is supported
+    if (Tok.getIdentifier() != "fcc0")
+      return MatchOperand_ParseFail;
+  } else
+    return MatchOperand_NoMatch;
+
+  MipsOperand *op = MipsOperand::CreateReg(Mips::FCC0, S,
+        Parser.getTok().getLoc());
+  op->setRegKind(MipsOperand::Kind_CCRRegs);
+  Operands.push_back(op);
+
+  Parser.Lex(); // Eat reg number
+  return MatchOperand_Success;
+}
+
 MCSymbolRefExpr::VariantKind MipsAsmParser::getVariantKind(StringRef Symbol) {
 
   MCSymbolRefExpr::VariantKind VK
@@ -1023,13 +1390,13 @@ parseMathOperation(StringRef Name, SMLoc NameLoc,
     // Read the first operand.
     if (ParseOperand(Operands, Name)) {
       SMLoc Loc = getLexer().getLoc();
-      Parser.EatToEndOfStatement();
+      Parser.eatToEndOfStatement();
       return Error(Loc, "unexpected token in argument list");
     }
 
     if (getLexer().isNot(AsmToken::Comma)) {
       SMLoc Loc = getLexer().getLoc();
-      Parser.EatToEndOfStatement();
+      Parser.eatToEndOfStatement();
       return Error(Loc, "unexpected token in argument list");
 
     }
@@ -1041,14 +1408,14 @@ parseMathOperation(StringRef Name, SMLoc NameLoc,
     // Parse and remember the operand.
     if (ParseOperand(Operands, Name)) {
       SMLoc Loc = getLexer().getLoc();
-      Parser.EatToEndOfStatement();
+      Parser.eatToEndOfStatement();
       return Error(Loc, "unexpected token in argument list");
     }
   }
 
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     SMLoc Loc = getLexer().getLoc();
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return Error(Loc, "unexpected token in argument list");
   }
 
@@ -1059,16 +1426,18 @@ parseMathOperation(StringRef Name, SMLoc NameLoc,
 bool MipsAsmParser::
 ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc,
                  SmallVectorImpl<MCParsedAsmOperand*> &Operands) {
+  StringRef Mnemonic;
   // floating point instructions: should register be treated as double?
   if (requestsDoubleOperand(Name)) {
     setFpFormat(FP_FORMAT_D);
   Operands.push_back(MipsOperand::CreateToken(Name, NameLoc));
+  Mnemonic = Name;
   }
   else {
     setDefaultFpFormat();
     // Create the leading tokens for the mnemonic, split by '.' characters.
     size_t Start = 0, Next = Name.find('.');
-    StringRef Mnemonic = Name.slice(Start, Next);
+    Mnemonic = Name.slice(Start, Next);
 
     Operands.push_back(MipsOperand::CreateToken(Mnemonic, NameLoc));
 
@@ -1108,9 +1477,9 @@ ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc,
   // Read the remaining operands.
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     // Read the first operand.
-    if (ParseOperand(Operands, Name)) {
+    if (ParseOperand(Operands, Mnemonic)) {
       SMLoc Loc = getLexer().getLoc();
-      Parser.EatToEndOfStatement();
+      Parser.eatToEndOfStatement();
       return Error(Loc, "unexpected token in argument list");
     }
 
@@ -1120,7 +1489,7 @@ ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc,
       // Parse and remember the operand.
       if (ParseOperand(Operands, Name)) {
         SMLoc Loc = getLexer().getLoc();
-        Parser.EatToEndOfStatement();
+        Parser.eatToEndOfStatement();
         return Error(Loc, "unexpected token in argument list");
       }
     }
@@ -1128,7 +1497,7 @@ ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc,
 
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     SMLoc Loc = getLexer().getLoc();
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return Error(Loc, "unexpected token in argument list");
   }
 
@@ -1138,18 +1507,18 @@ ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc,
 
 bool MipsAsmParser::reportParseError(StringRef ErrorMsg) {
    SMLoc Loc = getLexer().getLoc();
-   Parser.EatToEndOfStatement();
+   Parser.eatToEndOfStatement();
    return Error(Loc, ErrorMsg);
 }
 
 bool MipsAsmParser::parseSetNoAtDirective() {
-  // line should look like:
+  // Line should look like:
   //  .set noat
   // set at reg to 0
   Options.setATReg(0);
   // eat noat
   Parser.Lex();
-  // if this is not the end of the statement, report error
+  // If this is not the end of the statement, report error
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     reportParseError("unexpected token in statement");
     return false;
@@ -1161,28 +1530,39 @@ bool MipsAsmParser::parseSetAtDirective() {
   // line can be
   //  .set at - defaults to $1
   // or .set at=$reg
+  int AtRegNo;
   getParser().Lex();
   if (getLexer().is(AsmToken::EndOfStatement)) {
     Options.setATReg(1);
     Parser.Lex(); // Consume the EndOfStatement
     return false;
   } else if (getLexer().is(AsmToken::Equal)) {
-    getParser().Lex(); //eat '='
+    getParser().Lex(); // eat '='
     if (getLexer().isNot(AsmToken::Dollar)) {
       reportParseError("unexpected token in statement");
       return false;
     }
-    Parser.Lex(); // eat '$'
-    if (getLexer().isNot(AsmToken::Integer)) {
+    Parser.Lex(); // Eat '$'
+    const AsmToken &Reg = Parser.getTok();
+    if (Reg.is(AsmToken::Identifier)) {
+      AtRegNo = matchCPURegisterName(Reg.getIdentifier());
+    } else if (Reg.is(AsmToken::Integer)) {
+      AtRegNo = Reg.getIntVal();
+    } else {
       reportParseError("unexpected token in statement");
       return false;
     }
-    const AsmToken &Reg = Parser.getTok();
-    if (!Options.setATReg(Reg.getIntVal())) {
+
+    if ( AtRegNo < 1 || AtRegNo > 31) {
+      reportParseError("unexpected token in statement");
+      return false;
+    }
+
+    if (!Options.setATReg(AtRegNo)) {
       reportParseError("unexpected token in statement");
       return false;
     }
-    getParser().Lex(); //eat reg
+    getParser().Lex(); // Eat reg
 
     if (getLexer().isNot(AsmToken::EndOfStatement)) {
       reportParseError("unexpected token in statement");
@@ -1198,7 +1578,7 @@ bool MipsAsmParser::parseSetAtDirective() {
 
 bool MipsAsmParser::parseSetReorderDirective() {
   Parser.Lex();
-  // if this is not the end of the statement, report error
+  // If this is not the end of the statement, report error
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     reportParseError("unexpected token in statement");
     return false;
@@ -1247,6 +1627,31 @@ bool MipsAsmParser::parseSetNoMacroDirective() {
   Parser.Lex(); // Consume the EndOfStatement
   return false;
 }
+
+bool MipsAsmParser::parseSetAssignment() {
+  StringRef Name;
+  const MCExpr *Value;
+
+  if (Parser.parseIdentifier(Name))
+    reportParseError("expected identifier after .set");
+
+  if (getLexer().isNot(AsmToken::Comma))
+    return reportParseError("unexpected token in .set directive");
+  Lex(); //eat comma
+
+  if (Parser.parseExpression(Value))
+    reportParseError("expected valid expression after comma");
+
+  // check if the Name already exists as a symbol
+  MCSymbol *Sym = getContext().LookupSymbol(Name);
+  if (Sym) {
+    return reportParseError("symbol already defined");
+  }
+  Sym = getContext().GetOrCreateSymbol(Name);
+  Sym->setVariableValue(Value);
+
+  return false;
+}
 bool MipsAsmParser::parseDirectiveSet() {
 
   // get next token
@@ -1266,55 +1671,92 @@ bool MipsAsmParser::parseDirectiveSet() {
     return parseSetNoMacroDirective();
   } else if (Tok.getString() == "nomips16") {
     // ignore this directive for now
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return false;
   } else if (Tok.getString() == "nomicromips") {
     // ignore this directive for now
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
+    return false;
+  } else {
+    // it is just an identifier, look for assignment
+    parseSetAssignment();
     return false;
   }
+
   return true;
 }
 
+/// parseDirectiveWord
+///  ::= .word [ expression (, expression)* ]
+bool MipsAsmParser::parseDirectiveWord(unsigned Size, SMLoc L) {
+  if (getLexer().isNot(AsmToken::EndOfStatement)) {
+    for (;;) {
+      const MCExpr *Value;
+      if (getParser().parseExpression(Value))
+        return true;
+
+      getParser().getStreamer().EmitValue(Value, Size);
+
+      if (getLexer().is(AsmToken::EndOfStatement))
+        break;
+
+      // FIXME: Improve diagnostic.
+      if (getLexer().isNot(AsmToken::Comma))
+        return Error(L, "unexpected token in directive");
+      Parser.Lex();
+    }
+  }
+
+  Parser.Lex();
+  return false;
+}
+
 bool MipsAsmParser::ParseDirective(AsmToken DirectiveID) {
 
-  if (DirectiveID.getString() == ".ent") {
+  StringRef IDVal = DirectiveID.getString();
+
+  if ( IDVal == ".ent") {
     // ignore this directive for now
     Parser.Lex();
     return false;
   }
 
-  if (DirectiveID.getString() == ".end") {
+  if (IDVal == ".end") {
     // ignore this directive for now
     Parser.Lex();
     return false;
   }
 
-  if (DirectiveID.getString() == ".frame") {
+  if (IDVal == ".frame") {
     // ignore this directive for now
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return false;
   }
 
-  if (DirectiveID.getString() == ".set") {
+  if (IDVal == ".set") {
     return parseDirectiveSet();
   }
 
-  if (DirectiveID.getString() == ".fmask") {
+  if (IDVal == ".fmask") {
     // ignore this directive for now
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return false;
   }
 
-  if (DirectiveID.getString() == ".mask") {
+  if (IDVal == ".mask") {
     // ignore this directive for now
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
     return false;
   }
 
-  if (DirectiveID.getString() == ".gpword") {
+  if (IDVal == ".gpword") {
     // ignore this directive for now
-    Parser.EatToEndOfStatement();
+    Parser.eatToEndOfStatement();
+    return false;
+  }
+
+  if (IDVal == ".word") {
+    parseDirectiveWord(4, DirectiveID.getLoc());
     return false;
   }
 
diff --git a/lib/Target/Mips/CMakeLists.txt b/lib/Target/Mips/CMakeLists.txt
index ef56e752b2e4..cf8bb189e475 100644
--- a/lib/Target/Mips/CMakeLists.txt
+++ b/lib/Target/Mips/CMakeLists.txt
@@ -9,7 +9,6 @@ tablegen(LLVM MipsGenAsmWriter.inc -gen-asm-writer)
 tablegen(LLVM MipsGenDAGISel.inc -gen-dag-isel)
 tablegen(LLVM MipsGenCallingConv.inc -gen-callingconv)
 tablegen(LLVM MipsGenSubtargetInfo.inc -gen-subtarget)
-tablegen(LLVM MipsGenEDInfo.inc -gen-enhanced-disassembly-info)
 tablegen(LLVM MipsGenAsmMatcher.inc -gen-asm-matcher)
 tablegen(LLVM MipsGenMCPseudoLowering.inc -gen-pseudo-lowering)
 add_public_tablegen_target(MipsCommonTableGen)
@@ -17,10 +16,13 @@ add_public_tablegen_target(MipsCommonTableGen)
 add_llvm_target(MipsCodeGen
   Mips16FrameLowering.cpp
   Mips16InstrInfo.cpp
+  Mips16ISelDAGToDAG.cpp
+  Mips16ISelLowering.cpp
   Mips16RegisterInfo.cpp
   MipsAnalyzeImmediate.cpp
   MipsAsmPrinter.cpp
   MipsCodeEmitter.cpp
+  MipsConstantIslandPass.cpp
   MipsDelaySlotFiller.cpp
   MipsJITInfo.cpp
   MipsInstrInfo.cpp
@@ -33,6 +35,8 @@ add_llvm_target(MipsCodeGen
   MipsRegisterInfo.cpp
   MipsSEFrameLowering.cpp
   MipsSEInstrInfo.cpp
+  MipsSEISelDAGToDAG.cpp
+  MipsSEISelLowering.cpp
   MipsSERegisterInfo.cpp
   MipsSubtarget.cpp
   MipsTargetMachine.cpp
diff --git a/lib/Target/Mips/Disassembler/LLVMBuild.txt b/lib/Target/Mips/Disassembler/LLVMBuild.txt
index 048ad0ddac5b..7101c06d12ac 100644
--- a/lib/Target/Mips/Disassembler/LLVMBuild.txt
+++ b/lib/Target/Mips/Disassembler/LLVMBuild.txt
@@ -1,4 +1,4 @@
-;===- ./lib/Target/Mips/Disassembler/LLVMBuild.txt --------------*- Conf -*--===;
+;===- ./lib/Target/Mips/Disassembler/LLVMBuild.txt -------------*- Conf -*--===;
 ;
 ;                     The LLVM Compiler Infrastructure
 ;
diff --git a/lib/Target/Mips/Disassembler/Makefile b/lib/Target/Mips/Disassembler/Makefile
index a78feba1f8df..7900373dd2b2 100644
--- a/lib/Target/Mips/Disassembler/Makefile
+++ b/lib/Target/Mips/Disassembler/Makefile
@@ -1,4 +1,4 @@
-##===- lib/Target/Mips/Disassembler/Makefile ----------------*- Makefile -*-===##
+##===- lib/Target/Mips/Disassembler/Makefile ---------------*- Makefile -*-===##
 #
 #                     The LLVM Compiler Infrastructure
 #
diff --git a/lib/Target/Mips/Disassembler/MipsDisassembler.cpp b/lib/Target/Mips/Disassembler/MipsDisassembler.cpp
index 82dbcc5bcf7d..59e49d8ddc6c 100644
--- a/lib/Target/Mips/Disassembler/MipsDisassembler.cpp
+++ b/lib/Target/Mips/Disassembler/MipsDisassembler.cpp
@@ -12,18 +12,15 @@
 //===----------------------------------------------------------------------===//
 
 #include "Mips.h"
-#include "MipsSubtarget.h"
 #include "MipsRegisterInfo.h"
-#include "llvm/MC/EDInstInfo.h"
+#include "MipsSubtarget.h"
 #include "llvm/MC/MCDisassembler.h"
 #include "llvm/MC/MCFixedLenDisassembler.h"
-#include "llvm/Support/MemoryObject.h"
-#include "llvm/Support/TargetRegistry.h"
-#include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/Support/MathExtras.h"
-
-#include "MipsGenEDInfo.inc"
+#include "llvm/Support/MemoryObject.h"
+#include "llvm/Support/TargetRegistry.h"
 
 using namespace llvm;
 
@@ -42,9 +39,6 @@ public:
 
   virtual ~MipsDisassemblerBase() {}
 
-  /// getEDInfo - See MCDisassembler.
-  const EDInstInfo *getEDInfo() const;
-
   const MCRegisterInfo *getRegInfo() const { return RegInfo; }
 
 private:
@@ -92,10 +86,6 @@ public:
 
 } // end anonymous namespace
 
-const EDInstInfo *MipsDisassemblerBase::getEDInfo() const {
-  return instInfoMips;
-}
-
 // Forward declare these because the autogenerated code will reference them.
 // Definitions are further down.
 static DecodeStatus DecodeCPU64RegsRegisterClass(MCInst &Inst,
@@ -103,6 +93,11 @@ static DecodeStatus DecodeCPU64RegsRegisterClass(MCInst &Inst,
                                                  uint64_t Address,
                                                  const void *Decoder);
 
+static DecodeStatus DecodeCPU16RegsRegisterClass(MCInst &Inst,
+                                                 unsigned RegNo,
+                                                 uint64_t Address,
+                                                 const void *Decoder);
+
 static DecodeStatus DecodeCPURegsRegisterClass(MCInst &Inst,
                                                unsigned RegNo,
                                                uint64_t Address,
@@ -143,10 +138,10 @@ static DecodeStatus DecodeHWRegs64RegisterClass(MCInst &Inst,
                                                 uint64_t Address,
                                                 const void *Decoder);
 
-static DecodeStatus DecodeACRegsRegisterClass(MCInst &Inst,
-                                              unsigned RegNo,
-                                              uint64_t Address,
-                                              const void *Decoder);
+static DecodeStatus DecodeACRegsDSPRegisterClass(MCInst &Inst,
+                                                 unsigned RegNo,
+                                                 uint64_t Address,
+                                                 const void *Decoder);
 
 static DecodeStatus DecodeBranchTarget(MCInst &Inst,
                                        unsigned Offset,
@@ -332,6 +327,15 @@ static unsigned getReg(const void *D, unsigned RC, unsigned RegNo) {
   return *(Dis->getRegInfo()->getRegClass(RC).begin() + RegNo);
 }
 
+static DecodeStatus DecodeCPU16RegsRegisterClass(MCInst &Inst,
+                                                 unsigned RegNo,
+                                                 uint64_t Address,
+                                                 const void *Decoder) {
+
+  return MCDisassembler::Fail;
+
+}
+
 static DecodeStatus DecodeCPU64RegsRegisterClass(MCInst &Inst,
                                                  unsigned RegNo,
                                                  uint64_t Address,
@@ -480,14 +484,14 @@ static DecodeStatus DecodeHWRegs64RegisterClass(MCInst &Inst,
   return MCDisassembler::Success;
 }
 
-static DecodeStatus DecodeACRegsRegisterClass(MCInst &Inst,
-                                              unsigned RegNo,
-                                              uint64_t Address,
-                                              const void *Decoder) {
+static DecodeStatus DecodeACRegsDSPRegisterClass(MCInst &Inst,
+                                                 unsigned RegNo,
+                                                 uint64_t Address,
+                                                 const void *Decoder) {
   if (RegNo >= 4)
     return MCDisassembler::Fail;
 
-  unsigned Reg = getReg(Decoder, Mips::ACRegsRegClassID, RegNo);
+  unsigned Reg = getReg(Decoder, Mips::ACRegsDSPRegClassID, RegNo);
   Inst.addOperand(MCOperand::CreateReg(Reg));
   return MCDisassembler::Success;
 }
diff --git a/lib/Target/Mips/InstPrinter/MipsInstPrinter.cpp b/lib/Target/Mips/InstPrinter/MipsInstPrinter.cpp
index 68d3ac5f3bd0..fc23cd380352 100644
--- a/lib/Target/Mips/InstPrinter/MipsInstPrinter.cpp
+++ b/lib/Target/Mips/InstPrinter/MipsInstPrinter.cpp
@@ -23,6 +23,7 @@
 #include "llvm/Support/raw_ostream.h"
 using namespace llvm;
 
+#define PRINT_ALIAS_INSTR
 #include "MipsGenAsmWriter.inc"
 
 const char* Mips::MipsFCCToString(Mips::CondCode CC) {
@@ -78,7 +79,9 @@ void MipsInstPrinter::printInst(const MCInst *MI, raw_ostream &O,
     O << "\t.set\tmips32r2\n";
   }
 
-  printInstruction(MI, O);
+  // Try to print any aliases first.
+  if (!printAliasInstr(MI, O))
+    printInstruction(MI, O);
   printAnnotation(O, Annot);
 
   switch (MI->getOpcode()) {
@@ -149,6 +152,11 @@ static void printExpr(const MCExpr *Expr, raw_ostream &OS) {
     OS << ')';
 }
 
+void MipsInstPrinter::printCPURegs(const MCInst *MI, unsigned OpNo,
+                                   raw_ostream &O) {
+  printRegName(O, MI->getOperand(OpNo).getReg());
+}
+
 void MipsInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
                                    raw_ostream &O) {
   const MCOperand &Op = MI->getOperand(OpNo);
diff --git a/lib/Target/Mips/InstPrinter/MipsInstPrinter.h b/lib/Target/Mips/InstPrinter/MipsInstPrinter.h
index 3d8a6f918ff6..d1b561f9764e 100644
--- a/lib/Target/Mips/InstPrinter/MipsInstPrinter.h
+++ b/lib/Target/Mips/InstPrinter/MipsInstPrinter.h
@@ -87,6 +87,9 @@ public:
 
   virtual void printRegName(raw_ostream &OS, unsigned RegNo) const;
   virtual void printInst(const MCInst *MI, raw_ostream &O, StringRef Annot);
+  void printCPURegs(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+
+  bool printAliasInstr(const MCInst *MI, raw_ostream &OS);
 
 private:
   void printOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
diff --git a/lib/Target/Mips/MCTargetDesc/CMakeLists.txt b/lib/Target/Mips/MCTargetDesc/CMakeLists.txt
index be5d7e42532a..4212c94a5578 100644
--- a/lib/Target/Mips/MCTargetDesc/CMakeLists.txt
+++ b/lib/Target/Mips/MCTargetDesc/CMakeLists.txt
@@ -5,6 +5,8 @@ add_llvm_library(LLVMMipsDesc
   MipsMCCodeEmitter.cpp
   MipsMCTargetDesc.cpp
   MipsELFObjectWriter.cpp
+  MipsReginfo.cpp
+  MipsELFStreamer.cpp
   )
 
 add_dependencies(LLVMMipsDesc MipsCommonTableGen)
diff --git a/lib/Target/Mips/MCTargetDesc/MipsAsmBackend.cpp b/lib/Target/Mips/MCTargetDesc/MipsAsmBackend.cpp
index c078794899d2..0b13607a572d 100644
--- a/lib/Target/Mips/MCTargetDesc/MipsAsmBackend.cpp
+++ b/lib/Target/Mips/MCTargetDesc/MipsAsmBackend.cpp
@@ -37,6 +37,7 @@ static unsigned adjustFixupValue(unsigned Kind, uint64_t Value) {
   case FK_Data_4:
   case FK_Data_8:
   case Mips::fixup_Mips_LO16:
+  case Mips::fixup_Mips_GPREL16:
   case Mips::fixup_Mips_GPOFF_HI:
   case Mips::fixup_Mips_GPOFF_LO:
   case Mips::fixup_Mips_GOT_PAGE:
@@ -213,7 +214,7 @@ public:
   /// fixup requires the associated instruction to be relaxed.
   bool fixupNeedsRelaxation(const MCFixup &Fixup,
                             uint64_t Value,
-                            const MCInstFragment *DF,
+                            const MCRelaxableFragment *DF,
                             const MCAsmLayout &Layout) const {
     // FIXME.
     assert(0 && "RelaxInstruction() unimplemented");
diff --git a/lib/Target/Mips/MCTargetDesc/MipsBaseInfo.h b/lib/Target/Mips/MCTargetDesc/MipsBaseInfo.h
index 94e0d20d8835..7a55efd5c330 100644
--- a/lib/Target/Mips/MCTargetDesc/MipsBaseInfo.h
+++ b/lib/Target/Mips/MCTargetDesc/MipsBaseInfo.h
@@ -121,99 +121,6 @@ namespace MipsII {
   };
 }
 
-
-/// getMipsRegisterNumbering - Given the enum value for some register,
-/// return the number that it corresponds to.
-inline static unsigned getMipsRegisterNumbering(unsigned RegEnum)
-{
-  switch (RegEnum) {
-  case Mips::ZERO: case Mips::ZERO_64: case Mips::F0: case Mips::D0_64:
-  case Mips::D0:   case Mips::FCC0:    case Mips::AC0:
-    return 0;
-  case Mips::AT: case Mips::AT_64: case Mips::F1: case Mips::D1_64:
-  case Mips::AC1:
-    return 1;
-  case Mips::V0: case Mips::V0_64: case Mips::F2: case Mips::D2_64:
-  case Mips::D1: case Mips::AC2:
-    return 2;
-  case Mips::V1: case Mips::V1_64: case Mips::F3: case Mips::D3_64:
-  case Mips::AC3:
-    return 3;
-  case Mips::A0: case Mips::A0_64: case Mips::F4: case Mips::D4_64:
-  case Mips::D2:
-    return 4;
-  case Mips::A1: case Mips::A1_64: case Mips::F5: case Mips::D5_64:
-    return 5;
-  case Mips::A2: case Mips::A2_64: case Mips::F6: case Mips::D6_64:
-  case Mips::D3:
-    return 6;
-  case Mips::A3: case Mips::A3_64: case Mips::F7: case Mips::D7_64:
-    return 7;
-  case Mips::T0: case Mips::T0_64: case Mips::F8: case Mips::D8_64:
-  case Mips::D4:
-    return 8;
-  case Mips::T1: case Mips::T1_64: case Mips::F9: case Mips::D9_64:
-    return 9;
-  case Mips::T2: case Mips::T2_64: case Mips::F10: case Mips::D10_64:
-  case Mips::D5:
-    return 10;
-  case Mips::T3: case Mips::T3_64: case Mips::F11: case Mips::D11_64:
-    return 11;
-  case Mips::T4: case Mips::T4_64: case Mips::F12: case Mips::D12_64:
-  case Mips::D6:
-    return 12;
-  case Mips::T5: case Mips::T5_64: case Mips::F13: case Mips::D13_64:
-    return 13;
-  case Mips::T6: case Mips::T6_64: case Mips::F14: case Mips::D14_64:
-  case Mips::D7:
-    return 14;
-  case Mips::T7: case Mips::T7_64: case Mips::F15: case Mips::D15_64:
-    return 15;
-  case Mips::S0: case Mips::S0_64: case Mips::F16: case Mips::D16_64:
-  case Mips::D8:
-    return 16;
-  case Mips::S1: case Mips::S1_64: case Mips::F17: case Mips::D17_64:
-    return 17;
-  case Mips::S2: case Mips::S2_64: case Mips::F18: case Mips::D18_64:
-  case Mips::D9:
-    return 18;
-  case Mips::S3: case Mips::S3_64: case Mips::F19: case Mips::D19_64:
-    return 19;
-  case Mips::S4: case Mips::S4_64: case Mips::F20: case Mips::D20_64:
-  case Mips::D10:
-    return 20;
-  case Mips::S5: case Mips::S5_64: case Mips::F21: case Mips::D21_64:
-    return 21;
-  case Mips::S6: case Mips::S6_64: case Mips::F22: case Mips::D22_64:
-  case Mips::D11:
-    return 22;
-  case Mips::S7: case Mips::S7_64: case Mips::F23: case Mips::D23_64:
-    return 23;
-  case Mips::T8: case Mips::T8_64: case Mips::F24: case Mips::D24_64:
-  case Mips::D12:
-    return 24;
-  case Mips::T9: case Mips::T9_64: case Mips::F25: case Mips::D25_64:
-    return 25;
-  case Mips::K0: case Mips::K0_64: case Mips::F26: case Mips::D26_64:
-  case Mips::D13:
-    return 26;
-  case Mips::K1: case Mips::K1_64: case Mips::F27: case Mips::D27_64:
-    return 27;
-  case Mips::GP: case Mips::GP_64: case Mips::F28: case Mips::D28_64:
-  case Mips::D14:
-    return 28;
-  case Mips::SP: case Mips::SP_64: case Mips::F29: case Mips::D29_64:
-  case Mips::HWR29:
-    return 29;
-  case Mips::FP: case Mips::FP_64: case Mips::F30: case Mips::D30_64:
-  case Mips::D15:
-    return 30;
-  case Mips::RA: case Mips::RA_64: case Mips::F31: case Mips::D31_64:
-    return 31;
-  default: llvm_unreachable("Unknown register number!");
-  }
-}
-
 inline static std::pair<const MCSymbolRefExpr*, int64_t>
 MipsGetSymAndOffset(const MCFixup &Fixup) {
   MCFixupKind FixupKind = Fixup.getKind();
diff --git a/lib/Target/Mips/MCTargetDesc/MipsELFObjectWriter.cpp b/lib/Target/Mips/MCTargetDesc/MipsELFObjectWriter.cpp
index f82e203c23ca..6471b51583ce 100644
--- a/lib/Target/Mips/MCTargetDesc/MipsELFObjectWriter.cpp
+++ b/lib/Target/Mips/MCTargetDesc/MipsELFObjectWriter.cpp
@@ -42,7 +42,6 @@ namespace {
     virtual unsigned GetRelocType(const MCValue &Target, const MCFixup &Fixup,
                                   bool IsPCRel, bool IsRelocWithSymbol,
                                   int64_t Addend) const;
-    virtual unsigned getEFlags() const;
     virtual const MCSymbol *ExplicitRelSym(const MCAssembler &Asm,
                                            const MCValue &Target,
                                            const MCFragment &F,
@@ -61,19 +60,6 @@ MipsELFObjectWriter::MipsELFObjectWriter(bool _is64Bit, uint8_t OSABI,
 
 MipsELFObjectWriter::~MipsELFObjectWriter() {}
 
-// FIXME: get the real EABI Version from the Subtarget class.
-unsigned MipsELFObjectWriter::getEFlags() const {
-
-  // FIXME: We can't tell if we are PIC (dynamic) or CPIC (static)
-  unsigned Flag = ELF::EF_MIPS_NOREORDER;
-
-  if (is64Bit())
-    Flag |= ELF::EF_MIPS_ARCH_64R2;
-  else
-    Flag |= ELF::EF_MIPS_ARCH_32R2;
-  return Flag;
-}
-
 const MCSymbol *MipsELFObjectWriter::ExplicitRelSym(const MCAssembler &Asm,
                                                     const MCValue &Target,
                                                     const MCFragment &F,
@@ -108,7 +94,13 @@ unsigned MipsELFObjectWriter::GetRelocType(const MCValue &Target,
     Type = ELF::R_MIPS_64;
     break;
   case FK_GPRel_4:
-    Type = ELF::R_MIPS_GPREL32;
+    if (isN64()) {
+      Type = setRType((unsigned)ELF::R_MIPS_GPREL32, Type);
+      Type = setRType2((unsigned)ELF::R_MIPS_64, Type);
+      Type = setRType3((unsigned)ELF::R_MIPS_NONE, Type);
+    }
+    else
+      Type = ELF::R_MIPS_GPREL32;
     break;
   case Mips::fixup_Mips_GPREL16:
     Type = ELF::R_MIPS_GPREL16;
diff --git a/lib/Target/Mips/MCTargetDesc/MipsELFStreamer.cpp b/lib/Target/Mips/MCTargetDesc/MipsELFStreamer.cpp
new file mode 100644
index 000000000000..c33bc9ae3034
--- /dev/null
+++ b/lib/Target/Mips/MCTargetDesc/MipsELFStreamer.cpp
@@ -0,0 +1,89 @@
+//===-- MipsELFStreamer.cpp - MipsELFStreamer ---------------------------===//
+//
+//                       The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===-------------------------------------------------------------------===//
+#include "MCTargetDesc/MipsELFStreamer.h"
+#include "MipsSubtarget.h"
+#include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCELF.h"
+#include "llvm/MC/MCELFSymbolFlags.h"
+#include "llvm/MC/MCSymbol.h"
+#include "llvm/Support/ELF.h"
+#include "llvm/Support/ErrorHandling.h"
+
+namespace llvm {
+
+  MCELFStreamer* createMipsELFStreamer(MCContext &Context, MCAsmBackend &TAB,
+                                       raw_ostream &OS, MCCodeEmitter *Emitter,
+                                       bool RelaxAll, bool NoExecStack) {
+    MipsELFStreamer *S = new MipsELFStreamer(Context, TAB, OS, Emitter,
+                                             RelaxAll, NoExecStack);
+    return S;
+  }
+
+  // For llc. Set a group of ELF header flags
+  void
+  MipsELFStreamer::emitELFHeaderFlagsCG(const MipsSubtarget &Subtarget) {
+
+    if (hasRawTextSupport())
+      return;
+
+    // Update e_header flags
+    MCAssembler& MCA = getAssembler();
+    unsigned EFlags = MCA.getELFHeaderEFlags();
+
+    if (Subtarget.inMips16Mode())
+      EFlags |= ELF::EF_MIPS_ARCH_ASE_M16;
+    else
+      EFlags |= ELF::EF_MIPS_NOREORDER;
+
+    // Architecture
+    if (Subtarget.hasMips64r2())
+      EFlags |= ELF::EF_MIPS_ARCH_64R2;
+    else if (Subtarget.hasMips64())
+      EFlags |= ELF::EF_MIPS_ARCH_64;
+    else if (Subtarget.hasMips32r2())
+      EFlags |= ELF::EF_MIPS_ARCH_32R2;
+    else
+      EFlags |= ELF::EF_MIPS_ARCH_32;
+
+    if (Subtarget.inMicroMipsMode())
+      EFlags |= ELF::EF_MIPS_MICROMIPS;
+
+    // ABI
+    if (Subtarget.isABI_O32())
+      EFlags |= ELF::EF_MIPS_ABI_O32;
+
+    // Relocation Model
+    Reloc::Model RM = Subtarget.getRelocationModel();
+    if (RM == Reloc::PIC_ || RM == Reloc::Default)
+      EFlags |= ELF::EF_MIPS_PIC;
+    else if (RM == Reloc::Static)
+      ; // Do nothing for Reloc::Static
+    else
+      llvm_unreachable("Unsupported relocation model for e_flags");
+
+    MCA.setELFHeaderEFlags(EFlags);
+  }
+
+  // For llc. Set a symbol's STO flags
+  void
+  MipsELFStreamer::emitMipsSTOCG(const MipsSubtarget &Subtarget,
+                                 MCSymbol *Sym,
+                                 unsigned Val) {
+
+    if (hasRawTextSupport())
+      return;
+
+    MCSymbolData &Data = getOrCreateSymbolData(Sym);
+    // The "other" values are stored in the last 6 bits of the second byte
+    // The traditional defines for STO values assume the full byte and thus
+    // the shift to pack it.
+    MCELF::setOther(Data, Val >> 2);
+  }
+
+} // namespace llvm
diff --git a/lib/Target/Mips/MCTargetDesc/MipsELFStreamer.h b/lib/Target/Mips/MCTargetDesc/MipsELFStreamer.h
new file mode 100644
index 000000000000..b10ccc78e665
--- /dev/null
+++ b/lib/Target/Mips/MCTargetDesc/MipsELFStreamer.h
@@ -0,0 +1,43 @@
+//=== MipsELFStreamer.h - MipsELFStreamer ------------------------------===//
+//
+//                    The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENCE.TXT for details.
+//
+//===-------------------------------------------------------------------===//
+#ifndef MIPSELFSTREAMER_H_
+#define MIPSELFSTREAMER_H_
+
+#include "llvm/MC/MCELFStreamer.h"
+
+namespace llvm {
+class MipsAsmPrinter;
+class MipsSubtarget;
+class MCSymbol;
+
+class MipsELFStreamer : public MCELFStreamer {
+public:
+  MipsELFStreamer(MCContext &Context, MCAsmBackend &TAB,
+                  raw_ostream &OS, MCCodeEmitter *Emitter,
+                  bool RelaxAll, bool NoExecStack)
+    : MCELFStreamer(SK_MipsELFStreamer, Context, TAB, OS, Emitter) {
+  }
+
+  ~MipsELFStreamer() {}
+  void emitELFHeaderFlagsCG(const MipsSubtarget &Subtarget);
+  void emitMipsSTOCG(const MipsSubtarget &Subtarget,
+                     MCSymbol *Sym,
+                     unsigned Val);
+
+  static bool classof(const MCStreamer *S) {
+    return S->getKind() == SK_MipsELFStreamer;
+  }
+};
+
+  MCELFStreamer* createMipsELFStreamer(MCContext &Context, MCAsmBackend &TAB,
+                                       raw_ostream &OS, MCCodeEmitter *Emitter,
+                                       bool RelaxAll, bool NoExecStack);
+}
+
+#endif /* MIPSELFSTREAMER_H_ */
diff --git a/lib/Target/Mips/MCTargetDesc/MipsMCAsmInfo.cpp b/lib/Target/Mips/MCTargetDesc/MipsMCAsmInfo.cpp
index 9d67aa1856e3..5d4b32d30578 100644
--- a/lib/Target/Mips/MCTargetDesc/MipsMCAsmInfo.cpp
+++ b/lib/Target/Mips/MCTargetDesc/MipsMCAsmInfo.cpp
@@ -24,6 +24,11 @@ MipsMCAsmInfo::MipsMCAsmInfo(const Target &T, StringRef TT) {
       (TheTriple.getArch() == Triple::mips64))
     IsLittleEndian = false;
 
+  if ((TheTriple.getArch() == Triple::mips64el) ||
+      (TheTriple.getArch() == Triple::mips64)) {
+    PointerSize = CalleeSaveStackSlotSize = 8;
+  }
+
   AlignmentIsInBytes          = false;
   Data16bitsDirective         = "\t.2byte\t";
   Data32bitsDirective         = "\t.4byte\t";
@@ -34,7 +39,7 @@ MipsMCAsmInfo::MipsMCAsmInfo(const Target &T, StringRef TT) {
   GPRel32Directive            = "\t.gpword\t";
   GPRel64Directive            = "\t.gpdword\t";
   WeakRefDirective            = "\t.weak\t";
-
+  DebugLabelSuffix            = "=.";
   SupportsDebugInformation = true;
   ExceptionsType = ExceptionHandling::DwarfCFI;
   HasLEB128 = true;
diff --git a/lib/Target/Mips/MCTargetDesc/MipsMCCodeEmitter.cpp b/lib/Target/Mips/MCTargetDesc/MipsMCCodeEmitter.cpp
index da1e4552c9d0..e198a7c983f0 100644
--- a/lib/Target/Mips/MCTargetDesc/MipsMCCodeEmitter.cpp
+++ b/lib/Target/Mips/MCTargetDesc/MipsMCCodeEmitter.cpp
@@ -19,6 +19,7 @@
 #include "llvm/ADT/APFloat.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
@@ -33,11 +34,13 @@ class MipsMCCodeEmitter : public MCCodeEmitter {
   MipsMCCodeEmitter(const MipsMCCodeEmitter &) LLVM_DELETED_FUNCTION;
   void operator=(const MipsMCCodeEmitter &) LLVM_DELETED_FUNCTION;
   const MCInstrInfo &MCII;
+  MCContext &Ctx;
   bool IsLittleEndian;
 
 public:
-  MipsMCCodeEmitter(const MCInstrInfo &mcii, bool IsLittle) :
-            MCII(mcii), IsLittleEndian(IsLittle) {}
+  MipsMCCodeEmitter(const MCInstrInfo &mcii, MCContext &Ctx_,
+                    const MCSubtargetInfo &sti, bool IsLittle) :
+    MCII(mcii), Ctx(Ctx_), IsLittleEndian(IsLittle) {}
 
   ~MipsMCCodeEmitter() {}
 
@@ -93,7 +96,7 @@ MCCodeEmitter *llvm::createMipsMCCodeEmitterEB(const MCInstrInfo &MCII,
                                                const MCSubtargetInfo &STI,
                                                MCContext &Ctx)
 {
-  return new MipsMCCodeEmitter(MCII, false);
+  return new MipsMCCodeEmitter(MCII, Ctx, STI, false);
 }
 
 MCCodeEmitter *llvm::createMipsMCCodeEmitterEL(const MCInstrInfo &MCII,
@@ -101,7 +104,7 @@ MCCodeEmitter *llvm::createMipsMCCodeEmitterEL(const MCInstrInfo &MCII,
                                                const MCSubtargetInfo &STI,
                                                MCContext &Ctx)
 {
-  return new MipsMCCodeEmitter(MCII, true);
+  return new MipsMCCodeEmitter(MCII, Ctx, STI, true);
 }
 
 /// EncodeInstruction - Emit the instruction.
@@ -139,12 +142,6 @@ EncodeInstruction(const MCInst &MI, raw_ostream &OS,
     llvm_unreachable("unimplemented opcode in EncodeInstruction()");
 
   const MCInstrDesc &Desc = MCII.get(TmpInst.getOpcode());
-  uint64_t TSFlags = Desc.TSFlags;
-
-  // Pseudo instructions don't get encoded and shouldn't be here
-  // in the first place!
-  if ((TSFlags & MipsII::FormMask) == MipsII::Pseudo)
-    llvm_unreachable("Pseudo opcode found in EncodeInstruction()");
 
   // Get byte count of instruction
   unsigned Size = Desc.getSize();
@@ -163,8 +160,9 @@ getBranchTargetOpValue(const MCInst &MI, unsigned OpNo,
 
   const MCOperand &MO = MI.getOperand(OpNo);
 
-  // If the destination is an immediate, we have nothing to do.
-  if (MO.isImm()) return MO.getImm();
+  // If the destination is an immediate, divide by 4.
+  if (MO.isImm()) return MO.getImm() >> 2;
+
   assert(MO.isExpr() &&
          "getBranchTargetOpValue expects only expressions or immediates");
 
@@ -182,8 +180,9 @@ getJumpTargetOpValue(const MCInst &MI, unsigned OpNo,
                      SmallVectorImpl<MCFixup> &Fixups) const {
 
   const MCOperand &MO = MI.getOperand(OpNo);
-  // If the destination is an immediate, we have nothing to do.
-  if (MO.isImm()) return MO.getImm();
+  // If the destination is an immediate, divide by 4.
+  if (MO.isImm()) return MO.getImm()>>2;
+
   assert(MO.isExpr() &&
          "getJumpTargetOpValue expects only expressions or an immediate");
 
@@ -200,7 +199,7 @@ getMachineOpValue(const MCInst &MI, const MCOperand &MO,
                   SmallVectorImpl<MCFixup> &Fixups) const {
   if (MO.isReg()) {
     unsigned Reg = MO.getReg();
-    unsigned RegNo = getMipsRegisterNumbering(Reg);
+    unsigned RegNo = Ctx.getRegisterInfo().getEncodingValue(Reg);
     return RegNo;
   } else if (MO.isImm()) {
     return static_cast<unsigned>(MO.getImm());
diff --git a/lib/Target/Mips/MCTargetDesc/MipsMCTargetDesc.cpp b/lib/Target/Mips/MCTargetDesc/MipsMCTargetDesc.cpp
index f634f082be5a..be83b54b6124 100644
--- a/lib/Target/Mips/MCTargetDesc/MipsMCTargetDesc.cpp
+++ b/lib/Target/Mips/MCTargetDesc/MipsMCTargetDesc.cpp
@@ -11,15 +11,16 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "MipsMCAsmInfo.h"
+#include "MCTargetDesc/MipsELFStreamer.h"
 #include "MipsMCTargetDesc.h"
 #include "InstPrinter/MipsInstPrinter.h"
-#include "llvm/MC/MachineLocation.h"
+#include "MipsMCAsmInfo.h"
 #include "llvm/MC/MCCodeGenInfo.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/MC/MachineLocation.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
 
@@ -131,7 +132,7 @@ static MCStreamer *createMCStreamer(const Target &T, StringRef TT,
                                     bool NoExecStack) {
   Triple TheTriple(TT);
 
-  return createELFStreamer(Ctx, MAB, _OS, _Emitter, RelaxAll, NoExecStack);
+  return createMipsELFStreamer(Ctx, MAB, _OS, _Emitter, RelaxAll, NoExecStack);
 }
 
 extern "C" void LLVMInitializeMipsTargetMC() {
diff --git a/lib/Target/Mips/MCTargetDesc/MipsReginfo.cpp b/lib/Target/Mips/MCTargetDesc/MipsReginfo.cpp
new file mode 100644
index 000000000000..1dc9bcb36a5f
--- /dev/null
+++ b/lib/Target/Mips/MCTargetDesc/MipsReginfo.cpp
@@ -0,0 +1,80 @@
+//===-- MipsReginfo.cpp - Registerinfo handling  --------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+// .reginfo
+//    Elf32_Word ri_gprmask
+//    Elf32_Word ri_cprmask[4]
+//    Elf32_Word ri_gp_value
+//
+// .MIPS.options - N64
+//    Elf64_Byte    kind (ODK_REGINFO)
+//    Elf64_Byte    size (40 bytes)
+//    Elf64_Section section (0)
+//    Elf64_Word    info (unused)
+//    Elf64_Word    ri_gprmask ()
+//    Elf64_Word    ri_pad ()
+//    Elf64_Word[4] ri_cprmask ()
+//    Elf64_Addr    ri_gp_value ()
+//
+// .MIPS.options - N32
+//    Elf32_Byte    kind (ODK_REGINFO)
+//    Elf32_Byte    size (36 bytes)
+//    Elf32_Section section (0)
+//    Elf32_Word    info (unused)
+//    Elf32_Word    ri_gprmask ()
+//    Elf32_Word    ri_pad ()
+//    Elf32_Word[4] ri_cprmask ()
+//    Elf32_Addr    ri_gp_value ()
+//
+//===----------------------------------------------------------------------===//
+#include "MCTargetDesc/MipsReginfo.h"
+#include "MipsSubtarget.h"
+#include "MipsTargetObjectFile.h"
+#include "llvm/MC/MCStreamer.h"
+
+using namespace llvm;
+
+// Integrated assembler version
+void
+MipsReginfo::emitMipsReginfoSectionCG(MCStreamer &OS,
+    const TargetLoweringObjectFile &TLOF,
+    const MipsSubtarget &MST) const
+{
+
+  if (OS.hasRawTextSupport())
+    return;
+
+  const MipsTargetObjectFile &TLOFELF =
+      static_cast<const MipsTargetObjectFile &>(TLOF);
+  OS.SwitchSection(TLOFELF.getReginfoSection());
+
+  // .reginfo
+  if (MST.isABI_O32()) {
+    OS.EmitIntValue(0, 4); // ri_gprmask
+    OS.EmitIntValue(0, 4); // ri_cpr[0]mask
+    OS.EmitIntValue(0, 4); // ri_cpr[1]mask
+    OS.EmitIntValue(0, 4); // ri_cpr[2]mask
+    OS.EmitIntValue(0, 4); // ri_cpr[3]mask
+    OS.EmitIntValue(0, 4); // ri_gp_value
+  }
+  // .MIPS.options
+  else if (MST.isABI_N64()) {
+    OS.EmitIntValue(1, 1); // kind
+    OS.EmitIntValue(40, 1); // size
+    OS.EmitIntValue(0, 2); // section
+    OS.EmitIntValue(0, 4); // info
+    OS.EmitIntValue(0, 4); // ri_gprmask
+    OS.EmitIntValue(0, 4); // pad
+    OS.EmitIntValue(0, 4); // ri_cpr[0]mask
+    OS.EmitIntValue(0, 4); // ri_cpr[1]mask
+    OS.EmitIntValue(0, 4); // ri_cpr[2]mask
+    OS.EmitIntValue(0, 4); // ri_cpr[3]mask
+    OS.EmitIntValue(0, 8); // ri_gp_value
+  }
+  else llvm_unreachable("Unsupported abi for reginfo");
+}
+
diff --git a/lib/Target/Mips/MCTargetDesc/MipsReginfo.h b/lib/Target/Mips/MCTargetDesc/MipsReginfo.h
new file mode 100644
index 000000000000..039b8eaaf287
--- /dev/null
+++ b/lib/Target/Mips/MCTargetDesc/MipsReginfo.h
@@ -0,0 +1,31 @@
+//=== MipsReginfo.h - MipsReginfo -----------------------------------------===//
+//
+//                    The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENCE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MIPSREGINFO_H
+#define MIPSREGINFO_H
+
+namespace llvm {
+  class MCStreamer;
+  class TargetLoweringObjectFile;
+  class MipsSubtarget;
+
+  class MipsReginfo {
+    void anchor();
+  public:
+    MipsReginfo() {}
+
+    void emitMipsReginfoSectionCG(MCStreamer &OS,
+        const TargetLoweringObjectFile &TLOF,
+        const MipsSubtarget &MST) const;
+  };
+
+} // namespace llvm
+
+#endif
+
diff --git a/lib/Target/Mips/Makefile b/lib/Target/Mips/Makefile
index bd8c5173454e..bcf951e861b0 100644
--- a/lib/Target/Mips/Makefile
+++ b/lib/Target/Mips/Makefile
@@ -16,7 +16,7 @@ BUILT_SOURCES = MipsGenRegisterInfo.inc MipsGenInstrInfo.inc \
                 MipsGenAsmWriter.inc MipsGenCodeEmitter.inc \
                 MipsGenDAGISel.inc MipsGenCallingConv.inc \
                 MipsGenSubtargetInfo.inc MipsGenMCCodeEmitter.inc \
-                MipsGenEDInfo.inc MipsGenDisassemblerTables.inc \
+                MipsGenDisassemblerTables.inc \
                 MipsGenMCPseudoLowering.inc MipsGenAsmMatcher.inc
 
 DIRS = InstPrinter Disassembler AsmParser TargetInfo MCTargetDesc
diff --git a/lib/Target/Mips/Mips.h b/lib/Target/Mips/Mips.h
index 2963f7e7fa42..8c65bb4020b5 100644
--- a/lib/Target/Mips/Mips.h
+++ b/lib/Target/Mips/Mips.h
@@ -27,6 +27,7 @@ namespace llvm {
   FunctionPass *createMipsLongBranchPass(MipsTargetMachine &TM);
   FunctionPass *createMipsJITCodeEmitterPass(MipsTargetMachine &TM,
                                              JITCodeEmitter &JCE);
+  FunctionPass *createMipsConstantIslandPass(MipsTargetMachine &tm);
 
 } // end namespace llvm;
 
diff --git a/lib/Target/Mips/Mips.td b/lib/Target/Mips/Mips.td
index 90c01d5de0a9..eefb02a494ca 100644
--- a/lib/Target/Mips/Mips.td
+++ b/lib/Target/Mips/Mips.td
@@ -44,32 +44,29 @@ def FeatureN64         : SubtargetFeature<"n64", "MipsABI", "N64",
                                 "Enable n64 ABI">;
 def FeatureEABI        : SubtargetFeature<"eabi", "MipsABI", "EABI",
                                 "Enable eabi ABI">;
-def FeatureAndroid     : SubtargetFeature<"android", "IsAndroid", "true",
-                                "Target is android">;
 def FeatureVFPU        : SubtargetFeature<"vfpu", "HasVFPU",
                                 "true", "Enable vector FPU instructions.">;
 def FeatureSEInReg     : SubtargetFeature<"seinreg", "HasSEInReg", "true",
                                 "Enable 'signext in register' instructions.">;
 def FeatureCondMov     : SubtargetFeature<"condmov", "HasCondMov", "true",
                                 "Enable 'conditional move' instructions.">;
-def FeatureMulDivAdd   : SubtargetFeature<"muldivadd", "HasMulDivAdd", "true",
-                                "Enable 'multiply add/sub' instructions.">;
-def FeatureMinMax      : SubtargetFeature<"minmax", "HasMinMax", "true",
-                                "Enable 'min/max' instructions.">;
 def FeatureSwap        : SubtargetFeature<"swap", "HasSwap", "true",
                                 "Enable 'byte/half swap' instructions.">;
 def FeatureBitCount    : SubtargetFeature<"bitcount", "HasBitCount", "true",
                                 "Enable 'count leading bits' instructions.">;
+def FeatureFPIdx       : SubtargetFeature<"FPIdx", "HasFPIdx", "true",
+                                "Enable 'FP indexed load/store' instructions.">;
 def FeatureMips32      : SubtargetFeature<"mips32", "MipsArchVersion", "Mips32",
                                 "Mips32 ISA Support",
                                 [FeatureCondMov, FeatureBitCount]>;
 def FeatureMips32r2    : SubtargetFeature<"mips32r2", "MipsArchVersion",
                                 "Mips32r2", "Mips32r2 ISA Support",
-                                [FeatureMips32, FeatureSEInReg, FeatureSwap]>;
+                                [FeatureMips32, FeatureSEInReg, FeatureSwap,
+                                 FeatureFPIdx]>;
 def FeatureMips64      : SubtargetFeature<"mips64", "MipsArchVersion",
                                 "Mips64", "Mips64 ISA Support",
                                 [FeatureGP64Bit, FeatureFP64Bit,
-                                 FeatureMips32]>;
+                                 FeatureMips32, FeatureFPIdx]>;
 def FeatureMips64r2    : SubtargetFeature<"mips64r2", "MipsArchVersion",
                                 "Mips64r2", "Mips64r2 ISA Support",
                                 [FeatureMips64, FeatureMips32r2]>;
@@ -81,6 +78,9 @@ def FeatureDSP : SubtargetFeature<"dsp", "HasDSP", "true", "Mips DSP ASE">;
 def FeatureDSPR2 : SubtargetFeature<"dspr2", "HasDSPR2", "true",
                                     "Mips DSP-R2 ASE", [FeatureDSP]>;
 
+def FeatureMicroMips  : SubtargetFeature<"micromips", "InMicroMipsMode", "true",
+                                         "microMips mode">;
+
 //===----------------------------------------------------------------------===//
 // Mips processors supported.
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/Mips/Mips16FrameLowering.cpp b/lib/Target/Mips/Mips16FrameLowering.cpp
index 4e6b21feb55d..1bb6fe46295b 100644
--- a/lib/Target/Mips/Mips16FrameLowering.cpp
+++ b/lib/Target/Mips/Mips16FrameLowering.cpp
@@ -12,25 +12,26 @@
 //===----------------------------------------------------------------------===//
 
 #include "Mips16FrameLowering.h"
-#include "MipsInstrInfo.h"
 #include "MCTargetDesc/MipsBaseInfo.h"
-#include "llvm/Function.h"
+#include "Mips16InstrInfo.h"
+#include "MipsInstrInfo.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
 void Mips16FrameLowering::emitPrologue(MachineFunction &MF) const {
   MachineBasicBlock &MBB = MF.front();
   MachineFrameInfo *MFI = MF.getFrameInfo();
-  const MipsInstrInfo &TII =
-    *static_cast<const MipsInstrInfo*>(MF.getTarget().getInstrInfo());
+  const Mips16InstrInfo &TII =
+    *static_cast<const Mips16InstrInfo*>(MF.getTarget().getInstrInfo());
   MachineBasicBlock::iterator MBBI = MBB.begin();
   DebugLoc dl = MBBI != MBB.end() ? MBBI->getDebugLoc() : DebugLoc();
   uint64_t StackSize = MFI->getStackSize();
@@ -38,9 +39,35 @@ void Mips16FrameLowering::emitPrologue(MachineFunction &MF) const {
   // No need to allocate space on the stack.
   if (StackSize == 0 && !MFI->adjustsStack()) return;
 
+  MachineModuleInfo &MMI = MF.getMMI();
+  std::vector<MachineMove> &Moves = MMI.getFrameMoves();
+  MachineLocation DstML, SrcML;
+
   // Adjust stack.
-  if (isInt<16>(-StackSize))
-    BuildMI(MBB, MBBI, dl, TII.get(Mips::SaveRaF16)).addImm(StackSize);
+  TII.makeFrame(Mips::SP, StackSize, MBB, MBBI);
+
+  // emit ".cfi_def_cfa_offset StackSize"
+  MCSymbol *AdjustSPLabel = MMI.getContext().CreateTempSymbol();
+  BuildMI(MBB, MBBI, dl,
+          TII.get(TargetOpcode::PROLOG_LABEL)).addSym(AdjustSPLabel);
+  DstML = MachineLocation(MachineLocation::VirtualFP);
+  SrcML = MachineLocation(MachineLocation::VirtualFP, -StackSize);
+  Moves.push_back(MachineMove(AdjustSPLabel, DstML, SrcML));
+
+  MCSymbol *CSLabel = MMI.getContext().CreateTempSymbol();
+  BuildMI(MBB, MBBI, dl,
+          TII.get(TargetOpcode::PROLOG_LABEL)).addSym(CSLabel);
+  DstML = MachineLocation(MachineLocation::VirtualFP, -8);
+  SrcML = MachineLocation(Mips::S1);
+  Moves.push_back(MachineMove(CSLabel, DstML, SrcML));
+
+  DstML = MachineLocation(MachineLocation::VirtualFP, -12);
+  SrcML = MachineLocation(Mips::S0);
+  Moves.push_back(MachineMove(CSLabel, DstML, SrcML));
+
+  DstML = MachineLocation(MachineLocation::VirtualFP, -4);
+  SrcML = MachineLocation(Mips::RA);
+  Moves.push_back(MachineMove(CSLabel, DstML, SrcML));
 
   if (hasFP(MF))
     BuildMI(MBB, MBBI, dl, TII.get(Mips::MoveR3216), Mips::S0)
@@ -52,8 +79,8 @@ void Mips16FrameLowering::emitEpilogue(MachineFunction &MF,
                                  MachineBasicBlock &MBB) const {
   MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
   MachineFrameInfo *MFI = MF.getFrameInfo();
-  const MipsInstrInfo &TII =
-    *static_cast<const MipsInstrInfo*>(MF.getTarget().getInstrInfo());
+  const Mips16InstrInfo &TII =
+    *static_cast<const Mips16InstrInfo*>(MF.getTarget().getInstrInfo());
   DebugLoc dl = MBBI->getDebugLoc();
   uint64_t StackSize = MFI->getStackSize();
 
@@ -65,9 +92,8 @@ void Mips16FrameLowering::emitEpilogue(MachineFunction &MF,
       .addReg(Mips::S0);
 
   // Adjust stack.
-  if (isInt<16>(StackSize))
-    // assumes stacksize multiple of 8
-    BuildMI(MBB, MBBI, dl, TII.get(Mips::RestoreRaF16)).addImm(StackSize);
+  // assumes stacksize multiple of 8
+  TII.restoreFrame(Mips::SP, StackSize, MBB, MBBI);
 }
 
 bool Mips16FrameLowering::
@@ -113,6 +139,25 @@ bool Mips16FrameLowering::restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
   return true;
 }
 
+// Eliminate ADJCALLSTACKDOWN, ADJCALLSTACKUP pseudo instructions
+void Mips16FrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  if (!hasReservedCallFrame(MF)) {
+    int64_t Amount = I->getOperand(0).getImm();
+
+    if (I->getOpcode() == Mips::ADJCALLSTACKDOWN)
+      Amount = -Amount;
+
+    const Mips16InstrInfo &TII =
+      *static_cast<const Mips16InstrInfo*>(MF.getTarget().getInstrInfo());
+
+    TII.adjustStackPtr(Mips::SP, Amount, MBB, I);
+  }
+
+  MBB.erase(I);
+}
+
 bool
 Mips16FrameLowering::hasReservedCallFrame(const MachineFunction &MF) const {
   const MachineFrameInfo *MFI = MF.getFrameInfo();
diff --git a/lib/Target/Mips/Mips16FrameLowering.h b/lib/Target/Mips/Mips16FrameLowering.h
index 01db71e8def5..54fdb7871466 100644
--- a/lib/Target/Mips/Mips16FrameLowering.h
+++ b/lib/Target/Mips/Mips16FrameLowering.h
@@ -20,13 +20,17 @@ namespace llvm {
 class Mips16FrameLowering : public MipsFrameLowering {
 public:
   explicit Mips16FrameLowering(const MipsSubtarget &STI)
-    : MipsFrameLowering(STI) {}
+    : MipsFrameLowering(STI, 8) {}
 
   /// emitProlog/emitEpilog - These methods insert prolog and epilog code into
   /// the function.
   void emitPrologue(MachineFunction &MF) const;
   void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
 
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I) const;
+
   bool spillCalleeSavedRegisters(MachineBasicBlock &MBB,
                                  MachineBasicBlock::iterator MI,
                                  const std::vector<CalleeSavedInfo> &CSI,
diff --git a/lib/Target/Mips/Mips16ISelDAGToDAG.cpp b/lib/Target/Mips/Mips16ISelDAGToDAG.cpp
new file mode 100644
index 000000000000..00b3449300c5
--- /dev/null
+++ b/lib/Target/Mips/Mips16ISelDAGToDAG.cpp
@@ -0,0 +1,308 @@
+//===-- Mips16ISelDAGToDAG.cpp - A Dag to Dag Inst Selector for Mips16 ----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Subclass of MipsDAGToDAGISel specialized for mips16.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "mips-isel"
+#include "Mips16ISelDAGToDAG.h"
+#include "Mips.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
+#include "MipsAnalyzeImmediate.h"
+#include "MipsMachineFunction.h"
+#include "MipsRegisterInfo.h"
+#include "llvm/CodeGen/MachineConstantPool.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/SelectionDAGNodes.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Type.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetMachine.h"
+using namespace llvm;
+
+/// Select multiply instructions.
+std::pair<SDNode*, SDNode*>
+Mips16DAGToDAGISel::selectMULT(SDNode *N, unsigned Opc, DebugLoc DL, EVT Ty,
+                               bool HasLo, bool HasHi) {
+  SDNode *Lo = 0, *Hi = 0;
+  SDNode *Mul = CurDAG->getMachineNode(Opc, DL, MVT::Glue, N->getOperand(0),
+                                       N->getOperand(1));
+  SDValue InFlag = SDValue(Mul, 0);
+
+  if (HasLo) {
+    unsigned Opcode = Mips::Mflo16;
+    Lo = CurDAG->getMachineNode(Opcode, DL, Ty, MVT::Glue, InFlag);
+    InFlag = SDValue(Lo, 1);
+  }
+  if (HasHi) {
+    unsigned Opcode = Mips::Mfhi16;
+    Hi = CurDAG->getMachineNode(Opcode, DL, Ty, InFlag);
+  }
+  return std::make_pair(Lo, Hi);
+}
+
+void Mips16DAGToDAGISel::initGlobalBaseReg(MachineFunction &MF) {
+  MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
+
+  if (!MipsFI->globalBaseRegSet())
+    return;
+
+  MachineBasicBlock &MBB = MF.front();
+  MachineBasicBlock::iterator I = MBB.begin();
+  MachineRegisterInfo &RegInfo = MF.getRegInfo();
+  const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
+  DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
+  unsigned V0, V1, V2, GlobalBaseReg = MipsFI->getGlobalBaseReg();
+  const TargetRegisterClass *RC =
+    (const TargetRegisterClass*)&Mips::CPU16RegsRegClass;
+
+  V0 = RegInfo.createVirtualRegister(RC);
+  V1 = RegInfo.createVirtualRegister(RC);
+  V2 = RegInfo.createVirtualRegister(RC);
+
+  BuildMI(MBB, I, DL, TII.get(Mips::LiRxImmX16), V0)
+    .addExternalSymbol("_gp_disp", MipsII::MO_ABS_HI);
+  BuildMI(MBB, I, DL, TII.get(Mips::AddiuRxPcImmX16), V1)
+    .addExternalSymbol("_gp_disp", MipsII::MO_ABS_LO);
+  BuildMI(MBB, I, DL, TII.get(Mips::SllX16), V2).addReg(V0).addImm(16);
+  BuildMI(MBB, I, DL, TII.get(Mips::AdduRxRyRz16), GlobalBaseReg)
+    .addReg(V1).addReg(V2);
+}
+
+// Insert instructions to initialize the Mips16 SP Alias register in the
+// first MBB of the function.
+//
+void Mips16DAGToDAGISel::initMips16SPAliasReg(MachineFunction &MF) {
+  MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
+
+  if (!MipsFI->mips16SPAliasRegSet())
+    return;
+
+  MachineBasicBlock &MBB = MF.front();
+  MachineBasicBlock::iterator I = MBB.begin();
+  const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
+  DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
+  unsigned Mips16SPAliasReg = MipsFI->getMips16SPAliasReg();
+
+  BuildMI(MBB, I, DL, TII.get(Mips::MoveR3216), Mips16SPAliasReg)
+    .addReg(Mips::SP);
+}
+
+void Mips16DAGToDAGISel::processFunctionAfterISel(MachineFunction &MF) {
+  initGlobalBaseReg(MF);
+  initMips16SPAliasReg(MF);
+}
+
+/// getMips16SPAliasReg - Output the instructions required to put the
+/// SP into a Mips16 accessible aliased register.
+SDValue Mips16DAGToDAGISel::getMips16SPAliasReg() {
+  unsigned Mips16SPAliasReg =
+    MF->getInfo<MipsFunctionInfo>()->getMips16SPAliasReg();
+  return CurDAG->getRegister(Mips16SPAliasReg, TLI.getPointerTy());
+}
+
+void Mips16DAGToDAGISel::getMips16SPRefReg(SDNode *Parent, SDValue &AliasReg) {
+  SDValue AliasFPReg = CurDAG->getRegister(Mips::S0, TLI.getPointerTy());
+  if (Parent) {
+    switch (Parent->getOpcode()) {
+      case ISD::LOAD: {
+        LoadSDNode *SD = dyn_cast<LoadSDNode>(Parent);
+        switch (SD->getMemoryVT().getSizeInBits()) {
+        case 8:
+        case 16:
+          AliasReg = TM.getFrameLowering()->hasFP(*MF)?
+            AliasFPReg: getMips16SPAliasReg();
+          return;
+        }
+        break;
+      }
+      case ISD::STORE: {
+        StoreSDNode *SD = dyn_cast<StoreSDNode>(Parent);
+        switch (SD->getMemoryVT().getSizeInBits()) {
+        case 8:
+        case 16:
+          AliasReg = TM.getFrameLowering()->hasFP(*MF)?
+            AliasFPReg: getMips16SPAliasReg();
+          return;
+        }
+        break;
+      }
+    }
+  }
+  AliasReg = CurDAG->getRegister(Mips::SP, TLI.getPointerTy());
+  return;
+
+}
+
+bool Mips16DAGToDAGISel::selectAddr16(
+  SDNode *Parent, SDValue Addr, SDValue &Base, SDValue &Offset,
+  SDValue &Alias) {
+  EVT ValTy = Addr.getValueType();
+
+  Alias = CurDAG->getTargetConstant(0, ValTy);
+
+  // if Address is FI, get the TargetFrameIndex.
+  if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) {
+    Base   = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy);
+    Offset = CurDAG->getTargetConstant(0, ValTy);
+    getMips16SPRefReg(Parent, Alias);
+    return true;
+  }
+  // on PIC code Load GA
+  if (Addr.getOpcode() == MipsISD::Wrapper) {
+    Base   = Addr.getOperand(0);
+    Offset = Addr.getOperand(1);
+    return true;
+  }
+  if (TM.getRelocationModel() != Reloc::PIC_) {
+    if ((Addr.getOpcode() == ISD::TargetExternalSymbol ||
+        Addr.getOpcode() == ISD::TargetGlobalAddress))
+      return false;
+  }
+  // Addresses of the form FI+const or FI|const
+  if (CurDAG->isBaseWithConstantOffset(Addr)) {
+    ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Addr.getOperand(1));
+    if (isInt<16>(CN->getSExtValue())) {
+
+      // If the first operand is a FI, get the TargetFI Node
+      if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>
+                                  (Addr.getOperand(0))) {
+        Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy);
+        getMips16SPRefReg(Parent, Alias);
+      }
+      else
+        Base = Addr.getOperand(0);
+
+      Offset = CurDAG->getTargetConstant(CN->getZExtValue(), ValTy);
+      return true;
+    }
+  }
+  // Operand is a result from an ADD.
+  if (Addr.getOpcode() == ISD::ADD) {
+    // When loading from constant pools, load the lower address part in
+    // the instruction itself. Example, instead of:
+    //  lui $2, %hi($CPI1_0)
+    //  addiu $2, $2, %lo($CPI1_0)
+    //  lwc1 $f0, 0($2)
+    // Generate:
+    //  lui $2, %hi($CPI1_0)
+    //  lwc1 $f0, %lo($CPI1_0)($2)
+    if (Addr.getOperand(1).getOpcode() == MipsISD::Lo ||
+        Addr.getOperand(1).getOpcode() == MipsISD::GPRel) {
+      SDValue Opnd0 = Addr.getOperand(1).getOperand(0);
+      if (isa<ConstantPoolSDNode>(Opnd0) || isa<GlobalAddressSDNode>(Opnd0) ||
+          isa<JumpTableSDNode>(Opnd0)) {
+        Base = Addr.getOperand(0);
+        Offset = Opnd0;
+        return true;
+      }
+    }
+
+    // If an indexed floating point load/store can be emitted, return false.
+    const LSBaseSDNode *LS = dyn_cast<LSBaseSDNode>(Parent);
+
+    if (LS &&
+        (LS->getMemoryVT() == MVT::f32 || LS->getMemoryVT() == MVT::f64) &&
+        Subtarget.hasFPIdx())
+      return false;
+  }
+  Base   = Addr;
+  Offset = CurDAG->getTargetConstant(0, ValTy);
+  return true;
+}
+
+/// Select instructions not customized! Used for
+/// expanded, promoted and normal instructions
+std::pair<bool, SDNode*> Mips16DAGToDAGISel::selectNode(SDNode *Node) {
+  unsigned Opcode = Node->getOpcode();
+  DebugLoc DL = Node->getDebugLoc();
+
+  ///
+  // Instruction Selection not handled by the auto-generated
+  // tablegen selection should be handled here.
+  ///
+  EVT NodeTy = Node->getValueType(0);
+  unsigned MultOpc;
+
+  switch(Opcode) {
+  default: break;
+
+  case ISD::SUBE:
+  case ISD::ADDE: {
+    SDValue InFlag = Node->getOperand(2), CmpLHS;
+    unsigned Opc = InFlag.getOpcode(); (void)Opc;
+    assert(((Opc == ISD::ADDC || Opc == ISD::ADDE) ||
+            (Opc == ISD::SUBC || Opc == ISD::SUBE)) &&
+           "(ADD|SUB)E flag operand must come from (ADD|SUB)C/E insn");
+
+    unsigned MOp;
+    if (Opcode == ISD::ADDE) {
+      CmpLHS = InFlag.getValue(0);
+      MOp = Mips::AdduRxRyRz16;
+    } else {
+      CmpLHS = InFlag.getOperand(0);
+      MOp = Mips::SubuRxRyRz16;
+    }
+
+    SDValue Ops[] = { CmpLHS, InFlag.getOperand(1) };
+
+    SDValue LHS = Node->getOperand(0);
+    SDValue RHS = Node->getOperand(1);
+
+    EVT VT = LHS.getValueType();
+
+    unsigned Sltu_op = Mips::SltuRxRyRz16;
+    SDNode *Carry = CurDAG->getMachineNode(Sltu_op, DL, VT, Ops, 2);
+    unsigned Addu_op = Mips::AdduRxRyRz16;
+    SDNode *AddCarry = CurDAG->getMachineNode(Addu_op, DL, VT,
+                                              SDValue(Carry,0), RHS);
+
+    SDNode *Result = CurDAG->SelectNodeTo(Node, MOp, VT, MVT::Glue, LHS,
+                                          SDValue(AddCarry,0));
+    return std::make_pair(true, Result);
+  }
+
+  /// Mul with two results
+  case ISD::SMUL_LOHI:
+  case ISD::UMUL_LOHI: {
+    MultOpc = (Opcode == ISD::UMUL_LOHI ? Mips::MultuRxRy16 : Mips::MultRxRy16);
+    std::pair<SDNode*, SDNode*> LoHi = selectMULT(Node, MultOpc, DL, NodeTy,
+                                                  true, true);
+    if (!SDValue(Node, 0).use_empty())
+      ReplaceUses(SDValue(Node, 0), SDValue(LoHi.first, 0));
+
+    if (!SDValue(Node, 1).use_empty())
+      ReplaceUses(SDValue(Node, 1), SDValue(LoHi.second, 0));
+
+    return std::make_pair(true, (SDNode*)NULL);
+  }
+
+  case ISD::MULHS:
+  case ISD::MULHU: {
+    MultOpc = (Opcode == ISD::MULHU ? Mips::MultuRxRy16 : Mips::MultRxRy16);
+    SDNode *Result = selectMULT(Node, MultOpc, DL, NodeTy, false, true).second;
+    return std::make_pair(true, Result);
+  }
+  }
+
+  return std::make_pair(false, (SDNode*)NULL);
+}
+
+FunctionPass *llvm::createMips16ISelDag(MipsTargetMachine &TM) {
+  return new Mips16DAGToDAGISel(TM);
+}
diff --git a/lib/Target/Mips/Mips16ISelDAGToDAG.h b/lib/Target/Mips/Mips16ISelDAGToDAG.h
new file mode 100644
index 000000000000..baa85877d957
--- /dev/null
+++ b/lib/Target/Mips/Mips16ISelDAGToDAG.h
@@ -0,0 +1,51 @@
+//===---- Mips16ISelDAGToDAG.h - A Dag to Dag Inst Selector for Mips ------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Subclass of MipsDAGToDAGISel specialized for mips16.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MIPS16ISELDAGTODAG_H
+#define MIPS16ISELDAGTODAG_H
+
+#include "MipsISelDAGToDAG.h"
+
+namespace llvm {
+
+class Mips16DAGToDAGISel : public MipsDAGToDAGISel {
+public:
+  explicit Mips16DAGToDAGISel(MipsTargetMachine &TM) : MipsDAGToDAGISel(TM) {}
+
+private:
+  std::pair<SDNode*, SDNode*> selectMULT(SDNode *N, unsigned Opc, DebugLoc DL,
+                                         EVT Ty, bool HasLo, bool HasHi);
+
+  SDValue getMips16SPAliasReg();
+
+  void getMips16SPRefReg(SDNode *Parent, SDValue &AliasReg);
+
+  virtual bool selectAddr16(SDNode *Parent, SDValue N, SDValue &Base,
+                            SDValue &Offset, SDValue &Alias);
+
+  virtual std::pair<bool, SDNode*> selectNode(SDNode *Node);
+
+  virtual void processFunctionAfterISel(MachineFunction &MF);
+
+  // Insert instructions to initialize the global base register in the
+  // first MBB of the function.
+  void initGlobalBaseReg(MachineFunction &MF);
+
+  void initMips16SPAliasReg(MachineFunction &MF);
+};
+
+FunctionPass *createMips16ISelDag(MipsTargetMachine &TM);
+
+}
+
+#endif
diff --git a/lib/Target/Mips/Mips16ISelLowering.cpp b/lib/Target/Mips/Mips16ISelLowering.cpp
new file mode 100644
index 000000000000..23eb5375ac94
--- /dev/null
+++ b/lib/Target/Mips/Mips16ISelLowering.cpp
@@ -0,0 +1,689 @@
+//===-- Mips16ISelLowering.h - Mips16 DAG Lowering Interface ----*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Subclass of MipsTargetLowering specialized for mips16.
+//
+//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "mips-lower"
+#include "Mips16ISelLowering.h"
+#include "MipsRegisterInfo.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include <set>
+
+using namespace llvm;
+
+static cl::opt<bool>
+Mips16HardFloat("mips16-hard-float", cl::NotHidden,
+                cl::desc("MIPS: mips16 hard float enable."),
+                cl::init(false));
+
+static cl::opt<bool> DontExpandCondPseudos16(
+  "mips16-dont-expand-cond-pseudo",
+  cl::init(false),
+  cl::desc("Dont expand conditional move related "
+           "pseudos for Mips 16"),
+  cl::Hidden);
+
+namespace {
+  std::set<const char*, MipsTargetLowering::LTStr> NoHelperNeeded;
+}
+
+Mips16TargetLowering::Mips16TargetLowering(MipsTargetMachine &TM)
+  : MipsTargetLowering(TM) {
+  //
+  // set up as if mips32 and then revert so we can test the mechanism
+  // for switching
+  addRegisterClass(MVT::i32, &Mips::CPURegsRegClass);
+  addRegisterClass(MVT::f32, &Mips::FGR32RegClass);
+  computeRegisterProperties();
+  clearRegisterClasses();
+
+  // Set up the register classes
+  addRegisterClass(MVT::i32, &Mips::CPU16RegsRegClass);
+
+  if (Mips16HardFloat)
+    setMips16HardFloatLibCalls();
+
+  setOperationAction(ISD::MEMBARRIER,         MVT::Other, Expand);
+  setOperationAction(ISD::ATOMIC_FENCE,       MVT::Other, Expand);
+  setOperationAction(ISD::ATOMIC_CMP_SWAP,    MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_SWAP,        MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_LOAD_ADD,    MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_LOAD_SUB,    MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_LOAD_AND,    MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_LOAD_OR,     MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_LOAD_XOR,    MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_LOAD_NAND,   MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_LOAD_MIN,    MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_LOAD_MAX,    MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_LOAD_UMIN,   MVT::i32,   Expand);
+  setOperationAction(ISD::ATOMIC_LOAD_UMAX,   MVT::i32,   Expand);
+
+  computeRegisterProperties();
+}
+
+const MipsTargetLowering *
+llvm::createMips16TargetLowering(MipsTargetMachine &TM) {
+  return new Mips16TargetLowering(TM);
+}
+
+bool
+Mips16TargetLowering::allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const {
+  return false;
+}
+
+MachineBasicBlock *
+Mips16TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
+                                                  MachineBasicBlock *BB) const {
+  switch (MI->getOpcode()) {
+  default:
+    return MipsTargetLowering::EmitInstrWithCustomInserter(MI, BB);
+  case Mips::SelBeqZ:
+    return emitSel16(Mips::BeqzRxImm16, MI, BB);
+  case Mips::SelBneZ:
+    return emitSel16(Mips::BnezRxImm16, MI, BB);
+  case Mips::SelTBteqZCmpi:
+    return emitSeliT16(Mips::BteqzX16, Mips::CmpiRxImmX16, MI, BB);
+  case Mips::SelTBteqZSlti:
+    return emitSeliT16(Mips::BteqzX16, Mips::SltiRxImmX16, MI, BB);
+  case Mips::SelTBteqZSltiu:
+    return emitSeliT16(Mips::BteqzX16, Mips::SltiuRxImmX16, MI, BB);
+  case Mips::SelTBtneZCmpi:
+    return emitSeliT16(Mips::BtnezX16, Mips::CmpiRxImmX16, MI, BB);
+  case Mips::SelTBtneZSlti:
+    return emitSeliT16(Mips::BtnezX16, Mips::SltiRxImmX16, MI, BB);
+  case Mips::SelTBtneZSltiu:
+    return emitSeliT16(Mips::BtnezX16, Mips::SltiuRxImmX16, MI, BB);
+  case Mips::SelTBteqZCmp:
+    return emitSelT16(Mips::BteqzX16, Mips::CmpRxRy16, MI, BB);
+  case Mips::SelTBteqZSlt:
+    return emitSelT16(Mips::BteqzX16, Mips::SltRxRy16, MI, BB);
+  case Mips::SelTBteqZSltu:
+    return emitSelT16(Mips::BteqzX16, Mips::SltuRxRy16, MI, BB);
+  case Mips::SelTBtneZCmp:
+    return emitSelT16(Mips::BtnezX16, Mips::CmpRxRy16, MI, BB);
+  case Mips::SelTBtneZSlt:
+    return emitSelT16(Mips::BtnezX16, Mips::SltRxRy16, MI, BB);
+  case Mips::SelTBtneZSltu:
+    return emitSelT16(Mips::BtnezX16, Mips::SltuRxRy16, MI, BB);
+  case Mips::BteqzT8CmpX16:
+    return emitFEXT_T8I816_ins(Mips::BteqzX16, Mips::CmpRxRy16, MI, BB);
+  case Mips::BteqzT8SltX16:
+    return emitFEXT_T8I816_ins(Mips::BteqzX16, Mips::SltRxRy16, MI, BB);
+  case Mips::BteqzT8SltuX16:
+    // TBD: figure out a way to get this or remove the instruction
+    // altogether.
+    return emitFEXT_T8I816_ins(Mips::BteqzX16, Mips::SltuRxRy16, MI, BB);
+  case Mips::BtnezT8CmpX16:
+    return emitFEXT_T8I816_ins(Mips::BtnezX16, Mips::CmpRxRy16, MI, BB);
+  case Mips::BtnezT8SltX16:
+    return emitFEXT_T8I816_ins(Mips::BtnezX16, Mips::SltRxRy16, MI, BB);
+  case Mips::BtnezT8SltuX16:
+    // TBD: figure out a way to get this or remove the instruction
+    // altogether.
+    return emitFEXT_T8I816_ins(Mips::BtnezX16, Mips::SltuRxRy16, MI, BB);
+  case Mips::BteqzT8CmpiX16: return emitFEXT_T8I8I16_ins(
+    Mips::BteqzX16, Mips::CmpiRxImm16, Mips::CmpiRxImmX16, MI, BB);
+  case Mips::BteqzT8SltiX16: return emitFEXT_T8I8I16_ins(
+    Mips::BteqzX16, Mips::SltiRxImm16, Mips::SltiRxImmX16, MI, BB);
+  case Mips::BteqzT8SltiuX16: return emitFEXT_T8I8I16_ins(
+    Mips::BteqzX16, Mips::SltiuRxImm16, Mips::SltiuRxImmX16, MI, BB);
+  case Mips::BtnezT8CmpiX16: return emitFEXT_T8I8I16_ins(
+    Mips::BtnezX16, Mips::CmpiRxImm16, Mips::CmpiRxImmX16, MI, BB);
+  case Mips::BtnezT8SltiX16: return emitFEXT_T8I8I16_ins(
+    Mips::BtnezX16, Mips::SltiRxImm16, Mips::SltiRxImmX16, MI, BB);
+  case Mips::BtnezT8SltiuX16: return emitFEXT_T8I8I16_ins(
+    Mips::BtnezX16, Mips::SltiuRxImm16, Mips::SltiuRxImmX16, MI, BB);
+    break;
+  case Mips::SltCCRxRy16:
+    return emitFEXT_CCRX16_ins(Mips::SltRxRy16, MI, BB);
+    break;
+  case Mips::SltiCCRxImmX16:
+    return emitFEXT_CCRXI16_ins
+      (Mips::SltiRxImm16, Mips::SltiRxImmX16, MI, BB);
+  case Mips::SltiuCCRxImmX16:
+    return emitFEXT_CCRXI16_ins
+      (Mips::SltiuRxImm16, Mips::SltiuRxImmX16, MI, BB);
+  case Mips::SltuCCRxRy16:
+    return emitFEXT_CCRX16_ins
+      (Mips::SltuRxRy16, MI, BB);
+  }
+}
+
+bool Mips16TargetLowering::
+isEligibleForTailCallOptimization(const MipsCC &MipsCCInfo,
+                                  unsigned NextStackOffset,
+                                  const MipsFunctionInfo& FI) const {
+  // No tail call optimization for mips16.
+  return false;
+}
+
+void Mips16TargetLowering::setMips16LibcallName
+  (RTLIB::Libcall L, const char *Name) {
+  setLibcallName(L, Name);
+  NoHelperNeeded.insert(Name);
+}
+
+void Mips16TargetLowering::setMips16HardFloatLibCalls() {
+  setMips16LibcallName(RTLIB::ADD_F32, "__mips16_addsf3");
+  setMips16LibcallName(RTLIB::ADD_F64, "__mips16_adddf3");
+  setMips16LibcallName(RTLIB::SUB_F32, "__mips16_subsf3");
+  setMips16LibcallName(RTLIB::SUB_F64, "__mips16_subdf3");
+  setMips16LibcallName(RTLIB::MUL_F32, "__mips16_mulsf3");
+  setMips16LibcallName(RTLIB::MUL_F64, "__mips16_muldf3");
+  setMips16LibcallName(RTLIB::DIV_F32, "__mips16_divsf3");
+  setMips16LibcallName(RTLIB::DIV_F64, "__mips16_divdf3");
+  setMips16LibcallName(RTLIB::FPEXT_F32_F64, "__mips16_extendsfdf2");
+  setMips16LibcallName(RTLIB::FPROUND_F64_F32, "__mips16_truncdfsf2");
+  setMips16LibcallName(RTLIB::FPTOSINT_F32_I32, "__mips16_fix_truncsfsi");
+  setMips16LibcallName(RTLIB::FPTOSINT_F64_I32, "__mips16_fix_truncdfsi");
+  setMips16LibcallName(RTLIB::SINTTOFP_I32_F32, "__mips16_floatsisf");
+  setMips16LibcallName(RTLIB::SINTTOFP_I32_F64, "__mips16_floatsidf");
+  setMips16LibcallName(RTLIB::UINTTOFP_I32_F32, "__mips16_floatunsisf");
+  setMips16LibcallName(RTLIB::UINTTOFP_I32_F64, "__mips16_floatunsidf");
+  setMips16LibcallName(RTLIB::OEQ_F32, "__mips16_eqsf2");
+  setMips16LibcallName(RTLIB::OEQ_F64, "__mips16_eqdf2");
+  setMips16LibcallName(RTLIB::UNE_F32, "__mips16_nesf2");
+  setMips16LibcallName(RTLIB::UNE_F64, "__mips16_nedf2");
+  setMips16LibcallName(RTLIB::OGE_F32, "__mips16_gesf2");
+  setMips16LibcallName(RTLIB::OGE_F64, "__mips16_gedf2");
+  setMips16LibcallName(RTLIB::OLT_F32, "__mips16_ltsf2");
+  setMips16LibcallName(RTLIB::OLT_F64, "__mips16_ltdf2");
+  setMips16LibcallName(RTLIB::OLE_F32, "__mips16_lesf2");
+  setMips16LibcallName(RTLIB::OLE_F64, "__mips16_ledf2");
+  setMips16LibcallName(RTLIB::OGT_F32, "__mips16_gtsf2");
+  setMips16LibcallName(RTLIB::OGT_F64, "__mips16_gtdf2");
+  setMips16LibcallName(RTLIB::UO_F32, "__mips16_unordsf2");
+  setMips16LibcallName(RTLIB::UO_F64, "__mips16_unorddf2");
+  setMips16LibcallName(RTLIB::O_F32, "__mips16_unordsf2");
+  setMips16LibcallName(RTLIB::O_F64, "__mips16_unorddf2");
+}
+
+
+//
+// The Mips16 hard float is a crazy quilt inherited from gcc. I have a much
+// cleaner way to do all of this but it will have to wait until the traditional
+// gcc mechanism is completed.
+//
+// For Pic, in order for Mips16 code to call Mips32 code which according the abi
+// have either arguments or returned values placed in floating point registers,
+// we use a set of helper functions. (This includes functions which return type
+//  complex which on Mips are returned in a pair of floating point registers).
+//
+// This is an encoding that we inherited from gcc.
+// In Mips traditional O32, N32 ABI, floating point numbers are passed in
+// floating point argument registers 1,2 only when the first and optionally
+// the second arguments are float (sf) or double (df).
+// For Mips16 we are only concerned with the situations where floating point
+// arguments are being passed in floating point registers by the ABI, because
+// Mips16 mode code cannot execute floating point instructions to load those
+// values and hence helper functions are needed.
+// The possibilities are (), (sf), (sf, sf), (sf, df), (df), (df, sf), (df, df)
+// the helper function suffixs for these are:
+//                        0,  1,    5,        9,         2,   6,        10
+// this suffix can then be calculated as follows:
+// for a given argument Arg:
+//     Arg1x, Arg2x = 1 :  Arg is sf
+//                    2 :  Arg is df
+//                    0:   Arg is neither sf or df
+// So this stub is the string for number Arg1x + Arg2x*4.
+// However not all numbers between 0 and 10 are possible, we check anyway and
+// assert if the impossible exists.
+//
+
+unsigned int Mips16TargetLowering::getMips16HelperFunctionStubNumber
+  (ArgListTy &Args) const {
+  unsigned int resultNum = 0;
+  if (Args.size() >= 1) {
+    Type *t = Args[0].Ty;
+    if (t->isFloatTy()) {
+      resultNum = 1;
+    }
+    else if (t->isDoubleTy()) {
+      resultNum = 2;
+    }
+  }
+  if (resultNum) {
+    if (Args.size() >=2) {
+      Type *t = Args[1].Ty;
+      if (t->isFloatTy()) {
+        resultNum += 4;
+      }
+      else if (t->isDoubleTy()) {
+        resultNum += 8;
+      }
+    }
+  }
+  return resultNum;
+}
+
+//
+// prefixs are attached to stub numbers depending on the return type .
+// return type: float  sf_
+//              double df_
+//              single complex sc_
+//              double complext dc_
+//              others  NO PREFIX
+//
+//
+// The full name of a helper function is__mips16_call_stub +
+//    return type dependent prefix + stub number
+//
+//
+// This is something that probably should be in a different source file and
+// perhaps done differently but my main purpose is to not waste runtime
+// on something that we can enumerate in the source. Another possibility is
+// to have a python script to generate these mapping tables. This will do
+// for now. There are a whole series of helper function mapping arrays, one
+// for each return type class as outlined above. There there are 11 possible
+//  entries. Ones with 0 are ones which should never be selected
+//
+// All the arrays are similar except for ones which return neither
+// sf, df, sc, dc, in which only care about ones which have sf or df as a
+// first parameter.
+//
+#define P_ "__mips16_call_stub_"
+#define MAX_STUB_NUMBER 10
+#define T1 P "1", P "2", 0, 0, P "5", P "6", 0, 0, P "9", P "10"
+#define T P "0" , T1
+#define P P_
+static char const * vMips16Helper[MAX_STUB_NUMBER+1] =
+  {0, T1 };
+#undef P
+#define P P_ "sf_"
+static char const * sfMips16Helper[MAX_STUB_NUMBER+1] =
+  { T };
+#undef P
+#define P P_ "df_"
+static char const * dfMips16Helper[MAX_STUB_NUMBER+1] =
+  { T };
+#undef P
+#define P P_ "sc_"
+static char const * scMips16Helper[MAX_STUB_NUMBER+1] =
+  { T };
+#undef P
+#define P P_ "dc_"
+static char const * dcMips16Helper[MAX_STUB_NUMBER+1] =
+  { T };
+#undef P
+#undef P_
+
+
+const char* Mips16TargetLowering::
+  getMips16HelperFunction
+    (Type* RetTy, ArgListTy &Args, bool &needHelper) const {
+  const unsigned int stubNum = getMips16HelperFunctionStubNumber(Args);
+#ifndef NDEBUG
+  const unsigned int maxStubNum = 10;
+  assert(stubNum <= maxStubNum);
+  const bool validStubNum[maxStubNum+1] =
+    {true, true, true, false, false, true, true, false, false, true, true};
+  assert(validStubNum[stubNum]);
+#endif
+  const char *result;
+  if (RetTy->isFloatTy()) {
+    result = sfMips16Helper[stubNum];
+  }
+  else if (RetTy ->isDoubleTy()) {
+    result = dfMips16Helper[stubNum];
+  }
+  else if (RetTy->isStructTy()) {
+    // check if it's complex
+    if (RetTy->getNumContainedTypes() == 2) {
+      if ((RetTy->getContainedType(0)->isFloatTy()) &&
+          (RetTy->getContainedType(1)->isFloatTy())) {
+        result = scMips16Helper[stubNum];
+      }
+      else if ((RetTy->getContainedType(0)->isDoubleTy()) &&
+               (RetTy->getContainedType(1)->isDoubleTy())) {
+        result = dcMips16Helper[stubNum];
+      }
+      else {
+        llvm_unreachable("Uncovered condition");
+      }
+    }
+    else {
+      llvm_unreachable("Uncovered condition");
+    }
+  }
+  else {
+    if (stubNum == 0) {
+      needHelper = false;
+      return "";
+    }
+    result = vMips16Helper[stubNum];
+  }
+  needHelper = true;
+  return result;
+}
+
+void Mips16TargetLowering::
+getOpndList(SmallVectorImpl<SDValue> &Ops,
+            std::deque< std::pair<unsigned, SDValue> > &RegsToPass,
+            bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage,
+            CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const {
+  SelectionDAG &DAG = CLI.DAG;
+  const char* Mips16HelperFunction = 0;
+  bool NeedMips16Helper = false;
+
+  if (getTargetMachine().Options.UseSoftFloat && Mips16HardFloat) {
+    //
+    // currently we don't have symbols tagged with the mips16 or mips32
+    // qualifier so we will assume that we don't know what kind it is.
+    // and generate the helper
+    //
+    bool LookupHelper = true;
+    if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(CLI.Callee)) {
+      if (NoHelperNeeded.find(S->getSymbol()) != NoHelperNeeded.end()) {
+        LookupHelper = false;
+      }
+    }
+    if (LookupHelper) Mips16HelperFunction =
+      getMips16HelperFunction(CLI.RetTy, CLI.Args, NeedMips16Helper);
+
+  }
+
+  SDValue JumpTarget = Callee;
+
+  // T9 should contain the address of the callee function if
+  // -reloction-model=pic or it is an indirect call.
+  if (IsPICCall || !GlobalOrExternal) {
+    unsigned V0Reg = Mips::V0;
+    if (NeedMips16Helper) {
+      RegsToPass.push_front(std::make_pair(V0Reg, Callee));
+      JumpTarget = DAG.getExternalSymbol(Mips16HelperFunction, getPointerTy());
+      JumpTarget = getAddrGlobal(JumpTarget, DAG, MipsII::MO_GOT);
+    } else
+      RegsToPass.push_front(std::make_pair((unsigned)Mips::T9, Callee));
+  }
+
+  Ops.push_back(JumpTarget);
+
+  MipsTargetLowering::getOpndList(Ops, RegsToPass, IsPICCall, GlobalOrExternal,
+                                  InternalLinkage, CLI, Callee, Chain);
+}
+
+MachineBasicBlock *Mips16TargetLowering::
+emitSel16(unsigned Opc, MachineInstr *MI, MachineBasicBlock *BB) const {
+  if (DontExpandCondPseudos16)
+    return BB;
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  DebugLoc DL = MI->getDebugLoc();
+  // To "insert" a SELECT_CC instruction, we actually have to insert the
+  // diamond control-flow pattern.  The incoming instruction knows the
+  // destination vreg to set, the condition code register to branch on, the
+  // true/false values to select between, and a branch opcode to use.
+  const BasicBlock *LLVM_BB = BB->getBasicBlock();
+  MachineFunction::iterator It = BB;
+  ++It;
+
+  //  thisMBB:
+  //  ...
+  //   TrueVal = ...
+  //   setcc r1, r2, r3
+  //   bNE   r1, r0, copy1MBB
+  //   fallthrough --> copy0MBB
+  MachineBasicBlock *thisMBB  = BB;
+  MachineFunction *F = BB->getParent();
+  MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *sinkMBB  = F->CreateMachineBasicBlock(LLVM_BB);
+  F->insert(It, copy0MBB);
+  F->insert(It, sinkMBB);
+
+  // Transfer the remainder of BB and its successor edges to sinkMBB.
+  sinkMBB->splice(sinkMBB->begin(), BB,
+                  llvm::next(MachineBasicBlock::iterator(MI)),
+                  BB->end());
+  sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
+
+  // Next, add the true and fallthrough blocks as its successors.
+  BB->addSuccessor(copy0MBB);
+  BB->addSuccessor(sinkMBB);
+
+  BuildMI(BB, DL, TII->get(Opc)).addReg(MI->getOperand(3).getReg())
+    .addMBB(sinkMBB);
+
+  //  copy0MBB:
+  //   %FalseValue = ...
+  //   # fallthrough to sinkMBB
+  BB = copy0MBB;
+
+  // Update machine-CFG edges
+  BB->addSuccessor(sinkMBB);
+
+  //  sinkMBB:
+  //   %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ]
+  //  ...
+  BB = sinkMBB;
+
+  BuildMI(*BB, BB->begin(), DL,
+          TII->get(Mips::PHI), MI->getOperand(0).getReg())
+    .addReg(MI->getOperand(1).getReg()).addMBB(thisMBB)
+    .addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB);
+
+  MI->eraseFromParent();   // The pseudo instruction is gone now.
+  return BB;
+}
+
+MachineBasicBlock *Mips16TargetLowering::emitSelT16
+  (unsigned Opc1, unsigned Opc2,
+   MachineInstr *MI, MachineBasicBlock *BB) const {
+  if (DontExpandCondPseudos16)
+    return BB;
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  DebugLoc DL = MI->getDebugLoc();
+  // To "insert" a SELECT_CC instruction, we actually have to insert the
+  // diamond control-flow pattern.  The incoming instruction knows the
+  // destination vreg to set, the condition code register to branch on, the
+  // true/false values to select between, and a branch opcode to use.
+  const BasicBlock *LLVM_BB = BB->getBasicBlock();
+  MachineFunction::iterator It = BB;
+  ++It;
+
+  //  thisMBB:
+  //  ...
+  //   TrueVal = ...
+  //   setcc r1, r2, r3
+  //   bNE   r1, r0, copy1MBB
+  //   fallthrough --> copy0MBB
+  MachineBasicBlock *thisMBB  = BB;
+  MachineFunction *F = BB->getParent();
+  MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *sinkMBB  = F->CreateMachineBasicBlock(LLVM_BB);
+  F->insert(It, copy0MBB);
+  F->insert(It, sinkMBB);
+
+  // Transfer the remainder of BB and its successor edges to sinkMBB.
+  sinkMBB->splice(sinkMBB->begin(), BB,
+                  llvm::next(MachineBasicBlock::iterator(MI)),
+                  BB->end());
+  sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
+
+  // Next, add the true and fallthrough blocks as its successors.
+  BB->addSuccessor(copy0MBB);
+  BB->addSuccessor(sinkMBB);
+
+  BuildMI(BB, DL, TII->get(Opc2)).addReg(MI->getOperand(3).getReg())
+    .addReg(MI->getOperand(4).getReg());
+  BuildMI(BB, DL, TII->get(Opc1)).addMBB(sinkMBB);
+
+  //  copy0MBB:
+  //   %FalseValue = ...
+  //   # fallthrough to sinkMBB
+  BB = copy0MBB;
+
+  // Update machine-CFG edges
+  BB->addSuccessor(sinkMBB);
+
+  //  sinkMBB:
+  //   %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ]
+  //  ...
+  BB = sinkMBB;
+
+  BuildMI(*BB, BB->begin(), DL,
+          TII->get(Mips::PHI), MI->getOperand(0).getReg())
+    .addReg(MI->getOperand(1).getReg()).addMBB(thisMBB)
+    .addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB);
+
+  MI->eraseFromParent();   // The pseudo instruction is gone now.
+  return BB;
+
+}
+
+MachineBasicBlock *Mips16TargetLowering::emitSeliT16
+  (unsigned Opc1, unsigned Opc2,
+   MachineInstr *MI, MachineBasicBlock *BB) const {
+  if (DontExpandCondPseudos16)
+    return BB;
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  DebugLoc DL = MI->getDebugLoc();
+  // To "insert" a SELECT_CC instruction, we actually have to insert the
+  // diamond control-flow pattern.  The incoming instruction knows the
+  // destination vreg to set, the condition code register to branch on, the
+  // true/false values to select between, and a branch opcode to use.
+  const BasicBlock *LLVM_BB = BB->getBasicBlock();
+  MachineFunction::iterator It = BB;
+  ++It;
+
+  //  thisMBB:
+  //  ...
+  //   TrueVal = ...
+  //   setcc r1, r2, r3
+  //   bNE   r1, r0, copy1MBB
+  //   fallthrough --> copy0MBB
+  MachineBasicBlock *thisMBB  = BB;
+  MachineFunction *F = BB->getParent();
+  MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *sinkMBB  = F->CreateMachineBasicBlock(LLVM_BB);
+  F->insert(It, copy0MBB);
+  F->insert(It, sinkMBB);
+
+  // Transfer the remainder of BB and its successor edges to sinkMBB.
+  sinkMBB->splice(sinkMBB->begin(), BB,
+                  llvm::next(MachineBasicBlock::iterator(MI)),
+                  BB->end());
+  sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
+
+  // Next, add the true and fallthrough blocks as its successors.
+  BB->addSuccessor(copy0MBB);
+  BB->addSuccessor(sinkMBB);
+
+  BuildMI(BB, DL, TII->get(Opc2)).addReg(MI->getOperand(3).getReg())
+    .addImm(MI->getOperand(4).getImm());
+  BuildMI(BB, DL, TII->get(Opc1)).addMBB(sinkMBB);
+
+  //  copy0MBB:
+  //   %FalseValue = ...
+  //   # fallthrough to sinkMBB
+  BB = copy0MBB;
+
+  // Update machine-CFG edges
+  BB->addSuccessor(sinkMBB);
+
+  //  sinkMBB:
+  //   %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ]
+  //  ...
+  BB = sinkMBB;
+
+  BuildMI(*BB, BB->begin(), DL,
+          TII->get(Mips::PHI), MI->getOperand(0).getReg())
+    .addReg(MI->getOperand(1).getReg()).addMBB(thisMBB)
+    .addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB);
+
+  MI->eraseFromParent();   // The pseudo instruction is gone now.
+  return BB;
+
+}
+
+MachineBasicBlock
+  *Mips16TargetLowering::emitFEXT_T8I816_ins(unsigned BtOpc, unsigned CmpOpc,
+                                             MachineInstr *MI,
+                                             MachineBasicBlock *BB) const {
+  if (DontExpandCondPseudos16)
+    return BB;
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  unsigned regX = MI->getOperand(0).getReg();
+  unsigned regY = MI->getOperand(1).getReg();
+  MachineBasicBlock *target = MI->getOperand(2).getMBB();
+  BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(CmpOpc)).addReg(regX).addReg(regY);
+  BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(BtOpc)).addMBB(target);
+  MI->eraseFromParent();   // The pseudo instruction is gone now.
+  return BB;
+}
+
+MachineBasicBlock *Mips16TargetLowering::emitFEXT_T8I8I16_ins(
+  unsigned BtOpc, unsigned CmpiOpc, unsigned CmpiXOpc,
+  MachineInstr *MI,  MachineBasicBlock *BB) const {
+  if (DontExpandCondPseudos16)
+    return BB;
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  unsigned regX = MI->getOperand(0).getReg();
+  int64_t imm = MI->getOperand(1).getImm();
+  MachineBasicBlock *target = MI->getOperand(2).getMBB();
+  unsigned CmpOpc;
+  if (isUInt<8>(imm))
+    CmpOpc = CmpiOpc;
+  else if (isUInt<16>(imm))
+    CmpOpc = CmpiXOpc;
+  else
+    llvm_unreachable("immediate field not usable");
+  BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(CmpOpc)).addReg(regX).addImm(imm);
+  BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(BtOpc)).addMBB(target);
+  MI->eraseFromParent();   // The pseudo instruction is gone now.
+  return BB;
+}
+
+static unsigned Mips16WhichOp8uOr16simm
+  (unsigned shortOp, unsigned longOp, int64_t Imm) {
+  if (isUInt<8>(Imm))
+    return shortOp;
+  else if (isInt<16>(Imm))
+    return longOp;
+  else
+    llvm_unreachable("immediate field not usable");
+}
+
+MachineBasicBlock *Mips16TargetLowering::emitFEXT_CCRX16_ins(
+  unsigned SltOpc,
+  MachineInstr *MI,  MachineBasicBlock *BB) const {
+  if (DontExpandCondPseudos16)
+    return BB;
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  unsigned CC = MI->getOperand(0).getReg();
+  unsigned regX = MI->getOperand(1).getReg();
+  unsigned regY = MI->getOperand(2).getReg();
+  BuildMI(*BB, MI, MI->getDebugLoc(),
+		  TII->get(SltOpc)).addReg(regX).addReg(regY);
+  BuildMI(*BB, MI, MI->getDebugLoc(),
+          TII->get(Mips::MoveR3216), CC).addReg(Mips::T8);
+  MI->eraseFromParent();   // The pseudo instruction is gone now.
+  return BB;
+}
+
+MachineBasicBlock *Mips16TargetLowering::emitFEXT_CCRXI16_ins(
+  unsigned SltiOpc, unsigned SltiXOpc,
+  MachineInstr *MI,  MachineBasicBlock *BB )const {
+  if (DontExpandCondPseudos16)
+    return BB;
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  unsigned CC = MI->getOperand(0).getReg();
+  unsigned regX = MI->getOperand(1).getReg();
+  int64_t Imm = MI->getOperand(2).getImm();
+  unsigned SltOpc = Mips16WhichOp8uOr16simm(SltiOpc, SltiXOpc, Imm);
+  BuildMI(*BB, MI, MI->getDebugLoc(),
+          TII->get(SltOpc)).addReg(regX).addImm(Imm);
+  BuildMI(*BB, MI, MI->getDebugLoc(),
+          TII->get(Mips::MoveR3216), CC).addReg(Mips::T8);
+  MI->eraseFromParent();   // The pseudo instruction is gone now.
+  return BB;
+
+}
diff --git a/lib/Target/Mips/Mips16ISelLowering.h b/lib/Target/Mips/Mips16ISelLowering.h
new file mode 100644
index 000000000000..b23e2a1f37db
--- /dev/null
+++ b/lib/Target/Mips/Mips16ISelLowering.h
@@ -0,0 +1,80 @@
+//===-- Mips16ISelLowering.h - Mips16 DAG Lowering Interface ----*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Subclass of MipsTargetLowering specialized for mips16.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef Mips16ISELLOWERING_H
+#define Mips16ISELLOWERING_H
+
+#include "MipsISelLowering.h"
+
+namespace llvm {
+  class Mips16TargetLowering : public MipsTargetLowering  {
+  public:
+    explicit Mips16TargetLowering(MipsTargetMachine &TM);
+
+    virtual bool allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const;
+
+    virtual MachineBasicBlock *
+    EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *MBB) const;
+
+  private:
+    virtual bool
+    isEligibleForTailCallOptimization(const MipsCC &MipsCCInfo,
+                                      unsigned NextStackOffset,
+                                      const MipsFunctionInfo& FI) const;
+
+    void setMips16LibcallName(RTLIB::Libcall, const char *Name);
+
+    void setMips16HardFloatLibCalls();
+
+    unsigned int
+      getMips16HelperFunctionStubNumber(ArgListTy &Args) const;
+
+    const char *getMips16HelperFunction
+      (Type* RetTy, ArgListTy &Args, bool &needHelper) const;
+
+    virtual void
+    getOpndList(SmallVectorImpl<SDValue> &Ops,
+                std::deque< std::pair<unsigned, SDValue> > &RegsToPass,
+                bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage,
+                CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const;
+
+    MachineBasicBlock *emitSel16(unsigned Opc, MachineInstr *MI,
+                                 MachineBasicBlock *BB) const;
+
+    MachineBasicBlock *emitSeliT16(unsigned Opc1, unsigned Opc2,
+                                   MachineInstr *MI,
+                                   MachineBasicBlock *BB) const;
+
+    MachineBasicBlock *emitSelT16(unsigned Opc1, unsigned Opc2,
+                                  MachineInstr *MI,
+                                  MachineBasicBlock *BB) const;
+
+    MachineBasicBlock *emitFEXT_T8I816_ins(unsigned BtOpc, unsigned CmpOpc,
+                                           MachineInstr *MI,
+                                           MachineBasicBlock *BB) const;
+
+    MachineBasicBlock *emitFEXT_T8I8I16_ins(
+      unsigned BtOpc, unsigned CmpiOpc, unsigned CmpiXOpc,
+      MachineInstr *MI,  MachineBasicBlock *BB) const;
+
+    MachineBasicBlock *emitFEXT_CCRX16_ins(
+      unsigned SltOpc,
+      MachineInstr *MI,  MachineBasicBlock *BB) const;
+
+    MachineBasicBlock *emitFEXT_CCRXI16_ins(
+      unsigned SltiOpc, unsigned SltiXOpc,
+      MachineInstr *MI,  MachineBasicBlock *BB )const;
+  };
+}
+
+#endif // Mips16ISELLOWERING_H
diff --git a/lib/Target/Mips/Mips16InstrFormats.td b/lib/Target/Mips/Mips16InstrFormats.td
index 61602b62fb44..4ff62ef3b6f9 100644
--- a/lib/Target/Mips/Mips16InstrFormats.td
+++ b/lib/Target/Mips/Mips16InstrFormats.td
@@ -29,45 +29,13 @@
 //
 //===----------------------------------------------------------------------===//
 
-// Format specifies the encoding used by the instruction.  This is part of the
-// ad-hoc solution used to emit machine instruction encodings by our machine
-// code emitter.
-//
-class Format16<bits<5> val> {
-  bits<5> Value = val;
-}
-
-def Pseudo16          : Format16<0>;
-def FrmI16            : Format16<1>;
-def FrmRI16           : Format16<2>;
-def FrmRR16           : Format16<3>;
-def FrmRRI16          : Format16<4>;
-def FrmRRR16          : Format16<5>;
-def FrmRRI_A16        : Format16<6>;
-def FrmSHIFT16        : Format16<7>;
-def FrmI8_TYPE16      : Format16<8>;
-def FrmI8_MOVR3216    : Format16<9>;
-def FrmI8_MOV32R16    : Format16<10>;
-def FrmI8_SVRS16      : Format16<11>;
-def FrmJAL16          : Format16<12>;
-def FrmJALX16         : Format16<13>;
-def FrmEXT_I16        : Format16<14>;
-def FrmASMACRO16      : Format16<15>;
-def FrmEXT_RI16       : Format16<16>;
-def FrmEXT_RRI16      : Format16<17>;
-def FrmEXT_RRI_A16    : Format16<18>;
-def FrmEXT_SHIFT16    : Format16<19>;
-def FrmEXT_I816       : Format16<20>;
-def FrmEXT_I8_SVRS16  : Format16<21>;
-def FrmOther16        : Format16<22>; // Instruction w/ a custom format
 
 // Base class for Mips 16 Format
 // This class does not depend on the instruction size
 //
 class MipsInst16_Base<dag outs, dag ins, string asmstr, list<dag> pattern,
-                      InstrItinClass itin, Format16 f>: Instruction
+                      InstrItinClass itin>: Instruction
 {
-  Format16 Form = f;
 
   let Namespace = "Mips";
 
@@ -78,14 +46,6 @@ class MipsInst16_Base<dag outs, dag ins, string asmstr, list<dag> pattern,
   let Pattern     = pattern;
   let Itinerary   = itin;
 
-  //
-  // Attributes specific to Mips instructions...
-  //
-  bits<5> FormBits = Form.Value;
-
-  // TSFlags layout should be kept in sync with MipsInstrInfo.h.
-  let TSFlags{4-0}   = FormBits;
-
   let Predicates = [InMips16Mode];
 }
 
@@ -93,30 +53,35 @@ class MipsInst16_Base<dag outs, dag ins, string asmstr, list<dag> pattern,
 // Generic Mips 16 Format
 //
 class MipsInst16<dag outs, dag ins, string asmstr, list<dag> pattern,
-                 InstrItinClass itin, Format16 f>:
-  MipsInst16_Base<outs, ins, asmstr, pattern, itin, f>
+                 InstrItinClass itin>:
+  MipsInst16_Base<outs, ins, asmstr, pattern, itin>
 {
   field bits<16> Inst;
   bits<5> Opcode = 0;
 
   // Top 5 bits are the 'opcode' field
   let Inst{15-11} = Opcode;
+  
+  let Size=2;
+  field bits<16> SoftFail = 0;
 }
 
 //
 // For 32 bit extended instruction forms.
 //
 class MipsInst16_32<dag outs, dag ins, string asmstr, list<dag> pattern,
-                    InstrItinClass itin, Format16 f>:
-  MipsInst16_Base<outs, ins, asmstr, pattern, itin, f>
+                    InstrItinClass itin>:
+  MipsInst16_Base<outs, ins, asmstr, pattern, itin>
 {
   field bits<32> Inst;
-
+  
+  let Size=4;
+  field bits<32> SoftFail = 0;
 }
 
 class MipsInst16_EXTEND<dag outs, dag ins, string asmstr, list<dag> pattern,
-                        InstrItinClass itin, Format16 f>:
-  MipsInst16_32<outs, ins, asmstr, pattern, itin, f>
+                        InstrItinClass itin>:
+  MipsInst16_32<outs, ins, asmstr, pattern, itin>
 {
   let Inst{31-27} = 0b11110;
 }
@@ -125,7 +90,7 @@ class MipsInst16_EXTEND<dag outs, dag ins, string asmstr, list<dag> pattern,
 
 // Mips Pseudo Instructions Format
 class MipsPseudo16<dag outs, dag ins, string asmstr, list<dag> pattern>:
-  MipsInst16<outs, ins, asmstr, pattern, IIPseudo, Pseudo16> {
+  MipsInst16<outs, ins, asmstr, pattern, IIPseudo> {
   let isCodeGenOnly = 1;
   let isPseudo = 1;
 }
@@ -137,7 +102,7 @@ class MipsPseudo16<dag outs, dag ins, string asmstr, list<dag> pattern>:
 
 class FI16<bits<5> op, dag outs, dag ins, string asmstr, list<dag> pattern,
            InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmI16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<11> imm11;
 
@@ -152,7 +117,7 @@ class FI16<bits<5> op, dag outs, dag ins, string asmstr, list<dag> pattern,
 
 class FRI16<bits<5> op, dag outs, dag ins, string asmstr,
             list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmRI16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<3>  rx;
   bits<8>   imm8;
@@ -169,7 +134,7 @@ class FRI16<bits<5> op, dag outs, dag ins, string asmstr,
 
 class FRR16<bits<5> _funct, dag outs, dag ins, string asmstr,
             list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmRR16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<3>  rx;
   bits<3>  ry;
@@ -188,7 +153,7 @@ class FRR16<bits<5> _funct, dag outs, dag ins, string asmstr,
 //
 class FRR_SF16<bits<5> _funct, bits<3> _subfunct, dag outs, dag ins,
                string asmstr, list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmRR16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<3>  rx;
   bits<3>  subfunct;
@@ -208,7 +173,7 @@ class FRR_SF16<bits<5> _funct, bits<3> _subfunct, dag outs, dag ins,
 //
 class FC16<bits<5> _funct, dag outs, dag ins, string asmstr,
            list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmRR16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<6>  _code;  // code is a keyword in tablegen
   bits<5>  funct;
@@ -226,7 +191,7 @@ class FC16<bits<5> _funct, dag outs, dag ins, string asmstr,
 class FRR16_JALRC<bits<1> _nd, bits<1> _l, bits<1> r_a,
                   dag outs, dag ins, string asmstr,
                   list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmRR16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<3>  rx;
   bits<1>  nd;
@@ -252,7 +217,7 @@ class FRR16_JALRC<bits<1> _nd, bits<1> _l, bits<1> r_a,
 
 class FRRI16<bits<5> op, dag outs, dag ins, string asmstr,
              list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmRRI16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<3>  rx;
   bits<3>  ry;
@@ -272,7 +237,7 @@ class FRRI16<bits<5> op, dag outs, dag ins, string asmstr,
 
 class FRRR16<bits<2> _f, dag outs, dag ins, string asmstr,
              list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmRRR16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<3>  rx;
   bits<3>  ry;
@@ -294,7 +259,7 @@ class FRRR16<bits<2> _f, dag outs, dag ins, string asmstr,
 
 class FRRI_A16<bits<1> _f, dag outs, dag ins, string asmstr,
                list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmRRI_A16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<3>  rx;
   bits<3>  ry;
@@ -316,7 +281,7 @@ class FRRI_A16<bits<1> _f, dag outs, dag ins, string asmstr,
 
 class FSHIFT16<bits<2> _f, dag outs, dag ins, string asmstr,
                list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmSHIFT16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<3>  rx;
   bits<3>  ry;
@@ -338,7 +303,7 @@ class FSHIFT16<bits<2> _f, dag outs, dag ins, string asmstr,
 
 class FI816<bits<3> _func, dag outs, dag ins, string asmstr,
             list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmI8_TYPE16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<3>  func;
   bits<8>   imm8;
@@ -356,7 +321,7 @@ class FI816<bits<3> _func, dag outs, dag ins, string asmstr,
 
 class FI8_MOVR3216<dag outs, dag ins, string asmstr,
                    list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmI8_MOVR3216>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
 
   bits<4> ry;
@@ -378,7 +343,7 @@ class FI8_MOVR3216<dag outs, dag ins, string asmstr,
 
 class FI8_MOV32R16<dag outs, dag ins, string asmstr,
                    list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmI8_MOV32R16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
 
   bits<3>  func;
@@ -402,7 +367,7 @@ class FI8_MOV32R16<dag outs, dag ins, string asmstr,
 
 class FI8_SVRS16<bits<1> _s, dag outs, dag ins, string asmstr,
                  list<dag> pattern, InstrItinClass itin>:
-  MipsInst16<outs, ins, asmstr, pattern, itin, FrmI8_SVRS16>
+  MipsInst16<outs, ins, asmstr, pattern, itin>
 {
   bits<1> s;
   bits<1> ra = 0;
@@ -429,7 +394,7 @@ class FI8_SVRS16<bits<1> _s, dag outs, dag ins, string asmstr,
 
 class FJAL16<bits<1> _X, dag outs, dag ins, string asmstr,
              list<dag> pattern, InstrItinClass itin>:
-  MipsInst16_32<outs, ins, asmstr, pattern, itin, FrmJAL16>
+  MipsInst16_32<outs, ins, asmstr, pattern, itin>
 {
   bits<1> X;
   bits<26> imm26;
@@ -452,7 +417,7 @@ class FJAL16<bits<1> _X, dag outs, dag ins, string asmstr,
 
 class FEXT_I16<bits<5> _eop, dag outs, dag ins, string asmstr,
                list<dag> pattern, InstrItinClass itin>:
-  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin, FrmEXT_I16>
+  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin>
 {
   bits<16> imm16;
   bits<5> eop;
@@ -474,7 +439,7 @@ class FEXT_I16<bits<5> _eop, dag outs, dag ins, string asmstr,
 
 class FASMACRO16<dag outs, dag ins, string asmstr,
                  list<dag> pattern, InstrItinClass itin>:
-  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin, FrmASMACRO16>
+  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin>
 {
   bits<3> select;
   bits<3> p4;
@@ -503,7 +468,7 @@ class FASMACRO16<dag outs, dag ins, string asmstr,
 
 class FEXT_RI16<bits<5> _op, dag outs, dag ins, string asmstr,
                 list<dag> pattern, InstrItinClass itin>:
-  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin, FrmEXT_RI16>
+  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin>
 {
   bits<16> imm16;
   bits<5> op;
@@ -527,7 +492,7 @@ class FEXT_RI16<bits<5> _op, dag outs, dag ins, string asmstr,
 
 class FEXT_RRI16<bits<5> _op, dag outs, dag ins, string asmstr,
                  list<dag> pattern, InstrItinClass itin>:
-  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin, FrmEXT_RRI16>
+  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin>
 {
   bits<5> op;
   bits<16> imm16;
@@ -552,7 +517,7 @@ class FEXT_RRI16<bits<5> _op, dag outs, dag ins, string asmstr,
 
 class FEXT_RRI_A16<bits<1> _f, dag outs, dag ins, string asmstr,
                    list<dag> pattern, InstrItinClass itin>:
-  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin, FrmEXT_RRI_A16>
+  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin>
 {
   bits<15> imm15;
   bits<3> rx;
@@ -578,7 +543,7 @@ class FEXT_RRI_A16<bits<1> _f, dag outs, dag ins, string asmstr,
 
 class FEXT_SHIFT16<bits<2> _f, dag outs, dag ins, string asmstr,
                    list<dag> pattern, InstrItinClass itin>:
-  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin, FrmEXT_SHIFT16>
+  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin>
 {
   bits<6> sa6;
   bits<3> rx;
@@ -605,7 +570,7 @@ class FEXT_SHIFT16<bits<2> _f, dag outs, dag ins, string asmstr,
 
 class FEXT_I816<bits<3> _funct, dag outs, dag ins, string asmstr,
                 list<dag> pattern, InstrItinClass itin>:
-  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin, FrmEXT_I816>
+  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin>
 {
   bits<16> imm16;
   bits<5> I8;
@@ -630,7 +595,7 @@ class FEXT_I816<bits<3> _funct, dag outs, dag ins, string asmstr,
 
 class FEXT_I8_SVRS16<bits<1> s_, dag outs, dag ins, string asmstr,
                      list<dag> pattern, InstrItinClass itin>:
-  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin, FrmI8_SVRS16>
+  MipsInst16_EXTEND<outs, ins, asmstr, pattern, itin>
 {
   bits<3> xsregs =0;
   bits<8> framesize =0;
@@ -659,5 +624,3 @@ class FEXT_I8_SVRS16<bits<1> s_, dag outs, dag ins, string asmstr,
 
 }
 
-
-
diff --git a/lib/Target/Mips/Mips16InstrInfo.cpp b/lib/Target/Mips/Mips16InstrInfo.cpp
index 619646b3178a..17dd2c07967a 100644
--- a/lib/Target/Mips/Mips16InstrInfo.cpp
+++ b/lib/Target/Mips/Mips16InstrInfo.cpp
@@ -12,18 +12,29 @@
 //===----------------------------------------------------------------------===//
 
 #include "Mips16InstrInfo.h"
-#include "MipsTargetMachine.h"
-#include "MipsMachineFunction.h"
 #include "InstPrinter/MipsInstPrinter.h"
+#include "MipsMachineFunction.h"
+#include "MipsTargetMachine.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/StringRef.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
-#include "llvm/ADT/STLExtras.h"
-#include "llvm/ADT/StringRef.h"
 
 using namespace llvm;
 
+static cl::opt<bool> NeverUseSaveRestore(
+  "mips16-never-use-save-restore",
+  cl::init(false),
+  cl::desc("For testing ability to adjust stack pointer "
+           "without save/restore instruction"),
+  cl::Hidden);
+
+
 Mips16InstrInfo::Mips16InstrInfo(MipsTargetMachine &tm)
   : MipsInstrInfo(tm, Mips::BimmX16),
     RI(*tm.getSubtargetImpl(), *this) {}
@@ -87,10 +98,10 @@ void Mips16InstrInfo::copyPhysReg(MachineBasicBlock &MBB,
 }
 
 void Mips16InstrInfo::
-storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
-                    unsigned SrcReg, bool isKill, int FI,
-                    const TargetRegisterClass *RC,
-                    const TargetRegisterInfo *TRI) const {
+storeRegToStack(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
+                unsigned SrcReg, bool isKill, int FI,
+                const TargetRegisterClass *RC, const TargetRegisterInfo *TRI,
+                int64_t Offset) const {
   DebugLoc DL;
   if (I != MBB.end()) DL = I->getDebugLoc();
   MachineMemOperand *MMO = GetMemOperand(MBB, FI, MachineMemOperand::MOStore);
@@ -99,14 +110,13 @@ storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
     Opc = Mips::SwRxSpImmX16;
   assert(Opc && "Register class not handled!");
   BuildMI(MBB, I, DL, get(Opc)).addReg(SrcReg, getKillRegState(isKill))
-    .addFrameIndex(FI).addImm(0).addMemOperand(MMO);
+    .addFrameIndex(FI).addImm(Offset).addMemOperand(MMO);
 }
 
 void Mips16InstrInfo::
-loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
-                     unsigned DestReg, int FI,
-                     const TargetRegisterClass *RC,
-                     const TargetRegisterInfo *TRI) const {
+loadRegFromStack(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
+                 unsigned DestReg, int FI, const TargetRegisterClass *RC,
+                 const TargetRegisterInfo *TRI, int64_t Offset) const {
   DebugLoc DL;
   if (I != MBB.end()) DL = I->getDebugLoc();
   MachineMemOperand *MMO = GetMemOperand(MBB, FI, MachineMemOperand::MOLoad);
@@ -115,13 +125,12 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
   if (Mips::CPU16RegsRegClass.hasSubClassEq(RC))
     Opc = Mips::LwRxSpImmX16;
   assert(Opc && "Register class not handled!");
-  BuildMI(MBB, I, DL, get(Opc), DestReg).addFrameIndex(FI).addImm(0)
+  BuildMI(MBB, I, DL, get(Opc), DestReg).addFrameIndex(FI).addImm(Offset)
     .addMemOperand(MMO);
 }
 
 bool Mips16InstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
   MachineBasicBlock &MBB = *MI->getParent();
-
   switch(MI->getDesc().getOpcode()) {
   default:
     return false;
@@ -160,20 +169,215 @@ unsigned Mips16InstrInfo::GetOppositeBranchOpc(unsigned Opc) const {
   return 0;
 }
 
+// Adjust SP by FrameSize bytes. Save RA, S0, S1
+void Mips16InstrInfo::makeFrame(unsigned SP, int64_t FrameSize,
+                    MachineBasicBlock &MBB,
+                    MachineBasicBlock::iterator I) const {
+  DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
+  if (!NeverUseSaveRestore) {
+    if (isUInt<11>(FrameSize))
+      BuildMI(MBB, I, DL, get(Mips::SaveRaF16)).addImm(FrameSize);
+    else {
+      int Base = 2040; // should create template function like isUInt that
+                       // returns largest possible n bit unsigned integer
+      int64_t Remainder = FrameSize - Base;
+      BuildMI(MBB, I, DL, get(Mips::SaveRaF16)). addImm(Base);
+      if (isInt<16>(-Remainder))
+        BuildAddiuSpImm(MBB, I, -Remainder);
+      else
+        adjustStackPtrBig(SP, -Remainder, MBB, I, Mips::V0, Mips::V1);
+    }
+
+  }
+  else {
+    //
+    // sw ra, -4[sp]
+    // sw s1, -8[sp]
+    // sw s0, -12[sp]
+
+    MachineInstrBuilder MIB1 = BuildMI(MBB, I, DL, get(Mips::SwRxSpImmX16),
+                                       Mips::RA);
+    MIB1.addReg(Mips::SP);
+    MIB1.addImm(-4);
+    MachineInstrBuilder MIB2 = BuildMI(MBB, I, DL, get(Mips::SwRxSpImmX16),
+                                       Mips::S1);
+    MIB2.addReg(Mips::SP);
+    MIB2.addImm(-8);
+    MachineInstrBuilder MIB3 = BuildMI(MBB, I, DL, get(Mips::SwRxSpImmX16),
+                                       Mips::S0);
+    MIB3.addReg(Mips::SP);
+    MIB3.addImm(-12);
+    adjustStackPtrBig(SP, -FrameSize, MBB, I, Mips::V0, Mips::V1);
+  }
+}
+
+// Adjust SP by FrameSize bytes. Restore RA, S0, S1
+void Mips16InstrInfo::restoreFrame(unsigned SP, int64_t FrameSize,
+                                   MachineBasicBlock &MBB,
+                                   MachineBasicBlock::iterator I) const {
+  DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
+  if (!NeverUseSaveRestore) {
+    if (isUInt<11>(FrameSize))
+      BuildMI(MBB, I, DL, get(Mips::RestoreRaF16)).addImm(FrameSize);
+    else {
+      int Base = 2040; // should create template function like isUInt that
+                       // returns largest possible n bit unsigned integer
+      int64_t Remainder = FrameSize - Base;
+      if (isInt<16>(Remainder))
+        BuildAddiuSpImm(MBB, I, Remainder);
+      else
+        adjustStackPtrBig(SP, Remainder, MBB, I, Mips::A0, Mips::A1);
+      BuildMI(MBB, I, DL, get(Mips::RestoreRaF16)). addImm(Base);
+    }
+  }
+  else {
+    adjustStackPtrBig(SP, FrameSize, MBB, I, Mips::A0, Mips::A1);
+    // lw ra, -4[sp]
+    // lw s1, -8[sp]
+    // lw s0, -12[sp]
+    MachineInstrBuilder MIB1 = BuildMI(MBB, I, DL, get(Mips::LwRxSpImmX16),
+                                       Mips::A0);
+    MIB1.addReg(Mips::SP);
+    MIB1.addImm(-4);
+    MachineInstrBuilder MIB0 = BuildMI(MBB, I, DL, get(Mips::Move32R16),
+                                       Mips::RA);
+     MIB0.addReg(Mips::A0);
+    MachineInstrBuilder MIB2 = BuildMI(MBB, I, DL, get(Mips::LwRxSpImmX16),
+                                       Mips::S1);
+    MIB2.addReg(Mips::SP);
+    MIB2.addImm(-8);
+    MachineInstrBuilder MIB3 = BuildMI(MBB, I, DL, get(Mips::LwRxSpImmX16),
+                                       Mips::S0);
+    MIB3.addReg(Mips::SP);
+    MIB3.addImm(-12);
+  }
+
+}
+
+// Adjust SP by Amount bytes where bytes can be up to 32bit number.
+// This can only be called at times that we know that there is at least one free
+// register.
+// This is clearly safe at prologue and epilogue.
+//
+void Mips16InstrInfo::adjustStackPtrBig(unsigned SP, int64_t Amount,
+                                        MachineBasicBlock &MBB,
+                                        MachineBasicBlock::iterator I,
+                                        unsigned Reg1, unsigned Reg2) const {
+  DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
+//  MachineRegisterInfo &RegInfo = MBB.getParent()->getRegInfo();
+//  unsigned Reg1 = RegInfo.createVirtualRegister(&Mips::CPU16RegsRegClass);
+//  unsigned Reg2 = RegInfo.createVirtualRegister(&Mips::CPU16RegsRegClass);
+  //
+  // li reg1, constant
+  // move reg2, sp
+  // add reg1, reg1, reg2
+  // move sp, reg1
+  //
+  //
+  MachineInstrBuilder MIB1 = BuildMI(MBB, I, DL, get(Mips::LwConstant32), Reg1);
+  MIB1.addImm(Amount);
+  MachineInstrBuilder MIB2 = BuildMI(MBB, I, DL, get(Mips::MoveR3216), Reg2);
+  MIB2.addReg(Mips::SP, RegState::Kill);
+  MachineInstrBuilder MIB3 = BuildMI(MBB, I, DL, get(Mips::AdduRxRyRz16), Reg1);
+  MIB3.addReg(Reg1);
+  MIB3.addReg(Reg2, RegState::Kill);
+  MachineInstrBuilder MIB4 = BuildMI(MBB, I, DL, get(Mips::Move32R16),
+                                                     Mips::SP);
+  MIB4.addReg(Reg1, RegState::Kill);
+}
+
+void Mips16InstrInfo::adjustStackPtrBigUnrestricted(unsigned SP, int64_t Amount,
+                    MachineBasicBlock &MBB,
+                    MachineBasicBlock::iterator I) const {
+   assert(false && "adjust stack pointer amount exceeded");
+}
+
 /// Adjust SP by Amount bytes.
 void Mips16InstrInfo::adjustStackPtr(unsigned SP, int64_t Amount,
                                      MachineBasicBlock &MBB,
                                      MachineBasicBlock::iterator I) const {
-  DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
-  if (isInt<16>(Amount)) {
-    if (Amount < 0)
-      BuildMI(MBB, I, DL, get(Mips::SaveDecSpF16)). addImm(-Amount);
-    else if (Amount > 0)
-      BuildMI(MBB, I, DL, get(Mips::RestoreIncSpF16)).addImm(Amount);
+  if (isInt<16>(Amount))  // need to change to addiu sp, ....and isInt<16>
+    BuildAddiuSpImm(MBB, I, Amount);
+  else
+    adjustStackPtrBigUnrestricted(SP, Amount, MBB, I);
+}
+
+/// This function generates the sequence of instructions needed to get the
+/// result of adding register REG and immediate IMM.
+unsigned
+Mips16InstrInfo::loadImmediate(unsigned FrameReg,
+                               int64_t Imm, MachineBasicBlock &MBB,
+                               MachineBasicBlock::iterator II, DebugLoc DL,
+                               unsigned &NewImm) const {
+  //
+  // given original instruction is:
+  // Instr rx, T[offset] where offset is too big.
+  //
+  // lo = offset & 0xFFFF
+  // hi = ((offset >> 16) + (lo >> 15)) & 0xFFFF;
+  //
+  // let T = temporary register
+  // li T, hi
+  // shl T, 16
+  // add T, Rx, T
+  //
+  RegScavenger rs;
+  int32_t lo = Imm & 0xFFFF;
+  int32_t hi = ((Imm >> 16) + (lo >> 15)) & 0xFFFF;
+  NewImm = lo;
+  unsigned Reg =0;
+  unsigned SpReg = 0;
+  rs.enterBasicBlock(&MBB);
+  rs.forward(II);
+  //
+  // we use T0 for the first register, if we need to save something away.
+  // we use T1 for the second register, if we need to save something away.
+  //
+  unsigned FirstRegSaved =0, SecondRegSaved=0;
+  unsigned FirstRegSavedTo = 0, SecondRegSavedTo = 0;
+
+  Reg = rs.FindUnusedReg(&Mips::CPU16RegsRegClass);
+  if (Reg == 0) {
+    FirstRegSaved = Reg = Mips::V0;
+    FirstRegSavedTo = Mips::T0;
+    copyPhysReg(MBB, II, DL, FirstRegSavedTo, FirstRegSaved, true);
   }
   else
-    // not implemented for large values yet
-    assert(false && "adjust stack pointer amount exceeded");
+    rs.setUsed(Reg);
+  BuildMI(MBB, II, DL, get(Mips::LiRxImmX16), Reg).addImm(hi);
+  BuildMI(MBB, II, DL, get(Mips::SllX16), Reg).addReg(Reg).
+    addImm(16);
+  if (FrameReg == Mips::SP) {
+    SpReg = rs.FindUnusedReg(&Mips::CPU16RegsRegClass);
+    if (SpReg == 0) {
+      if (Reg != Mips::V1) {
+        SecondRegSaved = SpReg = Mips::V1;
+        SecondRegSavedTo = Mips::T1;
+      }
+      else {
+        SecondRegSaved = SpReg = Mips::V0;
+        SecondRegSavedTo = Mips::T0;
+      }
+      copyPhysReg(MBB, II, DL, SecondRegSavedTo, SecondRegSaved, true);
+    }
+    else
+      rs.setUsed(SpReg);
+
+    copyPhysReg(MBB, II, DL, SpReg, Mips::SP, false);
+    BuildMI(MBB, II, DL, get(Mips::  AdduRxRyRz16), Reg).addReg(SpReg)
+      .addReg(Reg);
+  }
+  else
+    BuildMI(MBB, II, DL, get(Mips::  AdduRxRyRz16), Reg).addReg(FrameReg)
+      .addReg(Reg, RegState::Kill);
+  if (FirstRegSaved || SecondRegSaved) {
+    II = llvm::next(II);
+    if (FirstRegSaved)
+      copyPhysReg(MBB, II, DL, FirstRegSaved, FirstRegSavedTo, true);
+    if (SecondRegSaved)
+      copyPhysReg(MBB, II, DL, SecondRegSaved, SecondRegSavedTo, true);
+  }
+  return Reg;
 }
 
 unsigned Mips16InstrInfo::GetAnalyzableBrOpc(unsigned Opc) const {
@@ -194,6 +398,20 @@ void Mips16InstrInfo::ExpandRetRA16(MachineBasicBlock &MBB,
   BuildMI(MBB, I, I->getDebugLoc(), get(Opc));
 }
 
+
+const MCInstrDesc &Mips16InstrInfo::AddiuSpImm(int64_t Imm) const {
+  if (validSpImm8(Imm))
+    return get(Mips::AddiuSpImm16);
+  else
+    return get(Mips::AddiuSpImmX16);
+}
+
+void Mips16InstrInfo::BuildAddiuSpImm
+  (MachineBasicBlock &MBB, MachineBasicBlock::iterator I, int64_t Imm) const {
+  DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
+  BuildMI(MBB, I, DL, AddiuSpImm(Imm)).addImm(Imm);
+}
+
 const MipsInstrInfo *llvm::createMips16InstrInfo(MipsTargetMachine &TM) {
   return new Mips16InstrInfo(TM);
 }
diff --git a/lib/Target/Mips/Mips16InstrInfo.h b/lib/Target/Mips/Mips16InstrInfo.h
index e06ccfe61c52..a77a9043bb17 100644
--- a/lib/Target/Mips/Mips16InstrInfo.h
+++ b/lib/Target/Mips/Mips16InstrInfo.h
@@ -14,8 +14,8 @@
 #ifndef MIPS16INSTRUCTIONINFO_H
 #define MIPS16INSTRUCTIONINFO_H
 
-#include "MipsInstrInfo.h"
 #include "Mips16RegisterInfo.h"
+#include "MipsInstrInfo.h"
 
 namespace llvm {
 
@@ -48,31 +48,75 @@ public:
                            unsigned DestReg, unsigned SrcReg,
                            bool KillSrc) const;
 
-  virtual void storeRegToStackSlot(MachineBasicBlock &MBB,
-                                   MachineBasicBlock::iterator MBBI,
-                                   unsigned SrcReg, bool isKill, int FrameIndex,
-                                   const TargetRegisterClass *RC,
-                                   const TargetRegisterInfo *TRI) const;
+  virtual void storeRegToStack(MachineBasicBlock &MBB,
+                               MachineBasicBlock::iterator MBBI,
+                               unsigned SrcReg, bool isKill, int FrameIndex,
+                               const TargetRegisterClass *RC,
+                               const TargetRegisterInfo *TRI,
+                               int64_t Offset) const;
 
-  virtual void loadRegFromStackSlot(MachineBasicBlock &MBB,
-                                    MachineBasicBlock::iterator MBBI,
-                                    unsigned DestReg, int FrameIndex,
-                                    const TargetRegisterClass *RC,
-                                    const TargetRegisterInfo *TRI) const;
+  virtual void loadRegFromStack(MachineBasicBlock &MBB,
+                                MachineBasicBlock::iterator MBBI,
+                                unsigned DestReg, int FrameIndex,
+                                const TargetRegisterClass *RC,
+                                const TargetRegisterInfo *TRI,
+                                int64_t Offset) const;
 
   virtual bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const;
 
   virtual unsigned GetOppositeBranchOpc(unsigned Opc) const;
 
+  // Adjust SP by FrameSize bytes. Save RA, S0, S1
+  void makeFrame(unsigned SP, int64_t FrameSize, MachineBasicBlock &MBB,
+                      MachineBasicBlock::iterator I) const;
+
+  // Adjust SP by FrameSize bytes. Restore RA, S0, S1
+  void restoreFrame(unsigned SP, int64_t FrameSize, MachineBasicBlock &MBB,
+                      MachineBasicBlock::iterator I) const;
+
+
   /// Adjust SP by Amount bytes.
   void adjustStackPtr(unsigned SP, int64_t Amount, MachineBasicBlock &MBB,
                       MachineBasicBlock::iterator I) const;
 
+  /// Emit a series of instructions to load an immediate.
+  // This is to adjust some FrameReg. We return the new register to be used
+  // in place of FrameReg and the adjusted immediate field (&NewImm)
+  //
+  unsigned loadImmediate(unsigned FrameReg,
+                         int64_t Imm, MachineBasicBlock &MBB,
+                         MachineBasicBlock::iterator II, DebugLoc DL,
+                         unsigned &NewImm) const;
+
+  static bool validSpImm8(int offset) {
+    return ((offset & 7) == 0) && isInt<11>(offset);
+  }
+
+  //
+  // build the proper one based on the Imm field
+  //
+
+  const MCInstrDesc& AddiuSpImm(int64_t Imm) const;
+
+  void BuildAddiuSpImm
+    (MachineBasicBlock &MBB, MachineBasicBlock::iterator I, int64_t Imm) const;
+
 private:
   virtual unsigned GetAnalyzableBrOpc(unsigned Opc) const;
 
   void ExpandRetRA16(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
                    unsigned Opc) const;
+
+  // Adjust SP by Amount bytes where bytes can be up to 32bit number.
+  void adjustStackPtrBig(unsigned SP, int64_t Amount, MachineBasicBlock &MBB,
+                         MachineBasicBlock::iterator I,
+                         unsigned Reg1, unsigned Reg2) const;
+
+  // Adjust SP by Amount bytes where bytes can be up to 32bit number.
+  void adjustStackPtrBigUnrestricted(unsigned SP, int64_t Amount,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I) const;
+
 };
 
 }
diff --git a/lib/Target/Mips/Mips16InstrInfo.td b/lib/Target/Mips/Mips16InstrInfo.td
index 5defc75ea6ef..aa51aaf46565 100644
--- a/lib/Target/Mips/Mips16InstrInfo.td
+++ b/lib/Target/Mips/Mips16InstrInfo.td
@@ -15,7 +15,7 @@
 // Mips Address
 //
 def addr16 :
-  ComplexPattern<iPTR, 3, "SelectAddr16", [frameindex], [SDNPWantParent]>;
+  ComplexPattern<iPTR, 3, "selectAddr16", [frameindex], [SDNPWantParent]>;
 
 //
 // Address operand
@@ -32,18 +32,76 @@ def mem16_ea : Operand<i32> {
 }
 
 //
+//
+// I8 instruction format
+//
+
+class FI816_ins_base<bits<3> _func, string asmstr,
+                     string asmstr2, InstrItinClass itin>:
+  FI816<_func, (outs), (ins simm16:$imm), !strconcat(asmstr, asmstr2),
+        [], itin>;
+
+
+class FI816_SP_ins<bits<3> _func, string asmstr,
+                   InstrItinClass itin>:
+  FI816_ins_base<_func, asmstr, "\t$$sp, $imm # 16 bit inst", itin>;
+
+//
+// RI instruction format
+//
+
+
+class FRI16_ins_base<bits<5> op, string asmstr, string asmstr2,
+                     InstrItinClass itin>:
+  FRI16<op, (outs CPU16Regs:$rx), (ins simm16:$imm),
+        !strconcat(asmstr, asmstr2), [], itin>;
+
+class FRI16_ins<bits<5> op, string asmstr,
+                InstrItinClass itin>:
+  FRI16_ins_base<op, asmstr, "\t$rx, $imm \t# 16 bit inst", itin>;
+
+class FRI16R_ins_base<bits<5> op, string asmstr, string asmstr2,
+                     InstrItinClass itin>:
+  FRI16<op, (outs), (ins CPU16Regs:$rx, simm16:$imm),
+        !strconcat(asmstr, asmstr2), [], itin>;
+
+class FRI16R_ins<bits<5> op, string asmstr,
+                InstrItinClass itin>:
+  FRI16R_ins_base<op, asmstr, "\t$rx, $imm \t# 16 bit inst", itin>;
+
+class F2RI16_ins<bits<5> _op, string asmstr,
+                     InstrItinClass itin>:
+  FRI16<_op, (outs CPU16Regs:$rx), (ins CPU16Regs:$rx_, simm16:$imm),
+        !strconcat(asmstr, "\t$rx, $imm\t# 16 bit inst"), [], itin> {
+  let Constraints = "$rx_ = $rx";
+}
+
+class FRI16_B_ins<bits<5> _op, string asmstr,
+                  InstrItinClass itin>:
+  FRI16<_op, (outs), (ins  CPU16Regs:$rx, brtarget:$imm),
+        !strconcat(asmstr, "\t$rx, $imm  # 16 bit inst"), [], itin>;
+//
 // Compare a register and immediate and place result in CC
 // Implicit use of T8
 //
 // EXT-CCRR Instruction format
 //
-class FEXT_CCRXI16_ins<bits<5> _op, string asmstr,
-                       InstrItinClass itin>:
-  FEXT_RI16<_op, (outs CPU16Regs:$cc), (ins CPU16Regs:$rx, simm16:$imm),
-            !strconcat(asmstr, "\t$rx, $imm\n\tmove\t$cc, $$t8"), [], itin> {
+class FEXT_CCRXI16_ins<string asmstr>:
+  MipsPseudo16<(outs CPU16Regs:$cc), (ins CPU16Regs:$rx, simm16:$imm),
+               !strconcat(asmstr, "\t$rx, $imm\n\tmove\t$cc, $$t8"), []> {
   let isCodeGenOnly=1;
+  let usesCustomInserter = 1;
 }
 
+// JAL and JALX instruction format
+//
+class FJAL16_ins<bits<1> _X, string asmstr,
+                 InstrItinClass itin>:
+  FJAL16<_X, (outs), (ins simm20:$imm),
+         !strconcat(asmstr, "\t$imm\n\tnop"),[],
+         itin>  {
+  let isCodeGenOnly=1;
+}
 //
 // EXT-I instruction format
 //
@@ -57,13 +115,17 @@ class FEXT_I16_ins<bits<5> eop, string asmstr, InstrItinClass itin> :
 
 class FEXT_I816_ins_base<bits<3> _func, string asmstr,
                          string asmstr2, InstrItinClass itin>:
-  FEXT_I816<_func, (outs), (ins uimm16:$imm), !strconcat(asmstr, asmstr2),
+  FEXT_I816<_func, (outs), (ins simm16:$imm), !strconcat(asmstr, asmstr2),
             [], itin>;
 
 class FEXT_I816_ins<bits<3> _func, string asmstr,
                     InstrItinClass itin>:
   FEXT_I816_ins_base<_func, asmstr, "\t$imm", itin>;
 
+class FEXT_I816_SP_ins<bits<3> _func, string asmstr,
+                       InstrItinClass itin>:
+      FEXT_I816_ins_base<_func, asmstr, "\t$$sp, $imm", itin>;
+
 //
 // Assembler formats in alphabetical order.
 // Natural and pseudos are mixed together.
@@ -73,10 +135,11 @@ class FEXT_I816_ins<bits<3> _func, string asmstr,
 //
 // CC-RR Instruction format
 //
-class FCCRR16_ins<bits<5> f, string asmstr, InstrItinClass itin> :
-  FRR16<f, (outs CPU16Regs:$cc), (ins CPU16Regs:$rx, CPU16Regs:$ry),
-        !strconcat(asmstr, "\t$rx, $ry\n\tmove\t$cc, $$t8"), [], itin> {
+class FCCRR16_ins<string asmstr> :
+  MipsPseudo16<(outs CPU16Regs:$cc), (ins CPU16Regs:$rx, CPU16Regs:$ry),
+               !strconcat(asmstr, "\t$rx, $ry\n\tmove\t$cc, $$t8"), []> {
   let isCodeGenOnly=1;
+  let usesCustomInserter = 1;
 }
 
 //
@@ -92,6 +155,15 @@ class FEXT_RI16_ins<bits<5> _op, string asmstr,
                     InstrItinClass itin>:
   FEXT_RI16_ins_base<_op, asmstr, "\t$rx, $imm", itin>;
 
+class FEXT_RI16R_ins_base<bits<5> _op, string asmstr, string asmstr2,
+                         InstrItinClass itin>:
+  FEXT_RI16<_op, (outs ), (ins CPU16Regs:$rx, simm16:$imm),
+                  !strconcat(asmstr, asmstr2), [], itin>;
+
+class FEXT_RI16R_ins<bits<5> _op, string asmstr,
+                    InstrItinClass itin>:
+  FEXT_RI16R_ins_base<_op, asmstr, "\t$rx, $imm", itin>;
+
 class FEXT_RI16_PC_ins<bits<5> _op, string asmstr, InstrItinClass itin>:
   FEXT_RI16_ins_base<_op, asmstr, "\t$rx, $$pc, $imm", itin>;
 
@@ -149,25 +221,25 @@ class FEXT_SHIFT16_ins<bits<2> _f, string asmstr, InstrItinClass itin>:
 //
 // EXT-T8I8
 //
-class FEXT_T8I816_ins<bits<3> _func, string asmstr, string asmstr2,
-                      InstrItinClass itin>:
-  FEXT_I816<_func, (outs),
-            (ins CPU16Regs:$rx, CPU16Regs:$ry, brtarget:$imm),
-            !strconcat(asmstr2, !strconcat("\t$rx, $ry\n\t",
-            !strconcat(asmstr, "\t$imm"))),[], itin> {
+class FEXT_T8I816_ins<string asmstr, string asmstr2>:
+  MipsPseudo16<(outs),
+               (ins CPU16Regs:$rx, CPU16Regs:$ry, brtarget:$imm),
+               !strconcat(asmstr2, !strconcat("\t$rx, $ry\n\t",
+               !strconcat(asmstr, "\t$imm"))),[]> {
   let isCodeGenOnly=1;
+  let usesCustomInserter = 1;
 }
 
 //
 // EXT-T8I8I
 //
-class FEXT_T8I8I16_ins<bits<3> _func, string asmstr, string asmstr2,
-                       InstrItinClass itin>:
-  FEXT_I816<_func, (outs),
-            (ins CPU16Regs:$rx, simm16:$imm, brtarget:$targ),
-            !strconcat(asmstr2, !strconcat("\t$rx, $imm\n\t",
-            !strconcat(asmstr, "\t$targ"))), [], itin> {
+class FEXT_T8I8I16_ins<string asmstr, string asmstr2>:
+  MipsPseudo16<(outs),
+               (ins CPU16Regs:$rx, simm16:$imm, brtarget:$targ),
+               !strconcat(asmstr2, !strconcat("\t$rx, $imm\n\t",
+               !strconcat(asmstr, "\t$targ"))), []> {
   let isCodeGenOnly=1;
+  let usesCustomInserter = 1;
 }
 //
 
@@ -215,9 +287,14 @@ class FRR16_ins<bits<5> f, string asmstr, InstrItinClass itin> :
         !strconcat(asmstr, "\t$rx, $ry"), [], itin> {
 }
 
-class FRRTR16_ins<bits<5> f, string asmstr, InstrItinClass itin> :
-  FRR16<f, (outs CPU16Regs:$rz), (ins CPU16Regs:$rx, CPU16Regs:$ry),
-        !strconcat(asmstr, "\t$rx, $ry\n\tmove\t$rz, $$t8"), [], itin> ;
+class FRR16R_ins<bits<5> f, string asmstr, InstrItinClass itin> :
+  FRR16<f, (outs), (ins  CPU16Regs:$rx, CPU16Regs:$ry),
+        !strconcat(asmstr, "\t$rx, $ry"), [], itin> {
+}
+
+class FRRTR16_ins<string asmstr> :
+  MipsPseudo16<(outs CPU16Regs:$rz), (ins CPU16Regs:$rx, CPU16Regs:$ry),
+               !strconcat(asmstr, "\t$rx, $ry\n\tmove\t$rz, $$t8"), []> ;
 
 //
 // maybe refactor but need a $zero as a dummy first parameter
@@ -253,7 +330,7 @@ class FRR16_JALRC_RA_only_ins<bits<1> nd_, bits<1> l_,
 
 class FRR16_JALRC_ins<bits<1> nd, bits<1> l, bits<1> ra,
                       string asmstr, InstrItinClass itin>:
-  FRR16_JALRC<nd, l, ra, (outs), (ins CPU16Regs:$rx), 
+  FRR16_JALRC<nd, l, ra, (outs), (ins CPU16Regs:$rx),
               !strconcat(asmstr, "\t $rx"), [], itin> ;
 
 //
@@ -292,13 +369,13 @@ class FRRR16_ins<bits<2> _f, string asmstr,  InstrItinClass itin> :
 //
 // So this pseudo class only has one operand, i.e. op
 //
-class Sel<bits<5> f1, string op, InstrItinClass itin>:
-  MipsInst16_32<(outs CPU16Regs:$rd_), (ins CPU16Regs:$rd, CPU16Regs:$rs,
-                CPU16Regs:$rt),
-                !strconcat(op, "\t$rt, .+4\n\t\n\tmove $rd, $rs"), [], itin,
-                Pseudo16> {
-  let isCodeGenOnly=1;
+class Sel<string op>:
+  MipsPseudo16<(outs CPU16Regs:$rd_), (ins CPU16Regs:$rd, CPU16Regs:$rs,
+               CPU16Regs:$rt),
+               !strconcat(op, "\t$rt, .+4\n\t\n\tmove $rd, $rs"), []> {
+  //let isCodeGenOnly=1;
   let Constraints = "$rd = $rd_";
+  let usesCustomInserter = 1;
 }
 
 //
@@ -316,16 +393,15 @@ class Sel<bits<5> f1, string op, InstrItinClass itin>:
 // move $rd, $rs
 //
 //
-class SeliT<bits<5> f1, string op1, bits<5> f2, string op2,
-                 InstrItinClass itin>:
-  MipsInst16_32<(outs CPU16Regs:$rd_), (ins CPU16Regs:$rd, CPU16Regs:$rs,
-                                        CPU16Regs:$rl, simm16:$imm),
-                 !strconcat(op2,
-                 !strconcat("\t$rl, $imm\n\t",
-                 !strconcat(op1, "\t.+4\n\tmove $rd, $rs"))), [], itin,
-                 Pseudo16> {
+class SeliT<string op1, string op2>:
+  MipsPseudo16<(outs CPU16Regs:$rd_), (ins CPU16Regs:$rd, CPU16Regs:$rs,
+                                       CPU16Regs:$rl, simm16:$imm),
+               !strconcat(op2,
+               !strconcat("\t$rl, $imm\n\t",
+               !strconcat(op1, "\t.+4\n\tmove $rd, $rs"))), []> {
   let isCodeGenOnly=1;
   let Constraints = "$rd = $rd_";
+  let usesCustomInserter = 1;
 }
 
 //
@@ -340,18 +416,30 @@ class SeliT<bits<5> f1, string op1, bits<5> f2, string op2,
 // move $rd, $rs
 //
 //
-class SelT<bits<5> f1, string op1, bits<5> f2, string op2,
-           InstrItinClass itin>:
-  MipsInst16_32<(outs CPU16Regs:$rd_), (ins CPU16Regs:$rd, CPU16Regs:$rs,
+class SelT<string op1, string op2>:
+  MipsPseudo16<(outs CPU16Regs:$rd_),
+               (ins CPU16Regs:$rd, CPU16Regs:$rs,
                 CPU16Regs:$rl, CPU16Regs:$rr),
-                !strconcat(op2,
-                !strconcat("\t$rl, $rr\n\t",
-                !strconcat(op1, "\t.+4\n\tmove $rd, $rs"))), [], itin,
-                Pseudo16> {
+               !strconcat(op2,
+               !strconcat("\t$rl, $rr\n\t",
+               !strconcat(op1, "\t.+4\n\tmove $rd, $rs"))), []> {
   let isCodeGenOnly=1;
   let Constraints = "$rd = $rd_";
+  let usesCustomInserter = 1;
 }
 
+//
+// 32 bit constant
+//
+def imm32: Operand<i32>;
+
+def Constant32:
+  MipsPseudo16<(outs), (ins imm32:$imm), "\t.word $imm", []>;
+
+def LwConstant32:
+  MipsPseudo16<(outs), (ins CPU16Regs:$rx, imm32:$imm),
+    "lw\t$rx, 1f\n\tb\t2f\n\t.align\t2\n1: \t.word\t$imm\n2:", []>;
+
 
 //
 // Some general instruction class info
@@ -385,14 +473,21 @@ class MayStore {
 }
 //
 
+
 // Format: ADDIU rx, immediate MIPS16e
 // Purpose: Add Immediate Unsigned Word (2-Operand, Extended)
 // To add a constant to a 32-bit integer.
 //
 def AddiuRxImmX16: FEXT_RI16_ins<0b01001, "addiu", IIAlu>;
 
+def AddiuRxRxImm16: F2RI16_ins<0b01001, "addiu", IIAlu>,
+  ArithLogic16Defs<0> {
+  let AddedComplexity = 5;
+}
 def AddiuRxRxImmX16: FEXT_2RI16_ins<0b01001, "addiu", IIAlu>,
-  ArithLogic16Defs<0>;
+  ArithLogic16Defs<0> {
+  let isCodeGenOnly = 1;
+}
 
 def AddiuRxRyOffMemX16:
   FEXT_RRI_A16_mem_ins<0, "addiu", mem16_ea, IIAlu>;
@@ -404,6 +499,25 @@ def AddiuRxRyOffMemX16:
 // To add a constant to the program counter.
 //
 def AddiuRxPcImmX16: FEXT_RI16_PC_ins<0b00001, "addiu", IIAlu>;
+
+//
+// Format: ADDIU sp, immediate MIPS16e
+// Purpose: Add Immediate Unsigned Word (2-Operand, SP-Relative, Extended)
+// To add a constant to the stack pointer.
+//
+def AddiuSpImm16
+  : FI816_SP_ins<0b011, "addiu", IIAlu> {
+  let Defs = [SP];
+  let Uses = [SP];
+  let AddedComplexity = 5;
+}
+
+def AddiuSpImmX16
+  : FEXT_I816_SP_ins<0b011, "addiu", IIAlu> {
+  let Defs = [SP];
+  let Uses = [SP];
+}
+
 //
 // Format: ADDU rz, rx, ry MIPS16e
 // Purpose: Add Unsigned Word (3-Operand)
@@ -422,6 +536,14 @@ def AndRxRxRy16: FRxRxRy16_ins<0b01100, "and", IIAlu>, ArithLogic16Defs<1>;
 
 //
 // Format: BEQZ rx, offset MIPS16e
+// Purpose: Branch on Equal to Zero
+// To test a GPR then do a PC-relative conditional branch.
+//
+def BeqzRxImm16: FRI16_B_ins<0b00100, "beqz", IIAlu>, cbranch16;
+
+
+//
+// Format: BEQZ rx, offset MIPS16e
 // Purpose: Branch on Equal to Zero (Extended)
 // To test a GPR then do a PC-relative conditional branch.
 //
@@ -435,6 +557,13 @@ def BimmX16: FEXT_I16_ins<0b00010, "b", IIAlu>, branch16;
 
 //
 // Format: BNEZ rx, offset MIPS16e
+// Purpose: Branch on Not Equal to Zero
+// To test a GPR then do a PC-relative conditional branch.
+//
+def BnezRxImm16: FRI16_B_ins<0b00101, "bnez", IIAlu>, cbranch16;
+
+//
+// Format: BNEZ rx, offset MIPS16e
 // Purpose: Branch on Not Equal to Zero (Extended)
 // To test a GPR then do a PC-relative conditional branch.
 //
@@ -445,20 +574,22 @@ def BnezRxImmX16: FEXT_RI16_B_ins<0b00101, "bnez", IIAlu>, cbranch16;
 // Purpose: Branch on T Equal to Zero (Extended)
 // To test special register T then do a PC-relative conditional branch.
 //
-def BteqzX16: FEXT_I816_ins<0b000, "bteqz", IIAlu>, cbranch16;
+def BteqzX16: FEXT_I816_ins<0b000, "bteqz", IIAlu>, cbranch16 {
+  let Uses = [T8];
+}
 
-def BteqzT8CmpX16: FEXT_T8I816_ins<0b000, "bteqz", "cmp", IIAlu>, cbranch16;
+def BteqzT8CmpX16: FEXT_T8I816_ins<"bteqz", "cmp">, cbranch16;
 
-def BteqzT8CmpiX16: FEXT_T8I8I16_ins<0b000, "bteqz", "cmpi", IIAlu>,
+def BteqzT8CmpiX16: FEXT_T8I8I16_ins<"bteqz", "cmpi">,
   cbranch16;
 
-def BteqzT8SltX16: FEXT_T8I816_ins<0b000, "bteqz", "slt", IIAlu>, cbranch16;
+def BteqzT8SltX16: FEXT_T8I816_ins<"bteqz", "slt">, cbranch16;
 
-def BteqzT8SltuX16: FEXT_T8I816_ins<0b000, "bteqz", "sltu", IIAlu>, cbranch16;
+def BteqzT8SltuX16: FEXT_T8I816_ins<"bteqz", "sltu">, cbranch16;
 
-def BteqzT8SltiX16: FEXT_T8I8I16_ins<0b000, "bteqz", "slti", IIAlu>, cbranch16;
+def BteqzT8SltiX16: FEXT_T8I8I16_ins<"bteqz", "slti">, cbranch16;
 
-def BteqzT8SltiuX16: FEXT_T8I8I16_ins<0b000, "bteqz", "sltiu", IIAlu>,
+def BteqzT8SltiuX16: FEXT_T8I8I16_ins<"bteqz", "sltiu">,
   cbranch16;
 
 //
@@ -466,22 +597,52 @@ def BteqzT8SltiuX16: FEXT_T8I8I16_ins<0b000, "bteqz", "sltiu", IIAlu>,
 // Purpose: Branch on T Not Equal to Zero (Extended)
 // To test special register T then do a PC-relative conditional branch.
 //
-def BtnezX16: FEXT_I816_ins<0b001, "btnez", IIAlu> ,cbranch16;
+def BtnezX16: FEXT_I816_ins<0b001, "btnez", IIAlu> ,cbranch16 {
+  let Uses = [T8];
+}
 
-def BtnezT8CmpX16: FEXT_T8I816_ins<0b000, "btnez", "cmp", IIAlu>, cbranch16;
+def BtnezT8CmpX16: FEXT_T8I816_ins<"btnez", "cmp">, cbranch16;
 
-def BtnezT8CmpiX16: FEXT_T8I8I16_ins<0b000, "btnez", "cmpi", IIAlu>, cbranch16;
+def BtnezT8CmpiX16: FEXT_T8I8I16_ins<"btnez", "cmpi">, cbranch16;
 
-def BtnezT8SltX16: FEXT_T8I816_ins<0b000, "btnez", "slt", IIAlu>, cbranch16;
+def BtnezT8SltX16: FEXT_T8I816_ins<"btnez", "slt">, cbranch16;
 
-def BtnezT8SltuX16: FEXT_T8I816_ins<0b000, "btnez", "sltu", IIAlu>, cbranch16;
+def BtnezT8SltuX16: FEXT_T8I816_ins<"btnez", "sltu">, cbranch16;
 
-def BtnezT8SltiX16: FEXT_T8I8I16_ins<0b000, "btnez", "slti", IIAlu>, cbranch16;
+def BtnezT8SltiX16: FEXT_T8I8I16_ins<"btnez", "slti">, cbranch16;
 
-def BtnezT8SltiuX16: FEXT_T8I8I16_ins<0b000, "btnez", "sltiu", IIAlu>,
+def BtnezT8SltiuX16: FEXT_T8I8I16_ins<"btnez", "sltiu">,
   cbranch16;
 
 //
+// Format: CMP rx, ry MIPS16e
+// Purpose: Compare
+// To compare the contents of two GPRs.
+//
+def CmpRxRy16: FRR16R_ins<0b01010, "cmp", IIAlu> {
+  let Defs = [T8];
+}
+
+//
+// Format: CMPI rx, immediate MIPS16e
+// Purpose: Compare Immediate
+// To compare a constant with the contents of a GPR.
+//
+def CmpiRxImm16: FRI16R_ins<0b01110, "cmpi", IIAlu> {
+  let Defs = [T8];
+}
+
+//
+// Format: CMPI rx, immediate MIPS16e
+// Purpose: Compare Immediate (Extended)
+// To compare a constant with the contents of a GPR.
+//
+def CmpiRxImmX16: FEXT_RI16R_ins<0b01110, "cmpi", IIAlu> {
+  let Defs = [T8];
+}
+
+
+//
 // Format: DIV rx, ry MIPS16e
 // Purpose: Divide Word
 // To divide 32-bit signed integers.
@@ -498,7 +659,19 @@ def DivRxRy16: FRR16_div_ins<0b11010, "div", IIAlu> {
 def DivuRxRy16: FRR16_div_ins<0b11011, "divu", IIAlu> {
   let Defs = [HI, LO];
 }
+//
+// Format: JAL target MIPS16e
+// Purpose: Jump and Link
+// To execute a procedure call within the current 256 MB-aligned
+// region and preserve the current ISA.
+//
 
+def Jal16 : FJAL16_ins<0b0, "jal", IIAlu> {
+  let isBranch = 1;
+  let hasDelaySlot = 0;  // not true, but we add the nop for now
+  let isTerminator=1;
+  let isBarrier=1;
+}
 
 //
 // Format: JR ra MIPS16e
@@ -515,7 +688,7 @@ def JrRa16: FRR16_JALRC_RA_only_ins<0, 0, "jr", IIAlu> {
   let isBarrier=1;
 }
 
-def JrcRa16: FRR16_JALRC_RA_only_ins<0, 0, "jrc", IIAlu> {
+def JrcRa16: FRR16_JALRC_RA_only_ins<1, 1, "jrc", IIAlu> {
   let isBranch = 1;
   let isIndirectBranch = 1;
   let isTerminator=1;
@@ -533,7 +706,9 @@ def JrcRx16: FRR16_JALRC_ins<1, 1, 0, "jrc", IIAlu> {
 // Purpose: Load Byte (Extended)
 // To load a byte from memory as a signed value.
 //
-def LbRxRyOffMemX16: FEXT_RRI16_mem_ins<0b10011, "lb", mem16, IILoad>, MayLoad;
+def LbRxRyOffMemX16: FEXT_RRI16_mem_ins<0b10011, "lb", mem16, IILoad>, MayLoad{
+  let isCodeGenOnly = 1;
+}
 
 //
 // Format: LBU ry, offset(rx) MIPS16e
@@ -541,14 +716,18 @@ def LbRxRyOffMemX16: FEXT_RRI16_mem_ins<0b10011, "lb", mem16, IILoad>, MayLoad;
 // To load a byte from memory as a unsigned value.
 //
 def LbuRxRyOffMemX16:
-  FEXT_RRI16_mem_ins<0b10100, "lbu", mem16, IILoad>, MayLoad;
+  FEXT_RRI16_mem_ins<0b10100, "lbu", mem16, IILoad>, MayLoad {
+  let isCodeGenOnly = 1;
+}
 
 //
 // Format: LH ry, offset(rx) MIPS16e
 // Purpose: Load Halfword signed (Extended)
 // To load a halfword from memory as a signed value.
 //
-def LhRxRyOffMemX16: FEXT_RRI16_mem_ins<0b10100, "lh", mem16, IILoad>, MayLoad;
+def LhRxRyOffMemX16: FEXT_RRI16_mem_ins<0b10100, "lh", mem16, IILoad>, MayLoad{
+  let isCodeGenOnly = 1;
+}
 
 //
 // Format: LHU ry, offset(rx) MIPS16e
@@ -556,7 +735,16 @@ def LhRxRyOffMemX16: FEXT_RRI16_mem_ins<0b10100, "lh", mem16, IILoad>, MayLoad;
 // To load a halfword from memory as an unsigned value.
 //
 def LhuRxRyOffMemX16:
-  FEXT_RRI16_mem_ins<0b10100, "lhu", mem16, IILoad>, MayLoad;
+  FEXT_RRI16_mem_ins<0b10100, "lhu", mem16, IILoad>, MayLoad {
+  let isCodeGenOnly = 1;
+}
+
+//
+// Format: LI rx, immediate MIPS16e
+// Purpose: Load Immediate
+// To load a constant into a GPR.
+//
+def LiRxImm16: FRI16_ins<0b01101, "li", IIAlu>;
 
 //
 // Format: LI rx, immediate MIPS16e
@@ -570,13 +758,17 @@ def LiRxImmX16: FEXT_RI16_ins<0b01101, "li", IIAlu>;
 // Purpose: Load Word (Extended)
 // To load a word from memory as a signed value.
 //
-def LwRxRyOffMemX16: FEXT_RRI16_mem_ins<0b10011, "lw", mem16, IILoad>, MayLoad;
+def LwRxRyOffMemX16: FEXT_RRI16_mem_ins<0b10011, "lw", mem16, IILoad>, MayLoad{
+  let isCodeGenOnly = 1;
+}
 
 // Format: LW rx, offset(sp) MIPS16e
 // Purpose: Load Word (SP-Relative, Extended)
 // To load an SP-relative word from memory as a signed value.
 //
-def LwRxSpImmX16: FEXT_RI16_SP_explicit_ins<0b10110, "lw", IILoad>, MayLoad;
+def LwRxSpImmX16: FEXT_RI16_SP_explicit_ins<0b10110, "lw", IILoad>, MayLoad{
+  let Uses = [SP];
+}
 
 //
 // Format: MOVE r32, rz MIPS16e
@@ -688,6 +880,8 @@ def RestoreRaF16:
   FI8_SVRS16<0b1, (outs), (ins uimm16:$frame_size),
              "restore\t$$ra,  $$s0, $$s1, $frame_size", [], IILoad >, MayLoad {
   let isCodeGenOnly = 1;
+  let Defs = [S0, S1, RA, SP];
+  let Uses = [SP];
 }
 
 // Use Restore to increment SP since SP is not a Mip 16 register, this
@@ -698,6 +892,8 @@ def RestoreIncSpF16:
   FI8_SVRS16<0b1, (outs), (ins uimm16:$frame_size),
              "restore\t$frame_size", [], IILoad >, MayLoad {
   let isCodeGenOnly = 1;
+  let Defs = [SP];
+  let Uses = [SP];
 }
 
 //
@@ -712,6 +908,8 @@ def SaveRaF16:
   FI8_SVRS16<0b1, (outs), (ins uimm16:$frame_size),
              "save\t$$ra, $$s0, $$s1, $frame_size", [], IIStore >, MayStore {
   let isCodeGenOnly = 1;
+  let Uses = [RA, SP, S0, S1];
+  let Defs = [SP];
 }
 
 //
@@ -723,6 +921,8 @@ def SaveDecSpF16:
   FI8_SVRS16<0b1, (outs), (ins uimm16:$frame_size),
              "save\t$frame_size", [], IIStore >, MayStore {
   let isCodeGenOnly = 1;
+  let Uses = [SP];
+  let Defs = [SP];
 }
 //
 // Format: SB ry, offset(rx) MIPS16e
@@ -741,7 +941,7 @@ def SbRxRyOffMemX16:
 // Purpose: if rt==0, do nothing
 //          else rs = rt
 //
-def SelBeqZ: Sel<0b00100, "beqz", IIAlu>;
+def SelBeqZ: Sel<"beqz">;
 
 //
 // Format:  SelTBteqZCmp rd, rs, rl, rr
@@ -749,7 +949,7 @@ def SelBeqZ: Sel<0b00100, "beqz", IIAlu>;
 //          If b==0 then do nothing.
 //          if b!=0 then rd = rs
 //
-def SelTBteqZCmp: SelT<0b000, "bteqz", 0b01010, "cmp", IIAlu>;
+def SelTBteqZCmp: SelT<"bteqz", "cmp">;
 
 //
 // Format:  SelTBteqZCmpi rd, rs, rl, rr
@@ -757,7 +957,7 @@ def SelTBteqZCmp: SelT<0b000, "bteqz", 0b01010, "cmp", IIAlu>;
 //          If b==0 then do nothing.
 //          if b!=0 then rd = rs
 //
-def SelTBteqZCmpi: SeliT<0b000, "bteqz", 0b01110, "cmpi", IIAlu>;
+def SelTBteqZCmpi: SeliT<"bteqz", "cmpi">;
 
 //
 // Format:  SelTBteqZSlt rd, rs, rl, rr
@@ -765,7 +965,7 @@ def SelTBteqZCmpi: SeliT<0b000, "bteqz", 0b01110, "cmpi", IIAlu>;
 //          If b==0 then do nothing.
 //          if b!=0 then rd = rs
 //
-def SelTBteqZSlt: SelT<0b000, "bteqz", 0b00010, "slt", IIAlu>;
+def SelTBteqZSlt: SelT<"bteqz", "slt">;
 
 //
 // Format:  SelTBteqZSlti rd, rs, rl, rr
@@ -773,7 +973,7 @@ def SelTBteqZSlt: SelT<0b000, "bteqz", 0b00010, "slt", IIAlu>;
 //          If b==0 then do nothing.
 //          if b!=0 then rd = rs
 //
-def SelTBteqZSlti: SeliT<0b000, "bteqz", 0b01010, "slti", IIAlu>;
+def SelTBteqZSlti: SeliT<"bteqz", "slti">;
 
 //
 // Format:  SelTBteqZSltu rd, rs, rl, rr
@@ -781,7 +981,7 @@ def SelTBteqZSlti: SeliT<0b000, "bteqz", 0b01010, "slti", IIAlu>;
 //          If b==0 then do nothing.
 //          if b!=0 then rd = rs
 //
-def SelTBteqZSltu: SelT<0b000, "bteqz", 0b00011, "sltu", IIAlu>;
+def SelTBteqZSltu: SelT<"bteqz", "sltu">;
 
 //
 // Format:  SelTBteqZSltiu rd, rs, rl, rr
@@ -789,14 +989,14 @@ def SelTBteqZSltu: SelT<0b000, "bteqz", 0b00011, "sltu", IIAlu>;
 //          If b==0 then do nothing.
 //          if b!=0 then rd = rs
 //
-def SelTBteqZSltiu: SeliT<0b000, "bteqz", 0b01011, "sltiu", IIAlu>;
+def SelTBteqZSltiu: SeliT<"bteqz", "sltiu">;
 
 //
 // Format: SelBnez rd, rs, rt
 // Purpose: if rt!=0, do nothing
 //          else rs = rt
 //
-def SelBneZ: Sel<0b00101, "bnez", IIAlu>;
+def SelBneZ: Sel<"bnez">;
 
 //
 // Format:  SelTBtneZCmp rd, rs, rl, rr
@@ -804,7 +1004,7 @@ def SelBneZ: Sel<0b00101, "bnez", IIAlu>;
 //          If b!=0 then do nothing.
 //          if b0=0 then rd = rs
 //
-def SelTBtneZCmp: SelT<0b001, "btnez", 0b01010, "cmp", IIAlu>;
+def SelTBtneZCmp: SelT<"btnez", "cmp">;
 
 //
 // Format:  SelTBtnezCmpi rd, rs, rl, rr
@@ -812,7 +1012,7 @@ def SelTBtneZCmp: SelT<0b001, "btnez", 0b01010, "cmp", IIAlu>;
 //          If b!=0 then do nothing.
 //          if b==0 then rd = rs
 //
-def SelTBtneZCmpi: SeliT<0b000, "btnez", 0b01110, "cmpi", IIAlu>;
+def SelTBtneZCmpi: SeliT<"btnez", "cmpi">;
 
 //
 // Format:  SelTBtneZSlt rd, rs, rl, rr
@@ -820,7 +1020,7 @@ def SelTBtneZCmpi: SeliT<0b000, "btnez", 0b01110, "cmpi", IIAlu>;
 //          If b!=0 then do nothing.
 //          if b==0 then rd = rs
 //
-def SelTBtneZSlt: SelT<0b001, "btnez", 0b00010, "slt", IIAlu>;
+def SelTBtneZSlt: SelT<"btnez", "slt">;
 
 //
 // Format:  SelTBtneZSlti rd, rs, rl, rr
@@ -828,7 +1028,7 @@ def SelTBtneZSlt: SelT<0b001, "btnez", 0b00010, "slt", IIAlu>;
 //          If b!=0 then do nothing.
 //          if b==0 then rd = rs
 //
-def SelTBtneZSlti: SeliT<0b001, "btnez", 0b01010, "slti", IIAlu>;
+def SelTBtneZSlti: SeliT<"btnez", "slti">;
 
 //
 // Format:  SelTBtneZSltu rd, rs, rl, rr
@@ -836,7 +1036,7 @@ def SelTBtneZSlti: SeliT<0b001, "btnez", 0b01010, "slti", IIAlu>;
 //          If b!=0 then do nothing.
 //          if b==0 then rd = rs
 //
-def SelTBtneZSltu: SelT<0b001, "btnez", 0b00011, "sltu", IIAlu>;
+def SelTBtneZSltu: SelT<"btnez", "sltu">;
 
 //
 // Format:  SelTBtneZSltiu rd, rs, rl, rr
@@ -844,7 +1044,7 @@ def SelTBtneZSltu: SelT<0b001, "btnez", 0b00011, "sltu", IIAlu>;
 //          If b!=0 then do nothing.
 //          if b==0 then rd = rs
 //
-def SelTBtneZSltiu: SeliT<0b001, "btnez", 0b01011, "sltiu", IIAlu>;
+def SelTBtneZSltiu: SeliT<"btnez", "sltiu">;
 //
 //
 // Format: SH ry, offset(rx) MIPS16e
@@ -868,39 +1068,78 @@ def SllX16: FEXT_SHIFT16_ins<0b00, "sll", IIAlu>;
 //
 def SllvRxRy16 : FRxRxRy16_ins<0b00100, "sllv", IIAlu>;
 
+// Format: SLTI rx, immediate MIPS16e
+// Purpose: Set on Less Than Immediate
+// To record the result of a less-than comparison with a constant.
+//
+//
+def SltiRxImm16: FRI16R_ins<0b01010, "slti", IIAlu> {
+  let Defs = [T8];
+}
+
 //
 // Format: SLTI rx, immediate MIPS16e
 // Purpose: Set on Less Than Immediate (Extended)
 // To record the result of a less-than comparison with a constant.
 //
-def SltiCCRxImmX16: FEXT_CCRXI16_ins<0b01010, "slti", IIAlu>;
+//
+def SltiRxImmX16: FEXT_RI16R_ins<0b01010, "slti", IIAlu> {
+  let Defs = [T8];
+}
 
+def SltiCCRxImmX16: FEXT_CCRXI16_ins<"slti">;
+
+// Format: SLTIU rx, immediate MIPS16e
+// Purpose: Set on Less Than Immediate Unsigned
+// To record the result of a less-than comparison with a constant.
+//
+//
+def SltiuRxImm16: FRI16R_ins<0b01011, "sltiu", IIAlu> {
+  let Defs = [T8];
+}
+
+//
+// Format: SLTI rx, immediate MIPS16e
+// Purpose: Set on Less Than Immediate Unsigned (Extended)
+// To record the result of a less-than comparison with a constant.
+//
+//
+def SltiuRxImmX16: FEXT_RI16R_ins<0b01011, "sltiu", IIAlu> {
+  let Defs = [T8];
+}
 //
 // Format: SLTIU rx, immediate MIPS16e
 // Purpose: Set on Less Than Immediate Unsigned (Extended)
 // To record the result of a less-than comparison with a constant.
 //
-def SltiuCCRxImmX16: FEXT_CCRXI16_ins<0b01011, "sltiu", IIAlu>;
+def SltiuCCRxImmX16: FEXT_CCRXI16_ins<"sltiu">;
 
 //
 // Format: SLT rx, ry MIPS16e
 // Purpose: Set on Less Than
 // To record the result of a less-than comparison.
 //
-def SltRxRy16: FRR16_ins<0b00010, "slt", IIAlu>;
+def SltRxRy16: FRR16R_ins<0b00010, "slt", IIAlu>{
+  let Defs = [T8];
+}
 
-def SltCCRxRy16: FCCRR16_ins<0b00010, "slt", IIAlu>;
+def SltCCRxRy16: FCCRR16_ins<"slt">;
 
 // Format: SLTU rx, ry MIPS16e
 // Purpose: Set on Less Than Unsigned
 // To record the result of an unsigned less-than comparison.
 //
-def SltuRxRyRz16: FRRTR16_ins<0b00011, "sltu", IIAlu> {
+def SltuRxRy16: FRR16R_ins<0b00011, "sltu", IIAlu>{
+  let Defs = [T8];
+}
+
+def SltuRxRyRz16: FRRTR16_ins<"sltu"> {
   let isCodeGenOnly=1;
+  let Defs = [T8];
 }
 
 
-def SltuCCRxRy16: FCCRR16_ins<0b00011, "sltu", IIAlu>;
+def SltuCCRxRy16: FCCRR16_ins<"sltu">;
 //
 // Format: SRAV ry, rx MIPS16e
 // Purpose: Shift Word Right Arithmetic Variable
@@ -996,6 +1235,7 @@ class ArithLogicI16_pat<SDNode OpNode, PatFrag imm_type, Instruction I> :
   Mips16Pat<(OpNode CPU16Regs:$in, imm_type:$imm),
             (I CPU16Regs:$in, imm_type:$imm)>;
 
+def: ArithLogicI16_pat<add, immSExt8, AddiuRxRxImm16>;
 def: ArithLogicI16_pat<add, immSExt16, AddiuRxRxImmX16>;
 def: ArithLogicI16_pat<shl, immZExt5, SllX16>;
 def: ArithLogicI16_pat<srl, immZExt5, SrlX16>;
@@ -1029,14 +1269,19 @@ def: StoreM16_pat<store, SwRxRyOffMemX16>;
 // Unconditional branch
 class UncondBranch16_pat<SDNode OpNode, Instruction I>:
   Mips16Pat<(OpNode bb:$imm16), (I bb:$imm16)> {
-    let Predicates = [RelocPIC, InMips16Mode];
+    let Predicates = [InMips16Mode];
   }
 
+def : Mips16Pat<(MipsJmpLink (i32 tglobaladdr:$dst)),
+                (Jal16 tglobaladdr:$dst)>;
+
+def : Mips16Pat<(MipsJmpLink (i32 texternalsym:$dst)),
+                (Jal16 texternalsym:$dst)>;
+
 // Indirect branch
 def: Mips16Pat<
-  (brind CPU16Regs:$rs), 
-  (JrcRx16 CPU16Regs:$rs)>;  
-
+  (brind CPU16Regs:$rs),
+  (JrcRx16 CPU16Regs:$rs)>;
 
 // Jump and Link (Call)
 let isCall=1, hasDelaySlot=0 in
@@ -1221,14 +1466,14 @@ def: Mips16Pat<(i32 immZExt16:$in), (LiRxImmX16 immZExt16:$in)>;
 // MipsDivRem
 //
 def: Mips16Pat
-  <(MipsDivRem CPU16Regs:$rx, CPU16Regs:$ry),
+  <(MipsDivRem16 CPU16Regs:$rx, CPU16Regs:$ry),
    (DivRxRy16 CPU16Regs:$rx, CPU16Regs:$ry)>;
 
 //
 // MipsDivRemU
 //
 def: Mips16Pat
-  <(MipsDivRemU CPU16Regs:$rx, CPU16Regs:$ry),
+  <(MipsDivRemU16 CPU16Regs:$rx, CPU16Regs:$ry),
    (DivuRxRy16 CPU16Regs:$rx, CPU16Regs:$ry)>;
 
 //  signed a,b
@@ -1464,7 +1709,7 @@ def: Mips16Pat
 //
 def: Mips16Pat
   <(setle CPU16Regs:$lhs, CPU16Regs:$rhs),
-   (XorRxRxRy16 (SltCCRxRy16 CPU16Regs:$rhs, CPU16Regs:$lhs), (LiRxImmX16 1))>;
+   (XorRxRxRy16 (SltCCRxRy16 CPU16Regs:$rhs, CPU16Regs:$lhs), (LiRxImm16 1))>;
 
 //
 // setlt
@@ -1524,7 +1769,11 @@ def: Mips16Pat<(add CPU16Regs:$hi, (MipsLo tglobaladdr:$lo)),
 
 // hi/lo relocs
 
-def : Mips16Pat<(MipsHi tglobaltlsaddr:$in), 
+def : Mips16Pat<(MipsHi tglobaladdr:$in),
+                (SllX16 (LiRxImmX16 tglobaladdr:$in), 16)>;
+def : Mips16Pat<(MipsHi tjumptable:$in),
+                (SllX16 (LiRxImmX16 tjumptable:$in), 16)>;
+def : Mips16Pat<(MipsHi tglobaltlsaddr:$in),
                 (SllX16 (LiRxImmX16 tglobaltlsaddr:$in), 16)>;
 
 // wrapper_pic
@@ -1539,4 +1788,4 @@ def : Wrapper16Pat<tglobaltlsaddr, AddiuRxRxImmX16, CPU16Regs>;
 def : Mips16Pat<(i32 (extloadi8   addr16:$src)),
                 (LbuRxRyOffMemX16  addr16:$src)>;
 def : Mips16Pat<(i32 (extloadi16  addr16:$src)),
-                (LhuRxRyOffMemX16  addr16:$src)>;
\ No newline at end of file
+                (LhuRxRyOffMemX16  addr16:$src)>;
diff --git a/lib/Target/Mips/Mips16RegisterInfo.cpp b/lib/Target/Mips/Mips16RegisterInfo.cpp
index d7397a32f074..6cca2276856d 100644
--- a/lib/Target/Mips/Mips16RegisterInfo.cpp
+++ b/lib/Target/Mips/Mips16RegisterInfo.cpp
@@ -1,3 +1,4 @@
+
 //===-- Mips16RegisterInfo.cpp - MIPS16 Register Information -== ----------===//
 //
 //                     The LLVM Compiler Infrastructure
@@ -14,28 +15,30 @@
 #include "Mips16RegisterInfo.h"
 #include "Mips16InstrInfo.h"
 #include "Mips.h"
+#include "Mips16InstrInfo.h"
 #include "MipsAnalyzeImmediate.h"
 #include "MipsInstrInfo.h"
-#include "MipsSubtarget.h"
 #include "MipsMachineFunction.h"
-#include "llvm/Constants.h"
-#include "llvm/DebugInfo.h"
-#include "llvm/Type.h"
-#include "llvm/Function.h"
-#include "llvm/CodeGen/ValueTypes.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunction.h"
+#include "MipsSubtarget.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/DebugInfo.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/STLExtras.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
@@ -43,25 +46,36 @@ Mips16RegisterInfo::Mips16RegisterInfo(const MipsSubtarget &ST,
     const Mips16InstrInfo &I)
   : MipsRegisterInfo(ST), TII(I) {}
 
-// This function eliminate ADJCALLSTACKDOWN,
-// ADJCALLSTACKUP pseudo instructions
-void Mips16RegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-
-  if (!TFI->hasReservedCallFrame(MF)) {
-    int64_t Amount = I->getOperand(0).getImm();
-
-    if (I->getOpcode() == Mips::ADJCALLSTACKDOWN)
-      Amount = -Amount;
+bool Mips16RegisterInfo::requiresRegisterScavenging
+  (const MachineFunction &MF) const {
+  return true;
+}
+bool Mips16RegisterInfo::requiresFrameIndexScavenging
+  (const MachineFunction &MF) const {
+  return true;
+}
 
-    const Mips16InstrInfo *II = static_cast<const Mips16InstrInfo*>(&TII);
+bool Mips16RegisterInfo::useFPForScavengingIndex
+  (const MachineFunction &MF) const {
+  return false;
+}
 
-    II->adjustStackPtr(Mips::SP, Amount, MBB, I);
-  }
+bool Mips16RegisterInfo::saveScavengerRegister
+  (MachineBasicBlock &MBB,
+   MachineBasicBlock::iterator I,
+   MachineBasicBlock::iterator &UseMI,
+   const TargetRegisterClass *RC,
+   unsigned Reg) const {
+  DebugLoc DL;
+  TII.copyPhysReg(MBB, I, DL, Mips::T0, Reg, true);
+  TII.copyPhysReg(MBB, UseMI, DL, Reg, Mips::T0, true);
+  return true;
+}
 
-  MBB.erase(I);
+const TargetRegisterClass *
+Mips16RegisterInfo::intRegClass(unsigned Size) const {
+  assert(Size == 4);
+  return &Mips::CPU16RegsRegClass;
 }
 
 void Mips16RegisterInfo::eliminateFI(MachineBasicBlock::iterator II,
@@ -114,13 +128,23 @@ void Mips16RegisterInfo::eliminateFI(MachineBasicBlock::iterator II,
   //   by adding the size of the stack:
   //   incoming argument, callee-saved register location or local variable.
   int64_t Offset;
+  bool IsKill = false;
   Offset = SPOffset + (int64_t)StackSize;
   Offset += MI.getOperand(OpNo + 1).getImm();
 
 
   DEBUG(errs() << "Offset     : " << Offset << "\n" << "<--------->\n");
 
-  MI.getOperand(OpNo).ChangeToRegister(FrameReg, false);
+  if (!MI.isDebugValue() && ( ((FrameReg != Mips::SP) && !isInt<16>(Offset)) ||
+      ((FrameReg == Mips::SP) && !isInt<15>(Offset)) )) {
+    MachineBasicBlock &MBB = *MI.getParent();
+    DebugLoc DL = II->getDebugLoc();
+    unsigned NewImm;
+    FrameReg = TII.loadImmediate(FrameReg, Offset, MBB, II, DL, NewImm);
+    Offset = SignExtend64<16>(NewImm);
+    IsKill = true;
+  }
+  MI.getOperand(OpNo).ChangeToRegister(FrameReg, false, false, IsKill);
   MI.getOperand(OpNo + 1).ChangeToImmediate(Offset);
 
 
diff --git a/lib/Target/Mips/Mips16RegisterInfo.h b/lib/Target/Mips/Mips16RegisterInfo.h
index 153def20d085..2b3d2b1a4ecb 100644
--- a/lib/Target/Mips/Mips16RegisterInfo.h
+++ b/lib/Target/Mips/Mips16RegisterInfo.h
@@ -22,11 +22,23 @@ class Mips16InstrInfo;
 class Mips16RegisterInfo : public MipsRegisterInfo {
   const Mips16InstrInfo &TII;
 public:
-  Mips16RegisterInfo(const MipsSubtarget &Subtarget, const Mips16InstrInfo &TII);
+  Mips16RegisterInfo(const MipsSubtarget &Subtarget,
+                     const Mips16InstrInfo &TII);
+
+  bool requiresRegisterScavenging(const MachineFunction &MF) const;
+
+  bool requiresFrameIndexScavenging(const MachineFunction &MF) const;
+
+  bool useFPForScavengingIndex(const MachineFunction &MF) const;
+
+  bool saveScavengerRegister(MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I,
+                                     MachineBasicBlock::iterator &UseMI,
+                                     const TargetRegisterClass *RC,
+                                     unsigned Reg) const;
+
+  virtual const TargetRegisterClass *intRegClass(unsigned Size) const;
 
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
 private:
   virtual void eliminateFI(MachineBasicBlock::iterator II, unsigned OpNo,
                            int FrameIndex, uint64_t StackSize,
diff --git a/lib/Target/Mips/Mips64InstrInfo.td b/lib/Target/Mips/Mips64InstrInfo.td
index 83322eac8c62..846a8224af35 100644
--- a/lib/Target/Mips/Mips64InstrInfo.td
+++ b/lib/Target/Mips/Mips64InstrInfo.td
@@ -34,192 +34,202 @@ def immZExt6 : ImmLeaf<i32, [{return Imm == (Imm & 0x3f);}]>;
 //===----------------------------------------------------------------------===//
 // Instructions specific format
 //===----------------------------------------------------------------------===//
-// Shifts
-// 64-bit shift instructions.
 let DecoderNamespace = "Mips64" in {
-class shift_rotate_imm64<bits<6> func, bits<5> isRotate, string instr_asm,
-                         SDNode OpNode>:
-  shift_rotate_imm<func, isRotate, instr_asm, OpNode, immZExt6, shamt,
-                   CPU64Regs>;
-
-// Mul, Div
-class Mult64<bits<6> func, string instr_asm, InstrItinClass itin>:
-  Mult<func, instr_asm, itin, CPU64Regs, [HI64, LO64]>;
-class Div64<SDNode op, bits<6> func, string instr_asm, InstrItinClass itin>:
-  Div<op, func, instr_asm, itin, CPU64Regs, [HI64, LO64]>;
-
-multiclass Atomic2Ops64<PatFrag Op, string Opstr> {
-  def #NAME# : Atomic2Ops<Op, Opstr, CPU64Regs, CPURegs>,
-               Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : Atomic2Ops<Op, Opstr, CPU64Regs, CPU64Regs>,
-               Requires<[IsN64, HasStandardEncoding]> {
+
+multiclass Atomic2Ops64<PatFrag Op> {
+  def NAME : Atomic2Ops<Op, CPU64Regs, CPURegs>,
+             Requires<[NotN64, HasStdEnc]>;
+  def _P8  : Atomic2Ops<Op, CPU64Regs, CPU64Regs>,
+             Requires<[IsN64, HasStdEnc]> {
     let isCodeGenOnly = 1;
   }
 }
 
-multiclass AtomicCmpSwap64<PatFrag Op, string Width>  {
-  def #NAME# : AtomicCmpSwap<Op, Width, CPU64Regs, CPURegs>,
-               Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : AtomicCmpSwap<Op, Width, CPU64Regs, CPU64Regs>,
-               Requires<[IsN64, HasStandardEncoding]> {
+multiclass AtomicCmpSwap64<PatFrag Op>  {
+  def NAME : AtomicCmpSwap<Op, CPU64Regs, CPURegs>,
+             Requires<[NotN64, HasStdEnc]>;
+  def _P8  : AtomicCmpSwap<Op, CPU64Regs, CPU64Regs>,
+             Requires<[IsN64, HasStdEnc]> {
     let isCodeGenOnly = 1;
   }
 }
 }
-let usesCustomInserter = 1, Predicates = [HasMips64, HasStandardEncoding],
+let usesCustomInserter = 1, Predicates = [HasStdEnc],
   DecoderNamespace = "Mips64" in {
-  defm ATOMIC_LOAD_ADD_I64  : Atomic2Ops64<atomic_load_add_64, "load_add_64">;
-  defm ATOMIC_LOAD_SUB_I64  : Atomic2Ops64<atomic_load_sub_64, "load_sub_64">;
-  defm ATOMIC_LOAD_AND_I64  : Atomic2Ops64<atomic_load_and_64, "load_and_64">;
-  defm ATOMIC_LOAD_OR_I64   : Atomic2Ops64<atomic_load_or_64, "load_or_64">;
-  defm ATOMIC_LOAD_XOR_I64  : Atomic2Ops64<atomic_load_xor_64, "load_xor_64">;
-  defm ATOMIC_LOAD_NAND_I64 : Atomic2Ops64<atomic_load_nand_64, "load_nand_64">;
-  defm ATOMIC_SWAP_I64      : Atomic2Ops64<atomic_swap_64, "swap_64">;
-  defm ATOMIC_CMP_SWAP_I64  : AtomicCmpSwap64<atomic_cmp_swap_64, "64">;
+  defm ATOMIC_LOAD_ADD_I64  : Atomic2Ops64<atomic_load_add_64>;
+  defm ATOMIC_LOAD_SUB_I64  : Atomic2Ops64<atomic_load_sub_64>;
+  defm ATOMIC_LOAD_AND_I64  : Atomic2Ops64<atomic_load_and_64>;
+  defm ATOMIC_LOAD_OR_I64   : Atomic2Ops64<atomic_load_or_64>;
+  defm ATOMIC_LOAD_XOR_I64  : Atomic2Ops64<atomic_load_xor_64>;
+  defm ATOMIC_LOAD_NAND_I64 : Atomic2Ops64<atomic_load_nand_64>;
+  defm ATOMIC_SWAP_I64      : Atomic2Ops64<atomic_swap_64>;
+  defm ATOMIC_CMP_SWAP_I64  : AtomicCmpSwap64<atomic_cmp_swap_64>;
+}
+
+/// Pseudo instructions for loading, storing and copying accumulator registers.
+let isPseudo = 1 in {
+  defm LOAD_AC128  : LoadM<"load_ac128", ACRegs128>;
+  defm STORE_AC128 : StoreM<"store_ac128", ACRegs128>;
 }
 
+def COPY_AC128 : PseudoSE<(outs ACRegs128:$dst), (ins ACRegs128:$src), []>;
+
 //===----------------------------------------------------------------------===//
 // Instruction definition
 //===----------------------------------------------------------------------===//
 let DecoderNamespace = "Mips64" in {
 /// Arithmetic Instructions (ALU Immediate)
-def DADDi    : ArithOverflowI<0x18, "daddi", add, simm16_64, immSExt16,
-                           CPU64Regs>;
-def DADDiu   : ArithLogicI<0x19, "daddiu", add, simm16_64, immSExt16,
-                           CPU64Regs>, IsAsCheapAsAMove;
-def DANDi    : ArithLogicI<0x0c, "andi", and, uimm16_64, immZExt16, CPU64Regs>;
-def SLTi64   : SetCC_I<0x0a, "slti", setlt, simm16_64, immSExt16, CPU64Regs>;
-def SLTiu64  : SetCC_I<0x0b, "sltiu", setult, simm16_64, immSExt16, CPU64Regs>;
-def ORi64    : ArithLogicI<0x0d, "ori", or, uimm16_64, immZExt16, CPU64Regs>;
-def XORi64   : ArithLogicI<0x0e, "xori", xor, uimm16_64, immZExt16, CPU64Regs>;
-def LUi64    : LoadUpper<0x0f, "lui", CPU64Regs, uimm16_64>;
+def DADDi   : ArithLogicI<"daddi", simm16_64, CPU64RegsOpnd>, ADDI_FM<0x18>;
+def DADDiu  : ArithLogicI<"daddiu", simm16_64, CPU64RegsOpnd, immSExt16, add>,
+              ADDI_FM<0x19>, IsAsCheapAsAMove;
+def DANDi   : ArithLogicI<"andi", uimm16_64, CPU64RegsOpnd, immZExt16, and>,
+              ADDI_FM<0xc>;
+def SLTi64  : SetCC_I<"slti", setlt, simm16_64, immSExt16, CPU64Regs>,
+              SLTI_FM<0xa>;
+def SLTiu64 : SetCC_I<"sltiu", setult, simm16_64, immSExt16, CPU64Regs>,
+              SLTI_FM<0xb>;
+def ORi64   : ArithLogicI<"ori", uimm16_64, CPU64RegsOpnd, immZExt16, or>,
+              ADDI_FM<0xd>;
+def XORi64  : ArithLogicI<"xori", uimm16_64, CPU64RegsOpnd, immZExt16, xor>,
+              ADDI_FM<0xe>;
+def LUi64   : LoadUpper<"lui", CPU64Regs, uimm16_64>, LUI_FM;
 
 /// Arithmetic Instructions (3-Operand, R-Type)
-def DADD     : ArithOverflowR<0x00, 0x2C, "dadd", IIAlu, CPU64Regs, 1>;
-def DADDu    : ArithLogicR<0x00, 0x2d, "daddu", add, IIAlu, CPU64Regs, 1>;
-def DSUBu    : ArithLogicR<0x00, 0x2f, "dsubu", sub, IIAlu, CPU64Regs>;
-def SLT64    : SetCC_R<0x00, 0x2a, "slt", setlt, CPU64Regs>;
-def SLTu64   : SetCC_R<0x00, 0x2b, "sltu", setult, CPU64Regs>;
-def AND64    : ArithLogicR<0x00, 0x24, "and", and, IIAlu, CPU64Regs, 1>;
-def OR64     : ArithLogicR<0x00, 0x25, "or", or, IIAlu, CPU64Regs, 1>;
-def XOR64    : ArithLogicR<0x00, 0x26, "xor", xor, IIAlu, CPU64Regs, 1>;
-def NOR64    : LogicNOR<0x00, 0x27, "nor", CPU64Regs>;
+def DADD   : ArithLogicR<"dadd", CPU64RegsOpnd>, ADD_FM<0, 0x2c>;
+def DADDu  : ArithLogicR<"daddu", CPU64RegsOpnd, 1, IIAlu, add>,
+                              ADD_FM<0, 0x2d>;
+def DSUBu  : ArithLogicR<"dsubu", CPU64RegsOpnd, 0, IIAlu, sub>,
+                              ADD_FM<0, 0x2f>;
+def SLT64  : SetCC_R<"slt", setlt, CPU64Regs>, ADD_FM<0, 0x2a>;
+def SLTu64 : SetCC_R<"sltu", setult, CPU64Regs>, ADD_FM<0, 0x2b>;
+def AND64  : ArithLogicR<"and", CPU64RegsOpnd, 1, IIAlu, and>, ADD_FM<0, 0x24>;
+def OR64   : ArithLogicR<"or", CPU64RegsOpnd, 1, IIAlu, or>, ADD_FM<0, 0x25>;
+def XOR64  : ArithLogicR<"xor", CPU64RegsOpnd, 1, IIAlu, xor>, ADD_FM<0, 0x26>;
+def NOR64  : LogicNOR<"nor", CPU64RegsOpnd>, ADD_FM<0, 0x27>;
 
 /// Shift Instructions
-def DSLL     : shift_rotate_imm64<0x38, 0x00, "dsll", shl>;
-def DSRL     : shift_rotate_imm64<0x3a, 0x00, "dsrl", srl>;
-def DSRA     : shift_rotate_imm64<0x3b, 0x00, "dsra", sra>;
-def DSLLV    : shift_rotate_reg<0x14, 0x00, "dsllv", shl, CPU64Regs>;
-def DSRLV    : shift_rotate_reg<0x16, 0x00, "dsrlv", srl, CPU64Regs>;
-def DSRAV    : shift_rotate_reg<0x17, 0x00, "dsrav", sra, CPU64Regs>;
-let Pattern = []<dag> in {
-  def DSLL32   : shift_rotate_imm64<0x3c, 0x00, "dsll32", shl>;
-  def DSRL32   : shift_rotate_imm64<0x3e, 0x00, "dsrl32", srl>;
-  def DSRA32   : shift_rotate_imm64<0x3f, 0x00, "dsra32", sra>;
-}
+def DSLL   : shift_rotate_imm<"dsll", shamt, CPU64RegsOpnd, shl, immZExt6>,
+             SRA_FM<0x38, 0>;
+def DSRL   : shift_rotate_imm<"dsrl", shamt, CPU64RegsOpnd, srl, immZExt6>,
+             SRA_FM<0x3a, 0>;
+def DSRA   : shift_rotate_imm<"dsra", shamt, CPU64RegsOpnd, sra, immZExt6>,
+             SRA_FM<0x3b, 0>;
+def DSLLV  : shift_rotate_reg<"dsllv", CPU64RegsOpnd, shl>, SRLV_FM<0x14, 0>;
+def DSRLV  : shift_rotate_reg<"dsrlv", CPU64RegsOpnd, srl>, SRLV_FM<0x16, 0>;
+def DSRAV  : shift_rotate_reg<"dsrav", CPU64RegsOpnd, sra>, SRLV_FM<0x17, 0>;
+def DSLL32 : shift_rotate_imm<"dsll32", shamt, CPU64RegsOpnd>, SRA_FM<0x3c, 0>;
+def DSRL32 : shift_rotate_imm<"dsrl32", shamt, CPU64RegsOpnd>, SRA_FM<0x3e, 0>;
+def DSRA32 : shift_rotate_imm<"dsra32", shamt, CPU64RegsOpnd>, SRA_FM<0x3f, 0>;
 }
 // Rotate Instructions
-let Predicates = [HasMips64r2, HasStandardEncoding],
+let Predicates = [HasMips64r2, HasStdEnc],
     DecoderNamespace = "Mips64" in {
-  def DROTR    : shift_rotate_imm64<0x3a, 0x01, "drotr", rotr>;
-  def DROTRV   : shift_rotate_reg<0x16, 0x01, "drotrv", rotr, CPU64Regs>;
+  def DROTR  : shift_rotate_imm<"drotr", shamt, CPU64RegsOpnd, rotr, immZExt6>,
+               SRA_FM<0x3a, 1>;
+  def DROTRV : shift_rotate_reg<"drotrv", CPU64RegsOpnd, rotr>,
+               SRLV_FM<0x16, 1>;
 }
 
 let DecoderNamespace = "Mips64" in {
 /// Load and Store Instructions
 ///  aligned
-defm LB64    : LoadM64<0x20, "lb",  sextloadi8>;
-defm LBu64   : LoadM64<0x24, "lbu", zextloadi8>;
-defm LH64    : LoadM64<0x21, "lh",  sextloadi16>;
-defm LHu64   : LoadM64<0x25, "lhu", zextloadi16>;
-defm LW64    : LoadM64<0x23, "lw",  sextloadi32>;
-defm LWu64   : LoadM64<0x27, "lwu", zextloadi32>;
-defm SB64    : StoreM64<0x28, "sb", truncstorei8>;
-defm SH64    : StoreM64<0x29, "sh", truncstorei16>;
-defm SW64    : StoreM64<0x2b, "sw", truncstorei32>;
-defm LD      : LoadM64<0x37, "ld",  load>;
-defm SD      : StoreM64<0x3f, "sd", store>;
+defm LB64  : LoadM<"lb", CPU64Regs, sextloadi8>, LW_FM<0x20>;
+defm LBu64 : LoadM<"lbu", CPU64Regs, zextloadi8>, LW_FM<0x24>;
+defm LH64  : LoadM<"lh", CPU64Regs, sextloadi16>, LW_FM<0x21>;
+defm LHu64 : LoadM<"lhu", CPU64Regs, zextloadi16>, LW_FM<0x25>;
+defm LW64  : LoadM<"lw", CPU64Regs, sextloadi32>, LW_FM<0x23>;
+defm LWu64 : LoadM<"lwu", CPU64Regs, zextloadi32>, LW_FM<0x27>;
+defm SB64  : StoreM<"sb", CPU64Regs, truncstorei8>, LW_FM<0x28>;
+defm SH64  : StoreM<"sh", CPU64Regs, truncstorei16>, LW_FM<0x29>;
+defm SW64  : StoreM<"sw", CPU64Regs, truncstorei32>, LW_FM<0x2b>;
+defm LD    : LoadM<"ld", CPU64Regs, load>, LW_FM<0x37>;
+defm SD    : StoreM<"sd", CPU64Regs, store>, LW_FM<0x3f>;
 
 /// load/store left/right
-let isCodeGenOnly = 1 in {
-  defm LWL64 : LoadLeftRightM64<0x22, "lwl", MipsLWL>;
-  defm LWR64 : LoadLeftRightM64<0x26, "lwr", MipsLWR>;
-  defm SWL64 : StoreLeftRightM64<0x2a, "swl", MipsSWL>;
-  defm SWR64 : StoreLeftRightM64<0x2e, "swr", MipsSWR>;
-}
-defm LDL   : LoadLeftRightM64<0x1a, "ldl", MipsLDL>;
-defm LDR   : LoadLeftRightM64<0x1b, "ldr", MipsLDR>;
-defm SDL   : StoreLeftRightM64<0x2c, "sdl", MipsSDL>;
-defm SDR   : StoreLeftRightM64<0x2d, "sdr", MipsSDR>;
+defm LWL64 : LoadLeftRightM<"lwl", MipsLWL, CPU64Regs>, LW_FM<0x22>;
+defm LWR64 : LoadLeftRightM<"lwr", MipsLWR, CPU64Regs>, LW_FM<0x26>;
+defm SWL64 : StoreLeftRightM<"swl", MipsSWL, CPU64Regs>, LW_FM<0x2a>;
+defm SWR64 : StoreLeftRightM<"swr", MipsSWR, CPU64Regs>, LW_FM<0x2e>;
+
+defm LDL   : LoadLeftRightM<"ldl", MipsLDL, CPU64Regs>, LW_FM<0x1a>;
+defm LDR   : LoadLeftRightM<"ldr", MipsLDR, CPU64Regs>, LW_FM<0x1b>;
+defm SDL   : StoreLeftRightM<"sdl", MipsSDL, CPU64Regs>, LW_FM<0x2c>;
+defm SDR   : StoreLeftRightM<"sdr", MipsSDR, CPU64Regs>, LW_FM<0x2d>;
 
 /// Load-linked, Store-conditional
-def LLD    : LLBase<0x34, "lld", CPU64Regs, mem>,
-             Requires<[NotN64, HasStandardEncoding]>;
-def LLD_P8 : LLBase<0x34, "lld", CPU64Regs, mem64>,
-             Requires<[IsN64, HasStandardEncoding]> {
-  let isCodeGenOnly = 1;
+let Predicates = [NotN64, HasStdEnc] in {
+  def LLD : LLBase<"lld", CPU64RegsOpnd, mem>, LW_FM<0x34>;
+  def SCD : SCBase<"scd", CPU64RegsOpnd, mem>, LW_FM<0x3c>;
 }
-def SCD    : SCBase<0x3c, "scd", CPU64Regs, mem>,
-             Requires<[NotN64, HasStandardEncoding]>;
-def SCD_P8 : SCBase<0x3c, "scd", CPU64Regs, mem64>,
-             Requires<[IsN64, HasStandardEncoding]> {
-  let isCodeGenOnly = 1;
+
+let Predicates = [IsN64, HasStdEnc], isCodeGenOnly = 1 in {
+  def LLD_P8 : LLBase<"lld", CPU64RegsOpnd, mem64>, LW_FM<0x34>;
+  def SCD_P8 : SCBase<"scd", CPU64RegsOpnd, mem64>, LW_FM<0x3c>;
 }
 
 /// Jump and Branch Instructions
-def JR64   : IndirectBranch<CPU64Regs>;
-def BEQ64  : CBranch<0x04, "beq", seteq, CPU64Regs>;
-def BNE64  : CBranch<0x05, "bne", setne, CPU64Regs>;
-def BGEZ64 : CBranchZero<0x01, 1, "bgez", setge, CPU64Regs>;
-def BGTZ64 : CBranchZero<0x07, 0, "bgtz", setgt, CPU64Regs>;
-def BLEZ64 : CBranchZero<0x06, 0, "blez", setle, CPU64Regs>;
-def BLTZ64 : CBranchZero<0x01, 0, "bltz", setlt, CPU64Regs>;
+def JR64   : IndirectBranch<CPU64Regs>, MTLO_FM<8>;
+def BEQ64  : CBranch<"beq", seteq, CPU64Regs>, BEQ_FM<4>;
+def BNE64  : CBranch<"bne", setne, CPU64Regs>, BEQ_FM<5>;
+def BGEZ64 : CBranchZero<"bgez", setge, CPU64Regs>, BGEZ_FM<1, 1>;
+def BGTZ64 : CBranchZero<"bgtz", setgt, CPU64Regs>, BGEZ_FM<7, 0>;
+def BLEZ64 : CBranchZero<"blez", setle, CPU64Regs>, BGEZ_FM<6, 0>;
+def BLTZ64 : CBranchZero<"bltz", setlt, CPU64Regs>, BGEZ_FM<1, 0>;
 }
 let DecoderNamespace = "Mips64" in
-def JALR64 : JumpLinkReg<0x00, 0x09, "jalr", CPU64Regs>;
-def TAILCALL64_R : JumpFR<CPU64Regs, MipsTailCall>, IsTailCall;
+def JALR64 : JumpLinkReg<"jalr", CPU64Regs>, JALR_FM;
+def JALR64Pseudo : JumpLinkRegPseudo<CPU64Regs, JALR64, RA_64>;
+def TAILCALL64_R : JumpFR<CPU64Regs, MipsTailCall>, MTLO_FM<8>, IsTailCall;
 
 let DecoderNamespace = "Mips64" in {
 /// Multiply and Divide Instructions.
-def DMULT    : Mult64<0x1c, "dmult", IIImul>;
-def DMULTu   : Mult64<0x1d, "dmultu", IIImul>;
-def DSDIV    : Div64<MipsDivRem, 0x1e, "ddiv", IIIdiv>;
-def DUDIV    : Div64<MipsDivRemU, 0x1f, "ddivu", IIIdiv>;
-
-def MTHI64 : MoveToLOHI<0x11, "mthi", CPU64Regs, [HI64]>;
-def MTLO64 : MoveToLOHI<0x13, "mtlo", CPU64Regs, [LO64]>;
-def MFHI64 : MoveFromLOHI<0x10, "mfhi", CPU64Regs, [HI64]>;
-def MFLO64 : MoveFromLOHI<0x12, "mflo", CPU64Regs, [LO64]>;
+def DMULT  : Mult<"dmult", IIImul, CPU64RegsOpnd, [HI64, LO64]>,
+             MULT_FM<0, 0x1c>;
+def DMULTu : Mult<"dmultu", IIImul, CPU64RegsOpnd, [HI64, LO64]>,
+             MULT_FM<0, 0x1d>;
+def PseudoDMULT  : MultDivPseudo<DMULT, ACRegs128, CPU64RegsOpnd, MipsMult,
+                                 IIImul>;
+def PseudoDMULTu : MultDivPseudo<DMULTu, ACRegs128, CPU64RegsOpnd, MipsMultu,
+                                 IIImul>;
+def DSDIV : Div<"ddiv", IIIdiv, CPU64RegsOpnd, [HI64, LO64]>, MULT_FM<0, 0x1e>;
+def DUDIV : Div<"ddivu", IIIdiv, CPU64RegsOpnd, [HI64, LO64]>, MULT_FM<0, 0x1f>;
+def PseudoDSDIV : MultDivPseudo<DSDIV, ACRegs128, CPU64RegsOpnd, MipsDivRem,
+                                IIIdiv, 0>;
+def PseudoDUDIV : MultDivPseudo<DUDIV, ACRegs128, CPU64RegsOpnd, MipsDivRemU,
+                                IIIdiv, 0>;
+
+def MTHI64 : MoveToLOHI<"mthi", CPU64Regs, [HI64]>, MTLO_FM<0x11>;
+def MTLO64 : MoveToLOHI<"mtlo", CPU64Regs, [LO64]>, MTLO_FM<0x13>;
+def MFHI64 : MoveFromLOHI<"mfhi", CPU64Regs, [HI64]>, MFLO_FM<0x10>;
+def MFLO64 : MoveFromLOHI<"mflo", CPU64Regs, [LO64]>, MFLO_FM<0x12>;
 
 /// Sign Ext In Register Instructions.
-def SEB64 : SignExtInReg<0x10, "seb", i8, CPU64Regs>;
-def SEH64 : SignExtInReg<0x18, "seh", i16, CPU64Regs>;
+def SEB64 : SignExtInReg<"seb", i8, CPU64Regs>, SEB_FM<0x10, 0x20>;
+def SEH64 : SignExtInReg<"seh", i16, CPU64Regs>, SEB_FM<0x18, 0x20>;
 
 /// Count Leading
-def DCLZ : CountLeading0<0x24, "dclz", CPU64Regs>;
-def DCLO : CountLeading1<0x25, "dclo", CPU64Regs>;
+def DCLZ : CountLeading0<"dclz", CPU64RegsOpnd>, CLO_FM<0x24>;
+def DCLO : CountLeading1<"dclo", CPU64RegsOpnd>, CLO_FM<0x25>;
 
 /// Double Word Swap Bytes/HalfWords
-def DSBH : SubwordSwap<0x24, 0x2, "dsbh", CPU64Regs>;
-def DSHD : SubwordSwap<0x24, 0x5, "dshd", CPU64Regs>;
+def DSBH : SubwordSwap<"dsbh", CPU64RegsOpnd>, SEB_FM<2, 0x24>;
+def DSHD : SubwordSwap<"dshd", CPU64RegsOpnd>, SEB_FM<5, 0x24>;
+
+def LEA_ADDiu64 : EffectiveAddress<"daddiu", CPU64Regs, mem_ea_64>, LW_FM<0x19>;
 
-def LEA_ADDiu64 : EffectiveAddress<0x19,"daddiu\t$rt, $addr", CPU64Regs, mem_ea_64>;
 }
-let Uses = [SP_64], DecoderNamespace = "Mips64" in
-def DynAlloc64 : EffectiveAddress<0x19,"daddiu\t$rt, $addr", CPU64Regs, mem_ea_64>,
-                 Requires<[IsN64, HasStandardEncoding]>;
 let DecoderNamespace = "Mips64" in {
-def RDHWR64 : ReadHardware<CPU64Regs, HWRegs64>;
+def RDHWR64 : ReadHardware<CPU64Regs, HW64RegsOpnd>, RDHWR_FM;
 
-def DEXT : ExtBase<3, "dext", CPU64Regs>;
+def DEXT : ExtBase<"dext", CPU64RegsOpnd>, EXT_FM<3>;
 let Pattern = []<dag> in {
-  def DEXTU : ExtBase<2, "dextu", CPU64Regs>;
-  def DEXTM : ExtBase<1, "dextm", CPU64Regs>;
+  def DEXTU : ExtBase<"dextu", CPU64RegsOpnd>, EXT_FM<2>;
+  def DEXTM : ExtBase<"dextm", CPU64RegsOpnd>, EXT_FM<1>;
 }
-def DINS : InsBase<7, "dins", CPU64Regs>;
+def DINS : InsBase<"dins", CPU64RegsOpnd>, EXT_FM<7>;
 let Pattern = []<dag> in {
-  def DINSU : InsBase<6, "dinsu", CPU64Regs>;
-  def DINSM : InsBase<5, "dinsm", CPU64Regs>;
+  def DINSU : InsBase<"dinsu", CPU64RegsOpnd>, EXT_FM<6>;
+  def DINSM : InsBase<"dinsm", CPU64RegsOpnd>, EXT_FM<5>;
 }
 
 let isCodeGenOnly = 1, rs = 0, shamt = 0 in {
@@ -236,13 +246,13 @@ let isCodeGenOnly = 1, rs = 0, shamt = 0 in {
 //===----------------------------------------------------------------------===//
 
 // extended loads
-let Predicates = [NotN64, HasStandardEncoding] in {
+let Predicates = [NotN64, HasStdEnc] in {
   def : MipsPat<(i64 (extloadi1  addr:$src)), (LB64 addr:$src)>;
   def : MipsPat<(i64 (extloadi8  addr:$src)), (LB64 addr:$src)>;
   def : MipsPat<(i64 (extloadi16 addr:$src)), (LH64 addr:$src)>;
   def : MipsPat<(i64 (extloadi32 addr:$src)), (LW64 addr:$src)>;
 }
-let Predicates = [IsN64, HasStandardEncoding] in {
+let Predicates = [IsN64, HasStdEnc] in {
   def : MipsPat<(i64 (extloadi1  addr:$src)), (LB64_P8 addr:$src)>;
   def : MipsPat<(i64 (extloadi8  addr:$src)), (LB64_P8 addr:$src)>;
   def : MipsPat<(i64 (extloadi16 addr:$src)), (LH64_P8 addr:$src)>;
@@ -293,14 +303,10 @@ defm : SetgtPats<CPU64Regs, SLT64, SLTu64>;
 defm : SetgePats<CPU64Regs, SLT64, SLTu64>;
 defm : SetgeImmPats<CPU64Regs, SLTi64, SLTiu64>;
 
-// select MipsDynAlloc
-def : MipsPat<(MipsDynAlloc addr:$f), (DynAlloc64 addr:$f)>,
-      Requires<[IsN64, HasStandardEncoding]>;
-
 // truncate
 def : MipsPat<(i32 (trunc CPU64Regs:$src)),
               (SLL (EXTRACT_SUBREG CPU64Regs:$src, sub_32), 0)>,
-      Requires<[IsN64, HasStandardEncoding]>;
+      Requires<[IsN64, HasStdEnc]>;
 
 // 32-to-64-bit extension
 def : MipsPat<(i64 (anyext CPURegs:$src)), (SLL64_32 CPURegs:$src)>;
@@ -314,37 +320,73 @@ def : MipsPat<(i64 (sext_inreg CPU64Regs:$src, i32)),
 // bswap MipsPattern
 def : MipsPat<(bswap CPU64Regs:$rt), (DSHD (DSBH CPU64Regs:$rt))>;
 
+// mflo/hi patterns.
+def : MipsPat<(i64 (ExtractLOHI ACRegs128:$ac, imm:$lohi_idx)),
+              (EXTRACT_SUBREG ACRegs128:$ac, imm:$lohi_idx)>;
+
 //===----------------------------------------------------------------------===//
 // Instruction aliases
 //===----------------------------------------------------------------------===//
-def : InstAlias<"move $dst,$src", (DADD CPU64Regs:$dst,CPU64Regs:$src,ZERO_64)>;
-
+def : InstAlias<"move $dst, $src",
+                (DADDu CPU64RegsOpnd:$dst,  CPU64RegsOpnd:$src, ZERO_64), 1>,
+      Requires<[HasMips64]>;
+def : InstAlias<"move $dst, $src",
+                (OR64 CPU64RegsOpnd:$dst, CPU64RegsOpnd:$src, ZERO_64), 1>,
+      Requires<[HasMips64]>;
+def : InstAlias<"and $rs, $rt, $imm",
+                (DANDi CPU64RegsOpnd:$rs, CPU64RegsOpnd:$rt, uimm16_64:$imm),
+                1>,
+      Requires<[HasMips64]>;
+def : InstAlias<"slt $rs, $rt, $imm",
+                (SLTi64 CPURegsOpnd:$rs, CPU64Regs:$rt, simm16_64:$imm), 1>,
+      Requires<[HasMips64]>;
+def : InstAlias<"xor $rs, $rt, $imm",
+                (XORi64 CPU64RegsOpnd:$rs, CPU64RegsOpnd:$rt, uimm16_64:$imm),
+                1>,
+      Requires<[HasMips64]>;
+def : InstAlias<"not $rt, $rs",
+                (NOR64 CPU64RegsOpnd:$rt, CPU64RegsOpnd:$rs, ZERO_64), 1>,
+      Requires<[HasMips64]>;
+def : InstAlias<"j $rs", (JR64 CPU64Regs:$rs), 0>, Requires<[HasMips64]>;
+def : InstAlias<"jalr $rs", (JALR64 RA_64, CPU64Regs:$rs)>,
+      Requires<[HasMips64]>;
+def : InstAlias<"jal $rs", (JALR64 RA_64, CPU64Regs:$rs), 0>,
+                 Requires<[HasMips64]>;
+def : InstAlias<"jal $rd,$rs", (JALR64 CPU64Regs:$rd, CPU64Regs:$rs), 0>,
+                 Requires<[HasMips64]>;
+def : InstAlias<"daddu $rs, $rt, $imm",
+                (DADDiu CPU64RegsOpnd:$rs, CPU64RegsOpnd:$rt, simm16_64:$imm),
+                1>;
+def : InstAlias<"dadd $rs, $rt, $imm",
+                (DADDi CPU64RegsOpnd:$rs, CPU64RegsOpnd:$rt, simm16_64:$imm),
+                1>;
+def : InstAlias<"or $rs, $rt, $imm",
+                (ORi64 CPU64RegsOpnd:$rs, CPU64RegsOpnd:$rt, uimm16_64:$imm),
+                1>, Requires<[HasMips64]>;
 /// Move between CPU and coprocessor registers
+
 let DecoderNamespace = "Mips64" in {
-def MFC0_3OP64  : MFC3OP<0x10, 0, (outs CPU64Regs:$rt), 
-                       (ins CPU64Regs:$rd, uimm16:$sel),"mfc0\t$rt, $rd, $sel">;
-def MTC0_3OP64  : MFC3OP<0x10, 4, (outs CPU64Regs:$rd, uimm16:$sel),
-                       (ins CPU64Regs:$rt),"mtc0\t$rt, $rd, $sel">;
-def MFC2_3OP64  : MFC3OP<0x12, 0, (outs CPU64Regs:$rt),
-                       (ins CPU64Regs:$rd, uimm16:$sel),"mfc2\t$rt, $rd, $sel">;
-def MTC2_3OP64  : MFC3OP<0x12, 4, (outs CPU64Regs:$rd, uimm16:$sel),
-                       (ins CPU64Regs:$rt),"mtc2\t$rt, $rd, $sel">;
-def DMFC0_3OP64  : MFC3OP<0x10, 1, (outs CPU64Regs:$rt), 
-                       (ins CPU64Regs:$rd, uimm16:$sel),"dmfc0\t$rt, $rd, $sel">;
-def DMTC0_3OP64  : MFC3OP<0x10, 5, (outs CPU64Regs:$rd, uimm16:$sel),
-                       (ins CPU64Regs:$rt),"dmtc0\t$rt, $rd, $sel">;
-def DMFC2_3OP64  : MFC3OP<0x12, 1, (outs CPU64Regs:$rt),
-                       (ins CPU64Regs:$rd, uimm16:$sel),"dmfc2\t$rt, $rd, $sel">;
-def DMTC2_3OP64  : MFC3OP<0x12, 5, (outs CPU64Regs:$rd, uimm16:$sel),
-                       (ins CPU64Regs:$rt),"dmtc2\t$rt, $rd, $sel">;
+def DMFC0_3OP64 : MFC3OP<(outs CPU64RegsOpnd:$rt),
+                         (ins CPU64RegsOpnd:$rd, uimm16:$sel),
+                         "dmfc0\t$rt, $rd, $sel">, MFC3OP_FM<0x10, 1>;
+def DMTC0_3OP64 : MFC3OP<(outs CPU64RegsOpnd:$rd, uimm16:$sel),
+                         (ins CPU64RegsOpnd:$rt),
+                         "dmtc0\t$rt, $rd, $sel">, MFC3OP_FM<0x10, 5>;
+def DMFC2_3OP64 : MFC3OP<(outs CPU64RegsOpnd:$rt),
+                         (ins CPU64RegsOpnd:$rd, uimm16:$sel),
+                         "dmfc2\t$rt, $rd, $sel">, MFC3OP_FM<0x12, 1>;
+def DMTC2_3OP64 : MFC3OP<(outs CPU64RegsOpnd:$rd, uimm16:$sel),
+                         (ins CPU64RegsOpnd:$rt),
+                         "dmtc2\t$rt, $rd, $sel">, MFC3OP_FM<0x12, 5>;
 }
+
 // Two operand (implicit 0 selector) versions:
-def : InstAlias<"mfc0 $rt, $rd", (MFC0_3OP64 CPU64Regs:$rt, CPU64Regs:$rd, 0)>;
-def : InstAlias<"mtc0 $rt, $rd", (MTC0_3OP64 CPU64Regs:$rd, 0, CPU64Regs:$rt)>;
-def : InstAlias<"mfc2 $rt, $rd", (MFC2_3OP64 CPU64Regs:$rt, CPU64Regs:$rd, 0)>;
-def : InstAlias<"mtc2 $rt, $rd", (MTC2_3OP64 CPU64Regs:$rd, 0, CPU64Regs:$rt)>;
-def : InstAlias<"dmfc0 $rt, $rd", (DMFC0_3OP64 CPU64Regs:$rt, CPU64Regs:$rd, 0)>;
-def : InstAlias<"dmtc0 $rt, $rd", (DMTC0_3OP64 CPU64Regs:$rd, 0, CPU64Regs:$rt)>;
-def : InstAlias<"dmfc2 $rt, $rd", (DMFC2_3OP64 CPU64Regs:$rt, CPU64Regs:$rd, 0)>;
-def : InstAlias<"dmtc2 $rt, $rd", (DMTC2_3OP64 CPU64Regs:$rd, 0, CPU64Regs:$rt)>;
+def : InstAlias<"dmfc0 $rt, $rd",
+                (DMFC0_3OP64 CPU64RegsOpnd:$rt, CPU64RegsOpnd:$rd, 0), 0>;
+def : InstAlias<"dmtc0 $rt, $rd",
+                (DMTC0_3OP64 CPU64RegsOpnd:$rd, 0, CPU64RegsOpnd:$rt), 0>;
+def : InstAlias<"dmfc2 $rt, $rd",
+                (DMFC2_3OP64 CPU64RegsOpnd:$rt, CPU64RegsOpnd:$rd, 0), 0>;
+def : InstAlias<"dmtc2 $rt, $rd",
+                (DMTC2_3OP64 CPU64RegsOpnd:$rd, 0, CPU64RegsOpnd:$rt), 0>;
 
diff --git a/lib/Target/Mips/MipsAsmPrinter.cpp b/lib/Target/Mips/MipsAsmPrinter.cpp
index bf2818d61df0..1876cb6ffae4 100644
--- a/lib/Target/Mips/MipsAsmPrinter.cpp
+++ b/lib/Target/Mips/MipsAsmPrinter.cpp
@@ -13,31 +13,33 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "mips-asm-printer"
+#include "InstPrinter/MipsInstPrinter.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
+#include "MCTargetDesc/MipsELFStreamer.h"
 #include "Mips.h"
 #include "MipsAsmPrinter.h"
 #include "MipsInstrInfo.h"
 #include "MipsMCInstLower.h"
-#include "InstPrinter/MipsInstPrinter.h"
-#include "MCTargetDesc/MipsBaseInfo.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/Twine.h"
-#include "llvm/BasicBlock.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineMemOperand.h"
-#include "llvm/InlineAsm.h"
-#include "llvm/Instructions.h"
+#include "llvm/IR/BasicBlock.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/InlineAsm.h"
+#include "llvm/IR/Instructions.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/ELF.h"
 #include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/Mangler.h"
-#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLoweringObjectFile.h"
 #include "llvm/Target/TargetOptions.h"
 
@@ -65,19 +67,28 @@ void MipsAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     return;
   }
 
-  // Do any auto-generated pseudo lowerings.
-  if (emitPseudoExpansionLowering(OutStreamer, MI))
-    return;
-
   MachineBasicBlock::const_instr_iterator I = MI;
   MachineBasicBlock::const_instr_iterator E = MI->getParent()->instr_end();
 
   do {
-    MCInst TmpInst0;
-    MCInstLowering.Lower(I++, TmpInst0);
+    // Do any auto-generated pseudo lowerings.
+    if (emitPseudoExpansionLowering(OutStreamer, &*I))
+      continue;
+
+    // The inMips16Mode() test is not permanent.
+    // Some instructions are marked as pseudo right now which
+    // would make the test fail for the wrong reason but
+    // that will be fixed soon. We need this here because we are
+    // removing another test for this situation downstream in the
+    // callchain.
+    //
+    if (I->isPseudo() && !Subtarget->inMips16Mode())
+      llvm_unreachable("Pseudo opcode found in EmitInstruction()");
 
+    MCInst TmpInst0;
+    MCInstLowering.Lower(I, TmpInst0);
     OutStreamer.EmitInstruction(TmpInst0);
-  } while ((I != E) && I->isInsideBundle()); // Delay slot check
+  } while ((++I != E) && I->isInsideBundle()); // Delay slot check
 }
 
 //===----------------------------------------------------------------------===//
@@ -139,7 +150,7 @@ void MipsAsmPrinter::printSavedRegsBitmask(raw_ostream &O) {
     if (Mips::CPURegsRegClass.contains(Reg))
       break;
 
-    unsigned RegNum = getMipsRegisterNumbering(Reg);
+    unsigned RegNum = TM.getRegisterInfo()->getEncodingValue(Reg);
     if (Mips::AFGR64RegClass.contains(Reg)) {
       FPUBitmask |= (3 << RegNum);
       CSFPRegsSize += AFGR64RegSize;
@@ -154,7 +165,7 @@ void MipsAsmPrinter::printSavedRegsBitmask(raw_ostream &O) {
   // Set CPU Bitmask.
   for (; i != e; ++i) {
     unsigned Reg = CSI[i].getReg();
-    unsigned RegNum = getMipsRegisterNumbering(Reg);
+    unsigned RegNum = TM.getRegisterInfo()->getEncodingValue(Reg);
     CPUBitmask |= (1 << RegNum);
   }
 
@@ -221,6 +232,11 @@ void MipsAsmPrinter::EmitFunctionEntryLabel() {
     // OutStreamer.EmitRawText(StringRef("\t.set\tnomicromips"));
     OutStreamer.EmitRawText("\t.ent\t" + Twine(CurrentFnSym->getName()));
   }
+
+  if (Subtarget->inMicroMipsMode())
+    if (MipsELFStreamer *MES = dyn_cast<MipsELFStreamer>(&OutStreamer))
+      MES->emitMipsSTOCG(*Subtarget, CurrentFnSym,
+      (unsigned)ELF::STO_MIPS_MICROMIPS);
   OutStreamer.EmitLabel(CurrentFnSym);
 }
 
@@ -236,10 +252,11 @@ void MipsAsmPrinter::EmitFunctionBodyStart() {
     raw_svector_ostream OS(Str);
     printSavedRegsBitmask(OS);
     OutStreamer.EmitRawText(OS.str());
-
-    OutStreamer.EmitRawText(StringRef("\t.set\tnoreorder"));
-    OutStreamer.EmitRawText(StringRef("\t.set\tnomacro"));
-    OutStreamer.EmitRawText(StringRef("\t.set\tnoat"));
+    if (!Subtarget->inMips16Mode()) {
+      OutStreamer.EmitRawText(StringRef("\t.set\tnoreorder"));
+      OutStreamer.EmitRawText(StringRef("\t.set\tnomacro"));
+      OutStreamer.EmitRawText(StringRef("\t.set\tnoat"));
+    }
   }
 }
 
@@ -250,9 +267,11 @@ void MipsAsmPrinter::EmitFunctionBodyEnd() {
   // always be at the function end, and we can't emit and
   // break with BB logic.
   if (OutStreamer.hasRawTextSupport()) {
-    OutStreamer.EmitRawText(StringRef("\t.set\tat"));
-    OutStreamer.EmitRawText(StringRef("\t.set\tmacro"));
-    OutStreamer.EmitRawText(StringRef("\t.set\treorder"));
+    if (!Subtarget->inMips16Mode()) {
+      OutStreamer.EmitRawText(StringRef("\t.set\tat"));
+      OutStreamer.EmitRawText(StringRef("\t.set\tmacro"));
+      OutStreamer.EmitRawText(StringRef("\t.set\treorder"));
+    }
     OutStreamer.EmitRawText("\t.end\t" + Twine(CurrentFnSym->getName()));
   }
 }
@@ -540,6 +559,18 @@ void MipsAsmPrinter::EmitStartOfAsmFile(Module &M) {
   // return to previous section
   if (OutStreamer.hasRawTextSupport())
     OutStreamer.EmitRawText(StringRef("\t.previous"));
+
+}
+
+void MipsAsmPrinter::EmitEndOfAsmFile(Module &M) {
+
+  if (OutStreamer.hasRawTextSupport()) return;
+
+  // Emit Mips ELF register info
+  Subtarget->getMReginfo().emitMipsReginfoSectionCG(
+             OutStreamer, getObjFileLowering(), *Subtarget);
+  if (MipsELFStreamer *MES = dyn_cast<MipsELFStreamer>(&OutStreamer))
+    MES->emitELFHeaderFlagsCG(*Subtarget);
 }
 
 MachineLocation
diff --git a/lib/Target/Mips/MipsAsmPrinter.h b/lib/Target/Mips/MipsAsmPrinter.h
index 94d8bfa10569..dbdaf266b75f 100644
--- a/lib/Target/Mips/MipsAsmPrinter.h
+++ b/lib/Target/Mips/MipsAsmPrinter.h
@@ -14,8 +14,8 @@
 #ifndef MIPSASMPRINTER_H
 #define MIPSASMPRINTER_H
 
-#include "MipsMachineFunction.h"
 #include "MipsMCInstLower.h"
+#include "MipsMachineFunction.h"
 #include "MipsSubtarget.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/Support/Compiler.h"
@@ -80,6 +80,7 @@ public:
   void printFCCOperand(const MachineInstr *MI, int opNum, raw_ostream &O,
                        const char *Modifier = 0);
   void EmitStartOfAsmFile(Module &M);
+  void EmitEndOfAsmFile(Module &M);
   virtual MachineLocation getDebugValueLocation(const MachineInstr *MI) const;
   void PrintDebugValueComment(const MachineInstr *MI, raw_ostream &OS);
 };
diff --git a/lib/Target/Mips/MipsCallingConv.td b/lib/Target/Mips/MipsCallingConv.td
index 78cf140def60..462def76cc80 100644
--- a/lib/Target/Mips/MipsCallingConv.td
+++ b/lib/Target/Mips/MipsCallingConv.td
@@ -96,6 +96,12 @@ def RetCC_MipsN : CallingConv<[
   CCIfType<[f64], CCAssignToReg<[D0_64, D2_64]>>
 ]>;
 
+// In soft-mode, register A0_64, instead of V1_64, is used to return a long
+// double value.
+def RetCC_F128Soft : CallingConv<[
+  CCIfType<[i64], CCAssignToReg<[V0_64, A0_64]>>
+]>;
+
 //===----------------------------------------------------------------------===//
 // Mips EABI Calling Convention
 //===----------------------------------------------------------------------===//
@@ -139,17 +145,6 @@ def RetCC_MipsEABI : CallingConv<[
 ]>;
 
 //===----------------------------------------------------------------------===//
-// Mips Android Calling Convention
-//===----------------------------------------------------------------------===//
-
-def RetCC_MipsAndroid : CallingConv<[
-  // f32 are returned in registers F0, F2, F1, F3
-  CCIfType<[f32], CCAssignToReg<[F0, F2, F1, F3]>>,
-
-  CCDelegateTo<RetCC_MipsO32>
-]>;
-
-//===----------------------------------------------------------------------===//
 // Mips FastCC Calling Convention
 //===----------------------------------------------------------------------===//
 def CC_MipsO32_FastCC : CallingConv<[
@@ -209,7 +204,6 @@ def RetCC_Mips : CallingConv<[
   CCIfSubtarget<"isABI_EABI()", CCDelegateTo<RetCC_MipsEABI>>,
   CCIfSubtarget<"isABI_N32()", CCDelegateTo<RetCC_MipsN>>,
   CCIfSubtarget<"isABI_N64()", CCDelegateTo<RetCC_MipsN>>,
-  CCIfSubtarget<"isAndroid()", CCDelegateTo<RetCC_MipsAndroid>>,
   CCDelegateTo<RetCC_MipsO32>
 ]>;
 
diff --git a/lib/Target/Mips/MipsCodeEmitter.cpp b/lib/Target/Mips/MipsCodeEmitter.cpp
index 05090b84dece..1d86d903c12e 100644
--- a/lib/Target/Mips/MipsCodeEmitter.cpp
+++ b/lib/Target/Mips/MipsCodeEmitter.cpp
@@ -14,22 +14,23 @@
 
 #define DEBUG_TYPE "jit"
 #include "Mips.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
 #include "MipsInstrInfo.h"
 #include "MipsRelocations.h"
 #include "MipsSubtarget.h"
 #include "MipsTargetMachine.h"
-#include "MCTargetDesc/MipsBaseInfo.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/JITCodeEmitter.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineOperand.h"
 #include "llvm/CodeGen/Passes.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
 #include "llvm/PassManager.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
@@ -62,67 +63,77 @@ class MipsCodeEmitter : public MachineFunctionPass {
 
   static char ID;
 
-  public:
-    MipsCodeEmitter(TargetMachine &tm, JITCodeEmitter &mce) :
-      MachineFunctionPass(ID), JTI(0),
-      II((const MipsInstrInfo *) tm.getInstrInfo()),
-      TD(tm.getDataLayout()), TM(tm), MCE(mce), MCPEs(0), MJTEs(0),
-      IsPIC(TM.getRelocationModel() == Reloc::PIC_) {
-    }
+public:
+  MipsCodeEmitter(TargetMachine &tm, JITCodeEmitter &mce)
+    : MachineFunctionPass(ID), JTI(0),
+      II((const MipsInstrInfo *) tm.getInstrInfo()), TD(tm.getDataLayout()),
+      TM(tm), MCE(mce), MCPEs(0), MJTEs(0),
+      IsPIC(TM.getRelocationModel() == Reloc::PIC_) {}
 
-    bool runOnMachineFunction(MachineFunction &MF);
+  bool runOnMachineFunction(MachineFunction &MF);
 
-    virtual const char *getPassName() const {
-      return "Mips Machine Code Emitter";
-    }
+  virtual const char *getPassName() const {
+    return "Mips Machine Code Emitter";
+  }
+
+  /// getBinaryCodeForInstr - This function, generated by the
+  /// CodeEmitterGenerator using TableGen, produces the binary encoding for
+  /// machine instructions.
+  uint64_t getBinaryCodeForInstr(const MachineInstr &MI) const;
 
-    /// getBinaryCodeForInstr - This function, generated by the
-    /// CodeEmitterGenerator using TableGen, produces the binary encoding for
-    /// machine instructions.
-    uint64_t getBinaryCodeForInstr(const MachineInstr &MI) const;
+  void emitInstruction(MachineBasicBlock::instr_iterator MI,
+                       MachineBasicBlock &MBB);
 
-    void emitInstruction(const MachineInstr &MI);
+private:
 
-  private:
+  void emitWord(unsigned Word);
 
-    void emitWord(unsigned Word);
+  /// Routines that handle operands which add machine relocations which are
+  /// fixed up by the relocation stage.
+  void emitGlobalAddress(const GlobalValue *GV, unsigned Reloc,
+                         bool MayNeedFarStub) const;
+  void emitExternalSymbolAddress(const char *ES, unsigned Reloc) const;
+  void emitConstPoolAddress(unsigned CPI, unsigned Reloc) const;
+  void emitJumpTableAddress(unsigned JTIndex, unsigned Reloc) const;
+  void emitMachineBasicBlock(MachineBasicBlock *BB, unsigned Reloc) const;
 
-    /// Routines that handle operands which add machine relocations which are
-    /// fixed up by the relocation stage.
-    void emitGlobalAddress(const GlobalValue *GV, unsigned Reloc,
-                           bool MayNeedFarStub) const;
-    void emitExternalSymbolAddress(const char *ES, unsigned Reloc) const;
-    void emitConstPoolAddress(unsigned CPI, unsigned Reloc) const;
-    void emitJumpTableAddress(unsigned JTIndex, unsigned Reloc) const;
-    void emitMachineBasicBlock(MachineBasicBlock *BB, unsigned Reloc) const;
+  /// getMachineOpValue - Return binary encoding of operand. If the machine
+  /// operand requires relocation, record the relocation and return zero.
+  unsigned getMachineOpValue(const MachineInstr &MI,
+                             const MachineOperand &MO) const;
 
-    /// getMachineOpValue - Return binary encoding of operand. If the machine
-    /// operand requires relocation, record the relocation and return zero.
-    unsigned getMachineOpValue(const MachineInstr &MI,
-                               const MachineOperand &MO) const;
+  unsigned getRelocation(const MachineInstr &MI,
+                         const MachineOperand &MO) const;
 
-    unsigned getRelocation(const MachineInstr &MI,
-                           const MachineOperand &MO) const;
+  unsigned getJumpTargetOpValue(const MachineInstr &MI, unsigned OpNo) const;
 
-    unsigned getJumpTargetOpValue(const MachineInstr &MI, unsigned OpNo) const;
+  unsigned getBranchTargetOpValue(const MachineInstr &MI, unsigned OpNo) const;
+  unsigned getMemEncoding(const MachineInstr &MI, unsigned OpNo) const;
+  unsigned getSizeExtEncoding(const MachineInstr &MI, unsigned OpNo) const;
+  unsigned getSizeInsEncoding(const MachineInstr &MI, unsigned OpNo) const;
 
-    unsigned getBranchTargetOpValue(const MachineInstr &MI,
-                                    unsigned OpNo) const;
-    unsigned getMemEncoding(const MachineInstr &MI, unsigned OpNo) const;
-    unsigned getSizeExtEncoding(const MachineInstr &MI, unsigned OpNo) const;
-    unsigned getSizeInsEncoding(const MachineInstr &MI, unsigned OpNo) const;
+  void emitGlobalAddressUnaligned(const GlobalValue *GV, unsigned Reloc,
+                                  int Offset) const;
 
-    void emitGlobalAddressUnaligned(const GlobalValue *GV, unsigned Reloc,
-                                    int Offset) const;
-  };
+  /// Expand pseudo instructions with accumulator register operands.
+  void expandACCInstr(MachineBasicBlock::instr_iterator &MI,
+                      MachineBasicBlock &MBB, unsigned Opc) const;
+
+  /// \brief Expand pseudo instruction. Return true if MI was expanded.
+  bool expandPseudos(MachineBasicBlock::instr_iterator &MI,
+                     MachineBasicBlock &MBB) const;
+};
 }
 
 char MipsCodeEmitter::ID = 0;
 
 bool MipsCodeEmitter::runOnMachineFunction(MachineFunction &MF) {
-  JTI = ((MipsTargetMachine&) MF.getTarget()).getJITInfo();
-  II = ((const MipsTargetMachine&) MF.getTarget()).getInstrInfo();
-  TD = ((const MipsTargetMachine&) MF.getTarget()).getDataLayout();
+  MipsTargetMachine &Target = static_cast<MipsTargetMachine &>(
+                                const_cast<TargetMachine &>(MF.getTarget()));
+
+  JTI = Target.getJITInfo();
+  II = Target.getInstrInfo();
+  TD = Target.getDataLayout();
   Subtarget = &TM.getSubtarget<MipsSubtarget> ();
   MCPEs = &MF.getConstantPool()->getConstants();
   MJTEs = 0;
@@ -139,8 +150,8 @@ bool MipsCodeEmitter::runOnMachineFunction(MachineFunction &MF) {
         MBB != E; ++MBB){
       MCE.StartMachineBasicBlock(MBB);
       for (MachineBasicBlock::instr_iterator I = MBB->instr_begin(),
-           E = MBB->instr_end(); I != E; ++I)
-        emitInstruction(*I);
+           E = MBB->instr_end(); I != E;)
+        emitInstruction(*I++, *MBB);
     }
   } while (MCE.finishFunction(MF));
 
@@ -209,7 +220,7 @@ unsigned MipsCodeEmitter::getSizeInsEncoding(const MachineInstr &MI,
 unsigned MipsCodeEmitter::getMachineOpValue(const MachineInstr &MI,
                                             const MachineOperand &MO) const {
   if (MO.isReg())
-    return getMipsRegisterNumbering(MO.getReg());
+    return TM.getRegisterInfo()->getEncodingValue(MO.getReg());
   else if (MO.isImm())
     return static_cast<unsigned>(MO.getImm());
   else if (MO.isGlobal())
@@ -265,19 +276,21 @@ void MipsCodeEmitter::emitMachineBasicBlock(MachineBasicBlock *BB,
                                              Reloc, BB));
 }
 
-void MipsCodeEmitter::emitInstruction(const MachineInstr &MI) {
-  DEBUG(errs() << "JIT: " << (void*)MCE.getCurrentPCValue() << ":\t" << MI);
+void MipsCodeEmitter::emitInstruction(MachineBasicBlock::instr_iterator MI,
+                                      MachineBasicBlock &MBB) {
+  DEBUG(errs() << "JIT: " << (void*)MCE.getCurrentPCValue() << ":\t" << *MI);
 
-  MCE.processDebugLoc(MI.getDebugLoc(), true);
-
-  // Skip pseudo instructions.
-  if ((MI.getDesc().TSFlags & MipsII::FormMask) == MipsII::Pseudo)
+  // Expand pseudo instruction. Skip if MI was not expanded.
+  if (((MI->getDesc().TSFlags & MipsII::FormMask) == MipsII::Pseudo) &&
+      !expandPseudos(MI, MBB))
     return;
 
-  emitWord(getBinaryCodeForInstr(MI));
+  MCE.processDebugLoc(MI->getDebugLoc(), true);
+
+  emitWord(getBinaryCodeForInstr(*MI));
   ++NumEmitted;  // Keep track of the # of mi's emitted
 
-  MCE.processDebugLoc(MI.getDebugLoc(), false);
+  MCE.processDebugLoc(MI->getDebugLoc(), false);
 }
 
 void MipsCodeEmitter::emitWord(unsigned Word) {
@@ -289,6 +302,57 @@ void MipsCodeEmitter::emitWord(unsigned Word) {
     MCE.emitWordBE(Word);
 }
 
+void MipsCodeEmitter::expandACCInstr(MachineBasicBlock::instr_iterator &MI,
+                                     MachineBasicBlock &MBB,
+                                     unsigned Opc) const {
+  // Expand "pseudomult $ac0, $t0, $t1" to "mult $t0, $t1".
+  BuildMI(MBB, &*MI, MI->getDebugLoc(), II->get(Opc))
+    .addReg(MI->getOperand(1).getReg()).addReg(MI->getOperand(2).getReg());
+}
+
+bool MipsCodeEmitter::expandPseudos(MachineBasicBlock::instr_iterator &MI,
+                                    MachineBasicBlock &MBB) const {
+  switch (MI->getOpcode()) {
+  case Mips::NOP:
+    BuildMI(MBB, &*MI, MI->getDebugLoc(), II->get(Mips::SLL), Mips::ZERO)
+      .addReg(Mips::ZERO).addImm(0);
+    break;
+  case Mips::JALRPseudo:
+    BuildMI(MBB, &*MI, MI->getDebugLoc(), II->get(Mips::JALR), Mips::RA)
+      .addReg(MI->getOperand(0).getReg());
+    break;
+  case Mips::PseudoMULT:
+    expandACCInstr(MI, MBB, Mips::MULT);
+    break;
+  case Mips::PseudoMULTu:
+    expandACCInstr(MI, MBB, Mips::MULTu);
+    break;
+  case Mips::PseudoSDIV:
+    expandACCInstr(MI, MBB, Mips::SDIV);
+    break;
+  case Mips::PseudoUDIV:
+    expandACCInstr(MI, MBB, Mips::UDIV);
+    break;
+  case Mips::PseudoMADD:
+    expandACCInstr(MI, MBB, Mips::MADD);
+    break;
+  case Mips::PseudoMADDU:
+    expandACCInstr(MI, MBB, Mips::MADDU);
+    break;
+  case Mips::PseudoMSUB:
+    expandACCInstr(MI, MBB, Mips::MSUB);
+    break;
+  case Mips::PseudoMSUBU:
+    expandACCInstr(MI, MBB, Mips::MSUBU);
+    break;
+  default:
+    return false;
+  }
+
+  (MI--)->eraseFromBundle();
+  return true;
+}
+
 /// createMipsJITCodeEmitterPass - Return a pass that emits the collected Mips
 /// code to the specified MCE object.
 FunctionPass *llvm::createMipsJITCodeEmitterPass(MipsTargetMachine &TM,
diff --git a/lib/Target/Mips/MipsCondMov.td b/lib/Target/Mips/MipsCondMov.td
index b12b1f2b5ad4..42e4c99f05d6 100644
--- a/lib/Target/Mips/MipsCondMov.td
+++ b/lib/Target/Mips/MipsCondMov.td
@@ -16,42 +16,37 @@
 // MipsISelLowering::EmitInstrWithCustomInserter if target does not have
 // conditional move instructions.
 // cond:int, data:int
-class CondMovIntInt<RegisterClass CRC, RegisterClass DRC, bits<6> funct,
-                    string instr_asm> :
-  FR<0, funct, (outs DRC:$rd), (ins DRC:$rs, CRC:$rt, DRC:$F),
-     !strconcat(instr_asm, "\t$rd, $rs, $rt"), [], NoItinerary> {
-  let shamt = 0;
+class CMov_I_I_FT<string opstr, RegisterClass CRC, RegisterClass DRC,
+                  InstrItinClass Itin> :
+  InstSE<(outs DRC:$rd), (ins DRC:$rs, CRC:$rt, DRC:$F),
+         !strconcat(opstr, "\t$rd, $rs, $rt"), [], Itin, FrmFR> {
   let Constraints = "$F = $rd";
 }
 
 // cond:int, data:float
-class CondMovIntFP<RegisterClass CRC, RegisterClass DRC, bits<5> fmt,
-                   bits<6> func, string instr_asm> :
-  FFR<0x11, func, fmt, (outs DRC:$fd), (ins DRC:$fs, CRC:$rt, DRC:$F),
-      !strconcat(instr_asm, "\t$fd, $fs, $rt"), []> {
-  bits<5> rt;
-  let ft = rt;
+class CMov_I_F_FT<string opstr, RegisterClass CRC, RegisterClass DRC,
+                  InstrItinClass Itin> :
+  InstSE<(outs DRC:$fd), (ins DRC:$fs, CRC:$rt, DRC:$F),
+         !strconcat(opstr, "\t$fd, $fs, $rt"), [], Itin, FrmFR> {
   let Constraints = "$F = $fd";
 }
 
 // cond:float, data:int
-class CondMovFPInt<RegisterClass RC, SDNode cmov, bits<1> tf,
-                   string instr_asm> :
-  FCMOV<tf, (outs RC:$rd), (ins RC:$rs, RC:$F),
-        !strconcat(instr_asm, "\t$rd, $rs, $$fcc0"),
-        [(set RC:$rd, (cmov RC:$rs, RC:$F))]> {
-  let cc = 0;
+class CMov_F_I_FT<string opstr, RegisterClass RC, InstrItinClass Itin,
+                  SDPatternOperator OpNode = null_frag> :
+  InstSE<(outs RC:$rd), (ins RC:$rs, RC:$F),
+         !strconcat(opstr, "\t$rd, $rs, $$fcc0"),
+         [(set RC:$rd, (OpNode RC:$rs, RC:$F))], Itin, FrmFR> {
   let Uses = [FCR31];
   let Constraints = "$F = $rd";
 }
 
 // cond:float, data:float
-class CondMovFPFP<RegisterClass RC, SDNode cmov, bits<5> fmt, bits<1> tf,
-                  string instr_asm> :
-  FFCMOV<fmt, tf, (outs RC:$fd), (ins RC:$fs, RC:$F),
-         !strconcat(instr_asm, "\t$fd, $fs, $$fcc0"),
-         [(set RC:$fd, (cmov RC:$fs, RC:$F))]> {
-  let cc = 0;
+class CMov_F_F_FT<string opstr, RegisterClass RC, InstrItinClass Itin,
+                  SDPatternOperator OpNode = null_frag> :
+  InstSE<(outs RC:$fd), (ins RC:$fs, RC:$F),
+         !strconcat(opstr, "\t$fd, $fs, $$fcc0"),
+         [(set RC:$fd, (OpNode RC:$fs, RC:$F))], Itin, FrmFR> {
   let Uses = [FCR31];
   let Constraints = "$F = $fd";
 }
@@ -63,21 +58,23 @@ multiclass MovzPats0<RegisterClass CRC, RegisterClass DRC,
                      Instruction SLTiuOp> {
   def : MipsPat<(select (i32 (setge CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F),
                 (MOVZInst DRC:$T, (SLTOp CRC:$lhs, CRC:$rhs), DRC:$F)>;
-  def : MipsPat<
-          (select (i32 (setuge CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F),
-          (MOVZInst DRC:$T, (SLTuOp CRC:$lhs, CRC:$rhs), DRC:$F)>;
-  def : MipsPat<
-          (select (i32 (setge CRC:$lhs, immSExt16:$rhs)), DRC:$T, DRC:$F),
-          (MOVZInst DRC:$T, (SLTiOp CRC:$lhs, immSExt16:$rhs), DRC:$F)>;
-  def : MipsPat<
-          (select (i32 (setuge CRC:$lh, immSExt16:$rh)), DRC:$T, DRC:$F),
-          (MOVZInst DRC:$T, (SLTiuOp CRC:$lh, immSExt16:$rh), DRC:$F)>;
-  def : MipsPat<
-          (select (i32 (setle CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F),
-          (MOVZInst DRC:$T, (SLTOp CRC:$rhs, CRC:$lhs), DRC:$F)>;
-  def : MipsPat<
-          (select (i32 (setule CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F),
-          (MOVZInst DRC:$T, (SLTuOp CRC:$rhs, CRC:$lhs), DRC:$F)>;
+  def : MipsPat<(select (i32 (setuge CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F),
+                (MOVZInst DRC:$T, (SLTuOp CRC:$lhs, CRC:$rhs), DRC:$F)>;
+  def : MipsPat<(select (i32 (setge CRC:$lhs, immSExt16:$rhs)), DRC:$T, DRC:$F),
+                (MOVZInst DRC:$T, (SLTiOp CRC:$lhs, immSExt16:$rhs), DRC:$F)>;
+  def : MipsPat<(select (i32 (setuge CRC:$lh, immSExt16:$rh)), DRC:$T, DRC:$F),
+                (MOVZInst DRC:$T, (SLTiuOp CRC:$lh, immSExt16:$rh), DRC:$F)>;
+  def : MipsPat<(select (i32 (setle CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F),
+                (MOVZInst DRC:$T, (SLTOp CRC:$rhs, CRC:$lhs), DRC:$F)>;
+  def : MipsPat<(select (i32 (setule CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F),
+                (MOVZInst DRC:$T, (SLTuOp CRC:$rhs, CRC:$lhs), DRC:$F)>;
+  def : MipsPat<(select (i32 (setgt CRC:$lhs, immSExt16Plus1:$rhs)),
+                        DRC:$T, DRC:$F),
+                (MOVZInst DRC:$T, (SLTiOp CRC:$lhs, (Plus1 imm:$rhs)), DRC:$F)>;
+  def : MipsPat<(select (i32 (setugt CRC:$lhs, immSExt16Plus1:$rhs)),
+                        DRC:$T, DRC:$F),
+                (MOVZInst DRC:$T, (SLTiuOp CRC:$lhs, (Plus1 imm:$rhs)),
+                          DRC:$F)>;
 }
 
 multiclass MovzPats1<RegisterClass CRC, RegisterClass DRC,
@@ -106,88 +103,110 @@ multiclass MovnPats<RegisterClass CRC, RegisterClass DRC, Instruction MOVNInst,
 }
 
 // Instantiation of instructions.
-def MOVZ_I_I     : CondMovIntInt<CPURegs, CPURegs, 0x0a, "movz">;
-let Predicates = [HasMips64, HasStandardEncoding],
+def MOVZ_I_I : CMov_I_I_FT<"movz", CPURegs, CPURegs, NoItinerary>,
+               ADD_FM<0, 0xa>;
+let Predicates = [HasStdEnc],
                   DecoderNamespace = "Mips64" in {
-  def MOVZ_I_I64   : CondMovIntInt<CPURegs, CPU64Regs, 0x0a, "movz">;
-  def MOVZ_I64_I   : CondMovIntInt<CPU64Regs, CPURegs, 0x0a, "movz"> {
+  def MOVZ_I_I64   : CMov_I_I_FT<"movz", CPURegs, CPU64Regs, NoItinerary>,
+                     ADD_FM<0, 0xa>;
+  def MOVZ_I64_I   : CMov_I_I_FT<"movz", CPU64Regs, CPURegs, NoItinerary>,
+                     ADD_FM<0, 0xa> {
     let isCodeGenOnly = 1;
   }
-  def MOVZ_I64_I64 : CondMovIntInt<CPU64Regs, CPU64Regs, 0x0a, "movz"> {
+  def MOVZ_I64_I64 : CMov_I_I_FT<"movz", CPU64Regs, CPU64Regs, NoItinerary>,
+                     ADD_FM<0, 0xa> {
     let isCodeGenOnly = 1;
   }
 }
 
-def MOVN_I_I     : CondMovIntInt<CPURegs, CPURegs, 0x0b, "movn">;
-let Predicates = [HasMips64, HasStandardEncoding],
+def MOVN_I_I       : CMov_I_I_FT<"movn", CPURegs, CPURegs, NoItinerary>,
+                     ADD_FM<0, 0xb>;
+let Predicates = [HasStdEnc],
                   DecoderNamespace = "Mips64" in {
-  def MOVN_I_I64   : CondMovIntInt<CPURegs, CPU64Regs, 0x0b, "movn">;
-  def MOVN_I64_I   : CondMovIntInt<CPU64Regs, CPURegs, 0x0b, "movn"> {
+  def MOVN_I_I64   : CMov_I_I_FT<"movn", CPURegs, CPU64Regs, NoItinerary>,
+                     ADD_FM<0, 0xb>;
+  def MOVN_I64_I   : CMov_I_I_FT<"movn", CPU64Regs, CPURegs, NoItinerary>,
+                     ADD_FM<0, 0xb> {
     let isCodeGenOnly = 1;
   }
-  def MOVN_I64_I64 : CondMovIntInt<CPU64Regs, CPU64Regs, 0x0b, "movn"> {
+  def MOVN_I64_I64 : CMov_I_I_FT<"movn", CPU64Regs, CPU64Regs, NoItinerary>,
+                     ADD_FM<0, 0xb> {
     let isCodeGenOnly = 1;
   }
 }
 
-def MOVZ_I_S   : CondMovIntFP<CPURegs, FGR32, 16, 18, "movz.s">;
-def MOVZ_I64_S : CondMovIntFP<CPU64Regs, FGR32, 16, 18, "movz.s">,
-                 Requires<[HasMips64, HasStandardEncoding]> {
+def MOVZ_I_S : CMov_I_F_FT<"movz.s", CPURegs, FGR32, IIFmove>,
+               CMov_I_F_FM<18, 16>;
+def MOVZ_I64_S : CMov_I_F_FT<"movz.s", CPU64Regs, FGR32, IIFmove>,
+                 CMov_I_F_FM<18, 16>, Requires<[HasMips64, HasStdEnc]> {
   let DecoderNamespace = "Mips64";
 }
 
-def MOVN_I_S   : CondMovIntFP<CPURegs, FGR32, 16, 19, "movn.s">;
-def MOVN_I64_S : CondMovIntFP<CPU64Regs, FGR32, 16, 19, "movn.s">,
-                 Requires<[HasMips64, HasStandardEncoding]> {
+def MOVN_I_S : CMov_I_F_FT<"movn.s", CPURegs, FGR32, IIFmove>,
+               CMov_I_F_FM<19, 16>;
+def MOVN_I64_S : CMov_I_F_FT<"movn.s", CPU64Regs, FGR32, IIFmove>,
+                 CMov_I_F_FM<19, 16>, Requires<[HasMips64, HasStdEnc]> {
   let DecoderNamespace = "Mips64";
 }
 
-let Predicates = [NotFP64bit, HasStandardEncoding] in {
-  def MOVZ_I_D32   : CondMovIntFP<CPURegs, AFGR64, 17, 18, "movz.d">;
-  def MOVN_I_D32   : CondMovIntFP<CPURegs, AFGR64, 17, 19, "movn.d">;
+let Predicates = [NotFP64bit, HasStdEnc] in {
+  def MOVZ_I_D32 : CMov_I_F_FT<"movz.d", CPURegs, AFGR64, IIFmove>,
+                   CMov_I_F_FM<18, 17>;
+  def MOVN_I_D32 : CMov_I_F_FT<"movn.d", CPURegs, AFGR64, IIFmove>,
+                   CMov_I_F_FM<19, 17>;
 }
-let Predicates = [IsFP64bit, HasStandardEncoding],
+let Predicates = [IsFP64bit, HasStdEnc],
                   DecoderNamespace = "Mips64" in {
-  def MOVZ_I_D64   : CondMovIntFP<CPURegs, FGR64, 17, 18, "movz.d">;
-  def MOVZ_I64_D64 : CondMovIntFP<CPU64Regs, FGR64, 17, 18, "movz.d"> {
+  def MOVZ_I_D64 : CMov_I_F_FT<"movz.d", CPURegs, FGR64, IIFmove>,
+                   CMov_I_F_FM<18, 17>;
+  def MOVZ_I64_D64 : CMov_I_F_FT<"movz.d", CPU64Regs, FGR64, IIFmove>,
+                     CMov_I_F_FM<18, 17> {
     let isCodeGenOnly = 1;
   }
-  def MOVN_I_D64   : CondMovIntFP<CPURegs, FGR64, 17, 19, "movn.d">;
-  def MOVN_I64_D64 : CondMovIntFP<CPU64Regs, FGR64, 17, 19, "movn.d"> {
+  def MOVN_I_D64 : CMov_I_F_FT<"movn.d", CPURegs, FGR64, IIFmove>,
+                   CMov_I_F_FM<19, 17>;
+  def MOVN_I64_D64 : CMov_I_F_FT<"movn.d", CPU64Regs, FGR64, IIFmove>,
+                     CMov_I_F_FM<19, 17> {
     let isCodeGenOnly = 1;
   }
 }
 
-def MOVT_I   : CondMovFPInt<CPURegs, MipsCMovFP_T, 1, "movt">;
-def MOVT_I64 : CondMovFPInt<CPU64Regs, MipsCMovFP_T, 1, "movt">,
-               Requires<[HasMips64, HasStandardEncoding]> {
+def MOVT_I : CMov_F_I_FT<"movt", CPURegs, IIAlu, MipsCMovFP_T>, CMov_F_I_FM<1>;
+def MOVT_I64 : CMov_F_I_FT<"movt", CPU64Regs, IIAlu, MipsCMovFP_T>,
+               CMov_F_I_FM<1>, Requires<[HasMips64, HasStdEnc]> {
   let DecoderNamespace = "Mips64";
 }
 
-def MOVF_I   : CondMovFPInt<CPURegs, MipsCMovFP_F, 0, "movf">;
-def MOVF_I64 : CondMovFPInt<CPU64Regs, MipsCMovFP_F, 0, "movf">,
-               Requires<[HasMips64, HasStandardEncoding]> {
+def MOVF_I : CMov_F_I_FT<"movf", CPURegs, IIAlu, MipsCMovFP_F>, CMov_F_I_FM<0>;
+def MOVF_I64 : CMov_F_I_FT<"movf", CPU64Regs, IIAlu, MipsCMovFP_F>,
+               CMov_F_I_FM<0>, Requires<[HasMips64, HasStdEnc]> {
   let DecoderNamespace = "Mips64";
 }
 
-def MOVT_S : CondMovFPFP<FGR32, MipsCMovFP_T, 16, 1, "movt.s">;
-def MOVF_S : CondMovFPFP<FGR32, MipsCMovFP_F, 16, 0, "movf.s">;
+def MOVT_S : CMov_F_F_FT<"movt.s", FGR32, IIFmove, MipsCMovFP_T>,
+             CMov_F_F_FM<16, 1>;
+def MOVF_S : CMov_F_F_FT<"movf.s", FGR32, IIFmove, MipsCMovFP_F>,
+             CMov_F_F_FM<16, 0>;
 
-let Predicates = [NotFP64bit, HasStandardEncoding] in {
-  def MOVT_D32 : CondMovFPFP<AFGR64, MipsCMovFP_T, 17, 1, "movt.d">;
-  def MOVF_D32 : CondMovFPFP<AFGR64, MipsCMovFP_F, 17, 0, "movf.d">;
+let Predicates = [NotFP64bit, HasStdEnc] in {
+  def MOVT_D32 : CMov_F_F_FT<"movt.d", AFGR64, IIFmove, MipsCMovFP_T>,
+                 CMov_F_F_FM<17, 1>;
+  def MOVF_D32 : CMov_F_F_FT<"movf.d", AFGR64, IIFmove, MipsCMovFP_F>,
+                 CMov_F_F_FM<17, 0>;
 }
-let Predicates = [IsFP64bit, HasStandardEncoding],
+let Predicates = [IsFP64bit, HasStdEnc],
     DecoderNamespace = "Mips64" in {
-  def MOVT_D64 : CondMovFPFP<FGR64, MipsCMovFP_T, 17, 1, "movt.d">;
-  def MOVF_D64 : CondMovFPFP<FGR64, MipsCMovFP_F, 17, 0, "movf.d">;
+  def MOVT_D64 : CMov_F_F_FT<"movt.d", FGR64, IIFmove, MipsCMovFP_T>,
+                 CMov_F_F_FM<17, 1>;
+  def MOVF_D64 : CMov_F_F_FT<"movf.d", FGR64, IIFmove, MipsCMovFP_F>,
+                 CMov_F_F_FM<17, 0>;
 }
 
 // Instantiation of conditional move patterns.
 defm : MovzPats0<CPURegs, CPURegs, MOVZ_I_I, SLT, SLTu, SLTi, SLTiu>;
 defm : MovzPats1<CPURegs, CPURegs, MOVZ_I_I, XOR>;
 defm : MovzPats2<CPURegs, CPURegs, MOVZ_I_I, XORi>;
-let Predicates = [HasMips64, HasStandardEncoding] in {
+let Predicates = [HasMips64, HasStdEnc] in {
   defm : MovzPats0<CPURegs, CPU64Regs, MOVZ_I_I64, SLT, SLTu, SLTi, SLTiu>;
   defm : MovzPats0<CPU64Regs, CPURegs, MOVZ_I_I, SLT64, SLTu64, SLTi64,
                    SLTiu64>;
@@ -202,7 +221,7 @@ let Predicates = [HasMips64, HasStandardEncoding] in {
 }
 
 defm : MovnPats<CPURegs, CPURegs, MOVN_I_I, XOR>;
-let Predicates = [HasMips64, HasStandardEncoding] in {
+let Predicates = [HasMips64, HasStdEnc] in {
   defm : MovnPats<CPURegs, CPU64Regs, MOVN_I_I64, XOR>;
   defm : MovnPats<CPU64Regs, CPURegs, MOVN_I64_I, XOR64>;
   defm : MovnPats<CPU64Regs, CPU64Regs, MOVN_I64_I64, XOR64>;
@@ -211,19 +230,19 @@ let Predicates = [HasMips64, HasStandardEncoding] in {
 defm : MovzPats0<CPURegs, FGR32, MOVZ_I_S, SLT, SLTu, SLTi, SLTiu>;
 defm : MovzPats1<CPURegs, FGR32, MOVZ_I_S, XOR>;
 defm : MovnPats<CPURegs, FGR32, MOVN_I_S, XOR>;
-let Predicates = [HasMips64, HasStandardEncoding] in {
+let Predicates = [HasMips64, HasStdEnc] in {
   defm : MovzPats0<CPU64Regs, FGR32, MOVZ_I_S, SLT64, SLTu64, SLTi64,
                    SLTiu64>;
   defm : MovzPats1<CPU64Regs, FGR32, MOVZ_I64_S, XOR64>;
   defm : MovnPats<CPU64Regs, FGR32, MOVN_I64_S, XOR64>;
 }
 
-let Predicates = [NotFP64bit, HasStandardEncoding] in {
+let Predicates = [NotFP64bit, HasStdEnc] in {
   defm : MovzPats0<CPURegs, AFGR64, MOVZ_I_D32, SLT, SLTu, SLTi, SLTiu>;
   defm : MovzPats1<CPURegs, AFGR64, MOVZ_I_D32, XOR>;
   defm : MovnPats<CPURegs, AFGR64, MOVN_I_D32, XOR>;
 }
-let Predicates = [IsFP64bit, HasStandardEncoding] in {
+let Predicates = [IsFP64bit, HasStdEnc] in {
   defm : MovzPats0<CPURegs, FGR64, MOVZ_I_D64, SLT, SLTu, SLTi, SLTiu>;
   defm : MovzPats0<CPU64Regs, FGR64, MOVZ_I_D64, SLT64, SLTu64, SLTi64,
                    SLTiu64>;
diff --git a/lib/Target/Mips/MipsConstantIslandPass.cpp b/lib/Target/Mips/MipsConstantIslandPass.cpp
new file mode 100644
index 000000000000..b5de1ebad22b
--- /dev/null
+++ b/lib/Target/Mips/MipsConstantIslandPass.cpp
@@ -0,0 +1,85 @@
+//===-- MipsConstantIslandPass.cpp - Emit Pc Relative loads----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+//
+// This pass is used to make Pc relative loads of constants.
+// For now, only Mips16 will use this. While it has the same name and
+// uses many ideas from the LLVM ARM Constant Island Pass, it's not intended
+// to reuse any of the code from the ARM version.
+//
+// Loading constants inline is expensive on Mips16 and it's in general better
+// to place the constant nearby in code space and then it can be loaded with a
+// simple 16 bit load instruction.
+//
+// The constants can be not just numbers but addresses of functions and labels.
+// This can be particularly helpful in static relocation mode for embedded
+// non linux targets.
+//
+//
+
+#define DEBUG_TYPE "mips-constant-islands"
+
+#include "Mips.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
+#include "MipsTargetMachine.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/IR/Function.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/MathExtras.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetRegisterInfo.h"
+
+using namespace llvm;
+
+namespace {
+  typedef MachineBasicBlock::iterator Iter;
+  typedef MachineBasicBlock::reverse_iterator ReverseIter;
+
+  class MipsConstantIslands : public MachineFunctionPass {
+
+  public:
+    static char ID;
+    MipsConstantIslands(TargetMachine &tm)
+      : MachineFunctionPass(ID), TM(tm),
+        TII(static_cast<const MipsInstrInfo*>(tm.getInstrInfo())),
+        IsPIC(TM.getRelocationModel() == Reloc::PIC_),
+        ABI(TM.getSubtarget<MipsSubtarget>().getTargetABI()) {}
+
+    virtual const char *getPassName() const {
+      return "Mips Constant Islands";
+    }
+
+    bool runOnMachineFunction(MachineFunction &F);
+
+  private:
+
+
+    const TargetMachine &TM;
+    const MipsInstrInfo *TII;
+    bool IsPIC;
+    unsigned ABI;
+
+  };
+
+  char MipsConstantIslands::ID = 0;
+} // end of anonymous namespace
+
+/// createMipsLongBranchPass - Returns a pass that converts branches to long
+/// branches.
+FunctionPass *llvm::createMipsConstantIslandPass(MipsTargetMachine &tm) {
+  return new MipsConstantIslands(tm);
+}
+
+bool MipsConstantIslands::runOnMachineFunction(MachineFunction &F) {
+  return true;
+}
+
diff --git a/lib/Target/Mips/MipsDSPInstrFormats.td b/lib/Target/Mips/MipsDSPInstrFormats.td
index 8e01d06596a1..a72a763fde06 100644
--- a/lib/Target/Mips/MipsDSPInstrFormats.td
+++ b/lib/Target/Mips/MipsDSPInstrFormats.td
@@ -24,8 +24,9 @@ class DSPInst : MipsInst<(outs), (ins), "", [], NoItinerary, FrmOther> {
   let Predicates = [HasDSP];
 }
 
-class PseudoDSP<dag outs, dag ins, list<dag> pattern>:
-  MipsPseudo<outs, ins, "", pattern> {
+class PseudoDSP<dag outs, dag ins, list<dag> pattern,
+                InstrItinClass itin = IIPseudo>:
+  MipsPseudo<outs, ins, pattern, itin> {
   let Predicates = [HasDSP];
 }
 
diff --git a/lib/Target/Mips/MipsDSPInstrInfo.td b/lib/Target/Mips/MipsDSPInstrInfo.td
index ef9402865b0d..3c116e1264b3 100644
--- a/lib/Target/Mips/MipsDSPInstrInfo.td
+++ b/lib/Target/Mips/MipsDSPInstrInfo.td
@@ -20,17 +20,18 @@ def immZExt10 : ImmLeaf<i32, [{return isUInt<10>(Imm);}]>;
 def immSExt6 : ImmLeaf<i32, [{return isInt<6>(Imm);}]>;
 
 // Mips-specific dsp nodes
-def SDT_MipsExtr : SDTypeProfile<1, 1, [SDTCisVT<0, i32>, SDTCisSameAs<0, 1>]>;
-def SDT_MipsShilo : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
-def SDT_MipsDPA : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisSameAs<0, 1>]>;
+def SDT_MipsExtr : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisSameAs<0, 1>,
+                                        SDTCisVT<2, untyped>]>;
+def SDT_MipsShilo : SDTypeProfile<1, 2, [SDTCisVT<0, untyped>,
+                                         SDTCisSameAs<0, 2>, SDTCisVT<1, i32>]>;
+def SDT_MipsDPA : SDTypeProfile<1, 3, [SDTCisVT<0, untyped>, SDTCisSameAs<0, 3>,
+                                       SDTCisVT<1, i32>, SDTCisSameAs<1, 2>]>;
 
 class MipsDSPBase<string Opc, SDTypeProfile Prof> :
-  SDNode<!strconcat("MipsISD::", Opc), Prof,
-         [SDNPHasChain, SDNPInGlue, SDNPOutGlue]>;
+  SDNode<!strconcat("MipsISD::", Opc), Prof>;
 
 class MipsDSPSideEffectBase<string Opc, SDTypeProfile Prof> :
-  SDNode<!strconcat("MipsISD::", Opc), Prof,
-         [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPSideEffect]>;
+  SDNode<!strconcat("MipsISD::", Opc), Prof, [SDNPHasChain, SDNPSideEffect]>;
 
 def MipsEXTP : MipsDSPSideEffectBase<"EXTP", SDT_MipsExtr>;
 def MipsEXTPDP : MipsDSPSideEffectBase<"EXTPDP", SDT_MipsExtr>;
@@ -40,7 +41,7 @@ def MipsEXTR_R_W : MipsDSPSideEffectBase<"EXTR_R_W", SDT_MipsExtr>;
 def MipsEXTR_RS_W : MipsDSPSideEffectBase<"EXTR_RS_W", SDT_MipsExtr>;
 
 def MipsSHILO : MipsDSPBase<"SHILO", SDT_MipsShilo>;
-def MipsMTHLIP : MipsDSPBase<"MTHLIP", SDT_MipsShilo>;
+def MipsMTHLIP : MipsDSPSideEffectBase<"MTHLIP", SDT_MipsShilo>;
 
 def MipsMULSAQ_S_W_PH : MipsDSPSideEffectBase<"MULSAQ_S_W_PH", SDT_MipsDPA>;
 def MipsMAQ_S_W_PHL : MipsDSPSideEffectBase<"MAQ_S_W_PHL", SDT_MipsDPA>;
@@ -75,10 +76,6 @@ def MipsMSUB_DSP : MipsDSPBase<"MSUB_DSP", SDT_MipsDPA>;
 def MipsMSUBU_DSP : MipsDSPBase<"MSUBU_DSP", SDT_MipsDPA>;
 
 // Flags.
-class IsCommutable {
-  bit isCommutable = 1;
-}
-
 class UseAC {
   list<Register> Uses = [AC0];
 }
@@ -387,7 +384,7 @@ class APPEND_DESC_BASE<string instr_asm, SDPatternOperator OpNode,
 class EXTR_W_TY1_R2_DESC_BASE<string instr_asm, SDPatternOperator OpNode,
                               InstrItinClass itin> {
   dag OutOperandList = (outs CPURegs:$rt);
-  dag InOperandList = (ins ACRegs:$ac, CPURegs:$shift_rs);
+  dag InOperandList = (ins ACRegsDSP:$ac, CPURegs:$shift_rs);
   string AsmString = !strconcat(instr_asm, "\t$rt, $ac, $shift_rs");
   InstrItinClass Itinerary = itin;
   list<Register> Defs = [DSPCtrl];
@@ -396,46 +393,40 @@ class EXTR_W_TY1_R2_DESC_BASE<string instr_asm, SDPatternOperator OpNode,
 class EXTR_W_TY1_R1_DESC_BASE<string instr_asm, SDPatternOperator OpNode,
                               InstrItinClass itin> {
   dag OutOperandList = (outs CPURegs:$rt);
-  dag InOperandList = (ins ACRegs:$ac, uimm16:$shift_rs);
+  dag InOperandList = (ins ACRegsDSP:$ac, uimm16:$shift_rs);
   string AsmString = !strconcat(instr_asm, "\t$rt, $ac, $shift_rs");
   InstrItinClass Itinerary = itin;
   list<Register> Defs = [DSPCtrl];
 }
 
-class SHILO_R1_PSEUDO_BASE<SDPatternOperator OpNode, InstrItinClass itin,
-                           Instruction realinst> :
-  PseudoDSP<(outs), (ins simm16:$shift), [(OpNode immSExt6:$shift)]>,
-  PseudoInstExpansion<(realinst AC0, simm16:$shift)> {
-  list<Register> Defs = [DSPCtrl, AC0];
-  list<Register> Uses = [AC0];
-  InstrItinClass Itinerary = itin;
-}
-
-class SHILO_R1_DESC_BASE<string instr_asm> {
-  dag OutOperandList = (outs ACRegs:$ac);
-  dag InOperandList = (ins simm16:$shift);
+class SHILO_R1_DESC_BASE<string instr_asm, SDPatternOperator OpNode> {
+  dag OutOperandList = (outs ACRegsDSP:$ac);
+  dag InOperandList = (ins simm16:$shift, ACRegsDSP:$acin);
   string AsmString = !strconcat(instr_asm, "\t$ac, $shift");
+  list<dag> Pattern = [(set ACRegsDSP:$ac,
+                        (OpNode immSExt6:$shift, ACRegsDSP:$acin))];
+  list<Register> Defs = [DSPCtrl];
+  string Constraints = "$acin = $ac";
 }
 
-class SHILO_R2_PSEUDO_BASE<SDPatternOperator OpNode, InstrItinClass itin,
-                           Instruction realinst> :
-  PseudoDSP<(outs), (ins CPURegs:$rs), [(OpNode CPURegs:$rs)]>,
-  PseudoInstExpansion<(realinst AC0, CPURegs:$rs)> {
-  list<Register> Defs = [DSPCtrl, AC0];
-  list<Register> Uses = [AC0];
-  InstrItinClass Itinerary = itin;
-}
-
-class SHILO_R2_DESC_BASE<string instr_asm> {
-  dag OutOperandList = (outs ACRegs:$ac);
-  dag InOperandList = (ins CPURegs:$rs);
+class SHILO_R2_DESC_BASE<string instr_asm, SDPatternOperator OpNode> {
+  dag OutOperandList = (outs ACRegsDSP:$ac);
+  dag InOperandList = (ins CPURegs:$rs, ACRegsDSP:$acin);
   string AsmString = !strconcat(instr_asm, "\t$ac, $rs");
+  list<dag> Pattern = [(set ACRegsDSP:$ac,
+                        (OpNode CPURegs:$rs, ACRegsDSP:$acin))];
+  list<Register> Defs = [DSPCtrl];
+  string Constraints = "$acin = $ac";
 }
 
-class MTHLIP_DESC_BASE<string instr_asm> {
-  dag OutOperandList = (outs ACRegs:$ac);
-  dag InOperandList = (ins CPURegs:$rs);
+class MTHLIP_DESC_BASE<string instr_asm, SDPatternOperator OpNode> {
+  dag OutOperandList = (outs ACRegsDSP:$ac);
+  dag InOperandList = (ins CPURegs:$rs, ACRegsDSP:$acin);
   string AsmString = !strconcat(instr_asm, "\t$rs, $ac");
+  list<dag> Pattern = [(set ACRegsDSP:$ac,
+                        (OpNode CPURegs:$rs, ACRegsDSP:$acin))];
+  list<Register> Uses = [DSPCtrl];
+  string Constraints = "$acin = $ac";
 }
 
 class RDDSP_DESC_BASE<string instr_asm, SDPatternOperator OpNode,
@@ -458,39 +449,41 @@ class WRDSP_DESC_BASE<string instr_asm, SDPatternOperator OpNode,
   list<Register> Defs = [DSPCtrl];
 }
 
-class DPA_W_PH_PSEUDO_BASE<SDPatternOperator OpNode, InstrItinClass itin,
-                           Instruction realinst> :
-  PseudoDSP<(outs), (ins CPURegs:$rs, CPURegs:$rt),
-            [(OpNode CPURegs:$rs, CPURegs:$rt)]>,
-  PseudoInstExpansion<(realinst AC0, CPURegs:$rs, CPURegs:$rt)> {
-  list<Register> Defs = [DSPCtrl, AC0];
-  list<Register> Uses = [AC0];
-  InstrItinClass Itinerary = itin;
+class DPA_W_PH_DESC_BASE<string instr_asm, SDPatternOperator OpNode> {
+  dag OutOperandList = (outs ACRegsDSP:$ac);
+  dag InOperandList = (ins CPURegs:$rs, CPURegs:$rt, ACRegsDSP:$acin);
+  string AsmString = !strconcat(instr_asm, "\t$ac, $rs, $rt");
+  list<dag> Pattern = [(set ACRegsDSP:$ac,
+                        (OpNode CPURegs:$rs, CPURegs:$rt, ACRegsDSP:$acin))];
+  list<Register> Defs = [DSPCtrl];
+  string Constraints = "$acin = $ac";
 }
 
-class DPA_W_PH_DESC_BASE<string instr_asm> {
-  dag OutOperandList = (outs ACRegs:$ac);
+class MULT_DESC_BASE<string instr_asm, SDPatternOperator OpNode,
+                     InstrItinClass itin> {
+  dag OutOperandList = (outs ACRegsDSP:$ac);
   dag InOperandList = (ins CPURegs:$rs, CPURegs:$rt);
   string AsmString = !strconcat(instr_asm, "\t$ac, $rs, $rt");
-}
-
-class MULT_PSEUDO_BASE<SDPatternOperator OpNode, InstrItinClass itin,
-                       Instruction realinst> :
-  PseudoDSP<(outs), (ins CPURegs:$rs, CPURegs:$rt),
-            [(OpNode CPURegs:$rs, CPURegs:$rt)]>,
-  PseudoInstExpansion<(realinst AC0, CPURegs:$rs, CPURegs:$rt)> {
-  list<Register> Defs = [DSPCtrl, AC0];
+  list<dag> Pattern = [(set ACRegsDSP:$ac, (OpNode CPURegs:$rs, CPURegs:$rt))];
   InstrItinClass Itinerary = itin;
+  int AddedComplexity = 20;
+  bit isCommutable = 1;
 }
 
-class MULT_DESC_BASE<string instr_asm> {
-  dag OutOperandList = (outs ACRegs:$ac);
-  dag InOperandList = (ins CPURegs:$rs, CPURegs:$rt);
+class MADD_DESC_BASE<string instr_asm, SDPatternOperator OpNode,
+                     InstrItinClass itin> {
+  dag OutOperandList = (outs ACRegsDSP:$ac);
+  dag InOperandList = (ins CPURegs:$rs, CPURegs:$rt, ACRegsDSP:$acin);
   string AsmString = !strconcat(instr_asm, "\t$ac, $rs, $rt");
+  list<dag> Pattern = [(set ACRegsDSP:$ac,
+                        (OpNode CPURegs:$rs, CPURegs:$rt, ACRegsDSP:$acin))];
+  InstrItinClass Itinerary = itin;
+  int AddedComplexity = 20;
+  string Constraints = "$acin = $ac";
 }
 
 class BPOSGE32_PSEUDO_DESC_BASE<SDPatternOperator OpNode, InstrItinClass itin> :
-  MipsPseudo<(outs CPURegs:$dst), (ins), "", [(set CPURegs:$dst, (OpNode))]> {
+  MipsPseudo<(outs CPURegs:$dst), (ins), [(set CPURegs:$dst, (OpNode))]> {
   list<Register> Uses = [DSPCtrl];
   bit usesCustomInserter = 1;
 }
@@ -721,44 +714,40 @@ class MULQ_RS_PH_DESC : ADDU_QB_DESC_BASE<"mulq_rs.ph", int_mips_mulq_rs_ph,
                                           NoItinerary, DSPRegs, DSPRegs>,
                         IsCommutable;
 
-class MULSAQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"mulsaq_s.w.ph">;
+class MULSAQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"mulsaq_s.w.ph",
+                                              MipsMULSAQ_S_W_PH>;
 
-class MAQ_S_W_PHL_DESC : DPA_W_PH_DESC_BASE<"maq_s.w.phl">;
+class MAQ_S_W_PHL_DESC : DPA_W_PH_DESC_BASE<"maq_s.w.phl", MipsMAQ_S_W_PHL>;
 
-class MAQ_S_W_PHR_DESC : DPA_W_PH_DESC_BASE<"maq_s.w.phr">;
+class MAQ_S_W_PHR_DESC : DPA_W_PH_DESC_BASE<"maq_s.w.phr", MipsMAQ_S_W_PHR>;
 
-class MAQ_SA_W_PHL_DESC : DPA_W_PH_DESC_BASE<"maq_sa.w.phl">;
+class MAQ_SA_W_PHL_DESC : DPA_W_PH_DESC_BASE<"maq_sa.w.phl", MipsMAQ_SA_W_PHL>;
 
-class MAQ_SA_W_PHR_DESC : DPA_W_PH_DESC_BASE<"maq_sa.w.phr">;
+class MAQ_SA_W_PHR_DESC : DPA_W_PH_DESC_BASE<"maq_sa.w.phr", MipsMAQ_SA_W_PHR>;
 
 // Dot product with accumulate/subtract
-class DPAU_H_QBL_DESC : DPA_W_PH_DESC_BASE<"dpau.h.qbl">;
-
-class DPAU_H_QBR_DESC : DPA_W_PH_DESC_BASE<"dpau.h.qbr">;
+class DPAU_H_QBL_DESC : DPA_W_PH_DESC_BASE<"dpau.h.qbl", MipsDPAU_H_QBL>;
 
-class DPSU_H_QBL_DESC : DPA_W_PH_DESC_BASE<"dpsu.h.qbl">;
+class DPAU_H_QBR_DESC : DPA_W_PH_DESC_BASE<"dpau.h.qbr", MipsDPAU_H_QBR>;
 
-class DPSU_H_QBR_DESC : DPA_W_PH_DESC_BASE<"dpsu.h.qbr">;
+class DPSU_H_QBL_DESC : DPA_W_PH_DESC_BASE<"dpsu.h.qbl", MipsDPSU_H_QBL>;
 
-class DPAQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaq_s.w.ph">;
+class DPSU_H_QBR_DESC : DPA_W_PH_DESC_BASE<"dpsu.h.qbr", MipsDPSU_H_QBR>;
 
-class DPSQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsq_s.w.ph">;
+class DPAQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaq_s.w.ph", MipsDPAQ_S_W_PH>;
 
-class DPAQ_SA_L_W_DESC : DPA_W_PH_DESC_BASE<"dpaq_sa.l.w">;
+class DPSQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsq_s.w.ph", MipsDPSQ_S_W_PH>;
 
-class DPSQ_SA_L_W_DESC : DPA_W_PH_DESC_BASE<"dpsq_sa.l.w">;
+class DPAQ_SA_L_W_DESC : DPA_W_PH_DESC_BASE<"dpaq_sa.l.w", MipsDPAQ_SA_L_W>;
 
-class MULT_DSP_DESC : MULT_DESC_BASE<"mult">;
+class DPSQ_SA_L_W_DESC : DPA_W_PH_DESC_BASE<"dpsq_sa.l.w", MipsDPSQ_SA_L_W>;
 
-class MULTU_DSP_DESC : MULT_DESC_BASE<"multu">;
-
-class MADD_DSP_DESC : MULT_DESC_BASE<"madd">;
-
-class MADDU_DSP_DESC : MULT_DESC_BASE<"maddu">;
-
-class MSUB_DSP_DESC : MULT_DESC_BASE<"msub">;
-
-class MSUBU_DSP_DESC : MULT_DESC_BASE<"msubu">;
+class MULT_DSP_DESC  : MULT_DESC_BASE<"mult", MipsMult, NoItinerary>;
+class MULTU_DSP_DESC : MULT_DESC_BASE<"multu", MipsMultu, NoItinerary>;
+class MADD_DSP_DESC  : MADD_DESC_BASE<"madd", MipsMAdd, NoItinerary>;
+class MADDU_DSP_DESC : MADD_DESC_BASE<"maddu", MipsMAddu, NoItinerary>;
+class MSUB_DSP_DESC  : MADD_DESC_BASE<"msub", MipsMSub, NoItinerary>;
+class MSUBU_DSP_DESC : MADD_DESC_BASE<"msubu", MipsMSubu, NoItinerary>;
 
 // Comparison
 class CMPU_EQ_QB_DESC : CMP_EQ_QB_R2_DESC_BASE<"cmpu.eq.qb",
@@ -871,11 +860,11 @@ class EXTR_S_H_DESC : EXTR_W_TY1_R1_DESC_BASE<"extr_s.h", MipsEXTR_S_H,
 class EXTRV_S_H_DESC : EXTR_W_TY1_R2_DESC_BASE<"extrv_s.h", MipsEXTR_S_H,
                                                NoItinerary>;
 
-class SHILO_DESC : SHILO_R1_DESC_BASE<"shilo">;
+class SHILO_DESC : SHILO_R1_DESC_BASE<"shilo", MipsSHILO>;
 
-class SHILOV_DESC : SHILO_R2_DESC_BASE<"shilov">;
+class SHILOV_DESC : SHILO_R2_DESC_BASE<"shilov", MipsSHILO>;
 
-class MTHLIP_DESC : MTHLIP_DESC_BASE<"mthlip">;
+class MTHLIP_DESC : MTHLIP_DESC_BASE<"mthlip", MipsMTHLIP>;
 
 class RDDSP_DESC : RDDSP_DESC_BASE<"rddsp", int_mips_rddsp, NoItinerary>;
 
@@ -979,23 +968,25 @@ class MULQ_S_PH_DESC : ADDU_QB_DESC_BASE<"mulq_s.ph", int_mips_mulq_s_ph,
                        IsCommutable;
 
 // Dot product with accumulate/subtract
-class DPA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpa.w.ph">;
+class DPA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpa.w.ph", MipsDPA_W_PH>;
 
-class DPS_W_PH_DESC : DPA_W_PH_DESC_BASE<"dps.w.ph">;
+class DPS_W_PH_DESC : DPA_W_PH_DESC_BASE<"dps.w.ph", MipsDPS_W_PH>;
 
-class DPAQX_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaqx_s.w.ph">;
+class DPAQX_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaqx_s.w.ph", MipsDPAQX_S_W_PH>;
 
-class DPAQX_SA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaqx_sa.w.ph">;
+class DPAQX_SA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaqx_sa.w.ph",
+                                              MipsDPAQX_SA_W_PH>;
 
-class DPAX_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpax.w.ph">;
+class DPAX_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpax.w.ph", MipsDPAX_W_PH>;
 
-class DPSX_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsx.w.ph">;
+class DPSX_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsx.w.ph", MipsDPSX_W_PH>;
 
-class DPSQX_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsqx_s.w.ph">;
+class DPSQX_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsqx_s.w.ph", MipsDPSQX_S_W_PH>;
 
-class DPSQX_SA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsqx_sa.w.ph">;
+class DPSQX_SA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsqx_sa.w.ph",
+                                              MipsDPSQX_SA_W_PH>;
 
-class MULSA_W_PH_DESC : DPA_W_PH_DESC_BASE<"mulsa.w.ph">;
+class MULSA_W_PH_DESC : DPA_W_PH_DESC_BASE<"mulsa.w.ph", MipsMULSA_W_PH>;
 
 // Precision reduce/expand
 class PRECR_QB_PH_DESC : CMP_EQ_QB_R3_DESC_BASE<"precr.qb.ph",
@@ -1210,71 +1201,14 @@ def PREPEND : PREPEND_ENC, PREPEND_DESC;
 }
 
 // Pseudos.
-def MULSAQ_S_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMULSAQ_S_W_PH, NoItinerary,
-                                                MULSAQ_S_W_PH>;
-def MAQ_S_W_PHL_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMAQ_S_W_PHL, NoItinerary,
-                                              MAQ_S_W_PHL>;
-def MAQ_S_W_PHR_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMAQ_S_W_PHR, NoItinerary,
-                                              MAQ_S_W_PHR>;
-def MAQ_SA_W_PHL_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMAQ_SA_W_PHL, NoItinerary,
-                                               MAQ_SA_W_PHL>;
-def MAQ_SA_W_PHR_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMAQ_SA_W_PHR, NoItinerary,
-                                               MAQ_SA_W_PHR>;
-def DPAU_H_QBL_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAU_H_QBL, NoItinerary,
-                                             DPAU_H_QBL>;
-def DPAU_H_QBR_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAU_H_QBR, NoItinerary,
-                                             DPAU_H_QBR>;
-def DPSU_H_QBL_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSU_H_QBL, NoItinerary,
-                                             DPSU_H_QBL>;
-def DPSU_H_QBR_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSU_H_QBR, NoItinerary,
-                                             DPSU_H_QBR>;
-def DPAQ_S_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAQ_S_W_PH, NoItinerary,
-                                              DPAQ_S_W_PH>;
-def DPSQ_S_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSQ_S_W_PH, NoItinerary,
-                                              DPSQ_S_W_PH>;
-def DPAQ_SA_L_W_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAQ_SA_L_W, NoItinerary,
-                                              DPAQ_SA_L_W>;
-def DPSQ_SA_L_W_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSQ_SA_L_W, NoItinerary,
-                                              DPSQ_SA_L_W>;
-
-def MULT_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMULT, NoItinerary, MULT_DSP>,
-                      IsCommutable;
-def MULTU_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMULTU, NoItinerary, MULTU_DSP>,
-                       IsCommutable;
-def MADD_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMADD_DSP, NoItinerary, MADD_DSP>,
-                      IsCommutable, UseAC;
-def MADDU_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMADDU_DSP, NoItinerary, MADDU_DSP>,
-                       IsCommutable, UseAC;
-def MSUB_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMSUB_DSP, NoItinerary, MSUB_DSP>,
-                      UseAC;
-def MSUBU_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMSUBU_DSP, NoItinerary, MSUBU_DSP>,
-                       UseAC;
-
-def SHILO_PSEUDO : SHILO_R1_PSEUDO_BASE<MipsSHILO, NoItinerary, SHILO>;
-def SHILOV_PSEUDO : SHILO_R2_PSEUDO_BASE<MipsSHILO, NoItinerary, SHILOV>;
-def MTHLIP_PSEUDO : SHILO_R2_PSEUDO_BASE<MipsMTHLIP, NoItinerary, MTHLIP>;
-
-let Predicates = [HasDSPR2] in {
-
-def DPA_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPA_W_PH, NoItinerary, DPA_W_PH>;
-def DPS_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPS_W_PH, NoItinerary, DPS_W_PH>;
-def DPAQX_S_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAQX_S_W_PH, NoItinerary,
-                                               DPAQX_S_W_PH>;
-def DPAQX_SA_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAQX_SA_W_PH, NoItinerary,
-                                                DPAQX_SA_W_PH>;
-def DPAX_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAX_W_PH, NoItinerary,
-                                            DPAX_W_PH>;
-def DPSX_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSX_W_PH, NoItinerary,
-                                            DPSX_W_PH>;
-def DPSQX_S_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSQX_S_W_PH, NoItinerary,
-                                               DPSQX_S_W_PH>;
-def DPSQX_SA_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSQX_SA_W_PH, NoItinerary,
-                                                DPSQX_SA_W_PH>;
-def MULSA_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMULSA_W_PH, NoItinerary,
-                                             MULSA_W_PH>;
-
+/// Pseudo instructions for loading, storing and copying accumulator registers.
+let isPseudo = 1 in {
+  defm LOAD_AC_DSP  : LoadM<"load_ac_dsp", ACRegsDSP>;
+  defm STORE_AC_DSP : StoreM<"store_ac_dsp", ACRegsDSP>;
 }
 
+def COPY_AC_DSP : PseudoSE<(outs ACRegsDSP:$dst), (ins ACRegsDSP:$src), []>;
+
 // Patterns.
 class DSPPat<dag pattern, dag result, Predicate pred = HasDSP> :
   Pat<pattern, result>, Requires<[pred]>;
@@ -1300,10 +1234,12 @@ def : DSPPat<(store (v4i8 DSPRegs:$val), addr:$a),
 
 // Extr patterns.
 class EXTR_W_TY1_R2_Pat<SDPatternOperator OpNode, Instruction Instr> :
-  DSPPat<(i32 (OpNode CPURegs:$rs)), (Instr AC0, CPURegs:$rs)>;
+  DSPPat<(i32 (OpNode CPURegs:$rs, ACRegsDSP:$ac)),
+         (Instr ACRegsDSP:$ac, CPURegs:$rs)>;
 
 class EXTR_W_TY1_R1_Pat<SDPatternOperator OpNode, Instruction Instr> :
-  DSPPat<(i32 (OpNode immZExt5:$shift)), (Instr AC0, immZExt5:$shift)>;
+  DSPPat<(i32 (OpNode immZExt5:$shift, ACRegsDSP:$ac)),
+         (Instr ACRegsDSP:$ac, immZExt5:$shift)>;
 
 def : EXTR_W_TY1_R1_Pat<MipsEXTP, EXTP>;
 def : EXTR_W_TY1_R2_Pat<MipsEXTP, EXTPV>;
@@ -1317,3 +1253,19 @@ def : EXTR_W_TY1_R1_Pat<MipsEXTR_RS_W, EXTR_RS_W>;
 def : EXTR_W_TY1_R2_Pat<MipsEXTR_RS_W, EXTRV_RS_W>;
 def : EXTR_W_TY1_R1_Pat<MipsEXTR_S_H, EXTR_S_H>;
 def : EXTR_W_TY1_R2_Pat<MipsEXTR_S_H, EXTRV_S_H>;
+
+// mflo/hi patterns.
+let AddedComplexity = 20 in
+def : DSPPat<(i32 (ExtractLOHI ACRegsDSP:$ac, imm:$lohi_idx)),
+             (EXTRACT_SUBREG ACRegsDSP:$ac, imm:$lohi_idx)>;
+
+// Indexed load patterns.
+class IndexedLoadPat<SDPatternOperator LoadNode, Instruction Instr> :
+  DSPPat<(i32 (LoadNode (add i32:$base, i32:$index))),
+         (Instr i32:$base, i32:$index)>;
+
+let AddedComplexity = 20 in {
+  def : IndexedLoadPat<zextloadi8, LBUX>;
+  def : IndexedLoadPat<sextloadi16, LHX>;
+  def : IndexedLoadPat<load, LWX>;
+}
diff --git a/lib/Target/Mips/MipsDelaySlotFiller.cpp b/lib/Target/Mips/MipsDelaySlotFiller.cpp
index e3c8ed75cf91..d07a595af38a 100644
--- a/lib/Target/Mips/MipsDelaySlotFiller.cpp
+++ b/lib/Target/Mips/MipsDelaySlotFiller.cpp
@@ -1,4 +1,4 @@
-//===-- DelaySlotFiller.cpp - Mips Delay Slot Filler ----------------------===//
+//===-- MipsDelaySlotFiller.cpp - Mips Delay Slot Filler ------------------===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,22 +7,28 @@
 //
 //===----------------------------------------------------------------------===//
 //
-// Simple pass to fills delay slots with useful instructions.
+// Simple pass to fill delay slots with useful instructions.
 //
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "delay-slot-filler"
 
 #include "Mips.h"
+#include "MipsInstrInfo.h"
 #include "MipsTargetMachine.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/CodeGen/MachineBranchProbabilityInfo.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/PseudoSourceValue.h"
 #include "llvm/Support/CommandLine.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/ADT/SmallSet.h"
-#include "llvm/ADT/Statistic.h"
 
 using namespace llvm;
 
@@ -33,27 +39,143 @@ STATISTIC(UsefulSlots, "Number of delay slots filled with instructions that"
 static cl::opt<bool> DisableDelaySlotFiller(
   "disable-mips-delay-filler",
   cl::init(false),
-  cl::desc("Disable the delay slot filler, which attempts to fill the Mips"
-           "delay slots with useful instructions."),
+  cl::desc("Fill all delay slots with NOPs."),
   cl::Hidden);
 
-// This option can be used to silence complaints by machine verifier passes.
-static cl::opt<bool> SkipDelaySlotFiller(
-  "skip-mips-delay-filler",
+static cl::opt<bool> DisableForwardSearch(
+  "disable-mips-df-forward-search",
+  cl::init(true),
+  cl::desc("Disallow MIPS delay filler to search forward."),
+  cl::Hidden);
+
+static cl::opt<bool> DisableSuccBBSearch(
+  "disable-mips-df-succbb-search",
+  cl::init(true),
+  cl::desc("Disallow MIPS delay filler to search successor basic blocks."),
+  cl::Hidden);
+
+static cl::opt<bool> DisableBackwardSearch(
+  "disable-mips-df-backward-search",
   cl::init(false),
-  cl::desc("Skip MIPS' delay slot filling pass."),
+  cl::desc("Disallow MIPS delay filler to search backward."),
   cl::Hidden);
 
 namespace {
-  struct Filler : public MachineFunctionPass {
-    typedef MachineBasicBlock::instr_iterator InstrIter;
-    typedef MachineBasicBlock::reverse_instr_iterator ReverseInstrIter;
+  typedef MachineBasicBlock::iterator Iter;
+  typedef MachineBasicBlock::reverse_iterator ReverseIter;
+  typedef SmallDenseMap<MachineBasicBlock*, MachineInstr*, 2> BB2BrMap;
+
+  /// \brief A functor comparing edge weight of two blocks.
+  struct CmpWeight {
+    CmpWeight(const MachineBasicBlock &S,
+              const MachineBranchProbabilityInfo &P) : Src(S), Prob(P) {}
+
+    bool operator()(const MachineBasicBlock *Dst0,
+                    const MachineBasicBlock *Dst1) const {
+      return Prob.getEdgeWeight(&Src, Dst0) < Prob.getEdgeWeight(&Src, Dst1);
+    }
 
-    TargetMachine &TM;
-    const TargetInstrInfo *TII;
-    InstrIter LastFiller;
+    const MachineBasicBlock &Src;
+    const MachineBranchProbabilityInfo &Prob;
+  };
 
-    static char ID;
+  class RegDefsUses {
+  public:
+    RegDefsUses(TargetMachine &TM);
+    void init(const MachineInstr &MI);
+
+    /// This function sets all caller-saved registers in Defs.
+    void setCallerSaved(const MachineInstr &MI);
+
+    /// This function sets all unallocatable registers in Defs.
+    void setUnallocatableRegs(const MachineFunction &MF);
+
+    /// Set bits in Uses corresponding to MBB's live-out registers except for
+    /// the registers that are live-in to SuccBB.
+    void addLiveOut(const MachineBasicBlock &MBB,
+                    const MachineBasicBlock &SuccBB);
+
+    bool update(const MachineInstr &MI, unsigned Begin, unsigned End);
+
+  private:
+    bool checkRegDefsUses(BitVector &NewDefs, BitVector &NewUses, unsigned Reg,
+                          bool IsDef) const;
+
+    /// Returns true if Reg or its alias is in RegSet.
+    bool isRegInSet(const BitVector &RegSet, unsigned Reg) const;
+
+    const TargetRegisterInfo &TRI;
+    BitVector Defs, Uses;
+  };
+
+  /// Base class for inspecting loads and stores.
+  class InspectMemInstr {
+  public:
+    InspectMemInstr(bool ForbidMemInstr_)
+      : OrigSeenLoad(false), OrigSeenStore(false), SeenLoad(false),
+        SeenStore(false), ForbidMemInstr(ForbidMemInstr_) {}
+
+    /// Return true if MI cannot be moved to delay slot.
+    bool hasHazard(const MachineInstr &MI);
+
+    virtual ~InspectMemInstr() {}
+
+  protected:
+    /// Flags indicating whether loads or stores have been seen.
+    bool OrigSeenLoad, OrigSeenStore, SeenLoad, SeenStore;
+
+    /// Memory instructions are not allowed to move to delay slot if this flag
+    /// is true.
+    bool ForbidMemInstr;
+
+  private:
+    virtual bool hasHazard_(const MachineInstr &MI) = 0;
+  };
+
+  /// This subclass rejects any memory instructions.
+  class NoMemInstr : public InspectMemInstr {
+  public:
+    NoMemInstr() : InspectMemInstr(true) {}
+  private:
+    virtual bool hasHazard_(const MachineInstr &MI) { return true; }
+  };
+
+  /// This subclass accepts loads from stacks and constant loads.
+  class LoadFromStackOrConst : public InspectMemInstr {
+  public:
+    LoadFromStackOrConst() : InspectMemInstr(false) {}
+  private:
+    virtual bool hasHazard_(const MachineInstr &MI);
+  };
+
+  /// This subclass uses memory dependence information to determine whether a
+  /// memory instruction can be moved to a delay slot.
+  class MemDefsUses : public InspectMemInstr {
+  public:
+    MemDefsUses(const MachineFrameInfo *MFI);
+
+  private:
+    virtual bool hasHazard_(const MachineInstr &MI);
+
+    /// Update Defs and Uses. Return true if there exist dependences that
+    /// disqualify the delay slot candidate between V and values in Uses and
+    /// Defs.
+    bool updateDefsUses(const Value *V, bool MayStore);
+
+    /// Get the list of underlying objects of MI's memory operand.
+    bool getUnderlyingObjects(const MachineInstr &MI,
+                              SmallVectorImpl<const Value *> &Objects) const;
+
+    const MachineFrameInfo *MFI;
+    SmallPtrSet<const Value*, 4> Uses, Defs;
+
+    /// Flags indicating whether loads or stores with no underlying objects have
+    /// been seen.
+    bool SeenNoObjLoad, SeenNoObjStore;
+  };
+
+  class Filler : public MachineFunctionPass {
+  public:
     Filler(TargetMachine &tm)
       : MachineFunctionPass(ID), TM(tm), TII(tm.getInstrInfo()) { }
 
@@ -61,11 +183,7 @@ namespace {
       return "Mips Delay Slot Filler";
     }
 
-    bool runOnMachineBasicBlock(MachineBasicBlock &MBB);
     bool runOnMachineFunction(MachineFunction &F) {
-      if (SkipDelaySlotFiller)
-        return false;
-
       bool Changed = false;
       for (MachineFunction::iterator FI = F.begin(), FE = F.end();
            FI != FE; ++FI)
@@ -73,66 +191,334 @@ namespace {
       return Changed;
     }
 
-    bool isDelayFiller(MachineBasicBlock &MBB,
-                       InstrIter candidate);
-
-    void insertCallUses(InstrIter MI,
-                        SmallSet<unsigned, 32> &RegDefs,
-                        SmallSet<unsigned, 32> &RegUses);
-
-    void insertDefsUses(InstrIter MI,
-                        SmallSet<unsigned, 32> &RegDefs,
-                        SmallSet<unsigned, 32> &RegUses);
-
-    bool IsRegInSet(SmallSet<unsigned, 32> &RegSet,
-                    unsigned Reg);
+    void getAnalysisUsage(AnalysisUsage &AU) const {
+      AU.addRequired<MachineBranchProbabilityInfo>();
+      MachineFunctionPass::getAnalysisUsage(AU);
+    }
 
-    bool delayHasHazard(InstrIter candidate,
-                        bool &sawLoad, bool &sawStore,
-                        SmallSet<unsigned, 32> &RegDefs,
-                        SmallSet<unsigned, 32> &RegUses);
+  private:
+    bool runOnMachineBasicBlock(MachineBasicBlock &MBB);
 
-    bool
-    findDelayInstr(MachineBasicBlock &MBB, InstrIter slot,
-                   InstrIter &Filler);
+    /// This function checks if it is valid to move Candidate to the delay slot
+    /// and returns true if it isn't. It also updates memory and register
+    /// dependence information.
+    bool delayHasHazard(const MachineInstr &Candidate, RegDefsUses &RegDU,
+                        InspectMemInstr &IM) const;
+
+    /// This function searches range [Begin, End) for an instruction that can be
+    /// moved to the delay slot. Returns true on success.
+    template<typename IterTy>
+    bool searchRange(MachineBasicBlock &MBB, IterTy Begin, IterTy End,
+                     RegDefsUses &RegDU, InspectMemInstr &IM,
+                     IterTy &Filler) const;
+
+    /// This function searches in the backward direction for an instruction that
+    /// can be moved to the delay slot. Returns true on success.
+    bool searchBackward(MachineBasicBlock &MBB, Iter Slot) const;
+
+    /// This function searches MBB in the forward direction for an instruction
+    /// that can be moved to the delay slot. Returns true on success.
+    bool searchForward(MachineBasicBlock &MBB, Iter Slot) const;
+
+    /// This function searches one of MBB's successor blocks for an instruction
+    /// that can be moved to the delay slot and inserts clones of the
+    /// instruction into the successor's predecessor blocks.
+    bool searchSuccBBs(MachineBasicBlock &MBB, Iter Slot) const;
+
+    /// Pick a successor block of MBB. Return NULL if MBB doesn't have a
+    /// successor block that is not a landing pad.
+    MachineBasicBlock *selectSuccBB(MachineBasicBlock &B) const;
+
+    /// This function analyzes MBB and returns an instruction with an unoccupied
+    /// slot that branches to Dst.
+    std::pair<MipsInstrInfo::BranchType, MachineInstr *>
+    getBranch(MachineBasicBlock &MBB, const MachineBasicBlock &Dst) const;
+
+    /// Examine Pred and see if it is possible to insert an instruction into
+    /// one of its branches delay slot or its end.
+    bool examinePred(MachineBasicBlock &Pred, const MachineBasicBlock &Succ,
+                     RegDefsUses &RegDU, bool &HasMultipleSuccs,
+                     BB2BrMap &BrMap) const;
+
+    bool terminateSearch(const MachineInstr &Candidate) const;
 
+    TargetMachine &TM;
+    const TargetInstrInfo *TII;
 
+    static char ID;
   };
   char Filler::ID = 0;
 } // end of anonymous namespace
 
+static bool hasUnoccupiedSlot(const MachineInstr *MI) {
+  return MI->hasDelaySlot() && !MI->isBundledWithSucc();
+}
+
+/// This function inserts clones of Filler into predecessor blocks.
+static void insertDelayFiller(Iter Filler, const BB2BrMap &BrMap) {
+  MachineFunction *MF = Filler->getParent()->getParent();
+
+  for (BB2BrMap::const_iterator I = BrMap.begin(); I != BrMap.end(); ++I) {
+    if (I->second) {
+      MIBundleBuilder(I->second).append(MF->CloneMachineInstr(&*Filler));
+      ++UsefulSlots;
+    } else {
+      I->first->insert(I->first->end(), MF->CloneMachineInstr(&*Filler));
+    }
+  }
+}
+
+/// This function adds registers Filler defines to MBB's live-in register list.
+static void addLiveInRegs(Iter Filler, MachineBasicBlock &MBB) {
+  for (unsigned I = 0, E = Filler->getNumOperands(); I != E; ++I) {
+    const MachineOperand &MO = Filler->getOperand(I);
+    unsigned R;
+
+    if (!MO.isReg() || !MO.isDef() || !(R = MO.getReg()))
+      continue;
+
+#ifndef NDEBUG
+    const MachineFunction &MF = *MBB.getParent();
+    assert(MF.getTarget().getRegisterInfo()->getAllocatableSet(MF).test(R) &&
+           "Shouldn't move an instruction with unallocatable registers across "
+           "basic block boundaries.");
+#endif
+
+    if (!MBB.isLiveIn(R))
+      MBB.addLiveIn(R);
+  }
+}
+
+RegDefsUses::RegDefsUses(TargetMachine &TM)
+  : TRI(*TM.getRegisterInfo()), Defs(TRI.getNumRegs(), false),
+    Uses(TRI.getNumRegs(), false) {}
+
+void RegDefsUses::init(const MachineInstr &MI) {
+  // Add all register operands which are explicit and non-variadic.
+  update(MI, 0, MI.getDesc().getNumOperands());
+
+  // If MI is a call, add RA to Defs to prevent users of RA from going into
+  // delay slot.
+  if (MI.isCall())
+    Defs.set(Mips::RA);
+
+  // Add all implicit register operands of branch instructions except
+  // register AT.
+  if (MI.isBranch()) {
+    update(MI, MI.getDesc().getNumOperands(), MI.getNumOperands());
+    Defs.reset(Mips::AT);
+  }
+}
+
+void RegDefsUses::setCallerSaved(const MachineInstr &MI) {
+  assert(MI.isCall());
+
+  // If MI is a call, add all caller-saved registers to Defs.
+  BitVector CallerSavedRegs(TRI.getNumRegs(), true);
+
+  CallerSavedRegs.reset(Mips::ZERO);
+  CallerSavedRegs.reset(Mips::ZERO_64);
+
+  for (const MCPhysReg *R = TRI.getCalleeSavedRegs(); *R; ++R)
+    for (MCRegAliasIterator AI(*R, &TRI, true); AI.isValid(); ++AI)
+      CallerSavedRegs.reset(*AI);
+
+  Defs |= CallerSavedRegs;
+}
+
+void RegDefsUses::setUnallocatableRegs(const MachineFunction &MF) {
+  BitVector AllocSet = TRI.getAllocatableSet(MF);
+
+  for (int R = AllocSet.find_first(); R != -1; R = AllocSet.find_next(R))
+    for (MCRegAliasIterator AI(R, &TRI, false); AI.isValid(); ++AI)
+      AllocSet.set(*AI);
+
+  AllocSet.set(Mips::ZERO);
+  AllocSet.set(Mips::ZERO_64);
+
+  Defs |= AllocSet.flip();
+}
+
+void RegDefsUses::addLiveOut(const MachineBasicBlock &MBB,
+                             const MachineBasicBlock &SuccBB) {
+  for (MachineBasicBlock::const_succ_iterator SI = MBB.succ_begin(),
+       SE = MBB.succ_end(); SI != SE; ++SI)
+    if (*SI != &SuccBB)
+      for (MachineBasicBlock::livein_iterator LI = (*SI)->livein_begin(),
+           LE = (*SI)->livein_end(); LI != LE; ++LI)
+        Uses.set(*LI);
+}
+
+bool RegDefsUses::update(const MachineInstr &MI, unsigned Begin, unsigned End) {
+  BitVector NewDefs(TRI.getNumRegs()), NewUses(TRI.getNumRegs());
+  bool HasHazard = false;
+
+  for (unsigned I = Begin; I != End; ++I) {
+    const MachineOperand &MO = MI.getOperand(I);
+
+    if (MO.isReg() && MO.getReg())
+      HasHazard |= checkRegDefsUses(NewDefs, NewUses, MO.getReg(), MO.isDef());
+  }
+
+  Defs |= NewDefs;
+  Uses |= NewUses;
+
+  return HasHazard;
+}
+
+bool RegDefsUses::checkRegDefsUses(BitVector &NewDefs, BitVector &NewUses,
+                                   unsigned Reg, bool IsDef) const {
+  if (IsDef) {
+    NewDefs.set(Reg);
+    // check whether Reg has already been defined or used.
+    return (isRegInSet(Defs, Reg) || isRegInSet(Uses, Reg));
+  }
+
+  NewUses.set(Reg);
+  // check whether Reg has already been defined.
+  return isRegInSet(Defs, Reg);
+}
+
+bool RegDefsUses::isRegInSet(const BitVector &RegSet, unsigned Reg) const {
+  // Check Reg and all aliased Registers.
+  for (MCRegAliasIterator AI(Reg, &TRI, true); AI.isValid(); ++AI)
+    if (RegSet.test(*AI))
+      return true;
+  return false;
+}
+
+bool InspectMemInstr::hasHazard(const MachineInstr &MI) {
+  if (!MI.mayStore() && !MI.mayLoad())
+    return false;
+
+  if (ForbidMemInstr)
+    return true;
+
+  OrigSeenLoad = SeenLoad;
+  OrigSeenStore = SeenStore;
+  SeenLoad |= MI.mayLoad();
+  SeenStore |= MI.mayStore();
+
+  // If MI is an ordered or volatile memory reference, disallow moving
+  // subsequent loads and stores to delay slot.
+  if (MI.hasOrderedMemoryRef() && (OrigSeenLoad || OrigSeenStore)) {
+    ForbidMemInstr = true;
+    return true;
+  }
+
+  return hasHazard_(MI);
+}
+
+bool LoadFromStackOrConst::hasHazard_(const MachineInstr &MI) {
+  if (MI.mayStore())
+    return true;
+
+  if (!MI.hasOneMemOperand() || !(*MI.memoperands_begin())->getValue())
+    return true;
+
+  const Value *V = (*MI.memoperands_begin())->getValue();
+
+  if (isa<FixedStackPseudoSourceValue>(V))
+    return false;
+
+  if (const PseudoSourceValue *PSV = dyn_cast<const PseudoSourceValue>(V))
+    return !PSV->PseudoSourceValue::isConstant(0) &&
+      (V != PseudoSourceValue::getStack());
+
+  return true;
+}
+
+MemDefsUses::MemDefsUses(const MachineFrameInfo *MFI_)
+  : InspectMemInstr(false), MFI(MFI_), SeenNoObjLoad(false),
+    SeenNoObjStore(false) {}
+
+bool MemDefsUses::hasHazard_(const MachineInstr &MI) {
+  bool HasHazard = false;
+  SmallVector<const Value *, 4> Objs;
+
+  // Check underlying object list.
+  if (getUnderlyingObjects(MI, Objs)) {
+    for (SmallVector<const Value *, 4>::const_iterator I = Objs.begin();
+         I != Objs.end(); ++I)
+      HasHazard |= updateDefsUses(*I, MI.mayStore());
+
+    return HasHazard;
+  }
+
+  // No underlying objects found.
+  HasHazard = MI.mayStore() && (OrigSeenLoad || OrigSeenStore);
+  HasHazard |= MI.mayLoad() || OrigSeenStore;
+
+  SeenNoObjLoad |= MI.mayLoad();
+  SeenNoObjStore |= MI.mayStore();
+
+  return HasHazard;
+}
+
+bool MemDefsUses::updateDefsUses(const Value *V, bool MayStore) {
+  if (MayStore)
+    return !Defs.insert(V) || Uses.count(V) || SeenNoObjStore || SeenNoObjLoad;
+
+  Uses.insert(V);
+  return Defs.count(V) || SeenNoObjStore;
+}
+
+bool MemDefsUses::
+getUnderlyingObjects(const MachineInstr &MI,
+                     SmallVectorImpl<const Value *> &Objects) const {
+  if (!MI.hasOneMemOperand() || !(*MI.memoperands_begin())->getValue())
+    return false;
+
+  const Value *V = (*MI.memoperands_begin())->getValue();
+
+  SmallVector<Value *, 4> Objs;
+  GetUnderlyingObjects(const_cast<Value *>(V), Objs);
+
+  for (SmallVector<Value*, 4>::iterator I = Objs.begin(), E = Objs.end();
+       I != E; ++I) {
+    if (const PseudoSourceValue *PSV = dyn_cast<PseudoSourceValue>(*I)) {
+      if (PSV->isAliased(MFI))
+        return false;
+    } else if (!isIdentifiedObject(V))
+      return false;
+
+    Objects.push_back(*I);
+  }
+
+  return true;
+}
+
 /// runOnMachineBasicBlock - Fill in delay slots for the given basic block.
 /// We assume there is only one delay slot per delayed instruction.
-bool Filler::
-runOnMachineBasicBlock(MachineBasicBlock &MBB) {
+bool Filler::runOnMachineBasicBlock(MachineBasicBlock &MBB) {
   bool Changed = false;
-  LastFiller = MBB.instr_end();
-
-  for (InstrIter I = MBB.instr_begin(); I != MBB.instr_end(); ++I)
-    if (I->hasDelaySlot()) {
-      ++FilledSlots;
-      Changed = true;
-
-      InstrIter D;
-
-      // Delay slot filling is disabled at -O0.
-      if (!DisableDelaySlotFiller && (TM.getOptLevel() != CodeGenOpt::None) &&
-          findDelayInstr(MBB, I, D)) {
-        MBB.splice(llvm::next(I), &MBB, D);
-        ++UsefulSlots;
-      } else
-        BuildMI(MBB, llvm::next(I), I->getDebugLoc(), TII->get(Mips::NOP));
-
-      // Record the filler instruction that filled the delay slot.
-      // The instruction after it will be visited in the next iteration.
-      LastFiller = ++I;
-
-      // Set InsideBundle bit so that the machine verifier doesn't expect this
-      // instruction to be a terminator.
-      LastFiller->setIsInsideBundle();
-     }
-  return Changed;
 
+  for (Iter I = MBB.begin(); I != MBB.end(); ++I) {
+    if (!hasUnoccupiedSlot(&*I))
+      continue;
+
+    ++FilledSlots;
+    Changed = true;
+
+    // Delay slot filling is disabled at -O0.
+    if (!DisableDelaySlotFiller && (TM.getOptLevel() != CodeGenOpt::None)) {
+      if (searchBackward(MBB, I))
+        continue;
+
+      if (I->isTerminator()) {
+        if (searchSuccBBs(MBB, I))
+          continue;
+      } else if (searchForward(MBB, I)) {
+        continue;
+      }
+    }
+
+    // Bundle the NOP to the instruction with the delay slot.
+    BuildMI(MBB, llvm::next(I), I->getDebugLoc(), TII->get(Mips::NOP));
+    MIBundleBuilder(MBB, I, llvm::next(llvm::next(I)));
+  }
+
+  return Changed;
 }
 
 /// createMipsDelaySlotFillerPass - Returns a pass that fills in delay
@@ -141,129 +527,195 @@ FunctionPass *llvm::createMipsDelaySlotFillerPass(MipsTargetMachine &tm) {
   return new Filler(tm);
 }
 
-bool Filler::findDelayInstr(MachineBasicBlock &MBB,
-                            InstrIter slot,
-                            InstrIter &Filler) {
-  SmallSet<unsigned, 32> RegDefs;
-  SmallSet<unsigned, 32> RegUses;
-
-  insertDefsUses(slot, RegDefs, RegUses);
-
-  bool sawLoad = false;
-  bool sawStore = false;
-
-  for (ReverseInstrIter I(slot); I != MBB.instr_rend(); ++I) {
+template<typename IterTy>
+bool Filler::searchRange(MachineBasicBlock &MBB, IterTy Begin, IterTy End,
+                         RegDefsUses &RegDU, InspectMemInstr& IM,
+                         IterTy &Filler) const {
+  for (IterTy I = Begin; I != End; ++I) {
     // skip debug value
     if (I->isDebugValue())
       continue;
 
-    // Convert to forward iterator.
-    InstrIter FI(llvm::next(I).base());
-
-    if (I->hasUnmodeledSideEffects()
-        || I->isInlineAsm()
-        || I->isLabel()
-        || FI == LastFiller
-        || I->isPseudo()
-        //
-        // Should not allow:
-        // ERET, DERET or WAIT, PAUSE. Need to add these to instruction
-        // list. TBD.
-        )
+    if (terminateSearch(*I))
       break;
 
-    if (delayHasHazard(FI, sawLoad, sawStore, RegDefs, RegUses)) {
-      insertDefsUses(FI, RegDefs, RegUses);
+    assert((!I->isCall() && !I->isReturn() && !I->isBranch()) &&
+           "Cannot put calls, returns or branches in delay slot.");
+
+    if (delayHasHazard(*I, RegDU, IM))
       continue;
-    }
 
-    Filler = FI;
+    Filler = I;
     return true;
   }
 
   return false;
 }
 
-bool Filler::delayHasHazard(InstrIter candidate,
-                            bool &sawLoad, bool &sawStore,
-                            SmallSet<unsigned, 32> &RegDefs,
-                            SmallSet<unsigned, 32> &RegUses) {
-  if (candidate->isImplicitDef() || candidate->isKill())
-    return true;
+bool Filler::searchBackward(MachineBasicBlock &MBB, Iter Slot) const {
+  if (DisableBackwardSearch)
+    return false;
 
-  // Loads or stores cannot be moved past a store to the delay slot
-  // and stores cannot be moved past a load.
-  if (candidate->mayLoad()) {
-    if (sawStore)
-      return true;
-    sawLoad = true;
-  }
+  RegDefsUses RegDU(TM);
+  MemDefsUses MemDU(MBB.getParent()->getFrameInfo());
+  ReverseIter Filler;
 
-  if (candidate->mayStore()) {
-    if (sawStore)
-      return true;
-    sawStore = true;
-    if (sawLoad)
-      return true;
+  RegDU.init(*Slot);
+
+  if (searchRange(MBB, ReverseIter(Slot), MBB.rend(), RegDU, MemDU, Filler)) {
+    MBB.splice(llvm::next(Slot), &MBB, llvm::next(Filler).base());
+    MIBundleBuilder(MBB, Slot, llvm::next(llvm::next(Slot)));
+    ++UsefulSlots;
+    return true;
   }
 
-  assert((!candidate->isCall() && !candidate->isReturn()) &&
-         "Cannot put calls or returns in delay slot.");
+  return false;
+}
 
-  for (unsigned i = 0, e = candidate->getNumOperands(); i!= e; ++i) {
-    const MachineOperand &MO = candidate->getOperand(i);
-    unsigned Reg;
+bool Filler::searchForward(MachineBasicBlock &MBB, Iter Slot) const {
+  // Can handle only calls.
+  if (DisableForwardSearch || !Slot->isCall())
+    return false;
 
-    if (!MO.isReg() || !(Reg = MO.getReg()))
-      continue; // skip
+  RegDefsUses RegDU(TM);
+  NoMemInstr NM;
+  Iter Filler;
 
-    if (MO.isDef()) {
-      // check whether Reg is defined or used before delay slot.
-      if (IsRegInSet(RegDefs, Reg) || IsRegInSet(RegUses, Reg))
-        return true;
-    }
-    if (MO.isUse()) {
-      // check whether Reg is defined before delay slot.
-      if (IsRegInSet(RegDefs, Reg))
-        return true;
-    }
+  RegDU.setCallerSaved(*Slot);
+
+  if (searchRange(MBB, llvm::next(Slot), MBB.end(), RegDU, NM, Filler)) {
+    MBB.splice(llvm::next(Slot), &MBB, Filler);
+    MIBundleBuilder(MBB, Slot, llvm::next(llvm::next(Slot)));
+    ++UsefulSlots;
+    return true;
   }
+
   return false;
 }
 
-// Insert Defs and Uses of MI into the sets RegDefs and RegUses.
-void Filler::insertDefsUses(InstrIter MI,
-                            SmallSet<unsigned, 32> &RegDefs,
-                            SmallSet<unsigned, 32> &RegUses) {
-  // If MI is a call or return, just examine the explicit non-variadic operands.
-  MCInstrDesc MCID = MI->getDesc();
-  unsigned e = MI->isCall() || MI->isReturn() ? MCID.getNumOperands() :
-                                                MI->getNumOperands();
+bool Filler::searchSuccBBs(MachineBasicBlock &MBB, Iter Slot) const {
+  if (DisableSuccBBSearch)
+    return false;
+
+  MachineBasicBlock *SuccBB = selectSuccBB(MBB);
+
+  if (!SuccBB)
+    return false;
+
+  RegDefsUses RegDU(TM);
+  bool HasMultipleSuccs = false;
+  BB2BrMap BrMap;
+  OwningPtr<InspectMemInstr> IM;
+  Iter Filler;
+
+  // Iterate over SuccBB's predecessor list.
+  for (MachineBasicBlock::pred_iterator PI = SuccBB->pred_begin(),
+       PE = SuccBB->pred_end(); PI != PE; ++PI)
+    if (!examinePred(**PI, *SuccBB, RegDU, HasMultipleSuccs, BrMap))
+      return false;
+
+  // Do not allow moving instructions which have unallocatable register operands
+  // across basic block boundaries.
+  RegDU.setUnallocatableRegs(*MBB.getParent());
+
+  // Only allow moving loads from stack or constants if any of the SuccBB's
+  // predecessors have multiple successors.
+  if (HasMultipleSuccs) {
+    IM.reset(new LoadFromStackOrConst());
+  } else {
+    const MachineFrameInfo *MFI = MBB.getParent()->getFrameInfo();
+    IM.reset(new MemDefsUses(MFI));
+  }
 
-  // Add RA to RegDefs to prevent users of RA from going into delay slot.
-  if (MI->isCall())
-    RegDefs.insert(Mips::RA);
+  if (!searchRange(MBB, SuccBB->begin(), SuccBB->end(), RegDU, *IM, Filler))
+    return false;
 
-  for (unsigned i = 0; i != e; ++i) {
-    const MachineOperand &MO = MI->getOperand(i);
-    unsigned Reg;
+  insertDelayFiller(Filler, BrMap);
+  addLiveInRegs(Filler, *SuccBB);
+  Filler->eraseFromParent();
 
-    if (!MO.isReg() || !(Reg = MO.getReg()))
-      continue;
+  return true;
+}
+
+MachineBasicBlock *Filler::selectSuccBB(MachineBasicBlock &B) const {
+  if (B.succ_empty())
+    return NULL;
+
+  // Select the successor with the larget edge weight.
+  CmpWeight Cmp(B, getAnalysis<MachineBranchProbabilityInfo>());
+  MachineBasicBlock *S = *std::max_element(B.succ_begin(), B.succ_end(), Cmp);
+  return S->isLandingPad() ? NULL : S;
+}
+
+std::pair<MipsInstrInfo::BranchType, MachineInstr *>
+Filler::getBranch(MachineBasicBlock &MBB, const MachineBasicBlock &Dst) const {
+  const MipsInstrInfo *TII =
+    static_cast<const MipsInstrInfo*>(TM.getInstrInfo());
+  MachineBasicBlock *TrueBB = 0, *FalseBB = 0;
+  SmallVector<MachineInstr*, 2> BranchInstrs;
+  SmallVector<MachineOperand, 2> Cond;
+
+  MipsInstrInfo::BranchType R =
+    TII->AnalyzeBranch(MBB, TrueBB, FalseBB, Cond, false, BranchInstrs);
+
+  if ((R == MipsInstrInfo::BT_None) || (R == MipsInstrInfo::BT_NoBranch))
+    return std::make_pair(R, (MachineInstr*)NULL);
+
+  if (R != MipsInstrInfo::BT_CondUncond) {
+    if (!hasUnoccupiedSlot(BranchInstrs[0]))
+      return std::make_pair(MipsInstrInfo::BT_None, (MachineInstr*)NULL);
+
+    assert(((R != MipsInstrInfo::BT_Uncond) || (TrueBB == &Dst)));
+
+    return std::make_pair(R, BranchInstrs[0]);
+  }
+
+  assert((TrueBB == &Dst) || (FalseBB == &Dst));
 
-    if (MO.isDef())
-      RegDefs.insert(Reg);
-    else if (MO.isUse())
-      RegUses.insert(Reg);
+  // Examine the conditional branch. See if its slot is occupied.
+  if (hasUnoccupiedSlot(BranchInstrs[0]))
+    return std::make_pair(MipsInstrInfo::BT_Cond, BranchInstrs[0]);
+
+  // If that fails, try the unconditional branch.
+  if (hasUnoccupiedSlot(BranchInstrs[1]) && (FalseBB == &Dst))
+    return std::make_pair(MipsInstrInfo::BT_Uncond, BranchInstrs[1]);
+
+  return std::make_pair(MipsInstrInfo::BT_None, (MachineInstr*)NULL);
+}
+
+bool Filler::examinePred(MachineBasicBlock &Pred, const MachineBasicBlock &Succ,
+                         RegDefsUses &RegDU, bool &HasMultipleSuccs,
+                         BB2BrMap &BrMap) const {
+  std::pair<MipsInstrInfo::BranchType, MachineInstr *> P =
+    getBranch(Pred, Succ);
+
+  // Return if either getBranch wasn't able to analyze the branches or there
+  // were no branches with unoccupied slots.
+  if (P.first == MipsInstrInfo::BT_None)
+    return false;
+
+  if ((P.first != MipsInstrInfo::BT_Uncond) &&
+      (P.first != MipsInstrInfo::BT_NoBranch)) {
+    HasMultipleSuccs = true;
+    RegDU.addLiveOut(Pred, Succ);
   }
+
+  BrMap[&Pred] = P.second;
+  return true;
 }
 
-//returns true if the Reg or its alias is in the RegSet.
-bool Filler::IsRegInSet(SmallSet<unsigned, 32> &RegSet, unsigned Reg) {
-  // Check Reg and all aliased Registers.
-  for (MCRegAliasIterator AI(Reg, TM.getRegisterInfo(), true);
-       AI.isValid(); ++AI)
-    if (RegSet.count(*AI))
-      return true;
-  return false;
+bool Filler::delayHasHazard(const MachineInstr &Candidate, RegDefsUses &RegDU,
+                            InspectMemInstr &IM) const {
+  bool HasHazard = (Candidate.isImplicitDef() || Candidate.isKill());
+
+  HasHazard |= IM.hasHazard(Candidate);
+  HasHazard |= RegDU.update(Candidate, 0, Candidate.getNumOperands());
+
+  return HasHazard;
+}
+
+bool Filler::terminateSearch(const MachineInstr &Candidate) const {
+  return (Candidate.isTerminator() || Candidate.isCall() ||
+          Candidate.isLabel() || Candidate.isInlineAsm() ||
+          Candidate.hasUnmodeledSideEffects());
 }
diff --git a/lib/Target/Mips/MipsFrameLowering.cpp b/lib/Target/Mips/MipsFrameLowering.cpp
index 2cad2a6264ab..eb9d49fefb2f 100644
--- a/lib/Target/Mips/MipsFrameLowering.cpp
+++ b/lib/Target/Mips/MipsFrameLowering.cpp
@@ -12,20 +12,20 @@
 //===----------------------------------------------------------------------===//
 
 #include "MipsFrameLowering.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
 #include "MipsAnalyzeImmediate.h"
 #include "MipsInstrInfo.h"
 #include "MipsMachineFunction.h"
 #include "MipsTargetMachine.h"
-#include "MCTargetDesc/MipsBaseInfo.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
diff --git a/lib/Target/Mips/MipsFrameLowering.h b/lib/Target/Mips/MipsFrameLowering.h
index df52d92da830..6a5f79d0dfc4 100644
--- a/lib/Target/Mips/MipsFrameLowering.h
+++ b/lib/Target/Mips/MipsFrameLowering.h
@@ -26,9 +26,8 @@ protected:
   const MipsSubtarget &STI;
 
 public:
-  explicit MipsFrameLowering(const MipsSubtarget &sti)
-    : TargetFrameLowering(StackGrowsDown, sti.hasMips64() ? 16 : 8, 0,
-                          sti.hasMips64() ? 16 : 8), STI(sti) {}
+  explicit MipsFrameLowering(const MipsSubtarget &sti, unsigned Alignment)
+    : TargetFrameLowering(StackGrowsDown, Alignment, 0, Alignment), STI(sti) {}
 
   static const MipsFrameLowering *create(MipsTargetMachine &TM,
                                          const MipsSubtarget &ST);
@@ -39,7 +38,7 @@ protected:
   uint64_t estimateStackSize(const MachineFunction &MF) const;
 };
 
-/// Create MipsInstrInfo objects.
+/// Create MipsFrameLowering objects.
 const MipsFrameLowering *createMips16FrameLowering(const MipsSubtarget &ST);
 const MipsFrameLowering *createMipsSEFrameLowering(const MipsSubtarget &ST);
 
diff --git a/lib/Target/Mips/MipsISelDAGToDAG.cpp b/lib/Target/Mips/MipsISelDAGToDAG.cpp
index c5fca7f4b27a..77b08cb11e0c 100644
--- a/lib/Target/Mips/MipsISelDAGToDAG.cpp
+++ b/lib/Target/Mips/MipsISelDAGToDAG.cpp
@@ -12,29 +12,29 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "mips-isel"
+#include "MipsISelDAGToDAG.h"
+#include "Mips16ISelDAGToDAG.h"
+#include "MipsSEISelDAGToDAG.h"
 #include "Mips.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
 #include "MipsAnalyzeImmediate.h"
 #include "MipsMachineFunction.h"
 #include "MipsRegisterInfo.h"
-#include "MipsSubtarget.h"
-#include "MipsTargetMachine.h"
-#include "MCTargetDesc/MipsBaseInfo.h"
-#include "llvm/GlobalValue.h"
-#include "llvm/Instructions.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/Support/CFG.h"
-#include "llvm/Type.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
-#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/CodeGen/SelectionDAGNodes.h"
-#include "llvm/Target/TargetMachine.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Type.h"
+#include "llvm/Support/CFG.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetMachine.h"
 using namespace llvm;
 
 //===----------------------------------------------------------------------===//
@@ -45,263 +45,11 @@ using namespace llvm;
 // MipsDAGToDAGISel - MIPS specific code to select MIPS machine
 // instructions for SelectionDAG operations.
 //===----------------------------------------------------------------------===//
-namespace {
-
-class MipsDAGToDAGISel : public SelectionDAGISel {
-
-  /// TM - Keep a reference to MipsTargetMachine.
-  MipsTargetMachine &TM;
-
-  /// Subtarget - Keep a pointer to the MipsSubtarget around so that we can
-  /// make the right decision when generating code for different targets.
-  const MipsSubtarget &Subtarget;
-
-public:
-  explicit MipsDAGToDAGISel(MipsTargetMachine &tm) :
-  SelectionDAGISel(tm),
-  TM(tm), Subtarget(tm.getSubtarget<MipsSubtarget>()) {}
-
-  // Pass Name
-  virtual const char *getPassName() const {
-    return "MIPS DAG->DAG Pattern Instruction Selection";
-  }
-
-  virtual bool runOnMachineFunction(MachineFunction &MF);
-
-private:
-  // Include the pieces autogenerated from the target description.
-  #include "MipsGenDAGISel.inc"
-
-  /// getTargetMachine - Return a reference to the TargetMachine, casted
-  /// to the target-specific type.
-  const MipsTargetMachine &getTargetMachine() {
-    return static_cast<const MipsTargetMachine &>(TM);
-  }
-
-  /// getInstrInfo - Return a reference to the TargetInstrInfo, casted
-  /// to the target-specific type.
-  const MipsInstrInfo *getInstrInfo() {
-    return getTargetMachine().getInstrInfo();
-  }
-
-  SDNode *getGlobalBaseReg();
-
-  SDValue getMips16SPAliasReg();
-
-  void getMips16SPRefReg(SDNode *parent, SDValue &AliasReg);
-
-  std::pair<SDNode*, SDNode*> SelectMULT(SDNode *N, unsigned Opc, DebugLoc dl,
-                                         EVT Ty, bool HasLo, bool HasHi);
-
-  SDNode *Select(SDNode *N);
-
-  // Complex Pattern.
-  bool SelectAddr(SDNode *Parent, SDValue N, SDValue &Base, SDValue &Offset);
-
-  bool SelectAddr16(SDNode *Parent, SDValue N, SDValue &Base, SDValue &Offset,
-       SDValue &Alias);
-
-  // getImm - Return a target constant with the specified value.
-  inline SDValue getImm(const SDNode *Node, unsigned Imm) {
-    return CurDAG->getTargetConstant(Imm, Node->getValueType(0));
-  }
-
-  void ProcessFunctionAfterISel(MachineFunction &MF);
-  bool ReplaceUsesWithZeroReg(MachineRegisterInfo *MRI, const MachineInstr&);
-  void InitGlobalBaseReg(MachineFunction &MF);
-  void InitMips16SPAliasReg(MachineFunction &MF);
-
-  virtual bool SelectInlineAsmMemoryOperand(const SDValue &Op,
-                                            char ConstraintCode,
-                                            std::vector<SDValue> &OutOps);
-};
-
-}
-
-// Insert instructions to initialize the global base register in the
-// first MBB of the function. When the ABI is O32 and the relocation model is
-// PIC, the necessary instructions are emitted later to prevent optimization
-// passes from moving them.
-void MipsDAGToDAGISel::InitGlobalBaseReg(MachineFunction &MF) {
-  MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
-
-  if (!MipsFI->globalBaseRegSet())
-    return;
-
-  MachineBasicBlock &MBB = MF.front();
-  MachineBasicBlock::iterator I = MBB.begin();
-  MachineRegisterInfo &RegInfo = MF.getRegInfo();
-  const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
-  DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
-  unsigned V0, V1, V2, GlobalBaseReg = MipsFI->getGlobalBaseReg();
-  const TargetRegisterClass *RC;
-
-  if (Subtarget.isABI_N64())
-    RC = (const TargetRegisterClass*)&Mips::CPU64RegsRegClass;
-  else if (Subtarget.inMips16Mode())
-    RC = (const TargetRegisterClass*)&Mips::CPU16RegsRegClass;
-  else
-    RC = (const TargetRegisterClass*)&Mips::CPURegsRegClass;
-
-  V0 = RegInfo.createVirtualRegister(RC);
-  V1 = RegInfo.createVirtualRegister(RC);
-  V2 = RegInfo.createVirtualRegister(RC);
-
-  if (Subtarget.isABI_N64()) {
-    MF.getRegInfo().addLiveIn(Mips::T9_64);
-    MBB.addLiveIn(Mips::T9_64);
-
-    // lui $v0, %hi(%neg(%gp_rel(fname)))
-    // daddu $v1, $v0, $t9
-    // daddiu $globalbasereg, $v1, %lo(%neg(%gp_rel(fname)))
-    const GlobalValue *FName = MF.getFunction();
-    BuildMI(MBB, I, DL, TII.get(Mips::LUi64), V0)
-      .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_HI);
-    BuildMI(MBB, I, DL, TII.get(Mips::DADDu), V1).addReg(V0)
-      .addReg(Mips::T9_64);
-    BuildMI(MBB, I, DL, TII.get(Mips::DADDiu), GlobalBaseReg).addReg(V1)
-      .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_LO);
-    return;
-  }
-
-  if (Subtarget.inMips16Mode()) {
-    BuildMI(MBB, I, DL, TII.get(Mips::LiRxImmX16), V0)
-      .addExternalSymbol("_gp_disp", MipsII::MO_ABS_HI);
-    BuildMI(MBB, I, DL, TII.get(Mips::AddiuRxPcImmX16), V1)
-      .addExternalSymbol("_gp_disp", MipsII::MO_ABS_LO);
-    BuildMI(MBB, I, DL, TII.get(Mips::SllX16), V2).addReg(V0).addImm(16);
-    BuildMI(MBB, I, DL, TII.get(Mips::AdduRxRyRz16), GlobalBaseReg)
-      .addReg(V1).addReg(V2);
-    return;
-  }
-
-  if (MF.getTarget().getRelocationModel() == Reloc::Static) {
-    // Set global register to __gnu_local_gp.
-    //
-    // lui   $v0, %hi(__gnu_local_gp)
-    // addiu $globalbasereg, $v0, %lo(__gnu_local_gp)
-    BuildMI(MBB, I, DL, TII.get(Mips::LUi), V0)
-      .addExternalSymbol("__gnu_local_gp", MipsII::MO_ABS_HI);
-    BuildMI(MBB, I, DL, TII.get(Mips::ADDiu), GlobalBaseReg).addReg(V0)
-      .addExternalSymbol("__gnu_local_gp", MipsII::MO_ABS_LO);
-    return;
-  }
-
-  MF.getRegInfo().addLiveIn(Mips::T9);
-  MBB.addLiveIn(Mips::T9);
-
-  if (Subtarget.isABI_N32()) {
-    // lui $v0, %hi(%neg(%gp_rel(fname)))
-    // addu $v1, $v0, $t9
-    // addiu $globalbasereg, $v1, %lo(%neg(%gp_rel(fname)))
-    const GlobalValue *FName = MF.getFunction();
-    BuildMI(MBB, I, DL, TII.get(Mips::LUi), V0)
-      .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_HI);
-    BuildMI(MBB, I, DL, TII.get(Mips::ADDu), V1).addReg(V0).addReg(Mips::T9);
-    BuildMI(MBB, I, DL, TII.get(Mips::ADDiu), GlobalBaseReg).addReg(V1)
-      .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_LO);
-    return;
-  }
-
-  assert(Subtarget.isABI_O32());
-
-  // For O32 ABI, the following instruction sequence is emitted to initialize
-  // the global base register:
-  //
-  //  0. lui   $2, %hi(_gp_disp)
-  //  1. addiu $2, $2, %lo(_gp_disp)
-  //  2. addu  $globalbasereg, $2, $t9
-  //
-  // We emit only the last instruction here.
-  //
-  // GNU linker requires that the first two instructions appear at the beginning
-  // of a function and no instructions be inserted before or between them.
-  // The two instructions are emitted during lowering to MC layer in order to
-  // avoid any reordering.
-  //
-  // Register $2 (Mips::V0) is added to the list of live-in registers to ensure
-  // the value instruction 1 (addiu) defines is valid when instruction 2 (addu)
-  // reads it.
-  MF.getRegInfo().addLiveIn(Mips::V0);
-  MBB.addLiveIn(Mips::V0);
-  BuildMI(MBB, I, DL, TII.get(Mips::ADDu), GlobalBaseReg)
-    .addReg(Mips::V0).addReg(Mips::T9);
-}
-
-// Insert instructions to initialize the Mips16 SP Alias register in the
-// first MBB of the function.
-//
-void MipsDAGToDAGISel::InitMips16SPAliasReg(MachineFunction &MF) {
-  MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
-
-  if (!MipsFI->mips16SPAliasRegSet())
-    return;
-
-  MachineBasicBlock &MBB = MF.front();
-  MachineBasicBlock::iterator I = MBB.begin();
-  const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
-  DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
-  unsigned Mips16SPAliasReg = MipsFI->getMips16SPAliasReg();
-
-  BuildMI(MBB, I, DL, TII.get(Mips::MoveR3216), Mips16SPAliasReg)
-    .addReg(Mips::SP);
-}
-
-
-bool MipsDAGToDAGISel::ReplaceUsesWithZeroReg(MachineRegisterInfo *MRI,
-                                              const MachineInstr& MI) {
-  unsigned DstReg = 0, ZeroReg = 0;
-
-  // Check if MI is "addiu $dst, $zero, 0" or "daddiu $dst, $zero, 0".
-  if ((MI.getOpcode() == Mips::ADDiu) &&
-      (MI.getOperand(1).getReg() == Mips::ZERO) &&
-      (MI.getOperand(2).getImm() == 0)) {
-    DstReg = MI.getOperand(0).getReg();
-    ZeroReg = Mips::ZERO;
-  } else if ((MI.getOpcode() == Mips::DADDiu) &&
-             (MI.getOperand(1).getReg() == Mips::ZERO_64) &&
-             (MI.getOperand(2).getImm() == 0)) {
-    DstReg = MI.getOperand(0).getReg();
-    ZeroReg = Mips::ZERO_64;
-  }
-
-  if (!DstReg)
-    return false;
-
-  // Replace uses with ZeroReg.
-  for (MachineRegisterInfo::use_iterator U = MRI->use_begin(DstReg),
-       E = MRI->use_end(); U != E;) {
-    MachineOperand &MO = U.getOperand();
-    unsigned OpNo = U.getOperandNo();
-    MachineInstr *MI = MO.getParent();
-    ++U;
-
-    // Do not replace if it is a phi's operand or is tied to def operand.
-    if (MI->isPHI() || MI->isRegTiedToDefOperand(OpNo) || MI->isPseudo())
-      continue;
-
-    MO.setReg(ZeroReg);
-  }
-
-  return true;
-}
-
-void MipsDAGToDAGISel::ProcessFunctionAfterISel(MachineFunction &MF) {
-  InitGlobalBaseReg(MF);
-  InitMips16SPAliasReg(MF);
-
-  MachineRegisterInfo *MRI = &MF.getRegInfo();
-
-  for (MachineFunction::iterator MFI = MF.begin(), MFE = MF.end(); MFI != MFE;
-       ++MFI)
-    for (MachineBasicBlock::iterator I = MFI->begin(); I != MFI->end(); ++I)
-      ReplaceUsesWithZeroReg(MRI, *I);
-}
 
 bool MipsDAGToDAGISel::runOnMachineFunction(MachineFunction &MF) {
   bool Ret = SelectionDAGISel::runOnMachineFunction(MF);
 
-  ProcessFunctionAfterISel(MF);
+  processFunctionAfterISel(MF);
 
   return Ret;
 }
@@ -313,230 +61,36 @@ SDNode *MipsDAGToDAGISel::getGlobalBaseReg() {
   return CurDAG->getRegister(GlobalBaseReg, TLI.getPointerTy()).getNode();
 }
 
-/// getMips16SPAliasReg - Output the instructions required to put the
-/// SP into a Mips16 accessible aliased register.
-SDValue MipsDAGToDAGISel::getMips16SPAliasReg() {
-  unsigned Mips16SPAliasReg =
-    MF->getInfo<MipsFunctionInfo>()->getMips16SPAliasReg();
-  return CurDAG->getRegister(Mips16SPAliasReg, TLI.getPointerTy());
-}
-
 /// ComplexPattern used on MipsInstrInfo
 /// Used on Mips Load/Store instructions
-bool MipsDAGToDAGISel::
-SelectAddr(SDNode *Parent, SDValue Addr, SDValue &Base, SDValue &Offset) {
-  EVT ValTy = Addr.getValueType();
-
-  // if Address is FI, get the TargetFrameIndex.
-  if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) {
-    Base   = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy);
-    Offset = CurDAG->getTargetConstant(0, ValTy);
-    return true;
-  }
-
-  // on PIC code Load GA
-  if (Addr.getOpcode() == MipsISD::Wrapper) {
-    Base   = Addr.getOperand(0);
-    Offset = Addr.getOperand(1);
-    return true;
-  }
-
-  if (TM.getRelocationModel() != Reloc::PIC_) {
-    if ((Addr.getOpcode() == ISD::TargetExternalSymbol ||
-        Addr.getOpcode() == ISD::TargetGlobalAddress))
-      return false;
-  }
-
-  // Addresses of the form FI+const or FI|const
-  if (CurDAG->isBaseWithConstantOffset(Addr)) {
-    ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Addr.getOperand(1));
-    if (isInt<16>(CN->getSExtValue())) {
-
-      // If the first operand is a FI, get the TargetFI Node
-      if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>
-                                  (Addr.getOperand(0)))
-        Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy);
-      else
-        Base = Addr.getOperand(0);
-
-      Offset = CurDAG->getTargetConstant(CN->getZExtValue(), ValTy);
-      return true;
-    }
-  }
-
-  // Operand is a result from an ADD.
-  if (Addr.getOpcode() == ISD::ADD) {
-    // When loading from constant pools, load the lower address part in
-    // the instruction itself. Example, instead of:
-    //  lui $2, %hi($CPI1_0)
-    //  addiu $2, $2, %lo($CPI1_0)
-    //  lwc1 $f0, 0($2)
-    // Generate:
-    //  lui $2, %hi($CPI1_0)
-    //  lwc1 $f0, %lo($CPI1_0)($2)
-    if (Addr.getOperand(1).getOpcode() == MipsISD::Lo ||
-        Addr.getOperand(1).getOpcode() == MipsISD::GPRel) {
-      SDValue Opnd0 = Addr.getOperand(1).getOperand(0);
-      if (isa<ConstantPoolSDNode>(Opnd0) || isa<GlobalAddressSDNode>(Opnd0) ||
-          isa<JumpTableSDNode>(Opnd0)) {
-        Base = Addr.getOperand(0);
-        Offset = Opnd0;
-        return true;
-      }
-    }
-
-    // If an indexed floating point load/store can be emitted, return false.
-    const LSBaseSDNode *LS = dyn_cast<LSBaseSDNode>(Parent);
-
-    if (LS &&
-        (LS->getMemoryVT() == MVT::f32 || LS->getMemoryVT() == MVT::f64) &&
-        Subtarget.hasMips32r2Or64())
-      return false;
-  }
-
-  Base   = Addr;
-  Offset = CurDAG->getTargetConstant(0, ValTy);
-  return true;
+bool MipsDAGToDAGISel::selectAddrRegImm(SDValue Addr, SDValue &Base,
+                                        SDValue &Offset) const {
+  llvm_unreachable("Unimplemented function.");
+  return false;
 }
 
-void MipsDAGToDAGISel::getMips16SPRefReg(SDNode *Parent, SDValue &AliasReg) {
-  SDValue AliasFPReg = CurDAG->getRegister(Mips::S0, TLI.getPointerTy());
-  if (Parent) {
-    switch (Parent->getOpcode()) {
-      case ISD::LOAD: {
-        LoadSDNode *SD = dyn_cast<LoadSDNode>(Parent);
-        switch (SD->getMemoryVT().getSizeInBits()) {
-        case 8:
-        case 16:
-          AliasReg = TM.getFrameLowering()->hasFP(*MF)?
-            AliasFPReg: getMips16SPAliasReg();
-          return;
-        }
-        break;
-      }
-      case ISD::STORE: {
-        StoreSDNode *SD = dyn_cast<StoreSDNode>(Parent);
-        switch (SD->getMemoryVT().getSizeInBits()) {
-        case 8:
-        case 16:
-          AliasReg = TM.getFrameLowering()->hasFP(*MF)?
-            AliasFPReg: getMips16SPAliasReg();
-          return;
-        }
-        break;
-      }
-    }
-  }
-  AliasReg = CurDAG->getRegister(Mips::SP, TLI.getPointerTy());
-  return;
-
+bool MipsDAGToDAGISel::selectAddrDefault(SDValue Addr, SDValue &Base,
+                                         SDValue &Offset) const {
+  llvm_unreachable("Unimplemented function.");
+  return false;
 }
-bool MipsDAGToDAGISel::SelectAddr16(
-  SDNode *Parent, SDValue Addr, SDValue &Base, SDValue &Offset,
-  SDValue &Alias) {
-  EVT ValTy = Addr.getValueType();
-
-  Alias = CurDAG->getTargetConstant(0, ValTy);
-
-  // if Address is FI, get the TargetFrameIndex.
-  if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) {
-    Base   = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy);
-    Offset = CurDAG->getTargetConstant(0, ValTy);
-    getMips16SPRefReg(Parent, Alias);
-    return true;
-  }
-  // on PIC code Load GA
-  if (Addr.getOpcode() == MipsISD::Wrapper) {
-    Base   = Addr.getOperand(0);
-    Offset = Addr.getOperand(1);
-    return true;
-  }
-  if (TM.getRelocationModel() != Reloc::PIC_) {
-    if ((Addr.getOpcode() == ISD::TargetExternalSymbol ||
-        Addr.getOpcode() == ISD::TargetGlobalAddress))
-      return false;
-  }
-  // Addresses of the form FI+const or FI|const
-  if (CurDAG->isBaseWithConstantOffset(Addr)) {
-    ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Addr.getOperand(1));
-    if (isInt<16>(CN->getSExtValue())) {
-
-      // If the first operand is a FI, get the TargetFI Node
-      if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>
-                                  (Addr.getOperand(0))) {
-        Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy);
-        getMips16SPRefReg(Parent, Alias);
-      }
-      else
-        Base = Addr.getOperand(0);
-
-      Offset = CurDAG->getTargetConstant(CN->getZExtValue(), ValTy);
-      return true;
-    }
-  }
-  // Operand is a result from an ADD.
-  if (Addr.getOpcode() == ISD::ADD) {
-    // When loading from constant pools, load the lower address part in
-    // the instruction itself. Example, instead of:
-    //  lui $2, %hi($CPI1_0)
-    //  addiu $2, $2, %lo($CPI1_0)
-    //  lwc1 $f0, 0($2)
-    // Generate:
-    //  lui $2, %hi($CPI1_0)
-    //  lwc1 $f0, %lo($CPI1_0)($2)
-    if (Addr.getOperand(1).getOpcode() == MipsISD::Lo ||
-        Addr.getOperand(1).getOpcode() == MipsISD::GPRel) {
-      SDValue Opnd0 = Addr.getOperand(1).getOperand(0);
-      if (isa<ConstantPoolSDNode>(Opnd0) || isa<GlobalAddressSDNode>(Opnd0) ||
-          isa<JumpTableSDNode>(Opnd0)) {
-        Base = Addr.getOperand(0);
-        Offset = Opnd0;
-        return true;
-      }
-    }
 
-    // If an indexed floating point load/store can be emitted, return false.
-    const LSBaseSDNode *LS = dyn_cast<LSBaseSDNode>(Parent);
-
-    if (LS &&
-        (LS->getMemoryVT() == MVT::f32 || LS->getMemoryVT() == MVT::f64) &&
-        Subtarget.hasMips32r2Or64())
-      return false;
-  }
-  Base   = Addr;
-  Offset = CurDAG->getTargetConstant(0, ValTy);
-  return true;
+bool MipsDAGToDAGISel::selectIntAddr(SDValue Addr, SDValue &Base,
+                                     SDValue &Offset) const {
+  llvm_unreachable("Unimplemented function.");
+  return false;
 }
 
-/// Select multiply instructions.
-std::pair<SDNode*, SDNode*>
-MipsDAGToDAGISel::SelectMULT(SDNode *N, unsigned Opc, DebugLoc dl, EVT Ty,
-                             bool HasLo, bool HasHi) {
-  SDNode *Lo = 0, *Hi = 0;
-  SDNode *Mul = CurDAG->getMachineNode(Opc, dl, MVT::Glue, N->getOperand(0),
-                                       N->getOperand(1));
-  SDValue InFlag = SDValue(Mul, 0);
-
-  if (HasLo) {
-    unsigned Opcode = Subtarget.inMips16Mode() ? Mips::Mflo16 :
-      (Ty == MVT::i32 ? Mips::MFLO : Mips::MFLO64);
-    Lo = CurDAG->getMachineNode(Opcode, dl, Ty, MVT::Glue, InFlag);
-    InFlag = SDValue(Lo, 1);
-  }
-  if (HasHi) {
-    unsigned Opcode = Subtarget.inMips16Mode() ? Mips::Mfhi16 :
-      (Ty == MVT::i32 ? Mips::MFHI : Mips::MFHI64);
-    Hi = CurDAG->getMachineNode(Opcode, dl, Ty, InFlag);
-  }
-  return std::make_pair(Lo, Hi);
+bool MipsDAGToDAGISel::selectAddr16(SDNode *Parent, SDValue N, SDValue &Base,
+                                    SDValue &Offset, SDValue &Alias) {
+  llvm_unreachable("Unimplemented function.");
+  return false;
 }
 
-
 /// Select instructions not customized! Used for
 /// expanded, promoted and normal instructions
 SDNode* MipsDAGToDAGISel::Select(SDNode *Node) {
   unsigned Opcode = Node->getOpcode();
-  DebugLoc dl = Node->getDebugLoc();
 
   // Dump information about the Node being selected
   DEBUG(errs() << "Selecting: "; Node->dump(CurDAG); errs() << "\n");
@@ -547,167 +101,19 @@ SDNode* MipsDAGToDAGISel::Select(SDNode *Node) {
     return NULL;
   }
 
-  ///
-  // Instruction Selection not handled by the auto-generated
-  // tablegen selection should be handled here.
-  ///
-  EVT NodeTy = Node->getValueType(0);
-  unsigned MultOpc;
+  // See if subclasses can handle this node.
+  std::pair<bool, SDNode*> Ret = selectNode(Node);
+
+  if (Ret.first)
+    return Ret.second;
 
   switch(Opcode) {
   default: break;
 
-  case ISD::SUBE:
-  case ISD::ADDE: {
-    bool inMips16Mode = Subtarget.inMips16Mode();
-    SDValue InFlag = Node->getOperand(2), CmpLHS;
-    unsigned Opc = InFlag.getOpcode(); (void)Opc;
-    assert(((Opc == ISD::ADDC || Opc == ISD::ADDE) ||
-            (Opc == ISD::SUBC || Opc == ISD::SUBE)) &&
-           "(ADD|SUB)E flag operand must come from (ADD|SUB)C/E insn");
-
-    unsigned MOp;
-    if (Opcode == ISD::ADDE) {
-      CmpLHS = InFlag.getValue(0);
-      if (inMips16Mode)
-        MOp = Mips::AdduRxRyRz16;
-      else
-        MOp = Mips::ADDu;
-    } else {
-      CmpLHS = InFlag.getOperand(0);
-      if (inMips16Mode)
-        MOp = Mips::SubuRxRyRz16;
-      else
-        MOp = Mips::SUBu;
-    }
-
-    SDValue Ops[] = { CmpLHS, InFlag.getOperand(1) };
-
-    SDValue LHS = Node->getOperand(0);
-    SDValue RHS = Node->getOperand(1);
-
-    EVT VT = LHS.getValueType();
-
-    unsigned Sltu_op = inMips16Mode? Mips::SltuRxRyRz16: Mips::SLTu;
-    SDNode *Carry = CurDAG->getMachineNode(Sltu_op, dl, VT, Ops, 2);
-    unsigned Addu_op = inMips16Mode? Mips::AdduRxRyRz16 : Mips::ADDu;
-    SDNode *AddCarry = CurDAG->getMachineNode(Addu_op, dl, VT,
-                                              SDValue(Carry,0), RHS);
-
-    return CurDAG->SelectNodeTo(Node, MOp, VT, MVT::Glue,
-                                LHS, SDValue(AddCarry,0));
-  }
-
-  /// Mul with two results
-  case ISD::SMUL_LOHI:
-  case ISD::UMUL_LOHI: {
-    if (NodeTy == MVT::i32) {
-      if (Subtarget.inMips16Mode())
-        MultOpc = (Opcode == ISD::UMUL_LOHI ? Mips::MultuRxRy16 :
-                   Mips::MultRxRy16);
-      else
-        MultOpc = (Opcode == ISD::UMUL_LOHI ? Mips::MULTu : Mips::MULT);
-    }
-    else
-      MultOpc = (Opcode == ISD::UMUL_LOHI ? Mips::DMULTu : Mips::DMULT);
-
-    std::pair<SDNode*, SDNode*> LoHi = SelectMULT(Node, MultOpc, dl, NodeTy,
-                                                  true, true);
-
-    if (!SDValue(Node, 0).use_empty())
-      ReplaceUses(SDValue(Node, 0), SDValue(LoHi.first, 0));
-
-    if (!SDValue(Node, 1).use_empty())
-      ReplaceUses(SDValue(Node, 1), SDValue(LoHi.second, 0));
-
-    return NULL;
-  }
-
-  /// Special Muls
-  case ISD::MUL: {
-    // Mips32 has a 32-bit three operand mul instruction.
-    if (Subtarget.hasMips32() && NodeTy == MVT::i32)
-      break;
-    return SelectMULT(Node, NodeTy == MVT::i32 ? Mips::MULT : Mips::DMULT,
-                      dl, NodeTy, true, false).first;
-  }
-  case ISD::MULHS:
-  case ISD::MULHU: {
-    if (NodeTy == MVT::i32) {
-      if (Subtarget.inMips16Mode())
-        MultOpc = (Opcode == ISD::MULHU ?
-                   Mips::MultuRxRy16 : Mips::MultRxRy16);
-      else
-        MultOpc = (Opcode == ISD::MULHU ? Mips::MULTu : Mips::MULT);
-    }
-    else
-      MultOpc = (Opcode == ISD::MULHU ? Mips::DMULTu : Mips::DMULT);
-
-    return SelectMULT(Node, MultOpc, dl, NodeTy, false, true).second;
-  }
-
   // Get target GOT address.
   case ISD::GLOBAL_OFFSET_TABLE:
     return getGlobalBaseReg();
 
-  case ISD::ConstantFP: {
-    ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(Node);
-    if (Node->getValueType(0) == MVT::f64 && CN->isExactlyValue(+0.0)) {
-      if (Subtarget.hasMips64()) {
-        SDValue Zero = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
-                                              Mips::ZERO_64, MVT::i64);
-        return CurDAG->getMachineNode(Mips::DMTC1, dl, MVT::f64, Zero);
-      }
-
-      SDValue Zero = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
-                                            Mips::ZERO, MVT::i32);
-      return CurDAG->getMachineNode(Mips::BuildPairF64, dl, MVT::f64, Zero,
-                                    Zero);
-    }
-    break;
-  }
-
-  case ISD::Constant: {
-    const ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Node);
-    unsigned Size = CN->getValueSizeInBits(0);
-
-    if (Size == 32)
-      break;
-
-    MipsAnalyzeImmediate AnalyzeImm;
-    int64_t Imm = CN->getSExtValue();
-
-    const MipsAnalyzeImmediate::InstSeq &Seq =
-      AnalyzeImm.Analyze(Imm, Size, false);
-
-    MipsAnalyzeImmediate::InstSeq::const_iterator Inst = Seq.begin();
-    DebugLoc DL = CN->getDebugLoc();
-    SDNode *RegOpnd;
-    SDValue ImmOpnd = CurDAG->getTargetConstant(SignExtend64<16>(Inst->ImmOpnd),
-                                                MVT::i64);
-
-    // The first instruction can be a LUi which is different from other
-    // instructions (ADDiu, ORI and SLL) in that it does not have a register
-    // operand.
-    if (Inst->Opc == Mips::LUi64)
-      RegOpnd = CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64, ImmOpnd);
-    else
-      RegOpnd =
-        CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64,
-                               CurDAG->getRegister(Mips::ZERO_64, MVT::i64),
-                               ImmOpnd);
-
-    // The remaining instructions in the sequence are handled here.
-    for (++Inst; Inst != Seq.end(); ++Inst) {
-      ImmOpnd = CurDAG->getTargetConstant(SignExtend64<16>(Inst->ImmOpnd),
-                                          MVT::i64);
-      RegOpnd = CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64,
-                                       SDValue(RegOpnd, 0), ImmOpnd);
-    }
-
-    return RegOpnd;
-  }
-
 #ifndef NDEBUG
   case ISD::LOAD:
   case ISD::STORE:
@@ -716,31 +122,6 @@ SDNode* MipsDAGToDAGISel::Select(SDNode *Node) {
            "Unexpected unaligned loads/stores.");
     break;
 #endif
-
-  case MipsISD::ThreadPointer: {
-    EVT PtrVT = TLI.getPointerTy();
-    unsigned RdhwrOpc, SrcReg, DestReg;
-
-    if (PtrVT == MVT::i32) {
-      RdhwrOpc = Mips::RDHWR;
-      SrcReg = Mips::HWR29;
-      DestReg = Mips::V1;
-    } else {
-      RdhwrOpc = Mips::RDHWR64;
-      SrcReg = Mips::HWR29_64;
-      DestReg = Mips::V1_64;
-    }
-
-    SDNode *Rdhwr =
-      CurDAG->getMachineNode(RdhwrOpc, Node->getDebugLoc(),
-                             Node->getValueType(0),
-                             CurDAG->getRegister(SrcReg, PtrVT));
-    SDValue Chain = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, DestReg,
-                                         SDValue(Rdhwr, 0));
-    SDValue ResNode = CurDAG->getCopyFromReg(Chain, dl, DestReg, PtrVT);
-    ReplaceUses(SDValue(Node, 0), ResNode);
-    return ResNode.getNode();
-  }
   }
 
   // Select the default instruction
@@ -766,5 +147,8 @@ SelectInlineAsmMemoryOperand(const SDValue &Op, char ConstraintCode,
 /// createMipsISelDag - This pass converts a legalized DAG into a
 /// MIPS-specific DAG, ready for instruction scheduling.
 FunctionPass *llvm::createMipsISelDag(MipsTargetMachine &TM) {
-  return new MipsDAGToDAGISel(TM);
+  if (TM.getSubtargetImpl()->inMips16Mode())
+    return llvm::createMips16ISelDag(TM);
+
+  return llvm::createMipsSEISelDag(TM);
 }
diff --git a/lib/Target/Mips/MipsISelDAGToDAG.h b/lib/Target/Mips/MipsISelDAGToDAG.h
new file mode 100644
index 000000000000..cf0f9c58aa9c
--- /dev/null
+++ b/lib/Target/Mips/MipsISelDAGToDAG.h
@@ -0,0 +1,93 @@
+//===---- MipsISelDAGToDAG.h - A Dag to Dag Inst Selector for Mips --------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines an instruction selector for the MIPS target.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MIPSISELDAGTODAG_H
+#define MIPSISELDAGTODAG_H
+
+#include "Mips.h"
+#include "MipsSubtarget.h"
+#include "MipsTargetMachine.h"
+#include "llvm/CodeGen/SelectionDAGISel.h"
+
+//===----------------------------------------------------------------------===//
+// Instruction Selector Implementation
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// MipsDAGToDAGISel - MIPS specific code to select MIPS machine
+// instructions for SelectionDAG operations.
+//===----------------------------------------------------------------------===//
+namespace llvm {
+
+class MipsDAGToDAGISel : public SelectionDAGISel {
+public:
+  explicit MipsDAGToDAGISel(MipsTargetMachine &TM)
+    : SelectionDAGISel(TM), Subtarget(TM.getSubtarget<MipsSubtarget>()) {}
+
+  // Pass Name
+  virtual const char *getPassName() const {
+    return "MIPS DAG->DAG Pattern Instruction Selection";
+  }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF);
+
+protected:
+  SDNode *getGlobalBaseReg();
+
+  /// Keep a pointer to the MipsSubtarget around so that we can make the right
+  /// decision when generating code for different targets.
+  const MipsSubtarget &Subtarget;
+
+private:
+  // Include the pieces autogenerated from the target description.
+  #include "MipsGenDAGISel.inc"
+
+  // Complex Pattern.
+  /// (reg + imm).
+  virtual bool selectAddrRegImm(SDValue Addr, SDValue &Base,
+                                SDValue &Offset) const;
+
+  /// Fall back on this function if all else fails.
+  virtual bool selectAddrDefault(SDValue Addr, SDValue &Base,
+                                 SDValue &Offset) const;
+
+  /// Match integer address pattern.
+  virtual bool selectIntAddr(SDValue Addr, SDValue &Base,
+                             SDValue &Offset) const;
+
+  virtual bool selectAddr16(SDNode *Parent, SDValue N, SDValue &Base,
+                            SDValue &Offset, SDValue &Alias);
+
+  virtual SDNode *Select(SDNode *N);
+
+  virtual std::pair<bool, SDNode*> selectNode(SDNode *Node) = 0;
+
+  // getImm - Return a target constant with the specified value.
+  inline SDValue getImm(const SDNode *Node, uint64_t Imm) {
+    return CurDAG->getTargetConstant(Imm, Node->getValueType(0));
+  }
+
+  virtual void processFunctionAfterISel(MachineFunction &MF) = 0;
+
+  virtual bool SelectInlineAsmMemoryOperand(const SDValue &Op,
+                                            char ConstraintCode,
+                                            std::vector<SDValue> &OutOps);
+};
+
+/// createMipsISelDag - This pass converts a legalized DAG into a
+/// MIPS-specific DAG, ready for instruction scheduling.
+FunctionPass *createMipsISelDag(MipsTargetMachine &TM);
+
+}
+
+#endif
diff --git a/lib/Target/Mips/MipsISelLowering.cpp b/lib/Target/Mips/MipsISelLowering.cpp
index b0dd0a766f70..e2219f257ecd 100644
--- a/lib/Target/Mips/MipsISelLowering.cpp
+++ b/lib/Target/Mips/MipsISelLowering.cpp
@@ -11,20 +11,14 @@
 // selection DAG.
 //
 //===----------------------------------------------------------------------===//
-
 #define DEBUG_TYPE "mips-lower"
 #include "MipsISelLowering.h"
+#include "InstPrinter/MipsInstPrinter.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
 #include "MipsMachineFunction.h"
+#include "MipsSubtarget.h"
 #include "MipsTargetMachine.h"
 #include "MipsTargetObjectFile.h"
-#include "MipsSubtarget.h"
-#include "InstPrinter/MipsInstPrinter.h"
-#include "MCTargetDesc/MipsBaseInfo.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/CallingConv.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
@@ -33,6 +27,10 @@
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
@@ -43,10 +41,6 @@ using namespace llvm;
 STATISTIC(NumTailCalls, "Number of tail calls");
 
 static cl::opt<bool>
-EnableMipsTailCalls("enable-mips-tail-calls", cl::Hidden,
-                    cl::desc("MIPS: Enable tail calls."), cl::init(false));
-
-static cl::opt<bool>
 LargeGOT("mxgot", cl::Hidden,
          cl::desc("MIPS: Enable GOT larger than 64k."), cl::init(false));
 
@@ -67,7 +61,7 @@ static const uint16_t Mips64DPRegs[8] = {
 // If I is a shifted mask, set the size (Size) and the first bit of the
 // mask (Pos), and return true.
 // For example, if I is 0x003ff800, (Pos, Size) = (11, 11).
-static bool IsShiftedMask(uint64_t I, uint64_t &Pos, uint64_t &Size) {
+static bool isShiftedMask(uint64_t I, uint64_t &Pos, uint64_t &Size) {
   if (!isShiftedMask_64(I))
      return false;
 
@@ -76,7 +70,7 @@ static bool IsShiftedMask(uint64_t I, uint64_t &Pos, uint64_t &Size) {
   return true;
 }
 
-static SDValue GetGlobalReg(SelectionDAG &DAG, EVT Ty) {
+SDValue MipsTargetLowering::getGlobalReg(SelectionDAG &DAG, EVT Ty) const {
   MipsFunctionInfo *FI = DAG.getMachineFunction().getInfo<MipsFunctionInfo>();
   return DAG.getRegister(FI->getGlobalBaseReg(), Ty);
 }
@@ -111,11 +105,12 @@ static SDValue getAddrNonPIC(SDValue Op, SelectionDAG &DAG) {
                      DAG.getNode(MipsISD::Lo, DL, Ty, Lo));
 }
 
-static SDValue getAddrLocal(SDValue Op, SelectionDAG &DAG, bool HasMips64) {
+SDValue MipsTargetLowering::getAddrLocal(SDValue Op, SelectionDAG &DAG,
+                                         bool HasMips64) const {
   DebugLoc DL = Op.getDebugLoc();
   EVT Ty = Op.getValueType();
   unsigned GOTFlag = HasMips64 ? MipsII::MO_GOT_PAGE : MipsII::MO_GOT;
-  SDValue GOT = DAG.getNode(MipsISD::Wrapper, DL, Ty, GetGlobalReg(DAG, Ty),
+  SDValue GOT = DAG.getNode(MipsISD::Wrapper, DL, Ty, getGlobalReg(DAG, Ty),
                             getTargetNode(Op, DAG, GOTFlag));
   SDValue Load = DAG.getLoad(Ty, DL, DAG.getEntryNode(), GOT,
                              MachinePointerInfo::getGOT(), false, false, false,
@@ -125,21 +120,23 @@ static SDValue getAddrLocal(SDValue Op, SelectionDAG &DAG, bool HasMips64) {
   return DAG.getNode(ISD::ADD, DL, Ty, Load, Lo);
 }
 
-static SDValue getAddrGlobal(SDValue Op, SelectionDAG &DAG, unsigned Flag) {
+SDValue MipsTargetLowering::getAddrGlobal(SDValue Op, SelectionDAG &DAG,
+                                          unsigned Flag) const {
   DebugLoc DL = Op.getDebugLoc();
   EVT Ty = Op.getValueType();
-  SDValue Tgt = DAG.getNode(MipsISD::Wrapper, DL, Ty, GetGlobalReg(DAG, Ty),
+  SDValue Tgt = DAG.getNode(MipsISD::Wrapper, DL, Ty, getGlobalReg(DAG, Ty),
                             getTargetNode(Op, DAG, Flag));
   return DAG.getLoad(Ty, DL, DAG.getEntryNode(), Tgt,
                      MachinePointerInfo::getGOT(), false, false, false, 0);
 }
 
-static SDValue getAddrGlobalLargeGOT(SDValue Op, SelectionDAG &DAG,
-                                     unsigned HiFlag, unsigned LoFlag) {
+SDValue MipsTargetLowering::getAddrGlobalLargeGOT(SDValue Op, SelectionDAG &DAG,
+                                                  unsigned HiFlag,
+                                                  unsigned LoFlag) const {
   DebugLoc DL = Op.getDebugLoc();
   EVT Ty = Op.getValueType();
   SDValue Hi = DAG.getNode(MipsISD::Hi, DL, Ty, getTargetNode(Op, DAG, HiFlag));
-  Hi = DAG.getNode(ISD::ADD, DL, Ty, Hi, GetGlobalReg(DAG, Ty));
+  Hi = DAG.getNode(ISD::ADD, DL, Ty, Hi, getGlobalReg(DAG, Ty));
   SDValue Wrapper = DAG.getNode(MipsISD::Wrapper, DL, Ty, Hi,
                                 getTargetNode(Op, DAG, LoFlag));
   return DAG.getLoad(Ty, DL, DAG.getEntryNode(), Wrapper,
@@ -155,21 +152,27 @@ const char *MipsTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case MipsISD::GPRel:             return "MipsISD::GPRel";
   case MipsISD::ThreadPointer:     return "MipsISD::ThreadPointer";
   case MipsISD::Ret:               return "MipsISD::Ret";
+  case MipsISD::EH_RETURN:         return "MipsISD::EH_RETURN";
   case MipsISD::FPBrcond:          return "MipsISD::FPBrcond";
   case MipsISD::FPCmp:             return "MipsISD::FPCmp";
   case MipsISD::CMovFP_T:          return "MipsISD::CMovFP_T";
   case MipsISD::CMovFP_F:          return "MipsISD::CMovFP_F";
   case MipsISD::FPRound:           return "MipsISD::FPRound";
+  case MipsISD::ExtractLOHI:       return "MipsISD::ExtractLOHI";
+  case MipsISD::InsertLOHI:        return "MipsISD::InsertLOHI";
+  case MipsISD::Mult:              return "MipsISD::Mult";
+  case MipsISD::Multu:             return "MipsISD::Multu";
   case MipsISD::MAdd:              return "MipsISD::MAdd";
   case MipsISD::MAddu:             return "MipsISD::MAddu";
   case MipsISD::MSub:              return "MipsISD::MSub";
   case MipsISD::MSubu:             return "MipsISD::MSubu";
   case MipsISD::DivRem:            return "MipsISD::DivRem";
   case MipsISD::DivRemU:           return "MipsISD::DivRemU";
+  case MipsISD::DivRem16:          return "MipsISD::DivRem16";
+  case MipsISD::DivRemU16:         return "MipsISD::DivRemU16";
   case MipsISD::BuildPairF64:      return "MipsISD::BuildPairF64";
   case MipsISD::ExtractElementF64: return "MipsISD::ExtractElementF64";
   case MipsISD::Wrapper:           return "MipsISD::Wrapper";
-  case MipsISD::DynAlloc:          return "MipsISD::DynAlloc";
   case MipsISD::Sync:              return "MipsISD::Sync";
   case MipsISD::Ext:               return "MipsISD::Ext";
   case MipsISD::Ins:               return "MipsISD::Ins";
@@ -191,7 +194,7 @@ const char *MipsTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case MipsISD::MTHLIP:            return "MipsISD::MTHLIP";
   case MipsISD::MULT:              return "MipsISD::MULT";
   case MipsISD::MULTU:             return "MipsISD::MULTU";
-  case MipsISD::MADD_DSP:          return "MipsISD::MADD_DSPDSP";
+  case MipsISD::MADD_DSP:          return "MipsISD::MADD_DSP";
   case MipsISD::MADDU_DSP:         return "MipsISD::MADDU_DSP";
   case MipsISD::MSUB_DSP:          return "MipsISD::MSUB_DSP";
   case MipsISD::MSUBU_DSP:         return "MipsISD::MSUBU_DSP";
@@ -205,50 +208,11 @@ MipsTargetLowering(MipsTargetMachine &TM)
     Subtarget(&TM.getSubtarget<MipsSubtarget>()),
     HasMips64(Subtarget->hasMips64()), IsN64(Subtarget->isABI_N64()),
     IsO32(Subtarget->isABI_O32()) {
-
   // Mips does not have i1 type, so use i32 for
   // setcc operations results (slt, sgt, ...).
   setBooleanContents(ZeroOrOneBooleanContent);
   setBooleanVectorContents(ZeroOrOneBooleanContent); // FIXME: Is this correct?
 
-  // Set up the register classes
-  addRegisterClass(MVT::i32, &Mips::CPURegsRegClass);
-
-  if (HasMips64)
-    addRegisterClass(MVT::i64, &Mips::CPU64RegsRegClass);
-
-  if (Subtarget->inMips16Mode()) {
-    addRegisterClass(MVT::i32, &Mips::CPU16RegsRegClass);
-  }
-
-  if (Subtarget->hasDSP()) {
-    MVT::SimpleValueType VecTys[2] = {MVT::v2i16, MVT::v4i8};
-
-    for (unsigned i = 0; i < array_lengthof(VecTys); ++i) {
-      addRegisterClass(VecTys[i], &Mips::DSPRegsRegClass);
-
-      // Expand all builtin opcodes.
-      for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc)
-        setOperationAction(Opc, VecTys[i], Expand);
-
-      setOperationAction(ISD::LOAD, VecTys[i], Legal);
-      setOperationAction(ISD::STORE, VecTys[i], Legal);
-      setOperationAction(ISD::BITCAST, VecTys[i], Legal);
-    }
-  }
-
-  if (!TM.Options.UseSoftFloat) {
-    addRegisterClass(MVT::f32, &Mips::FGR32RegClass);
-
-    // When dealing with single precision only, use libcalls
-    if (!Subtarget->isSingleFloat()) {
-      if (HasMips64)
-        addRegisterClass(MVT::f64, &Mips::FGR64RegClass);
-      else
-        addRegisterClass(MVT::f64, &Mips::AFGR64RegClass);
-    }
-  }
-
   // Load extented operations for i1 types must be promoted
   setLoadExtAction(ISD::EXTLOAD,  MVT::i1,  Promote);
   setLoadExtAction(ISD::ZEXTLOAD, MVT::i1,  Promote);
@@ -265,6 +229,7 @@ MipsTargetLowering(MipsTargetMachine &TM)
   AddPromotedToType(ISD::SETCC, MVT::i1, MVT::i32);
 
   // Mips Custom Operations
+  setOperationAction(ISD::BR_JT,              MVT::Other, Custom);
   setOperationAction(ISD::GlobalAddress,      MVT::i32,   Custom);
   setOperationAction(ISD::BlockAddress,       MVT::i32,   Custom);
   setOperationAction(ISD::GlobalTLSAddress,   MVT::i32,   Custom);
@@ -281,18 +246,6 @@ MipsTargetLowering(MipsTargetMachine &TM)
   setOperationAction(ISD::VASTART,            MVT::Other, Custom);
   setOperationAction(ISD::FCOPYSIGN,          MVT::f32,   Custom);
   setOperationAction(ISD::FCOPYSIGN,          MVT::f64,   Custom);
-  if (Subtarget->inMips16Mode()) {
-    setOperationAction(ISD::MEMBARRIER,         MVT::Other, Expand);
-    setOperationAction(ISD::ATOMIC_FENCE,       MVT::Other, Expand);
-  }
-  else {
-    setOperationAction(ISD::MEMBARRIER,         MVT::Other, Custom);
-    setOperationAction(ISD::ATOMIC_FENCE,       MVT::Other, Custom);
-  }
-  if (!Subtarget->inMips16Mode()) {
-    setOperationAction(ISD::LOAD,               MVT::i32, Custom);
-    setOperationAction(ISD::STORE,              MVT::i32, Custom);
-  }
 
   if (!TM.Options.NoNaNsFPMath) {
     setOperationAction(ISD::FABS,             MVT::f32,   Custom);
@@ -330,8 +283,10 @@ MipsTargetLowering(MipsTargetMachine &TM)
   setOperationAction(ISD::UREM, MVT::i64, Expand);
 
   // Operations not directly supported by Mips.
-  setOperationAction(ISD::BR_JT,             MVT::Other, Expand);
-  setOperationAction(ISD::BR_CC,             MVT::Other, Expand);
+  setOperationAction(ISD::BR_CC,             MVT::f32,   Expand);
+  setOperationAction(ISD::BR_CC,             MVT::f64,   Expand);
+  setOperationAction(ISD::BR_CC,             MVT::i32,   Expand);
+  setOperationAction(ISD::BR_CC,             MVT::i64,   Expand);
   setOperationAction(ISD::SELECT_CC,         MVT::Other, Expand);
   setOperationAction(ISD::UINT_TO_FP,        MVT::i32,   Expand);
   setOperationAction(ISD::UINT_TO_FP,        MVT::i64,   Expand);
@@ -361,6 +316,8 @@ MipsTargetLowering(MipsTargetMachine &TM)
   setOperationAction(ISD::FSIN,              MVT::f64,   Expand);
   setOperationAction(ISD::FCOS,              MVT::f32,   Expand);
   setOperationAction(ISD::FCOS,              MVT::f64,   Expand);
+  setOperationAction(ISD::FSINCOS,           MVT::f32,   Expand);
+  setOperationAction(ISD::FSINCOS,           MVT::f64,   Expand);
   setOperationAction(ISD::FPOWI,             MVT::f32,   Expand);
   setOperationAction(ISD::FPOW,              MVT::f32,   Expand);
   setOperationAction(ISD::FPOW,              MVT::f64,   Expand);
@@ -383,6 +340,8 @@ MipsTargetLowering(MipsTargetMachine &TM)
   setOperationAction(ISD::EHSELECTION,       MVT::i32, Expand);
   setOperationAction(ISD::EHSELECTION,       MVT::i64, Expand);
 
+  setOperationAction(ISD::EH_RETURN, MVT::Other, Custom);
+
   setOperationAction(ISD::VAARG,             MVT::Other, Expand);
   setOperationAction(ISD::VACOPY,            MVT::Other, Expand);
   setOperationAction(ISD::VAEND,             MVT::Other, Expand);
@@ -399,21 +358,6 @@ MipsTargetLowering(MipsTargetMachine &TM)
   setOperationAction(ISD::ATOMIC_STORE,      MVT::i32,    Expand);
   setOperationAction(ISD::ATOMIC_STORE,      MVT::i64,    Expand);
 
-  if (Subtarget->inMips16Mode()) {
-    setOperationAction(ISD::ATOMIC_CMP_SWAP,       MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_SWAP,           MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_LOAD_ADD,       MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_LOAD_SUB,       MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_LOAD_AND,       MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_LOAD_OR,        MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_LOAD_XOR,       MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_LOAD_NAND,      MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_LOAD_MIN,       MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_LOAD_MAX,       MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_LOAD_UMIN,      MVT::i32,    Expand);
-    setOperationAction(ISD::ATOMIC_LOAD_UMAX,      MVT::i32,    Expand);
-  }
-
   setInsertFencesForAtomic(true);
 
   if (!Subtarget->hasSEInReg()) {
@@ -438,8 +382,6 @@ MipsTargetLowering(MipsTargetMachine &TM)
     setTruncStoreAction(MVT::i64, MVT::i32, Custom);
   }
 
-  setTargetDAGCombine(ISD::ADDE);
-  setTargetDAGCombine(ISD::SUBE);
   setTargetDAGCombine(ISD::SDIVREM);
   setTargetDAGCombine(ISD::UDIVREM);
   setTargetDAGCombine(ISD::SELECT);
@@ -450,206 +392,27 @@ MipsTargetLowering(MipsTargetMachine &TM)
   setMinFunctionAlignment(HasMips64 ? 3 : 2);
 
   setStackPointerRegisterToSaveRestore(IsN64 ? Mips::SP_64 : Mips::SP);
-  computeRegisterProperties();
 
   setExceptionPointerRegister(IsN64 ? Mips::A0_64 : Mips::A0);
   setExceptionSelectorRegister(IsN64 ? Mips::A1_64 : Mips::A1);
 
-  maxStoresPerMemcpy = 16;
+  MaxStoresPerMemcpy = 16;
 }
 
-bool MipsTargetLowering::allowsUnalignedMemoryAccesses(EVT VT) const {
-  MVT::SimpleValueType SVT = VT.getSimpleVT().SimpleTy;
-
-  if (Subtarget->inMips16Mode())
-    return false;
+const MipsTargetLowering *MipsTargetLowering::create(MipsTargetMachine &TM) {
+  if (TM.getSubtargetImpl()->inMips16Mode())
+    return llvm::createMips16TargetLowering(TM);
 
-  switch (SVT) {
-  case MVT::i64:
-  case MVT::i32:
-    return true;
-  default:
-    return false;
-  }
+  return llvm::createMipsSETargetLowering(TM);
 }
 
 EVT MipsTargetLowering::getSetCCResultType(EVT VT) const {
-  return MVT::i32;
-}
-
-// SelectMadd -
-// Transforms a subgraph in CurDAG if the following pattern is found:
-//  (addc multLo, Lo0), (adde multHi, Hi0),
-// where,
-//  multHi/Lo: product of multiplication
-//  Lo0: initial value of Lo register
-//  Hi0: initial value of Hi register
-// Return true if pattern matching was successful.
-static bool SelectMadd(SDNode *ADDENode, SelectionDAG *CurDAG) {
-  // ADDENode's second operand must be a flag output of an ADDC node in order
-  // for the matching to be successful.
-  SDNode *ADDCNode = ADDENode->getOperand(2).getNode();
-
-  if (ADDCNode->getOpcode() != ISD::ADDC)
-    return false;
-
-  SDValue MultHi = ADDENode->getOperand(0);
-  SDValue MultLo = ADDCNode->getOperand(0);
-  SDNode *MultNode = MultHi.getNode();
-  unsigned MultOpc = MultHi.getOpcode();
-
-  // MultHi and MultLo must be generated by the same node,
-  if (MultLo.getNode() != MultNode)
-    return false;
-
-  // and it must be a multiplication.
-  if (MultOpc != ISD::SMUL_LOHI && MultOpc != ISD::UMUL_LOHI)
-    return false;
-
-  // MultLo amd MultHi must be the first and second output of MultNode
-  // respectively.
-  if (MultHi.getResNo() != 1 || MultLo.getResNo() != 0)
-    return false;
-
-  // Transform this to a MADD only if ADDENode and ADDCNode are the only users
-  // of the values of MultNode, in which case MultNode will be removed in later
-  // phases.
-  // If there exist users other than ADDENode or ADDCNode, this function returns
-  // here, which will result in MultNode being mapped to a single MULT
-  // instruction node rather than a pair of MULT and MADD instructions being
-  // produced.
-  if (!MultHi.hasOneUse() || !MultLo.hasOneUse())
-    return false;
-
-  SDValue Chain = CurDAG->getEntryNode();
-  DebugLoc dl = ADDENode->getDebugLoc();
-
-  // create MipsMAdd(u) node
-  MultOpc = MultOpc == ISD::UMUL_LOHI ? MipsISD::MAddu : MipsISD::MAdd;
-
-  SDValue MAdd = CurDAG->getNode(MultOpc, dl, MVT::Glue,
-                                 MultNode->getOperand(0),// Factor 0
-                                 MultNode->getOperand(1),// Factor 1
-                                 ADDCNode->getOperand(1),// Lo0
-                                 ADDENode->getOperand(1));// Hi0
-
-  // create CopyFromReg nodes
-  SDValue CopyFromLo = CurDAG->getCopyFromReg(Chain, dl, Mips::LO, MVT::i32,
-                                              MAdd);
-  SDValue CopyFromHi = CurDAG->getCopyFromReg(CopyFromLo.getValue(1), dl,
-                                              Mips::HI, MVT::i32,
-                                              CopyFromLo.getValue(2));
-
-  // replace uses of adde and addc here
-  if (!SDValue(ADDCNode, 0).use_empty())
-    CurDAG->ReplaceAllUsesOfValueWith(SDValue(ADDCNode, 0), CopyFromLo);
-
-  if (!SDValue(ADDENode, 0).use_empty())
-    CurDAG->ReplaceAllUsesOfValueWith(SDValue(ADDENode, 0), CopyFromHi);
-
-  return true;
-}
-
-// SelectMsub -
-// Transforms a subgraph in CurDAG if the following pattern is found:
-//  (addc Lo0, multLo), (sube Hi0, multHi),
-// where,
-//  multHi/Lo: product of multiplication
-//  Lo0: initial value of Lo register
-//  Hi0: initial value of Hi register
-// Return true if pattern matching was successful.
-static bool SelectMsub(SDNode *SUBENode, SelectionDAG *CurDAG) {
-  // SUBENode's second operand must be a flag output of an SUBC node in order
-  // for the matching to be successful.
-  SDNode *SUBCNode = SUBENode->getOperand(2).getNode();
-
-  if (SUBCNode->getOpcode() != ISD::SUBC)
-    return false;
-
-  SDValue MultHi = SUBENode->getOperand(1);
-  SDValue MultLo = SUBCNode->getOperand(1);
-  SDNode *MultNode = MultHi.getNode();
-  unsigned MultOpc = MultHi.getOpcode();
-
-  // MultHi and MultLo must be generated by the same node,
-  if (MultLo.getNode() != MultNode)
-    return false;
-
-  // and it must be a multiplication.
-  if (MultOpc != ISD::SMUL_LOHI && MultOpc != ISD::UMUL_LOHI)
-    return false;
-
-  // MultLo amd MultHi must be the first and second output of MultNode
-  // respectively.
-  if (MultHi.getResNo() != 1 || MultLo.getResNo() != 0)
-    return false;
-
-  // Transform this to a MSUB only if SUBENode and SUBCNode are the only users
-  // of the values of MultNode, in which case MultNode will be removed in later
-  // phases.
-  // If there exist users other than SUBENode or SUBCNode, this function returns
-  // here, which will result in MultNode being mapped to a single MULT
-  // instruction node rather than a pair of MULT and MSUB instructions being
-  // produced.
-  if (!MultHi.hasOneUse() || !MultLo.hasOneUse())
-    return false;
-
-  SDValue Chain = CurDAG->getEntryNode();
-  DebugLoc dl = SUBENode->getDebugLoc();
-
-  // create MipsSub(u) node
-  MultOpc = MultOpc == ISD::UMUL_LOHI ? MipsISD::MSubu : MipsISD::MSub;
-
-  SDValue MSub = CurDAG->getNode(MultOpc, dl, MVT::Glue,
-                                 MultNode->getOperand(0),// Factor 0
-                                 MultNode->getOperand(1),// Factor 1
-                                 SUBCNode->getOperand(0),// Lo0
-                                 SUBENode->getOperand(0));// Hi0
-
-  // create CopyFromReg nodes
-  SDValue CopyFromLo = CurDAG->getCopyFromReg(Chain, dl, Mips::LO, MVT::i32,
-                                              MSub);
-  SDValue CopyFromHi = CurDAG->getCopyFromReg(CopyFromLo.getValue(1), dl,
-                                              Mips::HI, MVT::i32,
-                                              CopyFromLo.getValue(2));
-
-  // replace uses of sube and subc here
-  if (!SDValue(SUBCNode, 0).use_empty())
-    CurDAG->ReplaceAllUsesOfValueWith(SDValue(SUBCNode, 0), CopyFromLo);
-
-  if (!SDValue(SUBENode, 0).use_empty())
-    CurDAG->ReplaceAllUsesOfValueWith(SDValue(SUBENode, 0), CopyFromHi);
-
-  return true;
-}
-
-static SDValue PerformADDECombine(SDNode *N, SelectionDAG &DAG,
-                                  TargetLowering::DAGCombinerInfo &DCI,
-                                  const MipsSubtarget *Subtarget) {
-  if (DCI.isBeforeLegalize())
-    return SDValue();
-
-  if (Subtarget->hasMips32() && N->getValueType(0) == MVT::i32 &&
-      SelectMadd(N, &DAG))
-    return SDValue(N, 0);
-
-  return SDValue();
-}
-
-static SDValue PerformSUBECombine(SDNode *N, SelectionDAG &DAG,
-                                  TargetLowering::DAGCombinerInfo &DCI,
-                                  const MipsSubtarget *Subtarget) {
-  if (DCI.isBeforeLegalize())
-    return SDValue();
-
-  if (Subtarget->hasMips32() && N->getValueType(0) == MVT::i32 &&
-      SelectMsub(N, &DAG))
-    return SDValue(N, 0);
-
-  return SDValue();
+  if (!VT.isVector())
+    return MVT::i32;
+  return VT.changeVectorElementTypeToInteger();
 }
 
-static SDValue PerformDivRemCombine(SDNode *N, SelectionDAG &DAG,
+static SDValue performDivRemCombine(SDNode *N, SelectionDAG &DAG,
                                     TargetLowering::DAGCombinerInfo &DCI,
                                     const MipsSubtarget *Subtarget) {
   if (DCI.isBeforeLegalizeOps())
@@ -658,18 +421,18 @@ static SDValue PerformDivRemCombine(SDNode *N, SelectionDAG &DAG,
   EVT Ty = N->getValueType(0);
   unsigned LO = (Ty == MVT::i32) ? Mips::LO : Mips::LO64;
   unsigned HI = (Ty == MVT::i32) ? Mips::HI : Mips::HI64;
-  unsigned opc = N->getOpcode() == ISD::SDIVREM ? MipsISD::DivRem :
-                                                  MipsISD::DivRemU;
-  DebugLoc dl = N->getDebugLoc();
+  unsigned Opc = N->getOpcode() == ISD::SDIVREM ? MipsISD::DivRem16 :
+                                                  MipsISD::DivRemU16;
+  DebugLoc DL = N->getDebugLoc();
 
-  SDValue DivRem = DAG.getNode(opc, dl, MVT::Glue,
+  SDValue DivRem = DAG.getNode(Opc, DL, MVT::Glue,
                                N->getOperand(0), N->getOperand(1));
   SDValue InChain = DAG.getEntryNode();
   SDValue InGlue = DivRem;
 
   // insert MFLO
   if (N->hasAnyUseOfValue(0)) {
-    SDValue CopyFromLo = DAG.getCopyFromReg(InChain, dl, LO, Ty,
+    SDValue CopyFromLo = DAG.getCopyFromReg(InChain, DL, LO, Ty,
                                             InGlue);
     DAG.ReplaceAllUsesOfValueWith(SDValue(N, 0), CopyFromLo);
     InChain = CopyFromLo.getValue(1);
@@ -678,7 +441,7 @@ static SDValue PerformDivRemCombine(SDNode *N, SelectionDAG &DAG,
 
   // insert MFHI
   if (N->hasAnyUseOfValue(1)) {
-    SDValue CopyFromHi = DAG.getCopyFromReg(InChain, dl,
+    SDValue CopyFromHi = DAG.getCopyFromReg(InChain, DL,
                                             HI, Ty, InGlue);
     DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), CopyFromHi);
   }
@@ -713,8 +476,9 @@ static Mips::CondCode FPCondCCodeToFCC(ISD::CondCode CC) {
 }
 
 
-// Returns true if condition code has to be inverted.
-static bool InvertFPCondCode(Mips::CondCode CC) {
+/// This function returns true if the floating point conditional branches and
+/// conditional moves which use condition code CC should be inverted.
+static bool invertFPCondCodeUser(Mips::CondCode CC) {
   if (CC >= Mips::FCOND_F && CC <= Mips::FCOND_NGT)
     return false;
 
@@ -726,7 +490,7 @@ static bool InvertFPCondCode(Mips::CondCode CC) {
 
 // Creates and returns an FPCmp node from a setcc node.
 // Returns Op if setcc is not a floating point comparison.
-static SDValue CreateFPCmp(SelectionDAG &DAG, const SDValue &Op) {
+static SDValue createFPCmp(SelectionDAG &DAG, const SDValue &Op) {
   // must be a SETCC node
   if (Op.getOpcode() != ISD::SETCC)
     return Op;
@@ -737,28 +501,27 @@ static SDValue CreateFPCmp(SelectionDAG &DAG, const SDValue &Op) {
     return Op;
 
   SDValue RHS = Op.getOperand(1);
-  DebugLoc dl = Op.getDebugLoc();
+  DebugLoc DL = Op.getDebugLoc();
 
   // Assume the 3rd operand is a CondCodeSDNode. Add code to check the type of
   // node if necessary.
   ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
 
-  return DAG.getNode(MipsISD::FPCmp, dl, MVT::Glue, LHS, RHS,
+  return DAG.getNode(MipsISD::FPCmp, DL, MVT::Glue, LHS, RHS,
                      DAG.getConstant(FPCondCCodeToFCC(CC), MVT::i32));
 }
 
 // Creates and returns a CMovFPT/F node.
-static SDValue CreateCMovFP(SelectionDAG &DAG, SDValue Cond, SDValue True,
+static SDValue createCMovFP(SelectionDAG &DAG, SDValue Cond, SDValue True,
                             SDValue False, DebugLoc DL) {
-  bool invert = InvertFPCondCode((Mips::CondCode)
-                                 cast<ConstantSDNode>(Cond.getOperand(2))
-                                 ->getSExtValue());
+  ConstantSDNode *CC = cast<ConstantSDNode>(Cond.getOperand(2));
+  bool invert = invertFPCondCodeUser((Mips::CondCode)CC->getSExtValue());
 
   return DAG.getNode((invert ? MipsISD::CMovFP_F : MipsISD::CMovFP_T), DL,
                      True.getValueType(), True, False, Cond);
 }
 
-static SDValue PerformSELECTCombine(SDNode *N, SelectionDAG &DAG,
+static SDValue performSELECTCombine(SDNode *N, SelectionDAG &DAG,
                                     TargetLowering::DAGCombinerInfo &DCI,
                                     const MipsSubtarget *Subtarget) {
   if (DCI.isBeforeLegalizeOps())
@@ -791,7 +554,7 @@ static SDValue PerformSELECTCombine(SDNode *N, SelectionDAG &DAG,
   return DAG.getNode(ISD::SELECT, DL, FalseTy, SetCC, False, True);
 }
 
-static SDValue PerformANDCombine(SDNode *N, SelectionDAG &DAG,
+static SDValue performANDCombine(SDNode *N, SelectionDAG &DAG,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const MipsSubtarget *Subtarget) {
   // Pattern match EXT.
@@ -817,7 +580,7 @@ static SDValue PerformANDCombine(SDNode *N, SelectionDAG &DAG,
 
   // Op's second operand must be a shifted mask.
   if (!(CN = dyn_cast<ConstantSDNode>(Mask)) ||
-      !IsShiftedMask(CN->getZExtValue(), SMPos, SMSize))
+      !isShiftedMask(CN->getZExtValue(), SMPos, SMSize))
     return SDValue();
 
   // Return if the shifted mask does not start at bit 0 or the sum of its size
@@ -831,7 +594,7 @@ static SDValue PerformANDCombine(SDNode *N, SelectionDAG &DAG,
                      DAG.getConstant(SMSize, MVT::i32));
 }
 
-static SDValue PerformORCombine(SDNode *N, SelectionDAG &DAG,
+static SDValue performORCombine(SDNode *N, SelectionDAG &DAG,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const MipsSubtarget *Subtarget) {
   // Pattern match INS.
@@ -850,7 +613,7 @@ static SDValue PerformORCombine(SDNode *N, SelectionDAG &DAG,
     return SDValue();
 
   if (!(CN = dyn_cast<ConstantSDNode>(And0.getOperand(1))) ||
-      !IsShiftedMask(~CN->getSExtValue(), SMPos0, SMSize0))
+      !isShiftedMask(~CN->getSExtValue(), SMPos0, SMSize0))
     return SDValue();
 
   // See if Op's second operand matches (and (shl $src, pos), mask1).
@@ -858,7 +621,7 @@ static SDValue PerformORCombine(SDNode *N, SelectionDAG &DAG,
     return SDValue();
 
   if (!(CN = dyn_cast<ConstantSDNode>(And1.getOperand(1))) ||
-      !IsShiftedMask(CN->getZExtValue(), SMPos1, SMSize1))
+      !isShiftedMask(CN->getZExtValue(), SMPos1, SMSize1))
     return SDValue();
 
   // The shift masks must have the same position and size.
@@ -885,7 +648,7 @@ static SDValue PerformORCombine(SDNode *N, SelectionDAG &DAG,
                      DAG.getConstant(SMSize0, MVT::i32), And0.getOperand(0));
 }
 
-static SDValue PerformADDCombine(SDNode *N, SelectionDAG &DAG,
+static SDValue performADDCombine(SDNode *N, SelectionDAG &DAG,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const MipsSubtarget *Subtarget) {
   // (add v0, (add v1, abs_lo(tjt))) => (add (add v0, v1), abs_lo(tjt))
@@ -915,25 +678,21 @@ static SDValue PerformADDCombine(SDNode *N, SelectionDAG &DAG,
 SDValue  MipsTargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI)
   const {
   SelectionDAG &DAG = DCI.DAG;
-  unsigned opc = N->getOpcode();
+  unsigned Opc = N->getOpcode();
 
-  switch (opc) {
+  switch (Opc) {
   default: break;
-  case ISD::ADDE:
-    return PerformADDECombine(N, DAG, DCI, Subtarget);
-  case ISD::SUBE:
-    return PerformSUBECombine(N, DAG, DCI, Subtarget);
   case ISD::SDIVREM:
   case ISD::UDIVREM:
-    return PerformDivRemCombine(N, DAG, DCI, Subtarget);
+    return performDivRemCombine(N, DAG, DCI, Subtarget);
   case ISD::SELECT:
-    return PerformSELECTCombine(N, DAG, DCI, Subtarget);
+    return performSELECTCombine(N, DAG, DCI, Subtarget);
   case ISD::AND:
-    return PerformANDCombine(N, DAG, DCI, Subtarget);
+    return performANDCombine(N, DAG, DCI, Subtarget);
   case ISD::OR:
-    return PerformORCombine(N, DAG, DCI, Subtarget);
+    return performORCombine(N, DAG, DCI, Subtarget);
   case ISD::ADD:
-    return PerformADDCombine(N, DAG, DCI, Subtarget);
+    return performADDCombine(N, DAG, DCI, Subtarget);
   }
 
   return SDValue();
@@ -964,30 +723,32 @@ LowerOperation(SDValue Op, SelectionDAG &DAG) const
 {
   switch (Op.getOpcode())
   {
-    case ISD::BRCOND:             return LowerBRCOND(Op, DAG);
-    case ISD::ConstantPool:       return LowerConstantPool(Op, DAG);
-    case ISD::GlobalAddress:      return LowerGlobalAddress(Op, DAG);
-    case ISD::BlockAddress:       return LowerBlockAddress(Op, DAG);
-    case ISD::GlobalTLSAddress:   return LowerGlobalTLSAddress(Op, DAG);
-    case ISD::JumpTable:          return LowerJumpTable(Op, DAG);
-    case ISD::SELECT:             return LowerSELECT(Op, DAG);
-    case ISD::SELECT_CC:          return LowerSELECT_CC(Op, DAG);
-    case ISD::SETCC:              return LowerSETCC(Op, DAG);
-    case ISD::VASTART:            return LowerVASTART(Op, DAG);
-    case ISD::FCOPYSIGN:          return LowerFCOPYSIGN(Op, DAG);
-    case ISD::FABS:               return LowerFABS(Op, DAG);
-    case ISD::FRAMEADDR:          return LowerFRAMEADDR(Op, DAG);
-    case ISD::RETURNADDR:         return LowerRETURNADDR(Op, DAG);
-    case ISD::MEMBARRIER:         return LowerMEMBARRIER(Op, DAG);
-    case ISD::ATOMIC_FENCE:       return LowerATOMIC_FENCE(Op, DAG);
-    case ISD::SHL_PARTS:          return LowerShiftLeftParts(Op, DAG);
-    case ISD::SRA_PARTS:          return LowerShiftRightParts(Op, DAG, true);
-    case ISD::SRL_PARTS:          return LowerShiftRightParts(Op, DAG, false);
-    case ISD::LOAD:               return LowerLOAD(Op, DAG);
-    case ISD::STORE:              return LowerSTORE(Op, DAG);
-    case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
-    case ISD::INTRINSIC_W_CHAIN:  return LowerINTRINSIC_W_CHAIN(Op, DAG);
-    case ISD::ADD:                return LowerADD(Op, DAG);
+  case ISD::BR_JT:              return lowerBR_JT(Op, DAG);
+  case ISD::BRCOND:             return lowerBRCOND(Op, DAG);
+  case ISD::ConstantPool:       return lowerConstantPool(Op, DAG);
+  case ISD::GlobalAddress:      return lowerGlobalAddress(Op, DAG);
+  case ISD::BlockAddress:       return lowerBlockAddress(Op, DAG);
+  case ISD::GlobalTLSAddress:   return lowerGlobalTLSAddress(Op, DAG);
+  case ISD::JumpTable:          return lowerJumpTable(Op, DAG);
+  case ISD::SELECT:             return lowerSELECT(Op, DAG);
+  case ISD::SELECT_CC:          return lowerSELECT_CC(Op, DAG);
+  case ISD::SETCC:              return lowerSETCC(Op, DAG);
+  case ISD::VASTART:            return lowerVASTART(Op, DAG);
+  case ISD::FCOPYSIGN:          return lowerFCOPYSIGN(Op, DAG);
+  case ISD::FABS:               return lowerFABS(Op, DAG);
+  case ISD::FRAMEADDR:          return lowerFRAMEADDR(Op, DAG);
+  case ISD::RETURNADDR:         return lowerRETURNADDR(Op, DAG);
+  case ISD::EH_RETURN:          return lowerEH_RETURN(Op, DAG);
+  case ISD::MEMBARRIER:         return lowerMEMBARRIER(Op, DAG);
+  case ISD::ATOMIC_FENCE:       return lowerATOMIC_FENCE(Op, DAG);
+  case ISD::SHL_PARTS:          return lowerShiftLeftParts(Op, DAG);
+  case ISD::SRA_PARTS:          return lowerShiftRightParts(Op, DAG, true);
+  case ISD::SRL_PARTS:          return lowerShiftRightParts(Op, DAG, false);
+  case ISD::LOAD:               return lowerLOAD(Op, DAG);
+  case ISD::STORE:              return lowerSTORE(Op, DAG);
+  case ISD::INTRINSIC_WO_CHAIN: return lowerINTRINSIC_WO_CHAIN(Op, DAG);
+  case ISD::INTRINSIC_W_CHAIN:  return lowerINTRINSIC_W_CHAIN(Op, DAG);
+  case ISD::ADD:                return lowerADD(Op, DAG);
   }
   return SDValue();
 }
@@ -996,287 +757,133 @@ LowerOperation(SDValue Op, SelectionDAG &DAG) const
 //  Lower helper functions
 //===----------------------------------------------------------------------===//
 
-// AddLiveIn - This helper function adds the specified physical register to the
+// addLiveIn - This helper function adds the specified physical register to the
 // MachineFunction as a live in value.  It also creates a corresponding
 // virtual register for it.
 static unsigned
-AddLiveIn(MachineFunction &MF, unsigned PReg, const TargetRegisterClass *RC)
+addLiveIn(MachineFunction &MF, unsigned PReg, const TargetRegisterClass *RC)
 {
-  assert(RC->contains(PReg) && "Not the correct regclass!");
   unsigned VReg = MF.getRegInfo().createVirtualRegister(RC);
   MF.getRegInfo().addLiveIn(PReg, VReg);
   return VReg;
 }
 
-// Get fp branch code (not opcode) from condition code.
-static Mips::FPBranchCode GetFPBranchCodeFromCond(Mips::CondCode CC) {
-  if (CC >= Mips::FCOND_F && CC <= Mips::FCOND_NGT)
-    return Mips::BRANCH_T;
-
-  assert((CC >= Mips::FCOND_T && CC <= Mips::FCOND_GT) &&
-         "Invalid CondCode.");
-
-  return Mips::BRANCH_F;
-}
-
-/*
-static MachineBasicBlock* ExpandCondMov(MachineInstr *MI, MachineBasicBlock *BB,
-                                        DebugLoc dl,
-                                        const MipsSubtarget *Subtarget,
-                                        const TargetInstrInfo *TII,
-                                        bool isFPCmp, unsigned Opc) {
-  // There is no need to expand CMov instructions if target has
-  // conditional moves.
-  if (Subtarget->hasCondMov())
-    return BB;
-
-  // To "insert" a SELECT_CC instruction, we actually have to insert the
-  // diamond control-flow pattern.  The incoming instruction knows the
-  // destination vreg to set, the condition code register to branch on, the
-  // true/false values to select between, and a branch opcode to use.
-  const BasicBlock *LLVM_BB = BB->getBasicBlock();
-  MachineFunction::iterator It = BB;
-  ++It;
-
-  //  thisMBB:
-  //  ...
-  //   TrueVal = ...
-  //   setcc r1, r2, r3
-  //   bNE   r1, r0, copy1MBB
-  //   fallthrough --> copy0MBB
-  MachineBasicBlock *thisMBB  = BB;
-  MachineFunction *F = BB->getParent();
-  MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
-  MachineBasicBlock *sinkMBB  = F->CreateMachineBasicBlock(LLVM_BB);
-  F->insert(It, copy0MBB);
-  F->insert(It, sinkMBB);
-
-  // Transfer the remainder of BB and its successor edges to sinkMBB.
-  sinkMBB->splice(sinkMBB->begin(), BB,
-                  llvm::next(MachineBasicBlock::iterator(MI)),
-                  BB->end());
-  sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
-
-  // Next, add the true and fallthrough blocks as its successors.
-  BB->addSuccessor(copy0MBB);
-  BB->addSuccessor(sinkMBB);
-
-  // Emit the right instruction according to the type of the operands compared
-  if (isFPCmp)
-    BuildMI(BB, dl, TII->get(Opc)).addMBB(sinkMBB);
-  else
-    BuildMI(BB, dl, TII->get(Opc)).addReg(MI->getOperand(2).getReg())
-      .addReg(Mips::ZERO).addMBB(sinkMBB);
-
-  //  copy0MBB:
-  //   %FalseValue = ...
-  //   # fallthrough to sinkMBB
-  BB = copy0MBB;
-
-  // Update machine-CFG edges
-  BB->addSuccessor(sinkMBB);
-
-  //  sinkMBB:
-  //   %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ]
-  //  ...
-  BB = sinkMBB;
-
-  if (isFPCmp)
-    BuildMI(*BB, BB->begin(), dl,
-            TII->get(Mips::PHI), MI->getOperand(0).getReg())
-      .addReg(MI->getOperand(2).getReg()).addMBB(thisMBB)
-      .addReg(MI->getOperand(1).getReg()).addMBB(copy0MBB);
-  else
-    BuildMI(*BB, BB->begin(), dl,
-            TII->get(Mips::PHI), MI->getOperand(0).getReg())
-      .addReg(MI->getOperand(3).getReg()).addMBB(thisMBB)
-      .addReg(MI->getOperand(1).getReg()).addMBB(copy0MBB);
-
-  MI->eraseFromParent();   // The pseudo instruction is gone now.
-  return BB;
-}
-*/
-
-MachineBasicBlock *
-MipsTargetLowering::EmitBPOSGE32(MachineInstr *MI, MachineBasicBlock *BB) const{
-  // $bb:
-  //  bposge32_pseudo $vr0
-  //  =>
-  // $bb:
-  //  bposge32 $tbb
-  // $fbb:
-  //  li $vr2, 0
-  //  b $sink
-  // $tbb:
-  //  li $vr1, 1
-  // $sink:
-  //  $vr0 = phi($vr2, $fbb, $vr1, $tbb)
-
-  MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
-  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
-  const TargetRegisterClass *RC = &Mips::CPURegsRegClass;
-  DebugLoc DL = MI->getDebugLoc();
-  const BasicBlock *LLVM_BB = BB->getBasicBlock();
-  MachineFunction::iterator It = llvm::next(MachineFunction::iterator(BB));
-  MachineFunction *F = BB->getParent();
-  MachineBasicBlock *FBB = F->CreateMachineBasicBlock(LLVM_BB);
-  MachineBasicBlock *TBB = F->CreateMachineBasicBlock(LLVM_BB);
-  MachineBasicBlock *Sink  = F->CreateMachineBasicBlock(LLVM_BB);
-  F->insert(It, FBB);
-  F->insert(It, TBB);
-  F->insert(It, Sink);
-
-  // Transfer the remainder of BB and its successor edges to Sink.
-  Sink->splice(Sink->begin(), BB, llvm::next(MachineBasicBlock::iterator(MI)),
-               BB->end());
-  Sink->transferSuccessorsAndUpdatePHIs(BB);
-
-  // Add successors.
-  BB->addSuccessor(FBB);
-  BB->addSuccessor(TBB);
-  FBB->addSuccessor(Sink);
-  TBB->addSuccessor(Sink);
-
-  // Insert the real bposge32 instruction to $BB.
-  BuildMI(BB, DL, TII->get(Mips::BPOSGE32)).addMBB(TBB);
-
-  // Fill $FBB.
-  unsigned VR2 = RegInfo.createVirtualRegister(RC);
-  BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::ADDiu), VR2)
-    .addReg(Mips::ZERO).addImm(0);
-  BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::B)).addMBB(Sink);
-
-  // Fill $TBB.
-  unsigned VR1 = RegInfo.createVirtualRegister(RC);
-  BuildMI(*TBB, TBB->end(), DL, TII->get(Mips::ADDiu), VR1)
-    .addReg(Mips::ZERO).addImm(1);
-
-  // Insert phi function to $Sink.
-  BuildMI(*Sink, Sink->begin(), DL, TII->get(Mips::PHI),
-          MI->getOperand(0).getReg())
-    .addReg(VR2).addMBB(FBB).addReg(VR1).addMBB(TBB);
-
-  MI->eraseFromParent();   // The pseudo instruction is gone now.
-  return Sink;
-}
-
 MachineBasicBlock *
 MipsTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
                                                 MachineBasicBlock *BB) const {
   switch (MI->getOpcode()) {
-  default: llvm_unreachable("Unexpected instr type to insert");
+  default:
+    llvm_unreachable("Unexpected instr type to insert");
   case Mips::ATOMIC_LOAD_ADD_I8:
   case Mips::ATOMIC_LOAD_ADD_I8_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 1, Mips::ADDu);
+    return emitAtomicBinaryPartword(MI, BB, 1, Mips::ADDu);
   case Mips::ATOMIC_LOAD_ADD_I16:
   case Mips::ATOMIC_LOAD_ADD_I16_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 2, Mips::ADDu);
+    return emitAtomicBinaryPartword(MI, BB, 2, Mips::ADDu);
   case Mips::ATOMIC_LOAD_ADD_I32:
   case Mips::ATOMIC_LOAD_ADD_I32_P8:
-    return EmitAtomicBinary(MI, BB, 4, Mips::ADDu);
+    return emitAtomicBinary(MI, BB, 4, Mips::ADDu);
   case Mips::ATOMIC_LOAD_ADD_I64:
   case Mips::ATOMIC_LOAD_ADD_I64_P8:
-    return EmitAtomicBinary(MI, BB, 8, Mips::DADDu);
+    return emitAtomicBinary(MI, BB, 8, Mips::DADDu);
 
   case Mips::ATOMIC_LOAD_AND_I8:
   case Mips::ATOMIC_LOAD_AND_I8_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 1, Mips::AND);
+    return emitAtomicBinaryPartword(MI, BB, 1, Mips::AND);
   case Mips::ATOMIC_LOAD_AND_I16:
   case Mips::ATOMIC_LOAD_AND_I16_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 2, Mips::AND);
+    return emitAtomicBinaryPartword(MI, BB, 2, Mips::AND);
   case Mips::ATOMIC_LOAD_AND_I32:
   case Mips::ATOMIC_LOAD_AND_I32_P8:
-    return EmitAtomicBinary(MI, BB, 4, Mips::AND);
+    return emitAtomicBinary(MI, BB, 4, Mips::AND);
   case Mips::ATOMIC_LOAD_AND_I64:
   case Mips::ATOMIC_LOAD_AND_I64_P8:
-    return EmitAtomicBinary(MI, BB, 8, Mips::AND64);
+    return emitAtomicBinary(MI, BB, 8, Mips::AND64);
 
   case Mips::ATOMIC_LOAD_OR_I8:
   case Mips::ATOMIC_LOAD_OR_I8_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 1, Mips::OR);
+    return emitAtomicBinaryPartword(MI, BB, 1, Mips::OR);
   case Mips::ATOMIC_LOAD_OR_I16:
   case Mips::ATOMIC_LOAD_OR_I16_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 2, Mips::OR);
+    return emitAtomicBinaryPartword(MI, BB, 2, Mips::OR);
   case Mips::ATOMIC_LOAD_OR_I32:
   case Mips::ATOMIC_LOAD_OR_I32_P8:
-    return EmitAtomicBinary(MI, BB, 4, Mips::OR);
+    return emitAtomicBinary(MI, BB, 4, Mips::OR);
   case Mips::ATOMIC_LOAD_OR_I64:
   case Mips::ATOMIC_LOAD_OR_I64_P8:
-    return EmitAtomicBinary(MI, BB, 8, Mips::OR64);
+    return emitAtomicBinary(MI, BB, 8, Mips::OR64);
 
   case Mips::ATOMIC_LOAD_XOR_I8:
   case Mips::ATOMIC_LOAD_XOR_I8_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 1, Mips::XOR);
+    return emitAtomicBinaryPartword(MI, BB, 1, Mips::XOR);
   case Mips::ATOMIC_LOAD_XOR_I16:
   case Mips::ATOMIC_LOAD_XOR_I16_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 2, Mips::XOR);
+    return emitAtomicBinaryPartword(MI, BB, 2, Mips::XOR);
   case Mips::ATOMIC_LOAD_XOR_I32:
   case Mips::ATOMIC_LOAD_XOR_I32_P8:
-    return EmitAtomicBinary(MI, BB, 4, Mips::XOR);
+    return emitAtomicBinary(MI, BB, 4, Mips::XOR);
   case Mips::ATOMIC_LOAD_XOR_I64:
   case Mips::ATOMIC_LOAD_XOR_I64_P8:
-    return EmitAtomicBinary(MI, BB, 8, Mips::XOR64);
+    return emitAtomicBinary(MI, BB, 8, Mips::XOR64);
 
   case Mips::ATOMIC_LOAD_NAND_I8:
   case Mips::ATOMIC_LOAD_NAND_I8_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 1, 0, true);
+    return emitAtomicBinaryPartword(MI, BB, 1, 0, true);
   case Mips::ATOMIC_LOAD_NAND_I16:
   case Mips::ATOMIC_LOAD_NAND_I16_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 2, 0, true);
+    return emitAtomicBinaryPartword(MI, BB, 2, 0, true);
   case Mips::ATOMIC_LOAD_NAND_I32:
   case Mips::ATOMIC_LOAD_NAND_I32_P8:
-    return EmitAtomicBinary(MI, BB, 4, 0, true);
+    return emitAtomicBinary(MI, BB, 4, 0, true);
   case Mips::ATOMIC_LOAD_NAND_I64:
   case Mips::ATOMIC_LOAD_NAND_I64_P8:
-    return EmitAtomicBinary(MI, BB, 8, 0, true);
+    return emitAtomicBinary(MI, BB, 8, 0, true);
 
   case Mips::ATOMIC_LOAD_SUB_I8:
   case Mips::ATOMIC_LOAD_SUB_I8_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 1, Mips::SUBu);
+    return emitAtomicBinaryPartword(MI, BB, 1, Mips::SUBu);
   case Mips::ATOMIC_LOAD_SUB_I16:
   case Mips::ATOMIC_LOAD_SUB_I16_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 2, Mips::SUBu);
+    return emitAtomicBinaryPartword(MI, BB, 2, Mips::SUBu);
   case Mips::ATOMIC_LOAD_SUB_I32:
   case Mips::ATOMIC_LOAD_SUB_I32_P8:
-    return EmitAtomicBinary(MI, BB, 4, Mips::SUBu);
+    return emitAtomicBinary(MI, BB, 4, Mips::SUBu);
   case Mips::ATOMIC_LOAD_SUB_I64:
   case Mips::ATOMIC_LOAD_SUB_I64_P8:
-    return EmitAtomicBinary(MI, BB, 8, Mips::DSUBu);
+    return emitAtomicBinary(MI, BB, 8, Mips::DSUBu);
 
   case Mips::ATOMIC_SWAP_I8:
   case Mips::ATOMIC_SWAP_I8_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 1, 0);
+    return emitAtomicBinaryPartword(MI, BB, 1, 0);
   case Mips::ATOMIC_SWAP_I16:
   case Mips::ATOMIC_SWAP_I16_P8:
-    return EmitAtomicBinaryPartword(MI, BB, 2, 0);
+    return emitAtomicBinaryPartword(MI, BB, 2, 0);
   case Mips::ATOMIC_SWAP_I32:
   case Mips::ATOMIC_SWAP_I32_P8:
-    return EmitAtomicBinary(MI, BB, 4, 0);
+    return emitAtomicBinary(MI, BB, 4, 0);
   case Mips::ATOMIC_SWAP_I64:
   case Mips::ATOMIC_SWAP_I64_P8:
-    return EmitAtomicBinary(MI, BB, 8, 0);
+    return emitAtomicBinary(MI, BB, 8, 0);
 
   case Mips::ATOMIC_CMP_SWAP_I8:
   case Mips::ATOMIC_CMP_SWAP_I8_P8:
-    return EmitAtomicCmpSwapPartword(MI, BB, 1);
+    return emitAtomicCmpSwapPartword(MI, BB, 1);
   case Mips::ATOMIC_CMP_SWAP_I16:
   case Mips::ATOMIC_CMP_SWAP_I16_P8:
-    return EmitAtomicCmpSwapPartword(MI, BB, 2);
+    return emitAtomicCmpSwapPartword(MI, BB, 2);
   case Mips::ATOMIC_CMP_SWAP_I32:
   case Mips::ATOMIC_CMP_SWAP_I32_P8:
-    return EmitAtomicCmpSwap(MI, BB, 4);
+    return emitAtomicCmpSwap(MI, BB, 4);
   case Mips::ATOMIC_CMP_SWAP_I64:
   case Mips::ATOMIC_CMP_SWAP_I64_P8:
-    return EmitAtomicCmpSwap(MI, BB, 8);
-  case Mips::BPOSGE32_PSEUDO:
-    return EmitBPOSGE32(MI, BB);
+    return emitAtomicCmpSwap(MI, BB, 8);
   }
 }
 
 // This function also handles Mips::ATOMIC_SWAP_I32 (when BinOpcode == 0), and
 // Mips::ATOMIC_LOAD_NAND_I32 (when Nand == true)
 MachineBasicBlock *
-MipsTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB,
+MipsTargetLowering::emitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB,
                                      unsigned Size, unsigned BinOpcode,
                                      bool Nand) const {
   assert((Size == 4 || Size == 8) && "Unsupported size for EmitAtomicBinary.");
@@ -1285,7 +892,7 @@ MipsTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB,
   MachineRegisterInfo &RegInfo = MF->getRegInfo();
   const TargetRegisterClass *RC = getRegClassFor(MVT::getIntegerVT(Size * 8));
   const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
-  DebugLoc dl = MI->getDebugLoc();
+  DebugLoc DL = MI->getDebugLoc();
   unsigned LL, SC, AND, NOR, ZERO, BEQ;
 
   if (Size == 4) {
@@ -1341,20 +948,20 @@ MipsTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB,
   //    sc success, storeval, 0(ptr)
   //    beq success, $0, loopMBB
   BB = loopMBB;
-  BuildMI(BB, dl, TII->get(LL), OldVal).addReg(Ptr).addImm(0);
+  BuildMI(BB, DL, TII->get(LL), OldVal).addReg(Ptr).addImm(0);
   if (Nand) {
     //  and andres, oldval, incr
     //  nor storeval, $0, andres
-    BuildMI(BB, dl, TII->get(AND), AndRes).addReg(OldVal).addReg(Incr);
-    BuildMI(BB, dl, TII->get(NOR), StoreVal).addReg(ZERO).addReg(AndRes);
+    BuildMI(BB, DL, TII->get(AND), AndRes).addReg(OldVal).addReg(Incr);
+    BuildMI(BB, DL, TII->get(NOR), StoreVal).addReg(ZERO).addReg(AndRes);
   } else if (BinOpcode) {
     //  <binop> storeval, oldval, incr
-    BuildMI(BB, dl, TII->get(BinOpcode), StoreVal).addReg(OldVal).addReg(Incr);
+    BuildMI(BB, DL, TII->get(BinOpcode), StoreVal).addReg(OldVal).addReg(Incr);
   } else {
     StoreVal = Incr;
   }
-  BuildMI(BB, dl, TII->get(SC), Success).addReg(StoreVal).addReg(Ptr).addImm(0);
-  BuildMI(BB, dl, TII->get(BEQ)).addReg(Success).addReg(ZERO).addMBB(loopMBB);
+  BuildMI(BB, DL, TII->get(SC), Success).addReg(StoreVal).addReg(Ptr).addImm(0);
+  BuildMI(BB, DL, TII->get(BEQ)).addReg(Success).addReg(ZERO).addMBB(loopMBB);
 
   MI->eraseFromParent();   // The instruction is gone now.
 
@@ -1362,7 +969,7 @@ MipsTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB,
 }
 
 MachineBasicBlock *
-MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI,
+MipsTargetLowering::emitAtomicBinaryPartword(MachineInstr *MI,
                                              MachineBasicBlock *BB,
                                              unsigned Size, unsigned BinOpcode,
                                              bool Nand) const {
@@ -1373,7 +980,7 @@ MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI,
   MachineRegisterInfo &RegInfo = MF->getRegInfo();
   const TargetRegisterClass *RC = getRegClassFor(MVT::i32);
   const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
-  DebugLoc dl = MI->getDebugLoc();
+  DebugLoc DL = MI->getDebugLoc();
   unsigned LL = IsN64 ? Mips::LL_P8 : Mips::LL;
   unsigned SC = IsN64 ? Mips::SC_P8 : Mips::SC;
 
@@ -1432,18 +1039,18 @@ MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI,
   //    sll     incr2,incr,shiftamt
 
   int64_t MaskImm = (Size == 1) ? 255 : 65535;
-  BuildMI(BB, dl, TII->get(Mips::ADDiu), MaskLSB2)
+  BuildMI(BB, DL, TII->get(Mips::ADDiu), MaskLSB2)
     .addReg(Mips::ZERO).addImm(-4);
-  BuildMI(BB, dl, TII->get(Mips::AND), AlignedAddr)
+  BuildMI(BB, DL, TII->get(Mips::AND), AlignedAddr)
     .addReg(Ptr).addReg(MaskLSB2);
-  BuildMI(BB, dl, TII->get(Mips::ANDi), PtrLSB2).addReg(Ptr).addImm(3);
-  BuildMI(BB, dl, TII->get(Mips::SLL), ShiftAmt).addReg(PtrLSB2).addImm(3);
-  BuildMI(BB, dl, TII->get(Mips::ORi), MaskUpper)
+  BuildMI(BB, DL, TII->get(Mips::ANDi), PtrLSB2).addReg(Ptr).addImm(3);
+  BuildMI(BB, DL, TII->get(Mips::SLL), ShiftAmt).addReg(PtrLSB2).addImm(3);
+  BuildMI(BB, DL, TII->get(Mips::ORi), MaskUpper)
     .addReg(Mips::ZERO).addImm(MaskImm);
-  BuildMI(BB, dl, TII->get(Mips::SLLV), Mask)
+  BuildMI(BB, DL, TII->get(Mips::SLLV), Mask)
     .addReg(ShiftAmt).addReg(MaskUpper);
-  BuildMI(BB, dl, TII->get(Mips::NOR), Mask2).addReg(Mips::ZERO).addReg(Mask);
-  BuildMI(BB, dl, TII->get(Mips::SLLV), Incr2).addReg(ShiftAmt).addReg(Incr);
+  BuildMI(BB, DL, TII->get(Mips::NOR), Mask2).addReg(Mips::ZERO).addReg(Mask);
+  BuildMI(BB, DL, TII->get(Mips::SLLV), Incr2).addReg(ShiftAmt).addReg(Incr);
 
   // atomic.load.binop
   // loopMBB:
@@ -1465,32 +1072,32 @@ MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI,
   //   beq     success,$0,loopMBB
 
   BB = loopMBB;
-  BuildMI(BB, dl, TII->get(LL), OldVal).addReg(AlignedAddr).addImm(0);
+  BuildMI(BB, DL, TII->get(LL), OldVal).addReg(AlignedAddr).addImm(0);
   if (Nand) {
     //  and andres, oldval, incr2
     //  nor binopres, $0, andres
     //  and newval, binopres, mask
-    BuildMI(BB, dl, TII->get(Mips::AND), AndRes).addReg(OldVal).addReg(Incr2);
-    BuildMI(BB, dl, TII->get(Mips::NOR), BinOpRes)
+    BuildMI(BB, DL, TII->get(Mips::AND), AndRes).addReg(OldVal).addReg(Incr2);
+    BuildMI(BB, DL, TII->get(Mips::NOR), BinOpRes)
       .addReg(Mips::ZERO).addReg(AndRes);
-    BuildMI(BB, dl, TII->get(Mips::AND), NewVal).addReg(BinOpRes).addReg(Mask);
+    BuildMI(BB, DL, TII->get(Mips::AND), NewVal).addReg(BinOpRes).addReg(Mask);
   } else if (BinOpcode) {
     //  <binop> binopres, oldval, incr2
     //  and newval, binopres, mask
-    BuildMI(BB, dl, TII->get(BinOpcode), BinOpRes).addReg(OldVal).addReg(Incr2);
-    BuildMI(BB, dl, TII->get(Mips::AND), NewVal).addReg(BinOpRes).addReg(Mask);
+    BuildMI(BB, DL, TII->get(BinOpcode), BinOpRes).addReg(OldVal).addReg(Incr2);
+    BuildMI(BB, DL, TII->get(Mips::AND), NewVal).addReg(BinOpRes).addReg(Mask);
   } else {// atomic.swap
     //  and newval, incr2, mask
-    BuildMI(BB, dl, TII->get(Mips::AND), NewVal).addReg(Incr2).addReg(Mask);
+    BuildMI(BB, DL, TII->get(Mips::AND), NewVal).addReg(Incr2).addReg(Mask);
   }
 
-  BuildMI(BB, dl, TII->get(Mips::AND), MaskedOldVal0)
+  BuildMI(BB, DL, TII->get(Mips::AND), MaskedOldVal0)
     .addReg(OldVal).addReg(Mask2);
-  BuildMI(BB, dl, TII->get(Mips::OR), StoreVal)
+  BuildMI(BB, DL, TII->get(Mips::OR), StoreVal)
     .addReg(MaskedOldVal0).addReg(NewVal);
-  BuildMI(BB, dl, TII->get(SC), Success)
+  BuildMI(BB, DL, TII->get(SC), Success)
     .addReg(StoreVal).addReg(AlignedAddr).addImm(0);
-  BuildMI(BB, dl, TII->get(Mips::BEQ))
+  BuildMI(BB, DL, TII->get(Mips::BEQ))
     .addReg(Success).addReg(Mips::ZERO).addMBB(loopMBB);
 
   //  sinkMBB:
@@ -1501,13 +1108,13 @@ MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI,
   BB = sinkMBB;
   int64_t ShiftImm = (Size == 1) ? 24 : 16;
 
-  BuildMI(BB, dl, TII->get(Mips::AND), MaskedOldVal1)
+  BuildMI(BB, DL, TII->get(Mips::AND), MaskedOldVal1)
     .addReg(OldVal).addReg(Mask);
-  BuildMI(BB, dl, TII->get(Mips::SRLV), SrlRes)
+  BuildMI(BB, DL, TII->get(Mips::SRLV), SrlRes)
       .addReg(ShiftAmt).addReg(MaskedOldVal1);
-  BuildMI(BB, dl, TII->get(Mips::SLL), SllRes)
+  BuildMI(BB, DL, TII->get(Mips::SLL), SllRes)
       .addReg(SrlRes).addImm(ShiftImm);
-  BuildMI(BB, dl, TII->get(Mips::SRA), Dest)
+  BuildMI(BB, DL, TII->get(Mips::SRA), Dest)
       .addReg(SllRes).addImm(ShiftImm);
 
   MI->eraseFromParent();   // The instruction is gone now.
@@ -1516,7 +1123,7 @@ MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI,
 }
 
 MachineBasicBlock *
-MipsTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI,
+MipsTargetLowering::emitAtomicCmpSwap(MachineInstr *MI,
                                       MachineBasicBlock *BB,
                                       unsigned Size) const {
   assert((Size == 4 || Size == 8) && "Unsupported size for EmitAtomicCmpSwap.");
@@ -1525,7 +1132,7 @@ MipsTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI,
   MachineRegisterInfo &RegInfo = MF->getRegInfo();
   const TargetRegisterClass *RC = getRegClassFor(MVT::getIntegerVT(Size * 8));
   const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
-  DebugLoc dl = MI->getDebugLoc();
+  DebugLoc DL = MI->getDebugLoc();
   unsigned LL, SC, ZERO, BNE, BEQ;
 
   if (Size == 4) {
@@ -1579,17 +1186,17 @@ MipsTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI,
   //   ll dest, 0(ptr)
   //   bne dest, oldval, exitMBB
   BB = loop1MBB;
-  BuildMI(BB, dl, TII->get(LL), Dest).addReg(Ptr).addImm(0);
-  BuildMI(BB, dl, TII->get(BNE))
+  BuildMI(BB, DL, TII->get(LL), Dest).addReg(Ptr).addImm(0);
+  BuildMI(BB, DL, TII->get(BNE))
     .addReg(Dest).addReg(OldVal).addMBB(exitMBB);
 
   // loop2MBB:
   //   sc success, newval, 0(ptr)
   //   beq success, $0, loop1MBB
   BB = loop2MBB;
-  BuildMI(BB, dl, TII->get(SC), Success)
+  BuildMI(BB, DL, TII->get(SC), Success)
     .addReg(NewVal).addReg(Ptr).addImm(0);
-  BuildMI(BB, dl, TII->get(BEQ))
+  BuildMI(BB, DL, TII->get(BEQ))
     .addReg(Success).addReg(ZERO).addMBB(loop1MBB);
 
   MI->eraseFromParent();   // The instruction is gone now.
@@ -1598,7 +1205,7 @@ MipsTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI,
 }
 
 MachineBasicBlock *
-MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI,
+MipsTargetLowering::emitAtomicCmpSwapPartword(MachineInstr *MI,
                                               MachineBasicBlock *BB,
                                               unsigned Size) const {
   assert((Size == 1 || Size == 2) &&
@@ -1608,7 +1215,7 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI,
   MachineRegisterInfo &RegInfo = MF->getRegInfo();
   const TargetRegisterClass *RC = getRegClassFor(MVT::i32);
   const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
-  DebugLoc dl = MI->getDebugLoc();
+  DebugLoc DL = MI->getDebugLoc();
   unsigned LL = IsN64 ? Mips::LL_P8 : Mips::LL;
   unsigned SC = IsN64 ? Mips::SC_P8 : Mips::SC;
 
@@ -1675,24 +1282,24 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI,
   //    andi    maskednewval,newval,255
   //    sll     shiftednewval,maskednewval,shiftamt
   int64_t MaskImm = (Size == 1) ? 255 : 65535;
-  BuildMI(BB, dl, TII->get(Mips::ADDiu), MaskLSB2)
+  BuildMI(BB, DL, TII->get(Mips::ADDiu), MaskLSB2)
     .addReg(Mips::ZERO).addImm(-4);
-  BuildMI(BB, dl, TII->get(Mips::AND), AlignedAddr)
+  BuildMI(BB, DL, TII->get(Mips::AND), AlignedAddr)
     .addReg(Ptr).addReg(MaskLSB2);
-  BuildMI(BB, dl, TII->get(Mips::ANDi), PtrLSB2).addReg(Ptr).addImm(3);
-  BuildMI(BB, dl, TII->get(Mips::SLL), ShiftAmt).addReg(PtrLSB2).addImm(3);
-  BuildMI(BB, dl, TII->get(Mips::ORi), MaskUpper)
+  BuildMI(BB, DL, TII->get(Mips::ANDi), PtrLSB2).addReg(Ptr).addImm(3);
+  BuildMI(BB, DL, TII->get(Mips::SLL), ShiftAmt).addReg(PtrLSB2).addImm(3);
+  BuildMI(BB, DL, TII->get(Mips::ORi), MaskUpper)
     .addReg(Mips::ZERO).addImm(MaskImm);
-  BuildMI(BB, dl, TII->get(Mips::SLLV), Mask)
+  BuildMI(BB, DL, TII->get(Mips::SLLV), Mask)
     .addReg(ShiftAmt).addReg(MaskUpper);
-  BuildMI(BB, dl, TII->get(Mips::NOR), Mask2).addReg(Mips::ZERO).addReg(Mask);
-  BuildMI(BB, dl, TII->get(Mips::ANDi), MaskedCmpVal)
+  BuildMI(BB, DL, TII->get(Mips::NOR), Mask2).addReg(Mips::ZERO).addReg(Mask);
+  BuildMI(BB, DL, TII->get(Mips::ANDi), MaskedCmpVal)
     .addReg(CmpVal).addImm(MaskImm);
-  BuildMI(BB, dl, TII->get(Mips::SLLV), ShiftedCmpVal)
+  BuildMI(BB, DL, TII->get(Mips::SLLV), ShiftedCmpVal)
     .addReg(ShiftAmt).addReg(MaskedCmpVal);
-  BuildMI(BB, dl, TII->get(Mips::ANDi), MaskedNewVal)
+  BuildMI(BB, DL, TII->get(Mips::ANDi), MaskedNewVal)
     .addReg(NewVal).addImm(MaskImm);
-  BuildMI(BB, dl, TII->get(Mips::SLLV), ShiftedNewVal)
+  BuildMI(BB, DL, TII->get(Mips::SLLV), ShiftedNewVal)
     .addReg(ShiftAmt).addReg(MaskedNewVal);
 
   //  loop1MBB:
@@ -1700,10 +1307,10 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI,
   //    and     maskedoldval0,oldval,mask
   //    bne     maskedoldval0,shiftedcmpval,sinkMBB
   BB = loop1MBB;
-  BuildMI(BB, dl, TII->get(LL), OldVal).addReg(AlignedAddr).addImm(0);
-  BuildMI(BB, dl, TII->get(Mips::AND), MaskedOldVal0)
+  BuildMI(BB, DL, TII->get(LL), OldVal).addReg(AlignedAddr).addImm(0);
+  BuildMI(BB, DL, TII->get(Mips::AND), MaskedOldVal0)
     .addReg(OldVal).addReg(Mask);
-  BuildMI(BB, dl, TII->get(Mips::BNE))
+  BuildMI(BB, DL, TII->get(Mips::BNE))
     .addReg(MaskedOldVal0).addReg(ShiftedCmpVal).addMBB(sinkMBB);
 
   //  loop2MBB:
@@ -1712,13 +1319,13 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI,
   //    sc      success,storeval,0(alignedaddr)
   //    beq     success,$0,loop1MBB
   BB = loop2MBB;
-  BuildMI(BB, dl, TII->get(Mips::AND), MaskedOldVal1)
+  BuildMI(BB, DL, TII->get(Mips::AND), MaskedOldVal1)
     .addReg(OldVal).addReg(Mask2);
-  BuildMI(BB, dl, TII->get(Mips::OR), StoreVal)
+  BuildMI(BB, DL, TII->get(Mips::OR), StoreVal)
     .addReg(MaskedOldVal1).addReg(ShiftedNewVal);
-  BuildMI(BB, dl, TII->get(SC), Success)
+  BuildMI(BB, DL, TII->get(SC), Success)
       .addReg(StoreVal).addReg(AlignedAddr).addImm(0);
-  BuildMI(BB, dl, TII->get(Mips::BEQ))
+  BuildMI(BB, DL, TII->get(Mips::BEQ))
       .addReg(Success).addReg(Mips::ZERO).addMBB(loop1MBB);
 
   //  sinkMBB:
@@ -1728,11 +1335,11 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI,
   BB = sinkMBB;
   int64_t ShiftImm = (Size == 1) ? 24 : 16;
 
-  BuildMI(BB, dl, TII->get(Mips::SRLV), SrlRes)
+  BuildMI(BB, DL, TII->get(Mips::SRLV), SrlRes)
       .addReg(ShiftAmt).addReg(MaskedOldVal0);
-  BuildMI(BB, dl, TII->get(Mips::SLL), SllRes)
+  BuildMI(BB, DL, TII->get(Mips::SLL), SllRes)
       .addReg(SrlRes).addImm(ShiftImm);
-  BuildMI(BB, dl, TII->get(Mips::SRA), Dest)
+  BuildMI(BB, DL, TII->get(Mips::SRA), Dest)
       .addReg(SllRes).addImm(ShiftImm);
 
   MI->eraseFromParent();   // The instruction is gone now.
@@ -1743,16 +1350,46 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI,
 //===----------------------------------------------------------------------===//
 //  Misc Lower Operation implementation
 //===----------------------------------------------------------------------===//
+SDValue MipsTargetLowering::lowerBR_JT(SDValue Op, SelectionDAG &DAG) const {
+  SDValue Chain = Op.getOperand(0);
+  SDValue Table = Op.getOperand(1);
+  SDValue Index = Op.getOperand(2);
+  DebugLoc DL = Op.getDebugLoc();
+  EVT PTy = getPointerTy();
+  unsigned EntrySize =
+    DAG.getMachineFunction().getJumpTableInfo()->getEntrySize(*getDataLayout());
+
+  Index = DAG.getNode(ISD::MUL, DL, PTy, Index,
+                      DAG.getConstant(EntrySize, PTy));
+  SDValue Addr = DAG.getNode(ISD::ADD, DL, PTy, Index, Table);
+
+  EVT MemVT = EVT::getIntegerVT(*DAG.getContext(), EntrySize * 8);
+  Addr = DAG.getExtLoad(ISD::SEXTLOAD, DL, PTy, Chain, Addr,
+                        MachinePointerInfo::getJumpTable(), MemVT, false, false,
+                        0);
+  Chain = Addr.getValue(1);
+
+  if ((getTargetMachine().getRelocationModel() == Reloc::PIC_) || IsN64) {
+    // For PIC, the sequence is:
+    // BRIND(load(Jumptable + index) + RelocBase)
+    // RelocBase can be JumpTable, GOT or some sort of global base.
+    Addr = DAG.getNode(ISD::ADD, DL, PTy, Addr,
+                       getPICJumpTableRelocBase(Table, DAG));
+  }
+
+  return DAG.getNode(ISD::BRIND, DL, MVT::Other, Chain, Addr);
+}
+
 SDValue MipsTargetLowering::
-LowerBRCOND(SDValue Op, SelectionDAG &DAG) const
+lowerBRCOND(SDValue Op, SelectionDAG &DAG) const
 {
   // The first operand is the chain, the second is the condition, the third is
   // the block to branch to if the condition is true.
   SDValue Chain = Op.getOperand(0);
   SDValue Dest = Op.getOperand(2);
-  DebugLoc dl = Op.getDebugLoc();
+  DebugLoc DL = Op.getDebugLoc();
 
-  SDValue CondRes = CreateFPCmp(DAG, Op.getOperand(1));
+  SDValue CondRes = createFPCmp(DAG, Op.getOperand(1));
 
   // Return if flag is not set by a floating point comparison.
   if (CondRes.getOpcode() != MipsISD::FPCmp)
@@ -1761,27 +1398,27 @@ LowerBRCOND(SDValue Op, SelectionDAG &DAG) const
   SDValue CCNode  = CondRes.getOperand(2);
   Mips::CondCode CC =
     (Mips::CondCode)cast<ConstantSDNode>(CCNode)->getZExtValue();
-  SDValue BrCode = DAG.getConstant(GetFPBranchCodeFromCond(CC), MVT::i32);
-
-  return DAG.getNode(MipsISD::FPBrcond, dl, Op.getValueType(), Chain, BrCode,
+  unsigned Opc = invertFPCondCodeUser(CC) ? Mips::BRANCH_F : Mips::BRANCH_T;
+  SDValue BrCode = DAG.getConstant(Opc, MVT::i32);
+  return DAG.getNode(MipsISD::FPBrcond, DL, Op.getValueType(), Chain, BrCode,
                      Dest, CondRes);
 }
 
 SDValue MipsTargetLowering::
-LowerSELECT(SDValue Op, SelectionDAG &DAG) const
+lowerSELECT(SDValue Op, SelectionDAG &DAG) const
 {
-  SDValue Cond = CreateFPCmp(DAG, Op.getOperand(0));
+  SDValue Cond = createFPCmp(DAG, Op.getOperand(0));
 
   // Return if flag is not set by a floating point comparison.
   if (Cond.getOpcode() != MipsISD::FPCmp)
     return Op;
 
-  return CreateCMovFP(DAG, Cond, Op.getOperand(1), Op.getOperand(2),
+  return createCMovFP(DAG, Cond, Op.getOperand(1), Op.getOperand(2),
                       Op.getDebugLoc());
 }
 
 SDValue MipsTargetLowering::
-LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const
+lowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const
 {
   DebugLoc DL = Op.getDebugLoc();
   EVT Ty = Op.getOperand(0).getValueType();
@@ -1793,8 +1430,8 @@ LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const
                      Op.getOperand(3));
 }
 
-SDValue MipsTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
-  SDValue Cond = CreateFPCmp(DAG, Op);
+SDValue MipsTargetLowering::lowerSETCC(SDValue Op, SelectionDAG &DAG) const {
+  SDValue Cond = createFPCmp(DAG, Op);
 
   assert(Cond.getOpcode() == MipsISD::FPCmp &&
          "Floating point operand expected.");
@@ -1802,13 +1439,13 @@ SDValue MipsTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
   SDValue True  = DAG.getConstant(1, MVT::i32);
   SDValue False = DAG.getConstant(0, MVT::i32);
 
-  return CreateCMovFP(DAG, Cond, True, False, Op.getDebugLoc());
+  return createCMovFP(DAG, Cond, True, False, Op.getDebugLoc());
 }
 
-SDValue MipsTargetLowering::LowerGlobalAddress(SDValue Op,
+SDValue MipsTargetLowering::lowerGlobalAddress(SDValue Op,
                                                SelectionDAG &DAG) const {
   // FIXME there isn't actually debug info here
-  DebugLoc dl = Op.getDebugLoc();
+  DebugLoc DL = Op.getDebugLoc();
   const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
 
   if (getTargetMachine().getRelocationModel() != Reloc::PIC_ && !IsN64) {
@@ -1817,12 +1454,12 @@ SDValue MipsTargetLowering::LowerGlobalAddress(SDValue Op,
 
     // %gp_rel relocation
     if (TLOF.IsGlobalInSmallSection(GV, getTargetMachine())) {
-      SDValue GA = DAG.getTargetGlobalAddress(GV, dl, MVT::i32, 0,
+      SDValue GA = DAG.getTargetGlobalAddress(GV, DL, MVT::i32, 0,
                                               MipsII::MO_GPREL);
-      SDValue GPRelNode = DAG.getNode(MipsISD::GPRel, dl,
+      SDValue GPRelNode = DAG.getNode(MipsISD::GPRel, DL,
                                       DAG.getVTList(MVT::i32), &GA, 1);
       SDValue GPReg = DAG.getRegister(Mips::GP, MVT::i32);
-      return DAG.getNode(ISD::ADD, dl, MVT::i32, GPReg, GPRelNode);
+      return DAG.getNode(ISD::ADD, DL, MVT::i32, GPReg, GPRelNode);
     }
 
     // %hi/%lo relocation
@@ -1840,7 +1477,7 @@ SDValue MipsTargetLowering::LowerGlobalAddress(SDValue Op,
                        HasMips64 ? MipsII::MO_GOT_DISP : MipsII::MO_GOT16);
 }
 
-SDValue MipsTargetLowering::LowerBlockAddress(SDValue Op,
+SDValue MipsTargetLowering::lowerBlockAddress(SDValue Op,
                                               SelectionDAG &DAG) const {
   if (getTargetMachine().getRelocationModel() != Reloc::PIC_ && !IsN64)
     return getAddrNonPIC(Op, DAG);
@@ -1849,14 +1486,14 @@ SDValue MipsTargetLowering::LowerBlockAddress(SDValue Op,
 }
 
 SDValue MipsTargetLowering::
-LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const
+lowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const
 {
   // If the relocation model is PIC, use the General Dynamic TLS Model or
   // Local Dynamic TLS model, otherwise use the Initial Exec or
   // Local Exec TLS Model.
 
   GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);
-  DebugLoc dl = GA->getDebugLoc();
+  DebugLoc DL = GA->getDebugLoc();
   const GlobalValue *GV = GA->getGlobal();
   EVT PtrVT = getPointerTy();
 
@@ -1867,9 +1504,9 @@ LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const
     unsigned Flag = (model == TLSModel::LocalDynamic) ? MipsII::MO_TLSLDM
                                                       : MipsII::MO_TLSGD;
 
-    SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, Flag);
-    SDValue Argument = DAG.getNode(MipsISD::Wrapper, dl, PtrVT,
-                                   GetGlobalReg(DAG, PtrVT), TGA);
+    SDValue TGA = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, Flag);
+    SDValue Argument = DAG.getNode(MipsISD::Wrapper, DL, PtrVT,
+                                   getGlobalReg(DAG, PtrVT), TGA);
     unsigned PtrSize = PtrVT.getSizeInBits();
     IntegerType *PtrTy = Type::getIntNTy(*DAG.getContext(), PtrSize);
 
@@ -1883,9 +1520,9 @@ LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const
 
     TargetLowering::CallLoweringInfo CLI(DAG.getEntryNode(), PtrTy,
                   false, false, false, false, 0, CallingConv::C,
-                  /*isTailCall=*/false, /*doesNotRet=*/false,
+                  /*IsTailCall=*/false, /*doesNotRet=*/false,
                   /*isReturnValueUsed=*/true,
-                  TlsGetAddr, Args, DAG, dl);
+                  TlsGetAddr, Args, DAG, DL);
     std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
 
     SDValue Ret = CallResult.first;
@@ -1893,44 +1530,44 @@ LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const
     if (model != TLSModel::LocalDynamic)
       return Ret;
 
-    SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+    SDValue TGAHi = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
                                                MipsII::MO_DTPREL_HI);
-    SDValue Hi = DAG.getNode(MipsISD::Hi, dl, PtrVT, TGAHi);
-    SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+    SDValue Hi = DAG.getNode(MipsISD::Hi, DL, PtrVT, TGAHi);
+    SDValue TGALo = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
                                                MipsII::MO_DTPREL_LO);
-    SDValue Lo = DAG.getNode(MipsISD::Lo, dl, PtrVT, TGALo);
-    SDValue Add = DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Ret);
-    return DAG.getNode(ISD::ADD, dl, PtrVT, Add, Lo);
+    SDValue Lo = DAG.getNode(MipsISD::Lo, DL, PtrVT, TGALo);
+    SDValue Add = DAG.getNode(ISD::ADD, DL, PtrVT, Hi, Ret);
+    return DAG.getNode(ISD::ADD, DL, PtrVT, Add, Lo);
   }
 
   SDValue Offset;
   if (model == TLSModel::InitialExec) {
     // Initial Exec TLS Model
-    SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+    SDValue TGA = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
                                              MipsII::MO_GOTTPREL);
-    TGA = DAG.getNode(MipsISD::Wrapper, dl, PtrVT, GetGlobalReg(DAG, PtrVT),
+    TGA = DAG.getNode(MipsISD::Wrapper, DL, PtrVT, getGlobalReg(DAG, PtrVT),
                       TGA);
-    Offset = DAG.getLoad(PtrVT, dl,
+    Offset = DAG.getLoad(PtrVT, DL,
                          DAG.getEntryNode(), TGA, MachinePointerInfo(),
                          false, false, false, 0);
   } else {
     // Local Exec TLS Model
     assert(model == TLSModel::LocalExec);
-    SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+    SDValue TGAHi = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
                                                MipsII::MO_TPREL_HI);
-    SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+    SDValue TGALo = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
                                                MipsII::MO_TPREL_LO);
-    SDValue Hi = DAG.getNode(MipsISD::Hi, dl, PtrVT, TGAHi);
-    SDValue Lo = DAG.getNode(MipsISD::Lo, dl, PtrVT, TGALo);
-    Offset = DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Lo);
+    SDValue Hi = DAG.getNode(MipsISD::Hi, DL, PtrVT, TGAHi);
+    SDValue Lo = DAG.getNode(MipsISD::Lo, DL, PtrVT, TGALo);
+    Offset = DAG.getNode(ISD::ADD, DL, PtrVT, Hi, Lo);
   }
 
-  SDValue ThreadPointer = DAG.getNode(MipsISD::ThreadPointer, dl, PtrVT);
-  return DAG.getNode(ISD::ADD, dl, PtrVT, ThreadPointer, Offset);
+  SDValue ThreadPointer = DAG.getNode(MipsISD::ThreadPointer, DL, PtrVT);
+  return DAG.getNode(ISD::ADD, DL, PtrVT, ThreadPointer, Offset);
 }
 
 SDValue MipsTargetLowering::
-LowerJumpTable(SDValue Op, SelectionDAG &DAG) const
+lowerJumpTable(SDValue Op, SelectionDAG &DAG) const
 {
   if (getTargetMachine().getRelocationModel() != Reloc::PIC_ && !IsN64)
     return getAddrNonPIC(Op, DAG);
@@ -1939,7 +1576,7 @@ LowerJumpTable(SDValue Op, SelectionDAG &DAG) const
 }
 
 SDValue MipsTargetLowering::
-LowerConstantPool(SDValue Op, SelectionDAG &DAG) const
+lowerConstantPool(SDValue Op, SelectionDAG &DAG) const
 {
   // gp_rel relocation
   // FIXME: we should reference the constant pool using small data sections,
@@ -1957,22 +1594,22 @@ LowerConstantPool(SDValue Op, SelectionDAG &DAG) const
   return getAddrLocal(Op, DAG, HasMips64);
 }
 
-SDValue MipsTargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG) const {
+SDValue MipsTargetLowering::lowerVASTART(SDValue Op, SelectionDAG &DAG) const {
   MachineFunction &MF = DAG.getMachineFunction();
   MipsFunctionInfo *FuncInfo = MF.getInfo<MipsFunctionInfo>();
 
-  DebugLoc dl = Op.getDebugLoc();
+  DebugLoc DL = Op.getDebugLoc();
   SDValue FI = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(),
                                  getPointerTy());
 
   // vastart just stores the address of the VarArgsFrameIndex slot into the
   // memory location argument.
   const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
-  return DAG.getStore(Op.getOperand(0), dl, FI, Op.getOperand(1),
+  return DAG.getStore(Op.getOperand(0), DL, FI, Op.getOperand(1),
                       MachinePointerInfo(SV), false, false, 0);
 }
 
-static SDValue LowerFCOPYSIGN32(SDValue Op, SelectionDAG &DAG, bool HasR2) {
+static SDValue lowerFCOPYSIGN32(SDValue Op, SelectionDAG &DAG, bool HasR2) {
   EVT TyX = Op.getOperand(0).getValueType();
   EVT TyY = Op.getOperand(1).getValueType();
   SDValue Const1 = DAG.getConstant(1, MVT::i32);
@@ -2017,7 +1654,7 @@ static SDValue LowerFCOPYSIGN32(SDValue Op, SelectionDAG &DAG, bool HasR2) {
   return DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64, LowX, Res);
 }
 
-static SDValue LowerFCOPYSIGN64(SDValue Op, SelectionDAG &DAG, bool HasR2) {
+static SDValue lowerFCOPYSIGN64(SDValue Op, SelectionDAG &DAG, bool HasR2) {
   unsigned WidthX = Op.getOperand(0).getValueSizeInBits();
   unsigned WidthY = Op.getOperand(1).getValueSizeInBits();
   EVT TyX = MVT::getIntegerVT(WidthX), TyY = MVT::getIntegerVT(WidthY);
@@ -2066,14 +1703,14 @@ static SDValue LowerFCOPYSIGN64(SDValue Op, SelectionDAG &DAG, bool HasR2) {
 }
 
 SDValue
-MipsTargetLowering::LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const {
+MipsTargetLowering::lowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const {
   if (Subtarget->hasMips64())
-    return LowerFCOPYSIGN64(Op, DAG, Subtarget->hasMips32r2());
+    return lowerFCOPYSIGN64(Op, DAG, Subtarget->hasMips32r2());
 
-  return LowerFCOPYSIGN32(Op, DAG, Subtarget->hasMips32r2());
+  return lowerFCOPYSIGN32(Op, DAG, Subtarget->hasMips32r2());
 }
 
-static SDValue LowerFABS32(SDValue Op, SelectionDAG &DAG, bool HasR2) {
+static SDValue lowerFABS32(SDValue Op, SelectionDAG &DAG, bool HasR2) {
   SDValue Res, Const1 = DAG.getConstant(1, MVT::i32);
   DebugLoc DL = Op.getDebugLoc();
 
@@ -2102,7 +1739,7 @@ static SDValue LowerFABS32(SDValue Op, SelectionDAG &DAG, bool HasR2) {
   return DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64, LowX, Res);
 }
 
-static SDValue LowerFABS64(SDValue Op, SelectionDAG &DAG, bool HasR2) {
+static SDValue lowerFABS64(SDValue Op, SelectionDAG &DAG, bool HasR2) {
   SDValue Res, Const1 = DAG.getConstant(1, MVT::i32);
   DebugLoc DL = Op.getDebugLoc();
 
@@ -2123,15 +1760,15 @@ static SDValue LowerFABS64(SDValue Op, SelectionDAG &DAG, bool HasR2) {
 }
 
 SDValue
-MipsTargetLowering::LowerFABS(SDValue Op, SelectionDAG &DAG) const {
+MipsTargetLowering::lowerFABS(SDValue Op, SelectionDAG &DAG) const {
   if (Subtarget->hasMips64() && (Op.getValueType() == MVT::f64))
-    return LowerFABS64(Op, DAG, Subtarget->hasMips32r2());
+    return lowerFABS64(Op, DAG, Subtarget->hasMips32r2());
 
-  return LowerFABS32(Op, DAG, Subtarget->hasMips32r2());
+  return lowerFABS32(Op, DAG, Subtarget->hasMips32r2());
 }
 
 SDValue MipsTargetLowering::
-LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const {
+lowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const {
   // check the depth
   assert((cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue() == 0) &&
          "Frame address can only be determined for current frame.");
@@ -2139,13 +1776,13 @@ LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const {
   MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
   MFI->setFrameAddressIsTaken(true);
   EVT VT = Op.getValueType();
-  DebugLoc dl = Op.getDebugLoc();
-  SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl,
+  DebugLoc DL = Op.getDebugLoc();
+  SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), DL,
                                          IsN64 ? Mips::FP_64 : Mips::FP, VT);
   return FrameAddr;
 }
 
-SDValue MipsTargetLowering::LowerRETURNADDR(SDValue Op,
+SDValue MipsTargetLowering::lowerRETURNADDR(SDValue Op,
                                             SelectionDAG &DAG) const {
   // check the depth
   assert((cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue() == 0) &&
@@ -2153,7 +1790,7 @@ SDValue MipsTargetLowering::LowerRETURNADDR(SDValue Op,
 
   MachineFunction &MF = DAG.getMachineFunction();
   MachineFrameInfo *MFI = MF.getFrameInfo();
-  EVT VT = Op.getValueType();
+  MVT VT = Op.getSimpleValueType();
   unsigned RA = IsN64 ? Mips::RA_64 : Mips::RA;
   MFI->setReturnAddressIsTaken(true);
 
@@ -2162,26 +1799,54 @@ SDValue MipsTargetLowering::LowerRETURNADDR(SDValue Op,
   return DAG.getCopyFromReg(DAG.getEntryNode(), Op.getDebugLoc(), Reg, VT);
 }
 
+// An EH_RETURN is the result of lowering llvm.eh.return which in turn is
+// generated from __builtin_eh_return (offset, handler)
+// The effect of this is to adjust the stack pointer by "offset"
+// and then branch to "handler".
+SDValue MipsTargetLowering::lowerEH_RETURN(SDValue Op, SelectionDAG &DAG)
+                                                                     const {
+  MachineFunction &MF = DAG.getMachineFunction();
+  MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
+
+  MipsFI->setCallsEhReturn();
+  SDValue Chain     = Op.getOperand(0);
+  SDValue Offset    = Op.getOperand(1);
+  SDValue Handler   = Op.getOperand(2);
+  DebugLoc DL       = Op.getDebugLoc();
+  EVT Ty = IsN64 ? MVT::i64 : MVT::i32;
+
+  // Store stack offset in V1, store jump target in V0. Glue CopyToReg and
+  // EH_RETURN nodes, so that instructions are emitted back-to-back.
+  unsigned OffsetReg = IsN64 ? Mips::V1_64 : Mips::V1;
+  unsigned AddrReg = IsN64 ? Mips::V0_64 : Mips::V0;
+  Chain = DAG.getCopyToReg(Chain, DL, OffsetReg, Offset, SDValue());
+  Chain = DAG.getCopyToReg(Chain, DL, AddrReg, Handler, Chain.getValue(1));
+  return DAG.getNode(MipsISD::EH_RETURN, DL, MVT::Other, Chain,
+                     DAG.getRegister(OffsetReg, Ty),
+                     DAG.getRegister(AddrReg, getPointerTy()),
+                     Chain.getValue(1));
+}
+
 // TODO: set SType according to the desired memory barrier behavior.
 SDValue
-MipsTargetLowering::LowerMEMBARRIER(SDValue Op, SelectionDAG &DAG) const {
+MipsTargetLowering::lowerMEMBARRIER(SDValue Op, SelectionDAG &DAG) const {
   unsigned SType = 0;
-  DebugLoc dl = Op.getDebugLoc();
-  return DAG.getNode(MipsISD::Sync, dl, MVT::Other, Op.getOperand(0),
+  DebugLoc DL = Op.getDebugLoc();
+  return DAG.getNode(MipsISD::Sync, DL, MVT::Other, Op.getOperand(0),
                      DAG.getConstant(SType, MVT::i32));
 }
 
-SDValue MipsTargetLowering::LowerATOMIC_FENCE(SDValue Op,
+SDValue MipsTargetLowering::lowerATOMIC_FENCE(SDValue Op,
                                               SelectionDAG &DAG) const {
   // FIXME: Need pseudo-fence for 'singlethread' fences
   // FIXME: Set SType for weaker fences where supported/appropriate.
   unsigned SType = 0;
-  DebugLoc dl = Op.getDebugLoc();
-  return DAG.getNode(MipsISD::Sync, dl, MVT::Other, Op.getOperand(0),
+  DebugLoc DL = Op.getDebugLoc();
+  return DAG.getNode(MipsISD::Sync, DL, MVT::Other, Op.getOperand(0),
                      DAG.getConstant(SType, MVT::i32));
 }
 
-SDValue MipsTargetLowering::LowerShiftLeftParts(SDValue Op,
+SDValue MipsTargetLowering::lowerShiftLeftParts(SDValue Op,
                                                 SelectionDAG &DAG) const {
   DebugLoc DL = Op.getDebugLoc();
   SDValue Lo = Op.getOperand(0), Hi = Op.getOperand(1);
@@ -2212,7 +1877,7 @@ SDValue MipsTargetLowering::LowerShiftLeftParts(SDValue Op,
   return DAG.getMergeValues(Ops, 2, DL);
 }
 
-SDValue MipsTargetLowering::LowerShiftRightParts(SDValue Op, SelectionDAG &DAG,
+SDValue MipsTargetLowering::lowerShiftRightParts(SDValue Op, SelectionDAG &DAG,
                                                  bool IsSRA) const {
   DebugLoc DL = Op.getDebugLoc();
   SDValue Lo = Op.getOperand(0), Hi = Op.getOperand(1);
@@ -2271,7 +1936,7 @@ static SDValue CreateLoadLR(unsigned Opc, SelectionDAG &DAG, LoadSDNode *LD,
 }
 
 // Expand an unaligned 32 or 64-bit integer load node.
-SDValue MipsTargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
+SDValue MipsTargetLowering::lowerLOAD(SDValue Op, SelectionDAG &DAG) const {
   LoadSDNode *LD = cast<LoadSDNode>(Op);
   EVT MemVT = LD->getMemoryVT();
 
@@ -2349,7 +2014,7 @@ static SDValue CreateStoreLR(unsigned Opc, SelectionDAG &DAG, StoreSDNode *SD,
 }
 
 // Expand an unaligned 32 or 64-bit integer store node.
-SDValue MipsTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
+SDValue MipsTargetLowering::lowerSTORE(SDValue Op, SelectionDAG &DAG) const {
   StoreSDNode *SD = cast<StoreSDNode>(Op);
   EVT MemVT = SD->getMemoryVT();
 
@@ -2385,6 +2050,22 @@ SDValue MipsTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
   return CreateStoreLR(MipsISD::SDR, DAG, SD, SDL, IsLittle ? 0 : 7);
 }
 
+static SDValue initAccumulator(SDValue In, DebugLoc DL, SelectionDAG &DAG) {
+  SDValue InLo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, In,
+                             DAG.getConstant(0, MVT::i32));
+  SDValue InHi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, In,
+                             DAG.getConstant(1, MVT::i32));
+  return DAG.getNode(MipsISD::InsertLOHI, DL, MVT::Untyped, InLo, InHi);
+}
+
+static SDValue extractLOHI(SDValue Op, DebugLoc DL, SelectionDAG &DAG) {
+  SDValue Lo = DAG.getNode(MipsISD::ExtractLOHI, DL, MVT::i32, Op,
+                           DAG.getConstant(Mips::sub_lo, MVT::i32));
+  SDValue Hi = DAG.getNode(MipsISD::ExtractLOHI, DL, MVT::i32, Op,
+                           DAG.getConstant(Mips::sub_hi, MVT::i32));
+  return DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Lo, Hi);
+}
+
 // This function expands mips intrinsic nodes which have 64-bit input operands
 // or output values.
 //
@@ -2397,140 +2078,143 @@ SDValue MipsTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
 // v1 = copy hi
 // out64 = merge-values (v0, v1)
 //
-static SDValue LowerDSPIntr(SDValue Op, SelectionDAG &DAG,
-                            unsigned Opc, bool HasI64In, bool HasI64Out) {
+static SDValue lowerDSPIntr(SDValue Op, SelectionDAG &DAG, unsigned Opc) {
   DebugLoc DL = Op.getDebugLoc();
   bool HasChainIn = Op->getOperand(0).getValueType() == MVT::Other;
-  SDValue Chain = HasChainIn ? Op->getOperand(0) : DAG.getEntryNode();
   SmallVector<SDValue, 3> Ops;
+  unsigned OpNo = 0;
 
-  if (HasI64In) {
-    SDValue InLo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32,
-                               Op->getOperand(1 + HasChainIn),
-                               DAG.getConstant(0, MVT::i32));
-    SDValue InHi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32,
-                               Op->getOperand(1 + HasChainIn),
-                               DAG.getConstant(1, MVT::i32));
+  // See if Op has a chain input.
+  if (HasChainIn)
+    Ops.push_back(Op->getOperand(OpNo++));
 
-    Chain = DAG.getCopyToReg(Chain, DL, Mips::LO, InLo, SDValue());
-    Chain = DAG.getCopyToReg(Chain, DL, Mips::HI, InHi, Chain.getValue(1));
+  // The next operand is the intrinsic opcode.
+  assert(Op->getOperand(OpNo).getOpcode() == ISD::TargetConstant);
 
-    Ops.push_back(Chain);
-    Ops.append(Op->op_begin() + HasChainIn + 2, Op->op_end());
-    Ops.push_back(Chain.getValue(1));
-  } else {
-    Ops.push_back(Chain);
-    Ops.append(Op->op_begin() + HasChainIn + 1, Op->op_end());
-  }
+  // See if the next operand has type i64.
+  SDValue Opnd = Op->getOperand(++OpNo), In64;
+
+  if (Opnd.getValueType() == MVT::i64)
+    In64 = initAccumulator(Opnd, DL, DAG);
+  else
+    Ops.push_back(Opnd);
 
-  if (!HasI64Out)
-    return DAG.getNode(Opc, DL, Op->value_begin(), Op->getNumValues(),
-                       Ops.begin(), Ops.size());
+  // Push the remaining operands.
+  for (++OpNo ; OpNo < Op->getNumOperands(); ++OpNo)
+    Ops.push_back(Op->getOperand(OpNo));
 
-  SDValue Intr = DAG.getNode(Opc, DL, DAG.getVTList(MVT::Other, MVT::Glue),
-                             Ops.begin(), Ops.size());
-  SDValue OutLo = DAG.getCopyFromReg(Intr.getValue(0), DL, Mips::LO, MVT::i32,
-                                     Intr.getValue(1));
-  SDValue OutHi = DAG.getCopyFromReg(OutLo.getValue(1), DL, Mips::HI, MVT::i32,
-                                     OutLo.getValue(2));
-  SDValue Out = DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, OutLo, OutHi);
+  // Add In64 to the end of the list.
+  if (In64.getNode())
+    Ops.push_back(In64);
+
+  // Scan output.
+  SmallVector<EVT, 2> ResTys;
+
+  for (SDNode::value_iterator I = Op->value_begin(), E = Op->value_end();
+       I != E; ++I)
+    ResTys.push_back((*I == MVT::i64) ? MVT::Untyped : *I);
+
+  // Create node.
+  SDValue Val = DAG.getNode(Opc, DL, ResTys, &Ops[0], Ops.size());
+  SDValue Out = (ResTys[0] == MVT::Untyped) ? extractLOHI(Val, DL, DAG) : Val;
 
   if (!HasChainIn)
     return Out;
 
-  SDValue Vals[] = { Out, OutHi.getValue(1) };
+  assert(Val->getValueType(1) == MVT::Other);
+  SDValue Vals[] = { Out, SDValue(Val.getNode(), 1) };
   return DAG.getMergeValues(Vals, 2, DL);
 }
 
-SDValue MipsTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
+SDValue MipsTargetLowering::lowerINTRINSIC_WO_CHAIN(SDValue Op,
                                                     SelectionDAG &DAG) const {
   switch (cast<ConstantSDNode>(Op->getOperand(0))->getZExtValue()) {
   default:
     return SDValue();
   case Intrinsic::mips_shilo:
-    return LowerDSPIntr(Op, DAG, MipsISD::SHILO, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::SHILO);
   case Intrinsic::mips_dpau_h_qbl:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBL, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBL);
   case Intrinsic::mips_dpau_h_qbr:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBR, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBR);
   case Intrinsic::mips_dpsu_h_qbl:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBL, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBL);
   case Intrinsic::mips_dpsu_h_qbr:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBR, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBR);
   case Intrinsic::mips_dpa_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPA_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPA_W_PH);
   case Intrinsic::mips_dps_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPS_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPS_W_PH);
   case Intrinsic::mips_dpax_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPAX_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPAX_W_PH);
   case Intrinsic::mips_dpsx_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPSX_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPSX_W_PH);
   case Intrinsic::mips_mulsa_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::MULSA_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MULSA_W_PH);
   case Intrinsic::mips_mult:
-    return LowerDSPIntr(Op, DAG, MipsISD::MULT, false, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::Mult);
   case Intrinsic::mips_multu:
-    return LowerDSPIntr(Op, DAG, MipsISD::MULTU, false, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::Multu);
   case Intrinsic::mips_madd:
-    return LowerDSPIntr(Op, DAG, MipsISD::MADD_DSP, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MAdd);
   case Intrinsic::mips_maddu:
-    return LowerDSPIntr(Op, DAG, MipsISD::MADDU_DSP, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MAddu);
   case Intrinsic::mips_msub:
-    return LowerDSPIntr(Op, DAG, MipsISD::MSUB_DSP, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MSub);
   case Intrinsic::mips_msubu:
-    return LowerDSPIntr(Op, DAG, MipsISD::MSUBU_DSP, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MSubu);
   }
 }
 
-SDValue MipsTargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
+SDValue MipsTargetLowering::lowerINTRINSIC_W_CHAIN(SDValue Op,
                                                    SelectionDAG &DAG) const {
   switch (cast<ConstantSDNode>(Op->getOperand(1))->getZExtValue()) {
   default:
     return SDValue();
   case Intrinsic::mips_extp:
-    return LowerDSPIntr(Op, DAG, MipsISD::EXTP, true, false);
+    return lowerDSPIntr(Op, DAG, MipsISD::EXTP);
   case Intrinsic::mips_extpdp:
-    return LowerDSPIntr(Op, DAG, MipsISD::EXTPDP, true, false);
+    return lowerDSPIntr(Op, DAG, MipsISD::EXTPDP);
   case Intrinsic::mips_extr_w:
-    return LowerDSPIntr(Op, DAG, MipsISD::EXTR_W, true, false);
+    return lowerDSPIntr(Op, DAG, MipsISD::EXTR_W);
   case Intrinsic::mips_extr_r_w:
-    return LowerDSPIntr(Op, DAG, MipsISD::EXTR_R_W, true, false);
+    return lowerDSPIntr(Op, DAG, MipsISD::EXTR_R_W);
   case Intrinsic::mips_extr_rs_w:
-    return LowerDSPIntr(Op, DAG, MipsISD::EXTR_RS_W, true, false);
+    return lowerDSPIntr(Op, DAG, MipsISD::EXTR_RS_W);
   case Intrinsic::mips_extr_s_h:
-    return LowerDSPIntr(Op, DAG, MipsISD::EXTR_S_H, true, false);
+    return lowerDSPIntr(Op, DAG, MipsISD::EXTR_S_H);
   case Intrinsic::mips_mthlip:
-    return LowerDSPIntr(Op, DAG, MipsISD::MTHLIP, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MTHLIP);
   case Intrinsic::mips_mulsaq_s_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::MULSAQ_S_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MULSAQ_S_W_PH);
   case Intrinsic::mips_maq_s_w_phl:
-    return LowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHL, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHL);
   case Intrinsic::mips_maq_s_w_phr:
-    return LowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHR, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHR);
   case Intrinsic::mips_maq_sa_w_phl:
-    return LowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHL, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHL);
   case Intrinsic::mips_maq_sa_w_phr:
-    return LowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHR, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHR);
   case Intrinsic::mips_dpaq_s_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPAQ_S_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPAQ_S_W_PH);
   case Intrinsic::mips_dpsq_s_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPSQ_S_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPSQ_S_W_PH);
   case Intrinsic::mips_dpaq_sa_l_w:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPAQ_SA_L_W, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPAQ_SA_L_W);
   case Intrinsic::mips_dpsq_sa_l_w:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPSQ_SA_L_W, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPSQ_SA_L_W);
   case Intrinsic::mips_dpaqx_s_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPAQX_S_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPAQX_S_W_PH);
   case Intrinsic::mips_dpaqx_sa_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPAQX_SA_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPAQX_SA_W_PH);
   case Intrinsic::mips_dpsqx_s_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPSQX_S_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPSQX_S_W_PH);
   case Intrinsic::mips_dpsqx_sa_w_ph:
-    return LowerDSPIntr(Op, DAG, MipsISD::DPSQX_SA_W_PH, true, true);
+    return lowerDSPIntr(Op, DAG, MipsISD::DPSQX_SA_W_PH);
   }
 }
 
-SDValue MipsTargetLowering::LowerADD(SDValue Op, SelectionDAG &DAG) const {
+SDValue MipsTargetLowering::lowerADD(SDValue Op, SelectionDAG &DAG) const {
   if (Op->getOperand(0).getOpcode() != ISD::FRAMEADDR
       || cast<ConstantSDNode>
         (Op->getOperand(0).getOperand(0))->getZExtValue() != 0
@@ -2667,28 +2351,6 @@ static unsigned getNextIntArgReg(unsigned Reg) {
   return (Reg == Mips::A0) ? Mips::A1 : Mips::A3;
 }
 
-/// IsEligibleForTailCallOptimization - Check whether the call is eligible
-/// for tail call optimization.
-bool MipsTargetLowering::
-IsEligibleForTailCallOptimization(const MipsCC &MipsCCInfo,
-                                  unsigned NextStackOffset,
-                                  const MipsFunctionInfo& FI) const {
-  if (!EnableMipsTailCalls)
-    return false;
-
-  // No tail call optimization for mips16.
-  if (Subtarget->inMips16Mode())
-    return false;
-
-  // Return false if either the callee or caller has a byval argument.
-  if (MipsCCInfo.hasByValArg() || FI.hasByvalArg())
-    return false;
-
-  // Return true if the callee's argument area is no larger than the
-  // caller's.
-  return NextStackOffset <= FI.getIncomingArgSize();
-}
-
 SDValue
 MipsTargetLowering::passArgOnStack(SDValue StackPtr, unsigned Offset,
                                    SDValue Chain, SDValue Arg, DebugLoc DL,
@@ -2707,21 +2369,65 @@ MipsTargetLowering::passArgOnStack(SDValue StackPtr, unsigned Offset,
                       /*isVolatile=*/ true, false, 0);
 }
 
+void MipsTargetLowering::
+getOpndList(SmallVectorImpl<SDValue> &Ops,
+            std::deque< std::pair<unsigned, SDValue> > &RegsToPass,
+            bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage,
+            CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const {
+  // Insert node "GP copy globalreg" before call to function.
+  //
+  // R_MIPS_CALL* operators (emitted when non-internal functions are called
+  // in PIC mode) allow symbols to be resolved via lazy binding.
+  // The lazy binding stub requires GP to point to the GOT.
+  if (IsPICCall && !InternalLinkage) {
+    unsigned GPReg = IsN64 ? Mips::GP_64 : Mips::GP;
+    EVT Ty = IsN64 ? MVT::i64 : MVT::i32;
+    RegsToPass.push_back(std::make_pair(GPReg, getGlobalReg(CLI.DAG, Ty)));
+  }
+
+  // Build a sequence of copy-to-reg nodes chained together with token
+  // chain and flag operands which copy the outgoing args into registers.
+  // The InFlag in necessary since all emitted instructions must be
+  // stuck together.
+  SDValue InFlag;
+
+  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
+    Chain = CLI.DAG.getCopyToReg(Chain, CLI.DL, RegsToPass[i].first,
+                                 RegsToPass[i].second, InFlag);
+    InFlag = Chain.getValue(1);
+  }
+
+  // Add argument registers to the end of the list so that they are
+  // known live into the call.
+  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
+    Ops.push_back(CLI.DAG.getRegister(RegsToPass[i].first,
+                                      RegsToPass[i].second.getValueType()));
+
+  // Add a register mask operand representing the call-preserved registers.
+  const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
+  const uint32_t *Mask = TRI->getCallPreservedMask(CLI.CallConv);
+  assert(Mask && "Missing call preserved mask for calling convention");
+  Ops.push_back(CLI.DAG.getRegisterMask(Mask));
+
+  if (InFlag.getNode())
+    Ops.push_back(InFlag);
+}
+
 /// LowerCall - functions arguments are copied from virtual regs to
 /// (physical regs)/(stack frame), CALLSEQ_START and CALLSEQ_END are emitted.
 SDValue
 MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                               SmallVectorImpl<SDValue> &InVals) const {
   SelectionDAG &DAG                     = CLI.DAG;
-  DebugLoc &dl                          = CLI.DL;
+  DebugLoc &DL                          = CLI.DL;
   SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
   SmallVector<SDValue, 32> &OutVals     = CLI.OutVals;
   SmallVector<ISD::InputArg, 32> &Ins   = CLI.Ins;
   SDValue Chain                         = CLI.Chain;
   SDValue Callee                        = CLI.Callee;
-  bool &isTailCall                      = CLI.IsTailCall;
+  bool &IsTailCall                      = CLI.IsTailCall;
   CallingConv::ID CallConv              = CLI.CallConv;
-  bool isVarArg                         = CLI.IsVarArg;
+  bool IsVarArg                         = CLI.IsVarArg;
 
   MachineFunction &MF = DAG.getMachineFunction();
   MachineFrameInfo *MFI = MF.getFrameInfo();
@@ -2730,22 +2436,24 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
 
   // Analyze operands of the call, assigning locations to each operand.
   SmallVector<CCValAssign, 16> ArgLocs;
-  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
+  CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(),
                  getTargetMachine(), ArgLocs, *DAG.getContext());
-  MipsCC MipsCCInfo(CallConv, isVarArg, IsO32, CCInfo);
+  MipsCC MipsCCInfo(CallConv, IsO32, CCInfo);
 
-  MipsCCInfo.analyzeCallOperands(Outs);
+  MipsCCInfo.analyzeCallOperands(Outs, IsVarArg,
+                                 getTargetMachine().Options.UseSoftFloat,
+                                 Callee.getNode(), CLI.Args);
 
   // Get a count of how many bytes are to be pushed on the stack.
   unsigned NextStackOffset = CCInfo.getNextStackOffset();
 
   // Check if it's really possible to do a tail call.
-  if (isTailCall)
-    isTailCall =
-      IsEligibleForTailCallOptimization(MipsCCInfo, NextStackOffset,
+  if (IsTailCall)
+    IsTailCall =
+      isEligibleForTailCallOptimization(MipsCCInfo, NextStackOffset,
                                         *MF.getInfo<MipsFunctionInfo>());
 
-  if (isTailCall)
+  if (IsTailCall)
     ++NumTailCalls;
 
   // Chain is the output chain of the last Load/Store or CopyToReg node.
@@ -2755,15 +2463,15 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   NextStackOffset = RoundUpToAlignment(NextStackOffset, StackAlignment);
   SDValue NextStackOffsetVal = DAG.getIntPtrConstant(NextStackOffset, true);
 
-  if (!isTailCall)
+  if (!IsTailCall)
     Chain = DAG.getCALLSEQ_START(Chain, NextStackOffsetVal);
 
-  SDValue StackPtr = DAG.getCopyFromReg(Chain, dl,
+  SDValue StackPtr = DAG.getCopyFromReg(Chain, DL,
                                         IsN64 ? Mips::SP_64 : Mips::SP,
                                         getPointerTy());
 
   // With EABI is it possible to have 16 args on registers.
-  SmallVector<std::pair<unsigned, SDValue>, 16> RegsToPass;
+  std::deque< std::pair<unsigned, SDValue> > RegsToPass;
   SmallVector<SDValue, 8> MemOpChains;
   MipsCC::byval_iterator ByValArg = MipsCCInfo.byval_begin();
 
@@ -2779,9 +2487,9 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       assert(Flags.getByValSize() &&
              "ByVal args of size 0 should have been ignored by front-end.");
       assert(ByValArg != MipsCCInfo.byval_end());
-      assert(!isTailCall &&
+      assert(!IsTailCall &&
              "Do not tail-call optimize if there is a byval argument.");
-      passByValArg(Chain, dl, RegsToPass, MemOpChains, StackPtr, MFI, DAG, Arg,
+      passByValArg(Chain, DL, RegsToPass, MemOpChains, StackPtr, MFI, DAG, Arg,
                    MipsCCInfo, *ByValArg, Flags, Subtarget->isLittle());
       ++ByValArg;
       continue;
@@ -2793,12 +2501,13 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     case CCValAssign::Full:
       if (VA.isRegLoc()) {
         if ((ValVT == MVT::f32 && LocVT == MVT::i32) ||
-            (ValVT == MVT::f64 && LocVT == MVT::i64))
-          Arg = DAG.getNode(ISD::BITCAST, dl, LocVT, Arg);
+            (ValVT == MVT::f64 && LocVT == MVT::i64) ||
+            (ValVT == MVT::i64 && LocVT == MVT::f64))
+          Arg = DAG.getNode(ISD::BITCAST, DL, LocVT, Arg);
         else if (ValVT == MVT::f64 && LocVT == MVT::i32) {
-          SDValue Lo = DAG.getNode(MipsISD::ExtractElementF64, dl, MVT::i32,
+          SDValue Lo = DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32,
                                    Arg, DAG.getConstant(0, MVT::i32));
-          SDValue Hi = DAG.getNode(MipsISD::ExtractElementF64, dl, MVT::i32,
+          SDValue Hi = DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32,
                                    Arg, DAG.getConstant(1, MVT::i32));
           if (!Subtarget->isLittle())
             std::swap(Lo, Hi);
@@ -2811,13 +2520,13 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       }
       break;
     case CCValAssign::SExt:
-      Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, LocVT, Arg);
+      Arg = DAG.getNode(ISD::SIGN_EXTEND, DL, LocVT, Arg);
       break;
     case CCValAssign::ZExt:
-      Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, LocVT, Arg);
+      Arg = DAG.getNode(ISD::ZERO_EXTEND, DL, LocVT, Arg);
       break;
     case CCValAssign::AExt:
-      Arg = DAG.getNode(ISD::ANY_EXTEND, dl, LocVT, Arg);
+      Arg = DAG.getNode(ISD::ANY_EXTEND, DL, LocVT, Arg);
       break;
     }
 
@@ -2834,25 +2543,27 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     // emit ISD::STORE whichs stores the
     // parameter value to a stack Location
     MemOpChains.push_back(passArgOnStack(StackPtr, VA.getLocMemOffset(),
-                                         Chain, Arg, dl, isTailCall, DAG));
+                                         Chain, Arg, DL, IsTailCall, DAG));
   }
 
   // Transform all store nodes into one single node because all store
   // nodes are independent of each other.
   if (!MemOpChains.empty())
-    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
+    Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other,
                         &MemOpChains[0], MemOpChains.size());
 
   // If the callee is a GlobalAddress/ExternalSymbol node (quite common, every
   // direct call is) turn it into a TargetGlobalAddress/TargetExternalSymbol
   // node so that legalize doesn't hack it.
   bool IsPICCall = (IsN64 || IsPIC); // true if calls are translated to jalr $25
-  bool GlobalOrExternal = false;
+  bool GlobalOrExternal = false, InternalLinkage = false;
   SDValue CalleeLo;
 
   if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
     if (IsPICCall) {
-      if (G->getGlobal()->hasInternalLinkage())
+      InternalLinkage = G->getGlobal()->hasInternalLinkage();
+
+      if (InternalLinkage)
         Callee = getAddrLocal(Callee, DAG, HasMips64);
       else if (LargeGOT)
         Callee = getAddrGlobalLargeGOT(Callee, DAG, MipsII::MO_CALL_HI16,
@@ -2860,7 +2571,7 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       else
         Callee = getAddrGlobal(Callee, DAG, MipsII::MO_GOT_CALL);
     } else
-      Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl, getPointerTy(), 0,
+      Callee = DAG.getTargetGlobalAddress(G->getGlobal(), DL, getPointerTy(), 0,
                                           MipsII::MO_NO_FLAG);
     GlobalOrExternal = true;
   }
@@ -2871,84 +2582,23 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     else if (LargeGOT)
       Callee = getAddrGlobalLargeGOT(Callee, DAG, MipsII::MO_CALL_HI16,
                                      MipsII::MO_CALL_LO16);
-    else if (HasMips64)
-      Callee = getAddrGlobal(Callee, DAG, MipsII::MO_GOT_DISP);
-    else // O32 & PIC
+    else // N64 || PIC
       Callee = getAddrGlobal(Callee, DAG, MipsII::MO_GOT_CALL);
 
     GlobalOrExternal = true;
   }
 
-  SDValue InFlag;
-
-  // T9 register operand.
-  SDValue T9;
-
-  // T9 should contain the address of the callee function if
-  // -reloction-model=pic or it is an indirect call.
-  if (IsPICCall || !GlobalOrExternal) {
-    // copy to T9
-    unsigned T9Reg = IsN64 ? Mips::T9_64 : Mips::T9;
-    Chain = DAG.getCopyToReg(Chain, dl, T9Reg, Callee, SDValue(0, 0));
-    InFlag = Chain.getValue(1);
-
-    if (Subtarget->inMips16Mode())
-      T9 = DAG.getRegister(T9Reg, getPointerTy());
-    else
-      Callee = DAG.getRegister(T9Reg, getPointerTy());
-  }
-
-  // Insert node "GP copy globalreg" before call to function.
-  // Lazy-binding stubs require GP to point to the GOT.
-  if (IsPICCall) {
-    unsigned GPReg = IsN64 ? Mips::GP_64 : Mips::GP;
-    EVT Ty = IsN64 ? MVT::i64 : MVT::i32;
-    RegsToPass.push_back(std::make_pair(GPReg, GetGlobalReg(DAG, Ty)));
-  }
-
-  // Build a sequence of copy-to-reg nodes chained together with token
-  // chain and flag operands which copy the outgoing args into registers.
-  // The InFlag in necessary since all emitted instructions must be
-  // stuck together.
-  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
-    Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
-                             RegsToPass[i].second, InFlag);
-    InFlag = Chain.getValue(1);
-  }
-
-  // MipsJmpLink = #chain, #target_address, #opt_in_flags...
-  //             = Chain, Callee, Reg#1, Reg#2, ...
-  //
-  // Returns a chain & a flag for retval copy to use.
+  SmallVector<SDValue, 8> Ops(1, Chain);
   SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
-  SmallVector<SDValue, 8> Ops;
-  Ops.push_back(Chain);
-  Ops.push_back(Callee);
 
-  // Add argument registers to the end of the list so that they are
-  // known live into the call.
-  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
-    Ops.push_back(DAG.getRegister(RegsToPass[i].first,
-                                  RegsToPass[i].second.getValueType()));
+  getOpndList(Ops, RegsToPass, IsPICCall, GlobalOrExternal, InternalLinkage,
+              CLI, Callee, Chain);
 
-  // Add T9 register operand.
-  if (T9.getNode())
-    Ops.push_back(T9);
+  if (IsTailCall)
+    return DAG.getNode(MipsISD::TailCall, DL, MVT::Other, &Ops[0], Ops.size());
 
-  // Add a register mask operand representing the call-preserved registers.
-  const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
-  const uint32_t *Mask = TRI->getCallPreservedMask(CallConv);
-  assert(Mask && "Missing call preserved mask for calling convention");
-  Ops.push_back(DAG.getRegisterMask(Mask));
-
-  if (InFlag.getNode())
-    Ops.push_back(InFlag);
-
-  if (isTailCall)
-    return DAG.getNode(MipsISD::TailCall, dl, MVT::Other, &Ops[0], Ops.size());
-
-  Chain  = DAG.getNode(MipsISD::JmpLink, dl, NodeTys, &Ops[0], Ops.size());
-  InFlag = Chain.getValue(1);
+  Chain  = DAG.getNode(MipsISD::JmpLink, DL, NodeTys, &Ops[0], Ops.size());
+  SDValue InFlag = Chain.getValue(1);
 
   // Create the CALLSEQ_END node.
   Chain = DAG.getCALLSEQ_END(Chain, NextStackOffsetVal,
@@ -2957,31 +2607,40 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
 
   // Handle result values, copying them out of physregs into vregs that we
   // return.
-  return LowerCallResult(Chain, InFlag, CallConv, isVarArg,
-                         Ins, dl, DAG, InVals);
+  return LowerCallResult(Chain, InFlag, CallConv, IsVarArg,
+                         Ins, DL, DAG, InVals, CLI.Callee.getNode(), CLI.RetTy);
 }
 
 /// LowerCallResult - Lower the result values of a call into the
 /// appropriate copies out of appropriate physical registers.
 SDValue
 MipsTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
-                                    CallingConv::ID CallConv, bool isVarArg,
+                                    CallingConv::ID CallConv, bool IsVarArg,
                                     const SmallVectorImpl<ISD::InputArg> &Ins,
-                                    DebugLoc dl, SelectionDAG &DAG,
-                                    SmallVectorImpl<SDValue> &InVals) const {
+                                    DebugLoc DL, SelectionDAG &DAG,
+                                    SmallVectorImpl<SDValue> &InVals,
+                                    const SDNode *CallNode,
+                                    const Type *RetTy) const {
   // Assign locations to each value returned by this call.
   SmallVector<CCValAssign, 16> RVLocs;
-  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
+  CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(),
                  getTargetMachine(), RVLocs, *DAG.getContext());
+  MipsCC MipsCCInfo(CallConv, IsO32, CCInfo);
 
-  CCInfo.AnalyzeCallResult(Ins, RetCC_Mips);
+  MipsCCInfo.analyzeCallResult(Ins, getTargetMachine().Options.UseSoftFloat,
+                               CallNode, RetTy);
 
   // Copy all of the result registers out of their specified physreg.
   for (unsigned i = 0; i != RVLocs.size(); ++i) {
-    Chain = DAG.getCopyFromReg(Chain, dl, RVLocs[i].getLocReg(),
-                               RVLocs[i].getValVT(), InFlag).getValue(1);
-    InFlag = Chain.getValue(2);
-    InVals.push_back(Chain.getValue(0));
+    SDValue Val = DAG.getCopyFromReg(Chain, DL, RVLocs[i].getLocReg(),
+                                     RVLocs[i].getLocVT(), InFlag);
+    Chain = Val.getValue(1);
+    InFlag = Val.getValue(2);
+
+    if (RVLocs[i].getValVT() != RVLocs[i].getLocVT())
+      Val = DAG.getNode(ISD::BITCAST, DL, RVLocs[i].getValVT(), Val);
+
+    InVals.push_back(Val);
   }
 
   return Chain;
@@ -2995,9 +2654,9 @@ MipsTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
 SDValue
 MipsTargetLowering::LowerFormalArguments(SDValue Chain,
                                          CallingConv::ID CallConv,
-                                         bool isVarArg,
+                                         bool IsVarArg,
                                       const SmallVectorImpl<ISD::InputArg> &Ins,
-                                         DebugLoc dl, SelectionDAG &DAG,
+                                         DebugLoc DL, SelectionDAG &DAG,
                                          SmallVectorImpl<SDValue> &InVals)
                                           const {
   MachineFunction &MF = DAG.getMachineFunction();
@@ -3011,16 +2670,17 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain,
 
   // Assign locations to all of the incoming arguments.
   SmallVector<CCValAssign, 16> ArgLocs;
-  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
+  CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(),
                  getTargetMachine(), ArgLocs, *DAG.getContext());
-  MipsCC MipsCCInfo(CallConv, isVarArg, IsO32, CCInfo);
+  MipsCC MipsCCInfo(CallConv, IsO32, CCInfo);
+  Function::const_arg_iterator FuncArg =
+    DAG.getMachineFunction().getFunction()->arg_begin();
+  bool UseSoftFloat = getTargetMachine().Options.UseSoftFloat;
 
-  MipsCCInfo.analyzeFormalArguments(Ins);
+  MipsCCInfo.analyzeFormalArguments(Ins, UseSoftFloat, FuncArg);
   MipsFI->setFormalArgInfo(CCInfo.getNextStackOffset(),
                            MipsCCInfo.hasByValArg());
 
-  Function::const_arg_iterator FuncArg =
-    DAG.getMachineFunction().getFunction()->arg_begin();
   unsigned CurArgIdx = 0;
   MipsCC::byval_iterator ByValArg = MipsCCInfo.byval_begin();
 
@@ -3036,7 +2696,7 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain,
       assert(Flags.getByValSize() &&
              "ByVal args of size 0 should have been ignored by front-end.");
       assert(ByValArg != MipsCCInfo.byval_end());
-      copyByValRegs(Chain, dl, OutChains, DAG, Flags, InVals, &*FuncArg,
+      copyByValRegs(Chain, DL, OutChains, DAG, Flags, InVals, &*FuncArg,
                     MipsCCInfo, *ByValArg);
       ++ByValArg;
       continue;
@@ -3049,7 +2709,8 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain,
       const TargetRegisterClass *RC;
 
       if (RegVT == MVT::i32)
-        RC = &Mips::CPURegsRegClass;
+        RC = Subtarget->inMips16Mode()? &Mips::CPU16RegsRegClass :
+                                        &Mips::CPURegsRegClass;
       else if (RegVT == MVT::i64)
         RC = &Mips::CPU64RegsRegClass;
       else if (RegVT == MVT::f32)
@@ -3061,8 +2722,8 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain,
 
       // Transform the arguments stored on
       // physical registers into virtual ones
-      unsigned Reg = AddLiveIn(DAG.getMachineFunction(), ArgReg, RC);
-      SDValue ArgValue = DAG.getCopyFromReg(Chain, dl, Reg, RegVT);
+      unsigned Reg = addLiveIn(DAG.getMachineFunction(), ArgReg, RC);
+      SDValue ArgValue = DAG.getCopyFromReg(Chain, DL, Reg, RegVT);
 
       // If this is an 8 or 16-bit value, it has been passed promoted
       // to 32 bits.  Insert an assert[sz]ext to capture this, then
@@ -3074,22 +2735,24 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain,
         else if (VA.getLocInfo() == CCValAssign::ZExt)
           Opcode = ISD::AssertZext;
         if (Opcode)
-          ArgValue = DAG.getNode(Opcode, dl, RegVT, ArgValue,
+          ArgValue = DAG.getNode(Opcode, DL, RegVT, ArgValue,
                                  DAG.getValueType(ValVT));
-        ArgValue = DAG.getNode(ISD::TRUNCATE, dl, ValVT, ArgValue);
+        ArgValue = DAG.getNode(ISD::TRUNCATE, DL, ValVT, ArgValue);
       }
 
-      // Handle floating point arguments passed in integer registers.
+      // Handle floating point arguments passed in integer registers and
+      // long double arguments passed in floating point registers.
       if ((RegVT == MVT::i32 && ValVT == MVT::f32) ||
-          (RegVT == MVT::i64 && ValVT == MVT::f64))
-        ArgValue = DAG.getNode(ISD::BITCAST, dl, ValVT, ArgValue);
+          (RegVT == MVT::i64 && ValVT == MVT::f64) ||
+          (RegVT == MVT::f64 && ValVT == MVT::i64))
+        ArgValue = DAG.getNode(ISD::BITCAST, DL, ValVT, ArgValue);
       else if (IsO32 && RegVT == MVT::i32 && ValVT == MVT::f64) {
-        unsigned Reg2 = AddLiveIn(DAG.getMachineFunction(),
+        unsigned Reg2 = addLiveIn(DAG.getMachineFunction(),
                                   getNextIntArgReg(ArgReg), RC);
-        SDValue ArgValue2 = DAG.getCopyFromReg(Chain, dl, Reg2, RegVT);
+        SDValue ArgValue2 = DAG.getCopyFromReg(Chain, DL, Reg2, RegVT);
         if (!Subtarget->isLittle())
           std::swap(ArgValue, ArgValue2);
-        ArgValue = DAG.getNode(MipsISD::BuildPairF64, dl, MVT::f64,
+        ArgValue = DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64,
                                ArgValue, ArgValue2);
       }
 
@@ -3105,7 +2768,7 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain,
 
       // Create load nodes to retrieve arguments from the stack
       SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
-      InVals.push_back(DAG.getLoad(ValVT, dl, Chain, FIN,
+      InVals.push_back(DAG.getLoad(ValVT, DL, Chain, FIN,
                                    MachinePointerInfo::getFixedStack(FI),
                                    false, false, false, 0));
     }
@@ -3121,18 +2784,18 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain,
         createVirtualRegister(getRegClassFor(IsN64 ? MVT::i64 : MVT::i32));
       MipsFI->setSRetReturnReg(Reg);
     }
-    SDValue Copy = DAG.getCopyToReg(DAG.getEntryNode(), dl, Reg, InVals[0]);
-    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Copy, Chain);
+    SDValue Copy = DAG.getCopyToReg(DAG.getEntryNode(), DL, Reg, InVals[0]);
+    Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Copy, Chain);
   }
 
-  if (isVarArg)
-    writeVarArgRegs(OutChains, MipsCCInfo, Chain, dl, DAG);
+  if (IsVarArg)
+    writeVarArgRegs(OutChains, MipsCCInfo, Chain, DL, DAG);
 
   // All stores are grouped in one node to allow the matching between
   // the size of Ins and InVals. This only happens when on varg functions
   if (!OutChains.empty()) {
     OutChains.push_back(Chain);
-    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
+    Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other,
                         &OutChains[0], OutChains.size());
   }
 
@@ -3145,80 +2808,80 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain,
 
 bool
 MipsTargetLowering::CanLowerReturn(CallingConv::ID CallConv,
-                                   MachineFunction &MF, bool isVarArg,
+                                   MachineFunction &MF, bool IsVarArg,
                                    const SmallVectorImpl<ISD::OutputArg> &Outs,
                                    LLVMContext &Context) const {
   SmallVector<CCValAssign, 16> RVLocs;
-  CCState CCInfo(CallConv, isVarArg, MF, getTargetMachine(),
+  CCState CCInfo(CallConv, IsVarArg, MF, getTargetMachine(),
                  RVLocs, Context);
   return CCInfo.CheckReturn(Outs, RetCC_Mips);
 }
 
 SDValue
 MipsTargetLowering::LowerReturn(SDValue Chain,
-                                CallingConv::ID CallConv, bool isVarArg,
+                                CallingConv::ID CallConv, bool IsVarArg,
                                 const SmallVectorImpl<ISD::OutputArg> &Outs,
                                 const SmallVectorImpl<SDValue> &OutVals,
-                                DebugLoc dl, SelectionDAG &DAG) const {
-
+                                DebugLoc DL, SelectionDAG &DAG) const {
   // CCValAssign - represent the assignment of
   // the return value to a location
   SmallVector<CCValAssign, 16> RVLocs;
+  MachineFunction &MF = DAG.getMachineFunction();
 
   // CCState - Info about the registers and stack slot.
-  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
-                 getTargetMachine(), RVLocs, *DAG.getContext());
+  CCState CCInfo(CallConv, IsVarArg, MF, getTargetMachine(), RVLocs,
+                 *DAG.getContext());
+  MipsCC MipsCCInfo(CallConv, IsO32, CCInfo);
 
-  // Analize return values.
-  CCInfo.AnalyzeReturn(Outs, RetCC_Mips);
-
-  // If this is the first return lowered for this function, add
-  // the regs to the liveout set for the function.
-  if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
-    for (unsigned i = 0; i != RVLocs.size(); ++i)
-      if (RVLocs[i].isRegLoc())
-        DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
-  }
+  // Analyze return values.
+  MipsCCInfo.analyzeReturn(Outs, getTargetMachine().Options.UseSoftFloat,
+                           MF.getFunction()->getReturnType());
 
   SDValue Flag;
+  SmallVector<SDValue, 4> RetOps(1, Chain);
 
   // Copy the result values into the output registers.
   for (unsigned i = 0; i != RVLocs.size(); ++i) {
+    SDValue Val = OutVals[i];
     CCValAssign &VA = RVLocs[i];
     assert(VA.isRegLoc() && "Can only return in registers!");
 
-    Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), OutVals[i], Flag);
+    if (RVLocs[i].getValVT() != RVLocs[i].getLocVT())
+      Val = DAG.getNode(ISD::BITCAST, DL, RVLocs[i].getLocVT(), Val);
+
+    Chain = DAG.getCopyToReg(Chain, DL, VA.getLocReg(), Val, Flag);
 
-    // guarantee that all emitted copies are
-    // stuck together, avoiding something bad
+    // Guarantee that all emitted copies are stuck together with flags.
     Flag = Chain.getValue(1);
+    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
   }
 
   // The mips ABIs for returning structs by value requires that we copy
   // the sret argument into $v0 for the return. We saved the argument into
   // a virtual register in the entry block, so now we copy the value out
   // and into $v0.
-  if (DAG.getMachineFunction().getFunction()->hasStructRetAttr()) {
-    MachineFunction &MF      = DAG.getMachineFunction();
+  if (MF.getFunction()->hasStructRetAttr()) {
     MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
     unsigned Reg = MipsFI->getSRetReturnReg();
 
     if (!Reg)
       llvm_unreachable("sret virtual register not created in the entry block");
-    SDValue Val = DAG.getCopyFromReg(Chain, dl, Reg, getPointerTy());
+    SDValue Val = DAG.getCopyFromReg(Chain, DL, Reg, getPointerTy());
     unsigned V0 = IsN64 ? Mips::V0_64 : Mips::V0;
 
-    Chain = DAG.getCopyToReg(Chain, dl, V0, Val, Flag);
+    Chain = DAG.getCopyToReg(Chain, DL, V0, Val, Flag);
     Flag = Chain.getValue(1);
-    MF.getRegInfo().addLiveOut(V0);
+    RetOps.push_back(DAG.getRegister(V0, getPointerTy()));
   }
 
-  // Return on Mips is always a "jr $ra"
+  RetOps[0] = Chain;  // Update chain.
+
+  // Add the flag if we have it.
   if (Flag.getNode())
-    return DAG.getNode(MipsISD::Ret, dl, MVT::Other, Chain, Flag);
+    RetOps.push_back(Flag);
 
-  // Return Void
-  return DAG.getNode(MipsISD::Ret, dl, MVT::Other, Chain);
+  // Return on Mips is always a "jr $ra"
+  return DAG.getNode(MipsISD::Ret, DL, MVT::Other, &RetOps[0], RetOps.size());
 }
 
 //===----------------------------------------------------------------------===//
@@ -3251,6 +2914,8 @@ getConstraintType(const std::string &Constraint) const
       case 'l':
       case 'x':
         return C_RegisterClass;
+      case 'R':
+        return C_Memory;
     }
   }
   return TargetLowering::getConstraintType(Constraint);
@@ -3299,6 +2964,9 @@ MipsTargetLowering::getSingleConstraintMatchWeight(
     if (isa<ConstantInt>(CallOperandVal))
       weight = CW_Constant;
     break;
+  case 'R':
+    weight = CW_Memory;
+    break;
   }
   return weight;
 }
@@ -3448,13 +3116,34 @@ void MipsTargetLowering::LowerAsmOperandForConstraint(SDValue Op,
 }
 
 bool
+MipsTargetLowering::isLegalAddressingMode(const AddrMode &AM, Type *Ty) const {
+  // No global is ever allowed as a base.
+  if (AM.BaseGV)
+    return false;
+
+  switch (AM.Scale) {
+  case 0: // "r+i" or just "i", depending on HasBaseReg.
+    break;
+  case 1:
+    if (!AM.HasBaseReg) // allow "r+i".
+      break;
+    return false; // disallow "r+r" or "r+r+i".
+  default:
+    return false;
+  }
+
+  return true;
+}
+
+bool
 MipsTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
   // The Mips target isn't yet aware of offsets.
   return false;
 }
 
 EVT MipsTargetLowering::getOptimalMemOpType(uint64_t Size, unsigned DstAlign,
-                                            unsigned SrcAlign, bool IsZeroVal,
+                                            unsigned SrcAlign,
+                                            bool IsMemset, bool ZeroMemset,
                                             bool MemcpyStrSrc,
                                             MachineFunction &MF) const {
   if (Subtarget->hasMips64())
@@ -3478,40 +3167,62 @@ unsigned MipsTargetLowering::getJumpTableEncoding() const {
   return TargetLowering::getJumpTableEncoding();
 }
 
-MipsTargetLowering::MipsCC::MipsCC(CallingConv::ID CallConv, bool IsVarArg,
-                                   bool IsO32, CCState &Info) : CCInfo(Info) {
-  UseRegsForByval = true;
+/// This function returns true if CallSym is a long double emulation routine.
+static bool isF128SoftLibCall(const char *CallSym) {
+  const char *const LibCalls[] =
+    {"__addtf3", "__divtf3", "__eqtf2", "__extenddftf2", "__extendsftf2",
+     "__fixtfdi", "__fixtfsi", "__fixtfti", "__fixunstfdi", "__fixunstfsi",
+     "__fixunstfti", "__floatditf", "__floatsitf", "__floattitf",
+     "__floatunditf", "__floatunsitf", "__floatuntitf", "__getf2", "__gttf2",
+     "__letf2", "__lttf2", "__multf3", "__netf2", "__powitf2", "__subtf3",
+     "__trunctfdf2", "__trunctfsf2", "__unordtf2",
+     "ceill", "copysignl", "cosl", "exp2l", "expl", "floorl", "fmal", "fmodl",
+     "log10l", "log2l", "logl", "nearbyintl", "powl", "rintl", "sinl", "sqrtl",
+     "truncl"};
 
-  if (IsO32) {
-    RegSize = 4;
-    NumIntArgRegs = array_lengthof(O32IntRegs);
-    ReservedArgArea = 16;
-    IntArgRegs = ShadowRegs = O32IntRegs;
-    FixedFn = VarFn = CC_MipsO32;
-  } else {
-    RegSize = 8;
-    NumIntArgRegs = array_lengthof(Mips64IntRegs);
-    ReservedArgArea = 0;
-    IntArgRegs = Mips64IntRegs;
-    ShadowRegs = Mips64DPRegs;
-    FixedFn = CC_MipsN;
-    VarFn = CC_MipsN_VarArg;
-  }
+  const char * const *End = LibCalls + array_lengthof(LibCalls);
 
-  if (CallConv == CallingConv::Fast) {
-    assert(!IsVarArg);
-    UseRegsForByval = false;
-    ReservedArgArea = 0;
-    FixedFn = VarFn = CC_Mips_FastCC;
-  }
+  // Check that LibCalls is sorted alphabetically.
+  MipsTargetLowering::LTStr Comp;
+
+#ifndef NDEBUG
+  for (const char * const *I = LibCalls; I < End - 1; ++I)
+    assert(Comp(*I, *(I + 1)));
+#endif
+
+  return std::binary_search(LibCalls, End, CallSym, Comp);
+}
+
+/// This function returns true if Ty is fp128 or i128 which was originally a
+/// fp128.
+static bool originalTypeIsF128(const Type *Ty, const SDNode *CallNode) {
+  if (Ty->isFP128Ty())
+    return true;
 
+  const ExternalSymbolSDNode *ES =
+    dyn_cast_or_null<const ExternalSymbolSDNode>(CallNode);
+
+  // If the Ty is i128 and the function being called is a long double emulation
+  // routine, then the original type is f128.
+  return (ES && Ty->isIntegerTy(128) && isF128SoftLibCall(ES->getSymbol()));
+}
+
+MipsTargetLowering::MipsCC::MipsCC(CallingConv::ID CC, bool IsO32_,
+                                   CCState &Info)
+  : CCInfo(Info), CallConv(CC), IsO32(IsO32_) {
   // Pre-allocate reserved argument area.
-  CCInfo.AllocateStack(ReservedArgArea, 1);
+  CCInfo.AllocateStack(reservedArgArea(), 1);
 }
 
 void MipsTargetLowering::MipsCC::
-analyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Args) {
+analyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Args,
+                    bool IsVarArg, bool IsSoftFloat, const SDNode *CallNode,
+                    std::vector<ArgListEntry> &FuncArgs) {
+  assert((CallConv != CallingConv::Fast || !IsVarArg) &&
+         "CallingConv::Fast shouldn't be used for vararg functions.");
+
   unsigned NumOpnds = Args.size();
+  llvm::CCAssignFn *FixedFn = fixedArgFn(), *VarFn = varArgFn();
 
   for (unsigned I = 0; I != NumOpnds; ++I) {
     MVT ArgVT = Args[I].VT;
@@ -3523,10 +3234,13 @@ analyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Args) {
       continue;
     }
 
-    if (Args[I].IsFixed)
-      R = FixedFn(I, ArgVT, ArgVT, CCValAssign::Full, ArgFlags, CCInfo);
-    else
+    if (IsVarArg && !Args[I].IsFixed)
       R = VarFn(I, ArgVT, ArgVT, CCValAssign::Full, ArgFlags, CCInfo);
+    else {
+      MVT RegVT = getRegVT(ArgVT, FuncArgs[Args[I].OrigArgIndex].Ty, CallNode,
+                           IsSoftFloat);
+      R = FixedFn(I, ArgVT, RegVT, CCValAssign::Full, ArgFlags, CCInfo);
+    }
 
     if (R) {
 #ifndef NDEBUG
@@ -3539,19 +3253,26 @@ analyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Args) {
 }
 
 void MipsTargetLowering::MipsCC::
-analyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Args) {
+analyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Args,
+                       bool IsSoftFloat, Function::const_arg_iterator FuncArg) {
   unsigned NumArgs = Args.size();
+  llvm::CCAssignFn *FixedFn = fixedArgFn();
+  unsigned CurArgIdx = 0;
 
   for (unsigned I = 0; I != NumArgs; ++I) {
     MVT ArgVT = Args[I].VT;
     ISD::ArgFlagsTy ArgFlags = Args[I].Flags;
+    std::advance(FuncArg, Args[I].OrigArgIndex - CurArgIdx);
+    CurArgIdx = Args[I].OrigArgIndex;
 
     if (ArgFlags.isByVal()) {
       handleByValArg(I, ArgVT, ArgVT, CCValAssign::Full, ArgFlags);
       continue;
     }
 
-    if (!FixedFn(I, ArgVT, ArgVT, CCValAssign::Full, ArgFlags, CCInfo))
+    MVT RegVT = getRegVT(ArgVT, FuncArg->getType(), 0, IsSoftFloat);
+
+    if (!FixedFn(I, ArgVT, RegVT, CCValAssign::Full, ArgFlags, CCInfo))
       continue;
 
 #ifndef NDEBUG
@@ -3562,6 +3283,44 @@ analyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Args) {
   }
 }
 
+template<typename Ty>
+void MipsTargetLowering::MipsCC::
+analyzeReturn(const SmallVectorImpl<Ty> &RetVals, bool IsSoftFloat,
+              const SDNode *CallNode, const Type *RetTy) const {
+  CCAssignFn *Fn;
+
+  if (IsSoftFloat && originalTypeIsF128(RetTy, CallNode))
+    Fn = RetCC_F128Soft;
+  else
+    Fn = RetCC_Mips;
+
+  for (unsigned I = 0, E = RetVals.size(); I < E; ++I) {
+    MVT VT = RetVals[I].VT;
+    ISD::ArgFlagsTy Flags = RetVals[I].Flags;
+    MVT RegVT = this->getRegVT(VT, RetTy, CallNode, IsSoftFloat);
+
+    if (Fn(I, VT, RegVT, CCValAssign::Full, Flags, this->CCInfo)) {
+#ifndef NDEBUG
+      dbgs() << "Call result #" << I << " has unhandled type "
+             << EVT(VT).getEVTString() << '\n';
+#endif
+      llvm_unreachable(0);
+    }
+  }
+}
+
+void MipsTargetLowering::MipsCC::
+analyzeCallResult(const SmallVectorImpl<ISD::InputArg> &Ins, bool IsSoftFloat,
+                  const SDNode *CallNode, const Type *RetTy) const {
+  analyzeReturn(Ins, IsSoftFloat, CallNode, RetTy);
+}
+
+void MipsTargetLowering::MipsCC::
+analyzeReturn(const SmallVectorImpl<ISD::OutputArg> &Outs, bool IsSoftFloat,
+              const Type *RetTy) const {
+  analyzeReturn(Outs, IsSoftFloat, 0, RetTy);
+}
+
 void
 MipsTargetLowering::MipsCC::handleByValArg(unsigned ValNo, MVT ValVT,
                                            MVT LocVT,
@@ -3570,11 +3329,12 @@ MipsTargetLowering::MipsCC::handleByValArg(unsigned ValNo, MVT ValVT,
   assert(ArgFlags.getByValSize() && "Byval argument's size shouldn't be 0.");
 
   struct ByValArgInfo ByVal;
+  unsigned RegSize = regSize();
   unsigned ByValSize = RoundUpToAlignment(ArgFlags.getByValSize(), RegSize);
   unsigned Align = std::min(std::max(ArgFlags.getByValAlign(), RegSize),
                             RegSize * 2);
 
-  if (UseRegsForByval)
+  if (useRegsForByval())
     allocateRegs(ByVal, ByValSize, Align);
 
   // Allocate space on caller's stack.
@@ -3585,9 +3345,38 @@ MipsTargetLowering::MipsCC::handleByValArg(unsigned ValNo, MVT ValVT,
   ByValArgs.push_back(ByVal);
 }
 
+unsigned MipsTargetLowering::MipsCC::numIntArgRegs() const {
+  return IsO32 ? array_lengthof(O32IntRegs) : array_lengthof(Mips64IntRegs);
+}
+
+unsigned MipsTargetLowering::MipsCC::reservedArgArea() const {
+  return (IsO32 && (CallConv != CallingConv::Fast)) ? 16 : 0;
+}
+
+const uint16_t *MipsTargetLowering::MipsCC::intArgRegs() const {
+  return IsO32 ? O32IntRegs : Mips64IntRegs;
+}
+
+llvm::CCAssignFn *MipsTargetLowering::MipsCC::fixedArgFn() const {
+  if (CallConv == CallingConv::Fast)
+    return CC_Mips_FastCC;
+
+  return IsO32 ? CC_MipsO32 : CC_MipsN;
+}
+
+llvm::CCAssignFn *MipsTargetLowering::MipsCC::varArgFn() const {
+  return IsO32 ? CC_MipsO32 : CC_MipsN_VarArg;
+}
+
+const uint16_t *MipsTargetLowering::MipsCC::shadowRegs() const {
+  return IsO32 ? O32IntRegs : Mips64DPRegs;
+}
+
 void MipsTargetLowering::MipsCC::allocateRegs(ByValArgInfo &ByVal,
                                               unsigned ByValSize,
                                               unsigned Align) {
+  unsigned RegSize = regSize(), NumIntArgRegs = numIntArgRegs();
+  const uint16_t *IntArgRegs = intArgRegs(), *ShadowRegs = shadowRegs();
   assert(!(ByValSize % RegSize) && !(Align % RegSize) &&
          "Byval argument's size and alignment should be a multiple of"
          "RegSize.");
@@ -3606,6 +3395,21 @@ void MipsTargetLowering::MipsCC::allocateRegs(ByValArgInfo &ByVal,
     CCInfo.AllocateReg(IntArgRegs[I], ShadowRegs[I]);
 }
 
+MVT MipsTargetLowering::MipsCC::getRegVT(MVT VT, const Type *OrigTy,
+                                         const SDNode *CallNode,
+                                         bool IsSoftFloat) const {
+  if (IsSoftFloat || IsO32)
+    return VT;
+
+  // Check if the original type was fp128.
+  if (originalTypeIsF128(OrigTy, CallNode)) {
+    assert(VT == MVT::i64);
+    return MVT::f64;
+  }
+
+  return VT;
+}
+
 void MipsTargetLowering::
 copyByValRegs(SDValue Chain, DebugLoc DL, std::vector<SDValue> &OutChains,
               SelectionDAG &DAG, const ISD::ArgFlagsTy &Flags,
@@ -3633,12 +3437,12 @@ copyByValRegs(SDValue Chain, DebugLoc DL, std::vector<SDValue> &OutChains,
     return;
 
   // Copy arg registers.
-  EVT RegTy = MVT::getIntegerVT(CC.regSize() * 8);
+  MVT RegTy = MVT::getIntegerVT(CC.regSize() * 8);
   const TargetRegisterClass *RC = getRegClassFor(RegTy);
 
   for (unsigned I = 0; I < ByVal.NumRegs; ++I) {
     unsigned ArgReg = CC.intArgRegs()[ByVal.FirstIdx + I];
-    unsigned VReg = AddLiveIn(MF, ArgReg, RC);
+    unsigned VReg = addLiveIn(MF, ArgReg, RC);
     unsigned Offset = I * CC.regSize();
     SDValue StorePtr = DAG.getNode(ISD::ADD, DL, PtrTy, FIN,
                                    DAG.getConstant(Offset, PtrTy));
@@ -3652,7 +3456,7 @@ copyByValRegs(SDValue Chain, DebugLoc DL, std::vector<SDValue> &OutChains,
 // Copy byVal arg to registers and stack.
 void MipsTargetLowering::
 passByValArg(SDValue Chain, DebugLoc DL,
-             SmallVector<std::pair<unsigned, SDValue>, 16> &RegsToPass,
+             std::deque< std::pair<unsigned, SDValue> > &RegsToPass,
              SmallVector<SDValue, 8> &MemOpChains, SDValue StackPtr,
              MachineFrameInfo *MFI, SelectionDAG &DAG, SDValue Arg,
              const MipsCC &CC, const ByValArgInfo &ByVal,
@@ -3755,7 +3559,7 @@ MipsTargetLowering::writeVarArgRegs(std::vector<SDValue> &OutChains,
   const CCState &CCInfo = CC.getCCInfo();
   unsigned Idx = CCInfo.getFirstUnallocated(ArgRegs, NumRegs);
   unsigned RegSize = CC.regSize();
-  EVT RegTy = MVT::getIntegerVT(RegSize * 8);
+  MVT RegTy = MVT::getIntegerVT(RegSize * 8);
   const TargetRegisterClass *RC = getRegClassFor(RegTy);
   MachineFunction &MF = DAG.getMachineFunction();
   MachineFrameInfo *MFI = MF.getFrameInfo();
@@ -3780,7 +3584,7 @@ MipsTargetLowering::writeVarArgRegs(std::vector<SDValue> &OutChains,
   // in the caller's stack frame, while for N32/64, it is allocated in the
   // callee's stack frame.
   for (unsigned I = Idx; I < NumRegs; ++I, VaArgOffset += RegSize) {
-    unsigned Reg = AddLiveIn(MF, ArgRegs[I], RC);
+    unsigned Reg = addLiveIn(MF, ArgRegs[I], RC);
     SDValue ArgValue = DAG.getCopyFromReg(Chain, DL, Reg, RegTy);
     FI = MFI->CreateFixedObject(RegSize, VaArgOffset, true);
     SDValue PtrOff = DAG.getFrameIndex(FI, getPointerTy());
diff --git a/lib/Target/Mips/MipsISelLowering.h b/lib/Target/Mips/MipsISelLowering.h
index 43f97e89a7bf..cab71a61e07a 100644
--- a/lib/Target/Mips/MipsISelLowering.h
+++ b/lib/Target/Mips/MipsISelLowering.h
@@ -19,7 +19,10 @@
 #include "MipsSubtarget.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Target/TargetLowering.h"
+#include <deque>
+#include <string>
 
 namespace llvm {
   namespace MipsISD {
@@ -63,6 +66,18 @@ namespace llvm {
       // Return
       Ret,
 
+      EH_RETURN,
+
+      // Node used to extract integer from accumulator.
+      ExtractLOHI,
+
+      // Node used to insert integers to accumulator.
+      InsertLOHI,
+
+      // Mult nodes.
+      Mult,
+      Multu,
+
       // MAdd/Sub nodes
       MAdd,
       MAddu,
@@ -72,6 +87,8 @@ namespace llvm {
       // DivRem(u)
       DivRem,
       DivRemU,
+      DivRem16,
+      DivRemU16,
 
       BuildPairF64,
       ExtractElementF64,
@@ -147,9 +164,9 @@ namespace llvm {
   public:
     explicit MipsTargetLowering(MipsTargetMachine &TM);
 
-    virtual MVT getShiftAmountTy(EVT LHSTy) const { return MVT::i32; }
+    static const MipsTargetLowering *create(MipsTargetMachine &TM);
 
-    virtual bool allowsUnalignedMemoryAccesses (EVT VT) const;
+    virtual MVT getScalarShiftAmountTy(EVT LHSTy) const { return MVT::i32; }
 
     virtual void LowerOperationWrapper(SDNode *N,
                                        SmallVectorImpl<SDValue> &Results,
@@ -172,7 +189,34 @@ namespace llvm {
     EVT getSetCCResultType(EVT VT) const;
 
     virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
-  private:
+
+    virtual MachineBasicBlock *
+    EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *MBB) const;
+
+    struct LTStr {
+      bool operator()(const char *S1, const char *S2) const {
+        return strcmp(S1, S2) < 0;
+      }
+    };
+
+  protected:
+    SDValue getGlobalReg(SelectionDAG &DAG, EVT Ty) const;
+
+    SDValue getAddrLocal(SDValue Op, SelectionDAG &DAG, bool HasMips64) const;
+
+    SDValue getAddrGlobal(SDValue Op, SelectionDAG &DAG, unsigned Flag) const;
+
+    SDValue getAddrGlobalLargeGOT(SDValue Op, SelectionDAG &DAG,
+                                  unsigned HiFlag, unsigned LoFlag) const;
+
+    /// This function fills Ops, which is the list of operands that will later
+    /// be used when a function call node is created. It also generates
+    /// copyToReg nodes to set up argument registers.
+    virtual void
+    getOpndList(SmallVectorImpl<SDValue> &Ops,
+                std::deque< std::pair<unsigned, SDValue> > &RegsToPass,
+                bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage,
+                CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const;
 
     /// ByValArgInfo - Byval argument information.
     struct ByValArgInfo {
@@ -187,53 +231,80 @@ namespace llvm {
     /// arguments and inquire about calling convention information.
     class MipsCC {
     public:
-      MipsCC(CallingConv::ID CallConv, bool IsVarArg, bool IsO32,
-             CCState &Info);
+      MipsCC(CallingConv::ID CallConv, bool IsO32, CCState &Info);
 
-      void analyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Outs);
-      void analyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Ins);
-      void handleByValArg(unsigned ValNo, MVT ValVT, MVT LocVT,
-                          CCValAssign::LocInfo LocInfo,
-                          ISD::ArgFlagsTy ArgFlags);
+      void analyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Outs,
+                               bool IsVarArg, bool IsSoftFloat,
+                               const SDNode *CallNode,
+                               std::vector<ArgListEntry> &FuncArgs);
+      void analyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Ins,
+                                  bool IsSoftFloat,
+                                  Function::const_arg_iterator FuncArg);
+
+      void analyzeCallResult(const SmallVectorImpl<ISD::InputArg> &Ins,
+                             bool IsSoftFloat, const SDNode *CallNode,
+                             const Type *RetTy) const;
+
+      void analyzeReturn(const SmallVectorImpl<ISD::OutputArg> &Outs,
+                         bool IsSoftFloat, const Type *RetTy) const;
 
       const CCState &getCCInfo() const { return CCInfo; }
 
       /// hasByValArg - Returns true if function has byval arguments.
       bool hasByValArg() const { return !ByValArgs.empty(); }
 
-      /// useRegsForByval - Returns true if the calling convention allows the
-      /// use of registers to pass byval arguments.
-      bool useRegsForByval() const { return UseRegsForByval; }
-
       /// regSize - Size (in number of bits) of integer registers.
-      unsigned regSize() const { return RegSize; }
+      unsigned regSize() const { return IsO32 ? 4 : 8; }
 
       /// numIntArgRegs - Number of integer registers available for calls.
-      unsigned numIntArgRegs() const { return NumIntArgRegs; }
+      unsigned numIntArgRegs() const;
 
       /// reservedArgArea - The size of the area the caller reserves for
       /// register arguments. This is 16-byte if ABI is O32.
-      unsigned reservedArgArea() const { return ReservedArgArea; }
+      unsigned reservedArgArea() const;
 
-      /// intArgRegs - Pointer to array of integer registers.
-      const uint16_t *intArgRegs() const { return IntArgRegs; }
+      /// Return pointer to array of integer argument registers.
+      const uint16_t *intArgRegs() const;
 
       typedef SmallVector<ByValArgInfo, 2>::const_iterator byval_iterator;
       byval_iterator byval_begin() const { return ByValArgs.begin(); }
       byval_iterator byval_end() const { return ByValArgs.end(); }
 
     private:
+      void handleByValArg(unsigned ValNo, MVT ValVT, MVT LocVT,
+                          CCValAssign::LocInfo LocInfo,
+                          ISD::ArgFlagsTy ArgFlags);
+
+      /// useRegsForByval - Returns true if the calling convention allows the
+      /// use of registers to pass byval arguments.
+      bool useRegsForByval() const { return CallConv != CallingConv::Fast; }
+
+      /// Return the function that analyzes fixed argument list functions.
+      llvm::CCAssignFn *fixedArgFn() const;
+
+      /// Return the function that analyzes variable argument list functions.
+      llvm::CCAssignFn *varArgFn() const;
+
+      const uint16_t *shadowRegs() const;
+
       void allocateRegs(ByValArgInfo &ByVal, unsigned ByValSize,
                         unsigned Align);
 
+      /// Return the type of the register which is used to pass an argument or
+      /// return a value. This function returns f64 if the argument is an i64
+      /// value which has been generated as a result of softening an f128 value.
+      /// Otherwise, it just returns VT.
+      MVT getRegVT(MVT VT, const Type *OrigTy, const SDNode *CallNode,
+                   bool IsSoftFloat) const;
+
+      template<typename Ty>
+      void analyzeReturn(const SmallVectorImpl<Ty> &RetVals, bool IsSoftFloat,
+                         const SDNode *CallNode, const Type *RetTy) const;
+
       CCState &CCInfo;
-      bool UseRegsForByval;
-      unsigned RegSize;
-      unsigned NumIntArgRegs;
-      unsigned ReservedArgArea;
-      const uint16_t *IntArgRegs, *ShadowRegs;
+      CallingConv::ID CallConv;
+      bool IsO32;
       SmallVector<ByValArgInfo, 2> ByValArgs;
-      llvm::CCAssignFn *FixedFn, *VarFn;
     };
 
     // Subtarget Info
@@ -241,44 +312,49 @@ namespace llvm {
 
     bool HasMips64, IsN64, IsO32;
 
+  private:
     // Lower Operand helpers
     SDValue LowerCallResult(SDValue Chain, SDValue InFlag,
                             CallingConv::ID CallConv, bool isVarArg,
                             const SmallVectorImpl<ISD::InputArg> &Ins,
                             DebugLoc dl, SelectionDAG &DAG,
-                            SmallVectorImpl<SDValue> &InVals) const;
+                            SmallVectorImpl<SDValue> &InVals,
+                            const SDNode *CallNode, const Type *RetTy) const;
 
     // Lower Operand specifics
-    SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerFABS(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerMEMBARRIER(SDValue Op, SelectionDAG& DAG) const;
-    SDValue LowerATOMIC_FENCE(SDValue Op, SelectionDAG& DAG) const;
-    SDValue LowerShiftLeftParts(SDValue Op, SelectionDAG& DAG) const;
-    SDValue LowerShiftRightParts(SDValue Op, SelectionDAG& DAG,
+    SDValue lowerBR_JT(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerJumpTable(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerSELECT(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerSETCC(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerVASTART(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerFABS(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerEH_RETURN(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerMEMBARRIER(SDValue Op, SelectionDAG& DAG) const;
+    SDValue lowerATOMIC_FENCE(SDValue Op, SelectionDAG& DAG) const;
+    SDValue lowerShiftLeftParts(SDValue Op, SelectionDAG& DAG) const;
+    SDValue lowerShiftRightParts(SDValue Op, SelectionDAG& DAG,
                                  bool IsSRA) const;
-    SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerADD(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerLOAD(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerSTORE(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerADD(SDValue Op, SelectionDAG &DAG) const;
 
-    /// IsEligibleForTailCallOptimization - Check whether the call is eligible
+    /// isEligibleForTailCallOptimization - Check whether the call is eligible
     /// for tail call optimization.
-    bool IsEligibleForTailCallOptimization(const MipsCC &MipsCCInfo,
-                                           unsigned NextStackOffset,
-                                           const MipsFunctionInfo& FI) const;
+    virtual bool
+    isEligibleForTailCallOptimization(const MipsCC &MipsCCInfo,
+                                      unsigned NextStackOffset,
+                                      const MipsFunctionInfo& FI) const = 0;
 
     /// copyByValArg - Copy argument registers which were used to pass a byval
     /// argument to the stack. Create a stack frame object for the byval
@@ -292,7 +368,7 @@ namespace llvm {
 
     /// passByValArg - Pass a byval argument in registers or on stack.
     void passByValArg(SDValue Chain, DebugLoc DL,
-                      SmallVector<std::pair<unsigned, SDValue>, 16> &RegsToPass,
+                      std::deque< std::pair<unsigned, SDValue> > &RegsToPass,
                       SmallVector<SDValue, 8> &MemOpChains, SDValue StackPtr,
                       MachineFrameInfo *MFI, SelectionDAG &DAG, SDValue Arg,
                       const MipsCC &CC, const ByValArgInfo &ByVal,
@@ -332,10 +408,6 @@ namespace llvm {
                   const SmallVectorImpl<SDValue> &OutVals,
                   DebugLoc dl, SelectionDAG &DAG) const;
 
-    virtual MachineBasicBlock *
-      EmitInstrWithCustomInserter(MachineInstr *MI,
-                                  MachineBasicBlock *MBB) const;
-
     // Inline asm support
     ConstraintType getConstraintType(const std::string &Constraint) const;
 
@@ -357,10 +429,13 @@ namespace llvm {
                                               std::vector<SDValue> &Ops,
                                               SelectionDAG &DAG) const;
 
+    virtual bool isLegalAddressingMode(const AddrMode &AM, Type *Ty) const;
+
     virtual bool isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const;
 
     virtual EVT getOptimalMemOpType(uint64_t Size, unsigned DstAlign,
-                                    unsigned SrcAlign, bool IsZeroVal,
+                                    unsigned SrcAlign,
+                                    bool IsMemset, bool ZeroMemset,
                                     bool MemcpyStrSrc,
                                     MachineFunction &MF) const;
 
@@ -371,18 +446,20 @@ namespace llvm {
 
     virtual unsigned getJumpTableEncoding() const;
 
-    MachineBasicBlock *EmitBPOSGE32(MachineInstr *MI,
-                                    MachineBasicBlock *BB) const;
-    MachineBasicBlock *EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB,
+    MachineBasicBlock *emitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB,
                     unsigned Size, unsigned BinOpcode, bool Nand = false) const;
-    MachineBasicBlock *EmitAtomicBinaryPartword(MachineInstr *MI,
+    MachineBasicBlock *emitAtomicBinaryPartword(MachineInstr *MI,
                     MachineBasicBlock *BB, unsigned Size, unsigned BinOpcode,
                     bool Nand = false) const;
-    MachineBasicBlock *EmitAtomicCmpSwap(MachineInstr *MI,
+    MachineBasicBlock *emitAtomicCmpSwap(MachineInstr *MI,
                                   MachineBasicBlock *BB, unsigned Size) const;
-    MachineBasicBlock *EmitAtomicCmpSwapPartword(MachineInstr *MI,
+    MachineBasicBlock *emitAtomicCmpSwapPartword(MachineInstr *MI,
                                   MachineBasicBlock *BB, unsigned Size) const;
   };
+
+  /// Create MipsTargetLowering objects.
+  const MipsTargetLowering *createMips16TargetLowering(MipsTargetMachine &TM);
+  const MipsTargetLowering *createMipsSETargetLowering(MipsTargetMachine &TM);
 }
 
 #endif // MipsISELLOWERING_H
diff --git a/lib/Target/Mips/MipsInstrFPU.td b/lib/Target/Mips/MipsInstrFPU.td
index 33ee02068946..6b23057c9cdb 100644
--- a/lib/Target/Mips/MipsInstrFPU.td
+++ b/lib/Target/Mips/MipsInstrFPU.td
@@ -86,272 +86,320 @@ def fpimm0neg : PatLeaf<(fpimm), [{
 // Only S32 and D32 are supported right now.
 //===----------------------------------------------------------------------===//
 
-// FP load.
-let DecoderMethod = "DecodeFMem" in {
-class FPLoad<bits<6> op, string opstr, RegisterClass RC, Operand MemOpnd>:
-  FMem<op, (outs RC:$ft), (ins MemOpnd:$addr),
-      !strconcat(opstr, "\t$ft, $addr"), [(set RC:$ft, (load addr:$addr))],
-      IILoad>;
-
-// FP store.
-class FPStore<bits<6> op, string opstr, RegisterClass RC, Operand MemOpnd>:
-  FMem<op, (outs), (ins RC:$ft, MemOpnd:$addr),
-      !strconcat(opstr, "\t$ft, $addr"), [(store RC:$ft, addr:$addr)],
-      IIStore>;
-}
-// FP indexed load.
-class FPIdxLoad<bits<6> funct, string opstr, RegisterClass DRC,
-                RegisterClass PRC, SDPatternOperator FOp = null_frag>:
-  FFMemIdx<funct, (outs DRC:$fd), (ins PRC:$base, PRC:$index),
-           !strconcat(opstr, "\t$fd, ${index}(${base})"),
-           [(set DRC:$fd, (FOp (add PRC:$base, PRC:$index)))]> {
-  let fs = 0;
-}
-
-// FP indexed store.
-class FPIdxStore<bits<6> funct, string opstr, RegisterClass DRC,
-                 RegisterClass PRC, SDPatternOperator FOp= null_frag>:
-  FFMemIdx<funct, (outs), (ins DRC:$fs, PRC:$base, PRC:$index),
-           !strconcat(opstr, "\t$fs, ${index}(${base})"),
-           [(FOp DRC:$fs, (add PRC:$base, PRC:$index))]> {
-  let fd = 0;
-}
-
-// Instructions that convert an FP value to 32-bit fixed point.
-multiclass FFR1_W_M<bits<6> funct, string opstr> {
-  def _S   : FFR1<funct, 16, opstr, "w.s", FGR32, FGR32>;
-  def _D32 : FFR1<funct, 17, opstr, "w.d", FGR32, AFGR64>,
-             Requires<[NotFP64bit, HasStandardEncoding]>;
-  def _D64 : FFR1<funct, 17, opstr, "w.d", FGR32, FGR64>,
-             Requires<[IsFP64bit, HasStandardEncoding]> {
-    let DecoderNamespace = "Mips64";
+class ADDS_FT<string opstr, RegisterClass RC, InstrItinClass Itin, bit IsComm,
+              SDPatternOperator OpNode= null_frag> :
+  InstSE<(outs RC:$fd), (ins RC:$fs, RC:$ft),
+         !strconcat(opstr, "\t$fd, $fs, $ft"),
+         [(set RC:$fd, (OpNode RC:$fs, RC:$ft))], Itin, FrmFR> {
+  let isCommutable = IsComm;
+}
+
+multiclass ADDS_M<string opstr, InstrItinClass Itin, bit IsComm,
+                  SDPatternOperator OpNode = null_frag> {
+  def _D32 : ADDS_FT<opstr, AFGR64, Itin, IsComm, OpNode>,
+             Requires<[NotFP64bit, HasStdEnc]>;
+  def _D64 : ADDS_FT<opstr, FGR64, Itin, IsComm, OpNode>,
+             Requires<[IsFP64bit, HasStdEnc]> {
+    string DecoderNamespace = "Mips64";
   }
 }
 
-// Instructions that convert an FP value to 64-bit fixed point.
-let Predicates = [IsFP64bit, HasStandardEncoding], DecoderNamespace = "Mips64" in
-multiclass FFR1_L_M<bits<6> funct, string opstr> {
-  def _S   : FFR1<funct, 16, opstr, "l.s", FGR64, FGR32>;
-  def _D64 : FFR1<funct, 17, opstr, "l.d", FGR64, FGR64>;
-}
+class ABSS_FT<string opstr, RegisterClass DstRC, RegisterClass SrcRC,
+              InstrItinClass Itin, SDPatternOperator OpNode= null_frag> :
+  InstSE<(outs DstRC:$fd), (ins SrcRC:$fs), !strconcat(opstr, "\t$fd, $fs"),
+         [(set DstRC:$fd, (OpNode SrcRC:$fs))], Itin, FrmFR>,
+  NeverHasSideEffects;
 
-// FP-to-FP conversion instructions.
-multiclass FFR1P_M<bits<6> funct, string opstr, SDNode OpNode> {
-  def _S   : FFR1P<funct, 16, opstr, "s", FGR32, FGR32, OpNode>;
-  def _D32 : FFR1P<funct, 17, opstr, "d", AFGR64, AFGR64, OpNode>,
-             Requires<[NotFP64bit, HasStandardEncoding]>;
-  def _D64 : FFR1P<funct, 17, opstr, "d", FGR64, FGR64, OpNode>,
-             Requires<[IsFP64bit, HasStandardEncoding]> {
-    let DecoderNamespace = "Mips64";
+multiclass ABSS_M<string opstr, InstrItinClass Itin,
+                  SDPatternOperator OpNode= null_frag> {
+  def _D32 : ABSS_FT<opstr, AFGR64, AFGR64, Itin, OpNode>,
+             Requires<[NotFP64bit, HasStdEnc]>;
+  def _D64 : ABSS_FT<opstr, FGR64, FGR64, Itin, OpNode>,
+             Requires<[IsFP64bit, HasStdEnc]> {
+    string DecoderNamespace = "Mips64";
   }
 }
 
-multiclass FFR2P_M<bits<6> funct, string opstr, SDNode OpNode, bit isComm = 0> {
-  let isCommutable = isComm in {
-  def _S   : FFR2P<funct, 16, opstr, "s", FGR32, OpNode>;
-  def _D32 : FFR2P<funct, 17, opstr, "d", AFGR64, OpNode>,
-             Requires<[NotFP64bit, HasStandardEncoding]>;
-  def _D64 : FFR2P<funct, 17, opstr, "d", FGR64, OpNode>,
-             Requires<[IsFP64bit, HasStandardEncoding]> {
+multiclass ROUND_M<string opstr, InstrItinClass Itin> {
+  def _D32 : ABSS_FT<opstr, FGR32, AFGR64, Itin>,
+             Requires<[NotFP64bit, HasStdEnc]>;
+  def _D64 : ABSS_FT<opstr, FGR32, FGR64, Itin>,
+             Requires<[IsFP64bit, HasStdEnc]> {
     let DecoderNamespace = "Mips64";
   }
 }
-}
 
-// FP madd/msub/nmadd/nmsub instruction classes.
-class FMADDSUB<bits<3> funct, bits<3> fmt, string opstr, string fmtstr,
-               SDNode OpNode, RegisterClass RC> :
-  FFMADDSUB<funct, fmt, (outs RC:$fd), (ins RC:$fr, RC:$fs, RC:$ft),
-            !strconcat(opstr, ".", fmtstr, "\t$fd, $fr, $fs, $ft"),
-            [(set RC:$fd, (OpNode (fmul RC:$fs, RC:$ft), RC:$fr))]>;
-
-class FNMADDSUB<bits<3> funct, bits<3> fmt, string opstr, string fmtstr,
-                SDNode OpNode, RegisterClass RC> :
-  FFMADDSUB<funct, fmt, (outs RC:$fd), (ins RC:$fr, RC:$fs, RC:$ft),
-            !strconcat(opstr, ".", fmtstr, "\t$fd, $fr, $fs, $ft"),
-            [(set RC:$fd, (fsub fpimm0, (OpNode (fmul RC:$fs, RC:$ft), RC:$fr)))]>;
+class MFC1_FT<string opstr, RegisterClass DstRC, RegisterClass SrcRC,
+              InstrItinClass Itin, SDPatternOperator OpNode= null_frag> :
+  InstSE<(outs DstRC:$rt), (ins SrcRC:$fs), !strconcat(opstr, "\t$rt, $fs"),
+         [(set DstRC:$rt, (OpNode SrcRC:$fs))], Itin, FrmFR>;
+
+class MTC1_FT<string opstr, RegisterClass DstRC, RegisterClass SrcRC,
+              InstrItinClass Itin, SDPatternOperator OpNode= null_frag> :
+  InstSE<(outs DstRC:$fs), (ins SrcRC:$rt), !strconcat(opstr, "\t$rt, $fs"),
+         [(set DstRC:$fs, (OpNode SrcRC:$rt))], Itin, FrmFR>;
+
+class MFC1_FT_CCR<string opstr, RegisterClass DstRC, RegisterOperand SrcRC,
+              InstrItinClass Itin, SDPatternOperator OpNode= null_frag> :
+  InstSE<(outs DstRC:$rt), (ins SrcRC:$fs), !strconcat(opstr, "\t$rt, $fs"),
+         [(set DstRC:$rt, (OpNode SrcRC:$fs))], Itin, FrmFR>;
+
+class MTC1_FT_CCR<string opstr, RegisterOperand DstRC, RegisterClass SrcRC,
+              InstrItinClass Itin, SDPatternOperator OpNode= null_frag> :
+  InstSE<(outs DstRC:$fs), (ins SrcRC:$rt), !strconcat(opstr, "\t$rt, $fs"),
+         [(set DstRC:$fs, (OpNode SrcRC:$rt))], Itin, FrmFR>;
+
+class LW_FT<string opstr, RegisterClass RC, InstrItinClass Itin,
+            Operand MemOpnd, SDPatternOperator OpNode= null_frag> :
+  InstSE<(outs RC:$rt), (ins MemOpnd:$addr), !strconcat(opstr, "\t$rt, $addr"),
+         [(set RC:$rt, (OpNode addrDefault:$addr))], Itin, FrmFI> {
+  let DecoderMethod = "DecodeFMem";
+}
+
+class SW_FT<string opstr, RegisterClass RC, InstrItinClass Itin,
+            Operand MemOpnd, SDPatternOperator OpNode= null_frag> :
+  InstSE<(outs), (ins RC:$rt, MemOpnd:$addr), !strconcat(opstr, "\t$rt, $addr"),
+         [(OpNode RC:$rt, addrDefault:$addr)], Itin, FrmFI> {
+  let DecoderMethod = "DecodeFMem";
+}
+
+class MADDS_FT<string opstr, RegisterClass RC, InstrItinClass Itin,
+               SDPatternOperator OpNode = null_frag> :
+  InstSE<(outs RC:$fd), (ins RC:$fr, RC:$fs, RC:$ft),
+         !strconcat(opstr, "\t$fd, $fr, $fs, $ft"),
+         [(set RC:$fd, (OpNode (fmul RC:$fs, RC:$ft), RC:$fr))], Itin, FrmFR>;
+
+class NMADDS_FT<string opstr, RegisterClass RC, InstrItinClass Itin,
+                SDPatternOperator OpNode = null_frag> :
+  InstSE<(outs RC:$fd), (ins RC:$fr, RC:$fs, RC:$ft),
+         !strconcat(opstr, "\t$fd, $fr, $fs, $ft"),
+         [(set RC:$fd, (fsub fpimm0, (OpNode (fmul RC:$fs, RC:$ft), RC:$fr)))],
+         Itin, FrmFR>;
+
+class LWXC1_FT<string opstr, RegisterClass DRC, RegisterClass PRC,
+               InstrItinClass Itin, SDPatternOperator OpNode = null_frag> :
+  InstSE<(outs DRC:$fd), (ins PRC:$base, PRC:$index),
+         !strconcat(opstr, "\t$fd, ${index}(${base})"),
+         [(set DRC:$fd, (OpNode (add PRC:$base, PRC:$index)))], Itin, FrmFI> {
+  let AddedComplexity = 20;
+}
+
+class SWXC1_FT<string opstr, RegisterClass DRC, RegisterClass PRC,
+               InstrItinClass Itin, SDPatternOperator OpNode = null_frag> :
+  InstSE<(outs), (ins DRC:$fs, PRC:$base, PRC:$index),
+         !strconcat(opstr, "\t$fs, ${index}(${base})"),
+         [(OpNode DRC:$fs, (add PRC:$base, PRC:$index))], Itin, FrmFI> {
+  let AddedComplexity = 20;
+}
+
+class BC1F_FT<string opstr, InstrItinClass Itin,
+              SDPatternOperator Op = null_frag>  :
+  InstSE<(outs), (ins brtarget:$offset), !strconcat(opstr, "\t$offset"),
+         [(MipsFPBrcond Op, bb:$offset)], Itin, FrmFI> {
+  let isBranch = 1;
+  let isTerminator = 1;
+  let hasDelaySlot = 1;
+  let Defs = [AT];
+  let Uses = [FCR31];
+}
+
+class CEQS_FT<string typestr, RegisterClass RC, InstrItinClass Itin,
+              SDPatternOperator OpNode = null_frag>  :
+  InstSE<(outs), (ins RC:$fs, RC:$ft, condcode:$cond),
+         !strconcat("c.$cond.", typestr, "\t$fs, $ft"),
+         [(OpNode RC:$fs, RC:$ft, imm:$cond)], Itin, FrmFR> {
+  let Defs = [FCR31];
+}
 
 //===----------------------------------------------------------------------===//
 // Floating Point Instructions
 //===----------------------------------------------------------------------===//
-defm ROUND_W : FFR1_W_M<0xc, "round">;
-defm ROUND_L : FFR1_L_M<0x8, "round">;
-defm TRUNC_W : FFR1_W_M<0xd, "trunc">;
-defm TRUNC_L : FFR1_L_M<0x9, "trunc">;
-defm CEIL_W  : FFR1_W_M<0xe, "ceil">;
-defm CEIL_L  : FFR1_L_M<0xa, "ceil">;
-defm FLOOR_W : FFR1_W_M<0xf, "floor">;
-defm FLOOR_L : FFR1_L_M<0xb, "floor">;
-defm CVT_W   : FFR1_W_M<0x24, "cvt">, NeverHasSideEffects;
-//defm CVT_L   : FFR1_L_M<0x25, "cvt">;
-
-def CVT_S_W : FFR1<0x20, 20, "cvt", "s.w", FGR32, FGR32>, NeverHasSideEffects;
-def CVT_L_S : FFR1<0x25, 16, "cvt", "l.s", FGR64, FGR32>, NeverHasSideEffects;
-def CVT_L_D64: FFR1<0x25, 17, "cvt", "l.d", FGR64, FGR64>, NeverHasSideEffects;
-
-let Predicates = [NotFP64bit, HasStandardEncoding], neverHasSideEffects = 1 in {
-  def CVT_S_D32 : FFR1<0x20, 17, "cvt", "s.d", FGR32, AFGR64>;
-  def CVT_D32_W : FFR1<0x21, 20, "cvt", "d.w", AFGR64, FGR32>;
-  def CVT_D32_S : FFR1<0x21, 16, "cvt", "d.s", AFGR64, FGR32>;
-}
-
-let Predicates = [IsFP64bit, HasStandardEncoding], DecoderNamespace = "Mips64",
-    neverHasSideEffects = 1 in {
- def CVT_S_D64 : FFR1<0x20, 17, "cvt", "s.d", FGR32, FGR64>;
- def CVT_S_L   : FFR1<0x20, 21, "cvt", "s.l", FGR32, FGR64>;
- def CVT_D64_W : FFR1<0x21, 20, "cvt", "d.w", FGR64, FGR32>;
- def CVT_D64_S : FFR1<0x21, 16, "cvt", "d.s", FGR64, FGR32>;
- def CVT_D64_L : FFR1<0x21, 21, "cvt", "d.l", FGR64, FGR64>;
-}
-
-let Predicates = [NoNaNsFPMath, HasStandardEncoding] in {
-  defm FABS    : FFR1P_M<0x5, "abs",  fabs>;
-  defm FNEG    : FFR1P_M<0x7, "neg",  fneg>;
-}
-defm FSQRT   : FFR1P_M<0x4, "sqrt", fsqrt>;
+def ROUND_W_S  : ABSS_FT<"round.w.s", FGR32, FGR32, IIFcvt>, ABSS_FM<0xc, 16>;
+def TRUNC_W_S  : ABSS_FT<"trunc.w.s", FGR32, FGR32, IIFcvt>, ABSS_FM<0xd, 16>;
+def CEIL_W_S   : ABSS_FT<"ceil.w.s", FGR32, FGR32, IIFcvt>, ABSS_FM<0xe, 16>;
+def FLOOR_W_S  : ABSS_FT<"floor.w.s", FGR32, FGR32, IIFcvt>, ABSS_FM<0xf, 16>;
+def CVT_W_S    : ABSS_FT<"cvt.w.s", FGR32, FGR32, IIFcvt>, ABSS_FM<0x24, 16>;
+
+defm ROUND_W : ROUND_M<"round.w.d", IIFcvt>, ABSS_FM<0xc, 17>;
+defm TRUNC_W : ROUND_M<"trunc.w.d", IIFcvt>, ABSS_FM<0xd, 17>;
+defm CEIL_W  : ROUND_M<"ceil.w.d", IIFcvt>, ABSS_FM<0xe, 17>;
+defm FLOOR_W : ROUND_M<"floor.w.d", IIFcvt>, ABSS_FM<0xf, 17>;
+defm CVT_W   : ROUND_M<"cvt.w.d", IIFcvt>, ABSS_FM<0x24, 17>;
+
+let Predicates = [IsFP64bit, HasStdEnc], DecoderNamespace = "Mips64" in {
+  def ROUND_L_S : ABSS_FT<"round.l.s", FGR64, FGR32, IIFcvt>, ABSS_FM<0x8, 16>;
+  def ROUND_L_D64 : ABSS_FT<"round.l.d", FGR64, FGR64, IIFcvt>,
+                    ABSS_FM<0x8, 17>;
+  def TRUNC_L_S : ABSS_FT<"trunc.l.s", FGR64, FGR32, IIFcvt>, ABSS_FM<0x9, 16>;
+  def TRUNC_L_D64 : ABSS_FT<"trunc.l.d", FGR64, FGR64, IIFcvt>,
+                    ABSS_FM<0x9, 17>;
+  def CEIL_L_S  : ABSS_FT<"ceil.l.s", FGR64, FGR32, IIFcvt>, ABSS_FM<0xa, 16>;
+  def CEIL_L_D64 : ABSS_FT<"ceil.l.d", FGR64, FGR64, IIFcvt>, ABSS_FM<0xa, 17>;
+  def FLOOR_L_S : ABSS_FT<"floor.l.s", FGR64, FGR32, IIFcvt>, ABSS_FM<0xb, 16>;
+  def FLOOR_L_D64 : ABSS_FT<"floor.l.d", FGR64, FGR64, IIFcvt>,
+                    ABSS_FM<0xb, 17>;
+}
+
+def CVT_S_W : ABSS_FT<"cvt.s.w", FGR32, FGR32, IIFcvt>, ABSS_FM<0x20, 20>;
+def CVT_L_S : ABSS_FT<"cvt.l.s", FGR64, FGR32, IIFcvt>, ABSS_FM<0x25, 16>;
+def CVT_L_D64: ABSS_FT<"cvt.l.d", FGR64, FGR64, IIFcvt>, ABSS_FM<0x25, 17>;
+
+let Predicates = [NotFP64bit, HasStdEnc] in {
+  def CVT_S_D32 : ABSS_FT<"cvt.s.d", FGR32, AFGR64, IIFcvt>, ABSS_FM<0x20, 17>;
+  def CVT_D32_W : ABSS_FT<"cvt.d.w", AFGR64, FGR32, IIFcvt>, ABSS_FM<0x21, 20>;
+  def CVT_D32_S : ABSS_FT<"cvt.d.s", AFGR64, FGR32, IIFcvt>, ABSS_FM<0x21, 16>;
+}
+
+let Predicates = [IsFP64bit, HasStdEnc], DecoderNamespace = "Mips64" in {
+ def CVT_S_D64 : ABSS_FT<"cvt.s.d", FGR32, FGR64, IIFcvt>, ABSS_FM<0x20, 17>;
+ def CVT_S_L   : ABSS_FT<"cvt.s.l", FGR32, FGR64, IIFcvt>, ABSS_FM<0x20, 21>;
+ def CVT_D64_W : ABSS_FT<"cvt.d.w", FGR64, FGR32, IIFcvt>, ABSS_FM<0x21, 20>;
+ def CVT_D64_S : ABSS_FT<"cvt.d.s", FGR64, FGR32, IIFcvt>, ABSS_FM<0x21, 16>;
+ def CVT_D64_L : ABSS_FT<"cvt.d.l", FGR64, FGR64, IIFcvt>, ABSS_FM<0x21, 21>;
+}
+
+let Predicates = [NoNaNsFPMath, HasStdEnc] in {
+  def FABS_S : ABSS_FT<"abs.s", FGR32, FGR32, IIFcvt, fabs>, ABSS_FM<0x5, 16>;
+  def FNEG_S : ABSS_FT<"neg.s", FGR32, FGR32, IIFcvt, fneg>, ABSS_FM<0x7, 16>;
+  defm FABS : ABSS_M<"abs.d", IIFcvt, fabs>, ABSS_FM<0x5, 17>;
+  defm FNEG : ABSS_M<"neg.d", IIFcvt, fneg>, ABSS_FM<0x7, 17>;
+}
+
+def  FSQRT_S : ABSS_FT<"sqrt.s", FGR32, FGR32, IIFsqrtSingle, fsqrt>,
+               ABSS_FM<0x4, 16>;
+defm FSQRT : ABSS_M<"sqrt.d", IIFsqrtDouble, fsqrt>, ABSS_FM<0x4, 17>;
 
 // The odd-numbered registers are only referenced when doing loads,
 // stores, and moves between floating-point and integer registers.
 // When defining instructions, we reference all 32-bit registers,
 // regardless of register aliasing.
 
-class FFRGPR<bits<5> _fmt, dag outs, dag ins, string asmstr, list<dag> pattern>:
-             FFR<0x11, 0x0, _fmt, outs, ins, asmstr, pattern> {
-  bits<5> rt;
-  let ft = rt;
-  let fd = 0;
-}
-
 /// Move Control Registers From/To CPU Registers
-def CFC1  : FFRGPR<0x2, (outs CPURegs:$rt), (ins CCR:$fs),
-                  "cfc1\t$rt, $fs", []>;
-
-def CTC1  : FFRGPR<0x6, (outs CCR:$fs), (ins CPURegs:$rt),
-                  "ctc1\t$rt, $fs", []>;
-
-def MFC1  : FFRGPR<0x00, (outs CPURegs:$rt), (ins FGR32:$fs),
-                  "mfc1\t$rt, $fs",
-                  [(set CPURegs:$rt, (bitconvert FGR32:$fs))]>;
-
-def MTC1  : FFRGPR<0x04, (outs FGR32:$fs), (ins CPURegs:$rt),
-                  "mtc1\t$rt, $fs",
-                  [(set FGR32:$fs, (bitconvert CPURegs:$rt))]>;
-
-def DMFC1 : FFRGPR<0x01, (outs CPU64Regs:$rt), (ins FGR64:$fs),
-                  "dmfc1\t$rt, $fs",
-                  [(set CPU64Regs:$rt, (bitconvert FGR64:$fs))]>;
-
-def DMTC1 : FFRGPR<0x05, (outs FGR64:$fs), (ins CPU64Regs:$rt),
-                  "dmtc1\t$rt, $fs",
-                  [(set FGR64:$fs, (bitconvert CPU64Regs:$rt))]>;
-
-def FMOV_S   : FFR1<0x6, 16, "mov", "s", FGR32, FGR32>;
-def FMOV_D32 : FFR1<0x6, 17, "mov", "d", AFGR64, AFGR64>,
-               Requires<[NotFP64bit, HasStandardEncoding]>;
-def FMOV_D64 : FFR1<0x6, 17, "mov", "d", FGR64, FGR64>,
-               Requires<[IsFP64bit, HasStandardEncoding]> {
+def CFC1 : MFC1_FT_CCR<"cfc1", CPURegs, CCROpnd, IIFmove>, MFC1_FM<2>;
+def CTC1 : MTC1_FT_CCR<"ctc1", CCROpnd, CPURegs, IIFmove>, MFC1_FM<6>;
+def MFC1 : MFC1_FT<"mfc1", CPURegs, FGR32, IIFmove, bitconvert>, MFC1_FM<0>;
+def MTC1 : MTC1_FT<"mtc1", FGR32, CPURegs, IIFmove, bitconvert>, MFC1_FM<4>;
+def DMFC1 : MFC1_FT<"dmfc1", CPU64Regs, FGR64, IIFmove, bitconvert>, MFC1_FM<1>;
+def DMTC1 : MTC1_FT<"dmtc1", FGR64, CPU64Regs, IIFmove, bitconvert>, MFC1_FM<5>;
+
+def FMOV_S   : ABSS_FT<"mov.s", FGR32, FGR32, IIFmove>, ABSS_FM<0x6, 16>;
+def FMOV_D32 : ABSS_FT<"mov.d", AFGR64, AFGR64, IIFmove>, ABSS_FM<0x6, 17>,
+               Requires<[NotFP64bit, HasStdEnc]>;
+def FMOV_D64 : ABSS_FT<"mov.d", FGR64, FGR64, IIFmove>, ABSS_FM<0x6, 17>,
+               Requires<[IsFP64bit, HasStdEnc]> {
   let DecoderNamespace = "Mips64";
 }
 
 /// Floating Point Memory Instructions
-let Predicates = [IsN64, HasStandardEncoding], DecoderNamespace = "Mips64" in {
-  def LWC1_P8   : FPLoad<0x31, "lwc1", FGR32, mem64>;
-  def SWC1_P8   : FPStore<0x39, "swc1", FGR32, mem64>;
-  def LDC164_P8 : FPLoad<0x35, "ldc1", FGR64, mem64> {
+let Predicates = [IsN64, HasStdEnc], DecoderNamespace = "Mips64" in {
+  def LWC1_P8 : LW_FT<"lwc1", FGR32, IILoad, mem64, load>, LW_FM<0x31>;
+  def SWC1_P8 : SW_FT<"swc1", FGR32, IIStore, mem64, store>, LW_FM<0x39>;
+  def LDC164_P8 : LW_FT<"ldc1", FGR64, IILoad, mem64, load>, LW_FM<0x35> {
     let isCodeGenOnly =1;
   }
-  def SDC164_P8 : FPStore<0x3d, "sdc1", FGR64, mem64> {
+  def SDC164_P8 : SW_FT<"sdc1", FGR64, IIStore, mem64, store>, LW_FM<0x3d> {
     let isCodeGenOnly =1;
   }
 }
 
-let Predicates = [NotN64, HasStandardEncoding] in {
-  def LWC1   : FPLoad<0x31, "lwc1", FGR32, mem>;
-  def SWC1   : FPStore<0x39, "swc1", FGR32, mem>;
+let Predicates = [NotN64, HasStdEnc] in {
+  def LWC1 : LW_FT<"lwc1", FGR32, IILoad, mem, load>, LW_FM<0x31>;
+  def SWC1 : SW_FT<"swc1", FGR32, IIStore, mem, store>, LW_FM<0x39>;
 }
 
-let Predicates = [NotN64, HasMips64, HasStandardEncoding],
+let Predicates = [NotN64, HasMips64, HasStdEnc],
   DecoderNamespace = "Mips64" in {
-  def LDC164 : FPLoad<0x35, "ldc1", FGR64, mem>;
-  def SDC164 : FPStore<0x3d, "sdc1", FGR64, mem>;
+  def LDC164 : LW_FT<"ldc1", FGR64, IILoad, mem, load>, LW_FM<0x35>;
+  def SDC164 : SW_FT<"sdc1", FGR64, IIStore, mem, store>, LW_FM<0x3d>;
 }
 
-let Predicates = [NotN64, NotMips64, HasStandardEncoding] in {
-  def LDC1   : FPLoad<0x35, "ldc1", AFGR64, mem>;
-  def SDC1   : FPStore<0x3d, "sdc1", AFGR64, mem>;
+let Predicates = [NotN64, NotMips64, HasStdEnc] in {
+  def LDC1 : LW_FT<"ldc1", AFGR64, IILoad, mem, load>, LW_FM<0x35>;
+  def SDC1 : SW_FT<"sdc1", AFGR64, IIStore, mem, store>, LW_FM<0x3d>;
 }
 
 // Indexed loads and stores.
-let Predicates = [HasMips32r2Or64, HasStandardEncoding] in {
-  def LWXC1 : FPIdxLoad<0x0, "lwxc1", FGR32, CPURegs, load>;
-  def SWXC1 : FPIdxStore<0x8, "swxc1", FGR32, CPURegs, store>;
+let Predicates = [HasFPIdx, HasStdEnc] in {
+  def LWXC1 : LWXC1_FT<"lwxc1", FGR32, CPURegs, IILoad, load>, LWXC1_FM<0>;
+  def SWXC1 : SWXC1_FT<"swxc1", FGR32, CPURegs, IIStore, store>, SWXC1_FM<8>;
 }
 
-let Predicates = [HasMips32r2, NotMips64, HasStandardEncoding] in {
-  def LDXC1 : FPIdxLoad<0x1, "ldxc1", AFGR64, CPURegs, load>;
-  def SDXC1 : FPIdxStore<0x9, "sdxc1", AFGR64, CPURegs, store>;
+let Predicates = [HasMips32r2, NotMips64, HasStdEnc] in {
+  def LDXC1 : LWXC1_FT<"ldxc1", AFGR64, CPURegs, IILoad, load>, LWXC1_FM<1>;
+  def SDXC1 : SWXC1_FT<"sdxc1", AFGR64, CPURegs, IIStore, store>, SWXC1_FM<9>;
 }
 
-let Predicates = [HasMips64, NotN64, HasStandardEncoding], DecoderNamespace="Mips64" in {
-  def LDXC164 : FPIdxLoad<0x1, "ldxc1", FGR64, CPURegs, load>;
-  def SDXC164 : FPIdxStore<0x9, "sdxc1", FGR64, CPURegs, store>;
+let Predicates = [HasMips64, NotN64, HasStdEnc], DecoderNamespace="Mips64" in {
+  def LDXC164 : LWXC1_FT<"ldxc1", FGR64, CPURegs, IILoad, load>, LWXC1_FM<1>;
+  def SDXC164 : SWXC1_FT<"sdxc1", FGR64, CPURegs, IIStore, store>, SWXC1_FM<9>;
 }
 
 // n64
-let Predicates = [IsN64, HasStandardEncoding], isCodeGenOnly=1 in {
-  def LWXC1_P8   : FPIdxLoad<0x0, "lwxc1", FGR32, CPU64Regs, load>;
-  def LDXC164_P8 : FPIdxLoad<0x1, "ldxc1", FGR64, CPU64Regs, load>;
-  def SWXC1_P8   : FPIdxStore<0x8, "swxc1", FGR32, CPU64Regs, store>;
-  def SDXC164_P8 : FPIdxStore<0x9, "sdxc1", FGR64, CPU64Regs, store>;
+let Predicates = [IsN64, HasStdEnc], isCodeGenOnly=1 in {
+  def LWXC1_P8 : LWXC1_FT<"lwxc1", FGR32, CPU64Regs, IILoad, load>, LWXC1_FM<0>;
+  def LDXC164_P8 : LWXC1_FT<"ldxc1", FGR64, CPU64Regs, IILoad, load>,
+                   LWXC1_FM<1>;
+  def SWXC1_P8 : SWXC1_FT<"swxc1", FGR32, CPU64Regs, IIStore, store>,
+                 SWXC1_FM<8>;
+  def SDXC164_P8 : SWXC1_FT<"sdxc1", FGR64, CPU64Regs, IIStore, store>,
+                   SWXC1_FM<9>;
 }
 
 // Load/store doubleword indexed unaligned.
-let Predicates = [NotMips64, HasStandardEncoding] in {
-  def LUXC1 : FPIdxLoad<0x5, "luxc1", AFGR64, CPURegs>;
-  def SUXC1 : FPIdxStore<0xd, "suxc1", AFGR64, CPURegs>;
+let Predicates = [NotMips64, HasStdEnc] in {
+  def LUXC1 : LWXC1_FT<"luxc1", AFGR64, CPURegs, IILoad>, LWXC1_FM<0x5>;
+  def SUXC1 : SWXC1_FT<"suxc1", AFGR64, CPURegs, IIStore>, SWXC1_FM<0xd>;
 }
 
-let Predicates = [HasMips64, HasStandardEncoding],
+let Predicates = [HasMips64, HasStdEnc],
   DecoderNamespace="Mips64" in {
-  def LUXC164 : FPIdxLoad<0x5, "luxc1", FGR64, CPURegs>;
-  def SUXC164 : FPIdxStore<0xd, "suxc1", FGR64, CPURegs>;
+  def LUXC164 : LWXC1_FT<"luxc1", FGR64, CPURegs, IILoad>, LWXC1_FM<0x5>;
+  def SUXC164 : SWXC1_FT<"suxc1", FGR64, CPURegs, IIStore>, SWXC1_FM<0xd>;
 }
 
 /// Floating-point Aritmetic
-defm FADD : FFR2P_M<0x00, "add", fadd, 1>;
-defm FDIV : FFR2P_M<0x03, "div", fdiv>;
-defm FMUL : FFR2P_M<0x02, "mul", fmul, 1>;
-defm FSUB : FFR2P_M<0x01, "sub", fsub>;
+def FADD_S : ADDS_FT<"add.s", FGR32, IIFadd, 1, fadd>, ADDS_FM<0x00, 16>;
+defm FADD : ADDS_M<"add.d", IIFadd, 1, fadd>, ADDS_FM<0x00, 17>;
+def FDIV_S : ADDS_FT<"div.s", FGR32, IIFdivSingle, 0, fdiv>, ADDS_FM<0x03, 16>;
+defm FDIV : ADDS_M<"div.d", IIFdivDouble, 0, fdiv>, ADDS_FM<0x03, 17>;
+def FMUL_S : ADDS_FT<"mul.s", FGR32, IIFmulSingle, 1, fmul>, ADDS_FM<0x02, 16>;
+defm FMUL : ADDS_M<"mul.d", IIFmulDouble, 1, fmul>, ADDS_FM<0x02, 17>;
+def FSUB_S : ADDS_FT<"sub.s", FGR32, IIFadd, 0, fsub>, ADDS_FM<0x01, 16>;
+defm FSUB : ADDS_M<"sub.d", IIFadd, 0, fsub>, ADDS_FM<0x01, 17>;
 
-let Predicates = [HasMips32r2, HasStandardEncoding] in {
-  def MADD_S : FMADDSUB<0x4, 0, "madd", "s", fadd, FGR32>;
-  def MSUB_S : FMADDSUB<0x5, 0, "msub", "s", fsub, FGR32>;
+let Predicates = [HasMips32r2, HasStdEnc] in {
+  def MADD_S : MADDS_FT<"madd.s", FGR32, IIFmulSingle, fadd>, MADDS_FM<4, 0>;
+  def MSUB_S : MADDS_FT<"msub.s", FGR32, IIFmulSingle, fsub>, MADDS_FM<5, 0>;
 }
 
-let Predicates = [HasMips32r2, NoNaNsFPMath, HasStandardEncoding] in {
-  def NMADD_S : FNMADDSUB<0x6, 0, "nmadd", "s", fadd, FGR32>;
-  def NMSUB_S : FNMADDSUB<0x7, 0, "nmsub", "s", fsub, FGR32>;
+let Predicates = [HasMips32r2, NoNaNsFPMath, HasStdEnc] in {
+  def NMADD_S : NMADDS_FT<"nmadd.s", FGR32, IIFmulSingle, fadd>, MADDS_FM<6, 0>;
+  def NMSUB_S : NMADDS_FT<"nmsub.s", FGR32, IIFmulSingle, fsub>, MADDS_FM<7, 0>;
 }
 
-let Predicates = [HasMips32r2, NotFP64bit, HasStandardEncoding] in {
-  def MADD_D32 : FMADDSUB<0x4, 1, "madd", "d", fadd, AFGR64>;
-  def MSUB_D32 : FMADDSUB<0x5, 1, "msub", "d", fsub, AFGR64>;
+let Predicates = [HasMips32r2, NotFP64bit, HasStdEnc] in {
+  def MADD_D32 : MADDS_FT<"madd.d", AFGR64, IIFmulDouble, fadd>, MADDS_FM<4, 1>;
+  def MSUB_D32 : MADDS_FT<"msub.d", AFGR64, IIFmulDouble, fsub>, MADDS_FM<5, 1>;
 }
 
-let Predicates = [HasMips32r2, NotFP64bit, NoNaNsFPMath, HasStandardEncoding] in {
-  def NMADD_D32 : FNMADDSUB<0x6, 1, "nmadd", "d", fadd, AFGR64>;
-  def NMSUB_D32 : FNMADDSUB<0x7, 1, "nmsub", "d", fsub, AFGR64>;
+let Predicates = [HasMips32r2, NotFP64bit, NoNaNsFPMath, HasStdEnc] in {
+  def NMADD_D32 : NMADDS_FT<"nmadd.d", AFGR64, IIFmulDouble, fadd>,
+                  MADDS_FM<6, 1>;
+  def NMSUB_D32 : NMADDS_FT<"nmsub.d", AFGR64, IIFmulDouble, fsub>,
+                  MADDS_FM<7, 1>;
 }
 
-let Predicates = [HasMips32r2, IsFP64bit, HasStandardEncoding], isCodeGenOnly=1 in {
-  def MADD_D64 : FMADDSUB<0x4, 1, "madd", "d", fadd, FGR64>;
-  def MSUB_D64 : FMADDSUB<0x5, 1, "msub", "d", fsub, FGR64>;
+let Predicates = [HasMips32r2, IsFP64bit, HasStdEnc], isCodeGenOnly=1 in {
+  def MADD_D64 : MADDS_FT<"madd.d", FGR64, IIFmulDouble, fadd>, MADDS_FM<4, 1>;
+  def MSUB_D64 : MADDS_FT<"msub.d", FGR64, IIFmulDouble, fsub>, MADDS_FM<5, 1>;
 }
 
-let Predicates = [HasMips32r2, IsFP64bit, NoNaNsFPMath, HasStandardEncoding],
+let Predicates = [HasMips32r2, IsFP64bit, NoNaNsFPMath, HasStdEnc],
     isCodeGenOnly=1 in {
-  def NMADD_D64 : FNMADDSUB<0x6, 1, "nmadd", "d", fadd, FGR64>;
-  def NMSUB_D64 : FNMADDSUB<0x7, 1, "nmsub", "d", fsub, FGR64>;
+  def NMADD_D64 : NMADDS_FT<"nmadd.d", FGR64, IIFmulDouble, fadd>,
+                  MADDS_FM<6, 1>;
+  def NMSUB_D64 : NMADDS_FT<"nmsub.d", FGR64, IIFmulDouble, fsub>,
+                  MADDS_FM<7, 1>;
 }
 
 //===----------------------------------------------------------------------===//
@@ -362,19 +410,9 @@ let Predicates = [HasMips32r2, IsFP64bit, NoNaNsFPMath, HasStandardEncoding],
 def MIPS_BRANCH_F  : PatLeaf<(i32 0)>;
 def MIPS_BRANCH_T  : PatLeaf<(i32 1)>;
 
-/// Floating Point Branch of False/True (Likely)
-let isBranch=1, isTerminator=1, hasDelaySlot=1, base=0x8, Uses=[FCR31] in
-  class FBRANCH<bits<1> nd, bits<1> tf, PatLeaf op, string asmstr> :
-      FFI<0x11, (outs), (ins brtarget:$dst), !strconcat(asmstr, "\t$dst"),
-        [(MipsFPBrcond op, bb:$dst)]> {
-  let Inst{20-18} = 0;
-  let Inst{17} = nd;
-  let Inst{16} = tf;
-}
-
 let DecoderMethod = "DecodeBC1" in {
-def BC1F  : FBRANCH<0, 0, MIPS_BRANCH_F,  "bc1f">;
-def BC1T  : FBRANCH<0, 1, MIPS_BRANCH_T,  "bc1t">;
+def BC1F : BC1F_FT<"bc1f", IIBranch, MIPS_BRANCH_F>, BC1F_FM<0, 0>;
+def BC1T : BC1F_FT<"bc1t", IIBranch, MIPS_BRANCH_T>, BC1F_FM<0, 1>;
 }
 //===----------------------------------------------------------------------===//
 // Floating Point Flag Conditions
@@ -398,33 +436,24 @@ def MIPS_FCOND_NGE  : PatLeaf<(i32 13)>;
 def MIPS_FCOND_LE   : PatLeaf<(i32 14)>;
 def MIPS_FCOND_NGT  : PatLeaf<(i32 15)>;
 
-class FCMP<bits<5> fmt, RegisterClass RC, string typestr> :
-  FCC<fmt, (outs), (ins RC:$fs, RC:$ft, condcode:$cc),
-      !strconcat("c.$cc.", typestr, "\t$fs, $ft"),
-      [(MipsFPCmp RC:$fs, RC:$ft, imm:$cc)]>;
-
 /// Floating Point Compare
-let Defs=[FCR31] in {
-  def FCMP_S32 : FCMP<0x10, FGR32, "s">;
-  def FCMP_D32 : FCMP<0x11, AFGR64, "d">,
-      Requires<[NotFP64bit, HasStandardEncoding]>;
-  def FCMP_D64 : FCMP<0x11, FGR64, "d">,
-      Requires<[IsFP64bit, HasStandardEncoding]> {
-    let DecoderNamespace = "Mips64";
-  }
-}
+def FCMP_S32 : CEQS_FT<"s", FGR32, IIFcmp, MipsFPCmp>, CEQS_FM<16>;
+def FCMP_D32 : CEQS_FT<"d", AFGR64, IIFcmp, MipsFPCmp>, CEQS_FM<17>,
+               Requires<[NotFP64bit, HasStdEnc]>;
+let DecoderNamespace = "Mips64" in
+def FCMP_D64 : CEQS_FT<"d", FGR64, IIFcmp, MipsFPCmp>, CEQS_FM<17>,
+               Requires<[IsFP64bit, HasStdEnc]>;
 
 //===----------------------------------------------------------------------===//
 // Floating Point Pseudo-Instructions
 //===----------------------------------------------------------------------===//
-def MOVCCRToCCR : PseudoSE<(outs CCR:$dst), (ins CCR:$src),
-                           "# MOVCCRToCCR", []>;
+def MOVCCRToCCR : PseudoSE<(outs CCR:$dst), (ins CCROpnd:$src), []>;
 
 // This pseudo instr gets expanded into 2 mtc1 instrs after register
 // allocation.
 def BuildPairF64 :
   PseudoSE<(outs AFGR64:$dst),
-           (ins CPURegs:$lo, CPURegs:$hi), "",
+           (ins CPURegs:$lo, CPURegs:$hi),
            [(set AFGR64:$dst, (MipsBuildPairF64 CPURegs:$lo, CPURegs:$hi))]>;
 
 // This pseudo instr gets expanded into 2 mfc1 instrs after register
@@ -432,7 +461,7 @@ def BuildPairF64 :
 // if n is 0, lower part of src is extracted.
 // if n is 1, higher part of src is extracted.
 def ExtractElementF64 :
-  PseudoSE<(outs CPURegs:$dst), (ins AFGR64:$src, i32imm:$n), "",
+  PseudoSE<(outs CPURegs:$dst), (ins AFGR64:$src, i32imm:$n),
            [(set CPURegs:$dst, (MipsExtractElementF64 AFGR64:$src, imm:$n))]>;
 
 //===----------------------------------------------------------------------===//
@@ -444,7 +473,7 @@ def : MipsPat<(f32 fpimm0neg), (FNEG_S (MTC1 ZERO))>;
 def : MipsPat<(f32 (sint_to_fp CPURegs:$src)), (CVT_S_W (MTC1 CPURegs:$src))>;
 def : MipsPat<(i32 (fp_to_sint FGR32:$src)), (MFC1 (TRUNC_W_S FGR32:$src))>;
 
-let Predicates = [NotFP64bit, HasStandardEncoding] in {
+let Predicates = [NotFP64bit, HasStdEnc] in {
   def : MipsPat<(f64 (sint_to_fp CPURegs:$src)),
                 (CVT_D32_W (MTC1 CPURegs:$src))>;
   def : MipsPat<(i32 (fp_to_sint AFGR64:$src)),
@@ -453,7 +482,7 @@ let Predicates = [NotFP64bit, HasStandardEncoding] in {
   def : MipsPat<(f64 (fextend FGR32:$src)), (CVT_D32_S FGR32:$src)>;
 }
 
-let Predicates = [IsFP64bit, HasStandardEncoding] in {
+let Predicates = [IsFP64bit, HasStdEnc] in {
   def : MipsPat<(f64 fpimm0), (DMTC1 ZERO_64)>;
   def : MipsPat<(f64 fpimm0neg), (FNEG_D64 (DMTC1 ZERO_64))>;
 
@@ -473,3 +502,28 @@ let Predicates = [IsFP64bit, HasStandardEncoding] in {
   def : MipsPat<(f32 (fround FGR64:$src)), (CVT_S_D64 FGR64:$src)>;
   def : MipsPat<(f64 (fextend FGR32:$src)), (CVT_D64_S FGR32:$src)>;
 }
+
+// Patterns for loads/stores with a reg+imm operand.
+let AddedComplexity = 40 in {
+  let Predicates = [IsN64, HasStdEnc] in {
+    def : LoadRegImmPat<LWC1_P8, f32, load>;
+    def : StoreRegImmPat<SWC1_P8, f32>;
+    def : LoadRegImmPat<LDC164_P8, f64, load>;
+    def : StoreRegImmPat<SDC164_P8, f64>;
+  }
+
+  let Predicates = [NotN64, HasStdEnc] in {
+    def : LoadRegImmPat<LWC1, f32, load>;
+    def : StoreRegImmPat<SWC1, f32>;
+  }
+
+  let Predicates = [NotN64, HasMips64, HasStdEnc] in {
+    def : LoadRegImmPat<LDC164, f64, load>;
+    def : StoreRegImmPat<SDC164, f64>;
+  }
+
+  let Predicates = [NotN64, NotMips64, HasStdEnc] in {
+    def : LoadRegImmPat<LDC1, f64, load>;
+    def : StoreRegImmPat<SDC1, f64>;
+  }
+}
diff --git a/lib/Target/Mips/MipsInstrFormats.td b/lib/Target/Mips/MipsInstrFormats.td
index 1ecbdc2474b3..ee432c875355 100644
--- a/lib/Target/Mips/MipsInstrFormats.td
+++ b/lib/Target/Mips/MipsInstrFormats.td
@@ -76,20 +76,22 @@ class MipsInst<dag outs, dag ins, string asmstr, list<dag> pattern,
 class InstSE<dag outs, dag ins, string asmstr, list<dag> pattern,
              InstrItinClass itin, Format f>:
   MipsInst<outs, ins, asmstr, pattern, itin, f> {
-  let Predicates = [HasStandardEncoding];
+  let Predicates = [HasStdEnc];
 }
 
 // Mips Pseudo Instructions Format
-class MipsPseudo<dag outs, dag ins, string asmstr, list<dag> pattern>:
-  MipsInst<outs, ins, asmstr, pattern, IIPseudo, Pseudo> {
+class MipsPseudo<dag outs, dag ins, list<dag> pattern,
+                 InstrItinClass itin = IIPseudo> :
+  MipsInst<outs, ins, "", pattern, itin, Pseudo> {
   let isCodeGenOnly = 1;
   let isPseudo = 1;
 }
 
 // Mips32/64 Pseudo Instruction Format
-class PseudoSE<dag outs, dag ins, string asmstr, list<dag> pattern>:
-  MipsPseudo<outs, ins, asmstr, pattern> {
-  let Predicates = [HasStandardEncoding];
+class PseudoSE<dag outs, dag ins, list<dag> pattern,
+               InstrItinClass itin = IIPseudo>:
+  MipsPseudo<outs, ins, pattern, itin> {
+  let Predicates = [HasStdEnc];
 }
 
 // Pseudo-instructions for alternate assembly syntax (never used by codegen).
@@ -161,30 +163,28 @@ class BranchBase<bits<6> op, dag outs, dag ins, string asmstr,
 // Format J instruction class in Mips : <|opcode|address|>
 //===----------------------------------------------------------------------===//
 
-class FJ<bits<6> op, dag outs, dag ins, string asmstr, list<dag> pattern,
-         InstrItinClass itin>: InstSE<outs, ins, asmstr, pattern, itin, FrmJ>
+class FJ<bits<6> op>
 {
-  bits<26> addr;
+  bits<26> target;
 
-  let Opcode = op;
+  bits<32> Inst;
 
-  let Inst{25-0} = addr;
+  let Inst{31-26} = op;
+  let Inst{25-0}  = target;
 }
 
- //===----------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
 // MFC instruction class in Mips : <|op|mf|rt|rd|0000000|sel|>
 //===----------------------------------------------------------------------===//
-class MFC3OP<bits<6> op, bits<5> _mfmt, dag outs, dag ins, string asmstr>:
-  InstSE<outs, ins, asmstr, [], NoItinerary, FrmFR>
+class MFC3OP_FM<bits<6> op, bits<5> mfmt>
 {
-  bits<5> mfmt;
   bits<5> rt;
   bits<5> rd;
   bits<3> sel;
 
-  let Opcode = op;
-  let mfmt = _mfmt;
+  bits<32> Inst;
 
+  let Inst{31-26} = op;
   let Inst{25-21} = mfmt;
   let Inst{20-16} = rt;
   let Inst{15-11} = rd;
@@ -192,6 +192,270 @@ class MFC3OP<bits<6> op, bits<5> _mfmt, dag outs, dag ins, string asmstr>:
   let Inst{2-0}   = sel;
 }
 
+class ADD_FM<bits<6> op, bits<6> funct> {
+  bits<5> rd;
+  bits<5> rs;
+  bits<5> rt;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = op;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = rt;
+  let Inst{15-11} = rd;
+  let Inst{10-6}  = 0;
+  let Inst{5-0}   = funct;
+}
+
+class ADDI_FM<bits<6> op> {
+  bits<5>  rs;
+  bits<5>  rt;
+  bits<16> imm16;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = op;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = rt;
+  let Inst{15-0}  = imm16;
+}
+
+class SRA_FM<bits<6> funct, bit rotate> {
+  bits<5> rd;
+  bits<5> rt;
+  bits<5> shamt;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0;
+  let Inst{25-22} = 0;
+  let Inst{21}    = rotate;
+  let Inst{20-16} = rt;
+  let Inst{15-11} = rd;
+  let Inst{10-6}  = shamt;
+  let Inst{5-0}   = funct;
+}
+
+class SRLV_FM<bits<6> funct, bit rotate> {
+  bits<5> rd;
+  bits<5> rt;
+  bits<5> rs;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = rt;
+  let Inst{15-11} = rd;
+  let Inst{10-7}  = 0;
+  let Inst{6}     = rotate;
+  let Inst{5-0}   = funct;
+}
+
+class BEQ_FM<bits<6> op> {
+  bits<5>  rs;
+  bits<5>  rt;
+  bits<16> offset;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = op;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = rt;
+  let Inst{15-0}  = offset;
+}
+
+class BGEZ_FM<bits<6> op, bits<5> funct> {
+  bits<5>  rs;
+  bits<16> offset;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = op;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = funct;
+  let Inst{15-0}  = offset;
+}
+
+class B_FM {
+  bits<16> offset;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 4;
+  let Inst{25-21} = 0;
+  let Inst{20-16} = 0;
+  let Inst{15-0}  = offset;
+}
+
+class SLTI_FM<bits<6> op> {
+  bits<5> rt;
+  bits<5> rs;
+  bits<16> imm16;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = op;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = rt;
+  let Inst{15-0}  = imm16;
+}
+
+class MFLO_FM<bits<6> funct> {
+  bits<5> rd;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0;
+  let Inst{25-16} = 0;
+  let Inst{15-11} = rd;
+  let Inst{10-6}  = 0;
+  let Inst{5-0}   = funct;
+}
+
+class MTLO_FM<bits<6> funct> {
+  bits<5> rs;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0;
+  let Inst{25-21} = rs;
+  let Inst{20-6}  = 0;
+  let Inst{5-0}   = funct;
+}
+
+class SEB_FM<bits<5> funct, bits<6> funct2> {
+  bits<5> rd;
+  bits<5> rt;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0x1f;
+  let Inst{25-21} = 0;
+  let Inst{20-16} = rt;
+  let Inst{15-11} = rd;
+  let Inst{10-6}  = funct;
+  let Inst{5-0}   = funct2;
+}
+
+class CLO_FM<bits<6> funct> {
+  bits<5> rd;
+  bits<5> rs;
+  bits<5> rt;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0x1c;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = rt;
+  let Inst{15-11} = rd;
+  let Inst{10-6}  = 0;
+  let Inst{5-0}   = funct;
+  let rt = rd;
+}
+
+class LUI_FM {
+  bits<5> rt;
+  bits<16> imm16;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0xf;
+  let Inst{25-21} = 0;
+  let Inst{20-16} = rt;
+  let Inst{15-0}  = imm16;
+}
+
+class JALR_FM {
+  bits<5> rd;
+  bits<5> rs;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = 0;
+  let Inst{15-11} = rd;
+  let Inst{10-6}  = 0;
+  let Inst{5-0}   = 9;
+}
+
+class BAL_FM {
+  bits<16> offset;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 1;
+  let Inst{25-21} = 0;
+  let Inst{20-16} = 0x11;
+  let Inst{15-0}  = offset;
+}
+
+class BGEZAL_FM<bits<5> funct> {
+  bits<5>  rs;
+  bits<16> offset;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 1;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = funct;
+  let Inst{15-0}  = offset;
+}
+
+class SYNC_FM {
+  bits<5> stype;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0;
+  let Inst{10-6}  = stype;
+  let Inst{5-0}   = 0xf;
+}
+
+class MULT_FM<bits<6> op, bits<6> funct> {
+  bits<5>  rs;
+  bits<5>  rt;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = op;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = rt;
+  let Inst{15-6}  = 0;
+  let Inst{5-0}   = funct;
+}
+
+class EXT_FM<bits<6> funct> {
+  bits<5> rt;
+  bits<5> rs;
+  bits<5> pos;
+  bits<5> size;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0x1f;
+  let Inst{25-21} = rs;
+  let Inst{20-16} = rt;
+  let Inst{15-11} = size;
+  let Inst{10-6}  = pos;
+  let Inst{5-0}   = funct;
+}
+
+class RDHWR_FM {
+  bits<5> rt;
+  bits<5> rd;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0x1f;
+  let Inst{25-21} = 0;
+  let Inst{20-16} = rt;
+  let Inst{15-11} = rd;
+  let Inst{10-6}  = 0;
+  let Inst{5-0}   = 0x3b;
+}
+
 //===----------------------------------------------------------------------===//
 //
 //  FLOATING POINT INSTRUCTION FORMATS
@@ -206,31 +470,6 @@ class MFC3OP<bits<6> op, bits<5> _mfmt, dag outs, dag ins, string asmstr>:
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
-// Format FR instruction class in Mips : <|opcode|fmt|ft|fs|fd|funct|>
-//===----------------------------------------------------------------------===//
-
-class FFR<bits<6> op, bits<6> _funct, bits<5> _fmt, dag outs, dag ins,
-          string asmstr, list<dag> pattern> :
-  InstSE<outs, ins, asmstr, pattern, NoItinerary, FrmFR>
-{
-  bits<5>  fd;
-  bits<5>  fs;
-  bits<5>  ft;
-  bits<5>  fmt;
-  bits<6>  funct;
-
-  let Opcode = op;
-  let funct  = _funct;
-  let fmt    = _fmt;
-
-  let Inst{25-21} = fmt;
-  let Inst{20-16} = ft;
-  let Inst{15-11} = fs;
-  let Inst{10-6}  = fd;
-  let Inst{5-0}   = funct;
-}
-
-//===----------------------------------------------------------------------===//
 // Format FI instruction class in Mips : <|opcode|base|ft|immediate|>
 //===----------------------------------------------------------------------===//
 
@@ -248,130 +487,179 @@ class FFI<bits<6> op, dag outs, dag ins, string asmstr, list<dag> pattern>:
   let Inst{15-0}  = imm16;
 }
 
-//===----------------------------------------------------------------------===//
-// Compare instruction class in Mips : <|010001|fmt|ft|fs|0000011|condcode|>
-//===----------------------------------------------------------------------===//
-
-class FCC<bits<5> _fmt, dag outs, dag ins, string asmstr, list<dag> pattern> :
-  InstSE<outs, ins, asmstr, pattern, NoItinerary, FrmOther>
-{
-  bits<5>  fs;
-  bits<5>  ft;
-  bits<4>  cc;
-  bits<5>  fmt;
+class ADDS_FM<bits<6> funct, bits<5> fmt> {
+  bits<5> fd;
+  bits<5> fs;
+  bits<5> ft;
 
-  let Opcode = 0x11;
-  let fmt    = _fmt;
+  bits<32> Inst;
 
+  let Inst{31-26} = 0x11;
   let Inst{25-21} = fmt;
   let Inst{20-16} = ft;
   let Inst{15-11} = fs;
-  let Inst{10-6}  = 0;
-  let Inst{5-4}   = 0b11;
-  let Inst{3-0}   = cc;
+  let Inst{10-6}  = fd;
+  let Inst{5-0}   = funct;
 }
 
+class ABSS_FM<bits<6> funct, bits<5> fmt> {
+  bits<5> fd;
+  bits<5> fs;
 
-class FCMOV<bits<1> _tf, dag outs, dag ins, string asmstr,
-            list<dag> pattern> :
-  InstSE<outs, ins, asmstr, pattern, NoItinerary, FrmOther>
-{
-  bits<5>  rd;
-  bits<5>  rs;
-  bits<3>  cc;
-  bits<1>  tf;
-
-  let Opcode = 0;
-  let tf = _tf;
+  bits<32> Inst;
 
-  let Inst{25-21} = rs;
-  let Inst{20-18} = cc;
-  let Inst{17} = 0;
-  let Inst{16} = tf;
-  let Inst{15-11} = rd;
-  let Inst{10-6}  = 0;
-  let Inst{5-0}   = 1;
+  let Inst{31-26} = 0x11;
+  let Inst{25-21} = fmt;
+  let Inst{20-16} = 0;
+  let Inst{15-11} = fs;
+  let Inst{10-6}  = fd;
+  let Inst{5-0}   = funct;
 }
 
-class FFCMOV<bits<5> _fmt, bits<1> _tf, dag outs, dag ins, string asmstr,
-             list<dag> pattern> :
-  InstSE<outs, ins, asmstr, pattern, NoItinerary, FrmOther>
-{
-  bits<5>  fd;
-  bits<5>  fs;
-  bits<3>  cc;
-  bits<5>  fmt;
-  bits<1>  tf;
+class MFC1_FM<bits<5> funct> {
+  bits<5> rt;
+  bits<5> fs;
 
-  let Opcode = 17;
-  let fmt = _fmt;
-  let tf = _tf;
+  bits<32> Inst;
 
-  let Inst{25-21} = fmt;
-  let Inst{20-18} = cc;
-  let Inst{17} = 0;
-  let Inst{16} = tf;
+  let Inst{31-26} = 0x11;
+  let Inst{25-21} = funct;
+  let Inst{20-16} = rt;
   let Inst{15-11} = fs;
-  let Inst{10-6}  = fd;
-  let Inst{5-0}   = 17;
+  let Inst{10-0}  = 0;
 }
 
-// FP unary instructions without patterns.
-class FFR1<bits<6> funct, bits<5> fmt, string opstr, string fmtstr,
-           RegisterClass DstRC, RegisterClass SrcRC> :
-  FFR<0x11, funct, fmt, (outs DstRC:$fd), (ins SrcRC:$fs),
-      !strconcat(opstr, ".", fmtstr, "\t$fd, $fs"), []> {
-  let ft = 0;
-}
+class LW_FM<bits<6> op> {
+  bits<5> rt;
+  bits<21> addr;
 
-// FP unary instructions with patterns.
-class FFR1P<bits<6> funct, bits<5> fmt, string opstr, string fmtstr,
-            RegisterClass DstRC, RegisterClass SrcRC, SDNode OpNode> :
-  FFR<0x11, funct, fmt, (outs DstRC:$fd), (ins SrcRC:$fs),
-      !strconcat(opstr, ".", fmtstr, "\t$fd, $fs"),
-      [(set DstRC:$fd, (OpNode SrcRC:$fs))]> {
-  let ft = 0;
-}
+  bits<32> Inst;
 
-class FFR2P<bits<6> funct, bits<5> fmt, string opstr,
-            string fmtstr, RegisterClass RC, SDNode OpNode> :
-  FFR<0x11, funct, fmt, (outs RC:$fd), (ins RC:$fs, RC:$ft),
-      !strconcat(opstr, ".", fmtstr, "\t$fd, $fs, $ft"),
-      [(set RC:$fd, (OpNode RC:$fs, RC:$ft))]>;
+  let Inst{31-26} = op;
+  let Inst{25-21} = addr{20-16};
+  let Inst{20-16} = rt;
+  let Inst{15-0}  = addr{15-0};
+}
 
-// Floating point madd/msub/nmadd/nmsub.
-class FFMADDSUB<bits<3> funct, bits<3> fmt, dag outs, dag ins, string asmstr,
-                list<dag> pattern>
-  : InstSE<outs, ins, asmstr, pattern, NoItinerary, FrmOther> {
+class MADDS_FM<bits<3> funct, bits<3> fmt> {
   bits<5> fd;
   bits<5> fr;
   bits<5> fs;
   bits<5> ft;
 
-  let Opcode = 0x13;
+  bits<32> Inst;
+
+  let Inst{31-26} = 0x13;
   let Inst{25-21} = fr;
   let Inst{20-16} = ft;
   let Inst{15-11} = fs;
-  let Inst{10-6} = fd;
-  let Inst{5-3} = funct;
-  let Inst{2-0} = fmt;
+  let Inst{10-6}  = fd;
+  let Inst{5-3}   = funct;
+  let Inst{2-0}   = fmt;
 }
 
-// FP indexed load/store instructions.
-class FFMemIdx<bits<6> funct, dag outs, dag ins, string asmstr,
-               list<dag> pattern> :
-  InstSE<outs, ins, asmstr, pattern, NoItinerary, FrmOther>
-{
-  bits<5>  base;
-  bits<5>  index;
-  bits<5>  fs;
-  bits<5>  fd;
+class LWXC1_FM<bits<6> funct> {
+  bits<5> fd;
+  bits<5> base;
+  bits<5> index;
 
-  let Opcode = 0x13;
+  bits<32> Inst;
 
+  let Inst{31-26} = 0x13;
+  let Inst{25-21} = base;
+  let Inst{20-16} = index;
+  let Inst{15-11} = 0;
+  let Inst{10-6}  = fd;
+  let Inst{5-0}   = funct;
+}
+
+class SWXC1_FM<bits<6> funct> {
+  bits<5> fs;
+  bits<5> base;
+  bits<5> index;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0x13;
   let Inst{25-21} = base;
   let Inst{20-16} = index;
   let Inst{15-11} = fs;
+  let Inst{10-6}  = 0;
+  let Inst{5-0}   = funct;
+}
+
+class BC1F_FM<bit nd, bit tf> {
+  bits<16> offset;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0x11;
+  let Inst{25-21} = 0x8;
+  let Inst{20-18} = 0; // cc
+  let Inst{17} = nd;
+  let Inst{16} = tf;
+  let Inst{15-0} = offset;
+}
+
+class CEQS_FM<bits<5> fmt> {
+  bits<5> fs;
+  bits<5> ft;
+  bits<4> cond;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0x11;
+  let Inst{25-21} = fmt;
+  let Inst{20-16} = ft;
+  let Inst{15-11} = fs;
+  let Inst{10-8} = 0; // cc
+  let Inst{7-4} = 0x3;
+  let Inst{3-0} = cond;
+}
+
+class CMov_I_F_FM<bits<6> funct, bits<5> fmt> {
+  bits<5> fd;
+  bits<5> fs;
+  bits<5> rt;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0x11;
+  let Inst{25-21} = fmt;
+  let Inst{20-16} = rt;
+  let Inst{15-11} = fs;
   let Inst{10-6} = fd;
   let Inst{5-0} = funct;
 }
+
+class CMov_F_I_FM<bit tf> {
+  bits<5> rd;
+  bits<5> rs;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0;
+  let Inst{25-21} = rs;
+  let Inst{20-18} = 0; // cc
+  let Inst{17} = 0;
+  let Inst{16} = tf;
+  let Inst{15-11} = rd;
+  let Inst{10-6} = 0;
+  let Inst{5-0} = 1;
+}
+
+class CMov_F_F_FM<bits<5> fmt, bit tf> {
+  bits<5> fd;
+  bits<5> fs;
+
+  bits<32> Inst;
+
+  let Inst{31-26} = 0x11;
+  let Inst{25-21} = fmt;
+  let Inst{20-18} = 0; // cc
+  let Inst{17} = 0;
+  let Inst{16} = tf;
+  let Inst{15-11} = fs;
+  let Inst{10-6} = fd;
+  let Inst{5-0} = 0x11;
+}
diff --git a/lib/Target/Mips/MipsInstrInfo.cpp b/lib/Target/Mips/MipsInstrInfo.cpp
index ca80d43f36f1..ad92d41209e9 100644
--- a/lib/Target/Mips/MipsInstrInfo.cpp
+++ b/lib/Target/Mips/MipsInstrInfo.cpp
@@ -11,16 +11,16 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "MipsAnalyzeImmediate.h"
 #include "MipsInstrInfo.h"
-#include "MipsTargetMachine.h"
-#include "MipsMachineFunction.h"
 #include "InstPrinter/MipsInstPrinter.h"
+#include "MipsAnalyzeImmediate.h"
+#include "MipsMachineFunction.h"
+#include "MipsTargetMachine.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
-#include "llvm/ADT/STLExtras.h"
 
 #define GET_INSTRINFO_CTOR
 #include "MipsGenInstrInfo.inc"
@@ -93,81 +93,11 @@ bool MipsInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
                                   MachineBasicBlock *&TBB,
                                   MachineBasicBlock *&FBB,
                                   SmallVectorImpl<MachineOperand> &Cond,
-                                  bool AllowModify) const
-{
-
-  MachineBasicBlock::reverse_iterator I = MBB.rbegin(), REnd = MBB.rend();
-
-  // Skip all the debug instructions.
-  while (I != REnd && I->isDebugValue())
-    ++I;
-
-  if (I == REnd || !isUnpredicatedTerminator(&*I)) {
-    // If this block ends with no branches (it just falls through to its succ)
-    // just return false, leaving TBB/FBB null.
-    TBB = FBB = NULL;
-    return false;
-  }
-
-  MachineInstr *LastInst = &*I;
-  unsigned LastOpc = LastInst->getOpcode();
-
-  // Not an analyzable branch (must be an indirect jump).
-  if (!GetAnalyzableBrOpc(LastOpc))
-    return true;
-
-  // Get the second to last instruction in the block.
-  unsigned SecondLastOpc = 0;
-  MachineInstr *SecondLastInst = NULL;
-
-  if (++I != REnd) {
-    SecondLastInst = &*I;
-    SecondLastOpc = GetAnalyzableBrOpc(SecondLastInst->getOpcode());
-
-    // Not an analyzable branch (must be an indirect jump).
-    if (isUnpredicatedTerminator(SecondLastInst) && !SecondLastOpc)
-      return true;
-  }
-
-  // If there is only one terminator instruction, process it.
-  if (!SecondLastOpc) {
-    // Unconditional branch
-    if (LastOpc == UncondBrOpc) {
-      TBB = LastInst->getOperand(0).getMBB();
-      return false;
-    }
-
-    // Conditional branch
-    AnalyzeCondBr(LastInst, LastOpc, TBB, Cond);
-    return false;
-  }
-
-  // If we reached here, there are two branches.
-  // If there are three terminators, we don't know what sort of block this is.
-  if (++I != REnd && isUnpredicatedTerminator(&*I))
-    return true;
-
-  // If second to last instruction is an unconditional branch,
-  // analyze it and remove the last instruction.
-  if (SecondLastOpc == UncondBrOpc) {
-    // Return if the last instruction cannot be removed.
-    if (!AllowModify)
-      return true;
-
-    TBB = SecondLastInst->getOperand(0).getMBB();
-    LastInst->eraseFromParent();
-    return false;
-  }
-
-  // Conditional branch followed by an unconditional branch.
-  // The last one must be unconditional.
-  if (LastOpc != UncondBrOpc)
-    return true;
-
-  AnalyzeCondBr(SecondLastInst, SecondLastOpc, TBB, Cond);
-  FBB = LastInst->getOperand(0).getMBB();
+                                  bool AllowModify) const {
+  SmallVector<MachineInstr*, 2> BranchInstrs;
+  BranchType BT = AnalyzeBranch(MBB, TBB, FBB, Cond, AllowModify, BranchInstrs);
 
-  return false;
+  return (BT == BT_None) || (BT == BT_Indirect);
 }
 
 void MipsInstrInfo::BuildCondBr(MachineBasicBlock &MBB,
@@ -256,6 +186,90 @@ ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const
   return false;
 }
 
+MipsInstrInfo::BranchType MipsInstrInfo::
+AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
+              MachineBasicBlock *&FBB, SmallVectorImpl<MachineOperand> &Cond,
+              bool AllowModify,
+              SmallVectorImpl<MachineInstr*> &BranchInstrs) const {
+
+  MachineBasicBlock::reverse_iterator I = MBB.rbegin(), REnd = MBB.rend();
+
+  // Skip all the debug instructions.
+  while (I != REnd && I->isDebugValue())
+    ++I;
+
+  if (I == REnd || !isUnpredicatedTerminator(&*I)) {
+    // This block ends with no branches (it just falls through to its succ).
+    // Leave TBB/FBB null.
+    TBB = FBB = NULL;
+    return BT_NoBranch;
+  }
+
+  MachineInstr *LastInst = &*I;
+  unsigned LastOpc = LastInst->getOpcode();
+  BranchInstrs.push_back(LastInst);
+
+  // Not an analyzable branch (e.g., indirect jump).
+  if (!GetAnalyzableBrOpc(LastOpc))
+    return LastInst->isIndirectBranch() ? BT_Indirect : BT_None;
+
+  // Get the second to last instruction in the block.
+  unsigned SecondLastOpc = 0;
+  MachineInstr *SecondLastInst = NULL;
+
+  if (++I != REnd) {
+    SecondLastInst = &*I;
+    SecondLastOpc = GetAnalyzableBrOpc(SecondLastInst->getOpcode());
+
+    // Not an analyzable branch (must be an indirect jump).
+    if (isUnpredicatedTerminator(SecondLastInst) && !SecondLastOpc)
+      return BT_None;
+  }
+
+  // If there is only one terminator instruction, process it.
+  if (!SecondLastOpc) {
+    // Unconditional branch
+    if (LastOpc == UncondBrOpc) {
+      TBB = LastInst->getOperand(0).getMBB();
+      return BT_Uncond;
+    }
+
+    // Conditional branch
+    AnalyzeCondBr(LastInst, LastOpc, TBB, Cond);
+    return BT_Cond;
+  }
+
+  // If we reached here, there are two branches.
+  // If there are three terminators, we don't know what sort of block this is.
+  if (++I != REnd && isUnpredicatedTerminator(&*I))
+    return BT_None;
+
+  BranchInstrs.insert(BranchInstrs.begin(), SecondLastInst);
+
+  // If second to last instruction is an unconditional branch,
+  // analyze it and remove the last instruction.
+  if (SecondLastOpc == UncondBrOpc) {
+    // Return if the last instruction cannot be removed.
+    if (!AllowModify)
+      return BT_None;
+
+    TBB = SecondLastInst->getOperand(0).getMBB();
+    LastInst->eraseFromParent();
+    BranchInstrs.pop_back();
+    return BT_Uncond;
+  }
+
+  // Conditional branch followed by an unconditional branch.
+  // The last one must be unconditional.
+  if (LastOpc != UncondBrOpc)
+    return BT_None;
+
+  AnalyzeCondBr(SecondLastInst, SecondLastOpc, TBB, Cond);
+  FBB = LastInst->getOperand(0).getMBB();
+
+  return BT_CondUncond;
+}
+
 /// Return the number of bytes of code the specified instruction may be.
 unsigned MipsInstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const {
   switch (MI->getOpcode()) {
diff --git a/lib/Target/Mips/MipsInstrInfo.h b/lib/Target/Mips/MipsInstrInfo.h
index aca2bc7ae98d..8c05d97beac2 100644
--- a/lib/Target/Mips/MipsInstrInfo.h
+++ b/lib/Target/Mips/MipsInstrInfo.h
@@ -31,6 +31,15 @@ protected:
   unsigned UncondBrOpc;
 
 public:
+  enum BranchType {
+    BT_None,       // Couldn't analyze branch.
+    BT_NoBranch,   // No branches found.
+    BT_Uncond,     // One unconditional branch.
+    BT_Cond,       // One conditional branch.
+    BT_CondUncond, // A conditional branch followed by an unconditional branch.
+    BT_Indirect    // One indirct branch.
+  };
+
   explicit MipsInstrInfo(MipsTargetMachine &TM, unsigned UncondBrOpc);
 
   static const MipsInstrInfo *create(MipsTargetMachine &TM);
@@ -51,6 +60,12 @@ public:
   virtual
   bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const;
 
+  BranchType AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
+                           MachineBasicBlock *&FBB,
+                           SmallVectorImpl<MachineOperand> &Cond,
+                           bool AllowModify,
+                           SmallVectorImpl<MachineInstr*> &BranchInstrs) const;
+
   virtual MachineInstr* emitFrameIndexDebugValue(MachineFunction &MF,
                                                  int FrameIx, uint64_t Offset,
                                                  const MDNode *MDPtr,
@@ -71,6 +86,36 @@ public:
   /// Return the number of bytes of code the specified instruction may be.
   unsigned GetInstSizeInBytes(const MachineInstr *MI) const;
 
+  virtual void storeRegToStackSlot(MachineBasicBlock &MBB,
+                                   MachineBasicBlock::iterator MBBI,
+                                   unsigned SrcReg, bool isKill, int FrameIndex,
+                                   const TargetRegisterClass *RC,
+                                   const TargetRegisterInfo *TRI) const {
+    storeRegToStack(MBB, MBBI, SrcReg, isKill, FrameIndex, RC, TRI, 0);
+  }
+
+  virtual void loadRegFromStackSlot(MachineBasicBlock &MBB,
+                                    MachineBasicBlock::iterator MBBI,
+                                    unsigned DestReg, int FrameIndex,
+                                    const TargetRegisterClass *RC,
+                                    const TargetRegisterInfo *TRI) const {
+    loadRegFromStack(MBB, MBBI, DestReg, FrameIndex, RC, TRI, 0);
+  }
+
+  virtual void storeRegToStack(MachineBasicBlock &MBB,
+                               MachineBasicBlock::iterator MI,
+                               unsigned SrcReg, bool isKill, int FrameIndex,
+                               const TargetRegisterClass *RC,
+                               const TargetRegisterInfo *TRI,
+                               int64_t Offset) const = 0;
+
+  virtual void loadRegFromStack(MachineBasicBlock &MBB,
+                                MachineBasicBlock::iterator MI,
+                                unsigned DestReg, int FrameIndex,
+                                const TargetRegisterClass *RC,
+                                const TargetRegisterInfo *TRI,
+                                int64_t Offset) const = 0;
+
 protected:
   bool isZeroImm(const MachineOperand &op) const;
 
diff --git a/lib/Target/Mips/MipsInstrInfo.td b/lib/Target/Mips/MipsInstrInfo.td
index aa8881997285..3a82e8171301 100644
--- a/lib/Target/Mips/MipsInstrInfo.td
+++ b/lib/Target/Mips/MipsInstrInfo.td
@@ -23,18 +23,19 @@ def SDT_MipsCMov         : SDTypeProfile<1, 4, [SDTCisSameAs<0, 1>,
                                                 SDTCisInt<4>]>;
 def SDT_MipsCallSeqStart : SDCallSeqStart<[SDTCisVT<0, i32>]>;
 def SDT_MipsCallSeqEnd   : SDCallSeqEnd<[SDTCisVT<0, i32>, SDTCisVT<1, i32>]>;
-def SDT_MipsMAddMSub     : SDTypeProfile<0, 4,
-                                         [SDTCisVT<0, i32>, SDTCisSameAs<0, 1>,
-                                          SDTCisSameAs<1, 2>,
-                                          SDTCisSameAs<2, 3>]>;
-def SDT_MipsDivRem       : SDTypeProfile<0, 2,
-                                         [SDTCisInt<0>,
-                                          SDTCisSameAs<0, 1>]>;
+def SDT_ExtractLOHI : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisVT<1, untyped>,
+                                           SDTCisVT<2, i32>]>;
+def SDT_InsertLOHI : SDTypeProfile<1, 2, [SDTCisVT<0, untyped>,
+                                          SDTCisVT<1, i32>, SDTCisSameAs<1, 2>]>;
+def SDT_MipsMultDiv : SDTypeProfile<1, 2, [SDTCisVT<0, untyped>, SDTCisInt<1>,
+                                    SDTCisSameAs<1, 2>]>;
+def SDT_MipsMAddMSub : SDTypeProfile<1, 3,
+                                     [SDTCisVT<0, untyped>, SDTCisSameAs<0, 3>,
+                                      SDTCisVT<1, i32>, SDTCisSameAs<1, 2>]>;
+def SDT_MipsDivRem16 : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>]>;
 
 def SDT_MipsThreadPointer : SDTypeProfile<1, 0, [SDTCisPtrTy<0>]>;
 
-def SDT_MipsDynAlloc    : SDTypeProfile<1, 1, [SDTCisVT<0, iPTR>,
-                                               SDTCisSameAs<0, 1>]>;
 def SDT_Sync             : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
 
 def SDT_Ext : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
@@ -74,7 +75,8 @@ def MipsTprelLo    : SDNode<"MipsISD::TprelLo", SDTIntUnaryOp>;
 def MipsThreadPointer: SDNode<"MipsISD::ThreadPointer", SDT_MipsThreadPointer>;
 
 // Return
-def MipsRet : SDNode<"MipsISD::Ret", SDTNone, [SDNPHasChain, SDNPOptInGlue]>;
+def MipsRet : SDNode<"MipsISD::Ret", SDTNone,
+                     [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
 
 // These are target-independent nodes, but have target-specific formats.
 def callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_MipsCallSeqStart,
@@ -83,20 +85,27 @@ def callseq_end   : SDNode<"ISD::CALLSEQ_END", SDT_MipsCallSeqEnd,
                            [SDNPHasChain, SDNPSideEffect,
                             SDNPOptInGlue, SDNPOutGlue]>;
 
+// Node used to extract integer from LO/HI register.
+def ExtractLOHI : SDNode<"MipsISD::ExtractLOHI", SDT_ExtractLOHI>;
+
+// Node used to insert 32-bit integers to LOHI register pair.
+def InsertLOHI : SDNode<"MipsISD::InsertLOHI", SDT_InsertLOHI>;
+
+// Mult nodes.
+def MipsMult  : SDNode<"MipsISD::Mult", SDT_MipsMultDiv>;
+def MipsMultu : SDNode<"MipsISD::Multu", SDT_MipsMultDiv>;
+
 // MAdd*/MSub* nodes
-def MipsMAdd      : SDNode<"MipsISD::MAdd", SDT_MipsMAddMSub,
-                           [SDNPOptInGlue, SDNPOutGlue]>;
-def MipsMAddu     : SDNode<"MipsISD::MAddu", SDT_MipsMAddMSub,
-                           [SDNPOptInGlue, SDNPOutGlue]>;
-def MipsMSub      : SDNode<"MipsISD::MSub", SDT_MipsMAddMSub,
-                           [SDNPOptInGlue, SDNPOutGlue]>;
-def MipsMSubu     : SDNode<"MipsISD::MSubu", SDT_MipsMAddMSub,
-                           [SDNPOptInGlue, SDNPOutGlue]>;
+def MipsMAdd  : SDNode<"MipsISD::MAdd", SDT_MipsMAddMSub>;
+def MipsMAddu : SDNode<"MipsISD::MAddu", SDT_MipsMAddMSub>;
+def MipsMSub  : SDNode<"MipsISD::MSub", SDT_MipsMAddMSub>;
+def MipsMSubu : SDNode<"MipsISD::MSubu", SDT_MipsMAddMSub>;
 
 // DivRem(u) nodes
-def MipsDivRem    : SDNode<"MipsISD::DivRem", SDT_MipsDivRem,
-                           [SDNPOutGlue]>;
-def MipsDivRemU   : SDNode<"MipsISD::DivRemU", SDT_MipsDivRem,
+def MipsDivRem    : SDNode<"MipsISD::DivRem", SDT_MipsMultDiv>;
+def MipsDivRemU   : SDNode<"MipsISD::DivRemU", SDT_MipsMultDiv>;
+def MipsDivRem16  : SDNode<"MipsISD::DivRem16", SDT_MipsDivRem16, [SDNPOutGlue]>;
+def MipsDivRemU16 : SDNode<"MipsISD::DivRemU16", SDT_MipsDivRem16,
                            [SDNPOutGlue]>;
 
 // Target constant nodes that are not part of any isel patterns and remain
@@ -111,10 +120,6 @@ def MipsDivRemU   : SDNode<"MipsISD::DivRemU", SDT_MipsDivRem,
 
 def MipsWrapper    : SDNode<"MipsISD::Wrapper", SDTIntBinOp>;
 
-// Pointer to dynamically allocated stack area.
-def MipsDynAlloc  : SDNode<"MipsISD::DynAlloc", SDT_MipsDynAlloc,
-                           [SDNPHasChain, SDNPInGlue]>;
-
 def MipsSync : SDNode<"MipsISD::Sync", SDT_Sync, [SDNPHasChain,SDNPSideEffect]>;
 
 def MipsExt :  SDNode<"MipsISD::Ext", SDT_Ext>;
@@ -148,14 +153,14 @@ def HasSwap     :     Predicate<"Subtarget.hasSwap()">,
                       AssemblerPredicate<"FeatureSwap">;
 def HasCondMov  :     Predicate<"Subtarget.hasCondMov()">,
                       AssemblerPredicate<"FeatureCondMov">;
+def HasFPIdx    :     Predicate<"Subtarget.hasFPIdx()">,
+                      AssemblerPredicate<"FeatureFPIdx">;
 def HasMips32    :    Predicate<"Subtarget.hasMips32()">,
                       AssemblerPredicate<"FeatureMips32">;
 def HasMips32r2  :    Predicate<"Subtarget.hasMips32r2()">,
                       AssemblerPredicate<"FeatureMips32r2">;
 def HasMips64    :    Predicate<"Subtarget.hasMips64()">,
                       AssemblerPredicate<"FeatureMips64">;
-def HasMips32r2Or64 : Predicate<"Subtarget.hasMips32r2Or64()">,
-                      AssemblerPredicate<"FeatureMips32r2,FeatureMips64">;
 def NotMips64    :    Predicate<"!Subtarget.hasMips64()">,
                       AssemblerPredicate<"!FeatureMips64">;
 def HasMips64r2  :    Predicate<"Subtarget.hasMips64r2()">,
@@ -172,11 +177,15 @@ def RelocPIC    :     Predicate<"TM.getRelocationModel() == Reloc::PIC_">,
                       AssemblerPredicate<"FeatureMips32">;
 def NoNaNsFPMath :    Predicate<"TM.Options.NoNaNsFPMath">,
                       AssemblerPredicate<"FeatureMips32">;
-def HasStandardEncoding : Predicate<"Subtarget.hasStandardEncoding()">,
-                          AssemblerPredicate<"!FeatureMips16">;
+def HasStdEnc :       Predicate<"Subtarget.hasStandardEncoding()">,
+                      AssemblerPredicate<"!FeatureMips16">;
 
 class MipsPat<dag pattern, dag result> : Pat<pattern, result> {
-  let Predicates = [HasStandardEncoding];
+  let Predicates = [HasStdEnc];
+}
+
+class IsCommutable {
+  bit isCommutable = 1;
 }
 
 class IsBranch {
@@ -234,6 +243,10 @@ def calltarget64: Operand<i64>;
 def simm16      : Operand<i32> {
   let DecoderMethod= "DecodeSimm16";
 }
+
+def simm20      : Operand<i32> {
+}
+
 def simm16_64   : Operand<i64>;
 def shamt       : Operand<i32>;
 
@@ -253,6 +266,7 @@ def mem : Operand<i32> {
   let MIOperandInfo = (ops CPURegs, simm16);
   let EncoderMethod = "getMemEncoding";
   let ParserMatchClass = MipsMemAsmOperand;
+  let OperandType = "OPERAND_MEMORY";
 }
 
 def mem64 : Operand<i64> {
@@ -260,18 +274,21 @@ def mem64 : Operand<i64> {
   let MIOperandInfo = (ops CPU64Regs, simm16_64);
   let EncoderMethod = "getMemEncoding";
   let ParserMatchClass = MipsMemAsmOperand;
+  let OperandType = "OPERAND_MEMORY";
 }
 
 def mem_ea : Operand<i32> {
   let PrintMethod = "printMemOperandEA";
   let MIOperandInfo = (ops CPURegs, simm16);
   let EncoderMethod = "getMemEncoding";
+  let OperandType = "OPERAND_MEMORY";
 }
 
 def mem_ea_64 : Operand<i64> {
   let PrintMethod = "printMemOperandEA";
   let MIOperandInfo = (ops CPU64Regs, simm16_64);
   let EncoderMethod = "getMemEncoding";
+  let OperandType = "OPERAND_MEMORY";
 }
 
 // size operand of ext instruction
@@ -296,10 +313,21 @@ def HI16 : SDNodeXForm<imm, [{
   return getImm(N, (N->getZExtValue() >> 16) & 0xFFFF);
 }]>;
 
+// Plus 1.
+def Plus1 : SDNodeXForm<imm, [{ return getImm(N, N->getSExtValue() + 1); }]>;
+
+// Node immediate fits as 16-bit sign extended on target immediate.
+// e.g. addi, andi
+def immSExt8  : PatLeaf<(imm), [{ return isInt<8>(N->getSExtValue()); }]>;
+
 // Node immediate fits as 16-bit sign extended on target immediate.
 // e.g. addi, andi
 def immSExt16  : PatLeaf<(imm), [{ return isInt<16>(N->getSExtValue()); }]>;
 
+// Node immediate fits as 15-bit sign extended on target immediate.
+// e.g. addi, andi
+def immSExt15  : PatLeaf<(imm), [{ return isInt<15>(N->getSExtValue()); }]>;
+
 // Node immediate fits as 16-bit zero extended on target immediate.
 // The LO16 param means that only the lower 16 bits of the node
 // immediate are caught.
@@ -320,113 +348,84 @@ def immLow16Zero : PatLeaf<(imm), [{
 // shamt field must fit in 5 bits.
 def immZExt5 : ImmLeaf<i32, [{return Imm == (Imm & 0x1f);}]>;
 
+// True if (N + 1) fits in 16-bit field.
+def immSExt16Plus1 : PatLeaf<(imm), [{
+  return isInt<17>(N->getSExtValue()) && isInt<16>(N->getSExtValue() + 1);
+}]>;
+
 // Mips Address Mode! SDNode frameindex could possibily be a match
 // since load and store instructions from stack used it.
 def addr :
-  ComplexPattern<iPTR, 2, "SelectAddr", [frameindex], [SDNPWantParent]>;
+  ComplexPattern<iPTR, 2, "selectIntAddr", [frameindex]>;
+
+def addrRegImm :
+  ComplexPattern<iPTR, 2, "selectAddrRegImm", [frameindex]>;
+
+def addrDefault :
+  ComplexPattern<iPTR, 2, "selectAddrDefault", [frameindex]>;
 
 //===----------------------------------------------------------------------===//
 // Instructions specific format
 //===----------------------------------------------------------------------===//
 
-/// Move Control Registers From/To CPU Registers
-def MFC0_3OP  : MFC3OP<0x10, 0, (outs CPURegs:$rt),
-                       (ins CPURegs:$rd, uimm16:$sel),"mfc0\t$rt, $rd, $sel">;
-def : InstAlias<"mfc0 $rt, $rd", (MFC0_3OP CPURegs:$rt, CPURegs:$rd, 0)>;
-
-def MTC0_3OP  : MFC3OP<0x10, 4, (outs CPURegs:$rd, uimm16:$sel),
-                       (ins CPURegs:$rt),"mtc0\t$rt, $rd, $sel">;
-def : InstAlias<"mtc0 $rt, $rd", (MTC0_3OP CPURegs:$rd, 0, CPURegs:$rt)>;
-
-def MFC2_3OP  : MFC3OP<0x12, 0, (outs CPURegs:$rt),
-                       (ins CPURegs:$rd, uimm16:$sel),"mfc2\t$rt, $rd, $sel">;
-def : InstAlias<"mfc2 $rt, $rd", (MFC2_3OP CPURegs:$rt, CPURegs:$rd, 0)>;
-
-def MTC2_3OP  : MFC3OP<0x12, 4, (outs CPURegs:$rd, uimm16:$sel),
-                       (ins CPURegs:$rt),"mtc2\t$rt, $rd, $sel">;
-def : InstAlias<"mtc2 $rt, $rd", (MTC2_3OP CPURegs:$rd, 0, CPURegs:$rt)>;
-
 // Arithmetic and logical instructions with 3 register operands.
-class ArithLogicR<bits<6> op, bits<6> func, string instr_asm, SDNode OpNode,
-                  InstrItinClass itin, RegisterClass RC, bit isComm = 0>:
-  FR<op, func, (outs RC:$rd), (ins RC:$rs, RC:$rt),
-     !strconcat(instr_asm, "\t$rd, $rs, $rt"),
-     [(set RC:$rd, (OpNode RC:$rs, RC:$rt))], itin> {
-  let shamt = 0;
+class ArithLogicR<string opstr, RegisterOperand RO, bit isComm = 0,
+                  InstrItinClass Itin = NoItinerary,
+                  SDPatternOperator OpNode = null_frag>:
+  InstSE<(outs RO:$rd), (ins RO:$rs, RO:$rt),
+         !strconcat(opstr, "\t$rd, $rs, $rt"),
+         [(set RO:$rd, (OpNode RO:$rs, RO:$rt))], Itin, FrmR> {
   let isCommutable = isComm;
   let isReMaterializable = 1;
-}
-
-class ArithOverflowR<bits<6> op, bits<6> func, string instr_asm,
-                    InstrItinClass itin, RegisterClass RC, bit isComm = 0>:
-  FR<op, func, (outs RC:$rd), (ins RC:$rs, RC:$rt),
-     !strconcat(instr_asm, "\t$rd, $rs, $rt"), [], itin> {
-  let shamt = 0;
-  let isCommutable = isComm;
+  string BaseOpcode;
+  string Arch;
 }
 
 // Arithmetic and logical instructions with 2 register operands.
-class ArithLogicI<bits<6> op, string instr_asm, SDNode OpNode,
-                  Operand Od, PatLeaf imm_type, RegisterClass RC> :
-  FI<op, (outs RC:$rt), (ins RC:$rs, Od:$imm16),
-     !strconcat(instr_asm, "\t$rt, $rs, $imm16"),
-     [(set RC:$rt, (OpNode RC:$rs, imm_type:$imm16))], IIAlu> {
+class ArithLogicI<string opstr, Operand Od, RegisterOperand RO,
+                  SDPatternOperator imm_type = null_frag,
+                  SDPatternOperator OpNode = null_frag> :
+  InstSE<(outs RO:$rt), (ins RO:$rs, Od:$imm16),
+         !strconcat(opstr, "\t$rt, $rs, $imm16"),
+         [(set RO:$rt, (OpNode RO:$rs, imm_type:$imm16))], IIAlu, FrmI> {
   let isReMaterializable = 1;
 }
 
-class ArithOverflowI<bits<6> op, string instr_asm, SDNode OpNode,
-                     Operand Od, PatLeaf imm_type, RegisterClass RC> :
-  FI<op, (outs RC:$rt), (ins RC:$rs, Od:$imm16),
-     !strconcat(instr_asm, "\t$rt, $rs, $imm16"), [], IIAlu>;
-
 // Arithmetic Multiply ADD/SUB
-let rd = 0, shamt = 0, Defs = [HI, LO], Uses = [HI, LO] in
-class MArithR<bits<6> func, string instr_asm, SDNode op, bit isComm = 0> :
-  FR<0x1c, func, (outs), (ins CPURegs:$rs, CPURegs:$rt),
-     !strconcat(instr_asm, "\t$rs, $rt"),
-     [(op CPURegs:$rs, CPURegs:$rt, LO, HI)], IIImul> {
-  let rd = 0;
-  let shamt = 0;
+class MArithR<string opstr, bit isComm = 0> :
+  InstSE<(outs), (ins CPURegsOpnd:$rs, CPURegsOpnd:$rt),
+         !strconcat(opstr, "\t$rs, $rt"), [], IIImul, FrmR> {
+  let Defs = [HI, LO];
+  let Uses = [HI, LO];
   let isCommutable = isComm;
 }
 
 //  Logical
-class LogicNOR<bits<6> op, bits<6> func, string instr_asm, RegisterClass RC>:
-  FR<op, func, (outs RC:$rd), (ins RC:$rs, RC:$rt),
-     !strconcat(instr_asm, "\t$rd, $rs, $rt"),
-     [(set RC:$rd, (not (or RC:$rs, RC:$rt)))], IIAlu> {
-  let shamt = 0;
+class LogicNOR<string opstr, RegisterOperand RC>:
+  InstSE<(outs RC:$rd), (ins RC:$rs, RC:$rt),
+         !strconcat(opstr, "\t$rd, $rs, $rt"),
+         [(set RC:$rd, (not (or RC:$rs, RC:$rt)))], IIAlu, FrmR> {
   let isCommutable = 1;
 }
 
 // Shifts
-class shift_rotate_imm<bits<6> func, bits<5> isRotate, string instr_asm,
-                       SDNode OpNode, PatFrag PF, Operand ImmOpnd,
-                       RegisterClass RC>:
-  FR<0x00, func, (outs RC:$rd), (ins RC:$rt, ImmOpnd:$shamt),
-     !strconcat(instr_asm, "\t$rd, $rt, $shamt"),
-     [(set RC:$rd, (OpNode RC:$rt, PF:$shamt))], IIAlu> {
-  let rs = isRotate;
-}
-
-// 32-bit shift instructions.
-class shift_rotate_imm32<bits<6> func, bits<5> isRotate, string instr_asm,
-                         SDNode OpNode>:
-  shift_rotate_imm<func, isRotate, instr_asm, OpNode, immZExt5, shamt, CPURegs>;
-
-class shift_rotate_reg<bits<6> func, bits<5> isRotate, string instr_asm,
-                       SDNode OpNode, RegisterClass RC>:
-  FR<0x00, func, (outs RC:$rd), (ins CPURegs:$rs, RC:$rt),
-     !strconcat(instr_asm, "\t$rd, $rt, $rs"),
-     [(set RC:$rd, (OpNode RC:$rt, CPURegs:$rs))], IIAlu> {
-  let shamt = isRotate;
-}
+class shift_rotate_imm<string opstr, Operand ImmOpnd,
+                       RegisterOperand RC, SDPatternOperator OpNode = null_frag,
+                       SDPatternOperator PF = null_frag> :
+  InstSE<(outs RC:$rd), (ins RC:$rt, ImmOpnd:$shamt),
+         !strconcat(opstr, "\t$rd, $rt, $shamt"),
+         [(set RC:$rd, (OpNode RC:$rt, PF:$shamt))], IIAlu, FrmR>;
+
+class shift_rotate_reg<string opstr, RegisterOperand RC,
+                       SDPatternOperator OpNode = null_frag>:
+  InstSE<(outs RC:$rd), (ins CPURegsOpnd:$rs, RC:$rt),
+         !strconcat(opstr, "\t$rd, $rt, $rs"),
+         [(set RC:$rd, (OpNode RC:$rt, CPURegsOpnd:$rs))], IIAlu, FrmR>;
 
 // Load Upper Imediate
-class LoadUpper<bits<6> op, string instr_asm, RegisterClass RC, Operand Imm>:
-  FI<op, (outs RC:$rt), (ins Imm:$imm16),
-     !strconcat(instr_asm, "\t$rt, $imm16"), [], IIAlu>, IsAsCheapAsAMove {
-  let rs = 0;
+class LoadUpper<string opstr, RegisterClass RC, Operand Imm>:
+  InstSE<(outs RC:$rt), (ins Imm:$imm16), !strconcat(opstr, "\t$rt, $imm16"),
+         [], IIAlu, FrmI>, IsAsCheapAsAMove {
   let neverHasSideEffects = 1;
   let isReMaterializable = 1;
 }
@@ -440,66 +439,40 @@ class FMem<bits<6> op, dag outs, dag ins, string asmstr, list<dag> pattern,
 }
 
 // Memory Load/Store
-let canFoldAsLoad = 1 in
-class LoadM<bits<6> op, string instr_asm, PatFrag OpNode, RegisterClass RC,
-            Operand MemOpnd, bit Pseudo>:
-  FMem<op, (outs RC:$rt), (ins MemOpnd:$addr),
-     !strconcat(instr_asm, "\t$rt, $addr"),
-     [(set RC:$rt, (OpNode addr:$addr))], IILoad> {
-  let isPseudo = Pseudo;
-}
-
-class StoreM<bits<6> op, string instr_asm, PatFrag OpNode, RegisterClass RC,
-             Operand MemOpnd, bit Pseudo>:
-  FMem<op, (outs), (ins RC:$rt, MemOpnd:$addr),
-     !strconcat(instr_asm, "\t$rt, $addr"),
-     [(OpNode RC:$rt, addr:$addr)], IIStore> {
-  let isPseudo = Pseudo;
-}
-
-// 32-bit load.
-multiclass LoadM32<bits<6> op, string instr_asm, PatFrag OpNode,
-                   bit Pseudo = 0> {
-  def #NAME# : LoadM<op, instr_asm, OpNode, CPURegs, mem, Pseudo>,
-               Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : LoadM<op, instr_asm, OpNode, CPURegs, mem64, Pseudo>,
-               Requires<[IsN64, HasStandardEncoding]> {
-    let DecoderNamespace = "Mips64";
-    let isCodeGenOnly = 1;
-  }
+class Load<string opstr, SDPatternOperator OpNode, RegisterClass RC,
+           Operand MemOpnd, ComplexPattern Addr> :
+  InstSE<(outs RC:$rt), (ins MemOpnd:$addr), !strconcat(opstr, "\t$rt, $addr"),
+         [(set RC:$rt, (OpNode Addr:$addr))], NoItinerary, FrmI> {
+  let DecoderMethod = "DecodeMem";
+  let canFoldAsLoad = 1;
+  let mayLoad = 1;
 }
 
-// 64-bit load.
-multiclass LoadM64<bits<6> op, string instr_asm, PatFrag OpNode,
-                   bit Pseudo = 0> {
-  def #NAME# : LoadM<op, instr_asm, OpNode, CPU64Regs, mem, Pseudo>,
-               Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : LoadM<op, instr_asm, OpNode, CPU64Regs, mem64, Pseudo>,
-               Requires<[IsN64, HasStandardEncoding]> {
-    let DecoderNamespace = "Mips64";
-    let isCodeGenOnly = 1;
-  }
+class Store<string opstr, SDPatternOperator OpNode, RegisterClass RC,
+            Operand MemOpnd, ComplexPattern Addr> :
+  InstSE<(outs), (ins RC:$rt, MemOpnd:$addr), !strconcat(opstr, "\t$rt, $addr"),
+         [(OpNode RC:$rt, Addr:$addr)], NoItinerary, FrmI> {
+  let DecoderMethod = "DecodeMem";
+  let mayStore = 1;
 }
 
-// 32-bit store.
-multiclass StoreM32<bits<6> op, string instr_asm, PatFrag OpNode,
-                    bit Pseudo = 0> {
-  def #NAME# : StoreM<op, instr_asm, OpNode, CPURegs, mem, Pseudo>,
-               Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : StoreM<op, instr_asm, OpNode, CPURegs, mem64, Pseudo>,
-               Requires<[IsN64, HasStandardEncoding]> {
+multiclass LoadM<string opstr, RegisterClass RC,
+                 SDPatternOperator OpNode = null_frag,
+                 ComplexPattern Addr = addr> {
+  def NAME : Load<opstr, OpNode, RC, mem, Addr>, Requires<[NotN64, HasStdEnc]>;
+  def _P8  : Load<opstr, OpNode, RC, mem64, Addr>,
+             Requires<[IsN64, HasStdEnc]> {
     let DecoderNamespace = "Mips64";
     let isCodeGenOnly = 1;
   }
 }
 
-// 64-bit store.
-multiclass StoreM64<bits<6> op, string instr_asm, PatFrag OpNode,
-                    bit Pseudo = 0> {
-  def #NAME# : StoreM<op, instr_asm, OpNode, CPU64Regs, mem, Pseudo>,
-               Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : StoreM<op, instr_asm, OpNode, CPU64Regs, mem64, Pseudo>,
-               Requires<[IsN64, HasStandardEncoding]> {
+multiclass StoreM<string opstr, RegisterClass RC,
+                  SDPatternOperator OpNode = null_frag,
+                  ComplexPattern Addr = addr> {
+  def NAME : Store<opstr, OpNode, RC, mem, Addr>, Requires<[NotN64, HasStdEnc]>;
+  def _P8  : Store<opstr, OpNode, RC, mem64, Addr>,
+             Requires<[IsN64, HasStdEnc]> {
     let DecoderNamespace = "Mips64";
     let isCodeGenOnly = 1;
   }
@@ -507,81 +480,58 @@ multiclass StoreM64<bits<6> op, string instr_asm, PatFrag OpNode,
 
 // Load/Store Left/Right
 let canFoldAsLoad = 1 in
-class LoadLeftRight<bits<6> op, string instr_asm, SDNode OpNode,
-                    RegisterClass RC, Operand MemOpnd> :
-  FMem<op, (outs RC:$rt), (ins MemOpnd:$addr, RC:$src),
-       !strconcat(instr_asm, "\t$rt, $addr"),
-       [(set RC:$rt, (OpNode addr:$addr, RC:$src))], IILoad> {
+class LoadLeftRight<string opstr, SDNode OpNode, RegisterClass RC,
+                    Operand MemOpnd> :
+  InstSE<(outs RC:$rt), (ins MemOpnd:$addr, RC:$src),
+         !strconcat(opstr, "\t$rt, $addr"),
+         [(set RC:$rt, (OpNode addr:$addr, RC:$src))], NoItinerary, FrmI> {
+  let DecoderMethod = "DecodeMem";
   string Constraints = "$src = $rt";
 }
 
-class StoreLeftRight<bits<6> op, string instr_asm, SDNode OpNode,
-                     RegisterClass RC, Operand MemOpnd>:
-  FMem<op, (outs), (ins RC:$rt, MemOpnd:$addr),
-       !strconcat(instr_asm, "\t$rt, $addr"), [(OpNode RC:$rt, addr:$addr)],
-       IIStore>;
-
-// 32-bit load left/right.
-multiclass LoadLeftRightM32<bits<6> op, string instr_asm, SDNode OpNode> {
-  def #NAME# : LoadLeftRight<op, instr_asm, OpNode, CPURegs, mem>,
-               Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : LoadLeftRight<op, instr_asm, OpNode, CPURegs, mem64>,
-               Requires<[IsN64, HasStandardEncoding]> {
-    let DecoderNamespace = "Mips64";
-    let isCodeGenOnly = 1;
-  }
-}
-
-// 64-bit load left/right.
-multiclass LoadLeftRightM64<bits<6> op, string instr_asm, SDNode OpNode> {
-  def #NAME# : LoadLeftRight<op, instr_asm, OpNode, CPU64Regs, mem>,
-               Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : LoadLeftRight<op, instr_asm, OpNode, CPU64Regs, mem64>,
-               Requires<[IsN64, HasStandardEncoding]> {
-    let DecoderNamespace = "Mips64";
-    let isCodeGenOnly = 1;
-  }
+class StoreLeftRight<string opstr, SDNode OpNode, RegisterClass RC,
+                     Operand MemOpnd>:
+  InstSE<(outs), (ins RC:$rt, MemOpnd:$addr), !strconcat(opstr, "\t$rt, $addr"),
+         [(OpNode RC:$rt, addr:$addr)], NoItinerary, FrmI> {
+  let DecoderMethod = "DecodeMem";
 }
 
-// 32-bit store left/right.
-multiclass StoreLeftRightM32<bits<6> op, string instr_asm, SDNode OpNode> {
-  def #NAME# : StoreLeftRight<op, instr_asm, OpNode, CPURegs, mem>,
-               Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : StoreLeftRight<op, instr_asm, OpNode, CPURegs, mem64>,
-               Requires<[IsN64, HasStandardEncoding]> {
+multiclass LoadLeftRightM<string opstr, SDNode OpNode, RegisterClass RC> {
+  def NAME : LoadLeftRight<opstr, OpNode, RC, mem>,
+             Requires<[NotN64, HasStdEnc]>;
+  def _P8  : LoadLeftRight<opstr, OpNode, RC, mem64>,
+             Requires<[IsN64, HasStdEnc]> {
     let DecoderNamespace = "Mips64";
     let isCodeGenOnly = 1;
   }
 }
 
-// 64-bit store left/right.
-multiclass StoreLeftRightM64<bits<6> op, string instr_asm, SDNode OpNode> {
-  def #NAME# : StoreLeftRight<op, instr_asm, OpNode, CPU64Regs, mem>,
-               Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : StoreLeftRight<op, instr_asm, OpNode, CPU64Regs, mem64>,
-               Requires<[IsN64, HasStandardEncoding]> {
+multiclass StoreLeftRightM<string opstr, SDNode OpNode, RegisterClass RC> {
+  def NAME : StoreLeftRight<opstr, OpNode, RC, mem>,
+             Requires<[NotN64, HasStdEnc]>;
+  def _P8  : StoreLeftRight<opstr, OpNode, RC, mem64>,
+             Requires<[IsN64, HasStdEnc]> {
     let DecoderNamespace = "Mips64";
     let isCodeGenOnly = 1;
   }
 }
 
 // Conditional Branch
-class CBranch<bits<6> op, string instr_asm, PatFrag cond_op, RegisterClass RC>:
-  BranchBase<op, (outs), (ins RC:$rs, RC:$rt, brtarget:$imm16),
-             !strconcat(instr_asm, "\t$rs, $rt, $imm16"),
-             [(brcond (i32 (cond_op RC:$rs, RC:$rt)), bb:$imm16)], IIBranch> {
+class CBranch<string opstr, PatFrag cond_op, RegisterClass RC> :
+  InstSE<(outs), (ins RC:$rs, RC:$rt, brtarget:$offset),
+         !strconcat(opstr, "\t$rs, $rt, $offset"),
+         [(brcond (i32 (cond_op RC:$rs, RC:$rt)), bb:$offset)], IIBranch,
+         FrmI> {
   let isBranch = 1;
   let isTerminator = 1;
   let hasDelaySlot = 1;
   let Defs = [AT];
 }
 
-class CBranchZero<bits<6> op, bits<5> _rt, string instr_asm, PatFrag cond_op,
-                  RegisterClass RC>:
-  BranchBase<op, (outs), (ins RC:$rs, brtarget:$imm16),
-             !strconcat(instr_asm, "\t$rs, $imm16"),
-             [(brcond (i32 (cond_op RC:$rs, 0)), bb:$imm16)], IIBranch> {
-  let rt = _rt;
+class CBranchZero<string opstr, PatFrag cond_op, RegisterClass RC> :
+  InstSE<(outs), (ins RC:$rs, brtarget:$offset),
+         !strconcat(opstr, "\t$rs, $offset"),
+         [(brcond (i32 (cond_op RC:$rs, 0)), bb:$offset)], IIBranch, FrmI> {
   let isBranch = 1;
   let isTerminator = 1;
   let hasDelaySlot = 1;
@@ -589,27 +539,23 @@ class CBranchZero<bits<6> op, bits<5> _rt, string instr_asm, PatFrag cond_op,
 }
 
 // SetCC
-class SetCC_R<bits<6> op, bits<6> func, string instr_asm, PatFrag cond_op,
-              RegisterClass RC>:
-  FR<op, func, (outs CPURegs:$rd), (ins RC:$rs, RC:$rt),
-     !strconcat(instr_asm, "\t$rd, $rs, $rt"),
-     [(set CPURegs:$rd, (cond_op RC:$rs, RC:$rt))],
-     IIAlu> {
-  let shamt = 0;
-}
+class SetCC_R<string opstr, PatFrag cond_op, RegisterClass RC> :
+  InstSE<(outs CPURegsOpnd:$rd), (ins RC:$rs, RC:$rt),
+         !strconcat(opstr, "\t$rd, $rs, $rt"),
+         [(set CPURegsOpnd:$rd, (cond_op RC:$rs, RC:$rt))], IIAlu, FrmR>;
 
-class SetCC_I<bits<6> op, string instr_asm, PatFrag cond_op, Operand Od,
-              PatLeaf imm_type, RegisterClass RC>:
-  FI<op, (outs CPURegs:$rt), (ins RC:$rs, Od:$imm16),
-     !strconcat(instr_asm, "\t$rt, $rs, $imm16"),
-     [(set CPURegs:$rt, (cond_op RC:$rs, imm_type:$imm16))],
-     IIAlu>;
+class SetCC_I<string opstr, PatFrag cond_op, Operand Od, PatLeaf imm_type,
+              RegisterClass RC>:
+  InstSE<(outs CPURegsOpnd:$rt), (ins RC:$rs, Od:$imm16),
+         !strconcat(opstr, "\t$rt, $rs, $imm16"),
+         [(set CPURegsOpnd:$rt, (cond_op RC:$rs, imm_type:$imm16))],
+         IIAlu, FrmI>;
 
 // Jump
-class JumpFJ<bits<6> op, DAGOperand opnd, string instr_asm,
-             SDPatternOperator operator, SDPatternOperator targetoperator>:
-  FJ<op, (outs), (ins opnd:$target), !strconcat(instr_asm, "\t$target"),
-     [(operator targetoperator:$target)], IIBranch> {
+class JumpFJ<DAGOperand opnd, string opstr, SDPatternOperator operator,
+             SDPatternOperator targetoperator> :
+  InstSE<(outs), (ins opnd:$target), !strconcat(opstr, "\t$target"),
+         [(operator targetoperator:$target)], IIBranch, FrmJ> {
   let isTerminator=1;
   let isBarrier=1;
   let hasDelaySlot = 1;
@@ -618,27 +564,21 @@ class JumpFJ<bits<6> op, DAGOperand opnd, string instr_asm,
 }
 
 // Unconditional branch
-class UncondBranch<bits<6> op, string instr_asm>:
-  BranchBase<op, (outs), (ins brtarget:$imm16),
-             !strconcat(instr_asm, "\t$imm16"), [(br bb:$imm16)], IIBranch> {
-  let rs = 0;
-  let rt = 0;
+class UncondBranch<string opstr> :
+  InstSE<(outs), (ins brtarget:$offset), !strconcat(opstr, "\t$offset"),
+         [(br bb:$offset)], IIBranch, FrmI> {
   let isBranch = 1;
   let isTerminator = 1;
   let isBarrier = 1;
   let hasDelaySlot = 1;
-  let Predicates = [RelocPIC, HasStandardEncoding];
+  let Predicates = [RelocPIC, HasStdEnc];
   let Defs = [AT];
 }
 
 // Base class for indirect branch and return instruction classes.
 let isTerminator=1, isBarrier=1, hasDelaySlot = 1 in
 class JumpFR<RegisterClass RC, SDPatternOperator operator = null_frag>:
-  FR<0, 0x8, (outs), (ins RC:$rs), "jr\t$rs", [(operator RC:$rs)], IIBranch> {
-  let rt = 0;
-  let rd = 0;
-  let shamt = 0;
-}
+  InstSE<(outs), (ins RC:$rs), "jr\t$rs", [(operator RC:$rs)], IIBranch, FrmR>;
 
 // Indirect branch
 class IndirectBranch<RegisterClass RC>: JumpFR<RC, brind> {
@@ -656,465 +596,523 @@ class RetBase<RegisterClass RC>: JumpFR<RC> {
 
 // Jump and Link (Call)
 let isCall=1, hasDelaySlot=1, Defs = [RA] in {
-  class JumpLink<bits<6> op, string instr_asm>:
-    FJ<op, (outs), (ins calltarget:$target),
-       !strconcat(instr_asm, "\t$target"), [(MipsJmpLink imm:$target)],
-       IIBranch> {
-       let DecoderMethod = "DecodeJumpTarget";
-       }
-
-  class JumpLinkReg<bits<6> op, bits<6> func, string instr_asm,
-                    RegisterClass RC>:
-    FR<op, func, (outs), (ins RC:$rs),
-       !strconcat(instr_asm, "\t$rs"), [(MipsJmpLink RC:$rs)], IIBranch> {
-    let rt = 0;
-    let rd = 31;
-    let shamt = 0;
+  class JumpLink<string opstr> :
+    InstSE<(outs), (ins calltarget:$target), !strconcat(opstr, "\t$target"),
+           [(MipsJmpLink imm:$target)], IIBranch, FrmJ> {
+    let DecoderMethod = "DecodeJumpTarget";
   }
 
-  class BranchLink<string instr_asm, bits<5> _rt, RegisterClass RC>:
-    FI<0x1, (outs), (ins RC:$rs, brtarget:$imm16),
-       !strconcat(instr_asm, "\t$rs, $imm16"), [], IIBranch> {
-    let rt = _rt;
-  }
+  class JumpLinkRegPseudo<RegisterClass RC, Instruction JALRInst,
+                          Register RetReg>:
+    PseudoSE<(outs), (ins RC:$rs), [(MipsJmpLink RC:$rs)], IIBranch>,
+    PseudoInstExpansion<(JALRInst RetReg, RC:$rs)>;
+
+  class JumpLinkReg<string opstr, RegisterClass RC>:
+    InstSE<(outs RC:$rd), (ins RC:$rs), !strconcat(opstr, "\t$rd, $rs"),
+           [], IIBranch, FrmR>;
+
+  class BGEZAL_FT<string opstr, RegisterOperand RO> :
+    InstSE<(outs), (ins RO:$rs, brtarget:$offset),
+           !strconcat(opstr, "\t$rs, $offset"), [], IIBranch, FrmI>;
+
+}
+
+class BAL_FT :
+  InstSE<(outs), (ins brtarget:$offset), "bal\t$offset", [], IIBranch, FrmI> {
+  let isBranch = 1;
+  let isTerminator = 1;
+  let isBarrier = 1;
+  let hasDelaySlot = 1;
+  let Defs = [RA];
 }
 
+// Sync
+let hasSideEffects = 1 in
+class SYNC_FT :
+  InstSE<(outs), (ins i32imm:$stype), "sync $stype", [(MipsSync imm:$stype)],
+         NoItinerary, FrmOther>;
+
 // Mul, Div
-class Mult<bits<6> func, string instr_asm, InstrItinClass itin,
-           RegisterClass RC, list<Register> DefRegs>:
-  FR<0x00, func, (outs), (ins RC:$rs, RC:$rt),
-     !strconcat(instr_asm, "\t$rs, $rt"), [], itin> {
-  let rd = 0;
-  let shamt = 0;
+class Mult<string opstr, InstrItinClass itin, RegisterOperand RO,
+           list<Register> DefRegs> :
+  InstSE<(outs), (ins RO:$rs, RO:$rt), !strconcat(opstr, "\t$rs, $rt"), [],
+         itin, FrmR> {
   let isCommutable = 1;
   let Defs = DefRegs;
   let neverHasSideEffects = 1;
 }
 
-class Mult32<bits<6> func, string instr_asm, InstrItinClass itin>:
-  Mult<func, instr_asm, itin, CPURegs, [HI, LO]>;
-
-class Div<SDNode op, bits<6> func, string instr_asm, InstrItinClass itin,
-          RegisterClass RC, list<Register> DefRegs>:
-  FR<0x00, func, (outs), (ins RC:$rs, RC:$rt),
-     !strconcat(instr_asm, "\t$$zero, $rs, $rt"),
-     [(op RC:$rs, RC:$rt)], itin> {
-  let rd = 0;
-  let shamt = 0;
+// Pseudo multiply/divide instruction with explicit accumulator register
+// operands.
+class MultDivPseudo<Instruction RealInst, RegisterClass R0, RegisterOperand R1,
+                    SDPatternOperator OpNode, InstrItinClass Itin,
+                    bit IsComm = 1, bit HasSideEffects = 0> :
+  PseudoSE<(outs R0:$ac), (ins R1:$rs, R1:$rt),
+           [(set R0:$ac, (OpNode R1:$rs, R1:$rt))], Itin>,
+  PseudoInstExpansion<(RealInst R1:$rs, R1:$rt)> {
+  let isCommutable = IsComm;
+  let hasSideEffects = HasSideEffects;
+}
+
+// Pseudo multiply add/sub instruction with explicit accumulator register
+// operands.
+class MAddSubPseudo<Instruction RealInst, SDPatternOperator OpNode>
+  : PseudoSE<(outs ACRegs:$ac),
+             (ins CPURegsOpnd:$rs, CPURegsOpnd:$rt, ACRegs:$acin),
+             [(set ACRegs:$ac,
+              (OpNode CPURegsOpnd:$rs, CPURegsOpnd:$rt, ACRegs:$acin))],
+             IIImul>,
+    PseudoInstExpansion<(RealInst CPURegsOpnd:$rs, CPURegsOpnd:$rt)> {
+  string Constraints = "$acin = $ac";
+}
+
+class Div<string opstr, InstrItinClass itin, RegisterOperand RO,
+          list<Register> DefRegs> :
+  InstSE<(outs), (ins RO:$rs, RO:$rt), !strconcat(opstr, "\t$$zero, $rs, $rt"),
+         [], itin, FrmR> {
   let Defs = DefRegs;
 }
 
-class Div32<SDNode op, bits<6> func, string instr_asm, InstrItinClass itin>:
-  Div<op, func, instr_asm, itin, CPURegs, [HI, LO]>;
-
 // Move from Hi/Lo
-class MoveFromLOHI<bits<6> func, string instr_asm, RegisterClass RC,
-                   list<Register> UseRegs>:
-  FR<0x00, func, (outs RC:$rd), (ins),
-     !strconcat(instr_asm, "\t$rd"), [], IIHiLo> {
-  let rs = 0;
-  let rt = 0;
-  let shamt = 0;
+class MoveFromLOHI<string opstr, RegisterClass RC, list<Register> UseRegs>:
+  InstSE<(outs RC:$rd), (ins), !strconcat(opstr, "\t$rd"), [], IIHiLo, FrmR> {
   let Uses = UseRegs;
   let neverHasSideEffects = 1;
 }
 
-class MoveToLOHI<bits<6> func, string instr_asm, RegisterClass RC,
-                 list<Register> DefRegs>:
-  FR<0x00, func, (outs), (ins RC:$rs),
-     !strconcat(instr_asm, "\t$rs"), [], IIHiLo> {
-  let rt = 0;
-  let rd = 0;
-  let shamt = 0;
+class MoveToLOHI<string opstr, RegisterClass RC, list<Register> DefRegs>:
+  InstSE<(outs), (ins RC:$rs), !strconcat(opstr, "\t$rs"), [], IIHiLo, FrmR> {
   let Defs = DefRegs;
   let neverHasSideEffects = 1;
 }
 
-class EffectiveAddress<bits<6> opc, string instr_asm, RegisterClass RC, Operand Mem> :
-  FMem<opc, (outs RC:$rt), (ins Mem:$addr),
-     instr_asm, [(set RC:$rt, addr:$addr)], IIAlu> {
- let isCodeGenOnly = 1;
+class EffectiveAddress<string opstr, RegisterClass RC, Operand Mem> :
+  InstSE<(outs RC:$rt), (ins Mem:$addr), !strconcat(opstr, "\t$rt, $addr"),
+         [(set RC:$rt, addr:$addr)], NoItinerary, FrmI> {
+  let isCodeGenOnly = 1;
+  let DecoderMethod = "DecodeMem";
 }
 
 // Count Leading Ones/Zeros in Word
-class CountLeading0<bits<6> func, string instr_asm, RegisterClass RC>:
-  FR<0x1c, func, (outs RC:$rd), (ins RC:$rs),
-     !strconcat(instr_asm, "\t$rd, $rs"),
-     [(set RC:$rd, (ctlz RC:$rs))], IIAlu>,
-     Requires<[HasBitCount, HasStandardEncoding]> {
-  let shamt = 0;
-  let rt = rd;
-}
+class CountLeading0<string opstr, RegisterOperand RO>:
+  InstSE<(outs RO:$rd), (ins RO:$rs), !strconcat(opstr, "\t$rd, $rs"),
+         [(set RO:$rd, (ctlz RO:$rs))], IIAlu, FrmR>,
+  Requires<[HasBitCount, HasStdEnc]>;
+
+class CountLeading1<string opstr, RegisterOperand RO>:
+  InstSE<(outs RO:$rd), (ins RO:$rs), !strconcat(opstr, "\t$rd, $rs"),
+         [(set RO:$rd, (ctlz (not RO:$rs)))], IIAlu, FrmR>,
+  Requires<[HasBitCount, HasStdEnc]>;
 
-class CountLeading1<bits<6> func, string instr_asm, RegisterClass RC>:
-  FR<0x1c, func, (outs RC:$rd), (ins RC:$rs),
-     !strconcat(instr_asm, "\t$rd, $rs"),
-     [(set RC:$rd, (ctlz (not RC:$rs)))], IIAlu>,
-     Requires<[HasBitCount, HasStandardEncoding]> {
-  let shamt = 0;
-  let rt = rd;
-}
 
 // Sign Extend in Register.
-class SignExtInReg<bits<5> sa, string instr_asm, ValueType vt,
-                   RegisterClass RC>:
-  FR<0x1f, 0x20, (outs RC:$rd), (ins RC:$rt),
-     !strconcat(instr_asm, "\t$rd, $rt"),
-     [(set RC:$rd, (sext_inreg RC:$rt, vt))], NoItinerary> {
-  let rs = 0;
-  let shamt = sa;
-  let Predicates = [HasSEInReg, HasStandardEncoding];
+class SignExtInReg<string opstr, ValueType vt, RegisterClass RC> :
+  InstSE<(outs RC:$rd), (ins RC:$rt), !strconcat(opstr, "\t$rd, $rt"),
+         [(set RC:$rd, (sext_inreg RC:$rt, vt))], NoItinerary, FrmR> {
+  let Predicates = [HasSEInReg, HasStdEnc];
 }
 
 // Subword Swap
-class SubwordSwap<bits<6> func, bits<5> sa, string instr_asm, RegisterClass RC>:
-  FR<0x1f, func, (outs RC:$rd), (ins RC:$rt),
-     !strconcat(instr_asm, "\t$rd, $rt"), [], NoItinerary> {
-  let rs = 0;
-  let shamt = sa;
-  let Predicates = [HasSwap, HasStandardEncoding];
+class SubwordSwap<string opstr, RegisterOperand RO>:
+  InstSE<(outs RO:$rd), (ins RO:$rt), !strconcat(opstr, "\t$rd, $rt"), [],
+         NoItinerary, FrmR> {
+  let Predicates = [HasSwap, HasStdEnc];
   let neverHasSideEffects = 1;
 }
 
 // Read Hardware
-class ReadHardware<RegisterClass CPURegClass, RegisterClass HWRegClass>
-  : FR<0x1f, 0x3b, (outs CPURegClass:$rt), (ins HWRegClass:$rd),
-       "rdhwr\t$rt, $rd", [], IIAlu> {
-  let rs = 0;
-  let shamt = 0;
-}
+class ReadHardware<RegisterClass CPURegClass, RegisterOperand RO> :
+  InstSE<(outs CPURegClass:$rt), (ins RO:$rd), "rdhwr\t$rt, $rd", [],
+         IIAlu, FrmR>;
 
 // Ext and Ins
-class ExtBase<bits<6> _funct, string instr_asm, RegisterClass RC>:
-  FR<0x1f, _funct, (outs RC:$rt), (ins RC:$rs, uimm16:$pos, size_ext:$sz),
-     !strconcat(instr_asm, " $rt, $rs, $pos, $sz"),
-     [(set RC:$rt, (MipsExt RC:$rs, imm:$pos, imm:$sz))], NoItinerary> {
-  bits<5> pos;
-  bits<5> sz;
-  let rd = sz;
-  let shamt = pos;
-  let Predicates = [HasMips32r2, HasStandardEncoding];
-}
-
-class InsBase<bits<6> _funct, string instr_asm, RegisterClass RC>:
-  FR<0x1f, _funct, (outs RC:$rt),
-     (ins RC:$rs, uimm16:$pos, size_ins:$sz, RC:$src),
-     !strconcat(instr_asm, " $rt, $rs, $pos, $sz"),
-     [(set RC:$rt, (MipsIns RC:$rs, imm:$pos, imm:$sz, RC:$src))],
-     NoItinerary> {
-  bits<5> pos;
-  bits<5> sz;
-  let rd = sz;
-  let shamt = pos;
-  let Predicates = [HasMips32r2, HasStandardEncoding];
+class ExtBase<string opstr, RegisterOperand RO>:
+  InstSE<(outs RO:$rt), (ins RO:$rs, uimm16:$pos, size_ext:$size),
+         !strconcat(opstr, " $rt, $rs, $pos, $size"),
+         [(set RO:$rt, (MipsExt RO:$rs, imm:$pos, imm:$size))], NoItinerary,
+         FrmR> {
+  let Predicates = [HasMips32r2, HasStdEnc];
+}
+
+class InsBase<string opstr, RegisterOperand RO>:
+  InstSE<(outs RO:$rt), (ins RO:$rs, uimm16:$pos, size_ins:$size, RO:$src),
+         !strconcat(opstr, " $rt, $rs, $pos, $size"),
+         [(set RO:$rt, (MipsIns RO:$rs, imm:$pos, imm:$size, RO:$src))],
+         NoItinerary, FrmR> {
+  let Predicates = [HasMips32r2, HasStdEnc];
   let Constraints = "$src = $rt";
 }
 
 // Atomic instructions with 2 source operands (ATOMIC_SWAP & ATOMIC_LOAD_*).
-class Atomic2Ops<PatFrag Op, string Opstr, RegisterClass DRC,
-                 RegisterClass PRC> :
+class Atomic2Ops<PatFrag Op, RegisterClass DRC, RegisterClass PRC> :
   PseudoSE<(outs DRC:$dst), (ins PRC:$ptr, DRC:$incr),
-           !strconcat("atomic_", Opstr, "\t$dst, $ptr, $incr"),
            [(set DRC:$dst, (Op PRC:$ptr, DRC:$incr))]>;
 
-multiclass Atomic2Ops32<PatFrag Op, string Opstr> {
-  def #NAME# : Atomic2Ops<Op, Opstr, CPURegs, CPURegs>,
-                          Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : Atomic2Ops<Op, Opstr, CPURegs, CPU64Regs>,
-                          Requires<[IsN64, HasStandardEncoding]> {
+multiclass Atomic2Ops32<PatFrag Op> {
+  def NAME : Atomic2Ops<Op, CPURegs, CPURegs>, Requires<[NotN64, HasStdEnc]>;
+  def _P8  : Atomic2Ops<Op, CPURegs, CPU64Regs>,
+             Requires<[IsN64, HasStdEnc]> {
     let DecoderNamespace = "Mips64";
   }
 }
 
 // Atomic Compare & Swap.
-class AtomicCmpSwap<PatFrag Op, string Width, RegisterClass DRC,
-                    RegisterClass PRC> :
+class AtomicCmpSwap<PatFrag Op, RegisterClass DRC, RegisterClass PRC> :
   PseudoSE<(outs DRC:$dst), (ins PRC:$ptr, DRC:$cmp, DRC:$swap),
-           !strconcat("atomic_cmp_swap_", Width, "\t$dst, $ptr, $cmp, $swap"),
            [(set DRC:$dst, (Op PRC:$ptr, DRC:$cmp, DRC:$swap))]>;
 
-multiclass AtomicCmpSwap32<PatFrag Op, string Width>  {
-  def #NAME# : AtomicCmpSwap<Op, Width, CPURegs, CPURegs>,
-                             Requires<[NotN64, HasStandardEncoding]>;
-  def _P8    : AtomicCmpSwap<Op, Width, CPURegs, CPU64Regs>,
-                             Requires<[IsN64, HasStandardEncoding]> {
+multiclass AtomicCmpSwap32<PatFrag Op>  {
+  def NAME : AtomicCmpSwap<Op, CPURegs, CPURegs>,
+             Requires<[NotN64, HasStdEnc]>;
+  def _P8  : AtomicCmpSwap<Op, CPURegs, CPU64Regs>,
+             Requires<[IsN64, HasStdEnc]> {
     let DecoderNamespace = "Mips64";
   }
 }
 
-class LLBase<bits<6> Opc, string opstring, RegisterClass RC, Operand Mem> :
-  FMem<Opc, (outs RC:$rt), (ins Mem:$addr),
-       !strconcat(opstring, "\t$rt, $addr"), [], IILoad> {
+class LLBase<string opstr, RegisterOperand RO, Operand Mem> :
+  InstSE<(outs RO:$rt), (ins Mem:$addr), !strconcat(opstr, "\t$rt, $addr"),
+         [], NoItinerary, FrmI> {
+  let DecoderMethod = "DecodeMem";
   let mayLoad = 1;
 }
 
-class SCBase<bits<6> Opc, string opstring, RegisterClass RC, Operand Mem> :
-  FMem<Opc, (outs RC:$dst), (ins RC:$rt, Mem:$addr),
-       !strconcat(opstring, "\t$rt, $addr"), [], IIStore> {
+class SCBase<string opstr, RegisterOperand RO, Operand Mem> :
+  InstSE<(outs RO:$dst), (ins RO:$rt, Mem:$addr),
+         !strconcat(opstr, "\t$rt, $addr"), [], NoItinerary, FrmI> {
+  let DecoderMethod = "DecodeMem";
   let mayStore = 1;
   let Constraints = "$rt = $dst";
 }
 
+class MFC3OP<dag outs, dag ins, string asmstr> :
+  InstSE<outs, ins, asmstr, [], NoItinerary, FrmFR>;
+
 //===----------------------------------------------------------------------===//
 // Pseudo instructions
 //===----------------------------------------------------------------------===//
 
 // Return RA.
 let isReturn=1, isTerminator=1, hasDelaySlot=1, isBarrier=1, hasCtrlDep=1 in
-def RetRA : PseudoSE<(outs), (ins), "", [(MipsRet)]>;
+def RetRA : PseudoSE<(outs), (ins), [(MipsRet)]>;
 
 let Defs = [SP], Uses = [SP], hasSideEffects = 1 in {
 def ADJCALLSTACKDOWN : MipsPseudo<(outs), (ins i32imm:$amt),
-                                  "!ADJCALLSTACKDOWN $amt",
                                   [(callseq_start timm:$amt)]>;
 def ADJCALLSTACKUP   : MipsPseudo<(outs), (ins i32imm:$amt1, i32imm:$amt2),
-                                  "!ADJCALLSTACKUP $amt1",
                                   [(callseq_end timm:$amt1, timm:$amt2)]>;
 }
 
-// When handling PIC code the assembler needs .cpload and .cprestore
-// directives. If the real instructions corresponding these directives
-// are used, we have the same behavior, but get also a bunch of warnings
-// from the assembler.
-let neverHasSideEffects = 1 in
-def CPRESTORE : PseudoSE<(outs), (ins i32imm:$loc, CPURegs:$gp),
-                         ".cprestore\t$loc", []>;
-
 let usesCustomInserter = 1 in {
-  defm ATOMIC_LOAD_ADD_I8   : Atomic2Ops32<atomic_load_add_8, "load_add_8">;
-  defm ATOMIC_LOAD_ADD_I16  : Atomic2Ops32<atomic_load_add_16, "load_add_16">;
-  defm ATOMIC_LOAD_ADD_I32  : Atomic2Ops32<atomic_load_add_32, "load_add_32">;
-  defm ATOMIC_LOAD_SUB_I8   : Atomic2Ops32<atomic_load_sub_8, "load_sub_8">;
-  defm ATOMIC_LOAD_SUB_I16  : Atomic2Ops32<atomic_load_sub_16, "load_sub_16">;
-  defm ATOMIC_LOAD_SUB_I32  : Atomic2Ops32<atomic_load_sub_32, "load_sub_32">;
-  defm ATOMIC_LOAD_AND_I8   : Atomic2Ops32<atomic_load_and_8, "load_and_8">;
-  defm ATOMIC_LOAD_AND_I16  : Atomic2Ops32<atomic_load_and_16, "load_and_16">;
-  defm ATOMIC_LOAD_AND_I32  : Atomic2Ops32<atomic_load_and_32, "load_and_32">;
-  defm ATOMIC_LOAD_OR_I8    : Atomic2Ops32<atomic_load_or_8, "load_or_8">;
-  defm ATOMIC_LOAD_OR_I16   : Atomic2Ops32<atomic_load_or_16, "load_or_16">;
-  defm ATOMIC_LOAD_OR_I32   : Atomic2Ops32<atomic_load_or_32, "load_or_32">;
-  defm ATOMIC_LOAD_XOR_I8   : Atomic2Ops32<atomic_load_xor_8, "load_xor_8">;
-  defm ATOMIC_LOAD_XOR_I16  : Atomic2Ops32<atomic_load_xor_16, "load_xor_16">;
-  defm ATOMIC_LOAD_XOR_I32  : Atomic2Ops32<atomic_load_xor_32, "load_xor_32">;
-  defm ATOMIC_LOAD_NAND_I8  : Atomic2Ops32<atomic_load_nand_8, "load_nand_8">;
-  defm ATOMIC_LOAD_NAND_I16 : Atomic2Ops32<atomic_load_nand_16, "load_nand_16">;
-  defm ATOMIC_LOAD_NAND_I32 : Atomic2Ops32<atomic_load_nand_32, "load_nand_32">;
-
-  defm ATOMIC_SWAP_I8       : Atomic2Ops32<atomic_swap_8, "swap_8">;
-  defm ATOMIC_SWAP_I16      : Atomic2Ops32<atomic_swap_16, "swap_16">;
-  defm ATOMIC_SWAP_I32      : Atomic2Ops32<atomic_swap_32, "swap_32">;
-
-  defm ATOMIC_CMP_SWAP_I8   : AtomicCmpSwap32<atomic_cmp_swap_8, "8">;
-  defm ATOMIC_CMP_SWAP_I16  : AtomicCmpSwap32<atomic_cmp_swap_16, "16">;
-  defm ATOMIC_CMP_SWAP_I32  : AtomicCmpSwap32<atomic_cmp_swap_32, "32">;
-}
+  defm ATOMIC_LOAD_ADD_I8   : Atomic2Ops32<atomic_load_add_8>;
+  defm ATOMIC_LOAD_ADD_I16  : Atomic2Ops32<atomic_load_add_16>;
+  defm ATOMIC_LOAD_ADD_I32  : Atomic2Ops32<atomic_load_add_32>;
+  defm ATOMIC_LOAD_SUB_I8   : Atomic2Ops32<atomic_load_sub_8>;
+  defm ATOMIC_LOAD_SUB_I16  : Atomic2Ops32<atomic_load_sub_16>;
+  defm ATOMIC_LOAD_SUB_I32  : Atomic2Ops32<atomic_load_sub_32>;
+  defm ATOMIC_LOAD_AND_I8   : Atomic2Ops32<atomic_load_and_8>;
+  defm ATOMIC_LOAD_AND_I16  : Atomic2Ops32<atomic_load_and_16>;
+  defm ATOMIC_LOAD_AND_I32  : Atomic2Ops32<atomic_load_and_32>;
+  defm ATOMIC_LOAD_OR_I8    : Atomic2Ops32<atomic_load_or_8>;
+  defm ATOMIC_LOAD_OR_I16   : Atomic2Ops32<atomic_load_or_16>;
+  defm ATOMIC_LOAD_OR_I32   : Atomic2Ops32<atomic_load_or_32>;
+  defm ATOMIC_LOAD_XOR_I8   : Atomic2Ops32<atomic_load_xor_8>;
+  defm ATOMIC_LOAD_XOR_I16  : Atomic2Ops32<atomic_load_xor_16>;
+  defm ATOMIC_LOAD_XOR_I32  : Atomic2Ops32<atomic_load_xor_32>;
+  defm ATOMIC_LOAD_NAND_I8  : Atomic2Ops32<atomic_load_nand_8>;
+  defm ATOMIC_LOAD_NAND_I16 : Atomic2Ops32<atomic_load_nand_16>;
+  defm ATOMIC_LOAD_NAND_I32 : Atomic2Ops32<atomic_load_nand_32>;
+
+  defm ATOMIC_SWAP_I8       : Atomic2Ops32<atomic_swap_8>;
+  defm ATOMIC_SWAP_I16      : Atomic2Ops32<atomic_swap_16>;
+  defm ATOMIC_SWAP_I32      : Atomic2Ops32<atomic_swap_32>;
+
+  defm ATOMIC_CMP_SWAP_I8   : AtomicCmpSwap32<atomic_cmp_swap_8>;
+  defm ATOMIC_CMP_SWAP_I16  : AtomicCmpSwap32<atomic_cmp_swap_16>;
+  defm ATOMIC_CMP_SWAP_I32  : AtomicCmpSwap32<atomic_cmp_swap_32>;
+}
+
+/// Pseudo instructions for loading, storing and copying accumulator registers.
+let isPseudo = 1 in {
+  defm LOAD_AC64  : LoadM<"load_ac64", ACRegs>;
+  defm STORE_AC64 : StoreM<"store_ac64", ACRegs>;
+}
+
+def COPY_AC64 : PseudoSE<(outs ACRegs:$dst), (ins ACRegs:$src), []>;
 
 //===----------------------------------------------------------------------===//
 // Instruction definition
 //===----------------------------------------------------------------------===//
-
-class LoadImm32< string instr_asm, Operand Od, RegisterClass RC> :
-  MipsAsmPseudoInst<(outs RC:$rt), (ins Od:$imm32),
-                     !strconcat(instr_asm, "\t$rt, $imm32")> ;
-def LoadImm32Reg : LoadImm32<"li", shamt,CPURegs>;
-
-class LoadAddress<string instr_asm, Operand MemOpnd, RegisterClass RC> :
-  MipsAsmPseudoInst<(outs RC:$rt), (ins MemOpnd:$addr),
-                     !strconcat(instr_asm, "\t$rt, $addr")> ;
-def LoadAddr32Reg : LoadAddress<"la", mem, CPURegs>;
-
-class LoadAddressImm<string instr_asm, Operand Od, RegisterClass RC> :
-  MipsAsmPseudoInst<(outs RC:$rt), (ins Od:$imm32),
-                     !strconcat(instr_asm, "\t$rt, $imm32")> ;
-def LoadAddr32Imm : LoadAddressImm<"la", shamt,CPURegs>;
-
 //===----------------------------------------------------------------------===//
 // MipsI Instructions
 //===----------------------------------------------------------------------===//
 
 /// Arithmetic Instructions (ALU Immediate)
-def ADDiu   : ArithLogicI<0x09, "addiu", add, simm16, immSExt16, CPURegs>,
-              IsAsCheapAsAMove;
-def ADDi    : ArithOverflowI<0x08, "addi", add, simm16, immSExt16, CPURegs>;
-def SLTi    : SetCC_I<0x0a, "slti", setlt, simm16, immSExt16, CPURegs>;
-def SLTiu   : SetCC_I<0x0b, "sltiu", setult, simm16, immSExt16, CPURegs>;
-def ANDi    : ArithLogicI<0x0c, "andi", and, uimm16, immZExt16, CPURegs>;
-def ORi     : ArithLogicI<0x0d, "ori", or, uimm16, immZExt16, CPURegs>;
-def XORi    : ArithLogicI<0x0e, "xori", xor, uimm16, immZExt16, CPURegs>;
-def LUi     : LoadUpper<0x0f, "lui", CPURegs, uimm16>;
+def ADDiu : ArithLogicI<"addiu", simm16, CPURegsOpnd, immSExt16, add>,
+            ADDI_FM<0x9>, IsAsCheapAsAMove;
+def ADDi  : ArithLogicI<"addi", simm16, CPURegsOpnd>, ADDI_FM<0x8>;
+def SLTi  : SetCC_I<"slti", setlt, simm16, immSExt16, CPURegs>, SLTI_FM<0xa>;
+def SLTiu : SetCC_I<"sltiu", setult, simm16, immSExt16, CPURegs>, SLTI_FM<0xb>;
+def ANDi  : ArithLogicI<"andi", uimm16, CPURegsOpnd, immZExt16, and>,
+            ADDI_FM<0xc>;
+def ORi   : ArithLogicI<"ori", uimm16, CPURegsOpnd, immZExt16, or>,
+            ADDI_FM<0xd>;
+def XORi  : ArithLogicI<"xori", uimm16, CPURegsOpnd, immZExt16, xor>,
+            ADDI_FM<0xe>;
+def LUi   : LoadUpper<"lui", CPURegs, uimm16>, LUI_FM;
 
 /// Arithmetic Instructions (3-Operand, R-Type)
-def ADDu    : ArithLogicR<0x00, 0x21, "addu", add, IIAlu, CPURegs, 1>;
-def SUBu    : ArithLogicR<0x00, 0x23, "subu", sub, IIAlu, CPURegs>;
-def ADD     : ArithOverflowR<0x00, 0x20, "add", IIAlu, CPURegs, 1>;
-def SUB     : ArithOverflowR<0x00, 0x22, "sub", IIAlu, CPURegs>;
-def SLT     : SetCC_R<0x00, 0x2a, "slt", setlt, CPURegs>;
-def SLTu    : SetCC_R<0x00, 0x2b, "sltu", setult, CPURegs>;
-def AND     : ArithLogicR<0x00, 0x24, "and", and, IIAlu, CPURegs, 1>;
-def OR      : ArithLogicR<0x00, 0x25, "or",  or, IIAlu, CPURegs, 1>;
-def XOR     : ArithLogicR<0x00, 0x26, "xor", xor, IIAlu, CPURegs, 1>;
-def NOR     : LogicNOR<0x00, 0x27, "nor", CPURegs>;
+def ADDu : ArithLogicR<"addu", CPURegsOpnd, 1, IIAlu, add>, ADD_FM<0, 0x21>;
+def SUBu : ArithLogicR<"subu", CPURegsOpnd, 0, IIAlu, sub>, ADD_FM<0, 0x23>;
+def MUL  : ArithLogicR<"mul", CPURegsOpnd, 1, IIImul, mul>, ADD_FM<0x1c, 2>;
+def ADD  : ArithLogicR<"add", CPURegsOpnd>, ADD_FM<0, 0x20>;
+def SUB  : ArithLogicR<"sub", CPURegsOpnd>, ADD_FM<0, 0x22>;
+def SLT  : SetCC_R<"slt", setlt, CPURegs>, ADD_FM<0, 0x2a>;
+def SLTu : SetCC_R<"sltu", setult, CPURegs>, ADD_FM<0, 0x2b>;
+def AND  : ArithLogicR<"and", CPURegsOpnd, 1, IIAlu, and>, ADD_FM<0, 0x24>;
+def OR   : ArithLogicR<"or", CPURegsOpnd, 1, IIAlu, or>, ADD_FM<0, 0x25>;
+def XOR  : ArithLogicR<"xor", CPURegsOpnd, 1, IIAlu, xor>, ADD_FM<0, 0x26>;
+def NOR  : LogicNOR<"nor", CPURegsOpnd>, ADD_FM<0, 0x27>;
 
 /// Shift Instructions
-def SLL     : shift_rotate_imm32<0x00, 0x00, "sll", shl>;
-def SRL     : shift_rotate_imm32<0x02, 0x00, "srl", srl>;
-def SRA     : shift_rotate_imm32<0x03, 0x00, "sra", sra>;
-def SLLV    : shift_rotate_reg<0x04, 0x00, "sllv", shl, CPURegs>;
-def SRLV    : shift_rotate_reg<0x06, 0x00, "srlv", srl, CPURegs>;
-def SRAV    : shift_rotate_reg<0x07, 0x00, "srav", sra, CPURegs>;
+def SLL  : shift_rotate_imm<"sll", shamt, CPURegsOpnd, shl, immZExt5>,
+           SRA_FM<0, 0>;
+def SRL  : shift_rotate_imm<"srl", shamt, CPURegsOpnd, srl, immZExt5>,
+           SRA_FM<2, 0>;
+def SRA  : shift_rotate_imm<"sra", shamt, CPURegsOpnd, sra, immZExt5>,
+           SRA_FM<3, 0>;
+def SLLV : shift_rotate_reg<"sllv", CPURegsOpnd, shl>, SRLV_FM<4, 0>;
+def SRLV : shift_rotate_reg<"srlv", CPURegsOpnd, srl>, SRLV_FM<6, 0>;
+def SRAV : shift_rotate_reg<"srav", CPURegsOpnd, sra>, SRLV_FM<7, 0>;
 
 // Rotate Instructions
-let Predicates = [HasMips32r2, HasStandardEncoding] in {
-    def ROTR    : shift_rotate_imm32<0x02, 0x01, "rotr", rotr>;
-    def ROTRV   : shift_rotate_reg<0x06, 0x01, "rotrv", rotr, CPURegs>;
+let Predicates = [HasMips32r2, HasStdEnc] in {
+  def ROTR  : shift_rotate_imm<"rotr", shamt, CPURegsOpnd, rotr, immZExt5>,
+              SRA_FM<2, 1>;
+  def ROTRV : shift_rotate_reg<"rotrv", CPURegsOpnd, rotr>, SRLV_FM<6, 1>;
 }
 
 /// Load and Store Instructions
 ///  aligned
-defm LB      : LoadM32<0x20, "lb",  sextloadi8>;
-defm LBu     : LoadM32<0x24, "lbu", zextloadi8>;
-defm LH      : LoadM32<0x21, "lh",  sextloadi16>;
-defm LHu     : LoadM32<0x25, "lhu", zextloadi16>;
-defm LW      : LoadM32<0x23, "lw",  load>;
-defm SB      : StoreM32<0x28, "sb", truncstorei8>;
-defm SH      : StoreM32<0x29, "sh", truncstorei16>;
-defm SW      : StoreM32<0x2b, "sw", store>;
+defm LB  : LoadM<"lb", CPURegs, sextloadi8>, LW_FM<0x20>;
+defm LBu : LoadM<"lbu", CPURegs, zextloadi8, addrDefault>, LW_FM<0x24>;
+defm LH  : LoadM<"lh", CPURegs, sextloadi16, addrDefault>, LW_FM<0x21>;
+defm LHu : LoadM<"lhu", CPURegs, zextloadi16>, LW_FM<0x25>;
+defm LW  : LoadM<"lw", CPURegs, load, addrDefault>, LW_FM<0x23>;
+defm SB  : StoreM<"sb", CPURegs, truncstorei8>, LW_FM<0x28>;
+defm SH  : StoreM<"sh", CPURegs, truncstorei16>, LW_FM<0x29>;
+defm SW  : StoreM<"sw", CPURegs, store>, LW_FM<0x2b>;
 
 /// load/store left/right
-defm LWL : LoadLeftRightM32<0x22, "lwl", MipsLWL>;
-defm LWR : LoadLeftRightM32<0x26, "lwr", MipsLWR>;
-defm SWL : StoreLeftRightM32<0x2a, "swl", MipsSWL>;
-defm SWR : StoreLeftRightM32<0x2e, "swr", MipsSWR>;
+defm LWL : LoadLeftRightM<"lwl", MipsLWL, CPURegs>, LW_FM<0x22>;
+defm LWR : LoadLeftRightM<"lwr", MipsLWR, CPURegs>, LW_FM<0x26>;
+defm SWL : StoreLeftRightM<"swl", MipsSWL, CPURegs>, LW_FM<0x2a>;
+defm SWR : StoreLeftRightM<"swr", MipsSWR, CPURegs>, LW_FM<0x2e>;
 
-let hasSideEffects = 1 in
-def SYNC : InstSE<(outs), (ins i32imm:$stype), "sync $stype",
-                  [(MipsSync imm:$stype)], NoItinerary, FrmOther>
-{
-  bits<5> stype;
-  let Opcode = 0;
-  let Inst{25-11} = 0;
-  let Inst{10-6} = stype;
-  let Inst{5-0} = 15;
-}
+def SYNC : SYNC_FT, SYNC_FM;
 
 /// Load-linked, Store-conditional
-def LL    : LLBase<0x30, "ll", CPURegs, mem>,
-            Requires<[NotN64, HasStandardEncoding]>;
-def LL_P8 : LLBase<0x30, "ll", CPURegs, mem64>,
-            Requires<[IsN64, HasStandardEncoding]> {
-  let DecoderNamespace = "Mips64";
+let Predicates = [NotN64, HasStdEnc] in {
+  def LL : LLBase<"ll", CPURegsOpnd, mem>, LW_FM<0x30>;
+  def SC : SCBase<"sc", CPURegsOpnd, mem>, LW_FM<0x38>;
 }
 
-def SC    : SCBase<0x38, "sc", CPURegs, mem>,
-            Requires<[NotN64, HasStandardEncoding]>;
-def SC_P8 : SCBase<0x38, "sc", CPURegs, mem64>,
-            Requires<[IsN64, HasStandardEncoding]> {
-  let DecoderNamespace = "Mips64";
+let Predicates = [IsN64, HasStdEnc], DecoderNamespace = "Mips64" in {
+  def LL_P8 : LLBase<"ll", CPURegsOpnd, mem64>, LW_FM<0x30>;
+  def SC_P8 : SCBase<"sc", CPURegsOpnd, mem64>, LW_FM<0x38>;
 }
 
 /// Jump and Branch Instructions
-def J       : JumpFJ<0x02, jmptarget, "j", br, bb>,
-              Requires<[RelocStatic, HasStandardEncoding]>, IsBranch;
-def JR      : IndirectBranch<CPURegs>;
-def B       : UncondBranch<0x04, "b">;
-def BEQ     : CBranch<0x04, "beq", seteq, CPURegs>;
-def BNE     : CBranch<0x05, "bne", setne, CPURegs>;
-def BGEZ    : CBranchZero<0x01, 1, "bgez", setge, CPURegs>;
-def BGTZ    : CBranchZero<0x07, 0, "bgtz", setgt, CPURegs>;
-def BLEZ    : CBranchZero<0x06, 0, "blez", setle, CPURegs>;
-def BLTZ    : CBranchZero<0x01, 0, "bltz", setlt, CPURegs>;
-
-let rt = 0, rs = 0, isBranch = 1, isTerminator = 1, isBarrier = 1,
-    hasDelaySlot = 1, Defs = [RA] in
-def BAL_BR: FI<0x1, (outs), (ins brtarget:$imm16), "bal\t$imm16", [], IIBranch>;
-
-def JAL  : JumpLink<0x03, "jal">;
-def JALR : JumpLinkReg<0x00, 0x09, "jalr", CPURegs>;
-def BGEZAL  : BranchLink<"bgezal", 0x11, CPURegs>;
-def BLTZAL  : BranchLink<"bltzal", 0x10, CPURegs>;
-def TAILCALL : JumpFJ<0x02, calltarget, "j", MipsTailCall, imm>, IsTailCall;
-def TAILCALL_R : JumpFR<CPURegs, MipsTailCall>, IsTailCall;
-
-def RET : RetBase<CPURegs>;
+def J       : JumpFJ<jmptarget, "j", br, bb>, FJ<2>,
+              Requires<[RelocStatic, HasStdEnc]>, IsBranch;
+def JR      : IndirectBranch<CPURegs>, MTLO_FM<8>;
+def B       : UncondBranch<"b">, B_FM;
+def BEQ     : CBranch<"beq", seteq, CPURegs>, BEQ_FM<4>;
+def BNE     : CBranch<"bne", setne, CPURegs>, BEQ_FM<5>;
+def BGEZ    : CBranchZero<"bgez", setge, CPURegs>, BGEZ_FM<1, 1>;
+def BGTZ    : CBranchZero<"bgtz", setgt, CPURegs>, BGEZ_FM<7, 0>;
+def BLEZ    : CBranchZero<"blez", setle, CPURegs>, BGEZ_FM<6, 0>;
+def BLTZ    : CBranchZero<"bltz", setlt, CPURegs>, BGEZ_FM<1, 0>;
+
+def BAL_BR: BAL_FT, BAL_FM;
+
+def JAL  : JumpLink<"jal">, FJ<3>;
+def JALR : JumpLinkReg<"jalr", CPURegs>, JALR_FM;
+def JALRPseudo : JumpLinkRegPseudo<CPURegs, JALR, RA>;
+def BGEZAL : BGEZAL_FT<"bgezal", CPURegsOpnd>, BGEZAL_FM<0x11>;
+def BLTZAL : BGEZAL_FT<"bltzal", CPURegsOpnd>, BGEZAL_FM<0x10>;
+def TAILCALL : JumpFJ<calltarget, "j", MipsTailCall, imm>, FJ<2>, IsTailCall;
+def TAILCALL_R : JumpFR<CPURegs, MipsTailCall>, MTLO_FM<8>, IsTailCall;
+
+def RET : RetBase<CPURegs>, MTLO_FM<8>;
+
+// Exception handling related node and instructions.
+// The conversion sequence is:
+// ISD::EH_RETURN -> MipsISD::EH_RETURN ->
+// MIPSeh_return -> (stack change + indirect branch)
+//
+// MIPSeh_return takes the place of regular return instruction
+// but takes two arguments (V1, V0) which are used for storing
+// the offset and return address respectively.
+def SDT_MipsEHRET : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisPtrTy<1>]>;
 
-/// Multiply and Divide Instructions.
-def MULT    : Mult32<0x18, "mult", IIImul>;
-def MULTu   : Mult32<0x19, "multu", IIImul>;
-def SDIV    : Div32<MipsDivRem, 0x1a, "div", IIIdiv>;
-def UDIV    : Div32<MipsDivRemU, 0x1b, "divu", IIIdiv>;
+def MIPSehret : SDNode<"MipsISD::EH_RETURN", SDT_MipsEHRET,
+                      [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
+
+let Uses = [V0, V1], isTerminator = 1, isReturn = 1, isBarrier = 1 in {
+  def MIPSeh_return32 : MipsPseudo<(outs), (ins CPURegs:$spoff, CPURegs:$dst),
+                                [(MIPSehret CPURegs:$spoff, CPURegs:$dst)]>;
+  def MIPSeh_return64 : MipsPseudo<(outs), (ins CPU64Regs:$spoff,
+                                                CPU64Regs:$dst),
+                                [(MIPSehret CPU64Regs:$spoff, CPU64Regs:$dst)]>;
+}
 
-def MTHI : MoveToLOHI<0x11, "mthi", CPURegs, [HI]>;
-def MTLO : MoveToLOHI<0x13, "mtlo", CPURegs, [LO]>;
-def MFHI : MoveFromLOHI<0x10, "mfhi", CPURegs, [HI]>;
-def MFLO : MoveFromLOHI<0x12, "mflo", CPURegs, [LO]>;
+/// Multiply and Divide Instructions.
+def MULT  : Mult<"mult", IIImul, CPURegsOpnd, [HI, LO]>, MULT_FM<0, 0x18>;
+def MULTu : Mult<"multu", IIImul, CPURegsOpnd, [HI, LO]>, MULT_FM<0, 0x19>;
+def PseudoMULT  : MultDivPseudo<MULT, ACRegs, CPURegsOpnd, MipsMult, IIImul>;
+def PseudoMULTu : MultDivPseudo<MULTu, ACRegs, CPURegsOpnd, MipsMultu, IIImul>;
+def SDIV  : Div<"div", IIIdiv, CPURegsOpnd, [HI, LO]>, MULT_FM<0, 0x1a>;
+def UDIV  : Div<"divu", IIIdiv, CPURegsOpnd, [HI, LO]>, MULT_FM<0, 0x1b>;
+def PseudoSDIV : MultDivPseudo<SDIV, ACRegs, CPURegsOpnd, MipsDivRem, IIIdiv, 0>;
+def PseudoUDIV : MultDivPseudo<UDIV, ACRegs, CPURegsOpnd, MipsDivRemU, IIIdiv,
+                               0>;
+
+def MTHI : MoveToLOHI<"mthi", CPURegs, [HI]>, MTLO_FM<0x11>;
+def MTLO : MoveToLOHI<"mtlo", CPURegs, [LO]>, MTLO_FM<0x13>;
+def MFHI : MoveFromLOHI<"mfhi", CPURegs, [HI]>, MFLO_FM<0x10>;
+def MFLO : MoveFromLOHI<"mflo", CPURegs, [LO]>, MFLO_FM<0x12>;
 
 /// Sign Ext In Register Instructions.
-def SEB : SignExtInReg<0x10, "seb", i8, CPURegs>;
-def SEH : SignExtInReg<0x18, "seh", i16, CPURegs>;
+def SEB : SignExtInReg<"seb", i8, CPURegs>, SEB_FM<0x10, 0x20>;
+def SEH : SignExtInReg<"seh", i16, CPURegs>, SEB_FM<0x18, 0x20>;
 
 /// Count Leading
-def CLZ : CountLeading0<0x20, "clz", CPURegs>;
-def CLO : CountLeading1<0x21, "clo", CPURegs>;
+def CLZ : CountLeading0<"clz", CPURegsOpnd>, CLO_FM<0x20>;
+def CLO : CountLeading1<"clo", CPURegsOpnd>, CLO_FM<0x21>;
 
 /// Word Swap Bytes Within Halfwords
-def WSBH : SubwordSwap<0x20, 0x2, "wsbh", CPURegs>;
+def WSBH : SubwordSwap<"wsbh", CPURegsOpnd>, SEB_FM<2, 0x20>;
 
-/// No operation
-let addr=0 in
-  def NOP   : FJ<0, (outs), (ins), "nop", [], IIAlu>;
+/// No operation.
+def NOP : PseudoSE<(outs), (ins), []>, PseudoInstExpansion<(SLL ZERO, ZERO, 0)>;
 
 // FrameIndexes are legalized when they are operands from load/store
 // instructions. The same not happens for stack address copies, so an
 // add op with mem ComplexPattern is used and the stack address copy
 // can be matched. It's similar to Sparc LEA_ADDRi
-def LEA_ADDiu : EffectiveAddress<0x09,"addiu\t$rt, $addr", CPURegs, mem_ea>;
-
-// DynAlloc node points to dynamically allocated stack space.
-// $sp is added to the list of implicitly used registers to prevent dead code
-// elimination from removing instructions that modify $sp.
-let Uses = [SP] in
-def DynAlloc : EffectiveAddress<0x09,"addiu\t$rt, $addr", CPURegs, mem_ea>;
+def LEA_ADDiu : EffectiveAddress<"addiu", CPURegs, mem_ea>, LW_FM<9>;
 
 // MADD*/MSUB*
-def MADD  : MArithR<0, "madd", MipsMAdd, 1>;
-def MADDU : MArithR<1, "maddu", MipsMAddu, 1>;
-def MSUB  : MArithR<4, "msub", MipsMSub>;
-def MSUBU : MArithR<5, "msubu", MipsMSubu>;
+def MADD  : MArithR<"madd", 1>, MULT_FM<0x1c, 0>;
+def MADDU : MArithR<"maddu", 1>, MULT_FM<0x1c, 1>;
+def MSUB  : MArithR<"msub">, MULT_FM<0x1c, 4>;
+def MSUBU : MArithR<"msubu">, MULT_FM<0x1c, 5>;
+def PseudoMADD  : MAddSubPseudo<MADD, MipsMAdd>;
+def PseudoMADDU : MAddSubPseudo<MADDU, MipsMAddu>;
+def PseudoMSUB  : MAddSubPseudo<MSUB, MipsMSub>;
+def PseudoMSUBU : MAddSubPseudo<MSUBU, MipsMSubu>;
 
-// MUL is a assembly macro in the current used ISAs. In recent ISA's
-// it is a real instruction.
-def MUL   : ArithLogicR<0x1c, 0x02, "mul", mul, IIImul, CPURegs, 1>,
-            Requires<[HasMips32, HasStandardEncoding]>;
+def RDHWR : ReadHardware<CPURegs, HWRegsOpnd>, RDHWR_FM;
 
-def RDHWR : ReadHardware<CPURegs, HWRegs>;
+def EXT : ExtBase<"ext", CPURegsOpnd>, EXT_FM<0>;
+def INS : InsBase<"ins", CPURegsOpnd>, EXT_FM<4>;
 
-def EXT : ExtBase<0, "ext", CPURegs>;
-def INS : InsBase<4, "ins", CPURegs>;
+/// Move Control Registers From/To CPU Registers
+def MFC0_3OP : MFC3OP<(outs CPURegsOpnd:$rt),
+                      (ins CPURegsOpnd:$rd, uimm16:$sel),
+                      "mfc0\t$rt, $rd, $sel">, MFC3OP_FM<0x10, 0>;
+
+def MTC0_3OP : MFC3OP<(outs CPURegsOpnd:$rd, uimm16:$sel),
+                      (ins CPURegsOpnd:$rt),
+                      "mtc0\t$rt, $rd, $sel">, MFC3OP_FM<0x10, 4>;
+
+def MFC2_3OP : MFC3OP<(outs CPURegsOpnd:$rt),
+                      (ins CPURegsOpnd:$rd, uimm16:$sel),
+                      "mfc2\t$rt, $rd, $sel">, MFC3OP_FM<0x12, 0>;
+
+def MTC2_3OP : MFC3OP<(outs CPURegsOpnd:$rd, uimm16:$sel),
+                      (ins CPURegsOpnd:$rt),
+                      "mtc2\t$rt, $rd, $sel">, MFC3OP_FM<0x12, 4>;
 
 //===----------------------------------------------------------------------===//
 // Instruction aliases
 //===----------------------------------------------------------------------===//
-def : InstAlias<"move $dst,$src", (ADD CPURegs:$dst,CPURegs:$src,ZERO)>;
-def : InstAlias<"bal $offset", (BGEZAL RA,brtarget:$offset)>;
-def : InstAlias<"addu $rs,$rt,$imm",
-                (ADDiu CPURegs:$rs,CPURegs:$rt,simm16:$imm)>;
-def : InstAlias<"add $rs,$rt,$imm",
-                (ADDi CPURegs:$rs,CPURegs:$rt,simm16:$imm)>;
-def : InstAlias<"and $rs,$rt,$imm",
-                (ANDi CPURegs:$rs,CPURegs:$rt,simm16:$imm)>;
-def : InstAlias<"j $rs", (JR CPURegs:$rs)>;
-def : InstAlias<"not $rt,$rs", (NOR CPURegs:$rt,CPURegs:$rs,ZERO)>;
-def : InstAlias<"neg $rt,$rs", (SUB CPURegs:$rt,ZERO,CPURegs:$rs)>;
-def : InstAlias<"negu $rt,$rs", (SUBu CPURegs:$rt,ZERO,CPURegs:$rs)>;
-def : InstAlias<"slt $rs,$rt,$imm",
-                (SLTi CPURegs:$rs,CPURegs:$rt,simm16:$imm)>;
-def : InstAlias<"xor $rs,$rt,$imm",
-                (XORi CPURegs:$rs,CPURegs:$rt,simm16:$imm)>;
+def : InstAlias<"move $dst, $src",
+                (ADDu CPURegsOpnd:$dst, CPURegsOpnd:$src,ZERO), 1>,
+      Requires<[NotMips64]>;
+def : InstAlias<"move $dst, $src",
+                (OR CPURegsOpnd:$dst, CPURegsOpnd:$src,ZERO), 1>,
+      Requires<[NotMips64]>;
+def : InstAlias<"bal $offset", (BGEZAL RA, brtarget:$offset), 1>;
+def : InstAlias<"addu $rs, $rt, $imm",
+                (ADDiu CPURegsOpnd:$rs, CPURegsOpnd:$rt, simm16:$imm), 0>;
+def : InstAlias<"add $rs, $rt, $imm",
+                (ADDi CPURegsOpnd:$rs, CPURegsOpnd:$rt, simm16:$imm), 0>;
+def : InstAlias<"and $rs, $rt, $imm",
+                (ANDi CPURegsOpnd:$rs, CPURegsOpnd:$rt, simm16:$imm), 0>;
+def : InstAlias<"j $rs", (JR CPURegs:$rs), 0>,
+      Requires<[NotMips64]>;
+def : InstAlias<"jalr $rs", (JALR RA, CPURegs:$rs)>, Requires<[NotMips64]>;
+def : InstAlias<"jal $rs", (JALR RA, CPURegs:$rs), 0>, Requires<[NotMips64]>;
+def : InstAlias<"jal $rd,$rs", (JALR CPURegs:$rd, CPURegs:$rs), 0>,
+                 Requires<[NotMips64]>;
+def : InstAlias<"not $rt, $rs",
+                (NOR CPURegsOpnd:$rt, CPURegsOpnd:$rs, ZERO), 1>;
+def : InstAlias<"neg $rt, $rs",
+                (SUB CPURegsOpnd:$rt, ZERO, CPURegsOpnd:$rs), 1>;
+def : InstAlias<"negu $rt, $rs",
+                (SUBu CPURegsOpnd:$rt, ZERO, CPURegsOpnd:$rs), 1>;
+def : InstAlias<"slt $rs, $rt, $imm",
+                (SLTi CPURegsOpnd:$rs, CPURegs:$rt, simm16:$imm), 0>;
+def : InstAlias<"xor $rs, $rt, $imm",
+                (XORi CPURegsOpnd:$rs, CPURegsOpnd:$rt, simm16:$imm), 0>,
+      Requires<[NotMips64]>;
+def : InstAlias<"or $rs, $rt, $imm",
+                (ORi CPURegsOpnd:$rs, CPURegsOpnd:$rt, simm16:$imm), 0>,
+                 Requires<[NotMips64]>;
+def : InstAlias<"nop", (SLL ZERO, ZERO, 0), 1>;
+def : InstAlias<"mfc0 $rt, $rd",
+                (MFC0_3OP CPURegsOpnd:$rt, CPURegsOpnd:$rd, 0), 0>;
+def : InstAlias<"mtc0 $rt, $rd",
+                (MTC0_3OP CPURegsOpnd:$rd, 0, CPURegsOpnd:$rt), 0>;
+def : InstAlias<"mfc2 $rt, $rd",
+                (MFC2_3OP CPURegsOpnd:$rt, CPURegsOpnd:$rd, 0), 0>;
+def : InstAlias<"mtc2 $rt, $rd",
+                (MTC2_3OP CPURegsOpnd:$rd, 0, CPURegsOpnd:$rt), 0>;
+
+//===----------------------------------------------------------------------===//
+// Assembler Pseudo Instructions
+//===----------------------------------------------------------------------===//
+
+class LoadImm32< string instr_asm, Operand Od, RegisterOperand RO> :
+  MipsAsmPseudoInst<(outs RO:$rt), (ins Od:$imm32),
+                     !strconcat(instr_asm, "\t$rt, $imm32")> ;
+def LoadImm32Reg : LoadImm32<"li", shamt,CPURegsOpnd>;
+
+class LoadAddress<string instr_asm, Operand MemOpnd, RegisterOperand RO> :
+  MipsAsmPseudoInst<(outs RO:$rt), (ins MemOpnd:$addr),
+                     !strconcat(instr_asm, "\t$rt, $addr")> ;
+def LoadAddr32Reg : LoadAddress<"la", mem, CPURegsOpnd>;
+
+class LoadAddressImm<string instr_asm, Operand Od, RegisterOperand RO> :
+  MipsAsmPseudoInst<(outs RO:$rt), (ins Od:$imm32),
+                     !strconcat(instr_asm, "\t$rt, $imm32")> ;
+def LoadAddr32Imm : LoadAddressImm<"la", shamt,CPURegsOpnd>;
+
+
 
 //===----------------------------------------------------------------------===//
 //  Arbitrary patterns that map to one or more instructions
 //===----------------------------------------------------------------------===//
 
+// Load/store pattern templates.
+class LoadRegImmPat<Instruction LoadInst, ValueType ValTy, PatFrag Node> :
+  MipsPat<(ValTy (Node addrRegImm:$a)), (LoadInst addrRegImm:$a)>;
+
+class StoreRegImmPat<Instruction StoreInst, ValueType ValTy> :
+  MipsPat<(store ValTy:$v, addrRegImm:$a), (StoreInst ValTy:$v, addrRegImm:$a)>;
+
 // Small immediates
 def : MipsPat<(i32 immSExt16:$in),
               (ADDiu ZERO, imm:$in)>;
@@ -1194,25 +1192,25 @@ def : WrapperPat<tglobaltlsaddr, ADDiu, CPURegs>;
 
 // Mips does not have "not", so we expand our way
 def : MipsPat<(not CPURegs:$in),
-              (NOR CPURegs:$in, ZERO)>;
+              (NOR CPURegsOpnd:$in, ZERO)>;
 
 // extended loads
-let Predicates = [NotN64, HasStandardEncoding] in {
+let Predicates = [NotN64, HasStdEnc] in {
   def : MipsPat<(i32 (extloadi1  addr:$src)), (LBu addr:$src)>;
   def : MipsPat<(i32 (extloadi8  addr:$src)), (LBu addr:$src)>;
   def : MipsPat<(i32 (extloadi16 addr:$src)), (LHu addr:$src)>;
 }
-let Predicates = [IsN64, HasStandardEncoding] in {
+let Predicates = [IsN64, HasStdEnc] in {
   def : MipsPat<(i32 (extloadi1  addr:$src)), (LBu_P8 addr:$src)>;
   def : MipsPat<(i32 (extloadi8  addr:$src)), (LBu_P8 addr:$src)>;
   def : MipsPat<(i32 (extloadi16 addr:$src)), (LHu_P8 addr:$src)>;
 }
 
 // peepholes
-let Predicates = [NotN64, HasStandardEncoding] in {
+let Predicates = [NotN64, HasStdEnc] in {
   def : MipsPat<(store (i32 0), addr:$dst), (SW ZERO, addr:$dst)>;
 }
-let Predicates = [IsN64, HasStandardEncoding] in {
+let Predicates = [IsN64, HasStdEnc] in {
   def : MipsPat<(store (i32 0), addr:$dst), (SW_P8 ZERO, addr:$dst)>;
 }
 
@@ -1289,12 +1287,27 @@ defm : SetgtPats<CPURegs, SLT, SLTu>;
 defm : SetgePats<CPURegs, SLT, SLTu>;
 defm : SetgeImmPats<CPURegs, SLTi, SLTiu>;
 
-// select MipsDynAlloc
-def : MipsPat<(MipsDynAlloc addr:$f), (DynAlloc addr:$f)>;
-
 // bswap pattern
 def : MipsPat<(bswap CPURegs:$rt), (ROTR (WSBH CPURegs:$rt), 16)>;
 
+// mflo/hi patterns.
+def : MipsPat<(i32 (ExtractLOHI ACRegs:$ac, imm:$lohi_idx)),
+              (EXTRACT_SUBREG ACRegs:$ac, imm:$lohi_idx)>;
+
+// Load halfword/word patterns.
+let AddedComplexity = 40 in {
+  let Predicates = [NotN64, HasStdEnc] in {
+    def : LoadRegImmPat<LBu, i32, zextloadi8>;
+    def : LoadRegImmPat<LH, i32, sextloadi16>;
+    def : LoadRegImmPat<LW, i32, load>;
+  }
+  let Predicates = [IsN64, HasStdEnc] in {
+    def : LoadRegImmPat<LBu_P8, i32, zextloadi8>;
+    def : LoadRegImmPat<LH_P8, i32, sextloadi16>;
+    def : LoadRegImmPat<LW_P8, i32, load>;
+  }
+}
+
 //===----------------------------------------------------------------------===//
 // Floating Point Support
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/Mips/MipsJITInfo.cpp b/lib/Target/Mips/MipsJITInfo.cpp
index da1119df8f9f..1b2a325d3ce6 100644
--- a/lib/Target/Mips/MipsJITInfo.cpp
+++ b/lib/Target/Mips/MipsJITInfo.cpp
@@ -16,12 +16,12 @@
 #include "MipsInstrInfo.h"
 #include "MipsRelocations.h"
 #include "MipsSubtarget.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/JITCodeEmitter.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/raw_ostream.h"
 #include "llvm/Support/Memory.h"
+#include "llvm/Support/raw_ostream.h"
 #include <cstdlib>
 using namespace llvm;
 
diff --git a/lib/Target/Mips/MipsLongBranch.cpp b/lib/Target/Mips/MipsLongBranch.cpp
index 5d9f0cffb749..2efe534053a2 100644
--- a/lib/Target/Mips/MipsLongBranch.cpp
+++ b/lib/Target/Mips/MipsLongBranch.cpp
@@ -10,21 +10,21 @@
 // This pass expands a branch or jump instruction into a long branch if its
 // offset is too large to fit into its immediate field.
 //
-// FIXME: 
-// 1. Fix pc-region jump instructions which cross 256MB segment boundaries. 
+// FIXME:
+// 1. Fix pc-region jump instructions which cross 256MB segment boundaries.
 // 2. If program has inline assembly statements whose size cannot be
-//    determined accurately, load branch target addresses from the GOT. 
+//    determined accurately, load branch target addresses from the GOT.
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "mips-long-branch"
 
 #include "Mips.h"
-#include "MipsTargetMachine.h"
 #include "MCTargetDesc/MipsBaseInfo.h"
+#include "MipsTargetMachine.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/Function.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Target/TargetInstrInfo.h"
@@ -258,7 +258,8 @@ void MipsLongBranch::expandToLongBranch(MBBInfo &I) {
     BalTgtMBB->addSuccessor(TgtMBB);
 
     int64_t TgtAddress = MBBInfos[TgtMBB->getNumber()].Address;
-    int64_t Offset = TgtAddress - (I.Address + I.Size - 20);
+    unsigned BalTgtMBBSize = 5;
+    int64_t Offset = TgtAddress - (I.Address + I.Size - BalTgtMBBSize * 4);
     int64_t Lo = SignExtend64<16>(Offset & 0xffff);
     int64_t Hi = SignExtend64<16>(((Offset + 0x8000) >> 16) & 0xffff);
 
@@ -283,9 +284,10 @@ void MipsLongBranch::expandToLongBranch(MBBInfo &I) {
         .addReg(Mips::SP).addImm(-8);
       BuildMI(*LongBrMBB, Pos, DL, TII->get(Mips::SW)).addReg(Mips::RA)
         .addReg(Mips::SP).addImm(0);
-      BuildMI(*LongBrMBB, Pos, DL, TII->get(Mips::BAL_BR)).addMBB(BalTgtMBB);
-      BuildMI(*LongBrMBB, Pos, DL, TII->get(Mips::LUi), Mips::AT).addImm(Hi)
-        ->setIsInsideBundle();
+
+      MIBundleBuilder(*LongBrMBB, Pos)
+        .append(BuildMI(*MF, DL, TII->get(Mips::BAL_BR)).addMBB(BalTgtMBB))
+        .append(BuildMI(*MF, DL, TII->get(Mips::LUi), Mips::AT).addImm(Hi));
 
       Pos = BalTgtMBB->begin();
 
@@ -295,9 +297,11 @@ void MipsLongBranch::expandToLongBranch(MBBInfo &I) {
         .addReg(Mips::RA).addReg(Mips::AT);
       BuildMI(*BalTgtMBB, Pos, DL, TII->get(Mips::LW), Mips::RA)
         .addReg(Mips::SP).addImm(0);
-      BuildMI(*BalTgtMBB, Pos, DL, TII->get(Mips::JR)).addReg(Mips::AT);
-      BuildMI(*BalTgtMBB, Pos, DL, TII->get(Mips::ADDiu), Mips::SP)
-        .addReg(Mips::SP).addImm(8)->setIsInsideBundle();
+
+      MIBundleBuilder(*BalTgtMBB, Pos)
+        .append(BuildMI(*MF, DL, TII->get(Mips::JR)).addReg(Mips::AT))
+        .append(BuildMI(*MF, DL, TII->get(Mips::ADDiu), Mips::SP)
+                .addReg(Mips::SP).addImm(8));
     } else {
       // $longbr:
       //  daddiu $sp, $sp, -16
@@ -335,9 +339,11 @@ void MipsLongBranch::expandToLongBranch(MBBInfo &I) {
         .addReg(Mips::AT_64).addImm(16);
       BuildMI(*LongBrMBB, Pos, DL, TII->get(Mips::DADDiu), Mips::AT_64)
         .addReg(Mips::AT_64).addImm(Hi);
-      BuildMI(*LongBrMBB, Pos, DL, TII->get(Mips::BAL_BR)).addMBB(BalTgtMBB);
-      BuildMI(*LongBrMBB, Pos, DL, TII->get(Mips::DSLL), Mips::AT_64)
-        .addReg(Mips::AT_64).addImm(16)->setIsInsideBundle();
+
+      MIBundleBuilder(*LongBrMBB, Pos)
+        .append(BuildMI(*MF, DL, TII->get(Mips::BAL_BR)).addMBB(BalTgtMBB))
+        .append(BuildMI(*MF, DL, TII->get(Mips::DSLL), Mips::AT_64)
+                .addReg(Mips::AT_64).addImm(16));
 
       Pos = BalTgtMBB->begin();
 
@@ -347,10 +353,15 @@ void MipsLongBranch::expandToLongBranch(MBBInfo &I) {
         .addReg(Mips::RA_64).addReg(Mips::AT_64);
       BuildMI(*BalTgtMBB, Pos, DL, TII->get(Mips::LD), Mips::RA_64)
         .addReg(Mips::SP_64).addImm(0);
-      BuildMI(*BalTgtMBB, Pos, DL, TII->get(Mips::JR64)).addReg(Mips::AT_64);
-      BuildMI(*BalTgtMBB, Pos, DL, TII->get(Mips::DADDiu), Mips::SP_64)
-        .addReg(Mips::SP_64).addImm(16)->setIsInsideBundle();
+
+      MIBundleBuilder(*BalTgtMBB, Pos)
+        .append(BuildMI(*MF, DL, TII->get(Mips::JR64)).addReg(Mips::AT_64))
+        .append(BuildMI(*MF, DL, TII->get(Mips::DADDiu), Mips::SP_64)
+                .addReg(Mips::SP_64).addImm(16));
     }
+
+    assert(BalTgtMBBSize == BalTgtMBB->size());
+    assert(LongBrMBB->size() + BalTgtMBBSize == LongBranchSeqSize);
   } else {
     // $longbr:
     //  j $tgt
@@ -359,8 +370,11 @@ void MipsLongBranch::expandToLongBranch(MBBInfo &I) {
     //
     Pos = LongBrMBB->begin();
     LongBrMBB->addSuccessor(TgtMBB);
-    BuildMI(*LongBrMBB, Pos, DL, TII->get(Mips::J)).addMBB(TgtMBB);
-    BuildMI(*LongBrMBB, Pos, DL, TII->get(Mips::NOP))->setIsInsideBundle();
+    MIBundleBuilder(*LongBrMBB, Pos)
+      .append(BuildMI(*MF, DL, TII->get(Mips::J)).addMBB(TgtMBB))
+      .append(BuildMI(*MF, DL, TII->get(Mips::NOP)));
+
+    assert(LongBrMBB->size() == LongBranchSeqSize);
   }
 
   if (I.Br->isUnconditionalBranch()) {
diff --git a/lib/Target/Mips/MipsMCInstLower.cpp b/lib/Target/Mips/MipsMCInstLower.cpp
index 4162f981d1df..d836975eb7d2 100644
--- a/lib/Target/Mips/MipsMCInstLower.cpp
+++ b/lib/Target/Mips/MipsMCInstLower.cpp
@@ -12,9 +12,9 @@
 //
 //===----------------------------------------------------------------------===//
 #include "MipsMCInstLower.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
 #include "MipsAsmPrinter.h"
 #include "MipsInstrInfo.h"
-#include "MCTargetDesc/MipsBaseInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineOperand.h"
diff --git a/lib/Target/Mips/MipsMachineFunction.cpp b/lib/Target/Mips/MipsMachineFunction.cpp
index 5ff19aba0267..59b23f7ad7c1 100644
--- a/lib/Target/Mips/MipsMachineFunction.cpp
+++ b/lib/Target/Mips/MipsMachineFunction.cpp
@@ -8,12 +8,12 @@
 //===----------------------------------------------------------------------===//
 
 #include "MipsMachineFunction.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
 #include "MipsInstrInfo.h"
 #include "MipsSubtarget.h"
-#include "MCTargetDesc/MipsBaseInfo.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/CommandLine.h"
 
 using namespace llvm;
@@ -56,4 +56,20 @@ unsigned MipsFunctionInfo::getMips16SPAliasReg() {
   return Mips16SPAliasReg = MF.getRegInfo().createVirtualRegister(RC);
 }
 
+void MipsFunctionInfo::createEhDataRegsFI() {
+  for (int I = 0; I < 4; ++I) {
+    const MipsSubtarget &ST = MF.getTarget().getSubtarget<MipsSubtarget>();
+    const TargetRegisterClass *RC = ST.isABI_N64() ?
+        &Mips::CPU64RegsRegClass : &Mips::CPURegsRegClass;
+
+    EhDataRegFI[I] = MF.getFrameInfo()->CreateStackObject(RC->getSize(),
+        RC->getAlignment(), false);
+  }
+}
+
+bool MipsFunctionInfo::isEhDataRegFI(int FI) const {
+  return CallsEhReturn && (FI == EhDataRegFI[0] || FI == EhDataRegFI[1]
+                        || FI == EhDataRegFI[2] || FI == EhDataRegFI[3]);
+}
+
 void MipsFunctionInfo::anchor() { }
diff --git a/lib/Target/Mips/MipsMachineFunction.h b/lib/Target/Mips/MipsMachineFunction.h
index bb45f92f18fd..b05b348037d9 100644
--- a/lib/Target/Mips/MipsMachineFunction.h
+++ b/lib/Target/Mips/MipsMachineFunction.h
@@ -15,8 +15,8 @@
 #define MIPS_MACHINE_FUNCTION_INFO_H
 
 #include "MipsSubtarget.h"
-#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/Target/TargetFrameLowering.h"
 #include "llvm/Target/TargetMachine.h"
 #include <utility>
@@ -53,10 +53,16 @@ class MipsFunctionInfo : public MachineFunctionInfo {
   /// Size of incoming argument area.
   unsigned IncomingArgSize;
 
+  /// CallsEhReturn - Whether the function calls llvm.eh.return.
+  bool CallsEhReturn;
+
+  /// Frame objects for spilling eh data registers.
+  int EhDataRegFI[4];
+
 public:
   MipsFunctionInfo(MachineFunction& MF)
    : MF(MF), SRetReturnReg(0), GlobalBaseReg(0), Mips16SPAliasReg(0),
-     VarArgsFrameIndex(0)
+     VarArgsFrameIndex(0), CallsEhReturn(false)
   {}
 
   unsigned getSRetReturnReg() const { return SRetReturnReg; }
@@ -78,6 +84,14 @@ public:
   }
 
   unsigned getIncomingArgSize() const { return IncomingArgSize; }
+
+  bool callsEhReturn() const { return CallsEhReturn; }
+  void setCallsEhReturn() { CallsEhReturn = true; }
+
+  void createEhDataRegsFI();
+  int getEhDataRegFI(unsigned Reg) const { return EhDataRegFI[Reg]; }
+  bool isEhDataRegFI(int FI) const;
+
 };
 
 } // end of namespace llvm
diff --git a/lib/Target/Mips/MipsRegisterInfo.cpp b/lib/Target/Mips/MipsRegisterInfo.cpp
index d8e0dd436a95..32507334e9c6 100644
--- a/lib/Target/Mips/MipsRegisterInfo.cpp
+++ b/lib/Target/Mips/MipsRegisterInfo.cpp
@@ -17,25 +17,25 @@
 #include "Mips.h"
 #include "MipsAnalyzeImmediate.h"
 #include "MipsInstrInfo.h"
-#include "MipsSubtarget.h"
 #include "MipsMachineFunction.h"
-#include "llvm/Constants.h"
-#include "llvm/DebugInfo.h"
-#include "llvm/Type.h"
-#include "llvm/CodeGen/ValueTypes.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunction.h"
+#include "MipsSubtarget.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/DebugInfo.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/STLExtras.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
 
 #define GET_REGINFO_TARGET_DESC
 #include "MipsGenRegisterInfo.inc"
@@ -47,6 +47,28 @@ MipsRegisterInfo::MipsRegisterInfo(const MipsSubtarget &ST)
 
 unsigned MipsRegisterInfo::getPICCallReg() { return Mips::T9; }
 
+
+unsigned
+MipsRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC,
+                                      MachineFunction &MF) const {
+  switch (RC->getID()) {
+  default:
+    return 0;
+  case Mips::CPURegsRegClassID:
+  case Mips::CPU64RegsRegClassID:
+  case Mips::DSPRegsRegClassID: {
+    const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
+    return 28 - TFI->hasFP(MF);
+  }
+  case Mips::FGR32RegClassID:
+    return 32;
+  case Mips::AFGR64RegClassID:
+    return 16;
+  case Mips::FGR64RegClassID:
+    return 32;
+  }
+}
+
 //===----------------------------------------------------------------------===//
 // Callee Saved Registers methods
 //===----------------------------------------------------------------------===//
@@ -155,21 +177,14 @@ MipsRegisterInfo::trackLivenessAfterRegAlloc(const MachineFunction &MF) const {
 // direct reference.
 void MipsRegisterInfo::
 eliminateFrameIndex(MachineBasicBlock::iterator II, int SPAdj,
-                    RegScavenger *RS) const {
+                    unsigned FIOperandNum, RegScavenger *RS) const {
   MachineInstr &MI = *II;
   MachineFunction &MF = *MI.getParent()->getParent();
 
-  unsigned i = 0;
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() &&
-           "Instr doesn't have FrameIndex operand!");
-  }
-
   DEBUG(errs() << "\nFunction : " << MF.getName() << "\n";
         errs() << "<--------->\n" << MI);
 
-  int FrameIndex = MI.getOperand(i).getIndex();
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
   uint64_t stackSize = MF.getFrameInfo()->getStackSize();
   int64_t spOffset = MF.getFrameInfo()->getObjectOffset(FrameIndex);
 
@@ -177,7 +192,7 @@ eliminateFrameIndex(MachineBasicBlock::iterator II, int SPAdj,
                << "spOffset   : " << spOffset << "\n"
                << "stackSize  : " << stackSize << "\n");
 
-  eliminateFI(MI, i, FrameIndex, stackSize, spOffset);
+  eliminateFI(MI, FIOperandNum, FrameIndex, stackSize, spOffset);
 }
 
 unsigned MipsRegisterInfo::
diff --git a/lib/Target/Mips/MipsRegisterInfo.h b/lib/Target/Mips/MipsRegisterInfo.h
index 78adf7f18bf2..5ed51241391f 100644
--- a/lib/Target/Mips/MipsRegisterInfo.h
+++ b/lib/Target/Mips/MipsRegisterInfo.h
@@ -42,6 +42,8 @@ public:
   void adjustMipsStackFrame(MachineFunction &MF) const;
 
   /// Code Generation virtual methods...
+  unsigned getRegPressureLimit(const TargetRegisterClass *RC,
+                               MachineFunction &MF) const;
   const uint16_t *getCalleeSavedRegs(const MachineFunction *MF = 0) const;
   const uint32_t *getCallPreservedMask(CallingConv::ID) const;
 
@@ -53,9 +55,11 @@ public:
 
   /// Stack Frame Processing Methods
   void eliminateFrameIndex(MachineBasicBlock::iterator II,
-                           int SPAdj, RegScavenger *RS = NULL) const;
+                           int SPAdj, unsigned FIOperandNum,
+                           RegScavenger *RS = NULL) const;
 
-  void processFunctionBeforeFrameFinalized(MachineFunction &MF) const;
+  void processFunctionBeforeFrameFinalized(MachineFunction &MF,
+                                       RegScavenger *RS = NULL) const;
 
   /// Debug information queries.
   unsigned getFrameRegister(const MachineFunction &MF) const;
@@ -64,6 +68,9 @@ public:
   unsigned getEHExceptionRegister() const;
   unsigned getEHHandlerRegister() const;
 
+  /// \brief Return GPR register class.
+  virtual const TargetRegisterClass *intRegClass(unsigned Size) const = 0;
+
 private:
   virtual void eliminateFI(MachineBasicBlock::iterator II, unsigned OpNo,
                            int FrameIndex, uint64_t StackSize,
diff --git a/lib/Target/Mips/MipsRegisterInfo.td b/lib/Target/Mips/MipsRegisterInfo.td
index 391c19e07e33..64458bcef7ef 100644
--- a/lib/Target/Mips/MipsRegisterInfo.td
+++ b/lib/Target/Mips/MipsRegisterInfo.td
@@ -18,54 +18,56 @@ def sub_lo     : SubRegIndex;
 def sub_hi     : SubRegIndex;
 }
 
+class Unallocatable {
+  bit isAllocatable = 0;
+}
+
 // We have banks of 32 registers each.
-class MipsReg<string n> : Register<n> {
-  field bits<5> Num;
+class MipsReg<bits<16> Enc, string n> : Register<n> {
+  let HWEncoding = Enc;
   let Namespace = "Mips";
 }
 
-class MipsRegWithSubRegs<string n, list<Register> subregs>
+class MipsRegWithSubRegs<bits<16> Enc, string n, list<Register> subregs>
   : RegisterWithSubRegs<n, subregs> {
-  field bits<5> Num;
+  let HWEncoding = Enc;
   let Namespace = "Mips";
 }
 
 // Mips CPU Registers
-class MipsGPRReg<bits<5> num, string n> : MipsReg<n> {
-  let Num = num;
-}
+class MipsGPRReg<bits<16> Enc, string n> : MipsReg<Enc, n>;
 
 // Mips 64-bit CPU Registers
-class Mips64GPRReg<bits<5> num, string n, list<Register> subregs>
-  : MipsRegWithSubRegs<n, subregs> {
-  let Num = num;
+class Mips64GPRReg<bits<16> Enc, string n, list<Register> subregs>
+  : MipsRegWithSubRegs<Enc, n, subregs> {
   let SubRegIndices = [sub_32];
 }
 
 // Mips 32-bit FPU Registers
-class FPR<bits<5> num, string n> : MipsReg<n> {
-  let Num = num;
-}
+class FPR<bits<16> Enc, string n> : MipsReg<Enc, n>;
 
 // Mips 64-bit (aliased) FPU Registers
-class AFPR<bits<5> num, string n, list<Register> subregs>
-  : MipsRegWithSubRegs<n, subregs> {
-  let Num = num;
+class AFPR<bits<16> Enc, string n, list<Register> subregs>
+  : MipsRegWithSubRegs<Enc, n, subregs> {
   let SubRegIndices = [sub_fpeven, sub_fpodd];
   let CoveredBySubRegs = 1;
 }
 
-class AFPR64<bits<5> num, string n, list<Register> subregs>
-  : MipsRegWithSubRegs<n, subregs> {
-  let Num = num;
+class AFPR64<bits<16> Enc, string n, list<Register> subregs>
+  : MipsRegWithSubRegs<Enc, n, subregs> {
   let SubRegIndices = [sub_32];
 }
 
-// Mips Hardware Registers
-class HWR<bits<5> num, string n> : MipsReg<n> {
-  let Num = num;
+// Accumulator Registers
+class ACC<bits<16> Enc, string n, list<Register> subregs>
+  : MipsRegWithSubRegs<Enc, n, subregs> {
+  let SubRegIndices = [sub_lo, sub_hi];
+  let CoveredBySubRegs = 1;
 }
 
+// Mips Hardware Registers
+class HWR<bits<16> Enc, string n> : MipsReg<Enc, n>;
+
 //===----------------------------------------------------------------------===//
 //  Registers
 //===----------------------------------------------------------------------===//
@@ -228,7 +230,13 @@ let Namespace = "Mips" in {
 
   // Hi/Lo registers
   def HI  : Register<"hi">, DwarfRegNum<[64]>;
+  def HI1 : Register<"hi1">, DwarfRegNum<[176]>;
+  def HI2 : Register<"hi2">, DwarfRegNum<[178]>;
+  def HI3 : Register<"hi3">, DwarfRegNum<[180]>;
   def LO  : Register<"lo">, DwarfRegNum<[65]>;
+  def LO1 : Register<"lo1">, DwarfRegNum<[177]>;
+  def LO2 : Register<"lo2">, DwarfRegNum<[179]>;
+  def LO3 : Register<"lo3">, DwarfRegNum<[181]>;
 
   let SubRegIndices = [sub_32] in {
   def HI64  : RegisterWithSubRegs<"hi", [HI]>;
@@ -239,21 +247,22 @@ let Namespace = "Mips" in {
   def FCR31 : Register<"31">;
 
   // fcc0 register
-  def FCC0 : Register<"fcc0">;
+  def FCC0 : MipsReg<0, "fcc0">;
 
   // PC register
   def PC : Register<"pc">;
 
   // Hardware register $29
-  def HWR29 : Register<"29">;
-  def HWR29_64 : Register<"29">;
+  def HWR29 : MipsReg<29, "29">;
+  def HWR29_64 : MipsReg<29, "29">;
 
   // Accum registers
-  let SubRegIndices = [sub_lo, sub_hi] in
-  def AC0 : RegisterWithSubRegs<"ac0", [LO, HI]>;
-  def AC1 : Register<"ac1">;
-  def AC2 : Register<"ac2">;
-  def AC3 : Register<"ac3">;
+  def AC0 : ACC<0, "ac0", [LO, HI]>;
+  def AC1 : ACC<1, "ac1", [LO1, HI1]>;
+  def AC2 : ACC<2, "ac2", [LO2, HI2]>;
+  def AC3 : ACC<3, "ac3", [LO3, HI3]>;
+
+  def AC0_64 : ACC<0, "ac0", [LO64, HI64]>;
 
   def DSPCtrl : Register<"dspctrl">;
 }
@@ -300,9 +309,9 @@ def CPU16Regs : RegisterClass<"Mips", [i32], 32, (add
   // Callee save
   S0, S1)>;
 
-def CPURAReg : RegisterClass<"Mips", [i32], 32, (add RA)>;
+def CPURAReg : RegisterClass<"Mips", [i32], 32, (add RA)>, Unallocatable;
 
-def CPUSPReg : RegisterClass<"Mips", [i32], 32, (add SP)>;
+def CPUSPReg : RegisterClass<"Mips", [i32], 32, (add SP)>, Unallocatable;
 
 // 64bit fp:
 // * FGR64  - 32 64-bit registers
@@ -328,15 +337,70 @@ def AFGR64 : RegisterClass<"Mips", [f64], 64, (add
 def FGR64 : RegisterClass<"Mips", [f64], 64, (sequence "D%u_64", 0, 31)>;
 
 // Condition Register for floating point operations
-def CCR  : RegisterClass<"Mips", [i32], 32, (add FCR31,FCC0)>;
+def CCR  : RegisterClass<"Mips", [i32], 32, (add FCR31,FCC0)>, Unallocatable;
 
 // Hi/Lo Registers
-def HILO : RegisterClass<"Mips", [i32], 32, (add HI, LO)>;
-def HILO64 : RegisterClass<"Mips", [i64], 64, (add HI64, LO64)>;
+def HILO : RegisterClass<"Mips", [i32], 32, (add HI, LO)>, Unallocatable;
+def HILO64 : RegisterClass<"Mips", [i64], 64, (add HI64, LO64)>, Unallocatable;
 
 // Hardware registers
-def HWRegs : RegisterClass<"Mips", [i32], 32, (add HWR29)>;
-def HWRegs64 : RegisterClass<"Mips", [i64], 32, (add HWR29_64)>;
+def HWRegs : RegisterClass<"Mips", [i32], 32, (add HWR29)>, Unallocatable;
+def HWRegs64 : RegisterClass<"Mips", [i64], 64, (add HWR29_64)>, Unallocatable;
 
 // Accumulator Registers
-def ACRegs : RegisterClass<"Mips", [i64], 64, (sequence "AC%u", 0, 3)>;
+def ACRegs : RegisterClass<"Mips", [untyped], 64, (add AC0)> {
+  let Size = 64;
+}
+
+def ACRegs128 : RegisterClass<"Mips", [untyped], 128, (add AC0_64)> {
+  let Size = 128;
+}
+
+def ACRegsDSP : RegisterClass<"Mips", [untyped], 64, (sequence "AC%u", 0, 3)> {
+  let Size = 64;
+}
+
+def CPURegsAsmOperand : AsmOperandClass {
+  let Name = "CPURegsAsm";
+  let ParserMethod = "parseCPURegs";
+}
+
+def CPU64RegsAsmOperand : AsmOperandClass {
+  let Name = "CPU64RegsAsm";
+  let ParserMethod = "parseCPU64Regs";
+}
+
+def CCRAsmOperand : AsmOperandClass {
+  let Name = "CCRAsm";
+  let ParserMethod = "parseCCRRegs";
+}
+
+def CPURegsOpnd : RegisterOperand<CPURegs, "printCPURegs"> {
+  let ParserMatchClass = CPURegsAsmOperand;
+}
+
+def CPU64RegsOpnd : RegisterOperand<CPU64Regs, "printCPURegs"> {
+  let ParserMatchClass = CPU64RegsAsmOperand;
+}
+
+def CCROpnd : RegisterOperand<CCR, "printCPURegs"> {
+  let ParserMatchClass = CCRAsmOperand;
+}
+
+def HWRegsAsmOperand : AsmOperandClass {
+  let Name = "HWRegsAsm";
+  let ParserMethod = "parseHWRegs";
+}
+
+def HW64RegsAsmOperand : AsmOperandClass {
+  let Name = "HW64RegsAsm";
+  let ParserMethod = "parseHW64Regs";
+}
+
+def HWRegsOpnd : RegisterOperand<HWRegs, "printCPURegs"> {
+  let ParserMatchClass = HWRegsAsmOperand;
+}
+
+def HW64RegsOpnd : RegisterOperand<HWRegs64, "printCPURegs"> {
+  let ParserMatchClass = HW64RegsAsmOperand;
+}
diff --git a/lib/Target/Mips/MipsSEFrameLowering.cpp b/lib/Target/Mips/MipsSEFrameLowering.cpp
index 03f5176b2974..68ec92188802 100644
--- a/lib/Target/Mips/MipsSEFrameLowering.cpp
+++ b/lib/Target/Mips/MipsSEFrameLowering.cpp
@@ -12,26 +12,187 @@
 //===----------------------------------------------------------------------===//
 
 #include "MipsSEFrameLowering.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
 #include "MipsAnalyzeImmediate.h"
-#include "MipsSEInstrInfo.h"
 #include "MipsMachineFunction.h"
-#include "MCTargetDesc/MipsBaseInfo.h"
-#include "llvm/Function.h"
+#include "MipsSEInstrInfo.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
+namespace {
+typedef MachineBasicBlock::iterator Iter;
+
+/// Helper class to expand accumulator pseudos.
+class ExpandACCPseudo {
+public:
+  ExpandACCPseudo(MachineFunction &MF);
+  bool expand();
+
+private:
+  bool expandInstr(MachineBasicBlock &MBB, Iter I);
+  void expandLoad(MachineBasicBlock &MBB, Iter I, unsigned RegSize);
+  void expandStore(MachineBasicBlock &MBB, Iter I, unsigned RegSize);
+  void expandCopy(MachineBasicBlock &MBB, Iter I, unsigned RegSize);
+
+  MachineFunction &MF;
+  const MipsSEInstrInfo &TII;
+  const MipsRegisterInfo &RegInfo;
+  MachineRegisterInfo &MRI;
+};
+}
+
+ExpandACCPseudo::ExpandACCPseudo(MachineFunction &MF_)
+  : MF(MF_),
+    TII(*static_cast<const MipsSEInstrInfo*>(MF.getTarget().getInstrInfo())),
+    RegInfo(TII.getRegisterInfo()), MRI(MF.getRegInfo()) {}
+
+bool ExpandACCPseudo::expand() {
+  bool Expanded = false;
+
+  for (MachineFunction::iterator BB = MF.begin(), BBEnd = MF.end();
+       BB != BBEnd; ++BB)
+    for (Iter I = BB->begin(), End = BB->end(); I != End;)
+      Expanded |= expandInstr(*BB, I++);
+
+  return Expanded;
+}
+
+bool ExpandACCPseudo::expandInstr(MachineBasicBlock &MBB, Iter I) {
+  switch(I->getOpcode()) {
+  case Mips::LOAD_AC64:
+  case Mips::LOAD_AC64_P8:
+  case Mips::LOAD_AC_DSP:
+  case Mips::LOAD_AC_DSP_P8:
+    expandLoad(MBB, I, 4);
+    break;
+  case Mips::LOAD_AC128:
+  case Mips::LOAD_AC128_P8:
+    expandLoad(MBB, I, 8);
+    break;
+  case Mips::STORE_AC64:
+  case Mips::STORE_AC64_P8:
+  case Mips::STORE_AC_DSP:
+  case Mips::STORE_AC_DSP_P8:
+    expandStore(MBB, I, 4);
+    break;
+  case Mips::STORE_AC128:
+  case Mips::STORE_AC128_P8:
+    expandStore(MBB, I, 8);
+    break;
+  case Mips::COPY_AC64:
+  case Mips::COPY_AC_DSP:
+    expandCopy(MBB, I, 4);
+    break;
+  case Mips::COPY_AC128:
+    expandCopy(MBB, I, 8);
+    break;
+  default:
+    return false;
+  }
+
+  MBB.erase(I);
+  return true;
+}
+
+void ExpandACCPseudo::expandLoad(MachineBasicBlock &MBB, Iter I,
+                                 unsigned RegSize) {
+  //  load $vr0, FI
+  //  copy lo, $vr0
+  //  load $vr1, FI + 4
+  //  copy hi, $vr1
+
+  assert(I->getOperand(0).isReg() && I->getOperand(1).isFI());
+
+  const TargetRegisterClass *RC = RegInfo.intRegClass(RegSize);
+  unsigned VR0 = MRI.createVirtualRegister(RC);
+  unsigned VR1 = MRI.createVirtualRegister(RC);
+  unsigned Dst = I->getOperand(0).getReg(), FI = I->getOperand(1).getIndex();
+  unsigned Lo = RegInfo.getSubReg(Dst, Mips::sub_lo);
+  unsigned Hi = RegInfo.getSubReg(Dst, Mips::sub_hi);
+  DebugLoc DL = I->getDebugLoc();
+  const MCInstrDesc &Desc = TII.get(TargetOpcode::COPY);
+
+  TII.loadRegFromStack(MBB, I, VR0, FI, RC, &RegInfo, 0);
+  BuildMI(MBB, I, DL, Desc, Lo).addReg(VR0, RegState::Kill);
+  TII.loadRegFromStack(MBB, I, VR1, FI, RC, &RegInfo, RegSize);
+  BuildMI(MBB, I, DL, Desc, Hi).addReg(VR1, RegState::Kill);
+}
+
+void ExpandACCPseudo::expandStore(MachineBasicBlock &MBB, Iter I,
+                                  unsigned RegSize) {
+  //  copy $vr0, lo
+  //  store $vr0, FI
+  //  copy $vr1, hi
+  //  store $vr1, FI + 4
+
+  assert(I->getOperand(0).isReg() && I->getOperand(1).isFI());
+
+  const TargetRegisterClass *RC = RegInfo.intRegClass(RegSize);
+  unsigned VR0 = MRI.createVirtualRegister(RC);
+  unsigned VR1 = MRI.createVirtualRegister(RC);
+  unsigned Src = I->getOperand(0).getReg(), FI = I->getOperand(1).getIndex();
+  unsigned SrcKill = getKillRegState(I->getOperand(0).isKill());
+  unsigned Lo = RegInfo.getSubReg(Src, Mips::sub_lo);
+  unsigned Hi = RegInfo.getSubReg(Src, Mips::sub_hi);
+  DebugLoc DL = I->getDebugLoc();
+
+  BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), VR0).addReg(Lo, SrcKill);
+  TII.storeRegToStack(MBB, I, VR0, true, FI, RC, &RegInfo, 0);
+  BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), VR1).addReg(Hi, SrcKill);
+  TII.storeRegToStack(MBB, I, VR1, true, FI, RC, &RegInfo, RegSize);
+}
+
+void ExpandACCPseudo::expandCopy(MachineBasicBlock &MBB, Iter I,
+                                 unsigned RegSize) {
+  //  copy $vr0, src_lo
+  //  copy dst_lo, $vr0
+  //  copy $vr1, src_hi
+  //  copy dst_hi, $vr1
+
+  const TargetRegisterClass *RC = RegInfo.intRegClass(RegSize);
+  unsigned VR0 = MRI.createVirtualRegister(RC);
+  unsigned VR1 = MRI.createVirtualRegister(RC);
+  unsigned Dst = I->getOperand(0).getReg(), Src = I->getOperand(1).getReg();
+  unsigned SrcKill = getKillRegState(I->getOperand(1).isKill());
+  unsigned DstLo = RegInfo.getSubReg(Dst, Mips::sub_lo);
+  unsigned DstHi = RegInfo.getSubReg(Dst, Mips::sub_hi);
+  unsigned SrcLo = RegInfo.getSubReg(Src, Mips::sub_lo);
+  unsigned SrcHi = RegInfo.getSubReg(Src, Mips::sub_hi);
+  DebugLoc DL = I->getDebugLoc();
+
+  BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), VR0).addReg(SrcLo, SrcKill);
+  BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), DstLo)
+    .addReg(VR0, RegState::Kill);
+  BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), VR1).addReg(SrcHi, SrcKill);
+  BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), DstHi)
+    .addReg(VR1, RegState::Kill);
+}
+
+unsigned MipsSEFrameLowering::ehDataReg(unsigned I) const {
+  static const unsigned EhDataReg[] = {
+    Mips::A0, Mips::A1, Mips::A2, Mips::A3
+  };
+  static const unsigned EhDataReg64[] = {
+    Mips::A0_64, Mips::A1_64, Mips::A2_64, Mips::A3_64
+  };
+
+  return STI.isABI_N64() ? EhDataReg64[I] : EhDataReg[I];
+}
+
 void MipsSEFrameLowering::emitPrologue(MachineFunction &MF) const {
   MachineBasicBlock &MBB   = MF.front();
   MachineFrameInfo *MFI    = MF.getFrameInfo();
+  MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
   const MipsRegisterInfo *RegInfo =
     static_cast<const MipsRegisterInfo*>(MF.getTarget().getRegisterInfo());
   const MipsSEInstrInfo &TII =
@@ -105,6 +266,30 @@ void MipsSEFrameLowering::emitPrologue(MachineFunction &MF) const {
     }
   }
 
+  if (MipsFI->callsEhReturn()) {
+    const TargetRegisterClass *RC = STI.isABI_N64() ?
+        &Mips::CPU64RegsRegClass : &Mips::CPURegsRegClass;
+
+    // Insert instructions that spill eh data registers.
+    for (int I = 0; I < 4; ++I) {
+      if (!MBB.isLiveIn(ehDataReg(I)))
+        MBB.addLiveIn(ehDataReg(I));
+      TII.storeRegToStackSlot(MBB, MBBI, ehDataReg(I), false,
+                              MipsFI->getEhDataRegFI(I), RC, RegInfo);
+    }
+
+    // Emit .cfi_offset directives for eh data registers.
+    MCSymbol *CSLabel2 = MMI.getContext().CreateTempSymbol();
+    BuildMI(MBB, MBBI, dl,
+            TII.get(TargetOpcode::PROLOG_LABEL)).addSym(CSLabel2);
+    for (int I = 0; I < 4; ++I) {
+      int64_t Offset = MFI->getObjectOffset(MipsFI->getEhDataRegFI(I));
+      DstML = MachineLocation(MachineLocation::VirtualFP, Offset);
+      SrcML = MachineLocation(ehDataReg(I));
+      Moves.push_back(MachineMove(CSLabel2, DstML, SrcML));
+    }
+  }
+
   // if framepointer enabled, set it to point to the stack pointer.
   if (hasFP(MF)) {
     // Insert instruction "move $fp, $sp" at this location.
@@ -124,6 +309,9 @@ void MipsSEFrameLowering::emitEpilogue(MachineFunction &MF,
                                        MachineBasicBlock &MBB) const {
   MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
   MachineFrameInfo *MFI            = MF.getFrameInfo();
+  MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
+  const MipsRegisterInfo *RegInfo =
+    static_cast<const MipsRegisterInfo*>(MF.getTarget().getRegisterInfo());
   const MipsSEInstrInfo &TII =
     *static_cast<const MipsSEInstrInfo*>(MF.getTarget().getInstrInfo());
   DebugLoc dl = MBBI->getDebugLoc();
@@ -144,6 +332,22 @@ void MipsSEFrameLowering::emitEpilogue(MachineFunction &MF,
     BuildMI(MBB, I, dl, TII.get(ADDu), SP).addReg(FP).addReg(ZERO);
   }
 
+  if (MipsFI->callsEhReturn()) {
+    const TargetRegisterClass *RC = STI.isABI_N64() ?
+        &Mips::CPU64RegsRegClass : &Mips::CPURegsRegClass;
+
+    // Find first instruction that restores a callee-saved register.
+    MachineBasicBlock::iterator I = MBBI;
+    for (unsigned i = 0; i < MFI->getCalleeSavedInfo().size(); ++i)
+      --I;
+
+    // Insert instructions that restore eh data registers.
+    for (int J = 0; J < 4; ++J) {
+      TII.loadRegFromStackSlot(MBB, I, ehDataReg(J), MipsFI->getEhDataRegFI(J),
+                               RC, RegInfo);
+    }
+  }
+
   // Get the number of bytes from FrameInfo
   uint64_t StackSize = MFI->getStackSize();
 
@@ -191,19 +395,59 @@ MipsSEFrameLowering::hasReservedCallFrame(const MachineFunction &MF) const {
 
   // Reserve call frame if the size of the maximum call frame fits into 16-bit
   // immediate field and there are no variable sized objects on the stack.
-  return isInt<16>(MFI->getMaxCallFrameSize()) && !MFI->hasVarSizedObjects();
+  // Make sure the second register scavenger spill slot can be accessed with one
+  // instruction.
+  return isInt<16>(MFI->getMaxCallFrameSize() + getStackAlignment()) &&
+    !MFI->hasVarSizedObjects();
+}
+
+// Eliminate ADJCALLSTACKDOWN, ADJCALLSTACKUP pseudo instructions
+void MipsSEFrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  const MipsSEInstrInfo &TII =
+    *static_cast<const MipsSEInstrInfo*>(MF.getTarget().getInstrInfo());
+
+  if (!hasReservedCallFrame(MF)) {
+    int64_t Amount = I->getOperand(0).getImm();
+
+    if (I->getOpcode() == Mips::ADJCALLSTACKDOWN)
+      Amount = -Amount;
+
+    unsigned SP = STI.isABI_N64() ? Mips::SP_64 : Mips::SP;
+    TII.adjustStackPtr(SP, Amount, MBB, I);
+  }
+
+  MBB.erase(I);
 }
 
 void MipsSEFrameLowering::
 processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
                                      RegScavenger *RS) const {
   MachineRegisterInfo &MRI = MF.getRegInfo();
+  MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
   unsigned FP = STI.isABI_N64() ? Mips::FP_64 : Mips::FP;
 
   // Mark $fp as used if function has dedicated frame pointer.
   if (hasFP(MF))
     MRI.setPhysRegUsed(FP);
 
+  // Create spill slots for eh data registers if function calls eh_return.
+  if (MipsFI->callsEhReturn())
+    MipsFI->createEhDataRegsFI();
+
+  // Expand pseudo instructions which load, store or copy accumulators.
+  // Add an emergency spill slot if a pseudo was expanded.
+  if (ExpandACCPseudo(MF).expand()) {
+    // The spill slot should be half the size of the accumulator. If target is
+    // mips64, it should be 64-bit, otherwise it should be 32-bt.
+    const TargetRegisterClass *RC = STI.hasMips64() ?
+      &Mips::CPU64RegsRegClass : &Mips::CPURegsRegClass;
+    int FI = MF.getFrameInfo()->CreateStackObject(RC->getSize(),
+                                                  RC->getAlignment(), false);
+    RS->addScavengingFrameIndex(FI);
+  }
+
   // Set scavenging frame index if necessary.
   uint64_t MaxSPOffset = MF.getInfo<MipsFunctionInfo>()->getIncomingArgSize() +
     estimateStackSize(MF);
@@ -215,7 +459,7 @@ processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
     &Mips::CPU64RegsRegClass : &Mips::CPURegsRegClass;
   int FI = MF.getFrameInfo()->CreateStackObject(RC->getSize(),
                                                 RC->getAlignment(), false);
-  RS->setScavengingFrameIndex(FI);
+  RS->addScavengingFrameIndex(FI);
 }
 
 const MipsFrameLowering *
diff --git a/lib/Target/Mips/MipsSEFrameLowering.h b/lib/Target/Mips/MipsSEFrameLowering.h
index 6481a0ac86d7..193a66cc65a7 100644
--- a/lib/Target/Mips/MipsSEFrameLowering.h
+++ b/lib/Target/Mips/MipsSEFrameLowering.h
@@ -21,13 +21,17 @@ namespace llvm {
 class MipsSEFrameLowering : public MipsFrameLowering {
 public:
   explicit MipsSEFrameLowering(const MipsSubtarget &STI)
-    : MipsFrameLowering(STI) {}
+    : MipsFrameLowering(STI, STI.hasMips64() ? 16 : 8) {}
 
   /// emitProlog/emitEpilog - These methods insert prolog and epilog code into
   /// the function.
   void emitPrologue(MachineFunction &MF) const;
   void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
 
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I) const;
+
   bool spillCalleeSavedRegisters(MachineBasicBlock &MBB,
                                  MachineBasicBlock::iterator MI,
                                  const std::vector<CalleeSavedInfo> &CSI,
@@ -37,6 +41,7 @@ public:
 
   void processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
                                             RegScavenger *RS) const;
+  unsigned ehDataReg(unsigned I) const;
 };
 
 } // End llvm namespace
diff --git a/lib/Target/Mips/MipsSEISelDAGToDAG.cpp b/lib/Target/Mips/MipsSEISelDAGToDAG.cpp
new file mode 100644
index 000000000000..d6d220750c61
--- /dev/null
+++ b/lib/Target/Mips/MipsSEISelDAGToDAG.cpp
@@ -0,0 +1,473 @@
+//===-- MipsSEISelDAGToDAG.cpp - A Dag to Dag Inst Selector for MipsSE ----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Subclass of MipsDAGToDAGISel specialized for mips32/64.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "mips-isel"
+#include "MipsSEISelDAGToDAG.h"
+#include "Mips.h"
+#include "MCTargetDesc/MipsBaseInfo.h"
+#include "MipsAnalyzeImmediate.h"
+#include "MipsMachineFunction.h"
+#include "MipsRegisterInfo.h"
+#include "llvm/CodeGen/MachineConstantPool.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/SelectionDAGNodes.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Type.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetMachine.h"
+using namespace llvm;
+
+
+bool MipsSEDAGToDAGISel::replaceUsesWithZeroReg(MachineRegisterInfo *MRI,
+                                                const MachineInstr& MI) {
+  unsigned DstReg = 0, ZeroReg = 0;
+
+  // Check if MI is "addiu $dst, $zero, 0" or "daddiu $dst, $zero, 0".
+  if ((MI.getOpcode() == Mips::ADDiu) &&
+      (MI.getOperand(1).getReg() == Mips::ZERO) &&
+      (MI.getOperand(2).getImm() == 0)) {
+    DstReg = MI.getOperand(0).getReg();
+    ZeroReg = Mips::ZERO;
+  } else if ((MI.getOpcode() == Mips::DADDiu) &&
+             (MI.getOperand(1).getReg() == Mips::ZERO_64) &&
+             (MI.getOperand(2).getImm() == 0)) {
+    DstReg = MI.getOperand(0).getReg();
+    ZeroReg = Mips::ZERO_64;
+  }
+
+  if (!DstReg)
+    return false;
+
+  // Replace uses with ZeroReg.
+  for (MachineRegisterInfo::use_iterator U = MRI->use_begin(DstReg),
+       E = MRI->use_end(); U != E;) {
+    MachineOperand &MO = U.getOperand();
+    unsigned OpNo = U.getOperandNo();
+    MachineInstr *MI = MO.getParent();
+    ++U;
+
+    // Do not replace if it is a phi's operand or is tied to def operand.
+    if (MI->isPHI() || MI->isRegTiedToDefOperand(OpNo) || MI->isPseudo())
+      continue;
+
+    MO.setReg(ZeroReg);
+  }
+
+  return true;
+}
+
+void MipsSEDAGToDAGISel::initGlobalBaseReg(MachineFunction &MF) {
+  MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
+
+  if (!MipsFI->globalBaseRegSet())
+    return;
+
+  MachineBasicBlock &MBB = MF.front();
+  MachineBasicBlock::iterator I = MBB.begin();
+  MachineRegisterInfo &RegInfo = MF.getRegInfo();
+  const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
+  DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc();
+  unsigned V0, V1, GlobalBaseReg = MipsFI->getGlobalBaseReg();
+  const TargetRegisterClass *RC;
+
+  if (Subtarget.isABI_N64())
+    RC = (const TargetRegisterClass*)&Mips::CPU64RegsRegClass;
+  else
+    RC = (const TargetRegisterClass*)&Mips::CPURegsRegClass;
+
+  V0 = RegInfo.createVirtualRegister(RC);
+  V1 = RegInfo.createVirtualRegister(RC);
+
+  if (Subtarget.isABI_N64()) {
+    MF.getRegInfo().addLiveIn(Mips::T9_64);
+    MBB.addLiveIn(Mips::T9_64);
+
+    // lui $v0, %hi(%neg(%gp_rel(fname)))
+    // daddu $v1, $v0, $t9
+    // daddiu $globalbasereg, $v1, %lo(%neg(%gp_rel(fname)))
+    const GlobalValue *FName = MF.getFunction();
+    BuildMI(MBB, I, DL, TII.get(Mips::LUi64), V0)
+      .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_HI);
+    BuildMI(MBB, I, DL, TII.get(Mips::DADDu), V1).addReg(V0)
+      .addReg(Mips::T9_64);
+    BuildMI(MBB, I, DL, TII.get(Mips::DADDiu), GlobalBaseReg).addReg(V1)
+      .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_LO);
+    return;
+  }
+
+  if (MF.getTarget().getRelocationModel() == Reloc::Static) {
+    // Set global register to __gnu_local_gp.
+    //
+    // lui   $v0, %hi(__gnu_local_gp)
+    // addiu $globalbasereg, $v0, %lo(__gnu_local_gp)
+    BuildMI(MBB, I, DL, TII.get(Mips::LUi), V0)
+      .addExternalSymbol("__gnu_local_gp", MipsII::MO_ABS_HI);
+    BuildMI(MBB, I, DL, TII.get(Mips::ADDiu), GlobalBaseReg).addReg(V0)
+      .addExternalSymbol("__gnu_local_gp", MipsII::MO_ABS_LO);
+    return;
+  }
+
+  MF.getRegInfo().addLiveIn(Mips::T9);
+  MBB.addLiveIn(Mips::T9);
+
+  if (Subtarget.isABI_N32()) {
+    // lui $v0, %hi(%neg(%gp_rel(fname)))
+    // addu $v1, $v0, $t9
+    // addiu $globalbasereg, $v1, %lo(%neg(%gp_rel(fname)))
+    const GlobalValue *FName = MF.getFunction();
+    BuildMI(MBB, I, DL, TII.get(Mips::LUi), V0)
+      .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_HI);
+    BuildMI(MBB, I, DL, TII.get(Mips::ADDu), V1).addReg(V0).addReg(Mips::T9);
+    BuildMI(MBB, I, DL, TII.get(Mips::ADDiu), GlobalBaseReg).addReg(V1)
+      .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_LO);
+    return;
+  }
+
+  assert(Subtarget.isABI_O32());
+
+  // For O32 ABI, the following instruction sequence is emitted to initialize
+  // the global base register:
+  //
+  //  0. lui   $2, %hi(_gp_disp)
+  //  1. addiu $2, $2, %lo(_gp_disp)
+  //  2. addu  $globalbasereg, $2, $t9
+  //
+  // We emit only the last instruction here.
+  //
+  // GNU linker requires that the first two instructions appear at the beginning
+  // of a function and no instructions be inserted before or between them.
+  // The two instructions are emitted during lowering to MC layer in order to
+  // avoid any reordering.
+  //
+  // Register $2 (Mips::V0) is added to the list of live-in registers to ensure
+  // the value instruction 1 (addiu) defines is valid when instruction 2 (addu)
+  // reads it.
+  MF.getRegInfo().addLiveIn(Mips::V0);
+  MBB.addLiveIn(Mips::V0);
+  BuildMI(MBB, I, DL, TII.get(Mips::ADDu), GlobalBaseReg)
+    .addReg(Mips::V0).addReg(Mips::T9);
+}
+
+void MipsSEDAGToDAGISel::processFunctionAfterISel(MachineFunction &MF) {
+  initGlobalBaseReg(MF);
+
+  MachineRegisterInfo *MRI = &MF.getRegInfo();
+
+  for (MachineFunction::iterator MFI = MF.begin(), MFE = MF.end(); MFI != MFE;
+       ++MFI)
+    for (MachineBasicBlock::iterator I = MFI->begin(); I != MFI->end(); ++I)
+      replaceUsesWithZeroReg(MRI, *I);
+}
+
+/// Select multiply instructions.
+std::pair<SDNode*, SDNode*>
+MipsSEDAGToDAGISel::selectMULT(SDNode *N, unsigned Opc, DebugLoc DL, EVT Ty,
+                               bool HasLo, bool HasHi) {
+  SDNode *Lo = 0, *Hi = 0;
+  SDNode *Mul = CurDAG->getMachineNode(Opc, DL, MVT::Glue, N->getOperand(0),
+                                       N->getOperand(1));
+  SDValue InFlag = SDValue(Mul, 0);
+
+  if (HasLo) {
+    unsigned Opcode = (Ty == MVT::i32 ? Mips::MFLO : Mips::MFLO64);
+    Lo = CurDAG->getMachineNode(Opcode, DL, Ty, MVT::Glue, InFlag);
+    InFlag = SDValue(Lo, 1);
+  }
+  if (HasHi) {
+    unsigned Opcode = (Ty == MVT::i32 ? Mips::MFHI : Mips::MFHI64);
+    Hi = CurDAG->getMachineNode(Opcode, DL, Ty, InFlag);
+  }
+  return std::make_pair(Lo, Hi);
+}
+
+SDNode *MipsSEDAGToDAGISel::selectAddESubE(unsigned MOp, SDValue InFlag,
+                                           SDValue CmpLHS, DebugLoc DL,
+                                           SDNode *Node) const {
+  unsigned Opc = InFlag.getOpcode(); (void)Opc;
+
+  assert(((Opc == ISD::ADDC || Opc == ISD::ADDE) ||
+          (Opc == ISD::SUBC || Opc == ISD::SUBE)) &&
+         "(ADD|SUB)E flag operand must come from (ADD|SUB)C/E insn");
+
+  SDValue Ops[] = { CmpLHS, InFlag.getOperand(1) };
+  SDValue LHS = Node->getOperand(0), RHS = Node->getOperand(1);
+  EVT VT = LHS.getValueType();
+
+  SDNode *Carry = CurDAG->getMachineNode(Mips::SLTu, DL, VT, Ops, 2);
+  SDNode *AddCarry = CurDAG->getMachineNode(Mips::ADDu, DL, VT,
+                                            SDValue(Carry, 0), RHS);
+  return CurDAG->SelectNodeTo(Node, MOp, VT, MVT::Glue, LHS,
+                              SDValue(AddCarry, 0));
+}
+
+/// ComplexPattern used on MipsInstrInfo
+/// Used on Mips Load/Store instructions
+bool MipsSEDAGToDAGISel::selectAddrRegImm(SDValue Addr, SDValue &Base,
+                                          SDValue &Offset) const {
+  EVT ValTy = Addr.getValueType();
+
+  // if Address is FI, get the TargetFrameIndex.
+  if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) {
+    Base   = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy);
+    Offset = CurDAG->getTargetConstant(0, ValTy);
+    return true;
+  }
+
+  // on PIC code Load GA
+  if (Addr.getOpcode() == MipsISD::Wrapper) {
+    Base   = Addr.getOperand(0);
+    Offset = Addr.getOperand(1);
+    return true;
+  }
+
+  if (TM.getRelocationModel() != Reloc::PIC_) {
+    if ((Addr.getOpcode() == ISD::TargetExternalSymbol ||
+        Addr.getOpcode() == ISD::TargetGlobalAddress))
+      return false;
+  }
+
+  // Addresses of the form FI+const or FI|const
+  if (CurDAG->isBaseWithConstantOffset(Addr)) {
+    ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Addr.getOperand(1));
+    if (isInt<16>(CN->getSExtValue())) {
+
+      // If the first operand is a FI, get the TargetFI Node
+      if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>
+                                  (Addr.getOperand(0)))
+        Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy);
+      else
+        Base = Addr.getOperand(0);
+
+      Offset = CurDAG->getTargetConstant(CN->getZExtValue(), ValTy);
+      return true;
+    }
+  }
+
+  // Operand is a result from an ADD.
+  if (Addr.getOpcode() == ISD::ADD) {
+    // When loading from constant pools, load the lower address part in
+    // the instruction itself. Example, instead of:
+    //  lui $2, %hi($CPI1_0)
+    //  addiu $2, $2, %lo($CPI1_0)
+    //  lwc1 $f0, 0($2)
+    // Generate:
+    //  lui $2, %hi($CPI1_0)
+    //  lwc1 $f0, %lo($CPI1_0)($2)
+    if (Addr.getOperand(1).getOpcode() == MipsISD::Lo ||
+        Addr.getOperand(1).getOpcode() == MipsISD::GPRel) {
+      SDValue Opnd0 = Addr.getOperand(1).getOperand(0);
+      if (isa<ConstantPoolSDNode>(Opnd0) || isa<GlobalAddressSDNode>(Opnd0) ||
+          isa<JumpTableSDNode>(Opnd0)) {
+        Base = Addr.getOperand(0);
+        Offset = Opnd0;
+        return true;
+      }
+    }
+  }
+
+  return false;
+}
+
+bool MipsSEDAGToDAGISel::selectAddrDefault(SDValue Addr, SDValue &Base,
+                                           SDValue &Offset) const {
+  Base = Addr;
+  Offset = CurDAG->getTargetConstant(0, Addr.getValueType());
+  return true;
+}
+
+bool MipsSEDAGToDAGISel::selectIntAddr(SDValue Addr, SDValue &Base,
+                                       SDValue &Offset) const {
+  return selectAddrRegImm(Addr, Base, Offset) ||
+    selectAddrDefault(Addr, Base, Offset);
+}
+
+std::pair<bool, SDNode*> MipsSEDAGToDAGISel::selectNode(SDNode *Node) {
+  unsigned Opcode = Node->getOpcode();
+  DebugLoc DL = Node->getDebugLoc();
+
+  ///
+  // Instruction Selection not handled by the auto-generated
+  // tablegen selection should be handled here.
+  ///
+  EVT NodeTy = Node->getValueType(0);
+  SDNode *Result;
+  unsigned MultOpc;
+
+  switch(Opcode) {
+  default: break;
+
+  case ISD::SUBE: {
+    SDValue InFlag = Node->getOperand(2);
+    Result = selectAddESubE(Mips::SUBu, InFlag, InFlag.getOperand(0), DL, Node);
+    return std::make_pair(true, Result);
+  }
+
+  case ISD::ADDE: {
+    SDValue InFlag = Node->getOperand(2);
+    Result = selectAddESubE(Mips::ADDu, InFlag, InFlag.getValue(0), DL, Node);
+    return std::make_pair(true, Result);
+  }
+
+  /// Mul with two results
+  case ISD::SMUL_LOHI:
+  case ISD::UMUL_LOHI: {
+    if (NodeTy == MVT::i32)
+      MultOpc = (Opcode == ISD::UMUL_LOHI ? Mips::MULTu : Mips::MULT);
+    else
+      MultOpc = (Opcode == ISD::UMUL_LOHI ? Mips::DMULTu : Mips::DMULT);
+
+    std::pair<SDNode*, SDNode*> LoHi = selectMULT(Node, MultOpc, DL, NodeTy,
+                                                  true, true);
+
+    if (!SDValue(Node, 0).use_empty())
+      ReplaceUses(SDValue(Node, 0), SDValue(LoHi.first, 0));
+
+    if (!SDValue(Node, 1).use_empty())
+      ReplaceUses(SDValue(Node, 1), SDValue(LoHi.second, 0));
+
+    return std::make_pair(true, (SDNode*)NULL);
+  }
+
+  /// Special Muls
+  case ISD::MUL: {
+    // Mips32 has a 32-bit three operand mul instruction.
+    if (Subtarget.hasMips32() && NodeTy == MVT::i32)
+      break;
+    MultOpc = NodeTy == MVT::i32 ? Mips::MULT : Mips::DMULT;
+    Result = selectMULT(Node, MultOpc, DL, NodeTy, true, false).first;
+    return std::make_pair(true, Result);
+  }
+  case ISD::MULHS:
+  case ISD::MULHU: {
+    if (NodeTy == MVT::i32)
+      MultOpc = (Opcode == ISD::MULHU ? Mips::MULTu : Mips::MULT);
+    else
+      MultOpc = (Opcode == ISD::MULHU ? Mips::DMULTu : Mips::DMULT);
+
+    Result = selectMULT(Node, MultOpc, DL, NodeTy, false, true).second;
+    return std::make_pair(true, Result);
+  }
+
+  case ISD::ConstantFP: {
+    ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(Node);
+    if (Node->getValueType(0) == MVT::f64 && CN->isExactlyValue(+0.0)) {
+      if (Subtarget.hasMips64()) {
+        SDValue Zero = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), DL,
+                                              Mips::ZERO_64, MVT::i64);
+        Result = CurDAG->getMachineNode(Mips::DMTC1, DL, MVT::f64, Zero);
+      } else {
+        SDValue Zero = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), DL,
+                                              Mips::ZERO, MVT::i32);
+        Result = CurDAG->getMachineNode(Mips::BuildPairF64, DL, MVT::f64, Zero,
+                                        Zero);
+      }
+
+      return std::make_pair(true, Result);
+    }
+    break;
+  }
+
+  case ISD::Constant: {
+    const ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Node);
+    unsigned Size = CN->getValueSizeInBits(0);
+
+    if (Size == 32)
+      break;
+
+    MipsAnalyzeImmediate AnalyzeImm;
+    int64_t Imm = CN->getSExtValue();
+
+    const MipsAnalyzeImmediate::InstSeq &Seq =
+      AnalyzeImm.Analyze(Imm, Size, false);
+
+    MipsAnalyzeImmediate::InstSeq::const_iterator Inst = Seq.begin();
+    DebugLoc DL = CN->getDebugLoc();
+    SDNode *RegOpnd;
+    SDValue ImmOpnd = CurDAG->getTargetConstant(SignExtend64<16>(Inst->ImmOpnd),
+                                                MVT::i64);
+
+    // The first instruction can be a LUi which is different from other
+    // instructions (ADDiu, ORI and SLL) in that it does not have a register
+    // operand.
+    if (Inst->Opc == Mips::LUi64)
+      RegOpnd = CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64, ImmOpnd);
+    else
+      RegOpnd =
+        CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64,
+                               CurDAG->getRegister(Mips::ZERO_64, MVT::i64),
+                               ImmOpnd);
+
+    // The remaining instructions in the sequence are handled here.
+    for (++Inst; Inst != Seq.end(); ++Inst) {
+      ImmOpnd = CurDAG->getTargetConstant(SignExtend64<16>(Inst->ImmOpnd),
+                                          MVT::i64);
+      RegOpnd = CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64,
+                                       SDValue(RegOpnd, 0), ImmOpnd);
+    }
+
+    return std::make_pair(true, RegOpnd);
+  }
+
+  case MipsISD::ThreadPointer: {
+    EVT PtrVT = TLI.getPointerTy();
+    unsigned RdhwrOpc, SrcReg, DestReg;
+
+    if (PtrVT == MVT::i32) {
+      RdhwrOpc = Mips::RDHWR;
+      SrcReg = Mips::HWR29;
+      DestReg = Mips::V1;
+    } else {
+      RdhwrOpc = Mips::RDHWR64;
+      SrcReg = Mips::HWR29_64;
+      DestReg = Mips::V1_64;
+    }
+
+    SDNode *Rdhwr =
+      CurDAG->getMachineNode(RdhwrOpc, Node->getDebugLoc(),
+                             Node->getValueType(0),
+                             CurDAG->getRegister(SrcReg, PtrVT));
+    SDValue Chain = CurDAG->getCopyToReg(CurDAG->getEntryNode(), DL, DestReg,
+                                         SDValue(Rdhwr, 0));
+    SDValue ResNode = CurDAG->getCopyFromReg(Chain, DL, DestReg, PtrVT);
+    ReplaceUses(SDValue(Node, 0), ResNode);
+    return std::make_pair(true, ResNode.getNode());
+  }
+
+  case MipsISD::InsertLOHI: {
+    unsigned RCID = Subtarget.hasDSP() ? Mips::ACRegsDSPRegClassID :
+                                         Mips::ACRegsRegClassID;
+    SDValue RegClass = CurDAG->getTargetConstant(RCID, MVT::i32);
+    SDValue LoIdx = CurDAG->getTargetConstant(Mips::sub_lo, MVT::i32);
+    SDValue HiIdx = CurDAG->getTargetConstant(Mips::sub_hi, MVT::i32);
+    const SDValue Ops[] = { RegClass, Node->getOperand(0), LoIdx,
+                            Node->getOperand(1), HiIdx };
+    SDNode *Res = CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, DL,
+                                         MVT::Untyped, Ops, 5);
+    return std::make_pair(true, Res);
+  }
+  }
+
+  return std::make_pair(false, (SDNode*)NULL);
+}
+
+FunctionPass *llvm::createMipsSEISelDag(MipsTargetMachine &TM) {
+  return new MipsSEDAGToDAGISel(TM);
+}
diff --git a/lib/Target/Mips/MipsSEISelDAGToDAG.h b/lib/Target/Mips/MipsSEISelDAGToDAG.h
new file mode 100644
index 000000000000..6137ab040bbc
--- /dev/null
+++ b/lib/Target/Mips/MipsSEISelDAGToDAG.h
@@ -0,0 +1,57 @@
+//===-- MipsSEISelDAGToDAG.h - A Dag to Dag Inst Selector for MipsSE -----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Subclass of MipsDAGToDAGISel specialized for mips32/64.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MIPSSEISELDAGTODAG_H
+#define MIPSSEISELDAGTODAG_H
+
+#include "MipsISelDAGToDAG.h"
+
+namespace llvm {
+
+class MipsSEDAGToDAGISel : public MipsDAGToDAGISel {
+
+public:
+  explicit MipsSEDAGToDAGISel(MipsTargetMachine &TM) : MipsDAGToDAGISel(TM) {}
+
+private:
+  bool replaceUsesWithZeroReg(MachineRegisterInfo *MRI, const MachineInstr&);
+
+  std::pair<SDNode*, SDNode*> selectMULT(SDNode *N, unsigned Opc, DebugLoc dl,
+                                         EVT Ty, bool HasLo, bool HasHi);
+
+  SDNode *selectAddESubE(unsigned MOp, SDValue InFlag, SDValue CmpLHS,
+                         DebugLoc DL, SDNode *Node) const;
+
+  virtual bool selectAddrRegImm(SDValue Addr, SDValue &Base,
+                                SDValue &Offset) const;
+
+  virtual bool selectAddrDefault(SDValue Addr, SDValue &Base,
+                                 SDValue &Offset) const;
+
+  virtual bool selectIntAddr(SDValue Addr, SDValue &Base,
+                             SDValue &Offset) const;
+
+  virtual std::pair<bool, SDNode*> selectNode(SDNode *Node);
+
+  virtual void processFunctionAfterISel(MachineFunction &MF);
+
+  // Insert instructions to initialize the global base register in the
+  // first MBB of the function.
+  void initGlobalBaseReg(MachineFunction &MF);
+};
+
+FunctionPass *createMipsSEISelDag(MipsTargetMachine &TM);
+
+}
+
+#endif
diff --git a/lib/Target/Mips/MipsSEISelLowering.cpp b/lib/Target/Mips/MipsSEISelLowering.cpp
new file mode 100644
index 000000000000..4f219218d31f
--- /dev/null
+++ b/lib/Target/Mips/MipsSEISelLowering.cpp
@@ -0,0 +1,442 @@
+//===-- MipsSEISelLowering.cpp - MipsSE DAG Lowering Interface --*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Subclass of MipsTargetLowering specialized for mips32/64.
+//
+//===----------------------------------------------------------------------===//
+#include "MipsSEISelLowering.h"
+#include "MipsRegisterInfo.h"
+#include "MipsTargetMachine.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Target/TargetInstrInfo.h"
+
+using namespace llvm;
+
+static cl::opt<bool>
+EnableMipsTailCalls("enable-mips-tail-calls", cl::Hidden,
+                    cl::desc("MIPS: Enable tail calls."), cl::init(false));
+
+MipsSETargetLowering::MipsSETargetLowering(MipsTargetMachine &TM)
+  : MipsTargetLowering(TM) {
+  // Set up the register classes
+  addRegisterClass(MVT::i32, &Mips::CPURegsRegClass);
+
+  if (HasMips64)
+    addRegisterClass(MVT::i64, &Mips::CPU64RegsRegClass);
+
+  if (Subtarget->hasDSP()) {
+    MVT::SimpleValueType VecTys[2] = {MVT::v2i16, MVT::v4i8};
+
+    for (unsigned i = 0; i < array_lengthof(VecTys); ++i) {
+      addRegisterClass(VecTys[i], &Mips::DSPRegsRegClass);
+
+      // Expand all builtin opcodes.
+      for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc)
+        setOperationAction(Opc, VecTys[i], Expand);
+
+      setOperationAction(ISD::LOAD, VecTys[i], Legal);
+      setOperationAction(ISD::STORE, VecTys[i], Legal);
+      setOperationAction(ISD::BITCAST, VecTys[i], Legal);
+    }
+  }
+
+  if (!TM.Options.UseSoftFloat) {
+    addRegisterClass(MVT::f32, &Mips::FGR32RegClass);
+
+    // When dealing with single precision only, use libcalls
+    if (!Subtarget->isSingleFloat()) {
+      if (HasMips64)
+        addRegisterClass(MVT::f64, &Mips::FGR64RegClass);
+      else
+        addRegisterClass(MVT::f64, &Mips::AFGR64RegClass);
+    }
+  }
+
+  setOperationAction(ISD::SMUL_LOHI,          MVT::i32, Custom);
+  setOperationAction(ISD::UMUL_LOHI,          MVT::i32, Custom);
+  setOperationAction(ISD::MULHS,              MVT::i32, Custom);
+  setOperationAction(ISD::MULHU,              MVT::i32, Custom);
+
+  if (HasMips64)
+    setOperationAction(ISD::MUL,              MVT::i64, Custom);
+
+  setOperationAction(ISD::SDIVREM, MVT::i32, Custom);
+  setOperationAction(ISD::UDIVREM, MVT::i32, Custom);
+  setOperationAction(ISD::SDIVREM, MVT::i64, Custom);
+  setOperationAction(ISD::UDIVREM, MVT::i64, Custom);
+  setOperationAction(ISD::MEMBARRIER,         MVT::Other, Custom);
+  setOperationAction(ISD::ATOMIC_FENCE,       MVT::Other, Custom);
+  setOperationAction(ISD::LOAD,               MVT::i32, Custom);
+  setOperationAction(ISD::STORE,              MVT::i32, Custom);
+
+  setTargetDAGCombine(ISD::ADDE);
+  setTargetDAGCombine(ISD::SUBE);
+
+  computeRegisterProperties();
+}
+
+const MipsTargetLowering *
+llvm::createMipsSETargetLowering(MipsTargetMachine &TM) {
+  return new MipsSETargetLowering(TM);
+}
+
+
+bool
+MipsSETargetLowering::allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const {
+  MVT::SimpleValueType SVT = VT.getSimpleVT().SimpleTy;
+
+  switch (SVT) {
+  case MVT::i64:
+  case MVT::i32:
+    if (Fast)
+      *Fast = true;
+    return true;
+  default:
+    return false;
+  }
+}
+
+SDValue MipsSETargetLowering::LowerOperation(SDValue Op,
+                                             SelectionDAG &DAG) const {
+  switch(Op.getOpcode()) {
+  case ISD::SMUL_LOHI: return lowerMulDiv(Op, MipsISD::Mult, true, true, DAG);
+  case ISD::UMUL_LOHI: return lowerMulDiv(Op, MipsISD::Multu, true, true, DAG);
+  case ISD::MULHS:     return lowerMulDiv(Op, MipsISD::Mult, false, true, DAG);
+  case ISD::MULHU:     return lowerMulDiv(Op, MipsISD::Multu, false, true, DAG);
+  case ISD::MUL:       return lowerMulDiv(Op, MipsISD::Mult, true, false, DAG);
+  case ISD::SDIVREM:   return lowerMulDiv(Op, MipsISD::DivRem, true, true, DAG);
+  case ISD::UDIVREM:   return lowerMulDiv(Op, MipsISD::DivRemU, true, true, DAG);
+  }
+
+  return MipsTargetLowering::LowerOperation(Op, DAG);
+}
+
+// selectMADD -
+// Transforms a subgraph in CurDAG if the following pattern is found:
+//  (addc multLo, Lo0), (adde multHi, Hi0),
+// where,
+//  multHi/Lo: product of multiplication
+//  Lo0: initial value of Lo register
+//  Hi0: initial value of Hi register
+// Return true if pattern matching was successful.
+static bool selectMADD(SDNode *ADDENode, SelectionDAG *CurDAG) {
+  // ADDENode's second operand must be a flag output of an ADDC node in order
+  // for the matching to be successful.
+  SDNode *ADDCNode = ADDENode->getOperand(2).getNode();
+
+  if (ADDCNode->getOpcode() != ISD::ADDC)
+    return false;
+
+  SDValue MultHi = ADDENode->getOperand(0);
+  SDValue MultLo = ADDCNode->getOperand(0);
+  SDNode *MultNode = MultHi.getNode();
+  unsigned MultOpc = MultHi.getOpcode();
+
+  // MultHi and MultLo must be generated by the same node,
+  if (MultLo.getNode() != MultNode)
+    return false;
+
+  // and it must be a multiplication.
+  if (MultOpc != ISD::SMUL_LOHI && MultOpc != ISD::UMUL_LOHI)
+    return false;
+
+  // MultLo amd MultHi must be the first and second output of MultNode
+  // respectively.
+  if (MultHi.getResNo() != 1 || MultLo.getResNo() != 0)
+    return false;
+
+  // Transform this to a MADD only if ADDENode and ADDCNode are the only users
+  // of the values of MultNode, in which case MultNode will be removed in later
+  // phases.
+  // If there exist users other than ADDENode or ADDCNode, this function returns
+  // here, which will result in MultNode being mapped to a single MULT
+  // instruction node rather than a pair of MULT and MADD instructions being
+  // produced.
+  if (!MultHi.hasOneUse() || !MultLo.hasOneUse())
+    return false;
+
+  DebugLoc DL = ADDENode->getDebugLoc();
+
+  // Initialize accumulator.
+  SDValue ACCIn = CurDAG->getNode(MipsISD::InsertLOHI, DL, MVT::Untyped,
+                                  ADDCNode->getOperand(1),
+                                  ADDENode->getOperand(1));
+
+  // create MipsMAdd(u) node
+  MultOpc = MultOpc == ISD::UMUL_LOHI ? MipsISD::MAddu : MipsISD::MAdd;
+
+  SDValue MAdd = CurDAG->getNode(MultOpc, DL, MVT::Untyped,
+                                 MultNode->getOperand(0),// Factor 0
+                                 MultNode->getOperand(1),// Factor 1
+                                 ACCIn);
+
+  // replace uses of adde and addc here
+  if (!SDValue(ADDCNode, 0).use_empty()) {
+    SDValue LoIdx = CurDAG->getConstant(Mips::sub_lo, MVT::i32);
+    SDValue LoOut = CurDAG->getNode(MipsISD::ExtractLOHI, DL, MVT::i32, MAdd,
+                                    LoIdx);
+    CurDAG->ReplaceAllUsesOfValueWith(SDValue(ADDCNode, 0), LoOut);
+  }
+  if (!SDValue(ADDENode, 0).use_empty()) {
+    SDValue HiIdx = CurDAG->getConstant(Mips::sub_hi, MVT::i32);
+    SDValue HiOut = CurDAG->getNode(MipsISD::ExtractLOHI, DL, MVT::i32, MAdd,
+                                    HiIdx);
+    CurDAG->ReplaceAllUsesOfValueWith(SDValue(ADDENode, 0), HiOut);
+  }
+
+  return true;
+}
+
+// selectMSUB -
+// Transforms a subgraph in CurDAG if the following pattern is found:
+//  (addc Lo0, multLo), (sube Hi0, multHi),
+// where,
+//  multHi/Lo: product of multiplication
+//  Lo0: initial value of Lo register
+//  Hi0: initial value of Hi register
+// Return true if pattern matching was successful.
+static bool selectMSUB(SDNode *SUBENode, SelectionDAG *CurDAG) {
+  // SUBENode's second operand must be a flag output of an SUBC node in order
+  // for the matching to be successful.
+  SDNode *SUBCNode = SUBENode->getOperand(2).getNode();
+
+  if (SUBCNode->getOpcode() != ISD::SUBC)
+    return false;
+
+  SDValue MultHi = SUBENode->getOperand(1);
+  SDValue MultLo = SUBCNode->getOperand(1);
+  SDNode *MultNode = MultHi.getNode();
+  unsigned MultOpc = MultHi.getOpcode();
+
+  // MultHi and MultLo must be generated by the same node,
+  if (MultLo.getNode() != MultNode)
+    return false;
+
+  // and it must be a multiplication.
+  if (MultOpc != ISD::SMUL_LOHI && MultOpc != ISD::UMUL_LOHI)
+    return false;
+
+  // MultLo amd MultHi must be the first and second output of MultNode
+  // respectively.
+  if (MultHi.getResNo() != 1 || MultLo.getResNo() != 0)
+    return false;
+
+  // Transform this to a MSUB only if SUBENode and SUBCNode are the only users
+  // of the values of MultNode, in which case MultNode will be removed in later
+  // phases.
+  // If there exist users other than SUBENode or SUBCNode, this function returns
+  // here, which will result in MultNode being mapped to a single MULT
+  // instruction node rather than a pair of MULT and MSUB instructions being
+  // produced.
+  if (!MultHi.hasOneUse() || !MultLo.hasOneUse())
+    return false;
+
+  DebugLoc DL = SUBENode->getDebugLoc();
+
+  // Initialize accumulator.
+  SDValue ACCIn = CurDAG->getNode(MipsISD::InsertLOHI, DL, MVT::Untyped,
+                                  SUBCNode->getOperand(0),
+                                  SUBENode->getOperand(0));
+
+  // create MipsSub(u) node
+  MultOpc = MultOpc == ISD::UMUL_LOHI ? MipsISD::MSubu : MipsISD::MSub;
+
+  SDValue MSub = CurDAG->getNode(MultOpc, DL, MVT::Glue,
+                                 MultNode->getOperand(0),// Factor 0
+                                 MultNode->getOperand(1),// Factor 1
+                                 ACCIn);
+
+  // replace uses of sube and subc here
+  if (!SDValue(SUBCNode, 0).use_empty()) {
+    SDValue LoIdx = CurDAG->getConstant(Mips::sub_lo, MVT::i32);
+    SDValue LoOut = CurDAG->getNode(MipsISD::ExtractLOHI, DL, MVT::i32, MSub,
+                                    LoIdx);
+    CurDAG->ReplaceAllUsesOfValueWith(SDValue(SUBCNode, 0), LoOut);
+  }
+  if (!SDValue(SUBENode, 0).use_empty()) {
+    SDValue HiIdx = CurDAG->getConstant(Mips::sub_hi, MVT::i32);
+    SDValue HiOut = CurDAG->getNode(MipsISD::ExtractLOHI, DL, MVT::i32, MSub,
+                                    HiIdx);
+    CurDAG->ReplaceAllUsesOfValueWith(SDValue(SUBENode, 0), HiOut);
+  }
+
+  return true;
+}
+
+static SDValue performADDECombine(SDNode *N, SelectionDAG &DAG,
+                                  TargetLowering::DAGCombinerInfo &DCI,
+                                  const MipsSubtarget *Subtarget) {
+  if (DCI.isBeforeLegalize())
+    return SDValue();
+
+  if (Subtarget->hasMips32() && N->getValueType(0) == MVT::i32 &&
+      selectMADD(N, &DAG))
+    return SDValue(N, 0);
+
+  return SDValue();
+}
+
+static SDValue performSUBECombine(SDNode *N, SelectionDAG &DAG,
+                                  TargetLowering::DAGCombinerInfo &DCI,
+                                  const MipsSubtarget *Subtarget) {
+  if (DCI.isBeforeLegalize())
+    return SDValue();
+
+  if (Subtarget->hasMips32() && N->getValueType(0) == MVT::i32 &&
+      selectMSUB(N, &DAG))
+    return SDValue(N, 0);
+
+  return SDValue();
+}
+
+SDValue
+MipsSETargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const {
+  SelectionDAG &DAG = DCI.DAG;
+
+  switch (N->getOpcode()) {
+  case ISD::ADDE:
+    return performADDECombine(N, DAG, DCI, Subtarget);
+  case ISD::SUBE:
+    return performSUBECombine(N, DAG, DCI, Subtarget);
+  default:
+    return MipsTargetLowering::PerformDAGCombine(N, DCI);
+  }
+}
+
+MachineBasicBlock *
+MipsSETargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
+                                                  MachineBasicBlock *BB) const {
+  switch (MI->getOpcode()) {
+  default:
+    return MipsTargetLowering::EmitInstrWithCustomInserter(MI, BB);
+  case Mips::BPOSGE32_PSEUDO:
+    return emitBPOSGE32(MI, BB);
+  }
+}
+
+bool MipsSETargetLowering::
+isEligibleForTailCallOptimization(const MipsCC &MipsCCInfo,
+                                  unsigned NextStackOffset,
+                                  const MipsFunctionInfo& FI) const {
+  if (!EnableMipsTailCalls)
+    return false;
+
+  // Return false if either the callee or caller has a byval argument.
+  if (MipsCCInfo.hasByValArg() || FI.hasByvalArg())
+    return false;
+
+  // Return true if the callee's argument area is no larger than the
+  // caller's.
+  return NextStackOffset <= FI.getIncomingArgSize();
+}
+
+void MipsSETargetLowering::
+getOpndList(SmallVectorImpl<SDValue> &Ops,
+            std::deque< std::pair<unsigned, SDValue> > &RegsToPass,
+            bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage,
+            CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const {
+  // T9 should contain the address of the callee function if
+  // -reloction-model=pic or it is an indirect call.
+  if (IsPICCall || !GlobalOrExternal) {
+    unsigned T9Reg = IsN64 ? Mips::T9_64 : Mips::T9;
+    RegsToPass.push_front(std::make_pair(T9Reg, Callee));
+  } else
+    Ops.push_back(Callee);
+
+  MipsTargetLowering::getOpndList(Ops, RegsToPass, IsPICCall, GlobalOrExternal,
+                                  InternalLinkage, CLI, Callee, Chain);
+}
+
+SDValue MipsSETargetLowering::lowerMulDiv(SDValue Op, unsigned NewOpc,
+                                          bool HasLo, bool HasHi,
+                                          SelectionDAG &DAG) const {
+  EVT Ty = Op.getOperand(0).getValueType();
+  DebugLoc DL = Op.getDebugLoc();
+  SDValue Mult = DAG.getNode(NewOpc, DL, MVT::Untyped,
+                             Op.getOperand(0), Op.getOperand(1));
+  SDValue Lo, Hi;
+
+  if (HasLo)
+    Lo = DAG.getNode(MipsISD::ExtractLOHI, DL, Ty, Mult,
+                     DAG.getConstant(Mips::sub_lo, MVT::i32));
+  if (HasHi)
+    Hi = DAG.getNode(MipsISD::ExtractLOHI, DL, Ty, Mult,
+                     DAG.getConstant(Mips::sub_hi, MVT::i32));
+
+  if (!HasLo || !HasHi)
+    return HasLo ? Lo : Hi;
+
+  SDValue Vals[] = { Lo, Hi };
+  return DAG.getMergeValues(Vals, 2, DL);
+}
+
+MachineBasicBlock * MipsSETargetLowering::
+emitBPOSGE32(MachineInstr *MI, MachineBasicBlock *BB) const{
+  // $bb:
+  //  bposge32_pseudo $vr0
+  //  =>
+  // $bb:
+  //  bposge32 $tbb
+  // $fbb:
+  //  li $vr2, 0
+  //  b $sink
+  // $tbb:
+  //  li $vr1, 1
+  // $sink:
+  //  $vr0 = phi($vr2, $fbb, $vr1, $tbb)
+
+  MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+  const TargetRegisterClass *RC = &Mips::CPURegsRegClass;
+  DebugLoc DL = MI->getDebugLoc();
+  const BasicBlock *LLVM_BB = BB->getBasicBlock();
+  MachineFunction::iterator It = llvm::next(MachineFunction::iterator(BB));
+  MachineFunction *F = BB->getParent();
+  MachineBasicBlock *FBB = F->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *TBB = F->CreateMachineBasicBlock(LLVM_BB);
+  MachineBasicBlock *Sink  = F->CreateMachineBasicBlock(LLVM_BB);
+  F->insert(It, FBB);
+  F->insert(It, TBB);
+  F->insert(It, Sink);
+
+  // Transfer the remainder of BB and its successor edges to Sink.
+  Sink->splice(Sink->begin(), BB, llvm::next(MachineBasicBlock::iterator(MI)),
+               BB->end());
+  Sink->transferSuccessorsAndUpdatePHIs(BB);
+
+  // Add successors.
+  BB->addSuccessor(FBB);
+  BB->addSuccessor(TBB);
+  FBB->addSuccessor(Sink);
+  TBB->addSuccessor(Sink);
+
+  // Insert the real bposge32 instruction to $BB.
+  BuildMI(BB, DL, TII->get(Mips::BPOSGE32)).addMBB(TBB);
+
+  // Fill $FBB.
+  unsigned VR2 = RegInfo.createVirtualRegister(RC);
+  BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::ADDiu), VR2)
+    .addReg(Mips::ZERO).addImm(0);
+  BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::B)).addMBB(Sink);
+
+  // Fill $TBB.
+  unsigned VR1 = RegInfo.createVirtualRegister(RC);
+  BuildMI(*TBB, TBB->end(), DL, TII->get(Mips::ADDiu), VR1)
+    .addReg(Mips::ZERO).addImm(1);
+
+  // Insert phi function to $Sink.
+  BuildMI(*Sink, Sink->begin(), DL, TII->get(Mips::PHI),
+          MI->getOperand(0).getReg())
+    .addReg(VR2).addMBB(FBB).addReg(VR1).addMBB(TBB);
+
+  MI->eraseFromParent();   // The pseudo instruction is gone now.
+  return Sink;
+}
diff --git a/lib/Target/Mips/MipsSEISelLowering.h b/lib/Target/Mips/MipsSEISelLowering.h
new file mode 100644
index 000000000000..186f6a343dee
--- /dev/null
+++ b/lib/Target/Mips/MipsSEISelLowering.h
@@ -0,0 +1,62 @@
+//===-- MipsSEISelLowering.h - MipsSE DAG Lowering Interface ----*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Subclass of MipsTargetLowering specialized for mips32/64.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MipsSEISELLOWERING_H
+#define MipsSEISELLOWERING_H
+
+#include "MipsISelLowering.h"
+#include "MipsRegisterInfo.h"
+
+namespace llvm {
+  class MipsSETargetLowering : public MipsTargetLowering  {
+  public:
+    explicit MipsSETargetLowering(MipsTargetMachine &TM);
+
+    virtual bool allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const;
+
+    virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
+
+    virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
+
+    virtual MachineBasicBlock *
+    EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *MBB) const;
+
+    virtual const TargetRegisterClass *getRepRegClassFor(MVT VT) const {
+      if (VT == MVT::Untyped)
+        return Subtarget->hasDSP() ? &Mips::ACRegsDSPRegClass :
+                                     &Mips::ACRegsRegClass;
+
+      return TargetLowering::getRepRegClassFor(VT);
+    }
+
+  private:
+    virtual bool
+    isEligibleForTailCallOptimization(const MipsCC &MipsCCInfo,
+                                      unsigned NextStackOffset,
+                                      const MipsFunctionInfo& FI) const;
+
+    virtual void
+    getOpndList(SmallVectorImpl<SDValue> &Ops,
+                std::deque< std::pair<unsigned, SDValue> > &RegsToPass,
+                bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage,
+                CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const;
+
+    SDValue lowerMulDiv(SDValue Op, unsigned NewOpc, bool HasLo, bool HasHi,
+                        SelectionDAG &DAG) const;
+
+    MachineBasicBlock *emitBPOSGE32(MachineInstr *MI,
+                                    MachineBasicBlock *BB) const;
+  };
+}
+
+#endif // MipsSEISELLOWERING_H
diff --git a/lib/Target/Mips/MipsSEInstrInfo.cpp b/lib/Target/Mips/MipsSEInstrInfo.cpp
index fb0f9df038c3..ca0315ed9f6e 100644
--- a/lib/Target/Mips/MipsSEInstrInfo.cpp
+++ b/lib/Target/Mips/MipsSEInstrInfo.cpp
@@ -12,14 +12,14 @@
 //===----------------------------------------------------------------------===//
 
 #include "MipsSEInstrInfo.h"
-#include "MipsTargetMachine.h"
-#include "MipsMachineFunction.h"
 #include "InstPrinter/MipsInstPrinter.h"
+#include "MipsMachineFunction.h"
+#include "MipsTargetMachine.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
-#include "llvm/ADT/STLExtras.h"
 
 using namespace llvm;
 
@@ -90,7 +90,7 @@ void MipsSEInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
 
   if (Mips::CPURegsRegClass.contains(DestReg)) { // Copy to CPU Reg.
     if (Mips::CPURegsRegClass.contains(SrcReg))
-      Opc = Mips::ADDu, ZeroReg = Mips::ZERO;
+      Opc = Mips::OR, ZeroReg = Mips::ZERO;
     else if (Mips::CCRRegClass.contains(SrcReg))
       Opc = Mips::CFC1;
     else if (Mips::FGR32RegClass.contains(SrcReg))
@@ -120,7 +120,7 @@ void MipsSEInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
     Opc = Mips::MOVCCRToCCR;
   else if (Mips::CPU64RegsRegClass.contains(DestReg)) { // Copy to CPU64 Reg.
     if (Mips::CPU64RegsRegClass.contains(SrcReg))
-      Opc = Mips::DADDu, ZeroReg = Mips::ZERO_64;
+      Opc = Mips::OR64, ZeroReg = Mips::ZERO_64;
     else if (SrcReg == Mips::HI64)
       Opc = Mips::MFHI64, SrcReg = 0;
     else if (SrcReg == Mips::LO64)
@@ -136,6 +136,12 @@ void MipsSEInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
     else if (Mips::FGR64RegClass.contains(DestReg))
       Opc = Mips::DMTC1;
   }
+  else if (Mips::ACRegsRegClass.contains(DestReg, SrcReg))
+    Opc = Mips::COPY_AC64;
+  else if (Mips::ACRegsDSPRegClass.contains(DestReg, SrcReg))
+    Opc = Mips::COPY_AC_DSP;
+  else if (Mips::ACRegs128RegClass.contains(DestReg, SrcReg))
+    Opc = Mips::COPY_AC128;
 
   assert(Opc && "Cannot copy registers");
 
@@ -144,18 +150,18 @@ void MipsSEInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
   if (DestReg)
     MIB.addReg(DestReg, RegState::Define);
 
-  if (ZeroReg)
-    MIB.addReg(ZeroReg);
-
   if (SrcReg)
     MIB.addReg(SrcReg, getKillRegState(KillSrc));
+
+  if (ZeroReg)
+    MIB.addReg(ZeroReg);
 }
 
 void MipsSEInstrInfo::
-storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
-                    unsigned SrcReg, bool isKill, int FI,
-                    const TargetRegisterClass *RC,
-                    const TargetRegisterInfo *TRI) const {
+storeRegToStack(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
+                unsigned SrcReg, bool isKill, int FI,
+                const TargetRegisterClass *RC, const TargetRegisterInfo *TRI,
+                int64_t Offset) const {
   DebugLoc DL;
   if (I != MBB.end()) DL = I->getDebugLoc();
   MachineMemOperand *MMO = GetMemOperand(MBB, FI, MachineMemOperand::MOStore);
@@ -166,6 +172,12 @@ storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
     Opc = IsN64 ? Mips::SW_P8 : Mips::SW;
   else if (Mips::CPU64RegsRegClass.hasSubClassEq(RC))
     Opc = IsN64 ? Mips::SD_P8 : Mips::SD;
+  else if (Mips::ACRegsRegClass.hasSubClassEq(RC))
+    Opc = IsN64 ? Mips::STORE_AC64_P8 : Mips::STORE_AC64;
+  else if (Mips::ACRegsDSPRegClass.hasSubClassEq(RC))
+    Opc = IsN64 ? Mips::STORE_AC_DSP_P8 : Mips::STORE_AC_DSP;
+  else if (Mips::ACRegs128RegClass.hasSubClassEq(RC))
+    Opc = IsN64 ? Mips::STORE_AC128_P8 : Mips::STORE_AC128;
   else if (Mips::FGR32RegClass.hasSubClassEq(RC))
     Opc = IsN64 ? Mips::SWC1_P8 : Mips::SWC1;
   else if (Mips::AFGR64RegClass.hasSubClassEq(RC))
@@ -175,15 +187,13 @@ storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
 
   assert(Opc && "Register class not handled!");
   BuildMI(MBB, I, DL, get(Opc)).addReg(SrcReg, getKillRegState(isKill))
-    .addFrameIndex(FI).addImm(0).addMemOperand(MMO);
+    .addFrameIndex(FI).addImm(Offset).addMemOperand(MMO);
 }
 
 void MipsSEInstrInfo::
-loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
-                     unsigned DestReg, int FI,
-                     const TargetRegisterClass *RC,
-                     const TargetRegisterInfo *TRI) const
-{
+loadRegFromStack(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
+                 unsigned DestReg, int FI, const TargetRegisterClass *RC,
+                 const TargetRegisterInfo *TRI, int64_t Offset) const {
   DebugLoc DL;
   if (I != MBB.end()) DL = I->getDebugLoc();
   MachineMemOperand *MMO = GetMemOperand(MBB, FI, MachineMemOperand::MOLoad);
@@ -193,6 +203,12 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
     Opc = IsN64 ? Mips::LW_P8 : Mips::LW;
   else if (Mips::CPU64RegsRegClass.hasSubClassEq(RC))
     Opc = IsN64 ? Mips::LD_P8 : Mips::LD;
+  else if (Mips::ACRegsRegClass.hasSubClassEq(RC))
+    Opc = IsN64 ? Mips::LOAD_AC64_P8 : Mips::LOAD_AC64;
+  else if (Mips::ACRegsDSPRegClass.hasSubClassEq(RC))
+    Opc = IsN64 ? Mips::LOAD_AC_DSP_P8 : Mips::LOAD_AC_DSP;
+  else if (Mips::ACRegs128RegClass.hasSubClassEq(RC))
+    Opc = IsN64 ? Mips::LOAD_AC128_P8 : Mips::LOAD_AC128;
   else if (Mips::FGR32RegClass.hasSubClassEq(RC))
     Opc = IsN64 ? Mips::LWC1_P8 : Mips::LWC1;
   else if (Mips::AFGR64RegClass.hasSubClassEq(RC))
@@ -201,7 +217,7 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
     Opc = IsN64 ? Mips::LDC164_P8 : Mips::LDC164;
 
   assert(Opc && "Register class not handled!");
-  BuildMI(MBB, I, DL, get(Opc), DestReg).addFrameIndex(FI).addImm(0)
+  BuildMI(MBB, I, DL, get(Opc), DestReg).addFrameIndex(FI).addImm(Offset)
     .addMemOperand(MMO);
 }
 
@@ -220,6 +236,10 @@ bool MipsSEInstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
   case Mips::ExtractElementF64:
     ExpandExtractElementF64(MBB, MI);
     break;
+  case Mips::MIPSeh_return32:
+  case Mips::MIPSeh_return64:
+    ExpandEhReturn(MBB, MI);
+    break;
   }
 
   MBB.erase(MI);
@@ -356,6 +376,35 @@ void MipsSEInstrInfo::ExpandBuildPairF64(MachineBasicBlock &MBB,
     .addReg(HiReg);
 }
 
+void MipsSEInstrInfo::ExpandEhReturn(MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I) const {
+  // This pseudo instruction is generated as part of the lowering of
+  // ISD::EH_RETURN. We convert it to a stack increment by OffsetReg, and
+  // indirect jump to TargetReg
+  const MipsSubtarget &STI = TM.getSubtarget<MipsSubtarget>();
+  unsigned ADDU = STI.isABI_N64() ? Mips::DADDu : Mips::ADDu;
+  unsigned OR = STI.isABI_N64() ? Mips::OR64 : Mips::OR;
+  unsigned JR = STI.isABI_N64() ? Mips::JR64 : Mips::JR;
+  unsigned SP = STI.isABI_N64() ? Mips::SP_64 : Mips::SP;
+  unsigned RA = STI.isABI_N64() ? Mips::RA_64 : Mips::RA;
+  unsigned T9 = STI.isABI_N64() ? Mips::T9_64 : Mips::T9;
+  unsigned ZERO = STI.isABI_N64() ? Mips::ZERO_64 : Mips::ZERO;
+  unsigned OffsetReg = I->getOperand(0).getReg();
+  unsigned TargetReg = I->getOperand(1).getReg();
+
+  // or   $ra, $v0, $zero
+  // addu $sp, $sp, $v1
+  // jr   $ra
+  if (TM.getRelocationModel() == Reloc::PIC_)
+    BuildMI(MBB, I, I->getDebugLoc(), TM.getInstrInfo()->get(OR), T9)
+        .addReg(TargetReg).addReg(ZERO);
+  BuildMI(MBB, I, I->getDebugLoc(), TM.getInstrInfo()->get(OR), RA)
+      .addReg(TargetReg).addReg(ZERO);
+  BuildMI(MBB, I, I->getDebugLoc(), TM.getInstrInfo()->get(ADDU), SP)
+      .addReg(SP).addReg(OffsetReg);
+  BuildMI(MBB, I, I->getDebugLoc(), TM.getInstrInfo()->get(JR)).addReg(RA);
+}
+
 const MipsInstrInfo *llvm::createMipsSEInstrInfo(MipsTargetMachine &TM) {
   return new MipsSEInstrInfo(TM);
 }
diff --git a/lib/Target/Mips/MipsSEInstrInfo.h b/lib/Target/Mips/MipsSEInstrInfo.h
index 55b78b2cfb97..0bf7876f0fe0 100644
--- a/lib/Target/Mips/MipsSEInstrInfo.h
+++ b/lib/Target/Mips/MipsSEInstrInfo.h
@@ -49,17 +49,19 @@ public:
                            unsigned DestReg, unsigned SrcReg,
                            bool KillSrc) const;
 
-  virtual void storeRegToStackSlot(MachineBasicBlock &MBB,
-                                   MachineBasicBlock::iterator MBBI,
-                                   unsigned SrcReg, bool isKill, int FrameIndex,
-                                   const TargetRegisterClass *RC,
-                                   const TargetRegisterInfo *TRI) const;
-
-  virtual void loadRegFromStackSlot(MachineBasicBlock &MBB,
-                                    MachineBasicBlock::iterator MBBI,
-                                    unsigned DestReg, int FrameIndex,
-                                    const TargetRegisterClass *RC,
-                                    const TargetRegisterInfo *TRI) const;
+  virtual void storeRegToStack(MachineBasicBlock &MBB,
+                               MachineBasicBlock::iterator MI,
+                               unsigned SrcReg, bool isKill, int FrameIndex,
+                               const TargetRegisterClass *RC,
+                               const TargetRegisterInfo *TRI,
+                               int64_t Offset) const;
+
+  virtual void loadRegFromStack(MachineBasicBlock &MBB,
+                                MachineBasicBlock::iterator MI,
+                                unsigned DestReg, int FrameIndex,
+                                const TargetRegisterClass *RC,
+                                const TargetRegisterInfo *TRI,
+                                int64_t Offset) const;
 
   virtual bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const;
 
@@ -85,6 +87,8 @@ private:
                                MachineBasicBlock::iterator I) const;
   void ExpandBuildPairF64(MachineBasicBlock &MBB,
                           MachineBasicBlock::iterator I) const;
+  void ExpandEhReturn(MachineBasicBlock &MBB,
+                      MachineBasicBlock::iterator I) const;
 };
 
 }
diff --git a/lib/Target/Mips/MipsSERegisterInfo.cpp b/lib/Target/Mips/MipsSERegisterInfo.cpp
index 56b9ba95e5de..96967380b29d 100644
--- a/lib/Target/Mips/MipsSERegisterInfo.cpp
+++ b/lib/Target/Mips/MipsSERegisterInfo.cpp
@@ -15,28 +15,28 @@
 #include "MipsSERegisterInfo.h"
 #include "Mips.h"
 #include "MipsAnalyzeImmediate.h"
+#include "MipsMachineFunction.h"
 #include "MipsSEInstrInfo.h"
 #include "MipsSubtarget.h"
-#include "MipsMachineFunction.h"
-#include "llvm/Constants.h"
-#include "llvm/DebugInfo.h"
-#include "llvm/Type.h"
-#include "llvm/Function.h"
-#include "llvm/CodeGen/ValueTypes.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/DebugInfo.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/STLExtras.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
@@ -54,26 +54,13 @@ requiresFrameIndexScavenging(const MachineFunction &MF) const {
   return true;
 }
 
-// This function eliminate ADJCALLSTACKDOWN,
-// ADJCALLSTACKUP pseudo instructions
-void MipsSERegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-
-  if (!TFI->hasReservedCallFrame(MF)) {
-    int64_t Amount = I->getOperand(0).getImm();
-
-    if (I->getOpcode() == Mips::ADJCALLSTACKDOWN)
-      Amount = -Amount;
-
-    const MipsSEInstrInfo *II = static_cast<const MipsSEInstrInfo*>(&TII);
-    unsigned SP = Subtarget.isABI_N64() ? Mips::SP_64 : Mips::SP;
+const TargetRegisterClass *
+MipsSERegisterInfo::intRegClass(unsigned Size) const {
+  if (Size == 4)
+    return &Mips::CPURegsRegClass;
 
-    II->adjustStackPtr(SP, Amount, MBB, I);
-  }
-
-  MBB.erase(I);
+  assert(Size == 8);
+  return &Mips::CPU64RegsRegClass;
 }
 
 void MipsSERegisterInfo::eliminateFI(MachineBasicBlock::iterator II,
@@ -83,6 +70,7 @@ void MipsSERegisterInfo::eliminateFI(MachineBasicBlock::iterator II,
   MachineInstr &MI = *II;
   MachineFunction &MF = *MI.getParent()->getParent();
   MachineFrameInfo *MFI = MF.getFrameInfo();
+  MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>();
 
   const std::vector<CalleeSavedInfo> &CSI = MFI->getCalleeSavedInfo();
   int MinCSFI = 0;
@@ -93,15 +81,18 @@ void MipsSERegisterInfo::eliminateFI(MachineBasicBlock::iterator II,
     MaxCSFI = CSI[CSI.size() - 1].getFrameIdx();
   }
 
+  bool EhDataRegFI = MipsFI->isEhDataRegFI(FrameIndex);
+
   // The following stack frame objects are always referenced relative to $sp:
   //  1. Outgoing arguments.
   //  2. Pointer to dynamically allocated stack space.
   //  3. Locations for callee-saved registers.
+  //  4. Locations for eh data registers.
   // Everything else is referenced relative to whatever register
   // getFrameRegister() returns.
   unsigned FrameReg;
 
-  if (FrameIndex >= MinCSFI && FrameIndex <= MaxCSFI)
+  if ((FrameIndex >= MinCSFI && FrameIndex <= MaxCSFI) || EhDataRegFI)
     FrameReg = Subtarget.isABI_N64() ? Mips::SP_64 : Mips::SP;
   else
     FrameReg = getFrameRegister(MF);
diff --git a/lib/Target/Mips/MipsSERegisterInfo.h b/lib/Target/Mips/MipsSERegisterInfo.h
index 7437bd36c333..2f7c37bb460d 100644
--- a/lib/Target/Mips/MipsSERegisterInfo.h
+++ b/lib/Target/Mips/MipsSERegisterInfo.h
@@ -31,9 +31,7 @@ public:
 
   bool requiresFrameIndexScavenging(const MachineFunction &MF) const;
 
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
+  virtual const TargetRegisterClass *intRegClass(unsigned Size) const;
 
 private:
   virtual void eliminateFI(MachineBasicBlock::iterator II, unsigned OpNo,
diff --git a/lib/Target/Mips/MipsSubtarget.cpp b/lib/Target/Mips/MipsSubtarget.cpp
index 930af4dda159..e11e5d142b74 100644
--- a/lib/Target/Mips/MipsSubtarget.cpp
+++ b/lib/Target/Mips/MipsSubtarget.cpp
@@ -26,13 +26,14 @@ void MipsSubtarget::anchor() { }
 
 MipsSubtarget::MipsSubtarget(const std::string &TT, const std::string &CPU,
                              const std::string &FS, bool little,
-                             Reloc::Model RM) :
+                             Reloc::Model _RM) :
   MipsGenSubtargetInfo(TT, CPU, FS),
   MipsArchVersion(Mips32), MipsABI(UnknownABI), IsLittle(little),
   IsSingleFloat(false), IsFP64bit(false), IsGP64bit(false), HasVFPU(false),
-  IsLinux(true), HasSEInReg(false), HasCondMov(false), HasMulDivAdd(false),
-  HasMinMax(false), HasSwap(false), HasBitCount(false), InMips16Mode(false),
-  HasDSP(false), HasDSPR2(false), IsAndroid(false)
+  IsLinux(true), HasSEInReg(false), HasCondMov(false), HasSwap(false),
+  HasBitCount(false), HasFPIdx(false),
+  InMips16Mode(false), InMicroMipsMode(false), HasDSP(false), HasDSPR2(false),
+  RM(_RM)
 {
   std::string CPUName = CPU;
   if (CPUName.empty())
diff --git a/lib/Target/Mips/MipsSubtarget.h b/lib/Target/Mips/MipsSubtarget.h
index ff69237ec2bd..7a2e47ce5a9d 100644
--- a/lib/Target/Mips/MipsSubtarget.h
+++ b/lib/Target/Mips/MipsSubtarget.h
@@ -14,8 +14,9 @@
 #ifndef MIPSSUBTARGET_H
 #define MIPSSUBTARGET_H
 
-#include "llvm/Target/TargetSubtargetInfo.h"
+#include "MCTargetDesc/MipsReginfo.h"
 #include "llvm/MC/MCInstrItineraries.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
 #include <string>
 
 #define GET_SUBTARGETINFO_HEADER
@@ -76,30 +77,32 @@ protected:
   // HasCondMov - Conditional mov (MOVZ, MOVN) instructions.
   bool HasCondMov;
 
-  // HasMulDivAdd - Multiply add and sub (MADD, MADDu, MSUB, MSUBu)
-  // instructions.
-  bool HasMulDivAdd;
-
-  // HasMinMax - MIN and MAX instructions.
-  bool HasMinMax;
-
   // HasSwap - Byte and half swap instructions.
   bool HasSwap;
 
   // HasBitCount - Count leading '1' and '0' bits.
   bool HasBitCount;
 
+  // HasFPIdx -- Floating point indexed load/store instructions.
+  bool HasFPIdx;
+
   // InMips16 -- can process Mips16 instructions
   bool InMips16Mode;
 
+  // InMicroMips -- can process MicroMips instructions
+  bool InMicroMipsMode;
+
   // HasDSP, HasDSPR2 -- supports DSP ASE.
   bool HasDSP, HasDSPR2;
 
-  // IsAndroid -- target is android
-  bool IsAndroid;
-
   InstrItineraryData InstrItins;
 
+  // The instance to the register info section object
+  MipsReginfo MRI;
+
+  // Relocation Model
+  Reloc::Model RM;
+
 public:
   virtual bool enablePostRAScheduler(CodeGenOpt::Level OptLevel,
                                      AntiDepBreakMode& Mode,
@@ -127,8 +130,6 @@ public:
   bool hasMips64() const { return MipsArchVersion >= Mips64; }
   bool hasMips64r2() const { return MipsArchVersion == Mips64r2; }
 
-  bool hasMips32r2Or64() const { return hasMips32r2() || hasMips64(); }
-
   bool isLittle() const { return IsLittle; }
   bool isFP64bit() const { return IsFP64bit; }
   bool isGP64bit() const { return IsGP64bit; }
@@ -137,9 +138,9 @@ public:
   bool isNotSingleFloat() const { return !IsSingleFloat; }
   bool hasVFPU() const { return HasVFPU; }
   bool inMips16Mode() const { return InMips16Mode; }
+  bool inMicroMipsMode() const { return InMicroMipsMode; }
   bool hasDSP() const { return HasDSP; }
   bool hasDSPR2() const { return HasDSPR2; }
-  bool isAndroid() const { return IsAndroid; }
   bool isLinux() const { return IsLinux; }
   bool useSmallSection() const { return UseSmallSection; }
 
@@ -148,10 +149,15 @@ public:
   /// Features related to the presence of specific instructions.
   bool hasSEInReg()   const { return HasSEInReg; }
   bool hasCondMov()   const { return HasCondMov; }
-  bool hasMulDivAdd() const { return HasMulDivAdd; }
-  bool hasMinMax()    const { return HasMinMax; }
   bool hasSwap()      const { return HasSwap; }
   bool hasBitCount()  const { return HasBitCount; }
+  bool hasFPIdx()     const { return HasFPIdx; }
+
+  // Grab MipsRegInfo object
+  const MipsReginfo &getMReginfo() const { return MRI; }
+
+  // Grab relocation model
+  Reloc::Model getRelocationModel() const {return RM;}
 };
 } // End llvm namespace
 
diff --git a/lib/Target/Mips/MipsTargetMachine.cpp b/lib/Target/Mips/MipsTargetMachine.cpp
index 983ee219412b..33363580aba7 100644
--- a/lib/Target/Mips/MipsTargetMachine.cpp
+++ b/lib/Target/Mips/MipsTargetMachine.cpp
@@ -15,8 +15,8 @@
 #include "Mips.h"
 #include "MipsFrameLowering.h"
 #include "MipsInstrInfo.h"
-#include "llvm/PassManager.h"
 #include "llvm/CodeGen/Passes.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
@@ -45,15 +45,16 @@ MipsTargetMachine(const Target &T, StringRef TT,
     Subtarget(TT, CPU, FS, isLittle, RM),
     DL(isLittle ?
                (Subtarget.isABI_N64() ?
-                "e-p:64:64:64-i8:8:32-i16:16:32-i64:64:64-f128:128:128-n32" :
-                "e-p:32:32:32-i8:8:32-i16:16:32-i64:64:64-n32") :
+                "e-p:64:64:64-i8:8:32-i16:16:32-i64:64:64-f128:128:128-"
+                "n32:64-S128" :
+                "e-p:32:32:32-i8:8:32-i16:16:32-i64:64:64-n32-S64") :
                (Subtarget.isABI_N64() ?
-                "E-p:64:64:64-i8:8:32-i16:16:32-i64:64:64-f128:128:128-n32" :
-                "E-p:32:32:32-i8:8:32-i16:16:32-i64:64:64-n32")),
+                "E-p:64:64:64-i8:8:32-i16:16:32-i64:64:64-f128:128:128-"
+                "n32:64-S128" :
+                "E-p:32:32:32-i8:8:32-i16:16:32-i64:64:64-n32-S64")),
     InstrInfo(MipsInstrInfo::create(*this)),
     FrameLowering(MipsFrameLowering::create(*this, Subtarget)),
-    TLInfo(*this), TSInfo(*this), JITInfo(),
-    STTI(&TLInfo), VTTI(&TLInfo) {
+    TLInfo(MipsTargetLowering::create(*this)), TSInfo(*this), JITInfo() {
 }
 
 void MipsebTargetMachine::anchor() { }
@@ -115,6 +116,8 @@ bool MipsPassConfig::addPreEmitPass() {
   // NOTE: long branch has not been implemented for mips16.
   if (TM.getSubtarget<MipsSubtarget>().hasStandardEncoding())
     addPass(createMipsLongBranchPass(TM));
+  if (TM.getSubtarget<MipsSubtarget>().inMips16Mode())
+    addPass(createMipsConstantIslandPass(TM));
 
   return true;
 }
diff --git a/lib/Target/Mips/MipsTargetMachine.h b/lib/Target/Mips/MipsTargetMachine.h
index b54f5cee6d4d..7e5f19226433 100644
--- a/lib/Target/Mips/MipsTargetMachine.h
+++ b/lib/Target/Mips/MipsTargetMachine.h
@@ -15,15 +15,15 @@
 #define MIPSTARGETMACHINE_H
 
 #include "MipsFrameLowering.h"
-#include "MipsInstrInfo.h"
 #include "MipsISelLowering.h"
+#include "MipsInstrInfo.h"
 #include "MipsJITInfo.h"
 #include "MipsSelectionDAGInfo.h"
 #include "MipsSubtarget.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/DataLayout.h"
+#include "llvm/ADT/OwningPtr.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetTransformImpl.h"
+#include "llvm/Target/TargetMachine.h"
 
 namespace llvm {
 class formatted_raw_ostream;
@@ -32,13 +32,11 @@ class MipsRegisterInfo;
 class MipsTargetMachine : public LLVMTargetMachine {
   MipsSubtarget       Subtarget;
   const DataLayout    DL; // Calculates type size & alignment
-  const MipsInstrInfo *InstrInfo;
-  const MipsFrameLowering *FrameLowering;
-  MipsTargetLowering  TLInfo;
+  OwningPtr<const MipsInstrInfo> InstrInfo;
+  OwningPtr<const MipsFrameLowering> FrameLowering;
+  OwningPtr<const MipsTargetLowering> TLInfo;
   MipsSelectionDAGInfo TSInfo;
   MipsJITInfo JITInfo;
-  ScalarTargetTransformImpl STTI;
-  VectorTargetTransformImpl VTTI;
 
 public:
   MipsTargetMachine(const Target &T, StringRef TT,
@@ -47,12 +45,12 @@ public:
                     CodeGenOpt::Level OL,
                     bool isLittle);
 
-  virtual ~MipsTargetMachine() { delete InstrInfo; }
+  virtual ~MipsTargetMachine() {}
 
   virtual const MipsInstrInfo *getInstrInfo() const
-  { return InstrInfo; }
+  { return InstrInfo.get(); }
   virtual const TargetFrameLowering *getFrameLowering() const
-  { return FrameLowering; }
+  { return FrameLowering.get(); }
   virtual const MipsSubtarget *getSubtargetImpl() const
   { return &Subtarget; }
   virtual const DataLayout *getDataLayout()    const
@@ -65,20 +63,13 @@ public:
   }
 
   virtual const MipsTargetLowering *getTargetLowering() const {
-    return &TLInfo;
+    return TLInfo.get();
   }
 
   virtual const MipsSelectionDAGInfo* getSelectionDAGInfo() const {
     return &TSInfo;
   }
 
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
-
   // Pass Pipeline Configuration
   virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
   virtual bool addCodeEmitter(PassManagerBase &PM, JITCodeEmitter &JCE);
diff --git a/lib/Target/Mips/MipsTargetObjectFile.cpp b/lib/Target/Mips/MipsTargetObjectFile.cpp
index 881908b82c91..4c748c5b57cd 100644
--- a/lib/Target/Mips/MipsTargetObjectFile.cpp
+++ b/lib/Target/Mips/MipsTargetObjectFile.cpp
@@ -9,14 +9,14 @@
 
 #include "MipsTargetObjectFile.h"
 #include "MipsSubtarget.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/GlobalVariable.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/GlobalVariable.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCSectionELF.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ELF.h"
+#include "llvm/Target/TargetMachine.h"
 using namespace llvm;
 
 static cl::opt<unsigned>
@@ -38,6 +38,20 @@ void MipsTargetObjectFile::Initialize(MCContext &Ctx, const TargetMachine &TM){
                                ELF::SHF_WRITE |ELF::SHF_ALLOC,
                                SectionKind::getBSS());
 
+  // Register info information
+  const MipsSubtarget &Subtarget = TM.getSubtarget<MipsSubtarget>();
+  if (Subtarget.isABI_N64() || Subtarget.isABI_N32())
+    ReginfoSection =
+      getContext().getELFSection(".MIPS.options",
+                                 ELF::SHT_MIPS_OPTIONS,
+                                 ELF::SHF_ALLOC |ELF::SHF_MIPS_NOSTRIP,
+                                 SectionKind::getMetadata());
+  else
+    ReginfoSection =
+      getContext().getELFSection(".reginfo",
+                                 ELF::SHT_MIPS_REGINFO,
+                                 ELF::SHF_ALLOC,
+                                 SectionKind::getMetadata());
 }
 
 // A address must be loaded from a small section if its size is less than the
diff --git a/lib/Target/Mips/MipsTargetObjectFile.h b/lib/Target/Mips/MipsTargetObjectFile.h
index c394a9dc02e4..c0e9140c829c 100644
--- a/lib/Target/Mips/MipsTargetObjectFile.h
+++ b/lib/Target/Mips/MipsTargetObjectFile.h
@@ -17,6 +17,7 @@ namespace llvm {
   class MipsTargetObjectFile : public TargetLoweringObjectFileELF {
     const MCSection *SmallDataSection;
     const MCSection *SmallBSSSection;
+    const MCSection *ReginfoSection;
   public:
 
     void Initialize(MCContext &Ctx, const TargetMachine &TM);
@@ -35,6 +36,7 @@ namespace llvm {
                                             const TargetMachine &TM) const;
 
     // TODO: Classify globals as mips wishes.
+    const MCSection *getReginfoSection() const { return ReginfoSection; }
   };
 } // end namespace llvm
 
diff --git a/lib/Target/Mips/TargetInfo/MipsTargetInfo.cpp b/lib/Target/Mips/TargetInfo/MipsTargetInfo.cpp
index 243632b20aac..3615c146a527 100644
--- a/lib/Target/Mips/TargetInfo/MipsTargetInfo.cpp
+++ b/lib/Target/Mips/TargetInfo/MipsTargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "Mips.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
diff --git a/lib/Target/NVPTX/CMakeLists.txt b/lib/Target/NVPTX/CMakeLists.txt
index 7cb16b4dd810..7da2fed4cd57 100644
--- a/lib/Target/NVPTX/CMakeLists.txt
+++ b/lib/Target/NVPTX/CMakeLists.txt
@@ -22,7 +22,7 @@ set(NVPTXCodeGen_sources
   NVPTXAllocaHoisting.cpp
   NVPTXAsmPrinter.cpp
   NVPTXUtilities.cpp
-  VectorElementize.cpp
+  NVVMReflect.cpp
   )
 
 add_llvm_target(NVPTXCodeGen ${NVPTXCodeGen_sources})
diff --git a/lib/Target/NVPTX/MCTargetDesc/NVPTXBaseInfo.h b/lib/Target/NVPTX/MCTargetDesc/NVPTXBaseInfo.h
index 454583850b71..b3e8b5d2622d 100644
--- a/lib/Target/NVPTX/MCTargetDesc/NVPTXBaseInfo.h
+++ b/lib/Target/NVPTX/MCTargetDesc/NVPTXBaseInfo.h
@@ -52,25 +52,24 @@ enum PropertyAnnotation {
 };
 
 const unsigned AnnotationNameLen = 8; // length of each annotation name
-const char
-PropertyAnnotationNames[PROPERTY_LAST + 1][AnnotationNameLen + 1] = {
-  "maxntidx",               // PROPERTY_MAXNTID_X
-  "maxntidy",               // PROPERTY_MAXNTID_Y
-  "maxntidz",               // PROPERTY_MAXNTID_Z
-  "reqntidx",               // PROPERTY_REQNTID_X
-  "reqntidy",               // PROPERTY_REQNTID_Y
-  "reqntidz",               // PROPERTY_REQNTID_Z
-  "minctasm",               // PROPERTY_MINNCTAPERSM
-  "texture",                // PROPERTY_ISTEXTURE
-  "surface",                // PROPERTY_ISSURFACE
-  "sampler",                // PROPERTY_ISSAMPLER
-  "rdoimage",               // PROPERTY_ISREADONLY_IMAGE_PARAM
-  "wroimage",               // PROPERTY_ISWRITEONLY_IMAGE_PARAM
-  "kernel",                 // PROPERTY_ISKERNEL_FUNCTION
-  "align",                  // PROPERTY_ALIGN
+const char PropertyAnnotationNames[PROPERTY_LAST + 1][AnnotationNameLen + 1] = {
+  "maxntidx",                         // PROPERTY_MAXNTID_X
+  "maxntidy",                         // PROPERTY_MAXNTID_Y
+  "maxntidz",                         // PROPERTY_MAXNTID_Z
+  "reqntidx",                         // PROPERTY_REQNTID_X
+  "reqntidy",                         // PROPERTY_REQNTID_Y
+  "reqntidz",                         // PROPERTY_REQNTID_Z
+  "minctasm",                         // PROPERTY_MINNCTAPERSM
+  "texture",                          // PROPERTY_ISTEXTURE
+  "surface",                          // PROPERTY_ISSURFACE
+  "sampler",                          // PROPERTY_ISSAMPLER
+  "rdoimage",                         // PROPERTY_ISREADONLY_IMAGE_PARAM
+  "wroimage",                         // PROPERTY_ISWRITEONLY_IMAGE_PARAM
+  "kernel",                           // PROPERTY_ISKERNEL_FUNCTION
+  "align",                            // PROPERTY_ALIGN
 
-  // last property
-  "proplast",               // PROPERTY_LAST
+              // last property
+  "proplast", // PROPERTY_LAST
 };
 
 // name of named metadata used for global annotations
@@ -80,9 +79,8 @@ PropertyAnnotationNames[PROPERTY_LAST + 1][AnnotationNameLen + 1] = {
 // compiling those .cpp files, hence __attribute__((unused)).
 __attribute__((unused))
 #endif
-static const char* NamedMDForAnnotations = "nvvm.annotations";
+    static const char *NamedMDForAnnotations = "nvvm.annotations";
 
 }
 
-
 #endif
diff --git a/lib/Target/NVPTX/MCTargetDesc/NVPTXMCAsmInfo.cpp b/lib/Target/NVPTX/MCTargetDesc/NVPTXMCAsmInfo.cpp
index 1d4166575da5..459cd96cb0cd 100644
--- a/lib/Target/NVPTX/MCTargetDesc/NVPTXMCAsmInfo.cpp
+++ b/lib/Target/NVPTX/MCTargetDesc/NVPTXMCAsmInfo.cpp
@@ -23,15 +23,15 @@ bool CompileForDebugging;
 // compile for debugging
 static cl::opt<bool, true>
 Debug("debug-compile", cl::desc("Compile for debugging"), cl::Hidden,
-      cl::location(CompileForDebugging),
-      cl::init(false));
+      cl::location(CompileForDebugging), cl::init(false));
 
-void NVPTXMCAsmInfo::anchor() { }
+void NVPTXMCAsmInfo::anchor() {}
 
 NVPTXMCAsmInfo::NVPTXMCAsmInfo(const Target &T, const StringRef &TT) {
   Triple TheTriple(TT);
-  if (TheTriple.getArch() == Triple::nvptx64)
-    PointerSize = 8;
+  if (TheTriple.getArch() == Triple::nvptx64) {
+    PointerSize = CalleeSaveStackSlotSize = 8;
+  }
 
   CommentString = "//";
 
@@ -54,7 +54,7 @@ NVPTXMCAsmInfo::NVPTXMCAsmInfo(const Target &T, const StringRef &TT) {
   Data32bitsDirective = " .b32 ";
   Data64bitsDirective = " .b64 ";
   PrivateGlobalPrefix = "";
-  ZeroDirective =  " .b8";
+  ZeroDirective = " .b8";
   AsciiDirective = " .b8";
   AscizDirective = " .b8";
 
diff --git a/lib/Target/NVPTX/MCTargetDesc/NVPTXMCTargetDesc.cpp b/lib/Target/NVPTX/MCTargetDesc/NVPTXMCTargetDesc.cpp
index 44aa01ca6e30..ccd29705df72 100644
--- a/lib/Target/NVPTX/MCTargetDesc/NVPTXMCTargetDesc.cpp
+++ b/lib/Target/NVPTX/MCTargetDesc/NVPTXMCTargetDesc.cpp
@@ -28,7 +28,6 @@
 #define GET_REGINFO_MC_DESC
 #include "NVPTXGenRegisterInfo.inc"
 
-
 using namespace llvm;
 
 static MCInstrInfo *createNVPTXMCInstrInfo() {
@@ -44,22 +43,20 @@ static MCRegisterInfo *createNVPTXMCRegisterInfo(StringRef TT) {
   return X;
 }
 
-static MCSubtargetInfo *createNVPTXMCSubtargetInfo(StringRef TT, StringRef CPU,
-                                                   StringRef FS) {
+static MCSubtargetInfo *
+createNVPTXMCSubtargetInfo(StringRef TT, StringRef CPU, StringRef FS) {
   MCSubtargetInfo *X = new MCSubtargetInfo();
   InitNVPTXMCSubtargetInfo(X, TT, CPU, FS);
   return X;
 }
 
-static MCCodeGenInfo *createNVPTXMCCodeGenInfo(StringRef TT, Reloc::Model RM,
-                                               CodeModel::Model CM,
-                                               CodeGenOpt::Level OL) {
+static MCCodeGenInfo *createNVPTXMCCodeGenInfo(
+    StringRef TT, Reloc::Model RM, CodeModel::Model CM, CodeGenOpt::Level OL) {
   MCCodeGenInfo *X = new MCCodeGenInfo();
   X->InitMCCodeGenInfo(RM, CM, OL);
   return X;
 }
 
-
 // Force static initialization.
 extern "C" void LLVMInitializeNVPTXTargetMC() {
   // Register the MC asm info.
diff --git a/lib/Target/NVPTX/ManagedStringPool.h b/lib/Target/NVPTX/ManagedStringPool.h
index b5684883fc95..d6c79b5110cc 100644
--- a/lib/Target/NVPTX/ManagedStringPool.h
+++ b/lib/Target/NVPTX/ManagedStringPool.h
@@ -12,7 +12,6 @@
 //
 //===----------------------------------------------------------------------===//
 
-
 #ifndef LLVM_SUPPORT_MANAGED_STRING_H
 #define LLVM_SUPPORT_MANAGED_STRING_H
 
diff --git a/lib/Target/NVPTX/NVPTX.h b/lib/Target/NVPTX/NVPTX.h
index a8d082a4d8b0..6a53a443bfb6 100644
--- a/lib/Target/NVPTX/NVPTX.h
+++ b/lib/Target/NVPTX/NVPTX.h
@@ -15,11 +15,11 @@
 #ifndef LLVM_TARGET_NVPTX_H
 #define LLVM_TARGET_NVPTX_H
 
-#include "llvm/Value.h"
-#include "llvm/Module.h"
+#include "MCTargetDesc/NVPTXBaseInfo.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Value.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Target/TargetMachine.h"
-#include "MCTargetDesc/NVPTXBaseInfo.h"
 #include <cassert>
 #include <iosfwd>
 
@@ -41,19 +41,24 @@ enum CondCodes {
 
 inline static const char *NVPTXCondCodeToString(NVPTXCC::CondCodes CC) {
   switch (CC) {
-  case NVPTXCC::NE:  return "ne";
-  case NVPTXCC::EQ:   return "eq";
-  case NVPTXCC::LT:   return "lt";
-  case NVPTXCC::LE:  return "le";
-  case NVPTXCC::GT:  return "gt";
-  case NVPTXCC::GE:   return "ge";
+  case NVPTXCC::NE:
+    return "ne";
+  case NVPTXCC::EQ:
+    return "eq";
+  case NVPTXCC::LT:
+    return "lt";
+  case NVPTXCC::LE:
+    return "le";
+  case NVPTXCC::GT:
+    return "gt";
+  case NVPTXCC::GE:
+    return "ge";
   }
   llvm_unreachable("Unknown condition code");
 }
 
-FunctionPass *createNVPTXISelDag(NVPTXTargetMachine &TM,
-                                 llvm::CodeGenOpt::Level OptLevel);
-FunctionPass *createVectorElementizePass(NVPTXTargetMachine &);
+FunctionPass *
+createNVPTXISelDag(NVPTXTargetMachine &TM, llvm::CodeGenOpt::Level OptLevel);
 FunctionPass *createLowerStructArgsPass(NVPTXTargetMachine &);
 FunctionPass *createNVPTXReMatPass(NVPTXTargetMachine &);
 FunctionPass *createNVPTXReMatBlockPass(NVPTXTargetMachine &);
@@ -63,8 +68,7 @@ bool isImageOrSamplerVal(const Value *, const Module *);
 extern Target TheNVPTXTarget32;
 extern Target TheNVPTXTarget64;
 
-namespace NVPTX
-{
+namespace NVPTX {
 enum DrvInterface {
   NVCL,
   CUDA,
@@ -103,7 +107,7 @@ enum LoadStore {
 };
 
 namespace PTXLdStInstCode {
-enum AddressSpace{
+enum AddressSpace {
   GENERIC = 0,
   GLOBAL = 1,
   CONSTANT = 2,
diff --git a/lib/Target/NVPTX/NVPTX.td b/lib/Target/NVPTX/NVPTX.td
index 7aee3595c625..d78b4e81a3e5 100644
--- a/lib/Target/NVPTX/NVPTX.td
+++ b/lib/Target/NVPTX/NVPTX.td
@@ -26,14 +26,6 @@ include "NVPTXInstrInfo.td"
 //===----------------------------------------------------------------------===//
 
 // SM Versions
-def SM10 : SubtargetFeature<"sm_10", "SmVersion", "10",
-                            "Target SM 1.0">;
-def SM11 : SubtargetFeature<"sm_11", "SmVersion", "11",
-                            "Target SM 1.1">;
-def SM12 : SubtargetFeature<"sm_12", "SmVersion", "12",
-                            "Target SM 1.2">;
-def SM13 : SubtargetFeature<"sm_13", "SmVersion", "13",
-                            "Target SM 1.3">;
 def SM20 : SubtargetFeature<"sm_20", "SmVersion", "20",
                             "Target SM 2.0">;
 def SM21 : SubtargetFeature<"sm_21", "SmVersion", "21",
@@ -56,10 +48,6 @@ def PTX31 : SubtargetFeature<"ptx31", "PTXVersion", "31",
 class Proc<string Name, list<SubtargetFeature> Features>
  : Processor<Name, NoItineraries, Features>;
 
-def : Proc<"sm_10", [SM10]>;
-def : Proc<"sm_11", [SM11]>;
-def : Proc<"sm_12", [SM12]>;
-def : Proc<"sm_13", [SM13]>;
 def : Proc<"sm_20", [SM20]>;
 def : Proc<"sm_21", [SM21]>;
 def : Proc<"sm_30", [SM30]>;
diff --git a/lib/Target/NVPTX/NVPTXAllocaHoisting.cpp b/lib/Target/NVPTX/NVPTXAllocaHoisting.cpp
index 668c39308f71..0f792ec6826e 100644
--- a/lib/Target/NVPTX/NVPTXAllocaHoisting.cpp
+++ b/lib/Target/NVPTX/NVPTXAllocaHoisting.cpp
@@ -11,17 +11,17 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/Function.h"
-#include "llvm/Instructions.h"
-#include "llvm/Constants.h"
 #include "NVPTXAllocaHoisting.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Instructions.h"
 
 namespace llvm {
 
 bool NVPTXAllocaHoisting::runOnFunction(Function &function) {
-  bool               functionModified    = false;
-  Function::iterator I                   = function.begin();
-  TerminatorInst    *firstTerminatorInst = (I++)->getTerminator();
+  bool functionModified = false;
+  Function::iterator I = function.begin();
+  TerminatorInst *firstTerminatorInst = (I++)->getTerminator();
 
   for (Function::iterator E = function.end(); I != E; ++I) {
     for (BasicBlock::iterator BI = I->begin(), BE = I->end(); BI != BE;) {
@@ -37,12 +37,10 @@ bool NVPTXAllocaHoisting::runOnFunction(Function &function) {
 }
 
 char NVPTXAllocaHoisting::ID = 1;
-RegisterPass<NVPTXAllocaHoisting> X("alloca-hoisting",
-                                    "Hoisting alloca instructions in non-entry "
-                                    "blocks to the entry block");
+RegisterPass<NVPTXAllocaHoisting>
+X("alloca-hoisting", "Hoisting alloca instructions in non-entry "
+                     "blocks to the entry block");
 
-FunctionPass *createAllocaHoisting() {
-  return new NVPTXAllocaHoisting();
-}
+FunctionPass *createAllocaHoisting() { return new NVPTXAllocaHoisting(); }
 
 } // end namespace llvm
diff --git a/lib/Target/NVPTX/NVPTXAllocaHoisting.h b/lib/Target/NVPTX/NVPTXAllocaHoisting.h
index c7cabf695311..19d73c5783cb 100644
--- a/lib/Target/NVPTX/NVPTXAllocaHoisting.h
+++ b/lib/Target/NVPTX/NVPTXAllocaHoisting.h
@@ -15,8 +15,8 @@
 #define NVPTX_ALLOCA_HOISTING_H_
 
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Pass.h"
-#include "llvm/DataLayout.h"
 
 namespace llvm {
 
diff --git a/lib/Target/NVPTX/NVPTXAsmPrinter.cpp b/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
index 0a885ce1c4a6..ce5d78afa332 100644
--- a/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
+++ b/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
@@ -13,40 +13,40 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTXAsmPrinter.h"
+#include "MCTargetDesc/NVPTXMCAsmInfo.h"
 #include "NVPTX.h"
 #include "NVPTXInstrInfo.h"
-#include "NVPTXTargetMachine.h"
+#include "NVPTXNumRegisters.h"
 #include "NVPTXRegisterInfo.h"
+#include "NVPTXTargetMachine.h"
 #include "NVPTXUtilities.h"
-#include "MCTargetDesc/NVPTXMCAsmInfo.h"
-#include "NVPTXNumRegisters.h"
+#include "cl_common_defines.h"
 #include "llvm/ADT/StringExtras.h"
-#include "llvm/DebugInfo.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/Module.h"
+#include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Assembly/Writer.h"
 #include "llvm/CodeGen/Analysis.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/DebugInfo.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Operator.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/Target/Mangler.h"
-#include "llvm/Target/TargetLoweringObjectFile.h"
-#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormattedStream.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Support/TimeValue.h"
-#include "llvm/Support/CommandLine.h"
-#include "llvm/Analysis/ConstantFolding.h"
 #include "llvm/Support/Path.h"
-#include "llvm/Assembly/Writer.h"
-#include "cl_common_defines.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/TimeValue.h"
+#include "llvm/Target/Mangler.h"
+#include "llvm/Target/TargetLoweringObjectFile.h"
 #include <sstream>
 using namespace llvm;
 
-
 #include "NVPTXGenAsmWriter.inc"
 
 bool RegAllocNilUsed = true;
@@ -58,21 +58,17 @@ EmitLineNumbers("nvptx-emit-line-numbers",
                 cl::desc("NVPTX Specific: Emit Line numbers even without -G"),
                 cl::init(true));
 
-namespace llvm  {
-bool InterleaveSrcInPtx = false;
-}
-
-static cl::opt<bool, true>InterleaveSrc("nvptx-emit-src",
-                                        cl::ZeroOrMore,
-                       cl::desc("NVPTX Specific: Emit source line in ptx file"),
-                                        cl::location(llvm::InterleaveSrcInPtx));
+namespace llvm { bool InterleaveSrcInPtx = false; }
 
+static cl::opt<bool, true>
+InterleaveSrc("nvptx-emit-src", cl::ZeroOrMore,
+              cl::desc("NVPTX Specific: Emit source line in ptx file"),
+              cl::location(llvm::InterleaveSrcInPtx));
 
 namespace {
 /// DiscoverDependentGlobals - Return a set of GlobalVariables on which \p V
 /// depends.
-void DiscoverDependentGlobals(Value *V,
-                              DenseSet<GlobalVariable*> &Globals) {
+void DiscoverDependentGlobals(Value *V, DenseSet<GlobalVariable *> &Globals) {
   if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
     Globals.insert(GV);
   else {
@@ -87,12 +83,12 @@ void DiscoverDependentGlobals(Value *V,
 /// VisitGlobalVariableForEmission - Add \p GV to the list of GlobalVariable
 /// instances to be emitted, but only after any dependents have been added
 /// first.
-void VisitGlobalVariableForEmission(GlobalVariable *GV,
-                                    SmallVectorImpl<GlobalVariable*> &Order,
-                                    DenseSet<GlobalVariable*> &Visited,
-                                    DenseSet<GlobalVariable*> &Visiting) {
+void VisitGlobalVariableForEmission(
+    GlobalVariable *GV, SmallVectorImpl<GlobalVariable *> &Order,
+    DenseSet<GlobalVariable *> &Visited, DenseSet<GlobalVariable *> &Visiting) {
   // Have we already visited this one?
-  if (Visited.count(GV)) return;
+  if (Visited.count(GV))
+    return;
 
   // Do we have a circular dependency?
   if (Visiting.count(GV))
@@ -102,12 +98,13 @@ void VisitGlobalVariableForEmission(GlobalVariable *GV,
   Visiting.insert(GV);
 
   // Make sure we visit all dependents first
-  DenseSet<GlobalVariable*> Others;
+  DenseSet<GlobalVariable *> Others;
   for (unsigned i = 0, e = GV->getNumOperands(); i != e; ++i)
     DiscoverDependentGlobals(GV->getOperand(i), Others);
-  
-  for (DenseSet<GlobalVariable*>::iterator I = Others.begin(),
-       E = Others.end(); I != E; ++I)
+
+  for (DenseSet<GlobalVariable *>::iterator I = Others.begin(),
+                                            E = Others.end();
+       I != E; ++I)
     VisitGlobalVariableForEmission(*I, Order, Visited, Visiting);
 
   // Now we can visit ourself
@@ -141,43 +138,35 @@ const MCExpr *nvptx::LowerConstant(const Constant *CV, AsmPrinter &AP) {
   if (CE == 0)
     llvm_unreachable("Unknown constant value to lower!");
 
-
   switch (CE->getOpcode()) {
   default:
     // If the code isn't optimized, there may be outstanding folding
     // opportunities. Attempt to fold the expression using DataLayout as a
     // last resort before giving up.
-    if (Constant *C =
-        ConstantFoldConstantExpression(CE, AP.TM.getDataLayout()))
+    if (Constant *C = ConstantFoldConstantExpression(CE, AP.TM.getDataLayout()))
       if (C != CE)
         return LowerConstant(C, AP);
 
     // Otherwise report the problem to the user.
     {
-        std::string S;
-        raw_string_ostream OS(S);
-        OS << "Unsupported expression in static initializer: ";
-        WriteAsOperand(OS, CE, /*PrintType=*/false,
-                       !AP.MF ? 0 : AP.MF->getFunction()->getParent());
-        report_fatal_error(OS.str());
+      std::string S;
+      raw_string_ostream OS(S);
+      OS << "Unsupported expression in static initializer: ";
+      WriteAsOperand(OS, CE, /*PrintType=*/ false,
+                     !AP.MF ? 0 : AP.MF->getFunction()->getParent());
+      report_fatal_error(OS.str());
     }
   case Instruction::GetElementPtr: {
     const DataLayout &TD = *AP.TM.getDataLayout();
     // Generate a symbolic expression for the byte address
-    const Constant *PtrVal = CE->getOperand(0);
-    SmallVector<Value*, 8> IdxVec(CE->op_begin()+1, CE->op_end());
-    int64_t Offset = TD.getIndexedOffset(PtrVal->getType(), IdxVec);
+    APInt OffsetAI(TD.getPointerSizeInBits(), 0);
+    cast<GEPOperator>(CE)->accumulateConstantOffset(TD, OffsetAI);
 
     const MCExpr *Base = LowerConstant(CE->getOperand(0), AP);
-    if (Offset == 0)
+    if (!OffsetAI)
       return Base;
 
-    // Truncate/sext the offset to the pointer size.
-    if (TD.getPointerSizeInBits() != 64) {
-      int SExtAmount = 64-TD.getPointerSizeInBits();
-      Offset = (Offset << SExtAmount) >> SExtAmount;
-    }
-
+    int64_t Offset = OffsetAI.getSExtValue();
     return MCBinaryExpr::CreateAdd(Base, MCConstantExpr::Create(Offset, Ctx),
                                    Ctx);
   }
@@ -187,7 +176,7 @@ const MCExpr *nvptx::LowerConstant(const Constant *CV, AsmPrinter &AP) {
     // expression properly.  This is important for differences between
     // blockaddress labels.  Since the two labels are in the same function, it
     // is reasonable to treat their delta as a 32-bit value.
-    // FALL THROUGH.
+  // FALL THROUGH.
   case Instruction::BitCast:
     return LowerConstant(CE->getOperand(0), AP);
 
@@ -197,7 +186,7 @@ const MCExpr *nvptx::LowerConstant(const Constant *CV, AsmPrinter &AP) {
     // integer type.  This promotes constant folding and simplifies this code.
     Constant *Op = CE->getOperand(0);
     Op = ConstantExpr::getIntegerCast(Op, TD.getIntPtrType(CV->getContext()),
-                                      false/*ZExt*/);
+                                      false /*ZExt*/);
     return LowerConstant(Op, AP);
   }
 
@@ -219,11 +208,12 @@ const MCExpr *nvptx::LowerConstant(const Constant *CV, AsmPrinter &AP) {
     // the high bits so we are sure to get a proper truncation if the input is
     // a constant expr.
     unsigned InBits = TD.getTypeAllocSizeInBits(Op->getType());
-    const MCExpr *MaskExpr = MCConstantExpr::Create(~0ULL >> (64-InBits), Ctx);
+    const MCExpr *MaskExpr =
+        MCConstantExpr::Create(~0ULL >> (64 - InBits), Ctx);
     return MCBinaryExpr::CreateAnd(OpExpr, MaskExpr, Ctx);
   }
 
-  // The MC library also has a right-shift operator, but it isn't consistently
+    // The MC library also has a right-shift operator, but it isn't consistently
   // signed or unsigned between different targets.
   case Instruction::Add:
   case Instruction::Sub:
@@ -237,24 +227,32 @@ const MCExpr *nvptx::LowerConstant(const Constant *CV, AsmPrinter &AP) {
     const MCExpr *LHS = LowerConstant(CE->getOperand(0), AP);
     const MCExpr *RHS = LowerConstant(CE->getOperand(1), AP);
     switch (CE->getOpcode()) {
-    default: llvm_unreachable("Unknown binary operator constant cast expr");
-    case Instruction::Add: return MCBinaryExpr::CreateAdd(LHS, RHS, Ctx);
-    case Instruction::Sub: return MCBinaryExpr::CreateSub(LHS, RHS, Ctx);
-    case Instruction::Mul: return MCBinaryExpr::CreateMul(LHS, RHS, Ctx);
-    case Instruction::SDiv: return MCBinaryExpr::CreateDiv(LHS, RHS, Ctx);
-    case Instruction::SRem: return MCBinaryExpr::CreateMod(LHS, RHS, Ctx);
-    case Instruction::Shl: return MCBinaryExpr::CreateShl(LHS, RHS, Ctx);
-    case Instruction::And: return MCBinaryExpr::CreateAnd(LHS, RHS, Ctx);
-    case Instruction::Or:  return MCBinaryExpr::CreateOr (LHS, RHS, Ctx);
-    case Instruction::Xor: return MCBinaryExpr::CreateXor(LHS, RHS, Ctx);
+    default:
+      llvm_unreachable("Unknown binary operator constant cast expr");
+    case Instruction::Add:
+      return MCBinaryExpr::CreateAdd(LHS, RHS, Ctx);
+    case Instruction::Sub:
+      return MCBinaryExpr::CreateSub(LHS, RHS, Ctx);
+    case Instruction::Mul:
+      return MCBinaryExpr::CreateMul(LHS, RHS, Ctx);
+    case Instruction::SDiv:
+      return MCBinaryExpr::CreateDiv(LHS, RHS, Ctx);
+    case Instruction::SRem:
+      return MCBinaryExpr::CreateMod(LHS, RHS, Ctx);
+    case Instruction::Shl:
+      return MCBinaryExpr::CreateShl(LHS, RHS, Ctx);
+    case Instruction::And:
+      return MCBinaryExpr::CreateAnd(LHS, RHS, Ctx);
+    case Instruction::Or:
+      return MCBinaryExpr::CreateOr(LHS, RHS, Ctx);
+    case Instruction::Xor:
+      return MCBinaryExpr::CreateXor(LHS, RHS, Ctx);
     }
   }
   }
 }
 
-
-void NVPTXAsmPrinter::emitLineNumberAsDotLoc(const MachineInstr &MI)
-{
+void NVPTXAsmPrinter::emitLineNumberAsDotLoc(const MachineInstr &MI) {
   if (!EmitLineNumbers)
     return;
   if (ignoreLoc(MI))
@@ -273,7 +271,6 @@ void NVPTXAsmPrinter::emitLineNumberAsDotLoc(const MachineInstr &MI)
   if (curLoc.isUnknown())
     return;
 
-
   const MachineFunction *MF = MI.getParent()->getParent();
   //const TargetMachine &TM = MF->getTarget();
 
@@ -294,14 +291,13 @@ void NVPTXAsmPrinter::emitLineNumberAsDotLoc(const MachineInstr &MI)
   if (filenameMap.find(fileName.str()) == filenameMap.end())
     return;
 
-
   // Emit the line from the source file.
   if (llvm::InterleaveSrcInPtx)
     this->emitSrcInText(fileName.str(), curLoc.getLine());
 
   std::stringstream temp;
-  temp << "\t.loc " << filenameMap[fileName.str()]
-       << " " << curLoc.getLine() << " " << curLoc.getCol();
+  temp << "\t.loc " << filenameMap[fileName.str()] << " " << curLoc.getLine()
+       << " " << curLoc.getCol();
   OutStreamer.EmitRawText(Twine(temp.str().c_str()));
 }
 
@@ -314,9 +310,7 @@ void NVPTXAsmPrinter::EmitInstruction(const MachineInstr *MI) {
   OutStreamer.EmitRawText(OS.str());
 }
 
-void NVPTXAsmPrinter::printReturnValStr(const Function *F,
-                                        raw_ostream &O)
-{
+void NVPTXAsmPrinter::printReturnValStr(const Function *F, raw_ostream &O) {
   const DataLayout *TD = TM.getDataLayout();
   const TargetLowering *TLI = TM.getTargetLowering();
 
@@ -334,53 +328,49 @@ void NVPTXAsmPrinter::printReturnValStr(const Function *F,
       unsigned size = 0;
       if (const IntegerType *ITy = dyn_cast<IntegerType>(Ty)) {
         size = ITy->getBitWidth();
-        if (size < 32) size = 32;
+        if (size < 32)
+          size = 32;
       } else {
-        assert(Ty->isFloatingPointTy() &&
-               "Floating point type expected here");
+        assert(Ty->isFloatingPointTy() && "Floating point type expected here");
         size = Ty->getPrimitiveSizeInBits();
       }
 
       O << ".param .b" << size << " func_retval0";
-    }
-    else if (isa<PointerType>(Ty)) {
+    } else if (isa<PointerType>(Ty)) {
       O << ".param .b" << TLI->getPointerTy().getSizeInBits()
-            << " func_retval0";
+        << " func_retval0";
     } else {
-      if ((Ty->getTypeID() == Type::StructTyID) ||
-          isa<VectorType>(Ty)) {
+      if ((Ty->getTypeID() == Type::StructTyID) || isa<VectorType>(Ty)) {
         SmallVector<EVT, 16> vtparts;
         ComputeValueVTs(*TLI, Ty, vtparts);
         unsigned totalsz = 0;
-        for (unsigned i=0,e=vtparts.size(); i!=e; ++i) {
+        for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
           unsigned elems = 1;
           EVT elemtype = vtparts[i];
           if (vtparts[i].isVector()) {
             elems = vtparts[i].getVectorNumElements();
             elemtype = vtparts[i].getVectorElementType();
           }
-          for (unsigned j=0, je=elems; j!=je; ++j) {
+          for (unsigned j = 0, je = elems; j != je; ++j) {
             unsigned sz = elemtype.getSizeInBits();
-            if (elemtype.isInteger() && (sz < 8)) sz = 8;
-            totalsz += sz/8;
+            if (elemtype.isInteger() && (sz < 8))
+              sz = 8;
+            totalsz += sz / 8;
           }
         }
         unsigned retAlignment = 0;
         if (!llvm::getAlign(*F, 0, retAlignment))
           retAlignment = TD->getABITypeAlignment(Ty);
-        O << ".param .align "
-            << retAlignment
-            << " .b8 func_retval0["
-            << totalsz << "]";
+        O << ".param .align " << retAlignment << " .b8 func_retval0[" << totalsz
+          << "]";
       } else
-        assert(false &&
-               "Unknown return type");
+        assert(false && "Unknown return type");
     }
   } else {
     SmallVector<EVT, 16> vtparts;
     ComputeValueVTs(*TLI, Ty, vtparts);
     unsigned idx = 0;
-    for (unsigned i=0,e=vtparts.size(); i!=e; ++i) {
+    for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
       unsigned elems = 1;
       EVT elemtype = vtparts[i];
       if (vtparts[i].isVector()) {
@@ -388,14 +378,16 @@ void NVPTXAsmPrinter::printReturnValStr(const Function *F,
         elemtype = vtparts[i].getVectorElementType();
       }
 
-      for (unsigned j=0, je=elems; j!=je; ++j) {
+      for (unsigned j = 0, je = elems; j != je; ++j) {
         unsigned sz = elemtype.getSizeInBits();
-        if (elemtype.isInteger() && (sz < 32)) sz = 32;
+        if (elemtype.isInteger() && (sz < 32))
+          sz = 32;
         O << ".reg .b" << sz << " func_retval" << idx;
-        if (j<je-1) O << ", ";
+        if (j < je - 1)
+          O << ", ";
         ++idx;
       }
-      if (i < e-1)
+      if (i < e - 1)
         O << ", ";
     }
   }
@@ -416,7 +408,7 @@ void NVPTXAsmPrinter::EmitFunctionEntryLabel() {
   // Set up
   MRI = &MF->getRegInfo();
   F = MF->getFunction();
-  emitLinkageDirective(F,O);
+  emitLinkageDirective(F, O);
   if (llvm::isKernelFunction(*F))
     O << ".entry ";
   else {
@@ -439,7 +431,7 @@ void NVPTXAsmPrinter::EmitFunctionEntryLabel() {
 void NVPTXAsmPrinter::EmitFunctionBodyStart() {
   const TargetRegisterInfo &TRI = *TM.getRegisterInfo();
   unsigned numRegClasses = TRI.getNumRegClasses();
-  VRidGlobal2LocalMap = new std::map<unsigned, unsigned>[numRegClasses+1];
+  VRidGlobal2LocalMap = new std::map<unsigned, unsigned>[numRegClasses + 1];
   OutStreamer.EmitRawText(StringRef("{\n"));
   setAndEmitFunctionVirtualRegisters(*MF);
 
@@ -451,54 +443,63 @@ void NVPTXAsmPrinter::EmitFunctionBodyStart() {
 
 void NVPTXAsmPrinter::EmitFunctionBodyEnd() {
   OutStreamer.EmitRawText(StringRef("}\n"));
-  delete []VRidGlobal2LocalMap;
+  delete[] VRidGlobal2LocalMap;
 }
 
-
-void
-NVPTXAsmPrinter::emitKernelFunctionDirectives(const Function& F,
-                                              raw_ostream &O) const {
+void NVPTXAsmPrinter::emitKernelFunctionDirectives(const Function &F,
+                                                   raw_ostream &O) const {
   // If the NVVM IR has some of reqntid* specified, then output
   // the reqntid directive, and set the unspecified ones to 1.
   // If none of reqntid* is specified, don't output reqntid directive.
   unsigned reqntidx, reqntidy, reqntidz;
   bool specified = false;
-  if (llvm::getReqNTIDx(F, reqntidx) == false) reqntidx = 1;
-  else specified = true;
-  if (llvm::getReqNTIDy(F, reqntidy) == false) reqntidy = 1;
-  else specified = true;
-  if (llvm::getReqNTIDz(F, reqntidz) == false) reqntidz = 1;
-  else specified = true;
+  if (llvm::getReqNTIDx(F, reqntidx) == false)
+    reqntidx = 1;
+  else
+    specified = true;
+  if (llvm::getReqNTIDy(F, reqntidy) == false)
+    reqntidy = 1;
+  else
+    specified = true;
+  if (llvm::getReqNTIDz(F, reqntidz) == false)
+    reqntidz = 1;
+  else
+    specified = true;
 
   if (specified)
-    O << ".reqntid " << reqntidx << ", "
-    << reqntidy << ", " << reqntidz << "\n";
+    O << ".reqntid " << reqntidx << ", " << reqntidy << ", " << reqntidz
+      << "\n";
 
   // If the NVVM IR has some of maxntid* specified, then output
   // the maxntid directive, and set the unspecified ones to 1.
   // If none of maxntid* is specified, don't output maxntid directive.
   unsigned maxntidx, maxntidy, maxntidz;
   specified = false;
-  if (llvm::getMaxNTIDx(F, maxntidx) == false) maxntidx = 1;
-  else specified = true;
-  if (llvm::getMaxNTIDy(F, maxntidy) == false) maxntidy = 1;
-  else specified = true;
-  if (llvm::getMaxNTIDz(F, maxntidz) == false) maxntidz = 1;
-  else specified = true;
+  if (llvm::getMaxNTIDx(F, maxntidx) == false)
+    maxntidx = 1;
+  else
+    specified = true;
+  if (llvm::getMaxNTIDy(F, maxntidy) == false)
+    maxntidy = 1;
+  else
+    specified = true;
+  if (llvm::getMaxNTIDz(F, maxntidz) == false)
+    maxntidz = 1;
+  else
+    specified = true;
 
   if (specified)
-    O << ".maxntid " << maxntidx << ", "
-    << maxntidy << ", " << maxntidz << "\n";
+    O << ".maxntid " << maxntidx << ", " << maxntidy << ", " << maxntidz
+      << "\n";
 
   unsigned mincta;
   if (llvm::getMinCTASm(F, mincta))
     O << ".minnctapersm " << mincta << "\n";
 }
 
-void
-NVPTXAsmPrinter::getVirtualRegisterName(unsigned vr, bool isVec,
-                                        raw_ostream &O) {
-  const TargetRegisterClass * RC = MRI->getRegClass(vr);
+void NVPTXAsmPrinter::getVirtualRegisterName(unsigned vr, bool isVec,
+                                             raw_ostream &O) {
+  const TargetRegisterClass *RC = MRI->getRegClass(vr);
   unsigned id = RC->getID();
 
   std::map<unsigned, unsigned> &regmap = VRidGlobal2LocalMap[id];
@@ -508,61 +509,41 @@ NVPTXAsmPrinter::getVirtualRegisterName(unsigned vr, bool isVec,
     O << getNVPTXRegClassStr(RC) << mapped_vr;
     return;
   }
-  // Vector virtual register
-  if (getNVPTXVectorSize(RC) == 4)
-    O << "{"
-    << getNVPTXRegClassStr(RC) << mapped_vr << "_0, "
-    << getNVPTXRegClassStr(RC) << mapped_vr << "_1, "
-    << getNVPTXRegClassStr(RC) << mapped_vr << "_2, "
-    << getNVPTXRegClassStr(RC) << mapped_vr << "_3"
-    << "}";
-  else if (getNVPTXVectorSize(RC) == 2)
-    O << "{"
-    << getNVPTXRegClassStr(RC) << mapped_vr << "_0, "
-    << getNVPTXRegClassStr(RC) << mapped_vr << "_1"
-    << "}";
-  else
-    llvm_unreachable("Unsupported vector size");
+  report_fatal_error("Bad register!");
 }
 
-void
-NVPTXAsmPrinter::emitVirtualRegister(unsigned int vr, bool isVec,
-                                     raw_ostream &O) {
+void NVPTXAsmPrinter::emitVirtualRegister(unsigned int vr, bool isVec,
+                                          raw_ostream &O) {
   getVirtualRegisterName(vr, isVec, O);
 }
 
-void NVPTXAsmPrinter::printVecModifiedImmediate(const MachineOperand &MO,
-                                                const char *Modifier,
-                                                raw_ostream &O) {
-  static const char vecelem[] = {'0', '1', '2', '3', '0', '1', '2', '3'};
-  int Imm = (int)MO.getImm();
-  if(0 == strcmp(Modifier, "vecelem"))
+void NVPTXAsmPrinter::printVecModifiedImmediate(
+    const MachineOperand &MO, const char *Modifier, raw_ostream &O) {
+  static const char vecelem[] = { '0', '1', '2', '3', '0', '1', '2', '3' };
+  int Imm = (int) MO.getImm();
+  if (0 == strcmp(Modifier, "vecelem"))
     O << "_" << vecelem[Imm];
-  else if(0 == strcmp(Modifier, "vecv4comm1")) {
-    if((Imm < 0) || (Imm > 3))
+  else if (0 == strcmp(Modifier, "vecv4comm1")) {
+    if ((Imm < 0) || (Imm > 3))
       O << "//";
-  }
-  else if(0 == strcmp(Modifier, "vecv4comm2")) {
-    if((Imm < 4) || (Imm > 7))
+  } else if (0 == strcmp(Modifier, "vecv4comm2")) {
+    if ((Imm < 4) || (Imm > 7))
       O << "//";
-  }
-  else if(0 == strcmp(Modifier, "vecv4pos")) {
-    if(Imm < 0) Imm = 0;
-    O << "_" << vecelem[Imm%4];
-  }
-  else if(0 == strcmp(Modifier, "vecv2comm1")) {
-    if((Imm < 0) || (Imm > 1))
+  } else if (0 == strcmp(Modifier, "vecv4pos")) {
+    if (Imm < 0)
+      Imm = 0;
+    O << "_" << vecelem[Imm % 4];
+  } else if (0 == strcmp(Modifier, "vecv2comm1")) {
+    if ((Imm < 0) || (Imm > 1))
       O << "//";
-  }
-  else if(0 == strcmp(Modifier, "vecv2comm2")) {
-    if((Imm < 2) || (Imm > 3))
+  } else if (0 == strcmp(Modifier, "vecv2comm2")) {
+    if ((Imm < 2) || (Imm > 3))
       O << "//";
-  }
-  else if(0 == strcmp(Modifier, "vecv2pos")) {
-    if(Imm < 0) Imm = 0;
-    O << "_" << vecelem[Imm%2];
-  }
-  else
+  } else if (0 == strcmp(Modifier, "vecv2pos")) {
+    if (Imm < 0)
+      Imm = 0;
+    O << "_" << vecelem[Imm % 2];
+  } else
     llvm_unreachable("Unknown Modifier on immediate operand");
 }
 
@@ -584,7 +565,7 @@ void NVPTXAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
           emitVirtualRegister(MO.getReg(), true, O);
         else
           llvm_unreachable(
-                 "Don't know how to handle the modifier on virtual register.");
+              "Don't know how to handle the modifier on virtual register.");
       }
     }
     return;
@@ -595,7 +576,8 @@ void NVPTXAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
     else if (strstr(Modifier, "vec") == Modifier)
       printVecModifiedImmediate(MO, Modifier, O);
     else
-      llvm_unreachable("Don't know how to handle modifier on immediate operand");
+      llvm_unreachable(
+          "Don't know how to handle modifier on immediate operand");
     return;
 
   case MachineOperand::MO_FPImmediate:
@@ -607,18 +589,16 @@ void NVPTXAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
     break;
 
   case MachineOperand::MO_ExternalSymbol: {
-    const char * symbname = MO.getSymbolName();
+    const char *symbname = MO.getSymbolName();
     if (strstr(symbname, ".PARAM") == symbname) {
       unsigned index;
-      sscanf(symbname+6, "%u[];", &index);
+      sscanf(symbname + 6, "%u[];", &index);
       printParamName(index, O);
-    }
-    else if (strstr(symbname, ".HLPPARAM") == symbname) {
+    } else if (strstr(symbname, ".HLPPARAM") == symbname) {
       unsigned index;
-      sscanf(symbname+9, "%u[];", &index);
+      sscanf(symbname + 9, "%u[];", &index);
       O << *CurrentFnSym << "_param_" << index << "_offset";
-    }
-    else
+    } else
       O << symbname;
     break;
   }
@@ -632,8 +612,8 @@ void NVPTXAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
   }
 }
 
-void NVPTXAsmPrinter::
-printImplicitDef(const MachineInstr *MI, raw_ostream &O) const {
+void NVPTXAsmPrinter::printImplicitDef(const MachineInstr *MI,
+                                       raw_ostream &O) const {
 #ifndef __OPTIMIZE__
   O << "\t// Implicit def :";
   //printOperand(MI, 0);
@@ -647,64 +627,69 @@ void NVPTXAsmPrinter::printMemOperand(const MachineInstr *MI, int opNum,
 
   if (Modifier && !strcmp(Modifier, "add")) {
     O << ", ";
-    printOperand(MI, opNum+1, O);
+    printOperand(MI, opNum + 1, O);
   } else {
-    if (MI->getOperand(opNum+1).isImm() &&
-        MI->getOperand(opNum+1).getImm() == 0)
+    if (MI->getOperand(opNum + 1).isImm() &&
+        MI->getOperand(opNum + 1).getImm() == 0)
       return; // don't print ',0' or '+0'
     O << "+";
-    printOperand(MI, opNum+1, O);
+    printOperand(MI, opNum + 1, O);
   }
 }
 
 void NVPTXAsmPrinter::printLdStCode(const MachineInstr *MI, int opNum,
-                                    raw_ostream &O, const char *Modifier)
-{
+                                    raw_ostream &O, const char *Modifier) {
   if (Modifier) {
     const MachineOperand &MO = MI->getOperand(opNum);
-    int Imm = (int)MO.getImm();
+    int Imm = (int) MO.getImm();
     if (!strcmp(Modifier, "volatile")) {
       if (Imm)
         O << ".volatile";
     } else if (!strcmp(Modifier, "addsp")) {
       switch (Imm) {
-      case NVPTX::PTXLdStInstCode::GLOBAL: O << ".global"; break;
-      case NVPTX::PTXLdStInstCode::SHARED: O << ".shared"; break;
-      case NVPTX::PTXLdStInstCode::LOCAL: O << ".local"; break;
-      case NVPTX::PTXLdStInstCode::PARAM: O << ".param"; break;
-      case NVPTX::PTXLdStInstCode::CONSTANT: O << ".const"; break;
+      case NVPTX::PTXLdStInstCode::GLOBAL:
+        O << ".global";
+        break;
+      case NVPTX::PTXLdStInstCode::SHARED:
+        O << ".shared";
+        break;
+      case NVPTX::PTXLdStInstCode::LOCAL:
+        O << ".local";
+        break;
+      case NVPTX::PTXLdStInstCode::PARAM:
+        O << ".param";
+        break;
+      case NVPTX::PTXLdStInstCode::CONSTANT:
+        O << ".const";
+        break;
       case NVPTX::PTXLdStInstCode::GENERIC:
         if (!nvptxSubtarget.hasGenericLdSt())
           O << ".global";
         break;
       default:
-        assert("wrong value");
+        llvm_unreachable("Wrong Address Space");
       }
-    }
-    else if (!strcmp(Modifier, "sign")) {
-      if (Imm==NVPTX::PTXLdStInstCode::Signed)
+    } else if (!strcmp(Modifier, "sign")) {
+      if (Imm == NVPTX::PTXLdStInstCode::Signed)
         O << "s";
-      else if (Imm==NVPTX::PTXLdStInstCode::Unsigned)
+      else if (Imm == NVPTX::PTXLdStInstCode::Unsigned)
         O << "u";
       else
         O << "f";
-    }
-    else if (!strcmp(Modifier, "vec")) {
-      if (Imm==NVPTX::PTXLdStInstCode::V2)
+    } else if (!strcmp(Modifier, "vec")) {
+      if (Imm == NVPTX::PTXLdStInstCode::V2)
         O << ".v2";
-      else if (Imm==NVPTX::PTXLdStInstCode::V4)
+      else if (Imm == NVPTX::PTXLdStInstCode::V4)
         O << ".v4";
-    }
-    else
-      assert("unknown modifier");
-  }
-  else
-    assert("unknown modifier");
+    } else
+      llvm_unreachable("Unknown Modifier");
+  } else
+    llvm_unreachable("Empty Modifier");
 }
 
-void NVPTXAsmPrinter::emitDeclaration (const Function *F, raw_ostream &O) {
+void NVPTXAsmPrinter::emitDeclaration(const Function *F, raw_ostream &O) {
 
-  emitLinkageDirective(F,O);
+  emitLinkageDirective(F, O);
   if (llvm::isKernelFunction(*F))
     O << ".entry ";
   else
@@ -715,8 +700,7 @@ void NVPTXAsmPrinter::emitDeclaration (const Function *F, raw_ostream &O) {
   O << ";\n";
 }
 
-static bool usedInGlobalVarDef(const Constant *C)
-{
+static bool usedInGlobalVarDef(const Constant *C) {
   if (!C)
     return false;
 
@@ -726,8 +710,8 @@ static bool usedInGlobalVarDef(const Constant *C)
     return true;
   }
 
-  for (Value::const_use_iterator ui=C->use_begin(), ue=C->use_end();
-      ui!=ue; ++ui) {
+  for (Value::const_use_iterator ui = C->use_begin(), ue = C->use_end();
+       ui != ue; ++ui) {
     const Constant *C = dyn_cast<Constant>(*ui);
     if (usedInGlobalVarDef(C))
       return true;
@@ -735,8 +719,7 @@ static bool usedInGlobalVarDef(const Constant *C)
   return false;
 }
 
-static bool usedInOneFunc(const User *U, Function const *&oneFunc)
-{
+static bool usedInOneFunc(const User *U, Function const *&oneFunc) {
   if (const GlobalVariable *othergv = dyn_cast<GlobalVariable>(U)) {
     if (othergv->getName().str() == "llvm.used")
       return true;
@@ -749,19 +732,17 @@ static bool usedInOneFunc(const User *U, Function const *&oneFunc)
         return false;
       oneFunc = curFunc;
       return true;
-    }
-    else
+    } else
       return false;
   }
 
   if (const MDNode *md = dyn_cast<MDNode>(U))
     if (md->hasName() && ((md->getName().str() == "llvm.dbg.gv") ||
-        (md->getName().str() == "llvm.dbg.sp")))
+                          (md->getName().str() == "llvm.dbg.sp")))
       return true;
 
-
-  for (User::const_use_iterator ui=U->use_begin(), ue=U->use_end();
-      ui!=ue; ++ui) {
+  for (User::const_use_iterator ui = U->use_begin(), ue = U->use_end();
+       ui != ue; ++ui) {
     if (usedInOneFunc(*ui, oneFunc) == false)
       return false;
   }
@@ -795,16 +776,18 @@ static bool canDemoteGlobalVar(const GlobalVariable *gv, Function const *&f) {
 
 static bool useFuncSeen(const Constant *C,
                         llvm::DenseMap<const Function *, bool> &seenMap) {
-  for (Value::const_use_iterator ui=C->use_begin(), ue=C->use_end();
-      ui!=ue; ++ui) {
+  for (Value::const_use_iterator ui = C->use_begin(), ue = C->use_end();
+       ui != ue; ++ui) {
     if (const Constant *cu = dyn_cast<Constant>(*ui)) {
       if (useFuncSeen(cu, seenMap))
         return true;
     } else if (const Instruction *I = dyn_cast<Instruction>(*ui)) {
       const BasicBlock *bb = I->getParent();
-      if (!bb) continue;
+      if (!bb)
+        continue;
       const Function *caller = bb->getParent();
-      if (!caller) continue;
+      if (!caller)
+        continue;
       if (seenMap.find(caller) != seenMap.end())
         return true;
     }
@@ -812,10 +795,9 @@ static bool useFuncSeen(const Constant *C,
   return false;
 }
 
-void NVPTXAsmPrinter::emitDeclarations (Module &M, raw_ostream &O) {
+void NVPTXAsmPrinter::emitDeclarations(Module &M, raw_ostream &O) {
   llvm::DenseMap<const Function *, bool> seenMap;
-  for (Module::const_iterator FI=M.begin(), FE=M.end();
-      FI!=FE; ++FI) {
+  for (Module::const_iterator FI = M.begin(), FE = M.end(); FI != FE; ++FI) {
     const Function *F = FI;
 
     if (F->isDeclaration()) {
@@ -827,8 +809,9 @@ void NVPTXAsmPrinter::emitDeclarations (Module &M, raw_ostream &O) {
       emitDeclaration(F, O);
       continue;
     }
-    for (Value::const_use_iterator iter=F->use_begin(),
-        iterEnd=F->use_end(); iter!=iterEnd; ++iter) {
+    for (Value::const_use_iterator iter = F->use_begin(),
+                                   iterEnd = F->use_end();
+         iter != iterEnd; ++iter) {
       if (const Constant *C = dyn_cast<Constant>(*iter)) {
         if (usedInGlobalVarDef(C)) {
           // The use is in the initialization of a global variable
@@ -847,12 +830,15 @@ void NVPTXAsmPrinter::emitDeclarations (Module &M, raw_ostream &O) {
         }
       }
 
-      if (!isa<Instruction>(*iter)) continue;
+      if (!isa<Instruction>(*iter))
+        continue;
       const Instruction *instr = cast<Instruction>(*iter);
       const BasicBlock *bb = instr->getParent();
-      if (!bb) continue;
+      if (!bb)
+        continue;
       const Function *caller = bb->getParent();
-      if (!caller) continue;
+      if (!caller)
+        continue;
 
       // If a caller has already been seen, then the caller is
       // appearing in the module before the callee. so print out
@@ -871,9 +857,10 @@ void NVPTXAsmPrinter::recordAndEmitFilenames(Module &M) {
   DebugInfoFinder DbgFinder;
   DbgFinder.processModule(M);
 
-  unsigned i=1;
+  unsigned i = 1;
   for (DebugInfoFinder::iterator I = DbgFinder.compile_unit_begin(),
-      E = DbgFinder.compile_unit_end(); I != E; ++I) {
+                                 E = DbgFinder.compile_unit_end();
+       I != E; ++I) {
     DICompileUnit DIUnit(*I);
     StringRef Filename(DIUnit.getFilename());
     StringRef Dirname(DIUnit.getDirectory());
@@ -890,7 +877,8 @@ void NVPTXAsmPrinter::recordAndEmitFilenames(Module &M) {
   }
 
   for (DebugInfoFinder::iterator I = DbgFinder.subprogram_begin(),
-      E = DbgFinder.subprogram_end(); I != E; ++I) {
+                                 E = DbgFinder.subprogram_end();
+       I != E; ++I) {
     DISubprogram SP(*I);
     StringRef Filename(SP.getFilename());
     StringRef Dirname(SP.getDirectory());
@@ -906,7 +894,7 @@ void NVPTXAsmPrinter::recordAndEmitFilenames(Module &M) {
   }
 }
 
-bool NVPTXAsmPrinter::doInitialization (Module &M) {
+bool NVPTXAsmPrinter::doInitialization(Module &M) {
 
   SmallString<128> Str1;
   raw_svector_ostream OS1(Str1);
@@ -918,8 +906,8 @@ bool NVPTXAsmPrinter::doInitialization (Module &M) {
   //bool Result = AsmPrinter::doInitialization(M);
 
   // Initialize TargetLoweringObjectFile.
-  const_cast<TargetLoweringObjectFile&>(getObjFileLowering())
-          .Initialize(OutContext, TM);
+  const_cast<TargetLoweringObjectFile &>(getObjFileLowering())
+      .Initialize(OutContext, TM);
 
   Mang = new Mangler(OutContext, *TM.getDataLayout());
 
@@ -927,11 +915,9 @@ bool NVPTXAsmPrinter::doInitialization (Module &M) {
   emitHeader(M, OS1);
   OutStreamer.EmitRawText(OS1.str());
 
-
   // Already commented out
   //bool Result = AsmPrinter::doInitialization(M);
 
-
   if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA)
     recordAndEmitFilenames(M);
 
@@ -945,16 +931,16 @@ bool NVPTXAsmPrinter::doInitialization (Module &M) {
   // global variable in order, and ensure that we emit it *after* its dependent
   // globals. We use a little extra memory maintaining both a set and a list to
   // have fast searches while maintaining a strict ordering.
-  SmallVector<GlobalVariable*,8> Globals;
-  DenseSet<GlobalVariable*> GVVisited;
-  DenseSet<GlobalVariable*> GVVisiting;
+  SmallVector<GlobalVariable *, 8> Globals;
+  DenseSet<GlobalVariable *> GVVisited;
+  DenseSet<GlobalVariable *> GVVisiting;
 
   // Visit each global variable, in order
-  for (Module::global_iterator I = M.global_begin(), E = M.global_end();
-       I != E; ++I)
+  for (Module::global_iterator I = M.global_begin(), E = M.global_end(); I != E;
+       ++I)
     VisitGlobalVariableForEmission(I, Globals, GVVisited, GVVisiting);
 
-  assert(GVVisited.size() == M.getGlobalList().size() && 
+  assert(GVVisited.size() == M.getGlobalList().size() &&
          "Missed a global variable");
   assert(GVVisiting.size() == 0 && "Did not fully process a global variable");
 
@@ -965,10 +951,10 @@ bool NVPTXAsmPrinter::doInitialization (Module &M) {
   OS2 << '\n';
 
   OutStreamer.EmitRawText(OS2.str());
-  return false;  // success
+  return false; // success
 }
 
-void NVPTXAsmPrinter::emitHeader (Module &M, raw_ostream &O) {
+void NVPTXAsmPrinter::emitHeader(Module &M, raw_ostream &O) {
   O << "//\n";
   O << "// Generated by LLVM NVPTX Back-End\n";
   O << "//\n";
@@ -1008,12 +994,12 @@ bool NVPTXAsmPrinter::doFinalization(Module &M) {
 
   Module::GlobalListType &global_list = M.getGlobalList();
   int i, n = global_list.size();
-  GlobalVariable **gv_array = new GlobalVariable* [n];
+  GlobalVariable **gv_array = new GlobalVariable *[n];
 
   // first, back-up GlobalVariable in gv_array
   i = 0;
   for (Module::global_iterator I = global_list.begin(), E = global_list.end();
-      I != E; ++I)
+       I != E; ++I)
     gv_array[i++] = &*I;
 
   // second, empty global_list
@@ -1024,13 +1010,12 @@ bool NVPTXAsmPrinter::doFinalization(Module &M) {
   bool ret = AsmPrinter::doFinalization(M);
 
   // now we restore global variables
-  for (i = 0; i < n; i ++)
+  for (i = 0; i < n; i++)
     global_list.insert(global_list.end(), gv_array[i]);
 
   delete[] gv_array;
   return ret;
 
-
   //bool Result = AsmPrinter::doFinalization(M);
   // Instead of calling the parents doFinalization, we may
   // clone parents doFinalization and customize here.
@@ -1050,8 +1035,8 @@ bool NVPTXAsmPrinter::doFinalization(Module &M) {
 // external without init                  -> .extern
 // appending                              -> not allowed, assert.
 
-void NVPTXAsmPrinter::emitLinkageDirective(const GlobalValue* V, raw_ostream &O)
-{
+void NVPTXAsmPrinter::emitLinkageDirective(const GlobalValue *V,
+                                           raw_ostream &O) {
   if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA) {
     if (V->hasExternalLinkage()) {
       if (isa<GlobalVariable>(V)) {
@@ -1078,8 +1063,7 @@ void NVPTXAsmPrinter::emitLinkageDirective(const GlobalValue* V, raw_ostream &O)
   }
 }
 
-
-void NVPTXAsmPrinter::printModuleLevelGV(GlobalVariable* GVar, raw_ostream &O,
+void NVPTXAsmPrinter::printModuleLevelGV(GlobalVariable *GVar, raw_ostream &O,
                                          bool processDemoted) {
 
   // Skip meta data
@@ -1130,30 +1114,48 @@ void NVPTXAsmPrinter::printModuleLevelGV(GlobalVariable* GVar, raw_ostream &O,
     if (Initializer)
       CI = dyn_cast<ConstantInt>(Initializer);
     if (CI) {
-      unsigned sample=CI->getZExtValue();
+      unsigned sample = CI->getZExtValue();
 
       O << " = { ";
 
-      for (int i =0, addr=((sample & __CLK_ADDRESS_MASK ) >>
-          __CLK_ADDRESS_BASE) ; i < 3 ; i++) {
+      for (int i = 0,
+               addr = ((sample & __CLK_ADDRESS_MASK) >> __CLK_ADDRESS_BASE);
+           i < 3; i++) {
         O << "addr_mode_" << i << " = ";
         switch (addr) {
-        case 0: O << "wrap"; break;
-        case 1: O << "clamp_to_border"; break;
-        case 2: O << "clamp_to_edge"; break;
-        case 3: O << "wrap"; break;
-        case 4: O << "mirror"; break;
+        case 0:
+          O << "wrap";
+          break;
+        case 1:
+          O << "clamp_to_border";
+          break;
+        case 2:
+          O << "clamp_to_edge";
+          break;
+        case 3:
+          O << "wrap";
+          break;
+        case 4:
+          O << "mirror";
+          break;
         }
-        O <<", ";
+        O << ", ";
       }
       O << "filter_mode = ";
-      switch (( sample & __CLK_FILTER_MASK ) >> __CLK_FILTER_BASE ) {
-      case 0: O << "nearest"; break;
-      case 1: O << "linear";  break;
-      case 2: assert ( 0 && "Anisotropic filtering is not supported");
-      default: O << "nearest"; break;
+      switch ((sample & __CLK_FILTER_MASK) >> __CLK_FILTER_BASE) {
+      case 0:
+        O << "nearest";
+        break;
+      case 1:
+        O << "linear";
+        break;
+      case 2:
+        assert(0 && "Anisotropic filtering is not supported");
+      default:
+        O << "nearest";
+        break;
       }
-      if (!(( sample &__CLK_NORMALIZED_MASK ) >> __CLK_NORMALIZED_BASE)) {
+      if (!((sample & __CLK_NORMALIZED_MASK) >> __CLK_NORMALIZED_BASE)) {
         O << ", force_unnormalized_coords = 1";
       }
       O << " }";
@@ -1195,7 +1197,6 @@ void NVPTXAsmPrinter::printModuleLevelGV(GlobalVariable* GVar, raw_ostream &O,
   else
     O << " .align " << GVar->getAlignment();
 
-
   if (ETy->isPrimitiveType() || ETy->isIntegerTy() || isa<PointerType>(ETy)) {
     O << " .";
     O << getPTXFundamentalTypeStr(ETy, false);
@@ -1205,17 +1206,17 @@ void NVPTXAsmPrinter::printModuleLevelGV(GlobalVariable* GVar, raw_ostream &O,
     // Ptx allows variable initilization only for constant and global state
     // spaces.
     if (((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
-        (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST_NOT_GEN) ||
-        (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST))
-        && GVar->hasInitializer()) {
+         (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST_NOT_GEN) ||
+         (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) &&
+        GVar->hasInitializer()) {
       Constant *Initializer = GVar->getInitializer();
       if (!Initializer->isNullValue()) {
-        O << " = " ;
+        O << " = ";
         printScalarConstant(Initializer, O);
       }
     }
   } else {
-    unsigned int ElementSize =0;
+    unsigned int ElementSize = 0;
 
     // Although PTX has direct support for struct type and array type and
     // LLVM IR is very similar to PTX, the LLVM CodeGen does not support for
@@ -1229,54 +1230,49 @@ void NVPTXAsmPrinter::printModuleLevelGV(GlobalVariable* GVar, raw_ostream &O,
       // Ptx allows variable initilization only for constant and
       // global state spaces.
       if (((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
-          (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST_NOT_GEN) ||
-          (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST))
-          && GVar->hasInitializer()) {
+           (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST_NOT_GEN) ||
+           (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) &&
+          GVar->hasInitializer()) {
         Constant *Initializer = GVar->getInitializer();
-        if (!isa<UndefValue>(Initializer) &&
-            !Initializer->isNullValue()) {
+        if (!isa<UndefValue>(Initializer) && !Initializer->isNullValue()) {
           AggBuffer aggBuffer(ElementSize, O, *this);
           bufferAggregateConstant(Initializer, &aggBuffer);
           if (aggBuffer.numSymbols) {
             if (nvptxSubtarget.is64Bit()) {
-              O << " .u64 " << *Mang->getSymbol(GVar) <<"[" ;
-              O << ElementSize/8;
-            }
-            else {
-              O << " .u32 " << *Mang->getSymbol(GVar) <<"[" ;
-              O << ElementSize/4;
+              O << " .u64 " << *Mang->getSymbol(GVar) << "[";
+              O << ElementSize / 8;
+            } else {
+              O << " .u32 " << *Mang->getSymbol(GVar) << "[";
+              O << ElementSize / 4;
             }
             O << "]";
-          }
-          else {
-            O << " .b8 " << *Mang->getSymbol(GVar) <<"[" ;
+          } else {
+            O << " .b8 " << *Mang->getSymbol(GVar) << "[";
             O << ElementSize;
             O << "]";
           }
-          O << " = {" ;
+          O << " = {";
           aggBuffer.print();
           O << "}";
-        }
-        else {
-          O << " .b8 " << *Mang->getSymbol(GVar) ;
+        } else {
+          O << " .b8 " << *Mang->getSymbol(GVar);
           if (ElementSize) {
-            O <<"[" ;
+            O << "[";
             O << ElementSize;
             O << "]";
           }
         }
-      }
-      else {
+      } else {
         O << " .b8 " << *Mang->getSymbol(GVar);
         if (ElementSize) {
-          O <<"[" ;
+          O << "[";
           O << ElementSize;
           O << "]";
         }
       }
       break;
     default:
-      assert( 0 && "type not supported yet");
+      assert(0 && "type not supported yet");
     }
 
   }
@@ -1289,7 +1285,7 @@ void NVPTXAsmPrinter::emitDemotedVars(const Function *f, raw_ostream &O) {
 
   std::vector<GlobalVariable *> &gvars = localDecls[f];
 
-  for (unsigned i=0, e=gvars.size(); i!=e; ++i) {
+  for (unsigned i = 0, e = gvars.size(); i != e; ++i) {
     O << "\t// demoted variable\n\t";
     printModuleLevelGV(gvars[i], O, true);
   }
@@ -1299,32 +1295,33 @@ void NVPTXAsmPrinter::emitPTXAddressSpace(unsigned int AddressSpace,
                                           raw_ostream &O) const {
   switch (AddressSpace) {
   case llvm::ADDRESS_SPACE_LOCAL:
-    O << "local" ;
+    O << "local";
     break;
   case llvm::ADDRESS_SPACE_GLOBAL:
-    O << "global" ;
+    O << "global";
     break;
   case llvm::ADDRESS_SPACE_CONST:
     // This logic should be consistent with that in
     // getCodeAddrSpace() (NVPTXISelDATToDAT.cpp)
     if (nvptxSubtarget.hasGenericLdSt())
-      O << "global" ;
+      O << "global";
     else
-      O << "const" ;
+      O << "const";
     break;
   case llvm::ADDRESS_SPACE_CONST_NOT_GEN:
-    O << "const" ;
+    O << "const";
     break;
   case llvm::ADDRESS_SPACE_SHARED:
-    O << "shared" ;
+    O << "shared";
     break;
   default:
-    llvm_unreachable("unexpected address space");
+    report_fatal_error("Bad address space found while emitting PTX");
+    break;
   }
 }
 
-std::string NVPTXAsmPrinter::getPTXFundamentalTypeStr(const Type *Ty,
-                                                      bool useB4PTR) const {
+std::string
+NVPTXAsmPrinter::getPTXFundamentalTypeStr(const Type *Ty, bool useB4PTR) const {
   switch (Ty->getTypeID()) {
   default:
     llvm_unreachable("unexpected type");
@@ -1348,17 +1345,20 @@ std::string NVPTXAsmPrinter::getPTXFundamentalTypeStr(const Type *Ty,
     return "f64";
   case Type::PointerTyID:
     if (nvptxSubtarget.is64Bit())
-      if (useB4PTR) return "b64";
-      else return "u64";
+      if (useB4PTR)
+        return "b64";
+      else
+        return "u64";
+    else if (useB4PTR)
+      return "b32";
     else
-      if (useB4PTR) return "b32";
-      else return "u32";
+      return "u32";
   }
   llvm_unreachable("unexpected type");
   return NULL;
 }
 
-void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable* GVar,
+void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable *GVar,
                                             raw_ostream &O) {
 
   const DataLayout *TD = TM.getDataLayout();
@@ -1382,7 +1382,7 @@ void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable* GVar,
     return;
   }
 
-  int64_t ElementSize =0;
+  int64_t ElementSize = 0;
 
   // Although PTX has direct support for struct type and array type and LLVM IR
   // is very similar to PTX, the LLVM CodeGen does not support for targets that
@@ -1393,22 +1393,19 @@ void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable* GVar,
   case Type::ArrayTyID:
   case Type::VectorTyID:
     ElementSize = TD->getTypeStoreSize(ETy);
-    O << " .b8 " << *Mang->getSymbol(GVar) <<"[" ;
+    O << " .b8 " << *Mang->getSymbol(GVar) << "[";
     if (ElementSize) {
-      O << itostr(ElementSize) ;
+      O << itostr(ElementSize);
     }
     O << "]";
     break;
   default:
-    assert( 0 && "type not supported yet");
+    assert(0 && "type not supported yet");
   }
-  return ;
+  return;
 }
 
-
-static unsigned int
-getOpenCLAlignment(const DataLayout *TD,
-                   Type *Ty) {
+static unsigned int getOpenCLAlignment(const DataLayout *TD, Type *Ty) {
   if (Ty->isPrimitiveType() || Ty->isIntegerTy() || isa<PointerType>(Ty))
     return TD->getPrefTypeAlignment(Ty);
 
@@ -1422,9 +1419,9 @@ getOpenCLAlignment(const DataLayout *TD,
     unsigned int numE = VTy->getNumElements();
     unsigned int alignE = TD->getPrefTypeAlignment(ETy);
     if (numE == 3)
-      return 4*alignE;
+      return 4 * alignE;
     else
-      return numE*alignE;
+      return numE * alignE;
   }
 
   const StructType *STy = dyn_cast<StructType>(Ty);
@@ -1432,7 +1429,7 @@ getOpenCLAlignment(const DataLayout *TD,
     unsigned int alignStruct = 1;
     // Go through each element of the struct and find the
     // largest alignment.
-    for (unsigned i=0, e=STy->getNumElements(); i != e; i++) {
+    for (unsigned i = 0, e = STy->getNumElements(); i != e; i++) {
       Type *ETy = STy->getElementType(i);
       unsigned int align = getOpenCLAlignment(TD, ETy);
       if (align > alignStruct)
@@ -1476,7 +1473,7 @@ void NVPTXAsmPrinter::printParamName(int paramIndex, raw_ostream &O) {
   }
 
   for (I = F->arg_begin(), E = F->arg_end(); I != E; ++I, i++) {
-    if (i==paramIndex) {
+    if (i == paramIndex) {
       printParamName(I, paramIndex, O);
       return;
     }
@@ -1484,10 +1481,9 @@ void NVPTXAsmPrinter::printParamName(int paramIndex, raw_ostream &O) {
   llvm_unreachable("paramIndex out of bound");
 }
 
-void NVPTXAsmPrinter::emitFunctionParamList(const Function *F,
-                                            raw_ostream &O) {
+void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
   const DataLayout *TD = TM.getDataLayout();
-  const AttrListPtr &PAL = F->getAttributes();
+  const AttributeSet &PAL = F->getAttributes();
   const TargetLowering *TLI = TM.getTargetLowering();
   Function::const_arg_iterator I, E;
   unsigned paramIndex = 0;
@@ -1499,7 +1495,7 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F,
   O << "(\n";
 
   for (I = F->arg_begin(), E = F->arg_end(); I != E; ++I, paramIndex++) {
-    const Type *Ty = I->getType();
+    Type *Ty = I->getType();
 
     if (!first)
       O << ",\n";
@@ -1514,15 +1510,28 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F,
           O << "\t.param .surfref " << *CurrentFnSym << "_param_" << paramIndex;
         else // Default image is read_only
           O << "\t.param .texref " << *CurrentFnSym << "_param_" << paramIndex;
-      }
-      else // Should be llvm::isSampler(*I)
+      } else // Should be llvm::isSampler(*I)
         O << "\t.param .samplerref " << *CurrentFnSym << "_param_"
-        << paramIndex;
+          << paramIndex;
       continue;
     }
 
-    if (PAL.getParamAttributes(paramIndex+1).
-          hasAttribute(Attributes::ByVal) == false) {
+    if (PAL.hasAttribute(paramIndex + 1, Attribute::ByVal) == false) {
+      if (Ty->isVectorTy()) {
+        // Just print .param .b8 .align <a> .param[size];
+        // <a> = PAL.getparamalignment
+        // size = typeallocsize of element type
+        unsigned align = PAL.getParamAlignment(paramIndex + 1);
+        if (align == 0)
+          align = TD->getABITypeAlignment(Ty);
+
+        unsigned sz = TD->getTypeAllocSize(Ty);
+        O << "\t.param .align " << align << " .b8 ";
+        printParamName(I, paramIndex, O);
+        O << "[" << sz << "]";
+
+        continue;
+      }
       // Just a scalar
       const PointerType *PTy = dyn_cast<PointerType>(Ty);
       if (isKernelFunc) {
@@ -1533,7 +1542,7 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F,
           if (nvptxSubtarget.getDrvInterface() != NVPTX::CUDA) {
             Type *ETy = PTy->getElementType();
             int addrSpace = PTy->getAddressSpace();
-            switch(addrSpace) {
+            switch (addrSpace) {
             default:
               O << ".ptr ";
               break;
@@ -1548,15 +1557,14 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F,
               O << ".ptr .global ";
               break;
             }
-            O << ".align " << (int)getOpenCLAlignment(TD, ETy) << " ";
+            O << ".align " << (int) getOpenCLAlignment(TD, ETy) << " ";
           }
           printParamName(I, paramIndex, O);
           continue;
         }
 
         // non-pointer scalar to kernel func
-        O << "\t.param ."
-            << getPTXFundamentalTypeStr(Ty) << " ";
+        O << "\t.param ." << getPTXFundamentalTypeStr(Ty) << " ";
         printParamName(I, paramIndex, O);
         continue;
       }
@@ -1565,9 +1573,9 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F,
       unsigned sz = 0;
       if (isa<IntegerType>(Ty)) {
         sz = cast<IntegerType>(Ty)->getBitWidth();
-        if (sz < 32) sz = 32;
-      }
-      else if (isa<PointerType>(Ty))
+        if (sz < 32)
+          sz = 32;
+      } else if (isa<PointerType>(Ty))
         sz = thePointerTy.getSizeInBits();
       else
         sz = Ty->getPrimitiveSizeInBits();
@@ -1581,21 +1589,19 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F,
 
     // param has byVal attribute. So should be a pointer
     const PointerType *PTy = dyn_cast<PointerType>(Ty);
-    assert(PTy &&
-           "Param with byval attribute should be a pointer type");
+    assert(PTy && "Param with byval attribute should be a pointer type");
     Type *ETy = PTy->getElementType();
 
     if (isABI || isKernelFunc) {
       // Just print .param .b8 .align <a> .param[size];
       // <a> = PAL.getparamalignment
       // size = typeallocsize of element type
-      unsigned align = PAL.getParamAlignment(paramIndex+1);
+      unsigned align = PAL.getParamAlignment(paramIndex + 1);
       if (align == 0)
         align = TD->getABITypeAlignment(ETy);
 
       unsigned sz = TD->getTypeAllocSize(ETy);
-      O << "\t.param .align " << align
-          << " .b8 ";
+      O << "\t.param .align " << align << " .b8 ";
       printParamName(I, paramIndex, O);
       O << "[" << sz << "]";
       continue;
@@ -1606,7 +1612,7 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F,
       // each vector element.
       SmallVector<EVT, 16> vtparts;
       ComputeValueVTs(*TLI, ETy, vtparts);
-      for (unsigned i=0,e=vtparts.size(); i!=e; ++i) {
+      for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
         unsigned elems = 1;
         EVT elemtype = vtparts[i];
         if (vtparts[i].isVector()) {
@@ -1614,15 +1620,17 @@ void NVPTXAsmPrinter::emitFunctionParamList(const Function *F,
           elemtype = vtparts[i].getVectorElementType();
         }
 
-        for (unsigned j=0,je=elems; j!=je; ++j) {
+        for (unsigned j = 0, je = elems; j != je; ++j) {
           unsigned sz = elemtype.getSizeInBits();
-          if (elemtype.isInteger() && (sz < 32)) sz = 32;
+          if (elemtype.isInteger() && (sz < 32))
+            sz = 32;
           O << "\t.reg .b" << sz << " ";
           printParamName(I, paramIndex, O);
-          if (j<je-1) O << ",\n";
+          if (j < je - 1)
+            O << ",\n";
           ++paramIndex;
         }
-        if (i<e-1)
+        if (i < e - 1)
           O << ",\n";
       }
       --paramIndex;
@@ -1639,9 +1647,8 @@ void NVPTXAsmPrinter::emitFunctionParamList(const MachineFunction &MF,
   emitFunctionParamList(F, O);
 }
 
-
-void NVPTXAsmPrinter::
-setAndEmitFunctionVirtualRegisters(const MachineFunction &MF) {
+void NVPTXAsmPrinter::setAndEmitFunctionVirtualRegisters(
+    const MachineFunction &MF) {
   SmallString<128> Str;
   raw_svector_ostream O(Str);
 
@@ -1654,14 +1661,12 @@ setAndEmitFunctionVirtualRegisters(const MachineFunction &MF) {
   const MachineFrameInfo *MFI = MF.getFrameInfo();
   int NumBytes = (int) MFI->getStackSize();
   if (NumBytes) {
-    O << "\t.local .align " << MFI->getMaxAlignment() << " .b8 \t"
-        << DEPOTNAME
-        << getFunctionNumber() << "[" << NumBytes << "];\n";
+    O << "\t.local .align " << MFI->getMaxAlignment() << " .b8 \t" << DEPOTNAME
+      << getFunctionNumber() << "[" << NumBytes << "];\n";
     if (nvptxSubtarget.is64Bit()) {
       O << "\t.reg .b64 \t%SP;\n";
       O << "\t.reg .b64 \t%SPL;\n";
-    }
-    else {
+    } else {
       O << "\t.reg .b32 \t%SP;\n";
       O << "\t.reg .b32 \t%SPL;\n";
     }
@@ -1672,12 +1677,12 @@ setAndEmitFunctionVirtualRegisters(const MachineFunction &MF) {
   // register number and the per class virtual register number.
   // We use the per class virtual register number in the ptx output.
   unsigned int numVRs = MRI->getNumVirtRegs();
-  for (unsigned i=0; i< numVRs; i++) {
+  for (unsigned i = 0; i < numVRs; i++) {
     unsigned int vr = TRI->index2VirtReg(i);
     const TargetRegisterClass *RC = MRI->getRegClass(vr);
     std::map<unsigned, unsigned> &regmap = VRidGlobal2LocalMap[RC->getID()];
     int n = regmap.size();
-    regmap.insert(std::make_pair(vr, n+1));
+    regmap.insert(std::make_pair(vr, n + 1));
   }
 
   // Emit register declarations
@@ -1721,23 +1726,20 @@ setAndEmitFunctionVirtualRegisters(const MachineFunction &MF) {
   OutStreamer.EmitRawText(O.str());
 }
 
-
 void NVPTXAsmPrinter::printFPConstant(const ConstantFP *Fp, raw_ostream &O) {
-  APFloat APF = APFloat(Fp->getValueAPF());  // make a copy
+  APFloat APF = APFloat(Fp->getValueAPF()); // make a copy
   bool ignored;
   unsigned int numHex;
   const char *lead;
 
-  if (Fp->getType()->getTypeID()==Type::FloatTyID) {
+  if (Fp->getType()->getTypeID() == Type::FloatTyID) {
     numHex = 8;
     lead = "0f";
-    APF.convert(APFloat::IEEEsingle, APFloat::rmNearestTiesToEven,
-                &ignored);
+    APF.convert(APFloat::IEEEsingle, APFloat::rmNearestTiesToEven, &ignored);
   } else if (Fp->getType()->getTypeID() == Type::DoubleTyID) {
     numHex = 16;
     lead = "0d";
-    APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven,
-                &ignored);
+    APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &ignored);
   } else
     llvm_unreachable("unsupported fp type");
 
@@ -1779,7 +1781,6 @@ void NVPTXAsmPrinter::printScalarConstant(Constant *CPV, raw_ostream &O) {
   llvm_unreachable("Not scalar type found in printScalarConstant()");
 }
 
-
 void NVPTXAsmPrinter::bufferLEByte(Constant *CPV, int Bytes,
                                    AggBuffer *aggBuffer) {
 
@@ -1787,7 +1788,7 @@ void NVPTXAsmPrinter::bufferLEByte(Constant *CPV, int Bytes,
 
   if (isa<UndefValue>(CPV) || CPV->isNullValue()) {
     int s = TD->getTypeAllocSize(CPV->getType());
-    if (s<Bytes)
+    if (s < Bytes)
       s = Bytes;
     aggBuffer->addZeros(s);
     return;
@@ -1798,28 +1799,26 @@ void NVPTXAsmPrinter::bufferLEByte(Constant *CPV, int Bytes,
 
   case Type::IntegerTyID: {
     const Type *ETy = CPV->getType();
-    if ( ETy == Type::getInt8Ty(CPV->getContext()) ){
+    if (ETy == Type::getInt8Ty(CPV->getContext())) {
       unsigned char c =
           (unsigned char)(dyn_cast<ConstantInt>(CPV))->getZExtValue();
       ptr = &c;
       aggBuffer->addBytes(ptr, 1, Bytes);
-    } else if ( ETy == Type::getInt16Ty(CPV->getContext()) ) {
-      short int16 =
-          (short)(dyn_cast<ConstantInt>(CPV))->getZExtValue();
-      ptr = (unsigned char*)&int16;
+    } else if (ETy == Type::getInt16Ty(CPV->getContext())) {
+      short int16 = (short)(dyn_cast<ConstantInt>(CPV))->getZExtValue();
+      ptr = (unsigned char *)&int16;
       aggBuffer->addBytes(ptr, 2, Bytes);
-    } else if ( ETy == Type::getInt32Ty(CPV->getContext()) ) {
+    } else if (ETy == Type::getInt32Ty(CPV->getContext())) {
       if (ConstantInt *constInt = dyn_cast<ConstantInt>(CPV)) {
-        int int32 =(int)(constInt->getZExtValue());
-        ptr = (unsigned char*)&int32;
+        int int32 = (int)(constInt->getZExtValue());
+        ptr = (unsigned char *)&int32;
         aggBuffer->addBytes(ptr, 4, Bytes);
         break;
       } else if (ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
-        if (ConstantInt *constInt =
-            dyn_cast<ConstantInt>(ConstantFoldConstantExpression(
-                Cexpr, TD))) {
-          int int32 =(int)(constInt->getZExtValue());
-          ptr = (unsigned char*)&int32;
+        if (ConstantInt *constInt = dyn_cast<ConstantInt>(
+                ConstantFoldConstantExpression(Cexpr, TD))) {
+          int int32 = (int)(constInt->getZExtValue());
+          ptr = (unsigned char *)&int32;
           aggBuffer->addBytes(ptr, 4, Bytes);
           break;
         }
@@ -1831,17 +1830,17 @@ void NVPTXAsmPrinter::bufferLEByte(Constant *CPV, int Bytes,
         }
       }
       llvm_unreachable("unsupported integer const type");
-    } else if (ETy == Type::getInt64Ty(CPV->getContext()) ) {
+    } else if (ETy == Type::getInt64Ty(CPV->getContext())) {
       if (ConstantInt *constInt = dyn_cast<ConstantInt>(CPV)) {
-        long long int64 =(long long)(constInt->getZExtValue());
-        ptr = (unsigned char*)&int64;
+        long long int64 = (long long)(constInt->getZExtValue());
+        ptr = (unsigned char *)&int64;
         aggBuffer->addBytes(ptr, 8, Bytes);
         break;
       } else if (ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
         if (ConstantInt *constInt = dyn_cast<ConstantInt>(
-            ConstantFoldConstantExpression(Cexpr, TD))) {
-          long long int64 =(long long)(constInt->getZExtValue());
-          ptr = (unsigned char*)&int64;
+                ConstantFoldConstantExpression(Cexpr, TD))) {
+          long long int64 = (long long)(constInt->getZExtValue());
+          ptr = (unsigned char *)&int64;
           aggBuffer->addBytes(ptr, 8, Bytes);
           break;
         }
@@ -1860,17 +1859,16 @@ void NVPTXAsmPrinter::bufferLEByte(Constant *CPV, int Bytes,
   case Type::FloatTyID:
   case Type::DoubleTyID: {
     ConstantFP *CFP = dyn_cast<ConstantFP>(CPV);
-    const Type* Ty = CFP->getType();
+    const Type *Ty = CFP->getType();
     if (Ty == Type::getFloatTy(CPV->getContext())) {
-      float float32 = (float)CFP->getValueAPF().convertToFloat();
-      ptr = (unsigned char*)&float32;
+      float float32 = (float) CFP->getValueAPF().convertToFloat();
+      ptr = (unsigned char *)&float32;
       aggBuffer->addBytes(ptr, 4, Bytes);
     } else if (Ty == Type::getDoubleTy(CPV->getContext())) {
       double float64 = CFP->getValueAPF().convertToDouble();
-      ptr = (unsigned char*)&float64;
+      ptr = (unsigned char *)&float64;
       aggBuffer->addBytes(ptr, 8, Bytes);
-    }
-    else {
+    } else {
       llvm_unreachable("unsupported fp const type");
     }
     break;
@@ -1878,8 +1876,7 @@ void NVPTXAsmPrinter::bufferLEByte(Constant *CPV, int Bytes,
   case Type::PointerTyID: {
     if (GlobalValue *GVar = dyn_cast<GlobalValue>(CPV)) {
       aggBuffer->addSymbol(GVar);
-    }
-    else if (ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
+    } else if (ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
       Value *v = Cexpr->stripPointerCasts();
       aggBuffer->addSymbol(v);
     }
@@ -1895,10 +1892,9 @@ void NVPTXAsmPrinter::bufferLEByte(Constant *CPV, int Bytes,
         isa<ConstantStruct>(CPV)) {
       int ElementSize = TD->getTypeAllocSize(CPV->getType());
       bufferAggregateConstant(CPV, aggBuffer);
-      if ( Bytes > ElementSize )
-        aggBuffer->addZeros(Bytes-ElementSize);
-    }
-    else if (isa<ConstantAggregateZero>(CPV))
+      if (Bytes > ElementSize)
+        aggBuffer->addZeros(Bytes - ElementSize);
+    } else if (isa<ConstantAggregateZero>(CPV))
       aggBuffer->addZeros(Bytes);
     else
       llvm_unreachable("Unexpected Constant type");
@@ -1924,7 +1920,7 @@ void NVPTXAsmPrinter::bufferAggregateConstant(Constant *CPV,
   }
 
   if (const ConstantDataSequential *CDS =
-      dyn_cast<ConstantDataSequential>(CPV)) {
+          dyn_cast<ConstantDataSequential>(CPV)) {
     if (CDS->getNumElements())
       for (unsigned i = 0; i < CDS->getNumElements(); ++i)
         bufferLEByte(cast<Constant>(CDS->getElementAsConstant(i)), 0,
@@ -1932,20 +1928,18 @@ void NVPTXAsmPrinter::bufferAggregateConstant(Constant *CPV,
     return;
   }
 
-
   if (isa<ConstantStruct>(CPV)) {
     if (CPV->getNumOperands()) {
       StructType *ST = cast<StructType>(CPV->getType());
       for (unsigned i = 0, e = CPV->getNumOperands(); i != e; ++i) {
-        if ( i == (e - 1))
+        if (i == (e - 1))
           Bytes = TD->getStructLayout(ST)->getElementOffset(0) +
-          TD->getTypeAllocSize(ST)
-          - TD->getStructLayout(ST)->getElementOffset(i);
+                  TD->getTypeAllocSize(ST) -
+                  TD->getStructLayout(ST)->getElementOffset(i);
         else
-          Bytes = TD->getStructLayout(ST)->getElementOffset(i+1) -
-          TD->getStructLayout(ST)->getElementOffset(i);
-        bufferLEByte(cast<Constant>(CPV->getOperand(i)), Bytes,
-                     aggBuffer);
+          Bytes = TD->getStructLayout(ST)->getElementOffset(i + 1) -
+                  TD->getStructLayout(ST)->getElementOffset(i);
+        bufferLEByte(cast<Constant>(CPV->getOperand(i)), Bytes, aggBuffer);
       }
     }
     return;
@@ -1956,15 +1950,13 @@ void NVPTXAsmPrinter::bufferAggregateConstant(Constant *CPV,
 // buildTypeNameMap - Run through symbol table looking for type names.
 //
 
-
 bool NVPTXAsmPrinter::isImageType(const Type *Ty) {
 
   std::map<const Type *, std::string>::iterator PI = TypeNameMap.find(Ty);
 
-  if (PI != TypeNameMap.end() &&
-      (!PI->second.compare("struct._image1d_t") ||
-          !PI->second.compare("struct._image2d_t") ||
-          !PI->second.compare("struct._image3d_t")))
+  if (PI != TypeNameMap.end() && (!PI->second.compare("struct._image1d_t") ||
+                                  !PI->second.compare("struct._image2d_t") ||
+                                  !PI->second.compare("struct._image3d_t")))
     return true;
 
   return false;
@@ -1974,10 +1966,10 @@ bool NVPTXAsmPrinter::isImageType(const Type *Ty) {
 ///
 bool NVPTXAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
                                       unsigned AsmVariant,
-                                      const char *ExtraCode,
-                                      raw_ostream &O) {
+                                      const char *ExtraCode, raw_ostream &O) {
   if (ExtraCode && ExtraCode[0]) {
-    if (ExtraCode[1] != 0) return true; // Unknown modifier.
+    if (ExtraCode[1] != 0)
+      return true; // Unknown modifier.
 
     switch (ExtraCode[0]) {
     default:
@@ -1993,13 +1985,11 @@ bool NVPTXAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
   return false;
 }
 
-bool NVPTXAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI,
-                                            unsigned OpNo,
-                                            unsigned AsmVariant,
-                                            const char *ExtraCode,
-                                            raw_ostream &O) {
+bool NVPTXAsmPrinter::PrintAsmMemoryOperand(
+    const MachineInstr *MI, unsigned OpNo, unsigned AsmVariant,
+    const char *ExtraCode, raw_ostream &O) {
   if (ExtraCode && ExtraCode[0])
-    return true;  // Unknown modifier
+    return true; // Unknown modifier
 
   O << '[';
   printMemOperand(MI, OpNo, O);
@@ -2008,71 +1998,69 @@ bool NVPTXAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI,
   return false;
 }
 
-bool NVPTXAsmPrinter::ignoreLoc(const MachineInstr &MI)
-{
-  switch(MI.getOpcode()) {
+bool NVPTXAsmPrinter::ignoreLoc(const MachineInstr &MI) {
+  switch (MI.getOpcode()) {
   default:
     return false;
-  case NVPTX::CallArgBeginInst:  case NVPTX::CallArgEndInst0:
-  case NVPTX::CallArgEndInst1:  case NVPTX::CallArgF32:
-  case NVPTX::CallArgF64:  case NVPTX::CallArgI16:
-  case NVPTX::CallArgI32:  case NVPTX::CallArgI32imm:
-  case NVPTX::CallArgI64:  case NVPTX::CallArgI8:
-  case NVPTX::CallArgParam:  case NVPTX::CallVoidInst:
-  case NVPTX::CallVoidInstReg:  case NVPTX::Callseq_End:
+  case NVPTX::CallArgBeginInst:
+  case NVPTX::CallArgEndInst0:
+  case NVPTX::CallArgEndInst1:
+  case NVPTX::CallArgF32:
+  case NVPTX::CallArgF64:
+  case NVPTX::CallArgI16:
+  case NVPTX::CallArgI32:
+  case NVPTX::CallArgI32imm:
+  case NVPTX::CallArgI64:
+  case NVPTX::CallArgI8:
+  case NVPTX::CallArgParam:
+  case NVPTX::CallVoidInst:
+  case NVPTX::CallVoidInstReg:
+  case NVPTX::Callseq_End:
   case NVPTX::CallVoidInstReg64:
-  case NVPTX::DeclareParamInst:  case NVPTX::DeclareRetMemInst:
-  case NVPTX::DeclareRetRegInst:  case NVPTX::DeclareRetScalarInst:
-  case NVPTX::DeclareScalarParamInst:  case NVPTX::DeclareScalarRegInst:
-  case NVPTX::StoreParamF32:  case NVPTX::StoreParamF64:
-  case NVPTX::StoreParamI16:  case NVPTX::StoreParamI32:
-  case NVPTX::StoreParamI64:  case NVPTX::StoreParamI8:
-  case NVPTX::StoreParamS32I8:  case NVPTX::StoreParamU32I8:
-  case NVPTX::StoreParamS32I16:  case NVPTX::StoreParamU32I16:
-  case NVPTX::StoreParamScalar2F32:  case NVPTX::StoreParamScalar2F64:
-  case NVPTX::StoreParamScalar2I16:  case NVPTX::StoreParamScalar2I32:
-  case NVPTX::StoreParamScalar2I64:  case NVPTX::StoreParamScalar2I8:
-  case NVPTX::StoreParamScalar4F32:  case NVPTX::StoreParamScalar4I16:
-  case NVPTX::StoreParamScalar4I32:  case NVPTX::StoreParamScalar4I8:
-  case NVPTX::StoreParamV2F32:  case NVPTX::StoreParamV2F64:
-  case NVPTX::StoreParamV2I16:  case NVPTX::StoreParamV2I32:
-  case NVPTX::StoreParamV2I64:  case NVPTX::StoreParamV2I8:
-  case NVPTX::StoreParamV4F32:  case NVPTX::StoreParamV4I16:
-  case NVPTX::StoreParamV4I32:  case NVPTX::StoreParamV4I8:
-  case NVPTX::StoreRetvalF32:  case NVPTX::StoreRetvalF64:
-  case NVPTX::StoreRetvalI16:  case NVPTX::StoreRetvalI32:
-  case NVPTX::StoreRetvalI64:  case NVPTX::StoreRetvalI8:
-  case NVPTX::StoreRetvalScalar2F32:  case NVPTX::StoreRetvalScalar2F64:
-  case NVPTX::StoreRetvalScalar2I16:  case NVPTX::StoreRetvalScalar2I32:
-  case NVPTX::StoreRetvalScalar2I64:  case NVPTX::StoreRetvalScalar2I8:
-  case NVPTX::StoreRetvalScalar4F32:  case NVPTX::StoreRetvalScalar4I16:
-  case NVPTX::StoreRetvalScalar4I32:  case NVPTX::StoreRetvalScalar4I8:
-  case NVPTX::StoreRetvalV2F32:  case NVPTX::StoreRetvalV2F64:
-  case NVPTX::StoreRetvalV2I16:  case NVPTX::StoreRetvalV2I32:
-  case NVPTX::StoreRetvalV2I64:  case NVPTX::StoreRetvalV2I8:
-  case NVPTX::StoreRetvalV4F32:  case NVPTX::StoreRetvalV4I16:
-  case NVPTX::StoreRetvalV4I32:  case NVPTX::StoreRetvalV4I8:
-  case NVPTX::LastCallArgF32:  case NVPTX::LastCallArgF64:
-  case NVPTX::LastCallArgI16:  case NVPTX::LastCallArgI32:
-  case NVPTX::LastCallArgI32imm:  case NVPTX::LastCallArgI64:
-  case NVPTX::LastCallArgI8:  case NVPTX::LastCallArgParam:
-  case NVPTX::LoadParamMemF32:  case NVPTX::LoadParamMemF64:
-  case NVPTX::LoadParamMemI16:  case NVPTX::LoadParamMemI32:
-  case NVPTX::LoadParamMemI64:  case NVPTX::LoadParamMemI8:
-  case NVPTX::LoadParamRegF32:  case NVPTX::LoadParamRegF64:
-  case NVPTX::LoadParamRegI16:  case NVPTX::LoadParamRegI32:
-  case NVPTX::LoadParamRegI64:  case NVPTX::LoadParamRegI8:
-  case NVPTX::LoadParamScalar2F32:  case NVPTX::LoadParamScalar2F64:
-  case NVPTX::LoadParamScalar2I16:  case NVPTX::LoadParamScalar2I32:
-  case NVPTX::LoadParamScalar2I64:  case NVPTX::LoadParamScalar2I8:
-  case NVPTX::LoadParamScalar4F32:  case NVPTX::LoadParamScalar4I16:
-  case NVPTX::LoadParamScalar4I32:  case NVPTX::LoadParamScalar4I8:
-  case NVPTX::LoadParamV2F32:  case NVPTX::LoadParamV2F64:
-  case NVPTX::LoadParamV2I16:  case NVPTX::LoadParamV2I32:
-  case NVPTX::LoadParamV2I64:  case NVPTX::LoadParamV2I8:
-  case NVPTX::LoadParamV4F32:  case NVPTX::LoadParamV4I16:
-  case NVPTX::LoadParamV4I32:  case NVPTX::LoadParamV4I8:
-  case NVPTX::PrototypeInst:   case NVPTX::DBG_VALUE:
+  case NVPTX::DeclareParamInst:
+  case NVPTX::DeclareRetMemInst:
+  case NVPTX::DeclareRetRegInst:
+  case NVPTX::DeclareRetScalarInst:
+  case NVPTX::DeclareScalarParamInst:
+  case NVPTX::DeclareScalarRegInst:
+  case NVPTX::StoreParamF32:
+  case NVPTX::StoreParamF64:
+  case NVPTX::StoreParamI16:
+  case NVPTX::StoreParamI32:
+  case NVPTX::StoreParamI64:
+  case NVPTX::StoreParamI8:
+  case NVPTX::StoreParamS32I8:
+  case NVPTX::StoreParamU32I8:
+  case NVPTX::StoreParamS32I16:
+  case NVPTX::StoreParamU32I16:
+  case NVPTX::StoreRetvalF32:
+  case NVPTX::StoreRetvalF64:
+  case NVPTX::StoreRetvalI16:
+  case NVPTX::StoreRetvalI32:
+  case NVPTX::StoreRetvalI64:
+  case NVPTX::StoreRetvalI8:
+  case NVPTX::LastCallArgF32:
+  case NVPTX::LastCallArgF64:
+  case NVPTX::LastCallArgI16:
+  case NVPTX::LastCallArgI32:
+  case NVPTX::LastCallArgI32imm:
+  case NVPTX::LastCallArgI64:
+  case NVPTX::LastCallArgI8:
+  case NVPTX::LastCallArgParam:
+  case NVPTX::LoadParamMemF32:
+  case NVPTX::LoadParamMemF64:
+  case NVPTX::LoadParamMemI16:
+  case NVPTX::LoadParamMemI32:
+  case NVPTX::LoadParamMemI64:
+  case NVPTX::LoadParamMemI8:
+  case NVPTX::LoadParamRegF32:
+  case NVPTX::LoadParamRegF64:
+  case NVPTX::LoadParamRegI16:
+  case NVPTX::LoadParamRegI32:
+  case NVPTX::LoadParamRegI64:
+  case NVPTX::LoadParamRegI8:
+  case NVPTX::PrototypeInst:
+  case NVPTX::DBG_VALUE:
     return true;
   }
   return false;
@@ -2084,10 +2072,9 @@ extern "C" void LLVMInitializeNVPTXBackendAsmPrinter() {
   RegisterAsmPrinter<NVPTXAsmPrinter> Y(TheNVPTXTarget64);
 }
 
-
 void NVPTXAsmPrinter::emitSrcInText(StringRef filename, unsigned line) {
   std::stringstream temp;
-  LineReader * reader = this->getReader(filename.str());
+  LineReader *reader = this->getReader(filename.str());
   temp << "\n//";
   temp << filename.str();
   temp << ":";
@@ -2098,29 +2085,26 @@ void NVPTXAsmPrinter::emitSrcInText(StringRef filename, unsigned line) {
   this->OutStreamer.EmitRawText(Twine(temp.str()));
 }
 
-
 LineReader *NVPTXAsmPrinter::getReader(std::string filename) {
-  if (reader == NULL)  {
-    reader =  new LineReader(filename);
+  if (reader == NULL) {
+    reader = new LineReader(filename);
   }
 
   if (reader->fileName() != filename) {
     delete reader;
-    reader =  new LineReader(filename);
+    reader = new LineReader(filename);
   }
 
   return reader;
 }
 
-
-std::string
-LineReader::readLine(unsigned lineNum) {
+std::string LineReader::readLine(unsigned lineNum) {
   if (lineNum < theCurLine) {
     theCurLine = 0;
-    fstr.seekg(0,std::ios::beg);
+    fstr.seekg(0, std::ios::beg);
   }
   while (theCurLine < lineNum) {
-    fstr.getline(buff,500);
+    fstr.getline(buff, 500);
     theCurLine++;
   }
   return buff;
diff --git a/lib/Target/NVPTX/NVPTXAsmPrinter.h b/lib/Target/NVPTX/NVPTXAsmPrinter.h
index 6488b1442580..6dc9fc0ffeff 100644
--- a/lib/Target/NVPTX/NVPTXAsmPrinter.h
+++ b/lib/Target/NVPTX/NVPTXAsmPrinter.h
@@ -16,19 +16,19 @@
 #define NVPTXASMPRINTER_H
 
 #include "NVPTX.h"
-#include "NVPTXTargetMachine.h"
 #include "NVPTXSubtarget.h"
-#include "llvm/Function.h"
+#include "NVPTXTargetMachine.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/CodeGen/AsmPrinter.h"
-#include "llvm/Support/CommandLine.h"
+#include "llvm/IR/Function.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCExpr.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/MC/MCSymbol.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/FormattedStream.h"
 #include "llvm/Target/Mangler.h"
-#include "llvm/ADT/SmallString.h"
-#include "llvm/ADT/StringExtras.h"
+#include "llvm/Target/TargetMachine.h"
 #include <fstream>
 
 // The ptx syntax and format is very different from that usually seem in a .s
@@ -43,15 +43,15 @@
 // This is defined in AsmPrinter.cpp.
 // Used to process the constant expressions in initializers.
 namespace nvptx {
-const llvm::MCExpr *LowerConstant(const llvm::Constant *CV,
-                                  llvm::AsmPrinter &AP) ;
+const llvm::MCExpr *
+LowerConstant(const llvm::Constant *CV, llvm::AsmPrinter &AP);
 }
 
 namespace llvm {
 
 class LineReader {
 private:
-  unsigned theCurLine ;
+  unsigned theCurLine;
   std::ifstream fstr;
   char buff[512];
   std::string theFileName;
@@ -63,17 +63,12 @@ public:
     theFileName = filename;
   }
   std::string fileName() { return theFileName; }
-  ~LineReader() {
-    fstr.close();
-  }
+  ~LineReader() { fstr.close(); }
   std::string readLine(unsigned line);
 };
 
-
-
 class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
 
-
   class AggBuffer {
     // Used to buffer the emitted string for initializing global
     // aggregates.
@@ -92,7 +87,7 @@ class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
     // Once we have this AggBuffer setup, we can choose how to print
     // it out.
   public:
-    unsigned size;   // size of the buffer in bytes
+    unsigned size;         // size of the buffer in bytes
     unsigned char *buffer; // the buffer
     unsigned numSymbols;   // number of symbol addresses
     SmallVector<unsigned, 4> symbolPosInBuffer;
@@ -105,33 +100,31 @@ class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
 
   public:
     AggBuffer(unsigned _size, raw_ostream &_O, NVPTXAsmPrinter &_AP)
-    :O(_O),AP(_AP) {
+        : O(_O), AP(_AP) {
       buffer = new unsigned char[_size];
       size = _size;
       curpos = 0;
       numSymbols = 0;
     }
-    ~AggBuffer() {
-      delete [] buffer;
-    }
+    ~AggBuffer() { delete[] buffer; }
     unsigned addBytes(unsigned char *Ptr, int Num, int Bytes) {
-      assert((curpos+Num) <= size);
-      assert((curpos+Bytes) <= size);
-      for ( int i= 0; i < Num; ++i) {
+      assert((curpos + Num) <= size);
+      assert((curpos + Bytes) <= size);
+      for (int i = 0; i < Num; ++i) {
         buffer[curpos] = Ptr[i];
-        curpos ++;
+        curpos++;
       }
-      for ( int i=Num; i < Bytes ; ++i) {
+      for (int i = Num; i < Bytes; ++i) {
         buffer[curpos] = 0;
-        curpos ++;
+        curpos++;
       }
       return curpos;
     }
     unsigned addZeros(int Num) {
-      assert((curpos+Num) <= size);
-      for ( int i= 0; i < Num; ++i) {
+      assert((curpos + Num) <= size);
+      for (int i = 0; i < Num; ++i) {
         buffer[curpos] = 0;
-        curpos ++;
+        curpos++;
       }
       return curpos;
     }
@@ -143,10 +136,10 @@ class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
     void print() {
       if (numSymbols == 0) {
         // print out in bytes
-        for (unsigned i=0; i<size; i++) {
+        for (unsigned i = 0; i < size; i++) {
           if (i)
             O << ", ";
-          O << (unsigned int)buffer[i];
+          O << (unsigned int) buffer[i];
         }
       } else {
         // print out in 4-bytes or 8-bytes
@@ -156,7 +149,7 @@ class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
         unsigned int nBytes = 4;
         if (AP.nvptxSubtarget.is64Bit())
           nBytes = 8;
-        for (pos=0; pos<size; pos+=nBytes) {
+        for (pos = 0; pos < size; pos += nBytes) {
           if (pos)
             O << ", ";
           if (pos == nextSymbolPos) {
@@ -164,22 +157,19 @@ class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
             if (GlobalValue *GVar = dyn_cast<GlobalValue>(v)) {
               MCSymbol *Name = AP.Mang->getSymbol(GVar);
               O << *Name;
-            }
-            else if (ConstantExpr *Cexpr =
-                dyn_cast<ConstantExpr>(v)) {
+            } else if (ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(v)) {
               O << *nvptx::LowerConstant(Cexpr, AP);
             } else
               llvm_unreachable("symbol type unknown");
             nSym++;
             if (nSym >= numSymbols)
-              nextSymbolPos = size+1;
+              nextSymbolPos = size + 1;
             else
               nextSymbolPos = symbolPosInBuffer[nSym];
-          } else
-            if (nBytes == 4)
-              O << *(unsigned int*)(buffer+pos);
-            else
-              O << *(unsigned long long*)(buffer+pos);
+          } else if (nBytes == 4)
+            O << *(unsigned int *)(buffer + pos);
+          else
+            O << *(unsigned long long *)(buffer + pos);
         }
       }
     }
@@ -189,10 +179,8 @@ class LLVM_LIBRARY_VISIBILITY NVPTXAsmPrinter : public AsmPrinter {
 
   virtual void emitSrcInText(StringRef filename, unsigned line);
 
-private :
-  virtual const char *getPassName() const {
-    return "NVPTX Assembly Printer";
-  }
+private:
+  virtual const char *getPassName() const { return "NVPTX Assembly Printer"; }
 
   const Function *F;
   std::string CurrentFnName;
@@ -207,31 +195,28 @@ private :
 
   void printGlobalVariable(const GlobalVariable *GVar);
   void printOperand(const MachineInstr *MI, int opNum, raw_ostream &O,
-                    const char *Modifier=0);
+                    const char *Modifier = 0);
   void printLdStCode(const MachineInstr *MI, int opNum, raw_ostream &O,
-                     const char *Modifier=0);
-  void printVecModifiedImmediate(const MachineOperand &MO,
-                                 const char *Modifier, raw_ostream &O);
+                     const char *Modifier = 0);
+  void printVecModifiedImmediate(const MachineOperand &MO, const char *Modifier,
+                                 raw_ostream &O);
   void printMemOperand(const MachineInstr *MI, int opNum, raw_ostream &O,
-                       const char *Modifier=0);
+                       const char *Modifier = 0);
   void printImplicitDef(const MachineInstr *MI, raw_ostream &O) const;
   // definition autogenerated.
   void printInstruction(const MachineInstr *MI, raw_ostream &O);
-  void printModuleLevelGV(GlobalVariable* GVar, raw_ostream &O,
-                          bool=false);
+  void printModuleLevelGV(GlobalVariable *GVar, raw_ostream &O, bool = false);
   void printParamName(int paramIndex, raw_ostream &O);
   void printParamName(Function::const_arg_iterator I, int paramIndex,
                       raw_ostream &O);
   void emitHeader(Module &M, raw_ostream &O);
-  void emitKernelFunctionDirectives(const Function& F,
-                                    raw_ostream &O) const;
+  void emitKernelFunctionDirectives(const Function &F, raw_ostream &O) const;
   void emitVirtualRegister(unsigned int vr, bool isVec, raw_ostream &O);
   void emitFunctionExternParamList(const MachineFunction &MF);
   void emitFunctionParamList(const Function *, raw_ostream &O);
   void emitFunctionParamList(const MachineFunction &MF, raw_ostream &O);
   void setAndEmitFunctionVirtualRegisters(const MachineFunction &MF);
-  void emitFunctionTempData(const MachineFunction &MF,
-                            unsigned &FrameSize);
+  void emitFunctionTempData(const MachineFunction &MF, unsigned &FrameSize);
   bool isImageType(const Type *Ty);
   bool PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
                        unsigned AsmVariant, const char *ExtraCode,
@@ -269,17 +254,16 @@ private:
   void recordAndEmitFilenames(Module &);
 
   void emitPTXGlobalVariable(const GlobalVariable *GVar, raw_ostream &O);
-  void emitPTXAddressSpace(unsigned int AddressSpace,
-                           raw_ostream &O) const;
-  std::string getPTXFundamentalTypeStr(const Type *Ty, bool=true) const ;
-  void printScalarConstant(Constant *CPV, raw_ostream &O) ;
-  void printFPConstant(const ConstantFP *Fp, raw_ostream &O) ;
-  void bufferLEByte(Constant *CPV, int Bytes, AggBuffer *aggBuffer) ;
-  void bufferAggregateConstant(Constant *CV, AggBuffer *aggBuffer) ;
+  void emitPTXAddressSpace(unsigned int AddressSpace, raw_ostream &O) const;
+  std::string getPTXFundamentalTypeStr(const Type *Ty, bool = true) const;
+  void printScalarConstant(Constant *CPV, raw_ostream &O);
+  void printFPConstant(const ConstantFP *Fp, raw_ostream &O);
+  void bufferLEByte(Constant *CPV, int Bytes, AggBuffer *aggBuffer);
+  void bufferAggregateConstant(Constant *CV, AggBuffer *aggBuffer);
 
   void printOperandProper(const MachineOperand &MO);
 
-  void emitLinkageDirective(const GlobalValue* V, raw_ostream &O);
+  void emitLinkageDirective(const GlobalValue *V, raw_ostream &O);
   void emitDeclarations(Module &, raw_ostream &O);
   void emitDeclaration(const Function *, raw_ostream &O);
 
@@ -289,10 +273,9 @@ private:
   LineReader *reader;
   LineReader *getReader(std::string);
 public:
-  NVPTXAsmPrinter(TargetMachine &TM,
-                  MCStreamer &Streamer)
-  : AsmPrinter(TM, Streamer),
-    nvptxSubtarget(TM.getSubtarget<NVPTXSubtarget>()) {
+  NVPTXAsmPrinter(TargetMachine &TM, MCStreamer &Streamer)
+      : AsmPrinter(TM, Streamer),
+        nvptxSubtarget(TM.getSubtarget<NVPTXSubtarget>()) {
     CurrentBankselLabelInBasicBlock = "";
     VRidGlobal2LocalMap = NULL;
     reader = NULL;
diff --git a/lib/Target/NVPTX/NVPTXFrameLowering.cpp b/lib/Target/NVPTX/NVPTXFrameLowering.cpp
index a9abc00bf3f6..6533da5102b0 100644
--- a/lib/Target/NVPTX/NVPTXFrameLowering.cpp
+++ b/lib/Target/NVPTX/NVPTXFrameLowering.cpp
@@ -17,17 +17,15 @@
 #include "NVPTXSubtarget.h"
 #include "NVPTXTargetMachine.h"
 #include "llvm/ADT/BitVector.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/MC/MachineLocation.h"
 #include "llvm/Target/TargetInstrInfo.h"
 
 using namespace llvm;
 
-bool NVPTXFrameLowering::hasFP(const MachineFunction &MF) const {
-  return true;
-}
+bool NVPTXFrameLowering::hasFP(const MachineFunction &MF) const { return true; }
 
 void NVPTXFrameLowering::emitPrologue(MachineFunction &MF) const {
   if (MF.getFrameInfo()->hasStackObjects()) {
@@ -42,35 +40,39 @@ void NVPTXFrameLowering::emitPrologue(MachineFunction &MF) const {
       // mov %SPL, %depot;
       // cvta.local %SP, %SPL;
       if (is64bit) {
-        MachineInstr *MI = BuildMI(MBB, MBBI, dl,
-                               tm.getInstrInfo()->get(NVPTX::cvta_local_yes_64),
-                                   NVPTX::VRFrame).addReg(NVPTX::VRFrameLocal);
-        BuildMI(MBB, MI, dl,
-                tm.getInstrInfo()->get(NVPTX::IMOV64rr), NVPTX::VRFrameLocal)
-        .addReg(NVPTX::VRDepot);
+        MachineInstr *MI = BuildMI(
+            MBB, MBBI, dl, tm.getInstrInfo()->get(NVPTX::cvta_local_yes_64),
+            NVPTX::VRFrame).addReg(NVPTX::VRFrameLocal);
+        BuildMI(MBB, MI, dl, tm.getInstrInfo()->get(NVPTX::IMOV64rr),
+                NVPTX::VRFrameLocal).addReg(NVPTX::VRDepot);
       } else {
-        MachineInstr *MI = BuildMI(MBB, MBBI, dl,
-                                  tm.getInstrInfo()->get(NVPTX::cvta_local_yes),
-                                   NVPTX::VRFrame).addReg(NVPTX::VRFrameLocal);
-        BuildMI(MBB, MI, dl,
-                tm.getInstrInfo()->get(NVPTX::IMOV32rr), NVPTX::VRFrameLocal)
-        .addReg(NVPTX::VRDepot);
+        MachineInstr *MI = BuildMI(
+            MBB, MBBI, dl, tm.getInstrInfo()->get(NVPTX::cvta_local_yes),
+            NVPTX::VRFrame).addReg(NVPTX::VRFrameLocal);
+        BuildMI(MBB, MI, dl, tm.getInstrInfo()->get(NVPTX::IMOV32rr),
+                NVPTX::VRFrameLocal).addReg(NVPTX::VRDepot);
       }
-    }
-    else {
+    } else {
       // mov %SP, %depot;
       if (is64bit)
-        BuildMI(MBB, MBBI, dl,
-                tm.getInstrInfo()->get(NVPTX::IMOV64rr), NVPTX::VRFrame)
-                .addReg(NVPTX::VRDepot);
+        BuildMI(MBB, MBBI, dl, tm.getInstrInfo()->get(NVPTX::IMOV64rr),
+                NVPTX::VRFrame).addReg(NVPTX::VRDepot);
       else
-        BuildMI(MBB, MBBI, dl,
-                tm.getInstrInfo()->get(NVPTX::IMOV32rr), NVPTX::VRFrame)
-                .addReg(NVPTX::VRDepot);
+        BuildMI(MBB, MBBI, dl, tm.getInstrInfo()->get(NVPTX::IMOV32rr),
+                NVPTX::VRFrame).addReg(NVPTX::VRDepot);
     }
   }
 }
 
 void NVPTXFrameLowering::emitEpilogue(MachineFunction &MF,
-                                      MachineBasicBlock &MBB) const {
+                                      MachineBasicBlock &MBB) const {}
+
+// This function eliminates ADJCALLSTACKDOWN,
+// ADJCALLSTACKUP pseudo instructions
+void NVPTXFrameLowering::eliminateCallFramePseudoInstr(
+    MachineFunction &MF, MachineBasicBlock &MBB,
+    MachineBasicBlock::iterator I) const {
+  // Simply discard ADJCALLSTACKDOWN,
+  // ADJCALLSTACKUP instructions.
+  MBB.erase(I);
 }
diff --git a/lib/Target/NVPTX/NVPTXFrameLowering.h b/lib/Target/NVPTX/NVPTXFrameLowering.h
index ee87b3997e78..819f1dd3f4be 100644
--- a/lib/Target/NVPTX/NVPTXFrameLowering.h
+++ b/lib/Target/NVPTX/NVPTXFrameLowering.h
@@ -16,7 +16,6 @@
 
 #include "llvm/Target/TargetFrameLowering.h"
 
-
 namespace llvm {
 class NVPTXTargetMachine;
 
@@ -26,13 +25,16 @@ class NVPTXFrameLowering : public TargetFrameLowering {
 
 public:
   explicit NVPTXFrameLowering(NVPTXTargetMachine &_tm, bool _is64bit)
-  : TargetFrameLowering(TargetFrameLowering::StackGrowsUp, 8, 0),
-    tm(_tm), is64bit(_is64bit) {}
+      : TargetFrameLowering(TargetFrameLowering::StackGrowsUp, 8, 0), tm(_tm),
+        is64bit(_is64bit) {}
 
   virtual bool hasFP(const MachineFunction &MF) const;
   virtual void emitPrologue(MachineFunction &MF) const;
-  virtual void emitEpilogue(MachineFunction &MF,
-                            MachineBasicBlock &MBB) const;
+  virtual void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
+
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I) const;
 };
 
 } // End llvm namespace
diff --git a/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp b/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp
index 4e92f0e785fe..e862988c85d1 100644
--- a/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp
+++ b/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp
@@ -11,42 +11,36 @@
 //
 //===----------------------------------------------------------------------===//
 
-
-#include "llvm/Instructions.h"
-#include "llvm/Support/raw_ostream.h"
 #include "NVPTXISelDAGToDAG.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetIntrinsicInfo.h"
-#include "llvm/GlobalValue.h"
 
 #undef DEBUG_TYPE
 #define DEBUG_TYPE "nvptx-isel"
 
 using namespace llvm;
 
-
-static cl::opt<bool>
-UseFMADInstruction("nvptx-mad-enable",
-                   cl::ZeroOrMore,
-                cl::desc("NVPTX Specific: Enable generating FMAD instructions"),
-                   cl::init(false));
+static cl::opt<bool> UseFMADInstruction(
+    "nvptx-mad-enable", cl::ZeroOrMore,
+    cl::desc("NVPTX Specific: Enable generating FMAD instructions"),
+    cl::init(false));
 
 static cl::opt<int>
-FMAContractLevel("nvptx-fma-level",
-                 cl::ZeroOrMore,
+FMAContractLevel("nvptx-fma-level", cl::ZeroOrMore,
                  cl::desc("NVPTX Specific: FMA contraction (0: don't do it"
-                     " 1: do it  2: do it aggressively"),
-                     cl::init(2));
+                          " 1: do it  2: do it aggressively"),
+                 cl::init(2));
 
-
-static cl::opt<int>
-UsePrecDivF32("nvptx-prec-divf32",
-              cl::ZeroOrMore,
-             cl::desc("NVPTX Specifies: 0 use div.approx, 1 use div.full, 2 use"
-                  " IEEE Compliant F32 div.rnd if avaiable."),
-                  cl::init(2));
+static cl::opt<int> UsePrecDivF32(
+    "nvptx-prec-divf32", cl::ZeroOrMore,
+    cl::desc("NVPTX Specifies: 0 use div.approx, 1 use div.full, 2 use"
+             " IEEE Compliant F32 div.rnd if avaiable."),
+    cl::init(2));
 
 /// createNVPTXISelDag - This pass converts a legalized DAG into a
 /// NVPTX-specific DAG, ready for instruction scheduling.
@@ -55,26 +49,22 @@ FunctionPass *llvm::createNVPTXISelDag(NVPTXTargetMachine &TM,
   return new NVPTXDAGToDAGISel(TM, OptLevel);
 }
 
-
 NVPTXDAGToDAGISel::NVPTXDAGToDAGISel(NVPTXTargetMachine &tm,
                                      CodeGenOpt::Level OptLevel)
-: SelectionDAGISel(tm, OptLevel),
-  Subtarget(tm.getSubtarget<NVPTXSubtarget>())
-{
+    : SelectionDAGISel(tm, OptLevel),
+      Subtarget(tm.getSubtarget<NVPTXSubtarget>()) {
   // Always do fma.f32 fpcontract if the target supports the instruction.
   // Always do fma.f64 fpcontract if the target supports the instruction.
   // Do mad.f32 is nvptx-mad-enable is specified and the target does not
   // support fma.f32.
 
   doFMADF32 = (OptLevel > 0) && UseFMADInstruction && !Subtarget.hasFMAF32();
-  doFMAF32 =  (OptLevel > 0) && Subtarget.hasFMAF32() &&
-      (FMAContractLevel>=1);
-  doFMAF64 =  (OptLevel > 0) && Subtarget.hasFMAF64() &&
-      (FMAContractLevel>=1);
-  doFMAF32AGG =  (OptLevel > 0) && Subtarget.hasFMAF32() &&
-      (FMAContractLevel==2);
-  doFMAF64AGG =  (OptLevel > 0) && Subtarget.hasFMAF64() &&
-      (FMAContractLevel==2);
+  doFMAF32 = (OptLevel > 0) && Subtarget.hasFMAF32() && (FMAContractLevel >= 1);
+  doFMAF64 = (OptLevel > 0) && Subtarget.hasFMAF64() && (FMAContractLevel >= 1);
+  doFMAF32AGG =
+      (OptLevel > 0) && Subtarget.hasFMAF32() && (FMAContractLevel == 2);
+  doFMAF64AGG =
+      (OptLevel > 0) && Subtarget.hasFMAF64() && (FMAContractLevel == 2);
 
   allowFMA = (FMAContractLevel >= 1) || UseFMADInstruction;
 
@@ -92,10 +82,10 @@ NVPTXDAGToDAGISel::NVPTXDAGToDAGISel(NVPTXTargetMachine &tm,
 
 /// Select - Select instructions not customized! Used for
 /// expanded, promoted and normal instructions.
-SDNode* NVPTXDAGToDAGISel::Select(SDNode *N) {
+SDNode *NVPTXDAGToDAGISel::Select(SDNode *N) {
 
   if (N->isMachineOpcode())
-    return NULL;   // Already selected.
+    return NULL; // Already selected.
 
   SDNode *ResNode = NULL;
   switch (N->getOpcode()) {
@@ -105,29 +95,48 @@ SDNode* NVPTXDAGToDAGISel::Select(SDNode *N) {
   case ISD::STORE:
     ResNode = SelectStore(N);
     break;
+  case NVPTXISD::LoadV2:
+  case NVPTXISD::LoadV4:
+    ResNode = SelectLoadVector(N);
+    break;
+  case NVPTXISD::LDGV2:
+  case NVPTXISD::LDGV4:
+  case NVPTXISD::LDUV2:
+  case NVPTXISD::LDUV4:
+    ResNode = SelectLDGLDUVector(N);
+    break;
+  case NVPTXISD::StoreV2:
+  case NVPTXISD::StoreV4:
+    ResNode = SelectStoreVector(N);
+    break;
+  default:
+    break;
   }
   if (ResNode)
     return ResNode;
   return SelectCode(N);
 }
 
-
-static unsigned int
-getCodeAddrSpace(MemSDNode *N, const NVPTXSubtarget &Subtarget)
-{
+static unsigned int getCodeAddrSpace(MemSDNode *N,
+                                     const NVPTXSubtarget &Subtarget) {
   const Value *Src = N->getSrcValue();
   if (!Src)
     return NVPTX::PTXLdStInstCode::LOCAL;
 
   if (const PointerType *PT = dyn_cast<PointerType>(Src->getType())) {
     switch (PT->getAddressSpace()) {
-    case llvm::ADDRESS_SPACE_LOCAL: return NVPTX::PTXLdStInstCode::LOCAL;
-    case llvm::ADDRESS_SPACE_GLOBAL: return NVPTX::PTXLdStInstCode::GLOBAL;
-    case llvm::ADDRESS_SPACE_SHARED: return NVPTX::PTXLdStInstCode::SHARED;
+    case llvm::ADDRESS_SPACE_LOCAL:
+      return NVPTX::PTXLdStInstCode::LOCAL;
+    case llvm::ADDRESS_SPACE_GLOBAL:
+      return NVPTX::PTXLdStInstCode::GLOBAL;
+    case llvm::ADDRESS_SPACE_SHARED:
+      return NVPTX::PTXLdStInstCode::SHARED;
     case llvm::ADDRESS_SPACE_CONST_NOT_GEN:
       return NVPTX::PTXLdStInstCode::CONSTANT;
-    case llvm::ADDRESS_SPACE_GENERIC: return NVPTX::PTXLdStInstCode::GENERIC;
-    case llvm::ADDRESS_SPACE_PARAM: return NVPTX::PTXLdStInstCode::PARAM;
+    case llvm::ADDRESS_SPACE_GENERIC:
+      return NVPTX::PTXLdStInstCode::GENERIC;
+    case llvm::ADDRESS_SPACE_PARAM:
+      return NVPTX::PTXLdStInstCode::PARAM;
     case llvm::ADDRESS_SPACE_CONST:
       // If the arch supports generic address space, translate it to GLOBAL
       // for correctness.
@@ -138,18 +147,18 @@ getCodeAddrSpace(MemSDNode *N, const NVPTXSubtarget &Subtarget)
         return NVPTX::PTXLdStInstCode::GLOBAL;
       else
         return NVPTX::PTXLdStInstCode::CONSTANT;
-    default: break;
+    default:
+      break;
     }
   }
   return NVPTX::PTXLdStInstCode::LOCAL;
 }
 
-
-SDNode* NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
+SDNode *NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
   DebugLoc dl = N->getDebugLoc();
   LoadSDNode *LD = cast<LoadSDNode>(N);
   EVT LoadedVT = LD->getMemoryVT();
-  SDNode *NVPTXLD= NULL;
+  SDNode *NVPTXLD = NULL;
 
   // do not support pre/post inc/dec
   if (LD->isIndexed())
@@ -189,7 +198,7 @@ SDNode* NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
   //          type is integer
   // Float  : ISD::NON_EXTLOAD or ISD::EXTLOAD and the type is float
   MVT ScalarVT = SimpleVT.getScalarType();
-  unsigned fromTypeWidth =  ScalarVT.getSizeInBits();
+  unsigned fromTypeWidth = ScalarVT.getSizeInBits();
   unsigned int fromType;
   if ((LD->getExtensionType() == ISD::SEXTLOAD))
     fromType = NVPTX::PTXLdStInstCode::Signed;
@@ -208,121 +217,166 @@ SDNode* NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
 
   if (SelectDirectAddr(N1, Addr)) {
     switch (TargetVT) {
-    case MVT::i8:    Opcode = NVPTX::LD_i8_avar; break;
-    case MVT::i16:   Opcode = NVPTX::LD_i16_avar; break;
-    case MVT::i32:   Opcode = NVPTX::LD_i32_avar; break;
-    case MVT::i64:   Opcode = NVPTX::LD_i64_avar; break;
-    case MVT::f32:   Opcode = NVPTX::LD_f32_avar; break;
-    case MVT::f64:   Opcode = NVPTX::LD_f64_avar; break;
-    case MVT::v2i8:  Opcode = NVPTX::LD_v2i8_avar; break;
-    case MVT::v2i16: Opcode = NVPTX::LD_v2i16_avar; break;
-    case MVT::v2i32: Opcode = NVPTX::LD_v2i32_avar; break;
-    case MVT::v2i64: Opcode = NVPTX::LD_v2i64_avar; break;
-    case MVT::v2f32: Opcode = NVPTX::LD_v2f32_avar; break;
-    case MVT::v2f64: Opcode = NVPTX::LD_v2f64_avar; break;
-    case MVT::v4i8:  Opcode = NVPTX::LD_v4i8_avar; break;
-    case MVT::v4i16: Opcode = NVPTX::LD_v4i16_avar; break;
-    case MVT::v4i32: Opcode = NVPTX::LD_v4i32_avar; break;
-    case MVT::v4f32: Opcode = NVPTX::LD_v4f32_avar; break;
-    default: return NULL;
+    case MVT::i8:
+      Opcode = NVPTX::LD_i8_avar;
+      break;
+    case MVT::i16:
+      Opcode = NVPTX::LD_i16_avar;
+      break;
+    case MVT::i32:
+      Opcode = NVPTX::LD_i32_avar;
+      break;
+    case MVT::i64:
+      Opcode = NVPTX::LD_i64_avar;
+      break;
+    case MVT::f32:
+      Opcode = NVPTX::LD_f32_avar;
+      break;
+    case MVT::f64:
+      Opcode = NVPTX::LD_f64_avar;
+      break;
+    default:
+      return NULL;
     }
-    SDValue Ops[] = { getI32Imm(isVolatile),
-                      getI32Imm(codeAddrSpace),
-                      getI32Imm(vecType),
-                      getI32Imm(fromType),
-                      getI32Imm(fromTypeWidth),
-                      Addr, Chain };
-    NVPTXLD = CurDAG->getMachineNode(Opcode, dl, TargetVT,
-                                     MVT::Other, Ops, 7);
-  } else if (Subtarget.is64Bit()?
-      SelectADDRsi64(N1.getNode(), N1, Base, Offset):
-      SelectADDRsi(N1.getNode(), N1, Base, Offset)) {
+    SDValue Ops[] = { getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
+                      getI32Imm(vecType), getI32Imm(fromType),
+                      getI32Imm(fromTypeWidth), Addr, Chain };
+    NVPTXLD = CurDAG->getMachineNode(Opcode, dl, TargetVT, MVT::Other, Ops, 7);
+  } else if (Subtarget.is64Bit()
+                 ? SelectADDRsi64(N1.getNode(), N1, Base, Offset)
+                 : SelectADDRsi(N1.getNode(), N1, Base, Offset)) {
     switch (TargetVT) {
-    case MVT::i8:    Opcode = NVPTX::LD_i8_asi; break;
-    case MVT::i16:   Opcode = NVPTX::LD_i16_asi; break;
-    case MVT::i32:   Opcode = NVPTX::LD_i32_asi; break;
-    case MVT::i64:   Opcode = NVPTX::LD_i64_asi; break;
-    case MVT::f32:   Opcode = NVPTX::LD_f32_asi; break;
-    case MVT::f64:   Opcode = NVPTX::LD_f64_asi; break;
-    case MVT::v2i8:  Opcode = NVPTX::LD_v2i8_asi; break;
-    case MVT::v2i16: Opcode = NVPTX::LD_v2i16_asi; break;
-    case MVT::v2i32: Opcode = NVPTX::LD_v2i32_asi; break;
-    case MVT::v2i64: Opcode = NVPTX::LD_v2i64_asi; break;
-    case MVT::v2f32: Opcode = NVPTX::LD_v2f32_asi; break;
-    case MVT::v2f64: Opcode = NVPTX::LD_v2f64_asi; break;
-    case MVT::v4i8:  Opcode = NVPTX::LD_v4i8_asi; break;
-    case MVT::v4i16: Opcode = NVPTX::LD_v4i16_asi; break;
-    case MVT::v4i32: Opcode = NVPTX::LD_v4i32_asi; break;
-    case MVT::v4f32: Opcode = NVPTX::LD_v4f32_asi; break;
-    default: return NULL;
+    case MVT::i8:
+      Opcode = NVPTX::LD_i8_asi;
+      break;
+    case MVT::i16:
+      Opcode = NVPTX::LD_i16_asi;
+      break;
+    case MVT::i32:
+      Opcode = NVPTX::LD_i32_asi;
+      break;
+    case MVT::i64:
+      Opcode = NVPTX::LD_i64_asi;
+      break;
+    case MVT::f32:
+      Opcode = NVPTX::LD_f32_asi;
+      break;
+    case MVT::f64:
+      Opcode = NVPTX::LD_f64_asi;
+      break;
+    default:
+      return NULL;
     }
-    SDValue Ops[] = { getI32Imm(isVolatile),
-                      getI32Imm(codeAddrSpace),
-                      getI32Imm(vecType),
-                      getI32Imm(fromType),
-                      getI32Imm(fromTypeWidth),
-                      Base, Offset, Chain };
-    NVPTXLD = CurDAG->getMachineNode(Opcode, dl, TargetVT,
-                                     MVT::Other, Ops, 8);
-  } else if (Subtarget.is64Bit()?
-      SelectADDRri64(N1.getNode(), N1, Base, Offset):
-      SelectADDRri(N1.getNode(), N1, Base, Offset)) {
-    switch (TargetVT) {
-    case MVT::i8:    Opcode = NVPTX::LD_i8_ari; break;
-    case MVT::i16:   Opcode = NVPTX::LD_i16_ari; break;
-    case MVT::i32:   Opcode = NVPTX::LD_i32_ari; break;
-    case MVT::i64:   Opcode = NVPTX::LD_i64_ari; break;
-    case MVT::f32:   Opcode = NVPTX::LD_f32_ari; break;
-    case MVT::f64:   Opcode = NVPTX::LD_f64_ari; break;
-    case MVT::v2i8:  Opcode = NVPTX::LD_v2i8_ari; break;
-    case MVT::v2i16: Opcode = NVPTX::LD_v2i16_ari; break;
-    case MVT::v2i32: Opcode = NVPTX::LD_v2i32_ari; break;
-    case MVT::v2i64: Opcode = NVPTX::LD_v2i64_ari; break;
-    case MVT::v2f32: Opcode = NVPTX::LD_v2f32_ari; break;
-    case MVT::v2f64: Opcode = NVPTX::LD_v2f64_ari; break;
-    case MVT::v4i8:  Opcode = NVPTX::LD_v4i8_ari; break;
-    case MVT::v4i16: Opcode = NVPTX::LD_v4i16_ari; break;
-    case MVT::v4i32: Opcode = NVPTX::LD_v4i32_ari; break;
-    case MVT::v4f32: Opcode = NVPTX::LD_v4f32_ari; break;
-    default: return NULL;
+    SDValue Ops[] = { getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
+                      getI32Imm(vecType), getI32Imm(fromType),
+                      getI32Imm(fromTypeWidth), Base, Offset, Chain };
+    NVPTXLD = CurDAG->getMachineNode(Opcode, dl, TargetVT, MVT::Other, Ops, 8);
+  } else if (Subtarget.is64Bit()
+                 ? SelectADDRri64(N1.getNode(), N1, Base, Offset)
+                 : SelectADDRri(N1.getNode(), N1, Base, Offset)) {
+    if (Subtarget.is64Bit()) {
+      switch (TargetVT) {
+      case MVT::i8:
+        Opcode = NVPTX::LD_i8_ari_64;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::LD_i16_ari_64;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::LD_i32_ari_64;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::LD_i64_ari_64;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::LD_f32_ari_64;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::LD_f64_ari_64;
+        break;
+      default:
+        return NULL;
+      }
+    } else {
+      switch (TargetVT) {
+      case MVT::i8:
+        Opcode = NVPTX::LD_i8_ari;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::LD_i16_ari;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::LD_i32_ari;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::LD_i64_ari;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::LD_f32_ari;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::LD_f64_ari;
+        break;
+      default:
+        return NULL;
+      }
     }
-    SDValue Ops[] = { getI32Imm(isVolatile),
-                      getI32Imm(codeAddrSpace),
-                      getI32Imm(vecType),
-                      getI32Imm(fromType),
-                      getI32Imm(fromTypeWidth),
-                      Base, Offset, Chain };
-    NVPTXLD = CurDAG->getMachineNode(Opcode, dl, TargetVT,
-                                     MVT::Other, Ops, 8);
-  }
-  else {
-    switch (TargetVT) {
-    case MVT::i8:    Opcode = NVPTX::LD_i8_areg; break;
-    case MVT::i16:   Opcode = NVPTX::LD_i16_areg; break;
-    case MVT::i32:   Opcode = NVPTX::LD_i32_areg; break;
-    case MVT::i64:   Opcode = NVPTX::LD_i64_areg; break;
-    case MVT::f32:   Opcode = NVPTX::LD_f32_areg; break;
-    case MVT::f64:   Opcode = NVPTX::LD_f64_areg; break;
-    case MVT::v2i8:  Opcode = NVPTX::LD_v2i8_areg; break;
-    case MVT::v2i16: Opcode = NVPTX::LD_v2i16_areg; break;
-    case MVT::v2i32: Opcode = NVPTX::LD_v2i32_areg; break;
-    case MVT::v2i64: Opcode = NVPTX::LD_v2i64_areg; break;
-    case MVT::v2f32: Opcode = NVPTX::LD_v2f32_areg; break;
-    case MVT::v2f64: Opcode = NVPTX::LD_v2f64_areg; break;
-    case MVT::v4i8:  Opcode = NVPTX::LD_v4i8_areg; break;
-    case MVT::v4i16: Opcode = NVPTX::LD_v4i16_areg; break;
-    case MVT::v4i32: Opcode = NVPTX::LD_v4i32_areg; break;
-    case MVT::v4f32: Opcode = NVPTX::LD_v4f32_areg; break;
-    default: return NULL;
+    SDValue Ops[] = { getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
+                      getI32Imm(vecType), getI32Imm(fromType),
+                      getI32Imm(fromTypeWidth), Base, Offset, Chain };
+    NVPTXLD = CurDAG->getMachineNode(Opcode, dl, TargetVT, MVT::Other, Ops, 8);
+  } else {
+    if (Subtarget.is64Bit()) {
+      switch (TargetVT) {
+      case MVT::i8:
+        Opcode = NVPTX::LD_i8_areg_64;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::LD_i16_areg_64;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::LD_i32_areg_64;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::LD_i64_areg_64;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::LD_f32_areg_64;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::LD_f64_areg_64;
+        break;
+      default:
+        return NULL;
+      }
+    } else {
+      switch (TargetVT) {
+      case MVT::i8:
+        Opcode = NVPTX::LD_i8_areg;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::LD_i16_areg;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::LD_i32_areg;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::LD_i64_areg;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::LD_f32_areg;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::LD_f64_areg;
+        break;
+      default:
+        return NULL;
+      }
     }
-    SDValue Ops[] = { getI32Imm(isVolatile),
-                      getI32Imm(codeAddrSpace),
-                      getI32Imm(vecType),
-                      getI32Imm(fromType),
-                      getI32Imm(fromTypeWidth),
-                      N1, Chain };
-    NVPTXLD = CurDAG->getMachineNode(Opcode, dl, TargetVT,
-                                     MVT::Other, Ops, 7);
+    SDValue Ops[] = { getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
+                      getI32Imm(vecType), getI32Imm(fromType),
+                      getI32Imm(fromTypeWidth), N1, Chain };
+    NVPTXLD = CurDAG->getMachineNode(Opcode, dl, TargetVT, MVT::Other, Ops, 7);
   }
 
   if (NVPTXLD != NULL) {
@@ -334,7 +388,590 @@ SDNode* NVPTXDAGToDAGISel::SelectLoad(SDNode *N) {
   return NVPTXLD;
 }
 
-SDNode* NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
+SDNode *NVPTXDAGToDAGISel::SelectLoadVector(SDNode *N) {
+
+  SDValue Chain = N->getOperand(0);
+  SDValue Op1 = N->getOperand(1);
+  SDValue Addr, Offset, Base;
+  unsigned Opcode;
+  DebugLoc DL = N->getDebugLoc();
+  SDNode *LD;
+  MemSDNode *MemSD = cast<MemSDNode>(N);
+  EVT LoadedVT = MemSD->getMemoryVT();
+
+  if (!LoadedVT.isSimple())
+    return NULL;
+
+  // Address Space Setting
+  unsigned int CodeAddrSpace = getCodeAddrSpace(MemSD, Subtarget);
+
+  // Volatile Setting
+  // - .volatile is only availalble for .global and .shared
+  bool IsVolatile = MemSD->isVolatile();
+  if (CodeAddrSpace != NVPTX::PTXLdStInstCode::GLOBAL &&
+      CodeAddrSpace != NVPTX::PTXLdStInstCode::SHARED &&
+      CodeAddrSpace != NVPTX::PTXLdStInstCode::GENERIC)
+    IsVolatile = false;
+
+  // Vector Setting
+  MVT SimpleVT = LoadedVT.getSimpleVT();
+
+  // Type Setting: fromType + fromTypeWidth
+  //
+  // Sign   : ISD::SEXTLOAD
+  // Unsign : ISD::ZEXTLOAD, ISD::NON_EXTLOAD or ISD::EXTLOAD and the
+  //          type is integer
+  // Float  : ISD::NON_EXTLOAD or ISD::EXTLOAD and the type is float
+  MVT ScalarVT = SimpleVT.getScalarType();
+  unsigned FromTypeWidth = ScalarVT.getSizeInBits();
+  unsigned int FromType;
+  // The last operand holds the original LoadSDNode::getExtensionType() value
+  unsigned ExtensionType = cast<ConstantSDNode>(
+      N->getOperand(N->getNumOperands() - 1))->getZExtValue();
+  if (ExtensionType == ISD::SEXTLOAD)
+    FromType = NVPTX::PTXLdStInstCode::Signed;
+  else if (ScalarVT.isFloatingPoint())
+    FromType = NVPTX::PTXLdStInstCode::Float;
+  else
+    FromType = NVPTX::PTXLdStInstCode::Unsigned;
+
+  unsigned VecType;
+
+  switch (N->getOpcode()) {
+  case NVPTXISD::LoadV2:
+    VecType = NVPTX::PTXLdStInstCode::V2;
+    break;
+  case NVPTXISD::LoadV4:
+    VecType = NVPTX::PTXLdStInstCode::V4;
+    break;
+  default:
+    return NULL;
+  }
+
+  EVT EltVT = N->getValueType(0);
+
+  if (SelectDirectAddr(Op1, Addr)) {
+    switch (N->getOpcode()) {
+    default:
+      return NULL;
+    case NVPTXISD::LoadV2:
+      switch (EltVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::LDV_i8_v2_avar;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::LDV_i16_v2_avar;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::LDV_i32_v2_avar;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::LDV_i64_v2_avar;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::LDV_f32_v2_avar;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::LDV_f64_v2_avar;
+        break;
+      }
+      break;
+    case NVPTXISD::LoadV4:
+      switch (EltVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::LDV_i8_v4_avar;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::LDV_i16_v4_avar;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::LDV_i32_v4_avar;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::LDV_f32_v4_avar;
+        break;
+      }
+      break;
+    }
+
+    SDValue Ops[] = { getI32Imm(IsVolatile), getI32Imm(CodeAddrSpace),
+                      getI32Imm(VecType), getI32Imm(FromType),
+                      getI32Imm(FromTypeWidth), Addr, Chain };
+    LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(), Ops, 7);
+  } else if (Subtarget.is64Bit()
+                 ? SelectADDRsi64(Op1.getNode(), Op1, Base, Offset)
+                 : SelectADDRsi(Op1.getNode(), Op1, Base, Offset)) {
+    switch (N->getOpcode()) {
+    default:
+      return NULL;
+    case NVPTXISD::LoadV2:
+      switch (EltVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::LDV_i8_v2_asi;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::LDV_i16_v2_asi;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::LDV_i32_v2_asi;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::LDV_i64_v2_asi;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::LDV_f32_v2_asi;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::LDV_f64_v2_asi;
+        break;
+      }
+      break;
+    case NVPTXISD::LoadV4:
+      switch (EltVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::LDV_i8_v4_asi;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::LDV_i16_v4_asi;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::LDV_i32_v4_asi;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::LDV_f32_v4_asi;
+        break;
+      }
+      break;
+    }
+
+    SDValue Ops[] = { getI32Imm(IsVolatile), getI32Imm(CodeAddrSpace),
+                      getI32Imm(VecType), getI32Imm(FromType),
+                      getI32Imm(FromTypeWidth), Base, Offset, Chain };
+    LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(), Ops, 8);
+  } else if (Subtarget.is64Bit()
+                 ? SelectADDRri64(Op1.getNode(), Op1, Base, Offset)
+                 : SelectADDRri(Op1.getNode(), Op1, Base, Offset)) {
+    if (Subtarget.is64Bit()) {
+      switch (N->getOpcode()) {
+      default:
+        return NULL;
+      case NVPTXISD::LoadV2:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::LDV_i8_v2_ari_64;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::LDV_i16_v2_ari_64;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::LDV_i32_v2_ari_64;
+          break;
+        case MVT::i64:
+          Opcode = NVPTX::LDV_i64_v2_ari_64;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::LDV_f32_v2_ari_64;
+          break;
+        case MVT::f64:
+          Opcode = NVPTX::LDV_f64_v2_ari_64;
+          break;
+        }
+        break;
+      case NVPTXISD::LoadV4:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::LDV_i8_v4_ari_64;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::LDV_i16_v4_ari_64;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::LDV_i32_v4_ari_64;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::LDV_f32_v4_ari_64;
+          break;
+        }
+        break;
+      }
+    } else {
+      switch (N->getOpcode()) {
+      default:
+        return NULL;
+      case NVPTXISD::LoadV2:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::LDV_i8_v2_ari;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::LDV_i16_v2_ari;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::LDV_i32_v2_ari;
+          break;
+        case MVT::i64:
+          Opcode = NVPTX::LDV_i64_v2_ari;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::LDV_f32_v2_ari;
+          break;
+        case MVT::f64:
+          Opcode = NVPTX::LDV_f64_v2_ari;
+          break;
+        }
+        break;
+      case NVPTXISD::LoadV4:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::LDV_i8_v4_ari;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::LDV_i16_v4_ari;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::LDV_i32_v4_ari;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::LDV_f32_v4_ari;
+          break;
+        }
+        break;
+      }
+    }
+
+    SDValue Ops[] = { getI32Imm(IsVolatile), getI32Imm(CodeAddrSpace),
+                      getI32Imm(VecType), getI32Imm(FromType),
+                      getI32Imm(FromTypeWidth), Base, Offset, Chain };
+
+    LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(), Ops, 8);
+  } else {
+    if (Subtarget.is64Bit()) {
+      switch (N->getOpcode()) {
+      default:
+        return NULL;
+      case NVPTXISD::LoadV2:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::LDV_i8_v2_areg_64;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::LDV_i16_v2_areg_64;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::LDV_i32_v2_areg_64;
+          break;
+        case MVT::i64:
+          Opcode = NVPTX::LDV_i64_v2_areg_64;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::LDV_f32_v2_areg_64;
+          break;
+        case MVT::f64:
+          Opcode = NVPTX::LDV_f64_v2_areg_64;
+          break;
+        }
+        break;
+      case NVPTXISD::LoadV4:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::LDV_i8_v4_areg_64;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::LDV_i16_v4_areg_64;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::LDV_i32_v4_areg_64;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::LDV_f32_v4_areg_64;
+          break;
+        }
+        break;
+      }
+    } else {
+      switch (N->getOpcode()) {
+      default:
+        return NULL;
+      case NVPTXISD::LoadV2:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::LDV_i8_v2_areg;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::LDV_i16_v2_areg;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::LDV_i32_v2_areg;
+          break;
+        case MVT::i64:
+          Opcode = NVPTX::LDV_i64_v2_areg;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::LDV_f32_v2_areg;
+          break;
+        case MVT::f64:
+          Opcode = NVPTX::LDV_f64_v2_areg;
+          break;
+        }
+        break;
+      case NVPTXISD::LoadV4:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::LDV_i8_v4_areg;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::LDV_i16_v4_areg;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::LDV_i32_v4_areg;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::LDV_f32_v4_areg;
+          break;
+        }
+        break;
+      }
+    }
+
+    SDValue Ops[] = { getI32Imm(IsVolatile), getI32Imm(CodeAddrSpace),
+                      getI32Imm(VecType), getI32Imm(FromType),
+                      getI32Imm(FromTypeWidth), Op1, Chain };
+    LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(), Ops, 7);
+  }
+
+  MachineSDNode::mmo_iterator MemRefs0 = MF->allocateMemRefsArray(1);
+  MemRefs0[0] = cast<MemSDNode>(N)->getMemOperand();
+  cast<MachineSDNode>(LD)->setMemRefs(MemRefs0, MemRefs0 + 1);
+
+  return LD;
+}
+
+SDNode *NVPTXDAGToDAGISel::SelectLDGLDUVector(SDNode *N) {
+
+  SDValue Chain = N->getOperand(0);
+  SDValue Op1 = N->getOperand(1);
+  unsigned Opcode;
+  DebugLoc DL = N->getDebugLoc();
+  SDNode *LD;
+
+  EVT RetVT = N->getValueType(0);
+
+  // Select opcode
+  if (Subtarget.is64Bit()) {
+    switch (N->getOpcode()) {
+    default:
+      return NULL;
+    case NVPTXISD::LDGV2:
+      switch (RetVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2i8_ELE_64;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2i16_ELE_64;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2i32_ELE_64;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2i64_ELE_64;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2f32_ELE_64;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2f64_ELE_64;
+        break;
+      }
+      break;
+    case NVPTXISD::LDGV4:
+      switch (RetVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::INT_PTX_LDG_G_v4i8_ELE_64;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::INT_PTX_LDG_G_v4i16_ELE_64;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::INT_PTX_LDG_G_v4i32_ELE_64;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::INT_PTX_LDG_G_v4f32_ELE_64;
+        break;
+      }
+      break;
+    case NVPTXISD::LDUV2:
+      switch (RetVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2i8_ELE_64;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2i16_ELE_64;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2i32_ELE_64;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2i64_ELE_64;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2f32_ELE_64;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2f64_ELE_64;
+        break;
+      }
+      break;
+    case NVPTXISD::LDUV4:
+      switch (RetVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::INT_PTX_LDU_G_v4i8_ELE_64;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::INT_PTX_LDU_G_v4i16_ELE_64;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::INT_PTX_LDU_G_v4i32_ELE_64;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::INT_PTX_LDU_G_v4f32_ELE_64;
+        break;
+      }
+      break;
+    }
+  } else {
+    switch (N->getOpcode()) {
+    default:
+      return NULL;
+    case NVPTXISD::LDGV2:
+      switch (RetVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2i8_ELE_32;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2i16_ELE_32;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2i32_ELE_32;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2i64_ELE_32;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2f32_ELE_32;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::INT_PTX_LDG_G_v2f64_ELE_32;
+        break;
+      }
+      break;
+    case NVPTXISD::LDGV4:
+      switch (RetVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::INT_PTX_LDG_G_v4i8_ELE_32;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::INT_PTX_LDG_G_v4i16_ELE_32;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::INT_PTX_LDG_G_v4i32_ELE_32;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::INT_PTX_LDG_G_v4f32_ELE_32;
+        break;
+      }
+      break;
+    case NVPTXISD::LDUV2:
+      switch (RetVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2i8_ELE_32;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2i16_ELE_32;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2i32_ELE_32;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2i64_ELE_32;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2f32_ELE_32;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::INT_PTX_LDU_G_v2f64_ELE_32;
+        break;
+      }
+      break;
+    case NVPTXISD::LDUV4:
+      switch (RetVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::INT_PTX_LDU_G_v4i8_ELE_32;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::INT_PTX_LDU_G_v4i16_ELE_32;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::INT_PTX_LDU_G_v4i32_ELE_32;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::INT_PTX_LDU_G_v4f32_ELE_32;
+        break;
+      }
+      break;
+    }
+  }
+
+  SDValue Ops[] = { Op1, Chain };
+  LD = CurDAG->getMachineNode(Opcode, DL, N->getVTList(), &Ops[0], 2);
+
+  MachineSDNode::mmo_iterator MemRefs0 = MF->allocateMemRefsArray(1);
+  MemRefs0[0] = cast<MemSDNode>(N)->getMemOperand();
+  cast<MachineSDNode>(LD)->setMemRefs(MemRefs0, MemRefs0 + 1);
+
+  return LD;
+}
+
+SDNode *NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
   DebugLoc dl = N->getDebugLoc();
   StoreSDNode *ST = cast<StoreSDNode>(N);
   EVT StoreVT = ST->getMemoryVT();
@@ -375,7 +1012,7 @@ SDNode* NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
   // - for integer type, always use 'u'
   //
   MVT ScalarVT = SimpleVT.getScalarType();
-  unsigned toTypeWidth =  ScalarVT.getSizeInBits();
+  unsigned toTypeWidth = ScalarVT.getSizeInBits();
   unsigned int toType;
   if (ScalarVT.isFloatingPoint())
     toType = NVPTX::PTXLdStInstCode::Float;
@@ -394,124 +1031,166 @@ SDNode* NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
 
   if (SelectDirectAddr(N2, Addr)) {
     switch (SourceVT) {
-    case MVT::i8:    Opcode = NVPTX::ST_i8_avar; break;
-    case MVT::i16:   Opcode = NVPTX::ST_i16_avar; break;
-    case MVT::i32:   Opcode = NVPTX::ST_i32_avar; break;
-    case MVT::i64:   Opcode = NVPTX::ST_i64_avar; break;
-    case MVT::f32:   Opcode = NVPTX::ST_f32_avar; break;
-    case MVT::f64:   Opcode = NVPTX::ST_f64_avar; break;
-    case MVT::v2i8:  Opcode = NVPTX::ST_v2i8_avar; break;
-    case MVT::v2i16: Opcode = NVPTX::ST_v2i16_avar; break;
-    case MVT::v2i32: Opcode = NVPTX::ST_v2i32_avar; break;
-    case MVT::v2i64: Opcode = NVPTX::ST_v2i64_avar; break;
-    case MVT::v2f32: Opcode = NVPTX::ST_v2f32_avar; break;
-    case MVT::v2f64: Opcode = NVPTX::ST_v2f64_avar; break;
-    case MVT::v4i8:  Opcode = NVPTX::ST_v4i8_avar; break;
-    case MVT::v4i16: Opcode = NVPTX::ST_v4i16_avar; break;
-    case MVT::v4i32: Opcode = NVPTX::ST_v4i32_avar; break;
-    case MVT::v4f32: Opcode = NVPTX::ST_v4f32_avar; break;
-    default: return NULL;
+    case MVT::i8:
+      Opcode = NVPTX::ST_i8_avar;
+      break;
+    case MVT::i16:
+      Opcode = NVPTX::ST_i16_avar;
+      break;
+    case MVT::i32:
+      Opcode = NVPTX::ST_i32_avar;
+      break;
+    case MVT::i64:
+      Opcode = NVPTX::ST_i64_avar;
+      break;
+    case MVT::f32:
+      Opcode = NVPTX::ST_f32_avar;
+      break;
+    case MVT::f64:
+      Opcode = NVPTX::ST_f64_avar;
+      break;
+    default:
+      return NULL;
     }
-    SDValue Ops[] = { N1,
-                      getI32Imm(isVolatile),
-                      getI32Imm(codeAddrSpace),
-                      getI32Imm(vecType),
-                      getI32Imm(toType),
-                      getI32Imm(toTypeWidth),
-                      Addr, Chain };
-    NVPTXST = CurDAG->getMachineNode(Opcode, dl,
-                                     MVT::Other, Ops, 8);
-  } else if (Subtarget.is64Bit()?
-      SelectADDRsi64(N2.getNode(), N2, Base, Offset):
-      SelectADDRsi(N2.getNode(), N2, Base, Offset)) {
+    SDValue Ops[] = { N1, getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
+                      getI32Imm(vecType), getI32Imm(toType),
+                      getI32Imm(toTypeWidth), Addr, Chain };
+    NVPTXST = CurDAG->getMachineNode(Opcode, dl, MVT::Other, Ops, 8);
+  } else if (Subtarget.is64Bit()
+                 ? SelectADDRsi64(N2.getNode(), N2, Base, Offset)
+                 : SelectADDRsi(N2.getNode(), N2, Base, Offset)) {
     switch (SourceVT) {
-    case MVT::i8:    Opcode = NVPTX::ST_i8_asi; break;
-    case MVT::i16:   Opcode = NVPTX::ST_i16_asi; break;
-    case MVT::i32:   Opcode = NVPTX::ST_i32_asi; break;
-    case MVT::i64:   Opcode = NVPTX::ST_i64_asi; break;
-    case MVT::f32:   Opcode = NVPTX::ST_f32_asi; break;
-    case MVT::f64:   Opcode = NVPTX::ST_f64_asi; break;
-    case MVT::v2i8:  Opcode = NVPTX::ST_v2i8_asi; break;
-    case MVT::v2i16: Opcode = NVPTX::ST_v2i16_asi; break;
-    case MVT::v2i32: Opcode = NVPTX::ST_v2i32_asi; break;
-    case MVT::v2i64: Opcode = NVPTX::ST_v2i64_asi; break;
-    case MVT::v2f32: Opcode = NVPTX::ST_v2f32_asi; break;
-    case MVT::v2f64: Opcode = NVPTX::ST_v2f64_asi; break;
-    case MVT::v4i8:  Opcode = NVPTX::ST_v4i8_asi; break;
-    case MVT::v4i16: Opcode = NVPTX::ST_v4i16_asi; break;
-    case MVT::v4i32: Opcode = NVPTX::ST_v4i32_asi; break;
-    case MVT::v4f32: Opcode = NVPTX::ST_v4f32_asi; break;
-    default: return NULL;
+    case MVT::i8:
+      Opcode = NVPTX::ST_i8_asi;
+      break;
+    case MVT::i16:
+      Opcode = NVPTX::ST_i16_asi;
+      break;
+    case MVT::i32:
+      Opcode = NVPTX::ST_i32_asi;
+      break;
+    case MVT::i64:
+      Opcode = NVPTX::ST_i64_asi;
+      break;
+    case MVT::f32:
+      Opcode = NVPTX::ST_f32_asi;
+      break;
+    case MVT::f64:
+      Opcode = NVPTX::ST_f64_asi;
+      break;
+    default:
+      return NULL;
     }
-    SDValue Ops[] = { N1,
-                      getI32Imm(isVolatile),
-                      getI32Imm(codeAddrSpace),
-                      getI32Imm(vecType),
-                      getI32Imm(toType),
-                      getI32Imm(toTypeWidth),
-                      Base, Offset, Chain };
-    NVPTXST = CurDAG->getMachineNode(Opcode, dl,
-                                     MVT::Other, Ops, 9);
-  } else if (Subtarget.is64Bit()?
-      SelectADDRri64(N2.getNode(), N2, Base, Offset):
-      SelectADDRri(N2.getNode(), N2, Base, Offset)) {
-    switch (SourceVT) {
-    case MVT::i8:    Opcode = NVPTX::ST_i8_ari; break;
-    case MVT::i16:   Opcode = NVPTX::ST_i16_ari; break;
-    case MVT::i32:   Opcode = NVPTX::ST_i32_ari; break;
-    case MVT::i64:   Opcode = NVPTX::ST_i64_ari; break;
-    case MVT::f32:   Opcode = NVPTX::ST_f32_ari; break;
-    case MVT::f64:   Opcode = NVPTX::ST_f64_ari; break;
-    case MVT::v2i8:  Opcode = NVPTX::ST_v2i8_ari; break;
-    case MVT::v2i16: Opcode = NVPTX::ST_v2i16_ari; break;
-    case MVT::v2i32: Opcode = NVPTX::ST_v2i32_ari; break;
-    case MVT::v2i64: Opcode = NVPTX::ST_v2i64_ari; break;
-    case MVT::v2f32: Opcode = NVPTX::ST_v2f32_ari; break;
-    case MVT::v2f64: Opcode = NVPTX::ST_v2f64_ari; break;
-    case MVT::v4i8:  Opcode = NVPTX::ST_v4i8_ari; break;
-    case MVT::v4i16: Opcode = NVPTX::ST_v4i16_ari; break;
-    case MVT::v4i32: Opcode = NVPTX::ST_v4i32_ari; break;
-    case MVT::v4f32: Opcode = NVPTX::ST_v4f32_ari; break;
-    default: return NULL;
+    SDValue Ops[] = { N1, getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
+                      getI32Imm(vecType), getI32Imm(toType),
+                      getI32Imm(toTypeWidth), Base, Offset, Chain };
+    NVPTXST = CurDAG->getMachineNode(Opcode, dl, MVT::Other, Ops, 9);
+  } else if (Subtarget.is64Bit()
+                 ? SelectADDRri64(N2.getNode(), N2, Base, Offset)
+                 : SelectADDRri(N2.getNode(), N2, Base, Offset)) {
+    if (Subtarget.is64Bit()) {
+      switch (SourceVT) {
+      case MVT::i8:
+        Opcode = NVPTX::ST_i8_ari_64;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::ST_i16_ari_64;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::ST_i32_ari_64;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::ST_i64_ari_64;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::ST_f32_ari_64;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::ST_f64_ari_64;
+        break;
+      default:
+        return NULL;
+      }
+    } else {
+      switch (SourceVT) {
+      case MVT::i8:
+        Opcode = NVPTX::ST_i8_ari;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::ST_i16_ari;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::ST_i32_ari;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::ST_i64_ari;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::ST_f32_ari;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::ST_f64_ari;
+        break;
+      default:
+        return NULL;
+      }
     }
-    SDValue Ops[] = { N1,
-                      getI32Imm(isVolatile),
-                      getI32Imm(codeAddrSpace),
-                      getI32Imm(vecType),
-                      getI32Imm(toType),
-                      getI32Imm(toTypeWidth),
-                      Base, Offset, Chain };
-    NVPTXST = CurDAG->getMachineNode(Opcode, dl,
-                                     MVT::Other, Ops, 9);
+    SDValue Ops[] = { N1, getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
+                      getI32Imm(vecType), getI32Imm(toType),
+                      getI32Imm(toTypeWidth), Base, Offset, Chain };
+    NVPTXST = CurDAG->getMachineNode(Opcode, dl, MVT::Other, Ops, 9);
   } else {
-    switch (SourceVT) {
-    case MVT::i8:    Opcode = NVPTX::ST_i8_areg; break;
-    case MVT::i16:   Opcode = NVPTX::ST_i16_areg; break;
-    case MVT::i32:   Opcode = NVPTX::ST_i32_areg; break;
-    case MVT::i64:   Opcode = NVPTX::ST_i64_areg; break;
-    case MVT::f32:   Opcode = NVPTX::ST_f32_areg; break;
-    case MVT::f64:   Opcode = NVPTX::ST_f64_areg; break;
-    case MVT::v2i8:  Opcode = NVPTX::ST_v2i8_areg; break;
-    case MVT::v2i16: Opcode = NVPTX::ST_v2i16_areg; break;
-    case MVT::v2i32: Opcode = NVPTX::ST_v2i32_areg; break;
-    case MVT::v2i64: Opcode = NVPTX::ST_v2i64_areg; break;
-    case MVT::v2f32: Opcode = NVPTX::ST_v2f32_areg; break;
-    case MVT::v2f64: Opcode = NVPTX::ST_v2f64_areg; break;
-    case MVT::v4i8:  Opcode = NVPTX::ST_v4i8_areg; break;
-    case MVT::v4i16: Opcode = NVPTX::ST_v4i16_areg; break;
-    case MVT::v4i32: Opcode = NVPTX::ST_v4i32_areg; break;
-    case MVT::v4f32: Opcode = NVPTX::ST_v4f32_areg; break;
-    default: return NULL;
+    if (Subtarget.is64Bit()) {
+      switch (SourceVT) {
+      case MVT::i8:
+        Opcode = NVPTX::ST_i8_areg_64;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::ST_i16_areg_64;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::ST_i32_areg_64;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::ST_i64_areg_64;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::ST_f32_areg_64;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::ST_f64_areg_64;
+        break;
+      default:
+        return NULL;
+      }
+    } else {
+      switch (SourceVT) {
+      case MVT::i8:
+        Opcode = NVPTX::ST_i8_areg;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::ST_i16_areg;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::ST_i32_areg;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::ST_i64_areg;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::ST_f32_areg;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::ST_f64_areg;
+        break;
+      default:
+        return NULL;
+      }
     }
-    SDValue Ops[] = { N1,
-                      getI32Imm(isVolatile),
-                      getI32Imm(codeAddrSpace),
-                      getI32Imm(vecType),
-                      getI32Imm(toType),
-                      getI32Imm(toTypeWidth),
-                      N2, Chain };
-    NVPTXST = CurDAG->getMachineNode(Opcode, dl,
-                                     MVT::Other, Ops, 8);
+    SDValue Ops[] = { N1, getI32Imm(isVolatile), getI32Imm(codeAddrSpace),
+                      getI32Imm(vecType), getI32Imm(toType),
+                      getI32Imm(toTypeWidth), N2, Chain };
+    NVPTXST = CurDAG->getMachineNode(Opcode, dl, MVT::Other, Ops, 8);
   }
 
   if (NVPTXST != NULL) {
@@ -523,12 +1202,388 @@ SDNode* NVPTXDAGToDAGISel::SelectStore(SDNode *N) {
   return NVPTXST;
 }
 
+SDNode *NVPTXDAGToDAGISel::SelectStoreVector(SDNode *N) {
+  SDValue Chain = N->getOperand(0);
+  SDValue Op1 = N->getOperand(1);
+  SDValue Addr, Offset, Base;
+  unsigned Opcode;
+  DebugLoc DL = N->getDebugLoc();
+  SDNode *ST;
+  EVT EltVT = Op1.getValueType();
+  MemSDNode *MemSD = cast<MemSDNode>(N);
+  EVT StoreVT = MemSD->getMemoryVT();
+
+  // Address Space Setting
+  unsigned CodeAddrSpace = getCodeAddrSpace(MemSD, Subtarget);
+
+  if (CodeAddrSpace == NVPTX::PTXLdStInstCode::CONSTANT) {
+    report_fatal_error("Cannot store to pointer that points to constant "
+                       "memory space");
+  }
+
+  // Volatile Setting
+  // - .volatile is only availalble for .global and .shared
+  bool IsVolatile = MemSD->isVolatile();
+  if (CodeAddrSpace != NVPTX::PTXLdStInstCode::GLOBAL &&
+      CodeAddrSpace != NVPTX::PTXLdStInstCode::SHARED &&
+      CodeAddrSpace != NVPTX::PTXLdStInstCode::GENERIC)
+    IsVolatile = false;
+
+  // Type Setting: toType + toTypeWidth
+  // - for integer type, always use 'u'
+  assert(StoreVT.isSimple() && "Store value is not simple");
+  MVT ScalarVT = StoreVT.getSimpleVT().getScalarType();
+  unsigned ToTypeWidth = ScalarVT.getSizeInBits();
+  unsigned ToType;
+  if (ScalarVT.isFloatingPoint())
+    ToType = NVPTX::PTXLdStInstCode::Float;
+  else
+    ToType = NVPTX::PTXLdStInstCode::Unsigned;
+
+  SmallVector<SDValue, 12> StOps;
+  SDValue N2;
+  unsigned VecType;
+
+  switch (N->getOpcode()) {
+  case NVPTXISD::StoreV2:
+    VecType = NVPTX::PTXLdStInstCode::V2;
+    StOps.push_back(N->getOperand(1));
+    StOps.push_back(N->getOperand(2));
+    N2 = N->getOperand(3);
+    break;
+  case NVPTXISD::StoreV4:
+    VecType = NVPTX::PTXLdStInstCode::V4;
+    StOps.push_back(N->getOperand(1));
+    StOps.push_back(N->getOperand(2));
+    StOps.push_back(N->getOperand(3));
+    StOps.push_back(N->getOperand(4));
+    N2 = N->getOperand(5);
+    break;
+  default:
+    return NULL;
+  }
+
+  StOps.push_back(getI32Imm(IsVolatile));
+  StOps.push_back(getI32Imm(CodeAddrSpace));
+  StOps.push_back(getI32Imm(VecType));
+  StOps.push_back(getI32Imm(ToType));
+  StOps.push_back(getI32Imm(ToTypeWidth));
+
+  if (SelectDirectAddr(N2, Addr)) {
+    switch (N->getOpcode()) {
+    default:
+      return NULL;
+    case NVPTXISD::StoreV2:
+      switch (EltVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::STV_i8_v2_avar;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::STV_i16_v2_avar;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::STV_i32_v2_avar;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::STV_i64_v2_avar;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::STV_f32_v2_avar;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::STV_f64_v2_avar;
+        break;
+      }
+      break;
+    case NVPTXISD::StoreV4:
+      switch (EltVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::STV_i8_v4_avar;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::STV_i16_v4_avar;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::STV_i32_v4_avar;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::STV_f32_v4_avar;
+        break;
+      }
+      break;
+    }
+    StOps.push_back(Addr);
+  } else if (Subtarget.is64Bit()
+                 ? SelectADDRsi64(N2.getNode(), N2, Base, Offset)
+                 : SelectADDRsi(N2.getNode(), N2, Base, Offset)) {
+    switch (N->getOpcode()) {
+    default:
+      return NULL;
+    case NVPTXISD::StoreV2:
+      switch (EltVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::STV_i8_v2_asi;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::STV_i16_v2_asi;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::STV_i32_v2_asi;
+        break;
+      case MVT::i64:
+        Opcode = NVPTX::STV_i64_v2_asi;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::STV_f32_v2_asi;
+        break;
+      case MVT::f64:
+        Opcode = NVPTX::STV_f64_v2_asi;
+        break;
+      }
+      break;
+    case NVPTXISD::StoreV4:
+      switch (EltVT.getSimpleVT().SimpleTy) {
+      default:
+        return NULL;
+      case MVT::i8:
+        Opcode = NVPTX::STV_i8_v4_asi;
+        break;
+      case MVT::i16:
+        Opcode = NVPTX::STV_i16_v4_asi;
+        break;
+      case MVT::i32:
+        Opcode = NVPTX::STV_i32_v4_asi;
+        break;
+      case MVT::f32:
+        Opcode = NVPTX::STV_f32_v4_asi;
+        break;
+      }
+      break;
+    }
+    StOps.push_back(Base);
+    StOps.push_back(Offset);
+  } else if (Subtarget.is64Bit()
+                 ? SelectADDRri64(N2.getNode(), N2, Base, Offset)
+                 : SelectADDRri(N2.getNode(), N2, Base, Offset)) {
+    if (Subtarget.is64Bit()) {
+      switch (N->getOpcode()) {
+      default:
+        return NULL;
+      case NVPTXISD::StoreV2:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::STV_i8_v2_ari_64;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::STV_i16_v2_ari_64;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::STV_i32_v2_ari_64;
+          break;
+        case MVT::i64:
+          Opcode = NVPTX::STV_i64_v2_ari_64;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::STV_f32_v2_ari_64;
+          break;
+        case MVT::f64:
+          Opcode = NVPTX::STV_f64_v2_ari_64;
+          break;
+        }
+        break;
+      case NVPTXISD::StoreV4:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::STV_i8_v4_ari_64;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::STV_i16_v4_ari_64;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::STV_i32_v4_ari_64;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::STV_f32_v4_ari_64;
+          break;
+        }
+        break;
+      }
+    } else {
+      switch (N->getOpcode()) {
+      default:
+        return NULL;
+      case NVPTXISD::StoreV2:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::STV_i8_v2_ari;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::STV_i16_v2_ari;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::STV_i32_v2_ari;
+          break;
+        case MVT::i64:
+          Opcode = NVPTX::STV_i64_v2_ari;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::STV_f32_v2_ari;
+          break;
+        case MVT::f64:
+          Opcode = NVPTX::STV_f64_v2_ari;
+          break;
+        }
+        break;
+      case NVPTXISD::StoreV4:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::STV_i8_v4_ari;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::STV_i16_v4_ari;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::STV_i32_v4_ari;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::STV_f32_v4_ari;
+          break;
+        }
+        break;
+      }
+    }
+    StOps.push_back(Base);
+    StOps.push_back(Offset);
+  } else {
+    if (Subtarget.is64Bit()) {
+      switch (N->getOpcode()) {
+      default:
+        return NULL;
+      case NVPTXISD::StoreV2:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::STV_i8_v2_areg_64;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::STV_i16_v2_areg_64;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::STV_i32_v2_areg_64;
+          break;
+        case MVT::i64:
+          Opcode = NVPTX::STV_i64_v2_areg_64;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::STV_f32_v2_areg_64;
+          break;
+        case MVT::f64:
+          Opcode = NVPTX::STV_f64_v2_areg_64;
+          break;
+        }
+        break;
+      case NVPTXISD::StoreV4:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::STV_i8_v4_areg_64;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::STV_i16_v4_areg_64;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::STV_i32_v4_areg_64;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::STV_f32_v4_areg_64;
+          break;
+        }
+        break;
+      }
+    } else {
+      switch (N->getOpcode()) {
+      default:
+        return NULL;
+      case NVPTXISD::StoreV2:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::STV_i8_v2_areg;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::STV_i16_v2_areg;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::STV_i32_v2_areg;
+          break;
+        case MVT::i64:
+          Opcode = NVPTX::STV_i64_v2_areg;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::STV_f32_v2_areg;
+          break;
+        case MVT::f64:
+          Opcode = NVPTX::STV_f64_v2_areg;
+          break;
+        }
+        break;
+      case NVPTXISD::StoreV4:
+        switch (EltVT.getSimpleVT().SimpleTy) {
+        default:
+          return NULL;
+        case MVT::i8:
+          Opcode = NVPTX::STV_i8_v4_areg;
+          break;
+        case MVT::i16:
+          Opcode = NVPTX::STV_i16_v4_areg;
+          break;
+        case MVT::i32:
+          Opcode = NVPTX::STV_i32_v4_areg;
+          break;
+        case MVT::f32:
+          Opcode = NVPTX::STV_f32_v4_areg;
+          break;
+        }
+        break;
+      }
+    }
+    StOps.push_back(N2);
+  }
+
+  StOps.push_back(Chain);
+
+  ST = CurDAG->getMachineNode(Opcode, DL, MVT::Other, &StOps[0], StOps.size());
+
+  MachineSDNode::mmo_iterator MemRefs0 = MF->allocateMemRefsArray(1);
+  MemRefs0[0] = cast<MemSDNode>(N)->getMemOperand();
+  cast<MachineSDNode>(ST)->setMemRefs(MemRefs0, MemRefs0 + 1);
+
+  return ST;
+}
+
 // SelectDirectAddr - Match a direct address for DAG.
 // A direct address could be a globaladdress or externalsymbol.
 bool NVPTXDAGToDAGISel::SelectDirectAddr(SDValue N, SDValue &Address) {
   // Return true if TGA or ES.
-  if (N.getOpcode() == ISD::TargetGlobalAddress
-      || N.getOpcode() == ISD::TargetExternalSymbol) {
+  if (N.getOpcode() == ISD::TargetGlobalAddress ||
+      N.getOpcode() == ISD::TargetExternalSymbol) {
     Address = N;
     return true;
   }
@@ -546,12 +1601,11 @@ bool NVPTXDAGToDAGISel::SelectDirectAddr(SDValue N, SDValue &Address) {
 }
 
 // symbol+offset
-bool NVPTXDAGToDAGISel::SelectADDRsi_imp(SDNode *OpNode, SDValue Addr,
-                                         SDValue &Base, SDValue &Offset,
-                                         MVT mvt) {
+bool NVPTXDAGToDAGISel::SelectADDRsi_imp(
+    SDNode *OpNode, SDValue Addr, SDValue &Base, SDValue &Offset, MVT mvt) {
   if (Addr.getOpcode() == ISD::ADD) {
     if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Addr.getOperand(1))) {
-      SDValue base=Addr.getOperand(0);
+      SDValue base = Addr.getOperand(0);
       if (SelectDirectAddr(base, Base)) {
         Offset = CurDAG->getTargetConstant(CN->getZExtValue(), mvt);
         return true;
@@ -574,9 +1628,8 @@ bool NVPTXDAGToDAGISel::SelectADDRsi64(SDNode *OpNode, SDValue Addr,
 }
 
 // register+offset
-bool NVPTXDAGToDAGISel::SelectADDRri_imp(SDNode *OpNode, SDValue Addr,
-                                         SDValue &Base, SDValue &Offset,
-                                         MVT mvt) {
+bool NVPTXDAGToDAGISel::SelectADDRri_imp(
+    SDNode *OpNode, SDValue Addr, SDValue &Base, SDValue &Offset, MVT mvt) {
   if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) {
     Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), mvt);
     Offset = CurDAG->getTargetConstant(0, mvt);
@@ -584,7 +1637,7 @@ bool NVPTXDAGToDAGISel::SelectADDRri_imp(SDNode *OpNode, SDValue Addr,
   }
   if (Addr.getOpcode() == ISD::TargetExternalSymbol ||
       Addr.getOpcode() == ISD::TargetGlobalAddress)
-    return false;  // direct calls.
+    return false; // direct calls.
 
   if (Addr.getOpcode() == ISD::ADD) {
     if (SelectDirectAddr(Addr.getOperand(0), Addr)) {
@@ -592,7 +1645,7 @@ bool NVPTXDAGToDAGISel::SelectADDRri_imp(SDNode *OpNode, SDValue Addr,
     }
     if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Addr.getOperand(1))) {
       if (FrameIndexSDNode *FIN =
-          dyn_cast<FrameIndexSDNode>(Addr.getOperand(0)))
+              dyn_cast<FrameIndexSDNode>(Addr.getOperand(0)))
         // Constant offset from frame ref.
         Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), mvt);
       else
@@ -624,8 +1677,7 @@ bool NVPTXDAGToDAGISel::ChkMemSDNodeAddressSpace(SDNode *N,
   // (See SelectionDAGNodes.h). So we need to check for both.
   if (MemSDNode *mN = dyn_cast<MemSDNode>(N)) {
     Src = mN->getSrcValue();
-  }
-  else if (MemSDNode *mN = dyn_cast<MemIntrinsicSDNode>(N)) {
+  } else if (MemSDNode *mN = dyn_cast<MemIntrinsicSDNode>(N)) {
     Src = mN->getSrcValue();
   }
   if (!Src)
@@ -637,13 +1689,13 @@ bool NVPTXDAGToDAGISel::ChkMemSDNodeAddressSpace(SDNode *N,
 
 /// SelectInlineAsmMemoryOperand - Implement addressing mode selection for
 /// inline asm expressions.
-bool NVPTXDAGToDAGISel::SelectInlineAsmMemoryOperand(const SDValue &Op,
-                                                     char ConstraintCode,
-                                                 std::vector<SDValue> &OutOps) {
+bool NVPTXDAGToDAGISel::SelectInlineAsmMemoryOperand(
+    const SDValue &Op, char ConstraintCode, std::vector<SDValue> &OutOps) {
   SDValue Op0, Op1;
   switch (ConstraintCode) {
-  default: return true;
-  case 'm':   // memory
+  default:
+    return true;
+  case 'm': // memory
     if (SelectDirectAddr(Op, Op0)) {
       OutOps.push_back(Op0);
       OutOps.push_back(CurDAG->getTargetConstant(0, MVT::i32));
@@ -666,10 +1718,8 @@ bool NVPTXDAGToDAGISel::SelectInlineAsmMemoryOperand(const SDValue &Op,
 // pattern matcher inserts a bunch of IMOVi8rr to convert
 // the imm to i8imm, and this causes instruction selection
 // to fail.
-bool NVPTXDAGToDAGISel::UndefOrImm(SDValue Op, SDValue N,
-                                   SDValue &Retval) {
-  if (!(N.getOpcode() == ISD::UNDEF) &&
-      !(N.getOpcode() == ISD::Constant))
+bool NVPTXDAGToDAGISel::UndefOrImm(SDValue Op, SDValue N, SDValue &Retval) {
+  if (!(N.getOpcode() == ISD::UNDEF) && !(N.getOpcode() == ISD::Constant))
     return false;
 
   if (N.getOpcode() == ISD::UNDEF)
diff --git a/lib/Target/NVPTX/NVPTXISelDAGToDAG.h b/lib/Target/NVPTX/NVPTXISelDAGToDAG.h
index ccd69b29dd42..70e8e464297d 100644
--- a/lib/Target/NVPTX/NVPTXISelDAGToDAG.h
+++ b/lib/Target/NVPTX/NVPTXISelDAGToDAG.h
@@ -18,8 +18,8 @@
 #include "NVPTXRegisterInfo.h"
 #include "NVPTXTargetMachine.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/Compiler.h"
-#include "llvm/Intrinsics.h"
 using namespace llvm;
 
 namespace {
@@ -64,16 +64,18 @@ public:
 
   const NVPTXSubtarget &Subtarget;
 
-  virtual bool SelectInlineAsmMemoryOperand(const SDValue &Op,
-                                            char ConstraintCode,
-                                            std::vector<SDValue> &OutOps);
+  virtual bool SelectInlineAsmMemoryOperand(
+      const SDValue &Op, char ConstraintCode, std::vector<SDValue> &OutOps);
 private:
-  // Include the pieces autogenerated from the target description.
+// Include the pieces autogenerated from the target description.
 #include "NVPTXGenDAGISel.inc"
 
   SDNode *Select(SDNode *N);
-  SDNode* SelectLoad(SDNode *N);
-  SDNode* SelectStore(SDNode *N);
+  SDNode *SelectLoad(SDNode *N);
+  SDNode *SelectLoadVector(SDNode *N);
+  SDNode *SelectLDGLDUVector(SDNode *N);
+  SDNode *SelectStore(SDNode *N);
+  SDNode *SelectStoreVector(SDNode *N);
 
   inline SDValue getI32Imm(unsigned Imm) {
     return CurDAG->getTargetConstant(Imm, MVT::i32);
@@ -96,7 +98,6 @@ private:
   bool SelectADDRsi64(SDNode *OpNode, SDValue Addr, SDValue &Base,
                       SDValue &Offset);
 
-
   bool ChkMemSDNodeAddressSpace(SDNode *N, unsigned int spN) const;
 
   bool UndefOrImm(SDValue Op, SDValue N, SDValue &Retval);
diff --git a/lib/Target/NVPTX/NVPTXISelLowering.cpp b/lib/Target/NVPTX/NVPTXISelLowering.cpp
index f1a99d77be9d..6e01a5a82071 100644
--- a/lib/Target/NVPTX/NVPTXISelLowering.cpp
+++ b/lib/Target/NVPTX/NVPTXISelLowering.cpp
@@ -11,30 +11,29 @@
 //
 //===----------------------------------------------------------------------===//
 
-
-#include "NVPTX.h"
 #include "NVPTXISelLowering.h"
+#include "NVPTX.h"
 #include "NVPTXTargetMachine.h"
 #include "NVPTXTargetObjectFile.h"
 #include "NVPTXUtilities.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/Support/CommandLine.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/GlobalValue.h"
-#include "llvm/Module.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/Analysis.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Module.h"
+#include "llvm/MC/MCSectionELF.h"
 #include "llvm/Support/CallSite.h"
-#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
-#include "llvm/MC/MCSectionELF.h"
 #include <sstream>
 
 #undef DEBUG_TYPE
@@ -44,28 +43,39 @@ using namespace llvm;
 
 static unsigned int uniqueCallSite = 0;
 
-static cl::opt<bool>
-RetainVectorOperands("nvptx-codegen-vectors",
-     cl::desc("NVPTX Specific: Retain LLVM's vectors and generate PTX vectors"),
-                     cl::init(true));
+static cl::opt<bool> sched4reg(
+    "nvptx-sched4reg",
+    cl::desc("NVPTX Specific: schedule for register pressue"), cl::init(false));
 
-static cl::opt<bool>
-sched4reg("nvptx-sched4reg",
-          cl::desc("NVPTX Specific: schedule for register pressue"),
-          cl::init(false));
+static bool IsPTXVectorType(MVT VT) {
+  switch (VT.SimpleTy) {
+  default:
+    return false;
+  case MVT::v2i8:
+  case MVT::v4i8:
+  case MVT::v2i16:
+  case MVT::v4i16:
+  case MVT::v2i32:
+  case MVT::v4i32:
+  case MVT::v2i64:
+  case MVT::v2f32:
+  case MVT::v4f32:
+  case MVT::v2f64:
+    return true;
+  }
+}
 
 // NVPTXTargetLowering Constructor.
 NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
-: TargetLowering(TM, new NVPTXTargetObjectFile()),
-  nvTM(&TM),
-  nvptxSubtarget(TM.getSubtarget<NVPTXSubtarget>()) {
+    : TargetLowering(TM, new NVPTXTargetObjectFile()), nvTM(&TM),
+      nvptxSubtarget(TM.getSubtarget<NVPTXSubtarget>()) {
 
   // always lower memset, memcpy, and memmove intrinsics to load/store
   // instructions, rather
   // then generating calls to memset, mempcy or memmove.
-  maxStoresPerMemset = (unsigned)0xFFFFFFFF;
-  maxStoresPerMemcpy = (unsigned)0xFFFFFFFF;
-  maxStoresPerMemmove = (unsigned)0xFFFFFFFF;
+  MaxStoresPerMemset = (unsigned) 0xFFFFFFFF;
+  MaxStoresPerMemcpy = (unsigned) 0xFFFFFFFF;
+  MaxStoresPerMemmove = (unsigned) 0xFFFFFFFF;
 
   setBooleanContents(ZeroOrNegativeOneBooleanContent);
 
@@ -87,82 +97,51 @@ NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
   addRegisterClass(MVT::f32, &NVPTX::Float32RegsRegClass);
   addRegisterClass(MVT::f64, &NVPTX::Float64RegsRegClass);
 
-  if (RetainVectorOperands) {
-    addRegisterClass(MVT::v2f32, &NVPTX::V2F32RegsRegClass);
-    addRegisterClass(MVT::v4f32, &NVPTX::V4F32RegsRegClass);
-    addRegisterClass(MVT::v2i32, &NVPTX::V2I32RegsRegClass);
-    addRegisterClass(MVT::v4i32, &NVPTX::V4I32RegsRegClass);
-    addRegisterClass(MVT::v2f64, &NVPTX::V2F64RegsRegClass);
-    addRegisterClass(MVT::v2i64, &NVPTX::V2I64RegsRegClass);
-    addRegisterClass(MVT::v2i16, &NVPTX::V2I16RegsRegClass);
-    addRegisterClass(MVT::v4i16, &NVPTX::V4I16RegsRegClass);
-    addRegisterClass(MVT::v2i8, &NVPTX::V2I8RegsRegClass);
-    addRegisterClass(MVT::v4i8, &NVPTX::V4I8RegsRegClass);
-
-    setOperationAction(ISD::BUILD_VECTOR, MVT::v4i32  , Custom);
-    setOperationAction(ISD::BUILD_VECTOR, MVT::v4f32  , Custom);
-    setOperationAction(ISD::BUILD_VECTOR, MVT::v4i16  , Custom);
-    setOperationAction(ISD::BUILD_VECTOR, MVT::v4i8   , Custom);
-    setOperationAction(ISD::BUILD_VECTOR, MVT::v2i64  , Custom);
-    setOperationAction(ISD::BUILD_VECTOR, MVT::v2f64  , Custom);
-    setOperationAction(ISD::BUILD_VECTOR, MVT::v2i32  , Custom);
-    setOperationAction(ISD::BUILD_VECTOR, MVT::v2f32  , Custom);
-    setOperationAction(ISD::BUILD_VECTOR, MVT::v2i16  , Custom);
-    setOperationAction(ISD::BUILD_VECTOR, MVT::v2i8   , Custom);
-
-    setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v4i32  , Custom);
-    setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v4f32  , Custom);
-    setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v4i16  , Custom);
-    setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v4i8   , Custom);
-    setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2i64  , Custom);
-    setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2f64  , Custom);
-    setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2i32  , Custom);
-    setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2f32  , Custom);
-    setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2i16  , Custom);
-    setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2i8   , Custom);
-  }
-
   // Operations not directly supported by NVPTX.
-  setOperationAction(ISD::SELECT_CC,         MVT::Other, Expand);
-  setOperationAction(ISD::BR_CC,             MVT::Other, Expand);
+  setOperationAction(ISD::SELECT_CC, MVT::Other, Expand);
+  setOperationAction(ISD::BR_CC, MVT::f32, Expand);
+  setOperationAction(ISD::BR_CC, MVT::f64, Expand);
+  setOperationAction(ISD::BR_CC, MVT::i1, Expand);
+  setOperationAction(ISD::BR_CC, MVT::i8, Expand);
+  setOperationAction(ISD::BR_CC, MVT::i16, Expand);
+  setOperationAction(ISD::BR_CC, MVT::i32, Expand);
+  setOperationAction(ISD::BR_CC, MVT::i64, Expand);
   setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i64, Expand);
   setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i32, Expand);
   setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16, Expand);
-  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8 , Expand);
-  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1 , Expand);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8, Expand);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
 
   if (nvptxSubtarget.hasROT64()) {
-    setOperationAction(ISD::ROTL , MVT::i64, Legal);
-    setOperationAction(ISD::ROTR , MVT::i64, Legal);
-  }
-  else {
-    setOperationAction(ISD::ROTL , MVT::i64, Expand);
-    setOperationAction(ISD::ROTR , MVT::i64, Expand);
+    setOperationAction(ISD::ROTL, MVT::i64, Legal);
+    setOperationAction(ISD::ROTR, MVT::i64, Legal);
+  } else {
+    setOperationAction(ISD::ROTL, MVT::i64, Expand);
+    setOperationAction(ISD::ROTR, MVT::i64, Expand);
   }
   if (nvptxSubtarget.hasROT32()) {
-    setOperationAction(ISD::ROTL , MVT::i32, Legal);
-    setOperationAction(ISD::ROTR , MVT::i32, Legal);
-  }
-  else {
-    setOperationAction(ISD::ROTL , MVT::i32, Expand);
-    setOperationAction(ISD::ROTR , MVT::i32, Expand);
+    setOperationAction(ISD::ROTL, MVT::i32, Legal);
+    setOperationAction(ISD::ROTR, MVT::i32, Legal);
+  } else {
+    setOperationAction(ISD::ROTL, MVT::i32, Expand);
+    setOperationAction(ISD::ROTR, MVT::i32, Expand);
   }
 
-  setOperationAction(ISD::ROTL , MVT::i16, Expand);
-  setOperationAction(ISD::ROTR , MVT::i16, Expand);
-  setOperationAction(ISD::ROTL , MVT::i8, Expand);
-  setOperationAction(ISD::ROTR , MVT::i8, Expand);
-  setOperationAction(ISD::BSWAP , MVT::i16, Expand);
-  setOperationAction(ISD::BSWAP , MVT::i32, Expand);
-  setOperationAction(ISD::BSWAP , MVT::i64, Expand);
+  setOperationAction(ISD::ROTL, MVT::i16, Expand);
+  setOperationAction(ISD::ROTR, MVT::i16, Expand);
+  setOperationAction(ISD::ROTL, MVT::i8, Expand);
+  setOperationAction(ISD::ROTR, MVT::i8, Expand);
+  setOperationAction(ISD::BSWAP, MVT::i16, Expand);
+  setOperationAction(ISD::BSWAP, MVT::i32, Expand);
+  setOperationAction(ISD::BSWAP, MVT::i64, Expand);
 
   // Indirect branch is not supported.
   // This also disables Jump Table creation.
-  setOperationAction(ISD::BR_JT,             MVT::Other, Expand);
-  setOperationAction(ISD::BRIND,             MVT::Other, Expand);
+  setOperationAction(ISD::BR_JT, MVT::Other, Expand);
+  setOperationAction(ISD::BRIND, MVT::Other, Expand);
 
-  setOperationAction(ISD::GlobalAddress   , MVT::i32  , Custom);
-  setOperationAction(ISD::GlobalAddress   , MVT::i64  , Custom);
+  setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
+  setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
 
   // We want to legalize constant related memmove and memcopy
   // intrinsics.
@@ -185,92 +164,114 @@ NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
   setTruncStoreAction(MVT::i8, MVT::i1, Expand);
 
   // This is legal in NVPTX
-  setOperationAction(ISD::ConstantFP,         MVT::f64, Legal);
-  setOperationAction(ISD::ConstantFP,         MVT::f32, Legal);
+  setOperationAction(ISD::ConstantFP, MVT::f64, Legal);
+  setOperationAction(ISD::ConstantFP, MVT::f32, Legal);
 
   // TRAP can be lowered to PTX trap
-  setOperationAction(ISD::TRAP,               MVT::Other, Legal);
-
-  // By default, CONCAT_VECTORS is implemented via store/load
-  // through stack. It is slow and uses local memory. We need
-  // to custom-lowering them.
-  setOperationAction(ISD::CONCAT_VECTORS, MVT::v4i32  , Custom);
-  setOperationAction(ISD::CONCAT_VECTORS, MVT::v4f32  , Custom);
-  setOperationAction(ISD::CONCAT_VECTORS, MVT::v4i16  , Custom);
-  setOperationAction(ISD::CONCAT_VECTORS, MVT::v4i8   , Custom);
-  setOperationAction(ISD::CONCAT_VECTORS, MVT::v2i64  , Custom);
-  setOperationAction(ISD::CONCAT_VECTORS, MVT::v2f64  , Custom);
-  setOperationAction(ISD::CONCAT_VECTORS, MVT::v2i32  , Custom);
-  setOperationAction(ISD::CONCAT_VECTORS, MVT::v2f32  , Custom);
-  setOperationAction(ISD::CONCAT_VECTORS, MVT::v2i16  , Custom);
-  setOperationAction(ISD::CONCAT_VECTORS, MVT::v2i8   , Custom);
-
-  // Expand vector int to float and float to int conversions
-  // - For SINT_TO_FP and UINT_TO_FP, the src type
-  //   (Node->getOperand(0).getValueType())
-  //   is used to determine the action, while for FP_TO_UINT and FP_TO_SINT,
-  //   the dest type (Node->getValueType(0)) is used.
-  //
-  //   See VectorLegalizer::LegalizeOp() (LegalizeVectorOps.cpp) for the vector
-  //   case, and
-  //   SelectionDAGLegalize::LegalizeOp() (LegalizeDAG.cpp) for the scalar case.
-  //
-  //   That is why v4i32 or v2i32 are used here.
-  //
-  //   The expansion for vectors happens in VectorLegalizer::LegalizeOp()
-  //   (LegalizeVectorOps.cpp).
-  setOperationAction(ISD::SINT_TO_FP, MVT::v4i32, Expand);
-  setOperationAction(ISD::SINT_TO_FP, MVT::v2i32, Expand);
-  setOperationAction(ISD::UINT_TO_FP, MVT::v4i32, Expand);
-  setOperationAction(ISD::UINT_TO_FP, MVT::v2i32, Expand);
-  setOperationAction(ISD::FP_TO_SINT, MVT::v2i32, Expand);
-  setOperationAction(ISD::FP_TO_SINT, MVT::v4i32, Expand);
-  setOperationAction(ISD::FP_TO_UINT, MVT::v2i32, Expand);
-  setOperationAction(ISD::FP_TO_UINT, MVT::v4i32, Expand);
+  setOperationAction(ISD::TRAP, MVT::Other, Legal);
+
+  // Register custom handling for vector loads/stores
+  for (int i = MVT::FIRST_VECTOR_VALUETYPE; i <= MVT::LAST_VECTOR_VALUETYPE;
+       ++i) {
+    MVT VT = (MVT::SimpleValueType) i;
+    if (IsPTXVectorType(VT)) {
+      setOperationAction(ISD::LOAD, VT, Custom);
+      setOperationAction(ISD::STORE, VT, Custom);
+      setOperationAction(ISD::INTRINSIC_W_CHAIN, VT, Custom);
+    }
+  }
 
   // Now deduce the information based on the above mentioned
   // actions
   computeRegisterProperties();
 }
 
-
 const char *NVPTXTargetLowering::getTargetNodeName(unsigned Opcode) const {
   switch (Opcode) {
-  default: return 0;
-  case NVPTXISD::CALL:            return "NVPTXISD::CALL";
-  case NVPTXISD::RET_FLAG:        return "NVPTXISD::RET_FLAG";
-  case NVPTXISD::Wrapper:         return "NVPTXISD::Wrapper";
-  case NVPTXISD::NVBuiltin:       return "NVPTXISD::NVBuiltin";
-  case NVPTXISD::DeclareParam:    return "NVPTXISD::DeclareParam";
+  default:
+    return 0;
+  case NVPTXISD::CALL:
+    return "NVPTXISD::CALL";
+  case NVPTXISD::RET_FLAG:
+    return "NVPTXISD::RET_FLAG";
+  case NVPTXISD::Wrapper:
+    return "NVPTXISD::Wrapper";
+  case NVPTXISD::NVBuiltin:
+    return "NVPTXISD::NVBuiltin";
+  case NVPTXISD::DeclareParam:
+    return "NVPTXISD::DeclareParam";
   case NVPTXISD::DeclareScalarParam:
     return "NVPTXISD::DeclareScalarParam";
-  case NVPTXISD::DeclareRet:      return "NVPTXISD::DeclareRet";
-  case NVPTXISD::DeclareRetParam: return "NVPTXISD::DeclareRetParam";
-  case NVPTXISD::PrintCall:       return "NVPTXISD::PrintCall";
-  case NVPTXISD::LoadParam:       return "NVPTXISD::LoadParam";
-  case NVPTXISD::StoreParam:      return "NVPTXISD::StoreParam";
-  case NVPTXISD::StoreParamS32:   return "NVPTXISD::StoreParamS32";
-  case NVPTXISD::StoreParamU32:   return "NVPTXISD::StoreParamU32";
-  case NVPTXISD::MoveToParam:     return "NVPTXISD::MoveToParam";
-  case NVPTXISD::CallArgBegin:    return "NVPTXISD::CallArgBegin";
-  case NVPTXISD::CallArg:         return "NVPTXISD::CallArg";
-  case NVPTXISD::LastCallArg:     return "NVPTXISD::LastCallArg";
-  case NVPTXISD::CallArgEnd:      return "NVPTXISD::CallArgEnd";
-  case NVPTXISD::CallVoid:        return "NVPTXISD::CallVoid";
-  case NVPTXISD::CallVal:         return "NVPTXISD::CallVal";
-  case NVPTXISD::CallSymbol:      return "NVPTXISD::CallSymbol";
-  case NVPTXISD::Prototype:       return "NVPTXISD::Prototype";
-  case NVPTXISD::MoveParam:       return "NVPTXISD::MoveParam";
-  case NVPTXISD::MoveRetval:      return "NVPTXISD::MoveRetval";
-  case NVPTXISD::MoveToRetval:    return "NVPTXISD::MoveToRetval";
-  case NVPTXISD::StoreRetval:     return "NVPTXISD::StoreRetval";
-  case NVPTXISD::PseudoUseParam:  return "NVPTXISD::PseudoUseParam";
-  case NVPTXISD::RETURN:          return "NVPTXISD::RETURN";
-  case NVPTXISD::CallSeqBegin:    return "NVPTXISD::CallSeqBegin";
-  case NVPTXISD::CallSeqEnd:      return "NVPTXISD::CallSeqEnd";
+  case NVPTXISD::DeclareRet:
+    return "NVPTXISD::DeclareRet";
+  case NVPTXISD::DeclareRetParam:
+    return "NVPTXISD::DeclareRetParam";
+  case NVPTXISD::PrintCall:
+    return "NVPTXISD::PrintCall";
+  case NVPTXISD::LoadParam:
+    return "NVPTXISD::LoadParam";
+  case NVPTXISD::StoreParam:
+    return "NVPTXISD::StoreParam";
+  case NVPTXISD::StoreParamS32:
+    return "NVPTXISD::StoreParamS32";
+  case NVPTXISD::StoreParamU32:
+    return "NVPTXISD::StoreParamU32";
+  case NVPTXISD::MoveToParam:
+    return "NVPTXISD::MoveToParam";
+  case NVPTXISD::CallArgBegin:
+    return "NVPTXISD::CallArgBegin";
+  case NVPTXISD::CallArg:
+    return "NVPTXISD::CallArg";
+  case NVPTXISD::LastCallArg:
+    return "NVPTXISD::LastCallArg";
+  case NVPTXISD::CallArgEnd:
+    return "NVPTXISD::CallArgEnd";
+  case NVPTXISD::CallVoid:
+    return "NVPTXISD::CallVoid";
+  case NVPTXISD::CallVal:
+    return "NVPTXISD::CallVal";
+  case NVPTXISD::CallSymbol:
+    return "NVPTXISD::CallSymbol";
+  case NVPTXISD::Prototype:
+    return "NVPTXISD::Prototype";
+  case NVPTXISD::MoveParam:
+    return "NVPTXISD::MoveParam";
+  case NVPTXISD::MoveRetval:
+    return "NVPTXISD::MoveRetval";
+  case NVPTXISD::MoveToRetval:
+    return "NVPTXISD::MoveToRetval";
+  case NVPTXISD::StoreRetval:
+    return "NVPTXISD::StoreRetval";
+  case NVPTXISD::PseudoUseParam:
+    return "NVPTXISD::PseudoUseParam";
+  case NVPTXISD::RETURN:
+    return "NVPTXISD::RETURN";
+  case NVPTXISD::CallSeqBegin:
+    return "NVPTXISD::CallSeqBegin";
+  case NVPTXISD::CallSeqEnd:
+    return "NVPTXISD::CallSeqEnd";
+  case NVPTXISD::LoadV2:
+    return "NVPTXISD::LoadV2";
+  case NVPTXISD::LoadV4:
+    return "NVPTXISD::LoadV4";
+  case NVPTXISD::LDGV2:
+    return "NVPTXISD::LDGV2";
+  case NVPTXISD::LDGV4:
+    return "NVPTXISD::LDGV4";
+  case NVPTXISD::LDUV2:
+    return "NVPTXISD::LDUV2";
+  case NVPTXISD::LDUV4:
+    return "NVPTXISD::LDUV4";
+  case NVPTXISD::StoreV2:
+    return "NVPTXISD::StoreV2";
+  case NVPTXISD::StoreV4:
+    return "NVPTXISD::StoreV4";
   }
 }
 
+bool NVPTXTargetLowering::shouldSplitVectorElementType(EVT VT) const {
+  return VT == MVT::i1;
+}
 
 SDValue
 NVPTXTargetLowering::LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const {
@@ -280,10 +281,9 @@ NVPTXTargetLowering::LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const {
   return DAG.getNode(NVPTXISD::Wrapper, dl, getPointerTy(), Op);
 }
 
-std::string NVPTXTargetLowering::getPrototype(Type *retTy,
-                                              const ArgListTy &Args,
-                                    const SmallVectorImpl<ISD::OutputArg> &Outs,
-                                              unsigned retAlignment) const {
+std::string NVPTXTargetLowering::getPrototype(
+    Type *retTy, const ArgListTy &Args,
+    const SmallVectorImpl<ISD::OutputArg> &Outs, unsigned retAlignment) const {
 
   bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
 
@@ -299,54 +299,47 @@ std::string NVPTXTargetLowering::getPrototype(Type *retTy,
         unsigned size = 0;
         if (const IntegerType *ITy = dyn_cast<IntegerType>(retTy)) {
           size = ITy->getBitWidth();
-          if (size < 32) size = 32;
-        }
-        else {
+          if (size < 32)
+            size = 32;
+        } else {
           assert(retTy->isFloatingPointTy() &&
                  "Floating point type expected here");
           size = retTy->getPrimitiveSizeInBits();
         }
 
         O << ".param .b" << size << " _";
-      }
-      else if (isa<PointerType>(retTy))
-        O << ".param .b" << getPointerTy().getSizeInBits()
-        << " _";
+      } else if (isa<PointerType>(retTy))
+        O << ".param .b" << getPointerTy().getSizeInBits() << " _";
       else {
         if ((retTy->getTypeID() == Type::StructTyID) ||
             isa<VectorType>(retTy)) {
           SmallVector<EVT, 16> vtparts;
           ComputeValueVTs(*this, retTy, vtparts);
           unsigned totalsz = 0;
-          for (unsigned i=0,e=vtparts.size(); i!=e; ++i) {
+          for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
             unsigned elems = 1;
             EVT elemtype = vtparts[i];
             if (vtparts[i].isVector()) {
               elems = vtparts[i].getVectorNumElements();
               elemtype = vtparts[i].getVectorElementType();
             }
-            for (unsigned j=0, je=elems; j!=je; ++j) {
+            for (unsigned j = 0, je = elems; j != je; ++j) {
               unsigned sz = elemtype.getSizeInBits();
-              if (elemtype.isInteger() && (sz < 8)) sz = 8;
-              totalsz += sz/8;
+              if (elemtype.isInteger() && (sz < 8))
+                sz = 8;
+              totalsz += sz / 8;
             }
           }
-          O << ".param .align "
-              << retAlignment
-              << " .b8 _["
-              << totalsz << "]";
-        }
-        else {
-          assert(false &&
-                 "Unknown return type");
+          O << ".param .align " << retAlignment << " .b8 _[" << totalsz << "]";
+        } else {
+          assert(false && "Unknown return type");
         }
       }
-    }
-    else {
+    } else {
       SmallVector<EVT, 16> vtparts;
       ComputeValueVTs(*this, retTy, vtparts);
       unsigned idx = 0;
-      for (unsigned i=0,e=vtparts.size(); i!=e; ++i) {
+      for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
         unsigned elems = 1;
         EVT elemtype = vtparts[i];
         if (vtparts[i].isVector()) {
@@ -354,14 +347,16 @@ std::string NVPTXTargetLowering::getPrototype(Type *retTy,
           elemtype = vtparts[i].getVectorElementType();
         }
 
-        for (unsigned j=0, je=elems; j!=je; ++j) {
+        for (unsigned j = 0, je = elems; j != je; ++j) {
           unsigned sz = elemtype.getSizeInBits();
-          if (elemtype.isInteger() && (sz < 32)) sz = 32;
+          if (elemtype.isInteger() && (sz < 32))
+            sz = 32;
           O << ".reg .b" << sz << " _";
-          if (j<je-1) O << ", ";
+          if (j < je - 1)
+            O << ", ";
           ++idx;
         }
-        if (i < e-1)
+        if (i < e - 1)
           O << ", ";
       }
     }
@@ -372,7 +367,7 @@ std::string NVPTXTargetLowering::getPrototype(Type *retTy,
   bool first = true;
   MVT thePointerTy = getPointerTy();
 
-  for (unsigned i=0,e=Args.size(); i!=e; ++i) {
+  for (unsigned i = 0, e = Args.size(); i != e; ++i) {
     const Type *Ty = Args[i].Ty;
     if (!first) {
       O << ", ";
@@ -383,9 +378,9 @@ std::string NVPTXTargetLowering::getPrototype(Type *retTy,
       unsigned sz = 0;
       if (isa<IntegerType>(Ty)) {
         sz = cast<IntegerType>(Ty)->getBitWidth();
-        if (sz < 32) sz = 32;
-      }
-      else if (isa<PointerType>(Ty))
+        if (sz < 32)
+          sz = 32;
+      } else if (isa<PointerType>(Ty))
         sz = thePointerTy.getSizeInBits();
       else
         sz = Ty->getPrimitiveSizeInBits();
@@ -397,23 +392,20 @@ std::string NVPTXTargetLowering::getPrototype(Type *retTy,
       continue;
     }
     const PointerType *PTy = dyn_cast<PointerType>(Ty);
-    assert(PTy &&
-           "Param with byval attribute should be a pointer type");
+    assert(PTy && "Param with byval attribute should be a pointer type");
     Type *ETy = PTy->getElementType();
 
     if (isABI) {
       unsigned align = Outs[i].Flags.getByValAlign();
       unsigned sz = getDataLayout()->getTypeAllocSize(ETy);
-      O << ".param .align " << align
-          << " .b8 ";
+      O << ".param .align " << align << " .b8 ";
       O << "_";
       O << "[" << sz << "]";
       continue;
-    }
-    else {
+    } else {
       SmallVector<EVT, 16> vtparts;
       ComputeValueVTs(*this, ETy, vtparts);
-      for (unsigned i=0,e=vtparts.size(); i!=e; ++i) {
+      for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
         unsigned elems = 1;
         EVT elemtype = vtparts[i];
         if (vtparts[i].isVector()) {
@@ -421,14 +413,16 @@ std::string NVPTXTargetLowering::getPrototype(Type *retTy,
           elemtype = vtparts[i].getVectorElementType();
         }
 
-        for (unsigned j=0,je=elems; j!=je; ++j) {
+        for (unsigned j = 0, je = elems; j != je; ++j) {
           unsigned sz = elemtype.getSizeInBits();
-          if (elemtype.isInteger() && (sz < 32)) sz = 32;
+          if (elemtype.isInteger() && (sz < 32))
+            sz = 32;
           O << ".reg .b" << sz << " ";
           O << "_";
-          if (j<je-1) O << ", ";
+          if (j < je - 1)
+            O << ", ";
         }
-        if (i<e-1)
+        if (i < e - 1)
           O << ", ";
       }
       continue;
@@ -438,27 +432,25 @@ std::string NVPTXTargetLowering::getPrototype(Type *retTy,
   return O.str();
 }
 
-
-SDValue
-NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
-                               SmallVectorImpl<SDValue> &InVals) const {
-  SelectionDAG &DAG                     = CLI.DAG;
-  DebugLoc &dl                          = CLI.DL;
+SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
+                                       SmallVectorImpl<SDValue> &InVals) const {
+  SelectionDAG &DAG = CLI.DAG;
+  DebugLoc &dl = CLI.DL;
   SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
-  SmallVector<SDValue, 32> &OutVals     = CLI.OutVals;
-  SmallVector<ISD::InputArg, 32> &Ins   = CLI.Ins;
-  SDValue Chain                         = CLI.Chain;
-  SDValue Callee                        = CLI.Callee;
-  bool &isTailCall                      = CLI.IsTailCall;
-  ArgListTy &Args                       = CLI.Args;
-  Type *retTy                           = CLI.RetTy;
-  ImmutableCallSite *CS                 = CLI.CS;
+  SmallVector<SDValue, 32> &OutVals = CLI.OutVals;
+  SmallVector<ISD::InputArg, 32> &Ins = CLI.Ins;
+  SDValue Chain = CLI.Chain;
+  SDValue Callee = CLI.Callee;
+  bool &isTailCall = CLI.IsTailCall;
+  ArgListTy &Args = CLI.Args;
+  Type *retTy = CLI.RetTy;
+  ImmutableCallSite *CS = CLI.CS;
 
   bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
 
   SDValue tempChain = Chain;
-  Chain = DAG.getCALLSEQ_START(Chain,
-                               DAG.getIntPtrConstant(uniqueCallSite, true));
+  Chain =
+      DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(uniqueCallSite, true));
   SDValue InFlag = Chain.getValue(1);
 
   assert((Outs.size() == Args.size()) &&
@@ -466,7 +458,7 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   unsigned paramCount = 0;
   // Declare the .params or .reg need to pass values
   // to the function
-  for (unsigned i=0, e=Outs.size(); i!=e; ++i) {
+  for (unsigned i = 0, e = Outs.size(); i != e; ++i) {
     EVT VT = Outs[i].VT;
 
     if (Outs[i].Flags.isByVal() == false) {
@@ -477,19 +469,20 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       if (isABI)
         isReg = 0;
       unsigned sz = VT.getSizeInBits();
-      if (VT.isInteger() && (sz < 32)) sz = 32;
+      if (VT.isInteger() && (sz < 32))
+        sz = 32;
       SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
       SDValue DeclareParamOps[] = { Chain,
                                     DAG.getConstant(paramCount, MVT::i32),
                                     DAG.getConstant(sz, MVT::i32),
-                                    DAG.getConstant(isReg, MVT::i32),
-                                    InFlag };
+                                    DAG.getConstant(isReg, MVT::i32), InFlag };
       Chain = DAG.getNode(NVPTXISD::DeclareScalarParam, dl, DeclareParamVTs,
                           DeclareParamOps, 5);
       InFlag = Chain.getValue(1);
       SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
       SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
-                             DAG.getConstant(0, MVT::i32), OutVals[i], InFlag };
+                                 DAG.getConstant(0, MVT::i32), OutVals[i],
+                                 InFlag };
 
       unsigned opcode = NVPTXISD::StoreParam;
       if (isReg)
@@ -509,8 +502,7 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     // struct or vector
     SmallVector<EVT, 16> vtparts;
     const PointerType *PTy = dyn_cast<PointerType>(Args[i].Ty);
-    assert(PTy &&
-           "Type of a byval parameter should be pointer");
+    assert(PTy && "Type of a byval parameter should be pointer");
     ComputeValueVTs(*this, PTy->getElementType(), vtparts);
 
     if (isABI) {
@@ -520,40 +512,41 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       // The ByValAlign in the Outs[i].Flags is alway set at this point, so we
       // don't need to
       // worry about natural alignment or not. See TargetLowering::LowerCallTo()
-      SDValue DeclareParamOps[] = { Chain,
-                       DAG.getConstant(Outs[i].Flags.getByValAlign(), MVT::i32),
-                                    DAG.getConstant(paramCount, MVT::i32),
-                                    DAG.getConstant(sz, MVT::i32),
-                                    InFlag };
+      SDValue DeclareParamOps[] = {
+        Chain, DAG.getConstant(Outs[i].Flags.getByValAlign(), MVT::i32),
+        DAG.getConstant(paramCount, MVT::i32), DAG.getConstant(sz, MVT::i32),
+        InFlag
+      };
       Chain = DAG.getNode(NVPTXISD::DeclareParam, dl, DeclareParamVTs,
                           DeclareParamOps, 5);
       InFlag = Chain.getValue(1);
       unsigned curOffset = 0;
-      for (unsigned j=0,je=vtparts.size(); j!=je; ++j) {
+      for (unsigned j = 0, je = vtparts.size(); j != je; ++j) {
         unsigned elems = 1;
         EVT elemtype = vtparts[j];
         if (vtparts[j].isVector()) {
           elems = vtparts[j].getVectorNumElements();
           elemtype = vtparts[j].getVectorElementType();
         }
-        for (unsigned k=0,ke=elems; k!=ke; ++k) {
+        for (unsigned k = 0, ke = elems; k != ke; ++k) {
           unsigned sz = elemtype.getSizeInBits();
-          if (elemtype.isInteger() && (sz < 8)) sz = 8;
-          SDValue srcAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(),
-                                        OutVals[i],
-                                        DAG.getConstant(curOffset,
-                                                        getPointerTy()));
-          SDValue theVal = DAG.getLoad(elemtype, dl, tempChain, srcAddr,
-                                MachinePointerInfo(), false, false, false, 0);
+          if (elemtype.isInteger() && (sz < 8))
+            sz = 8;
+          SDValue srcAddr =
+              DAG.getNode(ISD::ADD, dl, getPointerTy(), OutVals[i],
+                          DAG.getConstant(curOffset, getPointerTy()));
+          SDValue theVal =
+              DAG.getLoad(elemtype, dl, tempChain, srcAddr,
+                          MachinePointerInfo(), false, false, false, 0);
           SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
-          SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount,
-                                                            MVT::i32),
-                                           DAG.getConstant(curOffset, MVT::i32),
-                                                            theVal, InFlag };
+          SDValue CopyParamOps[] = { Chain,
+                                     DAG.getConstant(paramCount, MVT::i32),
+                                     DAG.getConstant(curOffset, MVT::i32),
+                                     theVal, InFlag };
           Chain = DAG.getNode(NVPTXISD::StoreParam, dl, CopyParamVTs,
                               CopyParamOps, 5);
           InFlag = Chain.getValue(1);
-          curOffset += sz/8;
+          curOffset += sz / 8;
         }
       }
       ++paramCount;
@@ -562,30 +555,31 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     // Non-abi, struct or vector
     // Declare a bunch or .reg .b<size> .param<n>
     unsigned curOffset = 0;
-    for (unsigned j=0,je=vtparts.size(); j!=je; ++j) {
+    for (unsigned j = 0, je = vtparts.size(); j != je; ++j) {
       unsigned elems = 1;
       EVT elemtype = vtparts[j];
       if (vtparts[j].isVector()) {
         elems = vtparts[j].getVectorNumElements();
         elemtype = vtparts[j].getVectorElementType();
       }
-      for (unsigned k=0,ke=elems; k!=ke; ++k) {
+      for (unsigned k = 0, ke = elems; k != ke; ++k) {
         unsigned sz = elemtype.getSizeInBits();
-        if (elemtype.isInteger() && (sz < 32)) sz = 32;
+        if (elemtype.isInteger() && (sz < 32))
+          sz = 32;
         SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
-        SDValue DeclareParamOps[] = { Chain, DAG.getConstant(paramCount,
-                                                             MVT::i32),
-                                                  DAG.getConstant(sz, MVT::i32),
-                                                   DAG.getConstant(1, MVT::i32),
-                                                             InFlag };
+        SDValue DeclareParamOps[] = { Chain,
+                                      DAG.getConstant(paramCount, MVT::i32),
+                                      DAG.getConstant(sz, MVT::i32),
+                                      DAG.getConstant(1, MVT::i32), InFlag };
         Chain = DAG.getNode(NVPTXISD::DeclareScalarParam, dl, DeclareParamVTs,
                             DeclareParamOps, 5);
         InFlag = Chain.getValue(1);
-        SDValue srcAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(), OutVals[i],
-                                      DAG.getConstant(curOffset,
-                                                      getPointerTy()));
-        SDValue theVal = DAG.getLoad(elemtype, dl, tempChain, srcAddr,
-                                  MachinePointerInfo(), false, false, false, 0);
+        SDValue srcAddr =
+            DAG.getNode(ISD::ADD, dl, getPointerTy(), OutVals[i],
+                        DAG.getConstant(curOffset, getPointerTy()));
+        SDValue theVal =
+            DAG.getLoad(elemtype, dl, tempChain, srcAddr, MachinePointerInfo(),
+                        false, false, false, 0);
         SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
         SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
                                    DAG.getConstant(0, MVT::i32), theVal,
@@ -610,20 +604,21 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     // Declare one .param .align 16 .b8 func_retval0[<size>] for ABI or
     // individual .reg .b<size> func_retval<0..> for non ABI
     unsigned resultsz = 0;
-    for (unsigned i=0,e=resvtparts.size(); i!=e; ++i) {
+    for (unsigned i = 0, e = resvtparts.size(); i != e; ++i) {
       unsigned elems = 1;
       EVT elemtype = resvtparts[i];
       if (resvtparts[i].isVector()) {
         elems = resvtparts[i].getVectorNumElements();
         elemtype = resvtparts[i].getVectorElementType();
       }
-      for (unsigned j=0,je=elems; j!=je; ++j) {
+      for (unsigned j = 0, je = elems; j != je; ++j) {
         unsigned sz = elemtype.getSizeInBits();
         if (isABI == false) {
-          if (elemtype.isInteger() && (sz < 32)) sz = 32;
-        }
-        else {
-          if (elemtype.isInteger() && (sz < 8)) sz = 8;
+          if (elemtype.isInteger() && (sz < 32))
+            sz = 32;
+        } else {
+          if (elemtype.isInteger() && (sz < 8))
+            sz = 8;
         }
         if (isABI == false) {
           SDVTList DeclareRetVTs = DAG.getVTList(MVT::Other, MVT::Glue);
@@ -641,7 +636,7 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     }
     if (isABI) {
       if (retTy->isPrimitiveType() || retTy->isIntegerTy() ||
-          retTy->isPointerTy() ) {
+          retTy->isPointerTy()) {
         // Scalar needs to be at least 32bit wide
         if (resultsz < 32)
           resultsz = 32;
@@ -652,8 +647,7 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
         Chain = DAG.getNode(NVPTXISD::DeclareRet, dl, DeclareRetVTs,
                             DeclareRetOps, 5);
         InFlag = Chain.getValue(1);
-      }
-      else {
+      } else {
         if (Func) { // direct call
           if (!llvm::getAlign(*(CS->getCalledFunction()), 0, retAlignment))
             retAlignment = getDataLayout()->getABITypeAlignment(retTy);
@@ -663,10 +657,10 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
             retAlignment = getDataLayout()->getABITypeAlignment(retTy);
         }
         SDVTList DeclareRetVTs = DAG.getVTList(MVT::Other, MVT::Glue);
-        SDValue DeclareRetOps[] = { Chain, DAG.getConstant(retAlignment,
-                                                           MVT::i32),
-                                          DAG.getConstant(resultsz/8, MVT::i32),
-                                         DAG.getConstant(0, MVT::i32), InFlag };
+        SDValue DeclareRetOps[] = { Chain,
+                                    DAG.getConstant(retAlignment, MVT::i32),
+                                    DAG.getConstant(resultsz / 8, MVT::i32),
+                                    DAG.getConstant(0, MVT::i32), InFlag };
         Chain = DAG.getNode(NVPTXISD::DeclareRetParam, dl, DeclareRetVTs,
                             DeclareRetOps, 5);
         InFlag = Chain.getValue(1);
@@ -684,24 +678,24 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     // INLINEASM SDNode.
     SDVTList InlineAsmVTs = DAG.getVTList(MVT::Other, MVT::Glue);
     std::string proto_string = getPrototype(retTy, Args, Outs, retAlignment);
-    const char *asmstr = nvTM->getManagedStrPool()->
-        getManagedString(proto_string.c_str())->c_str();
-    SDValue InlineAsmOps[] = { Chain,
-                               DAG.getTargetExternalSymbol(asmstr,
-                                                           getPointerTy()),
-                                                           DAG.getMDNode(0),
-                                   DAG.getTargetConstant(0, MVT::i32), InFlag };
+    const char *asmstr = nvTM->getManagedStrPool()
+        ->getManagedString(proto_string.c_str())->c_str();
+    SDValue InlineAsmOps[] = {
+      Chain, DAG.getTargetExternalSymbol(asmstr, getPointerTy()),
+      DAG.getMDNode(0), DAG.getTargetConstant(0, MVT::i32), InFlag
+    };
     Chain = DAG.getNode(ISD::INLINEASM, dl, InlineAsmVTs, InlineAsmOps, 5);
     InFlag = Chain.getValue(1);
   }
   // Op to just print "call"
   SDVTList PrintCallVTs = DAG.getVTList(MVT::Other, MVT::Glue);
-  SDValue PrintCallOps[] = { Chain,
-                             DAG.getConstant(isABI ? ((Ins.size()==0) ? 0 : 1)
-                                 : retCount, MVT::i32),
-                                   InFlag };
-  Chain = DAG.getNode(Func?(NVPTXISD::PrintCallUni):(NVPTXISD::PrintCall), dl,
-      PrintCallVTs, PrintCallOps, 3);
+  SDValue PrintCallOps[] = {
+    Chain,
+    DAG.getConstant(isABI ? ((Ins.size() == 0) ? 0 : 1) : retCount, MVT::i32),
+    InFlag
+  };
+  Chain = DAG.getNode(Func ? (NVPTXISD::PrintCallUni) : (NVPTXISD::PrintCall),
+                      dl, PrintCallVTs, PrintCallOps, 3);
   InFlag = Chain.getValue(1);
 
   // Ops to print out the function name
@@ -717,31 +711,28 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                       CallArgBeginOps, 2);
   InFlag = Chain.getValue(1);
 
-  for (unsigned i=0, e=paramCount; i!=e; ++i) {
+  for (unsigned i = 0, e = paramCount; i != e; ++i) {
     unsigned opcode;
-    if (i==(e-1))
+    if (i == (e - 1))
       opcode = NVPTXISD::LastCallArg;
     else
       opcode = NVPTXISD::CallArg;
     SDVTList CallArgVTs = DAG.getVTList(MVT::Other, MVT::Glue);
     SDValue CallArgOps[] = { Chain, DAG.getConstant(1, MVT::i32),
-                             DAG.getConstant(i, MVT::i32),
-                             InFlag };
+                             DAG.getConstant(i, MVT::i32), InFlag };
     Chain = DAG.getNode(opcode, dl, CallArgVTs, CallArgOps, 4);
     InFlag = Chain.getValue(1);
   }
   SDVTList CallArgEndVTs = DAG.getVTList(MVT::Other, MVT::Glue);
-  SDValue CallArgEndOps[] = { Chain,
-                              DAG.getConstant(Func ? 1 : 0, MVT::i32),
+  SDValue CallArgEndOps[] = { Chain, DAG.getConstant(Func ? 1 : 0, MVT::i32),
                               InFlag };
-  Chain = DAG.getNode(NVPTXISD::CallArgEnd, dl, CallArgEndVTs, CallArgEndOps,
-                      3);
+  Chain =
+      DAG.getNode(NVPTXISD::CallArgEnd, dl, CallArgEndVTs, CallArgEndOps, 3);
   InFlag = Chain.getValue(1);
 
   if (!Func) {
     SDVTList PrototypeVTs = DAG.getVTList(MVT::Other, MVT::Glue);
-    SDValue PrototypeOps[] = { Chain,
-                               DAG.getConstant(uniqueCallSite, MVT::i32),
+    SDValue PrototypeOps[] = { Chain, DAG.getConstant(uniqueCallSite, MVT::i32),
                                InFlag };
     Chain = DAG.getNode(NVPTXISD::Prototype, dl, PrototypeVTs, PrototypeOps, 3);
     InFlag = Chain.getValue(1);
@@ -751,33 +742,28 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   if (Ins.size() > 0) {
     if (isABI) {
       unsigned resoffset = 0;
-      for (unsigned i=0,e=Ins.size(); i!=e; ++i) {
+      for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
         unsigned sz = Ins[i].VT.getSizeInBits();
-        if (Ins[i].VT.isInteger() && (sz < 8)) sz = 8;
-        std::vector<EVT> LoadRetVTs;
-        LoadRetVTs.push_back(Ins[i].VT);
-        LoadRetVTs.push_back(MVT::Other); LoadRetVTs.push_back(MVT::Glue);
-        std::vector<SDValue> LoadRetOps;
-        LoadRetOps.push_back(Chain);
-        LoadRetOps.push_back(DAG.getConstant(1, MVT::i32));
-        LoadRetOps.push_back(DAG.getConstant(resoffset, MVT::i32));
-        LoadRetOps.push_back(InFlag);
+        if (Ins[i].VT.isInteger() && (sz < 8))
+          sz = 8;
+        EVT LoadRetVTs[] = { Ins[i].VT, MVT::Other, MVT::Glue };
+        SDValue LoadRetOps[] = { Chain, DAG.getConstant(1, MVT::i32),
+                                 DAG.getConstant(resoffset, MVT::i32), InFlag };
         SDValue retval = DAG.getNode(NVPTXISD::LoadParam, dl, LoadRetVTs,
-                                     &LoadRetOps[0], LoadRetOps.size());
+                                     LoadRetOps, array_lengthof(LoadRetOps));
         Chain = retval.getValue(1);
         InFlag = retval.getValue(2);
         InVals.push_back(retval);
-        resoffset += sz/8;
+        resoffset += sz / 8;
       }
-    }
-    else {
+    } else {
       SmallVector<EVT, 16> resvtparts;
       ComputeValueVTs(*this, retTy, resvtparts);
 
       assert(Ins.size() == resvtparts.size() &&
              "Unexpected number of return values in non-ABI case");
       unsigned paramNum = 0;
-      for (unsigned i=0,e=Ins.size(); i!=e; ++i) {
+      for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
         assert(EVT(Ins[i].VT) == resvtparts[i] &&
                "Unexpected EVT type in non-ABI case");
         unsigned numelems = 1;
@@ -787,17 +773,13 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
           elemtype = Ins[i].VT.getVectorElementType();
         }
         std::vector<SDValue> tempRetVals;
-        for (unsigned j=0; j<numelems; ++j) {
-          std::vector<EVT> MoveRetVTs;
-          MoveRetVTs.push_back(elemtype);
-          MoveRetVTs.push_back(MVT::Other); MoveRetVTs.push_back(MVT::Glue);
-          std::vector<SDValue> MoveRetOps;
-          MoveRetOps.push_back(Chain);
-          MoveRetOps.push_back(DAG.getConstant(0, MVT::i32));
-          MoveRetOps.push_back(DAG.getConstant(paramNum, MVT::i32));
-          MoveRetOps.push_back(InFlag);
+        for (unsigned j = 0; j < numelems; ++j) {
+          EVT MoveRetVTs[] = { elemtype, MVT::Other, MVT::Glue };
+          SDValue MoveRetOps[] = { Chain, DAG.getConstant(0, MVT::i32),
+                                   DAG.getConstant(paramNum, MVT::i32),
+                                   InFlag };
           SDValue retval = DAG.getNode(NVPTXISD::LoadParam, dl, MoveRetVTs,
-                                       &MoveRetOps[0], MoveRetOps.size());
+                                       MoveRetOps, array_lengthof(MoveRetOps));
           Chain = retval.getValue(1);
           InFlag = retval.getValue(2);
           tempRetVals.push_back(retval);
@@ -811,9 +793,8 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       }
     }
   }
-  Chain = DAG.getCALLSEQ_END(Chain,
-                             DAG.getIntPtrConstant(uniqueCallSite, true),
-                             DAG.getIntPtrConstant(uniqueCallSite+1, true),
+  Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(uniqueCallSite, true),
+                             DAG.getIntPtrConstant(uniqueCallSite + 1, true),
                              InFlag);
   uniqueCallSite++;
 
@@ -826,76 +807,183 @@ NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
 // By default CONCAT_VECTORS is lowered by ExpandVectorBuildThroughStack()
 // (see LegalizeDAG.cpp). This is slow and uses local memory.
 // We use extract/insert/build vector just as what LegalizeOp() does in llvm 2.5
-SDValue NVPTXTargetLowering::
-LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const {
+SDValue
+NVPTXTargetLowering::LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const {
   SDNode *Node = Op.getNode();
   DebugLoc dl = Node->getDebugLoc();
   SmallVector<SDValue, 8> Ops;
   unsigned NumOperands = Node->getNumOperands();
-  for (unsigned i=0; i < NumOperands; ++i) {
+  for (unsigned i = 0; i < NumOperands; ++i) {
     SDValue SubOp = Node->getOperand(i);
     EVT VVT = SubOp.getNode()->getValueType(0);
     EVT EltVT = VVT.getVectorElementType();
     unsigned NumSubElem = VVT.getVectorNumElements();
-    for (unsigned j=0; j < NumSubElem; ++j) {
+    for (unsigned j = 0; j < NumSubElem; ++j) {
       Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, SubOp,
                                 DAG.getIntPtrConstant(j)));
     }
   }
-  return DAG.getNode(ISD::BUILD_VECTOR, dl, Node->getValueType(0),
-                     &Ops[0], Ops.size());
+  return DAG.getNode(ISD::BUILD_VECTOR, dl, Node->getValueType(0), &Ops[0],
+                     Ops.size());
 }
 
-SDValue NVPTXTargetLowering::
-LowerOperation(SDValue Op, SelectionDAG &DAG) const {
+SDValue
+NVPTXTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   switch (Op.getOpcode()) {
-  case ISD::RETURNADDR: return SDValue();
-  case ISD::FRAMEADDR:  return SDValue();
-  case ISD::GlobalAddress:      return LowerGlobalAddress(Op, DAG);
-  case ISD::INTRINSIC_W_CHAIN: return Op;
+  case ISD::RETURNADDR:
+    return SDValue();
+  case ISD::FRAMEADDR:
+    return SDValue();
+  case ISD::GlobalAddress:
+    return LowerGlobalAddress(Op, DAG);
+  case ISD::INTRINSIC_W_CHAIN:
+    return Op;
   case ISD::BUILD_VECTOR:
   case ISD::EXTRACT_SUBVECTOR:
     return Op;
-  case ISD::CONCAT_VECTORS: return LowerCONCAT_VECTORS(Op, DAG);
-  case ISD::STORE: return LowerSTORE(Op, DAG);
-  case ISD::LOAD: return LowerLOAD(Op, DAG);
+  case ISD::CONCAT_VECTORS:
+    return LowerCONCAT_VECTORS(Op, DAG);
+  case ISD::STORE:
+    return LowerSTORE(Op, DAG);
+  case ISD::LOAD:
+    return LowerLOAD(Op, DAG);
   default:
     llvm_unreachable("Custom lowering not defined for operation");
   }
 }
 
+SDValue NVPTXTargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
+  if (Op.getValueType() == MVT::i1)
+    return LowerLOADi1(Op, DAG);
+  else
+    return SDValue();
+}
 
 // v = ld i1* addr
 //   =>
 // v1 = ld i8* addr
 // v = trunc v1 to i1
-SDValue NVPTXTargetLowering::
-LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
+SDValue NVPTXTargetLowering::LowerLOADi1(SDValue Op, SelectionDAG &DAG) const {
   SDNode *Node = Op.getNode();
   LoadSDNode *LD = cast<LoadSDNode>(Node);
   DebugLoc dl = Node->getDebugLoc();
-  assert(LD->getExtensionType() == ISD::NON_EXTLOAD) ;
+  assert(LD->getExtensionType() == ISD::NON_EXTLOAD);
   assert(Node->getValueType(0) == MVT::i1 &&
          "Custom lowering for i1 load only");
-  SDValue newLD = DAG.getLoad(MVT::i8, dl, LD->getChain(), LD->getBasePtr(),
-                              LD->getPointerInfo(),
-                              LD->isVolatile(), LD->isNonTemporal(),
-                              LD->isInvariant(),
-                              LD->getAlignment());
+  SDValue newLD =
+      DAG.getLoad(MVT::i8, dl, LD->getChain(), LD->getBasePtr(),
+                  LD->getPointerInfo(), LD->isVolatile(), LD->isNonTemporal(),
+                  LD->isInvariant(), LD->getAlignment());
   SDValue result = DAG.getNode(ISD::TRUNCATE, dl, MVT::i1, newLD);
   // The legalizer (the caller) is expecting two values from the legalized
   // load, so we build a MergeValues node for it. See ExpandUnalignedLoad()
   // in LegalizeDAG.cpp which also uses MergeValues.
-  SDValue Ops[] = {result, LD->getChain()};
+  SDValue Ops[] = { result, LD->getChain() };
   return DAG.getMergeValues(Ops, 2, dl);
 }
 
+SDValue NVPTXTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
+  EVT ValVT = Op.getOperand(1).getValueType();
+  if (ValVT == MVT::i1)
+    return LowerSTOREi1(Op, DAG);
+  else if (ValVT.isVector())
+    return LowerSTOREVector(Op, DAG);
+  else
+    return SDValue();
+}
+
+SDValue
+NVPTXTargetLowering::LowerSTOREVector(SDValue Op, SelectionDAG &DAG) const {
+  SDNode *N = Op.getNode();
+  SDValue Val = N->getOperand(1);
+  DebugLoc DL = N->getDebugLoc();
+  EVT ValVT = Val.getValueType();
+
+  if (ValVT.isVector()) {
+    // We only handle "native" vector sizes for now, e.g. <4 x double> is not
+    // legal.  We can (and should) split that into 2 stores of <2 x double> here
+    // but I'm leaving that as a TODO for now.
+    if (!ValVT.isSimple())
+      return SDValue();
+    switch (ValVT.getSimpleVT().SimpleTy) {
+    default:
+      return SDValue();
+    case MVT::v2i8:
+    case MVT::v2i16:
+    case MVT::v2i32:
+    case MVT::v2i64:
+    case MVT::v2f32:
+    case MVT::v2f64:
+    case MVT::v4i8:
+    case MVT::v4i16:
+    case MVT::v4i32:
+    case MVT::v4f32:
+      // This is a "native" vector type
+      break;
+    }
+
+    unsigned Opcode = 0;
+    EVT EltVT = ValVT.getVectorElementType();
+    unsigned NumElts = ValVT.getVectorNumElements();
+
+    // Since StoreV2 is a target node, we cannot rely on DAG type legalization.
+    // Therefore, we must ensure the type is legal.  For i1 and i8, we set the
+    // stored type to i16 and propogate the "real" type as the memory type.
+    bool NeedExt = false;
+    if (EltVT.getSizeInBits() < 16)
+      NeedExt = true;
+
+    switch (NumElts) {
+    default:
+      return SDValue();
+    case 2:
+      Opcode = NVPTXISD::StoreV2;
+      break;
+    case 4: {
+      Opcode = NVPTXISD::StoreV4;
+      break;
+    }
+    }
+
+    SmallVector<SDValue, 8> Ops;
+
+    // First is the chain
+    Ops.push_back(N->getOperand(0));
+
+    // Then the split values
+    for (unsigned i = 0; i < NumElts; ++i) {
+      SDValue ExtVal = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltVT, Val,
+                                   DAG.getIntPtrConstant(i));
+      if (NeedExt)
+        // ANY_EXTEND is correct here since the store will only look at the
+        // lower-order bits anyway.
+        ExtVal = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i16, ExtVal);
+      Ops.push_back(ExtVal);
+    }
+
+    // Then any remaining arguments
+    for (unsigned i = 2, e = N->getNumOperands(); i != e; ++i) {
+      Ops.push_back(N->getOperand(i));
+    }
+
+    MemSDNode *MemSD = cast<MemSDNode>(N);
+
+    SDValue NewSt = DAG.getMemIntrinsicNode(
+        Opcode, DL, DAG.getVTList(MVT::Other), &Ops[0], Ops.size(),
+        MemSD->getMemoryVT(), MemSD->getMemOperand());
+
+    //return DCI.CombineTo(N, NewSt, true);
+    return NewSt;
+  }
+
+  return SDValue();
+}
+
 // st i1 v, addr
 //    =>
 // v1 = zxt v to i8
 // st i8, addr
-SDValue NVPTXTargetLowering::
-LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
+SDValue NVPTXTargetLowering::LowerSTOREi1(SDValue Op, SelectionDAG &DAG) const {
   SDNode *Node = Op.getNode();
   DebugLoc dl = Node->getDebugLoc();
   StoreSDNode *ST = cast<StoreSDNode>(Node);
@@ -906,18 +994,14 @@ LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
   unsigned Alignment = ST->getAlignment();
   bool isVolatile = ST->isVolatile();
   bool isNonTemporal = ST->isNonTemporal();
-  Tmp3 = DAG.getNode(ISD::ZERO_EXTEND, dl,
-                     MVT::i8, Tmp3);
-  SDValue Result = DAG.getStore(Tmp1, dl, Tmp3, Tmp2,
-                                ST->getPointerInfo(), isVolatile,
-                                isNonTemporal, Alignment);
+  Tmp3 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i8, Tmp3);
+  SDValue Result = DAG.getStore(Tmp1, dl, Tmp3, Tmp2, ST->getPointerInfo(),
+                                isVolatile, isNonTemporal, Alignment);
   return Result;
 }
 
-
-SDValue
-NVPTXTargetLowering::getExtSymb(SelectionDAG &DAG, const char *inname, int idx,
-                                EVT v) const {
+SDValue NVPTXTargetLowering::getExtSymb(SelectionDAG &DAG, const char *inname,
+                                        int idx, EVT v) const {
   std::string *name = nvTM->getManagedStrPool()->getManagedString(inname);
   std::stringstream suffix;
   suffix << idx;
@@ -930,19 +1014,16 @@ NVPTXTargetLowering::getParamSymbol(SelectionDAG &DAG, int idx, EVT v) const {
   return getExtSymb(DAG, ".PARAM", idx, v);
 }
 
-SDValue
-NVPTXTargetLowering::getParamHelpSymbol(SelectionDAG &DAG, int idx) {
+SDValue NVPTXTargetLowering::getParamHelpSymbol(SelectionDAG &DAG, int idx) {
   return getExtSymb(DAG, ".HLPPARAM", idx);
 }
 
 // Check to see if the kernel argument is image*_t or sampler_t
 
 bool llvm::isImageOrSamplerVal(const Value *arg, const Module *context) {
-  static const char *const specialTypes[] = {
-                                             "struct._image2d_t",
-                                             "struct._image3d_t",
-                                             "struct._sampler_t"
-  };
+  static const char *const specialTypes[] = { "struct._image2d_t",
+                                              "struct._image3d_t",
+                                              "struct._sampler_t" };
 
   const Type *Ty = arg->getType();
   const PointerType *PTy = dyn_cast<PointerType>(Ty);
@@ -954,7 +1035,7 @@ bool llvm::isImageOrSamplerVal(const Value *arg, const Module *context) {
     return false;
 
   const StructType *STy = dyn_cast<StructType>(PTy->getElementType());
-  const std::string TypeName = STy ? STy->getName() : "";
+  const std::string TypeName = STy && !STy->isLiteral() ? STy->getName() : "";
 
   for (int i = 0, e = array_lengthof(specialTypes); i != e; ++i)
     if (TypeName == specialTypes[i])
@@ -963,17 +1044,15 @@ bool llvm::isImageOrSamplerVal(const Value *arg, const Module *context) {
   return false;
 }
 
-SDValue
-NVPTXTargetLowering::LowerFormalArguments(SDValue Chain,
-                                        CallingConv::ID CallConv, bool isVarArg,
-                                      const SmallVectorImpl<ISD::InputArg> &Ins,
-                                          DebugLoc dl, SelectionDAG &DAG,
-                                       SmallVectorImpl<SDValue> &InVals) const {
+SDValue NVPTXTargetLowering::LowerFormalArguments(
+    SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
+    const SmallVectorImpl<ISD::InputArg> &Ins, DebugLoc dl, SelectionDAG &DAG,
+    SmallVectorImpl<SDValue> &InVals) const {
   MachineFunction &MF = DAG.getMachineFunction();
   const DataLayout *TD = getDataLayout();
 
   const Function *F = MF.getFunction();
-  const AttrListPtr &PAL = F->getAttributes();
+  const AttributeSet &PAL = F->getAttributes();
 
   SDValue Root = DAG.getRoot();
   std::vector<SDValue> OutChains;
@@ -984,34 +1063,43 @@ NVPTXTargetLowering::LowerFormalArguments(SDValue Chain,
   std::vector<Type *> argTypes;
   std::vector<const Argument *> theArgs;
   for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
-      I != E; ++I) {
+       I != E; ++I) {
     theArgs.push_back(I);
     argTypes.push_back(I->getType());
   }
-  assert(argTypes.size() == Ins.size() &&
-         "Ins types and function types did not match");
+  //assert(argTypes.size() == Ins.size() &&
+  //       "Ins types and function types did not match");
 
   int idx = 0;
-  for (unsigned i=0, e=Ins.size(); i!=e; ++i, ++idx) {
+  for (unsigned i = 0, e = argTypes.size(); i != e; ++i, ++idx) {
     Type *Ty = argTypes[i];
     EVT ObjectVT = getValueType(Ty);
-    assert(ObjectVT == Ins[i].VT &&
-           "Ins type did not match function type");
+    //assert(ObjectVT == Ins[i].VT &&
+    //       "Ins type did not match function type");
 
     // If the kernel argument is image*_t or sampler_t, convert it to
     // a i32 constant holding the parameter position. This can later
     // matched in the AsmPrinter to output the correct mangled name.
-    if (isImageOrSamplerVal(theArgs[i],
-                           (theArgs[i]->getParent() ?
-                               theArgs[i]->getParent()->getParent() : 0))) {
+    if (isImageOrSamplerVal(
+            theArgs[i],
+            (theArgs[i]->getParent() ? theArgs[i]->getParent()->getParent()
+                                     : 0))) {
       assert(isKernel && "Only kernels can have image/sampler params");
-      InVals.push_back(DAG.getConstant(i+1, MVT::i32));
+      InVals.push_back(DAG.getConstant(i + 1, MVT::i32));
       continue;
     }
 
     if (theArgs[i]->use_empty()) {
       // argument is dead
-      InVals.push_back(DAG.getNode(ISD::UNDEF, dl, ObjectVT));
+      if (ObjectVT.isVector()) {
+        EVT EltVT = ObjectVT.getVectorElementType();
+        unsigned NumElts = ObjectVT.getVectorNumElements();
+        for (unsigned vi = 0; vi < NumElts; ++vi) {
+          InVals.push_back(DAG.getNode(ISD::UNDEF, dl, EltVT));
+        }
+      } else {
+        InVals.push_back(DAG.getNode(ISD::UNDEF, dl, ObjectVT));
+      }
       continue;
     }
 
@@ -1019,29 +1107,52 @@ NVPTXTargetLowering::LowerFormalArguments(SDValue Chain,
     // to newly created nodes. The SDNOdes for params have to
     // appear in the same order as their order of appearance
     // in the original function. "idx+1" holds that order.
-    if (PAL.getParamAttributes(i+1).hasAttribute(Attributes::ByVal) == false) {
+    if (PAL.hasAttribute(i + 1, Attribute::ByVal) == false) {
+      if (ObjectVT.isVector()) {
+        unsigned NumElts = ObjectVT.getVectorNumElements();
+        EVT EltVT = ObjectVT.getVectorElementType();
+        unsigned Offset = 0;
+        for (unsigned vi = 0; vi < NumElts; ++vi) {
+          SDValue A = getParamSymbol(DAG, idx, getPointerTy());
+          SDValue B = DAG.getIntPtrConstant(Offset);
+          SDValue Addr = DAG.getNode(ISD::ADD, dl, getPointerTy(),
+                                     //getParamSymbol(DAG, idx, EltVT),
+                                     //DAG.getConstant(Offset, getPointerTy()));
+                                     A, B);
+          Value *SrcValue = Constant::getNullValue(PointerType::get(
+              EltVT.getTypeForEVT(F->getContext()), llvm::ADDRESS_SPACE_PARAM));
+          SDValue Ld = DAG.getLoad(
+              EltVT, dl, Root, Addr, MachinePointerInfo(SrcValue), false, false,
+              false,
+              TD->getABITypeAlignment(EltVT.getTypeForEVT(F->getContext())));
+          Offset += EltVT.getStoreSizeInBits() / 8;
+          InVals.push_back(Ld);
+        }
+        continue;
+      }
+
       // A plain scalar.
       if (isABI || isKernel) {
         // If ABI, load from the param symbol
         SDValue Arg = getParamSymbol(DAG, idx);
-        Value *srcValue = new Argument(PointerType::get(ObjectVT.getTypeForEVT(
-            F->getContext()),
-            llvm::ADDRESS_SPACE_PARAM));
-        SDValue p = DAG.getLoad(ObjectVT, dl, Root, Arg,
-                                MachinePointerInfo(srcValue), false, false,
-                                false,
-                                TD->getABITypeAlignment(ObjectVT.getTypeForEVT(
-                                  F->getContext())));
+        // Conjure up a value that we can get the address space from.
+        // FIXME: Using a constant here is a hack.
+        Value *srcValue = Constant::getNullValue(
+            PointerType::get(ObjectVT.getTypeForEVT(F->getContext()),
+                             llvm::ADDRESS_SPACE_PARAM));
+        SDValue p = DAG.getLoad(
+            ObjectVT, dl, Root, Arg, MachinePointerInfo(srcValue), false, false,
+            false,
+            TD->getABITypeAlignment(ObjectVT.getTypeForEVT(F->getContext())));
         if (p.getNode())
-          DAG.AssignOrdering(p.getNode(), idx+1);
+          DAG.AssignOrdering(p.getNode(), idx + 1);
         InVals.push_back(p);
-      }
-      else {
+      } else {
         // If no ABI, just move the param symbol
         SDValue Arg = getParamSymbol(DAG, idx, ObjectVT);
         SDValue p = DAG.getNode(NVPTXISD::MoveParam, dl, ObjectVT, Arg);
         if (p.getNode())
-          DAG.AssignOrdering(p.getNode(), idx+1);
+          DAG.AssignOrdering(p.getNode(), idx + 1);
         InVals.push_back(p);
       }
       continue;
@@ -1058,47 +1169,49 @@ NVPTXTargetLowering::LowerFormalArguments(SDValue Chain,
       SDValue Arg = getParamSymbol(DAG, idx, getPointerTy());
       SDValue p = DAG.getNode(NVPTXISD::MoveParam, dl, ObjectVT, Arg);
       if (p.getNode())
-        DAG.AssignOrdering(p.getNode(), idx+1);
+        DAG.AssignOrdering(p.getNode(), idx + 1);
       if (isKernel)
         InVals.push_back(p);
       else {
-        SDValue p2 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, ObjectVT,
-                    DAG.getConstant(Intrinsic::nvvm_ptr_local_to_gen, MVT::i32),
-                                 p);
+        SDValue p2 = DAG.getNode(
+            ISD::INTRINSIC_WO_CHAIN, dl, ObjectVT,
+            DAG.getConstant(Intrinsic::nvvm_ptr_local_to_gen, MVT::i32), p);
         InVals.push_back(p2);
       }
     } else {
       // Have to move a set of param symbols to registers and
       // store them locally and return the local pointer in InVals
       const PointerType *elemPtrType = dyn_cast<PointerType>(argTypes[i]);
-      assert(elemPtrType &&
-             "Byval parameter should be a pointer type");
+      assert(elemPtrType && "Byval parameter should be a pointer type");
       Type *elemType = elemPtrType->getElementType();
       // Compute the constituent parts
       SmallVector<EVT, 16> vtparts;
       SmallVector<uint64_t, 16> offsets;
       ComputeValueVTs(*this, elemType, vtparts, &offsets, 0);
       unsigned totalsize = 0;
-      for (unsigned j=0, je=vtparts.size(); j!=je; ++j)
+      for (unsigned j = 0, je = vtparts.size(); j != je; ++j)
         totalsize += vtparts[j].getStoreSizeInBits();
-      SDValue localcopy =  DAG.getFrameIndex(MF.getFrameInfo()->
-                                      CreateStackObject(totalsize/8, 16, false),
-                                             getPointerTy());
+      SDValue localcopy = DAG.getFrameIndex(
+          MF.getFrameInfo()->CreateStackObject(totalsize / 8, 16, false),
+          getPointerTy());
       unsigned sizesofar = 0;
       std::vector<SDValue> theChains;
-      for (unsigned j=0, je=vtparts.size(); j!=je; ++j) {
+      for (unsigned j = 0, je = vtparts.size(); j != je; ++j) {
         unsigned numElems = 1;
-        if (vtparts[j].isVector()) numElems = vtparts[j].getVectorNumElements();
-        for (unsigned k=0, ke=numElems; k!=ke; ++k) {
+        if (vtparts[j].isVector())
+          numElems = vtparts[j].getVectorNumElements();
+        for (unsigned k = 0, ke = numElems; k != ke; ++k) {
           EVT tmpvt = vtparts[j];
-          if (tmpvt.isVector()) tmpvt = tmpvt.getVectorElementType();
+          if (tmpvt.isVector())
+            tmpvt = tmpvt.getVectorElementType();
           SDValue arg = DAG.getNode(NVPTXISD::MoveParam, dl, tmpvt,
                                     getParamSymbol(DAG, idx, tmpvt));
-          SDValue addr = DAG.getNode(ISD::ADD, dl, getPointerTy(), localcopy,
-                                    DAG.getConstant(sizesofar, getPointerTy()));
-          theChains.push_back(DAG.getStore(Chain, dl, arg, addr,
-                                        MachinePointerInfo(), false, false, 0));
-          sizesofar += tmpvt.getStoreSizeInBits()/8;
+          SDValue addr =
+              DAG.getNode(ISD::ADD, dl, getPointerTy(), localcopy,
+                          DAG.getConstant(sizesofar, getPointerTy()));
+          theChains.push_back(DAG.getStore(
+              Chain, dl, arg, addr, MachinePointerInfo(), false, false, 0));
+          sizesofar += tmpvt.getStoreSizeInBits() / 8;
           ++idx;
         }
       }
@@ -1118,43 +1231,42 @@ NVPTXTargetLowering::LowerFormalArguments(SDValue Chain,
   //}
 
   if (!OutChains.empty())
-    DAG.setRoot(DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
-                            &OutChains[0], OutChains.size()));
+    DAG.setRoot(DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &OutChains[0],
+                            OutChains.size()));
 
   return Chain;
 }
 
-SDValue
-NVPTXTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
-                                 bool isVarArg,
-                                 const SmallVectorImpl<ISD::OutputArg> &Outs,
-                                 const SmallVectorImpl<SDValue> &OutVals,
-                                 DebugLoc dl, SelectionDAG &DAG) const {
+SDValue NVPTXTargetLowering::LowerReturn(
+    SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
+    const SmallVectorImpl<ISD::OutputArg> &Outs,
+    const SmallVectorImpl<SDValue> &OutVals, DebugLoc dl,
+    SelectionDAG &DAG) const {
 
   bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
 
   unsigned sizesofar = 0;
   unsigned idx = 0;
-  for (unsigned i=0, e=Outs.size(); i!=e; ++i) {
+  for (unsigned i = 0, e = Outs.size(); i != e; ++i) {
     SDValue theVal = OutVals[i];
     EVT theValType = theVal.getValueType();
     unsigned numElems = 1;
-    if (theValType.isVector()) numElems = theValType.getVectorNumElements();
-    for (unsigned j=0,je=numElems; j!=je; ++j) {
+    if (theValType.isVector())
+      numElems = theValType.getVectorNumElements();
+    for (unsigned j = 0, je = numElems; j != je; ++j) {
       SDValue tmpval = theVal;
       if (theValType.isVector())
         tmpval = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
-                             theValType.getVectorElementType(),
-                             tmpval, DAG.getIntPtrConstant(j));
-      Chain = DAG.getNode(isABI ? NVPTXISD::StoreRetval :NVPTXISD::MoveToRetval,
-          dl, MVT::Other,
-          Chain,
-          DAG.getConstant(isABI ? sizesofar : idx, MVT::i32),
+                             theValType.getVectorElementType(), tmpval,
+                             DAG.getIntPtrConstant(j));
+      Chain = DAG.getNode(
+          isABI ? NVPTXISD::StoreRetval : NVPTXISD::MoveToRetval, dl,
+          MVT::Other, Chain, DAG.getConstant(isABI ? sizesofar : idx, MVT::i32),
           tmpval);
       if (theValType.isVector())
-        sizesofar += theValType.getVectorElementType().getStoreSizeInBits()/8;
+        sizesofar += theValType.getVectorElementType().getStoreSizeInBits() / 8;
       else
-        sizesofar += theValType.getStoreSizeInBits()/8;
+        sizesofar += theValType.getStoreSizeInBits() / 8;
       ++idx;
     }
   }
@@ -1162,12 +1274,9 @@ NVPTXTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
   return DAG.getNode(NVPTXISD::RET_FLAG, dl, MVT::Other, Chain);
 }
 
-void
-NVPTXTargetLowering::LowerAsmOperandForConstraint(SDValue Op,
-                                                  std::string &Constraint,
-                                                  std::vector<SDValue> &Ops,
-                                                  SelectionDAG &DAG) const
-{
+void NVPTXTargetLowering::LowerAsmOperandForConstraint(
+    SDValue Op, std::string &Constraint, std::vector<SDValue> &Ops,
+    SelectionDAG &DAG) const {
   if (Constraint.length() > 1)
     return;
   else
@@ -1177,8 +1286,7 @@ NVPTXTargetLowering::LowerAsmOperandForConstraint(SDValue Op,
 // NVPTX suuport vector of legal types of any length in Intrinsics because the
 // NVPTX specific type legalizer
 // will legalize them to the PTX supported length.
-bool
-NVPTXTargetLowering::isTypeSupportedInIntrinsic(MVT VT) const {
+bool NVPTXTargetLowering::isTypeSupportedInIntrinsic(MVT VT) const {
   if (isTypeLegal(VT))
     return true;
   if (VT.isVector()) {
@@ -1189,15 +1297,13 @@ NVPTXTargetLowering::isTypeSupportedInIntrinsic(MVT VT) const {
   return false;
 }
 
-
 // llvm.ptx.memcpy.const and llvm.ptx.memmove.const need to be modeled as
 // TgtMemIntrinsic
 // because we need the information that is only available in the "Value" type
 // of destination
 // pointer. In particular, the address space information.
-bool
-NVPTXTargetLowering::getTgtMemIntrinsic(IntrinsicInfo& Info, const CallInst &I,
-                                        unsigned Intrinsic) const {
+bool NVPTXTargetLowering::getTgtMemIntrinsic(
+    IntrinsicInfo &Info, const CallInst &I, unsigned Intrinsic) const {
   switch (Intrinsic) {
   default:
     return false;
@@ -1253,9 +1359,8 @@ NVPTXTargetLowering::getTgtMemIntrinsic(IntrinsicInfo& Info, const CallInst &I,
 /// Used to guide target specific optimizations, like loop strength reduction
 /// (LoopStrengthReduce.cpp) and memory optimization for address mode
 /// (CodeGenPrepare.cpp)
-bool
-NVPTXTargetLowering::isLegalAddressingMode(const AddrMode &AM,
-                                           Type *Ty) const {
+bool NVPTXTargetLowering::isLegalAddressingMode(const AddrMode &AM,
+                                                Type *Ty) const {
 
   // AddrMode - This represents an addressing mode of:
   //    BaseGV + BaseOffs + BaseReg + Scale*ScaleReg
@@ -1273,10 +1378,10 @@ NVPTXTargetLowering::isLegalAddressingMode(const AddrMode &AM,
   }
 
   switch (AM.Scale) {
-  case 0:  // "r", "r+i" or "i" is allowed
+  case 0: // "r", "r+i" or "i" is allowed
     break;
   case 1:
-    if (AM.HasBaseReg)  // "r+r+i" or "r+r" is not allowed.
+    if (AM.HasBaseReg) // "r+r+i" or "r+r" is not allowed.
       return false;
     // Otherwise we have r+i.
     break;
@@ -1313,8 +1418,7 @@ NVPTXTargetLowering::getConstraintType(const std::string &Constraint) const {
   return TargetLowering::getConstraintType(Constraint);
 }
 
-
-std::pair<unsigned, const TargetRegisterClass*>
+std::pair<unsigned, const TargetRegisterClass *>
 NVPTXTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
                                                   EVT VT) const {
   if (Constraint.size() == 1) {
@@ -1337,9 +1441,253 @@ NVPTXTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
   return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
 }
 
-
-
 /// getFunctionAlignment - Return the Log2 alignment of this function.
 unsigned NVPTXTargetLowering::getFunctionAlignment(const Function *) const {
   return 4;
 }
+
+/// ReplaceVectorLoad - Convert vector loads into multi-output scalar loads.
+static void ReplaceLoadVector(SDNode *N, SelectionDAG &DAG,
+                              SmallVectorImpl<SDValue> &Results) {
+  EVT ResVT = N->getValueType(0);
+  DebugLoc DL = N->getDebugLoc();
+
+  assert(ResVT.isVector() && "Vector load must have vector type");
+
+  // We only handle "native" vector sizes for now, e.g. <4 x double> is not
+  // legal.  We can (and should) split that into 2 loads of <2 x double> here
+  // but I'm leaving that as a TODO for now.
+  assert(ResVT.isSimple() && "Can only handle simple types");
+  switch (ResVT.getSimpleVT().SimpleTy) {
+  default:
+    return;
+  case MVT::v2i8:
+  case MVT::v2i16:
+  case MVT::v2i32:
+  case MVT::v2i64:
+  case MVT::v2f32:
+  case MVT::v2f64:
+  case MVT::v4i8:
+  case MVT::v4i16:
+  case MVT::v4i32:
+  case MVT::v4f32:
+    // This is a "native" vector type
+    break;
+  }
+
+  EVT EltVT = ResVT.getVectorElementType();
+  unsigned NumElts = ResVT.getVectorNumElements();
+
+  // Since LoadV2 is a target node, we cannot rely on DAG type legalization.
+  // Therefore, we must ensure the type is legal.  For i1 and i8, we set the
+  // loaded type to i16 and propogate the "real" type as the memory type.
+  bool NeedTrunc = false;
+  if (EltVT.getSizeInBits() < 16) {
+    EltVT = MVT::i16;
+    NeedTrunc = true;
+  }
+
+  unsigned Opcode = 0;
+  SDVTList LdResVTs;
+
+  switch (NumElts) {
+  default:
+    return;
+  case 2:
+    Opcode = NVPTXISD::LoadV2;
+    LdResVTs = DAG.getVTList(EltVT, EltVT, MVT::Other);
+    break;
+  case 4: {
+    Opcode = NVPTXISD::LoadV4;
+    EVT ListVTs[] = { EltVT, EltVT, EltVT, EltVT, MVT::Other };
+    LdResVTs = DAG.getVTList(ListVTs, 5);
+    break;
+  }
+  }
+
+  SmallVector<SDValue, 8> OtherOps;
+
+  // Copy regular operands
+  for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
+    OtherOps.push_back(N->getOperand(i));
+
+  LoadSDNode *LD = cast<LoadSDNode>(N);
+
+  // The select routine does not have access to the LoadSDNode instance, so
+  // pass along the extension information
+  OtherOps.push_back(DAG.getIntPtrConstant(LD->getExtensionType()));
+
+  SDValue NewLD = DAG.getMemIntrinsicNode(Opcode, DL, LdResVTs, &OtherOps[0],
+                                          OtherOps.size(), LD->getMemoryVT(),
+                                          LD->getMemOperand());
+
+  SmallVector<SDValue, 4> ScalarRes;
+
+  for (unsigned i = 0; i < NumElts; ++i) {
+    SDValue Res = NewLD.getValue(i);
+    if (NeedTrunc)
+      Res = DAG.getNode(ISD::TRUNCATE, DL, ResVT.getVectorElementType(), Res);
+    ScalarRes.push_back(Res);
+  }
+
+  SDValue LoadChain = NewLD.getValue(NumElts);
+
+  SDValue BuildVec =
+      DAG.getNode(ISD::BUILD_VECTOR, DL, ResVT, &ScalarRes[0], NumElts);
+
+  Results.push_back(BuildVec);
+  Results.push_back(LoadChain);
+}
+
+static void ReplaceINTRINSIC_W_CHAIN(SDNode *N, SelectionDAG &DAG,
+                                     SmallVectorImpl<SDValue> &Results) {
+  SDValue Chain = N->getOperand(0);
+  SDValue Intrin = N->getOperand(1);
+  DebugLoc DL = N->getDebugLoc();
+
+  // Get the intrinsic ID
+  unsigned IntrinNo = cast<ConstantSDNode>(Intrin.getNode())->getZExtValue();
+  switch (IntrinNo) {
+  default:
+    return;
+  case Intrinsic::nvvm_ldg_global_i:
+  case Intrinsic::nvvm_ldg_global_f:
+  case Intrinsic::nvvm_ldg_global_p:
+  case Intrinsic::nvvm_ldu_global_i:
+  case Intrinsic::nvvm_ldu_global_f:
+  case Intrinsic::nvvm_ldu_global_p: {
+    EVT ResVT = N->getValueType(0);
+
+    if (ResVT.isVector()) {
+      // Vector LDG/LDU
+
+      unsigned NumElts = ResVT.getVectorNumElements();
+      EVT EltVT = ResVT.getVectorElementType();
+
+      // Since LDU/LDG are target nodes, we cannot rely on DAG type legalization.
+      // Therefore, we must ensure the type is legal.  For i1 and i8, we set the
+      // loaded type to i16 and propogate the "real" type as the memory type.
+      bool NeedTrunc = false;
+      if (EltVT.getSizeInBits() < 16) {
+        EltVT = MVT::i16;
+        NeedTrunc = true;
+      }
+
+      unsigned Opcode = 0;
+      SDVTList LdResVTs;
+
+      switch (NumElts) {
+      default:
+        return;
+      case 2:
+        switch (IntrinNo) {
+        default:
+          return;
+        case Intrinsic::nvvm_ldg_global_i:
+        case Intrinsic::nvvm_ldg_global_f:
+        case Intrinsic::nvvm_ldg_global_p:
+          Opcode = NVPTXISD::LDGV2;
+          break;
+        case Intrinsic::nvvm_ldu_global_i:
+        case Intrinsic::nvvm_ldu_global_f:
+        case Intrinsic::nvvm_ldu_global_p:
+          Opcode = NVPTXISD::LDUV2;
+          break;
+        }
+        LdResVTs = DAG.getVTList(EltVT, EltVT, MVT::Other);
+        break;
+      case 4: {
+        switch (IntrinNo) {
+        default:
+          return;
+        case Intrinsic::nvvm_ldg_global_i:
+        case Intrinsic::nvvm_ldg_global_f:
+        case Intrinsic::nvvm_ldg_global_p:
+          Opcode = NVPTXISD::LDGV4;
+          break;
+        case Intrinsic::nvvm_ldu_global_i:
+        case Intrinsic::nvvm_ldu_global_f:
+        case Intrinsic::nvvm_ldu_global_p:
+          Opcode = NVPTXISD::LDUV4;
+          break;
+        }
+        EVT ListVTs[] = { EltVT, EltVT, EltVT, EltVT, MVT::Other };
+        LdResVTs = DAG.getVTList(ListVTs, 5);
+        break;
+      }
+      }
+
+      SmallVector<SDValue, 8> OtherOps;
+
+      // Copy regular operands
+
+      OtherOps.push_back(Chain); // Chain
+                                 // Skip operand 1 (intrinsic ID)
+                                 // Others
+      for (unsigned i = 2, e = N->getNumOperands(); i != e; ++i)
+        OtherOps.push_back(N->getOperand(i));
+
+      MemIntrinsicSDNode *MemSD = cast<MemIntrinsicSDNode>(N);
+
+      SDValue NewLD = DAG.getMemIntrinsicNode(
+          Opcode, DL, LdResVTs, &OtherOps[0], OtherOps.size(),
+          MemSD->getMemoryVT(), MemSD->getMemOperand());
+
+      SmallVector<SDValue, 4> ScalarRes;
+
+      for (unsigned i = 0; i < NumElts; ++i) {
+        SDValue Res = NewLD.getValue(i);
+        if (NeedTrunc)
+          Res =
+              DAG.getNode(ISD::TRUNCATE, DL, ResVT.getVectorElementType(), Res);
+        ScalarRes.push_back(Res);
+      }
+
+      SDValue LoadChain = NewLD.getValue(NumElts);
+
+      SDValue BuildVec =
+          DAG.getNode(ISD::BUILD_VECTOR, DL, ResVT, &ScalarRes[0], NumElts);
+
+      Results.push_back(BuildVec);
+      Results.push_back(LoadChain);
+    } else {
+      // i8 LDG/LDU
+      assert(ResVT.isSimple() && ResVT.getSimpleVT().SimpleTy == MVT::i8 &&
+             "Custom handling of non-i8 ldu/ldg?");
+
+      // Just copy all operands as-is
+      SmallVector<SDValue, 4> Ops;
+      for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
+        Ops.push_back(N->getOperand(i));
+
+      // Force output to i16
+      SDVTList LdResVTs = DAG.getVTList(MVT::i16, MVT::Other);
+
+      MemIntrinsicSDNode *MemSD = cast<MemIntrinsicSDNode>(N);
+
+      // We make sure the memory type is i8, which will be used during isel
+      // to select the proper instruction.
+      SDValue NewLD =
+          DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, LdResVTs, &Ops[0],
+                                  Ops.size(), MVT::i8, MemSD->getMemOperand());
+
+      Results.push_back(NewLD.getValue(0));
+      Results.push_back(NewLD.getValue(1));
+    }
+  }
+  }
+}
+
+void NVPTXTargetLowering::ReplaceNodeResults(
+    SDNode *N, SmallVectorImpl<SDValue> &Results, SelectionDAG &DAG) const {
+  switch (N->getOpcode()) {
+  default:
+    report_fatal_error("Unhandled custom legalization");
+  case ISD::LOAD:
+    ReplaceLoadVector(N, DAG, Results);
+    return;
+  case ISD::INTRINSIC_W_CHAIN:
+    ReplaceINTRINSIC_W_CHAIN(N, DAG, Results);
+    return;
+  }
+}
diff --git a/lib/Target/NVPTX/NVPTXISelLowering.h b/lib/Target/NVPTX/NVPTXISelLowering.h
index 94a177ceb00a..3cd49d38af76 100644
--- a/lib/Target/NVPTX/NVPTXISelLowering.h
+++ b/lib/Target/NVPTX/NVPTXISelLowering.h
@@ -58,7 +58,16 @@ enum NodeType {
   RETURN,
   CallSeqBegin,
   CallSeqEnd,
-  Dummy
+  Dummy,
+
+  LoadV2 = ISD::FIRST_TARGET_MEMORY_OPCODE,
+  LoadV4,
+  LDGV2, // LDG.v2
+  LDGV4, // LDG.v4
+  LDUV2, // LDU.v2
+  LDUV4, // LDU.v4
+  StoreV2,
+  StoreV4
 };
 }
 
@@ -78,7 +87,7 @@ public:
 
   bool isTypeSupportedInIntrinsic(MVT VT) const;
 
-  bool getTgtMemIntrinsic(IntrinsicInfo& Info, const CallInst &I,
+  bool getTgtMemIntrinsic(IntrinsicInfo &Info, const CallInst &I,
                           unsigned Intrinsic) const;
 
   /// isLegalAddressingMode - Return true if the addressing mode represented
@@ -92,18 +101,19 @@ public:
   virtual unsigned getFunctionAlignment(const Function *F) const;
 
   virtual EVT getSetCCResultType(EVT VT) const {
+    if (VT.isVector())
+      return MVT::getVectorVT(MVT::i1, VT.getVectorNumElements());
     return MVT::i1;
   }
 
   ConstraintType getConstraintType(const std::string &Constraint) const;
-  std::pair<unsigned, const TargetRegisterClass*>
+  std::pair<unsigned, const TargetRegisterClass *>
   getRegForInlineAsmConstraint(const std::string &Constraint, EVT VT) const;
 
-  virtual SDValue
-  LowerFormalArguments(SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
-                       const SmallVectorImpl<ISD::InputArg> &Ins, DebugLoc dl,
-                       SelectionDAG &DAG,
-                       SmallVectorImpl<SDValue> &InVals) const;
+  virtual SDValue LowerFormalArguments(
+      SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
+      const SmallVectorImpl<ISD::InputArg> &Ins, DebugLoc dl, SelectionDAG &DAG,
+      SmallVectorImpl<SDValue> &InVals) const;
 
   virtual SDValue
   LowerCall(CallLoweringInfo &CLI, SmallVectorImpl<SDValue> &InVals) const;
@@ -125,22 +135,29 @@ public:
   NVPTXTargetMachine *nvTM;
 
   // PTX always uses 32-bit shift amounts
-  virtual MVT getShiftAmountTy(EVT LHSTy) const {
-    return MVT::i32;
-  }
+  virtual MVT getScalarShiftAmountTy(EVT LHSTy) const { return MVT::i32; }
+
+  virtual bool shouldSplitVectorElementType(EVT VT) const;
 
 private:
-  const NVPTXSubtarget &nvptxSubtarget;  // cache the subtarget here
+  const NVPTXSubtarget &nvptxSubtarget; // cache the subtarget here
 
-  SDValue getExtSymb(SelectionDAG &DAG, const char *name, int idx, EVT =
-                         MVT::i32) const;
+  SDValue getExtSymb(SelectionDAG &DAG, const char *name, int idx,
+                     EVT = MVT::i32) const;
   SDValue getParamSymbol(SelectionDAG &DAG, int idx, EVT = MVT::i32) const;
   SDValue getParamHelpSymbol(SelectionDAG &DAG, int idx);
 
   SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const;
 
-  SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerLOADi1(SDValue Op, SelectionDAG &DAG) const;
+
+  SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSTOREi1(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSTOREVector(SDValue Op, SelectionDAG &DAG) const;
+
+  virtual void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue> &Results,
+                                  SelectionDAG &DAG) const;
 };
 } // namespace llvm
 
diff --git a/lib/Target/NVPTX/NVPTXInstrInfo.cpp b/lib/Target/NVPTX/NVPTXInstrInfo.cpp
index cd50deb26a23..33a63c26f4e2 100644
--- a/lib/Target/NVPTX/NVPTXInstrInfo.cpp
+++ b/lib/Target/NVPTX/NVPTXInstrInfo.cpp
@@ -16,108 +16,62 @@
 #include "NVPTXTargetMachine.h"
 #define GET_INSTRINFO_CTOR
 #include "NVPTXGenInstrInfo.inc"
-#include "llvm/Function.h"
+#include "llvm/IR/Function.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include <cstdio>
 
-
 using namespace llvm;
 
 // FIXME: Add the subtarget support on this constructor.
 NVPTXInstrInfo::NVPTXInstrInfo(NVPTXTargetMachine &tm)
-: NVPTXGenInstrInfo(),
-  TM(tm),
-  RegInfo(*this, *TM.getSubtargetImpl()) {}
-
+    : NVPTXGenInstrInfo(), TM(tm), RegInfo(*this, *TM.getSubtargetImpl()) {}
 
-void NVPTXInstrInfo::copyPhysReg (MachineBasicBlock &MBB,
-                                  MachineBasicBlock::iterator I, DebugLoc DL,
-                                  unsigned DestReg, unsigned SrcReg,
-                                  bool KillSrc) const {
+void NVPTXInstrInfo::copyPhysReg(
+    MachineBasicBlock &MBB, MachineBasicBlock::iterator I, DebugLoc DL,
+    unsigned DestReg, unsigned SrcReg, bool KillSrc) const {
   if (NVPTX::Int32RegsRegClass.contains(DestReg) &&
       NVPTX::Int32RegsRegClass.contains(SrcReg))
     BuildMI(MBB, I, DL, get(NVPTX::IMOV32rr), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
+        .addReg(SrcReg, getKillRegState(KillSrc));
   else if (NVPTX::Int8RegsRegClass.contains(DestReg) &&
-      NVPTX::Int8RegsRegClass.contains(SrcReg))
+           NVPTX::Int8RegsRegClass.contains(SrcReg))
     BuildMI(MBB, I, DL, get(NVPTX::IMOV8rr), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
+        .addReg(SrcReg, getKillRegState(KillSrc));
   else if (NVPTX::Int1RegsRegClass.contains(DestReg) &&
-      NVPTX::Int1RegsRegClass.contains(SrcReg))
+           NVPTX::Int1RegsRegClass.contains(SrcReg))
     BuildMI(MBB, I, DL, get(NVPTX::IMOV1rr), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
+        .addReg(SrcReg, getKillRegState(KillSrc));
   else if (NVPTX::Float32RegsRegClass.contains(DestReg) &&
-      NVPTX::Float32RegsRegClass.contains(SrcReg))
+           NVPTX::Float32RegsRegClass.contains(SrcReg))
     BuildMI(MBB, I, DL, get(NVPTX::FMOV32rr), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
+        .addReg(SrcReg, getKillRegState(KillSrc));
   else if (NVPTX::Int16RegsRegClass.contains(DestReg) &&
-      NVPTX::Int16RegsRegClass.contains(SrcReg))
+           NVPTX::Int16RegsRegClass.contains(SrcReg))
     BuildMI(MBB, I, DL, get(NVPTX::IMOV16rr), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
+        .addReg(SrcReg, getKillRegState(KillSrc));
   else if (NVPTX::Int64RegsRegClass.contains(DestReg) &&
-      NVPTX::Int64RegsRegClass.contains(SrcReg))
+           NVPTX::Int64RegsRegClass.contains(SrcReg))
     BuildMI(MBB, I, DL, get(NVPTX::IMOV64rr), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
+        .addReg(SrcReg, getKillRegState(KillSrc));
   else if (NVPTX::Float64RegsRegClass.contains(DestReg) &&
-      NVPTX::Float64RegsRegClass.contains(SrcReg))
+           NVPTX::Float64RegsRegClass.contains(SrcReg))
     BuildMI(MBB, I, DL, get(NVPTX::FMOV64rr), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (NVPTX::V4F32RegsRegClass.contains(DestReg) &&
-      NVPTX::V4F32RegsRegClass.contains(SrcReg))
-    BuildMI(MBB, I, DL, get(NVPTX::V4f32Mov), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (NVPTX::V4I32RegsRegClass.contains(DestReg) &&
-      NVPTX::V4I32RegsRegClass.contains(SrcReg))
-    BuildMI(MBB, I, DL, get(NVPTX::V4i32Mov), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (NVPTX::V2F32RegsRegClass.contains(DestReg) &&
-      NVPTX::V2F32RegsRegClass.contains(SrcReg))
-    BuildMI(MBB, I, DL, get(NVPTX::V2f32Mov), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (NVPTX::V2I32RegsRegClass.contains(DestReg) &&
-      NVPTX::V2I32RegsRegClass.contains(SrcReg))
-    BuildMI(MBB, I, DL, get(NVPTX::V2i32Mov), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (NVPTX::V4I8RegsRegClass.contains(DestReg) &&
-      NVPTX::V4I8RegsRegClass.contains(SrcReg))
-    BuildMI(MBB, I, DL, get(NVPTX::V4i8Mov), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (NVPTX::V2I8RegsRegClass.contains(DestReg) &&
-      NVPTX::V2I8RegsRegClass.contains(SrcReg))
-    BuildMI(MBB, I, DL, get(NVPTX::V2i8Mov), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (NVPTX::V4I16RegsRegClass.contains(DestReg) &&
-      NVPTX::V4I16RegsRegClass.contains(SrcReg))
-    BuildMI(MBB, I, DL, get(NVPTX::V4i16Mov), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (NVPTX::V2I16RegsRegClass.contains(DestReg) &&
-      NVPTX::V2I16RegsRegClass.contains(SrcReg))
-    BuildMI(MBB, I, DL, get(NVPTX::V2i16Mov), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (NVPTX::V2I64RegsRegClass.contains(DestReg) &&
-      NVPTX::V2I64RegsRegClass.contains(SrcReg))
-    BuildMI(MBB, I, DL, get(NVPTX::V2i64Mov), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
-  else if (NVPTX::V2F64RegsRegClass.contains(DestReg) &&
-      NVPTX::V2F64RegsRegClass.contains(SrcReg))
-    BuildMI(MBB, I, DL, get(NVPTX::V2f64Mov), DestReg)
-    .addReg(SrcReg, getKillRegState(KillSrc));
+        .addReg(SrcReg, getKillRegState(KillSrc));
   else {
     llvm_unreachable("Don't know how to copy a register");
   }
 }
 
-bool NVPTXInstrInfo::isMoveInstr(const MachineInstr &MI,
-                                 unsigned &SrcReg,
+bool NVPTXInstrInfo::isMoveInstr(const MachineInstr &MI, unsigned &SrcReg,
                                  unsigned &DestReg) const {
   // Look for the appropriate part of TSFlags
   bool isMove = false;
 
-  unsigned TSFlags = (MI.getDesc().TSFlags & NVPTX::SimpleMoveMask) >>
-      NVPTX::SimpleMoveShift;
+  unsigned TSFlags =
+      (MI.getDesc().TSFlags & NVPTX::SimpleMoveMask) >> NVPTX::SimpleMoveShift;
   isMove = (TSFlags == 1);
 
   if (isMove) {
@@ -134,10 +88,10 @@ bool NVPTXInstrInfo::isMoveInstr(const MachineInstr &MI,
   return false;
 }
 
-bool  NVPTXInstrInfo::isReadSpecialReg(MachineInstr &MI) const
-{
+bool NVPTXInstrInfo::isReadSpecialReg(MachineInstr &MI) const {
   switch (MI.getOpcode()) {
-  default: return false;
+  default:
+    return false;
   case NVPTX::INT_PTX_SREG_NTID_X:
   case NVPTX::INT_PTX_SREG_NTID_Y:
   case NVPTX::INT_PTX_SREG_NTID_Z:
@@ -155,12 +109,11 @@ bool  NVPTXInstrInfo::isReadSpecialReg(MachineInstr &MI) const
   }
 }
 
-
 bool NVPTXInstrInfo::isLoadInstr(const MachineInstr &MI,
                                  unsigned &AddrSpace) const {
   bool isLoad = false;
-  unsigned TSFlags = (MI.getDesc().TSFlags & NVPTX::isLoadMask) >>
-      NVPTX::isLoadShift;
+  unsigned TSFlags =
+      (MI.getDesc().TSFlags & NVPTX::isLoadMask) >> NVPTX::isLoadShift;
   isLoad = (TSFlags == 1);
   if (isLoad)
     AddrSpace = getLdStCodeAddrSpace(MI);
@@ -170,15 +123,14 @@ bool NVPTXInstrInfo::isLoadInstr(const MachineInstr &MI,
 bool NVPTXInstrInfo::isStoreInstr(const MachineInstr &MI,
                                   unsigned &AddrSpace) const {
   bool isStore = false;
-  unsigned TSFlags = (MI.getDesc().TSFlags & NVPTX::isStoreMask) >>
-      NVPTX::isStoreShift;
+  unsigned TSFlags =
+      (MI.getDesc().TSFlags & NVPTX::isStoreMask) >> NVPTX::isStoreShift;
   isStore = (TSFlags == 1);
   if (isStore)
     AddrSpace = getLdStCodeAddrSpace(MI);
   return isStore;
 }
 
-
 bool NVPTXInstrInfo::CanTailMerge(const MachineInstr *MI) const {
   unsigned addrspace = 0;
   if (MI->getOpcode() == NVPTX::INT_CUDA_SYNCTHREADS)
@@ -192,7 +144,6 @@ bool NVPTXInstrInfo::CanTailMerge(const MachineInstr *MI) const {
   return true;
 }
 
-
 /// AnalyzeBranch - Analyze the branching code at the end of MBB, returning
 /// true if it cannot be understood (e.g. it's a switch dispatch or isn't
 /// implemented for a target).  Upon success, this returns false and returns
@@ -216,11 +167,9 @@ bool NVPTXInstrInfo::CanTailMerge(const MachineInstr *MI) const {
 /// Note that RemoveBranch and InsertBranch must be implemented to support
 /// cases where this method returns success.
 ///
-bool NVPTXInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
-                                   MachineBasicBlock *&TBB,
-                                   MachineBasicBlock *&FBB,
-                                   SmallVectorImpl<MachineOperand> &Cond,
-                                   bool AllowModify) const {
+bool NVPTXInstrInfo::AnalyzeBranch(
+    MachineBasicBlock &MBB, MachineBasicBlock *&TBB, MachineBasicBlock *&FBB,
+    SmallVectorImpl<MachineOperand> &Cond, bool AllowModify) const {
   // If the block has no terminators, it just falls into the block after it.
   MachineBasicBlock::iterator I = MBB.end();
   if (I == MBB.begin() || !isUnpredicatedTerminator(--I))
@@ -248,14 +197,13 @@ bool NVPTXInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
   MachineInstr *SecondLastInst = I;
 
   // If there are three terminators, we don't know what sort of block this is.
-  if (SecondLastInst && I != MBB.begin() &&
-      isUnpredicatedTerminator(--I))
+  if (SecondLastInst && I != MBB.begin() && isUnpredicatedTerminator(--I))
     return true;
 
   // If the block ends with NVPTX::GOTO and NVPTX:CBranch, handle it.
   if (SecondLastInst->getOpcode() == NVPTX::CBranch &&
       LastInst->getOpcode() == NVPTX::GOTO) {
-    TBB =  SecondLastInst->getOperand(1).getMBB();
+    TBB = SecondLastInst->getOperand(1).getMBB();
     Cond.push_back(SecondLastInst->getOperand(0));
     FBB = LastInst->getOperand(0).getMBB();
     return false;
@@ -278,7 +226,8 @@ bool NVPTXInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
 
 unsigned NVPTXInstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
   MachineBasicBlock::iterator I = MBB.end();
-  if (I == MBB.begin()) return 0;
+  if (I == MBB.begin())
+    return 0;
   --I;
   if (I->getOpcode() != NVPTX::GOTO && I->getOpcode() != NVPTX::CBranch)
     return 0;
@@ -288,7 +237,8 @@ unsigned NVPTXInstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
 
   I = MBB.end();
 
-  if (I == MBB.begin()) return 1;
+  if (I == MBB.begin())
+    return 1;
   --I;
   if (I->getOpcode() != NVPTX::CBranch)
     return 1;
@@ -298,11 +248,9 @@ unsigned NVPTXInstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
   return 2;
 }
 
-unsigned
-NVPTXInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
-                             MachineBasicBlock *FBB,
-                             const SmallVectorImpl<MachineOperand> &Cond,
-                             DebugLoc DL) const {
+unsigned NVPTXInstrInfo::InsertBranch(
+    MachineBasicBlock &MBB, MachineBasicBlock *TBB, MachineBasicBlock *FBB,
+    const SmallVectorImpl<MachineOperand> &Cond, DebugLoc DL) const {
   // Shouldn't be a fall through.
   assert(TBB && "InsertBranch must not be told to insert a fallthrough");
   assert((Cond.size() == 1 || Cond.size() == 0) &&
@@ -310,17 +258,16 @@ NVPTXInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
 
   // One-way branch.
   if (FBB == 0) {
-    if (Cond.empty())   // Unconditional branch
+    if (Cond.empty()) // Unconditional branch
       BuildMI(&MBB, DL, get(NVPTX::GOTO)).addMBB(TBB);
-    else                // Conditional branch
-      BuildMI(&MBB, DL, get(NVPTX::CBranch))
-      .addReg(Cond[0].getReg()).addMBB(TBB);
+    else // Conditional branch
+      BuildMI(&MBB, DL, get(NVPTX::CBranch)).addReg(Cond[0].getReg())
+          .addMBB(TBB);
     return 1;
   }
 
   // Two-way Conditional Branch.
-  BuildMI(&MBB, DL, get(NVPTX::CBranch))
-  .addReg(Cond[0].getReg()).addMBB(TBB);
+  BuildMI(&MBB, DL, get(NVPTX::CBranch)).addReg(Cond[0].getReg()).addMBB(TBB);
   BuildMI(&MBB, DL, get(NVPTX::GOTO)).addMBB(FBB);
   return 2;
 }
diff --git a/lib/Target/NVPTX/NVPTXInstrInfo.h b/lib/Target/NVPTX/NVPTXInstrInfo.h
index 7b8e218b05b6..b1972e9b7254 100644
--- a/lib/Target/NVPTX/NVPTXInstrInfo.h
+++ b/lib/Target/NVPTX/NVPTXInstrInfo.h
@@ -23,8 +23,7 @@
 
 namespace llvm {
 
-class NVPTXInstrInfo : public NVPTXGenInstrInfo
-{
+class NVPTXInstrInfo : public NVPTXGenInstrInfo {
   NVPTXTargetMachine &TM;
   const NVPTXRegisterInfo RegInfo;
 public:
@@ -50,30 +49,26 @@ public:
    *                               const TargetRegisterClass *RC) const;
    */
 
-  virtual void copyPhysReg(MachineBasicBlock &MBB,
-                           MachineBasicBlock::iterator I, DebugLoc DL,
-                           unsigned DestReg, unsigned SrcReg,
-                           bool KillSrc) const ;
-  virtual bool isMoveInstr(const MachineInstr &MI,
-                           unsigned &SrcReg,
+  virtual void copyPhysReg(
+      MachineBasicBlock &MBB, MachineBasicBlock::iterator I, DebugLoc DL,
+      unsigned DestReg, unsigned SrcReg, bool KillSrc) const;
+  virtual bool isMoveInstr(const MachineInstr &MI, unsigned &SrcReg,
                            unsigned &DestReg) const;
   bool isLoadInstr(const MachineInstr &MI, unsigned &AddrSpace) const;
   bool isStoreInstr(const MachineInstr &MI, unsigned &AddrSpace) const;
   bool isReadSpecialReg(MachineInstr &MI) const;
 
-  virtual bool CanTailMerge(const MachineInstr *MI) const ;
+  virtual bool CanTailMerge(const MachineInstr *MI) const;
   // Branch analysis.
-  virtual bool AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
-                             MachineBasicBlock *&FBB,
-                             SmallVectorImpl<MachineOperand> &Cond,
-                             bool AllowModify) const;
+  virtual bool AnalyzeBranch(
+      MachineBasicBlock &MBB, MachineBasicBlock *&TBB, MachineBasicBlock *&FBB,
+      SmallVectorImpl<MachineOperand> &Cond, bool AllowModify) const;
   virtual unsigned RemoveBranch(MachineBasicBlock &MBB) const;
-  virtual unsigned InsertBranch(MachineBasicBlock &MBB,MachineBasicBlock *TBB,
-                                MachineBasicBlock *FBB,
-                                const SmallVectorImpl<MachineOperand> &Cond,
-                                DebugLoc DL) const;
+  virtual unsigned InsertBranch(
+      MachineBasicBlock &MBB, MachineBasicBlock *TBB, MachineBasicBlock *FBB,
+      const SmallVectorImpl<MachineOperand> &Cond, DebugLoc DL) const;
   unsigned getLdStCodeAddrSpace(const MachineInstr &MI) const {
-    return  MI.getOperand(2).getImm();
+    return MI.getOperand(2).getImm();
   }
 
 };
diff --git a/lib/Target/NVPTX/NVPTXInstrInfo.td b/lib/Target/NVPTX/NVPTXInstrInfo.td
index 8a410b872925..f43abe283b58 100644
--- a/lib/Target/NVPTX/NVPTXInstrInfo.td
+++ b/lib/Target/NVPTX/NVPTXInstrInfo.td
@@ -52,6 +52,7 @@ def hasAtomAddF32 : Predicate<"Subtarget.hasAtomAddF32()">;
 def hasVote : Predicate<"Subtarget.hasVote()">;
 def hasDouble : Predicate<"Subtarget.hasDouble()">;
 def reqPTX20 : Predicate<"Subtarget.reqPTX20()">;
+def hasLDG : Predicate<"Subtarget.hasLDG()">;
 def hasLDU : Predicate<"Subtarget.hasLDU()">;
 def hasGenericLdSt : Predicate<"Subtarget.hasGenericLdSt()">;
 
@@ -2153,11 +2154,21 @@ multiclass LD<NVPTXRegClass regclass> {
       i32imm:$fromWidth, Int32Regs:$addr),
 !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
            "$fromWidth \t$dst, [$addr];"), []>;
+  def _areg_64 : NVPTXInst<(outs regclass:$dst),
+    (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
+     i32imm:$fromWidth, Int64Regs:$addr),
+     !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}$fromWidth",
+                " \t$dst, [$addr];"), []>;
   def _ari : NVPTXInst<(outs regclass:$dst),
     (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
       i32imm:$fromWidth, Int32Regs:$addr, i32imm:$offset),
 !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
            "$fromWidth \t$dst, [$addr+$offset];"), []>;
+  def _ari_64 : NVPTXInst<(outs regclass:$dst),
+    (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
+     i32imm:$fromWidth, Int64Regs:$addr, i32imm:$offset),
+    !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}$fromWidth",
+               " \t$dst, [$addr+$offset];"), []>;
   def _asi : NVPTXInst<(outs regclass:$dst),
     (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
       i32imm:$fromWidth, imem:$addr, i32imm:$offset),
@@ -2174,19 +2185,6 @@ defm LD_f32 : LD<Float32Regs>;
 defm LD_f64 : LD<Float64Regs>;
 }
 
-let VecInstType=isVecLD.Value, mayLoad=1, neverHasSideEffects=1 in {
-defm LD_v2i8 : LD<V2I8Regs>;
-defm LD_v4i8 : LD<V4I8Regs>;
-defm LD_v2i16 : LD<V2I16Regs>;
-defm LD_v4i16 : LD<V4I16Regs>;
-defm LD_v2i32 : LD<V2I32Regs>;
-defm LD_v4i32 : LD<V4I32Regs>;
-defm LD_v2f32 : LD<V2F32Regs>;
-defm LD_v4f32 : LD<V4F32Regs>;
-defm LD_v2i64 : LD<V2I64Regs>;
-defm LD_v2f64 : LD<V2F64Regs>;
-}
-
 multiclass ST<NVPTXRegClass regclass> {
   def _avar : NVPTXInst<(outs),
     (ins regclass:$src, LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec,
@@ -2198,11 +2196,21 @@ multiclass ST<NVPTXRegClass regclass> {
       LdStCode:$Sign, i32imm:$toWidth, Int32Regs:$addr),
 !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}$toWidth",
            " \t[$addr], $src;"), []>;
+  def _areg_64 : NVPTXInst<(outs),
+    (ins regclass:$src, LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec,
+     LdStCode:$Sign, i32imm:$toWidth, Int64Regs:$addr),
+  !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}$toWidth ",
+               "\t[$addr], $src;"), []>;
   def _ari : NVPTXInst<(outs),
     (ins regclass:$src, LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec,
       LdStCode:$Sign, i32imm:$toWidth, Int32Regs:$addr, i32imm:$offset),
 !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}$toWidth",
            " \t[$addr+$offset], $src;"), []>;
+  def _ari_64 : NVPTXInst<(outs),
+    (ins regclass:$src, LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec,
+     LdStCode:$Sign, i32imm:$toWidth, Int64Regs:$addr, i32imm:$offset),
+  !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}$toWidth ",
+               "\t[$addr+$offset], $src;"), []>;
   def _asi : NVPTXInst<(outs),
     (ins regclass:$src, LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec,
       LdStCode:$Sign, i32imm:$toWidth, imem:$addr, i32imm:$offset),
@@ -2219,19 +2227,6 @@ defm ST_f32 : ST<Float32Regs>;
 defm ST_f64 : ST<Float64Regs>;
 }
 
-let VecInstType=isVecST.Value, mayStore=1, neverHasSideEffects=1 in {
-defm ST_v2i8 : ST<V2I8Regs>;
-defm ST_v4i8 : ST<V4I8Regs>;
-defm ST_v2i16 : ST<V2I16Regs>;
-defm ST_v4i16 : ST<V4I16Regs>;
-defm ST_v2i32 : ST<V2I32Regs>;
-defm ST_v4i32 : ST<V4I32Regs>;
-defm ST_v2f32 : ST<V2F32Regs>;
-defm ST_v4f32 : ST<V4F32Regs>;
-defm ST_v2i64 : ST<V2I64Regs>;
-defm ST_v2f64 : ST<V2F64Regs>;
-}
-
 // The following is used only in and after vector elementizations.
 // Vector elementization happens at the machine instruction level, so the
 // following instruction
@@ -2247,11 +2242,21 @@ multiclass LD_VEC<NVPTXRegClass regclass> {
       i32imm:$fromWidth, Int32Regs:$addr),
     !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
                "$fromWidth \t{{$dst1, $dst2}}, [$addr];"), []>;
+  def _v2_areg_64 : NVPTXInst<(outs regclass:$dst1, regclass:$dst2),
+    (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
+     i32imm:$fromWidth, Int64Regs:$addr),
+    !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
+               "$fromWidth \t{{$dst1, $dst2}}, [$addr];"), []>;
   def _v2_ari : NVPTXInst<(outs regclass:$dst1, regclass:$dst2),
     (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
       i32imm:$fromWidth, Int32Regs:$addr, i32imm:$offset),
     !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
                "$fromWidth \t{{$dst1, $dst2}}, [$addr+$offset];"), []>;
+  def _v2_ari_64 : NVPTXInst<(outs regclass:$dst1, regclass:$dst2),
+    (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
+     i32imm:$fromWidth, Int64Regs:$addr, i32imm:$offset),
+    !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
+               "$fromWidth \t{{$dst1, $dst2}}, [$addr+$offset];"), []>;
   def _v2_asi : NVPTXInst<(outs regclass:$dst1, regclass:$dst2),
     (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
       i32imm:$fromWidth, imem:$addr, i32imm:$offset),
@@ -2269,6 +2274,12 @@ multiclass LD_VEC<NVPTXRegClass regclass> {
       i32imm:$fromWidth, Int32Regs:$addr),
     !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
                "$fromWidth \t{{$dst1, $dst2, $dst3, $dst4}}, [$addr];"), []>;
+  def _v4_areg_64 : NVPTXInst<(outs regclass:$dst1, regclass:$dst2,
+                               regclass:$dst3, regclass:$dst4),
+    (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
+     i32imm:$fromWidth, Int64Regs:$addr),
+    !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
+               "$fromWidth \t{{$dst1, $dst2, $dst3, $dst4}}, [$addr];"), []>;
   def _v4_ari : NVPTXInst<(outs regclass:$dst1, regclass:$dst2, regclass:$dst3,
       regclass:$dst4),
     (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
@@ -2276,6 +2287,13 @@ multiclass LD_VEC<NVPTXRegClass regclass> {
     !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
                "$fromWidth \t{{$dst1, $dst2, $dst3, $dst4}}, [$addr+$offset];"),
                 []>;
+  def _v4_ari_64 : NVPTXInst<(outs regclass:$dst1, regclass:$dst2,
+                              regclass:$dst3, regclass:$dst4),
+    (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
+     i32imm:$fromWidth, Int64Regs:$addr, i32imm:$offset),
+    !strconcat("ld${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
+               "$fromWidth \t{{$dst1, $dst2, $dst3, $dst4}}, [$addr+$offset];"),
+    []>;
   def _v4_asi : NVPTXInst<(outs regclass:$dst1, regclass:$dst2, regclass:$dst3,
       regclass:$dst4),
     (ins LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
@@ -2304,12 +2322,23 @@ multiclass ST_VEC<NVPTXRegClass regclass> {
       LdStCode:$Vec, LdStCode:$Sign, i32imm:$fromWidth, Int32Regs:$addr),
     !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
                "$fromWidth \t[$addr], {{$src1, $src2}};"), []>;
+  def _v2_areg_64 : NVPTXInst<(outs),
+    (ins regclass:$src1, regclass:$src2, LdStCode:$isVol, LdStCode:$addsp,
+     LdStCode:$Vec, LdStCode:$Sign, i32imm:$fromWidth, Int64Regs:$addr),
+    !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
+               "$fromWidth \t[$addr], {{$src1, $src2}};"), []>;
   def _v2_ari : NVPTXInst<(outs),
     (ins regclass:$src1, regclass:$src2, LdStCode:$isVol, LdStCode:$addsp,
       LdStCode:$Vec, LdStCode:$Sign, i32imm:$fromWidth, Int32Regs:$addr,
       i32imm:$offset),
     !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
                "$fromWidth \t[$addr+$offset], {{$src1, $src2}};"), []>;
+  def _v2_ari_64 : NVPTXInst<(outs),
+    (ins regclass:$src1, regclass:$src2, LdStCode:$isVol, LdStCode:$addsp,
+     LdStCode:$Vec, LdStCode:$Sign, i32imm:$fromWidth, Int64Regs:$addr,
+     i32imm:$offset),
+    !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
+               "$fromWidth \t[$addr+$offset], {{$src1, $src2}};"), []>;
   def _v2_asi : NVPTXInst<(outs),
     (ins regclass:$src1, regclass:$src2, LdStCode:$isVol, LdStCode:$addsp,
       LdStCode:$Vec, LdStCode:$Sign, i32imm:$fromWidth, imem:$addr,
@@ -2328,6 +2357,12 @@ multiclass ST_VEC<NVPTXRegClass regclass> {
       i32imm:$fromWidth, Int32Regs:$addr),
     !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
                "$fromWidth \t[$addr], {{$src1, $src2, $src3, $src4}};"), []>;
+  def _v4_areg_64 : NVPTXInst<(outs),
+    (ins regclass:$src1, regclass:$src2, regclass:$src3, regclass:$src4,
+     LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
+     i32imm:$fromWidth, Int64Regs:$addr),
+    !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
+               "$fromWidth \t[$addr], {{$src1, $src2, $src3, $src4}};"), []>;
   def _v4_ari : NVPTXInst<(outs),
     (ins regclass:$src1, regclass:$src2, regclass:$src3, regclass:$src4,
       LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
@@ -2335,6 +2370,13 @@ multiclass ST_VEC<NVPTXRegClass regclass> {
     !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
                "$fromWidth \t[$addr+$offset], {{$src1, $src2, $src3, $src4}};"),
     []>;
+  def _v4_ari_64 : NVPTXInst<(outs),
+    (ins regclass:$src1, regclass:$src2, regclass:$src3, regclass:$src4,
+     LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
+     i32imm:$fromWidth, Int64Regs:$addr, i32imm:$offset),
+    !strconcat("st${isVol:volatile}${addsp:addsp}${Vec:vec}.${Sign:sign}",
+               "$fromWidth \t[$addr+$offset], {{$src1, $src2, $src3, $src4}};"),
+     []>;
   def _v4_asi : NVPTXInst<(outs),
     (ins regclass:$src1, regclass:$src2, regclass:$src3, regclass:$src4,
       LdStCode:$isVol, LdStCode:$addsp, LdStCode:$Vec, LdStCode:$Sign,
@@ -2822,8 +2864,6 @@ def trapinst : NVPTXInst<(outs), (ins),
                          "trap;",
                          [(trap)]>;
 
-include "NVPTXVector.td"
-
 include "NVPTXIntrinsics.td"
 
 
diff --git a/lib/Target/NVPTX/NVPTXIntrinsics.td b/lib/Target/NVPTX/NVPTXIntrinsics.td
index 028a94bfd1bb..49e2568dfa2c 100644
--- a/lib/Target/NVPTX/NVPTXIntrinsics.td
+++ b/lib/Target/NVPTX/NVPTXIntrinsics.td
@@ -1343,52 +1343,113 @@ defm INT_PTX_LDU_G_v4f32_ELE
   : VLDU_G_ELE_V4<"v4.f32 \t{{$dst1, $dst2, $dst3, $dst4}}, [$src];",
     Float32Regs>;
 
-// Vector ldu
-multiclass VLDU_G<string TyStr, NVPTXRegClass regclass, Intrinsic IntOp,
-  NVPTXInst eleInst, NVPTXInst eleInst64> {
- def _32:    NVPTXVecInst<(outs regclass:$result), (ins Int32Regs:$src),
-               !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp Int32Regs:$src))], eleInst>,
- Requires<[hasLDU]>;
- def _64:    NVPTXVecInst<(outs regclass:$result), (ins Int64Regs:$src),
-               !strconcat("ldu.global.", TyStr),
-         [(set regclass:$result, (IntOp Int64Regs:$src))], eleInst64>,
- Requires<[hasLDU]>;
+
+//-----------------------------------
+// Support for ldg on sm_35 or later 
+//-----------------------------------
+
+def ldg_i8 : PatFrag<(ops node:$ptr), (int_nvvm_ldg_global_i node:$ptr), [{
+  MemIntrinsicSDNode *M = cast<MemIntrinsicSDNode>(N);
+  return M->getMemoryVT() == MVT::i8;
+}]>;
+
+multiclass LDG_G<string TyStr, NVPTXRegClass regclass, Intrinsic IntOp> {
+  def areg: NVPTXInst<(outs regclass:$result), (ins Int32Regs:$src),
+               !strconcat("ld.global.nc.", TyStr),
+         [(set regclass:$result, (IntOp Int32Regs:$src))]>, Requires<[hasLDG]>;
+  def areg64: NVPTXInst<(outs regclass:$result), (ins Int64Regs:$src),
+               !strconcat("ld.global.nc.", TyStr),
+         [(set regclass:$result, (IntOp Int64Regs:$src))]>, Requires<[hasLDG]>;
+ def avar:  NVPTXInst<(outs regclass:$result), (ins imem:$src),
+               !strconcat("ld.global.nc.", TyStr),
+         [(set regclass:$result, (IntOp (Wrapper tglobaladdr:$src)))]>,
+         Requires<[hasLDG]>;
+ def ari :  NVPTXInst<(outs regclass:$result), (ins MEMri:$src),
+               !strconcat("ld.global.nc.", TyStr),
+         [(set regclass:$result, (IntOp ADDRri:$src))]>, Requires<[hasLDG]>;
+ def ari64 :  NVPTXInst<(outs regclass:$result), (ins MEMri64:$src),
+               !strconcat("ld.global.nc.", TyStr),
+         [(set regclass:$result, (IntOp ADDRri64:$src))]>, Requires<[hasLDG]>;
+}
+
+multiclass LDG_G_NOINTRIN<string TyStr, NVPTXRegClass regclass, PatFrag IntOp> {
+  def areg: NVPTXInst<(outs regclass:$result), (ins Int32Regs:$src),
+               !strconcat("ld.global.nc.", TyStr),
+         [(set regclass:$result, (IntOp Int32Regs:$src))]>, Requires<[hasLDG]>;
+  def areg64: NVPTXInst<(outs regclass:$result), (ins Int64Regs:$src),
+               !strconcat("ld.global.nc.", TyStr),
+         [(set regclass:$result, (IntOp Int64Regs:$src))]>, Requires<[hasLDG]>;
+ def avar:  NVPTXInst<(outs regclass:$result), (ins imem:$src),
+               !strconcat("ld.global.nc.", TyStr),
+         [(set regclass:$result, (IntOp (Wrapper tglobaladdr:$src)))]>,
+        Requires<[hasLDG]>;
+ def ari :  NVPTXInst<(outs regclass:$result), (ins MEMri:$src),
+               !strconcat("ld.global.nc.", TyStr),
+         [(set regclass:$result, (IntOp ADDRri:$src))]>, Requires<[hasLDG]>;
+ def ari64 :  NVPTXInst<(outs regclass:$result), (ins MEMri64:$src),
+               !strconcat("ld.global.nc.", TyStr),
+         [(set regclass:$result, (IntOp ADDRri64:$src))]>, Requires<[hasLDG]>;
+}
+
+defm INT_PTX_LDG_GLOBAL_i8
+  : LDG_G_NOINTRIN<"u8 \t$result, [$src];",  Int16Regs, ldg_i8>;
+defm INT_PTX_LDG_GLOBAL_i16
+  : LDG_G<"u16 \t$result, [$src];", Int16Regs,   int_nvvm_ldg_global_i>;
+defm INT_PTX_LDG_GLOBAL_i32
+  : LDG_G<"u32 \t$result, [$src];", Int32Regs,   int_nvvm_ldg_global_i>;
+defm INT_PTX_LDG_GLOBAL_i64
+  : LDG_G<"u64 \t$result, [$src];", Int64Regs,   int_nvvm_ldg_global_i>;
+defm INT_PTX_LDG_GLOBAL_f32
+  : LDG_G<"f32 \t$result, [$src];", Float32Regs, int_nvvm_ldg_global_f>;
+defm INT_PTX_LDG_GLOBAL_f64
+  : LDG_G<"f64 \t$result, [$src];", Float64Regs, int_nvvm_ldg_global_f>;
+defm INT_PTX_LDG_GLOBAL_p32
+  : LDG_G<"u32 \t$result, [$src];", Int32Regs,   int_nvvm_ldg_global_p>;
+defm INT_PTX_LDG_GLOBAL_p64
+  : LDG_G<"u64 \t$result, [$src];", Int64Regs,   int_nvvm_ldg_global_p>;
+
+// vector
+
+// Elementized vector ldg 
+multiclass VLDG_G_ELE_V2<string TyStr, NVPTXRegClass regclass> {
+ def _32:     NVPTXInst<(outs regclass:$dst1, regclass:$dst2),
+                     (ins Int32Regs:$src),
+                     !strconcat("ld.global.nc.", TyStr), []>;
+ def _64:     NVPTXInst<(outs regclass:$dst1, regclass:$dst2),
+                     (ins Int64Regs:$src),
+                     !strconcat("ld.global.nc.", TyStr), []>;
 }
 
-let VecInstType=isVecLD.Value in {
-defm INT_PTX_LDU_G_v2i8  : VLDU_G<"v2.u8 \t${result:vecfull}, [$src];",
-  V2I8Regs,  int_nvvm_ldu_global_i, INT_PTX_LDU_G_v2i8_ELE_32,
-  INT_PTX_LDU_G_v2i8_ELE_64>;
-defm INT_PTX_LDU_G_v4i8  : VLDU_G<"v4.u8 \t${result:vecfull}, [$src];",
-  V4I8Regs,  int_nvvm_ldu_global_i, INT_PTX_LDU_G_v4i8_ELE_32,
-  INT_PTX_LDU_G_v4i8_ELE_64>;
-defm INT_PTX_LDU_G_v2i16 : VLDU_G<"v2.u16 \t${result:vecfull}, [$src];",
-  V2I16Regs, int_nvvm_ldu_global_i, INT_PTX_LDU_G_v2i16_ELE_32,
-  INT_PTX_LDU_G_v2i16_ELE_64>;
-defm INT_PTX_LDU_G_v4i16 : VLDU_G<"v4.u16 \t${result:vecfull}, [$src];",
-  V4I16Regs, int_nvvm_ldu_global_i, INT_PTX_LDU_G_v4i16_ELE_32,
-  INT_PTX_LDU_G_v4i16_ELE_64>;
-defm INT_PTX_LDU_G_v2i32 : VLDU_G<"v2.u32 \t${result:vecfull}, [$src];",
-  V2I32Regs, int_nvvm_ldu_global_i, INT_PTX_LDU_G_v2i32_ELE_32,
-  INT_PTX_LDU_G_v2i32_ELE_64>;
-defm INT_PTX_LDU_G_v4i32 : VLDU_G<"v4.u32 \t${result:vecfull}, [$src];",
-  V4I32Regs, int_nvvm_ldu_global_i, INT_PTX_LDU_G_v4i32_ELE_32,
-  INT_PTX_LDU_G_v4i32_ELE_64>;
-defm INT_PTX_LDU_G_v2f32 : VLDU_G<"v2.f32 \t${result:vecfull}, [$src];",
-  V2F32Regs, int_nvvm_ldu_global_f, INT_PTX_LDU_G_v2f32_ELE_32,
-  INT_PTX_LDU_G_v2f32_ELE_64>;
-defm INT_PTX_LDU_G_v4f32 : VLDU_G<"v4.f32 \t${result:vecfull}, [$src];",
-  V4F32Regs, int_nvvm_ldu_global_f, INT_PTX_LDU_G_v4f32_ELE_32,
-  INT_PTX_LDU_G_v4f32_ELE_64>;
-defm INT_PTX_LDU_G_v2i64 : VLDU_G<"v2.u64 \t${result:vecfull}, [$src];",
-  V2I64Regs, int_nvvm_ldu_global_i, INT_PTX_LDU_G_v2i64_ELE_32,
-  INT_PTX_LDU_G_v2i64_ELE_64>;
-defm INT_PTX_LDU_G_v2f64 : VLDU_G<"v2.f64 \t${result:vecfull}, [$src];",
-  V2F64Regs, int_nvvm_ldu_global_f, INT_PTX_LDU_G_v2f64_ELE_32,
-  INT_PTX_LDU_G_v2f64_ELE_64>;
+multiclass VLDG_G_ELE_V4<string TyStr, NVPTXRegClass regclass> { 
+ def _32:    NVPTXInst<(outs regclass:$dst1, regclass:$dst2,
+                        regclass:$dst3, regclass:$dst4), (ins Int32Regs:$src),
+               !strconcat("ld.global.nc.", TyStr), []>;
+ def _64:    NVPTXInst<(outs regclass:$dst1, regclass:$dst2,
+                        regclass:$dst3, regclass:$dst4), (ins Int64Regs:$src),
+               !strconcat("ld.global.nc.", TyStr), []>;
 }
 
+// FIXME: 8-bit LDG should be fixed once LDG/LDU nodes are made into proper loads.
+defm INT_PTX_LDG_G_v2i8_ELE
+  : VLDG_G_ELE_V2<"v2.u8 \t{{$dst1, $dst2}}, [$src];",  Int16Regs>;
+defm INT_PTX_LDG_G_v2i16_ELE
+  : VLDG_G_ELE_V2<"v2.u16 \t{{$dst1, $dst2}}, [$src];", Int16Regs>;
+defm INT_PTX_LDG_G_v2i32_ELE
+  : VLDG_G_ELE_V2<"v2.u32 \t{{$dst1, $dst2}}, [$src];", Int32Regs>;
+defm INT_PTX_LDG_G_v2f32_ELE
+  : VLDG_G_ELE_V2<"v2.f32 \t{{$dst1, $dst2}}, [$src];", Float32Regs>;
+defm INT_PTX_LDG_G_v2i64_ELE
+  : VLDG_G_ELE_V2<"v2.u64 \t{{$dst1, $dst2}}, [$src];", Int64Regs>;
+defm INT_PTX_LDG_G_v2f64_ELE
+  : VLDG_G_ELE_V2<"v2.f64 \t{{$dst1, $dst2}}, [$src];", Float64Regs>;
+defm INT_PTX_LDG_G_v4i8_ELE
+  : VLDG_G_ELE_V4<"v4.u8 \t{{$dst1, $dst2, $dst3, $dst4}}, [$src];", Int16Regs>;
+defm INT_PTX_LDG_G_v4i16_ELE
+  : VLDG_G_ELE_V4<"v4.u16 \t{{$dst1, $dst2, $dst3, $dst4}}, [$src];", Int16Regs>;
+defm INT_PTX_LDG_G_v4i32_ELE
+  : VLDG_G_ELE_V4<"v4.u32 \t{{$dst1, $dst2, $dst3, $dst4}}, [$src];", Int32Regs>;
+defm INT_PTX_LDG_G_v4f32_ELE
+  : VLDG_G_ELE_V4<"v4.f32 \t{{$dst1, $dst2, $dst3, $dst4}}, [$src];", Float32Regs>;
 
 
 multiclass NG_TO_G<string Str, Intrinsic Intrin> {
diff --git a/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp b/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
index 9273931e9919..7c257b4c6a89 100644
--- a/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
+++ b/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
@@ -12,31 +12,28 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTXLowerAggrCopies.h"
-#include "llvm/Constants.h"
-#include "llvm/Function.h"
-#include "llvm/IRBuilder.h"
-#include "llvm/Instructions.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/LLVMContext.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/InstIterator.h"
-#include "llvm/DataLayout.h"
 
 using namespace llvm;
 
-namespace llvm {
-FunctionPass *createLowerAggrCopies();
-}
+namespace llvm { FunctionPass *createLowerAggrCopies(); }
 
 char NVPTXLowerAggrCopies::ID = 0;
 
 // Lower MemTransferInst or load-store pair to loop
-static void convertTransferToLoop(Instruction *splitAt, Value *srcAddr,
-                                  Value *dstAddr, Value *len,
-                                  //unsigned numLoads,
-                                  bool srcVolatile, bool dstVolatile,
-                                  LLVMContext &Context, Function &F) {
+static void convertTransferToLoop(
+    Instruction *splitAt, Value *srcAddr, Value *dstAddr, Value *len,
+    //unsigned numLoads,
+    bool srcVolatile, bool dstVolatile, LLVMContext &Context, Function &F) {
   Type *indType = len->getType();
 
   BasicBlock *origBB = splitAt->getParent();
@@ -48,10 +45,8 @@ static void convertTransferToLoop(Instruction *splitAt, Value *srcAddr,
 
   // srcAddr and dstAddr are expected to be pointer types,
   // so no check is made here.
-  unsigned srcAS =
-      dyn_cast<PointerType>(srcAddr->getType())->getAddressSpace();
-  unsigned dstAS =
-      dyn_cast<PointerType>(dstAddr->getType())->getAddressSpace();
+  unsigned srcAS = dyn_cast<PointerType>(srcAddr->getType())->getAddressSpace();
+  unsigned dstAS = dyn_cast<PointerType>(dstAddr->getType())->getAddressSpace();
 
   // Cast pointers to (char *)
   srcAddr = builder.CreateBitCast(srcAddr, Type::getInt8PtrTy(Context, srcAS));
@@ -86,12 +81,11 @@ static void convertMemSetToLoop(Instruction *splitAt, Value *dstAddr,
   origBB->getTerminator()->setSuccessor(0, loopBB);
   IRBuilder<> builder(origBB, origBB->getTerminator());
 
-  unsigned dstAS =
-      dyn_cast<PointerType>(dstAddr->getType())->getAddressSpace();
+  unsigned dstAS = dyn_cast<PointerType>(dstAddr->getType())->getAddressSpace();
 
   // Cast pointer to the type of value getting stored
-  dstAddr = builder.CreateBitCast(dstAddr,
-                                  PointerType::get(val->getType(), dstAS));
+  dstAddr =
+      builder.CreateBitCast(dstAddr, PointerType::get(val->getType(), dstAS));
 
   IRBuilder<> loop(loopBB);
   PHINode *ind = loop.CreatePHI(len->getType(), 0);
@@ -120,24 +114,26 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
   for (Function::iterator BI = F.begin(), BE = F.end(); BI != BE; ++BI) {
     //BasicBlock *bb = BI;
     for (BasicBlock::iterator II = BI->begin(), IE = BI->end(); II != IE;
-        ++II) {
-      if (LoadInst * load = dyn_cast<LoadInst>(II)) {
+         ++II) {
+      if (LoadInst *load = dyn_cast<LoadInst>(II)) {
 
-        if (load->hasOneUse() == false) continue;
+        if (load->hasOneUse() == false)
+          continue;
 
-        if (TD->getTypeStoreSize(load->getType()) < MaxAggrCopySize) continue;
+        if (TD->getTypeStoreSize(load->getType()) < MaxAggrCopySize)
+          continue;
 
         User *use = *(load->use_begin());
-        if (StoreInst * store = dyn_cast<StoreInst>(use)) {
+        if (StoreInst *store = dyn_cast<StoreInst>(use)) {
           if (store->getOperand(0) != load) //getValueOperand
-          continue;
+            continue;
           aggrLoads.push_back(load);
         }
-      } else if (MemTransferInst * intr = dyn_cast<MemTransferInst>(II)) {
+      } else if (MemTransferInst *intr = dyn_cast<MemTransferInst>(II)) {
         Value *len = intr->getLength();
         // If the number of elements being copied is greater
         // than MaxAggrCopySize, lower it to a loop
-        if (ConstantInt * len_int = dyn_cast < ConstantInt > (len)) {
+        if (ConstantInt *len_int = dyn_cast<ConstantInt>(len)) {
           if (len_int->getZExtValue() >= MaxAggrCopySize) {
             aggrMemcpys.push_back(intr);
           }
@@ -145,9 +141,9 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
           // turn variable length memcpy/memmov into loop
           aggrMemcpys.push_back(intr);
         }
-      } else if (MemSetInst * memsetintr = dyn_cast<MemSetInst>(II)) {
+      } else if (MemSetInst *memsetintr = dyn_cast<MemSetInst>(II)) {
         Value *len = memsetintr->getLength();
-        if (ConstantInt * len_int = dyn_cast<ConstantInt>(len)) {
+        if (ConstantInt *len_int = dyn_cast<ConstantInt>(len)) {
           if (len_int->getZExtValue() >= MaxAggrCopySize) {
             aggrMemsets.push_back(memsetintr);
           }
@@ -158,8 +154,9 @@ bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
       }
     }
   }
-  if ((aggrLoads.size() == 0) && (aggrMemcpys.size() == 0)
-      && (aggrMemsets.size() == 0)) return false;
+  if ((aggrLoads.size() == 0) && (aggrMemcpys.size() == 0) &&
+      (aggrMemsets.size() == 0))
+    return false;
 
   //
   // Do the transformation of an aggr load/copy/set to a loop
diff --git a/lib/Target/NVPTX/NVPTXLowerAggrCopies.h b/lib/Target/NVPTX/NVPTXLowerAggrCopies.h
index b150c69815dd..286e753fa92b 100644
--- a/lib/Target/NVPTX/NVPTXLowerAggrCopies.h
+++ b/lib/Target/NVPTX/NVPTXLowerAggrCopies.h
@@ -15,9 +15,9 @@
 #ifndef NVPTX_LOWER_AGGR_COPIES_H
 #define NVPTX_LOWER_AGGR_COPIES_H
 
-#include "llvm/Pass.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
-#include "llvm/DataLayout.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/Pass.h"
 
 namespace llvm {
 
diff --git a/lib/Target/NVPTX/NVPTXNumRegisters.h b/lib/Target/NVPTX/NVPTXNumRegisters.h
index b4a4dbce98a9..a95c16b1e67e 100644
--- a/lib/Target/NVPTX/NVPTXNumRegisters.h
+++ b/lib/Target/NVPTX/NVPTXNumRegisters.h
@@ -11,10 +11,6 @@
 #ifndef NVPTX_NUM_REGISTERS_H
 #define NVPTX_NUM_REGISTERS_H
 
-namespace llvm {
-
-const unsigned NVPTXNumRegisters = 396;
-
-}
+namespace llvm { const unsigned NVPTXNumRegisters = 396; }
 
 #endif
diff --git a/lib/Target/NVPTX/NVPTXRegisterInfo.cpp b/lib/Target/NVPTX/NVPTXRegisterInfo.cpp
index e3cd46f063bf..282465359b07 100644
--- a/lib/Target/NVPTX/NVPTXRegisterInfo.cpp
+++ b/lib/Target/NVPTX/NVPTXRegisterInfo.cpp
@@ -13,260 +13,88 @@
 
 #define DEBUG_TYPE "nvptx-reg-info"
 
-#include "NVPTX.h"
 #include "NVPTXRegisterInfo.h"
+#include "NVPTX.h"
 #include "NVPTXSubtarget.h"
 #include "llvm/ADT/BitVector.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/MC/MachineLocation.h"
 #include "llvm/Target/TargetInstrInfo.h"
 
-
 using namespace llvm;
 
-namespace llvm
-{
-std::string getNVPTXRegClassName (TargetRegisterClass const *RC) {
+namespace llvm {
+std::string getNVPTXRegClassName(TargetRegisterClass const *RC) {
   if (RC == &NVPTX::Float32RegsRegClass) {
     return ".f32";
   }
   if (RC == &NVPTX::Float64RegsRegClass) {
     return ".f64";
-  }
-  else if (RC == &NVPTX::Int64RegsRegClass) {
+  } else if (RC == &NVPTX::Int64RegsRegClass) {
     return ".s64";
-  }
-  else if (RC == &NVPTX::Int32RegsRegClass) {
+  } else if (RC == &NVPTX::Int32RegsRegClass) {
     return ".s32";
-  }
-  else if (RC == &NVPTX::Int16RegsRegClass) {
+  } else if (RC == &NVPTX::Int16RegsRegClass) {
     return ".s16";
   }
-  // Int8Regs become 16-bit registers in PTX
-  else if (RC == &NVPTX::Int8RegsRegClass) {
+      // Int8Regs become 16-bit registers in PTX
+      else if (RC == &NVPTX::Int8RegsRegClass) {
     return ".s16";
-  }
-  else if (RC == &NVPTX::Int1RegsRegClass) {
+  } else if (RC == &NVPTX::Int1RegsRegClass) {
     return ".pred";
-  }
-  else if (RC == &NVPTX::SpecialRegsRegClass) {
+  } else if (RC == &NVPTX::SpecialRegsRegClass) {
     return "!Special!";
-  }
-  else if (RC == &NVPTX::V2F32RegsRegClass) {
-    return ".v2.f32";
-  }
-  else if (RC == &NVPTX::V4F32RegsRegClass) {
-    return ".v4.f32";
-  }
-  else if (RC == &NVPTX::V2I32RegsRegClass) {
-    return ".v2.s32";
-  }
-  else if (RC == &NVPTX::V4I32RegsRegClass) {
-    return ".v4.s32";
-  }
-  else if (RC == &NVPTX::V2F64RegsRegClass) {
-    return ".v2.f64";
-  }
-  else if (RC == &NVPTX::V2I64RegsRegClass) {
-    return ".v2.s64";
-  }
-  else if (RC == &NVPTX::V2I16RegsRegClass) {
-    return ".v2.s16";
-  }
-  else if (RC == &NVPTX::V4I16RegsRegClass) {
-    return ".v4.s16";
-  }
-  else if (RC == &NVPTX::V2I8RegsRegClass) {
-    return ".v2.s16";
-  }
-  else if (RC == &NVPTX::V4I8RegsRegClass) {
-    return ".v4.s16";
-  }
-  else {
+  } else {
     return "INTERNAL";
   }
   return "";
 }
 
-std::string getNVPTXRegClassStr (TargetRegisterClass const *RC) {
+std::string getNVPTXRegClassStr(TargetRegisterClass const *RC) {
   if (RC == &NVPTX::Float32RegsRegClass) {
     return "%f";
   }
   if (RC == &NVPTX::Float64RegsRegClass) {
     return "%fd";
-  }
-  else if (RC == &NVPTX::Int64RegsRegClass) {
+  } else if (RC == &NVPTX::Int64RegsRegClass) {
     return "%rd";
-  }
-  else if (RC == &NVPTX::Int32RegsRegClass) {
+  } else if (RC == &NVPTX::Int32RegsRegClass) {
     return "%r";
-  }
-  else if (RC == &NVPTX::Int16RegsRegClass) {
+  } else if (RC == &NVPTX::Int16RegsRegClass) {
     return "%rs";
-  }
-  else if (RC == &NVPTX::Int8RegsRegClass) {
+  } else if (RC == &NVPTX::Int8RegsRegClass) {
     return "%rc";
-  }
-  else if (RC == &NVPTX::Int1RegsRegClass) {
+  } else if (RC == &NVPTX::Int1RegsRegClass) {
     return "%p";
-  }
-  else if (RC == &NVPTX::SpecialRegsRegClass) {
+  } else if (RC == &NVPTX::SpecialRegsRegClass) {
     return "!Special!";
-  }
-  else if (RC == &NVPTX::V2F32RegsRegClass) {
-    return "%v2f";
-  }
-  else if (RC == &NVPTX::V4F32RegsRegClass) {
-    return "%v4f";
-  }
-  else if (RC == &NVPTX::V2I32RegsRegClass) {
-    return "%v2r";
-  }
-  else if (RC == &NVPTX::V4I32RegsRegClass) {
-    return "%v4r";
-  }
-  else if (RC == &NVPTX::V2F64RegsRegClass) {
-    return "%v2fd";
-  }
-  else if (RC == &NVPTX::V2I64RegsRegClass) {
-    return "%v2rd";
-  }
-  else if (RC == &NVPTX::V2I16RegsRegClass) {
-    return "%v2s";
-  }
-  else if (RC == &NVPTX::V4I16RegsRegClass) {
-    return "%v4rs";
-  }
-  else if (RC == &NVPTX::V2I8RegsRegClass) {
-    return "%v2rc";
-  }
-  else if (RC == &NVPTX::V4I8RegsRegClass) {
-    return "%v4rc";
-  }
-  else {
+  } else {
     return "INTERNAL";
   }
   return "";
 }
-
-bool isNVPTXVectorRegClass(TargetRegisterClass const *RC) {
-  if (RC->getID() == NVPTX::V2F32RegsRegClassID)
-    return true;
-  if (RC->getID() == NVPTX::V2F64RegsRegClassID)
-    return true;
-  if (RC->getID() == NVPTX::V2I16RegsRegClassID)
-    return true;
-  if (RC->getID() == NVPTX::V2I32RegsRegClassID)
-    return true;
-  if (RC->getID() == NVPTX::V2I64RegsRegClassID)
-    return true;
-  if (RC->getID() == NVPTX::V2I8RegsRegClassID)
-    return true;
-  if (RC->getID() == NVPTX::V4F32RegsRegClassID)
-    return true;
-  if (RC->getID() == NVPTX::V4I16RegsRegClassID)
-    return true;
-  if (RC->getID() == NVPTX::V4I32RegsRegClassID)
-    return true;
-  if (RC->getID() == NVPTX::V4I8RegsRegClassID)
-    return true;
-  return false;
-}
-
-std::string getNVPTXElemClassName(TargetRegisterClass const *RC) {
-  if (RC->getID() == NVPTX::V2F32RegsRegClassID)
-    return getNVPTXRegClassName(&NVPTX::Float32RegsRegClass);
-  if (RC->getID() == NVPTX::V2F64RegsRegClassID)
-    return getNVPTXRegClassName(&NVPTX::Float64RegsRegClass);
-  if (RC->getID() == NVPTX::V2I16RegsRegClassID)
-    return getNVPTXRegClassName(&NVPTX::Int16RegsRegClass);
-  if (RC->getID() == NVPTX::V2I32RegsRegClassID)
-    return getNVPTXRegClassName(&NVPTX::Int32RegsRegClass);
-  if (RC->getID() == NVPTX::V2I64RegsRegClassID)
-    return getNVPTXRegClassName(&NVPTX::Int64RegsRegClass);
-  if (RC->getID() == NVPTX::V2I8RegsRegClassID)
-    return getNVPTXRegClassName(&NVPTX::Int8RegsRegClass);
-  if (RC->getID() == NVPTX::V4F32RegsRegClassID)
-    return getNVPTXRegClassName(&NVPTX::Float32RegsRegClass);
-  if (RC->getID() == NVPTX::V4I16RegsRegClassID)
-    return getNVPTXRegClassName(&NVPTX::Int16RegsRegClass);
-  if (RC->getID() == NVPTX::V4I32RegsRegClassID)
-    return getNVPTXRegClassName(&NVPTX::Int32RegsRegClass);
-  if (RC->getID() == NVPTX::V4I8RegsRegClassID)
-    return getNVPTXRegClassName(&NVPTX::Int8RegsRegClass);
-  llvm_unreachable("Not a vector register class");
-}
-
-const TargetRegisterClass *getNVPTXElemClass(TargetRegisterClass const *RC) {
-  if (RC->getID() == NVPTX::V2F32RegsRegClassID)
-    return (&NVPTX::Float32RegsRegClass);
-  if (RC->getID() == NVPTX::V2F64RegsRegClassID)
-    return (&NVPTX::Float64RegsRegClass);
-  if (RC->getID() == NVPTX::V2I16RegsRegClassID)
-    return (&NVPTX::Int16RegsRegClass);
-  if (RC->getID() == NVPTX::V2I32RegsRegClassID)
-    return (&NVPTX::Int32RegsRegClass);
-  if (RC->getID() == NVPTX::V2I64RegsRegClassID)
-    return (&NVPTX::Int64RegsRegClass);
-  if (RC->getID() == NVPTX::V2I8RegsRegClassID)
-    return (&NVPTX::Int8RegsRegClass);
-  if (RC->getID() == NVPTX::V4F32RegsRegClassID)
-    return (&NVPTX::Float32RegsRegClass);
-  if (RC->getID() == NVPTX::V4I16RegsRegClassID)
-    return (&NVPTX::Int16RegsRegClass);
-  if (RC->getID() == NVPTX::V4I32RegsRegClassID)
-    return (&NVPTX::Int32RegsRegClass);
-  if (RC->getID() == NVPTX::V4I8RegsRegClassID)
-    return (&NVPTX::Int8RegsRegClass);
-  llvm_unreachable("Not a vector register class");
-}
-
-int getNVPTXVectorSize(TargetRegisterClass const *RC) {
-  if (RC->getID() == NVPTX::V2F32RegsRegClassID)
-    return 2;
-  if (RC->getID() == NVPTX::V2F64RegsRegClassID)
-    return 2;
-  if (RC->getID() == NVPTX::V2I16RegsRegClassID)
-    return 2;
-  if (RC->getID() == NVPTX::V2I32RegsRegClassID)
-    return 2;
-  if (RC->getID() == NVPTX::V2I64RegsRegClassID)
-    return 2;
-  if (RC->getID() == NVPTX::V2I8RegsRegClassID)
-    return 2;
-  if (RC->getID() == NVPTX::V4F32RegsRegClassID)
-    return 4;
-  if (RC->getID() == NVPTX::V4I16RegsRegClassID)
-    return 4;
-  if (RC->getID() == NVPTX::V4I32RegsRegClassID)
-    return 4;
-  if (RC->getID() == NVPTX::V4I8RegsRegClassID)
-    return 4;
-  llvm_unreachable("Not a vector register class");
-}
 }
 
 NVPTXRegisterInfo::NVPTXRegisterInfo(const TargetInstrInfo &tii,
                                      const NVPTXSubtarget &st)
-  : NVPTXGenRegisterInfo(0),
-    Is64Bit(st.is64Bit()) {}
+    : NVPTXGenRegisterInfo(0), Is64Bit(st.is64Bit()) {}
 
 #define GET_REGINFO_TARGET_DESC
 #include "NVPTXGenRegisterInfo.inc"
 
 /// NVPTX Callee Saved Registers
-const uint16_t* NVPTXRegisterInfo::
-getCalleeSavedRegs(const MachineFunction *MF) const {
+const uint16_t *
+NVPTXRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
   static const uint16_t CalleeSavedRegs[] = { 0 };
   return CalleeSavedRegs;
 }
 
 // NVPTX Callee Saved Reg Classes
-const TargetRegisterClass* const*
+const TargetRegisterClass *const *
 NVPTXRegisterInfo::getCalleeSavedRegClasses(const MachineFunction *MF) const {
-  static const TargetRegisterClass * const CalleeSavedRegClasses[] = { 0 };
+  static const TargetRegisterClass *const CalleeSavedRegClasses[] = { 0 };
   return CalleeSavedRegClasses;
 }
 
@@ -275,34 +103,24 @@ BitVector NVPTXRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   return Reserved;
 }
 
-void NVPTXRegisterInfo::
-eliminateFrameIndex(MachineBasicBlock::iterator II,
-                    int SPAdj,
-                    RegScavenger *RS) const {
+void NVPTXRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
+                                            int SPAdj, unsigned FIOperandNum,
+                                            RegScavenger *RS) const {
   assert(SPAdj == 0 && "Unexpected");
 
-  unsigned i = 0;
   MachineInstr &MI = *II;
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() &&
-           "Instr doesn't have FrameIndex operand!");
-  }
-
-  int FrameIndex = MI.getOperand(i).getIndex();
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
 
   MachineFunction &MF = *MI.getParent()->getParent();
   int Offset = MF.getFrameInfo()->getObjectOffset(FrameIndex) +
-      MI.getOperand(i+1).getImm();
+               MI.getOperand(FIOperandNum + 1).getImm();
 
   // Using I0 as the frame pointer
-  MI.getOperand(i).ChangeToRegister(NVPTX::VRFrame, false);
-  MI.getOperand(i+1).ChangeToImmediate(Offset);
+  MI.getOperand(FIOperandNum).ChangeToRegister(NVPTX::VRFrame, false);
+  MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset);
 }
 
-
-int NVPTXRegisterInfo::
-getDwarfRegNum(unsigned RegNum, bool isEH) const {
+int NVPTXRegisterInfo::getDwarfRegNum(unsigned RegNum, bool isEH) const {
   return 0;
 }
 
@@ -310,16 +128,4 @@ unsigned NVPTXRegisterInfo::getFrameRegister(const MachineFunction &MF) const {
   return NVPTX::VRFrame;
 }
 
-unsigned NVPTXRegisterInfo::getRARegister() const {
-  return 0;
-}
-
-// This function eliminates ADJCALLSTACKDOWN,
-// ADJCALLSTACKUP pseudo instructions
-void NVPTXRegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  // Simply discard ADJCALLSTACKDOWN,
-  // ADJCALLSTACKUP instructions.
-  MBB.erase(I);
-}
+unsigned NVPTXRegisterInfo::getRARegister() const { return 0; }
diff --git a/lib/Target/NVPTX/NVPTXRegisterInfo.h b/lib/Target/NVPTX/NVPTXRegisterInfo.h
index 595178335ae2..d40682066142 100644
--- a/lib/Target/NVPTX/NVPTXRegisterInfo.h
+++ b/lib/Target/NVPTX/NVPTXRegisterInfo.h
@@ -17,7 +17,6 @@
 #include "ManagedStringPool.h"
 #include "llvm/Target/TargetRegisterInfo.h"
 
-
 #define GET_REGINFO_HEADER
 #include "NVPTXGenRegisterInfo.inc"
 #include "llvm/Target/TargetRegisterInfo.h"
@@ -33,34 +32,28 @@ class NVPTXRegisterInfo : public NVPTXGenRegisterInfo {
 private:
   bool Is64Bit;
   // Hold Strings that can be free'd all together with NVPTXRegisterInfo
-  ManagedStringPool     ManagedStrPool;
+  ManagedStringPool ManagedStrPool;
 
 public:
-  NVPTXRegisterInfo(const TargetInstrInfo &tii,
-                    const NVPTXSubtarget &st);
-
+  NVPTXRegisterInfo(const TargetInstrInfo &tii, const NVPTXSubtarget &st);
 
   //------------------------------------------------------
   // Pure virtual functions from TargetRegisterInfo
   //------------------------------------------------------
 
   // NVPTX callee saved registers
-  virtual const uint16_t*
+  virtual const uint16_t *
   getCalleeSavedRegs(const MachineFunction *MF = 0) const;
 
   // NVPTX callee saved register classes
-  virtual const TargetRegisterClass* const *
+  virtual const TargetRegisterClass *const *
   getCalleeSavedRegClasses(const MachineFunction *MF) const;
 
   virtual BitVector getReservedRegs(const MachineFunction &MF) const;
 
-  virtual void eliminateFrameIndex(MachineBasicBlock::iterator MI,
-                                   int SPAdj,
-                                   RegScavenger *RS=NULL) const;
-
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
+  virtual void eliminateFrameIndex(MachineBasicBlock::iterator MI, int SPAdj,
+                                   unsigned FIOperandNum,
+                                   RegScavenger *RS = NULL) const;
 
   virtual int getDwarfRegNum(unsigned RegNum, bool isEH) const;
   virtual unsigned getFrameRegister(const MachineFunction &MF) const;
@@ -78,15 +71,9 @@ public:
 
 };
 
-
-std::string getNVPTXRegClassName (const TargetRegisterClass *RC);
-std::string getNVPTXRegClassStr (const TargetRegisterClass *RC);
-bool isNVPTXVectorRegClass (const TargetRegisterClass *RC);
-std::string getNVPTXElemClassName (const TargetRegisterClass *RC);
-int getNVPTXVectorSize (const TargetRegisterClass *RC);
-const TargetRegisterClass *getNVPTXElemClass(const TargetRegisterClass *RC);
+std::string getNVPTXRegClassName(const TargetRegisterClass *RC);
+std::string getNVPTXRegClassStr(const TargetRegisterClass *RC);
 
 } // end namespace llvm
 
-
 #endif
diff --git a/lib/Target/NVPTX/NVPTXRegisterInfo.td b/lib/Target/NVPTX/NVPTXRegisterInfo.td
index ba158258b994..8d100d631683 100644
--- a/lib/Target/NVPTX/NVPTXRegisterInfo.td
+++ b/lib/Target/NVPTX/NVPTXRegisterInfo.td
@@ -37,9 +37,6 @@ foreach i = 0-395 in {
   def RL#i : NVPTXReg<"%rl"#i>; // 64-bit
   def F#i  : NVPTXReg<"%f"#i>;  // 32-bit float
   def FL#i : NVPTXReg<"%fl"#i>; // 64-bit float
-  // Vectors
-  foreach s = [ "2b8", "2b16", "2b32", "2b64", "4b8", "4b16", "4b32" ] in
-    def v#s#_#i : NVPTXReg<"%v"#s#"_"#i>;
 
   // Arguments
   def ia#i : NVPTXReg<"%ia"#i>;
@@ -65,44 +62,3 @@ def Float64ArgRegs : NVPTXRegClass<[f64], 64, (add (sequence "da%u", 0, 395))>;
 
 // Read NVPTXRegisterInfo.cpp to see how VRFrame and VRDepot are used.
 def SpecialRegs : NVPTXRegClass<[i32], 32, (add VRFrame, VRDepot)>;
-
-class NVPTXVecRegClass<list<ValueType> regTypes, int alignment, dag regList,
-                       NVPTXRegClass sClass,
-                       int e,
-                       string n>
-  : NVPTXRegClass<regTypes, alignment, regList>
-{
-  NVPTXRegClass scalarClass=sClass;
-  int elems=e;
-  string name=n;
-}
-def V2F32Regs
-  : NVPTXVecRegClass<[v2f32], 64, (add (sequence "v2b32_%u", 0, 395)),
-    Float32Regs, 2, ".v2.f32">;
-def V4F32Regs
-  : NVPTXVecRegClass<[v4f32], 128, (add (sequence "v4b32_%u", 0, 395)),
-    Float32Regs, 4, ".v4.f32">;
-def V2I32Regs
-  : NVPTXVecRegClass<[v2i32], 64, (add (sequence "v2b32_%u", 0, 395)),
-    Int32Regs, 2, ".v2.u32">;
-def V4I32Regs
-  : NVPTXVecRegClass<[v4i32], 128, (add (sequence "v4b32_%u", 0, 395)),
-    Int32Regs, 4, ".v4.u32">;
-def V2F64Regs
-  : NVPTXVecRegClass<[v2f64], 128, (add (sequence "v2b64_%u", 0, 395)),
-    Float64Regs, 2, ".v2.f64">;
-def V2I64Regs
-  : NVPTXVecRegClass<[v2i64], 128, (add (sequence "v2b64_%u", 0, 395)),
-    Int64Regs, 2, ".v2.u64">;
-def V2I16Regs
-  : NVPTXVecRegClass<[v2i16], 32, (add (sequence "v2b16_%u", 0, 395)),
-    Int16Regs, 2, ".v2.u16">;
-def V4I16Regs
-  : NVPTXVecRegClass<[v4i16], 64, (add (sequence "v4b16_%u", 0, 395)),
-    Int16Regs, 4, ".v4.u16">;
-def V2I8Regs
-  : NVPTXVecRegClass<[v2i8], 16, (add (sequence "v2b8_%u", 0, 395)),
-    Int8Regs, 2, ".v2.u8">;
-def V4I8Regs
-  : NVPTXVecRegClass<[v4i8], 32, (add (sequence "v4b8_%u", 0, 395)),
-    Int8Regs, 4, ".v4.u8">;
diff --git a/lib/Target/NVPTX/NVPTXSection.h b/lib/Target/NVPTX/NVPTXSection.h
index f1ca466266f6..e166be5a68e4 100644
--- a/lib/Target/NVPTX/NVPTXSection.h
+++ b/lib/Target/NVPTX/NVPTXSection.h
@@ -14,8 +14,8 @@
 #ifndef LLVM_NVPTXSECTION_H
 #define LLVM_NVPTXSECTION_H
 
+#include "llvm/IR/GlobalVariable.h"
 #include "llvm/MC/MCSection.h"
-#include "llvm/GlobalVariable.h"
 #include <vector>
 
 namespace llvm {
@@ -38,6 +38,8 @@ public:
   virtual bool isBaseAddressKnownZero() const { return true; }
   virtual bool UseCodeAlign() const { return false; }
   virtual bool isVirtualSection() const { return false; }
+  virtual std::string getLabelBeginName() const { return ""; }
+  virtual std::string getLabelEndName() const { return ""; }
 };
 
 } // end namespace llvm
diff --git a/lib/Target/NVPTX/NVPTXSplitBBatBar.cpp b/lib/Target/NVPTX/NVPTXSplitBBatBar.cpp
index 2836cad4f021..83dfe120899a 100644
--- a/lib/Target/NVPTX/NVPTXSplitBBatBar.cpp
+++ b/lib/Target/NVPTX/NVPTXSplitBBatBar.cpp
@@ -11,19 +11,17 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/Function.h"
-#include "llvm/Instructions.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/Support/InstIterator.h"
-#include "NVPTXUtilities.h"
 #include "NVPTXSplitBBatBar.h"
+#include "NVPTXUtilities.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/Support/InstIterator.h"
 
 using namespace llvm;
 
-namespace llvm {
-FunctionPass *createSplitBBatBarPass();
-}
+namespace llvm { FunctionPass *createSplitBBatBarPass(); }
 
 char NVPTXSplitBBatBar::ID = 0;
 
@@ -72,6 +70,4 @@ bool NVPTXSplitBBatBar::runOnFunction(Function &F) {
 // This interface will most likely not be necessary, because this pass will
 // not be invoked by the driver, but will be used as a prerequisite to
 // another pass.
-FunctionPass *llvm::createSplitBBatBarPass() {
-  return new NVPTXSplitBBatBar();
-}
+FunctionPass *llvm::createSplitBBatBarPass() { return new NVPTXSplitBBatBar(); }
diff --git a/lib/Target/NVPTX/NVPTXSplitBBatBar.h b/lib/Target/NVPTX/NVPTXSplitBBatBar.h
index 9e4d5a066d4c..bdafba9075a0 100644
--- a/lib/Target/NVPTX/NVPTXSplitBBatBar.h
+++ b/lib/Target/NVPTX/NVPTXSplitBBatBar.h
@@ -15,8 +15,8 @@
 #ifndef NVPTX_SPLIT_BB_AT_BAR_H
 #define NVPTX_SPLIT_BB_AT_BAR_H
 
-#include "llvm/Pass.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/Pass.h"
 
 namespace llvm {
 
diff --git a/lib/Target/NVPTX/NVPTXSubtarget.cpp b/lib/Target/NVPTX/NVPTXSubtarget.cpp
index 7b62cce2c65c..2dcd73dcff9c 100644
--- a/lib/Target/NVPTX/NVPTXSubtarget.cpp
+++ b/lib/Target/NVPTX/NVPTXSubtarget.cpp
@@ -22,27 +22,23 @@ using namespace llvm;
 // Select Driver Interface
 #include "llvm/Support/CommandLine.h"
 namespace {
-cl::opt<NVPTX::DrvInterface>
-DriverInterface(cl::desc("Choose driver interface:"),
-                cl::values(
-                    clEnumValN(NVPTX::NVCL, "drvnvcl", "Nvidia OpenCL driver"),
-                    clEnumValN(NVPTX::CUDA, "drvcuda", "Nvidia CUDA driver"),
-                    clEnumValN(NVPTX::TEST, "drvtest", "Plain Test"),
-                    clEnumValEnd),
-                    cl::init(NVPTX::NVCL));
+cl::opt<NVPTX::DrvInterface> DriverInterface(
+    cl::desc("Choose driver interface:"),
+    cl::values(clEnumValN(NVPTX::NVCL, "drvnvcl", "Nvidia OpenCL driver"),
+               clEnumValN(NVPTX::CUDA, "drvcuda", "Nvidia CUDA driver"),
+               clEnumValN(NVPTX::TEST, "drvtest", "Plain Test"), clEnumValEnd),
+    cl::init(NVPTX::NVCL));
 }
 
 NVPTXSubtarget::NVPTXSubtarget(const std::string &TT, const std::string &CPU,
                                const std::string &FS, bool is64Bit)
-: NVPTXGenSubtargetInfo(TT, CPU, FS),
-  Is64Bit(is64Bit),
-  PTXVersion(0),
-  SmVersion(10) {
+    : NVPTXGenSubtargetInfo(TT, CPU, FS), Is64Bit(is64Bit), PTXVersion(0),
+      SmVersion(20) {
 
   drvInterface = DriverInterface;
 
   // Provide the default CPU if none
-  std::string defCPU = "sm_10";
+  std::string defCPU = "sm_20";
 
   ParseSubtargetFeatures((CPU.empty() ? defCPU : CPU), FS);
 
diff --git a/lib/Target/NVPTX/NVPTXSubtarget.h b/lib/Target/NVPTX/NVPTXSubtarget.h
index 3cfd9718e541..670077daaa69 100644
--- a/lib/Target/NVPTX/NVPTXSubtarget.h
+++ b/lib/Target/NVPTX/NVPTXSubtarget.h
@@ -14,8 +14,8 @@
 #ifndef NVPTXSUBTARGET_H
 #define NVPTXSUBTARGET_H
 
-#include "llvm/Target/TargetSubtargetInfo.h"
 #include "NVPTX.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
 
 #define GET_SUBTARGETINFO_HEADER
 #include "NVPTXGenSubtargetInfo.inc"
@@ -25,7 +25,7 @@
 namespace llvm {
 
 class NVPTXSubtarget : public NVPTXGenSubtargetInfo {
-  
+
   std::string TargetName;
   NVPTX::DrvInterface drvInterface;
   bool Is64Bit;
@@ -57,16 +57,14 @@ public:
   bool hasF32FTZ() const { return SmVersion >= 20; }
   bool hasFMAF32() const { return SmVersion >= 20; }
   bool hasFMAF64() const { return SmVersion >= 13; }
+  bool hasLDG() const { return SmVersion >= 32; }
   bool hasLDU() const { return SmVersion >= 20; }
   bool hasGenericLdSt() const { return SmVersion >= 20; }
   inline bool hasHWROT32() const { return false; }
-  inline bool hasSWROT32() const {
-    return true;
-  }
-  inline bool hasROT32() const { return hasHWROT32() || hasSWROT32() ; }
+  inline bool hasSWROT32() const { return true; }
+  inline bool hasROT32() const { return hasHWROT32() || hasSWROT32(); }
   inline bool hasROT64() const { return SmVersion >= 20; }
 
-
   bool is64Bit() const { return Is64Bit; }
 
   unsigned int getSmVersion() const { return SmVersion; }
@@ -95,4 +93,4 @@ public:
 
 } // End llvm namespace
 
-#endif  // NVPTXSUBTARGET_H
+#endif // NVPTXSUBTARGET_H
diff --git a/lib/Target/NVPTX/NVPTXTargetMachine.cpp b/lib/Target/NVPTX/NVPTXTargetMachine.cpp
index cbb490003d37..67ca6b58e5a6 100644
--- a/lib/Target/NVPTX/NVPTXTargetMachine.cpp
+++ b/lib/Target/NVPTX/NVPTXTargetMachine.cpp
@@ -12,27 +12,30 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTXTargetMachine.h"
-#include "NVPTX.h"
-#include "NVPTXSplitBBatBar.h"
-#include "NVPTXLowerAggrCopies.h"
 #include "MCTargetDesc/NVPTXMCAsmInfo.h"
+#include "NVPTX.h"
 #include "NVPTXAllocaHoisting.h"
-#include "llvm/PassManager.h"
+#include "NVPTXLowerAggrCopies.h"
+#include "NVPTXSplitBBatBar.h"
+#include "llvm/ADT/OwningPtr.h"
 #include "llvm/Analysis/Passes.h"
 #include "llvm/Analysis/Verifier.h"
 #include "llvm/Assembly/PrintModulePass.h"
-#include "llvm/ADT/OwningPtr.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/Passes.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/PassManager.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/FormattedStream.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetLowering.h"
 #include "llvm/Target/TargetLoweringObjectFile.h"
@@ -41,14 +44,12 @@
 #include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Target/TargetSubtargetInfo.h"
 #include "llvm/Transforms/Scalar.h"
-#include "llvm/Support/CommandLine.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/FormattedStream.h"
-#include "llvm/Support/TargetRegistry.h"
-
 
 using namespace llvm;
 
+namespace llvm {
+void initializeNVVMReflectPass(PassRegistry&);
+}
 
 extern "C" void LLVMInitializeNVPTXTarget() {
   // Register the target.
@@ -58,53 +59,42 @@ extern "C" void LLVMInitializeNVPTXTarget() {
   RegisterMCAsmInfo<NVPTXMCAsmInfo> A(TheNVPTXTarget32);
   RegisterMCAsmInfo<NVPTXMCAsmInfo> B(TheNVPTXTarget64);
 
+  // FIXME: This pass is really intended to be invoked during IR optimization,
+  // but it's very NVPTX-specific.
+  initializeNVVMReflectPass(*PassRegistry::getPassRegistry());
 }
 
-NVPTXTargetMachine::NVPTXTargetMachine(const Target &T,
-                                       StringRef TT,
-                                       StringRef CPU,
-                                       StringRef FS,
-                                       const TargetOptions& Options,
-                                       Reloc::Model RM,
-                                       CodeModel::Model CM,
-                                       CodeGenOpt::Level OL,
-                                       bool is64bit)
-: LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
-  Subtarget(TT, CPU, FS, is64bit),
-  DL(Subtarget.getDataLayout()),
-  InstrInfo(*this), TLInfo(*this), TSInfo(*this), FrameLowering(*this,is64bit),
-  STTI(&TLInfo), VTTI(&TLInfo)
-/*FrameInfo(TargetFrameInfo::StackGrowsUp, 8, 0)*/ {
-}
-
-
+NVPTXTargetMachine::NVPTXTargetMachine(
+    const Target &T, StringRef TT, StringRef CPU, StringRef FS,
+    const TargetOptions &Options, Reloc::Model RM, CodeModel::Model CM,
+    CodeGenOpt::Level OL, bool is64bit)
+    : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+      Subtarget(TT, CPU, FS, is64bit), DL(Subtarget.getDataLayout()),
+      InstrInfo(*this), TLInfo(*this), TSInfo(*this),
+      FrameLowering(
+          *this, is64bit) /*FrameInfo(TargetFrameInfo::StackGrowsUp, 8, 0)*/ {}
 
 void NVPTXTargetMachine32::anchor() {}
 
-NVPTXTargetMachine32::NVPTXTargetMachine32(const Target &T, StringRef TT,
-                                           StringRef CPU, StringRef FS,
-                                           const TargetOptions &Options,
-                                           Reloc::Model RM, CodeModel::Model CM,
-                                           CodeGenOpt::Level OL)
-: NVPTXTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL, false) {
-}
+NVPTXTargetMachine32::NVPTXTargetMachine32(
+    const Target &T, StringRef TT, StringRef CPU, StringRef FS,
+    const TargetOptions &Options, Reloc::Model RM, CodeModel::Model CM,
+    CodeGenOpt::Level OL)
+    : NVPTXTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL, false) {}
 
 void NVPTXTargetMachine64::anchor() {}
 
-NVPTXTargetMachine64::NVPTXTargetMachine64(const Target &T, StringRef TT,
-                                           StringRef CPU, StringRef FS,
-                                           const TargetOptions &Options,
-                                           Reloc::Model RM, CodeModel::Model CM,
-                                           CodeGenOpt::Level OL)
-: NVPTXTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL, true) {
-}
-
+NVPTXTargetMachine64::NVPTXTargetMachine64(
+    const Target &T, StringRef TT, StringRef CPU, StringRef FS,
+    const TargetOptions &Options, Reloc::Model RM, CodeModel::Model CM,
+    CodeGenOpt::Level OL)
+    : NVPTXTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL, true) {}
 
 namespace llvm {
 class NVPTXPassConfig : public TargetPassConfig {
 public:
   NVPTXPassConfig(NVPTXTargetMachine *TM, PassManagerBase &PM)
-  : TargetPassConfig(TM, PM) {}
+      : TargetPassConfig(TM, PM) {}
 
   NVPTXTargetMachine &getNVPTXTargetMachine() const {
     return getTM<NVPTXTargetMachine>();
@@ -125,10 +115,7 @@ bool NVPTXPassConfig::addInstSelector() {
   addPass(createSplitBBatBarPass());
   addPass(createAllocaHoisting());
   addPass(createNVPTXISelDag(getNVPTXTargetMachine(), getOptLevel()));
-  addPass(createVectorElementizePass(getNVPTXTargetMachine()));
   return false;
 }
 
-bool NVPTXPassConfig::addPreRegAlloc() {
-  return false;
-}
+bool NVPTXPassConfig::addPreRegAlloc() { return false; }
diff --git a/lib/Target/NVPTX/NVPTXTargetMachine.h b/lib/Target/NVPTX/NVPTXTargetMachine.h
index 11bc9d4fa698..5fbcf735b48f 100644
--- a/lib/Target/NVPTX/NVPTXTargetMachine.h
+++ b/lib/Target/NVPTX/NVPTXTargetMachine.h
@@ -11,77 +11,64 @@
 //
 //===----------------------------------------------------------------------===//
 
-
 #ifndef NVPTX_TARGETMACHINE_H
 #define NVPTX_TARGETMACHINE_H
 
-#include "NVPTXInstrInfo.h"
+#include "ManagedStringPool.h"
+#include "NVPTXFrameLowering.h"
 #include "NVPTXISelLowering.h"
+#include "NVPTXInstrInfo.h"
 #include "NVPTXRegisterInfo.h"
 #include "NVPTXSubtarget.h"
-#include "NVPTXFrameLowering.h"
-#include "ManagedStringPool.h"
-#include "llvm/DataLayout.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Target/TargetFrameLowering.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetSelectionDAGInfo.h"
-#include "llvm/Target/TargetTransformImpl.h"
 
 namespace llvm {
 
 /// NVPTXTargetMachine
 ///
 class NVPTXTargetMachine : public LLVMTargetMachine {
-  NVPTXSubtarget        Subtarget;
-  const DataLayout      DL;       // Calculates type size & alignment
-  NVPTXInstrInfo        InstrInfo;
-  NVPTXTargetLowering   TLInfo;
-  TargetSelectionDAGInfo   TSInfo;
+  NVPTXSubtarget Subtarget;
+  const DataLayout DL; // Calculates type size & alignment
+  NVPTXInstrInfo InstrInfo;
+  NVPTXTargetLowering TLInfo;
+  TargetSelectionDAGInfo TSInfo;
 
   // NVPTX does not have any call stack frame, but need a NVPTX specific
   // FrameLowering class because TargetFrameLowering is abstract.
-  NVPTXFrameLowering       FrameLowering;
+  NVPTXFrameLowering FrameLowering;
 
   // Hold Strings that can be free'd all together with NVPTXTargetMachine
-  ManagedStringPool     ManagedStrPool;
-
-  ScalarTargetTransformImpl STTI;
-  VectorTargetTransformImpl VTTI;
+  ManagedStringPool ManagedStrPool;
 
   //bool addCommonCodeGenPasses(PassManagerBase &, CodeGenOpt::Level,
   //                            bool DisableVerify, MCContext *&OutCtx);
 
 public:
-  NVPTXTargetMachine(const Target &T, StringRef TT, StringRef CPU,
-                     StringRef FS, const TargetOptions &Options,
-                     Reloc::Model RM, CodeModel::Model CM,
-                     CodeGenOpt::Level OP,
-                     bool is64bit);
+  NVPTXTargetMachine(const Target &T, StringRef TT, StringRef CPU, StringRef FS,
+                     const TargetOptions &Options, Reloc::Model RM,
+                     CodeModel::Model CM, CodeGenOpt::Level OP, bool is64bit);
 
   virtual const TargetFrameLowering *getFrameLowering() const {
     return &FrameLowering;
   }
-  virtual const NVPTXInstrInfo *getInstrInfo() const  { return &InstrInfo; }
-  virtual const DataLayout *getDataLayout() const     { return &DL;}
-  virtual const NVPTXSubtarget *getSubtargetImpl() const { return &Subtarget;}
+  virtual const NVPTXInstrInfo *getInstrInfo() const { return &InstrInfo; }
+  virtual const DataLayout *getDataLayout() const { return &DL; }
+  virtual const NVPTXSubtarget *getSubtargetImpl() const { return &Subtarget; }
 
   virtual const NVPTXRegisterInfo *getRegisterInfo() const {
     return &(InstrInfo.getRegisterInfo());
   }
 
   virtual NVPTXTargetLowering *getTargetLowering() const {
-    return const_cast<NVPTXTargetLowering*>(&TLInfo);
+    return const_cast<NVPTXTargetLowering *>(&TLInfo);
   }
 
   virtual const TargetSelectionDAGInfo *getSelectionDAGInfo() const {
     return &TSInfo;
   }
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
 
   //virtual bool addInstSelector(PassManagerBase &PM,
   //                             CodeGenOpt::Level OptLevel);
@@ -89,22 +76,19 @@ public:
   //virtual bool addPreRegAlloc(PassManagerBase &, CodeGenOpt::Level);
 
   ManagedStringPool *getManagedStrPool() const {
-    return const_cast<ManagedStringPool*>(&ManagedStrPool);
+    return const_cast<ManagedStringPool *>(&ManagedStrPool);
   }
 
   virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
 
   // Emission of machine code through JITCodeEmitter is not supported.
-  virtual bool addPassesToEmitMachineCode(PassManagerBase &,
-                                          JITCodeEmitter &,
+  virtual bool addPassesToEmitMachineCode(PassManagerBase &, JITCodeEmitter &,
                                           bool = true) {
     return true;
   }
 
   // Emission of machine code through MCJIT is not supported.
-  virtual bool addPassesToEmitMC(PassManagerBase &,
-                                 MCContext *&,
-                                 raw_ostream &,
+  virtual bool addPassesToEmitMC(PassManagerBase &, MCContext *&, raw_ostream &,
                                  bool = true) {
     return true;
   }
@@ -129,7 +113,6 @@ public:
                        CodeGenOpt::Level OL);
 };
 
-
 } // end namespace llvm
 
 #endif
diff --git a/lib/Target/NVPTX/NVPTXTargetObjectFile.h b/lib/Target/NVPTX/NVPTXTargetObjectFile.h
index b5698a2fc08f..6ab0e08ad091 100644
--- a/lib/Target/NVPTX/NVPTXTargetObjectFile.h
+++ b/lib/Target/NVPTX/NVPTXTargetObjectFile.h
@@ -46,45 +46,43 @@ public:
   }
 
   virtual void Initialize(MCContext &ctx, const TargetMachine &TM) {
-    TextSection = new NVPTXSection(MCSection::SV_ELF,
-                                   SectionKind::getText());
-    DataSection = new NVPTXSection(MCSection::SV_ELF,
-                                   SectionKind::getDataRel());
-    BSSSection = new NVPTXSection(MCSection::SV_ELF,
-                                  SectionKind::getBSS());
-    ReadOnlySection = new NVPTXSection(MCSection::SV_ELF,
-                                       SectionKind::getReadOnly());
+    TextSection = new NVPTXSection(MCSection::SV_ELF, SectionKind::getText());
+    DataSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getDataRel());
+    BSSSection = new NVPTXSection(MCSection::SV_ELF, SectionKind::getBSS());
+    ReadOnlySection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getReadOnly());
 
-    StaticCtorSection = new NVPTXSection(MCSection::SV_ELF,
-                                         SectionKind::getMetadata());
-    StaticDtorSection = new NVPTXSection(MCSection::SV_ELF,
-                                         SectionKind::getMetadata());
-    LSDASection = new NVPTXSection(MCSection::SV_ELF,
-                                   SectionKind::getMetadata());
-    EHFrameSection = new NVPTXSection(MCSection::SV_ELF,
-                                      SectionKind::getMetadata());
-    DwarfAbbrevSection = new NVPTXSection(MCSection::SV_ELF,
-                                          SectionKind::getMetadata());
-    DwarfInfoSection = new NVPTXSection(MCSection::SV_ELF,
-                                        SectionKind::getMetadata());
-    DwarfLineSection = new NVPTXSection(MCSection::SV_ELF,
-                                        SectionKind::getMetadata());
-    DwarfFrameSection = new NVPTXSection(MCSection::SV_ELF,
-                                         SectionKind::getMetadata());
-    DwarfPubTypesSection = new NVPTXSection(MCSection::SV_ELF,
-                                            SectionKind::getMetadata());
-    DwarfDebugInlineSection = new NVPTXSection(MCSection::SV_ELF,
-                                               SectionKind::getMetadata());
-    DwarfStrSection = new NVPTXSection(MCSection::SV_ELF,
-                                       SectionKind::getMetadata());
-    DwarfLocSection = new NVPTXSection(MCSection::SV_ELF,
-                                       SectionKind::getMetadata());
-    DwarfARangesSection = new NVPTXSection(MCSection::SV_ELF,
-                                           SectionKind::getMetadata());
-    DwarfRangesSection = new NVPTXSection(MCSection::SV_ELF,
-                                          SectionKind::getMetadata());
-    DwarfMacroInfoSection = new NVPTXSection(MCSection::SV_ELF,
-                                             SectionKind::getMetadata());
+    StaticCtorSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    StaticDtorSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    LSDASection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    EHFrameSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfAbbrevSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfInfoSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfLineSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfFrameSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfPubTypesSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfDebugInlineSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfStrSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfLocSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfARangesSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfRangesSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
+    DwarfMacroInfoSection =
+        new NVPTXSection(MCSection::SV_ELF, SectionKind::getMetadata());
   }
 
   virtual const MCSection *getSectionForConstant(SectionKind Kind) const {
@@ -93,8 +91,7 @@ public:
 
   virtual const MCSection *
   getExplicitSectionGlobal(const GlobalValue *GV, SectionKind Kind,
-                           Mangler *Mang,
-                           const TargetMachine &TM) const {
+                           Mangler *Mang, const TargetMachine &TM) const {
     return DataSection;
   }
 
diff --git a/lib/Target/NVPTX/NVPTXUtilities.cpp b/lib/Target/NVPTX/NVPTXUtilities.cpp
index 3f52251cc1b2..6786eb02240c 100644
--- a/lib/Target/NVPTX/NVPTXUtilities.cpp
+++ b/lib/Target/NVPTX/NVPTXUtilities.cpp
@@ -12,11 +12,11 @@
 
 #include "NVPTXUtilities.h"
 #include "NVPTX.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/Function.h"
-#include "llvm/Module.h"
-#include "llvm/Constants.h"
-#include "llvm/Operator.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Operator.h"
 #include <algorithm>
 #include <cstring>
 #include <map>
@@ -34,7 +34,6 @@ typedef std::map<const Module *, global_val_annot_t> per_module_annot_t;
 
 ManagedStatic<per_module_annot_t> annotationCache;
 
-
 static void cacheAnnotationFromMD(const MDNode *md, key_val_pair_t &retval) {
   assert(md && "Invalid mdnode for annotation");
   assert((md->getNumOperands() % 2) == 1 && "Invalid number of operands");
@@ -46,7 +45,7 @@ static void cacheAnnotationFromMD(const MDNode *md, key_val_pair_t &retval) {
     assert(prop && "Annotation property not a string");
 
     // value
-    ConstantInt *Val = dyn_cast<ConstantInt>(md->getOperand(i+1));
+    ConstantInt *Val = dyn_cast<ConstantInt>(md->getOperand(i + 1));
     assert(Val && "Value operand not a constant int");
 
     std::string keyname = prop->getString().str();
@@ -120,9 +119,9 @@ bool llvm::findAllNVVMAnnotation(const GlobalValue *gv, std::string prop,
 bool llvm::isTexture(const llvm::Value &val) {
   if (const GlobalValue *gv = dyn_cast<GlobalValue>(&val)) {
     unsigned annot;
-    if (llvm::findOneNVVMAnnotation(gv,
-                       llvm::PropertyAnnotationNames[llvm::PROPERTY_ISTEXTURE],
-                                   annot)) {
+    if (llvm::findOneNVVMAnnotation(
+            gv, llvm::PropertyAnnotationNames[llvm::PROPERTY_ISTEXTURE],
+            annot)) {
       assert((annot == 1) && "Unexpected annotation on a texture symbol");
       return true;
     }
@@ -133,9 +132,9 @@ bool llvm::isTexture(const llvm::Value &val) {
 bool llvm::isSurface(const llvm::Value &val) {
   if (const GlobalValue *gv = dyn_cast<GlobalValue>(&val)) {
     unsigned annot;
-    if (llvm::findOneNVVMAnnotation(gv,
-                       llvm::PropertyAnnotationNames[llvm::PROPERTY_ISSURFACE],
-                                   annot)) {
+    if (llvm::findOneNVVMAnnotation(
+            gv, llvm::PropertyAnnotationNames[llvm::PROPERTY_ISSURFACE],
+            annot)) {
       assert((annot == 1) && "Unexpected annotation on a surface symbol");
       return true;
     }
@@ -146,9 +145,9 @@ bool llvm::isSurface(const llvm::Value &val) {
 bool llvm::isSampler(const llvm::Value &val) {
   if (const GlobalValue *gv = dyn_cast<GlobalValue>(&val)) {
     unsigned annot;
-    if (llvm::findOneNVVMAnnotation(gv,
-                       llvm::PropertyAnnotationNames[llvm::PROPERTY_ISSAMPLER],
-                                   annot)) {
+    if (llvm::findOneNVVMAnnotation(
+            gv, llvm::PropertyAnnotationNames[llvm::PROPERTY_ISSAMPLER],
+            annot)) {
       assert((annot == 1) && "Unexpected annotation on a sampler symbol");
       return true;
     }
@@ -156,9 +155,9 @@ bool llvm::isSampler(const llvm::Value &val) {
   if (const Argument *arg = dyn_cast<Argument>(&val)) {
     const Function *func = arg->getParent();
     std::vector<unsigned> annot;
-    if (llvm::findAllNVVMAnnotation(func,
-                       llvm::PropertyAnnotationNames[llvm::PROPERTY_ISSAMPLER],
-                                   annot)) {
+    if (llvm::findAllNVVMAnnotation(
+            func, llvm::PropertyAnnotationNames[llvm::PROPERTY_ISSAMPLER],
+            annot)) {
       if (std::find(annot.begin(), annot.end(), arg->getArgNo()) != annot.end())
         return true;
     }
@@ -171,8 +170,9 @@ bool llvm::isImageReadOnly(const llvm::Value &val) {
     const Function *func = arg->getParent();
     std::vector<unsigned> annot;
     if (llvm::findAllNVVMAnnotation(func,
-          llvm::PropertyAnnotationNames[llvm::PROPERTY_ISREADONLY_IMAGE_PARAM],
-                                   annot)) {
+                                    llvm::PropertyAnnotationNames[
+                                        llvm::PROPERTY_ISREADONLY_IMAGE_PARAM],
+                                    annot)) {
       if (std::find(annot.begin(), annot.end(), arg->getArgNo()) != annot.end())
         return true;
     }
@@ -185,8 +185,9 @@ bool llvm::isImageWriteOnly(const llvm::Value &val) {
     const Function *func = arg->getParent();
     std::vector<unsigned> annot;
     if (llvm::findAllNVVMAnnotation(func,
-         llvm::PropertyAnnotationNames[llvm::PROPERTY_ISWRITEONLY_IMAGE_PARAM],
-                                   annot)) {
+                                    llvm::PropertyAnnotationNames[
+                                        llvm::PROPERTY_ISWRITEONLY_IMAGE_PARAM],
+                                    annot)) {
       if (std::find(annot.begin(), annot.end(), arg->getArgNo()) != annot.end())
         return true;
     }
@@ -214,52 +215,44 @@ std::string llvm::getSamplerName(const llvm::Value &val) {
 }
 
 bool llvm::getMaxNTIDx(const Function &F, unsigned &x) {
-  return (llvm::findOneNVVMAnnotation(&F,
-                       llvm::PropertyAnnotationNames[llvm::PROPERTY_MAXNTID_X],
-                                      x));
+  return (llvm::findOneNVVMAnnotation(
+      &F, llvm::PropertyAnnotationNames[llvm::PROPERTY_MAXNTID_X], x));
 }
 
 bool llvm::getMaxNTIDy(const Function &F, unsigned &y) {
-  return (llvm::findOneNVVMAnnotation(&F,
-                       llvm::PropertyAnnotationNames[llvm::PROPERTY_MAXNTID_Y],
-                                      y));
+  return (llvm::findOneNVVMAnnotation(
+      &F, llvm::PropertyAnnotationNames[llvm::PROPERTY_MAXNTID_Y], y));
 }
 
 bool llvm::getMaxNTIDz(const Function &F, unsigned &z) {
-  return (llvm::findOneNVVMAnnotation(&F,
-                       llvm::PropertyAnnotationNames[llvm::PROPERTY_MAXNTID_Z],
-                                      z));
+  return (llvm::findOneNVVMAnnotation(
+      &F, llvm::PropertyAnnotationNames[llvm::PROPERTY_MAXNTID_Z], z));
 }
 
 bool llvm::getReqNTIDx(const Function &F, unsigned &x) {
-  return (llvm::findOneNVVMAnnotation(&F,
-                       llvm::PropertyAnnotationNames[llvm::PROPERTY_REQNTID_X],
-                                      x));
+  return (llvm::findOneNVVMAnnotation(
+      &F, llvm::PropertyAnnotationNames[llvm::PROPERTY_REQNTID_X], x));
 }
 
 bool llvm::getReqNTIDy(const Function &F, unsigned &y) {
-  return (llvm::findOneNVVMAnnotation(&F,
-                       llvm::PropertyAnnotationNames[llvm::PROPERTY_REQNTID_Y],
-                                      y));
+  return (llvm::findOneNVVMAnnotation(
+      &F, llvm::PropertyAnnotationNames[llvm::PROPERTY_REQNTID_Y], y));
 }
 
 bool llvm::getReqNTIDz(const Function &F, unsigned &z) {
-  return (llvm::findOneNVVMAnnotation(&F,
-                       llvm::PropertyAnnotationNames[llvm::PROPERTY_REQNTID_Z],
-                                      z));
+  return (llvm::findOneNVVMAnnotation(
+      &F, llvm::PropertyAnnotationNames[llvm::PROPERTY_REQNTID_Z], z));
 }
 
 bool llvm::getMinCTASm(const Function &F, unsigned &x) {
-  return (llvm::findOneNVVMAnnotation(&F,
-                    llvm::PropertyAnnotationNames[llvm::PROPERTY_MINNCTAPERSM],
-                                      x));
+  return (llvm::findOneNVVMAnnotation(
+      &F, llvm::PropertyAnnotationNames[llvm::PROPERTY_MINNCTAPERSM], x));
 }
 
 bool llvm::isKernelFunction(const Function &F) {
   unsigned x = 0;
-  bool retval = llvm::findOneNVVMAnnotation(&F,
-               llvm::PropertyAnnotationNames[llvm::PROPERTY_ISKERNEL_FUNCTION],
-                                            x);
+  bool retval = llvm::findOneNVVMAnnotation(
+      &F, llvm::PropertyAnnotationNames[llvm::PROPERTY_ISKERNEL_FUNCTION], x);
   if (retval == false) {
     // There is no NVVM metadata, check the calling convention
     if (F.getCallingConv() == llvm::CallingConv::PTX_Kernel)
@@ -267,20 +260,19 @@ bool llvm::isKernelFunction(const Function &F) {
     else
       return false;
   }
-  return (x==1);
+  return (x == 1);
 }
 
 bool llvm::getAlign(const Function &F, unsigned index, unsigned &align) {
   std::vector<unsigned> Vs;
-  bool retval = llvm::findAllNVVMAnnotation(&F,
-                           llvm::PropertyAnnotationNames[llvm::PROPERTY_ALIGN],
-                                            Vs);
+  bool retval = llvm::findAllNVVMAnnotation(
+      &F, llvm::PropertyAnnotationNames[llvm::PROPERTY_ALIGN], Vs);
   if (retval == false)
     return false;
-  for (int i=0, e=Vs.size(); i<e; i++) {
+  for (int i = 0, e = Vs.size(); i < e; i++) {
     unsigned v = Vs[i];
-    if ( (v >> 16) == index ) {
-      align =  v & 0xFFFF;
+    if ((v >> 16) == index) {
+      align = v & 0xFFFF;
       return true;
     }
   }
@@ -289,16 +281,15 @@ bool llvm::getAlign(const Function &F, unsigned index, unsigned &align) {
 
 bool llvm::getAlign(const CallInst &I, unsigned index, unsigned &align) {
   if (MDNode *alignNode = I.getMetadata("callalign")) {
-    for (int i=0, n = alignNode->getNumOperands();
-        i<n; i++) {
+    for (int i = 0, n = alignNode->getNumOperands(); i < n; i++) {
       if (const ConstantInt *CI =
-          dyn_cast<ConstantInt>(alignNode->getOperand(i))) {
+              dyn_cast<ConstantInt>(alignNode->getOperand(i))) {
         unsigned v = CI->getZExtValue();
-        if ( (v>>16) == index ) {
+        if ((v >> 16) == index) {
           align = v & 0xFFFF;
           return true;
         }
-        if ( (v>>16) > index ) {
+        if ((v >> 16) > index) {
           return false;
         }
       }
@@ -337,8 +328,8 @@ bool llvm::isMemorySpaceTransferIntrinsic(Intrinsic::ID id) {
 // consider several special intrinsics in striping pointer casts, and
 // provide an option to ignore GEP indicies for find out the base address only
 // which could be used in simple alias disambigurate.
-const Value *llvm::skipPointerTransfer(const Value *V,
-                                       bool ignore_GEP_indices) {
+const Value *
+llvm::skipPointerTransfer(const Value *V, bool ignore_GEP_indices) {
   V = V->stripPointerCasts();
   while (true) {
     if (const IntrinsicInst *IS = dyn_cast<IntrinsicInst>(V)) {
@@ -360,8 +351,8 @@ const Value *llvm::skipPointerTransfer(const Value *V,
 // - ignore GEP indicies for find out the base address only, and
 // - tracking PHINode
 // which could be used in simple alias disambigurate.
-const Value *llvm::skipPointerTransfer(const Value *V,
-                                       std::set<const Value *> &processed) {
+const Value *
+llvm::skipPointerTransfer(const Value *V, std::set<const Value *> &processed) {
   if (processed.find(V) != processed.end())
     return NULL;
   processed.insert(V);
@@ -406,7 +397,6 @@ const Value *llvm::skipPointerTransfer(const Value *V,
   return V;
 }
 
-
 // The following are some useful utilities for debuggung
 
 BasicBlock *llvm::getParentBlock(Value *v) {
diff --git a/lib/Target/NVPTX/NVPTXUtilities.h b/lib/Target/NVPTX/NVPTXUtilities.h
index fe6ad559e9df..a208004297d0 100644
--- a/lib/Target/NVPTX/NVPTXUtilities.h
+++ b/lib/Target/NVPTX/NVPTXUtilities.h
@@ -14,17 +14,16 @@
 #ifndef NVPTXUTILITIES_H
 #define NVPTXUTILITIES_H
 
-#include "llvm/Value.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/Function.h"
-#include "llvm/IntrinsicInst.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Value.h"
 #include <cstdarg>
 #include <set>
 #include <string>
 #include <vector>
 
-namespace llvm
-{
+namespace llvm {
 
 #define NVCL_IMAGE2D_READONLY_FUNCNAME "__is_image2D_readonly"
 #define NVCL_IMAGE3D_READONLY_FUNCNAME "__is_image3D_readonly"
@@ -64,8 +63,7 @@ bool isBarrierIntrinsic(llvm::Intrinsic::ID);
 /// to pass into type construction of CallInst ctors.  This turns a null
 /// terminated list of pointers (or other value types) into a real live vector.
 ///
-template<typename T>
-inline std::vector<T> make_vector(T A, ...) {
+template <typename T> inline std::vector<T> make_vector(T A, ...) {
   va_list Args;
   va_start(Args, A);
   std::vector<T> Result;
@@ -78,8 +76,8 @@ inline std::vector<T> make_vector(T A, ...) {
 
 bool isMemorySpaceTransferIntrinsic(Intrinsic::ID id);
 const Value *skipPointerTransfer(const Value *V, bool ignore_GEP_indices);
-const Value *skipPointerTransfer(const Value *V,
-                                 std::set<const Value *> &processed);
+const Value *
+skipPointerTransfer(const Value *V, std::set<const Value *> &processed);
 BasicBlock *getParentBlock(Value *v);
 Function *getParentFunction(Value *v);
 void dumpBlock(Value *v, char *blockName);
diff --git a/lib/Target/NVPTX/NVPTXutil.cpp b/lib/Target/NVPTX/NVPTXutil.cpp
index 6a0e5328f62f..5f074b33a2d4 100644
--- a/lib/Target/NVPTX/NVPTXutil.cpp
+++ b/lib/Target/NVPTX/NVPTXutil.cpp
@@ -18,8 +18,7 @@ using namespace llvm;
 
 namespace llvm {
 
-bool isParamLoad(const MachineInstr *MI)
-{
+bool isParamLoad(const MachineInstr *MI) {
   if ((MI->getOpcode() != NVPTX::LD_i32_avar) &&
       (MI->getOpcode() != NVPTX::LD_i64_avar))
     return false;
@@ -30,13 +29,11 @@ bool isParamLoad(const MachineInstr *MI)
   return true;
 }
 
-#define DATA_MASK     0x7f
-#define DIGIT_WIDTH   7
-#define MORE_BYTES    0x80
+#define DATA_MASK 0x7f
+#define DIGIT_WIDTH 7
+#define MORE_BYTES 0x80
 
-static int encode_leb128(uint64_t val, int *nbytes,
-                         char *space, int splen)
-{
+static int encode_leb128(uint64_t val, int *nbytes, char *space, int splen) {
   char *a;
   char *end = space + splen;
 
@@ -61,29 +58,30 @@ static int encode_leb128(uint64_t val, int *nbytes,
 #undef DIGIT_WIDTH
 #undef MORE_BYTES
 
-uint64_t encode_leb128(const char *str)
-{
-  union { uint64_t x; char a[8]; } temp64;
+uint64_t encode_leb128(const char *str) {
+  union {
+    uint64_t x;
+    char a[8];
+  } temp64;
 
   temp64.x = 0;
 
-  for (unsigned i=0,e=strlen(str); i!=e; ++i)
-    temp64.a[i] = str[e-1-i];
+  for (unsigned i = 0, e = strlen(str); i != e; ++i)
+    temp64.a[i] = str[e - 1 - i];
 
   char encoded[16];
   int nbytes;
 
   int retval = encode_leb128(temp64.x, &nbytes, encoded, 16);
 
-  (void)retval;
-  assert(retval == 0 &&
-         "Encoding to leb128 failed");
+  (void) retval;
+  assert(retval == 0 && "Encoding to leb128 failed");
 
   assert(nbytes <= 8 &&
          "Cannot support register names with leb128 encoding > 8 bytes");
 
   temp64.x = 0;
-  for (int i=0; i<nbytes; ++i)
+  for (int i = 0; i < nbytes; ++i)
     temp64.a[i] = encoded[i];
 
   return temp64.x;
diff --git a/lib/Target/NVPTX/NVVMReflect.cpp b/lib/Target/NVPTX/NVVMReflect.cpp
new file mode 100644
index 000000000000..0ad62ce39b0d
--- /dev/null
+++ b/lib/Target/NVPTX/NVVMReflect.cpp
@@ -0,0 +1,177 @@
+//===- NVVMReflect.cpp - NVVM Emulate conditional compilation -------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass replaces occurences of __nvvm_reflect("string") with an
+// integer based on -nvvm-reflect-list string=<int> option given to this pass.
+// If an undefined string value is seen in a call to __nvvm_reflect("string"),
+// a default value of 0 will be used.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/StringMap.h"
+#include "llvm/Pass.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Type.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_os_ostream.h"
+#include "llvm/Transforms/Scalar.h"
+#include <map>
+#include <sstream>
+#include <string>
+#include <vector>
+
+#define NVVM_REFLECT_FUNCTION "__nvvm_reflect"
+
+using namespace llvm;
+
+namespace llvm { void initializeNVVMReflectPass(PassRegistry &); }
+
+namespace {
+class LLVM_LIBRARY_VISIBILITY NVVMReflect : public ModulePass {
+private:
+  StringMap<int> VarMap;
+  typedef DenseMap<std::string, int>::iterator VarMapIter;
+  Function *ReflectFunction;
+
+public:
+  static char ID;
+  NVVMReflect() : ModulePass(ID) {
+    VarMap.clear();
+    ReflectFunction = 0;
+  }
+
+  void getAnalysisUsage(AnalysisUsage &AU) const { AU.setPreservesAll(); }
+  virtual bool runOnModule(Module &);
+
+  void setVarMap();
+};
+}
+
+static cl::opt<bool>
+NVVMReflectEnabled("nvvm-reflect-enable", cl::init(true),
+                   cl::desc("NVVM reflection, enabled by default"));
+
+char NVVMReflect::ID = 0;
+INITIALIZE_PASS(NVVMReflect, "nvvm-reflect",
+                "Replace occurences of __nvvm_reflect() calls with 0/1", false,
+                false)
+
+static cl::list<std::string>
+ReflectList("nvvm-reflect-list", cl::value_desc("name=<int>"),
+            cl::desc("A list of string=num assignments"),
+            cl::ValueRequired);
+
+/// The command line can look as follows :
+/// -nvvm-reflect-list a=1,b=2 -nvvm-reflect-list c=3,d=0 -R e=2
+/// The strings "a=1,b=2", "c=3,d=0", "e=2" are available in the
+/// ReflectList vector. First, each of ReflectList[i] is 'split'
+/// using "," as the delimiter. Then each of this part is split
+/// using "=" as the delimiter.
+void NVVMReflect::setVarMap() {
+  for (unsigned i = 0, e = ReflectList.size(); i != e; ++i) {
+    DEBUG(dbgs() << "Option : "  << ReflectList[i] << "\n");
+    SmallVector<StringRef, 4> NameValList;
+    StringRef(ReflectList[i]).split(NameValList, ",");
+    for (unsigned j = 0, ej = NameValList.size(); j != ej; ++j) {
+      SmallVector<StringRef, 2> NameValPair;
+      NameValList[j].split(NameValPair, "=");
+      assert(NameValPair.size() == 2 && "name=val expected");
+      std::stringstream ValStream(NameValPair[1]);
+      int Val;
+      ValStream >> Val;
+      assert((!(ValStream.fail())) && "integer value expected");
+      VarMap[NameValPair[0]] = Val;
+    }
+  }
+}
+
+bool NVVMReflect::runOnModule(Module &M) {
+  if (!NVVMReflectEnabled)
+    return false;
+
+  setVarMap();
+
+  ReflectFunction = M.getFunction(NVVM_REFLECT_FUNCTION);
+
+  // If reflect function is not used, then there will be
+  // no entry in the module.
+  if (ReflectFunction == 0)
+    return false;
+
+  // Validate _reflect function
+  assert(ReflectFunction->isDeclaration() &&
+         "_reflect function should not have a body");
+  assert(ReflectFunction->getReturnType()->isIntegerTy() &&
+         "_reflect's return type should be integer");
+
+  std::vector<Instruction *> ToRemove;
+
+  // Go through the uses of ReflectFunction in this Function.
+  // Each of them should a CallInst with a ConstantArray argument.
+  // First validate that. If the c-string corresponding to the
+  // ConstantArray can be found successfully, see if it can be
+  // found in VarMap. If so, replace the uses of CallInst with the
+  // value found in VarMap. If not, replace the use  with value 0.
+  for (Value::use_iterator I = ReflectFunction->use_begin(),
+                           E = ReflectFunction->use_end();
+       I != E; ++I) {
+    assert(isa<CallInst>(*I) && "Only a call instruction can use _reflect");
+    CallInst *Reflect = cast<CallInst>(*I);
+
+    assert((Reflect->getNumOperands() == 2) &&
+           "Only one operand expect for _reflect function");
+    // In cuda, we will have an extra constant-to-generic conversion of
+    // the string.
+    const Value *conv = Reflect->getArgOperand(0);
+    assert(isa<CallInst>(conv) && "Expected a const-to-gen conversion");
+    const CallInst *ConvCall = cast<CallInst>(conv);
+    const Value *str = ConvCall->getArgOperand(0);
+    assert(isa<ConstantExpr>(str) &&
+           "Format of _reflect function not recognized");
+    const ConstantExpr *GEP = cast<ConstantExpr>(str);
+
+    const Value *Sym = GEP->getOperand(0);
+    assert(isa<Constant>(Sym) && "Format of _reflect function not recognized");
+
+    const Constant *SymStr = cast<Constant>(Sym);
+
+    assert(isa<ConstantDataSequential>(SymStr->getOperand(0)) &&
+           "Format of _reflect function not recognized");
+
+    assert(cast<ConstantDataSequential>(SymStr->getOperand(0))->isCString() &&
+           "Format of _reflect function not recognized");
+
+    std::string ReflectArg =
+        cast<ConstantDataSequential>(SymStr->getOperand(0))->getAsString();
+
+    ReflectArg = ReflectArg.substr(0, ReflectArg.size() - 1);
+    DEBUG(dbgs() << "Arg of _reflect : " << ReflectArg << "\n");
+
+    int ReflectVal = 0; // The default value is 0
+    if (VarMap.find(ReflectArg) != VarMap.end()) {
+      ReflectVal = VarMap[ReflectArg];
+    }
+    Reflect->replaceAllUsesWith(
+        ConstantInt::get(Reflect->getType(), ReflectVal));
+    ToRemove.push_back(Reflect);
+  }
+  if (ToRemove.size() == 0)
+    return false;
+
+  for (unsigned i = 0, e = ToRemove.size(); i != e; ++i)
+    ToRemove[i]->eraseFromParent();
+  return true;
+}
diff --git a/lib/Target/NVPTX/TargetInfo/NVPTXTargetInfo.cpp b/lib/Target/NVPTX/TargetInfo/NVPTXTargetInfo.cpp
index f3624b9f23c7..cc7d4dc5ece7 100644
--- a/lib/Target/NVPTX/TargetInfo/NVPTXTargetInfo.cpp
+++ b/lib/Target/NVPTX/TargetInfo/NVPTXTargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "NVPTX.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
@@ -17,7 +17,7 @@ Target llvm::TheNVPTXTarget64;
 
 extern "C" void LLVMInitializeNVPTXTargetInfo() {
   RegisterTarget<Triple::nvptx> X(TheNVPTXTarget32, "nvptx",
-    "NVIDIA PTX 32-bit");
+                                  "NVIDIA PTX 32-bit");
   RegisterTarget<Triple::nvptx64> Y(TheNVPTXTarget64, "nvptx64",
-    "NVIDIA PTX 64-bit");
+                                    "NVIDIA PTX 64-bit");
 }
diff --git a/lib/Target/NVPTX/VectorElementize.cpp b/lib/Target/NVPTX/VectorElementize.cpp
deleted file mode 100644
index 8043e2de0972..000000000000
--- a/lib/Target/NVPTX/VectorElementize.cpp
+++ /dev/null
@@ -1,1248 +0,0 @@
-//===-- VectorElementize.cpp - Remove unreachable blocks for codegen --===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This pass converts operations on vector types to operations on their
-// element types.
-//
-// For generic binary and unary vector instructions, the conversion is simple.
-// Suppose we have
-//        av = bv Vop cv
-// where av, bv, and cv are vector virtual registers, and Vop is a vector op.
-// This gets converted to the following :
-//       a1 = b1 Sop c1
-//       a2 = b2 Sop c2
-//
-// VectorToScalarMap maintains the vector vreg to scalar vreg mapping.
-// For the above example, the map will look as follows:
-// av => [a1, a2]
-// bv => [b1, b2]
-//
-// In addition, initVectorInfo creates the following opcode->opcode map.
-// Vop => Sop
-// OtherVop => OtherSop
-// ...
-//
-// For vector specific instructions like vecbuild, vecshuffle etc, the
-// conversion is different. Look at comments near the functions with
-// prefix createVec<...>.
-//
-//===----------------------------------------------------------------------===//
-
-#include "llvm/CodeGen/Passes.h"
-#include "llvm/Constant.h"
-#include "llvm/Instructions.h"
-#include "llvm/Function.h"
-#include "llvm/Pass.h"
-#include "llvm/Type.h"
-#include "llvm/Support/CommandLine.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/Support/CFG.h"
-#include "llvm/Support/Compiler.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/ADT/DepthFirstIterator.h"
-#include "llvm/ADT/SmallPtrSet.h"
-#include "NVPTX.h"
-#include "NVPTXTargetMachine.h"
-
-using namespace llvm;
-
-namespace {
-
-class LLVM_LIBRARY_VISIBILITY VectorElementize : public MachineFunctionPass {
-  virtual bool runOnMachineFunction(MachineFunction &F);
-
-  NVPTXTargetMachine &TM;
-  MachineRegisterInfo *MRI;
-  const NVPTXRegisterInfo *RegInfo;
-  const NVPTXInstrInfo *InstrInfo;
-
-  llvm::DenseMap<const TargetRegisterClass *, const TargetRegisterClass *>
-  RegClassMap;
-  llvm::DenseMap<unsigned, bool> SimpleMoveMap;
-
-  llvm::DenseMap<unsigned, SmallVector<unsigned, 4> > VectorToScalarMap;
-
-  bool isVectorInstr(MachineInstr *);
-
-  SmallVector<unsigned, 4> getScalarRegisters(unsigned);
-  unsigned getScalarVersion(unsigned);
-  unsigned getScalarVersion(MachineInstr *);
-
-  bool isVectorRegister(unsigned);
-  const TargetRegisterClass *getScalarRegClass(const TargetRegisterClass *RC);
-  unsigned numCopiesNeeded(MachineInstr *);
-
-  void createLoadCopy(MachineFunction&, MachineInstr *,
-                      std::vector<MachineInstr *>&);
-  void createStoreCopy(MachineFunction&, MachineInstr *,
-                       std::vector<MachineInstr *>&);
-
-  void createVecDest(MachineFunction&, MachineInstr *,
-                     std::vector<MachineInstr *>&);
-
-  void createCopies(MachineFunction&, MachineInstr *,
-                    std::vector<MachineInstr *>&);
-
-  unsigned copyProp(MachineFunction&);
-  unsigned removeDeadMoves(MachineFunction&);
-
-  void elementize(MachineFunction&);
-
-  bool isSimpleMove(MachineInstr *);
-
-  void createVecShuffle(MachineFunction& F, MachineInstr *Instr,
-                        std::vector<MachineInstr *>& copies);
-
-  void createVecExtract(MachineFunction& F, MachineInstr *Instr,
-                        std::vector<MachineInstr *>& copies);
-
-  void createVecInsert(MachineFunction& F, MachineInstr *Instr,
-                       std::vector<MachineInstr *>& copies);
-
-  void createVecBuild(MachineFunction& F, MachineInstr *Instr,
-                      std::vector<MachineInstr *>& copies);
-
-public:
-
-  static char ID; // Pass identification, replacement for typeid
-  VectorElementize(NVPTXTargetMachine &tm)
-  : MachineFunctionPass(ID), TM(tm) {}
-
-  virtual const char *getPassName() const {
-    return "Convert LLVM vector types to their element types";
-  }
-};
-
-char VectorElementize::ID = 1;
-}
-
-static cl::opt<bool>
-RemoveRedundantMoves("nvptx-remove-redundant-moves",
-       cl::desc("NVPTX: Remove redundant moves introduced by vector lowering"),
-                     cl::init(true));
-
-#define VECINST(x) ((((x)->getDesc().TSFlags) & NVPTX::VecInstTypeMask) \
-    >> NVPTX::VecInstTypeShift)
-#define ISVECINST(x) (VECINST(x) != NVPTX::VecNOP)
-#define ISVECLOAD(x)    (VECINST(x) == NVPTX::VecLoad)
-#define ISVECSTORE(x)   (VECINST(x) == NVPTX::VecStore)
-#define ISVECBUILD(x)   (VECINST(x) == NVPTX::VecBuild)
-#define ISVECSHUFFLE(x) (VECINST(x) == NVPTX::VecShuffle)
-#define ISVECEXTRACT(x) (VECINST(x) == NVPTX::VecExtract)
-#define ISVECINSERT(x)  (VECINST(x) == NVPTX::VecInsert)
-#define ISVECDEST(x)     (VECINST(x) == NVPTX::VecDest)
-
-bool VectorElementize::isSimpleMove(MachineInstr *mi) {
-  if (mi->isCopy())
-    return true;
-  unsigned TSFlags = (mi->getDesc().TSFlags & NVPTX::SimpleMoveMask)
-        >> NVPTX::SimpleMoveShift;
-  return (TSFlags == 1);
-}
-
-bool VectorElementize::isVectorInstr(MachineInstr *mi) {
-  if ((mi->getOpcode() == NVPTX::PHI) ||
-      (mi->getOpcode() == NVPTX::IMPLICIT_DEF) || mi->isCopy()) {
-    MachineOperand dest = mi->getOperand(0);
-    return isVectorRegister(dest.getReg());
-  }
-  return ISVECINST(mi);
-}
-
-unsigned VectorElementize::getScalarVersion(MachineInstr *mi) {
-  return getScalarVersion(mi->getOpcode());
-}
-
-///=============================================================================
-///Instr is assumed to be a vector instruction. For most vector instructions,
-///the size of the destination vector register gives the number of scalar copies
-///needed. For VecStore, size of getOperand(1) gives the number of scalar copies
-///needed. For VecExtract, the dest is a scalar. So getOperand(1) gives the
-///number of scalar copies needed.
-///=============================================================================
-unsigned VectorElementize::numCopiesNeeded(MachineInstr *Instr) {
-  unsigned numDefs=0;
-  unsigned def;
-  for (unsigned i=0, e=Instr->getNumOperands(); i!=e; ++i) {
-    MachineOperand oper = Instr->getOperand(i);
-
-    if (!oper.isReg()) continue;
-    if (!oper.isDef()) continue;
-    def = i;
-    numDefs++;
-  }
-  assert((numDefs <= 1) && "Only 0 or 1 defs supported");
-
-  if (numDefs == 1) {
-    unsigned regnum = Instr->getOperand(def).getReg();
-    if (ISVECEXTRACT(Instr))
-      regnum = Instr->getOperand(1).getReg();
-    return getNVPTXVectorSize(MRI->getRegClass(regnum));
-  }
-  else if (numDefs == 0) {
-    assert(ISVECSTORE(Instr)
-           && "Only 0 def instruction supported is vector store");
-
-    unsigned regnum = Instr->getOperand(0).getReg();
-    return getNVPTXVectorSize(MRI->getRegClass(regnum));
-  }
-  return 1;
-}
-
-const TargetRegisterClass *VectorElementize::
-getScalarRegClass(const TargetRegisterClass *RC) {
-  assert(isNVPTXVectorRegClass(RC) &&
-         "Not a vector register class");
-  return getNVPTXElemClass(RC);
-}
-
-bool VectorElementize::isVectorRegister(unsigned reg) {
-  const TargetRegisterClass *RC=MRI->getRegClass(reg);
-  return isNVPTXVectorRegClass(RC);
-}
-
-///=============================================================================
-///For every vector register 'v' that is not already in the VectorToScalarMap,
-///create n scalar registers of the corresponding element type, where n
-///is 2 or 4 (getNVPTXVectorSize) and add it VectorToScalarMap.
-///=============================================================================
-SmallVector<unsigned, 4> VectorElementize::getScalarRegisters(unsigned regnum) {
-  assert(isVectorRegister(regnum) && "Expecting a vector register here");
-  // Create the scalar registers and put them in the map, if not already there.
-  if (VectorToScalarMap.find(regnum) == VectorToScalarMap.end()) {
-    const TargetRegisterClass *vecClass = MRI->getRegClass(regnum);
-    const TargetRegisterClass *scalarClass = getScalarRegClass(vecClass);
-
-    SmallVector<unsigned, 4> temp;
-
-    for (unsigned i=0, e=getNVPTXVectorSize(vecClass); i!=e; ++i)
-      temp.push_back(MRI->createVirtualRegister(scalarClass));
-
-    VectorToScalarMap[regnum] = temp;
-  }
-  return VectorToScalarMap[regnum];
-}
-
-///=============================================================================
-///For a vector load of the form
-///va <= ldv2 [addr]
-///the following multi output instruction is created :
-///[v1, v2] <= LD [addr]
-///Look at NVPTXVector.td for the definitions of multi output loads.
-///=============================================================================
-void VectorElementize::createLoadCopy(MachineFunction& F, MachineInstr *Instr,
-                                      std::vector<MachineInstr *>& copies) {
-  copies.push_back(F.CloneMachineInstr(Instr));
-
-  MachineInstr *copy=copies[0];
-  copy->setDesc(InstrInfo->get(getScalarVersion(copy)));
-
-  // Remove the dest, that should be a vector operand.
-  MachineOperand dest = copy->getOperand(0);
-  unsigned regnum = dest.getReg();
-
-  SmallVector<unsigned, 4> scalarRegs = getScalarRegisters(regnum);
-  copy->RemoveOperand(0);
-
-  std::vector<MachineOperand> otherOperands;
-  for (unsigned i=0, e=copy->getNumOperands(); i!=e; ++i)
-    otherOperands.push_back(copy->getOperand(i));
-
-  for (unsigned i=0, e=copy->getNumOperands(); i!=e; ++i)
-    copy->RemoveOperand(0);
-
-  for (unsigned i=0, e=scalarRegs.size(); i!=e; ++i) {
-    copy->addOperand(MachineOperand::CreateReg(scalarRegs[i], true));
-  }
-
-  for (unsigned i=0, e=otherOperands.size(); i!=e; ++i)
-    copy->addOperand(otherOperands[i]);
-
-}
-
-///=============================================================================
-///For a vector store of the form
-///stv2 va, [addr]
-///the following multi input instruction is created :
-///ST v1, v2, [addr]
-///Look at NVPTXVector.td for the definitions of multi input stores.
-///=============================================================================
-void VectorElementize::createStoreCopy(MachineFunction& F, MachineInstr *Instr,
-                                       std::vector<MachineInstr *>& copies) {
-  copies.push_back(F.CloneMachineInstr(Instr));
-
-  MachineInstr *copy=copies[0];
-  copy->setDesc(InstrInfo->get(getScalarVersion(copy)));
-
-  MachineOperand src = copy->getOperand(0);
-  unsigned regnum = src.getReg();
-
-  SmallVector<unsigned, 4> scalarRegs = getScalarRegisters(regnum);
-  copy->RemoveOperand(0);
-
-  std::vector<MachineOperand> otherOperands;
-  for (unsigned i=0, e=copy->getNumOperands(); i!=e; ++i)
-    otherOperands.push_back(copy->getOperand(i));
-
-  for (unsigned i=0, e=copy->getNumOperands(); i!=e; ++i)
-    copy->RemoveOperand(0);
-
-  for (unsigned i=0, e=scalarRegs.size(); i!=e; ++i)
-    copy->addOperand(MachineOperand::CreateReg(scalarRegs[i], false));
-
-  for (unsigned i=0, e=otherOperands.size(); i!=e; ++i)
-    copy->addOperand(otherOperands[i]);
-}
-
-///=============================================================================
-///va <= shufflev2 vb, vc, <i1>, <i2>
-///gets converted to 2 moves into a1 and a2. The source of the moves depend on
-///i1 and i2. i1, i2 can belong to the set {0, 1, 2, 3} for shufflev2. For
-///shufflev4 the set is {0,..7}. For example, if i1=3, i2=0, the move
-///instructions will be
-///a1 <= c2
-///a2 <= b1
-///=============================================================================
-void VectorElementize::createVecShuffle(MachineFunction& F, MachineInstr *Instr,
-                                        std::vector<MachineInstr *>& copies) {
-  unsigned numcopies=numCopiesNeeded(Instr);
-
-  unsigned destregnum = Instr->getOperand(0).getReg();
-  unsigned src1regnum = Instr->getOperand(1).getReg();
-  unsigned src2regnum = Instr->getOperand(2).getReg();
-
-  SmallVector<unsigned, 4> dest = getScalarRegisters(destregnum);
-  SmallVector<unsigned, 4> src1 = getScalarRegisters(src1regnum);
-  SmallVector<unsigned, 4> src2 = getScalarRegisters(src2regnum);
-
-  DebugLoc DL = Instr->getDebugLoc();
-
-  for (unsigned i=0; i<numcopies; i++) {
-    MachineInstr *copy = BuildMI(F, DL,
-                              InstrInfo->get(getScalarVersion(Instr)), dest[i]);
-    MachineOperand which=Instr->getOperand(3+i);
-    assert(which.isImm() && "Shuffle operand not a constant");
-
-    int src=which.getImm();
-    int elem=src%numcopies;
-
-    if (which.getImm() < numcopies)
-      copy->addOperand(MachineOperand::CreateReg(src1[elem], false));
-    else
-      copy->addOperand(MachineOperand::CreateReg(src2[elem], false));
-    copies.push_back(copy);
-  }
-}
-
-///=============================================================================
-///a <= extractv2 va, <i1>
-///gets turned into a simple move to the scalar register a. The source depends
-///on i1.
-///=============================================================================
-void VectorElementize::createVecExtract(MachineFunction& F, MachineInstr *Instr,
-                                        std::vector<MachineInstr *>& copies) {
-  unsigned srcregnum = Instr->getOperand(1).getReg();
-
-  SmallVector<unsigned, 4> src = getScalarRegisters(srcregnum);
-
-  MachineOperand which = Instr->getOperand(2);
-  assert(which.isImm() && "Extract operand not a constant");
-
-  DebugLoc DL = Instr->getDebugLoc();
-
-  MachineInstr *copy = BuildMI(F, DL, InstrInfo->get(getScalarVersion(Instr)),
-                               Instr->getOperand(0).getReg());
-  copy->addOperand(MachineOperand::CreateReg(src[which.getImm()], false));
-
-  copies.push_back(copy);
-}
-
-///=============================================================================
-///va <= vecinsertv2 vb, c, <i1>
-///This instruction copies all elements of vb to va, except the 'i1'th element.
-///The scalar value c becomes the 'i1'th element of va.
-///This gets translated to 2 (4 for vecinsertv4) moves.
-///=============================================================================
-void VectorElementize::createVecInsert(MachineFunction& F, MachineInstr *Instr,
-                                       std::vector<MachineInstr *>& copies) {
-  unsigned numcopies=numCopiesNeeded(Instr);
-
-  unsigned destregnum = Instr->getOperand(0).getReg();
-  unsigned srcregnum = Instr->getOperand(1).getReg();
-
-  SmallVector<unsigned, 4> dest = getScalarRegisters(destregnum);
-  SmallVector<unsigned, 4> src = getScalarRegisters(srcregnum);
-
-  MachineOperand which=Instr->getOperand(3);
-  assert(which.isImm() && "Insert operand not a constant");
-  unsigned int elem=which.getImm();
-
-  DebugLoc DL = Instr->getDebugLoc();
-
-  for (unsigned i=0; i<numcopies; i++) {
-    MachineInstr *copy = BuildMI(F, DL,
-                              InstrInfo->get(getScalarVersion(Instr)), dest[i]);
-
-    if (i != elem)
-      copy->addOperand(MachineOperand::CreateReg(src[i], false));
-    else
-      copy->addOperand(Instr->getOperand(2));
-
-    copies.push_back(copy);
-  }
-
-}
-
-///=============================================================================
-///va <= buildv2 b1, b2
-///gets translated to
-///a1 <= b1
-///a2 <= b2
-///=============================================================================
-void VectorElementize::createVecBuild(MachineFunction& F, MachineInstr *Instr,
-                                      std::vector<MachineInstr *>& copies) {
-  unsigned numcopies=numCopiesNeeded(Instr);
-
-  unsigned destregnum = Instr->getOperand(0).getReg();
-
-  SmallVector<unsigned, 4> dest = getScalarRegisters(destregnum);
-
-  DebugLoc DL = Instr->getDebugLoc();
-
-  for (unsigned i=0; i<numcopies; i++) {
-    MachineInstr *copy = BuildMI(F, DL,
-                              InstrInfo->get(getScalarVersion(Instr)), dest[i]);
-
-    copy->addOperand(Instr->getOperand(1+i));
-
-    copies.push_back(copy);
-  }
-
-}
-
-///=============================================================================
-///For a tex inst of the form
-///va <= op [scalar operands]
-///the following multi output instruction is created :
-///[v1, v2] <= op' [scalar operands]
-///=============================================================================
-void VectorElementize::createVecDest(MachineFunction& F, MachineInstr *Instr,
-                                     std::vector<MachineInstr *>& copies) {
-  copies.push_back(F.CloneMachineInstr(Instr));
-
-  MachineInstr *copy=copies[0];
-  copy->setDesc(InstrInfo->get(getScalarVersion(copy)));
-
-  // Remove the dest, that should be a vector operand.
-  MachineOperand dest = copy->getOperand(0);
-  unsigned regnum = dest.getReg();
-
-  SmallVector<unsigned, 4> scalarRegs = getScalarRegisters(regnum);
-  copy->RemoveOperand(0);
-
-  std::vector<MachineOperand> otherOperands;
-  for (unsigned i=0, e=copy->getNumOperands(); i!=e; ++i)
-    otherOperands.push_back(copy->getOperand(i));
-
-  for (unsigned i=0, e=copy->getNumOperands(); i!=e; ++i)
-    copy->RemoveOperand(0);
-
-  for (unsigned i=0, e=scalarRegs.size(); i!=e; ++i)
-    copy->addOperand(MachineOperand::CreateReg(scalarRegs[i], true));
-
-  for (unsigned i=0, e=otherOperands.size(); i!=e; ++i)
-    copy->addOperand(otherOperands[i]);
-}
-
-///=============================================================================
-///Look at the vector instruction type and dispatch to the createVec<...>
-///function that creates the scalar copies.
-///=============================================================================
-void VectorElementize::createCopies(MachineFunction& F, MachineInstr *Instr,
-                                    std::vector<MachineInstr *>& copies) {
-  if (ISVECLOAD(Instr)) {
-    createLoadCopy(F, Instr, copies);
-    return;
-  }
-  if (ISVECSTORE(Instr)) {
-    createStoreCopy(F, Instr, copies);
-    return;
-  }
-  if (ISVECSHUFFLE(Instr)) {
-    createVecShuffle(F, Instr, copies);
-    return;
-  }
-  if (ISVECEXTRACT(Instr)) {
-    createVecExtract(F, Instr, copies);
-    return;
-  }
-  if (ISVECINSERT(Instr)) {
-    createVecInsert(F, Instr, copies);
-    return;
-  }
-  if (ISVECDEST(Instr)) {
-    createVecDest(F, Instr, copies);
-    return;
-  }
-  if (ISVECBUILD(Instr)) {
-    createVecBuild(F, Instr, copies);
-    return;
-  }
-
-  unsigned numcopies=numCopiesNeeded(Instr);
-
-  for (unsigned i=0; i<numcopies; ++i)
-    copies.push_back(F.CloneMachineInstr(Instr));
-
-  for (unsigned i=0; i<numcopies; ++i) {
-    MachineInstr *copy = copies[i];
-
-    std::vector<MachineOperand> allOperands;
-    std::vector<bool> isDef;
-
-    for (unsigned j=0, e=copy->getNumOperands(); j!=e; ++j) {
-      MachineOperand oper = copy->getOperand(j);
-      allOperands.push_back(oper);
-      if (oper.isReg())
-        isDef.push_back(oper.isDef());
-      else
-        isDef.push_back(false);
-    }
-
-    for (unsigned j=0, e=copy->getNumOperands(); j!=e; ++j)
-      copy->RemoveOperand(0);
-
-    copy->setDesc(InstrInfo->get(getScalarVersion(Instr)));
-
-    for (unsigned j=0, e=allOperands.size(); j!=e; ++j) {
-      MachineOperand oper=allOperands[j];
-      if (oper.isReg()) {
-        unsigned regnum = oper.getReg();
-        if (isVectorRegister(regnum)) {
-
-          SmallVector<unsigned, 4> scalarRegs = getScalarRegisters(regnum);
-          copy->addOperand(MachineOperand::CreateReg(scalarRegs[i], isDef[j]));
-        }
-        else
-          copy->addOperand(oper);
-      }
-      else
-        copy->addOperand(oper);
-    }
-  }
-}
-
-///=============================================================================
-///Scan through all basic blocks, looking for vector instructions.
-///For each vector instruction I, insert the scalar copies before I, and
-///add I into toRemove vector. Finally remove all instructions in toRemove.
-///=============================================================================
-void VectorElementize::elementize(MachineFunction &F) {
-  for (MachineFunction::reverse_iterator BI=F.rbegin(), BE=F.rend();
-      BI!=BE; ++BI) {
-    MachineBasicBlock *BB = &*BI;
-
-    std::vector<MachineInstr *> copies;
-    std::vector<MachineInstr *> toRemove;
-
-    for (MachineBasicBlock::iterator II=BB->begin(), IE=BB->end();
-        II!=IE; ++II) {
-      MachineInstr *Instr = &*II;
-
-      if (!isVectorInstr(Instr))
-        continue;
-
-      copies.clear();
-      createCopies(F, Instr, copies);
-      for (unsigned i=0, e=copies.size(); i!=e; ++i)
-        BB->insert(II, copies[i]);
-
-      assert((copies.size() > 0) && "Problem in createCopies");
-      toRemove.push_back(Instr);
-    }
-    for (unsigned i=0, e=toRemove.size(); i!=e; ++i)
-      F.DeleteMachineInstr(toRemove[i]->getParent()->remove(toRemove[i]));
-  }
-}
-
-///=============================================================================
-///a <= b
-///...
-///...
-///x <= op(a, ...)
-///gets converted to
-///
-///x <= op(b, ...)
-///The original move is still present. This works on SSA form machine code.
-///Note that a <= b should be a simple vreg-to-vreg move instruction.
-///TBD : I didn't find a function that can do replaceOperand, so I remove
-///all operands and add all of them again, replacing the one while adding.
-///=============================================================================
-unsigned VectorElementize::copyProp(MachineFunction &F) {
-  unsigned numReplacements = 0;
-
-  for (MachineFunction::reverse_iterator BI=F.rbegin(), BE=F.rend(); BI!=BE;
-      ++BI) {
-    MachineBasicBlock *BB = &*BI;
-
-    for (MachineBasicBlock::iterator II=BB->begin(), IE=BB->end(); II!=IE;
-        ++II) {
-      MachineInstr *Instr = &*II;
-
-      // Don't do copy propagation on PHI as it will cause unnecessary
-      // live range overlap.
-      if ((Instr->getOpcode() == TargetOpcode::PHI) ||
-          (Instr->getOpcode() == TargetOpcode::DBG_VALUE))
-        continue;
-
-      bool needsReplacement = false;
-
-      for (unsigned i=0, e=Instr->getNumOperands(); i!=e; ++i) {
-        MachineOperand oper = Instr->getOperand(i);
-        if (!oper.isReg()) continue;
-        if (oper.isDef()) continue;
-        if (!RegInfo->isVirtualRegister(oper.getReg())) continue;
-
-        MachineInstr *defInstr = MRI->getVRegDef(oper.getReg());
-
-        if (!defInstr) continue;
-
-        if (!isSimpleMove(defInstr)) continue;
-
-        MachineOperand defSrc = defInstr->getOperand(1);
-        if (!defSrc.isReg()) continue;
-        if (!RegInfo->isVirtualRegister(defSrc.getReg())) continue;
-
-        needsReplacement = true;
-
-      }
-      if (!needsReplacement) continue;
-
-      numReplacements++;
-
-      std::vector<MachineOperand> operands;
-
-      for (unsigned i=0, e=Instr->getNumOperands(); i!=e; ++i) {
-        MachineOperand oper = Instr->getOperand(i);
-        bool flag = false;
-        do {
-          if (!(oper.isReg()))
-            break;
-          if (oper.isDef())
-            break;
-          if (!(RegInfo->isVirtualRegister(oper.getReg())))
-            break;
-          MachineInstr *defInstr = MRI->getVRegDef(oper.getReg());
-          if (!(isSimpleMove(defInstr)))
-            break;
-          MachineOperand defSrc = defInstr->getOperand(1);
-          if (!(defSrc.isReg()))
-            break;
-          if (!(RegInfo->isVirtualRegister(defSrc.getReg())))
-            break;
-          operands.push_back(defSrc);
-          flag = true;
-        } while (0);
-        if (flag == false)
-          operands.push_back(oper);
-      }
-
-      for (unsigned i=0, e=Instr->getNumOperands(); i!=e; ++i)
-        Instr->RemoveOperand(0);
-      for (unsigned i=0, e=operands.size(); i!=e; ++i)
-        Instr->addOperand(operands[i]);
-
-    }
-  }
-  return numReplacements;
-}
-
-///=============================================================================
-///Look for simple vreg-to-vreg instructions whose use_empty() is true, add
-///them to deadMoves vector. Then remove all instructions in deadMoves.
-///=============================================================================
-unsigned VectorElementize::removeDeadMoves(MachineFunction &F) {
-  std::vector<MachineInstr *> deadMoves;
-  for (MachineFunction::reverse_iterator BI=F.rbegin(), BE=F.rend(); BI!=BE;
-      ++BI) {
-    MachineBasicBlock *BB = &*BI;
-
-    for (MachineBasicBlock::iterator II=BB->begin(), IE=BB->end(); II!=IE;
-        ++II) {
-      MachineInstr *Instr = &*II;
-
-      if (!isSimpleMove(Instr)) continue;
-
-      MachineOperand dest = Instr->getOperand(0);
-      assert(dest.isReg() && "dest of move not a register");
-      assert(RegInfo->isVirtualRegister(dest.getReg()) &&
-             "dest of move not a virtual register");
-
-      if (MRI->use_empty(dest.getReg())) {
-        deadMoves.push_back(Instr);
-      }
-    }
-  }
-
-  for (unsigned i=0, e=deadMoves.size(); i!=e; ++i)
-    F.DeleteMachineInstr(deadMoves[i]->getParent()->remove(deadMoves[i]));
-
-  return deadMoves.size();
-}
-
-///=============================================================================
-///Main function for this pass.
-///=============================================================================
-bool VectorElementize::runOnMachineFunction(MachineFunction &F) {
-  MRI = &F.getRegInfo();
-
-  RegInfo = TM.getRegisterInfo();
-  InstrInfo = TM.getInstrInfo();
-
-  VectorToScalarMap.clear();
-
-  elementize(F);
-
-  if (RemoveRedundantMoves)
-    while (1) {
-      if (copyProp(F) == 0) break;
-      removeDeadMoves(F);
-    }
-
-  return true;
-}
-
-FunctionPass *llvm::createVectorElementizePass(NVPTXTargetMachine &tm) {
-  return new VectorElementize(tm);
-}
-
-unsigned VectorElementize::getScalarVersion(unsigned opcode) {
-  if (opcode == NVPTX::PHI)
-    return opcode;
-  if (opcode == NVPTX::IMPLICIT_DEF)
-    return opcode;
-  switch(opcode) {
-  default: llvm_unreachable("Scalar version not set, fix NVPTXVector.td");
-  case TargetOpcode::COPY: return TargetOpcode::COPY;
-  case NVPTX::AddCCCV2I32: return NVPTX::ADDCCCi32rr;
-  case NVPTX::AddCCCV4I32: return NVPTX::ADDCCCi32rr;
-  case NVPTX::AddCCV2I32: return NVPTX::ADDCCi32rr;
-  case NVPTX::AddCCV4I32: return NVPTX::ADDCCi32rr;
-  case NVPTX::Build_Vector2_f32: return NVPTX::FMOV32rr;
-  case NVPTX::Build_Vector2_f64: return NVPTX::FMOV64rr;
-  case NVPTX::Build_Vector2_i16: return NVPTX::IMOV16rr;
-  case NVPTX::Build_Vector2_i32: return NVPTX::IMOV32rr;
-  case NVPTX::Build_Vector2_i64: return NVPTX::IMOV64rr;
-  case NVPTX::Build_Vector2_i8: return NVPTX::IMOV8rr;
-  case NVPTX::Build_Vector4_f32: return NVPTX::FMOV32rr;
-  case NVPTX::Build_Vector4_i16: return NVPTX::IMOV16rr;
-  case NVPTX::Build_Vector4_i32: return NVPTX::IMOV32rr;
-  case NVPTX::Build_Vector4_i8: return NVPTX::IMOV8rr;
-  case NVPTX::CVTv2i16tov2i32: return NVPTX::Zint_extendext16to32;
-  case NVPTX::CVTv2i64tov2i32: return NVPTX::TRUNC_64to32;
-  case NVPTX::CVTv2i8tov2i32: return NVPTX::Zint_extendext8to32;
-  case NVPTX::CVTv4i16tov4i32: return NVPTX::Zint_extendext16to32;
-  case NVPTX::CVTv4i8tov4i32: return NVPTX::Zint_extendext8to32;
-  case NVPTX::F32MAD_ftzV2: return NVPTX::FMAD32_ftzrrr;
-  case NVPTX::F32MADV2: return NVPTX::FMAD32rrr;
-  case NVPTX::F32MAD_ftzV4: return NVPTX::FMAD32_ftzrrr;
-  case NVPTX::F32MADV4: return NVPTX::FMAD32rrr;
-  case NVPTX::F32FMA_ftzV2: return NVPTX::FMA32_ftzrrr;
-  case NVPTX::F32FMAV2: return NVPTX::FMA32rrr;
-  case NVPTX::F32FMA_ftzV4: return NVPTX::FMA32_ftzrrr;
-  case NVPTX::F32FMAV4: return NVPTX::FMA32rrr;
-  case NVPTX::F64FMAV2: return NVPTX::FMA64rrr;
-  case NVPTX::FVecEQV2F32: return NVPTX::FSetEQf32rr_toi32;
-  case NVPTX::FVecEQV2F64: return NVPTX::FSetEQf64rr_toi64;
-  case NVPTX::FVecEQV4F32: return NVPTX::FSetEQf32rr_toi32;
-  case NVPTX::FVecGEV2F32: return NVPTX::FSetGEf32rr_toi32;
-  case NVPTX::FVecGEV2F64: return NVPTX::FSetGEf64rr_toi64;
-  case NVPTX::FVecGEV4F32: return NVPTX::FSetGEf32rr_toi32;
-  case NVPTX::FVecGTV2F32: return NVPTX::FSetGTf32rr_toi32;
-  case NVPTX::FVecGTV2F64: return NVPTX::FSetGTf64rr_toi64;
-  case NVPTX::FVecGTV4F32: return NVPTX::FSetGTf32rr_toi32;
-  case NVPTX::FVecLEV2F32: return NVPTX::FSetLEf32rr_toi32;
-  case NVPTX::FVecLEV2F64: return NVPTX::FSetLEf64rr_toi64;
-  case NVPTX::FVecLEV4F32: return NVPTX::FSetLEf32rr_toi32;
-  case NVPTX::FVecLTV2F32: return NVPTX::FSetLTf32rr_toi32;
-  case NVPTX::FVecLTV2F64: return NVPTX::FSetLTf64rr_toi64;
-  case NVPTX::FVecLTV4F32: return NVPTX::FSetLTf32rr_toi32;
-  case NVPTX::FVecNANV2F32: return NVPTX::FSetNANf32rr_toi32;
-  case NVPTX::FVecNANV2F64: return NVPTX::FSetNANf64rr_toi64;
-  case NVPTX::FVecNANV4F32: return NVPTX::FSetNANf32rr_toi32;
-  case NVPTX::FVecNEV2F32: return NVPTX::FSetNEf32rr_toi32;
-  case NVPTX::FVecNEV2F64: return NVPTX::FSetNEf64rr_toi64;
-  case NVPTX::FVecNEV4F32: return NVPTX::FSetNEf32rr_toi32;
-  case NVPTX::FVecNUMV2F32: return NVPTX::FSetNUMf32rr_toi32;
-  case NVPTX::FVecNUMV2F64: return NVPTX::FSetNUMf64rr_toi64;
-  case NVPTX::FVecNUMV4F32: return NVPTX::FSetNUMf32rr_toi32;
-  case NVPTX::FVecUEQV2F32: return NVPTX::FSetUEQf32rr_toi32;
-  case NVPTX::FVecUEQV2F64: return NVPTX::FSetUEQf64rr_toi64;
-  case NVPTX::FVecUEQV4F32: return NVPTX::FSetUEQf32rr_toi32;
-  case NVPTX::FVecUGEV2F32: return NVPTX::FSetUGEf32rr_toi32;
-  case NVPTX::FVecUGEV2F64: return NVPTX::FSetUGEf64rr_toi64;
-  case NVPTX::FVecUGEV4F32: return NVPTX::FSetUGEf32rr_toi32;
-  case NVPTX::FVecUGTV2F32: return NVPTX::FSetUGTf32rr_toi32;
-  case NVPTX::FVecUGTV2F64: return NVPTX::FSetUGTf64rr_toi64;
-  case NVPTX::FVecUGTV4F32: return NVPTX::FSetUGTf32rr_toi32;
-  case NVPTX::FVecULEV2F32: return NVPTX::FSetULEf32rr_toi32;
-  case NVPTX::FVecULEV2F64: return NVPTX::FSetULEf64rr_toi64;
-  case NVPTX::FVecULEV4F32: return NVPTX::FSetULEf32rr_toi32;
-  case NVPTX::FVecULTV2F32: return NVPTX::FSetULTf32rr_toi32;
-  case NVPTX::FVecULTV2F64: return NVPTX::FSetULTf64rr_toi64;
-  case NVPTX::FVecULTV4F32: return NVPTX::FSetULTf32rr_toi32;
-  case NVPTX::FVecUNEV2F32: return NVPTX::FSetUNEf32rr_toi32;
-  case NVPTX::FVecUNEV2F64: return NVPTX::FSetUNEf64rr_toi64;
-  case NVPTX::FVecUNEV4F32: return NVPTX::FSetUNEf32rr_toi32;
-  case NVPTX::I16MADV2: return NVPTX::MAD16rrr;
-  case NVPTX::I16MADV4: return NVPTX::MAD16rrr;
-  case NVPTX::I32MADV2: return NVPTX::MAD32rrr;
-  case NVPTX::I32MADV4: return NVPTX::MAD32rrr;
-  case NVPTX::I64MADV2: return NVPTX::MAD64rrr;
-  case NVPTX::I8MADV2: return NVPTX::MAD8rrr;
-  case NVPTX::I8MADV4: return NVPTX::MAD8rrr;
-  case NVPTX::ShiftLV2I16: return NVPTX::SHLi16rr;
-  case NVPTX::ShiftLV2I32: return NVPTX::SHLi32rr;
-  case NVPTX::ShiftLV2I64: return NVPTX::SHLi64rr;
-  case NVPTX::ShiftLV2I8: return NVPTX::SHLi8rr;
-  case NVPTX::ShiftLV4I16: return NVPTX::SHLi16rr;
-  case NVPTX::ShiftLV4I32: return NVPTX::SHLi32rr;
-  case NVPTX::ShiftLV4I8: return NVPTX::SHLi8rr;
-  case NVPTX::ShiftRAV2I16: return NVPTX::SRAi16rr;
-  case NVPTX::ShiftRAV2I32: return NVPTX::SRAi32rr;
-  case NVPTX::ShiftRAV2I64: return NVPTX::SRAi64rr;
-  case NVPTX::ShiftRAV2I8: return NVPTX::SRAi8rr;
-  case NVPTX::ShiftRAV4I16: return NVPTX::SRAi16rr;
-  case NVPTX::ShiftRAV4I32: return NVPTX::SRAi32rr;
-  case NVPTX::ShiftRAV4I8: return NVPTX::SRAi8rr;
-  case NVPTX::ShiftRLV2I16: return NVPTX::SRLi16rr;
-  case NVPTX::ShiftRLV2I32: return NVPTX::SRLi32rr;
-  case NVPTX::ShiftRLV2I64: return NVPTX::SRLi64rr;
-  case NVPTX::ShiftRLV2I8: return NVPTX::SRLi8rr;
-  case NVPTX::ShiftRLV4I16: return NVPTX::SRLi16rr;
-  case NVPTX::ShiftRLV4I32: return NVPTX::SRLi32rr;
-  case NVPTX::ShiftRLV4I8: return NVPTX::SRLi8rr;
-  case NVPTX::SubCCCV2I32: return NVPTX::SUBCCCi32rr;
-  case NVPTX::SubCCCV4I32: return NVPTX::SUBCCCi32rr;
-  case NVPTX::SubCCV2I32: return NVPTX::SUBCCi32rr;
-  case NVPTX::SubCCV4I32: return NVPTX::SUBCCi32rr;
-  case NVPTX::V2F32Div_prec_ftz: return NVPTX::FDIV32rr_prec_ftz;
-  case NVPTX::V2F32Div_prec: return NVPTX::FDIV32rr_prec;
-  case NVPTX::V2F32Div_ftz: return NVPTX::FDIV32rr_ftz;
-  case NVPTX::V2F32Div: return NVPTX::FDIV32rr;
-  case NVPTX::V2F32_Select: return NVPTX::SELECTf32rr;
-  case NVPTX::V2F64Div: return NVPTX::FDIV64rr;
-  case NVPTX::V2F64_Select: return NVPTX::SELECTf64rr;
-  case NVPTX::V2I16_Select: return NVPTX::SELECTi16rr;
-  case NVPTX::V2I32_Select: return NVPTX::SELECTi32rr;
-  case NVPTX::V2I64_Select: return NVPTX::SELECTi64rr;
-  case NVPTX::V2I8_Select: return NVPTX::SELECTi8rr;
-  case NVPTX::V2f32Extract: return NVPTX::FMOV32rr;
-  case NVPTX::V2f32Insert: return NVPTX::FMOV32rr;
-  case NVPTX::V2f32Mov: return NVPTX::FMOV32rr;
-  case NVPTX::V2f64Extract: return NVPTX::FMOV64rr;
-  case NVPTX::V2f64Insert: return NVPTX::FMOV64rr;
-  case NVPTX::V2f64Mov: return NVPTX::FMOV64rr;
-  case NVPTX::V2i16Extract: return NVPTX::IMOV16rr;
-  case NVPTX::V2i16Insert: return NVPTX::IMOV16rr;
-  case NVPTX::V2i16Mov: return NVPTX::IMOV16rr;
-  case NVPTX::V2i32Extract: return NVPTX::IMOV32rr;
-  case NVPTX::V2i32Insert: return NVPTX::IMOV32rr;
-  case NVPTX::V2i32Mov: return NVPTX::IMOV32rr;
-  case NVPTX::V2i64Extract: return NVPTX::IMOV64rr;
-  case NVPTX::V2i64Insert: return NVPTX::IMOV64rr;
-  case NVPTX::V2i64Mov: return NVPTX::IMOV64rr;
-  case NVPTX::V2i8Extract: return NVPTX::IMOV8rr;
-  case NVPTX::V2i8Insert: return NVPTX::IMOV8rr;
-  case NVPTX::V2i8Mov: return NVPTX::IMOV8rr;
-  case NVPTX::V4F32Div_prec_ftz: return NVPTX::FDIV32rr_prec_ftz;
-  case NVPTX::V4F32Div_prec: return NVPTX::FDIV32rr_prec;
-  case NVPTX::V4F32Div_ftz: return NVPTX::FDIV32rr_ftz;
-  case NVPTX::V4F32Div: return NVPTX::FDIV32rr;
-  case NVPTX::V4F32_Select: return NVPTX::SELECTf32rr;
-  case NVPTX::V4I16_Select: return NVPTX::SELECTi16rr;
-  case NVPTX::V4I32_Select: return NVPTX::SELECTi32rr;
-  case NVPTX::V4I8_Select: return NVPTX::SELECTi8rr;
-  case NVPTX::V4f32Extract: return NVPTX::FMOV32rr;
-  case NVPTX::V4f32Insert: return NVPTX::FMOV32rr;
-  case NVPTX::V4f32Mov: return NVPTX::FMOV32rr;
-  case NVPTX::V4i16Extract: return NVPTX::IMOV16rr;
-  case NVPTX::V4i16Insert: return NVPTX::IMOV16rr;
-  case NVPTX::V4i16Mov: return NVPTX::IMOV16rr;
-  case NVPTX::V4i32Extract: return NVPTX::IMOV32rr;
-  case NVPTX::V4i32Insert: return NVPTX::IMOV32rr;
-  case NVPTX::V4i32Mov: return NVPTX::IMOV32rr;
-  case NVPTX::V4i8Extract: return NVPTX::IMOV8rr;
-  case NVPTX::V4i8Insert: return NVPTX::IMOV8rr;
-  case NVPTX::V4i8Mov: return NVPTX::IMOV8rr;
-  case NVPTX::VAddV2I16: return NVPTX::ADDi16rr;
-  case NVPTX::VAddV2I32: return NVPTX::ADDi32rr;
-  case NVPTX::VAddV2I64: return NVPTX::ADDi64rr;
-  case NVPTX::VAddV2I8: return NVPTX::ADDi8rr;
-  case NVPTX::VAddV4I16: return NVPTX::ADDi16rr;
-  case NVPTX::VAddV4I32: return NVPTX::ADDi32rr;
-  case NVPTX::VAddV4I8: return NVPTX::ADDi8rr;
-  case NVPTX::VAddfV2F32: return NVPTX::FADDf32rr;
-  case NVPTX::VAddfV2F32_ftz: return NVPTX::FADDf32rr_ftz;
-  case NVPTX::VAddfV2F64: return NVPTX::FADDf64rr;
-  case NVPTX::VAddfV4F32: return NVPTX::FADDf32rr;
-  case NVPTX::VAddfV4F32_ftz: return NVPTX::FADDf32rr_ftz;
-  case NVPTX::VAndV2I16: return NVPTX::ANDb16rr;
-  case NVPTX::VAndV2I32: return NVPTX::ANDb32rr;
-  case NVPTX::VAndV2I64: return NVPTX::ANDb64rr;
-  case NVPTX::VAndV2I8: return NVPTX::ANDb8rr;
-  case NVPTX::VAndV4I16: return NVPTX::ANDb16rr;
-  case NVPTX::VAndV4I32: return NVPTX::ANDb32rr;
-  case NVPTX::VAndV4I8: return NVPTX::ANDb8rr;
-  case NVPTX::VMulfV2F32_ftz: return NVPTX::FMULf32rr_ftz;
-  case NVPTX::VMulfV2F32: return NVPTX::FMULf32rr;
-  case NVPTX::VMulfV2F64: return NVPTX::FMULf64rr;
-  case NVPTX::VMulfV4F32_ftz: return NVPTX::FMULf32rr_ftz;
-  case NVPTX::VMulfV4F32: return NVPTX::FMULf32rr;
-  case NVPTX::VMultHSV2I16: return NVPTX::MULTHSi16rr;
-  case NVPTX::VMultHSV2I32: return NVPTX::MULTHSi32rr;
-  case NVPTX::VMultHSV2I64: return NVPTX::MULTHSi64rr;
-  case NVPTX::VMultHSV2I8: return NVPTX::MULTHSi8rr;
-  case NVPTX::VMultHSV4I16: return NVPTX::MULTHSi16rr;
-  case NVPTX::VMultHSV4I32: return NVPTX::MULTHSi32rr;
-  case NVPTX::VMultHSV4I8: return NVPTX::MULTHSi8rr;
-  case NVPTX::VMultHUV2I16: return NVPTX::MULTHUi16rr;
-  case NVPTX::VMultHUV2I32: return NVPTX::MULTHUi32rr;
-  case NVPTX::VMultHUV2I64: return NVPTX::MULTHUi64rr;
-  case NVPTX::VMultHUV2I8: return NVPTX::MULTHUi8rr;
-  case NVPTX::VMultHUV4I16: return NVPTX::MULTHUi16rr;
-  case NVPTX::VMultHUV4I32: return NVPTX::MULTHUi32rr;
-  case NVPTX::VMultHUV4I8: return NVPTX::MULTHUi8rr;
-  case NVPTX::VMultV2I16: return NVPTX::MULTi16rr;
-  case NVPTX::VMultV2I32: return NVPTX::MULTi32rr;
-  case NVPTX::VMultV2I64: return NVPTX::MULTi64rr;
-  case NVPTX::VMultV2I8: return NVPTX::MULTi8rr;
-  case NVPTX::VMultV4I16: return NVPTX::MULTi16rr;
-  case NVPTX::VMultV4I32: return NVPTX::MULTi32rr;
-  case NVPTX::VMultV4I8: return NVPTX::MULTi8rr;
-  case NVPTX::VNegV2I16: return NVPTX::INEG16;
-  case NVPTX::VNegV2I32: return NVPTX::INEG32;
-  case NVPTX::VNegV2I64: return NVPTX::INEG64;
-  case NVPTX::VNegV2I8: return NVPTX::INEG8;
-  case NVPTX::VNegV4I16: return NVPTX::INEG16;
-  case NVPTX::VNegV4I32: return NVPTX::INEG32;
-  case NVPTX::VNegV4I8: return NVPTX::INEG8;
-  case NVPTX::VNegv2f32: return NVPTX::FNEGf32;
-  case NVPTX::VNegv2f32_ftz: return NVPTX::FNEGf32_ftz;
-  case NVPTX::VNegv2f64: return NVPTX::FNEGf64;
-  case NVPTX::VNegv4f32: return NVPTX::FNEGf32;
-  case NVPTX::VNegv4f32_ftz: return NVPTX::FNEGf32_ftz;
-  case NVPTX::VNotV2I16: return NVPTX::NOT16;
-  case NVPTX::VNotV2I32: return NVPTX::NOT32;
-  case NVPTX::VNotV2I64: return NVPTX::NOT64;
-  case NVPTX::VNotV2I8: return NVPTX::NOT8;
-  case NVPTX::VNotV4I16: return NVPTX::NOT16;
-  case NVPTX::VNotV4I32: return NVPTX::NOT32;
-  case NVPTX::VNotV4I8: return NVPTX::NOT8;
-  case NVPTX::VOrV2I16: return NVPTX::ORb16rr;
-  case NVPTX::VOrV2I32: return NVPTX::ORb32rr;
-  case NVPTX::VOrV2I64: return NVPTX::ORb64rr;
-  case NVPTX::VOrV2I8: return NVPTX::ORb8rr;
-  case NVPTX::VOrV4I16: return NVPTX::ORb16rr;
-  case NVPTX::VOrV4I32: return NVPTX::ORb32rr;
-  case NVPTX::VOrV4I8: return NVPTX::ORb8rr;
-  case NVPTX::VSDivV2I16: return NVPTX::SDIVi16rr;
-  case NVPTX::VSDivV2I32: return NVPTX::SDIVi32rr;
-  case NVPTX::VSDivV2I64: return NVPTX::SDIVi64rr;
-  case NVPTX::VSDivV2I8: return NVPTX::SDIVi8rr;
-  case NVPTX::VSDivV4I16: return NVPTX::SDIVi16rr;
-  case NVPTX::VSDivV4I32: return NVPTX::SDIVi32rr;
-  case NVPTX::VSDivV4I8: return NVPTX::SDIVi8rr;
-  case NVPTX::VSRemV2I16: return NVPTX::SREMi16rr;
-  case NVPTX::VSRemV2I32: return NVPTX::SREMi32rr;
-  case NVPTX::VSRemV2I64: return NVPTX::SREMi64rr;
-  case NVPTX::VSRemV2I8: return NVPTX::SREMi8rr;
-  case NVPTX::VSRemV4I16: return NVPTX::SREMi16rr;
-  case NVPTX::VSRemV4I32: return NVPTX::SREMi32rr;
-  case NVPTX::VSRemV4I8: return NVPTX::SREMi8rr;
-  case NVPTX::VSubV2I16: return NVPTX::SUBi16rr;
-  case NVPTX::VSubV2I32: return NVPTX::SUBi32rr;
-  case NVPTX::VSubV2I64: return NVPTX::SUBi64rr;
-  case NVPTX::VSubV2I8: return NVPTX::SUBi8rr;
-  case NVPTX::VSubV4I16: return NVPTX::SUBi16rr;
-  case NVPTX::VSubV4I32: return NVPTX::SUBi32rr;
-  case NVPTX::VSubV4I8: return NVPTX::SUBi8rr;
-  case NVPTX::VSubfV2F32_ftz: return NVPTX::FSUBf32rr_ftz;
-  case NVPTX::VSubfV2F32: return NVPTX::FSUBf32rr;
-  case NVPTX::VSubfV2F64: return NVPTX::FSUBf64rr;
-  case NVPTX::VSubfV4F32_ftz: return NVPTX::FSUBf32rr_ftz;
-  case NVPTX::VSubfV4F32: return NVPTX::FSUBf32rr;
-  case NVPTX::VUDivV2I16: return NVPTX::UDIVi16rr;
-  case NVPTX::VUDivV2I32: return NVPTX::UDIVi32rr;
-  case NVPTX::VUDivV2I64: return NVPTX::UDIVi64rr;
-  case NVPTX::VUDivV2I8: return NVPTX::UDIVi8rr;
-  case NVPTX::VUDivV4I16: return NVPTX::UDIVi16rr;
-  case NVPTX::VUDivV4I32: return NVPTX::UDIVi32rr;
-  case NVPTX::VUDivV4I8: return NVPTX::UDIVi8rr;
-  case NVPTX::VURemV2I16: return NVPTX::UREMi16rr;
-  case NVPTX::VURemV2I32: return NVPTX::UREMi32rr;
-  case NVPTX::VURemV2I64: return NVPTX::UREMi64rr;
-  case NVPTX::VURemV2I8: return NVPTX::UREMi8rr;
-  case NVPTX::VURemV4I16: return NVPTX::UREMi16rr;
-  case NVPTX::VURemV4I32: return NVPTX::UREMi32rr;
-  case NVPTX::VURemV4I8: return NVPTX::UREMi8rr;
-  case NVPTX::VXorV2I16: return NVPTX::XORb16rr;
-  case NVPTX::VXorV2I32: return NVPTX::XORb32rr;
-  case NVPTX::VXorV2I64: return NVPTX::XORb64rr;
-  case NVPTX::VXorV2I8: return NVPTX::XORb8rr;
-  case NVPTX::VXorV4I16: return NVPTX::XORb16rr;
-  case NVPTX::VXorV4I32: return NVPTX::XORb32rr;
-  case NVPTX::VXorV4I8: return NVPTX::XORb8rr;
-  case NVPTX::VecSEQV2I16: return NVPTX::ISetSEQi16rr_toi16;
-  case NVPTX::VecSEQV2I32: return NVPTX::ISetSEQi32rr_toi32;
-  case NVPTX::VecSEQV2I64: return NVPTX::ISetSEQi64rr_toi64;
-  case NVPTX::VecSEQV2I8: return NVPTX::ISetSEQi8rr_toi8;
-  case NVPTX::VecSEQV4I16: return NVPTX::ISetSEQi16rr_toi16;
-  case NVPTX::VecSEQV4I32: return NVPTX::ISetSEQi32rr_toi32;
-  case NVPTX::VecSEQV4I8: return NVPTX::ISetSEQi8rr_toi8;
-  case NVPTX::VecSGEV2I16: return NVPTX::ISetSGEi16rr_toi16;
-  case NVPTX::VecSGEV2I32: return NVPTX::ISetSGEi32rr_toi32;
-  case NVPTX::VecSGEV2I64: return NVPTX::ISetSGEi64rr_toi64;
-  case NVPTX::VecSGEV2I8: return NVPTX::ISetSGEi8rr_toi8;
-  case NVPTX::VecSGEV4I16: return NVPTX::ISetSGEi16rr_toi16;
-  case NVPTX::VecSGEV4I32: return NVPTX::ISetSGEi32rr_toi32;
-  case NVPTX::VecSGEV4I8: return NVPTX::ISetSGEi8rr_toi8;
-  case NVPTX::VecSGTV2I16: return NVPTX::ISetSGTi16rr_toi16;
-  case NVPTX::VecSGTV2I32: return NVPTX::ISetSGTi32rr_toi32;
-  case NVPTX::VecSGTV2I64: return NVPTX::ISetSGTi64rr_toi64;
-  case NVPTX::VecSGTV2I8: return NVPTX::ISetSGTi8rr_toi8;
-  case NVPTX::VecSGTV4I16: return NVPTX::ISetSGTi16rr_toi16;
-  case NVPTX::VecSGTV4I32: return NVPTX::ISetSGTi32rr_toi32;
-  case NVPTX::VecSGTV4I8: return NVPTX::ISetSGTi8rr_toi8;
-  case NVPTX::VecSLEV2I16: return NVPTX::ISetSLEi16rr_toi16;
-  case NVPTX::VecSLEV2I32: return NVPTX::ISetSLEi32rr_toi32;
-  case NVPTX::VecSLEV2I64: return NVPTX::ISetSLEi64rr_toi64;
-  case NVPTX::VecSLEV2I8: return NVPTX::ISetSLEi8rr_toi8;
-  case NVPTX::VecSLEV4I16: return NVPTX::ISetSLEi16rr_toi16;
-  case NVPTX::VecSLEV4I32: return NVPTX::ISetSLEi32rr_toi32;
-  case NVPTX::VecSLEV4I8: return NVPTX::ISetSLEi8rr_toi8;
-  case NVPTX::VecSLTV2I16: return NVPTX::ISetSLTi16rr_toi16;
-  case NVPTX::VecSLTV2I32: return NVPTX::ISetSLTi32rr_toi32;
-  case NVPTX::VecSLTV2I64: return NVPTX::ISetSLTi64rr_toi64;
-  case NVPTX::VecSLTV2I8: return NVPTX::ISetSLTi8rr_toi8;
-  case NVPTX::VecSLTV4I16: return NVPTX::ISetSLTi16rr_toi16;
-  case NVPTX::VecSLTV4I32: return NVPTX::ISetSLTi32rr_toi32;
-  case NVPTX::VecSLTV4I8: return NVPTX::ISetSLTi8rr_toi8;
-  case NVPTX::VecSNEV2I16: return NVPTX::ISetSNEi16rr_toi16;
-  case NVPTX::VecSNEV2I32: return NVPTX::ISetSNEi32rr_toi32;
-  case NVPTX::VecSNEV2I64: return NVPTX::ISetSNEi64rr_toi64;
-  case NVPTX::VecSNEV2I8: return NVPTX::ISetSNEi8rr_toi8;
-  case NVPTX::VecSNEV4I16: return NVPTX::ISetSNEi16rr_toi16;
-  case NVPTX::VecSNEV4I32: return NVPTX::ISetSNEi32rr_toi32;
-  case NVPTX::VecSNEV4I8: return NVPTX::ISetSNEi8rr_toi8;
-  case NVPTX::VecShuffle_v2f32: return NVPTX::FMOV32rr;
-  case NVPTX::VecShuffle_v2f64: return NVPTX::FMOV64rr;
-  case NVPTX::VecShuffle_v2i16: return NVPTX::IMOV16rr;
-  case NVPTX::VecShuffle_v2i32: return NVPTX::IMOV32rr;
-  case NVPTX::VecShuffle_v2i64: return NVPTX::IMOV64rr;
-  case NVPTX::VecShuffle_v2i8: return NVPTX::IMOV8rr;
-  case NVPTX::VecShuffle_v4f32: return NVPTX::FMOV32rr;
-  case NVPTX::VecShuffle_v4i16: return NVPTX::IMOV16rr;
-  case NVPTX::VecShuffle_v4i32: return NVPTX::IMOV32rr;
-  case NVPTX::VecShuffle_v4i8: return NVPTX::IMOV8rr;
-  case NVPTX::VecUEQV2I16: return NVPTX::ISetUEQi16rr_toi16;
-  case NVPTX::VecUEQV2I32: return NVPTX::ISetUEQi32rr_toi32;
-  case NVPTX::VecUEQV2I64: return NVPTX::ISetUEQi64rr_toi64;
-  case NVPTX::VecUEQV2I8: return NVPTX::ISetUEQi8rr_toi8;
-  case NVPTX::VecUEQV4I16: return NVPTX::ISetUEQi16rr_toi16;
-  case NVPTX::VecUEQV4I32: return NVPTX::ISetUEQi32rr_toi32;
-  case NVPTX::VecUEQV4I8: return NVPTX::ISetUEQi8rr_toi8;
-  case NVPTX::VecUGEV2I16: return NVPTX::ISetUGEi16rr_toi16;
-  case NVPTX::VecUGEV2I32: return NVPTX::ISetUGEi32rr_toi32;
-  case NVPTX::VecUGEV2I64: return NVPTX::ISetUGEi64rr_toi64;
-  case NVPTX::VecUGEV2I8: return NVPTX::ISetUGEi8rr_toi8;
-  case NVPTX::VecUGEV4I16: return NVPTX::ISetUGEi16rr_toi16;
-  case NVPTX::VecUGEV4I32: return NVPTX::ISetUGEi32rr_toi32;
-  case NVPTX::VecUGEV4I8: return NVPTX::ISetUGEi8rr_toi8;
-  case NVPTX::VecUGTV2I16: return NVPTX::ISetUGTi16rr_toi16;
-  case NVPTX::VecUGTV2I32: return NVPTX::ISetUGTi32rr_toi32;
-  case NVPTX::VecUGTV2I64: return NVPTX::ISetUGTi64rr_toi64;
-  case NVPTX::VecUGTV2I8: return NVPTX::ISetUGTi8rr_toi8;
-  case NVPTX::VecUGTV4I16: return NVPTX::ISetUGTi16rr_toi16;
-  case NVPTX::VecUGTV4I32: return NVPTX::ISetUGTi32rr_toi32;
-  case NVPTX::VecUGTV4I8: return NVPTX::ISetUGTi8rr_toi8;
-  case NVPTX::VecULEV2I16: return NVPTX::ISetULEi16rr_toi16;
-  case NVPTX::VecULEV2I32: return NVPTX::ISetULEi32rr_toi32;
-  case NVPTX::VecULEV2I64: return NVPTX::ISetULEi64rr_toi64;
-  case NVPTX::VecULEV2I8: return NVPTX::ISetULEi8rr_toi8;
-  case NVPTX::VecULEV4I16: return NVPTX::ISetULEi16rr_toi16;
-  case NVPTX::VecULEV4I32: return NVPTX::ISetULEi32rr_toi32;
-  case NVPTX::VecULEV4I8: return NVPTX::ISetULEi8rr_toi8;
-  case NVPTX::VecULTV2I16: return NVPTX::ISetULTi16rr_toi16;
-  case NVPTX::VecULTV2I32: return NVPTX::ISetULTi32rr_toi32;
-  case NVPTX::VecULTV2I64: return NVPTX::ISetULTi64rr_toi64;
-  case NVPTX::VecULTV2I8: return NVPTX::ISetULTi8rr_toi8;
-  case NVPTX::VecULTV4I16: return NVPTX::ISetULTi16rr_toi16;
-  case NVPTX::VecULTV4I32: return NVPTX::ISetULTi32rr_toi32;
-  case NVPTX::VecULTV4I8: return NVPTX::ISetULTi8rr_toi8;
-  case NVPTX::VecUNEV2I16: return NVPTX::ISetUNEi16rr_toi16;
-  case NVPTX::VecUNEV2I32: return NVPTX::ISetUNEi32rr_toi32;
-  case NVPTX::VecUNEV2I64: return NVPTX::ISetUNEi64rr_toi64;
-  case NVPTX::VecUNEV2I8: return NVPTX::ISetUNEi8rr_toi8;
-  case NVPTX::VecUNEV4I16: return NVPTX::ISetUNEi16rr_toi16;
-  case NVPTX::VecUNEV4I32: return NVPTX::ISetUNEi32rr_toi32;
-  case NVPTX::VecUNEV4I8: return NVPTX::ISetUNEi8rr_toi8;
-  case NVPTX::INT_PTX_LDU_G_v2i8_32: return NVPTX::INT_PTX_LDU_G_v2i8_ELE_32;
-  case NVPTX::INT_PTX_LDU_G_v4i8_32: return NVPTX::INT_PTX_LDU_G_v4i8_ELE_32;
-  case NVPTX::INT_PTX_LDU_G_v2i16_32: return NVPTX::INT_PTX_LDU_G_v2i16_ELE_32;
-  case NVPTX::INT_PTX_LDU_G_v4i16_32: return NVPTX::INT_PTX_LDU_G_v4i16_ELE_32;
-  case NVPTX::INT_PTX_LDU_G_v2i32_32: return NVPTX::INT_PTX_LDU_G_v2i32_ELE_32;
-  case NVPTX::INT_PTX_LDU_G_v4i32_32: return NVPTX::INT_PTX_LDU_G_v4i32_ELE_32;
-  case NVPTX::INT_PTX_LDU_G_v2f32_32: return NVPTX::INT_PTX_LDU_G_v2f32_ELE_32;
-  case NVPTX::INT_PTX_LDU_G_v4f32_32: return NVPTX::INT_PTX_LDU_G_v4f32_ELE_32;
-  case NVPTX::INT_PTX_LDU_G_v2i64_32: return NVPTX::INT_PTX_LDU_G_v2i64_ELE_32;
-  case NVPTX::INT_PTX_LDU_G_v2f64_32: return NVPTX::INT_PTX_LDU_G_v2f64_ELE_32;
-  case NVPTX::INT_PTX_LDU_G_v2i8_64: return NVPTX::INT_PTX_LDU_G_v2i8_ELE_64;
-  case NVPTX::INT_PTX_LDU_G_v4i8_64: return NVPTX::INT_PTX_LDU_G_v4i8_ELE_64;
-  case NVPTX::INT_PTX_LDU_G_v2i16_64: return NVPTX::INT_PTX_LDU_G_v2i16_ELE_64;
-  case NVPTX::INT_PTX_LDU_G_v4i16_64: return NVPTX::INT_PTX_LDU_G_v4i16_ELE_64;
-  case NVPTX::INT_PTX_LDU_G_v2i32_64: return NVPTX::INT_PTX_LDU_G_v2i32_ELE_64;
-  case NVPTX::INT_PTX_LDU_G_v4i32_64: return NVPTX::INT_PTX_LDU_G_v4i32_ELE_64;
-  case NVPTX::INT_PTX_LDU_G_v2f32_64: return NVPTX::INT_PTX_LDU_G_v2f32_ELE_64;
-  case NVPTX::INT_PTX_LDU_G_v4f32_64: return NVPTX::INT_PTX_LDU_G_v4f32_ELE_64;
-  case NVPTX::INT_PTX_LDU_G_v2i64_64: return NVPTX::INT_PTX_LDU_G_v2i64_ELE_64;
-  case NVPTX::INT_PTX_LDU_G_v2f64_64: return NVPTX::INT_PTX_LDU_G_v2f64_ELE_64;
-
-  case NVPTX::LoadParamV4I32: return NVPTX::LoadParamScalar4I32;
-  case NVPTX::LoadParamV4I16: return NVPTX::LoadParamScalar4I16;
-  case NVPTX::LoadParamV4I8: return NVPTX::LoadParamScalar4I8;
-  case NVPTX::LoadParamV2I64: return NVPTX::LoadParamScalar2I64;
-  case NVPTX::LoadParamV2I32: return NVPTX::LoadParamScalar2I32;
-  case NVPTX::LoadParamV2I16: return NVPTX::LoadParamScalar2I16;
-  case NVPTX::LoadParamV2I8: return NVPTX::LoadParamScalar2I8;
-  case NVPTX::LoadParamV4F32: return NVPTX::LoadParamScalar4F32;
-  case NVPTX::LoadParamV2F32: return NVPTX::LoadParamScalar2F32;
-  case NVPTX::LoadParamV2F64: return NVPTX::LoadParamScalar2F64;
-  case NVPTX::StoreParamV4I32: return NVPTX::StoreParamScalar4I32;
-  case NVPTX::StoreParamV4I16: return NVPTX::StoreParamScalar4I16;
-  case NVPTX::StoreParamV4I8: return NVPTX::StoreParamScalar4I8;
-  case NVPTX::StoreParamV2I64: return NVPTX::StoreParamScalar2I64;
-  case NVPTX::StoreParamV2I32: return NVPTX::StoreParamScalar2I32;
-  case NVPTX::StoreParamV2I16: return NVPTX::StoreParamScalar2I16;
-  case NVPTX::StoreParamV2I8: return NVPTX::StoreParamScalar2I8;
-  case NVPTX::StoreParamV4F32: return NVPTX::StoreParamScalar4F32;
-  case NVPTX::StoreParamV2F32: return NVPTX::StoreParamScalar2F32;
-  case NVPTX::StoreParamV2F64: return NVPTX::StoreParamScalar2F64;
-  case NVPTX::StoreRetvalV4I32: return NVPTX::StoreRetvalScalar4I32;
-  case NVPTX::StoreRetvalV4I16: return NVPTX::StoreRetvalScalar4I16;
-  case NVPTX::StoreRetvalV4I8: return NVPTX::StoreRetvalScalar4I8;
-  case NVPTX::StoreRetvalV2I64: return NVPTX::StoreRetvalScalar2I64;
-  case NVPTX::StoreRetvalV2I32: return NVPTX::StoreRetvalScalar2I32;
-  case NVPTX::StoreRetvalV2I16: return NVPTX::StoreRetvalScalar2I16;
-  case NVPTX::StoreRetvalV2I8: return NVPTX::StoreRetvalScalar2I8;
-  case NVPTX::StoreRetvalV4F32: return NVPTX::StoreRetvalScalar4F32;
-  case NVPTX::StoreRetvalV2F32: return NVPTX::StoreRetvalScalar2F32;
-  case NVPTX::StoreRetvalV2F64: return NVPTX::StoreRetvalScalar2F64;
-  case NVPTX::VecI32toV4I8: return NVPTX::I32toV4I8;
-  case NVPTX::VecI64toV4I16: return NVPTX::I64toV4I16;
-  case NVPTX::VecI16toV2I8: return NVPTX::I16toV2I8;
-  case NVPTX::VecI32toV2I16: return NVPTX::I32toV2I16;
-  case NVPTX::VecI64toV2I32: return NVPTX::I64toV2I32;
-  case NVPTX::VecF64toV2F32: return NVPTX::F64toV2F32;
-
-  case NVPTX::LD_v2i8_avar: return NVPTX::LDV_i8_v2_avar;
-  case NVPTX::LD_v2i8_areg: return NVPTX::LDV_i8_v2_areg;
-  case NVPTX::LD_v2i8_ari:  return NVPTX::LDV_i8_v2_ari;
-  case NVPTX::LD_v2i8_asi:  return NVPTX::LDV_i8_v2_asi;
-  case NVPTX::LD_v4i8_avar: return NVPTX::LDV_i8_v4_avar;
-  case NVPTX::LD_v4i8_areg: return NVPTX::LDV_i8_v4_areg;
-  case NVPTX::LD_v4i8_ari:  return NVPTX::LDV_i8_v4_ari;
-  case NVPTX::LD_v4i8_asi:  return NVPTX::LDV_i8_v4_asi;
-
-  case NVPTX::LD_v2i16_avar: return NVPTX::LDV_i16_v2_avar;
-  case NVPTX::LD_v2i16_areg: return NVPTX::LDV_i16_v2_areg;
-  case NVPTX::LD_v2i16_ari:  return NVPTX::LDV_i16_v2_ari;
-  case NVPTX::LD_v2i16_asi:  return NVPTX::LDV_i16_v2_asi;
-  case NVPTX::LD_v4i16_avar: return NVPTX::LDV_i16_v4_avar;
-  case NVPTX::LD_v4i16_areg: return NVPTX::LDV_i16_v4_areg;
-  case NVPTX::LD_v4i16_ari:  return NVPTX::LDV_i16_v4_ari;
-  case NVPTX::LD_v4i16_asi:  return NVPTX::LDV_i16_v4_asi;
-
-  case NVPTX::LD_v2i32_avar: return NVPTX::LDV_i32_v2_avar;
-  case NVPTX::LD_v2i32_areg: return NVPTX::LDV_i32_v2_areg;
-  case NVPTX::LD_v2i32_ari:  return NVPTX::LDV_i32_v2_ari;
-  case NVPTX::LD_v2i32_asi:  return NVPTX::LDV_i32_v2_asi;
-  case NVPTX::LD_v4i32_avar: return NVPTX::LDV_i32_v4_avar;
-  case NVPTX::LD_v4i32_areg: return NVPTX::LDV_i32_v4_areg;
-  case NVPTX::LD_v4i32_ari:  return NVPTX::LDV_i32_v4_ari;
-  case NVPTX::LD_v4i32_asi:  return NVPTX::LDV_i32_v4_asi;
-
-  case NVPTX::LD_v2f32_avar: return NVPTX::LDV_f32_v2_avar;
-  case NVPTX::LD_v2f32_areg: return NVPTX::LDV_f32_v2_areg;
-  case NVPTX::LD_v2f32_ari:  return NVPTX::LDV_f32_v2_ari;
-  case NVPTX::LD_v2f32_asi:  return NVPTX::LDV_f32_v2_asi;
-  case NVPTX::LD_v4f32_avar: return NVPTX::LDV_f32_v4_avar;
-  case NVPTX::LD_v4f32_areg: return NVPTX::LDV_f32_v4_areg;
-  case NVPTX::LD_v4f32_ari:  return NVPTX::LDV_f32_v4_ari;
-  case NVPTX::LD_v4f32_asi:  return NVPTX::LDV_f32_v4_asi;
-
-  case NVPTX::LD_v2i64_avar: return NVPTX::LDV_i64_v2_avar;
-  case NVPTX::LD_v2i64_areg: return NVPTX::LDV_i64_v2_areg;
-  case NVPTX::LD_v2i64_ari:  return NVPTX::LDV_i64_v2_ari;
-  case NVPTX::LD_v2i64_asi:  return NVPTX::LDV_i64_v2_asi;
-  case NVPTX::LD_v2f64_avar: return NVPTX::LDV_f64_v2_avar;
-  case NVPTX::LD_v2f64_areg: return NVPTX::LDV_f64_v2_areg;
-  case NVPTX::LD_v2f64_ari:  return NVPTX::LDV_f64_v2_ari;
-  case NVPTX::LD_v2f64_asi:  return NVPTX::LDV_f64_v2_asi;
-
-  case NVPTX::ST_v2i8_avar: return NVPTX::STV_i8_v2_avar;
-  case NVPTX::ST_v2i8_areg: return NVPTX::STV_i8_v2_areg;
-  case NVPTX::ST_v2i8_ari:  return NVPTX::STV_i8_v2_ari;
-  case NVPTX::ST_v2i8_asi:  return NVPTX::STV_i8_v2_asi;
-  case NVPTX::ST_v4i8_avar: return NVPTX::STV_i8_v4_avar;
-  case NVPTX::ST_v4i8_areg: return NVPTX::STV_i8_v4_areg;
-  case NVPTX::ST_v4i8_ari:  return NVPTX::STV_i8_v4_ari;
-  case NVPTX::ST_v4i8_asi:  return NVPTX::STV_i8_v4_asi;
-
-  case NVPTX::ST_v2i16_avar: return NVPTX::STV_i16_v2_avar;
-  case NVPTX::ST_v2i16_areg: return NVPTX::STV_i16_v2_areg;
-  case NVPTX::ST_v2i16_ari:  return NVPTX::STV_i16_v2_ari;
-  case NVPTX::ST_v2i16_asi:  return NVPTX::STV_i16_v2_asi;
-  case NVPTX::ST_v4i16_avar: return NVPTX::STV_i16_v4_avar;
-  case NVPTX::ST_v4i16_areg: return NVPTX::STV_i16_v4_areg;
-  case NVPTX::ST_v4i16_ari:  return NVPTX::STV_i16_v4_ari;
-  case NVPTX::ST_v4i16_asi:  return NVPTX::STV_i16_v4_asi;
-
-  case NVPTX::ST_v2i32_avar: return NVPTX::STV_i32_v2_avar;
-  case NVPTX::ST_v2i32_areg: return NVPTX::STV_i32_v2_areg;
-  case NVPTX::ST_v2i32_ari:  return NVPTX::STV_i32_v2_ari;
-  case NVPTX::ST_v2i32_asi:  return NVPTX::STV_i32_v2_asi;
-  case NVPTX::ST_v4i32_avar: return NVPTX::STV_i32_v4_avar;
-  case NVPTX::ST_v4i32_areg: return NVPTX::STV_i32_v4_areg;
-  case NVPTX::ST_v4i32_ari:  return NVPTX::STV_i32_v4_ari;
-  case NVPTX::ST_v4i32_asi:  return NVPTX::STV_i32_v4_asi;
-
-  case NVPTX::ST_v2f32_avar: return NVPTX::STV_f32_v2_avar;
-  case NVPTX::ST_v2f32_areg: return NVPTX::STV_f32_v2_areg;
-  case NVPTX::ST_v2f32_ari:  return NVPTX::STV_f32_v2_ari;
-  case NVPTX::ST_v2f32_asi:  return NVPTX::STV_f32_v2_asi;
-  case NVPTX::ST_v4f32_avar: return NVPTX::STV_f32_v4_avar;
-  case NVPTX::ST_v4f32_areg: return NVPTX::STV_f32_v4_areg;
-  case NVPTX::ST_v4f32_ari:  return NVPTX::STV_f32_v4_ari;
-  case NVPTX::ST_v4f32_asi:  return NVPTX::STV_f32_v4_asi;
-
-  case NVPTX::ST_v2i64_avar: return NVPTX::STV_i64_v2_avar;
-  case NVPTX::ST_v2i64_areg: return NVPTX::STV_i64_v2_areg;
-  case NVPTX::ST_v2i64_ari:  return NVPTX::STV_i64_v2_ari;
-  case NVPTX::ST_v2i64_asi:  return NVPTX::STV_i64_v2_asi;
-  case NVPTX::ST_v2f64_avar: return NVPTX::STV_f64_v2_avar;
-  case NVPTX::ST_v2f64_areg: return NVPTX::STV_f64_v2_areg;
-  case NVPTX::ST_v2f64_ari:  return NVPTX::STV_f64_v2_ari;
-  case NVPTX::ST_v2f64_asi:  return NVPTX::STV_f64_v2_asi;
-  }
-  return 0;
-}
diff --git a/lib/Target/NVPTX/cl_common_defines.h b/lib/Target/NVPTX/cl_common_defines.h
index a7347efd7850..45cc0b8b67f2 100644
--- a/lib/Target/NVPTX/cl_common_defines.h
+++ b/lib/Target/NVPTX/cl_common_defines.h
@@ -24,22 +24,21 @@ enum {
   CLK_LUMINANCE = 0x10B9
 
 #if (__NV_CL_C_VERSION >= __NV_CL_C_VERSION_1_1)
-  ,
+      ,
   CLK_Rx = 0x10BA,
   CLK_RGx = 0x10BB,
   CLK_RGBx = 0x10BC
 #endif
 };
 
-
 typedef enum clk_channel_type {
   // valid formats for float return types
-  CLK_SNORM_INT8 = 0x10D0,            // four channel RGBA unorm8
-  CLK_SNORM_INT16 = 0x10D1,           // four channel RGBA unorm16
-  CLK_UNORM_INT8 = 0x10D2,            // four channel RGBA unorm8
-  CLK_UNORM_INT16 = 0x10D3,           // four channel RGBA unorm16
-  CLK_HALF_FLOAT = 0x10DD,            // four channel RGBA half
-  CLK_FLOAT = 0x10DE,                 // four channel RGBA float
+  CLK_SNORM_INT8 = 0x10D0,  // four channel RGBA unorm8
+  CLK_SNORM_INT16 = 0x10D1, // four channel RGBA unorm16
+  CLK_UNORM_INT8 = 0x10D2,  // four channel RGBA unorm8
+  CLK_UNORM_INT16 = 0x10D3, // four channel RGBA unorm16
+  CLK_HALF_FLOAT = 0x10DD,  // four channel RGBA half
+  CLK_FLOAT = 0x10DE,       // four channel RGBA float
 
 #if (__NV_CL_C_VERSION >= __NV_CL_C_VERSION_1_1)
   CLK_UNORM_SHORT_565 = 0x10D4,
@@ -48,7 +47,7 @@ typedef enum clk_channel_type {
 #endif
 
   // valid only for integer return types
-  CLK_SIGNED_INT8 =  0x10D7,
+  CLK_SIGNED_INT8 = 0x10D7,
   CLK_SIGNED_INT16 = 0x10D8,
   CLK_SIGNED_INT32 = 0x10D9,
   CLK_UNSIGNED_INT8 = 0x10DA,
@@ -56,70 +55,68 @@ typedef enum clk_channel_type {
   CLK_UNSIGNED_INT32 = 0x10DC,
 
   // CI SPI for CPU
-  __CLK_UNORM_INT8888 ,         // four channel ARGB unorm8
-  __CLK_UNORM_INT8888R,        // four channel BGRA unorm8
+  __CLK_UNORM_INT8888,  // four channel ARGB unorm8
+  __CLK_UNORM_INT8888R, // four channel BGRA unorm8
 
   __CLK_VALID_IMAGE_TYPE_COUNT,
   __CLK_INVALID_IMAGE_TYPE = __CLK_VALID_IMAGE_TYPE_COUNT,
-  __CLK_VALID_IMAGE_TYPE_MASK_BITS = 4,         // number of bits required to
-                                                // represent any image type
-  __CLK_VALID_IMAGE_TYPE_MASK = ( 1 << __CLK_VALID_IMAGE_TYPE_MASK_BITS ) - 1
-}clk_channel_type;
+  __CLK_VALID_IMAGE_TYPE_MASK_BITS = 4, // number of bits required to
+                                        // represent any image type
+  __CLK_VALID_IMAGE_TYPE_MASK = (1 << __CLK_VALID_IMAGE_TYPE_MASK_BITS) - 1
+} clk_channel_type;
 
 typedef enum clk_sampler_type {
-    __CLK_ADDRESS_BASE             = 0,
-    CLK_ADDRESS_NONE               = 0 << __CLK_ADDRESS_BASE,
-    CLK_ADDRESS_CLAMP              = 1 << __CLK_ADDRESS_BASE,
-    CLK_ADDRESS_CLAMP_TO_EDGE      = 2 << __CLK_ADDRESS_BASE,
-    CLK_ADDRESS_REPEAT             = 3 << __CLK_ADDRESS_BASE,
-    CLK_ADDRESS_MIRROR             = 4 << __CLK_ADDRESS_BASE,
+  __CLK_ADDRESS_BASE = 0,
+  CLK_ADDRESS_NONE = 0 << __CLK_ADDRESS_BASE,
+  CLK_ADDRESS_CLAMP = 1 << __CLK_ADDRESS_BASE,
+  CLK_ADDRESS_CLAMP_TO_EDGE = 2 << __CLK_ADDRESS_BASE,
+  CLK_ADDRESS_REPEAT = 3 << __CLK_ADDRESS_BASE,
+  CLK_ADDRESS_MIRROR = 4 << __CLK_ADDRESS_BASE,
 
 #if (__NV_CL_C_VERSION >= __NV_CL_C_VERSION_1_1)
-    CLK_ADDRESS_MIRRORED_REPEAT    = CLK_ADDRESS_MIRROR,
+  CLK_ADDRESS_MIRRORED_REPEAT = CLK_ADDRESS_MIRROR,
 #endif
-    __CLK_ADDRESS_MASK             = CLK_ADDRESS_NONE | CLK_ADDRESS_CLAMP |
-                                     CLK_ADDRESS_CLAMP_TO_EDGE |
-                                     CLK_ADDRESS_REPEAT | CLK_ADDRESS_MIRROR,
-    __CLK_ADDRESS_BITS             = 3,        // number of bits required to
-                                               // represent address info
-
-    __CLK_NORMALIZED_BASE          = __CLK_ADDRESS_BITS,
-    CLK_NORMALIZED_COORDS_FALSE    = 0,
-    CLK_NORMALIZED_COORDS_TRUE     = 1 << __CLK_NORMALIZED_BASE,
-    __CLK_NORMALIZED_MASK          = CLK_NORMALIZED_COORDS_FALSE |
-                                     CLK_NORMALIZED_COORDS_TRUE,
-    __CLK_NORMALIZED_BITS          = 1,        // number of bits required to
-                                               // represent normalization
-
-    __CLK_FILTER_BASE              = __CLK_NORMALIZED_BASE +
-                                     __CLK_NORMALIZED_BITS,
-    CLK_FILTER_NEAREST             = 0 << __CLK_FILTER_BASE,
-    CLK_FILTER_LINEAR              = 1 << __CLK_FILTER_BASE,
-    CLK_FILTER_ANISOTROPIC         = 2 << __CLK_FILTER_BASE,
-    __CLK_FILTER_MASK              = CLK_FILTER_NEAREST | CLK_FILTER_LINEAR |
-                                     CLK_FILTER_ANISOTROPIC,
-    __CLK_FILTER_BITS              = 2,        // number of bits required to
-                                               // represent address info
-
-    __CLK_MIP_BASE                 = __CLK_FILTER_BASE + __CLK_FILTER_BITS,
-    CLK_MIP_NEAREST                = 0 << __CLK_MIP_BASE,
-    CLK_MIP_LINEAR                 = 1 << __CLK_MIP_BASE,
-    CLK_MIP_ANISOTROPIC            = 2 << __CLK_MIP_BASE,
-    __CLK_MIP_MASK                 = CLK_MIP_NEAREST | CLK_MIP_LINEAR |
-                                     CLK_MIP_ANISOTROPIC,
-    __CLK_MIP_BITS                 = 2,
-
-    __CLK_SAMPLER_BITS             = __CLK_MIP_BASE + __CLK_MIP_BITS,
-    __CLK_SAMPLER_MASK             = __CLK_MIP_MASK | __CLK_FILTER_MASK |
-                                     __CLK_NORMALIZED_MASK | __CLK_ADDRESS_MASK,
-
-    __CLK_ANISOTROPIC_RATIO_BITS   = 5,
-    __CLK_ANISOTROPIC_RATIO_MASK   = (int) 0x80000000 >>
-                                      (__CLK_ANISOTROPIC_RATIO_BITS-1)
+  __CLK_ADDRESS_MASK =
+      CLK_ADDRESS_NONE | CLK_ADDRESS_CLAMP | CLK_ADDRESS_CLAMP_TO_EDGE |
+      CLK_ADDRESS_REPEAT | CLK_ADDRESS_MIRROR,
+  __CLK_ADDRESS_BITS = 3, // number of bits required to
+                          // represent address info
+
+  __CLK_NORMALIZED_BASE = __CLK_ADDRESS_BITS,
+  CLK_NORMALIZED_COORDS_FALSE = 0,
+  CLK_NORMALIZED_COORDS_TRUE = 1 << __CLK_NORMALIZED_BASE,
+  __CLK_NORMALIZED_MASK =
+      CLK_NORMALIZED_COORDS_FALSE | CLK_NORMALIZED_COORDS_TRUE,
+  __CLK_NORMALIZED_BITS = 1, // number of bits required to
+                             // represent normalization
+
+  __CLK_FILTER_BASE = __CLK_NORMALIZED_BASE + __CLK_NORMALIZED_BITS,
+  CLK_FILTER_NEAREST = 0 << __CLK_FILTER_BASE,
+  CLK_FILTER_LINEAR = 1 << __CLK_FILTER_BASE,
+  CLK_FILTER_ANISOTROPIC = 2 << __CLK_FILTER_BASE,
+  __CLK_FILTER_MASK =
+      CLK_FILTER_NEAREST | CLK_FILTER_LINEAR | CLK_FILTER_ANISOTROPIC,
+  __CLK_FILTER_BITS = 2, // number of bits required to
+                         // represent address info
+
+  __CLK_MIP_BASE = __CLK_FILTER_BASE + __CLK_FILTER_BITS,
+  CLK_MIP_NEAREST = 0 << __CLK_MIP_BASE,
+  CLK_MIP_LINEAR = 1 << __CLK_MIP_BASE,
+  CLK_MIP_ANISOTROPIC = 2 << __CLK_MIP_BASE,
+  __CLK_MIP_MASK = CLK_MIP_NEAREST | CLK_MIP_LINEAR | CLK_MIP_ANISOTROPIC,
+  __CLK_MIP_BITS = 2,
+
+  __CLK_SAMPLER_BITS = __CLK_MIP_BASE + __CLK_MIP_BITS,
+  __CLK_SAMPLER_MASK = __CLK_MIP_MASK | __CLK_FILTER_MASK |
+                       __CLK_NORMALIZED_MASK | __CLK_ADDRESS_MASK,
+
+  __CLK_ANISOTROPIC_RATIO_BITS = 5,
+  __CLK_ANISOTROPIC_RATIO_MASK =
+      (int) 0x80000000 >> (__CLK_ANISOTROPIC_RATIO_BITS - 1)
 } clk_sampler_type;
 
 // Memory synchronization
-#define CLK_LOCAL_MEM_FENCE     (1 << 0)
-#define CLK_GLOBAL_MEM_FENCE    (1 << 1)
+#define CLK_LOCAL_MEM_FENCE (1 << 0)
+#define CLK_GLOBAL_MEM_FENCE (1 << 1)
 
 #endif // __CL_COMMON_DEFINES_H__
diff --git a/lib/Target/NVPTX/gen-register-defs.py b/lib/Target/NVPTX/gen-register-defs.py
deleted file mode 100644
index ed0666823124..000000000000
--- a/lib/Target/NVPTX/gen-register-defs.py
+++ /dev/null
@@ -1,202 +0,0 @@
-#!/usr/bin/env python
-
-num_regs = 396
-
-outFile = open('NVPTXRegisterInfo.td', 'w')
-
-outFile.write('''
-//===-- NVPTXRegisterInfo.td - NVPTX Register defs ---------*- tablegen -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-//===----------------------------------------------------------------------===//
-//  Declarations that describe the PTX register file
-//===----------------------------------------------------------------------===//
-
-class NVPTXReg<string n> : Register<n> {
-  let Namespace = "NVPTX";
-}
-
-class NVPTXRegClass<list<ValueType> regTypes, int alignment, dag regList>
-     : RegisterClass <"NVPTX", regTypes, alignment, regList>;
-
-//===----------------------------------------------------------------------===//
-//  Registers
-//===----------------------------------------------------------------------===//
-
-// Special Registers used as stack pointer
-def VRFrame         : NVPTXReg<"%SP">;
-def VRFrameLocal    : NVPTXReg<"%SPL">;
-
-// Special Registers used as the stack
-def VRDepot  : NVPTXReg<"%Depot">;
-''')
-
-# Predicates
-outFile.write('''
-//===--- Predicate --------------------------------------------------------===//
-''')
-for i in range(0, num_regs):
-  outFile.write('def P%d : NVPTXReg<"%%p%d">;\n' % (i, i))
-
-# Int8
-outFile.write('''
-//===--- 8-bit ------------------------------------------------------------===//
-''')
-for i in range(0, num_regs):
-  outFile.write('def RC%d : NVPTXReg<"%%rc%d">;\n' % (i, i))
-
-# Int16
-outFile.write('''
-//===--- 16-bit -----------------------------------------------------------===//
-''')
-for i in range(0, num_regs):
-  outFile.write('def RS%d : NVPTXReg<"%%rs%d">;\n' % (i, i))
-
-# Int32
-outFile.write('''
-//===--- 32-bit -----------------------------------------------------------===//
-''')
-for i in range(0, num_regs):
-  outFile.write('def R%d : NVPTXReg<"%%r%d">;\n' % (i, i))
-
-# Int64
-outFile.write('''
-//===--- 64-bit -----------------------------------------------------------===//
-''')
-for i in range(0, num_regs):
-  outFile.write('def RL%d : NVPTXReg<"%%rl%d">;\n' % (i, i))
-
-# F32
-outFile.write('''
-//===--- 32-bit float -----------------------------------------------------===//
-''')
-for i in range(0, num_regs):
-  outFile.write('def F%d : NVPTXReg<"%%f%d">;\n' % (i, i))
-
-# F64
-outFile.write('''
-//===--- 64-bit float -----------------------------------------------------===//
-''')
-for i in range(0, num_regs):
-  outFile.write('def FL%d : NVPTXReg<"%%fl%d">;\n' % (i, i))
-
-# Vector registers
-outFile.write('''
-//===--- Vector -----------------------------------------------------------===//
-''')
-for i in range(0, num_regs):
-  outFile.write('def v2b8_%d : NVPTXReg<"%%v2b8_%d">;\n' % (i, i))
-for i in range(0, num_regs):
-  outFile.write('def v2b16_%d : NVPTXReg<"%%v2b16_%d">;\n' % (i, i))
-for i in range(0, num_regs):
-  outFile.write('def v2b32_%d : NVPTXReg<"%%v2b32_%d">;\n' % (i, i))
-for i in range(0, num_regs):
-  outFile.write('def v2b64_%d : NVPTXReg<"%%v2b64_%d">;\n' % (i, i))
-
-for i in range(0, num_regs):
-  outFile.write('def v4b8_%d : NVPTXReg<"%%v4b8_%d">;\n' % (i, i))
-for i in range(0, num_regs):
-  outFile.write('def v4b16_%d : NVPTXReg<"%%v4b16_%d">;\n' % (i, i))
-for i in range(0, num_regs):
-  outFile.write('def v4b32_%d : NVPTXReg<"%%v4b32_%d">;\n' % (i, i))
-
-# Argument registers
-outFile.write('''
-//===--- Arguments --------------------------------------------------------===//
-''')
-for i in range(0, num_regs):
-  outFile.write('def ia%d : NVPTXReg<"%%ia%d">;\n' % (i, i))
-for i in range(0, num_regs):
-  outFile.write('def la%d : NVPTXReg<"%%la%d">;\n' % (i, i))
-for i in range(0, num_regs):
-  outFile.write('def fa%d : NVPTXReg<"%%fa%d">;\n' % (i, i))
-for i in range(0, num_regs):
-  outFile.write('def da%d : NVPTXReg<"%%da%d">;\n' % (i, i))
-
-outFile.write('''
-//===----------------------------------------------------------------------===//
-//  Register classes
-//===----------------------------------------------------------------------===//
-''')
-
-outFile.write('def Int1Regs : NVPTXRegClass<[i1], 8, (add (sequence "P%%u", 0, %d))>;\n' % (num_regs-1))
-outFile.write('def Int8Regs : NVPTXRegClass<[i8], 8, (add (sequence "RC%%u", 0, %d))>;\n' % (num_regs-1))
-outFile.write('def Int16Regs : NVPTXRegClass<[i16], 16, (add (sequence "RS%%u", 0, %d))>;\n' % (num_regs-1))
-outFile.write('def Int32Regs : NVPTXRegClass<[i32], 32, (add (sequence "R%%u", 0, %d))>;\n' % (num_regs-1))
-outFile.write('def Int64Regs : NVPTXRegClass<[i64], 64, (add (sequence "RL%%u", 0, %d))>;\n' % (num_regs-1))
-
-outFile.write('def Float32Regs : NVPTXRegClass<[f32], 32, (add (sequence "F%%u", 0, %d))>;\n' % (num_regs-1))
-outFile.write('def Float64Regs : NVPTXRegClass<[f64], 64, (add (sequence "FL%%u", 0, %d))>;\n' % (num_regs-1))
-
-outFile.write('def Int32ArgRegs : NVPTXRegClass<[i32], 32, (add (sequence "ia%%u", 0, %d))>;\n' % (num_regs-1))
-outFile.write('def Int64ArgRegs : NVPTXRegClass<[i64], 64, (add (sequence "la%%u", 0, %d))>;\n' % (num_regs-1))
-outFile.write('def Float32ArgRegs : NVPTXRegClass<[f32], 32, (add (sequence "fa%%u", 0, %d))>;\n' % (num_regs-1))
-outFile.write('def Float64ArgRegs : NVPTXRegClass<[f64], 64, (add (sequence "da%%u", 0, %d))>;\n' % (num_regs-1))
-
-outFile.write('''
-// Read NVPTXRegisterInfo.cpp to see how VRFrame and VRDepot are used.
-def SpecialRegs : NVPTXRegClass<[i32], 32, (add VRFrame, VRDepot)>;
-''')
-
-outFile.write('''
-class NVPTXVecRegClass<list<ValueType> regTypes, int alignment, dag regList,
-                       NVPTXRegClass sClass,
-                       int e,
-                       string n>
-  : NVPTXRegClass<regTypes, alignment, regList>
-{
-  NVPTXRegClass scalarClass=sClass;
-  int elems=e;
-  string name=n;
-}
-''')
-
-
-outFile.write('def V2F32Regs\n  : NVPTXVecRegClass<[v2f32], 64, (add (sequence "v2b32_%%u", 0, %d)),\n    Float32Regs, 2, ".v2.f32">;\n' % (num_regs-1))
-outFile.write('def V4F32Regs\n  : NVPTXVecRegClass<[v4f32], 128, (add (sequence "v4b32_%%u", 0, %d)),\n    Float32Regs, 4, ".v4.f32">;\n' % (num_regs-1))
-
-outFile.write('def V2I32Regs\n  : NVPTXVecRegClass<[v2i32], 64, (add (sequence "v2b32_%%u", 0, %d)),\n    Int32Regs, 2, ".v2.u32">;\n' % (num_regs-1))
-outFile.write('def V4I32Regs\n  : NVPTXVecRegClass<[v4i32], 128, (add (sequence "v4b32_%%u", 0, %d)),\n    Int32Regs, 4, ".v4.u32">;\n' % (num_regs-1))
-
-outFile.write('def V2F64Regs\n  : NVPTXVecRegClass<[v2f64], 128, (add (sequence "v2b64_%%u", 0, %d)),\n    Float64Regs, 2, ".v2.f64">;\n' % (num_regs-1))
-outFile.write('def V2I64Regs\n  : NVPTXVecRegClass<[v2i64], 128, (add (sequence "v2b64_%%u", 0, %d)),\n    Int64Regs, 2, ".v2.u64">;\n' % (num_regs-1))
-
-outFile.write('def V2I16Regs\n  : NVPTXVecRegClass<[v2i16], 32, (add (sequence "v2b16_%%u", 0, %d)),\n    Int16Regs, 2, ".v2.u16">;\n' % (num_regs-1))
-outFile.write('def V4I16Regs\n  : NVPTXVecRegClass<[v4i16], 64, (add (sequence "v4b16_%%u", 0, %d)),\n    Int16Regs, 4, ".v4.u16">;\n' % (num_regs-1))
-
-outFile.write('def V2I8Regs\n  : NVPTXVecRegClass<[v2i8], 16, (add (sequence "v2b8_%%u", 0, %d)),\n    Int8Regs, 2, ".v2.u8">;\n' % (num_regs-1))
-outFile.write('def V4I8Regs\n  : NVPTXVecRegClass<[v4i8], 32, (add (sequence "v4b8_%%u", 0, %d)),\n    Int8Regs, 4, ".v4.u8">;\n' % (num_regs-1))
-
-outFile.close()
-
-
-outFile = open('NVPTXNumRegisters.h', 'w')
-outFile.write('''
-//===-- NVPTXNumRegisters.h - PTX Register Info ---------------------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef NVPTX_NUM_REGISTERS_H
-#define NVPTX_NUM_REGISTERS_H
-
-namespace llvm {
-
-const unsigned NVPTXNumRegisters = %d;
-
-}
-
-#endif
-''' % num_regs)
-
-outFile.close()
diff --git a/lib/Target/PowerPC/CMakeLists.txt b/lib/Target/PowerPC/CMakeLists.txt
index 192d18d66440..6036428fad93 100644
--- a/lib/Target/PowerPC/CMakeLists.txt
+++ b/lib/Target/PowerPC/CMakeLists.txt
@@ -26,6 +26,7 @@ add_llvm_target(PowerPCCodeGen
   PPCRegisterInfo.cpp
   PPCSubtarget.cpp
   PPCTargetMachine.cpp
+  PPCTargetTransformInfo.cpp
   PPCSelectionDAGInfo.cpp
   )
 
diff --git a/lib/Target/PowerPC/InstPrinter/PPCInstPrinter.cpp b/lib/Target/PowerPC/InstPrinter/PPCInstPrinter.cpp
index 3d583060d1ef..bacc108c62b4 100644
--- a/lib/Target/PowerPC/InstPrinter/PPCInstPrinter.cpp
+++ b/lib/Target/PowerPC/InstPrinter/PPCInstPrinter.cpp
@@ -13,7 +13,7 @@
 
 #define DEBUG_TYPE "asm-printer"
 #include "PPCInstPrinter.h"
-#include "MCTargetDesc/PPCBaseInfo.h"
+#include "MCTargetDesc/PPCMCTargetDesc.h"
 #include "MCTargetDesc/PPCPredicates.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
@@ -87,35 +87,9 @@ void PPCInstPrinter::printPredicateOperand(const MCInst *MI, unsigned OpNo,
                                            raw_ostream &O, 
                                            const char *Modifier) {
   unsigned Code = MI->getOperand(OpNo).getImm();
-  if (!Modifier) {
-    unsigned CCReg = MI->getOperand(OpNo+1).getReg();
-    unsigned RegNo;
-    switch (CCReg) {
-    default: llvm_unreachable("Unknown CR register");
-    case PPC::CR0: RegNo = 0; break;
-    case PPC::CR1: RegNo = 1; break;
-    case PPC::CR2: RegNo = 2; break;
-    case PPC::CR3: RegNo = 3; break;
-    case PPC::CR4: RegNo = 4; break;
-    case PPC::CR5: RegNo = 5; break;
-    case PPC::CR6: RegNo = 6; break;
-    case PPC::CR7: RegNo = 7; break;
-    }
-
-    // Print the CR bit number. The Code is ((BI << 5) | BO) for a
-    // BCC, but we must have the positive form here (BO == 12)
-    unsigned BI = Code >> 5;
-    assert((Code & 0xF) == 12 &&
-           "BO in predicate bit must have the positive form");
-
-    unsigned Value = 4*RegNo + BI;
-    O << Value;
-    return;
-  }
 
   if (StringRef(Modifier) == "cc") {
     switch ((PPC::Predicate)Code) {
-    case PPC::PRED_ALWAYS: return; // Don't print anything for always.
     case PPC::PRED_LT: O << "lt"; return;
     case PPC::PRED_LE: O << "le"; return;
     case PPC::PRED_EQ: O << "eq"; return;
@@ -129,8 +103,6 @@ void PPCInstPrinter::printPredicateOperand(const MCInst *MI, unsigned OpNo,
   
   assert(StringRef(Modifier) == "reg" &&
          "Need to specify 'cc' or 'reg' as predicate op modifier!");
-  // Don't print the register for 'always'.
-  if (Code == PPC::PRED_ALWAYS) return;
   printOperand(MI, OpNo+1, O);
 }
 
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp b/lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp
index 87ecb13a4c76..ec2657403e0c 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp
@@ -13,8 +13,8 @@
 #include "llvm/MC/MCELFObjectWriter.h"
 #include "llvm/MC/MCFixupKindInfo.h"
 #include "llvm/MC/MCMachObjectWriter.h"
-#include "llvm/MC/MCSectionMachO.h"
 #include "llvm/MC/MCObjectWriter.h"
+#include "llvm/MC/MCSectionMachO.h"
 #include "llvm/MC/MCValue.h"
 #include "llvm/Object/MachOFormat.h"
 #include "llvm/Support/ELF.h"
@@ -30,11 +30,9 @@ static unsigned adjustFixupValue(unsigned Kind, uint64_t Value) {
   case FK_Data_2:
   case FK_Data_4:
   case FK_Data_8:
-  case PPC::fixup_ppc_toc:
+  case PPC::fixup_ppc_tlsreg:
+  case PPC::fixup_ppc_nofixup:
     return Value;
-  case PPC::fixup_ppc_lo14:
-  case PPC::fixup_ppc_toc16_ds:
-    return (Value & 0xffff) << 2;
   case PPC::fixup_ppc_brcond14:
     return Value & 0xfffc;
   case PPC::fixup_ppc_br24:
@@ -46,8 +44,9 @@ static unsigned adjustFixupValue(unsigned Kind, uint64_t Value) {
   case PPC::fixup_ppc_ha16:
     return ((Value >> 16) + ((Value & 0x8000) ? 1 : 0)) & 0xffff;
   case PPC::fixup_ppc_lo16:
-  case PPC::fixup_ppc_toc16:
     return Value & 0xffff;
+  case PPC::fixup_ppc_lo16_ds:
+    return Value & 0xfffc;
   }
 }
 
@@ -61,7 +60,9 @@ public:
   void RecordRelocation(MachObjectWriter *Writer,
                         const MCAssembler &Asm, const MCAsmLayout &Layout,
                         const MCFragment *Fragment, const MCFixup &Fixup,
-                        MCValue Target, uint64_t &FixedValue) {}
+                        MCValue Target, uint64_t &FixedValue) {
+    llvm_unreachable("Relocation emission for MachO/PPC unimplemented!");
+  }
 };
 
 class PPCAsmBackend : public MCAsmBackend {
@@ -78,10 +79,9 @@ public:
       { "fixup_ppc_brcond14",    16,     14,   MCFixupKindInfo::FKF_IsPCRel },
       { "fixup_ppc_lo16",        16,     16,   0 },
       { "fixup_ppc_ha16",        16,     16,   0 },
-      { "fixup_ppc_lo14",        16,     14,   0 },
-      { "fixup_ppc_toc",          0,     64,   0 },
-      { "fixup_ppc_toc16",       16,     16,   0 },
-      { "fixup_ppc_toc16_ds",    16,     14,   0 }
+      { "fixup_ppc_lo16_ds",     16,     14,   0 },
+      { "fixup_ppc_tlsreg",       0,      0,   0 },
+      { "fixup_ppc_nofixup",      0,      0,   0 }
     };
 
     if (Kind < FirstTargetFixupKind)
@@ -92,6 +92,20 @@ public:
     return Infos[Kind - FirstTargetFixupKind];
   }
 
+  void applyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize,
+                  uint64_t Value) const {
+    Value = adjustFixupValue(Fixup.getKind(), Value);
+    if (!Value) return;           // Doesn't change encoding.
+
+    unsigned Offset = Fixup.getOffset();
+
+    // For each byte of the fragment that the fixup touches, mask in the bits
+    // from the fixup value. The Value has been "split up" into the appropriate
+    // bitfields above.
+    for (unsigned i = 0; i != 4; ++i)
+      Data[Offset + i] |= uint8_t((Value >> ((4 - i - 1)*8)) & 0xff);
+  }
+
   bool mayNeedRelaxation(const MCInst &Inst) const {
     // FIXME.
     return false;
@@ -99,7 +113,7 @@ public:
 
   bool fixupNeedsRelaxation(const MCFixup &Fixup,
                             uint64_t Value,
-                            const MCInstFragment *DF,
+                            const MCRelaxableFragment *DF,
                             const MCAsmLayout &Layout) const {
     // FIXME.
     llvm_unreachable("relaxInstruction() unimplemented");
@@ -135,11 +149,6 @@ namespace {
   public:
     DarwinPPCAsmBackend(const Target &T) : PPCAsmBackend(T) { }
 
-    void applyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize,
-                    uint64_t Value) const {
-      llvm_unreachable("UNIMP");
-    }
-
     MCObjectWriter *createObjectWriter(raw_ostream &OS) const {
       bool is64 = getPointerSize() == 8;
       return createMachObjectWriter(new PPCMachObjectWriter(
@@ -161,19 +170,6 @@ namespace {
     ELFPPCAsmBackend(const Target &T, uint8_t OSABI) :
       PPCAsmBackend(T), OSABI(OSABI) { }
 
-    void applyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize,
-                    uint64_t Value) const {
-      Value = adjustFixupValue(Fixup.getKind(), Value);
-      if (!Value) return;           // Doesn't change encoding.
-
-      unsigned Offset = Fixup.getOffset();
-
-      // For each byte of the fragment that the fixup touches, mask in the bits from
-      // the fixup value. The Value has been "split up" into the appropriate
-      // bitfields above.
-      for (unsigned i = 0; i != 4; ++i)
-        Data[Offset + i] |= uint8_t((Value >> ((4 - i - 1)*8)) & 0xff);
-    }
 
     MCObjectWriter *createObjectWriter(raw_ostream &OS) const {
       bool is64 = getPointerSize() == 8;
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCBaseInfo.h b/lib/Target/PowerPC/MCTargetDesc/PPCBaseInfo.h
deleted file mode 100644
index 9c975c089ea6..000000000000
--- a/lib/Target/PowerPC/MCTargetDesc/PPCBaseInfo.h
+++ /dev/null
@@ -1,70 +0,0 @@
-//===-- PPCBaseInfo.h - Top level definitions for PPC -----------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file contains small standalone helper functions and enum definitions for
-// the PPC target useful for the compiler back-end and the MC libraries.
-// As such, it deliberately does not include references to LLVM core
-// code gen types, passes, etc..
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef PPCBASEINFO_H
-#define PPCBASEINFO_H
-
-#include "PPCMCTargetDesc.h"
-#include "llvm/Support/ErrorHandling.h"
-
-namespace llvm {
-
-/// getPPCRegisterNumbering - Given the enum value for some register, e.g.
-/// PPC::F14, return the number that it corresponds to (e.g. 14).
-inline static unsigned getPPCRegisterNumbering(unsigned RegEnum) {
-  using namespace PPC;
-  switch (RegEnum) {
-  case 0: return 0;
-  case R0 :  case X0 :  case F0 :  case V0 : case CR0:  case CR0LT: return  0;
-  case R1 :  case X1 :  case F1 :  case V1 : case CR1:  case CR0GT: return  1;
-  case R2 :  case X2 :  case F2 :  case V2 : case CR2:  case CR0EQ: return  2;
-  case R3 :  case X3 :  case F3 :  case V3 : case CR3:  case CR0UN: return  3;
-  case R4 :  case X4 :  case F4 :  case V4 : case CR4:  case CR1LT: return  4;
-  case R5 :  case X5 :  case F5 :  case V5 : case CR5:  case CR1GT: return  5;
-  case R6 :  case X6 :  case F6 :  case V6 : case CR6:  case CR1EQ: return  6;
-  case R7 :  case X7 :  case F7 :  case V7 : case CR7:  case CR1UN: return  7;
-  case R8 :  case X8 :  case F8 :  case V8 : case CR2LT: return  8;
-  case R9 :  case X9 :  case F9 :  case V9 : case CR2GT: return  9;
-  case R10:  case X10:  case F10:  case V10: case CR2EQ: return 10;
-  case R11:  case X11:  case F11:  case V11: case CR2UN: return 11;
-  case R12:  case X12:  case F12:  case V12: case CR3LT: return 12;
-  case R13:  case X13:  case F13:  case V13: case CR3GT: return 13;
-  case R14:  case X14:  case F14:  case V14: case CR3EQ: return 14;
-  case R15:  case X15:  case F15:  case V15: case CR3UN: return 15;
-  case R16:  case X16:  case F16:  case V16: case CR4LT: return 16;
-  case R17:  case X17:  case F17:  case V17: case CR4GT: return 17;
-  case R18:  case X18:  case F18:  case V18: case CR4EQ: return 18;
-  case R19:  case X19:  case F19:  case V19: case CR4UN: return 19;
-  case R20:  case X20:  case F20:  case V20: case CR5LT: return 20;
-  case R21:  case X21:  case F21:  case V21: case CR5GT: return 21;
-  case R22:  case X22:  case F22:  case V22: case CR5EQ: return 22;
-  case R23:  case X23:  case F23:  case V23: case CR5UN: return 23;
-  case R24:  case X24:  case F24:  case V24: case CR6LT: return 24;
-  case R25:  case X25:  case F25:  case V25: case CR6GT: return 25;
-  case R26:  case X26:  case F26:  case V26: case CR6EQ: return 26;
-  case R27:  case X27:  case F27:  case V27: case CR6UN: return 27;
-  case R28:  case X28:  case F28:  case V28: case CR7LT: return 28;
-  case R29:  case X29:  case F29:  case V29: case CR7GT: return 29;
-  case R30:  case X30:  case F30:  case V30: case CR7EQ: return 30;
-  case R31:  case X31:  case F31:  case V31: case CR7UN: return 31;
-  default:
-    llvm_unreachable("Unhandled reg in PPCRegisterInfo::getRegisterNumbering!");
-  }
-}
-
-} // end namespace llvm;
-
-#endif
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCELFObjectWriter.cpp b/lib/Target/PowerPC/MCTargetDesc/PPCELFObjectWriter.cpp
index dc93f7124a52..84e4175e635b 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCELFObjectWriter.cpp
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCELFObjectWriter.cpp
@@ -7,12 +7,13 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "MCTargetDesc/PPCFixupKinds.h"
 #include "MCTargetDesc/PPCMCTargetDesc.h"
+#include "MCTargetDesc/PPCFixupKinds.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/MC/MCELFObjectWriter.h"
-#include "llvm/Support/ErrorHandling.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCValue.h"
+#include "llvm/Support/ErrorHandling.h"
 
 using namespace llvm;
 
@@ -33,9 +34,25 @@ namespace {
                                                     const MCFixup &Fixup,
                                                     bool IsPCRel) const;
     virtual void adjustFixupOffset(const MCFixup &Fixup, uint64_t &RelocOffset);
+
+    virtual void sortRelocs(const MCAssembler &Asm,
+                            std::vector<ELFRelocationEntry> &Relocs);
+  };
+
+  class PPCELFRelocationEntry : public ELFRelocationEntry {
+  public:
+    PPCELFRelocationEntry(const ELFRelocationEntry &RE);
+    bool operator<(const PPCELFRelocationEntry &RE) const {
+      return (RE.r_offset < r_offset ||
+              (RE.r_offset == r_offset && RE.Type > Type));
+    }
   };
 }
 
+PPCELFRelocationEntry::PPCELFRelocationEntry(const ELFRelocationEntry &RE)
+  : ELFRelocationEntry(RE.r_offset, RE.Index, RE.Type, RE.Symbol,
+                       RE.r_addend, *RE.Fixup) {}
+
 PPCELFObjectWriter::PPCELFObjectWriter(bool Is64Bit, uint8_t OSABI)
   : MCELFObjectTargetWriter(Is64Bit, OSABI,
                             Is64Bit ?  ELF::EM_PPC64 : ELF::EM_PPC,
@@ -60,9 +77,14 @@ unsigned PPCELFObjectWriter::getRelocTypeInner(const MCValue &Target,
     case PPC::fixup_ppc_br24:
       Type = ELF::R_PPC_REL24;
       break;
+    case FK_Data_4:
     case FK_PCRel_4:
       Type = ELF::R_PPC_REL32;
       break;
+    case FK_Data_8:
+    case FK_PCRel_8:
+      Type = ELF::R_PPC64_REL64;
+      break;
     }
   } else {
     switch ((unsigned)Fixup.getKind()) {
@@ -79,9 +101,24 @@ unsigned PPCELFObjectWriter::getRelocTypeInner(const MCValue &Target,
       case MCSymbolRefExpr::VK_PPC_TPREL16_HA:
         Type = ELF::R_PPC_TPREL16_HA;
         break;
+      case MCSymbolRefExpr::VK_PPC_DTPREL16_HA:
+        Type = ELF::R_PPC64_DTPREL16_HA;
+        break;
       case MCSymbolRefExpr::VK_None:
         Type = ELF::R_PPC_ADDR16_HA;
 	break;
+      case MCSymbolRefExpr::VK_PPC_TOC16_HA:
+        Type = ELF::R_PPC64_TOC16_HA;
+        break;
+      case MCSymbolRefExpr::VK_PPC_GOT_TPREL16_HA:
+        Type = ELF::R_PPC64_GOT_TPREL16_HA;
+        break;
+      case MCSymbolRefExpr::VK_PPC_GOT_TLSGD16_HA:
+        Type = ELF::R_PPC64_GOT_TLSGD16_HA;
+        break;
+      case MCSymbolRefExpr::VK_PPC_GOT_TLSLD16_HA:
+        Type = ELF::R_PPC64_GOT_TLSLD16_HA;
+        break;
       }
       break;
     case PPC::fixup_ppc_lo16:
@@ -90,22 +127,56 @@ unsigned PPCELFObjectWriter::getRelocTypeInner(const MCValue &Target,
       case MCSymbolRefExpr::VK_PPC_TPREL16_LO:
         Type = ELF::R_PPC_TPREL16_LO;
         break;
+      case MCSymbolRefExpr::VK_PPC_DTPREL16_LO:
+        Type = ELF::R_PPC64_DTPREL16_LO;
+        break;
       case MCSymbolRefExpr::VK_None:
         Type = ELF::R_PPC_ADDR16_LO;
 	break;
+      case MCSymbolRefExpr::VK_PPC_TOC_ENTRY:
+        Type = ELF::R_PPC64_TOC16;
+        break;
+      case MCSymbolRefExpr::VK_PPC_TOC16_LO:
+        Type = ELF::R_PPC64_TOC16_LO;
+        break;
+      case MCSymbolRefExpr::VK_PPC_GOT_TLSGD16_LO:
+        Type = ELF::R_PPC64_GOT_TLSGD16_LO;
+        break;
+      case MCSymbolRefExpr::VK_PPC_GOT_TLSLD16_LO:
+        Type = ELF::R_PPC64_GOT_TLSLD16_LO;
+        break;
       }
       break;
-    case PPC::fixup_ppc_lo14:
-      Type = ELF::R_PPC_ADDR14;
-      break;
-    case PPC::fixup_ppc_toc:
-      Type = ELF::R_PPC64_TOC;
+    case PPC::fixup_ppc_lo16_ds:
+      switch (Modifier) {
+      default: llvm_unreachable("Unsupported Modifier");
+      case MCSymbolRefExpr::VK_None:
+        Type = ELF::R_PPC64_ADDR16_DS;
+        break;
+      case MCSymbolRefExpr::VK_PPC_TOC_ENTRY:
+        Type = ELF::R_PPC64_TOC16_DS;
+	break;
+      case MCSymbolRefExpr::VK_PPC_TOC16_LO:
+        Type = ELF::R_PPC64_TOC16_LO_DS;
+        break;
+      case MCSymbolRefExpr::VK_PPC_GOT_TPREL16_LO:
+        Type = ELF::R_PPC64_GOT_TPREL16_LO_DS;
+        break;
+      }
       break;
-    case PPC::fixup_ppc_toc16:
-      Type = ELF::R_PPC64_TOC16;
+    case PPC::fixup_ppc_tlsreg:
+      Type = ELF::R_PPC64_TLS;
       break;
-    case PPC::fixup_ppc_toc16_ds:
-      Type = ELF::R_PPC64_TOC16_DS;
+    case PPC::fixup_ppc_nofixup:
+      switch (Modifier) {
+      default: llvm_unreachable("Unsupported Modifier");
+      case MCSymbolRefExpr::VK_PPC_TLSGD:
+        Type = ELF::R_PPC64_TLSGD;
+        break;
+      case MCSymbolRefExpr::VK_PPC_TLSLD:
+        Type = ELF::R_PPC64_TLSLD;
+        break;
+      }
       break;
     case FK_Data_8:
       switch (Modifier) {
@@ -162,8 +233,7 @@ adjustFixupOffset(const MCFixup &Fixup, uint64_t &RelocOffset) {
   switch ((unsigned)Fixup.getKind()) {
     case PPC::fixup_ppc_ha16:
     case PPC::fixup_ppc_lo16:
-    case PPC::fixup_ppc_toc16:
-    case PPC::fixup_ppc_toc16_ds:
+    case PPC::fixup_ppc_lo16_ds:
       RelocOffset += 2;
       break;
     default:
@@ -171,6 +241,34 @@ adjustFixupOffset(const MCFixup &Fixup, uint64_t &RelocOffset) {
   }
 }
 
+// The standard sorter only sorts on the r_offset field, but PowerPC can
+// have multiple relocations at the same offset.  Sort secondarily on the
+// relocation type to avoid nondeterminism.
+void PPCELFObjectWriter::sortRelocs(const MCAssembler &Asm,
+                                    std::vector<ELFRelocationEntry> &Relocs) {
+
+  // Copy to a temporary vector of relocation entries having a different
+  // sort function.
+  std::vector<PPCELFRelocationEntry> TmpRelocs;
+  
+  for (std::vector<ELFRelocationEntry>::iterator R = Relocs.begin();
+       R != Relocs.end(); ++R) {
+    TmpRelocs.push_back(PPCELFRelocationEntry(*R));
+  }
+
+  // Sort in place by ascending r_offset and descending r_type.
+  array_pod_sort(TmpRelocs.begin(), TmpRelocs.end());
+
+  // Copy back to the original vector.
+  unsigned I = 0;
+  for (std::vector<PPCELFRelocationEntry>::iterator R = TmpRelocs.begin();
+       R != TmpRelocs.end(); ++R, ++I) {
+    Relocs[I] = ELFRelocationEntry(R->r_offset, R->Index, R->Type,
+                                   R->Symbol, R->r_addend, *R->Fixup);
+  }
+}
+
+
 MCObjectWriter *llvm::createPPCELFObjectWriter(raw_ostream &OS,
                                                bool Is64Bit,
                                                uint8_t OSABI) {
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCFixupKinds.h b/lib/Target/PowerPC/MCTargetDesc/PPCFixupKinds.h
index 37b265e7fd38..86c44f57a5e2 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCFixupKinds.h
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCFixupKinds.h
@@ -12,6 +12,8 @@
 
 #include "llvm/MC/MCFixup.h"
 
+#undef PPC
+
 namespace llvm {
 namespace PPC {
 enum Fixups {
@@ -31,19 +33,16 @@ enum Fixups {
   /// like 'lis'.
   fixup_ppc_ha16,
   
-  /// fixup_ppc_lo14 - A 14-bit fixup corresponding to lo16(_foo) for instrs
-  /// like 'std'.
-  fixup_ppc_lo14,
-
-  /// fixup_ppc_toc - Insert value of TOC base (.TOC.).
-  fixup_ppc_toc,
+  /// fixup_ppc_lo16_ds - A 14-bit fixup corresponding to lo16(_foo) with
+  /// implied 2 zero bits for instrs like 'std'.
+  fixup_ppc_lo16_ds,
 
-  /// fixup_ppc_toc16 - A 16-bit signed fixup relative to the TOC base.
-  fixup_ppc_toc16,
+  /// fixup_ppc_tlsreg - Insert thread-pointer register number.
+  fixup_ppc_tlsreg,
 
-  /// fixup_ppc_toc16_ds - A 14-bit signed fixup relative to the TOC base with
-  /// implied 2 zero bits
-  fixup_ppc_toc16_ds,
+  /// fixup_ppc_nofixup - Not a true fixup, but ties a symbol to a call
+  /// to __tls_get_addr for the TLS general and local dynamic models.
+  fixup_ppc_nofixup,
   
   // Marker
   LastTargetFixupKind,
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCMCAsmInfo.cpp b/lib/Target/PowerPC/MCTargetDesc/PPCMCAsmInfo.cpp
index 215aa40c4afd..a25d7fe64f3a 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCMCAsmInfo.cpp
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCMCAsmInfo.cpp
@@ -17,8 +17,9 @@ using namespace llvm;
 void PPCMCAsmInfoDarwin::anchor() { }
 
 PPCMCAsmInfoDarwin::PPCMCAsmInfoDarwin(bool is64Bit) {
-  if (is64Bit)
-    PointerSize = 8;
+  if (is64Bit) {
+    PointerSize = CalleeSaveStackSlotSize = 8;
+  }
   IsLittleEndian = false;
 
   PCSymbol = ".";
@@ -35,8 +36,9 @@ PPCMCAsmInfoDarwin::PPCMCAsmInfoDarwin(bool is64Bit) {
 void PPCLinuxMCAsmInfo::anchor() { }
 
 PPCLinuxMCAsmInfo::PPCLinuxMCAsmInfo(bool is64Bit) {
-  if (is64Bit)
-    PointerSize = 8;
+  if (is64Bit) {
+    PointerSize = CalleeSaveStackSlotSize = 8;
+  }
   IsLittleEndian = false;
 
   // ".comm align is in bytes but .align is pow-2."
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCMCCodeEmitter.cpp b/lib/Target/PowerPC/MCTargetDesc/PPCMCCodeEmitter.cpp
index 21183024a509..2223cd623cb5 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCMCCodeEmitter.cpp
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCMCCodeEmitter.cpp
@@ -12,15 +12,17 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "mccodeemitter"
-#include "MCTargetDesc/PPCBaseInfo.h"
+#include "MCTargetDesc/PPCMCTargetDesc.h"
 #include "MCTargetDesc/PPCFixupKinds.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/MC/MCCodeEmitter.h"
-#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
-#include "llvm/ADT/Statistic.h"
-#include "llvm/Support/raw_ostream.h"
+#include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
 using namespace llvm;
 
 STATISTIC(MCNumEmitted, "Number of MC instructions emitted");
@@ -31,24 +33,17 @@ class PPCMCCodeEmitter : public MCCodeEmitter {
   void operator=(const PPCMCCodeEmitter &) LLVM_DELETED_FUNCTION;
 
   const MCSubtargetInfo &STI;
+  const MCContext &CTX;
   Triple TT;
 
 public:
   PPCMCCodeEmitter(const MCInstrInfo &mcii, const MCSubtargetInfo &sti,
                    MCContext &ctx)
-    : STI(sti), TT(STI.getTargetTriple()) {
+    : STI(sti), CTX(ctx), TT(STI.getTargetTriple()) {
   }
   
   ~PPCMCCodeEmitter() {}
 
-  bool is64BitMode() const {
-    return (STI.getFeatureBits() & PPC::Feature64Bit) != 0;
-  }
-
-  bool isSVR4ABI() const {
-    return TT.isMacOSX() == 0;
-  }
-
   unsigned getDirectBrEncoding(const MCInst &MI, unsigned OpNo,
                                SmallVectorImpl<MCFixup> &Fixups) const;
   unsigned getCondBrEncoding(const MCInst &MI, unsigned OpNo,
@@ -61,6 +56,8 @@ public:
                             SmallVectorImpl<MCFixup> &Fixups) const;
   unsigned getMemRIXEncoding(const MCInst &MI, unsigned OpNo,
                              SmallVectorImpl<MCFixup> &Fixups) const;
+  unsigned getTLSRegEncoding(const MCInst &MI, unsigned OpNo,
+                             SmallVectorImpl<MCFixup> &Fixups) const;
   unsigned get_crbitm_encoding(const MCInst &MI, unsigned OpNo,
                                SmallVectorImpl<MCFixup> &Fixups) const;
 
@@ -77,11 +74,11 @@ public:
                          SmallVectorImpl<MCFixup> &Fixups) const {
     uint64_t Bits = getBinaryCodeForInstr(MI, Fixups);
 
-    // BL8_NOPELF and BLA8_NOP_ELF is both size of 8 bacause of the
-    // following 'nop'.
+    // BL8_NOP etc. all have a size of 8 because of the following 'nop'.
     unsigned Size = 4; // FIXME: Have Desc.getSize() return the correct value!
     unsigned Opcode = MI.getOpcode();
-    if (Opcode == PPC::BL8_NOP_ELF || Opcode == PPC::BLA8_NOP_ELF)
+    if (Opcode == PPC::BL8_NOP || Opcode == PPC::BLA8_NOP ||
+        Opcode == PPC::BL8_NOP_TLSGD || Opcode == PPC::BL8_NOP_TLSLD)
       Size = 8;
     
     // Output the constant in big endian byte order.
@@ -114,6 +111,17 @@ getDirectBrEncoding(const MCInst &MI, unsigned OpNo,
   // Add a fixup for the branch target.
   Fixups.push_back(MCFixup::Create(0, MO.getExpr(),
                                    (MCFixupKind)PPC::fixup_ppc_br24));
+
+  // For special TLS calls, add another fixup for the symbol.  Apparently
+  // BL8_NOP, BL8_NOP_TLSGD, and BL8_NOP_TLSLD are sufficiently
+  // similar that TblGen will not generate a separate case for the latter
+  // two, so this is the only way to get the extra fixup generated.
+  unsigned Opcode = MI.getOpcode();
+  if (Opcode == PPC::BL8_NOP_TLSGD || Opcode == PPC::BL8_NOP_TLSLD) {
+    const MCOperand &MO2 = MI.getOperand(OpNo+1);
+    Fixups.push_back(MCFixup::Create(0, MO2.getExpr(),
+                                     (MCFixupKind)PPC::fixup_ppc_nofixup));
+  }
   return 0;
 }
 
@@ -162,12 +170,8 @@ unsigned PPCMCCodeEmitter::getMemRIEncoding(const MCInst &MI, unsigned OpNo,
     return (getMachineOpValue(MI, MO, Fixups) & 0xFFFF) | RegBits;
   
   // Add a fixup for the displacement field.
-  if (isSVR4ABI() && is64BitMode())
-    Fixups.push_back(MCFixup::Create(0, MO.getExpr(),
-                                     (MCFixupKind)PPC::fixup_ppc_toc16));
-  else
-    Fixups.push_back(MCFixup::Create(0, MO.getExpr(),
-                                     (MCFixupKind)PPC::fixup_ppc_lo16));
+  Fixups.push_back(MCFixup::Create(0, MO.getExpr(),
+                                   (MCFixupKind)PPC::fixup_ppc_lo16));
   return RegBits;
 }
 
@@ -183,17 +187,26 @@ unsigned PPCMCCodeEmitter::getMemRIXEncoding(const MCInst &MI, unsigned OpNo,
   if (MO.isImm())
     return (getMachineOpValue(MI, MO, Fixups) & 0x3FFF) | RegBits;
   
-  // Add a fixup for the branch target.
-  if (isSVR4ABI() && is64BitMode())
-    Fixups.push_back(MCFixup::Create(0, MO.getExpr(),
-                                     (MCFixupKind)PPC::fixup_ppc_toc16_ds));
-  else
-    Fixups.push_back(MCFixup::Create(0, MO.getExpr(),
-                                     (MCFixupKind)PPC::fixup_ppc_lo14));
+  // Add a fixup for the displacement field.
+  Fixups.push_back(MCFixup::Create(0, MO.getExpr(),
+                                   (MCFixupKind)PPC::fixup_ppc_lo16_ds));
   return RegBits;
 }
 
 
+unsigned PPCMCCodeEmitter::getTLSRegEncoding(const MCInst &MI, unsigned OpNo,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+  const MCOperand &MO = MI.getOperand(OpNo);
+  if (MO.isReg()) return getMachineOpValue(MI, MO, Fixups);
+  
+  // Add a fixup for the TLS register, which simply provides a relocation
+  // hint to the linker that this statement is part of a relocation sequence.
+  // Return the thread-pointer register's encoding.
+  Fixups.push_back(MCFixup::Create(0, MO.getExpr(),
+                                   (MCFixupKind)PPC::fixup_ppc_tlsreg));
+  return CTX.getRegisterInfo().getEncodingValue(PPC::X13);
+}
+
 unsigned PPCMCCodeEmitter::
 get_crbitm_encoding(const MCInst &MI, unsigned OpNo,
                     SmallVectorImpl<MCFixup> &Fixups) const {
@@ -202,7 +215,7 @@ get_crbitm_encoding(const MCInst &MI, unsigned OpNo,
           MI.getOpcode() == PPC::MFOCRF ||
           MI.getOpcode() == PPC::MTCRF8) &&
          (MO.getReg() >= PPC::CR0 && MO.getReg() <= PPC::CR7));
-  return 0x80 >> getPPCRegisterNumbering(MO.getReg());
+  return 0x80 >> CTX.getRegisterInfo().getEncodingValue(MO.getReg());
 }
 
 
@@ -214,7 +227,7 @@ getMachineOpValue(const MCInst &MI, const MCOperand &MO,
     // The GPR operand should come through here though.
     assert((MI.getOpcode() != PPC::MTCRF && MI.getOpcode() != PPC::MFOCRF) ||
            MO.getReg() < PPC::CR0 || MO.getReg() > PPC::CR7);
-    return getPPCRegisterNumbering(MO.getReg());
+    return CTX.getRegisterInfo().getEncodingValue(MO.getReg());
   }
   
   assert(MO.isImm() &&
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.cpp b/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.cpp
index 4c2578d5dc53..2209f936ec33 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.cpp
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.cpp
@@ -12,14 +12,14 @@
 //===----------------------------------------------------------------------===//
 
 #include "PPCMCTargetDesc.h"
-#include "PPCMCAsmInfo.h"
 #include "InstPrinter/PPCInstPrinter.h"
-#include "llvm/MC/MachineLocation.h"
+#include "PPCMCAsmInfo.h"
 #include "llvm/MC/MCCodeGenInfo.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/MC/MachineLocation.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
 
@@ -88,6 +88,11 @@ static MCCodeGenInfo *createPPCMCCodeGenInfo(StringRef TT, Reloc::Model RM,
     else
       RM = Reloc::Static;
   }
+  if (CM == CodeModel::Default) {
+    Triple T(TT);
+    if (!T.isOSDarwin() && T.getArch() == Triple::ppc64)
+      CM = CodeModel::Medium;
+  }
   X->InitMCCodeGenInfo(RM, CM, OL);
   return X;
 }
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.h b/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.h
index a0e4cf3005f2..38a7420d972d 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.h
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.h
@@ -14,6 +14,9 @@
 #ifndef PPCMCTARGETDESC_H
 #define PPCMCTARGETDESC_H
 
+// GCC #defines PPC on Linux but we use it as our namespace name
+#undef PPC
+
 #include "llvm/Support/DataTypes.h"
 
 namespace llvm {
@@ -44,6 +47,10 @@ MCObjectWriter *createPPCELFObjectWriter(raw_ostream &OS,
                                          uint8_t OSABI);
 } // End llvm namespace
 
+// Generated files will use "namespace PPC". To avoid symbol clash,
+// undefine PPC here. PPC may be predefined on some hosts.
+#undef PPC
+
 // Defines symbolic names for PowerPC registers.  This defines a mapping from
 // register name to register number.
 //
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.cpp b/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.cpp
index 12bb0a143406..d84eb9c6aa03 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.cpp
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.cpp
@@ -18,7 +18,6 @@ using namespace llvm;
 
 PPC::Predicate PPC::InvertPredicate(PPC::Predicate Opcode) {
   switch (Opcode) {
-  default: llvm_unreachable("Unknown PPC branch opcode!");
   case PPC::PRED_EQ: return PPC::PRED_NE;
   case PPC::PRED_NE: return PPC::PRED_EQ;
   case PPC::PRED_LT: return PPC::PRED_GE;
@@ -28,4 +27,5 @@ PPC::Predicate PPC::InvertPredicate(PPC::Predicate Opcode) {
   case PPC::PRED_NU: return PPC::PRED_UN;
   case PPC::PRED_UN: return PPC::PRED_NU;
   }
+  llvm_unreachable("Unknown PPC branch opcode!");
 }
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.h b/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.h
index f872e861bfa7..ad2b01812816 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.h
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.h
@@ -14,11 +14,17 @@
 #ifndef LLVM_TARGET_POWERPC_PPCPREDICATES_H
 #define LLVM_TARGET_POWERPC_PPCPREDICATES_H
 
+// GCC #defines PPC on Linux but we use it as our namespace name
+#undef PPC
+
+// Generated files will use "namespace PPC". To avoid symbol clash,
+// undefine PPC here. PPC may be predefined on some hosts.
+#undef PPC
+
 namespace llvm {
 namespace PPC {
   /// Predicate - These are "(BI << 5) | BO"  for various predicates.
   enum Predicate {
-    PRED_ALWAYS = (0 << 5) | 20,
     PRED_LT     = (0 << 5) | 12,
     PRED_LE     = (1 << 5) |  4,
     PRED_EQ     = (2 << 5) | 12,
diff --git a/lib/Target/PowerPC/PPC.h b/lib/Target/PowerPC/PPC.h
index 9103e1232505..446b6854fb5b 100644
--- a/lib/Target/PowerPC/PPC.h
+++ b/lib/Target/PowerPC/PPC.h
@@ -15,7 +15,6 @@
 #ifndef LLVM_TARGET_POWERPC_H
 #define LLVM_TARGET_POWERPC_H
 
-#include "MCTargetDesc/PPCBaseInfo.h"
 #include "MCTargetDesc/PPCMCTargetDesc.h"
 #include <string>
 
@@ -25,6 +24,7 @@
 namespace llvm {
   class PPCTargetMachine;
   class FunctionPass;
+  class ImmutablePass;
   class JITCodeEmitter;
   class MachineInstr;
   class AsmPrinter;
@@ -37,6 +37,9 @@ namespace llvm {
                                             JITCodeEmitter &MCE);
   void LowerPPCMachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI,
                                     AsmPrinter &AP, bool isDarwin);
+
+  /// \brief Creates an PPC-specific Target Transformation Info pass.
+  ImmutablePass *createPPCTargetTransformInfoPass(const PPCTargetMachine *TM);
   
   namespace PPCII {
     
@@ -53,25 +56,32 @@ namespace llvm {
     
     /// MO_PIC_FLAG - If this bit is set, the symbol reference is relative to
     /// the function's picbase, e.g. lo16(symbol-picbase).
-    MO_PIC_FLAG = 4,
+    MO_PIC_FLAG = 2,
 
     /// MO_NLP_FLAG - If this bit is set, the symbol reference is actually to
     /// the non_lazy_ptr for the global, e.g. lo16(symbol$non_lazy_ptr-picbase).
-    MO_NLP_FLAG = 8,
+    MO_NLP_FLAG = 4,
     
     /// MO_NLP_HIDDEN_FLAG - If this bit is set, the symbol reference is to a
     /// symbol with hidden visibility.  This causes a different kind of
     /// non-lazy-pointer to be generated.
-    MO_NLP_HIDDEN_FLAG = 16,
+    MO_NLP_HIDDEN_FLAG = 8,
 
     /// The next are not flags but distinct values.
-    MO_ACCESS_MASK = 224,
+    MO_ACCESS_MASK = 0xf0,
 
     /// MO_LO16, MO_HA16 - lo16(symbol) and ha16(symbol)
-    MO_LO16 = 32, MO_HA16 = 64,
+    MO_LO16 = 1 << 4,
+    MO_HA16 = 2 << 4,
+
+    MO_TPREL16_HA = 3 << 4,
+    MO_TPREL16_LO = 4 << 4,
 
-    MO_TPREL16_HA = 96,
-    MO_TPREL16_LO = 128
+    /// These values identify relocations on immediates folded
+    /// into memory operations.
+    MO_DTPREL16_LO = 5 << 4,
+    MO_TLSLD16_LO  = 6 << 4,
+    MO_TOC16_LO    = 7 << 4
   };
   } // end namespace PPCII
   
diff --git a/lib/Target/PowerPC/PPC.td b/lib/Target/PowerPC/PPC.td
index cb15dadb7e99..389216278ee4 100644
--- a/lib/Target/PowerPC/PPC.td
+++ b/lib/Target/PowerPC/PPC.td
@@ -39,7 +39,12 @@ def DirectiveE500mc : SubtargetFeature<"", "DarwinDirective",
                                        "PPC::DIR_E500mc", "">;
 def DirectiveE5500  : SubtargetFeature<"", "DarwinDirective", 
                                        "PPC::DIR_E5500", "">;
+def DirectivePwr3: SubtargetFeature<"", "DarwinDirective", "PPC::DIR_PWR3", "">;
+def DirectivePwr4: SubtargetFeature<"", "DarwinDirective", "PPC::DIR_PWR4", "">;
+def DirectivePwr5: SubtargetFeature<"", "DarwinDirective", "PPC::DIR_PWR5", "">;
+def DirectivePwr5x: SubtargetFeature<"", "DarwinDirective", "PPC::DIR_PWR5X", "">;
 def DirectivePwr6: SubtargetFeature<"", "DarwinDirective", "PPC::DIR_PWR6", "">;
+def DirectivePwr6x: SubtargetFeature<"", "DarwinDirective", "PPC::DIR_PWR6X", "">;
 def DirectivePwr7: SubtargetFeature<"", "DarwinDirective", "PPC::DIR_PWR7", "">;
 
 def Feature64Bit     : SubtargetFeature<"64bit","Has64BitSupport", "true",
@@ -52,12 +57,42 @@ def FeatureMFOCRF    : SubtargetFeature<"mfocrf","HasMFOCRF", "true",
                                         "Enable the MFOCRF instruction">;
 def FeatureFSqrt     : SubtargetFeature<"fsqrt","HasFSQRT", "true",
                                         "Enable the fsqrt instruction">;
+def FeatureFRE       : SubtargetFeature<"fre", "HasFRE", "true",
+                                        "Enable the fre instruction">;
+def FeatureFRES      : SubtargetFeature<"fres", "HasFRES", "true",
+                                        "Enable the fres instruction">;
+def FeatureFRSQRTE   : SubtargetFeature<"frsqrte", "HasFRSQRTE", "true",
+                                        "Enable the frsqrte instruction">;
+def FeatureFRSQRTES  : SubtargetFeature<"frsqrtes", "HasFRSQRTES", "true",
+                                        "Enable the frsqrtes instruction">;
+def FeatureRecipPrec : SubtargetFeature<"recipprec", "HasRecipPrec", "true",
+                              "Assume higher precision reciprocal estimates">;
 def FeatureSTFIWX    : SubtargetFeature<"stfiwx","HasSTFIWX", "true",
                                         "Enable the stfiwx instruction">;
+def FeatureLFIWAX    : SubtargetFeature<"lfiwax","HasLFIWAX", "true",
+                                        "Enable the lfiwax instruction">;
+def FeatureFPRND     : SubtargetFeature<"fprnd", "HasFPRND", "true",
+                                        "Enable the fri[mnpz] instructions">;
+def FeatureFPCVT     : SubtargetFeature<"fpcvt", "HasFPCVT", "true",
+  "Enable fc[ft]* (unsigned and single-precision) and lfiwzx instructions">;
 def FeatureISEL      : SubtargetFeature<"isel","HasISEL", "true",
                                         "Enable the isel instruction">;
+def FeaturePOPCNTD   : SubtargetFeature<"popcntd","HasPOPCNTD", "true",
+                                        "Enable the popcnt[dw] instructions">;
+def FeatureLDBRX     : SubtargetFeature<"ldbrx","HasLDBRX", "true",
+                                        "Enable the ldbrx instruction">;
 def FeatureBookE     : SubtargetFeature<"booke", "IsBookE", "true",
                                         "Enable Book E instructions">;
+def FeatureQPX       : SubtargetFeature<"qpx","HasQPX", "true",
+                                        "Enable QPX instructions">;
+
+// Note: Future features to add when support is extended to more
+// recent ISA levels:
+//
+// CMPB         p6, p6x, p7        cmpb
+// DFP          p6, p6x, p7        decimal floating-point instructions
+// POPCNTB      p5 through p7      popcntb and related instructions
+// VSX          p7                 vector-scalar instruction set
 
 //===----------------------------------------------------------------------===//
 // Register File Description
@@ -73,30 +108,46 @@ include "PPCInstrInfo.td"
 
 def : Processor<"generic", G3Itineraries, [Directive32]>;
 def : Processor<"440", PPC440Itineraries, [Directive440, FeatureISEL,
+                                           FeatureFRES, FeatureFRSQRTE,
                                            FeatureBookE]>;
 def : Processor<"450", PPC440Itineraries, [Directive440, FeatureISEL,
+                                           FeatureFRES, FeatureFRSQRTE,
                                            FeatureBookE]>;
 def : Processor<"601", G3Itineraries, [Directive601]>;
 def : Processor<"602", G3Itineraries, [Directive602]>;
-def : Processor<"603", G3Itineraries, [Directive603]>;
-def : Processor<"603e", G3Itineraries, [Directive603]>;
-def : Processor<"603ev", G3Itineraries, [Directive603]>;
-def : Processor<"604", G3Itineraries, [Directive604]>;
-def : Processor<"604e", G3Itineraries, [Directive604]>;
-def : Processor<"620", G3Itineraries, [Directive620]>;
-def : Processor<"750", G4Itineraries, [Directive750]>;
-def : Processor<"g3", G3Itineraries, [Directive750]>;
-def : Processor<"7400", G4Itineraries, [Directive7400, FeatureAltivec]>;
-def : Processor<"g4", G4Itineraries, [Directive7400, FeatureAltivec]>;
-def : Processor<"7450", G4PlusItineraries, [Directive7400, FeatureAltivec]>;
-def : Processor<"g4+", G4PlusItineraries, [Directive7400, FeatureAltivec]>;
-def : Processor<"970", G5Itineraries,
+def : Processor<"603", G3Itineraries, [Directive603,
+                                       FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"603e", G3Itineraries, [Directive603,
+                                        FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"603ev", G3Itineraries, [Directive603,
+                                         FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"604", G3Itineraries, [Directive604,
+                                       FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"604e", G3Itineraries, [Directive604,
+                                        FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"620", G3Itineraries, [Directive620,
+                                       FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"750", G4Itineraries, [Directive750,
+                                       FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"g3", G3Itineraries, [Directive750,
+                                      FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"7400", G4Itineraries, [Directive7400, FeatureAltivec,
+                                        FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"g4", G4Itineraries, [Directive7400, FeatureAltivec,
+                                      FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"7450", G4PlusItineraries, [Directive7400, FeatureAltivec,
+                                            FeatureFRES, FeatureFRSQRTE]>;
+def : Processor<"g4+", G4PlusItineraries, [Directive7400, FeatureAltivec,
+                                           FeatureFRES, FeatureFRSQRTE]>;
+def : ProcessorModel<"970", G5Model,
                   [Directive970, FeatureAltivec,
-                   FeatureMFOCRF, FeatureFSqrt, FeatureSTFIWX,
+                   FeatureMFOCRF, FeatureFSqrt,
+                   FeatureFRES, FeatureFRSQRTE, FeatureSTFIWX,
                    Feature64Bit /*, Feature64BitRegs */]>;
-def : Processor<"g5", G5Itineraries,
+def : ProcessorModel<"g5", G5Model,
                   [Directive970, FeatureAltivec,
                    FeatureMFOCRF, FeatureFSqrt, FeatureSTFIWX,
+                   FeatureFRES, FeatureFRSQRTE,
                    Feature64Bit /*, Feature64BitRegs */]>;
 def : ProcessorModel<"e500mc", PPCE500mcModel,
                   [DirectiveE500mc, FeatureMFOCRF,
@@ -104,23 +155,65 @@ def : ProcessorModel<"e500mc", PPCE500mcModel,
 def : ProcessorModel<"e5500", PPCE5500Model,
                   [DirectiveE5500, FeatureMFOCRF, Feature64Bit,
                    FeatureSTFIWX, FeatureBookE, FeatureISEL]>;
-def : Processor<"a2", PPCA2Itineraries, [DirectiveA2, FeatureBookE,
-                                         FeatureMFOCRF, FeatureFSqrt,
-                                         FeatureSTFIWX, FeatureISEL,
-                                         Feature64Bit
-                                     /*, Feature64BitRegs */]>;
-def : Processor<"pwr6", G5Itineraries,
+def : ProcessorModel<"a2", PPCA2Model,
+                  [DirectiveA2, FeatureBookE, FeatureMFOCRF,
+                   FeatureFSqrt, FeatureFRE, FeatureFRES,
+                   FeatureFRSQRTE, FeatureFRSQRTES, FeatureRecipPrec,
+                   FeatureSTFIWX, FeatureLFIWAX,
+                   FeatureFPRND, FeatureFPCVT, FeatureISEL,
+                   FeaturePOPCNTD, FeatureLDBRX, Feature64Bit
+               /*, Feature64BitRegs */]>;
+def : ProcessorModel<"a2q", PPCA2Model,
+                  [DirectiveA2, FeatureBookE, FeatureMFOCRF,
+                   FeatureFSqrt, FeatureFRE, FeatureFRES,
+                   FeatureFRSQRTE, FeatureFRSQRTES, FeatureRecipPrec,
+                   FeatureSTFIWX, FeatureLFIWAX,
+                   FeatureFPRND, FeatureFPCVT, FeatureISEL,
+                   FeaturePOPCNTD, FeatureLDBRX, Feature64Bit
+               /*, Feature64BitRegs */, FeatureQPX]>;
+def : ProcessorModel<"pwr3", G5Model,
+                  [DirectivePwr3, FeatureAltivec,
+                   FeatureFRES, FeatureFRSQRTE, FeatureMFOCRF,
+                   FeatureSTFIWX, Feature64Bit]>;
+def : ProcessorModel<"pwr4", G5Model,
+                  [DirectivePwr4, FeatureAltivec, FeatureMFOCRF,
+                   FeatureFSqrt, FeatureFRES, FeatureFRSQRTE,
+                   FeatureSTFIWX, Feature64Bit]>;
+def : ProcessorModel<"pwr5", G5Model,
+                  [DirectivePwr5, FeatureAltivec, FeatureMFOCRF,
+                   FeatureFSqrt, FeatureFRE, FeatureFRES,
+                   FeatureFRSQRTE, FeatureFRSQRTES,
+                   FeatureSTFIWX, Feature64Bit]>;
+def : ProcessorModel<"pwr5x", G5Model,
+                  [DirectivePwr5x, FeatureAltivec, FeatureMFOCRF,
+                   FeatureFSqrt, FeatureFRE, FeatureFRES,
+                   FeatureFRSQRTE, FeatureFRSQRTES,
+                   FeatureSTFIWX, FeatureFPRND, Feature64Bit]>;
+def : ProcessorModel<"pwr6", G5Model,
                   [DirectivePwr6, FeatureAltivec,
-                   FeatureMFOCRF, FeatureFSqrt, FeatureSTFIWX,
-                   Feature64Bit /*, Feature64BitRegs */]>;
-def : Processor<"pwr7", G5Itineraries,
+                   FeatureMFOCRF, FeatureFSqrt, FeatureFRE,
+                   FeatureFRES, FeatureFRSQRTE, FeatureFRSQRTES,
+                   FeatureRecipPrec, FeatureSTFIWX, FeatureLFIWAX,
+                   FeatureFPRND, Feature64Bit /*, Feature64BitRegs */]>;
+def : ProcessorModel<"pwr6x", G5Model,
+                  [DirectivePwr5x, FeatureAltivec, FeatureMFOCRF,
+                   FeatureFSqrt, FeatureFRE, FeatureFRES,
+                   FeatureFRSQRTE, FeatureFRSQRTES, FeatureRecipPrec,
+                   FeatureSTFIWX, FeatureLFIWAX,
+                   FeatureFPRND, Feature64Bit]>;
+def : ProcessorModel<"pwr7", G5Model,
                   [DirectivePwr7, FeatureAltivec,
-                   FeatureMFOCRF, FeatureFSqrt, FeatureSTFIWX,
-                   FeatureISEL, Feature64Bit /*, Feature64BitRegs */]>;
+                   FeatureMFOCRF, FeatureFSqrt, FeatureFRE,
+                   FeatureFRES, FeatureFRSQRTE, FeatureFRSQRTES,
+                   FeatureRecipPrec, FeatureSTFIWX, FeatureLFIWAX,
+                   FeatureFPRND, FeatureFPCVT, FeatureISEL,
+                   FeaturePOPCNTD, FeatureLDBRX,
+                   Feature64Bit /*, Feature64BitRegs */]>;
 def : Processor<"ppc", G3Itineraries, [Directive32]>;
-def : Processor<"ppc64", G5Itineraries,
+def : ProcessorModel<"ppc64", G5Model,
                   [Directive64, FeatureAltivec,
-                   FeatureMFOCRF, FeatureFSqrt, FeatureSTFIWX,
+                   FeatureMFOCRF, FeatureFSqrt, FeatureFRES,
+                   FeatureFRSQRTE, FeatureSTFIWX,
                    Feature64Bit /*, Feature64BitRegs */]>;
 
 
diff --git a/lib/Target/PowerPC/PPCAsmPrinter.cpp b/lib/Target/PowerPC/PPCAsmPrinter.cpp
index 15d690bd8970..96a9f0a39006 100644
--- a/lib/Target/PowerPC/PPCAsmPrinter.cpp
+++ b/lib/Target/PowerPC/PPCAsmPrinter.cpp
@@ -18,14 +18,13 @@
 
 #define DEBUG_TYPE "asmprinter"
 #include "PPC.h"
-#include "PPCTargetMachine.h"
-#include "PPCSubtarget.h"
 #include "InstPrinter/PPCInstPrinter.h"
 #include "MCTargetDesc/PPCPredicates.h"
-#include "llvm/Constants.h"
-#include "llvm/DebugInfo.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Module.h"
+#include "PPCSubtarget.h"
+#include "PPCTargetMachine.h"
+#include "llvm/ADT/MapVector.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/Assembly/Writer.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
@@ -33,28 +32,30 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfoImpls.h"
 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/DebugInfo.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Module.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCInstBuilder.h"
+#include "llvm/MC/MCSectionELF.h"
 #include "llvm/MC/MCSectionMachO.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/MC/MCSectionELF.h"
-#include "llvm/Target/Mangler.h"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetOptions.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
-#include "llvm/Support/MathExtras.h"
+#include "llvm/Support/ELF.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/MathExtras.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Support/ELF.h"
-#include "llvm/ADT/StringExtras.h"
-#include "llvm/ADT/SmallString.h"
-#include "llvm/ADT/MapVector.h"
+#include "llvm/Target/Mangler.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetOptions.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 using namespace llvm;
 
 namespace {
@@ -72,6 +73,7 @@ namespace {
       return "PowerPC Assembly Printer";
     }
 
+    MCSymbol *lookUpOrCreateTOCEntry(MCSymbol *Sym);
 
     virtual void EmitInstruction(const MachineInstr *MI);
 
@@ -309,6 +311,25 @@ bool PPCAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI, unsigned OpNo,
 }
 
 
+/// lookUpOrCreateTOCEntry -- Given a symbol, look up whether a TOC entry
+/// exists for it.  If not, create one.  Then return a symbol that references
+/// the TOC entry.
+MCSymbol *PPCAsmPrinter::lookUpOrCreateTOCEntry(MCSymbol *Sym) {
+
+  MCSymbol *&TOCEntry = TOC[Sym];
+
+  // To avoid name clash check if the name already exists.
+  while (TOCEntry == 0) {
+    if (OutContext.LookupSymbol(Twine(MAI->getPrivateGlobalPrefix()) +
+                                "C" + Twine(TOCLabelID++)) == 0) {
+      TOCEntry = GetTempSymbol("C", TOCLabelID);
+    }
+  }
+
+  return TOCEntry;
+}
+
+
 /// EmitInstruction -- Print out a single PowerPC MI in Darwin syntax to
 /// the current output stream.
 ///
@@ -349,14 +370,10 @@ void PPCAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     MCSymbol *PICBase = MF->getPICBaseSymbol();
     
     // Emit the 'bl'.
-    TmpInst.setOpcode(PPC::BL_Darwin); // Darwin vs SVR4 doesn't matter here.
-    
-    
-    // FIXME: We would like an efficient form for this, so we don't have to do
-    // a lot of extra uniquing.
-    TmpInst.addOperand(MCOperand::CreateExpr(MCSymbolRefExpr::
-                                             Create(PICBase, OutContext)));
-    OutStreamer.EmitInstruction(TmpInst);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::BL)
+      // FIXME: We would like an efficient form for this, so we don't have to do
+      // a lot of extra uniquing.
+      .addExpr(MCSymbolRefExpr::Create(PICBase, OutContext)));
     
     // Emit the label.
     OutStreamer.EmitLabel(PICBase);
@@ -382,14 +399,8 @@ void PPCAsmPrinter::EmitInstruction(const MachineInstr *MI) {
       MOSymbol = GetCPISymbol(MO.getIndex());
     else if (MO.isJTI())
       MOSymbol = GetJTISymbol(MO.getIndex());
-    MCSymbol *&TOCEntry = TOC[MOSymbol];
-    // To avoid name clash check if the name already exists.
-    while (TOCEntry == 0) {
-      if (OutContext.LookupSymbol(Twine(MAI->getPrivateGlobalPrefix()) +
-                                  "C" + Twine(TOCLabelID++)) == 0) {
-        TOCEntry = GetTempSymbol("C", TOCLabelID);
-      }
-    }
+
+    MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(MOSymbol);
 
     const MCExpr *Exp =
       MCSymbolRefExpr::Create(TOCEntry, MCSymbolRefExpr::VK_PPC_TOC_ENTRY,
@@ -399,15 +410,299 @@ void PPCAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     return;
   }
       
+  case PPC::ADDIStocHA: {
+    // Transform %Xd = ADDIStocHA %X2, <ga:@sym>
+    LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
+
+    // Change the opcode to ADDIS8.  If the global address is external,
+    // has common linkage, is a function address, or is a jump table
+    // address, then generate a TOC entry and reference that.  Otherwise
+    // reference the symbol directly.
+    TmpInst.setOpcode(PPC::ADDIS8);
+    const MachineOperand &MO = MI->getOperand(2);
+    assert((MO.isGlobal() || MO.isCPI() || MO.isJTI()) &&
+           "Invalid operand for ADDIStocHA!");
+    MCSymbol *MOSymbol = 0;
+    bool IsExternal = false;
+    bool IsFunction = false;
+    bool IsCommon = false;
+    bool IsAvailExt = false;
+
+    if (MO.isGlobal()) {
+      const GlobalValue *GValue = MO.getGlobal();
+      const GlobalAlias *GAlias = dyn_cast<GlobalAlias>(GValue);
+      const GlobalValue *RealGValue = GAlias ?
+        GAlias->resolveAliasedGlobal(false) : GValue;
+      MOSymbol = Mang->getSymbol(RealGValue);
+      const GlobalVariable *GVar = dyn_cast<GlobalVariable>(RealGValue);
+      IsExternal = GVar && !GVar->hasInitializer();
+      IsCommon = GVar && RealGValue->hasCommonLinkage();
+      IsFunction = !GVar;
+      IsAvailExt = GVar && RealGValue->hasAvailableExternallyLinkage();
+    } else if (MO.isCPI())
+      MOSymbol = GetCPISymbol(MO.getIndex());
+    else if (MO.isJTI())
+      MOSymbol = GetJTISymbol(MO.getIndex());
+
+    if (IsExternal || IsFunction || IsCommon || IsAvailExt || MO.isJTI())
+      MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
+
+    const MCExpr *Exp =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TOC16_HA,
+                              OutContext);
+    TmpInst.getOperand(2) = MCOperand::CreateExpr(Exp);
+    OutStreamer.EmitInstruction(TmpInst);
+    return;
+  }
+  case PPC::LDtocL: {
+    // Transform %Xd = LDtocL <ga:@sym>, %Xs
+    LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
+
+    // Change the opcode to LD.  If the global address is external, has
+    // common linkage, or is a jump table address, then reference the
+    // associated TOC entry.  Otherwise reference the symbol directly.
+    TmpInst.setOpcode(PPC::LD);
+    const MachineOperand &MO = MI->getOperand(1);
+    assert((MO.isGlobal() || MO.isJTI() || MO.isCPI()) &&
+           "Invalid operand for LDtocL!");
+    MCSymbol *MOSymbol = 0;
+
+    if (MO.isJTI())
+      MOSymbol = lookUpOrCreateTOCEntry(GetJTISymbol(MO.getIndex()));
+    else if (MO.isCPI())
+      MOSymbol = GetCPISymbol(MO.getIndex());
+    else if (MO.isGlobal()) {
+      const GlobalValue *GValue = MO.getGlobal();
+      const GlobalAlias *GAlias = dyn_cast<GlobalAlias>(GValue);
+      const GlobalValue *RealGValue = GAlias ?
+        GAlias->resolveAliasedGlobal(false) : GValue;
+      MOSymbol = Mang->getSymbol(RealGValue);
+      const GlobalVariable *GVar = dyn_cast<GlobalVariable>(RealGValue);
+    
+      if (!GVar || !GVar->hasInitializer() || RealGValue->hasCommonLinkage() ||
+          RealGValue->hasAvailableExternallyLinkage())
+        MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
+    }
+
+    const MCExpr *Exp =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TOC16_LO,
+                              OutContext);
+    TmpInst.getOperand(1) = MCOperand::CreateExpr(Exp);
+    OutStreamer.EmitInstruction(TmpInst);
+    return;
+  }
+  case PPC::ADDItocL: {
+    // Transform %Xd = ADDItocL %Xs, <ga:@sym>
+    LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
+
+    // Change the opcode to ADDI8.  If the global address is external, then
+    // generate a TOC entry and reference that.  Otherwise reference the
+    // symbol directly.
+    TmpInst.setOpcode(PPC::ADDI8);
+    const MachineOperand &MO = MI->getOperand(2);
+    assert((MO.isGlobal() || MO.isCPI()) && "Invalid operand for ADDItocL");
+    MCSymbol *MOSymbol = 0;
+    bool IsExternal = false;
+    bool IsFunction = false;
+
+    if (MO.isGlobal()) {
+      const GlobalValue *GValue = MO.getGlobal();
+      const GlobalAlias *GAlias = dyn_cast<GlobalAlias>(GValue);
+      const GlobalValue *RealGValue = GAlias ?
+        GAlias->resolveAliasedGlobal(false) : GValue;
+      MOSymbol = Mang->getSymbol(RealGValue);
+      const GlobalVariable *GVar = dyn_cast<GlobalVariable>(RealGValue);
+      IsExternal = GVar && !GVar->hasInitializer();
+      IsFunction = !GVar;
+    } else if (MO.isCPI())
+      MOSymbol = GetCPISymbol(MO.getIndex());
+
+    if (IsFunction || IsExternal)
+      MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
+
+    const MCExpr *Exp =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TOC16_LO,
+                              OutContext);
+    TmpInst.getOperand(2) = MCOperand::CreateExpr(Exp);
+    OutStreamer.EmitInstruction(TmpInst);
+    return;
+  }
+  case PPC::ADDISgotTprelHA: {
+    // Transform: %Xd = ADDISgotTprelHA %X2, <ga:@sym>
+    // Into:      %Xd = ADDIS8 %X2, sym@got@tlsgd@ha
+    assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
+    const MachineOperand &MO = MI->getOperand(2);
+    const GlobalValue *GValue = MO.getGlobal();
+    MCSymbol *MOSymbol = Mang->getSymbol(GValue);
+    const MCExpr *SymGotTprel =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TPREL16_HA,
+                              OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDIS8)
+                                .addReg(MI->getOperand(0).getReg())
+                                .addReg(PPC::X2)
+                                .addExpr(SymGotTprel));
+    return;
+  }
+  case PPC::LDgotTprelL: {
+    // Transform %Xd = LDgotTprelL <ga:@sym>, %Xs
+    LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
+
+    // Change the opcode to LD.
+    TmpInst.setOpcode(PPC::LD);
+    const MachineOperand &MO = MI->getOperand(1);
+    const GlobalValue *GValue = MO.getGlobal();
+    MCSymbol *MOSymbol = Mang->getSymbol(GValue);
+    const MCExpr *Exp =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TPREL16_LO,
+                              OutContext);
+    TmpInst.getOperand(1) = MCOperand::CreateExpr(Exp);
+    OutStreamer.EmitInstruction(TmpInst);
+    return;
+  }
+  case PPC::ADDIStlsgdHA: {
+    // Transform: %Xd = ADDIStlsgdHA %X2, <ga:@sym>
+    // Into:      %Xd = ADDIS8 %X2, sym@got@tlsgd@ha
+    assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
+    const MachineOperand &MO = MI->getOperand(2);
+    const GlobalValue *GValue = MO.getGlobal();
+    MCSymbol *MOSymbol = Mang->getSymbol(GValue);
+    const MCExpr *SymGotTlsGD =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSGD16_HA,
+                              OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDIS8)
+                                .addReg(MI->getOperand(0).getReg())
+                                .addReg(PPC::X2)
+                                .addExpr(SymGotTlsGD));
+    return;
+  }
+  case PPC::ADDItlsgdL: {
+    // Transform: %Xd = ADDItlsgdL %Xs, <ga:@sym>
+    // Into:      %Xd = ADDI8 %Xs, sym@got@tlsgd@l
+    assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
+    const MachineOperand &MO = MI->getOperand(2);
+    const GlobalValue *GValue = MO.getGlobal();
+    MCSymbol *MOSymbol = Mang->getSymbol(GValue);
+    const MCExpr *SymGotTlsGD =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSGD16_LO,
+                              OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDI8)
+                                .addReg(MI->getOperand(0).getReg())
+                                .addReg(MI->getOperand(1).getReg())
+                                .addExpr(SymGotTlsGD));
+    return;
+  }
+  case PPC::GETtlsADDR: {
+    // Transform: %X3 = GETtlsADDR %X3, <ga:@sym>
+    // Into:      BL8_NOP_TLSGD __tls_get_addr(sym@tlsgd)
+    assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
+
+    StringRef Name = "__tls_get_addr";
+    MCSymbol *TlsGetAddr = OutContext.GetOrCreateSymbol(Name);
+    const MCSymbolRefExpr *TlsRef = 
+      MCSymbolRefExpr::Create(TlsGetAddr, MCSymbolRefExpr::VK_None, OutContext);
+    const MachineOperand &MO = MI->getOperand(2);
+    const GlobalValue *GValue = MO.getGlobal();
+    MCSymbol *MOSymbol = Mang->getSymbol(GValue);
+    const MCExpr *SymVar =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TLSGD,
+                              OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::BL8_NOP_TLSGD)
+                                .addExpr(TlsRef)
+                                .addExpr(SymVar));
+    return;
+  }
+  case PPC::ADDIStlsldHA: {
+    // Transform: %Xd = ADDIStlsldHA %X2, <ga:@sym>
+    // Into:      %Xd = ADDIS8 %X2, sym@got@tlsld@ha
+    assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
+    const MachineOperand &MO = MI->getOperand(2);
+    const GlobalValue *GValue = MO.getGlobal();
+    MCSymbol *MOSymbol = Mang->getSymbol(GValue);
+    const MCExpr *SymGotTlsLD =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSLD16_HA,
+                              OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDIS8)
+                                .addReg(MI->getOperand(0).getReg())
+                                .addReg(PPC::X2)
+                                .addExpr(SymGotTlsLD));
+    return;
+  }
+  case PPC::ADDItlsldL: {
+    // Transform: %Xd = ADDItlsldL %Xs, <ga:@sym>
+    // Into:      %Xd = ADDI8 %Xs, sym@got@tlsld@l
+    assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
+    const MachineOperand &MO = MI->getOperand(2);
+    const GlobalValue *GValue = MO.getGlobal();
+    MCSymbol *MOSymbol = Mang->getSymbol(GValue);
+    const MCExpr *SymGotTlsLD =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_GOT_TLSLD16_LO,
+                              OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDI8)
+                                .addReg(MI->getOperand(0).getReg())
+                                .addReg(MI->getOperand(1).getReg())
+                                .addExpr(SymGotTlsLD));
+    return;
+  }
+  case PPC::GETtlsldADDR: {
+    // Transform: %X3 = GETtlsldADDR %X3, <ga:@sym>
+    // Into:      BL8_NOP_TLSLD __tls_get_addr(sym@tlsld)
+    assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
+
+    StringRef Name = "__tls_get_addr";
+    MCSymbol *TlsGetAddr = OutContext.GetOrCreateSymbol(Name);
+    const MCSymbolRefExpr *TlsRef = 
+      MCSymbolRefExpr::Create(TlsGetAddr, MCSymbolRefExpr::VK_None, OutContext);
+    const MachineOperand &MO = MI->getOperand(2);
+    const GlobalValue *GValue = MO.getGlobal();
+    MCSymbol *MOSymbol = Mang->getSymbol(GValue);
+    const MCExpr *SymVar =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_TLSLD,
+                              OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::BL8_NOP_TLSLD)
+                                .addExpr(TlsRef)
+                                .addExpr(SymVar));
+    return;
+  }
+  case PPC::ADDISdtprelHA: {
+    // Transform: %Xd = ADDISdtprelHA %X3, <ga:@sym>
+    // Into:      %Xd = ADDIS8 %X3, sym@dtprel@ha
+    assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
+    const MachineOperand &MO = MI->getOperand(2);
+    const GlobalValue *GValue = MO.getGlobal();
+    MCSymbol *MOSymbol = Mang->getSymbol(GValue);
+    const MCExpr *SymDtprel =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_DTPREL16_HA,
+                              OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDIS8)
+                                .addReg(MI->getOperand(0).getReg())
+                                .addReg(PPC::X3)
+                                .addExpr(SymDtprel));
+    return;
+  }
+  case PPC::ADDIdtprelL: {
+    // Transform: %Xd = ADDIdtprelL %Xs, <ga:@sym>
+    // Into:      %Xd = ADDI8 %Xs, sym@dtprel@l
+    assert(Subtarget.isPPC64() && "Not supported for 32-bit PowerPC");
+    const MachineOperand &MO = MI->getOperand(2);
+    const GlobalValue *GValue = MO.getGlobal();
+    MCSymbol *MOSymbol = Mang->getSymbol(GValue);
+    const MCExpr *SymDtprel =
+      MCSymbolRefExpr::Create(MOSymbol, MCSymbolRefExpr::VK_PPC_DTPREL16_LO,
+                              OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDI8)
+                                .addReg(MI->getOperand(0).getReg())
+                                .addReg(MI->getOperand(1).getReg())
+                                .addExpr(SymDtprel));
+    return;
+  }
   case PPC::MFCRpseud:
   case PPC::MFCR8pseud:
     // Transform: %R3 = MFCRpseud %CR7
     // Into:      %R3 = MFCR      ;; cr7
     OutStreamer.AddComment(PPCInstPrinter::
                            getRegisterName(MI->getOperand(1).getReg()));
-    TmpInst.setOpcode(Subtarget.isPPC64() ? PPC::MFCR8 : PPC::MFCR);
-    TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-    OutStreamer.EmitInstruction(TmpInst);
+    OutStreamer.EmitInstruction(MCInstBuilder(Subtarget.isPPC64() ? PPC::MFCR8 : PPC::MFCR)
+      .addReg(MI->getOperand(0).getReg()));
     return;
   case PPC::SYNC:
     // In Book E sync is called msync, handle this special case here...
@@ -438,14 +733,14 @@ void PPCLinuxAsmPrinter::EmitFunctionEntryLabel() {
   // Generates a R_PPC64_ADDR64 (from FK_DATA_8) relocation for the function
   // entry point.
   OutStreamer.EmitValue(MCSymbolRefExpr::Create(Symbol1, OutContext),
-                        8/*size*/, 0/*addrspace*/);
+			8 /*size*/);
   MCSymbol *Symbol2 = OutContext.GetOrCreateSymbol(StringRef(".TOC."));
   // Generates a R_PPC64_TOC relocation for TOC base insertion.
   OutStreamer.EmitValue(MCSymbolRefExpr::Create(Symbol2,
                         MCSymbolRefExpr::VK_PPC_TOC, OutContext),
-                        8/*size*/, 0/*addrspace*/);
+                        8/*size*/);
   // Emit a null environment pointer.
-  OutStreamer.EmitIntValue(0, 8 /* size */, 0 /* addrspace */);
+  OutStreamer.EmitIntValue(0, 8 /* size */);
   OutStreamer.SwitchSection(Current);
 
   MCSymbol *RealFnSym = OutContext.GetOrCreateSymbol(
@@ -474,6 +769,25 @@ bool PPCLinuxAsmPrinter::doFinalization(Module &M) {
     }
   }
 
+  MachineModuleInfoELF &MMIELF =
+    MMI->getObjFileInfo<MachineModuleInfoELF>();
+
+  MachineModuleInfoELF::SymbolListTy Stubs = MMIELF.GetGVStubList();
+  if (!Stubs.empty()) {
+    OutStreamer.SwitchSection(getObjFileLowering().getDataSection());
+    for (unsigned i = 0, e = Stubs.size(); i != e; ++i) {
+      // L_foo$stub:
+      OutStreamer.EmitLabel(Stubs[i].first);
+      //   .long _foo
+      OutStreamer.EmitValue(MCSymbolRefExpr::Create(Stubs[i].second.getPointer(),
+                                                    OutContext),
+                            isPPC64 ? 8 : 4/*size*/, 0/*addrspace*/);
+    }
+
+    Stubs.clear();
+    OutStreamer.AddBlankLine();
+  }
+
   return AsmPrinter::doFinalization(M);
 }
 
@@ -508,7 +822,12 @@ void PPCDarwinAsmPrinter::EmitStartOfAsmFile(Module &M) {
     "ppcA2",
     "ppce500mc",
     "ppce5500",
+    "power3",
+    "power4",
+    "power5",
+    "power5x",
     "power6",
+    "power6x",
     "power7",
     "ppc64"
   };
@@ -523,8 +842,11 @@ void PPCDarwinAsmPrinter::EmitStartOfAsmFile(Module &M) {
   assert(Directive <= PPC::DIR_64 && "Directive out of range.");
   
   // FIXME: This is a total hack, finish mc'izing the PPC backend.
-  if (OutStreamer.hasRawTextSupport())
+  if (OutStreamer.hasRawTextSupport()) {
+    assert(Directive < sizeof(CPUDirectives) / sizeof(*CPUDirectives) &&
+           "CPUDirectives[] might not be up-to-date!");
     OutStreamer.EmitRawText("\t.machine " + Twine(CPUDirectives[Directive]));
+  }
 
   // Prime text sections so they are adjacent.  This reduces the likelihood a
   // large data or debug section causes a branch to exceed 16M limit.
@@ -549,16 +871,13 @@ void PPCDarwinAsmPrinter::EmitStartOfAsmFile(Module &M) {
 
 static MCSymbol *GetLazyPtr(MCSymbol *Sym, MCContext &Ctx) {
   // Remove $stub suffix, add $lazy_ptr.
-  SmallString<128> TmpStr(Sym->getName().begin(), Sym->getName().end()-5);
-  TmpStr += "$lazy_ptr";
-  return Ctx.GetOrCreateSymbol(TmpStr.str());
+  StringRef NoStub = Sym->getName().substr(0, Sym->getName().size()-5);
+  return Ctx.GetOrCreateSymbol(NoStub + "$lazy_ptr");
 }
 
 static MCSymbol *GetAnonSym(MCSymbol *Sym, MCContext &Ctx) {
   // Add $tmp suffix to $stub, yielding $stub$tmp.
-  SmallString<128> TmpStr(Sym->getName().begin(), Sym->getName().end());
-  TmpStr += "$tmp";
-  return Ctx.GetOrCreateSymbol(TmpStr.str());
+  return Ctx.GetOrCreateSymbol(Sym->getName() + "$tmp");
 }
 
 void PPCDarwinAsmPrinter::
@@ -589,32 +908,51 @@ EmitFunctionStubs(const MachineModuleInfoMachO::SymbolListTy &Stubs) {
                                            
       OutStreamer.EmitLabel(Stub);
       OutStreamer.EmitSymbolAttribute(RawSym, MCSA_IndirectSymbol);
-      // FIXME: MCize this.
-      OutStreamer.EmitRawText(StringRef("\tmflr r0"));
-      OutStreamer.EmitRawText("\tbcl 20,31," + Twine(AnonSymbol->getName()));
+
+      const MCExpr *Anon = MCSymbolRefExpr::Create(AnonSymbol, OutContext);
+
+      // mflr r0
+      OutStreamer.EmitInstruction(MCInstBuilder(PPC::MFLR).addReg(PPC::R0));
+      // bcl 20, 31, AnonSymbol
+      OutStreamer.EmitInstruction(MCInstBuilder(PPC::BCLalways).addExpr(Anon));
       OutStreamer.EmitLabel(AnonSymbol);
-      OutStreamer.EmitRawText(StringRef("\tmflr r11"));
-      OutStreamer.EmitRawText("\taddis r11,r11,ha16("+Twine(LazyPtr->getName())+
-                              "-" + AnonSymbol->getName() + ")");
-      OutStreamer.EmitRawText(StringRef("\tmtlr r0"));
-      
-      if (isPPC64)
-        OutStreamer.EmitRawText("\tldu r12,lo16(" + Twine(LazyPtr->getName()) +
-                                "-" + AnonSymbol->getName() + ")(r11)");
-      else
-        OutStreamer.EmitRawText("\tlwzu r12,lo16(" + Twine(LazyPtr->getName()) +
-                                "-" + AnonSymbol->getName() + ")(r11)");
-      OutStreamer.EmitRawText(StringRef("\tmtctr r12"));
-      OutStreamer.EmitRawText(StringRef("\tbctr"));
-      
+      // mflr r11
+      OutStreamer.EmitInstruction(MCInstBuilder(PPC::MFLR).addReg(PPC::R11));
+      // addis r11, r11, ha16(LazyPtr - AnonSymbol)
+      const MCExpr *Sub =
+        MCBinaryExpr::CreateSub(MCSymbolRefExpr::Create(LazyPtr, OutContext),
+                                Anon, OutContext);
+      OutStreamer.EmitInstruction(MCInstBuilder(PPC::ADDIS)
+        .addReg(PPC::R11)
+        .addReg(PPC::R11)
+        .addExpr(Sub));
+      // mtlr r0
+      OutStreamer.EmitInstruction(MCInstBuilder(PPC::MTLR).addReg(PPC::R0));
+
+      // ldu r12, lo16(LazyPtr - AnonSymbol)(r11)
+      // lwzu r12, lo16(LazyPtr - AnonSymbol)(r11)
+      OutStreamer.EmitInstruction(MCInstBuilder(isPPC64 ? PPC::LDU : PPC::LWZU)
+        .addReg(PPC::R12)
+        .addExpr(Sub).addExpr(Sub)
+        .addReg(PPC::R11));
+      // mtctr r12
+      OutStreamer.EmitInstruction(MCInstBuilder(PPC::MTCTR).addReg(PPC::R12));
+      // bctr
+      OutStreamer.EmitInstruction(MCInstBuilder(PPC::BCTR));
+
       OutStreamer.SwitchSection(LSPSection);
       OutStreamer.EmitLabel(LazyPtr);
       OutStreamer.EmitSymbolAttribute(RawSym, MCSA_IndirectSymbol);
-      
-      if (isPPC64)
-        OutStreamer.EmitRawText(StringRef("\t.quad dyld_stub_binding_helper"));
-      else
-        OutStreamer.EmitRawText(StringRef("\t.long dyld_stub_binding_helper"));
+
+      MCSymbol *DyldStubBindingHelper =
+        OutContext.GetOrCreateSymbol(StringRef("dyld_stub_binding_helper"));
+      if (isPPC64) {
+        // .quad dyld_stub_binding_helper
+        OutStreamer.EmitSymbolValue(DyldStubBindingHelper, 8);
+      } else {
+        // .long dyld_stub_binding_helper
+        OutStreamer.EmitSymbolValue(DyldStubBindingHelper, 4);
+      }
     }
     OutStreamer.AddBlankLine();
     return;
@@ -634,23 +972,42 @@ EmitFunctionStubs(const MachineModuleInfoMachO::SymbolListTy &Stubs) {
     EmitAlignment(4);
     OutStreamer.EmitLabel(Stub);
     OutStreamer.EmitSymbolAttribute(RawSym, MCSA_IndirectSymbol);
-    OutStreamer.EmitRawText("\tlis r11,ha16(" + Twine(LazyPtr->getName()) +")");
-    if (isPPC64)
-      OutStreamer.EmitRawText("\tldu r12,lo16(" + Twine(LazyPtr->getName()) +
-                              ")(r11)");
-    else
-      OutStreamer.EmitRawText("\tlwzu r12,lo16(" + Twine(LazyPtr->getName()) +
-                              ")(r11)");
-    OutStreamer.EmitRawText(StringRef("\tmtctr r12"));
-    OutStreamer.EmitRawText(StringRef("\tbctr"));
+    // lis r11, ha16(LazyPtr)
+    const MCExpr *LazyPtrHa16 =
+      MCSymbolRefExpr::Create(LazyPtr, MCSymbolRefExpr::VK_PPC_DARWIN_HA16,
+                              OutContext);
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::LIS)
+      .addReg(PPC::R11)
+      .addExpr(LazyPtrHa16));
+
+    const MCExpr *LazyPtrLo16 =
+      MCSymbolRefExpr::Create(LazyPtr, MCSymbolRefExpr::VK_PPC_DARWIN_LO16,
+                              OutContext);
+    // ldu r12, lo16(LazyPtr)(r11)
+    // lwzu r12, lo16(LazyPtr)(r11)
+    OutStreamer.EmitInstruction(MCInstBuilder(isPPC64 ? PPC::LDU : PPC::LWZU)
+      .addReg(PPC::R12)
+      .addExpr(LazyPtrLo16).addExpr(LazyPtrLo16)
+      .addReg(PPC::R11));
+
+    // mtctr r12
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::MTCTR).addReg(PPC::R12));
+    // bctr
+    OutStreamer.EmitInstruction(MCInstBuilder(PPC::BCTR));
+
     OutStreamer.SwitchSection(LSPSection);
     OutStreamer.EmitLabel(LazyPtr);
     OutStreamer.EmitSymbolAttribute(RawSym, MCSA_IndirectSymbol);
-    
-    if (isPPC64)
-      OutStreamer.EmitRawText(StringRef("\t.quad dyld_stub_binding_helper"));
-    else
-      OutStreamer.EmitRawText(StringRef("\t.long dyld_stub_binding_helper"));
+
+    MCSymbol *DyldStubBindingHelper =
+      OutContext.GetOrCreateSymbol(StringRef("dyld_stub_binding_helper"));
+    if (isPPC64) {
+      // .quad dyld_stub_binding_helper
+      OutStreamer.EmitSymbolValue(DyldStubBindingHelper, 8);
+    } else {
+      // .long dyld_stub_binding_helper
+      OutStreamer.EmitSymbolValue(DyldStubBindingHelper, 4);
+    }
   }
   
   OutStreamer.AddBlankLine();
@@ -703,7 +1060,7 @@ bool PPCDarwinAsmPrinter::doFinalization(Module &M) {
 
       if (MCSym.getInt())
         // External to current translation unit.
-        OutStreamer.EmitIntValue(0, isPPC64 ? 8 : 4/*size*/, 0/*addrspace*/);
+        OutStreamer.EmitIntValue(0, isPPC64 ? 8 : 4/*size*/);
       else
         // Internal to current translation unit.
         //
@@ -713,7 +1070,7 @@ bool PPCDarwinAsmPrinter::doFinalization(Module &M) {
         // fill in the value for the NLP in those cases.
         OutStreamer.EmitValue(MCSymbolRefExpr::Create(MCSym.getPointer(),
                                                       OutContext),
-                              isPPC64 ? 8 : 4/*size*/, 0/*addrspace*/);
+                              isPPC64 ? 8 : 4/*size*/);
     }
 
     Stubs.clear();
@@ -732,7 +1089,7 @@ bool PPCDarwinAsmPrinter::doFinalization(Module &M) {
       OutStreamer.EmitValue(MCSymbolRefExpr::
                             Create(Stubs[i].second.getPointer(),
                                    OutContext),
-                            isPPC64 ? 8 : 4/*size*/, 0/*addrspace*/);
+                            isPPC64 ? 8 : 4/*size*/);
     }
 
     Stubs.clear();
diff --git a/lib/Target/PowerPC/PPCBranchSelector.cpp b/lib/Target/PowerPC/PPCBranchSelector.cpp
index 21a0fb200f20..bd1c37868110 100644
--- a/lib/Target/PowerPC/PPCBranchSelector.cpp
+++ b/lib/Target/PowerPC/PPCBranchSelector.cpp
@@ -17,21 +17,27 @@
 
 #define DEBUG_TYPE "ppc-branch-select"
 #include "PPC.h"
+#include "MCTargetDesc/PPCPredicates.h"
 #include "PPCInstrBuilder.h"
 #include "PPCInstrInfo.h"
-#include "MCTargetDesc/PPCPredicates.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/Support/MathExtras.h"
+#include "llvm/Target/TargetMachine.h"
 using namespace llvm;
 
 STATISTIC(NumExpanded, "Number of branches expanded to long format");
 
+namespace llvm {
+  void initializePPCBSelPass(PassRegistry&);
+}
+
 namespace {
   struct PPCBSel : public MachineFunctionPass {
     static char ID;
-    PPCBSel() : MachineFunctionPass(ID) {}
+    PPCBSel() : MachineFunctionPass(ID) {
+      initializePPCBSelPass(*PassRegistry::getPassRegistry());
+    }
 
     /// BlockSizes - The sizes of the basic blocks in the function.
     std::vector<unsigned> BlockSizes;
@@ -45,6 +51,9 @@ namespace {
   char PPCBSel::ID = 0;
 }
 
+INITIALIZE_PASS(PPCBSel, "ppc-branch-select", "PowerPC Branch Selector",
+                false, false)
+
 /// createPPCBranchSelectionPass - returns an instance of the Branch Selection
 /// Pass
 ///
diff --git a/lib/Target/PowerPC/PPCCTRLoops.cpp b/lib/Target/PowerPC/PPCCTRLoops.cpp
index 2a2abb171fb1..81a54d7015b0 100644
--- a/lib/Target/PowerPC/PPCCTRLoops.cpp
+++ b/lib/Target/PowerPC/PPCCTRLoops.cpp
@@ -31,20 +31,20 @@
 
 #define DEBUG_TYPE "ctrloops"
 #include "PPC.h"
-#include "PPCTargetMachine.h"
 #include "MCTargetDesc/PPCPredicates.h"
-#include "llvm/Constants.h"
-#include "llvm/PassSupport.h"
+#include "PPCTargetMachine.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/Statistic.h"
-#include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineLoopInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/PassSupport.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetInstrInfo.h"
@@ -54,6 +54,10 @@ using namespace llvm;
 
 STATISTIC(NumCTRLoops, "Number of loops converted to CTR loops");
 
+namespace llvm {
+  void initializePPCCTRLoopsPass(PassRegistry&);
+}
+
 namespace {
   class CountValue;
   struct PPCCTRLoops : public MachineFunctionPass {
@@ -64,7 +68,9 @@ namespace {
   public:
     static char ID;   // Pass identification, replacement for typeid
 
-    PPCCTRLoops() : MachineFunctionPass(ID) {}
+    PPCCTRLoops() : MachineFunctionPass(ID) {
+      initializePPCCTRLoopsPass(*PassRegistry::getPassRegistry());
+    }
 
     virtual bool runOnMachineFunction(MachineFunction &MF);
 
@@ -174,15 +180,32 @@ namespace {
   };
 } // end anonymous namespace
 
+INITIALIZE_PASS_BEGIN(PPCCTRLoops, "ppc-ctr-loops", "PowerPC CTR Loops",
+                      false, false)
+INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
+INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)
+INITIALIZE_PASS_END(PPCCTRLoops, "ppc-ctr-loops", "PowerPC CTR Loops",
+                    false, false)
 
 /// isCompareEquals - Returns true if the instruction is a compare equals
 /// instruction with an immediate operand.
-static bool isCompareEqualsImm(const MachineInstr *MI, bool &SignedCmp) {
-  if (MI->getOpcode() == PPC::CMPWI || MI->getOpcode() == PPC::CMPDI) {
+static bool isCompareEqualsImm(const MachineInstr *MI, bool &SignedCmp,
+                               bool &Int64Cmp) {
+  if (MI->getOpcode() == PPC::CMPWI) {
     SignedCmp = true;
+    Int64Cmp = false;
+    return true;
+  } else if (MI->getOpcode() == PPC::CMPDI) {
+    SignedCmp = true;
+    Int64Cmp = true;
+    return true;
+  } else if (MI->getOpcode() == PPC::CMPLWI) {
+    SignedCmp = false;
+    Int64Cmp = false;
     return true;
-  } else if (MI->getOpcode() == PPC::CMPLWI || MI->getOpcode() == PPC::CMPLDI) {
+  } else if (MI->getOpcode() == PPC::CMPLDI) {
     SignedCmp = false;
+    Int64Cmp = true;
     return true;
   }
 
@@ -341,9 +364,9 @@ CountValue *PPCCTRLoops::getTripCount(MachineLoop *L,
          RI = MRI->reg_begin(IV_Opnd->getReg()), RE = MRI->reg_end();
          RI != RE; ++RI) {
       IV_Opnd = &RI.getOperand();
-      bool SignedCmp;
+      bool SignedCmp, Int64Cmp;
       MachineInstr *MI = IV_Opnd->getParent();
-      if (L->contains(MI) && isCompareEqualsImm(MI, SignedCmp) &&
+      if (L->contains(MI) && isCompareEqualsImm(MI, SignedCmp, Int64Cmp) &&
           MI->getOperand(0).getReg() == PredReg) {
 
         OldInsts.push_back(MI);
@@ -368,14 +391,14 @@ CountValue *PPCCTRLoops::getTripCount(MachineLoop *L,
         assert(InitialValue->isReg() && "Expecting register for init value");
         unsigned InitialValueReg = InitialValue->getReg();
   
-        const MachineInstr *DefInstr = MRI->getVRegDef(InitialValueReg);
+        MachineInstr *DefInstr = MRI->getVRegDef(InitialValueReg);
   
         // Here we need to look for an immediate load (an li or lis/ori pair).
         if (DefInstr && (DefInstr->getOpcode() == PPC::ORI8 ||
                          DefInstr->getOpcode() == PPC::ORI)) {
-          int64_t start = (short) DefInstr->getOperand(2).getImm();
-          const MachineInstr *DefInstr2 =
-            MRI->getVRegDef(DefInstr->getOperand(0).getReg());
+          int64_t start = DefInstr->getOperand(2).getImm();
+          MachineInstr *DefInstr2 =
+            MRI->getVRegDef(DefInstr->getOperand(1).getReg());
           if (DefInstr2 && (DefInstr2->getOpcode() == PPC::LIS8 ||
                             DefInstr2->getOpcode() == PPC::LIS)) {
             DEBUG(dbgs() << "  initial constant: " << *DefInstr);
@@ -387,17 +410,33 @@ CountValue *PPCCTRLoops::getTripCount(MachineLoop *L,
             if ((count % iv_value) != 0) {
               return 0;
             }
-            return new CountValue(count/iv_value);
+
+            OldInsts.push_back(DefInstr);
+            OldInsts.push_back(DefInstr2);
+
+            // count/iv_value, the trip count, should be positive here. If it
+            // is negative, that indicates that the counter will wrap.
+            if (Int64Cmp)
+              return new CountValue(count/iv_value);
+            else
+              return new CountValue(uint32_t(count/iv_value));
           }
         } else if (DefInstr && (DefInstr->getOpcode() == PPC::LI8 ||
                                 DefInstr->getOpcode() == PPC::LI)) {
           DEBUG(dbgs() << "  initial constant: " << *DefInstr);
 
-          int64_t count = ImmVal - int64_t(short(DefInstr->getOperand(1).getImm()));
+          int64_t count = ImmVal -
+            int64_t(short(DefInstr->getOperand(1).getImm()));
           if ((count % iv_value) != 0) {
             return 0;
           }
-          return new CountValue(count/iv_value);
+
+          OldInsts.push_back(DefInstr);
+
+          if (Int64Cmp)
+            return new CountValue(count/iv_value);
+          else
+            return new CountValue(uint32_t(count/iv_value));
         } else if (iv_value == 1 || iv_value == -1) {
           // We can't determine a constant starting value.
           if (ImmVal == 0) {
@@ -405,8 +444,8 @@ CountValue *PPCCTRLoops::getTripCount(MachineLoop *L,
           }
           // FIXME: handle non-zero end value.
         }
-        // FIXME: handle non-unit increments (we might not want to introduce division
-        // but we can handle some 2^n cases with shifts).
+        // FIXME: handle non-unit increments (we might not want to introduce
+        // division but we can handle some 2^n cases with shifts).
   
       }
     }
@@ -477,9 +516,10 @@ bool PPCCTRLoops::isDead(const MachineInstr *MI,
     if (MO.isReg() && MO.isDef()) {
       unsigned Reg = MO.getReg();
       if (!MRI->use_nodbg_empty(Reg)) {
-        // This instruction has users, but if the only user is the phi node for the
-        // parent block, and the only use of that phi node is this instruction, then
-        // this instruction is dead: both it (and the phi node) can be removed.
+        // This instruction has users, but if the only user is the phi node for
+        // the parent block, and the only use of that phi node is this
+        // instruction, then this instruction is dead: both it (and the phi
+        // node) can be removed.
         MachineRegisterInfo::use_iterator I = MRI->use_begin(Reg);
         if (llvm::next(I) == MRI->use_end() &&
             I.getOperand().getParent()->isPHI()) {
@@ -582,6 +622,16 @@ bool PPCCTRLoops::convertToCTRLoop(MachineLoop *L) {
     DEBUG(dbgs() << "failed to get trip count!\n");
     return false;
   }
+
+  if (TripCount->isImm()) {
+    DEBUG(dbgs() << "constant trip count: " << TripCount->getImm() << "\n");
+
+    // FIXME: We currently can't form 64-bit constants
+    // (including 32-bit unsigned constants)
+    if (!isInt<32>(TripCount->getImm()))
+      return false;
+  }
+
   // Does the loop contain any invalid instructions?
   if (containsInvalidInstruction(L)) {
     return false;
@@ -635,7 +685,7 @@ bool PPCCTRLoops::convertToCTRLoop(MachineLoop *L) {
     const TargetRegisterClass *SrcRC =
       MF->getRegInfo().getRegClass(TripCount->getReg());
     CountReg = MF->getRegInfo().createVirtualRegister(RC);
-    unsigned CopyOp = (isPPC64 && SrcRC == GPRC) ?
+    unsigned CopyOp = (isPPC64 && GPRC->hasSubClassEq(SrcRC)) ?
                         (unsigned) PPC::EXTSW_32_64 :
                         (unsigned) TargetOpcode::COPY;
     BuildMI(*Preheader, InsertPos, dl,
@@ -652,13 +702,14 @@ bool PPCCTRLoops::convertToCTRLoop(MachineLoop *L) {
     // Put the trip count in a register for transfer into the count register.
 
     int64_t CountImm = TripCount->getImm();
-    assert(!TripCount->isNeg() && "Constant trip count must be positive");
+    if (TripCount->isNeg())
+      CountImm = -CountImm;
 
     CountReg = MF->getRegInfo().createVirtualRegister(RC);
-    if (CountImm > 0xFFFF) {
+    if (abs64(CountImm) > 0x7FFF) {
       BuildMI(*Preheader, InsertPos, dl,
               TII->get(isPPC64 ? PPC::LIS8 : PPC::LIS),
-              CountReg).addImm(CountImm >> 16);
+              CountReg).addImm((CountImm >> 16) & 0xFFFF);
       unsigned CountReg1 = CountReg;
       CountReg = MF->getRegInfo().createVirtualRegister(RC);
       BuildMI(*Preheader, InsertPos, dl,
diff --git a/lib/Target/PowerPC/PPCCallingConv.td b/lib/Target/PowerPC/PPCCallingConv.td
index 3f87e883b1e4..c8a29a3d2cfe 100644
--- a/lib/Target/PowerPC/PPCCallingConv.td
+++ b/lib/Target/PowerPC/PPCCallingConv.td
@@ -27,9 +27,10 @@ def RetCC_PPC : CallingConv<[
 
   CCIfType<[i32], CCAssignToReg<[R3, R4, R5, R6, R7, R8, R9, R10]>>,
   CCIfType<[i64], CCAssignToReg<[X3, X4, X5, X6]>>,
+  CCIfType<[i128], CCAssignToReg<[X3, X4, X5, X6]>>,
   
-  CCIfType<[f32], CCAssignToReg<[F1]>>,
-  CCIfType<[f64], CCAssignToReg<[F1, F2]>>,
+  CCIfType<[f32], CCAssignToReg<[F1, F2]>>,
+  CCIfType<[f64], CCAssignToReg<[F1, F2, F3, F4]>>,
   
   // Vector types are always returned in V2.
   CCIfType<[v16i8, v8i16, v4i32, v4f32], CCAssignToReg<[V2]>>
@@ -37,49 +38,20 @@ def RetCC_PPC : CallingConv<[
 
 
 //===----------------------------------------------------------------------===//
-// PowerPC Argument Calling Conventions
-//===----------------------------------------------------------------------===//
-/*
-def CC_PPC : CallingConv<[
-  // The first 8 integer arguments are passed in integer registers.
-  CCIfType<[i32], CCAssignToReg<[R3, R4, R5, R6, R7, R8, R9, R10]>>,
-  CCIfType<[i64], CCAssignToReg<[X3, X4, X5, X6, X7, X8, X9, X10]>>,
-  
-  // Common sub-targets passes FP values in F1 - F13
-  CCIfType<[f32, f64], 
-           CCAssignToReg<[F1, F2, F3, F4, F5, F6, F7, F8,F9,F10,F11,F12,F13]>>,
-           
-  // The first 12 Vector arguments are passed in altivec registers.
-  CCIfType<[v16i8, v8i16, v4i32, v4f32],
-              CCAssignToReg<[V2, V3, V4, V5, V6, V7, V8, V9, V10,V11,V12,V13]>>
-
-/*
-  // Integer/FP values get stored in stack slots that are 8 bytes in size and
-  // 8-byte aligned if there are no more registers to hold them.
-  CCIfType<[i32, i64, f32, f64], CCAssignToStack<8, 8>>,
-  
-  // Vectors get 16-byte stack slots that are 16-byte aligned.
-  CCIfType<[v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
-              CCAssignToStack<16, 16>>*/
-]>;
-
-*/
-
-//===----------------------------------------------------------------------===//
-// PowerPC System V Release 4 ABI
+// PowerPC System V Release 4 32-bit ABI
 //===----------------------------------------------------------------------===//
 
-def CC_PPC_SVR4_Common : CallingConv<[
+def CC_PPC32_SVR4_Common : CallingConv<[
   // The ABI requires i64 to be passed in two adjacent registers with the first
   // register having an odd register number.
-  CCIfType<[i32], CCIfSplit<CCCustom<"CC_PPC_SVR4_Custom_AlignArgRegs">>>,
+  CCIfType<[i32], CCIfSplit<CCCustom<"CC_PPC32_SVR4_Custom_AlignArgRegs">>>,
 
   // The first 8 integer arguments are passed in integer registers.
   CCIfType<[i32], CCAssignToReg<[R3, R4, R5, R6, R7, R8, R9, R10]>>,
 
   // Make sure the i64 words from a long double are either both passed in
   // registers or both passed on the stack.
-  CCIfType<[f64], CCIfSplit<CCCustom<"CC_PPC_SVR4_Custom_AlignFPArgRegs">>>,
+  CCIfType<[f64], CCIfSplit<CCCustom<"CC_PPC32_SVR4_Custom_AlignFPArgRegs">>>,
   
   // FP values are passed in F1 - F8.
   CCIfType<[f32, f64], CCAssignToReg<[F1, F2, F3, F4, F5, F6, F7, F8]>>,
@@ -100,18 +72,18 @@ def CC_PPC_SVR4_Common : CallingConv<[
 // This calling convention puts vector arguments always on the stack. It is used
 // to assign vector arguments which belong to the variable portion of the
 // parameter list of a variable argument function.
-def CC_PPC_SVR4_VarArg : CallingConv<[
-  CCDelegateTo<CC_PPC_SVR4_Common>
+def CC_PPC32_SVR4_VarArg : CallingConv<[
+  CCDelegateTo<CC_PPC32_SVR4_Common>
 ]>;
 
-// In contrast to CC_PPC_SVR4_VarArg, this calling convention first tries to put
-// vector arguments in vector registers before putting them on the stack.
-def CC_PPC_SVR4 : CallingConv<[
+// In contrast to CC_PPC32_SVR4_VarArg, this calling convention first tries to
+// put vector arguments in vector registers before putting them on the stack.
+def CC_PPC32_SVR4 : CallingConv<[
   // The first 12 Vector arguments are passed in AltiVec registers.
   CCIfType<[v16i8, v8i16, v4i32, v4f32],
            CCAssignToReg<[V2, V3, V4, V5, V6, V7, V8, V9, V10, V11, V12, V13]>>,
            
-  CCDelegateTo<CC_PPC_SVR4_Common>
+  CCDelegateTo<CC_PPC32_SVR4_Common>
 ]>;  
 
 // Helper "calling convention" to handle aggregate by value arguments.
@@ -122,15 +94,15 @@ def CC_PPC_SVR4 : CallingConv<[
 // Still, the address of the aggregate copy in the callers stack frame is passed
 // in a GPR (or in the parameter list area if all GPRs are allocated) from the
 // caller to the callee. The location for the address argument is assigned by
-// the CC_PPC_SVR4 calling convention.
+// the CC_PPC32_SVR4 calling convention.
 //
-// The only purpose of CC_PPC_SVR4_Custom_Dummy is to skip arguments which are
+// The only purpose of CC_PPC32_SVR4_Custom_Dummy is to skip arguments which are
 // not passed by value.
  
-def CC_PPC_SVR4_ByVal : CallingConv<[
+def CC_PPC32_SVR4_ByVal : CallingConv<[
   CCIfByVal<CCPassByVal<4, 4>>,
   
-  CCCustom<"CC_PPC_SVR4_Custom_Dummy">
+  CCCustom<"CC_PPC32_SVR4_Custom_Dummy">
 ]>;
 
 def CSR_Darwin32 : CalleeSavedRegs<(add R13, R14, R15, R16, R17, R18, R19, R20,
@@ -164,3 +136,9 @@ def CSR_SVR464   : CalleeSavedRegs<(add X14, X15, X16, X17, X18, X19, X20, VRSAV
                                         F27, F28, F29, F30, F31, CR2, CR3, CR4,
                                         V20, V21, V22, V23, V24, V25, V26, V27,
                                         V28, V29, V30, V31)>;
+
+def CSR_NoRegs : CalleeSavedRegs<(add VRSAVE)>;
+def CSR_NoRegs_Darwin : CalleeSavedRegs<(add)>;
+
+def CSR_NoRegs_Altivec : CalleeSavedRegs<(add (sequence "V%u", 0, 31), VRSAVE)>;
+
diff --git a/lib/Target/PowerPC/PPCCodeEmitter.cpp b/lib/Target/PowerPC/PPCCodeEmitter.cpp
index 252a2d159ec3..64787185138b 100644
--- a/lib/Target/PowerPC/PPCCodeEmitter.cpp
+++ b/lib/Target/PowerPC/PPCCodeEmitter.cpp
@@ -12,15 +12,15 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "PPCTargetMachine.h"
-#include "PPCRelocations.h"
 #include "PPC.h"
-#include "llvm/Module.h"
-#include "llvm/PassManager.h"
+#include "PPCRelocations.h"
+#include "PPCTargetMachine.h"
 #include "llvm/CodeGen/JITCodeEmitter.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
+#include "llvm/IR/Module.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetOptions.h"
@@ -68,6 +68,7 @@ namespace {
     unsigned getLO16Encoding(const MachineInstr &MI, unsigned OpNo) const;
     unsigned getMemRIEncoding(const MachineInstr &MI, unsigned OpNo) const;
     unsigned getMemRIXEncoding(const MachineInstr &MI, unsigned OpNo) const;
+    unsigned getTLSRegEncoding(const MachineInstr &MI, unsigned OpNo) const;
 
     const char *getPassName() const { return "PowerPC Machine Code Emitter"; }
 
@@ -141,7 +142,7 @@ unsigned PPCCodeEmitter::get_crbitm_encoding(const MachineInstr &MI,
   assert((MI.getOpcode() == PPC::MTCRF || MI.getOpcode() == PPC::MTCRF8 ||
             MI.getOpcode() == PPC::MFOCRF) &&
          (MO.getReg() >= PPC::CR0 && MO.getReg() <= PPC::CR7));
-  return 0x80 >> getPPCRegisterNumbering(MO.getReg());
+  return 0x80 >> TM.getRegisterInfo()->getEncodingValue(MO.getReg());
 }
 
 MachineRelocation PPCCodeEmitter::GetRelocation(const MachineOperand &MO, 
@@ -243,6 +244,13 @@ unsigned PPCCodeEmitter::getMemRIXEncoding(const MachineInstr &MI,
 }
 
 
+unsigned PPCCodeEmitter::getTLSRegEncoding(const MachineInstr &MI,
+                                           unsigned OpNo) const {
+  llvm_unreachable("TLS not supported on the old JIT.");
+  return 0;
+}
+
+
 unsigned PPCCodeEmitter::getMachineOpValue(const MachineInstr &MI,
                                            const MachineOperand &MO) const {
 
@@ -252,7 +260,7 @@ unsigned PPCCodeEmitter::getMachineOpValue(const MachineInstr &MI,
     assert((MI.getOpcode() != PPC::MTCRF && MI.getOpcode() != PPC::MTCRF8 &&
              MI.getOpcode() != PPC::MFOCRF) ||
            MO.getReg() < PPC::CR0 || MO.getReg() > PPC::CR7);
-    return getPPCRegisterNumbering(MO.getReg());
+    return TM.getRegisterInfo()->getEncodingValue(MO.getReg());
   }
   
   assert(MO.isImm() &&
diff --git a/lib/Target/PowerPC/PPCFrameLowering.cpp b/lib/Target/PowerPC/PPCFrameLowering.cpp
index caf7bf2be793..3244b904ee64 100644
--- a/lib/Target/PowerPC/PPCFrameLowering.cpp
+++ b/lib/Target/PowerPC/PPCFrameLowering.cpp
@@ -12,16 +12,16 @@
 //===----------------------------------------------------------------------===//
 
 #include "PPCFrameLowering.h"
-#include "PPCInstrInfo.h"
 #include "PPCInstrBuilder.h"
+#include "PPCInstrInfo.h"
 #include "PPCMachineFunctionInfo.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
@@ -103,6 +103,7 @@ static void RemoveVRSaveCode(MachineInstr *MI) {
 // transform this into the appropriate ORI instruction.
 static void HandleVRSaveUpdate(MachineInstr *MI, const TargetInstrInfo &TII) {
   MachineFunction *MF = MI->getParent()->getParent();
+  const TargetRegisterInfo *TRI = MF->getTarget().getRegisterInfo();
   DebugLoc dl = MI->getDebugLoc();
 
   unsigned UsedRegMask = 0;
@@ -115,16 +116,25 @@ static void HandleVRSaveUpdate(MachineInstr *MI, const TargetInstrInfo &TII) {
   for (MachineRegisterInfo::livein_iterator
        I = MF->getRegInfo().livein_begin(),
        E = MF->getRegInfo().livein_end(); I != E; ++I) {
-    unsigned RegNo = getPPCRegisterNumbering(I->first);
+    unsigned RegNo = TRI->getEncodingValue(I->first);
     if (VRRegNo[RegNo] == I->first)        // If this really is a vector reg.
       UsedRegMask &= ~(1 << (31-RegNo));   // Doesn't need to be marked.
   }
-  for (MachineRegisterInfo::liveout_iterator
-       I = MF->getRegInfo().liveout_begin(),
-       E = MF->getRegInfo().liveout_end(); I != E; ++I) {
-    unsigned RegNo = getPPCRegisterNumbering(*I);
-    if (VRRegNo[RegNo] == *I)              // If this really is a vector reg.
-      UsedRegMask &= ~(1 << (31-RegNo));   // Doesn't need to be marked.
+
+  // Live out registers appear as use operands on return instructions.
+  for (MachineFunction::const_iterator BI = MF->begin(), BE = MF->end();
+       UsedRegMask != 0 && BI != BE; ++BI) {
+    const MachineBasicBlock &MBB = *BI;
+    if (MBB.empty() || !MBB.back().isReturn())
+      continue;
+    const MachineInstr &Ret = MBB.back();
+    for (unsigned I = 0, E = Ret.getNumOperands(); I != E; ++I) {
+      const MachineOperand &MO = Ret.getOperand(I);
+      if (!MO.isReg() || !PPC::VRRCRegClass.contains(MO.getReg()))
+        continue;
+      unsigned RegNo = TRI->getEncodingValue(MO.getReg());
+      UsedRegMask &= ~(1 << (31-RegNo));
+    }
   }
 
   // If no registers are used, turn this into a copy.
@@ -179,13 +189,31 @@ static bool spillsCR(const MachineFunction &MF) {
   return FuncInfo->isCRSpilled();
 }
 
+static bool spillsVRSAVE(const MachineFunction &MF) {
+  const PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
+  return FuncInfo->isVRSAVESpilled();
+}
+
+static bool hasSpills(const MachineFunction &MF) {
+  const PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
+  return FuncInfo->hasSpills();
+}
+
+static bool hasNonRISpills(const MachineFunction &MF) {
+  const PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
+  return FuncInfo->hasNonRISpills();
+}
+
 /// determineFrameLayout - Determine the size of the frame and maximum call
 /// frame size.
-void PPCFrameLowering::determineFrameLayout(MachineFunction &MF) const {
+unsigned PPCFrameLowering::determineFrameLayout(MachineFunction &MF,
+                                                bool UpdateMF,
+                                                bool UseEstimate) const {
   MachineFrameInfo *MFI = MF.getFrameInfo();
 
   // Get the number of bytes to allocate from the FrameInfo
-  unsigned FrameSize = MFI->getStackSize();
+  unsigned FrameSize =
+    UseEstimate ? MFI->estimateStackSize(MF) : MFI->getStackSize();
 
   // Get the alignments provided by the target, and the maximum alignment
   // (if any) of the fixed frame objects.
@@ -198,13 +226,14 @@ void PPCFrameLowering::determineFrameLayout(MachineFunction &MF) const {
   // to adjust the stack pointer (we fit in the Red Zone).  For 64-bit
   // SVR4, we also require a stack frame if we need to spill the CR,
   // since this spill area is addressed relative to the stack pointer.
-  bool DisableRedZone = MF.getFunction()->getFnAttributes().
-    hasAttribute(Attributes::NoRedZone);
-  // FIXME SVR4 The 32-bit SVR4 ABI has no red zone.  However, it can
-  // still generate stackless code if all local vars are reg-allocated.
-  // Try: (FrameSize <= 224
-  //       || (FrameSize == 0 && Subtarget.isPPC32 && Subtarget.isSVR4ABI()))
+  // The 32-bit SVR4 ABI has no Red Zone. However, it can still generate
+  // stackless code if all local vars are reg-allocated.
+  bool DisableRedZone = MF.getFunction()->getAttributes().
+    hasAttribute(AttributeSet::FunctionIndex, Attribute::NoRedZone);
   if (!DisableRedZone &&
+      (Subtarget.isPPC64() ||                      // 32-bit SVR4, no stack-
+       !Subtarget.isSVR4ABI() ||                   //   allocated locals.
+	FrameSize == 0) &&
       FrameSize <= 224 &&                          // Fits in red zone.
       !MFI->hasVarSizedObjects() &&                // No dynamic alloca.
       !MFI->adjustsStack() &&                      // No calls.
@@ -213,8 +242,9 @@ void PPCFrameLowering::determineFrameLayout(MachineFunction &MF) const {
 	&& spillsCR(MF)) &&
       (!ALIGN_STACK || MaxAlign <= TargetAlign)) { // No special alignment.
     // No need for frame
-    MFI->setStackSize(0);
-    return;
+    if (UpdateMF)
+      MFI->setStackSize(0);
+    return 0;
   }
 
   // Get the maximum call frame size of all the calls.
@@ -231,7 +261,8 @@ void PPCFrameLowering::determineFrameLayout(MachineFunction &MF) const {
     maxCallFrameSize = (maxCallFrameSize + AlignMask) & ~AlignMask;
 
   // Update maximum call frame size.
-  MFI->setMaxCallFrameSize(maxCallFrameSize);
+  if (UpdateMF)
+    MFI->setMaxCallFrameSize(maxCallFrameSize);
 
   // Include call frame size in total.
   FrameSize += maxCallFrameSize;
@@ -240,7 +271,10 @@ void PPCFrameLowering::determineFrameLayout(MachineFunction &MF) const {
   FrameSize = (FrameSize + AlignMask) & ~AlignMask;
 
   // Update frame info.
-  MFI->setStackSize(FrameSize);
+  if (UpdateMF)
+    MFI->setStackSize(FrameSize);
+
+  return FrameSize;
 }
 
 // hasFP - Return true if the specified function actually has a dedicated frame
@@ -261,7 +295,8 @@ bool PPCFrameLowering::needsFP(const MachineFunction &MF) const {
 
   // Naked functions have no stack frame pushed, so we don't have a frame
   // pointer.
-  if (MF.getFunction()->getFnAttributes().hasAttribute(Attributes::Naked))
+  if (MF.getFunction()->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
+                                                     Attribute::Naked))
     return false;
 
   return MF.getTarget().Options.DisableFramePointerElim(MF) ||
@@ -270,6 +305,31 @@ bool PPCFrameLowering::needsFP(const MachineFunction &MF) const {
      MF.getInfo<PPCFunctionInfo>()->hasFastCall());
 }
 
+void PPCFrameLowering::replaceFPWithRealFP(MachineFunction &MF) const {
+  bool is31 = needsFP(MF);
+  unsigned FPReg  = is31 ? PPC::R31 : PPC::R1;
+  unsigned FP8Reg = is31 ? PPC::X31 : PPC::X1;
+
+  for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
+       BI != BE; ++BI)
+    for (MachineBasicBlock::iterator MBBI = BI->end(); MBBI != BI->begin(); ) {
+      --MBBI;
+      for (unsigned I = 0, E = MBBI->getNumOperands(); I != E; ++I) {
+        MachineOperand &MO = MBBI->getOperand(I);
+        if (!MO.isReg())
+          continue;
+
+        switch (MO.getReg()) {
+        case PPC::FP:
+          MO.setReg(FPReg);
+          break;
+        case PPC::FP8:
+          MO.setReg(FP8Reg);
+          break;
+        }
+      }
+    }
+}
 
 void PPCFrameLowering::emitPrologue(MachineFunction &MF) const {
   MachineBasicBlock &MBB = MF.front();   // Prolog goes in entry BB
@@ -300,13 +360,12 @@ void PPCFrameLowering::emitPrologue(MachineFunction &MF) const {
   MBBI = MBB.begin();
 
   // Work out frame sizes.
-  // FIXME: determineFrameLayout() may change the frame size. This should be
-  // moved upper, to some hook.
-  determineFrameLayout(MF);
-  unsigned FrameSize = MFI->getStackSize();
-
+  unsigned FrameSize = determineFrameLayout(MF);
   int NegFrameSize = -FrameSize;
 
+  if (MFI->isFrameAddressTaken())
+    replaceFPWithRealFP(MF);
+
   // Get processor type.
   bool isPPC64 = Subtarget.isPPC64();
   // Get operating system
@@ -769,14 +828,15 @@ static bool MustSaveLR(const MachineFunction &MF, unsigned LR) {
 
 void
 PPCFrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
-                                                   RegScavenger *RS) const {
+                                                   RegScavenger *) const {
   const TargetRegisterInfo *RegInfo = MF.getTarget().getRegisterInfo();
 
   //  Save and clear the LR state.
   PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
   unsigned LR = RegInfo->getRARegister();
   FI->setMustSaveLR(MustSaveLR(MF, LR));
-  MF.getRegInfo().setPhysRegUnused(LR);
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+  MRI.setPhysRegUnused(LR);
 
   //  Save R31 if necessary
   int FPSI = FI->getFramePointerSaveIndex();
@@ -801,29 +861,24 @@ PPCFrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
     MFI->CreateFixedObject(-1 * TCSPDelta, TCSPDelta, true);
   }
 
-  // Reserve a slot closest to SP or frame pointer if we have a dynalloc or
-  // a large stack, which will require scavenging a register to materialize a
-  // large offset.
-  // FIXME: this doesn't actually check stack size, so is a bit pessimistic
-  // FIXME: doesn't detect whether or not we need to spill vXX, which requires
-  //        r0 for now.
-
-  if (RegInfo->requiresRegisterScavenging(MF))
-    if (needsFP(MF) || spillsCR(MF)) {
-      const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
-      const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
-      const TargetRegisterClass *RC = isPPC64 ? G8RC : GPRC;
-      RS->setScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
-                                                         RC->getAlignment(),
-                                                         false));
-    }
+  // For 32-bit SVR4, allocate the nonvolatile CR spill slot iff the 
+  // function uses CR 2, 3, or 4.
+  if (!isPPC64 && !isDarwinABI && 
+      (MRI.isPhysRegUsed(PPC::CR2) ||
+       MRI.isPhysRegUsed(PPC::CR3) ||
+       MRI.isPhysRegUsed(PPC::CR4))) {
+    int FrameIdx = MFI->CreateFixedObject((uint64_t)4, (int64_t)-4, true);
+    FI->setCRSpillFrameIndex(FrameIdx);
+  }
 }
 
-void PPCFrameLowering::processFunctionBeforeFrameFinalized(MachineFunction &MF)
-                                                                        const {
+void PPCFrameLowering::processFunctionBeforeFrameFinalized(MachineFunction &MF,
+                                                       RegScavenger *RS) const {
   // Early exit if not using the SVR4 ABI.
-  if (!Subtarget.isSVR4ABI())
+  if (!Subtarget.isSVR4ABI()) {
+    addScavengingSpillSlot(MF, RS);
     return;
+  }
 
   // Get callee saved register information.
   MachineFrameInfo *FFI = MF.getFrameInfo();
@@ -831,6 +886,7 @@ void PPCFrameLowering::processFunctionBeforeFrameFinalized(MachineFunction &MF)
 
   // Early exit if no callee saved registers are modified!
   if (CSI.empty() && !needsFP(MF)) {
+    addScavengingSpillSlot(MF, RS);
     return;
   }
 
@@ -895,6 +951,7 @@ void PPCFrameLowering::processFunctionBeforeFrameFinalized(MachineFunction &MF)
   }
 
   PPCFunctionInfo *PFI = MF.getInfo<PPCFunctionInfo>();
+  const TargetRegisterInfo *TRI = MF.getTarget().getRegisterInfo();
 
   int64_t LowerBound = 0;
 
@@ -914,7 +971,7 @@ void PPCFrameLowering::processFunctionBeforeFrameFinalized(MachineFunction &MF)
       FFI->setObjectOffset(FI, LowerBound + FFI->getObjectOffset(FI));
     }
 
-    LowerBound -= (31 - getPPCRegisterNumbering(MinFPR) + 1) * 8;
+    LowerBound -= (31 - TRI->getEncodingValue(MinFPR) + 1) * 8;
   }
 
   // Check whether the frame pointer register is allocated. If so, make sure it
@@ -948,8 +1005,8 @@ void PPCFrameLowering::processFunctionBeforeFrameFinalized(MachineFunction &MF)
     }
 
     unsigned MinReg =
-      std::min<unsigned>(getPPCRegisterNumbering(MinGPR),
-                         getPPCRegisterNumbering(MinG8R));
+      std::min<unsigned>(TRI->getEncodingValue(MinGPR),
+                         TRI->getEncodingValue(MinG8R));
 
     if (Subtarget.isPPC64()) {
       LowerBound -= (31 - MinReg + 1) * 8;
@@ -1009,6 +1066,44 @@ void PPCFrameLowering::processFunctionBeforeFrameFinalized(MachineFunction &MF)
       FFI->setObjectOffset(FI, LowerBound + FFI->getObjectOffset(FI));
     }
   }
+
+  addScavengingSpillSlot(MF, RS);
+}
+
+void
+PPCFrameLowering::addScavengingSpillSlot(MachineFunction &MF,
+                                         RegScavenger *RS) const {
+  // Reserve a slot closest to SP or frame pointer if we have a dynalloc or
+  // a large stack, which will require scavenging a register to materialize a
+  // large offset.
+
+  // We need to have a scavenger spill slot for spills if the frame size is
+  // large. In case there is no free register for large-offset addressing,
+  // this slot is used for the necessary emergency spill. Also, we need the
+  // slot for dynamic stack allocations.
+
+  // The scavenger might be invoked if the frame offset does not fit into
+  // the 16-bit immediate. We don't know the complete frame size here
+  // because we've not yet computed callee-saved register spills or the
+  // needed alignment padding.
+  unsigned StackSize = determineFrameLayout(MF, false, true);
+  MachineFrameInfo *MFI = MF.getFrameInfo();
+  if (MFI->hasVarSizedObjects() || spillsCR(MF) || spillsVRSAVE(MF) ||
+      hasNonRISpills(MF) || (hasSpills(MF) && !isInt<16>(StackSize))) {
+    const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
+    const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
+    const TargetRegisterClass *RC = Subtarget.isPPC64() ? G8RC : GPRC;
+    RS->addScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
+                                                       RC->getAlignment(),
+                                                       false));
+
+    // These kinds of spills might need two registers.
+    if (spillsCR(MF) || spillsVRSAVE(MF))
+      RS->addScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
+                                                         RC->getAlignment(),
+                                                         false));
+
+  }
 }
 
 bool 
@@ -1046,8 +1141,8 @@ PPCFrameLowering::spillCalleeSavedRegisters(MachineBasicBlock &MBB,
       // save slot via GPR12 (available in the prolog for 32- and 64-bit).
       if (Subtarget.isPPC64()) {
 	// 64-bit:  SP+8
-	MBB.insert(MI, BuildMI(*MF, DL, TII.get(PPC::MFCR), PPC::X12));
-	MBB.insert(MI, BuildMI(*MF, DL, TII.get(PPC::STW))
+	MBB.insert(MI, BuildMI(*MF, DL, TII.get(PPC::MFCR8), PPC::X12));
+	MBB.insert(MI, BuildMI(*MF, DL, TII.get(PPC::STW8))
 			       .addReg(PPC::X12,
 				       getKillRegState(true))
 			       .addImm(8)
@@ -1087,7 +1182,7 @@ restoreCRs(bool isPPC64, bool CR2Spilled, bool CR3Spilled, bool CR4Spilled,
 
   if (isPPC64) {
     // 64-bit:  SP+8
-    MBB.insert(MI, BuildMI(*MF, DL, TII.get(PPC::LWZ), PPC::X12)
+    MBB.insert(MI, BuildMI(*MF, DL, TII.get(PPC::LWZ8), PPC::X12)
 	       .addImm(8)
 	       .addReg(PPC::X1));
     RestoreOp = PPC::MTCRF8;
@@ -1103,15 +1198,56 @@ restoreCRs(bool isPPC64, bool CR2Spilled, bool CR3Spilled, bool CR4Spilled,
   
   if (CR2Spilled)
     MBB.insert(MI, BuildMI(*MF, DL, TII.get(RestoreOp), PPC::CR2)
-	       .addReg(MoveReg));
+               .addReg(MoveReg, getKillRegState(!CR3Spilled && !CR4Spilled)));
 
   if (CR3Spilled)
     MBB.insert(MI, BuildMI(*MF, DL, TII.get(RestoreOp), PPC::CR3)
-	       .addReg(MoveReg));
+               .addReg(MoveReg, getKillRegState(!CR4Spilled)));
 
   if (CR4Spilled)
     MBB.insert(MI, BuildMI(*MF, DL, TII.get(RestoreOp), PPC::CR4)
-	       .addReg(MoveReg));
+               .addReg(MoveReg, getKillRegState(true)));
+}
+
+void PPCFrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  const PPCInstrInfo &TII =
+    *static_cast<const PPCInstrInfo*>(MF.getTarget().getInstrInfo());
+  if (MF.getTarget().Options.GuaranteedTailCallOpt &&
+      I->getOpcode() == PPC::ADJCALLSTACKUP) {
+    // Add (actually subtract) back the amount the callee popped on return.
+    if (int CalleeAmt =  I->getOperand(1).getImm()) {
+      bool is64Bit = Subtarget.isPPC64();
+      CalleeAmt *= -1;
+      unsigned StackReg = is64Bit ? PPC::X1 : PPC::R1;
+      unsigned TmpReg = is64Bit ? PPC::X0 : PPC::R0;
+      unsigned ADDIInstr = is64Bit ? PPC::ADDI8 : PPC::ADDI;
+      unsigned ADDInstr = is64Bit ? PPC::ADD8 : PPC::ADD4;
+      unsigned LISInstr = is64Bit ? PPC::LIS8 : PPC::LIS;
+      unsigned ORIInstr = is64Bit ? PPC::ORI8 : PPC::ORI;
+      MachineInstr *MI = I;
+      DebugLoc dl = MI->getDebugLoc();
+
+      if (isInt<16>(CalleeAmt)) {
+        BuildMI(MBB, I, dl, TII.get(ADDIInstr), StackReg)
+          .addReg(StackReg, RegState::Kill)
+          .addImm(CalleeAmt);
+      } else {
+        MachineBasicBlock::iterator MBBI = I;
+        BuildMI(MBB, MBBI, dl, TII.get(LISInstr), TmpReg)
+          .addImm(CalleeAmt >> 16);
+        BuildMI(MBB, MBBI, dl, TII.get(ORIInstr), TmpReg)
+          .addReg(TmpReg, RegState::Kill)
+          .addImm(CalleeAmt & 0xFFFF);
+        BuildMI(MBB, MBBI, dl, TII.get(ADDInstr), StackReg)
+          .addReg(StackReg, RegState::Kill)
+          .addReg(TmpReg);
+      }
+    }
+  }
+  // Simply discard ADJCALLSTACKDOWN, ADJCALLSTACKUP instructions.
+  MBB.erase(I);
 }
 
 bool 
diff --git a/lib/Target/PowerPC/PPCFrameLowering.h b/lib/Target/PowerPC/PPCFrameLowering.h
index 4d957b91c7bb..6f5f9368c6c6 100644
--- a/lib/Target/PowerPC/PPCFrameLowering.h
+++ b/lib/Target/PowerPC/PPCFrameLowering.h
@@ -15,9 +15,9 @@
 
 #include "PPC.h"
 #include "PPCSubtarget.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/Target/TargetFrameLowering.h"
 #include "llvm/Target/TargetMachine.h"
-#include "llvm/ADT/STLExtras.h"
 
 namespace llvm {
   class PPCSubtarget;
@@ -27,11 +27,14 @@ class PPCFrameLowering: public TargetFrameLowering {
 
 public:
   PPCFrameLowering(const PPCSubtarget &sti)
-    : TargetFrameLowering(TargetFrameLowering::StackGrowsDown, 16, 0),
+    : TargetFrameLowering(TargetFrameLowering::StackGrowsDown,
+        (sti.hasQPX() || sti.isBGQ()) ? 32 : 16, 0),
       Subtarget(sti) {
   }
 
-  void determineFrameLayout(MachineFunction &MF) const;
+  unsigned determineFrameLayout(MachineFunction &MF,
+                                bool UpdateMF = true,
+                                bool UseEstimate = false) const;
 
   /// emitProlog/emitEpilog - These methods insert prolog and epilog code into
   /// the function.
@@ -40,16 +43,23 @@ public:
 
   bool hasFP(const MachineFunction &MF) const;
   bool needsFP(const MachineFunction &MF) const;
+  void replaceFPWithRealFP(MachineFunction &MF) const;
 
   void processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
                                             RegScavenger *RS = NULL) const;
-  void processFunctionBeforeFrameFinalized(MachineFunction &MF) const;
+  void processFunctionBeforeFrameFinalized(MachineFunction &MF,
+                                       RegScavenger *RS = NULL) const;
+  void addScavengingSpillSlot(MachineFunction &MF, RegScavenger *RS) const;
 
   bool spillCalleeSavedRegisters(MachineBasicBlock &MBB,
                                  MachineBasicBlock::iterator MI,
                                  const std::vector<CalleeSavedInfo> &CSI,
                                  const TargetRegisterInfo *TRI) const;
 
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I) const;
+
   bool restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
                                    MachineBasicBlock::iterator MI,
                                    const std::vector<CalleeSavedInfo> &CSI,
@@ -139,6 +149,9 @@ public:
       return 0;
     }
 
+    // Note that the offsets here overlap, but this is fixed up in
+    // processFunctionBeforeFrameFinalized.
+
     static const SpillSlot Offsets[] = {
       // Floating-point register save area offsets.
       {PPC::F31, -8},
diff --git a/lib/Target/PowerPC/PPCHazardRecognizers.cpp b/lib/Target/PowerPC/PPCHazardRecognizers.cpp
index 6ed1fb9e6a3c..4bf1e3396429 100644
--- a/lib/Target/PowerPC/PPCHazardRecognizers.cpp
+++ b/lib/Target/PowerPC/PPCHazardRecognizers.cpp
@@ -179,7 +179,7 @@ getHazardType(SUnit *SU, int Stalls) {
   }
 
   // Do not allow MTCTR and BCTRL to be in the same dispatch group.
-  if (HasCTRSet && (Opcode == PPC::BCTRL_Darwin || Opcode == PPC::BCTRL_SVR4))
+  if (HasCTRSet && Opcode == PPC::BCTRL)
     return NoopHazard;
 
   // If this is a load following a store, make sure it's not to the same or
diff --git a/lib/Target/PowerPC/PPCISelDAGToDAG.cpp b/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
index 254fea67fc4e..95efc11b53c1 100644
--- a/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
+++ b/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
@@ -14,24 +14,30 @@
 
 #define DEBUG_TYPE "ppc-codegen"
 #include "PPC.h"
-#include "PPCTargetMachine.h"
 #include "MCTargetDesc/PPCPredicates.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "PPCTargetMachine.h"
 #include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Constants.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalValue.h"
-#include "llvm/Intrinsics.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/Debug.h"
-#include "llvm/Support/MathExtras.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetOptions.h"
 using namespace llvm;
 
+namespace llvm {
+  void initializePPCDAGToDAGISelPass(PassRegistry&);
+}
+
 namespace {
   //===--------------------------------------------------------------------===//
   /// PPCDAGToDAGISel - PPC specific code to select PPC machine
@@ -46,7 +52,9 @@ namespace {
     explicit PPCDAGToDAGISel(PPCTargetMachine &tm)
       : SelectionDAGISel(tm), TM(tm),
         PPCLowering(*TM.getTargetLowering()),
-        PPCSubTarget(*TM.getSubtargetImpl()) {}
+        PPCSubTarget(*TM.getSubtargetImpl()) {
+      initializePPCDAGToDAGISelPass(*PassRegistry::getPassRegistry());
+    }
 
     virtual bool runOnMachineFunction(MachineFunction &MF) {
       // Make sure we re-emit a set of the global base reg if necessary
@@ -59,6 +67,8 @@ namespace {
       return true;
     }
 
+    virtual void PostprocessISelDAG();
+
     /// getI32Imm - Return a target constant with the specified value, of type
     /// i32.
     inline SDValue getI32Imm(unsigned Imm) {
@@ -110,10 +120,10 @@ namespace {
     }
 
     /// SelectAddrImmOffs - Return true if the operand is valid for a preinc
-    /// immediate field.  Because preinc imms have already been validated, just
-    /// accept it.
+    /// immediate field.  Note that the operand at this point is already the
+    /// result of a prior SelectAddressRegImm call.
     bool SelectAddrImmOffs(SDValue N, SDValue &Out) const {
-      if (isa<ConstantSDNode>(N) || N.getOpcode() == PPCISD::Lo ||
+      if (N.getOpcode() == ISD::TargetConstant ||
           N.getOpcode() == ISD::TargetGlobalAddress) {
         Out = N;
         return true;
@@ -122,18 +132,6 @@ namespace {
       return false;
     }
 
-    /// SelectAddrIdxOffs - Return true if the operand is valid for a preinc
-    /// index field.  Because preinc imms have already been validated, just
-    /// accept it.
-    bool SelectAddrIdxOffs(SDValue N, SDValue &Out) const {
-      if (isa<ConstantSDNode>(N) || N.getOpcode() == PPCISD::Lo ||
-          N.getOpcode() == ISD::TargetGlobalAddress)
-        return false;
-
-      Out = N;
-      return true;
-    }
-
     /// SelectAddrIdx - Given the specified addressed, check to see if it can be
     /// represented as an indexed [r+r] operation.  Returns false if it can
     /// be represented by [r+imm], which are preferred.
@@ -154,6 +152,12 @@ namespace {
       return PPCLowering.SelectAddressRegImmShift(N, Disp, Base, *CurDAG);
     }
 
+    // Select an address into a single register.
+    bool SelectAddr(SDValue N, SDValue &Base) {
+      Base = N;
+      return true;
+    }
+
     /// SelectInlineAsmMemoryOperand - Implement addressing mode selection for
     /// inline asm expressions.  It is always correct to compute the value into
     /// a register.  The case of adding a (possibly relocatable) constant to a
@@ -1040,7 +1044,7 @@ SDNode *PPCDAGToDAGISel::Select(SDNode *N) {
       break;
 
     SDValue Offset = LD->getOffset();
-    if (isa<ConstantSDNode>(Offset) ||
+    if (Offset.getOpcode() == ISD::TargetConstant ||
         Offset.getOpcode() == ISD::TargetGlobalAddress) {
 
       unsigned Opcode;
@@ -1107,7 +1111,7 @@ SDNode *PPCDAGToDAGISel::Select(SDNode *N) {
 
       SDValue Chain = LD->getChain();
       SDValue Base = LD->getBasePtr();
-      SDValue Ops[] = { Offset, Base, Chain };
+      SDValue Ops[] = { Base, Offset, Chain };
       return CurDAG->getMachineNode(Opcode, dl, LD->getValueType(0),
                                     PPCLowering.getPointerTy(),
                                     MVT::Other, Ops, 3);
@@ -1268,11 +1272,277 @@ SDNode *PPCDAGToDAGISel::Select(SDNode *N) {
                                            Chain), 0);
     return CurDAG->SelectNodeTo(N, Reg, MVT::Other, Chain);
   }
+  case PPCISD::TOC_ENTRY: {
+    assert (PPCSubTarget.isPPC64() && "Only supported for 64-bit ABI");
+
+    // For medium and large code model, we generate two instructions as
+    // described below.  Otherwise we allow SelectCodeCommon to handle this,
+    // selecting one of LDtoc, LDtocJTI, and LDtocCPT.
+    CodeModel::Model CModel = TM.getCodeModel();
+    if (CModel != CodeModel::Medium && CModel != CodeModel::Large)
+      break;
+
+    // The first source operand is a TargetGlobalAddress or a
+    // TargetJumpTable.  If it is an externally defined symbol, a symbol
+    // with common linkage, a function address, or a jump table address,
+    // or if we are generating code for large code model, we generate:
+    //   LDtocL(<ga:@sym>, ADDIStocHA(%X2, <ga:@sym>))
+    // Otherwise we generate:
+    //   ADDItocL(ADDIStocHA(%X2, <ga:@sym>), <ga:@sym>)
+    SDValue GA = N->getOperand(0);
+    SDValue TOCbase = N->getOperand(1);
+    SDNode *Tmp = CurDAG->getMachineNode(PPC::ADDIStocHA, dl, MVT::i64,
+                                        TOCbase, GA);
+
+    if (isa<JumpTableSDNode>(GA) || CModel == CodeModel::Large)
+      return CurDAG->getMachineNode(PPC::LDtocL, dl, MVT::i64, GA,
+                                    SDValue(Tmp, 0));
+
+    if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(GA)) {
+      const GlobalValue *GValue = G->getGlobal();
+      const GlobalAlias *GAlias = dyn_cast<GlobalAlias>(GValue);
+      const GlobalValue *RealGValue = GAlias ?
+        GAlias->resolveAliasedGlobal(false) : GValue;
+      const GlobalVariable *GVar = dyn_cast<GlobalVariable>(RealGValue);
+      assert((GVar || isa<Function>(RealGValue)) &&
+             "Unexpected global value subclass!");
+
+      // An external variable is one without an initializer.  For these,
+      // for variables with common linkage, and for Functions, generate
+      // the LDtocL form.
+      if (!GVar || !GVar->hasInitializer() || RealGValue->hasCommonLinkage() ||
+          RealGValue->hasAvailableExternallyLinkage())
+        return CurDAG->getMachineNode(PPC::LDtocL, dl, MVT::i64, GA,
+                                      SDValue(Tmp, 0));
+    }
+
+    return CurDAG->getMachineNode(PPC::ADDItocL, dl, MVT::i64,
+                                  SDValue(Tmp, 0), GA);
+  }
+  case PPCISD::VADD_SPLAT: {
+    // This expands into one of three sequences, depending on whether
+    // the first operand is odd or even, positive or negative.
+    assert(isa<ConstantSDNode>(N->getOperand(0)) &&
+           isa<ConstantSDNode>(N->getOperand(1)) &&
+           "Invalid operand on VADD_SPLAT!");
+
+    int Elt     = N->getConstantOperandVal(0);
+    int EltSize = N->getConstantOperandVal(1);
+    unsigned Opc1, Opc2, Opc3;
+    EVT VT;
+
+    if (EltSize == 1) {
+      Opc1 = PPC::VSPLTISB;
+      Opc2 = PPC::VADDUBM;
+      Opc3 = PPC::VSUBUBM;
+      VT = MVT::v16i8;
+    } else if (EltSize == 2) {
+      Opc1 = PPC::VSPLTISH;
+      Opc2 = PPC::VADDUHM;
+      Opc3 = PPC::VSUBUHM;
+      VT = MVT::v8i16;
+    } else {
+      assert(EltSize == 4 && "Invalid element size on VADD_SPLAT!");
+      Opc1 = PPC::VSPLTISW;
+      Opc2 = PPC::VADDUWM;
+      Opc3 = PPC::VSUBUWM;
+      VT = MVT::v4i32;
+    }
+
+    if ((Elt & 1) == 0) {
+      // Elt is even, in the range [-32,-18] + [16,30].
+      //
+      // Convert: VADD_SPLAT elt, size
+      // Into:    tmp = VSPLTIS[BHW] elt
+      //          VADDU[BHW]M tmp, tmp
+      // Where:   [BHW] = B for size = 1, H for size = 2, W for size = 4
+      SDValue EltVal = getI32Imm(Elt >> 1);
+      SDNode *Tmp = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
+      SDValue TmpVal = SDValue(Tmp, 0);
+      return CurDAG->getMachineNode(Opc2, dl, VT, TmpVal, TmpVal);
+
+    } else if (Elt > 0) {
+      // Elt is odd and positive, in the range [17,31].
+      //
+      // Convert: VADD_SPLAT elt, size
+      // Into:    tmp1 = VSPLTIS[BHW] elt-16
+      //          tmp2 = VSPLTIS[BHW] -16
+      //          VSUBU[BHW]M tmp1, tmp2
+      SDValue EltVal = getI32Imm(Elt - 16);
+      SDNode *Tmp1 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
+      EltVal = getI32Imm(-16);
+      SDNode *Tmp2 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
+      return CurDAG->getMachineNode(Opc3, dl, VT, SDValue(Tmp1, 0),
+                                    SDValue(Tmp2, 0));
+
+    } else {
+      // Elt is odd and negative, in the range [-31,-17].
+      //
+      // Convert: VADD_SPLAT elt, size
+      // Into:    tmp1 = VSPLTIS[BHW] elt+16
+      //          tmp2 = VSPLTIS[BHW] -16
+      //          VADDU[BHW]M tmp1, tmp2
+      SDValue EltVal = getI32Imm(Elt + 16);
+      SDNode *Tmp1 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
+      EltVal = getI32Imm(-16);
+      SDNode *Tmp2 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
+      return CurDAG->getMachineNode(Opc2, dl, VT, SDValue(Tmp1, 0),
+                                    SDValue(Tmp2, 0));
+    }
+  }
   }
 
   return SelectCode(N);
 }
 
+/// PostProcessISelDAG - Perform some late peephole optimizations
+/// on the DAG representation.
+void PPCDAGToDAGISel::PostprocessISelDAG() {
+
+  // Skip peepholes at -O0.
+  if (TM.getOptLevel() == CodeGenOpt::None)
+    return;
+
+  // These optimizations are currently supported only for 64-bit SVR4.
+  if (PPCSubTarget.isDarwin() || !PPCSubTarget.isPPC64())
+    return;
+
+  SelectionDAG::allnodes_iterator Position(CurDAG->getRoot().getNode());
+  ++Position;
+
+  while (Position != CurDAG->allnodes_begin()) {
+    SDNode *N = --Position;
+    // Skip dead nodes and any non-machine opcodes.
+    if (N->use_empty() || !N->isMachineOpcode())
+      continue;
+
+    unsigned FirstOp;
+    unsigned StorageOpcode = N->getMachineOpcode();
+
+    switch (StorageOpcode) {
+    default: continue;
+
+    case PPC::LBZ:
+    case PPC::LBZ8:
+    case PPC::LD:
+    case PPC::LFD:
+    case PPC::LFS:
+    case PPC::LHA:
+    case PPC::LHA8:
+    case PPC::LHZ:
+    case PPC::LHZ8:
+    case PPC::LWA:
+    case PPC::LWZ:
+    case PPC::LWZ8:
+      FirstOp = 0;
+      break;
+
+    case PPC::STB:
+    case PPC::STB8:
+    case PPC::STD:
+    case PPC::STFD:
+    case PPC::STFS:
+    case PPC::STH:
+    case PPC::STH8:
+    case PPC::STW:
+    case PPC::STW8:
+      FirstOp = 1;
+      break;
+    }
+
+    // If this is a load or store with a zero offset, we may be able to
+    // fold an add-immediate into the memory operation.
+    if (!isa<ConstantSDNode>(N->getOperand(FirstOp)) ||
+        N->getConstantOperandVal(FirstOp) != 0)
+      continue;
+
+    SDValue Base = N->getOperand(FirstOp + 1);
+    if (!Base.isMachineOpcode())
+      continue;
+
+    unsigned Flags = 0;
+    bool ReplaceFlags = true;
+
+    // When the feeding operation is an add-immediate of some sort,
+    // determine whether we need to add relocation information to the
+    // target flags on the immediate operand when we fold it into the
+    // load instruction.
+    //
+    // For something like ADDItocL, the relocation information is
+    // inferred from the opcode; when we process it in the AsmPrinter,
+    // we add the necessary relocation there.  A load, though, can receive
+    // relocation from various flavors of ADDIxxx, so we need to carry
+    // the relocation information in the target flags.
+    switch (Base.getMachineOpcode()) {
+    default: continue;
+
+    case PPC::ADDI8:
+    case PPC::ADDI:
+      // In some cases (such as TLS) the relocation information
+      // is already in place on the operand, so copying the operand
+      // is sufficient.
+      ReplaceFlags = false;
+      // For these cases, the immediate may not be divisible by 4, in
+      // which case the fold is illegal for DS-form instructions.  (The
+      // other cases provide aligned addresses and are always safe.)
+      if ((StorageOpcode == PPC::LWA ||
+           StorageOpcode == PPC::LD  ||
+           StorageOpcode == PPC::STD) &&
+          (!isa<ConstantSDNode>(Base.getOperand(1)) ||
+           Base.getConstantOperandVal(1) % 4 != 0))
+        continue;
+      break;
+    case PPC::ADDIdtprelL:
+      Flags = PPCII::MO_DTPREL16_LO;
+      break;
+    case PPC::ADDItlsldL:
+      Flags = PPCII::MO_TLSLD16_LO;
+      break;
+    case PPC::ADDItocL:
+      Flags = PPCII::MO_TOC16_LO;
+      break;
+    }
+
+    // We found an opportunity.  Reverse the operands from the add
+    // immediate and substitute them into the load or store.  If
+    // needed, update the target flags for the immediate operand to
+    // reflect the necessary relocation information.
+    DEBUG(dbgs() << "Folding add-immediate into mem-op:\nBase:    ");
+    DEBUG(Base->dump(CurDAG));
+    DEBUG(dbgs() << "\nN: ");
+    DEBUG(N->dump(CurDAG));
+    DEBUG(dbgs() << "\n");
+
+    SDValue ImmOpnd = Base.getOperand(1);
+
+    // If the relocation information isn't already present on the
+    // immediate operand, add it now.
+    if (ReplaceFlags) {
+      if (GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(ImmOpnd)) {
+        DebugLoc dl = GA->getDebugLoc();
+        const GlobalValue *GV = GA->getGlobal();
+        ImmOpnd = CurDAG->getTargetGlobalAddress(GV, dl, MVT::i64, 0, Flags);
+      } else if (ConstantPoolSDNode *CP =
+                 dyn_cast<ConstantPoolSDNode>(ImmOpnd)) {
+        const Constant *C = CP->getConstVal();
+        ImmOpnd = CurDAG->getTargetConstantPool(C, MVT::i64,
+                                                CP->getAlignment(),
+                                                0, Flags);
+      }
+    }
+
+    if (FirstOp == 1) // Store
+      (void)CurDAG->UpdateNodeOperands(N, N->getOperand(0), ImmOpnd,
+                                       Base.getOperand(0), N->getOperand(3));
+    else // Load
+      (void)CurDAG->UpdateNodeOperands(N, ImmOpnd, Base.getOperand(0),
+                                       N->getOperand(2));
+
+    // The add-immediate may now be dead, in which case remove it.
+    if (Base.getNode()->use_empty())
+      CurDAG->RemoveDeadNode(Base.getNode());
+  }
+}
 
 
 /// createPPCISelDag - This pass converts a legalized DAG into a
@@ -1282,3 +1552,14 @@ FunctionPass *llvm::createPPCISelDag(PPCTargetMachine &TM) {
   return new PPCDAGToDAGISel(TM);
 }
 
+static void initializePassOnce(PassRegistry &Registry) {
+  const char *Name = "PowerPC DAG->DAG Pattern Instruction Selection";
+  PassInfo *PI = new PassInfo(Name, "ppc-codegen", &SelectionDAGISel::ID, 0,
+                              false, false);
+  Registry.registerPass(*PI, true);
+}
+
+void llvm::initializePPCDAGToDAGISelPass(PassRegistry &Registry) {
+  CALL_ONCE_INITIALIZATION(initializePassOnce);
+}
+
diff --git a/lib/Target/PowerPC/PPCISelLowering.cpp b/lib/Target/PowerPC/PPCISelLowering.cpp
index adf78d5233ae..16fc8a0e3726 100644
--- a/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -12,15 +12,10 @@
 //===----------------------------------------------------------------------===//
 
 #include "PPCISelLowering.h"
+#include "MCTargetDesc/PPCPredicates.h"
 #include "PPCMachineFunctionInfo.h"
 #include "PPCPerfectShuffle.h"
 #include "PPCTargetMachine.h"
-#include "MCTargetDesc/PPCPredicates.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
@@ -29,6 +24,11 @@
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/MathExtras.h"
@@ -36,20 +36,20 @@
 #include "llvm/Target/TargetOptions.h"
 using namespace llvm;
 
-static bool CC_PPC_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
-                                     CCValAssign::LocInfo &LocInfo,
-                                     ISD::ArgFlagsTy &ArgFlags,
-                                     CCState &State);
-static bool CC_PPC_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
-                                            MVT &LocVT,
-                                            CCValAssign::LocInfo &LocInfo,
-                                            ISD::ArgFlagsTy &ArgFlags,
-                                            CCState &State);
-static bool CC_PPC_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
+static bool CC_PPC32_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+                                       CCValAssign::LocInfo &LocInfo,
+                                       ISD::ArgFlagsTy &ArgFlags,
+                                       CCState &State);
+static bool CC_PPC32_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
                                               MVT &LocVT,
                                               CCValAssign::LocInfo &LocInfo,
                                               ISD::ArgFlagsTy &ArgFlags,
                                               CCState &State);
+static bool CC_PPC32_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
+                                                MVT &LocVT,
+                                                CCValAssign::LocInfo &LocInfo,
+                                                ISD::ArgFlagsTy &ArgFlags,
+                                                CCState &State);
 
 static cl::opt<bool> DisablePPCPreinc("disable-ppc-preinc",
 cl::desc("disable preincrement load/store generation on PPC"), cl::Hidden);
@@ -57,6 +57,9 @@ cl::desc("disable preincrement load/store generation on PPC"), cl::Hidden);
 static cl::opt<bool> DisableILPPref("disable-ppc-ilp-pref",
 cl::desc("disable setting the node scheduling preference to ILP on PPC"), cl::Hidden);
 
+static cl::opt<bool> DisablePPCUnaligned("disable-ppc-unaligned",
+cl::desc("disable unaligned load/store generation on PPC"), cl::Hidden);
+
 static TargetLoweringObjectFile *CreateTLOF(const PPCTargetMachine &TM) {
   if (TM.getSubtargetImpl()->isDarwin())
     return new TargetLoweringObjectFileMachO();
@@ -67,6 +70,7 @@ static TargetLoweringObjectFile *CreateTLOF(const PPCTargetMachine &TM) {
 PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
   : TargetLowering(TM, CreateTLOF(TM)), PPCSubTarget(*TM.getSubtargetImpl()) {
   const PPCSubtarget *Subtarget = &TM.getSubtarget<PPCSubtarget>();
+  PPCRegInfo = TM.getRegisterInfo();
 
   setPow2DivIsCheap();
 
@@ -112,6 +116,7 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
   setOperationAction(ISD::FTRUNC, MVT::ppcf128, Expand);
   setOperationAction(ISD::FRINT,  MVT::ppcf128, Expand);
   setOperationAction(ISD::FNEARBYINT, MVT::ppcf128, Expand);
+  setOperationAction(ISD::FREM, MVT::ppcf128, Expand);
 
   // PowerPC has no SREM/UREM instructions
   setOperationAction(ISD::SREM, MVT::i32, Expand);
@@ -132,11 +137,13 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
   // We don't support sin/cos/sqrt/fmod/pow
   setOperationAction(ISD::FSIN , MVT::f64, Expand);
   setOperationAction(ISD::FCOS , MVT::f64, Expand);
+  setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
   setOperationAction(ISD::FREM , MVT::f64, Expand);
   setOperationAction(ISD::FPOW , MVT::f64, Expand);
   setOperationAction(ISD::FMA  , MVT::f64, Legal);
   setOperationAction(ISD::FSIN , MVT::f32, Expand);
   setOperationAction(ISD::FCOS , MVT::f32, Expand);
+  setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
   setOperationAction(ISD::FREM , MVT::f32, Expand);
   setOperationAction(ISD::FPOW , MVT::f32, Expand);
   setOperationAction(ISD::FMA  , MVT::f32, Legal);
@@ -144,26 +151,58 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
   setOperationAction(ISD::FLT_ROUNDS_, MVT::i32, Custom);
 
   // If we're enabling GP optimizations, use hardware square root
-  if (!Subtarget->hasFSQRT()) {
+  if (!Subtarget->hasFSQRT() &&
+      !(TM.Options.UnsafeFPMath &&
+        Subtarget->hasFRSQRTE() && Subtarget->hasFRE()))
     setOperationAction(ISD::FSQRT, MVT::f64, Expand);
+
+  if (!Subtarget->hasFSQRT() &&
+      !(TM.Options.UnsafeFPMath &&
+        Subtarget->hasFRSQRTES() && Subtarget->hasFRES()))
     setOperationAction(ISD::FSQRT, MVT::f32, Expand);
-  }
 
   setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
   setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
 
+  if (Subtarget->hasFPRND()) {
+    setOperationAction(ISD::FFLOOR, MVT::f64, Legal);
+    setOperationAction(ISD::FCEIL,  MVT::f64, Legal);
+    setOperationAction(ISD::FTRUNC, MVT::f64, Legal);
+
+    setOperationAction(ISD::FFLOOR, MVT::f32, Legal);
+    setOperationAction(ISD::FCEIL,  MVT::f32, Legal);
+    setOperationAction(ISD::FTRUNC, MVT::f32, Legal);
+
+    // frin does not implement "ties to even." Thus, this is safe only in
+    // fast-math mode.
+    if (TM.Options.UnsafeFPMath) {
+      setOperationAction(ISD::FNEARBYINT, MVT::f64, Legal);
+      setOperationAction(ISD::FNEARBYINT, MVT::f32, Legal);
+
+      // These need to set FE_INEXACT, and use a custom inserter.
+      setOperationAction(ISD::FRINT, MVT::f64, Legal);
+      setOperationAction(ISD::FRINT, MVT::f32, Legal);
+    }
+  }
+
   // PowerPC does not have BSWAP, CTPOP or CTTZ
   setOperationAction(ISD::BSWAP, MVT::i32  , Expand);
-  setOperationAction(ISD::CTPOP, MVT::i32  , Expand);
   setOperationAction(ISD::CTTZ , MVT::i32  , Expand);
   setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i32, Expand);
   setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i32, Expand);
   setOperationAction(ISD::BSWAP, MVT::i64  , Expand);
-  setOperationAction(ISD::CTPOP, MVT::i64  , Expand);
   setOperationAction(ISD::CTTZ , MVT::i64  , Expand);
   setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i64, Expand);
   setOperationAction(ISD::CTLZ_ZERO_UNDEF, MVT::i64, Expand);
 
+  if (Subtarget->hasPOPCNTD()) {
+    setOperationAction(ISD::CTPOP, MVT::i32  , Legal);
+    setOperationAction(ISD::CTPOP, MVT::i64  , Legal);
+  } else {
+    setOperationAction(ISD::CTPOP, MVT::i32  , Expand);
+    setOperationAction(ISD::CTPOP, MVT::i64  , Expand);
+  }
+
   // PowerPC does not have ROTR
   setOperationAction(ISD::ROTR, MVT::i32   , Expand);
   setOperationAction(ISD::ROTR, MVT::i64   , Expand);
@@ -206,6 +245,14 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
   setOperationAction(ISD::EXCEPTIONADDR, MVT::i32, Expand);
   setOperationAction(ISD::EHSELECTION,   MVT::i32, Expand);
 
+  // NOTE: EH_SJLJ_SETJMP/_LONGJMP supported here is NOT intended to support
+  // SjLj exception handling but a light-weight setjmp/longjmp replacement to
+  // support continuation, user-level threading, and etc.. As a result, no
+  // other SjLj exception interfaces are implemented and please don't build
+  // your own exception handling based on them.
+  // LLVM/Clang supports zero-cost DWARF exception handling.
+  setOperationAction(ISD::EH_SJLJ_SETJMP, MVT::i32, Custom);
+  setOperationAction(ISD::EH_SJLJ_LONGJMP, MVT::Other, Custom);
 
   // We want to legalize GlobalAddress and ConstantPool nodes into the
   // appropriate instructions to materialize the address.
@@ -285,15 +332,28 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
     // We cannot do this with Promote because i64 is not a legal type.
     setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
 
-    // FIXME: disable this lowered code.  This generates 64-bit register values,
-    // and we don't model the fact that the top part is clobbered by calls.  We
-    // need to flag these together so that the value isn't live across a call.
-    //setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
+    if (PPCSubTarget.hasLFIWAX() || Subtarget->isPPC64())
+      setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
   } else {
     // PowerPC does not have FP_TO_UINT on 32-bit implementations.
     setOperationAction(ISD::FP_TO_UINT, MVT::i32, Expand);
   }
 
+  // With the instructions enabled under FPCVT, we can do everything.
+  if (PPCSubTarget.hasFPCVT()) {
+    if (Subtarget->has64BitSupport()) {
+      setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
+      setOperationAction(ISD::FP_TO_UINT, MVT::i64, Custom);
+      setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
+      setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
+    }
+
+    setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
+    setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
+    setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
+    setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
+  }
+
   if (Subtarget->use64BitRegs()) {
     // 64-bit PowerPC implementations can support i64 types directly
     addRegisterClass(MVT::i64, &PPC::G8RCRegClass);
@@ -347,6 +407,21 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
       setOperationAction(ISD::UREM, VT, Expand);
       setOperationAction(ISD::FDIV, VT, Expand);
       setOperationAction(ISD::FNEG, VT, Expand);
+      setOperationAction(ISD::FSQRT, VT, Expand);
+      setOperationAction(ISD::FLOG, VT, Expand);
+      setOperationAction(ISD::FLOG10, VT, Expand);
+      setOperationAction(ISD::FLOG2, VT, Expand);
+      setOperationAction(ISD::FEXP, VT, Expand);
+      setOperationAction(ISD::FEXP2, VT, Expand);
+      setOperationAction(ISD::FSIN, VT, Expand);
+      setOperationAction(ISD::FCOS, VT, Expand);
+      setOperationAction(ISD::FABS, VT, Expand);
+      setOperationAction(ISD::FPOWI, VT, Expand);
+      setOperationAction(ISD::FFLOOR, VT, Expand);
+      setOperationAction(ISD::FCEIL,  VT, Expand);
+      setOperationAction(ISD::FTRUNC, VT, Expand);
+      setOperationAction(ISD::FRINT,  VT, Expand);
+      setOperationAction(ISD::FNEARBYINT, VT, Expand);
       setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Expand);
       setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Expand);
       setOperationAction(ISD::BUILD_VECTOR, VT, Expand);
@@ -361,6 +436,7 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
       setOperationAction(ISD::CTLZ_ZERO_UNDEF, VT, Expand);
       setOperationAction(ISD::CTTZ, VT, Expand);
       setOperationAction(ISD::CTTZ_ZERO_UNDEF, VT, Expand);
+      setOperationAction(ISD::VSELECT, VT, Expand);
       setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Expand);
 
       for (unsigned j = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
@@ -373,12 +449,6 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
       setLoadExtAction(ISD::EXTLOAD, VT, Expand);
     }
 
-    for (unsigned i = (unsigned)MVT::FIRST_FP_VECTOR_VALUETYPE;
-         i <= (unsigned)MVT::LAST_FP_VECTOR_VALUETYPE; ++i) {
-      MVT::SimpleValueType VT = (MVT::SimpleValueType)i;
-      setOperationAction(ISD::FSQRT, VT, Expand);
-    }
-
     // We can custom expand all VECTOR_SHUFFLEs to VPERM, others we can handle
     // with merges, splats, etc.
     setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v16i8, Custom);
@@ -393,6 +463,10 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
     setOperationAction(ISD::FP_TO_UINT, MVT::v4i32, Legal);
     setOperationAction(ISD::SINT_TO_FP, MVT::v4i32, Legal);
     setOperationAction(ISD::UINT_TO_FP, MVT::v4i32, Legal);
+    setOperationAction(ISD::FFLOOR, MVT::v4f32, Legal);
+    setOperationAction(ISD::FCEIL, MVT::v4f32, Legal);
+    setOperationAction(ISD::FTRUNC, MVT::v4f32, Legal);
+    setOperationAction(ISD::FNEARBYINT, MVT::v4f32, Legal);
 
     addRegisterClass(MVT::v4f32, &PPC::VRRCRegClass);
     addRegisterClass(MVT::v4i32, &PPC::VRRCRegClass);
@@ -401,6 +475,12 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
 
     setOperationAction(ISD::MUL, MVT::v4f32, Legal);
     setOperationAction(ISD::FMA, MVT::v4f32, Legal);
+
+    if (TM.Options.UnsafeFPMath) {
+      setOperationAction(ISD::FDIV, MVT::v4f32, Legal);
+      setOperationAction(ISD::FSQRT, MVT::v4f32, Legal);
+    }
+
     setOperationAction(ISD::MUL, MVT::v4i32, Custom);
     setOperationAction(ISD::MUL, MVT::v8i16, Custom);
     setOperationAction(ISD::MUL, MVT::v16i8, Custom);
@@ -429,6 +509,8 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
 
   setOperationAction(ISD::ATOMIC_LOAD,  MVT::i32, Expand);
   setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Expand);
+  setOperationAction(ISD::ATOMIC_LOAD,  MVT::i64, Expand);
+  setOperationAction(ISD::ATOMIC_STORE, MVT::i64, Expand);
 
   setBooleanContents(ZeroOrOneBooleanContent);
   setBooleanVectorContents(ZeroOrOneBooleanContent); // FIXME: Is this correct?
@@ -449,6 +531,12 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
   setTargetDAGCombine(ISD::BR_CC);
   setTargetDAGCombine(ISD::BSWAP);
 
+  // Use reciprocal estimates.
+  if (TM.Options.UnsafeFPMath) {
+    setTargetDAGCombine(ISD::FDIV);
+    setTargetDAGCombine(ISD::FSQRT);
+  }
+
   // Darwin long double math library functions have $LDBL128 appended.
   if (Subtarget->isDarwin()) {
     setLibcallName(RTLIB::COS_PPCF128, "cosl$LDBL128");
@@ -482,15 +570,14 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
   // friends. Gcc uses same threshold of 128 bytes (= 32 word stores).
   if (Subtarget->getDarwinDirective() == PPC::DIR_E500mc ||
       Subtarget->getDarwinDirective() == PPC::DIR_E5500) {
-    maxStoresPerMemset = 32;
-    maxStoresPerMemsetOptSize = 16;
-    maxStoresPerMemcpy = 32;
-    maxStoresPerMemcpyOptSize = 8;
-    maxStoresPerMemmove = 32;
-    maxStoresPerMemmoveOptSize = 8;
+    MaxStoresPerMemset = 32;
+    MaxStoresPerMemsetOptSize = 16;
+    MaxStoresPerMemcpy = 32;
+    MaxStoresPerMemcpyOptSize = 8;
+    MaxStoresPerMemmove = 32;
+    MaxStoresPerMemmoveOptSize = 8;
 
     setPrefFunctionAlignment(4);
-    benefitFromCodePlacementOpt = true;
   }
 }
 
@@ -521,6 +608,8 @@ const char *PPCTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case PPCISD::FCFID:           return "PPCISD::FCFID";
   case PPCISD::FCTIDZ:          return "PPCISD::FCTIDZ";
   case PPCISD::FCTIWZ:          return "PPCISD::FCTIWZ";
+  case PPCISD::FRE:             return "PPCISD::FRE";
+  case PPCISD::FRSQRTE:         return "PPCISD::FRSQRTE";
   case PPCISD::STFIWX:          return "PPCISD::STFIWX";
   case PPCISD::VMADDFP:         return "PPCISD::VMADDFP";
   case PPCISD::VNMSUBFP:        return "PPCISD::VNMSUBFP";
@@ -536,16 +625,13 @@ const char *PPCTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case PPCISD::SRL:             return "PPCISD::SRL";
   case PPCISD::SRA:             return "PPCISD::SRA";
   case PPCISD::SHL:             return "PPCISD::SHL";
-  case PPCISD::EXTSW_32:        return "PPCISD::EXTSW_32";
-  case PPCISD::STD_32:          return "PPCISD::STD_32";
-  case PPCISD::CALL_SVR4:       return "PPCISD::CALL_SVR4";
-  case PPCISD::CALL_NOP_SVR4:   return "PPCISD::CALL_NOP_SVR4";
-  case PPCISD::CALL_Darwin:     return "PPCISD::CALL_Darwin";
-  case PPCISD::NOP:             return "PPCISD::NOP";
+  case PPCISD::CALL:            return "PPCISD::CALL";
+  case PPCISD::CALL_NOP:        return "PPCISD::CALL_NOP";
   case PPCISD::MTCTR:           return "PPCISD::MTCTR";
-  case PPCISD::BCTRL_Darwin:    return "PPCISD::BCTRL_Darwin";
-  case PPCISD::BCTRL_SVR4:      return "PPCISD::BCTRL_SVR4";
+  case PPCISD::BCTRL:           return "PPCISD::BCTRL";
   case PPCISD::RET_FLAG:        return "PPCISD::RET_FLAG";
+  case PPCISD::EH_SJLJ_SETJMP:  return "PPCISD::EH_SJLJ_SETJMP";
+  case PPCISD::EH_SJLJ_LONGJMP: return "PPCISD::EH_SJLJ_LONGJMP";
   case PPCISD::MFCR:            return "PPCISD::MFCR";
   case PPCISD::VCMP:            return "PPCISD::VCMP";
   case PPCISD::VCMPo:           return "PPCISD::VCMPo";
@@ -555,13 +641,25 @@ const char *PPCTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case PPCISD::STCX:            return "PPCISD::STCX";
   case PPCISD::COND_BRANCH:     return "PPCISD::COND_BRANCH";
   case PPCISD::MFFS:            return "PPCISD::MFFS";
-  case PPCISD::MTFSB0:          return "PPCISD::MTFSB0";
-  case PPCISD::MTFSB1:          return "PPCISD::MTFSB1";
   case PPCISD::FADDRTZ:         return "PPCISD::FADDRTZ";
-  case PPCISD::MTFSF:           return "PPCISD::MTFSF";
   case PPCISD::TC_RETURN:       return "PPCISD::TC_RETURN";
   case PPCISD::CR6SET:          return "PPCISD::CR6SET";
   case PPCISD::CR6UNSET:        return "PPCISD::CR6UNSET";
+  case PPCISD::ADDIS_TOC_HA:    return "PPCISD::ADDIS_TOC_HA";
+  case PPCISD::LD_TOC_L:        return "PPCISD::LD_TOC_L";
+  case PPCISD::ADDI_TOC_L:      return "PPCISD::ADDI_TOC_L";
+  case PPCISD::ADDIS_GOT_TPREL_HA: return "PPCISD::ADDIS_GOT_TPREL_HA";
+  case PPCISD::LD_GOT_TPREL_L:  return "PPCISD::LD_GOT_TPREL_L";
+  case PPCISD::ADD_TLS:         return "PPCISD::ADD_TLS";
+  case PPCISD::ADDIS_TLSGD_HA:  return "PPCISD::ADDIS_TLSGD_HA";
+  case PPCISD::ADDI_TLSGD_L:    return "PPCISD::ADDI_TLSGD_L";
+  case PPCISD::GET_TLS_ADDR:    return "PPCISD::GET_TLS_ADDR";
+  case PPCISD::ADDIS_TLSLD_HA:  return "PPCISD::ADDIS_TLSLD_HA";
+  case PPCISD::ADDI_TLSLD_L:    return "PPCISD::ADDI_TLSLD_L";
+  case PPCISD::GET_TLSLD_ADDR:  return "PPCISD::GET_TLSLD_ADDR";
+  case PPCISD::ADDIS_DTPREL_HA: return "PPCISD::ADDIS_DTPREL_HA";
+  case PPCISD::ADDI_DTPREL_L:   return "PPCISD::ADDI_DTPREL_L";
+  case PPCISD::VADD_SPLAT:      return "PPCISD::VADD_SPLAT";
   }
 }
 
@@ -995,7 +1093,7 @@ bool PPCTargetLowering::SelectAddressRegImm(SDValue N, SDValue &Disp,
     short Imm;
     if (isIntS16Immediate(CN, Imm)) {
       Disp = DAG.getTargetConstant(Imm, CN->getValueType(0));
-      Base = DAG.getRegister(PPCSubTarget.isPPC64() ? PPC::X0 : PPC::R0,
+      Base = DAG.getRegister(PPCSubTarget.isPPC64() ? PPC::ZERO8 : PPC::ZERO,
                              CN->getValueType(0));
       return true;
     }
@@ -1044,7 +1142,7 @@ bool PPCTargetLowering::SelectAddressRegRegOnly(SDValue N, SDValue &Base,
   }
 
   // Otherwise, do it the hard way, using R0 as the base register.
-  Base = DAG.getRegister(PPCSubTarget.isPPC64() ? PPC::X0 : PPC::R0,
+  Base = DAG.getRegister(PPCSubTarget.isPPC64() ? PPC::ZERO8 : PPC::ZERO,
                          N.getValueType());
   Index = N;
   return true;
@@ -1107,7 +1205,7 @@ bool PPCTargetLowering::SelectAddressRegImmShift(SDValue N, SDValue &Disp,
       short Imm;
       if (isIntS16Immediate(CN, Imm)) {
         Disp = DAG.getTargetConstant((unsigned short)Imm >> 2, getPointerTy());
-        Base = DAG.getRegister(PPCSubTarget.isPPC64() ? PPC::X0 : PPC::R0,
+        Base = DAG.getRegister(PPCSubTarget.isPPC64() ? PPC::ZERO8 : PPC::ZERO,
                                CN->getValueType(0));
         return true;
       }
@@ -1145,15 +1243,19 @@ bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
                                                   SelectionDAG &DAG) const {
   if (DisablePPCPreinc) return false;
 
+  bool isLoad = true;
   SDValue Ptr;
   EVT VT;
+  unsigned Alignment;
   if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
     Ptr = LD->getBasePtr();
     VT = LD->getMemoryVT();
-
+    Alignment = LD->getAlignment();
   } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
     Ptr = ST->getBasePtr();
     VT  = ST->getMemoryVT();
+    Alignment = ST->getAlignment();
+    isLoad = false;
   } else
     return false;
 
@@ -1161,7 +1263,25 @@ bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
   if (VT.isVector())
     return false;
 
-  if (SelectAddressRegReg(Ptr, Offset, Base, DAG)) {
+  if (SelectAddressRegReg(Ptr, Base, Offset, DAG)) {
+
+    // Common code will reject creating a pre-inc form if the base pointer
+    // is a frame index, or if N is a store and the base pointer is either
+    // the same as or a predecessor of the value being stored.  Check for
+    // those situations here, and try with swapped Base/Offset instead.
+    bool Swap = false;
+
+    if (isa<FrameIndexSDNode>(Base) || isa<RegisterSDNode>(Base))
+      Swap = true;
+    else if (!isLoad) {
+      SDValue Val = cast<StoreSDNode>(N)->getValue();
+      if (Val == Base || Base.getNode()->isPredecessorOf(Val.getNode()))
+        Swap = true;
+    }
+
+    if (Swap)
+      std::swap(Base, Offset);
+
     AM = ISD::PRE_INC;
     return true;
   }
@@ -1172,6 +1292,10 @@ bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
     if (!SelectAddressRegImm(Ptr, Offset, Base, DAG))
       return false;
   } else {
+    // LDU/STU need an address with at least 4-byte alignment.
+    if (Alignment < 4)
+      return false;
+
     // reg + imm * 4.
     if (!SelectAddressRegImmShift(Ptr, Offset, Base, DAG))
       return false;
@@ -1308,19 +1432,81 @@ SDValue PPCTargetLowering::LowerGlobalTLSAddress(SDValue Op,
   EVT PtrVT = getPointerTy();
   bool is64bit = PPCSubTarget.isPPC64();
 
-  TLSModel::Model model = getTargetMachine().getTLSModel(GV);
+  TLSModel::Model Model = getTargetMachine().getTLSModel(GV);
+
+  if (Model == TLSModel::LocalExec) {
+    SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+                                               PPCII::MO_TPREL16_HA);
+    SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
+                                               PPCII::MO_TPREL16_LO);
+    SDValue TLSReg = DAG.getRegister(is64bit ? PPC::X13 : PPC::R2,
+                                     is64bit ? MVT::i64 : MVT::i32);
+    SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, TGAHi, TLSReg);
+    return DAG.getNode(PPCISD::Lo, dl, PtrVT, TGALo, Hi);
+  }
+
+  if (!is64bit)
+    llvm_unreachable("only local-exec is currently supported for ppc32");
+
+  if (Model == TLSModel::InitialExec) {
+    SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, 0);
+    SDValue GOTReg = DAG.getRegister(PPC::X2, MVT::i64);
+    SDValue TPOffsetHi = DAG.getNode(PPCISD::ADDIS_GOT_TPREL_HA, dl,
+                                     PtrVT, GOTReg, TGA);
+    SDValue TPOffset = DAG.getNode(PPCISD::LD_GOT_TPREL_L, dl,
+                                   PtrVT, TGA, TPOffsetHi);
+    return DAG.getNode(PPCISD::ADD_TLS, dl, PtrVT, TPOffset, TGA);
+  }
+
+  if (Model == TLSModel::GeneralDynamic) {
+    SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, 0);
+    SDValue GOTReg = DAG.getRegister(PPC::X2, MVT::i64);
+    SDValue GOTEntryHi = DAG.getNode(PPCISD::ADDIS_TLSGD_HA, dl, PtrVT,
+                                     GOTReg, TGA);
+    SDValue GOTEntry = DAG.getNode(PPCISD::ADDI_TLSGD_L, dl, PtrVT,
+                                   GOTEntryHi, TGA);
+
+    // We need a chain node, and don't have one handy.  The underlying
+    // call has no side effects, so using the function entry node
+    // suffices.
+    SDValue Chain = DAG.getEntryNode();
+    Chain = DAG.getCopyToReg(Chain, dl, PPC::X3, GOTEntry);
+    SDValue ParmReg = DAG.getRegister(PPC::X3, MVT::i64);
+    SDValue TLSAddr = DAG.getNode(PPCISD::GET_TLS_ADDR, dl,
+                                  PtrVT, ParmReg, TGA);
+    // The return value from GET_TLS_ADDR really is in X3 already, but
+    // some hacks are needed here to tie everything together.  The extra
+    // copies dissolve during subsequent transforms.
+    Chain = DAG.getCopyToReg(Chain, dl, PPC::X3, TLSAddr);
+    return DAG.getCopyFromReg(Chain, dl, PPC::X3, PtrVT);
+  }
 
-  SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
-                                             PPCII::MO_TPREL16_HA);
-  SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
-                                             PPCII::MO_TPREL16_LO);
+  if (Model == TLSModel::LocalDynamic) {
+    SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, 0);
+    SDValue GOTReg = DAG.getRegister(PPC::X2, MVT::i64);
+    SDValue GOTEntryHi = DAG.getNode(PPCISD::ADDIS_TLSLD_HA, dl, PtrVT,
+                                     GOTReg, TGA);
+    SDValue GOTEntry = DAG.getNode(PPCISD::ADDI_TLSLD_L, dl, PtrVT,
+                                   GOTEntryHi, TGA);
+
+    // We need a chain node, and don't have one handy.  The underlying
+    // call has no side effects, so using the function entry node
+    // suffices.
+    SDValue Chain = DAG.getEntryNode();
+    Chain = DAG.getCopyToReg(Chain, dl, PPC::X3, GOTEntry);
+    SDValue ParmReg = DAG.getRegister(PPC::X3, MVT::i64);
+    SDValue TLSAddr = DAG.getNode(PPCISD::GET_TLSLD_ADDR, dl,
+                                  PtrVT, ParmReg, TGA);
+    // The return value from GET_TLSLD_ADDR really is in X3 already, but
+    // some hacks are needed here to tie everything together.  The extra
+    // copies dissolve during subsequent transforms.
+    Chain = DAG.getCopyToReg(Chain, dl, PPC::X3, TLSAddr);
+    SDValue DtvOffsetHi = DAG.getNode(PPCISD::ADDIS_DTPREL_HA, dl, PtrVT,
+                                      Chain, ParmReg, TGA);
+    return DAG.getNode(PPCISD::ADDI_DTPREL_L, dl, PtrVT, DtvOffsetHi, TGA);
+  }
 
-  if (model != TLSModel::LocalExec)
-    llvm_unreachable("only local-exec TLS mode supported");
-  SDValue TLSReg = DAG.getRegister(is64bit ? PPC::X13 : PPC::R2,
-                                   is64bit ? MVT::i64 : MVT::i32);
-  SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, TGAHi, TLSReg);
-  return DAG.getNode(PPCISD::Lo, dl, PtrVT, TGALo, Hi);
+  llvm_unreachable("Unknown TLS model!");
 }
 
 SDValue PPCTargetLowering::LowerGlobalAddress(SDValue Op,
@@ -1654,18 +1840,18 @@ SDValue PPCTargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG,
 
 #include "PPCGenCallingConv.inc"
 
-static bool CC_PPC_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
-                                     CCValAssign::LocInfo &LocInfo,
-                                     ISD::ArgFlagsTy &ArgFlags,
-                                     CCState &State) {
+static bool CC_PPC32_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+                                       CCValAssign::LocInfo &LocInfo,
+                                       ISD::ArgFlagsTy &ArgFlags,
+                                       CCState &State) {
   return true;
 }
 
-static bool CC_PPC_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
-                                            MVT &LocVT,
-                                            CCValAssign::LocInfo &LocInfo,
-                                            ISD::ArgFlagsTy &ArgFlags,
-                                            CCState &State) {
+static bool CC_PPC32_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
+                                              MVT &LocVT,
+                                              CCValAssign::LocInfo &LocInfo,
+                                              ISD::ArgFlagsTy &ArgFlags,
+                                              CCState &State) {
   static const uint16_t ArgRegs[] = {
     PPC::R3, PPC::R4, PPC::R5, PPC::R6,
     PPC::R7, PPC::R8, PPC::R9, PPC::R10,
@@ -1688,11 +1874,11 @@ static bool CC_PPC_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
   return false;
 }
 
-static bool CC_PPC_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
-                                              MVT &LocVT,
-                                              CCValAssign::LocInfo &LocInfo,
-                                              ISD::ArgFlagsTy &ArgFlags,
-                                              CCState &State) {
+static bool CC_PPC32_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
+                                                MVT &LocVT,
+                                                CCValAssign::LocInfo &LocInfo,
+                                                ISD::ArgFlagsTy &ArgFlags,
+                                                CCState &State) {
   static const uint16_t ArgRegs[] = {
     PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
     PPC::F8
@@ -1815,7 +2001,7 @@ PPCTargetLowering::LowerFormalArguments_32SVR4(
   // Reserve space for the linkage area on the stack.
   CCInfo.AllocateStack(PPCFrameLowering::getLinkageSize(false, false), PtrByteSize);
 
-  CCInfo.AnalyzeFormalArguments(Ins, CC_PPC_SVR4);
+  CCInfo.AnalyzeFormalArguments(Ins, CC_PPC32_SVR4);
 
   for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
     CCValAssign &VA = ArgLocs[i];
@@ -1876,7 +2062,7 @@ PPCTargetLowering::LowerFormalArguments_32SVR4(
   // Reserve stack space for the allocations in CCInfo.
   CCByValInfo.AllocateStack(CCInfo.getNextStackOffset(), PtrByteSize);
 
-  CCByValInfo.AnalyzeFormalArguments(Ins, CC_PPC_SVR4_ByVal);
+  CCByValInfo.AnalyzeFormalArguments(Ins, CC_PPC32_SVR4_ByVal);
 
   // Area that is at least reserved in the caller of this function.
   unsigned MinReservedArea = CCByValInfo.getNextStackOffset();
@@ -2068,13 +2254,16 @@ PPCTargetLowering::LowerFormalArguments_64SVR4(
   SmallVector<SDValue, 8> MemOps;
   unsigned nAltivecParamsAtEnd = 0;
   Function::const_arg_iterator FuncArg = MF.getFunction()->arg_begin();
-  for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo, ++FuncArg) {
+  unsigned CurArgIdx = 0;
+  for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo) {
     SDValue ArgVal;
     bool needsLoad = false;
     EVT ObjectVT = Ins[ArgNo].VT;
     unsigned ObjSize = ObjectVT.getSizeInBits()/8;
     unsigned ArgSize = ObjSize;
     ISD::ArgFlagsTy Flags = Ins[ArgNo].Flags;
+    std::advance(FuncArg, Ins[ArgNo].OrigArgIndex - CurArgIdx);
+    CurArgIdx = Ins[ArgNo].OrigArgIndex;
 
     unsigned CurArgOffset = ArgOffset;
 
@@ -2409,6 +2598,9 @@ PPCTargetLowering::LowerFormalArguments_Darwin(
 
   SmallVector<SDValue, 8> MemOps;
   unsigned nAltivecParamsAtEnd = 0;
+  // FIXME: FuncArg and Ins[ArgNo] must reference the same argument.
+  // When passing anonymous aggregates, this is currently not true.
+  // See LowerFormalArguments_64SVR4 for a fix.
   Function::const_arg_iterator FuncArg = MF.getFunction()->arg_begin();
   for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo, ++FuncArg) {
     SDValue ArgVal;
@@ -2995,7 +3187,7 @@ unsigned PrepareCall(SelectionDAG &DAG, SDValue &Callee, SDValue &InFlag,
   NodeTys.push_back(MVT::Other);   // Returns a chain
   NodeTys.push_back(MVT::Glue);    // Returns a flag for retval copy to use.
 
-  unsigned CallOpc = isSVR4ABI ? PPCISD::CALL_SVR4 : PPCISD::CALL_Darwin;
+  unsigned CallOpc = PPCISD::CALL;
 
   bool needIndirectCall = true;
   if (SDNode *Dest = isBLACompatibleAddress(Callee, DAG)) {
@@ -3128,8 +3320,11 @@ unsigned PrepareCall(SelectionDAG &DAG, SDValue &Callee, SDValue &InFlag,
     NodeTys.push_back(MVT::Other);
     NodeTys.push_back(MVT::Glue);
     Ops.push_back(Chain);
-    CallOpc = isSVR4ABI ? PPCISD::BCTRL_SVR4 : PPCISD::BCTRL_Darwin;
+    CallOpc = PPCISD::BCTRL;
     Callee.setNode(0);
+    // Add use of X11 (holding environment pointer)
+    if (isSVR4ABI && isPPC64)
+      Ops.push_back(DAG.getRegister(PPC::X11, PtrVT));
     // Add CTR register as callee so a bctr can be emitted later.
     if (isTailCall)
       Ops.push_back(DAG.getRegister(isPPC64 ? PPC::CTR8 : PPC::CTR, PtrVT));
@@ -3231,7 +3426,7 @@ PPCTargetLowering::FinishCall(CallingConv::ID CallConv, DebugLoc dl,
 
   // When performing tail call optimization the callee pops its arguments off
   // the stack. Account for this here so these bytes can be pushed back on in
-  // PPCRegisterInfo::eliminateCallFramePseudoInstr.
+  // PPCFrameLowering::eliminateCallFramePseudoInstr.
   int BytesCalleePops =
     (CallConv == CallingConv::Fast &&
      getTargetMachine().Options.GuaranteedTailCallOpt) ? NumBytes : 0;
@@ -3247,17 +3442,6 @@ PPCTargetLowering::FinishCall(CallingConv::ID CallConv, DebugLoc dl,
 
   // Emit tail call.
   if (isTailCall) {
-    // If this is the first return lowered for this function, add the regs
-    // to the liveout set for the function.
-    if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
-      SmallVector<CCValAssign, 16> RVLocs;
-      CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
-                     getTargetMachine(), RVLocs, *DAG.getContext());
-      CCInfo.AnalyzeCallResult(Ins, RetCC_PPC);
-      for (unsigned i = 0; i != RVLocs.size(); ++i)
-        DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
-    }
-
     assert(((Callee.getOpcode() == ISD::Register &&
              cast<RegisterSDNode>(Callee)->getReg() == PPC::CTR) ||
             Callee.getOpcode() == ISD::TargetExternalSymbol ||
@@ -3279,7 +3463,7 @@ PPCTargetLowering::FinishCall(CallingConv::ID CallConv, DebugLoc dl,
 
   bool needsTOCRestore = false;
   if (!isTailCall && PPCSubTarget.isSVR4ABI()&& PPCSubTarget.isPPC64()) {
-    if (CallOpc == PPCISD::BCTRL_SVR4) {
+    if (CallOpc == PPCISD::BCTRL) {
       // This is a call through a function pointer.
       // Restore the caller TOC from the save area into R2.
       // See PrepareCall() for more information about calls through function
@@ -3290,9 +3474,9 @@ PPCTargetLowering::FinishCall(CallingConv::ID CallConv, DebugLoc dl,
       // from allocating it), resulting in an additional register being
       // allocated and an unnecessary move instruction being generated.
       needsTOCRestore = true;
-    } else if ((CallOpc == PPCISD::CALL_SVR4) && !isLocalCall(Callee)) {
+    } else if ((CallOpc == PPCISD::CALL) && !isLocalCall(Callee)) {
       // Otherwise insert NOP for non-local calls.
-      CallOpc = PPCISD::CALL_NOP_SVR4;
+      CallOpc = PPCISD::CALL_NOP;
     }
   }
 
@@ -3401,11 +3585,11 @@ PPCTargetLowering::LowerCall_32SVR4(SDValue Chain, SDValue Callee,
       bool Result;
 
       if (Outs[i].IsFixed) {
-        Result = CC_PPC_SVR4(i, ArgVT, ArgVT, CCValAssign::Full, ArgFlags,
-                             CCInfo);
+        Result = CC_PPC32_SVR4(i, ArgVT, ArgVT, CCValAssign::Full, ArgFlags,
+                               CCInfo);
       } else {
-        Result = CC_PPC_SVR4_VarArg(i, ArgVT, ArgVT, CCValAssign::Full,
-                                    ArgFlags, CCInfo);
+        Result = CC_PPC32_SVR4_VarArg(i, ArgVT, ArgVT, CCValAssign::Full,
+                                      ArgFlags, CCInfo);
       }
 
       if (Result) {
@@ -3418,7 +3602,7 @@ PPCTargetLowering::LowerCall_32SVR4(SDValue Chain, SDValue Callee,
     }
   } else {
     // All arguments are treated the same.
-    CCInfo.AnalyzeCallOperands(Outs, CC_PPC_SVR4);
+    CCInfo.AnalyzeCallOperands(Outs, CC_PPC32_SVR4);
   }
 
   // Assign locations to all of the outgoing aggregate by value arguments.
@@ -3429,7 +3613,7 @@ PPCTargetLowering::LowerCall_32SVR4(SDValue Chain, SDValue Callee,
   // Reserve stack space for the allocations in CCInfo.
   CCByValInfo.AllocateStack(CCInfo.getNextStackOffset(), PtrByteSize);
 
-  CCByValInfo.AnalyzeCallOperands(Outs, CC_PPC_SVR4_ByVal);
+  CCByValInfo.AnalyzeCallOperands(Outs, CC_PPC32_SVR4_ByVal);
 
   // Size of the linkage area, parameter list area and the part of the local
   // space variable where copies of aggregates which are passed by value are
@@ -4323,14 +4507,8 @@ PPCTargetLowering::LowerReturn(SDValue Chain,
                  getTargetMachine(), RVLocs, *DAG.getContext());
   CCInfo.AnalyzeReturn(Outs, RetCC_PPC);
 
-  // If this is the first return lowered for this function, add the regs to the
-  // liveout set for the function.
-  if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
-    for (unsigned i = 0; i != RVLocs.size(); ++i)
-      DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
-  }
-
   SDValue Flag;
+  SmallVector<SDValue, 4> RetOps(1, Chain);
 
   // Copy the result values into the output registers.
   for (unsigned i = 0; i != RVLocs.size(); ++i) {
@@ -4355,12 +4533,17 @@ PPCTargetLowering::LowerReturn(SDValue Chain,
 
     Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), Arg, Flag);
     Flag = Chain.getValue(1);
+    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
   }
 
+  RetOps[0] = Chain;  // Update chain.
+
+  // Add the flag if we have it.
   if (Flag.getNode())
-    return DAG.getNode(PPCISD::RET_FLAG, dl, MVT::Other, Chain, Flag);
-  else
-    return DAG.getNode(PPCISD::RET_FLAG, dl, MVT::Other, Chain);
+    RetOps.push_back(Flag);
+
+  return DAG.getNode(PPCISD::RET_FLAG, dl, MVT::Other,
+                     &RetOps[0], RetOps.size());
 }
 
 SDValue PPCTargetLowering::LowerSTACKRESTORE(SDValue Op, SelectionDAG &DAG,
@@ -4466,6 +4649,21 @@ SDValue PPCTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op,
   return DAG.getNode(PPCISD::DYNALLOC, dl, VTs, Ops, 3);
 }
 
+SDValue PPCTargetLowering::lowerEH_SJLJ_SETJMP(SDValue Op,
+                                               SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  return DAG.getNode(PPCISD::EH_SJLJ_SETJMP, DL,
+                     DAG.getVTList(MVT::i32, MVT::Other),
+                     Op.getOperand(0), Op.getOperand(1));
+}
+
+SDValue PPCTargetLowering::lowerEH_SJLJ_LONGJMP(SDValue Op,
+                                                SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  return DAG.getNode(PPCISD::EH_SJLJ_LONGJMP, DL, MVT::Other,
+                     Op.getOperand(0), Op.getOperand(1));
+}
+
 /// LowerSELECT_CC - Lower floating point select_cc's into fsel instruction when
 /// possible.
 SDValue PPCTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const {
@@ -4553,37 +4751,72 @@ SDValue PPCTargetLowering::LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG,
   default: llvm_unreachable("Unhandled FP_TO_INT type in custom expander!");
   case MVT::i32:
     Tmp = DAG.getNode(Op.getOpcode()==ISD::FP_TO_SINT ? PPCISD::FCTIWZ :
-                                                         PPCISD::FCTIDZ,
+                        (PPCSubTarget.hasFPCVT() ? PPCISD::FCTIWUZ :
+                                                   PPCISD::FCTIDZ),
                       dl, MVT::f64, Src);
     break;
   case MVT::i64:
-    Tmp = DAG.getNode(PPCISD::FCTIDZ, dl, MVT::f64, Src);
+    assert((Op.getOpcode() == ISD::FP_TO_SINT || PPCSubTarget.hasFPCVT()) &&
+           "i64 FP_TO_UINT is supported only with FPCVT");
+    Tmp = DAG.getNode(Op.getOpcode()==ISD::FP_TO_SINT ? PPCISD::FCTIDZ :
+                                                        PPCISD::FCTIDUZ,
+                      dl, MVT::f64, Src);
     break;
   }
 
   // Convert the FP value to an int value through memory.
-  SDValue FIPtr = DAG.CreateStackTemporary(MVT::f64);
+  bool i32Stack = Op.getValueType() == MVT::i32 && PPCSubTarget.hasSTFIWX() &&
+    (Op.getOpcode() == ISD::FP_TO_SINT || PPCSubTarget.hasFPCVT());
+  SDValue FIPtr = DAG.CreateStackTemporary(i32Stack ? MVT::i32 : MVT::f64);
+  int FI = cast<FrameIndexSDNode>(FIPtr)->getIndex();
+  MachinePointerInfo MPI = MachinePointerInfo::getFixedStack(FI);
 
   // Emit a store to the stack slot.
-  SDValue Chain = DAG.getStore(DAG.getEntryNode(), dl, Tmp, FIPtr,
-                               MachinePointerInfo(), false, false, 0);
+  SDValue Chain;
+  if (i32Stack) {
+    MachineFunction &MF = DAG.getMachineFunction();
+    MachineMemOperand *MMO =
+      MF.getMachineMemOperand(MPI, MachineMemOperand::MOStore, 4, 4);
+    SDValue Ops[] = { DAG.getEntryNode(), Tmp, FIPtr };
+    Chain = DAG.getMemIntrinsicNode(PPCISD::STFIWX, dl,
+              DAG.getVTList(MVT::Other), Ops, array_lengthof(Ops),
+              MVT::i32, MMO);
+  } else
+    Chain = DAG.getStore(DAG.getEntryNode(), dl, Tmp, FIPtr,
+                         MPI, false, false, 0);
 
   // Result is a load from the stack slot.  If loading 4 bytes, make sure to
   // add in a bias.
-  if (Op.getValueType() == MVT::i32)
+  if (Op.getValueType() == MVT::i32 && !i32Stack) {
     FIPtr = DAG.getNode(ISD::ADD, dl, FIPtr.getValueType(), FIPtr,
                         DAG.getConstant(4, FIPtr.getValueType()));
-  return DAG.getLoad(Op.getValueType(), dl, Chain, FIPtr, MachinePointerInfo(),
+    MPI = MachinePointerInfo();
+  }
+
+  return DAG.getLoad(Op.getValueType(), dl, Chain, FIPtr, MPI,
                      false, false, false, 0);
 }
 
-SDValue PPCTargetLowering::LowerSINT_TO_FP(SDValue Op,
+SDValue PPCTargetLowering::LowerINT_TO_FP(SDValue Op,
                                            SelectionDAG &DAG) const {
   DebugLoc dl = Op.getDebugLoc();
   // Don't handle ppc_fp128 here; let it be lowered to a libcall.
   if (Op.getValueType() != MVT::f32 && Op.getValueType() != MVT::f64)
     return SDValue();
 
+  assert((Op.getOpcode() == ISD::SINT_TO_FP || PPCSubTarget.hasFPCVT()) &&
+         "UINT_TO_FP is supported only with FPCVT");
+
+  // If we have FCFIDS, then use it when converting to single-precision.
+  // Otherwise, convert to double-precision and then round.
+  unsigned FCFOp = (PPCSubTarget.hasFPCVT() && Op.getValueType() == MVT::f32) ?
+                   (Op.getOpcode() == ISD::UINT_TO_FP ?
+                    PPCISD::FCFIDUS : PPCISD::FCFIDS) :
+                   (Op.getOpcode() == ISD::UINT_TO_FP ?
+                    PPCISD::FCFIDU : PPCISD::FCFID);
+  MVT      FCFTy = (PPCSubTarget.hasFPCVT() && Op.getValueType() == MVT::f32) ?
+                   MVT::f32 : MVT::f64;
+
   if (Op.getOperand(0).getValueType() == MVT::i64) {
     SDValue SINT = Op.getOperand(0);
     // When converting to single-precision, we actually need to convert
@@ -4597,6 +4830,7 @@ SDValue PPCTargetLowering::LowerSINT_TO_FP(SDValue Op,
     // However, if -enable-unsafe-fp-math is in effect, accept double
     // rounding to avoid the extra overhead.
     if (Op.getValueType() == MVT::f32 &&
+        !PPCSubTarget.hasFPCVT() &&
         !DAG.getTarget().Options.UnsafeFPMath) {
 
       // Twiddle input to make sure the low 11 bits are zero.  (If this
@@ -4630,44 +4864,69 @@ SDValue PPCTargetLowering::LowerSINT_TO_FP(SDValue Op,
 
       SINT = DAG.getNode(ISD::SELECT, dl, MVT::i64, Cond, Round, SINT);
     }
+
     SDValue Bits = DAG.getNode(ISD::BITCAST, dl, MVT::f64, SINT);
-    SDValue FP = DAG.getNode(PPCISD::FCFID, dl, MVT::f64, Bits);
-    if (Op.getValueType() == MVT::f32)
+    SDValue FP = DAG.getNode(FCFOp, dl, FCFTy, Bits);
+
+    if (Op.getValueType() == MVT::f32 && !PPCSubTarget.hasFPCVT())
       FP = DAG.getNode(ISD::FP_ROUND, dl,
                        MVT::f32, FP, DAG.getIntPtrConstant(0));
     return FP;
   }
 
   assert(Op.getOperand(0).getValueType() == MVT::i32 &&
-         "Unhandled SINT_TO_FP type in custom expander!");
+         "Unhandled INT_TO_FP type in custom expander!");
   // Since we only generate this in 64-bit mode, we can take advantage of
   // 64-bit registers.  In particular, sign extend the input value into the
   // 64-bit register with extsw, store the WHOLE 64-bit value into the stack
   // then lfd it and fcfid it.
   MachineFunction &MF = DAG.getMachineFunction();
   MachineFrameInfo *FrameInfo = MF.getFrameInfo();
-  int FrameIdx = FrameInfo->CreateStackObject(8, 8, false);
   EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
-  SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
 
-  SDValue Ext64 = DAG.getNode(PPCISD::EXTSW_32, dl, MVT::i32,
+  SDValue Ld;
+  if (PPCSubTarget.hasLFIWAX() || PPCSubTarget.hasFPCVT()) {
+    int FrameIdx = FrameInfo->CreateStackObject(4, 4, false);
+    SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
+
+    SDValue Store = DAG.getStore(DAG.getEntryNode(), dl, Op.getOperand(0), FIdx,
+                                 MachinePointerInfo::getFixedStack(FrameIdx),
+                                 false, false, 0);
+
+    assert(cast<StoreSDNode>(Store)->getMemoryVT() == MVT::i32 &&
+           "Expected an i32 store");
+    MachineMemOperand *MMO =
+      MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FrameIdx),
+                              MachineMemOperand::MOLoad, 4, 4);
+    SDValue Ops[] = { Store, FIdx };
+    Ld = DAG.getMemIntrinsicNode(Op.getOpcode() == ISD::UINT_TO_FP ?
+                                   PPCISD::LFIWZX : PPCISD::LFIWAX,
+                                 dl, DAG.getVTList(MVT::f64, MVT::Other),
+                                 Ops, 2, MVT::i32, MMO);
+  } else {
+    assert(PPCSubTarget.isPPC64() &&
+           "i32->FP without LFIWAX supported only on PPC64");
+
+    int FrameIdx = FrameInfo->CreateStackObject(8, 8, false);
+    SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
+
+    SDValue Ext64 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i64,
                                 Op.getOperand(0));
 
-  // STD the extended value into the stack slot.
-  MachineMemOperand *MMO =
-    MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FrameIdx),
-                            MachineMemOperand::MOStore, 8, 8);
-  SDValue Ops[] = { DAG.getEntryNode(), Ext64, FIdx };
-  SDValue Store =
-    DAG.getMemIntrinsicNode(PPCISD::STD_32, dl, DAG.getVTList(MVT::Other),
-                            Ops, 4, MVT::i64, MMO);
-  // Load the value as a double.
-  SDValue Ld = DAG.getLoad(MVT::f64, dl, Store, FIdx, MachinePointerInfo(),
-                           false, false, false, 0);
+    // STD the extended value into the stack slot.
+    SDValue Store = DAG.getStore(DAG.getEntryNode(), dl, Ext64, FIdx,
+                                 MachinePointerInfo::getFixedStack(FrameIdx),
+                                 false, false, 0);
+
+    // Load the value as a double.
+    Ld = DAG.getLoad(MVT::f64, dl, Store, FIdx,
+                     MachinePointerInfo::getFixedStack(FrameIdx),
+                     false, false, false, 0);
+  }
 
   // FCFID it and return it.
-  SDValue FP = DAG.getNode(PPCISD::FCFID, dl, MVT::f64, Ld);
-  if (Op.getValueType() == MVT::f32)
+  SDValue FP = DAG.getNode(FCFOp, dl, FCFTy, Ld);
+  if (Op.getValueType() == MVT::f32 && !PPCSubTarget.hasFPCVT())
     FP = DAG.getNode(ISD::FP_ROUND, dl, MVT::f32, FP, DAG.getIntPtrConstant(0));
   return FP;
 }
@@ -4697,12 +4956,13 @@ SDValue PPCTargetLowering::LowerFLT_ROUNDS_(SDValue Op,
   MachineFunction &MF = DAG.getMachineFunction();
   EVT VT = Op.getValueType();
   EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
-  std::vector<EVT> NodeTys;
   SDValue MFFSreg, InFlag;
 
   // Save FP Control Word to register
-  NodeTys.push_back(MVT::f64);    // return register
-  NodeTys.push_back(MVT::Glue);   // unused in this context
+  EVT NodeTys[] = {
+    MVT::f64,    // return register
+    MVT::Glue    // unused in this context
+  };
   SDValue Chain = DAG.getNode(PPCISD::MFFS, dl, NodeTys, &InFlag, 0);
 
   // Save FP register to stack slot
@@ -4936,11 +5196,21 @@ SDValue PPCTargetLowering::LowerBUILD_VECTOR(SDValue Op,
   // Two instruction sequences.
 
   // If this value is in the range [-32,30] and is even, use:
-  //    tmp = VSPLTI[bhw], result = add tmp, tmp
-  if (SextVal >= -32 && SextVal <= 30 && (SextVal & 1) == 0) {
-    SDValue Res = BuildSplatI(SextVal >> 1, SplatSize, MVT::Other, DAG, dl);
-    Res = DAG.getNode(ISD::ADD, dl, Res.getValueType(), Res, Res);
-    return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Res);
+  //     VSPLTI[bhw](val/2) + VSPLTI[bhw](val/2)
+  // If this value is in the range [17,31] and is odd, use:
+  //     VSPLTI[bhw](val-16) - VSPLTI[bhw](-16)
+  // If this value is in the range [-31,-17] and is odd, use:
+  //     VSPLTI[bhw](val+16) + VSPLTI[bhw](-16)
+  // Note the last two are three-instruction sequences.
+  if (SextVal >= -32 && SextVal <= 31) {
+    // To avoid having these optimizations undone by constant folding,
+    // we convert to a pseudo that will be expanded later into one of
+    // the above forms.
+    SDValue Elt = DAG.getConstant(SextVal, MVT::i32);
+    EVT VT = Op.getValueType();
+    int Size = VT == MVT::v16i8 ? 1 : (VT == MVT::v8i16 ? 2 : 4);
+    SDValue EltSize = DAG.getConstant(Size, MVT::i32);
+    return DAG.getNode(PPCISD::VADD_SPLAT, dl, VT, Elt, EltSize);
   }
 
   // If this is 0x8000_0000 x 4, turn into vspltisw + vslw.  If it is
@@ -5036,23 +5306,6 @@ SDValue PPCTargetLowering::LowerBUILD_VECTOR(SDValue Op,
     }
   }
 
-  // Three instruction sequences.
-
-  // Odd, in range [17,31]:  (vsplti C)-(vsplti -16).
-  if (SextVal >= 0 && SextVal <= 31) {
-    SDValue LHS = BuildSplatI(SextVal-16, SplatSize, MVT::Other, DAG, dl);
-    SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG, dl);
-    LHS = DAG.getNode(ISD::SUB, dl, LHS.getValueType(), LHS, RHS);
-    return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), LHS);
-  }
-  // Odd, in range [-31,-17]:  (vsplti C)+(vsplti -16).
-  if (SextVal >= -31 && SextVal <= 0) {
-    SDValue LHS = BuildSplatI(SextVal+16, SplatSize, MVT::Other, DAG, dl);
-    SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG, dl);
-    LHS = DAG.getNode(ISD::ADD, dl, LHS.getValueType(), LHS, RHS);
-    return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), LHS);
-  }
-
   return SDValue();
 }
 
@@ -5326,9 +5579,7 @@ SDValue PPCTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
     Op.getOperand(3),  // RHS
     DAG.getConstant(CompareOpc, MVT::i32)
   };
-  std::vector<EVT> VTs;
-  VTs.push_back(Op.getOperand(2).getValueType());
-  VTs.push_back(MVT::Glue);
+  EVT VTs[] = { Op.getOperand(2).getValueType(), MVT::Glue };
   SDValue CompNode = DAG.getNode(PPCISD::VCMPo, dl, VTs, Ops, 3);
 
   // Now that we have the comparison, emit a copy from the CR to a GPR.
@@ -5470,11 +5721,15 @@ SDValue PPCTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::DYNAMIC_STACKALLOC:
     return LowerDYNAMIC_STACKALLOC(Op, DAG, PPCSubTarget);
 
+  case ISD::EH_SJLJ_SETJMP:     return lowerEH_SJLJ_SETJMP(Op, DAG);
+  case ISD::EH_SJLJ_LONGJMP:    return lowerEH_SJLJ_LONGJMP(Op, DAG);
+
   case ISD::SELECT_CC:          return LowerSELECT_CC(Op, DAG);
   case ISD::FP_TO_UINT:
   case ISD::FP_TO_SINT:         return LowerFP_TO_INT(Op, DAG,
                                                        Op.getDebugLoc());
-  case ISD::SINT_TO_FP:         return LowerSINT_TO_FP(Op, DAG);
+  case ISD::UINT_TO_FP:
+  case ISD::SINT_TO_FP:         return LowerINT_TO_FP(Op, DAG);
   case ISD::FLT_ROUNDS_:        return LowerFLT_ROUNDS_(Op, DAG);
 
   // Lower 64-bit shifts.
@@ -5528,50 +5783,8 @@ void PPCTargetLowering::ReplaceNodeResults(SDNode *N,
                              MVT::f64, N->getOperand(0),
                              DAG.getIntPtrConstant(1));
 
-    // This sequence changes FPSCR to do round-to-zero, adds the two halves
-    // of the long double, and puts FPSCR back the way it was.  We do not
-    // actually model FPSCR.
-    std::vector<EVT> NodeTys;
-    SDValue Ops[4], Result, MFFSreg, InFlag, FPreg;
-
-    NodeTys.push_back(MVT::f64);   // Return register
-    NodeTys.push_back(MVT::Glue);    // Returns a flag for later insns
-    Result = DAG.getNode(PPCISD::MFFS, dl, NodeTys, &InFlag, 0);
-    MFFSreg = Result.getValue(0);
-    InFlag = Result.getValue(1);
-
-    NodeTys.clear();
-    NodeTys.push_back(MVT::Glue);   // Returns a flag
-    Ops[0] = DAG.getConstant(31, MVT::i32);
-    Ops[1] = InFlag;
-    Result = DAG.getNode(PPCISD::MTFSB1, dl, NodeTys, Ops, 2);
-    InFlag = Result.getValue(0);
-
-    NodeTys.clear();
-    NodeTys.push_back(MVT::Glue);   // Returns a flag
-    Ops[0] = DAG.getConstant(30, MVT::i32);
-    Ops[1] = InFlag;
-    Result = DAG.getNode(PPCISD::MTFSB0, dl, NodeTys, Ops, 2);
-    InFlag = Result.getValue(0);
-
-    NodeTys.clear();
-    NodeTys.push_back(MVT::f64);    // result of add
-    NodeTys.push_back(MVT::Glue);   // Returns a flag
-    Ops[0] = Lo;
-    Ops[1] = Hi;
-    Ops[2] = InFlag;
-    Result = DAG.getNode(PPCISD::FADDRTZ, dl, NodeTys, Ops, 3);
-    FPreg = Result.getValue(0);
-    InFlag = Result.getValue(1);
-
-    NodeTys.clear();
-    NodeTys.push_back(MVT::f64);
-    Ops[0] = DAG.getConstant(1, MVT::i32);
-    Ops[1] = MFFSreg;
-    Ops[2] = FPreg;
-    Ops[3] = InFlag;
-    Result = DAG.getNode(PPCISD::MTFSF, dl, NodeTys, Ops, 4);
-    FPreg = Result.getValue(0);
+    // Add the two halves of the long double in round-to-zero mode.
+    SDValue FPreg = DAG.getNode(PPCISD::FADDRTZ, dl, MVT::f64, Lo, Hi);
 
     // We know the low half is about to be thrown away, so just use something
     // convenient.
@@ -5663,7 +5876,7 @@ PPCTargetLowering::EmitPartwordAtomicBinary(MachineInstr *MI,
   // registers without caring whether they're 32 or 64, but here we're
   // doing actual arithmetic on the addresses.
   bool is64bit = PPCSubTarget.isPPC64();
-  unsigned ZeroReg = is64bit ? PPC::X0 : PPC::R0;
+  unsigned ZeroReg = is64bit ? PPC::ZERO8 : PPC::ZERO;
 
   const BasicBlock *LLVM_BB = BB->getBasicBlock();
   MachineFunction *F = BB->getParent();
@@ -5767,7 +5980,7 @@ PPCTargetLowering::EmitPartwordAtomicBinary(MachineInstr *MI,
     .addReg(TmpReg).addReg(MaskReg);
   BuildMI(BB, dl, TII->get(is64bit ? PPC::OR8 : PPC::OR), Tmp4Reg)
     .addReg(Tmp3Reg).addReg(Tmp2Reg);
-  BuildMI(BB, dl, TII->get(is64bit ? PPC::STDCX : PPC::STWCX))
+  BuildMI(BB, dl, TII->get(PPC::STWCX))
     .addReg(Tmp4Reg).addReg(ZeroReg).addReg(PtrReg);
   BuildMI(BB, dl, TII->get(PPC::BCC))
     .addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB);
@@ -5782,9 +5995,238 @@ PPCTargetLowering::EmitPartwordAtomicBinary(MachineInstr *MI,
   return BB;
 }
 
+llvm::MachineBasicBlock*
+PPCTargetLowering::emitEHSjLjSetJmp(MachineInstr *MI,
+                                    MachineBasicBlock *MBB) const {
+  DebugLoc DL = MI->getDebugLoc();
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+
+  MachineFunction *MF = MBB->getParent();
+  MachineRegisterInfo &MRI = MF->getRegInfo();
+
+  const BasicBlock *BB = MBB->getBasicBlock();
+  MachineFunction::iterator I = MBB;
+  ++I;
+
+  // Memory Reference
+  MachineInstr::mmo_iterator MMOBegin = MI->memoperands_begin();
+  MachineInstr::mmo_iterator MMOEnd = MI->memoperands_end();
+
+  unsigned DstReg = MI->getOperand(0).getReg();
+  const TargetRegisterClass *RC = MRI.getRegClass(DstReg);
+  assert(RC->hasType(MVT::i32) && "Invalid destination!");
+  unsigned mainDstReg = MRI.createVirtualRegister(RC);
+  unsigned restoreDstReg = MRI.createVirtualRegister(RC);
+
+  MVT PVT = getPointerTy();
+  assert((PVT == MVT::i64 || PVT == MVT::i32) &&
+         "Invalid Pointer Size!");
+  // For v = setjmp(buf), we generate
+  //
+  // thisMBB:
+  //  SjLjSetup mainMBB
+  //  bl mainMBB
+  //  v_restore = 1
+  //  b sinkMBB
+  //
+  // mainMBB:
+  //  buf[LabelOffset] = LR
+  //  v_main = 0
+  //
+  // sinkMBB:
+  //  v = phi(main, restore)
+  //
+
+  MachineBasicBlock *thisMBB = MBB;
+  MachineBasicBlock *mainMBB = MF->CreateMachineBasicBlock(BB);
+  MachineBasicBlock *sinkMBB = MF->CreateMachineBasicBlock(BB);
+  MF->insert(I, mainMBB);
+  MF->insert(I, sinkMBB);
+
+  MachineInstrBuilder MIB;
+
+  // Transfer the remainder of BB and its successor edges to sinkMBB.
+  sinkMBB->splice(sinkMBB->begin(), MBB,
+                  llvm::next(MachineBasicBlock::iterator(MI)), MBB->end());
+  sinkMBB->transferSuccessorsAndUpdatePHIs(MBB);
+
+  // Note that the structure of the jmp_buf used here is not compatible
+  // with that used by libc, and is not designed to be. Specifically, it
+  // stores only those 'reserved' registers that LLVM does not otherwise
+  // understand how to spill. Also, by convention, by the time this
+  // intrinsic is called, Clang has already stored the frame address in the
+  // first slot of the buffer and stack address in the third. Following the
+  // X86 target code, we'll store the jump address in the second slot. We also
+  // need to save the TOC pointer (R2) to handle jumps between shared
+  // libraries, and that will be stored in the fourth slot. The thread
+  // identifier (R13) is not affected.
+
+  // thisMBB:
+  const int64_t LabelOffset = 1 * PVT.getStoreSize();
+  const int64_t TOCOffset   = 3 * PVT.getStoreSize();
+
+  // Prepare IP either in reg.
+  const TargetRegisterClass *PtrRC = getRegClassFor(PVT);
+  unsigned LabelReg = MRI.createVirtualRegister(PtrRC);
+  unsigned BufReg = MI->getOperand(1).getReg();
+
+  if (PPCSubTarget.isPPC64() && PPCSubTarget.isSVR4ABI()) {
+    MIB = BuildMI(*thisMBB, MI, DL, TII->get(PPC::STD))
+            .addReg(PPC::X2)
+            .addImm(TOCOffset / 4)
+            .addReg(BufReg);
+
+    MIB.setMemRefs(MMOBegin, MMOEnd);
+  }
+
+  // Setup
+  MIB = BuildMI(*thisMBB, MI, DL, TII->get(PPC::BCLalways)).addMBB(mainMBB);
+  MIB.addRegMask(PPCRegInfo->getNoPreservedMask());
+
+  BuildMI(*thisMBB, MI, DL, TII->get(PPC::LI), restoreDstReg).addImm(1);
+
+  MIB = BuildMI(*thisMBB, MI, DL, TII->get(PPC::EH_SjLj_Setup))
+          .addMBB(mainMBB);
+  MIB = BuildMI(*thisMBB, MI, DL, TII->get(PPC::B)).addMBB(sinkMBB);
+
+  thisMBB->addSuccessor(mainMBB, /* weight */ 0);
+  thisMBB->addSuccessor(sinkMBB, /* weight */ 1);
+
+  // mainMBB:
+  //  mainDstReg = 0
+  MIB = BuildMI(mainMBB, DL,
+    TII->get(PPCSubTarget.isPPC64() ? PPC::MFLR8 : PPC::MFLR), LabelReg);
+
+  // Store IP
+  if (PPCSubTarget.isPPC64()) {
+    MIB = BuildMI(mainMBB, DL, TII->get(PPC::STD))
+            .addReg(LabelReg)
+            .addImm(LabelOffset / 4)
+            .addReg(BufReg);
+  } else {
+    MIB = BuildMI(mainMBB, DL, TII->get(PPC::STW))
+            .addReg(LabelReg)
+            .addImm(LabelOffset)
+            .addReg(BufReg);
+  }
+
+  MIB.setMemRefs(MMOBegin, MMOEnd);
+
+  BuildMI(mainMBB, DL, TII->get(PPC::LI), mainDstReg).addImm(0);
+  mainMBB->addSuccessor(sinkMBB);
+
+  // sinkMBB:
+  BuildMI(*sinkMBB, sinkMBB->begin(), DL,
+          TII->get(PPC::PHI), DstReg)
+    .addReg(mainDstReg).addMBB(mainMBB)
+    .addReg(restoreDstReg).addMBB(thisMBB);
+
+  MI->eraseFromParent();
+  return sinkMBB;
+}
+
+MachineBasicBlock *
+PPCTargetLowering::emitEHSjLjLongJmp(MachineInstr *MI,
+                                     MachineBasicBlock *MBB) const {
+  DebugLoc DL = MI->getDebugLoc();
+  const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+
+  MachineFunction *MF = MBB->getParent();
+  MachineRegisterInfo &MRI = MF->getRegInfo();
+
+  // Memory Reference
+  MachineInstr::mmo_iterator MMOBegin = MI->memoperands_begin();
+  MachineInstr::mmo_iterator MMOEnd = MI->memoperands_end();
+
+  MVT PVT = getPointerTy();
+  assert((PVT == MVT::i64 || PVT == MVT::i32) &&
+         "Invalid Pointer Size!");
+
+  const TargetRegisterClass *RC =
+    (PVT == MVT::i64) ? &PPC::G8RCRegClass : &PPC::GPRCRegClass;
+  unsigned Tmp = MRI.createVirtualRegister(RC);
+  // Since FP is only updated here but NOT referenced, it's treated as GPR.
+  unsigned FP  = (PVT == MVT::i64) ? PPC::X31 : PPC::R31;
+  unsigned SP  = (PVT == MVT::i64) ? PPC::X1 : PPC::R1;
+
+  MachineInstrBuilder MIB;
+
+  const int64_t LabelOffset = 1 * PVT.getStoreSize();
+  const int64_t SPOffset    = 2 * PVT.getStoreSize();
+  const int64_t TOCOffset   = 3 * PVT.getStoreSize();
+
+  unsigned BufReg = MI->getOperand(0).getReg();
+
+  // Reload FP (the jumped-to function may not have had a
+  // frame pointer, and if so, then its r31 will be restored
+  // as necessary).
+  if (PVT == MVT::i64) {
+    MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LD), FP)
+            .addImm(0)
+            .addReg(BufReg);
+  } else {
+    MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LWZ), FP)
+            .addImm(0)
+            .addReg(BufReg);
+  }
+  MIB.setMemRefs(MMOBegin, MMOEnd);
+
+  // Reload IP
+  if (PVT == MVT::i64) {
+    MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LD), Tmp)
+            .addImm(LabelOffset / 4)
+            .addReg(BufReg);
+  } else {
+    MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LWZ), Tmp)
+            .addImm(LabelOffset)
+            .addReg(BufReg);
+  }
+  MIB.setMemRefs(MMOBegin, MMOEnd);
+
+  // Reload SP
+  if (PVT == MVT::i64) {
+    MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LD), SP)
+            .addImm(SPOffset / 4)
+            .addReg(BufReg);
+  } else {
+    MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LWZ), SP)
+            .addImm(SPOffset)
+            .addReg(BufReg);
+  }
+  MIB.setMemRefs(MMOBegin, MMOEnd);
+
+  // FIXME: When we also support base pointers, that register must also be
+  // restored here.
+
+  // Reload TOC
+  if (PVT == MVT::i64 && PPCSubTarget.isSVR4ABI()) {
+    MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LD), PPC::X2)
+            .addImm(TOCOffset / 4)
+            .addReg(BufReg);
+
+    MIB.setMemRefs(MMOBegin, MMOEnd);
+  }
+
+  // Jump
+  BuildMI(*MBB, MI, DL,
+          TII->get(PVT == MVT::i64 ? PPC::MTCTR8 : PPC::MTCTR)).addReg(Tmp);
+  BuildMI(*MBB, MI, DL, TII->get(PVT == MVT::i64 ? PPC::BCTR8 : PPC::BCTR));
+
+  MI->eraseFromParent();
+  return MBB;
+}
+
 MachineBasicBlock *
 PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
                                                MachineBasicBlock *BB) const {
+  if (MI->getOpcode() == PPC::EH_SjLj_SetJmp32 ||
+      MI->getOpcode() == PPC::EH_SjLj_SetJmp64) {
+    return emitEHSjLjSetJmp(MI, BB);
+  } else if (MI->getOpcode() == PPC::EH_SjLj_LongJmp32 ||
+             MI->getOpcode() == PPC::EH_SjLj_LongJmp64) {
+    return emitEHSjLjLongJmp(MI, BB);
+  }
+
   const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
 
   // To "insert" these instructions we actually have to insert their
@@ -5802,24 +6244,24 @@ PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
     unsigned SelectPred = MI->getOperand(4).getImm();
     DebugLoc dl = MI->getDebugLoc();
 
-    // The SelectPred is ((BI << 5) | BO) for a BCC
-    unsigned BO = SelectPred & 0xF;
-    assert((BO == 12 || BO == 4) && "invalid predicate BO field for isel");
-
-    unsigned TrueOpNo, FalseOpNo;
-    if (BO == 12) {
-      TrueOpNo = 2;
-      FalseOpNo = 3;
-    } else {
-      TrueOpNo = 3;
-      FalseOpNo = 2;
-      SelectPred = PPC::InvertPredicate((PPC::Predicate)SelectPred);
+    unsigned SubIdx;
+    bool SwapOps;
+    switch (SelectPred) {
+    default: llvm_unreachable("invalid predicate for isel");
+    case PPC::PRED_EQ: SubIdx = PPC::sub_eq; SwapOps = false; break;
+    case PPC::PRED_NE: SubIdx = PPC::sub_eq; SwapOps = true; break;
+    case PPC::PRED_LT: SubIdx = PPC::sub_lt; SwapOps = false; break;
+    case PPC::PRED_GE: SubIdx = PPC::sub_lt; SwapOps = true; break;
+    case PPC::PRED_GT: SubIdx = PPC::sub_gt; SwapOps = false; break;
+    case PPC::PRED_LE: SubIdx = PPC::sub_gt; SwapOps = true; break;
+    case PPC::PRED_UN: SubIdx = PPC::sub_un; SwapOps = false; break;
+    case PPC::PRED_NU: SubIdx = PPC::sub_un; SwapOps = true; break;
     }
 
     BuildMI(*BB, MI, dl, TII->get(OpCode), MI->getOperand(0).getReg())
-      .addReg(MI->getOperand(TrueOpNo).getReg())
-      .addReg(MI->getOperand(FalseOpNo).getReg())
-      .addImm(SelectPred).addReg(MI->getOperand(1).getReg());
+      .addReg(MI->getOperand(SwapOps? 3 : 2).getReg())
+      .addReg(MI->getOperand(SwapOps? 2 : 3).getReg())
+      .addReg(MI->getOperand(1).getReg(), 0, SubIdx);
   } else if (MI->getOpcode() == PPC::SELECT_CC_I4 ||
              MI->getOpcode() == PPC::SELECT_CC_I8 ||
              MI->getOpcode() == PPC::SELECT_CC_F4 ||
@@ -6052,7 +6494,7 @@ PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
     unsigned TmpDestReg = RegInfo.createVirtualRegister(RC);
     unsigned Ptr1Reg;
     unsigned TmpReg = RegInfo.createVirtualRegister(RC);
-    unsigned ZeroReg = is64bit ? PPC::X0 : PPC::R0;
+    unsigned ZeroReg = is64bit ? PPC::ZERO8 : PPC::ZERO;
     //  thisMBB:
     //   ...
     //   fallthrough --> loopMBB
@@ -6155,6 +6597,75 @@ PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
     BB = exitMBB;
     BuildMI(*BB, BB->begin(), dl, TII->get(PPC::SRW),dest).addReg(TmpReg)
       .addReg(ShiftReg);
+  } else if (MI->getOpcode() == PPC::FADDrtz) {
+    // This pseudo performs an FADD with rounding mode temporarily forced
+    // to round-to-zero.  We emit this via custom inserter since the FPSCR
+    // is not modeled at the SelectionDAG level.
+    unsigned Dest = MI->getOperand(0).getReg();
+    unsigned Src1 = MI->getOperand(1).getReg();
+    unsigned Src2 = MI->getOperand(2).getReg();
+    DebugLoc dl   = MI->getDebugLoc();
+
+    MachineRegisterInfo &RegInfo = F->getRegInfo();
+    unsigned MFFSReg = RegInfo.createVirtualRegister(&PPC::F8RCRegClass);
+
+    // Save FPSCR value.
+    BuildMI(*BB, MI, dl, TII->get(PPC::MFFS), MFFSReg);
+
+    // Set rounding mode to round-to-zero.
+    BuildMI(*BB, MI, dl, TII->get(PPC::MTFSB1)).addImm(31);
+    BuildMI(*BB, MI, dl, TII->get(PPC::MTFSB0)).addImm(30);
+
+    // Perform addition.
+    BuildMI(*BB, MI, dl, TII->get(PPC::FADD), Dest).addReg(Src1).addReg(Src2);
+
+    // Restore FPSCR value.
+    BuildMI(*BB, MI, dl, TII->get(PPC::MTFSF)).addImm(1).addReg(MFFSReg);
+  } else if (MI->getOpcode() == PPC::FRINDrint ||
+             MI->getOpcode() == PPC::FRINSrint) {
+    bool isf32 = MI->getOpcode() == PPC::FRINSrint;
+    unsigned Dest = MI->getOperand(0).getReg();
+    unsigned Src = MI->getOperand(1).getReg();
+    DebugLoc dl   = MI->getDebugLoc();
+
+    MachineRegisterInfo &RegInfo = F->getRegInfo();
+    unsigned CRReg = RegInfo.createVirtualRegister(&PPC::CRRCRegClass);
+
+    // Perform the rounding.
+    BuildMI(*BB, MI, dl, TII->get(isf32 ? PPC::FRINS : PPC::FRIND), Dest)
+      .addReg(Src);
+
+    // Compare the results.
+    BuildMI(*BB, MI, dl, TII->get(isf32 ? PPC::FCMPUS : PPC::FCMPUD), CRReg)
+      .addReg(Dest).addReg(Src);
+
+    // If the results were not equal, then set the FPSCR XX bit.
+    MachineBasicBlock *midMBB = F->CreateMachineBasicBlock(LLVM_BB);
+    MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
+    F->insert(It, midMBB);
+    F->insert(It, exitMBB);
+    exitMBB->splice(exitMBB->begin(), BB,
+                    llvm::next(MachineBasicBlock::iterator(MI)),
+                    BB->end());
+    exitMBB->transferSuccessorsAndUpdatePHIs(BB);
+
+    BuildMI(*BB, MI, dl, TII->get(PPC::BCC))
+      .addImm(PPC::PRED_EQ).addReg(CRReg).addMBB(exitMBB);
+
+    BB->addSuccessor(midMBB);
+    BB->addSuccessor(exitMBB);
+
+    BB = midMBB;
+
+    // Set the FPSCR XX bit (FE_INEXACT). Note that we cannot just set
+    // the FI bit here because that will not automatically set XX also,
+    // and XX is what libm interprets as the FE_INEXACT flag.
+    BuildMI(BB, dl, TII->get(PPC::MTFSB1)).addImm(/* 38 - 32 = */ 6);
+    BuildMI(BB, dl, TII->get(PPC::B)).addMBB(exitMBB);
+
+    BB->addSuccessor(exitMBB);
+
+    BB = exitMBB;
   } else {
     llvm_unreachable("Unexpected instr type to insert");
   }
@@ -6167,6 +6678,139 @@ PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
 // Target Optimization Hooks
 //===----------------------------------------------------------------------===//
 
+SDValue PPCTargetLowering::DAGCombineFastRecip(SDValue Op,
+                                               DAGCombinerInfo &DCI) const {
+  if (DCI.isAfterLegalizeVectorOps())
+    return SDValue();
+
+  EVT VT = Op.getValueType();
+
+  if ((VT == MVT::f32 && PPCSubTarget.hasFRES()) ||
+      (VT == MVT::f64 && PPCSubTarget.hasFRE())  ||
+      (VT == MVT::v4f32 && PPCSubTarget.hasAltivec())) {
+
+    // Newton iteration for a function: F(X) is X_{i+1} = X_i - F(X_i)/F'(X_i)
+    // For the reciprocal, we need to find the zero of the function:
+    //   F(X) = A X - 1 [which has a zero at X = 1/A]
+    //     =>
+    //   X_{i+1} = X_i (2 - A X_i) = X_i + X_i (1 - A X_i) [this second form
+    //     does not require additional intermediate precision]
+
+    // Convergence is quadratic, so we essentially double the number of digits
+    // correct after every iteration. The minimum architected relative
+    // accuracy is 2^-5. When hasRecipPrec(), this is 2^-14. IEEE float has
+    // 23 digits and double has 52 digits.
+    int Iterations = PPCSubTarget.hasRecipPrec() ? 1 : 3;
+    if (VT.getScalarType() == MVT::f64)
+      ++Iterations;
+
+    SelectionDAG &DAG = DCI.DAG;
+    DebugLoc dl = Op.getDebugLoc();
+
+    SDValue FPOne =
+      DAG.getConstantFP(1.0, VT.getScalarType());
+    if (VT.isVector()) {
+      assert(VT.getVectorNumElements() == 4 &&
+             "Unknown vector type");
+      FPOne = DAG.getNode(ISD::BUILD_VECTOR, dl, VT,
+                          FPOne, FPOne, FPOne, FPOne);
+    }
+
+    SDValue Est = DAG.getNode(PPCISD::FRE, dl, VT, Op);
+    DCI.AddToWorklist(Est.getNode());
+
+    // Newton iterations: Est = Est + Est (1 - Arg * Est)
+    for (int i = 0; i < Iterations; ++i) {
+      SDValue NewEst = DAG.getNode(ISD::FMUL, dl, VT, Op, Est);
+      DCI.AddToWorklist(NewEst.getNode());
+
+      NewEst = DAG.getNode(ISD::FSUB, dl, VT, FPOne, NewEst);
+      DCI.AddToWorklist(NewEst.getNode());
+
+      NewEst = DAG.getNode(ISD::FMUL, dl, VT, Est, NewEst);
+      DCI.AddToWorklist(NewEst.getNode());
+
+      Est = DAG.getNode(ISD::FADD, dl, VT, Est, NewEst);
+      DCI.AddToWorklist(Est.getNode());
+    }
+
+    return Est;
+  }
+
+  return SDValue();
+}
+
+SDValue PPCTargetLowering::DAGCombineFastRecipFSQRT(SDValue Op,
+                                             DAGCombinerInfo &DCI) const {
+  if (DCI.isAfterLegalizeVectorOps())
+    return SDValue();
+
+  EVT VT = Op.getValueType();
+
+  if ((VT == MVT::f32 && PPCSubTarget.hasFRSQRTES()) ||
+      (VT == MVT::f64 && PPCSubTarget.hasFRSQRTE())  ||
+      (VT == MVT::v4f32 && PPCSubTarget.hasAltivec())) {
+
+    // Newton iteration for a function: F(X) is X_{i+1} = X_i - F(X_i)/F'(X_i)
+    // For the reciprocal sqrt, we need to find the zero of the function:
+    //   F(X) = 1/X^2 - A [which has a zero at X = 1/sqrt(A)]
+    //     =>
+    //   X_{i+1} = X_i (1.5 - A X_i^2 / 2)
+    // As a result, we precompute A/2 prior to the iteration loop.
+
+    // Convergence is quadratic, so we essentially double the number of digits
+    // correct after every iteration. The minimum architected relative
+    // accuracy is 2^-5. When hasRecipPrec(), this is 2^-14. IEEE float has
+    // 23 digits and double has 52 digits.
+    int Iterations = PPCSubTarget.hasRecipPrec() ? 1 : 3;
+    if (VT.getScalarType() == MVT::f64)
+      ++Iterations;
+
+    SelectionDAG &DAG = DCI.DAG;
+    DebugLoc dl = Op.getDebugLoc();
+
+    SDValue FPThreeHalves =
+      DAG.getConstantFP(1.5, VT.getScalarType());
+    if (VT.isVector()) {
+      assert(VT.getVectorNumElements() == 4 &&
+             "Unknown vector type");
+      FPThreeHalves = DAG.getNode(ISD::BUILD_VECTOR, dl, VT,
+                                  FPThreeHalves, FPThreeHalves,
+                                  FPThreeHalves, FPThreeHalves);
+    }
+
+    SDValue Est = DAG.getNode(PPCISD::FRSQRTE, dl, VT, Op);
+    DCI.AddToWorklist(Est.getNode());
+
+    // We now need 0.5*Arg which we can write as (1.5*Arg - Arg) so that
+    // this entire sequence requires only one FP constant.
+    SDValue HalfArg = DAG.getNode(ISD::FMUL, dl, VT, FPThreeHalves, Op);
+    DCI.AddToWorklist(HalfArg.getNode());
+
+    HalfArg = DAG.getNode(ISD::FSUB, dl, VT, HalfArg, Op);
+    DCI.AddToWorklist(HalfArg.getNode());
+
+    // Newton iterations: Est = Est * (1.5 - HalfArg * Est * Est)
+    for (int i = 0; i < Iterations; ++i) {
+      SDValue NewEst = DAG.getNode(ISD::FMUL, dl, VT, Est, Est);
+      DCI.AddToWorklist(NewEst.getNode());
+
+      NewEst = DAG.getNode(ISD::FMUL, dl, VT, HalfArg, NewEst);
+      DCI.AddToWorklist(NewEst.getNode());
+
+      NewEst = DAG.getNode(ISD::FSUB, dl, VT, FPThreeHalves, NewEst);
+      DCI.AddToWorklist(NewEst.getNode());
+
+      Est = DAG.getNode(ISD::FMUL, dl, VT, Est, NewEst);
+      DCI.AddToWorklist(Est.getNode());
+    }
+
+    return Est;
+  }
+
+  return SDValue();
+}
+
 SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
                                              DAGCombinerInfo &DCI) const {
   const TargetMachine &TM = getTargetMachine();
@@ -6193,7 +6837,72 @@ SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
         return N->getOperand(0);
     }
     break;
+  case ISD::FDIV: {
+    assert(TM.Options.UnsafeFPMath &&
+           "Reciprocal estimates require UnsafeFPMath");
+
+    if (N->getOperand(1).getOpcode() == ISD::FSQRT) {
+      SDValue RV =
+        DAGCombineFastRecipFSQRT(N->getOperand(1).getOperand(0), DCI);
+      if (RV.getNode() != 0) {
+        DCI.AddToWorklist(RV.getNode());
+        return DAG.getNode(ISD::FMUL, dl, N->getValueType(0),
+                           N->getOperand(0), RV);
+      }
+    } else if (N->getOperand(1).getOpcode() == ISD::FP_EXTEND &&
+               N->getOperand(1).getOperand(0).getOpcode() == ISD::FSQRT) {
+      SDValue RV =
+        DAGCombineFastRecipFSQRT(N->getOperand(1).getOperand(0).getOperand(0),
+                                 DCI);
+      if (RV.getNode() != 0) {
+        DCI.AddToWorklist(RV.getNode());
+        RV = DAG.getNode(ISD::FP_EXTEND, N->getOperand(1).getDebugLoc(),
+                         N->getValueType(0), RV);
+        DCI.AddToWorklist(RV.getNode());
+        return DAG.getNode(ISD::FMUL, dl, N->getValueType(0),
+                           N->getOperand(0), RV);
+      }
+    } else if (N->getOperand(1).getOpcode() == ISD::FP_ROUND &&
+               N->getOperand(1).getOperand(0).getOpcode() == ISD::FSQRT) {
+      SDValue RV =
+        DAGCombineFastRecipFSQRT(N->getOperand(1).getOperand(0).getOperand(0),
+                                 DCI);
+      if (RV.getNode() != 0) {
+        DCI.AddToWorklist(RV.getNode());
+        RV = DAG.getNode(ISD::FP_ROUND, N->getOperand(1).getDebugLoc(),
+                         N->getValueType(0), RV,
+                         N->getOperand(1).getOperand(1));
+        DCI.AddToWorklist(RV.getNode());
+        return DAG.getNode(ISD::FMUL, dl, N->getValueType(0),
+                           N->getOperand(0), RV);
+      }
+    }
+
+    SDValue RV = DAGCombineFastRecip(N->getOperand(1), DCI);
+    if (RV.getNode() != 0) {
+      DCI.AddToWorklist(RV.getNode());
+      return DAG.getNode(ISD::FMUL, dl, N->getValueType(0),
+                         N->getOperand(0), RV);
+    }
+
+    }
+    break;
+  case ISD::FSQRT: {
+    assert(TM.Options.UnsafeFPMath &&
+           "Reciprocal estimates require UnsafeFPMath");
+
+    // Compute this as 1/(1/sqrt(X)), which is the reciprocal of the
+    // reciprocal sqrt.
+    SDValue RV = DAGCombineFastRecipFSQRT(N->getOperand(0), DCI);
+    if (RV.getNode() != 0) {
+      DCI.AddToWorklist(RV.getNode());
+      RV = DAGCombineFastRecip(RV, DCI);
+      if (RV.getNode() != 0)
+        return RV;
+    }
 
+    }
+    break;
   case ISD::SINT_TO_FP:
     if (TM.getSubtarget<PPCSubtarget>().has64BitSupport()) {
       if (N->getOperand(0).getOpcode() == ISD::FP_TO_SINT) {
@@ -6240,8 +6949,15 @@ SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
       Val = DAG.getNode(PPCISD::FCTIWZ, dl, MVT::f64, Val);
       DCI.AddToWorklist(Val.getNode());
 
-      Val = DAG.getNode(PPCISD::STFIWX, dl, MVT::Other, N->getOperand(0), Val,
-                        N->getOperand(2), N->getOperand(3));
+      SDValue Ops[] = {
+        N->getOperand(0), Val, N->getOperand(2),
+        DAG.getValueType(N->getOperand(1).getValueType())
+      };
+
+      Val = DAG.getMemIntrinsicNode(PPCISD::STFIWX, dl,
+              DAG.getVTList(MVT::Other), Ops, array_lengthof(Ops),
+              cast<StoreSDNode>(N)->getMemoryVT(),
+              cast<StoreSDNode>(N)->getMemOperand());
       DCI.AddToWorklist(Val.getNode());
       return Val;
     }
@@ -6251,7 +6967,10 @@ SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
         N->getOperand(1).getOpcode() == ISD::BSWAP &&
         N->getOperand(1).getNode()->hasOneUse() &&
         (N->getOperand(1).getValueType() == MVT::i32 ||
-         N->getOperand(1).getValueType() == MVT::i16)) {
+         N->getOperand(1).getValueType() == MVT::i16 ||
+         (TM.getSubtarget<PPCSubtarget>().hasLDBRX() &&
+          TM.getSubtarget<PPCSubtarget>().isPPC64() &&
+          N->getOperand(1).getValueType() == MVT::i64))) {
       SDValue BSwapOp = N->getOperand(1).getOperand(0);
       // Do an any-extend to 32-bits if this is a half-word input.
       if (BSwapOp.getValueType() == MVT::i16)
@@ -6272,7 +6991,10 @@ SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
     // Turn BSWAP (LOAD) -> lhbrx/lwbrx.
     if (ISD::isNON_EXTLoad(N->getOperand(0).getNode()) &&
         N->getOperand(0).hasOneUse() &&
-        (N->getValueType(0) == MVT::i32 || N->getValueType(0) == MVT::i16)) {
+        (N->getValueType(0) == MVT::i32 || N->getValueType(0) == MVT::i16 ||
+         (TM.getSubtarget<PPCSubtarget>().hasLDBRX() &&
+          TM.getSubtarget<PPCSubtarget>().isPPC64() &&
+          N->getValueType(0) == MVT::i64))) {
       SDValue Load = N->getOperand(0);
       LoadSDNode *LD = cast<LoadSDNode>(Load);
       // Create the byte-swapping load.
@@ -6283,8 +7005,9 @@ SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
       };
       SDValue BSLoad =
         DAG.getMemIntrinsicNode(PPCISD::LBRX, dl,
-                                DAG.getVTList(MVT::i32, MVT::Other), Ops, 3,
-                                LD->getMemoryVT(), LD->getMemOperand());
+                                DAG.getVTList(N->getValueType(0) == MVT::i64 ?
+                                              MVT::i64 : MVT::i32, MVT::Other),
+                                Ops, 3, LD->getMemoryVT(), LD->getMemOperand());
 
       // If this is an i16 load, insert the truncate.
       SDValue ResVal = BSLoad;
@@ -6384,14 +7107,12 @@ SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
       bool BranchOnWhenPredTrue = (CC == ISD::SETEQ) ^ (Val == 0);
 
       // Create the PPCISD altivec 'dot' comparison node.
-      std::vector<EVT> VTs;
       SDValue Ops[] = {
         LHS.getOperand(2),  // LHS of compare
         LHS.getOperand(3),  // RHS of compare
         DAG.getConstant(CompareOpc, MVT::i32)
       };
-      VTs.push_back(LHS.getOperand(2).getValueType());
-      VTs.push_back(MVT::Glue);
+      EVT VTs[] = { LHS.getOperand(2).getValueType(), MVT::Glue };
       SDValue CompNode = DAG.getNode(PPCISD::VCMPo, dl, VTs, Ops, 3);
 
       // Unpack the result based on how the target uses it.
@@ -6543,6 +7264,9 @@ PPCTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
     // GCC RS6000 Constraint Letters
     switch (Constraint[0]) {
     case 'b':   // R1-R31
+      if (VT == MVT::i64 && PPCSubTarget.isPPC64())
+        return std::make_pair(0U, &PPC::G8RC_NOX0RegClass);
+      return std::make_pair(0U, &PPC::GPRC_NOR0RegClass);
     case 'r':   // R0-R31
       if (VT == MVT::i64 && PPCSubTarget.isPPC64())
         return std::make_pair(0U, &PPC::G8RCRegClass);
@@ -6727,13 +7451,16 @@ SDValue PPCTargetLowering::LowerFRAMEADDR(SDValue Op,
   MachineFunction &MF = DAG.getMachineFunction();
   MachineFrameInfo *MFI = MF.getFrameInfo();
   MFI->setFrameAddressIsTaken(true);
-  bool is31 = (getTargetMachine().Options.DisableFramePointerElim(MF) ||
-               MFI->hasVarSizedObjects()) &&
-                  MFI->getStackSize() &&
-                  !MF.getFunction()->getFnAttributes().
-                    hasAttribute(Attributes::Naked);
-  unsigned FrameReg = isPPC64 ? (is31 ? PPC::X31 : PPC::X1) :
-                                (is31 ? PPC::R31 : PPC::R1);
+
+  // Naked functions never have a frame pointer, and so we use r1. For all
+  // other functions, this decision must be delayed until during PEI.
+  unsigned FrameReg;
+  if (MF.getFunction()->getAttributes().hasAttribute(
+        AttributeSet::FunctionIndex, Attribute::Naked))
+    FrameReg = isPPC64 ? PPC::X1 : PPC::R1;
+  else
+    FrameReg = isPPC64 ? PPC::FP8 : PPC::FP;
+
   SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl, FrameReg,
                                          PtrVT);
   while (Depth--)
@@ -6754,16 +7481,15 @@ PPCTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
 /// lowering. If DstAlign is zero that means it's safe to destination
 /// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
 /// means there isn't a need to check it against alignment requirement,
-/// probably because the source does not need to be loaded. If
-/// 'IsZeroVal' is true, that means it's safe to return a
-/// non-scalar-integer type, e.g. empty string source, constant, or loaded
-/// from memory. 'MemcpyStrSrc' indicates whether the memcpy source is
-/// constant so it does not need to be loaded.
+/// probably because the source does not need to be loaded. If 'IsMemset' is
+/// true, that means it's expanding a memset. If 'ZeroMemset' is true, that
+/// means it's a memset of zero. 'MemcpyStrSrc' indicates whether the memcpy
+/// source is constant so it does not need to be loaded.
 /// It returns EVT::Other if the type should be determined using generic
 /// target-independent logic.
 EVT PPCTargetLowering::getOptimalMemOpType(uint64_t Size,
                                            unsigned DstAlign, unsigned SrcAlign,
-                                           bool IsZeroVal,
+                                           bool IsMemset, bool ZeroMemset,
                                            bool MemcpyStrSrc,
                                            MachineFunction &MF) const {
   if (this->PPCSubTarget.isPPC64()) {
@@ -6773,6 +7499,32 @@ EVT PPCTargetLowering::getOptimalMemOpType(uint64_t Size,
   }
 }
 
+bool PPCTargetLowering::allowsUnalignedMemoryAccesses(EVT VT,
+                                                      bool *Fast) const {
+  if (DisablePPCUnaligned)
+    return false;
+
+  // PowerPC supports unaligned memory access for simple non-vector types.
+  // Although accessing unaligned addresses is not as efficient as accessing
+  // aligned addresses, it is generally more efficient than manual expansion,
+  // and generally only traps for software emulation when crossing page
+  // boundaries.
+
+  if (!VT.isSimple())
+    return false;
+
+  if (VT.getSimpleVT().isVector())
+    return false;
+
+  if (VT == MVT::ppcf128)
+    return false;
+
+  if (Fast)
+    *Fast = true;
+
+  return true;
+}
+
 /// isFMAFasterThanMulAndAdd - Return true if an FMA operation is faster than
 /// a pair of mul and add instructions. fmuladd intrinsics will be expanded to
 /// FMAs when this method returns true (and FMAs are legal), otherwise fmuladd
diff --git a/lib/Target/PowerPC/PPCISelLowering.h b/lib/Target/PowerPC/PPCISelLowering.h
index b3c7f9c28d40..7157b70d8622 100644
--- a/lib/Target/PowerPC/PPCISelLowering.h
+++ b/lib/Target/PowerPC/PPCISelLowering.h
@@ -16,9 +16,10 @@
 #define LLVM_TARGET_POWERPC_PPC32ISELLOWERING_H
 
 #include "PPC.h"
+#include "PPCRegisterInfo.h"
 #include "PPCSubtarget.h"
-#include "llvm/Target/TargetLowering.h"
 #include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/Target/TargetLowering.h"
 
 namespace llvm {
   namespace PPCISD {
@@ -35,14 +36,21 @@ namespace llvm {
       /// was temporarily in the f64 operand.
       FCFID,
 
+      /// Newer FCFID[US] integer-to-floating-point conversion instructions for
+      /// unsigned integers and single-precision outputs.
+      FCFIDU, FCFIDS, FCFIDUS,
+
       /// FCTI[D,W]Z - The FCTIDZ and FCTIWZ instructions, taking an f32 or f64
       /// operand, producing an f64 value containing the integer representation
       /// of that FP value.
       FCTIDZ, FCTIWZ,
 
-      /// STFIWX - The STFIWX instruction.  The first operand is an input token
-      /// chain, then an f64 value to store, then an address to store it to.
-      STFIWX,
+      /// Newer FCTI[D,W]UZ floating-point-to-integer conversion instructions for
+      /// unsigned integers.
+      FCTIDUZ, FCTIWUZ,
+
+      /// Reciprocal estimate instructions (unary FP ops).
+      FRE, FRSQRTE,
 
       // VMADDFP, VNMSUBFP - The VMADDFP and VNMSUBFP instructions, taking
       // three v4f32 operands and producing a v4f32 result.
@@ -90,17 +98,10 @@ namespace llvm {
       /// code.
       SRL, SRA, SHL,
 
-      /// EXTSW_32 - This is the EXTSW instruction for use with "32-bit"
-      /// registers.
-      EXTSW_32,
-
       /// CALL - A direct function call.
-      /// CALL_NOP_SVR4 is a call with the special  NOP which follows 64-bit
+      /// CALL_NOP is a call with the special NOP which follows 64-bit
       /// SVR4 calls.
-      CALL_Darwin, CALL_SVR4, CALL_NOP_SVR4,
-
-      /// NOP - Special NOP which follows 64-bit SVR4 calls.
-      NOP,
+      CALL, CALL_NOP,
 
       /// CHAIN,FLAG = MTCTR(VAL, CHAIN[, INFLAG]) - Directly corresponds to a
       /// MTCTR instruction.
@@ -108,7 +109,7 @@ namespace llvm {
 
       /// CHAIN,FLAG = BCTRL(CHAIN, INFLAG) - Directly corresponds to a
       /// BCTRL instruction.
-      BCTRL_Darwin, BCTRL_SVR4,
+      BCTRL,
 
       /// Return with a flag operand, matched by 'blr'
       RET_FLAG,
@@ -119,6 +120,12 @@ namespace llvm {
       /// are undefined.
       MFCR,
 
+      // EH_SJLJ_SETJMP - SjLj exception handling setjmp.
+      EH_SJLJ_SETJMP,
+
+      // EH_SJLJ_LONGJMP - SjLj exception handling longjmp.
+      EH_SJLJ_LONGJMP,
+
       /// RESVEC = VCMP(LHS, RHS, OPC) - Represents one of the altivec VCMP*
       /// instructions.  For lack of better number, we use the opcode number
       /// encoding for the OPC field to identify the compare.  For example, 838
@@ -138,26 +145,13 @@ namespace llvm {
       /// an optional input flag argument.
       COND_BRANCH,
 
-      // The following 5 instructions are used only as part of the
-      // long double-to-int conversion sequence.
-
-      /// OUTFLAG = MFFS F8RC - This moves the FPSCR (not modelled) into the
-      /// register.
-      MFFS,
-
-      /// OUTFLAG = MTFSB0 INFLAG - This clears a bit in the FPSCR.
-      MTFSB0,
-
-      /// OUTFLAG = MTFSB1 INFLAG - This sets a bit in the FPSCR.
-      MTFSB1,
-
-      /// F8RC, OUTFLAG = FADDRTZ F8RC, F8RC, INFLAG - This is an FADD done with
-      /// rounding towards zero.  It has flags added so it won't move past the
-      /// FPSCR-setting instructions.
+      /// F8RC = FADDRTZ F8RC, F8RC - This is an FADD done with rounding
+      /// towards zero.  Used only as part of the long double-to-int
+      /// conversion sequence.
       FADDRTZ,
 
-      /// MTFSF = F8RC, INFLAG - This moves the register into the FPSCR.
-      MTFSF,
+      /// F8RC = MFFS - This moves the FPSCR (not modeled) into the register.
+      MFFS,
 
       /// LARX = This corresponds to PPC l{w|d}arx instrcution: load and
       /// reserve indexed. This is used to implement atomic operations.
@@ -178,20 +172,111 @@ namespace llvm {
       CR6SET,
       CR6UNSET,
 
-      /// STD_32 - This is the STD instruction for use with "32-bit" registers.
-      STD_32 = ISD::FIRST_TARGET_MEMORY_OPCODE,
+      /// G8RC = ADDIS_GOT_TPREL_HA %X2, Symbol - Used by the initial-exec
+      /// TLS model, produces an ADDIS8 instruction that adds the GOT
+      /// base to sym@got@tprel@ha.
+      ADDIS_GOT_TPREL_HA,
+
+      /// G8RC = LD_GOT_TPREL_L Symbol, G8RReg - Used by the initial-exec
+      /// TLS model, produces a LD instruction with base register G8RReg
+      /// and offset sym@got@tprel@l.  This completes the addition that
+      /// finds the offset of "sym" relative to the thread pointer.
+      LD_GOT_TPREL_L,
+
+      /// G8RC = ADD_TLS G8RReg, Symbol - Used by the initial-exec TLS
+      /// model, produces an ADD instruction that adds the contents of
+      /// G8RReg to the thread pointer.  Symbol contains a relocation
+      /// sym@tls which is to be replaced by the thread pointer and
+      /// identifies to the linker that the instruction is part of a
+      /// TLS sequence.
+      ADD_TLS,
+
+      /// G8RC = ADDIS_TLSGD_HA %X2, Symbol - For the general-dynamic TLS
+      /// model, produces an ADDIS8 instruction that adds the GOT base
+      /// register to sym@got@tlsgd@ha.
+      ADDIS_TLSGD_HA,
+
+      /// G8RC = ADDI_TLSGD_L G8RReg, Symbol - For the general-dynamic TLS
+      /// model, produces an ADDI8 instruction that adds G8RReg to
+      /// sym@got@tlsgd@l.
+      ADDI_TLSGD_L,
+
+      /// G8RC = GET_TLS_ADDR %X3, Symbol - For the general-dynamic TLS
+      /// model, produces a call to __tls_get_addr(sym@tlsgd).
+      GET_TLS_ADDR,
+
+      /// G8RC = ADDIS_TLSLD_HA %X2, Symbol - For the local-dynamic TLS
+      /// model, produces an ADDIS8 instruction that adds the GOT base
+      /// register to sym@got@tlsld@ha.
+      ADDIS_TLSLD_HA,
+
+      /// G8RC = ADDI_TLSLD_L G8RReg, Symbol - For the local-dynamic TLS
+      /// model, produces an ADDI8 instruction that adds G8RReg to
+      /// sym@got@tlsld@l.
+      ADDI_TLSLD_L,
+
+      /// G8RC = GET_TLSLD_ADDR %X3, Symbol - For the local-dynamic TLS
+      /// model, produces a call to __tls_get_addr(sym@tlsld).
+      GET_TLSLD_ADDR,
+
+      /// G8RC = ADDIS_DTPREL_HA %X3, Symbol, Chain - For the
+      /// local-dynamic TLS model, produces an ADDIS8 instruction
+      /// that adds X3 to sym@dtprel@ha.  The Chain operand is needed 
+      /// to tie this in place following a copy to %X3 from the result
+      /// of a GET_TLSLD_ADDR.
+      ADDIS_DTPREL_HA,
+
+      /// G8RC = ADDI_DTPREL_L G8RReg, Symbol - For the local-dynamic TLS
+      /// model, produces an ADDI8 instruction that adds G8RReg to
+      /// sym@got@dtprel@l.
+      ADDI_DTPREL_L,
+
+      /// VRRC = VADD_SPLAT Elt, EltSize - Temporary node to be expanded
+      /// during instruction selection to optimize a BUILD_VECTOR into
+      /// operations on splats.  This is necessary to avoid losing these
+      /// optimizations due to constant folding.
+      VADD_SPLAT,
 
       /// CHAIN = STBRX CHAIN, GPRC, Ptr, Type - This is a
       /// byte-swapping store instruction.  It byte-swaps the low "Type" bits of
       /// the GPRC input, then stores it through Ptr.  Type can be either i16 or
       /// i32.
-      STBRX,
+      STBRX = ISD::FIRST_TARGET_MEMORY_OPCODE,
 
       /// GPRC, CHAIN = LBRX CHAIN, Ptr, Type - This is a
       /// byte-swapping load instruction.  It loads "Type" bits, byte swaps it,
       /// then puts it in the bottom bits of the GPRC.  TYPE can be either i16
       /// or i32.
-      LBRX
+      LBRX,
+
+      /// STFIWX - The STFIWX instruction.  The first operand is an input token
+      /// chain, then an f64 value to store, then an address to store it to.
+      STFIWX,
+
+      /// GPRC, CHAIN = LFIWAX CHAIN, Ptr - This is a floating-point
+      /// load which sign-extends from a 32-bit integer value into the
+      /// destination 64-bit register.
+      LFIWAX,
+
+      /// GPRC, CHAIN = LFIWZX CHAIN, Ptr - This is a floating-point
+      /// load which zero-extends from a 32-bit integer value into the
+      /// destination 64-bit register.
+      LFIWZX,
+
+      /// G8RC = ADDIS_TOC_HA %X2, Symbol - For medium and large code model,
+      /// produces an ADDIS8 instruction that adds the TOC base register to
+      /// sym@toc@ha.
+      ADDIS_TOC_HA,
+
+      /// G8RC = LD_TOC_L Symbol, G8RReg - For medium and large code model,
+      /// produces a LD instruction with base register G8RReg and offset
+      /// sym@toc@l.  Preceded by an ADDIS_TOC_HA to form a full 32-bit offset.
+      LD_TOC_L,
+
+      /// G8RC = ADDI_TOC_L G8RReg, Symbol - For medium code model, produces
+      /// an ADDI8 instruction that adds G8RReg to sym@toc@l.
+      /// Preceded by an ADDIS_TOC_HA to form a full 32-bit offset.
+      ADDI_TOC_L
     };
   }
 
@@ -241,6 +326,7 @@ namespace llvm {
 
   class PPCTargetLowering : public TargetLowering {
     const PPCSubtarget &PPCSubTarget;
+    const PPCRegisterInfo *PPCRegInfo;
 
   public:
     explicit PPCTargetLowering(PPCTargetMachine &TM);
@@ -249,7 +335,7 @@ namespace llvm {
     /// DAG node.
     virtual const char *getTargetNodeName(unsigned Opcode) const;
 
-    virtual MVT getShiftAmountTy(EVT LHSTy) const { return MVT::i32; }
+    virtual MVT getScalarShiftAmountTy(EVT LHSTy) const { return MVT::i32; }
 
     /// getSetCCResultType - Return the ISD::SETCC ValueType
     virtual EVT getSetCCResultType(EVT VT) const;
@@ -315,6 +401,12 @@ namespace llvm {
                                                 MachineBasicBlock *MBB,
                                             bool is8bit, unsigned Opcode) const;
 
+    MachineBasicBlock *emitEHSjLjSetJmp(MachineInstr *MI,
+                                        MachineBasicBlock *MBB) const;
+
+    MachineBasicBlock *emitEHSjLjLongJmp(MachineInstr *MI,
+                                         MachineBasicBlock *MBB) const;
+
     ConstraintType getConstraintType(const std::string &Constraint) const;
 
     /// Examine constraint string and operand type and determine a weight value.
@@ -358,18 +450,21 @@ namespace llvm {
     /// lowering. If DstAlign is zero that means it's safe to destination
     /// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
     /// means there isn't a need to check it against alignment requirement,
-    /// probably because the source does not need to be loaded. If
-    /// 'IsZeroVal' is true, that means it's safe to return a
-    /// non-scalar-integer type, e.g. empty string source, constant, or loaded
-    /// from memory. 'MemcpyStrSrc' indicates whether the memcpy source is
-    /// constant so it does not need to be loaded.
+    /// probably because the source does not need to be loaded. If 'IsMemset' is
+    /// true, that means it's expanding a memset. If 'ZeroMemset' is true, that
+    /// means it's a memset of zero. 'MemcpyStrSrc' indicates whether the memcpy
+    /// source is constant so it does not need to be loaded.
     /// It returns EVT::Other if the type should be determined using generic
     /// target-independent logic.
     virtual EVT
-    getOptimalMemOpType(uint64_t Size, unsigned DstAlign, unsigned SrcAlign,
-                        bool IsZeroVal, bool MemcpyStrSrc,
+    getOptimalMemOpType(uint64_t Size, unsigned DstAlign, unsigned SrcAlign, 
+                        bool IsMemset, bool ZeroMemset, bool MemcpyStrSrc,
                         MachineFunction &MF) const;
 
+    /// Is unaligned memory access allowed for the given type, and is it fast
+    /// relative to software emulation.
+    virtual bool allowsUnalignedMemoryAccesses(EVT VT, bool *Fast = 0) const;
+
     /// isFMAFasterThanMulAndAdd - Return true if an FMA operation is faster than
     /// a pair of mul and add instructions. fmuladd intrinsics will be expanded to
     /// FMAs when this method returns true (and FMAs are legal), otherwise fmuladd
@@ -415,7 +510,7 @@ namespace llvm {
                                       const PPCSubtarget &Subtarget) const;
     SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG, DebugLoc dl) const;
-    SDValue LowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerSHL_PARTS(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerSRL_PARTS(SDValue Op, SelectionDAG &DAG) const;
@@ -525,6 +620,12 @@ namespace llvm {
                      const SmallVectorImpl<ISD::InputArg> &Ins,
                      DebugLoc dl, SelectionDAG &DAG,
                      SmallVectorImpl<SDValue> &InVals) const;
+
+    SDValue lowerEH_SJLJ_SETJMP(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerEH_SJLJ_LONGJMP(SDValue Op, SelectionDAG &DAG) const;
+
+    SDValue DAGCombineFastRecip(SDValue Op, DAGCombinerInfo &DCI) const;
+    SDValue DAGCombineFastRecipFSQRT(SDValue Op, DAGCombinerInfo &DCI) const;
   };
 }
 
diff --git a/lib/Target/PowerPC/PPCInstr64Bit.td b/lib/Target/PowerPC/PPCInstr64Bit.td
index 9711452ec46a..fa5b65f0ba2d 100644
--- a/lib/Target/PowerPC/PPCInstr64Bit.td
+++ b/lib/Target/PowerPC/PPCInstr64Bit.td
@@ -30,8 +30,12 @@ def symbolLo64 : Operand<i64> {
   let EncoderMethod = "getLO16Encoding";
 }
 def tocentry : Operand<iPTR> {
-  let MIOperandInfo = (ops i32imm:$imm);
+  let MIOperandInfo = (ops i64imm:$imm);
 }
+def tlsreg : Operand<i64> {
+  let EncoderMethod = "getTLSRegEncoding";
+}
+def tlsgd : Operand<i64> {}
 
 //===----------------------------------------------------------------------===//
 // 64-bit transformation functions.
@@ -62,123 +66,112 @@ def HI48_64 : SDNodeXForm<imm, [{
 // Calls.
 //
 
+let isTerminator = 1, isBarrier = 1, PPC970_Unit = 7 in {
+  let isBranch = 1, isIndirectBranch = 1, Uses = [CTR8] in
+    def BCTR8 : XLForm_2_ext<19, 528, 20, 0, 0, (outs), (ins), "bctr", BrB, []>,
+        Requires<[In64BitMode]>;
+}
+
 let Defs = [LR8] in
   def MovePCtoLR8 : Pseudo<(outs), (ins), "#MovePCtoLR8", []>,
                     PPC970_Unit_BRU;
 
-// Darwin ABI Calls.
-let isCall = 1, PPC970_Unit = 7, Defs = [LR8] in {
-  // Convenient aliases for call instructions
-  let Uses = [RM] in {
-    def BL8_Darwin  : IForm<18, 0, 1,
-                            (outs), (ins calltarget:$func),
-                            "bl $func", BrB, []>;  // See Pat patterns below.
-    def BLA8_Darwin : IForm<18, 1, 1,
-                          (outs), (ins aaddr:$func),
-                          "bla $func", BrB, [(PPCcall_Darwin (i64 imm:$func))]>;
-  }
-  let Uses = [CTR8, RM] in {
-    def BCTRL8_Darwin : XLForm_2_ext<19, 528, 20, 0, 1, 
-                                  (outs), (ins),
-                                  "bctrl", BrB,
-                                  [(PPCbctrl_Darwin)]>, Requires<[In64BitMode]>;
+let isBranch = 1, isTerminator = 1, hasCtrlDep = 1, PPC970_Unit = 7 in {
+  let Defs = [CTR8], Uses = [CTR8] in {
+    def BDZ8  : BForm_1<16, 18, 0, 0, (outs), (ins condbrtarget:$dst),
+                        "bdz $dst">;
+    def BDNZ8 : BForm_1<16, 16, 0, 0, (outs), (ins condbrtarget:$dst),
+                        "bdnz $dst">;
   }
 }
 
-// ELF 64 ABI Calls = Darwin ABI Calls
-// Used to define BL8_ELF and BLA8_ELF
 let isCall = 1, PPC970_Unit = 7, Defs = [LR8] in {
   // Convenient aliases for call instructions
   let Uses = [RM] in {
-    def BL8_ELF  : IForm<18, 0, 1,
-                         (outs), (ins calltarget:$func),
-                         "bl $func", BrB, []>;  // See Pat patterns below.
+    def BL8  : IForm<18, 0, 1, (outs), (ins calltarget:$func),
+                     "bl $func", BrB, []>;  // See Pat patterns below.
 
-    let isCodeGenOnly = 1 in
-    def BL8_NOP_ELF  : IForm_and_DForm_4_zero<18, 0, 1, 24,
+    def BLA8 : IForm<18, 1, 1, (outs), (ins aaddr:$func),
+                     "bla $func", BrB, [(PPCcall (i64 imm:$func))]>;
+  }
+  let Uses = [RM], isCodeGenOnly = 1 in {
+    def BL8_NOP  : IForm_and_DForm_4_zero<18, 0, 1, 24,
                              (outs), (ins calltarget:$func),
                              "bl $func\n\tnop", BrB, []>;
 
-    def BLA8_ELF : IForm<18, 1, 1,
-                         (outs), (ins aaddr:$func),
-                         "bla $func", BrB, [(PPCcall_SVR4 (i64 imm:$func))]>;
+    def BL8_NOP_TLSGD : IForm_and_DForm_4_zero<18, 0, 1, 24,
+                                  (outs), (ins calltarget:$func, tlsgd:$sym),
+                                  "bl $func($sym)\n\tnop", BrB, []>;
 
-    let isCodeGenOnly = 1 in
-    def BLA8_NOP_ELF : IForm_and_DForm_4_zero<18, 1, 1, 24,
+    def BL8_NOP_TLSLD : IForm_and_DForm_4_zero<18, 0, 1, 24,
+                                  (outs), (ins calltarget:$func, tlsgd:$sym),
+                                  "bl $func($sym)\n\tnop", BrB, []>;
+
+    def BLA8_NOP : IForm_and_DForm_4_zero<18, 1, 1, 24,
                              (outs), (ins aaddr:$func),
                              "bla $func\n\tnop", BrB,
-                             [(PPCcall_nop_SVR4 (i64 imm:$func))]>;
+                             [(PPCcall_nop (i64 imm:$func))]>;
   }
-  let Uses = [X11, CTR8, RM] in {
-    def BCTRL8_ELF : XLForm_2_ext<19, 528, 20, 0, 1,
-                               (outs), (ins),
-                               "bctrl", BrB,
-                               [(PPCbctrl_SVR4)]>, Requires<[In64BitMode]>;
+  let Uses = [CTR8, RM] in {
+    def BCTRL8 : XLForm_2_ext<19, 528, 20, 0, 1, (outs), (ins),
+                              "bctrl", BrB, [(PPCbctrl)]>,
+                 Requires<[In64BitMode]>;
   }
 }
 
 
 // Calls
-def : Pat<(PPCcall_Darwin (i64 tglobaladdr:$dst)),
-          (BL8_Darwin tglobaladdr:$dst)>;
-def : Pat<(PPCcall_Darwin (i64 texternalsym:$dst)),
-          (BL8_Darwin texternalsym:$dst)>;
+def : Pat<(PPCcall (i64 tglobaladdr:$dst)),
+          (BL8 tglobaladdr:$dst)>;
+def : Pat<(PPCcall_nop (i64 tglobaladdr:$dst)),
+          (BL8_NOP tglobaladdr:$dst)>;
 
-def : Pat<(PPCcall_SVR4 (i64 tglobaladdr:$dst)),
-          (BL8_ELF tglobaladdr:$dst)>;
-def : Pat<(PPCcall_nop_SVR4 (i64 tglobaladdr:$dst)),
-          (BL8_NOP_ELF tglobaladdr:$dst)>;
-
-def : Pat<(PPCcall_SVR4 (i64 texternalsym:$dst)),
-          (BL8_ELF texternalsym:$dst)>;
-def : Pat<(PPCcall_nop_SVR4 (i64 texternalsym:$dst)),
-          (BL8_NOP_ELF texternalsym:$dst)>;
-
-def : Pat<(PPCnop),
-          (NOP)>;
+def : Pat<(PPCcall (i64 texternalsym:$dst)),
+          (BL8 texternalsym:$dst)>;
+def : Pat<(PPCcall_nop (i64 texternalsym:$dst)),
+          (BL8_NOP texternalsym:$dst)>;
 
 // Atomic operations
 let usesCustomInserter = 1 in {
   let Defs = [CR0] in {
     def ATOMIC_LOAD_ADD_I64 : Pseudo<
       (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$incr), "#ATOMIC_LOAD_ADD_I64",
-      [(set G8RC:$dst, (atomic_load_add_64 xoaddr:$ptr, G8RC:$incr))]>;
+      [(set i64:$dst, (atomic_load_add_64 xoaddr:$ptr, i64:$incr))]>;
     def ATOMIC_LOAD_SUB_I64 : Pseudo<
       (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$incr), "#ATOMIC_LOAD_SUB_I64",
-      [(set G8RC:$dst, (atomic_load_sub_64 xoaddr:$ptr, G8RC:$incr))]>;
+      [(set i64:$dst, (atomic_load_sub_64 xoaddr:$ptr, i64:$incr))]>;
     def ATOMIC_LOAD_OR_I64 : Pseudo<
       (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$incr), "#ATOMIC_LOAD_OR_I64",
-      [(set G8RC:$dst, (atomic_load_or_64 xoaddr:$ptr, G8RC:$incr))]>;
+      [(set i64:$dst, (atomic_load_or_64 xoaddr:$ptr, i64:$incr))]>;
     def ATOMIC_LOAD_XOR_I64 : Pseudo<
       (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$incr), "#ATOMIC_LOAD_XOR_I64",
-      [(set G8RC:$dst, (atomic_load_xor_64 xoaddr:$ptr, G8RC:$incr))]>;
+      [(set i64:$dst, (atomic_load_xor_64 xoaddr:$ptr, i64:$incr))]>;
     def ATOMIC_LOAD_AND_I64 : Pseudo<
       (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$incr), "#ATOMIC_LOAD_AND_i64",
-      [(set G8RC:$dst, (atomic_load_and_64 xoaddr:$ptr, G8RC:$incr))]>;
+      [(set i64:$dst, (atomic_load_and_64 xoaddr:$ptr, i64:$incr))]>;
     def ATOMIC_LOAD_NAND_I64 : Pseudo<
       (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$incr), "#ATOMIC_LOAD_NAND_I64",
-      [(set G8RC:$dst, (atomic_load_nand_64 xoaddr:$ptr, G8RC:$incr))]>;
+      [(set i64:$dst, (atomic_load_nand_64 xoaddr:$ptr, i64:$incr))]>;
 
     def ATOMIC_CMP_SWAP_I64 : Pseudo<
       (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$old, G8RC:$new), "#ATOMIC_CMP_SWAP_I64",
-      [(set G8RC:$dst, 
-                    (atomic_cmp_swap_64 xoaddr:$ptr, G8RC:$old, G8RC:$new))]>;
+      [(set i64:$dst, (atomic_cmp_swap_64 xoaddr:$ptr, i64:$old, i64:$new))]>;
 
     def ATOMIC_SWAP_I64 : Pseudo<
       (outs G8RC:$dst), (ins memrr:$ptr, G8RC:$new), "#ATOMIC_SWAP_I64",
-      [(set G8RC:$dst, (atomic_swap_64 xoaddr:$ptr, G8RC:$new))]>;
+      [(set i64:$dst, (atomic_swap_64 xoaddr:$ptr, i64:$new))]>;
   }
 }
 
 // Instructions to support atomic operations
 def LDARX : XForm_1<31,  84, (outs G8RC:$rD), (ins memrr:$ptr),
                    "ldarx $rD, $ptr", LdStLDARX,
-                   [(set G8RC:$rD, (PPClarx xoaddr:$ptr))]>;
+                   [(set i64:$rD, (PPClarx xoaddr:$ptr))]>;
 
 let Defs = [CR0] in
 def STDCX : XForm_1<31, 214, (outs), (ins G8RC:$rS, memrr:$dst),
                    "stdcx. $rS, $dst", LdStSTDCX,
-                   [(PPCstcx G8RC:$rS, xoaddr:$dst)]>,
+                   [(PPCstcx i64:$rS, xoaddr:$dst)]>,
                    isDOT;
 
 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [RM] in
@@ -197,17 +190,12 @@ def TCRETURNri8 : Pseudo<(outs), (ins CTRRC8:$dst, i32imm:$offset),
                  "#TC_RETURNr8 $dst $offset",
                  []>;
 
+let isCodeGenOnly = 1 in {
 
 let isTerminator = 1, isBarrier = 1, PPC970_Unit = 7, isBranch = 1,
-    isIndirectBranch = 1, isCall = 1, Uses = [CTR8, RM] in {
-  let isReturn = 1 in {
-    def TAILBCTR8 : XLForm_2_ext<19, 528, 20, 0, 0, (outs), (ins), "bctr", BrB, []>,
-        Requires<[In64BitMode]>;
-  }
-
-  def BCTR8 : XLForm_2_ext<19, 528, 20, 0, 0, (outs), (ins), "bctr", BrB, []>,
-      Requires<[In64BitMode]>;
-}
+    isIndirectBranch = 1, isCall = 1, isReturn = 1, Uses = [CTR8, RM] in
+def TAILBCTR8 : XLForm_2_ext<19, 528, 20, 0, 0, (outs), (ins), "bctr", BrB, []>,
+    Requires<[In64BitMode]>;
 
 
 let isBranch = 1, isTerminator = 1, hasCtrlDep = 1, PPC970_Unit = 7,
@@ -223,6 +211,8 @@ def TAILBA8   : IForm<18, 0, 0, (outs), (ins aaddr:$dst),
                   "ba $dst", BrB,
                   []>;
 
+}
+
 def : Pat<(PPCtc_return (i64 tglobaladdr:$dst),  imm:$imm),
           (TCRETURNdi8 tglobaladdr:$dst, imm:$imm)>;
 
@@ -232,20 +222,13 @@ def : Pat<(PPCtc_return (i64 texternalsym:$dst), imm:$imm),
 def : Pat<(PPCtc_return CTRRC8:$dst, imm:$imm),
           (TCRETURNri8 CTRRC8:$dst, imm:$imm)>;
 
-let isBranch = 1, isTerminator = 1, hasCtrlDep = 1, PPC970_Unit = 7 in {
-  let Defs = [CTR8], Uses = [CTR8] in {
-    def BDZ8  : BForm_1<16, 18, 0, 0, (outs), (ins condbrtarget:$dst),
-                        "bdz $dst">;
-    def BDNZ8 : BForm_1<16, 16, 0, 0, (outs), (ins condbrtarget:$dst),
-                        "bdnz $dst">;
-  }
-}
 
-// 64-but CR instructions
+// 64-bit CR instructions
 def MTCRF8 : XFXForm_5<31, 144, (outs crbitm:$FXM), (ins G8RC:$rS),
                       "mtcrf $FXM, $rS", BrMCRX>,
             PPC970_MicroCode, PPC970_Unit_CRU;
 
+let isCodeGenOnly = 1 in
 def MFCR8pseud: XFXForm_3<31, 19, (outs G8RC:$rT), (ins crbitm:$FXM),
                        "#MFCR8pseud", SprMFCR>,
             PPC970_MicroCode, PPC970_Unit_CRU;
@@ -254,6 +237,18 @@ def MFCR8 : XFXForm_3<31, 19, (outs G8RC:$rT), (ins),
                      "mfcr $rT", SprMFCR>,
                      PPC970_MicroCode, PPC970_Unit_CRU;
 
+let hasSideEffects = 1, isBarrier = 1, usesCustomInserter = 1 in {
+  def EH_SjLj_SetJmp64  : Pseudo<(outs GPRC:$dst), (ins memr:$buf),
+                            "#EH_SJLJ_SETJMP64",
+                            [(set i32:$dst, (PPCeh_sjlj_setjmp addr:$buf))]>,
+                          Requires<[In64BitMode]>;
+  let isTerminator = 1 in
+  def EH_SjLj_LongJmp64 : Pseudo<(outs), (ins memr:$buf),
+                            "#EH_SJLJ_LONGJMP64",
+                            [(PPCeh_sjlj_longjmp addr:$buf)]>,
+                          Requires<[In64BitMode]>;
+}
+
 //===----------------------------------------------------------------------===//
 // 64-bit SPR manipulation instrs.
 
@@ -262,13 +257,13 @@ def MFCTR8 : XFXForm_1_ext<31, 339, 9, (outs G8RC:$rT), (ins),
                            "mfctr $rT", SprMFSPR>,
              PPC970_DGroup_First, PPC970_Unit_FXU;
 }
-let Pattern = [(PPCmtctr G8RC:$rS)], Defs = [CTR8] in {
+let Pattern = [(PPCmtctr i64:$rS)], Defs = [CTR8] in {
 def MTCTR8 : XFXForm_7_ext<31, 467, 9, (outs), (ins G8RC:$rS),
                            "mtctr $rS", SprMTSPR>,
              PPC970_DGroup_First, PPC970_Unit_FXU;
 }
 
-let Pattern = [(set G8RC:$rT, readcyclecounter)] in
+let Pattern = [(set i64:$rT, readcyclecounter)] in
 def MFTB8 : XFXForm_1_ext<31, 339, 268, (outs G8RC:$rT), (ins),
                           "mfspr $rT, 268", SprMFTB>,
             PPC970_DGroup_First, PPC970_Unit_FXU;
@@ -279,8 +274,8 @@ def MFTB8 : XFXForm_1_ext<31, 339, 268, (outs G8RC:$rT), (ins),
 
 let Defs = [X1], Uses = [X1] in
 def DYNALLOC8 : Pseudo<(outs G8RC:$result), (ins G8RC:$negsize, memri:$fpsi),"#DYNALLOC8",
-                       [(set G8RC:$result,
-                             (PPCdynalloc G8RC:$negsize, iaddr:$fpsi))]>;
+                       [(set i64:$result,
+                             (PPCdynalloc i64:$negsize, iaddr:$fpsi))]>;
 
 let Defs = [LR8] in {
 def MTLR8  : XFXForm_7_ext<31, 467, 8, (outs), (ins G8RC:$rS),
@@ -302,126 +297,129 @@ let PPC970_Unit = 1 in {  // FXU Operations.
 let isReMaterializable = 1, isAsCheapAsAMove = 1, isMoveImm = 1 in {
 def LI8  : DForm_2_r0<14, (outs G8RC:$rD), (ins symbolLo64:$imm),
                       "li $rD, $imm", IntSimple,
-                      [(set G8RC:$rD, immSExt16:$imm)]>;
+                      [(set i64:$rD, immSExt16:$imm)]>;
 def LIS8 : DForm_2_r0<15, (outs G8RC:$rD), (ins symbolHi64:$imm),
                       "lis $rD, $imm", IntSimple,
-                      [(set G8RC:$rD, imm16ShiftedSExt:$imm)]>;
+                      [(set i64:$rD, imm16ShiftedSExt:$imm)]>;
 }
 
 // Logical ops.
 def NAND8: XForm_6<31, 476, (outs G8RC:$rA), (ins G8RC:$rS, G8RC:$rB),
                    "nand $rA, $rS, $rB", IntSimple,
-                   [(set G8RC:$rA, (not (and G8RC:$rS, G8RC:$rB)))]>;
+                   [(set i64:$rA, (not (and i64:$rS, i64:$rB)))]>;
 def AND8 : XForm_6<31,  28, (outs G8RC:$rA), (ins G8RC:$rS, G8RC:$rB),
                    "and $rA, $rS, $rB", IntSimple,
-                   [(set G8RC:$rA, (and G8RC:$rS, G8RC:$rB))]>;
+                   [(set i64:$rA, (and i64:$rS, i64:$rB))]>;
 def ANDC8: XForm_6<31,  60, (outs G8RC:$rA), (ins G8RC:$rS, G8RC:$rB),
                    "andc $rA, $rS, $rB", IntSimple,
-                   [(set G8RC:$rA, (and G8RC:$rS, (not G8RC:$rB)))]>;
+                   [(set i64:$rA, (and i64:$rS, (not i64:$rB)))]>;
 def OR8  : XForm_6<31, 444, (outs G8RC:$rA), (ins G8RC:$rS, G8RC:$rB),
                    "or $rA, $rS, $rB", IntSimple,
-                   [(set G8RC:$rA, (or G8RC:$rS, G8RC:$rB))]>;
+                   [(set i64:$rA, (or i64:$rS, i64:$rB))]>;
 def NOR8 : XForm_6<31, 124, (outs G8RC:$rA), (ins G8RC:$rS, G8RC:$rB),
                    "nor $rA, $rS, $rB", IntSimple,
-                   [(set G8RC:$rA, (not (or G8RC:$rS, G8RC:$rB)))]>;
+                   [(set i64:$rA, (not (or i64:$rS, i64:$rB)))]>;
 def ORC8 : XForm_6<31, 412, (outs G8RC:$rA), (ins G8RC:$rS, G8RC:$rB),
                    "orc $rA, $rS, $rB", IntSimple,
-                   [(set G8RC:$rA, (or G8RC:$rS, (not G8RC:$rB)))]>;
+                   [(set i64:$rA, (or i64:$rS, (not i64:$rB)))]>;
 def EQV8 : XForm_6<31, 284, (outs G8RC:$rA), (ins G8RC:$rS, G8RC:$rB),
                    "eqv $rA, $rS, $rB", IntSimple,
-                   [(set G8RC:$rA, (not (xor G8RC:$rS, G8RC:$rB)))]>;
+                   [(set i64:$rA, (not (xor i64:$rS, i64:$rB)))]>;
 def XOR8 : XForm_6<31, 316, (outs G8RC:$rA), (ins G8RC:$rS, G8RC:$rB),
                    "xor $rA, $rS, $rB", IntSimple,
-                   [(set G8RC:$rA, (xor G8RC:$rS, G8RC:$rB))]>;
+                   [(set i64:$rA, (xor i64:$rS, i64:$rB))]>;
 
 // Logical ops with immediate.
 def ANDIo8  : DForm_4<28, (outs G8RC:$dst), (ins G8RC:$src1, u16imm:$src2),
                       "andi. $dst, $src1, $src2", IntGeneral,
-                      [(set G8RC:$dst, (and G8RC:$src1, immZExt16:$src2))]>,
+                      [(set i64:$dst, (and i64:$src1, immZExt16:$src2))]>,
                       isDOT;
 def ANDISo8 : DForm_4<29, (outs G8RC:$dst), (ins G8RC:$src1, u16imm:$src2),
                      "andis. $dst, $src1, $src2", IntGeneral,
-                    [(set G8RC:$dst, (and G8RC:$src1,imm16ShiftedZExt:$src2))]>,
+                    [(set i64:$dst, (and i64:$src1, imm16ShiftedZExt:$src2))]>,
                      isDOT;
 def ORI8    : DForm_4<24, (outs G8RC:$dst), (ins G8RC:$src1, u16imm:$src2),
                       "ori $dst, $src1, $src2", IntSimple,
-                      [(set G8RC:$dst, (or G8RC:$src1, immZExt16:$src2))]>;
+                      [(set i64:$dst, (or i64:$src1, immZExt16:$src2))]>;
 def ORIS8   : DForm_4<25, (outs G8RC:$dst), (ins G8RC:$src1, u16imm:$src2),
                       "oris $dst, $src1, $src2", IntSimple,
-                    [(set G8RC:$dst, (or G8RC:$src1, imm16ShiftedZExt:$src2))]>;
+                    [(set i64:$dst, (or i64:$src1, imm16ShiftedZExt:$src2))]>;
 def XORI8   : DForm_4<26, (outs G8RC:$dst), (ins G8RC:$src1, u16imm:$src2),
                       "xori $dst, $src1, $src2", IntSimple,
-                      [(set G8RC:$dst, (xor G8RC:$src1, immZExt16:$src2))]>;
+                      [(set i64:$dst, (xor i64:$src1, immZExt16:$src2))]>;
 def XORIS8  : DForm_4<27, (outs G8RC:$dst), (ins G8RC:$src1, u16imm:$src2),
                       "xoris $dst, $src1, $src2", IntSimple,
-                   [(set G8RC:$dst, (xor G8RC:$src1, imm16ShiftedZExt:$src2))]>;
+                   [(set i64:$dst, (xor i64:$src1, imm16ShiftedZExt:$src2))]>;
 
 def ADD8  : XOForm_1<31, 266, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                      "add $rT, $rA, $rB", IntSimple,
-                     [(set G8RC:$rT, (add G8RC:$rA, G8RC:$rB))]>;
+                     [(set i64:$rT, (add i64:$rA, i64:$rB))]>;
+// ADD8 has a special form: reg = ADD8(reg, sym@tls) for use by the
+// initial-exec thread-local storage model.
+let isCodeGenOnly = 1 in
+def ADD8TLS  : XOForm_1<31, 266, 0, (outs G8RC:$rT), (ins G8RC:$rA, tlsreg:$rB),
+                        "add $rT, $rA, $rB@tls", IntSimple,
+                        [(set i64:$rT, (add i64:$rA, tglobaltlsaddr:$rB))]>;
                      
 let Defs = [CARRY] in {
 def ADDC8 : XOForm_1<31, 10, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                      "addc $rT, $rA, $rB", IntGeneral,
-                     [(set G8RC:$rT, (addc G8RC:$rA, G8RC:$rB))]>,
+                     [(set i64:$rT, (addc i64:$rA, i64:$rB))]>,
                      PPC970_DGroup_Cracked;
 def ADDIC8 : DForm_2<12, (outs G8RC:$rD), (ins G8RC:$rA, s16imm64:$imm),
                      "addic $rD, $rA, $imm", IntGeneral,
-                     [(set G8RC:$rD, (addc G8RC:$rA, immSExt16:$imm))]>;
+                     [(set i64:$rD, (addc i64:$rA, immSExt16:$imm))]>;
 }
-def ADDI8  : DForm_2<14, (outs G8RC:$rD), (ins G8RC:$rA, s16imm64:$imm),
-                     "addi $rD, $rA, $imm", IntSimple,
-                     [(set G8RC:$rD, (add G8RC:$rA, immSExt16:$imm))]>;
-def ADDI8L  : DForm_2<14, (outs G8RC:$rD), (ins G8RC:$rA, symbolLo64:$imm),
+def ADDI8  : DForm_2<14, (outs G8RC:$rD), (ins G8RC_NOX0:$rA, symbolLo64:$imm),
                      "addi $rD, $rA, $imm", IntSimple,
-                     [(set G8RC:$rD, (add G8RC:$rA, immSExt16:$imm))]>;
-def ADDIS8 : DForm_2<15, (outs G8RC:$rD), (ins G8RC:$rA, symbolHi64:$imm),
+                     [(set i64:$rD, (add i64:$rA, immSExt16:$imm))]>;
+def ADDIS8 : DForm_2<15, (outs G8RC:$rD), (ins G8RC_NOX0:$rA, symbolHi64:$imm),
                      "addis $rD, $rA, $imm", IntSimple,
-                     [(set G8RC:$rD, (add G8RC:$rA, imm16ShiftedSExt:$imm))]>;
+                     [(set i64:$rD, (add i64:$rA, imm16ShiftedSExt:$imm))]>;
 
 let Defs = [CARRY] in {
 def SUBFIC8: DForm_2< 8, (outs G8RC:$rD), (ins G8RC:$rA, s16imm64:$imm),
                      "subfic $rD, $rA, $imm", IntGeneral,
-                     [(set G8RC:$rD, (subc immSExt16:$imm, G8RC:$rA))]>;
+                     [(set i64:$rD, (subc immSExt16:$imm, i64:$rA))]>;
 def SUBFC8 : XOForm_1<31, 8, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                       "subfc $rT, $rA, $rB", IntGeneral,
-                      [(set G8RC:$rT, (subc G8RC:$rB, G8RC:$rA))]>,
+                      [(set i64:$rT, (subc i64:$rB, i64:$rA))]>,
                       PPC970_DGroup_Cracked;
 }
 def SUBF8 : XOForm_1<31, 40, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                      "subf $rT, $rA, $rB", IntGeneral,
-                     [(set G8RC:$rT, (sub G8RC:$rB, G8RC:$rA))]>;
+                     [(set i64:$rT, (sub i64:$rB, i64:$rA))]>;
 def NEG8    : XOForm_3<31, 104, 0, (outs G8RC:$rT), (ins G8RC:$rA),
                        "neg $rT, $rA", IntSimple,
-                       [(set G8RC:$rT, (ineg G8RC:$rA))]>;
+                       [(set i64:$rT, (ineg i64:$rA))]>;
 let Uses = [CARRY], Defs = [CARRY] in {
 def ADDE8   : XOForm_1<31, 138, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                        "adde $rT, $rA, $rB", IntGeneral,
-                       [(set G8RC:$rT, (adde G8RC:$rA, G8RC:$rB))]>;
+                       [(set i64:$rT, (adde i64:$rA, i64:$rB))]>;
 def ADDME8  : XOForm_3<31, 234, 0, (outs G8RC:$rT), (ins G8RC:$rA),
                        "addme $rT, $rA", IntGeneral,
-                       [(set G8RC:$rT, (adde G8RC:$rA, -1))]>;
+                       [(set i64:$rT, (adde i64:$rA, -1))]>;
 def ADDZE8  : XOForm_3<31, 202, 0, (outs G8RC:$rT), (ins G8RC:$rA),
                        "addze $rT, $rA", IntGeneral,
-                       [(set G8RC:$rT, (adde G8RC:$rA, 0))]>;
+                       [(set i64:$rT, (adde i64:$rA, 0))]>;
 def SUBFE8  : XOForm_1<31, 136, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                        "subfe $rT, $rA, $rB", IntGeneral,
-                       [(set G8RC:$rT, (sube G8RC:$rB, G8RC:$rA))]>;
+                       [(set i64:$rT, (sube i64:$rB, i64:$rA))]>;
 def SUBFME8 : XOForm_3<31, 232, 0, (outs G8RC:$rT), (ins G8RC:$rA),
                        "subfme $rT, $rA", IntGeneral,
-                       [(set G8RC:$rT, (sube -1, G8RC:$rA))]>;
+                       [(set i64:$rT, (sube -1, i64:$rA))]>;
 def SUBFZE8 : XOForm_3<31, 200, 0, (outs G8RC:$rT), (ins G8RC:$rA),
                        "subfze $rT, $rA", IntGeneral,
-                       [(set G8RC:$rT, (sube 0, G8RC:$rA))]>;
+                       [(set i64:$rT, (sube 0, i64:$rA))]>;
 }
 
 
 def MULHD : XOForm_1<31, 73, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                      "mulhd $rT, $rA, $rB", IntMulHW,
-                     [(set G8RC:$rT, (mulhs G8RC:$rA, G8RC:$rB))]>;
+                     [(set i64:$rT, (mulhs i64:$rA, i64:$rB))]>;
 def MULHDU : XOForm_1<31, 9, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                      "mulhdu $rT, $rA, $rB", IntMulHWU,
-                     [(set G8RC:$rT, (mulhu G8RC:$rA, G8RC:$rB))]>;
+                     [(set i64:$rT, (mulhu i64:$rA, i64:$rB))]>;
 
 def CMPD   : XForm_16_ext<31, 0, (outs CRRC:$crD), (ins G8RC:$rA, G8RC:$rB),
                           "cmpd $crD, $rA, $rB", IntCompare>, isPPC64;
@@ -434,54 +432,60 @@ def CMPLDI : DForm_6_ext<10, (outs CRRC:$dst), (ins G8RC:$src1, u16imm:$src2),
 
 def SLD  : XForm_6<31,  27, (outs G8RC:$rA), (ins G8RC:$rS, GPRC:$rB),
                    "sld $rA, $rS, $rB", IntRotateD,
-                   [(set G8RC:$rA, (PPCshl G8RC:$rS, GPRC:$rB))]>, isPPC64;
+                   [(set i64:$rA, (PPCshl i64:$rS, i32:$rB))]>, isPPC64;
 def SRD  : XForm_6<31, 539, (outs G8RC:$rA), (ins G8RC:$rS, GPRC:$rB),
                    "srd $rA, $rS, $rB", IntRotateD,
-                   [(set G8RC:$rA, (PPCsrl G8RC:$rS, GPRC:$rB))]>, isPPC64;
+                   [(set i64:$rA, (PPCsrl i64:$rS, i32:$rB))]>, isPPC64;
 let Defs = [CARRY] in {
 def SRAD : XForm_6<31, 794, (outs G8RC:$rA), (ins G8RC:$rS, GPRC:$rB),
                    "srad $rA, $rS, $rB", IntRotateD,
-                   [(set G8RC:$rA, (PPCsra G8RC:$rS, GPRC:$rB))]>, isPPC64;
+                   [(set i64:$rA, (PPCsra i64:$rS, i32:$rB))]>, isPPC64;
 }
                    
 def EXTSB8 : XForm_11<31, 954, (outs G8RC:$rA), (ins G8RC:$rS),
                       "extsb $rA, $rS", IntSimple,
-                      [(set G8RC:$rA, (sext_inreg G8RC:$rS, i8))]>;
+                      [(set i64:$rA, (sext_inreg i64:$rS, i8))]>;
 def EXTSH8 : XForm_11<31, 922, (outs G8RC:$rA), (ins G8RC:$rS),
                       "extsh $rA, $rS", IntSimple,
-                      [(set G8RC:$rA, (sext_inreg G8RC:$rS, i16))]>;
+                      [(set i64:$rA, (sext_inreg i64:$rS, i16))]>;
 
 def EXTSW  : XForm_11<31, 986, (outs G8RC:$rA), (ins G8RC:$rS),
                       "extsw $rA, $rS", IntSimple,
-                      [(set G8RC:$rA, (sext_inreg G8RC:$rS, i32))]>, isPPC64;
-/// EXTSW_32 - Just like EXTSW, but works on '32-bit' registers.
-def EXTSW_32 : XForm_11<31, 986, (outs GPRC:$rA), (ins GPRC:$rS),
-                      "extsw $rA, $rS", IntSimple,
-                      [(set GPRC:$rA, (PPCextsw_32 GPRC:$rS))]>, isPPC64;
+                      [(set i64:$rA, (sext_inreg i64:$rS, i32))]>, isPPC64;
 def EXTSW_32_64 : XForm_11<31, 986, (outs G8RC:$rA), (ins GPRC:$rS),
                       "extsw $rA, $rS", IntSimple,
-                      [(set G8RC:$rA, (sext GPRC:$rS))]>, isPPC64;
+                      [(set i64:$rA, (sext i32:$rS))]>, isPPC64;
 
 let Defs = [CARRY] in {
 def SRADI  : XSForm_1<31, 413, (outs G8RC:$rA), (ins G8RC:$rS, u6imm:$SH),
                       "sradi $rA, $rS, $SH", IntRotateDI,
-                      [(set G8RC:$rA, (sra G8RC:$rS, (i32 imm:$SH)))]>, isPPC64;
+                      [(set i64:$rA, (sra i64:$rS, (i32 imm:$SH)))]>, isPPC64;
 }
 def CNTLZD : XForm_11<31, 58, (outs G8RC:$rA), (ins G8RC:$rS),
                       "cntlzd $rA, $rS", IntGeneral,
-                      [(set G8RC:$rA, (ctlz G8RC:$rS))]>;
+                      [(set i64:$rA, (ctlz i64:$rS))]>;
+def POPCNTD : XForm_11<31, 506, (outs G8RC:$rA), (ins G8RC:$rS),
+                      "popcntd $rA, $rS", IntGeneral,
+                      [(set i64:$rA, (ctpop i64:$rS))]>;
+
+// popcntw also does a population count on the high 32 bits (storing the
+// results in the high 32-bits of the output). We'll ignore that here (which is
+// safe because we never separately use the high part of the 64-bit registers).
+def POPCNTW : XForm_11<31, 378, (outs GPRC:$rA), (ins GPRC:$rS),
+                      "popcntw $rA, $rS", IntGeneral,
+                      [(set i32:$rA, (ctpop i32:$rS))]>;
 
 def DIVD  : XOForm_1<31, 489, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                      "divd $rT, $rA, $rB", IntDivD,
-                     [(set G8RC:$rT, (sdiv G8RC:$rA, G8RC:$rB))]>, isPPC64,
+                     [(set i64:$rT, (sdiv i64:$rA, i64:$rB))]>, isPPC64,
                      PPC970_DGroup_First, PPC970_DGroup_Cracked;
 def DIVDU : XOForm_1<31, 457, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                      "divdu $rT, $rA, $rB", IntDivD,
-                     [(set G8RC:$rT, (udiv G8RC:$rA, G8RC:$rB))]>, isPPC64,
+                     [(set i64:$rT, (udiv i64:$rA, i64:$rB))]>, isPPC64,
                      PPC970_DGroup_First, PPC970_DGroup_Cracked;
 def MULLD : XOForm_1<31, 233, 0, (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB),
                      "mulld $rT, $rA, $rB", IntMulHD,
-                     [(set G8RC:$rT, (mul G8RC:$rA, G8RC:$rB))]>, isPPC64;
+                     [(set i64:$rT, (mul i64:$rA, i64:$rB))]>, isPPC64;
 
 
 let isCommutable = 1 in {
@@ -512,7 +516,7 @@ def RLWINM8 : MForm_2<21,
                      []>;
 
 def ISEL8   : AForm_4<31, 15,
-                     (outs G8RC:$rT), (ins G8RC:$rA, G8RC:$rB, pred:$cond),
+                     (outs G8RC:$rT), (ins G8RC_NOX0:$rA, G8RC:$rB, CRBITRC:$cond),
                      "isel $rT, $rA, $rB, $cond", IntGeneral,
                      []>;
 }  // End FXU Operations.
@@ -527,94 +531,96 @@ def ISEL8   : AForm_4<31, 15,
 let canFoldAsLoad = 1, PPC970_Unit = 2 in {
 def LHA8: DForm_1<42, (outs G8RC:$rD), (ins memri:$src),
                   "lha $rD, $src", LdStLHA,
-                  [(set G8RC:$rD, (sextloadi16 iaddr:$src))]>,
+                  [(set i64:$rD, (sextloadi16 iaddr:$src))]>,
                   PPC970_DGroup_Cracked;
 def LWA  : DSForm_1<58, 2, (outs G8RC:$rD), (ins memrix:$src),
                     "lwa $rD, $src", LdStLWA,
-                    [(set G8RC:$rD, (sextloadi32 ixaddr:$src))]>, isPPC64,
+                    [(set i64:$rD,
+                          (aligned4sextloadi32 ixaddr:$src))]>, isPPC64,
                     PPC970_DGroup_Cracked;
 def LHAX8: XForm_1<31, 343, (outs G8RC:$rD), (ins memrr:$src),
                    "lhax $rD, $src", LdStLHA,
-                   [(set G8RC:$rD, (sextloadi16 xaddr:$src))]>,
+                   [(set i64:$rD, (sextloadi16 xaddr:$src))]>,
                    PPC970_DGroup_Cracked;
 def LWAX : XForm_1<31, 341, (outs G8RC:$rD), (ins memrr:$src),
                    "lwax $rD, $src", LdStLHA,
-                   [(set G8RC:$rD, (sextloadi32 xaddr:$src))]>, isPPC64,
+                   [(set i64:$rD, (sextloadi32 xaddr:$src))]>, isPPC64,
                    PPC970_DGroup_Cracked;
 
 // Update forms.
-let mayLoad = 1 in
-def LHAU8 : DForm_1a<43, (outs G8RC:$rD, ptr_rc:$ea_result), (ins symbolLo:$disp,
-                            ptr_rc:$rA),
-                    "lhau $rD, $disp($rA)", LdStLHAU,
-                    []>, RegConstraint<"$rA = $ea_result">,
+let mayLoad = 1 in {
+def LHAU8 : DForm_1<43, (outs G8RC:$rD, ptr_rc_nor0:$ea_result),
+                    (ins memri:$addr),
+                    "lhau $rD, $addr", LdStLHAU,
+                    []>, RegConstraint<"$addr.reg = $ea_result">,
                     NoEncode<"$ea_result">;
 // NO LWAU!
 
-def LHAUX8 : XForm_1<31, 375, (outs G8RC:$rD, ptr_rc:$ea_result),
+def LHAUX8 : XForm_1<31, 375, (outs G8RC:$rD, ptr_rc_nor0:$ea_result),
                     (ins memrr:$addr),
                     "lhaux $rD, $addr", LdStLHAU,
-                    []>, RegConstraint<"$addr.offreg = $ea_result">,
+                    []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                     NoEncode<"$ea_result">;
-def LWAUX : XForm_1<31, 373, (outs G8RC:$rD, ptr_rc:$ea_result),
+def LWAUX : XForm_1<31, 373, (outs G8RC:$rD, ptr_rc_nor0:$ea_result),
                     (ins memrr:$addr),
                     "lwaux $rD, $addr", LdStLHAU,
-                    []>, RegConstraint<"$addr.offreg = $ea_result">,
+                    []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                     NoEncode<"$ea_result">, isPPC64;
 }
+}
 
 // Zero extending loads.
 let canFoldAsLoad = 1, PPC970_Unit = 2 in {
 def LBZ8 : DForm_1<34, (outs G8RC:$rD), (ins memri:$src),
                   "lbz $rD, $src", LdStLoad,
-                  [(set G8RC:$rD, (zextloadi8 iaddr:$src))]>;
+                  [(set i64:$rD, (zextloadi8 iaddr:$src))]>;
 def LHZ8 : DForm_1<40, (outs G8RC:$rD), (ins memri:$src),
                   "lhz $rD, $src", LdStLoad,
-                  [(set G8RC:$rD, (zextloadi16 iaddr:$src))]>;
+                  [(set i64:$rD, (zextloadi16 iaddr:$src))]>;
 def LWZ8 : DForm_1<32, (outs G8RC:$rD), (ins memri:$src),
                   "lwz $rD, $src", LdStLoad,
-                  [(set G8RC:$rD, (zextloadi32 iaddr:$src))]>, isPPC64;
+                  [(set i64:$rD, (zextloadi32 iaddr:$src))]>, isPPC64;
 
 def LBZX8 : XForm_1<31,  87, (outs G8RC:$rD), (ins memrr:$src),
                    "lbzx $rD, $src", LdStLoad,
-                   [(set G8RC:$rD, (zextloadi8 xaddr:$src))]>;
+                   [(set i64:$rD, (zextloadi8 xaddr:$src))]>;
 def LHZX8 : XForm_1<31, 279, (outs G8RC:$rD), (ins memrr:$src),
                    "lhzx $rD, $src", LdStLoad,
-                   [(set G8RC:$rD, (zextloadi16 xaddr:$src))]>;
+                   [(set i64:$rD, (zextloadi16 xaddr:$src))]>;
 def LWZX8 : XForm_1<31,  23, (outs G8RC:$rD), (ins memrr:$src),
                    "lwzx $rD, $src", LdStLoad,
-                   [(set G8RC:$rD, (zextloadi32 xaddr:$src))]>;
+                   [(set i64:$rD, (zextloadi32 xaddr:$src))]>;
                    
                    
 // Update forms.
 let mayLoad = 1 in {
-def LBZU8 : DForm_1<35, (outs G8RC:$rD, ptr_rc:$ea_result), (ins memri:$addr),
+def LBZU8 : DForm_1<35, (outs G8RC:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
                     "lbzu $rD, $addr", LdStLoadUpd,
                     []>, RegConstraint<"$addr.reg = $ea_result">,
                     NoEncode<"$ea_result">;
-def LHZU8 : DForm_1<41, (outs G8RC:$rD, ptr_rc:$ea_result), (ins memri:$addr),
+def LHZU8 : DForm_1<41, (outs G8RC:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
                     "lhzu $rD, $addr", LdStLoadUpd,
                     []>, RegConstraint<"$addr.reg = $ea_result">,
                     NoEncode<"$ea_result">;
-def LWZU8 : DForm_1<33, (outs G8RC:$rD, ptr_rc:$ea_result), (ins memri:$addr),
+def LWZU8 : DForm_1<33, (outs G8RC:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
                     "lwzu $rD, $addr", LdStLoadUpd,
                     []>, RegConstraint<"$addr.reg = $ea_result">,
                     NoEncode<"$ea_result">;
 
-def LBZUX8 : XForm_1<31, 119, (outs G8RC:$rD, ptr_rc:$ea_result),
+def LBZUX8 : XForm_1<31, 119, (outs G8RC:$rD, ptr_rc_nor0:$ea_result),
                    (ins memrr:$addr),
                    "lbzux $rD, $addr", LdStLoadUpd,
-                   []>, RegConstraint<"$addr.offreg = $ea_result">,
+                   []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                    NoEncode<"$ea_result">;
-def LHZUX8 : XForm_1<31, 311, (outs G8RC:$rD, ptr_rc:$ea_result),
+def LHZUX8 : XForm_1<31, 311, (outs G8RC:$rD, ptr_rc_nor0:$ea_result),
                    (ins memrr:$addr),
                    "lhzux $rD, $addr", LdStLoadUpd,
-                   []>, RegConstraint<"$addr.offreg = $ea_result">,
+                   []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                    NoEncode<"$ea_result">;
-def LWZUX8 : XForm_1<31, 55, (outs G8RC:$rD, ptr_rc:$ea_result),
+def LWZUX8 : XForm_1<31, 55, (outs G8RC:$rD, ptr_rc_nor0:$ea_result),
                    (ins memrr:$addr),
                    "lwzux $rD, $addr", LdStLoadUpd,
-                   []>, RegConstraint<"$addr.offreg = $ea_result">,
+                   []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                    NoEncode<"$ea_result">;
 }
 }
@@ -624,25 +630,28 @@ def LWZUX8 : XForm_1<31, 55, (outs G8RC:$rD, ptr_rc:$ea_result),
 let canFoldAsLoad = 1, PPC970_Unit = 2 in {
 def LD   : DSForm_1<58, 0, (outs G8RC:$rD), (ins memrix:$src),
                     "ld $rD, $src", LdStLD,
-                    [(set G8RC:$rD, (load ixaddr:$src))]>, isPPC64;
+                    [(set i64:$rD, (aligned4load ixaddr:$src))]>, isPPC64;
+// The following three definitions are selected for small code model only.
+// Otherwise, we need to create two instructions to form a 32-bit offset,
+// so we have a custom matcher for TOC_ENTRY in PPCDAGToDAGIsel::Select().
 def LDtoc: Pseudo<(outs G8RC:$rD), (ins tocentry:$disp, G8RC:$reg),
                   "#LDtoc",
-                  [(set G8RC:$rD,
-                     (PPCtoc_entry tglobaladdr:$disp, G8RC:$reg))]>, isPPC64;
+                  [(set i64:$rD,
+                     (PPCtoc_entry tglobaladdr:$disp, i64:$reg))]>, isPPC64;
 def LDtocJTI: Pseudo<(outs G8RC:$rD), (ins tocentry:$disp, G8RC:$reg),
                   "#LDtocJTI",
-                  [(set G8RC:$rD,
-                     (PPCtoc_entry tjumptable:$disp, G8RC:$reg))]>, isPPC64;
+                  [(set i64:$rD,
+                     (PPCtoc_entry tjumptable:$disp, i64:$reg))]>, isPPC64;
 def LDtocCPT: Pseudo<(outs G8RC:$rD), (ins tocentry:$disp, G8RC:$reg),
                   "#LDtocCPT",
-                  [(set G8RC:$rD,
-                     (PPCtoc_entry tconstpool:$disp, G8RC:$reg))]>, isPPC64;
+                  [(set i64:$rD,
+                     (PPCtoc_entry tconstpool:$disp, i64:$reg))]>, isPPC64;
 
-let hasSideEffects = 1 in { 
+let hasSideEffects = 1, isCodeGenOnly = 1 in {
 let RST = 2, DS = 2 in
 def LDinto_toc: DSForm_1a<58, 0, (outs), (ins G8RC:$reg),
                     "ld 2, 8($reg)", LdStLD,
-                    [(PPCload_toc G8RC:$reg)]>, isPPC64;
+                    [(PPCload_toc i64:$reg)]>, isPPC64;
                     
 let RST = 2, DS = 10, RA = 1 in
 def LDtoc_restore : DSForm_1a<58, 0, (outs), (ins),
@@ -651,18 +660,21 @@ def LDtoc_restore : DSForm_1a<58, 0, (outs), (ins),
 }
 def LDX  : XForm_1<31,  21, (outs G8RC:$rD), (ins memrr:$src),
                    "ldx $rD, $src", LdStLD,
-                   [(set G8RC:$rD, (load xaddr:$src))]>, isPPC64;
-                   
+                   [(set i64:$rD, (load xaddr:$src))]>, isPPC64;
+def LDBRX : XForm_1<31,  532, (outs G8RC:$rD), (ins memrr:$src),
+                   "ldbrx $rD, $src", LdStLoad,
+                   [(set i64:$rD, (PPClbrx xoaddr:$src, i64))]>, isPPC64;
+
 let mayLoad = 1 in
-def LDU  : DSForm_1<58, 1, (outs G8RC:$rD, ptr_rc:$ea_result), (ins memrix:$addr),
+def LDU  : DSForm_1<58, 1, (outs G8RC:$rD, ptr_rc_nor0:$ea_result), (ins memrix:$addr),
                     "ldu $rD, $addr", LdStLDU,
                     []>, RegConstraint<"$addr.reg = $ea_result">, isPPC64,
                     NoEncode<"$ea_result">;
 
-def LDUX : XForm_1<31, 53, (outs G8RC:$rD, ptr_rc:$ea_result),
+def LDUX : XForm_1<31, 53, (outs G8RC:$rD, ptr_rc_nor0:$ea_result),
                    (ins memrr:$addr),
                    "ldux $rD, $addr", LdStLDU,
-                   []>, RegConstraint<"$addr.offreg = $ea_result">,
+                   []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                    NoEncode<"$ea_result">, isPPC64;
 }
 
@@ -671,118 +683,168 @@ def : Pat<(PPCload ixaddr:$src),
 def : Pat<(PPCload xaddr:$src),
           (LDX xaddr:$src)>;
 
+// Support for medium and large code model.
+def ADDIStocHA: Pseudo<(outs G8RC:$rD), (ins G8RC_NOX0:$reg, tocentry:$disp),
+                       "#ADDIStocHA",
+                       [(set i64:$rD,
+                         (PPCaddisTocHA i64:$reg, tglobaladdr:$disp))]>,
+                       isPPC64;
+def LDtocL: Pseudo<(outs G8RC:$rD), (ins tocentry:$disp, G8RC_NOX0:$reg),
+                   "#LDtocL",
+                   [(set i64:$rD,
+                     (PPCldTocL tglobaladdr:$disp, i64:$reg))]>, isPPC64;
+def ADDItocL: Pseudo<(outs G8RC:$rD), (ins G8RC_NOX0:$reg, tocentry:$disp),
+                     "#ADDItocL",
+                     [(set i64:$rD,
+                       (PPCaddiTocL i64:$reg, tglobaladdr:$disp))]>, isPPC64;
+
+// Support for thread-local storage.
+def ADDISgotTprelHA: Pseudo<(outs G8RC:$rD), (ins G8RC_NOX0:$reg, symbolHi64:$disp),
+                         "#ADDISgotTprelHA",
+                         [(set i64:$rD,
+                           (PPCaddisGotTprelHA i64:$reg,
+                                               tglobaltlsaddr:$disp))]>,
+                  isPPC64;
+def LDgotTprelL: Pseudo<(outs G8RC:$rD), (ins symbolLo64:$disp, G8RC_NOX0:$reg),
+                        "#LDgotTprelL",
+                        [(set i64:$rD,
+                          (PPCldGotTprelL tglobaltlsaddr:$disp, i64:$reg))]>,
+                 isPPC64;
+def : Pat<(PPCaddTls i64:$in, tglobaltlsaddr:$g),
+          (ADD8TLS $in, tglobaltlsaddr:$g)>;
+def ADDIStlsgdHA: Pseudo<(outs G8RC:$rD), (ins G8RC_NOX0:$reg, symbolHi64:$disp),
+                         "#ADDIStlsgdHA",
+                         [(set i64:$rD,
+                           (PPCaddisTlsgdHA i64:$reg, tglobaltlsaddr:$disp))]>,
+                  isPPC64;
+def ADDItlsgdL : Pseudo<(outs G8RC:$rD), (ins G8RC_NOX0:$reg, symbolLo64:$disp),
+                       "#ADDItlsgdL",
+                       [(set i64:$rD,
+                         (PPCaddiTlsgdL i64:$reg, tglobaltlsaddr:$disp))]>,
+                 isPPC64;
+def GETtlsADDR : Pseudo<(outs G8RC:$rD), (ins G8RC:$reg, tlsgd:$sym),
+                        "#GETtlsADDR",
+                        [(set i64:$rD,
+                          (PPCgetTlsAddr i64:$reg, tglobaltlsaddr:$sym))]>,
+                 isPPC64;
+def ADDIStlsldHA: Pseudo<(outs G8RC:$rD), (ins G8RC_NOX0:$reg, symbolHi64:$disp),
+                         "#ADDIStlsldHA",
+                         [(set i64:$rD,
+                           (PPCaddisTlsldHA i64:$reg, tglobaltlsaddr:$disp))]>,
+                  isPPC64;
+def ADDItlsldL : Pseudo<(outs G8RC:$rD), (ins G8RC_NOX0:$reg, symbolLo64:$disp),
+                       "#ADDItlsldL",
+                       [(set i64:$rD,
+                         (PPCaddiTlsldL i64:$reg, tglobaltlsaddr:$disp))]>,
+                 isPPC64;
+def GETtlsldADDR : Pseudo<(outs G8RC:$rD), (ins G8RC:$reg, tlsgd:$sym),
+                          "#GETtlsldADDR",
+                          [(set i64:$rD,
+                            (PPCgetTlsldAddr i64:$reg, tglobaltlsaddr:$sym))]>,
+                   isPPC64;
+def ADDISdtprelHA: Pseudo<(outs G8RC:$rD), (ins G8RC_NOX0:$reg, symbolHi64:$disp),
+                          "#ADDISdtprelHA",
+                          [(set i64:$rD,
+                            (PPCaddisDtprelHA i64:$reg,
+                                              tglobaltlsaddr:$disp))]>,
+                   isPPC64;
+def ADDIdtprelL : Pseudo<(outs G8RC:$rD), (ins G8RC_NOX0:$reg, symbolLo64:$disp),
+                         "#ADDIdtprelL",
+                         [(set i64:$rD,
+                           (PPCaddiDtprelL i64:$reg, tglobaltlsaddr:$disp))]>,
+                  isPPC64;
+
 let PPC970_Unit = 2 in {
 // Truncating stores.                       
 def STB8 : DForm_1<38, (outs), (ins G8RC:$rS, memri:$src),
                    "stb $rS, $src", LdStStore,
-                   [(truncstorei8 G8RC:$rS, iaddr:$src)]>;
+                   [(truncstorei8 i64:$rS, iaddr:$src)]>;
 def STH8 : DForm_1<44, (outs), (ins G8RC:$rS, memri:$src),
                    "sth $rS, $src", LdStStore,
-                   [(truncstorei16 G8RC:$rS, iaddr:$src)]>;
+                   [(truncstorei16 i64:$rS, iaddr:$src)]>;
 def STW8 : DForm_1<36, (outs), (ins G8RC:$rS, memri:$src),
                    "stw $rS, $src", LdStStore,
-                   [(truncstorei32 G8RC:$rS, iaddr:$src)]>;
+                   [(truncstorei32 i64:$rS, iaddr:$src)]>;
 def STBX8 : XForm_8<31, 215, (outs), (ins G8RC:$rS, memrr:$dst),
                    "stbx $rS, $dst", LdStStore,
-                   [(truncstorei8 G8RC:$rS, xaddr:$dst)]>, 
+                   [(truncstorei8 i64:$rS, xaddr:$dst)]>,
                    PPC970_DGroup_Cracked;
 def STHX8 : XForm_8<31, 407, (outs), (ins G8RC:$rS, memrr:$dst),
                    "sthx $rS, $dst", LdStStore,
-                   [(truncstorei16 G8RC:$rS, xaddr:$dst)]>, 
+                   [(truncstorei16 i64:$rS, xaddr:$dst)]>,
                    PPC970_DGroup_Cracked;
 def STWX8 : XForm_8<31, 151, (outs), (ins G8RC:$rS, memrr:$dst),
                    "stwx $rS, $dst", LdStStore,
-                   [(truncstorei32 G8RC:$rS, xaddr:$dst)]>,
+                   [(truncstorei32 i64:$rS, xaddr:$dst)]>,
                    PPC970_DGroup_Cracked;
 // Normal 8-byte stores.
 def STD  : DSForm_1<62, 0, (outs), (ins G8RC:$rS, memrix:$dst),
                     "std $rS, $dst", LdStSTD,
-                    [(store G8RC:$rS, ixaddr:$dst)]>, isPPC64;
+                    [(aligned4store i64:$rS, ixaddr:$dst)]>, isPPC64;
 def STDX  : XForm_8<31, 149, (outs), (ins G8RC:$rS, memrr:$dst),
                    "stdx $rS, $dst", LdStSTD,
-                   [(store G8RC:$rS, xaddr:$dst)]>, isPPC64,
+                   [(store i64:$rS, xaddr:$dst)]>, isPPC64,
+                   PPC970_DGroup_Cracked;
+def STDBRX: XForm_8<31, 660, (outs), (ins G8RC:$rS, memrr:$dst),
+                   "stdbrx $rS, $dst", LdStStore,
+                   [(PPCstbrx i64:$rS, xoaddr:$dst, i64)]>, isPPC64,
                    PPC970_DGroup_Cracked;
 }
 
-let PPC970_Unit = 2 in {
-
-def STBU8 : DForm_1a<39, (outs ptr_rc:$ea_res), (ins G8RC:$rS,
-                             symbolLo:$ptroff, ptr_rc:$ptrreg),
-                    "stbu $rS, $ptroff($ptrreg)", LdStStoreUpd,
-                    [(set ptr_rc:$ea_res,
-                          (pre_truncsti8 G8RC:$rS, ptr_rc:$ptrreg, 
-                                         iaddroff:$ptroff))]>,
-                    RegConstraint<"$ptrreg = $ea_res">, NoEncode<"$ea_res">;
-def STHU8 : DForm_1a<45, (outs ptr_rc:$ea_res), (ins G8RC:$rS,
-                             symbolLo:$ptroff, ptr_rc:$ptrreg),
-                    "sthu $rS, $ptroff($ptrreg)", LdStStoreUpd,
-                    [(set ptr_rc:$ea_res,
-                        (pre_truncsti16 G8RC:$rS, ptr_rc:$ptrreg, 
-                                        iaddroff:$ptroff))]>,
-                    RegConstraint<"$ptrreg = $ea_res">, NoEncode<"$ea_res">;
-
-def STWU8 : DForm_1a<37, (outs ptr_rc:$ea_res), (ins G8RC:$rS,
-                             symbolLo:$ptroff, ptr_rc:$ptrreg),
-                    "stwu $rS, $ptroff($ptrreg)", LdStStoreUpd,
-                    [(set ptr_rc:$ea_res,
-                          (pre_truncsti32 G8RC:$rS, ptr_rc:$ptrreg,
-                                          iaddroff:$ptroff))]>,
-                    RegConstraint<"$ptrreg = $ea_res">, NoEncode<"$ea_res">;
-
-def STDU : DSForm_1a<62, 1, (outs ptr_rc:$ea_res), (ins G8RC:$rS,
-                                        s16immX4:$ptroff, ptr_rc:$ptrreg),
-                    "stdu $rS, $ptroff($ptrreg)", LdStSTDU,
-                    [(set ptr_rc:$ea_res, (pre_store G8RC:$rS, ptr_rc:$ptrreg, 
-                                                     iaddroff:$ptroff))]>,
-                    RegConstraint<"$ptrreg = $ea_res">, NoEncode<"$ea_res">,
-                    isPPC64;
-
-
-def STBUX8 : XForm_8<31, 247, (outs ptr_rc:$ea_res),
-                              (ins G8RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                    "stbux $rS, $ptroff, $ptrreg", LdStStoreUpd,
-                    [(set ptr_rc:$ea_res,
-                       (pre_truncsti8 G8RC:$rS,
-                                      ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
-                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+// Stores with Update (pre-inc).
+let PPC970_Unit = 2, mayStore = 1 in {
+def STBU8 : DForm_1<39, (outs ptr_rc_nor0:$ea_res), (ins G8RC:$rS, memri:$dst),
+                   "stbu $rS, $dst", LdStStoreUpd, []>,
+                   RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
+def STHU8 : DForm_1<45, (outs ptr_rc_nor0:$ea_res), (ins G8RC:$rS, memri:$dst),
+                   "sthu $rS, $dst", LdStStoreUpd, []>,
+                   RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
+def STWU8 : DForm_1<37, (outs ptr_rc_nor0:$ea_res), (ins G8RC:$rS, memri:$dst),
+                   "stwu $rS, $dst", LdStStoreUpd, []>,
+                   RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
+def STDU : DSForm_1<62, 1, (outs ptr_rc_nor0:$ea_res), (ins G8RC:$rS, memrix:$dst),
+                   "stdu $rS, $dst", LdStSTDU, []>,
+                   RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">,
+                   isPPC64;
+
+def STBUX8: XForm_8<31, 247, (outs ptr_rc_nor0:$ea_res), (ins G8RC:$rS, memrr:$dst),
+                    "stbux $rS, $dst", LdStStoreUpd, []>,
+                    RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
                     PPC970_DGroup_Cracked;
-
-def STHUX8 : XForm_8<31, 439, (outs ptr_rc:$ea_res),
-                              (ins G8RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                    "sthux $rS, $ptroff, $ptrreg", LdStStoreUpd,
-                    [(set ptr_rc:$ea_res,
-                       (pre_truncsti16 G8RC:$rS,
-                                       ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
-                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+def STHUX8: XForm_8<31, 439, (outs ptr_rc_nor0:$ea_res), (ins G8RC:$rS, memrr:$dst),
+                    "sthux $rS, $dst", LdStStoreUpd, []>,
+                    RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
                     PPC970_DGroup_Cracked;
-
-def STWUX8 : XForm_8<31, 183, (outs ptr_rc:$ea_res),
-                              (ins G8RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                    "stwux $rS, $ptroff, $ptrreg", LdStStoreUpd,
-                    [(set ptr_rc:$ea_res,
-                       (pre_truncsti32 G8RC:$rS,
-                                       ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
-                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+def STWUX8: XForm_8<31, 183, (outs ptr_rc_nor0:$ea_res), (ins G8RC:$rS, memrr:$dst),
+                    "stwux $rS, $dst", LdStStoreUpd, []>,
+                    RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
                     PPC970_DGroup_Cracked;
-
-def STDUX : XForm_8<31, 181, (outs ptr_rc:$ea_res),
-                              (ins G8RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                    "stdux $rS, $ptroff, $ptrreg", LdStSTDU,
-                    [(set ptr_rc:$ea_res,
-                       (pre_store G8RC:$rS, ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
-                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+def STDUX : XForm_8<31, 181, (outs ptr_rc_nor0:$ea_res), (ins G8RC:$rS, memrr:$dst),
+                    "stdux $rS, $dst", LdStSTDU, []>,
+                    RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
                     PPC970_DGroup_Cracked, isPPC64;
-
-// STD_32/STDX_32 - Just like STD/STDX, but uses a '32-bit' input register.
-def STD_32  : DSForm_1<62, 0, (outs), (ins GPRC:$rT, memrix:$dst),
-                       "std $rT, $dst", LdStSTD,
-                       [(PPCstd_32  GPRC:$rT, ixaddr:$dst)]>, isPPC64;
-def STDX_32  : XForm_8<31, 149, (outs), (ins GPRC:$rT, memrr:$dst),
-                       "stdx $rT, $dst", LdStSTD,
-                       [(PPCstd_32  GPRC:$rT, xaddr:$dst)]>, isPPC64,
-                       PPC970_DGroup_Cracked;
 }
 
+// Patterns to match the pre-inc stores.  We can't put the patterns on
+// the instruction definitions directly as ISel wants the address base
+// and offset to be separate operands, not a single complex operand.
+def : Pat<(pre_truncsti8 i64:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
+          (STBU8 $rS, iaddroff:$ptroff, $ptrreg)>;
+def : Pat<(pre_truncsti16 i64:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
+          (STHU8 $rS, iaddroff:$ptroff, $ptrreg)>;
+def : Pat<(pre_truncsti32 i64:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
+          (STWU8 $rS, iaddroff:$ptroff, $ptrreg)>;
+def : Pat<(aligned4pre_store i64:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
+          (STDU $rS, iaddroff:$ptroff, $ptrreg)>;
+
+def : Pat<(pre_truncsti8 i64:$rS, iPTR:$ptrreg, iPTR:$ptroff),
+          (STBUX8 $rS, $ptrreg, $ptroff)>;
+def : Pat<(pre_truncsti16 i64:$rS, iPTR:$ptrreg, iPTR:$ptroff),
+          (STHUX8 $rS, $ptrreg, $ptroff)>;
+def : Pat<(pre_truncsti32 i64:$rS, iPTR:$ptrreg, iPTR:$ptroff),
+          (STWUX8 $rS, $ptrreg, $ptroff)>;
+def : Pat<(pre_store i64:$rS, iPTR:$ptrreg, iPTR:$ptroff),
+          (STDUX $rS, $ptrreg, $ptroff)>;
 
 
 //===----------------------------------------------------------------------===//
@@ -793,10 +855,26 @@ def STDX_32  : XForm_8<31, 149, (outs), (ins GPRC:$rT, memrr:$dst),
 let PPC970_Unit = 3, Uses = [RM] in {  // FPU Operations.
 def FCFID  : XForm_26<63, 846, (outs F8RC:$frD), (ins F8RC:$frB),
                       "fcfid $frD, $frB", FPGeneral,
-                      [(set F8RC:$frD, (PPCfcfid F8RC:$frB))]>, isPPC64;
+                      [(set f64:$frD, (PPCfcfid f64:$frB))]>, isPPC64;
 def FCTIDZ : XForm_26<63, 815, (outs F8RC:$frD), (ins F8RC:$frB),
                       "fctidz $frD, $frB", FPGeneral,
-                      [(set F8RC:$frD, (PPCfctidz F8RC:$frB))]>, isPPC64;
+                      [(set f64:$frD, (PPCfctidz f64:$frB))]>, isPPC64;
+
+def FCFIDU  : XForm_26<63, 974, (outs F8RC:$frD), (ins F8RC:$frB),
+                      "fcfidu $frD, $frB", FPGeneral,
+                      [(set f64:$frD, (PPCfcfidu f64:$frB))]>, isPPC64;
+def FCFIDS  : XForm_26<59, 846, (outs F4RC:$frD), (ins F8RC:$frB),
+                      "fcfids $frD, $frB", FPGeneral,
+                      [(set f32:$frD, (PPCfcfids f64:$frB))]>, isPPC64;
+def FCFIDUS : XForm_26<59, 974, (outs F4RC:$frD), (ins F8RC:$frB),
+                      "fcfidus $frD, $frB", FPGeneral,
+                      [(set f32:$frD, (PPCfcfidus f64:$frB))]>, isPPC64;
+def FCTIDUZ : XForm_26<63, 943, (outs F8RC:$frD), (ins F8RC:$frB),
+                      "fctiduz $frD, $frB", FPGeneral,
+                      [(set f64:$frD, (PPCfctiduz f64:$frB))]>, isPPC64;
+def FCTIWUZ : XForm_26<63, 143, (outs F8RC:$frD), (ins F8RC:$frB),
+                      "fctiwuz $frD, $frB", FPGeneral,
+                      [(set f64:$frD, (PPCfctiwuz f64:$frB))]>, isPPC64;
 }
 
 
@@ -805,13 +883,13 @@ def FCTIDZ : XForm_26<63, 815, (outs F8RC:$frD), (ins F8RC:$frB),
 //
 
 // Extensions and truncates to/from 32-bit regs.
-def : Pat<(i64 (zext GPRC:$in)),
-          (RLDICL (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPRC:$in, sub_32),
+def : Pat<(i64 (zext i32:$in)),
+          (RLDICL (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $in, sub_32),
                   0, 32)>;
-def : Pat<(i64 (anyext GPRC:$in)),
-          (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GPRC:$in, sub_32)>;
-def : Pat<(i32 (trunc G8RC:$in)),
-          (EXTRACT_SUBREG G8RC:$in, sub_32)>;
+def : Pat<(i64 (anyext i32:$in)),
+          (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $in, sub_32)>;
+def : Pat<(i32 (trunc i64:$in)),
+          (EXTRACT_SUBREG $in, sub_32)>;
 
 // Extending loads with i64 targets.
 def : Pat<(zextloadi1 iaddr:$src),
@@ -838,24 +916,24 @@ def : Pat<(extloadi32 xaddr:$src),
 // Standard shifts.  These are represented separately from the real shifts above
 // so that we can distinguish between shifts that allow 6-bit and 7-bit shift
 // amounts.
-def : Pat<(sra G8RC:$rS, GPRC:$rB),
-          (SRAD G8RC:$rS, GPRC:$rB)>;
-def : Pat<(srl G8RC:$rS, GPRC:$rB),
-          (SRD G8RC:$rS, GPRC:$rB)>;
-def : Pat<(shl G8RC:$rS, GPRC:$rB),
-          (SLD G8RC:$rS, GPRC:$rB)>;
+def : Pat<(sra i64:$rS, i32:$rB),
+          (SRAD $rS, $rB)>;
+def : Pat<(srl i64:$rS, i32:$rB),
+          (SRD $rS, $rB)>;
+def : Pat<(shl i64:$rS, i32:$rB),
+          (SLD $rS, $rB)>;
 
 // SHL/SRL
-def : Pat<(shl G8RC:$in, (i32 imm:$imm)),
-          (RLDICR G8RC:$in, imm:$imm, (SHL64 imm:$imm))>;
-def : Pat<(srl G8RC:$in, (i32 imm:$imm)),
-          (RLDICL G8RC:$in, (SRL64 imm:$imm), imm:$imm)>;
+def : Pat<(shl i64:$in, (i32 imm:$imm)),
+          (RLDICR $in, imm:$imm, (SHL64 imm:$imm))>;
+def : Pat<(srl i64:$in, (i32 imm:$imm)),
+          (RLDICL $in, (SRL64 imm:$imm), imm:$imm)>;
 
 // ROTL
-def : Pat<(rotl G8RC:$in, GPRC:$sh),
-          (RLDCL G8RC:$in, GPRC:$sh, 0)>;
-def : Pat<(rotl G8RC:$in, (i32 imm:$imm)),
-          (RLDICL G8RC:$in, imm:$imm, 0)>;
+def : Pat<(rotl i64:$in, i32:$sh),
+          (RLDCL $in, $sh, 0)>;
+def : Pat<(rotl i64:$in, (i32 imm:$imm)),
+          (RLDICL $in, imm:$imm, 0)>;
 
 // Hi and Lo for Darwin Global Addresses.
 def : Pat<(PPChi tglobaladdr:$in, 0), (LIS8 tglobaladdr:$in)>;
@@ -866,15 +944,25 @@ def : Pat<(PPChi tjumptable:$in , 0), (LIS8 tjumptable:$in)>;
 def : Pat<(PPClo tjumptable:$in , 0), (LI8  tjumptable:$in)>;
 def : Pat<(PPChi tblockaddress:$in, 0), (LIS8 tblockaddress:$in)>;
 def : Pat<(PPClo tblockaddress:$in, 0), (LI8  tblockaddress:$in)>;
-def : Pat<(PPChi tglobaltlsaddr:$g, G8RC:$in),
-          (ADDIS8 G8RC:$in, tglobaltlsaddr:$g)>;
-def : Pat<(PPClo tglobaltlsaddr:$g, G8RC:$in),
-          (ADDI8L G8RC:$in, tglobaltlsaddr:$g)>;
-def : Pat<(add G8RC:$in, (PPChi tglobaladdr:$g, 0)),
-          (ADDIS8 G8RC:$in, tglobaladdr:$g)>;
-def : Pat<(add G8RC:$in, (PPChi tconstpool:$g, 0)),
-          (ADDIS8 G8RC:$in, tconstpool:$g)>;
-def : Pat<(add G8RC:$in, (PPChi tjumptable:$g, 0)),
-          (ADDIS8 G8RC:$in, tjumptable:$g)>;
-def : Pat<(add G8RC:$in, (PPChi tblockaddress:$g, 0)),
-          (ADDIS8 G8RC:$in, tblockaddress:$g)>;
+def : Pat<(PPChi tglobaltlsaddr:$g, i64:$in),
+          (ADDIS8 $in, tglobaltlsaddr:$g)>;
+def : Pat<(PPClo tglobaltlsaddr:$g, i64:$in),
+          (ADDI8 $in, tglobaltlsaddr:$g)>;
+def : Pat<(add i64:$in, (PPChi tglobaladdr:$g, 0)),
+          (ADDIS8 $in, tglobaladdr:$g)>;
+def : Pat<(add i64:$in, (PPChi tconstpool:$g, 0)),
+          (ADDIS8 $in, tconstpool:$g)>;
+def : Pat<(add i64:$in, (PPChi tjumptable:$g, 0)),
+          (ADDIS8 $in, tjumptable:$g)>;
+def : Pat<(add i64:$in, (PPChi tblockaddress:$g, 0)),
+          (ADDIS8 $in, tblockaddress:$g)>;
+
+// Patterns to match r+r indexed loads and stores for
+// addresses without at least 4-byte alignment.
+def : Pat<(i64 (unaligned4sextloadi32 xoaddr:$src)),
+          (LWAX xoaddr:$src)>;
+def : Pat<(i64 (unaligned4load xoaddr:$src)),
+          (LDX xoaddr:$src)>;
+def : Pat<(unaligned4store i64:$rS, xoaddr:$dst),
+          (STDX $rS, xoaddr:$dst)>;
+
diff --git a/lib/Target/PowerPC/PPCInstrAltivec.td b/lib/Target/PowerPC/PPCInstrAltivec.td
index ba58c3e4ac88..a5ba4c8aebef 100644
--- a/lib/Target/PowerPC/PPCInstrAltivec.td
+++ b/lib/Target/PowerPC/PPCInstrAltivec.td
@@ -158,34 +158,75 @@ def vecspltisw : PatLeaf<(build_vector), [{
   return PPC::get_VSPLTI_elt(N, 4, *CurDAG).getNode() != 0;
 }], VSPLTISW_get_imm>;
 
-def V_immneg0 : PatLeaf<(build_vector), [{
-  return PPC::isAllNegativeZeroVector(N);
-}]>;
-
 //===----------------------------------------------------------------------===//
 // Helpers for defining instructions that directly correspond to intrinsics.
 
-// VA1a_Int - A VAForm_1a intrinsic definition.
-class VA1a_Int<bits<6> xo, string opc, Intrinsic IntID>
+// VA1a_Int_Ty - A VAForm_1a intrinsic definition of specific type.
+class VA1a_Int_Ty<bits<6> xo, string opc, Intrinsic IntID, ValueType Ty>
   : VAForm_1a<xo, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB, VRRC:$vC),
               !strconcat(opc, " $vD, $vA, $vB, $vC"), VecFP,
-                       [(set VRRC:$vD, (IntID VRRC:$vA, VRRC:$vB, VRRC:$vC))]>;
+                       [(set Ty:$vD, (IntID Ty:$vA, Ty:$vB, Ty:$vC))]>;
 
-// VX1_Int - A VXForm_1 intrinsic definition.
-class VX1_Int<bits<11> xo, string opc, Intrinsic IntID>
+// VA1a_Int_Ty2 - A VAForm_1a intrinsic definition where the type of the
+// inputs doesn't match the type of the output.
+class VA1a_Int_Ty2<bits<6> xo, string opc, Intrinsic IntID, ValueType OutTy,
+                   ValueType InTy>
+  : VAForm_1a<xo, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB, VRRC:$vC),
+              !strconcat(opc, " $vD, $vA, $vB, $vC"), VecFP,
+                       [(set OutTy:$vD, (IntID InTy:$vA, InTy:$vB, InTy:$vC))]>;
+
+// VA1a_Int_Ty3 - A VAForm_1a intrinsic definition where there are two
+// input types and an output type.
+class VA1a_Int_Ty3<bits<6> xo, string opc, Intrinsic IntID, ValueType OutTy,
+                   ValueType In1Ty, ValueType In2Ty>
+  : VAForm_1a<xo, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB, VRRC:$vC),
+              !strconcat(opc, " $vD, $vA, $vB, $vC"), VecFP,
+                       [(set OutTy:$vD,
+                         (IntID In1Ty:$vA, In1Ty:$vB, In2Ty:$vC))]>;
+
+// VX1_Int_Ty - A VXForm_1 intrinsic definition of specific type.
+class VX1_Int_Ty<bits<11> xo, string opc, Intrinsic IntID, ValueType Ty>
+  : VXForm_1<xo, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
+             !strconcat(opc, " $vD, $vA, $vB"), VecFP,
+             [(set Ty:$vD, (IntID Ty:$vA, Ty:$vB))]>;
+
+// VX1_Int_Ty2 - A VXForm_1 intrinsic definition where the type of the
+// inputs doesn't match the type of the output.
+class VX1_Int_Ty2<bits<11> xo, string opc, Intrinsic IntID, ValueType OutTy,
+                  ValueType InTy>
   : VXForm_1<xo, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
              !strconcat(opc, " $vD, $vA, $vB"), VecFP,
-             [(set VRRC:$vD, (IntID VRRC:$vA, VRRC:$vB))]>;
+             [(set OutTy:$vD, (IntID InTy:$vA, InTy:$vB))]>;
 
-// VX2_Int - A VXForm_2 intrinsic definition.
-class VX2_Int<bits<11> xo, string opc, Intrinsic IntID>
+// VX1_Int_Ty3 - A VXForm_1 intrinsic definition where there are two
+// input types and an output type.
+class VX1_Int_Ty3<bits<11> xo, string opc, Intrinsic IntID, ValueType OutTy,
+                  ValueType In1Ty, ValueType In2Ty>
+  : VXForm_1<xo, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
+             !strconcat(opc, " $vD, $vA, $vB"), VecFP,
+             [(set OutTy:$vD, (IntID In1Ty:$vA, In2Ty:$vB))]>;
+
+// VX2_Int_SP - A VXForm_2 intrinsic definition of vector single-precision type.
+class VX2_Int_SP<bits<11> xo, string opc, Intrinsic IntID>
+  : VXForm_2<xo, (outs VRRC:$vD), (ins VRRC:$vB),
+             !strconcat(opc, " $vD, $vB"), VecFP,
+             [(set v4f32:$vD, (IntID v4f32:$vB))]>;
+
+// VX2_Int_Ty2 - A VXForm_2 intrinsic definition where the type of the
+// inputs doesn't match the type of the output.
+class VX2_Int_Ty2<bits<11> xo, string opc, Intrinsic IntID, ValueType OutTy,
+                  ValueType InTy>
   : VXForm_2<xo, (outs VRRC:$vD), (ins VRRC:$vB),
              !strconcat(opc, " $vD, $vB"), VecFP,
-             [(set VRRC:$vD, (IntID VRRC:$vB))]>;
+             [(set OutTy:$vD, (IntID InTy:$vB))]>;
 
 //===----------------------------------------------------------------------===//
 // Instruction Definitions.
 
+def HasAltivec : Predicate<"PPCSubTarget.hasAltivec()">;
+let Predicates = [HasAltivec] in {
+
+let isCodeGenOnly = 1 in {
 def DSS      : DSS_Form<822, (outs),
                         (ins u5imm:$ZERO0, u5imm:$STRM,u5imm:$ZERO1,u5imm:$ZERO2),
                         "dss $STRM", LdStLoad /*FIXME*/, []>;
@@ -217,129 +258,136 @@ def DSTST64  : DSS_Form<374, (outs),
 def DSTSTT64 : DSS_Form<374, (outs),
                         (ins u5imm:$ONE, u5imm:$STRM, G8RC:$rA, GPRC:$rB),
                         "dststt $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>;
+}
 
 def MFVSCR : VXForm_4<1540, (outs VRRC:$vD), (ins),
                       "mfvscr $vD", LdStStore,
-                      [(set VRRC:$vD, (int_ppc_altivec_mfvscr))]>; 
+                      [(set v8i16:$vD, (int_ppc_altivec_mfvscr))]>; 
 def MTVSCR : VXForm_5<1604, (outs), (ins VRRC:$vB),
                       "mtvscr $vB", LdStLoad,
-                      [(int_ppc_altivec_mtvscr VRRC:$vB)]>; 
+                      [(int_ppc_altivec_mtvscr v4i32:$vB)]>; 
 
 let canFoldAsLoad = 1, PPC970_Unit = 2 in {  // Loads.
 def LVEBX: XForm_1<31,   7, (outs VRRC:$vD), (ins memrr:$src),
                    "lvebx $vD, $src", LdStLoad,
-                   [(set VRRC:$vD, (int_ppc_altivec_lvebx xoaddr:$src))]>;
+                   [(set v16i8:$vD, (int_ppc_altivec_lvebx xoaddr:$src))]>;
 def LVEHX: XForm_1<31,  39, (outs VRRC:$vD), (ins memrr:$src),
                    "lvehx $vD, $src", LdStLoad,
-                   [(set VRRC:$vD, (int_ppc_altivec_lvehx xoaddr:$src))]>;
+                   [(set v8i16:$vD, (int_ppc_altivec_lvehx xoaddr:$src))]>;
 def LVEWX: XForm_1<31,  71, (outs VRRC:$vD), (ins memrr:$src),
                    "lvewx $vD, $src", LdStLoad,
-                   [(set VRRC:$vD, (int_ppc_altivec_lvewx xoaddr:$src))]>;
+                   [(set v4i32:$vD, (int_ppc_altivec_lvewx xoaddr:$src))]>;
 def LVX  : XForm_1<31, 103, (outs VRRC:$vD), (ins memrr:$src),
                    "lvx $vD, $src", LdStLoad,
-                   [(set VRRC:$vD, (int_ppc_altivec_lvx xoaddr:$src))]>;
+                   [(set v4i32:$vD, (int_ppc_altivec_lvx xoaddr:$src))]>;
 def LVXL : XForm_1<31, 359, (outs VRRC:$vD), (ins memrr:$src),
                    "lvxl $vD, $src", LdStLoad,
-                   [(set VRRC:$vD, (int_ppc_altivec_lvxl xoaddr:$src))]>;
+                   [(set v4i32:$vD, (int_ppc_altivec_lvxl xoaddr:$src))]>;
 }
 
 def LVSL : XForm_1<31,   6, (outs VRRC:$vD), (ins memrr:$src),
                    "lvsl $vD, $src", LdStLoad,
-                   [(set VRRC:$vD, (int_ppc_altivec_lvsl xoaddr:$src))]>,
+                   [(set v16i8:$vD, (int_ppc_altivec_lvsl xoaddr:$src))]>,
                    PPC970_Unit_LSU;
 def LVSR : XForm_1<31,  38, (outs VRRC:$vD), (ins memrr:$src),
                    "lvsr $vD, $src", LdStLoad,
-                   [(set VRRC:$vD, (int_ppc_altivec_lvsr xoaddr:$src))]>,
+                   [(set v16i8:$vD, (int_ppc_altivec_lvsr xoaddr:$src))]>,
                    PPC970_Unit_LSU;
 
 let PPC970_Unit = 2 in {   // Stores.
 def STVEBX: XForm_8<31, 135, (outs), (ins VRRC:$rS, memrr:$dst),
                    "stvebx $rS, $dst", LdStStore,
-                   [(int_ppc_altivec_stvebx VRRC:$rS, xoaddr:$dst)]>;
+                   [(int_ppc_altivec_stvebx v16i8:$rS, xoaddr:$dst)]>;
 def STVEHX: XForm_8<31, 167, (outs), (ins VRRC:$rS, memrr:$dst),
                    "stvehx $rS, $dst", LdStStore,
-                   [(int_ppc_altivec_stvehx VRRC:$rS, xoaddr:$dst)]>;
+                   [(int_ppc_altivec_stvehx v8i16:$rS, xoaddr:$dst)]>;
 def STVEWX: XForm_8<31, 199, (outs), (ins VRRC:$rS, memrr:$dst),
                    "stvewx $rS, $dst", LdStStore,
-                   [(int_ppc_altivec_stvewx VRRC:$rS, xoaddr:$dst)]>;
+                   [(int_ppc_altivec_stvewx v4i32:$rS, xoaddr:$dst)]>;
 def STVX  : XForm_8<31, 231, (outs), (ins VRRC:$rS, memrr:$dst),
                    "stvx $rS, $dst", LdStStore,
-                   [(int_ppc_altivec_stvx VRRC:$rS, xoaddr:$dst)]>;
+                   [(int_ppc_altivec_stvx v4i32:$rS, xoaddr:$dst)]>;
 def STVXL : XForm_8<31, 487, (outs), (ins VRRC:$rS, memrr:$dst),
                    "stvxl $rS, $dst", LdStStore,
-                   [(int_ppc_altivec_stvxl VRRC:$rS, xoaddr:$dst)]>;
+                   [(int_ppc_altivec_stvxl v4i32:$rS, xoaddr:$dst)]>;
 }
 
 let PPC970_Unit = 5 in {  // VALU Operations.
 // VA-Form instructions.  3-input AltiVec ops.
 def VMADDFP : VAForm_1<46, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vC, VRRC:$vB),
                        "vmaddfp $vD, $vA, $vC, $vB", VecFP,
-                       [(set VRRC:$vD, (fma VRRC:$vA, VRRC:$vC, VRRC:$vB))]>;
+                       [(set v4f32:$vD,
+                        (fma v4f32:$vA, v4f32:$vC, v4f32:$vB))]>;
+
+// FIXME: The fma+fneg pattern won't match because fneg is not legal.
 def VNMSUBFP: VAForm_1<47, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vC, VRRC:$vB),
                        "vnmsubfp $vD, $vA, $vC, $vB", VecFP,
-                       [(set VRRC:$vD, (fneg (fma VRRC:$vA, VRRC:$vC,
-                                                  (fneg VRRC:$vB))))]>; 
+                       [(set v4f32:$vD, (fneg (fma v4f32:$vA, v4f32:$vC,
+                                                  (fneg v4f32:$vB))))]>; 
+
+def VMHADDSHS  : VA1a_Int_Ty<32, "vmhaddshs", int_ppc_altivec_vmhaddshs, v8i16>;
+def VMHRADDSHS : VA1a_Int_Ty<33, "vmhraddshs", int_ppc_altivec_vmhraddshs,
+                             v8i16>;
+def VMLADDUHM  : VA1a_Int_Ty<34, "vmladduhm", int_ppc_altivec_vmladduhm, v8i16>;
 
-def VMHADDSHS  : VA1a_Int<32, "vmhaddshs",  int_ppc_altivec_vmhaddshs>;
-def VMHRADDSHS : VA1a_Int<33, "vmhraddshs", int_ppc_altivec_vmhraddshs>;
-def VMLADDUHM  : VA1a_Int<34, "vmladduhm",  int_ppc_altivec_vmladduhm>;
-def VPERM      : VA1a_Int<43, "vperm",      int_ppc_altivec_vperm>;
-def VSEL       : VA1a_Int<42, "vsel",       int_ppc_altivec_vsel>;
+def VPERM      : VA1a_Int_Ty3<43, "vperm", int_ppc_altivec_vperm,
+                              v4i32, v4i32, v16i8>;
+def VSEL       : VA1a_Int_Ty<42, "vsel",  int_ppc_altivec_vsel, v4i32>;
 
 // Shuffles.
 def VSLDOI  : VAForm_2<44, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB, u5imm:$SH),
                        "vsldoi $vD, $vA, $vB, $SH", VecFP,
-                       [(set VRRC:$vD, 
-                         (vsldoi_shuffle:$SH (v16i8 VRRC:$vA), VRRC:$vB))]>;
+                       [(set v16i8:$vD, 
+                         (vsldoi_shuffle:$SH v16i8:$vA, v16i8:$vB))]>;
 
 // VX-Form instructions.  AltiVec arithmetic ops.
 def VADDFP : VXForm_1<10, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vaddfp $vD, $vA, $vB", VecFP,
-                      [(set VRRC:$vD, (fadd VRRC:$vA, VRRC:$vB))]>;
+                      [(set v4f32:$vD, (fadd v4f32:$vA, v4f32:$vB))]>;
                       
 def VADDUBM : VXForm_1<0, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vaddubm $vD, $vA, $vB", VecGeneral,
-                      [(set VRRC:$vD, (add (v16i8 VRRC:$vA), VRRC:$vB))]>;
+                      [(set v16i8:$vD, (add v16i8:$vA, v16i8:$vB))]>;
 def VADDUHM : VXForm_1<64, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vadduhm $vD, $vA, $vB", VecGeneral,
-                      [(set VRRC:$vD, (add (v8i16 VRRC:$vA), VRRC:$vB))]>;
+                      [(set v8i16:$vD, (add v8i16:$vA, v8i16:$vB))]>;
 def VADDUWM : VXForm_1<128, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vadduwm $vD, $vA, $vB", VecGeneral,
-                      [(set VRRC:$vD, (add (v4i32 VRRC:$vA), VRRC:$vB))]>;
+                      [(set v4i32:$vD, (add v4i32:$vA, v4i32:$vB))]>;
                       
-def VADDCUW : VX1_Int<384, "vaddcuw", int_ppc_altivec_vaddcuw>;
-def VADDSBS : VX1_Int<768, "vaddsbs", int_ppc_altivec_vaddsbs>;
-def VADDSHS : VX1_Int<832, "vaddshs", int_ppc_altivec_vaddshs>;
-def VADDSWS : VX1_Int<896, "vaddsws", int_ppc_altivec_vaddsws>;
-def VADDUBS : VX1_Int<512, "vaddubs", int_ppc_altivec_vaddubs>;
-def VADDUHS : VX1_Int<576, "vadduhs", int_ppc_altivec_vadduhs>;
-def VADDUWS : VX1_Int<640, "vadduws", int_ppc_altivec_vadduws>;
+def VADDCUW : VX1_Int_Ty<384, "vaddcuw", int_ppc_altivec_vaddcuw, v4i32>;
+def VADDSBS : VX1_Int_Ty<768, "vaddsbs", int_ppc_altivec_vaddsbs, v16i8>;
+def VADDSHS : VX1_Int_Ty<832, "vaddshs", int_ppc_altivec_vaddshs, v8i16>;
+def VADDSWS : VX1_Int_Ty<896, "vaddsws", int_ppc_altivec_vaddsws, v4i32>;
+def VADDUBS : VX1_Int_Ty<512, "vaddubs", int_ppc_altivec_vaddubs, v16i8>;
+def VADDUHS : VX1_Int_Ty<576, "vadduhs", int_ppc_altivec_vadduhs, v8i16>;
+def VADDUWS : VX1_Int_Ty<640, "vadduws", int_ppc_altivec_vadduws, v4i32>;
                              
                              
 def VAND : VXForm_1<1028, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                     "vand $vD, $vA, $vB", VecFP,
-                    [(set VRRC:$vD, (and (v4i32 VRRC:$vA), VRRC:$vB))]>;
+                    [(set v4i32:$vD, (and v4i32:$vA, v4i32:$vB))]>;
 def VANDC : VXForm_1<1092, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                      "vandc $vD, $vA, $vB", VecFP,
-                     [(set VRRC:$vD, (and (v4i32 VRRC:$vA),
-                                          (vnot_ppc VRRC:$vB)))]>;
+                     [(set v4i32:$vD, (and v4i32:$vA,
+                                           (vnot_ppc v4i32:$vB)))]>;
 
 def VCFSX  : VXForm_1<842, (outs VRRC:$vD), (ins u5imm:$UIMM, VRRC:$vB),
                       "vcfsx $vD, $vB, $UIMM", VecFP,
-                      [(set VRRC:$vD,
-                             (int_ppc_altivec_vcfsx VRRC:$vB, imm:$UIMM))]>;
+                      [(set v4f32:$vD,
+                             (int_ppc_altivec_vcfsx v4i32:$vB, imm:$UIMM))]>;
 def VCFUX  : VXForm_1<778, (outs VRRC:$vD), (ins u5imm:$UIMM, VRRC:$vB),
                       "vcfux $vD, $vB, $UIMM", VecFP,
-                      [(set VRRC:$vD,
-                             (int_ppc_altivec_vcfux VRRC:$vB, imm:$UIMM))]>;
+                      [(set v4f32:$vD,
+                             (int_ppc_altivec_vcfux v4i32:$vB, imm:$UIMM))]>;
 def VCTSXS : VXForm_1<970, (outs VRRC:$vD), (ins u5imm:$UIMM, VRRC:$vB),
                       "vctsxs $vD, $vB, $UIMM", VecFP,
-                      [(set VRRC:$vD,
-                             (int_ppc_altivec_vctsxs VRRC:$vB, imm:$UIMM))]>;
+                      [(set v4i32:$vD,
+                             (int_ppc_altivec_vctsxs v4f32:$vB, imm:$UIMM))]>;
 def VCTUXS : VXForm_1<906, (outs VRRC:$vD), (ins u5imm:$UIMM, VRRC:$vB),
                       "vctuxs $vD, $vB, $UIMM", VecFP,
-                      [(set VRRC:$vD,
-                             (int_ppc_altivec_vctuxs VRRC:$vB, imm:$UIMM))]>;
+                      [(set v4i32:$vD,
+                             (int_ppc_altivec_vctuxs v4f32:$vB, imm:$UIMM))]>;
 
 // Defines with the UIM field set to 0 for floating-point
 // to integer (fp_to_sint/fp_to_uint) conversions and integer
@@ -347,203 +395,237 @@ def VCTUXS : VXForm_1<906, (outs VRRC:$vD), (ins u5imm:$UIMM, VRRC:$vB),
 let VA = 0 in {
 def VCFSX_0 : VXForm_1<842, (outs VRRC:$vD), (ins VRRC:$vB),
                        "vcfsx $vD, $vB, 0", VecFP,
-                       [(set VRRC:$vD,
-                             (int_ppc_altivec_vcfsx VRRC:$vB, 0))]>;
+                       [(set v4f32:$vD,
+                             (int_ppc_altivec_vcfsx v4i32:$vB, 0))]>;
 def VCTUXS_0 : VXForm_1<906, (outs VRRC:$vD), (ins VRRC:$vB),
                         "vctuxs $vD, $vB, 0", VecFP,
-                        [(set VRRC:$vD,
-                               (int_ppc_altivec_vctuxs VRRC:$vB, 0))]>;
+                        [(set v4i32:$vD,
+                               (int_ppc_altivec_vctuxs v4f32:$vB, 0))]>;
 def VCFUX_0 : VXForm_1<778, (outs VRRC:$vD), (ins VRRC:$vB),
                        "vcfux $vD, $vB, 0", VecFP,
-                       [(set VRRC:$vD,
-                               (int_ppc_altivec_vcfux VRRC:$vB, 0))]>;
+                       [(set v4f32:$vD,
+                               (int_ppc_altivec_vcfux v4i32:$vB, 0))]>;
 def VCTSXS_0 : VXForm_1<970, (outs VRRC:$vD), (ins VRRC:$vB),
                       "vctsxs $vD, $vB, 0", VecFP,
-                      [(set VRRC:$vD,
-                             (int_ppc_altivec_vctsxs VRRC:$vB, 0))]>;
+                      [(set v4i32:$vD,
+                             (int_ppc_altivec_vctsxs v4f32:$vB, 0))]>;
 }
-def VEXPTEFP : VX2_Int<394, "vexptefp", int_ppc_altivec_vexptefp>;
-def VLOGEFP  : VX2_Int<458, "vlogefp",  int_ppc_altivec_vlogefp>;
-
-def VAVGSB : VX1_Int<1282, "vavgsb", int_ppc_altivec_vavgsb>;
-def VAVGSH : VX1_Int<1346, "vavgsh", int_ppc_altivec_vavgsh>;
-def VAVGSW : VX1_Int<1410, "vavgsw", int_ppc_altivec_vavgsw>;
-def VAVGUB : VX1_Int<1026, "vavgub", int_ppc_altivec_vavgub>;
-def VAVGUH : VX1_Int<1090, "vavguh", int_ppc_altivec_vavguh>;
-def VAVGUW : VX1_Int<1154, "vavguw", int_ppc_altivec_vavguw>;
-
-def VMAXFP : VX1_Int<1034, "vmaxfp", int_ppc_altivec_vmaxfp>;
-def VMAXSB : VX1_Int< 258, "vmaxsb", int_ppc_altivec_vmaxsb>;
-def VMAXSH : VX1_Int< 322, "vmaxsh", int_ppc_altivec_vmaxsh>;
-def VMAXSW : VX1_Int< 386, "vmaxsw", int_ppc_altivec_vmaxsw>;
-def VMAXUB : VX1_Int<   2, "vmaxub", int_ppc_altivec_vmaxub>;
-def VMAXUH : VX1_Int<  66, "vmaxuh", int_ppc_altivec_vmaxuh>;
-def VMAXUW : VX1_Int< 130, "vmaxuw", int_ppc_altivec_vmaxuw>;
-def VMINFP : VX1_Int<1098, "vminfp", int_ppc_altivec_vminfp>;
-def VMINSB : VX1_Int< 770, "vminsb", int_ppc_altivec_vminsb>;
-def VMINSH : VX1_Int< 834, "vminsh", int_ppc_altivec_vminsh>;
-def VMINSW : VX1_Int< 898, "vminsw", int_ppc_altivec_vminsw>;
-def VMINUB : VX1_Int< 514, "vminub", int_ppc_altivec_vminub>;
-def VMINUH : VX1_Int< 578, "vminuh", int_ppc_altivec_vminuh>;
-def VMINUW : VX1_Int< 642, "vminuw", int_ppc_altivec_vminuw>;
+def VEXPTEFP : VX2_Int_SP<394, "vexptefp", int_ppc_altivec_vexptefp>;
+def VLOGEFP  : VX2_Int_SP<458, "vlogefp",  int_ppc_altivec_vlogefp>;
+
+def VAVGSB : VX1_Int_Ty<1282, "vavgsb", int_ppc_altivec_vavgsb, v16i8>;
+def VAVGSH : VX1_Int_Ty<1346, "vavgsh", int_ppc_altivec_vavgsh, v8i16>;
+def VAVGSW : VX1_Int_Ty<1410, "vavgsw", int_ppc_altivec_vavgsw, v4i32>;
+def VAVGUB : VX1_Int_Ty<1026, "vavgub", int_ppc_altivec_vavgub, v16i8>;
+def VAVGUH : VX1_Int_Ty<1090, "vavguh", int_ppc_altivec_vavguh, v8i16>;
+def VAVGUW : VX1_Int_Ty<1154, "vavguw", int_ppc_altivec_vavguw, v4i32>;
+
+def VMAXFP : VX1_Int_Ty<1034, "vmaxfp", int_ppc_altivec_vmaxfp, v4f32>;
+def VMAXSB : VX1_Int_Ty< 258, "vmaxsb", int_ppc_altivec_vmaxsb, v16i8>;
+def VMAXSH : VX1_Int_Ty< 322, "vmaxsh", int_ppc_altivec_vmaxsh, v8i16>;
+def VMAXSW : VX1_Int_Ty< 386, "vmaxsw", int_ppc_altivec_vmaxsw, v4i32>;
+def VMAXUB : VX1_Int_Ty<   2, "vmaxub", int_ppc_altivec_vmaxub, v16i8>;
+def VMAXUH : VX1_Int_Ty<  66, "vmaxuh", int_ppc_altivec_vmaxuh, v8i16>;
+def VMAXUW : VX1_Int_Ty< 130, "vmaxuw", int_ppc_altivec_vmaxuw, v4i32>;
+def VMINFP : VX1_Int_Ty<1098, "vminfp", int_ppc_altivec_vminfp, v4f32>;
+def VMINSB : VX1_Int_Ty< 770, "vminsb", int_ppc_altivec_vminsb, v16i8>;
+def VMINSH : VX1_Int_Ty< 834, "vminsh", int_ppc_altivec_vminsh, v8i16>;
+def VMINSW : VX1_Int_Ty< 898, "vminsw", int_ppc_altivec_vminsw, v4i32>;
+def VMINUB : VX1_Int_Ty< 514, "vminub", int_ppc_altivec_vminub, v16i8>;
+def VMINUH : VX1_Int_Ty< 578, "vminuh", int_ppc_altivec_vminuh, v8i16>;
+def VMINUW : VX1_Int_Ty< 642, "vminuw", int_ppc_altivec_vminuw, v4i32>;
 
 def VMRGHB : VXForm_1< 12, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vmrghb $vD, $vA, $vB", VecFP,
-                      [(set VRRC:$vD, (vmrghb_shuffle VRRC:$vA, VRRC:$vB))]>;
+                      [(set v16i8:$vD, (vmrghb_shuffle v16i8:$vA, v16i8:$vB))]>;
 def VMRGHH : VXForm_1< 76, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vmrghh $vD, $vA, $vB", VecFP,
-                      [(set VRRC:$vD, (vmrghh_shuffle VRRC:$vA, VRRC:$vB))]>;
+                      [(set v16i8:$vD, (vmrghh_shuffle v16i8:$vA, v16i8:$vB))]>;
 def VMRGHW : VXForm_1<140, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vmrghw $vD, $vA, $vB", VecFP,
-                      [(set VRRC:$vD, (vmrghw_shuffle VRRC:$vA, VRRC:$vB))]>;
+                      [(set v16i8:$vD, (vmrghw_shuffle v16i8:$vA, v16i8:$vB))]>;
 def VMRGLB : VXForm_1<268, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vmrglb $vD, $vA, $vB", VecFP,
-                      [(set VRRC:$vD, (vmrglb_shuffle VRRC:$vA, VRRC:$vB))]>;
+                      [(set v16i8:$vD, (vmrglb_shuffle v16i8:$vA, v16i8:$vB))]>;
 def VMRGLH : VXForm_1<332, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vmrglh $vD, $vA, $vB", VecFP,
-                      [(set VRRC:$vD, (vmrglh_shuffle VRRC:$vA, VRRC:$vB))]>;
+                      [(set v16i8:$vD, (vmrglh_shuffle v16i8:$vA, v16i8:$vB))]>;
 def VMRGLW : VXForm_1<396, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vmrglw $vD, $vA, $vB", VecFP,
-                      [(set VRRC:$vD, (vmrglw_shuffle VRRC:$vA, VRRC:$vB))]>;
-
-def VMSUMMBM : VA1a_Int<37, "vmsummbm", int_ppc_altivec_vmsummbm>;
-def VMSUMSHM : VA1a_Int<40, "vmsumshm", int_ppc_altivec_vmsumshm>;
-def VMSUMSHS : VA1a_Int<41, "vmsumshs", int_ppc_altivec_vmsumshs>;
-def VMSUMUBM : VA1a_Int<36, "vmsumubm", int_ppc_altivec_vmsumubm>;
-def VMSUMUHM : VA1a_Int<38, "vmsumuhm", int_ppc_altivec_vmsumuhm>;
-def VMSUMUHS : VA1a_Int<39, "vmsumuhs", int_ppc_altivec_vmsumuhs>;
-
-def VMULESB : VX1_Int<776, "vmulesb", int_ppc_altivec_vmulesb>;
-def VMULESH : VX1_Int<840, "vmulesh", int_ppc_altivec_vmulesh>;
-def VMULEUB : VX1_Int<520, "vmuleub", int_ppc_altivec_vmuleub>;
-def VMULEUH : VX1_Int<584, "vmuleuh", int_ppc_altivec_vmuleuh>;
-def VMULOSB : VX1_Int<264, "vmulosb", int_ppc_altivec_vmulosb>;
-def VMULOSH : VX1_Int<328, "vmulosh", int_ppc_altivec_vmulosh>;
-def VMULOUB : VX1_Int<  8, "vmuloub", int_ppc_altivec_vmuloub>;
-def VMULOUH : VX1_Int< 72, "vmulouh", int_ppc_altivec_vmulouh>;
+                      [(set v16i8:$vD, (vmrglw_shuffle v16i8:$vA, v16i8:$vB))]>;
+
+def VMSUMMBM : VA1a_Int_Ty3<37, "vmsummbm", int_ppc_altivec_vmsummbm,
+                            v4i32, v16i8, v4i32>;
+def VMSUMSHM : VA1a_Int_Ty3<40, "vmsumshm", int_ppc_altivec_vmsumshm,
+                            v4i32, v8i16, v4i32>;
+def VMSUMSHS : VA1a_Int_Ty3<41, "vmsumshs", int_ppc_altivec_vmsumshs,
+                            v4i32, v8i16, v4i32>;
+def VMSUMUBM : VA1a_Int_Ty3<36, "vmsumubm", int_ppc_altivec_vmsumubm,
+                            v4i32, v16i8, v4i32>;
+def VMSUMUHM : VA1a_Int_Ty3<38, "vmsumuhm", int_ppc_altivec_vmsumuhm,
+                            v4i32, v8i16, v4i32>;
+def VMSUMUHS : VA1a_Int_Ty3<39, "vmsumuhs", int_ppc_altivec_vmsumuhs,
+                            v4i32, v8i16, v4i32>;
+
+def VMULESB : VX1_Int_Ty2<776, "vmulesb", int_ppc_altivec_vmulesb,
+                          v8i16, v16i8>;
+def VMULESH : VX1_Int_Ty2<840, "vmulesh", int_ppc_altivec_vmulesh,
+                          v4i32, v8i16>;
+def VMULEUB : VX1_Int_Ty2<520, "vmuleub", int_ppc_altivec_vmuleub,
+                          v8i16, v16i8>;
+def VMULEUH : VX1_Int_Ty2<584, "vmuleuh", int_ppc_altivec_vmuleuh,
+                          v4i32, v8i16>;
+def VMULOSB : VX1_Int_Ty2<264, "vmulosb", int_ppc_altivec_vmulosb,
+                          v8i16, v16i8>;
+def VMULOSH : VX1_Int_Ty2<328, "vmulosh", int_ppc_altivec_vmulosh,
+                          v4i32, v8i16>;
+def VMULOUB : VX1_Int_Ty2<  8, "vmuloub", int_ppc_altivec_vmuloub,
+                          v8i16, v16i8>;
+def VMULOUH : VX1_Int_Ty2< 72, "vmulouh", int_ppc_altivec_vmulouh,
+                          v4i32, v8i16>;
                        
-def VREFP     : VX2_Int<266, "vrefp",     int_ppc_altivec_vrefp>;
-def VRFIM     : VX2_Int<714, "vrfim",     int_ppc_altivec_vrfim>;
-def VRFIN     : VX2_Int<522, "vrfin",     int_ppc_altivec_vrfin>;
-def VRFIP     : VX2_Int<650, "vrfip",     int_ppc_altivec_vrfip>;
-def VRFIZ     : VX2_Int<586, "vrfiz",     int_ppc_altivec_vrfiz>;
-def VRSQRTEFP : VX2_Int<330, "vrsqrtefp", int_ppc_altivec_vrsqrtefp>;
+def VREFP     : VX2_Int_SP<266, "vrefp",     int_ppc_altivec_vrefp>;
+def VRFIM     : VX2_Int_SP<714, "vrfim",     int_ppc_altivec_vrfim>;
+def VRFIN     : VX2_Int_SP<522, "vrfin",     int_ppc_altivec_vrfin>;
+def VRFIP     : VX2_Int_SP<650, "vrfip",     int_ppc_altivec_vrfip>;
+def VRFIZ     : VX2_Int_SP<586, "vrfiz",     int_ppc_altivec_vrfiz>;
+def VRSQRTEFP : VX2_Int_SP<330, "vrsqrtefp", int_ppc_altivec_vrsqrtefp>;
 
-def VSUBCUW : VX1_Int<74, "vsubcuw", int_ppc_altivec_vsubcuw>;
+def VSUBCUW : VX1_Int_Ty<74, "vsubcuw", int_ppc_altivec_vsubcuw, v4i32>;
 
 def VSUBFP  : VXForm_1<74, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vsubfp $vD, $vA, $vB", VecGeneral,
-                      [(set VRRC:$vD, (fsub VRRC:$vA, VRRC:$vB))]>;
+                      [(set v4f32:$vD, (fsub v4f32:$vA, v4f32:$vB))]>;
 def VSUBUBM : VXForm_1<1024, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vsububm $vD, $vA, $vB", VecGeneral,
-                      [(set VRRC:$vD, (sub (v16i8 VRRC:$vA), VRRC:$vB))]>;
+                      [(set v16i8:$vD, (sub v16i8:$vA, v16i8:$vB))]>;
 def VSUBUHM : VXForm_1<1088, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vsubuhm $vD, $vA, $vB", VecGeneral,
-                      [(set VRRC:$vD, (sub (v8i16 VRRC:$vA), VRRC:$vB))]>;
+                      [(set v8i16:$vD, (sub v8i16:$vA, v8i16:$vB))]>;
 def VSUBUWM : VXForm_1<1152, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vsubuwm $vD, $vA, $vB", VecGeneral,
-                      [(set VRRC:$vD, (sub (v4i32 VRRC:$vA), VRRC:$vB))]>;
+                      [(set v4i32:$vD, (sub v4i32:$vA, v4i32:$vB))]>;
                       
-def VSUBSBS : VX1_Int<1792, "vsubsbs" , int_ppc_altivec_vsubsbs>;
-def VSUBSHS : VX1_Int<1856, "vsubshs" , int_ppc_altivec_vsubshs>;
-def VSUBSWS : VX1_Int<1920, "vsubsws" , int_ppc_altivec_vsubsws>;
-def VSUBUBS : VX1_Int<1536, "vsububs" , int_ppc_altivec_vsububs>;
-def VSUBUHS : VX1_Int<1600, "vsubuhs" , int_ppc_altivec_vsubuhs>;
-def VSUBUWS : VX1_Int<1664, "vsubuws" , int_ppc_altivec_vsubuws>;
-def VSUMSWS : VX1_Int<1928, "vsumsws" , int_ppc_altivec_vsumsws>;
-def VSUM2SWS: VX1_Int<1672, "vsum2sws", int_ppc_altivec_vsum2sws>;
-def VSUM4SBS: VX1_Int<1672, "vsum4sbs", int_ppc_altivec_vsum4sbs>;
-def VSUM4SHS: VX1_Int<1608, "vsum4shs", int_ppc_altivec_vsum4shs>;
-def VSUM4UBS: VX1_Int<1544, "vsum4ubs", int_ppc_altivec_vsum4ubs>;
+def VSUBSBS : VX1_Int_Ty<1792, "vsubsbs" , int_ppc_altivec_vsubsbs, v16i8>;
+def VSUBSHS : VX1_Int_Ty<1856, "vsubshs" , int_ppc_altivec_vsubshs, v8i16>;
+def VSUBSWS : VX1_Int_Ty<1920, "vsubsws" , int_ppc_altivec_vsubsws, v4i32>;
+def VSUBUBS : VX1_Int_Ty<1536, "vsububs" , int_ppc_altivec_vsububs, v16i8>;
+def VSUBUHS : VX1_Int_Ty<1600, "vsubuhs" , int_ppc_altivec_vsubuhs, v8i16>;
+def VSUBUWS : VX1_Int_Ty<1664, "vsubuws" , int_ppc_altivec_vsubuws, v4i32>;
+
+def VSUMSWS : VX1_Int_Ty<1928, "vsumsws" , int_ppc_altivec_vsumsws, v4i32>;
+def VSUM2SWS: VX1_Int_Ty<1672, "vsum2sws", int_ppc_altivec_vsum2sws, v4i32>;
+
+def VSUM4SBS: VX1_Int_Ty3<1672, "vsum4sbs", int_ppc_altivec_vsum4sbs,
+                          v4i32, v16i8, v4i32>;
+def VSUM4SHS: VX1_Int_Ty3<1608, "vsum4shs", int_ppc_altivec_vsum4shs,
+                          v4i32, v8i16, v4i32>;
+def VSUM4UBS: VX1_Int_Ty3<1544, "vsum4ubs", int_ppc_altivec_vsum4ubs,
+                          v4i32, v16i8, v4i32>;
 
 def VNOR : VXForm_1<1284, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                     "vnor $vD, $vA, $vB", VecFP,
-                    [(set VRRC:$vD, (vnot_ppc (or (v4i32 VRRC:$vA),
-                                                  VRRC:$vB)))]>;
+                    [(set v4i32:$vD, (vnot_ppc (or v4i32:$vA,
+                                                   v4i32:$vB)))]>;
 def VOR : VXForm_1<1156, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vor $vD, $vA, $vB", VecFP,
-                      [(set VRRC:$vD, (or (v4i32 VRRC:$vA), VRRC:$vB))]>;
+                      [(set v4i32:$vD, (or v4i32:$vA, v4i32:$vB))]>;
 def VXOR : VXForm_1<1220, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                       "vxor $vD, $vA, $vB", VecFP,
-                      [(set VRRC:$vD, (xor (v4i32 VRRC:$vA), VRRC:$vB))]>;
+                      [(set v4i32:$vD, (xor v4i32:$vA, v4i32:$vB))]>;
+
+def VRLB   : VX1_Int_Ty<   4, "vrlb", int_ppc_altivec_vrlb, v16i8>;
+def VRLH   : VX1_Int_Ty<  68, "vrlh", int_ppc_altivec_vrlh, v8i16>;
+def VRLW   : VX1_Int_Ty< 132, "vrlw", int_ppc_altivec_vrlw, v4i32>;
 
-def VRLB   : VX1_Int<   4, "vrlb", int_ppc_altivec_vrlb>;
-def VRLH   : VX1_Int<  68, "vrlh", int_ppc_altivec_vrlh>;
-def VRLW   : VX1_Int< 132, "vrlw", int_ppc_altivec_vrlw>;
+def VSL    : VX1_Int_Ty< 452, "vsl" , int_ppc_altivec_vsl,  v4i32 >;
+def VSLO   : VX1_Int_Ty<1036, "vslo", int_ppc_altivec_vslo, v4i32>;
 
-def VSL    : VX1_Int< 452, "vsl" , int_ppc_altivec_vsl >;
-def VSLO   : VX1_Int<1036, "vslo", int_ppc_altivec_vslo>;
-def VSLB   : VX1_Int< 260, "vslb", int_ppc_altivec_vslb>;
-def VSLH   : VX1_Int< 324, "vslh", int_ppc_altivec_vslh>;
-def VSLW   : VX1_Int< 388, "vslw", int_ppc_altivec_vslw>;
+def VSLB   : VX1_Int_Ty< 260, "vslb", int_ppc_altivec_vslb, v16i8>;
+def VSLH   : VX1_Int_Ty< 324, "vslh", int_ppc_altivec_vslh, v8i16>;
+def VSLW   : VX1_Int_Ty< 388, "vslw", int_ppc_altivec_vslw, v4i32>;
 
 def VSPLTB : VXForm_1<524, (outs VRRC:$vD), (ins u5imm:$UIMM, VRRC:$vB),
                       "vspltb $vD, $vB, $UIMM", VecPerm,
-                      [(set VRRC:$vD,
-                        (vspltb_shuffle:$UIMM (v16i8 VRRC:$vB), (undef)))]>;
+                      [(set v16i8:$vD,
+                        (vspltb_shuffle:$UIMM v16i8:$vB, (undef)))]>;
 def VSPLTH : VXForm_1<588, (outs VRRC:$vD), (ins u5imm:$UIMM, VRRC:$vB),
                       "vsplth $vD, $vB, $UIMM", VecPerm,
-                      [(set VRRC:$vD,
-                        (vsplth_shuffle:$UIMM (v16i8 VRRC:$vB), (undef)))]>;
+                      [(set v16i8:$vD,
+                        (vsplth_shuffle:$UIMM v16i8:$vB, (undef)))]>;
 def VSPLTW : VXForm_1<652, (outs VRRC:$vD), (ins u5imm:$UIMM, VRRC:$vB),
                       "vspltw $vD, $vB, $UIMM", VecPerm,
-                      [(set VRRC:$vD, 
-                        (vspltw_shuffle:$UIMM (v16i8 VRRC:$vB), (undef)))]>;
+                      [(set v16i8:$vD, 
+                        (vspltw_shuffle:$UIMM v16i8:$vB, (undef)))]>;
 
-def VSR    : VX1_Int< 708, "vsr"  , int_ppc_altivec_vsr>;
-def VSRO   : VX1_Int<1100, "vsro" , int_ppc_altivec_vsro>;
-def VSRAB  : VX1_Int< 772, "vsrab", int_ppc_altivec_vsrab>;
-def VSRAH  : VX1_Int< 836, "vsrah", int_ppc_altivec_vsrah>;
-def VSRAW  : VX1_Int< 900, "vsraw", int_ppc_altivec_vsraw>;
-def VSRB   : VX1_Int< 516, "vsrb" , int_ppc_altivec_vsrb>;
-def VSRH   : VX1_Int< 580, "vsrh" , int_ppc_altivec_vsrh>;
-def VSRW   : VX1_Int< 644, "vsrw" , int_ppc_altivec_vsrw>;
+def VSR    : VX1_Int_Ty< 708, "vsr"  , int_ppc_altivec_vsr,  v4i32>;
+def VSRO   : VX1_Int_Ty<1100, "vsro" , int_ppc_altivec_vsro, v4i32>;
+
+def VSRAB  : VX1_Int_Ty< 772, "vsrab", int_ppc_altivec_vsrab, v16i8>;
+def VSRAH  : VX1_Int_Ty< 836, "vsrah", int_ppc_altivec_vsrah, v8i16>;
+def VSRAW  : VX1_Int_Ty< 900, "vsraw", int_ppc_altivec_vsraw, v4i32>;
+def VSRB   : VX1_Int_Ty< 516, "vsrb" , int_ppc_altivec_vsrb , v16i8>;
+def VSRH   : VX1_Int_Ty< 580, "vsrh" , int_ppc_altivec_vsrh , v8i16>;
+def VSRW   : VX1_Int_Ty< 644, "vsrw" , int_ppc_altivec_vsrw , v4i32>;
 
 
 def VSPLTISB : VXForm_3<780, (outs VRRC:$vD), (ins s5imm:$SIMM),
                        "vspltisb $vD, $SIMM", VecPerm,
-                       [(set VRRC:$vD, (v16i8 vecspltisb:$SIMM))]>;
+                       [(set v16i8:$vD, (v16i8 vecspltisb:$SIMM))]>;
 def VSPLTISH : VXForm_3<844, (outs VRRC:$vD), (ins s5imm:$SIMM),
                        "vspltish $vD, $SIMM", VecPerm,
-                       [(set VRRC:$vD, (v8i16 vecspltish:$SIMM))]>;
+                       [(set v8i16:$vD, (v8i16 vecspltish:$SIMM))]>;
 def VSPLTISW : VXForm_3<908, (outs VRRC:$vD), (ins s5imm:$SIMM),
                        "vspltisw $vD, $SIMM", VecPerm,
-                       [(set VRRC:$vD, (v4i32 vecspltisw:$SIMM))]>;
+                       [(set v4i32:$vD, (v4i32 vecspltisw:$SIMM))]>;
 
 // Vector Pack.
-def VPKPX   : VX1_Int<782, "vpkpx", int_ppc_altivec_vpkpx>;
-def VPKSHSS : VX1_Int<398, "vpkshss", int_ppc_altivec_vpkshss>;
-def VPKSHUS : VX1_Int<270, "vpkshus", int_ppc_altivec_vpkshus>;
-def VPKSWSS : VX1_Int<462, "vpkswss", int_ppc_altivec_vpkswss>;
-def VPKSWUS : VX1_Int<334, "vpkswus", int_ppc_altivec_vpkswus>;
+def VPKPX   : VX1_Int_Ty2<782, "vpkpx", int_ppc_altivec_vpkpx,
+                          v8i16, v4i32>;
+def VPKSHSS : VX1_Int_Ty2<398, "vpkshss", int_ppc_altivec_vpkshss,
+                          v16i8, v8i16>;
+def VPKSHUS : VX1_Int_Ty2<270, "vpkshus", int_ppc_altivec_vpkshus,
+                          v16i8, v8i16>;
+def VPKSWSS : VX1_Int_Ty2<462, "vpkswss", int_ppc_altivec_vpkswss,
+                          v16i8, v4i32>;
+def VPKSWUS : VX1_Int_Ty2<334, "vpkswus", int_ppc_altivec_vpkswus,
+                          v8i16, v4i32>;
 def VPKUHUM : VXForm_1<14, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                        "vpkuhum $vD, $vA, $vB", VecFP,
-                       [(set VRRC:$vD,
-                         (vpkuhum_shuffle (v16i8 VRRC:$vA), VRRC:$vB))]>;
-def VPKUHUS : VX1_Int<142, "vpkuhus", int_ppc_altivec_vpkuhus>;
+                       [(set v16i8:$vD,
+                         (vpkuhum_shuffle v16i8:$vA, v16i8:$vB))]>;
+def VPKUHUS : VX1_Int_Ty2<142, "vpkuhus", int_ppc_altivec_vpkuhus,
+                          v16i8, v8i16>;
 def VPKUWUM : VXForm_1<78, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
                        "vpkuwum $vD, $vA, $vB", VecFP,
-                       [(set VRRC:$vD,
-                         (vpkuwum_shuffle (v16i8 VRRC:$vA), VRRC:$vB))]>;
-def VPKUWUS : VX1_Int<206, "vpkuwus", int_ppc_altivec_vpkuwus>;
+                       [(set v16i8:$vD,
+                         (vpkuwum_shuffle v16i8:$vA, v16i8:$vB))]>;
+def VPKUWUS : VX1_Int_Ty2<206, "vpkuwus", int_ppc_altivec_vpkuwus,
+                          v8i16, v4i32>;
 
 // Vector Unpack.
-def VUPKHPX : VX2_Int<846, "vupkhpx", int_ppc_altivec_vupkhpx>;
-def VUPKHSB : VX2_Int<526, "vupkhsb", int_ppc_altivec_vupkhsb>;
-def VUPKHSH : VX2_Int<590, "vupkhsh", int_ppc_altivec_vupkhsh>;
-def VUPKLPX : VX2_Int<974, "vupklpx", int_ppc_altivec_vupklpx>;
-def VUPKLSB : VX2_Int<654, "vupklsb", int_ppc_altivec_vupklsb>;
-def VUPKLSH : VX2_Int<718, "vupklsh", int_ppc_altivec_vupklsh>;
+def VUPKHPX : VX2_Int_Ty2<846, "vupkhpx", int_ppc_altivec_vupkhpx,
+                          v4i32, v8i16>;
+def VUPKHSB : VX2_Int_Ty2<526, "vupkhsb", int_ppc_altivec_vupkhsb,
+                          v8i16, v16i8>;
+def VUPKHSH : VX2_Int_Ty2<590, "vupkhsh", int_ppc_altivec_vupkhsh,
+                          v4i32, v8i16>;
+def VUPKLPX : VX2_Int_Ty2<974, "vupklpx", int_ppc_altivec_vupklpx,
+                          v4i32, v8i16>;
+def VUPKLSB : VX2_Int_Ty2<654, "vupklsb", int_ppc_altivec_vupklsb,
+                          v8i16, v16i8>;
+def VUPKLSH : VX2_Int_Ty2<718, "vupklsh", int_ppc_altivec_vupklsh,
+                          v4i32, v8i16>;
 
 
 // Altivec Comparisons.
 
 class VCMP<bits<10> xo, string asmstr, ValueType Ty>
   : VXRForm_1<xo, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),asmstr,VecFPCompare,
-              [(set VRRC:$vD, (Ty (PPCvcmp VRRC:$vA, VRRC:$vB, xo)))]>;
+              [(set Ty:$vD, (Ty (PPCvcmp Ty:$vA, Ty:$vB, xo)))]>;
 class VCMPo<bits<10> xo, string asmstr, ValueType Ty>
   : VXRForm_1<xo, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),asmstr,VecFPCompare,
-              [(set VRRC:$vD, (Ty (PPCvcmp_o VRRC:$vA, VRRC:$vB, xo)))]> {
+              [(set Ty:$vD, (Ty (PPCvcmp_o Ty:$vA, Ty:$vB, xo)))]> {
   let Defs = [CR6];
   let RC = 1;
 }
@@ -582,10 +664,16 @@ def VCMPGTSWo : VCMPo<902, "vcmpgtsw. $vD, $vA, $vB", v4i32>;
 def VCMPGTUW  : VCMP <646, "vcmpgtuw $vD, $vA, $vB" , v4i32>;
 def VCMPGTUWo : VCMPo<646, "vcmpgtuw. $vD, $vA, $vB", v4i32>;
                       
+let isCodeGenOnly = 1 in
 def V_SET0 : VXForm_setzero<1220, (outs VRRC:$vD), (ins),
                       "vxor $vD, $vD, $vD", VecFP,
-                      [(set VRRC:$vD, (v4i32 immAllZerosV))]>;
+                      [(set v4i32:$vD, (v4i32 immAllZerosV))]>;
+let IMM=-1 in {
+def V_SETALLONES : VXForm_3<908, (outs VRRC:$vD), (ins),
+                      "vspltisw $vD, -1", VecFP,
+                      [(set v4i32:$vD, (v4i32 immAllOnesV))]>;
 }
+} // VALU Operations.
 
 //===----------------------------------------------------------------------===//
 // Additional Altivec Patterns
@@ -596,31 +684,31 @@ def : Pat<(int_ppc_altivec_dssall), (DSSALL 1, 0, 0, 0)>;
 def : Pat<(int_ppc_altivec_dss imm:$STRM), (DSS 0, imm:$STRM, 0, 0)>;
 
 //  * 32-bit
-def : Pat<(int_ppc_altivec_dst GPRC:$rA, GPRC:$rB, imm:$STRM),
-          (DST 0, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
-def : Pat<(int_ppc_altivec_dstt GPRC:$rA, GPRC:$rB, imm:$STRM),
-          (DSTT 1, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
-def : Pat<(int_ppc_altivec_dstst GPRC:$rA, GPRC:$rB, imm:$STRM),
-          (DSTST 0, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
-def : Pat<(int_ppc_altivec_dststt GPRC:$rA, GPRC:$rB, imm:$STRM),
-          (DSTSTT 1, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
+def : Pat<(int_ppc_altivec_dst i32:$rA, i32:$rB, imm:$STRM),
+          (DST 0, imm:$STRM, $rA, $rB)>;
+def : Pat<(int_ppc_altivec_dstt i32:$rA, i32:$rB, imm:$STRM),
+          (DSTT 1, imm:$STRM, $rA, $rB)>;
+def : Pat<(int_ppc_altivec_dstst i32:$rA, i32:$rB, imm:$STRM),
+          (DSTST 0, imm:$STRM, $rA, $rB)>;
+def : Pat<(int_ppc_altivec_dststt i32:$rA, i32:$rB, imm:$STRM),
+          (DSTSTT 1, imm:$STRM, $rA, $rB)>;
 
 //  * 64-bit
-def : Pat<(int_ppc_altivec_dst G8RC:$rA, GPRC:$rB, imm:$STRM),
-          (DST64 0, imm:$STRM, (i64 G8RC:$rA), GPRC:$rB)>;
-def : Pat<(int_ppc_altivec_dstt G8RC:$rA, GPRC:$rB, imm:$STRM),
-          (DSTT64 1, imm:$STRM, (i64 G8RC:$rA), GPRC:$rB)>;
-def : Pat<(int_ppc_altivec_dstst G8RC:$rA, GPRC:$rB, imm:$STRM),
-          (DSTST64 0, imm:$STRM, (i64 G8RC:$rA), GPRC:$rB)>;
-def : Pat<(int_ppc_altivec_dststt G8RC:$rA, GPRC:$rB, imm:$STRM),
-          (DSTSTT64 1, imm:$STRM, (i64 G8RC:$rA), GPRC:$rB)>;
+def : Pat<(int_ppc_altivec_dst i64:$rA, i32:$rB, imm:$STRM),
+          (DST64 0, imm:$STRM, $rA, $rB)>;
+def : Pat<(int_ppc_altivec_dstt i64:$rA, i32:$rB, imm:$STRM),
+          (DSTT64 1, imm:$STRM, $rA, $rB)>;
+def : Pat<(int_ppc_altivec_dstst i64:$rA, i32:$rB, imm:$STRM),
+          (DSTST64 0, imm:$STRM, $rA, $rB)>;
+def : Pat<(int_ppc_altivec_dststt i64:$rA, i32:$rB, imm:$STRM),
+          (DSTSTT64 1, imm:$STRM, $rA, $rB)>;
 
 // Loads.
 def : Pat<(v4i32 (load xoaddr:$src)), (LVX xoaddr:$src)>;
 
 // Stores.
-def : Pat<(store (v4i32 VRRC:$rS), xoaddr:$dst),
-          (STVX (v4i32 VRRC:$rS), xoaddr:$dst)>;
+def : Pat<(store v4i32:$rS, xoaddr:$dst),
+          (STVX $rS, xoaddr:$dst)>;
 
 // Bit conversions.
 def : Pat<(v16i8 (bitconvert (v8i16 VRRC:$src))), (v16i8 VRRC:$src)>;
@@ -642,82 +730,99 @@ def : Pat<(v4f32 (bitconvert (v4i32 VRRC:$src))), (v4f32 VRRC:$src)>;
 // Shuffles.
 
 // Match vsldoi(x,x), vpkuwum(x,x), vpkuhum(x,x)
-def:Pat<(vsldoi_unary_shuffle:$in (v16i8 VRRC:$vA), undef),
-        (VSLDOI VRRC:$vA, VRRC:$vA, (VSLDOI_unary_get_imm VRRC:$in))>;
-def:Pat<(vpkuwum_unary_shuffle (v16i8 VRRC:$vA), undef),
-        (VPKUWUM VRRC:$vA, VRRC:$vA)>;
-def:Pat<(vpkuhum_unary_shuffle (v16i8 VRRC:$vA), undef),
-        (VPKUHUM VRRC:$vA, VRRC:$vA)>;
+def:Pat<(vsldoi_unary_shuffle:$in v16i8:$vA, undef),
+        (VSLDOI $vA, $vA, (VSLDOI_unary_get_imm $in))>;
+def:Pat<(vpkuwum_unary_shuffle v16i8:$vA, undef),
+        (VPKUWUM $vA, $vA)>;
+def:Pat<(vpkuhum_unary_shuffle v16i8:$vA, undef),
+        (VPKUHUM $vA, $vA)>;
 
 // Match vmrg*(x,x)
-def:Pat<(vmrglb_unary_shuffle (v16i8 VRRC:$vA), undef),
-        (VMRGLB VRRC:$vA, VRRC:$vA)>;
-def:Pat<(vmrglh_unary_shuffle (v16i8 VRRC:$vA), undef),
-        (VMRGLH VRRC:$vA, VRRC:$vA)>;
-def:Pat<(vmrglw_unary_shuffle (v16i8 VRRC:$vA), undef),
-        (VMRGLW VRRC:$vA, VRRC:$vA)>;
-def:Pat<(vmrghb_unary_shuffle (v16i8 VRRC:$vA), undef),
-        (VMRGHB VRRC:$vA, VRRC:$vA)>;
-def:Pat<(vmrghh_unary_shuffle (v16i8 VRRC:$vA), undef),
-        (VMRGHH VRRC:$vA, VRRC:$vA)>;
-def:Pat<(vmrghw_unary_shuffle (v16i8 VRRC:$vA), undef),
-        (VMRGHW VRRC:$vA, VRRC:$vA)>;
+def:Pat<(vmrglb_unary_shuffle v16i8:$vA, undef),
+        (VMRGLB $vA, $vA)>;
+def:Pat<(vmrglh_unary_shuffle v16i8:$vA, undef),
+        (VMRGLH $vA, $vA)>;
+def:Pat<(vmrglw_unary_shuffle v16i8:$vA, undef),
+        (VMRGLW $vA, $vA)>;
+def:Pat<(vmrghb_unary_shuffle v16i8:$vA, undef),
+        (VMRGHB $vA, $vA)>;
+def:Pat<(vmrghh_unary_shuffle v16i8:$vA, undef),
+        (VMRGHH $vA, $vA)>;
+def:Pat<(vmrghw_unary_shuffle v16i8:$vA, undef),
+        (VMRGHW $vA, $vA)>;
 
 // Logical Operations
-def : Pat<(v4i32 (vnot_ppc VRRC:$vA)), (VNOR VRRC:$vA, VRRC:$vA)>;
+def : Pat<(vnot_ppc v4i32:$vA), (VNOR $vA, $vA)>;
 
-def : Pat<(v4i32 (vnot_ppc (or VRRC:$A, VRRC:$B))),
-          (VNOR VRRC:$A, VRRC:$B)>;
-def : Pat<(v4i32 (and VRRC:$A, (vnot_ppc VRRC:$B))),
-          (VANDC VRRC:$A, VRRC:$B)>;
+def : Pat<(vnot_ppc (or v4i32:$A, v4i32:$B)),
+          (VNOR $A, $B)>;
+def : Pat<(and v4i32:$A, (vnot_ppc v4i32:$B)),
+          (VANDC $A, $B)>;
 
-def : Pat<(fmul VRRC:$vA, VRRC:$vB),
-          (VMADDFP VRRC:$vA, VRRC:$vB, (v4i32 (V_SET0)))>; 
+def : Pat<(fmul v4f32:$vA, v4f32:$vB),
+          (VMADDFP $vA, $vB,
+             (v4i32 (VSLW (V_SETALLONES), (V_SETALLONES))))>; 
 
 // Fused multiply add and multiply sub for packed float.  These are represented
 // separately from the real instructions above, for operations that must have
 // the additional precision, such as Newton-Rhapson (used by divide, sqrt)
-def : Pat<(PPCvmaddfp VRRC:$A, VRRC:$B, VRRC:$C),
-          (VMADDFP VRRC:$A, VRRC:$B, VRRC:$C)>;
-def : Pat<(PPCvnmsubfp VRRC:$A, VRRC:$B, VRRC:$C),
-          (VNMSUBFP VRRC:$A, VRRC:$B, VRRC:$C)>;
+def : Pat<(PPCvmaddfp v4f32:$A, v4f32:$B, v4f32:$C),
+          (VMADDFP $A, $B, $C)>;
+def : Pat<(PPCvnmsubfp v4f32:$A, v4f32:$B, v4f32:$C),
+          (VNMSUBFP $A, $B, $C)>;
+
+def : Pat<(int_ppc_altivec_vmaddfp v4f32:$A, v4f32:$B, v4f32:$C),
+          (VMADDFP $A, $B, $C)>;
+def : Pat<(int_ppc_altivec_vnmsubfp v4f32:$A, v4f32:$B, v4f32:$C),
+          (VNMSUBFP $A, $B, $C)>;
 
-def : Pat<(int_ppc_altivec_vmaddfp VRRC:$A, VRRC:$B, VRRC:$C),
-          (VMADDFP VRRC:$A, VRRC:$B, VRRC:$C)>;
-def : Pat<(int_ppc_altivec_vnmsubfp VRRC:$A, VRRC:$B, VRRC:$C),
-          (VNMSUBFP VRRC:$A, VRRC:$B, VRRC:$C)>;
+def : Pat<(PPCvperm v16i8:$vA, v16i8:$vB, v16i8:$vC),
+          (VPERM $vA, $vB, $vC)>;
 
-def : Pat<(PPCvperm (v16i8 VRRC:$vA), VRRC:$vB, VRRC:$vC),
-          (VPERM VRRC:$vA, VRRC:$vB, VRRC:$vC)>;
+def : Pat<(PPCfre v4f32:$A), (VREFP $A)>;
+def : Pat<(PPCfrsqrte v4f32:$A), (VRSQRTEFP $A)>;
 
 // Vector shifts
-def : Pat<(v16i8 (shl (v16i8 VRRC:$vA), (v16i8 VRRC:$vB))),
-          (v16i8 (VSLB VRRC:$vA, VRRC:$vB))>;
-def : Pat<(v8i16 (shl (v8i16 VRRC:$vA), (v8i16 VRRC:$vB))),
-          (v8i16 (VSLH VRRC:$vA, VRRC:$vB))>;
-def : Pat<(v4i32 (shl (v4i32 VRRC:$vA), (v4i32 VRRC:$vB))),
-          (v4i32 (VSLW VRRC:$vA, VRRC:$vB))>;
-
-def : Pat<(v16i8 (srl (v16i8 VRRC:$vA), (v16i8 VRRC:$vB))),
-          (v16i8 (VSRB VRRC:$vA, VRRC:$vB))>;
-def : Pat<(v8i16 (srl (v8i16 VRRC:$vA), (v8i16 VRRC:$vB))),
-          (v8i16 (VSRH VRRC:$vA, VRRC:$vB))>;
-def : Pat<(v4i32 (srl (v4i32 VRRC:$vA), (v4i32 VRRC:$vB))),
-          (v4i32 (VSRW VRRC:$vA, VRRC:$vB))>;
-
-def : Pat<(v16i8 (sra (v16i8 VRRC:$vA), (v16i8 VRRC:$vB))),
-          (v16i8 (VSRAB VRRC:$vA, VRRC:$vB))>;
-def : Pat<(v8i16 (sra (v8i16 VRRC:$vA), (v8i16 VRRC:$vB))),
-          (v8i16 (VSRAH VRRC:$vA, VRRC:$vB))>;
-def : Pat<(v4i32 (sra (v4i32 VRRC:$vA), (v4i32 VRRC:$vB))),
-          (v4i32 (VSRAW VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v16i8 (shl v16i8:$vA, v16i8:$vB)),
+          (v16i8 (VSLB $vA, $vB))>;
+def : Pat<(v8i16 (shl v8i16:$vA, v8i16:$vB)),
+          (v8i16 (VSLH $vA, $vB))>;
+def : Pat<(v4i32 (shl v4i32:$vA, v4i32:$vB)),
+          (v4i32 (VSLW $vA, $vB))>;
+
+def : Pat<(v16i8 (srl v16i8:$vA, v16i8:$vB)),
+          (v16i8 (VSRB $vA, $vB))>;
+def : Pat<(v8i16 (srl v8i16:$vA, v8i16:$vB)),
+          (v8i16 (VSRH $vA, $vB))>;
+def : Pat<(v4i32 (srl v4i32:$vA, v4i32:$vB)),
+          (v4i32 (VSRW $vA, $vB))>;
+
+def : Pat<(v16i8 (sra v16i8:$vA, v16i8:$vB)),
+          (v16i8 (VSRAB $vA, $vB))>;
+def : Pat<(v8i16 (sra v8i16:$vA, v8i16:$vB)),
+          (v8i16 (VSRAH $vA, $vB))>;
+def : Pat<(v4i32 (sra v4i32:$vA, v4i32:$vB)),
+          (v4i32 (VSRAW $vA, $vB))>;
 
 // Float to integer and integer to float conversions
-def : Pat<(v4i32 (fp_to_sint (v4f32 VRRC:$vA))),
-           (VCTSXS_0 VRRC:$vA)>;
-def : Pat<(v4i32 (fp_to_uint (v4f32 VRRC:$vA))),
-           (VCTUXS_0 VRRC:$vA)>;
-def : Pat<(v4f32 (sint_to_fp (v4i32 VRRC:$vA))),
-           (VCFSX_0 VRRC:$vA)>;
-def : Pat<(v4f32 (uint_to_fp (v4i32 VRRC:$vA))),
-           (VCFUX_0 VRRC:$vA)>;
+def : Pat<(v4i32 (fp_to_sint v4f32:$vA)),
+           (VCTSXS_0 $vA)>;
+def : Pat<(v4i32 (fp_to_uint v4f32:$vA)),
+           (VCTUXS_0 $vA)>;
+def : Pat<(v4f32 (sint_to_fp v4i32:$vA)),
+           (VCFSX_0 $vA)>;
+def : Pat<(v4f32 (uint_to_fp v4i32:$vA)),
+           (VCFUX_0 $vA)>;
+
+// Floating-point rounding
+def : Pat<(v4f32 (ffloor v4f32:$vA)),
+          (VRFIM $vA)>;
+def : Pat<(v4f32 (fceil v4f32:$vA)),
+          (VRFIP $vA)>;
+def : Pat<(v4f32 (ftrunc v4f32:$vA)),
+          (VRFIZ $vA)>;
+def : Pat<(v4f32 (fnearbyint v4f32:$vA)),
+          (VRFIN $vA)>;
+
+} // end HasAltivec
+
diff --git a/lib/Target/PowerPC/PPCInstrFormats.td b/lib/Target/PowerPC/PPCInstrFormats.td
index c3c171cd21fc..400b7e367bfe 100644
--- a/lib/Target/PowerPC/PPCInstrFormats.td
+++ b/lib/Target/PowerPC/PPCInstrFormats.td
@@ -120,6 +120,18 @@ class BForm_1<bits<6> opcode, bits<5> bo, bit aa, bit lk, dag OOL, dag IOL,
   let CR = 0;
 }
 
+class BForm_2<bits<6> opcode, bits<5> bo, bits<5> bi, bit aa, bit lk,
+              dag OOL, dag IOL, string asmstr>
+  : I<opcode, OOL, IOL, asmstr, BrB> {
+  bits<14> BD;
+
+  let Inst{6-10}  = bo;
+  let Inst{11-15} = bi;
+  let Inst{16-29} = BD;
+  let Inst{30}    = aa;
+  let Inst{31}    = lk;
+}
+
 // 1.7.4 D-Form
 class DForm_base<bits<6> opcode, dag OOL, dag IOL, string asmstr,
                  InstrItinClass itin, list<dag> pattern> 
@@ -664,14 +676,13 @@ class XFXForm_7_ext<bits<6> opcode, bits<10> xo, bits<10> spr,
 // This is probably 1.7.9, but I don't have the reference that uses this
 // numbering scheme...
 class XFLForm<bits<6> opcode, bits<10> xo, dag OOL, dag IOL, string asmstr, 
-                      string cstr, InstrItinClass itin, list<dag>pattern>
+              InstrItinClass itin, list<dag>pattern>
   : I<opcode, OOL, IOL, asmstr, itin> {
   bits<8> FM;
   bits<5> rT;
 
   bit RC = 0;    // set by isDOT
   let Pattern = pattern;
-  let Constraints = cstr;
 
   let Inst{6} = 0;
   let Inst{7-14}  = FM;
@@ -765,16 +776,14 @@ class AForm_4<bits<6> opcode, bits<5> xo, dag OOL, dag IOL, string asmstr,
   bits<5> RT;
   bits<5> RA;
   bits<5> RB;
-  bits<7> BIBO;  // 2 bits of BI and 5 bits of BO (must be 12).
-  bits<3> CR;
+  bits<5> COND;
 
   let Pattern = pattern;
 
   let Inst{6-10}  = RT;
   let Inst{11-15} = RA;
   let Inst{16-20} = RB;
-  let Inst{21-23} = CR;
-  let Inst{24-25} = BIBO{6-5};
+  let Inst{21-25} = COND;
   let Inst{26-30} = xo;
   let Inst{31}    = 0;
 }
@@ -987,6 +996,7 @@ class VXRForm_1<bits<10> xo, dag OOL, dag IOL, string asmstr,
 //===----------------------------------------------------------------------===//
 class Pseudo<dag OOL, dag IOL, string asmstr, list<dag> pattern>
     : I<0, OOL, IOL, asmstr, NoItinerary> {
+  let isCodeGenOnly = 1;
   let PPC64 = 0;
   let Pattern = pattern;
   let Inst{31-0} = 0;
diff --git a/lib/Target/PowerPC/PPCInstrInfo.cpp b/lib/Target/PowerPC/PPCInstrInfo.cpp
index d9d68446f536..69c54ed084be 100644
--- a/lib/Target/PowerPC/PPCInstrInfo.cpp
+++ b/lib/Target/PowerPC/PPCInstrInfo.cpp
@@ -12,12 +12,13 @@
 //===----------------------------------------------------------------------===//
 
 #include "PPCInstrInfo.h"
+#include "MCTargetDesc/PPCPredicates.h"
 #include "PPC.h"
+#include "PPCHazardRecognizers.h"
 #include "PPCInstrBuilder.h"
 #include "PPCMachineFunctionInfo.h"
 #include "PPCTargetMachine.h"
-#include "PPCHazardRecognizers.h"
-#include "MCTargetDesc/PPCPredicates.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineMemOperand.h"
@@ -28,16 +29,10 @@
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/STLExtras.h"
 
 #define GET_INSTRINFO_CTOR
 #include "PPCGenInstrInfo.inc"
 
-namespace llvm {
-extern cl::opt<bool> DisablePPC32RS;
-extern cl::opt<bool> DisablePPC64RS;
-}
-
 using namespace llvm;
 
 static cl::
@@ -60,7 +55,7 @@ ScheduleHazardRecognizer *PPCInstrInfo::CreateTargetHazardRecognizer(
     return new PPCScoreboardHazardRecognizer(II, DAG);
   }
 
-  return TargetInstrInfoImpl::CreateTargetHazardRecognizer(TM, DAG);
+  return TargetInstrInfo::CreateTargetHazardRecognizer(TM, DAG);
 }
 
 /// CreateTargetPostRAHazardRecognizer - Return the postRA hazard recognizer
@@ -99,12 +94,18 @@ bool PPCInstrInfo::isCoalescableExtInstr(const MachineInstr &MI,
 
 unsigned PPCInstrInfo::isLoadFromStackSlot(const MachineInstr *MI,
                                            int &FrameIndex) const {
+  // Note: This list must be kept consistent with LoadRegFromStackSlot.
   switch (MI->getOpcode()) {
   default: break;
   case PPC::LD:
   case PPC::LWZ:
   case PPC::LFS:
   case PPC::LFD:
+  case PPC::RESTORE_CR:
+  case PPC::LVX:
+  case PPC::RESTORE_VRSAVE:
+    // Check for the operands added by addFrameReference (the immediate is the
+    // offset which defaults to 0).
     if (MI->getOperand(1).isImm() && !MI->getOperand(1).getImm() &&
         MI->getOperand(2).isFI()) {
       FrameIndex = MI->getOperand(2).getIndex();
@@ -117,12 +118,18 @@ unsigned PPCInstrInfo::isLoadFromStackSlot(const MachineInstr *MI,
 
 unsigned PPCInstrInfo::isStoreToStackSlot(const MachineInstr *MI,
                                           int &FrameIndex) const {
+  // Note: This list must be kept consistent with StoreRegToStackSlot.
   switch (MI->getOpcode()) {
   default: break;
   case PPC::STD:
   case PPC::STW:
   case PPC::STFS:
   case PPC::STFD:
+  case PPC::SPILL_CR:
+  case PPC::STVX:
+  case PPC::SPILL_VRSAVE:
+    // Check for the operands added by addFrameReference (the immediate is the
+    // offset which defaults to 0).
     if (MI->getOperand(1).isImm() && !MI->getOperand(1).getImm() &&
         MI->getOperand(2).isFI()) {
       FrameIndex = MI->getOperand(2).getIndex();
@@ -141,7 +148,7 @@ PPCInstrInfo::commuteInstruction(MachineInstr *MI, bool NewMI) const {
 
   // Normal instructions can be commuted the obvious way.
   if (MI->getOpcode() != PPC::RLWIMI)
-    return TargetInstrInfoImpl::commuteInstruction(MI, NewMI);
+    return TargetInstrInfo::commuteInstruction(MI, NewMI);
 
   // Cannot commute if it has a non-zero rotate count.
   if (MI->getOperand(3).getImm() != 0)
@@ -444,40 +451,22 @@ PPCInstrInfo::StoreRegToStackSlot(MachineFunction &MF,
                                   unsigned SrcReg, bool isKill,
                                   int FrameIdx,
                                   const TargetRegisterClass *RC,
-                                  SmallVectorImpl<MachineInstr*> &NewMIs) const{
+                                  SmallVectorImpl<MachineInstr*> &NewMIs,
+                                  bool &NonRI, bool &SpillsVRS) const{
+  // Note: If additional store instructions are added here,
+  // update isStoreToStackSlot.
+
   DebugLoc DL;
   if (PPC::GPRCRegClass.hasSubClassEq(RC)) {
-    if (SrcReg != PPC::LR) {
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STW))
-                                         .addReg(SrcReg,
-                                                 getKillRegState(isKill)),
-                                         FrameIdx));
-    } else {
-      // FIXME: this spills LR immediately to memory in one step.  To do this,
-      // we use R11, which we know cannot be used in the prolog/epilog.  This is
-      // a hack.
-      NewMIs.push_back(BuildMI(MF, DL, get(PPC::MFLR), PPC::R11));
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STW))
-                                         .addReg(PPC::R11,
-                                                 getKillRegState(isKill)),
-                                         FrameIdx));
-    }
+    NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STW))
+                                       .addReg(SrcReg,
+                                               getKillRegState(isKill)),
+                                       FrameIdx));
   } else if (PPC::G8RCRegClass.hasSubClassEq(RC)) {
-    if (SrcReg != PPC::LR8) {
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STD))
-                                         .addReg(SrcReg,
-                                                 getKillRegState(isKill)),
-                                         FrameIdx));
-    } else {
-      // FIXME: this spills LR immediately to memory in one step.  To do this,
-      // we use X11, which we know cannot be used in the prolog/epilog.  This is
-      // a hack.
-      NewMIs.push_back(BuildMI(MF, DL, get(PPC::MFLR8), PPC::X11));
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STD))
-                                         .addReg(PPC::X11,
-                                                 getKillRegState(isKill)),
-                                         FrameIdx));
-    }
+    NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STD))
+                                       .addReg(SrcReg,
+                                               getKillRegState(isKill)),
+                                       FrameIdx));
   } else if (PPC::F8RCRegClass.hasSubClassEq(RC)) {
     NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STFD))
                                        .addReg(SrcReg,
@@ -489,47 +478,11 @@ PPCInstrInfo::StoreRegToStackSlot(MachineFunction &MF,
                                                getKillRegState(isKill)),
                                        FrameIdx));
   } else if (PPC::CRRCRegClass.hasSubClassEq(RC)) {
-    if ((!DisablePPC32RS && !TM.getSubtargetImpl()->isPPC64()) ||
-        (!DisablePPC64RS && TM.getSubtargetImpl()->isPPC64())) {
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::SPILL_CR))
-                                         .addReg(SrcReg,
-                                                 getKillRegState(isKill)),
-                                         FrameIdx));
-      return true;
-    } else {
-      // FIXME: We need a scatch reg here.  The trouble with using R0 is that
-      // it's possible for the stack frame to be so big the save location is
-      // out of range of immediate offsets, necessitating another register.
-      // We hack this on Darwin by reserving R2.  It's probably broken on Linux
-      // at the moment.
-
-      bool is64Bit = TM.getSubtargetImpl()->isPPC64();
-      // We need to store the CR in the low 4-bits of the saved value.  First,
-      // issue a MFCR to save all of the CRBits.
-      unsigned ScratchReg = TM.getSubtargetImpl()->isDarwinABI() ?
-                              (is64Bit ? PPC::X2 : PPC::R2) :
-                              (is64Bit ? PPC::X0 : PPC::R0);
-      NewMIs.push_back(BuildMI(MF, DL, get(is64Bit ? PPC::MFCR8pseud :
-                                             PPC::MFCRpseud), ScratchReg)
-                               .addReg(SrcReg, getKillRegState(isKill)));
-
-      // If the saved register wasn't CR0, shift the bits left so that they are
-      // in CR0's slot.
-      if (SrcReg != PPC::CR0) {
-        unsigned ShiftBits = getPPCRegisterNumbering(SrcReg)*4;
-        // rlwinm scratch, scratch, ShiftBits, 0, 31.
-        NewMIs.push_back(BuildMI(MF, DL, get(is64Bit ? PPC::RLWINM8 :
-                           PPC::RLWINM), ScratchReg)
-                       .addReg(ScratchReg).addImm(ShiftBits)
-                       .addImm(0).addImm(31));
-      }
-
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(is64Bit ?
-                                           PPC::STW8 : PPC::STW))
-                                         .addReg(ScratchReg,
-                                                 getKillRegState(isKill)),
-                                         FrameIdx));
-    }
+    NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::SPILL_CR))
+                                       .addReg(SrcReg,
+                                               getKillRegState(isKill)),
+                                       FrameIdx));
+    return true;
   } else if (PPC::CRBITRCRegClass.hasSubClassEq(RC)) {
     // FIXME: We use CRi here because there is no mtcrf on a bit. Since the
     // backend currently only uses CR1EQ as an individual bit, this should
@@ -562,23 +515,22 @@ PPCInstrInfo::StoreRegToStackSlot(MachineFunction &MF,
       Reg = PPC::CR7;
 
     return StoreRegToStackSlot(MF, Reg, isKill, FrameIdx,
-                               &PPC::CRRCRegClass, NewMIs);
+                               &PPC::CRRCRegClass, NewMIs, NonRI, SpillsVRS);
 
   } else if (PPC::VRRCRegClass.hasSubClassEq(RC)) {
-    // We don't have indexed addressing for vector loads.  Emit:
-    // R0 = ADDI FI#
-    // STVX VAL, 0, R0
-    //
-    // FIXME: We use R0 here, because it isn't available for RA.
-    bool Is64Bit = TM.getSubtargetImpl()->isPPC64();
-    unsigned Instr = Is64Bit ? PPC::ADDI8 : PPC::ADDI;
-    unsigned GPR0  = Is64Bit ? PPC::X0    : PPC::R0;
-    NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Instr), GPR0),
-                                       FrameIdx, 0, 0));
-    NewMIs.push_back(BuildMI(MF, DL, get(PPC::STVX))
-                     .addReg(SrcReg, getKillRegState(isKill))
-                     .addReg(GPR0)
-                     .addReg(GPR0));
+    NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STVX))
+                                       .addReg(SrcReg,
+                                               getKillRegState(isKill)),
+                                       FrameIdx));
+    NonRI = true;
+  } else if (PPC::VRSAVERCRegClass.hasSubClassEq(RC)) {
+    assert(TM.getSubtargetImpl()->isDarwin() &&
+           "VRSAVE only needs spill/restore on Darwin");
+    NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::SPILL_VRSAVE))
+                                       .addReg(SrcReg,
+                                               getKillRegState(isKill)),
+                                       FrameIdx));
+    SpillsVRS = true;
   } else {
     llvm_unreachable("Unknown regclass!");
   }
@@ -595,10 +547,19 @@ PPCInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
   MachineFunction &MF = *MBB.getParent();
   SmallVector<MachineInstr*, 4> NewMIs;
 
-  if (StoreRegToStackSlot(MF, SrcReg, isKill, FrameIdx, RC, NewMIs)) {
-    PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
+  PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
+  FuncInfo->setHasSpills();
+
+  bool NonRI = false, SpillsVRS = false;
+  if (StoreRegToStackSlot(MF, SrcReg, isKill, FrameIdx, RC, NewMIs,
+                          NonRI, SpillsVRS))
     FuncInfo->setSpillsCR();
-  }
+
+  if (SpillsVRS)
+    FuncInfo->setSpillsVRSAVE();
+
+  if (NonRI)
+    FuncInfo->setHasNonRISpills();
 
   for (unsigned i = 0, e = NewMIs.size(); i != e; ++i)
     MBB.insert(MI, NewMIs[i]);
@@ -616,25 +577,17 @@ bool
 PPCInstrInfo::LoadRegFromStackSlot(MachineFunction &MF, DebugLoc DL,
                                    unsigned DestReg, int FrameIdx,
                                    const TargetRegisterClass *RC,
-                                   SmallVectorImpl<MachineInstr*> &NewMIs)const{
+                                   SmallVectorImpl<MachineInstr*> &NewMIs,
+                                   bool &NonRI, bool &SpillsVRS) const{
+  // Note: If additional load instructions are added here,
+  // update isLoadFromStackSlot.
+
   if (PPC::GPRCRegClass.hasSubClassEq(RC)) {
-    if (DestReg != PPC::LR) {
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LWZ),
-                                                 DestReg), FrameIdx));
-    } else {
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LWZ),
-                                                 PPC::R11), FrameIdx));
-      NewMIs.push_back(BuildMI(MF, DL, get(PPC::MTLR)).addReg(PPC::R11));
-    }
+    NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LWZ),
+                                               DestReg), FrameIdx));
   } else if (PPC::G8RCRegClass.hasSubClassEq(RC)) {
-    if (DestReg != PPC::LR8) {
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LD), DestReg),
-                                         FrameIdx));
-    } else {
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LD),
-                                                 PPC::X11), FrameIdx));
-      NewMIs.push_back(BuildMI(MF, DL, get(PPC::MTLR8)).addReg(PPC::X11));
-    }
+    NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LD), DestReg),
+                                       FrameIdx));
   } else if (PPC::F8RCRegClass.hasSubClassEq(RC)) {
     NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LFD), DestReg),
                                        FrameIdx));
@@ -642,37 +595,10 @@ PPCInstrInfo::LoadRegFromStackSlot(MachineFunction &MF, DebugLoc DL,
     NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LFS), DestReg),
                                        FrameIdx));
   } else if (PPC::CRRCRegClass.hasSubClassEq(RC)) {
-    if ((!DisablePPC32RS && !TM.getSubtargetImpl()->isPPC64()) ||
-        (!DisablePPC64RS && TM.getSubtargetImpl()->isPPC64())) {
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL,
-                                                 get(PPC::RESTORE_CR), DestReg)
-                                         , FrameIdx));
-      return true;
-    } else {
-      // FIXME: We need a scatch reg here.  The trouble with using R0 is that
-      // it's possible for the stack frame to be so big the save location is
-      // out of range of immediate offsets, necessitating another register.
-      // We hack this on Darwin by reserving R2.  It's probably broken on Linux
-      // at the moment.
-      unsigned ScratchReg = TM.getSubtargetImpl()->isDarwinABI() ?
-                                                            PPC::R2 : PPC::R0;
-      NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LWZ),
-                                         ScratchReg), FrameIdx));
-  
-      // If the reloaded register isn't CR0, shift the bits right so that they are
-      // in the right CR's slot.
-      if (DestReg != PPC::CR0) {
-        unsigned ShiftBits = getPPCRegisterNumbering(DestReg)*4;
-        // rlwinm r11, r11, 32-ShiftBits, 0, 31.
-        NewMIs.push_back(BuildMI(MF, DL, get(PPC::RLWINM), ScratchReg)
-                      .addReg(ScratchReg).addImm(32-ShiftBits).addImm(0)
-                      .addImm(31));
-      }
-  
-      NewMIs.push_back(BuildMI(MF, DL, get(TM.getSubtargetImpl()->isPPC64() ?
-                         PPC::MTCRF8 : PPC::MTCRF), DestReg)
-                       .addReg(ScratchReg));
-    }
+    NewMIs.push_back(addFrameReference(BuildMI(MF, DL,
+                                               get(PPC::RESTORE_CR), DestReg),
+                                       FrameIdx));
+    return true;
   } else if (PPC::CRBITRCRegClass.hasSubClassEq(RC)) {
 
     unsigned Reg = 0;
@@ -702,21 +628,20 @@ PPCInstrInfo::LoadRegFromStackSlot(MachineFunction &MF, DebugLoc DL,
       Reg = PPC::CR7;
 
     return LoadRegFromStackSlot(MF, DL, Reg, FrameIdx,
-                                &PPC::CRRCRegClass, NewMIs);
+                                &PPC::CRRCRegClass, NewMIs, NonRI, SpillsVRS);
 
   } else if (PPC::VRRCRegClass.hasSubClassEq(RC)) {
-    // We don't have indexed addressing for vector loads.  Emit:
-    // R0 = ADDI FI#
-    // Dest = LVX 0, R0
-    //
-    // FIXME: We use R0 here, because it isn't available for RA.
-    bool Is64Bit = TM.getSubtargetImpl()->isPPC64();
-    unsigned Instr = Is64Bit ? PPC::ADDI8 : PPC::ADDI;
-    unsigned GPR0  = Is64Bit ? PPC::X0    : PPC::R0;
-    NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Instr), GPR0),
-                                       FrameIdx, 0, 0));
-    NewMIs.push_back(BuildMI(MF, DL, get(PPC::LVX),DestReg).addReg(GPR0)
-                     .addReg(GPR0));
+    NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LVX), DestReg),
+                                       FrameIdx));
+    NonRI = true;
+  } else if (PPC::VRSAVERCRegClass.hasSubClassEq(RC)) {
+    assert(TM.getSubtargetImpl()->isDarwin() &&
+           "VRSAVE only needs spill/restore on Darwin");
+    NewMIs.push_back(addFrameReference(BuildMI(MF, DL,
+                                               get(PPC::RESTORE_VRSAVE),
+                                               DestReg),
+                                       FrameIdx));
+    SpillsVRS = true;
   } else {
     llvm_unreachable("Unknown regclass!");
   }
@@ -734,10 +659,21 @@ PPCInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
   SmallVector<MachineInstr*, 4> NewMIs;
   DebugLoc DL;
   if (MI != MBB.end()) DL = MI->getDebugLoc();
-  if (LoadRegFromStackSlot(MF, DL, DestReg, FrameIdx, RC, NewMIs)) {
-    PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
+
+  PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
+  FuncInfo->setHasSpills();
+
+  bool NonRI = false, SpillsVRS = false;
+  if (LoadRegFromStackSlot(MF, DL, DestReg, FrameIdx, RC, NewMIs,
+                           NonRI, SpillsVRS))
     FuncInfo->setSpillsCR();
-  }
+
+  if (SpillsVRS)
+    FuncInfo->setSpillsVRSAVE();
+
+  if (NonRI)
+    FuncInfo->setHasNonRISpills();
+
   for (unsigned i = 0, e = NewMIs.size(); i != e; ++i)
     MBB.insert(MI, NewMIs[i]);
 
@@ -786,8 +722,8 @@ unsigned PPCInstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const {
   case PPC::GC_LABEL:
   case PPC::DBG_VALUE:
     return 0;
-  case PPC::BL8_NOP_ELF:
-  case PPC::BLA8_NOP_ELF:
+  case PPC::BL8_NOP:
+  case PPC::BLA8_NOP:
     return 8;
   default:
     return 4; // PowerPC instructions are all 4 bytes
diff --git a/lib/Target/PowerPC/PPCInstrInfo.h b/lib/Target/PowerPC/PPCInstrInfo.h
index 374213ea435b..635e3480b06d 100644
--- a/lib/Target/PowerPC/PPCInstrInfo.h
+++ b/lib/Target/PowerPC/PPCInstrInfo.h
@@ -71,11 +71,13 @@ class PPCInstrInfo : public PPCGenInstrInfo {
   bool StoreRegToStackSlot(MachineFunction &MF,
                            unsigned SrcReg, bool isKill, int FrameIdx,
                            const TargetRegisterClass *RC,
-                           SmallVectorImpl<MachineInstr*> &NewMIs) const;
+                           SmallVectorImpl<MachineInstr*> &NewMIs,
+                           bool &NonRI, bool &SpillsVRS) const;
   bool LoadRegFromStackSlot(MachineFunction &MF, DebugLoc DL,
                             unsigned DestReg, int FrameIdx,
                             const TargetRegisterClass *RC,
-                            SmallVectorImpl<MachineInstr*> &NewMIs) const;
+                            SmallVectorImpl<MachineInstr*> &NewMIs,
+                            bool &NonRI, bool &SpillsVRS) const;
 public:
   explicit PPCInstrInfo(PPCTargetMachine &TM);
 
diff --git a/lib/Target/PowerPC/PPCInstrInfo.td b/lib/Target/PowerPC/PPCInstrInfo.td
index 6ee045a2c7c9..ab907622beeb 100644
--- a/lib/Target/PowerPC/PPCInstrInfo.td
+++ b/lib/Target/PowerPC/PPCInstrInfo.td
@@ -20,6 +20,10 @@ include "PPCInstrFormats.td"
 def SDT_PPCstfiwx : SDTypeProfile<0, 2, [ // stfiwx
   SDTCisVT<0, f64>, SDTCisPtrTy<1>
 ]>;
+def SDT_PPClfiwx : SDTypeProfile<1, 1, [ // lfiw[az]x
+  SDTCisVT<0, f64>, SDTCisPtrTy<1>
+]>;
+
 def SDT_PPCCallSeqStart : SDCallSeqStart<[ SDTCisVT<0, i32> ]>;
 def SDT_PPCCallSeqEnd   : SDCallSeqEnd<[ SDTCisVT<0, i32>,
                                          SDTCisVT<1, i32> ]>;
@@ -36,10 +40,10 @@ def SDT_PPCcondbr : SDTypeProfile<0, 3, [
 ]>;
 
 def SDT_PPClbrx : SDTypeProfile<1, 2, [
-  SDTCisVT<0, i32>, SDTCisPtrTy<1>, SDTCisVT<2, OtherVT>
+  SDTCisInt<0>, SDTCisPtrTy<1>, SDTCisVT<2, OtherVT>
 ]>;
 def SDT_PPCstbrx : SDTypeProfile<0, 3, [
-  SDTCisVT<0, i32>, SDTCisPtrTy<1>, SDTCisVT<2, OtherVT>
+  SDTCisInt<0>, SDTCisPtrTy<1>, SDTCisVT<2, OtherVT>
 ]>;
 
 def SDT_PPClarx : SDTypeProfile<1, 1, [
@@ -53,32 +57,36 @@ def SDT_PPCTC_ret : SDTypeProfile<0, 2, [
   SDTCisPtrTy<0>, SDTCisVT<1, i32>
 ]>;
 
-def SDT_PPCnop : SDTypeProfile<0, 0, []>;
 
 //===----------------------------------------------------------------------===//
 // PowerPC specific DAG Nodes.
 //
 
-def PPCfcfid  : SDNode<"PPCISD::FCFID" , SDTFPUnaryOp, []>;
+def PPCfre    : SDNode<"PPCISD::FRE",     SDTFPUnaryOp, []>;
+def PPCfrsqrte: SDNode<"PPCISD::FRSQRTE", SDTFPUnaryOp, []>;
+
+def PPCfcfid  : SDNode<"PPCISD::FCFID",   SDTFPUnaryOp, []>;
+def PPCfcfidu : SDNode<"PPCISD::FCFIDU",  SDTFPUnaryOp, []>;
+def PPCfcfids : SDNode<"PPCISD::FCFIDS",  SDTFPRoundOp, []>;
+def PPCfcfidus: SDNode<"PPCISD::FCFIDUS", SDTFPRoundOp, []>;
 def PPCfctidz : SDNode<"PPCISD::FCTIDZ", SDTFPUnaryOp, []>;
 def PPCfctiwz : SDNode<"PPCISD::FCTIWZ", SDTFPUnaryOp, []>;
+def PPCfctiduz: SDNode<"PPCISD::FCTIDUZ",SDTFPUnaryOp, []>;
+def PPCfctiwuz: SDNode<"PPCISD::FCTIWUZ",SDTFPUnaryOp, []>;
 def PPCstfiwx : SDNode<"PPCISD::STFIWX", SDT_PPCstfiwx,
                        [SDNPHasChain, SDNPMayStore]>;
+def PPClfiwax : SDNode<"PPCISD::LFIWAX", SDT_PPClfiwx,
+                       [SDNPHasChain, SDNPMayLoad]>;
+def PPClfiwzx : SDNode<"PPCISD::LFIWZX", SDT_PPClfiwx,
+                       [SDNPHasChain, SDNPMayLoad]>;
+
+// Extract FPSCR (not modeled at the DAG level).
+def PPCmffs   : SDNode<"PPCISD::MFFS",
+                       SDTypeProfile<1, 0, [SDTCisVT<0, f64>]>, []>;
+
+// Perform FADD in round-to-zero mode.
+def PPCfaddrtz: SDNode<"PPCISD::FADDRTZ", SDTFPBinOp, []>;
 
-// This sequence is used for long double->int conversions.  It changes the
-// bits in the FPSCR which is not modelled.  
-def PPCmffs   : SDNode<"PPCISD::MFFS", SDTypeProfile<1, 0, [SDTCisVT<0, f64>]>,
-                        [SDNPOutGlue]>;
-def PPCmtfsb0 : SDNode<"PPCISD::MTFSB0", SDTypeProfile<0, 1, [SDTCisInt<0>]>,
-                       [SDNPInGlue, SDNPOutGlue]>;
-def PPCmtfsb1 : SDNode<"PPCISD::MTFSB1", SDTypeProfile<0, 1, [SDTCisInt<0>]>,
-                       [SDNPInGlue, SDNPOutGlue]>;
-def PPCfaddrtz: SDNode<"PPCISD::FADDRTZ", SDTFPBinOp,
-                       [SDNPInGlue, SDNPOutGlue]>;
-def PPCmtfsf  : SDNode<"PPCISD::MTFSF", SDTypeProfile<1, 3, 
-                       [SDTCisVT<0, f64>, SDTCisInt<1>, SDTCisVT<2, f64>,
-                        SDTCisVT<3, f64>]>,
-                       [SDNPInGlue]>;
 
 def PPCfsel   : SDNode<"PPCISD::FSEL",  
    // Type constraint for fsel.
@@ -91,6 +99,20 @@ def PPCtoc_entry: SDNode<"PPCISD::TOC_ENTRY", SDTIntBinOp, [SDNPMayLoad]>;
 def PPCvmaddfp  : SDNode<"PPCISD::VMADDFP", SDTFPTernaryOp, []>;
 def PPCvnmsubfp : SDNode<"PPCISD::VNMSUBFP", SDTFPTernaryOp, []>;
 
+def PPCaddisGotTprelHA : SDNode<"PPCISD::ADDIS_GOT_TPREL_HA", SDTIntBinOp>;
+def PPCldGotTprelL : SDNode<"PPCISD::LD_GOT_TPREL_L", SDTIntBinOp,
+                            [SDNPMayLoad]>;
+def PPCaddTls     : SDNode<"PPCISD::ADD_TLS", SDTIntBinOp, []>;
+def PPCaddisTlsgdHA : SDNode<"PPCISD::ADDIS_TLSGD_HA", SDTIntBinOp>;
+def PPCaddiTlsgdL   : SDNode<"PPCISD::ADDI_TLSGD_L", SDTIntBinOp>;
+def PPCgetTlsAddr   : SDNode<"PPCISD::GET_TLS_ADDR", SDTIntBinOp>;
+def PPCaddisTlsldHA : SDNode<"PPCISD::ADDIS_TLSLD_HA", SDTIntBinOp>;
+def PPCaddiTlsldL   : SDNode<"PPCISD::ADDI_TLSLD_L", SDTIntBinOp>;
+def PPCgetTlsldAddr : SDNode<"PPCISD::GET_TLSLD_ADDR", SDTIntBinOp>;
+def PPCaddisDtprelHA : SDNode<"PPCISD::ADDIS_DTPREL_HA", SDTIntBinOp,
+                              [SDNPHasChain]>;
+def PPCaddiDtprelL   : SDNode<"PPCISD::ADDI_DTPREL_L", SDTIntBinOp>;
+
 def PPCvperm    : SDNode<"PPCISD::VPERM", SDT_PPCvperm, []>;
 
 // These nodes represent the 32-bit PPC shifts that operate on 6-bit shift
@@ -99,10 +121,6 @@ def PPCsrl        : SDNode<"PPCISD::SRL"       , SDTIntShiftOp>;
 def PPCsra        : SDNode<"PPCISD::SRA"       , SDTIntShiftOp>;
 def PPCshl        : SDNode<"PPCISD::SHL"       , SDTIntShiftOp>;
 
-def PPCextsw_32   : SDNode<"PPCISD::EXTSW_32"  , SDTIntUnaryOp>;
-def PPCstd_32     : SDNode<"PPCISD::STD_32"    , SDTStore,
-                           [SDNPHasChain, SDNPMayStore]>;
-
 // These are target-independent nodes, but have target-specific formats.
 def callseq_start : SDNode<"ISD::CALLSEQ_START", SDT_PPCCallSeqStart,
                            [SDNPHasChain, SDNPOutGlue]>;
@@ -110,16 +128,12 @@ def callseq_end   : SDNode<"ISD::CALLSEQ_END",   SDT_PPCCallSeqEnd,
                            [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
 
 def SDT_PPCCall   : SDTypeProfile<0, -1, [SDTCisInt<0>]>;
-def PPCcall_Darwin : SDNode<"PPCISD::CALL_Darwin", SDT_PPCCall,
-                            [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
-                             SDNPVariadic]>;
-def PPCcall_SVR4  : SDNode<"PPCISD::CALL_SVR4", SDT_PPCCall,
-                           [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
-                            SDNPVariadic]>;
-def PPCcall_nop_SVR4  : SDNode<"PPCISD::CALL_NOP_SVR4", SDT_PPCCall,
-                               [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
-                                SDNPVariadic]>;
-def PPCnop : SDNode<"PPCISD::NOP", SDT_PPCnop, [SDNPInGlue, SDNPOutGlue]>;
+def PPCcall  : SDNode<"PPCISD::CALL", SDT_PPCCall,
+                      [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
+                       SDNPVariadic]>;
+def PPCcall_nop  : SDNode<"PPCISD::CALL_NOP", SDT_PPCCall,
+                          [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
+                           SDNPVariadic]>;
 def PPCload   : SDNode<"PPCISD::LOAD", SDTypeProfile<1, 1, []>,
                        [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
 def PPCload_toc : SDNode<"PPCISD::LOAD_TOC", SDTypeProfile<0, 1, []>,
@@ -130,13 +144,9 @@ def PPCtoc_restore : SDNode<"PPCISD::TOC_RESTORE", SDTypeProfile<0, 0, []>,
                              SDNPInGlue, SDNPOutGlue]>;
 def PPCmtctr      : SDNode<"PPCISD::MTCTR", SDT_PPCCall,
                            [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
-def PPCbctrl_Darwin  : SDNode<"PPCISD::BCTRL_Darwin", SDTNone,
-                              [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
-                               SDNPVariadic]>;
-
-def PPCbctrl_SVR4  : SDNode<"PPCISD::BCTRL_SVR4", SDTNone,
-                            [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
-                             SDNPVariadic]>;
+def PPCbctrl : SDNode<"PPCISD::BCTRL", SDTNone,
+                      [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
+                       SDNPVariadic]>;
 
 def retflag       : SDNode<"PPCISD::RET_FLAG", SDTNone,
                            [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
@@ -144,6 +154,14 @@ def retflag       : SDNode<"PPCISD::RET_FLAG", SDTNone,
 def PPCtc_return : SDNode<"PPCISD::TC_RETURN", SDT_PPCTC_ret,
                         [SDNPHasChain,  SDNPOptInGlue, SDNPVariadic]>;
 
+def PPCeh_sjlj_setjmp  : SDNode<"PPCISD::EH_SJLJ_SETJMP",
+                                SDTypeProfile<1, 1, [SDTCisInt<0>,
+                                                     SDTCisPtrTy<1>]>,
+                                [SDNPHasChain, SDNPSideEffect]>;
+def PPCeh_sjlj_longjmp : SDNode<"PPCISD::EH_SJLJ_LONGJMP",
+                                SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>,
+                                [SDNPHasChain, SDNPSideEffect]>;
+
 def PPCvcmp       : SDNode<"PPCISD::VCMP" , SDT_PPCvcmp, []>;
 def PPCvcmp_o     : SDNode<"PPCISD::VCMPo", SDT_PPCvcmp, [SDNPOutGlue]>;
 
@@ -167,6 +185,12 @@ def PPClarx      : SDNode<"PPCISD::LARX", SDT_PPClarx,
 def PPCstcx      : SDNode<"PPCISD::STCX", SDT_PPCstcx,
                           [SDNPHasChain, SDNPMayStore]>;
 
+// Instructions to support medium and large code model
+def PPCaddisTocHA : SDNode<"PPCISD::ADDIS_TOC_HA", SDTIntBinOp, []>;
+def PPCldTocL     : SDNode<"PPCISD::LD_TOC_L", SDTIntBinOp, [SDNPMayLoad]>;
+def PPCaddiTocL   : SDNode<"PPCISD::ADDI_TOC_L", SDTIntBinOp, []>;
+
+
 // Instructions to support dynamic alloca.
 def SDTDynOp  : SDTypeProfile<1, 2, []>;
 def PPCdynalloc   : SDNode<"PPCISD::DYNALLOC", SDTDynOp, [SDNPHasChain]>;
@@ -258,6 +282,38 @@ def imm16ShiftedSExt : PatLeaf<(imm), [{
   return N->getZExtValue() == (uint64_t)(int)N->getZExtValue();
 }], HI16>;
 
+// Some r+i load/store instructions (such as LD, STD, LDU, etc.) that require
+// restricted memrix (offset/4) constants are alignment sensitive. If these
+// offsets are hidden behind TOC entries than the values of the lower-order
+// bits cannot be checked directly. As a result, we need to also incorporate
+// an alignment check into the relevant patterns.
+
+def aligned4load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+  return cast<LoadSDNode>(N)->getAlignment() >= 4;
+}]>;
+def aligned4store : PatFrag<(ops node:$val, node:$ptr),
+                            (store node:$val, node:$ptr), [{
+  return cast<StoreSDNode>(N)->getAlignment() >= 4;
+}]>;
+def aligned4sextloadi32 : PatFrag<(ops node:$ptr), (sextloadi32 node:$ptr), [{
+  return cast<LoadSDNode>(N)->getAlignment() >= 4;
+}]>;
+def aligned4pre_store : PatFrag<
+                          (ops node:$val, node:$base, node:$offset),
+                          (pre_store node:$val, node:$base, node:$offset), [{
+  return cast<StoreSDNode>(N)->getAlignment() >= 4;
+}]>;
+
+def unaligned4load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+  return cast<LoadSDNode>(N)->getAlignment() < 4;
+}]>;
+def unaligned4store : PatFrag<(ops node:$val, node:$ptr),
+                              (store node:$val, node:$ptr), [{
+  return cast<StoreSDNode>(N)->getAlignment() < 4;
+}]>;
+def unaligned4sextloadi32 : PatFrag<(ops node:$ptr), (sextloadi32 node:$ptr), [{
+  return cast<LoadSDNode>(N)->getAlignment() < 4;
+}]>;
 
 //===----------------------------------------------------------------------===//
 // PowerPC Flag Definitions.
@@ -294,9 +350,6 @@ def s16imm  : Operand<i32> {
 def u16imm  : Operand<i32> {
   let PrintMethod = "printU16ImmOperand";
 }
-def s16immX4  : Operand<i32> {   // Multiply imm by 4 before printing.
-  let PrintMethod = "printS16X4ImmOperand";
-}
 def directbrtarget : Operand<OtherVT> {
   let PrintMethod = "printBranchOperand";
   let EncoderMethod = "getDirectBrEncoding";
@@ -324,26 +377,37 @@ def crbitm: Operand<i8> {
   let EncoderMethod = "get_crbitm_encoding";
 }
 // Address operands
+// A version of ptr_rc which excludes R0 (or X0 in 64-bit mode).
+def ptr_rc_nor0 : PointerLikeRegClass<1>;
+
+def dispRI : Operand<iPTR>;
+def dispRIX : Operand<iPTR>;
+
 def memri : Operand<iPTR> {
   let PrintMethod = "printMemRegImm";
-  let MIOperandInfo = (ops i32imm:$imm, ptr_rc:$reg);
+  let MIOperandInfo = (ops dispRI:$imm, ptr_rc_nor0:$reg);
   let EncoderMethod = "getMemRIEncoding";
 }
 def memrr : Operand<iPTR> {
   let PrintMethod = "printMemRegReg";
-  let MIOperandInfo = (ops ptr_rc:$offreg, ptr_rc:$ptrreg);
+  let MIOperandInfo = (ops ptr_rc_nor0:$ptrreg, ptr_rc:$offreg);
 }
 def memrix : Operand<iPTR> {   // memri where the imm is shifted 2 bits.
   let PrintMethod = "printMemRegImmShifted";
-  let MIOperandInfo = (ops i32imm:$imm, ptr_rc:$reg);
+  let MIOperandInfo = (ops dispRIX:$imm, ptr_rc_nor0:$reg);
   let EncoderMethod = "getMemRIXEncoding";
 }
 
-// PowerPC Predicate operand.  20 = (0<<5)|20 = always, CR0 is a dummy reg
-// that doesn't matter.
-def pred : PredicateOperand<OtherVT, (ops imm, CRRC),
-                                     (ops (i32 20), (i32 zero_reg))> {
+// A single-register address. This is used with the SjLj
+// pseudo-instructions.
+def memr : Operand<iPTR> {
+  let MIOperandInfo = (ops ptr_rc:$ptrreg);
+}
+
+// PowerPC Predicate operand.
+def pred : Operand<OtherVT> {
   let PrintMethod = "printPredicateOperand";
+  let MIOperandInfo = (ops i32imm:$bibo, CRRC:$reg);
 }
 
 // Define PowerPC specific addressing mode.
@@ -352,9 +416,12 @@ def xaddr  : ComplexPattern<iPTR, 2, "SelectAddrIdx",    [], []>;
 def xoaddr : ComplexPattern<iPTR, 2, "SelectAddrIdxOnly",[], []>;
 def ixaddr : ComplexPattern<iPTR, 2, "SelectAddrImmShift", [], []>; // "std"
 
+// The address in a single register. This is used with the SjLj
+// pseudo-instructions.
+def addr   : ComplexPattern<iPTR, 1, "SelectAddr",[], []>;
+
 /// This is just the offset part of iaddr, used for preinc.
 def iaddroff : ComplexPattern<iPTR, 1, "SelectAddrImmOffs", [], []>;
-def xaddroff : ComplexPattern<iPTR, 1, "SelectAddrIdxOffs", [], []>;
 
 //===----------------------------------------------------------------------===//
 // PowerPC Instruction Predicate Definitions.
@@ -381,17 +448,22 @@ def UPDATE_VRSAVE    : Pseudo<(outs GPRC:$rD), (ins GPRC:$rS),
 
 let Defs = [R1], Uses = [R1] in
 def DYNALLOC : Pseudo<(outs GPRC:$result), (ins GPRC:$negsize, memri:$fpsi), "#DYNALLOC",
-                       [(set GPRC:$result,
-                             (PPCdynalloc GPRC:$negsize, iaddr:$fpsi))]>;
+                       [(set i32:$result,
+                             (PPCdynalloc i32:$negsize, iaddr:$fpsi))]>;
                          
 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation.  Expanded after
 // instruction selection into a branch sequence.
 let usesCustomInserter = 1,    // Expanded after instruction selection.
     PPC970_Single = 1 in {
-  def SELECT_CC_I4 : Pseudo<(outs GPRC:$dst), (ins CRRC:$cond, GPRC:$T, GPRC:$F,
+  // Note that SELECT_CC_I4 and SELECT_CC_I8 use the no-r0 register classes
+  // because either operand might become the first operand in an isel, and
+  // that operand cannot be r0.
+  def SELECT_CC_I4 : Pseudo<(outs GPRC:$dst), (ins CRRC:$cond,
+                              GPRC_NOR0:$T, GPRC_NOR0:$F,
                               i32imm:$BROPC), "#SELECT_CC_I4",
                               []>;
-  def SELECT_CC_I8 : Pseudo<(outs G8RC:$dst), (ins CRRC:$cond, G8RC:$T, G8RC:$F,
+  def SELECT_CC_I8 : Pseudo<(outs G8RC:$dst), (ins CRRC:$cond,
+                              G8RC_NOX0:$T, G8RC_NOX0:$F,
                               i32imm:$BROPC), "#SELECT_CC_I8",
                               []>;
   def SELECT_CC_F4  : Pseudo<(outs F4RC:$dst), (ins CRRC:$cond, F4RC:$T, F4RC:$F,
@@ -418,10 +490,9 @@ def RESTORE_CR : Pseudo<(outs CRRC:$cond), (ins memri:$F),
                      "#RESTORE_CR", []>;
 
 let isTerminator = 1, isBarrier = 1, PPC970_Unit = 7 in {
-  let isCodeGenOnly = 1, isReturn = 1, Uses = [LR, RM] in
-    def BLR : XLForm_2_br<19, 16, 0, (outs), (ins pred:$p),
-                          "b${p:cc}lr ${p:reg}", BrB, 
-                          [(retflag)]>;
+  let isReturn = 1, Uses = [LR, RM] in
+    def BLR : XLForm_2_ext<19, 16, 20, 0, 0, (outs), (ins), "blr", BrB,
+                           [(retflag)]>;
   let isBranch = 1, isIndirectBranch = 1, Uses = [CTR] in
     def BCTR : XLForm_2_ext<19, 528, 20, 0, 0, (outs), (ins), "bctr", BrB, []>;
 }
@@ -453,46 +524,29 @@ let isBranch = 1, isTerminator = 1, hasCtrlDep = 1, PPC970_Unit = 7 in {
   }
 }
 
-// Darwin ABI Calls.
-let isCall = 1, PPC970_Unit = 7, Defs = [LR] in {
-  // Convenient aliases for call instructions
-  let Uses = [RM] in {
-    def BL_Darwin  : IForm<18, 0, 1,
-                           (outs), (ins calltarget:$func), 
-                           "bl $func", BrB, []>;  // See Pat patterns below.
-    def BLA_Darwin : IForm<18, 1, 1, 
-                          (outs), (ins aaddr:$func),
-                          "bla $func", BrB, [(PPCcall_Darwin (i32 imm:$func))]>;
-  }
-  let Uses = [CTR, RM] in {
-    def BCTRL_Darwin : XLForm_2_ext<19, 528, 20, 0, 1, 
-                                  (outs), (ins),
-                                  "bctrl", BrB,
-                                  [(PPCbctrl_Darwin)]>, Requires<[In32BitMode]>;
+// The unconditional BCL used by the SjLj setjmp code.
+let isCall = 1, hasCtrlDep = 1, isCodeGenOnly = 1, PPC970_Unit = 7 in {
+  let Defs = [LR], Uses = [RM] in {
+    def BCLalways  : BForm_2<16, 20, 31, 0, 1, (outs), (ins condbrtarget:$dst),
+                            "bcl 20, 31, $dst">;
   }
 }
 
-// SVR4 ABI Calls.
 let isCall = 1, PPC970_Unit = 7, Defs = [LR] in {
   // Convenient aliases for call instructions
   let Uses = [RM] in {
-    def BL_SVR4  : IForm<18, 0, 1,
-                        (outs), (ins calltarget:$func), 
-                        "bl $func", BrB, []>;  // See Pat patterns below.
-    def BLA_SVR4 : IForm<18, 1, 1,
-                        (outs), (ins aaddr:$func),
-                        "bla $func", BrB,
-                        [(PPCcall_SVR4 (i32 imm:$func))]>;
+    def BL  : IForm<18, 0, 1, (outs), (ins calltarget:$func),
+                    "bl $func", BrB, []>;  // See Pat patterns below.
+    def BLA : IForm<18, 1, 1, (outs), (ins aaddr:$func),
+                    "bla $func", BrB, [(PPCcall (i32 imm:$func))]>;
   }
   let Uses = [CTR, RM] in {
-    def BCTRL_SVR4 : XLForm_2_ext<19, 528, 20, 0, 1,
-                                (outs), (ins),
-                                "bctrl", BrB,
-                                [(PPCbctrl_SVR4)]>, Requires<[In32BitMode]>;
+    def BCTRL : XLForm_2_ext<19, 528, 20, 0, 1, (outs), (ins),
+                             "bctrl", BrB, [(PPCbctrl)]>,
+                Requires<[In32BitMode]>;
   }
 }
 
-
 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, Uses = [RM] in
 def TCRETURNdi :Pseudo< (outs),
                         (ins calltarget:$dst, i32imm:$offset),
@@ -511,6 +565,8 @@ def TCRETURNri : Pseudo<(outs), (ins CTRRC:$dst, i32imm:$offset),
                  []>;
 
 
+let isCodeGenOnly = 1 in {
+
 let isTerminator = 1, isBarrier = 1, PPC970_Unit = 7, isBranch = 1,
     isIndirectBranch = 1, isCall = 1, isReturn = 1, Uses = [CTR, RM]  in
 def TAILBCTR : XLForm_2_ext<19, 528, 20, 0, 0, (outs), (ins), "bctr", BrB, []>,
@@ -524,6 +580,7 @@ def TAILB   : IForm<18, 0, 0, (outs), (ins calltarget:$dst),
                   "b $dst", BrB,
                   []>;
 
+}
 
 let isBranch = 1, isTerminator = 1, hasCtrlDep = 1, PPC970_Unit = 7,
     isBarrier = 1, isCall = 1, isReturn = 1, Uses = [RM] in
@@ -531,6 +588,22 @@ def TAILBA   : IForm<18, 0, 0, (outs), (ins aaddr:$dst),
                   "ba $dst", BrB,
                   []>;
 
+let hasSideEffects = 1, isBarrier = 1, usesCustomInserter = 1 in {
+  def EH_SjLj_SetJmp32  : Pseudo<(outs GPRC:$dst), (ins memr:$buf),
+                            "#EH_SJLJ_SETJMP32",
+                            [(set i32:$dst, (PPCeh_sjlj_setjmp addr:$buf))]>,
+                          Requires<[In32BitMode]>;
+  let isTerminator = 1 in
+  def EH_SjLj_LongJmp32 : Pseudo<(outs), (ins memr:$buf),
+                            "#EH_SJLJ_LONGJMP32",
+                            [(PPCeh_sjlj_longjmp addr:$buf)]>,
+                          Requires<[In32BitMode]>;
+}
+
+let isBranch = 1, isTerminator = 1 in {
+  def EH_SjLj_Setup : Pseudo<(outs), (ins directbrtarget:$dst),
+                        "#EH_SjLj_Setup\t$dst", []>;
+}
 
 // DCB* instructions.
 def DCBA   : DCB_Form<758, 0, (outs), (ins memrr:$dst),
@@ -566,93 +639,90 @@ let usesCustomInserter = 1 in {
   let Defs = [CR0] in {
     def ATOMIC_LOAD_ADD_I8 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_ADD_I8",
-      [(set GPRC:$dst, (atomic_load_add_8 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_add_8 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_SUB_I8 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_SUB_I8",
-      [(set GPRC:$dst, (atomic_load_sub_8 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_sub_8 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_AND_I8 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_AND_I8",
-      [(set GPRC:$dst, (atomic_load_and_8 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_and_8 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_OR_I8 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_OR_I8",
-      [(set GPRC:$dst, (atomic_load_or_8 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_or_8 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_XOR_I8 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "ATOMIC_LOAD_XOR_I8",
-      [(set GPRC:$dst, (atomic_load_xor_8 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_xor_8 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_NAND_I8 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_NAND_I8",
-      [(set GPRC:$dst, (atomic_load_nand_8 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_nand_8 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_ADD_I16 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_ADD_I16",
-      [(set GPRC:$dst, (atomic_load_add_16 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_add_16 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_SUB_I16 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_SUB_I16",
-      [(set GPRC:$dst, (atomic_load_sub_16 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_sub_16 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_AND_I16 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_AND_I16",
-      [(set GPRC:$dst, (atomic_load_and_16 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_and_16 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_OR_I16 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_OR_I16",
-      [(set GPRC:$dst, (atomic_load_or_16 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_or_16 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_XOR_I16 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_XOR_I16",
-      [(set GPRC:$dst, (atomic_load_xor_16 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_xor_16 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_NAND_I16 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_NAND_I16",
-      [(set GPRC:$dst, (atomic_load_nand_16 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_nand_16 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_ADD_I32 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_ADD_I32",
-      [(set GPRC:$dst, (atomic_load_add_32 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_add_32 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_SUB_I32 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_SUB_I32",
-      [(set GPRC:$dst, (atomic_load_sub_32 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_sub_32 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_AND_I32 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_AND_I32",
-      [(set GPRC:$dst, (atomic_load_and_32 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_and_32 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_OR_I32 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_OR_I32",
-      [(set GPRC:$dst, (atomic_load_or_32 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_or_32 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_XOR_I32 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_XOR_I32",
-      [(set GPRC:$dst, (atomic_load_xor_32 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_xor_32 xoaddr:$ptr, i32:$incr))]>;
     def ATOMIC_LOAD_NAND_I32 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$incr), "#ATOMIC_LOAD_NAND_I32",
-      [(set GPRC:$dst, (atomic_load_nand_32 xoaddr:$ptr, GPRC:$incr))]>;
+      [(set i32:$dst, (atomic_load_nand_32 xoaddr:$ptr, i32:$incr))]>;
 
     def ATOMIC_CMP_SWAP_I8 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$old, GPRC:$new), "#ATOMIC_CMP_SWAP_I8",
-      [(set GPRC:$dst, 
-                    (atomic_cmp_swap_8 xoaddr:$ptr, GPRC:$old, GPRC:$new))]>;
+      [(set i32:$dst, (atomic_cmp_swap_8 xoaddr:$ptr, i32:$old, i32:$new))]>;
     def ATOMIC_CMP_SWAP_I16 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$old, GPRC:$new), "#ATOMIC_CMP_SWAP_I16 $dst $ptr $old $new",
-      [(set GPRC:$dst, 
-                    (atomic_cmp_swap_16 xoaddr:$ptr, GPRC:$old, GPRC:$new))]>;
+      [(set i32:$dst, (atomic_cmp_swap_16 xoaddr:$ptr, i32:$old, i32:$new))]>;
     def ATOMIC_CMP_SWAP_I32 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$old, GPRC:$new), "#ATOMIC_CMP_SWAP_I32 $dst $ptr $old $new",
-      [(set GPRC:$dst, 
-                    (atomic_cmp_swap_32 xoaddr:$ptr, GPRC:$old, GPRC:$new))]>;
+      [(set i32:$dst, (atomic_cmp_swap_32 xoaddr:$ptr, i32:$old, i32:$new))]>;
 
     def ATOMIC_SWAP_I8 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$new), "#ATOMIC_SWAP_i8",
-      [(set GPRC:$dst, (atomic_swap_8 xoaddr:$ptr, GPRC:$new))]>;
+      [(set i32:$dst, (atomic_swap_8 xoaddr:$ptr, i32:$new))]>;
     def ATOMIC_SWAP_I16 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$new), "#ATOMIC_SWAP_I16",
-      [(set GPRC:$dst, (atomic_swap_16 xoaddr:$ptr, GPRC:$new))]>;
+      [(set i32:$dst, (atomic_swap_16 xoaddr:$ptr, i32:$new))]>;
     def ATOMIC_SWAP_I32 : Pseudo<
       (outs GPRC:$dst), (ins memrr:$ptr, GPRC:$new), "#ATOMIC_SWAP_I32",
-      [(set GPRC:$dst, (atomic_swap_32 xoaddr:$ptr, GPRC:$new))]>;
+      [(set i32:$dst, (atomic_swap_32 xoaddr:$ptr, i32:$new))]>;
   }
 }
 
 // Instructions to support atomic operations
 def LWARX : XForm_1<31,  20, (outs GPRC:$rD), (ins memrr:$src),
                    "lwarx $rD, $src", LdStLWARX,
-                   [(set GPRC:$rD, (PPClarx xoaddr:$src))]>;
+                   [(set i32:$rD, (PPClarx xoaddr:$src))]>;
 
 let Defs = [CR0] in
 def STWCX : XForm_1<31, 150, (outs), (ins GPRC:$rS, memrr:$dst),
                    "stwcx. $rS, $dst", LdStSTWCX,
-                   [(PPCstcx GPRC:$rS, xoaddr:$dst)]>,
+                   [(PPCstcx i32:$rS, xoaddr:$dst)]>,
                    isDOT;
 
 let isTerminator = 1, isBarrier = 1, hasCtrlDep = 1 in
@@ -666,94 +736,94 @@ def TRAP  : XForm_24<31, 4, (outs), (ins), "trap", LdStLoad, [(trap)]>;
 let canFoldAsLoad = 1, PPC970_Unit = 2 in {
 def LBZ : DForm_1<34, (outs GPRC:$rD), (ins memri:$src),
                   "lbz $rD, $src", LdStLoad,
-                  [(set GPRC:$rD, (zextloadi8 iaddr:$src))]>;
+                  [(set i32:$rD, (zextloadi8 iaddr:$src))]>;
 def LHA : DForm_1<42, (outs GPRC:$rD), (ins memri:$src),
                   "lha $rD, $src", LdStLHA,
-                  [(set GPRC:$rD, (sextloadi16 iaddr:$src))]>,
+                  [(set i32:$rD, (sextloadi16 iaddr:$src))]>,
                   PPC970_DGroup_Cracked;
 def LHZ : DForm_1<40, (outs GPRC:$rD), (ins memri:$src),
                   "lhz $rD, $src", LdStLoad,
-                  [(set GPRC:$rD, (zextloadi16 iaddr:$src))]>;
+                  [(set i32:$rD, (zextloadi16 iaddr:$src))]>;
 def LWZ : DForm_1<32, (outs GPRC:$rD), (ins memri:$src),
                   "lwz $rD, $src", LdStLoad,
-                  [(set GPRC:$rD, (load iaddr:$src))]>;
+                  [(set i32:$rD, (load iaddr:$src))]>;
 
 def LFS : DForm_1<48, (outs F4RC:$rD), (ins memri:$src),
                   "lfs $rD, $src", LdStLFD,
-                  [(set F4RC:$rD, (load iaddr:$src))]>;
+                  [(set f32:$rD, (load iaddr:$src))]>;
 def LFD : DForm_1<50, (outs F8RC:$rD), (ins memri:$src),
                   "lfd $rD, $src", LdStLFD,
-                  [(set F8RC:$rD, (load iaddr:$src))]>;
+                  [(set f64:$rD, (load iaddr:$src))]>;
 
 
 // Unindexed (r+i) Loads with Update (preinc).
 let mayLoad = 1 in {
-def LBZU : DForm_1<35, (outs GPRC:$rD, ptr_rc:$ea_result), (ins memri:$addr),
+def LBZU : DForm_1<35, (outs GPRC:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
                    "lbzu $rD, $addr", LdStLoadUpd,
                    []>, RegConstraint<"$addr.reg = $ea_result">,
                    NoEncode<"$ea_result">;
 
-def LHAU : DForm_1<43, (outs GPRC:$rD, ptr_rc:$ea_result), (ins memri:$addr),
+def LHAU : DForm_1<43, (outs GPRC:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
                    "lhau $rD, $addr", LdStLHAU,
                    []>, RegConstraint<"$addr.reg = $ea_result">,
                    NoEncode<"$ea_result">;
 
-def LHZU : DForm_1<41, (outs GPRC:$rD, ptr_rc:$ea_result), (ins memri:$addr),
+def LHZU : DForm_1<41, (outs GPRC:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
                    "lhzu $rD, $addr", LdStLoadUpd,
                    []>, RegConstraint<"$addr.reg = $ea_result">,
                    NoEncode<"$ea_result">;
 
-def LWZU : DForm_1<33, (outs GPRC:$rD, ptr_rc:$ea_result), (ins memri:$addr),
+def LWZU : DForm_1<33, (outs GPRC:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
                    "lwzu $rD, $addr", LdStLoadUpd,
                    []>, RegConstraint<"$addr.reg = $ea_result">,
                    NoEncode<"$ea_result">;
 
-def LFSU : DForm_1<49, (outs F4RC:$rD, ptr_rc:$ea_result), (ins memri:$addr),
+def LFSU : DForm_1<49, (outs F4RC:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
                   "lfsu $rD, $addr", LdStLFDU,
                   []>, RegConstraint<"$addr.reg = $ea_result">,
                    NoEncode<"$ea_result">;
 
-def LFDU : DForm_1<51, (outs F8RC:$rD, ptr_rc:$ea_result), (ins memri:$addr),
+def LFDU : DForm_1<51, (outs F8RC:$rD, ptr_rc_nor0:$ea_result), (ins memri:$addr),
                   "lfdu $rD, $addr", LdStLFDU,
                   []>, RegConstraint<"$addr.reg = $ea_result">,
                    NoEncode<"$ea_result">;
 
 
 // Indexed (r+r) Loads with Update (preinc).
-def LBZUX : XForm_1<31, 119, (outs GPRC:$rD, ptr_rc:$ea_result),
+def LBZUX : XForm_1<31, 119, (outs GPRC:$rD, ptr_rc_nor0:$ea_result),
                    (ins memrr:$addr),
                    "lbzux $rD, $addr", LdStLoadUpd,
-                   []>, RegConstraint<"$addr.offreg = $ea_result">,
+                   []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                    NoEncode<"$ea_result">;
 
-def LHAUX : XForm_1<31, 375, (outs GPRC:$rD, ptr_rc:$ea_result),
+def LHAUX : XForm_1<31, 375, (outs GPRC:$rD, ptr_rc_nor0:$ea_result),
                    (ins memrr:$addr),
                    "lhaux $rD, $addr", LdStLHAU,
-                   []>, RegConstraint<"$addr.offreg = $ea_result">,
+                   []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                    NoEncode<"$ea_result">;
 
-def LHZUX : XForm_1<31, 311, (outs GPRC:$rD, ptr_rc:$ea_result),
+def LHZUX : XForm_1<31, 311, (outs GPRC:$rD, ptr_rc_nor0:$ea_result),
                    (ins memrr:$addr),
                    "lhzux $rD, $addr", LdStLoadUpd,
-                   []>, RegConstraint<"$addr.offreg = $ea_result">,
+                   []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                    NoEncode<"$ea_result">;
 
-def LWZUX : XForm_1<31, 55, (outs GPRC:$rD, ptr_rc:$ea_result),
+def LWZUX : XForm_1<31, 55, (outs GPRC:$rD, ptr_rc_nor0:$ea_result),
                    (ins memrr:$addr),
                    "lwzux $rD, $addr", LdStLoadUpd,
-                   []>, RegConstraint<"$addr.offreg = $ea_result">,
+                   []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                    NoEncode<"$ea_result">;
 
-def LFSUX : XForm_1<31, 567, (outs F4RC:$rD, ptr_rc:$ea_result),
+def LFSUX : XForm_1<31, 567, (outs F4RC:$rD, ptr_rc_nor0:$ea_result),
                    (ins memrr:$addr),
                    "lfsux $rD, $addr", LdStLFDU,
-                   []>, RegConstraint<"$addr.offreg = $ea_result">,
+                   []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                    NoEncode<"$ea_result">;
 
-def LFDUX : XForm_1<31, 631, (outs F8RC:$rD, ptr_rc:$ea_result),
+def LFDUX : XForm_1<31, 631, (outs F8RC:$rD, ptr_rc_nor0:$ea_result),
                    (ins memrr:$addr),
                    "lfdux $rD, $addr", LdStLFDU,
-                   []>, RegConstraint<"$addr.offreg = $ea_result">,
+                   []>, RegConstraint<"$addr.ptrreg = $ea_result">,
                    NoEncode<"$ea_result">;
 }
 }
@@ -763,32 +833,39 @@ def LFDUX : XForm_1<31, 631, (outs F8RC:$rD, ptr_rc:$ea_result),
 let canFoldAsLoad = 1, PPC970_Unit = 2 in {
 def LBZX : XForm_1<31,  87, (outs GPRC:$rD), (ins memrr:$src),
                    "lbzx $rD, $src", LdStLoad,
-                   [(set GPRC:$rD, (zextloadi8 xaddr:$src))]>;
+                   [(set i32:$rD, (zextloadi8 xaddr:$src))]>;
 def LHAX : XForm_1<31, 343, (outs GPRC:$rD), (ins memrr:$src),
                    "lhax $rD, $src", LdStLHA,
-                   [(set GPRC:$rD, (sextloadi16 xaddr:$src))]>,
+                   [(set i32:$rD, (sextloadi16 xaddr:$src))]>,
                    PPC970_DGroup_Cracked;
 def LHZX : XForm_1<31, 279, (outs GPRC:$rD), (ins memrr:$src),
                    "lhzx $rD, $src", LdStLoad,
-                   [(set GPRC:$rD, (zextloadi16 xaddr:$src))]>;
+                   [(set i32:$rD, (zextloadi16 xaddr:$src))]>;
 def LWZX : XForm_1<31,  23, (outs GPRC:$rD), (ins memrr:$src),
                    "lwzx $rD, $src", LdStLoad,
-                   [(set GPRC:$rD, (load xaddr:$src))]>;
+                   [(set i32:$rD, (load xaddr:$src))]>;
                    
                    
 def LHBRX : XForm_1<31, 790, (outs GPRC:$rD), (ins memrr:$src),
                    "lhbrx $rD, $src", LdStLoad,
-                   [(set GPRC:$rD, (PPClbrx xoaddr:$src, i16))]>;
+                   [(set i32:$rD, (PPClbrx xoaddr:$src, i16))]>;
 def LWBRX : XForm_1<31,  534, (outs GPRC:$rD), (ins memrr:$src),
                    "lwbrx $rD, $src", LdStLoad,
-                   [(set GPRC:$rD, (PPClbrx xoaddr:$src, i32))]>;
+                   [(set i32:$rD, (PPClbrx xoaddr:$src, i32))]>;
 
 def LFSX   : XForm_25<31, 535, (outs F4RC:$frD), (ins memrr:$src),
                       "lfsx $frD, $src", LdStLFD,
-                      [(set F4RC:$frD, (load xaddr:$src))]>;
+                      [(set f32:$frD, (load xaddr:$src))]>;
 def LFDX   : XForm_25<31, 599, (outs F8RC:$frD), (ins memrr:$src),
                       "lfdx $frD, $src", LdStLFD,
-                      [(set F8RC:$frD, (load xaddr:$src))]>;
+                      [(set f64:$frD, (load xaddr:$src))]>;
+
+def LFIWAX : XForm_25<31, 855, (outs F8RC:$frD), (ins memrr:$src),
+                      "lfiwax $frD, $src", LdStLFD,
+                      [(set f64:$frD, (PPClfiwax xoaddr:$src))]>;
+def LFIWZX : XForm_25<31, 887, (outs F8RC:$frD), (ins memrr:$src),
+                      "lfiwzx $frD, $src", LdStLFD,
+                      [(set f64:$frD, (PPClfiwzx xoaddr:$src))]>;
 }
 
 //===----------------------------------------------------------------------===//
@@ -799,137 +876,128 @@ def LFDX   : XForm_25<31, 599, (outs F8RC:$frD), (ins memrr:$src),
 let PPC970_Unit = 2 in {
 def STB  : DForm_1<38, (outs), (ins GPRC:$rS, memri:$src),
                    "stb $rS, $src", LdStStore,
-                   [(truncstorei8 GPRC:$rS, iaddr:$src)]>;
+                   [(truncstorei8 i32:$rS, iaddr:$src)]>;
 def STH  : DForm_1<44, (outs), (ins GPRC:$rS, memri:$src),
                    "sth $rS, $src", LdStStore,
-                   [(truncstorei16 GPRC:$rS, iaddr:$src)]>;
+                   [(truncstorei16 i32:$rS, iaddr:$src)]>;
 def STW  : DForm_1<36, (outs), (ins GPRC:$rS, memri:$src),
                    "stw $rS, $src", LdStStore,
-                   [(store GPRC:$rS, iaddr:$src)]>;
+                   [(store i32:$rS, iaddr:$src)]>;
 def STFS : DForm_1<52, (outs), (ins F4RC:$rS, memri:$dst),
                    "stfs $rS, $dst", LdStSTFD,
-                   [(store F4RC:$rS, iaddr:$dst)]>;
+                   [(store f32:$rS, iaddr:$dst)]>;
 def STFD : DForm_1<54, (outs), (ins F8RC:$rS, memri:$dst),
                    "stfd $rS, $dst", LdStSTFD,
-                   [(store F8RC:$rS, iaddr:$dst)]>;
+                   [(store f64:$rS, iaddr:$dst)]>;
 }
 
 // Unindexed (r+i) Stores with Update (preinc).
-let PPC970_Unit = 2 in {
-def STBU  : DForm_1a<39, (outs ptr_rc:$ea_res), (ins GPRC:$rS,
-                             symbolLo:$ptroff, ptr_rc:$ptrreg),
-                    "stbu $rS, $ptroff($ptrreg)", LdStStoreUpd,
-                    [(set ptr_rc:$ea_res,
-                          (pre_truncsti8 GPRC:$rS, ptr_rc:$ptrreg, 
-                                         iaddroff:$ptroff))]>,
-                    RegConstraint<"$ptrreg = $ea_res">, NoEncode<"$ea_res">;
-def STHU  : DForm_1a<45, (outs ptr_rc:$ea_res), (ins GPRC:$rS,
-                             symbolLo:$ptroff, ptr_rc:$ptrreg),
-                    "sthu $rS, $ptroff($ptrreg)", LdStStoreUpd,
-                    [(set ptr_rc:$ea_res,
-                        (pre_truncsti16 GPRC:$rS, ptr_rc:$ptrreg, 
-                                        iaddroff:$ptroff))]>,
-                    RegConstraint<"$ptrreg = $ea_res">, NoEncode<"$ea_res">;
-def STWU  : DForm_1a<37, (outs ptr_rc:$ea_res), (ins GPRC:$rS,
-                             symbolLo:$ptroff, ptr_rc:$ptrreg),
-                    "stwu $rS, $ptroff($ptrreg)", LdStStoreUpd,
-                    [(set ptr_rc:$ea_res, (pre_store GPRC:$rS, ptr_rc:$ptrreg, 
-                                                     iaddroff:$ptroff))]>,
-                    RegConstraint<"$ptrreg = $ea_res">, NoEncode<"$ea_res">;
-def STFSU : DForm_1a<37, (outs ptr_rc:$ea_res), (ins F4RC:$rS,
-                             symbolLo:$ptroff, ptr_rc:$ptrreg),
-                    "stfsu $rS, $ptroff($ptrreg)", LdStSTFDU,
-                    [(set ptr_rc:$ea_res, (pre_store F4RC:$rS,  ptr_rc:$ptrreg, 
-                                          iaddroff:$ptroff))]>,
-                    RegConstraint<"$ptrreg = $ea_res">, NoEncode<"$ea_res">;
-def STFDU : DForm_1a<37, (outs ptr_rc:$ea_res), (ins F8RC:$rS,
-                             symbolLo:$ptroff, ptr_rc:$ptrreg),
-                    "stfdu $rS, $ptroff($ptrreg)", LdStSTFDU,
-                    [(set ptr_rc:$ea_res, (pre_store F8RC:$rS, ptr_rc:$ptrreg, 
-                                          iaddroff:$ptroff))]>,
-                    RegConstraint<"$ptrreg = $ea_res">, NoEncode<"$ea_res">;
+let PPC970_Unit = 2, mayStore = 1 in {
+def STBU  : DForm_1<39, (outs ptr_rc_nor0:$ea_res), (ins GPRC:$rS, memri:$dst),
+                    "stbu $rS, $dst", LdStStoreUpd, []>,
+                    RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
+def STHU  : DForm_1<45, (outs ptr_rc_nor0:$ea_res), (ins GPRC:$rS, memri:$dst),
+                    "sthu $rS, $dst", LdStStoreUpd, []>,
+                    RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
+def STWU  : DForm_1<37, (outs ptr_rc_nor0:$ea_res), (ins GPRC:$rS, memri:$dst),
+                    "stwu $rS, $dst", LdStStoreUpd, []>,
+                    RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
+def STFSU : DForm_1<37, (outs ptr_rc_nor0:$ea_res), (ins F4RC:$rS, memri:$dst),
+                    "stfsu $rS, $dst", LdStSTFDU, []>,
+                    RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
+def STFDU : DForm_1<37, (outs ptr_rc_nor0:$ea_res), (ins F8RC:$rS, memri:$dst),
+                    "stfdu $rS, $dst", LdStSTFDU, []>,
+                    RegConstraint<"$dst.reg = $ea_res">, NoEncode<"$ea_res">;
 }
 
+// Patterns to match the pre-inc stores.  We can't put the patterns on
+// the instruction definitions directly as ISel wants the address base
+// and offset to be separate operands, not a single complex operand.
+def : Pat<(pre_truncsti8 i32:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
+          (STBU $rS, iaddroff:$ptroff, $ptrreg)>;
+def : Pat<(pre_truncsti16 i32:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
+          (STHU $rS, iaddroff:$ptroff, $ptrreg)>;
+def : Pat<(pre_store i32:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
+          (STWU $rS, iaddroff:$ptroff, $ptrreg)>;
+def : Pat<(pre_store f32:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
+          (STFSU $rS, iaddroff:$ptroff, $ptrreg)>;
+def : Pat<(pre_store f64:$rS, iPTR:$ptrreg, iaddroff:$ptroff),
+          (STFDU $rS, iaddroff:$ptroff, $ptrreg)>;
 
 // Indexed (r+r) Stores.
-//
 let PPC970_Unit = 2 in {
 def STBX  : XForm_8<31, 215, (outs), (ins GPRC:$rS, memrr:$dst),
                    "stbx $rS, $dst", LdStStore,
-                   [(truncstorei8 GPRC:$rS, xaddr:$dst)]>, 
+                   [(truncstorei8 i32:$rS, xaddr:$dst)]>,
                    PPC970_DGroup_Cracked;
 def STHX  : XForm_8<31, 407, (outs), (ins GPRC:$rS, memrr:$dst),
                    "sthx $rS, $dst", LdStStore,
-                   [(truncstorei16 GPRC:$rS, xaddr:$dst)]>, 
+                   [(truncstorei16 i32:$rS, xaddr:$dst)]>,
                    PPC970_DGroup_Cracked;
 def STWX  : XForm_8<31, 151, (outs), (ins GPRC:$rS, memrr:$dst),
                    "stwx $rS, $dst", LdStStore,
-                   [(store GPRC:$rS, xaddr:$dst)]>,
-                   PPC970_DGroup_Cracked;
- 
-def STBUX : XForm_8<31, 247, (outs ptr_rc:$ea_res),
-                             (ins GPRC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                   "stbux $rS, $ptroff, $ptrreg", LdStStoreUpd,
-                   [(set ptr_rc:$ea_res,
-                      (pre_truncsti8 GPRC:$rS,
-                                     ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
-                   RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
+                   [(store i32:$rS, xaddr:$dst)]>,
                    PPC970_DGroup_Cracked;
  
-def STHUX : XForm_8<31, 439, (outs ptr_rc:$ea_res),
-                             (ins GPRC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                   "sthux $rS, $ptroff, $ptrreg", LdStStoreUpd,
-                   [(set ptr_rc:$ea_res,
-                      (pre_truncsti16 GPRC:$rS,
-                                      ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
-                   RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
-                   PPC970_DGroup_Cracked;
-                 
-def STWUX : XForm_8<31, 183, (outs ptr_rc:$ea_res),
-                             (ins GPRC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                   "stwux $rS, $ptroff, $ptrreg", LdStStoreUpd,
-                   [(set ptr_rc:$ea_res,
-                      (pre_store GPRC:$rS, ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
-                   RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
-                   PPC970_DGroup_Cracked;
-
-def STFSUX : XForm_8<31, 695, (outs ptr_rc:$ea_res),
-                              (ins F4RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                    "stfsux $rS, $ptroff, $ptrreg", LdStSTFDU,
-                    [(set ptr_rc:$ea_res,
-                       (pre_store F4RC:$rS, ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
-                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
-                    PPC970_DGroup_Cracked;
-
-def STFDUX : XForm_8<31, 759, (outs ptr_rc:$ea_res),
-                              (ins F8RC:$rS, ptr_rc:$ptroff, ptr_rc:$ptrreg),
-                    "stfdux $rS, $ptroff, $ptrreg", LdStSTFDU,
-                    [(set ptr_rc:$ea_res,
-                       (pre_store F8RC:$rS, ptr_rc:$ptrreg, xaddroff:$ptroff))]>,
-                    RegConstraint<"$ptroff = $ea_res">, NoEncode<"$ea_res">,
-                    PPC970_DGroup_Cracked;
-
 def STHBRX: XForm_8<31, 918, (outs), (ins GPRC:$rS, memrr:$dst),
                    "sthbrx $rS, $dst", LdStStore,
-                   [(PPCstbrx GPRC:$rS, xoaddr:$dst, i16)]>, 
+                   [(PPCstbrx i32:$rS, xoaddr:$dst, i16)]>,
                    PPC970_DGroup_Cracked;
 def STWBRX: XForm_8<31, 662, (outs), (ins GPRC:$rS, memrr:$dst),
                    "stwbrx $rS, $dst", LdStStore,
-                   [(PPCstbrx GPRC:$rS, xoaddr:$dst, i32)]>,
+                   [(PPCstbrx i32:$rS, xoaddr:$dst, i32)]>,
                    PPC970_DGroup_Cracked;
 
 def STFIWX: XForm_28<31, 983, (outs), (ins F8RC:$frS, memrr:$dst),
                      "stfiwx $frS, $dst", LdStSTFD,
-                     [(PPCstfiwx F8RC:$frS, xoaddr:$dst)]>;
+                     [(PPCstfiwx f64:$frS, xoaddr:$dst)]>;
                      
 def STFSX : XForm_28<31, 663, (outs), (ins F4RC:$frS, memrr:$dst),
                      "stfsx $frS, $dst", LdStSTFD,
-                     [(store F4RC:$frS, xaddr:$dst)]>;
+                     [(store f32:$frS, xaddr:$dst)]>;
 def STFDX : XForm_28<31, 727, (outs), (ins F8RC:$frS, memrr:$dst),
                      "stfdx $frS, $dst", LdStSTFD,
-                     [(store F8RC:$frS, xaddr:$dst)]>;
+                     [(store f64:$frS, xaddr:$dst)]>;
+}
+
+// Indexed (r+r) Stores with Update (preinc).
+let PPC970_Unit = 2, mayStore = 1 in {
+def STBUX : XForm_8<31, 247, (outs ptr_rc_nor0:$ea_res), (ins GPRC:$rS, memrr:$dst),
+                    "stbux $rS, $dst", LdStStoreUpd, []>,
+                    RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked;
+def STHUX : XForm_8<31, 439, (outs ptr_rc_nor0:$ea_res), (ins GPRC:$rS, memrr:$dst),
+                    "sthux $rS, $dst", LdStStoreUpd, []>,
+                    RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked;
+def STWUX : XForm_8<31, 183, (outs ptr_rc_nor0:$ea_res), (ins GPRC:$rS, memrr:$dst),
+                    "stwux $rS, $dst", LdStStoreUpd, []>,
+                    RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked;
+def STFSUX: XForm_8<31, 695, (outs ptr_rc_nor0:$ea_res), (ins F4RC:$rS, memrr:$dst),
+                    "stfsux $rS, $dst", LdStSTFDU, []>,
+                    RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked;
+def STFDUX: XForm_8<31, 759, (outs ptr_rc_nor0:$ea_res), (ins F8RC:$rS, memrr:$dst),
+                    "stfdux $rS, $dst", LdStSTFDU, []>,
+                    RegConstraint<"$dst.ptrreg = $ea_res">, NoEncode<"$ea_res">,
+                    PPC970_DGroup_Cracked;
 }
 
+// Patterns to match the pre-inc stores.  We can't put the patterns on
+// the instruction definitions directly as ISel wants the address base
+// and offset to be separate operands, not a single complex operand.
+def : Pat<(pre_truncsti8 i32:$rS, iPTR:$ptrreg, iPTR:$ptroff),
+          (STBUX $rS, $ptrreg, $ptroff)>;
+def : Pat<(pre_truncsti16 i32:$rS, iPTR:$ptrreg, iPTR:$ptroff),
+          (STHUX $rS, $ptrreg, $ptroff)>;
+def : Pat<(pre_store i32:$rS, iPTR:$ptrreg, iPTR:$ptroff),
+          (STWUX $rS, $ptrreg, $ptroff)>;
+def : Pat<(pre_store f32:$rS, iPTR:$ptrreg, iPTR:$ptroff),
+          (STFSUX $rS, $ptrreg, $ptroff)>;
+def : Pat<(pre_store f64:$rS, iPTR:$ptrreg, iPTR:$ptroff),
+          (STFDUX $rS, $ptrreg, $ptroff)>;
+
 def SYNC : XForm_24_sync<31, 598, (outs), (ins),
                         "sync", LdStSync,
                         [(int_ppc_sync)]>;
@@ -939,68 +1007,66 @@ def SYNC : XForm_24_sync<31, 598, (outs), (ins),
 //
 
 let PPC970_Unit = 1 in {  // FXU Operations.
-def ADDI   : DForm_2<14, (outs GPRC:$rD), (ins GPRC:$rA, s16imm:$imm),
-                     "addi $rD, $rA, $imm", IntSimple,
-                     [(set GPRC:$rD, (add GPRC:$rA, immSExt16:$imm))]>;
-def ADDIL  : DForm_2<14, (outs GPRC:$rD), (ins GPRC:$rA, symbolLo:$imm),
+def ADDI   : DForm_2<14, (outs GPRC:$rD), (ins GPRC_NOR0:$rA, symbolLo:$imm),
                      "addi $rD, $rA, $imm", IntSimple,
-                     [(set GPRC:$rD, (add GPRC:$rA, immSExt16:$imm))]>;
+                     [(set i32:$rD, (add i32:$rA, immSExt16:$imm))]>;
 let Defs = [CARRY] in {
 def ADDIC  : DForm_2<12, (outs GPRC:$rD), (ins GPRC:$rA, s16imm:$imm),
                      "addic $rD, $rA, $imm", IntGeneral,
-                     [(set GPRC:$rD, (addc GPRC:$rA, immSExt16:$imm))]>,
+                     [(set i32:$rD, (addc i32:$rA, immSExt16:$imm))]>,
                      PPC970_DGroup_Cracked;
 def ADDICo : DForm_2<13, (outs GPRC:$rD), (ins GPRC:$rA, s16imm:$imm),
                      "addic. $rD, $rA, $imm", IntGeneral,
                      []>;
 }
-def ADDIS  : DForm_2<15, (outs GPRC:$rD), (ins GPRC:$rA, symbolHi:$imm),
+def ADDIS  : DForm_2<15, (outs GPRC:$rD), (ins GPRC_NOR0:$rA, symbolHi:$imm),
                      "addis $rD, $rA, $imm", IntSimple,
-                     [(set GPRC:$rD, (add GPRC:$rA, imm16ShiftedSExt:$imm))]>;
-def LA     : DForm_2<14, (outs GPRC:$rD), (ins GPRC:$rA, symbolLo:$sym),
+                     [(set i32:$rD, (add i32:$rA, imm16ShiftedSExt:$imm))]>;
+let isCodeGenOnly = 1 in
+def LA     : DForm_2<14, (outs GPRC:$rD), (ins GPRC_NOR0:$rA, symbolLo:$sym),
                      "la $rD, $sym($rA)", IntGeneral,
-                     [(set GPRC:$rD, (add GPRC:$rA,
+                     [(set i32:$rD, (add i32:$rA,
                                           (PPClo tglobaladdr:$sym, 0)))]>;
 def MULLI  : DForm_2< 7, (outs GPRC:$rD), (ins GPRC:$rA, s16imm:$imm),
                      "mulli $rD, $rA, $imm", IntMulLI,
-                     [(set GPRC:$rD, (mul GPRC:$rA, immSExt16:$imm))]>;
+                     [(set i32:$rD, (mul i32:$rA, immSExt16:$imm))]>;
 let Defs = [CARRY] in {
 def SUBFIC : DForm_2< 8, (outs GPRC:$rD), (ins GPRC:$rA, s16imm:$imm),
                      "subfic $rD, $rA, $imm", IntGeneral,
-                     [(set GPRC:$rD, (subc immSExt16:$imm, GPRC:$rA))]>;
+                     [(set i32:$rD, (subc immSExt16:$imm, i32:$rA))]>;
 }
 
 let isReMaterializable = 1, isAsCheapAsAMove = 1, isMoveImm = 1 in {
   def LI  : DForm_2_r0<14, (outs GPRC:$rD), (ins symbolLo:$imm),
                        "li $rD, $imm", IntSimple,
-                       [(set GPRC:$rD, immSExt16:$imm)]>;
+                       [(set i32:$rD, immSExt16:$imm)]>;
   def LIS : DForm_2_r0<15, (outs GPRC:$rD), (ins symbolHi:$imm),
                        "lis $rD, $imm", IntSimple,
-                       [(set GPRC:$rD, imm16ShiftedSExt:$imm)]>;
+                       [(set i32:$rD, imm16ShiftedSExt:$imm)]>;
 }
 }
 
 let PPC970_Unit = 1 in {  // FXU Operations.
 def ANDIo : DForm_4<28, (outs GPRC:$dst), (ins GPRC:$src1, u16imm:$src2),
                     "andi. $dst, $src1, $src2", IntGeneral,
-                    [(set GPRC:$dst, (and GPRC:$src1, immZExt16:$src2))]>,
+                    [(set i32:$dst, (and i32:$src1, immZExt16:$src2))]>,
                     isDOT;
 def ANDISo : DForm_4<29, (outs GPRC:$dst), (ins GPRC:$src1, u16imm:$src2),
                     "andis. $dst, $src1, $src2", IntGeneral,
-                    [(set GPRC:$dst, (and GPRC:$src1,imm16ShiftedZExt:$src2))]>,
+                    [(set i32:$dst, (and i32:$src1, imm16ShiftedZExt:$src2))]>,
                     isDOT;
 def ORI   : DForm_4<24, (outs GPRC:$dst), (ins GPRC:$src1, u16imm:$src2),
                     "ori $dst, $src1, $src2", IntSimple,
-                    [(set GPRC:$dst, (or GPRC:$src1, immZExt16:$src2))]>;
+                    [(set i32:$dst, (or i32:$src1, immZExt16:$src2))]>;
 def ORIS  : DForm_4<25, (outs GPRC:$dst), (ins GPRC:$src1, u16imm:$src2),
                     "oris $dst, $src1, $src2", IntSimple,
-                    [(set GPRC:$dst, (or GPRC:$src1, imm16ShiftedZExt:$src2))]>;
+                    [(set i32:$dst, (or i32:$src1, imm16ShiftedZExt:$src2))]>;
 def XORI  : DForm_4<26, (outs GPRC:$dst), (ins GPRC:$src1, u16imm:$src2),
                     "xori $dst, $src1, $src2", IntSimple,
-                    [(set GPRC:$dst, (xor GPRC:$src1, immZExt16:$src2))]>;
+                    [(set i32:$dst, (xor i32:$src1, immZExt16:$src2))]>;
 def XORIS : DForm_4<27, (outs GPRC:$dst), (ins GPRC:$src1, u16imm:$src2),
                     "xoris $dst, $src1, $src2", IntSimple,
-                    [(set GPRC:$dst, (xor GPRC:$src1,imm16ShiftedZExt:$src2))]>;
+                    [(set i32:$dst, (xor i32:$src1, imm16ShiftedZExt:$src2))]>;
 def NOP   : DForm_4_zero<24, (outs), (ins), "nop", IntSimple,
                          []>;
 def CMPWI : DForm_5_ext<11, (outs CRRC:$crD), (ins GPRC:$rA, s16imm:$imm),
@@ -1013,38 +1079,38 @@ def CMPLWI : DForm_6_ext<10, (outs CRRC:$dst), (ins GPRC:$src1, u16imm:$src2),
 let PPC970_Unit = 1 in {  // FXU Operations.
 def NAND : XForm_6<31, 476, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "nand $rA, $rS, $rB", IntSimple,
-                   [(set GPRC:$rA, (not (and GPRC:$rS, GPRC:$rB)))]>;
+                   [(set i32:$rA, (not (and i32:$rS, i32:$rB)))]>;
 def AND  : XForm_6<31,  28, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "and $rA, $rS, $rB", IntSimple,
-                   [(set GPRC:$rA, (and GPRC:$rS, GPRC:$rB))]>;
+                   [(set i32:$rA, (and i32:$rS, i32:$rB))]>;
 def ANDC : XForm_6<31,  60, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "andc $rA, $rS, $rB", IntSimple,
-                   [(set GPRC:$rA, (and GPRC:$rS, (not GPRC:$rB)))]>;
+                   [(set i32:$rA, (and i32:$rS, (not i32:$rB)))]>;
 def OR   : XForm_6<31, 444, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "or $rA, $rS, $rB", IntSimple,
-                   [(set GPRC:$rA, (or GPRC:$rS, GPRC:$rB))]>;
+                   [(set i32:$rA, (or i32:$rS, i32:$rB))]>;
 def NOR  : XForm_6<31, 124, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "nor $rA, $rS, $rB", IntSimple,
-                   [(set GPRC:$rA, (not (or GPRC:$rS, GPRC:$rB)))]>;
+                   [(set i32:$rA, (not (or i32:$rS, i32:$rB)))]>;
 def ORC  : XForm_6<31, 412, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "orc $rA, $rS, $rB", IntSimple,
-                   [(set GPRC:$rA, (or GPRC:$rS, (not GPRC:$rB)))]>;
+                   [(set i32:$rA, (or i32:$rS, (not i32:$rB)))]>;
 def EQV  : XForm_6<31, 284, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "eqv $rA, $rS, $rB", IntSimple,
-                   [(set GPRC:$rA, (not (xor GPRC:$rS, GPRC:$rB)))]>;
+                   [(set i32:$rA, (not (xor i32:$rS, i32:$rB)))]>;
 def XOR  : XForm_6<31, 316, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "xor $rA, $rS, $rB", IntSimple,
-                   [(set GPRC:$rA, (xor GPRC:$rS, GPRC:$rB))]>;
+                   [(set i32:$rA, (xor i32:$rS, i32:$rB))]>;
 def SLW  : XForm_6<31,  24, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "slw $rA, $rS, $rB", IntGeneral,
-                   [(set GPRC:$rA, (PPCshl GPRC:$rS, GPRC:$rB))]>;
+                   [(set i32:$rA, (PPCshl i32:$rS, i32:$rB))]>;
 def SRW  : XForm_6<31, 536, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "srw $rA, $rS, $rB", IntGeneral,
-                   [(set GPRC:$rA, (PPCsrl GPRC:$rS, GPRC:$rB))]>;
+                   [(set i32:$rA, (PPCsrl i32:$rS, i32:$rB))]>;
 let Defs = [CARRY] in {
 def SRAW : XForm_6<31, 792, (outs GPRC:$rA), (ins GPRC:$rS, GPRC:$rB),
                    "sraw $rA, $rS, $rB", IntShift,
-                   [(set GPRC:$rA, (PPCsra GPRC:$rS, GPRC:$rB))]>;
+                   [(set i32:$rA, (PPCsra i32:$rS, i32:$rB))]>;
 }
 }
 
@@ -1052,17 +1118,17 @@ let PPC970_Unit = 1 in {  // FXU Operations.
 let Defs = [CARRY] in {
 def SRAWI : XForm_10<31, 824, (outs GPRC:$rA), (ins GPRC:$rS, u5imm:$SH), 
                      "srawi $rA, $rS, $SH", IntShift,
-                     [(set GPRC:$rA, (sra GPRC:$rS, (i32 imm:$SH)))]>;
+                     [(set i32:$rA, (sra i32:$rS, (i32 imm:$SH)))]>;
 }
 def CNTLZW : XForm_11<31,  26, (outs GPRC:$rA), (ins GPRC:$rS),
                       "cntlzw $rA, $rS", IntGeneral,
-                      [(set GPRC:$rA, (ctlz GPRC:$rS))]>;
+                      [(set i32:$rA, (ctlz i32:$rS))]>;
 def EXTSB  : XForm_11<31, 954, (outs GPRC:$rA), (ins GPRC:$rS),
                       "extsb $rA, $rS", IntSimple,
-                      [(set GPRC:$rA, (sext_inreg GPRC:$rS, i8))]>;
+                      [(set i32:$rA, (sext_inreg i32:$rS, i8))]>;
 def EXTSH  : XForm_11<31, 922, (outs GPRC:$rA), (ins GPRC:$rS),
                       "extsh $rA, $rS", IntSimple,
-                      [(set GPRC:$rA, (sext_inreg GPRC:$rS, i16))]>;
+                      [(set i32:$rA, (sext_inreg i32:$rS, i16))]>;
 
 def CMPW   : XForm_16_ext<31, 0, (outs CRRC:$crD), (ins GPRC:$rA, GPRC:$rB),
                           "cmpw $crD, $rA, $rB", IntCompare>;
@@ -1080,16 +1146,54 @@ def FCMPUD : XForm_17<63, 0, (outs CRRC:$crD), (ins F8RC:$fA, F8RC:$fB),
 let Uses = [RM] in {
   def FCTIWZ : XForm_26<63, 15, (outs F8RC:$frD), (ins F8RC:$frB),
                         "fctiwz $frD, $frB", FPGeneral,
-                        [(set F8RC:$frD, (PPCfctiwz F8RC:$frB))]>;
+                        [(set f64:$frD, (PPCfctiwz f64:$frB))]>;
+
   def FRSP   : XForm_26<63, 12, (outs F4RC:$frD), (ins F8RC:$frB),
                         "frsp $frD, $frB", FPGeneral,
-                        [(set F4RC:$frD, (fround F8RC:$frB))]>;
+                        [(set f32:$frD, (fround f64:$frB))]>;
+
+  // The frin -> nearbyint mapping is valid only in fast-math mode.
+  def FRIND  : XForm_26<63, 392, (outs F8RC:$frD), (ins F8RC:$frB),
+                        "frin $frD, $frB", FPGeneral,
+                        [(set f64:$frD, (fnearbyint f64:$frB))]>;
+  def FRINS  : XForm_26<63, 392, (outs F4RC:$frD), (ins F4RC:$frB),
+                        "frin $frD, $frB", FPGeneral,
+                        [(set f32:$frD, (fnearbyint f32:$frB))]>;
+
+  // These pseudos expand to rint but also set FE_INEXACT when the result does
+  // not equal the argument.
+  let usesCustomInserter = 1, Defs = [RM] in { // FIXME: Model FPSCR!
+    def FRINDrint : Pseudo<(outs F8RC:$frD), (ins F8RC:$frB),
+                            "#FRINDrint", [(set f64:$frD, (frint f64:$frB))]>;
+    def FRINSrint : Pseudo<(outs F4RC:$frD), (ins F4RC:$frB),
+                            "#FRINSrint", [(set f32:$frD, (frint f32:$frB))]>;
+  }
+
+  def FRIPD  : XForm_26<63, 456, (outs F8RC:$frD), (ins F8RC:$frB),
+                        "frip $frD, $frB", FPGeneral,
+                        [(set f64:$frD, (fceil f64:$frB))]>;
+  def FRIPS  : XForm_26<63, 456, (outs F4RC:$frD), (ins F4RC:$frB),
+                        "frip $frD, $frB", FPGeneral,
+                        [(set f32:$frD, (fceil f32:$frB))]>;
+  def FRIZD  : XForm_26<63, 424, (outs F8RC:$frD), (ins F8RC:$frB),
+                        "friz $frD, $frB", FPGeneral,
+                        [(set f64:$frD, (ftrunc f64:$frB))]>;
+  def FRIZS  : XForm_26<63, 424, (outs F4RC:$frD), (ins F4RC:$frB),
+                        "friz $frD, $frB", FPGeneral,
+                        [(set f32:$frD, (ftrunc f32:$frB))]>;
+  def FRIMD  : XForm_26<63, 488, (outs F8RC:$frD), (ins F8RC:$frB),
+                        "frim $frD, $frB", FPGeneral,
+                        [(set f64:$frD, (ffloor f64:$frB))]>;
+  def FRIMS  : XForm_26<63, 488, (outs F4RC:$frD), (ins F4RC:$frB),
+                        "frim $frD, $frB", FPGeneral,
+                        [(set f32:$frD, (ffloor f32:$frB))]>;
+
   def FSQRT  : XForm_26<63, 22, (outs F8RC:$frD), (ins F8RC:$frB),
                         "fsqrt $frD, $frB", FPSqrt,
-                        [(set F8RC:$frD, (fsqrt F8RC:$frB))]>;
+                        [(set f64:$frD, (fsqrt f64:$frB))]>;
   def FSQRTS : XForm_26<59, 22, (outs F4RC:$frD), (ins F4RC:$frB),
                         "fsqrts $frD, $frB", FPSqrt,
-                        [(set F4RC:$frD, (fsqrt F4RC:$frB))]>;
+                        [(set f32:$frD, (fsqrt f32:$frB))]>;
   }
 }
 
@@ -1099,31 +1203,44 @@ let Uses = [RM] in {
 /// sneak into a d-group with a store).
 def FMR   : XForm_26<63, 72, (outs F4RC:$frD), (ins F4RC:$frB),
                      "fmr $frD, $frB", FPGeneral,
-                     []>,  // (set F4RC:$frD, F4RC:$frB)
+                     []>,  // (set f32:$frD, f32:$frB)
                      PPC970_Unit_Pseudo;
 
 let PPC970_Unit = 3 in {  // FPU Operations.
 // These are artificially split into two different forms, for 4/8 byte FP.
 def FABSS  : XForm_26<63, 264, (outs F4RC:$frD), (ins F4RC:$frB),
                       "fabs $frD, $frB", FPGeneral,
-                      [(set F4RC:$frD, (fabs F4RC:$frB))]>;
+                      [(set f32:$frD, (fabs f32:$frB))]>;
 def FABSD  : XForm_26<63, 264, (outs F8RC:$frD), (ins F8RC:$frB),
                       "fabs $frD, $frB", FPGeneral,
-                      [(set F8RC:$frD, (fabs F8RC:$frB))]>;
+                      [(set f64:$frD, (fabs f64:$frB))]>;
 def FNABSS : XForm_26<63, 136, (outs F4RC:$frD), (ins F4RC:$frB),
                       "fnabs $frD, $frB", FPGeneral,
-                      [(set F4RC:$frD, (fneg (fabs F4RC:$frB)))]>;
+                      [(set f32:$frD, (fneg (fabs f32:$frB)))]>;
 def FNABSD : XForm_26<63, 136, (outs F8RC:$frD), (ins F8RC:$frB),
                       "fnabs $frD, $frB", FPGeneral,
-                      [(set F8RC:$frD, (fneg (fabs F8RC:$frB)))]>;
+                      [(set f64:$frD, (fneg (fabs f64:$frB)))]>;
 def FNEGS  : XForm_26<63, 40, (outs F4RC:$frD), (ins F4RC:$frB),
                       "fneg $frD, $frB", FPGeneral,
-                      [(set F4RC:$frD, (fneg F4RC:$frB))]>;
+                      [(set f32:$frD, (fneg f32:$frB))]>;
 def FNEGD  : XForm_26<63, 40, (outs F8RC:$frD), (ins F8RC:$frB),
                       "fneg $frD, $frB", FPGeneral,
-                      [(set F8RC:$frD, (fneg F8RC:$frB))]>;
+                      [(set f64:$frD, (fneg f64:$frB))]>;
+
+// Reciprocal estimates.
+def FRE      : XForm_26<63, 24, (outs F8RC:$frD), (ins F8RC:$frB),
+                        "fre $frD, $frB", FPGeneral,
+                        [(set f64:$frD, (PPCfre f64:$frB))]>;
+def FRES     : XForm_26<59, 24, (outs F4RC:$frD), (ins F4RC:$frB),
+                        "fres $frD, $frB", FPGeneral,
+                        [(set f32:$frD, (PPCfre f32:$frB))]>;
+def FRSQRTE  : XForm_26<63, 26, (outs F8RC:$frD), (ins F8RC:$frB),
+                        "frsqrte $frD, $frB", FPGeneral,
+                        [(set f64:$frD, (PPCfrsqrte f64:$frB))]>;
+def FRSQRTES : XForm_26<59, 26, (outs F4RC:$frD), (ins F4RC:$frB),
+                        "frsqrtes $frD, $frB", FPGeneral,
+                        [(set f32:$frD, (PPCfrsqrte f32:$frB))]>;
 }
-                      
 
 // XL-Form instructions.  condition register logical ops.
 //
@@ -1141,6 +1258,7 @@ def CROR  : XLForm_1<19, 449, (outs CRBITRC:$CRD),
                       "cror $CRD, $CRA, $CRB", BrCR,
                       []>;
 
+let isCodeGenOnly = 1 in {
 def CRSET  : XLForm_1_ext<19, 289, (outs CRBITRC:$dst), (ins),
               "creqv $dst, $dst, $dst", BrCR,
               []>;
@@ -1158,6 +1276,7 @@ def CR6UNSET: XLForm_1_ext<19, 193, (outs), (ins),
               "crxor 6, 6, 6", BrCR,
               [(PPCcr6unset)]>;
 }
+}
 
 // XFX-Form instructions.  Instructions that deal with SPRs.
 //
@@ -1166,7 +1285,7 @@ def MFCTR : XFXForm_1_ext<31, 339, 9, (outs GPRC:$rT), (ins),
                           "mfctr $rT", SprMFSPR>,
             PPC970_DGroup_First, PPC970_Unit_FXU;
 }
-let Defs = [CTR], Pattern = [(PPCmtctr GPRC:$rS)] in {
+let Defs = [CTR], Pattern = [(PPCmtctr i32:$rS)] in {
 def MTCTR : XFXForm_7_ext<31, 467, 9, (outs), (ins GPRC:$rS),
                           "mtctr $rS", SprMTSPR>,
             PPC970_DGroup_First, PPC970_Unit_FXU;
@@ -1193,6 +1312,29 @@ def MFVRSAVE : XFXForm_1_ext<31, 339, 256, (outs GPRC:$rT), (ins),
                              "mfspr $rT, 256", IntGeneral>,
                PPC970_DGroup_First, PPC970_Unit_FXU;
 
+let isCodeGenOnly = 1 in {
+  def MTVRSAVEv : XFXForm_7_ext<31, 467, 256,
+                                (outs VRSAVERC:$reg), (ins GPRC:$rS),
+                                "mtspr 256, $rS", IntGeneral>,
+                  PPC970_DGroup_Single, PPC970_Unit_FXU;
+  def MFVRSAVEv : XFXForm_1_ext<31, 339, 256, (outs GPRC:$rT),
+                                (ins VRSAVERC:$reg),
+                                "mfspr $rT, 256", IntGeneral>,
+                  PPC970_DGroup_First, PPC970_Unit_FXU;
+}
+
+// SPILL_VRSAVE - Indicate that we're dumping the VRSAVE register,
+// so we'll need to scavenge a register for it.
+let mayStore = 1 in
+def SPILL_VRSAVE : Pseudo<(outs), (ins VRSAVERC:$vrsave, memri:$F),
+                     "#SPILL_VRSAVE", []>;
+
+// RESTORE_VRSAVE - Indicate that we're restoring the VRSAVE register (previously
+// spilled), so we'll need to scavenge a register for it.
+let mayLoad = 1 in
+def RESTORE_VRSAVE : Pseudo<(outs VRSAVERC:$vrsave), (ins memri:$F),
+                     "#RESTORE_VRSAVE", []>;
+
 def MTCRF : XFXForm_5<31, 144, (outs crbitm:$FXM), (ins GPRC:$rS),
                       "mtcrf $FXM, $rS", BrMCRX>,
             PPC970_MicroCode, PPC970_Unit_CRU;
@@ -1207,6 +1349,7 @@ def MTCRF : XFXForm_5<31, 144, (outs crbitm:$FXM), (ins GPRC:$rS),
 // instruction to keep the register allocator from becoming confused.
 //
 // FIXME: Make this a real Pseudo instruction when the JIT switches to MC.
+let isCodeGenOnly = 1 in
 def MFCRpseud: XFXForm_3<31, 19, (outs GPRC:$rT), (ins crbitm:$FXM),
                        "#MFCRpseud", SprMFCR>,
             PPC970_MicroCode, PPC970_Unit_CRU;
@@ -1219,38 +1362,29 @@ def MFOCRF: XFXForm_5a<31, 19, (outs GPRC:$rT), (ins crbitm:$FXM),
                        "mfocrf $rT, $FXM", SprMFCR>,
             PPC970_DGroup_First, PPC970_Unit_CRU;
 
-// Instructions to manipulate FPSCR.  Only long double handling uses these.
-// FPSCR is not modelled; we use the SDNode Flag to keep things in order.
+// Pseudo instruction to perform FADD in round-to-zero mode.
+let usesCustomInserter = 1, Uses = [RM] in {
+  def FADDrtz: Pseudo<(outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRB), "",
+                      [(set f64:$FRT, (PPCfaddrtz f64:$FRA, f64:$FRB))]>;
+}
 
+// The above pseudo gets expanded to make use of the following instructions
+// to manipulate FPSCR.  Note that FPSCR is not modeled at the DAG level.
 let Uses = [RM], Defs = [RM] in { 
   def MTFSB0 : XForm_43<63, 70, (outs), (ins u5imm:$FM),
-                         "mtfsb0 $FM", IntMTFSB0,
-                        [(PPCmtfsb0 (i32 imm:$FM))]>,
+                        "mtfsb0 $FM", IntMTFSB0, []>,
                PPC970_DGroup_Single, PPC970_Unit_FPU;
   def MTFSB1 : XForm_43<63, 38, (outs), (ins u5imm:$FM),
-                         "mtfsb1 $FM", IntMTFSB0,
-                        [(PPCmtfsb1 (i32 imm:$FM))]>,
+                        "mtfsb1 $FM", IntMTFSB0, []>,
                PPC970_DGroup_Single, PPC970_Unit_FPU;
-  // MTFSF does not actually produce an FP result.  We pretend it copies
-  // input reg B to the output.  If we didn't do this it would look like the
-  // instruction had no outputs (because we aren't modelling the FPSCR) and
-  // it would be deleted.
-  def MTFSF  : XFLForm<63, 711, (outs F8RC:$FRA),
-                                (ins i32imm:$FM, F8RC:$rT, F8RC:$FRB),
-                         "mtfsf $FM, $rT", "$FRB = $FRA", IntMTFSB0,
-                         [(set F8RC:$FRA, (PPCmtfsf (i32 imm:$FM), 
-                                                     F8RC:$rT, F8RC:$FRB))]>,
+  def MTFSF  : XFLForm<63, 711, (outs), (ins i32imm:$FM, F8RC:$rT),
+                       "mtfsf $FM, $rT", IntMTFSB0, []>,
                PPC970_DGroup_Single, PPC970_Unit_FPU;
 }
 let Uses = [RM] in {
   def MFFS   : XForm_42<63, 583, (outs F8RC:$rT), (ins), 
                          "mffs $rT", IntMFFS,
-                         [(set F8RC:$rT, (PPCmffs))]>,
-               PPC970_DGroup_Single, PPC970_Unit_FPU;
-  def FADDrtz: AForm_2<63, 21,
-                      (outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRB),
-                      "fadd $FRT, $FRA, $FRB", FPAddSub,
-                      [(set F8RC:$FRT, (PPCfaddrtz F8RC:$FRA, F8RC:$FRB))]>,
+                         [(set f64:$rT, (PPCmffs))]>,
                PPC970_DGroup_Single, PPC970_Unit_FPU;
 }
 
@@ -1261,61 +1395,61 @@ let PPC970_Unit = 1 in {  // FXU Operations.
 //
 def ADD4  : XOForm_1<31, 266, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                      "add $rT, $rA, $rB", IntSimple,
-                     [(set GPRC:$rT, (add GPRC:$rA, GPRC:$rB))]>;
+                     [(set i32:$rT, (add i32:$rA, i32:$rB))]>;
 let Defs = [CARRY] in {
 def ADDC  : XOForm_1<31, 10, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                      "addc $rT, $rA, $rB", IntGeneral,
-                     [(set GPRC:$rT, (addc GPRC:$rA, GPRC:$rB))]>,
+                     [(set i32:$rT, (addc i32:$rA, i32:$rB))]>,
                      PPC970_DGroup_Cracked;
 }
 def DIVW  : XOForm_1<31, 491, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                      "divw $rT, $rA, $rB", IntDivW,
-                     [(set GPRC:$rT, (sdiv GPRC:$rA, GPRC:$rB))]>,
+                     [(set i32:$rT, (sdiv i32:$rA, i32:$rB))]>,
                      PPC970_DGroup_First, PPC970_DGroup_Cracked;
 def DIVWU : XOForm_1<31, 459, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                      "divwu $rT, $rA, $rB", IntDivW,
-                     [(set GPRC:$rT, (udiv GPRC:$rA, GPRC:$rB))]>,
+                     [(set i32:$rT, (udiv i32:$rA, i32:$rB))]>,
                      PPC970_DGroup_First, PPC970_DGroup_Cracked;
 def MULHW : XOForm_1<31, 75, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                      "mulhw $rT, $rA, $rB", IntMulHW,
-                     [(set GPRC:$rT, (mulhs GPRC:$rA, GPRC:$rB))]>;
+                     [(set i32:$rT, (mulhs i32:$rA, i32:$rB))]>;
 def MULHWU : XOForm_1<31, 11, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                      "mulhwu $rT, $rA, $rB", IntMulHWU,
-                     [(set GPRC:$rT, (mulhu GPRC:$rA, GPRC:$rB))]>;
+                     [(set i32:$rT, (mulhu i32:$rA, i32:$rB))]>;
 def MULLW : XOForm_1<31, 235, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                      "mullw $rT, $rA, $rB", IntMulHW,
-                     [(set GPRC:$rT, (mul GPRC:$rA, GPRC:$rB))]>;
+                     [(set i32:$rT, (mul i32:$rA, i32:$rB))]>;
 def SUBF  : XOForm_1<31, 40, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                      "subf $rT, $rA, $rB", IntGeneral,
-                     [(set GPRC:$rT, (sub GPRC:$rB, GPRC:$rA))]>;
+                     [(set i32:$rT, (sub i32:$rB, i32:$rA))]>;
 let Defs = [CARRY] in {
 def SUBFC : XOForm_1<31, 8, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                      "subfc $rT, $rA, $rB", IntGeneral,
-                     [(set GPRC:$rT, (subc GPRC:$rB, GPRC:$rA))]>,
+                     [(set i32:$rT, (subc i32:$rB, i32:$rA))]>,
                      PPC970_DGroup_Cracked;
 }
 def NEG    : XOForm_3<31, 104, 0, (outs GPRC:$rT), (ins GPRC:$rA),
                       "neg $rT, $rA", IntSimple,
-                      [(set GPRC:$rT, (ineg GPRC:$rA))]>;
+                      [(set i32:$rT, (ineg i32:$rA))]>;
 let Uses = [CARRY], Defs = [CARRY] in {
 def ADDE  : XOForm_1<31, 138, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                       "adde $rT, $rA, $rB", IntGeneral,
-                      [(set GPRC:$rT, (adde GPRC:$rA, GPRC:$rB))]>;
+                      [(set i32:$rT, (adde i32:$rA, i32:$rB))]>;
 def ADDME  : XOForm_3<31, 234, 0, (outs GPRC:$rT), (ins GPRC:$rA),
                       "addme $rT, $rA", IntGeneral,
-                      [(set GPRC:$rT, (adde GPRC:$rA, -1))]>;
+                      [(set i32:$rT, (adde i32:$rA, -1))]>;
 def ADDZE  : XOForm_3<31, 202, 0, (outs GPRC:$rT), (ins GPRC:$rA),
                       "addze $rT, $rA", IntGeneral,
-                      [(set GPRC:$rT, (adde GPRC:$rA, 0))]>;
+                      [(set i32:$rT, (adde i32:$rA, 0))]>;
 def SUBFE : XOForm_1<31, 136, 0, (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB),
                       "subfe $rT, $rA, $rB", IntGeneral,
-                      [(set GPRC:$rT, (sube GPRC:$rB, GPRC:$rA))]>;
+                      [(set i32:$rT, (sube i32:$rB, i32:$rA))]>;
 def SUBFME : XOForm_3<31, 232, 0, (outs GPRC:$rT), (ins GPRC:$rA),
                       "subfme $rT, $rA", IntGeneral,
-                      [(set GPRC:$rT, (sube -1, GPRC:$rA))]>;
+                      [(set i32:$rT, (sube -1, i32:$rA))]>;
 def SUBFZE : XOForm_3<31, 200, 0, (outs GPRC:$rT), (ins GPRC:$rA),
                       "subfze $rT, $rA", IntGeneral,
-                      [(set GPRC:$rT, (sube 0, GPRC:$rA))]>;
+                      [(set i32:$rT, (sube 0, i32:$rA))]>;
 }
 }
 
@@ -1327,43 +1461,41 @@ let Uses = [RM] in {
   def FMADD : AForm_1<63, 29, 
                       (outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRC, F8RC:$FRB),
                       "fmadd $FRT, $FRA, $FRC, $FRB", FPFused,
-                      [(set F8RC:$FRT,
-                            (fma F8RC:$FRA, F8RC:$FRC, F8RC:$FRB))]>;
+                      [(set f64:$FRT, (fma f64:$FRA, f64:$FRC, f64:$FRB))]>;
   def FMADDS : AForm_1<59, 29,
                       (outs F4RC:$FRT), (ins F4RC:$FRA, F4RC:$FRC, F4RC:$FRB),
                       "fmadds $FRT, $FRA, $FRC, $FRB", FPGeneral,
-                      [(set F4RC:$FRT,
-                            (fma F4RC:$FRA, F4RC:$FRC, F4RC:$FRB))]>;
+                      [(set f32:$FRT, (fma f32:$FRA, f32:$FRC, f32:$FRB))]>;
   def FMSUB : AForm_1<63, 28,
                       (outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRC, F8RC:$FRB),
                       "fmsub $FRT, $FRA, $FRC, $FRB", FPFused,
-                      [(set F8RC:$FRT,
-                            (fma F8RC:$FRA, F8RC:$FRC, (fneg F8RC:$FRB)))]>;
+                      [(set f64:$FRT,
+                            (fma f64:$FRA, f64:$FRC, (fneg f64:$FRB)))]>;
   def FMSUBS : AForm_1<59, 28,
                       (outs F4RC:$FRT), (ins F4RC:$FRA, F4RC:$FRC, F4RC:$FRB),
                       "fmsubs $FRT, $FRA, $FRC, $FRB", FPGeneral,
-                      [(set F4RC:$FRT,
-                            (fma F4RC:$FRA, F4RC:$FRC, (fneg F4RC:$FRB)))]>;
+                      [(set f32:$FRT,
+                            (fma f32:$FRA, f32:$FRC, (fneg f32:$FRB)))]>;
   def FNMADD : AForm_1<63, 31,
                       (outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRC, F8RC:$FRB),
                       "fnmadd $FRT, $FRA, $FRC, $FRB", FPFused,
-                      [(set F8RC:$FRT,
-                            (fneg (fma F8RC:$FRA, F8RC:$FRC, F8RC:$FRB)))]>;
+                      [(set f64:$FRT,
+                            (fneg (fma f64:$FRA, f64:$FRC, f64:$FRB)))]>;
   def FNMADDS : AForm_1<59, 31,
                       (outs F4RC:$FRT), (ins F4RC:$FRA, F4RC:$FRC, F4RC:$FRB),
                       "fnmadds $FRT, $FRA, $FRC, $FRB", FPGeneral,
-                      [(set F4RC:$FRT,
-                            (fneg (fma F4RC:$FRA, F4RC:$FRC, F4RC:$FRB)))]>;
+                      [(set f32:$FRT,
+                            (fneg (fma f32:$FRA, f32:$FRC, f32:$FRB)))]>;
   def FNMSUB : AForm_1<63, 30,
                       (outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRC, F8RC:$FRB),
                       "fnmsub $FRT, $FRA, $FRC, $FRB", FPFused,
-                      [(set F8RC:$FRT, (fneg (fma F8RC:$FRA, F8RC:$FRC,
-                                                  (fneg F8RC:$FRB))))]>;
+                      [(set f64:$FRT, (fneg (fma f64:$FRA, f64:$FRC,
+                                                 (fneg f64:$FRB))))]>;
   def FNMSUBS : AForm_1<59, 30,
                       (outs F4RC:$FRT), (ins F4RC:$FRA, F4RC:$FRC, F4RC:$FRB),
                       "fnmsubs $FRT, $FRA, $FRC, $FRB", FPGeneral,
-                      [(set F4RC:$FRT, (fneg (fma F4RC:$FRA, F4RC:$FRC,
-                                                  (fneg F4RC:$FRB))))]>;
+                      [(set f32:$FRT, (fneg (fma f32:$FRA, f32:$FRC,
+                                                 (fneg f32:$FRB))))]>;
 }
 // FSEL is artificially split into 4 and 8-byte forms for the result.  To avoid
 // having 4 of these, force the comparison to always be an 8-byte double (code
@@ -1372,50 +1504,50 @@ let Uses = [RM] in {
 def FSELD : AForm_1<63, 23,
                     (outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRC, F8RC:$FRB),
                     "fsel $FRT, $FRA, $FRC, $FRB", FPGeneral,
-                    [(set F8RC:$FRT, (PPCfsel F8RC:$FRA,F8RC:$FRC,F8RC:$FRB))]>;
+                    [(set f64:$FRT, (PPCfsel f64:$FRA, f64:$FRC, f64:$FRB))]>;
 def FSELS : AForm_1<63, 23,
                      (outs F4RC:$FRT), (ins F8RC:$FRA, F4RC:$FRC, F4RC:$FRB),
                      "fsel $FRT, $FRA, $FRC, $FRB", FPGeneral,
-                    [(set F4RC:$FRT, (PPCfsel F8RC:$FRA,F4RC:$FRC,F4RC:$FRB))]>;
+                    [(set f32:$FRT, (PPCfsel f64:$FRA, f32:$FRC, f32:$FRB))]>;
 let Uses = [RM] in {
   def FADD  : AForm_2<63, 21,
                       (outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRB),
                       "fadd $FRT, $FRA, $FRB", FPAddSub,
-                      [(set F8RC:$FRT, (fadd F8RC:$FRA, F8RC:$FRB))]>;
+                      [(set f64:$FRT, (fadd f64:$FRA, f64:$FRB))]>;
   def FADDS : AForm_2<59, 21,
                       (outs F4RC:$FRT), (ins F4RC:$FRA, F4RC:$FRB),
                       "fadds $FRT, $FRA, $FRB", FPGeneral,
-                      [(set F4RC:$FRT, (fadd F4RC:$FRA, F4RC:$FRB))]>;
+                      [(set f32:$FRT, (fadd f32:$FRA, f32:$FRB))]>;
   def FDIV  : AForm_2<63, 18,
                       (outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRB),
                       "fdiv $FRT, $FRA, $FRB", FPDivD,
-                      [(set F8RC:$FRT, (fdiv F8RC:$FRA, F8RC:$FRB))]>;
+                      [(set f64:$FRT, (fdiv f64:$FRA, f64:$FRB))]>;
   def FDIVS : AForm_2<59, 18,
                       (outs F4RC:$FRT), (ins F4RC:$FRA, F4RC:$FRB),
                       "fdivs $FRT, $FRA, $FRB", FPDivS,
-                      [(set F4RC:$FRT, (fdiv F4RC:$FRA, F4RC:$FRB))]>;
+                      [(set f32:$FRT, (fdiv f32:$FRA, f32:$FRB))]>;
   def FMUL  : AForm_3<63, 25,
                       (outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRC),
                       "fmul $FRT, $FRA, $FRC", FPFused,
-                      [(set F8RC:$FRT, (fmul F8RC:$FRA, F8RC:$FRC))]>;
+                      [(set f64:$FRT, (fmul f64:$FRA, f64:$FRC))]>;
   def FMULS : AForm_3<59, 25,
                       (outs F4RC:$FRT), (ins F4RC:$FRA, F4RC:$FRC),
                       "fmuls $FRT, $FRA, $FRC", FPGeneral,
-                      [(set F4RC:$FRT, (fmul F4RC:$FRA, F4RC:$FRC))]>;
+                      [(set f32:$FRT, (fmul f32:$FRA, f32:$FRC))]>;
   def FSUB  : AForm_2<63, 20,
                       (outs F8RC:$FRT), (ins F8RC:$FRA, F8RC:$FRB),
                       "fsub $FRT, $FRA, $FRB", FPAddSub,
-                      [(set F8RC:$FRT, (fsub F8RC:$FRA, F8RC:$FRB))]>;
+                      [(set f64:$FRT, (fsub f64:$FRA, f64:$FRB))]>;
   def FSUBS : AForm_2<59, 20,
                       (outs F4RC:$FRT), (ins F4RC:$FRA, F4RC:$FRB),
                       "fsubs $FRT, $FRA, $FRB", FPGeneral,
-                      [(set F4RC:$FRT, (fsub F4RC:$FRA, F4RC:$FRB))]>;
+                      [(set f32:$FRT, (fsub f32:$FRA, f32:$FRB))]>;
   }
 }
 
 let PPC970_Unit = 1 in {  // FXU Operations.
   def ISEL  : AForm_4<31, 15,
-                     (outs GPRC:$rT), (ins GPRC:$rA, GPRC:$rB, pred:$cond),
+                     (outs GPRC:$rT), (ins GPRC_NOR0:$rA, GPRC:$rB, CRBITRC:$cond),
                      "isel $rT, $rA, $rB, $cond", IntGeneral,
                      []>;
 }
@@ -1455,47 +1587,43 @@ def : Pat<(i32 imm:$imm),
           (ORI (LIS (HI16 imm:$imm)), (LO16 imm:$imm))>;
 
 // Implement the 'not' operation with the NOR instruction.
-def NOT : Pat<(not GPRC:$in),
-              (NOR GPRC:$in, GPRC:$in)>;
+def NOT : Pat<(not i32:$in),
+              (NOR $in, $in)>;
 
 // ADD an arbitrary immediate.
-def : Pat<(add GPRC:$in, imm:$imm),
-          (ADDIS (ADDI GPRC:$in, (LO16 imm:$imm)), (HA16 imm:$imm))>;
+def : Pat<(add i32:$in, imm:$imm),
+          (ADDIS (ADDI $in, (LO16 imm:$imm)), (HA16 imm:$imm))>;
 // OR an arbitrary immediate.
-def : Pat<(or GPRC:$in, imm:$imm),
-          (ORIS (ORI GPRC:$in, (LO16 imm:$imm)), (HI16 imm:$imm))>;
+def : Pat<(or i32:$in, imm:$imm),
+          (ORIS (ORI $in, (LO16 imm:$imm)), (HI16 imm:$imm))>;
 // XOR an arbitrary immediate.
-def : Pat<(xor GPRC:$in, imm:$imm),
-          (XORIS (XORI GPRC:$in, (LO16 imm:$imm)), (HI16 imm:$imm))>;
+def : Pat<(xor i32:$in, imm:$imm),
+          (XORIS (XORI $in, (LO16 imm:$imm)), (HI16 imm:$imm))>;
 // SUBFIC
-def : Pat<(sub  immSExt16:$imm, GPRC:$in),
-          (SUBFIC GPRC:$in, imm:$imm)>;
+def : Pat<(sub immSExt16:$imm, i32:$in),
+          (SUBFIC $in, imm:$imm)>;
 
 // SHL/SRL
-def : Pat<(shl GPRC:$in, (i32 imm:$imm)),
-          (RLWINM GPRC:$in, imm:$imm, 0, (SHL32 imm:$imm))>;
-def : Pat<(srl GPRC:$in, (i32 imm:$imm)),
-          (RLWINM GPRC:$in, (SRL32 imm:$imm), imm:$imm, 31)>;
+def : Pat<(shl i32:$in, (i32 imm:$imm)),
+          (RLWINM $in, imm:$imm, 0, (SHL32 imm:$imm))>;
+def : Pat<(srl i32:$in, (i32 imm:$imm)),
+          (RLWINM $in, (SRL32 imm:$imm), imm:$imm, 31)>;
 
 // ROTL
-def : Pat<(rotl GPRC:$in, GPRC:$sh),
-          (RLWNM GPRC:$in, GPRC:$sh, 0, 31)>;
-def : Pat<(rotl GPRC:$in, (i32 imm:$imm)),
-          (RLWINM GPRC:$in, imm:$imm, 0, 31)>;
+def : Pat<(rotl i32:$in, i32:$sh),
+          (RLWNM $in, $sh, 0, 31)>;
+def : Pat<(rotl i32:$in, (i32 imm:$imm)),
+          (RLWINM $in, imm:$imm, 0, 31)>;
 
 // RLWNM
-def : Pat<(and (rotl GPRC:$in, GPRC:$sh), maskimm32:$imm),
-          (RLWNM GPRC:$in, GPRC:$sh, (MB maskimm32:$imm), (ME maskimm32:$imm))>;
+def : Pat<(and (rotl i32:$in, i32:$sh), maskimm32:$imm),
+          (RLWNM $in, $sh, (MB maskimm32:$imm), (ME maskimm32:$imm))>;
 
 // Calls
-def : Pat<(PPCcall_Darwin (i32 tglobaladdr:$dst)),
-          (BL_Darwin tglobaladdr:$dst)>;
-def : Pat<(PPCcall_Darwin (i32 texternalsym:$dst)),
-          (BL_Darwin texternalsym:$dst)>;
-def : Pat<(PPCcall_SVR4 (i32 tglobaladdr:$dst)),
-          (BL_SVR4 tglobaladdr:$dst)>;
-def : Pat<(PPCcall_SVR4 (i32 texternalsym:$dst)),
-          (BL_SVR4 texternalsym:$dst)>;
+def : Pat<(PPCcall (i32 tglobaladdr:$dst)),
+          (BL tglobaladdr:$dst)>;
+def : Pat<(PPCcall (i32 texternalsym:$dst)),
+          (BL texternalsym:$dst)>;
 
 
 def : Pat<(PPCtc_return (i32 tglobaladdr:$dst),  imm:$imm),
@@ -1518,28 +1646,28 @@ def : Pat<(PPChi tjumptable:$in, 0), (LIS tjumptable:$in)>;
 def : Pat<(PPClo tjumptable:$in, 0), (LI tjumptable:$in)>;
 def : Pat<(PPChi tblockaddress:$in, 0), (LIS tblockaddress:$in)>;
 def : Pat<(PPClo tblockaddress:$in, 0), (LI tblockaddress:$in)>;
-def : Pat<(PPChi tglobaltlsaddr:$g, GPRC:$in),
-          (ADDIS GPRC:$in, tglobaltlsaddr:$g)>;
-def : Pat<(PPClo tglobaltlsaddr:$g, GPRC:$in),
-          (ADDIL GPRC:$in, tglobaltlsaddr:$g)>;
-def : Pat<(add GPRC:$in, (PPChi tglobaladdr:$g, 0)),
-          (ADDIS GPRC:$in, tglobaladdr:$g)>;
-def : Pat<(add GPRC:$in, (PPChi tconstpool:$g, 0)),
-          (ADDIS GPRC:$in, tconstpool:$g)>;
-def : Pat<(add GPRC:$in, (PPChi tjumptable:$g, 0)),
-          (ADDIS GPRC:$in, tjumptable:$g)>;
-def : Pat<(add GPRC:$in, (PPChi tblockaddress:$g, 0)),
-          (ADDIS GPRC:$in, tblockaddress:$g)>;
+def : Pat<(PPChi tglobaltlsaddr:$g, i32:$in),
+          (ADDIS $in, tglobaltlsaddr:$g)>;
+def : Pat<(PPClo tglobaltlsaddr:$g, i32:$in),
+          (ADDI $in, tglobaltlsaddr:$g)>;
+def : Pat<(add i32:$in, (PPChi tglobaladdr:$g, 0)),
+          (ADDIS $in, tglobaladdr:$g)>;
+def : Pat<(add i32:$in, (PPChi tconstpool:$g, 0)),
+          (ADDIS $in, tconstpool:$g)>;
+def : Pat<(add i32:$in, (PPChi tjumptable:$g, 0)),
+          (ADDIS $in, tjumptable:$g)>;
+def : Pat<(add i32:$in, (PPChi tblockaddress:$g, 0)),
+          (ADDIS $in, tblockaddress:$g)>;
 
 // Standard shifts.  These are represented separately from the real shifts above
 // so that we can distinguish between shifts that allow 5-bit and 6-bit shift
 // amounts.
-def : Pat<(sra GPRC:$rS, GPRC:$rB),
-          (SRAW GPRC:$rS, GPRC:$rB)>;
-def : Pat<(srl GPRC:$rS, GPRC:$rB),
-          (SRW GPRC:$rS, GPRC:$rB)>;
-def : Pat<(shl GPRC:$rS, GPRC:$rB),
-          (SLW GPRC:$rS, GPRC:$rB)>;
+def : Pat<(sra i32:$rS, i32:$rB),
+          (SRAW $rS, $rB)>;
+def : Pat<(srl i32:$rS, i32:$rB),
+          (SRW $rS, $rB)>;
+def : Pat<(shl i32:$rS, i32:$rB),
+          (SLW $rS, $rB)>;
 
 def : Pat<(zextloadi1 iaddr:$src),
           (LBZ iaddr:$src)>;
@@ -1562,8 +1690,8 @@ def : Pat<(f64 (extloadf32 iaddr:$src)),
 def : Pat<(f64 (extloadf32 xaddr:$src)),
           (COPY_TO_REGCLASS (LFSX xaddr:$src), F8RC)>;
 
-def : Pat<(f64 (fextend F4RC:$src)),
-          (COPY_TO_REGCLASS F4RC:$src, F8RC)>;
+def : Pat<(f64 (fextend f32:$src)),
+          (COPY_TO_REGCLASS $src, F8RC)>;
 
 // Memory barriers
 def : Pat<(membarrier (i32 imm /*ll*/),
@@ -1575,5 +1703,15 @@ def : Pat<(membarrier (i32 imm /*ll*/),
 
 def : Pat<(atomic_fence (imm), (imm)), (SYNC)>;
 
+// Additional FNMSUB patterns: -a*c + b == -(a*c - b)
+def : Pat<(fma (fneg f64:$A), f64:$C, f64:$B),
+          (FNMSUB $A, $C, $B)>;
+def : Pat<(fma f64:$A, (fneg f64:$C), f64:$B),
+          (FNMSUB $A, $C, $B)>;
+def : Pat<(fma (fneg f32:$A), f32:$C, f32:$B),
+          (FNMSUBS $A, $C, $B)>;
+def : Pat<(fma f32:$A, (fneg f32:$C), f32:$B),
+          (FNMSUBS $A, $C, $B)>;
+
 include "PPCInstrAltivec.td"
 include "PPCInstr64Bit.td"
diff --git a/lib/Target/PowerPC/PPCJITInfo.cpp b/lib/Target/PowerPC/PPCJITInfo.cpp
index aba27399d6da..cfcd7490ed0d 100644
--- a/lib/Target/PowerPC/PPCJITInfo.cpp
+++ b/lib/Target/PowerPC/PPCJITInfo.cpp
@@ -15,10 +15,10 @@
 #include "PPCJITInfo.h"
 #include "PPCRelocations.h"
 #include "PPCTargetMachine.h"
-#include "llvm/Function.h"
-#include "llvm/Support/Memory.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/Memory.h"
 #include "llvm/Support/raw_ostream.h"
 using namespace llvm;
 
@@ -115,7 +115,7 @@ asm(
     "lwz  r2, 208(r1)\n" // stub's frame
     "lwz  r4, 8(r2)\n" // stub's lr
     "li   r5, 0\n"       // 0 == 32 bit
-    "bl _PPCCompilationCallbackC\n"
+    "bl _LLVMPPCCompilationCallback\n"
     "mtctr r3\n"
     // Restore all int arg registers
     "lwz r10, 204(r1)\n"    "lwz r9,  200(r1)\n"
@@ -178,7 +178,7 @@ asm(
     "lwz  5, 104(1)\n" // stub's frame
     "lwz  4, 4(5)\n" // stub's lr
     "li   5, 0\n"       // 0 == 32 bit
-    "bl PPCCompilationCallbackC\n"
+    "bl LLVMPPCCompilationCallback\n"
     "mtctr 3\n"
     // Restore all int arg registers
     "lwz 10, 100(1)\n"   "lwz 9,  96(1)\n"
@@ -259,10 +259,10 @@ asm(
     "ld   4, 16(5)\n"  // stub's lr
     "li   5, 1\n"      // 1 == 64 bit
 #ifdef __ELF__
-    "bl PPCCompilationCallbackC\n"
+    "bl LLVMPPCCompilationCallback\n"
     "nop\n"
 #else
-    "bl _PPCCompilationCallbackC\n"
+    "bl _LLVMPPCCompilationCallback\n"
 #endif
     "mtctr 3\n"
     // Restore all int arg registers
@@ -292,9 +292,10 @@ void PPC64CompilationCallback() {
 #endif
 
 extern "C" {
-static void* LLVM_ATTRIBUTE_USED PPCCompilationCallbackC(unsigned *StubCallAddrPlus4,
-                                                         unsigned *OrigCallAddrPlus4,
-                                                         bool is64Bit) {
+LLVM_LIBRARY_VISIBILITY void *
+LLVMPPCCompilationCallback(unsigned *StubCallAddrPlus4,
+                           unsigned *OrigCallAddrPlus4,
+                           bool is64Bit) {
   // Adjust the pointer to the address of the call instruction in the stub
   // emitted by emitFunctionStub, rather than the instruction after it.
   unsigned *StubCallAddr = StubCallAddrPlus4 - 1;
diff --git a/lib/Target/PowerPC/PPCJITInfo.h b/lib/Target/PowerPC/PPCJITInfo.h
index 2f8243a597e6..46d4a08eb687 100644
--- a/lib/Target/PowerPC/PPCJITInfo.h
+++ b/lib/Target/PowerPC/PPCJITInfo.h
@@ -14,8 +14,8 @@
 #ifndef POWERPC_JITINFO_H
 #define POWERPC_JITINFO_H
 
-#include "llvm/Target/TargetJITInfo.h"
 #include "llvm/CodeGen/JITCodeEmitter.h"
+#include "llvm/Target/TargetJITInfo.h"
 
 namespace llvm {
   class PPCTargetMachine;
diff --git a/lib/Target/PowerPC/PPCMCInstLower.cpp b/lib/Target/PowerPC/PPCMCInstLower.cpp
index 19ec993ba00f..9b0df3e86a75 100644
--- a/lib/Target/PowerPC/PPCMCInstLower.cpp
+++ b/lib/Target/PowerPC/PPCMCInstLower.cpp
@@ -13,14 +13,15 @@
 //===----------------------------------------------------------------------===//
 
 #include "PPC.h"
+#include "llvm/ADT/SmallString.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineModuleInfoImpls.h"
+#include "llvm/IR/GlobalValue.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/Target/Mangler.h"
-#include "llvm/ADT/SmallString.h"
 using namespace llvm;
 
 static MachineModuleInfoMachO &getMachOMMI(AsmPrinter &AP) {
@@ -114,6 +115,12 @@ static MCOperand GetSymbolRef(const MachineOperand &MO, const MCSymbol *Symbol,
                                break;
     case PPCII::MO_TPREL16_LO: RefKind = MCSymbolRefExpr::VK_PPC_TPREL16_LO;
                                break;
+    case PPCII::MO_DTPREL16_LO: RefKind = MCSymbolRefExpr::VK_PPC_DTPREL16_LO;
+                                break;
+    case PPCII::MO_TLSLD16_LO: RefKind = MCSymbolRefExpr::VK_PPC_GOT_TLSLD16_LO;
+                               break;
+    case PPCII::MO_TOC16_LO: RefKind = MCSymbolRefExpr::VK_PPC_TOC16_LO;
+                             break;
    }
 
   // FIXME: This isn't right, but we don't have a good way to express this in
diff --git a/lib/Target/PowerPC/PPCMachineFunctionInfo.h b/lib/Target/PowerPC/PPCMachineFunctionInfo.h
index 24caffa3f0f2..ee18eadf6e5f 100644
--- a/lib/Target/PowerPC/PPCMachineFunctionInfo.h
+++ b/lib/Target/PowerPC/PPCMachineFunctionInfo.h
@@ -37,9 +37,19 @@ class PPCFunctionInfo : public MachineFunctionInfo {
   /// PEI.
   bool MustSaveLR;
 
+  /// Does this function have any stack spills.
+  bool HasSpills;
+
+  /// Does this function spill using instructions with only r+r (not r+i)
+  /// forms.
+  bool HasNonRISpills;
+
   /// SpillsCR - Indicates whether CR is spilled in the current function.
   bool SpillsCR;
 
+  /// Indicates whether VRSAVE is spilled in the current function.
+  bool SpillsVRSAVE;
+
   /// LRStoreRequired - The bool indicates whether there is some explicit use of
   /// the LR/LR8 stack slot that is not obvious from scanning the code.  This
   /// requires that the code generator produce a store of LR to the stack on
@@ -71,11 +81,17 @@ class PPCFunctionInfo : public MachineFunctionInfo {
   /// register for parameter passing.
   unsigned VarArgsNumFPR;
 
+  /// CRSpillFrameIndex - FrameIndex for CR spill slot for 32-bit SVR4.
+  int CRSpillFrameIndex;
+
 public:
   explicit PPCFunctionInfo(MachineFunction &MF) 
     : FramePointerSaveIndex(0),
       ReturnAddrSaveIndex(0),
+      HasSpills(false),
+      HasNonRISpills(false),
       SpillsCR(false),
+      SpillsVRSAVE(false),
       LRStoreRequired(false),
       MinReservedArea(0),
       TailCallSPDelta(0),
@@ -83,7 +99,8 @@ public:
       VarArgsFrameIndex(0),
       VarArgsStackOffset(0),
       VarArgsNumGPR(0),
-      VarArgsNumFPR(0) {}
+      VarArgsNumFPR(0),
+      CRSpillFrameIndex(0) {}
 
   int getFramePointerSaveIndex() const { return FramePointerSaveIndex; }
   void setFramePointerSaveIndex(int Idx) { FramePointerSaveIndex = Idx; }
@@ -105,9 +122,18 @@ public:
   void setMustSaveLR(bool U) { MustSaveLR = U; }
   bool mustSaveLR() const    { return MustSaveLR; }
 
+  void setHasSpills()      { HasSpills = true; }
+  bool hasSpills() const   { return HasSpills; }
+
+  void setHasNonRISpills()    { HasNonRISpills = true; }
+  bool hasNonRISpills() const { return HasNonRISpills; }
+
   void setSpillsCR()       { SpillsCR = true; }
   bool isCRSpilled() const { return SpillsCR; }
 
+  void setSpillsVRSAVE()       { SpillsVRSAVE = true; }
+  bool isVRSAVESpilled() const { return SpillsVRSAVE; }
+
   void setLRStoreRequired() { LRStoreRequired = true; }
   bool isLRStoreRequired() const { return LRStoreRequired; }
 
@@ -125,6 +151,9 @@ public:
 
   unsigned getVarArgsNumFPR() const { return VarArgsNumFPR; }
   void setVarArgsNumFPR(unsigned Num) { VarArgsNumFPR = Num; }
+
+  int getCRSpillFrameIndex() const { return CRSpillFrameIndex; }
+  void setCRSpillFrameIndex(int idx) { CRSpillFrameIndex = idx; }
 };
 
 } // end of namespace llvm
diff --git a/lib/Target/PowerPC/PPCRegisterInfo.cpp b/lib/Target/PowerPC/PPCRegisterInfo.cpp
index 459c3589d3f6..1d61a3a8eac2 100644
--- a/lib/Target/PowerPC/PPCRegisterInfo.cpp
+++ b/lib/Target/PowerPC/PPCRegisterInfo.cpp
@@ -15,63 +15,45 @@
 #define DEBUG_TYPE "reginfo"
 #include "PPCRegisterInfo.h"
 #include "PPC.h"
+#include "PPCFrameLowering.h"
 #include "PPCInstrBuilder.h"
 #include "PPCMachineFunctionInfo.h"
-#include "PPCFrameLowering.h"
 #include "PPCSubtarget.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
-#include "llvm/Function.h"
-#include "llvm/Type.h"
-#include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/CodeGen/MachineFunction.h"
-#include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/STLExtras.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
 #include <cstdlib>
 
 #define GET_REGINFO_TARGET_DESC
 #include "PPCGenRegisterInfo.inc"
 
-namespace llvm {
-cl::opt<bool> DisablePPC32RS("disable-ppc32-regscavenger",
-                                   cl::init(false),
-                                   cl::desc("Disable PPC32 register scavenger"),
-                                   cl::Hidden);
-cl::opt<bool> DisablePPC64RS("disable-ppc64-regscavenger",
-                                   cl::init(false),
-                                   cl::desc("Disable PPC64 register scavenger"),
-                                   cl::Hidden);
-}
-
 using namespace llvm;
 
-// FIXME (64-bit): Should be inlined.
-bool
-PPCRegisterInfo::requiresRegisterScavenging(const MachineFunction &) const {
-  return ((!DisablePPC32RS && !Subtarget.isPPC64()) ||
-          (!DisablePPC64RS && Subtarget.isPPC64()));
-}
-
 PPCRegisterInfo::PPCRegisterInfo(const PPCSubtarget &ST,
                                  const TargetInstrInfo &tii)
   : PPCGenRegisterInfo(ST.isPPC64() ? PPC::LR8 : PPC::LR,
                        ST.isPPC64() ? 0 : 1,
                        ST.isPPC64() ? 0 : 1),
-    Subtarget(ST), TII(tii), CRSpillFrameIdx(0) {
+    Subtarget(ST), TII(tii) {
   ImmToIdxMap[PPC::LD]   = PPC::LDX;    ImmToIdxMap[PPC::STD]  = PPC::STDX;
   ImmToIdxMap[PPC::LBZ]  = PPC::LBZX;   ImmToIdxMap[PPC::STB]  = PPC::STBX;
   ImmToIdxMap[PPC::LHZ]  = PPC::LHZX;   ImmToIdxMap[PPC::LHA]  = PPC::LHAX;
@@ -86,20 +68,20 @@ PPCRegisterInfo::PPCRegisterInfo(const PPCSubtarget &ST,
   ImmToIdxMap[PPC::LHZ8] = PPC::LHZX8; ImmToIdxMap[PPC::LWZ8] = PPC::LWZX8;
   ImmToIdxMap[PPC::STB8] = PPC::STBX8; ImmToIdxMap[PPC::STH8] = PPC::STHX8;
   ImmToIdxMap[PPC::STW8] = PPC::STWX8; ImmToIdxMap[PPC::STDU] = PPC::STDUX;
-  ImmToIdxMap[PPC::ADDI8] = PPC::ADD8; ImmToIdxMap[PPC::STD_32] = PPC::STDX_32;
+  ImmToIdxMap[PPC::ADDI8] = PPC::ADD8;
 }
 
-bool
-PPCRegisterInfo::trackLivenessAfterRegAlloc(const MachineFunction &MF) const {
-  return requiresRegisterScavenging(MF);
-}
-
-
 /// getPointerRegClass - Return the register class to use to hold pointers.
 /// This is used for addressing modes.
 const TargetRegisterClass *
 PPCRegisterInfo::getPointerRegClass(const MachineFunction &MF, unsigned Kind)
                                                                        const {
+  if (Kind == 1) {
+    if (Subtarget.isPPC64())
+      return &PPC::G8RC_NOX0RegClass;
+    return &PPC::GPRC_NOR0RegClass;
+  }
+
   if (Subtarget.isPPC64())
     return &PPC::G8RCRegClass;
   return &PPC::GPRCRegClass;
@@ -111,11 +93,6 @@ PPCRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
     return Subtarget.isPPC64() ? CSR_Darwin64_SaveList :
                                  CSR_Darwin32_SaveList;
 
-  // For 32-bit SVR4, also initialize the frame index associated with
-  // the CR spill slot.
-  if (!Subtarget.isPPC64())
-    CRSpillFrameIdx = 0;
-
   return Subtarget.isPPC64() ? CSR_SVR464_SaveList : CSR_SVR432_SaveList;
 }
 
@@ -128,12 +105,35 @@ PPCRegisterInfo::getCallPreservedMask(CallingConv::ID CC) const {
   return Subtarget.isPPC64() ? CSR_SVR464_RegMask : CSR_SVR432_RegMask;
 }
 
+const uint32_t*
+PPCRegisterInfo::getNoPreservedMask() const {
+  // The naming here is inverted: The CSR_NoRegs_Altivec has the
+  // Altivec registers masked so that they're not saved and restored around
+  // instructions with this preserved mask.
+
+  if (!Subtarget.hasAltivec())
+    return CSR_NoRegs_Altivec_RegMask;
+
+  if (Subtarget.isDarwin())
+    return CSR_NoRegs_Darwin_RegMask;
+  return CSR_NoRegs_RegMask;
+}
+
 BitVector PPCRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   BitVector Reserved(getNumRegs());
   const PPCFrameLowering *PPCFI =
     static_cast<const PPCFrameLowering*>(MF.getTarget().getFrameLowering());
 
-  Reserved.set(PPC::R0);
+  // The ZERO register is not really a register, but the representation of r0
+  // when used in instructions that treat r0 as the constant 0.
+  Reserved.set(PPC::ZERO);
+  Reserved.set(PPC::ZERO8);
+
+  // The FP register is also not really a register, but is the representation
+  // of the frame pointer register used by ISD::FRAMEADDR.
+  Reserved.set(PPC::FP);
+  Reserved.set(PPC::FP8);
+
   Reserved.set(PPC::R1);
   Reserved.set(PPC::LR);
   Reserved.set(PPC::LR8);
@@ -144,35 +144,21 @@ BitVector PPCRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
     Reserved.set(PPC::R2);  // System-reserved register
     Reserved.set(PPC::R13); // Small Data Area pointer register
   }
-  // Reserve R2 on Darwin to hack around the problem of save/restore of CR
-  // when the stack frame is too big to address directly; we need two regs.
-  // This is a hack.
-  if (Subtarget.isDarwinABI()) {
-    Reserved.set(PPC::R2);
-  }
   
   // On PPC64, r13 is the thread pointer. Never allocate this register.
-  // Note that this is over conservative, as it also prevents allocation of R31
-  // when the FP is not needed.
   if (Subtarget.isPPC64()) {
     Reserved.set(PPC::R13);
-    Reserved.set(PPC::R31);
 
-    Reserved.set(PPC::X0);
     Reserved.set(PPC::X1);
     Reserved.set(PPC::X13);
-    Reserved.set(PPC::X31);
+
+    if (PPCFI->needsFP(MF))
+      Reserved.set(PPC::X31);
 
     // The 64-bit SVR4 ABI reserves r2 for the TOC pointer.
     if (Subtarget.isSVR4ABI()) {
       Reserved.set(PPC::X2);
     }
-    // Reserve X2 on Darwin to hack around the problem of save/restore of CR
-    // when the stack frame is too big to address directly; we need two regs.
-    // This is a hack.
-    if (Subtarget.isDarwinABI()) {
-      Reserved.set(PPC::X2);
-    }
   }
 
   if (PPCFI->needsFP(MF))
@@ -190,6 +176,8 @@ PPCRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC,
   switch (RC->getID()) {
   default:
     return 0;
+  case PPC::G8RC_NOX0RegClassID:
+  case PPC::GPRC_NOR0RegClassID: 
   case PPC::G8RCRegClassID:
   case PPC::GPRCRegClassID: {
     unsigned FP = TFI->hasFP(MF) ? 1 : 0;
@@ -204,77 +192,10 @@ PPCRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC,
   }
 }
 
-bool
-PPCRegisterInfo::avoidWriteAfterWrite(const TargetRegisterClass *RC) const {
-  switch (RC->getID()) {
-  case PPC::G8RCRegClassID:
-  case PPC::GPRCRegClassID:
-  case PPC::F8RCRegClassID:
-  case PPC::F4RCRegClassID:
-  case PPC::VRRCRegClassID:
-    return true;
-  default:
-    return false;
-  }
-}
-
 //===----------------------------------------------------------------------===//
 // Stack Frame Processing methods
 //===----------------------------------------------------------------------===//
 
-void PPCRegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  if (MF.getTarget().Options.GuaranteedTailCallOpt &&
-      I->getOpcode() == PPC::ADJCALLSTACKUP) {
-    // Add (actually subtract) back the amount the callee popped on return.
-    if (int CalleeAmt =  I->getOperand(1).getImm()) {
-      bool is64Bit = Subtarget.isPPC64();
-      CalleeAmt *= -1;
-      unsigned StackReg = is64Bit ? PPC::X1 : PPC::R1;
-      unsigned TmpReg = is64Bit ? PPC::X0 : PPC::R0;
-      unsigned ADDIInstr = is64Bit ? PPC::ADDI8 : PPC::ADDI;
-      unsigned ADDInstr = is64Bit ? PPC::ADD8 : PPC::ADD4;
-      unsigned LISInstr = is64Bit ? PPC::LIS8 : PPC::LIS;
-      unsigned ORIInstr = is64Bit ? PPC::ORI8 : PPC::ORI;
-      MachineInstr *MI = I;
-      DebugLoc dl = MI->getDebugLoc();
-
-      if (isInt<16>(CalleeAmt)) {
-        BuildMI(MBB, I, dl, TII.get(ADDIInstr), StackReg)
-          .addReg(StackReg, RegState::Kill)
-          .addImm(CalleeAmt);
-      } else {
-        MachineBasicBlock::iterator MBBI = I;
-        BuildMI(MBB, MBBI, dl, TII.get(LISInstr), TmpReg)
-          .addImm(CalleeAmt >> 16);
-        BuildMI(MBB, MBBI, dl, TII.get(ORIInstr), TmpReg)
-          .addReg(TmpReg, RegState::Kill)
-          .addImm(CalleeAmt & 0xFFFF);
-        BuildMI(MBB, MBBI, dl, TII.get(ADDInstr), StackReg)
-          .addReg(StackReg, RegState::Kill)
-          .addReg(TmpReg);
-      }
-    }
-  }
-  // Simply discard ADJCALLSTACKDOWN, ADJCALLSTACKUP instructions.
-  MBB.erase(I);
-}
-
-/// findScratchRegister - Find a 'free' PPC register. Try for a call-clobbered
-/// register first and then a spilled callee-saved register if that fails.
-static
-unsigned findScratchRegister(MachineBasicBlock::iterator II, RegScavenger *RS,
-                             const TargetRegisterClass *RC, int SPAdj) {
-  assert(RS && "Register scavenging must be on");
-  unsigned Reg = RS->FindUnusedReg(RC);
-  // FIXME: move ARM callee-saved reg scan to target independent code, then 
-  // search for already spilled CS register here.
-  if (Reg == 0)
-    Reg = RS->scavengeRegister(RC, II, SPAdj);
-  return Reg;
-}
-
 /// lowerDynamicAlloc - Generate the code for allocating an object in the
 /// current frame.  The sequence of code with be in the general form
 ///
@@ -282,8 +203,7 @@ unsigned findScratchRegister(MachineBasicBlock::iterator II, RegScavenger *RS,
 ///   stwxu  R0, SP, Rnegsize   ; add and update the SP with the negated size
 ///   addi   Rnew, SP, \#maxCalFrameSize ; get the top of the allocation
 ///
-void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II,
-                                        int SPAdj, RegScavenger *RS) const {
+void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II) const {
   // Get the instruction.
   MachineInstr &MI = *II;
   // Get the instruction's basic block.
@@ -315,28 +235,16 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II,
   // Fortunately, a frame greater than 32K is rare.
   const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
   const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
-  const TargetRegisterClass *RC = LP64 ? G8RC : GPRC;
-
-  // FIXME (64-bit): Use "findScratchRegister"
-  unsigned Reg;
-  if (requiresRegisterScavenging(MF))
-    Reg = findScratchRegister(II, RS, RC, SPAdj);
-  else
-    Reg = PPC::R0;
+  unsigned Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC);
   
   if (MaxAlign < TargetAlign && isInt<16>(FrameSize)) {
     BuildMI(MBB, II, dl, TII.get(PPC::ADDI), Reg)
       .addReg(PPC::R31)
       .addImm(FrameSize);
   } else if (LP64) {
-    if (requiresRegisterScavenging(MF)) // FIXME (64-bit): Use "true" part.
-      BuildMI(MBB, II, dl, TII.get(PPC::LD), Reg)
-        .addImm(0)
-        .addReg(PPC::X1);
-    else
-      BuildMI(MBB, II, dl, TII.get(PPC::LD), PPC::X0)
-        .addImm(0)
-        .addReg(PPC::X1);
+    BuildMI(MBB, II, dl, TII.get(PPC::LD), Reg)
+      .addImm(0)
+      .addReg(PPC::X1);
   } else {
     BuildMI(MBB, II, dl, TII.get(PPC::LWZ), Reg)
       .addImm(0)
@@ -346,17 +254,10 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II,
   // Grow the stack and update the stack pointer link, then determine the
   // address of new allocated space.
   if (LP64) {
-    if (requiresRegisterScavenging(MF)) // FIXME (64-bit): Use "true" part.
-      BuildMI(MBB, II, dl, TII.get(PPC::STDUX), PPC::X1)
-        .addReg(Reg, RegState::Kill)
-        .addReg(PPC::X1)
-        .addReg(MI.getOperand(1).getReg());
-    else
-      BuildMI(MBB, II, dl, TII.get(PPC::STDUX), PPC::X1)
-        .addReg(PPC::X0, RegState::Kill)
-        .addReg(PPC::X1)
-        .addReg(MI.getOperand(1).getReg());
-
+    BuildMI(MBB, II, dl, TII.get(PPC::STDUX), PPC::X1)
+      .addReg(Reg, RegState::Kill)
+      .addReg(PPC::X1)
+      .addReg(MI.getOperand(1).getReg());
     if (!MI.getOperand(1).isKill())
       BuildMI(MBB, II, dl, TII.get(PPC::ADDI8), MI.getOperand(0).getReg())
         .addReg(PPC::X1)
@@ -398,23 +299,19 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II,
 ///   stw rA, FI               ; Store rA to the frame.
 ///
 void PPCRegisterInfo::lowerCRSpilling(MachineBasicBlock::iterator II,
-                                      unsigned FrameIndex, int SPAdj,
-                                      RegScavenger *RS) const {
+                                      unsigned FrameIndex) const {
   // Get the instruction.
   MachineInstr &MI = *II;       // ; SPILL_CR <SrcReg>, <offset>
   // Get the instruction's basic block.
   MachineBasicBlock &MBB = *MI.getParent();
+  MachineFunction &MF = *MBB.getParent();
   DebugLoc dl = MI.getDebugLoc();
 
-  // FIXME: Once LLVM supports creating virtual registers here, or the register
-  // scavenger can return multiple registers, stop using reserved registers
-  // here.
-  (void) SPAdj;
-  (void) RS;
-
   bool LP64 = Subtarget.isPPC64();
-  unsigned Reg = Subtarget.isDarwinABI() ?  (LP64 ? PPC::X2 : PPC::R2) :
-                                            (LP64 ? PPC::X0 : PPC::R0);
+  const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
+  const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
+
+  unsigned Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC);
   unsigned SrcReg = MI.getOperand(0).getReg();
 
   // We need to store the CR in the low 4-bits of the saved value. First, issue
@@ -424,16 +321,20 @@ void PPCRegisterInfo::lowerCRSpilling(MachineBasicBlock::iterator II,
     
   // If the saved register wasn't CR0, shift the bits left so that they are in
   // CR0's slot.
-  if (SrcReg != PPC::CR0)
+  if (SrcReg != PPC::CR0) {
+    unsigned Reg1 = Reg;
+    Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC);
+
     // rlwinm rA, rA, ShiftBits, 0, 31.
     BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::RLWINM8 : PPC::RLWINM), Reg)
-      .addReg(Reg, RegState::Kill)
-      .addImm(getPPCRegisterNumbering(SrcReg) * 4)
+      .addReg(Reg1, RegState::Kill)
+      .addImm(getEncodingValue(SrcReg) * 4)
       .addImm(0)
       .addImm(31);
+  }
 
   addFrameReference(BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::STW8 : PPC::STW))
-                    .addReg(Reg, getKillRegState(MI.getOperand(1).getImm())),
+                    .addReg(Reg, RegState::Kill),
                     FrameIndex);
 
   // Discard the pseudo instruction.
@@ -441,23 +342,19 @@ void PPCRegisterInfo::lowerCRSpilling(MachineBasicBlock::iterator II,
 }
 
 void PPCRegisterInfo::lowerCRRestore(MachineBasicBlock::iterator II,
-                                      unsigned FrameIndex, int SPAdj,
-                                      RegScavenger *RS) const {
+                                      unsigned FrameIndex) const {
   // Get the instruction.
   MachineInstr &MI = *II;       // ; <DestReg> = RESTORE_CR <offset>
   // Get the instruction's basic block.
   MachineBasicBlock &MBB = *MI.getParent();
+  MachineFunction &MF = *MBB.getParent();
   DebugLoc dl = MI.getDebugLoc();
 
-  // FIXME: Once LLVM supports creating virtual registers here, or the register
-  // scavenger can return multiple registers, stop using reserved registers
-  // here.
-  (void) SPAdj;
-  (void) RS;
-
   bool LP64 = Subtarget.isPPC64();
-  unsigned Reg = Subtarget.isDarwinABI() ?  (LP64 ? PPC::X2 : PPC::R2) :
-                                            (LP64 ? PPC::X0 : PPC::R0);
+  const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
+  const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
+
+  unsigned Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC);
   unsigned DestReg = MI.getOperand(0).getReg();
   assert(MI.definesRegister(DestReg) &&
     "RESTORE_CR does not define its destination");
@@ -468,15 +365,67 @@ void PPCRegisterInfo::lowerCRRestore(MachineBasicBlock::iterator II,
   // If the reloaded register isn't CR0, shift the bits right so that they are
   // in the right CR's slot.
   if (DestReg != PPC::CR0) {
-    unsigned ShiftBits = getPPCRegisterNumbering(DestReg)*4;
+    unsigned Reg1 = Reg;
+    Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC);
+
+    unsigned ShiftBits = getEncodingValue(DestReg)*4;
     // rlwinm r11, r11, 32-ShiftBits, 0, 31.
     BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::RLWINM8 : PPC::RLWINM), Reg)
-             .addReg(Reg).addImm(32-ShiftBits).addImm(0)
+             .addReg(Reg1, RegState::Kill).addImm(32-ShiftBits).addImm(0)
              .addImm(31);
   }
 
   BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::MTCRF8 : PPC::MTCRF), DestReg)
-             .addReg(Reg);
+             .addReg(Reg, RegState::Kill);
+
+  // Discard the pseudo instruction.
+  MBB.erase(II);
+}
+
+void PPCRegisterInfo::lowerVRSAVESpilling(MachineBasicBlock::iterator II,
+                                          unsigned FrameIndex) const {
+  // Get the instruction.
+  MachineInstr &MI = *II;       // ; SPILL_VRSAVE <SrcReg>, <offset>
+  // Get the instruction's basic block.
+  MachineBasicBlock &MBB = *MI.getParent();
+  MachineFunction &MF = *MBB.getParent();
+  DebugLoc dl = MI.getDebugLoc();
+
+  const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
+  unsigned Reg = MF.getRegInfo().createVirtualRegister(GPRC);
+  unsigned SrcReg = MI.getOperand(0).getReg();
+
+  BuildMI(MBB, II, dl, TII.get(PPC::MFVRSAVEv), Reg)
+          .addReg(SrcReg, getKillRegState(MI.getOperand(0).isKill()));
+    
+  addFrameReference(BuildMI(MBB, II, dl, TII.get(PPC::STW))
+                    .addReg(Reg, RegState::Kill),
+                    FrameIndex);
+
+  // Discard the pseudo instruction.
+  MBB.erase(II);
+}
+
+void PPCRegisterInfo::lowerVRSAVERestore(MachineBasicBlock::iterator II,
+                                         unsigned FrameIndex) const {
+  // Get the instruction.
+  MachineInstr &MI = *II;       // ; <DestReg> = RESTORE_VRSAVE <offset>
+  // Get the instruction's basic block.
+  MachineBasicBlock &MBB = *MI.getParent();
+  MachineFunction &MF = *MBB.getParent();
+  DebugLoc dl = MI.getDebugLoc();
+
+  const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
+  unsigned Reg = MF.getRegInfo().createVirtualRegister(GPRC);
+  unsigned DestReg = MI.getOperand(0).getReg();
+  assert(MI.definesRegister(DestReg) &&
+    "RESTORE_VRSAVE does not define its destination");
+
+  addFrameReference(BuildMI(MBB, II, dl, TII.get(PPC::LWZ),
+                              Reg), FrameIndex);
+
+  BuildMI(MBB, II, dl, TII.get(PPC::MTVRSAVEv), DestReg)
+             .addReg(Reg, RegState::Kill);
 
   // Discard the pseudo instruction.
   MBB.erase(II);
@@ -489,18 +438,14 @@ PPCRegisterInfo::hasReservedSpillSlot(const MachineFunction &MF,
   // For the nonvolatile condition registers (CR2, CR3, CR4) in an SVR4
   // ABI, return true to prevent allocating an additional frame slot.
   // For 64-bit, the CR save area is at SP+8; the value of FrameIdx = 0
-  // is arbitrary and will be subsequently ignored.  For 32-bit, we must
-  // create exactly one stack slot and return its FrameIdx for all
-  // nonvolatiles.
+  // is arbitrary and will be subsequently ignored.  For 32-bit, we have
+  // previously created the stack slot if needed, so return its FrameIdx.
   if (Subtarget.isSVR4ABI() && PPC::CR2 <= Reg && Reg <= PPC::CR4) {
-    if (Subtarget.isPPC64()) {
+    if (Subtarget.isPPC64())
       FrameIdx = 0;
-    } else if (CRSpillFrameIdx) {
-      FrameIdx = CRSpillFrameIdx;
-    } else {
-      MachineFrameInfo *MFI = ((MachineFunction &)MF).getFrameInfo();
-      FrameIdx = MFI->CreateFixedObject((uint64_t)4, (int64_t)-4, true);
-      CRSpillFrameIdx = FrameIdx;
+    else {
+      const PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
+      FrameIdx = FI->getCRSpillFrameIndex();
     }
     return true;
   }
@@ -509,7 +454,8 @@ PPCRegisterInfo::hasReservedSpillSlot(const MachineFunction &MF,
 
 void
 PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
-                                     int SPAdj, RegScavenger *RS) const {
+                                     int SPAdj, unsigned FIOperandNum,
+                                     RegScavenger *RS) const {
   assert(SPAdj == 0 && "Unexpected");
 
   // Get the instruction.
@@ -523,20 +469,13 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
   DebugLoc dl = MI.getDebugLoc();
 
-  // Find out which operand is the frame index.
-  unsigned FIOperandNo = 0;
-  while (!MI.getOperand(FIOperandNo).isFI()) {
-    ++FIOperandNo;
-    assert(FIOperandNo != MI.getNumOperands() &&
-           "Instr doesn't have FrameIndex operand!");
-  }
   // Take into account whether it's an add or mem instruction
-  unsigned OffsetOperandNo = (FIOperandNo == 2) ? 1 : 2;
+  unsigned OffsetOperandNo = (FIOperandNum == 2) ? 1 : 2;
   if (MI.isInlineAsm())
-    OffsetOperandNo = FIOperandNo-1;
+    OffsetOperandNo = FIOperandNum-1;
 
   // Get the frame index.
-  int FrameIndex = MI.getOperand(FIOperandNo).getIndex();
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
 
   // Get the frame pointer save index.  Users of this index are primarily
   // DYNALLOC instructions.
@@ -548,25 +487,29 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   // Special case for dynamic alloca.
   if (FPSI && FrameIndex == FPSI &&
       (OpC == PPC::DYNALLOC || OpC == PPC::DYNALLOC8)) {
-    lowerDynamicAlloc(II, SPAdj, RS);
+    lowerDynamicAlloc(II);
     return;
   }
 
-  // Special case for pseudo-ops SPILL_CR and RESTORE_CR.
-  if (requiresRegisterScavenging(MF)) {
-    if (OpC == PPC::SPILL_CR) {
-      lowerCRSpilling(II, FrameIndex, SPAdj, RS);
-      return;
-    } else if (OpC == PPC::RESTORE_CR) {
-      lowerCRRestore(II, FrameIndex, SPAdj, RS);
-      return;
-    }
+  // Special case for pseudo-ops SPILL_CR and RESTORE_CR, etc.
+  if (OpC == PPC::SPILL_CR) {
+    lowerCRSpilling(II, FrameIndex);
+    return;
+  } else if (OpC == PPC::RESTORE_CR) {
+    lowerCRRestore(II, FrameIndex);
+    return;
+  } else if (OpC == PPC::SPILL_VRSAVE) {
+    lowerVRSAVESpilling(II, FrameIndex);
+    return;
+  } else if (OpC == PPC::RESTORE_VRSAVE) {
+    lowerVRSAVERestore(II, FrameIndex);
+    return;
   }
 
   // Replace the FrameIndex with base register with GPR1 (SP) or GPR31 (FP).
 
   bool is64Bit = Subtarget.isPPC64();
-  MI.getOperand(FIOperandNo).ChangeToRegister(TFI->hasFP(MF) ?
+  MI.getOperand(FIOperandNum).ChangeToRegister(TFI->hasFP(MF) ?
                                               (is64Bit ? PPC::X31 : PPC::R31) :
                                                 (is64Bit ? PPC::X1 : PPC::R1),
                                               false);
@@ -579,11 +522,14 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   case PPC::LWA:
   case PPC::LD:
   case PPC::STD:
-  case PPC::STD_32:
     isIXAddr = true;
     break;
   }
-  
+
+  // If the instruction is not present in ImmToIdxMap, then it has no immediate
+  // form (and must be r+r).
+  bool noImmForm = !MI.isInlineAsm() && !ImmToIdxMap.count(OpC);
+
   // Now add the frame object offset to the offset from r1.
   int Offset = MFI->getObjectOffset(FrameIndex);
   if (!isIXAddr)
@@ -596,7 +542,8 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   // to Offset to get the correct offset.
   // Naked functions have stack size 0, although getStackSize may not reflect that
   // because we didn't call all the pieces that compute it for naked functions.
-  if (!MF.getFunction()->getFnAttributes().hasAttribute(Attributes::Naked))
+  if (!MF.getFunction()->getAttributes().
+        hasAttribute(AttributeSet::FunctionIndex, Attribute::Naked))
     Offset += MFI->getStackSize();
 
   // If we can, encode the offset directly into the instruction.  If this is a
@@ -606,7 +553,8 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   // only "std" to a stack slot that is at least 4-byte aligned, but it can
   // happen in invalid code.
   if (OpC == PPC::DBG_VALUE || // DBG_VALUE is always Reg+Imm
-      (isInt<16>(Offset) && (!isIXAddr || (Offset & 3) == 0))) {
+      (!noImmForm &&
+       isInt<16>(Offset) && (!isIXAddr || (Offset & 3) == 0))) {
     if (isIXAddr)
       Offset >>= 2;    // The actual encoded value has the low two bits zero.
     MI.getOperand(OffsetOperandNo).ChangeToImmediate(Offset);
@@ -616,19 +564,17 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   // The offset doesn't fit into a single register, scavenge one to build the
   // offset in.
 
-  unsigned SReg;
-  if (requiresRegisterScavenging(MF)) {
-    const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
-    const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
-    SReg = findScratchRegister(II, RS, is64Bit ? G8RC : GPRC, SPAdj);
-  } else
-    SReg = is64Bit ? PPC::X0 : PPC::R0;
+  const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
+  const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
+  const TargetRegisterClass *RC = is64Bit ? G8RC : GPRC;
+  unsigned SRegHi = MF.getRegInfo().createVirtualRegister(RC),
+           SReg = MF.getRegInfo().createVirtualRegister(RC);
 
   // Insert a set of rA with the full offset value before the ld, st, or add
-  BuildMI(MBB, II, dl, TII.get(is64Bit ? PPC::LIS8 : PPC::LIS), SReg)
+  BuildMI(MBB, II, dl, TII.get(is64Bit ? PPC::LIS8 : PPC::LIS), SRegHi)
     .addImm(Offset >> 16);
   BuildMI(MBB, II, dl, TII.get(is64Bit ? PPC::ORI8 : PPC::ORI), SReg)
-    .addReg(SReg, RegState::Kill)
+    .addReg(SRegHi, RegState::Kill)
     .addImm(Offset);
 
   // Convert into indexed form of the instruction:
@@ -637,7 +583,9 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   //   addi 0:rA 1:rB, 2, imm ==> add 0:rA, 1:rB, 2:r0
   unsigned OperandBase;
 
-  if (OpC != TargetOpcode::INLINEASM) {
+  if (noImmForm)
+    OperandBase = 1;
+  else if (OpC != TargetOpcode::INLINEASM) {
     assert(ImmToIdxMap.count(OpC) &&
            "No indexed form of load or store available!");
     unsigned NewOpcode = ImmToIdxMap.find(OpC)->second;
@@ -647,7 +595,7 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
     OperandBase = OffsetOperandNo;
   }
 
-  unsigned StackReg = MI.getOperand(FIOperandNo).getReg();
+  unsigned StackReg = MI.getOperand(FIOperandNum).getReg();
   MI.getOperand(OperandBase).ChangeToRegister(StackReg, false);
   MI.getOperand(OperandBase + 1).ChangeToRegister(SReg, false, false, true);
 }
diff --git a/lib/Target/PowerPC/PPCRegisterInfo.h b/lib/Target/PowerPC/PPCRegisterInfo.h
index a8fd796d9e97..7e6683eeb2ef 100644
--- a/lib/Target/PowerPC/PPCRegisterInfo.h
+++ b/lib/Target/PowerPC/PPCRegisterInfo.h
@@ -15,8 +15,8 @@
 #ifndef POWERPC32_REGISTERINFO_H
 #define POWERPC32_REGISTERINFO_H
 
+#include "llvm/ADT/DenseMap.h"
 #include "PPC.h"
-#include <map>
 
 #define GET_REGINFO_HEADER
 #include "PPCGenRegisterInfo.inc"
@@ -27,10 +27,9 @@ class TargetInstrInfo;
 class Type;
 
 class PPCRegisterInfo : public PPCGenRegisterInfo {
-  std::map<unsigned, unsigned> ImmToIdxMap;
+  DenseMap<unsigned, unsigned> ImmToIdxMap;
   const PPCSubtarget &Subtarget;
   const TargetInstrInfo &TII;
-  mutable int CRSpillFrameIdx;
 public:
   PPCRegisterInfo(const PPCSubtarget &SubTarget, const TargetInstrInfo &tii);
   
@@ -45,31 +44,38 @@ public:
   /// Code Generation virtual methods...
   const uint16_t *getCalleeSavedRegs(const MachineFunction* MF = 0) const;
   const uint32_t *getCallPreservedMask(CallingConv::ID CC) const;
+  const uint32_t *getNoPreservedMask() const;
 
   BitVector getReservedRegs(const MachineFunction &MF) const;
 
-  virtual bool avoidWriteAfterWrite(const TargetRegisterClass *RC) const;
+  /// We require the register scavenger.
+  bool requiresRegisterScavenging(const MachineFunction &MF) const {
+    return true;
+  }
+
+  bool requiresFrameIndexScavenging(const MachineFunction &MF) const {
+    return true;
+  }
+
+  bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const {
+    return true;
+  }
+
+  void lowerDynamicAlloc(MachineBasicBlock::iterator II) const;
+  void lowerCRSpilling(MachineBasicBlock::iterator II,
+                       unsigned FrameIndex) const;
+  void lowerCRRestore(MachineBasicBlock::iterator II,
+                      unsigned FrameIndex) const;
+  void lowerVRSAVESpilling(MachineBasicBlock::iterator II,
+                           unsigned FrameIndex) const;
+  void lowerVRSAVERestore(MachineBasicBlock::iterator II,
+                          unsigned FrameIndex) const;
 
-  /// requiresRegisterScavenging - We require a register scavenger.
-  /// FIXME (64-bit): Should be inlined.
-  bool requiresRegisterScavenging(const MachineFunction &MF) const;
-
-  bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const;
-
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
-
-  void lowerDynamicAlloc(MachineBasicBlock::iterator II,
-                         int SPAdj, RegScavenger *RS) const;
-  void lowerCRSpilling(MachineBasicBlock::iterator II, unsigned FrameIndex,
-                       int SPAdj, RegScavenger *RS) const;
-  void lowerCRRestore(MachineBasicBlock::iterator II, unsigned FrameIndex,
-                       int SPAdj, RegScavenger *RS) const;
   bool hasReservedSpillSlot(const MachineFunction &MF, unsigned Reg,
 			    int &FrameIdx) const;
   void eliminateFrameIndex(MachineBasicBlock::iterator II,
-                           int SPAdj, RegScavenger *RS = NULL) const;
+                           int SPAdj, unsigned FIOperandNum,
+                           RegScavenger *RS = NULL) const;
 
   // Debug information queries.
   unsigned getFrameRegister(const MachineFunction &MF) const;
diff --git a/lib/Target/PowerPC/PPCRegisterInfo.td b/lib/Target/PowerPC/PPCRegisterInfo.td
index 5ca387629b6c..57a25f5143fa 100644
--- a/lib/Target/PowerPC/PPCRegisterInfo.td
+++ b/lib/Target/PowerPC/PPCRegisterInfo.td
@@ -27,178 +27,72 @@ class PPCReg<string n> : Register<n> {
 
 // GPR - One of the 32 32-bit general-purpose registers
 class GPR<bits<5> num, string n> : PPCReg<n> {
-  field bits<5> Num = num;
+  let HWEncoding{4-0} = num;
 }
 
 // GP8 - One of the 32 64-bit general-purpose registers
 class GP8<GPR SubReg, string n> : PPCReg<n> {
-  field bits<5> Num = SubReg.Num;
+  let HWEncoding = SubReg.HWEncoding;
   let SubRegs = [SubReg];
   let SubRegIndices = [sub_32];
 }
 
 // SPR - One of the 32-bit special-purpose registers
 class SPR<bits<10> num, string n> : PPCReg<n> {
-  field bits<10> Num = num;
+  let HWEncoding{9-0} = num;
 }
 
 // FPR - One of the 32 64-bit floating-point registers
 class FPR<bits<5> num, string n> : PPCReg<n> {
-  field bits<5> Num = num;
+  let HWEncoding{4-0} = num;
 }
 
 // VR - One of the 32 128-bit vector registers
 class VR<bits<5> num, string n> : PPCReg<n> {
-  field bits<5> Num = num;
+  let HWEncoding{4-0} = num;
 }
 
 // CR - One of the 8 4-bit condition registers
 class CR<bits<3> num, string n, list<Register> subregs> : PPCReg<n> {
-  field bits<3> Num = num;
+  let HWEncoding{2-0} = num;
   let SubRegs = subregs;
 }
 
 // CRBIT - One of the 32 1-bit condition register fields
 class CRBIT<bits<5> num, string n> : PPCReg<n> {
-  field bits<5> Num = num;
+  let HWEncoding{4-0} = num;
 }
 
-
 // General-purpose registers
-def R0  : GPR< 0,  "r0">, DwarfRegNum<[-2, 0]>;
-def R1  : GPR< 1,  "r1">, DwarfRegNum<[-2, 1]>;
-def R2  : GPR< 2,  "r2">, DwarfRegNum<[-2, 2]>;
-def R3  : GPR< 3,  "r3">, DwarfRegNum<[-2, 3]>;
-def R4  : GPR< 4,  "r4">, DwarfRegNum<[-2, 4]>;
-def R5  : GPR< 5,  "r5">, DwarfRegNum<[-2, 5]>;
-def R6  : GPR< 6,  "r6">, DwarfRegNum<[-2, 6]>;
-def R7  : GPR< 7,  "r7">, DwarfRegNum<[-2, 7]>;
-def R8  : GPR< 8,  "r8">, DwarfRegNum<[-2, 8]>;
-def R9  : GPR< 9,  "r9">, DwarfRegNum<[-2, 9]>;
-def R10 : GPR<10, "r10">, DwarfRegNum<[-2, 10]>;
-def R11 : GPR<11, "r11">, DwarfRegNum<[-2, 11]>;
-def R12 : GPR<12, "r12">, DwarfRegNum<[-2, 12]>;
-def R13 : GPR<13, "r13">, DwarfRegNum<[-2, 13]>;
-def R14 : GPR<14, "r14">, DwarfRegNum<[-2, 14]>;
-def R15 : GPR<15, "r15">, DwarfRegNum<[-2, 15]>;
-def R16 : GPR<16, "r16">, DwarfRegNum<[-2, 16]>;
-def R17 : GPR<17, "r17">, DwarfRegNum<[-2, 17]>;
-def R18 : GPR<18, "r18">, DwarfRegNum<[-2, 18]>;
-def R19 : GPR<19, "r19">, DwarfRegNum<[-2, 19]>;
-def R20 : GPR<20, "r20">, DwarfRegNum<[-2, 20]>;
-def R21 : GPR<21, "r21">, DwarfRegNum<[-2, 21]>;
-def R22 : GPR<22, "r22">, DwarfRegNum<[-2, 22]>;
-def R23 : GPR<23, "r23">, DwarfRegNum<[-2, 23]>;
-def R24 : GPR<24, "r24">, DwarfRegNum<[-2, 24]>;
-def R25 : GPR<25, "r25">, DwarfRegNum<[-2, 25]>;
-def R26 : GPR<26, "r26">, DwarfRegNum<[-2, 26]>;
-def R27 : GPR<27, "r27">, DwarfRegNum<[-2, 27]>;
-def R28 : GPR<28, "r28">, DwarfRegNum<[-2, 28]>;
-def R29 : GPR<29, "r29">, DwarfRegNum<[-2, 29]>;
-def R30 : GPR<30, "r30">, DwarfRegNum<[-2, 30]>;
-def R31 : GPR<31, "r31">, DwarfRegNum<[-2, 31]>;
+foreach Index = 0-31 in {
+  def R#Index : GPR<Index, "r"#Index>, DwarfRegNum<[-2, Index]>;
+}
 
 // 64-bit General-purpose registers
-def X0  : GP8< R0,  "r0">, DwarfRegNum<[0, -2]>;
-def X1  : GP8< R1,  "r1">, DwarfRegNum<[1, -2]>;
-def X2  : GP8< R2,  "r2">, DwarfRegNum<[2, -2]>;
-def X3  : GP8< R3,  "r3">, DwarfRegNum<[3, -2]>;
-def X4  : GP8< R4,  "r4">, DwarfRegNum<[4, -2]>;
-def X5  : GP8< R5,  "r5">, DwarfRegNum<[5, -2]>;
-def X6  : GP8< R6,  "r6">, DwarfRegNum<[6, -2]>;
-def X7  : GP8< R7,  "r7">, DwarfRegNum<[7, -2]>;
-def X8  : GP8< R8,  "r8">, DwarfRegNum<[8, -2]>;
-def X9  : GP8< R9,  "r9">, DwarfRegNum<[9, -2]>;
-def X10 : GP8<R10, "r10">, DwarfRegNum<[10, -2]>;
-def X11 : GP8<R11, "r11">, DwarfRegNum<[11, -2]>;
-def X12 : GP8<R12, "r12">, DwarfRegNum<[12, -2]>;
-def X13 : GP8<R13, "r13">, DwarfRegNum<[13, -2]>;
-def X14 : GP8<R14, "r14">, DwarfRegNum<[14, -2]>;
-def X15 : GP8<R15, "r15">, DwarfRegNum<[15, -2]>;
-def X16 : GP8<R16, "r16">, DwarfRegNum<[16, -2]>;
-def X17 : GP8<R17, "r17">, DwarfRegNum<[17, -2]>;
-def X18 : GP8<R18, "r18">, DwarfRegNum<[18, -2]>;
-def X19 : GP8<R19, "r19">, DwarfRegNum<[19, -2]>;
-def X20 : GP8<R20, "r20">, DwarfRegNum<[20, -2]>;
-def X21 : GP8<R21, "r21">, DwarfRegNum<[21, -2]>;
-def X22 : GP8<R22, "r22">, DwarfRegNum<[22, -2]>;
-def X23 : GP8<R23, "r23">, DwarfRegNum<[23, -2]>;
-def X24 : GP8<R24, "r24">, DwarfRegNum<[24, -2]>;
-def X25 : GP8<R25, "r25">, DwarfRegNum<[25, -2]>;
-def X26 : GP8<R26, "r26">, DwarfRegNum<[26, -2]>;
-def X27 : GP8<R27, "r27">, DwarfRegNum<[27, -2]>;
-def X28 : GP8<R28, "r28">, DwarfRegNum<[28, -2]>;
-def X29 : GP8<R29, "r29">, DwarfRegNum<[29, -2]>;
-def X30 : GP8<R30, "r30">, DwarfRegNum<[30, -2]>;
-def X31 : GP8<R31, "r31">, DwarfRegNum<[31, -2]>;
+foreach Index = 0-31 in {
+  def X#Index : GP8<!cast<GPR>("R"#Index), "r"#Index>,
+                    DwarfRegNum<[Index, -2]>;
+}
 
 // Floating-point registers
-def F0  : FPR< 0,  "f0">, DwarfRegNum<[32, 32]>;
-def F1  : FPR< 1,  "f1">, DwarfRegNum<[33, 33]>;
-def F2  : FPR< 2,  "f2">, DwarfRegNum<[34, 34]>;
-def F3  : FPR< 3,  "f3">, DwarfRegNum<[35, 35]>;
-def F4  : FPR< 4,  "f4">, DwarfRegNum<[36, 36]>;
-def F5  : FPR< 5,  "f5">, DwarfRegNum<[37, 37]>;
-def F6  : FPR< 6,  "f6">, DwarfRegNum<[38, 38]>;
-def F7  : FPR< 7,  "f7">, DwarfRegNum<[39, 39]>;
-def F8  : FPR< 8,  "f8">, DwarfRegNum<[40, 40]>;
-def F9  : FPR< 9,  "f9">, DwarfRegNum<[41, 41]>;
-def F10 : FPR<10, "f10">, DwarfRegNum<[42, 42]>;
-def F11 : FPR<11, "f11">, DwarfRegNum<[43, 43]>;
-def F12 : FPR<12, "f12">, DwarfRegNum<[44, 44]>;
-def F13 : FPR<13, "f13">, DwarfRegNum<[45, 45]>;
-def F14 : FPR<14, "f14">, DwarfRegNum<[46, 46]>;
-def F15 : FPR<15, "f15">, DwarfRegNum<[47, 47]>;
-def F16 : FPR<16, "f16">, DwarfRegNum<[48, 48]>;
-def F17 : FPR<17, "f17">, DwarfRegNum<[49, 49]>;
-def F18 : FPR<18, "f18">, DwarfRegNum<[50, 50]>;
-def F19 : FPR<19, "f19">, DwarfRegNum<[51, 51]>;
-def F20 : FPR<20, "f20">, DwarfRegNum<[52, 52]>;
-def F21 : FPR<21, "f21">, DwarfRegNum<[53, 53]>;
-def F22 : FPR<22, "f22">, DwarfRegNum<[54, 54]>;
-def F23 : FPR<23, "f23">, DwarfRegNum<[55, 55]>;
-def F24 : FPR<24, "f24">, DwarfRegNum<[56, 56]>;
-def F25 : FPR<25, "f25">, DwarfRegNum<[57, 57]>;
-def F26 : FPR<26, "f26">, DwarfRegNum<[58, 58]>;
-def F27 : FPR<27, "f27">, DwarfRegNum<[59, 59]>;
-def F28 : FPR<28, "f28">, DwarfRegNum<[60, 60]>;
-def F29 : FPR<29, "f29">, DwarfRegNum<[61, 61]>;
-def F30 : FPR<30, "f30">, DwarfRegNum<[62, 62]>;
-def F31 : FPR<31, "f31">, DwarfRegNum<[63, 63]>;
+foreach Index = 0-31 in {
+  def F#Index : FPR<Index, "f"#Index>,
+                DwarfRegNum<[!add(Index, 32), !add(Index, 32)]>;
+}
 
 // Vector registers
-def V0  : VR< 0,  "v0">, DwarfRegNum<[77, 77]>;
-def V1  : VR< 1,  "v1">, DwarfRegNum<[78, 78]>;
-def V2  : VR< 2,  "v2">, DwarfRegNum<[79, 79]>;
-def V3  : VR< 3,  "v3">, DwarfRegNum<[80, 80]>;
-def V4  : VR< 4,  "v4">, DwarfRegNum<[81, 81]>;
-def V5  : VR< 5,  "v5">, DwarfRegNum<[82, 82]>;
-def V6  : VR< 6,  "v6">, DwarfRegNum<[83, 83]>;
-def V7  : VR< 7,  "v7">, DwarfRegNum<[84, 84]>;
-def V8  : VR< 8,  "v8">, DwarfRegNum<[85, 85]>;
-def V9  : VR< 9,  "v9">, DwarfRegNum<[86, 86]>;
-def V10 : VR<10, "v10">, DwarfRegNum<[87, 87]>;
-def V11 : VR<11, "v11">, DwarfRegNum<[88, 88]>;
-def V12 : VR<12, "v12">, DwarfRegNum<[89, 89]>;
-def V13 : VR<13, "v13">, DwarfRegNum<[90, 90]>;
-def V14 : VR<14, "v14">, DwarfRegNum<[91, 91]>;
-def V15 : VR<15, "v15">, DwarfRegNum<[92, 92]>;
-def V16 : VR<16, "v16">, DwarfRegNum<[93, 93]>;
-def V17 : VR<17, "v17">, DwarfRegNum<[94, 94]>;
-def V18 : VR<18, "v18">, DwarfRegNum<[95, 95]>;
-def V19 : VR<19, "v19">, DwarfRegNum<[96, 96]>;
-def V20 : VR<20, "v20">, DwarfRegNum<[97, 97]>;
-def V21 : VR<21, "v21">, DwarfRegNum<[98, 98]>;
-def V22 : VR<22, "v22">, DwarfRegNum<[99, 99]>;
-def V23 : VR<23, "v23">, DwarfRegNum<[100, 100]>;
-def V24 : VR<24, "v24">, DwarfRegNum<[101, 101]>;
-def V25 : VR<25, "v25">, DwarfRegNum<[102, 102]>;
-def V26 : VR<26, "v26">, DwarfRegNum<[103, 103]>;
-def V27 : VR<27, "v27">, DwarfRegNum<[104, 104]>;
-def V28 : VR<28, "v28">, DwarfRegNum<[105, 105]>;
-def V29 : VR<29, "v29">, DwarfRegNum<[106, 106]>;
-def V30 : VR<30, "v30">, DwarfRegNum<[107, 107]>;
-def V31 : VR<31, "v31">, DwarfRegNum<[108, 108]>;
+foreach Index = 0-31 in {
+  def V#Index : VR<Index, "v"#Index>,
+                DwarfRegNum<[!add(Index, 77), !add(Index, 77)]>;
+}
+
+// The reprsentation of r0 when treated as the constant 0.
+def ZERO  : GPR<0, "0">;
+def ZERO8 : GP8<ZERO, "0">;
+
+// Representations of the frame pointer used by ISD::FRAMEADDR.
+def FP   : GPR<0 /* arbitrary */, "**FRAME POINTER**">;
+def FP8  : GP8<FP, "**FRAME POINTER**">;
 
 // Condition register bits
 def CR0LT : CRBIT< 0, "0">;
@@ -278,11 +172,17 @@ def RM: SPR<512, "**ROUNDING MODE**">;
 // then nonvolatiles in reverse order since stmw/lmw save from rN to r31
 def GPRC : RegisterClass<"PPC", [i32], 32, (add (sequence "R%u", 2, 12),
                                                 (sequence "R%u", 30, 13),
-                                                R31, R0, R1, LR)>;
+                                                R31, R0, R1, FP)>;
 
 def G8RC : RegisterClass<"PPC", [i64], 64, (add (sequence "X%u", 2, 12),
                                                 (sequence "X%u", 30, 14),
-                                                X31, X13, X0, X1, LR8)>;
+                                                X31, X13, X0, X1, FP8)>;
+
+// For some instructions r0 is special (representing the value 0 instead of
+// the value in the r0 register), and we use these register subclasses to
+// prevent r0 from being allocated for use by those instructions.
+def GPRC_NOR0 : RegisterClass<"PPC", [i32], 32, (add (sub GPRC, R0), ZERO)>;
+def G8RC_NOX0 : RegisterClass<"PPC", [i64], 64, (add (sub G8RC, X0), ZERO8)>;
 
 // Allocate volatiles first, then non-volatiles in reverse order. With the SVR4
 // ABI the size of the Floating-point register save area is determined by the
diff --git a/lib/Target/PowerPC/PPCScheduleA2.td b/lib/Target/PowerPC/PPCScheduleA2.td
index ba63b5cd8faf..ae084aa0e8c1 100644
--- a/lib/Target/PowerPC/PPCScheduleA2.td
+++ b/lib/Target/PowerPC/PPCScheduleA2.td
@@ -749,3 +749,18 @@ def PPCA2Itineraries : ProcessorItineraries<
                               [15, 7],
                               [FPR_Bypass, FPR_Bypass]>
 ]>;
+
+// ===---------------------------------------------------------------------===//
+// A2 machine model for scheduling and other instruction cost heuristics.
+
+def PPCA2Model : SchedMachineModel {
+  let IssueWidth = 1;  // 2 micro-ops are dispatched per cycle.
+  let MinLatency = -1; // OperandCycles are interpreted as MinLatency.
+  let LoadLatency = 6; // Optimistic load latency assuming bypass.
+                       // This is overriden by OperandCycles if the
+                       // Itineraries are queried instead.
+  let MispredictPenalty = 6;
+
+  let Itineraries = PPCA2Itineraries;
+}
+
diff --git a/lib/Target/PowerPC/PPCScheduleG5.td b/lib/Target/PowerPC/PPCScheduleG5.td
index 7c02ea099c14..c64998d52a0c 100644
--- a/lib/Target/PowerPC/PPCScheduleG5.td
+++ b/lib/Target/PowerPC/PPCScheduleG5.td
@@ -92,3 +92,18 @@ def G5Itineraries : ProcessorItineraries<
   InstrItinData<VecVSL      , [InstrStage<2, [VIU1]>]>,
   InstrItinData<VecVSR      , [InstrStage<3, [VPU]>]>
 ]>;
+
+// ===---------------------------------------------------------------------===//
+// e5500 machine model for scheduling and other instruction cost heuristics.
+
+def G5Model : SchedMachineModel {
+  let IssueWidth = 4;  // 4 (non-branch) instructions are dispatched per cycle.
+  let MinLatency = 0;  // Out-of-order dispatch.
+  let LoadLatency = 3; // Optimistic load latency assuming bypass.
+                       // This is overriden by OperandCycles if the
+                       // Itineraries are queried instead.
+  let MispredictPenalty = 16;
+
+  let Itineraries = G5Itineraries;
+}
+
diff --git a/lib/Target/PowerPC/PPCSubtarget.cpp b/lib/Target/PowerPC/PPCSubtarget.cpp
index 9c8cb92cc7ea..a8f2b3f47d1b 100644
--- a/lib/Target/PowerPC/PPCSubtarget.cpp
+++ b/lib/Target/PowerPC/PPCSubtarget.cpp
@@ -12,12 +12,12 @@
 //===----------------------------------------------------------------------===//
 
 #include "PPCSubtarget.h"
-#include "PPCRegisterInfo.h"
 #include "PPC.h"
-#include "llvm/GlobalValue.h"
-#include "llvm/Target/TargetMachine.h"
+#include "PPCRegisterInfo.h"
+#include "llvm/IR/GlobalValue.h"
 #include "llvm/Support/Host.h"
 #include "llvm/Support/TargetRegistry.h"
+#include "llvm/Target/TargetMachine.h"
 #include <cstdlib>
 
 #define GET_SUBTARGETINFO_TARGET_DESC
@@ -36,9 +36,20 @@ PPCSubtarget::PPCSubtarget(const std::string &TT, const std::string &CPU,
   , Use64BitRegs(false)
   , IsPPC64(is64Bit)
   , HasAltivec(false)
+  , HasQPX(false)
   , HasFSQRT(false)
+  , HasFRE(false)
+  , HasFRES(false)
+  , HasFRSQRTE(false)
+  , HasFRSQRTES(false)
+  , HasRecipPrec(false)
   , HasSTFIWX(false)
+  , HasLFIWAX(false)
+  , HasFPRND(false)
+  , HasFPCVT(false)
   , HasISEL(false)
+  , HasPOPCNTD(false)
+  , HasLDBRX(false)
   , IsBookE(false)
   , HasLazyResolverStubs(false)
   , IsJITCodeModel(false)
@@ -82,6 +93,12 @@ PPCSubtarget::PPCSubtarget(const std::string &TT, const std::string &CPU,
   // Set up darwin-specific properties.
   if (isDarwin())
     HasLazyResolverStubs = true;
+
+  // QPX requires a 32-byte aligned stack. Note that we need to do this if
+  // we're compiling for a BG/Q system regardless of whether or not QPX
+  // is enabled because external functions will assume this alignment.
+  if (hasQPX() || isBGQ())
+    StackAlignment = 32;
 }
 
 /// SetJITMode - This is called to inform the subtarget info that we are
diff --git a/lib/Target/PowerPC/PPCSubtarget.h b/lib/Target/PowerPC/PPCSubtarget.h
index b9e22f43c39e..65b4d211fc6a 100644
--- a/lib/Target/PowerPC/PPCSubtarget.h
+++ b/lib/Target/PowerPC/PPCSubtarget.h
@@ -14,9 +14,9 @@
 #ifndef POWERPCSUBTARGET_H
 #define POWERPCSUBTARGET_H
 
-#include "llvm/Target/TargetSubtargetInfo.h"
-#include "llvm/MC/MCInstrItineraries.h"
 #include "llvm/ADT/Triple.h"
+#include "llvm/MC/MCInstrItineraries.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
 #include <string>
 
 #define GET_SUBTARGETINFO_HEADER
@@ -43,7 +43,12 @@ namespace PPC {
     DIR_A2,
     DIR_E500mc,
     DIR_E5500,
+    DIR_PWR3,
+    DIR_PWR4,
+    DIR_PWR5,
+    DIR_PWR5X,
     DIR_PWR6,
+    DIR_PWR6X,
     DIR_PWR7,
     DIR_64
   };
@@ -70,9 +75,17 @@ protected:
   bool Use64BitRegs;
   bool IsPPC64;
   bool HasAltivec;
+  bool HasQPX;
   bool HasFSQRT;
+  bool HasFRE, HasFRES, HasFRSQRTE, HasFRSQRTES;
+  bool HasRecipPrec;
   bool HasSTFIWX;
+  bool HasLFIWAX;
+  bool HasFPRND;
+  bool HasFPCVT;
   bool HasISEL;
+  bool HasPOPCNTD;
+  bool HasLDBRX;
   bool IsBookE;
   bool HasLazyResolverStubs;
   bool IsJITCodeModel;
@@ -148,10 +161,21 @@ public:
 
   // Specific obvious features.
   bool hasFSQRT() const { return HasFSQRT; }
+  bool hasFRE() const { return HasFRE; }
+  bool hasFRES() const { return HasFRES; }
+  bool hasFRSQRTE() const { return HasFRSQRTE; }
+  bool hasFRSQRTES() const { return HasFRSQRTES; }
+  bool hasRecipPrec() const { return HasRecipPrec; }
   bool hasSTFIWX() const { return HasSTFIWX; }
+  bool hasLFIWAX() const { return HasLFIWAX; }
+  bool hasFPRND() const { return HasFPRND; }
+  bool hasFPCVT() const { return HasFPCVT; }
   bool hasAltivec() const { return HasAltivec; }
+  bool hasQPX() const { return HasQPX; }
   bool hasMFOCRF() const { return HasMFOCRF; }
   bool hasISEL() const { return HasISEL; }
+  bool hasPOPCNTD() const { return HasPOPCNTD; }
+  bool hasLDBRX() const { return HasLDBRX; }
   bool isBookE() const { return IsBookE; }
 
   const Triple &getTargetTriple() const { return TargetTriple; }
@@ -160,6 +184,8 @@ public:
   bool isDarwin() const { return TargetTriple.isMacOSX(); }
   /// isBGP - True if this is a BG/P platform.
   bool isBGP() const { return TargetTriple.getVendor() == Triple::BGP; }
+  /// isBGQ - True if this is a BG/Q platform.
+  bool isBGQ() const { return TargetTriple.getVendor() == Triple::BGQ; }
 
   bool isDarwinABI() const { return isDarwin(); }
   bool isSVR4ABI() const { return !isDarwin(); }
diff --git a/lib/Target/PowerPC/PPCTargetMachine.cpp b/lib/Target/PowerPC/PPCTargetMachine.cpp
index 3fc977ee2b41..fe851c1b6fb8 100644
--- a/lib/Target/PowerPC/PPCTargetMachine.cpp
+++ b/lib/Target/PowerPC/PPCTargetMachine.cpp
@@ -13,13 +13,13 @@
 
 #include "PPCTargetMachine.h"
 #include "PPC.h"
-#include "llvm/PassManager.h"
-#include "llvm/MC/MCStreamer.h"
 #include "llvm/CodeGen/Passes.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/FormattedStream.h"
 #include "llvm/Support/TargetRegistry.h"
+#include "llvm/Target/TargetOptions.h"
 using namespace llvm;
 
 static cl::
@@ -43,8 +43,7 @@ PPCTargetMachine::PPCTargetMachine(const Target &T, StringRef TT,
     DL(Subtarget.getDataLayoutString()), InstrInfo(*this),
     FrameLowering(Subtarget), JITInfo(*this, is64Bit),
     TLInfo(*this), TSInfo(*this),
-    InstrItins(Subtarget.getInstrItineraryData()),
-    STTI(&TLInfo), VTTI(&TLInfo) {
+    InstrItins(Subtarget.getInstrItineraryData()) {
 
   // The binutils for the BG/P are too old for CFI.
   if (Subtarget.isBGP())
@@ -127,3 +126,12 @@ bool PPCTargetMachine::addCodeEmitter(PassManagerBase &PM,
 
   return false;
 }
+
+void PPCTargetMachine::addAnalysisPasses(PassManagerBase &PM) {
+  // Add first the target-independent BasicTTI pass, then our PPC pass. This
+  // allows the PPC pass to delegate to the target independent layer when
+  // appropriate.
+  PM.add(createBasicTargetTransformInfoPass(getTargetLowering()));
+  PM.add(createPPCTargetTransformInfoPass(this));
+}
+
diff --git a/lib/Target/PowerPC/PPCTargetMachine.h b/lib/Target/PowerPC/PPCTargetMachine.h
index c168433a71b3..606ccb314126 100644
--- a/lib/Target/PowerPC/PPCTargetMachine.h
+++ b/lib/Target/PowerPC/PPCTargetMachine.h
@@ -15,14 +15,13 @@
 #define PPC_TARGETMACHINE_H
 
 #include "PPCFrameLowering.h"
-#include "PPCSubtarget.h"
-#include "PPCJITInfo.h"
-#include "PPCInstrInfo.h"
 #include "PPCISelLowering.h"
+#include "PPCInstrInfo.h"
+#include "PPCJITInfo.h"
 #include "PPCSelectionDAGInfo.h"
+#include "PPCSubtarget.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetTransformImpl.h"
-#include "llvm/DataLayout.h"
 
 namespace llvm {
 
@@ -37,8 +36,6 @@ class PPCTargetMachine : public LLVMTargetMachine {
   PPCTargetLowering   TLInfo;
   PPCSelectionDAGInfo TSInfo;
   InstrItineraryData  InstrItins;
-  ScalarTargetTransformImpl STTI;
-  VectorTargetTransformImpl VTTI;
 
 public:
   PPCTargetMachine(const Target &T, StringRef TT,
@@ -66,17 +63,14 @@ public:
   virtual const InstrItineraryData *getInstrItineraryData() const {
     return &InstrItins;
   }
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
 
   // Pass Pipeline Configuration
   virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
   virtual bool addCodeEmitter(PassManagerBase &PM,
                               JITCodeEmitter &JCE);
+
+  /// \brief Register PPC analysis passes with a pass manager.
+  virtual void addAnalysisPasses(PassManagerBase &PM);
 };
 
 /// PPC32TargetMachine - PowerPC 32-bit target machine.
diff --git a/lib/Target/PowerPC/PPCTargetTransformInfo.cpp b/lib/Target/PowerPC/PPCTargetTransformInfo.cpp
new file mode 100644
index 000000000000..2504ba70c25a
--- /dev/null
+++ b/lib/Target/PowerPC/PPCTargetTransformInfo.cpp
@@ -0,0 +1,240 @@
+//===-- PPCTargetTransformInfo.cpp - PPC specific TTI pass ----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+/// \file
+/// This file implements a TargetTransformInfo analysis pass specific to the
+/// PPC target machine. It uses the target's detailed information to provide
+/// more precise answers to certain TTI queries, while letting the target
+/// independent and default TTI implementations handle the rest.
+///
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "ppctti"
+#include "PPC.h"
+#include "PPCTargetMachine.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/CostTable.h"
+using namespace llvm;
+
+// Declare the pass initialization routine locally as target-specific passes
+// don't havve a target-wide initialization entry point, and so we rely on the
+// pass constructor initialization.
+namespace llvm {
+void initializePPCTTIPass(PassRegistry &);
+}
+
+namespace {
+
+class PPCTTI : public ImmutablePass, public TargetTransformInfo {
+  const PPCTargetMachine *TM;
+  const PPCSubtarget *ST;
+  const PPCTargetLowering *TLI;
+
+  /// Estimate the overhead of scalarizing an instruction. Insert and Extract
+  /// are set if the result needs to be inserted and/or extracted from vectors.
+  unsigned getScalarizationOverhead(Type *Ty, bool Insert, bool Extract) const;
+
+public:
+  PPCTTI() : ImmutablePass(ID), TM(0), ST(0), TLI(0) {
+    llvm_unreachable("This pass cannot be directly constructed");
+  }
+
+  PPCTTI(const PPCTargetMachine *TM)
+      : ImmutablePass(ID), TM(TM), ST(TM->getSubtargetImpl()),
+        TLI(TM->getTargetLowering()) {
+    initializePPCTTIPass(*PassRegistry::getPassRegistry());
+  }
+
+  virtual void initializePass() {
+    pushTTIStack(this);
+  }
+
+  virtual void finalizePass() {
+    popTTIStack();
+  }
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+    TargetTransformInfo::getAnalysisUsage(AU);
+  }
+
+  /// Pass identification.
+  static char ID;
+
+  /// Provide necessary pointer adjustments for the two base classes.
+  virtual void *getAdjustedAnalysisPointer(const void *ID) {
+    if (ID == &TargetTransformInfo::ID)
+      return (TargetTransformInfo*)this;
+    return this;
+  }
+
+  /// \name Scalar TTI Implementations
+  /// @{
+  virtual PopcntSupportKind getPopcntSupport(unsigned TyWidth) const;
+
+  /// @}
+
+  /// \name Vector TTI Implementations
+  /// @{
+
+  virtual unsigned getNumberOfRegisters(bool Vector) const;
+  virtual unsigned getRegisterBitWidth(bool Vector) const;
+  virtual unsigned getMaximumUnrollFactor() const;
+  virtual unsigned getArithmeticInstrCost(unsigned Opcode, Type *Ty,
+                                          OperandValueKind,
+                                          OperandValueKind) const;
+  virtual unsigned getShuffleCost(ShuffleKind Kind, Type *Tp,
+                                  int Index, Type *SubTp) const;
+  virtual unsigned getCastInstrCost(unsigned Opcode, Type *Dst,
+                                    Type *Src) const;
+  virtual unsigned getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
+                                      Type *CondTy) const;
+  virtual unsigned getVectorInstrCost(unsigned Opcode, Type *Val,
+                                      unsigned Index) const;
+  virtual unsigned getMemoryOpCost(unsigned Opcode, Type *Src,
+                                   unsigned Alignment,
+                                   unsigned AddressSpace) const;
+
+  /// @}
+};
+
+} // end anonymous namespace
+
+INITIALIZE_AG_PASS(PPCTTI, TargetTransformInfo, "ppctti",
+                   "PPC Target Transform Info", true, true, false)
+char PPCTTI::ID = 0;
+
+ImmutablePass *
+llvm::createPPCTargetTransformInfoPass(const PPCTargetMachine *TM) {
+  return new PPCTTI(TM);
+}
+
+
+//===----------------------------------------------------------------------===//
+//
+// PPC cost model.
+//
+//===----------------------------------------------------------------------===//
+
+PPCTTI::PopcntSupportKind PPCTTI::getPopcntSupport(unsigned TyWidth) const {
+  assert(isPowerOf2_32(TyWidth) && "Ty width must be power of 2");
+  if (ST->hasPOPCNTD() && TyWidth <= 64)
+    return PSK_FastHardware;
+  return PSK_Software;
+}
+
+unsigned PPCTTI::getNumberOfRegisters(bool Vector) const {
+  if (Vector && !ST->hasAltivec())
+    return 0;
+  return 32;
+}
+
+unsigned PPCTTI::getRegisterBitWidth(bool Vector) const {
+  if (Vector) {
+    if (ST->hasAltivec()) return 128;
+    return 0;
+  }
+
+  if (ST->isPPC64())
+    return 64;
+  return 32;
+
+}
+
+unsigned PPCTTI::getMaximumUnrollFactor() const {
+  unsigned Directive = ST->getDarwinDirective();
+  // The 440 has no SIMD support, but floating-point instructions
+  // have a 5-cycle latency, so unroll by 5x for latency hiding.
+  if (Directive == PPC::DIR_440)
+    return 5;
+
+  // The A2 has no SIMD support, but floating-point instructions
+  // have a 6-cycle latency, so unroll by 6x for latency hiding.
+  if (Directive == PPC::DIR_A2)
+    return 6;
+
+  // FIXME: For lack of any better information, do no harm...
+  if (Directive == PPC::DIR_E500mc || Directive == PPC::DIR_E5500)
+    return 1;
+
+  // For most things, modern systems have two execution units (and
+  // out-of-order execution).
+  return 2;
+}
+
+unsigned PPCTTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty,
+                                        OperandValueKind Op1Info,
+                                        OperandValueKind Op2Info) const {
+  assert(TLI->InstructionOpcodeToISD(Opcode) && "Invalid opcode");
+
+  // Fallback to the default implementation.
+  return TargetTransformInfo::getArithmeticInstrCost(Opcode, Ty, Op1Info,
+                                                     Op2Info);
+}
+
+unsigned PPCTTI::getShuffleCost(ShuffleKind Kind, Type *Tp, int Index,
+                                Type *SubTp) const {
+  return TargetTransformInfo::getShuffleCost(Kind, Tp, Index, SubTp);
+}
+
+unsigned PPCTTI::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) const {
+  assert(TLI->InstructionOpcodeToISD(Opcode) && "Invalid opcode");
+
+  return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
+}
+
+unsigned PPCTTI::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
+                                    Type *CondTy) const {
+  return TargetTransformInfo::getCmpSelInstrCost(Opcode, ValTy, CondTy);
+}
+
+unsigned PPCTTI::getVectorInstrCost(unsigned Opcode, Type *Val,
+                                    unsigned Index) const {
+  assert(Val->isVectorTy() && "This must be a vector type");
+
+  int ISD = TLI->InstructionOpcodeToISD(Opcode);
+  assert(ISD && "Invalid opcode");
+
+  // Estimated cost of a load-hit-store delay.  This was obtained
+  // experimentally as a minimum needed to prevent unprofitable
+  // vectorization for the paq8p benchmark.  It may need to be
+  // raised further if other unprofitable cases remain.
+  unsigned LHSPenalty = 12;
+
+  // Vector element insert/extract with Altivec is very expensive,
+  // because they require store and reload with the attendant
+  // processor stall for load-hit-store.  Until VSX is available,
+  // these need to be estimated as very costly.
+  if (ISD == ISD::EXTRACT_VECTOR_ELT ||
+      ISD == ISD::INSERT_VECTOR_ELT)
+    return LHSPenalty +
+      TargetTransformInfo::getVectorInstrCost(Opcode, Val, Index);
+
+  return TargetTransformInfo::getVectorInstrCost(Opcode, Val, Index);
+}
+
+unsigned PPCTTI::getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
+                                 unsigned AddressSpace) const {
+  // Legalize the type.
+  std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Src);
+  assert((Opcode == Instruction::Load || Opcode == Instruction::Store) &&
+         "Invalid Opcode");
+
+  // Each load/store unit costs 1.
+  unsigned Cost = LT.first * 1;
+
+  // PPC in general does not support unaligned loads and stores. They'll need
+  // to be decomposed based on the alignment factor.
+  unsigned SrcBytes = LT.second.getStoreSize();
+  if (SrcBytes && Alignment && Alignment < SrcBytes)
+    Cost *= (SrcBytes/Alignment);
+
+  return Cost;
+}
+
diff --git a/lib/Target/PowerPC/README.txt b/lib/Target/PowerPC/README.txt
index b6763aa73802..cc2ff966332e 100644
--- a/lib/Target/PowerPC/README.txt
+++ b/lib/Target/PowerPC/README.txt
@@ -1,7 +1,6 @@
 //===- README.txt - Notes for improving PowerPC-specific code gen ---------===//
 
 TODO:
-* gpr0 allocation
 * lmw/stmw pass a la arm load store optimizer for prolog/epilog
 
 ===-------------------------------------------------------------------------===
@@ -204,12 +203,6 @@ http://gcc.gnu.org/ml/gcc-patches/2006-02/msg00133.html
 
 ===-------------------------------------------------------------------------===
 
-Implement Newton-Rhapson method for improving estimate instructions to the
-correct accuracy, and implementing divide as multiply by reciprocal when it has
-more than one use.  Itanium would want this too.
-
-===-------------------------------------------------------------------------===
-
 Compile offsets from allocas:
 
 int *%test() {
@@ -536,20 +529,6 @@ void func(unsigned int *ret, float dx, float dy, float dz, float dw) {
 
 ===-------------------------------------------------------------------------===
 
-Complete the signed i32 to FP conversion code using 64-bit registers
-transformation, good for PI.  See PPCISelLowering.cpp, this comment:
-
-     // FIXME: disable this lowered code.  This generates 64-bit register values,
-     // and we don't model the fact that the top part is clobbered by calls.  We
-     // need to flag these together so that the value isn't live across a call.
-     //setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
-
-Also, if the registers are spilled to the stack, we have to ensure that all
-64-bits of them are save/restored, otherwise we will miscompile the code.  It
-sounds like we need to get the 64-bit register classes going.
-
-===-------------------------------------------------------------------------===
-
 %struct.B = type { i8, [3 x i8] }
 
 define void @bar(%struct.B* %b) {
diff --git a/lib/Target/PowerPC/TargetInfo/PowerPCTargetInfo.cpp b/lib/Target/PowerPC/TargetInfo/PowerPCTargetInfo.cpp
index 5dc8568d83f2..fa44331b8af6 100644
--- a/lib/Target/PowerPC/TargetInfo/PowerPCTargetInfo.cpp
+++ b/lib/Target/PowerPC/TargetInfo/PowerPCTargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "PPC.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
diff --git a/lib/Target/R600/AMDGPU.h b/lib/Target/R600/AMDGPU.h
new file mode 100644
index 000000000000..0b01433cc926
--- /dev/null
+++ b/lib/Target/R600/AMDGPU.h
@@ -0,0 +1,51 @@
+//===-- AMDGPU.h - MachineFunction passes hw codegen --------------*- C++ -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPU_H
+#define AMDGPU_H
+
+#include "AMDGPUTargetMachine.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Target/TargetMachine.h"
+
+namespace llvm {
+
+class FunctionPass;
+class AMDGPUTargetMachine;
+
+// R600 Passes
+FunctionPass* createR600KernelParametersPass(const DataLayout *TD);
+FunctionPass *createR600ExpandSpecialInstrsPass(TargetMachine &tm);
+FunctionPass *createR600EmitClauseMarkers(TargetMachine &tm);
+FunctionPass *createR600ControlFlowFinalizer(TargetMachine &tm);
+
+// SI Passes
+FunctionPass *createSIAnnotateControlFlowPass();
+FunctionPass *createSILowerControlFlowPass(TargetMachine &tm);
+FunctionPass *createSICodeEmitterPass(formatted_raw_ostream &OS);
+FunctionPass *createSIInsertWaits(TargetMachine &tm);
+
+// Passes common to R600 and SI
+Pass *createAMDGPUStructurizeCFGPass();
+FunctionPass *createAMDGPUConvertToISAPass(TargetMachine &tm);
+FunctionPass* createAMDGPUIndirectAddressingPass(TargetMachine &tm);
+
+} // End namespace llvm
+
+namespace ShaderType {
+  enum Type {
+    PIXEL = 0,
+    VERTEX = 1,
+    GEOMETRY = 2,
+    COMPUTE = 3
+  };
+}
+
+#endif // AMDGPU_H
diff --git a/lib/Target/R600/AMDGPU.td b/lib/Target/R600/AMDGPU.td
new file mode 100644
index 000000000000..1a26c77d6bb2
--- /dev/null
+++ b/lib/Target/R600/AMDGPU.td
@@ -0,0 +1,41 @@
+//===-- AMDIL.td - AMDIL Tablegen files --*- tablegen -*-------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+
+// Include AMDIL TD files
+include "AMDILBase.td"
+
+
+def AMDGPUInstrInfo : InstrInfo {
+  let guessInstructionProperties = 1;
+}
+
+//===----------------------------------------------------------------------===//
+// Declare the target which we are implementing
+//===----------------------------------------------------------------------===//
+def AMDGPUAsmWriter : AsmWriter {
+    string AsmWriterClassName = "InstPrinter";
+    int Variant = 0;
+    bit isMCAsmWriter = 1;
+}
+
+def AMDGPU : Target {
+  // Pull in Instruction Info:
+  let InstructionSet = AMDGPUInstrInfo;
+  let AssemblyWriters = [AMDGPUAsmWriter];
+}
+
+// Include AMDGPU TD files
+include "R600Schedule.td"
+include "SISchedule.td"
+include "Processors.td"
+include "AMDGPUInstrInfo.td"
+include "AMDGPUIntrinsics.td"
+include "AMDGPURegisterInfo.td"
+include "AMDGPUInstructions.td"
+include "AMDGPUCallingConv.td"
diff --git a/lib/Target/R600/AMDGPUAsmPrinter.cpp b/lib/Target/R600/AMDGPUAsmPrinter.cpp
new file mode 100644
index 000000000000..f6001445f4b3
--- /dev/null
+++ b/lib/Target/R600/AMDGPUAsmPrinter.cpp
@@ -0,0 +1,145 @@
+//===-- AMDGPUAsmPrinter.cpp - AMDGPU Assebly printer  --------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+///
+/// The AMDGPUAsmPrinter is used to print both assembly string and also binary
+/// code.  When passed an MCAsmStreamer it prints assembly and when passed
+/// an MCObjectStreamer it outputs binary code.
+//
+//===----------------------------------------------------------------------===//
+//
+
+
+#include "AMDGPUAsmPrinter.h"
+#include "AMDGPU.h"
+#include "SIMachineFunctionInfo.h"
+#include "SIRegisterInfo.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Target/TargetLoweringObjectFile.h"
+
+using namespace llvm;
+
+
+static AsmPrinter *createAMDGPUAsmPrinterPass(TargetMachine &tm,
+                                              MCStreamer &Streamer) {
+  return new AMDGPUAsmPrinter(tm, Streamer);
+}
+
+extern "C" void LLVMInitializeR600AsmPrinter() {
+  TargetRegistry::RegisterAsmPrinter(TheAMDGPUTarget, createAMDGPUAsmPrinterPass);
+}
+
+/// We need to override this function so we can avoid
+/// the call to EmitFunctionHeader(), which the MCPureStreamer can't handle.
+bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
+  const AMDGPUSubtarget &STM = TM.getSubtarget<AMDGPUSubtarget>();
+  if (STM.dumpCode()) {
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
+    MF.dump();
+#endif
+  }
+  SetupMachineFunction(MF);
+  if (OutStreamer.hasRawTextSupport()) {
+    OutStreamer.EmitRawText("@" + MF.getName() + ":");
+  }
+  OutStreamer.SwitchSection(getObjFileLowering().getTextSection());
+  if (STM.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
+    EmitProgramInfo(MF);
+  }
+  EmitFunctionBody();
+  return false;
+}
+
+void AMDGPUAsmPrinter::EmitProgramInfo(MachineFunction &MF) {
+  unsigned MaxSGPR = 0;
+  unsigned MaxVGPR = 0;
+  bool VCCUsed = false;
+  const SIRegisterInfo * RI =
+                static_cast<const SIRegisterInfo*>(TM.getRegisterInfo());
+
+  for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
+                                                  BB != BB_E; ++BB) {
+    MachineBasicBlock &MBB = *BB;
+    for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
+                                                    I != E; ++I) {
+      MachineInstr &MI = *I;
+
+      unsigned numOperands = MI.getNumOperands();
+      for (unsigned op_idx = 0; op_idx < numOperands; op_idx++) {
+        MachineOperand & MO = MI.getOperand(op_idx);
+        unsigned maxUsed;
+        unsigned width = 0;
+        bool isSGPR = false;
+        unsigned reg;
+        unsigned hwReg;
+        if (!MO.isReg()) {
+          continue;
+        }
+        reg = MO.getReg();
+        if (reg == AMDGPU::VCC) {
+          VCCUsed = true;
+          continue;
+        }
+        switch (reg) {
+        default: break;
+        case AMDGPU::EXEC:
+        case AMDGPU::M0:
+          continue;
+        }
+
+        if (AMDGPU::SReg_32RegClass.contains(reg)) {
+          isSGPR = true;
+          width = 1;
+        } else if (AMDGPU::VReg_32RegClass.contains(reg)) {
+          isSGPR = false;
+          width = 1;
+        } else if (AMDGPU::SReg_64RegClass.contains(reg)) {
+          isSGPR = true;
+          width = 2;
+        } else if (AMDGPU::VReg_64RegClass.contains(reg)) {
+          isSGPR = false;
+          width = 2;
+        } else if (AMDGPU::SReg_128RegClass.contains(reg)) {
+          isSGPR = true;
+          width = 4;
+        } else if (AMDGPU::VReg_128RegClass.contains(reg)) {
+          isSGPR = false;
+          width = 4;
+        } else if (AMDGPU::SReg_256RegClass.contains(reg)) {
+          isSGPR = true;
+          width = 8;
+        } else if (AMDGPU::VReg_256RegClass.contains(reg)) {
+          isSGPR = false;
+          width = 8;
+        } else if (AMDGPU::VReg_512RegClass.contains(reg)) {
+          isSGPR = false;
+          width = 16;
+        } else {
+          assert(!"Unknown register class");
+        }
+        hwReg = RI->getEncodingValue(reg) & 0xff;
+        maxUsed = hwReg + width - 1;
+        if (isSGPR) {
+          MaxSGPR = maxUsed > MaxSGPR ? maxUsed : MaxSGPR;
+        } else {
+          MaxVGPR = maxUsed > MaxVGPR ? maxUsed : MaxVGPR;
+        }
+      }
+    }
+  }
+  if (VCCUsed) {
+    MaxSGPR += 2;
+  }
+  SIMachineFunctionInfo * MFI = MF.getInfo<SIMachineFunctionInfo>();
+  OutStreamer.EmitIntValue(MaxSGPR + 1, 4);
+  OutStreamer.EmitIntValue(MaxVGPR + 1, 4);
+  OutStreamer.EmitIntValue(MFI->PSInputAddr, 4);
+}
diff --git a/lib/Target/R600/AMDGPUAsmPrinter.h b/lib/Target/R600/AMDGPUAsmPrinter.h
new file mode 100644
index 000000000000..3812282b1798
--- /dev/null
+++ b/lib/Target/R600/AMDGPUAsmPrinter.h
@@ -0,0 +1,44 @@
+//===-- AMDGPUAsmPrinter.h - Print AMDGPU assembly code -------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief AMDGPU Assembly printer class.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPU_ASMPRINTER_H
+#define AMDGPU_ASMPRINTER_H
+
+#include "llvm/CodeGen/AsmPrinter.h"
+
+namespace llvm {
+
+class AMDGPUAsmPrinter : public AsmPrinter {
+
+public:
+  explicit AMDGPUAsmPrinter(TargetMachine &TM, MCStreamer &Streamer)
+    : AsmPrinter(TM, Streamer) { }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF);
+
+  virtual const char *getPassName() const {
+    return "AMDGPU Assembly Printer";
+  }
+
+  /// \brief Emit register usage information so that the GPU driver
+  /// can correctly setup the GPU state.
+  void EmitProgramInfo(MachineFunction &MF);
+
+  /// Implemented in AMDGPUMCInstLower.cpp
+  virtual void EmitInstruction(const MachineInstr *MI);
+};
+
+} // End anonymous llvm
+
+#endif //AMDGPU_ASMPRINTER_H
diff --git a/lib/Target/R600/AMDGPUCallingConv.td b/lib/Target/R600/AMDGPUCallingConv.td
new file mode 100644
index 000000000000..45ae37ef0c7f
--- /dev/null
+++ b/lib/Target/R600/AMDGPUCallingConv.td
@@ -0,0 +1,42 @@
+//===---- AMDCallingConv.td - Calling Conventions for Radeon GPUs ---------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This describes the calling conventions for the AMD Radeon GPUs.
+//
+//===----------------------------------------------------------------------===//
+
+// Inversion of CCIfInReg
+class CCIfNotInReg<CCAction A> : CCIf<"!ArgFlags.isInReg()", A> {}
+
+// Calling convention for SI
+def CC_SI : CallingConv<[
+
+  CCIfInReg<CCIfType<[f32, i32] , CCAssignToReg<[
+    SGPR0, SGPR1, SGPR2, SGPR3, SGPR4, SGPR5, SGPR6, SGPR7,
+    SGPR8, SGPR9, SGPR10, SGPR11, SGPR12, SGPR13, SGPR14, SGPR15
+  ]>>>,
+
+  CCIfInReg<CCIfType<[i64] , CCAssignToRegWithShadow<
+    [ SGPR0, SGPR2, SGPR4, SGPR6, SGPR8, SGPR10, SGPR12, SGPR14 ],
+    [ SGPR1, SGPR3, SGPR5, SGPR7, SGPR9, SGPR11, SGPR12, SGPR15 ]
+  >>>,
+
+  CCIfNotInReg<CCIfType<[f32, i32] , CCAssignToReg<[
+    VGPR0, VGPR1, VGPR2, VGPR3, VGPR4, VGPR5, VGPR6, VGPR7,
+    VGPR8, VGPR9, VGPR10, VGPR11, VGPR12, VGPR13, VGPR14, VGPR15,
+    VGPR16, VGPR17, VGPR18, VGPR19, VGPR20, VGPR21, VGPR22, VGPR23,
+    VGPR24, VGPR25, VGPR26, VGPR27, VGPR28, VGPR29, VGPR30, VGPR31
+  ]>>>
+
+]>;
+
+def CC_AMDGPU : CallingConv<[
+  CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>().device()"#
+       "->getGeneration() == AMDGPUDeviceInfo::HD7XXX", CCDelegateTo<CC_SI>>
+]>;
diff --git a/lib/Target/R600/AMDGPUConvertToISA.cpp b/lib/Target/R600/AMDGPUConvertToISA.cpp
new file mode 100644
index 000000000000..50297d1f60c8
--- /dev/null
+++ b/lib/Target/R600/AMDGPUConvertToISA.cpp
@@ -0,0 +1,62 @@
+//===-- AMDGPUConvertToISA.cpp - Lower AMDIL to HW ISA --------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief This pass lowers AMDIL machine instructions to the appropriate
+/// hardware instructions.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "AMDGPUInstrInfo.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+
+using namespace llvm;
+
+namespace {
+
+class AMDGPUConvertToISAPass : public MachineFunctionPass {
+
+private:
+  static char ID;
+  TargetMachine &TM;
+
+public:
+  AMDGPUConvertToISAPass(TargetMachine &tm) :
+    MachineFunctionPass(ID), TM(tm) { }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF);
+
+  virtual const char *getPassName() const {return "AMDGPU Convert to ISA";}
+
+};
+
+} // End anonymous namespace
+
+char AMDGPUConvertToISAPass::ID = 0;
+
+FunctionPass *llvm::createAMDGPUConvertToISAPass(TargetMachine &tm) {
+  return new AMDGPUConvertToISAPass(tm);
+}
+
+bool AMDGPUConvertToISAPass::runOnMachineFunction(MachineFunction &MF) {
+  const AMDGPUInstrInfo * TII =
+                      static_cast<const AMDGPUInstrInfo*>(TM.getInstrInfo());
+
+  for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
+                                                  BB != BB_E; ++BB) {
+    MachineBasicBlock &MBB = *BB;
+    for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
+                                                      I != E; ++I) {
+      MachineInstr &MI = *I;
+      TII->convertToISA(MI, MF, MBB.findDebugLoc(I));
+    }
+  }
+  return false;
+}
diff --git a/lib/Target/R600/AMDGPUFrameLowering.cpp b/lib/Target/R600/AMDGPUFrameLowering.cpp
new file mode 100644
index 000000000000..815d6f71c3be
--- /dev/null
+++ b/lib/Target/R600/AMDGPUFrameLowering.cpp
@@ -0,0 +1,122 @@
+//===----------------------- AMDGPUFrameLowering.cpp ----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+// Interface to describe a layout of a stack frame on a AMDIL target machine
+//
+//===----------------------------------------------------------------------===//
+#include "AMDGPUFrameLowering.h"
+#include "AMDGPURegisterInfo.h"
+#include "R600MachineFunctionInfo.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/Instructions.h"
+
+using namespace llvm;
+AMDGPUFrameLowering::AMDGPUFrameLowering(StackDirection D, unsigned StackAl,
+    int LAO, unsigned TransAl)
+  : TargetFrameLowering(D, StackAl, LAO, TransAl) { }
+
+AMDGPUFrameLowering::~AMDGPUFrameLowering() { }
+
+unsigned AMDGPUFrameLowering::getStackWidth(const MachineFunction &MF) const {
+
+  // XXX: Hardcoding to 1 for now.
+  //
+  // I think the StackWidth should stored as metadata associated with the
+  // MachineFunction.  This metadata can either be added by a frontend, or
+  // calculated by a R600 specific LLVM IR pass.
+  //
+  // The StackWidth determines how stack objects are laid out in memory.
+  // For a vector stack variable, like: int4 stack[2], the data will be stored
+  // in the following ways depending on the StackWidth.
+  //
+  // StackWidth = 1:
+  //
+  // T0.X = stack[0].x
+  // T1.X = stack[0].y
+  // T2.X = stack[0].z
+  // T3.X = stack[0].w
+  // T4.X = stack[1].x
+  // T5.X = stack[1].y
+  // T6.X = stack[1].z
+  // T7.X = stack[1].w
+  //
+  // StackWidth = 2:
+  //
+  // T0.X = stack[0].x
+  // T0.Y = stack[0].y
+  // T1.X = stack[0].z
+  // T1.Y = stack[0].w
+  // T2.X = stack[1].x
+  // T2.Y = stack[1].y
+  // T3.X = stack[1].z
+  // T3.Y = stack[1].w
+  // 
+  // StackWidth = 4:
+  // T0.X = stack[0].x
+  // T0.Y = stack[0].y
+  // T0.Z = stack[0].z
+  // T0.W = stack[0].w
+  // T1.X = stack[1].x
+  // T1.Y = stack[1].y
+  // T1.Z = stack[1].z
+  // T1.W = stack[1].w
+  return 1;
+}
+
+/// \returns The number of registers allocated for \p FI.
+int AMDGPUFrameLowering::getFrameIndexOffset(const MachineFunction &MF,
+                                         int FI) const {
+  const MachineFrameInfo *MFI = MF.getFrameInfo();
+  unsigned Offset = 0;
+  int UpperBound = FI == -1 ? MFI->getNumObjects() : FI;
+
+  for (int i = MFI->getObjectIndexBegin(); i < UpperBound; ++i) {
+    const AllocaInst *Alloca = MFI->getObjectAllocation(i);
+    unsigned ArrayElements;
+    const Type *AllocaType = Alloca->getAllocatedType();
+    const Type *ElementType;
+
+    if (AllocaType->isArrayTy()) {
+      ArrayElements = AllocaType->getArrayNumElements();
+      ElementType = AllocaType->getArrayElementType();
+    } else {
+      ArrayElements = 1;
+      ElementType = AllocaType;
+    }
+
+    unsigned VectorElements;
+    if (ElementType->isVectorTy()) {
+      VectorElements = ElementType->getVectorNumElements();
+    } else {
+      VectorElements = 1;
+    }
+
+    Offset += (VectorElements / getStackWidth(MF)) * ArrayElements;
+  }
+  return Offset;
+}
+
+const TargetFrameLowering::SpillSlot *
+AMDGPUFrameLowering::getCalleeSavedSpillSlots(unsigned &NumEntries) const {
+  NumEntries = 0;
+  return 0;
+}
+void
+AMDGPUFrameLowering::emitPrologue(MachineFunction &MF) const {
+}
+void
+AMDGPUFrameLowering::emitEpilogue(MachineFunction &MF,
+                                  MachineBasicBlock &MBB) const {
+}
+
+bool
+AMDGPUFrameLowering::hasFP(const MachineFunction &MF) const {
+  return false;
+}
diff --git a/lib/Target/R600/AMDGPUFrameLowering.h b/lib/Target/R600/AMDGPUFrameLowering.h
new file mode 100644
index 000000000000..cf5742ee0952
--- /dev/null
+++ b/lib/Target/R600/AMDGPUFrameLowering.h
@@ -0,0 +1,44 @@
+//===--------------------- AMDGPUFrameLowering.h ----------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Interface to describe a layout of a stack frame on a AMDIL target
+/// machine.
+//
+//===----------------------------------------------------------------------===//
+#ifndef AMDILFRAME_LOWERING_H
+#define AMDILFRAME_LOWERING_H
+
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/Target/TargetFrameLowering.h"
+
+namespace llvm {
+
+/// \brief Information about the stack frame layout on the AMDGPU targets.
+///
+/// It holds the direction of the stack growth, the known stack alignment on
+/// entry to each function, and the offset to the locals area.
+/// See TargetFrameInfo for more comments.
+class AMDGPUFrameLowering : public TargetFrameLowering {
+public:
+  AMDGPUFrameLowering(StackDirection D, unsigned StackAl, int LAO,
+                      unsigned TransAl = 1);
+  virtual ~AMDGPUFrameLowering();
+
+  /// \returns The number of 32-bit sub-registers that are used when storing
+  /// values to the stack.
+  virtual unsigned getStackWidth(const MachineFunction &MF) const;
+  virtual int getFrameIndexOffset(const MachineFunction &MF, int FI) const;
+  virtual const SpillSlot *getCalleeSavedSpillSlots(unsigned &NumEntries) const;
+  virtual void emitPrologue(MachineFunction &MF) const;
+  virtual void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
+  virtual bool hasFP(const MachineFunction &MF) const;
+};
+} // namespace llvm
+#endif // AMDILFRAME_LOWERING_H
diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
new file mode 100644
index 000000000000..a266df535d56
--- /dev/null
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -0,0 +1,414 @@
+//===-- AMDGPUISelLowering.cpp - AMDGPU Common DAG lowering functions -----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief This is the parent TargetLowering class for hardware code gen
+/// targets.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUISelLowering.h"
+#include "AMDGPURegisterInfo.h"
+#include "AMDILIntrinsicInfo.h"
+#include "AMDGPUSubtarget.h"
+#include "llvm/CodeGen/CallingConvLower.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+
+using namespace llvm;
+
+#include "AMDGPUGenCallingConv.inc"
+
+AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
+  TargetLowering(TM, new TargetLoweringObjectFileELF()) {
+
+  // Initialize target lowering borrowed from AMDIL
+  InitAMDILLowering();
+
+  // We need to custom lower some of the intrinsics
+  setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
+
+  // Library functions.  These default to Expand, but we have instructions
+  // for them.
+  setOperationAction(ISD::FCEIL,  MVT::f32, Legal);
+  setOperationAction(ISD::FEXP2,  MVT::f32, Legal);
+  setOperationAction(ISD::FPOW,   MVT::f32, Legal);
+  setOperationAction(ISD::FLOG2,  MVT::f32, Legal);
+  setOperationAction(ISD::FABS,   MVT::f32, Legal);
+  setOperationAction(ISD::FFLOOR, MVT::f32, Legal);
+  setOperationAction(ISD::FRINT,  MVT::f32, Legal);
+
+  // Lower floating point store/load to integer store/load to reduce the number
+  // of patterns in tablegen.
+  setOperationAction(ISD::STORE, MVT::f32, Promote);
+  AddPromotedToType(ISD::STORE, MVT::f32, MVT::i32);
+
+  setOperationAction(ISD::STORE, MVT::v4f32, Promote);
+  AddPromotedToType(ISD::STORE, MVT::v4f32, MVT::v4i32);
+
+  setOperationAction(ISD::LOAD, MVT::f32, Promote);
+  AddPromotedToType(ISD::LOAD, MVT::f32, MVT::i32);
+
+  setOperationAction(ISD::LOAD, MVT::v4f32, Promote);
+  AddPromotedToType(ISD::LOAD, MVT::v4f32, MVT::v4i32);
+
+  setOperationAction(ISD::MUL, MVT::i64, Expand);
+
+  setOperationAction(ISD::UDIV, MVT::i32, Expand);
+  setOperationAction(ISD::UDIVREM, MVT::i32, Custom);
+  setOperationAction(ISD::UREM, MVT::i32, Expand);
+}
+
+//===---------------------------------------------------------------------===//
+// TargetLowering Callbacks
+//===---------------------------------------------------------------------===//
+
+void AMDGPUTargetLowering::AnalyzeFormalArguments(CCState &State,
+                             const SmallVectorImpl<ISD::InputArg> &Ins) const {
+
+  State.AnalyzeFormalArguments(Ins, CC_AMDGPU);
+}
+
+SDValue AMDGPUTargetLowering::LowerReturn(
+                                     SDValue Chain,
+                                     CallingConv::ID CallConv,
+                                     bool isVarArg,
+                                     const SmallVectorImpl<ISD::OutputArg> &Outs,
+                                     const SmallVectorImpl<SDValue> &OutVals,
+                                     DebugLoc DL, SelectionDAG &DAG) const {
+  return DAG.getNode(AMDGPUISD::RET_FLAG, DL, MVT::Other, Chain);
+}
+
+//===---------------------------------------------------------------------===//
+// Target specific lowering
+//===---------------------------------------------------------------------===//
+
+SDValue AMDGPUTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG)
+    const {
+  switch (Op.getOpcode()) {
+  default:
+    Op.getNode()->dump();
+    assert(0 && "Custom lowering code for this"
+        "instruction is not implemented yet!");
+    break;
+  // AMDIL DAG lowering
+  case ISD::SDIV: return LowerSDIV(Op, DAG);
+  case ISD::SREM: return LowerSREM(Op, DAG);
+  case ISD::SIGN_EXTEND_INREG: return LowerSIGN_EXTEND_INREG(Op, DAG);
+  case ISD::BRCOND: return LowerBRCOND(Op, DAG);
+  // AMDGPU DAG lowering
+  case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
+  case ISD::UDIVREM: return LowerUDIVREM(Op, DAG);
+  }
+  return Op;
+}
+
+SDValue AMDGPUTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
+    SelectionDAG &DAG) const {
+  unsigned IntrinsicID = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
+  DebugLoc DL = Op.getDebugLoc();
+  EVT VT = Op.getValueType();
+
+  switch (IntrinsicID) {
+    default: return Op;
+    case AMDGPUIntrinsic::AMDIL_abs:
+      return LowerIntrinsicIABS(Op, DAG);
+    case AMDGPUIntrinsic::AMDIL_exp:
+      return DAG.getNode(ISD::FEXP2, DL, VT, Op.getOperand(1));
+    case AMDGPUIntrinsic::AMDGPU_lrp:
+      return LowerIntrinsicLRP(Op, DAG);
+    case AMDGPUIntrinsic::AMDIL_fraction:
+      return DAG.getNode(AMDGPUISD::FRACT, DL, VT, Op.getOperand(1));
+    case AMDGPUIntrinsic::AMDIL_max:
+      return DAG.getNode(AMDGPUISD::FMAX, DL, VT, Op.getOperand(1),
+                                                  Op.getOperand(2));
+    case AMDGPUIntrinsic::AMDGPU_imax:
+      return DAG.getNode(AMDGPUISD::SMAX, DL, VT, Op.getOperand(1),
+                                                  Op.getOperand(2));
+    case AMDGPUIntrinsic::AMDGPU_umax:
+      return DAG.getNode(AMDGPUISD::UMAX, DL, VT, Op.getOperand(1),
+                                                  Op.getOperand(2));
+    case AMDGPUIntrinsic::AMDIL_min:
+      return DAG.getNode(AMDGPUISD::FMIN, DL, VT, Op.getOperand(1),
+                                                  Op.getOperand(2));
+    case AMDGPUIntrinsic::AMDGPU_imin:
+      return DAG.getNode(AMDGPUISD::SMIN, DL, VT, Op.getOperand(1),
+                                                  Op.getOperand(2));
+    case AMDGPUIntrinsic::AMDGPU_umin:
+      return DAG.getNode(AMDGPUISD::UMIN, DL, VT, Op.getOperand(1),
+                                                  Op.getOperand(2));
+    case AMDGPUIntrinsic::AMDIL_round_nearest:
+      return DAG.getNode(ISD::FRINT, DL, VT, Op.getOperand(1));
+  }
+}
+
+///IABS(a) = SMAX(sub(0, a), a)
+SDValue AMDGPUTargetLowering::LowerIntrinsicIABS(SDValue Op,
+    SelectionDAG &DAG) const {
+
+  DebugLoc DL = Op.getDebugLoc();
+  EVT VT = Op.getValueType();
+  SDValue Neg = DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, VT),
+                                              Op.getOperand(1));
+
+  return DAG.getNode(AMDGPUISD::SMAX, DL, VT, Neg, Op.getOperand(1));
+}
+
+/// Linear Interpolation
+/// LRP(a, b, c) = muladd(a,  b, (1 - a) * c)
+SDValue AMDGPUTargetLowering::LowerIntrinsicLRP(SDValue Op,
+    SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT VT = Op.getValueType();
+  SDValue OneSubA = DAG.getNode(ISD::FSUB, DL, VT,
+                                DAG.getConstantFP(1.0f, MVT::f32),
+                                Op.getOperand(1));
+  SDValue OneSubAC = DAG.getNode(ISD::FMUL, DL, VT, OneSubA,
+                                                    Op.getOperand(3));
+  return DAG.getNode(ISD::FADD, DL, VT,
+      DAG.getNode(ISD::FMUL, DL, VT, Op.getOperand(1), Op.getOperand(2)),
+      OneSubAC);
+}
+
+/// \brief Generate Min/Max node
+SDValue AMDGPUTargetLowering::LowerMinMax(SDValue Op,
+    SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT VT = Op.getValueType();
+
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  SDValue True = Op.getOperand(2);
+  SDValue False = Op.getOperand(3);
+  SDValue CC = Op.getOperand(4);
+
+  if (VT != MVT::f32 ||
+      !((LHS == True && RHS == False) || (LHS == False && RHS == True))) {
+    return SDValue();
+  }
+
+  ISD::CondCode CCOpcode = cast<CondCodeSDNode>(CC)->get();
+  switch (CCOpcode) {
+  case ISD::SETOEQ:
+  case ISD::SETONE:
+  case ISD::SETUNE:
+  case ISD::SETNE:
+  case ISD::SETUEQ:
+  case ISD::SETEQ:
+  case ISD::SETFALSE:
+  case ISD::SETFALSE2:
+  case ISD::SETTRUE:
+  case ISD::SETTRUE2:
+  case ISD::SETUO:
+  case ISD::SETO:
+    assert(0 && "Operation should already be optimised !");
+  case ISD::SETULE:
+  case ISD::SETULT:
+  case ISD::SETOLE:
+  case ISD::SETOLT:
+  case ISD::SETLE:
+  case ISD::SETLT: {
+    if (LHS == True)
+      return DAG.getNode(AMDGPUISD::FMIN, DL, VT, LHS, RHS);
+    else
+      return DAG.getNode(AMDGPUISD::FMAX, DL, VT, LHS, RHS);
+  }
+  case ISD::SETGT:
+  case ISD::SETGE:
+  case ISD::SETUGE:
+  case ISD::SETOGE:
+  case ISD::SETUGT:
+  case ISD::SETOGT: {
+    if (LHS == True)
+      return DAG.getNode(AMDGPUISD::FMAX, DL, VT, LHS, RHS);
+    else
+      return DAG.getNode(AMDGPUISD::FMIN, DL, VT, LHS, RHS);
+  }
+  case ISD::SETCC_INVALID:
+    assert(0 && "Invalid setcc condcode !");
+  }
+  return Op;
+}
+
+
+
+SDValue AMDGPUTargetLowering::LowerUDIVREM(SDValue Op,
+    SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT VT = Op.getValueType();
+
+  SDValue Num = Op.getOperand(0);
+  SDValue Den = Op.getOperand(1);
+
+  SmallVector<SDValue, 8> Results;
+
+  // RCP =  URECIP(Den) = 2^32 / Den + e
+  // e is rounding error.
+  SDValue RCP = DAG.getNode(AMDGPUISD::URECIP, DL, VT, Den);
+
+  // RCP_LO = umulo(RCP, Den) */
+  SDValue RCP_LO = DAG.getNode(ISD::UMULO, DL, VT, RCP, Den);
+
+  // RCP_HI = mulhu (RCP, Den) */
+  SDValue RCP_HI = DAG.getNode(ISD::MULHU, DL, VT, RCP, Den);
+
+  // NEG_RCP_LO = -RCP_LO
+  SDValue NEG_RCP_LO = DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, VT),
+                                                     RCP_LO);
+
+  // ABS_RCP_LO = (RCP_HI == 0 ? NEG_RCP_LO : RCP_LO)
+  SDValue ABS_RCP_LO = DAG.getSelectCC(DL, RCP_HI, DAG.getConstant(0, VT),
+                                           NEG_RCP_LO, RCP_LO,
+                                           ISD::SETEQ);
+  // Calculate the rounding error from the URECIP instruction
+  // E = mulhu(ABS_RCP_LO, RCP)
+  SDValue E = DAG.getNode(ISD::MULHU, DL, VT, ABS_RCP_LO, RCP);
+
+  // RCP_A_E = RCP + E
+  SDValue RCP_A_E = DAG.getNode(ISD::ADD, DL, VT, RCP, E);
+
+  // RCP_S_E = RCP - E
+  SDValue RCP_S_E = DAG.getNode(ISD::SUB, DL, VT, RCP, E);
+
+  // Tmp0 = (RCP_HI == 0 ? RCP_A_E : RCP_SUB_E)
+  SDValue Tmp0 = DAG.getSelectCC(DL, RCP_HI, DAG.getConstant(0, VT),
+                                     RCP_A_E, RCP_S_E,
+                                     ISD::SETEQ);
+  // Quotient = mulhu(Tmp0, Num)
+  SDValue Quotient = DAG.getNode(ISD::MULHU, DL, VT, Tmp0, Num);
+
+  // Num_S_Remainder = Quotient * Den
+  SDValue Num_S_Remainder = DAG.getNode(ISD::UMULO, DL, VT, Quotient, Den);
+
+  // Remainder = Num - Num_S_Remainder
+  SDValue Remainder = DAG.getNode(ISD::SUB, DL, VT, Num, Num_S_Remainder);
+
+  // Remainder_GE_Den = (Remainder >= Den ? -1 : 0)
+  SDValue Remainder_GE_Den = DAG.getSelectCC(DL, Remainder, Den,
+                                                 DAG.getConstant(-1, VT),
+                                                 DAG.getConstant(0, VT),
+                                                 ISD::SETGE);
+  // Remainder_GE_Zero = (Remainder >= 0 ? -1 : 0)
+  SDValue Remainder_GE_Zero = DAG.getSelectCC(DL, Remainder,
+                                                  DAG.getConstant(0, VT),
+                                                  DAG.getConstant(-1, VT),
+                                                  DAG.getConstant(0, VT),
+                                                  ISD::SETGE);
+  // Tmp1 = Remainder_GE_Den & Remainder_GE_Zero
+  SDValue Tmp1 = DAG.getNode(ISD::AND, DL, VT, Remainder_GE_Den,
+                                               Remainder_GE_Zero);
+
+  // Calculate Division result:
+
+  // Quotient_A_One = Quotient + 1
+  SDValue Quotient_A_One = DAG.getNode(ISD::ADD, DL, VT, Quotient,
+                                                         DAG.getConstant(1, VT));
+
+  // Quotient_S_One = Quotient - 1
+  SDValue Quotient_S_One = DAG.getNode(ISD::SUB, DL, VT, Quotient,
+                                                         DAG.getConstant(1, VT));
+
+  // Div = (Tmp1 == 0 ? Quotient : Quotient_A_One)
+  SDValue Div = DAG.getSelectCC(DL, Tmp1, DAG.getConstant(0, VT),
+                                     Quotient, Quotient_A_One, ISD::SETEQ);
+
+  // Div = (Remainder_GE_Zero == 0 ? Quotient_S_One : Div)
+  Div = DAG.getSelectCC(DL, Remainder_GE_Zero, DAG.getConstant(0, VT),
+                            Quotient_S_One, Div, ISD::SETEQ);
+
+  // Calculate Rem result:
+
+  // Remainder_S_Den = Remainder - Den
+  SDValue Remainder_S_Den = DAG.getNode(ISD::SUB, DL, VT, Remainder, Den);
+
+  // Remainder_A_Den = Remainder + Den
+  SDValue Remainder_A_Den = DAG.getNode(ISD::ADD, DL, VT, Remainder, Den);
+
+  // Rem = (Tmp1 == 0 ? Remainder : Remainder_S_Den)
+  SDValue Rem = DAG.getSelectCC(DL, Tmp1, DAG.getConstant(0, VT),
+                                    Remainder, Remainder_S_Den, ISD::SETEQ);
+
+  // Rem = (Remainder_GE_Zero == 0 ? Remainder_A_Den : Rem)
+  Rem = DAG.getSelectCC(DL, Remainder_GE_Zero, DAG.getConstant(0, VT),
+                            Remainder_A_Den, Rem, ISD::SETEQ);
+  SDValue Ops[2];
+  Ops[0] = Div;
+  Ops[1] = Rem;
+  return DAG.getMergeValues(Ops, 2, DL);
+}
+
+//===----------------------------------------------------------------------===//
+// Helper functions
+//===----------------------------------------------------------------------===//
+
+bool AMDGPUTargetLowering::isHWTrueValue(SDValue Op) const {
+  if (ConstantFPSDNode * CFP = dyn_cast<ConstantFPSDNode>(Op)) {
+    return CFP->isExactlyValue(1.0);
+  }
+  if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op)) {
+    return C->isAllOnesValue();
+  }
+  return false;
+}
+
+bool AMDGPUTargetLowering::isHWFalseValue(SDValue Op) const {
+  if (ConstantFPSDNode * CFP = dyn_cast<ConstantFPSDNode>(Op)) {
+    return CFP->getValueAPF().isZero();
+  }
+  if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op)) {
+    return C->isNullValue();
+  }
+  return false;
+}
+
+SDValue AMDGPUTargetLowering::CreateLiveInRegister(SelectionDAG &DAG,
+                                                  const TargetRegisterClass *RC,
+                                                   unsigned Reg, EVT VT) const {
+  MachineFunction &MF = DAG.getMachineFunction();
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+  unsigned VirtualRegister;
+  if (!MRI.isLiveIn(Reg)) {
+    VirtualRegister = MRI.createVirtualRegister(RC);
+    MRI.addLiveIn(Reg, VirtualRegister);
+  } else {
+    VirtualRegister = MRI.getLiveInVirtReg(Reg);
+  }
+  return DAG.getRegister(VirtualRegister, VT);
+}
+
+#define NODE_NAME_CASE(node) case AMDGPUISD::node: return #node;
+
+const char* AMDGPUTargetLowering::getTargetNodeName(unsigned Opcode) const {
+  switch (Opcode) {
+  default: return 0;
+  // AMDIL DAG nodes
+  NODE_NAME_CASE(CALL);
+  NODE_NAME_CASE(UMUL);
+  NODE_NAME_CASE(DIV_INF);
+  NODE_NAME_CASE(RET_FLAG);
+  NODE_NAME_CASE(BRANCH_COND);
+
+  // AMDGPU DAG nodes
+  NODE_NAME_CASE(DWORDADDR)
+  NODE_NAME_CASE(FRACT)
+  NODE_NAME_CASE(FMAX)
+  NODE_NAME_CASE(SMAX)
+  NODE_NAME_CASE(UMAX)
+  NODE_NAME_CASE(FMIN)
+  NODE_NAME_CASE(SMIN)
+  NODE_NAME_CASE(UMIN)
+  NODE_NAME_CASE(URECIP)
+  NODE_NAME_CASE(EXPORT)
+  NODE_NAME_CASE(CONST_ADDRESS)
+  NODE_NAME_CASE(REGISTER_LOAD)
+  NODE_NAME_CASE(REGISTER_STORE)
+  }
+}
diff --git a/lib/Target/R600/AMDGPUISelLowering.h b/lib/Target/R600/AMDGPUISelLowering.h
new file mode 100644
index 000000000000..f31b6466bd46
--- /dev/null
+++ b/lib/Target/R600/AMDGPUISelLowering.h
@@ -0,0 +1,140 @@
+//===-- AMDGPUISelLowering.h - AMDGPU Lowering Interface --------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Interface definition of the TargetLowering class that is common
+/// to all AMD GPUs.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPUISELLOWERING_H
+#define AMDGPUISELLOWERING_H
+
+#include "llvm/Target/TargetLowering.h"
+
+namespace llvm {
+
+class MachineRegisterInfo;
+
+class AMDGPUTargetLowering : public TargetLowering {
+private:
+  SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerUDIVREM(SDValue Op, SelectionDAG &DAG) const;
+
+protected:
+
+  /// \brief Helper function that adds Reg to the LiveIn list of the DAG's
+  /// MachineFunction.
+  ///
+  /// \returns a RegisterSDNode representing Reg.
+  SDValue CreateLiveInRegister(SelectionDAG &DAG, const TargetRegisterClass *RC,
+                                                  unsigned Reg, EVT VT) const;
+
+  bool isHWTrueValue(SDValue Op) const;
+  bool isHWFalseValue(SDValue Op) const;
+
+  void AnalyzeFormalArguments(CCState &State,
+                              const SmallVectorImpl<ISD::InputArg> &Ins) const;
+
+public:
+  AMDGPUTargetLowering(TargetMachine &TM);
+
+  virtual SDValue LowerReturn(SDValue Chain, CallingConv::ID CallConv,
+                              bool isVarArg,
+                              const SmallVectorImpl<ISD::OutputArg> &Outs,
+                              const SmallVectorImpl<SDValue> &OutVals,
+                              DebugLoc DL, SelectionDAG &DAG) const;
+  virtual SDValue LowerCall(CallLoweringInfo &CLI,
+                            SmallVectorImpl<SDValue> &InVals) const {
+    CLI.Callee.dump();
+    llvm_unreachable("Undefined function");
+  }
+
+  virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerIntrinsicIABS(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerIntrinsicLRP(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerMinMax(SDValue Op, SelectionDAG &DAG) const;
+  virtual const char* getTargetNodeName(unsigned Opcode) const;
+
+  virtual SDNode *PostISelFolding(MachineSDNode *N, SelectionDAG &DAG) const {
+    return N;
+  }
+
+// Functions defined in AMDILISelLowering.cpp
+public:
+
+  /// \brief Determine which of the bits specified in \p Mask are known to be
+  /// either zero or one and return them in the \p KnownZero and \p KnownOne
+  /// bitsets.
+  virtual void computeMaskedBitsForTargetNode(const SDValue Op,
+                                              APInt &KnownZero,
+                                              APInt &KnownOne,
+                                              const SelectionDAG &DAG,
+                                              unsigned Depth = 0) const;
+
+  virtual bool getTgtMemIntrinsic(IntrinsicInfo &Info,
+                                  const CallInst &I, unsigned Intrinsic) const;
+
+  /// We want to mark f32/f64 floating point values as legal.
+  bool isFPImmLegal(const APFloat &Imm, EVT VT) const;
+
+  /// We don't want to shrink f64/f32 constants.
+  bool ShouldShrinkFPConstant(EVT VT) const;
+
+private:
+  void InitAMDILLowering();
+  SDValue LowerSREM(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSREM8(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSREM16(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSREM32(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSREM64(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSDIV(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSDIV24(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSDIV32(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSDIV64(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const;
+  EVT genIntType(uint32_t size = 32, uint32_t numEle = 1) const;
+  SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const;
+};
+
+namespace AMDGPUISD {
+
+enum {
+  // AMDIL ISD Opcodes
+  FIRST_NUMBER = ISD::BUILTIN_OP_END,
+  CALL,        // Function call based on a single integer
+  UMUL,        // 32bit unsigned multiplication
+  DIV_INF,      // Divide with infinity returned on zero divisor
+  RET_FLAG,
+  BRANCH_COND,
+  // End AMDIL ISD Opcodes
+  BITALIGN,
+  DWORDADDR,
+  FRACT,
+  FMAX,
+  SMAX,
+  UMAX,
+  FMIN,
+  SMIN,
+  UMIN,
+  URECIP,
+  EXPORT,
+  CONST_ADDRESS,
+  REGISTER_LOAD,
+  REGISTER_STORE,
+  LAST_AMDGPU_ISD_NUMBER
+};
+
+
+} // End namespace AMDGPUISD
+
+} // End namespace llvm
+
+#endif // AMDGPUISELLOWERING_H
diff --git a/lib/Target/R600/AMDGPUIndirectAddressing.cpp b/lib/Target/R600/AMDGPUIndirectAddressing.cpp
new file mode 100644
index 000000000000..ed6c8ec55dd2
--- /dev/null
+++ b/lib/Target/R600/AMDGPUIndirectAddressing.cpp
@@ -0,0 +1,343 @@
+//===-- AMDGPUIndirectAddressing.cpp - Indirect Adressing Support ---------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+///
+/// Instructions can use indirect addressing to index the register file as if it
+/// were memory.  This pass lowers RegisterLoad and RegisterStore instructions
+/// to either a COPY or a MOV that uses indirect addressing.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "R600InstrInfo.h"
+#include "R600MachineFunctionInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/Support/Debug.h"
+
+using namespace llvm;
+
+namespace {
+
+class AMDGPUIndirectAddressingPass : public MachineFunctionPass {
+
+private:
+  static char ID;
+  const AMDGPUInstrInfo *TII;
+
+  bool regHasExplicitDef(MachineRegisterInfo &MRI, unsigned Reg) const;
+
+public:
+  AMDGPUIndirectAddressingPass(TargetMachine &tm) :
+    MachineFunctionPass(ID),
+    TII(static_cast<const AMDGPUInstrInfo*>(tm.getInstrInfo()))
+    { }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF);
+
+  const char *getPassName() const { return "R600 Handle indirect addressing"; }
+
+};
+
+} // End anonymous namespace
+
+char AMDGPUIndirectAddressingPass::ID = 0;
+
+FunctionPass *llvm::createAMDGPUIndirectAddressingPass(TargetMachine &tm) {
+  return new AMDGPUIndirectAddressingPass(tm);
+}
+
+bool AMDGPUIndirectAddressingPass::runOnMachineFunction(MachineFunction &MF) {
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+
+  int IndirectBegin = TII->getIndirectIndexBegin(MF);
+  int IndirectEnd = TII->getIndirectIndexEnd(MF);
+
+  if (IndirectBegin == -1) {
+    // No indirect addressing, we can skip this pass
+    assert(IndirectEnd == -1);
+    return false;
+  }
+
+  // The map keeps track of the indirect address that is represented by
+  // each virtual register. The key is the register and the value is the
+  // indirect address it uses.
+  std::map<unsigned, unsigned> RegisterAddressMap;
+
+  // First pass - Lower all of the RegisterStore instructions and track which
+  // registers are live.
+  for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
+                                                      BB != BB_E; ++BB) {
+    // This map keeps track of the current live indirect registers.
+    // The key is the address and the value is the register
+    std::map<unsigned, unsigned> LiveAddressRegisterMap;
+    MachineBasicBlock &MBB = *BB;
+
+    for (MachineBasicBlock::iterator I = MBB.begin(), Next = llvm::next(I);
+                               I != MBB.end(); I = Next) {
+      Next = llvm::next(I);
+      MachineInstr &MI = *I;
+
+      if (!TII->isRegisterStore(MI)) {
+        continue;
+      }
+
+      // Lower RegisterStore
+
+      unsigned RegIndex = MI.getOperand(2).getImm();
+      unsigned Channel = MI.getOperand(3).getImm();
+      unsigned Address = TII->calculateIndirectAddress(RegIndex, Channel);
+      const TargetRegisterClass *IndirectStoreRegClass =
+                   TII->getIndirectAddrStoreRegClass(MI.getOperand(0).getReg());
+
+      if (MI.getOperand(1).getReg() == AMDGPU::INDIRECT_BASE_ADDR) {
+        // Direct register access.
+        unsigned DstReg = MRI.createVirtualRegister(IndirectStoreRegClass);
+
+        BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::COPY), DstReg)
+                .addOperand(MI.getOperand(0));
+
+        RegisterAddressMap[DstReg] = Address;
+        LiveAddressRegisterMap[Address] = DstReg;
+      } else {
+        // Indirect register access.
+        MachineInstrBuilder MOV = TII->buildIndirectWrite(BB, I,
+                                           MI.getOperand(0).getReg(), // Value
+                                           Address,
+                                           MI.getOperand(1).getReg()); // Offset
+        for (int i = IndirectBegin; i <= IndirectEnd; ++i) {
+          unsigned Addr = TII->calculateIndirectAddress(i, Channel);
+          unsigned DstReg = MRI.createVirtualRegister(IndirectStoreRegClass);
+          MOV.addReg(DstReg, RegState::Define | RegState::Implicit);
+          RegisterAddressMap[DstReg] = Addr;
+          LiveAddressRegisterMap[Addr] = DstReg;
+        }
+      }
+      MI.eraseFromParent();
+    }
+
+    // Update the live-ins of the succesor blocks
+    for (MachineBasicBlock::succ_iterator Succ = MBB.succ_begin(),
+                                          SuccEnd = MBB.succ_end();
+                                          SuccEnd != Succ; ++Succ) {
+      std::map<unsigned, unsigned>::const_iterator Key, KeyEnd;
+      for (Key = LiveAddressRegisterMap.begin(),
+           KeyEnd = LiveAddressRegisterMap.end(); KeyEnd != Key; ++Key) {
+        (*Succ)->addLiveIn(Key->second);
+      }
+    }
+  }
+
+  // Second pass - Lower the RegisterLoad instructions
+  for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
+                                                      BB != BB_E; ++BB) {
+    // Key is the address and the value is the register
+    std::map<unsigned, unsigned> LiveAddressRegisterMap;
+    MachineBasicBlock &MBB = *BB;
+
+    MachineBasicBlock::livein_iterator LI = MBB.livein_begin();
+    while (LI != MBB.livein_end()) {
+      std::vector<unsigned> PhiRegisters;
+
+      // Make sure this live in is used for indirect addressing
+      if (RegisterAddressMap.find(*LI) == RegisterAddressMap.end()) {
+        ++LI;
+        continue;
+      }
+
+      unsigned Address = RegisterAddressMap[*LI];
+      LiveAddressRegisterMap[Address] = *LI;
+      PhiRegisters.push_back(*LI);
+
+      // Check if there are other live in registers which map to the same
+      // indirect address.
+      for (MachineBasicBlock::livein_iterator LJ = llvm::next(LI),
+                                              LE = MBB.livein_end();
+                                              LJ != LE; ++LJ) {
+        unsigned Reg = *LJ;
+        if (RegisterAddressMap.find(Reg) == RegisterAddressMap.end()) {
+          continue;
+        }
+
+        if (RegisterAddressMap[Reg] == Address) {
+          PhiRegisters.push_back(Reg);
+        }
+      }
+
+      if (PhiRegisters.size() == 1) {
+        // We don't need to insert a Phi instruction, so we can just add the
+        // registers to the live list for the block.
+        LiveAddressRegisterMap[Address] = *LI;
+        MBB.removeLiveIn(*LI);
+      } else {
+        // We need to insert a PHI, because we have the same address being
+        // written in multiple predecessor blocks.
+        const TargetRegisterClass *PhiDstClass =
+                   TII->getIndirectAddrStoreRegClass(*(PhiRegisters.begin()));
+        unsigned PhiDstReg = MRI.createVirtualRegister(PhiDstClass);
+        MachineInstrBuilder Phi = BuildMI(MBB, MBB.begin(),
+                                          MBB.findDebugLoc(MBB.begin()),
+                                          TII->get(AMDGPU::PHI), PhiDstReg);
+
+        for (std::vector<unsigned>::const_iterator RI = PhiRegisters.begin(),
+                                                   RE = PhiRegisters.end();
+                                                   RI != RE; ++RI) {
+          unsigned Reg = *RI;
+          MachineInstr *DefInst = MRI.getVRegDef(Reg);
+          assert(DefInst);
+          MachineBasicBlock *RegBlock = DefInst->getParent();
+          Phi.addReg(Reg);
+          Phi.addMBB(RegBlock);
+          MBB.removeLiveIn(Reg);
+        }
+        RegisterAddressMap[PhiDstReg] = Address;
+        LiveAddressRegisterMap[Address] = PhiDstReg;
+      }
+      LI = MBB.livein_begin();
+    }
+
+    for (MachineBasicBlock::iterator I = MBB.begin(), Next = llvm::next(I);
+                               I != MBB.end(); I = Next) {
+      Next = llvm::next(I);
+      MachineInstr &MI = *I;
+
+      if (!TII->isRegisterLoad(MI)) {
+        if (MI.getOpcode() == AMDGPU::PHI) {
+          continue;
+        }
+        // Check for indirect register defs
+        for (unsigned OpIdx = 0, NumOperands = MI.getNumOperands();
+                                 OpIdx < NumOperands; ++OpIdx) {
+          MachineOperand &MO = MI.getOperand(OpIdx);
+          if (MO.isReg() && MO.isDef() &&
+              RegisterAddressMap.find(MO.getReg()) != RegisterAddressMap.end()) {
+            unsigned Reg = MO.getReg();
+            unsigned LiveAddress = RegisterAddressMap[Reg];
+            // Chain the live-ins
+            if (LiveAddressRegisterMap.find(LiveAddress) !=
+                                                     RegisterAddressMap.end()) {
+              MI.addOperand(MachineOperand::CreateReg(
+                                  LiveAddressRegisterMap[LiveAddress],
+                                  false, // isDef
+                                  true,  // isImp
+                                  true));  // isKill
+            }
+            LiveAddressRegisterMap[LiveAddress] = Reg;
+          }
+        }
+        continue;
+      }
+
+      const TargetRegisterClass *SuperIndirectRegClass =
+                                                TII->getSuperIndirectRegClass();
+      const TargetRegisterClass *IndirectLoadRegClass =
+                                             TII->getIndirectAddrLoadRegClass();
+      unsigned IndirectReg = MRI.createVirtualRegister(SuperIndirectRegClass);
+
+      unsigned RegIndex = MI.getOperand(2).getImm();
+      unsigned Channel = MI.getOperand(3).getImm();
+      unsigned Address = TII->calculateIndirectAddress(RegIndex, Channel);
+
+      if (MI.getOperand(1).getReg() == AMDGPU::INDIRECT_BASE_ADDR) {
+        // Direct register access
+        unsigned Reg = LiveAddressRegisterMap[Address];
+        unsigned AddrReg = IndirectLoadRegClass->getRegister(Address);
+
+        if (regHasExplicitDef(MRI, Reg)) {
+          // If the register we are reading from has an explicit def, then that
+          // means it was written via a direct register access (i.e. COPY
+          // or other instruction that doesn't use indirect addressing).  In
+          // this case we know where the value has been stored, so we can just
+          // issue a copy.
+          BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::COPY),
+                  MI.getOperand(0).getReg())
+                  .addReg(Reg);
+        } else {
+          // If the register we are reading has an implicit def, then that
+          // means it was written by an indirect register access (i.e. An
+          // instruction that uses indirect addressing. 
+          BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::COPY),
+                   MI.getOperand(0).getReg())
+                   .addReg(AddrReg)
+                   .addReg(Reg, RegState::Implicit);
+        }
+      } else {
+        // Indirect register access
+
+        // Note on REQ_SEQUENCE instructons: You can't actually use the register
+        // it defines unless  you have an instruction that takes the defined
+        // register class as an operand.
+
+        MachineInstrBuilder Sequence = BuildMI(MBB, I, MBB.findDebugLoc(I),
+                                               TII->get(AMDGPU::REG_SEQUENCE),
+                                               IndirectReg);
+        for (int i = IndirectBegin; i <= IndirectEnd; ++i) {
+          unsigned Addr = TII->calculateIndirectAddress(i, Channel);
+          if (LiveAddressRegisterMap.find(Addr) == LiveAddressRegisterMap.end()) {
+            continue;
+          }
+          unsigned Reg = LiveAddressRegisterMap[Addr];
+
+          // We only need to use REG_SEQUENCE for explicit defs, since the
+          // register coalescer won't do anything with the implicit defs.
+          if (!regHasExplicitDef(MRI, Reg)) {
+            continue;
+          }
+
+          // Insert a REQ_SEQUENCE instruction to force the register allocator
+          // to allocate the virtual register to the correct physical register.
+          Sequence.addReg(LiveAddressRegisterMap[Addr]);
+          Sequence.addImm(TII->getRegisterInfo().getIndirectSubReg(Addr));
+        }
+        MachineInstrBuilder Mov = TII->buildIndirectRead(BB, I,
+                                           MI.getOperand(0).getReg(), // Value
+                                           Address,
+                                           MI.getOperand(1).getReg()); // Offset
+
+
+
+        Mov.addReg(IndirectReg, RegState::Implicit | RegState::Kill);
+        Mov.addReg(LiveAddressRegisterMap[Address], RegState::Implicit);
+
+      }
+      MI.eraseFromParent();
+    }
+  }
+  return false;
+}
+
+bool AMDGPUIndirectAddressingPass::regHasExplicitDef(MachineRegisterInfo &MRI,
+                                                  unsigned Reg) const {
+  MachineInstr *DefInstr = MRI.getVRegDef(Reg);
+
+  if (!DefInstr) {
+    return false;
+  }
+
+  if (DefInstr->getOpcode() == AMDGPU::PHI) {
+    bool Explicit = false;
+    for (MachineInstr::const_mop_iterator I = DefInstr->operands_begin(),
+                                          E = DefInstr->operands_end();
+                                          I != E; ++I) {
+      const MachineOperand &MO = *I;
+      if (!MO.isReg() || MO.isDef()) {
+        continue;
+      }
+
+      Explicit = Explicit || regHasExplicitDef(MRI, MO.getReg());
+    }
+    return Explicit;
+  }
+
+  return DefInstr->getOperand(0).isReg() &&
+         DefInstr->getOperand(0).getReg() == Reg;
+}
diff --git a/lib/Target/R600/AMDGPUInstrInfo.cpp b/lib/Target/R600/AMDGPUInstrInfo.cpp
new file mode 100644
index 000000000000..30f736c84c25
--- /dev/null
+++ b/lib/Target/R600/AMDGPUInstrInfo.cpp
@@ -0,0 +1,267 @@
+//===-- AMDGPUInstrInfo.cpp - Base class for AMD GPU InstrInfo ------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Implementation of the TargetInstrInfo class that is common to all
+/// AMD GPUs.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUInstrInfo.h"
+#include "AMDGPURegisterInfo.h"
+#include "AMDGPUTargetMachine.h"
+#include "AMDIL.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+
+#define GET_INSTRINFO_CTOR
+#define GET_INSTRMAP_INFO
+#include "AMDGPUGenInstrInfo.inc"
+
+using namespace llvm;
+
+AMDGPUInstrInfo::AMDGPUInstrInfo(TargetMachine &tm)
+  : AMDGPUGenInstrInfo(0,0), RI(tm, *this), TM(tm) { }
+
+const AMDGPURegisterInfo &AMDGPUInstrInfo::getRegisterInfo() const {
+  return RI;
+}
+
+bool AMDGPUInstrInfo::isCoalescableExtInstr(const MachineInstr &MI,
+                                           unsigned &SrcReg, unsigned &DstReg,
+                                           unsigned &SubIdx) const {
+// TODO: Implement this function
+  return false;
+}
+
+unsigned AMDGPUInstrInfo::isLoadFromStackSlot(const MachineInstr *MI,
+                                             int &FrameIndex) const {
+// TODO: Implement this function
+  return 0;
+}
+
+unsigned AMDGPUInstrInfo::isLoadFromStackSlotPostFE(const MachineInstr *MI,
+                                                   int &FrameIndex) const {
+// TODO: Implement this function
+  return 0;
+}
+
+bool AMDGPUInstrInfo::hasLoadFromStackSlot(const MachineInstr *MI,
+                                          const MachineMemOperand *&MMO,
+                                          int &FrameIndex) const {
+// TODO: Implement this function
+  return false;
+}
+unsigned AMDGPUInstrInfo::isStoreFromStackSlot(const MachineInstr *MI,
+                                              int &FrameIndex) const {
+// TODO: Implement this function
+  return 0;
+}
+unsigned AMDGPUInstrInfo::isStoreFromStackSlotPostFE(const MachineInstr *MI,
+                                                    int &FrameIndex) const {
+// TODO: Implement this function
+  return 0;
+}
+bool AMDGPUInstrInfo::hasStoreFromStackSlot(const MachineInstr *MI,
+                                           const MachineMemOperand *&MMO,
+                                           int &FrameIndex) const {
+// TODO: Implement this function
+  return false;
+}
+
+MachineInstr *
+AMDGPUInstrInfo::convertToThreeAddress(MachineFunction::iterator &MFI,
+                                      MachineBasicBlock::iterator &MBBI,
+                                      LiveVariables *LV) const {
+// TODO: Implement this function
+  return NULL;
+}
+bool AMDGPUInstrInfo::getNextBranchInstr(MachineBasicBlock::iterator &iter,
+                                        MachineBasicBlock &MBB) const {
+  while (iter != MBB.end()) {
+    switch (iter->getOpcode()) {
+    default:
+      break;
+    case AMDGPU::BRANCH_COND_i32:
+    case AMDGPU::BRANCH_COND_f32:
+    case AMDGPU::BRANCH:
+      return true;
+    };
+    ++iter;
+  }
+  return false;
+}
+
+MachineBasicBlock::iterator skipFlowControl(MachineBasicBlock *MBB) {
+  MachineBasicBlock::iterator tmp = MBB->end();
+  if (!MBB->size()) {
+    return MBB->end();
+  }
+  while (--tmp) {
+    if (tmp->getOpcode() == AMDGPU::ENDLOOP
+        || tmp->getOpcode() == AMDGPU::ENDIF
+        || tmp->getOpcode() == AMDGPU::ELSE) {
+      if (tmp == MBB->begin()) {
+        return tmp;
+      } else {
+        continue;
+      }
+    }  else {
+      return ++tmp;
+    }
+  }
+  return MBB->end();
+}
+
+void
+AMDGPUInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
+                                    MachineBasicBlock::iterator MI,
+                                    unsigned SrcReg, bool isKill,
+                                    int FrameIndex,
+                                    const TargetRegisterClass *RC,
+                                    const TargetRegisterInfo *TRI) const {
+  assert(!"Not Implemented");
+}
+
+void
+AMDGPUInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator MI,
+                                     unsigned DestReg, int FrameIndex,
+                                     const TargetRegisterClass *RC,
+                                     const TargetRegisterInfo *TRI) const {
+  assert(!"Not Implemented");
+}
+
+MachineInstr *
+AMDGPUInstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
+                                      MachineInstr *MI,
+                                      const SmallVectorImpl<unsigned> &Ops,
+                                      int FrameIndex) const {
+// TODO: Implement this function
+  return 0;
+}
+MachineInstr*
+AMDGPUInstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
+                                      MachineInstr *MI,
+                                      const SmallVectorImpl<unsigned> &Ops,
+                                      MachineInstr *LoadMI) const {
+  // TODO: Implement this function
+  return 0;
+}
+bool
+AMDGPUInstrInfo::canFoldMemoryOperand(const MachineInstr *MI,
+                                     const SmallVectorImpl<unsigned> &Ops) const {
+  // TODO: Implement this function
+  return false;
+}
+bool
+AMDGPUInstrInfo::unfoldMemoryOperand(MachineFunction &MF, MachineInstr *MI,
+                                 unsigned Reg, bool UnfoldLoad,
+                                 bool UnfoldStore,
+                                 SmallVectorImpl<MachineInstr*> &NewMIs) const {
+  // TODO: Implement this function
+  return false;
+}
+
+bool
+AMDGPUInstrInfo::unfoldMemoryOperand(SelectionDAG &DAG, SDNode *N,
+                                    SmallVectorImpl<SDNode*> &NewNodes) const {
+  // TODO: Implement this function
+  return false;
+}
+
+unsigned
+AMDGPUInstrInfo::getOpcodeAfterMemoryUnfold(unsigned Opc,
+                                           bool UnfoldLoad, bool UnfoldStore,
+                                           unsigned *LoadRegIndex) const {
+  // TODO: Implement this function
+  return 0;
+}
+
+bool AMDGPUInstrInfo::shouldScheduleLoadsNear(SDNode *Load1, SDNode *Load2,
+                                             int64_t Offset1, int64_t Offset2,
+                                             unsigned NumLoads) const {
+  assert(Offset2 > Offset1
+         && "Second offset should be larger than first offset!");
+  // If we have less than 16 loads in a row, and the offsets are within 16,
+  // then schedule together.
+  // TODO: Make the loads schedule near if it fits in a cacheline
+  return (NumLoads < 16 && (Offset2 - Offset1) < 16);
+}
+
+bool
+AMDGPUInstrInfo::ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond)
+  const {
+  // TODO: Implement this function
+  return true;
+}
+void AMDGPUInstrInfo::insertNoop(MachineBasicBlock &MBB,
+                                MachineBasicBlock::iterator MI) const {
+  // TODO: Implement this function
+}
+
+bool AMDGPUInstrInfo::isPredicated(const MachineInstr *MI) const {
+  // TODO: Implement this function
+  return false;
+}
+bool
+AMDGPUInstrInfo::SubsumesPredicate(const SmallVectorImpl<MachineOperand> &Pred1,
+                                  const SmallVectorImpl<MachineOperand> &Pred2)
+  const {
+  // TODO: Implement this function
+  return false;
+}
+
+bool AMDGPUInstrInfo::DefinesPredicate(MachineInstr *MI,
+                                      std::vector<MachineOperand> &Pred) const {
+  // TODO: Implement this function
+  return false;
+}
+
+bool AMDGPUInstrInfo::isPredicable(MachineInstr *MI) const {
+  // TODO: Implement this function
+  return MI->getDesc().isPredicable();
+}
+
+bool
+AMDGPUInstrInfo::isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const {
+  // TODO: Implement this function
+  return true;
+}
+
+bool AMDGPUInstrInfo::isRegisterStore(const MachineInstr &MI) const {
+  return get(MI.getOpcode()).TSFlags & AMDGPU_FLAG_REGISTER_STORE;
+}
+
+bool AMDGPUInstrInfo::isRegisterLoad(const MachineInstr &MI) const {
+  return get(MI.getOpcode()).TSFlags & AMDGPU_FLAG_REGISTER_LOAD;
+}
+
+
+void AMDGPUInstrInfo::convertToISA(MachineInstr & MI, MachineFunction &MF,
+    DebugLoc DL) const {
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+  const AMDGPURegisterInfo & RI = getRegisterInfo();
+
+  for (unsigned i = 0; i < MI.getNumOperands(); i++) {
+    MachineOperand &MO = MI.getOperand(i);
+    // Convert dst regclass to one that is supported by the ISA
+    if (MO.isReg() && MO.isDef()) {
+      if (TargetRegisterInfo::isVirtualRegister(MO.getReg())) {
+        const TargetRegisterClass * oldRegClass = MRI.getRegClass(MO.getReg());
+        const TargetRegisterClass * newRegClass = RI.getISARegClass(oldRegClass);
+
+        assert(newRegClass);
+
+        MRI.setRegClass(MO.getReg(), newRegClass);
+      }
+    }
+  }
+}
diff --git a/lib/Target/R600/AMDGPUInstrInfo.h b/lib/Target/R600/AMDGPUInstrInfo.h
new file mode 100644
index 000000000000..3909e4e105ee
--- /dev/null
+++ b/lib/Target/R600/AMDGPUInstrInfo.h
@@ -0,0 +1,206 @@
+//===-- AMDGPUInstrInfo.h - AMDGPU Instruction Information ------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Contains the definition of a TargetInstrInfo class that is common
+/// to all AMD GPUs.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPUINSTRUCTIONINFO_H
+#define AMDGPUINSTRUCTIONINFO_H
+
+#include "AMDGPUInstrInfo.h"
+#include "AMDGPURegisterInfo.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include <map>
+
+#define GET_INSTRINFO_HEADER
+#define GET_INSTRINFO_ENUM
+#include "AMDGPUGenInstrInfo.inc"
+
+#define OPCODE_IS_ZERO_INT AMDGPU::PRED_SETE_INT
+#define OPCODE_IS_NOT_ZERO_INT AMDGPU::PRED_SETNE_INT
+#define OPCODE_IS_ZERO AMDGPU::PRED_SETE
+#define OPCODE_IS_NOT_ZERO AMDGPU::PRED_SETNE
+
+namespace llvm {
+
+class AMDGPUTargetMachine;
+class MachineFunction;
+class MachineInstr;
+class MachineInstrBuilder;
+
+class AMDGPUInstrInfo : public AMDGPUGenInstrInfo {
+private:
+  const AMDGPURegisterInfo RI;
+  bool getNextBranchInstr(MachineBasicBlock::iterator &iter,
+                          MachineBasicBlock &MBB) const;
+protected:
+  TargetMachine &TM;
+public:
+  explicit AMDGPUInstrInfo(TargetMachine &tm);
+
+  virtual const AMDGPURegisterInfo &getRegisterInfo() const = 0;
+
+  bool isCoalescableExtInstr(const MachineInstr &MI, unsigned &SrcReg,
+                             unsigned &DstReg, unsigned &SubIdx) const;
+
+  unsigned isLoadFromStackSlot(const MachineInstr *MI, int &FrameIndex) const;
+  unsigned isLoadFromStackSlotPostFE(const MachineInstr *MI,
+                                     int &FrameIndex) const;
+  bool hasLoadFromStackSlot(const MachineInstr *MI,
+                            const MachineMemOperand *&MMO,
+                            int &FrameIndex) const;
+  unsigned isStoreFromStackSlot(const MachineInstr *MI, int &FrameIndex) const;
+  unsigned isStoreFromStackSlotPostFE(const MachineInstr *MI,
+                                      int &FrameIndex) const;
+  bool hasStoreFromStackSlot(const MachineInstr *MI,
+                             const MachineMemOperand *&MMO,
+                             int &FrameIndex) const;
+
+  MachineInstr *
+  convertToThreeAddress(MachineFunction::iterator &MFI,
+                        MachineBasicBlock::iterator &MBBI,
+                        LiveVariables *LV) const;
+
+
+  virtual void copyPhysReg(MachineBasicBlock &MBB,
+                           MachineBasicBlock::iterator MI, DebugLoc DL,
+                           unsigned DestReg, unsigned SrcReg,
+                           bool KillSrc) const = 0;
+
+  void storeRegToStackSlot(MachineBasicBlock &MBB,
+                           MachineBasicBlock::iterator MI,
+                           unsigned SrcReg, bool isKill, int FrameIndex,
+                           const TargetRegisterClass *RC,
+                           const TargetRegisterInfo *TRI) const;
+  void loadRegFromStackSlot(MachineBasicBlock &MBB,
+                            MachineBasicBlock::iterator MI,
+                            unsigned DestReg, int FrameIndex,
+                            const TargetRegisterClass *RC,
+                            const TargetRegisterInfo *TRI) const;
+
+protected:
+  MachineInstr *foldMemoryOperandImpl(MachineFunction &MF,
+                                      MachineInstr *MI,
+                                      const SmallVectorImpl<unsigned> &Ops,
+                                      int FrameIndex) const;
+  MachineInstr *foldMemoryOperandImpl(MachineFunction &MF,
+                                      MachineInstr *MI,
+                                      const SmallVectorImpl<unsigned> &Ops,
+                                      MachineInstr *LoadMI) const;
+public:
+  bool canFoldMemoryOperand(const MachineInstr *MI,
+                            const SmallVectorImpl<unsigned> &Ops) const;
+  bool unfoldMemoryOperand(MachineFunction &MF, MachineInstr *MI,
+                           unsigned Reg, bool UnfoldLoad, bool UnfoldStore,
+                           SmallVectorImpl<MachineInstr *> &NewMIs) const;
+  bool unfoldMemoryOperand(SelectionDAG &DAG, SDNode *N,
+                           SmallVectorImpl<SDNode *> &NewNodes) const;
+  unsigned getOpcodeAfterMemoryUnfold(unsigned Opc,
+                                      bool UnfoldLoad, bool UnfoldStore,
+                                      unsigned *LoadRegIndex = 0) const;
+  bool shouldScheduleLoadsNear(SDNode *Load1, SDNode *Load2,
+                               int64_t Offset1, int64_t Offset2,
+                               unsigned NumLoads) const;
+
+  bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const;
+  void insertNoop(MachineBasicBlock &MBB,
+                  MachineBasicBlock::iterator MI) const;
+  bool isPredicated(const MachineInstr *MI) const;
+  bool SubsumesPredicate(const SmallVectorImpl<MachineOperand> &Pred1,
+                         const SmallVectorImpl<MachineOperand> &Pred2) const;
+  bool DefinesPredicate(MachineInstr *MI,
+                        std::vector<MachineOperand> &Pred) const;
+  bool isPredicable(MachineInstr *MI) const;
+  bool isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const;
+
+  // Helper functions that check the opcode for status information
+  bool isLoadInst(llvm::MachineInstr *MI) const;
+  bool isExtLoadInst(llvm::MachineInstr *MI) const;
+  bool isSWSExtLoadInst(llvm::MachineInstr *MI) const;
+  bool isSExtLoadInst(llvm::MachineInstr *MI) const;
+  bool isZExtLoadInst(llvm::MachineInstr *MI) const;
+  bool isAExtLoadInst(llvm::MachineInstr *MI) const;
+  bool isStoreInst(llvm::MachineInstr *MI) const;
+  bool isTruncStoreInst(llvm::MachineInstr *MI) const;
+  bool isRegisterStore(const MachineInstr &MI) const;
+  bool isRegisterLoad(const MachineInstr &MI) const;
+
+//===---------------------------------------------------------------------===//
+// Pure virtual funtions to be implemented by sub-classes.
+//===---------------------------------------------------------------------===//
+
+  virtual MachineInstr* getMovImmInstr(MachineFunction *MF, unsigned DstReg,
+                                       int64_t Imm) const = 0;
+  virtual unsigned getIEQOpcode() const = 0;
+  virtual bool isMov(unsigned opcode) const = 0;
+
+  /// \returns the smallest register index that will be accessed by an indirect
+  /// read or write or -1 if indirect addressing is not used by this program.
+  virtual int getIndirectIndexBegin(const MachineFunction &MF) const = 0;
+
+  /// \returns the largest register index that will be accessed by an indirect
+  /// read or write or -1 if indirect addressing is not used by this program.
+  virtual int getIndirectIndexEnd(const MachineFunction &MF) const = 0;
+
+  /// \brief Calculate the "Indirect Address" for the given \p RegIndex and
+  ///        \p Channel
+  ///
+  /// We model indirect addressing using a virtual address space that can be
+  /// accesed with loads and stores.  The "Indirect Address" is the memory
+  /// address in this virtual address space that maps to the given \p RegIndex
+  /// and \p Channel.
+  virtual unsigned calculateIndirectAddress(unsigned RegIndex,
+                                            unsigned Channel) const = 0;
+
+  /// \returns The register class to be used for storing values to an
+  /// "Indirect Address" .
+  virtual const TargetRegisterClass *getIndirectAddrStoreRegClass(
+                                                  unsigned SourceReg) const = 0;
+
+  /// \returns The register class to be used for loading values from
+  /// an "Indirect Address" .
+  virtual const TargetRegisterClass *getIndirectAddrLoadRegClass() const = 0;
+
+  /// \brief Build instruction(s) for an indirect register write.
+  ///
+  /// \returns The instruction that performs the indirect register write
+  virtual MachineInstrBuilder buildIndirectWrite(MachineBasicBlock *MBB,
+                                    MachineBasicBlock::iterator I,
+                                    unsigned ValueReg, unsigned Address,
+                                    unsigned OffsetReg) const = 0;
+
+  /// \brief Build instruction(s) for an indirect register read.
+  ///
+  /// \returns The instruction that performs the indirect register read
+  virtual MachineInstrBuilder buildIndirectRead(MachineBasicBlock *MBB,
+                                    MachineBasicBlock::iterator I,
+                                    unsigned ValueReg, unsigned Address,
+                                    unsigned OffsetReg) const = 0;
+
+  /// \returns the register class whose sub registers are the set of all
+  /// possible registers that can be used for indirect addressing.
+  virtual const TargetRegisterClass *getSuperIndirectRegClass() const = 0;
+
+
+  /// \brief Convert the AMDIL MachineInstr to a supported ISA
+  /// MachineInstr
+  virtual void convertToISA(MachineInstr & MI, MachineFunction &MF,
+    DebugLoc DL) const;
+
+};
+
+} // End llvm namespace
+
+#define AMDGPU_FLAG_REGISTER_LOAD  (UINT64_C(1) << 63)
+#define AMDGPU_FLAG_REGISTER_STORE (UINT64_C(1) << 62)
+
+#endif // AMDGPUINSTRINFO_H
diff --git a/lib/Target/R600/AMDGPUInstrInfo.td b/lib/Target/R600/AMDGPUInstrInfo.td
new file mode 100644
index 000000000000..b66ae879dc20
--- /dev/null
+++ b/lib/Target/R600/AMDGPUInstrInfo.td
@@ -0,0 +1,82 @@
+//===-- AMDGPUInstrInfo.td - AMDGPU DAG nodes --------------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains DAG node defintions for the AMDGPU target.
+//
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// AMDGPU DAG Profiles
+//===----------------------------------------------------------------------===//
+
+def AMDGPUDTIntTernaryOp : SDTypeProfile<1, 3, [
+  SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisInt<0>, SDTCisInt<3>
+]>;
+
+//===----------------------------------------------------------------------===//
+// AMDGPU DAG Nodes
+//
+
+// out = ((a << 32) | b) >> c)
+//
+// Can be used to optimize rtol:
+// rotl(a, b) = bitalign(a, a, 32 - b)
+def AMDGPUbitalign : SDNode<"AMDGPUISD::BITALIGN", AMDGPUDTIntTernaryOp>;
+
+// This argument to this node is a dword address.
+def AMDGPUdwordaddr : SDNode<"AMDGPUISD::DWORDADDR", SDTIntUnaryOp>;
+
+// out = a - floor(a)
+def AMDGPUfract : SDNode<"AMDGPUISD::FRACT", SDTFPUnaryOp>;
+
+// out = max(a, b) a and b are floats
+def AMDGPUfmax : SDNode<"AMDGPUISD::FMAX", SDTFPBinOp,
+  [SDNPCommutative, SDNPAssociative]
+>;
+
+// out = max(a, b) a and b are signed ints
+def AMDGPUsmax : SDNode<"AMDGPUISD::SMAX", SDTIntBinOp,
+  [SDNPCommutative, SDNPAssociative]
+>;
+
+// out = max(a, b) a and b are unsigned ints
+def AMDGPUumax : SDNode<"AMDGPUISD::UMAX", SDTIntBinOp,
+  [SDNPCommutative, SDNPAssociative]
+>;
+
+// out = min(a, b) a and b are floats
+def AMDGPUfmin : SDNode<"AMDGPUISD::FMIN", SDTFPBinOp,
+  [SDNPCommutative, SDNPAssociative]
+>;
+
+// out = min(a, b) a snd b are signed ints
+def AMDGPUsmin : SDNode<"AMDGPUISD::SMIN", SDTIntBinOp,
+  [SDNPCommutative, SDNPAssociative]
+>;
+
+// out = min(a, b) a and b are unsigned ints
+def AMDGPUumin : SDNode<"AMDGPUISD::UMIN", SDTIntBinOp,
+  [SDNPCommutative, SDNPAssociative]
+>;
+
+// urecip - This operation is a helper for integer division, it returns the
+// result of 1 / a as a fractional unsigned integer.
+// out = (2^32 / a) + e
+// e is rounding error
+def AMDGPUurecip : SDNode<"AMDGPUISD::URECIP", SDTIntUnaryOp>;
+
+def fpow : SDNode<"ISD::FPOW", SDTFPBinOp>;
+
+def AMDGPUregister_load : SDNode<"AMDGPUISD::REGISTER_LOAD",
+                          SDTypeProfile<1, 2, [SDTCisPtrTy<1>, SDTCisInt<2>]>,
+                          [SDNPHasChain, SDNPMayLoad]>;
+
+def AMDGPUregister_store : SDNode<"AMDGPUISD::REGISTER_STORE",
+                           SDTypeProfile<0, 3, [SDTCisPtrTy<1>, SDTCisInt<2>]>,
+                           [SDNPHasChain, SDNPMayStore]>;
diff --git a/lib/Target/R600/AMDGPUInstructions.td b/lib/Target/R600/AMDGPUInstructions.td
new file mode 100644
index 000000000000..e740348717c7
--- /dev/null
+++ b/lib/Target/R600/AMDGPUInstructions.td
@@ -0,0 +1,266 @@
+//===-- AMDGPUInstructions.td - Common instruction defs ---*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains instruction defs that are common to all hw codegen
+// targets.
+//
+//===----------------------------------------------------------------------===//
+
+class AMDGPUInst <dag outs, dag ins, string asm, list<dag> pattern> : Instruction {
+  field bit isRegisterLoad = 0;
+  field bit isRegisterStore = 0;
+
+  let Namespace = "AMDGPU";
+  let OutOperandList = outs;
+  let InOperandList = ins;
+  let AsmString = asm;
+  let Pattern = pattern;
+  let Itinerary = NullALU;
+
+  let TSFlags{63} = isRegisterLoad;
+  let TSFlags{62} = isRegisterStore;
+}
+
+class AMDGPUShaderInst <dag outs, dag ins, string asm, list<dag> pattern>
+    : AMDGPUInst<outs, ins, asm, pattern> {
+
+  field bits<32> Inst = 0xffffffff;
+
+}
+
+def InstFlag : OperandWithDefaultOps <i32, (ops (i32 0))>;
+
+def COND_EQ : PatLeaf <
+  (cond),
+  [{switch(N->get()){{default: return false;
+                     case ISD::SETOEQ: case ISD::SETUEQ:
+                     case ISD::SETEQ: return true;}}}]
+>;
+
+def COND_NE : PatLeaf <
+  (cond),
+  [{switch(N->get()){{default: return false;
+                     case ISD::SETONE: case ISD::SETUNE:
+                     case ISD::SETNE: return true;}}}]
+>;
+def COND_GT : PatLeaf <
+  (cond),
+  [{switch(N->get()){{default: return false;
+                     case ISD::SETOGT: case ISD::SETUGT:
+                     case ISD::SETGT: return true;}}}]
+>;
+
+def COND_GE : PatLeaf <
+  (cond),
+  [{switch(N->get()){{default: return false;
+                     case ISD::SETOGE: case ISD::SETUGE:
+                     case ISD::SETGE: return true;}}}]
+>;
+
+def COND_LT : PatLeaf <
+  (cond),
+  [{switch(N->get()){{default: return false;
+                     case ISD::SETOLT: case ISD::SETULT:
+                     case ISD::SETLT: return true;}}}]
+>;
+
+def COND_LE : PatLeaf <
+  (cond),
+  [{switch(N->get()){{default: return false;
+                     case ISD::SETOLE: case ISD::SETULE:
+                     case ISD::SETLE: return true;}}}]
+>;
+
+def COND_NULL : PatLeaf <
+  (cond),
+  [{return false;}]
+>;
+
+//===----------------------------------------------------------------------===//
+// Load/Store Pattern Fragments
+//===----------------------------------------------------------------------===//
+
+def zextloadi8_global : PatFrag<(ops node:$ptr), (zextloadi8 node:$ptr), [{
+    return isGlobalLoad(dyn_cast<LoadSDNode>(N));
+}]>;
+
+class Constants {
+int TWO_PI = 0x40c90fdb;
+int PI = 0x40490fdb;
+int TWO_PI_INV = 0x3e22f983;
+}
+def CONST : Constants;
+
+def FP_ZERO : PatLeaf <
+  (fpimm),
+  [{return N->getValueAPF().isZero();}]
+>;
+
+def FP_ONE : PatLeaf <
+  (fpimm),
+  [{return N->isExactlyValue(1.0);}]
+>;
+
+let isCodeGenOnly = 1, isPseudo = 1 in {
+
+let usesCustomInserter = 1  in {
+
+class CLAMP <RegisterClass rc> : AMDGPUShaderInst <
+  (outs rc:$dst),
+  (ins rc:$src0),
+  "CLAMP $dst, $src0",
+  [(set rc:$dst, (int_AMDIL_clamp rc:$src0, (f32 FP_ZERO), (f32 FP_ONE)))]
+>;
+
+class FABS <RegisterClass rc> : AMDGPUShaderInst <
+  (outs rc:$dst),
+  (ins rc:$src0),
+  "FABS $dst, $src0",
+  [(set rc:$dst, (fabs rc:$src0))]
+>;
+
+class FNEG <RegisterClass rc> : AMDGPUShaderInst <
+  (outs rc:$dst),
+  (ins rc:$src0),
+  "FNEG $dst, $src0",
+  [(set rc:$dst, (fneg rc:$src0))]
+>;
+
+} // usesCustomInserter = 1
+
+multiclass RegisterLoadStore <RegisterClass dstClass, Operand addrClass,
+                    ComplexPattern addrPat> {
+  def RegisterLoad : AMDGPUShaderInst <
+    (outs dstClass:$dst),
+    (ins addrClass:$addr, i32imm:$chan),
+    "RegisterLoad $dst, $addr",
+    [(set (i32 dstClass:$dst), (AMDGPUregister_load addrPat:$addr,
+                                                    (i32 timm:$chan)))]
+  > {
+    let isRegisterLoad = 1;
+  }
+
+  def RegisterStore : AMDGPUShaderInst <
+    (outs),
+    (ins dstClass:$val, addrClass:$addr, i32imm:$chan),
+    "RegisterStore $val, $addr",
+    [(AMDGPUregister_store (i32 dstClass:$val), addrPat:$addr, (i32 timm:$chan))]
+  > {
+    let isRegisterStore = 1;
+  }
+}
+
+} // End isCodeGenOnly = 1, isPseudo = 1
+
+/* Generic helper patterns for intrinsics */
+/* -------------------------------------- */
+
+class POW_Common <AMDGPUInst log_ieee, AMDGPUInst exp_ieee, AMDGPUInst mul,
+                  RegisterClass rc> : Pat <
+  (fpow rc:$src0, rc:$src1),
+  (exp_ieee (mul rc:$src1, (log_ieee rc:$src0)))
+>;
+
+/* Other helper patterns */
+/* --------------------- */
+
+/* Extract element pattern */
+class Extract_Element <ValueType sub_type, ValueType vec_type,
+                     RegisterClass vec_class, int sub_idx, 
+                     SubRegIndex sub_reg>: Pat<
+  (sub_type (vector_extract (vec_type vec_class:$src), sub_idx)),
+  (EXTRACT_SUBREG vec_class:$src, sub_reg)
+>;
+
+/* Insert element pattern */
+class Insert_Element <ValueType elem_type, ValueType vec_type,
+                      RegisterClass elem_class, RegisterClass vec_class,
+                      int sub_idx, SubRegIndex sub_reg> : Pat <
+
+  (vec_type (vector_insert (vec_type vec_class:$vec),
+                           (elem_type elem_class:$elem), sub_idx)),
+  (INSERT_SUBREG vec_class:$vec, elem_class:$elem, sub_reg)
+>;
+
+// Vector Build pattern
+class Vector1_Build <ValueType vecType, RegisterClass vectorClass,
+                     ValueType elemType, RegisterClass elemClass> : Pat <
+  (vecType (build_vector (elemType elemClass:$src))),
+  (vecType elemClass:$src)
+>;
+
+class Vector2_Build <ValueType vecType, RegisterClass vectorClass,
+                     ValueType elemType, RegisterClass elemClass> : Pat <
+  (vecType (build_vector (elemType elemClass:$sub0), (elemType elemClass:$sub1))),
+  (INSERT_SUBREG (INSERT_SUBREG
+  (vecType (IMPLICIT_DEF)), elemClass:$sub0, sub0), elemClass:$sub1, sub1)
+>;
+
+class Vector4_Build <ValueType vecType, RegisterClass vectorClass,
+                     ValueType elemType, RegisterClass elemClass> : Pat <
+  (vecType (build_vector (elemType elemClass:$x), (elemType elemClass:$y),
+                         (elemType elemClass:$z), (elemType elemClass:$w))),
+  (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG
+  (vecType (IMPLICIT_DEF)), elemClass:$x, sub0), elemClass:$y, sub1),
+                            elemClass:$z, sub2), elemClass:$w, sub3)
+>;
+
+class Vector8_Build <ValueType vecType, RegisterClass vectorClass,
+                     ValueType elemType, RegisterClass elemClass> : Pat <
+  (vecType (build_vector (elemType elemClass:$sub0), (elemType elemClass:$sub1),
+                         (elemType elemClass:$sub2), (elemType elemClass:$sub3),
+                         (elemType elemClass:$sub4), (elemType elemClass:$sub5),
+                         (elemType elemClass:$sub6), (elemType elemClass:$sub7))),
+  (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG
+  (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG
+  (vecType (IMPLICIT_DEF)), elemClass:$sub0, sub0), elemClass:$sub1, sub1),
+                            elemClass:$sub2, sub2), elemClass:$sub3, sub3),
+                            elemClass:$sub4, sub4), elemClass:$sub5, sub5),
+                            elemClass:$sub6, sub6), elemClass:$sub7, sub7)
+>;
+
+class Vector16_Build <ValueType vecType, RegisterClass vectorClass,
+                      ValueType elemType, RegisterClass elemClass> : Pat <
+  (vecType (build_vector (elemType elemClass:$sub0), (elemType elemClass:$sub1),
+                         (elemType elemClass:$sub2), (elemType elemClass:$sub3),
+                         (elemType elemClass:$sub4), (elemType elemClass:$sub5),
+                         (elemType elemClass:$sub6), (elemType elemClass:$sub7),
+                         (elemType elemClass:$sub8), (elemType elemClass:$sub9),
+                         (elemType elemClass:$sub10), (elemType elemClass:$sub11),
+                         (elemType elemClass:$sub12), (elemType elemClass:$sub13),
+                         (elemType elemClass:$sub14), (elemType elemClass:$sub15))),
+  (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG
+  (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG
+  (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG
+  (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG
+  (vecType (IMPLICIT_DEF)), elemClass:$sub0, sub0), elemClass:$sub1, sub1),
+                            elemClass:$sub2, sub2), elemClass:$sub3, sub3),
+                            elemClass:$sub4, sub4), elemClass:$sub5, sub5),
+                            elemClass:$sub6, sub6), elemClass:$sub7, sub7),
+                            elemClass:$sub8, sub8), elemClass:$sub9, sub9),
+                            elemClass:$sub10, sub10), elemClass:$sub11, sub11),
+                            elemClass:$sub12, sub12), elemClass:$sub13, sub13),
+                            elemClass:$sub14, sub14), elemClass:$sub15, sub15)
+>;
+
+// bitconvert pattern
+class BitConvert <ValueType dt, ValueType st, RegisterClass rc> : Pat <
+  (dt (bitconvert (st rc:$src0))),
+  (dt rc:$src0)
+>;
+
+class DwordAddrPat<ValueType vt, RegisterClass rc> : Pat <
+  (vt (AMDGPUdwordaddr (vt rc:$addr))),
+  (vt rc:$addr)
+>;
+
+include "R600Instructions.td"
+
+include "SIInstrInfo.td"
+
diff --git a/lib/Target/R600/AMDGPUIntrinsics.td b/lib/Target/R600/AMDGPUIntrinsics.td
new file mode 100644
index 000000000000..eecb25b04f79
--- /dev/null
+++ b/lib/Target/R600/AMDGPUIntrinsics.td
@@ -0,0 +1,60 @@
+//===-- AMDGPUIntrinsics.td - Common intrinsics  -*- tablegen -*-----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines intrinsics that are used by all hw codegen targets.
+//
+//===----------------------------------------------------------------------===//
+
+let TargetPrefix = "AMDGPU", isTarget = 1 in {
+
+  def int_AMDGPU_load_const : Intrinsic<[llvm_float_ty], [llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_load_imm : Intrinsic<[llvm_v4f32_ty], [llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_reserve_reg : Intrinsic<[], [llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_store_output : Intrinsic<[], [llvm_float_ty, llvm_i32_ty], []>;
+  def int_AMDGPU_swizzle : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty], [IntrNoMem]>;
+
+  def int_AMDGPU_arl : Intrinsic<[llvm_i32_ty], [llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_cndlt : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_div : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_dp4 : Intrinsic<[llvm_float_ty], [llvm_v4f32_ty, llvm_v4f32_ty], [IntrNoMem]>;
+  def int_AMDGPU_kill : Intrinsic<[], [llvm_float_ty], []>;
+  def int_AMDGPU_kilp : Intrinsic<[], [], []>;
+  def int_AMDGPU_lrp : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_mul : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_pow : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_rcp : Intrinsic<[llvm_float_ty], [llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_rsq : Intrinsic<[llvm_float_ty], [llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_seq : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_sgt : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_sge : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_sle : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_sne : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_mullit : Intrinsic<[llvm_v4f32_ty], [llvm_float_ty, llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_tex : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_txb : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_txf : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_txq : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_txd : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_txl : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_trunc : Intrinsic<[llvm_float_ty], [llvm_float_ty], [IntrNoMem]>;
+  def int_AMDGPU_ddx : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_ddy : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_imax : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_imin : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_umax : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_umin : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_cube : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty], [IntrNoMem]>;
+}
+
+let TargetPrefix = "TGSI", isTarget = 1 in {
+
+  def int_TGSI_lit_z : Intrinsic<[llvm_float_ty], [llvm_float_ty, llvm_float_ty, llvm_float_ty],[IntrNoMem]>;
+}
+
+include "SIIntrinsics.td"
diff --git a/lib/Target/R600/AMDGPUMCInstLower.cpp b/lib/Target/R600/AMDGPUMCInstLower.cpp
new file mode 100644
index 000000000000..1dc1c657dfe5
--- /dev/null
+++ b/lib/Target/R600/AMDGPUMCInstLower.cpp
@@ -0,0 +1,83 @@
+//===- AMDGPUMCInstLower.cpp - Lower AMDGPU MachineInstr to an MCInst -----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Code to lower AMDGPU MachineInstrs to their corresponding MCInst.
+//
+//===----------------------------------------------------------------------===//
+//
+
+#include "AMDGPUMCInstLower.h"
+#include "AMDGPUAsmPrinter.h"
+#include "R600InstrInfo.h"
+#include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/Support/ErrorHandling.h"
+
+using namespace llvm;
+
+AMDGPUMCInstLower::AMDGPUMCInstLower(MCContext &ctx):
+  Ctx(ctx)
+{ }
+
+void AMDGPUMCInstLower::lower(const MachineInstr *MI, MCInst &OutMI) const {
+  OutMI.setOpcode(MI->getOpcode());
+
+  for (unsigned i = 0, e = MI->getNumExplicitOperands(); i != e; ++i) {
+    const MachineOperand &MO = MI->getOperand(i);
+
+    MCOperand MCOp;
+    switch (MO.getType()) {
+    default:
+      llvm_unreachable("unknown operand type");
+    case MachineOperand::MO_FPImmediate: {
+      const APFloat &FloatValue = MO.getFPImm()->getValueAPF();
+      assert(&FloatValue.getSemantics() == &APFloat::IEEEsingle &&
+             "Only floating point immediates are supported at the moment.");
+      MCOp = MCOperand::CreateFPImm(FloatValue.convertToFloat());
+      break;
+    }
+    case MachineOperand::MO_Immediate:
+      MCOp = MCOperand::CreateImm(MO.getImm());
+      break;
+    case MachineOperand::MO_Register:
+      MCOp = MCOperand::CreateReg(MO.getReg());
+      break;
+    case MachineOperand::MO_MachineBasicBlock:
+      MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
+                                   MO.getMBB()->getSymbol(), Ctx));
+    }
+    OutMI.addOperand(MCOp);
+  }
+}
+
+void AMDGPUAsmPrinter::EmitInstruction(const MachineInstr *MI) {
+  AMDGPUMCInstLower MCInstLowering(OutContext);
+
+  if (MI->isBundle()) {
+    const MachineBasicBlock *MBB = MI->getParent();
+    MachineBasicBlock::const_instr_iterator I = MI;
+    ++I;
+    while (I != MBB->end() && I->isInsideBundle()) {
+      MCInst MCBundleInst;
+      const MachineInstr *BundledInst = I;
+      MCInstLowering.lower(BundledInst, MCBundleInst);
+      OutStreamer.EmitInstruction(MCBundleInst);
+      ++I;
+    }
+  } else {
+    MCInst TmpInst;
+    MCInstLowering.lower(MI, TmpInst);
+    OutStreamer.EmitInstruction(TmpInst);
+  }
+}
diff --git a/lib/Target/R600/AMDGPUMCInstLower.h b/lib/Target/R600/AMDGPUMCInstLower.h
new file mode 100644
index 000000000000..d7d538e92599
--- /dev/null
+++ b/lib/Target/R600/AMDGPUMCInstLower.h
@@ -0,0 +1,34 @@
+//===- AMDGPUMCInstLower.h MachineInstr Lowering Interface ------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPU_MCINSTLOWER_H
+#define AMDGPU_MCINSTLOWER_H
+
+namespace llvm {
+
+class MCInst;
+class MCContext;
+class MachineInstr;
+
+class AMDGPUMCInstLower {
+
+  MCContext &Ctx;
+
+public:
+  AMDGPUMCInstLower(MCContext &ctx);
+
+  /// \brief Lower a MachineInstr to an MCInst
+  void lower(const MachineInstr *MI, MCInst &OutMI) const;
+
+};
+
+} // End namespace llvm
+
+#endif //AMDGPU_MCINSTLOWER_H
diff --git a/lib/Target/R600/AMDGPUMachineFunction.cpp b/lib/Target/R600/AMDGPUMachineFunction.cpp
new file mode 100644
index 000000000000..0223ec8e4f3f
--- /dev/null
+++ b/lib/Target/R600/AMDGPUMachineFunction.cpp
@@ -0,0 +1,22 @@
+#include "AMDGPUMachineFunction.h"
+#include "llvm/IR/Attributes.h"
+#include "llvm/IR/Function.h"
+
+namespace llvm {
+
+const char *AMDGPUMachineFunction::ShaderTypeAttribute = "ShaderType";
+
+AMDGPUMachineFunction::AMDGPUMachineFunction(const MachineFunction &MF) :
+    MachineFunctionInfo() {
+  AttributeSet Set = MF.getFunction()->getAttributes();
+  Attribute A = Set.getAttribute(AttributeSet::FunctionIndex,
+                                 ShaderTypeAttribute);
+
+  if (A.isStringAttribute()) {
+    StringRef Str = A.getValueAsString();
+    if (Str.getAsInteger(0, ShaderType))
+      llvm_unreachable("Can't parse shader type!");
+  }
+}
+
+}
diff --git a/lib/Target/R600/AMDGPUMachineFunction.h b/lib/Target/R600/AMDGPUMachineFunction.h
new file mode 100644
index 000000000000..21c8c51dae45
--- /dev/null
+++ b/lib/Target/R600/AMDGPUMachineFunction.h
@@ -0,0 +1,29 @@
+//===-- R600MachineFunctionInfo.h - R600 Machine Function Info ----*- C++ -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPUMACHINEFUNCTION_H
+#define AMDGPUMACHINEFUNCTION_H
+
+#include "llvm/CodeGen/MachineFunction.h"
+
+namespace llvm {
+
+class AMDGPUMachineFunction : public MachineFunctionInfo {
+private:
+  static const char *ShaderTypeAttribute;
+public:
+  AMDGPUMachineFunction(const MachineFunction &MF);
+  unsigned ShaderType;
+};
+
+}
+#endif // AMDGPUMACHINEFUNCTION_H
diff --git a/lib/Target/R600/AMDGPURegisterInfo.cpp b/lib/Target/R600/AMDGPURegisterInfo.cpp
new file mode 100644
index 000000000000..fe994d2d05a1
--- /dev/null
+++ b/lib/Target/R600/AMDGPURegisterInfo.cpp
@@ -0,0 +1,75 @@
+//===-- AMDGPURegisterInfo.cpp - AMDGPU Register Information -------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Parent TargetRegisterInfo class common to all hw codegen targets.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPURegisterInfo.h"
+#include "AMDGPUTargetMachine.h"
+
+using namespace llvm;
+
+AMDGPURegisterInfo::AMDGPURegisterInfo(TargetMachine &tm,
+    const TargetInstrInfo &tii)
+: AMDGPUGenRegisterInfo(0),
+  TM(tm),
+  TII(tii)
+  { }
+
+//===----------------------------------------------------------------------===//
+// Function handling callbacks - Functions are a seldom used feature of GPUS, so
+// they are not supported at this time.
+//===----------------------------------------------------------------------===//
+
+const uint16_t AMDGPURegisterInfo::CalleeSavedReg = AMDGPU::NoRegister;
+
+const uint16_t* AMDGPURegisterInfo::getCalleeSavedRegs(const MachineFunction *MF)
+                                                                         const {
+  return &CalleeSavedReg;
+}
+
+void AMDGPURegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
+                                             int SPAdj,
+                                             unsigned FIOperandNum,
+                                             RegScavenger *RS) const {
+  assert(!"Subroutines not supported yet");
+}
+
+unsigned AMDGPURegisterInfo::getFrameRegister(const MachineFunction &MF) const {
+  assert(!"Subroutines not supported yet");
+  return 0;
+}
+
+unsigned AMDGPURegisterInfo::getIndirectSubReg(unsigned IndirectIndex) const {
+
+  switch(IndirectIndex) {
+  case 0: return AMDGPU::sub0;
+  case 1: return AMDGPU::sub1;
+  case 2: return AMDGPU::sub2;
+  case 3: return AMDGPU::sub3;
+  case 4: return AMDGPU::sub4;
+  case 5: return AMDGPU::sub5;
+  case 6: return AMDGPU::sub6;
+  case 7: return AMDGPU::sub7;
+  case 8: return AMDGPU::sub8;
+  case 9: return AMDGPU::sub9;
+  case 10: return AMDGPU::sub10;
+  case 11: return AMDGPU::sub11;
+  case 12: return AMDGPU::sub12;
+  case 13: return AMDGPU::sub13;
+  case 14: return AMDGPU::sub14;
+  case 15: return AMDGPU::sub15;
+  default: llvm_unreachable("indirect index out of range");
+  }
+}
+
+#define GET_REGINFO_TARGET_DESC
+#include "AMDGPUGenRegisterInfo.inc"
diff --git a/lib/Target/R600/AMDGPURegisterInfo.h b/lib/Target/R600/AMDGPURegisterInfo.h
new file mode 100644
index 000000000000..1fc88e7455b9
--- /dev/null
+++ b/lib/Target/R600/AMDGPURegisterInfo.h
@@ -0,0 +1,66 @@
+//===-- AMDGPURegisterInfo.h - AMDGPURegisterInfo Interface -*- C++ -*-----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief TargetRegisterInfo interface that is implemented by all hw codegen
+/// targets.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPUREGISTERINFO_H
+#define AMDGPUREGISTERINFO_H
+
+#include "llvm/ADT/BitVector.h"
+#include "llvm/Target/TargetRegisterInfo.h"
+
+#define GET_REGINFO_HEADER
+#define GET_REGINFO_ENUM
+#include "AMDGPUGenRegisterInfo.inc"
+
+namespace llvm {
+
+class AMDGPUTargetMachine;
+class TargetInstrInfo;
+
+struct AMDGPURegisterInfo : public AMDGPUGenRegisterInfo {
+  TargetMachine &TM;
+  const TargetInstrInfo &TII;
+  static const uint16_t CalleeSavedReg;
+
+  AMDGPURegisterInfo(TargetMachine &tm, const TargetInstrInfo &tii);
+
+  virtual BitVector getReservedRegs(const MachineFunction &MF) const {
+    assert(!"Unimplemented");  return BitVector();
+  }
+
+  /// \param RC is an AMDIL reg class.
+  ///
+  /// \returns The ISA reg class that is equivalent to \p RC.
+  virtual const TargetRegisterClass * getISARegClass(
+                                         const TargetRegisterClass * RC) const {
+    assert(!"Unimplemented"); return NULL;
+  }
+
+  virtual const TargetRegisterClass* getCFGStructurizerRegClass(MVT VT) const {
+    assert(!"Unimplemented"); return NULL;
+  }
+
+  const uint16_t* getCalleeSavedRegs(const MachineFunction *MF) const;
+  void eliminateFrameIndex(MachineBasicBlock::iterator MI, int SPAdj,
+                           unsigned FIOperandNum,
+                           RegScavenger *RS) const;
+  unsigned getFrameRegister(const MachineFunction &MF) const;
+
+  unsigned getIndirectSubReg(unsigned IndirectIndex) const;
+
+};
+
+} // End namespace llvm
+
+#endif // AMDIDSAREGISTERINFO_H
diff --git a/lib/Target/R600/AMDGPURegisterInfo.td b/lib/Target/R600/AMDGPURegisterInfo.td
new file mode 100644
index 000000000000..b5aca0347fb0
--- /dev/null
+++ b/lib/Target/R600/AMDGPURegisterInfo.td
@@ -0,0 +1,25 @@
+//===-- AMDGPURegisterInfo.td - AMDGPU register info -------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Tablegen register definitions common to all hw codegen targets.
+//
+//===----------------------------------------------------------------------===//
+
+let Namespace = "AMDGPU" in {
+
+foreach Index = 0-15 in {
+  def sub#Index : SubRegIndex;
+}
+
+def INDIRECT_BASE_ADDR : Register <"INDIRECT_BASE_ADDR">;
+
+}
+
+include "R600RegisterInfo.td"
+include "SIRegisterInfo.td"
diff --git a/lib/Target/R600/AMDGPUStructurizeCFG.cpp b/lib/Target/R600/AMDGPUStructurizeCFG.cpp
new file mode 100644
index 000000000000..dea43b874c6f
--- /dev/null
+++ b/lib/Target/R600/AMDGPUStructurizeCFG.cpp
@@ -0,0 +1,896 @@
+//===-- AMDGPUStructurizeCFG.cpp -  ------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// The pass implemented in this file transforms the programs control flow
+/// graph into a form that's suitable for code generation on hardware that
+/// implements control flow by execution masking. This currently includes all
+/// AMD GPUs but may as well be useful for other types of hardware.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "llvm/ADT/SCCIterator.h"
+#include "llvm/ADT/MapVector.h"
+#include "llvm/Analysis/RegionInfo.h"
+#include "llvm/Analysis/RegionIterator.h"
+#include "llvm/Analysis/RegionPass.h"
+#include "llvm/IR/Module.h"
+#include "llvm/Transforms/Utils/SSAUpdater.h"
+#include "llvm/Support/PatternMatch.h"
+
+using namespace llvm;
+using namespace llvm::PatternMatch;
+
+namespace {
+
+// Definition of the complex types used in this pass.
+
+typedef std::pair<BasicBlock *, Value *> BBValuePair;
+
+typedef SmallVector<RegionNode*, 8> RNVector;
+typedef SmallVector<BasicBlock*, 8> BBVector;
+typedef SmallVector<BranchInst*, 8> BranchVector;
+typedef SmallVector<BBValuePair, 2> BBValueVector;
+
+typedef SmallPtrSet<BasicBlock *, 8> BBSet;
+
+typedef MapVector<PHINode *, BBValueVector> PhiMap;
+typedef MapVector<BasicBlock *, BBVector> BB2BBVecMap;
+
+typedef DenseMap<DomTreeNode *, unsigned> DTN2UnsignedMap;
+typedef DenseMap<BasicBlock *, PhiMap> BBPhiMap;
+typedef DenseMap<BasicBlock *, Value *> BBPredicates;
+typedef DenseMap<BasicBlock *, BBPredicates> PredMap;
+typedef DenseMap<BasicBlock *, BasicBlock*> BB2BBMap;
+
+// The name for newly created blocks.
+
+static const char *FlowBlockName = "Flow";
+
+/// @brief Find the nearest common dominator for multiple BasicBlocks
+///
+/// Helper class for AMDGPUStructurizeCFG
+/// TODO: Maybe move into common code
+class NearestCommonDominator {
+
+  DominatorTree *DT;
+
+  DTN2UnsignedMap IndexMap;
+
+  BasicBlock *Result;
+  unsigned ResultIndex;
+  bool ExplicitMentioned;
+
+public:
+  /// \brief Start a new query
+  NearestCommonDominator(DominatorTree *DomTree) {
+    DT = DomTree;
+    Result = 0;
+  }
+
+  /// \brief Add BB to the resulting dominator
+  void addBlock(BasicBlock *BB, bool Remember = true) {
+
+    DomTreeNode *Node = DT->getNode(BB);
+
+    if (Result == 0) {
+      unsigned Numbering = 0;
+      for (;Node;Node = Node->getIDom())
+        IndexMap[Node] = ++Numbering;
+      Result = BB;
+      ResultIndex = 1;
+      ExplicitMentioned = Remember;
+      return;
+    }
+
+    for (;Node;Node = Node->getIDom())
+      if (IndexMap.count(Node))
+        break;
+      else
+        IndexMap[Node] = 0;
+
+    assert(Node && "Dominator tree invalid!");
+
+    unsigned Numbering = IndexMap[Node];
+    if (Numbering > ResultIndex) {
+      Result = Node->getBlock();
+      ResultIndex = Numbering;
+      ExplicitMentioned = Remember && (Result == BB);
+    } else if (Numbering == ResultIndex) {
+      ExplicitMentioned |= Remember;
+    }
+  }
+
+  /// \brief Is "Result" one of the BBs added with "Remember" = True?
+  bool wasResultExplicitMentioned() {
+    return ExplicitMentioned;
+  }
+
+  /// \brief Get the query result
+  BasicBlock *getResult() {
+    return Result;
+  }
+};
+
+/// @brief Transforms the control flow graph on one single entry/exit region
+/// at a time.
+///
+/// After the transform all "If"/"Then"/"Else" style control flow looks like
+/// this:
+///
+/// \verbatim
+/// 1
+/// ||
+/// | |
+/// 2 |
+/// | /
+/// |/   
+/// 3
+/// ||   Where:
+/// | |  1 = "If" block, calculates the condition
+/// 4 |  2 = "Then" subregion, runs if the condition is true
+/// | /  3 = "Flow" blocks, newly inserted flow blocks, rejoins the flow
+/// |/   4 = "Else" optional subregion, runs if the condition is false
+/// 5    5 = "End" block, also rejoins the control flow
+/// \endverbatim
+///
+/// Control flow is expressed as a branch where the true exit goes into the
+/// "Then"/"Else" region, while the false exit skips the region
+/// The condition for the optional "Else" region is expressed as a PHI node.
+/// The incomming values of the PHI node are true for the "If" edge and false
+/// for the "Then" edge.
+///
+/// Additionally to that even complicated loops look like this:
+///
+/// \verbatim
+/// 1
+/// ||
+/// | |
+/// 2 ^  Where:
+/// | /  1 = "Entry" block
+/// |/   2 = "Loop" optional subregion, with all exits at "Flow" block
+/// 3    3 = "Flow" block, with back edge to entry block
+/// |
+/// \endverbatim
+///
+/// The back edge of the "Flow" block is always on the false side of the branch
+/// while the true side continues the general flow. So the loop condition
+/// consist of a network of PHI nodes where the true incoming values expresses
+/// breaks and the false values expresses continue states.
+class AMDGPUStructurizeCFG : public RegionPass {
+
+  static char ID;
+
+  Type *Boolean;
+  ConstantInt *BoolTrue;
+  ConstantInt *BoolFalse;
+  UndefValue *BoolUndef;
+
+  Function *Func;
+  Region *ParentRegion;
+
+  DominatorTree *DT;
+
+  RNVector Order;
+  BBSet Visited;
+
+  BBPhiMap DeletedPhis;
+  BB2BBVecMap AddedPhis;
+
+  PredMap Predicates;
+  BranchVector Conditions;
+
+  BB2BBMap Loops;
+  PredMap LoopPreds;
+  BranchVector LoopConds;
+
+  RegionNode *PrevNode;
+
+  void orderNodes();
+
+  void analyzeLoops(RegionNode *N);
+
+  Value *invert(Value *Condition);
+
+  Value *buildCondition(BranchInst *Term, unsigned Idx, bool Invert);
+
+  void gatherPredicates(RegionNode *N);
+
+  void collectInfos();
+
+  void insertConditions(bool Loops);
+
+  void delPhiValues(BasicBlock *From, BasicBlock *To);
+
+  void addPhiValues(BasicBlock *From, BasicBlock *To);
+
+  void setPhiValues();
+
+  void killTerminator(BasicBlock *BB);
+
+  void changeExit(RegionNode *Node, BasicBlock *NewExit,
+                  bool IncludeDominator);
+
+  BasicBlock *getNextFlow(BasicBlock *Dominator);
+
+  BasicBlock *needPrefix(bool NeedEmpty);
+
+  BasicBlock *needPostfix(BasicBlock *Flow, bool ExitUseAllowed);
+
+  void setPrevNode(BasicBlock *BB);
+
+  bool dominatesPredicates(BasicBlock *BB, RegionNode *Node);
+
+  bool isPredictableTrue(RegionNode *Node);
+
+  void wireFlow(bool ExitUseAllowed, BasicBlock *LoopEnd);
+
+  void handleLoops(bool ExitUseAllowed, BasicBlock *LoopEnd);
+
+  void createFlow();
+
+  void rebuildSSA();
+
+public:
+  AMDGPUStructurizeCFG():
+    RegionPass(ID) {
+
+    initializeRegionInfoPass(*PassRegistry::getPassRegistry());
+  }
+
+  using Pass::doInitialization;
+  virtual bool doInitialization(Region *R, RGPassManager &RGM);
+
+  virtual bool runOnRegion(Region *R, RGPassManager &RGM);
+
+  virtual const char *getPassName() const {
+    return "AMDGPU simplify control flow";
+  }
+
+  void getAnalysisUsage(AnalysisUsage &AU) const {
+
+    AU.addRequired<DominatorTree>();
+    AU.addPreserved<DominatorTree>();
+    RegionPass::getAnalysisUsage(AU);
+  }
+
+};
+
+} // end anonymous namespace
+
+char AMDGPUStructurizeCFG::ID = 0;
+
+/// \brief Initialize the types and constants used in the pass
+bool AMDGPUStructurizeCFG::doInitialization(Region *R, RGPassManager &RGM) {
+  LLVMContext &Context = R->getEntry()->getContext();
+
+  Boolean = Type::getInt1Ty(Context);
+  BoolTrue = ConstantInt::getTrue(Context);
+  BoolFalse = ConstantInt::getFalse(Context);
+  BoolUndef = UndefValue::get(Boolean);
+
+  return false;
+}
+
+/// \brief Build up the general order of nodes
+void AMDGPUStructurizeCFG::orderNodes() {
+  scc_iterator<Region *> I = scc_begin(ParentRegion),
+                         E = scc_end(ParentRegion);
+  for (Order.clear(); I != E; ++I) {
+    std::vector<RegionNode *> &Nodes = *I;
+    Order.append(Nodes.begin(), Nodes.end());
+  }
+}
+
+/// \brief Determine the end of the loops
+void AMDGPUStructurizeCFG::analyzeLoops(RegionNode *N) {
+
+  if (N->isSubRegion()) {
+    // Test for exit as back edge
+    BasicBlock *Exit = N->getNodeAs<Region>()->getExit();
+    if (Visited.count(Exit))
+      Loops[Exit] = N->getEntry();
+
+  } else {
+    // Test for sucessors as back edge
+    BasicBlock *BB = N->getNodeAs<BasicBlock>();
+    BranchInst *Term = cast<BranchInst>(BB->getTerminator());
+
+    for (unsigned i = 0, e = Term->getNumSuccessors(); i != e; ++i) {
+      BasicBlock *Succ = Term->getSuccessor(i);
+
+      if (Visited.count(Succ))
+        Loops[Succ] = BB;
+    }
+  }
+}
+
+/// \brief Invert the given condition
+Value *AMDGPUStructurizeCFG::invert(Value *Condition) {
+
+  // First: Check if it's a constant
+  if (Condition == BoolTrue)
+    return BoolFalse;
+
+  if (Condition == BoolFalse)
+    return BoolTrue;
+
+  if (Condition == BoolUndef)
+    return BoolUndef;
+
+  // Second: If the condition is already inverted, return the original value
+  if (match(Condition, m_Not(m_Value(Condition))))
+    return Condition;
+
+  // Third: Check all the users for an invert
+  BasicBlock *Parent = cast<Instruction>(Condition)->getParent();
+  for (Value::use_iterator I = Condition->use_begin(),
+       E = Condition->use_end(); I != E; ++I) {
+
+    Instruction *User = dyn_cast<Instruction>(*I);
+    if (!User || User->getParent() != Parent)
+      continue;
+
+    if (match(*I, m_Not(m_Specific(Condition))))
+      return *I;
+  }
+
+  // Last option: Create a new instruction
+  return BinaryOperator::CreateNot(Condition, "", Parent->getTerminator());
+}
+
+/// \brief Build the condition for one edge
+Value *AMDGPUStructurizeCFG::buildCondition(BranchInst *Term, unsigned Idx,
+                                            bool Invert) {
+  Value *Cond = Invert ? BoolFalse : BoolTrue;
+  if (Term->isConditional()) {
+    Cond = Term->getCondition();
+
+    if (Idx != Invert)
+      Cond = invert(Cond);
+  }
+  return Cond;
+}
+
+/// \brief Analyze the predecessors of each block and build up predicates
+void AMDGPUStructurizeCFG::gatherPredicates(RegionNode *N) {
+
+  RegionInfo *RI = ParentRegion->getRegionInfo();
+  BasicBlock *BB = N->getEntry();
+  BBPredicates &Pred = Predicates[BB];
+  BBPredicates &LPred = LoopPreds[BB];
+
+  for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB);
+       PI != PE; ++PI) {
+
+    // Ignore it if it's a branch from outside into our region entry
+    if (!ParentRegion->contains(*PI))
+      continue;
+
+    Region *R = RI->getRegionFor(*PI);
+    if (R == ParentRegion) {
+
+      // It's a top level block in our region
+      BranchInst *Term = cast<BranchInst>((*PI)->getTerminator());
+      for (unsigned i = 0, e = Term->getNumSuccessors(); i != e; ++i) {
+        BasicBlock *Succ = Term->getSuccessor(i);
+        if (Succ != BB)
+          continue;
+
+        if (Visited.count(*PI)) {
+          // Normal forward edge
+          if (Term->isConditional()) {
+            // Try to treat it like an ELSE block
+            BasicBlock *Other = Term->getSuccessor(!i);
+            if (Visited.count(Other) && !Loops.count(Other) &&
+                !Pred.count(Other) && !Pred.count(*PI)) {
+
+              Pred[Other] = BoolFalse;
+              Pred[*PI] = BoolTrue;
+              continue;
+            }
+          }
+          Pred[*PI] = buildCondition(Term, i, false);
+ 
+        } else {
+          // Back edge
+          LPred[*PI] = buildCondition(Term, i, true);
+        }
+      }
+
+    } else {
+
+      // It's an exit from a sub region
+      while(R->getParent() != ParentRegion)
+        R = R->getParent();
+
+      // Edge from inside a subregion to its entry, ignore it
+      if (R == N)
+        continue;
+
+      BasicBlock *Entry = R->getEntry();
+      if (Visited.count(Entry))
+        Pred[Entry] = BoolTrue;
+      else
+        LPred[Entry] = BoolFalse;
+    }
+  }
+}
+
+/// \brief Collect various loop and predicate infos
+void AMDGPUStructurizeCFG::collectInfos() {
+
+  // Reset predicate
+  Predicates.clear();
+
+  // and loop infos
+  Loops.clear();
+  LoopPreds.clear();
+
+  // Reset the visited nodes
+  Visited.clear();
+
+  for (RNVector::reverse_iterator OI = Order.rbegin(), OE = Order.rend();
+       OI != OE; ++OI) {
+
+    // Analyze all the conditions leading to a node
+    gatherPredicates(*OI);
+
+    // Remember that we've seen this node
+    Visited.insert((*OI)->getEntry());
+
+    // Find the last back edges
+    analyzeLoops(*OI);
+  }
+}
+
+/// \brief Insert the missing branch conditions
+void AMDGPUStructurizeCFG::insertConditions(bool Loops) {
+  BranchVector &Conds = Loops ? LoopConds : Conditions;
+  Value *Default = Loops ? BoolTrue : BoolFalse;
+  SSAUpdater PhiInserter;
+
+  for (BranchVector::iterator I = Conds.begin(),
+       E = Conds.end(); I != E; ++I) {
+
+    BranchInst *Term = *I;
+    assert(Term->isConditional());
+
+    BasicBlock *Parent = Term->getParent();
+    BasicBlock *SuccTrue = Term->getSuccessor(0);
+    BasicBlock *SuccFalse = Term->getSuccessor(1);
+
+    PhiInserter.Initialize(Boolean, "");
+    PhiInserter.AddAvailableValue(&Func->getEntryBlock(), Default);
+    PhiInserter.AddAvailableValue(Loops ? SuccFalse : Parent, Default);
+
+    BBPredicates &Preds = Loops ? LoopPreds[SuccFalse] : Predicates[SuccTrue];
+
+    NearestCommonDominator Dominator(DT);
+    Dominator.addBlock(Parent, false);
+
+    Value *ParentValue = 0;
+    for (BBPredicates::iterator PI = Preds.begin(), PE = Preds.end();
+         PI != PE; ++PI) {
+
+      if (PI->first == Parent) {
+        ParentValue = PI->second;
+        break;
+      }
+      PhiInserter.AddAvailableValue(PI->first, PI->second);
+      Dominator.addBlock(PI->first);
+    }
+
+    if (ParentValue) {
+      Term->setCondition(ParentValue);
+    } else {
+      if (!Dominator.wasResultExplicitMentioned())
+        PhiInserter.AddAvailableValue(Dominator.getResult(), Default);
+
+      Term->setCondition(PhiInserter.GetValueInMiddleOfBlock(Parent));
+    }
+  }
+}
+
+/// \brief Remove all PHI values coming from "From" into "To" and remember
+/// them in DeletedPhis
+void AMDGPUStructurizeCFG::delPhiValues(BasicBlock *From, BasicBlock *To) {
+  PhiMap &Map = DeletedPhis[To];
+  for (BasicBlock::iterator I = To->begin(), E = To->end();
+       I != E && isa<PHINode>(*I);) {
+
+    PHINode &Phi = cast<PHINode>(*I++);
+    while (Phi.getBasicBlockIndex(From) != -1) {
+      Value *Deleted = Phi.removeIncomingValue(From, false);
+      Map[&Phi].push_back(std::make_pair(From, Deleted));
+    }
+  }
+}
+
+/// \brief Add a dummy PHI value as soon as we knew the new predecessor
+void AMDGPUStructurizeCFG::addPhiValues(BasicBlock *From, BasicBlock *To) {
+  for (BasicBlock::iterator I = To->begin(), E = To->end();
+       I != E && isa<PHINode>(*I);) {
+
+    PHINode &Phi = cast<PHINode>(*I++);
+    Value *Undef = UndefValue::get(Phi.getType());
+    Phi.addIncoming(Undef, From);
+  }
+  AddedPhis[To].push_back(From);
+}
+
+/// \brief Add the real PHI value as soon as everything is set up
+void AMDGPUStructurizeCFG::setPhiValues() {
+
+  SSAUpdater Updater;
+  for (BB2BBVecMap::iterator AI = AddedPhis.begin(), AE = AddedPhis.end();
+       AI != AE; ++AI) {
+
+    BasicBlock *To = AI->first;
+    BBVector &From = AI->second;
+
+    if (!DeletedPhis.count(To))
+      continue;
+
+    PhiMap &Map = DeletedPhis[To];
+    for (PhiMap::iterator PI = Map.begin(), PE = Map.end();
+         PI != PE; ++PI) {
+
+      PHINode *Phi = PI->first;
+      Value *Undef = UndefValue::get(Phi->getType());
+      Updater.Initialize(Phi->getType(), "");
+      Updater.AddAvailableValue(&Func->getEntryBlock(), Undef);
+      Updater.AddAvailableValue(To, Undef);
+
+      NearestCommonDominator Dominator(DT);
+      Dominator.addBlock(To, false);
+      for (BBValueVector::iterator VI = PI->second.begin(),
+           VE = PI->second.end(); VI != VE; ++VI) {
+
+        Updater.AddAvailableValue(VI->first, VI->second);
+        Dominator.addBlock(VI->first);
+      }
+
+      if (!Dominator.wasResultExplicitMentioned())
+        Updater.AddAvailableValue(Dominator.getResult(), Undef);
+
+      for (BBVector::iterator FI = From.begin(), FE = From.end();
+           FI != FE; ++FI) {
+
+        int Idx = Phi->getBasicBlockIndex(*FI);
+        assert(Idx != -1);
+        Phi->setIncomingValue(Idx, Updater.GetValueAtEndOfBlock(*FI));
+      }
+    }
+
+    DeletedPhis.erase(To);
+  }
+  assert(DeletedPhis.empty());
+}
+
+/// \brief Remove phi values from all successors and then remove the terminator.
+void AMDGPUStructurizeCFG::killTerminator(BasicBlock *BB) {
+  TerminatorInst *Term = BB->getTerminator();
+  if (!Term)
+    return;
+
+  for (succ_iterator SI = succ_begin(BB), SE = succ_end(BB);
+       SI != SE; ++SI) {
+
+    delPhiValues(BB, *SI);
+  }
+
+  Term->eraseFromParent();
+}
+
+/// \brief Let node exit(s) point to NewExit
+void AMDGPUStructurizeCFG::changeExit(RegionNode *Node, BasicBlock *NewExit,
+                                      bool IncludeDominator) {
+
+  if (Node->isSubRegion()) {
+    Region *SubRegion = Node->getNodeAs<Region>();
+    BasicBlock *OldExit = SubRegion->getExit();
+    BasicBlock *Dominator = 0;
+
+    // Find all the edges from the sub region to the exit
+    for (pred_iterator I = pred_begin(OldExit), E = pred_end(OldExit);
+         I != E;) {
+
+      BasicBlock *BB = *I++;
+      if (!SubRegion->contains(BB))
+        continue;
+
+      // Modify the edges to point to the new exit
+      delPhiValues(BB, OldExit);
+      BB->getTerminator()->replaceUsesOfWith(OldExit, NewExit);
+      addPhiValues(BB, NewExit);
+
+      // Find the new dominator (if requested)
+      if (IncludeDominator) {
+        if (!Dominator)
+          Dominator = BB;
+        else
+          Dominator = DT->findNearestCommonDominator(Dominator, BB);
+      }
+    }
+
+    // Change the dominator (if requested)
+    if (Dominator)
+      DT->changeImmediateDominator(NewExit, Dominator);
+
+    // Update the region info
+    SubRegion->replaceExit(NewExit);
+
+  } else {
+    BasicBlock *BB = Node->getNodeAs<BasicBlock>();
+    killTerminator(BB);
+    BranchInst::Create(NewExit, BB);
+    addPhiValues(BB, NewExit);
+    if (IncludeDominator)
+      DT->changeImmediateDominator(NewExit, BB);
+  }
+}
+
+/// \brief Create a new flow node and update dominator tree and region info
+BasicBlock *AMDGPUStructurizeCFG::getNextFlow(BasicBlock *Dominator) {
+  LLVMContext &Context = Func->getContext();
+  BasicBlock *Insert = Order.empty() ? ParentRegion->getExit() :
+                       Order.back()->getEntry();
+  BasicBlock *Flow = BasicBlock::Create(Context, FlowBlockName,
+                                        Func, Insert);
+  DT->addNewBlock(Flow, Dominator);
+  ParentRegion->getRegionInfo()->setRegionFor(Flow, ParentRegion);
+  return Flow;
+}
+
+/// \brief Create a new or reuse the previous node as flow node
+BasicBlock *AMDGPUStructurizeCFG::needPrefix(bool NeedEmpty) {
+
+  BasicBlock *Entry = PrevNode->getEntry();
+
+  if (!PrevNode->isSubRegion()) {
+    killTerminator(Entry);
+    if (!NeedEmpty || Entry->getFirstInsertionPt() == Entry->end())
+      return Entry;
+
+  } 
+
+  // create a new flow node
+  BasicBlock *Flow = getNextFlow(Entry);
+
+  // and wire it up
+  changeExit(PrevNode, Flow, true);
+  PrevNode = ParentRegion->getBBNode(Flow);
+  return Flow;
+}
+
+/// \brief Returns the region exit if possible, otherwise just a new flow node
+BasicBlock *AMDGPUStructurizeCFG::needPostfix(BasicBlock *Flow,
+                                              bool ExitUseAllowed) {
+
+  if (Order.empty() && ExitUseAllowed) {
+    BasicBlock *Exit = ParentRegion->getExit();
+    DT->changeImmediateDominator(Exit, Flow);
+    addPhiValues(Flow, Exit);
+    return Exit;
+  }
+  return getNextFlow(Flow);
+}
+
+/// \brief Set the previous node
+void AMDGPUStructurizeCFG::setPrevNode(BasicBlock *BB) {
+  PrevNode =  ParentRegion->contains(BB) ? ParentRegion->getBBNode(BB) : 0;
+}
+
+/// \brief Does BB dominate all the predicates of Node ?
+bool AMDGPUStructurizeCFG::dominatesPredicates(BasicBlock *BB, RegionNode *Node) {
+  BBPredicates &Preds = Predicates[Node->getEntry()];
+  for (BBPredicates::iterator PI = Preds.begin(), PE = Preds.end();
+       PI != PE; ++PI) {
+
+    if (!DT->dominates(BB, PI->first))
+      return false;
+  }
+  return true;
+}
+
+/// \brief Can we predict that this node will always be called?
+bool AMDGPUStructurizeCFG::isPredictableTrue(RegionNode *Node) {
+
+  BBPredicates &Preds = Predicates[Node->getEntry()];
+  bool Dominated = false;
+
+  // Regionentry is always true
+  if (PrevNode == 0)
+    return true;
+
+  for (BBPredicates::iterator I = Preds.begin(), E = Preds.end();
+       I != E; ++I) {
+
+    if (I->second != BoolTrue)
+      return false;
+
+    if (!Dominated && DT->dominates(I->first, PrevNode->getEntry()))
+      Dominated = true;
+  }
+
+  // TODO: The dominator check is too strict
+  return Dominated;
+}
+
+/// Take one node from the order vector and wire it up
+void AMDGPUStructurizeCFG::wireFlow(bool ExitUseAllowed,
+                                    BasicBlock *LoopEnd) {
+
+  RegionNode *Node = Order.pop_back_val();
+  Visited.insert(Node->getEntry());
+
+  if (isPredictableTrue(Node)) {
+    // Just a linear flow
+    if (PrevNode) {
+      changeExit(PrevNode, Node->getEntry(), true);
+    }
+    PrevNode = Node;
+
+  } else {
+    // Insert extra prefix node (or reuse last one)
+    BasicBlock *Flow = needPrefix(false);
+
+    // Insert extra postfix node (or use exit instead)
+    BasicBlock *Entry = Node->getEntry();
+    BasicBlock *Next = needPostfix(Flow, ExitUseAllowed);
+
+    // let it point to entry and next block
+    Conditions.push_back(BranchInst::Create(Entry, Next, BoolUndef, Flow));
+    addPhiValues(Flow, Entry);
+    DT->changeImmediateDominator(Entry, Flow);
+
+    PrevNode = Node;
+    while (!Order.empty() && !Visited.count(LoopEnd) &&
+           dominatesPredicates(Entry, Order.back())) {
+      handleLoops(false, LoopEnd);
+    }
+
+    changeExit(PrevNode, Next, false);
+    setPrevNode(Next);
+  }
+}
+
+void AMDGPUStructurizeCFG::handleLoops(bool ExitUseAllowed,
+                                       BasicBlock *LoopEnd) {
+  RegionNode *Node = Order.back();
+  BasicBlock *LoopStart = Node->getEntry();
+
+  if (!Loops.count(LoopStart)) {
+    wireFlow(ExitUseAllowed, LoopEnd);
+    return;
+  }
+
+  if (!isPredictableTrue(Node))
+    LoopStart = needPrefix(true);
+
+  LoopEnd = Loops[Node->getEntry()];
+  wireFlow(false, LoopEnd);
+  while (!Visited.count(LoopEnd)) {
+    handleLoops(false, LoopEnd);
+  }
+
+  // Create an extra loop end node
+  LoopEnd = needPrefix(false);
+  BasicBlock *Next = needPostfix(LoopEnd, ExitUseAllowed);
+  LoopConds.push_back(BranchInst::Create(Next, LoopStart,
+                                         BoolUndef, LoopEnd));
+  addPhiValues(LoopEnd, LoopStart);
+  setPrevNode(Next);
+}
+
+/// After this function control flow looks like it should be, but
+/// branches and PHI nodes only have undefined conditions.
+void AMDGPUStructurizeCFG::createFlow() {
+
+  BasicBlock *Exit = ParentRegion->getExit();
+  bool EntryDominatesExit = DT->dominates(ParentRegion->getEntry(), Exit);
+
+  DeletedPhis.clear();
+  AddedPhis.clear();
+  Conditions.clear();
+  LoopConds.clear();
+
+  PrevNode = 0;
+  Visited.clear();
+
+  while (!Order.empty()) {
+    handleLoops(EntryDominatesExit, 0);
+  }
+
+  if (PrevNode)
+    changeExit(PrevNode, Exit, EntryDominatesExit);
+  else
+    assert(EntryDominatesExit);
+}
+
+/// Handle a rare case where the disintegrated nodes instructions
+/// no longer dominate all their uses. Not sure if this is really nessasary
+void AMDGPUStructurizeCFG::rebuildSSA() {
+  SSAUpdater Updater;
+  for (Region::block_iterator I = ParentRegion->block_begin(),
+                              E = ParentRegion->block_end();
+       I != E; ++I) {
+
+    BasicBlock *BB = *I;
+    for (BasicBlock::iterator II = BB->begin(), IE = BB->end();
+         II != IE; ++II) {
+
+      bool Initialized = false;
+      for (Use *I = &II->use_begin().getUse(), *Next; I; I = Next) {
+
+        Next = I->getNext();
+
+        Instruction *User = cast<Instruction>(I->getUser());
+        if (User->getParent() == BB) {
+          continue;
+
+        } else if (PHINode *UserPN = dyn_cast<PHINode>(User)) {
+          if (UserPN->getIncomingBlock(*I) == BB)
+            continue;
+        }
+
+        if (DT->dominates(II, User))
+          continue;
+
+        if (!Initialized) {
+          Value *Undef = UndefValue::get(II->getType());
+          Updater.Initialize(II->getType(), "");
+          Updater.AddAvailableValue(&Func->getEntryBlock(), Undef);
+          Updater.AddAvailableValue(BB, II);
+          Initialized = true;
+        }
+        Updater.RewriteUseAfterInsertions(*I);
+      }
+    }
+  }
+}
+
+/// \brief Run the transformation for each region found
+bool AMDGPUStructurizeCFG::runOnRegion(Region *R, RGPassManager &RGM) {
+  if (R->isTopLevelRegion())
+    return false;
+
+  Func = R->getEntry()->getParent();
+  ParentRegion = R;
+
+  DT = &getAnalysis<DominatorTree>();
+
+  orderNodes();
+  collectInfos();
+  createFlow();
+  insertConditions(false);
+  insertConditions(true);
+  setPhiValues();
+  rebuildSSA();
+
+  // Cleanup
+  Order.clear();
+  Visited.clear();
+  DeletedPhis.clear();
+  AddedPhis.clear();
+  Predicates.clear();
+  Conditions.clear();
+  Loops.clear();
+  LoopPreds.clear();
+  LoopConds.clear();
+
+  return true;
+}
+
+/// \brief Create the pass
+Pass *llvm::createAMDGPUStructurizeCFGPass() {
+  return new AMDGPUStructurizeCFG();
+}
diff --git a/lib/Target/R600/AMDGPUSubtarget.cpp b/lib/Target/R600/AMDGPUSubtarget.cpp
new file mode 100644
index 000000000000..0f356a1c3f11
--- /dev/null
+++ b/lib/Target/R600/AMDGPUSubtarget.cpp
@@ -0,0 +1,87 @@
+//===-- AMDGPUSubtarget.cpp - AMDGPU Subtarget Information ----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Implements the AMDGPU specific subclass of TargetSubtarget.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUSubtarget.h"
+
+using namespace llvm;
+
+#define GET_SUBTARGETINFO_ENUM
+#define GET_SUBTARGETINFO_TARGET_DESC
+#define GET_SUBTARGETINFO_CTOR
+#include "AMDGPUGenSubtargetInfo.inc"
+
+AMDGPUSubtarget::AMDGPUSubtarget(StringRef TT, StringRef CPU, StringRef FS) :
+  AMDGPUGenSubtargetInfo(TT, CPU, FS), DumpCode(false) {
+    InstrItins = getInstrItineraryForCPU(CPU);
+
+  memset(CapsOverride, 0, sizeof(*CapsOverride)
+      * AMDGPUDeviceInfo::MaxNumberCapabilities);
+  // Default card
+  StringRef GPU = CPU;
+  Is64bit = false;
+  DefaultSize[0] = 64;
+  DefaultSize[1] = 1;
+  DefaultSize[2] = 1;
+  ParseSubtargetFeatures(GPU, FS);
+  DevName = GPU;
+  Device = AMDGPUDeviceInfo::getDeviceFromName(DevName, this, Is64bit);
+}
+
+AMDGPUSubtarget::~AMDGPUSubtarget() {
+  delete Device;
+}
+
+bool
+AMDGPUSubtarget::isOverride(AMDGPUDeviceInfo::Caps caps) const {
+  assert(caps < AMDGPUDeviceInfo::MaxNumberCapabilities &&
+      "Caps index is out of bounds!");
+  return CapsOverride[caps];
+}
+bool
+AMDGPUSubtarget::is64bit() const  {
+  return Is64bit;
+}
+bool
+AMDGPUSubtarget::isTargetELF() const {
+  return false;
+}
+size_t
+AMDGPUSubtarget::getDefaultSize(uint32_t dim) const {
+  if (dim > 3) {
+    return 1;
+  } else {
+    return DefaultSize[dim];
+  }
+}
+
+std::string
+AMDGPUSubtarget::getDataLayout() const {
+    if (!Device) {
+        return std::string("e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16"
+                "-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f80:32:32"
+                "-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64"
+                "-v96:128:128-v128:128:128-v192:256:256-v256:256:256"
+                "-v512:512:512-v1024:1024:1024-v2048:2048:2048-a0:0:64");
+    }
+    return Device->getDataLayout();
+}
+
+std::string
+AMDGPUSubtarget::getDeviceName() const {
+  return DevName;
+}
+const AMDGPUDevice *
+AMDGPUSubtarget::device() const {
+  return Device;
+}
diff --git a/lib/Target/R600/AMDGPUSubtarget.h b/lib/Target/R600/AMDGPUSubtarget.h
new file mode 100644
index 000000000000..1973fc6d544c
--- /dev/null
+++ b/lib/Target/R600/AMDGPUSubtarget.h
@@ -0,0 +1,65 @@
+//=====-- AMDGPUSubtarget.h - Define Subtarget for the AMDIL ---*- C++ -*-====//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief AMDGPU specific subclass of TargetSubtarget.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPUSUBTARGET_H
+#define AMDGPUSUBTARGET_H
+#include "AMDILDevice.h"
+#include "llvm/ADT/StringExtras.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
+
+#define GET_SUBTARGETINFO_HEADER
+#include "AMDGPUGenSubtargetInfo.inc"
+
+#define MAX_CB_SIZE (1 << 16)
+
+namespace llvm {
+
+class AMDGPUSubtarget : public AMDGPUGenSubtargetInfo {
+private:
+  bool CapsOverride[AMDGPUDeviceInfo::MaxNumberCapabilities];
+  const AMDGPUDevice *Device;
+  size_t DefaultSize[3];
+  std::string DevName;
+  bool Is64bit;
+  bool Is32on64bit;
+  bool DumpCode;
+  bool R600ALUInst;
+
+  InstrItineraryData InstrItins;
+
+public:
+  AMDGPUSubtarget(StringRef TT, StringRef CPU, StringRef FS);
+  virtual ~AMDGPUSubtarget();
+
+  const InstrItineraryData &getInstrItineraryData() const { return InstrItins; }
+  virtual void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
+
+  bool isOverride(AMDGPUDeviceInfo::Caps) const;
+  bool is64bit() const;
+
+  // Helper functions to simplify if statements
+  bool isTargetELF() const;
+  const AMDGPUDevice* device() const;
+  std::string getDataLayout() const;
+  std::string getDeviceName() const;
+  virtual size_t getDefaultSize(uint32_t dim) const;
+  bool dumpCode() const { return DumpCode; }
+  bool r600ALUEncoding() const { return R600ALUInst; }
+
+};
+
+} // End namespace llvm
+
+#endif // AMDGPUSUBTARGET_H
diff --git a/lib/Target/R600/AMDGPUTargetMachine.cpp b/lib/Target/R600/AMDGPUTargetMachine.cpp
new file mode 100644
index 000000000000..e7ea876e2abb
--- /dev/null
+++ b/lib/Target/R600/AMDGPUTargetMachine.cpp
@@ -0,0 +1,164 @@
+//===-- AMDGPUTargetMachine.cpp - TargetMachine for hw codegen targets-----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief The AMDGPU target machine contains all of the hardware specific
+/// information  needed to emit code for R600 and SI GPUs.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUTargetMachine.h"
+#include "AMDGPU.h"
+#include "R600ISelLowering.h"
+#include "R600InstrInfo.h"
+#include "R600MachineScheduler.h"
+#include "SIISelLowering.h"
+#include "SIInstrInfo.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/Verifier.h"
+#include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/PassManager.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/raw_os_ostream.h"
+#include "llvm/Transforms/IPO.h"
+#include "llvm/Transforms/Scalar.h"
+#include <llvm/CodeGen/Passes.h>
+
+using namespace llvm;
+
+extern "C" void LLVMInitializeR600Target() {
+  // Register the target
+  RegisterTargetMachine<AMDGPUTargetMachine> X(TheAMDGPUTarget);
+}
+
+static ScheduleDAGInstrs *createR600MachineScheduler(MachineSchedContext *C) {
+  return new ScheduleDAGMI(C, new R600SchedStrategy());
+}
+
+static MachineSchedRegistry
+SchedCustomRegistry("r600", "Run R600's custom scheduler",
+                    createR600MachineScheduler);
+
+AMDGPUTargetMachine::AMDGPUTargetMachine(const Target &T, StringRef TT,
+    StringRef CPU, StringRef FS,
+  TargetOptions Options,
+  Reloc::Model RM, CodeModel::Model CM,
+  CodeGenOpt::Level OptLevel
+)
+:
+  LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OptLevel),
+  Subtarget(TT, CPU, FS),
+  Layout(Subtarget.getDataLayout()),
+  FrameLowering(TargetFrameLowering::StackGrowsUp,
+      Subtarget.device()->getStackAlignment(), 0),
+  IntrinsicInfo(this),
+  InstrItins(&Subtarget.getInstrItineraryData()) {
+  // TLInfo uses InstrInfo so it must be initialized after.
+  if (Subtarget.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
+    InstrInfo = new R600InstrInfo(*this);
+    TLInfo = new R600TargetLowering(*this);
+  } else {
+    InstrInfo = new SIInstrInfo(*this);
+    TLInfo = new SITargetLowering(*this);
+  }
+}
+
+AMDGPUTargetMachine::~AMDGPUTargetMachine() {
+}
+
+namespace {
+class AMDGPUPassConfig : public TargetPassConfig {
+public:
+  AMDGPUPassConfig(AMDGPUTargetMachine *TM, PassManagerBase &PM)
+    : TargetPassConfig(TM, PM) {
+    const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+    if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
+      enablePass(&MachineSchedulerID);
+      MachineSchedRegistry::setDefault(createR600MachineScheduler);
+    }
+  }
+
+  AMDGPUTargetMachine &getAMDGPUTargetMachine() const {
+    return getTM<AMDGPUTargetMachine>();
+  }
+
+  virtual bool addPreISel();
+  virtual bool addInstSelector();
+  virtual bool addPreRegAlloc();
+  virtual bool addPostRegAlloc();
+  virtual bool addPreSched2();
+  virtual bool addPreEmitPass();
+};
+} // End of anonymous namespace
+
+TargetPassConfig *AMDGPUTargetMachine::createPassConfig(PassManagerBase &PM) {
+  return new AMDGPUPassConfig(this, PM);
+}
+
+bool
+AMDGPUPassConfig::addPreISel() {
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+  if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
+    addPass(createAMDGPUStructurizeCFGPass());
+    addPass(createSIAnnotateControlFlowPass());
+  }
+  return false;
+}
+
+bool AMDGPUPassConfig::addInstSelector() {
+  addPass(createAMDGPUPeepholeOpt(*TM));
+  addPass(createAMDGPUISelDag(getAMDGPUTargetMachine()));
+
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+  if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
+    // This callbacks this pass uses are not implemented yet on SI.
+    addPass(createAMDGPUIndirectAddressingPass(*TM));
+  }
+  return false;
+}
+
+bool AMDGPUPassConfig::addPreRegAlloc() {
+  addPass(createAMDGPUConvertToISAPass(*TM));
+  return false;
+}
+
+bool AMDGPUPassConfig::addPostRegAlloc() {
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+
+  if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
+    addPass(createSIInsertWaits(*TM));
+  }
+  return false;
+}
+
+bool AMDGPUPassConfig::addPreSched2() {
+
+  addPass(&IfConverterID);
+  return false;
+}
+
+bool AMDGPUPassConfig::addPreEmitPass() {
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+  if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
+    addPass(createAMDGPUCFGPreparationPass(*TM));
+    addPass(createAMDGPUCFGStructurizerPass(*TM));
+    addPass(createR600EmitClauseMarkers(*TM));
+    addPass(createR600ExpandSpecialInstrsPass(*TM));
+    addPass(createR600ControlFlowFinalizer(*TM));
+    addPass(&FinalizeMachineBundlesID);
+  } else {
+    addPass(createSILowerControlFlowPass(*TM));
+  }
+
+  return false;
+}
+
diff --git a/lib/Target/R600/AMDGPUTargetMachine.h b/lib/Target/R600/AMDGPUTargetMachine.h
new file mode 100644
index 000000000000..2afe7873a90c
--- /dev/null
+++ b/lib/Target/R600/AMDGPUTargetMachine.h
@@ -0,0 +1,70 @@
+//===-- AMDGPUTargetMachine.h - AMDGPU TargetMachine Interface --*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief The AMDGPU TargetMachine interface definition for hw codgen targets.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPU_TARGET_MACHINE_H
+#define AMDGPU_TARGET_MACHINE_H
+
+#include "AMDGPUFrameLowering.h"
+#include "AMDGPUInstrInfo.h"
+#include "AMDGPUSubtarget.h"
+#include "AMDILIntrinsicInfo.h"
+#include "R600ISelLowering.h"
+#include "llvm/ADT/OwningPtr.h"
+#include "llvm/IR/DataLayout.h"
+
+namespace llvm {
+
+MCAsmInfo* createMCAsmInfo(const Target &T, StringRef TT);
+
+class AMDGPUTargetMachine : public LLVMTargetMachine {
+
+  AMDGPUSubtarget Subtarget;
+  const DataLayout Layout;
+  AMDGPUFrameLowering FrameLowering;
+  AMDGPUIntrinsicInfo IntrinsicInfo;
+  const AMDGPUInstrInfo * InstrInfo;
+  AMDGPUTargetLowering * TLInfo;
+  const InstrItineraryData* InstrItins;
+
+public:
+   AMDGPUTargetMachine(const Target &T, StringRef TT, StringRef FS,
+                       StringRef CPU,
+                       TargetOptions Options,
+                       Reloc::Model RM, CodeModel::Model CM,
+                       CodeGenOpt::Level OL);
+   ~AMDGPUTargetMachine();
+   virtual const AMDGPUFrameLowering* getFrameLowering() const {
+     return &FrameLowering;
+   }
+   virtual const AMDGPUIntrinsicInfo* getIntrinsicInfo() const {
+     return &IntrinsicInfo;
+   }
+   virtual const AMDGPUInstrInfo *getInstrInfo() const {return InstrInfo;}
+   virtual const AMDGPUSubtarget *getSubtargetImpl() const {return &Subtarget; }
+   virtual const AMDGPURegisterInfo *getRegisterInfo() const {
+      return &InstrInfo->getRegisterInfo();
+   }
+   virtual AMDGPUTargetLowering * getTargetLowering() const {
+      return TLInfo;
+   }
+   virtual const InstrItineraryData* getInstrItineraryData() const {
+      return InstrItins;
+   }
+   virtual const DataLayout* getDataLayout() const { return &Layout; }
+   virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
+};
+
+} // End namespace llvm
+
+#endif // AMDGPU_TARGET_MACHINE_H
diff --git a/lib/Target/R600/AMDIL.h b/lib/Target/R600/AMDIL.h
new file mode 100644
index 000000000000..39ab664d1018
--- /dev/null
+++ b/lib/Target/R600/AMDIL.h
@@ -0,0 +1,121 @@
+//===-- AMDIL.h - Top-level interface for AMDIL representation --*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+/// This file contains the entry points for global functions defined in the LLVM
+/// AMDGPU back-end.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDIL_H
+#define AMDIL_H
+
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/Target/TargetMachine.h"
+
+#define ARENA_SEGMENT_RESERVED_UAVS 12
+#define DEFAULT_ARENA_UAV_ID 8
+#define DEFAULT_RAW_UAV_ID 7
+#define GLOBAL_RETURN_RAW_UAV_ID 11
+#define HW_MAX_NUM_CB 8
+#define MAX_NUM_UNIQUE_UAVS 8
+#define OPENCL_MAX_NUM_ATOMIC_COUNTERS 8
+#define OPENCL_MAX_READ_IMAGES 128
+#define OPENCL_MAX_WRITE_IMAGES 8
+#define OPENCL_MAX_SAMPLERS 16
+
+// The next two values can never be zero, as zero is the ID that is
+// used to assert against.
+#define DEFAULT_LDS_ID     1
+#define DEFAULT_GDS_ID     1
+#define DEFAULT_SCRATCH_ID 1
+#define DEFAULT_VEC_SLOTS  8
+
+#define OCL_DEVICE_RV710        0x0001
+#define OCL_DEVICE_RV730        0x0002
+#define OCL_DEVICE_RV770        0x0004
+#define OCL_DEVICE_CEDAR        0x0008
+#define OCL_DEVICE_REDWOOD      0x0010
+#define OCL_DEVICE_JUNIPER      0x0020
+#define OCL_DEVICE_CYPRESS      0x0040
+#define OCL_DEVICE_CAICOS       0x0080
+#define OCL_DEVICE_TURKS        0x0100
+#define OCL_DEVICE_BARTS        0x0200
+#define OCL_DEVICE_CAYMAN       0x0400
+#define OCL_DEVICE_ALL          0x3FFF
+
+/// The number of function ID's that are reserved for 
+/// internal compiler usage.
+const unsigned int RESERVED_FUNCS = 1024;
+
+namespace llvm {
+class AMDGPUInstrPrinter;
+class FunctionPass;
+class MCAsmInfo;
+class raw_ostream;
+class Target;
+class TargetMachine;
+
+// Instruction selection passes.
+FunctionPass*
+  createAMDGPUISelDag(TargetMachine &TM);
+FunctionPass*
+  createAMDGPUPeepholeOpt(TargetMachine &TM);
+
+// Pre emit passes.
+FunctionPass*
+  createAMDGPUCFGPreparationPass(TargetMachine &TM);
+FunctionPass*
+  createAMDGPUCFGStructurizerPass(TargetMachine &TM);
+
+extern Target TheAMDGPUTarget;
+} // end namespace llvm;
+
+// Include device information enumerations
+#include "AMDILDeviceInfo.h"
+
+namespace llvm {
+/// OpenCL uses address spaces to differentiate between
+/// various memory regions on the hardware. On the CPU
+/// all of the address spaces point to the same memory,
+/// however on the GPU, each address space points to
+/// a seperate piece of memory that is unique from other
+/// memory locations.
+namespace AMDGPUAS {
+enum AddressSpaces {
+  PRIVATE_ADDRESS  = 0, ///< Address space for private memory.
+  GLOBAL_ADDRESS   = 1, ///< Address space for global memory (RAT0, VTX0).
+  CONSTANT_ADDRESS = 2, ///< Address space for constant memory
+  LOCAL_ADDRESS    = 3, ///< Address space for local memory.
+  REGION_ADDRESS   = 4, ///< Address space for region memory.
+  ADDRESS_NONE     = 5, ///< Address space for unknown memory.
+  PARAM_D_ADDRESS  = 6, ///< Address space for direct addressible parameter memory (CONST0)
+  PARAM_I_ADDRESS  = 7, ///< Address space for indirect addressible parameter memory (VTX1)
+  CONSTANT_BUFFER_0 = 8,
+  CONSTANT_BUFFER_1 = 9,
+  CONSTANT_BUFFER_2 = 10,
+  CONSTANT_BUFFER_3 = 11,
+  CONSTANT_BUFFER_4 = 12,
+  CONSTANT_BUFFER_5 = 13,
+  CONSTANT_BUFFER_6 = 14,
+  CONSTANT_BUFFER_7 = 15,
+  CONSTANT_BUFFER_8 = 16,
+  CONSTANT_BUFFER_9 = 17,
+  CONSTANT_BUFFER_10 = 18,
+  CONSTANT_BUFFER_11 = 19,
+  CONSTANT_BUFFER_12 = 20,
+  CONSTANT_BUFFER_13 = 21,
+  CONSTANT_BUFFER_14 = 22,
+  CONSTANT_BUFFER_15 = 23,
+  LAST_ADDRESS     = 24
+};
+
+} // namespace AMDGPUAS
+
+} // end namespace llvm
+#endif // AMDIL_H
diff --git a/lib/Target/R600/AMDIL7XXDevice.cpp b/lib/Target/R600/AMDIL7XXDevice.cpp
new file mode 100644
index 000000000000..ea6ac34f570c
--- /dev/null
+++ b/lib/Target/R600/AMDIL7XXDevice.cpp
@@ -0,0 +1,115 @@
+//===-- AMDIL7XXDevice.cpp - Device Info for 7XX GPUs ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+// \file
+//==-----------------------------------------------------------------------===//
+#include "AMDIL7XXDevice.h"
+#include "AMDGPUSubtarget.h"
+#include "AMDILDevice.h"
+
+using namespace llvm;
+
+AMDGPU7XXDevice::AMDGPU7XXDevice(AMDGPUSubtarget *ST) : AMDGPUDevice(ST) {
+  setCaps();
+  std::string name = mSTM->getDeviceName();
+  if (name == "rv710") {
+    DeviceFlag = OCL_DEVICE_RV710;
+  } else if (name == "rv730") {
+    DeviceFlag = OCL_DEVICE_RV730;
+  } else {
+    DeviceFlag = OCL_DEVICE_RV770;
+  }
+}
+
+AMDGPU7XXDevice::~AMDGPU7XXDevice() {
+}
+
+void AMDGPU7XXDevice::setCaps() {
+  mSWBits.set(AMDGPUDeviceInfo::LocalMem);
+}
+
+size_t AMDGPU7XXDevice::getMaxLDSSize() const {
+  if (usesHardware(AMDGPUDeviceInfo::LocalMem)) {
+    return MAX_LDS_SIZE_700;
+  }
+  return 0;
+}
+
+size_t AMDGPU7XXDevice::getWavefrontSize() const {
+  return AMDGPUDevice::HalfWavefrontSize;
+}
+
+uint32_t AMDGPU7XXDevice::getGeneration() const {
+  return AMDGPUDeviceInfo::HD4XXX;
+}
+
+uint32_t AMDGPU7XXDevice::getResourceID(uint32_t DeviceID) const {
+  switch (DeviceID) {
+  default:
+    assert(0 && "ID type passed in is unknown!");
+    break;
+  case GLOBAL_ID:
+  case CONSTANT_ID:
+  case RAW_UAV_ID:
+  case ARENA_UAV_ID:
+    break;
+  case LDS_ID:
+    if (usesHardware(AMDGPUDeviceInfo::LocalMem)) {
+      return DEFAULT_LDS_ID;
+    }
+    break;
+  case SCRATCH_ID:
+    if (usesHardware(AMDGPUDeviceInfo::PrivateMem)) {
+      return DEFAULT_SCRATCH_ID;
+    }
+    break;
+  case GDS_ID:
+    assert(0 && "GDS UAV ID is not supported on this chip");
+    if (usesHardware(AMDGPUDeviceInfo::RegionMem)) {
+      return DEFAULT_GDS_ID;
+    }
+    break;
+  };
+
+  return 0;
+}
+
+uint32_t AMDGPU7XXDevice::getMaxNumUAVs() const {
+  return 1;
+}
+
+AMDGPU770Device::AMDGPU770Device(AMDGPUSubtarget *ST): AMDGPU7XXDevice(ST) {
+  setCaps();
+}
+
+AMDGPU770Device::~AMDGPU770Device() {
+}
+
+void AMDGPU770Device::setCaps() {
+  if (mSTM->isOverride(AMDGPUDeviceInfo::DoubleOps)) {
+    mSWBits.set(AMDGPUDeviceInfo::FMA);
+    mHWBits.set(AMDGPUDeviceInfo::DoubleOps);
+  }
+  mSWBits.set(AMDGPUDeviceInfo::BarrierDetect);
+  mHWBits.reset(AMDGPUDeviceInfo::LongOps);
+  mSWBits.set(AMDGPUDeviceInfo::LongOps);
+  mSWBits.set(AMDGPUDeviceInfo::LocalMem);
+}
+
+size_t AMDGPU770Device::getWavefrontSize() const {
+  return AMDGPUDevice::WavefrontSize;
+}
+
+AMDGPU710Device::AMDGPU710Device(AMDGPUSubtarget *ST) : AMDGPU7XXDevice(ST) {
+}
+
+AMDGPU710Device::~AMDGPU710Device() {
+}
+
+size_t AMDGPU710Device::getWavefrontSize() const {
+  return AMDGPUDevice::QuarterWavefrontSize;
+}
diff --git a/lib/Target/R600/AMDIL7XXDevice.h b/lib/Target/R600/AMDIL7XXDevice.h
new file mode 100644
index 000000000000..1cf4ca415a4c
--- /dev/null
+++ b/lib/Target/R600/AMDIL7XXDevice.h
@@ -0,0 +1,72 @@
+//==-- AMDIL7XXDevice.h - Define 7XX Device Device for AMDIL ---*- C++ -*--===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+/// \file
+/// \brief Interface for the subtarget data classes.
+///
+/// This file will define the interface that each generation needs to
+/// implement in order to correctly answer queries on the capabilities of the
+/// specific hardware.
+//===----------------------------------------------------------------------===//
+#ifndef AMDIL7XXDEVICEIMPL_H
+#define AMDIL7XXDEVICEIMPL_H
+#include "AMDILDevice.h"
+
+namespace llvm {
+class AMDGPUSubtarget;
+
+//===----------------------------------------------------------------------===//
+// 7XX generation of devices and their respective sub classes
+//===----------------------------------------------------------------------===//
+
+/// \brief The AMDGPU7XXDevice class represents the generic 7XX device.
+///
+/// All 7XX devices are derived from this class. The AMDGPU7XX device will only
+/// support the minimal features that are required to be considered OpenCL 1.0
+/// compliant and nothing more.
+class AMDGPU7XXDevice : public AMDGPUDevice {
+public:
+  AMDGPU7XXDevice(AMDGPUSubtarget *ST);
+  virtual ~AMDGPU7XXDevice();
+  virtual size_t getMaxLDSSize() const;
+  virtual size_t getWavefrontSize() const;
+  virtual uint32_t getGeneration() const;
+  virtual uint32_t getResourceID(uint32_t DeviceID) const;
+  virtual uint32_t getMaxNumUAVs() const;
+
+protected:
+  virtual void setCaps();
+};
+
+/// \brief The AMDGPU770Device class represents the RV770 chip and it's
+/// derivative cards.
+///
+/// The difference between this device and the base class is this device device
+/// adds support for double precision and has a larger wavefront size.
+class AMDGPU770Device : public AMDGPU7XXDevice {
+public:
+  AMDGPU770Device(AMDGPUSubtarget *ST);
+  virtual ~AMDGPU770Device();
+  virtual size_t getWavefrontSize() const;
+private:
+  virtual void setCaps();
+};
+
+/// \brief The AMDGPU710Device class derives from the 7XX base class.
+///
+/// This class is a smaller derivative, so we need to overload some of the
+/// functions in order to correctly specify this information.
+class AMDGPU710Device : public AMDGPU7XXDevice {
+public:
+  AMDGPU710Device(AMDGPUSubtarget *ST);
+  virtual ~AMDGPU710Device();
+  virtual size_t getWavefrontSize() const;
+};
+
+} // namespace llvm
+#endif // AMDILDEVICEIMPL_H
diff --git a/lib/Target/R600/AMDILBase.td b/lib/Target/R600/AMDILBase.td
new file mode 100644
index 000000000000..c12cedcf7fd5
--- /dev/null
+++ b/lib/Target/R600/AMDILBase.td
@@ -0,0 +1,85 @@
+//===- AMDIL.td - AMDIL Target Machine -------------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+// Target-independent interfaces which we are implementing
+//===----------------------------------------------------------------------===//
+
+include "llvm/Target/Target.td"
+
+// Dummy Instruction itineraries for pseudo instructions
+def ALU_NULL : FuncUnit;
+def NullALU : InstrItinClass;
+
+//===----------------------------------------------------------------------===//
+// AMDIL Subtarget features.
+//===----------------------------------------------------------------------===//
+def FeatureFP64     : SubtargetFeature<"fp64",
+        "CapsOverride[AMDGPUDeviceInfo::DoubleOps]",
+        "true",
+        "Enable 64bit double precision operations">;
+def FeatureByteAddress    : SubtargetFeature<"byte_addressable_store",
+        "CapsOverride[AMDGPUDeviceInfo::ByteStores]",
+        "true",
+        "Enable byte addressable stores">;
+def FeatureBarrierDetect : SubtargetFeature<"barrier_detect",
+        "CapsOverride[AMDGPUDeviceInfo::BarrierDetect]",
+        "true",
+        "Enable duplicate barrier detection(HD5XXX or later).">;
+def FeatureImages : SubtargetFeature<"images",
+        "CapsOverride[AMDGPUDeviceInfo::Images]",
+        "true",
+        "Enable image functions">;
+def FeatureMultiUAV : SubtargetFeature<"multi_uav",
+        "CapsOverride[AMDGPUDeviceInfo::MultiUAV]",
+        "true",
+        "Generate multiple UAV code(HD5XXX family or later)">;
+def FeatureMacroDB : SubtargetFeature<"macrodb",
+        "CapsOverride[AMDGPUDeviceInfo::MacroDB]",
+        "true",
+        "Use internal macrodb, instead of macrodb in driver">;
+def FeatureNoAlias : SubtargetFeature<"noalias",
+        "CapsOverride[AMDGPUDeviceInfo::NoAlias]",
+        "true",
+        "assert that all kernel argument pointers are not aliased">;
+def FeatureNoInline : SubtargetFeature<"no-inline",
+        "CapsOverride[AMDGPUDeviceInfo::NoInline]",
+        "true",
+        "specify whether to not inline functions">;
+
+def Feature64BitPtr : SubtargetFeature<"64BitPtr",
+        "Is64bit",
+        "false",
+        "Specify if 64bit addressing should be used.">;
+
+def Feature32on64BitPtr : SubtargetFeature<"64on32BitPtr",
+        "Is32on64bit",
+        "false",
+        "Specify if 64bit sized pointers with 32bit addressing should be used.">;
+def FeatureDebug : SubtargetFeature<"debug",
+        "CapsOverride[AMDGPUDeviceInfo::Debug]",
+        "true",
+        "Debug mode is enabled, so disable hardware accelerated address spaces.">;
+def FeatureDumpCode : SubtargetFeature <"DumpCode",
+        "DumpCode",
+        "true",
+        "Dump MachineInstrs in the CodeEmitter">;
+
+def FeatureR600ALUInst : SubtargetFeature<"R600ALUInst",
+        "R600ALUInst",
+        "false",
+        "Older version of ALU instructions encoding.">;
+
+
+//===----------------------------------------------------------------------===//
+// Register File, Calling Conv, Instruction Descriptions
+//===----------------------------------------------------------------------===//
+
+
+include "AMDILRegisterInfo.td"
+include "AMDILInstrInfo.td"
+
diff --git a/lib/Target/R600/AMDILCFGStructurizer.cpp b/lib/Target/R600/AMDILCFGStructurizer.cpp
new file mode 100644
index 000000000000..b0cd0f9756a4
--- /dev/null
+++ b/lib/Target/R600/AMDILCFGStructurizer.cpp
@@ -0,0 +1,3051 @@
+//===-- AMDILCFGStructurizer.cpp - CFG Structurizer -----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//==-----------------------------------------------------------------------===//
+
+#define DEBUGME 0
+#define DEBUG_TYPE "structcfg"
+
+#include "AMDGPUInstrInfo.h"
+#include "AMDIL.h"
+#include "llvm/ADT/SCCIterator.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/Analysis/DominatorInternals.h"
+#include "llvm/Analysis/Dominators.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineJumpTableInfo.h"
+#include "llvm/CodeGen/MachineLoopInfo.h"
+#include "llvm/CodeGen/MachinePostDominators.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/Target/TargetInstrInfo.h"
+
+using namespace llvm;
+
+// TODO: move-begin.
+
+//===----------------------------------------------------------------------===//
+//
+// Statistics for CFGStructurizer.
+//
+//===----------------------------------------------------------------------===//
+
+STATISTIC(numSerialPatternMatch,    "CFGStructurizer number of serial pattern "
+    "matched");
+STATISTIC(numIfPatternMatch,        "CFGStructurizer number of if pattern "
+    "matched");
+STATISTIC(numLoopbreakPatternMatch, "CFGStructurizer number of loop-break "
+    "pattern matched");
+STATISTIC(numLoopcontPatternMatch,  "CFGStructurizer number of loop-continue "
+    "pattern matched");
+STATISTIC(numLoopPatternMatch,      "CFGStructurizer number of loop pattern "
+    "matched");
+STATISTIC(numClonedBlock,           "CFGStructurizer cloned blocks");
+STATISTIC(numClonedInstr,           "CFGStructurizer cloned instructions");
+
+//===----------------------------------------------------------------------===//
+//
+// Miscellaneous utility for CFGStructurizer.
+//
+//===----------------------------------------------------------------------===//
+namespace llvmCFGStruct {
+#define SHOWNEWINSTR(i) \
+  if (DEBUGME) errs() << "New instr: " << *i << "\n"
+
+#define SHOWNEWBLK(b, msg) \
+if (DEBUGME) { \
+  errs() << msg << "BB" << b->getNumber() << "size " << b->size(); \
+  errs() << "\n"; \
+}
+
+#define SHOWBLK_DETAIL(b, msg) \
+if (DEBUGME) { \
+  if (b) { \
+  errs() << msg << "BB" << b->getNumber() << "size " << b->size(); \
+  b->print(errs()); \
+  errs() << "\n"; \
+  } \
+}
+
+#define INVALIDSCCNUM -1
+#define INVALIDREGNUM 0
+
+template<class LoopinfoT>
+void PrintLoopinfo(const LoopinfoT &LoopInfo, llvm::raw_ostream &OS) {
+  for (typename LoopinfoT::iterator iter = LoopInfo.begin(),
+       iterEnd = LoopInfo.end();
+       iter != iterEnd; ++iter) {
+    (*iter)->print(OS, 0);
+  }
+}
+
+template<class NodeT>
+void ReverseVector(SmallVector<NodeT *, DEFAULT_VEC_SLOTS> &Src) {
+  size_t sz = Src.size();
+  for (size_t i = 0; i < sz/2; ++i) {
+    NodeT *t = Src[i];
+    Src[i] = Src[sz - i - 1];
+    Src[sz - i - 1] = t;
+  }
+}
+
+} //end namespace llvmCFGStruct
+
+//===----------------------------------------------------------------------===//
+//
+// supporting data structure for CFGStructurizer
+//
+//===----------------------------------------------------------------------===//
+
+namespace llvmCFGStruct {
+template<class PassT>
+struct CFGStructTraits {
+};
+
+template <class InstrT>
+class BlockInformation {
+public:
+  bool isRetired;
+  int  sccNum;
+  //SmallVector<InstrT*, DEFAULT_VEC_SLOTS> succInstr;
+  //Instructions defining the corresponding successor.
+  BlockInformation() : isRetired(false), sccNum(INVALIDSCCNUM) {}
+};
+
+template <class BlockT, class InstrT, class RegiT>
+class LandInformation {
+public:
+  BlockT *landBlk;
+  std::set<RegiT> breakInitRegs;  //Registers that need to "reg = 0", before
+                                  //WHILELOOP(thisloop) init before entering
+                                  //thisloop.
+  std::set<RegiT> contInitRegs;   //Registers that need to "reg = 0", after
+                                  //WHILELOOP(thisloop) init after entering
+                                  //thisloop.
+  std::set<RegiT> endbranchInitRegs; //Init before entering this loop, at loop
+                                     //land block, branch cond on this reg.
+  std::set<RegiT> breakOnRegs;       //registers that need to "if (reg) break
+                                     //endif" after ENDLOOP(thisloop) break
+                                     //outerLoopOf(thisLoop).
+  std::set<RegiT> contOnRegs;       //registers that need to "if (reg) continue
+                                    //endif" after ENDLOOP(thisloop) continue on
+                                    //outerLoopOf(thisLoop).
+  LandInformation() : landBlk(NULL) {}
+};
+
+} //end of namespace llvmCFGStruct
+
+//===----------------------------------------------------------------------===//
+//
+// CFGStructurizer
+//
+//===----------------------------------------------------------------------===//
+
+namespace llvmCFGStruct {
+// bixia TODO: port it to BasicBlock, not just MachineBasicBlock.
+template<class PassT>
+class  CFGStructurizer {
+public:
+  typedef enum {
+    Not_SinglePath = 0,
+    SinglePath_InPath = 1,
+    SinglePath_NotInPath = 2
+  } PathToKind;
+
+public:
+  typedef typename PassT::InstructionType         InstrT;
+  typedef typename PassT::FunctionType            FuncT;
+  typedef typename PassT::DominatortreeType       DomTreeT;
+  typedef typename PassT::PostDominatortreeType   PostDomTreeT;
+  typedef typename PassT::DomTreeNodeType         DomTreeNodeT;
+  typedef typename PassT::LoopinfoType            LoopInfoT;
+
+  typedef GraphTraits<FuncT *>                    FuncGTraits;
+  //typedef FuncGTraits::nodes_iterator BlockIterator;
+  typedef typename FuncT::iterator                BlockIterator;
+
+  typedef typename FuncGTraits::NodeType          BlockT;
+  typedef GraphTraits<BlockT *>                   BlockGTraits;
+  typedef GraphTraits<Inverse<BlockT *> >         InvBlockGTraits;
+  //typedef BlockGTraits::succ_iterator InstructionIterator;
+  typedef typename BlockT::iterator               InstrIterator;
+
+  typedef CFGStructTraits<PassT>                  CFGTraits;
+  typedef BlockInformation<InstrT>                BlockInfo;
+  typedef std::map<BlockT *, BlockInfo *>         BlockInfoMap;
+
+  typedef int                                     RegiT;
+  typedef typename PassT::LoopType                LoopT;
+  typedef LandInformation<BlockT, InstrT, RegiT>  LoopLandInfo;
+        typedef std::map<LoopT *, LoopLandInfo *> LoopLandInfoMap;
+        //landing info for loop break
+  typedef SmallVector<BlockT *, 32>               BlockTSmallerVector;
+
+public:
+  CFGStructurizer();
+  ~CFGStructurizer();
+
+  /// Perform the CFG structurization
+  bool run(FuncT &Func, PassT &Pass, const AMDGPURegisterInfo *tri);
+
+  /// Perform the CFG preparation
+  bool prepare(FuncT &Func, PassT &Pass, const AMDGPURegisterInfo *tri);
+
+private:
+  void reversePredicateSetter(typename BlockT::iterator);
+  void   orderBlocks();
+  void   printOrderedBlocks(llvm::raw_ostream &OS);
+  int patternMatch(BlockT *CurBlock);
+  int patternMatchGroup(BlockT *CurBlock);
+
+  int serialPatternMatch(BlockT *CurBlock);
+  int ifPatternMatch(BlockT *CurBlock);
+  int switchPatternMatch(BlockT *CurBlock);
+  int loopendPatternMatch(BlockT *CurBlock);
+  int loopPatternMatch(BlockT *CurBlock);
+
+  int loopbreakPatternMatch(LoopT *LoopRep, BlockT *LoopHeader);
+  int loopcontPatternMatch(LoopT *LoopRep, BlockT *LoopHeader);
+  //int loopWithoutBreak(BlockT *);
+
+  void handleLoopbreak (BlockT *ExitingBlock, LoopT *ExitingLoop,
+                        BlockT *ExitBlock, LoopT *exitLoop, BlockT *landBlock);
+  void handleLoopcontBlock(BlockT *ContingBlock, LoopT *contingLoop,
+                           BlockT *ContBlock, LoopT *contLoop);
+  bool isSameloopDetachedContbreak(BlockT *Src1Block, BlockT *Src2Block);
+  int handleJumpintoIf(BlockT *HeadBlock, BlockT *TrueBlock,
+                       BlockT *FalseBlock);
+  int handleJumpintoIfImp(BlockT *HeadBlock, BlockT *TrueBlock,
+                          BlockT *FalseBlock);
+  int improveSimpleJumpintoIf(BlockT *HeadBlock, BlockT *TrueBlock,
+                              BlockT *FalseBlock, BlockT **LandBlockPtr);
+  void showImproveSimpleJumpintoIf(BlockT *HeadBlock, BlockT *TrueBlock,
+                                   BlockT *FalseBlock, BlockT *LandBlock,
+                                   bool Detail = false);
+  PathToKind singlePathTo(BlockT *SrcBlock, BlockT *DstBlock,
+                          bool AllowSideEntry = true);
+  BlockT *singlePathEnd(BlockT *srcBlock, BlockT *DstBlock,
+                        bool AllowSideEntry = true);
+  int cloneOnSideEntryTo(BlockT *PreBlock, BlockT *SrcBlock, BlockT *DstBlock);
+  void mergeSerialBlock(BlockT *DstBlock, BlockT *srcBlock);
+
+  void mergeIfthenelseBlock(InstrT *BranchInstr, BlockT *CurBlock,
+                            BlockT *TrueBlock, BlockT *FalseBlock,
+                            BlockT *LandBlock);
+  void mergeLooplandBlock(BlockT *DstBlock, LoopLandInfo *LoopLand);
+  void mergeLoopbreakBlock(BlockT *ExitingBlock, BlockT *ExitBlock,
+                           BlockT *ExitLandBlock, RegiT SetReg);
+  void settleLoopcontBlock(BlockT *ContingBlock, BlockT *ContBlock,
+                           RegiT SetReg);
+  BlockT *relocateLoopcontBlock(LoopT *ParentLoopRep, LoopT *LoopRep,
+                                std::set<BlockT*> &ExitBlockSet,
+                                BlockT *ExitLandBlk);
+  BlockT *addLoopEndbranchBlock(LoopT *LoopRep,
+                                BlockTSmallerVector &ExitingBlocks,
+                                BlockTSmallerVector &ExitBlocks);
+  BlockT *normalizeInfiniteLoopExit(LoopT *LoopRep);
+  void removeUnconditionalBranch(BlockT *SrcBlock);
+  void removeRedundantConditionalBranch(BlockT *SrcBlock);
+  void addDummyExitBlock(SmallVector<BlockT *, DEFAULT_VEC_SLOTS> &RetBlocks);
+
+  void removeSuccessor(BlockT *SrcBlock);
+  BlockT *cloneBlockForPredecessor(BlockT *CurBlock, BlockT *PredBlock);
+  BlockT *exitingBlock2ExitBlock (LoopT *LoopRep, BlockT *exitingBlock);
+
+  void migrateInstruction(BlockT *SrcBlock, BlockT *DstBlock,
+                          InstrIterator InsertPos);
+
+  void recordSccnum(BlockT *SrcBlock, int SCCNum);
+  int getSCCNum(BlockT *srcBlk);
+
+  void retireBlock(BlockT *DstBlock, BlockT *SrcBlock);
+  bool isRetiredBlock(BlockT *SrcBlock);
+  bool isActiveLoophead(BlockT *CurBlock);
+  bool needMigrateBlock(BlockT *Block);
+
+  BlockT *recordLoopLandBlock(LoopT *LoopRep, BlockT *LandBlock,
+                              BlockTSmallerVector &exitBlocks,
+                              std::set<BlockT*> &ExitBlockSet);
+  void setLoopLandBlock(LoopT *LoopRep, BlockT *Block = NULL);
+  BlockT *getLoopLandBlock(LoopT *LoopRep);
+  LoopLandInfo *getLoopLandInfo(LoopT *LoopRep);
+
+  void addLoopBreakOnReg(LoopT *LoopRep, RegiT RegNum);
+  void addLoopContOnReg(LoopT *LoopRep, RegiT RegNum);
+  void addLoopBreakInitReg(LoopT *LoopRep, RegiT RegNum);
+  void addLoopContInitReg(LoopT *LoopRep, RegiT RegNum);
+  void addLoopEndbranchInitReg(LoopT *LoopRep, RegiT RegNum);
+
+  bool hasBackEdge(BlockT *curBlock);
+  unsigned getLoopDepth  (LoopT *LoopRep);
+  int countActiveBlock(
+    typename SmallVector<BlockT *, DEFAULT_VEC_SLOTS>::const_iterator IterStart,
+    typename SmallVector<BlockT *, DEFAULT_VEC_SLOTS>::const_iterator IterEnd);
+    BlockT *findNearestCommonPostDom(std::set<BlockT *>&);
+  BlockT *findNearestCommonPostDom(BlockT *Block1, BlockT *Block2);
+
+private:
+  DomTreeT *domTree;
+  PostDomTreeT *postDomTree;
+  LoopInfoT *loopInfo;
+  PassT *passRep;
+  FuncT *funcRep;
+
+  BlockInfoMap blockInfoMap;
+  LoopLandInfoMap loopLandInfoMap;
+  SmallVector<BlockT *, DEFAULT_VEC_SLOTS> orderedBlks;
+  const AMDGPURegisterInfo *TRI;
+
+};  //template class CFGStructurizer
+
+template<class PassT> CFGStructurizer<PassT>::CFGStructurizer()
+  : domTree(NULL), postDomTree(NULL), loopInfo(NULL) {
+}
+
+template<class PassT> CFGStructurizer<PassT>::~CFGStructurizer() {
+  for (typename BlockInfoMap::iterator I = blockInfoMap.begin(),
+       E = blockInfoMap.end(); I != E; ++I) {
+    delete I->second;
+  }
+}
+
+template<class PassT>
+bool CFGStructurizer<PassT>::prepare(FuncT &func, PassT &pass,
+                                     const AMDGPURegisterInfo * tri) {
+  passRep = &pass;
+  funcRep = &func;
+  TRI = tri;
+
+  bool changed = false;
+
+  //FIXME: if not reducible flow graph, make it so ???
+
+  if (DEBUGME) {
+        errs() << "AMDGPUCFGStructurizer::prepare\n";
+  }
+
+  loopInfo = CFGTraits::getLoopInfo(pass);
+  if (DEBUGME) {
+    errs() << "LoopInfo:\n";
+    PrintLoopinfo(*loopInfo, errs());
+  }
+
+  orderBlocks();
+  if (DEBUGME) {
+    errs() << "Ordered blocks:\n";
+    printOrderedBlocks(errs());
+  }
+
+  SmallVector<BlockT *, DEFAULT_VEC_SLOTS> retBlks;
+
+  for (typename LoopInfoT::iterator iter = loopInfo->begin(),
+       iterEnd = loopInfo->end();
+       iter != iterEnd; ++iter) {
+    LoopT* loopRep = (*iter);
+    BlockTSmallerVector exitingBlks;
+    loopRep->getExitingBlocks(exitingBlks);
+    
+    if (exitingBlks.size() == 0) {
+      BlockT* dummyExitBlk = normalizeInfiniteLoopExit(loopRep);
+      if (dummyExitBlk != NULL)
+        retBlks.push_back(dummyExitBlk);
+    }
+  }
+
+  // Remove unconditional branch instr.
+  // Add dummy exit block iff there are multiple returns.
+
+  for (typename SmallVector<BlockT *, DEFAULT_VEC_SLOTS>::const_iterator
+       iterBlk = orderedBlks.begin(), iterEndBlk = orderedBlks.end();
+       iterBlk != iterEndBlk;
+       ++iterBlk) {
+    BlockT *curBlk = *iterBlk;
+    removeUnconditionalBranch(curBlk);
+    removeRedundantConditionalBranch(curBlk);
+    if (CFGTraits::isReturnBlock(curBlk)) {
+      retBlks.push_back(curBlk);
+    }
+    assert(curBlk->succ_size() <= 2);
+  } //for
+
+  if (retBlks.size() >= 2) {
+    addDummyExitBlock(retBlks);
+    changed = true;
+  }
+
+  return changed;
+} //CFGStructurizer::prepare
+
+template<class PassT>
+bool CFGStructurizer<PassT>::run(FuncT &func, PassT &pass,
+    const AMDGPURegisterInfo * tri) {
+  passRep = &pass;
+  funcRep = &func;
+  TRI = tri;
+
+  //Assume reducible CFG...
+  if (DEBUGME) {
+    errs() << "AMDGPUCFGStructurizer::run\n";
+    func.viewCFG();
+  }
+
+  domTree = CFGTraits::getDominatorTree(pass);
+  if (DEBUGME) {
+    domTree->print(errs(), (const llvm::Module*)0);
+  }
+
+  postDomTree = CFGTraits::getPostDominatorTree(pass);
+  if (DEBUGME) {
+    postDomTree->print(errs());
+  }
+
+  loopInfo = CFGTraits::getLoopInfo(pass);
+  if (DEBUGME) {
+    errs() << "LoopInfo:\n";
+    PrintLoopinfo(*loopInfo, errs());
+  }
+
+  orderBlocks();
+#ifdef STRESSTEST
+  //Use the worse block ordering to test the algorithm.
+  ReverseVector(orderedBlks);
+#endif
+
+  if (DEBUGME) {
+    errs() << "Ordered blocks:\n";
+    printOrderedBlocks(errs());
+  }
+  int numIter = 0;
+  bool finish = false;
+  BlockT *curBlk;
+  bool makeProgress = false;
+  int numRemainedBlk = countActiveBlock(orderedBlks.begin(),
+                                        orderedBlks.end());
+
+  do {
+    ++numIter;
+    if (DEBUGME) {
+      errs() << "numIter = " << numIter
+             << ", numRemaintedBlk = " << numRemainedBlk << "\n";
+    }
+
+    typename SmallVector<BlockT *, DEFAULT_VEC_SLOTS>::const_iterator
+      iterBlk = orderedBlks.begin();
+    typename SmallVector<BlockT *, DEFAULT_VEC_SLOTS>::const_iterator
+      iterBlkEnd = orderedBlks.end();
+
+    typename SmallVector<BlockT *, DEFAULT_VEC_SLOTS>::const_iterator
+      sccBeginIter = iterBlk;
+    BlockT *sccBeginBlk = NULL;
+    int sccNumBlk = 0;  // The number of active blocks, init to a
+                        // maximum possible number.
+    int sccNumIter;     // Number of iteration in this SCC.
+
+    while (iterBlk != iterBlkEnd) {
+      curBlk = *iterBlk;
+
+      if (sccBeginBlk == NULL) {
+        sccBeginIter = iterBlk;
+        sccBeginBlk = curBlk;
+        sccNumIter = 0;
+        sccNumBlk = numRemainedBlk; // Init to maximum possible number.
+        if (DEBUGME) {
+              errs() << "start processing SCC" << getSCCNum(sccBeginBlk);
+              errs() << "\n";
+        }
+      }
+
+      if (!isRetiredBlock(curBlk)) {
+        patternMatch(curBlk);
+      }
+
+      ++iterBlk;
+
+      bool contNextScc = true;
+      if (iterBlk == iterBlkEnd
+          || getSCCNum(sccBeginBlk) != getSCCNum(*iterBlk)) {
+        // Just finish one scc.
+        ++sccNumIter;
+        int sccRemainedNumBlk = countActiveBlock(sccBeginIter, iterBlk);
+        if (sccRemainedNumBlk != 1 && sccRemainedNumBlk >= sccNumBlk) {
+          if (DEBUGME) {
+            errs() << "Can't reduce SCC " << getSCCNum(curBlk)
+                   << ", sccNumIter = " << sccNumIter;
+            errs() << "doesn't make any progress\n";
+          }
+          contNextScc = true;
+        } else if (sccRemainedNumBlk != 1 && sccRemainedNumBlk < sccNumBlk) {
+          sccNumBlk = sccRemainedNumBlk;
+          iterBlk = sccBeginIter;
+          contNextScc = false;
+          if (DEBUGME) {
+            errs() << "repeat processing SCC" << getSCCNum(curBlk)
+                   << "sccNumIter = " << sccNumIter << "\n";
+            func.viewCFG();
+          }
+        } else {
+          // Finish the current scc.
+          contNextScc = true;
+        }
+      } else {
+        // Continue on next component in the current scc.
+        contNextScc = false;
+      }
+
+      if (contNextScc) {
+        sccBeginBlk = NULL;
+      }
+    } //while, "one iteration" over the function.
+
+    BlockT *entryBlk = FuncGTraits::nodes_begin(&func);
+    if (entryBlk->succ_size() == 0) {
+      finish = true;
+      if (DEBUGME) {
+        errs() << "Reduce to one block\n";
+      }
+    } else {
+      int newnumRemainedBlk
+        = countActiveBlock(orderedBlks.begin(), orderedBlks.end());
+      // consider cloned blocks ??
+      if (newnumRemainedBlk == 1 || newnumRemainedBlk < numRemainedBlk) {
+        makeProgress = true;
+        numRemainedBlk = newnumRemainedBlk;
+      } else {
+        makeProgress = false;
+        if (DEBUGME) {
+          errs() << "No progress\n";
+        }
+      }
+    }
+  } while (!finish && makeProgress);
+
+  // Misc wrap up to maintain the consistency of the Function representation.
+  CFGTraits::wrapup(FuncGTraits::nodes_begin(&func));
+
+  // Detach retired Block, release memory.
+  for (typename BlockInfoMap::iterator iterMap = blockInfoMap.begin(),
+       iterEndMap = blockInfoMap.end(); iterMap != iterEndMap; ++iterMap) {
+    if ((*iterMap).second && (*iterMap).second->isRetired) {
+      assert(((*iterMap).first)->getNumber() != -1);
+      if (DEBUGME) {
+        errs() << "Erase BB" << ((*iterMap).first)->getNumber() << "\n";
+      }
+      (*iterMap).first->eraseFromParent();  //Remove from the parent Function.
+    }
+    delete (*iterMap).second;
+  }
+  blockInfoMap.clear();
+
+  // clear loopLandInfoMap
+  for (typename LoopLandInfoMap::iterator iterMap = loopLandInfoMap.begin(),
+       iterEndMap = loopLandInfoMap.end(); iterMap != iterEndMap; ++iterMap) {
+    delete (*iterMap).second;
+  }
+  loopLandInfoMap.clear();
+
+  if (DEBUGME) {
+    func.viewCFG();
+  }
+
+  if (!finish) {
+    assert(!"IRREDUCIBL_CF");
+  }
+
+  return true;
+} //CFGStructurizer::run
+
+/// Print the ordered Blocks.
+///
+template<class PassT>
+void CFGStructurizer<PassT>::printOrderedBlocks(llvm::raw_ostream &os) {
+  size_t i = 0;
+  for (typename SmallVector<BlockT *, DEFAULT_VEC_SLOTS>::const_iterator
+      iterBlk = orderedBlks.begin(), iterBlkEnd = orderedBlks.end();
+       iterBlk != iterBlkEnd;
+       ++iterBlk, ++i) {
+    os << "BB" << (*iterBlk)->getNumber();
+    os << "(" << getSCCNum(*iterBlk) << "," << (*iterBlk)->size() << ")";
+    if (i != 0 && i % 10 == 0) {
+      os << "\n";
+    } else {
+      os << " ";
+    }
+  }
+} //printOrderedBlocks
+
+/// Compute the reversed DFS post order of Blocks
+///
+template<class PassT> void CFGStructurizer<PassT>::orderBlocks() {
+  int sccNum = 0;
+  BlockT *bb;
+  for (scc_iterator<FuncT *> sccIter = scc_begin(funcRep),
+       sccEnd = scc_end(funcRep); sccIter != sccEnd; ++sccIter, ++sccNum) {
+    std::vector<BlockT *> &sccNext = *sccIter;
+    for (typename std::vector<BlockT *>::const_iterator
+         blockIter = sccNext.begin(), blockEnd = sccNext.end();
+         blockIter != blockEnd; ++blockIter) {
+      bb = *blockIter;
+      orderedBlks.push_back(bb);
+      recordSccnum(bb, sccNum);
+    }
+  }
+
+  //walk through all the block in func to check for unreachable
+  for (BlockIterator blockIter1 = FuncGTraits::nodes_begin(funcRep),
+       blockEnd1 = FuncGTraits::nodes_end(funcRep);
+       blockIter1 != blockEnd1; ++blockIter1) {
+    BlockT *bb = &(*blockIter1);
+    sccNum = getSCCNum(bb);
+    if (sccNum == INVALIDSCCNUM) {
+      errs() << "unreachable block BB" << bb->getNumber() << "\n";
+    }
+  }
+} //orderBlocks
+
+template<class PassT> int CFGStructurizer<PassT>::patternMatch(BlockT *curBlk) {
+  int numMatch = 0;
+  int curMatch;
+
+  if (DEBUGME) {
+        errs() << "Begin patternMatch BB" << curBlk->getNumber() << "\n";
+  }
+
+  while ((curMatch = patternMatchGroup(curBlk)) > 0) {
+    numMatch += curMatch;
+  }
+
+  if (DEBUGME) {
+        errs() << "End patternMatch BB" << curBlk->getNumber()
+      << ", numMatch = " << numMatch << "\n";
+  }
+
+  return numMatch;
+} //patternMatch
+
+template<class PassT>
+int CFGStructurizer<PassT>::patternMatchGroup(BlockT *curBlk) {
+  int numMatch = 0;
+  numMatch += serialPatternMatch(curBlk);
+  numMatch += ifPatternMatch(curBlk);
+  numMatch += loopendPatternMatch(curBlk);
+  numMatch += loopPatternMatch(curBlk);
+  return numMatch;
+}//patternMatchGroup
+
+template<class PassT>
+int CFGStructurizer<PassT>::serialPatternMatch(BlockT *curBlk) {
+  if (curBlk->succ_size() != 1) {
+    return 0;
+  }
+
+  BlockT *childBlk = *curBlk->succ_begin();
+  if (childBlk->pred_size() != 1 || isActiveLoophead(childBlk)) {
+    return 0;
+  }
+
+  mergeSerialBlock(curBlk, childBlk);
+  ++numSerialPatternMatch;
+  return 1;
+} //serialPatternMatch
+
+template<class PassT>
+int CFGStructurizer<PassT>::ifPatternMatch(BlockT *curBlk) {
+  //two edges
+  if (curBlk->succ_size() != 2) {
+    return 0;
+  }
+
+  if (hasBackEdge(curBlk)) {
+    return 0;
+  }
+
+  InstrT *branchInstr = CFGTraits::getNormalBlockBranchInstr(curBlk);
+  if (branchInstr == NULL) {
+    return 0;
+  }
+
+  assert(CFGTraits::isCondBranch(branchInstr));
+
+  BlockT *trueBlk = CFGTraits::getTrueBranch(branchInstr);
+  BlockT *falseBlk = CFGTraits::getFalseBranch(curBlk, branchInstr);
+  BlockT *landBlk;
+  int cloned = 0;
+
+  // TODO: Simplify
+  if (trueBlk->succ_size() == 1 && falseBlk->succ_size() == 1
+    && *trueBlk->succ_begin() == *falseBlk->succ_begin()) {
+    landBlk = *trueBlk->succ_begin();
+  } else if (trueBlk->succ_size() == 0 && falseBlk->succ_size() == 0) {
+    landBlk = NULL;
+  } else if (trueBlk->succ_size() == 1 && *trueBlk->succ_begin() == falseBlk) {
+    landBlk = falseBlk;
+    falseBlk = NULL;
+  } else if (falseBlk->succ_size() == 1
+             && *falseBlk->succ_begin() == trueBlk) {
+    landBlk = trueBlk;
+    trueBlk = NULL;
+  } else if (falseBlk->succ_size() == 1
+             && isSameloopDetachedContbreak(trueBlk, falseBlk)) {
+    landBlk = *falseBlk->succ_begin();
+  } else if (trueBlk->succ_size() == 1
+    && isSameloopDetachedContbreak(falseBlk, trueBlk)) {
+    landBlk = *trueBlk->succ_begin();
+  } else {
+    return handleJumpintoIf(curBlk, trueBlk, falseBlk);
+  }
+
+  // improveSimpleJumpinfoIf can handle the case where landBlk == NULL but the
+  // new BB created for landBlk==NULL may introduce new challenge to the
+  // reduction process.
+  if (landBlk != NULL &&
+      ((trueBlk && trueBlk->pred_size() > 1)
+      || (falseBlk && falseBlk->pred_size() > 1))) {
+     cloned += improveSimpleJumpintoIf(curBlk, trueBlk, falseBlk, &landBlk);
+  }
+
+  if (trueBlk && trueBlk->pred_size() > 1) {
+    trueBlk = cloneBlockForPredecessor(trueBlk, curBlk);
+    ++cloned;
+  }
+
+  if (falseBlk && falseBlk->pred_size() > 1) {
+    falseBlk = cloneBlockForPredecessor(falseBlk, curBlk);
+    ++cloned;
+  }
+
+  mergeIfthenelseBlock(branchInstr, curBlk, trueBlk, falseBlk, landBlk);
+
+  ++numIfPatternMatch;
+
+  numClonedBlock += cloned;
+
+  return 1 + cloned;
+} //ifPatternMatch
+
+template<class PassT>
+int CFGStructurizer<PassT>::switchPatternMatch(BlockT *curBlk) {
+  return 0;
+} //switchPatternMatch
+
+template<class PassT>
+int CFGStructurizer<PassT>::loopendPatternMatch(BlockT *curBlk) {
+  LoopT *loopRep = loopInfo->getLoopFor(curBlk);
+  typename std::vector<LoopT *> nestedLoops;
+  while (loopRep) {
+    nestedLoops.push_back(loopRep);
+    loopRep = loopRep->getParentLoop();
+  }
+
+  if (nestedLoops.size() == 0) {
+    return 0;
+  }
+
+  // Process nested loop outside->inside, so "continue" to a outside loop won't
+  // be mistaken as "break" of the current loop.
+  int num = 0;
+  for (typename std::vector<LoopT *>::reverse_iterator
+       iter = nestedLoops.rbegin(), iterEnd = nestedLoops.rend();
+       iter != iterEnd; ++iter) {
+    loopRep = *iter;
+
+    if (getLoopLandBlock(loopRep) != NULL) {
+      continue;
+    }
+
+    BlockT *loopHeader = loopRep->getHeader();
+
+    int numBreak = loopbreakPatternMatch(loopRep, loopHeader);
+
+    if (numBreak == -1) {
+      break;
+    }
+
+    int numCont = loopcontPatternMatch(loopRep, loopHeader);
+    num += numBreak + numCont;
+  }
+
+  return num;
+} //loopendPatternMatch
+
+template<class PassT>
+int CFGStructurizer<PassT>::loopPatternMatch(BlockT *curBlk) {
+  if (curBlk->succ_size() != 0) {
+    return 0;
+  }
+
+  int numLoop = 0;
+  LoopT *loopRep = loopInfo->getLoopFor(curBlk);
+  while (loopRep && loopRep->getHeader() == curBlk) {
+    LoopLandInfo *loopLand = getLoopLandInfo(loopRep);
+    if (loopLand) {
+      BlockT *landBlk = loopLand->landBlk;
+      assert(landBlk);
+      if (!isRetiredBlock(landBlk)) {
+        mergeLooplandBlock(curBlk, loopLand);
+        ++numLoop;
+      }
+    }
+    loopRep = loopRep->getParentLoop();
+  }
+
+  numLoopPatternMatch += numLoop;
+
+  return numLoop;
+} //loopPatternMatch
+
+template<class PassT>
+int CFGStructurizer<PassT>::loopbreakPatternMatch(LoopT *loopRep,
+                                                  BlockT *loopHeader) {
+  BlockTSmallerVector exitingBlks;
+  loopRep->getExitingBlocks(exitingBlks);
+
+  if (DEBUGME) {
+    errs() << "Loop has " << exitingBlks.size() << " exiting blocks\n";
+  }
+
+  if (exitingBlks.size() == 0) {
+    setLoopLandBlock(loopRep);
+    return 0;
+  }
+
+  // Compute the corresponding exitBlks and exit block set.
+  BlockTSmallerVector exitBlks;
+  std::set<BlockT *> exitBlkSet;
+  for (typename BlockTSmallerVector::const_iterator iter = exitingBlks.begin(),
+       iterEnd = exitingBlks.end(); iter != iterEnd; ++iter) {
+    BlockT *exitingBlk = *iter;
+    BlockT *exitBlk = exitingBlock2ExitBlock(loopRep, exitingBlk);
+    exitBlks.push_back(exitBlk);
+    exitBlkSet.insert(exitBlk);  //non-duplicate insert
+  }
+
+  assert(exitBlkSet.size() > 0);
+  assert(exitBlks.size() == exitingBlks.size());
+
+  if (DEBUGME) {
+    errs() << "Loop has " << exitBlkSet.size() << " exit blocks\n";
+  }
+
+  // Find exitLandBlk.
+  BlockT *exitLandBlk = NULL;
+  int numCloned = 0;
+  int numSerial = 0;
+
+  if (exitBlkSet.size() == 1) {
+    exitLandBlk = *exitBlkSet.begin();
+  } else {
+    exitLandBlk = findNearestCommonPostDom(exitBlkSet);
+
+    if (exitLandBlk == NULL) {
+      return -1;
+    }
+
+    bool allInPath = true;
+    bool allNotInPath = true;
+    for (typename std::set<BlockT*>::const_iterator
+         iter = exitBlkSet.begin(),
+         iterEnd = exitBlkSet.end();
+         iter != iterEnd; ++iter) {
+      BlockT *exitBlk = *iter;
+
+      PathToKind pathKind = singlePathTo(exitBlk, exitLandBlk, true);
+      if (DEBUGME) {
+        errs() << "BB" << exitBlk->getNumber()
+               << " to BB" << exitLandBlk->getNumber() << " PathToKind="
+               << pathKind << "\n";
+      }
+
+      allInPath = allInPath && (pathKind == SinglePath_InPath);
+      allNotInPath = allNotInPath && (pathKind == SinglePath_NotInPath);
+
+      if (!allInPath && !allNotInPath) {
+        if (DEBUGME) {
+              errs() << "singlePath check fail\n";
+        }
+        return -1;
+      }
+    } // check all exit blocks
+
+    if (allNotInPath) {
+
+      // TODO: Simplify, maybe separate function?
+      LoopT *parentLoopRep = loopRep->getParentLoop();
+      BlockT *parentLoopHeader = NULL;
+      if (parentLoopRep)
+        parentLoopHeader = parentLoopRep->getHeader();
+
+      if (exitLandBlk == parentLoopHeader &&
+          (exitLandBlk = relocateLoopcontBlock(parentLoopRep,
+                                               loopRep,
+                                               exitBlkSet,
+                                               exitLandBlk)) != NULL) {
+        if (DEBUGME) {
+          errs() << "relocateLoopcontBlock success\n";
+        }
+      } else if ((exitLandBlk = addLoopEndbranchBlock(loopRep,
+                                                      exitingBlks,
+                                                      exitBlks)) != NULL) {
+        if (DEBUGME) {
+          errs() << "insertEndbranchBlock success\n";
+        }
+      } else {
+        if (DEBUGME) {
+          errs() << "loop exit fail\n";
+        }
+        return -1;
+      }
+    }
+
+    // Handle side entry to exit path.
+    exitBlks.clear();
+    exitBlkSet.clear();
+    for (typename BlockTSmallerVector::iterator iterExiting =
+           exitingBlks.begin(),
+         iterExitingEnd = exitingBlks.end();
+         iterExiting != iterExitingEnd; ++iterExiting) {
+      BlockT *exitingBlk = *iterExiting;
+      BlockT *exitBlk = exitingBlock2ExitBlock(loopRep, exitingBlk);
+      BlockT *newExitBlk = exitBlk;
+
+      if (exitBlk != exitLandBlk && exitBlk->pred_size() > 1) {
+        newExitBlk = cloneBlockForPredecessor(exitBlk, exitingBlk);
+        ++numCloned;
+      }
+
+      numCloned += cloneOnSideEntryTo(exitingBlk, newExitBlk, exitLandBlk);
+
+      exitBlks.push_back(newExitBlk);
+      exitBlkSet.insert(newExitBlk);
+    }
+
+    for (typename BlockTSmallerVector::iterator iterExit = exitBlks.begin(),
+         iterExitEnd = exitBlks.end();
+         iterExit != iterExitEnd; ++iterExit) {
+      BlockT *exitBlk = *iterExit;
+      numSerial += serialPatternMatch(exitBlk);
+    }
+
+    for (typename BlockTSmallerVector::iterator iterExit = exitBlks.begin(),
+         iterExitEnd = exitBlks.end();
+         iterExit != iterExitEnd; ++iterExit) {
+      BlockT *exitBlk = *iterExit;
+      if (exitBlk->pred_size() > 1) {
+        if (exitBlk != exitLandBlk) {
+          return -1;
+        }
+      } else {
+        if (exitBlk != exitLandBlk &&
+            (exitBlk->succ_size() != 1 ||
+            *exitBlk->succ_begin() != exitLandBlk)) {
+          return -1;
+        }
+      }
+    }
+  } // else
+
+  exitLandBlk = recordLoopLandBlock(loopRep, exitLandBlk, exitBlks, exitBlkSet);
+
+  // Fold break into the breaking block. Leverage across level breaks.
+  assert(exitingBlks.size() == exitBlks.size());
+  for (typename BlockTSmallerVector::const_iterator iterExit = exitBlks.begin(),
+       iterExiting = exitingBlks.begin(), iterExitEnd = exitBlks.end();
+       iterExit != iterExitEnd; ++iterExit, ++iterExiting) {
+    BlockT *exitBlk = *iterExit;
+    BlockT *exitingBlk = *iterExiting;
+    assert(exitBlk->pred_size() == 1 || exitBlk == exitLandBlk);
+    LoopT *exitingLoop = loopInfo->getLoopFor(exitingBlk);
+    handleLoopbreak(exitingBlk, exitingLoop, exitBlk, loopRep, exitLandBlk);
+  }
+
+  int numBreak = static_cast<int>(exitingBlks.size());
+  numLoopbreakPatternMatch += numBreak;
+  numClonedBlock += numCloned;
+  return numBreak + numSerial + numCloned;
+} //loopbreakPatternMatch
+
+template<class PassT>
+int CFGStructurizer<PassT>::loopcontPatternMatch(LoopT *loopRep,
+                                                 BlockT *loopHeader) {
+  int numCont = 0;
+  SmallVector<BlockT *, DEFAULT_VEC_SLOTS> contBlk;
+  for (typename InvBlockGTraits::ChildIteratorType iter =
+       InvBlockGTraits::child_begin(loopHeader),
+       iterEnd = InvBlockGTraits::child_end(loopHeader);
+       iter != iterEnd; ++iter) {
+    BlockT *curBlk = *iter;
+    if (loopRep->contains(curBlk)) {
+      handleLoopcontBlock(curBlk, loopInfo->getLoopFor(curBlk),
+                          loopHeader, loopRep);
+      contBlk.push_back(curBlk);
+      ++numCont;
+    }
+  }
+
+  for (typename SmallVector<BlockT *, DEFAULT_VEC_SLOTS>::iterator
+       iter = contBlk.begin(), iterEnd = contBlk.end();
+       iter != iterEnd; ++iter) {
+    (*iter)->removeSuccessor(loopHeader);
+  }
+
+  numLoopcontPatternMatch += numCont;
+
+  return numCont;
+} //loopcontPatternMatch
+
+
+template<class PassT>
+bool CFGStructurizer<PassT>::isSameloopDetachedContbreak(BlockT *src1Blk,
+                                                         BlockT *src2Blk) {
+  // return true iff src1Blk->succ_size() == 0 && src1Blk and src2Blk are in the
+  // same loop with LoopLandInfo without explicitly keeping track of
+  // loopContBlks and loopBreakBlks, this is a method to get the information.
+  //
+  if (src1Blk->succ_size() == 0) {
+    LoopT *loopRep = loopInfo->getLoopFor(src1Blk);
+    if (loopRep != NULL && loopRep == loopInfo->getLoopFor(src2Blk)) {
+      LoopLandInfo *&theEntry = loopLandInfoMap[loopRep];
+      if (theEntry != NULL) {
+        if (DEBUGME) {
+          errs() << "isLoopContBreakBlock yes src1 = BB"
+                 << src1Blk->getNumber()
+                 << " src2 = BB" << src2Blk->getNumber() << "\n";
+        }
+        return true;
+      }
+    }
+  }
+  return false;
+}  //isSameloopDetachedContbreak
+
+template<class PassT>
+int CFGStructurizer<PassT>::handleJumpintoIf(BlockT *headBlk,
+                                             BlockT *trueBlk,
+                                             BlockT *falseBlk) {
+  int num = handleJumpintoIfImp(headBlk, trueBlk, falseBlk);
+  if (num == 0) {
+    if (DEBUGME) {
+      errs() << "handleJumpintoIf swap trueBlk and FalseBlk" << "\n";
+    }
+    num = handleJumpintoIfImp(headBlk, falseBlk, trueBlk);
+  }
+  return num;
+}
+
+template<class PassT>
+int CFGStructurizer<PassT>::handleJumpintoIfImp(BlockT *headBlk,
+                                                BlockT *trueBlk,
+                                                BlockT *falseBlk) {
+  int num = 0;
+  BlockT *downBlk;
+
+  //trueBlk could be the common post dominator
+  downBlk = trueBlk;
+
+  if (DEBUGME) {
+    errs() << "handleJumpintoIfImp head = BB" << headBlk->getNumber()
+           << " true = BB" << trueBlk->getNumber()
+           << ", numSucc=" << trueBlk->succ_size()
+           << " false = BB" << falseBlk->getNumber() << "\n";
+  }
+
+  while (downBlk) {
+    if (DEBUGME) {
+      errs() << "check down = BB" << downBlk->getNumber();
+    }
+
+    if (singlePathTo(falseBlk, downBlk) == SinglePath_InPath) {
+      if (DEBUGME) {
+        errs() << " working\n";
+      }
+
+      num += cloneOnSideEntryTo(headBlk, trueBlk, downBlk);
+      num += cloneOnSideEntryTo(headBlk, falseBlk, downBlk);
+
+      numClonedBlock += num;
+      num += serialPatternMatch(*headBlk->succ_begin());
+      num += serialPatternMatch(*(++headBlk->succ_begin()));
+      num += ifPatternMatch(headBlk);
+      assert(num > 0);
+
+      break;
+    }
+    if (DEBUGME) {
+      errs() << " not working\n";
+    }
+    downBlk = (downBlk->succ_size() == 1) ? (*downBlk->succ_begin()) : NULL;
+  } // walk down the postDomTree
+
+  return num;
+} //handleJumpintoIf
+
+template<class PassT>
+void CFGStructurizer<PassT>::showImproveSimpleJumpintoIf(BlockT *headBlk,
+                                                         BlockT *trueBlk,
+                                                         BlockT *falseBlk,
+                                                         BlockT *landBlk,
+                                                         bool detail) {
+  errs() << "head = BB" << headBlk->getNumber()
+         << " size = " << headBlk->size();
+  if (detail) {
+    errs() << "\n";
+    headBlk->print(errs());
+    errs() << "\n";
+  }
+
+  if (trueBlk) {
+    errs() << ", true = BB" << trueBlk->getNumber() << " size = "
+           << trueBlk->size() << " numPred = " << trueBlk->pred_size();
+    if (detail) {
+      errs() << "\n";
+      trueBlk->print(errs());
+      errs() << "\n";
+    }
+  }
+  if (falseBlk) {
+    errs() << ", false = BB" << falseBlk->getNumber() << " size = "
+           << falseBlk->size() << " numPred = " << falseBlk->pred_size();
+    if (detail) {
+      errs() << "\n";
+      falseBlk->print(errs());
+      errs() << "\n";
+    }
+  }
+  if (landBlk) {
+    errs() << ", land = BB" << landBlk->getNumber() << " size = "
+           << landBlk->size() << " numPred = " << landBlk->pred_size();
+    if (detail) {
+      errs() << "\n";
+      landBlk->print(errs());
+      errs() << "\n";
+    }
+  }
+
+    errs() << "\n";
+} //showImproveSimpleJumpintoIf
+
+template<class PassT>
+int CFGStructurizer<PassT>::improveSimpleJumpintoIf(BlockT *headBlk,
+                                                    BlockT *trueBlk,
+                                                    BlockT *falseBlk,
+                                                    BlockT **plandBlk) {
+  bool migrateTrue = false;
+  bool migrateFalse = false;
+
+  BlockT *landBlk = *plandBlk;
+
+  assert((trueBlk == NULL || trueBlk->succ_size() <= 1)
+         && (falseBlk == NULL || falseBlk->succ_size() <= 1));
+
+  if (trueBlk == falseBlk) {
+    return 0;
+  }
+
+  migrateTrue = needMigrateBlock(trueBlk);
+  migrateFalse = needMigrateBlock(falseBlk);
+
+  if (!migrateTrue && !migrateFalse) {
+    return 0;
+  }
+
+  // If we need to migrate either trueBlk and falseBlk, migrate the rest that
+  // have more than one predecessors.  without doing this, its predecessor
+  // rather than headBlk will have undefined value in initReg.
+  if (!migrateTrue && trueBlk && trueBlk->pred_size() > 1) {
+    migrateTrue = true;
+  }
+  if (!migrateFalse && falseBlk && falseBlk->pred_size() > 1) {
+    migrateFalse = true;
+  }
+
+  if (DEBUGME) {
+    errs() << "before improveSimpleJumpintoIf: ";
+    showImproveSimpleJumpintoIf(headBlk, trueBlk, falseBlk, landBlk, 0);
+  }
+
+  // org: headBlk => if () {trueBlk} else {falseBlk} => landBlk
+  //
+  // new: headBlk => if () {initReg = 1; org trueBlk branch} else
+  //      {initReg = 0; org falseBlk branch }
+  //      => landBlk => if (initReg) {org trueBlk} else {org falseBlk}
+  //      => org landBlk
+  //      if landBlk->pred_size() > 2, put the about if-else inside
+  //      if (initReg !=2) {...}
+  //
+  // add initReg = initVal to headBlk
+
+  const TargetRegisterClass * I32RC = TRI->getCFGStructurizerRegClass(MVT::i32);
+  unsigned initReg =
+    funcRep->getRegInfo().createVirtualRegister(I32RC);
+  if (!migrateTrue || !migrateFalse) {
+    int initVal = migrateTrue ? 0 : 1;
+    CFGTraits::insertAssignInstrBefore(headBlk, passRep, initReg, initVal);
+  }
+
+  int numNewBlk = 0;
+
+  if (landBlk == NULL) {
+    landBlk = funcRep->CreateMachineBasicBlock();
+    funcRep->push_back(landBlk);  //insert to function
+
+    if (trueBlk) {
+      trueBlk->addSuccessor(landBlk);
+    } else {
+      headBlk->addSuccessor(landBlk);
+    }
+
+    if (falseBlk) {
+      falseBlk->addSuccessor(landBlk);
+    } else {
+      headBlk->addSuccessor(landBlk);
+    }
+
+    numNewBlk ++;
+  }
+
+  bool landBlkHasOtherPred = (landBlk->pred_size() > 2);
+
+  //insert AMDGPU::ENDIF to avoid special case "input landBlk == NULL"
+  typename BlockT::iterator insertPos =
+    CFGTraits::getInstrPos
+    (landBlk, CFGTraits::insertInstrBefore(landBlk, AMDGPU::ENDIF, passRep));
+
+  if (landBlkHasOtherPred) {
+    unsigned immReg =
+      funcRep->getRegInfo().createVirtualRegister(I32RC);
+    CFGTraits::insertAssignInstrBefore(insertPos, passRep, immReg, 2);
+    unsigned cmpResReg =
+      funcRep->getRegInfo().createVirtualRegister(I32RC);
+
+    CFGTraits::insertCompareInstrBefore(landBlk, insertPos, passRep, cmpResReg,
+                                        initReg, immReg);
+    CFGTraits::insertCondBranchBefore(landBlk, insertPos,
+                                      AMDGPU::IF_PREDICATE_SET, passRep,
+                                      cmpResReg, DebugLoc());
+  }
+
+  CFGTraits::insertCondBranchBefore(landBlk, insertPos, AMDGPU::IF_PREDICATE_SET,
+                                    passRep, initReg, DebugLoc());
+
+  if (migrateTrue) {
+    migrateInstruction(trueBlk, landBlk, insertPos);
+    // need to uncondionally insert the assignment to ensure a path from its
+    // predecessor rather than headBlk has valid value in initReg if
+    // (initVal != 1).
+    CFGTraits::insertAssignInstrBefore(trueBlk, passRep, initReg, 1);
+  }
+  CFGTraits::insertInstrBefore(insertPos, AMDGPU::ELSE, passRep);
+
+  if (migrateFalse) {
+    migrateInstruction(falseBlk, landBlk, insertPos);
+    // need to uncondionally insert the assignment to ensure a path from its
+    // predecessor rather than headBlk has valid value in initReg if
+    // (initVal != 0)
+    CFGTraits::insertAssignInstrBefore(falseBlk, passRep, initReg, 0);
+  }
+
+  if (landBlkHasOtherPred) {
+    // add endif
+    CFGTraits::insertInstrBefore(insertPos, AMDGPU::ENDIF, passRep);
+
+    // put initReg = 2 to other predecessors of landBlk
+    for (typename BlockT::pred_iterator predIter = landBlk->pred_begin(),
+         predIterEnd = landBlk->pred_end(); predIter != predIterEnd;
+         ++predIter) {
+      BlockT *curBlk = *predIter;
+      if (curBlk != trueBlk && curBlk != falseBlk) {
+        CFGTraits::insertAssignInstrBefore(curBlk, passRep, initReg, 2);
+      }
+    } //for
+  }
+  if (DEBUGME) {
+    errs() << "result from improveSimpleJumpintoIf: ";
+    showImproveSimpleJumpintoIf(headBlk, trueBlk, falseBlk, landBlk, 0);
+  }
+
+  // update landBlk
+  *plandBlk = landBlk;
+
+  return numNewBlk;
+} //improveSimpleJumpintoIf
+
+template<class PassT>
+void CFGStructurizer<PassT>::handleLoopbreak(BlockT *exitingBlk,
+                                              LoopT *exitingLoop,
+                                             BlockT *exitBlk,
+                                              LoopT *exitLoop,
+                                             BlockT *landBlk) {
+  if (DEBUGME) {
+    errs() << "Trying to break loop-depth = " << getLoopDepth(exitLoop)
+           << " from loop-depth = " << getLoopDepth(exitingLoop) << "\n";
+  }
+  const TargetRegisterClass * I32RC = TRI->getCFGStructurizerRegClass(MVT::i32);
+
+  RegiT initReg = INVALIDREGNUM;
+  if (exitingLoop != exitLoop) {
+    initReg = static_cast<int>
+      (funcRep->getRegInfo().createVirtualRegister(I32RC));
+    assert(initReg != INVALIDREGNUM);
+    addLoopBreakInitReg(exitLoop, initReg);
+    while (exitingLoop != exitLoop && exitingLoop) {
+      addLoopBreakOnReg(exitingLoop, initReg);
+      exitingLoop = exitingLoop->getParentLoop();
+    }
+    assert(exitingLoop == exitLoop);
+  }
+
+  mergeLoopbreakBlock(exitingBlk, exitBlk, landBlk, initReg);
+
+} //handleLoopbreak
+
+template<class PassT>
+void CFGStructurizer<PassT>::handleLoopcontBlock(BlockT *contingBlk,
+                                                  LoopT *contingLoop,
+                                                 BlockT *contBlk,
+                                                  LoopT *contLoop) {
+  if (DEBUGME) {
+    errs() << "loopcontPattern cont = BB" << contingBlk->getNumber()
+           << " header = BB" << contBlk->getNumber() << "\n";
+
+    errs() << "Trying to continue loop-depth = "
+           << getLoopDepth(contLoop)
+           << " from loop-depth = " << getLoopDepth(contingLoop) << "\n";
+  }
+
+  RegiT initReg = INVALIDREGNUM;
+  const TargetRegisterClass * I32RC = TRI->getCFGStructurizerRegClass(MVT::i32);
+  if (contingLoop != contLoop) {
+    initReg = static_cast<int>
+      (funcRep->getRegInfo().createVirtualRegister(I32RC));
+    assert(initReg != INVALIDREGNUM);
+    addLoopContInitReg(contLoop, initReg);
+    while (contingLoop && contingLoop->getParentLoop() != contLoop) {
+      addLoopBreakOnReg(contingLoop, initReg);  //not addLoopContOnReg
+      contingLoop = contingLoop->getParentLoop();
+    }
+    assert(contingLoop && contingLoop->getParentLoop() == contLoop);
+    addLoopContOnReg(contingLoop, initReg);
+  }
+
+  settleLoopcontBlock(contingBlk, contBlk, initReg);
+} //handleLoopcontBlock
+
+template<class PassT>
+void CFGStructurizer<PassT>::mergeSerialBlock(BlockT *dstBlk, BlockT *srcBlk) {
+  if (DEBUGME) {
+    errs() << "serialPattern BB" << dstBlk->getNumber()
+           << " <= BB" << srcBlk->getNumber() << "\n";
+  }
+  dstBlk->splice(dstBlk->end(), srcBlk, srcBlk->begin(), srcBlk->end());
+
+  dstBlk->removeSuccessor(srcBlk);
+  CFGTraits::cloneSuccessorList(dstBlk, srcBlk);
+
+  removeSuccessor(srcBlk);
+  retireBlock(dstBlk, srcBlk);
+} //mergeSerialBlock
+
+template<class PassT>
+void CFGStructurizer<PassT>::mergeIfthenelseBlock(InstrT *branchInstr,
+                                                  BlockT *curBlk,
+                                                  BlockT *trueBlk,
+                                                  BlockT *falseBlk,
+                                                  BlockT *landBlk) {
+  if (DEBUGME) {
+    errs() << "ifPattern BB" << curBlk->getNumber();
+    errs() << "{  ";
+    if (trueBlk) {
+      errs() << "BB" << trueBlk->getNumber();
+    }
+    errs() << "  } else ";
+    errs() << "{  ";
+    if (falseBlk) {
+      errs() << "BB" << falseBlk->getNumber();
+    }
+    errs() << "  }\n ";
+    errs() << "landBlock: ";
+    if (landBlk == NULL) {
+      errs() << "NULL";
+    } else {
+      errs() << "BB" << landBlk->getNumber();
+    }
+    errs() << "\n";
+  }
+
+  int oldOpcode = branchInstr->getOpcode();
+  DebugLoc branchDL = branchInstr->getDebugLoc();
+
+//    transform to
+//    if cond
+//       trueBlk
+//    else
+//       falseBlk
+//    endif
+//    landBlk
+
+  typename BlockT::iterator branchInstrPos =
+    CFGTraits::getInstrPos(curBlk, branchInstr);
+  CFGTraits::insertCondBranchBefore(branchInstrPos,
+                                    CFGTraits::getBranchNzeroOpcode(oldOpcode),
+                                    passRep,
+                                    branchDL);
+
+  if (trueBlk) {
+    curBlk->splice(branchInstrPos, trueBlk, trueBlk->begin(), trueBlk->end());
+    curBlk->removeSuccessor(trueBlk);
+    if (landBlk && trueBlk->succ_size()!=0) {
+      trueBlk->removeSuccessor(landBlk);
+    }
+    retireBlock(curBlk, trueBlk);
+  }
+  CFGTraits::insertInstrBefore(branchInstrPos, AMDGPU::ELSE, passRep);
+
+  if (falseBlk) {
+    curBlk->splice(branchInstrPos, falseBlk, falseBlk->begin(),
+                   falseBlk->end());
+    curBlk->removeSuccessor(falseBlk);
+    if (landBlk && falseBlk->succ_size() != 0) {
+      falseBlk->removeSuccessor(landBlk);
+    }
+    retireBlock(curBlk, falseBlk);
+  }
+  CFGTraits::insertInstrBefore(branchInstrPos, AMDGPU::ENDIF, passRep);
+
+  branchInstr->eraseFromParent();
+
+  if (landBlk && trueBlk && falseBlk) {
+    curBlk->addSuccessor(landBlk);
+  }
+
+} //mergeIfthenelseBlock
+
+template<class PassT>
+void CFGStructurizer<PassT>::mergeLooplandBlock(BlockT *dstBlk,
+                                                LoopLandInfo *loopLand) {
+  BlockT *landBlk = loopLand->landBlk;
+
+  if (DEBUGME) {
+    errs() << "loopPattern header = BB" << dstBlk->getNumber()
+           << " land = BB" << landBlk->getNumber() << "\n";
+  }
+
+  // Loop contInitRegs are init at the beginning of the loop.
+  for (typename std::set<RegiT>::const_iterator iter =
+         loopLand->contInitRegs.begin(),
+       iterEnd = loopLand->contInitRegs.end(); iter != iterEnd; ++iter) {
+    CFGTraits::insertAssignInstrBefore(dstBlk, passRep, *iter, 0);
+  }
+
+  /* we last inserterd the DebugLoc in the
+   * BREAK_LOGICALZ_i32 or AMDGPU::BREAK_LOGICALNZ statement in the current dstBlk.
+   * search for the DebugLoc in the that statement.
+   * if not found, we have to insert the empty/default DebugLoc */
+  InstrT *loopBreakInstr = CFGTraits::getLoopBreakInstr(dstBlk);
+  DebugLoc DLBreak = (loopBreakInstr) ? loopBreakInstr->getDebugLoc() : DebugLoc();
+
+  CFGTraits::insertInstrBefore(dstBlk, AMDGPU::WHILELOOP, passRep, DLBreak);
+  // Loop breakInitRegs are init before entering the loop.
+  for (typename std::set<RegiT>::const_iterator iter =
+         loopLand->breakInitRegs.begin(),
+       iterEnd = loopLand->breakInitRegs.end(); iter != iterEnd; ++iter) {
+    CFGTraits::insertAssignInstrBefore(dstBlk, passRep, *iter, 0);
+  }
+  // Loop endbranchInitRegs are init before entering the loop.
+  for (typename std::set<RegiT>::const_iterator iter =
+         loopLand->endbranchInitRegs.begin(),
+       iterEnd = loopLand->endbranchInitRegs.end(); iter != iterEnd; ++iter) {
+    CFGTraits::insertAssignInstrBefore(dstBlk, passRep, *iter, 0);
+  }
+
+  /* we last inserterd the DebugLoc in the continue statement in the current dstBlk
+   * search for the DebugLoc in the continue statement.
+   * if not found, we have to insert the empty/default DebugLoc */
+  InstrT *continueInstr = CFGTraits::getContinueInstr(dstBlk);
+  DebugLoc DLContinue = (continueInstr) ? continueInstr->getDebugLoc() : DebugLoc();
+
+  CFGTraits::insertInstrEnd(dstBlk, AMDGPU::ENDLOOP, passRep, DLContinue);
+  // Loop breakOnRegs are check after the ENDLOOP: break the loop outside this
+  // loop.
+  for (typename std::set<RegiT>::const_iterator iter =
+         loopLand->breakOnRegs.begin(),
+       iterEnd = loopLand->breakOnRegs.end(); iter != iterEnd; ++iter) {
+    CFGTraits::insertCondBranchEnd(dstBlk, AMDGPU::PREDICATED_BREAK, passRep,
+                                   *iter);
+  }
+
+  // Loop contOnRegs are check after the ENDLOOP: cont the loop outside this
+  // loop.
+  for (std::set<RegiT>::const_iterator iter = loopLand->contOnRegs.begin(),
+       iterEnd = loopLand->contOnRegs.end(); iter != iterEnd; ++iter) {
+    CFGTraits::insertCondBranchEnd(dstBlk, AMDGPU::CONTINUE_LOGICALNZ_i32,
+                                   passRep, *iter);
+  }
+
+  dstBlk->splice(dstBlk->end(), landBlk, landBlk->begin(), landBlk->end());
+
+  for (typename BlockT::succ_iterator iter = landBlk->succ_begin(),
+       iterEnd = landBlk->succ_end(); iter != iterEnd; ++iter) {
+    dstBlk->addSuccessor(*iter);  // *iter's predecessor is also taken care of.
+  }
+
+  removeSuccessor(landBlk);
+  retireBlock(dstBlk, landBlk);
+} //mergeLooplandBlock
+
+template<class PassT>
+void CFGStructurizer<PassT>::reversePredicateSetter(typename BlockT::iterator I) {
+  while (I--) {
+    if (I->getOpcode() == AMDGPU::PRED_X) {
+      switch (static_cast<MachineInstr *>(I)->getOperand(2).getImm()) {
+      case OPCODE_IS_ZERO_INT:
+        static_cast<MachineInstr *>(I)->getOperand(2).setImm(OPCODE_IS_NOT_ZERO_INT);
+        return;
+      case OPCODE_IS_NOT_ZERO_INT:
+        static_cast<MachineInstr *>(I)->getOperand(2).setImm(OPCODE_IS_ZERO_INT);
+        return;
+      case OPCODE_IS_ZERO:
+        static_cast<MachineInstr *>(I)->getOperand(2).setImm(OPCODE_IS_NOT_ZERO);
+        return;
+      case OPCODE_IS_NOT_ZERO:
+        static_cast<MachineInstr *>(I)->getOperand(2).setImm(OPCODE_IS_ZERO);
+        return;
+      default:
+        assert(0 && "PRED_X Opcode invalid!");
+      }
+    }
+  }
+}
+
+template<class PassT>
+void CFGStructurizer<PassT>::mergeLoopbreakBlock(BlockT *exitingBlk,
+                                                 BlockT *exitBlk,
+                                                 BlockT *exitLandBlk,
+                                                 RegiT  setReg) {
+  if (DEBUGME) {
+    errs() << "loopbreakPattern exiting = BB" << exitingBlk->getNumber()
+           << " exit = BB" << exitBlk->getNumber()
+           << " land = BB" << exitLandBlk->getNumber() << "\n";
+  }
+
+  InstrT *branchInstr = CFGTraits::getLoopendBlockBranchInstr(exitingBlk);
+  assert(branchInstr && CFGTraits::isCondBranch(branchInstr));
+
+  DebugLoc DL = branchInstr->getDebugLoc();
+
+  BlockT *trueBranch = CFGTraits::getTrueBranch(branchInstr);
+
+  //    transform exitingBlk to
+  //    if ( ) {
+  //       exitBlk (if exitBlk != exitLandBlk)
+  //       setReg = 1
+  //       break
+  //    }endif
+  //    successor = {orgSuccessor(exitingBlk) - exitBlk}
+
+  typename BlockT::iterator branchInstrPos =
+    CFGTraits::getInstrPos(exitingBlk, branchInstr);
+
+  if (exitBlk == exitLandBlk && setReg == INVALIDREGNUM) {
+    //break_logical
+
+    if (trueBranch != exitBlk) {
+      reversePredicateSetter(branchInstrPos);
+    }
+    CFGTraits::insertCondBranchBefore(branchInstrPos, AMDGPU::PREDICATED_BREAK, passRep, DL);
+  } else {
+    if (trueBranch != exitBlk) {
+      reversePredicateSetter(branchInstr);
+    }
+    CFGTraits::insertCondBranchBefore(branchInstrPos, AMDGPU::PREDICATED_BREAK, passRep, DL);
+    if (exitBlk != exitLandBlk) {
+      //splice is insert-before ...
+      exitingBlk->splice(branchInstrPos, exitBlk, exitBlk->begin(),
+                         exitBlk->end());
+    }
+    if (setReg != INVALIDREGNUM) {
+      CFGTraits::insertAssignInstrBefore(branchInstrPos, passRep, setReg, 1);
+    }
+    CFGTraits::insertInstrBefore(branchInstrPos, AMDGPU::BREAK, passRep);
+  } //if_logical
+
+  //now branchInst can be erase safely
+  branchInstr->eraseFromParent();
+
+  //now take care of successors, retire blocks
+  exitingBlk->removeSuccessor(exitBlk);
+  if (exitBlk != exitLandBlk) {
+    //splice is insert-before ...
+    exitBlk->removeSuccessor(exitLandBlk);
+    retireBlock(exitingBlk, exitBlk);
+  }
+
+} //mergeLoopbreakBlock
+
+template<class PassT>
+void CFGStructurizer<PassT>::settleLoopcontBlock(BlockT *contingBlk,
+                                                 BlockT *contBlk,
+                                                 RegiT   setReg) {
+  if (DEBUGME) {
+    errs() << "settleLoopcontBlock conting = BB"
+           << contingBlk->getNumber()
+           << ", cont = BB" << contBlk->getNumber() << "\n";
+  }
+
+  InstrT *branchInstr = CFGTraits::getLoopendBlockBranchInstr(contingBlk);
+  if (branchInstr) {
+    assert(CFGTraits::isCondBranch(branchInstr));
+    typename BlockT::iterator branchInstrPos =
+      CFGTraits::getInstrPos(contingBlk, branchInstr);
+    BlockT *trueBranch = CFGTraits::getTrueBranch(branchInstr);
+    int oldOpcode = branchInstr->getOpcode();
+    DebugLoc DL = branchInstr->getDebugLoc();
+
+    //    transform contingBlk to
+    //     if () {
+    //          move instr after branchInstr
+    //          continue
+    //        or
+    //          setReg = 1
+    //          break
+    //     }endif
+    //     successor = {orgSuccessor(contingBlk) - loopHeader}
+
+    bool useContinueLogical = 
+      (setReg == INVALIDREGNUM && (&*contingBlk->rbegin()) == branchInstr);
+
+    if (useContinueLogical == false) {
+      int branchOpcode =
+        trueBranch == contBlk ? CFGTraits::getBranchNzeroOpcode(oldOpcode)
+                              : CFGTraits::getBranchZeroOpcode(oldOpcode);
+
+      CFGTraits::insertCondBranchBefore(branchInstrPos, branchOpcode, passRep, DL);
+
+      if (setReg != INVALIDREGNUM) {
+        CFGTraits::insertAssignInstrBefore(branchInstrPos, passRep, setReg, 1);
+        // insertEnd to ensure phi-moves, if exist, go before the continue-instr.
+        CFGTraits::insertInstrEnd(contingBlk, AMDGPU::BREAK, passRep, DL);
+      } else {
+        // insertEnd to ensure phi-moves, if exist, go before the continue-instr.
+        CFGTraits::insertInstrEnd(contingBlk, AMDGPU::CONTINUE, passRep, DL);
+      }
+
+      CFGTraits::insertInstrEnd(contingBlk, AMDGPU::ENDIF, passRep, DL);
+    } else {
+      int branchOpcode =
+        trueBranch == contBlk ? CFGTraits::getContinueNzeroOpcode(oldOpcode)
+                              : CFGTraits::getContinueZeroOpcode(oldOpcode);
+
+      CFGTraits::insertCondBranchBefore(branchInstrPos, branchOpcode, passRep, DL);
+    }
+
+    branchInstr->eraseFromParent();
+  } else {
+    // if we've arrived here then we've already erased the branch instruction
+    // travel back up the basic block to see the last reference of our debug location
+    // we've just inserted that reference here so it should be representative
+    if (setReg != INVALIDREGNUM) {
+      CFGTraits::insertAssignInstrBefore(contingBlk, passRep, setReg, 1);
+      // insertEnd to ensure phi-moves, if exist, go before the continue-instr.
+      CFGTraits::insertInstrEnd(contingBlk, AMDGPU::BREAK, passRep, CFGTraits::getLastDebugLocInBB(contingBlk));
+    } else {
+      // insertEnd to ensure phi-moves, if exist, go before the continue-instr.
+      CFGTraits::insertInstrEnd(contingBlk, AMDGPU::CONTINUE, passRep, CFGTraits::getLastDebugLocInBB(contingBlk));
+    }
+  } //else
+
+} //settleLoopcontBlock
+
+// BBs in exitBlkSet are determined as in break-path for loopRep,
+// before we can put code for BBs as inside loop-body for loopRep
+// check whether those BBs are determined as cont-BB for parentLoopRep
+// earlier.
+// If so, generate a new BB newBlk
+//    (1) set newBlk common successor of BBs in exitBlkSet
+//    (2) change the continue-instr in BBs in exitBlkSet to break-instr
+//    (3) generate continue-instr in newBlk
+//
+template<class PassT>
+typename CFGStructurizer<PassT>::BlockT *
+CFGStructurizer<PassT>::relocateLoopcontBlock(LoopT *parentLoopRep,
+                                              LoopT *loopRep,
+                                              std::set<BlockT *> &exitBlkSet,
+                                              BlockT *exitLandBlk) {
+  std::set<BlockT *> endBlkSet;
+
+
+
+  for (typename std::set<BlockT *>::const_iterator iter = exitBlkSet.begin(),
+       iterEnd = exitBlkSet.end();
+       iter != iterEnd; ++iter) {
+    BlockT *exitBlk = *iter;
+    BlockT *endBlk = singlePathEnd(exitBlk, exitLandBlk);
+
+    if (endBlk == NULL || CFGTraits::getContinueInstr(endBlk) == NULL)
+      return NULL;
+
+    endBlkSet.insert(endBlk);
+  }
+
+  BlockT *newBlk = funcRep->CreateMachineBasicBlock();
+  funcRep->push_back(newBlk);  //insert to function
+  CFGTraits::insertInstrEnd(newBlk, AMDGPU::CONTINUE, passRep);
+  SHOWNEWBLK(newBlk, "New continue block: ");
+
+  for (typename std::set<BlockT*>::const_iterator iter = endBlkSet.begin(),
+       iterEnd = endBlkSet.end();
+       iter != iterEnd; ++iter) {
+      BlockT *endBlk = *iter;
+      InstrT *contInstr = CFGTraits::getContinueInstr(endBlk);
+      if (contInstr) {
+        contInstr->eraseFromParent();
+      }
+      endBlk->addSuccessor(newBlk);
+      if (DEBUGME) {
+        errs() << "Add new continue Block to BB"
+               << endBlk->getNumber() << " successors\n";
+      }
+  }
+
+  return newBlk;
+} //relocateLoopcontBlock
+
+
+// LoopEndbranchBlock is a BB created by the CFGStructurizer to use as
+// LoopLandBlock. This BB branch on the loop endBranchInit register to the
+// pathes corresponding to the loop exiting branches.
+
+template<class PassT>
+typename CFGStructurizer<PassT>::BlockT *
+CFGStructurizer<PassT>::addLoopEndbranchBlock(LoopT *loopRep,
+                                              BlockTSmallerVector &exitingBlks,
+                                              BlockTSmallerVector &exitBlks) {
+  const AMDGPUInstrInfo *tii =
+             static_cast<const AMDGPUInstrInfo *>(passRep->getTargetInstrInfo());
+  const TargetRegisterClass * I32RC = TRI->getCFGStructurizerRegClass(MVT::i32);
+
+  RegiT endBranchReg = static_cast<int>
+    (funcRep->getRegInfo().createVirtualRegister(I32RC));
+  assert(endBranchReg >= 0);
+
+  // reg = 0 before entering the loop
+  addLoopEndbranchInitReg(loopRep, endBranchReg);
+
+  uint32_t numBlks = static_cast<uint32_t>(exitingBlks.size());
+  assert(numBlks >=2 && numBlks == exitBlks.size());
+
+  BlockT *preExitingBlk = exitingBlks[0];
+  BlockT *preExitBlk = exitBlks[0];
+  BlockT *preBranchBlk = funcRep->CreateMachineBasicBlock();
+  funcRep->push_back(preBranchBlk);  //insert to function
+  SHOWNEWBLK(preBranchBlk, "New loopEndbranch block: ");
+
+  BlockT *newLandBlk = preBranchBlk;
+
+      CFGTraits::replaceInstrUseOfBlockWith(preExitingBlk, preExitBlk,
+        newLandBlk);
+  preExitingBlk->removeSuccessor(preExitBlk);
+  preExitingBlk->addSuccessor(newLandBlk);
+
+  //it is redundant to add reg = 0 to exitingBlks[0]
+
+  // For 1..n th exiting path (the last iteration handles two pathes) create the
+  // branch to the previous path and the current path.
+  for (uint32_t i = 1; i < numBlks; ++i) {
+    BlockT *curExitingBlk = exitingBlks[i];
+    BlockT *curExitBlk = exitBlks[i];
+    BlockT *curBranchBlk;
+
+    if (i == numBlks - 1) {
+      curBranchBlk = curExitBlk;
+    } else {
+      curBranchBlk = funcRep->CreateMachineBasicBlock();
+      funcRep->push_back(curBranchBlk);  //insert to function
+      SHOWNEWBLK(curBranchBlk, "New loopEndbranch block: ");
+    }
+
+    // Add reg = i to exitingBlks[i].
+    CFGTraits::insertAssignInstrBefore(curExitingBlk, passRep,
+                                       endBranchReg, i);
+
+    // Remove the edge (exitingBlks[i] exitBlks[i]) add new edge
+    // (exitingBlks[i], newLandBlk).
+    CFGTraits::replaceInstrUseOfBlockWith(curExitingBlk, curExitBlk,
+                                          newLandBlk);
+    curExitingBlk->removeSuccessor(curExitBlk);
+    curExitingBlk->addSuccessor(newLandBlk);
+
+    // add to preBranchBlk the branch instruction:
+    // if (endBranchReg == preVal)
+    //    preExitBlk
+    // else
+    //    curBranchBlk
+    //
+    // preValReg = i - 1
+
+  DebugLoc DL;
+  RegiT preValReg = static_cast<int>
+    (funcRep->getRegInfo().createVirtualRegister(I32RC));
+
+  preBranchBlk->insert(preBranchBlk->begin(),
+                       tii->getMovImmInstr(preBranchBlk->getParent(), preValReg,
+                       i - 1));
+
+  // condResReg = (endBranchReg == preValReg)
+    RegiT condResReg = static_cast<int>
+      (funcRep->getRegInfo().createVirtualRegister(I32RC));
+    BuildMI(preBranchBlk, DL, tii->get(tii->getIEQOpcode()), condResReg)
+      .addReg(endBranchReg).addReg(preValReg);
+
+    BuildMI(preBranchBlk, DL, tii->get(AMDGPU::BRANCH_COND_i32))
+      .addMBB(preExitBlk).addReg(condResReg);
+
+    preBranchBlk->addSuccessor(preExitBlk);
+    preBranchBlk->addSuccessor(curBranchBlk);
+
+    // Update preExitingBlk, preExitBlk, preBranchBlk.
+    preExitingBlk = curExitingBlk;
+    preExitBlk = curExitBlk;
+    preBranchBlk = curBranchBlk;
+
+  }  //end for 1 .. n blocks
+
+  return newLandBlk;
+} //addLoopEndbranchBlock
+
+template<class PassT>
+typename CFGStructurizer<PassT>::PathToKind
+CFGStructurizer<PassT>::singlePathTo(BlockT *srcBlk, BlockT *dstBlk,
+                                     bool allowSideEntry) {
+  assert(dstBlk);
+
+  if (srcBlk == dstBlk) {
+    return SinglePath_InPath;
+  }
+
+  while (srcBlk && srcBlk->succ_size() == 1) {
+    srcBlk = *srcBlk->succ_begin();
+    if (srcBlk == dstBlk) {
+      return SinglePath_InPath;
+    }
+
+    if (!allowSideEntry && srcBlk->pred_size() > 1) {
+      return Not_SinglePath;
+    }
+  }
+
+  if (srcBlk && srcBlk->succ_size()==0) {
+    return SinglePath_NotInPath;
+  }
+
+  return Not_SinglePath;
+} //singlePathTo
+
+// If there is a single path from srcBlk to dstBlk, return the last block before
+// dstBlk If there is a single path from srcBlk->end without dstBlk, return the
+// last block in the path Otherwise, return NULL
+template<class PassT>
+typename CFGStructurizer<PassT>::BlockT *
+CFGStructurizer<PassT>::singlePathEnd(BlockT *srcBlk, BlockT *dstBlk,
+                                      bool allowSideEntry) {
+  assert(dstBlk);
+
+  if (srcBlk == dstBlk) {
+    return srcBlk;
+  }
+
+  if (srcBlk->succ_size() == 0) {
+    return srcBlk;
+  }
+
+  while (srcBlk && srcBlk->succ_size() == 1) {
+    BlockT *preBlk = srcBlk;
+
+    srcBlk = *srcBlk->succ_begin();
+    if (srcBlk == NULL) {
+      return preBlk;
+    }
+
+    if (!allowSideEntry && srcBlk->pred_size() > 1) {
+      return NULL;
+    }
+  }
+
+  if (srcBlk && srcBlk->succ_size()==0) {
+    return srcBlk;
+  }
+
+  return NULL;
+
+} //singlePathEnd
+
+template<class PassT>
+int CFGStructurizer<PassT>::cloneOnSideEntryTo(BlockT *preBlk, BlockT *srcBlk,
+                                               BlockT *dstBlk) {
+  int cloned = 0;
+  assert(preBlk->isSuccessor(srcBlk));
+  while (srcBlk && srcBlk != dstBlk) {
+    assert(srcBlk->succ_size() == 1);
+    if (srcBlk->pred_size() > 1) {
+      srcBlk = cloneBlockForPredecessor(srcBlk, preBlk);
+      ++cloned;
+    }
+
+    preBlk = srcBlk;
+    srcBlk = *srcBlk->succ_begin();
+  }
+
+  return cloned;
+} //cloneOnSideEntryTo
+
+template<class PassT>
+typename CFGStructurizer<PassT>::BlockT *
+CFGStructurizer<PassT>::cloneBlockForPredecessor(BlockT *curBlk,
+                                                 BlockT *predBlk) {
+  assert(predBlk->isSuccessor(curBlk) &&
+         "succBlk is not a prececessor of curBlk");
+
+  BlockT *cloneBlk = CFGTraits::clone(curBlk);  //clone instructions
+  CFGTraits::replaceInstrUseOfBlockWith(predBlk, curBlk, cloneBlk);
+  //srcBlk, oldBlk, newBlk
+
+  predBlk->removeSuccessor(curBlk);
+  predBlk->addSuccessor(cloneBlk);
+
+  // add all successor to cloneBlk
+  CFGTraits::cloneSuccessorList(cloneBlk, curBlk);
+
+  numClonedInstr += curBlk->size();
+
+  if (DEBUGME) {
+    errs() << "Cloned block: " << "BB"
+           << curBlk->getNumber() << "size " << curBlk->size() << "\n";
+  }
+
+  SHOWNEWBLK(cloneBlk, "result of Cloned block: ");
+
+  return cloneBlk;
+} //cloneBlockForPredecessor
+
+template<class PassT>
+typename CFGStructurizer<PassT>::BlockT *
+CFGStructurizer<PassT>::exitingBlock2ExitBlock(LoopT *loopRep,
+                                               BlockT *exitingBlk) {
+  BlockT *exitBlk = NULL;
+
+  for (typename BlockT::succ_iterator iterSucc = exitingBlk->succ_begin(),
+       iterSuccEnd = exitingBlk->succ_end();
+       iterSucc != iterSuccEnd; ++iterSucc) {
+    BlockT *curBlk = *iterSucc;
+    if (!loopRep->contains(curBlk)) {
+      assert(exitBlk == NULL);
+      exitBlk = curBlk;
+    }
+  }
+
+  assert(exitBlk != NULL);
+
+  return exitBlk;
+} //exitingBlock2ExitBlock
+
+template<class PassT>
+void CFGStructurizer<PassT>::migrateInstruction(BlockT *srcBlk,
+                                                BlockT *dstBlk,
+                                                InstrIterator insertPos) {
+  InstrIterator spliceEnd;
+  //look for the input branchinstr, not the AMDGPU branchinstr
+  InstrT *branchInstr = CFGTraits::getNormalBlockBranchInstr(srcBlk);
+  if (branchInstr == NULL) {
+    if (DEBUGME) {
+      errs() << "migrateInstruction don't see branch instr\n" ;
+    }
+    spliceEnd = srcBlk->end();
+  } else {
+    if (DEBUGME) {
+      errs() << "migrateInstruction see branch instr\n" ;
+      branchInstr->dump();
+    }
+    spliceEnd = CFGTraits::getInstrPos(srcBlk, branchInstr);
+  }
+  if (DEBUGME) {
+    errs() << "migrateInstruction before splice dstSize = " << dstBlk->size()
+      << "srcSize = " << srcBlk->size() << "\n";
+  }
+
+  //splice insert before insertPos
+  dstBlk->splice(insertPos, srcBlk, srcBlk->begin(), spliceEnd);
+
+  if (DEBUGME) {
+    errs() << "migrateInstruction after splice dstSize = " << dstBlk->size()
+      << "srcSize = " << srcBlk->size() << "\n";
+  }
+} //migrateInstruction
+
+// normalizeInfiniteLoopExit change
+//   B1:
+//        uncond_br LoopHeader
+//
+// to
+//   B1:
+//        cond_br 1 LoopHeader dummyExit
+// and return the newly added dummy exit block
+// 
+template<class PassT>
+typename CFGStructurizer<PassT>::BlockT *
+CFGStructurizer<PassT>::normalizeInfiniteLoopExit(LoopT* LoopRep) {
+  BlockT *loopHeader;
+  BlockT *loopLatch;
+  loopHeader = LoopRep->getHeader();
+  loopLatch = LoopRep->getLoopLatch();
+  BlockT *dummyExitBlk = NULL;
+  const TargetRegisterClass * I32RC = TRI->getCFGStructurizerRegClass(MVT::i32);
+  if (loopHeader!=NULL && loopLatch!=NULL) {
+    InstrT *branchInstr = CFGTraits::getLoopendBlockBranchInstr(loopLatch);
+    if (branchInstr!=NULL && CFGTraits::isUncondBranch(branchInstr)) {
+      dummyExitBlk = funcRep->CreateMachineBasicBlock();
+      funcRep->push_back(dummyExitBlk);  //insert to function
+      SHOWNEWBLK(dummyExitBlk, "DummyExitBlock to normalize infiniteLoop: ");
+
+      if (DEBUGME) errs() << "Old branch instr: " << *branchInstr << "\n";
+
+      typename BlockT::iterator insertPos =
+        CFGTraits::getInstrPos(loopLatch, branchInstr);
+      unsigned immReg =
+        funcRep->getRegInfo().createVirtualRegister(I32RC);
+      CFGTraits::insertAssignInstrBefore(insertPos, passRep, immReg, 1);
+      InstrT *newInstr = 
+        CFGTraits::insertInstrBefore(insertPos, AMDGPU::BRANCH_COND_i32, passRep);
+      MachineInstrBuilder MIB(*funcRep, newInstr);
+      MIB.addMBB(loopHeader);
+      MIB.addReg(immReg, false);
+
+      SHOWNEWINSTR(newInstr);
+
+      branchInstr->eraseFromParent();
+      loopLatch->addSuccessor(dummyExitBlk);
+    }
+  }
+
+  return dummyExitBlk;
+} //normalizeInfiniteLoopExit
+
+template<class PassT>
+void CFGStructurizer<PassT>::removeUnconditionalBranch(BlockT *srcBlk) {
+  InstrT *branchInstr;
+
+  // I saw two unconditional branch in one basic block in example
+  // test_fc_do_while_or.c need to fix the upstream on this to remove the loop.
+  while ((branchInstr = CFGTraits::getLoopendBlockBranchInstr(srcBlk))
+          && CFGTraits::isUncondBranch(branchInstr)) {
+    if (DEBUGME) {
+          errs() << "Removing unconditional branch instruction" ;
+      branchInstr->dump();
+    }
+    branchInstr->eraseFromParent();
+  }
+} //removeUnconditionalBranch
+
+template<class PassT>
+void CFGStructurizer<PassT>::removeRedundantConditionalBranch(BlockT *srcBlk) {
+  if (srcBlk->succ_size() == 2) {
+    BlockT *blk1 = *srcBlk->succ_begin();
+    BlockT *blk2 = *(++srcBlk->succ_begin());
+
+    if (blk1 == blk2) {
+      InstrT *branchInstr = CFGTraits::getNormalBlockBranchInstr(srcBlk);
+      assert(branchInstr && CFGTraits::isCondBranch(branchInstr));
+      if (DEBUGME) {
+        errs() << "Removing unneeded conditional branch instruction" ;
+        branchInstr->dump();
+      }
+      branchInstr->eraseFromParent();
+      SHOWNEWBLK(blk1, "Removing redundant successor");
+      srcBlk->removeSuccessor(blk1);
+    }
+  }
+} //removeRedundantConditionalBranch
+
+template<class PassT>
+void CFGStructurizer<PassT>::addDummyExitBlock(SmallVector<BlockT*,
+                                               DEFAULT_VEC_SLOTS> &retBlks) {
+  BlockT *dummyExitBlk = funcRep->CreateMachineBasicBlock();
+  funcRep->push_back(dummyExitBlk);  //insert to function
+  CFGTraits::insertInstrEnd(dummyExitBlk, AMDGPU::RETURN, passRep);
+
+  for (typename SmallVector<BlockT *, DEFAULT_VEC_SLOTS>::iterator iter =
+         retBlks.begin(),
+       iterEnd = retBlks.end(); iter != iterEnd; ++iter) {
+    BlockT *curBlk = *iter;
+    InstrT *curInstr = CFGTraits::getReturnInstr(curBlk);
+    if (curInstr) {
+      curInstr->eraseFromParent();
+    }
+    curBlk->addSuccessor(dummyExitBlk);
+    if (DEBUGME) {
+      errs() << "Add dummyExitBlock to BB" << curBlk->getNumber()
+             << " successors\n";
+    }
+  } //for
+
+  SHOWNEWBLK(dummyExitBlk, "DummyExitBlock: ");
+} //addDummyExitBlock
+
+template<class PassT>
+void CFGStructurizer<PassT>::removeSuccessor(BlockT *srcBlk) {
+  while (srcBlk->succ_size()) {
+    srcBlk->removeSuccessor(*srcBlk->succ_begin());
+  }
+}
+
+template<class PassT>
+void CFGStructurizer<PassT>::recordSccnum(BlockT *srcBlk, int sccNum) {
+  BlockInfo *&srcBlkInfo = blockInfoMap[srcBlk];
+
+  if (srcBlkInfo == NULL) {
+    srcBlkInfo = new BlockInfo();
+  }
+
+  srcBlkInfo->sccNum = sccNum;
+}
+
+template<class PassT>
+int CFGStructurizer<PassT>::getSCCNum(BlockT *srcBlk) {
+  BlockInfo *srcBlkInfo = blockInfoMap[srcBlk];
+  return srcBlkInfo ? srcBlkInfo->sccNum : INVALIDSCCNUM;
+}
+
+template<class PassT>
+void CFGStructurizer<PassT>::retireBlock(BlockT *dstBlk, BlockT *srcBlk) {
+  if (DEBUGME) {
+        errs() << "Retiring BB" << srcBlk->getNumber() << "\n";
+  }
+
+  BlockInfo *&srcBlkInfo = blockInfoMap[srcBlk];
+
+  if (srcBlkInfo == NULL) {
+    srcBlkInfo = new BlockInfo();
+  }
+
+  srcBlkInfo->isRetired = true;
+  assert(srcBlk->succ_size() == 0 && srcBlk->pred_size() == 0
+         && "can't retire block yet");
+}
+
+template<class PassT>
+bool CFGStructurizer<PassT>::isRetiredBlock(BlockT *srcBlk) {
+  BlockInfo *srcBlkInfo = blockInfoMap[srcBlk];
+  return (srcBlkInfo && srcBlkInfo->isRetired);
+}
+
+template<class PassT>
+bool CFGStructurizer<PassT>::isActiveLoophead(BlockT *curBlk) {
+  LoopT *loopRep = loopInfo->getLoopFor(curBlk);
+  while (loopRep && loopRep->getHeader() == curBlk) {
+    LoopLandInfo *loopLand = getLoopLandInfo(loopRep);
+
+    if(loopLand == NULL)
+      return true;
+
+    BlockT *landBlk = loopLand->landBlk;
+    assert(landBlk);
+    if (!isRetiredBlock(landBlk)) {
+      return true;
+    }
+
+    loopRep = loopRep->getParentLoop();
+  }
+
+  return false;
+} //isActiveLoophead
+
+template<class PassT>
+bool CFGStructurizer<PassT>::needMigrateBlock(BlockT *blk) {
+  const unsigned blockSizeThreshold = 30;
+  const unsigned cloneInstrThreshold = 100;
+
+  bool multiplePreds = blk && (blk->pred_size() > 1);
+
+  if(!multiplePreds)
+    return false;
+
+  unsigned blkSize = blk->size();
+  return ((blkSize > blockSizeThreshold)
+          && (blkSize * (blk->pred_size() - 1) > cloneInstrThreshold));
+} //needMigrateBlock
+
+template<class PassT>
+typename CFGStructurizer<PassT>::BlockT *
+CFGStructurizer<PassT>::recordLoopLandBlock(LoopT *loopRep, BlockT *landBlk,
+                                            BlockTSmallerVector &exitBlks,
+                                            std::set<BlockT *> &exitBlkSet) {
+  SmallVector<BlockT *, DEFAULT_VEC_SLOTS> inpathBlks;  //in exit path blocks
+
+  for (typename BlockT::pred_iterator predIter = landBlk->pred_begin(),
+       predIterEnd = landBlk->pred_end();
+       predIter != predIterEnd; ++predIter) {
+    BlockT *curBlk = *predIter;
+    if (loopRep->contains(curBlk) || exitBlkSet.count(curBlk)) {
+      inpathBlks.push_back(curBlk);
+    }
+  } //for
+
+  //if landBlk has predecessors that are not in the given loop,
+  //create a new block
+  BlockT *newLandBlk = landBlk;
+  if (inpathBlks.size() != landBlk->pred_size()) {
+    newLandBlk = funcRep->CreateMachineBasicBlock();
+    funcRep->push_back(newLandBlk);  //insert to function
+    newLandBlk->addSuccessor(landBlk);
+    for (typename SmallVector<BlockT*, DEFAULT_VEC_SLOTS>::iterator iter =
+         inpathBlks.begin(),
+         iterEnd = inpathBlks.end(); iter != iterEnd; ++iter) {
+      BlockT *curBlk = *iter;
+      CFGTraits::replaceInstrUseOfBlockWith(curBlk, landBlk, newLandBlk);
+      //srcBlk, oldBlk, newBlk
+      curBlk->removeSuccessor(landBlk);
+      curBlk->addSuccessor(newLandBlk);
+    }
+    for (size_t i = 0, tot = exitBlks.size(); i < tot; ++i) {
+      if (exitBlks[i] == landBlk) {
+        exitBlks[i] = newLandBlk;
+      }
+    }
+    SHOWNEWBLK(newLandBlk, "NewLandingBlock: ");
+  }
+
+  setLoopLandBlock(loopRep, newLandBlk);
+
+  return newLandBlk;
+} // recordLoopbreakLand
+
+template<class PassT>
+void CFGStructurizer<PassT>::setLoopLandBlock(LoopT *loopRep, BlockT *blk) {
+  LoopLandInfo *&theEntry = loopLandInfoMap[loopRep];
+
+  if (theEntry == NULL) {
+    theEntry = new LoopLandInfo();
+  }
+  assert(theEntry->landBlk == NULL);
+
+  if (blk == NULL) {
+    blk = funcRep->CreateMachineBasicBlock();
+    funcRep->push_back(blk);  //insert to function
+    SHOWNEWBLK(blk, "DummyLandingBlock for loop without break: ");
+  }
+
+  theEntry->landBlk = blk;
+
+  if (DEBUGME) {
+    errs() << "setLoopLandBlock loop-header = BB"
+           << loopRep->getHeader()->getNumber()
+           << "  landing-block = BB" << blk->getNumber() << "\n";
+  }
+} // setLoopLandBlock
+
+template<class PassT>
+void CFGStructurizer<PassT>::addLoopBreakOnReg(LoopT *loopRep, RegiT regNum) {
+  LoopLandInfo *&theEntry = loopLandInfoMap[loopRep];
+
+  if (theEntry == NULL) {
+    theEntry = new LoopLandInfo();
+  }
+
+  theEntry->breakOnRegs.insert(regNum);
+
+  if (DEBUGME) {
+    errs() << "addLoopBreakOnReg loop-header = BB"
+           << loopRep->getHeader()->getNumber()
+           << "  regNum = " << regNum << "\n";
+  }
+} // addLoopBreakOnReg
+
+template<class PassT>
+void CFGStructurizer<PassT>::addLoopContOnReg(LoopT *loopRep, RegiT regNum) {
+  LoopLandInfo *&theEntry = loopLandInfoMap[loopRep];
+
+  if (theEntry == NULL) {
+    theEntry = new LoopLandInfo();
+  }
+  theEntry->contOnRegs.insert(regNum);
+
+  if (DEBUGME) {
+    errs() << "addLoopContOnReg loop-header = BB"
+           << loopRep->getHeader()->getNumber()
+           << "  regNum = " << regNum << "\n";
+  }
+} // addLoopContOnReg
+
+template<class PassT>
+void CFGStructurizer<PassT>::addLoopBreakInitReg(LoopT *loopRep, RegiT regNum) {
+  LoopLandInfo *&theEntry = loopLandInfoMap[loopRep];
+
+  if (theEntry == NULL) {
+    theEntry = new LoopLandInfo();
+  }
+  theEntry->breakInitRegs.insert(regNum);
+
+  if (DEBUGME) {
+    errs() << "addLoopBreakInitReg loop-header = BB"
+           << loopRep->getHeader()->getNumber()
+           << "  regNum = " << regNum << "\n";
+  }
+} // addLoopBreakInitReg
+
+template<class PassT>
+void CFGStructurizer<PassT>::addLoopContInitReg(LoopT *loopRep, RegiT regNum) {
+  LoopLandInfo *&theEntry = loopLandInfoMap[loopRep];
+
+  if (theEntry == NULL) {
+    theEntry = new LoopLandInfo();
+  }
+  theEntry->contInitRegs.insert(regNum);
+
+  if (DEBUGME) {
+    errs() << "addLoopContInitReg loop-header = BB"
+           << loopRep->getHeader()->getNumber()
+           << "  regNum = " << regNum << "\n";
+  }
+} // addLoopContInitReg
+
+template<class PassT>
+void CFGStructurizer<PassT>::addLoopEndbranchInitReg(LoopT *loopRep,
+                                                     RegiT regNum) {
+  LoopLandInfo *&theEntry = loopLandInfoMap[loopRep];
+
+  if (theEntry == NULL) {
+    theEntry = new LoopLandInfo();
+  }
+  theEntry->endbranchInitRegs.insert(regNum);
+
+  if (DEBUGME) {
+        errs() << "addLoopEndbranchInitReg loop-header = BB"
+      << loopRep->getHeader()->getNumber()
+      << "  regNum = " << regNum << "\n";
+  }
+} // addLoopEndbranchInitReg
+
+template<class PassT>
+typename CFGStructurizer<PassT>::LoopLandInfo *
+CFGStructurizer<PassT>::getLoopLandInfo(LoopT *loopRep) {
+  LoopLandInfo *&theEntry = loopLandInfoMap[loopRep];
+
+  return theEntry;
+} // getLoopLandInfo
+
+template<class PassT>
+typename CFGStructurizer<PassT>::BlockT *
+CFGStructurizer<PassT>::getLoopLandBlock(LoopT *loopRep) {
+  LoopLandInfo *&theEntry = loopLandInfoMap[loopRep];
+
+  return theEntry ? theEntry->landBlk : NULL;
+} // getLoopLandBlock
+
+
+template<class PassT>
+bool CFGStructurizer<PassT>::hasBackEdge(BlockT *curBlk) {
+  LoopT *loopRep = loopInfo->getLoopFor(curBlk);
+  if (loopRep == NULL)
+    return false;
+
+  BlockT *loopHeader = loopRep->getHeader();
+
+  return curBlk->isSuccessor(loopHeader);
+
+} //hasBackEdge
+
+template<class PassT>
+unsigned CFGStructurizer<PassT>::getLoopDepth(LoopT *loopRep) {
+  return loopRep ? loopRep->getLoopDepth() : 0;
+} //getLoopDepth
+
+template<class PassT>
+int CFGStructurizer<PassT>::countActiveBlock
+(typename SmallVector<BlockT*, DEFAULT_VEC_SLOTS>::const_iterator iterStart,
+ typename SmallVector<BlockT*, DEFAULT_VEC_SLOTS>::const_iterator iterEnd) {
+  int count = 0;
+  while (iterStart != iterEnd) {
+    if (!isRetiredBlock(*iterStart)) {
+      ++count;
+    }
+    ++iterStart;
+  }
+
+  return count;
+} //countActiveBlock
+
+// This is work around solution for findNearestCommonDominator not avaiable to
+// post dom a proper fix should go to Dominators.h.
+
+template<class PassT>
+typename CFGStructurizer<PassT>::BlockT*
+CFGStructurizer<PassT>::findNearestCommonPostDom(BlockT *blk1, BlockT *blk2) {
+
+  if (postDomTree->dominates(blk1, blk2)) {
+    return blk1;
+  }
+  if (postDomTree->dominates(blk2, blk1)) {
+    return blk2;
+  }
+
+  DomTreeNodeT *node1 = postDomTree->getNode(blk1);
+  DomTreeNodeT *node2 = postDomTree->getNode(blk2);
+
+  // Handle newly cloned node.
+  if (node1 == NULL && blk1->succ_size() == 1) {
+    return findNearestCommonPostDom(*blk1->succ_begin(), blk2);
+  }
+  if (node2 == NULL && blk2->succ_size() == 1) {
+    return findNearestCommonPostDom(blk1, *blk2->succ_begin());
+  }
+
+  if (node1 == NULL || node2 == NULL) {
+    return NULL;
+  }
+
+  node1 = node1->getIDom();
+  while (node1) {
+    if (postDomTree->dominates(node1, node2)) {
+      return node1->getBlock();
+    }
+    node1 = node1->getIDom();
+  }
+
+  return NULL;
+}
+
+template<class PassT>
+typename CFGStructurizer<PassT>::BlockT *
+CFGStructurizer<PassT>::findNearestCommonPostDom
+(typename std::set<BlockT *> &blks) {
+  BlockT *commonDom;
+  typename std::set<BlockT *>::const_iterator iter = blks.begin();
+  typename std::set<BlockT *>::const_iterator iterEnd = blks.end();
+  for (commonDom = *iter; iter != iterEnd && commonDom != NULL; ++iter) {
+    BlockT *curBlk = *iter;
+    if (curBlk != commonDom) {
+      commonDom = findNearestCommonPostDom(curBlk, commonDom);
+    }
+  }
+
+  if (DEBUGME) {
+    errs() << "Common post dominator for exit blocks is ";
+    if (commonDom) {
+          errs() << "BB" << commonDom->getNumber() << "\n";
+    } else {
+      errs() << "NULL\n";
+    }
+  }
+
+  return commonDom;
+} //findNearestCommonPostDom
+
+} //end namespace llvm
+
+//todo: move-end
+
+
+//===----------------------------------------------------------------------===//
+//
+// CFGStructurizer for AMDGPU
+//
+//===----------------------------------------------------------------------===//
+
+
+using namespace llvmCFGStruct;
+
+namespace llvm {
+class AMDGPUCFGStructurizer : public MachineFunctionPass {
+public:
+  typedef MachineInstr              InstructionType;
+  typedef MachineFunction           FunctionType;
+  typedef MachineBasicBlock         BlockType;
+  typedef MachineLoopInfo           LoopinfoType;
+  typedef MachineDominatorTree      DominatortreeType;
+  typedef MachinePostDominatorTree  PostDominatortreeType;
+  typedef MachineDomTreeNode        DomTreeNodeType;
+  typedef MachineLoop               LoopType;
+
+protected:
+  TargetMachine &TM;
+  const TargetInstrInfo *TII;
+  const AMDGPURegisterInfo *TRI;
+
+public:
+  AMDGPUCFGStructurizer(char &pid, TargetMachine &tm);
+  const TargetInstrInfo *getTargetInstrInfo() const;
+
+private:
+
+};
+
+} //end of namespace llvm
+AMDGPUCFGStructurizer::AMDGPUCFGStructurizer(char &pid, TargetMachine &tm)
+: MachineFunctionPass(pid), TM(tm), TII(tm.getInstrInfo()),
+  TRI(static_cast<const AMDGPURegisterInfo *>(tm.getRegisterInfo())) {
+}
+
+const TargetInstrInfo *AMDGPUCFGStructurizer::getTargetInstrInfo() const {
+  return TII;
+}
+//===----------------------------------------------------------------------===//
+//
+// CFGPrepare
+//
+//===----------------------------------------------------------------------===//
+
+
+using namespace llvmCFGStruct;
+
+namespace llvm {
+class AMDGPUCFGPrepare : public AMDGPUCFGStructurizer {
+public:
+  static char ID;
+
+public:
+  AMDGPUCFGPrepare(TargetMachine &tm);
+
+  virtual const char *getPassName() const;
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const;
+
+  bool runOnMachineFunction(MachineFunction &F);
+
+private:
+
+};
+
+char AMDGPUCFGPrepare::ID = 0;
+} //end of namespace llvm
+
+AMDGPUCFGPrepare::AMDGPUCFGPrepare(TargetMachine &tm)
+  : AMDGPUCFGStructurizer(ID, tm )  {
+}
+const char *AMDGPUCFGPrepare::getPassName() const {
+  return "AMD IL Control Flow Graph Preparation Pass";
+}
+
+void AMDGPUCFGPrepare::getAnalysisUsage(AnalysisUsage &AU) const {
+  AU.addPreserved<MachineFunctionAnalysis>();
+  AU.addRequired<MachineFunctionAnalysis>();
+  AU.addRequired<MachineDominatorTree>();
+  AU.addRequired<MachinePostDominatorTree>();
+  AU.addRequired<MachineLoopInfo>();
+}
+
+//===----------------------------------------------------------------------===//
+//
+// CFGPerform
+//
+//===----------------------------------------------------------------------===//
+
+
+using namespace llvmCFGStruct;
+
+namespace llvm {
+class AMDGPUCFGPerform : public AMDGPUCFGStructurizer {
+public:
+  static char ID;
+
+public:
+  AMDGPUCFGPerform(TargetMachine &tm);
+  virtual const char *getPassName() const;
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const;
+  bool runOnMachineFunction(MachineFunction &F);
+
+private:
+
+};
+
+char AMDGPUCFGPerform::ID = 0;
+} //end of namespace llvm
+
+  AMDGPUCFGPerform::AMDGPUCFGPerform(TargetMachine &tm)
+: AMDGPUCFGStructurizer(ID, tm) {
+}
+
+const char *AMDGPUCFGPerform::getPassName() const {
+  return "AMD IL Control Flow Graph structurizer Pass";
+}
+
+void AMDGPUCFGPerform::getAnalysisUsage(AnalysisUsage &AU) const {
+  AU.addPreserved<MachineFunctionAnalysis>();
+  AU.addRequired<MachineFunctionAnalysis>();
+  AU.addRequired<MachineDominatorTree>();
+  AU.addRequired<MachinePostDominatorTree>();
+  AU.addRequired<MachineLoopInfo>();
+}
+
+//===----------------------------------------------------------------------===//
+//
+// CFGStructTraits<AMDGPUCFGStructurizer>
+//
+//===----------------------------------------------------------------------===//
+
+namespace llvmCFGStruct {
+// this class is tailor to the AMDGPU backend
+template<>
+struct CFGStructTraits<AMDGPUCFGStructurizer> {
+  typedef int RegiT;
+
+  static int getBranchNzeroOpcode(int oldOpcode) {
+    switch(oldOpcode) {
+    case AMDGPU::JUMP_COND:
+    case AMDGPU::JUMP: return AMDGPU::IF_PREDICATE_SET;
+    case AMDGPU::BRANCH_COND_i32:
+    case AMDGPU::BRANCH_COND_f32: return AMDGPU::IF_LOGICALNZ_f32;
+    default:
+      assert(0 && "internal error");
+    }
+    return -1;
+  }
+
+  static int getBranchZeroOpcode(int oldOpcode) {
+    switch(oldOpcode) {
+    case AMDGPU::JUMP_COND:
+    case AMDGPU::JUMP: return AMDGPU::IF_PREDICATE_SET;
+    case AMDGPU::BRANCH_COND_i32:
+    case AMDGPU::BRANCH_COND_f32: return AMDGPU::IF_LOGICALZ_f32;
+    default:
+      assert(0 && "internal error");
+    }
+    return -1;
+  }
+
+  static int getContinueNzeroOpcode(int oldOpcode) {
+    switch(oldOpcode) {
+    case AMDGPU::JUMP_COND:
+    case AMDGPU::JUMP: return AMDGPU::CONTINUE_LOGICALNZ_i32;
+    default:
+      assert(0 && "internal error");
+    };
+    return -1;
+  }
+
+  static int getContinueZeroOpcode(int oldOpcode) {
+    switch(oldOpcode) {
+    case AMDGPU::JUMP_COND:
+    case AMDGPU::JUMP: return AMDGPU::CONTINUE_LOGICALZ_i32;
+    default:
+      assert(0 && "internal error");
+    }
+    return -1;
+  }
+
+  static MachineBasicBlock *getTrueBranch(MachineInstr *instr) {
+    return instr->getOperand(0).getMBB();
+  }
+
+  static void setTrueBranch(MachineInstr *instr, MachineBasicBlock *blk) {
+    instr->getOperand(0).setMBB(blk);
+  }
+
+  static MachineBasicBlock *
+  getFalseBranch(MachineBasicBlock *blk, MachineInstr *instr) {
+    assert(blk->succ_size() == 2);
+    MachineBasicBlock *trueBranch = getTrueBranch(instr);
+    MachineBasicBlock::succ_iterator iter = blk->succ_begin();
+    MachineBasicBlock::succ_iterator iterNext = iter;
+    ++iterNext;
+
+    return (*iter == trueBranch) ? *iterNext : *iter;
+  }
+
+  static bool isCondBranch(MachineInstr *instr) {
+    switch (instr->getOpcode()) {
+      case AMDGPU::JUMP_COND:
+      case AMDGPU::BRANCH_COND_i32:
+      case AMDGPU::BRANCH_COND_f32:
+      break;
+    default:
+      return false;
+    }
+    return true;
+  }
+
+  static bool isUncondBranch(MachineInstr *instr) {
+    switch (instr->getOpcode()) {
+    case AMDGPU::JUMP:
+    case AMDGPU::BRANCH:
+      return true;
+    default:
+      return false;
+    }
+    return true;
+  }
+
+  static DebugLoc getLastDebugLocInBB(MachineBasicBlock *blk) {
+    //get DebugLoc from the first MachineBasicBlock instruction with debug info
+    DebugLoc DL;
+    for (MachineBasicBlock::iterator iter = blk->begin(); iter != blk->end(); ++iter) {
+      MachineInstr *instr = &(*iter);
+      if (instr->getDebugLoc().isUnknown() == false) {
+        DL = instr->getDebugLoc();
+      }
+    }
+    return DL;
+  }
+
+  static MachineInstr *getNormalBlockBranchInstr(MachineBasicBlock *blk) {
+    MachineBasicBlock::reverse_iterator iter = blk->rbegin();
+    MachineInstr *instr = &*iter;
+    if (instr && (isCondBranch(instr) || isUncondBranch(instr))) {
+      return instr;
+    }
+    return NULL;
+  }
+
+  // The correct naming for this is getPossibleLoopendBlockBranchInstr.
+  //
+  // BB with backward-edge could have move instructions after the branch
+  // instruction.  Such move instruction "belong to" the loop backward-edge.
+  //
+  static MachineInstr *getLoopendBlockBranchInstr(MachineBasicBlock *blk) {
+    const AMDGPUInstrInfo * TII = static_cast<const AMDGPUInstrInfo *>(
+                                  blk->getParent()->getTarget().getInstrInfo());
+
+    for (MachineBasicBlock::reverse_iterator iter = blk->rbegin(),
+         iterEnd = blk->rend(); iter != iterEnd; ++iter) {
+      // FIXME: Simplify
+      MachineInstr *instr = &*iter;
+      if (instr) {
+        if (isCondBranch(instr) || isUncondBranch(instr)) {
+          return instr;
+        } else if (!TII->isMov(instr->getOpcode())) {
+          break;
+        }
+      }
+    }
+    return NULL;
+  }
+
+  static MachineInstr *getReturnInstr(MachineBasicBlock *blk) {
+    MachineBasicBlock::reverse_iterator iter = blk->rbegin();
+    if (iter != blk->rend()) {
+      MachineInstr *instr = &(*iter);
+      if (instr->getOpcode() == AMDGPU::RETURN) {
+        return instr;
+      }
+    }
+    return NULL;
+  }
+
+  static MachineInstr *getContinueInstr(MachineBasicBlock *blk) {
+    MachineBasicBlock::reverse_iterator iter = blk->rbegin();
+    if (iter != blk->rend()) {
+      MachineInstr *instr = &(*iter);
+      if (instr->getOpcode() == AMDGPU::CONTINUE) {
+        return instr;
+      }
+    }
+    return NULL;
+  }
+
+  static MachineInstr *getLoopBreakInstr(MachineBasicBlock *blk) {
+    for (MachineBasicBlock::iterator iter = blk->begin(); (iter != blk->end()); ++iter) {
+      MachineInstr *instr = &(*iter);
+      if (instr->getOpcode() == AMDGPU::PREDICATED_BREAK) {
+        return instr;
+      }
+    }
+    return NULL;
+  }
+
+  static bool isReturnBlock(MachineBasicBlock *blk) {
+    MachineInstr *instr = getReturnInstr(blk);
+    bool isReturn = (blk->succ_size() == 0);
+    if (instr) {
+      assert(isReturn);
+    } else if (isReturn) {
+      if (DEBUGME) {
+        errs() << "BB" << blk->getNumber()
+               <<" is return block without RETURN instr\n";
+      }
+    }
+
+    return  isReturn;
+  }
+
+  static MachineBasicBlock::iterator
+  getInstrPos(MachineBasicBlock *blk, MachineInstr *instr) {
+    assert(instr->getParent() == blk && "instruction doesn't belong to block");
+    MachineBasicBlock::iterator iter = blk->begin();
+    MachineBasicBlock::iterator iterEnd = blk->end();
+    while (&(*iter) != instr && iter != iterEnd) {
+      ++iter;
+    }
+
+    assert(iter != iterEnd);
+    return iter;
+  }//getInstrPos
+
+  static MachineInstr *insertInstrBefore(MachineBasicBlock *blk, int newOpcode,
+                                         AMDGPUCFGStructurizer *passRep) {
+    return insertInstrBefore(blk,newOpcode,passRep,DebugLoc());
+  } //insertInstrBefore
+
+  static MachineInstr *insertInstrBefore(MachineBasicBlock *blk, int newOpcode,
+                                         AMDGPUCFGStructurizer *passRep, DebugLoc DL) {
+    const TargetInstrInfo *tii = passRep->getTargetInstrInfo();
+    MachineInstr *newInstr =
+      blk->getParent()->CreateMachineInstr(tii->get(newOpcode), DL);
+
+    MachineBasicBlock::iterator res;
+    if (blk->begin() != blk->end()) {
+      blk->insert(blk->begin(), newInstr);
+    } else {
+      blk->push_back(newInstr);
+    }
+
+    SHOWNEWINSTR(newInstr);
+
+    return newInstr;
+  } //insertInstrBefore
+
+  static void insertInstrEnd(MachineBasicBlock *blk, int newOpcode,
+                             AMDGPUCFGStructurizer *passRep) {
+    insertInstrEnd(blk,newOpcode,passRep,DebugLoc());
+  } //insertInstrEnd
+
+  static void insertInstrEnd(MachineBasicBlock *blk, int newOpcode,
+                             AMDGPUCFGStructurizer *passRep, DebugLoc DL) {
+    const TargetInstrInfo *tii = passRep->getTargetInstrInfo();
+   MachineInstr *newInstr = blk->getParent()
+      ->CreateMachineInstr(tii->get(newOpcode), DL);
+
+    blk->push_back(newInstr);
+    //assume the instruction doesn't take any reg operand ...
+
+    SHOWNEWINSTR(newInstr);
+  } //insertInstrEnd
+
+  static MachineInstr *insertInstrBefore(MachineBasicBlock::iterator instrPos,
+                                         int newOpcode, 
+                                         AMDGPUCFGStructurizer *passRep) {
+    MachineInstr *oldInstr = &(*instrPos);
+    const TargetInstrInfo *tii = passRep->getTargetInstrInfo();
+    MachineBasicBlock *blk = oldInstr->getParent();
+    MachineInstr *newInstr =
+      blk->getParent()->CreateMachineInstr(tii->get(newOpcode),
+                                           DebugLoc());
+
+    blk->insert(instrPos, newInstr);
+    //assume the instruction doesn't take any reg operand ...
+
+    SHOWNEWINSTR(newInstr);
+    return newInstr;
+  } //insertInstrBefore
+
+  static void insertCondBranchBefore(MachineBasicBlock::iterator instrPos,
+                                     int newOpcode,
+                                     AMDGPUCFGStructurizer *passRep,
+                                     DebugLoc DL) {
+    MachineInstr *oldInstr = &(*instrPos);
+    const TargetInstrInfo *tii = passRep->getTargetInstrInfo();
+    MachineBasicBlock *blk = oldInstr->getParent();
+    MachineFunction *MF = blk->getParent();
+    MachineInstr *newInstr = MF->CreateMachineInstr(tii->get(newOpcode), DL);
+
+    blk->insert(instrPos, newInstr);
+    MachineInstrBuilder MIB(*MF, newInstr);
+    MIB.addReg(oldInstr->getOperand(1).getReg(), false);
+
+    SHOWNEWINSTR(newInstr);
+    //erase later oldInstr->eraseFromParent();
+  } //insertCondBranchBefore
+
+  static void insertCondBranchBefore(MachineBasicBlock *blk,
+                                     MachineBasicBlock::iterator insertPos,
+                                     int newOpcode,
+                                     AMDGPUCFGStructurizer *passRep,
+                                     RegiT regNum,
+                                     DebugLoc DL) {
+    const TargetInstrInfo *tii = passRep->getTargetInstrInfo();
+    MachineFunction *MF = blk->getParent();
+
+    MachineInstr *newInstr = MF->CreateMachineInstr(tii->get(newOpcode), DL);
+
+    //insert before
+    blk->insert(insertPos, newInstr);
+    MachineInstrBuilder(*MF, newInstr).addReg(regNum, false);
+
+    SHOWNEWINSTR(newInstr);
+  } //insertCondBranchBefore
+
+  static void insertCondBranchEnd(MachineBasicBlock *blk,
+                                  int newOpcode,
+                                  AMDGPUCFGStructurizer *passRep,
+                                  RegiT regNum) {
+    const TargetInstrInfo *tii = passRep->getTargetInstrInfo();
+    MachineFunction *MF = blk->getParent();
+    MachineInstr *newInstr =
+      MF->CreateMachineInstr(tii->get(newOpcode), DebugLoc());
+
+    blk->push_back(newInstr);
+    MachineInstrBuilder(*MF, newInstr).addReg(regNum, false);
+
+    SHOWNEWINSTR(newInstr);
+  } //insertCondBranchEnd
+
+
+  static void insertAssignInstrBefore(MachineBasicBlock::iterator instrPos,
+                                      AMDGPUCFGStructurizer *passRep,
+                                      RegiT regNum, int regVal) {
+    MachineInstr *oldInstr = &(*instrPos);
+    const AMDGPUInstrInfo *tii =
+             static_cast<const AMDGPUInstrInfo *>(passRep->getTargetInstrInfo());
+    MachineBasicBlock *blk = oldInstr->getParent();
+    MachineInstr *newInstr = tii->getMovImmInstr(blk->getParent(), regNum,
+                                                 regVal);
+    blk->insert(instrPos, newInstr);
+
+    SHOWNEWINSTR(newInstr);
+  } //insertAssignInstrBefore
+
+  static void insertAssignInstrBefore(MachineBasicBlock *blk,
+                                      AMDGPUCFGStructurizer *passRep,
+                                      RegiT regNum, int regVal) {
+    const AMDGPUInstrInfo *tii =
+             static_cast<const AMDGPUInstrInfo *>(passRep->getTargetInstrInfo());
+
+    MachineInstr *newInstr = tii->getMovImmInstr(blk->getParent(), regNum,
+                                                 regVal);
+    if (blk->begin() != blk->end()) {
+      blk->insert(blk->begin(), newInstr);
+    } else {
+      blk->push_back(newInstr);
+    }
+
+    SHOWNEWINSTR(newInstr);
+
+  } //insertInstrBefore
+
+  static void insertCompareInstrBefore(MachineBasicBlock *blk,
+                                       MachineBasicBlock::iterator instrPos,
+                                       AMDGPUCFGStructurizer *passRep,
+                                       RegiT dstReg, RegiT src1Reg,
+                                       RegiT src2Reg) {
+    const AMDGPUInstrInfo *tii =
+             static_cast<const AMDGPUInstrInfo *>(passRep->getTargetInstrInfo());
+    MachineFunction *MF = blk->getParent();
+    MachineInstr *newInstr =
+      MF->CreateMachineInstr(tii->get(tii->getIEQOpcode()), DebugLoc());
+
+    MachineInstrBuilder MIB(*MF, newInstr);
+    MIB.addReg(dstReg, RegState::Define); //set target
+    MIB.addReg(src1Reg); //set src value
+    MIB.addReg(src2Reg); //set src value
+
+    blk->insert(instrPos, newInstr);
+    SHOWNEWINSTR(newInstr);
+
+  } //insertCompareInstrBefore
+
+  static void cloneSuccessorList(MachineBasicBlock *dstBlk,
+                                 MachineBasicBlock *srcBlk) {
+    for (MachineBasicBlock::succ_iterator iter = srcBlk->succ_begin(),
+         iterEnd = srcBlk->succ_end(); iter != iterEnd; ++iter) {
+      dstBlk->addSuccessor(*iter);  // *iter's predecessor is also taken care of
+    }
+  } //cloneSuccessorList
+
+  static MachineBasicBlock *clone(MachineBasicBlock *srcBlk) {
+    MachineFunction *func = srcBlk->getParent();
+    MachineBasicBlock *newBlk = func->CreateMachineBasicBlock();
+    func->push_back(newBlk);  //insert to function
+    for (MachineBasicBlock::iterator iter = srcBlk->begin(),
+         iterEnd = srcBlk->end();
+         iter != iterEnd; ++iter) {
+      MachineInstr *instr = func->CloneMachineInstr(iter);
+      newBlk->push_back(instr);
+    }
+    return newBlk;
+  }
+
+  //MachineBasicBlock::ReplaceUsesOfBlockWith doesn't serve the purpose because
+  //the AMDGPU instruction is not recognized as terminator fix this and retire
+  //this routine
+  static void replaceInstrUseOfBlockWith(MachineBasicBlock *srcBlk,
+                                         MachineBasicBlock *oldBlk,
+                                         MachineBasicBlock *newBlk) {
+    MachineInstr *branchInstr = getLoopendBlockBranchInstr(srcBlk);
+    if (branchInstr && isCondBranch(branchInstr) &&
+        getTrueBranch(branchInstr) == oldBlk) {
+      setTrueBranch(branchInstr, newBlk);
+    }
+  }
+
+  static void wrapup(MachineBasicBlock *entryBlk) {
+    assert((!entryBlk->getParent()->getJumpTableInfo()
+            || entryBlk->getParent()->getJumpTableInfo()->isEmpty())
+           && "found a jump table");
+
+     //collect continue right before endloop
+     SmallVector<MachineInstr *, DEFAULT_VEC_SLOTS> contInstr;
+     MachineBasicBlock::iterator pre = entryBlk->begin();
+     MachineBasicBlock::iterator iterEnd = entryBlk->end();
+     MachineBasicBlock::iterator iter = pre;
+     while (iter != iterEnd) {
+       if (pre->getOpcode() == AMDGPU::CONTINUE
+           && iter->getOpcode() == AMDGPU::ENDLOOP) {
+         contInstr.push_back(pre);
+       }
+       pre = iter;
+       ++iter;
+     } //end while
+
+     //delete continue right before endloop
+     for (unsigned i = 0; i < contInstr.size(); ++i) {
+        contInstr[i]->eraseFromParent();
+     }
+
+     // TODO to fix up jump table so later phase won't be confused.  if
+     // (jumpTableInfo->isEmpty() == false) { need to clean the jump table, but
+     // there isn't such an interface yet.  alternatively, replace all the other
+     // blocks in the jump table with the entryBlk //}
+
+  } //wrapup
+
+  static MachineDominatorTree *getDominatorTree(AMDGPUCFGStructurizer &pass) {
+    return &pass.getAnalysis<MachineDominatorTree>();
+  }
+
+  static MachinePostDominatorTree*
+  getPostDominatorTree(AMDGPUCFGStructurizer &pass) {
+    return &pass.getAnalysis<MachinePostDominatorTree>();
+  }
+
+  static MachineLoopInfo *getLoopInfo(AMDGPUCFGStructurizer &pass) {
+    return &pass.getAnalysis<MachineLoopInfo>();
+  }
+}; // template class CFGStructTraits
+} //end of namespace llvm
+
+// createAMDGPUCFGPreparationPass- Returns a pass
+FunctionPass *llvm::createAMDGPUCFGPreparationPass(TargetMachine &tm
+                                                 ) {
+  return new AMDGPUCFGPrepare(tm );
+}
+
+bool AMDGPUCFGPrepare::runOnMachineFunction(MachineFunction &func) {
+  return llvmCFGStruct::CFGStructurizer<AMDGPUCFGStructurizer>().prepare(func,
+                                                                        *this,
+                                                                        TRI);
+}
+
+// createAMDGPUCFGStructurizerPass- Returns a pass
+FunctionPass *llvm::createAMDGPUCFGStructurizerPass(TargetMachine &tm
+                                                  ) {
+  return new AMDGPUCFGPerform(tm );
+}
+
+bool AMDGPUCFGPerform::runOnMachineFunction(MachineFunction &func) {
+  return llvmCFGStruct::CFGStructurizer<AMDGPUCFGStructurizer>().run(func,
+                                                                    *this,
+                                                                    TRI);
+}
diff --git a/lib/Target/R600/AMDILDevice.cpp b/lib/Target/R600/AMDILDevice.cpp
new file mode 100644
index 000000000000..db8e01ea4043
--- /dev/null
+++ b/lib/Target/R600/AMDILDevice.cpp
@@ -0,0 +1,132 @@
+//===-- AMDILDevice.cpp - Base class for AMDIL Devices --------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//==-----------------------------------------------------------------------===//
+#include "AMDILDevice.h"
+#include "AMDGPUSubtarget.h"
+
+using namespace llvm;
+// Default implementation for all of the classes.
+AMDGPUDevice::AMDGPUDevice(AMDGPUSubtarget *ST) : mSTM(ST) {
+  mHWBits.resize(AMDGPUDeviceInfo::MaxNumberCapabilities);
+  mSWBits.resize(AMDGPUDeviceInfo::MaxNumberCapabilities);
+  setCaps();
+  DeviceFlag = OCL_DEVICE_ALL;
+}
+
+AMDGPUDevice::~AMDGPUDevice() {
+    mHWBits.clear();
+    mSWBits.clear();
+}
+
+size_t AMDGPUDevice::getMaxGDSSize() const {
+  return 0;
+}
+
+uint32_t 
+AMDGPUDevice::getDeviceFlag() const {
+  return DeviceFlag;
+}
+
+size_t AMDGPUDevice::getMaxNumCBs() const {
+  if (usesHardware(AMDGPUDeviceInfo::ConstantMem)) {
+    return HW_MAX_NUM_CB;
+  }
+
+  return 0;
+}
+
+size_t AMDGPUDevice::getMaxCBSize() const {
+  if (usesHardware(AMDGPUDeviceInfo::ConstantMem)) {
+    return MAX_CB_SIZE;
+  }
+
+  return 0;
+}
+
+size_t AMDGPUDevice::getMaxScratchSize() const {
+  return 65536;
+}
+
+uint32_t AMDGPUDevice::getStackAlignment() const {
+  return 16;
+}
+
+void AMDGPUDevice::setCaps() {
+  mSWBits.set(AMDGPUDeviceInfo::HalfOps);
+  mSWBits.set(AMDGPUDeviceInfo::ByteOps);
+  mSWBits.set(AMDGPUDeviceInfo::ShortOps);
+  mSWBits.set(AMDGPUDeviceInfo::HW64BitDivMod);
+  if (mSTM->isOverride(AMDGPUDeviceInfo::NoInline)) {
+    mSWBits.set(AMDGPUDeviceInfo::NoInline);
+  }
+  if (mSTM->isOverride(AMDGPUDeviceInfo::MacroDB)) {
+    mSWBits.set(AMDGPUDeviceInfo::MacroDB);
+  }
+  if (mSTM->isOverride(AMDGPUDeviceInfo::Debug)) {
+    mSWBits.set(AMDGPUDeviceInfo::ConstantMem);
+  } else {
+    mHWBits.set(AMDGPUDeviceInfo::ConstantMem);
+  }
+  if (mSTM->isOverride(AMDGPUDeviceInfo::Debug)) {
+    mSWBits.set(AMDGPUDeviceInfo::PrivateMem);
+  } else {
+    mHWBits.set(AMDGPUDeviceInfo::PrivateMem);
+  }
+  if (mSTM->isOverride(AMDGPUDeviceInfo::BarrierDetect)) {
+    mSWBits.set(AMDGPUDeviceInfo::BarrierDetect);
+  }
+  mSWBits.set(AMDGPUDeviceInfo::ByteLDSOps);
+  mSWBits.set(AMDGPUDeviceInfo::LongOps);
+}
+
+AMDGPUDeviceInfo::ExecutionMode
+AMDGPUDevice::getExecutionMode(AMDGPUDeviceInfo::Caps Caps) const {
+  if (mHWBits[Caps]) {
+    assert(!mSWBits[Caps] && "Cannot set both SW and HW caps");
+    return AMDGPUDeviceInfo::Hardware;
+  }
+
+  if (mSWBits[Caps]) {
+    assert(!mHWBits[Caps] && "Cannot set both SW and HW caps");
+    return AMDGPUDeviceInfo::Software;
+  }
+
+  return AMDGPUDeviceInfo::Unsupported;
+
+}
+
+bool AMDGPUDevice::isSupported(AMDGPUDeviceInfo::Caps Mode) const {
+  return getExecutionMode(Mode) != AMDGPUDeviceInfo::Unsupported;
+}
+
+bool AMDGPUDevice::usesHardware(AMDGPUDeviceInfo::Caps Mode) const {
+  return getExecutionMode(Mode) == AMDGPUDeviceInfo::Hardware;
+}
+
+bool AMDGPUDevice::usesSoftware(AMDGPUDeviceInfo::Caps Mode) const {
+  return getExecutionMode(Mode) == AMDGPUDeviceInfo::Software;
+}
+
+std::string
+AMDGPUDevice::getDataLayout() const {
+  std::string DataLayout = std::string(
+   "e"
+   "-p:32:32:32"
+   "-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32"
+   "-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128"
+   "-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024-v2048:2048:2048"
+   "-n32:64"
+  );
+
+  if (usesHardware(AMDGPUDeviceInfo::DoubleOps)) {
+    DataLayout.append("-f64:64:64");
+  }
+
+  return DataLayout;
+}
diff --git a/lib/Target/R600/AMDILDevice.h b/lib/Target/R600/AMDILDevice.h
new file mode 100644
index 000000000000..97df98cafb2a
--- /dev/null
+++ b/lib/Target/R600/AMDILDevice.h
@@ -0,0 +1,117 @@
+//===---- AMDILDevice.h - Define Device Data for AMDGPU -----*- C++ -*------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Interface for the subtarget data classes.
+//
+/// This file will define the interface that each generation needs to
+/// implement in order to correctly answer queries on the capabilities of the
+/// specific hardware.
+//===----------------------------------------------------------------------===//
+#ifndef AMDILDEVICEIMPL_H
+#define AMDILDEVICEIMPL_H
+#include "AMDIL.h"
+#include "llvm/ADT/BitVector.h"
+
+namespace llvm {
+  class AMDGPUSubtarget;
+  class MCStreamer;
+//===----------------------------------------------------------------------===//
+// Interface for data that is specific to a single device
+//===----------------------------------------------------------------------===//
+class AMDGPUDevice {
+public:
+  AMDGPUDevice(AMDGPUSubtarget *ST);
+  virtual ~AMDGPUDevice();
+
+  // Enum values for the various memory types.
+  enum {
+    RAW_UAV_ID   = 0,
+    ARENA_UAV_ID = 1,
+    LDS_ID       = 2,
+    GDS_ID       = 3,
+    SCRATCH_ID   = 4,
+    CONSTANT_ID  = 5,
+    GLOBAL_ID    = 6,
+    MAX_IDS      = 7
+  } IO_TYPE_IDS;
+
+  /// \returns The max LDS size that the hardware supports.  Size is in
+  /// bytes.
+  virtual size_t getMaxLDSSize() const = 0;
+
+  /// \returns The max GDS size that the hardware supports if the GDS is
+  /// supported by the hardware.  Size is in bytes.
+  virtual size_t getMaxGDSSize() const;
+
+  /// \returns The max number of hardware constant address spaces that
+  /// are supported by this device.
+  virtual size_t getMaxNumCBs() const;
+
+  /// \returns The max number of bytes a single hardware constant buffer
+  /// can support.  Size is in bytes.
+  virtual size_t getMaxCBSize() const;
+
+  /// \returns The max number of bytes allowed by the hardware scratch
+  /// buffer.  Size is in bytes.
+  virtual size_t getMaxScratchSize() const;
+
+  /// \brief Get the flag that corresponds to the device.
+  virtual uint32_t getDeviceFlag() const;
+
+  /// \returns The number of work-items that exist in a single hardware
+  /// wavefront.
+  virtual size_t getWavefrontSize() const = 0;
+
+  /// \brief Get the generational name of this specific device.
+  virtual uint32_t getGeneration() const = 0;
+
+  /// \brief Get the stack alignment of this specific device.
+  virtual uint32_t getStackAlignment() const;
+
+  /// \brief Get the resource ID for this specific device.
+  virtual uint32_t getResourceID(uint32_t DeviceID) const = 0;
+
+  /// \brief Get the max number of UAV's for this device.
+  virtual uint32_t getMaxNumUAVs() const = 0;
+
+
+  // API utilizing more detailed capabilities of each family of
+  // cards. If a capability is supported, then either usesHardware or
+  // usesSoftware returned true.  If usesHardware returned true, then
+  // usesSoftware must return false for the same capability.  Hardware
+  // execution means that the feature is done natively by the hardware
+  // and is not emulated by the softare.  Software execution means
+  // that the feature could be done in the hardware, but there is
+  // software that emulates it with possibly using the hardware for
+  // support since the hardware does not fully comply with OpenCL
+  // specs.
+
+  bool isSupported(AMDGPUDeviceInfo::Caps Mode) const;
+  bool usesHardware(AMDGPUDeviceInfo::Caps Mode) const;
+  bool usesSoftware(AMDGPUDeviceInfo::Caps Mode) const;
+  virtual std::string getDataLayout() const;
+  static const unsigned int MAX_LDS_SIZE_700 = 16384;
+  static const unsigned int MAX_LDS_SIZE_800 = 32768;
+  static const unsigned int WavefrontSize = 64;
+  static const unsigned int HalfWavefrontSize = 32;
+  static const unsigned int QuarterWavefrontSize = 16;
+protected:
+  virtual void setCaps();
+  BitVector mHWBits;
+  llvm::BitVector mSWBits;
+  AMDGPUSubtarget *mSTM;
+  uint32_t DeviceFlag;
+private:
+  AMDGPUDeviceInfo::ExecutionMode
+  getExecutionMode(AMDGPUDeviceInfo::Caps Caps) const;
+};
+
+} // namespace llvm
+#endif // AMDILDEVICEIMPL_H
diff --git a/lib/Target/R600/AMDILDeviceInfo.cpp b/lib/Target/R600/AMDILDeviceInfo.cpp
new file mode 100644
index 000000000000..9605fbe63340
--- /dev/null
+++ b/lib/Target/R600/AMDILDeviceInfo.cpp
@@ -0,0 +1,94 @@
+//===-- AMDILDeviceInfo.cpp - AMDILDeviceInfo class -----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Function that creates DeviceInfo from a device name and other information.
+//
+//==-----------------------------------------------------------------------===//
+#include "AMDILDevices.h"
+#include "AMDGPUSubtarget.h"
+
+using namespace llvm;
+namespace llvm {
+namespace AMDGPUDeviceInfo {
+
+AMDGPUDevice* getDeviceFromName(const std::string &deviceName,
+                                AMDGPUSubtarget *ptr,
+                                bool is64bit, bool is64on32bit) {
+  if (deviceName.c_str()[2] == '7') {
+    switch (deviceName.c_str()[3]) {
+    case '1':
+      return new AMDGPU710Device(ptr);
+    case '7':
+      return new AMDGPU770Device(ptr);
+    default:
+      return new AMDGPU7XXDevice(ptr);
+    }
+  } else if (deviceName == "cypress") {
+#if DEBUG
+    assert(!is64bit && "This device does not support 64bit pointers!");
+    assert(!is64on32bit && "This device does not support 64bit"
+          " on 32bit pointers!");
+#endif
+    return new AMDGPUCypressDevice(ptr);
+  } else if (deviceName == "juniper") {
+#if DEBUG
+    assert(!is64bit && "This device does not support 64bit pointers!");
+    assert(!is64on32bit && "This device does not support 64bit"
+          " on 32bit pointers!");
+#endif
+    return new AMDGPUEvergreenDevice(ptr);
+  } else if (deviceName == "redwood") {
+#if DEBUG
+    assert(!is64bit && "This device does not support 64bit pointers!");
+    assert(!is64on32bit && "This device does not support 64bit"
+          " on 32bit pointers!");
+#endif
+    return new AMDGPURedwoodDevice(ptr);
+  } else if (deviceName == "cedar") {
+#if DEBUG
+    assert(!is64bit && "This device does not support 64bit pointers!");
+    assert(!is64on32bit && "This device does not support 64bit"
+          " on 32bit pointers!");
+#endif
+    return new AMDGPUCedarDevice(ptr);
+  } else if (deviceName == "barts" || deviceName == "turks") {
+#if DEBUG
+    assert(!is64bit && "This device does not support 64bit pointers!");
+    assert(!is64on32bit && "This device does not support 64bit"
+          " on 32bit pointers!");
+#endif
+    return new AMDGPUNIDevice(ptr);
+  } else if (deviceName == "cayman") {
+#if DEBUG
+    assert(!is64bit && "This device does not support 64bit pointers!");
+    assert(!is64on32bit && "This device does not support 64bit"
+          " on 32bit pointers!");
+#endif
+    return new AMDGPUCaymanDevice(ptr);
+  } else if (deviceName == "caicos") {
+#if DEBUG
+    assert(!is64bit && "This device does not support 64bit pointers!");
+    assert(!is64on32bit && "This device does not support 64bit"
+          " on 32bit pointers!");
+#endif
+    return new AMDGPUNIDevice(ptr);
+  } else if (deviceName == "SI") {
+    return new AMDGPUSIDevice(ptr);
+  } else {
+#if DEBUG
+    assert(!is64bit && "This device does not support 64bit pointers!");
+    assert(!is64on32bit && "This device does not support 64bit"
+          " on 32bit pointers!");
+#endif
+    return new AMDGPU7XXDevice(ptr);
+  }
+}
+} // End namespace AMDGPUDeviceInfo
+} // End namespace llvm
diff --git a/lib/Target/R600/AMDILDeviceInfo.h b/lib/Target/R600/AMDILDeviceInfo.h
new file mode 100644
index 000000000000..4b2c3a53c79f
--- /dev/null
+++ b/lib/Target/R600/AMDILDeviceInfo.h
@@ -0,0 +1,88 @@
+//===-- AMDILDeviceInfo.h - Constants for describing devices --------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//==-----------------------------------------------------------------------===//
+#ifndef AMDILDEVICEINFO_H
+#define AMDILDEVICEINFO_H
+
+
+#include <string>
+
+namespace llvm {
+  class AMDGPUDevice;
+  class AMDGPUSubtarget;
+  namespace AMDGPUDeviceInfo {
+    /// Each Capabilities can be executed using a hardware instruction,
+    /// emulated with a sequence of software instructions, or not
+    /// supported at all.
+    enum ExecutionMode {
+      Unsupported = 0, ///< Unsupported feature on the card(Default value)
+       /// This is the execution mode that is set if the feature is emulated in
+       /// software.
+      Software,
+      /// This execution mode is set if the feature exists natively in hardware
+      Hardware
+    };
+
+    enum Caps {
+      HalfOps          = 0x1,  ///< Half float is supported or not.
+      DoubleOps        = 0x2,  ///< Double is supported or not.
+      ByteOps          = 0x3,  ///< Byte(char) is support or not.
+      ShortOps         = 0x4,  ///< Short is supported or not.
+      LongOps          = 0x5,  ///< Long is supported or not.
+      Images           = 0x6,  ///< Images are supported or not.
+      ByteStores       = 0x7,  ///< ByteStores available(!HD4XXX).
+      ConstantMem      = 0x8,  ///< Constant/CB memory.
+      LocalMem         = 0x9,  ///< Local/LDS memory.
+      PrivateMem       = 0xA,  ///< Scratch/Private/Stack memory.
+      RegionMem        = 0xB,  ///< OCL GDS Memory Extension.
+      FMA              = 0xC,  ///< Use HW FMA or SW FMA.
+      ArenaSegment     = 0xD,  ///< Use for Arena UAV per pointer 12-1023.
+      MultiUAV         = 0xE,  ///< Use for UAV per Pointer 0-7.
+      Reserved0        = 0xF,  ///< ReservedFlag
+      NoAlias          = 0x10, ///< Cached loads.
+      Signed24BitOps   = 0x11, ///< Peephole Optimization.
+      /// Debug mode implies that no hardware features or optimizations
+      /// are performned and that all memory access go through a single
+      /// uav(Arena on HD5XXX/HD6XXX and Raw on HD4XXX).
+      Debug            = 0x12,
+      CachedMem        = 0x13, ///< Cached mem is available or not.
+      BarrierDetect    = 0x14, ///< Detect duplicate barriers.
+      Reserved1        = 0x15, ///< Reserved flag
+      ByteLDSOps       = 0x16, ///< Flag to specify if byte LDS ops are available.
+      ArenaVectors     = 0x17, ///< Flag to specify if vector loads from arena work.
+      TmrReg           = 0x18, ///< Flag to specify if Tmr register is supported.
+      NoInline         = 0x19, ///< Flag to specify that no inlining should occur.
+      MacroDB          = 0x1A, ///< Flag to specify that backend handles macrodb.
+      HW64BitDivMod    = 0x1B, ///< Flag for backend to generate 64bit div/mod.
+      ArenaUAV         = 0x1C, ///< Flag to specify that arena uav is supported.
+      PrivateUAV       = 0x1D, ///< Flag to specify that private memory uses uav's.
+      /// If more capabilities are required, then
+      /// this number needs to be increased.
+      /// All capabilities must come before this
+      /// number.
+      MaxNumberCapabilities = 0x20
+    };
+    /// These have to be in order with the older generations
+    /// having the lower number enumerations.
+    enum Generation {
+      HD4XXX = 0, ///< 7XX based devices.
+      HD5XXX, ///< Evergreen based devices.
+      HD6XXX, ///< NI/Evergreen+ based devices.
+      HD7XXX, ///< Southern Islands based devices.
+      HDTEST, ///< Experimental feature testing device.
+      HDNUMGEN
+    };
+
+
+  AMDGPUDevice*
+    getDeviceFromName(const std::string &name, AMDGPUSubtarget *ptr,
+                      bool is64bit = false, bool is64on32bit = false);
+  } // namespace AMDILDeviceInfo
+} // namespace llvm
+#endif // AMDILDEVICEINFO_H
diff --git a/lib/Target/R600/AMDILDevices.h b/lib/Target/R600/AMDILDevices.h
new file mode 100644
index 000000000000..636fa6d35947
--- /dev/null
+++ b/lib/Target/R600/AMDILDevices.h
@@ -0,0 +1,19 @@
+//===-- AMDILDevices.h - Consolidate AMDIL Device headers -----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//==-----------------------------------------------------------------------===//
+#ifndef AMDIL_DEVICES_H
+#define AMDIL_DEVICES_H
+// Include all of the device specific header files
+#include "AMDIL7XXDevice.h"
+#include "AMDILDevice.h"
+#include "AMDILEvergreenDevice.h"
+#include "AMDILNIDevice.h"
+#include "AMDILSIDevice.h"
+
+#endif // AMDIL_DEVICES_H
diff --git a/lib/Target/R600/AMDILEvergreenDevice.cpp b/lib/Target/R600/AMDILEvergreenDevice.cpp
new file mode 100644
index 000000000000..c5213a041005
--- /dev/null
+++ b/lib/Target/R600/AMDILEvergreenDevice.cpp
@@ -0,0 +1,169 @@
+//===-- AMDILEvergreenDevice.cpp - Device Info for Evergreen --------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//==-----------------------------------------------------------------------===//
+#include "AMDILEvergreenDevice.h"
+
+using namespace llvm;
+
+AMDGPUEvergreenDevice::AMDGPUEvergreenDevice(AMDGPUSubtarget *ST)
+: AMDGPUDevice(ST) {
+  setCaps();
+  std::string name = ST->getDeviceName();
+  if (name == "cedar") {
+    DeviceFlag = OCL_DEVICE_CEDAR;
+  } else if (name == "redwood") {
+    DeviceFlag = OCL_DEVICE_REDWOOD;
+  } else if (name == "cypress") {
+    DeviceFlag = OCL_DEVICE_CYPRESS;
+  } else {
+    DeviceFlag = OCL_DEVICE_JUNIPER;
+  }
+}
+
+AMDGPUEvergreenDevice::~AMDGPUEvergreenDevice() {
+}
+
+size_t AMDGPUEvergreenDevice::getMaxLDSSize() const {
+  if (usesHardware(AMDGPUDeviceInfo::LocalMem)) {
+    return MAX_LDS_SIZE_800;
+  } else {
+    return 0;
+  }
+}
+size_t AMDGPUEvergreenDevice::getMaxGDSSize() const {
+  if (usesHardware(AMDGPUDeviceInfo::RegionMem)) {
+    return MAX_LDS_SIZE_800;
+  } else {
+    return 0;
+  }
+}
+uint32_t AMDGPUEvergreenDevice::getMaxNumUAVs() const {
+  return 12;
+}
+
+uint32_t AMDGPUEvergreenDevice::getResourceID(uint32_t id) const {
+  switch(id) {
+  default:
+    assert(0 && "ID type passed in is unknown!");
+    break;
+  case CONSTANT_ID:
+  case RAW_UAV_ID:
+    return GLOBAL_RETURN_RAW_UAV_ID;
+  case GLOBAL_ID:
+  case ARENA_UAV_ID:
+    return DEFAULT_ARENA_UAV_ID;
+  case LDS_ID:
+    if (usesHardware(AMDGPUDeviceInfo::LocalMem)) {
+      return DEFAULT_LDS_ID;
+    } else {
+      return DEFAULT_ARENA_UAV_ID;
+    }
+  case GDS_ID:
+    if (usesHardware(AMDGPUDeviceInfo::RegionMem)) {
+      return DEFAULT_GDS_ID;
+    } else {
+      return DEFAULT_ARENA_UAV_ID;
+    }
+  case SCRATCH_ID:
+    if (usesHardware(AMDGPUDeviceInfo::PrivateMem)) {
+      return DEFAULT_SCRATCH_ID;
+    } else {
+      return DEFAULT_ARENA_UAV_ID;
+    }
+  };
+  return 0;
+}
+
+size_t AMDGPUEvergreenDevice::getWavefrontSize() const {
+  return AMDGPUDevice::WavefrontSize;
+}
+
+uint32_t AMDGPUEvergreenDevice::getGeneration() const {
+  return AMDGPUDeviceInfo::HD5XXX;
+}
+
+void AMDGPUEvergreenDevice::setCaps() {
+  mSWBits.set(AMDGPUDeviceInfo::ArenaSegment);
+  mHWBits.set(AMDGPUDeviceInfo::ArenaUAV);
+  mHWBits.set(AMDGPUDeviceInfo::HW64BitDivMod);
+  mSWBits.reset(AMDGPUDeviceInfo::HW64BitDivMod);
+  mSWBits.set(AMDGPUDeviceInfo::Signed24BitOps);
+  if (mSTM->isOverride(AMDGPUDeviceInfo::ByteStores)) {
+    mHWBits.set(AMDGPUDeviceInfo::ByteStores);
+  }
+  if (mSTM->isOverride(AMDGPUDeviceInfo::Debug)) {
+    mSWBits.set(AMDGPUDeviceInfo::LocalMem);
+    mSWBits.set(AMDGPUDeviceInfo::RegionMem);
+  } else {
+    mHWBits.set(AMDGPUDeviceInfo::LocalMem);
+    mHWBits.set(AMDGPUDeviceInfo::RegionMem);
+  }
+  mHWBits.set(AMDGPUDeviceInfo::Images);
+  if (mSTM->isOverride(AMDGPUDeviceInfo::NoAlias)) {
+    mHWBits.set(AMDGPUDeviceInfo::NoAlias);
+  }
+  mHWBits.set(AMDGPUDeviceInfo::CachedMem);
+  if (mSTM->isOverride(AMDGPUDeviceInfo::MultiUAV)) {
+    mHWBits.set(AMDGPUDeviceInfo::MultiUAV);
+  }
+  mHWBits.set(AMDGPUDeviceInfo::ByteLDSOps);
+  mSWBits.reset(AMDGPUDeviceInfo::ByteLDSOps);
+  mHWBits.set(AMDGPUDeviceInfo::ArenaVectors);
+  mHWBits.set(AMDGPUDeviceInfo::LongOps);
+  mSWBits.reset(AMDGPUDeviceInfo::LongOps);
+  mHWBits.set(AMDGPUDeviceInfo::TmrReg);
+}
+
+AMDGPUCypressDevice::AMDGPUCypressDevice(AMDGPUSubtarget *ST)
+  : AMDGPUEvergreenDevice(ST) {
+  setCaps();
+}
+
+AMDGPUCypressDevice::~AMDGPUCypressDevice() {
+}
+
+void AMDGPUCypressDevice::setCaps() {
+  if (mSTM->isOverride(AMDGPUDeviceInfo::DoubleOps)) {
+    mHWBits.set(AMDGPUDeviceInfo::DoubleOps);
+    mHWBits.set(AMDGPUDeviceInfo::FMA);
+  }
+}
+
+
+AMDGPUCedarDevice::AMDGPUCedarDevice(AMDGPUSubtarget *ST)
+  : AMDGPUEvergreenDevice(ST) {
+  setCaps();
+}
+
+AMDGPUCedarDevice::~AMDGPUCedarDevice() {
+}
+
+void AMDGPUCedarDevice::setCaps() {
+  mSWBits.set(AMDGPUDeviceInfo::FMA);
+}
+
+size_t AMDGPUCedarDevice::getWavefrontSize() const {
+  return AMDGPUDevice::QuarterWavefrontSize;
+}
+
+AMDGPURedwoodDevice::AMDGPURedwoodDevice(AMDGPUSubtarget *ST)
+  : AMDGPUEvergreenDevice(ST) {
+  setCaps();
+}
+
+AMDGPURedwoodDevice::~AMDGPURedwoodDevice() {
+}
+
+void AMDGPURedwoodDevice::setCaps() {
+  mSWBits.set(AMDGPUDeviceInfo::FMA);
+}
+
+size_t AMDGPURedwoodDevice::getWavefrontSize() const {
+  return AMDGPUDevice::HalfWavefrontSize;
+}
diff --git a/lib/Target/R600/AMDILEvergreenDevice.h b/lib/Target/R600/AMDILEvergreenDevice.h
new file mode 100644
index 000000000000..ea90f774a856
--- /dev/null
+++ b/lib/Target/R600/AMDILEvergreenDevice.h
@@ -0,0 +1,93 @@
+//==- AMDILEvergreenDevice.h - Define Evergreen Device for AMDIL -*- C++ -*--=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Interface for the subtarget data classes.
+///
+/// This file will define the interface that each generation needs to
+/// implement in order to correctly answer queries on the capabilities of the
+/// specific hardware.
+//===----------------------------------------------------------------------===//
+#ifndef AMDILEVERGREENDEVICE_H
+#define AMDILEVERGREENDEVICE_H
+#include "AMDGPUSubtarget.h"
+#include "AMDILDevice.h"
+
+namespace llvm {
+  class AMDGPUSubtarget;
+//===----------------------------------------------------------------------===//
+// Evergreen generation of devices and their respective sub classes
+//===----------------------------------------------------------------------===//
+
+
+/// \brief The AMDGPUEvergreenDevice is the base device class for all of the Evergreen
+/// series of cards.
+///
+/// This class contains information required to differentiate
+/// the Evergreen device from the generic AMDGPUDevice. This device represents
+/// that capabilities of the 'Juniper' cards, also known as the HD57XX.
+class AMDGPUEvergreenDevice : public AMDGPUDevice {
+public:
+  AMDGPUEvergreenDevice(AMDGPUSubtarget *ST);
+  virtual ~AMDGPUEvergreenDevice();
+  virtual size_t getMaxLDSSize() const;
+  virtual size_t getMaxGDSSize() const;
+  virtual size_t getWavefrontSize() const;
+  virtual uint32_t getGeneration() const;
+  virtual uint32_t getMaxNumUAVs() const;
+  virtual uint32_t getResourceID(uint32_t) const;
+protected:
+  virtual void setCaps();
+};
+
+/// The AMDGPUCypressDevice is similiar to the AMDGPUEvergreenDevice, except it has
+/// support for double precision operations. This device is used to represent
+/// both the Cypress and Hemlock cards, which are commercially known as HD58XX
+/// and HD59XX cards.
+class AMDGPUCypressDevice : public AMDGPUEvergreenDevice {
+public:
+  AMDGPUCypressDevice(AMDGPUSubtarget *ST);
+  virtual ~AMDGPUCypressDevice();
+private:
+  virtual void setCaps();
+};
+
+
+/// \brief The AMDGPUCedarDevice is the class that represents all of the 'Cedar' based
+/// devices.
+///
+/// This class differs from the base AMDGPUEvergreenDevice in that the
+/// device is a ~quarter of the 'Juniper'. These are commercially known as the
+/// HD54XX and HD53XX series of cards.
+class AMDGPUCedarDevice : public AMDGPUEvergreenDevice {
+public:
+  AMDGPUCedarDevice(AMDGPUSubtarget *ST);
+  virtual ~AMDGPUCedarDevice();
+  virtual size_t getWavefrontSize() const;
+private:
+  virtual void setCaps();
+};
+
+/// \brief The AMDGPURedwoodDevice is the class the represents all of the 'Redwood' based
+/// devices.
+///
+/// This class differs from the base class, in that these devices are
+/// considered about half of a 'Juniper' device. These are commercially known as
+/// the HD55XX and HD56XX series of cards.
+class AMDGPURedwoodDevice : public AMDGPUEvergreenDevice {
+public:
+  AMDGPURedwoodDevice(AMDGPUSubtarget *ST);
+  virtual ~AMDGPURedwoodDevice();
+  virtual size_t getWavefrontSize() const;
+private:
+  virtual void setCaps();
+};
+  
+} // namespace llvm
+#endif // AMDILEVERGREENDEVICE_H
diff --git a/lib/Target/R600/AMDILISelDAGToDAG.cpp b/lib/Target/R600/AMDILISelDAGToDAG.cpp
new file mode 100644
index 000000000000..fa8f62de9c0a
--- /dev/null
+++ b/lib/Target/R600/AMDILISelDAGToDAG.cpp
@@ -0,0 +1,643 @@
+//===-- AMDILISelDAGToDAG.cpp - A dag to dag inst selector for AMDIL ------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Defines an instruction selector for the AMDGPU target.
+//
+//===----------------------------------------------------------------------===//
+#include "AMDGPUInstrInfo.h"
+#include "AMDGPUISelLowering.h" // For AMDGPUISD
+#include "AMDGPURegisterInfo.h"
+#include "AMDILDevices.h"
+#include "R600InstrInfo.h"
+#include "SIISelLowering.h"
+#include "llvm/ADT/ValueMap.h"
+#include "llvm/CodeGen/PseudoSourceValue.h"
+#include "llvm/CodeGen/SelectionDAGISel.h"
+#include "llvm/Support/Compiler.h"
+#include "llvm/CodeGen/SelectionDAG.h"
+#include <list>
+#include <queue>
+
+using namespace llvm;
+
+//===----------------------------------------------------------------------===//
+// Instruction Selector Implementation
+//===----------------------------------------------------------------------===//
+
+namespace {
+/// AMDGPU specific code to select AMDGPU machine instructions for
+/// SelectionDAG operations.
+class AMDGPUDAGToDAGISel : public SelectionDAGISel {
+  // Subtarget - Keep a pointer to the AMDGPU Subtarget around so that we can
+  // make the right decision when generating code for different targets.
+  const AMDGPUSubtarget &Subtarget;
+public:
+  AMDGPUDAGToDAGISel(TargetMachine &TM);
+  virtual ~AMDGPUDAGToDAGISel();
+
+  SDNode *Select(SDNode *N);
+  virtual const char *getPassName() const;
+  virtual void PostprocessISelDAG();
+
+private:
+  inline SDValue getSmallIPtrImm(unsigned Imm);
+  bool FoldOperands(unsigned, const R600InstrInfo *, std::vector<SDValue> &);
+
+  // Complex pattern selectors
+  bool SelectADDRParam(SDValue Addr, SDValue& R1, SDValue& R2);
+  bool SelectADDR(SDValue N, SDValue &R1, SDValue &R2);
+  bool SelectADDR64(SDValue N, SDValue &R1, SDValue &R2);
+
+  static bool checkType(const Value *ptr, unsigned int addrspace);
+  static const Value *getBasePointerValue(const Value *V);
+
+  static bool isGlobalStore(const StoreSDNode *N);
+  static bool isPrivateStore(const StoreSDNode *N);
+  static bool isLocalStore(const StoreSDNode *N);
+  static bool isRegionStore(const StoreSDNode *N);
+
+  static bool isCPLoad(const LoadSDNode *N);
+  static bool isConstantLoad(const LoadSDNode *N, int cbID);
+  static bool isGlobalLoad(const LoadSDNode *N);
+  static bool isParamLoad(const LoadSDNode *N);
+  static bool isPrivateLoad(const LoadSDNode *N);
+  static bool isLocalLoad(const LoadSDNode *N);
+  static bool isRegionLoad(const LoadSDNode *N);
+
+  bool SelectGlobalValueConstantOffset(SDValue Addr, SDValue& IntPtr);
+  bool SelectGlobalValueVariableOffset(SDValue Addr,
+      SDValue &BaseReg, SDValue& Offset);
+  bool SelectADDRVTX_READ(SDValue Addr, SDValue &Base, SDValue &Offset);
+  bool SelectADDRIndirect(SDValue Addr, SDValue &Base, SDValue &Offset);
+
+  // Include the pieces autogenerated from the target description.
+#include "AMDGPUGenDAGISel.inc"
+};
+}  // end anonymous namespace
+
+/// \brief This pass converts a legalized DAG into a AMDGPU-specific
+// DAG, ready for instruction scheduling.
+FunctionPass *llvm::createAMDGPUISelDag(TargetMachine &TM
+                                       ) {
+  return new AMDGPUDAGToDAGISel(TM);
+}
+
+AMDGPUDAGToDAGISel::AMDGPUDAGToDAGISel(TargetMachine &TM
+                                     )
+  : SelectionDAGISel(TM), Subtarget(TM.getSubtarget<AMDGPUSubtarget>()) {
+}
+
+AMDGPUDAGToDAGISel::~AMDGPUDAGToDAGISel() {
+}
+
+SDValue AMDGPUDAGToDAGISel::getSmallIPtrImm(unsigned int Imm) {
+  return CurDAG->getTargetConstant(Imm, MVT::i32);
+}
+
+bool AMDGPUDAGToDAGISel::SelectADDRParam(
+    SDValue Addr, SDValue& R1, SDValue& R2) {
+
+  if (Addr.getOpcode() == ISD::FrameIndex) {
+    if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) {
+      R1 = CurDAG->getTargetFrameIndex(FIN->getIndex(), MVT::i32);
+      R2 = CurDAG->getTargetConstant(0, MVT::i32);
+    } else {
+      R1 = Addr;
+      R2 = CurDAG->getTargetConstant(0, MVT::i32);
+    }
+  } else if (Addr.getOpcode() == ISD::ADD) {
+    R1 = Addr.getOperand(0);
+    R2 = Addr.getOperand(1);
+  } else {
+    R1 = Addr;
+    R2 = CurDAG->getTargetConstant(0, MVT::i32);
+  }
+  return true;
+}
+
+bool AMDGPUDAGToDAGISel::SelectADDR(SDValue Addr, SDValue& R1, SDValue& R2) {
+  if (Addr.getOpcode() == ISD::TargetExternalSymbol ||
+      Addr.getOpcode() == ISD::TargetGlobalAddress) {
+    return false;
+  }
+  return SelectADDRParam(Addr, R1, R2);
+}
+
+
+bool AMDGPUDAGToDAGISel::SelectADDR64(SDValue Addr, SDValue& R1, SDValue& R2) {
+  if (Addr.getOpcode() == ISD::TargetExternalSymbol ||
+      Addr.getOpcode() == ISD::TargetGlobalAddress) {
+    return false;
+  }
+
+  if (Addr.getOpcode() == ISD::FrameIndex) {
+    if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) {
+      R1 = CurDAG->getTargetFrameIndex(FIN->getIndex(), MVT::i64);
+      R2 = CurDAG->getTargetConstant(0, MVT::i64);
+    } else {
+      R1 = Addr;
+      R2 = CurDAG->getTargetConstant(0, MVT::i64);
+    }
+  } else if (Addr.getOpcode() == ISD::ADD) {
+    R1 = Addr.getOperand(0);
+    R2 = Addr.getOperand(1);
+  } else {
+    R1 = Addr;
+    R2 = CurDAG->getTargetConstant(0, MVT::i64);
+  }
+  return true;
+}
+
+SDNode *AMDGPUDAGToDAGISel::Select(SDNode *N) {
+  unsigned int Opc = N->getOpcode();
+  if (N->isMachineOpcode()) {
+    return NULL;   // Already selected.
+  }
+  switch (Opc) {
+  default: break;
+  case ISD::BUILD_VECTOR: {
+    const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
+    if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
+      break;
+    }
+    // BUILD_VECTOR is usually lowered into an IMPLICIT_DEF + 4 INSERT_SUBREG
+    // that adds a 128 bits reg copy when going through TwoAddressInstructions
+    // pass. We want to avoid 128 bits copies as much as possible because they
+    // can't be bundled by our scheduler.
+    SDValue RegSeqArgs[9] = {
+      CurDAG->getTargetConstant(AMDGPU::R600_Reg128RegClassID, MVT::i32),
+      SDValue(), CurDAG->getTargetConstant(AMDGPU::sub0, MVT::i32),
+      SDValue(), CurDAG->getTargetConstant(AMDGPU::sub1, MVT::i32),
+      SDValue(), CurDAG->getTargetConstant(AMDGPU::sub2, MVT::i32),
+      SDValue(), CurDAG->getTargetConstant(AMDGPU::sub3, MVT::i32)
+    };
+    bool IsRegSeq = true;
+    for (unsigned i = 0; i < N->getNumOperands(); i++) {
+      if (dyn_cast<RegisterSDNode>(N->getOperand(i))) {
+        IsRegSeq = false;
+        break;
+      }
+      RegSeqArgs[2 * i + 1] = N->getOperand(i);
+    }
+    if (!IsRegSeq)
+      break;
+    return CurDAG->SelectNodeTo(N, AMDGPU::REG_SEQUENCE, N->getVTList(),
+        RegSeqArgs, 2 * N->getNumOperands() + 1);
+  }
+  case ISD::ConstantFP:
+  case ISD::Constant: {
+    const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
+    // XXX: Custom immediate lowering not implemented yet.  Instead we use
+    // pseudo instructions defined in SIInstructions.td
+    if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
+      break;
+    }
+    const R600InstrInfo *TII = static_cast<const R600InstrInfo*>(TM.getInstrInfo());
+
+    uint64_t ImmValue = 0;
+    unsigned ImmReg = AMDGPU::ALU_LITERAL_X;
+
+    if (N->getOpcode() == ISD::ConstantFP) {
+      // XXX: 64-bit Immediates not supported yet
+      assert(N->getValueType(0) != MVT::f64);
+
+      ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(N);
+      APFloat Value = C->getValueAPF();
+      float FloatValue = Value.convertToFloat();
+      if (FloatValue == 0.0) {
+        ImmReg = AMDGPU::ZERO;
+      } else if (FloatValue == 0.5) {
+        ImmReg = AMDGPU::HALF;
+      } else if (FloatValue == 1.0) {
+        ImmReg = AMDGPU::ONE;
+      } else {
+        ImmValue = Value.bitcastToAPInt().getZExtValue();
+      }
+    } else {
+      // XXX: 64-bit Immediates not supported yet
+      assert(N->getValueType(0) != MVT::i64);
+
+      ConstantSDNode *C = dyn_cast<ConstantSDNode>(N);
+      if (C->getZExtValue() == 0) {
+        ImmReg = AMDGPU::ZERO;
+      } else if (C->getZExtValue() == 1) {
+        ImmReg = AMDGPU::ONE_INT;
+      } else {
+        ImmValue = C->getZExtValue();
+      }
+    }
+
+    for (SDNode::use_iterator Use = N->use_begin(), Next = llvm::next(Use);
+                              Use != SDNode::use_end(); Use = Next) {
+      Next = llvm::next(Use);
+      std::vector<SDValue> Ops;
+      for (unsigned i = 0; i < Use->getNumOperands(); ++i) {
+        Ops.push_back(Use->getOperand(i));
+      }
+
+      if (!Use->isMachineOpcode()) {
+          if (ImmReg == AMDGPU::ALU_LITERAL_X) {
+            // We can only use literal constants (e.g. AMDGPU::ZERO,
+            // AMDGPU::ONE, etc) in machine opcodes.
+            continue;
+          }
+      } else {
+        if (!TII->isALUInstr(Use->getMachineOpcode()) ||
+            (TII->get(Use->getMachineOpcode()).TSFlags &
+            R600_InstFlag::VECTOR)) {
+          continue;
+        }
+
+        int ImmIdx = TII->getOperandIdx(Use->getMachineOpcode(), R600Operands::IMM);
+        assert(ImmIdx != -1);
+
+        // subtract one from ImmIdx, because the DST operand is usually index
+        // 0 for MachineInstrs, but we have no DST in the Ops vector.
+        ImmIdx--;
+
+        // Check that we aren't already using an immediate.
+        // XXX: It's possible for an instruction to have more than one
+        // immediate operand, but this is not supported yet.
+        if (ImmReg == AMDGPU::ALU_LITERAL_X) {
+          ConstantSDNode *C = dyn_cast<ConstantSDNode>(Use->getOperand(ImmIdx));
+          assert(C);
+
+          if (C->getZExtValue() != 0) {
+            // This instruction is already using an immediate.
+            continue;
+          }
+
+          // Set the immediate value
+          Ops[ImmIdx] = CurDAG->getTargetConstant(ImmValue, MVT::i32);
+        }
+      }
+      // Set the immediate register
+      Ops[Use.getOperandNo()] = CurDAG->getRegister(ImmReg, MVT::i32);
+
+      CurDAG->UpdateNodeOperands(*Use, Ops.data(), Use->getNumOperands());
+    }
+    break;
+  }
+  }
+  SDNode *Result = SelectCode(N);
+
+  // Fold operands of selected node
+
+  const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
+  if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
+    const R600InstrInfo *TII =
+        static_cast<const R600InstrInfo*>(TM.getInstrInfo());
+    if (Result && Result->isMachineOpcode() &&
+        !(TII->get(Result->getMachineOpcode()).TSFlags & R600_InstFlag::VECTOR)
+        && TII->isALUInstr(Result->getMachineOpcode())) {
+      // Fold FNEG/FABS/CONST_ADDRESS
+      // TODO: Isel can generate multiple MachineInst, we need to recursively
+      // parse Result
+      bool IsModified = false;
+      do {
+        std::vector<SDValue> Ops;
+        for(SDNode::op_iterator I = Result->op_begin(), E = Result->op_end();
+            I != E; ++I)
+          Ops.push_back(*I);
+        IsModified = FoldOperands(Result->getMachineOpcode(), TII, Ops);
+        if (IsModified) {
+          Result = CurDAG->UpdateNodeOperands(Result, Ops.data(), Ops.size());
+        }
+      } while (IsModified);
+
+      // If node has a single use which is CLAMP_R600, folds it
+      if (Result->hasOneUse() && Result->isMachineOpcode()) {
+        SDNode *PotentialClamp = *Result->use_begin();
+        if (PotentialClamp->isMachineOpcode() &&
+            PotentialClamp->getMachineOpcode() == AMDGPU::CLAMP_R600) {
+          unsigned ClampIdx =
+            TII->getOperandIdx(Result->getMachineOpcode(), R600Operands::CLAMP);
+          std::vector<SDValue> Ops;
+          unsigned NumOp = Result->getNumOperands();
+          for (unsigned i = 0; i < NumOp; ++i) {
+            Ops.push_back(Result->getOperand(i));
+          }
+          Ops[ClampIdx - 1] = CurDAG->getTargetConstant(1, MVT::i32);
+          Result = CurDAG->SelectNodeTo(PotentialClamp,
+              Result->getMachineOpcode(), PotentialClamp->getVTList(),
+              Ops.data(), NumOp);
+        }
+      }
+    }
+  }
+
+  return Result;
+}
+
+bool AMDGPUDAGToDAGISel::FoldOperands(unsigned Opcode,
+    const R600InstrInfo *TII, std::vector<SDValue> &Ops) {
+  int OperandIdx[] = {
+    TII->getOperandIdx(Opcode, R600Operands::SRC0),
+    TII->getOperandIdx(Opcode, R600Operands::SRC1),
+    TII->getOperandIdx(Opcode, R600Operands::SRC2)
+  };
+  int SelIdx[] = {
+    TII->getOperandIdx(Opcode, R600Operands::SRC0_SEL),
+    TII->getOperandIdx(Opcode, R600Operands::SRC1_SEL),
+    TII->getOperandIdx(Opcode, R600Operands::SRC2_SEL)
+  };
+  int NegIdx[] = {
+    TII->getOperandIdx(Opcode, R600Operands::SRC0_NEG),
+    TII->getOperandIdx(Opcode, R600Operands::SRC1_NEG),
+    TII->getOperandIdx(Opcode, R600Operands::SRC2_NEG)
+  };
+  int AbsIdx[] = {
+    TII->getOperandIdx(Opcode, R600Operands::SRC0_ABS),
+    TII->getOperandIdx(Opcode, R600Operands::SRC1_ABS),
+    -1
+  };
+
+  for (unsigned i = 0; i < 3; i++) {
+    if (OperandIdx[i] < 0)
+      return false;
+    SDValue Operand = Ops[OperandIdx[i] - 1];
+    switch (Operand.getOpcode()) {
+    case AMDGPUISD::CONST_ADDRESS: {
+      SDValue CstOffset;
+      if (Operand.getValueType().isVector() ||
+          !SelectGlobalValueConstantOffset(Operand.getOperand(0), CstOffset))
+        break;
+
+      // Gather others constants values
+      std::vector<unsigned> Consts;
+      for (unsigned j = 0; j < 3; j++) {
+        int SrcIdx = OperandIdx[j];
+        if (SrcIdx < 0)
+          break;
+        if (RegisterSDNode *Reg = dyn_cast<RegisterSDNode>(Ops[SrcIdx - 1])) {
+          if (Reg->getReg() == AMDGPU::ALU_CONST) {
+            ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(Ops[SelIdx[j] - 1]);
+            Consts.push_back(Cst->getZExtValue());
+          }
+        }
+      }
+
+      ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(CstOffset);
+      Consts.push_back(Cst->getZExtValue());
+      if (!TII->fitsConstReadLimitations(Consts))
+        break;
+
+      Ops[OperandIdx[i] - 1] = CurDAG->getRegister(AMDGPU::ALU_CONST, MVT::f32);
+      Ops[SelIdx[i] - 1] = CstOffset;
+      return true;
+      }
+    case ISD::FNEG:
+      if (NegIdx[i] < 0)
+        break;
+      Ops[OperandIdx[i] - 1] = Operand.getOperand(0);
+      Ops[NegIdx[i] - 1] = CurDAG->getTargetConstant(1, MVT::i32);
+      return true;
+    case ISD::FABS:
+      if (AbsIdx[i] < 0)
+        break;
+      Ops[OperandIdx[i] - 1] = Operand.getOperand(0);
+      Ops[AbsIdx[i] - 1] = CurDAG->getTargetConstant(1, MVT::i32);
+      return true;
+    case ISD::BITCAST:
+      Ops[OperandIdx[i] - 1] = Operand.getOperand(0);
+      return true;
+    default:
+      break;
+    }
+  }
+  return false;
+}
+
+bool AMDGPUDAGToDAGISel::checkType(const Value *ptr, unsigned int addrspace) {
+  if (!ptr) {
+    return false;
+  }
+  Type *ptrType = ptr->getType();
+  return dyn_cast<PointerType>(ptrType)->getAddressSpace() == addrspace;
+}
+
+const Value * AMDGPUDAGToDAGISel::getBasePointerValue(const Value *V) {
+  if (!V) {
+    return NULL;
+  }
+  const Value *ret = NULL;
+  ValueMap<const Value *, bool> ValueBitMap;
+  std::queue<const Value *, std::list<const Value *> > ValueQueue;
+  ValueQueue.push(V);
+  while (!ValueQueue.empty()) {
+    V = ValueQueue.front();
+    if (ValueBitMap.find(V) == ValueBitMap.end()) {
+      ValueBitMap[V] = true;
+      if (dyn_cast<Argument>(V) && dyn_cast<PointerType>(V->getType())) {
+        ret = V;
+        break;
+      } else if (dyn_cast<GlobalVariable>(V)) {
+        ret = V;
+        break;
+      } else if (dyn_cast<Constant>(V)) {
+        const ConstantExpr *CE = dyn_cast<ConstantExpr>(V);
+        if (CE) {
+          ValueQueue.push(CE->getOperand(0));
+        }
+      } else if (const AllocaInst *AI = dyn_cast<AllocaInst>(V)) {
+        ret = AI;
+        break;
+      } else if (const Instruction *I = dyn_cast<Instruction>(V)) {
+        uint32_t numOps = I->getNumOperands();
+        for (uint32_t x = 0; x < numOps; ++x) {
+          ValueQueue.push(I->getOperand(x));
+        }
+      } else {
+        assert(!"Found a Value that we didn't know how to handle!");
+      }
+    }
+    ValueQueue.pop();
+  }
+  return ret;
+}
+
+bool AMDGPUDAGToDAGISel::isGlobalStore(const StoreSDNode *N) {
+  return checkType(N->getSrcValue(), AMDGPUAS::GLOBAL_ADDRESS);
+}
+
+bool AMDGPUDAGToDAGISel::isPrivateStore(const StoreSDNode *N) {
+  return (!checkType(N->getSrcValue(), AMDGPUAS::LOCAL_ADDRESS)
+          && !checkType(N->getSrcValue(), AMDGPUAS::GLOBAL_ADDRESS)
+          && !checkType(N->getSrcValue(), AMDGPUAS::REGION_ADDRESS));
+}
+
+bool AMDGPUDAGToDAGISel::isLocalStore(const StoreSDNode *N) {
+  return checkType(N->getSrcValue(), AMDGPUAS::LOCAL_ADDRESS);
+}
+
+bool AMDGPUDAGToDAGISel::isRegionStore(const StoreSDNode *N) {
+  return checkType(N->getSrcValue(), AMDGPUAS::REGION_ADDRESS);
+}
+
+bool AMDGPUDAGToDAGISel::isConstantLoad(const LoadSDNode *N, int cbID) {
+  if (checkType(N->getSrcValue(), AMDGPUAS::CONSTANT_ADDRESS)) {
+    return true;
+  }
+  MachineMemOperand *MMO = N->getMemOperand();
+  const Value *V = MMO->getValue();
+  const Value *BV = getBasePointerValue(V);
+  if (MMO
+      && MMO->getValue()
+      && ((V && dyn_cast<GlobalValue>(V))
+          || (BV && dyn_cast<GlobalValue>(
+                        getBasePointerValue(MMO->getValue()))))) {
+    return checkType(N->getSrcValue(), AMDGPUAS::PRIVATE_ADDRESS);
+  } else {
+    return false;
+  }
+}
+
+bool AMDGPUDAGToDAGISel::isGlobalLoad(const LoadSDNode *N) {
+  return checkType(N->getSrcValue(), AMDGPUAS::GLOBAL_ADDRESS);
+}
+
+bool AMDGPUDAGToDAGISel::isParamLoad(const LoadSDNode *N) {
+  return checkType(N->getSrcValue(), AMDGPUAS::PARAM_I_ADDRESS);
+}
+
+bool AMDGPUDAGToDAGISel::isLocalLoad(const  LoadSDNode *N) {
+  return checkType(N->getSrcValue(), AMDGPUAS::LOCAL_ADDRESS);
+}
+
+bool AMDGPUDAGToDAGISel::isRegionLoad(const  LoadSDNode *N) {
+  return checkType(N->getSrcValue(), AMDGPUAS::REGION_ADDRESS);
+}
+
+bool AMDGPUDAGToDAGISel::isCPLoad(const LoadSDNode *N) {
+  MachineMemOperand *MMO = N->getMemOperand();
+  if (checkType(N->getSrcValue(), AMDGPUAS::PRIVATE_ADDRESS)) {
+    if (MMO) {
+      const Value *V = MMO->getValue();
+      const PseudoSourceValue *PSV = dyn_cast<PseudoSourceValue>(V);
+      if (PSV && PSV == PseudoSourceValue::getConstantPool()) {
+        return true;
+      }
+    }
+  }
+  return false;
+}
+
+bool AMDGPUDAGToDAGISel::isPrivateLoad(const LoadSDNode *N) {
+  if (checkType(N->getSrcValue(), AMDGPUAS::PRIVATE_ADDRESS)) {
+    // Check to make sure we are not a constant pool load or a constant load
+    // that is marked as a private load
+    if (isCPLoad(N) || isConstantLoad(N, -1)) {
+      return false;
+    }
+  }
+  if (!checkType(N->getSrcValue(), AMDGPUAS::LOCAL_ADDRESS)
+      && !checkType(N->getSrcValue(), AMDGPUAS::GLOBAL_ADDRESS)
+      && !checkType(N->getSrcValue(), AMDGPUAS::REGION_ADDRESS)
+      && !checkType(N->getSrcValue(), AMDGPUAS::CONSTANT_ADDRESS)
+      && !checkType(N->getSrcValue(), AMDGPUAS::PARAM_D_ADDRESS)
+      && !checkType(N->getSrcValue(), AMDGPUAS::PARAM_I_ADDRESS)) {
+    return true;
+  }
+  return false;
+}
+
+const char *AMDGPUDAGToDAGISel::getPassName() const {
+  return "AMDGPU DAG->DAG Pattern Instruction Selection";
+}
+
+#ifdef DEBUGTMP
+#undef INT64_C
+#endif
+#undef DEBUGTMP
+
+///==== AMDGPU Functions ====///
+
+bool AMDGPUDAGToDAGISel::SelectGlobalValueConstantOffset(SDValue Addr,
+    SDValue& IntPtr) {
+  if (ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(Addr)) {
+    IntPtr = CurDAG->getIntPtrConstant(Cst->getZExtValue() / 4, true);
+    return true;
+  }
+  return false;
+}
+
+bool AMDGPUDAGToDAGISel::SelectGlobalValueVariableOffset(SDValue Addr,
+    SDValue& BaseReg, SDValue &Offset) {
+  if (!dyn_cast<ConstantSDNode>(Addr)) {
+    BaseReg = Addr;
+    Offset = CurDAG->getIntPtrConstant(0, true);
+    return true;
+  }
+  return false;
+}
+
+bool AMDGPUDAGToDAGISel::SelectADDRVTX_READ(SDValue Addr, SDValue &Base,
+                                           SDValue &Offset) {
+  ConstantSDNode * IMMOffset;
+
+  if (Addr.getOpcode() == ISD::ADD
+      && (IMMOffset = dyn_cast<ConstantSDNode>(Addr.getOperand(1)))
+      && isInt<16>(IMMOffset->getZExtValue())) {
+
+      Base = Addr.getOperand(0);
+      Offset = CurDAG->getTargetConstant(IMMOffset->getZExtValue(), MVT::i32);
+      return true;
+  // If the pointer address is constant, we can move it to the offset field.
+  } else if ((IMMOffset = dyn_cast<ConstantSDNode>(Addr))
+             && isInt<16>(IMMOffset->getZExtValue())) {
+    Base = CurDAG->getCopyFromReg(CurDAG->getEntryNode(),
+                                  CurDAG->getEntryNode().getDebugLoc(),
+                                  AMDGPU::ZERO, MVT::i32);
+    Offset = CurDAG->getTargetConstant(IMMOffset->getZExtValue(), MVT::i32);
+    return true;
+  }
+
+  // Default case, no offset
+  Base = Addr;
+  Offset = CurDAG->getTargetConstant(0, MVT::i32);
+  return true;
+}
+
+bool AMDGPUDAGToDAGISel::SelectADDRIndirect(SDValue Addr, SDValue &Base,
+                                            SDValue &Offset) {
+  ConstantSDNode *C;
+
+  if ((C = dyn_cast<ConstantSDNode>(Addr))) {
+    Base = CurDAG->getRegister(AMDGPU::INDIRECT_BASE_ADDR, MVT::i32);
+    Offset = CurDAG->getTargetConstant(C->getZExtValue(), MVT::i32);
+  } else if ((Addr.getOpcode() == ISD::ADD || Addr.getOpcode() == ISD::OR) &&
+            (C = dyn_cast<ConstantSDNode>(Addr.getOperand(1)))) {
+    Base = Addr.getOperand(0);
+    Offset = CurDAG->getTargetConstant(C->getZExtValue(), MVT::i32);
+  } else {
+    Base = Addr;
+    Offset = CurDAG->getTargetConstant(0, MVT::i32);
+  }
+
+  return true;
+}
+
+void AMDGPUDAGToDAGISel::PostprocessISelDAG() {
+
+  // Go over all selected nodes and try to fold them a bit more
+  const AMDGPUTargetLowering& Lowering = ((const AMDGPUTargetLowering&)TLI);
+  for (SelectionDAG::allnodes_iterator I = CurDAG->allnodes_begin(),
+       E = CurDAG->allnodes_end(); I != E; ++I) {
+
+    MachineSDNode *Node = dyn_cast<MachineSDNode>(I);
+    if (!Node)
+      continue;
+
+    SDNode *ResNode = Lowering.PostISelFolding(Node, *CurDAG);
+    if (ResNode != Node)
+      ReplaceUses(Node, ResNode);
+  }
+}
+
diff --git a/lib/Target/R600/AMDILISelLowering.cpp b/lib/Target/R600/AMDILISelLowering.cpp
new file mode 100644
index 000000000000..922cac12b98e
--- /dev/null
+++ b/lib/Target/R600/AMDILISelLowering.cpp
@@ -0,0 +1,647 @@
+//===-- AMDILISelLowering.cpp - AMDIL DAG Lowering Implementation ---------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief TargetLowering functions borrowed from AMDIL.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUISelLowering.h"
+#include "AMDGPURegisterInfo.h"
+#include "AMDGPUSubtarget.h"
+#include "AMDILDevices.h"
+#include "AMDILIntrinsicInfo.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/PseudoSourceValue.h"
+#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/CodeGen/SelectionDAGNodes.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetOptions.h"
+
+using namespace llvm;
+//===----------------------------------------------------------------------===//
+// TargetLowering Implementation Help Functions End
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// TargetLowering Class Implementation Begins
+//===----------------------------------------------------------------------===//
+void AMDGPUTargetLowering::InitAMDILLowering() {
+  int types[] = {
+    (int)MVT::i8,
+    (int)MVT::i16,
+    (int)MVT::i32,
+    (int)MVT::f32,
+    (int)MVT::f64,
+    (int)MVT::i64,
+    (int)MVT::v2i8,
+    (int)MVT::v4i8,
+    (int)MVT::v2i16,
+    (int)MVT::v4i16,
+    (int)MVT::v4f32,
+    (int)MVT::v4i32,
+    (int)MVT::v2f32,
+    (int)MVT::v2i32,
+    (int)MVT::v2f64,
+    (int)MVT::v2i64
+  };
+
+  int IntTypes[] = {
+    (int)MVT::i8,
+    (int)MVT::i16,
+    (int)MVT::i32,
+    (int)MVT::i64
+  };
+
+  int FloatTypes[] = {
+    (int)MVT::f32,
+    (int)MVT::f64
+  };
+
+  int VectorTypes[] = {
+    (int)MVT::v2i8,
+    (int)MVT::v4i8,
+    (int)MVT::v2i16,
+    (int)MVT::v4i16,
+    (int)MVT::v4f32,
+    (int)MVT::v4i32,
+    (int)MVT::v2f32,
+    (int)MVT::v2i32,
+    (int)MVT::v2f64,
+    (int)MVT::v2i64
+  };
+  size_t NumTypes = sizeof(types) / sizeof(*types);
+  size_t NumFloatTypes = sizeof(FloatTypes) / sizeof(*FloatTypes);
+  size_t NumIntTypes = sizeof(IntTypes) / sizeof(*IntTypes);
+  size_t NumVectorTypes = sizeof(VectorTypes) / sizeof(*VectorTypes);
+
+  const AMDGPUSubtarget &STM = getTargetMachine().getSubtarget<AMDGPUSubtarget>();
+  // These are the current register classes that are
+  // supported
+
+  for (unsigned int x  = 0; x < NumTypes; ++x) {
+    MVT::SimpleValueType VT = (MVT::SimpleValueType)types[x];
+
+    //FIXME: SIGN_EXTEND_INREG is not meaningful for floating point types
+    // We cannot sextinreg, expand to shifts
+    setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Custom);
+    setOperationAction(ISD::SUBE, VT, Expand);
+    setOperationAction(ISD::SUBC, VT, Expand);
+    setOperationAction(ISD::ADDE, VT, Expand);
+    setOperationAction(ISD::ADDC, VT, Expand);
+    setOperationAction(ISD::BRCOND, VT, Custom);
+    setOperationAction(ISD::BR_JT, VT, Expand);
+    setOperationAction(ISD::BRIND, VT, Expand);
+    // TODO: Implement custom UREM/SREM routines
+    setOperationAction(ISD::SREM, VT, Expand);
+    setOperationAction(ISD::SMUL_LOHI, VT, Expand);
+    setOperationAction(ISD::UMUL_LOHI, VT, Expand);
+    if (VT != MVT::i64 && VT != MVT::v2i64) {
+      setOperationAction(ISD::SDIV, VT, Custom);
+    }
+  }
+  for (unsigned int x = 0; x < NumFloatTypes; ++x) {
+    MVT::SimpleValueType VT = (MVT::SimpleValueType)FloatTypes[x];
+
+    // IL does not have these operations for floating point types
+    setOperationAction(ISD::FP_ROUND_INREG, VT, Expand);
+    setOperationAction(ISD::SETOLT, VT, Expand);
+    setOperationAction(ISD::SETOGE, VT, Expand);
+    setOperationAction(ISD::SETOGT, VT, Expand);
+    setOperationAction(ISD::SETOLE, VT, Expand);
+    setOperationAction(ISD::SETULT, VT, Expand);
+    setOperationAction(ISD::SETUGE, VT, Expand);
+    setOperationAction(ISD::SETUGT, VT, Expand);
+    setOperationAction(ISD::SETULE, VT, Expand);
+  }
+
+  for (unsigned int x = 0; x < NumIntTypes; ++x) {
+    MVT::SimpleValueType VT = (MVT::SimpleValueType)IntTypes[x];
+
+    // GPU also does not have divrem function for signed or unsigned
+    setOperationAction(ISD::SDIVREM, VT, Expand);
+
+    // GPU does not have [S|U]MUL_LOHI functions as a single instruction
+    setOperationAction(ISD::SMUL_LOHI, VT, Expand);
+    setOperationAction(ISD::UMUL_LOHI, VT, Expand);
+
+    // GPU doesn't have a rotl, rotr, or byteswap instruction
+    setOperationAction(ISD::ROTR, VT, Expand);
+    setOperationAction(ISD::BSWAP, VT, Expand);
+
+    // GPU doesn't have any counting operators
+    setOperationAction(ISD::CTPOP, VT, Expand);
+    setOperationAction(ISD::CTTZ, VT, Expand);
+    setOperationAction(ISD::CTLZ, VT, Expand);
+  }
+
+  for (unsigned int ii = 0; ii < NumVectorTypes; ++ii) {
+    MVT::SimpleValueType VT = (MVT::SimpleValueType)VectorTypes[ii];
+
+    setOperationAction(ISD::VECTOR_SHUFFLE, VT, Expand);
+    setOperationAction(ISD::SDIVREM, VT, Expand);
+    setOperationAction(ISD::SMUL_LOHI, VT, Expand);
+    // setOperationAction(ISD::VSETCC, VT, Expand);
+    setOperationAction(ISD::SELECT_CC, VT, Expand);
+
+  }
+  if (STM.device()->isSupported(AMDGPUDeviceInfo::LongOps)) {
+    setOperationAction(ISD::MULHU, MVT::i64, Expand);
+    setOperationAction(ISD::MULHU, MVT::v2i64, Expand);
+    setOperationAction(ISD::MULHS, MVT::i64, Expand);
+    setOperationAction(ISD::MULHS, MVT::v2i64, Expand);
+    setOperationAction(ISD::ADD, MVT::v2i64, Expand);
+    setOperationAction(ISD::SREM, MVT::v2i64, Expand);
+    setOperationAction(ISD::Constant          , MVT::i64  , Legal);
+    setOperationAction(ISD::SDIV, MVT::v2i64, Expand);
+    setOperationAction(ISD::TRUNCATE, MVT::v2i64, Expand);
+    setOperationAction(ISD::SIGN_EXTEND, MVT::v2i64, Expand);
+    setOperationAction(ISD::ZERO_EXTEND, MVT::v2i64, Expand);
+    setOperationAction(ISD::ANY_EXTEND, MVT::v2i64, Expand);
+  }
+  if (STM.device()->isSupported(AMDGPUDeviceInfo::DoubleOps)) {
+    // we support loading/storing v2f64 but not operations on the type
+    setOperationAction(ISD::FADD, MVT::v2f64, Expand);
+    setOperationAction(ISD::FSUB, MVT::v2f64, Expand);
+    setOperationAction(ISD::FMUL, MVT::v2f64, Expand);
+    setOperationAction(ISD::FP_ROUND_INREG, MVT::v2f64, Expand);
+    setOperationAction(ISD::FP_EXTEND, MVT::v2f64, Expand);
+    setOperationAction(ISD::ConstantFP        , MVT::f64  , Legal);
+    // We want to expand vector conversions into their scalar
+    // counterparts.
+    setOperationAction(ISD::TRUNCATE, MVT::v2f64, Expand);
+    setOperationAction(ISD::SIGN_EXTEND, MVT::v2f64, Expand);
+    setOperationAction(ISD::ZERO_EXTEND, MVT::v2f64, Expand);
+    setOperationAction(ISD::ANY_EXTEND, MVT::v2f64, Expand);
+    setOperationAction(ISD::FABS, MVT::f64, Expand);
+    setOperationAction(ISD::FABS, MVT::v2f64, Expand);
+  }
+  // TODO: Fix the UDIV24 algorithm so it works for these
+  // types correctly. This needs vector comparisons
+  // for this to work correctly.
+  setOperationAction(ISD::UDIV, MVT::v2i8, Expand);
+  setOperationAction(ISD::UDIV, MVT::v4i8, Expand);
+  setOperationAction(ISD::UDIV, MVT::v2i16, Expand);
+  setOperationAction(ISD::UDIV, MVT::v4i16, Expand);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Custom);
+  setOperationAction(ISD::SUBC, MVT::Other, Expand);
+  setOperationAction(ISD::ADDE, MVT::Other, Expand);
+  setOperationAction(ISD::ADDC, MVT::Other, Expand);
+  setOperationAction(ISD::BRCOND, MVT::Other, Custom);
+  setOperationAction(ISD::BR_JT, MVT::Other, Expand);
+  setOperationAction(ISD::BRIND, MVT::Other, Expand);
+  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::Other, Expand);
+
+
+  // Use the default implementation.
+  setOperationAction(ISD::ConstantFP        , MVT::f32    , Legal);
+  setOperationAction(ISD::Constant          , MVT::i32    , Legal);
+
+  setSchedulingPreference(Sched::RegPressure);
+  setPow2DivIsCheap(false);
+  setSelectIsExpensive(true);
+  setJumpIsExpensive(true);
+
+  MaxStoresPerMemcpy  = 4096;
+  MaxStoresPerMemmove = 4096;
+  MaxStoresPerMemset  = 4096;
+
+}
+
+bool
+AMDGPUTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
+    const CallInst &I, unsigned Intrinsic) const {
+  return false;
+}
+
+// The backend supports 32 and 64 bit floating point immediates
+bool
+AMDGPUTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
+  if (VT.getScalarType().getSimpleVT().SimpleTy == MVT::f32
+      || VT.getScalarType().getSimpleVT().SimpleTy == MVT::f64) {
+    return true;
+  } else {
+    return false;
+  }
+}
+
+bool
+AMDGPUTargetLowering::ShouldShrinkFPConstant(EVT VT) const {
+  if (VT.getScalarType().getSimpleVT().SimpleTy == MVT::f32
+      || VT.getScalarType().getSimpleVT().SimpleTy == MVT::f64) {
+    return false;
+  } else {
+    return true;
+  }
+}
+
+
+// isMaskedValueZeroForTargetNode - Return true if 'Op & Mask' is known to
+// be zero. Op is expected to be a target specific node. Used by DAG
+// combiner.
+
+void
+AMDGPUTargetLowering::computeMaskedBitsForTargetNode(
+    const SDValue Op,
+    APInt &KnownZero,
+    APInt &KnownOne,
+    const SelectionDAG &DAG,
+    unsigned Depth) const {
+  APInt KnownZero2;
+  APInt KnownOne2;
+  KnownZero = KnownOne = APInt(KnownOne.getBitWidth(), 0); // Don't know anything
+  switch (Op.getOpcode()) {
+    default: break;
+    case ISD::SELECT_CC:
+             DAG.ComputeMaskedBits(
+                 Op.getOperand(1),
+                 KnownZero,
+                 KnownOne,
+                 Depth + 1
+                 );
+             DAG.ComputeMaskedBits(
+                 Op.getOperand(0),
+                 KnownZero2,
+                 KnownOne2
+                 );
+             assert((KnownZero & KnownOne) == 0
+                 && "Bits known to be one AND zero?");
+             assert((KnownZero2 & KnownOne2) == 0
+                 && "Bits known to be one AND zero?");
+             // Only known if known in both the LHS and RHS
+             KnownOne &= KnownOne2;
+             KnownZero &= KnownZero2;
+             break;
+  };
+}
+
+//===----------------------------------------------------------------------===//
+//                           Other Lowering Hooks
+//===----------------------------------------------------------------------===//
+
+SDValue
+AMDGPUTargetLowering::LowerSDIV(SDValue Op, SelectionDAG &DAG) const {
+  EVT OVT = Op.getValueType();
+  SDValue DST;
+  if (OVT.getScalarType() == MVT::i64) {
+    DST = LowerSDIV64(Op, DAG);
+  } else if (OVT.getScalarType() == MVT::i32) {
+    DST = LowerSDIV32(Op, DAG);
+  } else if (OVT.getScalarType() == MVT::i16
+      || OVT.getScalarType() == MVT::i8) {
+    DST = LowerSDIV24(Op, DAG);
+  } else {
+    DST = SDValue(Op.getNode(), 0);
+  }
+  return DST;
+}
+
+SDValue
+AMDGPUTargetLowering::LowerSREM(SDValue Op, SelectionDAG &DAG) const {
+  EVT OVT = Op.getValueType();
+  SDValue DST;
+  if (OVT.getScalarType() == MVT::i64) {
+    DST = LowerSREM64(Op, DAG);
+  } else if (OVT.getScalarType() == MVT::i32) {
+    DST = LowerSREM32(Op, DAG);
+  } else if (OVT.getScalarType() == MVT::i16) {
+    DST = LowerSREM16(Op, DAG);
+  } else if (OVT.getScalarType() == MVT::i8) {
+    DST = LowerSREM8(Op, DAG);
+  } else {
+    DST = SDValue(Op.getNode(), 0);
+  }
+  return DST;
+}
+
+SDValue
+AMDGPUTargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const {
+  SDValue Data = Op.getOperand(0);
+  VTSDNode *BaseType = cast<VTSDNode>(Op.getOperand(1));
+  DebugLoc DL = Op.getDebugLoc();
+  EVT DVT = Data.getValueType();
+  EVT BVT = BaseType->getVT();
+  unsigned baseBits = BVT.getScalarType().getSizeInBits();
+  unsigned srcBits = DVT.isSimple() ? DVT.getScalarType().getSizeInBits() : 1;
+  unsigned shiftBits = srcBits - baseBits;
+  if (srcBits < 32) {
+    // If the op is less than 32 bits, then it needs to extend to 32bits
+    // so it can properly keep the upper bits valid.
+    EVT IVT = genIntType(32, DVT.isVector() ? DVT.getVectorNumElements() : 1);
+    Data = DAG.getNode(ISD::ZERO_EXTEND, DL, IVT, Data);
+    shiftBits = 32 - baseBits;
+    DVT = IVT;
+  }
+  SDValue Shift = DAG.getConstant(shiftBits, DVT);
+  // Shift left by 'Shift' bits.
+  Data = DAG.getNode(ISD::SHL, DL, DVT, Data, Shift);
+  // Signed shift Right by 'Shift' bits.
+  Data = DAG.getNode(ISD::SRA, DL, DVT, Data, Shift);
+  if (srcBits < 32) {
+    // Once the sign extension is done, the op needs to be converted to
+    // its original type.
+    Data = DAG.getSExtOrTrunc(Data, DL, Op.getOperand(0).getValueType());
+  }
+  return Data;
+}
+EVT
+AMDGPUTargetLowering::genIntType(uint32_t size, uint32_t numEle) const {
+  int iSize = (size * numEle);
+  int vEle = (iSize >> ((size == 64) ? 6 : 5));
+  if (!vEle) {
+    vEle = 1;
+  }
+  if (size == 64) {
+    if (vEle == 1) {
+      return EVT(MVT::i64);
+    } else {
+      return EVT(MVT::getVectorVT(MVT::i64, vEle));
+    }
+  } else {
+    if (vEle == 1) {
+      return EVT(MVT::i32);
+    } else {
+      return EVT(MVT::getVectorVT(MVT::i32, vEle));
+    }
+  }
+}
+
+SDValue
+AMDGPUTargetLowering::LowerBRCOND(SDValue Op, SelectionDAG &DAG) const {
+  SDValue Chain = Op.getOperand(0);
+  SDValue Cond  = Op.getOperand(1);
+  SDValue Jump  = Op.getOperand(2);
+  SDValue Result;
+  Result = DAG.getNode(
+      AMDGPUISD::BRANCH_COND,
+      Op.getDebugLoc(),
+      Op.getValueType(),
+      Chain, Jump, Cond);
+  return Result;
+}
+
+SDValue
+AMDGPUTargetLowering::LowerSDIV24(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT OVT = Op.getValueType();
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  MVT INTTY;
+  MVT FLTTY;
+  if (!OVT.isVector()) {
+    INTTY = MVT::i32;
+    FLTTY = MVT::f32;
+  } else if (OVT.getVectorNumElements() == 2) {
+    INTTY = MVT::v2i32;
+    FLTTY = MVT::v2f32;
+  } else if (OVT.getVectorNumElements() == 4) {
+    INTTY = MVT::v4i32;
+    FLTTY = MVT::v4f32;
+  }
+  unsigned bitsize = OVT.getScalarType().getSizeInBits();
+  // char|short jq = ia ^ ib;
+  SDValue jq = DAG.getNode(ISD::XOR, DL, OVT, LHS, RHS);
+
+  // jq = jq >> (bitsize - 2)
+  jq = DAG.getNode(ISD::SRA, DL, OVT, jq, DAG.getConstant(bitsize - 2, OVT)); 
+
+  // jq = jq | 0x1
+  jq = DAG.getNode(ISD::OR, DL, OVT, jq, DAG.getConstant(1, OVT));
+
+  // jq = (int)jq
+  jq = DAG.getSExtOrTrunc(jq, DL, INTTY);
+
+  // int ia = (int)LHS;
+  SDValue ia = DAG.getSExtOrTrunc(LHS, DL, INTTY);
+
+  // int ib, (int)RHS;
+  SDValue ib = DAG.getSExtOrTrunc(RHS, DL, INTTY);
+
+  // float fa = (float)ia;
+  SDValue fa = DAG.getNode(ISD::SINT_TO_FP, DL, FLTTY, ia);
+
+  // float fb = (float)ib;
+  SDValue fb = DAG.getNode(ISD::SINT_TO_FP, DL, FLTTY, ib);
+
+  // float fq = native_divide(fa, fb);
+  SDValue fq = DAG.getNode(AMDGPUISD::DIV_INF, DL, FLTTY, fa, fb);
+
+  // fq = trunc(fq);
+  fq = DAG.getNode(ISD::FTRUNC, DL, FLTTY, fq);
+
+  // float fqneg = -fq;
+  SDValue fqneg = DAG.getNode(ISD::FNEG, DL, FLTTY, fq);
+
+  // float fr = mad(fqneg, fb, fa);
+  SDValue fr = DAG.getNode(ISD::FADD, DL, FLTTY,
+      DAG.getNode(ISD::MUL, DL, FLTTY, fqneg, fb), fa);
+
+  // int iq = (int)fq;
+  SDValue iq = DAG.getNode(ISD::FP_TO_SINT, DL, INTTY, fq);
+
+  // fr = fabs(fr);
+  fr = DAG.getNode(ISD::FABS, DL, FLTTY, fr);
+
+  // fb = fabs(fb);
+  fb = DAG.getNode(ISD::FABS, DL, FLTTY, fb);
+
+  // int cv = fr >= fb;
+  SDValue cv;
+  if (INTTY == MVT::i32) {
+    cv = DAG.getSetCC(DL, INTTY, fr, fb, ISD::SETOGE);
+  } else {
+    cv = DAG.getSetCC(DL, INTTY, fr, fb, ISD::SETOGE);
+  }
+  // jq = (cv ? jq : 0);
+  jq = DAG.getNode(ISD::SELECT, DL, OVT, cv, jq, 
+      DAG.getConstant(0, OVT));
+  // dst = iq + jq;
+  iq = DAG.getSExtOrTrunc(iq, DL, OVT);
+  iq = DAG.getNode(ISD::ADD, DL, OVT, iq, jq);
+  return iq;
+}
+
+SDValue
+AMDGPUTargetLowering::LowerSDIV32(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT OVT = Op.getValueType();
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  // The LowerSDIV32 function generates equivalent to the following IL.
+  // mov r0, LHS
+  // mov r1, RHS
+  // ilt r10, r0, 0
+  // ilt r11, r1, 0
+  // iadd r0, r0, r10
+  // iadd r1, r1, r11
+  // ixor r0, r0, r10
+  // ixor r1, r1, r11
+  // udiv r0, r0, r1
+  // ixor r10, r10, r11
+  // iadd r0, r0, r10
+  // ixor DST, r0, r10
+
+  // mov r0, LHS
+  SDValue r0 = LHS;
+
+  // mov r1, RHS
+  SDValue r1 = RHS;
+
+  // ilt r10, r0, 0
+  SDValue r10 = DAG.getSelectCC(DL,
+      r0, DAG.getConstant(0, OVT),
+      DAG.getConstant(-1, MVT::i32),
+      DAG.getConstant(0, MVT::i32),
+      ISD::SETLT);
+
+  // ilt r11, r1, 0
+  SDValue r11 = DAG.getSelectCC(DL,
+      r1, DAG.getConstant(0, OVT),
+      DAG.getConstant(-1, MVT::i32),
+      DAG.getConstant(0, MVT::i32),
+      ISD::SETLT);
+
+  // iadd r0, r0, r10
+  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
+
+  // iadd r1, r1, r11
+  r1 = DAG.getNode(ISD::ADD, DL, OVT, r1, r11);
+
+  // ixor r0, r0, r10
+  r0 = DAG.getNode(ISD::XOR, DL, OVT, r0, r10);
+
+  // ixor r1, r1, r11
+  r1 = DAG.getNode(ISD::XOR, DL, OVT, r1, r11);
+
+  // udiv r0, r0, r1
+  r0 = DAG.getNode(ISD::UDIV, DL, OVT, r0, r1);
+
+  // ixor r10, r10, r11
+  r10 = DAG.getNode(ISD::XOR, DL, OVT, r10, r11);
+
+  // iadd r0, r0, r10
+  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
+
+  // ixor DST, r0, r10
+  SDValue DST = DAG.getNode(ISD::XOR, DL, OVT, r0, r10); 
+  return DST;
+}
+
+SDValue
+AMDGPUTargetLowering::LowerSDIV64(SDValue Op, SelectionDAG &DAG) const {
+  return SDValue(Op.getNode(), 0);
+}
+
+SDValue
+AMDGPUTargetLowering::LowerSREM8(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT OVT = Op.getValueType();
+  MVT INTTY = MVT::i32;
+  if (OVT == MVT::v2i8) {
+    INTTY = MVT::v2i32;
+  } else if (OVT == MVT::v4i8) {
+    INTTY = MVT::v4i32;
+  }
+  SDValue LHS = DAG.getSExtOrTrunc(Op.getOperand(0), DL, INTTY);
+  SDValue RHS = DAG.getSExtOrTrunc(Op.getOperand(1), DL, INTTY);
+  LHS = DAG.getNode(ISD::SREM, DL, INTTY, LHS, RHS);
+  LHS = DAG.getSExtOrTrunc(LHS, DL, OVT);
+  return LHS;
+}
+
+SDValue
+AMDGPUTargetLowering::LowerSREM16(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT OVT = Op.getValueType();
+  MVT INTTY = MVT::i32;
+  if (OVT == MVT::v2i16) {
+    INTTY = MVT::v2i32;
+  } else if (OVT == MVT::v4i16) {
+    INTTY = MVT::v4i32;
+  }
+  SDValue LHS = DAG.getSExtOrTrunc(Op.getOperand(0), DL, INTTY);
+  SDValue RHS = DAG.getSExtOrTrunc(Op.getOperand(1), DL, INTTY);
+  LHS = DAG.getNode(ISD::SREM, DL, INTTY, LHS, RHS);
+  LHS = DAG.getSExtOrTrunc(LHS, DL, OVT);
+  return LHS;
+}
+
+SDValue
+AMDGPUTargetLowering::LowerSREM32(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT OVT = Op.getValueType();
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  // The LowerSREM32 function generates equivalent to the following IL.
+  // mov r0, LHS
+  // mov r1, RHS
+  // ilt r10, r0, 0
+  // ilt r11, r1, 0
+  // iadd r0, r0, r10
+  // iadd r1, r1, r11
+  // ixor r0, r0, r10
+  // ixor r1, r1, r11
+  // udiv r20, r0, r1
+  // umul r20, r20, r1
+  // sub r0, r0, r20
+  // iadd r0, r0, r10
+  // ixor DST, r0, r10
+
+  // mov r0, LHS
+  SDValue r0 = LHS;
+
+  // mov r1, RHS
+  SDValue r1 = RHS;
+
+  // ilt r10, r0, 0
+  SDValue r10 = DAG.getSetCC(DL, OVT, r0, DAG.getConstant(0, OVT), ISD::SETLT);
+
+  // ilt r11, r1, 0
+  SDValue r11 = DAG.getSetCC(DL, OVT, r1, DAG.getConstant(0, OVT), ISD::SETLT);
+
+  // iadd r0, r0, r10
+  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
+
+  // iadd r1, r1, r11
+  r1 = DAG.getNode(ISD::ADD, DL, OVT, r1, r11);
+
+  // ixor r0, r0, r10
+  r0 = DAG.getNode(ISD::XOR, DL, OVT, r0, r10);
+
+  // ixor r1, r1, r11
+  r1 = DAG.getNode(ISD::XOR, DL, OVT, r1, r11);
+
+  // udiv r20, r0, r1
+  SDValue r20 = DAG.getNode(ISD::UREM, DL, OVT, r0, r1);
+
+  // umul r20, r20, r1
+  r20 = DAG.getNode(AMDGPUISD::UMUL, DL, OVT, r20, r1);
+
+  // sub r0, r0, r20
+  r0 = DAG.getNode(ISD::SUB, DL, OVT, r0, r20);
+
+  // iadd r0, r0, r10
+  r0 = DAG.getNode(ISD::ADD, DL, OVT, r0, r10);
+
+  // ixor DST, r0, r10
+  SDValue DST = DAG.getNode(ISD::XOR, DL, OVT, r0, r10); 
+  return DST;
+}
+
+SDValue
+AMDGPUTargetLowering::LowerSREM64(SDValue Op, SelectionDAG &DAG) const {
+  return SDValue(Op.getNode(), 0);
+}
diff --git a/lib/Target/R600/AMDILInstrInfo.td b/lib/Target/R600/AMDILInstrInfo.td
new file mode 100644
index 000000000000..110f1476513b
--- /dev/null
+++ b/lib/Target/R600/AMDILInstrInfo.td
@@ -0,0 +1,207 @@
+//===------------ AMDILInstrInfo.td - AMDIL Target ------*-tablegen-*------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+// This file describes the AMDIL instructions in TableGen format.
+//
+//===----------------------------------------------------------------------===//
+// AMDIL Instruction Predicate Definitions
+// Predicate that is set to true if the hardware supports double precision
+// divide
+def HasHWDDiv                 : Predicate<"Subtarget.device()"
+                           "->getGeneration() > AMDGPUDeviceInfo::HD4XXX && "
+              "Subtarget.device()->usesHardware(AMDGPUDeviceInfo::DoubleOps)">;
+
+// Predicate that is set to true if the hardware supports double, but not double
+// precision divide in hardware
+def HasSWDDiv             : Predicate<"Subtarget.device()"
+                           "->getGeneration() == AMDGPUDeviceInfo::HD4XXX &&"
+              "Subtarget.device()->usesHardware(AMDGPUDeviceInfo::DoubleOps)">;
+
+// Predicate that is set to true if the hardware support 24bit signed
+// math ops. Otherwise a software expansion to 32bit math ops is used instead.
+def HasHWSign24Bit          : Predicate<"Subtarget.device()"
+                            "->getGeneration() > AMDGPUDeviceInfo::HD5XXX">;
+
+// Predicate that is set to true if 64bit operations are supported or not
+def HasHW64Bit              : Predicate<"Subtarget.device()"
+                            "->usesHardware(AMDGPUDeviceInfo::LongOps)">;
+def HasSW64Bit              : Predicate<"Subtarget.device()"
+                            "->usesSoftware(AMDGPUDeviceInfo::LongOps)">;
+
+// Predicate that is set to true if the timer register is supported
+def HasTmrRegister          : Predicate<"Subtarget.device()"
+                            "->isSupported(AMDGPUDeviceInfo::TmrReg)">;
+// Predicate that is true if we are at least evergreen series
+def HasDeviceIDInst         : Predicate<"Subtarget.device()"
+                            "->getGeneration() >= AMDGPUDeviceInfo::HD5XXX">;
+
+// Predicate that is true if we have region address space.
+def hasRegionAS             : Predicate<"Subtarget.device()"
+                            "->usesHardware(AMDGPUDeviceInfo::RegionMem)">;
+
+// Predicate that is false if we don't have region address space.
+def noRegionAS             : Predicate<"!Subtarget.device()"
+                            "->isSupported(AMDGPUDeviceInfo::RegionMem)">;
+
+
+// Predicate that is set to true if 64bit Mul is supported in the IL or not
+def HasHW64Mul              : Predicate<"Subtarget.calVersion()" 
+                                          ">= CAL_VERSION_SC_139"
+                                          "&& Subtarget.device()"
+                                          "->getGeneration() >="
+                                          "AMDGPUDeviceInfo::HD5XXX">;
+def HasSW64Mul              : Predicate<"Subtarget.calVersion()" 
+                                          "< CAL_VERSION_SC_139">;
+// Predicate that is set to true if 64bit Div/Mod is supported in the IL or not
+def HasHW64DivMod           : Predicate<"Subtarget.device()"
+                            "->usesHardware(AMDGPUDeviceInfo::HW64BitDivMod)">;
+def HasSW64DivMod           : Predicate<"Subtarget.device()"
+                            "->usesSoftware(AMDGPUDeviceInfo::HW64BitDivMod)">;
+
+// Predicate that is set to true if 64bit pointer are used.
+def Has64BitPtr             : Predicate<"Subtarget.is64bit()">;
+def Has32BitPtr             : Predicate<"!Subtarget.is64bit()">;
+//===--------------------------------------------------------------------===//
+// Custom Operands
+//===--------------------------------------------------------------------===//
+def brtarget   : Operand<OtherVT>;
+
+//===--------------------------------------------------------------------===//
+// Custom Selection DAG Type Profiles
+//===--------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+// Generic Profile Types
+//===----------------------------------------------------------------------===//
+
+def SDTIL_GenBinaryOp : SDTypeProfile<1, 2, [
+    SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>
+    ]>;
+def SDTIL_GenTernaryOp : SDTypeProfile<1, 3, [
+    SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>, SDTCisSameAs<2, 3>
+    ]>;
+def SDTIL_GenVecBuild : SDTypeProfile<1, 1, [
+    SDTCisEltOfVec<1, 0>
+    ]>;
+
+//===----------------------------------------------------------------------===//
+// Flow Control Profile Types
+//===----------------------------------------------------------------------===//
+// Branch instruction where second and third are basic blocks
+def SDTIL_BRCond : SDTypeProfile<0, 2, [
+    SDTCisVT<0, OtherVT>
+    ]>;
+
+//===--------------------------------------------------------------------===//
+// Custom Selection DAG Nodes
+//===--------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+// Flow Control DAG Nodes
+//===----------------------------------------------------------------------===//
+def IL_brcond      : SDNode<"AMDGPUISD::BRANCH_COND", SDTIL_BRCond, [SDNPHasChain]>;
+
+//===----------------------------------------------------------------------===//
+// Call/Return DAG Nodes
+//===----------------------------------------------------------------------===//
+def IL_retflag       : SDNode<"AMDGPUISD::RET_FLAG", SDTNone,
+    [SDNPHasChain, SDNPOptInGlue]>;
+
+//===--------------------------------------------------------------------===//
+// Instructions
+//===--------------------------------------------------------------------===//
+// Floating point math functions
+def IL_div_inf      : SDNode<"AMDGPUISD::DIV_INF", SDTIL_GenBinaryOp>;
+
+//===----------------------------------------------------------------------===//
+// Integer functions
+//===----------------------------------------------------------------------===//
+def IL_umul        : SDNode<"AMDGPUISD::UMUL"    , SDTIntBinOp,
+    [SDNPCommutative, SDNPAssociative]>;
+
+//===--------------------------------------------------------------------===//
+// Custom Pattern DAG Nodes
+//===--------------------------------------------------------------------===//
+def global_store : PatFrag<(ops node:$val, node:$ptr),
+    (store node:$val, node:$ptr), [{
+        return isGlobalStore(dyn_cast<StoreSDNode>(N));
+}]>;
+
+//===----------------------------------------------------------------------===//
+// Load pattern fragments
+//===----------------------------------------------------------------------===//
+// Global address space loads
+def global_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+    return isGlobalLoad(dyn_cast<LoadSDNode>(N));
+}]>;
+// Constant address space loads
+def constant_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+    return isConstantLoad(dyn_cast<LoadSDNode>(N), -1);
+}]>;
+
+//===----------------------------------------------------------------------===//
+// Complex addressing mode patterns
+//===----------------------------------------------------------------------===//
+def ADDR : ComplexPattern<i32, 2, "SelectADDR", [], []>;
+def ADDRF : ComplexPattern<i32, 2, "SelectADDR", [frameindex], []>;
+def ADDR64 : ComplexPattern<i64, 2, "SelectADDR64", [], []>;
+def ADDR64F : ComplexPattern<i64, 2, "SelectADDR64", [frameindex], []>;
+
+//===----------------------------------------------------------------------===//
+// Instruction format classes
+//===----------------------------------------------------------------------===//
+class ILFormat<dag outs, dag ins, string asmstr, list<dag> pattern>
+: Instruction {
+
+     let Namespace = "AMDGPU";
+     dag OutOperandList = outs;
+     dag InOperandList = ins;
+     let Pattern = pattern;
+     let AsmString = !strconcat(asmstr, "\n");
+     let isPseudo = 1;
+     let Itinerary = NullALU;
+     bit hasIEEEFlag = 0;
+     bit hasZeroOpFlag = 0;
+     let mayLoad = 0;
+     let mayStore = 0;
+     let hasSideEffects = 0;
+}
+
+//===--------------------------------------------------------------------===//
+// Multiclass Instruction formats
+//===--------------------------------------------------------------------===//
+// Multiclass that handles branch instructions
+multiclass BranchConditional<SDNode Op> {
+    def _i32 : ILFormat<(outs),
+  (ins brtarget:$target, GPRI32:$src0),
+        "; i32 Pseudo branch instruction",
+  [(Op bb:$target, GPRI32:$src0)]>;
+    def _f32 : ILFormat<(outs),
+  (ins brtarget:$target, GPRF32:$src0),
+        "; f32 Pseudo branch instruction",
+  [(Op bb:$target, GPRF32:$src0)]>;
+}
+
+// Only scalar types should generate flow control
+multiclass BranchInstr<string name> {
+  def _i32 : ILFormat<(outs), (ins GPRI32:$src),
+      !strconcat(name, " $src"), []>;
+  def _f32 : ILFormat<(outs), (ins GPRF32:$src),
+      !strconcat(name, " $src"), []>;
+}
+// Only scalar types should generate flow control
+multiclass BranchInstr2<string name> {
+  def _i32 : ILFormat<(outs), (ins GPRI32:$src0, GPRI32:$src1),
+      !strconcat(name, " $src0, $src1"), []>;
+  def _f32 : ILFormat<(outs), (ins GPRF32:$src0, GPRF32:$src1),
+      !strconcat(name, " $src0, $src1"), []>;
+}
+
+//===--------------------------------------------------------------------===//
+// Intrinsics support
+//===--------------------------------------------------------------------===//
+include "AMDILIntrinsics.td"
diff --git a/lib/Target/R600/AMDILIntrinsicInfo.cpp b/lib/Target/R600/AMDILIntrinsicInfo.cpp
new file mode 100644
index 000000000000..4ddb057d80a7
--- /dev/null
+++ b/lib/Target/R600/AMDILIntrinsicInfo.cpp
@@ -0,0 +1,79 @@
+//===- AMDILIntrinsicInfo.cpp - AMDGPU Intrinsic Information ------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief AMDGPU Implementation of the IntrinsicInfo class.
+//
+//===-----------------------------------------------------------------------===//
+
+#include "AMDILIntrinsicInfo.h"
+#include "AMDGPUSubtarget.h"
+#include "AMDIL.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Module.h"
+
+using namespace llvm;
+
+#define GET_LLVM_INTRINSIC_FOR_GCC_BUILTIN
+#include "AMDGPUGenIntrinsics.inc"
+#undef GET_LLVM_INTRINSIC_FOR_GCC_BUILTIN
+
+AMDGPUIntrinsicInfo::AMDGPUIntrinsicInfo(TargetMachine *tm) 
+  : TargetIntrinsicInfo() {
+}
+
+std::string 
+AMDGPUIntrinsicInfo::getName(unsigned int IntrID, Type **Tys,
+    unsigned int numTys) const  {
+  static const char* const names[] = {
+#define GET_INTRINSIC_NAME_TABLE
+#include "AMDGPUGenIntrinsics.inc"
+#undef GET_INTRINSIC_NAME_TABLE
+  };
+
+  if (IntrID < Intrinsic::num_intrinsics) {
+    return 0;
+  }
+  assert(IntrID < AMDGPUIntrinsic::num_AMDGPU_intrinsics
+      && "Invalid intrinsic ID");
+
+  std::string Result(names[IntrID - Intrinsic::num_intrinsics]);
+  return Result;
+}
+
+unsigned int
+AMDGPUIntrinsicInfo::lookupName(const char *Name, unsigned int Len) const  {
+#define GET_FUNCTION_RECOGNIZER
+#include "AMDGPUGenIntrinsics.inc"
+#undef GET_FUNCTION_RECOGNIZER
+  AMDGPUIntrinsic::ID IntrinsicID
+    = (AMDGPUIntrinsic::ID)Intrinsic::not_intrinsic;
+  IntrinsicID = getIntrinsicForGCCBuiltin("AMDGPU", Name);
+
+  if (IntrinsicID != (AMDGPUIntrinsic::ID)Intrinsic::not_intrinsic) {
+    return IntrinsicID;
+  }
+  return 0;
+}
+
+bool 
+AMDGPUIntrinsicInfo::isOverloaded(unsigned id) const  {
+  // Overload Table
+#define GET_INTRINSIC_OVERLOAD_TABLE
+#include "AMDGPUGenIntrinsics.inc"
+#undef GET_INTRINSIC_OVERLOAD_TABLE
+}
+
+Function*
+AMDGPUIntrinsicInfo::getDeclaration(Module *M, unsigned IntrID,
+    Type **Tys,
+    unsigned numTys) const  {
+  llvm_unreachable("Not implemented");
+}
diff --git a/lib/Target/R600/AMDILIntrinsicInfo.h b/lib/Target/R600/AMDILIntrinsicInfo.h
new file mode 100644
index 000000000000..35559e23fceb
--- /dev/null
+++ b/lib/Target/R600/AMDILIntrinsicInfo.h
@@ -0,0 +1,49 @@
+//===- AMDILIntrinsicInfo.h - AMDGPU Intrinsic Information ------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Interface for the AMDGPU Implementation of the Intrinsic Info class.
+//
+//===-----------------------------------------------------------------------===//
+#ifndef AMDIL_INTRINSICS_H
+#define AMDIL_INTRINSICS_H
+
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/Target/TargetIntrinsicInfo.h"
+
+namespace llvm {
+class TargetMachine;
+
+namespace AMDGPUIntrinsic {
+enum ID {
+  last_non_AMDGPU_intrinsic = Intrinsic::num_intrinsics - 1,
+#define GET_INTRINSIC_ENUM_VALUES
+#include "AMDGPUGenIntrinsics.inc"
+#undef GET_INTRINSIC_ENUM_VALUES
+      , num_AMDGPU_intrinsics
+};
+
+} // end namespace AMDGPUIntrinsic
+
+class AMDGPUIntrinsicInfo : public TargetIntrinsicInfo {
+public:
+  AMDGPUIntrinsicInfo(TargetMachine *tm);
+  std::string getName(unsigned int IntrId, Type **Tys = 0,
+                      unsigned int numTys = 0) const;
+  unsigned int lookupName(const char *Name, unsigned int Len) const;
+  bool isOverloaded(unsigned int IID) const;
+  Function *getDeclaration(Module *M, unsigned int ID,
+                           Type **Tys = 0,
+                           unsigned int numTys = 0) const;
+};
+
+} // end namespace llvm
+
+#endif // AMDIL_INTRINSICS_H
+
diff --git a/lib/Target/R600/AMDILIntrinsics.td b/lib/Target/R600/AMDILIntrinsics.td
new file mode 100644
index 000000000000..6ec3559af24c
--- /dev/null
+++ b/lib/Target/R600/AMDILIntrinsics.td
@@ -0,0 +1,232 @@
+//===- AMDILIntrinsics.td - Defines AMDIL Intrinscs -*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+// This file defines all of the amdil-specific intrinsics
+//
+//===---------------------------------------------------------------===//
+//===--------------------------------------------------------------------===//
+// Intrinsic classes
+// Generic versions of the above classes but for Target specific intrinsics
+// instead of SDNode patterns.
+//===--------------------------------------------------------------------===//
+let TargetPrefix = "AMDIL", isTarget = 1 in {
+     class VoidIntLong :
+          Intrinsic<[llvm_i64_ty], [], []>;
+     class VoidIntInt :
+          Intrinsic<[llvm_i32_ty], [], []>;
+     class VoidIntBool :
+          Intrinsic<[llvm_i32_ty], [], []>;
+     class UnaryIntInt :
+          Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>], [IntrNoMem]>;
+     class UnaryIntFloat :
+          Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
+     class ConvertIntFTOI :
+          Intrinsic<[llvm_anyint_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
+     class ConvertIntITOF :
+          Intrinsic<[llvm_anyfloat_ty], [llvm_anyint_ty], [IntrNoMem]>;
+     class UnaryIntNoRetInt :
+          Intrinsic<[], [llvm_anyint_ty], []>;
+     class UnaryIntNoRetFloat :
+          Intrinsic<[], [llvm_anyfloat_ty], []>;
+     class BinaryIntInt :
+          Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
+     class BinaryIntFloat :
+          Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
+     class BinaryIntNoRetInt :
+          Intrinsic<[], [llvm_anyint_ty, LLVMMatchType<0>], []>;
+     class BinaryIntNoRetFloat :
+          Intrinsic<[], [llvm_anyfloat_ty, LLVMMatchType<0>], []>;
+     class TernaryIntInt :
+          Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>,
+          LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
+     class TernaryIntFloat :
+          Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>,
+          LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
+     class QuaternaryIntInt :
+          Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>,
+          LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>], [IntrNoMem]>;
+     class UnaryAtomicInt :
+          Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty], [IntrReadWriteArgMem]>;
+     class BinaryAtomicInt :
+          Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty, llvm_i32_ty], [IntrReadWriteArgMem]>;
+     class TernaryAtomicInt :
+          Intrinsic<[llvm_i32_ty], [llvm_ptr_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty]>;
+     class UnaryAtomicIntNoRet :
+          Intrinsic<[], [llvm_ptr_ty, llvm_i32_ty], [IntrReadWriteArgMem]>;
+     class BinaryAtomicIntNoRet :
+          Intrinsic<[], [llvm_ptr_ty, llvm_i32_ty, llvm_i32_ty], [IntrReadWriteArgMem]>;
+     class TernaryAtomicIntNoRet :
+          Intrinsic<[], [llvm_ptr_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrReadWriteArgMem]>;
+}
+
+let TargetPrefix = "AMDIL", isTarget = 1 in {
+  def int_AMDIL_abs : GCCBuiltin<"__amdil_abs">, UnaryIntInt;
+
+  def int_AMDIL_bit_extract_i32 : GCCBuiltin<"__amdil_ibit_extract">,
+          TernaryIntInt;
+  def int_AMDIL_bit_extract_u32 : GCCBuiltin<"__amdil_ubit_extract">,
+          TernaryIntInt;
+  def int_AMDIL_bit_reverse_u32 : GCCBuiltin<"__amdil_ubit_reverse">,
+          UnaryIntInt;
+  def int_AMDIL_bit_count_i32 : GCCBuiltin<"__amdil_count_bits">,
+          UnaryIntInt;
+  def int_AMDIL_bit_find_first_lo : GCCBuiltin<"__amdil_ffb_lo">,
+          UnaryIntInt;
+  def int_AMDIL_bit_find_first_hi : GCCBuiltin<"__amdil_ffb_hi">,
+          UnaryIntInt;
+  def int_AMDIL_bit_find_first_sgn : GCCBuiltin<"__amdil_ffb_signed">,
+          UnaryIntInt;
+  def int_AMDIL_media_bitalign : GCCBuiltin<"__amdil_bitalign">,
+                    TernaryIntInt;
+  def int_AMDIL_media_bytealign : GCCBuiltin<"__amdil_bytealign">,
+                    TernaryIntInt;
+  def int_AMDIL_bit_insert_u32 : GCCBuiltin<"__amdil_ubit_insert">,
+                    QuaternaryIntInt;
+  def int_AMDIL_bfi : GCCBuiltin<"__amdil_bfi">,
+      TernaryIntInt;
+  def int_AMDIL_bfm : GCCBuiltin<"__amdil_bfm">,
+      BinaryIntInt;
+  def int_AMDIL_mulhi_i32 : GCCBuiltin<"__amdil_imul_high">,
+          BinaryIntInt;
+  def int_AMDIL_mulhi_u32 : GCCBuiltin<"__amdil_umul_high">,
+          BinaryIntInt;
+  def int_AMDIL_mul24_i32 : GCCBuiltin<"__amdil_imul24">,
+          BinaryIntInt;
+  def int_AMDIL_mul24_u32 : GCCBuiltin<"__amdil_umul24">,
+          BinaryIntInt;
+  def int_AMDIL_mulhi24_i32 : GCCBuiltin<"__amdil_imul24_high">,
+          BinaryIntInt;
+  def int_AMDIL_mulhi24_u32 : GCCBuiltin<"__amdil_umul24_high">,
+          BinaryIntInt;
+  def int_AMDIL_carry_i32 : GCCBuiltin<"__amdil_carry">,
+          BinaryIntInt;
+  def int_AMDIL_borrow_i32 : GCCBuiltin<"__amdil_borrow">,
+          BinaryIntInt;
+  def int_AMDIL_min_i32 : GCCBuiltin<"__amdil_imin">,
+          BinaryIntInt;
+  def int_AMDIL_min_u32 : GCCBuiltin<"__amdil_umin">,
+          BinaryIntInt;
+  def int_AMDIL_min     : GCCBuiltin<"__amdil_min">,
+          BinaryIntFloat;
+  def int_AMDIL_max_i32 : GCCBuiltin<"__amdil_imax">,
+          BinaryIntInt;
+  def int_AMDIL_max_u32 : GCCBuiltin<"__amdil_umax">,
+          BinaryIntInt;
+  def int_AMDIL_max     : GCCBuiltin<"__amdil_max">,
+          BinaryIntFloat;
+  def int_AMDIL_media_lerp_u4 : GCCBuiltin<"__amdil_u4lerp">,
+          TernaryIntInt;
+  def int_AMDIL_media_sad : GCCBuiltin<"__amdil_sad">,
+          TernaryIntInt;
+  def int_AMDIL_media_sad_hi : GCCBuiltin<"__amdil_sadhi">,
+          TernaryIntInt;
+  def int_AMDIL_fraction : GCCBuiltin<"__amdil_fraction">,
+          UnaryIntFloat;
+  def int_AMDIL_clamp : GCCBuiltin<"__amdil_clamp">,
+          TernaryIntFloat;
+  def int_AMDIL_pireduce : GCCBuiltin<"__amdil_pireduce">,
+          UnaryIntFloat;
+  def int_AMDIL_round_nearest : GCCBuiltin<"__amdil_round_nearest">,
+          UnaryIntFloat;
+  def int_AMDIL_round_neginf : GCCBuiltin<"__amdil_round_neginf">,
+          UnaryIntFloat;
+  def int_AMDIL_round_zero : GCCBuiltin<"__amdil_round_zero">,
+          UnaryIntFloat;
+  def int_AMDIL_acos : GCCBuiltin<"__amdil_acos">,
+          UnaryIntFloat;
+  def int_AMDIL_atan : GCCBuiltin<"__amdil_atan">,
+          UnaryIntFloat;
+  def int_AMDIL_asin : GCCBuiltin<"__amdil_asin">,
+          UnaryIntFloat;
+  def int_AMDIL_cos : GCCBuiltin<"__amdil_cos">,
+          UnaryIntFloat;
+  def int_AMDIL_cos_vec : GCCBuiltin<"__amdil_cos_vec">,
+          UnaryIntFloat;
+  def int_AMDIL_tan : GCCBuiltin<"__amdil_tan">,
+          UnaryIntFloat;
+  def int_AMDIL_sin : GCCBuiltin<"__amdil_sin">,
+          UnaryIntFloat;
+  def int_AMDIL_sin_vec : GCCBuiltin<"__amdil_sin_vec">,
+          UnaryIntFloat;
+  def int_AMDIL_pow : GCCBuiltin<"__amdil_pow">, BinaryIntFloat;
+  def int_AMDIL_div : GCCBuiltin<"__amdil_div">, BinaryIntFloat;
+  def int_AMDIL_udiv : GCCBuiltin<"__amdil_udiv">, BinaryIntInt;
+  def int_AMDIL_sqrt: GCCBuiltin<"__amdil_sqrt">,
+          UnaryIntFloat;
+  def int_AMDIL_sqrt_vec: GCCBuiltin<"__amdil_sqrt_vec">,
+          UnaryIntFloat;
+  def int_AMDIL_exp : GCCBuiltin<"__amdil_exp">,
+          UnaryIntFloat;
+  def int_AMDIL_exp_vec : GCCBuiltin<"__amdil_exp_vec">,
+          UnaryIntFloat;
+  def int_AMDIL_exn : GCCBuiltin<"__amdil_exn">,
+          UnaryIntFloat;
+  def int_AMDIL_log_vec : GCCBuiltin<"__amdil_log_vec">,
+          UnaryIntFloat;
+  def int_AMDIL_ln : GCCBuiltin<"__amdil_ln">,
+          UnaryIntFloat;
+  def int_AMDIL_sign: GCCBuiltin<"__amdil_sign">,
+          UnaryIntFloat;
+  def int_AMDIL_fma: GCCBuiltin<"__amdil_fma">,
+          TernaryIntFloat;
+  def int_AMDIL_rsq : GCCBuiltin<"__amdil_rsq">,
+          UnaryIntFloat;
+  def int_AMDIL_rsq_vec : GCCBuiltin<"__amdil_rsq_vec">,
+          UnaryIntFloat;
+  def int_AMDIL_length : GCCBuiltin<"__amdil_length">,
+          UnaryIntFloat;
+  def int_AMDIL_lerp : GCCBuiltin<"__amdil_lerp">,
+          TernaryIntFloat;
+  def int_AMDIL_media_sad4 : GCCBuiltin<"__amdil_sad4">,
+      Intrinsic<[llvm_i32_ty], [llvm_v4i32_ty,
+           llvm_v4i32_ty, llvm_i32_ty], []>;
+
+  def int_AMDIL_frexp_f64 : GCCBuiltin<"__amdil_frexp">,
+        Intrinsic<[llvm_v2i64_ty], [llvm_double_ty], []>;
+ def int_AMDIL_ldexp : GCCBuiltin<"__amdil_ldexp">,
+    Intrinsic<[llvm_anyfloat_ty], [llvm_anyfloat_ty, llvm_anyint_ty], []>;
+  def int_AMDIL_drcp : GCCBuiltin<"__amdil_rcp">,
+      Intrinsic<[llvm_double_ty], [llvm_double_ty], []>;
+  def int_AMDIL_convert_f16_f32 : GCCBuiltin<"__amdil_half_to_float">,
+      ConvertIntITOF;
+  def int_AMDIL_convert_f32_f16 : GCCBuiltin<"__amdil_float_to_half">,
+      ConvertIntFTOI;
+  def int_AMDIL_convert_f32_i32_rpi : GCCBuiltin<"__amdil_float_to_int_rpi">,
+      ConvertIntFTOI;
+  def int_AMDIL_convert_f32_i32_flr : GCCBuiltin<"__amdil_float_to_int_flr">,
+      ConvertIntFTOI;
+  def int_AMDIL_convert_f32_f16_near : GCCBuiltin<"__amdil_float_to_half_near">,
+      ConvertIntFTOI;
+  def int_AMDIL_convert_f32_f16_neg_inf : GCCBuiltin<"__amdil_float_to_half_neg_inf">,
+      ConvertIntFTOI;
+  def int_AMDIL_convert_f32_f16_plus_inf : GCCBuiltin<"__amdil_float_to_half_plus_inf">,
+      ConvertIntFTOI;
+ def int_AMDIL_media_convert_f2v4u8 : GCCBuiltin<"__amdil_f_2_u4">,
+      Intrinsic<[llvm_i32_ty], [llvm_v4f32_ty], []>;
+  def int_AMDIL_media_unpack_byte_0 : GCCBuiltin<"__amdil_unpack_0">,
+      ConvertIntITOF;
+  def int_AMDIL_media_unpack_byte_1 : GCCBuiltin<"__amdil_unpack_1">,
+      ConvertIntITOF;
+  def int_AMDIL_media_unpack_byte_2 : GCCBuiltin<"__amdil_unpack_2">,
+      ConvertIntITOF;
+  def int_AMDIL_media_unpack_byte_3 : GCCBuiltin<"__amdil_unpack_3">,
+      ConvertIntITOF;
+  def int_AMDIL_dp2_add : GCCBuiltin<"__amdil_dp2_add">,
+        Intrinsic<[llvm_float_ty], [llvm_v2f32_ty,
+          llvm_v2f32_ty, llvm_float_ty], []>;
+  def int_AMDIL_dp2 : GCCBuiltin<"__amdil_dp2">,
+        Intrinsic<[llvm_float_ty], [llvm_v2f32_ty,
+          llvm_v2f32_ty], []>;
+  def int_AMDIL_dp3 : GCCBuiltin<"__amdil_dp3">,
+        Intrinsic<[llvm_float_ty], [llvm_v4f32_ty,
+          llvm_v4f32_ty], []>;
+  def int_AMDIL_dp4 : GCCBuiltin<"__amdil_dp4">,
+        Intrinsic<[llvm_float_ty], [llvm_v4f32_ty,
+          llvm_v4f32_ty], []>;
+}
diff --git a/lib/Target/R600/AMDILNIDevice.cpp b/lib/Target/R600/AMDILNIDevice.cpp
new file mode 100644
index 000000000000..47c3f7f209d6
--- /dev/null
+++ b/lib/Target/R600/AMDILNIDevice.cpp
@@ -0,0 +1,65 @@
+//===-- AMDILNIDevice.cpp - Device Info for Northern Islands devices ------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//==-----------------------------------------------------------------------===//
+#include "AMDILNIDevice.h"
+#include "AMDGPUSubtarget.h"
+#include "AMDILEvergreenDevice.h"
+
+using namespace llvm;
+
+AMDGPUNIDevice::AMDGPUNIDevice(AMDGPUSubtarget *ST)
+  : AMDGPUEvergreenDevice(ST) {
+  std::string name = ST->getDeviceName();
+  if (name == "caicos") {
+    DeviceFlag = OCL_DEVICE_CAICOS;
+  } else if (name == "turks") {
+    DeviceFlag = OCL_DEVICE_TURKS;
+  } else if (name == "cayman") {
+    DeviceFlag = OCL_DEVICE_CAYMAN;
+  } else {
+    DeviceFlag = OCL_DEVICE_BARTS;
+  }
+}
+AMDGPUNIDevice::~AMDGPUNIDevice() {
+}
+
+size_t
+AMDGPUNIDevice::getMaxLDSSize() const {
+  if (usesHardware(AMDGPUDeviceInfo::LocalMem)) {
+    return MAX_LDS_SIZE_900;
+  } else {
+    return 0;
+  }
+}
+
+uint32_t
+AMDGPUNIDevice::getGeneration() const {
+  return AMDGPUDeviceInfo::HD6XXX;
+}
+
+
+AMDGPUCaymanDevice::AMDGPUCaymanDevice(AMDGPUSubtarget *ST)
+  : AMDGPUNIDevice(ST) {
+  setCaps();
+}
+
+AMDGPUCaymanDevice::~AMDGPUCaymanDevice() {
+}
+
+void
+AMDGPUCaymanDevice::setCaps() {
+  if (mSTM->isOverride(AMDGPUDeviceInfo::DoubleOps)) {
+    mHWBits.set(AMDGPUDeviceInfo::DoubleOps);
+    mHWBits.set(AMDGPUDeviceInfo::FMA);
+  }
+  mHWBits.set(AMDGPUDeviceInfo::Signed24BitOps);
+  mSWBits.reset(AMDGPUDeviceInfo::Signed24BitOps);
+  mSWBits.set(AMDGPUDeviceInfo::ArenaSegment);
+}
+
diff --git a/lib/Target/R600/AMDILNIDevice.h b/lib/Target/R600/AMDILNIDevice.h
new file mode 100644
index 000000000000..24a640845eab
--- /dev/null
+++ b/lib/Target/R600/AMDILNIDevice.h
@@ -0,0 +1,57 @@
+//===------- AMDILNIDevice.h - Define NI Device for AMDIL -*- C++ -*------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+/// \file
+/// \brief Interface for the subtarget data classes.
+///
+/// This file will define the interface that each generation needs to
+/// implement in order to correctly answer queries on the capabilities of the
+/// specific hardware.
+//===---------------------------------------------------------------------===//
+#ifndef AMDILNIDEVICE_H
+#define AMDILNIDEVICE_H
+#include "AMDGPUSubtarget.h"
+#include "AMDILEvergreenDevice.h"
+
+namespace llvm {
+
+class AMDGPUSubtarget;
+//===---------------------------------------------------------------------===//
+// NI generation of devices and their respective sub classes
+//===---------------------------------------------------------------------===//
+
+/// \brief The AMDGPUNIDevice is the base class for all Northern Island series of
+/// cards.
+///
+/// It is very similiar to the AMDGPUEvergreenDevice, with the major
+/// exception being differences in wavefront size and hardware capabilities.  The
+/// NI devices are all 64 wide wavefronts and also add support for signed 24 bit
+/// integer operations
+class AMDGPUNIDevice : public AMDGPUEvergreenDevice {
+public:
+  AMDGPUNIDevice(AMDGPUSubtarget*);
+  virtual ~AMDGPUNIDevice();
+  virtual size_t getMaxLDSSize() const;
+  virtual uint32_t getGeneration() const;
+};
+
+/// Just as the AMDGPUCypressDevice is the double capable version of the
+/// AMDGPUEvergreenDevice, the AMDGPUCaymanDevice is the double capable version
+/// of the AMDGPUNIDevice.  The other major difference is that the Cayman Device
+/// has 4 wide ALU's, whereas the rest of the NI family is a 5 wide.
+class AMDGPUCaymanDevice: public AMDGPUNIDevice {
+public:
+  AMDGPUCaymanDevice(AMDGPUSubtarget*);
+  virtual ~AMDGPUCaymanDevice();
+private:
+  virtual void setCaps();
+};
+
+static const unsigned int MAX_LDS_SIZE_900 = AMDGPUDevice::MAX_LDS_SIZE_800;
+} // namespace llvm
+#endif // AMDILNIDEVICE_H
diff --git a/lib/Target/R600/AMDILPeepholeOptimizer.cpp b/lib/Target/R600/AMDILPeepholeOptimizer.cpp
new file mode 100644
index 000000000000..3a28038666f7
--- /dev/null
+++ b/lib/Target/R600/AMDILPeepholeOptimizer.cpp
@@ -0,0 +1,1215 @@
+//===-- AMDILPeepholeOptimizer.cpp - AMDGPU Peephole optimizations ---------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//==-----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "PeepholeOpt"
+#ifdef DEBUG
+#define DEBUGME (DebugFlag && isCurrentDebugType(DEBUG_TYPE))
+#else
+#define DEBUGME 0
+#endif
+
+#include "AMDILDevices.h"
+#include "AMDGPUInstrInfo.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/StringExtras.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/ADT/Twine.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Module.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/MathExtras.h"
+
+#include <sstream>
+
+#if 0
+STATISTIC(PointerAssignments, "Number of dynamic pointer "
+    "assigments discovered");
+STATISTIC(PointerSubtract, "Number of pointer subtractions discovered");
+#endif
+
+using namespace llvm;
+// The Peephole optimization pass is used to do simple last minute optimizations
+// that are required for correct code or to remove redundant functions
+namespace {
+
+class OpaqueType;
+
+class LLVM_LIBRARY_VISIBILITY AMDGPUPeepholeOpt : public FunctionPass {
+public:
+  TargetMachine &TM;
+  static char ID;
+  AMDGPUPeepholeOpt(TargetMachine &tm);
+  ~AMDGPUPeepholeOpt();
+  const char *getPassName() const;
+  bool runOnFunction(Function &F);
+  bool doInitialization(Module &M);
+  bool doFinalization(Module &M);
+  void getAnalysisUsage(AnalysisUsage &AU) const;
+protected:
+private:
+  // Function to initiate all of the instruction level optimizations.
+  bool instLevelOptimizations(BasicBlock::iterator *inst);
+  // Quick check to see if we need to dump all of the pointers into the
+  // arena. If this is correct, then we set all pointers to exist in arena. This
+  // is a workaround for aliasing of pointers in a struct/union.
+  bool dumpAllIntoArena(Function &F);
+  // Because I don't want to invalidate any pointers while in the
+  // safeNestedForEachFunction. I push atomic conversions to a vector and handle
+  // it later. This function does the conversions if required.
+  void doAtomicConversionIfNeeded(Function &F);
+  // Because __amdil_is_constant cannot be properly evaluated if
+  // optimizations are disabled, the call's are placed in a vector
+  // and evaluated after the __amdil_image* functions are evaluated
+  // which should allow the __amdil_is_constant function to be
+  // evaluated correctly.
+  void doIsConstCallConversionIfNeeded();
+  bool mChanged;
+  bool mDebug;
+  bool mConvertAtomics;
+  CodeGenOpt::Level optLevel;
+  // Run a series of tests to see if we can optimize a CALL instruction.
+  bool optimizeCallInst(BasicBlock::iterator *bbb);
+  // A peephole optimization to optimize bit extract sequences.
+  bool optimizeBitExtract(Instruction *inst);
+  // A peephole optimization to optimize bit insert sequences.
+  bool optimizeBitInsert(Instruction *inst);
+  bool setupBitInsert(Instruction *base, 
+                      Instruction *&src, 
+                      Constant *&mask, 
+                      Constant *&shift);
+  // Expand the bit field insert instruction on versions of OpenCL that
+  // don't support it.
+  bool expandBFI(CallInst *CI);
+  // Expand the bit field mask instruction on version of OpenCL that 
+  // don't support it.
+  bool expandBFM(CallInst *CI);
+  // On 7XX and 8XX operations, we do not have 24 bit signed operations. So in
+  // this case we need to expand them. These functions check for 24bit functions
+  // and then expand.
+  bool isSigned24BitOps(CallInst *CI);
+  void expandSigned24BitOps(CallInst *CI);
+  // One optimization that can occur is that if the required workgroup size is
+  // specified then the result of get_local_size is known at compile time and
+  // can be returned accordingly.
+  bool isRWGLocalOpt(CallInst *CI);
+  // On northern island cards, the division is slightly less accurate than on
+  // previous generations, so we need to utilize a more accurate division. So we
+  // can translate the accurate divide to a normal divide on all other cards.
+  bool convertAccurateDivide(CallInst *CI);
+  void expandAccurateDivide(CallInst *CI);
+  // If the alignment is set incorrectly, it can produce really inefficient
+  // code. This checks for this scenario and fixes it if possible.
+  bool correctMisalignedMemOp(Instruction *inst);
+
+  // If we are in no opt mode, then we need to make sure that
+  // local samplers are properly propagated as constant propagation 
+  // doesn't occur and we need to know the value of kernel defined
+  // samplers at compile time.
+  bool propagateSamplerInst(CallInst *CI);
+
+  // Helper functions
+
+  // Group of functions that recursively calculate the size of a structure based
+  // on it's sub-types.
+  size_t getTypeSize(Type * const T, bool dereferencePtr = false);
+  size_t getTypeSize(StructType * const ST, bool dereferencePtr = false);
+  size_t getTypeSize(IntegerType * const IT, bool dereferencePtr = false);
+  size_t getTypeSize(FunctionType * const FT,bool dereferencePtr = false);
+  size_t getTypeSize(ArrayType * const AT, bool dereferencePtr = false);
+  size_t getTypeSize(VectorType * const VT, bool dereferencePtr = false);
+  size_t getTypeSize(PointerType * const PT, bool dereferencePtr = false);
+  size_t getTypeSize(OpaqueType * const OT, bool dereferencePtr = false);
+
+  LLVMContext *mCTX;
+  Function *mF;
+  const AMDGPUSubtarget *mSTM;
+  SmallVector< std::pair<CallInst *, Function *>, 16> atomicFuncs;
+  SmallVector<CallInst *, 16> isConstVec;
+}; // class AMDGPUPeepholeOpt
+  char AMDGPUPeepholeOpt::ID = 0;
+
+// A template function that has two levels of looping before calling the
+// function with a pointer to the current iterator.
+template<class InputIterator, class SecondIterator, class Function>
+Function safeNestedForEach(InputIterator First, InputIterator Last,
+                              SecondIterator S, Function F) {
+  for ( ; First != Last; ++First) {
+    SecondIterator sf, sl;
+    for (sf = First->begin(), sl = First->end();
+         sf != sl; )  {
+      if (!F(&sf)) {
+        ++sf;
+      } 
+    }
+  }
+  return F;
+}
+
+} // anonymous namespace
+
+namespace llvm {
+  FunctionPass *
+  createAMDGPUPeepholeOpt(TargetMachine &tm) {
+    return new AMDGPUPeepholeOpt(tm);
+  }
+} // llvm namespace
+
+AMDGPUPeepholeOpt::AMDGPUPeepholeOpt(TargetMachine &tm)
+  : FunctionPass(ID), TM(tm)  {
+  mDebug = DEBUGME;
+  optLevel = TM.getOptLevel();
+
+}
+
+AMDGPUPeepholeOpt::~AMDGPUPeepholeOpt()  {
+}
+
+const char *
+AMDGPUPeepholeOpt::getPassName() const  {
+  return "AMDGPU PeepHole Optimization Pass";
+}
+
+bool 
+containsPointerType(Type *Ty)  {
+  if (!Ty) {
+    return false;
+  }
+  switch(Ty->getTypeID()) {
+  default:
+    return false;
+  case Type::StructTyID: {
+    const StructType *ST = dyn_cast<StructType>(Ty);
+    for (StructType::element_iterator stb = ST->element_begin(),
+           ste = ST->element_end(); stb != ste; ++stb) {
+      if (!containsPointerType(*stb)) {
+        continue;
+      }
+      return true;
+    }
+    break;
+  }
+  case Type::VectorTyID:
+  case Type::ArrayTyID:
+    return containsPointerType(dyn_cast<SequentialType>(Ty)->getElementType());
+  case Type::PointerTyID:
+    return true;
+  };
+  return false;
+}
+
+bool 
+AMDGPUPeepholeOpt::dumpAllIntoArena(Function &F)  {
+  bool dumpAll = false;
+  for (Function::const_arg_iterator cab = F.arg_begin(),
+       cae = F.arg_end(); cab != cae; ++cab) {
+    const Argument *arg = cab;
+    const PointerType *PT = dyn_cast<PointerType>(arg->getType());
+    if (!PT) {
+      continue;
+    }
+    Type *DereferencedType = PT->getElementType();
+    if (!dyn_cast<StructType>(DereferencedType) 
+        ) {
+      continue;
+    }
+    if (!containsPointerType(DereferencedType)) {
+      continue;
+    }
+    // FIXME: Because a pointer inside of a struct/union may be aliased to
+    // another pointer we need to take the conservative approach and place all
+    // pointers into the arena until more advanced detection is implemented.
+    dumpAll = true;
+  }
+  return dumpAll;
+}
+void
+AMDGPUPeepholeOpt::doIsConstCallConversionIfNeeded() {
+  if (isConstVec.empty()) {
+    return;
+  }
+  for (unsigned x = 0, y = isConstVec.size(); x < y; ++x) {
+    CallInst *CI = isConstVec[x];
+    Constant *CV = dyn_cast<Constant>(CI->getOperand(0));
+    Type *aType = Type::getInt32Ty(*mCTX);
+    Value *Val = (CV != NULL) ? ConstantInt::get(aType, 1)
+      : ConstantInt::get(aType, 0);
+    CI->replaceAllUsesWith(Val);
+    CI->eraseFromParent();
+  }
+  isConstVec.clear();
+}
+void 
+AMDGPUPeepholeOpt::doAtomicConversionIfNeeded(Function &F)  {
+  // Don't do anything if we don't have any atomic operations.
+  if (atomicFuncs.empty()) {
+    return;
+  }
+  // Change the function name for the atomic if it is required
+  uint32_t size = atomicFuncs.size();
+  for (uint32_t x = 0; x < size; ++x) {
+    atomicFuncs[x].first->setOperand(
+        atomicFuncs[x].first->getNumOperands()-1, 
+        atomicFuncs[x].second);
+
+  }
+  mChanged = true;
+  if (mConvertAtomics) {
+    return;
+  }
+}
+
+bool 
+AMDGPUPeepholeOpt::runOnFunction(Function &MF)  {
+  mChanged = false;
+  mF = &MF;
+  mSTM = &TM.getSubtarget<AMDGPUSubtarget>();
+  if (mDebug) {
+    MF.dump();
+  }
+  mCTX = &MF.getType()->getContext();
+  mConvertAtomics = true;
+  safeNestedForEach(MF.begin(), MF.end(), MF.begin()->begin(),
+     std::bind1st(std::mem_fun(&AMDGPUPeepholeOpt::instLevelOptimizations),
+                  this));
+
+  doAtomicConversionIfNeeded(MF);
+  doIsConstCallConversionIfNeeded();
+
+  if (mDebug) {
+    MF.dump();
+  }
+  return mChanged;
+}
+
+bool 
+AMDGPUPeepholeOpt::optimizeCallInst(BasicBlock::iterator *bbb)  {
+  Instruction *inst = (*bbb);
+  CallInst *CI = dyn_cast<CallInst>(inst);
+  if (!CI) {
+    return false;
+  }
+  if (isSigned24BitOps(CI)) {
+    expandSigned24BitOps(CI);
+    ++(*bbb);
+    CI->eraseFromParent();
+    return true;
+  }
+  if (propagateSamplerInst(CI)) {
+    return false;
+  }
+  if (expandBFI(CI) || expandBFM(CI)) {
+    ++(*bbb);
+    CI->eraseFromParent();
+    return true;
+  }
+  if (convertAccurateDivide(CI)) {
+    expandAccurateDivide(CI);
+    ++(*bbb);
+    CI->eraseFromParent();
+    return true;
+  }
+
+  StringRef calleeName = CI->getOperand(CI->getNumOperands()-1)->getName();
+  if (calleeName.startswith("__amdil_is_constant")) {
+    // If we do not have optimizations, then this
+    // cannot be properly evaluated, so we add the
+    // call instruction to a vector and process
+    // them at the end of processing after the
+    // samplers have been correctly handled.
+    if (optLevel == CodeGenOpt::None) {
+      isConstVec.push_back(CI);
+      return false;
+    } else {
+      Constant *CV = dyn_cast<Constant>(CI->getOperand(0));
+      Type *aType = Type::getInt32Ty(*mCTX);
+      Value *Val = (CV != NULL) ? ConstantInt::get(aType, 1)
+        : ConstantInt::get(aType, 0);
+      CI->replaceAllUsesWith(Val);
+      ++(*bbb);
+      CI->eraseFromParent();
+      return true;
+    }
+  }
+
+  if (calleeName.equals("__amdil_is_asic_id_i32")) {
+    ConstantInt *CV = dyn_cast<ConstantInt>(CI->getOperand(0));
+    Type *aType = Type::getInt32Ty(*mCTX);
+    Value *Val = CV;
+    if (Val) {
+      Val = ConstantInt::get(aType, 
+          mSTM->device()->getDeviceFlag() & CV->getZExtValue());
+    } else {
+      Val = ConstantInt::get(aType, 0);
+    }
+    CI->replaceAllUsesWith(Val);
+    ++(*bbb);
+    CI->eraseFromParent();
+    return true;
+  }
+  Function *F = dyn_cast<Function>(CI->getOperand(CI->getNumOperands()-1));
+  if (!F) {
+    return false;
+  } 
+  if (F->getName().startswith("__atom") && !CI->getNumUses() 
+      && F->getName().find("_xchg") == StringRef::npos) {
+    std::string buffer(F->getName().str() + "_noret");
+    F = dyn_cast<Function>(
+          F->getParent()->getOrInsertFunction(buffer, F->getFunctionType()));
+    atomicFuncs.push_back(std::make_pair(CI, F));
+  }
+  
+  if (!mSTM->device()->isSupported(AMDGPUDeviceInfo::ArenaSegment)
+      && !mSTM->device()->isSupported(AMDGPUDeviceInfo::MultiUAV)) {
+    return false;
+  }
+  if (!mConvertAtomics) {
+    return false;
+  }
+  StringRef name = F->getName();
+  if (name.startswith("__atom") && name.find("_g") != StringRef::npos) {
+    mConvertAtomics = false;
+  }
+  return false;
+}
+
+bool
+AMDGPUPeepholeOpt::setupBitInsert(Instruction *base, 
+    Instruction *&src, 
+    Constant *&mask, 
+    Constant *&shift) {
+  if (!base) {
+    if (mDebug) {
+      dbgs() << "Null pointer passed into function.\n";
+    }
+    return false;
+  }
+  bool andOp = false;
+  if (base->getOpcode() == Instruction::Shl) {
+    shift = dyn_cast<Constant>(base->getOperand(1));
+  } else if (base->getOpcode() == Instruction::And) {
+    mask = dyn_cast<Constant>(base->getOperand(1));
+    andOp = true;
+  } else {
+    if (mDebug) {
+      dbgs() << "Failed setup with no Shl or And instruction on base opcode!\n";
+    }
+    // If the base is neither a Shl or a And, we don't fit any of the patterns above.
+    return false;
+  }
+  src = dyn_cast<Instruction>(base->getOperand(0));
+  if (!src) {
+    if (mDebug) {
+      dbgs() << "Failed setup since the base operand is not an instruction!\n";
+    }
+    return false;
+  }
+  // If we find an 'and' operation, then we don't need to
+  // find the next operation as we already know the
+  // bits that are valid at this point.
+  if (andOp) {
+    return true;
+  }
+  if (src->getOpcode() == Instruction::Shl && !shift) {
+    shift = dyn_cast<Constant>(src->getOperand(1));
+    src = dyn_cast<Instruction>(src->getOperand(0));
+  } else if (src->getOpcode() == Instruction::And && !mask) {
+    mask = dyn_cast<Constant>(src->getOperand(1));
+  }
+  if (!mask && !shift) {
+    if (mDebug) {
+      dbgs() << "Failed setup since both mask and shift are NULL!\n";
+    }
+    // Did not find a constant mask or a shift.
+    return false;
+  }
+  return true;
+}
+bool
+AMDGPUPeepholeOpt::optimizeBitInsert(Instruction *inst)  {
+  if (!inst) {
+    return false;
+  }
+  if (!inst->isBinaryOp()) {
+    return false;
+  }
+  if (inst->getOpcode() != Instruction::Or) {
+    return false;
+  }
+  if (optLevel == CodeGenOpt::None) {
+    return false;
+  }
+  // We want to do an optimization on a sequence of ops that in the end equals a
+  // single ISA instruction.
+  // The base pattern for this optimization is - ((A & B) << C) | ((D & E) << F)
+  // Some simplified versions of this pattern are as follows:
+  // (A & B) | (D & E) when B & E == 0 && C == 0 && F == 0
+  // ((A & B) << C) | (D & E) when B ^ E == 0 && (1 << C) >= E
+  // (A & B) | ((D & E) << F) when B ^ E == 0 && (1 << F) >= B
+  // (A & B) | (D << F) when (1 << F) >= B
+  // (A << C) | (D & E) when (1 << C) >= E
+  if (mSTM->device()->getGeneration() == AMDGPUDeviceInfo::HD4XXX) {
+    // The HD4XXX hardware doesn't support the ubit_insert instruction.
+    return false;
+  }
+  Type *aType = inst->getType();
+  bool isVector = aType->isVectorTy();
+  int numEle = 1;
+  // This optimization only works on 32bit integers.
+  if (aType->getScalarType()
+      != Type::getInt32Ty(inst->getContext())) {
+    return false;
+  }
+  if (isVector) {
+    const VectorType *VT = dyn_cast<VectorType>(aType);
+    numEle = VT->getNumElements();
+    // We currently cannot support more than 4 elements in a intrinsic and we
+    // cannot support Vec3 types.
+    if (numEle > 4 || numEle == 3) {
+      return false;
+    }
+  }
+  // TODO: Handle vectors.
+  if (isVector) {
+    if (mDebug) {
+      dbgs() << "!!! Vectors are not supported yet!\n";
+    }
+    return false;
+  }
+  Instruction *LHSSrc = NULL, *RHSSrc = NULL;
+  Constant *LHSMask = NULL, *RHSMask = NULL;
+  Constant *LHSShift = NULL, *RHSShift = NULL;
+  Instruction *LHS = dyn_cast<Instruction>(inst->getOperand(0));
+  Instruction *RHS = dyn_cast<Instruction>(inst->getOperand(1));
+  if (!setupBitInsert(LHS, LHSSrc, LHSMask, LHSShift)) {
+    if (mDebug) {
+      dbgs() << "Found an OR Operation that failed setup!\n";
+      inst->dump();
+      if (LHS) { LHS->dump(); }
+      if (LHSSrc) { LHSSrc->dump(); }
+      if (LHSMask) { LHSMask->dump(); }
+      if (LHSShift) { LHSShift->dump(); }
+    }
+    // There was an issue with the setup for BitInsert.
+    return false;
+  }
+  if (!setupBitInsert(RHS, RHSSrc, RHSMask, RHSShift)) {
+    if (mDebug) {
+      dbgs() << "Found an OR Operation that failed setup!\n";
+      inst->dump();
+      if (RHS) { RHS->dump(); }
+      if (RHSSrc) { RHSSrc->dump(); }
+      if (RHSMask) { RHSMask->dump(); }
+      if (RHSShift) { RHSShift->dump(); }
+    }
+    // There was an issue with the setup for BitInsert.
+    return false;
+  }
+  if (mDebug) {
+    dbgs() << "Found an OR operation that can possible be optimized to ubit insert!\n";
+    dbgs() << "Op:        "; inst->dump();
+    dbgs() << "LHS:       "; if (LHS) { LHS->dump(); } else { dbgs() << "(None)\n"; }
+    dbgs() << "LHS Src:   "; if (LHSSrc) { LHSSrc->dump(); } else { dbgs() << "(None)\n"; }
+    dbgs() << "LHS Mask:  "; if (LHSMask) { LHSMask->dump(); } else { dbgs() << "(None)\n"; }
+    dbgs() << "LHS Shift: "; if (LHSShift) { LHSShift->dump(); } else { dbgs() << "(None)\n"; }
+    dbgs() << "RHS:       "; if (RHS) { RHS->dump(); } else { dbgs() << "(None)\n"; }
+    dbgs() << "RHS Src:   "; if (RHSSrc) { RHSSrc->dump(); } else { dbgs() << "(None)\n"; }
+    dbgs() << "RHS Mask:  "; if (RHSMask) { RHSMask->dump(); } else { dbgs() << "(None)\n"; }
+    dbgs() << "RHS Shift: "; if (RHSShift) { RHSShift->dump(); } else { dbgs() << "(None)\n"; }
+  }
+  Constant *offset = NULL;
+  Constant *width = NULL;
+  uint32_t lhsMaskVal = 0, rhsMaskVal = 0;
+  uint32_t lhsShiftVal = 0, rhsShiftVal = 0;
+  uint32_t lhsMaskWidth = 0, rhsMaskWidth = 0;
+  uint32_t lhsMaskOffset = 0, rhsMaskOffset = 0;
+  lhsMaskVal = (LHSMask 
+      ? dyn_cast<ConstantInt>(LHSMask)->getZExtValue() : 0);
+  rhsMaskVal = (RHSMask 
+      ? dyn_cast<ConstantInt>(RHSMask)->getZExtValue() : 0);
+  lhsShiftVal = (LHSShift 
+      ? dyn_cast<ConstantInt>(LHSShift)->getZExtValue() : 0);
+  rhsShiftVal = (RHSShift 
+      ? dyn_cast<ConstantInt>(RHSShift)->getZExtValue() : 0);
+  lhsMaskWidth = lhsMaskVal ? CountPopulation_32(lhsMaskVal) : 32 - lhsShiftVal;
+  rhsMaskWidth = rhsMaskVal ? CountPopulation_32(rhsMaskVal) : 32 - rhsShiftVal;
+  lhsMaskOffset = lhsMaskVal ? CountTrailingZeros_32(lhsMaskVal) : lhsShiftVal;
+  rhsMaskOffset = rhsMaskVal ? CountTrailingZeros_32(rhsMaskVal) : rhsShiftVal;
+  // TODO: Handle the case of A & B | D & ~B(i.e. inverted masks).
+  if ((lhsMaskVal || rhsMaskVal) && !(lhsMaskVal ^ rhsMaskVal)) {
+    return false;
+  }
+  if (lhsMaskOffset >= (rhsMaskWidth + rhsMaskOffset)) {
+    offset = ConstantInt::get(aType, lhsMaskOffset, false);
+    width = ConstantInt::get(aType, lhsMaskWidth, false);
+    RHSSrc = RHS;
+    if (!isMask_32(lhsMaskVal) && !isShiftedMask_32(lhsMaskVal)) {
+      return false;
+    }
+    if (!LHSShift) {
+      LHSSrc = BinaryOperator::Create(Instruction::LShr, LHSSrc, offset,
+          "MaskShr", LHS);
+    } else if (lhsShiftVal != lhsMaskOffset) {
+      LHSSrc = BinaryOperator::Create(Instruction::LShr, LHSSrc, offset,
+          "MaskShr", LHS);
+    }
+    if (mDebug) {
+      dbgs() << "Optimizing LHS!\n";
+    }
+  } else if (rhsMaskOffset >= (lhsMaskWidth + lhsMaskOffset)) {
+    offset = ConstantInt::get(aType, rhsMaskOffset, false);
+    width = ConstantInt::get(aType, rhsMaskWidth, false);
+    LHSSrc = RHSSrc;
+    RHSSrc = LHS;
+    if (!isMask_32(rhsMaskVal) && !isShiftedMask_32(rhsMaskVal)) {
+      return false;
+    }
+    if (!RHSShift) {
+      LHSSrc = BinaryOperator::Create(Instruction::LShr, LHSSrc, offset,
+          "MaskShr", RHS);
+    } else if (rhsShiftVal != rhsMaskOffset) {
+      LHSSrc = BinaryOperator::Create(Instruction::LShr, LHSSrc, offset,
+          "MaskShr", RHS);
+    }
+    if (mDebug) {
+      dbgs() << "Optimizing RHS!\n";
+    }
+  } else {
+    if (mDebug) {
+      dbgs() << "Failed constraint 3!\n";
+    }
+    return false;
+  }
+  if (mDebug) {
+    dbgs() << "Width:  "; if (width) { width->dump(); } else { dbgs() << "(0)\n"; }
+    dbgs() << "Offset: "; if (offset) { offset->dump(); } else { dbgs() << "(0)\n"; }
+    dbgs() << "LHSSrc: "; if (LHSSrc) { LHSSrc->dump(); } else { dbgs() << "(0)\n"; }
+    dbgs() << "RHSSrc: "; if (RHSSrc) { RHSSrc->dump(); } else { dbgs() << "(0)\n"; }
+  }
+  if (!offset || !width) {
+    if (mDebug) {
+      dbgs() << "Either width or offset are NULL, failed detection!\n";
+    }
+    return false;
+  }
+  // Lets create the function signature.
+  std::vector<Type *> callTypes;
+  callTypes.push_back(aType);
+  callTypes.push_back(aType);
+  callTypes.push_back(aType);
+  callTypes.push_back(aType);
+  FunctionType *funcType = FunctionType::get(aType, callTypes, false);
+  std::string name = "__amdil_ubit_insert";
+  if (isVector) { name += "_v" + itostr(numEle) + "u32"; } else { name += "_u32"; }
+  Function *Func = 
+    dyn_cast<Function>(inst->getParent()->getParent()->getParent()->
+        getOrInsertFunction(StringRef(name), funcType));
+  Value *Operands[4] = {
+    width,
+    offset,
+    LHSSrc,
+    RHSSrc
+  };
+  CallInst *CI = CallInst::Create(Func, Operands, "BitInsertOpt");
+  if (mDebug) {
+    dbgs() << "Old Inst: ";
+    inst->dump();
+    dbgs() << "New Inst: ";
+    CI->dump();
+    dbgs() << "\n\n";
+  }
+  CI->insertBefore(inst);
+  inst->replaceAllUsesWith(CI);
+  return true;
+}
+
+bool 
+AMDGPUPeepholeOpt::optimizeBitExtract(Instruction *inst)  {
+  if (!inst) {
+    return false;
+  }
+  if (!inst->isBinaryOp()) {
+    return false;
+  }
+  if (inst->getOpcode() != Instruction::And) {
+    return false;
+  }
+  if (optLevel == CodeGenOpt::None) {
+    return false;
+  }
+  // We want to do some simple optimizations on Shift right/And patterns. The
+  // basic optimization is to turn (A >> B) & C where A is a 32bit type, B is a
+  // value smaller than 32 and C is a mask. If C is a constant value, then the
+  // following transformation can occur. For signed integers, it turns into the
+  // function call dst = __amdil_ibit_extract(log2(C), B, A) For unsigned
+  // integers, it turns into the function call dst =
+  // __amdil_ubit_extract(log2(C), B, A) The function __amdil_[u|i]bit_extract
+  // can be found in Section 7.9 of the ATI IL spec of the stream SDK for
+  // Evergreen hardware.
+  if (mSTM->device()->getGeneration() == AMDGPUDeviceInfo::HD4XXX) {
+    // This does not work on HD4XXX hardware.
+    return false;
+  }
+  Type *aType = inst->getType();
+  bool isVector = aType->isVectorTy();
+
+  // XXX Support vector types
+  if (isVector) {
+    return false;
+  }
+  int numEle = 1;
+  // This only works on 32bit integers
+  if (aType->getScalarType()
+      != Type::getInt32Ty(inst->getContext())) {
+    return false;
+  }
+  if (isVector) {
+    const VectorType *VT = dyn_cast<VectorType>(aType);
+    numEle = VT->getNumElements();
+    // We currently cannot support more than 4 elements in a intrinsic and we
+    // cannot support Vec3 types.
+    if (numEle > 4 || numEle == 3) {
+      return false;
+    }
+  }
+  BinaryOperator *ShiftInst = dyn_cast<BinaryOperator>(inst->getOperand(0));
+  // If the first operand is not a shift instruction, then we can return as it
+  // doesn't match this pattern.
+  if (!ShiftInst || !ShiftInst->isShift()) {
+    return false;
+  }
+  // If we are a shift left, then we need don't match this pattern.
+  if (ShiftInst->getOpcode() == Instruction::Shl) {
+    return false;
+  }
+  bool isSigned = ShiftInst->isArithmeticShift();
+  Constant *AndMask = dyn_cast<Constant>(inst->getOperand(1));
+  Constant *ShrVal = dyn_cast<Constant>(ShiftInst->getOperand(1));
+  // Lets make sure that the shift value and the and mask are constant integers.
+  if (!AndMask || !ShrVal) {
+    return false;
+  }
+  Constant *newMaskConst;
+  Constant *shiftValConst;
+  if (isVector) {
+    // Handle the vector case
+    std::vector<Constant *> maskVals;
+    std::vector<Constant *> shiftVals;
+    ConstantVector *AndMaskVec = dyn_cast<ConstantVector>(AndMask);
+    ConstantVector *ShrValVec = dyn_cast<ConstantVector>(ShrVal);
+    Type *scalarType = AndMaskVec->getType()->getScalarType();
+    assert(AndMaskVec->getNumOperands() ==
+           ShrValVec->getNumOperands() && "cannot have a "
+           "combination where the number of elements to a "
+           "shift and an and are different!");
+    for (size_t x = 0, y = AndMaskVec->getNumOperands(); x < y; ++x) {
+      ConstantInt *AndCI = dyn_cast<ConstantInt>(AndMaskVec->getOperand(x));
+      ConstantInt *ShiftIC = dyn_cast<ConstantInt>(ShrValVec->getOperand(x));
+      if (!AndCI || !ShiftIC) {
+        return false;
+      }
+      uint32_t maskVal = (uint32_t)AndCI->getZExtValue();
+      if (!isMask_32(maskVal)) {
+        return false;
+      }
+      maskVal = (uint32_t)CountTrailingOnes_32(maskVal);
+      uint32_t shiftVal = (uint32_t)ShiftIC->getZExtValue();
+      // If the mask or shiftval is greater than the bitcount, then break out.
+      if (maskVal >= 32 || shiftVal >= 32) {
+        return false;
+      }
+      // If the mask val is greater than the the number of original bits left
+      // then this optimization is invalid.
+      if (maskVal > (32 - shiftVal)) {
+        return false;
+      }
+      maskVals.push_back(ConstantInt::get(scalarType, maskVal, isSigned));
+      shiftVals.push_back(ConstantInt::get(scalarType, shiftVal, isSigned));
+    }
+    newMaskConst = ConstantVector::get(maskVals);
+    shiftValConst = ConstantVector::get(shiftVals);
+  } else {
+    // Handle the scalar case
+    uint32_t maskVal = (uint32_t)dyn_cast<ConstantInt>(AndMask)->getZExtValue();
+    // This must be a mask value where all lower bits are set to 1 and then any
+    // bit higher is set to 0.
+    if (!isMask_32(maskVal)) {
+      return false;
+    }
+    maskVal = (uint32_t)CountTrailingOnes_32(maskVal);
+    // Count the number of bits set in the mask, this is the width of the
+    // resulting bit set that is extracted from the source value.
+    uint32_t shiftVal = (uint32_t)dyn_cast<ConstantInt>(ShrVal)->getZExtValue();
+    // If the mask or shift val is greater than the bitcount, then break out.
+    if (maskVal >= 32 || shiftVal >= 32) {
+      return false;
+    }
+    // If the mask val is greater than the the number of original bits left then
+    // this optimization is invalid.
+    if (maskVal > (32 - shiftVal)) {
+      return false;
+    }
+    newMaskConst = ConstantInt::get(aType, maskVal, isSigned);
+    shiftValConst = ConstantInt::get(aType, shiftVal, isSigned);
+  }
+  // Lets create the function signature.
+  std::vector<Type *> callTypes;
+  callTypes.push_back(aType);
+  callTypes.push_back(aType);
+  callTypes.push_back(aType);
+  FunctionType *funcType = FunctionType::get(aType, callTypes, false);
+  std::string name = "llvm.AMDGPU.bit.extract.u32";
+  if (isVector) {
+    name += ".v" + itostr(numEle) + "i32";
+  } else {
+    name += ".";
+  }
+  // Lets create the function.
+  Function *Func = 
+    dyn_cast<Function>(inst->getParent()->getParent()->getParent()->
+                       getOrInsertFunction(StringRef(name), funcType));
+  Value *Operands[3] = {
+    ShiftInst->getOperand(0),
+    shiftValConst,
+    newMaskConst
+  };
+  // Lets create the Call with the operands
+  CallInst *CI = CallInst::Create(Func, Operands, "ByteExtractOpt");
+  CI->setDoesNotAccessMemory();
+  CI->insertBefore(inst);
+  inst->replaceAllUsesWith(CI);
+  return true;
+}
+
+bool
+AMDGPUPeepholeOpt::expandBFI(CallInst *CI) {
+  if (!CI) {
+    return false;
+  }
+  Value *LHS = CI->getOperand(CI->getNumOperands() - 1);
+  if (!LHS->getName().startswith("__amdil_bfi")) {
+    return false;
+  }
+  Type* type = CI->getOperand(0)->getType();
+  Constant *negOneConst = NULL;
+  if (type->isVectorTy()) {
+    std::vector<Constant *> negOneVals;
+    negOneConst = ConstantInt::get(CI->getContext(), 
+        APInt(32, StringRef("-1"), 10));
+    for (size_t x = 0,
+        y = dyn_cast<VectorType>(type)->getNumElements(); x < y; ++x) {
+      negOneVals.push_back(negOneConst);
+    }
+    negOneConst = ConstantVector::get(negOneVals);
+  } else {
+    negOneConst = ConstantInt::get(CI->getContext(), 
+        APInt(32, StringRef("-1"), 10));
+  }
+  // __amdil_bfi => (A & B) | (~A & C)
+  BinaryOperator *lhs = 
+    BinaryOperator::Create(Instruction::And, CI->getOperand(0),
+        CI->getOperand(1), "bfi_and", CI);
+  BinaryOperator *rhs =
+    BinaryOperator::Create(Instruction::Xor, CI->getOperand(0), negOneConst,
+        "bfi_not", CI);
+  rhs = BinaryOperator::Create(Instruction::And, rhs, CI->getOperand(2),
+      "bfi_and", CI);
+  lhs = BinaryOperator::Create(Instruction::Or, lhs, rhs, "bfi_or", CI);
+  CI->replaceAllUsesWith(lhs);
+  return true;
+}
+
+bool
+AMDGPUPeepholeOpt::expandBFM(CallInst *CI) {
+  if (!CI) {
+    return false;
+  }
+  Value *LHS = CI->getOperand(CI->getNumOperands() - 1);
+  if (!LHS->getName().startswith("__amdil_bfm")) {
+    return false;
+  }
+  // __amdil_bfm => ((1 << (src0 & 0x1F)) - 1) << (src1 & 0x1f)
+  Constant *newMaskConst = NULL;
+  Constant *newShiftConst = NULL;
+  Type* type = CI->getOperand(0)->getType();
+  if (type->isVectorTy()) {
+    std::vector<Constant*> newMaskVals, newShiftVals;
+    newMaskConst = ConstantInt::get(Type::getInt32Ty(*mCTX), 0x1F);
+    newShiftConst = ConstantInt::get(Type::getInt32Ty(*mCTX), 1);
+    for (size_t x = 0,
+        y = dyn_cast<VectorType>(type)->getNumElements(); x < y; ++x) {
+      newMaskVals.push_back(newMaskConst);
+      newShiftVals.push_back(newShiftConst);
+    }
+    newMaskConst = ConstantVector::get(newMaskVals);
+    newShiftConst = ConstantVector::get(newShiftVals);
+  } else {
+    newMaskConst = ConstantInt::get(Type::getInt32Ty(*mCTX), 0x1F);
+    newShiftConst = ConstantInt::get(Type::getInt32Ty(*mCTX), 1);
+  }
+  BinaryOperator *lhs =
+    BinaryOperator::Create(Instruction::And, CI->getOperand(0),
+        newMaskConst, "bfm_mask", CI);
+  lhs = BinaryOperator::Create(Instruction::Shl, newShiftConst,
+      lhs, "bfm_shl", CI);
+  lhs = BinaryOperator::Create(Instruction::Sub, lhs,
+      newShiftConst, "bfm_sub", CI);
+  BinaryOperator *rhs =
+    BinaryOperator::Create(Instruction::And, CI->getOperand(1),
+        newMaskConst, "bfm_mask", CI);
+  lhs = BinaryOperator::Create(Instruction::Shl, lhs, rhs, "bfm_shl", CI);
+  CI->replaceAllUsesWith(lhs);
+  return true;
+}
+
+bool
+AMDGPUPeepholeOpt::instLevelOptimizations(BasicBlock::iterator *bbb)  {
+  Instruction *inst = (*bbb);
+  if (optimizeCallInst(bbb)) {
+    return true;
+  }
+  if (optimizeBitExtract(inst)) {
+    return false;
+  }
+  if (optimizeBitInsert(inst)) {
+    return false;
+  }
+  if (correctMisalignedMemOp(inst)) {
+    return false;
+  }
+  return false;
+}
+bool
+AMDGPUPeepholeOpt::correctMisalignedMemOp(Instruction *inst) {
+  LoadInst *linst = dyn_cast<LoadInst>(inst);
+  StoreInst *sinst = dyn_cast<StoreInst>(inst);
+  unsigned alignment;
+  Type* Ty = inst->getType();
+  if (linst) {
+    alignment = linst->getAlignment();
+    Ty = inst->getType();
+  } else if (sinst) {
+    alignment = sinst->getAlignment();
+    Ty = sinst->getValueOperand()->getType();
+  } else {
+    return false;
+  }
+  unsigned size = getTypeSize(Ty);
+  if (size == alignment || size < alignment) {
+    return false;
+  }
+  if (!Ty->isStructTy()) {
+    return false;
+  }
+  if (alignment < 4) {
+    if (linst) {
+      linst->setAlignment(0);
+      return true;
+    } else if (sinst) {
+      sinst->setAlignment(0);
+      return true;
+    }
+  }
+  return false;
+}
+bool 
+AMDGPUPeepholeOpt::isSigned24BitOps(CallInst *CI)  {
+  if (!CI) {
+    return false;
+  }
+  Value *LHS = CI->getOperand(CI->getNumOperands() - 1);
+  std::string namePrefix = LHS->getName().substr(0, 14);
+  if (namePrefix != "__amdil_imad24" && namePrefix != "__amdil_imul24"
+      && namePrefix != "__amdil__imul24_high") {
+    return false;
+  }
+  if (mSTM->device()->usesHardware(AMDGPUDeviceInfo::Signed24BitOps)) {
+    return false;
+  }
+  return true;
+}
+
+void 
+AMDGPUPeepholeOpt::expandSigned24BitOps(CallInst *CI)  {
+  assert(isSigned24BitOps(CI) && "Must be a "
+      "signed 24 bit operation to call this function!");
+  Value *LHS = CI->getOperand(CI->getNumOperands()-1);
+  // On 7XX and 8XX we do not have signed 24bit, so we need to
+  // expand it to the following:
+  // imul24 turns into 32bit imul
+  // imad24 turns into 32bit imad
+  // imul24_high turns into 32bit imulhigh
+  if (LHS->getName().substr(0, 14) == "__amdil_imad24") {
+    Type *aType = CI->getOperand(0)->getType();
+    bool isVector = aType->isVectorTy();
+    int numEle = isVector ? dyn_cast<VectorType>(aType)->getNumElements() : 1;
+    std::vector<Type*> callTypes;
+    callTypes.push_back(CI->getOperand(0)->getType());
+    callTypes.push_back(CI->getOperand(1)->getType());
+    callTypes.push_back(CI->getOperand(2)->getType());
+    FunctionType *funcType =
+      FunctionType::get(CI->getOperand(0)->getType(), callTypes, false);
+    std::string name = "__amdil_imad";
+    if (isVector) {
+      name += "_v" + itostr(numEle) + "i32";
+    } else {
+      name += "_i32";
+    }
+    Function *Func = dyn_cast<Function>(
+                       CI->getParent()->getParent()->getParent()->
+                       getOrInsertFunction(StringRef(name), funcType));
+    Value *Operands[3] = {
+      CI->getOperand(0),
+      CI->getOperand(1),
+      CI->getOperand(2)
+    };
+    CallInst *nCI = CallInst::Create(Func, Operands, "imad24");
+    nCI->insertBefore(CI);
+    CI->replaceAllUsesWith(nCI);
+  } else if (LHS->getName().substr(0, 14) == "__amdil_imul24") {
+    BinaryOperator *mulOp =
+      BinaryOperator::Create(Instruction::Mul, CI->getOperand(0),
+          CI->getOperand(1), "imul24", CI);
+    CI->replaceAllUsesWith(mulOp);
+  } else if (LHS->getName().substr(0, 19) == "__amdil_imul24_high") {
+    Type *aType = CI->getOperand(0)->getType();
+
+    bool isVector = aType->isVectorTy();
+    int numEle = isVector ? dyn_cast<VectorType>(aType)->getNumElements() : 1;
+    std::vector<Type*> callTypes;
+    callTypes.push_back(CI->getOperand(0)->getType());
+    callTypes.push_back(CI->getOperand(1)->getType());
+    FunctionType *funcType =
+      FunctionType::get(CI->getOperand(0)->getType(), callTypes, false);
+    std::string name = "__amdil_imul_high";
+    if (isVector) {
+      name += "_v" + itostr(numEle) + "i32";
+    } else {
+      name += "_i32";
+    }
+    Function *Func = dyn_cast<Function>(
+                       CI->getParent()->getParent()->getParent()->
+                       getOrInsertFunction(StringRef(name), funcType));
+    Value *Operands[2] = {
+      CI->getOperand(0),
+      CI->getOperand(1)
+    };
+    CallInst *nCI = CallInst::Create(Func, Operands, "imul24_high");
+    nCI->insertBefore(CI);
+    CI->replaceAllUsesWith(nCI);
+  }
+}
+
+bool 
+AMDGPUPeepholeOpt::isRWGLocalOpt(CallInst *CI)  {
+  return (CI != NULL
+          && CI->getOperand(CI->getNumOperands() - 1)->getName() 
+          == "__amdil_get_local_size_int");
+}
+
+bool 
+AMDGPUPeepholeOpt::convertAccurateDivide(CallInst *CI)  {
+  if (!CI) {
+    return false;
+  }
+  if (mSTM->device()->getGeneration() == AMDGPUDeviceInfo::HD6XXX
+      && (mSTM->getDeviceName() == "cayman")) {
+    return false;
+  }
+  return CI->getOperand(CI->getNumOperands() - 1)->getName().substr(0, 20) 
+      == "__amdil_improved_div";
+}
+
+void 
+AMDGPUPeepholeOpt::expandAccurateDivide(CallInst *CI)  {
+  assert(convertAccurateDivide(CI)
+         && "expanding accurate divide can only happen if it is expandable!");
+  BinaryOperator *divOp =
+    BinaryOperator::Create(Instruction::FDiv, CI->getOperand(0),
+                           CI->getOperand(1), "fdiv32", CI);
+  CI->replaceAllUsesWith(divOp);
+}
+
+bool
+AMDGPUPeepholeOpt::propagateSamplerInst(CallInst *CI) {
+  if (optLevel != CodeGenOpt::None) {
+    return false;
+  }
+
+  if (!CI) {
+    return false;
+  }
+
+  unsigned funcNameIdx = 0;
+  funcNameIdx = CI->getNumOperands() - 1;
+  StringRef calleeName = CI->getOperand(funcNameIdx)->getName();
+  if (calleeName != "__amdil_image2d_read_norm"
+   && calleeName != "__amdil_image2d_read_unnorm"
+   && calleeName != "__amdil_image3d_read_norm"
+   && calleeName != "__amdil_image3d_read_unnorm") {
+    return false;
+  }
+
+  unsigned samplerIdx = 2;
+  samplerIdx = 1;
+  Value *sampler = CI->getOperand(samplerIdx);
+  LoadInst *lInst = dyn_cast<LoadInst>(sampler);
+  if (!lInst) {
+    return false;
+  }
+
+  if (lInst->getPointerAddressSpace() != AMDGPUAS::PRIVATE_ADDRESS) {
+    return false;
+  }
+
+  GlobalVariable *gv = dyn_cast<GlobalVariable>(lInst->getPointerOperand());
+  // If we are loading from what is not a global value, then we
+  // fail and return.
+  if (!gv) {
+    return false;
+  }
+
+  // If we don't have an initializer or we have an initializer and
+  // the initializer is not a 32bit integer, we fail.
+  if (!gv->hasInitializer() 
+      || !gv->getInitializer()->getType()->isIntegerTy(32)) {
+      return false;
+  }
+
+  // Now that we have the global variable initializer, lets replace
+  // all uses of the load instruction with the samplerVal and
+  // reparse the __amdil_is_constant() function.
+  Constant *samplerVal = gv->getInitializer();
+  lInst->replaceAllUsesWith(samplerVal);
+  return true;
+}
+
+bool 
+AMDGPUPeepholeOpt::doInitialization(Module &M)  {
+  return false;
+}
+
+bool 
+AMDGPUPeepholeOpt::doFinalization(Module &M)  {
+  return false;
+}
+
+void 
+AMDGPUPeepholeOpt::getAnalysisUsage(AnalysisUsage &AU) const  {
+  AU.addRequired<MachineFunctionAnalysis>();
+  FunctionPass::getAnalysisUsage(AU);
+  AU.setPreservesAll();
+}
+
+size_t AMDGPUPeepholeOpt::getTypeSize(Type * const T, bool dereferencePtr) {
+  size_t size = 0;
+  if (!T) {
+    return size;
+  }
+  switch (T->getTypeID()) {
+  case Type::X86_FP80TyID:
+  case Type::FP128TyID:
+  case Type::PPC_FP128TyID:
+  case Type::LabelTyID:
+    assert(0 && "These types are not supported by this backend");
+  default:
+  case Type::FloatTyID:
+  case Type::DoubleTyID:
+    size = T->getPrimitiveSizeInBits() >> 3;
+    break;
+  case Type::PointerTyID:
+    size = getTypeSize(dyn_cast<PointerType>(T), dereferencePtr);
+    break;
+  case Type::IntegerTyID:
+    size = getTypeSize(dyn_cast<IntegerType>(T), dereferencePtr);
+    break;
+  case Type::StructTyID:
+    size = getTypeSize(dyn_cast<StructType>(T), dereferencePtr);
+    break;
+  case Type::ArrayTyID:
+    size = getTypeSize(dyn_cast<ArrayType>(T), dereferencePtr);
+    break;
+  case Type::FunctionTyID:
+    size = getTypeSize(dyn_cast<FunctionType>(T), dereferencePtr);
+    break;
+  case Type::VectorTyID:
+    size = getTypeSize(dyn_cast<VectorType>(T), dereferencePtr);
+    break;
+  };
+  return size;
+}
+
+size_t AMDGPUPeepholeOpt::getTypeSize(StructType * const ST,
+    bool dereferencePtr) {
+  size_t size = 0;
+  if (!ST) {
+    return size;
+  }
+  Type *curType;
+  StructType::element_iterator eib;
+  StructType::element_iterator eie;
+  for (eib = ST->element_begin(), eie = ST->element_end(); eib != eie; ++eib) {
+    curType = *eib;
+    size += getTypeSize(curType, dereferencePtr);
+  }
+  return size;
+}
+
+size_t AMDGPUPeepholeOpt::getTypeSize(IntegerType * const IT,
+    bool dereferencePtr) {
+  return IT ? (IT->getBitWidth() >> 3) : 0;
+}
+
+size_t AMDGPUPeepholeOpt::getTypeSize(FunctionType * const FT,
+    bool dereferencePtr) {
+    assert(0 && "Should not be able to calculate the size of an function type");
+    return 0;
+}
+
+size_t AMDGPUPeepholeOpt::getTypeSize(ArrayType * const AT,
+    bool dereferencePtr) {
+  return (size_t)(AT ? (getTypeSize(AT->getElementType(),
+                                    dereferencePtr) * AT->getNumElements())
+                     : 0);
+}
+
+size_t AMDGPUPeepholeOpt::getTypeSize(VectorType * const VT,
+    bool dereferencePtr) {
+  return VT ? (VT->getBitWidth() >> 3) : 0;
+}
+
+size_t AMDGPUPeepholeOpt::getTypeSize(PointerType * const PT,
+    bool dereferencePtr) {
+  if (!PT) {
+    return 0;
+  }
+  Type *CT = PT->getElementType();
+  if (CT->getTypeID() == Type::StructTyID &&
+      PT->getAddressSpace() == AMDGPUAS::PRIVATE_ADDRESS) {
+    return getTypeSize(dyn_cast<StructType>(CT));
+  } else if (dereferencePtr) {
+    size_t size = 0;
+    for (size_t x = 0, y = PT->getNumContainedTypes(); x < y; ++x) {
+      size += getTypeSize(PT->getContainedType(x), dereferencePtr);
+    }
+    return size;
+  } else {
+    return 4;
+  }
+}
+
+size_t AMDGPUPeepholeOpt::getTypeSize(OpaqueType * const OT,
+    bool dereferencePtr) {
+  //assert(0 && "Should not be able to calculate the size of an opaque type");
+  return 4;
+}
diff --git a/lib/Target/R600/AMDILRegisterInfo.td b/lib/Target/R600/AMDILRegisterInfo.td
new file mode 100644
index 000000000000..b9d033432e8c
--- /dev/null
+++ b/lib/Target/R600/AMDILRegisterInfo.td
@@ -0,0 +1,107 @@
+//===- AMDILRegisterInfo.td - AMDIL Register defs ----------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+//  Declarations that describe the AMDIL register file
+//
+//===----------------------------------------------------------------------===//
+
+class AMDILReg<bits<16> num, string n> : Register<n> {
+  field bits<16> Value;
+  let Value = num;
+  let Namespace = "AMDGPU";
+}
+
+// We will start with 8 registers for each class before expanding to more
+// Since the swizzle is added based on the register class, we can leave it
+// off here and just specify different registers for different register classes
+def R1 : AMDILReg<1, "r1">, DwarfRegNum<[1]>;
+def R2 : AMDILReg<2, "r2">, DwarfRegNum<[2]>;
+def R3 : AMDILReg<3, "r3">, DwarfRegNum<[3]>;
+def R4 : AMDILReg<4, "r4">, DwarfRegNum<[4]>;
+def R5 : AMDILReg<5, "r5">, DwarfRegNum<[5]>;
+def R6 : AMDILReg<6, "r6">, DwarfRegNum<[6]>;
+def R7 : AMDILReg<7, "r7">, DwarfRegNum<[7]>;
+def R8 : AMDILReg<8, "r8">, DwarfRegNum<[8]>;
+def R9 : AMDILReg<9, "r9">, DwarfRegNum<[9]>;
+def R10 : AMDILReg<10, "r10">, DwarfRegNum<[10]>;
+def R11 : AMDILReg<11, "r11">, DwarfRegNum<[11]>;
+def R12 : AMDILReg<12, "r12">, DwarfRegNum<[12]>;
+def R13 : AMDILReg<13, "r13">, DwarfRegNum<[13]>;
+def R14 : AMDILReg<14, "r14">, DwarfRegNum<[14]>;
+def R15 : AMDILReg<15, "r15">, DwarfRegNum<[15]>;
+def R16 : AMDILReg<16, "r16">, DwarfRegNum<[16]>;
+def R17 : AMDILReg<17, "r17">, DwarfRegNum<[17]>;
+def R18 : AMDILReg<18, "r18">, DwarfRegNum<[18]>;
+def R19 : AMDILReg<19, "r19">, DwarfRegNum<[19]>;
+def R20 : AMDILReg<20, "r20">, DwarfRegNum<[20]>;
+
+// All registers between 1000 and 1024 are reserved and cannot be used
+// unless commented in this section
+// r1021-r1025 are used to dynamically calculate the local/group/thread/region/region_local ID's
+// r1020 is used to hold the frame index for local arrays
+// r1019 is used to hold the dynamic stack allocation pointer
+// r1018 is used as a temporary register for handwritten code
+// r1017 is used as a temporary register for handwritten code
+// r1016 is used as a temporary register for load/store code
+// r1015 is used as a temporary register for data segment offset
+// r1014 is used as a temporary register for store code
+// r1013 is used as the section data pointer register
+// r1012-r1010 and r1001-r1008 are used for temporary I/O registers
+// r1009 is used as the frame pointer register
+// r999 is used as the mem register.
+// r998 is used as the return address register.
+//def R1025 : AMDILReg<1025, "r1025">, DwarfRegNum<[1025]>;
+//def R1024 : AMDILReg<1024, "r1024">, DwarfRegNum<[1024]>;
+//def R1023 : AMDILReg<1023, "r1023">, DwarfRegNum<[1023]>;
+//def R1022 : AMDILReg<1022, "r1022">, DwarfRegNum<[1022]>;
+//def R1021 : AMDILReg<1021, "r1021">, DwarfRegNum<[1021]>;
+//def R1020 : AMDILReg<1020, "r1020">, DwarfRegNum<[1020]>;
+def SP : AMDILReg<1019, "r1019">, DwarfRegNum<[1019]>;
+def T1 : AMDILReg<1018, "r1018">, DwarfRegNum<[1018]>;
+def T2 : AMDILReg<1017, "r1017">, DwarfRegNum<[1017]>;
+def T3 : AMDILReg<1016, "r1016">, DwarfRegNum<[1016]>;
+def T4 : AMDILReg<1015, "r1015">, DwarfRegNum<[1015]>;
+def T5 : AMDILReg<1014, "r1014">, DwarfRegNum<[1014]>;
+def SDP : AMDILReg<1013, "r1013">, DwarfRegNum<[1013]>;
+def R1012: AMDILReg<1012, "r1012">, DwarfRegNum<[1012]>;
+def R1011: AMDILReg<1011, "r1011">, DwarfRegNum<[1011]>;
+def R1010: AMDILReg<1010, "r1010">, DwarfRegNum<[1010]>;
+def DFP : AMDILReg<1009, "r1009">, DwarfRegNum<[1009]>;
+def R1008: AMDILReg<1008, "r1008">, DwarfRegNum<[1008]>;
+def R1007: AMDILReg<1007, "r1007">, DwarfRegNum<[1007]>;
+def R1006: AMDILReg<1006, "r1006">, DwarfRegNum<[1006]>;
+def R1005: AMDILReg<1005, "r1005">, DwarfRegNum<[1005]>;
+def R1004: AMDILReg<1004, "r1004">, DwarfRegNum<[1004]>;
+def R1003: AMDILReg<1003, "r1003">, DwarfRegNum<[1003]>;
+def R1002: AMDILReg<1002, "r1002">, DwarfRegNum<[1002]>;
+def R1001: AMDILReg<1001, "r1001">, DwarfRegNum<[1001]>;
+def MEM : AMDILReg<999, "mem">, DwarfRegNum<[999]>;
+def RA : AMDILReg<998, "r998">, DwarfRegNum<[998]>;
+def FP : AMDILReg<997, "r997">, DwarfRegNum<[997]>;
+def GPRI16 : RegisterClass<"AMDGPU", [i16], 16,
+  (add (sequence "R%u", 1, 20), RA, SP, T1, T2, T3, T4, T5, SDP, R1010, R1011, R1001, R1002, R1003, R1004, R1005, R1006, R1007, R1008, MEM, R1012)> {
+        let AltOrders = [(add (sequence "R%u", 1, 20))];
+        let AltOrderSelect = [{
+          return 1;
+        }];
+    }
+def GPRI32 : RegisterClass<"AMDGPU", [i32], 32,
+  (add (sequence "R%u", 1, 20), RA, SP, T1, T2, T3, T4, T5, SDP, R1010, R1011, R1001, R1002, R1003, R1004, R1005, R1006, R1007, R1008, MEM, R1012)> {
+        let AltOrders = [(add (sequence "R%u", 1, 20))];
+        let AltOrderSelect = [{
+          return 1;
+        }];
+    }
+def GPRF32 : RegisterClass<"AMDGPU", [f32], 32,
+  (add (sequence "R%u", 1, 20), RA, SP, T1, T2, T3, T4, T5, SDP, R1010, R1011, R1001, R1002, R1003, R1004, R1005, R1006, R1007, R1008, MEM, R1012)> {
+        let AltOrders = [(add (sequence "R%u", 1, 20))];
+        let AltOrderSelect = [{
+          return 1;
+        }];
+    }
diff --git a/lib/Target/R600/AMDILSIDevice.cpp b/lib/Target/R600/AMDILSIDevice.cpp
new file mode 100644
index 000000000000..0d1de3d11eb4
--- /dev/null
+++ b/lib/Target/R600/AMDILSIDevice.cpp
@@ -0,0 +1,48 @@
+//===-- AMDILSIDevice.cpp - Device Info for Southern Islands GPUs ---------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//==-----------------------------------------------------------------------===//
+#include "AMDILSIDevice.h"
+#include "AMDGPUSubtarget.h"
+#include "AMDILEvergreenDevice.h"
+#include "AMDILNIDevice.h"
+
+using namespace llvm;
+
+AMDGPUSIDevice::AMDGPUSIDevice(AMDGPUSubtarget *ST)
+  : AMDGPUEvergreenDevice(ST) {
+}
+AMDGPUSIDevice::~AMDGPUSIDevice() {
+}
+
+size_t
+AMDGPUSIDevice::getMaxLDSSize() const {
+  if (usesHardware(AMDGPUDeviceInfo::LocalMem)) {
+    return MAX_LDS_SIZE_900;
+  } else {
+    return 0;
+  }
+}
+
+uint32_t
+AMDGPUSIDevice::getGeneration() const {
+  return AMDGPUDeviceInfo::HD7XXX;
+}
+
+std::string
+AMDGPUSIDevice::getDataLayout() const {
+  return std::string(
+    "e"
+    "-p:64:64:64"
+    "-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64"
+    "-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128"
+    "-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"
+    "-v2048:2048:2048"
+    "-n32:64"
+  );
+}
diff --git a/lib/Target/R600/AMDILSIDevice.h b/lib/Target/R600/AMDILSIDevice.h
new file mode 100644
index 000000000000..5b2cb2502211
--- /dev/null
+++ b/lib/Target/R600/AMDILSIDevice.h
@@ -0,0 +1,39 @@
+//===------- AMDILSIDevice.h - Define SI Device for AMDIL -*- C++ -*------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//==-----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Interface for the subtarget data classes.
+///
+/// This file will define the interface that each generation needs to
+/// implement in order to correctly answer queries on the capabilities of the
+/// specific hardware.
+//===---------------------------------------------------------------------===//
+#ifndef AMDILSIDEVICE_H
+#define AMDILSIDEVICE_H
+#include "AMDILEvergreenDevice.h"
+
+namespace llvm {
+class AMDGPUSubtarget;
+//===---------------------------------------------------------------------===//
+// SI generation of devices and their respective sub classes
+//===---------------------------------------------------------------------===//
+
+/// \brief The AMDGPUSIDevice is the base class for all Southern Island series
+/// of cards.
+class AMDGPUSIDevice : public AMDGPUEvergreenDevice {
+public:
+  AMDGPUSIDevice(AMDGPUSubtarget*);
+  virtual ~AMDGPUSIDevice();
+  virtual size_t getMaxLDSSize() const;
+  virtual uint32_t getGeneration() const;
+  virtual std::string getDataLayout() const;
+};
+
+} // namespace llvm
+#endif // AMDILSIDEVICE_H
diff --git a/lib/Target/R600/CMakeLists.txt b/lib/Target/R600/CMakeLists.txt
new file mode 100644
index 000000000000..8efba5846bef
--- /dev/null
+++ b/lib/Target/R600/CMakeLists.txt
@@ -0,0 +1,59 @@
+set(LLVM_TARGET_DEFINITIONS AMDGPU.td)
+
+tablegen(LLVM AMDGPUGenRegisterInfo.inc -gen-register-info)
+tablegen(LLVM AMDGPUGenInstrInfo.inc -gen-instr-info)
+tablegen(LLVM AMDGPUGenDAGISel.inc -gen-dag-isel)
+tablegen(LLVM AMDGPUGenCallingConv.inc -gen-callingconv)
+tablegen(LLVM AMDGPUGenSubtargetInfo.inc -gen-subtarget)
+tablegen(LLVM AMDGPUGenIntrinsics.inc -gen-tgt-intrinsic)
+tablegen(LLVM AMDGPUGenMCCodeEmitter.inc -gen-emitter -mc-emitter)
+tablegen(LLVM AMDGPUGenDFAPacketizer.inc -gen-dfa-packetizer)
+tablegen(LLVM AMDGPUGenAsmWriter.inc -gen-asm-writer)
+add_public_tablegen_target(AMDGPUCommonTableGen)
+
+add_llvm_target(R600CodeGen
+  AMDIL7XXDevice.cpp
+  AMDILCFGStructurizer.cpp
+  AMDILDevice.cpp
+  AMDILDeviceInfo.cpp
+  AMDILEvergreenDevice.cpp
+  AMDILIntrinsicInfo.cpp
+  AMDILISelDAGToDAG.cpp
+  AMDILISelLowering.cpp
+  AMDILNIDevice.cpp
+  AMDILPeepholeOptimizer.cpp
+  AMDILSIDevice.cpp
+  AMDGPUAsmPrinter.cpp
+  AMDGPUFrameLowering.cpp
+  AMDGPUIndirectAddressing.cpp
+  AMDGPUMCInstLower.cpp
+  AMDGPUMachineFunction.cpp
+  AMDGPUSubtarget.cpp
+  AMDGPUStructurizeCFG.cpp
+  AMDGPUTargetMachine.cpp
+  AMDGPUISelLowering.cpp
+  AMDGPUConvertToISA.cpp
+  AMDGPUInstrInfo.cpp
+  AMDGPURegisterInfo.cpp
+  R600ControlFlowFinalizer.cpp
+  R600EmitClauseMarkers.cpp
+  R600ExpandSpecialInstrs.cpp
+  R600InstrInfo.cpp
+  R600ISelLowering.cpp
+  R600MachineFunctionInfo.cpp
+  R600MachineScheduler.cpp
+  R600RegisterInfo.cpp
+  SIAnnotateControlFlow.cpp
+  SIInsertWaits.cpp
+  SIInstrInfo.cpp
+  SIISelLowering.cpp
+  SILowerControlFlow.cpp
+  SIMachineFunctionInfo.cpp
+  SIRegisterInfo.cpp
+  )
+
+add_dependencies(LLVMR600CodeGen intrinsics_gen)
+
+add_subdirectory(InstPrinter)
+add_subdirectory(TargetInfo)
+add_subdirectory(MCTargetDesc)
diff --git a/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.cpp b/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.cpp
new file mode 100644
index 000000000000..10547a598805
--- /dev/null
+++ b/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.cpp
@@ -0,0 +1,172 @@
+//===-- AMDGPUInstPrinter.cpp - AMDGPU MC Inst -> ASM ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+// \file
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUInstPrinter.h"
+#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCExpr.h"
+
+using namespace llvm;
+
+void AMDGPUInstPrinter::printInst(const MCInst *MI, raw_ostream &OS,
+                             StringRef Annot) {
+  printInstruction(MI, OS);
+
+  printAnnotation(OS, Annot);
+}
+
+void AMDGPUInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
+                                     raw_ostream &O) {
+
+  const MCOperand &Op = MI->getOperand(OpNo);
+  if (Op.isReg()) {
+    switch (Op.getReg()) {
+    // This is the default predicate state, so we don't need to print it.
+    case AMDGPU::PRED_SEL_OFF: break;
+    default: O << getRegisterName(Op.getReg()); break;
+    }
+  } else if (Op.isImm()) {
+    O << Op.getImm();
+  } else if (Op.isFPImm()) {
+    O << Op.getFPImm();
+  } else if (Op.isExpr()) {
+    const MCExpr *Exp = Op.getExpr();
+    Exp->print(O);
+  } else {
+    assert(!"unknown operand type in printOperand");
+  }
+}
+
+void AMDGPUInstPrinter::printInterpSlot(const MCInst *MI, unsigned OpNum,
+                                        raw_ostream &O) {
+  unsigned Imm = MI->getOperand(OpNum).getImm();
+
+  if (Imm == 2) {
+    O << "P0";
+  } else if (Imm == 1) {
+    O << "P20";
+  } else if (Imm == 0) {
+    O << "P10";
+  } else {
+    assert(!"Invalid interpolation parameter slot");
+  }
+}
+
+void AMDGPUInstPrinter::printMemOperand(const MCInst *MI, unsigned OpNo,
+                                        raw_ostream &O) {
+  printOperand(MI, OpNo, O);
+  O  << ", ";
+  printOperand(MI, OpNo + 1, O);
+}
+
+void AMDGPUInstPrinter::printIfSet(const MCInst *MI, unsigned OpNo,
+                                    raw_ostream &O, StringRef Asm) {
+  const MCOperand &Op = MI->getOperand(OpNo);
+  assert(Op.isImm());
+  if (Op.getImm() == 1) {
+    O << Asm;
+  }
+}
+
+void AMDGPUInstPrinter::printAbs(const MCInst *MI, unsigned OpNo,
+                                 raw_ostream &O) {
+  printIfSet(MI, OpNo, O, "|");
+}
+
+void AMDGPUInstPrinter::printClamp(const MCInst *MI, unsigned OpNo,
+                                   raw_ostream &O) {
+  printIfSet(MI, OpNo, O, "_SAT");
+}
+
+void AMDGPUInstPrinter::printLiteral(const MCInst *MI, unsigned OpNo,
+                                     raw_ostream &O) {
+  union Literal {
+    float f;
+    int32_t i;
+  } L;
+
+  L.i = MI->getOperand(OpNo).getImm();
+  O << L.i << "(" << L.f << ")";
+}
+
+void AMDGPUInstPrinter::printLast(const MCInst *MI, unsigned OpNo,
+                                  raw_ostream &O) {
+  printIfSet(MI, OpNo, O, " *");
+}
+
+void AMDGPUInstPrinter::printNeg(const MCInst *MI, unsigned OpNo,
+                                 raw_ostream &O) {
+  printIfSet(MI, OpNo, O, "-");
+}
+
+void AMDGPUInstPrinter::printOMOD(const MCInst *MI, unsigned OpNo,
+                                  raw_ostream &O) {
+  switch (MI->getOperand(OpNo).getImm()) {
+  default: break;
+  case 1:
+    O << " * 2.0";
+    break;
+  case 2:
+    O << " * 4.0";
+    break;
+  case 3:
+    O << " / 2.0";
+    break;
+  }
+}
+
+void AMDGPUInstPrinter::printRel(const MCInst *MI, unsigned OpNo,
+                                 raw_ostream &O) {
+  printIfSet(MI, OpNo, O, "+");
+}
+
+void AMDGPUInstPrinter::printUpdateExecMask(const MCInst *MI, unsigned OpNo,
+                                            raw_ostream &O) {
+  printIfSet(MI, OpNo, O, "ExecMask,");
+}
+
+void AMDGPUInstPrinter::printUpdatePred(const MCInst *MI, unsigned OpNo,
+                                        raw_ostream &O) {
+  printIfSet(MI, OpNo, O, "Pred,");
+}
+
+void AMDGPUInstPrinter::printWrite(const MCInst *MI, unsigned OpNo,
+                                       raw_ostream &O) {
+  const MCOperand &Op = MI->getOperand(OpNo);
+  if (Op.getImm() == 0) {
+    O << " (MASKED)";
+  }
+}
+
+void AMDGPUInstPrinter::printSel(const MCInst *MI, unsigned OpNo,
+                                  raw_ostream &O) {
+  const char * chans = "XYZW";
+  int sel = MI->getOperand(OpNo).getImm();
+
+  int chan = sel & 3;
+  sel >>= 2;
+
+  if (sel >= 512) {
+    sel -= 512;
+    int cb = sel >> 12;
+    sel &= 4095;
+    O << cb << "[" << sel << "]";
+  } else if (sel >= 448) {
+    sel -= 448;
+    O << sel;
+  } else if (sel >= 0){
+    O << sel;
+  }
+
+  if (sel >= 0)
+    O << "." << chans[chan];
+}
+
+#include "AMDGPUGenAsmWriter.inc"
diff --git a/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.h b/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.h
new file mode 100644
index 000000000000..767a7082cc2c
--- /dev/null
+++ b/lib/Target/R600/InstPrinter/AMDGPUInstPrinter.h
@@ -0,0 +1,54 @@
+//===-- AMDGPUInstPrinter.h - AMDGPU MC Inst -> ASM interface ---*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPUINSTPRINTER_H
+#define AMDGPUINSTPRINTER_H
+
+#include "llvm/ADT/StringRef.h"
+#include "llvm/MC/MCInstPrinter.h"
+#include "llvm/Support/raw_ostream.h"
+
+namespace llvm {
+
+class AMDGPUInstPrinter : public MCInstPrinter {
+public:
+  AMDGPUInstPrinter(const MCAsmInfo &MAI, const MCInstrInfo &MII,
+                     const MCRegisterInfo &MRI)
+    : MCInstPrinter(MAI, MII, MRI) {}
+
+  //Autogenerated by tblgen
+  void printInstruction(const MCInst *MI, raw_ostream &O);
+  static const char *getRegisterName(unsigned RegNo);
+
+  virtual void printInst(const MCInst *MI, raw_ostream &O, StringRef Annot);
+
+private:
+  void printOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printInterpSlot(const MCInst *MI, unsigned OpNum, raw_ostream &O);
+  void printMemOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printIfSet(const MCInst *MI, unsigned OpNo, raw_ostream &O, StringRef Asm);
+  void printAbs(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printClamp(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printLiteral(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printLast(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printNeg(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printOMOD(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printRel(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printUpdateExecMask(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printUpdatePred(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printWrite(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printSel(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+};
+
+} // End namespace llvm
+
+#endif // AMDGPUINSTRPRINTER_H
diff --git a/lib/Target/R600/InstPrinter/CMakeLists.txt b/lib/Target/R600/InstPrinter/CMakeLists.txt
new file mode 100644
index 000000000000..069c55ba948e
--- /dev/null
+++ b/lib/Target/R600/InstPrinter/CMakeLists.txt
@@ -0,0 +1,7 @@
+include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
+
+add_llvm_library(LLVMR600AsmPrinter
+  AMDGPUInstPrinter.cpp
+  )
+
+add_dependencies(LLVMR600AsmPrinter AMDGPUCommonTableGen)
diff --git a/lib/Target/R600/InstPrinter/LLVMBuild.txt b/lib/Target/R600/InstPrinter/LLVMBuild.txt
new file mode 100644
index 000000000000..ec0be89f104c
--- /dev/null
+++ b/lib/Target/R600/InstPrinter/LLVMBuild.txt
@@ -0,0 +1,24 @@
+;===- ./lib/Target/R600/InstPrinter/LLVMBuild.txt -----------*- Conf -*--===;
+;
+;                     The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+;   http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[component_0]
+type = Library
+name = R600AsmPrinter
+parent = R600
+required_libraries = MC Support
+add_to_library_groups = R600
+
diff --git a/lib/Target/R600/InstPrinter/Makefile b/lib/Target/R600/InstPrinter/Makefile
new file mode 100644
index 000000000000..a794cc1124ed
--- /dev/null
+++ b/lib/Target/R600/InstPrinter/Makefile
@@ -0,0 +1,15 @@
+#===- lib/Target/R600/AsmPrinter/Makefile ------------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+LEVEL = ../../../..
+LIBRARYNAME = LLVMR600AsmPrinter
+
+# Hack: we need to include 'main' x86 target directory to grab private headers
+CPP.Flags += -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/Target/CellSPU/LLVMBuild.txt b/lib/Target/R600/LLVMBuild.txt
index 277620bf4e59..f2a7554e5269 100644
--- a/lib/Target/CellSPU/LLVMBuild.txt
+++ b/lib/Target/R600/LLVMBuild.txt
@@ -1,4 +1,4 @@
-;===- ./lib/Target/CellSPU/LLVMBuild.txt -----------------------*- Conf -*--===;
+;===- ./lib/Target/AMDIL/LLVMBuild.txt -------------------------*- Conf -*--===;
 ;
 ;                     The LLVM Compiler Infrastructure
 ;
@@ -16,17 +16,17 @@
 ;===------------------------------------------------------------------------===;
 
 [common]
-subdirectories = MCTargetDesc TargetInfo
+subdirectories = InstPrinter MCTargetDesc TargetInfo
 
 [component_0]
 type = TargetGroup
-name = CellSPU
+name = R600
 parent = Target
 has_asmprinter = 1
 
 [component_1]
 type = Library
-name = CellSPUCodeGen
-parent = CellSPU
-required_libraries = AsmPrinter CellSPUDesc CellSPUInfo CodeGen Core MC SelectionDAG Support Target
-add_to_library_groups = CellSPU
+name = R600CodeGen
+parent = R600
+required_libraries = AsmPrinter CodeGen Core SelectionDAG Support Target MC R600AsmPrinter R600Desc R600Info 
+add_to_library_groups = R600
diff --git a/lib/Target/R600/MCTargetDesc/AMDGPUAsmBackend.cpp b/lib/Target/R600/MCTargetDesc/AMDGPUAsmBackend.cpp
new file mode 100644
index 000000000000..98fca432670d
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/AMDGPUAsmBackend.cpp
@@ -0,0 +1,90 @@
+//===-- AMDGPUAsmBackend.cpp - AMDGPU Assembler Backend -------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//===----------------------------------------------------------------------===//
+
+#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/MC/MCAsmBackend.h"
+#include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCObjectWriter.h"
+#include "llvm/MC/MCValue.h"
+#include "llvm/Support/TargetRegistry.h"
+
+using namespace llvm;
+
+namespace {
+
+class AMDGPUMCObjectWriter : public MCObjectWriter {
+public:
+  AMDGPUMCObjectWriter(raw_ostream &OS) : MCObjectWriter(OS, true) { }
+  virtual void ExecutePostLayoutBinding(MCAssembler &Asm,
+                                        const MCAsmLayout &Layout) {
+    //XXX: Implement if necessary.
+  }
+  virtual void RecordRelocation(const MCAssembler &Asm,
+                                const MCAsmLayout &Layout,
+                                const MCFragment *Fragment,
+                                const MCFixup &Fixup,
+                                MCValue Target, uint64_t &FixedValue) {
+    assert(!"Not implemented");
+  }
+
+  virtual void WriteObject(MCAssembler &Asm, const MCAsmLayout &Layout);
+
+};
+
+class AMDGPUAsmBackend : public MCAsmBackend {
+public:
+  AMDGPUAsmBackend(const Target &T)
+    : MCAsmBackend() {}
+
+  virtual AMDGPUMCObjectWriter *createObjectWriter(raw_ostream &OS) const;
+  virtual unsigned getNumFixupKinds() const { return 0; };
+  virtual void applyFixup(const MCFixup &Fixup, char *Data, unsigned DataSize,
+                          uint64_t Value) const;
+  virtual bool fixupNeedsRelaxation(const MCFixup &Fixup, uint64_t Value,
+                                    const MCRelaxableFragment *DF,
+                                    const MCAsmLayout &Layout) const {
+    return false;
+  }
+  virtual void relaxInstruction(const MCInst &Inst, MCInst &Res) const {
+    assert(!"Not implemented");
+  }
+  virtual bool mayNeedRelaxation(const MCInst &Inst) const { return false; }
+  virtual bool writeNopData(uint64_t Count, MCObjectWriter *OW) const {
+    return true;
+  }
+};
+
+} //End anonymous namespace
+
+void AMDGPUMCObjectWriter::WriteObject(MCAssembler &Asm,
+                                       const MCAsmLayout &Layout) {
+  for (MCAssembler::iterator I = Asm.begin(), E = Asm.end(); I != E; ++I) {
+    Asm.writeSectionData(I, Layout);
+  }
+}
+
+MCAsmBackend *llvm::createAMDGPUAsmBackend(const Target &T, StringRef TT,
+                                           StringRef CPU) {
+  return new AMDGPUAsmBackend(T);
+}
+
+AMDGPUMCObjectWriter * AMDGPUAsmBackend::createObjectWriter(
+                                                        raw_ostream &OS) const {
+  return new AMDGPUMCObjectWriter(OS);
+}
+
+void AMDGPUAsmBackend::applyFixup(const MCFixup &Fixup, char *Data,
+                                  unsigned DataSize, uint64_t Value) const {
+
+  uint16_t *Dst = (uint16_t*)(Data + Fixup.getOffset());
+  assert(Fixup.getKind() == FK_PCRel_4);
+  *Dst = (Value - 4) / 4;
+}
diff --git a/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.cpp b/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.cpp
new file mode 100644
index 000000000000..b7cdd7c8cde9
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.cpp
@@ -0,0 +1,83 @@
+//===-- MCTargetDesc/AMDGPUMCAsmInfo.cpp - Assembly Info ------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUMCAsmInfo.h"
+
+using namespace llvm;
+AMDGPUMCAsmInfo::AMDGPUMCAsmInfo(const Target &T, StringRef &TT) : MCAsmInfo() {
+  HasSingleParameterDotFile = false;
+  WeakDefDirective = 0;
+  //===------------------------------------------------------------------===//
+  HasSubsectionsViaSymbols = true;
+  HasMachoZeroFillDirective = false;
+  HasMachoTBSSDirective = false;
+  HasStaticCtorDtorReferenceInStaticMode = false;
+  LinkerRequiresNonEmptyDwarfLines = true;
+  MaxInstLength = 16;
+  PCSymbol = "$";
+  SeparatorString = "\n";
+  CommentColumn = 40;
+  CommentString = ";";
+  LabelSuffix = ":";
+  GlobalPrefix = "@";
+  PrivateGlobalPrefix = ";.";
+  LinkerPrivateGlobalPrefix = "!";
+  InlineAsmStart = ";#ASMSTART";
+  InlineAsmEnd = ";#ASMEND";
+  AssemblerDialect = 0;
+  AllowQuotesInName = false;
+  AllowNameToStartWithDigit = false;
+  AllowPeriodsInName = false;
+
+  //===--- Data Emission Directives -------------------------------------===//
+  ZeroDirective = ".zero";
+  AsciiDirective = ".ascii\t";
+  AscizDirective = ".asciz\t";
+  Data8bitsDirective = ".byte\t";
+  Data16bitsDirective = ".short\t";
+  Data32bitsDirective = ".long\t";
+  Data64bitsDirective = ".quad\t";
+  GPRel32Directive = 0;
+  SunStyleELFSectionSwitchSyntax = true;
+  UsesELFSectionDirectiveForBSS = true;
+  HasMicrosoftFastStdCallMangling = false;
+
+  //===--- Alignment Information ----------------------------------------===//
+  AlignDirective = ".align\t";
+  AlignmentIsInBytes = true;
+  TextAlignFillValue = 0;
+
+  //===--- Global Variable Emission Directives --------------------------===//
+  GlobalDirective = ".global";
+  ExternDirective = ".extern";
+  HasSetDirective = false;
+  HasAggressiveSymbolFolding = true;
+  COMMDirectiveAlignmentIsInBytes = false;
+  HasDotTypeDotSizeDirective = false;
+  HasNoDeadStrip = true;
+  HasSymbolResolver = false;
+  WeakRefDirective = ".weakref\t";
+  LinkOnceDirective = 0;
+  //===--- Dwarf Emission Directives -----------------------------------===//
+  HasLEB128 = true;
+  SupportsDebugInformation = true;
+  DwarfSectionOffsetDirective = ".offset";
+
+}
+
+const char*
+AMDGPUMCAsmInfo::getDataASDirective(unsigned int Size, unsigned int AS) const {
+  return 0;
+}
+
+const MCSection*
+AMDGPUMCAsmInfo::getNonexecutableStackSection(MCContext &CTX) const {
+  return 0;
+}
diff --git a/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.h b/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.h
new file mode 100644
index 000000000000..3ad0fa6824ab
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/AMDGPUMCAsmInfo.h
@@ -0,0 +1,30 @@
+//===-- MCTargetDesc/AMDGPUMCAsmInfo.h - AMDGPU MCAsm Interface  ----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPUMCASMINFO_H
+#define AMDGPUMCASMINFO_H
+
+#include "llvm/MC/MCAsmInfo.h"
+namespace llvm {
+
+class Target;
+class StringRef;
+
+class AMDGPUMCAsmInfo : public MCAsmInfo {
+public:
+  explicit AMDGPUMCAsmInfo(const Target &T, StringRef &TT);
+  const char* getDataASDirective(unsigned int Size, unsigned int AS) const;
+  const MCSection* getNonexecutableStackSection(MCContext &CTX) const;
+};
+} // namespace llvm
+#endif // AMDGPUMCASMINFO_H
diff --git a/lib/Target/R600/MCTargetDesc/AMDGPUMCCodeEmitter.h b/lib/Target/R600/MCTargetDesc/AMDGPUMCCodeEmitter.h
new file mode 100644
index 000000000000..cd3a7ce65aa5
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/AMDGPUMCCodeEmitter.h
@@ -0,0 +1,40 @@
+//===-- AMDGPUCodeEmitter.h - AMDGPU Code Emitter interface -----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief CodeEmitter interface for R600 and SI codegen.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef AMDGPUCODEEMITTER_H
+#define AMDGPUCODEEMITTER_H
+
+#include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/Support/raw_ostream.h"
+
+namespace llvm {
+
+class MCInst;
+class MCOperand;
+
+class AMDGPUMCCodeEmitter : public MCCodeEmitter {
+public:
+
+  uint64_t getBinaryCodeForInstr(const MCInst &MI,
+                                 SmallVectorImpl<MCFixup> &Fixups) const;
+
+  virtual uint64_t getMachineOpValue(const MCInst &MI, const MCOperand &MO,
+                                     SmallVectorImpl<MCFixup> &Fixups) const {
+    return 0;
+  }
+};
+
+} // End namespace llvm
+
+#endif // AMDGPUCODEEMITTER_H
diff --git a/lib/Target/R600/MCTargetDesc/AMDGPUMCTargetDesc.cpp b/lib/Target/R600/MCTargetDesc/AMDGPUMCTargetDesc.cpp
new file mode 100644
index 000000000000..072ee49b6311
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/AMDGPUMCTargetDesc.cpp
@@ -0,0 +1,113 @@
+//===-- AMDGPUMCTargetDesc.cpp - AMDGPU Target Descriptions ---------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief This file provides AMDGPU specific target descriptions.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUMCTargetDesc.h"
+#include "AMDGPUMCAsmInfo.h"
+#include "InstPrinter/AMDGPUInstPrinter.h"
+#include "llvm/MC/MCCodeGenInfo.h"
+#include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/MC/MachineLocation.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/TargetRegistry.h"
+
+#define GET_INSTRINFO_MC_DESC
+#include "AMDGPUGenInstrInfo.inc"
+
+#define GET_SUBTARGETINFO_MC_DESC
+#include "AMDGPUGenSubtargetInfo.inc"
+
+#define GET_REGINFO_MC_DESC
+#include "AMDGPUGenRegisterInfo.inc"
+
+using namespace llvm;
+
+static MCInstrInfo *createAMDGPUMCInstrInfo() {
+  MCInstrInfo *X = new MCInstrInfo();
+  InitAMDGPUMCInstrInfo(X);
+  return X;
+}
+
+static MCRegisterInfo *createAMDGPUMCRegisterInfo(StringRef TT) {
+  MCRegisterInfo *X = new MCRegisterInfo();
+  InitAMDGPUMCRegisterInfo(X, 0);
+  return X;
+}
+
+static MCSubtargetInfo *createAMDGPUMCSubtargetInfo(StringRef TT, StringRef CPU,
+                                                   StringRef FS) {
+  MCSubtargetInfo * X = new MCSubtargetInfo();
+  InitAMDGPUMCSubtargetInfo(X, TT, CPU, FS);
+  return X;
+}
+
+static MCCodeGenInfo *createAMDGPUMCCodeGenInfo(StringRef TT, Reloc::Model RM,
+                                               CodeModel::Model CM,
+                                               CodeGenOpt::Level OL) {
+  MCCodeGenInfo *X = new MCCodeGenInfo();
+  X->InitMCCodeGenInfo(RM, CM, OL);
+  return X;
+}
+
+static MCInstPrinter *createAMDGPUMCInstPrinter(const Target &T,
+                                                unsigned SyntaxVariant,
+                                                const MCAsmInfo &MAI,
+                                                const MCInstrInfo &MII,
+                                                const MCRegisterInfo &MRI,
+                                                const MCSubtargetInfo &STI) {
+  return new AMDGPUInstPrinter(MAI, MII, MRI);
+}
+
+static MCCodeEmitter *createAMDGPUMCCodeEmitter(const MCInstrInfo &MCII,
+                                                const MCRegisterInfo &MRI,
+                                                const MCSubtargetInfo &STI,
+                                                MCContext &Ctx) {
+  if (STI.getFeatureBits() & AMDGPU::Feature64BitPtr) {
+    return createSIMCCodeEmitter(MCII, MRI, STI, Ctx);
+  } else {
+    return createR600MCCodeEmitter(MCII, MRI, STI, Ctx);
+  }
+}
+
+static MCStreamer *createMCStreamer(const Target &T, StringRef TT,
+                                    MCContext &Ctx, MCAsmBackend &MAB,
+                                    raw_ostream &_OS,
+                                    MCCodeEmitter *_Emitter,
+                                    bool RelaxAll,
+                                    bool NoExecStack) {
+  return createPureStreamer(Ctx, MAB, _OS, _Emitter);
+}
+
+extern "C" void LLVMInitializeR600TargetMC() {
+
+  RegisterMCAsmInfo<AMDGPUMCAsmInfo> Y(TheAMDGPUTarget);
+
+  TargetRegistry::RegisterMCCodeGenInfo(TheAMDGPUTarget, createAMDGPUMCCodeGenInfo);
+
+  TargetRegistry::RegisterMCInstrInfo(TheAMDGPUTarget, createAMDGPUMCInstrInfo);
+
+  TargetRegistry::RegisterMCRegInfo(TheAMDGPUTarget, createAMDGPUMCRegisterInfo);
+
+  TargetRegistry::RegisterMCSubtargetInfo(TheAMDGPUTarget, createAMDGPUMCSubtargetInfo);
+
+  TargetRegistry::RegisterMCInstPrinter(TheAMDGPUTarget, createAMDGPUMCInstPrinter);
+
+  TargetRegistry::RegisterMCCodeEmitter(TheAMDGPUTarget, createAMDGPUMCCodeEmitter);
+
+  TargetRegistry::RegisterMCAsmBackend(TheAMDGPUTarget, createAMDGPUAsmBackend);
+
+  TargetRegistry::RegisterMCObjectStreamer(TheAMDGPUTarget, createMCStreamer);
+}
diff --git a/lib/Target/R600/MCTargetDesc/AMDGPUMCTargetDesc.h b/lib/Target/R600/MCTargetDesc/AMDGPUMCTargetDesc.h
new file mode 100644
index 000000000000..363a4af3f3a4
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/AMDGPUMCTargetDesc.h
@@ -0,0 +1,55 @@
+//===-- AMDGPUMCTargetDesc.h - AMDGPU Target Descriptions -----*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Provides AMDGPU specific target descriptions.
+//
+//===----------------------------------------------------------------------===//
+//
+
+#ifndef AMDGPUMCTARGETDESC_H
+#define AMDGPUMCTARGETDESC_H
+
+#include "llvm/ADT/StringRef.h"
+
+namespace llvm {
+class MCAsmBackend;
+class MCCodeEmitter;
+class MCContext;
+class MCInstrInfo;
+class MCRegisterInfo;
+class MCSubtargetInfo;
+class Target;
+
+extern Target TheAMDGPUTarget;
+
+MCCodeEmitter *createR600MCCodeEmitter(const MCInstrInfo &MCII,
+                                       const MCRegisterInfo &MRI,
+                                       const MCSubtargetInfo &STI,
+                                       MCContext &Ctx);
+
+MCCodeEmitter *createSIMCCodeEmitter(const MCInstrInfo &MCII,
+                                     const MCRegisterInfo &MRI,
+                                     const MCSubtargetInfo &STI,
+                                     MCContext &Ctx);
+
+MCAsmBackend *createAMDGPUAsmBackend(const Target &T, StringRef TT,
+                                     StringRef CPU);
+} // End llvm namespace
+
+#define GET_REGINFO_ENUM
+#include "AMDGPUGenRegisterInfo.inc"
+
+#define GET_INSTRINFO_ENUM
+#include "AMDGPUGenInstrInfo.inc"
+
+#define GET_SUBTARGETINFO_ENUM
+#include "AMDGPUGenSubtargetInfo.inc"
+
+#endif // AMDGPUMCTARGETDESC_H
diff --git a/lib/Target/R600/MCTargetDesc/CMakeLists.txt b/lib/Target/R600/MCTargetDesc/CMakeLists.txt
new file mode 100644
index 000000000000..37e714c2e7b8
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/CMakeLists.txt
@@ -0,0 +1,10 @@
+
+add_llvm_library(LLVMR600Desc
+  AMDGPUAsmBackend.cpp
+  AMDGPUMCTargetDesc.cpp
+  AMDGPUMCAsmInfo.cpp
+  R600MCCodeEmitter.cpp
+  SIMCCodeEmitter.cpp
+  )
+
+add_dependencies(LLVMR600Desc AMDGPUCommonTableGen)
diff --git a/lib/Target/R600/MCTargetDesc/LLVMBuild.txt b/lib/Target/R600/MCTargetDesc/LLVMBuild.txt
new file mode 100644
index 000000000000..b1beab0bb301
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/LLVMBuild.txt
@@ -0,0 +1,23 @@
+;===- ./lib/Target/R600/MCTargetDesc/LLVMBuild.txt ------------*- Conf -*--===;
+;
+;                     The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+;   http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[component_0]
+type = Library
+name = R600Desc
+parent = R600
+required_libraries = R600AsmPrinter R600Info MC
+add_to_library_groups = R600
diff --git a/lib/Target/R600/MCTargetDesc/Makefile b/lib/Target/R600/MCTargetDesc/Makefile
new file mode 100644
index 000000000000..8894a7607f4f
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/Makefile
@@ -0,0 +1,16 @@
+##===- lib/Target/AMDGPU/TargetDesc/Makefile ----------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+
+LEVEL = ../../../..
+LIBRARYNAME = LLVMR600Desc
+
+# Hack: we need to include 'main' target directory to grab private headers
+CPP.Flags += -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp b/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp
new file mode 100644
index 000000000000..927bcbd8305c
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/R600MCCodeEmitter.cpp
@@ -0,0 +1,585 @@
+//===- R600MCCodeEmitter.cpp - Code Emitter for R600->Cayman GPU families -===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+///
+/// This code emitter outputs bytecode that is understood by the r600g driver
+/// in the Mesa [1] project.  The bytecode is very similar to the hardware's ISA,
+/// but it still needs to be run through a finalizer in order to be executed
+/// by the GPU.
+///
+/// [1] http://www.mesa3d.org/
+//
+//===----------------------------------------------------------------------===//
+
+#include "R600Defines.h"
+#include "MCTargetDesc/AMDGPUMCCodeEmitter.h"
+#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
+#include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/Support/raw_ostream.h"
+#include <stdio.h>
+
+#define SRC_BYTE_COUNT 11
+#define DST_BYTE_COUNT 5
+
+using namespace llvm;
+
+namespace {
+
+class R600MCCodeEmitter : public AMDGPUMCCodeEmitter {
+  R600MCCodeEmitter(const R600MCCodeEmitter &) LLVM_DELETED_FUNCTION;
+  void operator=(const R600MCCodeEmitter &) LLVM_DELETED_FUNCTION;
+  const MCInstrInfo &MCII;
+  const MCRegisterInfo &MRI;
+  const MCSubtargetInfo &STI;
+  MCContext &Ctx;
+
+public:
+
+  R600MCCodeEmitter(const MCInstrInfo &mcii, const MCRegisterInfo &mri,
+                    const MCSubtargetInfo &sti, MCContext &ctx)
+    : MCII(mcii), MRI(mri), STI(sti), Ctx(ctx) { }
+
+  /// \brief Encode the instruction and write it to the OS.
+  virtual void EncodeInstruction(const MCInst &MI, raw_ostream &OS,
+                         SmallVectorImpl<MCFixup> &Fixups) const;
+
+  /// \returns the encoding for an MCOperand.
+  virtual uint64_t getMachineOpValue(const MCInst &MI, const MCOperand &MO,
+                                     SmallVectorImpl<MCFixup> &Fixups) const;
+private:
+
+  void EmitALUInstr(const MCInst &MI, SmallVectorImpl<MCFixup> &Fixups,
+                    raw_ostream &OS) const;
+  void EmitSrc(const MCInst &MI, unsigned OpIdx, raw_ostream &OS) const;
+  void EmitSrcISA(const MCInst &MI, unsigned RegOpIdx, unsigned SelOpIdx,
+                    raw_ostream &OS) const;
+  void EmitDst(const MCInst &MI, raw_ostream &OS) const;
+  void EmitFCInstr(const MCInst &MI, raw_ostream &OS) const;
+
+  void EmitNullBytes(unsigned int byteCount, raw_ostream &OS) const;
+
+  void EmitByte(unsigned int byte, raw_ostream &OS) const;
+
+  void EmitTwoBytes(uint32_t bytes, raw_ostream &OS) const;
+
+  void Emit(uint32_t value, raw_ostream &OS) const;
+  void Emit(uint64_t value, raw_ostream &OS) const;
+
+  unsigned getHWRegChan(unsigned reg) const;
+  unsigned getHWReg(unsigned regNo) const;
+
+  bool isFCOp(unsigned opcode) const;
+  bool isTexOp(unsigned opcode) const;
+  bool isFlagSet(const MCInst &MI, unsigned Operand, unsigned Flag) const;
+
+};
+
+} // End anonymous namespace
+
+enum RegElement {
+  ELEMENT_X = 0,
+  ELEMENT_Y,
+  ELEMENT_Z,
+  ELEMENT_W
+};
+
+enum InstrTypes {
+  INSTR_ALU = 0,
+  INSTR_TEX,
+  INSTR_FC,
+  INSTR_NATIVE,
+  INSTR_VTX,
+  INSTR_EXPORT,
+  INSTR_CFALU
+};
+
+enum FCInstr {
+  FC_IF_PREDICATE = 0,
+  FC_ELSE,
+  FC_ENDIF,
+  FC_BGNLOOP,
+  FC_ENDLOOP,
+  FC_BREAK_PREDICATE,
+  FC_CONTINUE
+};
+
+enum TextureTypes {
+  TEXTURE_1D = 1,
+  TEXTURE_2D,
+  TEXTURE_3D,
+  TEXTURE_CUBE,
+  TEXTURE_RECT,
+  TEXTURE_SHADOW1D,
+  TEXTURE_SHADOW2D,
+  TEXTURE_SHADOWRECT,
+  TEXTURE_1D_ARRAY,
+  TEXTURE_2D_ARRAY,
+  TEXTURE_SHADOW1D_ARRAY,
+  TEXTURE_SHADOW2D_ARRAY
+};
+
+MCCodeEmitter *llvm::createR600MCCodeEmitter(const MCInstrInfo &MCII,
+                                           const MCRegisterInfo &MRI,
+                                           const MCSubtargetInfo &STI,
+                                           MCContext &Ctx) {
+  return new R600MCCodeEmitter(MCII, MRI, STI, Ctx);
+}
+
+void R600MCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+  if (isFCOp(MI.getOpcode())){
+    EmitFCInstr(MI, OS);
+  } else if (MI.getOpcode() == AMDGPU::RETURN ||
+    MI.getOpcode() == AMDGPU::BUNDLE ||
+    MI.getOpcode() == AMDGPU::KILL) {
+    return;
+  } else {
+    switch(MI.getOpcode()) {
+    case AMDGPU::STACK_SIZE: {
+      EmitByte(MI.getOperand(0).getImm(), OS);
+      break;
+    }
+    case AMDGPU::RAT_WRITE_CACHELESS_32_eg:
+    case AMDGPU::RAT_WRITE_CACHELESS_128_eg: {
+      uint64_t inst = getBinaryCodeForInstr(MI, Fixups);
+      EmitByte(INSTR_NATIVE, OS);
+      Emit(inst, OS);
+      break;
+    }
+    case AMDGPU::CONSTANT_LOAD_eg:
+    case AMDGPU::VTX_READ_PARAM_8_eg:
+    case AMDGPU::VTX_READ_PARAM_16_eg:
+    case AMDGPU::VTX_READ_PARAM_32_eg:
+    case AMDGPU::VTX_READ_PARAM_128_eg:
+    case AMDGPU::VTX_READ_GLOBAL_8_eg:
+    case AMDGPU::VTX_READ_GLOBAL_32_eg:
+    case AMDGPU::VTX_READ_GLOBAL_128_eg:
+    case AMDGPU::TEX_VTX_CONSTBUF:
+    case AMDGPU::TEX_VTX_TEXBUF : {
+      uint64_t InstWord01 = getBinaryCodeForInstr(MI, Fixups);
+      uint32_t InstWord2 = MI.getOperand(2).getImm(); // Offset
+
+      EmitByte(INSTR_VTX, OS);
+      Emit(InstWord01, OS);
+      Emit(InstWord2, OS);
+      break;
+    }
+    case AMDGPU::TEX_LD:
+    case AMDGPU::TEX_GET_TEXTURE_RESINFO:
+    case AMDGPU::TEX_SAMPLE:
+    case AMDGPU::TEX_SAMPLE_C:
+    case AMDGPU::TEX_SAMPLE_L:
+    case AMDGPU::TEX_SAMPLE_C_L:
+    case AMDGPU::TEX_SAMPLE_LB:
+    case AMDGPU::TEX_SAMPLE_C_LB:
+    case AMDGPU::TEX_SAMPLE_G:
+    case AMDGPU::TEX_SAMPLE_C_G:
+    case AMDGPU::TEX_GET_GRADIENTS_H:
+    case AMDGPU::TEX_GET_GRADIENTS_V:
+    case AMDGPU::TEX_SET_GRADIENTS_H:
+    case AMDGPU::TEX_SET_GRADIENTS_V: {
+      unsigned Opcode = MI.getOpcode();
+      bool HasOffsets = (Opcode == AMDGPU::TEX_LD);
+      unsigned OpOffset = HasOffsets ? 3 : 0;
+      int64_t Sampler = MI.getOperand(OpOffset + 3).getImm();
+      int64_t TextureType = MI.getOperand(OpOffset + 4).getImm();
+
+      uint32_t SrcSelect[4] = {0, 1, 2, 3};
+      uint32_t Offsets[3] = {0, 0, 0};
+      uint64_t CoordType[4] = {1, 1, 1, 1};
+
+      if (HasOffsets)
+        for (unsigned i = 0; i < 3; i++) {
+          int SignedOffset = MI.getOperand(i + 2).getImm();
+          Offsets[i] = (SignedOffset & 0x1F);
+        }
+          
+
+      if (TextureType == TEXTURE_RECT ||
+          TextureType == TEXTURE_SHADOWRECT) {
+        CoordType[ELEMENT_X] = 0;
+        CoordType[ELEMENT_Y] = 0;
+      }
+
+      if (TextureType == TEXTURE_1D_ARRAY ||
+          TextureType == TEXTURE_SHADOW1D_ARRAY) {
+        if (Opcode == AMDGPU::TEX_SAMPLE_C_L ||
+            Opcode == AMDGPU::TEX_SAMPLE_C_LB) {
+          CoordType[ELEMENT_Y] = 0;
+        } else {
+          CoordType[ELEMENT_Z] = 0;
+          SrcSelect[ELEMENT_Z] = ELEMENT_Y;
+        }
+      } else if (TextureType == TEXTURE_2D_ARRAY ||
+          TextureType == TEXTURE_SHADOW2D_ARRAY) {
+        CoordType[ELEMENT_Z] = 0;
+      }
+
+
+      if ((TextureType == TEXTURE_SHADOW1D ||
+          TextureType == TEXTURE_SHADOW2D ||
+          TextureType == TEXTURE_SHADOWRECT ||
+          TextureType == TEXTURE_SHADOW1D_ARRAY) &&
+          Opcode != AMDGPU::TEX_SAMPLE_C_L &&
+          Opcode != AMDGPU::TEX_SAMPLE_C_LB) {
+        SrcSelect[ELEMENT_W] = ELEMENT_Z;
+      }
+
+      uint64_t Word01 = getBinaryCodeForInstr(MI, Fixups) |
+          CoordType[ELEMENT_X] << 60 | CoordType[ELEMENT_Y] << 61 |
+          CoordType[ELEMENT_Z] << 62 | CoordType[ELEMENT_W] << 63;
+      uint32_t Word2 = Sampler << 15 | SrcSelect[ELEMENT_X] << 20 |
+          SrcSelect[ELEMENT_Y] << 23 | SrcSelect[ELEMENT_Z] << 26 |
+          SrcSelect[ELEMENT_W] << 29 | Offsets[0] << 0 | Offsets[1] << 5 |
+          Offsets[2] << 10;
+
+      EmitByte(INSTR_TEX, OS);
+      Emit(Word01, OS);
+      Emit(Word2, OS);
+      break;
+    }
+    case AMDGPU::EG_ExportSwz:
+    case AMDGPU::R600_ExportSwz:
+    case AMDGPU::EG_ExportBuf:
+    case AMDGPU::R600_ExportBuf: {
+      uint64_t Inst = getBinaryCodeForInstr(MI, Fixups);
+      EmitByte(INSTR_EXPORT, OS);
+      Emit(Inst, OS);
+      break;
+    }
+    case AMDGPU::CF_ALU:
+    case AMDGPU::CF_ALU_PUSH_BEFORE: {
+      uint64_t Inst = getBinaryCodeForInstr(MI, Fixups);
+      EmitByte(INSTR_CFALU, OS);
+      Emit(Inst, OS);
+      break;
+    }
+    case AMDGPU::CF_TC:
+    case AMDGPU::CF_VC:
+    case AMDGPU::CF_CALL_FS:
+      return;
+    case AMDGPU::WHILE_LOOP:
+    case AMDGPU::END_LOOP:
+    case AMDGPU::LOOP_BREAK:
+    case AMDGPU::CF_CONTINUE:
+    case AMDGPU::CF_JUMP:
+    case AMDGPU::CF_ELSE:
+    case AMDGPU::POP: {
+      uint64_t Inst = getBinaryCodeForInstr(MI, Fixups);
+      EmitByte(INSTR_NATIVE, OS);
+      Emit(Inst, OS);
+      break;
+    }
+    default:
+      EmitALUInstr(MI, Fixups, OS);
+      break;
+    }
+  }
+}
+
+void R600MCCodeEmitter::EmitALUInstr(const MCInst &MI,
+                                     SmallVectorImpl<MCFixup> &Fixups,
+                                     raw_ostream &OS) const {
+  const MCInstrDesc &MCDesc = MCII.get(MI.getOpcode());
+
+  // Emit instruction type
+  EmitByte(INSTR_ALU, OS);
+
+  uint64_t InstWord01 = getBinaryCodeForInstr(MI, Fixups);
+
+  //older alu have different encoding for instructions with one or two src
+  //parameters.
+  if ((STI.getFeatureBits() & AMDGPU::FeatureR600ALUInst) &&
+      !(MCDesc.TSFlags & R600_InstFlag::OP3)) {
+    uint64_t ISAOpCode = InstWord01 & (0x3FFULL << 39);
+    InstWord01 &= ~(0x3FFULL << 39);
+    InstWord01 |= ISAOpCode << 1;
+  }
+
+  unsigned SrcNum = MCDesc.TSFlags & R600_InstFlag::OP3 ? 3 :
+      MCDesc.TSFlags & R600_InstFlag::OP2 ? 2 : 1;
+
+  EmitByte(SrcNum, OS);
+
+  const unsigned SrcOps[3][2] = {
+      {R600Operands::SRC0, R600Operands::SRC0_SEL},
+      {R600Operands::SRC1, R600Operands::SRC1_SEL},
+      {R600Operands::SRC2, R600Operands::SRC2_SEL}
+  };
+
+  for (unsigned SrcIdx = 0; SrcIdx < SrcNum; ++SrcIdx) {
+    unsigned RegOpIdx = R600Operands::ALUOpTable[SrcNum-1][SrcOps[SrcIdx][0]];
+    unsigned SelOpIdx = R600Operands::ALUOpTable[SrcNum-1][SrcOps[SrcIdx][1]];
+    EmitSrcISA(MI, RegOpIdx, SelOpIdx, OS);
+  }
+
+  Emit(InstWord01, OS);
+  return;
+}
+
+void R600MCCodeEmitter::EmitSrc(const MCInst &MI, unsigned OpIdx,
+                                raw_ostream &OS) const {
+  const MCOperand &MO = MI.getOperand(OpIdx);
+  union {
+    float f;
+    uint32_t i;
+  } Value;
+  Value.i = 0;
+  // Emit the source select (2 bytes).  For GPRs, this is the register index.
+  // For other potential instruction operands, (e.g. constant registers) the
+  // value of the source select is defined in the r600isa docs.
+  if (MO.isReg()) {
+    unsigned reg = MO.getReg();
+    EmitTwoBytes(getHWReg(reg), OS);
+    if (reg == AMDGPU::ALU_LITERAL_X) {
+      unsigned ImmOpIndex = MI.getNumOperands() - 1;
+      MCOperand ImmOp = MI.getOperand(ImmOpIndex);
+      if (ImmOp.isFPImm()) {
+        Value.f = ImmOp.getFPImm();
+      } else {
+        assert(ImmOp.isImm());
+        Value.i = ImmOp.getImm();
+      }
+    }
+  } else {
+    // XXX: Handle other operand types.
+    EmitTwoBytes(0, OS);
+  }
+
+  // Emit the source channel (1 byte)
+  if (MO.isReg()) {
+    EmitByte(getHWRegChan(MO.getReg()), OS);
+  } else {
+    EmitByte(0, OS);
+  }
+
+  // XXX: Emit isNegated (1 byte)
+  if ((!(isFlagSet(MI, OpIdx, MO_FLAG_ABS)))
+      && (isFlagSet(MI, OpIdx, MO_FLAG_NEG) ||
+     (MO.isReg() &&
+      (MO.getReg() == AMDGPU::NEG_ONE || MO.getReg() == AMDGPU::NEG_HALF)))){
+    EmitByte(1, OS);
+  } else {
+    EmitByte(0, OS);
+  }
+
+  // Emit isAbsolute (1 byte)
+  if (isFlagSet(MI, OpIdx, MO_FLAG_ABS)) {
+    EmitByte(1, OS);
+  } else {
+    EmitByte(0, OS);
+  }
+
+  // XXX: Emit relative addressing mode (1 byte)
+  EmitByte(0, OS);
+
+  // Emit kc_bank, This will be adjusted later by r600_asm
+  EmitByte(0, OS);
+
+  // Emit the literal value, if applicable (4 bytes).
+  Emit(Value.i, OS);
+
+}
+
+void R600MCCodeEmitter::EmitSrcISA(const MCInst &MI, unsigned RegOpIdx,
+                                   unsigned SelOpIdx, raw_ostream &OS) const {
+  const MCOperand &RegMO = MI.getOperand(RegOpIdx);
+  const MCOperand &SelMO = MI.getOperand(SelOpIdx);
+
+  union {
+    float f;
+    uint32_t i;
+  } InlineConstant;
+  InlineConstant.i = 0;
+  // Emit source type (1 byte) and source select (4 bytes). For GPRs type is 0
+  // and select is 0 (GPR index is encoded in the instr encoding. For constants
+  // type is 1 and select is the original const select passed from the driver.
+  unsigned Reg = RegMO.getReg();
+  if (Reg == AMDGPU::ALU_CONST) {
+    EmitByte(1, OS);
+    uint32_t Sel = SelMO.getImm();
+    Emit(Sel, OS);
+  } else {
+    EmitByte(0, OS);
+    Emit((uint32_t)0, OS);
+  }
+
+  if (Reg == AMDGPU::ALU_LITERAL_X) {
+    unsigned ImmOpIndex = MI.getNumOperands() - 1;
+    MCOperand ImmOp = MI.getOperand(ImmOpIndex);
+    if (ImmOp.isFPImm()) {
+      InlineConstant.f = ImmOp.getFPImm();
+    } else {
+      assert(ImmOp.isImm());
+      InlineConstant.i = ImmOp.getImm();
+    }
+  }
+
+  // Emit the literal value, if applicable (4 bytes).
+  Emit(InlineConstant.i, OS);
+}
+
+void R600MCCodeEmitter::EmitFCInstr(const MCInst &MI, raw_ostream &OS) const {
+
+  // Emit instruction type
+  EmitByte(INSTR_FC, OS);
+
+  // Emit SRC
+  unsigned NumOperands = MI.getNumOperands();
+  if (NumOperands > 0) {
+    assert(NumOperands == 1);
+    EmitSrc(MI, 0, OS);
+  } else {
+    EmitNullBytes(SRC_BYTE_COUNT, OS);
+  }
+
+  // Emit FC Instruction
+  enum FCInstr instr;
+  switch (MI.getOpcode()) {
+  case AMDGPU::PREDICATED_BREAK:
+    instr = FC_BREAK_PREDICATE;
+    break;
+  case AMDGPU::CONTINUE:
+    instr = FC_CONTINUE;
+    break;
+  case AMDGPU::IF_PREDICATE_SET:
+    instr = FC_IF_PREDICATE;
+    break;
+  case AMDGPU::ELSE:
+    instr = FC_ELSE;
+    break;
+  case AMDGPU::ENDIF:
+    instr = FC_ENDIF;
+    break;
+  case AMDGPU::ENDLOOP:
+    instr = FC_ENDLOOP;
+    break;
+  case AMDGPU::WHILELOOP:
+    instr = FC_BGNLOOP;
+    break;
+  default:
+    abort();
+    break;
+  }
+  EmitByte(instr, OS);
+}
+
+void R600MCCodeEmitter::EmitNullBytes(unsigned int ByteCount,
+                                      raw_ostream &OS) const {
+
+  for (unsigned int i = 0; i < ByteCount; i++) {
+    EmitByte(0, OS);
+  }
+}
+
+void R600MCCodeEmitter::EmitByte(unsigned int Byte, raw_ostream &OS) const {
+  OS.write((uint8_t) Byte & 0xff);
+}
+
+void R600MCCodeEmitter::EmitTwoBytes(unsigned int Bytes,
+                                     raw_ostream &OS) const {
+  OS.write((uint8_t) (Bytes & 0xff));
+  OS.write((uint8_t) ((Bytes >> 8) & 0xff));
+}
+
+void R600MCCodeEmitter::Emit(uint32_t Value, raw_ostream &OS) const {
+  for (unsigned i = 0; i < 4; i++) {
+    OS.write((uint8_t) ((Value >> (8 * i)) & 0xff));
+  }
+}
+
+void R600MCCodeEmitter::Emit(uint64_t Value, raw_ostream &OS) const {
+  for (unsigned i = 0; i < 8; i++) {
+    EmitByte((Value >> (8 * i)) & 0xff, OS);
+  }
+}
+
+unsigned R600MCCodeEmitter::getHWRegChan(unsigned reg) const {
+  return MRI.getEncodingValue(reg) >> HW_CHAN_SHIFT;
+}
+
+unsigned R600MCCodeEmitter::getHWReg(unsigned RegNo) const {
+  return MRI.getEncodingValue(RegNo) & HW_REG_MASK;
+}
+
+uint64_t R600MCCodeEmitter::getMachineOpValue(const MCInst &MI,
+                                              const MCOperand &MO,
+                                        SmallVectorImpl<MCFixup> &Fixup) const {
+  if (MO.isReg()) {
+    if (HAS_NATIVE_OPERANDS(MCII.get(MI.getOpcode()).TSFlags)) {
+      return MRI.getEncodingValue(MO.getReg());
+    } else {
+      return getHWReg(MO.getReg());
+    }
+  } else if (MO.isImm()) {
+    return MO.getImm();
+  } else {
+    assert(0);
+    return 0;
+  }
+}
+
+//===----------------------------------------------------------------------===//
+// Encoding helper functions
+//===----------------------------------------------------------------------===//
+
+bool R600MCCodeEmitter::isFCOp(unsigned opcode) const {
+  switch(opcode) {
+  default: return false;
+  case AMDGPU::PREDICATED_BREAK:
+  case AMDGPU::CONTINUE:
+  case AMDGPU::IF_PREDICATE_SET:
+  case AMDGPU::ELSE:
+  case AMDGPU::ENDIF:
+  case AMDGPU::ENDLOOP:
+  case AMDGPU::WHILELOOP:
+    return true;
+  }
+}
+
+bool R600MCCodeEmitter::isTexOp(unsigned opcode) const {
+  switch(opcode) {
+  default: return false;
+  case AMDGPU::TEX_LD:
+  case AMDGPU::TEX_GET_TEXTURE_RESINFO:
+  case AMDGPU::TEX_SAMPLE:
+  case AMDGPU::TEX_SAMPLE_C:
+  case AMDGPU::TEX_SAMPLE_L:
+  case AMDGPU::TEX_SAMPLE_C_L:
+  case AMDGPU::TEX_SAMPLE_LB:
+  case AMDGPU::TEX_SAMPLE_C_LB:
+  case AMDGPU::TEX_SAMPLE_G:
+  case AMDGPU::TEX_SAMPLE_C_G:
+  case AMDGPU::TEX_GET_GRADIENTS_H:
+  case AMDGPU::TEX_GET_GRADIENTS_V:
+  case AMDGPU::TEX_SET_GRADIENTS_H:
+  case AMDGPU::TEX_SET_GRADIENTS_V:
+    return true;
+  }
+}
+
+bool R600MCCodeEmitter::isFlagSet(const MCInst &MI, unsigned Operand,
+                                  unsigned Flag) const {
+  const MCInstrDesc &MCDesc = MCII.get(MI.getOpcode());
+  unsigned FlagIndex = GET_FLAG_OPERAND_IDX(MCDesc.TSFlags);
+  if (FlagIndex == 0) {
+    return false;
+  }
+  assert(MI.getOperand(FlagIndex).isImm());
+  return !!((MI.getOperand(FlagIndex).getImm() >>
+            (NUM_MO_FLAGS * Operand)) & Flag);
+}
+
+#include "AMDGPUGenMCCodeEmitter.inc"
diff --git a/lib/Target/R600/MCTargetDesc/SIMCCodeEmitter.cpp b/lib/Target/R600/MCTargetDesc/SIMCCodeEmitter.cpp
new file mode 100644
index 000000000000..5af83209a0d5
--- /dev/null
+++ b/lib/Target/R600/MCTargetDesc/SIMCCodeEmitter.cpp
@@ -0,0 +1,201 @@
+//===-- SIMCCodeEmitter.cpp - SI Code Emitter -------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief The SI code emitter produces machine code that can be executed
+/// directly on the GPU device.
+//
+//===----------------------------------------------------------------------===//
+
+#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
+#include "MCTargetDesc/AMDGPUMCCodeEmitter.h"
+#include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCFixup.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+namespace {
+
+/// \brief Helper type used in encoding
+typedef union {
+  int32_t I;
+  float F;
+} IntFloatUnion;
+
+class SIMCCodeEmitter : public  AMDGPUMCCodeEmitter {
+  SIMCCodeEmitter(const SIMCCodeEmitter &) LLVM_DELETED_FUNCTION;
+  void operator=(const SIMCCodeEmitter &) LLVM_DELETED_FUNCTION;
+  const MCInstrInfo &MCII;
+  const MCRegisterInfo &MRI;
+
+  /// \brief Can this operand also contain immediate values?
+  bool isSrcOperand(const MCInstrDesc &Desc, unsigned OpNo) const;
+
+  /// \brief Encode an fp or int literal
+  uint32_t getLitEncoding(const MCOperand &MO) const;
+
+public:
+  SIMCCodeEmitter(const MCInstrInfo &mcii, const MCRegisterInfo &mri,
+                  const MCSubtargetInfo &sti, MCContext &ctx)
+    : MCII(mcii), MRI(mri) { }
+
+  ~SIMCCodeEmitter() { }
+
+  /// \breif Encode the instruction and write it to the OS.
+  virtual void EncodeInstruction(const MCInst &MI, raw_ostream &OS,
+                         SmallVectorImpl<MCFixup> &Fixups) const;
+
+  /// \returns the encoding for an MCOperand.
+  virtual uint64_t getMachineOpValue(const MCInst &MI, const MCOperand &MO,
+                                     SmallVectorImpl<MCFixup> &Fixups) const;
+};
+
+} // End anonymous namespace
+
+MCCodeEmitter *llvm::createSIMCCodeEmitter(const MCInstrInfo &MCII,
+                                           const MCRegisterInfo &MRI,
+                                           const MCSubtargetInfo &STI,
+                                           MCContext &Ctx) {
+  return new SIMCCodeEmitter(MCII, MRI, STI, Ctx);
+}
+
+bool SIMCCodeEmitter::isSrcOperand(const MCInstrDesc &Desc,
+                                   unsigned OpNo) const {
+
+  unsigned RegClass = Desc.OpInfo[OpNo].RegClass;
+  return (AMDGPU::SSrc_32RegClassID == RegClass) ||
+         (AMDGPU::SSrc_64RegClassID == RegClass) ||
+         (AMDGPU::VSrc_32RegClassID == RegClass) ||
+         (AMDGPU::VSrc_64RegClassID == RegClass);
+}
+
+uint32_t SIMCCodeEmitter::getLitEncoding(const MCOperand &MO) const {
+
+  IntFloatUnion Imm;
+  if (MO.isImm())
+    Imm.I = MO.getImm();
+  else if (MO.isFPImm())
+    Imm.F = MO.getFPImm();
+  else
+    return ~0;
+
+  if (Imm.I >= 0 && Imm.I <= 64)
+    return 128 + Imm.I;
+
+  if (Imm.I >= -16 && Imm.I <= -1)
+    return 192 + abs(Imm.I);
+
+  if (Imm.F == 0.5f)
+    return 240;
+
+  if (Imm.F == -0.5f)
+    return 241;
+
+  if (Imm.F == 1.0f)
+    return 242;
+
+  if (Imm.F == -1.0f)
+    return 243;
+
+  if (Imm.F == 2.0f)
+    return 244;
+
+  if (Imm.F == -2.0f)
+    return 245;
+
+  if (Imm.F == 4.0f)
+    return 246;
+
+  if (Imm.F == -4.0f)
+    return 247;
+
+  return 255;
+}
+
+void SIMCCodeEmitter::EncodeInstruction(const MCInst &MI, raw_ostream &OS,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+
+  uint64_t Encoding = getBinaryCodeForInstr(MI, Fixups);
+  const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
+  unsigned bytes = Desc.getSize();
+
+  for (unsigned i = 0; i < bytes; i++) {
+    OS.write((uint8_t) ((Encoding >> (8 * i)) & 0xff));
+  }
+
+  if (bytes > 4)
+    return;
+
+  // Check for additional literals in SRC0/1/2 (Op 1/2/3)
+  for (unsigned i = 0, e = MI.getNumOperands(); i < e; ++i) {
+
+    // Check if this operand should be encoded as [SV]Src
+    if (!isSrcOperand(Desc, i))
+      continue;
+
+    // Is this operand a literal immediate?
+    const MCOperand &Op = MI.getOperand(i);
+    if (getLitEncoding(Op) != 255)
+      continue;
+
+    // Yes! Encode it
+    IntFloatUnion Imm;
+    if (Op.isImm())
+      Imm.I = Op.getImm();
+    else
+      Imm.F = Op.getFPImm();
+
+    for (unsigned j = 0; j < 4; j++) {
+      OS.write((uint8_t) ((Imm.I >> (8 * j)) & 0xff));
+    }
+
+    // Only one literal value allowed
+    break;
+  }
+}
+
+uint64_t SIMCCodeEmitter::getMachineOpValue(const MCInst &MI,
+                                            const MCOperand &MO,
+                                       SmallVectorImpl<MCFixup> &Fixups) const {
+  if (MO.isReg())
+    return MRI.getEncodingValue(MO.getReg());
+
+  if (MO.isExpr()) {
+    const MCExpr *Expr = MO.getExpr();
+    MCFixupKind Kind = MCFixupKind(FK_PCRel_4);
+    Fixups.push_back(MCFixup::Create(0, Expr, Kind, MI.getLoc()));
+    return 0;
+  }
+
+  // Figure out the operand number, needed for isSrcOperand check
+  unsigned OpNo = 0;
+  for (unsigned e = MI.getNumOperands(); OpNo < e; ++OpNo) {
+    if (&MO == &MI.getOperand(OpNo))
+      break;
+  }
+
+  const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
+  if (isSrcOperand(Desc, OpNo)) {
+    uint32_t Enc = getLitEncoding(MO);
+    if (Enc != ~0U && (Enc != 255 || Desc.getSize() == 4))
+      return Enc;
+
+  } else if (MO.isImm())
+    return MO.getImm();
+
+  llvm_unreachable("Encoding of this operand type is not supported yet.");
+  return 0;
+}
+
diff --git a/lib/Target/R600/Makefile b/lib/Target/R600/Makefile
new file mode 100644
index 000000000000..1b3ebbe8c8f3
--- /dev/null
+++ b/lib/Target/R600/Makefile
@@ -0,0 +1,23 @@
+##===- lib/Target/R600/Makefile ---------------------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+
+LEVEL = ../../..
+LIBRARYNAME = LLVMR600CodeGen
+TARGET = AMDGPU
+
+# Make sure that tblgen is run, first thing.
+BUILT_SOURCES = AMDGPUGenRegisterInfo.inc AMDGPUGenInstrInfo.inc \
+		AMDGPUGenDAGISel.inc  AMDGPUGenSubtargetInfo.inc \
+		AMDGPUGenMCCodeEmitter.inc AMDGPUGenCallingConv.inc \
+		AMDGPUGenIntrinsics.inc AMDGPUGenDFAPacketizer.inc \
+		AMDGPUGenAsmWriter.inc
+
+DIRS = InstPrinter TargetInfo MCTargetDesc
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/Target/R600/Processors.td b/lib/Target/R600/Processors.td
new file mode 100644
index 000000000000..868810c613b3
--- /dev/null
+++ b/lib/Target/R600/Processors.td
@@ -0,0 +1,30 @@
+//===-- Processors.td - TODO: Add brief description -------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// AMDIL processors supported.
+//
+//===----------------------------------------------------------------------===//
+
+class Proc<string Name, ProcessorItineraries itin, list<SubtargetFeature> Features>
+: Processor<Name, itin, Features>;
+def : Proc<"",           R600_EG_Itin, [FeatureR600ALUInst]>;
+def : Proc<"r600",       R600_EG_Itin, [FeatureR600ALUInst]>;
+def : Proc<"rv710",      R600_EG_Itin, []>;
+def : Proc<"rv730",      R600_EG_Itin, []>;
+def : Proc<"rv770",      R600_EG_Itin, [FeatureFP64]>;
+def : Proc<"cedar",      R600_EG_Itin, [FeatureByteAddress, FeatureImages]>;
+def : Proc<"redwood",    R600_EG_Itin, [FeatureByteAddress, FeatureImages]>;
+def : Proc<"juniper",    R600_EG_Itin, [FeatureByteAddress, FeatureImages]>;
+def : Proc<"cypress",    R600_EG_Itin, [FeatureByteAddress, FeatureImages, FeatureFP64]>;
+def : Proc<"barts",      R600_EG_Itin, [FeatureByteAddress, FeatureImages]>;
+def : Proc<"turks",      R600_EG_Itin, [FeatureByteAddress, FeatureImages]>;
+def : Proc<"caicos",     R600_EG_Itin, [FeatureByteAddress, FeatureImages]>;
+def : Proc<"cayman",     R600_EG_Itin, [FeatureByteAddress, FeatureImages, FeatureFP64]>;
+def : Proc<"SI", SI_Itin, [Feature64BitPtr]>;
+
diff --git a/lib/Target/R600/R600ControlFlowFinalizer.cpp b/lib/Target/R600/R600ControlFlowFinalizer.cpp
new file mode 100644
index 000000000000..3a6c7eac730f
--- /dev/null
+++ b/lib/Target/R600/R600ControlFlowFinalizer.cpp
@@ -0,0 +1,268 @@
+//===-- R600ControlFlowFinalizer.cpp - Finalize Control Flow Inst----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// This pass compute turns all control flow pseudo instructions into native one
+/// computing their address on the fly ; it also sets STACK_SIZE info.
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "r600cf"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+
+#include "AMDGPU.h"
+#include "R600Defines.h"
+#include "R600InstrInfo.h"
+#include "R600MachineFunctionInfo.h"
+#include "R600RegisterInfo.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+
+namespace llvm {
+
+class R600ControlFlowFinalizer : public MachineFunctionPass {
+
+private:
+  static char ID;
+  const R600InstrInfo *TII;
+  unsigned MaxFetchInst;
+
+  bool isFetch(const MachineInstr *MI) const {
+    switch (MI->getOpcode()) {
+    case AMDGPU::TEX_VTX_CONSTBUF:
+    case AMDGPU::TEX_VTX_TEXBUF:
+    case AMDGPU::TEX_LD:
+    case AMDGPU::TEX_GET_TEXTURE_RESINFO:
+    case AMDGPU::TEX_GET_GRADIENTS_H:
+    case AMDGPU::TEX_GET_GRADIENTS_V:
+    case AMDGPU::TEX_SET_GRADIENTS_H:
+    case AMDGPU::TEX_SET_GRADIENTS_V:
+    case AMDGPU::TEX_SAMPLE:
+    case AMDGPU::TEX_SAMPLE_C:
+    case AMDGPU::TEX_SAMPLE_L:
+    case AMDGPU::TEX_SAMPLE_C_L:
+    case AMDGPU::TEX_SAMPLE_LB:
+    case AMDGPU::TEX_SAMPLE_C_LB:
+    case AMDGPU::TEX_SAMPLE_G:
+    case AMDGPU::TEX_SAMPLE_C_G:
+    case AMDGPU::TXD:
+    case AMDGPU::TXD_SHADOW:
+     return true;
+    default:
+      return false;
+    }
+  }
+
+  bool IsTrivialInst(MachineInstr *MI) const {
+    switch (MI->getOpcode()) {
+    case AMDGPU::KILL:
+    case AMDGPU::RETURN:
+      return true;
+    default:
+      return false;
+    }
+  }
+
+  MachineBasicBlock::iterator
+  MakeFetchClause(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
+      unsigned CfAddress) const {
+    MachineBasicBlock::iterator ClauseHead = I;
+    unsigned AluInstCount = 0;
+    for (MachineBasicBlock::iterator E = MBB.end(); I != E; ++I) {
+      if (IsTrivialInst(I))
+        continue;
+      if (!isFetch(I))
+        break;
+      AluInstCount ++;
+      if (AluInstCount > MaxFetchInst)
+        break;
+    }
+    BuildMI(MBB, ClauseHead, MBB.findDebugLoc(ClauseHead),
+        TII->get(AMDGPU::CF_TC))
+        .addImm(CfAddress) // ADDR
+        .addImm(AluInstCount); // COUNT
+    return I;
+  }
+  void CounterPropagateAddr(MachineInstr *MI, unsigned Addr) const {
+    MI->getOperand(0).setImm(Addr + MI->getOperand(0).getImm());
+  }
+  void CounterPropagateAddr(std::set<MachineInstr *> MIs, unsigned Addr)
+      const {
+    for (std::set<MachineInstr *>::iterator It = MIs.begin(), E = MIs.end();
+        It != E; ++It) {
+      MachineInstr *MI = *It;
+      CounterPropagateAddr(MI, Addr);
+    }
+  }
+
+public:
+  R600ControlFlowFinalizer(TargetMachine &tm) : MachineFunctionPass(ID),
+    TII (static_cast<const R600InstrInfo *>(tm.getInstrInfo())) {
+      const AMDGPUSubtarget &ST = tm.getSubtarget<AMDGPUSubtarget>();
+      if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD4XXX)
+        MaxFetchInst = 8;
+      else
+        MaxFetchInst = 16;
+  }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF) {
+    unsigned MaxStack = 0;
+    unsigned CurrentStack = 0;
+    for (MachineFunction::iterator MB = MF.begin(), ME = MF.end(); MB != ME;
+        ++MB) {
+      MachineBasicBlock &MBB = *MB;
+      unsigned CfCount = 0;
+      std::vector<std::pair<unsigned, std::set<MachineInstr *> > > LoopStack;
+      std::vector<MachineInstr * > IfThenElseStack;
+      R600MachineFunctionInfo *MFI = MF.getInfo<R600MachineFunctionInfo>();
+      if (MFI->ShaderType == 1) {
+        BuildMI(MBB, MBB.begin(), MBB.findDebugLoc(MBB.begin()),
+            TII->get(AMDGPU::CF_CALL_FS));
+        CfCount++;
+      }
+      for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
+          I != E;) {
+        if (isFetch(I)) {
+          DEBUG(dbgs() << CfCount << ":"; I->dump(););
+          I = MakeFetchClause(MBB, I, 0);
+          CfCount++;
+          continue;
+        }
+
+        MachineBasicBlock::iterator MI = I;
+        I++;
+        switch (MI->getOpcode()) {
+        case AMDGPU::CF_ALU_PUSH_BEFORE:
+          CurrentStack++;
+          MaxStack = std::max(MaxStack, CurrentStack);
+        case AMDGPU::CF_ALU:
+        case AMDGPU::EG_ExportBuf:
+        case AMDGPU::EG_ExportSwz:
+        case AMDGPU::R600_ExportBuf:
+        case AMDGPU::R600_ExportSwz:
+          DEBUG(dbgs() << CfCount << ":"; MI->dump(););
+          CfCount++;
+          break;
+        case AMDGPU::WHILELOOP: {
+          CurrentStack++;
+          MaxStack = std::max(MaxStack, CurrentStack);
+          MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
+              TII->get(AMDGPU::WHILE_LOOP))
+              .addImm(2);
+          std::pair<unsigned, std::set<MachineInstr *> > Pair(CfCount,
+              std::set<MachineInstr *>());
+          Pair.second.insert(MIb);
+          LoopStack.push_back(Pair);
+          MI->eraseFromParent();
+          CfCount++;
+          break;
+        }
+        case AMDGPU::ENDLOOP: {
+          CurrentStack--;
+          std::pair<unsigned, std::set<MachineInstr *> > Pair =
+              LoopStack.back();
+          LoopStack.pop_back();
+          CounterPropagateAddr(Pair.second, CfCount);
+          BuildMI(MBB, MI, MBB.findDebugLoc(MI), TII->get(AMDGPU::END_LOOP))
+              .addImm(Pair.first + 1);
+          MI->eraseFromParent();
+          CfCount++;
+          break;
+        }
+        case AMDGPU::IF_PREDICATE_SET: {
+          MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
+              TII->get(AMDGPU::CF_JUMP))
+              .addImm(0)
+              .addImm(0);
+          IfThenElseStack.push_back(MIb);
+          DEBUG(dbgs() << CfCount << ":"; MIb->dump(););
+          MI->eraseFromParent();
+          CfCount++;
+          break;
+        }
+        case AMDGPU::ELSE: {
+          MachineInstr * JumpInst = IfThenElseStack.back();
+          IfThenElseStack.pop_back();
+          CounterPropagateAddr(JumpInst, CfCount);
+          MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
+              TII->get(AMDGPU::CF_ELSE))
+              .addImm(0)
+              .addImm(1);
+          DEBUG(dbgs() << CfCount << ":"; MIb->dump(););
+          IfThenElseStack.push_back(MIb);
+          MI->eraseFromParent();
+          CfCount++;
+          break;
+        }
+        case AMDGPU::ENDIF: {
+          CurrentStack--;
+          MachineInstr *IfOrElseInst = IfThenElseStack.back();
+          IfThenElseStack.pop_back();
+          CounterPropagateAddr(IfOrElseInst, CfCount + 1);
+          MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
+              TII->get(AMDGPU::POP))
+              .addImm(CfCount + 1)
+              .addImm(1);
+          DEBUG(dbgs() << CfCount << ":"; MIb->dump(););
+          MI->eraseFromParent();
+          CfCount++;
+          break;
+        }
+        case AMDGPU::PREDICATED_BREAK: {
+          CurrentStack--;
+          CfCount += 3;
+          BuildMI(MBB, MI, MBB.findDebugLoc(MI), TII->get(AMDGPU::CF_JUMP))
+              .addImm(CfCount)
+              .addImm(1);
+          MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
+              TII->get(AMDGPU::LOOP_BREAK))
+              .addImm(0);
+          BuildMI(MBB, MI, MBB.findDebugLoc(MI), TII->get(AMDGPU::POP))
+              .addImm(CfCount)
+              .addImm(1);
+          LoopStack.back().second.insert(MIb);
+          MI->eraseFromParent();
+          break;
+        }
+        case AMDGPU::CONTINUE: {
+          MachineInstr *MIb = BuildMI(MBB, MI, MBB.findDebugLoc(MI),
+              TII->get(AMDGPU::CF_CONTINUE))
+              .addImm(0);
+          LoopStack.back().second.insert(MIb);
+          MI->eraseFromParent();
+          CfCount++;
+          break;
+        }
+        default:
+          break;
+        }
+      }
+      BuildMI(MBB, MBB.begin(), MBB.findDebugLoc(MBB.begin()),
+          TII->get(AMDGPU::STACK_SIZE))
+          .addImm(MaxStack);
+    }
+
+    return false;
+  }
+
+  const char *getPassName() const {
+    return "R600 Control Flow Finalizer Pass";
+  }
+};
+
+char R600ControlFlowFinalizer::ID = 0;
+
+}
+
+
+llvm::FunctionPass *llvm::createR600ControlFlowFinalizer(TargetMachine &TM) {
+  return new R600ControlFlowFinalizer(TM);
+}
+
diff --git a/lib/Target/R600/R600Defines.h b/lib/Target/R600/R600Defines.h
new file mode 100644
index 000000000000..16cfcf59eb3d
--- /dev/null
+++ b/lib/Target/R600/R600Defines.h
@@ -0,0 +1,97 @@
+//===-- R600Defines.h - R600 Helper Macros ----------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//===----------------------------------------------------------------------===//
+
+#ifndef R600DEFINES_H_
+#define R600DEFINES_H_
+
+#include "llvm/MC/MCRegisterInfo.h"
+
+// Operand Flags
+#define MO_FLAG_CLAMP (1 << 0)
+#define MO_FLAG_NEG   (1 << 1)
+#define MO_FLAG_ABS   (1 << 2)
+#define MO_FLAG_MASK  (1 << 3)
+#define MO_FLAG_PUSH  (1 << 4)
+#define MO_FLAG_NOT_LAST  (1 << 5)
+#define MO_FLAG_LAST  (1 << 6)
+#define NUM_MO_FLAGS 7
+
+/// \brief Helper for getting the operand index for the instruction flags
+/// operand.
+#define GET_FLAG_OPERAND_IDX(Flags) (((Flags) >> 7) & 0x3)
+
+namespace R600_InstFlag {
+  enum TIF {
+    TRANS_ONLY = (1 << 0),
+    TEX = (1 << 1),
+    REDUCTION = (1 << 2),
+    FC = (1 << 3),
+    TRIG = (1 << 4),
+    OP3 = (1 << 5),
+    VECTOR = (1 << 6),
+    //FlagOperand bits 7, 8
+    NATIVE_OPERANDS = (1 << 9),
+    OP1 = (1 << 10),
+    OP2 = (1 << 11)
+  };
+}
+
+#define HAS_NATIVE_OPERANDS(Flags) ((Flags) & R600_InstFlag::NATIVE_OPERANDS)
+
+/// \brief Defines for extracting register infomation from register encoding
+#define HW_REG_MASK 0x1ff
+#define HW_CHAN_SHIFT 9
+
+#define GET_REG_CHAN(reg) ((reg) >> HW_CHAN_SHIFT)
+#define GET_REG_INDEX(reg) ((reg) & HW_REG_MASK)
+
+namespace R600Operands {
+  enum Ops {
+    DST,
+    UPDATE_EXEC_MASK,
+    UPDATE_PREDICATE,
+    WRITE,
+    OMOD,
+    DST_REL,
+    CLAMP,
+    SRC0,
+    SRC0_NEG,
+    SRC0_REL,
+    SRC0_ABS,
+    SRC0_SEL,
+    SRC1,
+    SRC1_NEG,
+    SRC1_REL,
+    SRC1_ABS,
+    SRC1_SEL,
+    SRC2,
+    SRC2_NEG,
+    SRC2_REL,
+    SRC2_SEL,
+    LAST,
+    PRED_SEL,
+    IMM,
+    COUNT
+ };
+
+  const static int ALUOpTable[3][R600Operands::COUNT] = {
+//            W        C     S  S  S  S     S  S  S  S     S  S  S
+//            R  O  D  L  S  R  R  R  R  S  R  R  R  R  S  R  R  R  L  P
+//   D  U     I  M  R  A  R  C  C  C  C  R  C  C  C  C  R  C  C  C  A  R  I
+//   S  E  U  T  O  E  M  C  0  0  0  0  C  1  1  1  1  C  2  2  2  S  E  M
+//   T  M  P  E  D  L  P  0  N  R  A  S  1  N  R  A  S  2  N  R  S  T  D  M
+    {0,-1,-1, 1, 2, 3, 4, 5, 6, 7, 8, 9,-1,-1,-1,-1,-1,-1,-1,-1,-1,10,11,12},
+    {0, 1, 2, 3, 4 ,5 ,6 ,7, 8, 9,10,11,12,13,14,15,16,-1,-1,-1,-1,17,18,19},
+    {0,-1,-1,-1,-1, 1, 2, 3, 4, 5,-1, 6, 7, 8, 9,-1,10,11,12,13,14,15,16,17}
+  };
+
+}
+
+#endif // R600DEFINES_H_
diff --git a/lib/Target/R600/R600EmitClauseMarkers.cpp b/lib/Target/R600/R600EmitClauseMarkers.cpp
new file mode 100644
index 000000000000..3fdc678b9ef1
--- /dev/null
+++ b/lib/Target/R600/R600EmitClauseMarkers.cpp
@@ -0,0 +1,255 @@
+//===-- R600EmitClauseMarkers.cpp - Emit CF_ALU ---------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// Add CF_ALU. R600 Alu instructions are grouped in clause which can hold
+/// 128 Alu instructions ; these instructions can access up to 4 prefetched
+/// 4 lines of 16 registers from constant buffers. Such ALU clauses are
+/// initiated by CF_ALU instructions.
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "R600Defines.h"
+#include "R600InstrInfo.h"
+#include "R600MachineFunctionInfo.h"
+#include "R600RegisterInfo.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+
+namespace llvm {
+
+class R600EmitClauseMarkersPass : public MachineFunctionPass {
+
+private:
+  static char ID;
+  const R600InstrInfo *TII;
+
+  unsigned OccupiedDwords(MachineInstr *MI) const {
+    switch (MI->getOpcode()) {
+    case AMDGPU::INTERP_PAIR_XY:
+    case AMDGPU::INTERP_PAIR_ZW:
+    case AMDGPU::INTERP_VEC_LOAD:
+    case AMDGPU::DOT4_eg_pseudo:
+    case AMDGPU::DOT4_r600_pseudo:
+      return 4;
+    case AMDGPU::KILL:
+      return 0;
+    default:
+      break;
+    }
+
+    if(TII->isVector(*MI) ||
+        TII->isCubeOp(MI->getOpcode()) ||
+        TII->isReductionOp(MI->getOpcode()))
+      return 4;
+
+    unsigned NumLiteral = 0;
+    for (MachineInstr::mop_iterator It = MI->operands_begin(),
+        E = MI->operands_end(); It != E; ++It) {
+      MachineOperand &MO = *It;
+      if (MO.isReg() && MO.getReg() == AMDGPU::ALU_LITERAL_X)
+        ++NumLiteral;
+    }
+    return 1 + NumLiteral;
+  }
+
+  bool isALU(const MachineInstr *MI) const {
+    if (TII->isALUInstr(MI->getOpcode()))
+      return true;
+    if (TII->isVector(*MI) || TII->isCubeOp(MI->getOpcode()))
+      return true;
+    switch (MI->getOpcode()) {
+    case AMDGPU::PRED_X:
+    case AMDGPU::INTERP_PAIR_XY:
+    case AMDGPU::INTERP_PAIR_ZW:
+    case AMDGPU::INTERP_VEC_LOAD:
+    case AMDGPU::COPY:
+    case AMDGPU::DOT4_eg_pseudo:
+    case AMDGPU::DOT4_r600_pseudo:
+      return true;
+    default:
+      return false;
+    }
+  }
+
+  bool IsTrivialInst(MachineInstr *MI) const {
+    switch (MI->getOpcode()) {
+    case AMDGPU::KILL:
+    case AMDGPU::RETURN:
+      return true;
+    default:
+      return false;
+    }
+  }
+
+  // Register Idx, then Const value
+  std::vector<std::pair<unsigned, unsigned> > ExtractConstRead(MachineInstr *MI)
+      const {
+    const R600Operands::Ops OpTable[3][2] = {
+      {R600Operands::SRC0, R600Operands::SRC0_SEL},
+      {R600Operands::SRC1, R600Operands::SRC1_SEL},
+      {R600Operands::SRC2, R600Operands::SRC2_SEL},
+    };
+    std::vector<std::pair<unsigned, unsigned> > Result;
+
+    if (!TII->isALUInstr(MI->getOpcode()))
+      return Result;
+    for (unsigned j = 0; j < 3; j++) {
+      int SrcIdx = TII->getOperandIdx(MI->getOpcode(), OpTable[j][0]);
+      if (SrcIdx < 0)
+        break;
+      if (MI->getOperand(SrcIdx).getReg() == AMDGPU::ALU_CONST) {
+        unsigned Const = MI->getOperand(
+            TII->getOperandIdx(MI->getOpcode(), OpTable[j][1])).getImm();
+        Result.push_back(std::pair<unsigned, unsigned>(SrcIdx, Const));
+      }
+    }
+    return Result;
+  }
+
+  std::pair<unsigned, unsigned> getAccessedBankLine(unsigned Sel) const {
+    // Sel is (512 + (kc_bank << 12) + ConstIndex) << 2
+    // (See also R600ISelLowering.cpp)
+    // ConstIndex value is in [0, 4095];
+    return std::pair<unsigned, unsigned>(
+        ((Sel >> 2) - 512) >> 12, // KC_BANK
+        // Line Number of ConstIndex
+        // A line contains 16 constant registers however KCX bank can lock
+        // two line at the same time ; thus we want to get an even line number.
+        // Line number can be retrieved with (>>4), using (>>5) <<1 generates
+        // an even number.
+        ((((Sel >> 2) - 512) & 4095) >> 5) << 1);
+  }
+
+  bool SubstituteKCacheBank(MachineInstr *MI,
+      std::vector<std::pair<unsigned, unsigned> > &CachedConsts) const {
+    std::vector<std::pair<unsigned, unsigned> > UsedKCache;
+    std::vector<std::pair<unsigned, unsigned> > Consts = ExtractConstRead(MI);
+    assert(TII->isALUInstr(MI->getOpcode()) && "Can't assign Const");
+    for (unsigned i = 0, n = Consts.size(); i < n; ++i) {
+      unsigned Sel = Consts[i].second;
+      unsigned Chan = Sel & 3, Index = ((Sel >> 2) - 512) & 31;
+      unsigned KCacheIndex = Index * 4 + Chan;
+      const std::pair<unsigned, unsigned> &BankLine = getAccessedBankLine(Sel);
+      if (CachedConsts.empty()) {
+        CachedConsts.push_back(BankLine);
+        UsedKCache.push_back(std::pair<unsigned, unsigned>(0, KCacheIndex));
+        continue;
+      }
+      if (CachedConsts[0] == BankLine) {
+        UsedKCache.push_back(std::pair<unsigned, unsigned>(0, KCacheIndex));
+        continue;
+      }
+      if (CachedConsts.size() == 1) {
+        CachedConsts.push_back(BankLine);
+        UsedKCache.push_back(std::pair<unsigned, unsigned>(1, KCacheIndex));
+        continue;
+      }
+      if (CachedConsts[1] == BankLine) {
+        UsedKCache.push_back(std::pair<unsigned, unsigned>(1, KCacheIndex));
+        continue;
+      }
+      return false;
+    }
+
+    for (unsigned i = 0, n = Consts.size(); i < n; ++i) {
+      switch(UsedKCache[i].first) {
+      case 0:
+        MI->getOperand(Consts[i].first).setReg(
+            AMDGPU::R600_KC0RegClass.getRegister(UsedKCache[i].second));
+        break;
+      case 1:
+        MI->getOperand(Consts[i].first).setReg(
+            AMDGPU::R600_KC1RegClass.getRegister(UsedKCache[i].second));
+        break;
+      default:
+        llvm_unreachable("Wrong Cache Line");
+      }
+    }
+    return true;
+  }
+
+  MachineBasicBlock::iterator
+  MakeALUClause(MachineBasicBlock &MBB, MachineBasicBlock::iterator I) const {
+    MachineBasicBlock::iterator ClauseHead = I;
+    std::vector<std::pair<unsigned, unsigned> > KCacheBanks;
+    bool PushBeforeModifier = false;
+    unsigned AluInstCount = 0;
+    for (MachineBasicBlock::iterator E = MBB.end(); I != E; ++I) {
+      if (IsTrivialInst(I))
+        continue;
+      if (!isALU(I))
+        break;
+      if (AluInstCount > TII->getMaxAlusPerClause())
+        break;
+      if (I->getOpcode() == AMDGPU::PRED_X) {
+        if (TII->getFlagOp(I).getImm() & MO_FLAG_PUSH)
+          PushBeforeModifier = true;
+        AluInstCount ++;
+        continue;
+      }
+      if (I->getOpcode() == AMDGPU::KILLGT) {
+        I++;
+        break;
+      }
+      if (TII->isALUInstr(I->getOpcode()) &&
+          !SubstituteKCacheBank(I, KCacheBanks))
+        break;
+      AluInstCount += OccupiedDwords(I);
+    }
+    unsigned Opcode = PushBeforeModifier ?
+        AMDGPU::CF_ALU_PUSH_BEFORE : AMDGPU::CF_ALU;
+    BuildMI(MBB, ClauseHead, MBB.findDebugLoc(ClauseHead), TII->get(Opcode))
+        .addImm(0) // ADDR
+        .addImm(KCacheBanks.empty()?0:KCacheBanks[0].first) // KB0
+        .addImm((KCacheBanks.size() < 2)?0:KCacheBanks[1].first) // KB1
+        .addImm(KCacheBanks.empty()?0:2) // KM0
+        .addImm((KCacheBanks.size() < 2)?0:2) // KM1
+        .addImm(KCacheBanks.empty()?0:KCacheBanks[0].second) // KLINE0
+        .addImm((KCacheBanks.size() < 2)?0:KCacheBanks[1].second) // KLINE1
+        .addImm(AluInstCount); // COUNT
+    return I;
+  }
+
+public:
+  R600EmitClauseMarkersPass(TargetMachine &tm) : MachineFunctionPass(ID),
+    TII (static_cast<const R600InstrInfo *>(tm.getInstrInfo())) { }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF) {
+    for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
+                                                    BB != BB_E; ++BB) {
+      MachineBasicBlock &MBB = *BB;
+      MachineBasicBlock::iterator I = MBB.begin();
+      if (I->getOpcode() == AMDGPU::CF_ALU)
+        continue; // BB was already parsed
+      for (MachineBasicBlock::iterator E = MBB.end(); I != E;) {
+        if (isALU(I))
+          I = MakeALUClause(MBB, I);
+        else
+          ++I;
+      }
+    }
+    return false;
+  }
+
+  const char *getPassName() const {
+    return "R600 Emit Clause Markers Pass";
+  }
+};
+
+char R600EmitClauseMarkersPass::ID = 0;
+
+}
+
+
+llvm::FunctionPass *llvm::createR600EmitClauseMarkers(TargetMachine &TM) {
+  return new R600EmitClauseMarkersPass(TM);
+}
+
diff --git a/lib/Target/R600/R600ExpandSpecialInstrs.cpp b/lib/Target/R600/R600ExpandSpecialInstrs.cpp
new file mode 100644
index 000000000000..f8c900f72776
--- /dev/null
+++ b/lib/Target/R600/R600ExpandSpecialInstrs.cpp
@@ -0,0 +1,297 @@
+//===-- R600ExpandSpecialInstrs.cpp - Expand special instructions ---------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// Vector, Reduction, and Cube instructions need to fill the entire instruction
+/// group to work correctly.  This pass expands these individual instructions
+/// into several instructions that will completely fill the instruction group.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "R600Defines.h"
+#include "R600InstrInfo.h"
+#include "R600MachineFunctionInfo.h"
+#include "R600RegisterInfo.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+
+using namespace llvm;
+
+namespace {
+
+class R600ExpandSpecialInstrsPass : public MachineFunctionPass {
+
+private:
+  static char ID;
+  const R600InstrInfo *TII;
+
+  bool ExpandInputPerspective(MachineInstr& MI);
+  bool ExpandInputConstant(MachineInstr& MI);
+
+public:
+  R600ExpandSpecialInstrsPass(TargetMachine &tm) : MachineFunctionPass(ID),
+    TII (static_cast<const R600InstrInfo *>(tm.getInstrInfo())) { }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF);
+
+  const char *getPassName() const {
+    return "R600 Expand special instructions pass";
+  }
+};
+
+} // End anonymous namespace
+
+char R600ExpandSpecialInstrsPass::ID = 0;
+
+FunctionPass *llvm::createR600ExpandSpecialInstrsPass(TargetMachine &TM) {
+  return new R600ExpandSpecialInstrsPass(TM);
+}
+
+bool R600ExpandSpecialInstrsPass::runOnMachineFunction(MachineFunction &MF) {
+
+  const R600RegisterInfo &TRI = TII->getRegisterInfo();
+
+  for (MachineFunction::iterator BB = MF.begin(), BB_E = MF.end();
+                                                  BB != BB_E; ++BB) {
+    MachineBasicBlock &MBB = *BB;
+    MachineBasicBlock::iterator I = MBB.begin();
+    while (I != MBB.end()) {
+      MachineInstr &MI = *I;
+      I = llvm::next(I);
+
+      switch (MI.getOpcode()) {
+      default: break;
+      // Expand PRED_X to one of the PRED_SET instructions.
+      case AMDGPU::PRED_X: {
+        uint64_t Flags = MI.getOperand(3).getImm();
+        // The native opcode used by PRED_X is stored as an immediate in the
+        // third operand.
+        MachineInstr *PredSet = TII->buildDefaultInstruction(MBB, I,
+                                            MI.getOperand(2).getImm(), // opcode
+                                            MI.getOperand(0).getReg(), // dst
+                                            MI.getOperand(1).getReg(), // src0
+                                            AMDGPU::ZERO);             // src1
+        TII->addFlag(PredSet, 0, MO_FLAG_MASK);
+        if (Flags & MO_FLAG_PUSH) {
+          TII->setImmOperand(PredSet, R600Operands::UPDATE_EXEC_MASK, 1);
+        } else {
+          TII->setImmOperand(PredSet, R600Operands::UPDATE_PREDICATE, 1);
+        }
+        MI.eraseFromParent();
+        continue;
+        }
+      case AMDGPU::BREAK: {
+        MachineInstr *PredSet = TII->buildDefaultInstruction(MBB, I,
+                                          AMDGPU::PRED_SETE_INT,
+                                          AMDGPU::PREDICATE_BIT,
+                                          AMDGPU::ZERO,
+                                          AMDGPU::ZERO);
+        TII->addFlag(PredSet, 0, MO_FLAG_MASK);
+        TII->setImmOperand(PredSet, R600Operands::UPDATE_EXEC_MASK, 1);
+
+        BuildMI(MBB, I, MBB.findDebugLoc(I),
+                TII->get(AMDGPU::PREDICATED_BREAK))
+                .addReg(AMDGPU::PREDICATE_BIT);
+        MI.eraseFromParent();
+        continue;
+        }
+
+      case AMDGPU::INTERP_PAIR_XY: {
+        MachineInstr *BMI;
+        unsigned PReg = AMDGPU::R600_ArrayBaseRegClass.getRegister(
+                MI.getOperand(2).getImm());
+
+        for (unsigned Chan = 0; Chan < 4; ++Chan) {
+          unsigned DstReg;
+
+          if (Chan < 2)
+            DstReg = MI.getOperand(Chan).getReg();
+          else
+            DstReg = Chan == 2 ? AMDGPU::T0_Z : AMDGPU::T0_W;
+
+          BMI = TII->buildDefaultInstruction(MBB, I, AMDGPU::INTERP_XY,
+              DstReg, MI.getOperand(3 + (Chan % 2)).getReg(), PReg);
+
+          if (Chan > 0) {
+            BMI->bundleWithPred();
+          }
+          if (Chan >= 2)
+            TII->addFlag(BMI, 0, MO_FLAG_MASK);
+          if (Chan != 3)
+            TII->addFlag(BMI, 0, MO_FLAG_NOT_LAST);
+        }
+
+        MI.eraseFromParent();
+        continue;
+        }
+
+      case AMDGPU::INTERP_PAIR_ZW: {
+        MachineInstr *BMI;
+        unsigned PReg = AMDGPU::R600_ArrayBaseRegClass.getRegister(
+                MI.getOperand(2).getImm());
+
+        for (unsigned Chan = 0; Chan < 4; ++Chan) {
+          unsigned DstReg;
+
+          if (Chan < 2)
+            DstReg = Chan == 0 ? AMDGPU::T0_X : AMDGPU::T0_Y;
+          else
+            DstReg = MI.getOperand(Chan-2).getReg();
+
+          BMI = TII->buildDefaultInstruction(MBB, I, AMDGPU::INTERP_ZW,
+              DstReg, MI.getOperand(3 + (Chan % 2)).getReg(), PReg);
+
+          if (Chan > 0) {
+            BMI->bundleWithPred();
+          }
+          if (Chan < 2)
+            TII->addFlag(BMI, 0, MO_FLAG_MASK);
+          if (Chan != 3)
+            TII->addFlag(BMI, 0, MO_FLAG_NOT_LAST);
+        }
+
+        MI.eraseFromParent();
+        continue;
+        }
+
+      case AMDGPU::INTERP_VEC_LOAD: {
+        const R600RegisterInfo &TRI = TII->getRegisterInfo();
+        MachineInstr *BMI;
+        unsigned PReg = AMDGPU::R600_ArrayBaseRegClass.getRegister(
+                MI.getOperand(1).getImm());
+        unsigned DstReg = MI.getOperand(0).getReg();
+
+        for (unsigned Chan = 0; Chan < 4; ++Chan) {
+          BMI = TII->buildDefaultInstruction(MBB, I, AMDGPU::INTERP_LOAD_P0,
+              TRI.getSubReg(DstReg, TRI.getSubRegFromChannel(Chan)), PReg);
+          if (Chan > 0) {
+            BMI->bundleWithPred();
+          }
+          if (Chan != 3)
+            TII->addFlag(BMI, 0, MO_FLAG_NOT_LAST);
+        }
+
+        MI.eraseFromParent();
+        continue;
+        }
+      }
+
+      bool IsReduction = TII->isReductionOp(MI.getOpcode());
+      bool IsVector = TII->isVector(MI);
+      bool IsCube = TII->isCubeOp(MI.getOpcode());
+      if (!IsReduction && !IsVector && !IsCube) {
+        continue;
+      }
+
+      // Expand the instruction
+      //
+      // Reduction instructions:
+      // T0_X = DP4 T1_XYZW, T2_XYZW
+      // becomes:
+      // TO_X = DP4 T1_X, T2_X
+      // TO_Y (write masked) = DP4 T1_Y, T2_Y
+      // TO_Z (write masked) = DP4 T1_Z, T2_Z
+      // TO_W (write masked) = DP4 T1_W, T2_W
+      //
+      // Vector instructions:
+      // T0_X = MULLO_INT T1_X, T2_X
+      // becomes:
+      // T0_X = MULLO_INT T1_X, T2_X
+      // T0_Y (write masked) = MULLO_INT T1_X, T2_X
+      // T0_Z (write masked) = MULLO_INT T1_X, T2_X
+      // T0_W (write masked) = MULLO_INT T1_X, T2_X
+      //
+      // Cube instructions:
+      // T0_XYZW = CUBE T1_XYZW
+      // becomes:
+      // TO_X = CUBE T1_Z, T1_Y
+      // T0_Y = CUBE T1_Z, T1_X
+      // T0_Z = CUBE T1_X, T1_Z
+      // T0_W = CUBE T1_Y, T1_Z
+      for (unsigned Chan = 0; Chan < 4; Chan++) {
+        unsigned DstReg = MI.getOperand(
+                            TII->getOperandIdx(MI, R600Operands::DST)).getReg();
+        unsigned Src0 = MI.getOperand(
+                           TII->getOperandIdx(MI, R600Operands::SRC0)).getReg();
+        unsigned Src1 = 0;
+
+        // Determine the correct source registers
+        if (!IsCube) {
+          int Src1Idx = TII->getOperandIdx(MI, R600Operands::SRC1);
+          if (Src1Idx != -1) {
+            Src1 = MI.getOperand(Src1Idx).getReg();
+          }
+        }
+        if (IsReduction) {
+          unsigned SubRegIndex = TRI.getSubRegFromChannel(Chan);
+          Src0 = TRI.getSubReg(Src0, SubRegIndex);
+          Src1 = TRI.getSubReg(Src1, SubRegIndex);
+        } else if (IsCube) {
+          static const int CubeSrcSwz[] = {2, 2, 0, 1};
+          unsigned SubRegIndex0 = TRI.getSubRegFromChannel(CubeSrcSwz[Chan]);
+          unsigned SubRegIndex1 = TRI.getSubRegFromChannel(CubeSrcSwz[3 - Chan]);
+          Src1 = TRI.getSubReg(Src0, SubRegIndex1);
+          Src0 = TRI.getSubReg(Src0, SubRegIndex0);
+        }
+
+        // Determine the correct destination registers;
+        bool Mask = false;
+        bool NotLast = true;
+        if (IsCube) {
+          unsigned SubRegIndex = TRI.getSubRegFromChannel(Chan);
+          DstReg = TRI.getSubReg(DstReg, SubRegIndex);
+        } else {
+          // Mask the write if the original instruction does not write to
+          // the current Channel.
+          Mask = (Chan != TRI.getHWRegChan(DstReg));
+          unsigned DstBase = TRI.getEncodingValue(DstReg) & HW_REG_MASK;
+          DstReg = AMDGPU::R600_TReg32RegClass.getRegister((DstBase * 4) + Chan);
+        }
+
+        // Set the IsLast bit
+        NotLast = (Chan != 3 );
+
+        // Add the new instruction
+        unsigned Opcode = MI.getOpcode();
+        switch (Opcode) {
+        case AMDGPU::CUBE_r600_pseudo:
+          Opcode = AMDGPU::CUBE_r600_real;
+          break;
+        case AMDGPU::CUBE_eg_pseudo:
+          Opcode = AMDGPU::CUBE_eg_real;
+          break;
+        case AMDGPU::DOT4_r600_pseudo:
+          Opcode = AMDGPU::DOT4_r600_real;
+          break;
+        case AMDGPU::DOT4_eg_pseudo:
+          Opcode = AMDGPU::DOT4_eg_real;
+          break;
+        default:
+          break;
+        }
+
+        MachineInstr *NewMI =
+          TII->buildDefaultInstruction(MBB, I, Opcode, DstReg, Src0, Src1);
+
+        if (Chan != 0)
+          NewMI->bundleWithPred();
+        if (Mask) {
+          TII->addFlag(NewMI, 0, MO_FLAG_MASK);
+        }
+        if (NotLast) {
+          TII->addFlag(NewMI, 0, MO_FLAG_NOT_LAST);
+        }
+      }
+      MI.eraseFromParent();
+    }
+  }
+  return false;
+}
diff --git a/lib/Target/R600/R600ISelLowering.cpp b/lib/Target/R600/R600ISelLowering.cpp
new file mode 100644
index 000000000000..53e6e51dd2b1
--- /dev/null
+++ b/lib/Target/R600/R600ISelLowering.cpp
@@ -0,0 +1,1106 @@
+//===-- R600ISelLowering.cpp - R600 DAG Lowering Implementation -----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Custom DAG lowering for R600
+//
+//===----------------------------------------------------------------------===//
+
+#include "R600ISelLowering.h"
+#include "R600Defines.h"
+#include "R600InstrInfo.h"
+#include "R600MachineFunctionInfo.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/IR/Argument.h"
+#include "llvm/IR/Function.h"
+
+using namespace llvm;
+
+R600TargetLowering::R600TargetLowering(TargetMachine &TM) :
+    AMDGPUTargetLowering(TM),
+    TII(static_cast<const R600InstrInfo*>(TM.getInstrInfo())) {
+  addRegisterClass(MVT::v4f32, &AMDGPU::R600_Reg128RegClass);
+  addRegisterClass(MVT::f32, &AMDGPU::R600_Reg32RegClass);
+  addRegisterClass(MVT::v4i32, &AMDGPU::R600_Reg128RegClass);
+  addRegisterClass(MVT::i32, &AMDGPU::R600_Reg32RegClass);
+  computeRegisterProperties();
+
+  setOperationAction(ISD::FADD, MVT::v4f32, Expand);
+  setOperationAction(ISD::FMUL, MVT::v4f32, Expand);
+  setOperationAction(ISD::FDIV, MVT::v4f32, Expand);
+  setOperationAction(ISD::FSUB, MVT::v4f32, Expand);
+
+  setOperationAction(ISD::ADD,  MVT::v4i32, Expand);
+  setOperationAction(ISD::AND,  MVT::v4i32, Expand);
+  setOperationAction(ISD::FP_TO_SINT, MVT::v4i32, Expand);
+  setOperationAction(ISD::FP_TO_UINT, MVT::v4i32, Expand);
+  setOperationAction(ISD::SINT_TO_FP, MVT::v4i32, Expand);
+  setOperationAction(ISD::UINT_TO_FP, MVT::v4i32, Expand);
+  setOperationAction(ISD::UDIV, MVT::v4i32, Expand);
+  setOperationAction(ISD::UREM, MVT::v4i32, Expand);
+  setOperationAction(ISD::SETCC, MVT::v4i32, Expand);
+
+  setOperationAction(ISD::BR_CC, MVT::i32, Expand);
+  setOperationAction(ISD::BR_CC, MVT::f32, Expand);
+
+  setOperationAction(ISD::FSUB, MVT::f32, Expand);
+
+  setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
+  setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
+  setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::i1, Custom);
+
+  setOperationAction(ISD::ROTL, MVT::i32, Custom);
+
+  setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
+  setOperationAction(ISD::SELECT_CC, MVT::i32, Custom);
+
+  setOperationAction(ISD::SETCC, MVT::i32, Expand);
+  setOperationAction(ISD::SETCC, MVT::f32, Expand);
+  setOperationAction(ISD::FP_TO_UINT, MVT::i1, Custom);
+
+  setOperationAction(ISD::SELECT, MVT::i32, Custom);
+  setOperationAction(ISD::SELECT, MVT::f32, Custom);
+
+  // Legalize loads and stores to the private address space.
+  setOperationAction(ISD::LOAD, MVT::i32, Custom);
+  setOperationAction(ISD::LOAD, MVT::v2i32, Custom);
+  setOperationAction(ISD::LOAD, MVT::v4i32, Custom);
+  setLoadExtAction(ISD::EXTLOAD, MVT::v4i8, Custom);
+  setLoadExtAction(ISD::EXTLOAD, MVT::i8, Custom);
+  setLoadExtAction(ISD::ZEXTLOAD, MVT::i8, Custom);
+  setLoadExtAction(ISD::ZEXTLOAD, MVT::v4i8, Custom);
+  setOperationAction(ISD::STORE, MVT::i8, Custom);
+  setOperationAction(ISD::STORE, MVT::i32, Custom);
+  setOperationAction(ISD::STORE, MVT::v2i32, Custom);
+  setOperationAction(ISD::STORE, MVT::v4i32, Custom);
+
+  setOperationAction(ISD::LOAD, MVT::i32, Custom);
+  setOperationAction(ISD::LOAD, MVT::v4i32, Custom);
+  setOperationAction(ISD::FrameIndex, MVT::i32, Custom);
+
+  setTargetDAGCombine(ISD::FP_ROUND);
+  setTargetDAGCombine(ISD::FP_TO_SINT);
+  setTargetDAGCombine(ISD::EXTRACT_VECTOR_ELT);
+  setTargetDAGCombine(ISD::SELECT_CC);
+
+  setBooleanContents(ZeroOrNegativeOneBooleanContent);
+  setSchedulingPreference(Sched::VLIW);
+}
+
+MachineBasicBlock * R600TargetLowering::EmitInstrWithCustomInserter(
+    MachineInstr * MI, MachineBasicBlock * BB) const {
+  MachineFunction * MF = BB->getParent();
+  MachineRegisterInfo &MRI = MF->getRegInfo();
+  MachineBasicBlock::iterator I = *MI;
+
+  switch (MI->getOpcode()) {
+  default: return AMDGPUTargetLowering::EmitInstrWithCustomInserter(MI, BB);
+  case AMDGPU::CLAMP_R600: {
+    MachineInstr *NewMI = TII->buildDefaultInstruction(*BB, I,
+                                                   AMDGPU::MOV,
+                                                   MI->getOperand(0).getReg(),
+                                                   MI->getOperand(1).getReg());
+    TII->addFlag(NewMI, 0, MO_FLAG_CLAMP);
+    break;
+  }
+
+  case AMDGPU::FABS_R600: {
+    MachineInstr *NewMI = TII->buildDefaultInstruction(*BB, I,
+                                                    AMDGPU::MOV,
+                                                    MI->getOperand(0).getReg(),
+                                                    MI->getOperand(1).getReg());
+    TII->addFlag(NewMI, 0, MO_FLAG_ABS);
+    break;
+  }
+
+  case AMDGPU::FNEG_R600: {
+    MachineInstr *NewMI = TII->buildDefaultInstruction(*BB, I,
+                                                    AMDGPU::MOV,
+                                                    MI->getOperand(0).getReg(),
+                                                    MI->getOperand(1).getReg());
+    TII->addFlag(NewMI, 0, MO_FLAG_NEG);
+    break;
+  }
+
+  case AMDGPU::MASK_WRITE: {
+    unsigned maskedRegister = MI->getOperand(0).getReg();
+    assert(TargetRegisterInfo::isVirtualRegister(maskedRegister));
+    MachineInstr * defInstr = MRI.getVRegDef(maskedRegister);
+    TII->addFlag(defInstr, 0, MO_FLAG_MASK);
+    break;
+  }
+
+  case AMDGPU::MOV_IMM_F32:
+    TII->buildMovImm(*BB, I, MI->getOperand(0).getReg(),
+                     MI->getOperand(1).getFPImm()->getValueAPF()
+                         .bitcastToAPInt().getZExtValue());
+    break;
+  case AMDGPU::MOV_IMM_I32:
+    TII->buildMovImm(*BB, I, MI->getOperand(0).getReg(),
+                     MI->getOperand(1).getImm());
+    break;
+  case AMDGPU::CONST_COPY: {
+    MachineInstr *NewMI = TII->buildDefaultInstruction(*BB, MI, AMDGPU::MOV,
+        MI->getOperand(0).getReg(), AMDGPU::ALU_CONST);
+    TII->setImmOperand(NewMI, R600Operands::SRC0_SEL,
+        MI->getOperand(1).getImm());
+    break;
+  }
+
+  case AMDGPU::RAT_WRITE_CACHELESS_32_eg:
+  case AMDGPU::RAT_WRITE_CACHELESS_128_eg: {
+    unsigned EOP = (llvm::next(I)->getOpcode() == AMDGPU::RETURN) ? 1 : 0;
+
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(MI->getOpcode()))
+            .addOperand(MI->getOperand(0))
+            .addOperand(MI->getOperand(1))
+            .addImm(EOP); // Set End of program bit
+    break;
+  }
+
+  case AMDGPU::TXD: {
+    unsigned T0 = MRI.createVirtualRegister(&AMDGPU::R600_Reg128RegClass);
+    unsigned T1 = MRI.createVirtualRegister(&AMDGPU::R600_Reg128RegClass);
+
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::TEX_SET_GRADIENTS_H), T0)
+            .addOperand(MI->getOperand(3))
+            .addOperand(MI->getOperand(4))
+            .addOperand(MI->getOperand(5))
+            .addOperand(MI->getOperand(6));
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::TEX_SET_GRADIENTS_V), T1)
+            .addOperand(MI->getOperand(2))
+            .addOperand(MI->getOperand(4))
+            .addOperand(MI->getOperand(5))
+            .addOperand(MI->getOperand(6));
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::TEX_SAMPLE_G))
+            .addOperand(MI->getOperand(0))
+            .addOperand(MI->getOperand(1))
+            .addOperand(MI->getOperand(4))
+            .addOperand(MI->getOperand(5))
+            .addOperand(MI->getOperand(6))
+            .addReg(T0, RegState::Implicit)
+            .addReg(T1, RegState::Implicit);
+    break;
+  }
+
+  case AMDGPU::TXD_SHADOW: {
+    unsigned T0 = MRI.createVirtualRegister(&AMDGPU::R600_Reg128RegClass);
+    unsigned T1 = MRI.createVirtualRegister(&AMDGPU::R600_Reg128RegClass);
+
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::TEX_SET_GRADIENTS_H), T0)
+            .addOperand(MI->getOperand(3))
+            .addOperand(MI->getOperand(4))
+            .addOperand(MI->getOperand(5))
+            .addOperand(MI->getOperand(6));
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::TEX_SET_GRADIENTS_V), T1)
+            .addOperand(MI->getOperand(2))
+            .addOperand(MI->getOperand(4))
+            .addOperand(MI->getOperand(5))
+            .addOperand(MI->getOperand(6));
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::TEX_SAMPLE_C_G))
+            .addOperand(MI->getOperand(0))
+            .addOperand(MI->getOperand(1))
+            .addOperand(MI->getOperand(4))
+            .addOperand(MI->getOperand(5))
+            .addOperand(MI->getOperand(6))
+            .addReg(T0, RegState::Implicit)
+            .addReg(T1, RegState::Implicit);
+    break;
+  }
+
+  case AMDGPU::BRANCH:
+      BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::JUMP))
+              .addOperand(MI->getOperand(0));
+      break;
+
+  case AMDGPU::BRANCH_COND_f32: {
+    MachineInstr *NewMI =
+      BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::PRED_X),
+              AMDGPU::PREDICATE_BIT)
+              .addOperand(MI->getOperand(1))
+              .addImm(OPCODE_IS_NOT_ZERO)
+              .addImm(0); // Flags
+    TII->addFlag(NewMI, 0, MO_FLAG_PUSH);
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::JUMP_COND))
+            .addOperand(MI->getOperand(0))
+            .addReg(AMDGPU::PREDICATE_BIT, RegState::Kill);
+    break;
+  }
+
+  case AMDGPU::BRANCH_COND_i32: {
+    MachineInstr *NewMI =
+      BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::PRED_X),
+            AMDGPU::PREDICATE_BIT)
+            .addOperand(MI->getOperand(1))
+            .addImm(OPCODE_IS_NOT_ZERO_INT)
+            .addImm(0); // Flags
+    TII->addFlag(NewMI, 0, MO_FLAG_PUSH);
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(AMDGPU::JUMP_COND))
+           .addOperand(MI->getOperand(0))
+            .addReg(AMDGPU::PREDICATE_BIT, RegState::Kill);
+    break;
+  }
+
+  case AMDGPU::EG_ExportSwz:
+  case AMDGPU::R600_ExportSwz: {
+    // Instruction is left unmodified if its not the last one of its type
+    bool isLastInstructionOfItsType = true;
+    unsigned InstExportType = MI->getOperand(1).getImm();
+    for (MachineBasicBlock::iterator NextExportInst = llvm::next(I),
+         EndBlock = BB->end(); NextExportInst != EndBlock;
+         NextExportInst = llvm::next(NextExportInst)) {
+      if (NextExportInst->getOpcode() == AMDGPU::EG_ExportSwz ||
+          NextExportInst->getOpcode() == AMDGPU::R600_ExportSwz) {
+        unsigned CurrentInstExportType = NextExportInst->getOperand(1)
+            .getImm();
+        if (CurrentInstExportType == InstExportType) {
+          isLastInstructionOfItsType = false;
+          break;
+        }
+      }
+    }
+    bool EOP = (llvm::next(I)->getOpcode() == AMDGPU::RETURN)? 1 : 0;
+    if (!EOP && !isLastInstructionOfItsType)
+      return BB;
+    unsigned CfInst = (MI->getOpcode() == AMDGPU::EG_ExportSwz)? 84 : 40;
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(MI->getOpcode()))
+            .addOperand(MI->getOperand(0))
+            .addOperand(MI->getOperand(1))
+            .addOperand(MI->getOperand(2))
+            .addOperand(MI->getOperand(3))
+            .addOperand(MI->getOperand(4))
+            .addOperand(MI->getOperand(5))
+            .addOperand(MI->getOperand(6))
+            .addImm(CfInst)
+            .addImm(EOP);
+    break;
+  }
+  case AMDGPU::RETURN: {
+    // RETURN instructions must have the live-out registers as implicit uses,
+    // otherwise they appear dead.
+    R600MachineFunctionInfo *MFI = MF->getInfo<R600MachineFunctionInfo>();
+    MachineInstrBuilder MIB(*MF, MI);
+    for (unsigned i = 0, e = MFI->LiveOuts.size(); i != e; ++i)
+      MIB.addReg(MFI->LiveOuts[i], RegState::Implicit);
+    return BB;
+  }
+  }
+
+  MI->eraseFromParent();
+  return BB;
+}
+
+//===----------------------------------------------------------------------===//
+// Custom DAG Lowering Operations
+//===----------------------------------------------------------------------===//
+
+using namespace llvm::Intrinsic;
+using namespace llvm::AMDGPUIntrinsic;
+
+SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
+  switch (Op.getOpcode()) {
+  default: return AMDGPUTargetLowering::LowerOperation(Op, DAG);
+  case ISD::ROTL: return LowerROTL(Op, DAG);
+  case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
+  case ISD::SELECT: return LowerSELECT(Op, DAG);
+  case ISD::STORE: return LowerSTORE(Op, DAG);
+  case ISD::LOAD: return LowerLOAD(Op, DAG);
+  case ISD::FrameIndex: return LowerFrameIndex(Op, DAG);
+  case ISD::INTRINSIC_VOID: {
+    SDValue Chain = Op.getOperand(0);
+    unsigned IntrinsicID =
+                         cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
+    switch (IntrinsicID) {
+    case AMDGPUIntrinsic::AMDGPU_store_output: {
+      MachineFunction &MF = DAG.getMachineFunction();
+      R600MachineFunctionInfo *MFI = MF.getInfo<R600MachineFunctionInfo>();
+      int64_t RegIndex = cast<ConstantSDNode>(Op.getOperand(3))->getZExtValue();
+      unsigned Reg = AMDGPU::R600_TReg32RegClass.getRegister(RegIndex);
+      MFI->LiveOuts.push_back(Reg);
+      return DAG.getCopyToReg(Chain, Op.getDebugLoc(), Reg, Op.getOperand(2));
+    }
+    case AMDGPUIntrinsic::R600_store_swizzle: {
+      const SDValue Args[8] = {
+        Chain,
+        Op.getOperand(2), // Export Value
+        Op.getOperand(3), // ArrayBase
+        Op.getOperand(4), // Type
+        DAG.getConstant(0, MVT::i32), // SWZ_X
+        DAG.getConstant(1, MVT::i32), // SWZ_Y
+        DAG.getConstant(2, MVT::i32), // SWZ_Z
+        DAG.getConstant(3, MVT::i32) // SWZ_W
+      };
+      return DAG.getNode(AMDGPUISD::EXPORT, Op.getDebugLoc(), Op.getValueType(),
+          Args, 8);
+    }
+
+    // default for switch(IntrinsicID)
+    default: break;
+    }
+    // break out of case ISD::INTRINSIC_VOID in switch(Op.getOpcode())
+    break;
+  }
+  case ISD::INTRINSIC_WO_CHAIN: {
+    unsigned IntrinsicID =
+                         cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
+    EVT VT = Op.getValueType();
+    DebugLoc DL = Op.getDebugLoc();
+    switch(IntrinsicID) {
+    default: return AMDGPUTargetLowering::LowerOperation(Op, DAG);
+    case AMDGPUIntrinsic::R600_load_input: {
+      int64_t RegIndex = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
+      unsigned Reg = AMDGPU::R600_TReg32RegClass.getRegister(RegIndex);
+      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass, Reg, VT);
+    }
+
+    case AMDGPUIntrinsic::R600_interp_input: {
+      int slot = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
+      int ijb = cast<ConstantSDNode>(Op.getOperand(2))->getSExtValue();
+      MachineSDNode *interp;
+      if (ijb < 0) {
+        interp = DAG.getMachineNode(AMDGPU::INTERP_VEC_LOAD, DL,
+            MVT::v4f32, DAG.getTargetConstant(slot / 4 , MVT::i32));
+        return DAG.getTargetExtractSubreg(
+            TII->getRegisterInfo().getSubRegFromChannel(slot % 4),
+            DL, MVT::f32, SDValue(interp, 0));
+      }
+
+      if (slot % 4 < 2)
+        interp = DAG.getMachineNode(AMDGPU::INTERP_PAIR_XY, DL,
+            MVT::f32, MVT::f32, DAG.getTargetConstant(slot / 4 , MVT::i32),
+            CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+                AMDGPU::R600_TReg32RegClass.getRegister(2 * ijb + 1), MVT::f32),
+            CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+                AMDGPU::R600_TReg32RegClass.getRegister(2 * ijb), MVT::f32));
+      else
+        interp = DAG.getMachineNode(AMDGPU::INTERP_PAIR_ZW, DL,
+            MVT::f32, MVT::f32, DAG.getTargetConstant(slot / 4 , MVT::i32),
+            CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+                AMDGPU::R600_TReg32RegClass.getRegister(2 * ijb + 1), MVT::f32),
+            CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+                AMDGPU::R600_TReg32RegClass.getRegister(2 * ijb), MVT::f32));
+
+      return SDValue(interp, slot % 2);
+    }
+
+    case r600_read_ngroups_x:
+      return LowerImplicitParameter(DAG, VT, DL, 0);
+    case r600_read_ngroups_y:
+      return LowerImplicitParameter(DAG, VT, DL, 1);
+    case r600_read_ngroups_z:
+      return LowerImplicitParameter(DAG, VT, DL, 2);
+    case r600_read_global_size_x:
+      return LowerImplicitParameter(DAG, VT, DL, 3);
+    case r600_read_global_size_y:
+      return LowerImplicitParameter(DAG, VT, DL, 4);
+    case r600_read_global_size_z:
+      return LowerImplicitParameter(DAG, VT, DL, 5);
+    case r600_read_local_size_x:
+      return LowerImplicitParameter(DAG, VT, DL, 6);
+    case r600_read_local_size_y:
+      return LowerImplicitParameter(DAG, VT, DL, 7);
+    case r600_read_local_size_z:
+      return LowerImplicitParameter(DAG, VT, DL, 8);
+
+    case r600_read_tgid_x:
+      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+                                  AMDGPU::T1_X, VT);
+    case r600_read_tgid_y:
+      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+                                  AMDGPU::T1_Y, VT);
+    case r600_read_tgid_z:
+      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+                                  AMDGPU::T1_Z, VT);
+    case r600_read_tidig_x:
+      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+                                  AMDGPU::T0_X, VT);
+    case r600_read_tidig_y:
+      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+                                  AMDGPU::T0_Y, VT);
+    case r600_read_tidig_z:
+      return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
+                                  AMDGPU::T0_Z, VT);
+    }
+    // break out of case ISD::INTRINSIC_WO_CHAIN in switch(Op.getOpcode())
+    break;
+  }
+  } // end switch(Op.getOpcode())
+  return SDValue();
+}
+
+void R600TargetLowering::ReplaceNodeResults(SDNode *N,
+                                            SmallVectorImpl<SDValue> &Results,
+                                            SelectionDAG &DAG) const {
+  switch (N->getOpcode()) {
+  default: return;
+  case ISD::FP_TO_UINT: Results.push_back(LowerFPTOUINT(N->getOperand(0), DAG));
+    return;
+  case ISD::LOAD: {
+    SDNode *Node = LowerLOAD(SDValue(N, 0), DAG).getNode();
+    Results.push_back(SDValue(Node, 0));
+    Results.push_back(SDValue(Node, 1));
+    // XXX: LLVM seems not to replace Chain Value inside CustomWidenLowerNode
+    // function
+    DAG.ReplaceAllUsesOfValueWith(SDValue(N,1), SDValue(Node, 1));
+    return;
+  }
+  case ISD::STORE:
+    SDNode *Node = LowerSTORE(SDValue(N, 0), DAG).getNode();
+    Results.push_back(SDValue(Node, 0));
+    return;
+  }
+}
+
+SDValue R600TargetLowering::LowerFPTOUINT(SDValue Op, SelectionDAG &DAG) const {
+  return DAG.getNode(
+      ISD::SETCC,
+      Op.getDebugLoc(),
+      MVT::i1,
+      Op, DAG.getConstantFP(0.0f, MVT::f32),
+      DAG.getCondCode(ISD::SETNE)
+      );
+}
+
+SDValue R600TargetLowering::LowerImplicitParameter(SelectionDAG &DAG, EVT VT,
+                                                   DebugLoc DL,
+                                                   unsigned DwordOffset) const {
+  unsigned ByteOffset = DwordOffset * 4;
+  PointerType * PtrType = PointerType::get(VT.getTypeForEVT(*DAG.getContext()),
+                                      AMDGPUAS::PARAM_I_ADDRESS);
+
+  // We shouldn't be using an offset wider than 16-bits for implicit parameters.
+  assert(isInt<16>(ByteOffset));
+
+  return DAG.getLoad(VT, DL, DAG.getEntryNode(),
+                     DAG.getConstant(ByteOffset, MVT::i32), // PTR
+                     MachinePointerInfo(ConstantPointerNull::get(PtrType)),
+                     false, false, false, 0);
+}
+
+SDValue R600TargetLowering::LowerFrameIndex(SDValue Op, SelectionDAG &DAG) const {
+
+  MachineFunction &MF = DAG.getMachineFunction();
+  const AMDGPUFrameLowering *TFL =
+   static_cast<const AMDGPUFrameLowering*>(getTargetMachine().getFrameLowering());
+
+  FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Op);
+  assert(FIN);
+
+  unsigned FrameIndex = FIN->getIndex();
+  unsigned Offset = TFL->getFrameIndexOffset(MF, FrameIndex);
+  return DAG.getConstant(Offset * 4 * TFL->getStackWidth(MF), MVT::i32);
+}
+
+SDValue R600TargetLowering::LowerROTL(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT VT = Op.getValueType();
+
+  return DAG.getNode(AMDGPUISD::BITALIGN, DL, VT,
+                     Op.getOperand(0),
+                     Op.getOperand(0),
+                     DAG.getNode(ISD::SUB, DL, VT,
+                                 DAG.getConstant(32, MVT::i32),
+                                 Op.getOperand(1)));
+}
+
+bool R600TargetLowering::isZero(SDValue Op) const {
+  if(ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(Op)) {
+    return Cst->isNullValue();
+  } else if(ConstantFPSDNode *CstFP = dyn_cast<ConstantFPSDNode>(Op)){
+    return CstFP->isZero();
+  } else {
+    return false;
+  }
+}
+
+SDValue R600TargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  EVT VT = Op.getValueType();
+
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  SDValue True = Op.getOperand(2);
+  SDValue False = Op.getOperand(3);
+  SDValue CC = Op.getOperand(4);
+  SDValue Temp;
+
+  // LHS and RHS are guaranteed to be the same value type
+  EVT CompareVT = LHS.getValueType();
+
+  // Check if we can lower this to a native operation.
+
+  // Try to lower to a SET* instruction:
+  //
+  // SET* can match the following patterns:
+  //
+  // select_cc f32, f32, -1,  0, cc_any
+  // select_cc f32, f32, 1.0f, 0.0f, cc_any
+  // select_cc i32, i32, -1,  0, cc_any
+  //
+
+  // Move hardware True/False values to the correct operand.
+  if (isHWTrueValue(False) && isHWFalseValue(True)) {
+    ISD::CondCode CCOpcode = cast<CondCodeSDNode>(CC)->get();
+    std::swap(False, True);
+    CC = DAG.getCondCode(ISD::getSetCCInverse(CCOpcode, CompareVT == MVT::i32));
+  }
+
+  if (isHWTrueValue(True) && isHWFalseValue(False) &&
+      (CompareVT == VT || VT == MVT::i32)) {
+    // This can be matched by a SET* instruction.
+    return DAG.getNode(ISD::SELECT_CC, DL, VT, LHS, RHS, True, False, CC);
+  }
+
+  // Try to lower to a CND* instruction:
+  //
+  // CND* can match the following patterns:
+  //
+  // select_cc f32, 0.0, f32, f32, cc_any
+  // select_cc f32, 0.0, i32, i32, cc_any
+  // select_cc i32, 0,   f32, f32, cc_any
+  // select_cc i32, 0,   i32, i32, cc_any
+  //
+  if (isZero(LHS) || isZero(RHS)) {
+    SDValue Cond = (isZero(LHS) ? RHS : LHS);
+    SDValue Zero = (isZero(LHS) ? LHS : RHS);
+    ISD::CondCode CCOpcode = cast<CondCodeSDNode>(CC)->get();
+    if (CompareVT != VT) {
+      // Bitcast True / False to the correct types.  This will end up being
+      // a nop, but it allows us to define only a single pattern in the
+      // .TD files for each CND* instruction rather than having to have
+      // one pattern for integer True/False and one for fp True/False
+      True = DAG.getNode(ISD::BITCAST, DL, CompareVT, True);
+      False = DAG.getNode(ISD::BITCAST, DL, CompareVT, False);
+    }
+    if (isZero(LHS)) {
+      CCOpcode = ISD::getSetCCSwappedOperands(CCOpcode);
+    }
+
+    switch (CCOpcode) {
+    case ISD::SETONE:
+    case ISD::SETUNE:
+    case ISD::SETNE:
+    case ISD::SETULE:
+    case ISD::SETULT:
+    case ISD::SETOLE:
+    case ISD::SETOLT:
+    case ISD::SETLE:
+    case ISD::SETLT:
+      CCOpcode = ISD::getSetCCInverse(CCOpcode, CompareVT == MVT::i32);
+      Temp = True;
+      True = False;
+      False = Temp;
+      break;
+    default:
+      break;
+    }
+    SDValue SelectNode = DAG.getNode(ISD::SELECT_CC, DL, CompareVT,
+        Cond, Zero,
+        True, False,
+        DAG.getCondCode(CCOpcode));
+    return DAG.getNode(ISD::BITCAST, DL, VT, SelectNode);
+  }
+
+
+  // Possible Min/Max pattern
+  SDValue MinMax = LowerMinMax(Op, DAG);
+  if (MinMax.getNode()) {
+    return MinMax;
+  }
+
+  // If we make it this for it means we have no native instructions to handle
+  // this SELECT_CC, so we must lower it.
+  SDValue HWTrue, HWFalse;
+
+  if (CompareVT == MVT::f32) {
+    HWTrue = DAG.getConstantFP(1.0f, CompareVT);
+    HWFalse = DAG.getConstantFP(0.0f, CompareVT);
+  } else if (CompareVT == MVT::i32) {
+    HWTrue = DAG.getConstant(-1, CompareVT);
+    HWFalse = DAG.getConstant(0, CompareVT);
+  }
+  else {
+    assert(!"Unhandled value type in LowerSELECT_CC");
+  }
+
+  // Lower this unsupported SELECT_CC into a combination of two supported
+  // SELECT_CC operations.
+  SDValue Cond = DAG.getNode(ISD::SELECT_CC, DL, CompareVT, LHS, RHS, HWTrue, HWFalse, CC);
+
+  return DAG.getNode(ISD::SELECT_CC, DL, VT,
+      Cond, HWFalse,
+      True, False,
+      DAG.getCondCode(ISD::SETNE));
+}
+
+SDValue R600TargetLowering::LowerSELECT(SDValue Op, SelectionDAG &DAG) const {
+  return DAG.getNode(ISD::SELECT_CC,
+      Op.getDebugLoc(),
+      Op.getValueType(),
+      Op.getOperand(0),
+      DAG.getConstant(0, MVT::i32),
+      Op.getOperand(1),
+      Op.getOperand(2),
+      DAG.getCondCode(ISD::SETNE));
+}
+
+/// LLVM generates byte-addresed pointers.  For indirect addressing, we need to
+/// convert these pointers to a register index.  Each register holds
+/// 16 bytes, (4 x 32bit sub-register), but we need to take into account the
+/// \p StackWidth, which tells us how many of the 4 sub-registrers will be used
+/// for indirect addressing.
+SDValue R600TargetLowering::stackPtrToRegIndex(SDValue Ptr,
+                                               unsigned StackWidth,
+                                               SelectionDAG &DAG) const {
+  unsigned SRLPad;
+  switch(StackWidth) {
+  case 1:
+    SRLPad = 2;
+    break;
+  case 2:
+    SRLPad = 3;
+    break;
+  case 4:
+    SRLPad = 4;
+    break;
+  default: llvm_unreachable("Invalid stack width");
+  }
+
+  return DAG.getNode(ISD::SRL, Ptr.getDebugLoc(), Ptr.getValueType(), Ptr,
+                     DAG.getConstant(SRLPad, MVT::i32));
+}
+
+void R600TargetLowering::getStackAddress(unsigned StackWidth,
+                                         unsigned ElemIdx,
+                                         unsigned &Channel,
+                                         unsigned &PtrIncr) const {
+  switch (StackWidth) {
+  default:
+  case 1:
+    Channel = 0;
+    if (ElemIdx > 0) {
+      PtrIncr = 1;
+    } else {
+      PtrIncr = 0;
+    }
+    break;
+  case 2:
+    Channel = ElemIdx % 2;
+    if (ElemIdx == 2) {
+      PtrIncr = 1;
+    } else {
+      PtrIncr = 0;
+    }
+    break;
+  case 4:
+    Channel = ElemIdx;
+    PtrIncr = 0;
+    break;
+  }
+}
+
+SDValue R600TargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  StoreSDNode *StoreNode = cast<StoreSDNode>(Op);
+  SDValue Chain = Op.getOperand(0);
+  SDValue Value = Op.getOperand(1);
+  SDValue Ptr = Op.getOperand(2);
+
+  if (StoreNode->getAddressSpace() == AMDGPUAS::GLOBAL_ADDRESS &&
+      Ptr->getOpcode() != AMDGPUISD::DWORDADDR) {
+    // Convert pointer from byte address to dword address.
+    Ptr = DAG.getNode(AMDGPUISD::DWORDADDR, DL, Ptr.getValueType(),
+                      DAG.getNode(ISD::SRL, DL, Ptr.getValueType(),
+                                  Ptr, DAG.getConstant(2, MVT::i32)));
+
+    if (StoreNode->isTruncatingStore() || StoreNode->isIndexed()) {
+      assert(!"Truncated and indexed stores not supported yet");
+    } else {
+      Chain = DAG.getStore(Chain, DL, Value, Ptr, StoreNode->getMemOperand());
+    }
+    return Chain;
+  }
+
+  EVT ValueVT = Value.getValueType();
+
+  if (StoreNode->getAddressSpace() != AMDGPUAS::PRIVATE_ADDRESS) {
+    return SDValue();
+  }
+
+  // Lowering for indirect addressing
+
+  const MachineFunction &MF = DAG.getMachineFunction();
+  const AMDGPUFrameLowering *TFL = static_cast<const AMDGPUFrameLowering*>(
+                                         getTargetMachine().getFrameLowering());
+  unsigned StackWidth = TFL->getStackWidth(MF);
+
+  Ptr = stackPtrToRegIndex(Ptr, StackWidth, DAG);
+
+  if (ValueVT.isVector()) {
+    unsigned NumElemVT = ValueVT.getVectorNumElements();
+    EVT ElemVT = ValueVT.getVectorElementType();
+    SDValue Stores[4];
+
+    assert(NumElemVT >= StackWidth && "Stack width cannot be greater than "
+                                      "vector width in load");
+
+    for (unsigned i = 0; i < NumElemVT; ++i) {
+      unsigned Channel, PtrIncr;
+      getStackAddress(StackWidth, i, Channel, PtrIncr);
+      Ptr = DAG.getNode(ISD::ADD, DL, MVT::i32, Ptr,
+                        DAG.getConstant(PtrIncr, MVT::i32));
+      SDValue Elem = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, ElemVT,
+                                 Value, DAG.getConstant(i, MVT::i32));
+
+      Stores[i] = DAG.getNode(AMDGPUISD::REGISTER_STORE, DL, MVT::Other,
+                              Chain, Elem, Ptr,
+                              DAG.getTargetConstant(Channel, MVT::i32));
+    }
+     Chain =  DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Stores, NumElemVT);
+   } else {
+    if (ValueVT == MVT::i8) {
+      Value = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i32, Value);
+    }
+    Chain = DAG.getNode(AMDGPUISD::REGISTER_STORE, DL, MVT::Other, Chain, Value, Ptr,
+    DAG.getTargetConstant(0, MVT::i32)); // Channel 
+  }
+
+  return Chain;
+}
+
+// return (512 + (kc_bank << 12)
+static int
+ConstantAddressBlock(unsigned AddressSpace) {
+  switch (AddressSpace) {
+  case AMDGPUAS::CONSTANT_BUFFER_0:
+    return 512;
+  case AMDGPUAS::CONSTANT_BUFFER_1:
+    return 512 + 4096;
+  case AMDGPUAS::CONSTANT_BUFFER_2:
+    return 512 + 4096 * 2;
+  case AMDGPUAS::CONSTANT_BUFFER_3:
+    return 512 + 4096 * 3;
+  case AMDGPUAS::CONSTANT_BUFFER_4:
+    return 512 + 4096 * 4;
+  case AMDGPUAS::CONSTANT_BUFFER_5:
+    return 512 + 4096 * 5;
+  case AMDGPUAS::CONSTANT_BUFFER_6:
+    return 512 + 4096 * 6;
+  case AMDGPUAS::CONSTANT_BUFFER_7:
+    return 512 + 4096 * 7;
+  case AMDGPUAS::CONSTANT_BUFFER_8:
+    return 512 + 4096 * 8;
+  case AMDGPUAS::CONSTANT_BUFFER_9:
+    return 512 + 4096 * 9;
+  case AMDGPUAS::CONSTANT_BUFFER_10:
+    return 512 + 4096 * 10;
+  case AMDGPUAS::CONSTANT_BUFFER_11:
+    return 512 + 4096 * 11;
+  case AMDGPUAS::CONSTANT_BUFFER_12:
+    return 512 + 4096 * 12;
+  case AMDGPUAS::CONSTANT_BUFFER_13:
+    return 512 + 4096 * 13;
+  case AMDGPUAS::CONSTANT_BUFFER_14:
+    return 512 + 4096 * 14;
+  case AMDGPUAS::CONSTANT_BUFFER_15:
+    return 512 + 4096 * 15;
+  default:
+    return -1;
+  }
+}
+
+SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
+{
+  EVT VT = Op.getValueType();
+  DebugLoc DL = Op.getDebugLoc();
+  LoadSDNode *LoadNode = cast<LoadSDNode>(Op);
+  SDValue Chain = Op.getOperand(0);
+  SDValue Ptr = Op.getOperand(1);
+  SDValue LoweredLoad;
+
+  int ConstantBlock = ConstantAddressBlock(LoadNode->getAddressSpace());
+  if (ConstantBlock > -1) {
+    SDValue Result;
+    if (dyn_cast<ConstantExpr>(LoadNode->getSrcValue()) ||
+        dyn_cast<Constant>(LoadNode->getSrcValue()) ||
+        dyn_cast<ConstantSDNode>(Ptr)) {
+      SDValue Slots[4];
+      for (unsigned i = 0; i < 4; i++) {
+        // We want Const position encoded with the following formula :
+        // (((512 + (kc_bank << 12) + const_index) << 2) + chan)
+        // const_index is Ptr computed by llvm using an alignment of 16.
+        // Thus we add (((512 + (kc_bank << 12)) + chan ) * 4 here and
+        // then div by 4 at the ISel step
+        SDValue NewPtr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr,
+            DAG.getConstant(4 * i + ConstantBlock * 16, MVT::i32));
+        Slots[i] = DAG.getNode(AMDGPUISD::CONST_ADDRESS, DL, MVT::i32, NewPtr);
+      }
+      Result = DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v4i32, Slots, 4);
+    } else {
+      // non constant ptr cant be folded, keeps it as a v4f32 load
+      Result = DAG.getNode(AMDGPUISD::CONST_ADDRESS, DL, MVT::v4i32,
+          DAG.getNode(ISD::SRL, DL, MVT::i32, Ptr, DAG.getConstant(4, MVT::i32)),
+          DAG.getConstant(LoadNode->getAddressSpace() -
+	                  AMDGPUAS::CONSTANT_BUFFER_0, MVT::i32)
+          );
+    }
+
+    if (!VT.isVector()) {
+      Result = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::i32, Result,
+          DAG.getConstant(0, MVT::i32));
+    }
+
+    SDValue MergedValues[2] = {
+        Result,
+        Chain
+    };
+    return DAG.getMergeValues(MergedValues, 2, DL);
+  }
+
+  if (LoadNode->getAddressSpace() != AMDGPUAS::PRIVATE_ADDRESS) {
+    return SDValue();
+  }
+
+  // Lowering for indirect addressing
+  const MachineFunction &MF = DAG.getMachineFunction();
+  const AMDGPUFrameLowering *TFL = static_cast<const AMDGPUFrameLowering*>(
+                                         getTargetMachine().getFrameLowering());
+  unsigned StackWidth = TFL->getStackWidth(MF);
+
+  Ptr = stackPtrToRegIndex(Ptr, StackWidth, DAG);
+
+  if (VT.isVector()) {
+    unsigned NumElemVT = VT.getVectorNumElements();
+    EVT ElemVT = VT.getVectorElementType();
+    SDValue Loads[4];
+
+    assert(NumElemVT >= StackWidth && "Stack width cannot be greater than "
+                                      "vector width in load");
+
+    for (unsigned i = 0; i < NumElemVT; ++i) {
+      unsigned Channel, PtrIncr;
+      getStackAddress(StackWidth, i, Channel, PtrIncr);
+      Ptr = DAG.getNode(ISD::ADD, DL, MVT::i32, Ptr,
+                        DAG.getConstant(PtrIncr, MVT::i32));
+      Loads[i] = DAG.getNode(AMDGPUISD::REGISTER_LOAD, DL, ElemVT,
+                             Chain, Ptr,
+                             DAG.getTargetConstant(Channel, MVT::i32),
+                             Op.getOperand(2));
+    }
+    for (unsigned i = NumElemVT; i < 4; ++i) {
+      Loads[i] = DAG.getUNDEF(ElemVT);
+    }
+    EVT TargetVT = EVT::getVectorVT(*DAG.getContext(), ElemVT, 4);
+    LoweredLoad = DAG.getNode(ISD::BUILD_VECTOR, DL, TargetVT, Loads, 4);
+  } else {
+    LoweredLoad = DAG.getNode(AMDGPUISD::REGISTER_LOAD, DL, VT,
+                              Chain, Ptr,
+                              DAG.getTargetConstant(0, MVT::i32), // Channel
+                              Op.getOperand(2));
+  }
+
+  SDValue Ops[2];
+  Ops[0] = LoweredLoad;
+  Ops[1] = Chain;
+
+  return DAG.getMergeValues(Ops, 2, DL);
+}
+
+/// XXX Only kernel functions are supported, so we can assume for now that
+/// every function is a kernel function, but in the future we should use
+/// separate calling conventions for kernel and non-kernel functions.
+SDValue R600TargetLowering::LowerFormalArguments(
+                                      SDValue Chain,
+                                      CallingConv::ID CallConv,
+                                      bool isVarArg,
+                                      const SmallVectorImpl<ISD::InputArg> &Ins,
+                                      DebugLoc DL, SelectionDAG &DAG,
+                                      SmallVectorImpl<SDValue> &InVals) const {
+  unsigned ParamOffsetBytes = 36;
+  Function::const_arg_iterator FuncArg =
+                            DAG.getMachineFunction().getFunction()->arg_begin();
+  for (unsigned i = 0, e = Ins.size(); i < e; ++i, ++FuncArg) {
+    EVT VT = Ins[i].VT;
+    Type *ArgType = FuncArg->getType();
+    unsigned ArgSizeInBits = ArgType->isPointerTy() ?
+                             32 : ArgType->getPrimitiveSizeInBits();
+    unsigned ArgBytes = ArgSizeInBits >> 3;
+    EVT ArgVT;
+    if (ArgSizeInBits < VT.getSizeInBits()) {
+      assert(!ArgType->isFloatTy() &&
+             "Extending floating point arguments not supported yet");
+      ArgVT = MVT::getIntegerVT(ArgSizeInBits);
+    } else {
+      ArgVT = VT;
+    }
+    PointerType *PtrTy = PointerType::get(VT.getTypeForEVT(*DAG.getContext()),
+                                                    AMDGPUAS::PARAM_I_ADDRESS);
+    SDValue Arg = DAG.getExtLoad(ISD::ZEXTLOAD, DL, VT, DAG.getRoot(),
+                                DAG.getConstant(ParamOffsetBytes, MVT::i32),
+                                       MachinePointerInfo(UndefValue::get(PtrTy)),
+                                       ArgVT, false, false, ArgBytes);
+    InVals.push_back(Arg);
+    ParamOffsetBytes += ArgBytes;
+  }
+  return Chain;
+}
+
+EVT R600TargetLowering::getSetCCResultType(EVT VT) const {
+   if (!VT.isVector()) return MVT::i32;
+   return VT.changeVectorElementTypeToInteger();
+}
+
+//===----------------------------------------------------------------------===//
+// Custom DAG Optimizations
+//===----------------------------------------------------------------------===//
+
+SDValue R600TargetLowering::PerformDAGCombine(SDNode *N,
+                                              DAGCombinerInfo &DCI) const {
+  SelectionDAG &DAG = DCI.DAG;
+
+  switch (N->getOpcode()) {
+  // (f32 fp_round (f64 uint_to_fp a)) -> (f32 uint_to_fp a)
+  case ISD::FP_ROUND: {
+      SDValue Arg = N->getOperand(0);
+      if (Arg.getOpcode() == ISD::UINT_TO_FP && Arg.getValueType() == MVT::f64) {
+        return DAG.getNode(ISD::UINT_TO_FP, N->getDebugLoc(), N->getValueType(0),
+                           Arg.getOperand(0));
+      }
+      break;
+    }
+
+  // (i32 fp_to_sint (fneg (select_cc f32, f32, 1.0, 0.0 cc))) ->
+  // (i32 select_cc f32, f32, -1, 0 cc)
+  //
+  // Mesa's GLSL frontend generates the above pattern a lot and we can lower
+  // this to one of the SET*_DX10 instructions.
+  case ISD::FP_TO_SINT: {
+    SDValue FNeg = N->getOperand(0);
+    if (FNeg.getOpcode() != ISD::FNEG) {
+      return SDValue();
+    }
+    SDValue SelectCC = FNeg.getOperand(0);
+    if (SelectCC.getOpcode() != ISD::SELECT_CC ||
+        SelectCC.getOperand(0).getValueType() != MVT::f32 || // LHS
+        SelectCC.getOperand(2).getValueType() != MVT::f32 || // True
+        !isHWTrueValue(SelectCC.getOperand(2)) ||
+        !isHWFalseValue(SelectCC.getOperand(3))) {
+      return SDValue();
+    }
+
+    return DAG.getNode(ISD::SELECT_CC, N->getDebugLoc(), N->getValueType(0),
+                           SelectCC.getOperand(0), // LHS
+                           SelectCC.getOperand(1), // RHS
+                           DAG.getConstant(-1, MVT::i32), // True
+                           DAG.getConstant(0, MVT::i32),  // Flase
+                           SelectCC.getOperand(4)); // CC
+
+    break;
+  }
+  // Extract_vec (Build_vector) generated by custom lowering
+  // also needs to be customly combined
+  case ISD::EXTRACT_VECTOR_ELT: {
+    SDValue Arg = N->getOperand(0);
+    if (Arg.getOpcode() == ISD::BUILD_VECTOR) {
+      if (ConstantSDNode *Const = dyn_cast<ConstantSDNode>(N->getOperand(1))) {
+        unsigned Element = Const->getZExtValue();
+        return Arg->getOperand(Element);
+      }
+    }
+    if (Arg.getOpcode() == ISD::BITCAST &&
+        Arg.getOperand(0).getOpcode() == ISD::BUILD_VECTOR) {
+      if (ConstantSDNode *Const = dyn_cast<ConstantSDNode>(N->getOperand(1))) {
+        unsigned Element = Const->getZExtValue();
+        return DAG.getNode(ISD::BITCAST, N->getDebugLoc(), N->getVTList(),
+            Arg->getOperand(0).getOperand(Element));
+      }
+    }
+  }
+
+  case ISD::SELECT_CC: {
+    // fold selectcc (selectcc x, y, a, b, cc), b, a, b, seteq ->
+    //      selectcc x, y, a, b, inv(cc)
+    //
+    // fold selectcc (selectcc x, y, a, b, cc), b, a, b, setne ->
+    //      selectcc x, y, a, b, cc
+    SDValue LHS = N->getOperand(0);
+    if (LHS.getOpcode() != ISD::SELECT_CC) {
+      return SDValue();
+    }
+
+    SDValue RHS = N->getOperand(1);
+    SDValue True = N->getOperand(2);
+    SDValue False = N->getOperand(3);
+    ISD::CondCode NCC = cast<CondCodeSDNode>(N->getOperand(4))->get();
+
+    if (LHS.getOperand(2).getNode() != True.getNode() ||
+        LHS.getOperand(3).getNode() != False.getNode() ||
+        RHS.getNode() != False.getNode()) {
+      return SDValue();
+    }
+
+    switch (NCC) {
+    default: return SDValue();
+    case ISD::SETNE: return LHS;
+    case ISD::SETEQ: {
+      ISD::CondCode LHSCC = cast<CondCodeSDNode>(LHS.getOperand(4))->get();
+      LHSCC = ISD::getSetCCInverse(LHSCC,
+                                  LHS.getOperand(0).getValueType().isInteger());
+      return DAG.getSelectCC(N->getDebugLoc(),
+                             LHS.getOperand(0),
+                             LHS.getOperand(1),
+                             LHS.getOperand(2),
+                             LHS.getOperand(3),
+                             LHSCC);
+    }
+    }
+  }
+  case AMDGPUISD::EXPORT: {
+    SDValue Arg = N->getOperand(1);
+    if (Arg.getOpcode() != ISD::BUILD_VECTOR)
+      break;
+    SDValue NewBldVec[4] = {
+        DAG.getUNDEF(MVT::f32),
+        DAG.getUNDEF(MVT::f32),
+        DAG.getUNDEF(MVT::f32),
+        DAG.getUNDEF(MVT::f32)
+      };
+    SDValue NewArgs[8] = {
+      N->getOperand(0), // Chain
+      SDValue(),
+      N->getOperand(2), // ArrayBase
+      N->getOperand(3), // Type
+      N->getOperand(4), // SWZ_X
+      N->getOperand(5), // SWZ_Y
+      N->getOperand(6), // SWZ_Z
+      N->getOperand(7) // SWZ_W
+    };
+    for (unsigned i = 0; i < Arg.getNumOperands(); i++) {
+      if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Arg.getOperand(i))) {
+        if (C->isZero()) {
+          NewArgs[4 + i] = DAG.getConstant(4, MVT::i32); // SEL_0
+        } else if (C->isExactlyValue(1.0)) {
+          NewArgs[4 + i] = DAG.getConstant(5, MVT::i32); // SEL_0
+        } else {
+          NewBldVec[i] = Arg.getOperand(i);
+        }
+      } else {
+        NewBldVec[i] = Arg.getOperand(i);
+      }
+    }
+    DebugLoc DL = N->getDebugLoc();
+    NewArgs[1] = DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v4f32, NewBldVec, 4);
+    return DAG.getNode(AMDGPUISD::EXPORT, DL, N->getVTList(), NewArgs, 8);
+  }
+  }
+  return SDValue();
+}
diff --git a/lib/Target/R600/R600ISelLowering.h b/lib/Target/R600/R600ISelLowering.h
new file mode 100644
index 000000000000..2c09acb9af30
--- /dev/null
+++ b/lib/Target/R600/R600ISelLowering.h
@@ -0,0 +1,74 @@
+//===-- R600ISelLowering.h - R600 DAG Lowering Interface -*- C++ -*--------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief R600 DAG Lowering interface definition
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef R600ISELLOWERING_H
+#define R600ISELLOWERING_H
+
+#include "AMDGPUISelLowering.h"
+
+namespace llvm {
+
+class R600InstrInfo;
+
+class R600TargetLowering : public AMDGPUTargetLowering {
+public:
+  R600TargetLowering(TargetMachine &TM);
+  virtual MachineBasicBlock * EmitInstrWithCustomInserter(MachineInstr *MI,
+      MachineBasicBlock * BB) const;
+  virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
+  virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
+  void ReplaceNodeResults(SDNode * N,
+      SmallVectorImpl<SDValue> &Results,
+      SelectionDAG &DAG) const;
+  virtual SDValue LowerFormalArguments(
+                                      SDValue Chain,
+                                      CallingConv::ID CallConv,
+                                      bool isVarArg,
+                                      const SmallVectorImpl<ISD::InputArg> &Ins,
+                                      DebugLoc DL, SelectionDAG &DAG,
+                                      SmallVectorImpl<SDValue> &InVals) const;
+  virtual EVT getSetCCResultType(EVT VT) const;
+private:
+  const R600InstrInfo * TII;
+
+  /// Each OpenCL kernel has nine implicit parameters that are stored in the
+  /// first nine dwords of a Vertex Buffer.  These implicit parameters are
+  /// lowered to load instructions which retreive the values from the Vertex
+  /// Buffer.
+  SDValue LowerImplicitParameter(SelectionDAG &DAG, EVT VT,
+                                 DebugLoc DL, unsigned DwordOffset) const;
+
+  void lowerImplicitParameter(MachineInstr *MI, MachineBasicBlock &BB,
+      MachineRegisterInfo & MRI, unsigned dword_offset) const;
+
+  /// \brief Lower ROTL opcode to BITALIGN
+  SDValue LowerROTL(SDValue Op, SelectionDAG &DAG) const;
+
+  SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFPTOUINT(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFrameIndex(SDValue Op, SelectionDAG &DAG) const;
+
+  SDValue stackPtrToRegIndex(SDValue Ptr, unsigned StackWidth,
+                                          SelectionDAG &DAG) const;
+  void getStackAddress(unsigned StackWidth, unsigned ElemIdx,
+                       unsigned &Channel, unsigned &PtrIncr) const;
+  bool isZero(SDValue Op) const;
+};
+
+} // End namespace llvm;
+
+#endif // R600ISELLOWERING_H
diff --git a/lib/Target/R600/R600InstrInfo.cpp b/lib/Target/R600/R600InstrInfo.cpp
new file mode 100644
index 000000000000..b232188a2641
--- /dev/null
+++ b/lib/Target/R600/R600InstrInfo.cpp
@@ -0,0 +1,841 @@
+//===-- R600InstrInfo.cpp - R600 Instruction Information ------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief R600 Implementation of TargetInstrInfo.
+//
+//===----------------------------------------------------------------------===//
+
+#include "R600InstrInfo.h"
+#include "AMDGPUSubtarget.h"
+#include "AMDGPUTargetMachine.h"
+#include "R600Defines.h"
+#include "R600MachineFunctionInfo.h"
+#include "R600RegisterInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+
+#define GET_INSTRINFO_CTOR
+#include "AMDGPUGenDFAPacketizer.inc"
+
+using namespace llvm;
+
+R600InstrInfo::R600InstrInfo(AMDGPUTargetMachine &tm)
+  : AMDGPUInstrInfo(tm),
+    RI(tm, *this)
+  { }
+
+const R600RegisterInfo &R600InstrInfo::getRegisterInfo() const {
+  return RI;
+}
+
+bool R600InstrInfo::isTrig(const MachineInstr &MI) const {
+  return get(MI.getOpcode()).TSFlags & R600_InstFlag::TRIG;
+}
+
+bool R600InstrInfo::isVector(const MachineInstr &MI) const {
+  return get(MI.getOpcode()).TSFlags & R600_InstFlag::VECTOR;
+}
+
+void
+R600InstrInfo::copyPhysReg(MachineBasicBlock &MBB,
+                           MachineBasicBlock::iterator MI, DebugLoc DL,
+                           unsigned DestReg, unsigned SrcReg,
+                           bool KillSrc) const {
+  if (AMDGPU::R600_Reg128RegClass.contains(DestReg)
+      && AMDGPU::R600_Reg128RegClass.contains(SrcReg)) {
+    for (unsigned I = 0; I < 4; I++) {
+      unsigned SubRegIndex = RI.getSubRegFromChannel(I);
+      buildDefaultInstruction(MBB, MI, AMDGPU::MOV,
+                              RI.getSubReg(DestReg, SubRegIndex),
+                              RI.getSubReg(SrcReg, SubRegIndex))
+                              .addReg(DestReg,
+                                      RegState::Define | RegState::Implicit);
+    }
+  } else {
+
+    // We can't copy vec4 registers
+    assert(!AMDGPU::R600_Reg128RegClass.contains(DestReg)
+           && !AMDGPU::R600_Reg128RegClass.contains(SrcReg));
+
+    MachineInstr *NewMI = buildDefaultInstruction(MBB, MI, AMDGPU::MOV,
+                                                  DestReg, SrcReg);
+    NewMI->getOperand(getOperandIdx(*NewMI, R600Operands::SRC0))
+                                    .setIsKill(KillSrc);
+  }
+}
+
+MachineInstr * R600InstrInfo::getMovImmInstr(MachineFunction *MF,
+                                             unsigned DstReg, int64_t Imm) const {
+  MachineInstr * MI = MF->CreateMachineInstr(get(AMDGPU::MOV), DebugLoc());
+  MachineInstrBuilder MIB(*MF, MI);
+  MIB.addReg(DstReg, RegState::Define);
+  MIB.addReg(AMDGPU::ALU_LITERAL_X);
+  MIB.addImm(Imm);
+  MIB.addReg(0); // PREDICATE_BIT
+
+  return MI;
+}
+
+unsigned R600InstrInfo::getIEQOpcode() const {
+  return AMDGPU::SETE_INT;
+}
+
+bool R600InstrInfo::isMov(unsigned Opcode) const {
+
+
+  switch(Opcode) {
+  default: return false;
+  case AMDGPU::MOV:
+  case AMDGPU::MOV_IMM_F32:
+  case AMDGPU::MOV_IMM_I32:
+    return true;
+  }
+}
+
+// Some instructions act as place holders to emulate operations that the GPU
+// hardware does automatically. This function can be used to check if
+// an opcode falls into this category.
+bool R600InstrInfo::isPlaceHolderOpcode(unsigned Opcode) const {
+  switch (Opcode) {
+  default: return false;
+  case AMDGPU::RETURN:
+    return true;
+  }
+}
+
+bool R600InstrInfo::isReductionOp(unsigned Opcode) const {
+  switch(Opcode) {
+    default: return false;
+    case AMDGPU::DOT4_r600_pseudo:
+    case AMDGPU::DOT4_eg_pseudo:
+      return true;
+  }
+}
+
+bool R600InstrInfo::isCubeOp(unsigned Opcode) const {
+  switch(Opcode) {
+    default: return false;
+    case AMDGPU::CUBE_r600_pseudo:
+    case AMDGPU::CUBE_r600_real:
+    case AMDGPU::CUBE_eg_pseudo:
+    case AMDGPU::CUBE_eg_real:
+      return true;
+  }
+}
+
+bool R600InstrInfo::isALUInstr(unsigned Opcode) const {
+  unsigned TargetFlags = get(Opcode).TSFlags;
+
+  return ((TargetFlags & R600_InstFlag::OP1) |
+          (TargetFlags & R600_InstFlag::OP2) |
+          (TargetFlags & R600_InstFlag::OP3));
+}
+
+bool
+R600InstrInfo::fitsConstReadLimitations(const std::vector<unsigned> &Consts)
+    const {
+  assert (Consts.size() <= 12 && "Too many operands in instructions group");
+  unsigned Pair1 = 0, Pair2 = 0;
+  for (unsigned i = 0, n = Consts.size(); i < n; ++i) {
+    unsigned ReadConstHalf = Consts[i] & 2;
+    unsigned ReadConstIndex = Consts[i] & (~3);
+    unsigned ReadHalfConst = ReadConstIndex | ReadConstHalf;
+    if (!Pair1) {
+      Pair1 = ReadHalfConst;
+      continue;
+    }
+    if (Pair1 == ReadHalfConst)
+      continue;
+    if (!Pair2) {
+      Pair2 = ReadHalfConst;
+      continue;
+    }
+    if (Pair2 != ReadHalfConst)
+      return false;
+  }
+  return true;
+}
+
+bool
+R600InstrInfo::canBundle(const std::vector<MachineInstr *> &MIs) const {
+  std::vector<unsigned> Consts;
+  for (unsigned i = 0, n = MIs.size(); i < n; i++) {
+    const MachineInstr *MI = MIs[i];
+
+    const R600Operands::Ops OpTable[3][2] = {
+      {R600Operands::SRC0, R600Operands::SRC0_SEL},
+      {R600Operands::SRC1, R600Operands::SRC1_SEL},
+      {R600Operands::SRC2, R600Operands::SRC2_SEL},
+    };
+
+    if (!isALUInstr(MI->getOpcode()))
+      continue;
+
+    for (unsigned j = 0; j < 3; j++) {
+      int SrcIdx = getOperandIdx(MI->getOpcode(), OpTable[j][0]);
+      if (SrcIdx < 0)
+        break;
+      if (MI->getOperand(SrcIdx).getReg() == AMDGPU::ALU_CONST) {
+        unsigned Const = MI->getOperand(
+            getOperandIdx(MI->getOpcode(), OpTable[j][1])).getImm();
+        Consts.push_back(Const);
+      }
+    }
+  }
+  return fitsConstReadLimitations(Consts);
+}
+
+DFAPacketizer *R600InstrInfo::CreateTargetScheduleState(const TargetMachine *TM,
+    const ScheduleDAG *DAG) const {
+  const InstrItineraryData *II = TM->getInstrItineraryData();
+  return TM->getSubtarget<AMDGPUSubtarget>().createDFAPacketizer(II);
+}
+
+static bool
+isPredicateSetter(unsigned Opcode) {
+  switch (Opcode) {
+  case AMDGPU::PRED_X:
+    return true;
+  default:
+    return false;
+  }
+}
+
+static MachineInstr *
+findFirstPredicateSetterFrom(MachineBasicBlock &MBB,
+                             MachineBasicBlock::iterator I) {
+  while (I != MBB.begin()) {
+    --I;
+    MachineInstr *MI = I;
+    if (isPredicateSetter(MI->getOpcode()))
+      return MI;
+  }
+
+  return NULL;
+}
+
+static
+bool isJump(unsigned Opcode) {
+  return Opcode == AMDGPU::JUMP || Opcode == AMDGPU::JUMP_COND;
+}
+
+bool
+R600InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
+                             MachineBasicBlock *&TBB,
+                             MachineBasicBlock *&FBB,
+                             SmallVectorImpl<MachineOperand> &Cond,
+                             bool AllowModify) const {
+  // Most of the following comes from the ARM implementation of AnalyzeBranch
+
+  // If the block has no terminators, it just falls into the block after it.
+  MachineBasicBlock::iterator I = MBB.end();
+  if (I == MBB.begin())
+    return false;
+  --I;
+  while (I->isDebugValue()) {
+    if (I == MBB.begin())
+      return false;
+    --I;
+  }
+  if (!isJump(static_cast<MachineInstr *>(I)->getOpcode())) {
+    return false;
+  }
+
+  // Get the last instruction in the block.
+  MachineInstr *LastInst = I;
+
+  // If there is only one terminator instruction, process it.
+  unsigned LastOpc = LastInst->getOpcode();
+  if (I == MBB.begin() ||
+          !isJump(static_cast<MachineInstr *>(--I)->getOpcode())) {
+    if (LastOpc == AMDGPU::JUMP) {
+      TBB = LastInst->getOperand(0).getMBB();
+      return false;
+    } else if (LastOpc == AMDGPU::JUMP_COND) {
+      MachineInstr *predSet = I;
+      while (!isPredicateSetter(predSet->getOpcode())) {
+        predSet = --I;
+      }
+      TBB = LastInst->getOperand(0).getMBB();
+      Cond.push_back(predSet->getOperand(1));
+      Cond.push_back(predSet->getOperand(2));
+      Cond.push_back(MachineOperand::CreateReg(AMDGPU::PRED_SEL_ONE, false));
+      return false;
+    }
+    return true;  // Can't handle indirect branch.
+  }
+
+  // Get the instruction before it if it is a terminator.
+  MachineInstr *SecondLastInst = I;
+  unsigned SecondLastOpc = SecondLastInst->getOpcode();
+
+  // If the block ends with a B and a Bcc, handle it.
+  if (SecondLastOpc == AMDGPU::JUMP_COND && LastOpc == AMDGPU::JUMP) {
+    MachineInstr *predSet = --I;
+    while (!isPredicateSetter(predSet->getOpcode())) {
+      predSet = --I;
+    }
+    TBB = SecondLastInst->getOperand(0).getMBB();
+    FBB = LastInst->getOperand(0).getMBB();
+    Cond.push_back(predSet->getOperand(1));
+    Cond.push_back(predSet->getOperand(2));
+    Cond.push_back(MachineOperand::CreateReg(AMDGPU::PRED_SEL_ONE, false));
+    return false;
+  }
+
+  // Otherwise, can't handle this.
+  return true;
+}
+
+int R600InstrInfo::getBranchInstr(const MachineOperand &op) const {
+  const MachineInstr *MI = op.getParent();
+
+  switch (MI->getDesc().OpInfo->RegClass) {
+  default: // FIXME: fallthrough??
+  case AMDGPU::GPRI32RegClassID: return AMDGPU::BRANCH_COND_i32;
+  case AMDGPU::GPRF32RegClassID: return AMDGPU::BRANCH_COND_f32;
+  };
+}
+
+unsigned
+R600InstrInfo::InsertBranch(MachineBasicBlock &MBB,
+                            MachineBasicBlock *TBB,
+                            MachineBasicBlock *FBB,
+                            const SmallVectorImpl<MachineOperand> &Cond,
+                            DebugLoc DL) const {
+  assert(TBB && "InsertBranch must not be told to insert a fallthrough");
+
+  if (FBB == 0) {
+    if (Cond.empty()) {
+      BuildMI(&MBB, DL, get(AMDGPU::JUMP)).addMBB(TBB);
+      return 1;
+    } else {
+      MachineInstr *PredSet = findFirstPredicateSetterFrom(MBB, MBB.end());
+      assert(PredSet && "No previous predicate !");
+      addFlag(PredSet, 0, MO_FLAG_PUSH);
+      PredSet->getOperand(2).setImm(Cond[1].getImm());
+
+      BuildMI(&MBB, DL, get(AMDGPU::JUMP_COND))
+             .addMBB(TBB)
+             .addReg(AMDGPU::PREDICATE_BIT, RegState::Kill);
+      return 1;
+    }
+  } else {
+    MachineInstr *PredSet = findFirstPredicateSetterFrom(MBB, MBB.end());
+    assert(PredSet && "No previous predicate !");
+    addFlag(PredSet, 0, MO_FLAG_PUSH);
+    PredSet->getOperand(2).setImm(Cond[1].getImm());
+    BuildMI(&MBB, DL, get(AMDGPU::JUMP_COND))
+            .addMBB(TBB)
+            .addReg(AMDGPU::PREDICATE_BIT, RegState::Kill);
+    BuildMI(&MBB, DL, get(AMDGPU::JUMP)).addMBB(FBB);
+    return 2;
+  }
+}
+
+unsigned
+R600InstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
+
+  // Note : we leave PRED* instructions there.
+  // They may be needed when predicating instructions.
+
+  MachineBasicBlock::iterator I = MBB.end();
+
+  if (I == MBB.begin()) {
+    return 0;
+  }
+  --I;
+  switch (I->getOpcode()) {
+  default:
+    return 0;
+  case AMDGPU::JUMP_COND: {
+    MachineInstr *predSet = findFirstPredicateSetterFrom(MBB, I);
+    clearFlag(predSet, 0, MO_FLAG_PUSH);
+    I->eraseFromParent();
+    break;
+  }
+  case AMDGPU::JUMP:
+    I->eraseFromParent();
+    break;
+  }
+  I = MBB.end();
+
+  if (I == MBB.begin()) {
+    return 1;
+  }
+  --I;
+  switch (I->getOpcode()) {
+    // FIXME: only one case??
+  default:
+    return 1;
+  case AMDGPU::JUMP_COND: {
+    MachineInstr *predSet = findFirstPredicateSetterFrom(MBB, I);
+    clearFlag(predSet, 0, MO_FLAG_PUSH);
+    I->eraseFromParent();
+    break;
+  }
+  case AMDGPU::JUMP:
+    I->eraseFromParent();
+    break;
+  }
+  return 2;
+}
+
+bool
+R600InstrInfo::isPredicated(const MachineInstr *MI) const {
+  int idx = MI->findFirstPredOperandIdx();
+  if (idx < 0)
+    return false;
+
+  unsigned Reg = MI->getOperand(idx).getReg();
+  switch (Reg) {
+  default: return false;
+  case AMDGPU::PRED_SEL_ONE:
+  case AMDGPU::PRED_SEL_ZERO:
+  case AMDGPU::PREDICATE_BIT:
+    return true;
+  }
+}
+
+bool
+R600InstrInfo::isPredicable(MachineInstr *MI) const {
+  // XXX: KILL* instructions can be predicated, but they must be the last
+  // instruction in a clause, so this means any instructions after them cannot
+  // be predicated.  Until we have proper support for instruction clauses in the
+  // backend, we will mark KILL* instructions as unpredicable.
+
+  if (MI->getOpcode() == AMDGPU::KILLGT) {
+    return false;
+  } else if (isVector(*MI)) {
+    return false;
+  } else {
+    return AMDGPUInstrInfo::isPredicable(MI);
+  }
+}
+
+
+bool
+R600InstrInfo::isProfitableToIfCvt(MachineBasicBlock &MBB,
+                                   unsigned NumCyles,
+                                   unsigned ExtraPredCycles,
+                                   const BranchProbability &Probability) const{
+  return true;
+}
+
+bool
+R600InstrInfo::isProfitableToIfCvt(MachineBasicBlock &TMBB,
+                                   unsigned NumTCycles,
+                                   unsigned ExtraTCycles,
+                                   MachineBasicBlock &FMBB,
+                                   unsigned NumFCycles,
+                                   unsigned ExtraFCycles,
+                                   const BranchProbability &Probability) const {
+  return true;
+}
+
+bool
+R600InstrInfo::isProfitableToDupForIfCvt(MachineBasicBlock &MBB,
+                                         unsigned NumCyles,
+                                         const BranchProbability &Probability)
+                                         const {
+  return true;
+}
+
+bool
+R600InstrInfo::isProfitableToUnpredicate(MachineBasicBlock &TMBB,
+                                         MachineBasicBlock &FMBB) const {
+  return false;
+}
+
+
+bool
+R600InstrInfo::ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const {
+  MachineOperand &MO = Cond[1];
+  switch (MO.getImm()) {
+  case OPCODE_IS_ZERO_INT:
+    MO.setImm(OPCODE_IS_NOT_ZERO_INT);
+    break;
+  case OPCODE_IS_NOT_ZERO_INT:
+    MO.setImm(OPCODE_IS_ZERO_INT);
+    break;
+  case OPCODE_IS_ZERO:
+    MO.setImm(OPCODE_IS_NOT_ZERO);
+    break;
+  case OPCODE_IS_NOT_ZERO:
+    MO.setImm(OPCODE_IS_ZERO);
+    break;
+  default:
+    return true;
+  }
+
+  MachineOperand &MO2 = Cond[2];
+  switch (MO2.getReg()) {
+  case AMDGPU::PRED_SEL_ZERO:
+    MO2.setReg(AMDGPU::PRED_SEL_ONE);
+    break;
+  case AMDGPU::PRED_SEL_ONE:
+    MO2.setReg(AMDGPU::PRED_SEL_ZERO);
+    break;
+  default:
+    return true;
+  }
+  return false;
+}
+
+bool
+R600InstrInfo::DefinesPredicate(MachineInstr *MI,
+                                std::vector<MachineOperand> &Pred) const {
+  return isPredicateSetter(MI->getOpcode());
+}
+
+
+bool
+R600InstrInfo::SubsumesPredicate(const SmallVectorImpl<MachineOperand> &Pred1,
+                       const SmallVectorImpl<MachineOperand> &Pred2) const {
+  return false;
+}
+
+
+bool
+R600InstrInfo::PredicateInstruction(MachineInstr *MI,
+                      const SmallVectorImpl<MachineOperand> &Pred) const {
+  int PIdx = MI->findFirstPredOperandIdx();
+
+  if (PIdx != -1) {
+    MachineOperand &PMO = MI->getOperand(PIdx);
+    PMO.setReg(Pred[2].getReg());
+    MachineInstrBuilder MIB(*MI->getParent()->getParent(), MI);
+    MIB.addReg(AMDGPU::PREDICATE_BIT, RegState::Implicit);
+    return true;
+  }
+
+  return false;
+}
+
+unsigned int R600InstrInfo::getInstrLatency(const InstrItineraryData *ItinData,
+                                            const MachineInstr *MI,
+                                            unsigned *PredCost) const {
+  if (PredCost)
+    *PredCost = 2;
+  return 2;
+}
+
+int R600InstrInfo::getIndirectIndexBegin(const MachineFunction &MF) const {
+  const MachineRegisterInfo &MRI = MF.getRegInfo();
+  const MachineFrameInfo *MFI = MF.getFrameInfo();
+  int Offset = 0;
+
+  if (MFI->getNumObjects() == 0) {
+    return -1;
+  }
+
+  if (MRI.livein_empty()) {
+    return 0;
+  }
+
+  for (MachineRegisterInfo::livein_iterator LI = MRI.livein_begin(),
+                                            LE = MRI.livein_end();
+                                            LI != LE; ++LI) {
+    Offset = std::max(Offset,
+                      GET_REG_INDEX(RI.getEncodingValue(LI->first)));
+  }
+
+  return Offset + 1;
+}
+
+int R600InstrInfo::getIndirectIndexEnd(const MachineFunction &MF) const {
+  int Offset = 0;
+  const MachineFrameInfo *MFI = MF.getFrameInfo();
+
+  // Variable sized objects are not supported
+  assert(!MFI->hasVarSizedObjects());
+
+  if (MFI->getNumObjects() == 0) {
+    return -1;
+  }
+
+  Offset = TM.getFrameLowering()->getFrameIndexOffset(MF, -1);
+
+  return getIndirectIndexBegin(MF) + Offset;
+}
+
+std::vector<unsigned> R600InstrInfo::getIndirectReservedRegs(
+                                             const MachineFunction &MF) const {
+  const AMDGPUFrameLowering *TFL =
+                 static_cast<const AMDGPUFrameLowering*>(TM.getFrameLowering());
+  std::vector<unsigned> Regs;
+
+  unsigned StackWidth = TFL->getStackWidth(MF);
+  int End = getIndirectIndexEnd(MF);
+
+  if (End == -1) {
+    return Regs;
+  }
+
+  for (int Index = getIndirectIndexBegin(MF); Index <= End; ++Index) {
+    unsigned SuperReg = AMDGPU::R600_Reg128RegClass.getRegister(Index);
+    Regs.push_back(SuperReg);
+    for (unsigned Chan = 0; Chan < StackWidth; ++Chan) {
+      unsigned Reg = AMDGPU::R600_TReg32RegClass.getRegister((4 * Index) + Chan);
+      Regs.push_back(Reg);
+    }
+  }
+  return Regs;
+}
+
+unsigned R600InstrInfo::calculateIndirectAddress(unsigned RegIndex,
+                                                 unsigned Channel) const {
+  // XXX: Remove when we support a stack width > 2
+  assert(Channel == 0);
+  return RegIndex;
+}
+
+const TargetRegisterClass * R600InstrInfo::getIndirectAddrStoreRegClass(
+                                                     unsigned SourceReg) const {
+  return &AMDGPU::R600_TReg32RegClass;
+}
+
+const TargetRegisterClass *R600InstrInfo::getIndirectAddrLoadRegClass() const {
+  return &AMDGPU::TRegMemRegClass;
+}
+
+MachineInstrBuilder R600InstrInfo::buildIndirectWrite(MachineBasicBlock *MBB,
+                                       MachineBasicBlock::iterator I,
+                                       unsigned ValueReg, unsigned Address,
+                                       unsigned OffsetReg) const {
+  unsigned AddrReg = AMDGPU::R600_AddrRegClass.getRegister(Address);
+  MachineInstr *MOVA = buildDefaultInstruction(*MBB, I, AMDGPU::MOVA_INT_eg,
+                                               AMDGPU::AR_X, OffsetReg);
+  setImmOperand(MOVA, R600Operands::WRITE, 0);
+
+  MachineInstrBuilder Mov = buildDefaultInstruction(*MBB, I, AMDGPU::MOV,
+                                      AddrReg, ValueReg)
+                                      .addReg(AMDGPU::AR_X, RegState::Implicit);
+  setImmOperand(Mov, R600Operands::DST_REL, 1);
+  return Mov;
+}
+
+MachineInstrBuilder R600InstrInfo::buildIndirectRead(MachineBasicBlock *MBB,
+                                       MachineBasicBlock::iterator I,
+                                       unsigned ValueReg, unsigned Address,
+                                       unsigned OffsetReg) const {
+  unsigned AddrReg = AMDGPU::R600_AddrRegClass.getRegister(Address);
+  MachineInstr *MOVA = buildDefaultInstruction(*MBB, I, AMDGPU::MOVA_INT_eg,
+                                                       AMDGPU::AR_X,
+                                                       OffsetReg);
+  setImmOperand(MOVA, R600Operands::WRITE, 0);
+  MachineInstrBuilder Mov = buildDefaultInstruction(*MBB, I, AMDGPU::MOV,
+                                      ValueReg,
+                                      AddrReg)
+                                      .addReg(AMDGPU::AR_X, RegState::Implicit);
+  setImmOperand(Mov, R600Operands::SRC0_REL, 1);
+
+  return Mov;
+}
+
+const TargetRegisterClass *R600InstrInfo::getSuperIndirectRegClass() const {
+  return &AMDGPU::IndirectRegRegClass;
+}
+
+unsigned R600InstrInfo::getMaxAlusPerClause() const {
+  return 115;
+}
+
+MachineInstrBuilder R600InstrInfo::buildDefaultInstruction(MachineBasicBlock &MBB,
+                                                  MachineBasicBlock::iterator I,
+                                                  unsigned Opcode,
+                                                  unsigned DstReg,
+                                                  unsigned Src0Reg,
+                                                  unsigned Src1Reg) const {
+  MachineInstrBuilder MIB = BuildMI(MBB, I, MBB.findDebugLoc(I), get(Opcode),
+    DstReg);           // $dst
+
+  if (Src1Reg) {
+    MIB.addImm(0)     // $update_exec_mask
+       .addImm(0);    // $update_predicate
+  }
+  MIB.addImm(1)        // $write
+     .addImm(0)        // $omod
+     .addImm(0)        // $dst_rel
+     .addImm(0)        // $dst_clamp
+     .addReg(Src0Reg)  // $src0
+     .addImm(0)        // $src0_neg
+     .addImm(0)        // $src0_rel
+     .addImm(0)        // $src0_abs
+     .addImm(-1);       // $src0_sel
+
+  if (Src1Reg) {
+    MIB.addReg(Src1Reg) // $src1
+       .addImm(0)       // $src1_neg
+       .addImm(0)       // $src1_rel
+       .addImm(0)       // $src1_abs
+       .addImm(-1);      // $src1_sel
+  }
+
+  //XXX: The r600g finalizer expects this to be 1, once we've moved the
+  //scheduling to the backend, we can change the default to 0.
+  MIB.addImm(1)        // $last
+      .addReg(AMDGPU::PRED_SEL_OFF) // $pred_sel
+      .addImm(0);        // $literal
+
+  return MIB;
+}
+
+MachineInstr *R600InstrInfo::buildMovImm(MachineBasicBlock &BB,
+                                         MachineBasicBlock::iterator I,
+                                         unsigned DstReg,
+                                         uint64_t Imm) const {
+  MachineInstr *MovImm = buildDefaultInstruction(BB, I, AMDGPU::MOV, DstReg,
+                                                  AMDGPU::ALU_LITERAL_X);
+  setImmOperand(MovImm, R600Operands::IMM, Imm);
+  return MovImm;
+}
+
+int R600InstrInfo::getOperandIdx(const MachineInstr &MI,
+                                 R600Operands::Ops Op) const {
+  return getOperandIdx(MI.getOpcode(), Op);
+}
+
+int R600InstrInfo::getOperandIdx(unsigned Opcode,
+                                 R600Operands::Ops Op) const {
+  unsigned TargetFlags = get(Opcode).TSFlags;
+  unsigned OpTableIdx;
+
+  if (!HAS_NATIVE_OPERANDS(TargetFlags)) {
+    switch (Op) {
+    case R600Operands::DST: return 0;
+    case R600Operands::SRC0: return 1;
+    case R600Operands::SRC1: return 2;
+    case R600Operands::SRC2: return 3;
+    default:
+      assert(!"Unknown operand type for instruction");
+      return -1;
+    }
+  }
+
+  if (TargetFlags & R600_InstFlag::OP1) {
+    OpTableIdx = 0;
+  } else if (TargetFlags & R600_InstFlag::OP2) {
+    OpTableIdx = 1;
+  } else {
+    assert((TargetFlags & R600_InstFlag::OP3) && "OP1, OP2, or OP3 not defined "
+                                                 "for this instruction");
+    OpTableIdx = 2;
+  }
+
+  return R600Operands::ALUOpTable[OpTableIdx][Op];
+}
+
+void R600InstrInfo::setImmOperand(MachineInstr *MI, R600Operands::Ops Op,
+                                  int64_t Imm) const {
+  int Idx = getOperandIdx(*MI, Op);
+  assert(Idx != -1 && "Operand not supported for this instruction.");
+  assert(MI->getOperand(Idx).isImm());
+  MI->getOperand(Idx).setImm(Imm);
+}
+
+//===----------------------------------------------------------------------===//
+// Instruction flag getters/setters
+//===----------------------------------------------------------------------===//
+
+bool R600InstrInfo::hasFlagOperand(const MachineInstr &MI) const {
+  return GET_FLAG_OPERAND_IDX(get(MI.getOpcode()).TSFlags) != 0;
+}
+
+MachineOperand &R600InstrInfo::getFlagOp(MachineInstr *MI, unsigned SrcIdx,
+                                         unsigned Flag) const {
+  unsigned TargetFlags = get(MI->getOpcode()).TSFlags;
+  int FlagIndex = 0;
+  if (Flag != 0) {
+    // If we pass something other than the default value of Flag to this
+    // function, it means we are want to set a flag on an instruction
+    // that uses native encoding.
+    assert(HAS_NATIVE_OPERANDS(TargetFlags));
+    bool IsOP3 = (TargetFlags & R600_InstFlag::OP3) == R600_InstFlag::OP3;
+    switch (Flag) {
+    case MO_FLAG_CLAMP:
+      FlagIndex = getOperandIdx(*MI, R600Operands::CLAMP);
+      break;
+    case MO_FLAG_MASK:
+      FlagIndex = getOperandIdx(*MI, R600Operands::WRITE);
+      break;
+    case MO_FLAG_NOT_LAST:
+    case MO_FLAG_LAST:
+      FlagIndex = getOperandIdx(*MI, R600Operands::LAST);
+      break;
+    case MO_FLAG_NEG:
+      switch (SrcIdx) {
+      case 0: FlagIndex = getOperandIdx(*MI, R600Operands::SRC0_NEG); break;
+      case 1: FlagIndex = getOperandIdx(*MI, R600Operands::SRC1_NEG); break;
+      case 2: FlagIndex = getOperandIdx(*MI, R600Operands::SRC2_NEG); break;
+      }
+      break;
+
+    case MO_FLAG_ABS:
+      assert(!IsOP3 && "Cannot set absolute value modifier for OP3 "
+                       "instructions.");
+      (void)IsOP3;
+      switch (SrcIdx) {
+      case 0: FlagIndex = getOperandIdx(*MI, R600Operands::SRC0_ABS); break;
+      case 1: FlagIndex = getOperandIdx(*MI, R600Operands::SRC1_ABS); break;
+      }
+      break;
+
+    default:
+      FlagIndex = -1;
+      break;
+    }
+    assert(FlagIndex != -1 && "Flag not supported for this instruction");
+  } else {
+      FlagIndex = GET_FLAG_OPERAND_IDX(TargetFlags);
+      assert(FlagIndex != 0 &&
+         "Instruction flags not supported for this instruction");
+  }
+
+  MachineOperand &FlagOp = MI->getOperand(FlagIndex);
+  assert(FlagOp.isImm());
+  return FlagOp;
+}
+
+void R600InstrInfo::addFlag(MachineInstr *MI, unsigned Operand,
+                            unsigned Flag) const {
+  unsigned TargetFlags = get(MI->getOpcode()).TSFlags;
+  if (Flag == 0) {
+    return;
+  }
+  if (HAS_NATIVE_OPERANDS(TargetFlags)) {
+    MachineOperand &FlagOp = getFlagOp(MI, Operand, Flag);
+    if (Flag == MO_FLAG_NOT_LAST) {
+      clearFlag(MI, Operand, MO_FLAG_LAST);
+    } else if (Flag == MO_FLAG_MASK) {
+      clearFlag(MI, Operand, Flag);
+    } else {
+      FlagOp.setImm(1);
+    }
+  } else {
+      MachineOperand &FlagOp = getFlagOp(MI, Operand);
+      FlagOp.setImm(FlagOp.getImm() | (Flag << (NUM_MO_FLAGS * Operand)));
+  }
+}
+
+void R600InstrInfo::clearFlag(MachineInstr *MI, unsigned Operand,
+                              unsigned Flag) const {
+  unsigned TargetFlags = get(MI->getOpcode()).TSFlags;
+  if (HAS_NATIVE_OPERANDS(TargetFlags)) {
+    MachineOperand &FlagOp = getFlagOp(MI, Operand, Flag);
+    FlagOp.setImm(0);
+  } else {
+    MachineOperand &FlagOp = getFlagOp(MI);
+    unsigned InstFlags = FlagOp.getImm();
+    InstFlags &= ~(Flag << (NUM_MO_FLAGS * Operand));
+    FlagOp.setImm(InstFlags);
+  }
+}
diff --git a/lib/Target/R600/R600InstrInfo.h b/lib/Target/R600/R600InstrInfo.h
new file mode 100644
index 000000000000..dbae90013d22
--- /dev/null
+++ b/lib/Target/R600/R600InstrInfo.h
@@ -0,0 +1,204 @@
+//===-- R600InstrInfo.h - R600 Instruction Info Interface -------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Interface definition for R600InstrInfo
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef R600INSTRUCTIONINFO_H_
+#define R600INSTRUCTIONINFO_H_
+
+#include "AMDGPUInstrInfo.h"
+#include "AMDIL.h"
+#include "R600Defines.h"
+#include "R600RegisterInfo.h"
+#include <map>
+
+namespace llvm {
+
+  class AMDGPUTargetMachine;
+  class DFAPacketizer;
+  class ScheduleDAG;
+  class MachineFunction;
+  class MachineInstr;
+  class MachineInstrBuilder;
+
+  class R600InstrInfo : public AMDGPUInstrInfo {
+  private:
+  const R600RegisterInfo RI;
+
+  int getBranchInstr(const MachineOperand &op) const;
+
+  public:
+  explicit R600InstrInfo(AMDGPUTargetMachine &tm);
+
+  const R600RegisterInfo &getRegisterInfo() const;
+  virtual void copyPhysReg(MachineBasicBlock &MBB,
+                           MachineBasicBlock::iterator MI, DebugLoc DL,
+                           unsigned DestReg, unsigned SrcReg,
+                           bool KillSrc) const;
+
+  bool isTrig(const MachineInstr &MI) const;
+  bool isPlaceHolderOpcode(unsigned opcode) const;
+  bool isReductionOp(unsigned opcode) const;
+  bool isCubeOp(unsigned opcode) const;
+
+  /// \returns true if this \p Opcode represents an ALU instruction.
+  bool isALUInstr(unsigned Opcode) const;
+
+  bool fitsConstReadLimitations(const std::vector<unsigned>&) const;
+  bool canBundle(const std::vector<MachineInstr *> &) const;
+
+  /// \breif Vector instructions are instructions that must fill all
+  /// instruction slots within an instruction group.
+  bool isVector(const MachineInstr &MI) const;
+
+  virtual MachineInstr * getMovImmInstr(MachineFunction *MF, unsigned DstReg,
+                                        int64_t Imm) const;
+
+  virtual unsigned getIEQOpcode() const;
+  virtual bool isMov(unsigned Opcode) const;
+
+  DFAPacketizer *CreateTargetScheduleState(const TargetMachine *TM,
+                                           const ScheduleDAG *DAG) const;
+
+  bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const;
+
+  bool AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB, MachineBasicBlock *&FBB,
+                     SmallVectorImpl<MachineOperand> &Cond, bool AllowModify) const;
+
+  unsigned InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB, MachineBasicBlock *FBB, const SmallVectorImpl<MachineOperand> &Cond, DebugLoc DL) const;
+
+  unsigned RemoveBranch(MachineBasicBlock &MBB) const;
+
+  bool isPredicated(const MachineInstr *MI) const;
+
+  bool isPredicable(MachineInstr *MI) const;
+
+  bool
+   isProfitableToDupForIfCvt(MachineBasicBlock &MBB, unsigned NumCyles,
+                             const BranchProbability &Probability) const;
+
+  bool isProfitableToIfCvt(MachineBasicBlock &MBB, unsigned NumCyles,
+                           unsigned ExtraPredCycles,
+                           const BranchProbability &Probability) const ;
+
+  bool
+   isProfitableToIfCvt(MachineBasicBlock &TMBB,
+                       unsigned NumTCycles, unsigned ExtraTCycles,
+                       MachineBasicBlock &FMBB,
+                       unsigned NumFCycles, unsigned ExtraFCycles,
+                       const BranchProbability &Probability) const;
+
+  bool DefinesPredicate(MachineInstr *MI,
+                                  std::vector<MachineOperand> &Pred) const;
+
+  bool SubsumesPredicate(const SmallVectorImpl<MachineOperand> &Pred1,
+                         const SmallVectorImpl<MachineOperand> &Pred2) const;
+
+  bool isProfitableToUnpredicate(MachineBasicBlock &TMBB,
+                                          MachineBasicBlock &FMBB) const;
+
+  bool PredicateInstruction(MachineInstr *MI,
+                        const SmallVectorImpl<MachineOperand> &Pred) const;
+
+  unsigned int getInstrLatency(const InstrItineraryData *ItinData,
+                               const MachineInstr *MI,
+                               unsigned *PredCost = 0) const;
+
+  virtual int getInstrLatency(const InstrItineraryData *ItinData,
+                              SDNode *Node) const { return 1;}
+
+  /// \returns a list of all the registers that may be accesed using indirect
+  /// addressing.
+  std::vector<unsigned> getIndirectReservedRegs(const MachineFunction &MF) const;
+
+  virtual int getIndirectIndexBegin(const MachineFunction &MF) const;
+
+  virtual int getIndirectIndexEnd(const MachineFunction &MF) const;
+
+
+  virtual unsigned calculateIndirectAddress(unsigned RegIndex,
+                                            unsigned Channel) const;
+
+  virtual const TargetRegisterClass *getIndirectAddrStoreRegClass(
+                                                      unsigned SourceReg) const;
+
+  virtual const TargetRegisterClass *getIndirectAddrLoadRegClass() const;
+
+  virtual MachineInstrBuilder buildIndirectWrite(MachineBasicBlock *MBB,
+                                  MachineBasicBlock::iterator I,
+                                  unsigned ValueReg, unsigned Address,
+                                  unsigned OffsetReg) const;
+
+  virtual MachineInstrBuilder buildIndirectRead(MachineBasicBlock *MBB,
+                                  MachineBasicBlock::iterator I,
+                                  unsigned ValueReg, unsigned Address,
+                                  unsigned OffsetReg) const;
+
+  virtual const TargetRegisterClass *getSuperIndirectRegClass() const;
+
+  unsigned getMaxAlusPerClause() const;
+
+  ///buildDefaultInstruction - This function returns a MachineInstr with
+  /// all the instruction modifiers initialized to their default values.
+  /// You can use this function to avoid manually specifying each instruction
+  /// modifier operand when building a new instruction.
+  ///
+  /// \returns a MachineInstr with all the instruction modifiers initialized
+  /// to their default values.
+  MachineInstrBuilder buildDefaultInstruction(MachineBasicBlock &MBB,
+                                              MachineBasicBlock::iterator I,
+                                              unsigned Opcode,
+                                              unsigned DstReg,
+                                              unsigned Src0Reg,
+                                              unsigned Src1Reg = 0) const;
+
+  MachineInstr *buildMovImm(MachineBasicBlock &BB,
+                                  MachineBasicBlock::iterator I,
+                                  unsigned DstReg,
+                                  uint64_t Imm) const;
+
+  /// \brief Get the index of Op in the MachineInstr.
+  ///
+  /// \returns -1 if the Instruction does not contain the specified \p Op.
+  int getOperandIdx(const MachineInstr &MI, R600Operands::Ops Op) const;
+
+  /// \brief Get the index of \p Op for the given Opcode.
+  ///
+  /// \returns -1 if the Instruction does not contain the specified \p Op.
+  int getOperandIdx(unsigned Opcode, R600Operands::Ops Op) const;
+
+  /// \brief Helper function for setting instruction flag values.
+  void setImmOperand(MachineInstr *MI, R600Operands::Ops Op, int64_t Imm) const;
+
+  /// \returns true if this instruction has an operand for storing target flags.
+  bool hasFlagOperand(const MachineInstr &MI) const;
+
+  ///\brief Add one of the MO_FLAG* flags to the specified \p Operand.
+  void addFlag(MachineInstr *MI, unsigned Operand, unsigned Flag) const;
+
+  ///\brief Determine if the specified \p Flag is set on this \p Operand.
+  bool isFlagSet(const MachineInstr &MI, unsigned Operand, unsigned Flag) const;
+
+  /// \param SrcIdx The register source to set the flag on (e.g src0, src1, src2)
+  /// \param Flag The flag being set.
+  ///
+  /// \returns the operand containing the flags for this instruction.
+  MachineOperand &getFlagOp(MachineInstr *MI, unsigned SrcIdx = 0,
+                            unsigned Flag = 0) const;
+
+  /// \brief Clear the specified flag on the instruction.
+  void clearFlag(MachineInstr *MI, unsigned Operand, unsigned Flag) const;
+};
+
+} // End llvm namespace
+
+#endif // R600INSTRINFO_H_
diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td
new file mode 100644
index 000000000000..663b41a66d6f
--- /dev/null
+++ b/lib/Target/R600/R600Instructions.td
@@ -0,0 +1,2267 @@
+//===-- R600Instructions.td - R600 Instruction defs  -------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// R600 Tablegen instruction definitions
+//
+//===----------------------------------------------------------------------===//
+
+include "R600Intrinsics.td"
+
+class InstR600 <bits<11> inst, dag outs, dag ins, string asm, list<dag> pattern,
+                InstrItinClass itin>
+    : AMDGPUInst <outs, ins, asm, pattern> {
+
+  field bits<64> Inst;
+  bit Trig = 0;
+  bit Op3 = 0;
+  bit isVector = 0;
+  bits<2> FlagOperandIdx = 0;
+  bit Op1 = 0;
+  bit Op2 = 0;
+  bit HasNativeOperands = 0;
+
+  bits<11> op_code = inst;
+  //let Inst = inst;
+  let Namespace = "AMDGPU";
+  let OutOperandList = outs;
+  let InOperandList = ins;
+  let AsmString = asm;
+  let Pattern = pattern;
+  let Itinerary = itin;
+
+  let TSFlags{4} = Trig;
+  let TSFlags{5} = Op3;
+
+  // Vector instructions are instructions that must fill all slots in an
+  // instruction group
+  let TSFlags{6} = isVector;
+  let TSFlags{8-7} = FlagOperandIdx;
+  let TSFlags{9} = HasNativeOperands;
+  let TSFlags{10} = Op1;
+  let TSFlags{11} = Op2;
+}
+
+class InstR600ISA <dag outs, dag ins, string asm, list<dag> pattern> :
+    AMDGPUInst <outs, ins, asm, pattern> {
+  field bits<64> Inst;
+
+  let Namespace = "AMDGPU";
+}
+
+def MEMxi : Operand<iPTR> {
+  let MIOperandInfo = (ops R600_TReg32_X:$ptr, i32imm:$index);
+  let PrintMethod = "printMemOperand";
+}
+
+def MEMrr : Operand<iPTR> {
+  let MIOperandInfo = (ops R600_Reg32:$ptr, R600_Reg32:$index);
+}
+
+// Operands for non-registers
+
+class InstFlag<string PM = "printOperand", int Default = 0>
+    : OperandWithDefaultOps <i32, (ops (i32 Default))> {
+  let PrintMethod = PM;
+}
+
+// src_sel for ALU src operands, see also ALU_CONST, ALU_PARAM registers
+def SEL : OperandWithDefaultOps <i32, (ops (i32 -1))> {
+  let PrintMethod = "printSel";
+}
+
+def LITERAL : InstFlag<"printLiteral">;
+
+def WRITE : InstFlag <"printWrite", 1>;
+def OMOD : InstFlag <"printOMOD">;
+def REL : InstFlag <"printRel">;
+def CLAMP : InstFlag <"printClamp">;
+def NEG : InstFlag <"printNeg">;
+def ABS : InstFlag <"printAbs">;
+def UEM : InstFlag <"printUpdateExecMask">;
+def UP : InstFlag <"printUpdatePred">;
+
+// XXX: The r600g finalizer in Mesa expects last to be one in most cases.
+// Once we start using the packetizer in this backend we should have this
+// default to 0.
+def LAST : InstFlag<"printLast", 1>;
+
+def FRAMEri : Operand<iPTR> {
+  let MIOperandInfo = (ops R600_Reg32:$ptr, i32imm:$index);
+}
+
+def ADDRParam : ComplexPattern<i32, 2, "SelectADDRParam", [], []>;
+def ADDRDWord : ComplexPattern<i32, 1, "SelectADDRDWord", [], []>;
+def ADDRVTX_READ : ComplexPattern<i32, 2, "SelectADDRVTX_READ", [], []>;
+def ADDRGA_CONST_OFFSET : ComplexPattern<i32, 1, "SelectGlobalValueConstantOffset", [], []>;
+def ADDRGA_VAR_OFFSET : ComplexPattern<i32, 2, "SelectGlobalValueVariableOffset", [], []>;
+def ADDRIndirect : ComplexPattern<iPTR, 2, "SelectADDRIndirect", [], []>;
+
+class R600ALU_Word0 {
+  field bits<32> Word0;
+
+  bits<11> src0;
+  bits<1>  src0_neg;
+  bits<1>  src0_rel;
+  bits<11> src1;
+  bits<1>  src1_rel;
+  bits<1>  src1_neg;
+  bits<3>  index_mode = 0;
+  bits<2>  pred_sel;
+  bits<1>  last;
+
+  bits<9>  src0_sel  = src0{8-0};
+  bits<2>  src0_chan = src0{10-9};
+  bits<9>  src1_sel  = src1{8-0};
+  bits<2>  src1_chan = src1{10-9};
+
+  let Word0{8-0}   = src0_sel;
+  let Word0{9}     = src0_rel;
+  let Word0{11-10} = src0_chan;
+  let Word0{12}    = src0_neg;
+  let Word0{21-13} = src1_sel;
+  let Word0{22}    = src1_rel;
+  let Word0{24-23} = src1_chan;
+  let Word0{25}    = src1_neg;
+  let Word0{28-26} = index_mode;
+  let Word0{30-29} = pred_sel;
+  let Word0{31}    = last;
+}
+
+class R600ALU_Word1 {
+  field bits<32> Word1;
+
+  bits<11> dst;
+  bits<3>  bank_swizzle = 0;
+  bits<1>  dst_rel;
+  bits<1>  clamp;
+
+  bits<7>  dst_sel  = dst{6-0};
+  bits<2>  dst_chan = dst{10-9};
+
+  let Word1{20-18} = bank_swizzle;
+  let Word1{27-21} = dst_sel;
+  let Word1{28}    = dst_rel;
+  let Word1{30-29} = dst_chan;
+  let Word1{31}    = clamp;
+}
+
+class R600ALU_Word1_OP2 <bits<11> alu_inst> : R600ALU_Word1{
+
+  bits<1>  src0_abs;
+  bits<1>  src1_abs;
+  bits<1>  update_exec_mask;
+  bits<1>  update_pred;
+  bits<1>  write;
+  bits<2>  omod;
+
+  let Word1{0}     = src0_abs;
+  let Word1{1}     = src1_abs;
+  let Word1{2}     = update_exec_mask;
+  let Word1{3}     = update_pred;
+  let Word1{4}     = write;
+  let Word1{6-5}   = omod;
+  let Word1{17-7}  = alu_inst;
+}
+
+class R600ALU_Word1_OP3 <bits<5> alu_inst> : R600ALU_Word1{
+
+  bits<11> src2;
+  bits<1>  src2_rel;
+  bits<1>  src2_neg;
+
+  bits<9>  src2_sel = src2{8-0};
+  bits<2>  src2_chan = src2{10-9};
+
+  let Word1{8-0}   = src2_sel;
+  let Word1{9}     = src2_rel;
+  let Word1{11-10} = src2_chan;
+  let Word1{12}    = src2_neg;
+  let Word1{17-13} = alu_inst;
+}
+
+class VTX_WORD0 {
+  field bits<32> Word0;
+  bits<7> SRC_GPR;
+  bits<5> VC_INST;
+  bits<2> FETCH_TYPE;
+  bits<1> FETCH_WHOLE_QUAD;
+  bits<8> BUFFER_ID;
+  bits<1> SRC_REL;
+  bits<2> SRC_SEL_X;
+  bits<6> MEGA_FETCH_COUNT;
+
+  let Word0{4-0}   = VC_INST;
+  let Word0{6-5}   = FETCH_TYPE;
+  let Word0{7}     = FETCH_WHOLE_QUAD;
+  let Word0{15-8}  = BUFFER_ID;
+  let Word0{22-16} = SRC_GPR;
+  let Word0{23}    = SRC_REL;
+  let Word0{25-24} = SRC_SEL_X;
+  let Word0{31-26} = MEGA_FETCH_COUNT;
+}
+
+class VTX_WORD1_GPR {
+  field bits<32> Word1;
+  bits<7> DST_GPR;
+  bits<1> DST_REL;
+  bits<3> DST_SEL_X;
+  bits<3> DST_SEL_Y;
+  bits<3> DST_SEL_Z;
+  bits<3> DST_SEL_W;
+  bits<1> USE_CONST_FIELDS;
+  bits<6> DATA_FORMAT;
+  bits<2> NUM_FORMAT_ALL;
+  bits<1> FORMAT_COMP_ALL;
+  bits<1> SRF_MODE_ALL;
+
+  let Word1{6-0} = DST_GPR;
+  let Word1{7}    = DST_REL;
+  let Word1{8}    = 0; // Reserved
+  let Word1{11-9} = DST_SEL_X;
+  let Word1{14-12} = DST_SEL_Y;
+  let Word1{17-15} = DST_SEL_Z;
+  let Word1{20-18} = DST_SEL_W;
+  let Word1{21}    = USE_CONST_FIELDS;
+  let Word1{27-22} = DATA_FORMAT;
+  let Word1{29-28} = NUM_FORMAT_ALL;
+  let Word1{30}    = FORMAT_COMP_ALL;
+  let Word1{31}    = SRF_MODE_ALL;
+}
+
+class TEX_WORD0 {
+  field bits<32> Word0;
+
+  bits<5> TEX_INST;
+  bits<2> INST_MOD;
+  bits<1> FETCH_WHOLE_QUAD;
+  bits<8> RESOURCE_ID;
+  bits<7> SRC_GPR;
+  bits<1> SRC_REL;
+  bits<1> ALT_CONST;
+  bits<2> RESOURCE_INDEX_MODE;
+  bits<2> SAMPLER_INDEX_MODE;
+
+  let Word0{4-0} = TEX_INST;
+  let Word0{6-5} = INST_MOD;
+  let Word0{7} = FETCH_WHOLE_QUAD;
+  let Word0{15-8} = RESOURCE_ID;
+  let Word0{22-16} = SRC_GPR;
+  let Word0{23} = SRC_REL;
+  let Word0{24} = ALT_CONST;
+  let Word0{26-25} = RESOURCE_INDEX_MODE;
+  let Word0{28-27} = SAMPLER_INDEX_MODE;
+}
+
+class TEX_WORD1 {
+  field bits<32> Word1;
+
+  bits<7> DST_GPR;
+  bits<1> DST_REL;
+  bits<3> DST_SEL_X;
+  bits<3> DST_SEL_Y;
+  bits<3> DST_SEL_Z;
+  bits<3> DST_SEL_W;
+  bits<7> LOD_BIAS;
+  bits<1> COORD_TYPE_X;
+  bits<1> COORD_TYPE_Y;
+  bits<1> COORD_TYPE_Z;
+  bits<1> COORD_TYPE_W;
+
+  let Word1{6-0} = DST_GPR;
+  let Word1{7} = DST_REL;
+  let Word1{11-9} = DST_SEL_X;
+  let Word1{14-12} = DST_SEL_Y;
+  let Word1{17-15} = DST_SEL_Z;
+  let Word1{20-18} = DST_SEL_W;
+  let Word1{27-21} = LOD_BIAS;
+  let Word1{28} = COORD_TYPE_X;
+  let Word1{29} = COORD_TYPE_Y;
+  let Word1{30} = COORD_TYPE_Z;
+  let Word1{31} = COORD_TYPE_W;
+}
+
+class TEX_WORD2 {
+  field bits<32> Word2;
+
+  bits<5> OFFSET_X;
+  bits<5> OFFSET_Y;
+  bits<5> OFFSET_Z;
+  bits<5> SAMPLER_ID;
+  bits<3> SRC_SEL_X;
+  bits<3> SRC_SEL_Y;
+  bits<3> SRC_SEL_Z;
+  bits<3> SRC_SEL_W;
+
+  let Word2{4-0} = OFFSET_X;
+  let Word2{9-5} = OFFSET_Y;
+  let Word2{14-10} = OFFSET_Z;
+  let Word2{19-15} = SAMPLER_ID;
+  let Word2{22-20} = SRC_SEL_X;
+  let Word2{25-23} = SRC_SEL_Y;
+  let Word2{28-26} = SRC_SEL_Z;
+  let Word2{31-29} = SRC_SEL_W;
+}
+
+/*
+XXX: R600 subtarget uses a slightly different encoding than the other
+subtargets.  We currently handle this in R600MCCodeEmitter, but we may
+want to use these instruction classes in the future.
+
+class R600ALU_Word1_OP2_r600 : R600ALU_Word1_OP2 {
+
+  bits<1>  fog_merge;
+  bits<10> alu_inst;
+
+  let Inst{37}    = fog_merge;
+  let Inst{39-38} = omod;
+  let Inst{49-40} = alu_inst;
+}
+
+class R600ALU_Word1_OP2_r700 : R600ALU_Word1_OP2 {
+
+  bits<11> alu_inst;
+
+  let Inst{38-37} = omod;
+  let Inst{49-39} = alu_inst;
+}
+*/
+
+def R600_Pred : PredicateOperand<i32, (ops R600_Predicate),
+                                     (ops PRED_SEL_OFF)>;
+
+
+let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
+
+// Class for instructions with only one source register.
+// If you add new ins to this instruction, make sure they are listed before
+// $literal, because the backend currently assumes that the last operand is
+// a literal.  Also be sure to update the enum R600Op1OperandIndex::ROI in
+// R600Defines.h, R600InstrInfo::buildDefaultInstruction(),
+// and R600InstrInfo::getOperandIdx().
+class R600_1OP <bits<11> inst, string opName, list<dag> pattern,
+                InstrItinClass itin = AnyALU> :
+    InstR600 <0,
+              (outs R600_Reg32:$dst),
+              (ins WRITE:$write, OMOD:$omod, REL:$dst_rel, CLAMP:$clamp,
+                   R600_Reg32:$src0, NEG:$src0_neg, REL:$src0_rel, ABS:$src0_abs, SEL:$src0_sel,
+                   LAST:$last, R600_Pred:$pred_sel, LITERAL:$literal),
+              !strconcat("  ", opName,
+                   "$clamp $dst$write$dst_rel$omod, "
+                   "$src0_neg$src0_abs$src0$src0_abs$src0_rel, "
+                   "$literal $pred_sel$last"),
+              pattern,
+              itin>,
+    R600ALU_Word0,
+    R600ALU_Word1_OP2 <inst> {
+
+  let src1 = 0;
+  let src1_rel = 0;
+  let src1_neg = 0;
+  let src1_abs = 0;
+  let update_exec_mask = 0;
+  let update_pred = 0;
+  let HasNativeOperands = 1;
+  let Op1 = 1;
+  let DisableEncoding = "$literal";
+
+  let Inst{31-0}  = Word0;
+  let Inst{63-32} = Word1;
+}
+
+class R600_1OP_Helper <bits<11> inst, string opName, SDPatternOperator node,
+                    InstrItinClass itin = AnyALU> :
+    R600_1OP <inst, opName,
+              [(set R600_Reg32:$dst, (node R600_Reg32:$src0))]
+>;
+
+// If you add our change the operands for R600_2OP instructions, you must
+// also update the R600Op2OperandIndex::ROI enum in R600Defines.h,
+// R600InstrInfo::buildDefaultInstruction(), and R600InstrInfo::getOperandIdx().
+class R600_2OP <bits<11> inst, string opName, list<dag> pattern,
+                InstrItinClass itin = AnyALU> :
+  InstR600 <inst,
+          (outs R600_Reg32:$dst),
+          (ins UEM:$update_exec_mask, UP:$update_pred, WRITE:$write,
+               OMOD:$omod, REL:$dst_rel, CLAMP:$clamp,
+               R600_Reg32:$src0, NEG:$src0_neg, REL:$src0_rel, ABS:$src0_abs, SEL:$src0_sel,
+               R600_Reg32:$src1, NEG:$src1_neg, REL:$src1_rel, ABS:$src1_abs, SEL:$src1_sel,
+               LAST:$last, R600_Pred:$pred_sel, LITERAL:$literal),
+          !strconcat("  ", opName,
+                "$clamp $update_exec_mask$update_pred$dst$write$dst_rel$omod, "
+                "$src0_neg$src0_abs$src0$src0_abs$src0_rel, "
+                "$src1_neg$src1_abs$src1$src1_abs$src1_rel, "
+                "$literal $pred_sel$last"),
+          pattern,
+          itin>,
+    R600ALU_Word0,
+    R600ALU_Word1_OP2 <inst> {
+
+  let HasNativeOperands = 1;
+  let Op2 = 1;
+  let DisableEncoding = "$literal";
+
+  let Inst{31-0}  = Word0;
+  let Inst{63-32} = Word1;
+}
+
+class R600_2OP_Helper <bits<11> inst, string opName, SDPatternOperator node,
+                       InstrItinClass itim = AnyALU> :
+    R600_2OP <inst, opName,
+              [(set R600_Reg32:$dst, (node R600_Reg32:$src0,
+                                           R600_Reg32:$src1))]
+>;
+
+// If you add our change the operands for R600_3OP instructions, you must
+// also update the R600Op3OperandIndex::ROI enum in R600Defines.h,
+// R600InstrInfo::buildDefaultInstruction(), and
+// R600InstrInfo::getOperandIdx().
+class R600_3OP <bits<5> inst, string opName, list<dag> pattern,
+                InstrItinClass itin = AnyALU> :
+  InstR600 <0,
+          (outs R600_Reg32:$dst),
+          (ins REL:$dst_rel, CLAMP:$clamp,
+               R600_Reg32:$src0, NEG:$src0_neg, REL:$src0_rel, SEL:$src0_sel,
+               R600_Reg32:$src1, NEG:$src1_neg, REL:$src1_rel, SEL:$src1_sel,
+               R600_Reg32:$src2, NEG:$src2_neg, REL:$src2_rel, SEL:$src2_sel,
+               LAST:$last, R600_Pred:$pred_sel, LITERAL:$literal),
+          !strconcat("  ", opName, "$clamp $dst$dst_rel, "
+                             "$src0_neg$src0$src0_rel, "
+                             "$src1_neg$src1$src1_rel, "
+                             "$src2_neg$src2$src2_rel, "
+                             "$literal $pred_sel$last"),
+          pattern,
+          itin>,
+    R600ALU_Word0,
+    R600ALU_Word1_OP3<inst>{
+
+  let HasNativeOperands = 1;
+  let DisableEncoding = "$literal";
+  let Op3 = 1;
+
+  let Inst{31-0}  = Word0;
+  let Inst{63-32} = Word1;
+}
+
+class R600_REDUCTION <bits<11> inst, dag ins, string asm, list<dag> pattern,
+                      InstrItinClass itin = VecALU> :
+  InstR600 <inst,
+          (outs R600_Reg32:$dst),
+          ins,
+          asm,
+          pattern,
+          itin>;
+
+class R600_TEX <bits<11> inst, string opName, list<dag> pattern,
+                InstrItinClass itin = AnyALU> :
+  InstR600 <inst,
+          (outs R600_Reg128:$DST_GPR),
+          (ins R600_Reg128:$SRC_GPR, i32imm:$RESOURCE_ID, i32imm:$SAMPLER_ID, i32imm:$textureTarget),
+          !strconcat(opName, "$DST_GPR, $SRC_GPR, $RESOURCE_ID, $SAMPLER_ID, $textureTarget"),
+          pattern,
+          itin>, TEX_WORD0, TEX_WORD1, TEX_WORD2 {
+    let Inst{31-0} = Word0;
+    let Inst{63-32} = Word1;
+
+    let TEX_INST = inst{4-0};
+    let SRC_REL = 0;
+    let DST_REL = 0;
+    let DST_SEL_X = 0;
+    let DST_SEL_Y = 1;
+    let DST_SEL_Z = 2;
+    let DST_SEL_W = 3;
+    let LOD_BIAS = 0;
+
+    let INST_MOD = 0;
+    let FETCH_WHOLE_QUAD = 0;
+    let ALT_CONST = 0;
+    let SAMPLER_INDEX_MODE = 0;
+
+    let COORD_TYPE_X = 0;
+    let COORD_TYPE_Y = 0;
+    let COORD_TYPE_Z = 0;
+    let COORD_TYPE_W = 0;
+  }
+
+} // End mayLoad = 1, mayStore = 0, hasSideEffects = 0
+
+def TEX_SHADOW : PatLeaf<
+  (imm),
+  [{uint32_t TType = (uint32_t)N->getZExtValue();
+    return (TType >= 6 && TType <= 8) || (TType >= 11 && TType <= 13);
+  }]
+>;
+
+def TEX_RECT : PatLeaf<
+  (imm),
+  [{uint32_t TType = (uint32_t)N->getZExtValue();
+    return TType == 5;
+  }]
+>;
+
+def TEX_ARRAY : PatLeaf<
+  (imm),
+  [{uint32_t TType = (uint32_t)N->getZExtValue();
+    return TType == 9 || TType == 10 || TType == 15 || TType == 16;
+  }]
+>;
+
+def TEX_SHADOW_ARRAY : PatLeaf<
+  (imm),
+  [{uint32_t TType = (uint32_t)N->getZExtValue();
+    return TType == 11 || TType == 12 || TType == 17;
+  }]
+>;
+
+class EG_CF_RAT <bits <8> cf_inst, bits <6> rat_inst, bits<4> rat_id, dag outs,
+                 dag ins, string asm, list<dag> pattern> :
+    InstR600ISA <outs, ins, asm, pattern> {
+  bits<7>  RW_GPR;
+  bits<7>  INDEX_GPR;
+
+  bits<2>  RIM;
+  bits<2>  TYPE;
+  bits<1>  RW_REL;
+  bits<2>  ELEM_SIZE;
+
+  bits<12> ARRAY_SIZE;
+  bits<4>  COMP_MASK;
+  bits<4>  BURST_COUNT;
+  bits<1>  VPM;
+  bits<1>  eop;
+  bits<1>  MARK;
+  bits<1>  BARRIER;
+
+  // CF_ALLOC_EXPORT_WORD0_RAT
+  let Inst{3-0}   = rat_id;
+  let Inst{9-4}   = rat_inst;
+  let Inst{10}    = 0; // Reserved
+  let Inst{12-11} = RIM;
+  let Inst{14-13} = TYPE;
+  let Inst{21-15} = RW_GPR;
+  let Inst{22}    = RW_REL;
+  let Inst{29-23} = INDEX_GPR;
+  let Inst{31-30} = ELEM_SIZE;
+
+  // CF_ALLOC_EXPORT_WORD1_BUF
+  let Inst{43-32} = ARRAY_SIZE;
+  let Inst{47-44} = COMP_MASK;
+  let Inst{51-48} = BURST_COUNT;
+  let Inst{52}    = VPM;
+  let Inst{53}    = eop;
+  let Inst{61-54} = cf_inst;
+  let Inst{62}    = MARK;
+  let Inst{63}    = BARRIER;
+}
+
+class LoadParamFrag <PatFrag load_type> : PatFrag <
+  (ops node:$ptr), (load_type node:$ptr),
+  [{ return isParamLoad(dyn_cast<LoadSDNode>(N)); }]
+>;
+
+def load_param : LoadParamFrag<load>;
+def load_param_zexti8 : LoadParamFrag<zextloadi8>;
+def load_param_zexti16 : LoadParamFrag<zextloadi16>;
+
+def isR600 : Predicate<"Subtarget.device()"
+                            "->getGeneration() == AMDGPUDeviceInfo::HD4XXX">;
+def isR700 : Predicate<"Subtarget.device()"
+                            "->getGeneration() == AMDGPUDeviceInfo::HD4XXX &&"
+                            "Subtarget.device()->getDeviceFlag()"
+                            ">= OCL_DEVICE_RV710">;
+def isEG : Predicate<
+  "Subtarget.device()->getGeneration() >= AMDGPUDeviceInfo::HD5XXX && "
+  "Subtarget.device()->getGeneration() < AMDGPUDeviceInfo::HD7XXX && "
+  "Subtarget.device()->getDeviceFlag() != OCL_DEVICE_CAYMAN">;
+
+def isCayman : Predicate<"Subtarget.device()"
+                            "->getDeviceFlag() == OCL_DEVICE_CAYMAN">;
+def isEGorCayman : Predicate<"Subtarget.device()"
+                            "->getGeneration() == AMDGPUDeviceInfo::HD5XXX"
+                            "|| Subtarget.device()->getGeneration() =="
+                            "AMDGPUDeviceInfo::HD6XXX">;
+
+def isR600toCayman : Predicate<
+                     "Subtarget.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX">;
+
+//===----------------------------------------------------------------------===//
+// R600 SDNodes
+//===----------------------------------------------------------------------===//
+
+def INTERP_PAIR_XY :  AMDGPUShaderInst <
+  (outs R600_TReg32_X:$dst0, R600_TReg32_Y:$dst1),
+  (ins i32imm:$src0, R600_Reg32:$src1, R600_Reg32:$src2),
+  "INTERP_PAIR_XY $src0 $src1 $src2 : $dst0 dst1",
+  []>;
+
+def INTERP_PAIR_ZW :  AMDGPUShaderInst <
+  (outs R600_TReg32_Z:$dst0, R600_TReg32_W:$dst1),
+  (ins i32imm:$src0, R600_Reg32:$src1, R600_Reg32:$src2),
+  "INTERP_PAIR_ZW $src0 $src1 $src2 : $dst0 dst1",
+  []>;
+
+def CONST_ADDRESS: SDNode<"AMDGPUISD::CONST_ADDRESS",
+  SDTypeProfile<1, -1, [SDTCisInt<0>, SDTCisPtrTy<1>]>,
+  [SDNPVariadic]
+>;
+
+//===----------------------------------------------------------------------===//
+// Interpolation Instructions
+//===----------------------------------------------------------------------===//
+
+def INTERP_VEC_LOAD :  AMDGPUShaderInst <
+  (outs R600_Reg128:$dst),
+  (ins i32imm:$src0),
+  "INTERP_LOAD $src0 : $dst",
+  []>;
+
+def INTERP_XY : R600_2OP <0xD6, "INTERP_XY", []> {
+  let bank_swizzle = 5;
+}
+
+def INTERP_ZW : R600_2OP <0xD7, "INTERP_ZW", []> {
+  let bank_swizzle = 5;
+}
+
+def INTERP_LOAD_P0 : R600_1OP <0xE0, "INTERP_LOAD_P0", []>;
+
+//===----------------------------------------------------------------------===//
+// Export Instructions
+//===----------------------------------------------------------------------===//
+
+def ExportType : SDTypeProfile<0, 7, [SDTCisFP<0>, SDTCisInt<1>]>;
+
+def EXPORT: SDNode<"AMDGPUISD::EXPORT", ExportType,
+  [SDNPHasChain, SDNPSideEffect]>;
+
+class ExportWord0 {
+  field bits<32> Word0;
+
+  bits<13> arraybase;
+  bits<2> type;
+  bits<7> gpr;
+  bits<2> elem_size;
+
+  let Word0{12-0} = arraybase;
+  let Word0{14-13} = type;
+  let Word0{21-15} = gpr;
+  let Word0{22} = 0; // RW_REL
+  let Word0{29-23} = 0; // INDEX_GPR
+  let Word0{31-30} = elem_size;
+}
+
+class ExportSwzWord1 {
+  field bits<32> Word1;
+
+  bits<3> sw_x;
+  bits<3> sw_y;
+  bits<3> sw_z;
+  bits<3> sw_w;
+  bits<1> eop;
+  bits<8> inst;
+
+  let Word1{2-0} = sw_x;
+  let Word1{5-3} = sw_y;
+  let Word1{8-6} = sw_z;
+  let Word1{11-9} = sw_w;
+}
+
+class ExportBufWord1 {
+  field bits<32> Word1;
+
+  bits<12> arraySize;
+  bits<4> compMask;
+  bits<1> eop;
+  bits<8> inst;
+
+  let Word1{11-0} = arraySize;
+  let Word1{15-12} = compMask;
+}
+
+multiclass ExportPattern<Instruction ExportInst, bits<8> cf_inst> {
+  def : Pat<(int_R600_store_pixel_depth R600_Reg32:$reg),
+    (ExportInst
+        (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), R600_Reg32:$reg, sub0),
+        0, 61, 0, 7, 7, 7, cf_inst, 0)
+  >;
+
+  def : Pat<(int_R600_store_pixel_stencil R600_Reg32:$reg),
+    (ExportInst
+        (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), R600_Reg32:$reg, sub0),
+        0, 61, 7, 0, 7, 7, cf_inst, 0)
+  >;
+
+  def : Pat<(int_R600_store_dummy (i32 imm:$type)),
+    (ExportInst
+        (v4f32 (IMPLICIT_DEF)), imm:$type, 0, 7, 7, 7, 7, cf_inst, 0)
+  >;
+
+  def : Pat<(int_R600_store_dummy 1),
+    (ExportInst
+        (v4f32 (IMPLICIT_DEF)), 1, 60, 7, 7, 7, 7, cf_inst, 0)
+  >;
+
+  def : Pat<(EXPORT (v4f32 R600_Reg128:$src), (i32 imm:$base), (i32 imm:$type),
+    (i32 imm:$swz_x), (i32 imm:$swz_y), (i32 imm:$swz_z), (i32 imm:$swz_w)),
+        (ExportInst R600_Reg128:$src, imm:$type, imm:$base,
+        imm:$swz_x, imm:$swz_y, imm:$swz_z, imm:$swz_w, cf_inst, 0)
+  >;
+
+}
+
+multiclass SteamOutputExportPattern<Instruction ExportInst,
+    bits<8> buf0inst, bits<8> buf1inst, bits<8> buf2inst, bits<8> buf3inst> {
+// Stream0
+  def : Pat<(int_R600_store_stream_output (v4f32 R600_Reg128:$src),
+      (i32 imm:$arraybase), (i32 0), (i32 imm:$mask)),
+      (ExportInst R600_Reg128:$src, 0, imm:$arraybase,
+      4095, imm:$mask, buf0inst, 0)>;
+// Stream1
+  def : Pat<(int_R600_store_stream_output (v4f32 R600_Reg128:$src),
+      (i32 imm:$arraybase), (i32 1), (i32 imm:$mask)),
+      (ExportInst R600_Reg128:$src, 0, imm:$arraybase,
+      4095, imm:$mask, buf1inst, 0)>;
+// Stream2
+  def : Pat<(int_R600_store_stream_output (v4f32 R600_Reg128:$src),
+      (i32 imm:$arraybase), (i32 2), (i32 imm:$mask)),
+      (ExportInst R600_Reg128:$src, 0, imm:$arraybase,
+      4095, imm:$mask, buf2inst, 0)>;
+// Stream3
+  def : Pat<(int_R600_store_stream_output (v4f32 R600_Reg128:$src),
+      (i32 imm:$arraybase), (i32 3), (i32 imm:$mask)),
+      (ExportInst R600_Reg128:$src, 0, imm:$arraybase,
+      4095, imm:$mask, buf3inst, 0)>;
+}
+
+let usesCustomInserter = 1 in {
+
+class ExportSwzInst : InstR600ISA<(
+    outs),
+    (ins R600_Reg128:$gpr, i32imm:$type, i32imm:$arraybase,
+    i32imm:$sw_x, i32imm:$sw_y, i32imm:$sw_z, i32imm:$sw_w, i32imm:$inst,
+    i32imm:$eop),
+    !strconcat("EXPORT", " $gpr"),
+    []>, ExportWord0, ExportSwzWord1 {
+  let elem_size = 3;
+  let Inst{31-0} = Word0;
+  let Inst{63-32} = Word1;
+}
+
+} // End usesCustomInserter = 1
+
+class ExportBufInst : InstR600ISA<(
+    outs),
+    (ins R600_Reg128:$gpr, i32imm:$type, i32imm:$arraybase,
+    i32imm:$arraySize, i32imm:$compMask, i32imm:$inst, i32imm:$eop),
+    !strconcat("EXPORT", " $gpr"),
+    []>, ExportWord0, ExportBufWord1 {
+  let elem_size = 0;
+  let Inst{31-0} = Word0;
+  let Inst{63-32} = Word1;
+}
+
+//===----------------------------------------------------------------------===//
+// Control Flow Instructions
+//===----------------------------------------------------------------------===//
+
+class CF_ALU_WORD0 {
+  field bits<32> Word0;
+
+  bits<22> ADDR;
+  bits<4> KCACHE_BANK0;
+  bits<4> KCACHE_BANK1;
+  bits<2> KCACHE_MODE0;
+
+  let Word0{21-0} = ADDR;
+  let Word0{25-22} = KCACHE_BANK0;
+  let Word0{29-26} = KCACHE_BANK1;
+  let Word0{31-30} = KCACHE_MODE0;
+}
+
+class CF_ALU_WORD1 {
+  field bits<32> Word1;
+
+  bits<2> KCACHE_MODE1;
+  bits<8> KCACHE_ADDR0;
+  bits<8> KCACHE_ADDR1;
+  bits<7> COUNT;
+  bits<1> ALT_CONST;
+  bits<4> CF_INST;
+  bits<1> WHOLE_QUAD_MODE;
+  bits<1> BARRIER;
+
+  let Word1{1-0} = KCACHE_MODE1;
+  let Word1{9-2} = KCACHE_ADDR0;
+  let Word1{17-10} = KCACHE_ADDR1;
+  let Word1{24-18} = COUNT;
+  let Word1{25} = ALT_CONST;
+  let Word1{29-26} = CF_INST;
+  let Word1{30} = WHOLE_QUAD_MODE;
+  let Word1{31} = BARRIER;
+}
+
+class ALU_CLAUSE<bits<4> inst, string OpName> : AMDGPUInst <(outs),
+(ins i32imm:$ADDR, i32imm:$KCACHE_BANK0, i32imm:$KCACHE_BANK1, i32imm:$KCACHE_MODE0, i32imm:$KCACHE_MODE1,
+i32imm:$KCACHE_ADDR0, i32imm:$KCACHE_ADDR1, i32imm:$COUNT),
+!strconcat(OpName, " $COUNT, @$ADDR, "
+"KC0[CB$KCACHE_BANK0:$KCACHE_ADDR0-$KCACHE_ADDR0+32]"
+", KC1[CB$KCACHE_BANK1:$KCACHE_ADDR1-$KCACHE_ADDR1+32]"),
+[] >, CF_ALU_WORD0, CF_ALU_WORD1 {
+  field bits<64> Inst;
+
+  let CF_INST = inst;
+  let ALT_CONST = 0;
+  let WHOLE_QUAD_MODE = 0;
+  let BARRIER = 1;
+
+  let Inst{31-0} = Word0;
+  let Inst{63-32} = Word1;
+}
+
+class CF_WORD0 {
+  field bits<32> Word0;
+
+  bits<24> ADDR;
+  bits<3> JUMPTABLE_SEL;
+
+  let Word0{23-0} = ADDR;
+  let Word0{26-24} = JUMPTABLE_SEL;
+}
+
+class CF_WORD1 {
+  field bits<32> Word1;
+
+  bits<3> POP_COUNT;
+  bits<5> CF_CONST;
+  bits<2> COND;
+  bits<6> COUNT;
+  bits<1> VALID_PIXEL_MODE;
+  bits<8> CF_INST;
+  bits<1> BARRIER;
+
+  let Word1{2-0} = POP_COUNT;
+  let Word1{7-3} = CF_CONST;
+  let Word1{9-8} = COND;
+  let Word1{15-10} = COUNT;
+  let Word1{20} = VALID_PIXEL_MODE;
+  let Word1{29-22} = CF_INST;
+  let Word1{31} = BARRIER;
+}
+
+class CF_CLAUSE <bits<8> inst, dag ins, string AsmPrint> : AMDGPUInst <(outs),
+ins, AsmPrint, [] >, CF_WORD0, CF_WORD1 {
+  field bits<64> Inst;
+
+  let CF_INST = inst;
+  let BARRIER = 1;
+  let JUMPTABLE_SEL = 0;
+  let CF_CONST = 0;
+  let VALID_PIXEL_MODE = 0;
+  let COND = 0;
+
+  let Inst{31-0} = Word0;
+  let Inst{63-32} = Word1;
+}
+
+def CF_TC : CF_CLAUSE<1, (ins i32imm:$ADDR, i32imm:$COUNT),
+"TEX $COUNT @$ADDR"> {
+  let POP_COUNT = 0;
+}
+
+def CF_VC : CF_CLAUSE<2, (ins i32imm:$ADDR, i32imm:$COUNT),
+"VTX $COUNT @$ADDR"> {
+  let POP_COUNT = 0;
+}
+
+def WHILE_LOOP : CF_CLAUSE<6, (ins i32imm:$ADDR), "LOOP_START_DX10 @$ADDR"> {
+  let POP_COUNT = 0;
+  let COUNT = 0;
+}
+
+def END_LOOP : CF_CLAUSE<5, (ins i32imm:$ADDR), "END_LOOP @$ADDR"> {
+  let POP_COUNT = 0;
+  let COUNT = 0;
+}
+
+def LOOP_BREAK : CF_CLAUSE<9, (ins i32imm:$ADDR), "LOOP_BREAK @$ADDR"> {
+  let POP_COUNT = 0;
+  let COUNT = 0;
+}
+
+def CF_CONTINUE : CF_CLAUSE<8, (ins i32imm:$ADDR), "CONTINUE @$ADDR"> {
+  let POP_COUNT = 0;
+  let COUNT = 0;
+}
+
+def CF_JUMP : CF_CLAUSE<10, (ins i32imm:$ADDR, i32imm:$POP_COUNT), "JUMP @$ADDR POP:$POP_COUNT"> {
+  let COUNT = 0;
+}
+
+def CF_ELSE : CF_CLAUSE<13, (ins i32imm:$ADDR, i32imm:$POP_COUNT), "ELSE @$ADDR POP:$POP_COUNT"> {
+  let COUNT = 0;
+}
+
+def CF_CALL_FS : CF_CLAUSE<19, (ins), "CALL_FS"> {
+  let ADDR = 0;
+  let COUNT = 0;
+  let POP_COUNT = 0;
+}
+
+def POP : CF_CLAUSE<14, (ins i32imm:$ADDR, i32imm:$POP_COUNT), "POP @$ADDR POP:$POP_COUNT"> {
+  let COUNT = 0;
+}
+
+def CF_ALU : ALU_CLAUSE<8, "ALU">;
+def CF_ALU_PUSH_BEFORE : ALU_CLAUSE<9, "ALU_PUSH_BEFORE">;
+
+def STACK_SIZE : AMDGPUInst <(outs),
+(ins i32imm:$num), "nstack $num", [] > {
+  field bits<8> Inst;
+  bits<8> num;
+  let Inst = num;
+}
+
+let Predicates = [isR600toCayman] in {
+
+//===----------------------------------------------------------------------===//
+// Common Instructions R600, R700, Evergreen, Cayman
+//===----------------------------------------------------------------------===//
+
+def ADD : R600_2OP_Helper <0x0, "ADD", fadd>;
+// Non-IEEE MUL: 0 * anything = 0
+def MUL : R600_2OP_Helper <0x1, "MUL NON-IEEE", int_AMDGPU_mul>;
+def MUL_IEEE : R600_2OP_Helper <0x2, "MUL_IEEE", fmul>;
+def MAX : R600_2OP_Helper <0x3, "MAX", AMDGPUfmax>;
+def MIN : R600_2OP_Helper <0x4, "MIN", AMDGPUfmin>;
+
+// For the SET* instructions there is a naming conflict in TargetSelectionDAG.td,
+// so some of the instruction names don't match the asm string.
+// XXX: Use the defs in TargetSelectionDAG.td instead of intrinsics.
+def SETE : R600_2OP <
+  0x08, "SETE",
+  [(set R600_Reg32:$dst,
+   (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, FP_ONE, FP_ZERO,
+             COND_EQ))]
+>;
+
+def SGT : R600_2OP <
+  0x09, "SETGT",
+  [(set R600_Reg32:$dst,
+   (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, FP_ONE, FP_ZERO,
+              COND_GT))]
+>;
+
+def SGE : R600_2OP <
+  0xA, "SETGE",
+  [(set R600_Reg32:$dst,
+   (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, FP_ONE, FP_ZERO,
+              COND_GE))]
+>;
+
+def SNE : R600_2OP <
+  0xB, "SETNE",
+  [(set R600_Reg32:$dst,
+   (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, FP_ONE, FP_ZERO,
+    COND_NE))]
+>;
+
+def SETE_DX10 : R600_2OP <
+  0xC, "SETE_DX10",
+  [(set R600_Reg32:$dst,
+   (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, (i32 -1), (i32 0),
+    COND_EQ))]
+>;
+
+def SETGT_DX10 : R600_2OP <
+  0xD, "SETGT_DX10",
+  [(set R600_Reg32:$dst,
+   (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, (i32 -1), (i32 0),
+    COND_GT))]
+>;
+
+def SETGE_DX10 : R600_2OP <
+  0xE, "SETGE_DX10",
+  [(set R600_Reg32:$dst,
+   (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, (i32 -1), (i32 0),
+    COND_GE))]
+>;
+
+def SETNE_DX10 : R600_2OP <
+  0xF, "SETNE_DX10",
+  [(set R600_Reg32:$dst,
+    (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, (i32 -1), (i32 0),
+     COND_NE))]
+>;
+
+def FRACT : R600_1OP_Helper <0x10, "FRACT", AMDGPUfract>;
+def TRUNC : R600_1OP_Helper <0x11, "TRUNC", int_AMDGPU_trunc>;
+def CEIL : R600_1OP_Helper <0x12, "CEIL", fceil>;
+def RNDNE : R600_1OP_Helper <0x13, "RNDNE", frint>;
+def FLOOR : R600_1OP_Helper <0x14, "FLOOR", ffloor>;
+
+def MOV : R600_1OP <0x19, "MOV", []>;
+
+let isPseudo = 1, isCodeGenOnly = 1, usesCustomInserter = 1 in {
+
+class MOV_IMM <ValueType vt, Operand immType> : AMDGPUInst <
+  (outs R600_Reg32:$dst),
+  (ins immType:$imm),
+  "",
+  []
+>;
+
+} // end let isPseudo = 1, isCodeGenOnly = 1, usesCustomInserter = 1
+
+def MOV_IMM_I32 : MOV_IMM<i32, i32imm>;
+def : Pat <
+  (imm:$val),
+  (MOV_IMM_I32 imm:$val)
+>;
+
+def MOV_IMM_F32 : MOV_IMM<f32, f32imm>;
+def : Pat <
+  (fpimm:$val),
+  (MOV_IMM_F32  fpimm:$val)
+>;
+
+def PRED_SETE : R600_2OP <0x20, "PRED_SETE", []>;
+def PRED_SETGT : R600_2OP <0x21, "PRED_SETGT", []>;
+def PRED_SETGE : R600_2OP <0x22, "PRED_SETGE", []>;
+def PRED_SETNE : R600_2OP <0x23, "PRED_SETNE", []>;
+
+let hasSideEffects = 1 in {
+
+def KILLGT : R600_2OP <0x2D, "KILLGT", []>;
+
+} // end hasSideEffects
+
+def AND_INT : R600_2OP_Helper <0x30, "AND_INT", and>;
+def OR_INT : R600_2OP_Helper <0x31, "OR_INT", or>;
+def XOR_INT : R600_2OP_Helper <0x32, "XOR_INT", xor>;
+def NOT_INT : R600_1OP_Helper <0x33, "NOT_INT", not>;
+def ADD_INT : R600_2OP_Helper <0x34, "ADD_INT", add>;
+def SUB_INT : R600_2OP_Helper <0x35, "SUB_INT", sub>;
+def MAX_INT : R600_2OP_Helper <0x36, "MAX_INT", AMDGPUsmax>;
+def MIN_INT : R600_2OP_Helper <0x37, "MIN_INT", AMDGPUsmin>;
+def MAX_UINT : R600_2OP_Helper <0x38, "MAX_UINT", AMDGPUumax>;
+def MIN_UINT : R600_2OP_Helper <0x39, "MIN_UINT", AMDGPUumin>;
+
+def SETE_INT : R600_2OP <
+  0x3A, "SETE_INT",
+  [(set (i32 R600_Reg32:$dst),
+   (selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETEQ))]
+>;
+
+def SETGT_INT : R600_2OP <
+  0x3B, "SETGT_INT",
+  [(set (i32 R600_Reg32:$dst),
+   (selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETGT))]
+>;
+
+def SETGE_INT : R600_2OP <
+  0x3C, "SETGE_INT",
+  [(set (i32 R600_Reg32:$dst),
+   (selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETGE))]
+>;
+
+def SETNE_INT : R600_2OP <
+  0x3D, "SETNE_INT",
+  [(set (i32 R600_Reg32:$dst),
+   (selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETNE))]
+>;
+
+def SETGT_UINT : R600_2OP <
+  0x3E, "SETGT_UINT",
+  [(set (i32 R600_Reg32:$dst),
+   (selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETUGT))]
+>;
+
+def SETGE_UINT : R600_2OP <
+  0x3F, "SETGE_UINT",
+  [(set (i32 R600_Reg32:$dst),
+    (selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETUGE))]
+>;
+
+def PRED_SETE_INT : R600_2OP <0x42, "PRED_SETE_INT", []>;
+def PRED_SETGT_INT : R600_2OP <0x43, "PRED_SETGE_INT", []>;
+def PRED_SETGE_INT : R600_2OP <0x44, "PRED_SETGE_INT", []>;
+def PRED_SETNE_INT : R600_2OP <0x45, "PRED_SETNE_INT", []>;
+
+def CNDE_INT : R600_3OP <
+  0x1C, "CNDE_INT",
+  [(set (i32 R600_Reg32:$dst),
+   (selectcc (i32 R600_Reg32:$src0), 0,
+       (i32 R600_Reg32:$src1), (i32 R600_Reg32:$src2),
+       COND_EQ))]
+>;
+
+def CNDGE_INT : R600_3OP <
+  0x1E, "CNDGE_INT",
+  [(set (i32 R600_Reg32:$dst),
+   (selectcc (i32 R600_Reg32:$src0), 0,
+       (i32 R600_Reg32:$src1), (i32 R600_Reg32:$src2),
+       COND_GE))]
+>;
+
+def CNDGT_INT : R600_3OP <
+  0x1D, "CNDGT_INT",
+  [(set (i32 R600_Reg32:$dst),
+   (selectcc (i32 R600_Reg32:$src0), 0,
+       (i32 R600_Reg32:$src1), (i32 R600_Reg32:$src2),
+       COND_GT))]
+>;
+
+//===----------------------------------------------------------------------===//
+// Texture instructions
+//===----------------------------------------------------------------------===//
+
+def TEX_LD : R600_TEX <
+  0x03, "TEX_LD",
+  [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txf R600_Reg128:$SRC_GPR,
+      imm:$OFFSET_X, imm:$OFFSET_Y, imm:$OFFSET_Z, imm:$RESOURCE_ID,
+      imm:$SAMPLER_ID, imm:$textureTarget))]
+> {
+let AsmString = "TEX_LD $DST_GPR, $SRC_GPR, $OFFSET_X, $OFFSET_Y, $OFFSET_Z,"
+    "$RESOURCE_ID, $SAMPLER_ID, $textureTarget";
+let InOperandList = (ins R600_Reg128:$SRC_GPR, i32imm:$OFFSET_X,
+    i32imm:$OFFSET_Y, i32imm:$OFFSET_Z, i32imm:$RESOURCE_ID, i32imm:$SAMPLER_ID,
+    i32imm:$textureTarget);
+}
+
+def TEX_GET_TEXTURE_RESINFO : R600_TEX <
+  0x04, "TEX_GET_TEXTURE_RESINFO",
+  [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txq R600_Reg128:$SRC_GPR,
+      imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))]
+>;
+
+def TEX_GET_GRADIENTS_H : R600_TEX <
+  0x07, "TEX_GET_GRADIENTS_H",
+  [(set R600_Reg128:$DST_GPR, (int_AMDGPU_ddx R600_Reg128:$SRC_GPR,
+      imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))]
+>;
+
+def TEX_GET_GRADIENTS_V : R600_TEX <
+  0x08, "TEX_GET_GRADIENTS_V",
+  [(set R600_Reg128:$DST_GPR, (int_AMDGPU_ddy R600_Reg128:$SRC_GPR,
+      imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))]
+>;
+
+def TEX_SET_GRADIENTS_H : R600_TEX <
+  0x0B, "TEX_SET_GRADIENTS_H",
+  []
+>;
+
+def TEX_SET_GRADIENTS_V : R600_TEX <
+  0x0C, "TEX_SET_GRADIENTS_V",
+  []
+>;
+
+def TEX_SAMPLE : R600_TEX <
+  0x10, "TEX_SAMPLE",
+  [(set R600_Reg128:$DST_GPR, (int_AMDGPU_tex R600_Reg128:$SRC_GPR,
+      imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))]
+>;
+
+def TEX_SAMPLE_C : R600_TEX <
+  0x18, "TEX_SAMPLE_C",
+  [(set R600_Reg128:$DST_GPR, (int_AMDGPU_tex R600_Reg128:$SRC_GPR,
+      imm:$RESOURCE_ID, imm:$SAMPLER_ID, TEX_SHADOW:$textureTarget))]
+>;
+
+def TEX_SAMPLE_L : R600_TEX <
+  0x11, "TEX_SAMPLE_L",
+  [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txl R600_Reg128:$SRC_GPR,
+      imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))]
+>;
+
+def TEX_SAMPLE_C_L : R600_TEX <
+  0x19, "TEX_SAMPLE_C_L",
+  [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txl R600_Reg128:$SRC_GPR,
+      imm:$RESOURCE_ID, imm:$SAMPLER_ID, TEX_SHADOW:$textureTarget))]
+>;
+
+def TEX_SAMPLE_LB : R600_TEX <
+  0x12, "TEX_SAMPLE_LB",
+  [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txb R600_Reg128:$SRC_GPR,
+      imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))]
+>;
+
+def TEX_SAMPLE_C_LB : R600_TEX <
+  0x1A, "TEX_SAMPLE_C_LB",
+  [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txb R600_Reg128:$SRC_GPR,
+      imm:$RESOURCE_ID, imm:$SAMPLER_ID, TEX_SHADOW:$textureTarget))]
+>;
+
+def TEX_SAMPLE_G : R600_TEX <
+  0x14, "TEX_SAMPLE_G",
+  []
+>;
+
+def TEX_SAMPLE_C_G : R600_TEX <
+  0x1C, "TEX_SAMPLE_C_G",
+  []
+>;
+
+//===----------------------------------------------------------------------===//
+// Helper classes for common instructions
+//===----------------------------------------------------------------------===//
+
+class MUL_LIT_Common <bits<5> inst> : R600_3OP <
+  inst, "MUL_LIT",
+  []
+>;
+
+class MULADD_Common <bits<5> inst> : R600_3OP <
+  inst, "MULADD",
+  []
+>;
+
+class MULADD_IEEE_Common <bits<5> inst> : R600_3OP <
+  inst, "MULADD_IEEE",
+  [(set (f32 R600_Reg32:$dst),
+   (fadd (fmul R600_Reg32:$src0, R600_Reg32:$src1), R600_Reg32:$src2))]
+>;
+
+class CNDE_Common <bits<5> inst> : R600_3OP <
+  inst, "CNDE",
+  [(set R600_Reg32:$dst,
+   (selectcc (f32 R600_Reg32:$src0), FP_ZERO,
+       (f32 R600_Reg32:$src1), (f32 R600_Reg32:$src2),
+       COND_EQ))]
+>;
+
+class CNDGT_Common <bits<5> inst> : R600_3OP <
+  inst, "CNDGT",
+  [(set R600_Reg32:$dst,
+   (selectcc (f32 R600_Reg32:$src0), FP_ZERO,
+       (f32 R600_Reg32:$src1), (f32 R600_Reg32:$src2),
+       COND_GT))]
+>;
+
+class CNDGE_Common <bits<5> inst> : R600_3OP <
+  inst, "CNDGE",
+  [(set R600_Reg32:$dst,
+   (selectcc (f32 R600_Reg32:$src0), FP_ZERO,
+       (f32 R600_Reg32:$src1), (f32 R600_Reg32:$src2),
+       COND_GE))]
+>;
+
+multiclass DOT4_Common <bits<11> inst> {
+
+  def _pseudo : R600_REDUCTION <inst,
+    (ins R600_Reg128:$src0, R600_Reg128:$src1),
+    "DOT4 $dst $src0, $src1",
+    [(set R600_Reg32:$dst, (int_AMDGPU_dp4 R600_Reg128:$src0, R600_Reg128:$src1))]
+  >;
+
+  def _real : R600_2OP <inst, "DOT4", []>;
+}
+
+let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
+multiclass CUBE_Common <bits<11> inst> {
+
+  def _pseudo : InstR600 <
+    inst,
+    (outs R600_Reg128:$dst),
+    (ins R600_Reg128:$src),
+    "CUBE $dst $src",
+    [(set R600_Reg128:$dst, (int_AMDGPU_cube R600_Reg128:$src))],
+    VecALU
+  > {
+    let isPseudo = 1;
+  }
+
+  def _real : R600_2OP <inst, "CUBE", []>;
+}
+} // End mayLoad = 0, mayStore = 0, hasSideEffects = 0
+
+class EXP_IEEE_Common <bits<11> inst> : R600_1OP_Helper <
+  inst, "EXP_IEEE", fexp2
+>;
+
+class FLT_TO_INT_Common <bits<11> inst> : R600_1OP_Helper <
+  inst, "FLT_TO_INT", fp_to_sint
+>;
+
+class INT_TO_FLT_Common <bits<11> inst> : R600_1OP_Helper <
+  inst, "INT_TO_FLT", sint_to_fp
+>;
+
+class FLT_TO_UINT_Common <bits<11> inst> : R600_1OP_Helper <
+  inst, "FLT_TO_UINT", fp_to_uint
+>;
+
+class UINT_TO_FLT_Common <bits<11> inst> : R600_1OP_Helper <
+  inst, "UINT_TO_FLT", uint_to_fp
+>;
+
+class LOG_CLAMPED_Common <bits<11> inst> : R600_1OP <
+  inst, "LOG_CLAMPED", []
+>;
+
+class LOG_IEEE_Common <bits<11> inst> : R600_1OP_Helper <
+  inst, "LOG_IEEE", flog2
+>;
+
+class LSHL_Common <bits<11> inst> : R600_2OP_Helper <inst, "LSHL", shl>;
+class LSHR_Common <bits<11> inst> : R600_2OP_Helper <inst, "LSHR", srl>;
+class ASHR_Common <bits<11> inst> : R600_2OP_Helper <inst, "ASHR", sra>;
+class MULHI_INT_Common <bits<11> inst> : R600_2OP_Helper <
+  inst, "MULHI_INT", mulhs
+>;
+class MULHI_UINT_Common <bits<11> inst> : R600_2OP_Helper <
+  inst, "MULHI", mulhu
+>;
+class MULLO_INT_Common <bits<11> inst> : R600_2OP_Helper <
+  inst, "MULLO_INT", mul
+>;
+class MULLO_UINT_Common <bits<11> inst> : R600_2OP <inst, "MULLO_UINT", []>;
+
+class RECIP_CLAMPED_Common <bits<11> inst> : R600_1OP <
+  inst, "RECIP_CLAMPED", []
+>;
+
+class RECIP_IEEE_Common <bits<11> inst> : R600_1OP <
+  inst, "RECIP_IEEE", [(set R600_Reg32:$dst, (fdiv FP_ONE, R600_Reg32:$src0))]
+>;
+
+class RECIP_UINT_Common <bits<11> inst> : R600_1OP_Helper <
+  inst, "RECIP_UINT", AMDGPUurecip
+>;
+
+class RECIPSQRT_CLAMPED_Common <bits<11> inst> : R600_1OP_Helper <
+  inst, "RECIPSQRT_CLAMPED", int_AMDGPU_rsq
+>;
+
+class RECIPSQRT_IEEE_Common <bits<11> inst> : R600_1OP <
+  inst, "RECIPSQRT_IEEE", []
+>;
+
+class SIN_Common <bits<11> inst> : R600_1OP <
+  inst, "SIN", []>{
+  let Trig = 1;
+}
+
+class COS_Common <bits<11> inst> : R600_1OP <
+  inst, "COS", []> {
+  let Trig = 1;
+}
+
+//===----------------------------------------------------------------------===//
+// Helper patterns for complex intrinsics
+//===----------------------------------------------------------------------===//
+
+multiclass DIV_Common <InstR600 recip_ieee> {
+def : Pat<
+  (int_AMDGPU_div R600_Reg32:$src0, R600_Reg32:$src1),
+  (MUL_IEEE R600_Reg32:$src0, (recip_ieee R600_Reg32:$src1))
+>;
+
+def : Pat<
+  (fdiv R600_Reg32:$src0, R600_Reg32:$src1),
+  (MUL_IEEE R600_Reg32:$src0, (recip_ieee R600_Reg32:$src1))
+>;
+}
+
+class TGSI_LIT_Z_Common <InstR600 mul_lit, InstR600 log_clamped, InstR600 exp_ieee> : Pat <
+  (int_TGSI_lit_z R600_Reg32:$src_x, R600_Reg32:$src_y, R600_Reg32:$src_w),
+  (exp_ieee (mul_lit (log_clamped (MAX R600_Reg32:$src_y, (f32 ZERO))), R600_Reg32:$src_w, R600_Reg32:$src_x))
+>;
+
+//===----------------------------------------------------------------------===//
+// R600 / R700 Instructions
+//===----------------------------------------------------------------------===//
+
+let Predicates = [isR600] in {
+
+  def MUL_LIT_r600 : MUL_LIT_Common<0x0C>;
+  def MULADD_r600 : MULADD_Common<0x10>;
+  def MULADD_IEEE_r600 : MULADD_IEEE_Common<0x14>;
+  def CNDE_r600 : CNDE_Common<0x18>;
+  def CNDGT_r600 : CNDGT_Common<0x19>;
+  def CNDGE_r600 : CNDGE_Common<0x1A>;
+  defm DOT4_r600 : DOT4_Common<0x50>;
+  defm CUBE_r600 : CUBE_Common<0x52>;
+  def EXP_IEEE_r600 : EXP_IEEE_Common<0x61>;
+  def LOG_CLAMPED_r600 : LOG_CLAMPED_Common<0x62>;
+  def LOG_IEEE_r600 : LOG_IEEE_Common<0x63>;
+  def RECIP_CLAMPED_r600 : RECIP_CLAMPED_Common<0x64>;
+  def RECIP_IEEE_r600 : RECIP_IEEE_Common<0x66>;
+  def RECIPSQRT_CLAMPED_r600 : RECIPSQRT_CLAMPED_Common<0x67>;
+  def RECIPSQRT_IEEE_r600 : RECIPSQRT_IEEE_Common<0x69>;
+  def FLT_TO_INT_r600 : FLT_TO_INT_Common<0x6b>;
+  def INT_TO_FLT_r600 : INT_TO_FLT_Common<0x6c>;
+  def FLT_TO_UINT_r600 : FLT_TO_UINT_Common<0x79>;
+  def UINT_TO_FLT_r600 : UINT_TO_FLT_Common<0x6d>;
+  def SIN_r600 : SIN_Common<0x6E>;
+  def COS_r600 : COS_Common<0x6F>;
+  def ASHR_r600 : ASHR_Common<0x70>;
+  def LSHR_r600 : LSHR_Common<0x71>;
+  def LSHL_r600 : LSHL_Common<0x72>;
+  def MULLO_INT_r600 : MULLO_INT_Common<0x73>;
+  def MULHI_INT_r600 : MULHI_INT_Common<0x74>;
+  def MULLO_UINT_r600 : MULLO_UINT_Common<0x75>;
+  def MULHI_UINT_r600 : MULHI_UINT_Common<0x76>;
+  def RECIP_UINT_r600 : RECIP_UINT_Common <0x78>;
+
+  defm DIV_r600 : DIV_Common<RECIP_IEEE_r600>;
+  def : POW_Common <LOG_IEEE_r600, EXP_IEEE_r600, MUL, R600_Reg32>;
+  def TGSI_LIT_Z_r600 : TGSI_LIT_Z_Common<MUL_LIT_r600, LOG_CLAMPED_r600, EXP_IEEE_r600>;
+
+  def : Pat<(fsqrt R600_Reg32:$src),
+    (MUL R600_Reg32:$src, (RECIPSQRT_CLAMPED_r600 R600_Reg32:$src))>;
+
+  def R600_ExportSwz : ExportSwzInst {
+    let Word1{20-17} = 1; // BURST_COUNT
+    let Word1{21} = eop;
+    let Word1{22} = 1; // VALID_PIXEL_MODE
+    let Word1{30-23} = inst;
+    let Word1{31} = 1; // BARRIER
+  }
+  defm : ExportPattern<R600_ExportSwz, 39>;
+
+  def R600_ExportBuf : ExportBufInst {
+    let Word1{20-17} = 1; // BURST_COUNT
+    let Word1{21} = eop;
+    let Word1{22} = 1; // VALID_PIXEL_MODE
+    let Word1{30-23} = inst;
+    let Word1{31} = 1; // BARRIER
+  }
+  defm : SteamOutputExportPattern<R600_ExportBuf, 0x20, 0x21, 0x22, 0x23>;
+}
+
+// Helper pattern for normalizing inputs to triginomic instructions for R700+
+// cards.
+class COS_PAT <InstR600 trig> : Pat<
+  (fcos R600_Reg32:$src),
+  (trig (MUL_IEEE (MOV_IMM_I32 CONST.TWO_PI_INV), R600_Reg32:$src))
+>;
+
+class SIN_PAT <InstR600 trig> : Pat<
+  (fsin R600_Reg32:$src),
+  (trig (MUL_IEEE (MOV_IMM_I32 CONST.TWO_PI_INV), R600_Reg32:$src))
+>;
+
+//===----------------------------------------------------------------------===//
+// R700 Only instructions
+//===----------------------------------------------------------------------===//
+
+let Predicates = [isR700] in {
+  def SIN_r700 : SIN_Common<0x6E>;
+  def COS_r700 : COS_Common<0x6F>;
+
+  // R700 normalizes inputs to SIN/COS the same as EG
+  def : SIN_PAT <SIN_r700>;
+  def : COS_PAT <COS_r700>;
+}
+
+//===----------------------------------------------------------------------===//
+// Evergreen Only instructions
+//===----------------------------------------------------------------------===//
+
+let Predicates = [isEG] in {
+
+def RECIP_IEEE_eg : RECIP_IEEE_Common<0x86>;
+defm DIV_eg : DIV_Common<RECIP_IEEE_eg>;
+
+def MULLO_INT_eg : MULLO_INT_Common<0x8F>;
+def MULHI_INT_eg : MULHI_INT_Common<0x90>;
+def MULLO_UINT_eg : MULLO_UINT_Common<0x91>;
+def MULHI_UINT_eg : MULHI_UINT_Common<0x92>;
+def RECIP_UINT_eg : RECIP_UINT_Common<0x94>;
+def RECIPSQRT_CLAMPED_eg : RECIPSQRT_CLAMPED_Common<0x87>;
+def EXP_IEEE_eg : EXP_IEEE_Common<0x81>;
+def LOG_IEEE_eg : LOG_IEEE_Common<0x83>;
+def RECIP_CLAMPED_eg : RECIP_CLAMPED_Common<0x84>;
+def RECIPSQRT_IEEE_eg : RECIPSQRT_IEEE_Common<0x89>;
+def SIN_eg : SIN_Common<0x8D>;
+def COS_eg : COS_Common<0x8E>;
+
+def : POW_Common <LOG_IEEE_eg, EXP_IEEE_eg, MUL, R600_Reg32>;
+def : SIN_PAT <SIN_eg>;
+def : COS_PAT <COS_eg>;
+def : Pat<(fsqrt R600_Reg32:$src),
+  (MUL R600_Reg32:$src, (RECIPSQRT_CLAMPED_eg R600_Reg32:$src))>;
+} // End Predicates = [isEG]
+
+//===----------------------------------------------------------------------===//
+// Evergreen / Cayman Instructions
+//===----------------------------------------------------------------------===//
+
+let Predicates = [isEGorCayman] in {
+
+  // BFE_UINT - bit_extract, an optimization for mask and shift
+  // Src0 = Input
+  // Src1 = Offset
+  // Src2 = Width
+  //
+  // bit_extract = (Input << (32 - Offset - Width)) >> (32 - Width)
+  //
+  // Example Usage:
+  // (Offset, Width)
+  //
+  // (0, 8)           = (Input << 24) >> 24  = (Input &  0xff)       >> 0
+  // (8, 8)           = (Input << 16) >> 24  = (Input &  0xffff)     >> 8
+  // (16,8)           = (Input <<  8) >> 24  = (Input &  0xffffff)   >> 16
+  // (24,8)           = (Input <<  0) >> 24  = (Input &  0xffffffff) >> 24
+  def BFE_UINT_eg : R600_3OP <0x4, "BFE_UINT",
+    [(set R600_Reg32:$dst, (int_AMDIL_bit_extract_u32 R600_Reg32:$src0,
+                                                      R600_Reg32:$src1,
+                                                      R600_Reg32:$src2))],
+    VecALU
+  >;
+
+  def BIT_ALIGN_INT_eg : R600_3OP <0xC, "BIT_ALIGN_INT",
+    [(set R600_Reg32:$dst, (AMDGPUbitalign R600_Reg32:$src0, R600_Reg32:$src1,
+                                          R600_Reg32:$src2))],
+    VecALU
+  >;
+
+  def MULADD_eg : MULADD_Common<0x14>;
+  def MULADD_IEEE_eg : MULADD_IEEE_Common<0x18>;
+  def ASHR_eg : ASHR_Common<0x15>;
+  def LSHR_eg : LSHR_Common<0x16>;
+  def LSHL_eg : LSHL_Common<0x17>;
+  def CNDE_eg : CNDE_Common<0x19>;
+  def CNDGT_eg : CNDGT_Common<0x1A>;
+  def CNDGE_eg : CNDGE_Common<0x1B>;
+  def MUL_LIT_eg : MUL_LIT_Common<0x1F>;
+  def LOG_CLAMPED_eg : LOG_CLAMPED_Common<0x82>;
+  defm DOT4_eg : DOT4_Common<0xBE>;
+  defm CUBE_eg : CUBE_Common<0xC0>;
+
+let hasSideEffects = 1 in {
+  def MOVA_INT_eg : R600_1OP <0xCC, "MOVA_INT", []>;
+}
+
+  def TGSI_LIT_Z_eg : TGSI_LIT_Z_Common<MUL_LIT_eg, LOG_CLAMPED_eg, EXP_IEEE_eg>;
+
+  def FLT_TO_INT_eg : FLT_TO_INT_Common<0x50> {
+    let Pattern = [];
+  }
+
+  def INT_TO_FLT_eg : INT_TO_FLT_Common<0x9B>;
+
+  def FLT_TO_UINT_eg : FLT_TO_UINT_Common<0x9A> {
+    let Pattern = [];
+  }
+
+  def UINT_TO_FLT_eg : UINT_TO_FLT_Common<0x9C>;
+
+  // TRUNC is used for the FLT_TO_INT instructions to work around a
+  // perceived problem where the rounding modes are applied differently
+  // depending on the instruction and the slot they are in.
+  // See:
+  // https://bugs.freedesktop.org/show_bug.cgi?id=50232
+  // Mesa commit: a1a0974401c467cb86ef818f22df67c21774a38c
+  //
+  // XXX: Lowering SELECT_CC will sometimes generate fp_to_[su]int nodes,
+  // which do not need to be truncated since the fp values are 0.0f or 1.0f.
+  // We should look into handling these cases separately.
+  def : Pat<(fp_to_sint R600_Reg32:$src0),
+    (FLT_TO_INT_eg (TRUNC R600_Reg32:$src0))>;
+
+  def : Pat<(fp_to_uint R600_Reg32:$src0),
+    (FLT_TO_UINT_eg (TRUNC R600_Reg32:$src0))>;
+
+  def EG_ExportSwz : ExportSwzInst {
+    let Word1{19-16} = 1; // BURST_COUNT
+    let Word1{20} = 1; // VALID_PIXEL_MODE
+    let Word1{21} = eop;
+    let Word1{29-22} = inst;
+    let Word1{30} = 0; // MARK
+    let Word1{31} = 1; // BARRIER
+  }
+  defm : ExportPattern<EG_ExportSwz, 83>;
+
+  def EG_ExportBuf : ExportBufInst {
+    let Word1{19-16} = 1; // BURST_COUNT
+    let Word1{20} = 1; // VALID_PIXEL_MODE
+    let Word1{21} = eop;
+    let Word1{29-22} = inst;
+    let Word1{30} = 0; // MARK
+    let Word1{31} = 1; // BARRIER
+  }
+  defm : SteamOutputExportPattern<EG_ExportBuf, 0x40, 0x41, 0x42, 0x43>;
+
+//===----------------------------------------------------------------------===//
+// Memory read/write instructions
+//===----------------------------------------------------------------------===//
+let usesCustomInserter = 1 in {
+
+class RAT_WRITE_CACHELESS_eg <dag ins, bits<4> comp_mask, string name,
+                              list<dag> pattern>
+    : EG_CF_RAT <0x57, 0x2, 0, (outs), ins,
+                 !strconcat(name, " $rw_gpr, $index_gpr, $eop"), pattern> {
+  let RIM         = 0;
+  // XXX: Have a separate instruction for non-indexed writes.
+  let TYPE        = 1;
+  let RW_REL      = 0;
+  let ELEM_SIZE   = 0;
+
+  let ARRAY_SIZE  = 0;
+  let COMP_MASK   = comp_mask;
+  let BURST_COUNT = 0;
+  let VPM         = 0;
+  let MARK        = 0;
+  let BARRIER     = 1;
+}
+
+} // End usesCustomInserter = 1
+
+// 32-bit store
+def RAT_WRITE_CACHELESS_32_eg : RAT_WRITE_CACHELESS_eg <
+  (ins R600_TReg32_X:$rw_gpr, R600_TReg32_X:$index_gpr, InstFlag:$eop),
+  0x1, "RAT_WRITE_CACHELESS_32_eg",
+  [(global_store (i32 R600_TReg32_X:$rw_gpr), R600_TReg32_X:$index_gpr)]
+>;
+
+//128-bit store
+def RAT_WRITE_CACHELESS_128_eg : RAT_WRITE_CACHELESS_eg <
+  (ins R600_Reg128:$rw_gpr, R600_TReg32_X:$index_gpr, InstFlag:$eop),
+  0xf, "RAT_WRITE_CACHELESS_128",
+  [(global_store (v4i32 R600_Reg128:$rw_gpr), R600_TReg32_X:$index_gpr)]
+>;
+
+class VTX_READ_eg <string name, bits<8> buffer_id, dag outs, list<dag> pattern>
+    : InstR600ISA <outs, (ins MEMxi:$ptr), name#" $dst, $ptr", pattern>,
+      VTX_WORD1_GPR, VTX_WORD0 {
+
+  // Static fields
+  let VC_INST = 0;
+  let FETCH_TYPE = 2;
+  let FETCH_WHOLE_QUAD = 0;
+  let BUFFER_ID = buffer_id;
+  let SRC_REL = 0;
+  // XXX: We can infer this field based on the SRC_GPR.  This would allow us
+  // to store vertex addresses in any channel, not just X.
+  let SRC_SEL_X = 0;
+  let DST_REL = 0;
+  // The docs say that if this bit is set, then DATA_FORMAT, NUM_FORMAT_ALL,
+  // FORMAT_COMP_ALL, SRF_MODE_ALL, and ENDIAN_SWAP fields will be ignored,
+  // however, based on my testing if USE_CONST_FIELDS is set, then all
+  // these fields need to be set to 0.
+  let USE_CONST_FIELDS = 0;
+  let NUM_FORMAT_ALL = 1;
+  let FORMAT_COMP_ALL = 0;
+  let SRF_MODE_ALL = 0;
+
+  let Inst{31-0} = Word0;
+  let Inst{63-32} = Word1;
+  // LLVM can only encode 64-bit instructions, so these fields are manually
+  // encoded in R600CodeEmitter
+  //
+  // bits<16> OFFSET;
+  // bits<2>  ENDIAN_SWAP = 0;
+  // bits<1>  CONST_BUF_NO_STRIDE = 0;
+  // bits<1>  MEGA_FETCH = 0;
+  // bits<1>  ALT_CONST = 0;
+  // bits<2>  BUFFER_INDEX_MODE = 0;
+
+
+
+  // VTX_WORD2 (LLVM can only encode 64-bit instructions, so WORD2 encoding
+  // is done in R600CodeEmitter
+  //
+  // Inst{79-64} = OFFSET;
+  // Inst{81-80} = ENDIAN_SWAP;
+  // Inst{82}    = CONST_BUF_NO_STRIDE;
+  // Inst{83}    = MEGA_FETCH;
+  // Inst{84}    = ALT_CONST;
+  // Inst{86-85} = BUFFER_INDEX_MODE;
+  // Inst{95-86} = 0; Reserved
+
+  // VTX_WORD3 (Padding)
+  //
+  // Inst{127-96} = 0;
+}
+
+class VTX_READ_8_eg <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_eg <"VTX_READ_8", buffer_id, (outs R600_TReg32_X:$dst),
+                   pattern> {
+
+  let MEGA_FETCH_COUNT = 1;
+  let DST_SEL_X = 0;
+  let DST_SEL_Y = 7;   // Masked
+  let DST_SEL_Z = 7;   // Masked
+  let DST_SEL_W = 7;   // Masked
+  let DATA_FORMAT = 1; // FMT_8
+}
+
+class VTX_READ_16_eg <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_eg <"VTX_READ_16", buffer_id, (outs R600_TReg32_X:$dst),
+                    pattern> {
+  let MEGA_FETCH_COUNT = 2;
+  let DST_SEL_X = 0;
+  let DST_SEL_Y = 7;   // Masked
+  let DST_SEL_Z = 7;   // Masked
+  let DST_SEL_W = 7;   // Masked
+  let DATA_FORMAT = 5; // FMT_16
+
+}
+
+class VTX_READ_32_eg <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_eg <"VTX_READ_32", buffer_id, (outs R600_TReg32_X:$dst),
+                   pattern> {
+
+  let MEGA_FETCH_COUNT = 4;
+  let DST_SEL_X        = 0;
+  let DST_SEL_Y        = 7;   // Masked
+  let DST_SEL_Z        = 7;   // Masked
+  let DST_SEL_W        = 7;   // Masked
+  let DATA_FORMAT      = 0xD; // COLOR_32
+
+  // This is not really necessary, but there were some GPU hangs that appeared
+  // to be caused by ALU instructions in the next instruction group that wrote
+  // to the $ptr registers of the VTX_READ.
+  // e.g.
+  // %T3_X<def> = VTX_READ_PARAM_32_eg %T2_X<kill>, 24
+  // %T2_X<def> = MOV %ZERO
+  //Adding this constraint prevents this from happening.
+  let Constraints = "$ptr.ptr = $dst";
+}
+
+class VTX_READ_128_eg <bits<8> buffer_id, list<dag> pattern>
+    : VTX_READ_eg <"VTX_READ_128", buffer_id, (outs R600_Reg128:$dst),
+                   pattern> {
+
+  let MEGA_FETCH_COUNT = 16;
+  let DST_SEL_X        =  0;
+  let DST_SEL_Y        =  1;
+  let DST_SEL_Z        =  2;
+  let DST_SEL_W        =  3;
+  let DATA_FORMAT      =  0x22; // COLOR_32_32_32_32
+
+  // XXX: Need to force VTX_READ_128 instructions to write to the same register
+  // that holds its buffer address to avoid potential hangs.  We can't use
+  // the same constraint as VTX_READ_32_eg, because the $ptr.ptr and $dst
+  // registers are different sizes.
+}
+
+//===----------------------------------------------------------------------===//
+// VTX Read from parameter memory space
+//===----------------------------------------------------------------------===//
+
+def VTX_READ_PARAM_8_eg : VTX_READ_8_eg <0,
+  [(set (i32 R600_TReg32_X:$dst), (load_param_zexti8 ADDRVTX_READ:$ptr))]
+>;
+
+def VTX_READ_PARAM_16_eg : VTX_READ_16_eg <0,
+  [(set (i32 R600_TReg32_X:$dst), (load_param_zexti16 ADDRVTX_READ:$ptr))]
+>;
+
+def VTX_READ_PARAM_32_eg : VTX_READ_32_eg <0,
+  [(set (i32 R600_TReg32_X:$dst), (load_param ADDRVTX_READ:$ptr))]
+>;
+
+def VTX_READ_PARAM_128_eg : VTX_READ_128_eg <0,
+  [(set (v4i32 R600_Reg128:$dst), (load_param ADDRVTX_READ:$ptr))]
+>;
+
+//===----------------------------------------------------------------------===//
+// VTX Read from global memory space
+//===----------------------------------------------------------------------===//
+
+// 8-bit reads
+def VTX_READ_GLOBAL_8_eg : VTX_READ_8_eg <1,
+  [(set (i32 R600_TReg32_X:$dst), (zextloadi8_global ADDRVTX_READ:$ptr))]
+>;
+
+// 32-bit reads
+def VTX_READ_GLOBAL_32_eg : VTX_READ_32_eg <1,
+  [(set (i32 R600_TReg32_X:$dst), (global_load ADDRVTX_READ:$ptr))]
+>;
+
+// 128-bit reads
+def VTX_READ_GLOBAL_128_eg : VTX_READ_128_eg <1,
+  [(set (v4i32 R600_Reg128:$dst), (global_load ADDRVTX_READ:$ptr))]
+>;
+
+//===----------------------------------------------------------------------===//
+// Constant Loads
+// XXX: We are currently storing all constants in the global address space.
+//===----------------------------------------------------------------------===//
+
+def CONSTANT_LOAD_eg : VTX_READ_32_eg <1,
+  [(set (i32 R600_TReg32_X:$dst), (constant_load ADDRVTX_READ:$ptr))]
+>;
+
+}
+
+//===----------------------------------------------------------------------===//
+// Regist loads and stores - for indirect addressing
+//===----------------------------------------------------------------------===//
+
+defm R600_ : RegisterLoadStore <R600_Reg32, FRAMEri, ADDRIndirect>;
+
+let Predicates = [isCayman] in {
+
+let isVector = 1 in {
+
+def RECIP_IEEE_cm : RECIP_IEEE_Common<0x86>;
+
+def MULLO_INT_cm : MULLO_INT_Common<0x8F>;
+def MULHI_INT_cm : MULHI_INT_Common<0x90>;
+def MULLO_UINT_cm : MULLO_UINT_Common<0x91>;
+def MULHI_UINT_cm : MULHI_UINT_Common<0x92>;
+def RECIPSQRT_CLAMPED_cm : RECIPSQRT_CLAMPED_Common<0x87>;
+def EXP_IEEE_cm : EXP_IEEE_Common<0x81>;
+def LOG_IEEE_cm : LOG_IEEE_Common<0x83>;
+def RECIP_CLAMPED_cm : RECIP_CLAMPED_Common<0x84>;
+def RECIPSQRT_IEEE_cm : RECIPSQRT_IEEE_Common<0x89>;
+def SIN_cm : SIN_Common<0x8D>;
+def COS_cm : COS_Common<0x8E>;
+} // End isVector = 1
+
+def : POW_Common <LOG_IEEE_cm, EXP_IEEE_cm, MUL, R600_Reg32>;
+def : SIN_PAT <SIN_cm>;
+def : COS_PAT <COS_cm>;
+
+defm DIV_cm : DIV_Common<RECIP_IEEE_cm>;
+
+// RECIP_UINT emulation for Cayman
+def : Pat <
+  (AMDGPUurecip R600_Reg32:$src0),
+  (FLT_TO_UINT_eg (MUL_IEEE (RECIP_IEEE_cm (UINT_TO_FLT_eg R600_Reg32:$src0)),
+                            (MOV_IMM_I32 0x4f800000)))
+>;
+
+
+def : Pat<(fsqrt R600_Reg32:$src),
+  (MUL R600_Reg32:$src, (RECIPSQRT_CLAMPED_cm R600_Reg32:$src))>;
+
+} // End isCayman
+
+//===----------------------------------------------------------------------===//
+// Branch Instructions
+//===----------------------------------------------------------------------===//
+
+
+def IF_PREDICATE_SET  : ILFormat<(outs), (ins GPRI32:$src),
+  "IF_PREDICATE_SET $src", []>;
+
+def PREDICATED_BREAK : ILFormat<(outs), (ins GPRI32:$src),
+  "PREDICATED_BREAK $src", []>;
+
+//===----------------------------------------------------------------------===//
+// Pseudo instructions
+//===----------------------------------------------------------------------===//
+
+let isPseudo = 1 in {
+
+def PRED_X : InstR600 <
+  0, (outs R600_Predicate_Bit:$dst),
+  (ins R600_Reg32:$src0, i32imm:$src1, i32imm:$flags),
+  "", [], NullALU> {
+  let FlagOperandIdx = 3;
+}
+
+let isTerminator = 1, isBranch = 1 in {
+def JUMP_COND : InstR600 <0x10,
+          (outs),
+          (ins brtarget:$target, R600_Predicate_Bit:$p),
+          "JUMP $target ($p)",
+          [], AnyALU
+  >;
+
+def JUMP : InstR600 <0x10,
+          (outs),
+          (ins brtarget:$target),
+          "JUMP $target",
+          [], AnyALU
+  >
+{
+  let isPredicable = 1;
+  let isBarrier = 1;
+}
+
+}  // End isTerminator = 1, isBranch = 1
+
+let usesCustomInserter = 1 in {
+
+let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in {
+
+def MASK_WRITE : AMDGPUShaderInst <
+    (outs),
+    (ins R600_Reg32:$src),
+    "MASK_WRITE $src",
+    []
+>;
+
+} // End mayLoad = 0, mayStore = 0, hasSideEffects = 1
+
+
+def TXD: AMDGPUShaderInst <
+  (outs R600_Reg128:$dst),
+  (ins R600_Reg128:$src0, R600_Reg128:$src1, R600_Reg128:$src2, i32imm:$resourceId, i32imm:$samplerId, i32imm:$textureTarget),
+  "TXD $dst, $src0, $src1, $src2, $resourceId, $samplerId, $textureTarget",
+  [(set R600_Reg128:$dst, (int_AMDGPU_txd R600_Reg128:$src0, R600_Reg128:$src1, R600_Reg128:$src2, imm:$resourceId, imm:$samplerId, imm:$textureTarget))]
+>;
+
+def TXD_SHADOW: AMDGPUShaderInst <
+  (outs R600_Reg128:$dst),
+  (ins R600_Reg128:$src0, R600_Reg128:$src1, R600_Reg128:$src2, i32imm:$resourceId, i32imm:$samplerId, i32imm:$textureTarget),
+  "TXD_SHADOW $dst, $src0, $src1, $src2, $resourceId, $samplerId, $textureTarget",
+  [(set R600_Reg128:$dst, (int_AMDGPU_txd R600_Reg128:$src0, R600_Reg128:$src1, R600_Reg128:$src2, imm:$resourceId, imm:$samplerId, TEX_SHADOW:$textureTarget))]
+>;
+
+} // End isPseudo = 1
+} // End usesCustomInserter = 1
+
+def CLAMP_R600 :  CLAMP <R600_Reg32>;
+def FABS_R600 : FABS<R600_Reg32>;
+def FNEG_R600 : FNEG<R600_Reg32>;
+
+//===---------------------------------------------------------------------===//
+// Return instruction
+//===---------------------------------------------------------------------===//
+let isTerminator = 1, isReturn = 1, hasCtrlDep = 1,
+    usesCustomInserter = 1 in {
+  def RETURN          : ILFormat<(outs), (ins variable_ops),
+      "RETURN", [(IL_retflag)]>;
+}
+
+
+//===----------------------------------------------------------------------===//
+// Constant Buffer Addressing Support
+//===----------------------------------------------------------------------===//
+
+let usesCustomInserter = 1, isCodeGenOnly = 1, isPseudo = 1, Namespace = "AMDGPU"  in {
+def CONST_COPY : Instruction {
+  let OutOperandList = (outs R600_Reg32:$dst);
+  let InOperandList = (ins i32imm:$src);
+  let Pattern =
+      [(set R600_Reg32:$dst, (CONST_ADDRESS ADDRGA_CONST_OFFSET:$src))];
+  let AsmString = "CONST_COPY";
+  let neverHasSideEffects = 1;
+  let isAsCheapAsAMove = 1;
+  let Itinerary = NullALU;
+}
+} // end usesCustomInserter = 1, isCodeGenOnly = 1, isPseudo = 1, Namespace = "AMDGPU"
+
+def TEX_VTX_CONSTBUF :
+  InstR600ISA <(outs R600_Reg128:$dst), (ins MEMxi:$ptr, i32imm:$BUFFER_ID), "VTX_READ_eg $dst, $ptr",
+      [(set R600_Reg128:$dst, (CONST_ADDRESS ADDRGA_VAR_OFFSET:$ptr, (i32 imm:$BUFFER_ID)))]>,
+  VTX_WORD1_GPR, VTX_WORD0 {
+
+  let VC_INST = 0;
+  let FETCH_TYPE = 2;
+  let FETCH_WHOLE_QUAD = 0;
+  let SRC_REL = 0;
+  let SRC_SEL_X = 0;
+  let DST_REL = 0;
+  let USE_CONST_FIELDS = 0;
+  let NUM_FORMAT_ALL = 2;
+  let FORMAT_COMP_ALL = 1;
+  let SRF_MODE_ALL = 1;
+  let MEGA_FETCH_COUNT = 16;
+  let DST_SEL_X        = 0;
+  let DST_SEL_Y        = 1;
+  let DST_SEL_Z        = 2;
+  let DST_SEL_W        = 3;
+  let DATA_FORMAT      = 35;
+
+  let Inst{31-0} = Word0;
+  let Inst{63-32} = Word1;
+
+// LLVM can only encode 64-bit instructions, so these fields are manually
+// encoded in R600CodeEmitter
+//
+// bits<16> OFFSET;
+// bits<2>  ENDIAN_SWAP = 0;
+// bits<1>  CONST_BUF_NO_STRIDE = 0;
+// bits<1>  MEGA_FETCH = 0;
+// bits<1>  ALT_CONST = 0;
+// bits<2>  BUFFER_INDEX_MODE = 0;
+
+
+
+// VTX_WORD2 (LLVM can only encode 64-bit instructions, so WORD2 encoding
+// is done in R600CodeEmitter
+//
+// Inst{79-64} = OFFSET;
+// Inst{81-80} = ENDIAN_SWAP;
+// Inst{82}    = CONST_BUF_NO_STRIDE;
+// Inst{83}    = MEGA_FETCH;
+// Inst{84}    = ALT_CONST;
+// Inst{86-85} = BUFFER_INDEX_MODE;
+// Inst{95-86} = 0; Reserved
+
+// VTX_WORD3 (Padding)
+//
+// Inst{127-96} = 0;
+}
+
+def TEX_VTX_TEXBUF:
+  InstR600ISA <(outs R600_Reg128:$dst), (ins MEMxi:$ptr, i32imm:$BUFFER_ID), "TEX_VTX_EXPLICIT_READ $dst, $ptr",
+      [(set R600_Reg128:$dst, (int_R600_load_texbuf ADDRGA_VAR_OFFSET:$ptr, imm:$BUFFER_ID))]>,
+VTX_WORD1_GPR, VTX_WORD0 {
+
+let VC_INST = 0;
+let FETCH_TYPE = 2;
+let FETCH_WHOLE_QUAD = 0;
+let SRC_REL = 0;
+let SRC_SEL_X = 0;
+let DST_REL = 0;
+let USE_CONST_FIELDS = 1;
+let NUM_FORMAT_ALL = 0;
+let FORMAT_COMP_ALL = 0;
+let SRF_MODE_ALL = 1;
+let MEGA_FETCH_COUNT = 16;
+let DST_SEL_X        = 0;
+let DST_SEL_Y        = 1;
+let DST_SEL_Z        = 2;
+let DST_SEL_W        = 3;
+let DATA_FORMAT      = 0;
+
+let Inst{31-0} = Word0;
+let Inst{63-32} = Word1;
+
+// LLVM can only encode 64-bit instructions, so these fields are manually
+// encoded in R600CodeEmitter
+//
+// bits<16> OFFSET;
+// bits<2>  ENDIAN_SWAP = 0;
+// bits<1>  CONST_BUF_NO_STRIDE = 0;
+// bits<1>  MEGA_FETCH = 0;
+// bits<1>  ALT_CONST = 0;
+// bits<2>  BUFFER_INDEX_MODE = 0;
+
+
+
+// VTX_WORD2 (LLVM can only encode 64-bit instructions, so WORD2 encoding
+// is done in R600CodeEmitter
+//
+// Inst{79-64} = OFFSET;
+// Inst{81-80} = ENDIAN_SWAP;
+// Inst{82}    = CONST_BUF_NO_STRIDE;
+// Inst{83}    = MEGA_FETCH;
+// Inst{84}    = ALT_CONST;
+// Inst{86-85} = BUFFER_INDEX_MODE;
+// Inst{95-86} = 0; Reserved
+
+// VTX_WORD3 (Padding)
+//
+// Inst{127-96} = 0;
+}
+
+
+
+//===--------------------------------------------------------------------===//
+// Instructions support
+//===--------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
+// Custom Inserter for Branches and returns, this eventually will be a
+// seperate pass
+//===---------------------------------------------------------------------===//
+let isTerminator = 1, usesCustomInserter = 1, isBranch = 1, isBarrier = 1 in {
+  def BRANCH : ILFormat<(outs), (ins brtarget:$target),
+      "; Pseudo unconditional branch instruction",
+      [(br bb:$target)]>;
+  defm BRANCH_COND : BranchConditional<IL_brcond>;
+}
+
+//===---------------------------------------------------------------------===//
+// Flow and Program control Instructions
+//===---------------------------------------------------------------------===//
+let isTerminator=1 in {
+  def SWITCH      : ILFormat< (outs), (ins GPRI32:$src),
+  !strconcat("SWITCH", " $src"), []>;
+  def CASE        : ILFormat< (outs), (ins GPRI32:$src),
+      !strconcat("CASE", " $src"), []>;
+  def BREAK       : ILFormat< (outs), (ins),
+      "BREAK", []>;
+  def CONTINUE    : ILFormat< (outs), (ins),
+      "CONTINUE", []>;
+  def DEFAULT     : ILFormat< (outs), (ins),
+      "DEFAULT", []>;
+  def ELSE        : ILFormat< (outs), (ins),
+      "ELSE", []>;
+  def ENDSWITCH   : ILFormat< (outs), (ins),
+      "ENDSWITCH", []>;
+  def ENDMAIN     : ILFormat< (outs), (ins),
+      "ENDMAIN", []>;
+  def END         : ILFormat< (outs), (ins),
+      "END", []>;
+  def ENDFUNC     : ILFormat< (outs), (ins),
+      "ENDFUNC", []>;
+  def ENDIF       : ILFormat< (outs), (ins),
+      "ENDIF", []>;
+  def WHILELOOP   : ILFormat< (outs), (ins),
+      "WHILE", []>;
+  def ENDLOOP     : ILFormat< (outs), (ins),
+      "ENDLOOP", []>;
+  def FUNC        : ILFormat< (outs), (ins),
+      "FUNC", []>;
+  def RETDYN      : ILFormat< (outs), (ins),
+      "RET_DYN", []>;
+  // This opcode has custom swizzle pattern encoded in Swizzle Encoder
+  defm IF_LOGICALNZ  : BranchInstr<"IF_LOGICALNZ">;
+  // This opcode has custom swizzle pattern encoded in Swizzle Encoder
+  defm IF_LOGICALZ   : BranchInstr<"IF_LOGICALZ">;
+  // This opcode has custom swizzle pattern encoded in Swizzle Encoder
+  defm BREAK_LOGICALNZ : BranchInstr<"BREAK_LOGICALNZ">;
+  // This opcode has custom swizzle pattern encoded in Swizzle Encoder
+  defm BREAK_LOGICALZ : BranchInstr<"BREAK_LOGICALZ">;
+  // This opcode has custom swizzle pattern encoded in Swizzle Encoder
+  defm CONTINUE_LOGICALNZ : BranchInstr<"CONTINUE_LOGICALNZ">;
+  // This opcode has custom swizzle pattern encoded in Swizzle Encoder
+  defm CONTINUE_LOGICALZ : BranchInstr<"CONTINUE_LOGICALZ">;
+  defm IFC         : BranchInstr2<"IFC">;
+  defm BREAKC      : BranchInstr2<"BREAKC">;
+  defm CONTINUEC   : BranchInstr2<"CONTINUEC">;
+}
+
+//===----------------------------------------------------------------------===//
+// ISel Patterns
+//===----------------------------------------------------------------------===//
+
+// CND*_INT Pattterns for f32 True / False values
+
+class CND_INT_f32 <InstR600 cnd, CondCode cc> : Pat <
+  (selectcc (i32 R600_Reg32:$src0), 0, (f32 R600_Reg32:$src1),
+                                            R600_Reg32:$src2, cc),
+  (cnd R600_Reg32:$src0, R600_Reg32:$src1, R600_Reg32:$src2)
+>;
+
+def : CND_INT_f32 <CNDE_INT,  SETEQ>;
+def : CND_INT_f32 <CNDGT_INT, SETGT>;
+def : CND_INT_f32 <CNDGE_INT, SETGE>;
+
+//CNDGE_INT extra pattern
+def : Pat <
+  (selectcc (i32 R600_Reg32:$src0), -1, (i32 R600_Reg32:$src1),
+                                        (i32 R600_Reg32:$src2), COND_GT),
+  (CNDGE_INT R600_Reg32:$src0, R600_Reg32:$src1, R600_Reg32:$src2)
+>;
+
+// KIL Patterns
+def KILP : Pat <
+  (int_AMDGPU_kilp),
+  (MASK_WRITE (KILLGT (f32 ONE), (f32 ZERO)))
+>;
+
+def KIL : Pat <
+  (int_AMDGPU_kill R600_Reg32:$src0),
+  (MASK_WRITE (KILLGT (f32 ZERO), (f32 R600_Reg32:$src0)))
+>;
+
+// SGT Reverse args
+def : Pat <
+  (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, FP_ONE, FP_ZERO, COND_LT),
+  (SGT R600_Reg32:$src1, R600_Reg32:$src0)
+>;
+
+// SGE Reverse args
+def : Pat <
+  (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, FP_ONE, FP_ZERO, COND_LE),
+  (SGE R600_Reg32:$src1, R600_Reg32:$src0)
+>;
+
+// SETGT_DX10 reverse args
+def : Pat <
+  (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, COND_LT),
+  (SETGT_DX10 R600_Reg32:$src1, R600_Reg32:$src0)
+>;
+
+// SETGE_DX10 reverse args
+def : Pat <
+  (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, COND_LE),
+  (SETGE_DX10 R600_Reg32:$src1, R600_Reg32:$src0)
+>;
+
+// SETGT_INT reverse args
+def : Pat <
+  (selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETLT),
+  (SETGT_INT R600_Reg32:$src1, R600_Reg32:$src0)
+>;
+
+// SETGE_INT reverse args
+def : Pat <
+  (selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETLE),
+  (SETGE_INT R600_Reg32:$src1, R600_Reg32:$src0)
+>;
+
+// SETGT_UINT reverse args
+def : Pat <
+  (selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETULT),
+  (SETGT_UINT R600_Reg32:$src1, R600_Reg32:$src0)
+>;
+
+// SETGE_UINT reverse args
+def : Pat <
+  (selectcc (i32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETULE),
+  (SETGE_UINT R600_Reg32:$src1, R600_Reg32:$src0)
+>;
+
+// The next two patterns are special cases for handling 'true if ordered' and
+// 'true if unordered' conditionals.  The assumption here is that the behavior of
+// SETE and SNE conforms to the Direct3D 10 rules for floating point values
+// described here:
+// http://msdn.microsoft.com/en-us/library/windows/desktop/cc308050.aspx#alpha_32_bit
+// We assume that  SETE returns false when one of the operands is NAN and
+// SNE returns true when on of the operands is NAN
+
+//SETE - 'true if ordered'
+def : Pat <
+  (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, FP_ONE, FP_ZERO, SETO),
+  (SETE R600_Reg32:$src0, R600_Reg32:$src1)
+>;
+
+//SETE_DX10 - 'true if ordered'
+def : Pat <
+  (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETO),
+  (SETE_DX10 R600_Reg32:$src0, R600_Reg32:$src1)
+>;
+
+//SNE - 'true if unordered'
+def : Pat <
+  (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, FP_ONE, FP_ZERO, SETUO),
+  (SNE R600_Reg32:$src0, R600_Reg32:$src1)
+>;
+
+//SETNE_DX10 - 'true if ordered'
+def : Pat <
+  (selectcc (f32 R600_Reg32:$src0), R600_Reg32:$src1, -1, 0, SETUO),
+  (SETNE_DX10 R600_Reg32:$src0, R600_Reg32:$src1)
+>;
+
+def : Extract_Element <f32, v4f32, R600_Reg128, 0, sub0>;
+def : Extract_Element <f32, v4f32, R600_Reg128, 1, sub1>;
+def : Extract_Element <f32, v4f32, R600_Reg128, 2, sub2>;
+def : Extract_Element <f32, v4f32, R600_Reg128, 3, sub3>;
+
+def : Insert_Element <f32, v4f32, R600_Reg32, R600_Reg128, 0, sub0>;
+def : Insert_Element <f32, v4f32, R600_Reg32, R600_Reg128, 1, sub1>;
+def : Insert_Element <f32, v4f32, R600_Reg32, R600_Reg128, 2, sub2>;
+def : Insert_Element <f32, v4f32, R600_Reg32, R600_Reg128, 3, sub3>;
+
+def : Extract_Element <i32, v4i32, R600_Reg128, 0, sub0>;
+def : Extract_Element <i32, v4i32, R600_Reg128, 1, sub1>;
+def : Extract_Element <i32, v4i32, R600_Reg128, 2, sub2>;
+def : Extract_Element <i32, v4i32, R600_Reg128, 3, sub3>;
+
+def : Insert_Element <i32, v4i32, R600_Reg32, R600_Reg128, 0, sub0>;
+def : Insert_Element <i32, v4i32, R600_Reg32, R600_Reg128, 1, sub1>;
+def : Insert_Element <i32, v4i32, R600_Reg32, R600_Reg128, 2, sub2>;
+def : Insert_Element <i32, v4i32, R600_Reg32, R600_Reg128, 3, sub3>;
+
+def : Vector4_Build <v4f32, R600_Reg128, f32, R600_Reg32>;
+def : Vector4_Build <v4i32, R600_Reg128, i32, R600_Reg32>;
+
+// bitconvert patterns
+
+def : BitConvert <i32, f32, R600_Reg32>;
+def : BitConvert <f32, i32, R600_Reg32>;
+def : BitConvert <v4f32, v4i32, R600_Reg128>;
+def : BitConvert <v4i32, v4f32, R600_Reg128>;
+
+// DWORDADDR pattern
+def : DwordAddrPat  <i32, R600_Reg32>;
+
+} // End isR600toCayman Predicate
diff --git a/lib/Target/R600/R600Intrinsics.td b/lib/Target/R600/R600Intrinsics.td
new file mode 100644
index 000000000000..dc8980aef146
--- /dev/null
+++ b/lib/Target/R600/R600Intrinsics.td
@@ -0,0 +1,31 @@
+//===-- R600Intrinsics.td - R600 Instrinsic defs -------*- tablegen -*-----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// R600 Intrinsic Definitions
+//
+//===----------------------------------------------------------------------===//
+
+let TargetPrefix = "R600", isTarget = 1 in {
+  def int_R600_load_input :
+    Intrinsic<[llvm_float_ty], [llvm_i32_ty], [IntrNoMem]>;
+  def int_R600_interp_input :
+    Intrinsic<[llvm_float_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_R600_load_texbuf :
+    Intrinsic<[llvm_v4f32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_R600_store_swizzle :
+    Intrinsic<[], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty], []>;
+  def int_R600_store_stream_output :
+    Intrinsic<[], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], []>;
+  def int_R600_store_pixel_depth :
+      Intrinsic<[], [llvm_float_ty], []>;
+  def int_R600_store_pixel_stencil :
+      Intrinsic<[], [llvm_float_ty], []>;
+  def int_R600_store_dummy :
+      Intrinsic<[], [llvm_i32_ty], []>;
+}
diff --git a/lib/Target/R600/R600MachineFunctionInfo.cpp b/lib/Target/R600/R600MachineFunctionInfo.cpp
new file mode 100644
index 000000000000..018b40363363
--- /dev/null
+++ b/lib/Target/R600/R600MachineFunctionInfo.cpp
@@ -0,0 +1,18 @@
+//===-- R600MachineFunctionInfo.cpp - R600 Machine Function Info-*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//===----------------------------------------------------------------------===//
+
+#include "R600MachineFunctionInfo.h"
+
+using namespace llvm;
+
+R600MachineFunctionInfo::R600MachineFunctionInfo(const MachineFunction &MF)
+  : AMDGPUMachineFunction(MF) { }
+
+
diff --git a/lib/Target/R600/R600MachineFunctionInfo.h b/lib/Target/R600/R600MachineFunctionInfo.h
new file mode 100644
index 000000000000..99c1f91b09b1
--- /dev/null
+++ b/lib/Target/R600/R600MachineFunctionInfo.h
@@ -0,0 +1,32 @@
+//===-- R600MachineFunctionInfo.h - R600 Machine Function Info ----*- C++ -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+//===----------------------------------------------------------------------===//
+
+#ifndef R600MACHINEFUNCTIONINFO_H
+#define R600MACHINEFUNCTIONINFO_H
+
+#include "llvm/ADT/BitVector.h"
+#include "llvm/CodeGen/SelectionDAG.h"
+#include "AMDGPUMachineFunction.h"
+#include <vector>
+
+namespace llvm {
+
+class R600MachineFunctionInfo : public AMDGPUMachineFunction {
+public:
+  R600MachineFunctionInfo(const MachineFunction &MF);
+  SmallVector<unsigned, 4> LiveOuts;
+  std::vector<unsigned> IndirectRegs;
+};
+
+} // End llvm namespace
+
+#endif //R600MACHINEFUNCTIONINFO_H
diff --git a/lib/Target/R600/R600MachineScheduler.cpp b/lib/Target/R600/R600MachineScheduler.cpp
new file mode 100644
index 000000000000..a777142a9e70
--- /dev/null
+++ b/lib/Target/R600/R600MachineScheduler.cpp
@@ -0,0 +1,427 @@
+//===-- R600MachineScheduler.cpp - R600 Scheduler Interface -*- C++ -*-----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief R600 Machine Scheduler interface
+// TODO: Scheduling is optimised for VLIW4 arch, modify it to support TRANS slot
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "misched"
+
+#include "R600MachineScheduler.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/LiveIntervalAnalysis.h"
+#include "llvm/Pass.h"
+#include "llvm/PassManager.h"
+#include "llvm/Support/raw_ostream.h"
+#include <set>
+
+using namespace llvm;
+
+void R600SchedStrategy::initialize(ScheduleDAGMI *dag) {
+
+  DAG = dag;
+  TII = static_cast<const R600InstrInfo*>(DAG->TII);
+  TRI = static_cast<const R600RegisterInfo*>(DAG->TRI);
+  MRI = &DAG->MRI;
+  Available[IDAlu]->clear();
+  Available[IDFetch]->clear();
+  Available[IDOther]->clear();
+  CurInstKind = IDOther;
+  CurEmitted = 0;
+  OccupedSlotsMask = 15;
+  InstKindLimit[IDAlu] = TII->getMaxAlusPerClause();
+
+
+  const AMDGPUSubtarget &ST = DAG->TM.getSubtarget<AMDGPUSubtarget>();
+  if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD5XXX) {
+    InstKindLimit[IDFetch] = 7; // 8 minus 1 for security
+  } else {
+    InstKindLimit[IDFetch] = 15; // 16 minus 1 for security
+  }
+}
+
+void R600SchedStrategy::MoveUnits(ReadyQueue *QSrc, ReadyQueue *QDst)
+{
+  if (QSrc->empty())
+    return;
+  for (ReadyQueue::iterator I = QSrc->begin(),
+      E = QSrc->end(); I != E; ++I) {
+    (*I)->NodeQueueId &= ~QSrc->getID();
+    QDst->push(*I);
+  }
+  QSrc->clear();
+}
+
+SUnit* R600SchedStrategy::pickNode(bool &IsTopNode) {
+  SUnit *SU = 0;
+  IsTopNode = true;
+  NextInstKind = IDOther;
+
+  // check if we might want to switch current clause type
+  bool AllowSwitchToAlu = (CurInstKind == IDOther) ||
+      (CurEmitted > InstKindLimit[CurInstKind]) ||
+      (Available[CurInstKind]->empty());
+  bool AllowSwitchFromAlu = (CurEmitted > InstKindLimit[CurInstKind]) &&
+      (!Available[IDFetch]->empty() || !Available[IDOther]->empty());
+
+  if ((AllowSwitchToAlu && CurInstKind != IDAlu) ||
+      (!AllowSwitchFromAlu && CurInstKind == IDAlu)) {
+    // try to pick ALU
+    SU = pickAlu();
+    if (SU) {
+      if (CurEmitted >  InstKindLimit[IDAlu])
+        CurEmitted = 0;
+      NextInstKind = IDAlu;
+    }
+  }
+
+  if (!SU) {
+    // try to pick FETCH
+    SU = pickOther(IDFetch);
+    if (SU)
+      NextInstKind = IDFetch;
+  }
+
+  // try to pick other
+  if (!SU) {
+    SU = pickOther(IDOther);
+    if (SU)
+      NextInstKind = IDOther;
+  }
+
+  DEBUG(
+      if (SU) {
+        dbgs() << "picked node: ";
+        SU->dump(DAG);
+      } else {
+        dbgs() << "NO NODE ";
+        for (int i = 0; i < IDLast; ++i) {
+          Available[i]->dump();
+          Pending[i]->dump();
+        }
+        for (unsigned i = 0; i < DAG->SUnits.size(); i++) {
+          const SUnit &S = DAG->SUnits[i];
+          if (!S.isScheduled)
+            S.dump(DAG);
+        }
+      }
+  );
+
+  return SU;
+}
+
+void R600SchedStrategy::schedNode(SUnit *SU, bool IsTopNode) {
+
+  DEBUG(dbgs() << "scheduled: ");
+  DEBUG(SU->dump(DAG));
+
+  if (NextInstKind != CurInstKind) {
+    DEBUG(dbgs() << "Instruction Type Switch\n");
+    if (NextInstKind != IDAlu)
+      OccupedSlotsMask = 15;
+    CurEmitted = 0;
+    CurInstKind = NextInstKind;
+  }
+
+  if (CurInstKind == IDAlu) {
+    switch (getAluKind(SU)) {
+    case AluT_XYZW:
+      CurEmitted += 4;
+      break;
+    case AluDiscarded:
+      break;
+    default: {
+      ++CurEmitted;
+      for (MachineInstr::mop_iterator It = SU->getInstr()->operands_begin(),
+          E = SU->getInstr()->operands_end(); It != E; ++It) {
+        MachineOperand &MO = *It;
+        if (MO.isReg() && MO.getReg() == AMDGPU::ALU_LITERAL_X)
+          ++CurEmitted;
+      }
+    }
+    }
+  } else {
+    ++CurEmitted;
+  }
+
+
+  DEBUG(dbgs() << CurEmitted << " Instructions Emitted in this clause\n");
+
+  if (CurInstKind != IDFetch) {
+    MoveUnits(Pending[IDFetch], Available[IDFetch]);
+  }
+  MoveUnits(Pending[IDOther], Available[IDOther]);
+}
+
+void R600SchedStrategy::releaseTopNode(SUnit *SU) {
+  int IK = getInstKind(SU);
+
+  DEBUG(dbgs() << IK << " <= ");
+  DEBUG(SU->dump(DAG));
+
+  Pending[IK]->push(SU);
+}
+
+void R600SchedStrategy::releaseBottomNode(SUnit *SU) {
+}
+
+bool R600SchedStrategy::regBelongsToClass(unsigned Reg,
+                                          const TargetRegisterClass *RC) const {
+  if (!TargetRegisterInfo::isVirtualRegister(Reg)) {
+    return RC->contains(Reg);
+  } else {
+    return MRI->getRegClass(Reg) == RC;
+  }
+}
+
+R600SchedStrategy::AluKind R600SchedStrategy::getAluKind(SUnit *SU) const {
+  MachineInstr *MI = SU->getInstr();
+
+    switch (MI->getOpcode()) {
+    case AMDGPU::INTERP_PAIR_XY:
+    case AMDGPU::INTERP_PAIR_ZW:
+    case AMDGPU::INTERP_VEC_LOAD:
+      return AluT_XYZW;
+    case AMDGPU::COPY:
+      if (TargetRegisterInfo::isPhysicalRegister(MI->getOperand(1).getReg())) {
+        // %vregX = COPY Tn_X is likely to be discarded in favor of an
+        // assignement of Tn_X to %vregX, don't considers it in scheduling
+        return AluDiscarded;
+      }
+      else if (MI->getOperand(1).isUndef()) {
+        // MI will become a KILL, don't considers it in scheduling
+        return AluDiscarded;
+      }
+    default:
+      break;
+    }
+
+    // Does the instruction take a whole IG ?
+    if(TII->isVector(*MI) ||
+        TII->isCubeOp(MI->getOpcode()) ||
+        TII->isReductionOp(MI->getOpcode()))
+      return AluT_XYZW;
+
+    // Is the result already assigned to a channel ?
+    unsigned DestSubReg = MI->getOperand(0).getSubReg();
+    switch (DestSubReg) {
+    case AMDGPU::sub0:
+      return AluT_X;
+    case AMDGPU::sub1:
+      return AluT_Y;
+    case AMDGPU::sub2:
+      return AluT_Z;
+    case AMDGPU::sub3:
+      return AluT_W;
+    default:
+      break;
+    }
+
+    // Is the result already member of a X/Y/Z/W class ?
+    unsigned DestReg = MI->getOperand(0).getReg();
+    if (regBelongsToClass(DestReg, &AMDGPU::R600_TReg32_XRegClass) ||
+        regBelongsToClass(DestReg, &AMDGPU::R600_AddrRegClass))
+      return AluT_X;
+    if (regBelongsToClass(DestReg, &AMDGPU::R600_TReg32_YRegClass))
+      return AluT_Y;
+    if (regBelongsToClass(DestReg, &AMDGPU::R600_TReg32_ZRegClass))
+      return AluT_Z;
+    if (regBelongsToClass(DestReg, &AMDGPU::R600_TReg32_WRegClass))
+      return AluT_W;
+    if (regBelongsToClass(DestReg, &AMDGPU::R600_Reg128RegClass))
+      return AluT_XYZW;
+
+    return AluAny;
+
+}
+
+int R600SchedStrategy::getInstKind(SUnit* SU) {
+  int Opcode = SU->getInstr()->getOpcode();
+
+  if (TII->isALUInstr(Opcode)) {
+    return IDAlu;
+  }
+
+  switch (Opcode) {
+  case AMDGPU::COPY:
+  case AMDGPU::CONST_COPY:
+  case AMDGPU::INTERP_PAIR_XY:
+  case AMDGPU::INTERP_PAIR_ZW:
+  case AMDGPU::INTERP_VEC_LOAD:
+  case AMDGPU::DOT4_eg_pseudo:
+  case AMDGPU::DOT4_r600_pseudo:
+    return IDAlu;
+  case AMDGPU::TEX_VTX_CONSTBUF:
+  case AMDGPU::TEX_VTX_TEXBUF:
+  case AMDGPU::TEX_LD:
+  case AMDGPU::TEX_GET_TEXTURE_RESINFO:
+  case AMDGPU::TEX_GET_GRADIENTS_H:
+  case AMDGPU::TEX_GET_GRADIENTS_V:
+  case AMDGPU::TEX_SET_GRADIENTS_H:
+  case AMDGPU::TEX_SET_GRADIENTS_V:
+  case AMDGPU::TEX_SAMPLE:
+  case AMDGPU::TEX_SAMPLE_C:
+  case AMDGPU::TEX_SAMPLE_L:
+  case AMDGPU::TEX_SAMPLE_C_L:
+  case AMDGPU::TEX_SAMPLE_LB:
+  case AMDGPU::TEX_SAMPLE_C_LB:
+  case AMDGPU::TEX_SAMPLE_G:
+  case AMDGPU::TEX_SAMPLE_C_G:
+  case AMDGPU::TXD:
+  case AMDGPU::TXD_SHADOW:
+    return IDFetch;
+  default:
+    DEBUG(
+        dbgs() << "other inst: ";
+        SU->dump(DAG);
+    );
+    return IDOther;
+  }
+}
+
+SUnit *R600SchedStrategy::PopInst(std::multiset<SUnit *, CompareSUnit> &Q) {
+  if (Q.empty())
+    return NULL;
+  for (std::set<SUnit *, CompareSUnit>::iterator It = Q.begin(), E = Q.end();
+      It != E; ++It) {
+    SUnit *SU = *It;
+    InstructionsGroupCandidate.push_back(SU->getInstr());
+    if (TII->canBundle(InstructionsGroupCandidate)) {
+      InstructionsGroupCandidate.pop_back();
+      Q.erase(It);
+      return SU;
+    } else {
+      InstructionsGroupCandidate.pop_back();
+    }
+  }
+  return NULL;
+}
+
+void R600SchedStrategy::LoadAlu() {
+  ReadyQueue *QSrc = Pending[IDAlu];
+  for (ReadyQueue::iterator I = QSrc->begin(),
+        E = QSrc->end(); I != E; ++I) {
+      (*I)->NodeQueueId &= ~QSrc->getID();
+      AluKind AK = getAluKind(*I);
+      AvailableAlus[AK].insert(*I);
+    }
+    QSrc->clear();
+}
+
+void R600SchedStrategy::PrepareNextSlot() {
+  DEBUG(dbgs() << "New Slot\n");
+  assert (OccupedSlotsMask && "Slot wasn't filled");
+  OccupedSlotsMask = 0;
+  InstructionsGroupCandidate.clear();
+  LoadAlu();
+}
+
+void R600SchedStrategy::AssignSlot(MachineInstr* MI, unsigned Slot) {
+  unsigned DestReg = MI->getOperand(0).getReg();
+  // PressureRegister crashes if an operand is def and used in the same inst
+  // and we try to constraint its regclass
+  for (MachineInstr::mop_iterator It = MI->operands_begin(),
+      E = MI->operands_end(); It != E; ++It) {
+    MachineOperand &MO = *It;
+    if (MO.isReg() && !MO.isDef() &&
+        MO.getReg() == MI->getOperand(0).getReg())
+      return;
+  }
+  // Constrains the regclass of DestReg to assign it to Slot
+  switch (Slot) {
+  case 0:
+    MRI->constrainRegClass(DestReg, &AMDGPU::R600_TReg32_XRegClass);
+    break;
+  case 1:
+    MRI->constrainRegClass(DestReg, &AMDGPU::R600_TReg32_YRegClass);
+    break;
+  case 2:
+    MRI->constrainRegClass(DestReg, &AMDGPU::R600_TReg32_ZRegClass);
+    break;
+  case 3:
+    MRI->constrainRegClass(DestReg, &AMDGPU::R600_TReg32_WRegClass);
+    break;
+  }
+}
+
+SUnit *R600SchedStrategy::AttemptFillSlot(unsigned Slot) {
+  static const AluKind IndexToID[] = {AluT_X, AluT_Y, AluT_Z, AluT_W};
+  SUnit *SlotedSU = PopInst(AvailableAlus[IndexToID[Slot]]);
+  SUnit *UnslotedSU = PopInst(AvailableAlus[AluAny]);
+  if (!UnslotedSU) {
+    return SlotedSU;
+  } else if (!SlotedSU) {
+    AssignSlot(UnslotedSU->getInstr(), Slot);
+    return UnslotedSU;
+  } else {
+    //Determine which one to pick (the lesser one)
+    if (CompareSUnit()(SlotedSU, UnslotedSU)) {
+      AvailableAlus[AluAny].insert(UnslotedSU);
+      return SlotedSU;
+    } else {
+      AvailableAlus[IndexToID[Slot]].insert(SlotedSU);
+      AssignSlot(UnslotedSU->getInstr(), Slot);
+      return UnslotedSU;
+    }
+  }
+}
+
+bool R600SchedStrategy::isAvailablesAluEmpty() const {
+  return Pending[IDAlu]->empty() && AvailableAlus[AluAny].empty() &&
+      AvailableAlus[AluT_XYZW].empty() && AvailableAlus[AluT_X].empty() &&
+      AvailableAlus[AluT_Y].empty() && AvailableAlus[AluT_Z].empty() &&
+      AvailableAlus[AluT_W].empty() && AvailableAlus[AluDiscarded].empty();
+}
+
+SUnit* R600SchedStrategy::pickAlu() {
+  while (!isAvailablesAluEmpty()) {
+    if (!OccupedSlotsMask) {
+      // Flush physical reg copies (RA will discard them)
+      if (!AvailableAlus[AluDiscarded].empty()) {
+        OccupedSlotsMask = 15;
+        return PopInst(AvailableAlus[AluDiscarded]);
+      }
+      // If there is a T_XYZW alu available, use it
+      if (!AvailableAlus[AluT_XYZW].empty()) {
+        OccupedSlotsMask = 15;
+        return PopInst(AvailableAlus[AluT_XYZW]);
+      }
+    }
+    for (unsigned Chan = 0; Chan < 4; ++Chan) {
+      bool isOccupied = OccupedSlotsMask & (1 << Chan);
+      if (!isOccupied) {
+        SUnit *SU = AttemptFillSlot(Chan);
+        if (SU) {
+          OccupedSlotsMask |= (1 << Chan);
+          InstructionsGroupCandidate.push_back(SU->getInstr());
+          return SU;
+        }
+      }
+    }
+    PrepareNextSlot();
+  }
+  return NULL;
+}
+
+SUnit* R600SchedStrategy::pickOther(int QID) {
+  SUnit *SU = 0;
+  ReadyQueue *AQ = Available[QID];
+
+  if (AQ->empty()) {
+    MoveUnits(Pending[QID], AQ);
+  }
+  if (!AQ->empty()) {
+    SU = *AQ->begin();
+    AQ->remove(AQ->begin());
+  }
+  return SU;
+}
+
diff --git a/lib/Target/R600/R600MachineScheduler.h b/lib/Target/R600/R600MachineScheduler.h
new file mode 100644
index 000000000000..3d0367fd8ebf
--- /dev/null
+++ b/lib/Target/R600/R600MachineScheduler.h
@@ -0,0 +1,120 @@
+//===-- R600MachineScheduler.h - R600 Scheduler Interface -*- C++ -*-------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief R600 Machine Scheduler interface
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef R600MACHINESCHEDULER_H_
+#define R600MACHINESCHEDULER_H_
+
+#include "R600InstrInfo.h"
+#include "llvm/CodeGen/MachineScheduler.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/ADT/PriorityQueue.h"
+
+using namespace llvm;
+
+namespace llvm {
+
+class CompareSUnit {
+public:
+  bool operator()(const SUnit *S1, const SUnit *S2) {
+    return S1->getDepth() > S2->getDepth();
+  }
+};
+
+class R600SchedStrategy : public MachineSchedStrategy {
+
+  const ScheduleDAGMI *DAG;
+  const R600InstrInfo *TII;
+  const R600RegisterInfo *TRI;
+  MachineRegisterInfo *MRI;
+
+  enum InstQueue {
+    QAlu = 1,
+    QFetch = 2,
+    QOther = 4
+  };
+
+  enum InstKind {
+    IDAlu,
+    IDFetch,
+    IDOther,
+    IDLast
+  };
+
+  enum AluKind {
+    AluAny,
+    AluT_X,
+    AluT_Y,
+    AluT_Z,
+    AluT_W,
+    AluT_XYZW,
+    AluDiscarded, // LLVM Instructions that are going to be eliminated
+    AluLast
+  };
+
+  ReadyQueue *Available[IDLast], *Pending[IDLast];
+  std::multiset<SUnit *, CompareSUnit> AvailableAlus[AluLast];
+
+  InstKind CurInstKind;
+  int CurEmitted;
+  InstKind NextInstKind;
+
+  int InstKindLimit[IDLast];
+
+  int OccupedSlotsMask;
+
+public:
+  R600SchedStrategy() :
+    DAG(0), TII(0), TRI(0), MRI(0) {
+    Available[IDAlu] = new ReadyQueue(QAlu, "AAlu");
+    Available[IDFetch] = new ReadyQueue(QFetch, "AFetch");
+    Available[IDOther] = new ReadyQueue(QOther, "AOther");
+    Pending[IDAlu] = new ReadyQueue(QAlu<<4, "PAlu");
+    Pending[IDFetch] = new ReadyQueue(QFetch<<4, "PFetch");
+    Pending[IDOther] = new ReadyQueue(QOther<<4, "POther");
+  }
+
+  virtual ~R600SchedStrategy() {
+    for (unsigned I = 0; I < IDLast; ++I) {
+      delete Available[I];
+      delete Pending[I];
+    }
+  }
+
+  virtual void initialize(ScheduleDAGMI *dag);
+  virtual SUnit *pickNode(bool &IsTopNode);
+  virtual void schedNode(SUnit *SU, bool IsTopNode);
+  virtual void releaseTopNode(SUnit *SU);
+  virtual void releaseBottomNode(SUnit *SU);
+
+private:
+  std::vector<MachineInstr *> InstructionsGroupCandidate;
+
+  int getInstKind(SUnit *SU);
+  bool regBelongsToClass(unsigned Reg, const TargetRegisterClass *RC) const;
+  AluKind getAluKind(SUnit *SU) const;
+  void LoadAlu();
+  bool isAvailablesAluEmpty() const;
+  SUnit *AttemptFillSlot (unsigned Slot);
+  void PrepareNextSlot();
+  SUnit *PopInst(std::multiset<SUnit *, CompareSUnit> &Q);
+
+  void AssignSlot(MachineInstr *MI, unsigned Slot);
+  SUnit* pickAlu();
+  SUnit* pickOther(int QID);
+  void MoveUnits(ReadyQueue *QSrc, ReadyQueue *QDst);
+};
+
+} // namespace llvm
+
+#endif /* R600MACHINESCHEDULER_H_ */
diff --git a/lib/Target/R600/R600RegisterInfo.cpp b/lib/Target/R600/R600RegisterInfo.cpp
new file mode 100644
index 000000000000..bbd7995d7d51
--- /dev/null
+++ b/lib/Target/R600/R600RegisterInfo.cpp
@@ -0,0 +1,99 @@
+//===-- R600RegisterInfo.cpp - R600 Register Information ------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief R600 implementation of the TargetRegisterInfo class.
+//
+//===----------------------------------------------------------------------===//
+
+#include "R600RegisterInfo.h"
+#include "AMDGPUTargetMachine.h"
+#include "R600Defines.h"
+#include "R600InstrInfo.h"
+#include "R600MachineFunctionInfo.h"
+
+using namespace llvm;
+
+R600RegisterInfo::R600RegisterInfo(AMDGPUTargetMachine &tm,
+    const TargetInstrInfo &tii)
+: AMDGPURegisterInfo(tm, tii),
+  TM(tm),
+  TII(tii)
+  { }
+
+BitVector R600RegisterInfo::getReservedRegs(const MachineFunction &MF) const {
+  BitVector Reserved(getNumRegs());
+
+  Reserved.set(AMDGPU::ZERO);
+  Reserved.set(AMDGPU::HALF);
+  Reserved.set(AMDGPU::ONE);
+  Reserved.set(AMDGPU::ONE_INT);
+  Reserved.set(AMDGPU::NEG_HALF);
+  Reserved.set(AMDGPU::NEG_ONE);
+  Reserved.set(AMDGPU::PV_X);
+  Reserved.set(AMDGPU::ALU_LITERAL_X);
+  Reserved.set(AMDGPU::ALU_CONST);
+  Reserved.set(AMDGPU::PREDICATE_BIT);
+  Reserved.set(AMDGPU::PRED_SEL_OFF);
+  Reserved.set(AMDGPU::PRED_SEL_ZERO);
+  Reserved.set(AMDGPU::PRED_SEL_ONE);
+
+  for (TargetRegisterClass::iterator I = AMDGPU::R600_AddrRegClass.begin(),
+                        E = AMDGPU::R600_AddrRegClass.end(); I != E; ++I) {
+    Reserved.set(*I);
+  }
+
+  for (TargetRegisterClass::iterator I = AMDGPU::TRegMemRegClass.begin(),
+                                     E = AMDGPU::TRegMemRegClass.end();
+                                     I !=  E; ++I) {
+    Reserved.set(*I);
+  }
+
+  const R600InstrInfo *RII = static_cast<const R600InstrInfo*>(&TII);
+  std::vector<unsigned> IndirectRegs = RII->getIndirectReservedRegs(MF);
+  for (std::vector<unsigned>::iterator I = IndirectRegs.begin(),
+                                       E = IndirectRegs.end();
+                                       I != E; ++I) {
+    Reserved.set(*I);
+  }
+  return Reserved;
+}
+
+const TargetRegisterClass *
+R600RegisterInfo::getISARegClass(const TargetRegisterClass * rc) const {
+  switch (rc->getID()) {
+  case AMDGPU::GPRF32RegClassID:
+  case AMDGPU::GPRI32RegClassID:
+    return &AMDGPU::R600_Reg32RegClass;
+  default: return rc;
+  }
+}
+
+unsigned R600RegisterInfo::getHWRegChan(unsigned reg) const {
+  return this->getEncodingValue(reg) >> HW_CHAN_SHIFT;
+}
+
+const TargetRegisterClass * R600RegisterInfo::getCFGStructurizerRegClass(
+                                                                   MVT VT) const {
+  switch(VT.SimpleTy) {
+  default:
+  case MVT::i32: return &AMDGPU::R600_TReg32RegClass;
+  }
+}
+
+unsigned R600RegisterInfo::getSubRegFromChannel(unsigned Channel) const {
+  switch (Channel) {
+    default: assert(!"Invalid channel index"); return 0;
+    case 0: return AMDGPU::sub0;
+    case 1: return AMDGPU::sub1;
+    case 2: return AMDGPU::sub2;
+    case 3: return AMDGPU::sub3;
+  }
+}
+
diff --git a/lib/Target/R600/R600RegisterInfo.h b/lib/Target/R600/R600RegisterInfo.h
new file mode 100644
index 000000000000..f9ca918f246b
--- /dev/null
+++ b/lib/Target/R600/R600RegisterInfo.h
@@ -0,0 +1,55 @@
+//===-- R600RegisterInfo.h - R600 Register Info Interface ------*- C++ -*--===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Interface definition for R600RegisterInfo
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef R600REGISTERINFO_H_
+#define R600REGISTERINFO_H_
+
+#include "AMDGPURegisterInfo.h"
+#include "AMDGPUTargetMachine.h"
+
+namespace llvm {
+
+class R600TargetMachine;
+class TargetInstrInfo;
+
+struct R600RegisterInfo : public AMDGPURegisterInfo {
+  AMDGPUTargetMachine &TM;
+  const TargetInstrInfo &TII;
+
+  R600RegisterInfo(AMDGPUTargetMachine &tm, const TargetInstrInfo &tii);
+
+  virtual BitVector getReservedRegs(const MachineFunction &MF) const;
+
+  /// \param RC is an AMDIL reg class.
+  ///
+  /// \returns the R600 reg class that is equivalent to \p RC.
+  virtual const TargetRegisterClass *getISARegClass(
+    const TargetRegisterClass *RC) const;
+
+  /// \brief get the HW encoding for a register's channel.
+  unsigned getHWRegChan(unsigned reg) const;
+
+  /// \brief get the register class of the specified type to use in the
+  /// CFGStructurizer
+  virtual const TargetRegisterClass * getCFGStructurizerRegClass(MVT VT) const;
+
+  /// \returns the sub reg enum value for the given \p Channel
+  /// (e.g. getSubRegFromChannel(0) -> AMDGPU::sel_x)
+  unsigned getSubRegFromChannel(unsigned Channel) const;
+
+};
+
+} // End namespace llvm
+
+#endif // AMDIDSAREGISTERINFO_H_
diff --git a/lib/Target/R600/R600RegisterInfo.td b/lib/Target/R600/R600RegisterInfo.td
new file mode 100644
index 000000000000..03f49761ea40
--- /dev/null
+++ b/lib/Target/R600/R600RegisterInfo.td
@@ -0,0 +1,209 @@
+
+class R600Reg <string name, bits<16> encoding> : Register<name> {
+  let Namespace = "AMDGPU";
+  let HWEncoding = encoding;
+}
+
+class R600RegWithChan <string name, bits<9> sel, string chan> :
+    Register <name> {
+
+  field bits<2> chan_encoding = !if(!eq(chan, "X"), 0,
+                                !if(!eq(chan, "Y"), 1,
+                                !if(!eq(chan, "Z"), 2,
+                                !if(!eq(chan, "W"), 3, 0))));
+  let HWEncoding{8-0}  = sel;
+  let HWEncoding{10-9} = chan_encoding;
+  let Namespace = "AMDGPU";
+}
+
+class R600Reg_128<string n, list<Register> subregs, bits<16> encoding> :
+    RegisterWithSubRegs<n, subregs> {
+  let Namespace = "AMDGPU";
+  let SubRegIndices = [sub0, sub1, sub2, sub3];
+  let HWEncoding = encoding;
+}
+
+foreach Index = 0-127 in {
+  foreach Chan = [ "X", "Y", "Z", "W" ] in {
+    // 32-bit Temporary Registers
+    def T#Index#_#Chan : R600RegWithChan <"T"#Index#"."#Chan, Index, Chan>;
+
+    // Indirect addressing offset registers
+    def Addr#Index#_#Chan : R600RegWithChan <"T("#Index#" + AR.x)."#Chan,
+                                              Index, Chan>;
+    def TRegMem#Index#_#Chan : R600RegWithChan <"T"#Index#"."#Chan, Index,
+                                                Chan>;
+  }
+  // 128-bit Temporary Registers
+  def T#Index#_XYZW : R600Reg_128 <"T"#Index#".XYZW",
+                                   [!cast<Register>("T"#Index#"_X"),
+                                    !cast<Register>("T"#Index#"_Y"),
+                                    !cast<Register>("T"#Index#"_Z"),
+                                    !cast<Register>("T"#Index#"_W")],
+                                   Index>;
+}
+
+// KCACHE_BANK0
+foreach Index = 159-128 in {
+  foreach Chan = [ "X", "Y", "Z", "W" ] in {
+    // 32-bit Temporary Registers
+    def KC0_#Index#_#Chan : R600RegWithChan <"KC0["#Index#"-128]."#Chan, Index, Chan>;
+  }
+  // 128-bit Temporary Registers
+  def KC0_#Index#_XYZW : R600Reg_128 <"KC0["#Index#"-128].XYZW",
+                                 [!cast<Register>("KC0_"#Index#"_X"),
+                                  !cast<Register>("KC0_"#Index#"_Y"),
+                                  !cast<Register>("KC0_"#Index#"_Z"),
+                                  !cast<Register>("KC0_"#Index#"_W")],
+                                 Index>;
+}
+
+// KCACHE_BANK1
+foreach Index = 191-160 in {
+  foreach Chan = [ "X", "Y", "Z", "W" ] in {
+    // 32-bit Temporary Registers
+    def KC1_#Index#_#Chan : R600RegWithChan <"KC1["#Index#"-160]."#Chan, Index, Chan>;
+  }
+  // 128-bit Temporary Registers
+  def KC1_#Index#_XYZW : R600Reg_128 <"KC1["#Index#"-160].XYZW",
+                                 [!cast<Register>("KC1_"#Index#"_X"),
+                                  !cast<Register>("KC1_"#Index#"_Y"),
+                                  !cast<Register>("KC1_"#Index#"_Z"),
+                                  !cast<Register>("KC1_"#Index#"_W")],
+                                 Index>;
+}
+
+
+// Array Base Register holding input in FS
+foreach Index = 448-480 in {
+  def ArrayBase#Index :  R600Reg<"ARRAY_BASE", Index>;
+}
+
+
+// Special Registers
+
+def ZERO : R600Reg<"0.0", 248>;
+def ONE : R600Reg<"1.0", 249>;
+def NEG_ONE : R600Reg<"-1.0", 249>;
+def ONE_INT : R600Reg<"1", 250>;
+def HALF : R600Reg<"0.5", 252>;
+def NEG_HALF : R600Reg<"-0.5", 252>;
+def ALU_LITERAL_X : R600Reg<"literal.x", 253>;
+def PV_X : R600Reg<"pv.x", 254>;
+def PREDICATE_BIT : R600Reg<"PredicateBit", 0>;
+def PRED_SEL_OFF: R600Reg<"Pred_sel_off", 0>;
+def PRED_SEL_ZERO : R600Reg<"Pred_sel_zero", 2>;
+def PRED_SEL_ONE : R600Reg<"Pred_sel_one", 3>;
+def AR_X : R600Reg<"AR.x", 0>;
+
+def R600_ArrayBase : RegisterClass <"AMDGPU", [f32, i32], 32,
+                          (add (sequence "ArrayBase%u", 448, 480))>;
+// special registers for ALU src operands
+// const buffer reference, SRCx_SEL contains index
+def ALU_CONST : R600Reg<"CBuf", 0>;
+// interpolation param reference, SRCx_SEL contains index
+def ALU_PARAM : R600Reg<"Param", 0>;
+
+let isAllocatable = 0 in {
+
+// XXX: Only use the X channel, until we support wider stack widths
+def R600_Addr : RegisterClass <"AMDGPU", [i32], 127, (add (sequence "Addr%u_X", 0, 127))>;
+
+} // End isAllocatable = 0
+
+def R600_KC0_X : RegisterClass <"AMDGPU", [f32, i32], 32,
+                              (add (sequence "KC0_%u_X", 128, 159))>;
+
+def R600_KC0_Y : RegisterClass <"AMDGPU", [f32, i32], 32,
+                              (add (sequence "KC0_%u_Y", 128, 159))>;
+
+def R600_KC0_Z : RegisterClass <"AMDGPU", [f32, i32], 32,
+                              (add (sequence "KC0_%u_Z", 128, 159))>;
+
+def R600_KC0_W : RegisterClass <"AMDGPU", [f32, i32], 32,
+                              (add (sequence "KC0_%u_W", 128, 159))>;
+
+def R600_KC0 : RegisterClass <"AMDGPU", [f32, i32], 32,
+                                   (interleave R600_KC0_X, R600_KC0_Y,
+                                               R600_KC0_Z, R600_KC0_W)>;
+
+def R600_KC1_X : RegisterClass <"AMDGPU", [f32, i32], 32,
+                              (add (sequence "KC1_%u_X", 160, 191))>;
+
+def R600_KC1_Y : RegisterClass <"AMDGPU", [f32, i32], 32,
+                              (add (sequence "KC1_%u_Y", 160, 191))>;
+
+def R600_KC1_Z : RegisterClass <"AMDGPU", [f32, i32], 32,
+                              (add (sequence "KC1_%u_Z", 160, 191))>;
+
+def R600_KC1_W : RegisterClass <"AMDGPU", [f32, i32], 32,
+                              (add (sequence "KC1_%u_W", 160, 191))>;
+
+def R600_KC1 : RegisterClass <"AMDGPU", [f32, i32], 32,
+                                   (interleave R600_KC1_X, R600_KC1_Y,
+                                               R600_KC1_Z, R600_KC1_W)>;
+
+def R600_TReg32_X : RegisterClass <"AMDGPU", [f32, i32], 32,
+                                   (add (sequence "T%u_X", 0, 127), AR_X)>;
+
+def R600_TReg32_Y : RegisterClass <"AMDGPU", [f32, i32], 32,
+                                   (add (sequence "T%u_Y", 0, 127))>;
+
+def R600_TReg32_Z : RegisterClass <"AMDGPU", [f32, i32], 32,
+                                   (add (sequence "T%u_Z", 0, 127))>;
+
+def R600_TReg32_W : RegisterClass <"AMDGPU", [f32, i32], 32,
+                                   (add (sequence "T%u_W", 0, 127))>;
+
+def R600_TReg32 : RegisterClass <"AMDGPU", [f32, i32], 32,
+                                   (interleave R600_TReg32_X, R600_TReg32_Y,
+                                               R600_TReg32_Z, R600_TReg32_W)>;
+
+def R600_Reg32 : RegisterClass <"AMDGPU", [f32, i32], 32, (add
+    R600_TReg32,
+    R600_ArrayBase,
+    R600_Addr,
+    ZERO, HALF, ONE, ONE_INT, PV_X, ALU_LITERAL_X, NEG_ONE, NEG_HALF,
+    ALU_CONST, ALU_PARAM
+    )>;
+
+def R600_Predicate : RegisterClass <"AMDGPU", [i32], 32, (add
+    PRED_SEL_OFF, PRED_SEL_ZERO, PRED_SEL_ONE)>;
+
+def R600_Predicate_Bit: RegisterClass <"AMDGPU", [i32], 32, (add
+    PREDICATE_BIT)>;
+
+def R600_Reg128 : RegisterClass<"AMDGPU", [v4f32, v4i32], 128,
+                                (add (sequence "T%u_XYZW", 0, 127))> {
+  let CopyCost = -1;
+}
+
+//===----------------------------------------------------------------------===//
+// Register classes for indirect addressing
+//===----------------------------------------------------------------------===//
+
+// Super register for all the Indirect Registers.  This register class is used
+// by the REG_SEQUENCE instruction to specify the registers to use for direct
+// reads / writes which may be written / read by an indirect address.
+class IndirectSuper<string n, list<Register> subregs> :
+    RegisterWithSubRegs<n, subregs> {
+  let Namespace = "AMDGPU";
+  let SubRegIndices =
+ [sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7,
+  sub8, sub9, sub10, sub11, sub12, sub13, sub14, sub15];
+}
+
+def IndirectSuperReg : IndirectSuper<"Indirect",
+  [TRegMem0_X, TRegMem1_X, TRegMem2_X, TRegMem3_X, TRegMem4_X, TRegMem5_X,
+   TRegMem6_X, TRegMem7_X, TRegMem8_X, TRegMem9_X, TRegMem10_X, TRegMem11_X,
+   TRegMem12_X, TRegMem13_X, TRegMem14_X, TRegMem15_X]
+>;
+
+def IndirectReg : RegisterClass<"AMDGPU", [f32, i32], 32, (add IndirectSuperReg)>;
+
+// This register class defines the registers that are the storage units for
+// the "Indirect Addressing" pseudo memory space.
+// XXX: Only use the X channel, until we support wider stack widths
+def TRegMem : RegisterClass<"AMDGPU", [f32, i32], 32,
+  (add (sequence "TRegMem%u_X", 0, 16))
+>;
diff --git a/lib/Target/R600/R600Schedule.td b/lib/Target/R600/R600Schedule.td
new file mode 100644
index 000000000000..7ede181c51dc
--- /dev/null
+++ b/lib/Target/R600/R600Schedule.td
@@ -0,0 +1,36 @@
+//===-- R600Schedule.td - R600 Scheduling definitions ------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// R600 has a VLIW architecture.  On pre-cayman cards there are 5 instruction
+// slots ALU.X, ALU.Y, ALU.Z, ALU.W, and TRANS.  For cayman cards, the TRANS
+// slot has been removed. 
+//
+//===----------------------------------------------------------------------===//
+
+
+def ALU_X : FuncUnit;
+def ALU_Y : FuncUnit;
+def ALU_Z : FuncUnit;
+def ALU_W : FuncUnit;
+def TRANS : FuncUnit;
+
+def AnyALU : InstrItinClass;
+def VecALU : InstrItinClass;
+def TransALU : InstrItinClass;
+
+def R600_EG_Itin : ProcessorItineraries <
+  [ALU_X, ALU_Y, ALU_Z, ALU_W, TRANS, ALU_NULL],
+  [],
+  [
+    InstrItinData<AnyALU, [InstrStage<1, [ALU_X, ALU_Y, ALU_Z, ALU_W, TRANS]>]>,
+    InstrItinData<VecALU, [InstrStage<1, [ALU_X, ALU_Y, ALU_X, ALU_W]>]>,
+    InstrItinData<TransALU, [InstrStage<1, [TRANS]>]>,
+    InstrItinData<NullALU, [InstrStage<1, [ALU_NULL]>]>
+  ]
+>;
diff --git a/lib/Target/R600/SIAnnotateControlFlow.cpp b/lib/Target/R600/SIAnnotateControlFlow.cpp
new file mode 100644
index 000000000000..2477e2a9dcc3
--- /dev/null
+++ b/lib/Target/R600/SIAnnotateControlFlow.cpp
@@ -0,0 +1,329 @@
+//===-- SIAnnotateControlFlow.cpp -  ------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// Annotates the control flow with hardware specific intrinsics.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "llvm/ADT/DepthFirstIterator.h"
+#include "llvm/Analysis/Dominators.h"
+#include "llvm/IR/Module.h"
+#include "llvm/Pass.h"
+#include "llvm/Transforms/Utils/BasicBlockUtils.h"
+#include "llvm/Transforms/Utils/SSAUpdater.h"
+
+using namespace llvm;
+
+namespace {
+
+// Complex types used in this pass
+typedef std::pair<BasicBlock *, Value *> StackEntry;
+typedef SmallVector<StackEntry, 16> StackVector;
+
+// Intrinsic names the control flow is annotated with
+static const char *IfIntrinsic = "llvm.SI.if";
+static const char *ElseIntrinsic = "llvm.SI.else";
+static const char *BreakIntrinsic = "llvm.SI.break";
+static const char *IfBreakIntrinsic = "llvm.SI.if.break";
+static const char *ElseBreakIntrinsic = "llvm.SI.else.break";
+static const char *LoopIntrinsic = "llvm.SI.loop";
+static const char *EndCfIntrinsic = "llvm.SI.end.cf";
+
+class SIAnnotateControlFlow : public FunctionPass {
+
+  static char ID;
+
+  Type *Boolean;
+  Type *Void;
+  Type *Int64;
+  Type *ReturnStruct;
+
+  ConstantInt *BoolTrue;
+  ConstantInt *BoolFalse;
+  UndefValue *BoolUndef;
+  Constant *Int64Zero;
+
+  Constant *If;
+  Constant *Else;
+  Constant *Break;
+  Constant *IfBreak;
+  Constant *ElseBreak;
+  Constant *Loop;
+  Constant *EndCf;
+
+  DominatorTree *DT;
+  StackVector Stack;
+  SSAUpdater PhiInserter;
+
+  bool isTopOfStack(BasicBlock *BB);
+
+  Value *popSaved();
+
+  void push(BasicBlock *BB, Value *Saved);
+
+  bool isElse(PHINode *Phi);
+
+  void eraseIfUnused(PHINode *Phi);
+
+  void openIf(BranchInst *Term);
+
+  void insertElse(BranchInst *Term);
+
+  void handleLoopCondition(Value *Cond);
+
+  void handleLoop(BranchInst *Term);
+
+  void closeControlFlow(BasicBlock *BB);
+
+public:
+  SIAnnotateControlFlow():
+    FunctionPass(ID) { }
+
+  virtual bool doInitialization(Module &M);
+
+  virtual bool runOnFunction(Function &F);
+
+  virtual const char *getPassName() const {
+    return "SI annotate control flow";
+  }
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+    AU.addRequired<DominatorTree>();
+    AU.addPreserved<DominatorTree>();
+    FunctionPass::getAnalysisUsage(AU);
+  }
+
+};
+
+} // end anonymous namespace
+
+char SIAnnotateControlFlow::ID = 0;
+
+/// \brief Initialize all the types and constants used in the pass
+bool SIAnnotateControlFlow::doInitialization(Module &M) {
+  LLVMContext &Context = M.getContext();
+
+  Void = Type::getVoidTy(Context);
+  Boolean = Type::getInt1Ty(Context);
+  Int64 = Type::getInt64Ty(Context);
+  ReturnStruct = StructType::get(Boolean, Int64, (Type *)0);
+
+  BoolTrue = ConstantInt::getTrue(Context);
+  BoolFalse = ConstantInt::getFalse(Context);
+  BoolUndef = UndefValue::get(Boolean);
+  Int64Zero = ConstantInt::get(Int64, 0);
+
+  If = M.getOrInsertFunction(
+    IfIntrinsic, ReturnStruct, Boolean, (Type *)0);
+
+  Else = M.getOrInsertFunction(
+    ElseIntrinsic, ReturnStruct, Int64, (Type *)0);
+
+  Break = M.getOrInsertFunction(
+    BreakIntrinsic, Int64, Int64, (Type *)0);
+
+  IfBreak = M.getOrInsertFunction(
+    IfBreakIntrinsic, Int64, Boolean, Int64, (Type *)0);
+
+  ElseBreak = M.getOrInsertFunction(
+    ElseBreakIntrinsic, Int64, Int64, Int64, (Type *)0);
+
+  Loop = M.getOrInsertFunction(
+    LoopIntrinsic, Boolean, Int64, (Type *)0);
+
+  EndCf = M.getOrInsertFunction(
+    EndCfIntrinsic, Void, Int64, (Type *)0);
+
+  return false;
+}
+
+/// \brief Is BB the last block saved on the stack ?
+bool SIAnnotateControlFlow::isTopOfStack(BasicBlock *BB) {
+  return !Stack.empty() && Stack.back().first == BB;
+}
+
+/// \brief Pop the last saved value from the control flow stack
+Value *SIAnnotateControlFlow::popSaved() {
+  return Stack.pop_back_val().second;
+}
+
+/// \brief Push a BB and saved value to the control flow stack
+void SIAnnotateControlFlow::push(BasicBlock *BB, Value *Saved) {
+  Stack.push_back(std::make_pair(BB, Saved));
+}
+
+/// \brief Can the condition represented by this PHI node treated like
+/// an "Else" block?
+bool SIAnnotateControlFlow::isElse(PHINode *Phi) {
+  BasicBlock *IDom = DT->getNode(Phi->getParent())->getIDom()->getBlock();
+  for (unsigned i = 0, e = Phi->getNumIncomingValues(); i != e; ++i) {
+    if (Phi->getIncomingBlock(i) == IDom) {
+
+      if (Phi->getIncomingValue(i) != BoolTrue)
+        return false;
+
+    } else {
+      if (Phi->getIncomingValue(i) != BoolFalse)
+        return false;
+ 
+    }
+  }
+  return true;
+}
+
+// \brief Erase "Phi" if it is not used any more
+void SIAnnotateControlFlow::eraseIfUnused(PHINode *Phi) {
+  if (!Phi->hasNUsesOrMore(1))
+    Phi->eraseFromParent();
+}
+
+/// \brief Open a new "If" block
+void SIAnnotateControlFlow::openIf(BranchInst *Term) {
+  Value *Ret = CallInst::Create(If, Term->getCondition(), "", Term);
+  Term->setCondition(ExtractValueInst::Create(Ret, 0, "", Term));
+  push(Term->getSuccessor(1), ExtractValueInst::Create(Ret, 1, "", Term));
+}
+
+/// \brief Close the last "If" block and open a new "Else" block
+void SIAnnotateControlFlow::insertElse(BranchInst *Term) {
+  Value *Ret = CallInst::Create(Else, popSaved(), "", Term);
+  Term->setCondition(ExtractValueInst::Create(Ret, 0, "", Term));
+  push(Term->getSuccessor(1), ExtractValueInst::Create(Ret, 1, "", Term));
+}
+
+/// \brief Recursively handle the condition leading to a loop
+void SIAnnotateControlFlow::handleLoopCondition(Value *Cond) {
+  if (PHINode *Phi = dyn_cast<PHINode>(Cond)) {
+
+    // Handle all non constant incoming values first
+    for (unsigned i = 0, e = Phi->getNumIncomingValues(); i != e; ++i) {
+      Value *Incoming = Phi->getIncomingValue(i);
+      if (isa<ConstantInt>(Incoming))
+        continue;
+
+      Phi->setIncomingValue(i, BoolFalse);
+      handleLoopCondition(Incoming);
+    }
+
+    BasicBlock *Parent = Phi->getParent();
+    BasicBlock *IDom = DT->getNode(Parent)->getIDom()->getBlock();
+
+    for (unsigned i = 0, e = Phi->getNumIncomingValues(); i != e; ++i) {
+
+      Value *Incoming = Phi->getIncomingValue(i);
+      if (Incoming != BoolTrue)
+        continue;
+
+      BasicBlock *From = Phi->getIncomingBlock(i);
+      if (From == IDom) {
+        CallInst *OldEnd = dyn_cast<CallInst>(Parent->getFirstInsertionPt());
+        if (OldEnd && OldEnd->getCalledFunction() == EndCf) {
+          Value *Args[] = {
+            OldEnd->getArgOperand(0),
+            PhiInserter.GetValueAtEndOfBlock(Parent)
+          };
+          Value *Ret = CallInst::Create(ElseBreak, Args, "", OldEnd);
+          PhiInserter.AddAvailableValue(Parent, Ret);
+          continue;
+        }
+      }
+
+      TerminatorInst *Insert = From->getTerminator();
+      Value *Arg = PhiInserter.GetValueAtEndOfBlock(From);
+      Value *Ret = CallInst::Create(Break, Arg, "", Insert);
+      PhiInserter.AddAvailableValue(From, Ret);
+    }
+    eraseIfUnused(Phi);
+
+  } else if (Instruction *Inst = dyn_cast<Instruction>(Cond)) {
+    BasicBlock *Parent = Inst->getParent();
+    TerminatorInst *Insert = Parent->getTerminator();
+    Value *Args[] = { Cond, PhiInserter.GetValueAtEndOfBlock(Parent) };
+    Value *Ret = CallInst::Create(IfBreak, Args, "", Insert);
+    PhiInserter.AddAvailableValue(Parent, Ret);
+
+  } else {
+    assert(0 && "Unhandled loop condition!");
+  }
+}
+
+/// \brief Handle a back edge (loop)
+void SIAnnotateControlFlow::handleLoop(BranchInst *Term) {
+  BasicBlock *Target = Term->getSuccessor(1);
+  PHINode *Broken = PHINode::Create(Int64, 0, "", &Target->front());
+
+  PhiInserter.Initialize(Int64, "");
+  PhiInserter.AddAvailableValue(Target, Broken);
+
+  Value *Cond = Term->getCondition();
+  Term->setCondition(BoolTrue);
+  handleLoopCondition(Cond);
+
+  BasicBlock *BB = Term->getParent();
+  Value *Arg = PhiInserter.GetValueAtEndOfBlock(BB);
+  for (pred_iterator PI = pred_begin(Target), PE = pred_end(Target);
+       PI != PE; ++PI) {
+
+    Broken->addIncoming(*PI == BB ? Arg : Int64Zero, *PI);
+  }
+
+  Term->setCondition(CallInst::Create(Loop, Arg, "", Term));
+  push(Term->getSuccessor(0), Arg);
+}
+
+/// \brief Close the last opened control flow
+void SIAnnotateControlFlow::closeControlFlow(BasicBlock *BB) {
+  CallInst::Create(EndCf, popSaved(), "", BB->getFirstInsertionPt());
+}
+
+/// \brief Annotate the control flow with intrinsics so the backend can
+/// recognize if/then/else and loops.
+bool SIAnnotateControlFlow::runOnFunction(Function &F) {
+  DT = &getAnalysis<DominatorTree>();
+
+  for (df_iterator<BasicBlock *> I = df_begin(&F.getEntryBlock()),
+       E = df_end(&F.getEntryBlock()); I != E; ++I) {
+
+    BranchInst *Term = dyn_cast<BranchInst>((*I)->getTerminator());
+
+    if (!Term || Term->isUnconditional()) {
+      if (isTopOfStack(*I))
+        closeControlFlow(*I);
+      continue;
+    }
+
+    if (I.nodeVisited(Term->getSuccessor(1))) {
+      if (isTopOfStack(*I))
+        closeControlFlow(*I);
+      handleLoop(Term);
+      continue;
+    }
+
+    if (isTopOfStack(*I)) {
+      PHINode *Phi = dyn_cast<PHINode>(Term->getCondition());
+      if (Phi && Phi->getParent() == *I && isElse(Phi)) {
+        insertElse(Term);
+        eraseIfUnused(Phi);
+        continue;
+      }
+      closeControlFlow(*I);
+    }
+    openIf(Term);
+  }
+
+  assert(Stack.empty());
+  return true;
+}
+
+/// \brief Create the annotation pass
+FunctionPass *llvm::createSIAnnotateControlFlowPass() {
+  return new SIAnnotateControlFlow();
+}
diff --git a/lib/Target/R600/SIISelLowering.cpp b/lib/Target/R600/SIISelLowering.cpp
new file mode 100644
index 000000000000..6f0c30761506
--- /dev/null
+++ b/lib/Target/R600/SIISelLowering.cpp
@@ -0,0 +1,670 @@
+//===-- SIISelLowering.cpp - SI DAG Lowering Implementation ---------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Custom DAG lowering for SI
+//
+//===----------------------------------------------------------------------===//
+
+#include "SIISelLowering.h"
+#include "AMDIL.h"
+#include "AMDGPU.h"
+#include "AMDILIntrinsicInfo.h"
+#include "SIInstrInfo.h"
+#include "SIMachineFunctionInfo.h"
+#include "SIRegisterInfo.h"
+#include "llvm/IR/Function.h"
+#include "llvm/CodeGen/CallingConvLower.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/SelectionDAG.h"
+
+using namespace llvm;
+
+SITargetLowering::SITargetLowering(TargetMachine &TM) :
+    AMDGPUTargetLowering(TM),
+    TII(static_cast<const SIInstrInfo*>(TM.getInstrInfo())),
+    TRI(TM.getRegisterInfo()) {
+
+  addRegisterClass(MVT::i1, &AMDGPU::SReg_64RegClass);
+  addRegisterClass(MVT::i64, &AMDGPU::SReg_64RegClass);
+
+  addRegisterClass(MVT::v16i8, &AMDGPU::SReg_128RegClass);
+  addRegisterClass(MVT::v32i8, &AMDGPU::SReg_256RegClass);
+  addRegisterClass(MVT::v64i8, &AMDGPU::SReg_512RegClass);
+
+  addRegisterClass(MVT::i32, &AMDGPU::VReg_32RegClass);
+  addRegisterClass(MVT::f32, &AMDGPU::VReg_32RegClass);
+
+  addRegisterClass(MVT::v1i32, &AMDGPU::VReg_32RegClass);
+
+  addRegisterClass(MVT::v2i32, &AMDGPU::VReg_64RegClass);
+  addRegisterClass(MVT::v2f32, &AMDGPU::VReg_64RegClass);
+
+  addRegisterClass(MVT::v4i32, &AMDGPU::VReg_128RegClass);
+  addRegisterClass(MVT::v4f32, &AMDGPU::VReg_128RegClass);
+
+  addRegisterClass(MVT::v8i32, &AMDGPU::VReg_256RegClass);
+  addRegisterClass(MVT::v8f32, &AMDGPU::VReg_256RegClass);
+
+  addRegisterClass(MVT::v16i32, &AMDGPU::VReg_512RegClass);
+  addRegisterClass(MVT::v16f32, &AMDGPU::VReg_512RegClass);
+
+  computeRegisterProperties();
+
+  setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v8i32, Expand);
+  setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v8f32, Expand);
+  setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v16i32, Expand);
+  setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v16f32, Expand);
+
+  setOperationAction(ISD::ADD, MVT::i64, Legal);
+  setOperationAction(ISD::ADD, MVT::i32, Legal);
+
+  setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
+  setOperationAction(ISD::SELECT_CC, MVT::i32, Custom);
+
+  setOperationAction(ISD::SELECT_CC, MVT::Other, Expand);
+  setTargetDAGCombine(ISD::SELECT_CC);
+
+  setTargetDAGCombine(ISD::SETCC);
+
+  setSchedulingPreference(Sched::RegPressure);
+}
+
+SDValue SITargetLowering::LowerFormalArguments(
+                                      SDValue Chain,
+                                      CallingConv::ID CallConv,
+                                      bool isVarArg,
+                                      const SmallVectorImpl<ISD::InputArg> &Ins,
+                                      DebugLoc DL, SelectionDAG &DAG,
+                                      SmallVectorImpl<SDValue> &InVals) const {
+
+  const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
+
+  MachineFunction &MF = DAG.getMachineFunction();
+  FunctionType *FType = MF.getFunction()->getFunctionType();
+  SIMachineFunctionInfo *Info = MF.getInfo<SIMachineFunctionInfo>();
+
+  assert(CallConv == CallingConv::C);
+
+  SmallVector<ISD::InputArg, 16> Splits;
+  uint32_t Skipped = 0;
+
+  for (unsigned i = 0, e = Ins.size(), PSInputNum = 0; i != e; ++i) {
+    const ISD::InputArg &Arg = Ins[i];
+   
+    // First check if it's a PS input addr 
+    if (Info->ShaderType == ShaderType::PIXEL && !Arg.Flags.isInReg()) {
+
+      assert((PSInputNum <= 15) && "Too many PS inputs!");
+
+      if (!Arg.Used) {
+        // We can savely skip PS inputs
+        Skipped |= 1 << i;
+        ++PSInputNum;
+        continue;
+      }
+
+      Info->PSInputAddr |= 1 << PSInputNum++;
+    }
+
+    // Second split vertices into their elements
+    if (Arg.VT.isVector()) {
+      ISD::InputArg NewArg = Arg;
+      NewArg.Flags.setSplit();
+      NewArg.VT = Arg.VT.getVectorElementType();
+
+      // We REALLY want the ORIGINAL number of vertex elements here, e.g. a
+      // three or five element vertex only needs three or five registers,
+      // NOT four or eigth.
+      Type *ParamType = FType->getParamType(Arg.OrigArgIndex);
+      unsigned NumElements = ParamType->getVectorNumElements();
+
+      for (unsigned j = 0; j != NumElements; ++j) {
+        Splits.push_back(NewArg);
+        NewArg.PartOffset += NewArg.VT.getStoreSize();
+      }
+
+    } else {
+      Splits.push_back(Arg);
+    }
+  }
+
+  SmallVector<CCValAssign, 16> ArgLocs;
+  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
+                 getTargetMachine(), ArgLocs, *DAG.getContext());
+
+  // At least one interpolation mode must be enabled or else the GPU will hang.
+  if (Info->ShaderType == ShaderType::PIXEL && (Info->PSInputAddr & 0x7F) == 0) {
+    Info->PSInputAddr |= 1;
+    CCInfo.AllocateReg(AMDGPU::VGPR0);
+    CCInfo.AllocateReg(AMDGPU::VGPR1);
+  }
+
+  AnalyzeFormalArguments(CCInfo, Splits);
+
+  for (unsigned i = 0, e = Ins.size(), ArgIdx = 0; i != e; ++i) {
+
+    if (Skipped & (1 << i)) {
+      InVals.push_back(SDValue());
+      continue;
+    }
+
+    CCValAssign &VA = ArgLocs[ArgIdx++];
+    assert(VA.isRegLoc() && "Parameter must be in a register!");
+
+    unsigned Reg = VA.getLocReg();
+    MVT VT = VA.getLocVT();
+
+    if (VT == MVT::i64) {
+      // For now assume it is a pointer
+      Reg = TRI->getMatchingSuperReg(Reg, AMDGPU::sub0,
+                                     &AMDGPU::SReg_64RegClass);
+      Reg = MF.addLiveIn(Reg, &AMDGPU::SReg_64RegClass);
+      InVals.push_back(DAG.getCopyFromReg(Chain, DL, Reg, VT));
+      continue;
+    }
+
+    const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg, VT);
+
+    Reg = MF.addLiveIn(Reg, RC);
+    SDValue Val = DAG.getCopyFromReg(Chain, DL, Reg, VT);
+
+    const ISD::InputArg &Arg = Ins[i];
+    if (Arg.VT.isVector()) {
+
+      // Build a vector from the registers
+      Type *ParamType = FType->getParamType(Arg.OrigArgIndex);
+      unsigned NumElements = ParamType->getVectorNumElements();
+
+      SmallVector<SDValue, 4> Regs;
+      Regs.push_back(Val);
+      for (unsigned j = 1; j != NumElements; ++j) {
+        Reg = ArgLocs[ArgIdx++].getLocReg();
+        Reg = MF.addLiveIn(Reg, RC);
+        Regs.push_back(DAG.getCopyFromReg(Chain, DL, Reg, VT));
+      }
+
+      // Fill up the missing vector elements
+      NumElements = Arg.VT.getVectorNumElements() - NumElements;
+      for (unsigned j = 0; j != NumElements; ++j)
+        Regs.push_back(DAG.getUNDEF(VT));
+ 
+      InVals.push_back(DAG.getNode(ISD::BUILD_VECTOR, DL, Arg.VT,
+                                   Regs.data(), Regs.size()));
+      continue;
+    }
+
+    InVals.push_back(Val);
+  }
+  return Chain;
+}
+
+MachineBasicBlock * SITargetLowering::EmitInstrWithCustomInserter(
+    MachineInstr * MI, MachineBasicBlock * BB) const {
+
+  switch (MI->getOpcode()) {
+  default:
+    return AMDGPUTargetLowering::EmitInstrWithCustomInserter(MI, BB);
+  case AMDGPU::BRANCH: return BB;
+  }
+  return BB;
+}
+
+EVT SITargetLowering::getSetCCResultType(EVT VT) const {
+  return MVT::i1;
+}
+
+MVT SITargetLowering::getScalarShiftAmountTy(EVT VT) const {
+  return MVT::i32;
+}
+
+//===----------------------------------------------------------------------===//
+// Custom DAG Lowering Operations
+//===----------------------------------------------------------------------===//
+
+SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
+  switch (Op.getOpcode()) {
+  default: return AMDGPUTargetLowering::LowerOperation(Op, DAG);
+  case ISD::BRCOND: return LowerBRCOND(Op, DAG);
+  case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
+  }
+  return SDValue();
+}
+
+/// \brief Helper function for LowerBRCOND
+static SDNode *findUser(SDValue Value, unsigned Opcode) {
+
+  SDNode *Parent = Value.getNode();
+  for (SDNode::use_iterator I = Parent->use_begin(), E = Parent->use_end();
+       I != E; ++I) {
+
+    if (I.getUse().get() != Value)
+      continue;
+
+    if (I->getOpcode() == Opcode)
+      return *I;
+  }
+  return 0;
+}
+
+/// This transforms the control flow intrinsics to get the branch destination as
+/// last parameter, also switches branch target with BR if the need arise
+SDValue SITargetLowering::LowerBRCOND(SDValue BRCOND,
+                                      SelectionDAG &DAG) const {
+
+  DebugLoc DL = BRCOND.getDebugLoc();
+
+  SDNode *Intr = BRCOND.getOperand(1).getNode();
+  SDValue Target = BRCOND.getOperand(2);
+  SDNode *BR = 0;
+
+  if (Intr->getOpcode() == ISD::SETCC) {
+    // As long as we negate the condition everything is fine
+    SDNode *SetCC = Intr;
+    assert(SetCC->getConstantOperandVal(1) == 1);
+    assert(cast<CondCodeSDNode>(SetCC->getOperand(2).getNode())->get() ==
+           ISD::SETNE);
+    Intr = SetCC->getOperand(0).getNode();
+
+  } else {
+    // Get the target from BR if we don't negate the condition
+    BR = findUser(BRCOND, ISD::BR);
+    Target = BR->getOperand(1);
+  }
+
+  assert(Intr->getOpcode() == ISD::INTRINSIC_W_CHAIN);
+
+  // Build the result and
+  SmallVector<EVT, 4> Res;
+  for (unsigned i = 1, e = Intr->getNumValues(); i != e; ++i)
+    Res.push_back(Intr->getValueType(i));
+
+  // operands of the new intrinsic call
+  SmallVector<SDValue, 4> Ops;
+  Ops.push_back(BRCOND.getOperand(0));
+  for (unsigned i = 1, e = Intr->getNumOperands(); i != e; ++i)
+    Ops.push_back(Intr->getOperand(i));
+  Ops.push_back(Target);
+
+  // build the new intrinsic call
+  SDNode *Result = DAG.getNode(
+    Res.size() > 1 ? ISD::INTRINSIC_W_CHAIN : ISD::INTRINSIC_VOID, DL,
+    DAG.getVTList(Res.data(), Res.size()), Ops.data(), Ops.size()).getNode();
+
+  if (BR) {
+    // Give the branch instruction our target
+    SDValue Ops[] = {
+      BR->getOperand(0),
+      BRCOND.getOperand(2)
+    };
+    DAG.MorphNodeTo(BR, ISD::BR, BR->getVTList(), Ops, 2);
+  }
+
+  SDValue Chain = SDValue(Result, Result->getNumValues() - 1);
+
+  // Copy the intrinsic results to registers
+  for (unsigned i = 1, e = Intr->getNumValues() - 1; i != e; ++i) {
+    SDNode *CopyToReg = findUser(SDValue(Intr, i), ISD::CopyToReg);
+    if (!CopyToReg)
+      continue;
+
+    Chain = DAG.getCopyToReg(
+      Chain, DL,
+      CopyToReg->getOperand(1),
+      SDValue(Result, i - 1),
+      SDValue());
+
+    DAG.ReplaceAllUsesWith(SDValue(CopyToReg, 0), CopyToReg->getOperand(0));
+  }
+
+  // Remove the old intrinsic from the chain
+  DAG.ReplaceAllUsesOfValueWith(
+    SDValue(Intr, Intr->getNumValues() - 1),
+    Intr->getOperand(0));
+
+  return Chain;
+}
+
+SDValue SITargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const {
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+  SDValue True = Op.getOperand(2);
+  SDValue False = Op.getOperand(3);
+  SDValue CC = Op.getOperand(4);
+  EVT VT = Op.getValueType();
+  DebugLoc DL = Op.getDebugLoc();
+
+  // Possible Min/Max pattern
+  SDValue MinMax = LowerMinMax(Op, DAG);
+  if (MinMax.getNode()) {
+    return MinMax;
+  }
+
+  SDValue Cond = DAG.getNode(ISD::SETCC, DL, MVT::i1, LHS, RHS, CC);
+  return DAG.getNode(ISD::SELECT, DL, VT, Cond, True, False);
+}
+
+//===----------------------------------------------------------------------===//
+// Custom DAG optimizations
+//===----------------------------------------------------------------------===//
+
+SDValue SITargetLowering::PerformDAGCombine(SDNode *N,
+                                            DAGCombinerInfo &DCI) const {
+  SelectionDAG &DAG = DCI.DAG;
+  DebugLoc DL = N->getDebugLoc();
+  EVT VT = N->getValueType(0);
+
+  switch (N->getOpcode()) {
+    default: break;
+    case ISD::SELECT_CC: {
+      N->dump();
+      ConstantSDNode *True, *False;
+      // i1 selectcc(l, r, -1, 0, cc) -> i1 setcc(l, r, cc)
+      if ((True = dyn_cast<ConstantSDNode>(N->getOperand(2)))
+          && (False = dyn_cast<ConstantSDNode>(N->getOperand(3)))
+          && True->isAllOnesValue()
+          && False->isNullValue()
+          && VT == MVT::i1) {
+        return DAG.getNode(ISD::SETCC, DL, VT, N->getOperand(0),
+                           N->getOperand(1), N->getOperand(4));
+
+      }
+      break;
+    }
+    case ISD::SETCC: {
+      SDValue Arg0 = N->getOperand(0);
+      SDValue Arg1 = N->getOperand(1);
+      SDValue CC = N->getOperand(2);
+      ConstantSDNode * C = NULL;
+      ISD::CondCode CCOp = dyn_cast<CondCodeSDNode>(CC)->get();
+
+      // i1 setcc (sext(i1), 0, setne) -> i1 setcc(i1, 0, setne)
+      if (VT == MVT::i1
+          && Arg0.getOpcode() == ISD::SIGN_EXTEND
+          && Arg0.getOperand(0).getValueType() == MVT::i1
+          && (C = dyn_cast<ConstantSDNode>(Arg1))
+          && C->isNullValue()
+          && CCOp == ISD::SETNE) {
+        return SimplifySetCC(VT, Arg0.getOperand(0),
+                             DAG.getConstant(0, MVT::i1), CCOp, true, DCI, DL);
+      }
+      break;
+    }
+  }
+  return SDValue();
+}
+
+/// \brief Test if RegClass is one of the VSrc classes 
+static bool isVSrc(unsigned RegClass) {
+  return AMDGPU::VSrc_32RegClassID == RegClass ||
+         AMDGPU::VSrc_64RegClassID == RegClass;
+}
+
+/// \brief Test if RegClass is one of the SSrc classes 
+static bool isSSrc(unsigned RegClass) {
+  return AMDGPU::SSrc_32RegClassID == RegClass ||
+         AMDGPU::SSrc_64RegClassID == RegClass;
+}
+
+/// \brief Analyze the possible immediate value Op
+///
+/// Returns -1 if it isn't an immediate, 0 if it's and inline immediate
+/// and the immediate value if it's a literal immediate
+int32_t SITargetLowering::analyzeImmediate(const SDNode *N) const {
+
+  union {
+    int32_t I;
+    float F;
+  } Imm;
+
+  if (const ConstantSDNode *Node = dyn_cast<ConstantSDNode>(N))
+    Imm.I = Node->getSExtValue();
+  else if (const ConstantFPSDNode *Node = dyn_cast<ConstantFPSDNode>(N))
+    Imm.F = Node->getValueAPF().convertToFloat();
+  else
+    return -1; // It isn't an immediate
+
+  if ((Imm.I >= -16 && Imm.I <= 64) ||
+      Imm.F == 0.5f || Imm.F == -0.5f ||
+      Imm.F == 1.0f || Imm.F == -1.0f ||
+      Imm.F == 2.0f || Imm.F == -2.0f ||
+      Imm.F == 4.0f || Imm.F == -4.0f)
+    return 0; // It's an inline immediate
+
+  return Imm.I; // It's a literal immediate
+}
+
+/// \brief Try to fold an immediate directly into an instruction
+bool SITargetLowering::foldImm(SDValue &Operand, int32_t &Immediate,
+                               bool &ScalarSlotUsed) const {
+
+  MachineSDNode *Mov = dyn_cast<MachineSDNode>(Operand);
+  if (Mov == 0 || !TII->isMov(Mov->getMachineOpcode()))
+    return false;
+
+  const SDValue &Op = Mov->getOperand(0);
+  int32_t Value = analyzeImmediate(Op.getNode());
+  if (Value == -1) {
+    // Not an immediate at all
+    return false;
+
+  } else if (Value == 0) {
+    // Inline immediates can always be fold
+    Operand = Op;
+    return true;
+
+  } else if (Value == Immediate) {
+    // Already fold literal immediate
+    Operand = Op;
+    return true;
+
+  } else if (!ScalarSlotUsed && !Immediate) {
+    // Fold this literal immediate
+    ScalarSlotUsed = true;
+    Immediate = Value;
+    Operand = Op;
+    return true;
+
+  }
+
+  return false;
+}
+
+/// \brief Does "Op" fit into register class "RegClass" ?
+bool SITargetLowering::fitsRegClass(SelectionDAG &DAG, SDValue &Op,
+                                    unsigned RegClass) const {
+
+  MachineRegisterInfo &MRI = DAG.getMachineFunction().getRegInfo(); 
+  SDNode *Node = Op.getNode();
+
+  const TargetRegisterClass *OpClass;
+  if (MachineSDNode *MN = dyn_cast<MachineSDNode>(Node)) {
+    const MCInstrDesc &Desc = TII->get(MN->getMachineOpcode());
+    int OpClassID = Desc.OpInfo[Op.getResNo()].RegClass;
+    if (OpClassID == -1)
+      OpClass = getRegClassFor(Op.getSimpleValueType());
+    else
+      OpClass = TRI->getRegClass(OpClassID);
+
+  } else if (Node->getOpcode() == ISD::CopyFromReg) {
+    RegisterSDNode *Reg = cast<RegisterSDNode>(Node->getOperand(1).getNode());
+    OpClass = MRI.getRegClass(Reg->getReg());
+
+  } else
+    return false;
+
+  return TRI->getRegClass(RegClass)->hasSubClassEq(OpClass);
+}
+
+/// \brief Make sure that we don't exeed the number of allowed scalars
+void SITargetLowering::ensureSRegLimit(SelectionDAG &DAG, SDValue &Operand,
+                                       unsigned RegClass,
+                                       bool &ScalarSlotUsed) const {
+
+  // First map the operands register class to a destination class
+  if (RegClass == AMDGPU::VSrc_32RegClassID)
+    RegClass = AMDGPU::VReg_32RegClassID;
+  else if (RegClass == AMDGPU::VSrc_64RegClassID)
+    RegClass = AMDGPU::VReg_64RegClassID;
+  else
+    return;
+
+  // Nothing todo if they fit naturaly
+  if (fitsRegClass(DAG, Operand, RegClass))
+    return;
+
+  // If the scalar slot isn't used yet use it now
+  if (!ScalarSlotUsed) {
+    ScalarSlotUsed = true;
+    return;
+  }
+
+  // This is a conservative aproach, it is possible that we can't determine
+  // the correct register class and copy too often, but better save than sorry.
+  SDValue RC = DAG.getTargetConstant(RegClass, MVT::i32);
+  SDNode *Node = DAG.getMachineNode(TargetOpcode::COPY_TO_REGCLASS, DebugLoc(),
+                                    Operand.getValueType(), Operand, RC);
+  Operand = SDValue(Node, 0);
+}
+
+SDNode *SITargetLowering::PostISelFolding(MachineSDNode *Node,
+                                          SelectionDAG &DAG) const {
+
+  // Original encoding (either e32 or e64)
+  int Opcode = Node->getMachineOpcode();
+  const MCInstrDesc *Desc = &TII->get(Opcode);
+
+  unsigned NumDefs = Desc->getNumDefs();
+  unsigned NumOps = Desc->getNumOperands();
+
+  // Commuted opcode if available
+  int OpcodeRev = Desc->isCommutable() ? TII->commuteOpcode(Opcode) : -1;
+  const MCInstrDesc *DescRev = OpcodeRev == -1 ? 0 : &TII->get(OpcodeRev);
+
+  assert(!DescRev || DescRev->getNumDefs() == NumDefs);
+  assert(!DescRev || DescRev->getNumOperands() == NumOps);
+
+  // e64 version if available, -1 otherwise
+  int OpcodeE64 = AMDGPU::getVOPe64(Opcode);
+  const MCInstrDesc *DescE64 = OpcodeE64 == -1 ? 0 : &TII->get(OpcodeE64);
+
+  assert(!DescE64 || DescE64->getNumDefs() == NumDefs);
+  assert(!DescE64 || DescE64->getNumOperands() == (NumOps + 4));
+
+  int32_t Immediate = Desc->getSize() == 4 ? 0 : -1;
+  bool HaveVSrc = false, HaveSSrc = false;
+
+  // First figure out what we alread have in this instruction
+  for (unsigned i = 0, e = Node->getNumOperands(), Op = NumDefs;
+       i != e && Op < NumOps; ++i, ++Op) {
+
+    unsigned RegClass = Desc->OpInfo[Op].RegClass;
+    if (isVSrc(RegClass))
+      HaveVSrc = true;
+    else if (isSSrc(RegClass))
+      HaveSSrc = true;
+    else
+      continue;
+
+    int32_t Imm = analyzeImmediate(Node->getOperand(i).getNode());
+    if (Imm != -1 && Imm != 0) {
+      // Literal immediate
+      Immediate = Imm;
+    }
+  }
+
+  // If we neither have VSrc nor SSrc it makes no sense to continue
+  if (!HaveVSrc && !HaveSSrc)
+    return Node;
+
+  // No scalar allowed when we have both VSrc and SSrc
+  bool ScalarSlotUsed = HaveVSrc && HaveSSrc;
+
+  // Second go over the operands and try to fold them
+  std::vector<SDValue> Ops;
+  bool Promote2e64 = false;
+  for (unsigned i = 0, e = Node->getNumOperands(), Op = NumDefs;
+       i != e && Op < NumOps; ++i, ++Op) {
+
+    const SDValue &Operand = Node->getOperand(i);
+    Ops.push_back(Operand);
+
+    // Already folded immediate ?
+    if (isa<ConstantSDNode>(Operand.getNode()) ||
+        isa<ConstantFPSDNode>(Operand.getNode()))
+      continue;
+
+    // Is this a VSrc or SSrc operand ?
+    unsigned RegClass = Desc->OpInfo[Op].RegClass;
+    if (isVSrc(RegClass) || isSSrc(RegClass)) {
+      // Try to fold the immediates
+      if (!foldImm(Ops[i], Immediate, ScalarSlotUsed)) {
+        // Folding didn't worked, make sure we don't hit the SReg limit
+        ensureSRegLimit(DAG, Ops[i], RegClass, ScalarSlotUsed);
+      }
+      continue;
+    }
+
+    if (i == 1 && DescRev && fitsRegClass(DAG, Ops[0], RegClass)) {
+
+      unsigned OtherRegClass = Desc->OpInfo[NumDefs].RegClass;
+      assert(isVSrc(OtherRegClass) || isSSrc(OtherRegClass));
+
+      // Test if it makes sense to swap operands
+      if (foldImm(Ops[1], Immediate, ScalarSlotUsed) ||
+          (!fitsRegClass(DAG, Ops[1], RegClass) &&
+           fitsRegClass(DAG, Ops[1], OtherRegClass))) {
+
+        // Swap commutable operands
+        SDValue Tmp = Ops[1];
+        Ops[1] = Ops[0];
+        Ops[0] = Tmp;
+
+        Desc = DescRev;
+        DescRev = 0;
+        continue;
+      }
+    }
+
+    if (DescE64 && !Immediate) {
+
+      // Test if it makes sense to switch to e64 encoding
+      unsigned OtherRegClass = DescE64->OpInfo[Op].RegClass;
+      if (!isVSrc(OtherRegClass) && !isSSrc(OtherRegClass))
+        continue;
+
+      int32_t TmpImm = -1;
+      if (foldImm(Ops[i], TmpImm, ScalarSlotUsed) ||
+          (!fitsRegClass(DAG, Ops[i], RegClass) &&
+           fitsRegClass(DAG, Ops[1], OtherRegClass))) {
+
+        // Switch to e64 encoding
+        Immediate = -1;
+        Promote2e64 = true;
+        Desc = DescE64;
+        DescE64 = 0;
+      }
+    }
+  }
+
+  if (Promote2e64) {
+    // Add the modifier flags while promoting
+    for (unsigned i = 0; i < 4; ++i)
+      Ops.push_back(DAG.getTargetConstant(0, MVT::i32));
+  }
+
+  // Add optional chain and glue
+  for (unsigned i = NumOps - NumDefs, e = Node->getNumOperands(); i < e; ++i)
+    Ops.push_back(Node->getOperand(i));
+
+  // Create a complete new instruction
+  return DAG.getMachineNode(Desc->Opcode, Node->getDebugLoc(),
+                            Node->getVTList(), Ops.data(), Ops.size());
+}
diff --git a/lib/Target/R600/SIISelLowering.h b/lib/Target/R600/SIISelLowering.h
new file mode 100644
index 000000000000..5ad2f40f0f3a
--- /dev/null
+++ b/lib/Target/R600/SIISelLowering.h
@@ -0,0 +1,58 @@
+//===-- SIISelLowering.h - SI DAG Lowering Interface ------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief SI DAG Lowering interface definition
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef SIISELLOWERING_H
+#define SIISELLOWERING_H
+
+#include "AMDGPUISelLowering.h"
+#include "SIInstrInfo.h"
+
+namespace llvm {
+
+class SITargetLowering : public AMDGPUTargetLowering {
+  const SIInstrInfo * TII;
+  const TargetRegisterInfo * TRI;
+
+  SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
+
+  bool foldImm(SDValue &Operand, int32_t &Immediate,
+               bool &ScalarSlotUsed) const;
+  bool fitsRegClass(SelectionDAG &DAG, SDValue &Op, unsigned RegClass) const;
+  void ensureSRegLimit(SelectionDAG &DAG, SDValue &Operand, 
+                       unsigned RegClass, bool &ScalarSlotUsed) const;
+
+public:
+  SITargetLowering(TargetMachine &tm);
+
+  SDValue LowerFormalArguments(SDValue Chain, CallingConv::ID CallConv,
+                               bool isVarArg,
+                               const SmallVectorImpl<ISD::InputArg> &Ins,
+                               DebugLoc DL, SelectionDAG &DAG,
+                               SmallVectorImpl<SDValue> &InVals) const;
+
+  virtual MachineBasicBlock * EmitInstrWithCustomInserter(MachineInstr * MI,
+                                              MachineBasicBlock * BB) const;
+  virtual EVT getSetCCResultType(EVT VT) const;
+  virtual MVT getScalarShiftAmountTy(EVT VT) const;
+  virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
+  virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
+  virtual SDNode *PostISelFolding(MachineSDNode *N, SelectionDAG &DAG) const;
+
+  int32_t analyzeImmediate(const SDNode *N) const;
+};
+
+} // End namespace llvm
+
+#endif //SIISELLOWERING_H
diff --git a/lib/Target/R600/SIInsertWaits.cpp b/lib/Target/R600/SIInsertWaits.cpp
new file mode 100644
index 000000000000..98bd3dbb6646
--- /dev/null
+++ b/lib/Target/R600/SIInsertWaits.cpp
@@ -0,0 +1,358 @@
+//===-- SILowerControlFlow.cpp - Use predicates for control flow ----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Insert wait instructions for memory reads and writes.
+///
+/// Memory reads and writes are issued asynchronously, so we need to insert
+/// S_WAITCNT instructions when we want to access any of their results or
+/// overwrite any register that's used asynchronously.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "SIInstrInfo.h"
+#include "SIMachineFunctionInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+
+using namespace llvm;
+
+namespace {
+
+/// \brief One variable for each of the hardware counters
+typedef union {
+  struct {
+    unsigned VM;
+    unsigned EXP;
+    unsigned LGKM;
+  } Named;
+  unsigned Array[3];
+
+} Counters;
+
+typedef Counters RegCounters[512];
+typedef std::pair<unsigned, unsigned> RegInterval;
+
+class SIInsertWaits : public MachineFunctionPass {
+
+private:
+  static char ID;
+  const SIInstrInfo *TII;
+  const SIRegisterInfo &TRI;
+  const MachineRegisterInfo *MRI;
+
+  /// \brief Constant hardware limits
+  static const Counters WaitCounts;
+
+  /// \brief Constant zero value
+  static const Counters ZeroCounts;
+
+  /// \brief Counter values we have already waited on.
+  Counters WaitedOn;
+
+  /// \brief Counter values for last instruction issued.
+  Counters LastIssued;
+
+  /// \brief Registers used by async instructions.
+  RegCounters UsedRegs;
+
+  /// \brief Registers defined by async instructions.
+  RegCounters DefinedRegs;
+
+  /// \brief Different export instruction types seen since last wait.
+  unsigned ExpInstrTypesSeen;
+
+  /// \brief Get increment/decrement amount for this instruction.
+  Counters getHwCounts(MachineInstr &MI);
+
+  /// \brief Is operand relevant for async execution?
+  bool isOpRelevant(MachineOperand &Op);
+
+  /// \brief Get register interval an operand affects.
+  RegInterval getRegInterval(MachineOperand &Op);
+
+  /// \brief Handle instructions async components
+  void pushInstruction(MachineInstr &MI);
+
+  /// \brief Insert the actual wait instruction
+  bool insertWait(MachineBasicBlock &MBB,
+                  MachineBasicBlock::iterator I,
+                  const Counters &Counts);
+
+  /// \brief Do we need def2def checks?
+  bool unorderedDefines(MachineInstr &MI);
+
+  /// \brief Resolve all operand dependencies to counter requirements
+  Counters handleOperands(MachineInstr &MI);
+
+public:
+  SIInsertWaits(TargetMachine &tm) :
+    MachineFunctionPass(ID),
+    TII(static_cast<const SIInstrInfo*>(tm.getInstrInfo())),
+    TRI(TII->getRegisterInfo()) { }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF);
+
+  const char *getPassName() const {
+    return "SI insert wait  instructions";
+  }
+
+};
+
+} // End anonymous namespace
+
+char SIInsertWaits::ID = 0;
+
+const Counters SIInsertWaits::WaitCounts = { { 15, 7, 7 } };
+const Counters SIInsertWaits::ZeroCounts = { { 0, 0, 0 } };
+
+FunctionPass *llvm::createSIInsertWaits(TargetMachine &tm) {
+  return new SIInsertWaits(tm);
+}
+
+Counters SIInsertWaits::getHwCounts(MachineInstr &MI) {
+
+  uint64_t TSFlags = TII->get(MI.getOpcode()).TSFlags;
+  Counters Result;
+
+  Result.Named.VM = !!(TSFlags & SIInstrFlags::VM_CNT);
+
+  // Only consider stores or EXP for EXP_CNT
+  Result.Named.EXP = !!(TSFlags & SIInstrFlags::EXP_CNT &&
+      (MI.getOpcode() == AMDGPU::EXP || MI.getDesc().mayStore()));
+
+  // LGKM may uses larger values
+  if (TSFlags & SIInstrFlags::LGKM_CNT) {
+
+    MachineOperand &Op = MI.getOperand(0);
+    assert(Op.isReg() && "First LGKM operand must be a register!");
+
+    unsigned Reg = Op.getReg();
+    unsigned Size = TRI.getMinimalPhysRegClass(Reg)->getSize();
+    Result.Named.LGKM = Size > 4 ? 2 : 1;
+
+  } else {
+    Result.Named.LGKM = 0;
+  }
+
+  return Result;
+}
+
+bool SIInsertWaits::isOpRelevant(MachineOperand &Op) {
+
+  // Constants are always irrelevant
+  if (!Op.isReg())
+    return false;
+
+  // Defines are always relevant
+  if (Op.isDef())
+    return true;
+
+  // For exports all registers are relevant
+  MachineInstr &MI = *Op.getParent();
+  if (MI.getOpcode() == AMDGPU::EXP)
+    return true;
+
+  // For stores the stored value is also relevant
+  if (!MI.getDesc().mayStore())
+    return false;
+
+  for (MachineInstr::mop_iterator I = MI.operands_begin(),
+       E = MI.operands_end(); I != E; ++I) {
+
+    if (I->isReg() && I->isUse())
+      return Op.isIdenticalTo(*I);
+  }
+
+  return false;
+}
+
+RegInterval SIInsertWaits::getRegInterval(MachineOperand &Op) {
+
+  if (!Op.isReg())
+    return std::make_pair(0, 0);
+
+  unsigned Reg = Op.getReg();
+  unsigned Size = TRI.getMinimalPhysRegClass(Reg)->getSize();
+
+  assert(Size >= 4);
+
+  RegInterval Result;
+  Result.first = TRI.getEncodingValue(Reg);
+  Result.second = Result.first + Size / 4;
+
+  return Result;
+}
+
+void SIInsertWaits::pushInstruction(MachineInstr &MI) {
+
+  // Get the hardware counter increments and sum them up
+  Counters Increment = getHwCounts(MI);
+  unsigned Sum = 0;
+
+  for (unsigned i = 0; i < 3; ++i) {
+    LastIssued.Array[i] += Increment.Array[i];
+    Sum += Increment.Array[i];
+  }
+
+  // If we don't increase anything then that's it
+  if (Sum == 0)
+    return;
+
+  // Remember which export instructions we have seen
+  if (Increment.Named.EXP) {
+    ExpInstrTypesSeen |= MI.getOpcode() == AMDGPU::EXP ? 1 : 2;
+  }
+
+  for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
+
+    MachineOperand &Op = MI.getOperand(i);
+    if (!isOpRelevant(Op))
+      continue;
+
+    RegInterval Interval = getRegInterval(Op);
+    for (unsigned j = Interval.first; j < Interval.second; ++j) {
+
+      // Remember which registers we define
+      if (Op.isDef())
+        DefinedRegs[j] = LastIssued;
+
+      // and which one we are using
+      if (Op.isUse())
+        UsedRegs[j] = LastIssued;
+    }
+  }
+}
+
+bool SIInsertWaits::insertWait(MachineBasicBlock &MBB,
+                               MachineBasicBlock::iterator I,
+                               const Counters &Required) {
+
+  // End of program? No need to wait on anything
+  if (I != MBB.end() && I->getOpcode() == AMDGPU::S_ENDPGM)
+    return false;
+
+  // Figure out if the async instructions execute in order
+  bool Ordered[3];
+
+  // VM_CNT is always ordered
+  Ordered[0] = true;
+
+  // EXP_CNT is unordered if we have both EXP & VM-writes
+  Ordered[1] = ExpInstrTypesSeen == 3;
+
+  // LGKM_CNT is handled as always unordered. TODO: Handle LDS and GDS
+  Ordered[2] = false;
+
+  // The values we are going to put into the S_WAITCNT instruction
+  Counters Counts = WaitCounts;
+
+  // Do we really need to wait?
+  bool NeedWait = false;
+
+  for (unsigned i = 0; i < 3; ++i) {
+
+    if (Required.Array[i] <= WaitedOn.Array[i])
+      continue;
+
+    NeedWait = true;
+    
+    if (Ordered[i]) {
+      unsigned Value = LastIssued.Array[i] - Required.Array[i];
+
+      // adjust the value to the real hardware posibilities
+      Counts.Array[i] = std::min(Value, WaitCounts.Array[i]);
+
+    } else
+      Counts.Array[i] = 0;
+
+    // Remember on what we have waited on
+    WaitedOn.Array[i] = LastIssued.Array[i] - Counts.Array[i];
+  }
+
+  if (!NeedWait)
+    return false;
+
+  // Reset EXP_CNT instruction types
+  if (Counts.Named.EXP == 0)
+    ExpInstrTypesSeen = 0;
+
+  // Build the wait instruction
+  BuildMI(MBB, I, DebugLoc(), TII->get(AMDGPU::S_WAITCNT))
+          .addImm((Counts.Named.VM & 0xF) |
+                  ((Counts.Named.EXP & 0x7) << 4) |
+                  ((Counts.Named.LGKM & 0x7) << 8));
+
+  return true;
+}
+
+/// \brief helper function for handleOperands
+static void increaseCounters(Counters &Dst, const Counters &Src) {
+
+  for (unsigned i = 0; i < 3; ++i)
+    Dst.Array[i] = std::max(Dst.Array[i], Src.Array[i]);
+}
+
+Counters SIInsertWaits::handleOperands(MachineInstr &MI) {
+
+  Counters Result = ZeroCounts;
+
+  // For each register affected by this
+  // instruction increase the result sequence
+  for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
+
+    MachineOperand &Op = MI.getOperand(i);
+    RegInterval Interval = getRegInterval(Op);
+    for (unsigned j = Interval.first; j < Interval.second; ++j) {
+
+      if (Op.isDef()) {
+        increaseCounters(Result, UsedRegs[j]);
+        increaseCounters(Result, DefinedRegs[j]);
+      }
+
+      if (Op.isUse())
+        increaseCounters(Result, DefinedRegs[j]);
+    }
+  }
+
+  return Result;
+}
+
+bool SIInsertWaits::runOnMachineFunction(MachineFunction &MF) {
+
+  bool Changes = false;
+
+  MRI = &MF.getRegInfo();
+
+  WaitedOn = ZeroCounts;
+  LastIssued = ZeroCounts;
+
+  memset(&UsedRegs, 0, sizeof(UsedRegs));
+  memset(&DefinedRegs, 0, sizeof(DefinedRegs));
+
+  for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
+       BI != BE; ++BI) {
+
+    MachineBasicBlock &MBB = *BI;
+    for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
+         I != E; ++I) {
+
+      Changes |= insertWait(MBB, I, handleOperands(*I));
+      pushInstruction(*I);
+    }
+
+    // Wait for everything at the end of the MBB
+    Changes |= insertWait(MBB, MBB.getFirstTerminator(), LastIssued);
+  }
+
+  return Changes;
+}
diff --git a/lib/Target/R600/SIInstrFormats.td b/lib/Target/R600/SIInstrFormats.td
new file mode 100644
index 000000000000..3891ddb2dbe2
--- /dev/null
+++ b/lib/Target/R600/SIInstrFormats.td
@@ -0,0 +1,426 @@
+//===-- SIInstrFormats.td - SI Instruction Encodings ----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// SI Instruction format definitions.
+//
+//===----------------------------------------------------------------------===//
+
+class InstSI <dag outs, dag ins, string asm, list<dag> pattern> :
+    AMDGPUInst<outs, ins, asm, pattern> {
+
+  field bits<1> VM_CNT = 0;
+  field bits<1> EXP_CNT = 0;
+  field bits<1> LGKM_CNT = 0;
+
+  let TSFlags{0} = VM_CNT;
+  let TSFlags{1} = EXP_CNT;
+  let TSFlags{2} = LGKM_CNT;
+}
+
+class Enc32 <dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI <outs, ins, asm, pattern> {
+
+  field bits<32> Inst;
+  let Size = 4;
+}
+
+class Enc64 <dag outs, dag ins, string asm, list<dag> pattern> :
+    InstSI <outs, ins, asm, pattern> {
+
+  field bits<64> Inst;
+  let Size = 8;
+}
+
+//===----------------------------------------------------------------------===//
+// Scalar operations
+//===----------------------------------------------------------------------===//
+
+class SOP1 <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    Enc32<outs, ins, asm, pattern> {
+
+  bits<7> SDST;
+  bits<8> SSRC0;
+
+  let Inst{7-0} = SSRC0;
+  let Inst{15-8} = op;
+  let Inst{22-16} = SDST;
+  let Inst{31-23} = 0x17d; //encoding;
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+}
+
+class SOP2 <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    Enc32 <outs, ins, asm, pattern> {
+  
+  bits<7> SDST;
+  bits<8> SSRC0;
+  bits<8> SSRC1;
+
+  let Inst{7-0} = SSRC0;
+  let Inst{15-8} = SSRC1;
+  let Inst{22-16} = SDST;
+  let Inst{29-23} = op;
+  let Inst{31-30} = 0x2; // encoding
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+}
+
+class SOPC <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
+  Enc32<outs, ins, asm, pattern> {
+
+  bits<8> SSRC0;
+  bits<8> SSRC1;
+
+  let Inst{7-0} = SSRC0;
+  let Inst{15-8} = SSRC1;
+  let Inst{22-16} = op;
+  let Inst{31-23} = 0x17e;
+
+  let DisableEncoding = "$dst";
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+}
+
+class SOPK <bits<5> op, dag outs, dag ins, string asm, list<dag> pattern> :
+   Enc32 <outs, ins , asm, pattern> {
+
+  bits <7> SDST;
+  bits <16> SIMM16;
+  
+  let Inst{15-0} = SIMM16;
+  let Inst{22-16} = SDST;
+  let Inst{27-23} = op;
+  let Inst{31-28} = 0xb; //encoding
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+}
+
+class SOPP <bits<7> op, dag ins, string asm, list<dag> pattern> : Enc32 <
+  (outs),
+  ins,
+  asm,
+  pattern > {
+
+  bits <16> SIMM16;
+
+  let Inst{15-0} = SIMM16;
+  let Inst{22-16} = op;
+  let Inst{31-23} = 0x17f; // encoding
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+}
+
+class SMRD <bits<5> op, bits<1> imm, dag outs, dag ins, string asm,
+            list<dag> pattern> : Enc32<outs, ins, asm, pattern> {
+
+  bits<7> SDST;
+  bits<7> SBASE;
+  bits<8> OFFSET;
+  
+  let Inst{7-0} = OFFSET;
+  let Inst{8} = imm;
+  let Inst{14-9} = SBASE{6-1};
+  let Inst{21-15} = SDST;
+  let Inst{26-22} = op;
+  let Inst{31-27} = 0x18; //encoding
+
+  let LGKM_CNT = 1;
+}
+
+//===----------------------------------------------------------------------===//
+// Vector ALU operations
+//===----------------------------------------------------------------------===//
+    
+let Uses = [EXEC] in {
+
+class VOP1 <bits<8> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    Enc32 <outs, ins, asm, pattern> {
+
+  bits<8> VDST;
+  bits<9> SRC0;
+  
+  let Inst{8-0} = SRC0;
+  let Inst{16-9} = op;
+  let Inst{24-17} = VDST;
+  let Inst{31-25} = 0x3f; //encoding
+  
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+}
+
+class VOP2 <bits<6> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    Enc32 <outs, ins, asm, pattern> {
+
+  bits<8> VDST;
+  bits<9> SRC0;
+  bits<8> VSRC1;
+  
+  let Inst{8-0} = SRC0;
+  let Inst{16-9} = VSRC1;
+  let Inst{24-17} = VDST;
+  let Inst{30-25} = op;
+  let Inst{31} = 0x0; //encoding
+  
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+}
+
+class VOP3 <bits<9> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    Enc64 <outs, ins, asm, pattern> {
+
+  bits<8> VDST;
+  bits<9> SRC0;
+  bits<9> SRC1;
+  bits<9> SRC2;
+  bits<3> ABS; 
+  bits<1> CLAMP;
+  bits<2> OMOD;
+  bits<3> NEG;
+
+  let Inst{7-0} = VDST;
+  let Inst{10-8} = ABS;
+  let Inst{11} = CLAMP;
+  let Inst{25-17} = op;
+  let Inst{31-26} = 0x34; //encoding
+  let Inst{40-32} = SRC0;
+  let Inst{49-41} = SRC1;
+  let Inst{58-50} = SRC2;
+  let Inst{60-59} = OMOD;
+  let Inst{63-61} = NEG;
+  
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+}
+
+class VOP3b <bits<9> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    Enc64 <outs, ins, asm, pattern> {
+
+  bits<8> VDST;
+  bits<9> SRC0;
+  bits<9> SRC1;
+  bits<9> SRC2;
+  bits<7> SDST;
+  bits<2> OMOD;
+  bits<3> NEG;
+
+  let Inst{7-0} = VDST;
+  let Inst{14-8} = SDST;
+  let Inst{25-17} = op;
+  let Inst{31-26} = 0x34; //encoding
+  let Inst{40-32} = SRC0;
+  let Inst{49-41} = SRC1;
+  let Inst{58-50} = SRC2;
+  let Inst{60-59} = OMOD;
+  let Inst{63-61} = NEG;
+
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+}
+
+class VOPC <bits<8> op, dag ins, string asm, list<dag> pattern> :
+    Enc32 <(outs VCCReg:$dst), ins, asm, pattern> {
+
+  bits<9> SRC0;
+  bits<8> VSRC1;
+
+  let Inst{8-0} = SRC0;
+  let Inst{16-9} = VSRC1;
+  let Inst{24-17} = op;
+  let Inst{31-25} = 0x3e;
+ 
+  let DisableEncoding = "$dst";
+  let mayLoad = 0;
+  let mayStore = 0;
+  let hasSideEffects = 0;
+}
+
+class VINTRP <bits <2> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    Enc32 <outs, ins, asm, pattern> {
+
+  bits<8> VDST;
+  bits<8> VSRC;
+  bits<2> ATTRCHAN;
+  bits<6> ATTR;
+
+  let Inst{7-0} = VSRC;
+  let Inst{9-8} = ATTRCHAN;
+  let Inst{15-10} = ATTR;
+  let Inst{17-16} = op;
+  let Inst{25-18} = VDST;
+  let Inst{31-26} = 0x32; // encoding
+
+  let neverHasSideEffects = 1;
+  let mayLoad = 1;
+  let mayStore = 0;
+}
+
+} // End Uses = [EXEC]
+
+//===----------------------------------------------------------------------===//
+// Vector I/O operations
+//===----------------------------------------------------------------------===//
+
+let Uses = [EXEC] in {
+
+class MUBUF <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    Enc64<outs, ins, asm, pattern> {
+
+  bits<8> VDATA;
+  bits<12> OFFSET;
+  bits<1> OFFEN;
+  bits<1> IDXEN;
+  bits<1> GLC;
+  bits<1> ADDR64;
+  bits<1> LDS;
+  bits<8> VADDR;
+  bits<7> SRSRC;
+  bits<1> SLC;
+  bits<1> TFE;
+  bits<8> SOFFSET;
+
+  let Inst{11-0} = OFFSET;
+  let Inst{12} = OFFEN;
+  let Inst{13} = IDXEN;
+  let Inst{14} = GLC;
+  let Inst{15} = ADDR64;
+  let Inst{16} = LDS;
+  let Inst{24-18} = op;
+  let Inst{31-26} = 0x38; //encoding
+  let Inst{39-32} = VADDR;
+  let Inst{47-40} = VDATA;
+  let Inst{52-48} = SRSRC{6-2};
+  let Inst{54} = SLC;
+  let Inst{55} = TFE;
+  let Inst{63-56} = SOFFSET;
+
+  let VM_CNT = 1;
+  let EXP_CNT = 1;
+
+  let neverHasSideEffects = 1;
+}
+
+class MTBUF <bits<3> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    Enc64<outs, ins, asm, pattern> {
+
+  bits<8> VDATA;
+  bits<12> OFFSET;
+  bits<1> OFFEN;
+  bits<1> IDXEN;
+  bits<1> GLC;
+  bits<1> ADDR64;
+  bits<4> DFMT;
+  bits<3> NFMT;
+  bits<8> VADDR;
+  bits<7> SRSRC;
+  bits<1> SLC;
+  bits<1> TFE;
+  bits<8> SOFFSET;
+
+  let Inst{11-0} = OFFSET;
+  let Inst{12} = OFFEN;
+  let Inst{13} = IDXEN;
+  let Inst{14} = GLC;
+  let Inst{15} = ADDR64;
+  let Inst{18-16} = op;
+  let Inst{22-19} = DFMT;
+  let Inst{25-23} = NFMT;
+  let Inst{31-26} = 0x3a; //encoding
+  let Inst{39-32} = VADDR;
+  let Inst{47-40} = VDATA;
+  let Inst{52-48} = SRSRC{6-2};
+  let Inst{54} = SLC;
+  let Inst{55} = TFE;
+  let Inst{63-56} = SOFFSET;
+
+  let VM_CNT = 1;
+  let EXP_CNT = 1;
+
+  let neverHasSideEffects = 1;
+}
+
+class MIMG <bits<7> op, dag outs, dag ins, string asm, list<dag> pattern> :
+    Enc64 <outs, ins, asm, pattern> {
+
+  bits<8> VDATA;
+  bits<4> DMASK;
+  bits<1> UNORM;
+  bits<1> GLC;
+  bits<1> DA;
+  bits<1> R128;
+  bits<1> TFE;
+  bits<1> LWE;
+  bits<1> SLC;
+  bits<8> VADDR;
+  bits<7> SRSRC;
+  bits<7> SSAMP; 
+
+  let Inst{11-8} = DMASK;
+  let Inst{12} = UNORM;
+  let Inst{13} = GLC;
+  let Inst{14} = DA;
+  let Inst{15} = R128;
+  let Inst{16} = TFE;
+  let Inst{17} = LWE;
+  let Inst{24-18} = op;
+  let Inst{25} = SLC;
+  let Inst{31-26} = 0x3c;
+  let Inst{39-32} = VADDR;
+  let Inst{47-40} = VDATA;
+  let Inst{52-48} = SRSRC{6-2};
+  let Inst{57-53} = SSAMP{6-2};
+
+  let VM_CNT = 1;
+  let EXP_CNT = 1;
+}
+
+def EXP : Enc64<
+  (outs),
+  (ins i32imm:$en, i32imm:$tgt, i32imm:$compr, i32imm:$done, i32imm:$vm,
+       VReg_32:$src0, VReg_32:$src1, VReg_32:$src2, VReg_32:$src3),
+  "EXP $en, $tgt, $compr, $done, $vm, $src0, $src1, $src2, $src3",
+  [] > {
+
+  bits<4> EN;
+  bits<6> TGT;
+  bits<1> COMPR;
+  bits<1> DONE;
+  bits<1> VM;
+  bits<8> VSRC0;
+  bits<8> VSRC1;
+  bits<8> VSRC2;
+  bits<8> VSRC3;
+
+  let Inst{3-0} = EN;
+  let Inst{9-4} = TGT;
+  let Inst{10} = COMPR;
+  let Inst{11} = DONE;
+  let Inst{12} = VM;
+  let Inst{31-26} = 0x3e;
+  let Inst{39-32} = VSRC0;
+  let Inst{47-40} = VSRC1;
+  let Inst{55-48} = VSRC2;
+  let Inst{63-56} = VSRC3;
+
+  let EXP_CNT = 1;
+}
+
+} // End Uses = [EXEC]
diff --git a/lib/Target/R600/SIInstrInfo.cpp b/lib/Target/R600/SIInstrInfo.cpp
new file mode 100644
index 000000000000..0bfcef562f04
--- /dev/null
+++ b/lib/Target/R600/SIInstrInfo.cpp
@@ -0,0 +1,264 @@
+//===-- SIInstrInfo.cpp - SI Instruction Information  ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief SI Implementation of TargetInstrInfo.
+//
+//===----------------------------------------------------------------------===//
+
+
+#include "SIInstrInfo.h"
+#include "AMDGPUTargetMachine.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/MC/MCInstrDesc.h"
+#include <stdio.h>
+
+using namespace llvm;
+
+SIInstrInfo::SIInstrInfo(AMDGPUTargetMachine &tm)
+  : AMDGPUInstrInfo(tm),
+    RI(tm, *this)
+    { }
+
+const SIRegisterInfo &SIInstrInfo::getRegisterInfo() const {
+  return RI;
+}
+
+void
+SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
+                         MachineBasicBlock::iterator MI, DebugLoc DL,
+                         unsigned DestReg, unsigned SrcReg,
+                         bool KillSrc) const {
+
+  // If we are trying to copy to or from SCC, there is a bug somewhere else in
+  // the backend.  While it may be theoretically possible to do this, it should
+  // never be necessary.
+  assert(DestReg != AMDGPU::SCC && SrcReg != AMDGPU::SCC);
+
+  const int16_t Sub0_15[] = {
+    AMDGPU::sub0, AMDGPU::sub1, AMDGPU::sub2, AMDGPU::sub3,
+    AMDGPU::sub4, AMDGPU::sub5, AMDGPU::sub6, AMDGPU::sub7,
+    AMDGPU::sub8, AMDGPU::sub9, AMDGPU::sub10, AMDGPU::sub11,
+    AMDGPU::sub12, AMDGPU::sub13, AMDGPU::sub14, AMDGPU::sub15, 0
+  };
+
+  const int16_t Sub0_7[] = {
+    AMDGPU::sub0, AMDGPU::sub1, AMDGPU::sub2, AMDGPU::sub3,
+    AMDGPU::sub4, AMDGPU::sub5, AMDGPU::sub6, AMDGPU::sub7, 0
+  };
+
+  const int16_t Sub0_3[] = {
+    AMDGPU::sub0, AMDGPU::sub1, AMDGPU::sub2, AMDGPU::sub3, 0
+  };
+
+  const int16_t Sub0_1[] = {
+    AMDGPU::sub0, AMDGPU::sub1, 0
+  };
+
+  unsigned Opcode;
+  const int16_t *SubIndices;
+
+  if (AMDGPU::M0 == DestReg) {
+    // Check if M0 isn't already set to this value
+    for (MachineBasicBlock::reverse_iterator E = MBB.rend(),
+      I = MachineBasicBlock::reverse_iterator(MI); I != E; ++I) {
+
+      if (!I->definesRegister(AMDGPU::M0))
+        continue;
+
+      unsigned Opc = I->getOpcode();
+      if (Opc != TargetOpcode::COPY && Opc != AMDGPU::S_MOV_B32)
+        break;
+
+      if (!I->readsRegister(SrcReg))
+        break;
+
+      // The copy isn't necessary
+      return;
+    }
+  }
+
+  if (AMDGPU::SReg_32RegClass.contains(DestReg)) {
+    assert(AMDGPU::SReg_32RegClass.contains(SrcReg));
+    BuildMI(MBB, MI, DL, get(AMDGPU::S_MOV_B32), DestReg)
+            .addReg(SrcReg, getKillRegState(KillSrc));
+    return;
+
+  } else if (AMDGPU::SReg_64RegClass.contains(DestReg)) {
+    assert(AMDGPU::SReg_64RegClass.contains(SrcReg));
+    BuildMI(MBB, MI, DL, get(AMDGPU::S_MOV_B64), DestReg)
+            .addReg(SrcReg, getKillRegState(KillSrc));
+    return;
+
+  } else if (AMDGPU::SReg_128RegClass.contains(DestReg)) {
+    assert(AMDGPU::SReg_128RegClass.contains(SrcReg));
+    Opcode = AMDGPU::S_MOV_B32;
+    SubIndices = Sub0_3;
+
+  } else if (AMDGPU::SReg_256RegClass.contains(DestReg)) {
+    assert(AMDGPU::SReg_256RegClass.contains(SrcReg));
+    Opcode = AMDGPU::S_MOV_B32;
+    SubIndices = Sub0_7;
+
+  } else if (AMDGPU::SReg_512RegClass.contains(DestReg)) {
+    assert(AMDGPU::SReg_512RegClass.contains(SrcReg));
+    Opcode = AMDGPU::S_MOV_B32;
+    SubIndices = Sub0_15;
+
+  } else if (AMDGPU::VReg_32RegClass.contains(DestReg)) {
+    assert(AMDGPU::VReg_32RegClass.contains(SrcReg) ||
+	   AMDGPU::SReg_32RegClass.contains(SrcReg));
+    BuildMI(MBB, MI, DL, get(AMDGPU::V_MOV_B32_e32), DestReg)
+            .addReg(SrcReg, getKillRegState(KillSrc));
+    return;
+
+  } else if (AMDGPU::VReg_64RegClass.contains(DestReg)) {
+    assert(AMDGPU::VReg_64RegClass.contains(SrcReg) ||
+	   AMDGPU::SReg_64RegClass.contains(SrcReg));
+    Opcode = AMDGPU::V_MOV_B32_e32;
+    SubIndices = Sub0_1;
+
+  } else if (AMDGPU::VReg_128RegClass.contains(DestReg)) {
+    assert(AMDGPU::VReg_128RegClass.contains(SrcReg) ||
+	   AMDGPU::SReg_128RegClass.contains(SrcReg));
+    Opcode = AMDGPU::V_MOV_B32_e32;
+    SubIndices = Sub0_3;
+
+  } else if (AMDGPU::VReg_256RegClass.contains(DestReg)) {
+    assert(AMDGPU::VReg_256RegClass.contains(SrcReg) ||
+	   AMDGPU::SReg_256RegClass.contains(SrcReg));
+    Opcode = AMDGPU::V_MOV_B32_e32;
+    SubIndices = Sub0_7;
+
+  } else if (AMDGPU::VReg_512RegClass.contains(DestReg)) {
+    assert(AMDGPU::VReg_512RegClass.contains(SrcReg) ||
+	   AMDGPU::SReg_512RegClass.contains(SrcReg));
+    Opcode = AMDGPU::V_MOV_B32_e32;
+    SubIndices = Sub0_15;
+
+  } else {
+    llvm_unreachable("Can't copy register!");
+  }
+
+  while (unsigned SubIdx = *SubIndices++) {
+    MachineInstrBuilder Builder = BuildMI(MBB, MI, DL,
+      get(Opcode), RI.getSubReg(DestReg, SubIdx));
+
+    Builder.addReg(RI.getSubReg(SrcReg, SubIdx), getKillRegState(KillSrc));
+
+    if (*SubIndices)
+      Builder.addReg(DestReg, RegState::Define | RegState::Implicit);
+  }
+}
+
+unsigned SIInstrInfo::commuteOpcode(unsigned Opcode) const {
+
+  int NewOpc;
+
+  // Try to map original to commuted opcode
+  if ((NewOpc = AMDGPU::getCommuteRev(Opcode)) != -1)
+    return NewOpc;
+
+  // Try to map commuted to original opcode
+  if ((NewOpc = AMDGPU::getCommuteOrig(Opcode)) != -1)
+    return NewOpc;
+
+  return Opcode;
+}
+
+MachineInstr *SIInstrInfo::commuteInstruction(MachineInstr *MI,
+                                              bool NewMI) const {
+
+  if (MI->getNumOperands() < 3 || !MI->getOperand(1).isReg() ||
+      !MI->getOperand(2).isReg())
+    return 0;
+
+  MI = TargetInstrInfo::commuteInstruction(MI, NewMI);
+
+  if (MI)
+    MI->setDesc(get(commuteOpcode(MI->getOpcode())));
+
+  return MI;
+}
+
+MachineInstr * SIInstrInfo::getMovImmInstr(MachineFunction *MF, unsigned DstReg,
+                                           int64_t Imm) const {
+  MachineInstr * MI = MF->CreateMachineInstr(get(AMDGPU::V_MOV_B32_e32), DebugLoc());
+  MachineInstrBuilder MIB(*MF, MI);
+  MIB.addReg(DstReg, RegState::Define);
+  MIB.addImm(Imm);
+
+  return MI;
+
+}
+
+bool SIInstrInfo::isMov(unsigned Opcode) const {
+  switch(Opcode) {
+  default: return false;
+  case AMDGPU::S_MOV_B32:
+  case AMDGPU::S_MOV_B64:
+  case AMDGPU::V_MOV_B32_e32:
+  case AMDGPU::V_MOV_B32_e64:
+    return true;
+  }
+}
+
+bool
+SIInstrInfo::isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const {
+  return RC != &AMDGPU::EXECRegRegClass;
+}
+
+//===----------------------------------------------------------------------===//
+// Indirect addressing callbacks
+//===----------------------------------------------------------------------===//
+
+unsigned SIInstrInfo::calculateIndirectAddress(unsigned RegIndex,
+                                                 unsigned Channel) const {
+  assert(Channel == 0);
+  return RegIndex;
+}
+
+
+int SIInstrInfo::getIndirectIndexBegin(const MachineFunction &MF) const {
+  llvm_unreachable("Unimplemented");
+}
+
+int SIInstrInfo::getIndirectIndexEnd(const MachineFunction &MF) const {
+  llvm_unreachable("Unimplemented");
+}
+
+const TargetRegisterClass *SIInstrInfo::getIndirectAddrStoreRegClass(
+                                                     unsigned SourceReg) const {
+  llvm_unreachable("Unimplemented");
+}
+
+const TargetRegisterClass *SIInstrInfo::getIndirectAddrLoadRegClass() const {
+  llvm_unreachable("Unimplemented");
+}
+
+MachineInstrBuilder SIInstrInfo::buildIndirectWrite(
+                                   MachineBasicBlock *MBB,
+                                   MachineBasicBlock::iterator I,
+                                   unsigned ValueReg,
+                                   unsigned Address, unsigned OffsetReg) const {
+  llvm_unreachable("Unimplemented");
+}
+
+MachineInstrBuilder SIInstrInfo::buildIndirectRead(
+                                   MachineBasicBlock *MBB,
+                                   MachineBasicBlock::iterator I,
+                                   unsigned ValueReg,
+                                   unsigned Address, unsigned OffsetReg) const {
+  llvm_unreachable("Unimplemented");
+}
+
+const TargetRegisterClass *SIInstrInfo::getSuperIndirectRegClass() const {
+  llvm_unreachable("Unimplemented");
+}
diff --git a/lib/Target/R600/SIInstrInfo.h b/lib/Target/R600/SIInstrInfo.h
new file mode 100644
index 000000000000..d4e60e508634
--- /dev/null
+++ b/lib/Target/R600/SIInstrInfo.h
@@ -0,0 +1,97 @@
+//===-- SIInstrInfo.h - SI Instruction Info Interface ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Interface definition for SIInstrInfo.
+//
+//===----------------------------------------------------------------------===//
+
+
+#ifndef SIINSTRINFO_H
+#define SIINSTRINFO_H
+
+#include "AMDGPUInstrInfo.h"
+#include "SIRegisterInfo.h"
+
+namespace llvm {
+
+class SIInstrInfo : public AMDGPUInstrInfo {
+private:
+  const SIRegisterInfo RI;
+
+public:
+  explicit SIInstrInfo(AMDGPUTargetMachine &tm);
+
+  const SIRegisterInfo &getRegisterInfo() const;
+
+  virtual void copyPhysReg(MachineBasicBlock &MBB,
+                           MachineBasicBlock::iterator MI, DebugLoc DL,
+                           unsigned DestReg, unsigned SrcReg,
+                           bool KillSrc) const;
+
+  unsigned commuteOpcode(unsigned Opcode) const;
+
+  virtual MachineInstr *commuteInstruction(MachineInstr *MI,
+                                           bool NewMI=false) const;
+
+  virtual MachineInstr * getMovImmInstr(MachineFunction *MF, unsigned DstReg,
+                                        int64_t Imm) const;
+
+  virtual unsigned getIEQOpcode() const { assert(!"Implement"); return 0;}
+  virtual bool isMov(unsigned Opcode) const;
+
+  virtual bool isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const;
+
+  virtual int getIndirectIndexBegin(const MachineFunction &MF) const;
+
+  virtual int getIndirectIndexEnd(const MachineFunction &MF) const;
+
+  virtual unsigned calculateIndirectAddress(unsigned RegIndex,
+                                            unsigned Channel) const;
+
+  virtual const TargetRegisterClass *getIndirectAddrStoreRegClass(
+                                                      unsigned SourceReg) const;
+
+  virtual const TargetRegisterClass *getIndirectAddrLoadRegClass() const;
+
+  virtual MachineInstrBuilder buildIndirectWrite(MachineBasicBlock *MBB,
+                                                 MachineBasicBlock::iterator I,
+                                                 unsigned ValueReg,
+                                                 unsigned Address,
+                                                 unsigned OffsetReg) const;
+
+  virtual MachineInstrBuilder buildIndirectRead(MachineBasicBlock *MBB,
+                                                MachineBasicBlock::iterator I,
+                                                unsigned ValueReg,
+                                                unsigned Address,
+                                                unsigned OffsetReg) const;
+
+  virtual const TargetRegisterClass *getSuperIndirectRegClass() const;
+  };
+
+namespace AMDGPU {
+
+  int getVOPe64(uint16_t Opcode);
+  int getCommuteRev(uint16_t Opcode);
+  int getCommuteOrig(uint16_t Opcode);
+
+} // End namespace AMDGPU
+
+} // End namespace llvm
+
+namespace SIInstrFlags {
+  enum Flags {
+    // First 4 bits are the instruction encoding
+    VM_CNT = 1 << 0,
+    EXP_CNT = 1 << 1,
+    LGKM_CNT = 1 << 2
+  };
+}
+
+#endif //SIINSTRINFO_H
diff --git a/lib/Target/R600/SIInstrInfo.td b/lib/Target/R600/SIInstrInfo.td
new file mode 100644
index 000000000000..617f0b871c25
--- /dev/null
+++ b/lib/Target/R600/SIInstrInfo.td
@@ -0,0 +1,356 @@
+//===-- SIInstrInfo.td - SI Instruction Infos -------------*- tablegen -*--===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// SI DAG Nodes
+//===----------------------------------------------------------------------===//
+
+// SMRD takes a 64bit memory address and can only add an 32bit offset
+def SIadd64bit32bit : SDNode<"ISD::ADD",
+  SDTypeProfile<1, 2, [SDTCisSameAs<0, 1>, SDTCisVT<0, i64>, SDTCisVT<2, i32>]>
+>;
+
+// Transformation function, extract the lower 32bit of a 64bit immediate
+def LO32 : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(N->getZExtValue() & 0xffffffff, MVT::i32);
+}]>;
+
+// Transformation function, extract the upper 32bit of a 64bit immediate
+def HI32 : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(N->getZExtValue() >> 32, MVT::i32);
+}]>;
+
+def IMM8bitDWORD : ImmLeaf <
+  i32, [{
+    return (Imm & ~0x3FC) == 0;
+  }], SDNodeXForm<imm, [{
+    return CurDAG->getTargetConstant(
+      N->getZExtValue() >> 2, MVT::i32);
+  }]>
+>;
+
+def IMM12bit : ImmLeaf <
+  i16,
+  [{return isUInt<12>(Imm);}]
+>;
+
+class InlineImm <ValueType vt> : PatLeaf <(vt imm), [{
+  return ((const SITargetLowering &)TLI).analyzeImmediate(N) == 0;
+}]>;
+
+//===----------------------------------------------------------------------===//
+// SI assembler operands
+//===----------------------------------------------------------------------===//
+
+def SIOperand {
+  int ZERO = 0x80;
+  int VCC = 0x6A;
+}
+
+include "SIInstrFormats.td"
+
+//===----------------------------------------------------------------------===//
+//
+// SI Instruction multiclass helpers.
+//
+// Instructions with _32 take 32-bit operands.
+// Instructions with _64 take 64-bit operands.
+//
+// VOP_* instructions can use either a 32-bit or 64-bit encoding.  The 32-bit
+// encoding is the standard encoding, but instruction that make use of
+// any of the instruction modifiers must use the 64-bit encoding.
+//
+// Instructions with _e32 use the 32-bit encoding.
+// Instructions with _e64 use the 64-bit encoding.
+//
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// Scalar classes
+//===----------------------------------------------------------------------===//
+
+class SOP1_32 <bits<8> op, string opName, list<dag> pattern> : SOP1 <
+  op, (outs SReg_32:$dst), (ins SSrc_32:$src0),
+  opName#" $dst, $src0", pattern
+>;
+
+class SOP1_64 <bits<8> op, string opName, list<dag> pattern> : SOP1 <
+  op, (outs SReg_64:$dst), (ins SSrc_64:$src0),
+  opName#" $dst, $src0", pattern
+>;
+
+class SOP2_32 <bits<7> op, string opName, list<dag> pattern> : SOP2 <
+  op, (outs SReg_32:$dst), (ins SSrc_32:$src0, SSrc_32:$src1),
+  opName#" $dst, $src0, $src1", pattern
+>;
+
+class SOP2_64 <bits<7> op, string opName, list<dag> pattern> : SOP2 <
+  op, (outs SReg_64:$dst), (ins SSrc_64:$src0, SSrc_64:$src1),
+  opName#" $dst, $src0, $src1", pattern
+>;
+
+class SOPC_32 <bits<7> op, string opName, list<dag> pattern> : SOPC <
+  op, (outs SCCReg:$dst), (ins SSrc_32:$src0, SSrc_32:$src1),
+  opName#" $dst, $src0, $src1", pattern
+>;
+
+class SOPC_64 <bits<7> op, string opName, list<dag> pattern> : SOPC <
+  op, (outs SCCReg:$dst), (ins SSrc_64:$src0, SSrc_64:$src1),
+  opName#" $dst, $src0, $src1", pattern
+>;
+
+class SOPK_32 <bits<5> op, string opName, list<dag> pattern> : SOPK <
+  op, (outs SReg_32:$dst), (ins i16imm:$src0),
+  opName#" $dst, $src0", pattern
+>;
+
+class SOPK_64 <bits<5> op, string opName, list<dag> pattern> : SOPK <
+  op, (outs SReg_64:$dst), (ins i16imm:$src0),
+  opName#" $dst, $src0", pattern
+>;
+
+multiclass SMRD_Helper <bits<5> op, string asm, RegisterClass baseClass,
+                        RegisterClass dstClass> {
+  def _IMM : SMRD <
+    op, 1, (outs dstClass:$dst),
+    (ins baseClass:$sbase, i32imm:$offset),
+    asm#" $dst, $sbase, $offset", []
+  >;
+
+  def _SGPR : SMRD <
+    op, 0, (outs dstClass:$dst),
+    (ins baseClass:$sbase, SReg_32:$soff),
+    asm#" $dst, $sbase, $soff", []
+  >;
+}
+
+//===----------------------------------------------------------------------===//
+// Vector ALU classes
+//===----------------------------------------------------------------------===//
+
+class VOP <string opName> {
+  string OpName = opName;
+}
+
+class VOP2_REV <string revOp, bit isOrig> {
+  string RevOp = revOp;
+  bit IsOrig = isOrig;
+}
+
+multiclass VOP1_Helper <bits<8> op, RegisterClass drc, RegisterClass src,
+                        string opName, list<dag> pattern> {
+
+  def _e32 : VOP1 <
+    op, (outs drc:$dst), (ins src:$src0),
+    opName#"_e32 $dst, $src0", pattern
+  >, VOP <opName>;
+
+  def _e64 : VOP3 <
+    {1, 1, op{6}, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
+    (outs drc:$dst),
+    (ins src:$src0,
+         i32imm:$abs, i32imm:$clamp,
+         i32imm:$omod, i32imm:$neg),
+    opName#"_e64 $dst, $src0, $abs, $clamp, $omod, $neg", []
+  >, VOP <opName> {
+    let SRC1 = SIOperand.ZERO;
+    let SRC2 = SIOperand.ZERO;
+  }
+}
+
+multiclass VOP1_32 <bits<8> op, string opName, list<dag> pattern>
+  : VOP1_Helper <op, VReg_32, VSrc_32, opName, pattern>;
+
+multiclass VOP1_64 <bits<8> op, string opName, list<dag> pattern>
+  : VOP1_Helper <op, VReg_64, VSrc_64, opName, pattern>;
+
+multiclass VOP2_Helper <bits<6> op, RegisterClass vrc, RegisterClass arc,
+                        string opName, list<dag> pattern, string revOp> {
+  def _e32 : VOP2 <
+    op, (outs vrc:$dst), (ins arc:$src0, vrc:$src1),
+    opName#"_e32 $dst, $src0, $src1", pattern
+  >, VOP <opName>, VOP2_REV<revOp#"_e32", !eq(revOp, opName)>;
+
+  def _e64 : VOP3 <
+    {1, 0, 0, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
+    (outs vrc:$dst),
+    (ins arc:$src0, arc:$src1,
+         i32imm:$abs, i32imm:$clamp,
+         i32imm:$omod, i32imm:$neg),
+    opName#"_e64 $dst, $src0, $src1, $abs, $clamp, $omod, $neg", []
+  >, VOP <opName>, VOP2_REV<revOp#"_e64", !eq(revOp, opName)> {
+    let SRC2 = SIOperand.ZERO;
+  }
+}
+
+multiclass VOP2_32 <bits<6> op, string opName, list<dag> pattern,
+                    string revOp = opName>
+  : VOP2_Helper <op, VReg_32, VSrc_32, opName, pattern, revOp>;
+
+multiclass VOP2_64 <bits<6> op, string opName, list<dag> pattern,
+                    string revOp = opName>
+  : VOP2_Helper <op, VReg_64, VSrc_64, opName, pattern, revOp>;
+
+multiclass VOP2b_32 <bits<6> op, string opName, list<dag> pattern,
+                     string revOp = opName> {
+
+  def _e32 : VOP2 <
+    op, (outs VReg_32:$dst), (ins VSrc_32:$src0, VReg_32:$src1),
+    opName#"_e32 $dst, $src0, $src1", pattern
+  >, VOP <opName>, VOP2_REV<revOp#"_e32", !eq(revOp, opName)>;
+
+  def _e64 : VOP3b <
+    {1, 0, 0, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
+    (outs VReg_32:$dst),
+    (ins VSrc_32:$src0, VSrc_32:$src1,
+         i32imm:$abs, i32imm:$clamp,
+         i32imm:$omod, i32imm:$neg),
+    opName#"_e64 $dst, $src0, $src1, $abs, $clamp, $omod, $neg", []
+  >, VOP <opName>, VOP2_REV<revOp#"_e64", !eq(revOp, opName)> {
+    let SRC2 = SIOperand.ZERO;
+    /* the VOP2 variant puts the carry out into VCC, the VOP3 variant
+       can write it into any SGPR. We currently don't use the carry out,
+       so for now hardcode it to VCC as well */
+    let SDST = SIOperand.VCC;
+  }
+}
+
+multiclass VOPC_Helper <bits<8> op, RegisterClass vrc, RegisterClass arc,
+                        string opName, ValueType vt, PatLeaf cond> {
+
+  def _e32 : VOPC <
+    op, (ins arc:$src0, vrc:$src1),
+    opName#"_e32 $dst, $src0, $src1", []
+  >, VOP <opName>;
+
+  def _e64 : VOP3 <
+    {0, op{7}, op{6}, op{5}, op{4}, op{3}, op{2}, op{1}, op{0}},
+    (outs SReg_64:$dst),
+    (ins arc:$src0, arc:$src1,
+         InstFlag:$abs, InstFlag:$clamp,
+         InstFlag:$omod, InstFlag:$neg),
+    opName#"_e64 $dst, $src0, $src1, $abs, $clamp, $omod, $neg",
+    !if(!eq(!cast<string>(cond), "COND_NULL"), []<dag>,
+      [(set SReg_64:$dst, (i1 (setcc (vt arc:$src0), arc:$src1, cond)))]
+    )
+  >, VOP <opName> {
+    let SRC2 = SIOperand.ZERO;
+  }
+}
+
+multiclass VOPC_32 <bits<8> op, string opName,
+  ValueType vt = untyped, PatLeaf cond = COND_NULL>
+  : VOPC_Helper <op, VReg_32, VSrc_32, opName, vt, cond>;
+
+multiclass VOPC_64 <bits<8> op, string opName,
+  ValueType vt = untyped, PatLeaf cond = COND_NULL>
+  : VOPC_Helper <op, VReg_64, VSrc_64, opName, vt, cond>;
+
+class VOP3_32 <bits<9> op, string opName, list<dag> pattern> : VOP3 <
+  op, (outs VReg_32:$dst),
+  (ins VSrc_32:$src0, VSrc_32:$src1, VSrc_32:$src2,
+   i32imm:$abs, i32imm:$clamp, i32imm:$omod, i32imm:$neg),
+  opName#" $dst, $src0, $src1, $src2, $abs, $clamp, $omod, $neg", pattern
+>, VOP <opName>;
+
+class VOP3_64 <bits<9> op, string opName, list<dag> pattern> : VOP3 <
+  op, (outs VReg_64:$dst),
+  (ins VSrc_64:$src0, VSrc_64:$src1, VSrc_64:$src2,
+   i32imm:$abs, i32imm:$clamp, i32imm:$omod, i32imm:$neg),
+  opName#" $dst, $src0, $src1, $src2, $abs, $clamp, $omod, $neg", pattern
+>, VOP <opName>;
+
+//===----------------------------------------------------------------------===//
+// Vector I/O classes
+//===----------------------------------------------------------------------===//
+
+class MTBUF_Store_Helper <bits<3> op, string asm, RegisterClass regClass> : MTBUF <
+  op,
+  (outs),
+  (ins regClass:$vdata, i16imm:$offset, i1imm:$offen, i1imm:$idxen, i1imm:$glc,
+   i1imm:$addr64, i8imm:$dfmt, i8imm:$nfmt, VReg_32:$vaddr,
+   SReg_128:$srsrc, i1imm:$slc, i1imm:$tfe, SSrc_32:$soffset),
+  asm#" $vdata, $offset, $offen, $idxen, $glc, $addr64, $dfmt,"
+     #" $nfmt, $vaddr, $srsrc, $slc, $tfe, $soffset",
+  []> {
+  let mayStore = 1;
+  let mayLoad = 0;
+}
+
+class MUBUF_Load_Helper <bits<7> op, string asm, RegisterClass regClass> : MUBUF <
+  op,
+  (outs regClass:$dst),
+  (ins i16imm:$offset, i1imm:$offen, i1imm:$idxen, i1imm:$glc, i1imm:$addr64,
+       i1imm:$lds, VReg_32:$vaddr, SReg_128:$srsrc, i1imm:$slc,
+       i1imm:$tfe, SSrc_32:$soffset),
+  asm#" $dst, $offset, $offen, $idxen, $glc, $addr64, "
+     #"$lds, $vaddr, $srsrc, $slc, $tfe, $soffset",
+  []> {
+  let mayLoad = 1;
+  let mayStore = 0;
+}
+
+class MTBUF_Load_Helper <bits<3> op, string asm, RegisterClass regClass> : MTBUF <
+  op,
+  (outs regClass:$dst),
+  (ins i16imm:$offset, i1imm:$offen, i1imm:$idxen, i1imm:$glc, i1imm:$addr64,
+       i8imm:$dfmt, i8imm:$nfmt, VReg_32:$vaddr, SReg_128:$srsrc,
+       i1imm:$slc, i1imm:$tfe, SSrc_32:$soffset),
+  asm#" $dst, $offset, $offen, $idxen, $glc, $addr64, $dfmt,"
+     #" $nfmt, $vaddr, $srsrc, $slc, $tfe, $soffset",
+  []> {
+  let mayLoad = 1;
+  let mayStore = 0;
+}
+
+class MIMG_Load_Helper <bits<7> op, string asm> : MIMG <
+  op,
+  (outs VReg_128:$vdata),
+  (ins i32imm:$dmask, i1imm:$unorm, i1imm:$glc, i1imm:$da, i1imm:$r128,
+       i1imm:$tfe, i1imm:$lwe, i1imm:$slc, unknown:$vaddr,
+       SReg_256:$srsrc, SReg_128:$ssamp),
+  asm#" $vdata, $dmask, $unorm, $glc, $da, $r128,"
+     #" $tfe, $lwe, $slc, $vaddr, $srsrc, $ssamp",
+  []> {
+  let mayLoad = 1;
+  let mayStore = 0;
+}
+
+//===----------------------------------------------------------------------===//
+// Vector instruction mappings
+//===----------------------------------------------------------------------===//
+
+// Maps an opcode in e32 form to its e64 equivalent
+def getVOPe64 : InstrMapping {
+  let FilterClass = "VOP";
+  let RowFields = ["OpName"];
+  let ColFields = ["Size"];
+  let KeyCol = ["4"];
+  let ValueCols = [["8"]];
+}
+
+// Maps an original opcode to its commuted version
+def getCommuteRev : InstrMapping {
+  let FilterClass = "VOP2_REV";
+  let RowFields = ["RevOp"];
+  let ColFields = ["IsOrig"];
+  let KeyCol = ["1"];
+  let ValueCols = [["0"]];
+}
+
+// Maps an commuted opcode to its original version
+def getCommuteOrig : InstrMapping {
+  let FilterClass = "VOP2_REV";
+  let RowFields = ["RevOp"];
+  let ColFields = ["IsOrig"];
+  let KeyCol = ["0"];
+  let ValueCols = [["1"]];
+}
+
+include "SIInstructions.td"
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
new file mode 100644
index 000000000000..4f734f91245a
--- /dev/null
+++ b/lib/Target/R600/SIInstructions.td
@@ -0,0 +1,1607 @@
+//===-- SIInstructions.td - SI Instruction Defintions ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+// This file was originally auto-generated from a GPU register header file and
+// all the instruction definitions were originally commented out.  Instructions
+// that are not yet supported remain commented out.
+//===----------------------------------------------------------------------===//
+
+class InterpSlots {
+int P0 = 2;
+int P10 = 0;
+int P20 = 1;
+}
+def INTERP : InterpSlots;
+
+def InterpSlot : Operand<i32> {
+  let PrintMethod = "printInterpSlot";
+}
+
+def isSI : Predicate<"Subtarget.device()"
+                            "->getGeneration() == AMDGPUDeviceInfo::HD7XXX">;
+
+let Predicates = [isSI] in {
+
+let neverHasSideEffects = 1 in {
+
+let isMoveImm = 1 in {
+def S_MOV_B32 : SOP1_32 <0x00000003, "S_MOV_B32", []>;
+def S_MOV_B64 : SOP1_64 <0x00000004, "S_MOV_B64", []>;
+def S_CMOV_B32 : SOP1_32 <0x00000005, "S_CMOV_B32", []>;
+def S_CMOV_B64 : SOP1_64 <0x00000006, "S_CMOV_B64", []>;
+} // End isMoveImm = 1
+
+def S_NOT_B32 : SOP1_32 <0x00000007, "S_NOT_B32", []>;
+def S_NOT_B64 : SOP1_64 <0x00000008, "S_NOT_B64", []>;
+def S_WQM_B32 : SOP1_32 <0x00000009, "S_WQM_B32", []>;
+def S_WQM_B64 : SOP1_64 <0x0000000a, "S_WQM_B64", []>;
+def S_BREV_B32 : SOP1_32 <0x0000000b, "S_BREV_B32", []>;
+def S_BREV_B64 : SOP1_64 <0x0000000c, "S_BREV_B64", []>;
+} // End neverHasSideEffects = 1
+
+////def S_BCNT0_I32_B32 : SOP1_BCNT0 <0x0000000d, "S_BCNT0_I32_B32", []>;
+////def S_BCNT0_I32_B64 : SOP1_BCNT0 <0x0000000e, "S_BCNT0_I32_B64", []>;
+////def S_BCNT1_I32_B32 : SOP1_BCNT1 <0x0000000f, "S_BCNT1_I32_B32", []>;
+////def S_BCNT1_I32_B64 : SOP1_BCNT1 <0x00000010, "S_BCNT1_I32_B64", []>;
+////def S_FF0_I32_B32 : SOP1_FF0 <0x00000011, "S_FF0_I32_B32", []>;
+////def S_FF0_I32_B64 : SOP1_FF0 <0x00000012, "S_FF0_I32_B64", []>;
+////def S_FF1_I32_B32 : SOP1_FF1 <0x00000013, "S_FF1_I32_B32", []>;
+////def S_FF1_I32_B64 : SOP1_FF1 <0x00000014, "S_FF1_I32_B64", []>;
+//def S_FLBIT_I32_B32 : SOP1_32 <0x00000015, "S_FLBIT_I32_B32", []>;
+//def S_FLBIT_I32_B64 : SOP1_32 <0x00000016, "S_FLBIT_I32_B64", []>;
+def S_FLBIT_I32 : SOP1_32 <0x00000017, "S_FLBIT_I32", []>;
+//def S_FLBIT_I32_I64 : SOP1_32 <0x00000018, "S_FLBIT_I32_I64", []>;
+//def S_SEXT_I32_I8 : SOP1_32 <0x00000019, "S_SEXT_I32_I8", []>;
+//def S_SEXT_I32_I16 : SOP1_32 <0x0000001a, "S_SEXT_I32_I16", []>;
+////def S_BITSET0_B32 : SOP1_BITSET0 <0x0000001b, "S_BITSET0_B32", []>;
+////def S_BITSET0_B64 : SOP1_BITSET0 <0x0000001c, "S_BITSET0_B64", []>;
+////def S_BITSET1_B32 : SOP1_BITSET1 <0x0000001d, "S_BITSET1_B32", []>;
+////def S_BITSET1_B64 : SOP1_BITSET1 <0x0000001e, "S_BITSET1_B64", []>;
+def S_GETPC_B64 : SOP1_64 <0x0000001f, "S_GETPC_B64", []>;
+def S_SETPC_B64 : SOP1_64 <0x00000020, "S_SETPC_B64", []>;
+def S_SWAPPC_B64 : SOP1_64 <0x00000021, "S_SWAPPC_B64", []>;
+def S_RFE_B64 : SOP1_64 <0x00000022, "S_RFE_B64", []>;
+
+let hasSideEffects = 1, Uses = [EXEC], Defs = [EXEC] in {
+
+def S_AND_SAVEEXEC_B64 : SOP1_64 <0x00000024, "S_AND_SAVEEXEC_B64", []>;
+def S_OR_SAVEEXEC_B64 : SOP1_64 <0x00000025, "S_OR_SAVEEXEC_B64", []>;
+def S_XOR_SAVEEXEC_B64 : SOP1_64 <0x00000026, "S_XOR_SAVEEXEC_B64", []>;
+def S_ANDN2_SAVEEXEC_B64 : SOP1_64 <0x00000027, "S_ANDN2_SAVEEXEC_B64", []>;
+def S_ORN2_SAVEEXEC_B64 : SOP1_64 <0x00000028, "S_ORN2_SAVEEXEC_B64", []>;
+def S_NAND_SAVEEXEC_B64 : SOP1_64 <0x00000029, "S_NAND_SAVEEXEC_B64", []>;
+def S_NOR_SAVEEXEC_B64 : SOP1_64 <0x0000002a, "S_NOR_SAVEEXEC_B64", []>;
+def S_XNOR_SAVEEXEC_B64 : SOP1_64 <0x0000002b, "S_XNOR_SAVEEXEC_B64", []>;
+
+} // End hasSideEffects = 1
+
+def S_QUADMASK_B32 : SOP1_32 <0x0000002c, "S_QUADMASK_B32", []>;
+def S_QUADMASK_B64 : SOP1_64 <0x0000002d, "S_QUADMASK_B64", []>;
+def S_MOVRELS_B32 : SOP1_32 <0x0000002e, "S_MOVRELS_B32", []>;
+def S_MOVRELS_B64 : SOP1_64 <0x0000002f, "S_MOVRELS_B64", []>;
+def S_MOVRELD_B32 : SOP1_32 <0x00000030, "S_MOVRELD_B32", []>;
+def S_MOVRELD_B64 : SOP1_64 <0x00000031, "S_MOVRELD_B64", []>;
+//def S_CBRANCH_JOIN : SOP1_ <0x00000032, "S_CBRANCH_JOIN", []>;
+def S_MOV_REGRD_B32 : SOP1_32 <0x00000033, "S_MOV_REGRD_B32", []>;
+def S_ABS_I32 : SOP1_32 <0x00000034, "S_ABS_I32", []>;
+def S_MOV_FED_B32 : SOP1_32 <0x00000035, "S_MOV_FED_B32", []>;
+def S_MOVK_I32 : SOPK_32 <0x00000000, "S_MOVK_I32", []>;
+def S_CMOVK_I32 : SOPK_32 <0x00000002, "S_CMOVK_I32", []>;
+
+/*
+This instruction is disabled for now until we can figure out how to teach
+the instruction selector to correctly use the  S_CMP* vs V_CMP*
+instructions.
+
+When this instruction is enabled the code generator sometimes produces this
+invalid sequence:
+
+SCC = S_CMPK_EQ_I32 SGPR0, imm
+VCC = COPY SCC
+VGPR0 = V_CNDMASK VCC, VGPR0, VGPR1
+
+def S_CMPK_EQ_I32 : SOPK <
+  0x00000003, (outs SCCReg:$dst), (ins SReg_32:$src0, i32imm:$src1),
+  "S_CMPK_EQ_I32",
+  [(set SCCReg:$dst, (setcc SReg_32:$src0, imm:$src1, SETEQ))]
+>;
+*/
+
+let isCompare = 1 in {
+def S_CMPK_LG_I32 : SOPK_32 <0x00000004, "S_CMPK_LG_I32", []>;
+def S_CMPK_GT_I32 : SOPK_32 <0x00000005, "S_CMPK_GT_I32", []>;
+def S_CMPK_GE_I32 : SOPK_32 <0x00000006, "S_CMPK_GE_I32", []>;
+def S_CMPK_LT_I32 : SOPK_32 <0x00000007, "S_CMPK_LT_I32", []>;
+def S_CMPK_LE_I32 : SOPK_32 <0x00000008, "S_CMPK_LE_I32", []>;
+def S_CMPK_EQ_U32 : SOPK_32 <0x00000009, "S_CMPK_EQ_U32", []>;
+def S_CMPK_LG_U32 : SOPK_32 <0x0000000a, "S_CMPK_LG_U32", []>;
+def S_CMPK_GT_U32 : SOPK_32 <0x0000000b, "S_CMPK_GT_U32", []>;
+def S_CMPK_GE_U32 : SOPK_32 <0x0000000c, "S_CMPK_GE_U32", []>;
+def S_CMPK_LT_U32 : SOPK_32 <0x0000000d, "S_CMPK_LT_U32", []>;
+def S_CMPK_LE_U32 : SOPK_32 <0x0000000e, "S_CMPK_LE_U32", []>;
+} // End isCompare = 1
+
+def S_ADDK_I32 : SOPK_32 <0x0000000f, "S_ADDK_I32", []>;
+def S_MULK_I32 : SOPK_32 <0x00000010, "S_MULK_I32", []>;
+//def S_CBRANCH_I_FORK : SOPK_ <0x00000011, "S_CBRANCH_I_FORK", []>;
+def S_GETREG_B32 : SOPK_32 <0x00000012, "S_GETREG_B32", []>;
+def S_SETREG_B32 : SOPK_32 <0x00000013, "S_SETREG_B32", []>;
+def S_GETREG_REGRD_B32 : SOPK_32 <0x00000014, "S_GETREG_REGRD_B32", []>;
+//def S_SETREG_IMM32_B32 : SOPK_32 <0x00000015, "S_SETREG_IMM32_B32", []>;
+//def EXP : EXP_ <0x00000000, "EXP", []>;
+
+let isCompare = 1 in {
+
+defm V_CMP_F_F32 : VOPC_32 <0x00000000, "V_CMP_F_F32">;
+defm V_CMP_LT_F32 : VOPC_32 <0x00000001, "V_CMP_LT_F32", f32, COND_LT>;
+defm V_CMP_EQ_F32 : VOPC_32 <0x00000002, "V_CMP_EQ_F32", f32, COND_EQ>;
+defm V_CMP_LE_F32 : VOPC_32 <0x00000003, "V_CMP_LE_F32", f32, COND_LE>;
+defm V_CMP_GT_F32 : VOPC_32 <0x00000004, "V_CMP_GT_F32", f32, COND_GT>;
+defm V_CMP_LG_F32 : VOPC_32 <0x00000005, "V_CMP_LG_F32", f32, COND_NE>;
+defm V_CMP_GE_F32 : VOPC_32 <0x00000006, "V_CMP_GE_F32", f32, COND_GE>;
+defm V_CMP_O_F32 : VOPC_32 <0x00000007, "V_CMP_O_F32">;
+defm V_CMP_U_F32 : VOPC_32 <0x00000008, "V_CMP_U_F32">;
+defm V_CMP_NGE_F32 : VOPC_32 <0x00000009, "V_CMP_NGE_F32">;
+defm V_CMP_NLG_F32 : VOPC_32 <0x0000000a, "V_CMP_NLG_F32">;
+defm V_CMP_NGT_F32 : VOPC_32 <0x0000000b, "V_CMP_NGT_F32">;
+defm V_CMP_NLE_F32 : VOPC_32 <0x0000000c, "V_CMP_NLE_F32">;
+defm V_CMP_NEQ_F32 : VOPC_32 <0x0000000d, "V_CMP_NEQ_F32", f32, COND_NE>;
+defm V_CMP_NLT_F32 : VOPC_32 <0x0000000e, "V_CMP_NLT_F32">;
+defm V_CMP_TRU_F32 : VOPC_32 <0x0000000f, "V_CMP_TRU_F32">;
+
+let hasSideEffects = 1, Defs = [EXEC] in {
+
+defm V_CMPX_F_F32 : VOPC_32 <0x00000010, "V_CMPX_F_F32">;
+defm V_CMPX_LT_F32 : VOPC_32 <0x00000011, "V_CMPX_LT_F32">;
+defm V_CMPX_EQ_F32 : VOPC_32 <0x00000012, "V_CMPX_EQ_F32">;
+defm V_CMPX_LE_F32 : VOPC_32 <0x00000013, "V_CMPX_LE_F32">;
+defm V_CMPX_GT_F32 : VOPC_32 <0x00000014, "V_CMPX_GT_F32">;
+defm V_CMPX_LG_F32 : VOPC_32 <0x00000015, "V_CMPX_LG_F32">;
+defm V_CMPX_GE_F32 : VOPC_32 <0x00000016, "V_CMPX_GE_F32">;
+defm V_CMPX_O_F32 : VOPC_32 <0x00000017, "V_CMPX_O_F32">;
+defm V_CMPX_U_F32 : VOPC_32 <0x00000018, "V_CMPX_U_F32">;
+defm V_CMPX_NGE_F32 : VOPC_32 <0x00000019, "V_CMPX_NGE_F32">;
+defm V_CMPX_NLG_F32 : VOPC_32 <0x0000001a, "V_CMPX_NLG_F32">;
+defm V_CMPX_NGT_F32 : VOPC_32 <0x0000001b, "V_CMPX_NGT_F32">;
+defm V_CMPX_NLE_F32 : VOPC_32 <0x0000001c, "V_CMPX_NLE_F32">;
+defm V_CMPX_NEQ_F32 : VOPC_32 <0x0000001d, "V_CMPX_NEQ_F32">;
+defm V_CMPX_NLT_F32 : VOPC_32 <0x0000001e, "V_CMPX_NLT_F32">;
+defm V_CMPX_TRU_F32 : VOPC_32 <0x0000001f, "V_CMPX_TRU_F32">;
+
+} // End hasSideEffects = 1, Defs = [EXEC]
+
+defm V_CMP_F_F64 : VOPC_64 <0x00000020, "V_CMP_F_F64">;
+defm V_CMP_LT_F64 : VOPC_64 <0x00000021, "V_CMP_LT_F64">;
+defm V_CMP_EQ_F64 : VOPC_64 <0x00000022, "V_CMP_EQ_F64">;
+defm V_CMP_LE_F64 : VOPC_64 <0x00000023, "V_CMP_LE_F64">;
+defm V_CMP_GT_F64 : VOPC_64 <0x00000024, "V_CMP_GT_F64">;
+defm V_CMP_LG_F64 : VOPC_64 <0x00000025, "V_CMP_LG_F64">;
+defm V_CMP_GE_F64 : VOPC_64 <0x00000026, "V_CMP_GE_F64">;
+defm V_CMP_O_F64 : VOPC_64 <0x00000027, "V_CMP_O_F64">;
+defm V_CMP_U_F64 : VOPC_64 <0x00000028, "V_CMP_U_F64">;
+defm V_CMP_NGE_F64 : VOPC_64 <0x00000029, "V_CMP_NGE_F64">;
+defm V_CMP_NLG_F64 : VOPC_64 <0x0000002a, "V_CMP_NLG_F64">;
+defm V_CMP_NGT_F64 : VOPC_64 <0x0000002b, "V_CMP_NGT_F64">;
+defm V_CMP_NLE_F64 : VOPC_64 <0x0000002c, "V_CMP_NLE_F64">;
+defm V_CMP_NEQ_F64 : VOPC_64 <0x0000002d, "V_CMP_NEQ_F64">;
+defm V_CMP_NLT_F64 : VOPC_64 <0x0000002e, "V_CMP_NLT_F64">;
+defm V_CMP_TRU_F64 : VOPC_64 <0x0000002f, "V_CMP_TRU_F64">;
+
+let hasSideEffects = 1, Defs = [EXEC] in {
+
+defm V_CMPX_F_F64 : VOPC_64 <0x00000030, "V_CMPX_F_F64">;
+defm V_CMPX_LT_F64 : VOPC_64 <0x00000031, "V_CMPX_LT_F64">;
+defm V_CMPX_EQ_F64 : VOPC_64 <0x00000032, "V_CMPX_EQ_F64">;
+defm V_CMPX_LE_F64 : VOPC_64 <0x00000033, "V_CMPX_LE_F64">;
+defm V_CMPX_GT_F64 : VOPC_64 <0x00000034, "V_CMPX_GT_F64">;
+defm V_CMPX_LG_F64 : VOPC_64 <0x00000035, "V_CMPX_LG_F64">;
+defm V_CMPX_GE_F64 : VOPC_64 <0x00000036, "V_CMPX_GE_F64">;
+defm V_CMPX_O_F64 : VOPC_64 <0x00000037, "V_CMPX_O_F64">;
+defm V_CMPX_U_F64 : VOPC_64 <0x00000038, "V_CMPX_U_F64">;
+defm V_CMPX_NGE_F64 : VOPC_64 <0x00000039, "V_CMPX_NGE_F64">;
+defm V_CMPX_NLG_F64 : VOPC_64 <0x0000003a, "V_CMPX_NLG_F64">;
+defm V_CMPX_NGT_F64 : VOPC_64 <0x0000003b, "V_CMPX_NGT_F64">;
+defm V_CMPX_NLE_F64 : VOPC_64 <0x0000003c, "V_CMPX_NLE_F64">;
+defm V_CMPX_NEQ_F64 : VOPC_64 <0x0000003d, "V_CMPX_NEQ_F64">;
+defm V_CMPX_NLT_F64 : VOPC_64 <0x0000003e, "V_CMPX_NLT_F64">;
+defm V_CMPX_TRU_F64 : VOPC_64 <0x0000003f, "V_CMPX_TRU_F64">;
+
+} // End hasSideEffects = 1, Defs = [EXEC]
+
+defm V_CMPS_F_F32 : VOPC_32 <0x00000040, "V_CMPS_F_F32">;
+defm V_CMPS_LT_F32 : VOPC_32 <0x00000041, "V_CMPS_LT_F32">;
+defm V_CMPS_EQ_F32 : VOPC_32 <0x00000042, "V_CMPS_EQ_F32">;
+defm V_CMPS_LE_F32 : VOPC_32 <0x00000043, "V_CMPS_LE_F32">;
+defm V_CMPS_GT_F32 : VOPC_32 <0x00000044, "V_CMPS_GT_F32">;
+defm V_CMPS_LG_F32 : VOPC_32 <0x00000045, "V_CMPS_LG_F32">;
+defm V_CMPS_GE_F32 : VOPC_32 <0x00000046, "V_CMPS_GE_F32">;
+defm V_CMPS_O_F32 : VOPC_32 <0x00000047, "V_CMPS_O_F32">;
+defm V_CMPS_U_F32 : VOPC_32 <0x00000048, "V_CMPS_U_F32">;
+defm V_CMPS_NGE_F32 : VOPC_32 <0x00000049, "V_CMPS_NGE_F32">;
+defm V_CMPS_NLG_F32 : VOPC_32 <0x0000004a, "V_CMPS_NLG_F32">;
+defm V_CMPS_NGT_F32 : VOPC_32 <0x0000004b, "V_CMPS_NGT_F32">;
+defm V_CMPS_NLE_F32 : VOPC_32 <0x0000004c, "V_CMPS_NLE_F32">;
+defm V_CMPS_NEQ_F32 : VOPC_32 <0x0000004d, "V_CMPS_NEQ_F32">;
+defm V_CMPS_NLT_F32 : VOPC_32 <0x0000004e, "V_CMPS_NLT_F32">;
+defm V_CMPS_TRU_F32 : VOPC_32 <0x0000004f, "V_CMPS_TRU_F32">;
+
+let hasSideEffects = 1, Defs = [EXEC] in {
+
+defm V_CMPSX_F_F32 : VOPC_32 <0x00000050, "V_CMPSX_F_F32">;
+defm V_CMPSX_LT_F32 : VOPC_32 <0x00000051, "V_CMPSX_LT_F32">;
+defm V_CMPSX_EQ_F32 : VOPC_32 <0x00000052, "V_CMPSX_EQ_F32">;
+defm V_CMPSX_LE_F32 : VOPC_32 <0x00000053, "V_CMPSX_LE_F32">;
+defm V_CMPSX_GT_F32 : VOPC_32 <0x00000054, "V_CMPSX_GT_F32">;
+defm V_CMPSX_LG_F32 : VOPC_32 <0x00000055, "V_CMPSX_LG_F32">;
+defm V_CMPSX_GE_F32 : VOPC_32 <0x00000056, "V_CMPSX_GE_F32">;
+defm V_CMPSX_O_F32 : VOPC_32 <0x00000057, "V_CMPSX_O_F32">;
+defm V_CMPSX_U_F32 : VOPC_32 <0x00000058, "V_CMPSX_U_F32">;
+defm V_CMPSX_NGE_F32 : VOPC_32 <0x00000059, "V_CMPSX_NGE_F32">;
+defm V_CMPSX_NLG_F32 : VOPC_32 <0x0000005a, "V_CMPSX_NLG_F32">;
+defm V_CMPSX_NGT_F32 : VOPC_32 <0x0000005b, "V_CMPSX_NGT_F32">;
+defm V_CMPSX_NLE_F32 : VOPC_32 <0x0000005c, "V_CMPSX_NLE_F32">;
+defm V_CMPSX_NEQ_F32 : VOPC_32 <0x0000005d, "V_CMPSX_NEQ_F32">;
+defm V_CMPSX_NLT_F32 : VOPC_32 <0x0000005e, "V_CMPSX_NLT_F32">;
+defm V_CMPSX_TRU_F32 : VOPC_32 <0x0000005f, "V_CMPSX_TRU_F32">;
+
+} // End hasSideEffects = 1, Defs = [EXEC]
+
+defm V_CMPS_F_F64 : VOPC_64 <0x00000060, "V_CMPS_F_F64">;
+defm V_CMPS_LT_F64 : VOPC_64 <0x00000061, "V_CMPS_LT_F64">;
+defm V_CMPS_EQ_F64 : VOPC_64 <0x00000062, "V_CMPS_EQ_F64">;
+defm V_CMPS_LE_F64 : VOPC_64 <0x00000063, "V_CMPS_LE_F64">;
+defm V_CMPS_GT_F64 : VOPC_64 <0x00000064, "V_CMPS_GT_F64">;
+defm V_CMPS_LG_F64 : VOPC_64 <0x00000065, "V_CMPS_LG_F64">;
+defm V_CMPS_GE_F64 : VOPC_64 <0x00000066, "V_CMPS_GE_F64">;
+defm V_CMPS_O_F64 : VOPC_64 <0x00000067, "V_CMPS_O_F64">;
+defm V_CMPS_U_F64 : VOPC_64 <0x00000068, "V_CMPS_U_F64">;
+defm V_CMPS_NGE_F64 : VOPC_64 <0x00000069, "V_CMPS_NGE_F64">;
+defm V_CMPS_NLG_F64 : VOPC_64 <0x0000006a, "V_CMPS_NLG_F64">;
+defm V_CMPS_NGT_F64 : VOPC_64 <0x0000006b, "V_CMPS_NGT_F64">;
+defm V_CMPS_NLE_F64 : VOPC_64 <0x0000006c, "V_CMPS_NLE_F64">;
+defm V_CMPS_NEQ_F64 : VOPC_64 <0x0000006d, "V_CMPS_NEQ_F64">;
+defm V_CMPS_NLT_F64 : VOPC_64 <0x0000006e, "V_CMPS_NLT_F64">;
+defm V_CMPS_TRU_F64 : VOPC_64 <0x0000006f, "V_CMPS_TRU_F64">;
+
+let hasSideEffects = 1, Defs = [EXEC] in {
+
+defm V_CMPSX_F_F64 : VOPC_64 <0x00000070, "V_CMPSX_F_F64">;
+defm V_CMPSX_LT_F64 : VOPC_64 <0x00000071, "V_CMPSX_LT_F64">;
+defm V_CMPSX_EQ_F64 : VOPC_64 <0x00000072, "V_CMPSX_EQ_F64">;
+defm V_CMPSX_LE_F64 : VOPC_64 <0x00000073, "V_CMPSX_LE_F64">;
+defm V_CMPSX_GT_F64 : VOPC_64 <0x00000074, "V_CMPSX_GT_F64">;
+defm V_CMPSX_LG_F64 : VOPC_64 <0x00000075, "V_CMPSX_LG_F64">;
+defm V_CMPSX_GE_F64 : VOPC_64 <0x00000076, "V_CMPSX_GE_F64">;
+defm V_CMPSX_O_F64 : VOPC_64 <0x00000077, "V_CMPSX_O_F64">;
+defm V_CMPSX_U_F64 : VOPC_64 <0x00000078, "V_CMPSX_U_F64">;
+defm V_CMPSX_NGE_F64 : VOPC_64 <0x00000079, "V_CMPSX_NGE_F64">;
+defm V_CMPSX_NLG_F64 : VOPC_64 <0x0000007a, "V_CMPSX_NLG_F64">;
+defm V_CMPSX_NGT_F64 : VOPC_64 <0x0000007b, "V_CMPSX_NGT_F64">;
+defm V_CMPSX_NLE_F64 : VOPC_64 <0x0000007c, "V_CMPSX_NLE_F64">;
+defm V_CMPSX_NEQ_F64 : VOPC_64 <0x0000007d, "V_CMPSX_NEQ_F64">;
+defm V_CMPSX_NLT_F64 : VOPC_64 <0x0000007e, "V_CMPSX_NLT_F64">;
+defm V_CMPSX_TRU_F64 : VOPC_64 <0x0000007f, "V_CMPSX_TRU_F64">;
+
+} // End hasSideEffects = 1, Defs = [EXEC]
+
+defm V_CMP_F_I32 : VOPC_32 <0x00000080, "V_CMP_F_I32">;
+defm V_CMP_LT_I32 : VOPC_32 <0x00000081, "V_CMP_LT_I32", i32, COND_LT>;
+defm V_CMP_EQ_I32 : VOPC_32 <0x00000082, "V_CMP_EQ_I32", i32, COND_EQ>;
+defm V_CMP_LE_I32 : VOPC_32 <0x00000083, "V_CMP_LE_I32", i32, COND_LE>;
+defm V_CMP_GT_I32 : VOPC_32 <0x00000084, "V_CMP_GT_I32", i32, COND_GT>;
+defm V_CMP_NE_I32 : VOPC_32 <0x00000085, "V_CMP_NE_I32", i32, COND_NE>;
+defm V_CMP_GE_I32 : VOPC_32 <0x00000086, "V_CMP_GE_I32", i32, COND_GE>;
+defm V_CMP_T_I32 : VOPC_32 <0x00000087, "V_CMP_T_I32">;
+
+let hasSideEffects = 1, Defs = [EXEC] in {
+
+defm V_CMPX_F_I32 : VOPC_32 <0x00000090, "V_CMPX_F_I32">;
+defm V_CMPX_LT_I32 : VOPC_32 <0x00000091, "V_CMPX_LT_I32">;
+defm V_CMPX_EQ_I32 : VOPC_32 <0x00000092, "V_CMPX_EQ_I32">;
+defm V_CMPX_LE_I32 : VOPC_32 <0x00000093, "V_CMPX_LE_I32">;
+defm V_CMPX_GT_I32 : VOPC_32 <0x00000094, "V_CMPX_GT_I32">;
+defm V_CMPX_NE_I32 : VOPC_32 <0x00000095, "V_CMPX_NE_I32">;
+defm V_CMPX_GE_I32 : VOPC_32 <0x00000096, "V_CMPX_GE_I32">;
+defm V_CMPX_T_I32 : VOPC_32 <0x00000097, "V_CMPX_T_I32">;
+
+} // End hasSideEffects = 1, Defs = [EXEC]
+
+defm V_CMP_F_I64 : VOPC_64 <0x000000a0, "V_CMP_F_I64">;
+defm V_CMP_LT_I64 : VOPC_64 <0x000000a1, "V_CMP_LT_I64">;
+defm V_CMP_EQ_I64 : VOPC_64 <0x000000a2, "V_CMP_EQ_I64">;
+defm V_CMP_LE_I64 : VOPC_64 <0x000000a3, "V_CMP_LE_I64">;
+defm V_CMP_GT_I64 : VOPC_64 <0x000000a4, "V_CMP_GT_I64">;
+defm V_CMP_NE_I64 : VOPC_64 <0x000000a5, "V_CMP_NE_I64">;
+defm V_CMP_GE_I64 : VOPC_64 <0x000000a6, "V_CMP_GE_I64">;
+defm V_CMP_T_I64 : VOPC_64 <0x000000a7, "V_CMP_T_I64">;
+
+let hasSideEffects = 1, Defs = [EXEC] in {
+
+defm V_CMPX_F_I64 : VOPC_64 <0x000000b0, "V_CMPX_F_I64">;
+defm V_CMPX_LT_I64 : VOPC_64 <0x000000b1, "V_CMPX_LT_I64">;
+defm V_CMPX_EQ_I64 : VOPC_64 <0x000000b2, "V_CMPX_EQ_I64">;
+defm V_CMPX_LE_I64 : VOPC_64 <0x000000b3, "V_CMPX_LE_I64">;
+defm V_CMPX_GT_I64 : VOPC_64 <0x000000b4, "V_CMPX_GT_I64">;
+defm V_CMPX_NE_I64 : VOPC_64 <0x000000b5, "V_CMPX_NE_I64">;
+defm V_CMPX_GE_I64 : VOPC_64 <0x000000b6, "V_CMPX_GE_I64">;
+defm V_CMPX_T_I64 : VOPC_64 <0x000000b7, "V_CMPX_T_I64">;
+
+} // End hasSideEffects = 1, Defs = [EXEC]
+
+defm V_CMP_F_U32 : VOPC_32 <0x000000c0, "V_CMP_F_U32">;
+defm V_CMP_LT_U32 : VOPC_32 <0x000000c1, "V_CMP_LT_U32">;
+defm V_CMP_EQ_U32 : VOPC_32 <0x000000c2, "V_CMP_EQ_U32">;
+defm V_CMP_LE_U32 : VOPC_32 <0x000000c3, "V_CMP_LE_U32">;
+defm V_CMP_GT_U32 : VOPC_32 <0x000000c4, "V_CMP_GT_U32">;
+defm V_CMP_NE_U32 : VOPC_32 <0x000000c5, "V_CMP_NE_U32">;
+defm V_CMP_GE_U32 : VOPC_32 <0x000000c6, "V_CMP_GE_U32">;
+defm V_CMP_T_U32 : VOPC_32 <0x000000c7, "V_CMP_T_U32">;
+
+let hasSideEffects = 1, Defs = [EXEC] in {
+
+defm V_CMPX_F_U32 : VOPC_32 <0x000000d0, "V_CMPX_F_U32">;
+defm V_CMPX_LT_U32 : VOPC_32 <0x000000d1, "V_CMPX_LT_U32">;
+defm V_CMPX_EQ_U32 : VOPC_32 <0x000000d2, "V_CMPX_EQ_U32">;
+defm V_CMPX_LE_U32 : VOPC_32 <0x000000d3, "V_CMPX_LE_U32">;
+defm V_CMPX_GT_U32 : VOPC_32 <0x000000d4, "V_CMPX_GT_U32">;
+defm V_CMPX_NE_U32 : VOPC_32 <0x000000d5, "V_CMPX_NE_U32">;
+defm V_CMPX_GE_U32 : VOPC_32 <0x000000d6, "V_CMPX_GE_U32">;
+defm V_CMPX_T_U32 : VOPC_32 <0x000000d7, "V_CMPX_T_U32">;
+
+} // End hasSideEffects = 1, Defs = [EXEC]
+
+defm V_CMP_F_U64 : VOPC_64 <0x000000e0, "V_CMP_F_U64">;
+defm V_CMP_LT_U64 : VOPC_64 <0x000000e1, "V_CMP_LT_U64">;
+defm V_CMP_EQ_U64 : VOPC_64 <0x000000e2, "V_CMP_EQ_U64">;
+defm V_CMP_LE_U64 : VOPC_64 <0x000000e3, "V_CMP_LE_U64">;
+defm V_CMP_GT_U64 : VOPC_64 <0x000000e4, "V_CMP_GT_U64">;
+defm V_CMP_NE_U64 : VOPC_64 <0x000000e5, "V_CMP_NE_U64">;
+defm V_CMP_GE_U64 : VOPC_64 <0x000000e6, "V_CMP_GE_U64">;
+defm V_CMP_T_U64 : VOPC_64 <0x000000e7, "V_CMP_T_U64">;
+
+let hasSideEffects = 1, Defs = [EXEC] in {
+
+defm V_CMPX_F_U64 : VOPC_64 <0x000000f0, "V_CMPX_F_U64">;
+defm V_CMPX_LT_U64 : VOPC_64 <0x000000f1, "V_CMPX_LT_U64">;
+defm V_CMPX_EQ_U64 : VOPC_64 <0x000000f2, "V_CMPX_EQ_U64">;
+defm V_CMPX_LE_U64 : VOPC_64 <0x000000f3, "V_CMPX_LE_U64">;
+defm V_CMPX_GT_U64 : VOPC_64 <0x000000f4, "V_CMPX_GT_U64">;
+defm V_CMPX_NE_U64 : VOPC_64 <0x000000f5, "V_CMPX_NE_U64">;
+defm V_CMPX_GE_U64 : VOPC_64 <0x000000f6, "V_CMPX_GE_U64">;
+defm V_CMPX_T_U64 : VOPC_64 <0x000000f7, "V_CMPX_T_U64">;
+
+} // End hasSideEffects = 1, Defs = [EXEC]
+
+defm V_CMP_CLASS_F32 : VOPC_32 <0x00000088, "V_CMP_CLASS_F32">;
+
+let hasSideEffects = 1, Defs = [EXEC] in {
+defm V_CMPX_CLASS_F32 : VOPC_32 <0x00000098, "V_CMPX_CLASS_F32">;
+} // End hasSideEffects = 1, Defs = [EXEC]
+
+defm V_CMP_CLASS_F64 : VOPC_64 <0x000000a8, "V_CMP_CLASS_F64">;
+
+let hasSideEffects = 1, Defs = [EXEC] in {
+defm V_CMPX_CLASS_F64 : VOPC_64 <0x000000b8, "V_CMPX_CLASS_F64">;
+} // End hasSideEffects = 1, Defs = [EXEC]
+
+} // End isCompare = 1
+
+//def BUFFER_LOAD_FORMAT_X : MUBUF_ <0x00000000, "BUFFER_LOAD_FORMAT_X", []>;
+//def BUFFER_LOAD_FORMAT_XY : MUBUF_ <0x00000001, "BUFFER_LOAD_FORMAT_XY", []>;
+//def BUFFER_LOAD_FORMAT_XYZ : MUBUF_ <0x00000002, "BUFFER_LOAD_FORMAT_XYZ", []>;
+def BUFFER_LOAD_FORMAT_XYZW : MUBUF_Load_Helper <0x00000003, "BUFFER_LOAD_FORMAT_XYZW", VReg_128>;
+//def BUFFER_STORE_FORMAT_X : MUBUF_ <0x00000004, "BUFFER_STORE_FORMAT_X", []>;
+//def BUFFER_STORE_FORMAT_XY : MUBUF_ <0x00000005, "BUFFER_STORE_FORMAT_XY", []>;
+//def BUFFER_STORE_FORMAT_XYZ : MUBUF_ <0x00000006, "BUFFER_STORE_FORMAT_XYZ", []>;
+//def BUFFER_STORE_FORMAT_XYZW : MUBUF_ <0x00000007, "BUFFER_STORE_FORMAT_XYZW", []>;
+//def BUFFER_LOAD_UBYTE : MUBUF_ <0x00000008, "BUFFER_LOAD_UBYTE", []>;
+//def BUFFER_LOAD_SBYTE : MUBUF_ <0x00000009, "BUFFER_LOAD_SBYTE", []>;
+//def BUFFER_LOAD_USHORT : MUBUF_ <0x0000000a, "BUFFER_LOAD_USHORT", []>;
+//def BUFFER_LOAD_SSHORT : MUBUF_ <0x0000000b, "BUFFER_LOAD_SSHORT", []>;
+def BUFFER_LOAD_DWORD : MUBUF_Load_Helper <0x0000000c, "BUFFER_LOAD_DWORD", VReg_32>;
+def BUFFER_LOAD_DWORDX2 : MUBUF_Load_Helper <0x0000000d, "BUFFER_LOAD_DWORDX2", VReg_64>;
+def BUFFER_LOAD_DWORDX4 : MUBUF_Load_Helper <0x0000000e, "BUFFER_LOAD_DWORDX4", VReg_128>;
+//def BUFFER_STORE_BYTE : MUBUF_ <0x00000018, "BUFFER_STORE_BYTE", []>;
+//def BUFFER_STORE_SHORT : MUBUF_ <0x0000001a, "BUFFER_STORE_SHORT", []>;
+//def BUFFER_STORE_DWORD : MUBUF_ <0x0000001c, "BUFFER_STORE_DWORD", []>;
+//def BUFFER_STORE_DWORDX2 : MUBUF_DWORDX2 <0x0000001d, "BUFFER_STORE_DWORDX2", []>;
+//def BUFFER_STORE_DWORDX4 : MUBUF_DWORDX4 <0x0000001e, "BUFFER_STORE_DWORDX4", []>;
+//def BUFFER_ATOMIC_SWAP : MUBUF_ <0x00000030, "BUFFER_ATOMIC_SWAP", []>;
+//def BUFFER_ATOMIC_CMPSWAP : MUBUF_ <0x00000031, "BUFFER_ATOMIC_CMPSWAP", []>;
+//def BUFFER_ATOMIC_ADD : MUBUF_ <0x00000032, "BUFFER_ATOMIC_ADD", []>;
+//def BUFFER_ATOMIC_SUB : MUBUF_ <0x00000033, "BUFFER_ATOMIC_SUB", []>;
+//def BUFFER_ATOMIC_RSUB : MUBUF_ <0x00000034, "BUFFER_ATOMIC_RSUB", []>;
+//def BUFFER_ATOMIC_SMIN : MUBUF_ <0x00000035, "BUFFER_ATOMIC_SMIN", []>;
+//def BUFFER_ATOMIC_UMIN : MUBUF_ <0x00000036, "BUFFER_ATOMIC_UMIN", []>;
+//def BUFFER_ATOMIC_SMAX : MUBUF_ <0x00000037, "BUFFER_ATOMIC_SMAX", []>;
+//def BUFFER_ATOMIC_UMAX : MUBUF_ <0x00000038, "BUFFER_ATOMIC_UMAX", []>;
+//def BUFFER_ATOMIC_AND : MUBUF_ <0x00000039, "BUFFER_ATOMIC_AND", []>;
+//def BUFFER_ATOMIC_OR : MUBUF_ <0x0000003a, "BUFFER_ATOMIC_OR", []>;
+//def BUFFER_ATOMIC_XOR : MUBUF_ <0x0000003b, "BUFFER_ATOMIC_XOR", []>;
+//def BUFFER_ATOMIC_INC : MUBUF_ <0x0000003c, "BUFFER_ATOMIC_INC", []>;
+//def BUFFER_ATOMIC_DEC : MUBUF_ <0x0000003d, "BUFFER_ATOMIC_DEC", []>;
+//def BUFFER_ATOMIC_FCMPSWAP : MUBUF_ <0x0000003e, "BUFFER_ATOMIC_FCMPSWAP", []>;
+//def BUFFER_ATOMIC_FMIN : MUBUF_ <0x0000003f, "BUFFER_ATOMIC_FMIN", []>;
+//def BUFFER_ATOMIC_FMAX : MUBUF_ <0x00000040, "BUFFER_ATOMIC_FMAX", []>;
+//def BUFFER_ATOMIC_SWAP_X2 : MUBUF_X2 <0x00000050, "BUFFER_ATOMIC_SWAP_X2", []>;
+//def BUFFER_ATOMIC_CMPSWAP_X2 : MUBUF_X2 <0x00000051, "BUFFER_ATOMIC_CMPSWAP_X2", []>;
+//def BUFFER_ATOMIC_ADD_X2 : MUBUF_X2 <0x00000052, "BUFFER_ATOMIC_ADD_X2", []>;
+//def BUFFER_ATOMIC_SUB_X2 : MUBUF_X2 <0x00000053, "BUFFER_ATOMIC_SUB_X2", []>;
+//def BUFFER_ATOMIC_RSUB_X2 : MUBUF_X2 <0x00000054, "BUFFER_ATOMIC_RSUB_X2", []>;
+//def BUFFER_ATOMIC_SMIN_X2 : MUBUF_X2 <0x00000055, "BUFFER_ATOMIC_SMIN_X2", []>;
+//def BUFFER_ATOMIC_UMIN_X2 : MUBUF_X2 <0x00000056, "BUFFER_ATOMIC_UMIN_X2", []>;
+//def BUFFER_ATOMIC_SMAX_X2 : MUBUF_X2 <0x00000057, "BUFFER_ATOMIC_SMAX_X2", []>;
+//def BUFFER_ATOMIC_UMAX_X2 : MUBUF_X2 <0x00000058, "BUFFER_ATOMIC_UMAX_X2", []>;
+//def BUFFER_ATOMIC_AND_X2 : MUBUF_X2 <0x00000059, "BUFFER_ATOMIC_AND_X2", []>;
+//def BUFFER_ATOMIC_OR_X2 : MUBUF_X2 <0x0000005a, "BUFFER_ATOMIC_OR_X2", []>;
+//def BUFFER_ATOMIC_XOR_X2 : MUBUF_X2 <0x0000005b, "BUFFER_ATOMIC_XOR_X2", []>;
+//def BUFFER_ATOMIC_INC_X2 : MUBUF_X2 <0x0000005c, "BUFFER_ATOMIC_INC_X2", []>;
+//def BUFFER_ATOMIC_DEC_X2 : MUBUF_X2 <0x0000005d, "BUFFER_ATOMIC_DEC_X2", []>;
+//def BUFFER_ATOMIC_FCMPSWAP_X2 : MUBUF_X2 <0x0000005e, "BUFFER_ATOMIC_FCMPSWAP_X2", []>;
+//def BUFFER_ATOMIC_FMIN_X2 : MUBUF_X2 <0x0000005f, "BUFFER_ATOMIC_FMIN_X2", []>;
+//def BUFFER_ATOMIC_FMAX_X2 : MUBUF_X2 <0x00000060, "BUFFER_ATOMIC_FMAX_X2", []>;
+//def BUFFER_WBINVL1_SC : MUBUF_WBINVL1 <0x00000070, "BUFFER_WBINVL1_SC", []>;
+//def BUFFER_WBINVL1 : MUBUF_WBINVL1 <0x00000071, "BUFFER_WBINVL1", []>;
+//def TBUFFER_LOAD_FORMAT_X : MTBUF_ <0x00000000, "TBUFFER_LOAD_FORMAT_X", []>;
+//def TBUFFER_LOAD_FORMAT_XY : MTBUF_ <0x00000001, "TBUFFER_LOAD_FORMAT_XY", []>;
+//def TBUFFER_LOAD_FORMAT_XYZ : MTBUF_ <0x00000002, "TBUFFER_LOAD_FORMAT_XYZ", []>;
+def TBUFFER_LOAD_FORMAT_XYZW : MTBUF_Load_Helper <0x00000003, "TBUFFER_LOAD_FORMAT_XYZW", VReg_128>;
+//def TBUFFER_STORE_FORMAT_X : MTBUF_ <0x00000004, "TBUFFER_STORE_FORMAT_X", []>;
+//def TBUFFER_STORE_FORMAT_XY : MTBUF_ <0x00000005, "TBUFFER_STORE_FORMAT_XY", []>;
+//def TBUFFER_STORE_FORMAT_XYZ : MTBUF_ <0x00000006, "TBUFFER_STORE_FORMAT_XYZ", []>;
+//def TBUFFER_STORE_FORMAT_XYZW : MTBUF_ <0x00000007, "TBUFFER_STORE_FORMAT_XYZW", []>;
+
+let mayLoad = 1 in {
+
+defm S_LOAD_DWORD : SMRD_Helper <0x00, "S_LOAD_DWORD", SReg_64, SReg_32>;
+defm S_LOAD_DWORDX2 : SMRD_Helper <0x01, "S_LOAD_DWORDX2", SReg_64, SReg_64>;
+defm S_LOAD_DWORDX4 : SMRD_Helper <0x02, "S_LOAD_DWORDX4", SReg_64, SReg_128>;
+defm S_LOAD_DWORDX8 : SMRD_Helper <0x03, "S_LOAD_DWORDX8", SReg_64, SReg_256>;
+defm S_LOAD_DWORDX16 : SMRD_Helper <0x04, "S_LOAD_DWORDX16", SReg_64, SReg_512>;
+
+defm S_BUFFER_LOAD_DWORD : SMRD_Helper <
+  0x08, "S_BUFFER_LOAD_DWORD", SReg_128, SReg_32
+>;
+
+defm S_BUFFER_LOAD_DWORDX2 : SMRD_Helper <
+  0x09, "S_BUFFER_LOAD_DWORDX2", SReg_128, SReg_64
+>;
+
+defm S_BUFFER_LOAD_DWORDX4 : SMRD_Helper <
+  0x0a, "S_BUFFER_LOAD_DWORDX4", SReg_128, SReg_128
+>;
+
+defm S_BUFFER_LOAD_DWORDX8 : SMRD_Helper <
+  0x0b, "S_BUFFER_LOAD_DWORDX8", SReg_128, SReg_256
+>;
+
+defm S_BUFFER_LOAD_DWORDX16 : SMRD_Helper <
+  0x0c, "S_BUFFER_LOAD_DWORDX16", SReg_128, SReg_512
+>;
+
+} // mayLoad = 1
+
+//def S_MEMTIME : SMRD_ <0x0000001e, "S_MEMTIME", []>;
+//def S_DCACHE_INV : SMRD_ <0x0000001f, "S_DCACHE_INV", []>;
+//def IMAGE_LOAD : MIMG_NoPattern_ <"IMAGE_LOAD", 0x00000000>;
+//def IMAGE_LOAD_MIP : MIMG_NoPattern_ <"IMAGE_LOAD_MIP", 0x00000001>;
+//def IMAGE_LOAD_PCK : MIMG_NoPattern_ <"IMAGE_LOAD_PCK", 0x00000002>;
+//def IMAGE_LOAD_PCK_SGN : MIMG_NoPattern_ <"IMAGE_LOAD_PCK_SGN", 0x00000003>;
+//def IMAGE_LOAD_MIP_PCK : MIMG_NoPattern_ <"IMAGE_LOAD_MIP_PCK", 0x00000004>;
+//def IMAGE_LOAD_MIP_PCK_SGN : MIMG_NoPattern_ <"IMAGE_LOAD_MIP_PCK_SGN", 0x00000005>;
+//def IMAGE_STORE : MIMG_NoPattern_ <"IMAGE_STORE", 0x00000008>;
+//def IMAGE_STORE_MIP : MIMG_NoPattern_ <"IMAGE_STORE_MIP", 0x00000009>;
+//def IMAGE_STORE_PCK : MIMG_NoPattern_ <"IMAGE_STORE_PCK", 0x0000000a>;
+//def IMAGE_STORE_MIP_PCK : MIMG_NoPattern_ <"IMAGE_STORE_MIP_PCK", 0x0000000b>;
+//def IMAGE_GET_RESINFO : MIMG_NoPattern_ <"IMAGE_GET_RESINFO", 0x0000000e>;
+//def IMAGE_ATOMIC_SWAP : MIMG_NoPattern_ <"IMAGE_ATOMIC_SWAP", 0x0000000f>;
+//def IMAGE_ATOMIC_CMPSWAP : MIMG_NoPattern_ <"IMAGE_ATOMIC_CMPSWAP", 0x00000010>;
+//def IMAGE_ATOMIC_ADD : MIMG_NoPattern_ <"IMAGE_ATOMIC_ADD", 0x00000011>;
+//def IMAGE_ATOMIC_SUB : MIMG_NoPattern_ <"IMAGE_ATOMIC_SUB", 0x00000012>;
+//def IMAGE_ATOMIC_RSUB : MIMG_NoPattern_ <"IMAGE_ATOMIC_RSUB", 0x00000013>;
+//def IMAGE_ATOMIC_SMIN : MIMG_NoPattern_ <"IMAGE_ATOMIC_SMIN", 0x00000014>;
+//def IMAGE_ATOMIC_UMIN : MIMG_NoPattern_ <"IMAGE_ATOMIC_UMIN", 0x00000015>;
+//def IMAGE_ATOMIC_SMAX : MIMG_NoPattern_ <"IMAGE_ATOMIC_SMAX", 0x00000016>;
+//def IMAGE_ATOMIC_UMAX : MIMG_NoPattern_ <"IMAGE_ATOMIC_UMAX", 0x00000017>;
+//def IMAGE_ATOMIC_AND : MIMG_NoPattern_ <"IMAGE_ATOMIC_AND", 0x00000018>;
+//def IMAGE_ATOMIC_OR : MIMG_NoPattern_ <"IMAGE_ATOMIC_OR", 0x00000019>;
+//def IMAGE_ATOMIC_XOR : MIMG_NoPattern_ <"IMAGE_ATOMIC_XOR", 0x0000001a>;
+//def IMAGE_ATOMIC_INC : MIMG_NoPattern_ <"IMAGE_ATOMIC_INC", 0x0000001b>;
+//def IMAGE_ATOMIC_DEC : MIMG_NoPattern_ <"IMAGE_ATOMIC_DEC", 0x0000001c>;
+//def IMAGE_ATOMIC_FCMPSWAP : MIMG_NoPattern_ <"IMAGE_ATOMIC_FCMPSWAP", 0x0000001d>;
+//def IMAGE_ATOMIC_FMIN : MIMG_NoPattern_ <"IMAGE_ATOMIC_FMIN", 0x0000001e>;
+//def IMAGE_ATOMIC_FMAX : MIMG_NoPattern_ <"IMAGE_ATOMIC_FMAX", 0x0000001f>;
+def IMAGE_SAMPLE : MIMG_Load_Helper <0x00000020, "IMAGE_SAMPLE">; 
+//def IMAGE_SAMPLE_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_CL", 0x00000021>;
+def IMAGE_SAMPLE_D : MIMG_Load_Helper <0x00000022, "IMAGE_SAMPLE_D">;
+//def IMAGE_SAMPLE_D_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_D_CL", 0x00000023>;
+def IMAGE_SAMPLE_L : MIMG_Load_Helper <0x00000024, "IMAGE_SAMPLE_L">;
+def IMAGE_SAMPLE_B : MIMG_Load_Helper <0x00000025, "IMAGE_SAMPLE_B">;
+//def IMAGE_SAMPLE_B_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_B_CL", 0x00000026>;
+//def IMAGE_SAMPLE_LZ : MIMG_NoPattern_ <"IMAGE_SAMPLE_LZ", 0x00000027>;
+def IMAGE_SAMPLE_C : MIMG_Load_Helper <0x00000028, "IMAGE_SAMPLE_C">;
+//def IMAGE_SAMPLE_C_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CL", 0x00000029>;
+//def IMAGE_SAMPLE_C_D : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D", 0x0000002a>;
+//def IMAGE_SAMPLE_C_D_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D_CL", 0x0000002b>;
+def IMAGE_SAMPLE_C_L : MIMG_Load_Helper <0x0000002c, "IMAGE_SAMPLE_C_L">;
+def IMAGE_SAMPLE_C_B : MIMG_Load_Helper <0x0000002d, "IMAGE_SAMPLE_C_B">;
+//def IMAGE_SAMPLE_C_B_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_B_CL", 0x0000002e>;
+//def IMAGE_SAMPLE_C_LZ : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_LZ", 0x0000002f>;
+//def IMAGE_SAMPLE_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_O", 0x00000030>;
+//def IMAGE_SAMPLE_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_CL_O", 0x00000031>;
+//def IMAGE_SAMPLE_D_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_D_O", 0x00000032>;
+//def IMAGE_SAMPLE_D_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_D_CL_O", 0x00000033>;
+//def IMAGE_SAMPLE_L_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_L_O", 0x00000034>;
+//def IMAGE_SAMPLE_B_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_B_O", 0x00000035>;
+//def IMAGE_SAMPLE_B_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_B_CL_O", 0x00000036>;
+//def IMAGE_SAMPLE_LZ_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_LZ_O", 0x00000037>;
+//def IMAGE_SAMPLE_C_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_O", 0x00000038>;
+//def IMAGE_SAMPLE_C_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CL_O", 0x00000039>;
+//def IMAGE_SAMPLE_C_D_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D_O", 0x0000003a>;
+//def IMAGE_SAMPLE_C_D_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_D_CL_O", 0x0000003b>;
+//def IMAGE_SAMPLE_C_L_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_L_O", 0x0000003c>;
+//def IMAGE_SAMPLE_C_B_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_B_O", 0x0000003d>;
+//def IMAGE_SAMPLE_C_B_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_B_CL_O", 0x0000003e>;
+//def IMAGE_SAMPLE_C_LZ_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_LZ_O", 0x0000003f>;
+//def IMAGE_GATHER4 : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4", 0x00000040>;
+//def IMAGE_GATHER4_CL : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_CL", 0x00000041>;
+//def IMAGE_GATHER4_L : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_L", 0x00000044>;
+//def IMAGE_GATHER4_B : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_B", 0x00000045>;
+//def IMAGE_GATHER4_B_CL : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_B_CL", 0x00000046>;
+//def IMAGE_GATHER4_LZ : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_LZ", 0x00000047>;
+//def IMAGE_GATHER4_C : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C", 0x00000048>;
+//def IMAGE_GATHER4_C_CL : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_CL", 0x00000049>;
+//def IMAGE_GATHER4_C_L : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_L", 0x0000004c>;
+//def IMAGE_GATHER4_C_B : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_B", 0x0000004d>;
+//def IMAGE_GATHER4_C_B_CL : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_B_CL", 0x0000004e>;
+//def IMAGE_GATHER4_C_LZ : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_LZ", 0x0000004f>;
+//def IMAGE_GATHER4_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_O", 0x00000050>;
+//def IMAGE_GATHER4_CL_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_CL_O", 0x00000051>;
+//def IMAGE_GATHER4_L_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_L_O", 0x00000054>;
+//def IMAGE_GATHER4_B_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_B_O", 0x00000055>;
+//def IMAGE_GATHER4_B_CL_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_B_CL_O", 0x00000056>;
+//def IMAGE_GATHER4_LZ_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_LZ_O", 0x00000057>;
+//def IMAGE_GATHER4_C_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_O", 0x00000058>;
+//def IMAGE_GATHER4_C_CL_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_CL_O", 0x00000059>;
+//def IMAGE_GATHER4_C_L_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_L_O", 0x0000005c>;
+//def IMAGE_GATHER4_C_B_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_B_O", 0x0000005d>;
+//def IMAGE_GATHER4_C_B_CL_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_B_CL_O", 0x0000005e>;
+//def IMAGE_GATHER4_C_LZ_O : MIMG_NoPattern_GATHER4 <"IMAGE_GATHER4_C_LZ_O", 0x0000005f>;
+//def IMAGE_GET_LOD : MIMG_NoPattern_ <"IMAGE_GET_LOD", 0x00000060>;
+//def IMAGE_SAMPLE_CD : MIMG_NoPattern_ <"IMAGE_SAMPLE_CD", 0x00000068>;
+//def IMAGE_SAMPLE_CD_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_CD_CL", 0x00000069>;
+//def IMAGE_SAMPLE_C_CD : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CD", 0x0000006a>;
+//def IMAGE_SAMPLE_C_CD_CL : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CD_CL", 0x0000006b>;
+//def IMAGE_SAMPLE_CD_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_CD_O", 0x0000006c>;
+//def IMAGE_SAMPLE_CD_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_CD_CL_O", 0x0000006d>;
+//def IMAGE_SAMPLE_C_CD_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CD_O", 0x0000006e>;
+//def IMAGE_SAMPLE_C_CD_CL_O : MIMG_NoPattern_ <"IMAGE_SAMPLE_C_CD_CL_O", 0x0000006f>;
+//def IMAGE_RSRC256 : MIMG_NoPattern_RSRC256 <"IMAGE_RSRC256", 0x0000007e>;
+//def IMAGE_SAMPLER : MIMG_NoPattern_ <"IMAGE_SAMPLER", 0x0000007f>;
+//def V_NOP : VOP1_ <0x00000000, "V_NOP", []>;
+
+
+let neverHasSideEffects = 1, isMoveImm = 1 in {
+defm V_MOV_B32 : VOP1_32 <0x00000001, "V_MOV_B32", []>;
+} // End neverHasSideEffects = 1, isMoveImm = 1
+
+defm V_READFIRSTLANE_B32 : VOP1_32 <0x00000002, "V_READFIRSTLANE_B32", []>;
+//defm V_CVT_I32_F64 : VOP1_32 <0x00000003, "V_CVT_I32_F64", []>;
+//defm V_CVT_F64_I32 : VOP1_64 <0x00000004, "V_CVT_F64_I32", []>;
+defm V_CVT_F32_I32 : VOP1_32 <0x00000005, "V_CVT_F32_I32",
+  [(set VReg_32:$dst, (sint_to_fp VSrc_32:$src0))]
+>;
+//defm V_CVT_F32_U32 : VOP1_32 <0x00000006, "V_CVT_F32_U32", []>;
+//defm V_CVT_U32_F32 : VOP1_32 <0x00000007, "V_CVT_U32_F32", []>;
+defm V_CVT_I32_F32 : VOP1_32 <0x00000008, "V_CVT_I32_F32",
+  [(set (i32 VReg_32:$dst), (fp_to_sint VSrc_32:$src0))]
+>;
+defm V_MOV_FED_B32 : VOP1_32 <0x00000009, "V_MOV_FED_B32", []>;
+////def V_CVT_F16_F32 : VOP1_F16 <0x0000000a, "V_CVT_F16_F32", []>;
+//defm V_CVT_F32_F16 : VOP1_32 <0x0000000b, "V_CVT_F32_F16", []>;
+//defm V_CVT_RPI_I32_F32 : VOP1_32 <0x0000000c, "V_CVT_RPI_I32_F32", []>;
+//defm V_CVT_FLR_I32_F32 : VOP1_32 <0x0000000d, "V_CVT_FLR_I32_F32", []>;
+//defm V_CVT_OFF_F32_I4 : VOP1_32 <0x0000000e, "V_CVT_OFF_F32_I4", []>;
+//defm V_CVT_F32_F64 : VOP1_32 <0x0000000f, "V_CVT_F32_F64", []>;
+//defm V_CVT_F64_F32 : VOP1_64 <0x00000010, "V_CVT_F64_F32", []>;
+//defm V_CVT_F32_UBYTE0 : VOP1_32 <0x00000011, "V_CVT_F32_UBYTE0", []>;
+//defm V_CVT_F32_UBYTE1 : VOP1_32 <0x00000012, "V_CVT_F32_UBYTE1", []>;
+//defm V_CVT_F32_UBYTE2 : VOP1_32 <0x00000013, "V_CVT_F32_UBYTE2", []>;
+//defm V_CVT_F32_UBYTE3 : VOP1_32 <0x00000014, "V_CVT_F32_UBYTE3", []>;
+//defm V_CVT_U32_F64 : VOP1_32 <0x00000015, "V_CVT_U32_F64", []>;
+//defm V_CVT_F64_U32 : VOP1_64 <0x00000016, "V_CVT_F64_U32", []>;
+defm V_FRACT_F32 : VOP1_32 <0x00000020, "V_FRACT_F32",
+  [(set VReg_32:$dst, (AMDGPUfract VSrc_32:$src0))]
+>;
+defm V_TRUNC_F32 : VOP1_32 <0x00000021, "V_TRUNC_F32", []>;
+defm V_CEIL_F32 : VOP1_32 <0x00000022, "V_CEIL_F32",
+  [(set VReg_32:$dst, (fceil VSrc_32:$src0))]
+>;
+defm V_RNDNE_F32 : VOP1_32 <0x00000023, "V_RNDNE_F32",
+  [(set VReg_32:$dst, (frint VSrc_32:$src0))]
+>;
+defm V_FLOOR_F32 : VOP1_32 <0x00000024, "V_FLOOR_F32",
+  [(set VReg_32:$dst, (ffloor VSrc_32:$src0))]
+>;
+defm V_EXP_F32 : VOP1_32 <0x00000025, "V_EXP_F32",
+  [(set VReg_32:$dst, (fexp2 VSrc_32:$src0))]
+>;
+defm V_LOG_CLAMP_F32 : VOP1_32 <0x00000026, "V_LOG_CLAMP_F32", []>;
+defm V_LOG_F32 : VOP1_32 <0x00000027, "V_LOG_F32",
+  [(set VReg_32:$dst, (flog2 VSrc_32:$src0))]
+>;
+defm V_RCP_CLAMP_F32 : VOP1_32 <0x00000028, "V_RCP_CLAMP_F32", []>;
+defm V_RCP_LEGACY_F32 : VOP1_32 <0x00000029, "V_RCP_LEGACY_F32", []>;
+defm V_RCP_F32 : VOP1_32 <0x0000002a, "V_RCP_F32",
+  [(set VReg_32:$dst, (fdiv FP_ONE, VSrc_32:$src0))]
+>;
+defm V_RCP_IFLAG_F32 : VOP1_32 <0x0000002b, "V_RCP_IFLAG_F32", []>;
+defm V_RSQ_CLAMP_F32 : VOP1_32 <0x0000002c, "V_RSQ_CLAMP_F32", []>;
+defm V_RSQ_LEGACY_F32 : VOP1_32 <
+  0x0000002d, "V_RSQ_LEGACY_F32",
+  [(set VReg_32:$dst, (int_AMDGPU_rsq VSrc_32:$src0))]
+>;
+defm V_RSQ_F32 : VOP1_32 <0x0000002e, "V_RSQ_F32", []>;
+defm V_RCP_F64 : VOP1_64 <0x0000002f, "V_RCP_F64", []>;
+defm V_RCP_CLAMP_F64 : VOP1_64 <0x00000030, "V_RCP_CLAMP_F64", []>;
+defm V_RSQ_F64 : VOP1_64 <0x00000031, "V_RSQ_F64", []>;
+defm V_RSQ_CLAMP_F64 : VOP1_64 <0x00000032, "V_RSQ_CLAMP_F64", []>;
+defm V_SQRT_F32 : VOP1_32 <0x00000033, "V_SQRT_F32", []>;
+defm V_SQRT_F64 : VOP1_64 <0x00000034, "V_SQRT_F64", []>;
+defm V_SIN_F32 : VOP1_32 <0x00000035, "V_SIN_F32", []>;
+defm V_COS_F32 : VOP1_32 <0x00000036, "V_COS_F32", []>;
+defm V_NOT_B32 : VOP1_32 <0x00000037, "V_NOT_B32", []>;
+defm V_BFREV_B32 : VOP1_32 <0x00000038, "V_BFREV_B32", []>;
+defm V_FFBH_U32 : VOP1_32 <0x00000039, "V_FFBH_U32", []>;
+defm V_FFBL_B32 : VOP1_32 <0x0000003a, "V_FFBL_B32", []>;
+defm V_FFBH_I32 : VOP1_32 <0x0000003b, "V_FFBH_I32", []>;
+//defm V_FREXP_EXP_I32_F64 : VOP1_32 <0x0000003c, "V_FREXP_EXP_I32_F64", []>;
+defm V_FREXP_MANT_F64 : VOP1_64 <0x0000003d, "V_FREXP_MANT_F64", []>;
+defm V_FRACT_F64 : VOP1_64 <0x0000003e, "V_FRACT_F64", []>;
+//defm V_FREXP_EXP_I32_F32 : VOP1_32 <0x0000003f, "V_FREXP_EXP_I32_F32", []>;
+defm V_FREXP_MANT_F32 : VOP1_32 <0x00000040, "V_FREXP_MANT_F32", []>;
+//def V_CLREXCP : VOP1_ <0x00000041, "V_CLREXCP", []>;
+defm V_MOVRELD_B32 : VOP1_32 <0x00000042, "V_MOVRELD_B32", []>;
+defm V_MOVRELS_B32 : VOP1_32 <0x00000043, "V_MOVRELS_B32", []>;
+defm V_MOVRELSD_B32 : VOP1_32 <0x00000044, "V_MOVRELSD_B32", []>;
+
+def V_INTERP_P1_F32 : VINTRP <
+  0x00000000,
+  (outs VReg_32:$dst),
+  (ins VReg_32:$i, i32imm:$attr_chan, i32imm:$attr, M0Reg:$m0),
+  "V_INTERP_P1_F32 $dst, $i, $attr_chan, $attr, [$m0]",
+  []> {
+  let DisableEncoding = "$m0";
+}
+
+def V_INTERP_P2_F32 : VINTRP <
+  0x00000001,
+  (outs VReg_32:$dst),
+  (ins VReg_32:$src0, VReg_32:$j, i32imm:$attr_chan, i32imm:$attr, M0Reg:$m0),
+  "V_INTERP_P2_F32 $dst, [$src0], $j, $attr_chan, $attr, [$m0]",
+  []> {
+
+  let Constraints = "$src0 = $dst";
+  let DisableEncoding = "$src0,$m0";
+
+}
+
+def V_INTERP_MOV_F32 : VINTRP <
+  0x00000002,
+  (outs VReg_32:$dst),
+  (ins InterpSlot:$src0, i32imm:$attr_chan, i32imm:$attr, M0Reg:$m0),
+  "V_INTERP_MOV_F32 $dst, $src0, $attr_chan, $attr, [$m0]",
+  []> {
+  let DisableEncoding = "$m0";
+}
+
+//def S_NOP : SOPP_ <0x00000000, "S_NOP", []>;
+
+let isTerminator = 1 in {
+
+def S_ENDPGM : SOPP <0x00000001, (ins), "S_ENDPGM",
+  [(IL_retflag)]> {
+  let SIMM16 = 0;
+  let isBarrier = 1;
+  let hasCtrlDep = 1;
+}
+
+let isBranch = 1 in {
+def S_BRANCH : SOPP <
+  0x00000002, (ins brtarget:$target), "S_BRANCH $target",
+  [(br bb:$target)]> {
+  let isBarrier = 1;
+}
+
+let DisableEncoding = "$scc" in {
+def S_CBRANCH_SCC0 : SOPP <
+  0x00000004, (ins brtarget:$target, SCCReg:$scc),
+  "S_CBRANCH_SCC0 $target", []
+>;
+def S_CBRANCH_SCC1 : SOPP <
+  0x00000005, (ins brtarget:$target, SCCReg:$scc),
+  "S_CBRANCH_SCC1 $target",
+  []
+>;
+} // End DisableEncoding = "$scc"
+
+def S_CBRANCH_VCCZ : SOPP <
+  0x00000006, (ins brtarget:$target, VCCReg:$vcc),
+  "S_CBRANCH_VCCZ $target",
+  []
+>;
+def S_CBRANCH_VCCNZ : SOPP <
+  0x00000007, (ins brtarget:$target, VCCReg:$vcc),
+  "S_CBRANCH_VCCNZ $target",
+  []
+>;
+
+let DisableEncoding = "$exec" in {
+def S_CBRANCH_EXECZ : SOPP <
+  0x00000008, (ins brtarget:$target, EXECReg:$exec),
+  "S_CBRANCH_EXECZ $target",
+  []
+>;
+def S_CBRANCH_EXECNZ : SOPP <
+  0x00000009, (ins brtarget:$target, EXECReg:$exec),
+  "S_CBRANCH_EXECNZ $target",
+  []
+>;
+} // End DisableEncoding = "$exec"
+
+
+} // End isBranch = 1
+} // End isTerminator = 1
+
+//def S_BARRIER : SOPP_ <0x0000000a, "S_BARRIER", []>;
+let hasSideEffects = 1 in {
+def S_WAITCNT : SOPP <0x0000000c, (ins i32imm:$simm16), "S_WAITCNT $simm16",
+  []
+>;
+} // End hasSideEffects
+//def S_SETHALT : SOPP_ <0x0000000d, "S_SETHALT", []>;
+//def S_SLEEP : SOPP_ <0x0000000e, "S_SLEEP", []>;
+//def S_SETPRIO : SOPP_ <0x0000000f, "S_SETPRIO", []>;
+//def S_SENDMSG : SOPP_ <0x00000010, "S_SENDMSG", []>;
+//def S_SENDMSGHALT : SOPP_ <0x00000011, "S_SENDMSGHALT", []>;
+//def S_TRAP : SOPP_ <0x00000012, "S_TRAP", []>;
+//def S_ICACHE_INV : SOPP_ <0x00000013, "S_ICACHE_INV", []>;
+//def S_INCPERFLEVEL : SOPP_ <0x00000014, "S_INCPERFLEVEL", []>;
+//def S_DECPERFLEVEL : SOPP_ <0x00000015, "S_DECPERFLEVEL", []>;
+//def S_TTRACEDATA : SOPP_ <0x00000016, "S_TTRACEDATA", []>;
+
+def V_CNDMASK_B32_e32 : VOP2 <0x00000000, (outs VReg_32:$dst),
+  (ins VSrc_32:$src0, VReg_32:$src1, VCCReg:$vcc),
+  "V_CNDMASK_B32_e32 $dst, $src0, $src1, [$vcc]",
+  []
+>{
+  let DisableEncoding = "$vcc";
+}
+
+def V_CNDMASK_B32_e64 : VOP3 <0x00000100, (outs VReg_32:$dst),
+  (ins VSrc_32:$src0, VSrc_32:$src1, SSrc_64:$src2,
+   InstFlag:$abs, InstFlag:$clamp, InstFlag:$omod, InstFlag:$neg),
+  "V_CNDMASK_B32_e64 $dst, $src0, $src1, $src2, $abs, $clamp, $omod, $neg",
+  [(set (i32 VReg_32:$dst), (select (i1 SSrc_64:$src2),
+   VSrc_32:$src1, VSrc_32:$src0))]
+>;
+
+//f32 pattern for V_CNDMASK_B32_e64
+def : Pat <
+  (f32 (select (i1 SSrc_64:$src2), VSrc_32:$src1, VSrc_32:$src0)),
+  (V_CNDMASK_B32_e64 VSrc_32:$src0, VSrc_32:$src1, SSrc_64:$src2)
+>;
+
+defm V_READLANE_B32 : VOP2_32 <0x00000001, "V_READLANE_B32", []>;
+defm V_WRITELANE_B32 : VOP2_32 <0x00000002, "V_WRITELANE_B32", []>;
+
+let isCommutable = 1 in {
+defm V_ADD_F32 : VOP2_32 <0x00000003, "V_ADD_F32",
+  [(set VReg_32:$dst, (fadd VSrc_32:$src0, VReg_32:$src1))]
+>;
+
+defm V_SUB_F32 : VOP2_32 <0x00000004, "V_SUB_F32",
+  [(set VReg_32:$dst, (fsub VSrc_32:$src0, VReg_32:$src1))]
+>;
+defm V_SUBREV_F32 : VOP2_32 <0x00000005, "V_SUBREV_F32", [], "V_SUB_F32">;
+} // End isCommutable = 1
+
+defm V_MAC_LEGACY_F32 : VOP2_32 <0x00000006, "V_MAC_LEGACY_F32", []>;
+
+let isCommutable = 1 in {
+
+defm V_MUL_LEGACY_F32 : VOP2_32 <
+  0x00000007, "V_MUL_LEGACY_F32",
+  [(set VReg_32:$dst, (int_AMDGPU_mul VSrc_32:$src0, VReg_32:$src1))]
+>;
+
+defm V_MUL_F32 : VOP2_32 <0x00000008, "V_MUL_F32",
+  [(set VReg_32:$dst, (fmul VSrc_32:$src0, VReg_32:$src1))]
+>;
+
+} // End isCommutable = 1
+
+//defm V_MUL_I32_I24 : VOP2_32 <0x00000009, "V_MUL_I32_I24", []>;
+//defm V_MUL_HI_I32_I24 : VOP2_32 <0x0000000a, "V_MUL_HI_I32_I24", []>;
+//defm V_MUL_U32_U24 : VOP2_32 <0x0000000b, "V_MUL_U32_U24", []>;
+//defm V_MUL_HI_U32_U24 : VOP2_32 <0x0000000c, "V_MUL_HI_U32_U24", []>;
+
+let isCommutable = 1 in {
+
+defm V_MIN_LEGACY_F32 : VOP2_32 <0x0000000d, "V_MIN_LEGACY_F32",
+  [(set VReg_32:$dst, (AMDGPUfmin VSrc_32:$src0, VReg_32:$src1))]
+>;
+
+defm V_MAX_LEGACY_F32 : VOP2_32 <0x0000000e, "V_MAX_LEGACY_F32",
+  [(set VReg_32:$dst, (AMDGPUfmax VSrc_32:$src0, VReg_32:$src1))]
+>;
+
+defm V_MIN_F32 : VOP2_32 <0x0000000f, "V_MIN_F32", []>;
+defm V_MAX_F32 : VOP2_32 <0x00000010, "V_MAX_F32", []>;
+defm V_MIN_I32 : VOP2_32 <0x00000011, "V_MIN_I32", []>;
+defm V_MAX_I32 : VOP2_32 <0x00000012, "V_MAX_I32", []>;
+defm V_MIN_U32 : VOP2_32 <0x00000013, "V_MIN_U32", []>;
+defm V_MAX_U32 : VOP2_32 <0x00000014, "V_MAX_U32", []>;
+
+defm V_LSHR_B32 : VOP2_32 <0x00000015, "V_LSHR_B32",
+  [(set VReg_32:$dst, (srl VSrc_32:$src0, (i32 VReg_32:$src1)))]
+>;
+defm V_LSHRREV_B32 : VOP2_32 <0x00000016, "V_LSHRREV_B32", [], "V_LSHR_B32">;
+
+defm V_ASHR_I32 : VOP2_32 <0x00000017, "V_ASHR_I32",
+  [(set VReg_32:$dst, (sra VSrc_32:$src0, (i32 VReg_32:$src1)))]
+>;
+defm V_ASHRREV_I32 : VOP2_32 <0x00000018, "V_ASHRREV_I32", [], "V_ASHR_I32">;
+
+defm V_LSHL_B32 : VOP2_32 <0x00000019, "V_LSHL_B32",
+  [(set VReg_32:$dst, (shl VSrc_32:$src0, (i32 VReg_32:$src1)))]
+>;
+defm V_LSHLREV_B32 : VOP2_32 <0x0000001a, "V_LSHLREV_B32", [], "V_LSHL_B32">;
+
+defm V_AND_B32 : VOP2_32 <0x0000001b, "V_AND_B32",
+  [(set VReg_32:$dst, (and VSrc_32:$src0, VReg_32:$src1))]
+>;
+defm V_OR_B32 : VOP2_32 <0x0000001c, "V_OR_B32",
+  [(set VReg_32:$dst, (or VSrc_32:$src0, VReg_32:$src1))]
+>;
+defm V_XOR_B32 : VOP2_32 <0x0000001d, "V_XOR_B32",
+  [(set VReg_32:$dst, (xor VSrc_32:$src0, VReg_32:$src1))]
+>;
+
+} // End isCommutable = 1
+
+defm V_BFM_B32 : VOP2_32 <0x0000001e, "V_BFM_B32", []>;
+defm V_MAC_F32 : VOP2_32 <0x0000001f, "V_MAC_F32", []>;
+defm V_MADMK_F32 : VOP2_32 <0x00000020, "V_MADMK_F32", []>;
+defm V_MADAK_F32 : VOP2_32 <0x00000021, "V_MADAK_F32", []>;
+//defm V_BCNT_U32_B32 : VOP2_32 <0x00000022, "V_BCNT_U32_B32", []>;
+//defm V_MBCNT_LO_U32_B32 : VOP2_32 <0x00000023, "V_MBCNT_LO_U32_B32", []>;
+//defm V_MBCNT_HI_U32_B32 : VOP2_32 <0x00000024, "V_MBCNT_HI_U32_B32", []>;
+
+let isCommutable = 1, Defs = [VCC] in { // Carry-out goes to VCC
+defm V_ADD_I32 : VOP2b_32 <0x00000025, "V_ADD_I32",
+  [(set VReg_32:$dst, (add (i32 VSrc_32:$src0), (i32 VReg_32:$src1)))]
+>;
+
+defm V_SUB_I32 : VOP2b_32 <0x00000026, "V_SUB_I32",
+  [(set VReg_32:$dst, (sub (i32 VSrc_32:$src0), (i32 VReg_32:$src1)))]
+>;
+defm V_SUBREV_I32 : VOP2b_32 <0x00000027, "V_SUBREV_I32", [], "V_SUB_I32">;
+
+let Uses = [VCC] in { // Carry-out comes from VCC
+defm V_ADDC_U32 : VOP2b_32 <0x00000028, "V_ADDC_U32", []>;
+defm V_SUBB_U32 : VOP2b_32 <0x00000029, "V_SUBB_U32", []>;
+defm V_SUBBREV_U32 : VOP2b_32 <0x0000002a, "V_SUBBREV_U32", [], "V_SUBB_U32">;
+} // End Uses = [VCC]
+} // End isCommutable = 1, Defs = [VCC]
+
+defm V_LDEXP_F32 : VOP2_32 <0x0000002b, "V_LDEXP_F32", []>;
+////def V_CVT_PKACCUM_U8_F32 : VOP2_U8 <0x0000002c, "V_CVT_PKACCUM_U8_F32", []>;
+////def V_CVT_PKNORM_I16_F32 : VOP2_I16 <0x0000002d, "V_CVT_PKNORM_I16_F32", []>;
+////def V_CVT_PKNORM_U16_F32 : VOP2_U16 <0x0000002e, "V_CVT_PKNORM_U16_F32", []>;
+defm V_CVT_PKRTZ_F16_F32 : VOP2_32 <0x0000002f, "V_CVT_PKRTZ_F16_F32",
+ [(set VReg_32:$dst, (int_SI_packf16 VSrc_32:$src0, VReg_32:$src1))]
+>;
+////def V_CVT_PK_U16_U32 : VOP2_U16 <0x00000030, "V_CVT_PK_U16_U32", []>;
+////def V_CVT_PK_I16_I32 : VOP2_I16 <0x00000031, "V_CVT_PK_I16_I32", []>;
+def S_CMP_EQ_I32 : SOPC_32 <0x00000000, "S_CMP_EQ_I32", []>;
+def S_CMP_LG_I32 : SOPC_32 <0x00000001, "S_CMP_LG_I32", []>;
+def S_CMP_GT_I32 : SOPC_32 <0x00000002, "S_CMP_GT_I32", []>;
+def S_CMP_GE_I32 : SOPC_32 <0x00000003, "S_CMP_GE_I32", []>;
+def S_CMP_LT_I32 : SOPC_32 <0x00000004, "S_CMP_LT_I32", []>;
+def S_CMP_LE_I32 : SOPC_32 <0x00000005, "S_CMP_LE_I32", []>;
+def S_CMP_EQ_U32 : SOPC_32 <0x00000006, "S_CMP_EQ_U32", []>;
+def S_CMP_LG_U32 : SOPC_32 <0x00000007, "S_CMP_LG_U32", []>;
+def S_CMP_GT_U32 : SOPC_32 <0x00000008, "S_CMP_GT_U32", []>;
+def S_CMP_GE_U32 : SOPC_32 <0x00000009, "S_CMP_GE_U32", []>;
+def S_CMP_LT_U32 : SOPC_32 <0x0000000a, "S_CMP_LT_U32", []>;
+def S_CMP_LE_U32 : SOPC_32 <0x0000000b, "S_CMP_LE_U32", []>;
+////def S_BITCMP0_B32 : SOPC_BITCMP0 <0x0000000c, "S_BITCMP0_B32", []>;
+////def S_BITCMP1_B32 : SOPC_BITCMP1 <0x0000000d, "S_BITCMP1_B32", []>;
+////def S_BITCMP0_B64 : SOPC_BITCMP0 <0x0000000e, "S_BITCMP0_B64", []>;
+////def S_BITCMP1_B64 : SOPC_BITCMP1 <0x0000000f, "S_BITCMP1_B64", []>;
+//def S_SETVSKIP : SOPC_ <0x00000010, "S_SETVSKIP", []>;
+
+let neverHasSideEffects = 1 in {
+
+def V_MAD_LEGACY_F32 : VOP3_32 <0x00000140, "V_MAD_LEGACY_F32", []>;
+def V_MAD_F32 : VOP3_32 <0x00000141, "V_MAD_F32", []>;
+//def V_MAD_I32_I24 : VOP3_32 <0x00000142, "V_MAD_I32_I24", []>;
+//def V_MAD_U32_U24 : VOP3_32 <0x00000143, "V_MAD_U32_U24", []>;
+
+} // End neverHasSideEffects
+def V_CUBEID_F32 : VOP3_32 <0x00000144, "V_CUBEID_F32", []>;
+def V_CUBESC_F32 : VOP3_32 <0x00000145, "V_CUBESC_F32", []>;
+def V_CUBETC_F32 : VOP3_32 <0x00000146, "V_CUBETC_F32", []>;
+def V_CUBEMA_F32 : VOP3_32 <0x00000147, "V_CUBEMA_F32", []>;
+def V_BFE_U32 : VOP3_32 <0x00000148, "V_BFE_U32", []>;
+def V_BFE_I32 : VOP3_32 <0x00000149, "V_BFE_I32", []>;
+def V_BFI_B32 : VOP3_32 <0x0000014a, "V_BFI_B32", []>;
+def V_FMA_F32 : VOP3_32 <0x0000014b, "V_FMA_F32", []>;
+def V_FMA_F64 : VOP3_64 <0x0000014c, "V_FMA_F64", []>;
+//def V_LERP_U8 : VOP3_U8 <0x0000014d, "V_LERP_U8", []>;
+def V_ALIGNBIT_B32 : VOP3_32 <0x0000014e, "V_ALIGNBIT_B32", []>;
+def V_ALIGNBYTE_B32 : VOP3_32 <0x0000014f, "V_ALIGNBYTE_B32", []>;
+def V_MULLIT_F32 : VOP3_32 <0x00000150, "V_MULLIT_F32", []>;
+////def V_MIN3_F32 : VOP3_MIN3 <0x00000151, "V_MIN3_F32", []>;
+////def V_MIN3_I32 : VOP3_MIN3 <0x00000152, "V_MIN3_I32", []>;
+////def V_MIN3_U32 : VOP3_MIN3 <0x00000153, "V_MIN3_U32", []>;
+////def V_MAX3_F32 : VOP3_MAX3 <0x00000154, "V_MAX3_F32", []>;
+////def V_MAX3_I32 : VOP3_MAX3 <0x00000155, "V_MAX3_I32", []>;
+////def V_MAX3_U32 : VOP3_MAX3 <0x00000156, "V_MAX3_U32", []>;
+////def V_MED3_F32 : VOP3_MED3 <0x00000157, "V_MED3_F32", []>;
+////def V_MED3_I32 : VOP3_MED3 <0x00000158, "V_MED3_I32", []>;
+////def V_MED3_U32 : VOP3_MED3 <0x00000159, "V_MED3_U32", []>;
+//def V_SAD_U8 : VOP3_U8 <0x0000015a, "V_SAD_U8", []>;
+//def V_SAD_HI_U8 : VOP3_U8 <0x0000015b, "V_SAD_HI_U8", []>;
+//def V_SAD_U16 : VOP3_U16 <0x0000015c, "V_SAD_U16", []>;
+def V_SAD_U32 : VOP3_32 <0x0000015d, "V_SAD_U32", []>;
+////def V_CVT_PK_U8_F32 : VOP3_U8 <0x0000015e, "V_CVT_PK_U8_F32", []>;
+def V_DIV_FIXUP_F32 : VOP3_32 <0x0000015f, "V_DIV_FIXUP_F32", []>;
+def V_DIV_FIXUP_F64 : VOP3_64 <0x00000160, "V_DIV_FIXUP_F64", []>;
+def V_LSHL_B64 : VOP3_64 <0x00000161, "V_LSHL_B64", []>;
+def V_LSHR_B64 : VOP3_64 <0x00000162, "V_LSHR_B64", []>;
+def V_ASHR_I64 : VOP3_64 <0x00000163, "V_ASHR_I64", []>;
+def V_ADD_F64 : VOP3_64 <0x00000164, "V_ADD_F64", []>;
+def V_MUL_F64 : VOP3_64 <0x00000165, "V_MUL_F64", []>;
+def V_MIN_F64 : VOP3_64 <0x00000166, "V_MIN_F64", []>;
+def V_MAX_F64 : VOP3_64 <0x00000167, "V_MAX_F64", []>;
+def V_LDEXP_F64 : VOP3_64 <0x00000168, "V_LDEXP_F64", []>;
+
+let isCommutable = 1 in {
+
+def V_MUL_LO_U32 : VOP3_32 <0x00000169, "V_MUL_LO_U32", []>;
+def V_MUL_HI_U32 : VOP3_32 <0x0000016a, "V_MUL_HI_U32", []>;
+def V_MUL_LO_I32 : VOP3_32 <0x0000016b, "V_MUL_LO_I32", []>;
+def V_MUL_HI_I32 : VOP3_32 <0x0000016c, "V_MUL_HI_I32", []>;
+
+} // isCommutable = 1
+
+def : Pat <
+  (mul VSrc_32:$src0, VReg_32:$src1),
+  (V_MUL_LO_I32 VSrc_32:$src0, VReg_32:$src1, (i32 0), 0, 0, 0, 0)
+>;
+
+def : Pat <
+  (mulhu VSrc_32:$src0, VReg_32:$src1),
+  (V_MUL_HI_U32 VSrc_32:$src0, VReg_32:$src1, (i32 0), 0, 0, 0, 0)
+>;
+
+def : Pat <
+  (mulhs VSrc_32:$src0, VReg_32:$src1),
+  (V_MUL_HI_I32 VSrc_32:$src0, VReg_32:$src1, (i32 0), 0, 0, 0, 0)
+>;
+
+def V_DIV_SCALE_F32 : VOP3_32 <0x0000016d, "V_DIV_SCALE_F32", []>;
+def V_DIV_SCALE_F64 : VOP3_64 <0x0000016e, "V_DIV_SCALE_F64", []>;
+def V_DIV_FMAS_F32 : VOP3_32 <0x0000016f, "V_DIV_FMAS_F32", []>;
+def V_DIV_FMAS_F64 : VOP3_64 <0x00000170, "V_DIV_FMAS_F64", []>;
+//def V_MSAD_U8 : VOP3_U8 <0x00000171, "V_MSAD_U8", []>;
+//def V_QSAD_U8 : VOP3_U8 <0x00000172, "V_QSAD_U8", []>;
+//def V_MQSAD_U8 : VOP3_U8 <0x00000173, "V_MQSAD_U8", []>;
+def V_TRIG_PREOP_F64 : VOP3_64 <0x00000174, "V_TRIG_PREOP_F64", []>;
+def S_ADD_U32 : SOP2_32 <0x00000000, "S_ADD_U32", []>;
+def S_SUB_U32 : SOP2_32 <0x00000001, "S_SUB_U32", []>;
+def S_ADD_I32 : SOP2_32 <0x00000002, "S_ADD_I32", []>;
+def S_SUB_I32 : SOP2_32 <0x00000003, "S_SUB_I32", []>;
+def S_ADDC_U32 : SOP2_32 <0x00000004, "S_ADDC_U32", []>;
+def S_SUBB_U32 : SOP2_32 <0x00000005, "S_SUBB_U32", []>;
+def S_MIN_I32 : SOP2_32 <0x00000006, "S_MIN_I32", []>;
+def S_MIN_U32 : SOP2_32 <0x00000007, "S_MIN_U32", []>;
+def S_MAX_I32 : SOP2_32 <0x00000008, "S_MAX_I32", []>;
+def S_MAX_U32 : SOP2_32 <0x00000009, "S_MAX_U32", []>;
+
+def S_CSELECT_B32 : SOP2 <
+  0x0000000a, (outs SReg_32:$dst),
+  (ins SReg_32:$src0, SReg_32:$src1, SCCReg:$scc), "S_CSELECT_B32",
+  [(set (i32 SReg_32:$dst), (select (i1 SCCReg:$scc),
+                                     SReg_32:$src0, SReg_32:$src1))]
+>;
+
+def S_CSELECT_B64 : SOP2_64 <0x0000000b, "S_CSELECT_B64", []>;
+
+// f32 pattern for S_CSELECT_B32
+def : Pat <
+  (f32 (select (i1 SCCReg:$scc), SReg_32:$src0, SReg_32:$src1)),
+  (S_CSELECT_B32 SReg_32:$src0, SReg_32:$src1, SCCReg:$scc)
+>;
+
+def S_AND_B32 : SOP2_32 <0x0000000e, "S_AND_B32", []>;
+
+def S_AND_B64 : SOP2_64 <0x0000000f, "S_AND_B64",
+  [(set SReg_64:$dst, (i64 (and SSrc_64:$src0, SSrc_64:$src1)))]
+>;
+
+def : Pat <
+  (i1 (and SSrc_64:$src0, SSrc_64:$src1)),
+  (S_AND_B64 SSrc_64:$src0, SSrc_64:$src1)
+>;
+
+def S_OR_B32 : SOP2_32 <0x00000010, "S_OR_B32", []>;
+def S_OR_B64 : SOP2_64 <0x00000011, "S_OR_B64", []>;
+def : Pat <
+  (i1 (or SSrc_64:$src0, SSrc_64:$src1)),
+  (S_OR_B64 SSrc_64:$src0, SSrc_64:$src1)
+>;
+def S_XOR_B32 : SOP2_32 <0x00000012, "S_XOR_B32", []>;
+def S_XOR_B64 : SOP2_64 <0x00000013, "S_XOR_B64", []>;
+def S_ANDN2_B32 : SOP2_32 <0x00000014, "S_ANDN2_B32", []>;
+def S_ANDN2_B64 : SOP2_64 <0x00000015, "S_ANDN2_B64", []>;
+def S_ORN2_B32 : SOP2_32 <0x00000016, "S_ORN2_B32", []>;
+def S_ORN2_B64 : SOP2_64 <0x00000017, "S_ORN2_B64", []>;
+def S_NAND_B32 : SOP2_32 <0x00000018, "S_NAND_B32", []>;
+def S_NAND_B64 : SOP2_64 <0x00000019, "S_NAND_B64", []>;
+def S_NOR_B32 : SOP2_32 <0x0000001a, "S_NOR_B32", []>;
+def S_NOR_B64 : SOP2_64 <0x0000001b, "S_NOR_B64", []>;
+def S_XNOR_B32 : SOP2_32 <0x0000001c, "S_XNOR_B32", []>;
+def S_XNOR_B64 : SOP2_64 <0x0000001d, "S_XNOR_B64", []>;
+def S_LSHL_B32 : SOP2_32 <0x0000001e, "S_LSHL_B32", []>;
+def S_LSHL_B64 : SOP2_64 <0x0000001f, "S_LSHL_B64", []>;
+def S_LSHR_B32 : SOP2_32 <0x00000020, "S_LSHR_B32", []>;
+def S_LSHR_B64 : SOP2_64 <0x00000021, "S_LSHR_B64", []>;
+def S_ASHR_I32 : SOP2_32 <0x00000022, "S_ASHR_I32", []>;
+def S_ASHR_I64 : SOP2_64 <0x00000023, "S_ASHR_I64", []>;
+def S_BFM_B32 : SOP2_32 <0x00000024, "S_BFM_B32", []>;
+def S_BFM_B64 : SOP2_64 <0x00000025, "S_BFM_B64", []>;
+def S_MUL_I32 : SOP2_32 <0x00000026, "S_MUL_I32", []>;
+def S_BFE_U32 : SOP2_32 <0x00000027, "S_BFE_U32", []>;
+def S_BFE_I32 : SOP2_32 <0x00000028, "S_BFE_I32", []>;
+def S_BFE_U64 : SOP2_64 <0x00000029, "S_BFE_U64", []>;
+def S_BFE_I64 : SOP2_64 <0x0000002a, "S_BFE_I64", []>;
+//def S_CBRANCH_G_FORK : SOP2_ <0x0000002b, "S_CBRANCH_G_FORK", []>;
+def S_ABSDIFF_I32 : SOP2_32 <0x0000002c, "S_ABSDIFF_I32", []>;
+
+let isCodeGenOnly = 1, isPseudo = 1 in {
+
+def LOAD_CONST : AMDGPUShaderInst <
+  (outs GPRF32:$dst),
+  (ins i32imm:$src),
+  "LOAD_CONST $dst, $src",
+  [(set GPRF32:$dst, (int_AMDGPU_load_const imm:$src))]
+>;
+
+// SI Psuedo instructions. These are used by the CFG structurizer pass
+// and should be lowered to ISA instructions prior to codegen.
+
+let mayLoad = 1, mayStore = 1, hasSideEffects = 1,
+    Uses = [EXEC], Defs = [EXEC] in {
+
+let isBranch = 1, isTerminator = 1 in {
+
+def SI_IF : InstSI <
+  (outs SReg_64:$dst),
+  (ins SReg_64:$vcc, brtarget:$target),
+  "SI_IF $dst, $vcc, $target",
+  [(set SReg_64:$dst, (int_SI_if SReg_64:$vcc, bb:$target))]
+>;
+
+def SI_ELSE : InstSI <
+  (outs SReg_64:$dst),
+  (ins SReg_64:$src, brtarget:$target),
+  "SI_ELSE $dst, $src, $target",
+  [(set SReg_64:$dst, (int_SI_else SReg_64:$src, bb:$target))]> {
+
+  let Constraints = "$src = $dst";
+}
+
+def SI_LOOP : InstSI <
+  (outs),
+  (ins SReg_64:$saved, brtarget:$target),
+  "SI_LOOP $saved, $target",
+  [(int_SI_loop SReg_64:$saved, bb:$target)]
+>;
+
+} // end isBranch = 1, isTerminator = 1
+
+def SI_BREAK : InstSI <
+  (outs SReg_64:$dst),
+  (ins SReg_64:$src),
+  "SI_ELSE $dst, $src",
+  [(set SReg_64:$dst, (int_SI_break SReg_64:$src))]
+>;
+
+def SI_IF_BREAK : InstSI <
+  (outs SReg_64:$dst),
+  (ins SReg_64:$vcc, SReg_64:$src),
+  "SI_IF_BREAK $dst, $vcc, $src",
+  [(set SReg_64:$dst, (int_SI_if_break SReg_64:$vcc, SReg_64:$src))]
+>;
+
+def SI_ELSE_BREAK : InstSI <
+  (outs SReg_64:$dst),
+  (ins SReg_64:$src0, SReg_64:$src1),
+  "SI_ELSE_BREAK $dst, $src0, $src1",
+  [(set SReg_64:$dst, (int_SI_else_break SReg_64:$src0, SReg_64:$src1))]
+>;
+
+def SI_END_CF : InstSI <
+  (outs),
+  (ins SReg_64:$saved),
+  "SI_END_CF $saved",
+  [(int_SI_end_cf SReg_64:$saved)]
+>;
+
+def SI_KILL : InstSI <
+  (outs),
+  (ins VReg_32:$src),
+  "SI_KIL $src",
+  [(int_AMDGPU_kill VReg_32:$src)]
+>;
+
+} // end mayLoad = 1, mayStore = 1, hasSideEffects = 1
+  // Uses = [EXEC], Defs = [EXEC]
+
+let Uses = [EXEC], Defs = [EXEC,VCC,M0] in {
+
+def SI_INDIRECT_SRC : InstSI <
+  (outs VReg_32:$dst, SReg_64:$temp),
+  (ins unknown:$src, VSrc_32:$idx, i32imm:$off),
+  "SI_INDIRECT_SRC $dst, $temp, $src, $idx, $off",
+  []
+>;
+
+class SI_INDIRECT_DST<RegisterClass rc> : InstSI <
+  (outs rc:$dst, SReg_64:$temp),
+  (ins unknown:$src, VSrc_32:$idx, i32imm:$off, VReg_32:$val),
+  "SI_INDIRECT_DST $dst, $temp, $src, $idx, $off, $val",
+  []
+> {
+  let Constraints = "$src = $dst";
+}
+
+def SI_INDIRECT_DST_V2 : SI_INDIRECT_DST<VReg_64>;
+def SI_INDIRECT_DST_V4 : SI_INDIRECT_DST<VReg_128>;
+def SI_INDIRECT_DST_V8 : SI_INDIRECT_DST<VReg_256>;
+def SI_INDIRECT_DST_V16 : SI_INDIRECT_DST<VReg_512>;
+
+} // Uses = [EXEC,VCC,M0], Defs = [EXEC,VCC,M0]
+
+} // end IsCodeGenOnly, isPseudo
+
+def : Pat<
+  (int_AMDGPU_cndlt VReg_32:$src0, VReg_32:$src1, VReg_32:$src2),
+  (V_CNDMASK_B32_e64 VReg_32:$src2, VReg_32:$src1, (V_CMP_GT_F32_e64 0, VReg_32:$src0))
+>;
+
+def : Pat <
+  (int_AMDGPU_kilp),
+  (SI_KILL (V_MOV_B32_e32 0xbf800000))
+>;
+
+/* int_SI_vs_load_input */
+def : Pat<
+  (int_SI_vs_load_input SReg_128:$tlst, IMM12bit:$attr_offset,
+                        VReg_32:$buf_idx_vgpr),
+  (BUFFER_LOAD_FORMAT_XYZW imm:$attr_offset, 0, 1, 0, 0, 0,
+                           VReg_32:$buf_idx_vgpr, SReg_128:$tlst,
+                           0, 0, 0)
+>;
+
+/* int_SI_export */
+def : Pat <
+  (int_SI_export imm:$en, imm:$vm, imm:$done, imm:$tgt, imm:$compr,
+                 VReg_32:$src0,VReg_32:$src1, VReg_32:$src2, VReg_32:$src3),
+  (EXP imm:$en, imm:$tgt, imm:$compr, imm:$done, imm:$vm,
+       VReg_32:$src0, VReg_32:$src1, VReg_32:$src2, VReg_32:$src3)
+>;
+
+
+/* int_SI_sample for simple 1D texture lookup */
+def : Pat <
+  (int_SI_sample imm:$writemask, VReg_32:$addr,
+                 SReg_256:$rsrc, SReg_128:$sampler, imm),
+  (IMAGE_SAMPLE imm:$writemask, 0, 0, 0, 0, 0, 0, 0, VReg_32:$addr,
+                SReg_256:$rsrc, SReg_128:$sampler)
+>;
+
+class SamplePattern<Intrinsic name, MIMG opcode, RegisterClass addr_class,
+                    ValueType addr_type> : Pat <
+    (name imm:$writemask, (addr_type addr_class:$addr),
+          SReg_256:$rsrc, SReg_128:$sampler, imm),
+    (opcode imm:$writemask, 0, 0, 0, 0, 0, 0, 0, addr_class:$addr,
+          SReg_256:$rsrc, SReg_128:$sampler)
+>;
+
+class SampleRectPattern<Intrinsic name, MIMG opcode, RegisterClass addr_class,
+                        ValueType addr_type> : Pat <
+    (name imm:$writemask, (addr_type addr_class:$addr),
+          SReg_256:$rsrc, SReg_128:$sampler, TEX_RECT),
+    (opcode imm:$writemask, 1, 0, 0, 0, 0, 0, 0, addr_class:$addr,
+          SReg_256:$rsrc, SReg_128:$sampler)
+>;
+
+class SampleArrayPattern<Intrinsic name, MIMG opcode, RegisterClass addr_class,
+                         ValueType addr_type> : Pat <
+    (name imm:$writemask, (addr_type addr_class:$addr),
+          SReg_256:$rsrc, SReg_128:$sampler, TEX_ARRAY),
+    (opcode imm:$writemask, 0, 0, 1, 0, 0, 0, 0, addr_class:$addr,
+          SReg_256:$rsrc, SReg_128:$sampler)
+>;
+
+class SampleShadowPattern<Intrinsic name, MIMG opcode,
+                          RegisterClass addr_class, ValueType addr_type> : Pat <
+    (name imm:$writemask, (addr_type addr_class:$addr),
+          SReg_256:$rsrc, SReg_128:$sampler, TEX_SHADOW),
+    (opcode imm:$writemask, 0, 0, 0, 0, 0, 0, 0, addr_class:$addr,
+          SReg_256:$rsrc, SReg_128:$sampler)
+>;
+
+class SampleShadowArrayPattern<Intrinsic name, MIMG opcode,
+                               RegisterClass addr_class, ValueType addr_type> : Pat <
+    (name imm:$writemask, (addr_type addr_class:$addr),
+          SReg_256:$rsrc, SReg_128:$sampler, TEX_SHADOW_ARRAY),
+    (opcode imm:$writemask, 0, 0, 1, 0, 0, 0, 0, addr_class:$addr,
+          SReg_256:$rsrc, SReg_128:$sampler)
+>;
+
+/* int_SI_sample* for texture lookups consuming more address parameters */
+multiclass SamplePatterns<RegisterClass addr_class, ValueType addr_type> {
+  def : SamplePattern <int_SI_sample, IMAGE_SAMPLE, addr_class, addr_type>;
+  def : SampleRectPattern <int_SI_sample, IMAGE_SAMPLE, addr_class, addr_type>;
+  def : SampleArrayPattern <int_SI_sample, IMAGE_SAMPLE, addr_class, addr_type>;
+  def : SampleShadowPattern <int_SI_sample, IMAGE_SAMPLE_C, addr_class, addr_type>;
+  def : SampleShadowArrayPattern <int_SI_sample, IMAGE_SAMPLE_C, addr_class, addr_type>;
+
+  def : SamplePattern <int_SI_samplel, IMAGE_SAMPLE_L, addr_class, addr_type>;
+  def : SampleArrayPattern <int_SI_samplel, IMAGE_SAMPLE_L, addr_class, addr_type>;
+  def : SampleShadowPattern <int_SI_samplel, IMAGE_SAMPLE_C_L, addr_class, addr_type>;
+  def : SampleShadowArrayPattern <int_SI_samplel, IMAGE_SAMPLE_C_L, addr_class, addr_type>;
+
+  def : SamplePattern <int_SI_sampleb, IMAGE_SAMPLE_B, addr_class, addr_type>;
+  def : SampleArrayPattern <int_SI_sampleb, IMAGE_SAMPLE_B, addr_class, addr_type>;
+  def : SampleShadowPattern <int_SI_sampleb, IMAGE_SAMPLE_C_B, addr_class, addr_type>;
+  def : SampleShadowArrayPattern <int_SI_sampleb, IMAGE_SAMPLE_C_B, addr_class, addr_type>;
+}
+
+defm : SamplePatterns<VReg_64, v2i32>;
+defm : SamplePatterns<VReg_128, v4i32>;
+defm : SamplePatterns<VReg_256, v8i32>;
+defm : SamplePatterns<VReg_512, v16i32>;
+
+/********** ============================================ **********/
+/********** Extraction, Insertion, Building and Casting  **********/
+/********** ============================================ **********/
+
+foreach Index = 0-2 in {
+  def Extract_Element_v2i32_#Index : Extract_Element <
+    i32, v2i32, VReg_64, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+  def Insert_Element_v2i32_#Index : Insert_Element <
+    i32, v2i32, VReg_32, VReg_64, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+
+  def Extract_Element_v2f32_#Index : Extract_Element <
+    f32, v2f32, VReg_64, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+  def Insert_Element_v2f32_#Index : Insert_Element <
+    f32, v2f32, VReg_32, VReg_64, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+}
+
+foreach Index = 0-3 in {
+  def Extract_Element_v4i32_#Index : Extract_Element <
+    i32, v4i32, VReg_128, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+  def Insert_Element_v4i32_#Index : Insert_Element <
+    i32, v4i32, VReg_32, VReg_128, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+
+  def Extract_Element_v4f32_#Index : Extract_Element <
+    f32, v4f32, VReg_128, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+  def Insert_Element_v4f32_#Index : Insert_Element <
+    f32, v4f32, VReg_32, VReg_128, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+}
+
+foreach Index = 0-7 in {
+  def Extract_Element_v8i32_#Index : Extract_Element <
+    i32, v8i32, VReg_256, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+  def Insert_Element_v8i32_#Index : Insert_Element <
+    i32, v8i32, VReg_32, VReg_256, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+
+  def Extract_Element_v8f32_#Index : Extract_Element <
+    f32, v8f32, VReg_256, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+  def Insert_Element_v8f32_#Index : Insert_Element <
+    f32, v8f32, VReg_32, VReg_256, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+}
+
+foreach Index = 0-15 in {
+  def Extract_Element_v16i32_#Index : Extract_Element <
+    i32, v16i32, VReg_512, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+  def Insert_Element_v16i32_#Index : Insert_Element <
+    i32, v16i32, VReg_32, VReg_512, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+
+  def Extract_Element_v16f32_#Index : Extract_Element <
+    f32, v16f32, VReg_512, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+  def Insert_Element_v16f32_#Index : Insert_Element <
+    f32, v16f32, VReg_32, VReg_512, Index, !cast<SubRegIndex>(sub#Index)
+  >;
+}
+
+def : Vector1_Build <v1i32, VReg_32, i32, VReg_32>;
+def : Vector2_Build <v2i32, VReg_64, i32, VReg_32>;
+def : Vector2_Build <v2f32, VReg_64, f32, VReg_32>;
+def : Vector4_Build <v4i32, VReg_128, i32, VReg_32>;
+def : Vector4_Build <v4f32, VReg_128, f32, VReg_32>;
+def : Vector8_Build <v8i32, VReg_256, i32, VReg_32>;
+def : Vector8_Build <v8f32, VReg_256, f32, VReg_32>;
+def : Vector16_Build <v16i32, VReg_512, i32, VReg_32>;
+def : Vector16_Build <v16f32, VReg_512, f32, VReg_32>;
+
+def : BitConvert <i32, f32, SReg_32>;
+def : BitConvert <i32, f32, VReg_32>;
+
+def : BitConvert <f32, i32, SReg_32>;
+def : BitConvert <f32, i32, VReg_32>;
+
+/********** =================== **********/
+/********** Src & Dst modifiers **********/
+/********** =================== **********/
+
+def : Pat <
+  (int_AMDIL_clamp VReg_32:$src, (f32 FP_ZERO), (f32 FP_ONE)),
+  (V_ADD_F32_e64 VReg_32:$src, (i32 0 /* SRC1 */),
+   0 /* ABS */, 1 /* CLAMP */, 0 /* OMOD */, 0 /* NEG */)
+>;
+
+def : Pat <
+  (fabs VReg_32:$src),
+  (V_ADD_F32_e64 VReg_32:$src, (i32 0 /* SRC1 */),
+   1 /* ABS */, 0 /* CLAMP */, 0 /* OMOD */, 0 /* NEG */)
+>;
+
+def : Pat <
+  (fneg VReg_32:$src),
+  (V_ADD_F32_e64 VReg_32:$src, (i32 0 /* SRC1 */),
+   0 /* ABS */, 0 /* CLAMP */, 0 /* OMOD */, 1 /* NEG */)
+>;
+
+/********** ================== **********/
+/********** Immediate Patterns **********/
+/********** ================== **********/
+
+def : Pat <
+  (i32 imm:$imm),
+  (V_MOV_B32_e32 imm:$imm)
+>;
+
+def : Pat <
+  (f32 fpimm:$imm),
+  (V_MOV_B32_e32 fpimm:$imm)
+>;
+
+def : Pat <
+  (i1 imm:$imm),
+  (S_MOV_B64 imm:$imm)
+>;
+
+def : Pat <
+  (i64 InlineImm<i64>:$imm),
+  (S_MOV_B64 InlineImm<i64>:$imm)
+>;
+
+// i64 immediates aren't supported in hardware, split it into two 32bit values
+def : Pat <
+  (i64 imm:$imm),
+  (INSERT_SUBREG (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
+    (S_MOV_B32 (i32 (LO32 imm:$imm))), sub0),
+    (S_MOV_B32 (i32 (HI32 imm:$imm))), sub1)
+>;
+
+/********** ===================== **********/
+/********** Interpolation Paterns **********/
+/********** ===================== **********/
+
+def : Pat <
+  (int_SI_fs_constant imm:$attr_chan, imm:$attr, M0Reg:$params),
+  (V_INTERP_MOV_F32 INTERP.P0, imm:$attr_chan, imm:$attr, M0Reg:$params)
+>;
+
+def : Pat <
+  (int_SI_fs_interp imm:$attr_chan, imm:$attr, M0Reg:$params, VReg_64:$ij),
+  (V_INTERP_P2_F32 (V_INTERP_P1_F32 (EXTRACT_SUBREG VReg_64:$ij, sub0),
+                                    imm:$attr_chan, imm:$attr, M0Reg:$params),
+                   (EXTRACT_SUBREG VReg_64:$ij, sub1),
+                   imm:$attr_chan, imm:$attr, M0Reg:$params)
+>;
+
+/********** ================== **********/
+/********** Intrinsic Patterns **********/
+/********** ================== **********/
+
+/* llvm.AMDGPU.pow */
+def : POW_Common <V_LOG_F32_e32, V_EXP_F32_e32, V_MUL_LEGACY_F32_e32, VReg_32>;
+
+def : Pat <
+  (int_AMDGPU_div VSrc_32:$src0, VSrc_32:$src1),
+  (V_MUL_LEGACY_F32_e32 VSrc_32:$src0, (V_RCP_LEGACY_F32_e32 VSrc_32:$src1))
+>;
+
+def : Pat<
+  (fdiv VSrc_32:$src0, VSrc_32:$src1),
+  (V_MUL_F32_e32 VSrc_32:$src0, (V_RCP_F32_e32 VSrc_32:$src1))
+>;
+
+def : Pat <
+  (fcos VSrc_32:$src0),
+  (V_COS_F32_e32 (V_MUL_F32_e32 VSrc_32:$src0, (V_MOV_B32_e32 CONST.TWO_PI_INV)))
+>;
+
+def : Pat <
+  (fsin VSrc_32:$src0),
+  (V_SIN_F32_e32 (V_MUL_F32_e32 VSrc_32:$src0, (V_MOV_B32_e32 CONST.TWO_PI_INV)))
+>;
+
+def : Pat <
+  (int_AMDGPU_cube VReg_128:$src),
+  (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
+    (V_CUBETC_F32 (EXTRACT_SUBREG VReg_128:$src, sub0),
+                  (EXTRACT_SUBREG VReg_128:$src, sub1),
+                  (EXTRACT_SUBREG VReg_128:$src, sub2),
+                  0, 0, 0, 0), sub0),
+    (V_CUBESC_F32 (EXTRACT_SUBREG VReg_128:$src, sub0),
+                  (EXTRACT_SUBREG VReg_128:$src, sub1),
+                  (EXTRACT_SUBREG VReg_128:$src, sub2),
+                  0, 0, 0, 0), sub1),
+    (V_CUBEMA_F32 (EXTRACT_SUBREG VReg_128:$src, sub0),
+                  (EXTRACT_SUBREG VReg_128:$src, sub1),
+                  (EXTRACT_SUBREG VReg_128:$src, sub2),
+                  0, 0, 0, 0), sub2),
+    (V_CUBEID_F32 (EXTRACT_SUBREG VReg_128:$src, sub0),
+                  (EXTRACT_SUBREG VReg_128:$src, sub1),
+                  (EXTRACT_SUBREG VReg_128:$src, sub2),
+                  0, 0, 0, 0), sub3)
+>;
+
+def : Pat <
+  (i32 (sext (i1 SReg_64:$src0))),
+  (V_CNDMASK_B32_e64 (i32 0), (i32 -1), SReg_64:$src0)
+>;
+
+// 1. Offset as 8bit DWORD immediate
+def : Pat <
+  (int_SI_load_const SReg_128:$sbase, IMM8bitDWORD:$offset),
+  (S_BUFFER_LOAD_DWORD_IMM SReg_128:$sbase, IMM8bitDWORD:$offset)
+>;
+
+// 2. Offset loaded in an 32bit SGPR
+def : Pat <
+  (int_SI_load_const SReg_128:$sbase, imm:$offset),
+  (S_BUFFER_LOAD_DWORD_SGPR SReg_128:$sbase, (S_MOV_B32 imm:$offset))
+>;
+
+// 3. Offset in an 32Bit VGPR
+def : Pat <
+  (int_SI_load_const SReg_128:$sbase, VReg_32:$voff),
+  (BUFFER_LOAD_DWORD 0, 1, 0, 0, 0, 0, VReg_32:$voff, SReg_128:$sbase, 0, 0, 0)
+>;
+
+/********** ================== **********/
+/**********   VOP3 Patterns    **********/
+/********** ================== **********/
+
+def : Pat <(f32 (fadd (fmul VSrc_32:$src0, VSrc_32:$src1), VSrc_32:$src2)),
+           (V_MAD_F32 VSrc_32:$src0, VSrc_32:$src1, VSrc_32:$src2,
+            0, 0, 0, 0)>;
+
+/********** ================== **********/
+/**********   SMRD Patterns    **********/
+/********** ================== **********/
+
+multiclass SMRD_Pattern <SMRD Instr_IMM, SMRD Instr_SGPR, ValueType vt> {
+  // 1. Offset as 8bit DWORD immediate
+  def : Pat <
+    (constant_load (SIadd64bit32bit SReg_64:$sbase, IMM8bitDWORD:$offset)),
+    (vt (Instr_IMM SReg_64:$sbase, IMM8bitDWORD:$offset))
+  >;
+
+  // 2. Offset loaded in an 32bit SGPR
+  def : Pat <
+    (constant_load (SIadd64bit32bit SReg_64:$sbase, imm:$offset)),
+    (vt (Instr_SGPR SReg_64:$sbase, (S_MOV_B32 imm:$offset)))
+  >;
+
+  // 3. No offset at all
+  def : Pat <
+    (constant_load SReg_64:$sbase),
+    (vt (Instr_IMM SReg_64:$sbase, 0))
+  >;
+}
+
+defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, f32>;
+defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, i32>;
+defm : SMRD_Pattern <S_LOAD_DWORDX4_IMM, S_LOAD_DWORDX4_SGPR, v16i8>;
+defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v32i8>;
+
+/********** ====================== **********/
+/**********   Indirect adressing   **********/
+/********** ====================== **********/
+
+multiclass SI_INDIRECT_Pattern <RegisterClass rc, ValueType vt,
+                                SI_INDIRECT_DST IndDst> {
+  // 1. Extract with offset
+  def : Pat<
+    (vector_extract (vt rc:$vec),
+      (i64 (zext (i32 (add VReg_32:$idx, imm:$off))))
+    ),
+    (f32 (SI_INDIRECT_SRC (IMPLICIT_DEF), rc:$vec, VReg_32:$idx, imm:$off))
+  >;
+
+  // 2. Extract without offset
+  def : Pat<
+    (vector_extract (vt rc:$vec),
+      (i64 (zext (i32 VReg_32:$idx)))
+    ),
+    (f32 (SI_INDIRECT_SRC (IMPLICIT_DEF), rc:$vec, VReg_32:$idx, 0))
+  >;
+
+  // 3. Insert with offset
+  def : Pat<
+    (vector_insert (vt rc:$vec), (f32 VReg_32:$val),
+      (i64 (zext (i32 (add VReg_32:$idx, imm:$off))))
+    ),
+    (vt (IndDst (IMPLICIT_DEF), rc:$vec, VReg_32:$idx, imm:$off, VReg_32:$val))
+  >;
+
+  // 4. Insert without offset
+  def : Pat<
+    (vector_insert (vt rc:$vec), (f32 VReg_32:$val),
+      (i64 (zext (i32 VReg_32:$idx)))
+    ),
+    (vt (IndDst (IMPLICIT_DEF), rc:$vec, VReg_32:$idx, 0, VReg_32:$val))
+  >;
+}
+
+defm : SI_INDIRECT_Pattern <VReg_64, v2f32, SI_INDIRECT_DST_V2>;
+defm : SI_INDIRECT_Pattern <VReg_128, v4f32, SI_INDIRECT_DST_V4>;
+defm : SI_INDIRECT_Pattern <VReg_256, v8f32, SI_INDIRECT_DST_V8>;
+defm : SI_INDIRECT_Pattern <VReg_512, v16f32, SI_INDIRECT_DST_V16>;
+
+/********** =============== **********/
+/**********   Conditions    **********/
+/********** =============== **********/
+
+def : Pat<
+  (i1 (setcc f32:$src0, f32:$src1, SETO)),
+  (V_CMP_O_F32_e64 f32:$src0, f32:$src1)
+>;
+
+def : Pat<
+  (i1 (setcc f32:$src0, f32:$src1, SETUO)),
+  (V_CMP_U_F32_e64 f32:$src0, f32:$src1)
+>;
+
+} // End isSI predicate
diff --git a/lib/Target/R600/SIIntrinsics.td b/lib/Target/R600/SIIntrinsics.td
new file mode 100644
index 000000000000..0af378edfe2e
--- /dev/null
+++ b/lib/Target/R600/SIIntrinsics.td
@@ -0,0 +1,42 @@
+//===-- SIIntrinsics.td - SI Intrinsic defs ----------------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// SI Intrinsic Definitions
+//
+//===----------------------------------------------------------------------===//
+
+
+let TargetPrefix = "SI", isTarget = 1 in {
+
+  def int_SI_packf16 : Intrinsic <[llvm_i32_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+  def int_SI_export : Intrinsic <[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_float_ty, llvm_float_ty, llvm_float_ty, llvm_float_ty], []>;
+  def int_SI_load_const : Intrinsic <[llvm_float_ty], [llvm_v16i8_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_SI_vs_load_input : Intrinsic <[llvm_v4f32_ty], [llvm_v16i8_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]> ;
+
+  class Sample : Intrinsic <[llvm_v4f32_ty], [llvm_i32_ty, llvm_anyvector_ty, llvm_v32i8_ty, llvm_v16i8_ty, llvm_i32_ty], [IntrNoMem]>;
+
+  def int_SI_sample : Sample;
+  def int_SI_sampleb : Sample;
+  def int_SI_samplel : Sample;
+
+  /* Interpolation Intrinsics */
+
+  def int_SI_fs_constant : Intrinsic <[llvm_float_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_SI_fs_interp : Intrinsic <[llvm_float_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_v2i32_ty], [IntrNoMem]>;
+
+  /* Control flow Intrinsics */
+
+  def int_SI_if : Intrinsic<[llvm_i64_ty], [llvm_i1_ty, llvm_empty_ty], []>;
+  def int_SI_else : Intrinsic<[llvm_i64_ty], [llvm_i64_ty, llvm_empty_ty], []>;
+  def int_SI_break : Intrinsic<[llvm_i64_ty], [llvm_i64_ty], []>;
+  def int_SI_if_break : Intrinsic<[llvm_i64_ty], [llvm_i1_ty, llvm_i64_ty], []>;
+  def int_SI_else_break : Intrinsic<[llvm_i64_ty], [llvm_i64_ty, llvm_i64_ty], []>;
+  def int_SI_loop : Intrinsic<[], [llvm_i64_ty, llvm_empty_ty], []>;
+  def int_SI_end_cf : Intrinsic<[], [llvm_i64_ty], []>;
+}
diff --git a/lib/Target/R600/SILowerControlFlow.cpp b/lib/Target/R600/SILowerControlFlow.cpp
new file mode 100644
index 000000000000..2b60eb9fb375
--- /dev/null
+++ b/lib/Target/R600/SILowerControlFlow.cpp
@@ -0,0 +1,501 @@
+//===-- SILowerControlFlow.cpp - Use predicates for control flow ----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief This pass lowers the pseudo control flow instructions to real
+/// machine instructions.
+///
+/// All control flow is handled using predicated instructions and
+/// a predicate stack.  Each Scalar ALU controls the operations of 64 Vector
+/// ALUs.  The Scalar ALU can update the predicate for any of the Vector ALUs
+/// by writting to the 64-bit EXEC register (each bit corresponds to a
+/// single vector ALU).  Typically, for predicates, a vector ALU will write
+/// to its bit of the VCC register (like EXEC VCC is 64-bits, one for each
+/// Vector ALU) and then the ScalarALU will AND the VCC register with the
+/// EXEC to update the predicates.
+///
+/// For example:
+/// %VCC = V_CMP_GT_F32 %VGPR1, %VGPR2
+/// %SGPR0 = SI_IF %VCC
+///   %VGPR0 = V_ADD_F32 %VGPR0, %VGPR0
+/// %SGPR0 = SI_ELSE %SGPR0
+///   %VGPR0 = V_SUB_F32 %VGPR0, %VGPR0
+/// SI_END_CF %SGPR0
+///
+/// becomes:
+///
+/// %SGPR0 = S_AND_SAVEEXEC_B64 %VCC  // Save and update the exec mask
+/// %SGPR0 = S_XOR_B64 %SGPR0, %EXEC  // Clear live bits from saved exec mask
+/// S_CBRANCH_EXECZ label0            // This instruction is an optional
+///                                   // optimization which allows us to
+///                                   // branch if all the bits of
+///                                   // EXEC are zero.
+/// %VGPR0 = V_ADD_F32 %VGPR0, %VGPR0 // Do the IF block of the branch
+///
+/// label0:
+/// %SGPR0 = S_OR_SAVEEXEC_B64 %EXEC   // Restore the exec mask for the Then block
+/// %EXEC = S_XOR_B64 %SGPR0, %EXEC    // Clear live bits from saved exec mask
+/// S_BRANCH_EXECZ label1              // Use our branch optimization
+///                                    // instruction again.
+/// %VGPR0 = V_SUB_F32 %VGPR0, %VGPR   // Do the THEN block
+/// label1:
+/// %EXEC = S_OR_B64 %EXEC, %SGPR0     // Re-enable saved exec mask bits
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "SIInstrInfo.h"
+#include "SIMachineFunctionInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+
+using namespace llvm;
+
+namespace {
+
+class SILowerControlFlowPass : public MachineFunctionPass {
+
+private:
+  static const unsigned SkipThreshold = 12;
+
+  static char ID;
+  const TargetRegisterInfo *TRI;
+  const TargetInstrInfo *TII;
+
+  bool shouldSkip(MachineBasicBlock *From, MachineBasicBlock *To);
+
+  void Skip(MachineInstr &From, MachineOperand &To);
+  void SkipIfDead(MachineInstr &MI);
+
+  void If(MachineInstr &MI);
+  void Else(MachineInstr &MI);
+  void Break(MachineInstr &MI);
+  void IfBreak(MachineInstr &MI);
+  void ElseBreak(MachineInstr &MI);
+  void Loop(MachineInstr &MI);
+  void EndCf(MachineInstr &MI);
+
+  void Kill(MachineInstr &MI);
+  void Branch(MachineInstr &MI);
+
+  void LoadM0(MachineInstr &MI, MachineInstr *MovRel);
+  void IndirectSrc(MachineInstr &MI);
+  void IndirectDst(MachineInstr &MI);
+
+public:
+  SILowerControlFlowPass(TargetMachine &tm) :
+    MachineFunctionPass(ID), TRI(tm.getRegisterInfo()),
+    TII(tm.getInstrInfo()) { }
+
+  virtual bool runOnMachineFunction(MachineFunction &MF);
+
+  const char *getPassName() const {
+    return "SI Lower control flow instructions";
+  }
+
+};
+
+} // End anonymous namespace
+
+char SILowerControlFlowPass::ID = 0;
+
+FunctionPass *llvm::createSILowerControlFlowPass(TargetMachine &tm) {
+  return new SILowerControlFlowPass(tm);
+}
+
+bool SILowerControlFlowPass::shouldSkip(MachineBasicBlock *From,
+                                        MachineBasicBlock *To) {
+
+  unsigned NumInstr = 0;
+
+  for (MachineBasicBlock *MBB = From; MBB != To && !MBB->succ_empty();
+       MBB = *MBB->succ_begin()) {
+
+    for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end();
+         NumInstr < SkipThreshold && I != E; ++I) {
+
+      if (I->isBundle() || !I->isBundled())
+        if (++NumInstr >= SkipThreshold)
+          return true;
+    }
+  }
+
+  return false;
+}
+
+void SILowerControlFlowPass::Skip(MachineInstr &From, MachineOperand &To) {
+
+  if (!shouldSkip(*From.getParent()->succ_begin(), To.getMBB()))
+    return;
+
+  DebugLoc DL = From.getDebugLoc();
+  BuildMI(*From.getParent(), &From, DL, TII->get(AMDGPU::S_CBRANCH_EXECZ))
+          .addOperand(To)
+          .addReg(AMDGPU::EXEC);
+}
+
+void SILowerControlFlowPass::SkipIfDead(MachineInstr &MI) {
+
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+
+  if (!shouldSkip(&MBB, &MBB.getParent()->back()))
+    return;
+
+  MachineBasicBlock::iterator Insert = &MI;
+  ++Insert;
+
+  // If the exec mask is non-zero, skip the next two instructions
+  BuildMI(MBB, Insert, DL, TII->get(AMDGPU::S_CBRANCH_EXECNZ))
+          .addImm(3)
+          .addReg(AMDGPU::EXEC);
+
+  // Exec mask is zero: Export to NULL target...
+  BuildMI(MBB, Insert, DL, TII->get(AMDGPU::EXP))
+          .addImm(0)
+          .addImm(0x09) // V_008DFC_SQ_EXP_NULL
+          .addImm(0)
+          .addImm(1)
+          .addImm(1)
+          .addReg(AMDGPU::VGPR0)
+          .addReg(AMDGPU::VGPR0)
+          .addReg(AMDGPU::VGPR0)
+          .addReg(AMDGPU::VGPR0);
+
+  // ... and terminate wavefront
+  BuildMI(MBB, Insert, DL, TII->get(AMDGPU::S_ENDPGM));
+}
+
+void SILowerControlFlowPass::If(MachineInstr &MI) {
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+  unsigned Reg = MI.getOperand(0).getReg();
+  unsigned Vcc = MI.getOperand(1).getReg();
+
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_AND_SAVEEXEC_B64), Reg)
+          .addReg(Vcc);
+
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_XOR_B64), Reg)
+          .addReg(AMDGPU::EXEC)
+          .addReg(Reg);
+
+  Skip(MI, MI.getOperand(2));
+
+  MI.eraseFromParent();
+}
+
+void SILowerControlFlowPass::Else(MachineInstr &MI) {
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+  unsigned Dst = MI.getOperand(0).getReg();
+  unsigned Src = MI.getOperand(1).getReg();
+
+  BuildMI(MBB, MBB.getFirstNonPHI(), DL,
+          TII->get(AMDGPU::S_OR_SAVEEXEC_B64), Dst)
+          .addReg(Src); // Saved EXEC
+
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_XOR_B64), AMDGPU::EXEC)
+          .addReg(AMDGPU::EXEC)
+          .addReg(Dst);
+
+  Skip(MI, MI.getOperand(2));
+
+  MI.eraseFromParent();
+}
+
+void SILowerControlFlowPass::Break(MachineInstr &MI) {
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+
+  unsigned Dst = MI.getOperand(0).getReg();
+  unsigned Src = MI.getOperand(1).getReg();
+ 
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_OR_B64), Dst)
+          .addReg(AMDGPU::EXEC)
+          .addReg(Src);
+
+  MI.eraseFromParent();
+}
+
+void SILowerControlFlowPass::IfBreak(MachineInstr &MI) {
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+
+  unsigned Dst = MI.getOperand(0).getReg();
+  unsigned Vcc = MI.getOperand(1).getReg();
+  unsigned Src = MI.getOperand(2).getReg();
+ 
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_OR_B64), Dst)
+          .addReg(Vcc)
+          .addReg(Src);
+
+  MI.eraseFromParent();
+}
+
+void SILowerControlFlowPass::ElseBreak(MachineInstr &MI) {
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+
+  unsigned Dst = MI.getOperand(0).getReg();
+  unsigned Saved = MI.getOperand(1).getReg();
+  unsigned Src = MI.getOperand(2).getReg();
+ 
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_OR_B64), Dst)
+          .addReg(Saved)
+          .addReg(Src);
+
+  MI.eraseFromParent();
+}
+
+void SILowerControlFlowPass::Loop(MachineInstr &MI) {
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+  unsigned Src = MI.getOperand(0).getReg();
+
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_ANDN2_B64), AMDGPU::EXEC)
+          .addReg(AMDGPU::EXEC)
+          .addReg(Src);
+
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_CBRANCH_EXECNZ))
+          .addOperand(MI.getOperand(1))
+          .addReg(AMDGPU::EXEC);
+
+  MI.eraseFromParent();
+}
+
+void SILowerControlFlowPass::EndCf(MachineInstr &MI) {
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+  unsigned Reg = MI.getOperand(0).getReg();
+
+  BuildMI(MBB, MBB.getFirstNonPHI(), DL,
+          TII->get(AMDGPU::S_OR_B64), AMDGPU::EXEC)
+          .addReg(AMDGPU::EXEC)
+          .addReg(Reg);
+
+  MI.eraseFromParent();
+}
+
+void SILowerControlFlowPass::Branch(MachineInstr &MI) {
+  MachineBasicBlock *Next = MI.getParent()->getNextNode();
+  MachineBasicBlock *Target = MI.getOperand(0).getMBB();
+  if (Target == Next)
+    MI.eraseFromParent();
+  else
+    assert(0);
+}
+
+void SILowerControlFlowPass::Kill(MachineInstr &MI) {
+
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+
+  // Kill is only allowed in pixel shaders
+  assert(MBB.getParent()->getInfo<SIMachineFunctionInfo>()->ShaderType ==
+         ShaderType::PIXEL);
+
+  // Clear this pixel from the exec mask if the operand is negative
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_CMPX_LE_F32_e32), AMDGPU::VCC)
+          .addImm(0)
+          .addOperand(MI.getOperand(0));
+
+  MI.eraseFromParent();
+}
+
+void SILowerControlFlowPass::LoadM0(MachineInstr &MI, MachineInstr *MovRel) {
+
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+  MachineBasicBlock::iterator I = MI;
+
+  unsigned Save = MI.getOperand(1).getReg();
+  unsigned Idx = MI.getOperand(3).getReg();
+
+  if (AMDGPU::SReg_32RegClass.contains(Idx)) {
+    BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B32), AMDGPU::M0)
+            .addReg(Idx);
+    MBB.insert(I, MovRel);
+    MI.eraseFromParent();
+    return;
+  }
+
+  assert(AMDGPU::SReg_64RegClass.contains(Save));
+  assert(AMDGPU::VReg_32RegClass.contains(Idx));
+
+  // Save the EXEC mask
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B64), Save)
+          .addReg(AMDGPU::EXEC);
+
+  // Read the next variant into VCC (lower 32 bits) <- also loop target
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32_e32), AMDGPU::VCC)
+          .addReg(Idx);
+
+  // Move index from VCC into M0
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B32), AMDGPU::M0)
+          .addReg(AMDGPU::VCC);
+
+  // Compare the just read M0 value to all possible Idx values
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_CMP_EQ_U32_e32), AMDGPU::VCC)
+          .addReg(AMDGPU::M0)
+          .addReg(Idx);
+
+  // Update EXEC, save the original EXEC value to VCC
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_AND_SAVEEXEC_B64), AMDGPU::VCC)
+          .addReg(AMDGPU::VCC);
+
+  // Do the actual move
+  MBB.insert(I, MovRel);
+
+  // Update EXEC, switch all done bits to 0 and all todo bits to 1
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_XOR_B64), AMDGPU::EXEC)
+          .addReg(AMDGPU::EXEC)
+          .addReg(AMDGPU::VCC);
+
+  // Loop back to V_READFIRSTLANE_B32 if there are still variants to cover
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_CBRANCH_EXECNZ))
+          .addImm(-7)
+          .addReg(AMDGPU::EXEC);
+
+  // Restore EXEC
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B64), AMDGPU::EXEC)
+          .addReg(Save);
+
+  MI.eraseFromParent();
+}
+
+void SILowerControlFlowPass::IndirectSrc(MachineInstr &MI) {
+
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+
+  unsigned Dst = MI.getOperand(0).getReg();
+  unsigned Vec = MI.getOperand(2).getReg();
+  unsigned Off = MI.getOperand(4).getImm();
+
+  MachineInstr *MovRel = 
+    BuildMI(*MBB.getParent(), DL, TII->get(AMDGPU::V_MOVRELS_B32_e32), Dst)
+            .addReg(TRI->getSubReg(Vec, AMDGPU::sub0) + Off)
+            .addReg(AMDGPU::M0, RegState::Implicit)
+            .addReg(Vec, RegState::Implicit);
+
+  LoadM0(MI, MovRel);
+}
+
+void SILowerControlFlowPass::IndirectDst(MachineInstr &MI) {
+
+  MachineBasicBlock &MBB = *MI.getParent();
+  DebugLoc DL = MI.getDebugLoc();
+
+  unsigned Dst = MI.getOperand(0).getReg();
+  unsigned Off = MI.getOperand(4).getImm();
+  unsigned Val = MI.getOperand(5).getReg();
+
+  MachineInstr *MovRel = 
+    BuildMI(*MBB.getParent(), DL, TII->get(AMDGPU::V_MOVRELD_B32_e32))
+            .addReg(TRI->getSubReg(Dst, AMDGPU::sub0) + Off, RegState::Define)
+            .addReg(Val)
+            .addReg(AMDGPU::M0, RegState::Implicit)
+            .addReg(Dst, RegState::Implicit);
+
+  LoadM0(MI, MovRel);
+}
+
+bool SILowerControlFlowPass::runOnMachineFunction(MachineFunction &MF) {
+
+  bool HaveKill = false;
+  bool NeedWQM = false;
+  unsigned Depth = 0;
+
+  for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
+       BI != BE; ++BI) {
+
+    MachineBasicBlock &MBB = *BI;
+    for (MachineBasicBlock::iterator I = MBB.begin(), Next = llvm::next(I);
+         I != MBB.end(); I = Next) {
+
+      Next = llvm::next(I);
+      MachineInstr &MI = *I;
+      switch (MI.getOpcode()) {
+        default: break;
+        case AMDGPU::SI_IF:
+          ++Depth;
+          If(MI);
+          break;
+
+        case AMDGPU::SI_ELSE:
+          Else(MI);
+          break;
+
+        case AMDGPU::SI_BREAK:
+          Break(MI);
+          break;
+
+        case AMDGPU::SI_IF_BREAK:
+          IfBreak(MI);
+          break;
+
+        case AMDGPU::SI_ELSE_BREAK:
+          ElseBreak(MI);
+          break;
+
+        case AMDGPU::SI_LOOP:
+          ++Depth;
+          Loop(MI);
+          break;
+
+        case AMDGPU::SI_END_CF:
+          if (--Depth == 0 && HaveKill) {
+            SkipIfDead(MI);
+            HaveKill = false;
+          }
+          EndCf(MI);
+          break;
+
+        case AMDGPU::SI_KILL:
+          if (Depth == 0)
+            SkipIfDead(MI);
+          else
+            HaveKill = true;
+          Kill(MI);
+          break;
+
+        case AMDGPU::S_BRANCH:
+          Branch(MI);
+          break;
+
+        case AMDGPU::SI_INDIRECT_SRC:
+          IndirectSrc(MI);
+          break;
+
+        case AMDGPU::SI_INDIRECT_DST_V2:
+        case AMDGPU::SI_INDIRECT_DST_V4:
+        case AMDGPU::SI_INDIRECT_DST_V8:
+        case AMDGPU::SI_INDIRECT_DST_V16:
+          IndirectDst(MI);
+          break;
+
+        case AMDGPU::V_INTERP_P1_F32:
+        case AMDGPU::V_INTERP_P2_F32:
+        case AMDGPU::V_INTERP_MOV_F32:
+          NeedWQM = true;
+          break;
+
+      }
+    }
+  }
+
+  if (NeedWQM) {
+    MachineBasicBlock &MBB = MF.front();
+    BuildMI(MBB, MBB.getFirstNonPHI(), DebugLoc(), TII->get(AMDGPU::S_WQM_B64),
+            AMDGPU::EXEC).addReg(AMDGPU::EXEC);
+  }
+
+  return true;
+}
diff --git a/lib/Target/R600/SIMachineFunctionInfo.cpp b/lib/Target/R600/SIMachineFunctionInfo.cpp
new file mode 100644
index 000000000000..ee0e30755f01
--- /dev/null
+++ b/lib/Target/R600/SIMachineFunctionInfo.cpp
@@ -0,0 +1,18 @@
+//===-- SIMachineFunctionInfo.cpp - SI Machine Function Info -------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+/// \file
+//===----------------------------------------------------------------------===//
+
+
+#include "SIMachineFunctionInfo.h"
+
+using namespace llvm;
+
+SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
+  : AMDGPUMachineFunction(MF),
+    PSInputAddr(0) { }
diff --git a/lib/Target/R600/SIMachineFunctionInfo.h b/lib/Target/R600/SIMachineFunctionInfo.h
new file mode 100644
index 000000000000..6da9f7f9a14d
--- /dev/null
+++ b/lib/Target/R600/SIMachineFunctionInfo.h
@@ -0,0 +1,33 @@
+//===- SIMachineFunctionInfo.h - SIMachineFunctionInfo interface -*- C++ -*-==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+//
+//===----------------------------------------------------------------------===//
+
+
+#ifndef SIMACHINEFUNCTIONINFO_H_
+#define SIMACHINEFUNCTIONINFO_H_
+
+#include "AMDGPUMachineFunction.h"
+
+namespace llvm {
+
+/// This class keeps track of the SPI_SP_INPUT_ADDR config register, which
+/// tells the hardware which interpolation parameters to load.
+class SIMachineFunctionInfo : public AMDGPUMachineFunction {
+public:
+  SIMachineFunctionInfo(const MachineFunction &MF);
+  unsigned PSInputAddr;
+};
+
+} // End namespace llvm
+
+
+#endif //_SIMACHINEFUNCTIONINFO_H_
diff --git a/lib/Target/R600/SIRegisterInfo.cpp b/lib/Target/R600/SIRegisterInfo.cpp
new file mode 100644
index 000000000000..99278ae8dceb
--- /dev/null
+++ b/lib/Target/R600/SIRegisterInfo.cpp
@@ -0,0 +1,53 @@
+//===-- SIRegisterInfo.cpp - SI Register Information ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief SI implementation of the TargetRegisterInfo class.
+//
+//===----------------------------------------------------------------------===//
+
+
+#include "SIRegisterInfo.h"
+#include "AMDGPUTargetMachine.h"
+
+using namespace llvm;
+
+SIRegisterInfo::SIRegisterInfo(AMDGPUTargetMachine &tm,
+    const TargetInstrInfo &tii)
+: AMDGPURegisterInfo(tm, tii),
+  TM(tm),
+  TII(tii)
+  { }
+
+BitVector SIRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
+  BitVector Reserved(getNumRegs());
+  return Reserved;
+}
+
+unsigned SIRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC,
+                                             MachineFunction &MF) const {
+  return RC->getNumRegs();
+}
+
+const TargetRegisterClass *
+SIRegisterInfo::getISARegClass(const TargetRegisterClass * rc) const {
+  switch (rc->getID()) {
+  case AMDGPU::GPRF32RegClassID:
+    return &AMDGPU::VReg_32RegClass;
+  default: return rc;
+  }
+}
+
+const TargetRegisterClass * SIRegisterInfo::getCFGStructurizerRegClass(
+                                                                   MVT VT) const {
+  switch(VT.SimpleTy) {
+    default:
+    case MVT::i32: return &AMDGPU::VReg_32RegClass;
+  }
+}
diff --git a/lib/Target/R600/SIRegisterInfo.h b/lib/Target/R600/SIRegisterInfo.h
new file mode 100644
index 000000000000..caec22841345
--- /dev/null
+++ b/lib/Target/R600/SIRegisterInfo.h
@@ -0,0 +1,50 @@
+//===-- SIRegisterInfo.h - SI Register Info Interface ----------*- C++ -*--===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief Interface definition for SIRegisterInfo
+//
+//===----------------------------------------------------------------------===//
+
+
+#ifndef SIREGISTERINFO_H_
+#define SIREGISTERINFO_H_
+
+#include "AMDGPURegisterInfo.h"
+
+namespace llvm {
+
+class AMDGPUTargetMachine;
+class TargetInstrInfo;
+
+struct SIRegisterInfo : public AMDGPURegisterInfo {
+  AMDGPUTargetMachine &TM;
+  const TargetInstrInfo &TII;
+
+  SIRegisterInfo(AMDGPUTargetMachine &tm, const TargetInstrInfo &tii);
+
+  virtual BitVector getReservedRegs(const MachineFunction &MF) const;
+
+  virtual unsigned getRegPressureLimit(const TargetRegisterClass *RC,
+                                       MachineFunction &MF) const;
+
+  /// \param RC is an AMDIL reg class.
+  ///
+  /// \returns the SI register class that is equivalent to \p RC.
+  virtual const TargetRegisterClass *
+    getISARegClass(const TargetRegisterClass *RC) const;
+
+  /// \brief get the register class of the specified type to use in the
+  /// CFGStructurizer
+  virtual const TargetRegisterClass * getCFGStructurizerRegClass(MVT VT) const;
+};
+
+} // End namespace llvm
+
+#endif // SIREGISTERINFO_H_
diff --git a/lib/Target/R600/SIRegisterInfo.td b/lib/Target/R600/SIRegisterInfo.td
new file mode 100644
index 000000000000..4f14931a9c48
--- /dev/null
+++ b/lib/Target/R600/SIRegisterInfo.td
@@ -0,0 +1,182 @@
+//===-- SIRegisterInfo.td - SI Register defs ---------------*- tablegen -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+//  Declarations that describe the SI registers
+//===----------------------------------------------------------------------===//
+
+class SIReg <string n, bits<16> encoding = 0> : Register<n> {
+  let Namespace = "AMDGPU";
+  let HWEncoding = encoding;
+}
+
+// Special Registers
+def VCC : SIReg<"VCC", 106>;
+def EXEC : SIReg<"EXEC", 126>;
+def SCC : SIReg<"SCC", 253>;
+def M0 : SIReg <"M0", 124>;
+
+// SGPR registers
+foreach Index = 0-101 in {
+  def SGPR#Index : SIReg <"SGPR"#Index, Index>;
+}
+
+// VGPR registers
+foreach Index = 0-255 in {
+  def VGPR#Index : SIReg <"VGPR"#Index, Index> {
+    let HWEncoding{8} = 1;
+  }
+}
+
+//===----------------------------------------------------------------------===//
+//  Groupings using register classes and tuples
+//===----------------------------------------------------------------------===//
+
+// SGPR 32-bit registers
+def SGPR_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
+                            (add (sequence "SGPR%u", 0, 101))>;
+
+// SGPR 64-bit registers
+def SGPR_64 : RegisterTuples<[sub0, sub1],
+                             [(add (decimate (trunc SGPR_32, 101), 2)),
+                              (add (decimate (shl SGPR_32, 1), 2))]>;
+
+// SGPR 128-bit registers
+def SGPR_128 : RegisterTuples<[sub0, sub1, sub2, sub3],
+                              [(add (decimate (trunc SGPR_32, 99), 4)),
+                               (add (decimate (shl SGPR_32, 1), 4)),
+                               (add (decimate (shl SGPR_32, 2), 4)),
+                               (add (decimate (shl SGPR_32, 3), 4))]>;
+
+// SGPR 256-bit registers
+def SGPR_256 : RegisterTuples<[sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7],
+                              [(add (decimate (trunc SGPR_32, 95), 4)),
+                               (add (decimate (shl SGPR_32, 1), 4)),
+                               (add (decimate (shl SGPR_32, 2), 4)),
+                               (add (decimate (shl SGPR_32, 3), 4)),
+                               (add (decimate (shl SGPR_32, 4), 4)),
+                               (add (decimate (shl SGPR_32, 5), 4)),
+                               (add (decimate (shl SGPR_32, 6), 4)),
+                               (add (decimate (shl SGPR_32, 7), 4))]>;
+
+// SGPR 512-bit registers
+def SGPR_512 : RegisterTuples<[sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7,
+                               sub8, sub9, sub10, sub11, sub12, sub13, sub14, sub15],
+                              [(add (decimate (trunc SGPR_32, 87), 4)),
+                               (add (decimate (shl SGPR_32, 1), 4)),
+                               (add (decimate (shl SGPR_32, 2), 4)),
+                               (add (decimate (shl SGPR_32, 3), 4)),
+                               (add (decimate (shl SGPR_32, 4), 4)),
+                               (add (decimate (shl SGPR_32, 5), 4)),
+                               (add (decimate (shl SGPR_32, 6), 4)),
+                               (add (decimate (shl SGPR_32, 7), 4)),
+                               (add (decimate (shl SGPR_32, 8), 4)),
+                               (add (decimate (shl SGPR_32, 9), 4)),
+                               (add (decimate (shl SGPR_32, 10), 4)),
+                               (add (decimate (shl SGPR_32, 11), 4)),
+                               (add (decimate (shl SGPR_32, 12), 4)),
+                               (add (decimate (shl SGPR_32, 13), 4)),
+                               (add (decimate (shl SGPR_32, 14), 4)),
+                               (add (decimate (shl SGPR_32, 15), 4))]>;
+
+// VGPR 32-bit registers
+def VGPR_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
+                            (add (sequence "VGPR%u", 0, 255))>;
+
+// VGPR 64-bit registers
+def VGPR_64 : RegisterTuples<[sub0, sub1],
+                             [(add (trunc VGPR_32, 255)),
+                              (add (shl VGPR_32, 1))]>;
+
+// VGPR 128-bit registers
+def VGPR_128 : RegisterTuples<[sub0, sub1, sub2, sub3],
+                              [(add (trunc VGPR_32, 253)),
+                               (add (shl VGPR_32, 1)),
+                               (add (shl VGPR_32, 2)),
+                               (add (shl VGPR_32, 3))]>;
+
+// VGPR 256-bit registers
+def VGPR_256 : RegisterTuples<[sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7],
+                              [(add (trunc VGPR_32, 249)),
+                               (add (shl VGPR_32, 1)),
+                               (add (shl VGPR_32, 2)),
+                               (add (shl VGPR_32, 3)),
+                               (add (shl VGPR_32, 4)),
+                               (add (shl VGPR_32, 5)),
+                               (add (shl VGPR_32, 6)),
+                               (add (shl VGPR_32, 7))]>;
+
+// VGPR 512-bit registers
+def VGPR_512 : RegisterTuples<[sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7,
+                               sub8, sub9, sub10, sub11, sub12, sub13, sub14, sub15],
+                              [(add (trunc VGPR_32, 241)),
+                               (add (shl VGPR_32, 1)),
+                               (add (shl VGPR_32, 2)),
+                               (add (shl VGPR_32, 3)),
+                               (add (shl VGPR_32, 4)),
+                               (add (shl VGPR_32, 5)),
+                               (add (shl VGPR_32, 6)),
+                               (add (shl VGPR_32, 7)),
+                               (add (shl VGPR_32, 8)),
+                               (add (shl VGPR_32, 9)),
+                               (add (shl VGPR_32, 10)),
+                               (add (shl VGPR_32, 11)),
+                               (add (shl VGPR_32, 12)),
+                               (add (shl VGPR_32, 13)),
+                               (add (shl VGPR_32, 14)),
+                               (add (shl VGPR_32, 15))]>;
+
+//===----------------------------------------------------------------------===//
+//  Register classes used as source and destination
+//===----------------------------------------------------------------------===//
+
+// Special register classes for predicates and the M0 register
+def SCCReg : RegisterClass<"AMDGPU", [i32, i1], 32, (add SCC)>;
+def VCCReg : RegisterClass<"AMDGPU", [i64, i1], 64, (add VCC)>;
+def EXECReg : RegisterClass<"AMDGPU", [i64, i1], 64, (add EXEC)>;
+def M0Reg : RegisterClass<"AMDGPU", [i32], 32, (add M0)>;
+
+// Register class for all scalar registers (SGPRs + Special Registers)
+def SReg_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
+  (add SGPR_32, M0Reg)
+>;
+
+def SReg_64 : RegisterClass<"AMDGPU", [i64, i1], 64,
+  (add SGPR_64, VCCReg, EXECReg)
+>;
+
+def SReg_128 : RegisterClass<"AMDGPU", [v16i8], 128, (add SGPR_128)>;
+
+def SReg_256 : RegisterClass<"AMDGPU", [v32i8], 256, (add SGPR_256)>;
+
+def SReg_512 : RegisterClass<"AMDGPU", [v64i8], 512, (add SGPR_512)>;
+
+// Register class for all vector registers (VGPRs + Interploation Registers)
+def VReg_32 : RegisterClass<"AMDGPU", [i32, f32, v1i32], 32, (add VGPR_32)>;
+
+def VReg_64 : RegisterClass<"AMDGPU", [i64, f64, v2i32, v2f32], 64, (add VGPR_64)>;
+
+def VReg_128 : RegisterClass<"AMDGPU", [v4i32, v4f32], 128, (add VGPR_128)>;
+
+def VReg_256 : RegisterClass<"AMDGPU", [v8i32, v8f32], 256, (add VGPR_256)>;
+
+def VReg_512 : RegisterClass<"AMDGPU", [v16i32, v16f32], 512, (add VGPR_512)>;
+
+//===----------------------------------------------------------------------===//
+//  [SV]Src_* register classes, can have either an immediate or an register
+//===----------------------------------------------------------------------===//
+
+def SSrc_32 : RegisterClass<"AMDGPU", [i32, f32], 32, (add SReg_32)>;
+
+def SSrc_64 : RegisterClass<"AMDGPU", [i64, f64, i1], 64, (add SReg_64)>;
+
+def VSrc_32 : RegisterClass<"AMDGPU", [i32, f32], 32, (add VReg_32, SReg_32)>;
+
+def VSrc_64 : RegisterClass<"AMDGPU", [i64, f64], 64, (add VReg_64, SReg_64)>;
+
diff --git a/lib/Target/R600/SISchedule.td b/lib/Target/R600/SISchedule.td
new file mode 100644
index 000000000000..28b65b825855
--- /dev/null
+++ b/lib/Target/R600/SISchedule.td
@@ -0,0 +1,15 @@
+//===-- SISchedule.td - SI Scheduling definitons -------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// TODO: This is just a place holder for now.
+//
+//===----------------------------------------------------------------------===//
+
+
+def SI_Itin : ProcessorItineraries <[], [], []>;
diff --git a/lib/Target/R600/TargetInfo/AMDGPUTargetInfo.cpp b/lib/Target/R600/TargetInfo/AMDGPUTargetInfo.cpp
new file mode 100644
index 000000000000..46b1f18c6263
--- /dev/null
+++ b/lib/Target/R600/TargetInfo/AMDGPUTargetInfo.cpp
@@ -0,0 +1,26 @@
+//===-- TargetInfo/AMDGPUTargetInfo.cpp - TargetInfo for AMDGPU -----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "llvm/Support/TargetRegistry.h"
+
+using namespace llvm;
+
+/// \brief The target for the AMDGPU backend
+Target llvm::TheAMDGPUTarget;
+
+/// \brief Extern function to initialize the targets for the AMDGPU backend
+extern "C" void LLVMInitializeR600TargetInfo() {
+  RegisterTarget<Triple::r600, false>
+    R600(TheAMDGPUTarget, "r600", "AMD GPUs HD2XXX-HD6XXX");
+}
diff --git a/lib/Target/R600/TargetInfo/CMakeLists.txt b/lib/Target/R600/TargetInfo/CMakeLists.txt
new file mode 100644
index 000000000000..3d1584eba346
--- /dev/null
+++ b/lib/Target/R600/TargetInfo/CMakeLists.txt
@@ -0,0 +1,7 @@
+include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
+
+add_llvm_library(LLVMR600Info
+  AMDGPUTargetInfo.cpp
+  )
+
+add_dependencies(LLVMR600Info AMDGPUCommonTableGen intrinsics_gen)
diff --git a/lib/Target/R600/TargetInfo/LLVMBuild.txt b/lib/Target/R600/TargetInfo/LLVMBuild.txt
new file mode 100644
index 000000000000..4c6fea4aa08c
--- /dev/null
+++ b/lib/Target/R600/TargetInfo/LLVMBuild.txt
@@ -0,0 +1,23 @@
+;===- ./lib/Target/R600/TargetInfo/LLVMBuild.txt --------------*- Conf -*--===;
+;
+;                     The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+;   http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[component_0]
+type = Library
+name = R600Info
+parent = R600
+required_libraries = MC Support
+add_to_library_groups = R600
diff --git a/lib/Target/R600/TargetInfo/Makefile b/lib/Target/R600/TargetInfo/Makefile
new file mode 100644
index 000000000000..b8ac4e782302
--- /dev/null
+++ b/lib/Target/R600/TargetInfo/Makefile
@@ -0,0 +1,15 @@
+##===- lib/Target/AMDGPU/TargetInfo/Makefile ----------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+LEVEL = ../../../..
+LIBRARYNAME = LLVMR600Info
+
+# Hack: we need to include 'main' target directory to grab private headers
+CPPFLAGS = -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/Target/README.txt b/lib/Target/README.txt
index 8165f5b8cc97..a9aab86abdac 100644
--- a/lib/Target/README.txt
+++ b/lib/Target/README.txt
@@ -262,22 +262,7 @@ unsigned countbits_slow(unsigned v) {
     c += v & 1;
   return c;
 }
-unsigned countbits_fast(unsigned v){
-  unsigned c;
-  for (c = 0; v; c++)
-    v &= v - 1; // clear the least significant bit set
-  return c;
-}
 
-BITBOARD = unsigned long long
-int PopCnt(register BITBOARD a) {
-  register int c=0;
-  while(a) {
-    c++;
-    a &= a - 1;
-  }
-  return c;
-}
 unsigned int popcount(unsigned int input) {
   unsigned int count = 0;
   for (unsigned int i =  0; i < 4 * 8; i++)
diff --git a/lib/Target/Sparc/DelaySlotFiller.cpp b/lib/Target/Sparc/DelaySlotFiller.cpp
index 7bf8c3f85eca..6123773d5f4b 100644
--- a/lib/Target/Sparc/DelaySlotFiller.cpp
+++ b/lib/Target/Sparc/DelaySlotFiller.cpp
@@ -14,14 +14,14 @@
 
 #define DEBUG_TYPE "delay-slot-filler"
 #include "Sparc.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/Support/CommandLine.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/ADT/SmallSet.h"
-#include "llvm/ADT/Statistic.h"
 
 using namespace llvm;
 
diff --git a/lib/Target/Sparc/FPMover.cpp b/lib/Target/Sparc/FPMover.cpp
index 9a729bd87044..1325b98cf0ee 100644
--- a/lib/Target/Sparc/FPMover.cpp
+++ b/lib/Target/Sparc/FPMover.cpp
@@ -14,14 +14,14 @@
 #define DEBUG_TYPE "fpmover"
 #include "Sparc.h"
 #include "SparcSubtarget.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/ADT/Statistic.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
 using namespace llvm;
 
 STATISTIC(NumFpDs , "Number of instructions translated");
diff --git a/lib/Target/Sparc/MCTargetDesc/SparcMCAsmInfo.cpp b/lib/Target/Sparc/MCTargetDesc/SparcMCAsmInfo.cpp
index f5e10fc3a465..3d4bfdcd5e6d 100644
--- a/lib/Target/Sparc/MCTargetDesc/SparcMCAsmInfo.cpp
+++ b/lib/Target/Sparc/MCTargetDesc/SparcMCAsmInfo.cpp
@@ -21,8 +21,9 @@ void SparcELFMCAsmInfo::anchor() { }
 SparcELFMCAsmInfo::SparcELFMCAsmInfo(const Target &T, StringRef TT) {
   IsLittleEndian = false;
   Triple TheTriple(TT);
-  if (TheTriple.getArch() == Triple::sparcv9)
-    PointerSize = 8;
+  if (TheTriple.getArch() == Triple::sparcv9) {
+    PointerSize = CalleeSaveStackSlotSize = 8;
+  }
 
   Data16bitsDirective = "\t.half\t";
   Data32bitsDirective = "\t.word\t";
diff --git a/lib/Target/Sparc/SparcAsmPrinter.cpp b/lib/Target/Sparc/SparcAsmPrinter.cpp
index 25548625e760..e14b3cbf161d 100644
--- a/lib/Target/Sparc/SparcAsmPrinter.cpp
+++ b/lib/Target/Sparc/SparcAsmPrinter.cpp
@@ -16,15 +16,15 @@
 #include "Sparc.h"
 #include "SparcInstrInfo.h"
 #include "SparcTargetMachine.h"
+#include "llvm/ADT/SmallString.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/Target/Mangler.h"
-#include "llvm/ADT/SmallString.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/Mangler.h"
 using namespace llvm;
 
 namespace {
diff --git a/lib/Target/Sparc/SparcCallingConv.td b/lib/Target/Sparc/SparcCallingConv.td
index d4712208126f..b38ac616dcf4 100644
--- a/lib/Target/Sparc/SparcCallingConv.td
+++ b/lib/Target/Sparc/SparcCallingConv.td
@@ -22,6 +22,14 @@ def RetCC_Sparc32 : CallingConv<[
   CCIfType<[f64], CCAssignToReg<[D0, D1]>>
 ]>;
 
+// Sparc 64-bit C return-value convention.
+def RetCC_Sparc64 : CallingConv<[
+  CCIfType<[i32], CCPromoteToType<i64>>,
+  CCIfType<[i64], CCAssignToReg<[I0, I1, I2, I3, I4, I5]>>,
+  CCIfType<[f32], CCAssignToReg<[F0, F1, F2, F3]>>,
+  CCIfType<[f64], CCAssignToReg<[D0, D1]>>
+]>;
+
 // Sparc 32-bit C Calling convention.
 def CC_Sparc32 : CallingConv<[
   //Custom assign SRet to [sp+64].
@@ -34,3 +42,15 @@ def CC_Sparc32 : CallingConv<[
   // Alternatively, they are assigned to the stack in 4-byte aligned units.
   CCAssignToStack<4, 4>
 ]>;
+
+// Sparc 64-bit C Calling convention.
+def CC_Sparc64 : CallingConv<[
+  // All integers are promoted to i64 by the caller.
+  CCIfType<[i32], CCPromoteToType<i64>>,
+  // Integer arguments get passed in integer registers if there is space.
+  CCIfType<[i64], CCAssignToReg<[I0, I1, I2, I3, I4, I5]>>,
+  // FIXME: Floating point arguments.
+
+  // Alternatively, they are assigned to the stack in 8-byte aligned units.
+  CCAssignToStack<8, 8>
+]>;
diff --git a/lib/Target/Sparc/SparcFrameLowering.cpp b/lib/Target/Sparc/SparcFrameLowering.cpp
index 716c79f43a26..a0dae6e9480c 100644
--- a/lib/Target/Sparc/SparcFrameLowering.cpp
+++ b/lib/Target/Sparc/SparcFrameLowering.cpp
@@ -14,15 +14,15 @@
 #include "SparcFrameLowering.h"
 #include "SparcInstrInfo.h"
 #include "SparcMachineFunctionInfo.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
@@ -67,6 +67,22 @@ void SparcFrameLowering::emitPrologue(MachineFunction &MF) const {
   }
 }
 
+void SparcFrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  MachineInstr &MI = *I;
+  DebugLoc dl = MI.getDebugLoc();
+  int Size = MI.getOperand(0).getImm();
+  if (MI.getOpcode() == SP::ADJCALLSTACKDOWN)
+    Size = -Size;
+  const SparcInstrInfo &TII =
+    *static_cast<const SparcInstrInfo*>(MF.getTarget().getInstrInfo());
+  if (Size)
+    BuildMI(MBB, I, dl, TII.get(SP::ADDri), SP::O6).addReg(SP::O6).addImm(Size);
+  MBB.erase(I);
+}
+
+
 void SparcFrameLowering::emitEpilogue(MachineFunction &MF,
                                   MachineBasicBlock &MBB) const {
   MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
diff --git a/lib/Target/Sparc/SparcFrameLowering.h b/lib/Target/Sparc/SparcFrameLowering.h
index 6b593c95bb10..464233e7da35 100644
--- a/lib/Target/Sparc/SparcFrameLowering.h
+++ b/lib/Target/Sparc/SparcFrameLowering.h
@@ -32,6 +32,10 @@ public:
   void emitPrologue(MachineFunction &MF) const;
   void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
 
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator I) const;
+
   bool hasFP(const MachineFunction &MF) const { return false; }
 };
 
diff --git a/lib/Target/Sparc/SparcISelDAGToDAG.cpp b/lib/Target/Sparc/SparcISelDAGToDAG.cpp
index 93710c4e0b0f..5fa545d30160 100644
--- a/lib/Target/Sparc/SparcISelDAGToDAG.cpp
+++ b/lib/Target/Sparc/SparcISelDAGToDAG.cpp
@@ -12,8 +12,8 @@
 //===----------------------------------------------------------------------===//
 
 #include "SparcTargetMachine.h"
-#include "llvm/Intrinsics.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
diff --git a/lib/Target/Sparc/SparcISelLowering.cpp b/lib/Target/Sparc/SparcISelLowering.cpp
index 8e5619e6bc8d..325f13424b42 100644
--- a/lib/Target/Sparc/SparcISelLowering.cpp
+++ b/lib/Target/Sparc/SparcISelLowering.cpp
@@ -13,11 +13,8 @@
 //===----------------------------------------------------------------------===//
 
 #include "SparcISelLowering.h"
-#include "SparcTargetMachine.h"
 #include "SparcMachineFunctionInfo.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Module.h"
+#include "SparcTargetMachine.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -25,6 +22,9 @@
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/ErrorHandling.h"
 using namespace llvm;
 
@@ -93,17 +93,13 @@ SparcTargetLowering::LowerReturn(SDValue Chain,
                  DAG.getTarget(), RVLocs, *DAG.getContext());
 
   // Analize return values.
-  CCInfo.AnalyzeReturn(Outs, RetCC_Sparc32);
-
-  // If this is the first return lowered for this function, add the regs to the
-  // liveout set for the function.
-  if (MF.getRegInfo().liveout_empty()) {
-    for (unsigned i = 0; i != RVLocs.size(); ++i)
-      if (RVLocs[i].isRegLoc())
-        MF.getRegInfo().addLiveOut(RVLocs[i].getLocReg());
-  }
+  CCInfo.AnalyzeReturn(Outs, Subtarget->is64Bit() ?
+                             RetCC_Sparc64 : RetCC_Sparc32);
 
   SDValue Flag;
+  SmallVector<SDValue, 4> RetOps(1, Chain);
+  // Make room for the return address offset.
+  RetOps.push_back(SDValue());
 
   // Copy the result values into the output registers.
   for (unsigned i = 0; i != RVLocs.size(); ++i) {
@@ -115,6 +111,7 @@ SparcTargetLowering::LowerReturn(SDValue Chain,
 
     // Guarantee that all emitted copies are stuck together with flags.
     Flag = Chain.getValue(1);
+    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
   }
 
   unsigned RetAddrOffset = 8; //Call Inst + Delay Slot
@@ -127,32 +124,47 @@ SparcTargetLowering::LowerReturn(SDValue Chain,
     SDValue Val = DAG.getCopyFromReg(Chain, dl, Reg, getPointerTy());
     Chain = DAG.getCopyToReg(Chain, dl, SP::I0, Val, Flag);
     Flag = Chain.getValue(1);
-    if (MF.getRegInfo().liveout_empty())
-      MF.getRegInfo().addLiveOut(SP::I0);
+    RetOps.push_back(DAG.getRegister(SP::I0, getPointerTy()));
     RetAddrOffset = 12; // CallInst + Delay Slot + Unimp
   }
 
-  SDValue RetAddrOffsetNode = DAG.getConstant(RetAddrOffset, MVT::i32);
+  RetOps[0] = Chain;  // Update chain.
+  RetOps[1] = DAG.getConstant(RetAddrOffset, MVT::i32);
 
+  // Add the flag if we have it.
   if (Flag.getNode())
-    return DAG.getNode(SPISD::RET_FLAG, dl, MVT::Other, Chain,
-                       RetAddrOffsetNode, Flag);
-  return DAG.getNode(SPISD::RET_FLAG, dl, MVT::Other, Chain,
-                     RetAddrOffsetNode);
+    RetOps.push_back(Flag);
+
+  return DAG.getNode(SPISD::RET_FLAG, dl, MVT::Other,
+                     &RetOps[0], RetOps.size());
 }
 
-/// LowerFormalArguments - V8 uses a very simple ABI, where all values are
+SDValue SparcTargetLowering::
+LowerFormalArguments(SDValue Chain,
+                     CallingConv::ID CallConv,
+                     bool IsVarArg,
+                     const SmallVectorImpl<ISD::InputArg> &Ins,
+                     DebugLoc DL,
+                     SelectionDAG &DAG,
+                     SmallVectorImpl<SDValue> &InVals) const {
+  if (Subtarget->is64Bit())
+    return LowerFormalArguments_64(Chain, CallConv, IsVarArg, Ins,
+                                   DL, DAG, InVals);
+  return LowerFormalArguments_32(Chain, CallConv, IsVarArg, Ins,
+                                 DL, DAG, InVals);
+}
+
+/// LowerFormalArguments32 - V8 uses a very simple ABI, where all values are
 /// passed in either one or two GPRs, including FP values.  TODO: we should
 /// pass FP values in FP registers for fastcc functions.
-SDValue
-SparcTargetLowering::LowerFormalArguments(SDValue Chain,
-                                          CallingConv::ID CallConv, bool isVarArg,
-                                          const SmallVectorImpl<ISD::InputArg>
-                                            &Ins,
-                                          DebugLoc dl, SelectionDAG &DAG,
-                                          SmallVectorImpl<SDValue> &InVals)
-                                            const {
-
+SDValue SparcTargetLowering::
+LowerFormalArguments_32(SDValue Chain,
+                        CallingConv::ID CallConv,
+                        bool isVarArg,
+                        const SmallVectorImpl<ISD::InputArg> &Ins,
+                        DebugLoc dl,
+                        SelectionDAG &DAG,
+                        SmallVectorImpl<SDValue> &InVals) const {
   MachineFunction &MF = DAG.getMachineFunction();
   MachineRegisterInfo &RegInfo = MF.getRegInfo();
   SparcMachineFunctionInfo *FuncInfo = MF.getInfo<SparcMachineFunctionInfo>();
@@ -344,6 +356,63 @@ SparcTargetLowering::LowerFormalArguments(SDValue Chain,
   return Chain;
 }
 
+// Lower formal arguments for the 64 bit ABI.
+SDValue SparcTargetLowering::
+LowerFormalArguments_64(SDValue Chain,
+                        CallingConv::ID CallConv,
+                        bool IsVarArg,
+                        const SmallVectorImpl<ISD::InputArg> &Ins,
+                        DebugLoc DL,
+                        SelectionDAG &DAG,
+                        SmallVectorImpl<SDValue> &InVals) const {
+  MachineFunction &MF = DAG.getMachineFunction();
+
+  // Analyze arguments according to CC_Sparc64.
+  SmallVector<CCValAssign, 16> ArgLocs;
+  CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(),
+                 getTargetMachine(), ArgLocs, *DAG.getContext());
+  CCInfo.AnalyzeFormalArguments(Ins, CC_Sparc64);
+
+  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
+    CCValAssign &VA = ArgLocs[i];
+    if (VA.isRegLoc()) {
+      // This argument is passed in a register.
+      // All integer register arguments are promoted by the caller to i64.
+
+      // Create a virtual register for the promoted live-in value.
+      unsigned VReg = MF.addLiveIn(VA.getLocReg(),
+                                   getRegClassFor(VA.getLocVT()));
+      SDValue Arg = DAG.getCopyFromReg(Chain, DL, VReg, VA.getLocVT());
+
+      // The caller promoted the argument, so insert an Assert?ext SDNode so we
+      // won't promote the value again in this function.
+      switch (VA.getLocInfo()) {
+      case CCValAssign::SExt:
+        Arg = DAG.getNode(ISD::AssertSext, DL, VA.getLocVT(), Arg,
+                          DAG.getValueType(VA.getValVT()));
+        break;
+      case CCValAssign::ZExt:
+        Arg = DAG.getNode(ISD::AssertZext, DL, VA.getLocVT(), Arg,
+                          DAG.getValueType(VA.getValVT()));
+        break;
+      default:
+        break;
+      }
+
+      // Truncate the register down to the argument type.
+      if (VA.isExtInLoc())
+        Arg = DAG.getNode(ISD::TRUNCATE, DL, VA.getValVT(), Arg);
+
+      InVals.push_back(Arg);
+      continue;
+    }
+
+    // The registers are exhausted. This argument was passed on the stack.
+    assert(VA.isMemLoc());
+  }
+  return Chain;
+}
+
 SDValue
 SparcTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                                SmallVectorImpl<SDValue> &InVals) const {
@@ -692,11 +761,14 @@ static SPCC::CondCodes FPCondCCodeToFCC(ISD::CondCode CC) {
 
 SparcTargetLowering::SparcTargetLowering(TargetMachine &TM)
   : TargetLowering(TM, new TargetLoweringObjectFileELF()) {
+  Subtarget = &TM.getSubtarget<SparcSubtarget>();
 
   // Set up the register classes.
   addRegisterClass(MVT::i32, &SP::IntRegsRegClass);
   addRegisterClass(MVT::f32, &SP::FPRegsRegClass);
   addRegisterClass(MVT::f64, &SP::DFPRegsRegClass);
+  if (Subtarget->is64Bit())
+    addRegisterClass(MVT::i64, &SP::I64RegsRegClass);
 
   // Turn FP extload into load/fextend
   setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
@@ -752,6 +824,11 @@ SparcTargetLowering::SparcTargetLowering(TargetMachine &TM)
   setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
   setOperationAction(ISD::SELECT_CC, MVT::f64, Custom);
 
+  if (Subtarget->is64Bit()) {
+    setOperationAction(ISD::BR_CC, MVT::i64, Custom);
+    setOperationAction(ISD::SELECT_CC, MVT::i64, Custom);
+  }
+
   // FIXME: There are instructions available for ATOMIC_FENCE
   // on SparcV8 and later.
   setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand);
@@ -759,10 +836,12 @@ SparcTargetLowering::SparcTargetLowering(TargetMachine &TM)
 
   setOperationAction(ISD::FSIN , MVT::f64, Expand);
   setOperationAction(ISD::FCOS , MVT::f64, Expand);
+  setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
   setOperationAction(ISD::FREM , MVT::f64, Expand);
   setOperationAction(ISD::FMA  , MVT::f64, Expand);
   setOperationAction(ISD::FSIN , MVT::f32, Expand);
   setOperationAction(ISD::FCOS , MVT::f32, Expand);
+  setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
   setOperationAction(ISD::FREM , MVT::f32, Expand);
   setOperationAction(ISD::FMA  , MVT::f32, Expand);
   setOperationAction(ISD::CTPOP, MVT::i32, Expand);
@@ -819,8 +898,10 @@ const char *SparcTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case SPISD::CMPICC:     return "SPISD::CMPICC";
   case SPISD::CMPFCC:     return "SPISD::CMPFCC";
   case SPISD::BRICC:      return "SPISD::BRICC";
+  case SPISD::BRXCC:      return "SPISD::BRXCC";
   case SPISD::BRFCC:      return "SPISD::BRFCC";
   case SPISD::SELECT_ICC: return "SPISD::SELECT_ICC";
+  case SPISD::SELECT_XCC: return "SPISD::SELECT_XCC";
   case SPISD::SELECT_FCC: return "SPISD::SELECT_FCC";
   case SPISD::Hi:         return "SPISD::Hi";
   case SPISD::Lo:         return "SPISD::Lo";
@@ -847,6 +928,7 @@ void SparcTargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
   switch (Op.getOpcode()) {
   default: break;
   case SPISD::SELECT_ICC:
+  case SPISD::SELECT_XCC:
   case SPISD::SELECT_FCC:
     DAG.ComputeMaskedBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1);
     DAG.ComputeMaskedBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1);
@@ -867,7 +949,8 @@ static void LookThroughSetCC(SDValue &LHS, SDValue &RHS,
   if (isa<ConstantSDNode>(RHS) &&
       cast<ConstantSDNode>(RHS)->isNullValue() &&
       CC == ISD::SETNE &&
-      ((LHS.getOpcode() == SPISD::SELECT_ICC &&
+      (((LHS.getOpcode() == SPISD::SELECT_ICC ||
+         LHS.getOpcode() == SPISD::SELECT_XCC) &&
         LHS.getOperand(3).getOpcode() == SPISD::CMPICC) ||
        (LHS.getOpcode() == SPISD::SELECT_FCC &&
         LHS.getOperand(3).getOpcode() == SPISD::CMPFCC)) &&
@@ -955,14 +1038,13 @@ static SDValue LowerBR_CC(SDValue Op, SelectionDAG &DAG) {
 
   // Get the condition flag.
   SDValue CompareFlag;
-  if (LHS.getValueType() == MVT::i32) {
-    std::vector<EVT> VTs;
-    VTs.push_back(MVT::i32);
-    VTs.push_back(MVT::Glue);
+  if (LHS.getValueType().isInteger()) {
+    EVT VTs[] = { LHS.getValueType(), MVT::Glue };
     SDValue Ops[2] = { LHS, RHS };
     CompareFlag = DAG.getNode(SPISD::CMPICC, dl, VTs, Ops, 2).getValue(1);
     if (SPCC == ~0U) SPCC = IntCondCCodeToICC(CC);
-    Opc = SPISD::BRICC;
+    // 32-bit compares use the icc flags, 64-bit uses the xcc flags.
+    Opc = LHS.getValueType() == MVT::i32 ? SPISD::BRICC : SPISD::BRXCC;
   } else {
     CompareFlag = DAG.getNode(SPISD::CMPFCC, dl, MVT::Glue, LHS, RHS);
     if (SPCC == ~0U) SPCC = FPCondCCodeToFCC(CC);
@@ -986,13 +1068,13 @@ static SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) {
   LookThroughSetCC(LHS, RHS, CC, SPCC);
 
   SDValue CompareFlag;
-  if (LHS.getValueType() == MVT::i32) {
-    std::vector<EVT> VTs;
-    VTs.push_back(LHS.getValueType());   // subcc returns a value
-    VTs.push_back(MVT::Glue);
+  if (LHS.getValueType().isInteger()) {
+    // subcc returns a value
+    EVT VTs[] = { LHS.getValueType(), MVT::Glue };
     SDValue Ops[2] = { LHS, RHS };
     CompareFlag = DAG.getNode(SPISD::CMPICC, dl, VTs, Ops, 2).getValue(1);
-    Opc = SPISD::SELECT_ICC;
+    Opc = LHS.getValueType() == MVT::i32 ?
+          SPISD::SELECT_ICC : SPISD::SELECT_XCC;
     if (SPCC == ~0U) SPCC = IntCondCCodeToICC(CC);
   } else {
     CompareFlag = DAG.getNode(SPISD::CMPFCC, dl, MVT::Glue, LHS, RHS);
diff --git a/lib/Target/Sparc/SparcISelLowering.h b/lib/Target/Sparc/SparcISelLowering.h
index 09148ea54027..aa2ef711a080 100644
--- a/lib/Target/Sparc/SparcISelLowering.h
+++ b/lib/Target/Sparc/SparcISelLowering.h
@@ -19,14 +19,18 @@
 #include "llvm/Target/TargetLowering.h"
 
 namespace llvm {
+  class SparcSubtarget;
+
   namespace SPISD {
     enum {
       FIRST_NUMBER = ISD::BUILTIN_OP_END,
-      CMPICC,      // Compare two GPR operands, set icc.
+      CMPICC,      // Compare two GPR operands, set icc+xcc.
       CMPFCC,      // Compare two FP operands, set fcc.
       BRICC,       // Branch to dest on icc condition
+      BRXCC,       // Branch to dest on xcc condition (64-bit only).
       BRFCC,       // Branch to dest on fcc condition
       SELECT_ICC,  // Select between two values using the current ICC flags.
+      SELECT_XCC,  // Select between two values using the current XCC flags.
       SELECT_FCC,  // Select between two values using the current FCC flags.
 
       Hi, Lo,      // Hi/Lo operations, typically on a global address.
@@ -42,6 +46,7 @@ namespace llvm {
   }
 
   class SparcTargetLowering : public TargetLowering {
+    const SparcSubtarget *Subtarget;
   public:
     SparcTargetLowering(TargetMachine &TM);
     virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
@@ -74,6 +79,18 @@ namespace llvm {
                            const SmallVectorImpl<ISD::InputArg> &Ins,
                            DebugLoc dl, SelectionDAG &DAG,
                            SmallVectorImpl<SDValue> &InVals) const;
+    SDValue LowerFormalArguments_32(SDValue Chain,
+                                    CallingConv::ID CallConv,
+                                    bool isVarArg,
+                                    const SmallVectorImpl<ISD::InputArg> &Ins,
+                                    DebugLoc dl, SelectionDAG &DAG,
+                                    SmallVectorImpl<SDValue> &InVals) const;
+    SDValue LowerFormalArguments_64(SDValue Chain,
+                                    CallingConv::ID CallConv,
+                                    bool isVarArg,
+                                    const SmallVectorImpl<ISD::InputArg> &Ins,
+                                    DebugLoc dl, SelectionDAG &DAG,
+                                    SmallVectorImpl<SDValue> &InVals) const;
 
     virtual SDValue
       LowerCall(TargetLowering::CallLoweringInfo &CLI,
diff --git a/lib/Target/Sparc/SparcInstr64Bit.td b/lib/Target/Sparc/SparcInstr64Bit.td
new file mode 100644
index 000000000000..ca1153b3fe8f
--- /dev/null
+++ b/lib/Target/Sparc/SparcInstr64Bit.td
@@ -0,0 +1,285 @@
+//===-- SparcInstr64Bit.td - 64-bit instructions for Sparc Target ---------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains instruction definitions and patterns needed for 64-bit
+// code generation on SPARC v9.
+//
+// Some SPARC v9 instructions are defined in SparcInstrInfo.td because they can
+// also be used in 32-bit code running on a SPARC v9 CPU.
+//
+//===----------------------------------------------------------------------===//
+
+let Predicates = [Is64Bit] in {
+// The same integer registers are used for i32 and i64 values.
+// When registers hold i32 values, the high bits are don't care.
+// This give us free trunc and anyext.
+def : Pat<(i64 (anyext i32:$val)), (COPY_TO_REGCLASS $val, I64Regs)>;
+def : Pat<(i32 (trunc i64:$val)), (COPY_TO_REGCLASS $val, IntRegs)>;
+
+} // Predicates = [Is64Bit]
+
+
+//===----------------------------------------------------------------------===//
+// 64-bit Shift Instructions.
+//===----------------------------------------------------------------------===//
+//
+// The 32-bit shift instructions are still available. The left shift srl
+// instructions shift all 64 bits, but it only accepts a 5-bit shift amount.
+//
+// The srl instructions only shift the low 32 bits and clear the high 32 bits.
+// Finally, sra shifts the low 32 bits and sign-extends to 64 bits.
+
+let Predicates = [Is64Bit] in {
+
+def : Pat<(i64 (zext i32:$val)), (SRLri $val, 0)>;
+def : Pat<(i64 (sext i32:$val)), (SRAri $val, 0)>;
+
+defm SLLX : F3_S<"sllx", 0b100101, 1, shl, i64, I64Regs>;
+defm SRLX : F3_S<"srlx", 0b100110, 1, srl, i64, I64Regs>;
+defm SRAX : F3_S<"srax", 0b100111, 1, sra, i64, I64Regs>;
+
+} // Predicates = [Is64Bit]
+
+
+//===----------------------------------------------------------------------===//
+// 64-bit Immediates.
+//===----------------------------------------------------------------------===//
+//
+// All 32-bit immediates can be materialized with sethi+or, but 64-bit
+// immediates may require more code. There may be a point where it is
+// preferable to use a constant pool load instead, depending on the
+// microarchitecture.
+
+// The %g0 register is constant 0.
+// This is useful for stx %g0, [...], for example.
+def : Pat<(i64 0), (i64 G0)>, Requires<[Is64Bit]>;
+
+// Single-instruction patterns.
+
+// The ALU instructions want their simm13 operands as i32 immediates.
+def as_i32imm : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(N->getSExtValue(), MVT::i32);
+}]>;
+def : Pat<(i64 simm13:$val), (ORri (i64 G0), (as_i32imm $val))>;
+def : Pat<(i64 SETHIimm:$val), (SETHIi (HI22 $val))>;
+
+// Double-instruction patterns.
+
+// All unsigned i32 immediates can be handled by sethi+or.
+def uimm32 : PatLeaf<(imm), [{ return isUInt<32>(N->getZExtValue()); }]>;
+def : Pat<(i64 uimm32:$val), (ORri (SETHIi (HI22 $val)), (LO10 $val))>,
+      Requires<[Is64Bit]>;
+
+// All negative i33 immediates can be handled by sethi+xor.
+def nimm33 : PatLeaf<(imm), [{
+  int64_t Imm = N->getSExtValue();
+  return Imm < 0 && isInt<33>(Imm);
+}]>;
+// Bits 10-31 inverted. Same as assembler's %hix.
+def HIX22 : SDNodeXForm<imm, [{
+  uint64_t Val = (~N->getZExtValue() >> 10) & ((1u << 22) - 1);
+  return CurDAG->getTargetConstant(Val, MVT::i32);
+}]>;
+// Bits 0-9 with ones in bits 10-31. Same as assembler's %lox.
+def LOX10 : SDNodeXForm<imm, [{
+  return CurDAG->getTargetConstant(~(~N->getZExtValue() & 0x3ff), MVT::i32);
+}]>;
+def : Pat<(i64 nimm33:$val), (XORri (SETHIi (HIX22 $val)), (LOX10 $val))>,
+      Requires<[Is64Bit]>;
+
+// More possible patterns:
+//
+//   (sllx sethi, n)
+//   (sllx simm13, n)
+//
+// 3 instrs:
+//
+//   (xor (sllx sethi), simm13)
+//   (sllx (xor sethi, simm13))
+//
+// 4 instrs:
+//
+//   (or sethi, (sllx sethi))
+//   (xnor sethi, (sllx sethi))
+//
+// 5 instrs:
+//
+//   (or (sllx sethi), (or sethi, simm13))
+//   (xnor (sllx sethi), (or sethi, simm13))
+//   (or (sllx sethi), (sllx sethi))
+//   (xnor (sllx sethi), (sllx sethi))
+//
+// Worst case is 6 instrs:
+//
+//   (or (sllx (or sethi, simmm13)), (or sethi, simm13))
+
+// Bits 42-63, same as assembler's %hh.
+def HH22 : SDNodeXForm<imm, [{
+  uint64_t Val = (N->getZExtValue() >> 42) & ((1u << 22) - 1);
+  return CurDAG->getTargetConstant(Val, MVT::i32);
+}]>;
+// Bits 32-41, same as assembler's %hm.
+def HM10 : SDNodeXForm<imm, [{
+  uint64_t Val = (N->getZExtValue() >> 32) & ((1u << 10) - 1);
+  return CurDAG->getTargetConstant(Val, MVT::i32);
+}]>;
+def : Pat<(i64 imm:$val),
+          (ORrr (SLLXri (ORri (SETHIi (HH22 $val)), (HM10 $val)), (i64 32)),
+                (ORri (SETHIi (HI22 $val)), (LO10 $val)))>,
+      Requires<[Is64Bit]>;
+
+
+//===----------------------------------------------------------------------===//
+// 64-bit Integer Arithmetic and Logic.
+//===----------------------------------------------------------------------===//
+
+let Predicates = [Is64Bit] in {
+
+// Register-register instructions.
+
+def : Pat<(and i64:$a, i64:$b), (ANDrr $a, $b)>;
+def : Pat<(or  i64:$a, i64:$b), (ORrr  $a, $b)>;
+def : Pat<(xor i64:$a, i64:$b), (XORrr $a, $b)>;
+
+def : Pat<(and i64:$a, (not i64:$b)), (ANDNrr $a, $b)>;
+def : Pat<(or  i64:$a, (not i64:$b)), (ORNrr  $a, $b)>;
+def : Pat<(xor i64:$a, (not i64:$b)), (XNORrr $a, $b)>;
+
+def : Pat<(add i64:$a, i64:$b), (ADDrr $a, $b)>;
+def : Pat<(sub i64:$a, i64:$b), (SUBrr $a, $b)>;
+
+// Add/sub with carry were renamed to addc/subc in SPARC v9.
+def : Pat<(adde i64:$a, i64:$b), (ADDXrr $a, $b)>;
+def : Pat<(sube i64:$a, i64:$b), (SUBXrr $a, $b)>;
+
+def : Pat<(addc i64:$a, i64:$b), (ADDCCrr $a, $b)>;
+def : Pat<(subc i64:$a, i64:$b), (SUBCCrr $a, $b)>;
+
+def : Pat<(SPcmpicc i64:$a, i64:$b), (SUBCCrr $a, $b)>;
+
+// Register-immediate instructions.
+
+def : Pat<(and i64:$a, (i64 simm13:$b)), (ANDri $a, (as_i32imm $b))>;
+def : Pat<(or  i64:$a, (i64 simm13:$b)), (ORri  $a, (as_i32imm $b))>;
+def : Pat<(xor i64:$a, (i64 simm13:$b)), (XORri $a, (as_i32imm $b))>;
+
+def : Pat<(add i64:$a, (i64 simm13:$b)), (ADDri $a, (as_i32imm $b))>;
+def : Pat<(sub i64:$a, (i64 simm13:$b)), (SUBri $a, (as_i32imm $b))>;
+
+def : Pat<(SPcmpicc i64:$a, (i64 simm13:$b)), (SUBCCri $a, (as_i32imm $b))>;
+
+} // Predicates = [Is64Bit]
+
+
+//===----------------------------------------------------------------------===//
+// 64-bit Loads and Stores.
+//===----------------------------------------------------------------------===//
+//
+// All the 32-bit loads and stores are available. The extending loads are sign
+// or zero-extending to 64 bits. The LDrr and LDri instructions load 32 bits
+// zero-extended to i64. Their mnemonic is lduw in SPARC v9 (Load Unsigned
+// Word).
+//
+// SPARC v9 adds 64-bit loads as well as a sign-extending ldsw i32 loads.
+
+let Predicates = [Is64Bit] in {
+
+// 64-bit loads.
+def LDXrr  : F3_1<3, 0b001011,
+                  (outs I64Regs:$dst), (ins MEMrr:$addr),
+                  "ldx [$addr], $dst",
+                  [(set i64:$dst, (load ADDRrr:$addr))]>;
+def LDXri  : F3_2<3, 0b001011,
+                  (outs I64Regs:$dst), (ins MEMri:$addr),
+                  "ldx [$addr], $dst",
+                  [(set i64:$dst, (load ADDRri:$addr))]>;
+
+// Extending loads to i64.
+def : Pat<(i64 (zextloadi8 ADDRrr:$addr)), (LDUBrr ADDRrr:$addr)>;
+def : Pat<(i64 (zextloadi8 ADDRri:$addr)), (LDUBri ADDRri:$addr)>;
+def : Pat<(i64 (sextloadi8 ADDRrr:$addr)), (LDSBrr ADDRrr:$addr)>;
+def : Pat<(i64 (sextloadi8 ADDRri:$addr)), (LDSBri ADDRri:$addr)>;
+
+def : Pat<(i64 (zextloadi16 ADDRrr:$addr)), (LDUHrr ADDRrr:$addr)>;
+def : Pat<(i64 (zextloadi16 ADDRri:$addr)), (LDUHri ADDRri:$addr)>;
+def : Pat<(i64 (sextloadi16 ADDRrr:$addr)), (LDSHrr ADDRrr:$addr)>;
+def : Pat<(i64 (sextloadi16 ADDRri:$addr)), (LDSHri ADDRri:$addr)>;
+
+def : Pat<(i64 (zextloadi32 ADDRrr:$addr)), (LDrr ADDRrr:$addr)>;
+def : Pat<(i64 (zextloadi32 ADDRri:$addr)), (LDri ADDRri:$addr)>;
+
+// Sign-extending load of i32 into i64 is a new SPARC v9 instruction.
+def LDSWrr : F3_1<3, 0b001011,
+                 (outs I64Regs:$dst), (ins MEMrr:$addr),
+                 "ldsw [$addr], $dst",
+                 [(set i64:$dst, (sextloadi32 ADDRrr:$addr))]>;
+def LDSWri : F3_2<3, 0b001011,
+                 (outs I64Regs:$dst), (ins MEMri:$addr),
+                 "ldsw [$addr], $dst",
+                 [(set i64:$dst, (sextloadi32 ADDRri:$addr))]>;
+
+// 64-bit stores.
+def STXrr  : F3_1<3, 0b001110,
+                 (outs), (ins MEMrr:$addr, I64Regs:$src),
+                 "stx $src, [$addr]",
+                 [(store i64:$src, ADDRrr:$addr)]>;
+def STXri  : F3_2<3, 0b001110,
+                 (outs), (ins MEMri:$addr, I64Regs:$src),
+                 "stx $src, [$addr]",
+                 [(store i64:$src, ADDRri:$addr)]>;
+
+// Truncating stores from i64 are identical to the i32 stores.
+def : Pat<(truncstorei8  i64:$src, ADDRrr:$addr), (STBrr ADDRrr:$addr, $src)>;
+def : Pat<(truncstorei8  i64:$src, ADDRri:$addr), (STBri ADDRri:$addr, $src)>;
+def : Pat<(truncstorei16 i64:$src, ADDRrr:$addr), (STHrr ADDRrr:$addr, $src)>;
+def : Pat<(truncstorei16 i64:$src, ADDRri:$addr), (STHri ADDRri:$addr, $src)>;
+def : Pat<(truncstorei32 i64:$src, ADDRrr:$addr), (STrr  ADDRrr:$addr, $src)>;
+def : Pat<(truncstorei32 i64:$src, ADDRri:$addr), (STri  ADDRri:$addr, $src)>;
+
+} // Predicates = [Is64Bit]
+
+
+//===----------------------------------------------------------------------===//
+// 64-bit Conditionals.
+//===----------------------------------------------------------------------===//
+//
+// Flag-setting instructions like subcc and addcc set both icc and xcc flags.
+// The icc flags correspond to the 32-bit result, and the xcc are for the
+// full 64-bit result.
+//
+// We reuse CMPICC SDNodes for compares, but use new BRXCC branch nodes for
+// 64-bit compares. See LowerBR_CC.
+
+let Predicates = [Is64Bit] in {
+
+let Uses = [ICC] in
+def BPXCC : BranchSP<0, (ins brtarget:$dst, CCOp:$cc),
+                     "bp$cc %xcc, $dst",
+                     [(SPbrxcc bb:$dst, imm:$cc)]>;
+
+// Conditional moves on %xcc.
+let Uses = [ICC], Constraints = "$f = $rd" in {
+def MOVXCCrr : Pseudo<(outs IntRegs:$rd),
+                      (ins IntRegs:$rs2, IntRegs:$f, CCOp:$cond),
+                      "mov$cond %xcc, $rs2, $rd",
+                      [(set i32:$rd,
+                       (SPselectxcc i32:$rs2, i32:$f, imm:$cond))]>;
+def MOVXCCri : Pseudo<(outs IntRegs:$rd),
+                      (ins i32imm:$i, IntRegs:$f, CCOp:$cond),
+                      "mov$cond %xcc, $i, $rd",
+                      [(set i32:$rd,
+                       (SPselecticc simm11:$i, i32:$f, imm:$cond))]>;
+} // Uses, Constraints
+
+def : Pat<(SPselectxcc i64:$t, i64:$f, imm:$cond),
+          (MOVXCCrr $t, $f, imm:$cond)>;
+def : Pat<(SPselectxcc (i64 simm11:$t), i64:$f, imm:$cond),
+          (MOVXCCri (as_i32imm $t), $f, imm:$cond)>;
+
+} // Predicates = [Is64Bit]
diff --git a/lib/Target/Sparc/SparcInstrFormats.td b/lib/Target/Sparc/SparcInstrFormats.td
index dce331228b8f..f1018569153c 100644
--- a/lib/Target/Sparc/SparcInstrFormats.td
+++ b/lib/Target/Sparc/SparcInstrFormats.td
@@ -111,4 +111,41 @@ class F3_3<bits<2> opVal, bits<6> op3val, bits<9> opfval, dag outs, dag ins,
   let Inst{4-0}  = rs2;
 }
 
+// Shift by register rs2.
+class F3_Sr<bits<2> opVal, bits<6> op3val, bit xVal, dag outs, dag ins,
+            string asmstr, list<dag> pattern> : F3<outs, ins, asmstr, pattern> {
+  bit x = xVal;           // 1 for 64-bit shifts.
+  bits<5> rs2;
+
+  let op         = opVal;
+  let op3        = op3val;
+
+  let Inst{13}   = 0;     // i field = 0
+  let Inst{12}   = x;     // extended registers.
+  let Inst{4-0}  = rs2;
+}
 
+// Shift by immediate.
+class F3_Si<bits<2> opVal, bits<6> op3val, bit xVal, dag outs, dag ins,
+            string asmstr, list<dag> pattern> : F3<outs, ins, asmstr, pattern> {
+  bit x = xVal;           // 1 for 64-bit shifts.
+  bits<6> shcnt;          // shcnt32 / shcnt64.
+
+  let op         = opVal;
+  let op3        = op3val;
+
+  let Inst{13}   = 1;     // i field = 1
+  let Inst{12}   = x;     // extended registers.
+  let Inst{5-0}  = shcnt;
+}
+
+// Define rr and ri shift instructions with patterns.
+multiclass F3_S<string OpcStr, bits<6> Op3Val, bit XVal, SDNode OpNode,
+                ValueType VT, RegisterClass RC> {
+  def rr : F3_Sr<2, Op3Val, XVal, (outs RC:$rd), (ins RC:$rs, RC:$rs2),
+                 !strconcat(OpcStr, " $rs, $rs2, $rd"),
+                 [(set VT:$rd, (OpNode VT:$rs, VT:$rs2))]>;
+  def ri : F3_Si<2, Op3Val, XVal, (outs RC:$rd), (ins RC:$rs, unknown:$shcnt),
+                 !strconcat(OpcStr, " $rs, $shcnt, $rd"),
+                 [(set VT:$rd, (OpNode VT:$rs, (VT imm:$shcnt)))]>;
+}
diff --git a/lib/Target/Sparc/SparcInstrInfo.cpp b/lib/Target/Sparc/SparcInstrInfo.cpp
index f8674d0bd660..39d7329f2663 100644
--- a/lib/Target/Sparc/SparcInstrInfo.cpp
+++ b/lib/Target/Sparc/SparcInstrInfo.cpp
@@ -15,12 +15,12 @@
 #include "Sparc.h"
 #include "SparcMachineFunctionInfo.h"
 #include "SparcSubtarget.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/SmallVector.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
-#include "llvm/ADT/STLExtras.h"
-#include "llvm/ADT/SmallVector.h"
 
 #define GET_INSTRINFO_CTOR
 #include "SparcGenInstrInfo.inc"
diff --git a/lib/Target/Sparc/SparcInstrInfo.td b/lib/Target/Sparc/SparcInstrInfo.td
index e64c140e4921..5ff439583c5c 100644
--- a/lib/Target/Sparc/SparcInstrInfo.td
+++ b/lib/Target/Sparc/SparcInstrInfo.td
@@ -21,6 +21,12 @@ include "SparcInstrFormats.td"
 // Feature predicates.
 //===----------------------------------------------------------------------===//
 
+// True when generating 32-bit code.
+def Is32Bit : Predicate<"!Subtarget.is64Bit()">;
+
+// True when generating 64-bit code. This also implies HasV9.
+def Is64Bit : Predicate<"Subtarget.is64Bit()">;
+
 // HasV9 - This predicate is true when the target processor supports V9
 // instructions.  Note that the machine may be running in 32-bit mode.
 def HasV9   : Predicate<"Subtarget.isV9()">;
@@ -63,17 +69,17 @@ def SETHIimm : PatLeaf<(imm), [{
 }], HI22>;
 
 // Addressing modes.
-def ADDRrr : ComplexPattern<i32, 2, "SelectADDRrr", [], []>;
-def ADDRri : ComplexPattern<i32, 2, "SelectADDRri", [frameindex], []>;
+def ADDRrr : ComplexPattern<iPTR, 2, "SelectADDRrr", [], []>;
+def ADDRri : ComplexPattern<iPTR, 2, "SelectADDRri", [frameindex], []>;
 
 // Address operands
-def MEMrr : Operand<i32> {
+def MEMrr : Operand<iPTR> {
   let PrintMethod = "printMemOperand";
-  let MIOperandInfo = (ops IntRegs, IntRegs);
+  let MIOperandInfo = (ops ptr_rc, ptr_rc);
 }
-def MEMri : Operand<i32> {
+def MEMri : Operand<iPTR> {
   let PrintMethod = "printMemOperand";
-  let MIOperandInfo = (ops IntRegs, i32imm);
+  let MIOperandInfo = (ops ptr_rc, i32imm);
 }
 
 // Branch targets have OtherVT type.
@@ -98,6 +104,7 @@ SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisVT<1, f32>]>;
 def SPcmpicc : SDNode<"SPISD::CMPICC", SDTIntBinOp, [SDNPOutGlue]>;
 def SPcmpfcc : SDNode<"SPISD::CMPFCC", SDTSPcmpfcc, [SDNPOutGlue]>;
 def SPbricc : SDNode<"SPISD::BRICC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
+def SPbrxcc : SDNode<"SPISD::BRXCC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
 def SPbrfcc : SDNode<"SPISD::BRFCC", SDTSPbrcc, [SDNPHasChain, SDNPInGlue]>;
 
 def SPhi    : SDNode<"SPISD::Hi", SDTIntUnaryOp>;
@@ -107,6 +114,7 @@ def SPftoi  : SDNode<"SPISD::FTOI", SDTSPFTOI>;
 def SPitof  : SDNode<"SPISD::ITOF", SDTSPITOF>;
 
 def SPselecticc : SDNode<"SPISD::SELECT_ICC", SDTSPselectcc, [SDNPInGlue]>;
+def SPselectxcc : SDNode<"SPISD::SELECT_XCC", SDTSPselectcc, [SDNPInGlue]>;
 def SPselectfcc : SDNode<"SPISD::SELECT_FCC", SDTSPselectcc, [SDNPInGlue]>;
 
 //  These are target-independent nodes, but have target-specific formats.
@@ -126,7 +134,7 @@ def call          : SDNode<"SPISD::CALL", SDT_SPCall,
 
 def SDT_SPRet     : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
 def retflag       : SDNode<"SPISD::RET_FLAG", SDT_SPRet,
-                           [SDNPHasChain, SDNPOptInGlue]>;
+                           [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
 
 def flushw        : SDNode<"SPISD::FLUSHW", SDTNone,
                            [SDNPHasChain, SDNPSideEffect, SDNPMayStore]>;
@@ -182,11 +190,11 @@ multiclass F3_12<string OpcStr, bits<6> Op3Val, SDNode OpNode> {
   def rr  : F3_1<2, Op3Val, 
                  (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
                  !strconcat(OpcStr, " $b, $c, $dst"),
-                 [(set IntRegs:$dst, (OpNode IntRegs:$b, IntRegs:$c))]>;
+                 [(set i32:$dst, (OpNode i32:$b, i32:$c))]>;
   def ri  : F3_2<2, Op3Val,
                  (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
                  !strconcat(OpcStr, " $b, $c, $dst"),
-                 [(set IntRegs:$dst, (OpNode IntRegs:$b, simm13:$c))]>;
+                 [(set i32:$dst, (OpNode i32:$b, (i32 simm13:$c)))]>;
 }
 
 /// F3_12np multiclass - Define a normal F3_1/F3_2 pattern in one shot, with no
@@ -243,10 +251,10 @@ let Predicates = [HasNoV9] in {  // Only emit these in V8 mode.
                       "!FpMOVD $src, $dst", []>;
   def FpNEGD : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$src),
                       "!FpNEGD $src, $dst",
-                      [(set DFPRegs:$dst, (fneg DFPRegs:$src))]>;
+                      [(set f64:$dst, (fneg f64:$src))]>;
   def FpABSD : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$src),
                       "!FpABSD $src, $dst",
-                      [(set DFPRegs:$dst, (fabs DFPRegs:$src))]>;
+                      [(set f64:$dst, (fabs f64:$src))]>;
 }
 
 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation.  Expanded after
@@ -257,19 +265,16 @@ let Uses = [ICC], usesCustomInserter = 1 in {
   def SELECT_CC_Int_ICC
    : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, i32imm:$Cond),
             "; SELECT_CC_Int_ICC PSEUDO!",
-            [(set IntRegs:$dst, (SPselecticc IntRegs:$T, IntRegs:$F,
-                                             imm:$Cond))]>;
+            [(set i32:$dst, (SPselecticc i32:$T, i32:$F, imm:$Cond))]>;
   def SELECT_CC_FP_ICC
    : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, i32imm:$Cond),
             "; SELECT_CC_FP_ICC PSEUDO!",
-            [(set FPRegs:$dst, (SPselecticc FPRegs:$T, FPRegs:$F,
-                                            imm:$Cond))]>;
+            [(set f32:$dst, (SPselecticc f32:$T, f32:$F, imm:$Cond))]>;
 
   def SELECT_CC_DFP_ICC
    : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, i32imm:$Cond),
             "; SELECT_CC_DFP_ICC PSEUDO!",
-            [(set DFPRegs:$dst, (SPselecticc DFPRegs:$T, DFPRegs:$F,
-                                             imm:$Cond))]>;
+            [(set f64:$dst, (SPselecticc f64:$T, f64:$F, imm:$Cond))]>;
 }
 
 let usesCustomInserter = 1, Uses = [FCC] in {
@@ -277,19 +282,16 @@ let usesCustomInserter = 1, Uses = [FCC] in {
   def SELECT_CC_Int_FCC
    : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, i32imm:$Cond),
             "; SELECT_CC_Int_FCC PSEUDO!",
-            [(set IntRegs:$dst, (SPselectfcc IntRegs:$T, IntRegs:$F,
-                                             imm:$Cond))]>;
+            [(set i32:$dst, (SPselectfcc i32:$T, i32:$F, imm:$Cond))]>;
 
   def SELECT_CC_FP_FCC
    : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, i32imm:$Cond),
             "; SELECT_CC_FP_FCC PSEUDO!",
-            [(set FPRegs:$dst, (SPselectfcc FPRegs:$T, FPRegs:$F,
-                                            imm:$Cond))]>;
+            [(set f32:$dst, (SPselectfcc f32:$T, f32:$F, imm:$Cond))]>;
   def SELECT_CC_DFP_FCC
    : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, i32imm:$Cond),
             "; SELECT_CC_DFP_FCC PSEUDO!",
-            [(set DFPRegs:$dst, (SPselectfcc DFPRegs:$T, DFPRegs:$F,
-                                             imm:$Cond))]>;
+            [(set f64:$dst, (SPselectfcc f64:$T, f64:$F, imm:$Cond))]>;
 }
 
 
@@ -309,111 +311,111 @@ let isReturn = 1, isTerminator = 1, hasDelaySlot = 1, isBarrier = 1 in {
 def LDSBrr : F3_1<3, 0b001001,
                   (outs IntRegs:$dst), (ins MEMrr:$addr),
                   "ldsb [$addr], $dst",
-                  [(set IntRegs:$dst, (sextloadi8 ADDRrr:$addr))]>;
+                  [(set i32:$dst, (sextloadi8 ADDRrr:$addr))]>;
 def LDSBri : F3_2<3, 0b001001,
                   (outs IntRegs:$dst), (ins MEMri:$addr),
                   "ldsb [$addr], $dst",
-                  [(set IntRegs:$dst, (sextloadi8 ADDRri:$addr))]>;
+                  [(set i32:$dst, (sextloadi8 ADDRri:$addr))]>;
 def LDSHrr : F3_1<3, 0b001010,
                   (outs IntRegs:$dst), (ins MEMrr:$addr),
                   "ldsh [$addr], $dst",
-                  [(set IntRegs:$dst, (sextloadi16 ADDRrr:$addr))]>;
+                  [(set i32:$dst, (sextloadi16 ADDRrr:$addr))]>;
 def LDSHri : F3_2<3, 0b001010,
                   (outs IntRegs:$dst), (ins MEMri:$addr),
                   "ldsh [$addr], $dst",
-                  [(set IntRegs:$dst, (sextloadi16 ADDRri:$addr))]>;
+                  [(set i32:$dst, (sextloadi16 ADDRri:$addr))]>;
 def LDUBrr : F3_1<3, 0b000001,
                   (outs IntRegs:$dst), (ins MEMrr:$addr),
                   "ldub [$addr], $dst",
-                  [(set IntRegs:$dst, (zextloadi8 ADDRrr:$addr))]>;
+                  [(set i32:$dst, (zextloadi8 ADDRrr:$addr))]>;
 def LDUBri : F3_2<3, 0b000001,
                   (outs IntRegs:$dst), (ins MEMri:$addr),
                   "ldub [$addr], $dst",
-                  [(set IntRegs:$dst, (zextloadi8 ADDRri:$addr))]>;
+                  [(set i32:$dst, (zextloadi8 ADDRri:$addr))]>;
 def LDUHrr : F3_1<3, 0b000010,
                   (outs IntRegs:$dst), (ins MEMrr:$addr),
                   "lduh [$addr], $dst",
-                  [(set IntRegs:$dst, (zextloadi16 ADDRrr:$addr))]>;
+                  [(set i32:$dst, (zextloadi16 ADDRrr:$addr))]>;
 def LDUHri : F3_2<3, 0b000010,
                   (outs IntRegs:$dst), (ins MEMri:$addr),
                   "lduh [$addr], $dst",
-                  [(set IntRegs:$dst, (zextloadi16 ADDRri:$addr))]>;
+                  [(set i32:$dst, (zextloadi16 ADDRri:$addr))]>;
 def LDrr   : F3_1<3, 0b000000,
                   (outs IntRegs:$dst), (ins MEMrr:$addr),
                   "ld [$addr], $dst",
-                  [(set IntRegs:$dst, (load ADDRrr:$addr))]>;
+                  [(set i32:$dst, (load ADDRrr:$addr))]>;
 def LDri   : F3_2<3, 0b000000,
                   (outs IntRegs:$dst), (ins MEMri:$addr),
                   "ld [$addr], $dst",
-                  [(set IntRegs:$dst, (load ADDRri:$addr))]>;
+                  [(set i32:$dst, (load ADDRri:$addr))]>;
 
 // Section B.2 - Load Floating-point Instructions, p. 92
 def LDFrr  : F3_1<3, 0b100000,
                   (outs FPRegs:$dst), (ins MEMrr:$addr),
                   "ld [$addr], $dst",
-                  [(set FPRegs:$dst, (load ADDRrr:$addr))]>;
+                  [(set f32:$dst, (load ADDRrr:$addr))]>;
 def LDFri  : F3_2<3, 0b100000,
                   (outs FPRegs:$dst), (ins MEMri:$addr),
                   "ld [$addr], $dst",
-                  [(set FPRegs:$dst, (load ADDRri:$addr))]>;
+                  [(set f32:$dst, (load ADDRri:$addr))]>;
 def LDDFrr : F3_1<3, 0b100011,
                   (outs DFPRegs:$dst), (ins MEMrr:$addr),
                   "ldd [$addr], $dst",
-                  [(set DFPRegs:$dst, (load ADDRrr:$addr))]>;
+                  [(set f64:$dst, (load ADDRrr:$addr))]>;
 def LDDFri : F3_2<3, 0b100011,
                   (outs DFPRegs:$dst), (ins MEMri:$addr),
                   "ldd [$addr], $dst",
-                  [(set DFPRegs:$dst, (load ADDRri:$addr))]>;
+                  [(set f64:$dst, (load ADDRri:$addr))]>;
 
 // Section B.4 - Store Integer Instructions, p. 95
 def STBrr : F3_1<3, 0b000101,
                  (outs), (ins MEMrr:$addr, IntRegs:$src),
                  "stb $src, [$addr]",
-                 [(truncstorei8 IntRegs:$src, ADDRrr:$addr)]>;
+                 [(truncstorei8 i32:$src, ADDRrr:$addr)]>;
 def STBri : F3_2<3, 0b000101,
                  (outs), (ins MEMri:$addr, IntRegs:$src),
                  "stb $src, [$addr]",
-                 [(truncstorei8 IntRegs:$src, ADDRri:$addr)]>;
+                 [(truncstorei8 i32:$src, ADDRri:$addr)]>;
 def STHrr : F3_1<3, 0b000110,
                  (outs), (ins MEMrr:$addr, IntRegs:$src),
                  "sth $src, [$addr]",
-                 [(truncstorei16 IntRegs:$src, ADDRrr:$addr)]>;
+                 [(truncstorei16 i32:$src, ADDRrr:$addr)]>;
 def STHri : F3_2<3, 0b000110,
                  (outs), (ins MEMri:$addr, IntRegs:$src),
                  "sth $src, [$addr]",
-                 [(truncstorei16 IntRegs:$src, ADDRri:$addr)]>;
+                 [(truncstorei16 i32:$src, ADDRri:$addr)]>;
 def STrr  : F3_1<3, 0b000100,
                  (outs), (ins MEMrr:$addr, IntRegs:$src),
                  "st $src, [$addr]",
-                 [(store IntRegs:$src, ADDRrr:$addr)]>;
+                 [(store i32:$src, ADDRrr:$addr)]>;
 def STri  : F3_2<3, 0b000100,
                  (outs), (ins MEMri:$addr, IntRegs:$src),
                  "st $src, [$addr]",
-                 [(store IntRegs:$src, ADDRri:$addr)]>;
+                 [(store i32:$src, ADDRri:$addr)]>;
 
 // Section B.5 - Store Floating-point Instructions, p. 97
 def STFrr   : F3_1<3, 0b100100,
                    (outs), (ins MEMrr:$addr, FPRegs:$src),
                    "st $src, [$addr]",
-                   [(store FPRegs:$src, ADDRrr:$addr)]>;
+                   [(store f32:$src, ADDRrr:$addr)]>;
 def STFri   : F3_2<3, 0b100100,
                    (outs), (ins MEMri:$addr, FPRegs:$src),
                    "st $src, [$addr]",
-                   [(store FPRegs:$src, ADDRri:$addr)]>;
+                   [(store f32:$src, ADDRri:$addr)]>;
 def STDFrr  : F3_1<3, 0b100111,
                    (outs), (ins MEMrr:$addr, DFPRegs:$src),
                    "std  $src, [$addr]",
-                   [(store DFPRegs:$src, ADDRrr:$addr)]>;
+                   [(store f64:$src, ADDRrr:$addr)]>;
 def STDFri  : F3_2<3, 0b100111,
                    (outs), (ins MEMri:$addr, DFPRegs:$src),
                    "std $src, [$addr]",
-                   [(store DFPRegs:$src, ADDRri:$addr)]>;
+                   [(store f64:$src, ADDRri:$addr)]>;
 
 // Section B.9 - SETHI Instruction, p. 104
 def SETHIi: F2_1<0b100,
                  (outs IntRegs:$dst), (ins i32imm:$src),
                  "sethi $src, $dst",
-                 [(set IntRegs:$dst, SETHIimm:$src)]>;
+                 [(set i32:$dst, SETHIimm:$src)]>;
 
 // Section B.10 - NOP Instruction, p. 105
 // (It's a special case of SETHI)
@@ -426,7 +428,7 @@ defm AND    : F3_12<"and", 0b000001, and>;
 def ANDNrr  : F3_1<2, 0b000101,
                    (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
                    "andn $b, $c, $dst",
-                   [(set IntRegs:$dst, (and IntRegs:$b, (not IntRegs:$c)))]>;
+                   [(set i32:$dst, (and i32:$b, (not i32:$c)))]>;
 def ANDNri  : F3_2<2, 0b000101,
                    (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
                    "andn $b, $c, $dst", []>;
@@ -436,7 +438,7 @@ defm OR     : F3_12<"or", 0b000010, or>;
 def ORNrr   : F3_1<2, 0b000110,
                    (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
                    "orn $b, $c, $dst",
-                   [(set IntRegs:$dst, (or IntRegs:$b, (not IntRegs:$c)))]>;
+                   [(set i32:$dst, (or i32:$b, (not i32:$c)))]>;
 def ORNri   : F3_2<2, 0b000110,
                    (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
                    "orn $b, $c, $dst", []>;
@@ -445,7 +447,7 @@ defm XOR    : F3_12<"xor", 0b000011, xor>;
 def XNORrr  : F3_1<2, 0b000111,
                    (outs IntRegs:$dst), (ins IntRegs:$b, IntRegs:$c),
                    "xnor $b, $c, $dst",
-                   [(set IntRegs:$dst, (not (xor IntRegs:$b, IntRegs:$c)))]>;
+                   [(set i32:$dst, (not (xor i32:$b, i32:$c)))]>;
 def XNORri  : F3_2<2, 0b000111,
                    (outs IntRegs:$dst), (ins IntRegs:$b, i32imm:$c),
                    "xnor $b, $c, $dst", []>;
@@ -462,7 +464,7 @@ defm ADD   : F3_12<"add", 0b000000, add>;
 def LEA_ADDri   : F3_2<2, 0b000000,
                    (outs IntRegs:$dst), (ins MEMri:$addr),
                    "add ${addr:arith}, $dst",
-                   [(set IntRegs:$dst, ADDRri:$addr)]>;
+                   [(set i32:$dst, ADDRri:$addr)]>;
 
 let Defs = [ICC] in                   
   defm ADDCC  : F3_12<"addcc", 0b010000, addc>;
@@ -603,11 +605,11 @@ def FDTOI : F3_3<2, 0b110100, 0b011010010,
 def FSTOD : F3_3<2, 0b110100, 0b011001001, 
                  (outs DFPRegs:$dst), (ins FPRegs:$src),
                  "fstod $src, $dst",
-                 [(set DFPRegs:$dst, (fextend FPRegs:$src))]>;
+                 [(set f64:$dst, (fextend f32:$src))]>;
 def FDTOS : F3_3<2, 0b110100, 0b011000110,
                  (outs FPRegs:$dst), (ins DFPRegs:$src),
                  "fdtos $src, $dst",
-                 [(set FPRegs:$dst, (fround DFPRegs:$src))]>;
+                 [(set f32:$dst, (fround f64:$src))]>;
 
 // Floating-point Move Instructions, p. 144
 def FMOVS : F3_3<2, 0b110100, 0b000000001,
@@ -616,22 +618,22 @@ def FMOVS : F3_3<2, 0b110100, 0b000000001,
 def FNEGS : F3_3<2, 0b110100, 0b000000101, 
                  (outs FPRegs:$dst), (ins FPRegs:$src),
                  "fnegs $src, $dst",
-                 [(set FPRegs:$dst, (fneg FPRegs:$src))]>;
+                 [(set f32:$dst, (fneg f32:$src))]>;
 def FABSS : F3_3<2, 0b110100, 0b000001001, 
                  (outs FPRegs:$dst), (ins FPRegs:$src),
                  "fabss $src, $dst",
-                 [(set FPRegs:$dst, (fabs FPRegs:$src))]>;
+                 [(set f32:$dst, (fabs f32:$src))]>;
 
 
 // Floating-point Square Root Instructions, p.145
 def FSQRTS : F3_3<2, 0b110100, 0b000101001, 
                   (outs FPRegs:$dst), (ins FPRegs:$src),
                   "fsqrts $src, $dst",
-                  [(set FPRegs:$dst, (fsqrt FPRegs:$src))]>;
+                  [(set f32:$dst, (fsqrt f32:$src))]>;
 def FSQRTD : F3_3<2, 0b110100, 0b000101010, 
                   (outs DFPRegs:$dst), (ins DFPRegs:$src),
                   "fsqrtd $src, $dst",
-                  [(set DFPRegs:$dst, (fsqrt DFPRegs:$src))]>;
+                  [(set f64:$dst, (fsqrt f64:$src))]>;
 
 
 
@@ -639,42 +641,42 @@ def FSQRTD : F3_3<2, 0b110100, 0b000101010,
 def FADDS  : F3_3<2, 0b110100, 0b001000001,
                   (outs FPRegs:$dst), (ins FPRegs:$src1, FPRegs:$src2),
                   "fadds $src1, $src2, $dst",
-                  [(set FPRegs:$dst, (fadd FPRegs:$src1, FPRegs:$src2))]>;
+                  [(set f32:$dst, (fadd f32:$src1, f32:$src2))]>;
 def FADDD  : F3_3<2, 0b110100, 0b001000010,
                   (outs DFPRegs:$dst), (ins DFPRegs:$src1, DFPRegs:$src2),
                   "faddd $src1, $src2, $dst",
-                  [(set DFPRegs:$dst, (fadd DFPRegs:$src1, DFPRegs:$src2))]>;
+                  [(set f64:$dst, (fadd f64:$src1, f64:$src2))]>;
 def FSUBS  : F3_3<2, 0b110100, 0b001000101,
                   (outs FPRegs:$dst), (ins FPRegs:$src1, FPRegs:$src2),
                   "fsubs $src1, $src2, $dst",
-                  [(set FPRegs:$dst, (fsub FPRegs:$src1, FPRegs:$src2))]>;
+                  [(set f32:$dst, (fsub f32:$src1, f32:$src2))]>;
 def FSUBD  : F3_3<2, 0b110100, 0b001000110,
                   (outs DFPRegs:$dst), (ins DFPRegs:$src1, DFPRegs:$src2),
                   "fsubd $src1, $src2, $dst",
-                  [(set DFPRegs:$dst, (fsub DFPRegs:$src1, DFPRegs:$src2))]>;
+                  [(set f64:$dst, (fsub f64:$src1, f64:$src2))]>;
 
 // Floating-point Multiply and Divide Instructions, p. 147
 def FMULS  : F3_3<2, 0b110100, 0b001001001,
                   (outs FPRegs:$dst), (ins FPRegs:$src1, FPRegs:$src2),
                   "fmuls $src1, $src2, $dst",
-                  [(set FPRegs:$dst, (fmul FPRegs:$src1, FPRegs:$src2))]>;
+                  [(set f32:$dst, (fmul f32:$src1, f32:$src2))]>;
 def FMULD  : F3_3<2, 0b110100, 0b001001010,
                   (outs DFPRegs:$dst), (ins DFPRegs:$src1, DFPRegs:$src2),
                   "fmuld $src1, $src2, $dst",
-                  [(set DFPRegs:$dst, (fmul DFPRegs:$src1, DFPRegs:$src2))]>;
+                  [(set f64:$dst, (fmul f64:$src1, f64:$src2))]>;
 def FSMULD : F3_3<2, 0b110100, 0b001101001,
                   (outs DFPRegs:$dst), (ins FPRegs:$src1, FPRegs:$src2),
                   "fsmuld $src1, $src2, $dst",
-                  [(set DFPRegs:$dst, (fmul (fextend FPRegs:$src1),
-                                            (fextend FPRegs:$src2)))]>;
+                  [(set f64:$dst, (fmul (fextend f32:$src1),
+                                        (fextend f32:$src2)))]>;
 def FDIVS  : F3_3<2, 0b110100, 0b001001101,
                  (outs FPRegs:$dst), (ins FPRegs:$src1, FPRegs:$src2),
                  "fdivs $src1, $src2, $dst",
-                 [(set FPRegs:$dst, (fdiv FPRegs:$src1, FPRegs:$src2))]>;
+                 [(set f32:$dst, (fdiv f32:$src1, f32:$src2))]>;
 def FDIVD  : F3_3<2, 0b110100, 0b001001110,
                  (outs DFPRegs:$dst), (ins DFPRegs:$src1, DFPRegs:$src2),
                  "fdivd $src1, $src2, $dst",
-                 [(set DFPRegs:$dst, (fdiv DFPRegs:$src1, DFPRegs:$src2))]>;
+                 [(set f64:$dst, (fdiv f64:$src1, f64:$src2))]>;
 
 // Floating-point Compare Instructions, p. 148
 // Note: the 2nd template arg is different for these guys.
@@ -685,11 +687,11 @@ let Defs = [FCC] in {
   def FCMPS  : F3_3<2, 0b110101, 0b001010001,
                    (outs), (ins FPRegs:$src1, FPRegs:$src2),
                    "fcmps $src1, $src2\n\tnop",
-                   [(SPcmpfcc FPRegs:$src1, FPRegs:$src2)]>;
+                   [(SPcmpfcc f32:$src1, f32:$src2)]>;
   def FCMPD  : F3_3<2, 0b110101, 0b001010010,
                    (outs), (ins DFPRegs:$src1, DFPRegs:$src2),
                    "fcmpd $src1, $src2\n\tnop",
-                   [(SPcmpfcc DFPRegs:$src1, DFPRegs:$src2)]>;
+                   [(SPcmpfcc f64:$src1, f64:$src2)]>;
 }
 
 //===----------------------------------------------------------------------===//
@@ -704,52 +706,45 @@ let Predicates = [HasV9], Constraints = "$T = $dst" in {
     def MOVICCrr
       : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, CCOp:$cc),
                "mov$cc %icc, $F, $dst",
-               [(set IntRegs:$dst,
-                           (SPselecticc IntRegs:$F, IntRegs:$T, imm:$cc))]>;
+               [(set i32:$dst, (SPselecticc i32:$F, i32:$T, imm:$cc))]>;
     def MOVICCri
       : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, i32imm:$F, CCOp:$cc),
                "mov$cc %icc, $F, $dst",
-               [(set IntRegs:$dst,
-                            (SPselecticc simm11:$F, IntRegs:$T, imm:$cc))]>;
+               [(set i32:$dst, (SPselecticc simm11:$F, i32:$T, imm:$cc))]>;
   }
 
   let Uses = [FCC] in {
     def MOVFCCrr
       : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, IntRegs:$F, CCOp:$cc),
                "mov$cc %fcc0, $F, $dst",
-               [(set IntRegs:$dst,
-                           (SPselectfcc IntRegs:$F, IntRegs:$T, imm:$cc))]>;
+               [(set i32:$dst, (SPselectfcc i32:$F, i32:$T, imm:$cc))]>;
     def MOVFCCri
       : Pseudo<(outs IntRegs:$dst), (ins IntRegs:$T, i32imm:$F, CCOp:$cc),
                "mov$cc %fcc0, $F, $dst",
-               [(set IntRegs:$dst,
-                            (SPselectfcc simm11:$F, IntRegs:$T, imm:$cc))]>;
+               [(set i32:$dst, (SPselectfcc simm11:$F, i32:$T, imm:$cc))]>;
   }
 
   let Uses = [ICC] in {
     def FMOVS_ICC
       : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, CCOp:$cc),
                "fmovs$cc %icc, $F, $dst",
-               [(set FPRegs:$dst,
-                           (SPselecticc FPRegs:$F, FPRegs:$T, imm:$cc))]>;
+               [(set f32:$dst,
+                           (SPselecticc f32:$F, f32:$T, imm:$cc))]>;
     def FMOVD_ICC
       : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, CCOp:$cc),
                "fmovd$cc %icc, $F, $dst",
-               [(set DFPRegs:$dst,
-                           (SPselecticc DFPRegs:$F, DFPRegs:$T, imm:$cc))]>;
+               [(set f64:$dst, (SPselecticc f64:$F, f64:$T, imm:$cc))]>;
   }
 
   let Uses = [FCC] in {
     def FMOVS_FCC
       : Pseudo<(outs FPRegs:$dst), (ins FPRegs:$T, FPRegs:$F, CCOp:$cc),
                "fmovs$cc %fcc0, $F, $dst",
-               [(set FPRegs:$dst,
-                           (SPselectfcc FPRegs:$F, FPRegs:$T, imm:$cc))]>;
+               [(set f32:$dst, (SPselectfcc f32:$F, f32:$T, imm:$cc))]>;
     def FMOVD_FCC
       : Pseudo<(outs DFPRegs:$dst), (ins DFPRegs:$T, DFPRegs:$F, CCOp:$cc),
                "fmovd$cc %fcc0, $F, $dst",
-               [(set DFPRegs:$dst,
-                           (SPselectfcc DFPRegs:$F, DFPRegs:$T, imm:$cc))]>;
+               [(set f64:$dst, (SPselectfcc f64:$F, f64:$T, imm:$cc))]>;
   }
 
 }
@@ -762,11 +757,11 @@ let Predicates = [HasV9] in {
   def FNEGD : F3_3<2, 0b110100, 0b000000110, 
                    (outs DFPRegs:$dst), (ins DFPRegs:$src),
                    "fnegd $src, $dst",
-                   [(set DFPRegs:$dst, (fneg DFPRegs:$src))]>;
+                   [(set f64:$dst, (fneg f64:$src))]>;
   def FABSD : F3_3<2, 0b110100, 0b000001010, 
                    (outs DFPRegs:$dst), (ins DFPRegs:$src),
                    "fabsd $src, $dst",
-                   [(set DFPRegs:$dst, (fabs DFPRegs:$src))]>;
+                   [(set f64:$dst, (fabs f64:$src))]>;
 }
 
 // POPCrr - This does a ctpop of a 64-bit register.  As such, we have to clear
@@ -774,8 +769,8 @@ let Predicates = [HasV9] in {
 def POPCrr : F3_1<2, 0b101110, 
                   (outs IntRegs:$dst), (ins IntRegs:$src),
                   "popc $src, $dst", []>, Requires<[HasV9]>;
-def : Pat<(ctpop IntRegs:$src),
-          (POPCrr (SLLri IntRegs:$src, 0))>;
+def : Pat<(ctpop i32:$src),
+          (POPCrr (SLLri $src, 0))>;
 
 //===----------------------------------------------------------------------===//
 // Non-Instruction Patterns
@@ -783,28 +778,28 @@ def : Pat<(ctpop IntRegs:$src),
 
 // Small immediates.
 def : Pat<(i32 simm13:$val),
-          (ORri G0, imm:$val)>;
+          (ORri (i32 G0), imm:$val)>;
 // Arbitrary immediates.
 def : Pat<(i32 imm:$val),
           (ORri (SETHIi (HI22 imm:$val)), (LO10 imm:$val))>;
 
 // subc
-def : Pat<(subc IntRegs:$b, IntRegs:$c),
-          (SUBCCrr IntRegs:$b, IntRegs:$c)>;
-def : Pat<(subc IntRegs:$b, simm13:$val),
-          (SUBCCri IntRegs:$b, imm:$val)>;
+def : Pat<(subc i32:$b, i32:$c),
+          (SUBCCrr $b, $c)>;
+def : Pat<(subc i32:$b, simm13:$val),
+          (SUBCCri $b, imm:$val)>;
 
 // Global addresses, constant pool entries
 def : Pat<(SPhi tglobaladdr:$in), (SETHIi tglobaladdr:$in)>;
-def : Pat<(SPlo tglobaladdr:$in), (ORri G0, tglobaladdr:$in)>;
+def : Pat<(SPlo tglobaladdr:$in), (ORri (i32 G0), tglobaladdr:$in)>;
 def : Pat<(SPhi tconstpool:$in), (SETHIi tconstpool:$in)>;
-def : Pat<(SPlo tconstpool:$in), (ORri G0, tconstpool:$in)>;
+def : Pat<(SPlo tconstpool:$in), (ORri (i32 G0), tconstpool:$in)>;
 
 // Add reg, lo.  This is used when taking the addr of a global/constpool entry.
-def : Pat<(add IntRegs:$r, (SPlo tglobaladdr:$in)),
-          (ADDri IntRegs:$r, tglobaladdr:$in)>;
-def : Pat<(add IntRegs:$r, (SPlo tconstpool:$in)),
-          (ADDri IntRegs:$r, tconstpool:$in)>;
+def : Pat<(add i32:$r, (SPlo tglobaladdr:$in)),
+          (ADDri $r, tglobaladdr:$in)>;
+def : Pat<(add i32:$r, (SPlo tconstpool:$in)),
+          (ADDri $r, tconstpool:$in)>;
 
 // Calls: 
 def : Pat<(call tglobaladdr:$dst),
@@ -823,3 +818,5 @@ def : Pat<(i32 (extloadi16 ADDRri:$src)), (LDUHri ADDRri:$src)>;
 // zextload bool -> zextload byte
 def : Pat<(i32 (zextloadi1 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
 def : Pat<(i32 (zextloadi1 ADDRri:$src)), (LDUBri ADDRri:$src)>;
+
+include "SparcInstr64Bit.td"
diff --git a/lib/Target/Sparc/SparcRegisterInfo.cpp b/lib/Target/Sparc/SparcRegisterInfo.cpp
index ff8d3c533f3d..db9b30eb4330 100644
--- a/lib/Target/Sparc/SparcRegisterInfo.cpp
+++ b/lib/Target/Sparc/SparcRegisterInfo.cpp
@@ -14,14 +14,14 @@
 #include "SparcRegisterInfo.h"
 #include "Sparc.h"
 #include "SparcSubtarget.h"
-#include "llvm/Type.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/STLExtras.h"
 
 #define GET_REGINFO_TARGET_DESC
 #include "SparcGenRegisterInfo.inc"
@@ -56,45 +56,33 @@ BitVector SparcRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
   return Reserved;
 }
 
-void SparcRegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  MachineInstr &MI = *I;
-  DebugLoc dl = MI.getDebugLoc();
-  int Size = MI.getOperand(0).getImm();
-  if (MI.getOpcode() == SP::ADJCALLSTACKDOWN)
-    Size = -Size;
-  if (Size)
-    BuildMI(MBB, I, dl, TII.get(SP::ADDri), SP::O6).addReg(SP::O6).addImm(Size);
-  MBB.erase(I);
+const TargetRegisterClass*
+SparcRegisterInfo::getPointerRegClass(const MachineFunction &MF,
+                                      unsigned Kind) const {
+  return Subtarget.is64Bit() ? &SP::I64RegsRegClass : &SP::IntRegsRegClass;
 }
 
 void
 SparcRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
-                                       int SPAdj, RegScavenger *RS) const {
+                                       int SPAdj, unsigned FIOperandNum,
+                                       RegScavenger *RS) const {
   assert(SPAdj == 0 && "Unexpected");
 
-  unsigned i = 0;
   MachineInstr &MI = *II;
   DebugLoc dl = MI.getDebugLoc();
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
-  }
-
-  int FrameIndex = MI.getOperand(i).getIndex();
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
 
   // Addressable stack objects are accessed using neg. offsets from %fp
   MachineFunction &MF = *MI.getParent()->getParent();
   int Offset = MF.getFrameInfo()->getObjectOffset(FrameIndex) +
-               MI.getOperand(i+1).getImm();
+               MI.getOperand(FIOperandNum + 1).getImm();
 
   // Replace frame index with a frame pointer reference.
   if (Offset >= -4096 && Offset <= 4095) {
     // If the offset is small enough to fit in the immediate field, directly
     // encode it.
-    MI.getOperand(i).ChangeToRegister(SP::I6, false);
-    MI.getOperand(i+1).ChangeToImmediate(Offset);
+    MI.getOperand(FIOperandNum).ChangeToRegister(SP::I6, false);
+    MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset);
   } else {
     // Otherwise, emit a G1 = SETHI %hi(offset).  FIXME: it would be better to 
     // scavenge a register here instead of reserving G1 all of the time.
@@ -104,8 +92,8 @@ SparcRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
     BuildMI(*MI.getParent(), II, dl, TII.get(SP::ADDrr), SP::G1).addReg(SP::G1)
       .addReg(SP::I6);
     // Insert: G1+%lo(offset) into the user.
-    MI.getOperand(i).ChangeToRegister(SP::G1, false);
-    MI.getOperand(i+1).ChangeToImmediate(Offset & ((1 << 10)-1));
+    MI.getOperand(FIOperandNum).ChangeToRegister(SP::G1, false);
+    MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset & ((1 << 10)-1));
   }
 }
 
diff --git a/lib/Target/Sparc/SparcRegisterInfo.h b/lib/Target/Sparc/SparcRegisterInfo.h
index 9515ad33dcc2..f91df5398953 100644
--- a/lib/Target/Sparc/SparcRegisterInfo.h
+++ b/lib/Target/Sparc/SparcRegisterInfo.h
@@ -36,14 +36,15 @@ struct SparcRegisterInfo : public SparcGenRegisterInfo {
 
   BitVector getReservedRegs(const MachineFunction &MF) const;
 
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
+  const TargetRegisterClass *getPointerRegClass(const MachineFunction &MF,
+                                                unsigned Kind) const;
 
   void eliminateFrameIndex(MachineBasicBlock::iterator II,
-                           int SPAdj, RegScavenger *RS = NULL) const;
+                           int SPAdj, unsigned FIOperandNum,
+                           RegScavenger *RS = NULL) const;
 
-  void processFunctionBeforeFrameFinalized(MachineFunction &MF) const;
+  void processFunctionBeforeFrameFinalized(MachineFunction &MF,
+                                       RegScavenger *RS = NULL) const;
 
   // Debug information queries.
   unsigned getFrameRegister(const MachineFunction &MF) const;
diff --git a/lib/Target/Sparc/SparcRegisterInfo.td b/lib/Target/Sparc/SparcRegisterInfo.td
index 81bff6c51c9d..497e7c5d5612 100644
--- a/lib/Target/Sparc/SparcRegisterInfo.td
+++ b/lib/Target/Sparc/SparcRegisterInfo.td
@@ -43,7 +43,7 @@ class Rd<bits<5> num, string n, list<Register> subregs> : SparcReg<n> {
 }
 
 // Control Registers
-def ICC : SparcCtrlReg<"ICC">;
+def ICC : SparcCtrlReg<"ICC">; // This represents icc and xcc in 64-bit code.
 def FCC : SparcCtrlReg<"FCC">;
 
 // Y register
@@ -140,7 +140,10 @@ def D15 : Rd<30, "F30", [F30, F31]>, DwarfRegNum<[87]>;
 // FIXME: the register order should be defined in terms of the preferred
 // allocation order...
 //
-def IntRegs : RegisterClass<"SP", [i32], 32,
+// This register class should not be used to hold i64 values, use the I64Regs
+// register class for that. The i64 type is included here to allow i64 patterns
+// using the integer instructions.
+def IntRegs : RegisterClass<"SP", [i32, i64], 32,
                             (add L0, L1, L2, L3, L4, L5, L6,
                                  L7, I0, I1, I2, I3, I4, I5,
                                  O0, O1, O2, O3, O4, O5, O7,
@@ -155,6 +158,13 @@ def IntRegs : RegisterClass<"SP", [i32], 32,
                                  G5, G6, G7 // reserved for kernel
                                  )>;
 
+// Register class for 64-bit mode, with a 64-bit spill slot size.
+// These are the same as the 32-bit registers, so TableGen will consider this
+// to be a sub-class of IntRegs. That works out because requiring a 64-bit
+// spill slot is a stricter constraint than only requiring a 32-bit spill slot.
+def I64Regs : RegisterClass<"SP", [i64], 64, (add IntRegs)>;
+
+// Floating point register classes.
 def FPRegs : RegisterClass<"SP", [f32], 32, (sequence "F%u", 0, 31)>;
 
 def DFPRegs : RegisterClass<"SP", [f64], 64, (sequence "D%u", 0, 15)>;
diff --git a/lib/Target/Sparc/SparcTargetMachine.cpp b/lib/Target/Sparc/SparcTargetMachine.cpp
index 45c962471dda..60bceb708fbc 100644
--- a/lib/Target/Sparc/SparcTargetMachine.cpp
+++ b/lib/Target/Sparc/SparcTargetMachine.cpp
@@ -12,8 +12,8 @@
 
 #include "SparcTargetMachine.h"
 #include "Sparc.h"
-#include "llvm/PassManager.h"
 #include "llvm/CodeGen/Passes.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
@@ -36,7 +36,7 @@ SparcTargetMachine::SparcTargetMachine(const Target &T, StringRef TT,
     DL(Subtarget.getDataLayout()),
     InstrInfo(Subtarget),
     TLInfo(*this), TSInfo(*this),
-    FrameLowering(Subtarget), STTI(&TLInfo), VTTI(&TLInfo) {
+    FrameLowering(Subtarget) {
 }
 
 namespace {
diff --git a/lib/Target/Sparc/SparcTargetMachine.h b/lib/Target/Sparc/SparcTargetMachine.h
index 0fbe2d7cda36..081075de2dc8 100644
--- a/lib/Target/Sparc/SparcTargetMachine.h
+++ b/lib/Target/Sparc/SparcTargetMachine.h
@@ -14,15 +14,14 @@
 #ifndef SPARCTARGETMACHINE_H
 #define SPARCTARGETMACHINE_H
 
-#include "SparcInstrInfo.h"
-#include "SparcISelLowering.h"
 #include "SparcFrameLowering.h"
+#include "SparcISelLowering.h"
+#include "SparcInstrInfo.h"
 #include "SparcSelectionDAGInfo.h"
 #include "SparcSubtarget.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/DataLayout.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetTransformImpl.h"
+#include "llvm/Target/TargetMachine.h"
 
 namespace llvm {
 
@@ -33,8 +32,6 @@ class SparcTargetMachine : public LLVMTargetMachine {
   SparcTargetLowering TLInfo;
   SparcSelectionDAGInfo TSInfo;
   SparcFrameLowering FrameLowering;
-  ScalarTargetTransformImpl STTI;
-  VectorTargetTransformImpl VTTI;
 public:
   SparcTargetMachine(const Target &T, StringRef TT,
                      StringRef CPU, StringRef FS, const TargetOptions &Options,
@@ -55,12 +52,6 @@ public:
   virtual const SparcSelectionDAGInfo* getSelectionDAGInfo() const {
     return &TSInfo;
   }
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
   virtual const DataLayout       *getDataLayout() const { return &DL; }
 
   // Pass Pipeline Configuration
diff --git a/lib/Target/Sparc/TargetInfo/SparcTargetInfo.cpp b/lib/Target/Sparc/TargetInfo/SparcTargetInfo.cpp
index c9d5b7bdfb3d..bb714632349a 100644
--- a/lib/Target/Sparc/TargetInfo/SparcTargetInfo.cpp
+++ b/lib/Target/Sparc/TargetInfo/SparcTargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "Sparc.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
diff --git a/lib/Target/Target.cpp b/lib/Target/Target.cpp
index 393178a4692e..9a78ebc3facb 100644
--- a/lib/Target/Target.cpp
+++ b/lib/Target/Target.cpp
@@ -14,11 +14,11 @@
 
 #include "llvm-c/Target.h"
 #include "llvm-c/Initialization.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/LLVMContext.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/PassManager.h"
-#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
-#include "llvm/LLVMContext.h"
 #include <cstring>
 
 using namespace llvm;
@@ -26,7 +26,6 @@ using namespace llvm;
 void llvm::initializeTarget(PassRegistry &Registry) {
   initializeDataLayoutPass(Registry);
   initializeTargetLibraryInfoPass(Registry);
-  initializeTargetTransformInfoPass(Registry);
 }
 
 void LLVMInitializeTarget(LLVMPassRegistryRef R) {
diff --git a/lib/Target/TargetInstrInfo.cpp b/lib/Target/TargetInstrInfo.cpp
deleted file mode 100644
index f1d1d07c38ae..000000000000
--- a/lib/Target/TargetInstrInfo.cpp
+++ /dev/null
@@ -1,88 +0,0 @@
-//===-- TargetInstrInfo.cpp - Target Instruction Information --------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file implements the TargetInstrInfo class.
-//
-//===----------------------------------------------------------------------===//
-
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/MC/MCAsmInfo.h"
-#include "llvm/MC/MCInstrItineraries.h"
-#include "llvm/Support/ErrorHandling.h"
-#include <cctype>
-using namespace llvm;
-
-//===----------------------------------------------------------------------===//
-//  TargetInstrInfo
-//
-// Methods that depend on CodeGen are implemented in
-// TargetInstrInfoImpl.cpp. Invoking them without linking libCodeGen raises a
-// link error.
-// ===----------------------------------------------------------------------===//
-
-TargetInstrInfo::~TargetInstrInfo() {
-}
-
-const TargetRegisterClass*
-TargetInstrInfo::getRegClass(const MCInstrDesc &MCID, unsigned OpNum,
-                             const TargetRegisterInfo *TRI,
-                             const MachineFunction &MF) const {
-  if (OpNum >= MCID.getNumOperands())
-    return 0;
-
-  short RegClass = MCID.OpInfo[OpNum].RegClass;
-  if (MCID.OpInfo[OpNum].isLookupPtrRegClass())
-    return TRI->getPointerRegClass(MF, RegClass);
-
-  // Instructions like INSERT_SUBREG do not have fixed register classes.
-  if (RegClass < 0)
-    return 0;
-
-  // Otherwise just look it up normally.
-  return TRI->getRegClass(RegClass);
-}
-
-/// insertNoop - Insert a noop into the instruction stream at the specified
-/// point.
-void TargetInstrInfo::insertNoop(MachineBasicBlock &MBB,
-                                 MachineBasicBlock::iterator MI) const {
-  llvm_unreachable("Target didn't implement insertNoop!");
-}
-
-/// Measure the specified inline asm to determine an approximation of its
-/// length.
-/// Comments (which run till the next SeparatorString or newline) do not
-/// count as an instruction.
-/// Any other non-whitespace text is considered an instruction, with
-/// multiple instructions separated by SeparatorString or newlines.
-/// Variable-length instructions are not handled here; this function
-/// may be overloaded in the target code to do that.
-unsigned TargetInstrInfo::getInlineAsmLength(const char *Str,
-                                             const MCAsmInfo &MAI) const {
-
-
-  // Count the number of instructions in the asm.
-  bool atInsnStart = true;
-  unsigned Length = 0;
-  for (; *Str; ++Str) {
-    if (*Str == '\n' || strncmp(Str, MAI.getSeparatorString(),
-                                strlen(MAI.getSeparatorString())) == 0)
-      atInsnStart = true;
-    if (atInsnStart && !std::isspace(*Str)) {
-      Length += MAI.getMaxInstLength();
-      atInsnStart = false;
-    }
-    if (atInsnStart && strncmp(Str, MAI.getCommentString(),
-                               strlen(MAI.getCommentString())) == 0)
-      atInsnStart = false;
-  }
-
-  return Length;
-}
diff --git a/lib/Target/TargetIntrinsicInfo.cpp b/lib/Target/TargetIntrinsicInfo.cpp
index e049a1d3b62f..64bd56f6e7df 100644
--- a/lib/Target/TargetIntrinsicInfo.cpp
+++ b/lib/Target/TargetIntrinsicInfo.cpp
@@ -12,8 +12,8 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/Target/TargetIntrinsicInfo.h"
-#include "llvm/Function.h"
 #include "llvm/ADT/StringMap.h"
+#include "llvm/IR/Function.h"
 using namespace llvm;
 
 TargetIntrinsicInfo::TargetIntrinsicInfo() {
diff --git a/lib/Target/TargetLibraryInfo.cpp b/lib/Target/TargetLibraryInfo.cpp
index 6d4eab12045c..ee88ce77c09f 100644
--- a/lib/Target/TargetLibraryInfo.cpp
+++ b/lib/Target/TargetLibraryInfo.cpp
@@ -24,6 +24,8 @@ void TargetLibraryInfo::anchor() { }
 
 const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] =
   {
+    "_IO_getc",
+    "_IO_putc",
     "_ZdaPv",
     "_ZdlPv",
     "_Znaj",
@@ -38,7 +40,14 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] =
     "__cxa_guard_abort",
     "__cxa_guard_acquire",
     "__cxa_guard_release",
+    "__isoc99_scanf",
+    "__isoc99_sscanf",
     "__memcpy_chk",
+    "__strdup",
+    "__strndup",
+    "__strtok_r",
+    "abs",
+    "access",
     "acos",
     "acosf",
     "acosh",
@@ -60,6 +69,13 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] =
     "atanhf",
     "atanhl",
     "atanl",
+    "atof",
+    "atoi",
+    "atol",
+    "atoll",
+    "bcmp",
+    "bcopy",
+    "bzero",
     "calloc",
     "cbrt",
     "cbrtf",
@@ -67,6 +83,10 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] =
     "ceil",
     "ceilf",
     "ceill",
+    "chmod",
+    "chown",
+    "clearerr",
+    "closedir",
     "copysign",
     "copysignf",
     "copysignl",
@@ -76,6 +96,7 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] =
     "coshf",
     "coshl",
     "cosl",
+    "ctermid",
     "exp",
     "exp10",
     "exp10f",
@@ -91,18 +112,67 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] =
     "fabs",
     "fabsf",
     "fabsl",
+    "fclose",
+    "fdopen",
+    "feof",
+    "ferror",
+    "fflush",
+    "ffs",
+    "ffsl",
+    "ffsll",
+    "fgetc",
+    "fgetpos",
+    "fgets",
+    "fileno",
     "fiprintf",
+    "flockfile",
     "floor",
     "floorf",
     "floorl",
     "fmod",
     "fmodf",
     "fmodl",
+    "fopen",
+    "fopen64",
+    "fprintf",
     "fputc",
     "fputs",
+    "fread",
     "free",
+    "frexp",
+    "frexpf",
+    "frexpl",
+    "fscanf",
+    "fseek",
+    "fseeko",
+    "fseeko64",
+    "fsetpos",
+    "fstat",
+    "fstat64",
+    "fstatvfs",
+    "fstatvfs64",
+    "ftell",
+    "ftello",
+    "ftello64",
+    "ftrylockfile",
+    "funlockfile",
     "fwrite",
+    "getc",
+    "getc_unlocked",
+    "getchar",
+    "getenv",
+    "getitimer",
+    "getlogin_r",
+    "getpwnam",
+    "gets",
+    "htonl",
+    "htons",
     "iprintf",
+    "isascii",
+    "isdigit",
+    "labs",
+    "lchown",
+    "llabs",
     "log",
     "log10",
     "log10f",
@@ -118,30 +188,64 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] =
     "logbl",
     "logf",
     "logl",
+    "lstat",
+    "lstat64",
     "malloc",
+    "memalign",
+    "memccpy",
     "memchr",
     "memcmp",
     "memcpy",
     "memmove",
+    "memrchr",
     "memset",
     "memset_pattern16",
+    "mkdir",
+    "mktime",
+    "modf",
+    "modff",
+    "modfl",
     "nearbyint",
     "nearbyintf",
     "nearbyintl",
+    "ntohl",
+    "ntohs",
+    "open",
+    "open64",
+    "opendir",
+    "pclose",
+    "perror",
+    "popen",
     "posix_memalign",
     "pow",
     "powf",
     "powl",
+    "pread",
+    "printf",
+    "putc",
     "putchar",
     "puts",
+    "pwrite",
+    "qsort",
+    "read",
+    "readlink",
     "realloc",
     "reallocf",
+    "realpath",
+    "remove",
+    "rename",
+    "rewind",
     "rint",
     "rintf",
     "rintl",
+    "rmdir",
     "round",
     "roundf",
     "roundl",
+    "scanf",
+    "setbuf",
+    "setitimer",
+    "setvbuf",
     "sin",
     "sinf",
     "sinh",
@@ -149,17 +253,28 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] =
     "sinhl",
     "sinl",
     "siprintf",
+    "snprintf",
+    "sprintf",
     "sqrt",
     "sqrtf",
     "sqrtl",
+    "sscanf",
+    "stat",
+    "stat64",
+    "statvfs",
+    "statvfs64",
     "stpcpy",
+    "stpncpy",
+    "strcasecmp",
     "strcat",
     "strchr",
     "strcmp",
+    "strcoll",
     "strcpy",
     "strcspn",
     "strdup",
     "strlen",
+    "strncasecmp",
     "strncat",
     "strncmp",
     "strncpy",
@@ -171,21 +286,43 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] =
     "strstr",
     "strtod",
     "strtof",
+    "strtok",
+    "strtok_r",
     "strtol",
     "strtold",
     "strtoll",
     "strtoul",
     "strtoull",
+    "strxfrm",
+    "system",
     "tan",
     "tanf",
     "tanh",
     "tanhf",
     "tanhl",
     "tanl",
+    "times",
+    "tmpfile",
+    "tmpfile64",
+    "toascii",
     "trunc",
     "truncf",
     "truncl",
-    "valloc"
+    "uname",
+    "ungetc",
+    "unlink",
+    "unsetenv",
+    "utime",
+    "utimes",
+    "valloc",
+    "vfprintf",
+    "vfscanf",
+    "vprintf",
+    "vscanf",
+    "vsnprintf",
+    "vsprintf",
+    "vsscanf",
+    "write"
   };
 
 /// initialize - Initialize the set of available library functions based on the
@@ -247,7 +384,9 @@ static void initialize(TargetLibraryInfo &TLI, const Triple &T,
     TLI.setUnavailable(LibFunc::fabsl);
     TLI.setUnavailable(LibFunc::floorl);
     TLI.setUnavailable(LibFunc::fmodl);
+    TLI.setUnavailable(LibFunc::frexpl);
     TLI.setUnavailable(LibFunc::logl);
+    TLI.setUnavailable(LibFunc::modfl);
     TLI.setUnavailable(LibFunc::powl);
     TLI.setUnavailable(LibFunc::sinl);
     TLI.setUnavailable(LibFunc::sinhl);
@@ -324,9 +463,116 @@ static void initialize(TargetLibraryInfo &TLI, const Triple &T,
       TLI.setUnavailable(LibFunc::tanhf);
     }
 
-    // Win32 does *not* provide stpcpy.  It is provided on POSIX systems:
-    // http://pubs.opengroup.org/onlinepubs/9699919799/functions/stpcpy.html
+    // Win32 does *not* provide provide these functions, but they are
+    // generally available on POSIX-compliant systems:
+    TLI.setUnavailable(LibFunc::access);
+    TLI.setUnavailable(LibFunc::bcmp);
+    TLI.setUnavailable(LibFunc::bcopy);
+    TLI.setUnavailable(LibFunc::bzero);
+    TLI.setUnavailable(LibFunc::chmod);
+    TLI.setUnavailable(LibFunc::chown);
+    TLI.setUnavailable(LibFunc::closedir);
+    TLI.setUnavailable(LibFunc::ctermid);
+    TLI.setUnavailable(LibFunc::fdopen);
+    TLI.setUnavailable(LibFunc::ffs);
+    TLI.setUnavailable(LibFunc::fileno);
+    TLI.setUnavailable(LibFunc::flockfile);
+    TLI.setUnavailable(LibFunc::fseeko);
+    TLI.setUnavailable(LibFunc::fstat);
+    TLI.setUnavailable(LibFunc::fstatvfs);
+    TLI.setUnavailable(LibFunc::ftello);
+    TLI.setUnavailable(LibFunc::ftrylockfile);
+    TLI.setUnavailable(LibFunc::funlockfile);
+    TLI.setUnavailable(LibFunc::getc_unlocked);
+    TLI.setUnavailable(LibFunc::getitimer);
+    TLI.setUnavailable(LibFunc::getlogin_r);
+    TLI.setUnavailable(LibFunc::getpwnam);
+    TLI.setUnavailable(LibFunc::htonl);
+    TLI.setUnavailable(LibFunc::htons);
+    TLI.setUnavailable(LibFunc::lchown);
+    TLI.setUnavailable(LibFunc::lstat);
+    TLI.setUnavailable(LibFunc::memccpy);
+    TLI.setUnavailable(LibFunc::mkdir);
+    TLI.setUnavailable(LibFunc::ntohl);
+    TLI.setUnavailable(LibFunc::ntohs);
+    TLI.setUnavailable(LibFunc::open);
+    TLI.setUnavailable(LibFunc::opendir);
+    TLI.setUnavailable(LibFunc::pclose);
+    TLI.setUnavailable(LibFunc::popen);
+    TLI.setUnavailable(LibFunc::pread);
+    TLI.setUnavailable(LibFunc::pwrite);
+    TLI.setUnavailable(LibFunc::read);
+    TLI.setUnavailable(LibFunc::readlink);
+    TLI.setUnavailable(LibFunc::realpath);
+    TLI.setUnavailable(LibFunc::rmdir);
+    TLI.setUnavailable(LibFunc::setitimer);
+    TLI.setUnavailable(LibFunc::stat);
+    TLI.setUnavailable(LibFunc::statvfs);
     TLI.setUnavailable(LibFunc::stpcpy);
+    TLI.setUnavailable(LibFunc::stpncpy);
+    TLI.setUnavailable(LibFunc::strcasecmp);
+    TLI.setUnavailable(LibFunc::strncasecmp);
+    TLI.setUnavailable(LibFunc::times);
+    TLI.setUnavailable(LibFunc::uname);
+    TLI.setUnavailable(LibFunc::unlink);
+    TLI.setUnavailable(LibFunc::unsetenv);
+    TLI.setUnavailable(LibFunc::utime);
+    TLI.setUnavailable(LibFunc::utimes);
+    TLI.setUnavailable(LibFunc::write);
+
+    // Win32 does *not* provide provide these functions, but they are
+    // specified by C99:
+    TLI.setUnavailable(LibFunc::atoll);
+    TLI.setUnavailable(LibFunc::frexpf);
+    TLI.setUnavailable(LibFunc::llabs);
+  }
+
+  // ffsl is available on at least Darwin, Mac OS X, iOS, FreeBSD, and
+  // Linux (GLIBC):
+  // http://developer.apple.com/library/mac/#documentation/Darwin/Reference/ManPages/man3/ffsl.3.html
+  // http://svn.freebsd.org/base/user/eri/pf45/head/lib/libc/string/ffsl.c
+  // http://www.gnu.org/software/gnulib/manual/html_node/ffsl.html
+  switch (T.getOS()) {
+  case Triple::Darwin:
+  case Triple::MacOSX:
+  case Triple::IOS:
+  case Triple::FreeBSD:
+  case Triple::Linux:
+    break;
+  default:
+    TLI.setUnavailable(LibFunc::ffsl);
+  }
+
+  // ffsll is available on at least FreeBSD and Linux (GLIBC):
+  // http://svn.freebsd.org/base/user/eri/pf45/head/lib/libc/string/ffsll.c
+  // http://www.gnu.org/software/gnulib/manual/html_node/ffsll.html
+  switch (T.getOS()) {
+  case Triple::FreeBSD:
+  case Triple::Linux:
+    break;
+  default:
+    TLI.setUnavailable(LibFunc::ffsll);
+  }
+
+  // The following functions are available on at least Linux:
+  if (T.getOS() != Triple::Linux) {
+    TLI.setUnavailable(LibFunc::dunder_strdup);
+    TLI.setUnavailable(LibFunc::dunder_strtok_r);
+    TLI.setUnavailable(LibFunc::dunder_isoc99_scanf);
+    TLI.setUnavailable(LibFunc::dunder_isoc99_sscanf);
+    TLI.setUnavailable(LibFunc::under_IO_getc);
+    TLI.setUnavailable(LibFunc::under_IO_putc);
+    TLI.setUnavailable(LibFunc::memalign);
+    TLI.setUnavailable(LibFunc::fopen64);
+    TLI.setUnavailable(LibFunc::fseeko64);
+    TLI.setUnavailable(LibFunc::fstat64);
+    TLI.setUnavailable(LibFunc::fstatvfs64);
+    TLI.setUnavailable(LibFunc::ftello64);
+    TLI.setUnavailable(LibFunc::lstat64);
+    TLI.setUnavailable(LibFunc::open64);
+    TLI.setUnavailable(LibFunc::stat64);
+    TLI.setUnavailable(LibFunc::statvfs64);
+    TLI.setUnavailable(LibFunc::tmpfile64);
   }
 }
 
@@ -351,11 +597,40 @@ TargetLibraryInfo::TargetLibraryInfo(const TargetLibraryInfo &TLI)
   CustomNames = TLI.CustomNames;
 }
 
+namespace {
+struct StringComparator {
+  /// Compare two strings and return true if LHS is lexicographically less than
+  /// RHS. Requires that RHS doesn't contain any zero bytes.
+  bool operator()(const char *LHS, StringRef RHS) const {
+    // Compare prefixes with strncmp. If prefixes match we know that LHS is
+    // greater or equal to RHS as RHS can't contain any '\0'.
+    return std::strncmp(LHS, RHS.data(), RHS.size()) < 0;
+  }
+
+  // Provided for compatibility with MSVC's debug mode.
+  bool operator()(StringRef LHS, const char *RHS) const { return LHS < RHS; }
+  bool operator()(StringRef LHS, StringRef RHS) const { return LHS < RHS; }
+  bool operator()(const char *LHS, const char *RHS) const {
+    return std::strcmp(LHS, RHS) < 0;
+  }
+};
+}
+
 bool TargetLibraryInfo::getLibFunc(StringRef funcName,
                                    LibFunc::Func &F) const {
   const char **Start = &StandardNames[0];
   const char **End = &StandardNames[LibFunc::NumLibFuncs];
-  const char **I = std::lower_bound(Start, End, funcName);
+
+  // Filter out empty names and names containing null bytes, those can't be in
+  // our table.
+  if (funcName.empty() || funcName.find('\0') != StringRef::npos)
+    return false;
+
+  // Check for \01 prefix that is used to mangle __asm declarations and
+  // strip it if present.
+  if (funcName.front() == '\01')
+    funcName = funcName.substr(1);
+  const char **I = std::lower_bound(Start, End, funcName, StringComparator());
   if (I != End && *I == funcName) {
     F = (LibFunc::Func)(I - Start);
     return true;
diff --git a/lib/Target/TargetLoweringObjectFile.cpp b/lib/Target/TargetLoweringObjectFile.cpp
index 9d7e2b825f41..f5121e34f77f 100644
--- a/lib/Target/TargetLoweringObjectFile.cpp
+++ b/lib/Target/TargetLoweringObjectFile.cpp
@@ -13,21 +13,21 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/Target/TargetLoweringObjectFile.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalVariable.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/Target/Mangler.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetOptions.h"
 #include "llvm/Support/Dwarf.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/Mangler.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
 using namespace llvm;
 
 //===----------------------------------------------------------------------===//
@@ -285,35 +285,35 @@ TargetLoweringObjectFile::getSectionForConstant(SectionKind Kind) const {
   return DataSection;
 }
 
-/// getExprForDwarfGlobalReference - Return an MCExpr to use for a
+/// getTTypeGlobalReference - Return an MCExpr to use for a
 /// reference to the specified global variable from exception
 /// handling information.
 const MCExpr *TargetLoweringObjectFile::
-getExprForDwarfGlobalReference(const GlobalValue *GV, Mangler *Mang,
-                               MachineModuleInfo *MMI, unsigned Encoding,
-                               MCStreamer &Streamer) const {
-  const MCSymbol *Sym = Mang->getSymbol(GV);
-  return getExprForDwarfReference(Sym, Encoding, Streamer);
+getTTypeGlobalReference(const GlobalValue *GV, Mangler *Mang,
+                        MachineModuleInfo *MMI, unsigned Encoding,
+                        MCStreamer &Streamer) const {
+  const MCSymbolRefExpr *Ref =
+    MCSymbolRefExpr::Create(Mang->getSymbol(GV), getContext());
+
+  return getTTypeReference(Ref, Encoding, Streamer);
 }
 
 const MCExpr *TargetLoweringObjectFile::
-getExprForDwarfReference(const MCSymbol *Sym, unsigned Encoding,
-                         MCStreamer &Streamer) const {
-  const MCExpr *Res = MCSymbolRefExpr::Create(Sym, getContext());
-
+getTTypeReference(const MCSymbolRefExpr *Sym, unsigned Encoding,
+                  MCStreamer &Streamer) const {
   switch (Encoding & 0x70) {
   default:
     report_fatal_error("We do not support this DWARF encoding yet!");
   case dwarf::DW_EH_PE_absptr:
     // Do nothing special
-    return Res;
+    return Sym;
   case dwarf::DW_EH_PE_pcrel: {
     // Emit a label to the streamer for the current position.  This gives us
     // .-foo addressing.
     MCSymbol *PCSym = getContext().CreateTempSymbol();
     Streamer.EmitLabel(PCSym);
     const MCExpr *PC = MCSymbolRefExpr::Create(PCSym, getContext());
-    return MCBinaryExpr::CreateSub(Res, PC, getContext());
+    return MCBinaryExpr::CreateSub(Sym, PC, getContext());
   }
   }
 }
diff --git a/lib/Target/TargetMachine.cpp b/lib/Target/TargetMachine.cpp
index 382571982b96..e7282519d597 100644
--- a/lib/Target/TargetMachine.cpp
+++ b/lib/Target/TargetMachine.cpp
@@ -11,12 +11,14 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/GlobalAlias.h"
-#include "llvm/GlobalValue.h"
-#include "llvm/GlobalVariable.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/GlobalVariable.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCCodeGenInfo.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/Support/CommandLine.h"
 using namespace llvm;
 
@@ -61,6 +63,30 @@ TargetMachine::~TargetMachine() {
   delete AsmInfo;
 }
 
+/// \brief Reset the target options based on the function's attributes.
+void TargetMachine::resetTargetOptions(const MachineFunction *MF) const {
+  const Function *F = MF->getFunction();
+  TargetOptions &TO = MF->getTarget().Options;
+  
+#define RESET_OPTION(X, Y)                                              \
+  do {                                                                  \
+    if (F->hasFnAttribute(Y))                                           \
+      TO.X =                                                            \
+        (F->getAttributes().                                            \
+           getAttribute(AttributeSet::FunctionIndex,                    \
+                        Y).getValueAsString() == "true");               \
+  } while (0)
+
+  RESET_OPTION(NoFramePointerElim, "no-frame-pointer-elim");
+  RESET_OPTION(NoFramePointerElimNonLeaf, "no-frame-pointer-elim-non-leaf");
+  RESET_OPTION(LessPreciseFPMADOption, "less-precise-fpmad");
+  RESET_OPTION(UnsafeFPMath, "unsafe-fp-math");
+  RESET_OPTION(NoInfsFPMath, "no-infs-fp-math");
+  RESET_OPTION(NoNaNsFPMath, "no-nans-fp-math");
+  RESET_OPTION(UseSoftFloat, "use-soft-float");
+  RESET_OPTION(DisableTailCalls, "disable-tail-calls");
+}
+
 /// getRelocationModel - Returns the code generation relocation model. The
 /// choices are static, PIC, and dynamic-no-pic, and target default.
 Reloc::Model TargetMachine::getRelocationModel() const {
diff --git a/lib/Target/TargetMachineC.cpp b/lib/Target/TargetMachineC.cpp
index f69c2abd50d2..79f74bd66127 100644
--- a/lib/Target/TargetMachineC.cpp
+++ b/lib/Target/TargetMachineC.cpp
@@ -11,17 +11,17 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "llvm-c/TargetMachine.h"
 #include "llvm-c/Core.h"
 #include "llvm-c/Target.h"
-#include "llvm-c/TargetMachine.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Support/TargetRegistry.h"
-#include "llvm/Support/raw_ostream.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Module.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/CodeGen.h"
 #include "llvm/Support/FormattedStream.h"
-#include "llvm/Module.h"
-#include "llvm/PassManager.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetMachine.h"
 #include <cassert>
 #include <cstdlib>
 #include <cstring>
@@ -184,7 +184,7 @@ LLVMBool LLVMTargetMachineEmitToFile(LLVMTargetMachineRef T, LLVMModuleRef M,
   }
 
   if (TM->addPassesToEmitFile(pass, destf, ft)) {
-    error = "No DataLayout in TargetMachine";
+    error = "TargetMachine can't emit a file of this type";
     *ErrorMessage = strdup(error.c_str());
     return true;
   }
diff --git a/lib/Target/TargetRegisterInfo.cpp b/lib/Target/TargetRegisterInfo.cpp
deleted file mode 100644
index be8b58289039..000000000000
--- a/lib/Target/TargetRegisterInfo.cpp
+++ /dev/null
@@ -1,248 +0,0 @@
-//===- TargetRegisterInfo.cpp - Target Register Information Implementation ===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file implements the TargetRegisterInfo interface.
-//
-//===----------------------------------------------------------------------===//
-
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/Support/raw_ostream.h"
-
-using namespace llvm;
-
-TargetRegisterInfo::TargetRegisterInfo(const TargetRegisterInfoDesc *ID,
-                             regclass_iterator RCB, regclass_iterator RCE,
-                             const char *const *SRINames,
-                             const unsigned *SRILaneMasks)
-  : InfoDesc(ID), SubRegIndexNames(SRINames),
-    SubRegIndexLaneMasks(SRILaneMasks),
-    RegClassBegin(RCB), RegClassEnd(RCE) {
-}
-
-TargetRegisterInfo::~TargetRegisterInfo() {}
-
-void PrintReg::print(raw_ostream &OS) const {
-  if (!Reg)
-    OS << "%noreg";
-  else if (TargetRegisterInfo::isStackSlot(Reg))
-    OS << "SS#" << TargetRegisterInfo::stackSlot2Index(Reg);
-  else if (TargetRegisterInfo::isVirtualRegister(Reg))
-    OS << "%vreg" << TargetRegisterInfo::virtReg2Index(Reg);
-  else if (TRI && Reg < TRI->getNumRegs())
-    OS << '%' << TRI->getName(Reg);
-  else
-    OS << "%physreg" << Reg;
-  if (SubIdx) {
-    if (TRI)
-      OS << ':' << TRI->getSubRegIndexName(SubIdx);
-    else
-      OS << ":sub(" << SubIdx << ')';
-  }
-}
-
-void PrintRegUnit::print(raw_ostream &OS) const {
-  // Generic printout when TRI is missing.
-  if (!TRI) {
-    OS << "Unit~" << Unit;
-    return;
-  }
-
-  // Check for invalid register units.
-  if (Unit >= TRI->getNumRegUnits()) {
-    OS << "BadUnit~" << Unit;
-    return;
-  }
-
-  // Normal units have at least one root.
-  MCRegUnitRootIterator Roots(Unit, TRI);
-  assert(Roots.isValid() && "Unit has no roots.");
-  OS << TRI->getName(*Roots);
-  for (++Roots; Roots.isValid(); ++Roots)
-    OS << '~' << TRI->getName(*Roots);
-}
-
-/// getAllocatableClass - Return the maximal subclass of the given register
-/// class that is alloctable, or NULL.
-const TargetRegisterClass *
-TargetRegisterInfo::getAllocatableClass(const TargetRegisterClass *RC) const {
-  if (!RC || RC->isAllocatable())
-    return RC;
-
-  const unsigned *SubClass = RC->getSubClassMask();
-  for (unsigned Base = 0, BaseE = getNumRegClasses();
-       Base < BaseE; Base += 32) {
-    unsigned Idx = Base;
-    for (unsigned Mask = *SubClass++; Mask; Mask >>= 1) {
-      unsigned Offset = CountTrailingZeros_32(Mask);
-      const TargetRegisterClass *SubRC = getRegClass(Idx + Offset);
-      if (SubRC->isAllocatable())
-        return SubRC;
-      Mask >>= Offset;
-      Idx += Offset + 1;
-    }
-  }
-  return NULL;
-}
-
-/// getMinimalPhysRegClass - Returns the Register Class of a physical
-/// register of the given type, picking the most sub register class of
-/// the right type that contains this physreg.
-const TargetRegisterClass *
-TargetRegisterInfo::getMinimalPhysRegClass(unsigned reg, EVT VT) const {
-  assert(isPhysicalRegister(reg) && "reg must be a physical register");
-
-  // Pick the most sub register class of the right type that contains
-  // this physreg.
-  const TargetRegisterClass* BestRC = 0;
-  for (regclass_iterator I = regclass_begin(), E = regclass_end(); I != E; ++I){
-    const TargetRegisterClass* RC = *I;
-    if ((VT == MVT::Other || RC->hasType(VT)) && RC->contains(reg) &&
-        (!BestRC || BestRC->hasSubClass(RC)))
-      BestRC = RC;
-  }
-
-  assert(BestRC && "Couldn't find the register class");
-  return BestRC;
-}
-
-/// getAllocatableSetForRC - Toggle the bits that represent allocatable
-/// registers for the specific register class.
-static void getAllocatableSetForRC(const MachineFunction &MF,
-                                   const TargetRegisterClass *RC, BitVector &R){
-  assert(RC->isAllocatable() && "invalid for nonallocatable sets");
-  ArrayRef<uint16_t> Order = RC->getRawAllocationOrder(MF);
-  for (unsigned i = 0; i != Order.size(); ++i)
-    R.set(Order[i]);
-}
-
-BitVector TargetRegisterInfo::getAllocatableSet(const MachineFunction &MF,
-                                          const TargetRegisterClass *RC) const {
-  BitVector Allocatable(getNumRegs());
-  if (RC) {
-    // A register class with no allocatable subclass returns an empty set.
-    const TargetRegisterClass *SubClass = getAllocatableClass(RC);
-    if (SubClass)
-      getAllocatableSetForRC(MF, SubClass, Allocatable);
-  } else {
-    for (TargetRegisterInfo::regclass_iterator I = regclass_begin(),
-         E = regclass_end(); I != E; ++I)
-      if ((*I)->isAllocatable())
-        getAllocatableSetForRC(MF, *I, Allocatable);
-  }
-
-  // Mask out the reserved registers
-  BitVector Reserved = getReservedRegs(MF);
-  Allocatable &= Reserved.flip();
-
-  return Allocatable;
-}
-
-static inline
-const TargetRegisterClass *firstCommonClass(const uint32_t *A,
-                                            const uint32_t *B,
-                                            const TargetRegisterInfo *TRI) {
-  for (unsigned I = 0, E = TRI->getNumRegClasses(); I < E; I += 32)
-    if (unsigned Common = *A++ & *B++)
-      return TRI->getRegClass(I + CountTrailingZeros_32(Common));
-  return 0;
-}
-
-const TargetRegisterClass *
-TargetRegisterInfo::getCommonSubClass(const TargetRegisterClass *A,
-                                      const TargetRegisterClass *B) const {
-  // First take care of the trivial cases.
-  if (A == B)
-    return A;
-  if (!A || !B)
-    return 0;
-
-  // Register classes are ordered topologically, so the largest common
-  // sub-class it the common sub-class with the smallest ID.
-  return firstCommonClass(A->getSubClassMask(), B->getSubClassMask(), this);
-}
-
-const TargetRegisterClass *
-TargetRegisterInfo::getMatchingSuperRegClass(const TargetRegisterClass *A,
-                                             const TargetRegisterClass *B,
-                                             unsigned Idx) const {
-  assert(A && B && "Missing register class");
-  assert(Idx && "Bad sub-register index");
-
-  // Find Idx in the list of super-register indices.
-  for (SuperRegClassIterator RCI(B, this); RCI.isValid(); ++RCI)
-    if (RCI.getSubReg() == Idx)
-      // The bit mask contains all register classes that are projected into B
-      // by Idx. Find a class that is also a sub-class of A.
-      return firstCommonClass(RCI.getMask(), A->getSubClassMask(), this);
-  return 0;
-}
-
-const TargetRegisterClass *TargetRegisterInfo::
-getCommonSuperRegClass(const TargetRegisterClass *RCA, unsigned SubA,
-                       const TargetRegisterClass *RCB, unsigned SubB,
-                       unsigned &PreA, unsigned &PreB) const {
-  assert(RCA && SubA && RCB && SubB && "Invalid arguments");
-
-  // Search all pairs of sub-register indices that project into RCA and RCB
-  // respectively. This is quadratic, but usually the sets are very small. On
-  // most targets like X86, there will only be a single sub-register index
-  // (e.g., sub_16bit projecting into GR16).
-  //
-  // The worst case is a register class like DPR on ARM.
-  // We have indices dsub_0..dsub_7 projecting into that class.
-  //
-  // It is very common that one register class is a sub-register of the other.
-  // Arrange for RCA to be the larger register so the answer will be found in
-  // the first iteration. This makes the search linear for the most common
-  // case.
-  const TargetRegisterClass *BestRC = 0;
-  unsigned *BestPreA = &PreA;
-  unsigned *BestPreB = &PreB;
-  if (RCA->getSize() < RCB->getSize()) {
-    std::swap(RCA, RCB);
-    std::swap(SubA, SubB);
-    std::swap(BestPreA, BestPreB);
-  }
-
-  // Also terminate the search one we have found a register class as small as
-  // RCA.
-  unsigned MinSize = RCA->getSize();
-
-  for (SuperRegClassIterator IA(RCA, this, true); IA.isValid(); ++IA) {
-    unsigned FinalA = composeSubRegIndices(IA.getSubReg(), SubA);
-    for (SuperRegClassIterator IB(RCB, this, true); IB.isValid(); ++IB) {
-      // Check if a common super-register class exists for this index pair.
-      const TargetRegisterClass *RC =
-        firstCommonClass(IA.getMask(), IB.getMask(), this);
-      if (!RC || RC->getSize() < MinSize)
-        continue;
-
-      // The indexes must compose identically: PreA+SubA == PreB+SubB.
-      unsigned FinalB = composeSubRegIndices(IB.getSubReg(), SubB);
-      if (FinalA != FinalB)
-        continue;
-
-      // Is RC a better candidate than BestRC?
-      if (BestRC && RC->getSize() >= BestRC->getSize())
-        continue;
-
-      // Yes, RC is the smallest super-register seen so far.
-      BestRC = RC;
-      *BestPreA = IA.getSubReg();
-      *BestPreB = IB.getSubReg();
-
-      // Bail early if we reached MinSize. We won't find a better candidate.
-      if (BestRC->getSize() == MinSize)
-        return BestRC;
-    }
-  }
-  return BestRC;
-}
diff --git a/lib/Target/TargetSubtargetInfo.cpp b/lib/Target/TargetSubtargetInfo.cpp
index 59ffdea00ea6..af0cef62d552 100644
--- a/lib/Target/TargetSubtargetInfo.cpp
+++ b/lib/Target/TargetSubtargetInfo.cpp
@@ -22,6 +22,10 @@ TargetSubtargetInfo::TargetSubtargetInfo() {}
 
 TargetSubtargetInfo::~TargetSubtargetInfo() {}
 
+bool TargetSubtargetInfo::enableMachineScheduler() const {
+  return false;
+}
+
 bool TargetSubtargetInfo::enablePostRAScheduler(
           CodeGenOpt::Level OptLevel,
           AntiDepBreakMode& Mode,
diff --git a/lib/Target/TargetTransformImpl.cpp b/lib/Target/TargetTransformImpl.cpp
deleted file mode 100644
index b36e6f858f72..000000000000
--- a/lib/Target/TargetTransformImpl.cpp
+++ /dev/null
@@ -1,353 +0,0 @@
-// llvm/Target/TargetTransformImpl.cpp - Target Loop Trans Info ---*- C++ -*-=//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#include "llvm/Target/TargetTransformImpl.h"
-#include "llvm/Target/TargetLowering.h"
-#include <utility>
-
-using namespace llvm;
-
-//===----------------------------------------------------------------------===//
-//
-// Calls used by scalar transformations.
-//
-//===----------------------------------------------------------------------===//
-
-bool ScalarTargetTransformImpl::isLegalAddImmediate(int64_t imm) const {
-  return TLI->isLegalAddImmediate(imm);
-}
-
-bool ScalarTargetTransformImpl::isLegalICmpImmediate(int64_t imm) const {
-  return TLI->isLegalICmpImmediate(imm);
-}
-
-bool ScalarTargetTransformImpl::isLegalAddressingMode(const AddrMode &AM,
-                                                      Type *Ty) const {
-  return TLI->isLegalAddressingMode(AM, Ty);
-}
-
-bool ScalarTargetTransformImpl::isTruncateFree(Type *Ty1, Type *Ty2) const {
-  return TLI->isTruncateFree(Ty1, Ty2);
-}
-
-bool ScalarTargetTransformImpl::isTypeLegal(Type *Ty) const {
-  EVT T = TLI->getValueType(Ty);
-  return TLI->isTypeLegal(T);
-}
-
-unsigned ScalarTargetTransformImpl::getJumpBufAlignment() const {
-  return TLI->getJumpBufAlignment();
-}
-
-unsigned ScalarTargetTransformImpl::getJumpBufSize() const {
-  return TLI->getJumpBufSize();
-}
-
-bool ScalarTargetTransformImpl::shouldBuildLookupTables() const {
-  return TLI->supportJumpTables() &&
-      (TLI->isOperationLegalOrCustom(ISD::BR_JT, MVT::Other) ||
-       TLI->isOperationLegalOrCustom(ISD::BRIND, MVT::Other));
-}
-
-//===----------------------------------------------------------------------===//
-//
-// Calls used by the vectorizers.
-//
-//===----------------------------------------------------------------------===//
-int VectorTargetTransformImpl::InstructionOpcodeToISD(unsigned Opcode) const {
-  enum InstructionOpcodes {
-#define HANDLE_INST(NUM, OPCODE, CLASS) OPCODE = NUM,
-#define LAST_OTHER_INST(NUM) InstructionOpcodesCount = NUM
-#include "llvm/Instruction.def"
-  };
-  switch (static_cast<InstructionOpcodes>(Opcode)) {
-  case Ret:            return 0;
-  case Br:             return 0;
-  case Switch:         return 0;
-  case IndirectBr:     return 0;
-  case Invoke:         return 0;
-  case Resume:         return 0;
-  case Unreachable:    return 0;
-  case Add:            return ISD::ADD;
-  case FAdd:           return ISD::FADD;
-  case Sub:            return ISD::SUB;
-  case FSub:           return ISD::FSUB;
-  case Mul:            return ISD::MUL;
-  case FMul:           return ISD::FMUL;
-  case UDiv:           return ISD::UDIV;
-  case SDiv:           return ISD::UDIV;
-  case FDiv:           return ISD::FDIV;
-  case URem:           return ISD::UREM;
-  case SRem:           return ISD::SREM;
-  case FRem:           return ISD::FREM;
-  case Shl:            return ISD::SHL;
-  case LShr:           return ISD::SRL;
-  case AShr:           return ISD::SRA;
-  case And:            return ISD::AND;
-  case Or:             return ISD::OR;
-  case Xor:            return ISD::XOR;
-  case Alloca:         return 0;
-  case Load:           return ISD::LOAD;
-  case Store:          return ISD::STORE;
-  case GetElementPtr:  return 0;
-  case Fence:          return 0;
-  case AtomicCmpXchg:  return 0;
-  case AtomicRMW:      return 0;
-  case Trunc:          return ISD::TRUNCATE;
-  case ZExt:           return ISD::ZERO_EXTEND;
-  case SExt:           return ISD::SIGN_EXTEND;
-  case FPToUI:         return ISD::FP_TO_UINT;
-  case FPToSI:         return ISD::FP_TO_SINT;
-  case UIToFP:         return ISD::UINT_TO_FP;
-  case SIToFP:         return ISD::SINT_TO_FP;
-  case FPTrunc:        return ISD::FP_ROUND;
-  case FPExt:          return ISD::FP_EXTEND;
-  case PtrToInt:       return ISD::BITCAST;
-  case IntToPtr:       return ISD::BITCAST;
-  case BitCast:        return ISD::BITCAST;
-  case ICmp:           return ISD::SETCC;
-  case FCmp:           return ISD::SETCC;
-  case PHI:            return 0;
-  case Call:           return 0;
-  case Select:         return ISD::SELECT;
-  case UserOp1:        return 0;
-  case UserOp2:        return 0;
-  case VAArg:          return 0;
-  case ExtractElement: return ISD::EXTRACT_VECTOR_ELT;
-  case InsertElement:  return ISD::INSERT_VECTOR_ELT;
-  case ShuffleVector:  return ISD::VECTOR_SHUFFLE;
-  case ExtractValue:   return ISD::MERGE_VALUES;
-  case InsertValue:    return ISD::MERGE_VALUES;
-  case LandingPad:     return 0;
-  }
-
-  llvm_unreachable("Unknown instruction type encountered!");
-}
-
-std::pair<unsigned, MVT>
-VectorTargetTransformImpl::getTypeLegalizationCost(Type *Ty) const {
-
-  LLVMContext &C = Ty->getContext();
-  EVT MTy = TLI->getValueType(Ty);
-
-  unsigned Cost = 1;
-  // We keep legalizing the type until we find a legal kind. We assume that
-  // the only operation that costs anything is the split. After splitting
-  // we need to handle two types.
-  while (true) {
-    TargetLowering::LegalizeKind LK = TLI->getTypeConversion(C, MTy);
-
-    if (LK.first == TargetLowering::TypeLegal)
-      return std::make_pair(Cost, MTy.getSimpleVT());
-
-    if (LK.first == TargetLowering::TypeSplitVector ||
-        LK.first == TargetLowering::TypeExpandInteger)
-      Cost *= 2;
-
-    // Keep legalizing the type.
-    MTy = LK.second;
-  }
-}
-
-unsigned
-VectorTargetTransformImpl::getScalarizationOverhead(Type *Ty,
-                                                    bool Insert,
-                                                    bool Extract) const {
-  assert (Ty->isVectorTy() && "Can only scalarize vectors");
-  unsigned Cost = 0;
-
-  for (int i = 0, e = Ty->getVectorNumElements(); i < e; ++i) {
-    if (Insert)
-      Cost += getVectorInstrCost(Instruction::InsertElement, Ty, i);
-    if (Extract)
-      Cost += getVectorInstrCost(Instruction::ExtractElement, Ty, i);
-  }
-
-  return Cost;
-}
-
-unsigned VectorTargetTransformImpl::getArithmeticInstrCost(unsigned Opcode,
-                                                           Type *Ty) const {
-  // Check if any of the operands are vector operands.
-  int ISD = InstructionOpcodeToISD(Opcode);
-  assert(ISD && "Invalid opcode");
-
-  std::pair<unsigned, MVT> LT = getTypeLegalizationCost(Ty);
-
-  if (!TLI->isOperationExpand(ISD, LT.second)) {
-    // The operation is legal. Assume it costs 1. Multiply
-    // by the type-legalization overhead.
-    return LT.first * 1;
-  }
-
-  // Else, assume that we need to scalarize this op.
-  if (Ty->isVectorTy()) {
-    unsigned Num = Ty->getVectorNumElements();
-    unsigned Cost = getArithmeticInstrCost(Opcode, Ty->getScalarType());
-    // return the cost of multiple scalar invocation plus the cost of inserting
-    // and extracting the values.
-    return getScalarizationOverhead(Ty, true, true) + Num * Cost;
-  }
-
-  // We don't know anything about this scalar instruction.
-  return 1;
-}
-
-unsigned VectorTargetTransformImpl::getBroadcastCost(Type *Tp) const {
-  return 1;
-}
-
-unsigned VectorTargetTransformImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
-                                  Type *Src) const {
-  int ISD = InstructionOpcodeToISD(Opcode);
-  assert(ISD && "Invalid opcode");
-
-  std::pair<unsigned, MVT> SrcLT = getTypeLegalizationCost(Src);
-  std::pair<unsigned, MVT> DstLT = getTypeLegalizationCost(Dst);
-
-  // Handle scalar conversions.
-  if (!Src->isVectorTy() && !Dst->isVectorTy()) {
-
-    // Scalar bitcasts are usually free.
-    if (Opcode == Instruction::BitCast)
-      return 0;
-
-    if (Opcode == Instruction::Trunc &&
-        TLI->isTruncateFree(SrcLT.second, DstLT.second))
-      return 0;
-
-    if (Opcode == Instruction::ZExt &&
-        TLI->isZExtFree(SrcLT.second, DstLT.second))
-      return 0;
-
-    // Just check the op cost. If the operation is legal then assume it costs 1.
-    if (!TLI->isOperationExpand(ISD, DstLT.second))
-      return  1;
-
-    // Assume that illegal scalar instruction are expensive.
-    return 4;
-  }
-
-  // Check vector-to-vector casts.
-  if (Dst->isVectorTy() && Src->isVectorTy()) {
-
-    // If the cast is between same-sized registers, then the check is simple.
-    if (SrcLT.first == DstLT.first &&
-        SrcLT.second.getSizeInBits() == DstLT.second.getSizeInBits()) {
-
-      // Bitcast between types that are legalized to the same type are free.
-      if (Opcode == Instruction::BitCast || Opcode == Instruction::Trunc)
-        return 0;
-
-      // Assume that Zext is done using AND.
-      if (Opcode == Instruction::ZExt)
-        return 1;
-
-      // Assume that sext is done using SHL and SRA.
-      if (Opcode == Instruction::SExt)
-        return 2;
-
-      // Just check the op cost. If the operation is legal then assume it costs
-      // 1 and multiply by the type-legalization overhead.
-      if (!TLI->isOperationExpand(ISD, DstLT.second))
-        return SrcLT.first * 1;
-    }
-
-    // If we are converting vectors and the operation is illegal, or
-    // if the vectors are legalized to different types, estimate the
-    // scalarization costs.
-    unsigned Num = Dst->getVectorNumElements();
-    unsigned Cost = getCastInstrCost(Opcode, Dst->getScalarType(),
-                                     Src->getScalarType());
-
-    // Return the cost of multiple scalar invocation plus the cost of
-    // inserting and extracting the values.
-    return getScalarizationOverhead(Dst, true, true) + Num * Cost;
-  }
-
-  // We already handled vector-to-vector and scalar-to-scalar conversions. This 
-  // is where we handle bitcast between vectors and scalars. We need to assume
-  //  that the conversion is scalarized in one way or another.
-  if (Opcode == Instruction::BitCast)
-    // Illegal bitcasts are done by storing and loading from a stack slot.
-    return (Src->isVectorTy()? getScalarizationOverhead(Src, false, true):0) +
-           (Dst->isVectorTy()? getScalarizationOverhead(Dst, true, false):0);
-
-  llvm_unreachable("Unhandled cast");
- }
-
-unsigned VectorTargetTransformImpl::getCFInstrCost(unsigned Opcode) const {
-  return 1;
-}
-
-unsigned VectorTargetTransformImpl::getCmpSelInstrCost(unsigned Opcode,
-                                                       Type *ValTy,
-                                                       Type *CondTy) const {
-  int ISD = InstructionOpcodeToISD(Opcode);
-  assert(ISD && "Invalid opcode");
-
-  // Selects on vectors are actually vector selects.
-  if (ISD == ISD::SELECT) {
-    assert(CondTy && "CondTy must exist");
-    if (CondTy->isVectorTy())
-      ISD = ISD::VSELECT;
-  }
-
-  std::pair<unsigned, MVT> LT = getTypeLegalizationCost(ValTy);
-
-  if (!TLI->isOperationExpand(ISD, LT.second)) {
-    // The operation is legal. Assume it costs 1. Multiply
-    // by the type-legalization overhead.
-    return LT.first * 1;
-  }
-
-  // Otherwise, assume that the cast is scalarized.
-  if (ValTy->isVectorTy()) {
-    unsigned Num = ValTy->getVectorNumElements();
-    if (CondTy)
-      CondTy = CondTy->getScalarType();
-    unsigned Cost = getCmpSelInstrCost(Opcode, ValTy->getScalarType(),
-                                       CondTy);
-
-    // Return the cost of multiple scalar invocation plus the cost of inserting
-    // and extracting the values.
-    return getScalarizationOverhead(ValTy, true, false) + Num * Cost;
-  }
-
-  // Unknown scalar opcode.
-  return 1;
-}
-
-unsigned VectorTargetTransformImpl::getVectorInstrCost(unsigned Opcode,
-                                                       Type *Val,
-                                                       unsigned Index) const {
-  return 1;
-}
-
-unsigned
-VectorTargetTransformImpl::getInstrCost(unsigned Opcode, Type *Ty1,
-                                        Type *Ty2) const {
-  return 1;
-}
-
-unsigned
-VectorTargetTransformImpl::getMemoryOpCost(unsigned Opcode, Type *Src,
-                                           unsigned Alignment,
-                                           unsigned AddressSpace) const {
-  std::pair<unsigned, MVT> LT = getTypeLegalizationCost(Src);
-
-  // Assume that all loads of legal types cost 1.
-  return LT.first;
-}
-
-unsigned
-VectorTargetTransformImpl::getNumberOfParts(Type *Tp) const {
-  std::pair<unsigned, MVT> LT = getTypeLegalizationCost(Tp);
-  return LT.first;
-}
diff --git a/lib/Target/X86/AsmParser/CMakeLists.txt b/lib/Target/X86/AsmParser/CMakeLists.txt
index 47489bb06c4e..54204d4b6390 100644
--- a/lib/Target/X86/AsmParser/CMakeLists.txt
+++ b/lib/Target/X86/AsmParser/CMakeLists.txt
@@ -1,7 +1,6 @@
 include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
 
 add_llvm_library(LLVMX86AsmParser
-  X86AsmLexer.cpp
   X86AsmParser.cpp
   )
 
diff --git a/lib/Target/X86/AsmParser/X86AsmLexer.cpp b/lib/Target/X86/AsmParser/X86AsmLexer.cpp
deleted file mode 100644
index 66ad35370936..000000000000
--- a/lib/Target/X86/AsmParser/X86AsmLexer.cpp
+++ /dev/null
@@ -1,159 +0,0 @@
-//===-- X86AsmLexer.cpp - Tokenize X86 assembly to AsmTokens --------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#include "MCTargetDesc/X86BaseInfo.h"
-#include "llvm/MC/MCAsmInfo.h"
-#include "llvm/MC/MCParser/MCAsmLexer.h"
-#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
-#include "llvm/MC/MCTargetAsmLexer.h"
-#include "llvm/Support/TargetRegistry.h"
-#include "llvm/ADT/SmallVector.h"
-
-using namespace llvm;
-
-namespace {
-
-class X86AsmLexer : public MCTargetAsmLexer {
-  const MCAsmInfo &AsmInfo;
-
-  bool tentativeIsValid;
-  AsmToken tentativeToken;
-
-  const AsmToken &lexTentative() {
-    tentativeToken = getLexer()->Lex();
-    tentativeIsValid = true;
-    return tentativeToken;
-  }
-
-  const AsmToken &lexDefinite() {
-    if (tentativeIsValid) {
-      tentativeIsValid = false;
-      return tentativeToken;
-    }
-    return getLexer()->Lex();
-  }
-
-  AsmToken LexTokenATT();
-  AsmToken LexTokenIntel();
-protected:
-  AsmToken LexToken() {
-    if (!Lexer) {
-      SetError(SMLoc(), "No MCAsmLexer installed");
-      return AsmToken(AsmToken::Error, "", 0);
-    }
-
-    switch (AsmInfo.getAssemblerDialect()) {
-    default:
-      SetError(SMLoc(), "Unhandled dialect");
-      return AsmToken(AsmToken::Error, "", 0);
-    case 0:
-      return LexTokenATT();
-    case 1:
-      return LexTokenIntel();
-    }
-  }
-public:
-  X86AsmLexer(const Target &T, const MCRegisterInfo &MRI, const MCAsmInfo &MAI)
-    : MCTargetAsmLexer(T), AsmInfo(MAI), tentativeIsValid(false) {
-  }
-};
-
-} // end anonymous namespace
-
-#define GET_REGISTER_MATCHER
-#include "X86GenAsmMatcher.inc"
-
-AsmToken X86AsmLexer::LexTokenATT() {
-  AsmToken lexedToken = lexDefinite();
-
-  switch (lexedToken.getKind()) {
-  default:
-    return lexedToken;
-  case AsmToken::Error:
-    SetError(Lexer->getErrLoc(), Lexer->getErr());
-    return lexedToken;
-
-  case AsmToken::Percent: {
-    const AsmToken &nextToken = lexTentative();
-    if (nextToken.getKind() != AsmToken::Identifier)
-      return lexedToken;
-
-    if (unsigned regID = MatchRegisterName(nextToken.getString())) {
-      lexDefinite();
-
-      // FIXME: This is completely wrong when there is a space or other
-      // punctuation between the % and the register name.
-      StringRef regStr(lexedToken.getString().data(),
-                       lexedToken.getString().size() +
-                       nextToken.getString().size());
-
-      return AsmToken(AsmToken::Register, regStr,
-                      static_cast<int64_t>(regID));
-    }
-
-    // Match register name failed.  If this is "db[0-7]", match it as an alias
-    // for dr[0-7].
-    if (nextToken.getString().size() == 3 &&
-        nextToken.getString().startswith("db")) {
-      int RegNo = -1;
-      switch (nextToken.getString()[2]) {
-      case '0': RegNo = X86::DR0; break;
-      case '1': RegNo = X86::DR1; break;
-      case '2': RegNo = X86::DR2; break;
-      case '3': RegNo = X86::DR3; break;
-      case '4': RegNo = X86::DR4; break;
-      case '5': RegNo = X86::DR5; break;
-      case '6': RegNo = X86::DR6; break;
-      case '7': RegNo = X86::DR7; break;
-      }
-
-      if (RegNo != -1) {
-        lexDefinite();
-
-        // FIXME: This is completely wrong when there is a space or other
-        // punctuation between the % and the register name.
-        StringRef regStr(lexedToken.getString().data(),
-                         lexedToken.getString().size() +
-                         nextToken.getString().size());
-        return AsmToken(AsmToken::Register, regStr,
-                        static_cast<int64_t>(RegNo));
-      }
-    }
-
-
-    return lexedToken;
-  }
-  }
-}
-
-AsmToken X86AsmLexer::LexTokenIntel() {
-  const AsmToken &lexedToken = lexDefinite();
-
-  switch(lexedToken.getKind()) {
-  default:
-    return lexedToken;
-  case AsmToken::Error:
-    SetError(Lexer->getErrLoc(), Lexer->getErr());
-    return lexedToken;
-  case AsmToken::Identifier: {
-    unsigned regID = MatchRegisterName(lexedToken.getString().lower());
-
-    if (regID)
-      return AsmToken(AsmToken::Register,
-                      lexedToken.getString(),
-                      static_cast<int64_t>(regID));
-    return lexedToken;
-  }
-  }
-}
-
-extern "C" void LLVMInitializeX86AsmLexer() {
-  RegisterMCAsmLexer<X86AsmLexer> X(TheX86_32Target);
-  RegisterMCAsmLexer<X86AsmLexer> Y(TheX86_64Target);
-}
diff --git a/lib/Target/X86/AsmParser/X86AsmParser.cpp b/lib/Target/X86/AsmParser/X86AsmParser.cpp
index ce446e75737c..e4623228b397 100644
--- a/lib/Target/X86/AsmParser/X86AsmParser.cpp
+++ b/lib/Target/X86/AsmParser/X86AsmParser.cpp
@@ -8,21 +8,22 @@
 //===----------------------------------------------------------------------===//
 
 #include "MCTargetDesc/X86BaseInfo.h"
-#include "llvm/MC/MCTargetAsmParser.h"
-#include "llvm/MC/MCStreamer.h"
-#include "llvm/MC/MCExpr.h"
-#include "llvm/MC/MCSymbol.h"
-#include "llvm/MC/MCInst.h"
-#include "llvm/MC/MCRegisterInfo.h"
-#include "llvm/MC/MCSubtargetInfo.h"
-#include "llvm/MC/MCParser/MCAsmLexer.h"
-#include "llvm/MC/MCParser/MCAsmParser.h"
-#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
 #include "llvm/ADT/APFloat.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/ADT/Twine.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCParser/MCAsmLexer.h"
+#include "llvm/MC/MCParser/MCAsmParser.h"
+#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/MC/MCSymbol.h"
+#include "llvm/MC/MCTargetAsmParser.h"
 #include "llvm/Support/SourceMgr.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
@@ -57,11 +58,18 @@ private:
   X86Operand *ParseATTOperand();
   X86Operand *ParseIntelOperand();
   X86Operand *ParseIntelOffsetOfOperator(SMLoc StartLoc);
-  X86Operand *ParseIntelTypeOperator(SMLoc StartLoc);
-  X86Operand *ParseIntelMemOperand(unsigned SegReg, SMLoc StartLoc);
-  X86Operand *ParseIntelBracExpression(unsigned SegReg, unsigned Size);
+  X86Operand *ParseIntelOperator(SMLoc StartLoc, unsigned OpKind);
+  X86Operand *ParseIntelMemOperand(unsigned SegReg, uint64_t ImmDisp,
+                                   SMLoc StartLoc);
+  X86Operand *ParseIntelBracExpression(unsigned SegReg, uint64_t ImmDisp,
+                                       unsigned Size);
+  X86Operand *ParseIntelVarWithQualifier(const MCExpr *&Disp,
+                                         SMLoc &IdentStart);
   X86Operand *ParseMemOperand(unsigned SegReg, SMLoc StartLoc);
 
+  X86Operand *CreateMemForInlineAsm(const MCExpr *Disp, SMLoc Start, SMLoc End,
+                                    SMLoc SizeDirLoc, unsigned Size);
+
   bool ParseIntelDotOperator(const MCExpr *Disp, const MCExpr **NewDisp,
                              SmallString<64> &Err);
 
@@ -168,31 +176,35 @@ struct X86Operand : public MCParsedAsmOperand {
 
   SMLoc StartLoc, EndLoc;
   SMLoc OffsetOfLoc;
+  bool AddressOf;
+
+  struct TokOp {
+    const char *Data;
+    unsigned Length;
+  };
+
+  struct RegOp {
+    unsigned RegNo;
+  };
+
+  struct ImmOp {
+    const MCExpr *Val;
+  };
+
+  struct MemOp {
+    unsigned SegReg;
+    const MCExpr *Disp;
+    unsigned BaseReg;
+    unsigned IndexReg;
+    unsigned Scale;
+    unsigned Size;
+  };
 
   union {
-    struct {
-      const char *Data;
-      unsigned Length;
-    } Tok;
-
-    struct {
-      unsigned RegNo;
-    } Reg;
-
-    struct {
-      const MCExpr *Val;
-      bool NeedAsmRewrite;
-    } Imm;
-
-    struct {
-      unsigned SegReg;
-      const MCExpr *Disp;
-      unsigned BaseReg;
-      unsigned IndexReg;
-      unsigned Scale;
-      unsigned Size;
-      bool NeedSizeDir;
-    } Mem;
+    struct TokOp Tok;
+    struct RegOp Reg;
+    struct ImmOp Imm;
+    struct MemOp Mem;
   };
 
   X86Operand(KindTy K, SMLoc Start, SMLoc End)
@@ -230,11 +242,6 @@ struct X86Operand : public MCParsedAsmOperand {
     return Imm.Val;
   }
 
-  bool needAsmRewrite() const {
-    assert(Kind == Immediate && "Invalid access!");
-    return Imm.NeedAsmRewrite;
-  }
-
   const MCExpr *getMemDisp() const {
     assert(Kind == Memory && "Invalid access!");
     return Mem.Disp;
@@ -331,18 +338,12 @@ struct X86Operand : public MCParsedAsmOperand {
     return isImmSExti64i32Value(CE->getValue());
   }
 
-  unsigned getMemSize() const {
-    assert(Kind == Memory && "Invalid access!");
-    return Mem.Size;
-  }
-
   bool isOffsetOf() const {
     return OffsetOfLoc.getPointer();
   }
 
-  bool needSizeDirective() const {
-    assert(Kind == Memory && "Invalid access!");
-    return Mem.NeedSizeDir;
+  bool needAddressOf() const {
+    return AddressOf;
   }
 
   bool isMem() const { return Kind == Memory; }
@@ -463,7 +464,7 @@ struct X86Operand : public MCParsedAsmOperand {
   }
 
   static X86Operand *CreateToken(StringRef Str, SMLoc Loc) {
-    SMLoc EndLoc = SMLoc::getFromPointer(Loc.getPointer() + Str.size() - 1);
+    SMLoc EndLoc = SMLoc::getFromPointer(Loc.getPointer() + Str.size());
     X86Operand *Res = new X86Operand(Token, Loc, EndLoc);
     Res->Tok.Data = Str.data();
     Res->Tok.Length = Str.size();
@@ -471,24 +472,24 @@ struct X86Operand : public MCParsedAsmOperand {
   }
 
   static X86Operand *CreateReg(unsigned RegNo, SMLoc StartLoc, SMLoc EndLoc,
+                               bool AddressOf = false,
                                SMLoc OffsetOfLoc = SMLoc()) {
     X86Operand *Res = new X86Operand(Register, StartLoc, EndLoc);
     Res->Reg.RegNo = RegNo;
+    Res->AddressOf = AddressOf;
     Res->OffsetOfLoc = OffsetOfLoc;
     return Res;
   }
 
-  static X86Operand *CreateImm(const MCExpr *Val, SMLoc StartLoc, SMLoc EndLoc,
-                               bool NeedRewrite = true){
+  static X86Operand *CreateImm(const MCExpr *Val, SMLoc StartLoc, SMLoc EndLoc){
     X86Operand *Res = new X86Operand(Immediate, StartLoc, EndLoc);
     Res->Imm.Val = Val;
-    Res->Imm.NeedAsmRewrite = NeedRewrite;
     return Res;
   }
 
   /// Create an absolute memory operand.
   static X86Operand *CreateMem(const MCExpr *Disp, SMLoc StartLoc, SMLoc EndLoc,
-                               unsigned Size = 0, bool NeedSizeDir = false){
+                               unsigned Size = 0) {
     X86Operand *Res = new X86Operand(Memory, StartLoc, EndLoc);
     Res->Mem.SegReg   = 0;
     Res->Mem.Disp     = Disp;
@@ -496,7 +497,7 @@ struct X86Operand : public MCParsedAsmOperand {
     Res->Mem.IndexReg = 0;
     Res->Mem.Scale    = 1;
     Res->Mem.Size     = Size;
-    Res->Mem.NeedSizeDir = NeedSizeDir;
+    Res->AddressOf = false;
     return Res;
   }
 
@@ -504,7 +505,7 @@ struct X86Operand : public MCParsedAsmOperand {
   static X86Operand *CreateMem(unsigned SegReg, const MCExpr *Disp,
                                unsigned BaseReg, unsigned IndexReg,
                                unsigned Scale, SMLoc StartLoc, SMLoc EndLoc,
-                               unsigned Size = 0, bool NeedSizeDir = false) {
+                               unsigned Size = 0) {
     // We should never just have a displacement, that should be parsed as an
     // absolute memory operand.
     assert((SegReg || BaseReg || IndexReg) && "Invalid memory operand!");
@@ -519,7 +520,7 @@ struct X86Operand : public MCParsedAsmOperand {
     Res->Mem.IndexReg = IndexReg;
     Res->Mem.Scale    = Scale;
     Res->Mem.Size     = Size;
-    Res->Mem.NeedSizeDir = NeedSizeDir;
+    Res->AddressOf = false;
     return Res;
   }
 };
@@ -558,10 +559,12 @@ bool X86AsmParser::ParseRegister(unsigned &RegNo,
     Parser.Lex(); // Eat percent token.
 
   const AsmToken &Tok = Parser.getTok();
+  EndLoc = Tok.getEndLoc();
+
   if (Tok.isNot(AsmToken::Identifier)) {
     if (isParsingIntelSyntax()) return true;
     return Error(StartLoc, "invalid register name",
-                 SMRange(StartLoc, Tok.getEndLoc()));
+                 SMRange(StartLoc, EndLoc));
   }
 
   RegNo = MatchRegisterName(Tok.getString());
@@ -582,13 +585,12 @@ bool X86AsmParser::ParseRegister(unsigned &RegNo,
         X86II::isX86_64ExtendedReg(RegNo))
       return Error(StartLoc, "register %"
                    + Tok.getString() + " is only available in 64-bit mode",
-                   SMRange(StartLoc, Tok.getEndLoc()));
+                   SMRange(StartLoc, EndLoc));
   }
 
   // Parse "%st" as "%st(0)" and "%st(1)", which is multiple tokens.
   if (RegNo == 0 && (Tok.getString() == "st" || Tok.getString() == "ST")) {
     RegNo = X86::ST0;
-    EndLoc = Tok.getLoc();
     Parser.Lex(); // Eat 'st'
 
     // Check to see if we have '(4)' after %st.
@@ -615,11 +617,13 @@ bool X86AsmParser::ParseRegister(unsigned &RegNo,
     if (getParser().Lex().isNot(AsmToken::RParen))
       return Error(Parser.getTok().getLoc(), "expected ')'");
 
-    EndLoc = Tok.getLoc();
+    EndLoc = Parser.getTok().getEndLoc();
     Parser.Lex(); // Eat ')'
     return false;
   }
 
+  EndLoc = Parser.getTok().getEndLoc();
+
   // If this is "db[0-7]", match it as an alias
   // for dr[0-7].
   if (RegNo == 0 && Tok.getString().size() == 3 &&
@@ -636,7 +640,7 @@ bool X86AsmParser::ParseRegister(unsigned &RegNo,
     }
 
     if (RegNo != 0) {
-      EndLoc = Tok.getLoc();
+      EndLoc = Parser.getTok().getEndLoc();
       Parser.Lex(); // Eat it.
       return false;
     }
@@ -645,10 +649,9 @@ bool X86AsmParser::ParseRegister(unsigned &RegNo,
   if (RegNo == 0) {
     if (isParsingIntelSyntax()) return true;
     return Error(StartLoc, "invalid register name",
-                 SMRange(StartLoc, Tok.getEndLoc()));
+                 SMRange(StartLoc, EndLoc));
   }
 
-  EndLoc = Tok.getEndLoc();
   Parser.Lex(); // Eat identifier token.
   return false;
 }
@@ -673,115 +676,354 @@ static unsigned getIntelMemOperandSize(StringRef OpStr) {
   return Size;
 }
 
-X86Operand *X86AsmParser::ParseIntelBracExpression(unsigned SegReg, 
+enum IntelBracExprState {
+  IBES_START,
+  IBES_LBRAC,
+  IBES_RBRAC,
+  IBES_REGISTER,
+  IBES_REGISTER_STAR,
+  IBES_REGISTER_STAR_INTEGER,
+  IBES_INTEGER,
+  IBES_INTEGER_STAR,
+  IBES_INDEX_REGISTER,
+  IBES_IDENTIFIER,
+  IBES_DISP_EXPR,
+  IBES_MINUS,
+  IBES_ERROR
+};
+
+class IntelBracExprStateMachine {
+  IntelBracExprState State;
+  unsigned BaseReg, IndexReg, Scale;
+  int64_t Disp;
+
+  unsigned TmpReg;
+  int64_t TmpInteger;
+
+  bool isPlus;
+
+public:
+  IntelBracExprStateMachine(MCAsmParser &parser, int64_t disp) :
+    State(IBES_START), BaseReg(0), IndexReg(0), Scale(1), Disp(disp),
+    TmpReg(0), TmpInteger(0), isPlus(true) {}
+
+  unsigned getBaseReg() { return BaseReg; }
+  unsigned getIndexReg() { return IndexReg; }
+  unsigned getScale() { return Scale; }
+  int64_t getDisp() { return Disp; }
+  bool isValidEndState() { return State == IBES_RBRAC; }
+
+  void onPlus() {
+    switch (State) {
+    default:
+      State = IBES_ERROR;
+      break;
+    case IBES_INTEGER:
+      State = IBES_START;
+      if (isPlus)
+        Disp += TmpInteger;
+      else
+        Disp -= TmpInteger;
+      break;
+    case IBES_REGISTER:
+      State = IBES_START;
+      // If we already have a BaseReg, then assume this is the IndexReg with a
+      // scale of 1.
+      if (!BaseReg) {
+        BaseReg = TmpReg;
+      } else {
+        assert (!IndexReg && "BaseReg/IndexReg already set!");
+        IndexReg = TmpReg;
+        Scale = 1;
+      }
+      break;
+    case IBES_INDEX_REGISTER:
+      State = IBES_START;
+      break;
+    }
+    isPlus = true;
+  }
+  void onMinus() {
+    switch (State) {
+    default:
+      State = IBES_ERROR;
+      break;
+    case IBES_START:
+      State = IBES_MINUS;
+      break;
+    case IBES_INTEGER:
+      State = IBES_START;
+      if (isPlus)
+        Disp += TmpInteger;
+      else
+        Disp -= TmpInteger;
+      break;
+    case IBES_REGISTER:
+      State = IBES_START;
+      // If we already have a BaseReg, then assume this is the IndexReg with a
+      // scale of 1.
+      if (!BaseReg) {
+        BaseReg = TmpReg;
+      } else {
+        assert (!IndexReg && "BaseReg/IndexReg already set!");
+        IndexReg = TmpReg;
+        Scale = 1;
+      }
+      break;
+    case IBES_INDEX_REGISTER:
+      State = IBES_START;
+      break;
+    }
+    isPlus = false;
+  }
+  void onRegister(unsigned Reg) {
+    switch (State) {
+    default:
+      State = IBES_ERROR;
+      break;
+    case IBES_START:
+      State = IBES_REGISTER;
+      TmpReg = Reg;
+      break;
+    case IBES_INTEGER_STAR:
+      assert (!IndexReg && "IndexReg already set!");
+      State = IBES_INDEX_REGISTER;
+      IndexReg = Reg;
+      Scale = TmpInteger;
+      break;
+    }
+  }
+  void onDispExpr() {
+    switch (State) {
+    default:
+      State = IBES_ERROR;
+      break;
+    case IBES_START:
+      State = IBES_DISP_EXPR;
+      break;
+    }
+  }
+  void onInteger(int64_t TmpInt) {
+    switch (State) {
+    default:
+      State = IBES_ERROR;
+      break;
+    case IBES_START:
+      State = IBES_INTEGER;
+      TmpInteger = TmpInt;
+      break;
+    case IBES_MINUS:
+      State = IBES_INTEGER;
+      TmpInteger = TmpInt;
+      break;
+    case IBES_REGISTER_STAR:
+      assert (!IndexReg && "IndexReg already set!");
+      State = IBES_INDEX_REGISTER;
+      IndexReg = TmpReg;
+      Scale = TmpInt;
+      break;
+    }
+  }
+  void onStar() {
+    switch (State) {
+    default:
+      State = IBES_ERROR;
+      break;
+    case IBES_INTEGER:
+      State = IBES_INTEGER_STAR;
+      break;
+    case IBES_REGISTER:
+      State = IBES_REGISTER_STAR;
+      break;
+    }
+  }
+  void onLBrac() {
+    switch (State) {
+    default:
+      State = IBES_ERROR;
+      break;
+    case IBES_RBRAC:
+      State = IBES_START;
+      isPlus = true;
+      break;
+    }
+  }
+  void onRBrac() {
+    switch (State) {
+    default:
+      State = IBES_ERROR;
+      break;
+    case IBES_DISP_EXPR:
+      State = IBES_RBRAC;
+      break;
+    case IBES_INTEGER:
+      State = IBES_RBRAC;
+      if (isPlus)
+        Disp += TmpInteger;
+      else
+        Disp -= TmpInteger;
+      break;
+    case IBES_REGISTER:
+      State = IBES_RBRAC;
+      // If we already have a BaseReg, then assume this is the IndexReg with a
+      // scale of 1.
+      if (!BaseReg) {
+        BaseReg = TmpReg;
+      } else {
+        assert (!IndexReg && "BaseReg/IndexReg already set!");
+        IndexReg = TmpReg;
+        Scale = 1;
+      }
+      break;
+    case IBES_INDEX_REGISTER:
+      State = IBES_RBRAC;
+      break;
+    }
+  }
+};
+
+X86Operand *X86AsmParser::CreateMemForInlineAsm(const MCExpr *Disp, SMLoc Start,
+                                                SMLoc End, SMLoc SizeDirLoc,
+                                                unsigned Size) {
+  bool NeedSizeDir = false;
+  bool IsVarDecl = false;
+  if (const MCSymbolRefExpr *SymRef = dyn_cast<MCSymbolRefExpr>(Disp)) {
+    const MCSymbol &Sym = SymRef->getSymbol();
+    // FIXME: The SemaLookup will fail if the name is anything other then an
+    // identifier.
+    // FIXME: Pass a valid SMLoc.
+    unsigned tLength, tSize, tType;
+    SemaCallback->LookupInlineAsmIdentifier(Sym.getName(), NULL, tLength,
+                                            tSize, tType, IsVarDecl);
+    if (!Size) {
+      Size = tType * 8; // Size is in terms of bits in this context.
+      NeedSizeDir = Size > 0;
+    }
+  }
+
+  // If this is not a VarDecl then assume it is a FuncDecl or some other label
+  // reference.  We need an 'r' constraint here, so we need to create register
+  // operand to ensure proper matching.  Just pick a GPR based on the size of
+  // a pointer.
+  if (!IsVarDecl) {
+    unsigned RegNo = is64BitMode() ? X86::RBX : X86::EBX;
+    return X86Operand::CreateReg(RegNo, Start, End, /*AddressOf=*/true);
+  }
+
+  if (NeedSizeDir)
+    InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_SizeDirective, SizeDirLoc,
+                                                /*Len*/0, Size));  
+
+  // When parsing inline assembly we set the base register to a non-zero value
+  // as we don't know the actual value at this time.  This is necessary to
+  // get the matching correct in some cases.
+  return X86Operand::CreateMem(/*SegReg*/0, Disp, /*BaseReg*/1, /*IndexReg*/0,
+                               /*Scale*/1, Start, End, Size);
+}
+
+X86Operand *X86AsmParser::ParseIntelBracExpression(unsigned SegReg,
+                                                   uint64_t ImmDisp,
                                                    unsigned Size) {
-  unsigned BaseReg = 0, IndexReg = 0, Scale = 1;
   const AsmToken &Tok = Parser.getTok();
-  SMLoc Start = Tok.getLoc(), End;
-
-  const MCExpr *Disp = MCConstantExpr::Create(0, getContext());
-  // Parse [ BaseReg + Scale*IndexReg + Disp ] or [ symbol ]
+  SMLoc Start = Tok.getLoc(), End = Tok.getEndLoc();
 
   // Eat '['
   if (getLexer().isNot(AsmToken::LBrac))
     return ErrorOperand(Start, "Expected '[' token!");
   Parser.Lex();
 
+  unsigned TmpReg = 0;
+
+  // Try to handle '[' 'Symbol' ']'
   if (getLexer().is(AsmToken::Identifier)) {
-    // Parse BaseReg
-    if (ParseRegister(BaseReg, Start, End)) {
-      // Handle '[' 'symbol' ']'
-      if (getParser().ParseExpression(Disp, End)) return 0;
+    if (ParseRegister(TmpReg, Start, End)) {
+      const MCExpr *Disp;
+      SMLoc IdentStart = Tok.getLoc();
+      if (getParser().parseExpression(Disp, End))
+        return 0;
+
+      if (X86Operand *Err = ParseIntelVarWithQualifier(Disp, IdentStart))
+        return Err;
+
       if (getLexer().isNot(AsmToken::RBrac))
-        return ErrorOperand(Start, "Expected ']' token!");
-      Parser.Lex();
-      End = Tok.getLoc();
-      return X86Operand::CreateMem(Disp, Start, End, Size);
-    }
-  } else if (getLexer().is(AsmToken::Integer)) {
-      int64_t Val = Tok.getIntVal();
+        return ErrorOperand(Parser.getTok().getLoc(), "Expected ']' token!");
+
+      // FIXME: We don't handle 'ImmDisp' '[' 'Symbol' ']'.
+      if (ImmDisp)
+        return ErrorOperand(Start, "Unsupported immediate displacement!");
+
+      // Adjust the EndLoc due to the ']'.
+      End = SMLoc::getFromPointer(Parser.getTok().getEndLoc().getPointer()-1);
       Parser.Lex();
-      SMLoc Loc = Tok.getLoc();
-      if (getLexer().is(AsmToken::RBrac)) {
-        // Handle '[' number ']'
-        Parser.Lex();
-        End = Tok.getLoc();
-        const MCExpr *Disp = MCConstantExpr::Create(Val, getContext());
-        if (SegReg)
-          return X86Operand::CreateMem(SegReg, Disp, 0, 0, Scale,
-                                       Start, End, Size);
+      if (!isParsingInlineAsm())
         return X86Operand::CreateMem(Disp, Start, End, Size);
-      } else if (getLexer().is(AsmToken::Star)) {
-        // Handle '[' Scale*IndexReg ']'
-        Parser.Lex();
-        SMLoc IdxRegLoc = Tok.getLoc();
-        if (ParseRegister(IndexReg, IdxRegLoc, End))
-          return ErrorOperand(IdxRegLoc, "Expected register");
-        Scale = Val;
-      } else
-        return ErrorOperand(Loc, "Unexpected token");
-  }
 
-  // Parse ][ as a plus.
-  bool ExpectRBrac = true;
-  if (getLexer().is(AsmToken::RBrac)) {
-    ExpectRBrac = false;
-    Parser.Lex();
-    End = Tok.getLoc();
+      // We want the size directive before the '['.
+      SMLoc SizeDirLoc = SMLoc::getFromPointer(Start.getPointer()-1);
+      return CreateMemForInlineAsm(Disp, Start, End, SizeDirLoc, Size);
+    }
   }
 
-  if (getLexer().is(AsmToken::Plus) || getLexer().is(AsmToken::Minus) ||
-      getLexer().is(AsmToken::LBrac)) {
-    ExpectRBrac = true;
-    bool isPlus = getLexer().is(AsmToken::Plus) ||
-      getLexer().is(AsmToken::LBrac);
-    Parser.Lex(); 
-    SMLoc PlusLoc = Tok.getLoc();
-    if (getLexer().is(AsmToken::Integer)) {
+  // Parse [ BaseReg + Scale*IndexReg + Disp ].  We may have already parsed an
+  // immediate displacement before the bracketed expression.
+  bool Done = false;
+  IntelBracExprStateMachine SM(Parser, ImmDisp);
+
+  // If we parsed a register, then the end loc has already been set and
+  // the identifier has already been lexed.  We also need to update the
+  // state.
+  if (TmpReg)
+    SM.onRegister(TmpReg);
+
+  const MCExpr *Disp = 0;
+  while (!Done) {
+    bool UpdateLocLex = true;
+
+    // The period in the dot operator (e.g., [ebx].foo.bar) is parsed as an
+    // identifier.  Don't try an parse it as a register.
+    if (Tok.getString().startswith("."))
+      break;
+
+    switch (getLexer().getKind()) {
+    default: {
+      if (SM.isValidEndState()) {
+        Done = true;
+        break;
+      }
+      return ErrorOperand(Tok.getLoc(), "Unexpected token!");
+    }
+    case AsmToken::Identifier: {
+      // This could be a register or a displacement expression.
+      if(!ParseRegister(TmpReg, Start, End)) {
+        SM.onRegister(TmpReg);
+        UpdateLocLex = false;
+        break;
+      } else if (!getParser().parseExpression(Disp, End)) {
+        SM.onDispExpr();
+        UpdateLocLex = false;
+        break;
+      }
+      return ErrorOperand(Tok.getLoc(), "Unexpected identifier!");
+    }
+    case AsmToken::Integer: {
       int64_t Val = Tok.getIntVal();
-      Parser.Lex();
-      if (getLexer().is(AsmToken::Star)) {
-        Parser.Lex();
-        SMLoc IdxRegLoc = Tok.getLoc();
-        if (ParseRegister(IndexReg, IdxRegLoc, End))
-          return ErrorOperand(IdxRegLoc, "Expected register");
-        Scale = Val;
-      } else if (getLexer().is(AsmToken::RBrac)) {
-        const MCExpr *ValExpr = MCConstantExpr::Create(Val, getContext());
-        Disp = isPlus ? ValExpr : MCConstantExpr::Create(0-Val, getContext());
-      } else
-        return ErrorOperand(PlusLoc, "unexpected token after +");
-    } else if (getLexer().is(AsmToken::Identifier)) {
-      // This could be an index register or a displacement expression.
-      End = Tok.getLoc();
-      if (!IndexReg)
-        ParseRegister(IndexReg, Start, End);
-      else if (getParser().ParseExpression(Disp, End)) return 0;
+      SM.onInteger(Val);
+      break;
     }
-  }
-  
-  // Parse ][ as a plus.
-  if (getLexer().is(AsmToken::RBrac)) {
-    ExpectRBrac = false;
-    Parser.Lex();
-    End = Tok.getLoc();
-    if (getLexer().is(AsmToken::LBrac)) {
-      ExpectRBrac = true;
-      Parser.Lex();
-      if (getParser().ParseExpression(Disp, End))
-        return 0;
+    case AsmToken::Plus:    SM.onPlus(); break;
+    case AsmToken::Minus:   SM.onMinus(); break;
+    case AsmToken::Star:    SM.onStar(); break;
+    case AsmToken::LBrac:   SM.onLBrac(); break;
+    case AsmToken::RBrac:   SM.onRBrac(); break;
+    }
+    if (!Done && UpdateLocLex) {
+      End = Tok.getLoc();
+      Parser.Lex(); // Consume the token.
     }
-  } else if (ExpectRBrac) {
-      if (getParser().ParseExpression(Disp, End))
-        return 0;
   }
 
-  if (ExpectRBrac) {
-    if (getLexer().isNot(AsmToken::RBrac))
-      return ErrorOperand(End, "expected ']' token!");
-    Parser.Lex();
-    End = Tok.getLoc();
-  }
+  if (!Disp)
+    Disp = MCConstantExpr::Create(SM.getDisp(), getContext());
 
   // Parse the dot operator (e.g., [ebx].foo.bar).
   if (Tok.getString().startswith(".")) {
@@ -790,22 +1032,73 @@ X86Operand *X86AsmParser::ParseIntelBracExpression(unsigned SegReg,
     if (ParseIntelDotOperator(Disp, &NewDisp, Err))
       return ErrorOperand(Tok.getLoc(), Err);
     
+    End = Parser.getTok().getEndLoc();
     Parser.Lex();  // Eat the field.
     Disp = NewDisp;
   }
 
-  End = Tok.getLoc();
+  int BaseReg = SM.getBaseReg();
+  int IndexReg = SM.getIndexReg();
 
   // handle [-42]
-  if (!BaseReg && !IndexReg)
-    return X86Operand::CreateMem(Disp, Start, End, Size);
+  if (!BaseReg && !IndexReg) {
+    if (!SegReg)
+      return X86Operand::CreateMem(Disp, Start, End);
+    else
+      return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, Start, End, Size);
+  }
 
+  int Scale = SM.getScale();
   return X86Operand::CreateMem(SegReg, Disp, BaseReg, IndexReg, Scale,
                                Start, End, Size);
 }
 
+// Inline assembly may use variable names with namespace alias qualifiers.
+X86Operand *X86AsmParser::ParseIntelVarWithQualifier(const MCExpr *&Disp,
+                                                     SMLoc &IdentStart) {
+  // We should only see Foo::Bar if we're parsing inline assembly.
+  if (!isParsingInlineAsm())
+    return 0;
+
+  // If we don't see a ':' then there can't be a qualifier.
+  if (getLexer().isNot(AsmToken::Colon))
+    return 0;
+
+
+  bool Done = false;
+  const AsmToken &Tok = Parser.getTok();
+  SMLoc IdentEnd = Tok.getEndLoc();
+  while (!Done) {
+    switch (getLexer().getKind()) {
+    default:
+      Done = true; 
+      break;
+    case AsmToken::Colon:
+      getLexer().Lex(); // Consume ':'.
+      if (getLexer().isNot(AsmToken::Colon))
+        return ErrorOperand(Tok.getLoc(), "Expected ':' token!");
+      getLexer().Lex(); // Consume second ':'.
+      if (getLexer().isNot(AsmToken::Identifier))
+        return ErrorOperand(Tok.getLoc(), "Expected an identifier token!");
+      break;
+    case AsmToken::Identifier:
+      IdentEnd = Tok.getEndLoc();
+      getLexer().Lex(); // Consume the identifier.
+      break;
+    }
+  }
+  size_t Len = IdentEnd.getPointer() - IdentStart.getPointer();
+  StringRef Identifier(IdentStart.getPointer(), Len);
+  MCSymbol *Sym = getContext().GetOrCreateSymbol(Identifier);
+  MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None;
+  Disp = MCSymbolRefExpr::Create(Sym, Variant, getParser().getContext());
+  return 0;
+}
+
 /// ParseIntelMemOperand - Parse intel style memory operand.
-X86Operand *X86AsmParser::ParseIntelMemOperand(unsigned SegReg, SMLoc Start) {
+X86Operand *X86AsmParser::ParseIntelMemOperand(unsigned SegReg,
+                                               uint64_t ImmDisp,
+                                               SMLoc Start) {
   const AsmToken &Tok = Parser.getTok();
   SMLoc End;
 
@@ -817,8 +1110,21 @@ X86Operand *X86AsmParser::ParseIntelMemOperand(unsigned SegReg, SMLoc Start) {
     Parser.Lex();
   }
 
+  // Parse ImmDisp [ BaseReg + Scale*IndexReg + Disp ].
+  if (getLexer().is(AsmToken::Integer)) {
+    const AsmToken &IntTok = Parser.getTok();
+    if (isParsingInlineAsm())
+      InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_ImmPrefix,
+                                                  IntTok.getLoc()));
+    uint64_t ImmDisp = IntTok.getIntVal();
+    Parser.Lex(); // Eat the integer.
+    if (getLexer().isNot(AsmToken::LBrac))
+      return ErrorOperand(Start, "Expected '[' token!");
+    return ParseIntelBracExpression(SegReg, ImmDisp, Size);
+  }
+
   if (getLexer().is(AsmToken::LBrac))
-    return ParseIntelBracExpression(SegReg, Size);
+    return ParseIntelBracExpression(SegReg, ImmDisp, Size);
 
   if (!ParseRegister(SegReg, Start, End)) {
     // Handel SegReg : [ ... ]
@@ -827,32 +1133,21 @@ X86Operand *X86AsmParser::ParseIntelMemOperand(unsigned SegReg, SMLoc Start) {
     Parser.Lex(); // Eat :
     if (getLexer().isNot(AsmToken::LBrac))
       return ErrorOperand(Start, "Expected '[' token!");
-    return ParseIntelBracExpression(SegReg, Size);
+    return ParseIntelBracExpression(SegReg, ImmDisp, Size);
   }
 
   const MCExpr *Disp = MCConstantExpr::Create(0, getParser().getContext());
-  if (getParser().ParseExpression(Disp, End)) return 0;
-  End = Parser.getTok().getLoc();
+  SMLoc IdentStart = Tok.getLoc();
+  if (getParser().parseExpression(Disp, End))
+    return 0;
 
-  bool NeedSizeDir = false;
-  if (!Size && isParsingInlineAsm()) {
-    if (const MCSymbolRefExpr *SymRef = dyn_cast<MCSymbolRefExpr>(Disp)) {
-      const MCSymbol &Sym = SymRef->getSymbol();
-      // FIXME: The SemaLookup will fail if the name is anything other then an
-      // identifier.
-      // FIXME: Pass a valid SMLoc.
-      SemaCallback->LookupInlineAsmIdentifier(Sym.getName(), NULL, Size);
-      NeedSizeDir = Size > 0;
-    }
-  }
   if (!isParsingInlineAsm())
     return X86Operand::CreateMem(Disp, Start, End, Size);
-  else
-    // When parsing inline assembly we set the base register to a non-zero value
-    // as we don't know the actual value at this time.  This is necessary to
-    // get the matching correct in some cases.
-    return X86Operand::CreateMem(/*SegReg*/0, Disp, /*BaseReg*/1, /*IndexReg*/0,
-                                 /*Scale*/1, Start, End, Size, NeedSizeDir);
+
+  if (X86Operand *Err = ParseIntelVarWithQualifier(Disp, IdentStart))
+    return Err;
+
+  return CreateMemForInlineAsm(Disp, Start, End, Start, Size);
 }
 
 /// Parse the '.' operator.
@@ -918,11 +1213,9 @@ X86Operand *X86AsmParser::ParseIntelOffsetOfOperator(SMLoc Start) {
 
   SMLoc End;
   const MCExpr *Val;
-  if (getParser().ParseExpression(Val, End))
+  if (getParser().parseExpression(Val, End))
     return ErrorOperand(Start, "Unable to parse expression!");
 
-  End = Parser.getTok().getLoc();
-
   // Don't emit the offset operator.
   InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Skip, OffsetOfLoc, 7));
 
@@ -930,13 +1223,23 @@ X86Operand *X86AsmParser::ParseIntelOffsetOfOperator(SMLoc Start) {
   // register operand to ensure proper matching.  Just pick a GPR based on
   // the size of a pointer.
   unsigned RegNo = is64BitMode() ? X86::RBX : X86::EBX;
-  return X86Operand::CreateReg(RegNo, Start, End, OffsetOfLoc);
+  return X86Operand::CreateReg(RegNo, Start, End, /*GetAddress=*/true,
+                               OffsetOfLoc);
 }
 
-/// Parse the 'TYPE' operator.  The TYPE operator returns the size of a C or
-/// C++ type or variable. If the variable is an array, TYPE returns the size of
-/// a single element of the array.
-X86Operand *X86AsmParser::ParseIntelTypeOperator(SMLoc Start) {
+enum IntelOperatorKind {
+  IOK_LENGTH,
+  IOK_SIZE,
+  IOK_TYPE
+};
+
+/// Parse the 'LENGTH', 'TYPE' and 'SIZE' operators.  The LENGTH operator
+/// returns the number of elements in an array.  It returns the value 1 for
+/// non-array variables.  The SIZE operator returns the size of a C or C++
+/// variable.  A variable's size is the product of its LENGTH and TYPE.  The
+/// TYPE operator returns the size of a C or C++ type or variable. If the
+/// variable is an array, TYPE returns the size of a single element.
+X86Operand *X86AsmParser::ParseIntelOperator(SMLoc Start, unsigned OpKind) {
   SMLoc TypeLoc = Start;
   Parser.Lex(); // Eat offset.
   Start = Parser.getTok().getLoc();
@@ -944,76 +1247,92 @@ X86Operand *X86AsmParser::ParseIntelTypeOperator(SMLoc Start) {
 
   SMLoc End;
   const MCExpr *Val;
-  if (getParser().ParseExpression(Val, End))
+  if (getParser().parseExpression(Val, End))
     return 0;
 
-  End = Parser.getTok().getLoc();
-
-  unsigned Size = 0;
+  unsigned Length = 0, Size = 0, Type = 0;
   if (const MCSymbolRefExpr *SymRef = dyn_cast<MCSymbolRefExpr>(Val)) {
     const MCSymbol &Sym = SymRef->getSymbol();
     // FIXME: The SemaLookup will fail if the name is anything other then an
     // identifier.
     // FIXME: Pass a valid SMLoc.
-    if (!SemaCallback->LookupInlineAsmIdentifier(Sym.getName(), NULL, Size))
-      return ErrorOperand(Start, "Unable to lookup TYPE of expr!");
-
-    Size /= 8; // Size is in terms of bits, but we want bytes in the context.
+    bool IsVarDecl;
+    if (!SemaCallback->LookupInlineAsmIdentifier(Sym.getName(), NULL, Length,
+                                                 Size, Type, IsVarDecl))
+      return ErrorOperand(Start, "Unable to lookup expr!");
+  }
+  unsigned CVal;
+  switch(OpKind) {
+  default: llvm_unreachable("Unexpected operand kind!");
+  case IOK_LENGTH: CVal = Length; break;
+  case IOK_SIZE: CVal = Size; break;
+  case IOK_TYPE: CVal = Type; break;
   }
 
   // Rewrite the type operator and the C or C++ type or variable in terms of an
   // immediate.  E.g. TYPE foo -> $$4
   unsigned Len = End.getPointer() - TypeLoc.getPointer();
-  InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Imm, TypeLoc, Len, Size));
+  InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Imm, TypeLoc, Len, CVal));
 
-  const MCExpr *Imm = MCConstantExpr::Create(Size, getContext());
-  return X86Operand::CreateImm(Imm, Start, End, /*NeedAsmRewrite*/false);
+  const MCExpr *Imm = MCConstantExpr::Create(CVal, getContext());
+  return X86Operand::CreateImm(Imm, Start, End);
 }
 
 X86Operand *X86AsmParser::ParseIntelOperand() {
   SMLoc Start = Parser.getTok().getLoc(), End;
-
-  // offset operator.
   StringRef AsmTokStr = Parser.getTok().getString();
-  if ((AsmTokStr == "offset" || AsmTokStr == "OFFSET") &&
-      isParsingInlineAsm())
-    return ParseIntelOffsetOfOperator(Start);
-
-  // Type directive.
-  if ((AsmTokStr == "type" || AsmTokStr == "TYPE") &&
-      isParsingInlineAsm())
-    return ParseIntelTypeOperator(Start);
-
-  // Unsupported directives.
-  if (isParsingIntelSyntax() &&
-      (AsmTokStr == "size" || AsmTokStr == "SIZE" ||
-       AsmTokStr == "length" || AsmTokStr == "LENGTH"))
-      return ErrorOperand(Start, "Unsupported directive!");
-
-  // immediate.
+
+  // Offset, length, type and size operators.
+  if (isParsingInlineAsm()) {
+    if (AsmTokStr == "offset" || AsmTokStr == "OFFSET")
+      return ParseIntelOffsetOfOperator(Start);
+    if (AsmTokStr == "length" || AsmTokStr == "LENGTH")
+      return ParseIntelOperator(Start, IOK_LENGTH);
+    if (AsmTokStr == "size" || AsmTokStr == "SIZE")
+      return ParseIntelOperator(Start, IOK_SIZE);
+    if (AsmTokStr == "type" || AsmTokStr == "TYPE")
+      return ParseIntelOperator(Start, IOK_TYPE);
+  }
+
+  // Immediate.
   if (getLexer().is(AsmToken::Integer) || getLexer().is(AsmToken::Real) ||
       getLexer().is(AsmToken::Minus)) {
     const MCExpr *Val;
-    if (!getParser().ParseExpression(Val, End)) {
-      End = Parser.getTok().getLoc();
-      return X86Operand::CreateImm(Val, Start, End);
+    bool isInteger = getLexer().is(AsmToken::Integer);
+    if (!getParser().parseExpression(Val, End)) {
+      if (isParsingInlineAsm())
+        InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_ImmPrefix, Start));
+      // Immediate.
+      if (getLexer().isNot(AsmToken::LBrac))
+        return X86Operand::CreateImm(Val, Start, End);
+
+      // Only positive immediates are valid.
+      if (!isInteger) {
+        Error(Parser.getTok().getLoc(), "expected a positive immediate "
+              "displacement before bracketed expr.");
+        return 0;
+      }
+
+      // Parse ImmDisp [ BaseReg + Scale*IndexReg + Disp ].
+      if (uint64_t ImmDisp = dyn_cast<MCConstantExpr>(Val)->getValue())
+        return ParseIntelMemOperand(/*SegReg=*/0, ImmDisp, Start);
     }
   }
 
-  // register
+  // Register.
   unsigned RegNo = 0;
   if (!ParseRegister(RegNo, Start, End)) {
     // If this is a segment register followed by a ':', then this is the start
     // of a memory reference, otherwise this is a normal register reference.
     if (getLexer().isNot(AsmToken::Colon))
-      return X86Operand::CreateReg(RegNo, Start, Parser.getTok().getLoc());
+      return X86Operand::CreateReg(RegNo, Start, End);
 
     getParser().Lex(); // Eat the colon.
-    return ParseIntelMemOperand(RegNo, Start);
+    return ParseIntelMemOperand(/*SegReg=*/RegNo, /*Disp=*/0, Start);
   }
 
-  // mem operand
-  return ParseIntelMemOperand(0, Start);
+  // Memory operand.
+  return ParseIntelMemOperand(/*SegReg=*/0, /*Disp=*/0, Start);
 }
 
 X86Operand *X86AsmParser::ParseATTOperand() {
@@ -1037,7 +1356,6 @@ X86Operand *X86AsmParser::ParseATTOperand() {
     if (getLexer().isNot(AsmToken::Colon))
       return X86Operand::CreateReg(RegNo, Start, End);
 
-
     getParser().Lex(); // Eat the colon.
     return ParseMemOperand(RegNo, Start);
   }
@@ -1046,7 +1364,7 @@ X86Operand *X86AsmParser::ParseATTOperand() {
     SMLoc Start = Parser.getTok().getLoc(), End;
     Parser.Lex();
     const MCExpr *Val;
-    if (getParser().ParseExpression(Val, End))
+    if (getParser().parseExpression(Val, End))
       return 0;
     return X86Operand::CreateImm(Val, Start, End);
   }
@@ -1064,7 +1382,7 @@ X86Operand *X86AsmParser::ParseMemOperand(unsigned SegReg, SMLoc MemStart) {
   const MCExpr *Disp = MCConstantExpr::Create(0, getParser().getContext());
   if (getLexer().isNot(AsmToken::LParen)) {
     SMLoc ExprEnd;
-    if (getParser().ParseExpression(Disp, ExprEnd)) return 0;
+    if (getParser().parseExpression(Disp, ExprEnd)) return 0;
 
     // After parsing the base expression we could either have a parenthesized
     // memory address or not.  If not, return now.  If so, eat the (.
@@ -1090,7 +1408,7 @@ X86Operand *X86AsmParser::ParseMemOperand(unsigned SegReg, SMLoc MemStart) {
       SMLoc ExprEnd;
 
       // It must be an parenthesized expression, parse it now.
-      if (getParser().ParseParenExpression(Disp, ExprEnd))
+      if (getParser().parseParenExpression(Disp, ExprEnd))
         return 0;
 
       // After parsing the base expression we could either have a parenthesized
@@ -1150,7 +1468,7 @@ X86Operand *X86AsmParser::ParseMemOperand(unsigned SegReg, SMLoc MemStart) {
           SMLoc Loc = Parser.getTok().getLoc();
 
           int64_t ScaleVal;
-          if (getParser().ParseAbsoluteExpression(ScaleVal)){
+          if (getParser().parseAbsoluteExpression(ScaleVal)){
             Error(Loc, "expected scale expression");
             return 0;
           }
@@ -1169,7 +1487,7 @@ X86Operand *X86AsmParser::ParseMemOperand(unsigned SegReg, SMLoc MemStart) {
       SMLoc Loc = Parser.getTok().getLoc();
 
       int64_t Value;
-      if (getParser().ParseAbsoluteExpression(Value))
+      if (getParser().parseAbsoluteExpression(Value))
         return 0;
 
       if (Value != 1)
@@ -1183,7 +1501,7 @@ X86Operand *X86AsmParser::ParseMemOperand(unsigned SegReg, SMLoc MemStart) {
     Error(Parser.getTok().getLoc(), "unexpected token in memory operand");
     return 0;
   }
-  SMLoc MemEnd = Parser.getTok().getLoc();
+  SMLoc MemEnd = Parser.getTok().getEndLoc();
   Parser.Lex(); // Eat the ')'.
 
   // If we have both a base register and an index register make sure they are
@@ -1310,7 +1628,7 @@ ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc,
     if (X86Operand *Op = ParseOperand())
       Operands.push_back(Op);
     else {
-      Parser.EatToEndOfStatement();
+      Parser.eatToEndOfStatement();
       return true;
     }
 
@@ -1321,14 +1639,14 @@ ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc,
       if (X86Operand *Op = ParseOperand())
         Operands.push_back(Op);
       else {
-        Parser.EatToEndOfStatement();
+        Parser.eatToEndOfStatement();
         return true;
       }
     }
 
     if (getLexer().isNot(AsmToken::EndOfStatement)) {
       SMLoc Loc = getLexer().getLoc();
-      Parser.EatToEndOfStatement();
+      Parser.eatToEndOfStatement();
       return Error(Loc, "unexpected token in argument list");
     }
   }
@@ -1509,245 +1827,78 @@ ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc,
   return false;
 }
 
-bool X86AsmParser::
-processInstruction(MCInst &Inst,
-                   const SmallVectorImpl<MCParsedAsmOperand*> &Ops) {
-  switch (Inst.getOpcode()) {
-  default: return false;
-  case X86::AND16i16: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti16i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::AND16ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::AND32i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti32i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::AND32ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::AND64i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti64i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::AND64ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::XOR16i16: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti16i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::XOR16ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::XOR32i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti32i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::XOR32ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::XOR64i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti64i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::XOR64ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::OR16i16: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti16i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::OR16ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::OR32i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti32i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::OR32ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::OR64i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti64i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::OR64ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::CMP16i16: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti16i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::CMP16ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::CMP32i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti32i8Value(Inst.getOperand(0).getImm()))
-      return false;
-
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::CMP32ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::CMP64i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti64i8Value(Inst.getOperand(0).getImm()))
-      return false;
+static bool convertToSExti8(MCInst &Inst, unsigned Opcode, unsigned Reg,
+                            bool isCmp) {
+  MCInst TmpInst;
+  TmpInst.setOpcode(Opcode);
+  if (!isCmp)
+    TmpInst.addOperand(MCOperand::CreateReg(Reg));
+  TmpInst.addOperand(MCOperand::CreateReg(Reg));
+  TmpInst.addOperand(Inst.getOperand(0));
+  Inst = TmpInst;
+  return true;
+}
 
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::CMP64ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::ADD16i16: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti16i8Value(Inst.getOperand(0).getImm()))
-      return false;
+static bool convert16i16to16ri8(MCInst &Inst, unsigned Opcode,
+                                bool isCmp = false) {
+  if (!Inst.getOperand(0).isImm() ||
+      !isImmSExti16i8Value(Inst.getOperand(0).getImm()))
+    return false;
 
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::ADD16ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::ADD32i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti32i8Value(Inst.getOperand(0).getImm()))
-      return false;
+  return convertToSExti8(Inst, Opcode, X86::AX, isCmp);
+}
 
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::ADD32ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::ADD64i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti64i8Value(Inst.getOperand(0).getImm()))
-      return false;
+static bool convert32i32to32ri8(MCInst &Inst, unsigned Opcode,
+                                bool isCmp = false) {
+  if (!Inst.getOperand(0).isImm() ||
+      !isImmSExti32i8Value(Inst.getOperand(0).getImm()))
+    return false;
 
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::ADD64ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::SUB16i16: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti16i8Value(Inst.getOperand(0).getImm()))
-      return false;
+  return convertToSExti8(Inst, Opcode, X86::EAX, isCmp);
+}
 
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::SUB16ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::AX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::SUB32i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti32i8Value(Inst.getOperand(0).getImm()))
-      return false;
+static bool convert64i32to64ri8(MCInst &Inst, unsigned Opcode,
+                                bool isCmp = false) {
+  if (!Inst.getOperand(0).isImm() ||
+      !isImmSExti64i8Value(Inst.getOperand(0).getImm()))
+    return false;
 
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::SUB32ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::EAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
-  case X86::SUB64i32: {
-    if (!Inst.getOperand(0).isImm() ||
-        !isImmSExti64i8Value(Inst.getOperand(0).getImm()))
-      return false;
+  return convertToSExti8(Inst, Opcode, X86::RAX, isCmp);
+}
 
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::SUB64ri8);
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(MCOperand::CreateReg(X86::RAX));
-    TmpInst.addOperand(Inst.getOperand(0));
-    Inst = TmpInst;
-    return true;
-  }
+bool X86AsmParser::
+processInstruction(MCInst &Inst,
+                   const SmallVectorImpl<MCParsedAsmOperand*> &Ops) {
+  switch (Inst.getOpcode()) {
+  default: return false;
+  case X86::AND16i16: return convert16i16to16ri8(Inst, X86::AND16ri8);
+  case X86::AND32i32: return convert32i32to32ri8(Inst, X86::AND32ri8);
+  case X86::AND64i32: return convert64i32to64ri8(Inst, X86::AND64ri8);
+  case X86::XOR16i16: return convert16i16to16ri8(Inst, X86::XOR16ri8);
+  case X86::XOR32i32: return convert32i32to32ri8(Inst, X86::XOR32ri8);
+  case X86::XOR64i32: return convert64i32to64ri8(Inst, X86::XOR64ri8);
+  case X86::OR16i16:  return convert16i16to16ri8(Inst, X86::OR16ri8);
+  case X86::OR32i32:  return convert32i32to32ri8(Inst, X86::OR32ri8);
+  case X86::OR64i32:  return convert64i32to64ri8(Inst, X86::OR64ri8);
+  case X86::CMP16i16: return convert16i16to16ri8(Inst, X86::CMP16ri8, true);
+  case X86::CMP32i32: return convert32i32to32ri8(Inst, X86::CMP32ri8, true);
+  case X86::CMP64i32: return convert64i32to64ri8(Inst, X86::CMP64ri8, true);
+  case X86::ADD16i16: return convert16i16to16ri8(Inst, X86::ADD16ri8);
+  case X86::ADD32i32: return convert32i32to32ri8(Inst, X86::ADD32ri8);
+  case X86::ADD64i32: return convert64i32to64ri8(Inst, X86::ADD64ri8);
+  case X86::SUB16i16: return convert16i16to16ri8(Inst, X86::SUB16ri8);
+  case X86::SUB32i32: return convert32i32to32ri8(Inst, X86::SUB32ri8);
+  case X86::SUB64i32: return convert64i32to64ri8(Inst, X86::SUB64ri8);
+  case X86::ADC16i16: return convert16i16to16ri8(Inst, X86::ADC16ri8);
+  case X86::ADC32i32: return convert32i32to32ri8(Inst, X86::ADC32ri8);
+  case X86::ADC64i32: return convert64i32to64ri8(Inst, X86::ADC64ri8);
+  case X86::SBB16i16: return convert16i16to16ri8(Inst, X86::SBB16ri8);
+  case X86::SBB32i32: return convert32i32to32ri8(Inst, X86::SBB32ri8);
+  case X86::SBB64i32: return convert64i32to64ri8(Inst, X86::SBB64ri8);
   }
 }
 
+static const char *getSubtargetFeatureName(unsigned Val);
 bool X86AsmParser::
 MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
                         SmallVectorImpl<MCParsedAsmOperand*> &Operands,
@@ -1809,10 +1960,21 @@ MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
       Out.EmitInstruction(Inst);
     Opcode = Inst.getOpcode();
     return false;
-  case Match_MissingFeature:
-    Error(IDLoc, "instruction requires a CPU feature not currently enabled",
-          EmptyRanges, MatchingInlineAsm);
-    return true;
+  case Match_MissingFeature: {
+    assert(ErrorInfo && "Unknown missing feature!");
+    // Special case the error message for the very common case where only
+    // a single subtarget feature is missing.
+    std::string Msg = "instruction requires:";
+    unsigned Mask = 1;
+    for (unsigned i = 0; i < (sizeof(ErrorInfo)*8-1); ++i) {
+      if (ErrorInfo & Mask) {
+        Msg += " ";
+        Msg += getSubtargetFeatureName(ErrorInfo & Mask);
+      }
+      Mask <<= 1;
+    }
+    return Error(IDLoc, Msg, EmptyRanges, MatchingInlineAsm);
+  }
   case Match_InvalidOperand:
     WasOriginallyInvalidOperand = true;
     break;
@@ -1843,19 +2005,32 @@ MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
   // Check for the various suffix matches.
   Tmp[Base.size()] = Suffixes[0];
   unsigned ErrorInfoIgnore;
+  unsigned ErrorInfoMissingFeature = 0; // Init suppresses compiler warnings.
   unsigned Match1, Match2, Match3, Match4;
 
   Match1 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore,
                                 isParsingIntelSyntax());
+  // If this returned as a missing feature failure, remember that.
+  if (Match1 == Match_MissingFeature)
+    ErrorInfoMissingFeature = ErrorInfoIgnore;
   Tmp[Base.size()] = Suffixes[1];
   Match2 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore,
                                 isParsingIntelSyntax());
+  // If this returned as a missing feature failure, remember that.
+  if (Match2 == Match_MissingFeature)
+    ErrorInfoMissingFeature = ErrorInfoIgnore;
   Tmp[Base.size()] = Suffixes[2];
   Match3 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore,
                                 isParsingIntelSyntax());
+  // If this returned as a missing feature failure, remember that.
+  if (Match3 == Match_MissingFeature)
+    ErrorInfoMissingFeature = ErrorInfoIgnore;
   Tmp[Base.size()] = Suffixes[3];
   Match4 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore,
                                 isParsingIntelSyntax());
+  // If this returned as a missing feature failure, remember that.
+  if (Match4 == Match_MissingFeature)
+    ErrorInfoMissingFeature = ErrorInfoIgnore;
 
   // Restore the old token.
   Op->setTokenValue(Base);
@@ -1936,9 +2111,16 @@ MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
   // missing feature.
   if ((Match1 == Match_MissingFeature) + (Match2 == Match_MissingFeature) +
       (Match3 == Match_MissingFeature) + (Match4 == Match_MissingFeature) == 1){
-    Error(IDLoc, "instruction requires a CPU feature not currently enabled",
-          EmptyRanges, MatchingInlineAsm);
-    return true;
+    std::string Msg = "instruction requires:";
+    unsigned Mask = 1;
+    for (unsigned i = 0; i < (sizeof(ErrorInfoMissingFeature)*8-1); ++i) {
+      if (ErrorInfoMissingFeature & Mask) {
+        Msg += " ";
+        Msg += getSubtargetFeatureName(ErrorInfoMissingFeature & Mask);
+      }
+      Mask <<= 1;
+    }
+    return Error(IDLoc, Msg, EmptyRanges, MatchingInlineAsm);
   }
 
   // If one instruction matched with an invalid operand, report this as an
@@ -1986,10 +2168,10 @@ bool X86AsmParser::ParseDirectiveWord(unsigned Size, SMLoc L) {
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     for (;;) {
       const MCExpr *Value;
-      if (getParser().ParseExpression(Value))
+      if (getParser().parseExpression(Value))
         return true;
 
-      getParser().getStreamer().EmitValue(Value, Size, 0 /*addrspace*/);
+      getParser().getStreamer().EmitValue(Value, Size);
 
       if (getLexer().is(AsmToken::EndOfStatement))
         break;
@@ -2027,16 +2209,13 @@ bool X86AsmParser::ParseDirectiveCode(StringRef IDVal, SMLoc L) {
   return false;
 }
 
-
-extern "C" void LLVMInitializeX86AsmLexer();
-
 // Force static initialization.
 extern "C" void LLVMInitializeX86AsmParser() {
   RegisterMCAsmParser<X86AsmParser> X(TheX86_32Target);
   RegisterMCAsmParser<X86AsmParser> Y(TheX86_64Target);
-  LLVMInitializeX86AsmLexer();
 }
 
 #define GET_REGISTER_MATCHER
 #define GET_MATCHER_IMPLEMENTATION
+#define GET_SUBTARGET_FEATURE_NAME
 #include "X86GenAsmMatcher.inc"
diff --git a/lib/Target/X86/CMakeLists.txt b/lib/Target/X86/CMakeLists.txt
index f4d03a602cf5..d14899d28a23 100644
--- a/lib/Target/X86/CMakeLists.txt
+++ b/lib/Target/X86/CMakeLists.txt
@@ -10,7 +10,6 @@ tablegen(LLVM X86GenDAGISel.inc -gen-dag-isel)
 tablegen(LLVM X86GenFastISel.inc -gen-fast-isel)
 tablegen(LLVM X86GenCallingConv.inc -gen-callingconv)
 tablegen(LLVM X86GenSubtargetInfo.inc -gen-subtarget)
-tablegen(LLVM X86GenEDInfo.inc -gen-enhanced-disassembly-info)
 add_public_tablegen_target(X86CommonTableGen)
 
 set(sources
@@ -26,11 +25,13 @@ set(sources
   X86JITInfo.cpp
   X86MCInstLower.cpp
   X86MachineFunctionInfo.cpp
+  X86PadShortFunction.cpp
   X86RegisterInfo.cpp
   X86SelectionDAGInfo.cpp
   X86Subtarget.cpp
   X86TargetMachine.cpp
   X86TargetObjectFile.cpp
+  X86TargetTransformInfo.cpp
   X86VZeroUpper.cpp
   )
 
diff --git a/lib/Target/X86/Disassembler/X86Disassembler.cpp b/lib/Target/X86/Disassembler/X86Disassembler.cpp
index f13692739a17..ca6f80ce3e58 100644
--- a/lib/Target/X86/Disassembler/X86Disassembler.cpp
+++ b/lib/Target/X86/Disassembler/X86Disassembler.cpp
@@ -16,11 +16,9 @@
 
 #include "X86Disassembler.h"
 #include "X86DisassemblerDecoder.h"
-
-#include "llvm/MC/EDInstInfo.h"
-#include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCDisassembler.h"
+#include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCSubtargetInfo.h"
@@ -33,7 +31,6 @@
 #include "X86GenRegisterInfo.inc"
 #define GET_INSTRINFO_ENUM
 #include "X86GenInstrInfo.inc"
-#include "X86GenEDInfo.inc"
 
 using namespace llvm;
 using namespace llvm::X86Disassembler;
@@ -84,10 +81,6 @@ X86GenericDisassembler::~X86GenericDisassembler() {
   delete MII;
 }
 
-const EDInstInfo *X86GenericDisassembler::getEDInfo() const {
-  return instInfoX86;
-}
-
 /// regionReader - a callback function that wraps the readByte method from
 ///   MemoryObject.
 ///
diff --git a/lib/Target/X86/Disassembler/X86Disassembler.h b/lib/Target/X86/Disassembler/X86Disassembler.h
index 981701f52764..b92427a7e91a 100644
--- a/lib/Target/X86/Disassembler/X86Disassembler.h
+++ b/lib/Target/X86/Disassembler/X86Disassembler.h
@@ -95,8 +95,6 @@ class MCSubtargetInfo;
 class MemoryObject;
 class raw_ostream;
 
-struct EDInstInfo;
-
 namespace X86Disassembler {
 
 /// X86GenericDisassembler - Generic disassembler for all X86 platforms.
@@ -122,8 +120,6 @@ public:
                               raw_ostream &vStream,
                               raw_ostream &cStream) const;
 
-  /// getEDInfo - See MCDisassembler.
-  const EDInstInfo *getEDInfo() const;
 private:
   DisassemblerMode              fMode;
 };
diff --git a/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c b/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c
index 85d8a991dd6e..e40edba6d689 100644
--- a/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c
+++ b/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c
@@ -61,7 +61,7 @@ static int modRMRequired(OpcodeType type,
                          InstructionContext insnContext,
                          uint8_t opcode) {
   const struct ContextDecision* decision = 0;
-  
+
   switch (type) {
   case ONEBYTE:
     decision = &ONEBYTE_SYM;
@@ -102,7 +102,7 @@ static InstrUID decode(OpcodeType type,
                        uint8_t opcode,
                        uint8_t modRM) {
   const struct ModRMDecision* dec = 0;
-  
+
   switch (type) {
   case ONEBYTE:
     dec = &ONEBYTE_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
@@ -123,7 +123,7 @@ static InstrUID decode(OpcodeType type,
     dec = &THREEBYTEA7_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
     break;
   }
-  
+
   switch (dec->modrm_type) {
   default:
     debug("Corrupt table!  Unknown modrm_type");
@@ -171,10 +171,10 @@ static const struct InstructionSpecifier *specifierForUID(InstrUID uid) {
  */
 static int consumeByte(struct InternalInstruction* insn, uint8_t* byte) {
   int ret = insn->reader(insn->readerArg, byte, insn->readerCursor);
-  
+
   if (!ret)
     ++(insn->readerCursor);
-  
+
   return ret;
 }
 
@@ -238,19 +238,19 @@ CONSUME_FUNC(consumeUInt64, uint64_t)
  */
 static void dbgprintf(struct InternalInstruction* insn,
                       const char* format,
-                      ...) {  
+                      ...) {
   char buffer[256];
   va_list ap;
-  
+
   if (!insn->dlog)
     return;
-    
+
   va_start(ap, format);
   (void)vsnprintf(buffer, sizeof(buffer), format, ap);
   va_end(ap);
-  
+
   insn->dlog(insn->dlogArg, buffer);
-  
+
   return;
 }
 
@@ -305,27 +305,40 @@ static int readPrefixes(struct InternalInstruction* insn) {
   BOOL prefixGroups[4] = { FALSE };
   uint64_t prefixLocation;
   uint8_t byte = 0;
-  
+
   BOOL hasAdSize = FALSE;
   BOOL hasOpSize = FALSE;
-  
+
   dbgprintf(insn, "readPrefixes()");
-    
+
   while (isPrefix) {
     prefixLocation = insn->readerCursor;
-    
+
     if (consumeByte(insn, &byte))
       return -1;
 
     /*
-     * If the first byte is a LOCK prefix break and let it be disassembled
-     * as a lock "instruction", by creating an <MCInst #xxxx LOCK_PREFIX>.
-     * FIXME there is currently no way to get the disassembler to print the
-     * lock prefix if it is not the first byte.
+     * If the byte is a LOCK/REP/REPNE prefix and not a part of the opcode, then
+     * break and let it be disassembled as a normal "instruction".
      */
-    if (insn->readerCursor - 1 == insn->startLocation && byte == 0xf0)
-      break;
-    
+    if (insn->readerCursor - 1 == insn->startLocation
+        && (byte == 0xf0 || byte == 0xf2 || byte == 0xf3)) {
+      uint8_t nextByte;
+      if (byte == 0xf0)
+        break;
+      if (lookAtByte(insn, &nextByte))
+        return -1;
+      if (insn->mode == MODE_64BIT && (nextByte & 0xf0) == 0x40) {
+        if (consumeByte(insn, &nextByte))
+          return -1;
+        if (lookAtByte(insn, &nextByte))
+          return -1;
+        unconsumeByte(insn);
+      }
+      if (nextByte != 0x0f && nextByte != 0x90)
+        break;
+    }
+
     switch (byte) {
     case 0xf0:  /* LOCK */
     case 0xf2:  /* REPNE/REPNZ */
@@ -387,21 +400,21 @@ static int readPrefixes(struct InternalInstruction* insn) {
       isPrefix = FALSE;
       break;
     }
-    
+
     if (isPrefix)
       dbgprintf(insn, "Found prefix 0x%hhx", byte);
   }
-    
+
   insn->vexSize = 0;
-  
+
   if (byte == 0xc4) {
     uint8_t byte1;
-      
+
     if (lookAtByte(insn, &byte1)) {
       dbgprintf(insn, "Couldn't read second byte of VEX");
       return -1;
     }
-    
+
     if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) {
       insn->vexSize = 3;
       insn->necessaryPrefixLocation = insn->readerCursor - 1;
@@ -410,67 +423,67 @@ static int readPrefixes(struct InternalInstruction* insn) {
       unconsumeByte(insn);
       insn->necessaryPrefixLocation = insn->readerCursor - 1;
     }
-    
+
     if (insn->vexSize == 3) {
       insn->vexPrefix[0] = byte;
       consumeByte(insn, &insn->vexPrefix[1]);
       consumeByte(insn, &insn->vexPrefix[2]);
 
       /* We simulate the REX prefix for simplicity's sake */
-   
+
       if (insn->mode == MODE_64BIT) {
-        insn->rexPrefix = 0x40 
+        insn->rexPrefix = 0x40
                         | (wFromVEX3of3(insn->vexPrefix[2]) << 3)
                         | (rFromVEX2of3(insn->vexPrefix[1]) << 2)
                         | (xFromVEX2of3(insn->vexPrefix[1]) << 1)
                         | (bFromVEX2of3(insn->vexPrefix[1]) << 0);
       }
-    
+
       switch (ppFromVEX3of3(insn->vexPrefix[2]))
       {
       default:
         break;
       case VEX_PREFIX_66:
-        hasOpSize = TRUE;      
+        hasOpSize = TRUE;
         break;
       }
-    
+
       dbgprintf(insn, "Found VEX prefix 0x%hhx 0x%hhx 0x%hhx", insn->vexPrefix[0], insn->vexPrefix[1], insn->vexPrefix[2]);
     }
   }
   else if (byte == 0xc5) {
     uint8_t byte1;
-    
+
     if (lookAtByte(insn, &byte1)) {
       dbgprintf(insn, "Couldn't read second byte of VEX");
       return -1;
     }
-      
+
     if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) {
       insn->vexSize = 2;
     }
     else {
       unconsumeByte(insn);
     }
-    
+
     if (insn->vexSize == 2) {
       insn->vexPrefix[0] = byte;
       consumeByte(insn, &insn->vexPrefix[1]);
-        
+
       if (insn->mode == MODE_64BIT) {
-        insn->rexPrefix = 0x40 
+        insn->rexPrefix = 0x40
                         | (rFromVEX2of2(insn->vexPrefix[1]) << 2);
       }
-        
+
       switch (ppFromVEX2of2(insn->vexPrefix[1]))
       {
       default:
         break;
       case VEX_PREFIX_66:
-        hasOpSize = TRUE;      
+        hasOpSize = TRUE;
         break;
       }
-         
+
       dbgprintf(insn, "Found VEX prefix 0x%hhx 0x%hhx", insn->vexPrefix[0], insn->vexPrefix[1]);
     }
   }
@@ -478,17 +491,17 @@ static int readPrefixes(struct InternalInstruction* insn) {
     if (insn->mode == MODE_64BIT) {
       if ((byte & 0xf0) == 0x40) {
         uint8_t opcodeByte;
-          
+
         if (lookAtByte(insn, &opcodeByte) || ((opcodeByte & 0xf0) == 0x40)) {
           dbgprintf(insn, "Redundant REX prefix");
           return -1;
         }
-          
+
         insn->rexPrefix = byte;
         insn->necessaryPrefixLocation = insn->readerCursor - 2;
-          
+
         dbgprintf(insn, "Found REX prefix 0x%hhx", byte);
-      } else {                
+      } else {
         unconsumeByte(insn);
         insn->necessaryPrefixLocation = insn->readerCursor - 1;
       }
@@ -526,7 +539,7 @@ static int readPrefixes(struct InternalInstruction* insn) {
       insn->immediateSize      = (hasOpSize ? 2 : 4);
     }
   }
-  
+
   return 0;
 }
 
@@ -537,22 +550,22 @@ static int readPrefixes(struct InternalInstruction* insn) {
  * @param insn  - The instruction whose opcode is to be read.
  * @return      - 0 if the opcode could be read successfully; nonzero otherwise.
  */
-static int readOpcode(struct InternalInstruction* insn) {  
+static int readOpcode(struct InternalInstruction* insn) {
   /* Determine the length of the primary opcode */
-  
+
   uint8_t current;
-  
+
   dbgprintf(insn, "readOpcode()");
-  
+
   insn->opcodeType = ONEBYTE;
-    
+
   if (insn->vexSize == 3)
   {
     switch (mmmmmFromVEX2of3(insn->vexPrefix[1]))
     {
     default:
       dbgprintf(insn, "Unhandled m-mmmm field for instruction (0x%hhx)", mmmmmFromVEX2of3(insn->vexPrefix[1]));
-      return -1;      
+      return -1;
     case 0:
       break;
     case VEX_LOB_0F:
@@ -564,7 +577,7 @@ static int readOpcode(struct InternalInstruction* insn) {
       insn->threeByteEscape = 0x38;
       insn->opcodeType = THREEBYTE_38;
       return consumeByte(insn, &insn->opcode);
-    case VEX_LOB_0F3A:    
+    case VEX_LOB_0F3A:
       insn->twoByteEscape = 0x0f;
       insn->threeByteEscape = 0x3a;
       insn->opcodeType = THREEBYTE_3A;
@@ -577,68 +590,68 @@ static int readOpcode(struct InternalInstruction* insn) {
     insn->opcodeType = TWOBYTE;
     return consumeByte(insn, &insn->opcode);
   }
-    
+
   if (consumeByte(insn, &current))
     return -1;
-  
+
   if (current == 0x0f) {
     dbgprintf(insn, "Found a two-byte escape prefix (0x%hhx)", current);
-    
+
     insn->twoByteEscape = current;
-    
+
     if (consumeByte(insn, &current))
       return -1;
-    
+
     if (current == 0x38) {
       dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current);
-      
+
       insn->threeByteEscape = current;
-      
+
       if (consumeByte(insn, &current))
         return -1;
-      
+
       insn->opcodeType = THREEBYTE_38;
     } else if (current == 0x3a) {
       dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current);
-      
+
       insn->threeByteEscape = current;
-      
+
       if (consumeByte(insn, &current))
         return -1;
-      
+
       insn->opcodeType = THREEBYTE_3A;
     } else if (current == 0xa6) {
       dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current);
-      
+
       insn->threeByteEscape = current;
-      
+
       if (consumeByte(insn, &current))
         return -1;
-      
+
       insn->opcodeType = THREEBYTE_A6;
     } else if (current == 0xa7) {
       dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current);
-      
+
       insn->threeByteEscape = current;
-      
+
       if (consumeByte(insn, &current))
         return -1;
-      
+
       insn->opcodeType = THREEBYTE_A7;
     } else {
       dbgprintf(insn, "Didn't find a three-byte escape prefix");
-      
+
       insn->opcodeType = TWOBYTE;
     }
   }
-  
+
   /*
    * At this point we have consumed the full opcode.
    * Anything we consume from here on must be unconsumed.
    */
-  
+
   insn->opcode = current;
-  
+
   return 0;
 }
 
@@ -660,19 +673,19 @@ static int getIDWithAttrMask(uint16_t* instructionID,
                              struct InternalInstruction* insn,
                              uint8_t attrMask) {
   BOOL hasModRMExtension;
-  
+
   uint8_t instructionClass;
 
   instructionClass = contextForAttrs(attrMask);
-  
+
   hasModRMExtension = modRMRequired(insn->opcodeType,
                                     instructionClass,
                                     insn->opcode);
-  
+
   if (hasModRMExtension) {
     if (readModRM(insn))
       return -1;
-    
+
     *instructionID = decode(insn->opcodeType,
                             instructionClass,
                             insn->opcode,
@@ -683,7 +696,7 @@ static int getIDWithAttrMask(uint16_t* instructionID,
                             insn->opcode,
                             0);
   }
-      
+
   return 0;
 }
 
@@ -696,7 +709,7 @@ static int getIDWithAttrMask(uint16_t* instructionID,
  */
 static BOOL is16BitEquivalent(const char* orig, const char* equiv) {
   off_t i;
-  
+
   for (i = 0;; i++) {
     if (orig[i] == '\0' && equiv[i] == '\0')
       return TRUE;
@@ -715,8 +728,8 @@ static BOOL is16BitEquivalent(const char* orig, const char* equiv) {
 }
 
 /*
- * getID - Determines the ID of an instruction, consuming the ModR/M byte as 
- *   appropriate for extended and escape opcodes.  Determines the attributes and 
+ * getID - Determines the ID of an instruction, consuming the ModR/M byte as
+ *   appropriate for extended and escape opcodes.  Determines the attributes and
  *   context for the instruction before doing so.
  *
  * @param insn  - The instruction whose ID is to be determined.
@@ -726,21 +739,21 @@ static BOOL is16BitEquivalent(const char* orig, const char* equiv) {
 static int getID(struct InternalInstruction* insn, const void *miiArg) {
   uint8_t attrMask;
   uint16_t instructionID;
-  
+
   dbgprintf(insn, "getID()");
-    
+
   attrMask = ATTR_NONE;
 
   if (insn->mode == MODE_64BIT)
     attrMask |= ATTR_64BIT;
-    
+
   if (insn->vexSize) {
     attrMask |= ATTR_VEX;
 
     if (insn->vexSize == 3) {
       switch (ppFromVEX3of3(insn->vexPrefix[2])) {
       case VEX_PREFIX_66:
-        attrMask |= ATTR_OPSIZE;    
+        attrMask |= ATTR_OPSIZE;
         break;
       case VEX_PREFIX_F3:
         attrMask |= ATTR_XS;
@@ -749,14 +762,14 @@ static int getID(struct InternalInstruction* insn, const void *miiArg) {
         attrMask |= ATTR_XD;
         break;
       }
-    
+
       if (lFromVEX3of3(insn->vexPrefix[2]))
         attrMask |= ATTR_VEXL;
     }
     else if (insn->vexSize == 2) {
       switch (ppFromVEX2of2(insn->vexPrefix[1])) {
       case VEX_PREFIX_66:
-        attrMask |= ATTR_OPSIZE;    
+        attrMask |= ATTR_OPSIZE;
         break;
       case VEX_PREFIX_F3:
         attrMask |= ATTR_XS;
@@ -765,7 +778,7 @@ static int getID(struct InternalInstruction* insn, const void *miiArg) {
         attrMask |= ATTR_XD;
         break;
       }
-    
+
       if (lFromVEX2of2(insn->vexPrefix[1]))
         attrMask |= ATTR_VEXL;
     }
@@ -836,26 +849,26 @@ static int getID(struct InternalInstruction* insn, const void *miiArg) {
      * conservative, but in the specific case where OpSize is present but not
      * in the right place we check if there's a 16-bit operation.
      */
-    
+
     const struct InstructionSpecifier *spec;
     uint16_t instructionIDWithOpsize;
     const char *specName, *specWithOpSizeName;
-    
+
     spec = specifierForUID(instructionID);
-    
+
     if (getIDWithAttrMask(&instructionIDWithOpsize,
                           insn,
                           attrMask | ATTR_OPSIZE)) {
-      /* 
+      /*
        * ModRM required with OpSize but not present; give up and return version
        * without OpSize set
        */
-      
+
       insn->instructionID = instructionID;
       insn->spec = spec;
       return 0;
     }
-    
+
     specName = x86DisassemblerGetInstrName(instructionID, miiArg);
     specWithOpSizeName =
       x86DisassemblerGetInstrName(instructionIDWithOpsize, miiArg);
@@ -882,10 +895,10 @@ static int getID(struct InternalInstruction* insn, const void *miiArg) {
     const struct InstructionSpecifier *specWithNewOpcode;
 
     spec = specifierForUID(instructionID);
-    
+
     /* Borrow opcode from one of the other XCHGar opcodes */
     insn->opcode = 0x91;
-   
+
     if (getIDWithAttrMask(&instructionIDWithNewOpcode,
                           insn,
                           attrMask)) {
@@ -906,10 +919,10 @@ static int getID(struct InternalInstruction* insn, const void *miiArg) {
 
     return 0;
   }
-  
+
   insn->instructionID = instructionID;
   insn->spec = specifierForUID(insn->instructionID);
-  
+
   return 0;
 }
 
@@ -924,14 +937,14 @@ static int readSIB(struct InternalInstruction* insn) {
   SIBIndex sibIndexBase = 0;
   SIBBase sibBaseBase = 0;
   uint8_t index, base;
-  
+
   dbgprintf(insn, "readSIB()");
-  
+
   if (insn->consumedSIB)
     return 0;
-  
+
   insn->consumedSIB = TRUE;
-  
+
   switch (insn->addressSize) {
   case 2:
     dbgprintf(insn, "SIB-based addressing doesn't work in 16-bit mode");
@@ -949,9 +962,9 @@ static int readSIB(struct InternalInstruction* insn) {
 
   if (consumeByte(insn, &insn->sib))
     return -1;
-  
+
   index = indexFromSIB(insn->sib) | (xFromREX(insn->rexPrefix) << 3);
-  
+
   switch (index) {
   case 0x4:
     insn->sibIndex = SIB_INDEX_NONE;
@@ -963,7 +976,7 @@ static int readSIB(struct InternalInstruction* insn) {
       insn->sibIndex = SIB_INDEX_NONE;
     break;
   }
-  
+
   switch (scaleFromSIB(insn->sib)) {
   case 0:
     insn->sibScale = 1;
@@ -978,9 +991,9 @@ static int readSIB(struct InternalInstruction* insn) {
     insn->sibScale = 8;
     break;
   }
-  
+
   base = baseFromSIB(insn->sib) | (bFromREX(insn->rexPrefix) << 3);
-  
+
   switch (base) {
   case 0x5:
     switch (modFromModRM(insn->modRM)) {
@@ -990,12 +1003,12 @@ static int readSIB(struct InternalInstruction* insn) {
       break;
     case 0x1:
       insn->eaDisplacement = EA_DISP_8;
-      insn->sibBase = (insn->addressSize == 4 ? 
+      insn->sibBase = (insn->addressSize == 4 ?
                        SIB_BASE_EBP : SIB_BASE_RBP);
       break;
     case 0x2:
       insn->eaDisplacement = EA_DISP_32;
-      insn->sibBase = (insn->addressSize == 4 ? 
+      insn->sibBase = (insn->addressSize == 4 ?
                        SIB_BASE_EBP : SIB_BASE_RBP);
       break;
     case 0x3:
@@ -1007,7 +1020,7 @@ static int readSIB(struct InternalInstruction* insn) {
     insn->sibBase = (SIBBase)(sibBaseBase + base);
     break;
   }
-  
+
   return 0;
 }
 
@@ -1015,22 +1028,22 @@ static int readSIB(struct InternalInstruction* insn) {
  * readDisplacement - Consumes the displacement of an instruction.
  *
  * @param insn  - The instruction whose displacement is to be read.
- * @return      - 0 if the displacement byte was successfully read; nonzero 
+ * @return      - 0 if the displacement byte was successfully read; nonzero
  *                otherwise.
  */
-static int readDisplacement(struct InternalInstruction* insn) {  
+static int readDisplacement(struct InternalInstruction* insn) {
   int8_t d8;
   int16_t d16;
   int32_t d32;
-  
+
   dbgprintf(insn, "readDisplacement()");
-  
+
   if (insn->consumedDisplacement)
     return 0;
-  
+
   insn->consumedDisplacement = TRUE;
   insn->displacementOffset = insn->readerCursor - insn->startLocation;
-  
+
   switch (insn->eaDisplacement) {
   case EA_DISP_NONE:
     insn->consumedDisplacement = FALSE;
@@ -1051,7 +1064,7 @@ static int readDisplacement(struct InternalInstruction* insn) {
     insn->displacement = d32;
     break;
   }
-  
+
   insn->consumedDisplacement = TRUE;
   return 0;
 }
@@ -1063,22 +1076,22 @@ static int readDisplacement(struct InternalInstruction* insn) {
  * @param insn  - The instruction whose addressing information is to be read.
  * @return      - 0 if the information was successfully read; nonzero otherwise.
  */
-static int readModRM(struct InternalInstruction* insn) {  
+static int readModRM(struct InternalInstruction* insn) {
   uint8_t mod, rm, reg;
-  
+
   dbgprintf(insn, "readModRM()");
-  
+
   if (insn->consumedModRM)
     return 0;
-  
+
   if (consumeByte(insn, &insn->modRM))
     return -1;
   insn->consumedModRM = TRUE;
-  
+
   mod     = modFromModRM(insn->modRM);
   rm      = rmFromModRM(insn->modRM);
   reg     = regFromModRM(insn->modRM);
-  
+
   /*
    * This goes by insn->registerSize to pick the correct register, which messes
    * up if we're using (say) XMM or 8-bit register operands.  That gets fixed in
@@ -1098,16 +1111,16 @@ static int readModRM(struct InternalInstruction* insn) {
     insn->eaRegBase = EA_REG_RAX;
     break;
   }
-  
+
   reg |= rFromREX(insn->rexPrefix) << 3;
   rm  |= bFromREX(insn->rexPrefix) << 3;
-  
+
   insn->reg = (Reg)(insn->regBase + reg);
-  
+
   switch (insn->addressSize) {
   case 2:
     insn->eaBaseBase = EA_BASE_BX_SI;
-     
+
     switch (mod) {
     case 0x0:
       if (rm == 0x6) {
@@ -1142,14 +1155,14 @@ static int readModRM(struct InternalInstruction* insn) {
   case 4:
   case 8:
     insn->eaBaseBase = (insn->addressSize == 4 ? EA_BASE_EAX : EA_BASE_RAX);
-    
+
     switch (mod) {
     case 0x0:
       insn->eaDisplacement = EA_DISP_NONE; /* readSIB may override this */
       switch (rm) {
       case 0x4:
       case 0xc:   /* in case REXW.b is set */
-        insn->eaBase = (insn->addressSize == 4 ? 
+        insn->eaBase = (insn->addressSize == 4 ?
                         EA_BASE_sib : EA_BASE_sib64);
         readSIB(insn);
         if (readDisplacement(insn))
@@ -1191,7 +1204,7 @@ static int readModRM(struct InternalInstruction* insn) {
     }
     break;
   } /* switch (insn->addressSize) */
-  
+
   return 0;
 }
 
@@ -1274,12 +1287,12 @@ GENERIC_FIXUP_FUNC(fixupRMValue,  insn->eaRegBase,  EA_REG)
  * @return      - 0 if fixup was successful; -1 if the register returned was
  *                invalid for its class.
  */
-static int fixupReg(struct InternalInstruction *insn, 
+static int fixupReg(struct InternalInstruction *insn,
                     const struct OperandSpecifier *op) {
   uint8_t valid;
-  
+
   dbgprintf(insn, "fixupReg()");
-  
+
   switch ((OperandEncoding)op->encoding) {
   default:
     debug("Expected a REG or R/M encoding in fixupReg");
@@ -1311,12 +1324,12 @@ static int fixupReg(struct InternalInstruction *insn,
     }
     break;
   }
-  
+
   return 0;
 }
 
 /*
- * readOpcodeModifier - Reads an operand from the opcode field of an 
+ * readOpcodeModifier - Reads an operand from the opcode field of an
  *   instruction.  Handles AddRegFrm instructions.
  *
  * @param insn    - The instruction whose opcode field is to be read.
@@ -1326,12 +1339,12 @@ static int fixupReg(struct InternalInstruction *insn,
  */
 static int readOpcodeModifier(struct InternalInstruction* insn) {
   dbgprintf(insn, "readOpcodeModifier()");
-  
+
   if (insn->consumedOpcodeModifier)
     return 0;
-  
+
   insn->consumedOpcodeModifier = TRUE;
-  
+
   switch (insn->spec->modifierType) {
   default:
     debug("Unknown modifier type.");
@@ -1345,11 +1358,11 @@ static int readOpcodeModifier(struct InternalInstruction* insn) {
   case MODIFIER_MODRM:
     insn->opcodeModifier = insn->modRM - insn->spec->modifierBase;
     return 0;
-  }  
+  }
 }
 
 /*
- * readOpcodeRegister - Reads an operand from the opcode field of an 
+ * readOpcodeRegister - Reads an operand from the opcode field of an
  *   instruction and interprets it appropriately given the operand width.
  *   Handles AddRegFrm instructions.
  *
@@ -1364,39 +1377,39 @@ static int readOpcodeRegister(struct InternalInstruction* insn, uint8_t size) {
 
   if (readOpcodeModifier(insn))
     return -1;
-  
+
   if (size == 0)
     size = insn->registerSize;
-  
+
   switch (size) {
   case 1:
-    insn->opcodeRegister = (Reg)(MODRM_REG_AL + ((bFromREX(insn->rexPrefix) << 3) 
+    insn->opcodeRegister = (Reg)(MODRM_REG_AL + ((bFromREX(insn->rexPrefix) << 3)
                                                   | insn->opcodeModifier));
-    if (insn->rexPrefix && 
+    if (insn->rexPrefix &&
         insn->opcodeRegister >= MODRM_REG_AL + 0x4 &&
         insn->opcodeRegister < MODRM_REG_AL + 0x8) {
       insn->opcodeRegister = (Reg)(MODRM_REG_SPL
                                    + (insn->opcodeRegister - MODRM_REG_AL - 4));
     }
-      
+
     break;
   case 2:
     insn->opcodeRegister = (Reg)(MODRM_REG_AX
-                                 + ((bFromREX(insn->rexPrefix) << 3) 
+                                 + ((bFromREX(insn->rexPrefix) << 3)
                                     | insn->opcodeModifier));
     break;
   case 4:
     insn->opcodeRegister = (Reg)(MODRM_REG_EAX
-                                 + ((bFromREX(insn->rexPrefix) << 3) 
+                                 + ((bFromREX(insn->rexPrefix) << 3)
                                     | insn->opcodeModifier));
     break;
   case 8:
-    insn->opcodeRegister = (Reg)(MODRM_REG_RAX 
-                                 + ((bFromREX(insn->rexPrefix) << 3) 
+    insn->opcodeRegister = (Reg)(MODRM_REG_RAX
+                                 + ((bFromREX(insn->rexPrefix) << 3)
                                     | insn->opcodeModifier));
     break;
   }
-  
+
   return 0;
 }
 
@@ -1414,20 +1427,20 @@ static int readImmediate(struct InternalInstruction* insn, uint8_t size) {
   uint16_t imm16;
   uint32_t imm32;
   uint64_t imm64;
-  
+
   dbgprintf(insn, "readImmediate()");
-  
+
   if (insn->numImmediatesConsumed == 2) {
     debug("Already consumed two immediates");
     return -1;
   }
-  
+
   if (size == 0)
     size = insn->immediateSize;
   else
     insn->immediateSize = size;
   insn->immediateOffset = insn->readerCursor - insn->startLocation;
-  
+
   switch (size) {
   case 1:
     if (consumeByte(insn, &imm8))
@@ -1450,9 +1463,9 @@ static int readImmediate(struct InternalInstruction* insn, uint8_t size) {
     insn->immediates[insn->numImmediatesConsumed] = imm64;
     break;
   }
-  
+
   insn->numImmediatesConsumed++;
-  
+
   return 0;
 }
 
@@ -1465,7 +1478,7 @@ static int readImmediate(struct InternalInstruction* insn, uint8_t size) {
  */
 static int readVVVV(struct InternalInstruction* insn) {
   dbgprintf(insn, "readVVVV()");
-        
+
   if (insn->vexSize == 3)
     insn->vvvv = vvvvFromVEX3of3(insn->vexPrefix[2]);
   else if (insn->vexSize == 2)
@@ -1490,14 +1503,14 @@ static int readOperands(struct InternalInstruction* insn) {
   int index;
   int hasVVVV, needVVVV;
   int sawRegImm = 0;
-  
+
   dbgprintf(insn, "readOperands()");
 
   /* If non-zero vvvv specified, need to make sure one of the operands
      uses it. */
   hasVVVV = !readVVVV(insn);
   needVVVV = hasVVVV && (insn->vvvv != 0);
-  
+
   for (index = 0; index < X86_MAX_OPERANDS; ++index) {
     switch (x86OperandSets[insn->spec->operands][index].encoding) {
     case ENCODING_NONE:
@@ -1599,7 +1612,7 @@ static int readOperands(struct InternalInstruction* insn) {
 
   /* If we didn't find ENCODING_VVVV operand, but non-zero vvvv present, fail */
   if (needVVVV) return -1;
-  
+
   return 0;
 }
 
@@ -1607,7 +1620,7 @@ static int readOperands(struct InternalInstruction* insn) {
  * decodeInstruction - Reads and interprets a full instruction provided by the
  *   user.
  *
- * @param insn      - A pointer to the instruction to be populated.  Must be 
+ * @param insn      - A pointer to the instruction to be populated.  Must be
  *                    pre-allocated.
  * @param reader    - The function to be used to read the instruction's bytes.
  * @param readerArg - A generic argument to be passed to the reader to store
@@ -1632,7 +1645,7 @@ int decodeInstruction(struct InternalInstruction* insn,
                       uint64_t startLoc,
                       DisassemblerMode mode) {
   memset(insn, 0, sizeof(struct InternalInstruction));
-    
+
   insn->reader = reader;
   insn->readerArg = readerArg;
   insn->dlog = logger;
@@ -1641,7 +1654,7 @@ int decodeInstruction(struct InternalInstruction* insn,
   insn->readerCursor = startLoc;
   insn->mode = mode;
   insn->numImmediatesConsumed = 0;
-  
+
   if (readPrefixes(insn)       ||
       readOpcode(insn)         ||
       getID(insn, miiArg)      ||
@@ -1650,14 +1663,14 @@ int decodeInstruction(struct InternalInstruction* insn,
     return -1;
 
   insn->operands = &x86OperandSets[insn->spec->operands][0];
-  
+
   insn->length = insn->readerCursor - insn->startLocation;
-  
+
   dbgprintf(insn, "Read from 0x%llx to 0x%llx: length %zu",
             startLoc, insn->readerCursor, insn->length);
-    
+
   if (insn->length > 15)
     dbgprintf(insn, "Instruction exceeds 15-byte limit");
-  
+
   return 0;
 }
diff --git a/lib/Target/X86/InstPrinter/X86ATTInstPrinter.cpp b/lib/Target/X86/InstPrinter/X86ATTInstPrinter.cpp
index a4bd1147bc51..e357710b20eb 100644
--- a/lib/Target/X86/InstPrinter/X86ATTInstPrinter.cpp
+++ b/lib/Target/X86/InstPrinter/X86ATTInstPrinter.cpp
@@ -14,12 +14,12 @@
 
 #define DEBUG_TYPE "asm-printer"
 #include "X86ATTInstPrinter.h"
-#include "X86InstComments.h"
 #include "MCTargetDesc/X86BaseInfo.h"
 #include "MCTargetDesc/X86MCTargetDesc.h"
-#include "llvm/MC/MCInst.h"
+#include "X86InstComments.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
 #include "llvm/Support/ErrorHandling.h"
@@ -131,7 +131,7 @@ void X86ATTInstPrinter::printPCRelImm(const MCInst *MI, unsigned OpNo,
                                       raw_ostream &O) {
   const MCOperand &Op = MI->getOperand(OpNo);
   if (Op.isImm())
-    O << Op.getImm();
+    O << formatImm(Op.getImm());
   else {
     assert(Op.isExpr() && "unknown pcrel immediate operand");
     // If a symbolic branch target was added as a constant expression then print
@@ -157,7 +157,7 @@ void X86ATTInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
   } else if (Op.isImm()) {
     // Print X86 immediates as signed values.
     O << markup("<imm:")
-      << '$' << (int64_t)Op.getImm()
+      << '$' << formatImm((int64_t)Op.getImm())
       << markup(">");
     
     if (CommentStream && (Op.getImm() > 255 || Op.getImm() < -256))
@@ -189,7 +189,7 @@ void X86ATTInstPrinter::printMemReference(const MCInst *MI, unsigned Op,
   if (DispSpec.isImm()) {
     int64_t DispVal = DispSpec.getImm();
     if (DispVal || (!IndexReg.getReg() && !BaseReg.getReg()))
-      O << DispVal;
+      O << formatImm(DispVal);
   } else {
     assert(DispSpec.isExpr() && "non-immediate displacement for LEA?");
     O << *DispSpec.getExpr();
@@ -207,7 +207,7 @@ void X86ATTInstPrinter::printMemReference(const MCInst *MI, unsigned Op,
       if (ScaleVal != 1) {
         O << ','
 	  << markup("<imm:")
-          << ScaleVal
+          << ScaleVal // never printed in hex.
 	  << markup(">");
       }
     }
diff --git a/lib/Target/X86/InstPrinter/X86InstComments.cpp b/lib/Target/X86/InstPrinter/X86InstComments.cpp
index 64ac5e685f76..0f6eeb19bccd 100644
--- a/lib/Target/X86/InstPrinter/X86InstComments.cpp
+++ b/lib/Target/X86/InstPrinter/X86InstComments.cpp
@@ -34,10 +34,6 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
 
   switch (MI->getOpcode()) {
   case X86::INSERTPSrr:
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    DecodeINSERTPSMask(MI->getOperand(3).getImm(), ShuffleMask);
-    break;
   case X86::VINSERTPSrr:
     DestName = getRegName(MI->getOperand(0).getReg());
     Src1Name = getRegName(MI->getOperand(1).getReg());
@@ -46,10 +42,6 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     break;
 
   case X86::MOVLHPSrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    DecodeMOVLHPSMask(2, ShuffleMask);
-    break;
   case X86::VMOVLHPSrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     Src1Name = getRegName(MI->getOperand(1).getReg());
@@ -58,10 +50,6 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     break;
 
   case X86::MOVHLPSrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    DecodeMOVHLPSMask(2, ShuffleMask);
-    break;
   case X86::VMOVHLPSrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     Src1Name = getRegName(MI->getOperand(1).getReg());
@@ -69,6 +57,29 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     DecodeMOVHLPSMask(2, ShuffleMask);
     break;
 
+  case X86::PALIGNR128rr:
+  case X86::VPALIGNR128rr:
+    Src1Name = getRegName(MI->getOperand(2).getReg());
+    // FALL THROUGH.
+  case X86::PALIGNR128rm:
+  case X86::VPALIGNR128rm:
+    Src2Name = getRegName(MI->getOperand(1).getReg());
+    DestName = getRegName(MI->getOperand(0).getReg());
+    DecodePALIGNRMask(MVT::v16i8,
+                      MI->getOperand(MI->getNumOperands()-1).getImm(),
+                      ShuffleMask);
+    break;
+  case X86::VPALIGNR256rr:
+    Src1Name = getRegName(MI->getOperand(2).getReg());
+    // FALL THROUGH.
+  case X86::VPALIGNR256rm:
+    Src2Name = getRegName(MI->getOperand(1).getReg());
+    DestName = getRegName(MI->getOperand(0).getReg());
+    DecodePALIGNRMask(MVT::v32i8,
+                      MI->getOperand(MI->getNumOperands()-1).getImm(),
+                      ShuffleMask);
+    break;
+
   case X86::PSHUFDri:
   case X86::VPSHUFDri:
     Src1Name = getRegName(MI->getOperand(1).getReg());
@@ -131,15 +142,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     break;
 
   case X86::PUNPCKHBWrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::PUNPCKHBWrm:
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKHMask(MVT::v16i8, ShuffleMask);
-    break;
   case X86::VPUNPCKHBWrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::PUNPCKHBWrm:
   case X86::VPUNPCKHBWrm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
@@ -154,15 +160,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     DecodeUNPCKHMask(MVT::v32i8, ShuffleMask);
     break;
   case X86::PUNPCKHWDrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::PUNPCKHWDrm:
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKHMask(MVT::v8i16, ShuffleMask);
-    break;
   case X86::VPUNPCKHWDrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::PUNPCKHWDrm:
   case X86::VPUNPCKHWDrm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
@@ -177,15 +178,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     DecodeUNPCKHMask(MVT::v16i16, ShuffleMask);
     break;
   case X86::PUNPCKHDQrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::PUNPCKHDQrm:
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKHMask(MVT::v4i32, ShuffleMask);
-    break;
   case X86::VPUNPCKHDQrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::PUNPCKHDQrm:
   case X86::VPUNPCKHDQrm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
@@ -200,15 +196,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     DecodeUNPCKHMask(MVT::v8i32, ShuffleMask);
     break;
   case X86::PUNPCKHQDQrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::PUNPCKHQDQrm:
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKHMask(MVT::v2i64, ShuffleMask);
-    break;
   case X86::VPUNPCKHQDQrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::PUNPCKHQDQrm:
   case X86::VPUNPCKHQDQrm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
@@ -224,15 +215,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     break;
 
   case X86::PUNPCKLBWrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::PUNPCKLBWrm:
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKLMask(MVT::v16i8, ShuffleMask);
-    break;
   case X86::VPUNPCKLBWrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::PUNPCKLBWrm:
   case X86::VPUNPCKLBWrm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
@@ -247,15 +233,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     DecodeUNPCKLMask(MVT::v32i8, ShuffleMask);
     break;
   case X86::PUNPCKLWDrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::PUNPCKLWDrm:
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKLMask(MVT::v8i16, ShuffleMask);
-    break;
   case X86::VPUNPCKLWDrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::PUNPCKLWDrm:
   case X86::VPUNPCKLWDrm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
@@ -270,15 +251,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     DecodeUNPCKLMask(MVT::v16i16, ShuffleMask);
     break;
   case X86::PUNPCKLDQrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::PUNPCKLDQrm:
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKLMask(MVT::v4i32, ShuffleMask);
-    break;
   case X86::VPUNPCKLDQrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::PUNPCKLDQrm:
   case X86::VPUNPCKLDQrm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
@@ -293,15 +269,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     DecodeUNPCKLMask(MVT::v8i32, ShuffleMask);
     break;
   case X86::PUNPCKLQDQrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::PUNPCKLQDQrm:
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKLMask(MVT::v2i64, ShuffleMask);
-    break;
   case X86::VPUNPCKLQDQrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::PUNPCKLQDQrm:
   case X86::VPUNPCKLQDQrm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
@@ -317,16 +288,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     break;
 
   case X86::SHUFPDrri:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::SHUFPDrmi:
-    DecodeSHUFPMask(MVT::v2f64, MI->getOperand(MI->getNumOperands()-1).getImm(),
-                    ShuffleMask);
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    break;
   case X86::VSHUFPDrri:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::SHUFPDrmi:
   case X86::VSHUFPDrmi:
     DecodeSHUFPMask(MVT::v2f64, MI->getOperand(MI->getNumOperands()-1).getImm(),
                     ShuffleMask);
@@ -344,16 +309,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     break;
 
   case X86::SHUFPSrri:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::SHUFPSrmi:
-    DecodeSHUFPMask(MVT::v4f32, MI->getOperand(MI->getNumOperands()-1).getImm(),
-                    ShuffleMask);
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    break;
   case X86::VSHUFPSrri:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::SHUFPSrmi:
   case X86::VSHUFPSrmi:
     DecodeSHUFPMask(MVT::v4f32, MI->getOperand(MI->getNumOperands()-1).getImm(),
                     ShuffleMask);
@@ -371,15 +330,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     break;
 
   case X86::UNPCKLPDrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::UNPCKLPDrm:
-    DecodeUNPCKLMask(MVT::v2f64, ShuffleMask);
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    break;
   case X86::VUNPCKLPDrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::UNPCKLPDrm:
   case X86::VUNPCKLPDrm:
     DecodeUNPCKLMask(MVT::v2f64, ShuffleMask);
     Src1Name = getRegName(MI->getOperand(1).getReg());
@@ -394,15 +348,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   case X86::UNPCKLPSrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::UNPCKLPSrm:
-    DecodeUNPCKLMask(MVT::v4f32, ShuffleMask);
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    break;
   case X86::VUNPCKLPSrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::UNPCKLPSrm:
   case X86::VUNPCKLPSrm:
     DecodeUNPCKLMask(MVT::v4f32, ShuffleMask);
     Src1Name = getRegName(MI->getOperand(1).getReg());
@@ -417,15 +366,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   case X86::UNPCKHPDrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::UNPCKHPDrm:
-    DecodeUNPCKHMask(MVT::v2f64, ShuffleMask);
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    break;
   case X86::VUNPCKHPDrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::UNPCKHPDrm:
   case X86::VUNPCKHPDrm:
     DecodeUNPCKHMask(MVT::v2f64, ShuffleMask);
     Src1Name = getRegName(MI->getOperand(1).getReg());
@@ -440,15 +384,10 @@ void llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   case X86::UNPCKHPSrr:
-    Src2Name = getRegName(MI->getOperand(2).getReg());
-    // FALL THROUGH.
-  case X86::UNPCKHPSrm:
-    DecodeUNPCKHMask(MVT::v4f32, ShuffleMask);
-    Src1Name = getRegName(MI->getOperand(0).getReg());
-    break;
   case X86::VUNPCKHPSrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
+  case X86::UNPCKHPSrm:
   case X86::VUNPCKHPSrm:
     DecodeUNPCKHMask(MVT::v4f32, ShuffleMask);
     Src1Name = getRegName(MI->getOperand(1).getReg());
diff --git a/lib/Target/X86/InstPrinter/X86IntelInstPrinter.cpp b/lib/Target/X86/InstPrinter/X86IntelInstPrinter.cpp
index d67aec7f10ef..141f4a4dd856 100644
--- a/lib/Target/X86/InstPrinter/X86IntelInstPrinter.cpp
+++ b/lib/Target/X86/InstPrinter/X86IntelInstPrinter.cpp
@@ -14,11 +14,11 @@
 
 #define DEBUG_TYPE "asm-printer"
 #include "X86IntelInstPrinter.h"
-#include "X86InstComments.h"
 #include "MCTargetDesc/X86BaseInfo.h"
 #include "MCTargetDesc/X86MCTargetDesc.h"
-#include "llvm/MC/MCInst.h"
+#include "X86InstComments.h"
 #include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormattedStream.h"
diff --git a/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp b/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp
index 467edadc7e09..598ddee56d21 100644
--- a/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp
+++ b/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp
@@ -113,7 +113,7 @@ public:
 
   bool fixupNeedsRelaxation(const MCFixup &Fixup,
                             uint64_t Value,
-                            const MCInstFragment *DF,
+                            const MCRelaxableFragment *DF,
                             const MCAsmLayout &Layout) const;
 
   void relaxInstruction(const MCInst &Inst, MCInst &Res) const;
@@ -255,7 +255,7 @@ bool X86AsmBackend::mayNeedRelaxation(const MCInst &Inst) const {
 
 bool X86AsmBackend::fixupNeedsRelaxation(const MCFixup &Fixup,
                                          uint64_t Value,
-                                         const MCInstFragment *DF,
+                                         const MCRelaxableFragment *DF,
                                          const MCAsmLayout &Layout) const {
   // Relax if the value is too big for a (signed) i8.
   return int64_t(Value) != int64_t(int8_t(Value));
@@ -279,9 +279,9 @@ void X86AsmBackend::relaxInstruction(const MCInst &Inst, MCInst &Res) const {
   Res.setOpcode(RelaxedOp);
 }
 
-/// writeNopData - Write optimal nops to the output file for the \p Count
-/// bytes.  This returns the number of bytes written.  It may return 0 if
-/// the \p Count is more than the maximum optimal nops.
+/// \brief Write a sequence of optimal nops to the output, covering \p Count
+/// bytes.
+/// \return - true on success, false on failure
 bool X86AsmBackend::writeNopData(uint64_t Count, MCObjectWriter *OW) const {
   static const uint8_t Nops[10][10] = {
     // nop
@@ -315,18 +315,18 @@ bool X86AsmBackend::writeNopData(uint64_t Count, MCObjectWriter *OW) const {
     return true;
   }
 
-  // Write an optimal sequence for the first 15 bytes.
-  const uint64_t OptimalCount = (Count < 16) ? Count : 15;
-  const uint64_t Prefixes = OptimalCount <= 10 ? 0 : OptimalCount - 10;
-  for (uint64_t i = 0, e = Prefixes; i != e; i++)
-    OW->Write8(0x66);
-  const uint64_t Rest = OptimalCount - Prefixes;
-  for (uint64_t i = 0, e = Rest; i != e; i++)
-    OW->Write8(Nops[Rest - 1][i]);
-
-  // Finish with single byte nops.
-  for (uint64_t i = OptimalCount, e = Count; i != e; ++i)
-   OW->Write8(0x90);
+  // 15 is the longest single nop instruction.  Emit as many 15-byte nops as
+  // needed, then emit a nop of the remaining length.
+  do {
+    const uint8_t ThisNopLength = (uint8_t) std::min(Count, (uint64_t) 15);
+    const uint8_t Prefixes = ThisNopLength <= 10 ? 0 : ThisNopLength - 10;
+    for (uint8_t i = 0; i < Prefixes; i++)
+      OW->Write8(0x66);
+    const uint8_t Rest = ThisNopLength - Prefixes;
+    for (uint8_t i = 0; i < Rest; i++)
+      OW->Write8(Nops[Rest - 1][i]);
+    Count -= ThisNopLength;
+  } while (Count != 0);
 
   return true;
 }
diff --git a/lib/Target/X86/MCTargetDesc/X86BaseInfo.h b/lib/Target/X86/MCTargetDesc/X86BaseInfo.h
index 7ea1961dec90..36695600707e 100644
--- a/lib/Target/X86/MCTargetDesc/X86BaseInfo.h
+++ b/lib/Target/X86/MCTargetDesc/X86BaseInfo.h
@@ -104,7 +104,7 @@ namespace X86II {
 
     /// MO_TLSLD - On a symbol operand this indicates that the immediate is
     /// the offset of the GOT entry with the TLS index for the module that
-    /// contains the symbol. When this index is passed to a call to to
+    /// contains the symbol. When this index is passed to a call to
     /// __tls_get_addr, the function will return the base address of the TLS
     /// block for the symbol. Used in the x86-64 local dynamic TLS access model.
     ///
@@ -114,7 +114,7 @@ namespace X86II {
 
     /// MO_TLSLDM - On a symbol operand this indicates that the immediate is
     /// the offset of the GOT entry with the TLS index for the module that
-    /// contains the symbol. When this index is passed to a call to to
+    /// contains the symbol. When this index is passed to a call to
     /// ___tls_get_addr, the function will return the base address of the TLS
     /// block for the symbol. Used in the IA32 local dynamic TLS access model.
     ///
@@ -276,9 +276,9 @@ namespace X86II {
     MRM_C1 = 33, MRM_C2 = 34, MRM_C3 = 35, MRM_C4 = 36,
     MRM_C8 = 37, MRM_C9 = 38, MRM_E8 = 39, MRM_F0 = 40,
     MRM_F8 = 41, MRM_F9 = 42, MRM_D0 = 45, MRM_D1 = 46,
-    MRM_D4 = 47, MRM_D5 = 48, MRM_D8 = 49, MRM_D9 = 50,
-    MRM_DA = 51, MRM_DB = 52, MRM_DC = 53, MRM_DD = 54,
-    MRM_DE = 55, MRM_DF = 56,
+    MRM_D4 = 47, MRM_D5 = 48, MRM_D6 = 49, MRM_D8 = 50,
+    MRM_D9 = 51, MRM_DA = 52, MRM_DB = 53, MRM_DC = 54,
+    MRM_DD = 55, MRM_DE = 56, MRM_DF = 57,
 
     /// RawFrmImm8 - This is used for the ENTER instruction, which has two
     /// immediates, the first of which is a 16-bit immediate (specified by
@@ -574,16 +574,13 @@ namespace X86II {
         ++FirstMemOp;// Skip the register dest (which is encoded in VEX_VVVV).
       return FirstMemOp;
     }
-    case X86II::MRM_C1: case X86II::MRM_C2:
-    case X86II::MRM_C3: case X86II::MRM_C4:
-    case X86II::MRM_C8: case X86II::MRM_C9:
-    case X86II::MRM_E8: case X86II::MRM_F0:
-    case X86II::MRM_F8: case X86II::MRM_F9:
-    case X86II::MRM_D0: case X86II::MRM_D1:
-    case X86II::MRM_D4: case X86II::MRM_D5:
-    case X86II::MRM_D8: case X86II::MRM_D9:
-    case X86II::MRM_DA: case X86II::MRM_DB:
-    case X86II::MRM_DC: case X86II::MRM_DD:
+    case X86II::MRM_C1: case X86II::MRM_C2: case X86II::MRM_C3:
+    case X86II::MRM_C4: case X86II::MRM_C8: case X86II::MRM_C9:
+    case X86II::MRM_E8: case X86II::MRM_F0: case X86II::MRM_F8:
+    case X86II::MRM_F9: case X86II::MRM_D0: case X86II::MRM_D1:
+    case X86II::MRM_D4: case X86II::MRM_D5: case X86II::MRM_D6:
+    case X86II::MRM_D8: case X86II::MRM_D9: case X86II::MRM_DA:
+    case X86II::MRM_DB: case X86II::MRM_DC: case X86II::MRM_DD:
     case X86II::MRM_DE: case X86II::MRM_DF:
       return -1;
     }
diff --git a/lib/Target/X86/MCTargetDesc/X86MCAsmInfo.cpp b/lib/Target/X86/MCTargetDesc/X86MCAsmInfo.cpp
index 16488eb7ae7e..7815ae98c9bd 100644
--- a/lib/Target/X86/MCTargetDesc/X86MCAsmInfo.cpp
+++ b/lib/Target/X86/MCTargetDesc/X86MCAsmInfo.cpp
@@ -44,7 +44,7 @@ void X86MCAsmInfoDarwin::anchor() { }
 X86MCAsmInfoDarwin::X86MCAsmInfoDarwin(const Triple &T) {
   bool is64Bit = T.getArch() == Triple::x86_64;
   if (is64Bit)
-    PointerSize = 8;
+    PointerSize = CalleeSaveStackSlotSize = 8;
 
   AssemblerDialect = AsmWriterFlavor;
 
@@ -76,8 +76,16 @@ X86_64MCAsmInfoDarwin::X86_64MCAsmInfoDarwin(const Triple &Triple)
 void X86ELFMCAsmInfo::anchor() { }
 
 X86ELFMCAsmInfo::X86ELFMCAsmInfo(const Triple &T) {
-  if (T.getArch() == Triple::x86_64)
-    PointerSize = 8;
+  bool is64Bit = T.getArch() == Triple::x86_64;
+  bool isX32 = T.getEnvironment() == Triple::GNUX32;
+
+  // For ELF, x86-64 pointer size depends on the ABI.
+  // For x86-64 without the x32 ABI, pointer size is 8. For x86 and for x86-64
+  // with the x32 ABI, pointer size remains the default 4.
+  PointerSize = (is64Bit && !isX32) ? 8 : 4;
+
+  // OTOH, stack slot size is always 8 for x86-64, even with the x32 ABI.
+  CalleeSaveStackSlotSize = is64Bit ? 8 : 4;
 
   AssemblerDialect = AsmWriterFlavor;
 
diff --git a/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp b/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
index 122204ae75c8..776cee1e35cc 100644
--- a/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
+++ b/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
@@ -446,6 +446,7 @@ void X86MCCodeEmitter::EmitVEXOpcodePrefix(uint64_t TSFlags, unsigned &CurByte,
                                            raw_ostream &OS) const {
   bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V;
   bool HasVEX_4VOp3 = (TSFlags >> X86II::VEXShift) & X86II::VEX_4VOp3;
+  bool HasMemOp4 = (TSFlags >> X86II::VEXShift) & X86II::MemOp4;
 
   // VEX_R: opcode externsion equivalent to REX.R in
   // 1's complement (inverted) form
@@ -650,12 +651,19 @@ void X86MCCodeEmitter::EmitVEXOpcodePrefix(uint64_t TSFlags, unsigned &CurByte,
     //  dst(ModR/M), src1(ModR/M)
     //  dst(ModR/M), src1(ModR/M), imm8
     //
+    //  FMA4:
+    //  dst(ModR/M.reg), src1(VEX_4V), src2(ModR/M), src3(VEX_I8IMM)
+    //  dst(ModR/M.reg), src1(VEX_4V), src2(VEX_I8IMM), src3(ModR/M),
     if (X86II::isX86_64ExtendedReg(MI.getOperand(CurOp).getReg()))
       VEX_R = 0x0;
     CurOp++;
 
     if (HasVEX_4V)
       VEX_4V = getVEXRegisterEncoding(MI, CurOp++);
+
+    if (HasMemOp4) // Skip second register source (encoded in I8IMM)
+      CurOp++;
+
     if (X86II::isX86_64ExtendedReg(MI.getOperand(CurOp).getReg()))
       VEX_B = 0x0;
     CurOp++;
@@ -666,9 +674,15 @@ void X86MCCodeEmitter::EmitVEXOpcodePrefix(uint64_t TSFlags, unsigned &CurByte,
     // MRMDestReg instructions forms:
     //  dst(ModR/M), src(ModR/M)
     //  dst(ModR/M), src(ModR/M), imm8
-    if (X86II::isX86_64ExtendedReg(MI.getOperand(0).getReg()))
+    //  dst(ModR/M), src1(VEX_4V), src2(ModR/M)
+    if (X86II::isX86_64ExtendedReg(MI.getOperand(CurOp).getReg()))
       VEX_B = 0x0;
-    if (X86II::isX86_64ExtendedReg(MI.getOperand(1).getReg()))
+    CurOp++;
+
+    if (HasVEX_4V)
+      VEX_4V = getVEXRegisterEncoding(MI, CurOp++);
+
+    if (X86II::isX86_64ExtendedReg(MI.getOperand(CurOp).getReg()))
       VEX_R = 0x0;
     break;
   case X86II::MRM0r: case X86II::MRM1r:
@@ -1038,9 +1052,14 @@ EncodeInstruction(const MCInst &MI, raw_ostream &OS,
 
   case X86II::MRMDestReg:
     EmitByte(BaseOpcode, CurByte, OS);
+    SrcRegNum = CurOp + 1;
+
+    if (HasVEX_4V) // Skip 1st src (which is encoded in VEX_VVVV)
+      ++SrcRegNum;
+
     EmitRegModRMByte(MI.getOperand(CurOp),
-                     GetX86RegNum(MI.getOperand(CurOp+1)), CurByte, OS);
-    CurOp += 2;
+                     GetX86RegNum(MI.getOperand(SrcRegNum)), CurByte, OS);
+    CurOp = SrcRegNum + 1;
     break;
 
   case X86II::MRMDestMem:
@@ -1117,16 +1136,13 @@ EncodeInstruction(const MCInst &MI, raw_ostream &OS,
                      TSFlags, CurByte, OS, Fixups);
     CurOp += X86::AddrNumOperands;
     break;
-  case X86II::MRM_C1: case X86II::MRM_C2:
-  case X86II::MRM_C3: case X86II::MRM_C4:
-  case X86II::MRM_C8: case X86II::MRM_C9:
-  case X86II::MRM_D0: case X86II::MRM_D1:
-  case X86II::MRM_D4: case X86II::MRM_D5:
-  case X86II::MRM_D8: case X86II::MRM_D9:
-  case X86II::MRM_DA: case X86II::MRM_DB:
-  case X86II::MRM_DC: case X86II::MRM_DD:
-  case X86II::MRM_DE: case X86II::MRM_DF:
-  case X86II::MRM_E8: case X86II::MRM_F0:
+  case X86II::MRM_C1: case X86II::MRM_C2: case X86II::MRM_C3:
+  case X86II::MRM_C4: case X86II::MRM_C8: case X86II::MRM_C9:
+  case X86II::MRM_D0: case X86II::MRM_D1: case X86II::MRM_D4:
+  case X86II::MRM_D5: case X86II::MRM_D6: case X86II::MRM_D8:
+  case X86II::MRM_D9: case X86II::MRM_DA: case X86II::MRM_DB:
+  case X86II::MRM_DC: case X86II::MRM_DD: case X86II::MRM_DE:
+  case X86II::MRM_DF: case X86II::MRM_E8: case X86II::MRM_F0:
   case X86II::MRM_F8: case X86II::MRM_F9:
     EmitByte(BaseOpcode, CurByte, OS);
 
@@ -1143,6 +1159,7 @@ EncodeInstruction(const MCInst &MI, raw_ostream &OS,
     case X86II::MRM_D1: MRM = 0xD1; break;
     case X86II::MRM_D4: MRM = 0xD4; break;
     case X86II::MRM_D5: MRM = 0xD5; break;
+    case X86II::MRM_D6: MRM = 0xD6; break;
     case X86II::MRM_D8: MRM = 0xD8; break;
     case X86II::MRM_D9: MRM = 0xD9; break;
     case X86II::MRM_DA: MRM = 0xDA; break;
diff --git a/lib/Target/X86/MCTargetDesc/X86MCTargetDesc.cpp b/lib/Target/X86/MCTargetDesc/X86MCTargetDesc.cpp
index 287c9f137a58..5e84530cd729 100644
--- a/lib/Target/X86/MCTargetDesc/X86MCTargetDesc.cpp
+++ b/lib/Target/X86/MCTargetDesc/X86MCTargetDesc.cpp
@@ -12,19 +12,19 @@
 //===----------------------------------------------------------------------===//
 
 #include "X86MCTargetDesc.h"
-#include "X86MCAsmInfo.h"
 #include "InstPrinter/X86ATTInstPrinter.h"
 #include "InstPrinter/X86IntelInstPrinter.h"
-#include "llvm/MC/MachineLocation.h"
+#include "X86MCAsmInfo.h"
+#include "llvm/ADT/Triple.h"
 #include "llvm/MC/MCCodeGenInfo.h"
 #include "llvm/MC/MCInstrAnalysis.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/MC/MCRegisterInfo.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSubtargetInfo.h"
-#include "llvm/ADT/Triple.h"
-#include "llvm/Support/Host.h"
+#include "llvm/MC/MachineLocation.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/Host.h"
 #include "llvm/Support/TargetRegistry.h"
 
 #define GET_REGINFO_MC_DESC
@@ -257,7 +257,8 @@ static MCRegisterInfo *createX86MCRegisterInfo(StringRef TT) {
   MCRegisterInfo *X = new MCRegisterInfo();
   InitX86MCRegisterInfo(X, RA,
                         X86_MC::getDwarfRegFlavour(TT, false),
-                        X86_MC::getDwarfRegFlavour(TT, true));
+                        X86_MC::getDwarfRegFlavour(TT, true),
+                        RA);
   X86_MC::InitLLVM2SEHRegisterMapping(X);
   return X;
 }
diff --git a/lib/Target/X86/MCTargetDesc/X86MachObjectWriter.cpp b/lib/Target/X86/MCTargetDesc/X86MachObjectWriter.cpp
index 7ff058edbc23..64f005c469bc 100644
--- a/lib/Target/X86/MCTargetDesc/X86MachObjectWriter.cpp
+++ b/lib/Target/X86/MCTargetDesc/X86MachObjectWriter.cpp
@@ -7,18 +7,18 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "MCTargetDesc/X86FixupKinds.h"
 #include "MCTargetDesc/X86MCTargetDesc.h"
-#include "llvm/MC/MCAssembler.h"
+#include "MCTargetDesc/X86FixupKinds.h"
+#include "llvm/ADT/Twine.h"
 #include "llvm/MC/MCAsmLayout.h"
+#include "llvm/MC/MCAssembler.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCMachObjectWriter.h"
 #include "llvm/MC/MCSectionMachO.h"
 #include "llvm/MC/MCValue.h"
-#include "llvm/ADT/Twine.h"
+#include "llvm/Object/MachOFormat.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/Format.h"
-#include "llvm/Object/MachOFormat.h"
 
 using namespace llvm;
 using namespace llvm::object;
diff --git a/lib/Target/X86/Makefile b/lib/Target/X86/Makefile
index 949661eb99e9..e518fecf044f 100644
--- a/lib/Target/X86/Makefile
+++ b/lib/Target/X86/Makefile
@@ -16,8 +16,7 @@ BUILT_SOURCES = X86GenRegisterInfo.inc X86GenInstrInfo.inc \
 		X86GenAsmWriter.inc X86GenAsmMatcher.inc \
                 X86GenAsmWriter1.inc X86GenDAGISel.inc  \
                 X86GenDisassemblerTables.inc X86GenFastISel.inc \
-                X86GenCallingConv.inc X86GenSubtargetInfo.inc \
-		X86GenEDInfo.inc
+                X86GenCallingConv.inc X86GenSubtargetInfo.inc
 
 DIRS = InstPrinter AsmParser Disassembler TargetInfo MCTargetDesc Utils
 
diff --git a/lib/Target/X86/README-SSE.txt b/lib/Target/X86/README-SSE.txt
index 40110353fc62..496b704ee85f 100644
--- a/lib/Target/X86/README-SSE.txt
+++ b/lib/Target/X86/README-SSE.txt
@@ -953,3 +953,12 @@ similarly, v[0]-v[1] should match to hsubpd, and {v[0]-v[1], w[0]-w[1]} should
 turn into hsubpd also.
 
 //===---------------------------------------------------------------------===//
+
+define <2 x i32> @foo(<2 x double> %in) {
+  %x = fptosi <2 x double> %in to <2 x i32>
+  ret <2 x i32> %x
+}
+
+Should compile into cvttpd2dq instead of being scalarized into 2 cvttsd2si.
+
+//===---------------------------------------------------------------------===//
diff --git a/lib/Target/X86/README.txt b/lib/Target/X86/README.txt
index 6a8a4fdf2520..b4285a071879 100644
--- a/lib/Target/X86/README.txt
+++ b/lib/Target/X86/README.txt
@@ -1568,43 +1568,6 @@ The second one is done for: Atom, Pentium Pro, all AMDs, Pentium 4, Nocona,
   Core 2, and "Generic"
 
 //===---------------------------------------------------------------------===//
-
-Testcase:
-int a(int x) { return (x & 127) > 31; }
-
-Current output:
-	movl	4(%esp), %eax
-	andl	$127, %eax
-	cmpl	$31, %eax
-	seta	%al
-	movzbl	%al, %eax
-	ret
-
-Ideal output:
-	xorl	%eax, %eax
-	testl	$96, 4(%esp)
-	setne	%al
-	ret
-
-This should definitely be done in instcombine, canonicalizing the range
-condition into a != condition.  We get this IR:
-
-define i32 @a(i32 %x) nounwind readnone {
-entry:
-	%0 = and i32 %x, 127		; <i32> [#uses=1]
-	%1 = icmp ugt i32 %0, 31		; <i1> [#uses=1]
-	%2 = zext i1 %1 to i32		; <i32> [#uses=1]
-	ret i32 %2
-}
-
-Instcombine prefers to strength reduce relational comparisons to equality
-comparisons when possible, this should be another case of that.  This could
-be handled pretty easily in InstCombiner::visitICmpInstWithInstAndIntCst, but it
-looks like InstCombiner::visitICmpInstWithInstAndIntCst should really already
-be redesigned to use ComputeMaskedBits and friends.
-
-
-//===---------------------------------------------------------------------===//
 Testcase:
 int x(int a) { return (a&0xf0)>>4; }
 
diff --git a/lib/Target/X86/TargetInfo/X86TargetInfo.cpp b/lib/Target/X86/TargetInfo/X86TargetInfo.cpp
index 52a67f763b0a..815d23588f11 100644
--- a/lib/Target/X86/TargetInfo/X86TargetInfo.cpp
+++ b/lib/Target/X86/TargetInfo/X86TargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "X86.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
diff --git a/lib/Target/X86/Utils/X86ShuffleDecode.cpp b/lib/Target/X86/Utils/X86ShuffleDecode.cpp
index 8b87c1f9c8ad..bbd490411f2d 100644
--- a/lib/Target/X86/Utils/X86ShuffleDecode.cpp
+++ b/lib/Target/X86/Utils/X86ShuffleDecode.cpp
@@ -61,6 +61,24 @@ void DecodeMOVLHPSMask(unsigned NElts, SmallVectorImpl<int> &ShuffleMask) {
     ShuffleMask.push_back(NElts+i);
 }
 
+void DecodePALIGNRMask(MVT VT, unsigned Imm,
+                       SmallVectorImpl<int> &ShuffleMask) {
+  unsigned NumElts = VT.getVectorNumElements();
+  unsigned Offset = Imm * (VT.getVectorElementType().getSizeInBits() / 8);
+
+  unsigned NumLanes = VT.getSizeInBits() / 128;
+  unsigned NumLaneElts = NumElts / NumLanes;
+
+  for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
+    for (unsigned i = 0; i != NumLaneElts; ++i) {
+      unsigned Base = i + Offset;
+      // if i+offset is out of this lane then we actually need the other source
+      if (Base >= NumLaneElts) Base += NumElts - NumLaneElts;
+      ShuffleMask.push_back(Base + l);
+    }
+  }
+}
+
 /// DecodePSHUFMask - This decodes the shuffle masks for pshufd, and vpermilp*.
 /// VT indicates the type of the vector allowing it to handle different
 /// datatypes and vector widths.
diff --git a/lib/Target/X86/Utils/X86ShuffleDecode.h b/lib/Target/X86/Utils/X86ShuffleDecode.h
index 70d8171a8154..017ab325ec51 100644
--- a/lib/Target/X86/Utils/X86ShuffleDecode.h
+++ b/lib/Target/X86/Utils/X86ShuffleDecode.h
@@ -35,6 +35,8 @@ void DecodeMOVHLPSMask(unsigned NElts, SmallVectorImpl<int> &ShuffleMask);
 // <0,2> or <0,1,4,5>
 void DecodeMOVLHPSMask(unsigned NElts, SmallVectorImpl<int> &ShuffleMask);
 
+void DecodePALIGNRMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+
 void DecodePSHUFMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
 
 void DecodePSHUFHWMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
diff --git a/lib/Target/X86/X86.h b/lib/Target/X86/X86.h
index dce5b4d2b008..1f9919f15955 100644
--- a/lib/Target/X86/X86.h
+++ b/lib/Target/X86/X86.h
@@ -63,11 +63,12 @@ FunctionPass *createX86JITCodeEmitterPass(X86TargetMachine &TM,
 ///
 FunctionPass *createEmitX86CodeToMemory();
 
-/// createX86MaxStackAlignmentHeuristicPass - This function returns a pass
-/// which determines whether the frame pointer register should be
-/// reserved in case dynamic stack alignment is later required.
-///
-FunctionPass *createX86MaxStackAlignmentHeuristicPass();
+/// \brief Creates an X86-specific Target Transformation Info pass.
+ImmutablePass *createX86TargetTransformInfoPass(const X86TargetMachine *TM);
+
+/// createX86PadShortFunctions - Return a pass that pads short functions
+/// with NOOPs. This will prevent a stall when returning on the Atom.
+FunctionPass *createX86PadShortFunctions();
 
 } // End llvm namespace
 
diff --git a/lib/Target/X86/X86.td b/lib/Target/X86/X86.td
index 8ad0bc08ac57..1dcc344e7f0d 100644
--- a/lib/Target/X86/X86.td
+++ b/lib/Target/X86/X86.td
@@ -120,11 +120,25 @@ def FeatureBMI2    : SubtargetFeature<"bmi2", "HasBMI2", "true",
                                       "Support BMI2 instructions">;
 def FeatureRTM     : SubtargetFeature<"rtm", "HasRTM", "true",
                                       "Support RTM instructions">;
+def FeatureHLE     : SubtargetFeature<"hle", "HasHLE", "true",
+                                      "Support HLE">;
+def FeatureADX     : SubtargetFeature<"adx", "HasADX", "true",
+                                      "Support ADX instructions">;
+def FeaturePRFCHW  : SubtargetFeature<"prfchw", "HasPRFCHW", "true",
+                                      "Support PRFCHW instructions">;
+def FeatureRDSEED  : SubtargetFeature<"rdseed", "HasRDSEED", "true",
+                                      "Support RDSEED instruction">;
 def FeatureLeaForSP : SubtargetFeature<"lea-sp", "UseLeaForSP", "true",
                                      "Use LEA for adjusting the stack pointer">;
 def FeatureSlowDivide : SubtargetFeature<"idiv-to-divb",
-                          "HasSlowDivide", "true",
-                          "Use small divide for positive values less than 256">;
+                                     "HasSlowDivide", "true",
+                                     "Use small divide for positive values less than 256">;
+def FeaturePadShortFunctions : SubtargetFeature<"pad-short-functions",
+                                     "PadShortFunctions", "true",
+                                     "Pad short functions">;
+def FeatureCallRegIndirect : SubtargetFeature<"call-reg-indirect",
+                                     "CallRegIndirect", "true",
+                                     "Call register indirect">;
 
 //===----------------------------------------------------------------------===//
 // X86 processors supported.
@@ -138,9 +152,6 @@ def ProcIntelAtom : SubtargetFeature<"atom", "X86ProcFamily", "IntelAtom",
 class Proc<string Name, list<SubtargetFeature> Features>
  : ProcessorModel<Name, GenericModel, Features>;
 
-class AtomProc<string Name, list<SubtargetFeature> Features>
- : ProcessorModel<Name, AtomModel, Features>;
-
 def : Proc<"generic",         []>;
 def : Proc<"i386",            []>;
 def : Proc<"i486",            []>;
@@ -155,47 +166,63 @@ def : Proc<"pentium3m",       [FeatureSSE1, FeatureSlowBTMem]>;
 def : Proc<"pentium-m",       [FeatureSSE2, FeatureSlowBTMem]>;
 def : Proc<"pentium4",        [FeatureSSE2]>;
 def : Proc<"pentium4m",       [FeatureSSE2, FeatureSlowBTMem]>;
-def : Proc<"x86-64",          [FeatureSSE2, Feature64Bit, FeatureSlowBTMem]>;
-def : Proc<"yonah",           [FeatureSSE3, FeatureSlowBTMem]>;
-def : Proc<"prescott",        [FeatureSSE3, FeatureSlowBTMem]>;
-def : Proc<"nocona",          [FeatureSSE3, FeatureCMPXCHG16B,
-                               FeatureSlowBTMem]>;
-def : Proc<"core2",           [FeatureSSSE3, FeatureCMPXCHG16B,
-                               FeatureSlowBTMem]>;
-def : Proc<"penryn",          [FeatureSSE41, FeatureCMPXCHG16B,
-                               FeatureSlowBTMem]>;
-def : AtomProc<"atom",        [ProcIntelAtom, FeatureSSSE3, FeatureCMPXCHG16B,
-                               FeatureMOVBE, FeatureSlowBTMem, FeatureLeaForSP,
-                               FeatureSlowDivide]>;
+def : Proc<"x86-64",          [FeatureSSE2, Feature64Bit, FeatureSlowBTMem,
+                               FeatureFastUAMem]>;
+// Intel Core Duo.
+def : ProcessorModel<"yonah", SandyBridgeModel,
+                     [FeatureSSE3, FeatureSlowBTMem]>;
+
+// NetBurst.
+def : Proc<"prescott", [FeatureSSE3, FeatureSlowBTMem]>;
+def : Proc<"nocona",   [FeatureSSE3, FeatureCMPXCHG16B, FeatureSlowBTMem]>;
+
+// Intel Core 2 Solo/Duo.
+def : ProcessorModel<"core2", SandyBridgeModel,
+                     [FeatureSSSE3, FeatureCMPXCHG16B, FeatureSlowBTMem]>;
+def : ProcessorModel<"penryn", SandyBridgeModel,
+                     [FeatureSSE41, FeatureCMPXCHG16B, FeatureSlowBTMem]>;
+
+// Atom.
+def : ProcessorModel<"atom", AtomModel,
+                     [ProcIntelAtom, FeatureSSSE3, FeatureCMPXCHG16B,
+                      FeatureMOVBE, FeatureSlowBTMem, FeatureLeaForSP,
+                      FeatureSlowDivide,
+                      FeatureCallRegIndirect,
+                      FeaturePadShortFunctions]>;
+
 // "Arrandale" along with corei3 and corei5
-def : Proc<"corei7",          [FeatureSSE42, FeatureCMPXCHG16B,
-                               FeatureSlowBTMem, FeatureFastUAMem,
-                               FeaturePOPCNT, FeatureAES]>;
-def : Proc<"nehalem",         [FeatureSSE42,  FeatureCMPXCHG16B,
-                               FeatureSlowBTMem, FeatureFastUAMem,
-                               FeaturePOPCNT]>;
+def : ProcessorModel<"corei7", SandyBridgeModel,
+                     [FeatureSSE42, FeatureCMPXCHG16B, FeatureSlowBTMem,
+                      FeatureFastUAMem, FeaturePOPCNT, FeatureAES]>;
+
+def : ProcessorModel<"nehalem", SandyBridgeModel,
+                     [FeatureSSE42,  FeatureCMPXCHG16B, FeatureSlowBTMem,
+                      FeatureFastUAMem, FeaturePOPCNT]>;
 // Westmere is a similar machine to nehalem with some additional features.
 // Westmere is the corei3/i5/i7 path from nehalem to sandybridge
-def : Proc<"westmere",        [FeatureSSE42, FeatureCMPXCHG16B,
-                               FeatureSlowBTMem, FeatureFastUAMem,
-                               FeaturePOPCNT, FeatureAES, FeaturePCLMUL]>;
+def : ProcessorModel<"westmere", SandyBridgeModel,
+                     [FeatureSSE42, FeatureCMPXCHG16B, FeatureSlowBTMem,
+                      FeatureFastUAMem, FeaturePOPCNT, FeatureAES,
+                      FeaturePCLMUL]>;
 // Sandy Bridge
 // SSE is not listed here since llvm treats AVX as a reimplementation of SSE,
 // rather than a superset.
-def : Proc<"corei7-avx",      [FeatureAVX, FeatureCMPXCHG16B, FeaturePOPCNT,
-                               FeatureAES, FeaturePCLMUL]>;
+def : ProcessorModel<"corei7-avx", SandyBridgeModel,
+                     [FeatureAVX, FeatureCMPXCHG16B, FeatureFastUAMem,
+                      FeaturePOPCNT, FeatureAES, FeaturePCLMUL]>;
 // Ivy Bridge
-def : Proc<"core-avx-i",      [FeatureAVX, FeatureCMPXCHG16B, FeaturePOPCNT,
-                               FeatureAES, FeaturePCLMUL,
-                               FeatureRDRAND, FeatureF16C, FeatureFSGSBase]>;
+def : ProcessorModel<"core-avx-i", SandyBridgeModel,
+                     [FeatureAVX, FeatureCMPXCHG16B, FeatureFastUAMem,
+                      FeaturePOPCNT, FeatureAES, FeaturePCLMUL, FeatureRDRAND,
+                      FeatureF16C, FeatureFSGSBase]>;
 
 // Haswell
-def : Proc<"core-avx2",       [FeatureAVX2, FeatureCMPXCHG16B, FeaturePOPCNT,
-                               FeatureAES, FeaturePCLMUL, FeatureRDRAND,
-                               FeatureF16C, FeatureFSGSBase,
-                               FeatureMOVBE, FeatureLZCNT, FeatureBMI,
-                               FeatureBMI2, FeatureFMA,
-                               FeatureRTM]>;
+def : ProcessorModel<"core-avx2", HaswellModel,
+                     [FeatureAVX2, FeatureCMPXCHG16B, FeatureFastUAMem,
+                      FeaturePOPCNT, FeatureAES, FeaturePCLMUL, FeatureRDRAND,
+                      FeatureF16C, FeatureFSGSBase, FeatureMOVBE, FeatureLZCNT,
+                      FeatureBMI, FeatureBMI2, FeatureFMA, FeatureRTM,
+                      FeatureHLE]>;
 
 def : Proc<"k6",              [FeatureMMX]>;
 def : Proc<"k6-2",            [Feature3DNow]>;
diff --git a/lib/Target/X86/X86AsmPrinter.cpp b/lib/Target/X86/X86AsmPrinter.cpp
index fdd712520b44..6b228b0b0329 100644
--- a/lib/Target/X86/X86AsmPrinter.cpp
+++ b/lib/Target/X86/X86AsmPrinter.cpp
@@ -13,33 +13,33 @@
 //===----------------------------------------------------------------------===//
 
 #include "X86AsmPrinter.h"
+#include "InstPrinter/X86ATTInstPrinter.h"
 #include "X86.h"
 #include "X86COFFMachineModuleInfo.h"
 #include "X86MachineFunctionInfo.h"
 #include "X86TargetMachine.h"
-#include "InstPrinter/X86ATTInstPrinter.h"
-#include "llvm/CallingConv.h"
-#include "llvm/DebugInfo.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Module.h"
-#include "llvm/Type.h"
+#include "llvm/ADT/SmallString.h"
 #include "llvm/Assembly/Writer.h"
+#include "llvm/CodeGen/MachineJumpTableInfo.h"
+#include "llvm/CodeGen/MachineModuleInfoImpls.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/DebugInfo.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Type.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCSectionMachO.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/CodeGen/MachineJumpTableInfo.h"
-#include "llvm/CodeGen/MachineModuleInfoImpls.h"
-#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
-#include "llvm/Target/Mangler.h"
-#include "llvm/Target/TargetOptions.h"
 #include "llvm/Support/COFF.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
-#include "llvm/ADT/SmallString.h"
+#include "llvm/Target/Mangler.h"
+#include "llvm/Target/TargetOptions.h"
 using namespace llvm;
 
 //===----------------------------------------------------------------------===//
@@ -201,7 +201,7 @@ void X86AsmPrinter::printSymbolOperand(const MachineOperand &MO,
   case X86II::MO_TLVP_PIC_BASE:
     O << "@TLVP" << '-' << *MF->getPICBaseSymbol();
     break;
-  case X86II::MO_SECREL:      O << "@SECREL";      break;
+  case X86II::MO_SECREL:    O << "@SECREL32";  break;
   }
 }
 
@@ -252,14 +252,15 @@ void X86AsmPrinter::printOperand(const MachineInstr *MI, unsigned OpNo,
   }
 
   case MachineOperand::MO_Immediate:
-    O << '$' << MO.getImm();
+    if (AsmVariant == 0) O << '$';
+    O << MO.getImm();
     return;
 
   case MachineOperand::MO_JumpTableIndex:
   case MachineOperand::MO_ConstantPoolIndex:
   case MachineOperand::MO_GlobalAddress:
   case MachineOperand::MO_ExternalSymbol: {
-    O << '$';
+    if (AsmVariant == 0) O << '$';
     printSymbolOperand(MO, O);
     break;
   }
@@ -355,19 +356,23 @@ void X86AsmPrinter::printIntelMemReference(const MachineInstr *MI, unsigned Op,
     NeedPlus = true;
   }
 
-  assert (DispSpec.isImm() && "Displacement is not an immediate!");
-  int64_t DispVal = DispSpec.getImm();
-  if (DispVal || (!IndexReg.getReg() && !BaseReg.getReg())) {
-    if (NeedPlus) {
-      if (DispVal > 0)
-        O << " + ";
-      else {
-        O << " - ";
-        DispVal = -DispVal;
+  if (!DispSpec.isImm()) {
+    if (NeedPlus) O << " + ";
+    printOperand(MI, Op+3, O, Modifier, AsmVariant);
+  } else {
+    int64_t DispVal = DispSpec.getImm();
+    if (DispVal || (!IndexReg.getReg() && !BaseReg.getReg())) {
+      if (NeedPlus) {
+        if (DispVal > 0)
+          O << " + ";
+        else {
+          O << " - ";
+          DispVal = -DispVal;
+        }
       }
+      O << DispVal;
     }
-    O << DispVal;
-  }  
+  }
   O << ']';
 }
 
@@ -543,7 +548,7 @@ void X86AsmPrinter::EmitEndOfAsmFile(Module &M) {
                                         MCSA_IndirectSymbol);
         // hlt; hlt; hlt; hlt; hlt     hlt = 0xf4.
         const char HltInsts[] = "\xf4\xf4\xf4\xf4\xf4";
-        OutStreamer.EmitBytes(StringRef(HltInsts, 5), 0/*addrspace*/);
+        OutStreamer.EmitBytes(StringRef(HltInsts, 5));
       }
 
       Stubs.clear();
@@ -569,7 +574,7 @@ void X86AsmPrinter::EmitEndOfAsmFile(Module &M) {
         // .long 0
         if (MCSym.getInt())
           // External to current translation unit.
-          OutStreamer.EmitIntValue(0, 4/*size*/, 0/*addrspace*/);
+          OutStreamer.EmitIntValue(0, 4/*size*/);
         else
           // Internal to current translation unit.
           //
@@ -578,8 +583,7 @@ void X86AsmPrinter::EmitEndOfAsmFile(Module &M) {
           // using NLPs.  However, sometimes the types are local to the file. So
           // we need to fill in the value for the NLP in those cases.
           OutStreamer.EmitValue(MCSymbolRefExpr::Create(MCSym.getPointer(),
-                                                        OutContext),
-                                4/*size*/, 0/*addrspace*/);
+                                                        OutContext), 4/*size*/);
       }
       Stubs.clear();
       OutStreamer.AddBlankLine();
@@ -596,8 +600,7 @@ void X86AsmPrinter::EmitEndOfAsmFile(Module &M) {
         // .long _foo
         OutStreamer.EmitValue(MCSymbolRefExpr::
                               Create(Stubs[i].second.getPointer(),
-                                     OutContext),
-                              4/*size*/, 0/*addrspace*/);
+                                     OutContext), 4/*size*/);
       }
       Stubs.clear();
       OutStreamer.AddBlankLine();
@@ -663,7 +666,7 @@ void X86AsmPrinter::EmitEndOfAsmFile(Module &M) {
           name += ",DATA";
         else
         name += ",data";
-        OutStreamer.EmitBytes(name, 0);
+        OutStreamer.EmitBytes(name);
       }
 
       for (unsigned i = 0, e = DLLExportedFns.size(); i != e; ++i) {
@@ -672,7 +675,7 @@ void X86AsmPrinter::EmitEndOfAsmFile(Module &M) {
         else
           name = " -export:";
         name += DLLExportedFns[i]->getName();
-        OutStreamer.EmitBytes(name, 0);
+        OutStreamer.EmitBytes(name);
       }
     }
   }
@@ -692,7 +695,7 @@ void X86AsmPrinter::EmitEndOfAsmFile(Module &M) {
       for (unsigned i = 0, e = Stubs.size(); i != e; ++i) {
         OutStreamer.EmitLabel(Stubs[i].first);
         OutStreamer.EmitSymbolValue(Stubs[i].second.getPointer(),
-                                    TD->getPointerSize(), 0);
+                                    TD->getPointerSize());
       }
       Stubs.clear();
     }
diff --git a/lib/Target/X86/X86AsmPrinter.h b/lib/Target/X86/X86AsmPrinter.h
index 61eb14e036d0..bc7496bad144 100644
--- a/lib/Target/X86/X86AsmPrinter.h
+++ b/lib/Target/X86/X86AsmPrinter.h
@@ -1,4 +1,4 @@
-//===-- X86AsmPrinter.h - Convert X86 LLVM code to assembly -----*- C++ -*-===//
+//===-- X86AsmPrinter.h - X86 implementation of AsmPrinter ------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -6,10 +6,6 @@
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
-//
-// AT&T assembly code printer class.
-//
-//===----------------------------------------------------------------------===//
 
 #ifndef X86ASMPRINTER_H
 #define X86ASMPRINTER_H
@@ -35,7 +31,7 @@ class LLVM_LIBRARY_VISIBILITY X86AsmPrinter : public AsmPrinter {
   }
 
   virtual const char *getPassName() const LLVM_OVERRIDE {
-    return "X86 AT&T-Style Assembly Printer";
+    return "X86 Assembly / Object Emitter";
   }
 
   const X86Subtarget &getSubtarget() const { return *Subtarget; }
diff --git a/lib/Target/X86/X86COFFMachineModuleInfo.h b/lib/Target/X86/X86COFFMachineModuleInfo.h
index a5a8dc18e41d..0dfeb42f1a4d 100644
--- a/lib/Target/X86/X86COFFMachineModuleInfo.h
+++ b/lib/Target/X86/X86COFFMachineModuleInfo.h
@@ -15,8 +15,8 @@
 #define X86COFF_MACHINEMODULEINFO_H
 
 #include "X86MachineFunctionInfo.h"
-#include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/ADT/DenseSet.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
 
 namespace llvm {
   class X86MachineFunctionInfo;
diff --git a/lib/Target/X86/X86CallingConv.td b/lib/Target/X86/X86CallingConv.td
index 6786756c7faf..9eafbd55a5ae 100644
--- a/lib/Target/X86/X86CallingConv.td
+++ b/lib/Target/X86/X86CallingConv.td
@@ -103,6 +103,15 @@ def RetCC_Intel_OCL_BI : CallingConv<[
   CCDelegateTo<RetCC_X86Common>
 ]>;
 
+// X86-32 HiPE return-value convention.
+def RetCC_X86_32_HiPE : CallingConv<[
+  // Promote all types to i32
+  CCIfType<[i8, i16], CCPromoteToType<i32>>,
+
+  // Return: HP, P, VAL1, VAL2
+  CCIfType<[i32], CCAssignToReg<[ESI, EBP, EAX, EDX]>>
+]>;
+
 // X86-64 C return-value convention.
 def RetCC_X86_64_C : CallingConv<[
   // The X86-64 calling convention always returns FP values in XMM0.
@@ -123,17 +132,30 @@ def RetCC_X86_Win64_C : CallingConv<[
   CCDelegateTo<RetCC_X86_64_C>
 ]>;
 
+// X86-64 HiPE return-value convention.
+def RetCC_X86_64_HiPE : CallingConv<[
+  // Promote all types to i64
+  CCIfType<[i8, i16, i32], CCPromoteToType<i64>>,
+
+  // Return: HP, P, VAL1, VAL2
+  CCIfType<[i64], CCAssignToReg<[R15, RBP, RAX, RDX]>>
+]>;
 
 // This is the root return-value convention for the X86-32 backend.
 def RetCC_X86_32 : CallingConv<[
   // If FastCC, use RetCC_X86_32_Fast.
   CCIfCC<"CallingConv::Fast", CCDelegateTo<RetCC_X86_32_Fast>>,
+  // If HiPE, use RetCC_X86_32_HiPE.
+  CCIfCC<"CallingConv::HiPE", CCDelegateTo<RetCC_X86_32_HiPE>>,
+
   // Otherwise, use RetCC_X86_32_C.
   CCDelegateTo<RetCC_X86_32_C>
 ]>;
 
 // This is the root return-value convention for the X86-64 backend.
 def RetCC_X86_64 : CallingConv<[
+  // HiPE uses RetCC_X86_64_HiPE
+  CCIfCC<"CallingConv::HiPE", CCDelegateTo<RetCC_X86_64_HiPE>>,
   // Mingw64 and native Win64 use Win64 CC
   CCIfSubtarget<"isTargetWin64()", CCDelegateTo<RetCC_X86_Win64_C>>,
 
@@ -254,29 +276,6 @@ def CC_X86_Win64_C : CallingConv<[
   CCIfType<[f80], CCAssignToStack<0, 0>>
 ]>;
 
-// X86-64 Intel OpenCL built-ins calling convention.
-def CC_Intel_OCL_BI : CallingConv<[
-  CCIfType<[i32], CCIfSubtarget<"isTargetWin32()", CCAssignToStack<4, 4>>>,
-
-  CCIfType<[i32], CCIfSubtarget<"isTargetWin64()", CCAssignToReg<[ECX, EDX, R8D, R9D]>>>,
-  CCIfType<[i64], CCIfSubtarget<"isTargetWin64()", CCAssignToReg<[RCX, RDX, R8,  R9 ]>>>,
-
-  CCIfType<[i32], CCAssignToReg<[EDI, ESI, EDX, ECX]>>,
-  CCIfType<[i64], CCAssignToReg<[RDI, RSI, RDX, RCX]>>,
-
- // The SSE vector arguments are passed in XMM registers.
-  CCIfType<[f32, f64, v4i32, v2i64, v4f32, v2f64],
-           CCAssignToReg<[XMM0, XMM1, XMM2, XMM3]>>,
-  
-  // The 256-bit vector arguments are passed in YMM registers.
-  CCIfType<[v8f32, v4f64, v8i32, v4i64],
-                CCAssignToReg<[YMM0, YMM1, YMM2, YMM3]>>,
-  
-  CCIfSubtarget<"isTargetWin64()", CCDelegateTo<CC_X86_Win64_C>>,
-  CCDelegateTo<CC_X86_64_C>
-]>;
-
-
 def CC_X86_64_GHC : CallingConv<[
   // Promote i8/i16/i32 arguments to i64.
   CCIfType<[i8, i16, i32], CCPromoteToType<i64>>,
@@ -291,6 +290,18 @@ def CC_X86_64_GHC : CallingConv<[
             CCAssignToReg<[XMM1, XMM2, XMM3, XMM4, XMM5, XMM6]>>>
 ]>;
 
+def CC_X86_64_HiPE : CallingConv<[
+  // Promote i8/i16/i32 arguments to i64.
+  CCIfType<[i8, i16, i32], CCPromoteToType<i64>>,
+
+  // Pass in VM's registers: HP, P, ARG0, ARG1, ARG2, ARG3
+  CCIfType<[i64], CCAssignToReg<[R15, RBP, RSI, RDX, RCX, R8]>>,
+
+  // Integer/FP values get stored in stack slots that are 8 bytes in size and
+  // 8-byte aligned if there are no more registers to hold them.
+  CCIfType<[i32, i64, f32, f64], CCAssignToStack<8, 8>>
+]>;
+
 //===----------------------------------------------------------------------===//
 // X86 C Calling Convention
 //===----------------------------------------------------------------------===//
@@ -376,8 +387,8 @@ def CC_X86_32_ThisCall : CallingConv<[
   // Promote i8/i16 arguments to i32.
   CCIfType<[i8, i16], CCPromoteToType<i32>>,
 
-  // Pass sret arguments indirectly through EAX
-  CCIfSRet<CCAssignToReg<[EAX]>>,
+  // Pass sret arguments indirectly through stack.
+  CCIfSRet<CCAssignToStack<4, 4>>,
 
   // The first integer argument is passed in ECX
   CCIfType<[i32], CCAssignToReg<[ECX]>>,
@@ -422,6 +433,42 @@ def CC_X86_32_GHC : CallingConv<[
   CCIfType<[i32], CCAssignToReg<[EBX, EBP, EDI, ESI]>>
 ]>;
 
+def CC_X86_32_HiPE : CallingConv<[
+  // Promote i8/i16 arguments to i32.
+  CCIfType<[i8, i16], CCPromoteToType<i32>>,
+
+  // Pass in VM's registers: HP, P, ARG0, ARG1, ARG2
+  CCIfType<[i32], CCAssignToReg<[ESI, EBP, EAX, EDX, ECX]>>,
+
+  // Integer/Float values get stored in stack slots that are 4 bytes in
+  // size and 4-byte aligned.
+  CCIfType<[i32, f32], CCAssignToStack<4, 4>>
+]>;
+
+// X86-64 Intel OpenCL built-ins calling convention.
+def CC_Intel_OCL_BI : CallingConv<[
+
+  CCIfType<[i32], CCIfSubtarget<"isTargetWin64()", CCAssignToReg<[ECX, EDX, R8D, R9D]>>>,
+  CCIfType<[i64], CCIfSubtarget<"isTargetWin64()", CCAssignToReg<[RCX, RDX, R8,  R9 ]>>>,
+
+  CCIfType<[i32], CCIfSubtarget<"is64Bit()", CCAssignToReg<[EDI, ESI, EDX, ECX]>>>,
+  CCIfType<[i64], CCIfSubtarget<"is64Bit()", CCAssignToReg<[RDI, RSI, RDX, RCX]>>>,
+
+  CCIfType<[i32], CCAssignToStack<4, 4>>,
+
+  // The SSE vector arguments are passed in XMM registers.
+  CCIfType<[f32, f64, v4i32, v2i64, v4f32, v2f64],
+           CCAssignToReg<[XMM0, XMM1, XMM2, XMM3]>>,
+
+  // The 256-bit vector arguments are passed in YMM registers.
+  CCIfType<[v8f32, v4f64, v8i32, v4i64],
+           CCAssignToReg<[YMM0, YMM1, YMM2, YMM3]>>,
+
+  CCIfSubtarget<"isTargetWin64()", CCDelegateTo<CC_X86_Win64_C>>,
+  CCIfSubtarget<"is64Bit()",       CCDelegateTo<CC_X86_64_C>>,
+  CCDelegateTo<CC_X86_32_C>
+]>;
+
 //===----------------------------------------------------------------------===//
 // X86 Root Argument Calling Conventions
 //===----------------------------------------------------------------------===//
@@ -432,6 +479,7 @@ def CC_X86_32 : CallingConv<[
   CCIfCC<"CallingConv::X86_ThisCall", CCDelegateTo<CC_X86_32_ThisCall>>,
   CCIfCC<"CallingConv::Fast", CCDelegateTo<CC_X86_32_FastCC>>,
   CCIfCC<"CallingConv::GHC", CCDelegateTo<CC_X86_32_GHC>>,
+  CCIfCC<"CallingConv::HiPE", CCDelegateTo<CC_X86_32_HiPE>>,
 
   // Otherwise, drop to normal X86-32 CC
   CCDelegateTo<CC_X86_32_C>
@@ -440,6 +488,7 @@ def CC_X86_32 : CallingConv<[
 // This is the root argument convention for the X86-64 backend.
 def CC_X86_64 : CallingConv<[
   CCIfCC<"CallingConv::GHC", CCDelegateTo<CC_X86_64_GHC>>,
+  CCIfCC<"CallingConv::HiPE", CCDelegateTo<CC_X86_64_HiPE>>,
 
   // Mingw64 and native Win64 use Win64 CC
   CCIfSubtarget<"isTargetWin64()", CCDelegateTo<CC_X86_Win64_C>>,
@@ -470,6 +519,9 @@ def CSR_64EHRet : CalleeSavedRegs<(add RAX, RDX, CSR_64)>;
 def CSR_Win64 : CalleeSavedRegs<(add RBX, RBP, RDI, RSI, R12, R13, R14, R15,
                                      (sequence "XMM%u", 6, 15))>;
 
+def CSR_MostRegs_64 : CalleeSavedRegs<(add RBX, RCX, RDX, RSI, RDI, R8, R9, R10,
+                                           R11, R12, R13, R14, R15, RBP,
+                                           (sequence "XMM%u", 0, 15))>;
 
 // Standard C + YMM6-15
 def CSR_Win64_Intel_OCL_BI_AVX : CalleeSavedRegs<(add RBX, RBP, RDI, RSI, R12,
diff --git a/lib/Target/X86/X86CodeEmitter.cpp b/lib/Target/X86/X86CodeEmitter.cpp
index 44db563818b1..2518e02e2a40 100644
--- a/lib/Target/X86/X86CodeEmitter.cpp
+++ b/lib/Target/X86/X86CodeEmitter.cpp
@@ -13,23 +13,23 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "x86-emitter"
+#include "X86.h"
 #include "X86InstrInfo.h"
 #include "X86JITInfo.h"
+#include "X86Relocations.h"
 #include "X86Subtarget.h"
 #include "X86TargetMachine.h"
-#include "X86Relocations.h"
-#include "X86.h"
-#include "llvm/LLVMContext.h"
-#include "llvm/PassManager.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/JITCodeEmitter.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/Passes.h"
-#include "llvm/ADT/Statistic.h"
+#include "llvm/IR/LLVMContext.h"
 #include "llvm/MC/MCCodeEmitter.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
@@ -124,7 +124,7 @@ template<class CodeEmitter>
 } // end anonymous namespace.
 
 /// createX86CodeEmitterPass - Return a pass that emits the collected X86 code
-/// to the specified templated MachineCodeEmitter object.
+/// to the specified JITCodeEmitter object.
 FunctionPass *llvm::createX86JITCodeEmitterPass(X86TargetMachine &TM,
                                                 JITCodeEmitter &JCE) {
   return new Emitter<JITCodeEmitter>(TM, JCE);
@@ -816,6 +816,7 @@ void Emitter<CodeEmitter>::emitVEXOpcodePrefix(uint64_t TSFlags,
                                                const MCInstrDesc *Desc) const {
   bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V;
   bool HasVEX_4VOp3 = (TSFlags >> X86II::VEXShift) & X86II::VEX_4VOp3;
+  bool HasMemOp4 = (TSFlags >> X86II::VEXShift) & X86II::MemOp4;
 
   // VEX_R: opcode externsion equivalent to REX.R in
   // 1's complement (inverted) form
@@ -1032,6 +1033,10 @@ void Emitter<CodeEmitter>::emitVEXOpcodePrefix(uint64_t TSFlags,
 
       if (HasVEX_4V)
         VEX_4V = getVEXRegisterEncoding(MI, CurOp++);
+
+      if (HasMemOp4) // Skip second register source (encoded in I8IMM)
+        CurOp++;
+
       if (X86II::isX86_64ExtendedReg(MI.getOperand(CurOp).getReg()))
         VEX_B = 0x0;
       CurOp++;
@@ -1042,9 +1047,15 @@ void Emitter<CodeEmitter>::emitVEXOpcodePrefix(uint64_t TSFlags,
       // MRMDestReg instructions forms:
       //  dst(ModR/M), src(ModR/M)
       //  dst(ModR/M), src(ModR/M), imm8
-      if (X86II::isX86_64ExtendedReg(MI.getOperand(0).getReg()))
+      //  dst(ModR/M), src1(VEX_4V), src2(ModR/M)
+      if (X86II::isX86_64ExtendedReg(MI.getOperand(CurOp).getReg()))
         VEX_B = 0x0;
-      if (X86II::isX86_64ExtendedReg(MI.getOperand(1).getReg()))
+      CurOp++;
+
+      if (HasVEX_4V)
+        VEX_4V = getVEXRegisterEncoding(MI, CurOp++);
+
+      if (X86II::isX86_64ExtendedReg(MI.getOperand(CurOp).getReg()))
         VEX_R = 0x0;
       break;
     case X86II::MRM0r: case X86II::MRM1r:
@@ -1279,9 +1290,14 @@ void Emitter<CodeEmitter>::emitInstruction(MachineInstr &MI,
 
   case X86II::MRMDestReg: {
     MCE.emitByte(BaseOpcode);
+
+    unsigned SrcRegNum = CurOp+1;
+    if (HasVEX_4V) // Skip 1st src (which is encoded in VEX_VVVV)
+      SrcRegNum++;
+
     emitRegModRMByte(MI.getOperand(CurOp).getReg(),
-                     getX86RegNum(MI.getOperand(CurOp+1).getReg()));
-    CurOp += 2;
+                     getX86RegNum(MI.getOperand(SrcRegNum).getReg()));
+    CurOp = SrcRegNum + 1;
     break;
   }
   case X86II::MRMDestMem: {
diff --git a/lib/Target/X86/X86CompilationCallback_Win64.asm b/lib/Target/X86/X86CompilationCallback_Win64.asm
index f321778db24b..69b4c71651d7 100644
--- a/lib/Target/X86/X86CompilationCallback_Win64.asm
+++ b/lib/Target/X86/X86CompilationCallback_Win64.asm
@@ -11,7 +11,7 @@
 ;;
 ;;===----------------------------------------------------------------------===
 
-extrn X86CompilationCallback2: PROC
+extrn LLVMX86CompilationCallback2: PROC
 
 .code
 X86CompilationCallback proc
@@ -42,7 +42,7 @@ X86CompilationCallback proc
     ; Pass prev frame and return address.
     mov     rcx, rbp
     mov     rdx, qword ptr [rbp+8]
-    call    X86CompilationCallback2
+    call    LLVMX86CompilationCallback2
 
     ; Restore all XMM arg registers.
     movaps  xmm3, [rsp+48+32]
diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp
index d4627c74cb1c..cadec682a435 100644
--- a/lib/Target/X86/X86FastISel.cpp
+++ b/lib/Target/X86/X86FastISel.cpp
@@ -14,24 +14,24 @@
 //===----------------------------------------------------------------------===//
 
 #include "X86.h"
-#include "X86InstrBuilder.h"
 #include "X86ISelLowering.h"
+#include "X86InstrBuilder.h"
 #include "X86RegisterInfo.h"
 #include "X86Subtarget.h"
 #include "X86TargetMachine.h"
-#include "llvm/CallingConv.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/GlobalAlias.h"
-#include "llvm/Instructions.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/Operator.h"
 #include "llvm/CodeGen/Analysis.h"
 #include "llvm/CodeGen/FastISel.h"
 #include "llvm/CodeGen/FunctionLoweringInfo.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Operator.h"
 #include "llvm/Support/CallSite.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/GetElementPtrTypeIterator.h"
@@ -75,6 +75,8 @@ public:
   virtual bool TryToFoldLoad(MachineInstr *MI, unsigned OpNo,
                              const LoadInst *LI);
 
+  virtual bool FastLowerArguments();
+
 #include "X86GenFastISel.inc"
 
 private:
@@ -297,7 +299,7 @@ bool X86FastISel::X86FastEmitStore(EVT VT, const Value *Val,
     case MVT::i32: Opc = X86::MOV32mi; break;
     case MVT::i64:
       // Must be a 32-bit sign extended value.
-      if ((int)CI->getSExtValue() == CI->getSExtValue())
+      if (isInt<32>(CI->getSExtValue()))
         Opc = X86::MOV64mi32;
       break;
     }
@@ -326,12 +328,11 @@ bool X86FastISel::X86FastEmitExtend(ISD::NodeType Opc, EVT DstVT,
                                     unsigned &ResultReg) {
   unsigned RR = FastEmit_r(SrcVT.getSimpleVT(), DstVT.getSimpleVT(), Opc,
                            Src, /*TODO: Kill=*/false);
-
-  if (RR != 0) {
-    ResultReg = RR;
-    return true;
-  } else
+  if (RR == 0)
     return false;
+
+  ResultReg = RR;
+  return true;
 }
 
 /// X86SelectAddress - Attempt to fill in an address from the given value.
@@ -727,7 +728,7 @@ bool X86FastISel::X86SelectRet(const Instruction *I) {
 
   // Don't handle popping bytes on return for now.
   if (X86MFInfo->getBytesToPopOnReturn() != 0)
-    return 0;
+    return false;
 
   // fastcc with -tailcallopt is intended to provide a guaranteed
   // tail call optimization. Fastisel doesn't know how to do that.
@@ -738,10 +739,12 @@ bool X86FastISel::X86SelectRet(const Instruction *I) {
   if (F.isVarArg())
     return false;
 
+  // Build a list of return value registers.
+  SmallVector<unsigned, 4> RetRegs;
+
   if (Ret->getNumOperands() > 0) {
     SmallVector<ISD::OutputArg, 4> Outs;
-    GetReturnInfo(F.getReturnType(), F.getAttributes().getRetAttributes(),
-                  Outs, TLI);
+    GetReturnInfo(F.getReturnType(), F.getAttributes(), Outs, TLI);
 
     // Analyze operands of the call, assigning locations to each operand.
     SmallVector<CCValAssign, 16> ValLocs;
@@ -806,25 +809,30 @@ bool X86FastISel::X86SelectRet(const Instruction *I) {
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
             DstReg).addReg(SrcReg);
 
-    // Mark the register as live out of the function.
-    MRI.addLiveOut(VA.getLocReg());
+    // Add register to return instruction.
+    RetRegs.push_back(VA.getLocReg());
   }
 
   // The x86-64 ABI for returning structs by value requires that we copy
   // the sret argument into %rax for the return. We saved the argument into
   // a virtual register in the entry block, so now we copy the value out
-  // and into %rax.
-  if (Subtarget->is64Bit() && F.hasStructRetAttr()) {
+  // and into %rax. We also do the same with %eax for Win32.
+  if (F.hasStructRetAttr() &&
+      (Subtarget->is64Bit() || Subtarget->isTargetWindows())) {
     unsigned Reg = X86MFInfo->getSRetReturnReg();
     assert(Reg &&
            "SRetReturnReg should have been set in LowerFormalArguments()!");
+    unsigned RetReg = Subtarget->is64Bit() ? X86::RAX : X86::EAX;
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
-            X86::RAX).addReg(Reg);
-    MRI.addLiveOut(X86::RAX);
+            RetReg).addReg(Reg);
+    RetRegs.push_back(RetReg);
   }
 
   // Now emit the RET.
-  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::RET));
+  MachineInstrBuilder MIB =
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(X86::RET));
+  for (unsigned i = 0, e = RetRegs.size(); i != e; ++i)
+    MIB.addReg(RetRegs[i], RegState::Implicit);
   return true;
 }
 
@@ -1373,7 +1381,6 @@ bool X86FastISel::TryEmitSmallMemcpy(X86AddressMode DestAM,
     else if (Len >= 2)
       VT = MVT::i16;
     else {
-      assert(Len == 1);
       VT = MVT::i8;
     }
 
@@ -1517,6 +1524,81 @@ bool X86FastISel::X86VisitIntrinsicCall(const IntrinsicInst &I) {
   }
 }
 
+bool X86FastISel::FastLowerArguments() {
+  if (!FuncInfo.CanLowerReturn)
+    return false;
+
+  if (Subtarget->isTargetWin64())
+    return false;
+
+  const Function *F = FuncInfo.Fn;
+  if (F->isVarArg())
+    return false;
+
+  CallingConv::ID CC = F->getCallingConv();
+  if (CC != CallingConv::C)
+    return false;
+  
+  if (!Subtarget->is64Bit())
+    return false;
+  
+  // Only handle simple cases. i.e. Up to 6 i32/i64 scalar arguments.
+  unsigned Idx = 1;
+  for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
+       I != E; ++I, ++Idx) {
+    if (Idx > 6)
+      return false;
+
+    if (F->getAttributes().hasAttribute(Idx, Attribute::ByVal) ||
+        F->getAttributes().hasAttribute(Idx, Attribute::InReg) ||
+        F->getAttributes().hasAttribute(Idx, Attribute::StructRet) ||
+        F->getAttributes().hasAttribute(Idx, Attribute::Nest))
+      return false;
+
+    Type *ArgTy = I->getType();
+    if (ArgTy->isStructTy() || ArgTy->isArrayTy() || ArgTy->isVectorTy())
+      return false;
+
+    EVT ArgVT = TLI.getValueType(ArgTy);
+    if (!ArgVT.isSimple()) return false;
+    switch (ArgVT.getSimpleVT().SimpleTy) {
+    case MVT::i32:
+    case MVT::i64:
+      break;
+    default:
+      return false;
+    }
+  }
+
+  static const uint16_t GPR32ArgRegs[] = {
+    X86::EDI, X86::ESI, X86::EDX, X86::ECX, X86::R8D, X86::R9D
+  };
+  static const uint16_t GPR64ArgRegs[] = {
+    X86::RDI, X86::RSI, X86::RDX, X86::RCX, X86::R8 , X86::R9
+  };
+
+  Idx = 0;
+  const TargetRegisterClass *RC32 = TLI.getRegClassFor(MVT::i32);
+  const TargetRegisterClass *RC64 = TLI.getRegClassFor(MVT::i64);
+  for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
+       I != E; ++I, ++Idx) {
+    if (I->use_empty())
+      continue;
+    bool is32Bit = TLI.getValueType(I->getType()) == MVT::i32;
+    const TargetRegisterClass *RC = is32Bit ? RC32 : RC64;
+    unsigned SrcReg = is32Bit ? GPR32ArgRegs[Idx] : GPR64ArgRegs[Idx];
+    unsigned DstReg = FuncInfo.MF->addLiveIn(SrcReg, RC);
+    // FIXME: Unfortunately it's necessary to emit a copy from the livein copy.
+    // Without this, EmitLiveInCopies may eliminate the livein if its only
+    // use is a bitcast (which isn't turned into an instruction).
+    unsigned ResultReg = createResultReg(RC);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
+            ResultReg).addReg(DstReg, getKillRegState(true));
+    UpdateValueMap(I, ResultReg);
+  }
+  return true;
+}
+
 bool X86FastISel::X86SelectCall(const Instruction *I) {
   const CallInst *CI = cast<CallInst>(I);
   const Value *Callee = CI->getCalledValue();
@@ -1529,6 +1611,10 @@ bool X86FastISel::X86SelectCall(const Instruction *I) {
   if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI))
     return X86VisitIntrinsicCall(*II);
 
+  // Allow SelectionDAG isel to handle tail calls.
+  if (cast<CallInst>(I)->isTailCall())
+    return false;
+
   return DoSelectCall(I, 0);
 }
 
@@ -1541,9 +1627,9 @@ static unsigned computeBytesPoppedByCallee(const X86Subtarget &Subtarget,
   CallingConv::ID CC = CS.getCallingConv();
   if (CC == CallingConv::Fast || CC == CallingConv::GHC)
     return 0;
-  if (!CS.paramHasAttr(1, Attributes::StructRet))
+  if (!CS.paramHasAttr(1, Attribute::StructRet))
     return 0;
-  if (CS.paramHasAttr(1, Attributes::InReg))
+  if (CS.paramHasAttr(1, Attribute::InReg))
     return 0;
   return 4;
 }
@@ -1581,8 +1667,7 @@ bool X86FastISel::DoSelectCall(const Instruction *I, const char *MemIntName) {
 
   // Check whether the function can return without sret-demotion.
   SmallVector<ISD::OutputArg, 4> Outs;
-  GetReturnInfo(I->getType(), CS.getAttributes().getRetAttributes(),
-                Outs, TLI);
+  GetReturnInfo(I->getType(), CS.getAttributes(), Outs, TLI);
   bool CanLowerReturn = TLI.CanLowerReturn(CS.getCallingConv(),
                                            *FuncInfo.MF, FTy->isVarArg(),
                                            Outs, FTy->getContext());
@@ -1622,12 +1707,12 @@ bool X86FastISel::DoSelectCall(const Instruction *I, const char *MemIntName) {
     Value *ArgVal = *i;
     ISD::ArgFlagsTy Flags;
     unsigned AttrInd = i - CS.arg_begin() + 1;
-    if (CS.paramHasAttr(AttrInd, Attributes::SExt))
+    if (CS.paramHasAttr(AttrInd, Attribute::SExt))
       Flags.setSExt();
-    if (CS.paramHasAttr(AttrInd, Attributes::ZExt))
+    if (CS.paramHasAttr(AttrInd, Attribute::ZExt))
       Flags.setZExt();
 
-    if (CS.paramHasAttr(AttrInd, Attributes::ByVal)) {
+    if (CS.paramHasAttr(AttrInd, Attribute::ByVal)) {
       PointerType *Ty = cast<PointerType>(ArgVal->getType());
       Type *ElementTy = Ty->getElementType();
       unsigned FrameSize = TD.getTypeAllocSize(ElementTy);
@@ -1641,9 +1726,9 @@ bool X86FastISel::DoSelectCall(const Instruction *I, const char *MemIntName) {
         return false;
     }
 
-    if (CS.paramHasAttr(AttrInd, Attributes::InReg))
+    if (CS.paramHasAttr(AttrInd, Attribute::InReg))
       Flags.setInReg();
-    if (CS.paramHasAttr(AttrInd, Attributes::Nest))
+    if (CS.paramHasAttr(AttrInd, Attribute::Nest))
       Flags.setNest();
 
     // If this is an i1/i8/i16 argument, promote to i32 to avoid an extra
@@ -1905,17 +1990,17 @@ bool X86FastISel::DoSelectCall(const Instruction *I, const char *MemIntName) {
   ComputeValueVTs(TLI, I->getType(), RetTys);
   for (unsigned i = 0, e = RetTys.size(); i != e; ++i) {
     EVT VT = RetTys[i];
-    EVT RegisterVT = TLI.getRegisterType(I->getParent()->getContext(), VT);
+    MVT RegisterVT = TLI.getRegisterType(I->getParent()->getContext(), VT);
     unsigned NumRegs = TLI.getNumRegisters(I->getParent()->getContext(), VT);
     for (unsigned j = 0; j != NumRegs; ++j) {
       ISD::InputArg MyFlags;
-      MyFlags.VT = RegisterVT.getSimpleVT();
+      MyFlags.VT = RegisterVT;
       MyFlags.Used = !CS.getInstruction()->use_empty();
-      if (CS.paramHasAttr(0, Attributes::SExt))
+      if (CS.paramHasAttr(0, Attribute::SExt))
         MyFlags.Flags.setSExt();
-      if (CS.paramHasAttr(0, Attributes::ZExt))
+      if (CS.paramHasAttr(0, Attribute::ZExt))
         MyFlags.Flags.setZExt();
-      if (CS.paramHasAttr(0, Attributes::InReg))
+      if (CS.paramHasAttr(0, Attribute::InReg))
         MyFlags.Flags.setInReg();
       Ins.push_back(MyFlags);
     }
@@ -2154,13 +2239,13 @@ unsigned X86FastISel::TargetMaterializeAlloca(const AllocaInst *C) {
 unsigned X86FastISel::TargetMaterializeFloatZero(const ConstantFP *CF) {
   MVT VT;
   if (!isTypeLegal(CF->getType(), VT))
-    return false;
+    return 0;
 
   // Get opcode and regclass for the given zero.
   unsigned Opc = 0;
   const TargetRegisterClass *RC = NULL;
   switch (VT.SimpleTy) {
-  default: return false;
+  default: return 0;
   case MVT::f32:
     if (X86ScalarSSEf32) {
       Opc = X86::FsFLD0SS;
@@ -2181,7 +2266,7 @@ unsigned X86FastISel::TargetMaterializeFloatZero(const ConstantFP *CF) {
     break;
   case MVT::f80:
     // No f80 support yet.
-    return false;
+    return 0;
   }
 
   unsigned ResultReg = createResultReg(RC);
diff --git a/lib/Target/X86/X86FloatingPoint.cpp b/lib/Target/X86/X86FloatingPoint.cpp
index 791f5982af7c..0585b43a4640 100644
--- a/lib/Target/X86/X86FloatingPoint.cpp
+++ b/lib/Target/X86/X86FloatingPoint.cpp
@@ -26,17 +26,17 @@
 #define DEBUG_TYPE "x86-codegen"
 #include "X86.h"
 #include "X86InstrInfo.h"
-#include "llvm/InlineAsm.h"
 #include "llvm/ADT/DepthFirstIterator.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
-#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/EdgeBundles.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/Passes.h"
+#include "llvm/IR/InlineAsm.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
@@ -111,7 +111,7 @@ namespace {
     EdgeBundles *Bundles;
 
     // Return a bitmask of FP registers in block's live-in list.
-    unsigned calcLiveInMask(MachineBasicBlock *MBB) {
+    static unsigned calcLiveInMask(MachineBasicBlock *MBB) {
       unsigned Mask = 0;
       for (MachineBasicBlock::livein_iterator I = MBB->livein_begin(),
            E = MBB->livein_end(); I != E; ++I) {
@@ -198,7 +198,7 @@ namespace {
     }
 
     /// getScratchReg - Return an FP register that is not currently in use.
-    unsigned getScratchReg() {
+    unsigned getScratchReg() const {
       for (int i = NumFPRegs - 1; i >= 8; --i)
         if (!isLive(i))
           return i;
@@ -206,7 +206,7 @@ namespace {
     }
 
     /// isScratchReg - Returns trus if RegNo is a scratch FP register.
-    bool isScratchReg(unsigned RegNo) {
+    static bool isScratchReg(unsigned RegNo) {
       return RegNo > 8 && RegNo < NumFPRegs;
     }
 
@@ -311,7 +311,7 @@ namespace {
     void handleSpecialFP(MachineBasicBlock::iterator &I);
 
     // Check if a COPY instruction is using FP registers.
-    bool isFPCopy(MachineInstr *MI) {
+    static bool isFPCopy(MachineInstr *MI) {
       unsigned DstReg = MI->getOperand(0).getReg();
       unsigned SrcReg = MI->getOperand(1).getReg();
 
diff --git a/lib/Target/X86/X86FrameLowering.cpp b/lib/Target/X86/X86FrameLowering.cpp
index 369589d469a6..54cbd40274a7 100644
--- a/lib/Target/X86/X86FrameLowering.cpp
+++ b/lib/Target/X86/X86FrameLowering.cpp
@@ -17,18 +17,18 @@
 #include "X86MachineFunctionInfo.h"
 #include "X86Subtarget.h"
 #include "X86TargetMachine.h"
-#include "llvm/Function.h"
+#include "llvm/ADT/SmallSet.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetOptions.h"
 #include "llvm/Support/CommandLine.h"
-#include "llvm/ADT/SmallSet.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
@@ -50,13 +50,13 @@ bool X86FrameLowering::hasFP(const MachineFunction &MF) const {
   return (MF.getTarget().Options.DisableFramePointerElim(MF) ||
           RegInfo->needsStackRealignment(MF) ||
           MFI->hasVarSizedObjects() ||
-          MFI->isFrameAddressTaken() ||
+          MFI->isFrameAddressTaken() || MF.hasMSInlineAsm() ||
           MF.getInfo<X86MachineFunctionInfo>()->getForceFramePointer() ||
           MMI.callsUnwindInit() || MMI.callsEHReturn());
 }
 
-static unsigned getSUBriOpcode(unsigned is64Bit, int64_t Imm) {
-  if (is64Bit) {
+static unsigned getSUBriOpcode(unsigned IsLP64, int64_t Imm) {
+  if (IsLP64) {
     if (isInt<8>(Imm))
       return X86::SUB64ri8;
     return X86::SUB64ri32;
@@ -67,8 +67,8 @@ static unsigned getSUBriOpcode(unsigned is64Bit, int64_t Imm) {
   }
 }
 
-static unsigned getADDriOpcode(unsigned is64Bit, int64_t Imm) {
-  if (is64Bit) {
+static unsigned getADDriOpcode(unsigned IsLP64, int64_t Imm) {
+  if (IsLP64) {
     if (isInt<8>(Imm))
       return X86::ADD64ri8;
     return X86::ADD64ri32;
@@ -79,8 +79,8 @@ static unsigned getADDriOpcode(unsigned is64Bit, int64_t Imm) {
   }
 }
 
-static unsigned getLEArOpcode(unsigned is64Bit) {
-  return is64Bit ? X86::LEA64r : X86::LEA32r;
+static unsigned getLEArOpcode(unsigned IsLP64) {
+  return IsLP64 ? X86::LEA64r : X86::LEA32r;
 }
 
 /// findDeadCallerSavedReg - Return a caller-saved register that isn't live
@@ -145,17 +145,17 @@ static unsigned findDeadCallerSavedReg(MachineBasicBlock &MBB,
 static
 void emitSPUpdate(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
                   unsigned StackPtr, int64_t NumBytes,
-                  bool Is64Bit, bool UseLEA,
+                  bool Is64Bit, bool IsLP64, bool UseLEA,
                   const TargetInstrInfo &TII, const TargetRegisterInfo &TRI) {
   bool isSub = NumBytes < 0;
   uint64_t Offset = isSub ? -NumBytes : NumBytes;
   unsigned Opc;
   if (UseLEA)
-    Opc = getLEArOpcode(Is64Bit);
+    Opc = getLEArOpcode(IsLP64);
   else
     Opc = isSub
-      ? getSUBriOpcode(Is64Bit, Offset)
-      : getADDriOpcode(Is64Bit, Offset);
+      ? getSUBriOpcode(IsLP64, Offset)
+      : getADDriOpcode(IsLP64, Offset);
 
   uint64_t Chunk = (1LL << 31) - 1;
   DebugLoc DL = MBB.findDebugLoc(MBBI);
@@ -625,6 +625,22 @@ uint32_t X86FrameLowering::getCompactUnwindEncoding(MachineFunction &MF) const {
   return CompactUnwindEncoding;
 }
 
+/// usesTheStack - This function checks if any of the users of EFLAGS
+/// copies the EFLAGS. We know that the code that lowers COPY of EFLAGS has
+/// to use the stack, and if we don't adjust the stack we clobber the first
+/// frame index.
+/// See X86InstrInfo::copyPhysReg.
+static bool usesTheStack(MachineFunction &MF) {
+  MachineRegisterInfo &MRI = MF.getRegInfo();
+
+  for (MachineRegisterInfo::reg_iterator ri = MRI.reg_begin(X86::EFLAGS),
+       re = MRI.reg_end(); ri != re; ++ri)
+    if (ri->isCopy())
+      return true;
+
+  return false;
+}
+
 /// emitPrologue - Push callee-saved registers onto the stack, which
 /// automatically adjust the stack pointer. Adjust the stack pointer to allocate
 /// space for local variables. Also emit labels used by the exception handler to
@@ -644,6 +660,7 @@ void X86FrameLowering::emitPrologue(MachineFunction &MF) const {
   uint64_t StackSize = MFI->getStackSize();    // Number of bytes to allocate.
   bool HasFP = hasFP(MF);
   bool Is64Bit = STI.is64Bit();
+  bool IsLP64 = STI.isTarget64BitLP64();
   bool IsWin64 = STI.isTargetWin64();
   bool UseLEA = STI.useLeaForSP();
   unsigned StackAlign = getStackAlignment();
@@ -673,12 +690,15 @@ void X86FrameLowering::emitPrologue(MachineFunction &MF) const {
   // If this is x86-64 and the Red Zone is not disabled, if we are a leaf
   // function, and use up to 128 bytes of stack space, don't have a frame
   // pointer, calls, or dynamic alloca then we do not need to adjust the
-  // stack pointer (we fit in the Red Zone).
-  if (Is64Bit && !Fn->getFnAttributes().hasAttribute(Attributes::NoRedZone) &&
+  // stack pointer (we fit in the Red Zone). We also check that we don't
+  // push and pop from the stack.
+  if (Is64Bit && !Fn->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
+                                                   Attribute::NoRedZone) &&
       !RegInfo->needsStackRealignment(MF) &&
       !MFI->hasVarSizedObjects() &&                     // No dynamic alloca.
       !MFI->adjustsStack() &&                           // No calls.
       !IsWin64 &&                                       // Win64 has no Red Zone
+      !usesTheStack(MF) &&                              // Don't push and pop.
       !MF.getTarget().Options.EnableSegmentedStacks) {  // Regular stack
     uint64_t MinSize = X86FI->getCalleeSavedFrameSize();
     if (HasFP) MinSize += SlotSize;
@@ -692,7 +712,7 @@ void X86FrameLowering::emitPrologue(MachineFunction &MF) const {
   if (TailCallReturnAddrDelta < 0) {
     MachineInstr *MI =
       BuildMI(MBB, MBBI, DL,
-              TII.get(getSUBriOpcode(Is64Bit, -TailCallReturnAddrDelta)),
+              TII.get(getSUBriOpcode(IsLP64, -TailCallReturnAddrDelta)),
               StackPtr)
         .addReg(StackPtr)
         .addImm(-TailCallReturnAddrDelta)
@@ -908,7 +928,7 @@ void X86FrameLowering::emitPrologue(MachineFunction &MF) const {
     // MSVC x64's __chkstk needs to adjust %rsp.
     // FIXME: %rax preserves the offset and should be available.
     if (isSPUpdateNeeded)
-      emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit,
+      emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit, IsLP64,
                    UseLEA, TII, *RegInfo);
 
     if (isEAXAlive) {
@@ -920,7 +940,7 @@ void X86FrameLowering::emitPrologue(MachineFunction &MF) const {
         MBB.insert(MBBI, MI);
     }
   } else if (NumBytes)
-    emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit,
+    emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit, IsLP64,
                  UseLEA, TII, *RegInfo);
 
   // If we need a base pointer, set it up here. It's whatever the value
@@ -977,6 +997,7 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
   unsigned RetOpcode = MBBI->getOpcode();
   DebugLoc DL = MBBI->getDebugLoc();
   bool Is64Bit = STI.is64Bit();
+  bool IsLP64 = STI.isTarget64BitLP64();
   bool UseLEA = STI.useLeaForSP();
   unsigned StackAlign = getStackAlignment();
   unsigned SlotSize = RegInfo->getSlotSize();
@@ -1062,7 +1083,7 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
     if (RegInfo->needsStackRealignment(MF))
       MBBI = FirstCSPop;
     if (CSSize != 0) {
-      unsigned Opc = getLEArOpcode(Is64Bit);
+      unsigned Opc = getLEArOpcode(IsLP64);
       addRegOffset(BuildMI(MBB, MBBI, DL, TII.get(Opc), StackPtr),
                    FramePtr, false, -CSSize);
     } else {
@@ -1072,7 +1093,8 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
     }
   } else if (NumBytes) {
     // Adjust stack pointer back: ESP += numbytes.
-    emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, UseLEA, TII, *RegInfo);
+    emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, IsLP64, UseLEA,
+                 TII, *RegInfo);
   }
 
   // We're returning from function via eh_return.
@@ -1107,7 +1129,8 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
     if (Offset) {
       // Check for possible merge with preceding ADD instruction.
       Offset += mergeSPUpdates(MBB, MBBI, StackPtr, true);
-      emitSPUpdate(MBB, MBBI, StackPtr, Offset, Is64Bit, UseLEA, TII, *RegInfo);
+      emitSPUpdate(MBB, MBBI, StackPtr, Offset, Is64Bit, IsLP64,
+                   UseLEA, TII, *RegInfo);
     }
 
     // Jump to label or value in register.
@@ -1138,7 +1161,7 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
     }
 
     MachineInstr *NewMI = prior(MBBI);
-    NewMI->copyImplicitOps(MBBI);
+    NewMI->copyImplicitOps(MF, MBBI);
 
     // Delete the pseudo instruction TCRETURN.
     MBB.erase(MBBI);
@@ -1150,7 +1173,8 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
 
     // Check for possible merge with preceding ADD instruction.
     delta += mergeSPUpdates(MBB, MBBI, StackPtr, true);
-    emitSPUpdate(MBB, MBBI, StackPtr, delta, Is64Bit, UseLEA, TII, *RegInfo);
+    emitSPUpdate(MBB, MBBI, StackPtr, delta, Is64Bit, IsLP64, UseLEA, TII,
+                 *RegInfo);
   }
 }
 
@@ -1362,17 +1386,25 @@ HasNestArgument(const MachineFunction *MF) {
   return false;
 }
 
-
-/// GetScratchRegister - Get a register for performing work in the segmented
-/// stack prologue. Depending on platform and the properties of the function
-/// either one or two registers will be needed. Set primary to true for
-/// the first register, false for the second.
+/// GetScratchRegister - Get a temp register for performing work in the
+/// segmented stack and the Erlang/HiPE stack prologue. Depending on platform
+/// and the properties of the function either one or two registers will be
+/// needed. Set primary to true for the first register, false for the second.
 static unsigned
 GetScratchRegister(bool Is64Bit, const MachineFunction &MF, bool Primary) {
+  CallingConv::ID CallingConvention = MF.getFunction()->getCallingConv();
+
+  // Erlang stuff.
+  if (CallingConvention == CallingConv::HiPE) {
+    if (Is64Bit)
+      return Primary ? X86::R14 : X86::R13;
+    else
+      return Primary ? X86::EBX : X86::EDI;
+  }
+
   if (Is64Bit)
     return Primary ? X86::R11 : X86::R12;
 
-  CallingConv::ID CallingConvention = MF.getFunction()->getCallingConv();
   bool IsNested = HasNestArgument(&MF);
 
   if (CallingConvention == CallingConv::X86_FastCall ||
@@ -1400,7 +1432,6 @@ X86FrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
   bool Is64Bit = STI.is64Bit();
   unsigned TlsReg, TlsOffset;
   DebugLoc DL;
-  const X86Subtarget *ST = &MF.getTarget().getSubtarget<X86Subtarget>();
 
   unsigned ScratchReg = GetScratchRegister(Is64Bit, MF, true);
   assert(!MF.getRegInfo().isLiveIn(ScratchReg) &&
@@ -1408,8 +1439,8 @@ X86FrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
 
   if (MF.getFunction()->isVarArg())
     report_fatal_error("Segmented stacks do not support vararg functions.");
-  if (!ST->isTargetLinux() && !ST->isTargetDarwin() &&
-      !ST->isTargetWin32() && !ST->isTargetFreeBSD())
+  if (!STI.isTargetLinux() && !STI.isTargetDarwin() &&
+      !STI.isTargetWin32() && !STI.isTargetFreeBSD())
     report_fatal_error("Segmented stacks not supported on this platform.");
 
   MachineBasicBlock *allocMBB = MF.CreateMachineBasicBlock();
@@ -1447,13 +1478,13 @@ X86FrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
 
   // Read the limit off the current stacklet off the stack_guard location.
   if (Is64Bit) {
-    if (ST->isTargetLinux()) {
+    if (STI.isTargetLinux()) {
       TlsReg = X86::FS;
       TlsOffset = 0x70;
-    } else if (ST->isTargetDarwin()) {
+    } else if (STI.isTargetDarwin()) {
       TlsReg = X86::GS;
       TlsOffset = 0x60 + 90*8; // See pthread_machdep.h. Steal TLS slot 90.
-    } else if (ST->isTargetFreeBSD()) {
+    } else if (STI.isTargetFreeBSD()) {
       TlsReg = X86::FS;
       TlsOffset = 0x18;
     } else {
@@ -1469,16 +1500,16 @@ X86FrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
     BuildMI(checkMBB, DL, TII.get(X86::CMP64rm)).addReg(ScratchReg)
       .addReg(0).addImm(1).addReg(0).addImm(TlsOffset).addReg(TlsReg);
   } else {
-    if (ST->isTargetLinux()) {
+    if (STI.isTargetLinux()) {
       TlsReg = X86::GS;
       TlsOffset = 0x30;
-    } else if (ST->isTargetDarwin()) {
+    } else if (STI.isTargetDarwin()) {
       TlsReg = X86::GS;
       TlsOffset = 0x48 + 90*4;
-    } else if (ST->isTargetWin32()) {
+    } else if (STI.isTargetWin32()) {
       TlsReg = X86::FS;
       TlsOffset = 0x14; // pvArbitrary, reserved for application use
-    } else if (ST->isTargetFreeBSD()) {
+    } else if (STI.isTargetFreeBSD()) {
       report_fatal_error("Segmented stacks not supported on FreeBSD i386.");
     } else {
       report_fatal_error("Segmented stacks not supported on this platform.");
@@ -1490,10 +1521,10 @@ X86FrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
       BuildMI(checkMBB, DL, TII.get(X86::LEA32r), ScratchReg).addReg(X86::ESP)
         .addImm(1).addReg(0).addImm(-StackSize).addReg(0);
 
-    if (ST->isTargetLinux() || ST->isTargetWin32()) {
+    if (STI.isTargetLinux() || STI.isTargetWin32()) {
       BuildMI(checkMBB, DL, TII.get(X86::CMP32rm)).addReg(ScratchReg)
         .addReg(0).addImm(0).addReg(0).addImm(TlsOffset).addReg(TlsReg);
-    } else if (ST->isTargetDarwin()) {
+    } else if (STI.isTargetDarwin()) {
 
       // TlsOffset doesn't fit into a mod r/m byte so we need an extra register
       unsigned ScratchReg2;
@@ -1579,3 +1610,228 @@ X86FrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
   MF.verify();
 #endif
 }
+
+/// Erlang programs may need a special prologue to handle the stack size they
+/// might need at runtime. That is because Erlang/OTP does not implement a C
+/// stack but uses a custom implementation of hybrid stack/heap architecture.
+/// (for more information see Eric Stenman's Ph.D. thesis:
+/// http://publications.uu.se/uu/fulltext/nbn_se_uu_diva-2688.pdf)
+///
+/// CheckStack:
+///	  temp0 = sp - MaxStack
+///	  if( temp0 < SP_LIMIT(P) ) goto IncStack else goto OldStart
+/// OldStart:
+///	  ...
+/// IncStack:
+///	  call inc_stack   # doubles the stack space
+///	  temp0 = sp - MaxStack
+///	  if( temp0 < SP_LIMIT(P) ) goto IncStack else goto OldStart
+void X86FrameLowering::adjustForHiPEPrologue(MachineFunction &MF) const {
+  const X86InstrInfo &TII = *TM.getInstrInfo();
+  MachineFrameInfo *MFI = MF.getFrameInfo();
+  const unsigned SlotSize = TM.getRegisterInfo()->getSlotSize();
+  const bool Is64Bit = STI.is64Bit();
+  DebugLoc DL;
+  // HiPE-specific values
+  const unsigned HipeLeafWords = 24;
+  const unsigned CCRegisteredArgs = Is64Bit ? 6 : 5;
+  const unsigned Guaranteed = HipeLeafWords * SlotSize;
+  unsigned CallerStkArity = MF.getFunction()->arg_size() > CCRegisteredArgs ?
+                            MF.getFunction()->arg_size() - CCRegisteredArgs : 0;
+  unsigned MaxStack = MFI->getStackSize() + CallerStkArity*SlotSize + SlotSize;
+
+  assert(STI.isTargetLinux() &&
+         "HiPE prologue is only supported on Linux operating systems.");
+
+  // Compute the largest caller's frame that is needed to fit the callees'
+  // frames. This 'MaxStack' is computed from:
+  //
+  // a) the fixed frame size, which is the space needed for all spilled temps,
+  // b) outgoing on-stack parameter areas, and
+  // c) the minimum stack space this function needs to make available for the
+  //    functions it calls (a tunable ABI property).
+  if (MFI->hasCalls()) {
+    unsigned MoreStackForCalls = 0;
+
+    for (MachineFunction::iterator MBBI = MF.begin(), MBBE = MF.end();
+         MBBI != MBBE; ++MBBI)
+      for (MachineBasicBlock::iterator MI = MBBI->begin(), ME = MBBI->end();
+           MI != ME; ++MI) {
+        if (!MI->isCall())
+          continue;
+
+        // Get callee operand.
+        const MachineOperand &MO = MI->getOperand(0);
+
+        // Only take account of global function calls (no closures etc.).
+        if (!MO.isGlobal())
+          continue;
+
+        const Function *F = dyn_cast<Function>(MO.getGlobal());
+        if (!F)
+          continue;
+
+        // Do not update 'MaxStack' for primitive and built-in functions
+        // (encoded with names either starting with "erlang."/"bif_" or not
+        // having a ".", such as a simple <Module>.<Function>.<Arity>, or an
+        // "_", such as the BIF "suspend_0") as they are executed on another
+        // stack.
+        if (F->getName().find("erlang.") != StringRef::npos ||
+            F->getName().find("bif_") != StringRef::npos ||
+            F->getName().find_first_of("._") == StringRef::npos)
+          continue;
+
+        unsigned CalleeStkArity =
+          F->arg_size() > CCRegisteredArgs ? F->arg_size()-CCRegisteredArgs : 0;
+        if (HipeLeafWords - 1 > CalleeStkArity)
+          MoreStackForCalls = std::max(MoreStackForCalls,
+                               (HipeLeafWords - 1 - CalleeStkArity) * SlotSize);
+      }
+    MaxStack += MoreStackForCalls;
+  }
+
+  // If the stack frame needed is larger than the guaranteed then runtime checks
+  // and calls to "inc_stack_0" BIF should be inserted in the assembly prologue.
+  if (MaxStack > Guaranteed) {
+    MachineBasicBlock &prologueMBB = MF.front();
+    MachineBasicBlock *stackCheckMBB = MF.CreateMachineBasicBlock();
+    MachineBasicBlock *incStackMBB = MF.CreateMachineBasicBlock();
+
+    for (MachineBasicBlock::livein_iterator I = prologueMBB.livein_begin(),
+           E = prologueMBB.livein_end(); I != E; I++) {
+      stackCheckMBB->addLiveIn(*I);
+      incStackMBB->addLiveIn(*I);
+    }
+
+    MF.push_front(incStackMBB);
+    MF.push_front(stackCheckMBB);
+
+    unsigned ScratchReg, SPReg, PReg, SPLimitOffset;
+    unsigned LEAop, CMPop, CALLop;
+    if (Is64Bit) {
+      SPReg = X86::RSP;
+      PReg  = X86::RBP;
+      LEAop = X86::LEA64r;
+      CMPop = X86::CMP64rm;
+      CALLop = X86::CALL64pcrel32;
+      SPLimitOffset = 0x90;
+    } else {
+      SPReg = X86::ESP;
+      PReg  = X86::EBP;
+      LEAop = X86::LEA32r;
+      CMPop = X86::CMP32rm;
+      CALLop = X86::CALLpcrel32;
+      SPLimitOffset = 0x4c;
+    }
+
+    ScratchReg = GetScratchRegister(Is64Bit, MF, true);
+    assert(!MF.getRegInfo().isLiveIn(ScratchReg) &&
+           "HiPE prologue scratch register is live-in");
+
+    // Create new MBB for StackCheck:
+    addRegOffset(BuildMI(stackCheckMBB, DL, TII.get(LEAop), ScratchReg),
+                 SPReg, false, -MaxStack);
+    // SPLimitOffset is in a fixed heap location (pointed by BP).
+    addRegOffset(BuildMI(stackCheckMBB, DL, TII.get(CMPop))
+                 .addReg(ScratchReg), PReg, false, SPLimitOffset);
+    BuildMI(stackCheckMBB, DL, TII.get(X86::JAE_4)).addMBB(&prologueMBB);
+
+    // Create new MBB for IncStack:
+    BuildMI(incStackMBB, DL, TII.get(CALLop)).
+      addExternalSymbol("inc_stack_0");
+    addRegOffset(BuildMI(incStackMBB, DL, TII.get(LEAop), ScratchReg),
+                 SPReg, false, -MaxStack);
+    addRegOffset(BuildMI(incStackMBB, DL, TII.get(CMPop))
+                 .addReg(ScratchReg), PReg, false, SPLimitOffset);
+    BuildMI(incStackMBB, DL, TII.get(X86::JLE_4)).addMBB(incStackMBB);
+
+    stackCheckMBB->addSuccessor(&prologueMBB, 99);
+    stackCheckMBB->addSuccessor(incStackMBB, 1);
+    incStackMBB->addSuccessor(&prologueMBB, 99);
+    incStackMBB->addSuccessor(incStackMBB, 1);
+  }
+#ifdef XDEBUG
+  MF.verify();
+#endif
+}
+
+void X86FrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  const X86InstrInfo &TII = *TM.getInstrInfo();
+  const X86RegisterInfo &RegInfo = *TM.getRegisterInfo();
+  unsigned StackPtr = RegInfo.getStackRegister();
+  bool reseveCallFrame = hasReservedCallFrame(MF);
+  int Opcode = I->getOpcode();
+  bool isDestroy = Opcode == TII.getCallFrameDestroyOpcode();
+  bool IsLP64 = STI.isTarget64BitLP64();
+  DebugLoc DL = I->getDebugLoc();
+  uint64_t Amount = !reseveCallFrame ? I->getOperand(0).getImm() : 0;
+  uint64_t CalleeAmt = isDestroy ? I->getOperand(1).getImm() : 0;
+  I = MBB.erase(I);
+
+  if (!reseveCallFrame) {
+    // If the stack pointer can be changed after prologue, turn the
+    // adjcallstackup instruction into a 'sub ESP, <amt>' and the
+    // adjcallstackdown instruction into 'add ESP, <amt>'
+    // TODO: consider using push / pop instead of sub + store / add
+    if (Amount == 0)
+      return;
+
+    // We need to keep the stack aligned properly.  To do this, we round the
+    // amount of space needed for the outgoing arguments up to the next
+    // alignment boundary.
+    unsigned StackAlign = TM.getFrameLowering()->getStackAlignment();
+    Amount = (Amount + StackAlign - 1) / StackAlign * StackAlign;
+
+    MachineInstr *New = 0;
+    if (Opcode == TII.getCallFrameSetupOpcode()) {
+      New = BuildMI(MF, DL, TII.get(getSUBriOpcode(IsLP64, Amount)),
+                    StackPtr)
+        .addReg(StackPtr)
+        .addImm(Amount);
+    } else {
+      assert(Opcode == TII.getCallFrameDestroyOpcode());
+
+      // Factor out the amount the callee already popped.
+      Amount -= CalleeAmt;
+
+      if (Amount) {
+        unsigned Opc = getADDriOpcode(IsLP64, Amount);
+        New = BuildMI(MF, DL, TII.get(Opc), StackPtr)
+          .addReg(StackPtr).addImm(Amount);
+      }
+    }
+
+    if (New) {
+      // The EFLAGS implicit def is dead.
+      New->getOperand(3).setIsDead();
+
+      // Replace the pseudo instruction with a new instruction.
+      MBB.insert(I, New);
+    }
+
+    return;
+  }
+
+  if (Opcode == TII.getCallFrameDestroyOpcode() && CalleeAmt) {
+    // If we are performing frame pointer elimination and if the callee pops
+    // something off the stack pointer, add it back.  We do this until we have
+    // more advanced stack pointer tracking ability.
+    unsigned Opc = getSUBriOpcode(IsLP64, CalleeAmt);
+    MachineInstr *New = BuildMI(MF, DL, TII.get(Opc), StackPtr)
+      .addReg(StackPtr).addImm(CalleeAmt);
+
+    // The EFLAGS implicit def is dead.
+    New->getOperand(3).setIsDead();
+
+    // We are not tracking the stack pointer adjustment by the callee, so make
+    // sure we restore the stack pointer immediately after the call, there may
+    // be spill code inserted between the CALL and ADJCALLSTACKUP instructions.
+    MachineBasicBlock::iterator B = MBB.begin();
+    while (I != B && !llvm::prior(I)->isCall())
+      --I;
+    MBB.insert(I, New);
+  }
+}
+
diff --git a/lib/Target/X86/X86FrameLowering.h b/lib/Target/X86/X86FrameLowering.h
index dc515dc39c79..3f08b9a2e8d2 100644
--- a/lib/Target/X86/X86FrameLowering.h
+++ b/lib/Target/X86/X86FrameLowering.h
@@ -43,6 +43,8 @@ public:
 
   void adjustForSegmentedStacks(MachineFunction &MF) const;
 
+  void adjustForHiPEPrologue(MachineFunction &MF) const;
+
   void processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
                                             RegScavenger *RS = NULL) const;
 
@@ -63,6 +65,10 @@ public:
   int getFrameIndexReference(const MachineFunction &MF, int FI,
                              unsigned &FrameReg) const;
   uint32_t getCompactUnwindEncoding(MachineFunction &MF) const;
+
+  void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                     MachineBasicBlock &MBB,
+                                     MachineBasicBlock::iterator MI) const;
 };
 
 } // End llvm namespace
diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index 99f557417b7c..6041669f8182 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -19,24 +19,21 @@
 #include "X86RegisterInfo.h"
 #include "X86Subtarget.h"
 #include "X86TargetMachine.h"
-#include "llvm/Instructions.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/Type.h"
-#include "llvm/CodeGen/FunctionLoweringInfo.h"
-#include "llvm/CodeGen/MachineConstantPool.h"
-#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Support/CFG.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/Statistic.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
 using namespace llvm;
 
 STATISTIC(NumLoadMoved, "Number of loads moved below TokenFactor");
@@ -283,13 +280,13 @@ namespace {
 
     /// getTargetMachine - Return a reference to the TargetMachine, casted
     /// to the target-specific type.
-    const X86TargetMachine &getTargetMachine() {
+    const X86TargetMachine &getTargetMachine() const {
       return static_cast<const X86TargetMachine &>(TM);
     }
 
     /// getInstrInfo - Return a reference to the TargetInstrInfo, casted
     /// to the target-specific type.
-    const X86InstrInfo *getInstrInfo() {
+    const X86InstrInfo *getInstrInfo() const {
       return getTargetMachine().getInstrInfo();
     }
   };
@@ -423,6 +420,11 @@ static bool isCalleeLoad(SDValue Callee, SDValue &Chain, bool HasCallSeq) {
 
   if (!Chain.getNumOperands())
     return false;
+  // Since we are not checking for AA here, conservatively abort if the chain
+  // writes to memory. It's not safe to move the callee (a load) across a store.
+  if (isa<MemSDNode>(Chain.getNode()) &&
+      cast<MemSDNode>(Chain.getNode())->writeMem())
+    return false;
   if (Chain.getOperand(0).getNode() == Callee.getNode())
     return true;
   if (Chain.getOperand(0).getOpcode() == ISD::TokenFactor &&
@@ -434,17 +436,19 @@ static bool isCalleeLoad(SDValue Callee, SDValue &Chain, bool HasCallSeq) {
 
 void X86DAGToDAGISel::PreprocessISelDAG() {
   // OptForSize is used in pattern predicates that isel is matching.
-  OptForSize = MF->getFunction()->getFnAttributes().
-    hasAttribute(Attributes::OptimizeForSize);
+  OptForSize = MF->getFunction()->getAttributes().
+    hasAttribute(AttributeSet::FunctionIndex, Attribute::OptimizeForSize);
 
   for (SelectionDAG::allnodes_iterator I = CurDAG->allnodes_begin(),
        E = CurDAG->allnodes_end(); I != E; ) {
     SDNode *N = I++;  // Preincrement iterator to avoid invalidation issues.
 
     if (OptLevel != CodeGenOpt::None &&
-        (N->getOpcode() == X86ISD::CALL ||
+        // Only does this when target favors doesn't favor register indirect
+        // call.
+        ((N->getOpcode() == X86ISD::CALL && !Subtarget->callRegIndirect()) ||
          (N->getOpcode() == X86ISD::TC_RETURN &&
-          // Only does this if load can be foled into TC_RETURN.
+          // Only does this if load can be folded into TC_RETURN.
           (Subtarget->is64Bit() ||
            getTargetMachine().getRelocationModel() != Reloc::PIC_)))) {
       /// Also try moving call address load from outside callseq_start to just
@@ -1040,8 +1044,8 @@ bool X86DAGToDAGISel::MatchAddressRecursively(SDValue N, X86ISelAddressMode &AM,
         AM.IndexReg = ShVal;
         return false;
       }
-    break;
     }
+    break;
 
   case ISD::SRL: {
     // Scale must not be used already.
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index b35fb514bf94..69341869aa3e 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -14,20 +14,15 @@
 
 #define DEBUG_TYPE "x86-isel"
 #include "X86ISelLowering.h"
+#include "Utils/X86ShuffleDecode.h"
 #include "X86.h"
 #include "X86InstrBuilder.h"
 #include "X86TargetMachine.h"
 #include "X86TargetObjectFile.h"
-#include "Utils/X86ShuffleDecode.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/GlobalAlias.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/Function.h"
-#include "llvm/Instructions.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/LLVMContext.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/StringExtras.h"
+#include "llvm/ADT/VariadicFunction.h"
 #include "llvm/CodeGen/IntrinsicLowering.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -35,14 +30,19 @@
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/LLVMContext.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/ADT/SmallSet.h"
-#include "llvm/ADT/Statistic.h"
-#include "llvm/ADT/StringExtras.h"
-#include "llvm/ADT/VariadicFunction.h"
 #include "llvm/Support/CallSite.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
@@ -85,6 +85,11 @@ static SDValue Extract128BitVector(SDValue Vec, unsigned IdxVal,
   unsigned NormalizedIdxVal = (((IdxVal * ElVT.getSizeInBits()) / 128)
                                * ElemsPerChunk);
 
+  // If the input is a buildvector just emit a smaller one.
+  if (Vec.getOpcode() == ISD::BUILD_VECTOR)
+    return DAG.getNode(ISD::BUILD_VECTOR, dl, ResultVT,
+                       Vec->op_begin()+NormalizedIdxVal, ElemsPerChunk);
+
   SDValue VecIdx = DAG.getIntPtrConstant(NormalizedIdxVal);
   SDValue Result = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, ResultVT, Vec,
                                VecIdx);
@@ -181,9 +186,12 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setSchedulingPreference(Sched::RegPressure);
   setStackPointerRegisterToSaveRestore(RegInfo->getStackRegister());
 
-  // Bypass i32 with i8 on Atom when compiling with O2
-  if (Subtarget->hasSlowDivide() && TM.getOptLevel() >= CodeGenOpt::Default)
+  // Bypass expensive divides on Atom when compiling with O2
+  if (Subtarget->hasSlowDivide() && TM.getOptLevel() >= CodeGenOpt::Default) {
     addBypassSlowDiv(32, 8);
+    if (Subtarget->is64Bit())
+      addBypassSlowDiv(64, 16);
+  }
 
   if (Subtarget->isTargetWindows() && !Subtarget->isTargetCygMing()) {
     // Setup Windows compiler runtime calls.
@@ -368,7 +376,13 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
 
   setOperationAction(ISD::BR_JT            , MVT::Other, Expand);
   setOperationAction(ISD::BRCOND           , MVT::Other, Custom);
-  setOperationAction(ISD::BR_CC            , MVT::Other, Expand);
+  setOperationAction(ISD::BR_CC            , MVT::f32,   Expand);
+  setOperationAction(ISD::BR_CC            , MVT::f64,   Expand);
+  setOperationAction(ISD::BR_CC            , MVT::f80,   Expand);
+  setOperationAction(ISD::BR_CC            , MVT::i8,    Expand);
+  setOperationAction(ISD::BR_CC            , MVT::i16,   Expand);
+  setOperationAction(ISD::BR_CC            , MVT::i32,   Expand);
+  setOperationAction(ISD::BR_CC            , MVT::i64,   Expand);
   setOperationAction(ISD::SELECT_CC        , MVT::Other, Expand);
   if (Subtarget->is64Bit())
     setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i32, Legal);
@@ -456,7 +470,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::SETCC         , MVT::i64  , Custom);
   }
   setOperationAction(ISD::EH_RETURN       , MVT::Other, Custom);
-  // NOTE: EH_SJLJ_SETJMP/_LONGJMP supported here is NOT intened to support
+  // NOTE: EH_SJLJ_SETJMP/_LONGJMP supported here is NOT intended to support
   // SjLj exception handling but a light-weight setjmp/longjmp replacement to
   // support continuation, user-level threading, and etc.. As a result, no
   // other SjLj exception interfaces are implemented and please don't build
@@ -605,10 +619,12 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FGETSIGN, MVT::i32, Custom);
 
     // We don't support sin/cos/fmod
-    setOperationAction(ISD::FSIN , MVT::f64, Expand);
-    setOperationAction(ISD::FCOS , MVT::f64, Expand);
-    setOperationAction(ISD::FSIN , MVT::f32, Expand);
-    setOperationAction(ISD::FCOS , MVT::f32, Expand);
+    setOperationAction(ISD::FSIN   , MVT::f64, Expand);
+    setOperationAction(ISD::FCOS   , MVT::f64, Expand);
+    setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
+    setOperationAction(ISD::FSIN   , MVT::f32, Expand);
+    setOperationAction(ISD::FCOS   , MVT::f32, Expand);
+    setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
 
     // Expand FP immediates into loads from the stack, except for the special
     // cases we handle.
@@ -633,8 +649,9 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FCOPYSIGN, MVT::f32, Custom);
 
     // We don't support sin/cos/fmod
-    setOperationAction(ISD::FSIN , MVT::f32, Expand);
-    setOperationAction(ISD::FCOS , MVT::f32, Expand);
+    setOperationAction(ISD::FSIN   , MVT::f32, Expand);
+    setOperationAction(ISD::FCOS   , MVT::f32, Expand);
+    setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
 
     // Special cases we handle for FP constants.
     addLegalFPImmediate(APFloat(+0.0f)); // xorps
@@ -644,8 +661,9 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     addLegalFPImmediate(APFloat(-1.0)); // FLD1/FCHS
 
     if (!TM.Options.UnsafeFPMath) {
-      setOperationAction(ISD::FSIN           , MVT::f64  , Expand);
-      setOperationAction(ISD::FCOS           , MVT::f64  , Expand);
+      setOperationAction(ISD::FSIN   , MVT::f64, Expand);
+      setOperationAction(ISD::FCOS   , MVT::f64, Expand);
+      setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
     }
   } else if (!TM.Options.UseSoftFloat) {
     // f32 and f64 in x87.
@@ -659,10 +677,12 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
 
     if (!TM.Options.UnsafeFPMath) {
-      setOperationAction(ISD::FSIN           , MVT::f32  , Expand);
-      setOperationAction(ISD::FSIN           , MVT::f64  , Expand);
-      setOperationAction(ISD::FCOS           , MVT::f32  , Expand);
-      setOperationAction(ISD::FCOS           , MVT::f64  , Expand);
+      setOperationAction(ISD::FSIN   , MVT::f64, Expand);
+      setOperationAction(ISD::FSIN   , MVT::f32, Expand);
+      setOperationAction(ISD::FCOS   , MVT::f64, Expand);
+      setOperationAction(ISD::FCOS   , MVT::f32, Expand);
+      setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
+      setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
     }
     addLegalFPImmediate(APFloat(+0.0)); // FLD0
     addLegalFPImmediate(APFloat(+1.0)); // FLD1
@@ -699,8 +719,9 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     }
 
     if (!TM.Options.UnsafeFPMath) {
-      setOperationAction(ISD::FSIN           , MVT::f80  , Expand);
-      setOperationAction(ISD::FCOS           , MVT::f80  , Expand);
+      setOperationAction(ISD::FSIN   , MVT::f80, Expand);
+      setOperationAction(ISD::FCOS   , MVT::f80, Expand);
+      setOperationAction(ISD::FSINCOS, MVT::f80, Expand);
     }
 
     setOperationAction(ISD::FFLOOR, MVT::f80, Expand);
@@ -725,74 +746,81 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
   // First set operation action for all vector types to either promote
   // (for widening) or expand (for scalarization). Then we will selectively
   // turn on ones that can be effectively codegen'd.
-  for (int VT = MVT::FIRST_VECTOR_VALUETYPE;
-           VT <= MVT::LAST_VECTOR_VALUETYPE; ++VT) {
-    setOperationAction(ISD::ADD , (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SUB , (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FADD, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FNEG, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FSUB, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::MUL , (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FMUL, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SDIV, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::UDIV, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FDIV, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SREM, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::UREM, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::LOAD, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::VECTOR_SHUFFLE, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::EXTRACT_VECTOR_ELT,(MVT::SimpleValueType)VT,Expand);
-    setOperationAction(ISD::INSERT_VECTOR_ELT,(MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::EXTRACT_SUBVECTOR,(MVT::SimpleValueType)VT,Expand);
-    setOperationAction(ISD::INSERT_SUBVECTOR,(MVT::SimpleValueType)VT,Expand);
-    setOperationAction(ISD::FABS, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FSIN, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FCOS, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FREM, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FMA,  (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FPOWI, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FSQRT, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FCOPYSIGN, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FFLOOR, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SMUL_LOHI, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::UMUL_LOHI, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SDIVREM, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::UDIVREM, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FPOW, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::CTPOP, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::CTTZ, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::CTTZ_ZERO_UNDEF, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::CTLZ, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::CTLZ_ZERO_UNDEF, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SHL, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SRA, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SRL, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::ROTL, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::ROTR, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::BSWAP, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SETCC, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FLOG, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FLOG2, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FLOG10, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FEXP, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FEXP2, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FP_TO_UINT, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::FP_TO_SINT, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::UINT_TO_FP, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SINT_TO_FP, (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SIGN_EXTEND_INREG, (MVT::SimpleValueType)VT,Expand);
-    setOperationAction(ISD::TRUNCATE,  (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::SIGN_EXTEND,  (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::ZERO_EXTEND,  (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::ANY_EXTEND,  (MVT::SimpleValueType)VT, Expand);
-    setOperationAction(ISD::VSELECT,  (MVT::SimpleValueType)VT, Expand);
+  for (int i = MVT::FIRST_VECTOR_VALUETYPE;
+           i <= MVT::LAST_VECTOR_VALUETYPE; ++i) {
+    MVT VT = (MVT::SimpleValueType)i;
+    setOperationAction(ISD::ADD , VT, Expand);
+    setOperationAction(ISD::SUB , VT, Expand);
+    setOperationAction(ISD::FADD, VT, Expand);
+    setOperationAction(ISD::FNEG, VT, Expand);
+    setOperationAction(ISD::FSUB, VT, Expand);
+    setOperationAction(ISD::MUL , VT, Expand);
+    setOperationAction(ISD::FMUL, VT, Expand);
+    setOperationAction(ISD::SDIV, VT, Expand);
+    setOperationAction(ISD::UDIV, VT, Expand);
+    setOperationAction(ISD::FDIV, VT, Expand);
+    setOperationAction(ISD::SREM, VT, Expand);
+    setOperationAction(ISD::UREM, VT, Expand);
+    setOperationAction(ISD::LOAD, VT, Expand);
+    setOperationAction(ISD::VECTOR_SHUFFLE, VT, Expand);
+    setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT,Expand);
+    setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Expand);
+    setOperationAction(ISD::EXTRACT_SUBVECTOR, VT,Expand);
+    setOperationAction(ISD::INSERT_SUBVECTOR, VT,Expand);
+    setOperationAction(ISD::FABS, VT, Expand);
+    setOperationAction(ISD::FSIN, VT, Expand);
+    setOperationAction(ISD::FSINCOS, VT, Expand);
+    setOperationAction(ISD::FCOS, VT, Expand);
+    setOperationAction(ISD::FSINCOS, VT, Expand);
+    setOperationAction(ISD::FREM, VT, Expand);
+    setOperationAction(ISD::FMA,  VT, Expand);
+    setOperationAction(ISD::FPOWI, VT, Expand);
+    setOperationAction(ISD::FSQRT, VT, Expand);
+    setOperationAction(ISD::FCOPYSIGN, VT, Expand);
+    setOperationAction(ISD::FFLOOR, VT, Expand);
+    setOperationAction(ISD::FCEIL, VT, Expand);
+    setOperationAction(ISD::FTRUNC, VT, Expand);
+    setOperationAction(ISD::FRINT, VT, Expand);
+    setOperationAction(ISD::FNEARBYINT, VT, Expand);
+    setOperationAction(ISD::SMUL_LOHI, VT, Expand);
+    setOperationAction(ISD::UMUL_LOHI, VT, Expand);
+    setOperationAction(ISD::SDIVREM, VT, Expand);
+    setOperationAction(ISD::UDIVREM, VT, Expand);
+    setOperationAction(ISD::FPOW, VT, Expand);
+    setOperationAction(ISD::CTPOP, VT, Expand);
+    setOperationAction(ISD::CTTZ, VT, Expand);
+    setOperationAction(ISD::CTTZ_ZERO_UNDEF, VT, Expand);
+    setOperationAction(ISD::CTLZ, VT, Expand);
+    setOperationAction(ISD::CTLZ_ZERO_UNDEF, VT, Expand);
+    setOperationAction(ISD::SHL, VT, Expand);
+    setOperationAction(ISD::SRA, VT, Expand);
+    setOperationAction(ISD::SRL, VT, Expand);
+    setOperationAction(ISD::ROTL, VT, Expand);
+    setOperationAction(ISD::ROTR, VT, Expand);
+    setOperationAction(ISD::BSWAP, VT, Expand);
+    setOperationAction(ISD::SETCC, VT, Expand);
+    setOperationAction(ISD::FLOG, VT, Expand);
+    setOperationAction(ISD::FLOG2, VT, Expand);
+    setOperationAction(ISD::FLOG10, VT, Expand);
+    setOperationAction(ISD::FEXP, VT, Expand);
+    setOperationAction(ISD::FEXP2, VT, Expand);
+    setOperationAction(ISD::FP_TO_UINT, VT, Expand);
+    setOperationAction(ISD::FP_TO_SINT, VT, Expand);
+    setOperationAction(ISD::UINT_TO_FP, VT, Expand);
+    setOperationAction(ISD::SINT_TO_FP, VT, Expand);
+    setOperationAction(ISD::SIGN_EXTEND_INREG, VT,Expand);
+    setOperationAction(ISD::TRUNCATE, VT, Expand);
+    setOperationAction(ISD::SIGN_EXTEND, VT, Expand);
+    setOperationAction(ISD::ZERO_EXTEND, VT, Expand);
+    setOperationAction(ISD::ANY_EXTEND, VT, Expand);
+    setOperationAction(ISD::VSELECT, VT, Expand);
     for (int InnerVT = MVT::FIRST_VECTOR_VALUETYPE;
              InnerVT <= MVT::LAST_VECTOR_VALUETYPE; ++InnerVT)
-      setTruncStoreAction((MVT::SimpleValueType)VT,
+      setTruncStoreAction(VT,
                           (MVT::SimpleValueType)InnerVT, Expand);
-    setLoadExtAction(ISD::SEXTLOAD, (MVT::SimpleValueType)VT, Expand);
-    setLoadExtAction(ISD::ZEXTLOAD, (MVT::SimpleValueType)VT, Expand);
-    setLoadExtAction(ISD::EXTLOAD, (MVT::SimpleValueType)VT, Expand);
+    setLoadExtAction(ISD::SEXTLOAD, VT, Expand);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, Expand);
+    setLoadExtAction(ISD::EXTLOAD, VT, Expand);
   }
 
   // FIXME: In order to prevent SSE instructions being expanded to MMX ones
@@ -865,6 +893,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::ADD,                MVT::v8i16, Legal);
     setOperationAction(ISD::ADD,                MVT::v4i32, Legal);
     setOperationAction(ISD::ADD,                MVT::v2i64, Legal);
+    setOperationAction(ISD::MUL,                MVT::v4i32, Custom);
     setOperationAction(ISD::MUL,                MVT::v2i64, Custom);
     setOperationAction(ISD::SUB,                MVT::v16i8, Legal);
     setOperationAction(ISD::SUB,                MVT::v8i16, Legal);
@@ -973,7 +1002,15 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FNEARBYINT,         MVT::f64,   Legal);
 
     setOperationAction(ISD::FFLOOR,             MVT::v4f32, Legal);
+    setOperationAction(ISD::FCEIL,              MVT::v4f32, Legal);
+    setOperationAction(ISD::FTRUNC,             MVT::v4f32, Legal);
+    setOperationAction(ISD::FRINT,              MVT::v4f32, Legal);
+    setOperationAction(ISD::FNEARBYINT,         MVT::v4f32, Legal);
     setOperationAction(ISD::FFLOOR,             MVT::v2f64, Legal);
+    setOperationAction(ISD::FCEIL,              MVT::v2f64, Legal);
+    setOperationAction(ISD::FTRUNC,             MVT::v2f64, Legal);
+    setOperationAction(ISD::FRINT,              MVT::v2f64, Legal);
+    setOperationAction(ISD::FNEARBYINT,         MVT::v2f64, Legal);
 
     // FIXME: Do we need to handle scalar-to-vector here?
     setOperationAction(ISD::MUL,                MVT::v4i32, Legal);
@@ -1016,26 +1053,21 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::SRA,               MVT::v8i16, Custom);
     setOperationAction(ISD::SRA,               MVT::v16i8, Custom);
 
-    if (Subtarget->hasAVX2()) {
-      setOperationAction(ISD::SRL,             MVT::v2i64, Legal);
-      setOperationAction(ISD::SRL,             MVT::v4i32, Legal);
+    // In the customized shift lowering, the legal cases in AVX2 will be
+    // recognized.
+    setOperationAction(ISD::SRL,               MVT::v2i64, Custom);
+    setOperationAction(ISD::SRL,               MVT::v4i32, Custom);
 
-      setOperationAction(ISD::SHL,             MVT::v2i64, Legal);
-      setOperationAction(ISD::SHL,             MVT::v4i32, Legal);
-
-      setOperationAction(ISD::SRA,             MVT::v4i32, Legal);
-    } else {
-      setOperationAction(ISD::SRL,             MVT::v2i64, Custom);
-      setOperationAction(ISD::SRL,             MVT::v4i32, Custom);
+    setOperationAction(ISD::SHL,               MVT::v2i64, Custom);
+    setOperationAction(ISD::SHL,               MVT::v4i32, Custom);
 
-      setOperationAction(ISD::SHL,             MVT::v2i64, Custom);
-      setOperationAction(ISD::SHL,             MVT::v4i32, Custom);
+    setOperationAction(ISD::SRA,               MVT::v4i32, Custom);
 
-      setOperationAction(ISD::SRA,             MVT::v4i32, Custom);
-    }
+    setOperationAction(ISD::SDIV,              MVT::v8i16, Custom);
+    setOperationAction(ISD::SDIV,              MVT::v4i32, Custom);
   }
 
-  if (!TM.Options.UseSoftFloat && Subtarget->hasAVX()) {
+  if (!TM.Options.UseSoftFloat && Subtarget->hasFp256()) {
     addRegisterClass(MVT::v32i8,  &X86::VR256RegClass);
     addRegisterClass(MVT::v16i16, &X86::VR256RegClass);
     addRegisterClass(MVT::v8i32,  &X86::VR256RegClass);
@@ -1053,6 +1085,10 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FDIV,               MVT::v8f32, Legal);
     setOperationAction(ISD::FSQRT,              MVT::v8f32, Legal);
     setOperationAction(ISD::FFLOOR,             MVT::v8f32, Legal);
+    setOperationAction(ISD::FCEIL,              MVT::v8f32, Legal);
+    setOperationAction(ISD::FTRUNC,             MVT::v8f32, Legal);
+    setOperationAction(ISD::FRINT,              MVT::v8f32, Legal);
+    setOperationAction(ISD::FNEARBYINT,         MVT::v8f32, Legal);
     setOperationAction(ISD::FNEG,               MVT::v8f32, Custom);
     setOperationAction(ISD::FABS,               MVT::v8f32, Custom);
 
@@ -1062,14 +1098,20 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FDIV,               MVT::v4f64, Legal);
     setOperationAction(ISD::FSQRT,              MVT::v4f64, Legal);
     setOperationAction(ISD::FFLOOR,             MVT::v4f64, Legal);
+    setOperationAction(ISD::FCEIL,              MVT::v4f64, Legal);
+    setOperationAction(ISD::FTRUNC,             MVT::v4f64, Legal);
+    setOperationAction(ISD::FRINT,              MVT::v4f64, Legal);
+    setOperationAction(ISD::FNEARBYINT,         MVT::v4f64, Legal);
     setOperationAction(ISD::FNEG,               MVT::v4f64, Custom);
     setOperationAction(ISD::FABS,               MVT::v4f64, Custom);
 
     setOperationAction(ISD::TRUNCATE,           MVT::v8i16, Custom);
+    setOperationAction(ISD::TRUNCATE,           MVT::v4i32, Custom);
 
     setOperationAction(ISD::FP_TO_SINT,         MVT::v8i16, Custom);
 
     setOperationAction(ISD::FP_TO_SINT,         MVT::v8i32, Legal);
+    setOperationAction(ISD::SINT_TO_FP,         MVT::v8i16, Promote);
     setOperationAction(ISD::SINT_TO_FP,         MVT::v8i32, Legal);
     setOperationAction(ISD::FP_ROUND,           MVT::v4f32, Legal);
 
@@ -1088,6 +1130,8 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::SRA,               MVT::v16i16, Custom);
     setOperationAction(ISD::SRA,               MVT::v32i8, Custom);
 
+    setOperationAction(ISD::SDIV,              MVT::v16i16, Custom);
+
     setOperationAction(ISD::SETCC,             MVT::v32i8, Custom);
     setOperationAction(ISD::SETCC,             MVT::v16i16, Custom);
     setOperationAction(ISD::SETCC,             MVT::v8i32, Custom);
@@ -1102,16 +1146,23 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::VSELECT,           MVT::v8i32, Legal);
     setOperationAction(ISD::VSELECT,           MVT::v8f32, Legal);
 
+    setOperationAction(ISD::SIGN_EXTEND,       MVT::v4i64, Custom);
+    setOperationAction(ISD::SIGN_EXTEND,       MVT::v8i32, Custom);
+    setOperationAction(ISD::ZERO_EXTEND,       MVT::v4i64, Custom);
+    setOperationAction(ISD::ZERO_EXTEND,       MVT::v8i32, Custom);
+    setOperationAction(ISD::ANY_EXTEND,        MVT::v4i64, Custom);
+    setOperationAction(ISD::ANY_EXTEND,        MVT::v8i32, Custom);
+
     if (Subtarget->hasFMA() || Subtarget->hasFMA4()) {
-      setOperationAction(ISD::FMA,             MVT::v8f32, Custom);
-      setOperationAction(ISD::FMA,             MVT::v4f64, Custom);
-      setOperationAction(ISD::FMA,             MVT::v4f32, Custom);
-      setOperationAction(ISD::FMA,             MVT::v2f64, Custom);
-      setOperationAction(ISD::FMA,             MVT::f32, Custom);
-      setOperationAction(ISD::FMA,             MVT::f64, Custom);
+      setOperationAction(ISD::FMA,             MVT::v8f32, Legal);
+      setOperationAction(ISD::FMA,             MVT::v4f64, Legal);
+      setOperationAction(ISD::FMA,             MVT::v4f32, Legal);
+      setOperationAction(ISD::FMA,             MVT::v2f64, Legal);
+      setOperationAction(ISD::FMA,             MVT::f32, Legal);
+      setOperationAction(ISD::FMA,             MVT::f64, Legal);
     }
 
-    if (Subtarget->hasAVX2()) {
+    if (Subtarget->hasInt256()) {
       setOperationAction(ISD::ADD,             MVT::v4i64, Legal);
       setOperationAction(ISD::ADD,             MVT::v8i32, Legal);
       setOperationAction(ISD::ADD,             MVT::v16i16, Legal);
@@ -1129,13 +1180,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
 
       setOperationAction(ISD::VSELECT,         MVT::v32i8, Legal);
 
-      setOperationAction(ISD::SRL,             MVT::v4i64, Legal);
-      setOperationAction(ISD::SRL,             MVT::v8i32, Legal);
-
-      setOperationAction(ISD::SHL,             MVT::v4i64, Legal);
-      setOperationAction(ISD::SHL,             MVT::v8i32, Legal);
-
-      setOperationAction(ISD::SRA,             MVT::v8i32, Legal);
+      setOperationAction(ISD::SDIV,            MVT::v8i32, Custom);
     } else {
       setOperationAction(ISD::ADD,             MVT::v4i64, Custom);
       setOperationAction(ISD::ADD,             MVT::v8i32, Custom);
@@ -1151,15 +1196,17 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
       setOperationAction(ISD::MUL,             MVT::v8i32, Custom);
       setOperationAction(ISD::MUL,             MVT::v16i16, Custom);
       // Don't lower v32i8 because there is no 128-bit byte mul
+    }
 
-      setOperationAction(ISD::SRL,             MVT::v4i64, Custom);
-      setOperationAction(ISD::SRL,             MVT::v8i32, Custom);
+    // In the customized shift lowering, the legal cases in AVX2 will be
+    // recognized.
+    setOperationAction(ISD::SRL,               MVT::v4i64, Custom);
+    setOperationAction(ISD::SRL,               MVT::v8i32, Custom);
 
-      setOperationAction(ISD::SHL,             MVT::v4i64, Custom);
-      setOperationAction(ISD::SHL,             MVT::v8i32, Custom);
+    setOperationAction(ISD::SHL,               MVT::v4i64, Custom);
+    setOperationAction(ISD::SHL,               MVT::v8i32, Custom);
 
-      setOperationAction(ISD::SRA,             MVT::v8i32, Custom);
-    }
+    setOperationAction(ISD::SRA,               MVT::v8i32, Custom);
 
     // Custom lower several nodes for 256-bit types.
     for (int i = MVT::FIRST_VECTOR_VALUETYPE;
@@ -1217,7 +1264,6 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
   setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
 
-
   // Only custom-lower 64-bit SADDO and friends on 64-bit because we don't
   // handle type legalization for these operations here.
   //
@@ -1246,6 +1292,19 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setLibcallName(RTLIB::SRA_I128, 0);
   }
 
+  // Combine sin / cos into one node or libcall if possible.
+  if (Subtarget->hasSinCos()) {
+    setLibcallName(RTLIB::SINCOS_F32, "sincosf");
+    setLibcallName(RTLIB::SINCOS_F64, "sincos");
+    if (Subtarget->isTargetDarwin()) {
+      // For MacOSX, we don't want to the normal expansion of a libcall to
+      // sincos. We want to issue a libcall to __sincos_stret to avoid memory
+      // traffic.
+      setOperationAction(ISD::FSINCOS, MVT::f64, Custom);
+      setOperationAction(ISD::FSINCOS, MVT::f32, Custom);
+    }
+  }
+
   // We have target-specific dag combine patterns for the following nodes:
   setTargetDAGCombine(ISD::VECTOR_SHUFFLE);
   setTargetDAGCombine(ISD::EXTRACT_VECTOR_ELT);
@@ -1266,6 +1325,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
   setTargetDAGCombine(ISD::ZERO_EXTEND);
   setTargetDAGCombine(ISD::ANY_EXTEND);
   setTargetDAGCombine(ISD::SIGN_EXTEND);
+  setTargetDAGCombine(ISD::SIGN_EXTEND_INREG);
   setTargetDAGCombine(ISD::TRUNCATE);
   setTargetDAGCombine(ISD::SINT_TO_FP);
   setTargetDAGCombine(ISD::SETCC);
@@ -1277,28 +1337,25 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
 
   // On Darwin, -Os means optimize for size without hurting performance,
   // do not reduce the limit.
-  maxStoresPerMemset = 16; // For @llvm.memset -> sequence of stores
-  maxStoresPerMemsetOptSize = Subtarget->isTargetDarwin() ? 16 : 8;
-  maxStoresPerMemcpy = 8; // For @llvm.memcpy -> sequence of stores
-  maxStoresPerMemcpyOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
-  maxStoresPerMemmove = 8; // For @llvm.memmove -> sequence of stores
-  maxStoresPerMemmoveOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
+  MaxStoresPerMemset = 16; // For @llvm.memset -> sequence of stores
+  MaxStoresPerMemsetOptSize = Subtarget->isTargetDarwin() ? 16 : 8;
+  MaxStoresPerMemcpy = 8; // For @llvm.memcpy -> sequence of stores
+  MaxStoresPerMemcpyOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
+  MaxStoresPerMemmove = 8; // For @llvm.memmove -> sequence of stores
+  MaxStoresPerMemmoveOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
   setPrefLoopAlignment(4); // 2^4 bytes.
-  benefitFromCodePlacementOpt = true;
 
   // Predictable cmov don't hurt on atom because it's in-order.
-  predictableSelectIsExpensive = !Subtarget->isAtom();
+  PredictableSelectIsExpensive = !Subtarget->isAtom();
 
   setPrefFunctionAlignment(4); // 2^4 bytes.
 }
 
-
 EVT X86TargetLowering::getSetCCResultType(EVT VT) const {
   if (!VT.isVector()) return MVT::i8;
   return VT.changeVectorElementTypeToInteger();
 }
 
-
 /// getMaxByValAlign - Helper for getByValTypeAlignment to determine
 /// the desired ByVal argument alignment.
 static void getMaxByValAlign(Type *Ty, unsigned &MaxAlign) {
@@ -1348,34 +1405,30 @@ unsigned X86TargetLowering::getByValTypeAlignment(Type *Ty) const {
 /// lowering. If DstAlign is zero that means it's safe to destination
 /// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
 /// means there isn't a need to check it against alignment requirement,
-/// probably because the source does not need to be loaded. If
-/// 'IsZeroVal' is true, that means it's safe to return a
-/// non-scalar-integer type, e.g. empty string source, constant, or loaded
-/// from memory. 'MemcpyStrSrc' indicates whether the memcpy source is
-/// constant so it does not need to be loaded.
+/// probably because the source does not need to be loaded. If 'IsMemset' is
+/// true, that means it's expanding a memset. If 'ZeroMemset' is true, that
+/// means it's a memset of zero. 'MemcpyStrSrc' indicates whether the memcpy
+/// source is constant so it does not need to be loaded.
 /// It returns EVT::Other if the type should be determined using generic
 /// target-independent logic.
 EVT
 X86TargetLowering::getOptimalMemOpType(uint64_t Size,
                                        unsigned DstAlign, unsigned SrcAlign,
-                                       bool IsZeroVal,
+                                       bool IsMemset, bool ZeroMemset,
                                        bool MemcpyStrSrc,
                                        MachineFunction &MF) const {
-  // FIXME: This turns off use of xmm stores for memset/memcpy on targets like
-  // linux.  This is because the stack realignment code can't handle certain
-  // cases like PR2962.  This should be removed when PR2962 is fixed.
   const Function *F = MF.getFunction();
-  if (IsZeroVal &&
-      !F->getFnAttributes().hasAttribute(Attributes::NoImplicitFloat)) {
+  if ((!IsMemset || ZeroMemset) &&
+      !F->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
+                                       Attribute::NoImplicitFloat)) {
     if (Size >= 16 &&
         (Subtarget->isUnalignedMemAccessFast() ||
          ((DstAlign == 0 || DstAlign >= 16) &&
-          (SrcAlign == 0 || SrcAlign >= 16))) &&
-        Subtarget->getStackAlignment() >= 16) {
-      if (Subtarget->getStackAlignment() >= 32) {
-        if (Subtarget->hasAVX2())
+          (SrcAlign == 0 || SrcAlign >= 16)))) {
+      if (Size >= 32) {
+        if (Subtarget->hasInt256())
           return MVT::v8i32;
-        if (Subtarget->hasAVX())
+        if (Subtarget->hasFp256())
           return MVT::v8f32;
       }
       if (Subtarget->hasSSE2())
@@ -1384,7 +1437,6 @@ X86TargetLowering::getOptimalMemOpType(uint64_t Size,
         return MVT::v4f32;
     } else if (!MemcpyStrSrc && Size >= 8 &&
                !Subtarget->is64Bit() &&
-               Subtarget->getStackAlignment() >= 8 &&
                Subtarget->hasSSE2()) {
       // Do not use f64 to lower memcpy if source is string constant. It's
       // better to use i32 to avoid the loads.
@@ -1396,6 +1448,21 @@ X86TargetLowering::getOptimalMemOpType(uint64_t Size,
   return MVT::i32;
 }
 
+bool X86TargetLowering::isSafeMemOpType(MVT VT) const {
+  if (VT == MVT::f32)
+    return X86ScalarSSEf32;
+  else if (VT == MVT::f64)
+    return X86ScalarSSEf64;
+  return true;
+}
+
+bool
+X86TargetLowering::allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const {
+  if (Fast)
+    *Fast = Subtarget->isUnalignedMemAccessFast();
+  return true;
+}
+
 /// getJumpTableEncoding - Return the entry encoding for a jump table in the
 /// current function.  The returned value is a member of the
 /// MachineJumpTableInfo::JTEntryKind enum.
@@ -1449,10 +1516,10 @@ getPICJumpTableRelocBaseExpr(const MachineFunction *MF, unsigned JTI,
 
 // FIXME: Why this routine is here? Move to RegInfo!
 std::pair<const TargetRegisterClass*, uint8_t>
-X86TargetLowering::findRepresentativeClass(EVT VT) const{
+X86TargetLowering::findRepresentativeClass(MVT VT) const{
   const TargetRegisterClass *RRC = 0;
   uint8_t Cost = 1;
-  switch (VT.getSimpleVT().SimpleTy) {
+  switch (VT.SimpleTy) {
   default:
     return TargetLowering::findRepresentativeClass(VT);
   case MVT::i8: case MVT::i16: case MVT::i32: case MVT::i64:
@@ -1494,7 +1561,6 @@ bool X86TargetLowering::getStackCookieLocation(unsigned &AddressSpace,
   return true;
 }
 
-
 //===----------------------------------------------------------------------===//
 //               Return Value Calling Convention Implementation
 //===----------------------------------------------------------------------===//
@@ -1526,14 +1592,7 @@ X86TargetLowering::LowerReturn(SDValue Chain,
                  RVLocs, *DAG.getContext());
   CCInfo.AnalyzeReturn(Outs, RetCC_X86);
 
-  // Add the regs to the liveout set for the function.
-  MachineRegisterInfo &MRI = DAG.getMachineFunction().getRegInfo();
-  for (unsigned i = 0; i != RVLocs.size(); ++i)
-    if (RVLocs[i].isRegLoc() && !MRI.isLiveOut(RVLocs[i].getLocReg()))
-      MRI.addLiveOut(RVLocs[i].getLocReg());
-
   SDValue Flag;
-
   SmallVector<SDValue, 6> RetOps;
   RetOps.push_back(Chain); // Operand #0 = Chain (updated below)
   // Operand #1 = Bytes To Pop
@@ -1602,14 +1661,16 @@ X86TargetLowering::LowerReturn(SDValue Chain,
 
     Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), ValToCopy, Flag);
     Flag = Chain.getValue(1);
+    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
   }
 
-  // The x86-64 ABI for returning structs by value requires that we copy
-  // the sret argument into %rax for the return. We saved the argument into
-  // a virtual register in the entry block, so now we copy the value out
-  // and into %rax.
-  if (Subtarget->is64Bit() &&
-      DAG.getMachineFunction().getFunction()->hasStructRetAttr()) {
+  // The x86-64 ABIs require that for returning structs by value we copy
+  // the sret argument into %rax/%eax (depending on ABI) for the return.
+  // Win32 requires us to put the sret argument to %eax as well.
+  // We saved the argument into a virtual register in the entry block,
+  // so now we copy the value out and into %rax/%eax.
+  if (DAG.getMachineFunction().getFunction()->hasStructRetAttr() &&
+      (Subtarget->is64Bit() || Subtarget->isTargetWindows())) {
     MachineFunction &MF = DAG.getMachineFunction();
     X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
     unsigned Reg = FuncInfo->getSRetReturnReg();
@@ -1617,11 +1678,14 @@ X86TargetLowering::LowerReturn(SDValue Chain,
            "SRetReturnReg should have been set in LowerFormalArguments().");
     SDValue Val = DAG.getCopyFromReg(Chain, dl, Reg, getPointerTy());
 
-    Chain = DAG.getCopyToReg(Chain, dl, X86::RAX, Val, Flag);
+    unsigned RetValReg
+        = (Subtarget->is64Bit() && !Subtarget->isTarget64BitILP32()) ?
+          X86::RAX : X86::EAX;
+    Chain = DAG.getCopyToReg(Chain, dl, RetValReg, Val, Flag);
     Flag = Chain.getValue(1);
 
-    // RAX now acts like a return value.
-    MRI.addLiveOut(X86::RAX);
+    // RAX/EAX now acts like a return value.
+    RetOps.push_back(DAG.getRegister(RetValReg, getPointerTy()));
   }
 
   RetOps[0] = Chain;  // Update chain.
@@ -1666,8 +1730,8 @@ bool X86TargetLowering::isUsedByReturnOnly(SDNode *N, SDValue &Chain) const {
   return true;
 }
 
-EVT
-X86TargetLowering::getTypeForExtArgOrReturn(LLVMContext &Context, EVT VT,
+MVT
+X86TargetLowering::getTypeForExtArgOrReturn(MVT VT,
                                             ISD::NodeType ExtendKind) const {
   MVT ReturnMVT;
   // TODO: Is this also valid on 32-bit?
@@ -1676,7 +1740,7 @@ X86TargetLowering::getTypeForExtArgOrReturn(LLVMContext &Context, EVT VT,
   else
     ReturnMVT = MVT::i32;
 
-  EVT MinVT = getRegisterType(Context, ReturnMVT);
+  MVT MinVT = getRegisterType(ReturnMVT);
   return VT.bitsLT(MinVT) ? MinVT : VT;
 }
 
@@ -1698,7 +1762,7 @@ X86TargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
   CCInfo.AnalyzeCallResult(Ins, RetCC_X86);
 
   // Copy all of the result registers out of their specified physreg.
-  for (unsigned i = 0; i != RVLocs.size(); ++i) {
+  for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
     CCValAssign &VA = RVLocs[i];
     EVT CopyVT = VA.getValVT();
 
@@ -1742,7 +1806,6 @@ X86TargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
   return Chain;
 }
 
-
 //===----------------------------------------------------------------------===//
 //                C & StdCall & Fast Calling Convention implementation
 //===----------------------------------------------------------------------===//
@@ -1806,7 +1869,8 @@ CreateCopyOfByValArgument(SDValue Src, SDValue Dst, SDValue Chain,
 /// IsTailCallConvention - Return true if the calling convention is one that
 /// supports tail call optimization.
 static bool IsTailCallConvention(CallingConv::ID CC) {
-  return (CC == CallingConv::Fast || CC == CallingConv::GHC);
+  return (CC == CallingConv::Fast || CC == CallingConv::GHC ||
+          CC == CallingConv::HiPE);
 }
 
 bool X86TargetLowering::mayBeEmittedAsTailCall(CallInst *CI) const {
@@ -1893,7 +1957,7 @@ X86TargetLowering::LowerFormalArguments(SDValue Chain,
   bool IsWin64 = Subtarget->isTargetWin64();
 
   assert(!(isVarArg && IsTailCallConvention(CallConv)) &&
-         "Var args not supported with calling convention fastcc or ghc");
+         "Var args not supported with calling convention fastcc, ghc or hipe");
 
   // Assign locations to all of the incoming arguments.
   SmallVector<CCValAssign, 16> ArgLocs;
@@ -1955,10 +2019,9 @@ X86TargetLowering::LowerFormalArguments(SDValue Chain,
 
       if (VA.isExtInLoc()) {
         // Handle MMX values passed in XMM regs.
-        if (RegVT.isVector()) {
-          ArgValue = DAG.getNode(X86ISD::MOVDQ2Q, dl, VA.getValVT(),
-                                 ArgValue);
-        } else
+        if (RegVT.isVector())
+          ArgValue = DAG.getNode(X86ISD::MOVDQ2Q, dl, VA.getValVT(), ArgValue);
+        else
           ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
       }
     } else {
@@ -1974,14 +2037,18 @@ X86TargetLowering::LowerFormalArguments(SDValue Chain,
     InVals.push_back(ArgValue);
   }
 
-  // The x86-64 ABI for returning structs by value requires that we copy
-  // the sret argument into %rax for the return. Save the argument into
-  // a virtual register so that we can access it from the return points.
-  if (Is64Bit && MF.getFunction()->hasStructRetAttr()) {
+  // The x86-64 ABIs require that for returning structs by value we copy
+  // the sret argument into %rax/%eax (depending on ABI) for the return.
+  // Win32 requires us to put the sret argument to %eax as well.
+  // Save the argument into a virtual register so that we can access it
+  // from the return points.
+  if (MF.getFunction()->hasStructRetAttr() &&
+      (Subtarget->is64Bit() || Subtarget->isTargetWindows())) {
     X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
     unsigned Reg = FuncInfo->getSRetReturnReg();
     if (!Reg) {
-      Reg = MF.getRegInfo().createVirtualRegister(getRegClassFor(MVT::i64));
+      MVT PtrTy = getPointerTy();
+      Reg = MF.getRegInfo().createVirtualRegister(getRegClassFor(PtrTy));
       FuncInfo->setSRetReturnReg(Reg);
     }
     SDValue Copy = DAG.getCopyToReg(DAG.getEntryNode(), dl, Reg, InVals[0]);
@@ -2034,8 +2101,8 @@ X86TargetLowering::LowerFormalArguments(SDValue Chain,
       unsigned NumIntRegs = CCInfo.getFirstUnallocated(GPR64ArgRegs,
                                                        TotalNumIntRegs);
 
-      bool NoImplicitFloatOps = Fn->getFnAttributes().
-        hasAttribute(Attributes::NoImplicitFloat);
+      bool NoImplicitFloatOps = Fn->getAttributes().
+        hasAttribute(AttributeSet::FunctionIndex, Attribute::NoImplicitFloat);
       assert(!(NumXMMRegs && !Subtarget->hasSSE1()) &&
              "SSE register cannot be used when SSE is disabled!");
       assert(!(NumXMMRegs && MF.getTarget().Options.UseSoftFloat &&
@@ -2238,7 +2305,7 @@ X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   }
 
   assert(!(isVarArg && IsTailCallConvention(CallConv)) &&
-         "Var args not supported with calling convention fastcc or ghc");
+         "Var args not supported with calling convention fastcc, ghc or hipe");
 
   // Analyze operands of the call, assigning locations to each operand.
   SmallVector<CCValAssign, 16> ArgLocs;
@@ -2513,8 +2580,9 @@ X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
         OpFlags = X86II::MO_DARWIN_STUB;
       } else if (Subtarget->isPICStyleRIPRel() &&
                  isa<Function>(GV) &&
-                 cast<Function>(GV)->getFnAttributes().
-                   hasAttribute(Attributes::NonLazyBind)) {
+                 cast<Function>(GV)->getAttributes().
+                   hasAttribute(AttributeSet::FunctionIndex,
+                                Attribute::NonLazyBind)) {
         // If the function is marked as non-lazy, generate an indirect call
         // which loads from the GOT directly. This avoids runtime overhead
         // at the cost of eager binding (and one extra byte of encoding).
@@ -2594,8 +2662,7 @@ X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     // This isn't right, although it's probably harmless on x86; liveouts
     // should be computed from returns not tail calls.  Consider a void
     // function making a tail call to a function returning int.
-    return DAG.getNode(X86ISD::TC_RETURN, dl,
-                       NodeTys, &Ops[0], Ops.size());
+    return DAG.getNode(X86ISD::TC_RETURN, dl, NodeTys, &Ops[0], Ops.size());
   }
 
   Chain = DAG.getNode(X86ISD::CALL, dl, NodeTys, &Ops[0], Ops.size());
@@ -2632,7 +2699,6 @@ X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
                          Ins, dl, DAG, InVals);
 }
 
-
 //===----------------------------------------------------------------------===//
 //                Fast Calling Convention (tail call) implementation
 //===----------------------------------------------------------------------===//
@@ -2754,7 +2820,7 @@ X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
                                     const SmallVectorImpl<ISD::OutputArg> &Outs,
                                     const SmallVectorImpl<SDValue> &OutVals,
                                     const SmallVectorImpl<ISD::InputArg> &Ins,
-                                                     SelectionDAG& DAG) const {
+                                                     SelectionDAG &DAG) const {
   if (!IsTailCallConvention(CalleeCC) &&
       CalleeCC != CallingConv::C)
     return false;
@@ -2793,7 +2859,7 @@ X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
 
   // An stdcall caller is expected to clean up its arguments; the callee
   // isn't going to do that.
-  if (!CCMatch && CallerCC==CallingConv::X86_StdCall)
+  if (!CCMatch && CallerCC == CallingConv::X86_StdCall)
     return false;
 
   // Do not sibcall optimize vararg calls unless all arguments are passed via
@@ -2913,9 +2979,15 @@ X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
     // callee-saved registers are restored. These happen to be the same
     // registers used to pass 'inreg' arguments so watch out for those.
     if (!Subtarget->is64Bit() &&
-        !isa<GlobalAddressSDNode>(Callee) &&
-        !isa<ExternalSymbolSDNode>(Callee)) {
+        ((!isa<GlobalAddressSDNode>(Callee) &&
+          !isa<ExternalSymbolSDNode>(Callee)) ||
+         getTargetMachine().getRelocationModel() == Reloc::PIC_)) {
       unsigned NumInRegs = 0;
+      // In PIC we need an extra register to formulate the address computation
+      // for the callee.
+      unsigned MaxInRegs =
+          (getTargetMachine().getRelocationModel() == Reloc::PIC_) ? 2 : 3;
+
       for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
         CCValAssign &VA = ArgLocs[i];
         if (!VA.isRegLoc())
@@ -2924,7 +2996,7 @@ X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
         switch (Reg) {
         default: break;
         case X86::EAX: case X86::EDX: case X86::ECX:
-          if (++NumInRegs == 3)
+          if (++NumInRegs == MaxInRegs)
             return false;
           break;
         }
@@ -2941,7 +3013,6 @@ X86TargetLowering::createFastISel(FunctionLoweringInfo &funcInfo,
   return X86::createFastISel(funcInfo, libInfo);
 }
 
-
 //===----------------------------------------------------------------------===//
 //                           Other Lowering Hooks
 //===----------------------------------------------------------------------===//
@@ -2961,7 +3032,7 @@ static bool isTargetShuffle(unsigned Opcode) {
   case X86ISD::PSHUFHW:
   case X86ISD::PSHUFLW:
   case X86ISD::SHUFP:
-  case X86ISD::PALIGN:
+  case X86ISD::PALIGNR:
   case X86ISD::MOVLHPS:
   case X86ISD::MOVLHPD:
   case X86ISD::MOVHLPS:
@@ -3011,7 +3082,7 @@ static SDValue getTargetShuffleNode(unsigned Opc, DebugLoc dl, EVT VT,
                                     SelectionDAG &DAG) {
   switch(Opc) {
   default: llvm_unreachable("Unknown x86 shuffle node");
-  case X86ISD::PALIGN:
+  case X86ISD::PALIGNR:
   case X86ISD::SHUFP:
   case X86ISD::VPERM2X128:
     return DAG.getNode(Opc, dl, VT, V1, V2,
@@ -3052,7 +3123,6 @@ SDValue X86TargetLowering::getReturnAddressFrameIndex(SelectionDAG &DAG) const {
   return DAG.getFrameIndex(ReturnAddrIndex, getPointerTy());
 }
 
-
 bool X86::isOffsetSuitableForCodeModel(int64_t Offset, CodeModel::Model M,
                                        bool hasSymbolicDisplacement) {
   // Offset should fit into 32 bit immediate field.
@@ -3103,6 +3173,8 @@ bool X86::isCalleePop(CallingConv::ID CallingConv,
     return TailCallOpt;
   case CallingConv::GHC:
     return TailCallOpt;
+  case CallingConv::HiPE:
+    return TailCallOpt;
   }
 }
 
@@ -3233,9 +3305,7 @@ static bool isUndefOrInRange(int Val, int Low, int Hi) {
 /// isUndefOrEqual - Val is either less than zero (undef) or equal to the
 /// specified value.
 static bool isUndefOrEqual(int Val, int CmpVal) {
-  if (Val < 0 || Val == CmpVal)
-    return true;
-  return false;
+  return (Val < 0 || Val == CmpVal);
 }
 
 /// isSequentialOrUndefInRange - Return true if every element in Mask, beginning
@@ -3262,8 +3332,8 @@ static bool isPSHUFDMask(ArrayRef<int> Mask, EVT VT) {
 
 /// isPSHUFHWMask - Return true if the node specifies a shuffle of elements that
 /// is suitable for input to PSHUFHW.
-static bool isPSHUFHWMask(ArrayRef<int> Mask, EVT VT, bool HasAVX2) {
-  if (VT != MVT::v8i16 && (!HasAVX2 || VT != MVT::v16i16))
+static bool isPSHUFHWMask(ArrayRef<int> Mask, EVT VT, bool HasInt256) {
+  if (VT != MVT::v8i16 && (!HasInt256 || VT != MVT::v16i16))
     return false;
 
   // Lower quadword copied in order or undef.
@@ -3291,8 +3361,8 @@ static bool isPSHUFHWMask(ArrayRef<int> Mask, EVT VT, bool HasAVX2) {
 
 /// isPSHUFLWMask - Return true if the node specifies a shuffle of elements that
 /// is suitable for input to PSHUFLW.
-static bool isPSHUFLWMask(ArrayRef<int> Mask, EVT VT, bool HasAVX2) {
-  if (VT != MVT::v8i16 && (!HasAVX2 || VT != MVT::v16i16))
+static bool isPSHUFLWMask(ArrayRef<int> Mask, EVT VT, bool HasInt256) {
+  if (VT != MVT::v8i16 && (!HasInt256 || VT != MVT::v16i16))
     return false;
 
   // Upper quadword copied in order.
@@ -3322,8 +3392,8 @@ static bool isPSHUFLWMask(ArrayRef<int> Mask, EVT VT, bool HasAVX2) {
 /// is suitable for input to PALIGNR.
 static bool isPALIGNRMask(ArrayRef<int> Mask, EVT VT,
                           const X86Subtarget *Subtarget) {
-  if ((VT.getSizeInBits() == 128 && !Subtarget->hasSSSE3()) ||
-      (VT.getSizeInBits() == 256 && !Subtarget->hasAVX2()))
+  if ((VT.is128BitVector() && !Subtarget->hasSSSE3()) ||
+      (VT.is256BitVector() && !Subtarget->hasInt256()))
     return false;
 
   unsigned NumElts = VT.getVectorNumElements();
@@ -3410,9 +3480,9 @@ static void CommuteVectorShuffleMask(SmallVectorImpl<int> &Mask,
 /// specifies a shuffle of elements that is suitable for input to 128/256-bit
 /// SHUFPS and SHUFPD. If Commuted is true, then it checks for sources to be
 /// reverse of what x86 shuffles want.
-static bool isSHUFPMask(ArrayRef<int> Mask, EVT VT, bool HasAVX,
+static bool isSHUFPMask(ArrayRef<int> Mask, EVT VT, bool HasFp256,
                         bool Commuted = false) {
-  if (!HasAVX && VT.getSizeInBits() == 256)
+  if (!HasFp256 && VT.is256BitVector())
     return false;
 
   unsigned NumElems = VT.getVectorNumElements();
@@ -3547,7 +3617,7 @@ static bool isMOVLHPSMask(ArrayRef<int> Mask, EVT VT) {
 static
 SDValue Compact8x32ShuffleNode(ShuffleVectorSDNode *SVOp,
                                SelectionDAG &DAG) {
-  EVT VT = SVOp->getValueType(0);
+  MVT VT = SVOp->getValueType(0).getSimpleVT();
   DebugLoc dl = SVOp->getDebugLoc();
 
   if (VT != MVT::v8i32 && VT != MVT::v8f32)
@@ -3591,14 +3661,14 @@ SDValue Compact8x32ShuffleNode(ShuffleVectorSDNode *SVOp,
 /// isUNPCKLMask - Return true if the specified VECTOR_SHUFFLE operand
 /// specifies a shuffle of elements that is suitable for input to UNPCKL.
 static bool isUNPCKLMask(ArrayRef<int> Mask, EVT VT,
-                         bool HasAVX2, bool V2IsSplat = false) {
+                         bool HasInt256, bool V2IsSplat = false) {
   unsigned NumElts = VT.getVectorNumElements();
 
   assert((VT.is128BitVector() || VT.is256BitVector()) &&
          "Unsupported vector type for unpckh");
 
-  if (VT.getSizeInBits() == 256 && NumElts != 4 && NumElts != 8 &&
-      (!HasAVX2 || (NumElts != 16 && NumElts != 32)))
+  if (VT.is256BitVector() && NumElts != 4 && NumElts != 8 &&
+      (!HasInt256 || (NumElts != 16 && NumElts != 32)))
     return false;
 
   // Handle 128 and 256-bit vector lengths. AVX defines UNPCK* to operate
@@ -3630,14 +3700,14 @@ static bool isUNPCKLMask(ArrayRef<int> Mask, EVT VT,
 /// isUNPCKHMask - Return true if the specified VECTOR_SHUFFLE operand
 /// specifies a shuffle of elements that is suitable for input to UNPCKH.
 static bool isUNPCKHMask(ArrayRef<int> Mask, EVT VT,
-                         bool HasAVX2, bool V2IsSplat = false) {
+                         bool HasInt256, bool V2IsSplat = false) {
   unsigned NumElts = VT.getVectorNumElements();
 
   assert((VT.is128BitVector() || VT.is256BitVector()) &&
          "Unsupported vector type for unpckh");
 
-  if (VT.getSizeInBits() == 256 && NumElts != 4 && NumElts != 8 &&
-      (!HasAVX2 || (NumElts != 16 && NumElts != 32)))
+  if (VT.is256BitVector() && NumElts != 4 && NumElts != 8 &&
+      (!HasInt256 || (NumElts != 16 && NumElts != 32)))
     return false;
 
   // Handle 128 and 256-bit vector lengths. AVX defines UNPCK* to operate
@@ -3667,22 +3737,22 @@ static bool isUNPCKHMask(ArrayRef<int> Mask, EVT VT,
 /// isUNPCKL_v_undef_Mask - Special case of isUNPCKLMask for canonical form
 /// of vector_shuffle v, v, <0, 4, 1, 5>, i.e. vector_shuffle v, undef,
 /// <0, 0, 1, 1>
-static bool isUNPCKL_v_undef_Mask(ArrayRef<int> Mask, EVT VT,
-                                  bool HasAVX2) {
+static bool isUNPCKL_v_undef_Mask(ArrayRef<int> Mask, EVT VT, bool HasInt256) {
   unsigned NumElts = VT.getVectorNumElements();
+  bool Is256BitVec = VT.is256BitVector();
 
   assert((VT.is128BitVector() || VT.is256BitVector()) &&
          "Unsupported vector type for unpckh");
 
-  if (VT.getSizeInBits() == 256 && NumElts != 4 && NumElts != 8 &&
-      (!HasAVX2 || (NumElts != 16 && NumElts != 32)))
+  if (Is256BitVec && NumElts != 4 && NumElts != 8 &&
+      (!HasInt256 || (NumElts != 16 && NumElts != 32)))
     return false;
 
   // For 256-bit i64/f64, use MOVDDUPY instead, so reject the matching pattern
   // FIXME: Need a better way to get rid of this, there's no latency difference
   // between UNPCKLPD and MOVDDUP, the later should always be checked first and
   // the former later. We should also remove the "_undef" special mask.
-  if (NumElts == 4 && VT.getSizeInBits() == 256)
+  if (NumElts == 4 && Is256BitVec)
     return false;
 
   // Handle 128 and 256-bit vector lengths. AVX defines UNPCK* to operate
@@ -3710,14 +3780,14 @@ static bool isUNPCKL_v_undef_Mask(ArrayRef<int> Mask, EVT VT,
 /// isUNPCKH_v_undef_Mask - Special case of isUNPCKHMask for canonical form
 /// of vector_shuffle v, v, <2, 6, 3, 7>, i.e. vector_shuffle v, undef,
 /// <2, 2, 3, 3>
-static bool isUNPCKH_v_undef_Mask(ArrayRef<int> Mask, EVT VT, bool HasAVX2) {
+static bool isUNPCKH_v_undef_Mask(ArrayRef<int> Mask, EVT VT, bool HasInt256) {
   unsigned NumElts = VT.getVectorNumElements();
 
   assert((VT.is128BitVector() || VT.is256BitVector()) &&
          "Unsupported vector type for unpckh");
 
-  if (VT.getSizeInBits() == 256 && NumElts != 4 && NumElts != 8 &&
-      (!HasAVX2 || (NumElts != 16 && NumElts != 32)))
+  if (VT.is256BitVector() && NumElts != 4 && NumElts != 8 &&
+      (!HasInt256 || (NumElts != 16 && NumElts != 32)))
     return false;
 
   // Handle 128 and 256-bit vector lengths. AVX defines UNPCK* to operate
@@ -3766,8 +3836,8 @@ static bool isMOVLMask(ArrayRef<int> Mask, EVT VT) {
 ///   vector_shuffle <4, 5, 6, 7, 12, 13, 14, 15>
 /// The first half comes from the second half of V1 and the second half from the
 /// the second half of V2.
-static bool isVPERM2X128Mask(ArrayRef<int> Mask, EVT VT, bool HasAVX) {
-  if (!HasAVX || !VT.is256BitVector())
+static bool isVPERM2X128Mask(ArrayRef<int> Mask, EVT VT, bool HasFp256) {
+  if (!HasFp256 || !VT.is256BitVector())
     return false;
 
   // The shuffle result is divided into half A and half B. In total the two
@@ -3798,7 +3868,7 @@ static bool isVPERM2X128Mask(ArrayRef<int> Mask, EVT VT, bool HasAVX) {
 /// getShuffleVPERM2X128Immediate - Return the appropriate immediate to shuffle
 /// the specified VECTOR_MASK mask with VPERM2F128/VPERM2I128 instructions.
 static unsigned getShuffleVPERM2X128Immediate(ShuffleVectorSDNode *SVOp) {
-  EVT VT = SVOp->getValueType(0);
+  MVT VT = SVOp->getValueType(0).getSimpleVT();
 
   unsigned HalfSize = VT.getVectorNumElements()/2;
 
@@ -3826,13 +3896,13 @@ static unsigned getShuffleVPERM2X128Immediate(ShuffleVectorSDNode *SVOp) {
 /// to the same elements of the low, but to the higher half of the source.
 /// In VPERMILPD the two lanes could be shuffled independently of each other
 /// with the same restriction that lanes can't be crossed. Also handles PSHUFDY.
-static bool isVPERMILPMask(ArrayRef<int> Mask, EVT VT, bool HasAVX) {
-  if (!HasAVX)
+static bool isVPERMILPMask(ArrayRef<int> Mask, EVT VT, bool HasFp256) {
+  if (!HasFp256)
     return false;
 
   unsigned NumElts = VT.getVectorNumElements();
   // Only match 256-bit with 32/64-bit types
-  if (VT.getSizeInBits() != 256 || (NumElts != 4 && NumElts != 8))
+  if (!VT.is256BitVector() || (NumElts != 4 && NumElts != 8))
     return false;
 
   unsigned NumLanes = VT.getSizeInBits()/128;
@@ -3888,8 +3958,8 @@ static bool isMOVSHDUPMask(ArrayRef<int> Mask, EVT VT,
 
   unsigned NumElems = VT.getVectorNumElements();
 
-  if ((VT.getSizeInBits() == 128 && NumElems != 4) ||
-      (VT.getSizeInBits() == 256 && NumElems != 8))
+  if ((VT.is128BitVector() && NumElems != 4) ||
+      (VT.is256BitVector() && NumElems != 8))
     return false;
 
   // "i+1" is the value the indexed mask element must have
@@ -3911,8 +3981,8 @@ static bool isMOVSLDUPMask(ArrayRef<int> Mask, EVT VT,
 
   unsigned NumElems = VT.getVectorNumElements();
 
-  if ((VT.getSizeInBits() == 128 && NumElems != 4) ||
-      (VT.getSizeInBits() == 256 && NumElems != 8))
+  if ((VT.is128BitVector() && NumElems != 4) ||
+      (VT.is256BitVector() && NumElems != 8))
     return false;
 
   // "i" is the value the indexed mask element must have
@@ -3927,8 +3997,8 @@ static bool isMOVSLDUPMask(ArrayRef<int> Mask, EVT VT,
 /// isMOVDDUPYMask - Return true if the specified VECTOR_SHUFFLE operand
 /// specifies a shuffle of elements that is suitable for input to 256-bit
 /// version of MOVDDUP.
-static bool isMOVDDUPYMask(ArrayRef<int> Mask, EVT VT, bool HasAVX) {
-  if (!HasAVX || !VT.is256BitVector())
+static bool isMOVDDUPYMask(ArrayRef<int> Mask, EVT VT, bool HasFp256) {
+  if (!HasFp256 || !VT.is256BitVector())
     return false;
 
   unsigned NumElts = VT.getVectorNumElements();
@@ -3972,9 +4042,8 @@ bool X86::isVEXTRACTF128Index(SDNode *N) {
   uint64_t Index =
     cast<ConstantSDNode>(N->getOperand(1).getNode())->getZExtValue();
 
-  unsigned VL = N->getValueType(0).getVectorNumElements();
-  unsigned VBits = N->getValueType(0).getSizeInBits();
-  unsigned ElSize = VBits / VL;
+  MVT VT = N->getValueType(0).getSimpleVT();
+  unsigned ElSize = VT.getVectorElementType().getSizeInBits();
   bool Result = (Index * ElSize) % 128 == 0;
 
   return Result;
@@ -3991,9 +4060,8 @@ bool X86::isVINSERTF128Index(SDNode *N) {
   uint64_t Index =
     cast<ConstantSDNode>(N->getOperand(2).getNode())->getZExtValue();
 
-  unsigned VL = N->getValueType(0).getVectorNumElements();
-  unsigned VBits = N->getValueType(0).getSizeInBits();
-  unsigned ElSize = VBits / VL;
+  MVT VT = N->getValueType(0).getSimpleVT();
+  unsigned ElSize = VT.getVectorElementType().getSizeInBits();
   bool Result = (Index * ElSize) % 128 == 0;
 
   return Result;
@@ -4003,7 +4071,7 @@ bool X86::isVINSERTF128Index(SDNode *N) {
 /// the specified VECTOR_SHUFFLE mask with PSHUF* and SHUFP* instructions.
 /// Handles 128-bit and 256-bit.
 static unsigned getShuffleSHUFImmediate(ShuffleVectorSDNode *N) {
-  EVT VT = N->getValueType(0);
+  MVT VT = N->getValueType(0).getSimpleVT();
 
   assert((VT.is128BitVector() || VT.is256BitVector()) &&
          "Unsupported vector type for PSHUF/SHUFP");
@@ -4033,7 +4101,7 @@ static unsigned getShuffleSHUFImmediate(ShuffleVectorSDNode *N) {
 /// getShufflePSHUFHWImmediate - Return the appropriate immediate to shuffle
 /// the specified VECTOR_SHUFFLE mask with the PSHUFHW instruction.
 static unsigned getShufflePSHUFHWImmediate(ShuffleVectorSDNode *N) {
-  EVT VT = N->getValueType(0);
+  MVT VT = N->getValueType(0).getSimpleVT();
 
   assert((VT == MVT::v8i16 || VT == MVT::v16i16) &&
          "Unsupported vector type for PSHUFHW");
@@ -4057,7 +4125,7 @@ static unsigned getShufflePSHUFHWImmediate(ShuffleVectorSDNode *N) {
 /// getShufflePSHUFLWImmediate - Return the appropriate immediate to shuffle
 /// the specified VECTOR_SHUFFLE mask with the PSHUFLW instruction.
 static unsigned getShufflePSHUFLWImmediate(ShuffleVectorSDNode *N) {
-  EVT VT = N->getValueType(0);
+  MVT VT = N->getValueType(0).getSimpleVT();
 
   assert((VT == MVT::v8i16 || VT == MVT::v16i16) &&
          "Unsupported vector type for PSHUFHW");
@@ -4081,7 +4149,7 @@ static unsigned getShufflePSHUFLWImmediate(ShuffleVectorSDNode *N) {
 /// getShufflePALIGNRImmediate - Return the appropriate immediate to shuffle
 /// the specified VECTOR_SHUFFLE mask with the PALIGNR instruction.
 static unsigned getShufflePALIGNRImmediate(ShuffleVectorSDNode *SVOp) {
-  EVT VT = SVOp->getValueType(0);
+  MVT VT = SVOp->getValueType(0).getSimpleVT();
   unsigned EltSize = VT.getVectorElementType().getSizeInBits() >> 3;
 
   unsigned NumElts = VT.getVectorNumElements();
@@ -4112,8 +4180,8 @@ unsigned X86::getExtractVEXTRACTF128Immediate(SDNode *N) {
   uint64_t Index =
     cast<ConstantSDNode>(N->getOperand(1).getNode())->getZExtValue();
 
-  EVT VecVT = N->getOperand(0).getValueType();
-  EVT ElVT = VecVT.getVectorElementType();
+  MVT VecVT = N->getOperand(0).getValueType().getSimpleVT();
+  MVT ElVT = VecVT.getVectorElementType();
 
   unsigned NumElemsPerChunk = 128 / ElVT.getSizeInBits();
   return Index / NumElemsPerChunk;
@@ -4129,8 +4197,8 @@ unsigned X86::getInsertVINSERTF128Immediate(SDNode *N) {
   uint64_t Index =
     cast<ConstantSDNode>(N->getOperand(2).getNode())->getZExtValue();
 
-  EVT VecVT = N->getValueType(0);
-  EVT ElVT = VecVT.getVectorElementType();
+  MVT VecVT = N->getValueType(0).getSimpleVT();
+  MVT ElVT = VecVT.getVectorElementType();
 
   unsigned NumElemsPerChunk = 128 / ElVT.getSizeInBits();
   return Index / NumElemsPerChunk;
@@ -4140,7 +4208,7 @@ unsigned X86::getInsertVINSERTF128Immediate(SDNode *N) {
 /// the specified VECTOR_SHUFFLE mask with VPERMQ and VPERMPD instructions.
 /// Handles 256-bit.
 static unsigned getShuffleCLImmediate(ShuffleVectorSDNode *N) {
-  EVT VT = N->getValueType(0);
+  MVT VT = N->getValueType(0).getSimpleVT();
 
   unsigned NumElts = VT.getVectorNumElements();
 
@@ -4160,17 +4228,18 @@ static unsigned getShuffleCLImmediate(ShuffleVectorSDNode *N) {
 /// isZeroNode - Returns true if Elt is a constant zero or a floating point
 /// constant +0.0.
 bool X86::isZeroNode(SDValue Elt) {
-  return ((isa<ConstantSDNode>(Elt) &&
-           cast<ConstantSDNode>(Elt)->isNullValue()) ||
-          (isa<ConstantFPSDNode>(Elt) &&
-           cast<ConstantFPSDNode>(Elt)->getValueAPF().isPosZero()));
+  if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Elt))
+    return CN->isNullValue();
+  if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(Elt))
+    return CFP->getValueAPF().isPosZero();
+  return false;
 }
 
 /// CommuteVectorShuffle - Swap vector_shuffle operands as well as values in
 /// their permute mask.
 static SDValue CommuteVectorShuffle(ShuffleVectorSDNode *SVOp,
                                     SelectionDAG &DAG) {
-  EVT VT = SVOp->getValueType(0);
+  MVT VT = SVOp->getValueType(0).getSimpleVT();
   unsigned NumElems = VT.getVectorNumElements();
   SmallVector<int, 8> MaskVec;
 
@@ -4319,12 +4388,11 @@ static bool isZeroShuffle(ShuffleVectorSDNode *N) {
 static SDValue getZeroVector(EVT VT, const X86Subtarget *Subtarget,
                              SelectionDAG &DAG, DebugLoc dl) {
   assert(VT.isVector() && "Expected a vector type");
-  unsigned Size = VT.getSizeInBits();
 
   // Always build SSE zero vectors as <4 x i32> bitcasted
   // to their dest type. This ensures they get CSE'd.
   SDValue Vec;
-  if (Size == 128) {  // SSE
+  if (VT.is128BitVector()) {  // SSE
     if (Subtarget->hasSSE2()) {  // SSE2
       SDValue Cst = DAG.getTargetConstant(0, MVT::i32);
       Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst);
@@ -4332,8 +4400,8 @@ static SDValue getZeroVector(EVT VT, const X86Subtarget *Subtarget,
       SDValue Cst = DAG.getTargetConstantFP(+0.0, MVT::f32);
       Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4f32, Cst, Cst, Cst, Cst);
     }
-  } else if (Size == 256) { // AVX
-    if (Subtarget->hasAVX2()) { // AVX2
+  } else if (VT.is256BitVector()) { // AVX
+    if (Subtarget->hasInt256()) { // AVX2
       SDValue Cst = DAG.getTargetConstant(0, MVT::i32);
       SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
       Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops, 8);
@@ -4354,22 +4422,21 @@ static SDValue getZeroVector(EVT VT, const X86Subtarget *Subtarget,
 /// Always build ones vectors as <4 x i32> or <8 x i32>. For 256-bit types with
 /// no AVX2 supprt, use two <4 x i32> inserted in a <8 x i32> appropriately.
 /// Then bitcast to their original type, ensuring they get CSE'd.
-static SDValue getOnesVector(EVT VT, bool HasAVX2, SelectionDAG &DAG,
+static SDValue getOnesVector(MVT VT, bool HasInt256, SelectionDAG &DAG,
                              DebugLoc dl) {
   assert(VT.isVector() && "Expected a vector type");
-  unsigned Size = VT.getSizeInBits();
 
   SDValue Cst = DAG.getTargetConstant(~0U, MVT::i32);
   SDValue Vec;
-  if (Size == 256) {
-    if (HasAVX2) { // AVX2
+  if (VT.is256BitVector()) {
+    if (HasInt256) { // AVX2
       SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
       Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops, 8);
     } else { // AVX
       Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst);
       Vec = Concat128BitVectors(Vec, Vec, MVT::v8i32, 8, DAG, dl);
     }
-  } else if (Size == 128) {
+  } else if (VT.is128BitVector()) {
     Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst);
   } else
     llvm_unreachable("Unexpected vector type");
@@ -4448,14 +4515,13 @@ static SDValue PromoteSplati8i16(SDValue V, SelectionDAG &DAG, int &EltNo) {
 static SDValue getLegalSplat(SelectionDAG &DAG, SDValue V, int EltNo) {
   EVT VT = V.getValueType();
   DebugLoc dl = V.getDebugLoc();
-  unsigned Size = VT.getSizeInBits();
 
-  if (Size == 128) {
+  if (VT.is128BitVector()) {
     V = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32, V);
     int SplatMask[4] = { EltNo, EltNo, EltNo, EltNo };
     V = DAG.getVectorShuffle(MVT::v4f32, dl, V, DAG.getUNDEF(MVT::v4f32),
                              &SplatMask[0]);
-  } else if (Size == 256) {
+  } else if (VT.is256BitVector()) {
     // To use VPERMILPS to splat scalars, the second half of indicies must
     // refer to the higher part, which is a duplication of the lower one,
     // because VPERMILPS can only handle in-lane permutations.
@@ -4479,14 +4545,14 @@ static SDValue PromoteSplat(ShuffleVectorSDNode *SV, SelectionDAG &DAG) {
 
   int EltNo = SV->getSplatIndex();
   int NumElems = SrcVT.getVectorNumElements();
-  unsigned Size = SrcVT.getSizeInBits();
+  bool Is256BitVec = SrcVT.is256BitVector();
 
-  assert(((Size == 128 && NumElems > 4) || Size == 256) &&
-          "Unknown how to promote splat for type");
+  assert(((SrcVT.is128BitVector() && NumElems > 4) || Is256BitVec) &&
+         "Unknown how to promote splat for type");
 
   // Extract the 128-bit part containing the splat element and update
   // the splat element index when it refers to the higher register.
-  if (Size == 256) {
+  if (Is256BitVec) {
     V1 = Extract128BitVector(V1, EltNo, DAG, dl);
     if (EltNo >= NumElems/2)
       EltNo -= NumElems/2;
@@ -4503,7 +4569,7 @@ static SDValue PromoteSplat(ShuffleVectorSDNode *SV, SelectionDAG &DAG) {
   // Recreate the 256-bit vector and place the same 128-bit vector
   // into the low and high part. This is necessary because we want
   // to use VPERM* to shuffle the vectors
-  if (Size == 256) {
+  if (Is256BitVec) {
     V1 = DAG.getNode(ISD::CONCAT_VECTORS, dl, SrcVT, V1, V1);
   }
 
@@ -4555,6 +4621,10 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT,
   case X86ISD::MOVLHPS:
     DecodeMOVLHPSMask(NumElems, Mask);
     break;
+  case X86ISD::PALIGNR:
+    ImmN = N->getOperand(N->getNumOperands()-1);
+    DecodePALIGNRMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    break;
   case X86ISD::PSHUFD:
   case X86ISD::VPERMILP:
     ImmN = N->getOperand(N->getNumOperands()-1);
@@ -4598,7 +4668,6 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT,
   case X86ISD::MOVLPS:
   case X86ISD::MOVSHDUP:
   case X86ISD::MOVSLDUP:
-  case X86ISD::PALIGN:
     // Not yet implemented
     return false;
   default: llvm_unreachable("unknown target shuffle node");
@@ -4893,7 +4962,7 @@ static SDValue getVShift(bool isLeft, EVT VT, SDValue SrcOp,
   return DAG.getNode(ISD::BITCAST, dl, VT,
                      DAG.getNode(Opc, dl, ShVT, SrcOp,
                              DAG.getConstant(NumBits,
-                                  TLI.getShiftAmountTy(SrcOp.getValueType()))));
+                                  TLI.getScalarShiftAmountTy(SrcOp.getValueType()))));
 }
 
 SDValue
@@ -5063,10 +5132,10 @@ static SDValue EltsFromConsecutiveLoads(EVT VT, SmallVectorImpl<SDValue> &Elts,
 /// or SDValue() otherwise.
 SDValue
 X86TargetLowering::LowerVectorBroadcast(SDValue Op, SelectionDAG &DAG) const {
-  if (!Subtarget->hasAVX())
+  if (!Subtarget->hasFp256())
     return SDValue();
 
-  EVT VT = Op.getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
   DebugLoc dl = Op.getDebugLoc();
 
   assert((VT.is128BitVector() || VT.is256BitVector()) &&
@@ -5109,7 +5178,7 @@ X86TargetLowering::LowerVectorBroadcast(SDValue Op, SelectionDAG &DAG) const {
       if (Sc.getOpcode() != ISD::SCALAR_TO_VECTOR &&
           Sc.getOpcode() != ISD::BUILD_VECTOR) {
 
-        if (!Subtarget->hasAVX2())
+        if (!Subtarget->hasInt256())
           return SDValue();
 
         // Use the register form of the broadcast instruction available on AVX2.
@@ -5136,7 +5205,7 @@ X86TargetLowering::LowerVectorBroadcast(SDValue Op, SelectionDAG &DAG) const {
   // Handle the broadcasting a single constant scalar from the constant pool
   // into a vector. On Sandybridge it is still better to load a constant vector
   // from the constant pool and not to broadcast it from a scalar.
-  if (ConstSplatVal && Subtarget->hasAVX2()) {
+  if (ConstSplatVal && Subtarget->hasInt256()) {
     EVT CVT = Ld.getValueType();
     assert(!CVT.isVector() && "Must not broadcast a vector type");
     unsigned ScalarSize = CVT.getSizeInBits();
@@ -5164,7 +5233,7 @@ X86TargetLowering::LowerVectorBroadcast(SDValue Op, SelectionDAG &DAG) const {
   unsigned ScalarSize = Ld.getValueType().getSizeInBits();
 
   // Handle AVX2 in-register broadcasts.
-  if (!IsLoad && Subtarget->hasAVX2() &&
+  if (!IsLoad && Subtarget->hasInt256() &&
       (ScalarSize == 32 || (Is256 && ScalarSize == 64)))
     return DAG.getNode(X86ISD::VBROADCAST, dl, VT, Ld);
 
@@ -5177,7 +5246,7 @@ X86TargetLowering::LowerVectorBroadcast(SDValue Op, SelectionDAG &DAG) const {
 
   // The integer check is needed for the 64-bit into 128-bit so it doesn't match
   // double since there is no vbroadcastsd xmm
-  if (Subtarget->hasAVX2() && Ld.getValueType().isInteger()) {
+  if (Subtarget->hasInt256() && Ld.getValueType().isInteger()) {
     if (ScalarSize == 8 || ScalarSize == 16 || ScalarSize == 64)
       return DAG.getNode(X86ISD::VBROADCAST, dl, VT, Ld);
   }
@@ -5264,8 +5333,8 @@ SDValue
 X86TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const {
   DebugLoc dl = Op.getDebugLoc();
 
-  EVT VT = Op.getValueType();
-  EVT ExtVT = VT.getVectorElementType();
+  MVT VT = Op.getValueType().getSimpleVT();
+  MVT ExtVT = VT.getVectorElementType();
   unsigned NumElems = Op.getNumOperands();
 
   // Vectors containing all zeros can be matched by pxor and xorps later
@@ -5281,11 +5350,11 @@ X86TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const {
   // Vectors containing all ones can be matched by pcmpeqd on 128-bit width
   // vectors or broken into v4i32 operations on 256-bit vectors. AVX2 can use
   // vpcmpeqd on 256-bit vectors.
-  if (ISD::isBuildVectorAllOnes(Op.getNode())) {
-    if (VT == MVT::v4i32 || (VT == MVT::v8i32 && Subtarget->hasAVX2()))
+  if (Subtarget->hasSSE2() && ISD::isBuildVectorAllOnes(Op.getNode())) {
+    if (VT == MVT::v4i32 || (VT == MVT::v8i32 && Subtarget->hasInt256()))
       return Op;
 
-    return getOnesVector(VT, Subtarget->hasAVX2(), DAG, dl);
+    return getOnesVector(VT, Subtarget->hasInt256(), DAG, dl);
   }
 
   SDValue Broadcast = LowerVectorBroadcast(Op, DAG);
@@ -5596,7 +5665,7 @@ X86TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const {
 // to create 256-bit vectors from two other 128-bit ones.
 static SDValue LowerAVXCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) {
   DebugLoc dl = Op.getDebugLoc();
-  EVT ResVT = Op.getValueType();
+  MVT ResVT = Op.getValueType().getSimpleVT();
 
   assert(ResVT.is256BitVector() && "Value type must be 256-bit wide");
 
@@ -5623,63 +5692,51 @@ LowerVECTOR_SHUFFLEtoBlend(ShuffleVectorSDNode *SVOp,
   SDValue V2 = SVOp->getOperand(1);
   DebugLoc dl = SVOp->getDebugLoc();
   MVT VT = SVOp->getValueType(0).getSimpleVT();
+  MVT EltVT = VT.getVectorElementType();
   unsigned NumElems = VT.getVectorNumElements();
 
-  if (!Subtarget->hasSSE41())
+  if (!Subtarget->hasSSE41() || EltVT == MVT::i8)
+    return SDValue();
+  if (!Subtarget->hasInt256() && VT == MVT::v16i16)
     return SDValue();
 
-  unsigned ISDNo = 0;
-  MVT OpTy;
-
-  switch (VT.SimpleTy) {
-  default: return SDValue();
-  case MVT::v8i16:
-    ISDNo = X86ISD::BLENDPW;
-    OpTy = MVT::v8i16;
-    break;
-  case MVT::v4i32:
-  case MVT::v4f32:
-    ISDNo = X86ISD::BLENDPS;
-    OpTy = MVT::v4f32;
-    break;
-  case MVT::v2i64:
-  case MVT::v2f64:
-    ISDNo = X86ISD::BLENDPD;
-    OpTy = MVT::v2f64;
-    break;
-  case MVT::v8i32:
-  case MVT::v8f32:
-    if (!Subtarget->hasAVX())
-      return SDValue();
-    ISDNo = X86ISD::BLENDPS;
-    OpTy = MVT::v8f32;
-    break;
-  case MVT::v4i64:
-  case MVT::v4f64:
-    if (!Subtarget->hasAVX())
-      return SDValue();
-    ISDNo = X86ISD::BLENDPD;
-    OpTy = MVT::v4f64;
-    break;
-  }
-  assert(ISDNo && "Invalid Op Number");
+  // Check the mask for BLEND and build the value.
+  unsigned MaskValue = 0;
+  // There are 2 lanes if (NumElems > 8), and 1 lane otherwise.
+  unsigned NumLanes = (NumElems-1)/8 + 1;
+  unsigned NumElemsInLane = NumElems / NumLanes;
 
-  unsigned MaskVals = 0;
+  // Blend for v16i16 should be symetric for the both lanes.
+  for (unsigned i = 0; i < NumElemsInLane; ++i) {
 
-  for (unsigned i = 0; i != NumElems; ++i) {
+    int SndLaneEltIdx = (NumLanes == 2) ?
+      SVOp->getMaskElt(i + NumElemsInLane) : -1;
     int EltIdx = SVOp->getMaskElt(i);
-    if (EltIdx == (int)i || EltIdx < 0)
-      MaskVals |= (1<<i);
-    else if (EltIdx == (int)(i + NumElems))
-      continue; // Bit is set to zero;
+
+    if ((EltIdx < 0 || EltIdx == (int)i) &&
+        (SndLaneEltIdx < 0 || SndLaneEltIdx == (int)(i + NumElemsInLane)))
+      continue;
+
+    if (((unsigned)EltIdx == (i + NumElems)) &&
+        (SndLaneEltIdx < 0 ||
+         (unsigned)SndLaneEltIdx == i + NumElems + NumElemsInLane))
+      MaskValue |= (1<<i);
     else
       return SDValue();
   }
 
-  V1 = DAG.getNode(ISD::BITCAST, dl, OpTy, V1);
-  V2 = DAG.getNode(ISD::BITCAST, dl, OpTy, V2);
-  SDValue Ret =  DAG.getNode(ISDNo, dl, OpTy, V1, V2,
-                             DAG.getConstant(MaskVals, MVT::i32));
+  // Convert i32 vectors to floating point if it is not AVX2.
+  // AVX2 introduced VPBLENDD instruction for 128 and 256-bit vectors.
+  MVT BlendVT = VT;
+  if (EltVT == MVT::i64 || (EltVT == MVT::i32 && !Subtarget->hasInt256())) {
+    BlendVT = MVT::getVectorVT(MVT::getFloatingPointVT(EltVT.getSizeInBits()),
+                               NumElems);
+    V1 = DAG.getNode(ISD::BITCAST, dl, VT, V1);
+    V2 = DAG.getNode(ISD::BITCAST, dl, VT, V2);
+  }
+
+  SDValue Ret = DAG.getNode(X86ISD::BLENDI, dl, BlendVT, V1, V2,
+                            DAG.getConstant(MaskValue, MVT::i32));
   return DAG.getNode(ISD::BITCAST, dl, VT, Ret);
 }
 
@@ -5814,6 +5871,11 @@ LowerVECTOR_SHUFFLEv8i16(SDValue Op, const X86Subtarget *Subtarget,
     }
   }
 
+  // Promote splats to a larger type which usually leads to more efficient code.
+  // FIXME: Is this true if pshufb is available?
+  if (SVOp->isSplat())
+    return PromoteSplat(SVOp, DAG);
+
   // If we have SSSE3, and all words of the result are from 1 input vector,
   // case 2 is generated, otherwise case 3 is generated.  If no SSSE3
   // is present, fall back to case 4.
@@ -5829,7 +5891,7 @@ LowerVECTOR_SHUFFLEv8i16(SDValue Op, const X86Subtarget *Subtarget,
       int EltIdx = MaskVals[i] * 2;
       int Idx0 = (TwoInputs && (EltIdx >= 16)) ? 0x80 : EltIdx;
       int Idx1 = (TwoInputs && (EltIdx >= 16)) ? 0x80 : EltIdx+1;
-      pshufbMask.push_back(DAG.getConstant(Idx0,   MVT::i8));
+      pshufbMask.push_back(DAG.getConstant(Idx0, MVT::i8));
       pshufbMask.push_back(DAG.getConstant(Idx1, MVT::i8));
     }
     V1 = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, V1);
@@ -5947,6 +6009,11 @@ SDValue LowerVECTOR_SHUFFLEv16i8(ShuffleVectorSDNode *SVOp,
   DebugLoc dl = SVOp->getDebugLoc();
   ArrayRef<int> MaskVals = SVOp->getMask();
 
+  // Promote splats to a larger type which usually leads to more efficient code.
+  // FIXME: Is this true if pshufb is available?
+  if (SVOp->isSplat())
+    return PromoteSplat(SVOp, DAG);
+
   // If we have SSSE3, case 1 is generated when all result bytes come from
   // one of  the inputs.  Otherwise, case 2 is generated.  If no SSSE3 is
   // present, fall back to case 3.
@@ -6065,7 +6132,7 @@ static
 SDValue LowerVECTOR_SHUFFLEv32i8(ShuffleVectorSDNode *SVOp,
                                  const X86Subtarget *Subtarget,
                                  SelectionDAG &DAG) {
-  EVT VT = SVOp->getValueType(0);
+  MVT VT = SVOp->getValueType(0).getSimpleVT();
   SDValue V1 = SVOp->getOperand(0);
   SDValue V2 = SVOp->getOperand(1);
   DebugLoc dl = SVOp->getDebugLoc();
@@ -6079,7 +6146,7 @@ SDValue LowerVECTOR_SHUFFLEv32i8(ShuffleVectorSDNode *SVOp,
   // (1) one of input vector is undefined or zeroinitializer.
   // The mask value 0x80 puts 0 in the corresponding slot of the vector.
   // And (2) the mask indexes don't cross the 128-bit lane.
-  if (VT != MVT::v32i8 || !Subtarget->hasAVX2() ||
+  if (VT != MVT::v32i8 || !Subtarget->hasInt256() ||
       (!V2IsUndef && !V2IsAllZero && !V1IsAllZero))
     return SDValue();
 
@@ -6112,8 +6179,9 @@ SDValue LowerVECTOR_SHUFFLEv32i8(ShuffleVectorSDNode *SVOp,
 /// vector_shuffle X, Y, <2, 3, | 10, 11, | 0, 1, | 14, 15>
 static
 SDValue RewriteAsNarrowerShuffle(ShuffleVectorSDNode *SVOp,
-                                 SelectionDAG &DAG, DebugLoc dl) {
+                                 SelectionDAG &DAG) {
   MVT VT = SVOp->getValueType(0).getSimpleVT();
+  DebugLoc dl = SVOp->getDebugLoc();
   unsigned NumElems = VT.getVectorNumElements();
   MVT NewVT;
   unsigned Scale;
@@ -6149,7 +6217,7 @@ SDValue RewriteAsNarrowerShuffle(ShuffleVectorSDNode *SVOp,
 
 /// getVZextMovL - Return a zero-extending vector move low node.
 ///
-static SDValue getVZextMovL(EVT VT, EVT OpVT,
+static SDValue getVZextMovL(MVT VT, EVT OpVT,
                             SDValue SrcOp, SelectionDAG &DAG,
                             const X86Subtarget *Subtarget, DebugLoc dl) {
   if (VT == MVT::v2f64 || VT == MVT::v4f32) {
@@ -6191,14 +6259,14 @@ LowerVECTOR_SHUFFLE_256(ShuffleVectorSDNode *SVOp, SelectionDAG &DAG) {
   if (NewOp.getNode())
     return NewOp;
 
-  EVT VT = SVOp->getValueType(0);
+  MVT VT = SVOp->getValueType(0).getSimpleVT();
 
   unsigned NumElems = VT.getVectorNumElements();
   unsigned NumLaneElems = NumElems / 2;
 
   DebugLoc dl = SVOp->getDebugLoc();
-  MVT EltVT = VT.getVectorElementType().getSimpleVT();
-  EVT NVT = MVT::getVectorVT(EltVT, NumLaneElems);
+  MVT EltVT = VT.getVectorElementType();
+  MVT NVT = MVT::getVectorVT(EltVT, NumLaneElems);
   SDValue Output[2];
 
   SmallVector<int, 16> Mask;
@@ -6303,7 +6371,7 @@ LowerVECTOR_SHUFFLE_128v4(ShuffleVectorSDNode *SVOp, SelectionDAG &DAG) {
   SDValue V1 = SVOp->getOperand(0);
   SDValue V2 = SVOp->getOperand(1);
   DebugLoc dl = SVOp->getDebugLoc();
-  EVT VT = SVOp->getValueType(0);
+  MVT VT = SVOp->getValueType(0).getSimpleVT();
 
   assert(VT.is128BitVector() && "Unsupported vector size");
 
@@ -6452,23 +6520,6 @@ static bool MayFoldVectorLoad(SDValue V) {
   return MayFoldLoad(V);
 }
 
-// FIXME: the version above should always be used. Since there's
-// a bug where several vector shuffles can't be folded because the
-// DAG is not updated during lowering and a node claims to have two
-// uses while it only has one, use this version, and let isel match
-// another instruction if the load really happens to have more than
-// one use. Remove this version after this bug get fixed.
-// rdar://8434668, PR8156
-static bool RelaxedMayFoldVectorLoad(SDValue V) {
-  if (V.hasOneUse() && V.getOpcode() == ISD::BITCAST)
-    V = V.getOperand(0);
-  if (V.hasOneUse() && V.getOpcode() == ISD::SCALAR_TO_VECTOR)
-    V = V.getOperand(0);
-  if (ISD::isNormalLoad(V.getNode()))
-    return true;
-  return false;
-}
-
 static
 SDValue getMOVDDup(SDValue &Op, DebugLoc &dl, SDValue V1, SelectionDAG &DAG) {
   EVT VT = Op.getValueType();
@@ -6574,7 +6625,7 @@ SDValue getMOVLP(SDValue &Op, DebugLoc &dl, SelectionDAG &DAG, bool HasSSE2) {
 
 // Reduce a vector shuffle to zext.
 SDValue
-X86TargetLowering::lowerVectorIntExtend(SDValue Op, SelectionDAG &DAG) const {
+X86TargetLowering::LowerVectorIntExtend(SDValue Op, SelectionDAG &DAG) const {
   // PMOVZX is only available from SSE41.
   if (!Subtarget->hasSSE41())
     return SDValue();
@@ -6582,7 +6633,7 @@ X86TargetLowering::lowerVectorIntExtend(SDValue Op, SelectionDAG &DAG) const {
   EVT VT = Op.getValueType();
 
   // Only AVX2 support 256-bit vector integer extending.
-  if (!Subtarget->hasAVX2() && VT.is256BitVector())
+  if (!Subtarget->hasInt256() && VT.is256BitVector())
     return SDValue();
 
   ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
@@ -6618,9 +6669,10 @@ X86TargetLowering::lowerVectorIntExtend(SDValue Op, SelectionDAG &DAG) const {
       return SDValue();
   }
 
+  LLVMContext *Context = DAG.getContext();
   unsigned NBits = VT.getVectorElementType().getSizeInBits() << Shift;
-  EVT NeVT = EVT::getIntegerVT(*DAG.getContext(), NBits);
-  EVT NVT = EVT::getVectorVT(*DAG.getContext(), NeVT, NumElems >> Shift);
+  EVT NeVT = EVT::getIntegerVT(*Context, NBits);
+  EVT NVT = EVT::getVectorVT(*Context, NeVT, NumElems >> Shift);
 
   if (!isTypeLegal(NVT))
     return SDValue();
@@ -6639,8 +6691,21 @@ X86TargetLowering::lowerVectorIntExtend(SDValue Op, SelectionDAG &DAG) const {
     // If it's foldable, i.e. normal load with single use, we will let code
     // selection to fold it. Otherwise, we will short the conversion sequence.
     if (CIdx && CIdx->getZExtValue() == 0 &&
-        (!ISD::isNormalLoad(V.getNode()) || !V.hasOneUse()))
+        (!ISD::isNormalLoad(V.getNode()) || !V.hasOneUse())) {
+      if (V.getValueSizeInBits() > V1.getValueSizeInBits()) {
+        // The "ext_vec_elt" node is wider than the result node.
+        // In this case we should extract subvector from V.
+        // (bitcast (sclr2vec (ext_vec_elt x))) -> (bitcast (extract_subvector x)).
+        unsigned Ratio = V.getValueSizeInBits() / V1.getValueSizeInBits();
+        EVT FullVT = V.getValueType();
+        EVT SubVecVT = EVT::getVectorVT(*Context, 
+                                        FullVT.getVectorElementType(),
+                                        FullVT.getVectorNumElements()/Ratio);
+        V = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, SubVecVT, V, 
+                        DAG.getIntPtrConstant(0));
+      }
       V1 = DAG.getNode(ISD::BITCAST, DL, V1.getValueType(), V);
+    }
   }
 
   return DAG.getNode(ISD::BITCAST, DL, VT,
@@ -6650,7 +6715,7 @@ X86TargetLowering::lowerVectorIntExtend(SDValue Op, SelectionDAG &DAG) const {
 SDValue
 X86TargetLowering::NormalizeVectorShuffle(SDValue Op, SelectionDAG &DAG) const {
   ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
-  EVT VT = Op.getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
   DebugLoc dl = Op.getDebugLoc();
   SDValue V1 = Op.getOperand(0);
   SDValue V2 = Op.getOperand(1);
@@ -6660,25 +6725,14 @@ X86TargetLowering::NormalizeVectorShuffle(SDValue Op, SelectionDAG &DAG) const {
 
   // Handle splat operations
   if (SVOp->isSplat()) {
-    unsigned NumElem = VT.getVectorNumElements();
-    int Size = VT.getSizeInBits();
-
     // Use vbroadcast whenever the splat comes from a foldable load
     SDValue Broadcast = LowerVectorBroadcast(Op, DAG);
     if (Broadcast.getNode())
       return Broadcast;
-
-    // Handle splats by matching through known shuffle masks
-    if ((Size == 128 && NumElem <= 4) ||
-        (Size == 256 && NumElem < 8))
-      return SDValue();
-
-    // All remaning splats are promoted to target supported vector shuffles.
-    return PromoteSplat(SVOp, DAG);
   }
 
   // Check integer expanding shuffles.
-  SDValue NewOp = lowerVectorIntExtend(Op, DAG);
+  SDValue NewOp = LowerVectorIntExtend(Op, DAG);
   if (NewOp.getNode())
     return NewOp;
 
@@ -6686,7 +6740,7 @@ X86TargetLowering::NormalizeVectorShuffle(SDValue Op, SelectionDAG &DAG) const {
   // do it!
   if (VT == MVT::v8i16  || VT == MVT::v16i8 ||
       VT == MVT::v16i16 || VT == MVT::v32i8) {
-    SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG, dl);
+    SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG);
     if (NewOp.getNode())
       return DAG.getNode(ISD::BITCAST, dl, VT, NewOp);
   } else if ((VT == MVT::v4i32 ||
@@ -6694,18 +6748,18 @@ X86TargetLowering::NormalizeVectorShuffle(SDValue Op, SelectionDAG &DAG) const {
     // FIXME: Figure out a cleaner way to do this.
     // Try to make use of movq to zero out the top part.
     if (ISD::isBuildVectorAllZeros(V2.getNode())) {
-      SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG, dl);
+      SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG);
       if (NewOp.getNode()) {
-        EVT NewVT = NewOp.getValueType();
+        MVT NewVT = NewOp.getValueType().getSimpleVT();
         if (isCommutedMOVLMask(cast<ShuffleVectorSDNode>(NewOp)->getMask(),
                                NewVT, true, false))
           return getVZextMovL(VT, NewVT, NewOp.getOperand(0),
                               DAG, Subtarget, dl);
       }
     } else if (ISD::isBuildVectorAllZeros(V1.getNode())) {
-      SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG, dl);
+      SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG);
       if (NewOp.getNode()) {
-        EVT NewVT = NewOp.getValueType();
+        MVT NewVT = NewOp.getValueType().getSimpleVT();
         if (isMOVLMask(cast<ShuffleVectorSDNode>(NewOp)->getMask(), NewVT))
           return getVZextMovL(VT, NewVT, NewOp.getOperand(1),
                               DAG, Subtarget, dl);
@@ -6720,7 +6774,7 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
   ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
   SDValue V1 = Op.getOperand(0);
   SDValue V2 = Op.getOperand(1);
-  EVT VT = Op.getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
   DebugLoc dl = Op.getDebugLoc();
   unsigned NumElems = VT.getVectorNumElements();
   bool V1IsUndef = V1.getOpcode() == ISD::UNDEF;
@@ -6728,11 +6782,11 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
   bool V1IsSplat = false;
   bool V2IsSplat = false;
   bool HasSSE2 = Subtarget->hasSSE2();
-  bool HasAVX    = Subtarget->hasAVX();
-  bool HasAVX2   = Subtarget->hasAVX2();
+  bool HasFp256    = Subtarget->hasFp256();
+  bool HasInt256   = Subtarget->hasInt256();
   MachineFunction &MF = DAG.getMachineFunction();
-  bool OptForSize = MF.getFunction()->getFnAttributes().
-    hasAttribute(Attributes::OptimizeForSize);
+  bool OptForSize = MF.getFunction()->getAttributes().
+    hasAttribute(AttributeSet::FunctionIndex, Attribute::OptimizeForSize);
 
   assert(VT.getSizeInBits() != 64 && "Can't lower MMX shuffles");
 
@@ -6766,20 +6820,20 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
 
   // NOTE: isPSHUFDMask can also match both masks below (unpckl_undef and
   // unpckh_undef). Only use pshufd if speed is more important than size.
-  if (OptForSize && isUNPCKL_v_undef_Mask(M, VT, HasAVX2))
+  if (OptForSize && isUNPCKL_v_undef_Mask(M, VT, HasInt256))
     return getTargetShuffleNode(X86ISD::UNPCKL, dl, VT, V1, V1, DAG);
-  if (OptForSize && isUNPCKH_v_undef_Mask(M, VT, HasAVX2))
+  if (OptForSize && isUNPCKH_v_undef_Mask(M, VT, HasInt256))
     return getTargetShuffleNode(X86ISD::UNPCKH, dl, VT, V1, V1, DAG);
 
   if (isMOVDDUPMask(M, VT) && Subtarget->hasSSE3() &&
-      V2IsUndef && RelaxedMayFoldVectorLoad(V1))
+      V2IsUndef && MayFoldVectorLoad(V1))
     return getMOVDDup(Op, dl, V1, DAG);
 
   if (isMOVHLPS_v_undef_Mask(M, VT))
     return getMOVHighToLow(Op, dl, DAG);
 
   // Use to match splats
-  if (HasSSE2 && isUNPCKHMask(M, VT, HasAVX2) && V2IsUndef &&
+  if (HasSSE2 && isUNPCKHMask(M, VT, HasInt256) && V2IsUndef &&
       (VT == MVT::v2f64 || VT == MVT::v2i64))
     return getTargetShuffleNode(X86ISD::UNPCKH, dl, VT, V1, V1, DAG);
 
@@ -6792,12 +6846,13 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
 
     unsigned TargetMask = getShuffleSHUFImmediate(SVOp);
 
-    if (HasAVX && (VT == MVT::v4f32 || VT == MVT::v2f64))
-      return getTargetShuffleNode(X86ISD::VPERMILP, dl, VT, V1, TargetMask, DAG);
-
     if (HasSSE2 && (VT == MVT::v4f32 || VT == MVT::v4i32))
       return getTargetShuffleNode(X86ISD::PSHUFD, dl, VT, V1, TargetMask, DAG);
 
+    if (HasFp256 && (VT == MVT::v4f32 || VT == MVT::v2f64))
+      return getTargetShuffleNode(X86ISD::VPERMILP, dl, VT, V1, TargetMask,
+                                  DAG);
+
     return getTargetShuffleNode(X86ISD::SHUFP, dl, VT, V1, V1,
                                 TargetMask, DAG);
   }
@@ -6810,7 +6865,7 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
   if (isShift && ShVal.hasOneUse()) {
     // If the shifted value has multiple uses, it may be cheaper to use
     // v_set0 + movlhps or movhlps, etc.
-    EVT EltVT = VT.getVectorElementType();
+    MVT EltVT = VT.getVectorElementType();
     ShAmt *= EltVT.getSizeInBits();
     return getVShift(isLeft, VT, ShVal, ShAmt, DAG, *this, dl);
   }
@@ -6828,7 +6883,7 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
   }
 
   // FIXME: fold these into legal mask.
-  if (isMOVLHPSMask(M, VT) && !isUNPCKLMask(M, VT, HasAVX2))
+  if (isMOVLHPSMask(M, VT) && !isUNPCKLMask(M, VT, HasInt256))
     return getMOVLowToHigh(Op, dl, DAG, HasSSE2);
 
   if (isMOVHLPSMask(M, VT))
@@ -6849,7 +6904,7 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
 
   if (isShift) {
     // No better options. Use a vshldq / vsrldq.
-    EVT EltVT = VT.getVectorElementType();
+    MVT EltVT = VT.getVectorElementType();
     ShAmt *= EltVT.getSizeInBits();
     return getVShift(isLeft, VT, ShVal, ShAmt, DAG, *this, dl);
   }
@@ -6878,10 +6933,10 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
     return getMOVL(DAG, dl, VT, V2, V1);
   }
 
-  if (isUNPCKLMask(M, VT, HasAVX2))
+  if (isUNPCKLMask(M, VT, HasInt256))
     return getTargetShuffleNode(X86ISD::UNPCKL, dl, VT, V1, V2, DAG);
 
-  if (isUNPCKHMask(M, VT, HasAVX2))
+  if (isUNPCKHMask(M, VT, HasInt256))
     return getTargetShuffleNode(X86ISD::UNPCKH, dl, VT, V1, V2, DAG);
 
   if (V2IsSplat) {
@@ -6890,9 +6945,9 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
     // new vector_shuffle with the corrected mask.p
     SmallVector<int, 8> NewMask(M.begin(), M.end());
     NormalizeMask(NewMask, NumElems);
-    if (isUNPCKLMask(NewMask, VT, HasAVX2, true))
+    if (isUNPCKLMask(NewMask, VT, HasInt256, true))
       return getTargetShuffleNode(X86ISD::UNPCKL, dl, VT, V1, V2, DAG);
-    if (isUNPCKHMask(NewMask, VT, HasAVX2, true))
+    if (isUNPCKHMask(NewMask, VT, HasInt256, true))
       return getTargetShuffleNode(X86ISD::UNPCKH, dl, VT, V1, V2, DAG);
   }
 
@@ -6904,15 +6959,15 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
     std::swap(V1IsSplat, V2IsSplat);
     Commuted = false;
 
-    if (isUNPCKLMask(M, VT, HasAVX2))
+    if (isUNPCKLMask(M, VT, HasInt256))
       return getTargetShuffleNode(X86ISD::UNPCKL, dl, VT, V1, V2, DAG);
 
-    if (isUNPCKHMask(M, VT, HasAVX2))
+    if (isUNPCKHMask(M, VT, HasInt256))
       return getTargetShuffleNode(X86ISD::UNPCKH, dl, VT, V1, V2, DAG);
   }
 
   // Normalize the node to match x86 shuffle ops if needed
-  if (!V2IsUndef && (isSHUFPMask(M, VT, HasAVX, /* Commuted */ true)))
+  if (!V2IsUndef && (isSHUFPMask(M, VT, HasFp256, /* Commuted */ true)))
     return CommuteVectorShuffle(SVOp, DAG);
 
   // The checks below are all present in isShuffleMaskLegal, but they are
@@ -6920,7 +6975,7 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
   // nodes, and remove one by one until they don't return Op anymore.
 
   if (isPALIGNRMask(M, VT, Subtarget))
-    return getTargetShuffleNode(X86ISD::PALIGN, dl, VT, V1, V2,
+    return getTargetShuffleNode(X86ISD::PALIGNR, dl, VT, V1, V2,
                                 getShufflePALIGNRImmediate(SVOp),
                                 DAG);
 
@@ -6930,23 +6985,23 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
       return getTargetShuffleNode(X86ISD::UNPCKL, dl, VT, V1, V1, DAG);
   }
 
-  if (isPSHUFHWMask(M, VT, HasAVX2))
+  if (isPSHUFHWMask(M, VT, HasInt256))
     return getTargetShuffleNode(X86ISD::PSHUFHW, dl, VT, V1,
                                 getShufflePSHUFHWImmediate(SVOp),
                                 DAG);
 
-  if (isPSHUFLWMask(M, VT, HasAVX2))
+  if (isPSHUFLWMask(M, VT, HasInt256))
     return getTargetShuffleNode(X86ISD::PSHUFLW, dl, VT, V1,
                                 getShufflePSHUFLWImmediate(SVOp),
                                 DAG);
 
-  if (isSHUFPMask(M, VT, HasAVX))
+  if (isSHUFPMask(M, VT, HasFp256))
     return getTargetShuffleNode(X86ISD::SHUFP, dl, VT, V1, V2,
                                 getShuffleSHUFImmediate(SVOp), DAG);
 
-  if (isUNPCKL_v_undef_Mask(M, VT, HasAVX2))
+  if (isUNPCKL_v_undef_Mask(M, VT, HasInt256))
     return getTargetShuffleNode(X86ISD::UNPCKL, dl, VT, V1, V1, DAG);
-  if (isUNPCKH_v_undef_Mask(M, VT, HasAVX2))
+  if (isUNPCKH_v_undef_Mask(M, VT, HasInt256))
     return getTargetShuffleNode(X86ISD::UNPCKH, dl, VT, V1, V1, DAG);
 
   //===--------------------------------------------------------------------===//
@@ -6955,12 +7010,12 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
   //
 
   // Handle VMOVDDUPY permutations
-  if (V2IsUndef && isMOVDDUPYMask(M, VT, HasAVX))
+  if (V2IsUndef && isMOVDDUPYMask(M, VT, HasFp256))
     return getTargetShuffleNode(X86ISD::MOVDDUP, dl, VT, V1, DAG);
 
   // Handle VPERMILPS/D* permutations
-  if (isVPERMILPMask(M, VT, HasAVX)) {
-    if (HasAVX2 && VT == MVT::v8i32)
+  if (isVPERMILPMask(M, VT, HasFp256)) {
+    if (HasInt256 && VT == MVT::v8i32)
       return getTargetShuffleNode(X86ISD::PSHUFD, dl, VT, V1,
                                   getShuffleSHUFImmediate(SVOp), DAG);
     return getTargetShuffleNode(X86ISD::VPERMILP, dl, VT, V1,
@@ -6968,7 +7023,7 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
   }
 
   // Handle VPERM2F128/VPERM2I128 permutations
-  if (isVPERM2X128Mask(M, VT, HasAVX))
+  if (isVPERM2X128Mask(M, VT, HasFp256))
     return getTargetShuffleNode(X86ISD::VPERM2X128, dl, VT, V1,
                                 V2, getShuffleVPERM2X128Immediate(SVOp), DAG);
 
@@ -6976,7 +7031,7 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
   if (BlendOp.getNode())
     return BlendOp;
 
-  if (V2IsUndef && HasAVX2 && (VT == MVT::v8i32 || VT == MVT::v8f32)) {
+  if (V2IsUndef && HasInt256 && (VT == MVT::v8i32 || VT == MVT::v8f32)) {
     SmallVector<SDValue, 8> permclMask;
     for (unsigned i = 0; i != 8; ++i) {
       permclMask.push_back(DAG.getConstant((M[i]>=0) ? M[i] : 0, MVT::i32));
@@ -6988,11 +7043,10 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
                        DAG.getNode(ISD::BITCAST, dl, VT, Mask), V1);
   }
 
-  if (V2IsUndef && HasAVX2 && (VT == MVT::v4i64 || VT == MVT::v4f64))
+  if (V2IsUndef && HasInt256 && (VT == MVT::v4i64 || VT == MVT::v4f64))
     return getTargetShuffleNode(X86ISD::VPERMI, dl, VT, V1,
                                 getShuffleCLImmediate(SVOp), DAG);
 
-
   //===--------------------------------------------------------------------===//
   // Since no target specific shuffle was selected for this generic one,
   // lower it into other known shuffles. FIXME: this isn't true yet, but
@@ -7030,13 +7084,11 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
   return SDValue();
 }
 
-SDValue
-X86TargetLowering::LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op,
-                                                SelectionDAG &DAG) const {
-  EVT VT = Op.getValueType();
+static SDValue LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) {
+  MVT VT = Op.getValueType().getSimpleVT();
   DebugLoc dl = Op.getDebugLoc();
 
-  if (!Op.getOperand(0).getValueType().is128BitVector())
+  if (!Op.getOperand(0).getValueType().getSimpleVT().is128BitVector())
     return SDValue();
 
   if (VT.getSizeInBits() == 8) {
@@ -7094,7 +7146,6 @@ X86TargetLowering::LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op,
   return SDValue();
 }
 
-
 SDValue
 X86TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
                                            SelectionDAG &DAG) const {
@@ -7102,7 +7153,7 @@ X86TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
     return SDValue();
 
   SDValue Vec = Op.getOperand(0);
-  EVT VecVT = Vec.getValueType();
+  MVT VecVT = Vec.getValueType().getSimpleVT();
 
   // If this is a 256-bit vector result, first extract the 128-bit vector and
   // then extract the element from the 128-bit vector.
@@ -7129,7 +7180,7 @@ X86TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
       return Res;
   }
 
-  EVT VT = Op.getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
   DebugLoc dl = Op.getDebugLoc();
   // TODO: handle v16i8.
   if (VT.getSizeInBits() == 16) {
@@ -7142,7 +7193,7 @@ X86TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
                                                  MVT::v4i32, Vec),
                                      Op.getOperand(1)));
     // Transform it so it match pextrw which produces a 32-bit result.
-    EVT EltVT = MVT::i32;
+    MVT EltVT = MVT::i32;
     SDValue Extract = DAG.getNode(X86ISD::PEXTRW, dl, EltVT,
                                   Op.getOperand(0), Op.getOperand(1));
     SDValue Assert  = DAG.getNode(ISD::AssertZext, dl, EltVT, Extract,
@@ -7157,7 +7208,7 @@ X86TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
 
     // SHUFPS the element to the lowest double word, then movss.
     int Mask[4] = { static_cast<int>(Idx), -1, -1, -1 };
-    EVT VVT = Op.getOperand(0).getValueType();
+    MVT VVT = Op.getOperand(0).getValueType().getSimpleVT();
     SDValue Vec = DAG.getVectorShuffle(VVT, dl, Op.getOperand(0),
                                        DAG.getUNDEF(VVT), Mask);
     return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, VT, Vec,
@@ -7176,7 +7227,7 @@ X86TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
     // Note if the lower 64 bits of the result of the UNPCKHPD is then stored
     // to a f64mem, the whole operation is folded into a single MOVHPDmr.
     int Mask[2] = { 1, -1 };
-    EVT VVT = Op.getOperand(0).getValueType();
+    MVT VVT = Op.getOperand(0).getValueType().getSimpleVT();
     SDValue Vec = DAG.getVectorShuffle(VVT, dl, Op.getOperand(0),
                                        DAG.getUNDEF(VVT), Mask);
     return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, VT, Vec,
@@ -7186,11 +7237,9 @@ X86TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
   return SDValue();
 }
 
-SDValue
-X86TargetLowering::LowerINSERT_VECTOR_ELT_SSE4(SDValue Op,
-                                               SelectionDAG &DAG) const {
-  EVT VT = Op.getValueType();
-  EVT EltVT = VT.getVectorElementType();
+static SDValue LowerINSERT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) {
+  MVT VT = Op.getValueType().getSimpleVT();
+  MVT EltVT = VT.getVectorElementType();
   DebugLoc dl = Op.getDebugLoc();
 
   SDValue N0 = Op.getOperand(0);
@@ -7243,8 +7292,8 @@ X86TargetLowering::LowerINSERT_VECTOR_ELT_SSE4(SDValue Op,
 
 SDValue
 X86TargetLowering::LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const {
-  EVT VT = Op.getValueType();
-  EVT EltVT = VT.getVectorElementType();
+  MVT VT = Op.getValueType().getSimpleVT();
+  MVT EltVT = VT.getVectorElementType();
 
   DebugLoc dl = Op.getDebugLoc();
   SDValue N0 = Op.getOperand(0);
@@ -7292,7 +7341,7 @@ X86TargetLowering::LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const {
 static SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) {
   LLVMContext *Context = DAG.getContext();
   DebugLoc dl = Op.getDebugLoc();
-  EVT OpVT = Op.getValueType();
+  MVT OpVT = Op.getValueType().getSimpleVT();
 
   // If this is a 256-bit vector result, first insert into a 128-bit
   // vector and then insert into the 256-bit vector.
@@ -7323,7 +7372,7 @@ static SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) {
 // upper bits of a vector.
 static SDValue LowerEXTRACT_SUBVECTOR(SDValue Op, const X86Subtarget *Subtarget,
                                       SelectionDAG &DAG) {
-  if (Subtarget->hasAVX()) {
+  if (Subtarget->hasFp256()) {
     DebugLoc dl = Op.getNode()->getDebugLoc();
     SDValue Vec = Op.getNode()->getOperand(0);
     SDValue Idx = Op.getNode()->getOperand(1);
@@ -7343,7 +7392,7 @@ static SDValue LowerEXTRACT_SUBVECTOR(SDValue Op, const X86Subtarget *Subtarget,
 // the upper bits of a vector.
 static SDValue LowerINSERT_SUBVECTOR(SDValue Op, const X86Subtarget *Subtarget,
                                      SelectionDAG &DAG) {
-  if (Subtarget->hasAVX()) {
+  if (Subtarget->hasFp256()) {
     DebugLoc dl = Op.getNode()->getDebugLoc();
     SDValue Vec = Op.getNode()->getOperand(0);
     SDValue SubVec = Op.getNode()->getOperand(1);
@@ -7459,7 +7508,6 @@ X86TargetLowering::LowerExternalSymbol(SDValue Op, SelectionDAG &DAG) const {
   DebugLoc DL = Op.getDebugLoc();
   Result = DAG.getNode(WrapperKind, DL, getPointerTy(), Result);
 
-
   // With PIC, the address is actually $g + Offset.
   if (getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
       !Subtarget->is64Bit()) {
@@ -7508,8 +7556,7 @@ X86TargetLowering::LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const {
 
 SDValue
 X86TargetLowering::LowerGlobalAddress(const GlobalValue *GV, DebugLoc dl,
-                                      int64_t Offset,
-                                      SelectionDAG &DAG) const {
+                                      int64_t Offset, SelectionDAG &DAG) const {
   // Create the TargetGlobalAddress node, folding in the constant
   // offset if it is legal.
   unsigned char OpFlags =
@@ -7729,7 +7776,7 @@ X86TargetLowering::LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const {
       case TLSModel::LocalExec:
         return LowerToTLSExecModel(GA, DAG, getPointerTy(), model,
                                    Subtarget->is64Bit(),
-                         getTargetMachine().getRelocationModel() == Reloc::PIC_);
+                        getTargetMachine().getRelocationModel() == Reloc::PIC_);
     }
     llvm_unreachable("Unknown TLS model.");
   }
@@ -7779,7 +7826,7 @@ X86TargetLowering::LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const {
                               Chain.getValue(1));
   }
 
-  if (Subtarget->isTargetWindows()) {
+  if (Subtarget->isTargetWindows() || Subtarget->isTargetMingw()) {
     // Just use the implicit TLS architecture
     // Need to generate someting similar to:
     //   mov     rdx, qword [gs:abs 58H]; Load pointer to ThreadLocalStorage
@@ -7799,18 +7846,19 @@ X86TargetLowering::LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const {
     SDValue Chain = DAG.getEntryNode();
 
     // Get the Thread Pointer, which is %fs:__tls_array (32-bit) or
-    // %gs:0x58 (64-bit).
+    // %gs:0x58 (64-bit). On MinGW, __tls_array is not available, so directly
+    // use its literal value of 0x2C.
     Value *Ptr = Constant::getNullValue(Subtarget->is64Bit()
                                         ? Type::getInt8PtrTy(*DAG.getContext(),
                                                              256)
                                         : Type::getInt32PtrTy(*DAG.getContext(),
                                                               257));
 
-    SDValue ThreadPointer = DAG.getLoad(getPointerTy(), dl, Chain,
-                                        Subtarget->is64Bit()
-                                        ? DAG.getIntPtrConstant(0x58)
-                                        : DAG.getExternalSymbol("_tls_array",
-                                                                getPointerTy()),
+    SDValue TlsArray = Subtarget->is64Bit() ? DAG.getIntPtrConstant(0x58) :
+      (Subtarget->isTargetMingw() ? DAG.getIntPtrConstant(0x2C) :
+        DAG.getExternalSymbol("_tls_array", getPointerTy()));
+
+    SDValue ThreadPointer = DAG.getLoad(getPointerTy(), dl, Chain, TlsArray,
                                         MachinePointerInfo(Ptr),
                                         false, false, false, 0);
 
@@ -7846,7 +7894,6 @@ X86TargetLowering::LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const {
   llvm_unreachable("TLS not implemented for this target.");
 }
 
-
 /// LowerShiftParts - Lower SRA_PARTS and friends, which return two i32 values
 /// and take a 2 x i32 value to shift plus a shift amount.
 SDValue X86TargetLowering::LowerShiftParts(SDValue Op, SelectionDAG &DAG) const{
@@ -8013,9 +8060,11 @@ SDValue X86TargetLowering::LowerUINT_TO_FP_i64(SDValue Op,
 
   SmallVector<Constant*,2> CV1;
   CV1.push_back(
-        ConstantFP::get(*Context, APFloat(APInt(64, 0x4330000000000000ULL))));
+    ConstantFP::get(*Context, APFloat(APFloat::IEEEdouble,
+                                      APInt(64, 0x4330000000000000ULL))));
   CV1.push_back(
-        ConstantFP::get(*Context, APFloat(APInt(64, 0x4530000000000000ULL))));
+    ConstantFP::get(*Context, APFloat(APFloat::IEEEdouble,
+                                      APInt(64, 0x4530000000000000ULL))));
   Constant *C1 = ConstantVector::get(CV1);
   SDValue CPIdx1 = DAG.getConstantPool(C1, getPointerTy(), 16);
 
@@ -8109,7 +8158,8 @@ SDValue X86TargetLowering::lowerUINT_TO_FP_vec(SDValue Op,
           SVT == MVT::v8i8 || SVT == MVT::v8i16) &&
          "Custom UINT_TO_FP is not supported!");
 
-  EVT NVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32, SVT.getVectorNumElements());
+  EVT NVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32,
+                             SVT.getVectorNumElements());
   return DAG.getNode(ISD::SINT_TO_FP, dl, Op.getValueType(),
                      DAG.getNode(ISD::ZERO_EXTEND, dl, NVT, N0));
 }
@@ -8202,8 +8252,9 @@ SDValue X86TargetLowering::LowerUINT_TO_FP(SDValue Op,
   return DAG.getNode(ISD::FP_ROUND, dl, DstVT, Add, DAG.getIntPtrConstant(0));
 }
 
-std::pair<SDValue,SDValue> X86TargetLowering::
-FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG, bool IsSigned, bool IsReplace) const {
+std::pair<SDValue,SDValue>
+X86TargetLowering:: FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG,
+                                    bool IsSigned, bool IsReplace) const {
   DebugLoc DL = Op.getDebugLoc();
 
   EVT DstTy = Op.getValueType();
@@ -8295,46 +8346,197 @@ FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG, bool IsSigned, bool IsReplace) co
   }
 }
 
-SDValue X86TargetLowering::lowerZERO_EXTEND(SDValue Op, SelectionDAG &DAG) const {
+static SDValue LowerAVXExtend(SDValue Op, SelectionDAG &DAG,
+                              const X86Subtarget *Subtarget) {
+  MVT VT = Op->getValueType(0).getSimpleVT();
+  SDValue In = Op->getOperand(0);
+  MVT InVT = In.getValueType().getSimpleVT();
+  DebugLoc dl = Op->getDebugLoc();
+
+  // Optimize vectors in AVX mode:
+  //
+  //   v8i16 -> v8i32
+  //   Use vpunpcklwd for 4 lower elements  v8i16 -> v4i32.
+  //   Use vpunpckhwd for 4 upper elements  v8i16 -> v4i32.
+  //   Concat upper and lower parts.
+  //
+  //   v4i32 -> v4i64
+  //   Use vpunpckldq for 4 lower elements  v4i32 -> v2i64.
+  //   Use vpunpckhdq for 4 upper elements  v4i32 -> v2i64.
+  //   Concat upper and lower parts.
+  //
+
+  if (((VT != MVT::v8i32) || (InVT != MVT::v8i16)) &&
+      ((VT != MVT::v4i64) || (InVT != MVT::v4i32)))
+    return SDValue();
+
+  if (Subtarget->hasInt256())
+    return DAG.getNode(X86ISD::VZEXT_MOVL, dl, VT, In);
+
+  SDValue ZeroVec = getZeroVector(InVT, Subtarget, DAG, dl);
+  SDValue Undef = DAG.getUNDEF(InVT);
+  bool NeedZero = Op.getOpcode() == ISD::ZERO_EXTEND;
+  SDValue OpLo = getUnpackl(DAG, dl, InVT, In, NeedZero ? ZeroVec : Undef);
+  SDValue OpHi = getUnpackh(DAG, dl, InVT, In, NeedZero ? ZeroVec : Undef);
+
+  MVT HVT = MVT::getVectorVT(VT.getVectorElementType(),
+                             VT.getVectorNumElements()/2);
+
+  OpLo = DAG.getNode(ISD::BITCAST, dl, HVT, OpLo);
+  OpHi = DAG.getNode(ISD::BITCAST, dl, HVT, OpHi);
+
+  return DAG.getNode(ISD::CONCAT_VECTORS, dl, VT, OpLo, OpHi);
+}
+
+SDValue X86TargetLowering::LowerANY_EXTEND(SDValue Op,
+                                           SelectionDAG &DAG) const {
+  if (Subtarget->hasFp256()) {
+    SDValue Res = LowerAVXExtend(Op, DAG, Subtarget);
+    if (Res.getNode())
+      return Res;
+  }
+
+  return SDValue();
+}
+SDValue X86TargetLowering::LowerZERO_EXTEND(SDValue Op,
+                                            SelectionDAG &DAG) const {
   DebugLoc DL = Op.getDebugLoc();
-  EVT VT = Op.getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
   SDValue In = Op.getOperand(0);
-  EVT SVT = In.getValueType();
+  MVT SVT = In.getValueType().getSimpleVT();
+
+  if (Subtarget->hasFp256()) {
+    SDValue Res = LowerAVXExtend(Op, DAG, Subtarget);
+    if (Res.getNode())
+      return Res;
+  }
 
   if (!VT.is256BitVector() || !SVT.is128BitVector() ||
       VT.getVectorNumElements() != SVT.getVectorNumElements())
     return SDValue();
 
-  assert(Subtarget->hasAVX() && "256-bit vector is observed without AVX!");
+  assert(Subtarget->hasFp256() && "256-bit vector is observed without AVX!");
 
   // AVX2 has better support of integer extending.
-  if (Subtarget->hasAVX2())
+  if (Subtarget->hasInt256())
     return DAG.getNode(X86ISD::VZEXT, DL, VT, In);
 
   SDValue Lo = DAG.getNode(X86ISD::VZEXT, DL, MVT::v4i32, In);
   static const int Mask[] = {4, 5, 6, 7, -1, -1, -1, -1};
   SDValue Hi = DAG.getNode(X86ISD::VZEXT, DL, MVT::v4i32,
-                           DAG.getVectorShuffle(MVT::v8i16, DL, In, DAG.getUNDEF(MVT::v8i16), &Mask[0]));
+                           DAG.getVectorShuffle(MVT::v8i16, DL, In,
+                                                DAG.getUNDEF(MVT::v8i16),
+                                                &Mask[0]));
 
   return DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v8i32, Lo, Hi);
 }
 
-SDValue X86TargetLowering::lowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const {
+SDValue X86TargetLowering::LowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const {
   DebugLoc DL = Op.getDebugLoc();
-  EVT VT = Op.getValueType();
-  EVT SVT = Op.getOperand(0).getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
+  SDValue In = Op.getOperand(0);
+  MVT SVT = In.getValueType().getSimpleVT();
 
-  if (!VT.is128BitVector() || !SVT.is256BitVector() ||
-      VT.getVectorNumElements() != SVT.getVectorNumElements())
+  if ((VT == MVT::v4i32) && (SVT == MVT::v4i64)) {
+    // On AVX2, v4i64 -> v4i32 becomes VPERMD.
+    if (Subtarget->hasInt256()) {
+      static const int ShufMask[] = {0, 2, 4, 6, -1, -1, -1, -1};
+      In = DAG.getNode(ISD::BITCAST, DL, MVT::v8i32, In);
+      In = DAG.getVectorShuffle(MVT::v8i32, DL, In, DAG.getUNDEF(MVT::v8i32),
+                                ShufMask);
+      return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, VT, In,
+                         DAG.getIntPtrConstant(0));
+    }
+
+    // On AVX, v4i64 -> v4i32 becomes a sequence that uses PSHUFD and MOVLHPS.
+    SDValue OpLo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v2i64, In,
+                               DAG.getIntPtrConstant(0));
+    SDValue OpHi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v2i64, In,
+                               DAG.getIntPtrConstant(2));
+
+    OpLo = DAG.getNode(ISD::BITCAST, DL, MVT::v4i32, OpLo);
+    OpHi = DAG.getNode(ISD::BITCAST, DL, MVT::v4i32, OpHi);
+
+    // The PSHUFD mask:
+    static const int ShufMask1[] = {0, 2, 0, 0};
+    SDValue Undef = DAG.getUNDEF(VT);
+    OpLo = DAG.getVectorShuffle(VT, DL, OpLo, Undef, ShufMask1);
+    OpHi = DAG.getVectorShuffle(VT, DL, OpHi, Undef, ShufMask1);
+
+    // The MOVLHPS mask:
+    static const int ShufMask2[] = {0, 1, 4, 5};
+    return DAG.getVectorShuffle(VT, DL, OpLo, OpHi, ShufMask2);
+  }
+
+  if ((VT == MVT::v8i16) && (SVT == MVT::v8i32)) {
+    // On AVX2, v8i32 -> v8i16 becomed PSHUFB.
+    if (Subtarget->hasInt256()) {
+      In = DAG.getNode(ISD::BITCAST, DL, MVT::v32i8, In);
+
+      SmallVector<SDValue,32> pshufbMask;
+      for (unsigned i = 0; i < 2; ++i) {
+        pshufbMask.push_back(DAG.getConstant(0x0, MVT::i8));
+        pshufbMask.push_back(DAG.getConstant(0x1, MVT::i8));
+        pshufbMask.push_back(DAG.getConstant(0x4, MVT::i8));
+        pshufbMask.push_back(DAG.getConstant(0x5, MVT::i8));
+        pshufbMask.push_back(DAG.getConstant(0x8, MVT::i8));
+        pshufbMask.push_back(DAG.getConstant(0x9, MVT::i8));
+        pshufbMask.push_back(DAG.getConstant(0xc, MVT::i8));
+        pshufbMask.push_back(DAG.getConstant(0xd, MVT::i8));
+        for (unsigned j = 0; j < 8; ++j)
+          pshufbMask.push_back(DAG.getConstant(0x80, MVT::i8));
+      }
+      SDValue BV = DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v32i8,
+                               &pshufbMask[0], 32);
+      In = DAG.getNode(X86ISD::PSHUFB, DL, MVT::v32i8, In, BV);
+      In = DAG.getNode(ISD::BITCAST, DL, MVT::v4i64, In);
+
+      static const int ShufMask[] = {0,  2,  -1,  -1};
+      In = DAG.getVectorShuffle(MVT::v4i64, DL,  In, DAG.getUNDEF(MVT::v4i64),
+                                &ShufMask[0]);
+      In = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v2i64, In,
+                       DAG.getIntPtrConstant(0));
+      return DAG.getNode(ISD::BITCAST, DL, VT, In);
+    }
+
+    SDValue OpLo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v4i32, In,
+                               DAG.getIntPtrConstant(0));
+
+    SDValue OpHi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v4i32, In,
+                               DAG.getIntPtrConstant(4));
+
+    OpLo = DAG.getNode(ISD::BITCAST, DL, MVT::v16i8, OpLo);
+    OpHi = DAG.getNode(ISD::BITCAST, DL, MVT::v16i8, OpHi);
+
+    // The PSHUFB mask:
+    static const int ShufMask1[] = {0,  1,  4,  5,  8,  9, 12, 13,
+                                   -1, -1, -1, -1, -1, -1, -1, -1};
+
+    SDValue Undef = DAG.getUNDEF(MVT::v16i8);
+    OpLo = DAG.getVectorShuffle(MVT::v16i8, DL, OpLo, Undef, ShufMask1);
+    OpHi = DAG.getVectorShuffle(MVT::v16i8, DL, OpHi, Undef, ShufMask1);
+
+    OpLo = DAG.getNode(ISD::BITCAST, DL, MVT::v4i32, OpLo);
+    OpHi = DAG.getNode(ISD::BITCAST, DL, MVT::v4i32, OpHi);
+
+    // The MOVLHPS Mask:
+    static const int ShufMask2[] = {0, 1, 4, 5};
+    SDValue res = DAG.getVectorShuffle(MVT::v4i32, DL, OpLo, OpHi, ShufMask2);
+    return DAG.getNode(ISD::BITCAST, DL, MVT::v8i16, res);
+  }
+
+  // Handle truncation of V256 to V128 using shuffles.
+  if (!VT.is128BitVector() || !SVT.is256BitVector())
     return SDValue();
 
-  assert(Subtarget->hasAVX() && "256-bit vector is observed without AVX!");
+  assert(VT.getVectorNumElements() != SVT.getVectorNumElements() &&
+         "Invalid op");
+  assert(Subtarget->hasFp256() && "256-bit vector without AVX!");
 
   unsigned NumElems = VT.getVectorNumElements();
   EVT NVT = EVT::getVectorVT(*DAG.getContext(), VT.getVectorElementType(),
                              NumElems * 2);
 
-  SDValue In = Op.getOperand(0);
   SmallVector<int, 16> MaskVec(NumElems * 2, -1);
   // Prepare truncation shuffle mask
   for (unsigned i = 0; i != NumElems; ++i)
@@ -8348,9 +8550,10 @@ SDValue X86TargetLowering::lowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const {
 
 SDValue X86TargetLowering::LowerFP_TO_SINT(SDValue Op,
                                            SelectionDAG &DAG) const {
-  if (Op.getValueType().isVector()) {
-    if (Op.getValueType() == MVT::v8i16)
-      return DAG.getNode(ISD::TRUNCATE, Op.getDebugLoc(), Op.getValueType(),
+  MVT VT = Op.getValueType().getSimpleVT();
+  if (VT.isVector()) {
+    if (VT == MVT::v8i16)
+      return DAG.getNode(ISD::TRUNCATE, Op.getDebugLoc(), VT,
                          DAG.getNode(ISD::FP_TO_SINT, Op.getDebugLoc(),
                                      MVT::v8i32, Op.getOperand(0)));
     return SDValue();
@@ -8389,12 +8592,11 @@ SDValue X86TargetLowering::LowerFP_TO_UINT(SDValue Op,
   return FIST;
 }
 
-SDValue X86TargetLowering::lowerFP_EXTEND(SDValue Op,
-                                          SelectionDAG &DAG) const {
+static SDValue LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) {
   DebugLoc DL = Op.getDebugLoc();
-  EVT VT = Op.getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
   SDValue In = Op.getOperand(0);
-  EVT SVT = In.getValueType();
+  MVT SVT = In.getValueType().getSimpleVT();
 
   assert(SVT == MVT::v2f32 && "Only customize MVT::v2f32 type legalization!");
 
@@ -8406,8 +8608,8 @@ SDValue X86TargetLowering::lowerFP_EXTEND(SDValue Op,
 SDValue X86TargetLowering::LowerFABS(SDValue Op, SelectionDAG &DAG) const {
   LLVMContext *Context = DAG.getContext();
   DebugLoc dl = Op.getDebugLoc();
-  EVT VT = Op.getValueType();
-  EVT EltVT = VT;
+  MVT VT = Op.getValueType().getSimpleVT();
+  MVT EltVT = VT;
   unsigned NumElts = VT == MVT::f64 ? 2 : 4;
   if (VT.isVector()) {
     EltVT = VT.getVectorElementType();
@@ -8415,9 +8617,11 @@ SDValue X86TargetLowering::LowerFABS(SDValue Op, SelectionDAG &DAG) const {
   }
   Constant *C;
   if (EltVT == MVT::f64)
-    C = ConstantFP::get(*Context, APFloat(APInt(64, ~(1ULL << 63))));
+    C = ConstantFP::get(*Context, APFloat(APFloat::IEEEdouble,
+                                          APInt(64, ~(1ULL << 63))));
   else
-    C = ConstantFP::get(*Context, APFloat(APInt(32, ~(1U << 31))));
+    C = ConstantFP::get(*Context, APFloat(APFloat::IEEEsingle,
+                                          APInt(32, ~(1U << 31))));
   C = ConstantVector::getSplat(NumElts, C);
   SDValue CPIdx = DAG.getConstantPool(C, getPointerTy());
   unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
@@ -8438,8 +8642,8 @@ SDValue X86TargetLowering::LowerFABS(SDValue Op, SelectionDAG &DAG) const {
 SDValue X86TargetLowering::LowerFNEG(SDValue Op, SelectionDAG &DAG) const {
   LLVMContext *Context = DAG.getContext();
   DebugLoc dl = Op.getDebugLoc();
-  EVT VT = Op.getValueType();
-  EVT EltVT = VT;
+  MVT VT = Op.getValueType().getSimpleVT();
+  MVT EltVT = VT;
   unsigned NumElts = VT == MVT::f64 ? 2 : 4;
   if (VT.isVector()) {
     EltVT = VT.getVectorElementType();
@@ -8447,9 +8651,11 @@ SDValue X86TargetLowering::LowerFNEG(SDValue Op, SelectionDAG &DAG) const {
   }
   Constant *C;
   if (EltVT == MVT::f64)
-    C = ConstantFP::get(*Context, APFloat(APInt(64, 1ULL << 63)));
+    C = ConstantFP::get(*Context, APFloat(APFloat::IEEEdouble,
+                                          APInt(64, 1ULL << 63)));
   else
-    C = ConstantFP::get(*Context, APFloat(APInt(32, 1U << 31)));
+    C = ConstantFP::get(*Context, APFloat(APFloat::IEEEsingle,
+                                          APInt(32, 1U << 31)));
   C = ConstantVector::getSplat(NumElts, C);
   SDValue CPIdx = DAG.getConstantPool(C, getPointerTy());
   unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
@@ -8473,8 +8679,8 @@ SDValue X86TargetLowering::LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const {
   SDValue Op0 = Op.getOperand(0);
   SDValue Op1 = Op.getOperand(1);
   DebugLoc dl = Op.getDebugLoc();
-  EVT VT = Op.getValueType();
-  EVT SrcVT = Op1.getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
+  MVT SrcVT = Op1.getValueType().getSimpleVT();
 
   // If second operand is smaller, extend it first.
   if (SrcVT.bitsLT(VT)) {
@@ -8493,13 +8699,15 @@ SDValue X86TargetLowering::LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const {
   // First get the sign bit of second operand.
   SmallVector<Constant*,4> CV;
   if (SrcVT == MVT::f64) {
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(64, 1ULL << 63))));
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(64, 0))));
+    const fltSemantics &Sem = APFloat::IEEEdouble;
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem, APInt(64, 1ULL << 63))));
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem, APInt(64, 0))));
   } else {
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 1U << 31))));
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
+    const fltSemantics &Sem = APFloat::IEEEsingle;
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem, APInt(32, 1U << 31))));
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem, APInt(32, 0))));
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem, APInt(32, 0))));
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem, APInt(32, 0))));
   }
   Constant *C = ConstantVector::get(CV);
   SDValue CPIdx = DAG.getConstantPool(C, getPointerTy(), 16);
@@ -8522,13 +8730,17 @@ SDValue X86TargetLowering::LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const {
   // Clear first operand sign bit.
   CV.clear();
   if (VT == MVT::f64) {
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(64, ~(1ULL << 63)))));
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(64, 0))));
+    const fltSemantics &Sem = APFloat::IEEEdouble;
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem,
+                                                   APInt(64, ~(1ULL << 63)))));
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem, APInt(64, 0))));
   } else {
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, ~(1U << 31)))));
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
-    CV.push_back(ConstantFP::get(*Context, APFloat(APInt(32, 0))));
+    const fltSemantics &Sem = APFloat::IEEEsingle;
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem,
+                                                   APInt(32, ~(1U << 31)))));
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem, APInt(32, 0))));
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem, APInt(32, 0))));
+    CV.push_back(ConstantFP::get(*Context, APFloat(Sem, APInt(32, 0))));
   }
   C = ConstantVector::get(CV);
   CPIdx = DAG.getConstantPool(C, getPointerTy(), 16);
@@ -8544,7 +8756,7 @@ SDValue X86TargetLowering::LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const {
 static SDValue LowerFGETSIGN(SDValue Op, SelectionDAG &DAG) {
   SDValue N0 = Op.getOperand(0);
   DebugLoc dl = Op.getDebugLoc();
-  EVT VT = Op.getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
 
   // Lower ISD::FGETSIGN to (AND (X86ISD::FGETSIGNx86 ...) 1).
   SDValue xFGETSIGN = DAG.getNode(X86ISD::FGETSIGNx86, dl, VT, N0,
@@ -8554,7 +8766,8 @@ static SDValue LowerFGETSIGN(SDValue Op, SelectionDAG &DAG) {
 
 // LowerVectorAllZeroTest - Check whether an OR'd tree is PTEST-able.
 //
-SDValue X86TargetLowering::LowerVectorAllZeroTest(SDValue Op, SelectionDAG &DAG) const {
+SDValue X86TargetLowering::LowerVectorAllZeroTest(SDValue Op,
+                                                  SelectionDAG &DAG) const {
   assert(Op.getOpcode() == ISD::OR && "Only check OR'd tree.");
 
   if (!Subtarget->hasSSE41())
@@ -8899,6 +9112,11 @@ SDValue X86TargetLowering::ConvertCmpIfNecessary(SDValue Cmp,
   return DAG.getNode(X86ISD::SAHF, dl, MVT::i32, TruncSrl);
 }
 
+static bool isAllOnes(SDValue V) {
+  ConstantSDNode *C = dyn_cast<ConstantSDNode>(V);
+  return C && C->isAllOnesValue();
+}
+
 /// LowerToBT - Result of 'and' is compared against zero. Turn it into a BT node
 /// if it's possible.
 SDValue X86TargetLowering::LowerToBT(SDValue And, ISD::CondCode CC,
@@ -8947,6 +9165,14 @@ SDValue X86TargetLowering::LowerToBT(SDValue And, ISD::CondCode CC,
   }
 
   if (LHS.getNode()) {
+    // If the LHS is of the form (x ^ -1) then replace the LHS with x and flip
+    // the condition code later.
+    bool Invert = false;
+    if (LHS.getOpcode() == ISD::XOR && isAllOnes(LHS.getOperand(1))) {
+      Invert = true;
+      LHS = LHS.getOperand(0);
+    }
+
     // If LHS is i8, promote it to i32 with any_extend.  There is no i8 BT
     // instruction.  Since the shift amount is in-range-or-undefined, we know
     // that doing a bittest on the i32 value is ok.  We extend to i32 because
@@ -8962,7 +9188,10 @@ SDValue X86TargetLowering::LowerToBT(SDValue And, ISD::CondCode CC,
       RHS = DAG.getNode(ISD::ANY_EXTEND, dl, LHS.getValueType(), RHS);
 
     SDValue BT = DAG.getNode(X86ISD::BT, dl, MVT::i32, LHS, RHS);
-    unsigned Cond = CC == ISD::SETEQ ? X86::COND_AE : X86::COND_B;
+    X86::CondCode Cond = CC == ISD::SETEQ ? X86::COND_AE : X86::COND_B;
+    // Flip the condition if the LHS was a not instruction
+    if (Invert)
+      Cond = X86::GetOppositeBranchCondition(Cond);
     return DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
                        DAG.getConstant(Cond, MVT::i8), BT);
   }
@@ -8970,65 +9199,10 @@ SDValue X86TargetLowering::LowerToBT(SDValue And, ISD::CondCode CC,
   return SDValue();
 }
 
-SDValue X86TargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
-
-  if (Op.getValueType().isVector()) return LowerVSETCC(Op, DAG);
-
-  assert(Op.getValueType() == MVT::i8 && "SetCC type must be 8-bit integer");
-  SDValue Op0 = Op.getOperand(0);
-  SDValue Op1 = Op.getOperand(1);
-  DebugLoc dl = Op.getDebugLoc();
-  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
-
-  // Optimize to BT if possible.
-  // Lower (X & (1 << N)) == 0 to BT(X, N).
-  // Lower ((X >>u N) & 1) != 0 to BT(X, N).
-  // Lower ((X >>s N) & 1) != 0 to BT(X, N).
-  if (Op0.getOpcode() == ISD::AND && Op0.hasOneUse() &&
-      Op1.getOpcode() == ISD::Constant &&
-      cast<ConstantSDNode>(Op1)->isNullValue() &&
-      (CC == ISD::SETEQ || CC == ISD::SETNE)) {
-    SDValue NewSetCC = LowerToBT(Op0, CC, dl, DAG);
-    if (NewSetCC.getNode())
-      return NewSetCC;
-  }
-
-  // Look for X == 0, X == 1, X != 0, or X != 1.  We can simplify some forms of
-  // these.
-  if (Op1.getOpcode() == ISD::Constant &&
-      (cast<ConstantSDNode>(Op1)->getZExtValue() == 1 ||
-       cast<ConstantSDNode>(Op1)->isNullValue()) &&
-      (CC == ISD::SETEQ || CC == ISD::SETNE)) {
-
-    // If the input is a setcc, then reuse the input setcc or use a new one with
-    // the inverted condition.
-    if (Op0.getOpcode() == X86ISD::SETCC) {
-      X86::CondCode CCode = (X86::CondCode)Op0.getConstantOperandVal(0);
-      bool Invert = (CC == ISD::SETNE) ^
-        cast<ConstantSDNode>(Op1)->isNullValue();
-      if (!Invert) return Op0;
-
-      CCode = X86::GetOppositeBranchCondition(CCode);
-      return DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
-                         DAG.getConstant(CCode, MVT::i8), Op0.getOperand(1));
-    }
-  }
-
-  bool isFP = Op1.getValueType().isFloatingPoint();
-  unsigned X86CC = TranslateX86CC(CC, isFP, Op0, Op1, DAG);
-  if (X86CC == X86::COND_INVALID)
-    return SDValue();
-
-  SDValue EFLAGS = EmitCmp(Op0, Op1, X86CC, DAG);
-  EFLAGS = ConvertCmpIfNecessary(EFLAGS, DAG);
-  return DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
-                     DAG.getConstant(X86CC, MVT::i8), EFLAGS);
-}
-
 // Lower256IntVSETCC - Break a VSETCC 256-bit integer VSETCC into two new 128
 // ones, and then concatenate the result back.
 static SDValue Lower256IntVSETCC(SDValue Op, SelectionDAG &DAG) {
-  EVT VT = Op.getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
 
   assert(VT.is256BitVector() && Op.getOpcode() == ISD::SETCC &&
          "Unsupported value type for operation");
@@ -9048,27 +9222,27 @@ static SDValue Lower256IntVSETCC(SDValue Op, SelectionDAG &DAG) {
   SDValue RHS2 = Extract128BitVector(RHS, NumElems/2, DAG, dl);
 
   // Issue the operation on the smaller types and concatenate the result back
-  MVT EltVT = VT.getVectorElementType().getSimpleVT();
-  EVT NewVT = MVT::getVectorVT(EltVT, NumElems/2);
+  MVT EltVT = VT.getVectorElementType();
+  MVT NewVT = MVT::getVectorVT(EltVT, NumElems/2);
   return DAG.getNode(ISD::CONCAT_VECTORS, dl, VT,
                      DAG.getNode(Op.getOpcode(), dl, NewVT, LHS1, RHS1, CC),
                      DAG.getNode(Op.getOpcode(), dl, NewVT, LHS2, RHS2, CC));
 }
 
-
-SDValue X86TargetLowering::LowerVSETCC(SDValue Op, SelectionDAG &DAG) const {
+static SDValue LowerVSETCC(SDValue Op, const X86Subtarget *Subtarget,
+                           SelectionDAG &DAG) {
   SDValue Cond;
   SDValue Op0 = Op.getOperand(0);
   SDValue Op1 = Op.getOperand(1);
   SDValue CC = Op.getOperand(2);
-  EVT VT = Op.getValueType();
+  MVT VT = Op.getValueType().getSimpleVT();
   ISD::CondCode SetCCOpcode = cast<CondCodeSDNode>(CC)->get();
-  bool isFP = Op.getOperand(1).getValueType().isFloatingPoint();
+  bool isFP = Op.getOperand(1).getValueType().getSimpleVT().isFloatingPoint();
   DebugLoc dl = Op.getDebugLoc();
 
   if (isFP) {
 #ifndef NDEBUG
-    EVT EltVT = Op0.getValueType().getVectorElementType();
+    MVT EltVT = Op0.getValueType().getVectorElementType().getSimpleVT();
     assert(EltVT == MVT::f32 || EltVT == MVT::f64);
 #endif
 
@@ -9133,7 +9307,7 @@ SDValue X86TargetLowering::LowerVSETCC(SDValue Op, SelectionDAG &DAG) const {
   }
 
   // Break 256-bit integer vector compare into smaller ones.
-  if (VT.is256BitVector() && !Subtarget->hasAVX2())
+  if (VT.is256BitVector() && !Subtarget->hasInt256())
     return Lower256IntVSETCC(Op, DAG);
 
   // We are handling one of the integer comparisons here.  Since SSE only has
@@ -9163,8 +9337,28 @@ SDValue X86TargetLowering::LowerVSETCC(SDValue Op, SelectionDAG &DAG) const {
   if (VT == MVT::v2i64) {
     if (Opc == X86ISD::PCMPGT && !Subtarget->hasSSE42())
       return SDValue();
-    if (Opc == X86ISD::PCMPEQ && !Subtarget->hasSSE41())
-      return SDValue();
+    if (Opc == X86ISD::PCMPEQ && !Subtarget->hasSSE41()) {
+      // If pcmpeqq is missing but pcmpeqd is available synthesize pcmpeqq with
+      // pcmpeqd + pshufd + pand.
+      assert(Subtarget->hasSSE2() && !FlipSigns && "Don't know how to lower!");
+
+      // First cast everything to the right type,
+      Op0 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, Op0);
+      Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, Op1);
+
+      // Do the compare.
+      SDValue Result = DAG.getNode(Opc, dl, MVT::v4i32, Op0, Op1);
+
+      // Make sure the lower and upper halves are both all-ones.
+      const int Mask[] = { 1, 0, 3, 2 };
+      SDValue Shuf = DAG.getVectorShuffle(MVT::v4i32, dl, Result, Result, Mask);
+      Result = DAG.getNode(ISD::AND, dl, MVT::v4i32, Result, Shuf);
+
+      if (Invert)
+        Result = DAG.getNOT(dl, Result, MVT::v4i32);
+
+      return DAG.getNode(ISD::BITCAST, dl, VT, Result);
+    }
   }
 
   // Since SSE has no unsigned integer comparisons, we need to flip  the sign
@@ -9189,6 +9383,63 @@ SDValue X86TargetLowering::LowerVSETCC(SDValue Op, SelectionDAG &DAG) const {
   return Result;
 }
 
+SDValue X86TargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
+
+  MVT VT = Op.getValueType().getSimpleVT();
+
+  if (VT.isVector()) return LowerVSETCC(Op, Subtarget, DAG);
+
+  assert(VT == MVT::i8 && "SetCC type must be 8-bit integer");
+  SDValue Op0 = Op.getOperand(0);
+  SDValue Op1 = Op.getOperand(1);
+  DebugLoc dl = Op.getDebugLoc();
+  ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
+
+  // Optimize to BT if possible.
+  // Lower (X & (1 << N)) == 0 to BT(X, N).
+  // Lower ((X >>u N) & 1) != 0 to BT(X, N).
+  // Lower ((X >>s N) & 1) != 0 to BT(X, N).
+  if (Op0.getOpcode() == ISD::AND && Op0.hasOneUse() &&
+      Op1.getOpcode() == ISD::Constant &&
+      cast<ConstantSDNode>(Op1)->isNullValue() &&
+      (CC == ISD::SETEQ || CC == ISD::SETNE)) {
+    SDValue NewSetCC = LowerToBT(Op0, CC, dl, DAG);
+    if (NewSetCC.getNode())
+      return NewSetCC;
+  }
+
+  // Look for X == 0, X == 1, X != 0, or X != 1.  We can simplify some forms of
+  // these.
+  if (Op1.getOpcode() == ISD::Constant &&
+      (cast<ConstantSDNode>(Op1)->getZExtValue() == 1 ||
+       cast<ConstantSDNode>(Op1)->isNullValue()) &&
+      (CC == ISD::SETEQ || CC == ISD::SETNE)) {
+
+    // If the input is a setcc, then reuse the input setcc or use a new one with
+    // the inverted condition.
+    if (Op0.getOpcode() == X86ISD::SETCC) {
+      X86::CondCode CCode = (X86::CondCode)Op0.getConstantOperandVal(0);
+      bool Invert = (CC == ISD::SETNE) ^
+        cast<ConstantSDNode>(Op1)->isNullValue();
+      if (!Invert) return Op0;
+
+      CCode = X86::GetOppositeBranchCondition(CCode);
+      return DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
+                         DAG.getConstant(CCode, MVT::i8), Op0.getOperand(1));
+    }
+  }
+
+  bool isFP = Op1.getValueType().getSimpleVT().isFloatingPoint();
+  unsigned X86CC = TranslateX86CC(CC, isFP, Op0, Op1, DAG);
+  if (X86CC == X86::COND_INVALID)
+    return SDValue();
+
+  SDValue EFLAGS = EmitCmp(Op0, Op1, X86CC, DAG);
+  EFLAGS = ConvertCmpIfNecessary(EFLAGS, DAG);
+  return DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
+                     DAG.getConstant(X86CC, MVT::i8), EFLAGS);
+}
+
 // isX86LogicalCmp - Return true if opcode is a X86 logical comparison.
 static bool isX86LogicalCmp(SDValue Op) {
   unsigned Opc = Op.getNode()->getOpcode();
@@ -9220,11 +9471,6 @@ static bool isZero(SDValue V) {
   return C && C->isNullValue();
 }
 
-static bool isAllOnes(SDValue V) {
-  ConstantSDNode *C = dyn_cast<ConstantSDNode>(V);
-  return C && C->isAllOnesValue();
-}
-
 static bool isTruncWithZeroHighBitsInput(SDValue V, SelectionDAG &DAG) {
   if (V.getOpcode() != ISD::TRUNCATE)
     return false;
@@ -9316,7 +9562,7 @@ SDValue X86TargetLowering::LowerSELECT(SDValue Op, SelectionDAG &DAG) const {
 
     SDValue Cmp = Cond.getOperand(1);
     unsigned Opc = Cmp.getOpcode();
-    EVT VT = Op.getValueType();
+    MVT VT = Op.getValueType().getSimpleVT();
 
     bool IllegalFPCMov = false;
     if (VT.isFloatingPoint() && !VT.isVector() &&
@@ -9425,6 +9671,53 @@ SDValue X86TargetLowering::LowerSELECT(SDValue Op, SelectionDAG &DAG) const {
   return DAG.getNode(X86ISD::CMOV, DL, VTs, Ops, array_lengthof(Ops));
 }
 
+SDValue X86TargetLowering::LowerSIGN_EXTEND(SDValue Op,
+                                            SelectionDAG &DAG) const {
+  MVT VT = Op->getValueType(0).getSimpleVT();
+  SDValue In = Op->getOperand(0);
+  MVT InVT = In.getValueType().getSimpleVT();
+  DebugLoc dl = Op->getDebugLoc();
+
+  if ((VT != MVT::v4i64 || InVT != MVT::v4i32) &&
+      (VT != MVT::v8i32 || InVT != MVT::v8i16))
+    return SDValue();
+
+  if (Subtarget->hasInt256())
+    return DAG.getNode(X86ISD::VSEXT_MOVL, dl, VT, In);
+
+  // Optimize vectors in AVX mode
+  // Sign extend  v8i16 to v8i32 and
+  //              v4i32 to v4i64
+  //
+  // Divide input vector into two parts
+  // for v4i32 the shuffle mask will be { 0, 1, -1, -1} {2, 3, -1, -1}
+  // use vpmovsx instruction to extend v4i32 -> v2i64; v8i16 -> v4i32
+  // concat the vectors to original VT
+
+  unsigned NumElems = InVT.getVectorNumElements();
+  SDValue Undef = DAG.getUNDEF(InVT);
+
+  SmallVector<int,8> ShufMask1(NumElems, -1);
+  for (unsigned i = 0; i != NumElems/2; ++i)
+    ShufMask1[i] = i;
+
+  SDValue OpLo = DAG.getVectorShuffle(InVT, dl, In, Undef, &ShufMask1[0]);
+
+  SmallVector<int,8> ShufMask2(NumElems, -1);
+  for (unsigned i = 0; i != NumElems/2; ++i)
+    ShufMask2[i] = i + NumElems/2;
+
+  SDValue OpHi = DAG.getVectorShuffle(InVT, dl, In, Undef, &ShufMask2[0]);
+
+  MVT HalfVT = MVT::getVectorVT(VT.getScalarType(),
+                                VT.getVectorNumElements()/2);
+
+  OpLo = DAG.getNode(X86ISD::VSEXT_MOVL, dl, HalfVT, OpLo);
+  OpHi = DAG.getNode(X86ISD::VSEXT_MOVL, dl, HalfVT, OpHi);
+
+  return DAG.getNode(ISD::CONCAT_VECTORS, dl, VT, OpLo, OpHi);
+}
+
 // isAndOrOfSingleUseSetCCs - Return true if node is an ISD::AND or
 // ISD::OR of two X86ISD::SETCC nodes each of which has no other use apart
 // from the AND / OR.
@@ -9713,7 +10006,6 @@ SDValue X86TargetLowering::LowerBRCOND(SDValue Op, SelectionDAG &DAG) const {
                      Chain, Dest, CC, Cond);
 }
 
-
 // Lower dynamic stack allocation to _alloca call for Cygwin/Mingw targets.
 // Calls to _alloca is needed to probe the stack when allocating more than 4k
 // bytes in one go. Touching the stack at 4K increments is necessary to ensure
@@ -9876,8 +10168,9 @@ SDValue X86TargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG) const {
     // Sanity Check: Make sure using fp_offset makes sense.
     assert(!getTargetMachine().Options.UseSoftFloat &&
            !(DAG.getMachineFunction()
-                .getFunction()->getFnAttributes()
-                .hasAttribute(Attributes::NoImplicitFloat)) &&
+                .getFunction()->getAttributes()
+                .hasAttribute(AttributeSet::FunctionIndex,
+                              Attribute::NoImplicitFloat)) &&
            Subtarget->hasSSE1());
   }
 
@@ -9925,7 +10218,7 @@ static SDValue LowerVACOPY(SDValue Op, const X86Subtarget *Subtarget,
                        MachinePointerInfo(DstSV), MachinePointerInfo(SrcSV));
 }
 
-// getTargetVShiftNOde - Handle vector element shifts where the shift amount
+// getTargetVShiftNode - Handle vector element shifts where the shift amount
 // may or may not be a constant. Takes immediate version of shift as input.
 static SDValue getTargetVShiftNode(unsigned Opc, DebugLoc dl, EVT VT,
                                    SDValue SrcOp, SDValue ShAmt,
@@ -10082,6 +10375,14 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) {
     return DAG.getNode(X86ISD::PMULUDQ, dl, Op.getValueType(),
                        Op.getOperand(1), Op.getOperand(2));
 
+  // SSE2/AVX2 sub with unsigned saturation intrinsics
+  case Intrinsic::x86_sse2_psubus_b:
+  case Intrinsic::x86_sse2_psubus_w:
+  case Intrinsic::x86_avx2_psubus_b:
+  case Intrinsic::x86_avx2_psubus_w:
+    return DAG.getNode(X86ISD::SUBUS, dl, Op.getValueType(),
+                       Op.getOperand(1), Op.getOperand(2));
+
   // SSE3/AVX horizontal add/sub intrinsics
   case Intrinsic::x86_sse3_hadd_ps:
   case Intrinsic::x86_sse3_hadd_pd:
@@ -10131,6 +10432,100 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) {
                        Op.getOperand(1), Op.getOperand(2));
   }
 
+  // SSE2/SSE41/AVX2 integer max/min intrinsics.
+  case Intrinsic::x86_sse2_pmaxu_b:
+  case Intrinsic::x86_sse41_pmaxuw:
+  case Intrinsic::x86_sse41_pmaxud:
+  case Intrinsic::x86_avx2_pmaxu_b:
+  case Intrinsic::x86_avx2_pmaxu_w:
+  case Intrinsic::x86_avx2_pmaxu_d:
+  case Intrinsic::x86_sse2_pminu_b:
+  case Intrinsic::x86_sse41_pminuw:
+  case Intrinsic::x86_sse41_pminud:
+  case Intrinsic::x86_avx2_pminu_b:
+  case Intrinsic::x86_avx2_pminu_w:
+  case Intrinsic::x86_avx2_pminu_d:
+  case Intrinsic::x86_sse41_pmaxsb:
+  case Intrinsic::x86_sse2_pmaxs_w:
+  case Intrinsic::x86_sse41_pmaxsd:
+  case Intrinsic::x86_avx2_pmaxs_b:
+  case Intrinsic::x86_avx2_pmaxs_w:
+  case Intrinsic::x86_avx2_pmaxs_d:
+  case Intrinsic::x86_sse41_pminsb:
+  case Intrinsic::x86_sse2_pmins_w:
+  case Intrinsic::x86_sse41_pminsd:
+  case Intrinsic::x86_avx2_pmins_b:
+  case Intrinsic::x86_avx2_pmins_w:
+  case Intrinsic::x86_avx2_pmins_d: {
+    unsigned Opcode;
+    switch (IntNo) {
+    default: llvm_unreachable("Impossible intrinsic");  // Can't reach here.
+    case Intrinsic::x86_sse2_pmaxu_b:
+    case Intrinsic::x86_sse41_pmaxuw:
+    case Intrinsic::x86_sse41_pmaxud:
+    case Intrinsic::x86_avx2_pmaxu_b:
+    case Intrinsic::x86_avx2_pmaxu_w:
+    case Intrinsic::x86_avx2_pmaxu_d:
+      Opcode = X86ISD::UMAX;
+      break;
+    case Intrinsic::x86_sse2_pminu_b:
+    case Intrinsic::x86_sse41_pminuw:
+    case Intrinsic::x86_sse41_pminud:
+    case Intrinsic::x86_avx2_pminu_b:
+    case Intrinsic::x86_avx2_pminu_w:
+    case Intrinsic::x86_avx2_pminu_d:
+      Opcode = X86ISD::UMIN;
+      break;
+    case Intrinsic::x86_sse41_pmaxsb:
+    case Intrinsic::x86_sse2_pmaxs_w:
+    case Intrinsic::x86_sse41_pmaxsd:
+    case Intrinsic::x86_avx2_pmaxs_b:
+    case Intrinsic::x86_avx2_pmaxs_w:
+    case Intrinsic::x86_avx2_pmaxs_d:
+      Opcode = X86ISD::SMAX;
+      break;
+    case Intrinsic::x86_sse41_pminsb:
+    case Intrinsic::x86_sse2_pmins_w:
+    case Intrinsic::x86_sse41_pminsd:
+    case Intrinsic::x86_avx2_pmins_b:
+    case Intrinsic::x86_avx2_pmins_w:
+    case Intrinsic::x86_avx2_pmins_d:
+      Opcode = X86ISD::SMIN;
+      break;
+    }
+    return DAG.getNode(Opcode, dl, Op.getValueType(),
+                       Op.getOperand(1), Op.getOperand(2));
+  }
+
+  // SSE/SSE2/AVX floating point max/min intrinsics.
+  case Intrinsic::x86_sse_max_ps:
+  case Intrinsic::x86_sse2_max_pd:
+  case Intrinsic::x86_avx_max_ps_256:
+  case Intrinsic::x86_avx_max_pd_256:
+  case Intrinsic::x86_sse_min_ps:
+  case Intrinsic::x86_sse2_min_pd:
+  case Intrinsic::x86_avx_min_ps_256:
+  case Intrinsic::x86_avx_min_pd_256: {
+    unsigned Opcode;
+    switch (IntNo) {
+    default: llvm_unreachable("Impossible intrinsic");  // Can't reach here.
+    case Intrinsic::x86_sse_max_ps:
+    case Intrinsic::x86_sse2_max_pd:
+    case Intrinsic::x86_avx_max_ps_256:
+    case Intrinsic::x86_avx_max_pd_256:
+      Opcode = X86ISD::FMAX;
+      break;
+    case Intrinsic::x86_sse_min_ps:
+    case Intrinsic::x86_sse2_min_pd:
+    case Intrinsic::x86_avx_min_ps_256:
+    case Intrinsic::x86_avx_min_pd_256:
+      Opcode = X86ISD::FMIN;
+      break;
+    }
+    return DAG.getNode(Opcode, dl, Op.getValueType(),
+                       Op.getOperand(1), Op.getOperand(2));
+  }
+
   // AVX2 variable shift intrinsics
   case Intrinsic::x86_avx2_psllv_d:
   case Intrinsic::x86_avx2_psllv_q:
@@ -10198,6 +10593,12 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) {
     return DAG.getNode(X86ISD::VPERMV, dl, Op.getValueType(),
                        Op.getOperand(2), Op.getOperand(1));
 
+  case Intrinsic::x86_sse_sqrt_ps:
+  case Intrinsic::x86_sse2_sqrt_pd:
+  case Intrinsic::x86_avx_sqrt_ps_256:
+  case Intrinsic::x86_avx_sqrt_pd_256:
+    return DAG.getNode(ISD::FSQRT, dl, Op.getValueType(), Op.getOperand(1));
+
   // ptest and testp intrinsics. The intrinsic these come from are designed to
   // return an integer value, not just an instruction so lower it to the ptest
   // or testp pattern and a setcc for the result.
@@ -10513,16 +10914,23 @@ static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) {
   switch (IntNo) {
   default: return SDValue();    // Don't custom lower most intrinsics.
 
-  // RDRAND intrinsics.
+  // RDRAND/RDSEED intrinsics.
   case Intrinsic::x86_rdrand_16:
   case Intrinsic::x86_rdrand_32:
-  case Intrinsic::x86_rdrand_64: {
+  case Intrinsic::x86_rdrand_64:
+  case Intrinsic::x86_rdseed_16:
+  case Intrinsic::x86_rdseed_32:
+  case Intrinsic::x86_rdseed_64: {
+    unsigned Opcode = (IntNo == Intrinsic::x86_rdseed_16 ||
+                       IntNo == Intrinsic::x86_rdseed_32 ||
+                       IntNo == Intrinsic::x86_rdseed_64) ? X86ISD::RDSEED :
+                                                            X86ISD::RDRAND;
     // Emit the node with the right value type.
     SDVTList VTs = DAG.getVTList(Op->getValueType(0), MVT::Glue, MVT::Other);
-    SDValue Result = DAG.getNode(X86ISD::RDRAND, dl, VTs, Op.getOperand(0));
+    SDValue Result = DAG.getNode(Opcode, dl, VTs, Op.getOperand(0));
 
-    // If the value returned by RDRAND was valid (CF=1), return 1. Otherwise
-    // return the value from Rand, which is always 0, casted to i32.
+    // If the value returned by RDRAND/RDSEED was valid (CF=1), return 1.
+    // Otherwise return the value from Rand, which is always 0, casted to i32.
     SDValue Ops[] = { DAG.getZExtOrTrunc(Result, dl, Op->getValueType(1)),
                       DAG.getConstant(1, Op->getValueType(1)),
                       DAG.getConstant(X86::COND_B, MVT::i32),
@@ -10535,6 +10943,18 @@ static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) {
     return DAG.getNode(ISD::MERGE_VALUES, dl, Op->getVTList(), Result, isValid,
                        SDValue(Result.getNode(), 2));
   }
+
+  // XTEST intrinsics.
+  case Intrinsic::x86_xtest: {
+    SDVTList VTs = DAG.getVTList(Op->getValueType(0), MVT::Other);
+    SDValue InTrans = DAG.getNode(X86ISD::XTEST, dl, VTs, Op.getOperand(0));
+    SDValue SetCC = DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
+                                DAG.getConstant(X86::COND_NE, MVT::i8),
+                                InTrans);
+    SDValue Ret = DAG.getNode(ISD::ZERO_EXTEND, dl, Op->getValueType(0), SetCC);
+    return DAG.getNode(ISD::MERGE_VALUES, dl, Op->getVTList(),
+                       Ret, SDValue(InTrans.getNode(), 1));
+  }
   }
 }
 
@@ -10710,7 +11130,7 @@ SDValue X86TargetLowering::LowerINIT_TRAMPOLINE(SDValue Op,
 
       // Check that ECX wasn't needed by an 'inreg' parameter.
       FunctionType *FTy = Func->getFunctionType();
-      const AttrListPtr &Attrs = Func->getAttributes();
+      const AttributeSet &Attrs = Func->getAttributes();
 
       if (!Attrs.isEmpty() && !Func->isVarArg()) {
         unsigned InRegCount = 0;
@@ -10718,7 +11138,7 @@ SDValue X86TargetLowering::LowerINIT_TRAMPOLINE(SDValue Op,
 
         for (FunctionType::param_iterator I = FTy->param_begin(),
              E = FTy->param_end(); I != E; ++I, ++Idx)
-          if (Attrs.getParamAttributes(Idx).hasAttribute(Attributes::InReg))
+          if (Attrs.hasAttribute(Idx, Attribute::InReg))
             // FIXME: should only count parameters that are lowered to integers.
             InRegCount += (TD->getTypeSizeInBits(*I) + 31) / 32;
 
@@ -10808,7 +11228,6 @@ SDValue X86TargetLowering::LowerFLT_ROUNDS_(SDValue Op,
   int SSFI = MF.getFrameInfo()->CreateStackObject(2, StackAlignment, false);
   SDValue StackSlot = DAG.getFrameIndex(SSFI, getPointerTy());
 
-
   MachineMemOperand *MMO =
    MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(SSFI),
                            MachineMemOperand::MOStore, 2, 2);
@@ -10841,7 +11260,6 @@ SDValue X86TargetLowering::LowerFLT_ROUNDS_(SDValue Op,
                             DAG.getConstant(1, MVT::i16)),
                 DAG.getConstant(3, MVT::i16));
 
-
   return DAG.getNode((VT.getSizeInBits() < 16 ?
                       ISD::TRUNCATE : ISD::ZERO_EXTEND), DL, VT, RetVal);
 }
@@ -10970,17 +11388,43 @@ static SDValue LowerSUB(SDValue Op, SelectionDAG &DAG) {
 
 static SDValue LowerMUL(SDValue Op, const X86Subtarget *Subtarget,
                         SelectionDAG &DAG) {
+  DebugLoc dl = Op.getDebugLoc();
   EVT VT = Op.getValueType();
 
   // Decompose 256-bit ops into smaller 128-bit ops.
-  if (VT.is256BitVector() && !Subtarget->hasAVX2())
+  if (VT.is256BitVector() && !Subtarget->hasInt256())
     return Lower256IntArith(Op, DAG);
 
+  SDValue A = Op.getOperand(0);
+  SDValue B = Op.getOperand(1);
+
+  // Lower v4i32 mul as 2x shuffle, 2x pmuludq, 2x shuffle.
+  if (VT == MVT::v4i32) {
+    assert(Subtarget->hasSSE2() && !Subtarget->hasSSE41() &&
+           "Should not custom lower when pmuldq is available!");
+
+    // Extract the odd parts.
+    const int UnpackMask[] = { 1, -1, 3, -1 };
+    SDValue Aodds = DAG.getVectorShuffle(VT, dl, A, A, UnpackMask);
+    SDValue Bodds = DAG.getVectorShuffle(VT, dl, B, B, UnpackMask);
+
+    // Multiply the even parts.
+    SDValue Evens = DAG.getNode(X86ISD::PMULUDQ, dl, MVT::v2i64, A, B);
+    // Now multiply odd parts.
+    SDValue Odds = DAG.getNode(X86ISD::PMULUDQ, dl, MVT::v2i64, Aodds, Bodds);
+
+    Evens = DAG.getNode(ISD::BITCAST, dl, VT, Evens);
+    Odds = DAG.getNode(ISD::BITCAST, dl, VT, Odds);
+
+    // Merge the two vectors back together with a shuffle. This expands into 2
+    // shuffles.
+    const int ShufMask[] = { 0, 4, 2, 6 };
+    return DAG.getVectorShuffle(VT, dl, Evens, Odds, ShufMask);
+  }
+
   assert((VT == MVT::v2i64 || VT == MVT::v4i64) &&
          "Only know how to lower V2I64/V4I64 multiply");
 
-  DebugLoc dl = Op.getDebugLoc();
-
   //  Ahi = psrlqi(a, 32);
   //  Bhi = psrlqi(b, 32);
   //
@@ -10992,9 +11436,6 @@ static SDValue LowerMUL(SDValue Op, const X86Subtarget *Subtarget,
   //  AhiBlo = psllqi(AhiBlo, 32);
   //  return AloBlo + AloBhi + AhiBlo;
 
-  SDValue A = Op.getOperand(0);
-  SDValue B = Op.getOperand(1);
-
   SDValue ShAmt = DAG.getConstant(32, MVT::i32);
 
   SDValue Ahi = DAG.getNode(X86ISD::VSRLI, dl, VT, A, ShAmt);
@@ -11018,16 +11459,55 @@ static SDValue LowerMUL(SDValue Op, const X86Subtarget *Subtarget,
   return DAG.getNode(ISD::ADD, dl, VT, Res, AhiBlo);
 }
 
-SDValue X86TargetLowering::LowerShift(SDValue Op, SelectionDAG &DAG) const {
+SDValue X86TargetLowering::LowerSDIV(SDValue Op, SelectionDAG &DAG) const {
+  EVT VT = Op.getValueType();
+  EVT EltTy = VT.getVectorElementType();
+  unsigned NumElts = VT.getVectorNumElements();
+  SDValue N0 = Op.getOperand(0);
+  DebugLoc dl = Op.getDebugLoc();
+
+  // Lower sdiv X, pow2-const.
+  BuildVectorSDNode *C = dyn_cast<BuildVectorSDNode>(Op.getOperand(1));
+  if (!C)
+    return SDValue();
+
+  APInt SplatValue, SplatUndef;
+  unsigned MinSplatBits;
+  bool HasAnyUndefs;
+  if (!C->isConstantSplat(SplatValue, SplatUndef, MinSplatBits, HasAnyUndefs))
+    return SDValue();
+
+  if ((SplatValue != 0) &&
+      (SplatValue.isPowerOf2() || (-SplatValue).isPowerOf2())) {
+    unsigned lg2 = SplatValue.countTrailingZeros();
+    // Splat the sign bit.
+    SDValue Sz = DAG.getConstant(EltTy.getSizeInBits()-1, MVT::i32);
+    SDValue SGN = getTargetVShiftNode(X86ISD::VSRAI, dl, VT, N0, Sz, DAG);
+    // Add (N0 < 0) ? abs2 - 1 : 0;
+    SDValue Amt = DAG.getConstant(EltTy.getSizeInBits() - lg2, MVT::i32);
+    SDValue SRL = getTargetVShiftNode(X86ISD::VSRLI, dl, VT, SGN, Amt, DAG);
+    SDValue ADD = DAG.getNode(ISD::ADD, dl, VT, N0, SRL);
+    SDValue Lg2Amt = DAG.getConstant(lg2, MVT::i32);
+    SDValue SRA = getTargetVShiftNode(X86ISD::VSRAI, dl, VT, ADD, Lg2Amt, DAG);
+
+    // If we're dividing by a positive value, we're done.  Otherwise, we must
+    // negate the result.
+    if (SplatValue.isNonNegative())
+      return SRA;
+
+    SmallVector<SDValue, 16> V(NumElts, DAG.getConstant(0, EltTy));
+    SDValue Zero = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &V[0], NumElts);
+    return DAG.getNode(ISD::SUB, dl, VT, Zero, SRA);
+  }
+  return SDValue();
+}
 
+static SDValue LowerScalarImmediateShift(SDValue Op, SelectionDAG &DAG,
+                                         const X86Subtarget *Subtarget) {
   EVT VT = Op.getValueType();
   DebugLoc dl = Op.getDebugLoc();
   SDValue R = Op.getOperand(0);
   SDValue Amt = Op.getOperand(1);
-  LLVMContext *Context = DAG.getContext();
-
-  if (!Subtarget->hasSSE2())
-    return SDValue();
 
   // Optimize shl/srl/sra with constant shift amount.
   if (isSplatVector(Amt.getNode())) {
@@ -11036,7 +11516,7 @@ SDValue X86TargetLowering::LowerShift(SDValue Op, SelectionDAG &DAG) const {
       uint64_t ShiftAmt = C->getZExtValue();
 
       if (VT == MVT::v2i64 || VT == MVT::v4i32 || VT == MVT::v8i16 ||
-          (Subtarget->hasAVX2() &&
+          (Subtarget->hasInt256() &&
            (VT == MVT::v4i64 || VT == MVT::v8i32 || VT == MVT::v16i16))) {
         if (Op.getOpcode() == ISD::SHL)
           return DAG.getNode(X86ISD::VSHLI, dl, VT, R,
@@ -11093,7 +11573,7 @@ SDValue X86TargetLowering::LowerShift(SDValue Op, SelectionDAG &DAG) const {
         llvm_unreachable("Unknown shift opcode.");
       }
 
-      if (Subtarget->hasAVX2() && VT == MVT::v32i8) {
+      if (Subtarget->hasInt256() && VT == MVT::v32i8) {
         if (Op.getOpcode() == ISD::SHL) {
           // Make a large shift.
           SDValue SHL = DAG.getNode(X86ISD::VSHLI, dl, MVT::v16i16, R,
@@ -11139,19 +11619,229 @@ SDValue X86TargetLowering::LowerShift(SDValue Op, SelectionDAG &DAG) const {
     }
   }
 
+  // Special case in 32-bit mode, where i64 is expanded into high and low parts.
+  if (!Subtarget->is64Bit() &&
+      (VT == MVT::v2i64 || (Subtarget->hasInt256() && VT == MVT::v4i64)) &&
+      Amt.getOpcode() == ISD::BITCAST &&
+      Amt.getOperand(0).getOpcode() == ISD::BUILD_VECTOR) {
+    Amt = Amt.getOperand(0);
+    unsigned Ratio = Amt.getValueType().getVectorNumElements() /
+                     VT.getVectorNumElements();
+    unsigned RatioInLog2 = Log2_32_Ceil(Ratio);
+    uint64_t ShiftAmt = 0;
+    for (unsigned i = 0; i != Ratio; ++i) {
+      ConstantSDNode *C = dyn_cast<ConstantSDNode>(Amt.getOperand(i));
+      if (C == 0)
+        return SDValue();
+      // 6 == Log2(64)
+      ShiftAmt |= C->getZExtValue() << (i * (1 << (6 - RatioInLog2)));
+    }
+    // Check remaining shift amounts.
+    for (unsigned i = Ratio; i != Amt.getNumOperands(); i += Ratio) {
+      uint64_t ShAmt = 0;
+      for (unsigned j = 0; j != Ratio; ++j) {
+        ConstantSDNode *C =
+          dyn_cast<ConstantSDNode>(Amt.getOperand(i + j));
+        if (C == 0)
+          return SDValue();
+        // 6 == Log2(64)
+        ShAmt |= C->getZExtValue() << (j * (1 << (6 - RatioInLog2)));
+      }
+      if (ShAmt != ShiftAmt)
+        return SDValue();
+    }
+    switch (Op.getOpcode()) {
+    default:
+      llvm_unreachable("Unknown shift opcode!");
+    case ISD::SHL:
+      return DAG.getNode(X86ISD::VSHLI, dl, VT, R,
+                         DAG.getConstant(ShiftAmt, MVT::i32));
+    case ISD::SRL:
+      return DAG.getNode(X86ISD::VSRLI, dl, VT, R,
+                         DAG.getConstant(ShiftAmt, MVT::i32));
+    case ISD::SRA:
+      return DAG.getNode(X86ISD::VSRAI, dl, VT, R,
+                         DAG.getConstant(ShiftAmt, MVT::i32));
+    }
+  }
+
+  return SDValue();
+}
+
+static SDValue LowerScalarVariableShift(SDValue Op, SelectionDAG &DAG,
+                                        const X86Subtarget* Subtarget) {
+  EVT VT = Op.getValueType();
+  DebugLoc dl = Op.getDebugLoc();
+  SDValue R = Op.getOperand(0);
+  SDValue Amt = Op.getOperand(1);
+
+  if ((VT == MVT::v2i64 && Op.getOpcode() != ISD::SRA) ||
+      VT == MVT::v4i32 || VT == MVT::v8i16 ||
+      (Subtarget->hasInt256() &&
+       ((VT == MVT::v4i64 && Op.getOpcode() != ISD::SRA) ||
+        VT == MVT::v8i32 || VT == MVT::v16i16))) {
+    SDValue BaseShAmt;
+    EVT EltVT = VT.getVectorElementType();
+
+    if (Amt.getOpcode() == ISD::BUILD_VECTOR) {
+      unsigned NumElts = VT.getVectorNumElements();
+      unsigned i, j;
+      for (i = 0; i != NumElts; ++i) {
+        if (Amt.getOperand(i).getOpcode() == ISD::UNDEF)
+          continue;
+        break;
+      }
+      for (j = i; j != NumElts; ++j) {
+        SDValue Arg = Amt.getOperand(j);
+        if (Arg.getOpcode() == ISD::UNDEF) continue;
+        if (Arg != Amt.getOperand(i))
+          break;
+      }
+      if (i != NumElts && j == NumElts)
+        BaseShAmt = Amt.getOperand(i);
+    } else {
+      if (Amt.getOpcode() == ISD::EXTRACT_SUBVECTOR)
+        Amt = Amt.getOperand(0);
+      if (Amt.getOpcode() == ISD::VECTOR_SHUFFLE &&
+               cast<ShuffleVectorSDNode>(Amt)->isSplat()) {
+        SDValue InVec = Amt.getOperand(0);
+        if (InVec.getOpcode() == ISD::BUILD_VECTOR) {
+          unsigned NumElts = InVec.getValueType().getVectorNumElements();
+          unsigned i = 0;
+          for (; i != NumElts; ++i) {
+            SDValue Arg = InVec.getOperand(i);
+            if (Arg.getOpcode() == ISD::UNDEF) continue;
+            BaseShAmt = Arg;
+            break;
+          }
+        } else if (InVec.getOpcode() == ISD::INSERT_VECTOR_ELT) {
+           if (ConstantSDNode *C =
+               dyn_cast<ConstantSDNode>(InVec.getOperand(2))) {
+             unsigned SplatIdx =
+               cast<ShuffleVectorSDNode>(Amt)->getSplatIndex();
+             if (C->getZExtValue() == SplatIdx)
+               BaseShAmt = InVec.getOperand(1);
+           }
+        }
+        if (BaseShAmt.getNode() == 0)
+          BaseShAmt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, Amt,
+                                  DAG.getIntPtrConstant(0));
+      }
+    }
+
+    if (BaseShAmt.getNode()) {
+      if (EltVT.bitsGT(MVT::i32))
+        BaseShAmt = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, BaseShAmt);
+      else if (EltVT.bitsLT(MVT::i32))
+        BaseShAmt = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, BaseShAmt);
+
+      switch (Op.getOpcode()) {
+      default:
+        llvm_unreachable("Unknown shift opcode!");
+      case ISD::SHL:
+        switch (VT.getSimpleVT().SimpleTy) {
+        default: return SDValue();
+        case MVT::v2i64:
+        case MVT::v4i32:
+        case MVT::v8i16:
+        case MVT::v4i64:
+        case MVT::v8i32:
+        case MVT::v16i16:
+          return getTargetVShiftNode(X86ISD::VSHLI, dl, VT, R, BaseShAmt, DAG);
+        }
+      case ISD::SRA:
+        switch (VT.getSimpleVT().SimpleTy) {
+        default: return SDValue();
+        case MVT::v4i32:
+        case MVT::v8i16:
+        case MVT::v8i32:
+        case MVT::v16i16:
+          return getTargetVShiftNode(X86ISD::VSRAI, dl, VT, R, BaseShAmt, DAG);
+        }
+      case ISD::SRL:
+        switch (VT.getSimpleVT().SimpleTy) {
+        default: return SDValue();
+        case MVT::v2i64:
+        case MVT::v4i32:
+        case MVT::v8i16:
+        case MVT::v4i64:
+        case MVT::v8i32:
+        case MVT::v16i16:
+          return getTargetVShiftNode(X86ISD::VSRLI, dl, VT, R, BaseShAmt, DAG);
+        }
+      }
+    }
+  }
+
+  // Special case in 32-bit mode, where i64 is expanded into high and low parts.
+  if (!Subtarget->is64Bit() &&
+      (VT == MVT::v2i64 || (Subtarget->hasInt256() && VT == MVT::v4i64)) &&
+      Amt.getOpcode() == ISD::BITCAST &&
+      Amt.getOperand(0).getOpcode() == ISD::BUILD_VECTOR) {
+    Amt = Amt.getOperand(0);
+    unsigned Ratio = Amt.getValueType().getVectorNumElements() /
+                     VT.getVectorNumElements();
+    std::vector<SDValue> Vals(Ratio);
+    for (unsigned i = 0; i != Ratio; ++i)
+      Vals[i] = Amt.getOperand(i);
+    for (unsigned i = Ratio; i != Amt.getNumOperands(); i += Ratio) {
+      for (unsigned j = 0; j != Ratio; ++j)
+        if (Vals[j] != Amt.getOperand(i + j))
+          return SDValue();
+    }
+    switch (Op.getOpcode()) {
+    default:
+      llvm_unreachable("Unknown shift opcode!");
+    case ISD::SHL:
+      return DAG.getNode(X86ISD::VSHL, dl, VT, R, Op.getOperand(1));
+    case ISD::SRL:
+      return DAG.getNode(X86ISD::VSRL, dl, VT, R, Op.getOperand(1));
+    case ISD::SRA:
+      return DAG.getNode(X86ISD::VSRA, dl, VT, R, Op.getOperand(1));
+    }
+  }
+
+  return SDValue();
+}
+
+SDValue X86TargetLowering::LowerShift(SDValue Op, SelectionDAG &DAG) const {
+
+  EVT VT = Op.getValueType();
+  DebugLoc dl = Op.getDebugLoc();
+  SDValue R = Op.getOperand(0);
+  SDValue Amt = Op.getOperand(1);
+  SDValue V;
+
+  if (!Subtarget->hasSSE2())
+    return SDValue();
+
+  V = LowerScalarImmediateShift(Op, DAG, Subtarget);
+  if (V.getNode())
+    return V;
+
+  V = LowerScalarVariableShift(Op, DAG, Subtarget);
+  if (V.getNode())
+      return V;
+
+  // AVX2 has VPSLLV/VPSRAV/VPSRLV.
+  if (Subtarget->hasInt256()) {
+    if (Op.getOpcode() == ISD::SRL &&
+        (VT == MVT::v2i64 || VT == MVT::v4i32 ||
+         VT == MVT::v4i64 || VT == MVT::v8i32))
+      return Op;
+    if (Op.getOpcode() == ISD::SHL &&
+        (VT == MVT::v2i64 || VT == MVT::v4i32 ||
+         VT == MVT::v4i64 || VT == MVT::v8i32))
+      return Op;
+    if (Op.getOpcode() == ISD::SRA && (VT == MVT::v4i32 || VT == MVT::v8i32))
+      return Op;
+  }
+
   // Lower SHL with variable shift amount.
   if (VT == MVT::v4i32 && Op->getOpcode() == ISD::SHL) {
-    Op = DAG.getNode(X86ISD::VSHLI, dl, VT, Op.getOperand(1),
-                     DAG.getConstant(23, MVT::i32));
+    Op = DAG.getNode(ISD::SHL, dl, VT, Amt, DAG.getConstant(23, VT));
 
-    const uint32_t CV[] = { 0x3f800000U, 0x3f800000U, 0x3f800000U, 0x3f800000U};
-    Constant *C = ConstantDataVector::get(*Context, CV);
-    SDValue CPIdx = DAG.getConstantPool(C, getPointerTy(), 16);
-    SDValue Addend = DAG.getLoad(VT, dl, DAG.getEntryNode(), CPIdx,
-                                 MachinePointerInfo::getConstantPool(),
-                                 false, false, false, 16);
-
-    Op = DAG.getNode(ISD::ADD, dl, VT, Op, Addend);
+    Op = DAG.getNode(ISD::ADD, dl, VT, Op, DAG.getConstant(0x3f800000U, VT));
     Op = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32, Op);
     Op = DAG.getNode(ISD::FP_TO_SINT, dl, VT, Op);
     return DAG.getNode(ISD::MUL, dl, VT, Op, R);
@@ -11160,8 +11850,7 @@ SDValue X86TargetLowering::LowerShift(SDValue Op, SelectionDAG &DAG) const {
     assert(Subtarget->hasSSE2() && "Need SSE2 for pslli/pcmpeq.");
 
     // a = a << 5;
-    Op = DAG.getNode(X86ISD::VSHLI, dl, MVT::v8i16, Op.getOperand(1),
-                     DAG.getConstant(5, MVT::i32));
+    Op = DAG.getNode(ISD::SHL, dl, VT, Amt, DAG.getConstant(5, VT));
     Op = DAG.getNode(ISD::BITCAST, dl, VT, Op);
 
     // Turn 'a' into a mask suitable for VSELECT
@@ -11336,9 +12025,9 @@ SDValue X86TargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op,
     default: return SDValue();
     case MVT::v8i32:
     case MVT::v16i16:
-      if (!Subtarget->hasAVX())
+      if (!Subtarget->hasFp256())
         return SDValue();
-      if (!Subtarget->hasAVX2()) {
+      if (!Subtarget->hasInt256()) {
         // needs to be split
         unsigned NumElems = VT.getVectorNumElements();
 
@@ -11364,14 +12053,28 @@ SDValue X86TargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op,
       // fall through
     case MVT::v4i32:
     case MVT::v8i16: {
-      SDValue Tmp1 = getTargetVShiftNode(X86ISD::VSHLI, dl, VT,
-                                         Op.getOperand(0), ShAmt, DAG);
+      // (sext (vzext x)) -> (vsext x)
+      SDValue Op0 = Op.getOperand(0);
+      SDValue Op00 = Op0.getOperand(0);
+      SDValue Tmp1;
+      // Hopefully, this VECTOR_SHUFFLE is just a VZEXT.
+      if (Op0.getOpcode() == ISD::BITCAST &&
+          Op00.getOpcode() == ISD::VECTOR_SHUFFLE)
+        Tmp1 = LowerVectorIntExtend(Op00, DAG);
+      if (Tmp1.getNode()) {
+        SDValue Tmp1Op0 = Tmp1.getOperand(0);
+        assert(Tmp1Op0.getOpcode() == X86ISD::VZEXT &&
+               "This optimization is invalid without a VZEXT.");
+        return DAG.getNode(X86ISD::VSEXT, dl, VT, Tmp1Op0.getOperand(0));
+      }
+
+      // If the above didn't work, then just use Shift-Left + Shift-Right.
+      Tmp1 = getTargetVShiftNode(X86ISD::VSHLI, dl, VT, Op0, ShAmt, DAG);
       return getTargetVShiftNode(X86ISD::VSRAI, dl, VT, Tmp1, ShAmt, DAG);
     }
   }
 }
 
-
 static SDValue LowerMEMBARRIER(SDValue Op, const X86Subtarget *Subtarget,
                               SelectionDAG &DAG) {
   DebugLoc dl = Op.getDebugLoc();
@@ -11456,7 +12159,6 @@ static SDValue LowerATOMIC_FENCE(SDValue Op, const X86Subtarget *Subtarget,
   return DAG.getNode(X86ISD::MEMBARRIER, dl, MVT::Other, Op.getOperand(0));
 }
 
-
 static SDValue LowerCMP_SWAP(SDValue Op, const X86Subtarget *Subtarget,
                              SelectionDAG &DAG) {
   EVT T = Op.getValueType();
@@ -11595,6 +12297,43 @@ static SDValue LowerADDC_ADDE_SUBC_SUBE(SDValue Op, SelectionDAG &DAG) {
                      Op.getOperand(1), Op.getOperand(2));
 }
 
+SDValue X86TargetLowering::LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const {
+  assert(Subtarget->isTargetDarwin() && Subtarget->is64Bit());
+
+  // For MacOSX, we want to call an alternative entry point: __sincos_stret,
+  // which returns the values in two XMM registers.
+  DebugLoc dl = Op.getDebugLoc();
+  SDValue Arg = Op.getOperand(0);
+  EVT ArgVT = Arg.getValueType();
+  Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
+
+  ArgListTy Args;
+  ArgListEntry Entry;
+
+  Entry.Node = Arg;
+  Entry.Ty = ArgTy;
+  Entry.isSExt = false;
+  Entry.isZExt = false;
+  Args.push_back(Entry);
+
+  // Only optimize x86_64 for now. i386 is a bit messy. For f32,
+  // the small struct {f32, f32} is returned in (eax, edx). For f64,
+  // the results are returned via SRet in memory.
+  const char *LibcallName = (ArgVT == MVT::f64)
+    ? "__sincos_stret" : "__sincosf_stret";
+  SDValue Callee = DAG.getExternalSymbol(LibcallName, getPointerTy());
+
+  StructType *RetTy = StructType::get(ArgTy, ArgTy, NULL);
+  TargetLowering::
+    CallLoweringInfo CLI(DAG.getEntryNode(), RetTy,
+                         false, false, false, false, 0,
+                         CallingConv::C, /*isTaillCall=*/false,
+                         /*doesNotRet=*/false, /*isReturnValueUsed*/true,
+                         Callee, Args, DAG, dl);
+  std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
+  return CallResult.first;
+}
+
 /// LowerOperation - Provide custom lowering hooks for some operations.
 ///
 SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
@@ -11624,11 +12363,13 @@ SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::SRL_PARTS:          return LowerShiftParts(Op, DAG);
   case ISD::SINT_TO_FP:         return LowerSINT_TO_FP(Op, DAG);
   case ISD::UINT_TO_FP:         return LowerUINT_TO_FP(Op, DAG);
-  case ISD::TRUNCATE:           return lowerTRUNCATE(Op, DAG);
-  case ISD::ZERO_EXTEND:        return lowerZERO_EXTEND(Op, DAG);
+  case ISD::TRUNCATE:           return LowerTRUNCATE(Op, DAG);
+  case ISD::ZERO_EXTEND:        return LowerZERO_EXTEND(Op, DAG);
+  case ISD::SIGN_EXTEND:        return LowerSIGN_EXTEND(Op, DAG);
+  case ISD::ANY_EXTEND:         return LowerANY_EXTEND(Op, DAG);
   case ISD::FP_TO_SINT:         return LowerFP_TO_SINT(Op, DAG);
   case ISD::FP_TO_UINT:         return LowerFP_TO_UINT(Op, DAG);
-  case ISD::FP_EXTEND:          return lowerFP_EXTEND(Op, DAG);
+  case ISD::FP_EXTEND:          return LowerFP_EXTEND(Op, DAG);
   case ISD::FABS:               return LowerFABS(Op, DAG);
   case ISD::FNEG:               return LowerFNEG(Op, DAG);
   case ISD::FCOPYSIGN:          return LowerFCOPYSIGN(Op, DAG);
@@ -11674,6 +12415,8 @@ SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::SUBE:               return LowerADDC_ADDE_SUBC_SUBE(Op, DAG);
   case ISD::ADD:                return LowerADD(Op, DAG);
   case ISD::SUB:                return LowerSUB(Op, DAG);
+  case ISD::SDIV:               return LowerSDIV(Op, DAG);
+  case ISD::FSINCOS:            return LowerFSINCOS(Op, DAG);
   }
 }
 
@@ -11727,6 +12470,7 @@ void X86TargetLowering::ReplaceNodeResults(SDNode *N,
                                            SmallVectorImpl<SDValue>&Results,
                                            SelectionDAG &DAG) const {
   DebugLoc dl = N->getDebugLoc();
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   switch (N->getOpcode()) {
   default:
     llvm_unreachable("Do not know how to custom type legalize this operation!");
@@ -11760,7 +12504,8 @@ void X86TargetLowering::ReplaceNodeResults(SDNode *N,
     return;
   }
   case ISD::UINT_TO_FP: {
-    if (N->getOperand(0).getValueType() != MVT::v2i32 &&
+    assert(Subtarget->hasSSE2() && "Requires at least SSE2!");
+    if (N->getOperand(0).getValueType() != MVT::v2i32 ||
         N->getValueType(0) != MVT::v2f32)
       return;
     SDValue ZExtIn = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::v2i64,
@@ -11776,6 +12521,8 @@ void X86TargetLowering::ReplaceNodeResults(SDNode *N,
     return;
   }
   case ISD::FP_ROUND: {
+    if (!TLI.isTypeLegal(N->getOperand(0).getValueType()))
+        return;
     SDValue V = DAG.getNode(X86ISD::VFPROUND, dl, MVT::v4f32, N->getOperand(0));
     Results.push_back(V);
     return;
@@ -11942,13 +12689,16 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::ANDNP:              return "X86ISD::ANDNP";
   case X86ISD::PSIGN:              return "X86ISD::PSIGN";
   case X86ISD::BLENDV:             return "X86ISD::BLENDV";
-  case X86ISD::BLENDPW:            return "X86ISD::BLENDPW";
-  case X86ISD::BLENDPS:            return "X86ISD::BLENDPS";
-  case X86ISD::BLENDPD:            return "X86ISD::BLENDPD";
+  case X86ISD::BLENDI:             return "X86ISD::BLENDI";
+  case X86ISD::SUBUS:              return "X86ISD::SUBUS";
   case X86ISD::HADD:               return "X86ISD::HADD";
   case X86ISD::HSUB:               return "X86ISD::HSUB";
   case X86ISD::FHADD:              return "X86ISD::FHADD";
   case X86ISD::FHSUB:              return "X86ISD::FHSUB";
+  case X86ISD::UMAX:               return "X86ISD::UMAX";
+  case X86ISD::UMIN:               return "X86ISD::UMIN";
+  case X86ISD::SMAX:               return "X86ISD::SMAX";
+  case X86ISD::SMIN:               return "X86ISD::SMIN";
   case X86ISD::FMAX:               return "X86ISD::FMAX";
   case X86ISD::FMIN:               return "X86ISD::FMIN";
   case X86ISD::FMAXC:              return "X86ISD::FMAXC";
@@ -12001,14 +12751,13 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::OR:                 return "X86ISD::OR";
   case X86ISD::XOR:                return "X86ISD::XOR";
   case X86ISD::AND:                return "X86ISD::AND";
-  case X86ISD::ANDN:               return "X86ISD::ANDN";
   case X86ISD::BLSI:               return "X86ISD::BLSI";
   case X86ISD::BLSMSK:             return "X86ISD::BLSMSK";
   case X86ISD::BLSR:               return "X86ISD::BLSR";
   case X86ISD::MUL_IMM:            return "X86ISD::MUL_IMM";
   case X86ISD::PTEST:              return "X86ISD::PTEST";
   case X86ISD::TESTP:              return "X86ISD::TESTP";
-  case X86ISD::PALIGN:             return "X86ISD::PALIGN";
+  case X86ISD::PALIGNR:            return "X86ISD::PALIGNR";
   case X86ISD::PSHUFD:             return "X86ISD::PSHUFD";
   case X86ISD::PSHUFHW:            return "X86ISD::PSHUFHW";
   case X86ISD::PSHUFLW:            return "X86ISD::PSHUFLW";
@@ -12039,6 +12788,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::WIN_FTOL:           return "X86ISD::WIN_FTOL";
   case X86ISD::SAHF:               return "X86ISD::SAHF";
   case X86ISD::RDRAND:             return "X86ISD::RDRAND";
+  case X86ISD::RDSEED:             return "X86ISD::RDSEED";
   case X86ISD::FMADD:              return "X86ISD::FMADD";
   case X86ISD::FMSUB:              return "X86ISD::FMSUB";
   case X86ISD::FNMADD:             return "X86ISD::FNMADD";
@@ -12047,6 +12797,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::FMSUBADD:           return "X86ISD::FMSUBADD";
   case X86ISD::PCMPESTRI:          return "X86ISD::PCMPESTRI";
   case X86ISD::PCMPISTRI:          return "X86ISD::PCMPISTRI";
+  case X86ISD::XTEST:              return "X86ISD::XTEST";
   }
 }
 
@@ -12104,24 +12855,21 @@ bool X86TargetLowering::isLegalAddressingMode(const AddrMode &AM,
   return true;
 }
 
-
 bool X86TargetLowering::isTruncateFree(Type *Ty1, Type *Ty2) const {
   if (!Ty1->isIntegerTy() || !Ty2->isIntegerTy())
     return false;
   unsigned NumBits1 = Ty1->getPrimitiveSizeInBits();
   unsigned NumBits2 = Ty2->getPrimitiveSizeInBits();
-  if (NumBits1 <= NumBits2)
-    return false;
-  return true;
+  return NumBits1 > NumBits2;
 }
 
 bool X86TargetLowering::isLegalICmpImmediate(int64_t Imm) const {
-  return Imm == (int32_t)Imm;
+  return isInt<32>(Imm);
 }
 
 bool X86TargetLowering::isLegalAddImmediate(int64_t Imm) const {
   // Can also use sub to handle negated immediates.
-  return Imm == (int32_t)Imm;
+  return isInt<32>(Imm);
 }
 
 bool X86TargetLowering::isTruncateFree(EVT VT1, EVT VT2) const {
@@ -12129,9 +12877,7 @@ bool X86TargetLowering::isTruncateFree(EVT VT1, EVT VT2) const {
     return false;
   unsigned NumBits1 = VT1.getSizeInBits();
   unsigned NumBits2 = VT2.getSizeInBits();
-  if (NumBits1 <= NumBits2)
-    return false;
-  return true;
+  return NumBits1 > NumBits2;
 }
 
 bool X86TargetLowering::isZExtFree(Type *Ty1, Type *Ty2) const {
@@ -12144,6 +12890,30 @@ bool X86TargetLowering::isZExtFree(EVT VT1, EVT VT2) const {
   return VT1 == MVT::i32 && VT2 == MVT::i64 && Subtarget->is64Bit();
 }
 
+bool X86TargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
+  EVT VT1 = Val.getValueType();
+  if (isZExtFree(VT1, VT2))
+    return true;
+
+  if (Val.getOpcode() != ISD::LOAD)
+    return false;
+
+  if (!VT1.isSimple() || !VT1.isInteger() ||
+      !VT2.isSimple() || !VT2.isInteger())
+    return false;
+
+  switch (VT1.getSimpleVT().SimpleTy) {
+  default: break;
+  case MVT::i8:
+  case MVT::i16:
+  case MVT::i32:
+    // X86 has 8, 16, and 32-bit zero-extending loads.
+    return true;
+  }
+
+  return false;
+}
+
 bool X86TargetLowering::isNarrowingProfitable(EVT VT1, EVT VT2) const {
   // i16 instructions are longer (0x66 prefix) and potentially slower.
   return !(VT1 == MVT::i32 && VT2 == MVT::i16);
@@ -12164,15 +12934,15 @@ X86TargetLowering::isShuffleMaskLegal(const SmallVectorImpl<int> &M,
   return (VT.getVectorNumElements() == 2 ||
           ShuffleVectorSDNode::isSplatMask(&M[0], VT) ||
           isMOVLMask(M, VT) ||
-          isSHUFPMask(M, VT, Subtarget->hasAVX()) ||
+          isSHUFPMask(M, VT, Subtarget->hasFp256()) ||
           isPSHUFDMask(M, VT) ||
-          isPSHUFHWMask(M, VT, Subtarget->hasAVX2()) ||
-          isPSHUFLWMask(M, VT, Subtarget->hasAVX2()) ||
+          isPSHUFHWMask(M, VT, Subtarget->hasInt256()) ||
+          isPSHUFLWMask(M, VT, Subtarget->hasInt256()) ||
           isPALIGNRMask(M, VT, Subtarget) ||
-          isUNPCKLMask(M, VT, Subtarget->hasAVX2()) ||
-          isUNPCKHMask(M, VT, Subtarget->hasAVX2()) ||
-          isUNPCKL_v_undef_Mask(M, VT, Subtarget->hasAVX2()) ||
-          isUNPCKH_v_undef_Mask(M, VT, Subtarget->hasAVX2()));
+          isUNPCKLMask(M, VT, Subtarget->hasInt256()) ||
+          isUNPCKHMask(M, VT, Subtarget->hasInt256()) ||
+          isUNPCKL_v_undef_Mask(M, VT, Subtarget->hasInt256()) ||
+          isUNPCKH_v_undef_Mask(M, VT, Subtarget->hasInt256()));
 }
 
 bool
@@ -12185,8 +12955,8 @@ X86TargetLowering::isVectorClearMaskLegal(const SmallVectorImpl<int> &Mask,
   if (NumElts == 4 && VT.is128BitVector()) {
     return (isMOVLMask(Mask, VT)  ||
             isCommutedMOVLMask(Mask, VT, true) ||
-            isSHUFPMask(Mask, VT, Subtarget->hasAVX()) ||
-            isSHUFPMask(Mask, VT, Subtarget->hasAVX(), /* Commuted */ true));
+            isSHUFPMask(Mask, VT, Subtarget->hasFp256()) ||
+            isSHUFPMask(Mask, VT, Subtarget->hasFp256(), /* Commuted */ true));
   }
   return false;
 }
@@ -12379,13 +13149,16 @@ static unsigned getPseudoCMOVOpc(EVT VT) {
 // to
 //
 //    ...
-//    EAX = LOAD MI.addr
+//    t1 = LOAD MI.addr
 // loop:
-//    t1 = OP MI.val, EAX
-//    LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined]
+//    t4 = phi(t1, t3 / loop)
+//    t2 = OP MI.val, t4
+//    EAX = t4
+//    LCMPXCHG [MI.addr], t2, [EAX is implicitly used & defined]
+//    t3 = EAX
 //    JNE loop
 // sink:
-//    dst = EAX
+//    dst = t3
 //    ...
 MachineBasicBlock *
 X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
@@ -12400,7 +13173,7 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
   MachineFunction::iterator I = MBB;
   ++I;
 
-  assert(MI->getNumOperands() <= X86::AddrNumOperands + 2 &&
+  assert(MI->getNumOperands() <= X86::AddrNumOperands + 4 &&
          "Unexpected number of operands");
 
   assert(MI->hasOneMemOperand() &&
@@ -12422,7 +13195,11 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
 
   const TargetRegisterClass *RC = MRI.getRegClass(DstReg);
   MVT::SimpleValueType VT = *RC->vt_begin();
-  unsigned AccPhyReg = getX86SubSuperRegister(X86::EAX, VT);
+  unsigned t1 = MRI.createVirtualRegister(RC);
+  unsigned t2 = MRI.createVirtualRegister(RC);
+  unsigned t3 = MRI.createVirtualRegister(RC);
+  unsigned t4 = MRI.createVirtualRegister(RC);
+  unsigned PhyReg = getX86SubSuperRegister(X86::EAX, VT);
 
   unsigned LCMPXCHGOpc = getCmpXChgOpcode(VT);
   unsigned LOADOpc = getLoadOpcode(VT);
@@ -12430,12 +13207,16 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
   // For the atomic load-arith operator, we generate
   //
   //  thisMBB:
-  //    EAX = LOAD [MI.addr]
+  //    t1 = LOAD [MI.addr]
   //  mainMBB:
+  //    t4 = phi(t1 / thisMBB, t3 / mainMBB)
   //    t1 = OP MI.val, EAX
+  //    EAX = t4
   //    LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined]
+  //    t3 = EAX
   //    JNE mainMBB
   //  sinkMBB:
+  //    dst = t3
 
   MachineBasicBlock *thisMBB = MBB;
   MachineBasicBlock *mainMBB = MF->CreateMachineBasicBlock(BB);
@@ -12451,23 +13232,34 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
   sinkMBB->transferSuccessorsAndUpdatePHIs(MBB);
 
   // thisMBB:
-  MIB = BuildMI(thisMBB, DL, TII->get(LOADOpc), AccPhyReg);
-  for (unsigned i = 0; i < X86::AddrNumOperands; ++i)
-    MIB.addOperand(MI->getOperand(MemOpndSlot + i));
-  MIB.setMemRefs(MMOBegin, MMOEnd);
+  MIB = BuildMI(thisMBB, DL, TII->get(LOADOpc), t1);
+  for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
+    MachineOperand NewMO = MI->getOperand(MemOpndSlot + i);
+    if (NewMO.isReg())
+      NewMO.setIsKill(false);
+    MIB.addOperand(NewMO);
+  }
+  for (MachineInstr::mmo_iterator MMOI = MMOBegin; MMOI != MMOEnd; ++MMOI) {
+    unsigned flags = (*MMOI)->getFlags();
+    flags = (flags & ~MachineMemOperand::MOStore) | MachineMemOperand::MOLoad;
+    MachineMemOperand *MMO =
+      MF->getMachineMemOperand((*MMOI)->getPointerInfo(), flags,
+                               (*MMOI)->getSize(),
+                               (*MMOI)->getBaseAlignment(),
+                               (*MMOI)->getTBAAInfo(),
+                               (*MMOI)->getRanges());
+    MIB.addMemOperand(MMO);
+  }
 
   thisMBB->addSuccessor(mainMBB);
 
   // mainMBB:
   MachineBasicBlock *origMainMBB = mainMBB;
-  mainMBB->addLiveIn(AccPhyReg);
 
-  // Copy AccPhyReg as it is used more than once.
-  unsigned AccReg = MRI.createVirtualRegister(RC);
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), AccReg)
-    .addReg(AccPhyReg);
+  // Add a PHI.
+  MachineInstr *Phi = BuildMI(mainMBB, DL, TII->get(X86::PHI), t4)
+                        .addReg(t1).addMBB(thisMBB).addReg(t3).addMBB(mainMBB);
 
-  unsigned t1 = MRI.createVirtualRegister(RC);
   unsigned Opc = MI->getOpcode();
   switch (Opc) {
   default:
@@ -12485,20 +13277,20 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
   case X86::ATOMXOR32:
   case X86::ATOMXOR64: {
     unsigned ARITHOpc = getNonAtomicOpcode(Opc);
-    BuildMI(mainMBB, DL, TII->get(ARITHOpc), t1).addReg(SrcReg)
-      .addReg(AccReg);
+    BuildMI(mainMBB, DL, TII->get(ARITHOpc), t2).addReg(SrcReg)
+      .addReg(t4);
     break;
   }
   case X86::ATOMNAND8:
   case X86::ATOMNAND16:
   case X86::ATOMNAND32:
   case X86::ATOMNAND64: {
-    unsigned t2 = MRI.createVirtualRegister(RC);
+    unsigned Tmp = MRI.createVirtualRegister(RC);
     unsigned NOTOpc;
     unsigned ANDOpc = getNonAtomicOpcodeWithExtraOpc(Opc, NOTOpc);
-    BuildMI(mainMBB, DL, TII->get(ANDOpc), t2).addReg(SrcReg)
-      .addReg(AccReg);
-    BuildMI(mainMBB, DL, TII->get(NOTOpc), t1).addReg(t2);
+    BuildMI(mainMBB, DL, TII->get(ANDOpc), Tmp).addReg(SrcReg)
+      .addReg(t4);
+    BuildMI(mainMBB, DL, TII->get(NOTOpc), t2).addReg(Tmp);
     break;
   }
   case X86::ATOMMAX8:
@@ -12522,20 +13314,22 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
 
     BuildMI(mainMBB, DL, TII->get(CMPOpc))
       .addReg(SrcReg)
-      .addReg(AccReg);
+      .addReg(t4);
 
     if (Subtarget->hasCMov()) {
       if (VT != MVT::i8) {
         // Native support
-        BuildMI(mainMBB, DL, TII->get(CMOVOpc), t1)
+        BuildMI(mainMBB, DL, TII->get(CMOVOpc), t2)
           .addReg(SrcReg)
-          .addReg(AccReg);
+          .addReg(t4);
       } else {
         // Promote i8 to i32 to use CMOV32
-        const TargetRegisterClass *RC32 = getRegClassFor(MVT::i32);
+        const TargetRegisterInfo* TRI = getTargetMachine().getRegisterInfo();
+        const TargetRegisterClass *RC32 =
+          TRI->getSubClassWithSubReg(getRegClassFor(MVT::i32), X86::sub_8bit);
         unsigned SrcReg32 = MRI.createVirtualRegister(RC32);
         unsigned AccReg32 = MRI.createVirtualRegister(RC32);
-        unsigned t2 = MRI.createVirtualRegister(RC32);
+        unsigned Tmp = MRI.createVirtualRegister(RC32);
 
         unsigned Undef = MRI.createVirtualRegister(RC32);
         BuildMI(mainMBB, DL, TII->get(TargetOpcode::IMPLICIT_DEF), Undef);
@@ -12546,15 +13340,15 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
           .addImm(X86::sub_8bit);
         BuildMI(mainMBB, DL, TII->get(TargetOpcode::INSERT_SUBREG), AccReg32)
           .addReg(Undef)
-          .addReg(AccReg)
+          .addReg(t4)
           .addImm(X86::sub_8bit);
 
-        BuildMI(mainMBB, DL, TII->get(CMOVOpc), t2)
+        BuildMI(mainMBB, DL, TII->get(CMOVOpc), Tmp)
           .addReg(SrcReg32)
           .addReg(AccReg32);
 
-        BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), t1)
-          .addReg(t2, 0, X86::sub_8bit);
+        BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), t2)
+          .addReg(Tmp, 0, X86::sub_8bit);
       }
     } else {
       // Use pseudo select and lower them.
@@ -12563,36 +13357,47 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
       unsigned SelOpc = getPseudoCMOVOpc(VT);
       X86::CondCode CC = X86::getCondFromCMovOpc(CMOVOpc);
       assert(CC != X86::COND_INVALID && "Invalid atomic-load-op transformation!");
-      MIB = BuildMI(mainMBB, DL, TII->get(SelOpc), t1)
-              .addReg(SrcReg).addReg(AccReg)
+      MIB = BuildMI(mainMBB, DL, TII->get(SelOpc), t2)
+              .addReg(SrcReg).addReg(t4)
               .addImm(CC);
       mainMBB = EmitLoweredSelect(MIB, mainMBB);
+      // Replace the original PHI node as mainMBB is changed after CMOV
+      // lowering.
+      BuildMI(*origMainMBB, Phi, DL, TII->get(X86::PHI), t4)
+        .addReg(t1).addMBB(thisMBB).addReg(t3).addMBB(mainMBB);
+      Phi->eraseFromParent();
     }
     break;
   }
   }
 
-  // Copy AccPhyReg back from virtual register.
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), AccPhyReg)
-    .addReg(AccReg);
+  // Copy PhyReg back from virtual register.
+  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), PhyReg)
+    .addReg(t4);
 
   MIB = BuildMI(mainMBB, DL, TII->get(LCMPXCHGOpc));
-  for (unsigned i = 0; i < X86::AddrNumOperands; ++i)
-    MIB.addOperand(MI->getOperand(MemOpndSlot + i));
-  MIB.addReg(t1);
+  for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
+    MachineOperand NewMO = MI->getOperand(MemOpndSlot + i);
+    if (NewMO.isReg())
+      NewMO.setIsKill(false);
+    MIB.addOperand(NewMO);
+  }
+  MIB.addReg(t2);
   MIB.setMemRefs(MMOBegin, MMOEnd);
 
+  // Copy PhyReg back to virtual register.
+  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), t3)
+    .addReg(PhyReg);
+
   BuildMI(mainMBB, DL, TII->get(X86::JNE_4)).addMBB(origMainMBB);
 
   mainMBB->addSuccessor(origMainMBB);
   mainMBB->addSuccessor(sinkMBB);
 
   // sinkMBB:
-  sinkMBB->addLiveIn(AccPhyReg);
-
   BuildMI(*sinkMBB, sinkMBB->begin(), DL,
           TII->get(TargetOpcode::COPY), DstReg)
-    .addReg(AccPhyReg);
+    .addReg(t3);
 
   MI->eraseFromParent();
   return sinkMBB;
@@ -12609,15 +13414,24 @@ X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
 // to
 //
 //    ...
-//    EAX = LOAD [MI.addr + 0]
-//    EDX = LOAD [MI.addr + 4]
+//    t1L = LOAD [MI.addr + 0]
+//    t1H = LOAD [MI.addr + 4]
 // loop:
-//    EBX = OP MI.val.lo, EAX
-//    ECX = OP MI.val.hi, EDX
+//    t4L = phi(t1L, t3L / loop)
+//    t4H = phi(t1H, t3H / loop)
+//    t2L = OP MI.val.lo, t4L
+//    t2H = OP MI.val.hi, t4H
+//    EAX = t4L
+//    EDX = t4H
+//    EBX = t2L
+//    ECX = t2H
 //    LCMPXCHG8B [MI.addr], [ECX:EBX & EDX:EAX are implicitly used and EDX:EAX is implicitly defined]
+//    t3L = EAX
+//    t3H = EDX
 //    JNE loop
 // sink:
-//    dst = EDX:EAX
+//    dstL = t3L
+//    dstH = t3H
 //    ...
 MachineBasicBlock *
 X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
@@ -12632,7 +13446,7 @@ X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
   MachineFunction::iterator I = MBB;
   ++I;
 
-  assert(MI->getNumOperands() <= X86::AddrNumOperands + 4 &&
+  assert(MI->getNumOperands() <= X86::AddrNumOperands + 7 &&
          "Unexpected number of operands");
 
   assert(MI->hasOneMemOperand() &&
@@ -12658,20 +13472,37 @@ X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
   const TargetRegisterClass *RC = &X86::GR32RegClass;
   const TargetRegisterClass *RC8 = &X86::GR8RegClass;
 
+  unsigned t1L = MRI.createVirtualRegister(RC);
+  unsigned t1H = MRI.createVirtualRegister(RC);
+  unsigned t2L = MRI.createVirtualRegister(RC);
+  unsigned t2H = MRI.createVirtualRegister(RC);
+  unsigned t3L = MRI.createVirtualRegister(RC);
+  unsigned t3H = MRI.createVirtualRegister(RC);
+  unsigned t4L = MRI.createVirtualRegister(RC);
+  unsigned t4H = MRI.createVirtualRegister(RC);
+
   unsigned LCMPXCHGOpc = X86::LCMPXCHG8B;
   unsigned LOADOpc = X86::MOV32rm;
 
   // For the atomic load-arith operator, we generate
   //
   //  thisMBB:
-  //    EAX = LOAD [MI.addr + 0]
-  //    EDX = LOAD [MI.addr + 4]
+  //    t1L = LOAD [MI.addr + 0]
+  //    t1H = LOAD [MI.addr + 4]
   //  mainMBB:
-  //    EBX = OP MI.vallo, EAX
-  //    ECX = OP MI.valhi, EDX
+  //    t4L = phi(t1L / thisMBB, t3L / mainMBB)
+  //    t4H = phi(t1H / thisMBB, t3H / mainMBB)
+  //    t2L = OP MI.val.lo, t4L
+  //    t2H = OP MI.val.hi, t4H
+  //    EBX = t2L
+  //    ECX = t2H
   //    LCMPXCHG8B [MI.addr], [ECX:EBX & EDX:EAX are implicitly used and EDX:EAX is implicitly defined]
-  //    JNE mainMBB
+  //    t3L = EAX
+  //    t3H = EDX
+  //    JNE loop
   //  sinkMBB:
+  //    dstL = t3L
+  //    dstH = t3H
 
   MachineBasicBlock *thisMBB = MBB;
   MachineBasicBlock *mainMBB = MF->CreateMachineBasicBlock(BB);
@@ -12688,35 +13519,50 @@ X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
 
   // thisMBB:
   // Lo
-  MIB = BuildMI(thisMBB, DL, TII->get(LOADOpc), X86::EAX);
-  for (unsigned i = 0; i < X86::AddrNumOperands; ++i)
-    MIB.addOperand(MI->getOperand(MemOpndSlot + i));
-  MIB.setMemRefs(MMOBegin, MMOEnd);
+  MIB = BuildMI(thisMBB, DL, TII->get(LOADOpc), t1L);
+  for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
+    MachineOperand NewMO = MI->getOperand(MemOpndSlot + i);
+    if (NewMO.isReg())
+      NewMO.setIsKill(false);
+    MIB.addOperand(NewMO);
+  }
+  for (MachineInstr::mmo_iterator MMOI = MMOBegin; MMOI != MMOEnd; ++MMOI) {
+    unsigned flags = (*MMOI)->getFlags();
+    flags = (flags & ~MachineMemOperand::MOStore) | MachineMemOperand::MOLoad;
+    MachineMemOperand *MMO =
+      MF->getMachineMemOperand((*MMOI)->getPointerInfo(), flags,
+                               (*MMOI)->getSize(),
+                               (*MMOI)->getBaseAlignment(),
+                               (*MMOI)->getTBAAInfo(),
+                               (*MMOI)->getRanges());
+    MIB.addMemOperand(MMO);
+  };
+  MachineInstr *LowMI = MIB;
+
   // Hi
-  MIB = BuildMI(thisMBB, DL, TII->get(LOADOpc), X86::EDX);
+  MIB = BuildMI(thisMBB, DL, TII->get(LOADOpc), t1H);
   for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
-    if (i == X86::AddrDisp)
+    if (i == X86::AddrDisp) {
       MIB.addDisp(MI->getOperand(MemOpndSlot + i), 4); // 4 == sizeof(i32)
-    else
-      MIB.addOperand(MI->getOperand(MemOpndSlot + i));
+    } else {
+      MachineOperand NewMO = MI->getOperand(MemOpndSlot + i);
+      if (NewMO.isReg())
+        NewMO.setIsKill(false);
+      MIB.addOperand(NewMO);
+    }
   }
-  MIB.setMemRefs(MMOBegin, MMOEnd);
+  MIB.setMemRefs(LowMI->memoperands_begin(), LowMI->memoperands_end());
 
   thisMBB->addSuccessor(mainMBB);
 
   // mainMBB:
   MachineBasicBlock *origMainMBB = mainMBB;
-  mainMBB->addLiveIn(X86::EAX);
-  mainMBB->addLiveIn(X86::EDX);
-
-  // Copy EDX:EAX as they are used more than once.
-  unsigned LoReg = MRI.createVirtualRegister(RC);
-  unsigned HiReg = MRI.createVirtualRegister(RC);
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), LoReg).addReg(X86::EAX);
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), HiReg).addReg(X86::EDX);
 
-  unsigned t1L = MRI.createVirtualRegister(RC);
-  unsigned t1H = MRI.createVirtualRegister(RC);
+  // Add PHIs.
+  MachineInstr *PhiL = BuildMI(mainMBB, DL, TII->get(X86::PHI), t4L)
+                        .addReg(t1L).addMBB(thisMBB).addReg(t3L).addMBB(mainMBB);
+  MachineInstr *PhiH = BuildMI(mainMBB, DL, TII->get(X86::PHI), t4H)
+                        .addReg(t1H).addMBB(thisMBB).addReg(t3H).addMBB(mainMBB);
 
   unsigned Opc = MI->getOpcode();
   switch (Opc) {
@@ -12729,19 +13575,23 @@ X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
   case X86::ATOMSUB6432: {
     unsigned HiOpc;
     unsigned LoOpc = getNonAtomic6432Opcode(Opc, HiOpc);
-    BuildMI(mainMBB, DL, TII->get(LoOpc), t1L).addReg(LoReg).addReg(SrcLoReg);
-    BuildMI(mainMBB, DL, TII->get(HiOpc), t1H).addReg(HiReg).addReg(SrcHiReg);
+    BuildMI(mainMBB, DL, TII->get(LoOpc), t2L).addReg(t4L)
+      .addReg(SrcLoReg);
+    BuildMI(mainMBB, DL, TII->get(HiOpc), t2H).addReg(t4H)
+      .addReg(SrcHiReg);
     break;
   }
   case X86::ATOMNAND6432: {
     unsigned HiOpc, NOTOpc;
     unsigned LoOpc = getNonAtomic6432OpcodeWithExtraOpc(Opc, HiOpc, NOTOpc);
-    unsigned t2L = MRI.createVirtualRegister(RC);
-    unsigned t2H = MRI.createVirtualRegister(RC);
-    BuildMI(mainMBB, DL, TII->get(LoOpc), t2L).addReg(SrcLoReg).addReg(LoReg);
-    BuildMI(mainMBB, DL, TII->get(HiOpc), t2H).addReg(SrcHiReg).addReg(HiReg);
-    BuildMI(mainMBB, DL, TII->get(NOTOpc), t1L).addReg(t2L);
-    BuildMI(mainMBB, DL, TII->get(NOTOpc), t1H).addReg(t2H);
+    unsigned TmpL = MRI.createVirtualRegister(RC);
+    unsigned TmpH = MRI.createVirtualRegister(RC);
+    BuildMI(mainMBB, DL, TII->get(LoOpc), TmpL).addReg(SrcLoReg)
+      .addReg(t4L);
+    BuildMI(mainMBB, DL, TII->get(HiOpc), TmpH).addReg(SrcHiReg)
+      .addReg(t4H);
+    BuildMI(mainMBB, DL, TII->get(NOTOpc), t2L).addReg(TmpL);
+    BuildMI(mainMBB, DL, TII->get(NOTOpc), t2H).addReg(TmpH);
     break;
   }
   case X86::ATOMMAX6432:
@@ -12757,12 +13607,12 @@ X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
     unsigned cc = MRI.createVirtualRegister(RC);
     // cl := cmp src_lo, lo
     BuildMI(mainMBB, DL, TII->get(X86::CMP32rr))
-      .addReg(SrcLoReg).addReg(LoReg);
+      .addReg(SrcLoReg).addReg(t4L);
     BuildMI(mainMBB, DL, TII->get(LoOpc), cL);
     BuildMI(mainMBB, DL, TII->get(X86::MOVZX32rr8), cL32).addReg(cL);
     // ch := cmp src_hi, hi
     BuildMI(mainMBB, DL, TII->get(X86::CMP32rr))
-      .addReg(SrcHiReg).addReg(HiReg);
+      .addReg(SrcHiReg).addReg(t4H);
     BuildMI(mainMBB, DL, TII->get(HiOpc), cH);
     BuildMI(mainMBB, DL, TII->get(X86::MOVZX32rr8), cH32).addReg(cH);
     // cc := if (src_hi == hi) ? cl : ch;
@@ -12777,58 +13627,74 @@ X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
     }
     BuildMI(mainMBB, DL, TII->get(X86::TEST32rr)).addReg(cc).addReg(cc);
     if (Subtarget->hasCMov()) {
-      BuildMI(mainMBB, DL, TII->get(X86::CMOVNE32rr), t1L)
-        .addReg(SrcLoReg).addReg(LoReg);
-      BuildMI(mainMBB, DL, TII->get(X86::CMOVNE32rr), t1H)
-        .addReg(SrcHiReg).addReg(HiReg);
+      BuildMI(mainMBB, DL, TII->get(X86::CMOVNE32rr), t2L)
+        .addReg(SrcLoReg).addReg(t4L);
+      BuildMI(mainMBB, DL, TII->get(X86::CMOVNE32rr), t2H)
+        .addReg(SrcHiReg).addReg(t4H);
     } else {
-      MIB = BuildMI(mainMBB, DL, TII->get(X86::CMOV_GR32), t1L)
-              .addReg(SrcLoReg).addReg(LoReg)
+      MIB = BuildMI(mainMBB, DL, TII->get(X86::CMOV_GR32), t2L)
+              .addReg(SrcLoReg).addReg(t4L)
               .addImm(X86::COND_NE);
       mainMBB = EmitLoweredSelect(MIB, mainMBB);
-      MIB = BuildMI(mainMBB, DL, TII->get(X86::CMOV_GR32), t1H)
-              .addReg(SrcHiReg).addReg(HiReg)
+      // As the lowered CMOV won't clobber EFLAGS, we could reuse it for the
+      // 2nd CMOV lowering.
+      mainMBB->addLiveIn(X86::EFLAGS);
+      MIB = BuildMI(mainMBB, DL, TII->get(X86::CMOV_GR32), t2H)
+              .addReg(SrcHiReg).addReg(t4H)
               .addImm(X86::COND_NE);
       mainMBB = EmitLoweredSelect(MIB, mainMBB);
+      // Replace the original PHI node as mainMBB is changed after CMOV
+      // lowering.
+      BuildMI(*origMainMBB, PhiL, DL, TII->get(X86::PHI), t4L)
+        .addReg(t1L).addMBB(thisMBB).addReg(t3L).addMBB(mainMBB);
+      BuildMI(*origMainMBB, PhiH, DL, TII->get(X86::PHI), t4H)
+        .addReg(t1H).addMBB(thisMBB).addReg(t3H).addMBB(mainMBB);
+      PhiL->eraseFromParent();
+      PhiH->eraseFromParent();
     }
     break;
   }
   case X86::ATOMSWAP6432: {
     unsigned HiOpc;
     unsigned LoOpc = getNonAtomic6432Opcode(Opc, HiOpc);
-    BuildMI(mainMBB, DL, TII->get(LoOpc), t1L).addReg(SrcLoReg);
-    BuildMI(mainMBB, DL, TII->get(HiOpc), t1H).addReg(SrcHiReg);
+    BuildMI(mainMBB, DL, TII->get(LoOpc), t2L).addReg(SrcLoReg);
+    BuildMI(mainMBB, DL, TII->get(HiOpc), t2H).addReg(SrcHiReg);
     break;
   }
   }
 
   // Copy EDX:EAX back from HiReg:LoReg
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::EAX).addReg(LoReg);
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::EDX).addReg(HiReg);
+  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::EAX).addReg(t4L);
+  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::EDX).addReg(t4H);
   // Copy ECX:EBX from t1H:t1L
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::EBX).addReg(t1L);
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::ECX).addReg(t1H);
+  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::EBX).addReg(t2L);
+  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::ECX).addReg(t2H);
 
   MIB = BuildMI(mainMBB, DL, TII->get(LCMPXCHGOpc));
-  for (unsigned i = 0; i < X86::AddrNumOperands; ++i)
-    MIB.addOperand(MI->getOperand(MemOpndSlot + i));
+  for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
+    MachineOperand NewMO = MI->getOperand(MemOpndSlot + i);
+    if (NewMO.isReg())
+      NewMO.setIsKill(false);
+    MIB.addOperand(NewMO);
+  }
   MIB.setMemRefs(MMOBegin, MMOEnd);
 
+  // Copy EDX:EAX back to t3H:t3L
+  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), t3L).addReg(X86::EAX);
+  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), t3H).addReg(X86::EDX);
+
   BuildMI(mainMBB, DL, TII->get(X86::JNE_4)).addMBB(origMainMBB);
 
   mainMBB->addSuccessor(origMainMBB);
   mainMBB->addSuccessor(sinkMBB);
 
   // sinkMBB:
-  sinkMBB->addLiveIn(X86::EAX);
-  sinkMBB->addLiveIn(X86::EDX);
-
   BuildMI(*sinkMBB, sinkMBB->begin(), DL,
           TII->get(TargetOpcode::COPY), DstLoReg)
-    .addReg(X86::EAX);
+    .addReg(t3L);
   BuildMI(*sinkMBB, sinkMBB->begin(), DL,
           TII->get(TargetOpcode::COPY), DstHiReg)
-    .addReg(X86::EDX);
+    .addReg(t3H);
 
   MI->eraseFromParent();
   return sinkMBB;
@@ -13239,7 +14105,7 @@ X86TargetLowering::EmitVAStartSaveXMMRegsWithCustomInserter(
     MBB->addSuccessor(EndMBB);
   }
 
-  unsigned MOVOpc = Subtarget->hasAVX() ? X86::VMOVAPSmr : X86::MOVAPSmr;
+  unsigned MOVOpc = Subtarget->hasFp256() ? X86::VMOVAPSmr : X86::MOVAPSmr;
   // In the XMM save block, save all the XMM argument registers.
   for (int i = 3, e = MI->getNumOperands(); i != e; ++i) {
     int64_t Offset = (i - 3) * 16 + VarArgsFPOffset;
@@ -14203,6 +15069,18 @@ static SDValue PerformShuffleCombine256(SDNode *N, SelectionDAG &DAG,
                                   Ld->getAlignment(),
                                   false/*isVolatile*/, true/*ReadMem*/,
                                   false/*WriteMem*/);
+
+        // Make sure the newly-created LOAD is in the same position as Ld in
+        // terms of dependency. We create a TokenFactor for Ld and ResNode,
+        // and update uses of Ld's output chain to use the TokenFactor.
+        if (Ld->hasAnyUseOfValue(1)) {
+          SDValue NewChain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
+                             SDValue(Ld, 1), SDValue(ResNode.getNode(), 1));
+          DAG.ReplaceAllUsesOfValueWith(SDValue(Ld, 1), NewChain);
+          DAG.UpdateNodeOperands(NewChain.getNode(), SDValue(Ld, 1),
+                                 SDValue(ResNode.getNode(), 1));
+        }
+
         return DAG.getNode(ISD::BITCAST, dl, VT, ResNode);
       }
     }
@@ -14248,7 +15126,7 @@ static SDValue PerformShuffleCombine(SDNode *N, SelectionDAG &DAG,
     return SDValue();
 
   // Combine 256-bit vector shuffles. This is only profitable when in AVX mode
-  if (Subtarget->hasAVX() && VT.is256BitVector() &&
+  if (Subtarget->hasFp256() && VT.is256BitVector() &&
       N->getOpcode() == ISD::VECTOR_SHUFFLE)
     return PerformShuffleCombine256(N, DAG, DCI, Subtarget);
 
@@ -14266,127 +15144,12 @@ static SDValue PerformShuffleCombine(SDNode *N, SelectionDAG &DAG,
   return EltsFromConsecutiveLoads(VT, Elts, dl, DAG);
 }
 
-
 /// PerformTruncateCombine - Converts truncate operation to
 /// a sequence of vector shuffle operations.
 /// It is possible when we truncate 256-bit vector to 128-bit vector
 static SDValue PerformTruncateCombine(SDNode *N, SelectionDAG &DAG,
                                       TargetLowering::DAGCombinerInfo &DCI,
                                       const X86Subtarget *Subtarget)  {
-  if (!DCI.isBeforeLegalizeOps())
-    return SDValue();
-
-  if (!Subtarget->hasAVX())
-    return SDValue();
-
-  EVT VT = N->getValueType(0);
-  SDValue Op = N->getOperand(0);
-  EVT OpVT = Op.getValueType();
-  DebugLoc dl = N->getDebugLoc();
-
-  if ((VT == MVT::v4i32) && (OpVT == MVT::v4i64)) {
-
-    if (Subtarget->hasAVX2()) {
-      // AVX2: v4i64 -> v4i32
-
-      // VPERMD
-      static const int ShufMask[] = {0, 2, 4, 6, -1, -1, -1, -1};
-
-      Op = DAG.getNode(ISD::BITCAST, dl, MVT::v8i32, Op);
-      Op = DAG.getVectorShuffle(MVT::v8i32, dl, Op, DAG.getUNDEF(MVT::v8i32),
-                                ShufMask);
-
-      return DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, Op,
-                         DAG.getIntPtrConstant(0));
-    }
-
-    // AVX: v4i64 -> v4i32
-    SDValue OpLo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v2i64, Op,
-                               DAG.getIntPtrConstant(0));
-
-    SDValue OpHi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v2i64, Op,
-                               DAG.getIntPtrConstant(2));
-
-    OpLo = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, OpLo);
-    OpHi = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, OpHi);
-
-    // PSHUFD
-    static const int ShufMask1[] = {0, 2, 0, 0};
-
-    SDValue Undef = DAG.getUNDEF(VT);
-    OpLo = DAG.getVectorShuffle(VT, dl, OpLo, Undef, ShufMask1);
-    OpHi = DAG.getVectorShuffle(VT, dl, OpHi, Undef, ShufMask1);
-
-    // MOVLHPS
-    static const int ShufMask2[] = {0, 1, 4, 5};
-
-    return DAG.getVectorShuffle(VT, dl, OpLo, OpHi, ShufMask2);
-  }
-
-  if ((VT == MVT::v8i16) && (OpVT == MVT::v8i32)) {
-
-    if (Subtarget->hasAVX2()) {
-      // AVX2: v8i32 -> v8i16
-
-      Op = DAG.getNode(ISD::BITCAST, dl, MVT::v32i8, Op);
-
-      // PSHUFB
-      SmallVector<SDValue,32> pshufbMask;
-      for (unsigned i = 0; i < 2; ++i) {
-        pshufbMask.push_back(DAG.getConstant(0x0, MVT::i8));
-        pshufbMask.push_back(DAG.getConstant(0x1, MVT::i8));
-        pshufbMask.push_back(DAG.getConstant(0x4, MVT::i8));
-        pshufbMask.push_back(DAG.getConstant(0x5, MVT::i8));
-        pshufbMask.push_back(DAG.getConstant(0x8, MVT::i8));
-        pshufbMask.push_back(DAG.getConstant(0x9, MVT::i8));
-        pshufbMask.push_back(DAG.getConstant(0xc, MVT::i8));
-        pshufbMask.push_back(DAG.getConstant(0xd, MVT::i8));
-        for (unsigned j = 0; j < 8; ++j)
-          pshufbMask.push_back(DAG.getConstant(0x80, MVT::i8));
-      }
-      SDValue BV = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v32i8,
-                               &pshufbMask[0], 32);
-      Op = DAG.getNode(X86ISD::PSHUFB, dl, MVT::v32i8, Op, BV);
-
-      Op = DAG.getNode(ISD::BITCAST, dl, MVT::v4i64, Op);
-
-      static const int ShufMask[] = {0,  2,  -1,  -1};
-      Op = DAG.getVectorShuffle(MVT::v4i64, dl,  Op, DAG.getUNDEF(MVT::v4i64),
-                                &ShufMask[0]);
-
-      Op = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v2i64, Op,
-                       DAG.getIntPtrConstant(0));
-
-      return DAG.getNode(ISD::BITCAST, dl, VT, Op);
-    }
-
-    SDValue OpLo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i32, Op,
-                               DAG.getIntPtrConstant(0));
-
-    SDValue OpHi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i32, Op,
-                               DAG.getIntPtrConstant(4));
-
-    OpLo = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, OpLo);
-    OpHi = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, OpHi);
-
-    // PSHUFB
-    static const int ShufMask1[] = {0,  1,  4,  5,  8,  9, 12, 13,
-                                   -1, -1, -1, -1, -1, -1, -1, -1};
-
-    SDValue Undef = DAG.getUNDEF(MVT::v16i8);
-    OpLo = DAG.getVectorShuffle(MVT::v16i8, dl, OpLo, Undef, ShufMask1);
-    OpHi = DAG.getVectorShuffle(MVT::v16i8, dl, OpHi, Undef, ShufMask1);
-
-    OpLo = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, OpLo);
-    OpHi = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, OpHi);
-
-    // MOVLHPS
-    static const int ShufMask2[] = {0, 1, 4, 5};
-
-    SDValue res = DAG.getVectorShuffle(MVT::v4i32, dl, OpLo, OpHi, ShufMask2);
-    return DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, res);
-  }
-
   return SDValue();
 }
 
@@ -14581,6 +15344,76 @@ static SDValue PerformEXTRACT_VECTOR_ELTCombine(SDNode *N, SelectionDAG &DAG,
   return SDValue();
 }
 
+/// \brief Matches a VSELECT onto min/max or return 0 if the node doesn't match.
+static unsigned matchIntegerMINMAX(SDValue Cond, EVT VT, SDValue LHS,
+                                   SDValue RHS, SelectionDAG &DAG,
+                                   const X86Subtarget *Subtarget) {
+  if (!VT.isVector())
+    return 0;
+
+  switch (VT.getSimpleVT().SimpleTy) {
+  default: return 0;
+  case MVT::v32i8:
+  case MVT::v16i16:
+  case MVT::v8i32:
+    if (!Subtarget->hasAVX2())
+      return 0;
+  case MVT::v16i8:
+  case MVT::v8i16:
+  case MVT::v4i32:
+    if (!Subtarget->hasSSE2())
+      return 0;
+  }
+
+  // SSE2 has only a small subset of the operations.
+  bool hasUnsigned = Subtarget->hasSSE41() ||
+                     (Subtarget->hasSSE2() && VT == MVT::v16i8);
+  bool hasSigned = Subtarget->hasSSE41() ||
+                   (Subtarget->hasSSE2() && VT == MVT::v8i16);
+
+  ISD::CondCode CC = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
+
+  // Check for x CC y ? x : y.
+  if (DAG.isEqualTo(LHS, Cond.getOperand(0)) &&
+      DAG.isEqualTo(RHS, Cond.getOperand(1))) {
+    switch (CC) {
+    default: break;
+    case ISD::SETULT:
+    case ISD::SETULE:
+      return hasUnsigned ? X86ISD::UMIN : 0;
+    case ISD::SETUGT:
+    case ISD::SETUGE:
+      return hasUnsigned ? X86ISD::UMAX : 0;
+    case ISD::SETLT:
+    case ISD::SETLE:
+      return hasSigned ? X86ISD::SMIN : 0;
+    case ISD::SETGT:
+    case ISD::SETGE:
+      return hasSigned ? X86ISD::SMAX : 0;
+    }
+  // Check for x CC y ? y : x -- a min/max with reversed arms.
+  } else if (DAG.isEqualTo(LHS, Cond.getOperand(1)) &&
+             DAG.isEqualTo(RHS, Cond.getOperand(0))) {
+    switch (CC) {
+    default: break;
+    case ISD::SETULT:
+    case ISD::SETULE:
+      return hasUnsigned ? X86ISD::UMAX : 0;
+    case ISD::SETUGT:
+    case ISD::SETUGE:
+      return hasUnsigned ? X86ISD::UMIN : 0;
+    case ISD::SETLT:
+    case ISD::SETLE:
+      return hasSigned ? X86ISD::SMAX : 0;
+    case ISD::SETGT:
+    case ISD::SETGE:
+      return hasSigned ? X86ISD::SMIN : 0;
+    }
+  }
+
+  return 0;
+}
+
 /// PerformSELECTCombine - Do target-specific dag combines on SELECT and VSELECT
 /// nodes.
 static SDValue PerformSELECTCombine(SDNode *N, SelectionDAG &DAG,
@@ -14861,6 +15694,67 @@ static SDValue PerformSELECTCombine(SDNode *N, SelectionDAG &DAG,
     }
   }
 
+  // Match VSELECTs into subs with unsigned saturation.
+  if (!DCI.isBeforeLegalize() &&
+      N->getOpcode() == ISD::VSELECT && Cond.getOpcode() == ISD::SETCC &&
+      // psubus is available in SSE2 and AVX2 for i8 and i16 vectors.
+      ((Subtarget->hasSSE2() && (VT == MVT::v16i8 || VT == MVT::v8i16)) ||
+       (Subtarget->hasAVX2() && (VT == MVT::v32i8 || VT == MVT::v16i16)))) {
+    ISD::CondCode CC = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
+
+    // Check if one of the arms of the VSELECT is a zero vector. If it's on the
+    // left side invert the predicate to simplify logic below.
+    SDValue Other;
+    if (ISD::isBuildVectorAllZeros(LHS.getNode())) {
+      Other = RHS;
+      CC = ISD::getSetCCInverse(CC, true);
+    } else if (ISD::isBuildVectorAllZeros(RHS.getNode())) {
+      Other = LHS;
+    }
+
+    if (Other.getNode() && Other->getNumOperands() == 2 &&
+        DAG.isEqualTo(Other->getOperand(0), Cond.getOperand(0))) {
+      SDValue OpLHS = Other->getOperand(0), OpRHS = Other->getOperand(1);
+      SDValue CondRHS = Cond->getOperand(1);
+
+      // Look for a general sub with unsigned saturation first.
+      // x >= y ? x-y : 0 --> subus x, y
+      // x >  y ? x-y : 0 --> subus x, y
+      if ((CC == ISD::SETUGE || CC == ISD::SETUGT) &&
+          Other->getOpcode() == ISD::SUB && DAG.isEqualTo(OpRHS, CondRHS))
+        return DAG.getNode(X86ISD::SUBUS, DL, VT, OpLHS, OpRHS);
+
+      // If the RHS is a constant we have to reverse the const canonicalization.
+      // x > C-1 ? x+-C : 0 --> subus x, C
+      if (CC == ISD::SETUGT && Other->getOpcode() == ISD::ADD &&
+          isSplatVector(CondRHS.getNode()) && isSplatVector(OpRHS.getNode())) {
+        APInt A = cast<ConstantSDNode>(OpRHS.getOperand(0))->getAPIntValue();
+        if (CondRHS.getConstantOperandVal(0) == -A-1)
+          return DAG.getNode(X86ISD::SUBUS, DL, VT, OpLHS,
+                             DAG.getConstant(-A, VT));
+      }
+
+      // Another special case: If C was a sign bit, the sub has been
+      // canonicalized into a xor.
+      // FIXME: Would it be better to use ComputeMaskedBits to determine whether
+      //        it's safe to decanonicalize the xor?
+      // x s< 0 ? x^C : 0 --> subus x, C
+      if (CC == ISD::SETLT && Other->getOpcode() == ISD::XOR &&
+          ISD::isBuildVectorAllZeros(CondRHS.getNode()) &&
+          isSplatVector(OpRHS.getNode())) {
+        APInt A = cast<ConstantSDNode>(OpRHS.getOperand(0))->getAPIntValue();
+        if (A.isSignBit())
+          return DAG.getNode(X86ISD::SUBUS, DL, VT, OpLHS, OpRHS);
+      }
+    }
+  }
+
+  // Try to match a min/max vector operation.
+  if (!DCI.isBeforeLegalize() &&
+      N->getOpcode() == ISD::VSELECT && Cond.getOpcode() == ISD::SETCC)
+    if (unsigned Op = matchIntegerMINMAX(Cond, VT, LHS, RHS, DAG, Subtarget))
+      return DAG.getNode(Op, DL, N->getValueType(0), LHS, RHS);
+
   // If we know that this node is legal then we know that it is going to be
   // matched by one of the SSE/AVX BLEND instructions. These instructions only
   // depend on the highest bit in each word. Try to use SimplifyDemandedBits
@@ -14935,8 +15829,9 @@ static SDValue checkBoolTestSetCCCombine(SDValue Cmp, X86::CondCode &CC) {
     // Quit if the constant is neither 0 or 1.
     return SDValue();
 
-  // Skip 'zext' node.
-  if (SetCC.getOpcode() == ISD::ZERO_EXTEND)
+  // Skip 'zext' or 'trunc' node.
+  if (SetCC.getOpcode() == ISD::ZERO_EXTEND ||
+      SetCC.getOpcode() == ISD::TRUNCATE)
     SetCC = SetCC.getOperand(0);
 
   switch (SetCC.getOpcode()) {
@@ -14955,9 +15850,15 @@ static SDValue checkBoolTestSetCCCombine(SDValue Cmp, X86::CondCode &CC) {
       return SDValue();
     // Quit if false value is not a constant.
     if (!FVal) {
-      // A special case for rdrand, where 0 is set if false cond is found.
       SDValue Op = SetCC.getOperand(0);
-      if (Op.getOpcode() != X86ISD::RDRAND)
+      // Skip 'zext' or 'trunc' node.
+      if (Op.getOpcode() == ISD::ZERO_EXTEND ||
+          Op.getOpcode() == ISD::TRUNCATE)
+        Op = Op.getOperand(0);
+      // A special case for rdrand/rdseed, where 0 is set if false cond is
+      // found.
+      if ((Op.getOpcode() != X86ISD::RDRAND &&
+           Op.getOpcode() != X86ISD::RDSEED) || Op.getResNo() != 0)
         return SDValue();
     }
     // Quit if false value is not the constant 0 or 1.
@@ -15137,7 +16038,7 @@ static SDValue PerformCMOVCombine(SDNode *N, SelectionDAG &DAG,
     ConstantSDNode *CmpAgainst = 0;
     if ((Cond.getOpcode() == X86ISD::CMP || Cond.getOpcode() == X86ISD::SUB) &&
         (CmpAgainst = dyn_cast<ConstantSDNode>(Cond.getOperand(1))) &&
-        dyn_cast<ConstantSDNode>(Cond.getOperand(0)) == 0) {
+        !isa<ConstantSDNode>(Cond.getOperand(0))) {
 
       if (CC == X86::COND_NE &&
           CmpAgainst == dyn_cast<ConstantSDNode>(FalseOp)) {
@@ -15158,7 +16059,6 @@ static SDValue PerformCMOVCombine(SDNode *N, SelectionDAG &DAG,
   return SDValue();
 }
 
-
 /// PerformMulCombine - Optimize a single multiply with constant into two
 /// in order to implement it with two cheaper instructions, e.g.
 /// LEA + SHL, LEA + LEA.
@@ -15247,7 +16147,6 @@ static SDValue PerformSHLCombine(SDNode *N, SelectionDAG &DAG) {
     }
   }
 
-
   // Hardware support for vector shifts is sparse which makes us scalarize the
   // vector operations in many cases. Also, on sandybridge ADD is faster than
   // shl.
@@ -15271,127 +16170,14 @@ static SDValue PerformSHLCombine(SDNode *N, SelectionDAG &DAG) {
 static SDValue PerformShiftCombine(SDNode* N, SelectionDAG &DAG,
                                    TargetLowering::DAGCombinerInfo &DCI,
                                    const X86Subtarget *Subtarget) {
-  EVT VT = N->getValueType(0);
   if (N->getOpcode() == ISD::SHL) {
     SDValue V = PerformSHLCombine(N, DAG);
     if (V.getNode()) return V;
   }
 
-  // On X86 with SSE2 support, we can transform this to a vector shift if
-  // all elements are shifted by the same amount.  We can't do this in legalize
-  // because the a constant vector is typically transformed to a constant pool
-  // so we have no knowledge of the shift amount.
-  if (!Subtarget->hasSSE2())
-    return SDValue();
-
-  if (VT != MVT::v2i64 && VT != MVT::v4i32 && VT != MVT::v8i16 &&
-      (!Subtarget->hasAVX2() ||
-       (VT != MVT::v4i64 && VT != MVT::v8i32 && VT != MVT::v16i16)))
-    return SDValue();
-
-  SDValue ShAmtOp = N->getOperand(1);
-  EVT EltVT = VT.getVectorElementType();
-  DebugLoc DL = N->getDebugLoc();
-  SDValue BaseShAmt = SDValue();
-  if (ShAmtOp.getOpcode() == ISD::BUILD_VECTOR) {
-    unsigned NumElts = VT.getVectorNumElements();
-    unsigned i = 0;
-    for (; i != NumElts; ++i) {
-      SDValue Arg = ShAmtOp.getOperand(i);
-      if (Arg.getOpcode() == ISD::UNDEF) continue;
-      BaseShAmt = Arg;
-      break;
-    }
-    // Handle the case where the build_vector is all undef
-    // FIXME: Should DAG allow this?
-    if (i == NumElts)
-      return SDValue();
-
-    for (; i != NumElts; ++i) {
-      SDValue Arg = ShAmtOp.getOperand(i);
-      if (Arg.getOpcode() == ISD::UNDEF) continue;
-      if (Arg != BaseShAmt) {
-        return SDValue();
-      }
-    }
-  } else if (ShAmtOp.getOpcode() == ISD::VECTOR_SHUFFLE &&
-             cast<ShuffleVectorSDNode>(ShAmtOp)->isSplat()) {
-    SDValue InVec = ShAmtOp.getOperand(0);
-    if (InVec.getOpcode() == ISD::BUILD_VECTOR) {
-      unsigned NumElts = InVec.getValueType().getVectorNumElements();
-      unsigned i = 0;
-      for (; i != NumElts; ++i) {
-        SDValue Arg = InVec.getOperand(i);
-        if (Arg.getOpcode() == ISD::UNDEF) continue;
-        BaseShAmt = Arg;
-        break;
-      }
-    } else if (InVec.getOpcode() == ISD::INSERT_VECTOR_ELT) {
-       if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(InVec.getOperand(2))) {
-         unsigned SplatIdx= cast<ShuffleVectorSDNode>(ShAmtOp)->getSplatIndex();
-         if (C->getZExtValue() == SplatIdx)
-           BaseShAmt = InVec.getOperand(1);
-       }
-    }
-    if (BaseShAmt.getNode() == 0) {
-      // Don't create instructions with illegal types after legalize
-      // types has run.
-      if (!DAG.getTargetLoweringInfo().isTypeLegal(EltVT) &&
-          !DCI.isBeforeLegalize())
-        return SDValue();
-
-      BaseShAmt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltVT, ShAmtOp,
-                              DAG.getIntPtrConstant(0));
-    }
-  } else
-    return SDValue();
-
-  // The shift amount is an i32.
-  if (EltVT.bitsGT(MVT::i32))
-    BaseShAmt = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, BaseShAmt);
-  else if (EltVT.bitsLT(MVT::i32))
-    BaseShAmt = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i32, BaseShAmt);
-
-  // The shift amount is identical so we can do a vector shift.
-  SDValue  ValOp = N->getOperand(0);
-  switch (N->getOpcode()) {
-  default:
-    llvm_unreachable("Unknown shift opcode!");
-  case ISD::SHL:
-    switch (VT.getSimpleVT().SimpleTy) {
-    default: return SDValue();
-    case MVT::v2i64:
-    case MVT::v4i32:
-    case MVT::v8i16:
-    case MVT::v4i64:
-    case MVT::v8i32:
-    case MVT::v16i16:
-      return getTargetVShiftNode(X86ISD::VSHLI, DL, VT, ValOp, BaseShAmt, DAG);
-    }
-  case ISD::SRA:
-    switch (VT.getSimpleVT().SimpleTy) {
-    default: return SDValue();
-    case MVT::v4i32:
-    case MVT::v8i16:
-    case MVT::v8i32:
-    case MVT::v16i16:
-      return getTargetVShiftNode(X86ISD::VSRAI, DL, VT, ValOp, BaseShAmt, DAG);
-    }
-  case ISD::SRL:
-    switch (VT.getSimpleVT().SimpleTy) {
-    default: return SDValue();
-    case MVT::v2i64:
-    case MVT::v4i32:
-    case MVT::v8i16:
-    case MVT::v4i64:
-    case MVT::v8i32:
-    case MVT::v16i16:
-      return getTargetVShiftNode(X86ISD::VSRLI, DL, VT, ValOp, BaseShAmt, DAG);
-    }
-  }
+  return SDValue();
 }
 
-
 // CMPEQCombine - Recognize the distinctive  (AND (setcc ...) (setcc ..))
 // where both setccs reference the same FP CMP, and rewrite for CMPEQSS
 // and friends.  Likewise for OR -> CMPNEQSS.
@@ -15420,8 +16206,7 @@ static SDValue CMPEQCombine(SDNode *N, SelectionDAG &DAG,
     if (VT == MVT::f32 || VT == MVT::f64) {
       bool ExpectingFlags = false;
       // Check for any users that want flags:
-      for (SDNode::use_iterator UI = N->use_begin(),
-             UE = N->use_end();
+      for (SDNode::use_iterator UI = N->use_begin(), UE = N->use_end();
            !ExpectingFlags && UI != UE; ++UI)
         switch (UI->getOpcode()) {
         default:
@@ -15500,9 +16285,92 @@ static bool CanFoldXORWithAllOnes(const SDNode *N) {
   return false;
 }
 
+// On AVX/AVX2 the type v8i1 is legalized to v8i16, which is an XMM sized
+// register. In most cases we actually compare or select YMM-sized registers
+// and mixing the two types creates horrible code. This method optimizes
+// some of the transition sequences.
+static SDValue WidenMaskArithmetic(SDNode *N, SelectionDAG &DAG,
+                                 TargetLowering::DAGCombinerInfo &DCI,
+                                 const X86Subtarget *Subtarget) {
+  EVT VT = N->getValueType(0);
+  if (!VT.is256BitVector())
+    return SDValue();
+
+  assert((N->getOpcode() == ISD::ANY_EXTEND ||
+          N->getOpcode() == ISD::ZERO_EXTEND ||
+          N->getOpcode() == ISD::SIGN_EXTEND) && "Invalid Node");
+
+  SDValue Narrow = N->getOperand(0);
+  EVT NarrowVT = Narrow->getValueType(0);
+  if (!NarrowVT.is128BitVector())
+    return SDValue();
+
+  if (Narrow->getOpcode() != ISD::XOR &&
+      Narrow->getOpcode() != ISD::AND &&
+      Narrow->getOpcode() != ISD::OR)
+    return SDValue();
+
+  SDValue N0  = Narrow->getOperand(0);
+  SDValue N1  = Narrow->getOperand(1);
+  DebugLoc DL = Narrow->getDebugLoc();
+
+  // The Left side has to be a trunc.
+  if (N0.getOpcode() != ISD::TRUNCATE)
+    return SDValue();
+
+  // The type of the truncated inputs.
+  EVT WideVT = N0->getOperand(0)->getValueType(0);
+  if (WideVT != VT)
+    return SDValue();
+
+  // The right side has to be a 'trunc' or a constant vector.
+  bool RHSTrunc = N1.getOpcode() == ISD::TRUNCATE;
+  bool RHSConst = (isSplatVector(N1.getNode()) &&
+                   isa<ConstantSDNode>(N1->getOperand(0)));
+  if (!RHSTrunc && !RHSConst)
+    return SDValue();
+
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+
+  if (!TLI.isOperationLegalOrPromote(Narrow->getOpcode(), WideVT))
+    return SDValue();
+
+  // Set N0 and N1 to hold the inputs to the new wide operation.
+  N0 = N0->getOperand(0);
+  if (RHSConst) {
+    N1 = DAG.getNode(ISD::ZERO_EXTEND, DL, WideVT.getScalarType(),
+                     N1->getOperand(0));
+    SmallVector<SDValue, 8> C(WideVT.getVectorNumElements(), N1);
+    N1 = DAG.getNode(ISD::BUILD_VECTOR, DL, WideVT, &C[0], C.size());
+  } else if (RHSTrunc) {
+    N1 = N1->getOperand(0);
+  }
+
+  // Generate the wide operation.
+  SDValue Op = DAG.getNode(Narrow->getOpcode(), DL, WideVT, N0, N1);
+  unsigned Opcode = N->getOpcode();
+  switch (Opcode) {
+  case ISD::ANY_EXTEND:
+    return Op;
+  case ISD::ZERO_EXTEND: {
+    unsigned InBits = NarrowVT.getScalarType().getSizeInBits();
+    APInt Mask = APInt::getAllOnesValue(InBits);
+    Mask = Mask.zext(VT.getScalarType().getSizeInBits());
+    return DAG.getNode(ISD::AND, DL, VT,
+                       Op, DAG.getConstant(Mask, VT));
+  }
+  case ISD::SIGN_EXTEND:
+    return DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, VT,
+                       Op, DAG.getValueType(NarrowVT));
+  default:
+    llvm_unreachable("Unexpected opcode");
+  }
+}
+
 static SDValue PerformAndCombine(SDNode *N, SelectionDAG &DAG,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const X86Subtarget *Subtarget) {
+  EVT VT = N->getValueType(0);
   if (DCI.isBeforeLegalizeOps())
     return SDValue();
 
@@ -15510,9 +16378,7 @@ static SDValue PerformAndCombine(SDNode *N, SelectionDAG &DAG,
   if (R.getNode())
     return R;
 
-  EVT VT = N->getValueType(0);
-
-  // Create ANDN, BLSI, and BLSR instructions
+  // Create BLSI, and BLSR instructions
   // BLSI is X & (-X)
   // BLSR is X & (X-1)
   if (Subtarget->hasBMI() && (VT == MVT::i32 || VT == MVT::i64)) {
@@ -15520,13 +16386,6 @@ static SDValue PerformAndCombine(SDNode *N, SelectionDAG &DAG,
     SDValue N1 = N->getOperand(1);
     DebugLoc DL = N->getDebugLoc();
 
-    // Check LHS for not
-    if (N0.getOpcode() == ISD::XOR && isAllOnes(N0.getOperand(1)))
-      return DAG.getNode(X86ISD::ANDN, DL, VT, N0.getOperand(0), N1);
-    // Check RHS for not
-    if (N1.getOpcode() == ISD::XOR && isAllOnes(N1.getOperand(1)))
-      return DAG.getNode(X86ISD::ANDN, DL, VT, N1.getOperand(0), N0);
-
     // Check LHS for neg
     if (N0.getOpcode() == ISD::SUB && N0.getOperand(1) == N1 &&
         isZero(N0.getOperand(0)))
@@ -15579,6 +16438,7 @@ static SDValue PerformAndCombine(SDNode *N, SelectionDAG &DAG,
 static SDValue PerformOrCombine(SDNode *N, SelectionDAG &DAG,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const X86Subtarget *Subtarget) {
+  EVT VT = N->getValueType(0);
   if (DCI.isBeforeLegalizeOps())
     return SDValue();
 
@@ -15586,15 +16446,13 @@ static SDValue PerformOrCombine(SDNode *N, SelectionDAG &DAG,
   if (R.getNode())
     return R;
 
-  EVT VT = N->getValueType(0);
-
   SDValue N0 = N->getOperand(0);
   SDValue N1 = N->getOperand(1);
 
   // look for psign/blend
   if (VT == MVT::v2i64 || VT == MVT::v4i64) {
     if (!Subtarget->hasSSSE3() ||
-        (VT == MVT::v4i64 && !Subtarget->hasAVX2()))
+        (VT == MVT::v4i64 && !Subtarget->hasInt256()))
       return SDValue();
 
     // Canonicalize pandn to RHS
@@ -15628,13 +16486,19 @@ static SDValue PerformOrCombine(SDNode *N, SelectionDAG &DAG,
       // Validate that the Mask operand is a vector sra node.
       // FIXME: what to do for bytes, since there is a psignb/pblendvb, but
       // there is no psrai.b
-      if (Mask.getOpcode() != X86ISD::VSRAI)
-        return SDValue();
-
-      // Check that the SRA is all signbits.
-      SDValue SraC = Mask.getOperand(1);
-      unsigned SraAmt  = cast<ConstantSDNode>(SraC)->getZExtValue();
       unsigned EltBits = MaskVT.getVectorElementType().getSizeInBits();
+      unsigned SraAmt = ~0;
+      if (Mask.getOpcode() == ISD::SRA) {
+        SDValue Amt = Mask.getOperand(1);
+        if (isSplatVector(Amt.getNode())) {
+          SDValue SclrAmt = Amt->getOperand(0);
+          if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(SclrAmt))
+            SraAmt = C->getZExtValue();
+        }
+      } else if (Mask.getOpcode() == X86ISD::VSRAI) {
+        SDValue SraC = Mask.getOperand(1);
+        SraAmt  = cast<ConstantSDNode>(SraC)->getZExtValue();
+      }
       if ((SraAmt + 1) != EltBits)
         return SDValue();
 
@@ -15762,6 +16626,7 @@ static SDValue performIntegerAbsCombine(SDNode *N, SelectionDAG &DAG) {
 static SDValue PerformXorCombine(SDNode *N, SelectionDAG &DAG,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const X86Subtarget *Subtarget) {
+  EVT VT = N->getValueType(0);
   if (DCI.isBeforeLegalizeOps())
     return SDValue();
 
@@ -15775,8 +16640,6 @@ static SDValue PerformXorCombine(SDNode *N, SelectionDAG &DAG,
   if (!Subtarget->hasBMI())
     return SDValue();
 
-  EVT VT = N->getValueType(0);
-
   if (VT != MVT::i32 && VT != MVT::i64)
     return SDValue();
 
@@ -15807,23 +16670,61 @@ static SDValue PerformLOADCombine(SDNode *N, SelectionDAG &DAG,
   EVT MemVT = Ld->getMemoryVT();
   DebugLoc dl = Ld->getDebugLoc();
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  unsigned RegSz = RegVT.getSizeInBits();
 
+  // On Sandybridge unaligned 256bit loads are inefficient.
   ISD::LoadExtType Ext = Ld->getExtensionType();
+  unsigned Alignment = Ld->getAlignment();
+  bool IsAligned = Alignment == 0 || Alignment >= MemVT.getSizeInBits()/8;
+  if (RegVT.is256BitVector() && !Subtarget->hasInt256() &&
+      !DCI.isBeforeLegalizeOps() && !IsAligned && Ext == ISD::NON_EXTLOAD) {
+    unsigned NumElems = RegVT.getVectorNumElements();
+    if (NumElems < 2)
+      return SDValue();
+
+    SDValue Ptr = Ld->getBasePtr();
+    SDValue Increment = DAG.getConstant(16, TLI.getPointerTy());
+
+    EVT HalfVT = EVT::getVectorVT(*DAG.getContext(), MemVT.getScalarType(),
+                                  NumElems/2);
+    SDValue Load1 = DAG.getLoad(HalfVT, dl, Ld->getChain(), Ptr,
+                                Ld->getPointerInfo(), Ld->isVolatile(),
+                                Ld->isNonTemporal(), Ld->isInvariant(),
+                                Alignment);
+    Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, Increment);
+    SDValue Load2 = DAG.getLoad(HalfVT, dl, Ld->getChain(), Ptr,
+                                Ld->getPointerInfo(), Ld->isVolatile(),
+                                Ld->isNonTemporal(), Ld->isInvariant(),
+                                std::min(16U, Alignment));
+    SDValue TF = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
+                             Load1.getValue(1),
+                             Load2.getValue(1));
+
+    SDValue NewVec = DAG.getUNDEF(RegVT);
+    NewVec = Insert128BitVector(NewVec, Load1, 0, DAG, dl);
+    NewVec = Insert128BitVector(NewVec, Load2, NumElems/2, DAG, dl);
+    return DCI.CombineTo(N, NewVec, TF, true);
+  }
 
   // If this is a vector EXT Load then attempt to optimize it using a
-  // shuffle. We need SSSE3 shuffles.
+  // shuffle. If SSSE3 is not available we may emit an illegal shuffle but the
+  // expansion is still better than scalar code.
+  // We generate X86ISD::VSEXT for SEXTLOADs if it's available, otherwise we'll
+  // emit a shuffle and a arithmetic shift.
   // TODO: It is possible to support ZExt by zeroing the undef values
   // during the shuffle phase or after the shuffle.
-  if (RegVT.isVector() && RegVT.isInteger() &&
-      Ext == ISD::EXTLOAD && Subtarget->hasSSSE3()) {
+  if (RegVT.isVector() && RegVT.isInteger() && Subtarget->hasSSE2() &&
+      (Ext == ISD::EXTLOAD || Ext == ISD::SEXTLOAD)) {
     assert(MemVT != RegVT && "Cannot extend to the same type");
     assert(MemVT.isVector() && "Must load a vector from memory");
 
     unsigned NumElems = RegVT.getVectorNumElements();
-    unsigned RegSz = RegVT.getSizeInBits();
     unsigned MemSz = MemVT.getSizeInBits();
     assert(RegSz > MemSz && "Register size must be greater than the mem size");
 
+    if (Ext == ISD::SEXTLOAD && RegSz == 256 && !Subtarget->hasInt256())
+      return SDValue();
+
     // All sizes must be a power of two.
     if (!isPowerOf2_32(RegSz * MemSz * NumElems))
       return SDValue();
@@ -15847,16 +16748,23 @@ static SDValue PerformLOADCombine(SDNode *N, SelectionDAG &DAG,
     // Calculate the number of scalar loads that we need to perform
     // in order to load our vector from memory.
     unsigned NumLoads = MemSz / SclrLoadTy.getSizeInBits();
+    if (Ext == ISD::SEXTLOAD && NumLoads > 1)
+      return SDValue();
+
+    unsigned loadRegZize = RegSz;
+    if (Ext == ISD::SEXTLOAD && RegSz == 256)
+      loadRegZize /= 2;
 
     // Represent our vector as a sequence of elements which are the
     // largest scalar that we can load.
     EVT LoadUnitVecVT = EVT::getVectorVT(*DAG.getContext(), SclrLoadTy,
-      RegSz/SclrLoadTy.getSizeInBits());
+      loadRegZize/SclrLoadTy.getSizeInBits());
 
     // Represent the data using the same element type that is stored in
     // memory. In practice, we ''widen'' MemVT.
-    EVT WideVecVT = EVT::getVectorVT(*DAG.getContext(), MemVT.getScalarType(),
-                                  RegSz/MemVT.getScalarType().getSizeInBits());
+    EVT WideVecVT =
+          EVT::getVectorVT(*DAG.getContext(), MemVT.getScalarType(),
+                       loadRegZize/MemVT.getScalarType().getSizeInBits());
 
     assert(WideVecVT.getSizeInBits() == LoadUnitVecVT.getSizeInBits() &&
       "Invalid vector type");
@@ -15897,6 +16805,39 @@ static SDValue PerformLOADCombine(SDNode *N, SelectionDAG &DAG,
     SDValue SlicedVec = DAG.getNode(ISD::BITCAST, dl, WideVecVT, Res);
     unsigned SizeRatio = RegSz/MemSz;
 
+    if (Ext == ISD::SEXTLOAD) {
+      // If we have SSE4.1 we can directly emit a VSEXT node.
+      if (Subtarget->hasSSE41()) {
+        SDValue Sext = DAG.getNode(X86ISD::VSEXT, dl, RegVT, SlicedVec);
+        return DCI.CombineTo(N, Sext, TF, true);
+      }
+
+      // Otherwise we'll shuffle the small elements in the high bits of the
+      // larger type and perform an arithmetic shift. If the shift is not legal
+      // it's better to scalarize.
+      if (!TLI.isOperationLegalOrCustom(ISD::SRA, RegVT))
+        return SDValue();
+
+      // Redistribute the loaded elements into the different locations.
+      SmallVector<int, 8> ShuffleVec(NumElems * SizeRatio, -1);
+      for (unsigned i = 0; i != NumElems; ++i)
+        ShuffleVec[i*SizeRatio + SizeRatio-1] = i;
+
+      SDValue Shuff = DAG.getVectorShuffle(WideVecVT, dl, SlicedVec,
+                                           DAG.getUNDEF(WideVecVT),
+                                           &ShuffleVec[0]);
+
+      Shuff = DAG.getNode(ISD::BITCAST, dl, RegVT, Shuff);
+
+      // Build the arithmetic shift.
+      unsigned Amt = RegVT.getVectorElementType().getSizeInBits() -
+                     MemVT.getVectorElementType().getSizeInBits();
+      Shuff = DAG.getNode(ISD::SRA, dl, RegVT, Shuff,
+                          DAG.getConstant(Amt, RegVT));
+
+      return DCI.CombineTo(N, Shuff, TF, true);
+    }
+
     // Redistribute the loaded elements into the different locations.
     SmallVector<int, 8> ShuffleVec(NumElems * SizeRatio, -1);
     for (unsigned i = 0; i != NumElems; ++i)
@@ -15930,11 +16871,16 @@ static SDValue PerformSTORECombine(SDNode *N, SelectionDAG &DAG,
   // On Sandy Bridge, 256-bit memory operations are executed by two
   // 128-bit ports. However, on Haswell it is better to issue a single 256-bit
   // memory  operation.
-  if (VT.is256BitVector() && !Subtarget->hasAVX2() &&
-      StoredVal.getNode()->getOpcode() == ISD::CONCAT_VECTORS &&
-      StoredVal.getNumOperands() == 2) {
-    SDValue Value0 = StoredVal.getOperand(0);
-    SDValue Value1 = StoredVal.getOperand(1);
+  unsigned Alignment = St->getAlignment();
+  bool IsAligned = Alignment == 0 || Alignment >= VT.getSizeInBits()/8;
+  if (VT.is256BitVector() && !Subtarget->hasInt256() &&
+      StVT == VT && !IsAligned) {
+    unsigned NumElems = VT.getVectorNumElements();
+    if (NumElems < 2)
+      return SDValue();
+
+    SDValue Value0 = Extract128BitVector(StoredVal, 0, DAG, dl);
+    SDValue Value1 = Extract128BitVector(StoredVal, NumElems/2, DAG, dl);
 
     SDValue Stride = DAG.getConstant(16, TLI.getPointerTy());
     SDValue Ptr0 = St->getBasePtr();
@@ -15942,10 +16888,11 @@ static SDValue PerformSTORECombine(SDNode *N, SelectionDAG &DAG,
 
     SDValue Ch0 = DAG.getStore(St->getChain(), dl, Value0, Ptr0,
                                 St->getPointerInfo(), St->isVolatile(),
-                                St->isNonTemporal(), St->getAlignment());
+                                St->isNonTemporal(), Alignment);
     SDValue Ch1 = DAG.getStore(St->getChain(), dl, Value1, Ptr1,
                                 St->getPointerInfo(), St->isVolatile(),
-                                St->isNonTemporal(), St->getAlignment());
+                                St->isNonTemporal(),
+                                std::min(16U, Alignment));
     return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Ch0, Ch1);
   }
 
@@ -16030,7 +16977,6 @@ static SDValue PerformSTORECombine(SDNode *N, SelectionDAG &DAG,
                                Chains.size());
   }
 
-
   // Turn load->store of MMX types into GPR load/stores.  This avoids clobbering
   // the FP state in cases where an emms may be missing.
   // A preferable solution to the general problem is to figure out the right
@@ -16041,8 +16987,8 @@ static SDValue PerformSTORECombine(SDNode *N, SelectionDAG &DAG,
     return SDValue();
 
   const Function *F = DAG.getMachineFunction().getFunction();
-  bool NoImplicitFloatOps = F->getFnAttributes().
-    hasAttribute(Attributes::NoImplicitFloat);
+  bool NoImplicitFloatOps = F->getAttributes().
+    hasAttribute(AttributeSet::FunctionIndex, Attribute::NoImplicitFloat);
   bool F64IsLegal = !DAG.getTarget().Options.UseSoftFloat && !NoImplicitFloatOps
                      && Subtarget->hasSSE2();
   if ((VT.isVector() ||
@@ -16278,7 +17224,7 @@ static SDValue PerformFADDCombine(SDNode *N, SelectionDAG &DAG,
 
   // Try to synthesize horizontal adds from adds of shuffles.
   if (((Subtarget->hasSSE3() && (VT == MVT::v4f32 || VT == MVT::v2f64)) ||
-       (Subtarget->hasAVX() && (VT == MVT::v8f32 || VT == MVT::v4f64))) &&
+       (Subtarget->hasFp256() && (VT == MVT::v8f32 || VT == MVT::v4f64))) &&
       isHorizontalBinOp(LHS, RHS, true))
     return DAG.getNode(X86ISD::FHADD, N->getDebugLoc(), VT, LHS, RHS);
   return SDValue();
@@ -16293,7 +17239,7 @@ static SDValue PerformFSUBCombine(SDNode *N, SelectionDAG &DAG,
 
   // Try to synthesize horizontal subs from subs of shuffles.
   if (((Subtarget->hasSSE3() && (VT == MVT::v4f32 || VT == MVT::v2f64)) ||
-       (Subtarget->hasAVX() && (VT == MVT::v8f32 || VT == MVT::v4f64))) &&
+       (Subtarget->hasFp256() && (VT == MVT::v8f32 || VT == MVT::v4f64))) &&
       isHorizontalBinOp(LHS, RHS, false))
     return DAG.getNode(X86ISD::FHSUB, N->getDebugLoc(), VT, LHS, RHS);
   return SDValue();
@@ -16336,7 +17282,6 @@ static SDValue PerformFMinFMaxCombine(SDNode *N, SelectionDAG &DAG) {
                      N->getOperand(0), N->getOperand(1));
 }
 
-
 /// PerformFANDCombine - Do target-specific dag combines on X86ISD::FAND nodes.
 static SDValue PerformFANDCombine(SDNode *N, SelectionDAG &DAG) {
   // FAND(0.0, x) -> 0.0
@@ -16382,58 +17327,57 @@ static SDValue PerformVZEXT_MOVLCombine(SDNode *N, SelectionDAG &DAG) {
   return SDValue();
 }
 
-static SDValue PerformSExtCombine(SDNode *N, SelectionDAG &DAG,
-                                  TargetLowering::DAGCombinerInfo &DCI,
-                                  const X86Subtarget *Subtarget) {
-  if (!DCI.isBeforeLegalizeOps())
-    return SDValue();
-
-  if (!Subtarget->hasAVX())
+static SDValue PerformSIGN_EXTEND_INREGCombine(SDNode *N, SelectionDAG &DAG, 
+                                               const X86Subtarget *Subtarget) {
+  EVT VT = N->getValueType(0);
+  if (!VT.isVector())
     return SDValue();
 
-  EVT VT = N->getValueType(0);
-  SDValue Op = N->getOperand(0);
-  EVT OpVT = Op.getValueType();
+  SDValue N0 = N->getOperand(0);
+  SDValue N1 = N->getOperand(1);
+  EVT ExtraVT = cast<VTSDNode>(N1)->getVT();
   DebugLoc dl = N->getDebugLoc();
 
-  if ((VT == MVT::v4i64 && OpVT == MVT::v4i32) ||
-      (VT == MVT::v8i32 && OpVT == MVT::v8i16)) {
-
-    if (Subtarget->hasAVX2())
-      return DAG.getNode(X86ISD::VSEXT_MOVL, dl, VT, Op);
-
-    // Optimize vectors in AVX mode
-    // Sign extend  v8i16 to v8i32 and
-    //              v4i32 to v4i64
-    //
-    // Divide input vector into two parts
-    // for v4i32 the shuffle mask will be { 0, 1, -1, -1} {2, 3, -1, -1}
-    // use vpmovsx instruction to extend v4i32 -> v2i64; v8i16 -> v4i32
-    // concat the vectors to original VT
-
-    unsigned NumElems = OpVT.getVectorNumElements();
-    SDValue Undef = DAG.getUNDEF(OpVT);
-
-    SmallVector<int,8> ShufMask1(NumElems, -1);
-    for (unsigned i = 0; i != NumElems/2; ++i)
-      ShufMask1[i] = i;
-
-    SDValue OpLo = DAG.getVectorShuffle(OpVT, dl, Op, Undef, &ShufMask1[0]);
+  // The SIGN_EXTEND_INREG to v4i64 is expensive operation on the
+  // both SSE and AVX2 since there is no sign-extended shift right
+  // operation on a vector with 64-bit elements.
+  //(sext_in_reg (v4i64 anyext (v4i32 x )), ExtraVT) ->
+  // (v4i64 sext (v4i32 sext_in_reg (v4i32 x , ExtraVT)))
+  if (VT == MVT::v4i64 && (N0.getOpcode() == ISD::ANY_EXTEND ||
+      N0.getOpcode() == ISD::SIGN_EXTEND)) {
+    SDValue N00 = N0.getOperand(0);
 
-    SmallVector<int,8> ShufMask2(NumElems, -1);
-    for (unsigned i = 0; i != NumElems/2; ++i)
-      ShufMask2[i] = i + NumElems/2;
+    // EXTLOAD has a better solution on AVX2, 
+    // it may be replaced with X86ISD::VSEXT node.
+    if (N00.getOpcode() == ISD::LOAD && Subtarget->hasInt256())
+      if (!ISD::isNormalLoad(N00.getNode()))
+        return SDValue();
 
-    SDValue OpHi = DAG.getVectorShuffle(OpVT, dl, Op, Undef, &ShufMask2[0]);
+    if (N00.getValueType() == MVT::v4i32 && ExtraVT.getSizeInBits() < 128) {
+        SDValue Tmp = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, MVT::v4i32, 
+                                  N00, N1);
+      return DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v4i64, Tmp);
+    }
+  }
+  return SDValue();
+}
 
-    EVT HalfVT = EVT::getVectorVT(*DAG.getContext(), VT.getScalarType(),
-                                  VT.getVectorNumElements()/2);
+static SDValue PerformSExtCombine(SDNode *N, SelectionDAG &DAG,
+                                  TargetLowering::DAGCombinerInfo &DCI,
+                                  const X86Subtarget *Subtarget) {
+  if (!DCI.isBeforeLegalizeOps())
+    return SDValue();
 
-    OpLo = DAG.getNode(X86ISD::VSEXT_MOVL, dl, HalfVT, OpLo);
-    OpHi = DAG.getNode(X86ISD::VSEXT_MOVL, dl, HalfVT, OpHi);
+  if (!Subtarget->hasFp256())
+    return SDValue();
 
-    return DAG.getNode(ISD::CONCAT_VECTORS, dl, VT, OpLo, OpHi);
+  EVT VT = N->getValueType(0);
+  if (VT.isVector() && VT.getSizeInBits() == 256) {
+    SDValue R = WidenMaskArithmetic(N, DAG, DCI, Subtarget);
+    if (R.getNode())
+      return R;
   }
+
   return SDValue();
 }
 
@@ -16487,58 +17431,26 @@ static SDValue PerformZExtCombine(SDNode *N, SelectionDAG &DAG,
   DebugLoc dl = N->getDebugLoc();
   SDValue N0 = N->getOperand(0);
   EVT VT = N->getValueType(0);
-  EVT OpVT = N0.getValueType();
 
   if (N0.getOpcode() == ISD::AND &&
       N0.hasOneUse() &&
       N0.getOperand(0).hasOneUse()) {
     SDValue N00 = N0.getOperand(0);
-    if (N00.getOpcode() != X86ISD::SETCC_CARRY)
-      return SDValue();
-    ConstantSDNode *C = dyn_cast<ConstantSDNode>(N0.getOperand(1));
-    if (!C || C->getZExtValue() != 1)
-      return SDValue();
-    return DAG.getNode(ISD::AND, dl, VT,
-                       DAG.getNode(X86ISD::SETCC_CARRY, dl, VT,
-                                   N00.getOperand(0), N00.getOperand(1)),
-                       DAG.getConstant(1, VT));
+    if (N00.getOpcode() == X86ISD::SETCC_CARRY) {
+      ConstantSDNode *C = dyn_cast<ConstantSDNode>(N0.getOperand(1));
+      if (!C || C->getZExtValue() != 1)
+        return SDValue();
+      return DAG.getNode(ISD::AND, dl, VT,
+                         DAG.getNode(X86ISD::SETCC_CARRY, dl, VT,
+                                     N00.getOperand(0), N00.getOperand(1)),
+                         DAG.getConstant(1, VT));
+    }
   }
 
-  // Optimize vectors in AVX mode:
-  //
-  //   v8i16 -> v8i32
-  //   Use vpunpcklwd for 4 lower elements  v8i16 -> v4i32.
-  //   Use vpunpckhwd for 4 upper elements  v8i16 -> v4i32.
-  //   Concat upper and lower parts.
-  //
-  //   v4i32 -> v4i64
-  //   Use vpunpckldq for 4 lower elements  v4i32 -> v2i64.
-  //   Use vpunpckhdq for 4 upper elements  v4i32 -> v2i64.
-  //   Concat upper and lower parts.
-  //
-  if (!DCI.isBeforeLegalizeOps())
-    return SDValue();
-
-  if (!Subtarget->hasAVX())
-    return SDValue();
-
-  if (((VT == MVT::v8i32) && (OpVT == MVT::v8i16)) ||
-      ((VT == MVT::v4i64) && (OpVT == MVT::v4i32)))  {
-
-    if (Subtarget->hasAVX2())
-      return DAG.getNode(X86ISD::VZEXT_MOVL, dl, VT, N0);
-
-    SDValue ZeroVec = getZeroVector(OpVT, Subtarget, DAG, dl);
-    SDValue OpLo = getUnpackl(DAG, dl, OpVT, N0, ZeroVec);
-    SDValue OpHi = getUnpackh(DAG, dl, OpVT, N0, ZeroVec);
-
-    EVT HVT = EVT::getVectorVT(*DAG.getContext(), VT.getVectorElementType(),
-                               VT.getVectorNumElements()/2);
-
-    OpLo = DAG.getNode(ISD::BITCAST, dl, HVT, OpLo);
-    OpHi = DAG.getNode(ISD::BITCAST, dl, HVT, OpHi);
-
-    return DAG.getNode(ISD::CONCAT_VECTORS, dl, VT, OpLo, OpHi);
+  if (VT.is256BitVector()) {
+    SDValue R = WidenMaskArithmetic(N, DAG, DCI, Subtarget);
+    if (R.getNode())
+      return R;
   }
 
   return SDValue();
@@ -16570,8 +17482,8 @@ static SDValue PerformISDSETCCCombine(SDNode *N, SelectionDAG &DAG) {
   return SDValue();
 }
 
-// Helper function of PerformSETCCCombine. It is to materialize "setb reg" 
-// as "sbb reg,reg", since it can be extended without zext and produces 
+// Helper function of PerformSETCCCombine. It is to materialize "setb reg"
+// as "sbb reg,reg", since it can be extended without zext and produces
 // an all-ones bit which is more useful than 0/1 in some cases.
 static SDValue MaterializeSETB(DebugLoc DL, SDValue EFLAGS, SelectionDAG &DAG) {
   return DAG.getNode(ISD::AND, DL, MVT::i8,
@@ -16589,13 +17501,13 @@ static SDValue PerformSETCCCombine(SDNode *N, SelectionDAG &DAG,
   SDValue EFLAGS = N->getOperand(1);
 
   if (CC == X86::COND_A) {
-    // Try to convert COND_A into COND_B in an attempt to facilitate 
+    // Try to convert COND_A into COND_B in an attempt to facilitate
     // materializing "setb reg".
     //
     // Do not flip "e > c", where "c" is a constant, because Cmp instruction
     // cannot take an immediate as its first operand.
     //
-    if (EFLAGS.getOpcode() == X86ISD::SUB && EFLAGS.hasOneUse() && 
+    if (EFLAGS.getOpcode() == X86ISD::SUB && EFLAGS.hasOneUse() &&
         EFLAGS.getValueType().isInteger() &&
         !isa<ConstantSDNode>(EFLAGS.getOperand(1))) {
       SDValue NewSub = DAG.getNode(X86ISD::SUB, EFLAGS.getDebugLoc(),
@@ -16751,7 +17663,7 @@ static SDValue PerformAddCombine(SDNode *N, SelectionDAG &DAG,
 
   // Try to synthesize horizontal adds from adds of shuffles.
   if (((Subtarget->hasSSSE3() && (VT == MVT::v8i16 || VT == MVT::v4i32)) ||
-       (Subtarget->hasAVX2() && (VT == MVT::v16i16 || VT == MVT::v8i32))) &&
+       (Subtarget->hasInt256() && (VT == MVT::v16i16 || VT == MVT::v8i32))) &&
       isHorizontalBinOp(Op0, Op1, true))
     return DAG.getNode(X86ISD::HADD, N->getDebugLoc(), VT, Op0, Op1);
 
@@ -16784,7 +17696,7 @@ static SDValue PerformSubCombine(SDNode *N, SelectionDAG &DAG,
   // Try to synthesize horizontal adds from adds of shuffles.
   EVT VT = N->getValueType(0);
   if (((Subtarget->hasSSSE3() && (VT == MVT::v8i16 || VT == MVT::v4i32)) ||
-       (Subtarget->hasAVX2() && (VT == MVT::v16i16 || VT == MVT::v8i32))) &&
+       (Subtarget->hasInt256() && (VT == MVT::v16i16 || VT == MVT::v8i32))) &&
       isHorizontalBinOp(Op0, Op1, true))
     return DAG.getNode(X86ISD::HSUB, N->getDebugLoc(), VT, Op0, Op1);
 
@@ -16803,7 +17715,8 @@ static SDValue performVZEXTCombine(SDNode *N, SelectionDAG &DAG,
   if (In.getOpcode() != X86ISD::VZEXT)
     return SDValue();
 
-  return DAG.getNode(X86ISD::VZEXT, N->getDebugLoc(), N->getValueType(0), In.getOperand(0));
+  return DAG.getNode(X86ISD::VZEXT, N->getDebugLoc(), N->getValueType(0),
+                     In.getOperand(0));
 }
 
 SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
@@ -16841,13 +17754,14 @@ SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
   case ISD::ANY_EXTEND:
   case ISD::ZERO_EXTEND:    return PerformZExtCombine(N, DAG, DCI, Subtarget);
   case ISD::SIGN_EXTEND:    return PerformSExtCombine(N, DAG, DCI, Subtarget);
+  case ISD::SIGN_EXTEND_INREG: return PerformSIGN_EXTEND_INREGCombine(N, DAG, Subtarget);
   case ISD::TRUNCATE:       return PerformTruncateCombine(N, DAG,DCI,Subtarget);
   case ISD::SETCC:          return PerformISDSETCCCombine(N, DAG);
   case X86ISD::SETCC:       return PerformSETCCCombine(N, DAG, DCI, Subtarget);
   case X86ISD::BRCOND:      return PerformBrCondCombine(N, DAG, DCI, Subtarget);
   case X86ISD::VZEXT:       return performVZEXTCombine(N, DAG, DCI, Subtarget);
   case X86ISD::SHUFP:       // Handle all target specific shuffles
-  case X86ISD::PALIGN:
+  case X86ISD::PALIGNR:
   case X86ISD::UNPCKH:
   case X86ISD::UNPCKL:
   case X86ISD::MOVHLPS:
@@ -17030,7 +17944,7 @@ bool X86TargetLowering::ExpandInlineAsm(CallInst *CI) const {
       AsmPieces.clear();
       const std::string &ConstraintsStr = IA->getConstraintString();
       SplitString(StringRef(ConstraintsStr).substr(5), AsmPieces, ",");
-      std::sort(AsmPieces.begin(), AsmPieces.end());
+      array_pod_sort(AsmPieces.begin(), AsmPieces.end());
       if (AsmPieces.size() == 4 &&
           AsmPieces[0] == "~{cc}" &&
           AsmPieces[1] == "~{dirflag}" &&
@@ -17048,7 +17962,7 @@ bool X86TargetLowering::ExpandInlineAsm(CallInst *CI) const {
       AsmPieces.clear();
       const std::string &ConstraintsStr = IA->getConstraintString();
       SplitString(StringRef(ConstraintsStr).substr(5), AsmPieces, ",");
-      std::sort(AsmPieces.begin(), AsmPieces.end());
+      array_pod_sort(AsmPieces.begin(), AsmPieces.end());
       if (AsmPieces.size() == 4 &&
           AsmPieces[0] == "~{cc}" &&
           AsmPieces[1] == "~{dirflag}" &&
@@ -17074,8 +17988,6 @@ bool X86TargetLowering::ExpandInlineAsm(CallInst *CI) const {
   return false;
 }
 
-
-
 /// getConstraintType - Given a constraint letter, return the type of
 /// constraint it is for this target.
 X86TargetLowering::ConstraintType
@@ -17152,17 +18064,17 @@ TargetLowering::ConstraintWeight
   case 'f':
   case 't':
   case 'u':
-      if (type->isFloatingPointTy())
-        weight = CW_SpecificReg;
-      break;
+    if (type->isFloatingPointTy())
+      weight = CW_SpecificReg;
+    break;
   case 'y':
-      if (type->isX86_MMXTy() && Subtarget->hasMMX())
-        weight = CW_SpecificReg;
-      break;
+    if (type->isX86_MMXTy() && Subtarget->hasMMX())
+      weight = CW_SpecificReg;
+    break;
   case 'x':
   case 'Y':
     if (((type->getPrimitiveSizeInBits() == 128) && Subtarget->hasSSE1()) ||
-        ((type->getPrimitiveSizeInBits() == 256) && Subtarget->hasAVX()))
+        ((type->getPrimitiveSizeInBits() == 256) && Subtarget->hasFp256()))
       weight = CW_Register;
     break;
   case 'I':
@@ -17530,7 +18442,7 @@ X86TargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
   // really want an 8-bit or 32-bit register, map to the appropriate register
   // class and return the appropriate register.
   if (Res.second == &X86::GR16RegClass) {
-    if (VT == MVT::i8) {
+    if (VT == MVT::i8 || VT == MVT::i1) {
       unsigned DestReg = 0;
       switch (Res.first) {
       default: break;
@@ -17543,7 +18455,7 @@ X86TargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
         Res.first = DestReg;
         Res.second = &X86::GR8RegClass;
       }
-    } else if (VT == MVT::i32) {
+    } else if (VT == MVT::i32 || VT == MVT::f32) {
       unsigned DestReg = 0;
       switch (Res.first) {
       default: break;
@@ -17560,7 +18472,7 @@ X86TargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
         Res.first = DestReg;
         Res.second = &X86::GR32RegClass;
       }
-    } else if (VT == MVT::i64) {
+    } else if (VT == MVT::i64 || VT == MVT::f64) {
       unsigned DestReg = 0;
       switch (Res.first) {
       default: break;
@@ -17598,207 +18510,3 @@ X86TargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
 
   return Res;
 }
-
-//===----------------------------------------------------------------------===//
-//
-// X86 cost model.
-//
-//===----------------------------------------------------------------------===//
-
-struct X86CostTblEntry {
-  int ISD;
-  MVT Type;
-  unsigned Cost;
-};
-
-static int
-FindInTable(const X86CostTblEntry *Tbl, unsigned len, int ISD, MVT Ty) {
-  for (unsigned int i = 0; i < len; ++i)
-    if (Tbl[i].ISD == ISD && Tbl[i].Type == Ty)
-      return i;
-
-  // Could not find an entry.
-  return -1;
-}
-
-struct X86TypeConversionCostTblEntry {
-  int ISD;
-  MVT Dst;
-  MVT Src;
-  unsigned Cost;
-};
-
-static int
-FindInConvertTable(const X86TypeConversionCostTblEntry *Tbl, unsigned len,
-                   int ISD, MVT Dst, MVT Src) {
-  for (unsigned int i = 0; i < len; ++i)
-    if (Tbl[i].ISD == ISD && Tbl[i].Src == Src && Tbl[i].Dst == Dst)
-      return i;
-
-  // Could not find an entry.
-  return -1;
-}
-
-unsigned
-X86VectorTargetTransformInfo::getArithmeticInstrCost(unsigned Opcode,
-                                                     Type *Ty) const {
-  // Legalize the type.
-  std::pair<unsigned, MVT> LT = getTypeLegalizationCost(Ty);
-
-  int ISD = InstructionOpcodeToISD(Opcode);
-  assert(ISD && "Invalid opcode");
-
-  const X86Subtarget &ST = TLI->getTargetMachine().getSubtarget<X86Subtarget>();
-
-  static const X86CostTblEntry AVX1CostTable[] = {
-    // We don't have to scalarize unsupported ops. We can issue two half-sized
-    // operations and we only need to extract the upper YMM half.
-    // Two ops + 1 extract + 1 insert = 4.
-    { ISD::MUL,     MVT::v8i32,    4 },
-    { ISD::SUB,     MVT::v8i32,    4 },
-    { ISD::ADD,     MVT::v8i32,    4 },
-    { ISD::MUL,     MVT::v4i64,    4 },
-    { ISD::SUB,     MVT::v4i64,    4 },
-    { ISD::ADD,     MVT::v4i64,    4 },
-    };
-
-  // Look for AVX1 lowering tricks.
-  if (ST.hasAVX()) {
-    int Idx = FindInTable(AVX1CostTable, array_lengthof(AVX1CostTable), ISD,
-                          LT.second);
-    if (Idx != -1)
-      return LT.first * AVX1CostTable[Idx].Cost;
-  }
-  // Fallback to the default implementation.
-  return VectorTargetTransformImpl::getArithmeticInstrCost(Opcode, Ty);
-}
-
-unsigned
-X86VectorTargetTransformInfo::getVectorInstrCost(unsigned Opcode, Type *Val,
-                                                 unsigned Index) const {
-  assert(Val->isVectorTy() && "This must be a vector type");
-
-  if (Index != -1U) {
-    // Legalize the type.
-    std::pair<unsigned, MVT> LT = getTypeLegalizationCost(Val);
-
-    // This type is legalized to a scalar type.
-    if (!LT.second.isVector())
-      return 0;
-
-    // The type may be split. Normalize the index to the new type.
-    unsigned Width = LT.second.getVectorNumElements();
-    Index = Index % Width;
-
-    // Floating point scalars are already located in index #0.
-    if (Val->getScalarType()->isFloatingPointTy() && Index == 0)
-      return 0;
-  }
-
-  return VectorTargetTransformImpl::getVectorInstrCost(Opcode, Val, Index);
-}
-
-unsigned X86VectorTargetTransformInfo::getCmpSelInstrCost(unsigned Opcode,
-                                                          Type *ValTy,
-                                                          Type *CondTy) const {
-  // Legalize the type.
-  std::pair<unsigned, MVT> LT = getTypeLegalizationCost(ValTy);
-
-  MVT MTy = LT.second;
-
-  int ISD = InstructionOpcodeToISD(Opcode);
-  assert(ISD && "Invalid opcode");
-
-  const X86Subtarget &ST =
-  TLI->getTargetMachine().getSubtarget<X86Subtarget>();
-
-  static const X86CostTblEntry SSE42CostTbl[] = {
-    { ISD::SETCC,   MVT::v2f64,   1 },
-    { ISD::SETCC,   MVT::v4f32,   1 },
-    { ISD::SETCC,   MVT::v2i64,   1 },
-    { ISD::SETCC,   MVT::v4i32,   1 },
-    { ISD::SETCC,   MVT::v8i16,   1 },
-    { ISD::SETCC,   MVT::v16i8,   1 },
-  };
-
-  static const X86CostTblEntry AVX1CostTbl[] = {
-    { ISD::SETCC,   MVT::v4f64,   1 },
-    { ISD::SETCC,   MVT::v8f32,   1 },
-    // AVX1 does not support 8-wide integer compare.
-    { ISD::SETCC,   MVT::v4i64,   4 },
-    { ISD::SETCC,   MVT::v8i32,   4 },
-    { ISD::SETCC,   MVT::v16i16,  4 },
-    { ISD::SETCC,   MVT::v32i8,   4 },
-  };
-
-  static const X86CostTblEntry AVX2CostTbl[] = {
-    { ISD::SETCC,   MVT::v4i64,   1 },
-    { ISD::SETCC,   MVT::v8i32,   1 },
-    { ISD::SETCC,   MVT::v16i16,  1 },
-    { ISD::SETCC,   MVT::v32i8,   1 },
-  };
-
-  if (ST.hasSSE42()) {
-    int Idx = FindInTable(SSE42CostTbl, array_lengthof(SSE42CostTbl), ISD, MTy);
-    if (Idx != -1)
-      return LT.first * SSE42CostTbl[Idx].Cost;
-  }
-
-  if (ST.hasAVX()) {
-    int Idx = FindInTable(AVX1CostTbl, array_lengthof(AVX1CostTbl), ISD, MTy);
-    if (Idx != -1)
-      return LT.first * AVX1CostTbl[Idx].Cost;
-  }
-
-  if (ST.hasAVX2()) {
-    int Idx = FindInTable(AVX2CostTbl, array_lengthof(AVX2CostTbl), ISD, MTy);
-    if (Idx != -1)
-      return LT.first * AVX2CostTbl[Idx].Cost;
-  }
-
-  return VectorTargetTransformImpl::getCmpSelInstrCost(Opcode, ValTy, CondTy);
-}
-
-unsigned X86VectorTargetTransformInfo::getCastInstrCost(unsigned Opcode,
-                                                        Type *Dst,
-                                                        Type *Src) const {
-  int ISD = InstructionOpcodeToISD(Opcode);
-  assert(ISD && "Invalid opcode");
-
-  EVT SrcTy = TLI->getValueType(Src);
-  EVT DstTy = TLI->getValueType(Dst);
-
-  if (!SrcTy.isSimple() || !DstTy.isSimple())
-    return VectorTargetTransformImpl::getCastInstrCost(Opcode, Dst, Src);
-
-  const X86Subtarget &ST = TLI->getTargetMachine().getSubtarget<X86Subtarget>();
-
-  static const X86TypeConversionCostTblEntry AVXConversionTbl[] = {
-    { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i16, 1 },
-    { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i16, 1 },
-    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i32, 1 },
-    { ISD::ZERO_EXTEND, MVT::v4i64, MVT::v4i32, 1 },
-    { ISD::TRUNCATE,    MVT::v4i32, MVT::v4i64, 1 },
-    { ISD::TRUNCATE,    MVT::v8i16, MVT::v8i32, 1 },
-    { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i8,  1 },
-    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i8,  1 },
-    { ISD::UINT_TO_FP,  MVT::v8f32, MVT::v8i8,  1 },
-    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i8,  1 },
-    { ISD::FP_TO_SINT,  MVT::v8i8,  MVT::v8f32, 1 },
-    { ISD::FP_TO_SINT,  MVT::v4i8,  MVT::v4f32, 1 },
-    { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i1,  6 },
-    { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i1,  9 },
-    { ISD::TRUNCATE,    MVT::v8i32, MVT::v8i64, 3 },
-  };
-
-  if (ST.hasAVX()) {
-    int Idx = FindInConvertTable(AVXConversionTbl,
-                                 array_lengthof(AVXConversionTbl),
-                                 ISD, DstTy.getSimpleVT(), SrcTy.getSimpleVT());
-    if (Idx != -1)
-      return AVXConversionTbl[Idx].Cost;
-  }
-
-  return VectorTargetTransformImpl::getCastInstrCost(Opcode, Dst, Src);
-}
-
diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index 465c6036ada6..5725f7aea581 100644
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@@ -15,15 +15,14 @@
 #ifndef X86ISELLOWERING_H
 #define X86ISELLOWERING_H
 
-#include "X86Subtarget.h"
-#include "X86RegisterInfo.h"
 #include "X86MachineFunctionInfo.h"
-#include "llvm/Target/TargetLowering.h"
-#include "llvm/Target/TargetTransformImpl.h"
-#include "llvm/Target/TargetOptions.h"
+#include "X86RegisterInfo.h"
+#include "X86Subtarget.h"
+#include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/FastISel.h"
 #include "llvm/CodeGen/SelectionDAG.h"
-#include "llvm/CodeGen/CallingConvLower.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/TargetOptions.h"
 
 namespace llvm {
   namespace X86ISD {
@@ -176,13 +175,14 @@ namespace llvm {
       /// PSIGN - Copy integer sign.
       PSIGN,
 
-      /// BLENDV - Blend where the selector is an XMM.
+      /// BLENDV - Blend where the selector is a register.
       BLENDV,
 
-      /// BLENDxx - Blend where the selector is an immediate.
-      BLENDPW,
-      BLENDPS,
-      BLENDPD,
+      /// BLENDI - Blend where the selector is an immediate.
+      BLENDI,
+
+      // SUBUS - Integer sub with unsigned saturation.
+      SUBUS,
 
       /// HADD - Integer horizontal add.
       HADD,
@@ -196,6 +196,12 @@ namespace llvm {
       /// FHSUB - Floating point horizontal sub.
       FHSUB,
 
+      /// UMAX, UMIN - Unsigned integer max and min.
+      UMAX, UMIN,
+
+      /// SMAX, SMIN - Signed integer max and min.
+      SMAX, SMIN,
+
       /// FMAX, FMIN - Floating point max and min.
       ///
       FMAX, FMIN,
@@ -228,11 +234,8 @@ namespace llvm {
       // EH_SJLJ_LONGJMP - SjLj exception handling longjmp.
       EH_SJLJ_LONGJMP,
 
-      /// TC_RETURN - Tail call return.
-      ///   operand #0 chain
-      ///   operand #1 callee (register or absolute)
-      ///   operand #2 stack adjustment
-      ///   operand #3 optional in flag
+      /// TC_RETURN - Tail call return. See X86TargetLowering::LowerCall for
+      /// the list of operands.
       TC_RETURN,
 
       // VZEXT_MOVL - Vector move low and zero extend.
@@ -272,8 +275,6 @@ namespace llvm {
       ADD, SUB, ADC, SBB, SMUL,
       INC, DEC, OR, XOR, AND,
 
-      ANDN, // ANDN - Bitwise AND NOT with FLAGS results.
-
       BLSI,   // BLSI - Extract lowest set isolated bit
       BLSMSK, // BLSMSK - Get mask up to lowest set bit
       BLSR,   // BLSR - Reset lowest set bit
@@ -290,7 +291,7 @@ namespace llvm {
       TESTP,
 
       // Several flavors of instructions with vector shuffle behaviors.
-      PALIGN,
+      PALIGNR,
       PSHUFD,
       PSHUFHW,
       PSHUFLW,
@@ -355,10 +356,17 @@ namespace llvm {
       // RDRAND - Get a random integer and indicate whether it is valid in CF.
       RDRAND,
 
+      // RDSEED - Get a NIST SP800-90B & C compliant random integer and
+      // indicate whether it is valid in CF.
+      RDSEED,
+
       // PCMP*STRI
       PCMPISTRI,
       PCMPESTRI,
 
+      // XTEST - Test if in transactional execution.
+      XTEST,
+
       // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG,
       // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG -
       // Atomic 64-bit binary operations.
@@ -470,7 +478,7 @@ namespace llvm {
 
     virtual unsigned getJumpTableEncoding() const;
 
-    virtual MVT getShiftAmountTy(EVT LHSTy) const { return MVT::i8; }
+    virtual MVT getScalarShiftAmountTy(EVT LHSTy) const { return MVT::i8; }
 
     virtual const MCExpr *
     LowerCustomJumpTableEntry(const MachineJumpTableInfo *MJTI,
@@ -496,23 +504,29 @@ namespace llvm {
     /// lowering. If DstAlign is zero that means it's safe to destination
     /// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
     /// means there isn't a need to check it against alignment requirement,
-    /// probably because the source does not need to be loaded. If
-    /// 'IsZeroVal' is true, that means it's safe to return a
-    /// non-scalar-integer type, e.g. empty string source, constant, or loaded
-    /// from memory. 'MemcpyStrSrc' indicates whether the memcpy source is
-    /// constant so it does not need to be loaded.
+    /// probably because the source does not need to be loaded. If 'IsMemset' is
+    /// true, that means it's expanding a memset. If 'ZeroMemset' is true, that
+    /// means it's a memset of zero. 'MemcpyStrSrc' indicates whether the memcpy
+    /// source is constant so it does not need to be loaded.
     /// It returns EVT::Other if the type should be determined using generic
     /// target-independent logic.
     virtual EVT
-    getOptimalMemOpType(uint64_t Size, unsigned DstAlign, unsigned SrcAlign,
-                        bool IsZeroVal, bool MemcpyStrSrc,
+    getOptimalMemOpType(uint64_t Size, unsigned DstAlign, unsigned SrcAlign, 
+                        bool IsMemset, bool ZeroMemset, bool MemcpyStrSrc,
                         MachineFunction &MF) const;
 
+    /// isSafeMemOpType - Returns true if it's safe to use load / store of the
+    /// specified type to expand memcpy / memset inline. This is mostly true
+    /// for all types except for some special cases. For example, on X86
+    /// targets without SSE2 f64 load / store are done with fldl / fstpl which
+    /// also does type conversion. Note the specified type doesn't have to be
+    /// legal as the hook is used before type legalization.
+    virtual bool isSafeMemOpType(MVT VT) const;
+
     /// allowsUnalignedMemoryAccesses - Returns true if the target allows
-    /// unaligned memory accesses. of the specified type.
-    virtual bool allowsUnalignedMemoryAccesses(EVT VT) const {
-      return true;
-    }
+    /// unaligned memory accesses. of the specified type. Returns whether it
+    /// is "fast" by reference in the second argument.
+    virtual bool allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const;
 
     /// LowerOperation - Provide custom lowering hooks for some operations.
     ///
@@ -630,6 +644,7 @@ namespace llvm {
     /// result out to 64 bits.
     virtual bool isZExtFree(Type *Ty1, Type *Ty2) const;
     virtual bool isZExtFree(EVT VT1, EVT VT2) const;
+    virtual bool isZExtFree(SDValue Val, EVT VT2) const;
 
     /// isFMAFasterThanMulAndAdd - Return true if an FMA operation is faster than
     /// a pair of mul and add instructions. fmuladd intrinsics will be expanded to
@@ -710,7 +725,7 @@ namespace llvm {
 
   protected:
     std::pair<const TargetRegisterClass*, uint8_t>
-    findRepresentativeClass(EVT VT) const;
+    findRepresentativeClass(MVT VT) const;
 
   private:
     /// Subtarget - Keep a pointer to the X86Subtarget around so that we can
@@ -783,9 +798,7 @@ namespace llvm {
     SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerINSERT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerGlobalAddress(const GlobalValue *GV, DebugLoc dl,
@@ -800,18 +813,18 @@ namespace llvm {
     SDValue LowerUINT_TO_FP_i64(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerUINT_TO_FP_i32(SDValue Op, SelectionDAG &DAG) const;
     SDValue lowerUINT_TO_FP_vec(SDValue Op, SelectionDAG &DAG) const;
-    SDValue lowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const;
-    SDValue lowerZERO_EXTEND(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerZERO_EXTEND(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerSIGN_EXTEND(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerANY_EXTEND(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFP_TO_UINT(SDValue Op, SelectionDAG &DAG) const;
-    SDValue lowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFABS(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFNEG(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerToBT(SDValue And, ISD::CondCode CC,
                       DebugLoc dl, SelectionDAG &DAG) const;
     SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG) const;
-    SDValue LowerVSETCC(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerMEMSET(SDValue Op, SelectionDAG &DAG) const;
@@ -828,8 +841,9 @@ namespace llvm {
     SDValue LowerINIT_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerShift(SDValue Op, SelectionDAG &DAG) const;
-
+    SDValue LowerSDIV(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const;
 
     // Utility functions to help LowerVECTOR_SHUFFLE & LowerBUILD_VECTOR
     SDValue LowerVectorBroadcast(SDValue Op, SelectionDAG &DAG) const;
@@ -838,7 +852,7 @@ namespace llvm {
 
     SDValue LowerVectorAllZeroTest(SDValue Op, SelectionDAG &DAG) const;
 
-    SDValue lowerVectorIntExtend(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerVectorIntExtend(SDValue Op, SelectionDAG &DAG) const;
 
     virtual SDValue
       LowerFormalArguments(SDValue Chain,
@@ -861,9 +875,8 @@ namespace llvm {
 
     virtual bool mayBeEmittedAsTailCall(CallInst *CI) const;
 
-    virtual EVT
-    getTypeForExtArgOrReturn(LLVMContext &Context, EVT VT,
-                             ISD::NodeType ExtendKind) const;
+    virtual MVT
+    getTypeForExtArgOrReturn(MVT VT, ISD::NodeType ExtendKind) const;
 
     virtual bool
     CanLowerReturn(CallingConv::ID CallConv, MachineFunction &MF,
@@ -932,23 +945,6 @@ namespace llvm {
     FastISel *createFastISel(FunctionLoweringInfo &funcInfo,
                              const TargetLibraryInfo *libInfo);
   }
-
-  class X86VectorTargetTransformInfo : public VectorTargetTransformImpl {
-  public:
-    explicit X86VectorTargetTransformInfo(const TargetLowering *TL) :
-    VectorTargetTransformImpl(TL) {}
-
-    virtual unsigned getArithmeticInstrCost(unsigned Opcode, Type *Ty) const;
-
-    virtual unsigned getVectorInstrCost(unsigned Opcode, Type *Val,
-                                        unsigned Index) const;
-
-    unsigned getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
-                                Type *CondTy) const;
-
-    virtual unsigned getCastInstrCost(unsigned Opcode, Type *Dst,
-                                      Type *Src) const;
-  };
 }
 
 #endif    // X86ISELLOWERING_H
diff --git a/lib/Target/X86/X86Instr3DNow.td b/lib/Target/X86/X86Instr3DNow.td
index 54b91c3edb8b..ba1aede3c1a0 100644
--- a/lib/Target/X86/X86Instr3DNow.td
+++ b/lib/Target/X86/X86Instr3DNow.td
@@ -84,15 +84,16 @@ defm PI2FD    : I3DNow_conv_rm_int<0x0D, "pi2fd">;
 defm PMULHRW  : I3DNow_binop_rm_int<0xB7, "pmulhrw">;
 
 
-def FEMMS : I3DNow<0x0E, RawFrm, (outs), (ins), "femms", [(int_x86_mmx_femms)]>;
+def FEMMS : I3DNow<0x0E, RawFrm, (outs), (ins), "femms",
+                   [(int_x86_mmx_femms)]>;
 
-def PREFETCH  : I3DNow<0x0D, MRM0m, (outs), (ins i32mem:$addr),
-                       "prefetch $addr", []>;
+def PREFETCH : I3DNow<0x0D, MRM0m, (outs), (ins i8mem:$addr),
+                      "prefetch\t$addr",
+                      [(prefetch addr:$addr, (i32 0), imm, (i32 1))]>;
 
-// FIXME: Diassembler gets a bogus decode conflict.
-let isAsmParserOnly = 1 in
-def PREFETCHW : I3DNow<0x0D, MRM1m, (outs), (ins i16mem:$addr),
-                       "prefetchw $addr", []>;
+def PREFETCHW : I<0x0D, MRM1m, (outs), (ins i8mem:$addr), "prefetchw\t$addr",
+                  [(prefetch addr:$addr, (i32 1), (i32 3), (i32 1))]>, TB,
+                Requires<[HasPrefetchW]>;
 
 // "3DNowA" instructions
 defm PF2IW    : I3DNow_conv_rm_int<0x1C, "pf2iw", "a">;
diff --git a/lib/Target/X86/X86InstrArithmetic.td b/lib/Target/X86/X86InstrArithmetic.td
index f790611b8f8c..225e9720da0c 100644
--- a/lib/Target/X86/X86InstrArithmetic.td
+++ b/lib/Target/X86/X86InstrArithmetic.td
@@ -14,7 +14,7 @@
 
 //===----------------------------------------------------------------------===//
 // LEA - Load Effective Address
-
+let SchedRW = [WriteLEA] in {
 let neverHasSideEffects = 1 in
 def LEA16r   : I<0x8D, MRMSrcMem,
                  (outs GR16:$dst), (ins i32mem:$src),
@@ -29,48 +29,59 @@ def LEA32r   : I<0x8D, MRMSrcMem,
 def LEA64_32r : I<0x8D, MRMSrcMem,
                   (outs GR32:$dst), (ins lea64_32mem:$src),
                   "lea{l}\t{$src|$dst}, {$dst|$src}",
-                  [(set GR32:$dst, lea32addr:$src)], IIC_LEA>,
+                  [(set GR32:$dst, lea64_32addr:$src)], IIC_LEA>,
                   Requires<[In64BitMode]>;
 
 let isReMaterializable = 1 in
-def LEA64r   : RI<0x8D, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
+def LEA64r   : RI<0x8D, MRMSrcMem, (outs GR64:$dst), (ins lea64mem:$src),
                   "lea{q}\t{$src|$dst}, {$dst|$src}",
                   [(set GR64:$dst, lea64addr:$src)], IIC_LEA>;
-
-
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 //  Fixed-Register Multiplication and Division Instructions.
 //
 
+// SchedModel info for instruction that loads one value and gets the second
+// (and possibly third) value from a register.
+// This is used for instructions that put the memory operands before other
+// uses.
+class SchedLoadReg<SchedWrite SW> : Sched<[SW,
+  // Memory operand.
+  ReadDefault, ReadDefault, ReadDefault, ReadDefault, ReadDefault,
+  // Register reads (implicit or explicit).
+  ReadAfterLd, ReadAfterLd]>;
+
 // Extra precision multiplication
 
 // AL is really implied by AX, but the registers in Defs must match the
 // SDNode results (i8, i32).
+// AL,AH = AL*GR8
 let Defs = [AL,EFLAGS,AX], Uses = [AL] in
 def MUL8r  : I<0xF6, MRM4r, (outs),  (ins GR8:$src), "mul{b}\t$src",
                // FIXME: Used for 8-bit mul, ignore result upper 8 bits.
                // This probably ought to be moved to a def : Pat<> if the
                // syntax can be accepted.
                [(set AL, (mul AL, GR8:$src)),
-                (implicit EFLAGS)], IIC_MUL8>;     // AL,AH = AL*GR8
-
+                (implicit EFLAGS)], IIC_MUL8>, Sched<[WriteIMul]>;
+// AX,DX = AX*GR16
 let Defs = [AX,DX,EFLAGS], Uses = [AX], neverHasSideEffects = 1 in
 def MUL16r : I<0xF7, MRM4r, (outs),  (ins GR16:$src),
-               "mul{w}\t$src", 
-               [], IIC_MUL16_REG>, OpSize;    // AX,DX = AX*GR16
-
+               "mul{w}\t$src",
+               [], IIC_MUL16_REG>, OpSize, Sched<[WriteIMul]>;
+// EAX,EDX = EAX*GR32
 let Defs = [EAX,EDX,EFLAGS], Uses = [EAX], neverHasSideEffects = 1 in
 def MUL32r : I<0xF7, MRM4r, (outs),  (ins GR32:$src),
-               "mul{l}\t$src",   // EAX,EDX = EAX*GR32
+               "mul{l}\t$src",
                [/*(set EAX, EDX, EFLAGS, (X86umul_flag EAX, GR32:$src))*/],
-               IIC_MUL32_REG>;
+               IIC_MUL32_REG>, Sched<[WriteIMul]>;
+// RAX,RDX = RAX*GR64
 let Defs = [RAX,RDX,EFLAGS], Uses = [RAX], neverHasSideEffects = 1 in
 def MUL64r : RI<0xF7, MRM4r, (outs), (ins GR64:$src),
-                "mul{q}\t$src",          // RAX,RDX = RAX*GR64
+                "mul{q}\t$src",
                 [/*(set RAX, RDX, EFLAGS, (X86umul_flag RAX, GR64:$src))*/],
-                IIC_MUL64>;
-
+                IIC_MUL64>, Sched<[WriteIMul]>;
+// AL,AH = AL*[mem8]
 let Defs = [AL,EFLAGS,AX], Uses = [AL] in
 def MUL8m  : I<0xF6, MRM4m, (outs), (ins i8mem :$src),
                "mul{b}\t$src",
@@ -78,51 +89,60 @@ def MUL8m  : I<0xF6, MRM4m, (outs), (ins i8mem :$src),
                // This probably ought to be moved to a def : Pat<> if the
                // syntax can be accepted.
                [(set AL, (mul AL, (loadi8 addr:$src))),
-                (implicit EFLAGS)], IIC_MUL8>;   // AL,AH = AL*[mem8]
-
+                (implicit EFLAGS)], IIC_MUL8>, SchedLoadReg<WriteIMulLd>;
+// AX,DX = AX*[mem16]
 let mayLoad = 1, neverHasSideEffects = 1 in {
 let Defs = [AX,DX,EFLAGS], Uses = [AX] in
 def MUL16m : I<0xF7, MRM4m, (outs), (ins i16mem:$src),
                "mul{w}\t$src",
-               [], IIC_MUL16_MEM>, OpSize; // AX,DX = AX*[mem16]
-
+               [], IIC_MUL16_MEM>, OpSize, SchedLoadReg<WriteIMulLd>;
+// EAX,EDX = EAX*[mem32]
 let Defs = [EAX,EDX,EFLAGS], Uses = [EAX] in
 def MUL32m : I<0xF7, MRM4m, (outs), (ins i32mem:$src),
               "mul{l}\t$src",
-              [], IIC_MUL32_MEM>;          // EAX,EDX = EAX*[mem32]
+              [], IIC_MUL32_MEM>, SchedLoadReg<WriteIMulLd>;
+// RAX,RDX = RAX*[mem64]
 let Defs = [RAX,RDX,EFLAGS], Uses = [RAX] in
 def MUL64m : RI<0xF7, MRM4m, (outs), (ins i64mem:$src),
-                "mul{q}\t$src", [], IIC_MUL64>;   // RAX,RDX = RAX*[mem64]
+                "mul{q}\t$src", [], IIC_MUL64>, SchedLoadReg<WriteIMulLd>;
 }
 
 let neverHasSideEffects = 1 in {
+// AL,AH = AL*GR8
 let Defs = [AL,EFLAGS,AX], Uses = [AL] in
 def IMUL8r  : I<0xF6, MRM5r, (outs),  (ins GR8:$src), "imul{b}\t$src", [],
-              IIC_IMUL8>; // AL,AH = AL*GR8
+              IIC_IMUL8>, Sched<[WriteIMul]>;
+// AX,DX = AX*GR16
 let Defs = [AX,DX,EFLAGS], Uses = [AX] in
 def IMUL16r : I<0xF7, MRM5r, (outs),  (ins GR16:$src), "imul{w}\t$src", [],
-              IIC_IMUL16_RR>, OpSize;    // AX,DX = AX*GR16
+              IIC_IMUL16_RR>, OpSize, Sched<[WriteIMul]>;
+// EAX,EDX = EAX*GR32
 let Defs = [EAX,EDX,EFLAGS], Uses = [EAX] in
 def IMUL32r : I<0xF7, MRM5r, (outs),  (ins GR32:$src), "imul{l}\t$src", [],
-              IIC_IMUL32_RR>; // EAX,EDX = EAX*GR32
+              IIC_IMUL32_RR>, Sched<[WriteIMul]>;
+// RAX,RDX = RAX*GR64
 let Defs = [RAX,RDX,EFLAGS], Uses = [RAX] in
 def IMUL64r : RI<0xF7, MRM5r, (outs), (ins GR64:$src), "imul{q}\t$src", [],
-              IIC_IMUL64_RR>; // RAX,RDX = RAX*GR64
+              IIC_IMUL64_RR>, Sched<[WriteIMul]>;
 
 let mayLoad = 1 in {
+// AL,AH = AL*[mem8]
 let Defs = [AL,EFLAGS,AX], Uses = [AL] in
 def IMUL8m  : I<0xF6, MRM5m, (outs), (ins i8mem :$src),
-                "imul{b}\t$src", [], IIC_IMUL8>;    // AL,AH = AL*[mem8]
+                "imul{b}\t$src", [], IIC_IMUL8>, SchedLoadReg<WriteIMulLd>;
+// AX,DX = AX*[mem16]
 let Defs = [AX,DX,EFLAGS], Uses = [AX] in
 def IMUL16m : I<0xF7, MRM5m, (outs), (ins i16mem:$src),
-                "imul{w}\t$src", [], IIC_IMUL16_MEM>, OpSize;
-                // AX,DX = AX*[mem16]
+                "imul{w}\t$src", [], IIC_IMUL16_MEM>, OpSize,
+              SchedLoadReg<WriteIMulLd>;
+// EAX,EDX = EAX*[mem32]
 let Defs = [EAX,EDX,EFLAGS], Uses = [EAX] in
 def IMUL32m : I<0xF7, MRM5m, (outs), (ins i32mem:$src),
-                "imul{l}\t$src", [], IIC_IMUL32_MEM>;  // EAX,EDX = EAX*[mem32]
+                "imul{l}\t$src", [], IIC_IMUL32_MEM>, SchedLoadReg<WriteIMulLd>;
+// RAX,RDX = RAX*[mem64]
 let Defs = [RAX,RDX,EFLAGS], Uses = [RAX] in
 def IMUL64m : RI<0xF7, MRM5m, (outs), (ins i64mem:$src),
-                 "imul{q}\t$src", [], IIC_IMUL64>;  // RAX,RDX = RAX*[mem64]
+                 "imul{q}\t$src", [], IIC_IMUL64>, SchedLoadReg<WriteIMulLd>;
 }
 } // neverHasSideEffects
 
@@ -130,7 +150,8 @@ def IMUL64m : RI<0xF7, MRM5m, (outs), (ins i64mem:$src),
 let Defs = [EFLAGS] in {
 let Constraints = "$src1 = $dst" in {
 
-let isCommutable = 1 in {  // X = IMUL Y, Z --> X = IMUL Z, Y
+let isCommutable = 1, SchedRW = [WriteIMul] in {
+// X = IMUL Y, Z --> X = IMUL Z, Y
 // Register-Register Signed Integer Multiply
 def IMUL16rr : I<0xAF, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src1,GR16:$src2),
                  "imul{w}\t{$src2, $dst|$dst, $src2}",
@@ -148,9 +169,10 @@ def IMUL64rr : RI<0xAF, MRMSrcReg, (outs GR64:$dst),
                   [(set GR64:$dst, EFLAGS,
                         (X86smul_flag GR64:$src1, GR64:$src2))], IIC_IMUL64_RR>,
                  TB;
-}
+} // isCommutable, SchedRW
 
 // Register-Memory Signed Integer Multiply
+let SchedRW = [WriteIMulLd, ReadAfterLd] in {
 def IMUL16rm : I<0xAF, MRMSrcMem, (outs GR16:$dst),
                                   (ins GR16:$src1, i16mem:$src2),
                  "imul{w}\t{$src2, $dst|$dst, $src2}",
@@ -158,7 +180,7 @@ def IMUL16rm : I<0xAF, MRMSrcMem, (outs GR16:$dst),
                        (X86smul_flag GR16:$src1, (load addr:$src2)))],
                        IIC_IMUL16_RM>,
                TB, OpSize;
-def IMUL32rm : I<0xAF, MRMSrcMem, (outs GR32:$dst), 
+def IMUL32rm : I<0xAF, MRMSrcMem, (outs GR32:$dst),
                  (ins GR32:$src1, i32mem:$src2),
                  "imul{l}\t{$src2, $dst|$dst, $src2}",
                  [(set GR32:$dst, EFLAGS,
@@ -172,18 +194,20 @@ def IMUL64rm : RI<0xAF, MRMSrcMem, (outs GR64:$dst),
                         (X86smul_flag GR64:$src1, (load addr:$src2)))],
                         IIC_IMUL64_RM>,
                TB;
+} // SchedRW
 } // Constraints = "$src1 = $dst"
 
 } // Defs = [EFLAGS]
 
 // Surprisingly enough, these are not two address instructions!
 let Defs = [EFLAGS] in {
+let SchedRW = [WriteIMul] in {
 // Register-Integer Signed Integer Multiply
 def IMUL16rri  : Ii16<0x69, MRMSrcReg,                      // GR16 = GR16*I16
                       (outs GR16:$dst), (ins GR16:$src1, i16imm:$src2),
                       "imul{w}\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                      [(set GR16:$dst, EFLAGS, 
-                            (X86smul_flag GR16:$src1, imm:$src2))], 
+                      [(set GR16:$dst, EFLAGS,
+                            (X86smul_flag GR16:$src1, imm:$src2))],
                             IIC_IMUL16_RRI>, OpSize;
 def IMUL16rri8 : Ii8<0x6B, MRMSrcReg,                       // GR16 = GR16*I8
                      (outs GR16:$dst), (ins GR16:$src1, i16i8imm:$src2),
@@ -216,9 +240,10 @@ def IMUL64rri8 : RIi8<0x6B, MRMSrcReg,                      // GR64 = GR64*I8
                       [(set GR64:$dst, EFLAGS,
                             (X86smul_flag GR64:$src1, i64immSExt8:$src2))],
                             IIC_IMUL64_RRI>;
-
+} // SchedRW
 
 // Memory-Integer Signed Integer Multiply
+let SchedRW = [WriteIMulLd] in {
 def IMUL16rmi  : Ii16<0x69, MRMSrcMem,                     // GR16 = [mem16]*I16
                       (outs GR16:$dst), (ins i16mem:$src1, i16imm:$src2),
                       "imul{w}\t{$src2, $src1, $dst|$dst, $src1, $src2}",
@@ -260,12 +285,15 @@ def IMUL64rmi8 : RIi8<0x6B, MRMSrcMem,                      // GR64 = [mem64]*I8
                             (X86smul_flag (load addr:$src1),
                                           i64immSExt8:$src2))],
                                           IIC_IMUL64_RMI>;
+} // SchedRW
 } // Defs = [EFLAGS]
 
 
 
 
 // unsigned division/remainder
+let hasSideEffects = 1 in { // so that we don't speculatively execute
+let SchedRW = [WriteIDiv] in {
 let Defs = [AL,EFLAGS,AX], Uses = [AX] in
 def DIV8r  : I<0xF6, MRM6r, (outs),  (ins GR8:$src),    // AX/r8 = AL,AH
                "div{b}\t$src", [], IIC_DIV8_REG>;
@@ -279,24 +307,30 @@ def DIV32r : I<0xF7, MRM6r, (outs),  (ins GR32:$src),   // EDX:EAX/r32 = EAX,EDX
 let Defs = [RAX,RDX,EFLAGS], Uses = [RAX,RDX] in
 def DIV64r : RI<0xF7, MRM6r, (outs), (ins GR64:$src),
                 "div{q}\t$src", [], IIC_DIV64>;
+} // SchedRW
 
 let mayLoad = 1 in {
 let Defs = [AL,EFLAGS,AX], Uses = [AX] in
 def DIV8m  : I<0xF6, MRM6m, (outs), (ins i8mem:$src),   // AX/[mem8] = AL,AH
-               "div{b}\t$src", [], IIC_DIV8_MEM>;
+               "div{b}\t$src", [], IIC_DIV8_MEM>,
+             SchedLoadReg<WriteIDivLd>;
 let Defs = [AX,DX,EFLAGS], Uses = [AX,DX] in
 def DIV16m : I<0xF7, MRM6m, (outs), (ins i16mem:$src),  // DX:AX/[mem16] = AX,DX
-               "div{w}\t$src", [], IIC_DIV16>, OpSize;
+               "div{w}\t$src", [], IIC_DIV16>, OpSize,
+             SchedLoadReg<WriteIDivLd>;
 let Defs = [EAX,EDX,EFLAGS], Uses = [EAX,EDX] in    // EDX:EAX/[mem32] = EAX,EDX
 def DIV32m : I<0xF7, MRM6m, (outs), (ins i32mem:$src),
-               "div{l}\t$src", [], IIC_DIV32>;
+               "div{l}\t$src", [], IIC_DIV32>,
+             SchedLoadReg<WriteIDivLd>;
 // RDX:RAX/[mem64] = RAX,RDX
 let Defs = [RAX,RDX,EFLAGS], Uses = [RAX,RDX] in
 def DIV64m : RI<0xF7, MRM6m, (outs), (ins i64mem:$src),
-                "div{q}\t$src", [], IIC_DIV64>;
+                "div{q}\t$src", [], IIC_DIV64>,
+             SchedLoadReg<WriteIDivLd>;
 }
 
 // Signed division/remainder.
+let SchedRW = [WriteIDiv] in {
 let Defs = [AL,EFLAGS,AX], Uses = [AX] in
 def IDIV8r : I<0xF6, MRM7r, (outs),  (ins GR8:$src),    // AX/r8 = AL,AH
                "idiv{b}\t$src", [], IIC_IDIV8>;
@@ -310,21 +344,27 @@ def IDIV32r: I<0xF7, MRM7r, (outs),  (ins GR32:$src),   // EDX:EAX/r32 = EAX,EDX
 let Defs = [RAX,RDX,EFLAGS], Uses = [RAX,RDX] in
 def IDIV64r: RI<0xF7, MRM7r, (outs), (ins GR64:$src),
                 "idiv{q}\t$src", [], IIC_IDIV64>;
+} // SchedRW
 
 let mayLoad = 1 in {
 let Defs = [AL,EFLAGS,AX], Uses = [AX] in
 def IDIV8m : I<0xF6, MRM7m, (outs), (ins i8mem:$src),   // AX/[mem8] = AL,AH
-               "idiv{b}\t$src", [], IIC_IDIV8>;
+               "idiv{b}\t$src", [], IIC_IDIV8>,
+             SchedLoadReg<WriteIDivLd>;
 let Defs = [AX,DX,EFLAGS], Uses = [AX,DX] in
 def IDIV16m: I<0xF7, MRM7m, (outs), (ins i16mem:$src),  // DX:AX/[mem16] = AX,DX
-               "idiv{w}\t$src", [], IIC_IDIV16>, OpSize;
+               "idiv{w}\t$src", [], IIC_IDIV16>, OpSize,
+             SchedLoadReg<WriteIDivLd>;
 let Defs = [EAX,EDX,EFLAGS], Uses = [EAX,EDX] in    // EDX:EAX/[mem32] = EAX,EDX
-def IDIV32m: I<0xF7, MRM7m, (outs), (ins i32mem:$src), 
-               "idiv{l}\t$src", [], IIC_IDIV32>;
+def IDIV32m: I<0xF7, MRM7m, (outs), (ins i32mem:$src),
+               "idiv{l}\t$src", [], IIC_IDIV32>,
+             SchedLoadReg<WriteIDivLd>;
 let Defs = [RAX,RDX,EFLAGS], Uses = [RAX,RDX] in // RDX:RAX/[mem64] = RAX,RDX
 def IDIV64m: RI<0xF7, MRM7m, (outs), (ins i64mem:$src),
-                "idiv{q}\t$src", [], IIC_IDIV64>;
+                "idiv{q}\t$src", [], IIC_IDIV64>,
+             SchedLoadReg<WriteIDivLd>;
 }
+} // hasSideEffects = 0
 
 //===----------------------------------------------------------------------===//
 //  Two address Instructions.
@@ -333,7 +373,7 @@ def IDIV64m: RI<0xF7, MRM7m, (outs), (ins i64mem:$src),
 // unary instructions
 let CodeSize = 2 in {
 let Defs = [EFLAGS] in {
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteALU] in {
 def NEG8r  : I<0xF6, MRM3r, (outs GR8 :$dst), (ins GR8 :$src1),
                "neg{b}\t$dst",
                [(set GR8:$dst, (ineg GR8:$src1)),
@@ -349,8 +389,10 @@ def NEG32r : I<0xF7, MRM3r, (outs GR32:$dst), (ins GR32:$src1),
 def NEG64r : RI<0xF7, MRM3r, (outs GR64:$dst), (ins GR64:$src1), "neg{q}\t$dst",
                 [(set GR64:$dst, (ineg GR64:$src1)),
                  (implicit EFLAGS)], IIC_UNARY_REG>;
-} // Constraints = "$src1 = $dst"
+} // Constraints = "$src1 = $dst", SchedRW
 
+// Read-modify-write negate.
+let SchedRW = [WriteALULd, WriteRMW] in {
 def NEG8m  : I<0xF6, MRM3m, (outs), (ins i8mem :$dst),
                "neg{b}\t$dst",
                [(store (ineg (loadi8 addr:$dst)), addr:$dst),
@@ -366,12 +408,13 @@ def NEG32m : I<0xF7, MRM3m, (outs), (ins i32mem:$dst),
 def NEG64m : RI<0xF7, MRM3m, (outs), (ins i64mem:$dst), "neg{q}\t$dst",
                 [(store (ineg (loadi64 addr:$dst)), addr:$dst),
                  (implicit EFLAGS)], IIC_UNARY_MEM>;
+} // SchedRW
 } // Defs = [EFLAGS]
 
 
 // Note: NOT does not set EFLAGS!
 
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteALU] in {
 // Match xor -1 to not. Favors these over a move imm + xor to save code size.
 let AddedComplexity = 15 in {
 def NOT8r  : I<0xF6, MRM2r, (outs GR8 :$dst), (ins GR8 :$src1),
@@ -386,8 +429,9 @@ def NOT32r : I<0xF7, MRM2r, (outs GR32:$dst), (ins GR32:$src1),
 def NOT64r : RI<0xF7, MRM2r, (outs GR64:$dst), (ins GR64:$src1), "not{q}\t$dst",
                 [(set GR64:$dst, (not GR64:$src1))], IIC_UNARY_REG>;
 }
-} // Constraints = "$src1 = $dst"
+} // Constraints = "$src1 = $dst", SchedRW
 
+let SchedRW = [WriteALULd, WriteRMW] in {
 def NOT8m  : I<0xF6, MRM2m, (outs), (ins i8mem :$dst),
                "not{b}\t$dst",
                [(store (not (loadi8 addr:$dst)), addr:$dst)], IIC_UNARY_MEM>;
@@ -400,11 +444,12 @@ def NOT32m : I<0xF7, MRM2m, (outs), (ins i32mem:$dst),
                [(store (not (loadi32 addr:$dst)), addr:$dst)], IIC_UNARY_MEM>;
 def NOT64m : RI<0xF7, MRM2m, (outs), (ins i64mem:$dst), "not{q}\t$dst",
                 [(store (not (loadi64 addr:$dst)), addr:$dst)], IIC_UNARY_MEM>;
+} // SchedRW
 } // CodeSize
 
 // TODO: inc/dec is slow for P4, but fast for Pentium-M.
 let Defs = [EFLAGS] in {
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteALU] in {
 let CodeSize = 2 in
 def INC8r  : I<0xFE, MRM0r, (outs GR8 :$dst), (ins GR8 :$src1),
                "inc{b}\t$dst",
@@ -412,11 +457,11 @@ def INC8r  : I<0xFE, MRM0r, (outs GR8 :$dst), (ins GR8 :$src1),
                IIC_UNARY_REG>;
 
 let isConvertibleToThreeAddress = 1, CodeSize = 1 in {  // Can xform into LEA.
-def INC16r : I<0x40, AddRegFrm, (outs GR16:$dst), (ins GR16:$src1), 
+def INC16r : I<0x40, AddRegFrm, (outs GR16:$dst), (ins GR16:$src1),
                "inc{w}\t$dst",
                [(set GR16:$dst, EFLAGS, (X86inc_flag GR16:$src1))], IIC_UNARY_REG>,
              OpSize, Requires<[In32BitMode]>;
-def INC32r : I<0x40, AddRegFrm, (outs GR32:$dst), (ins GR32:$src1), 
+def INC32r : I<0x40, AddRegFrm, (outs GR32:$dst), (ins GR32:$src1),
                "inc{l}\t$dst",
                [(set GR32:$dst, EFLAGS, (X86inc_flag GR32:$src1))],
                IIC_UNARY_REG>,
@@ -430,31 +475,31 @@ def INC64r : RI<0xFF, MRM0r, (outs GR64:$dst), (ins GR64:$src1), "inc{q}\t$dst",
 // In 64-bit mode, single byte INC and DEC cannot be encoded.
 let isConvertibleToThreeAddress = 1, CodeSize = 2 in {
 // Can transform into LEA.
-def INC64_16r : I<0xFF, MRM0r, (outs GR16:$dst), (ins GR16:$src1), 
+def INC64_16r : I<0xFF, MRM0r, (outs GR16:$dst), (ins GR16:$src1),
                   "inc{w}\t$dst",
                   [(set GR16:$dst, EFLAGS, (X86inc_flag GR16:$src1))],
                   IIC_UNARY_REG>,
                 OpSize, Requires<[In64BitMode]>;
-def INC64_32r : I<0xFF, MRM0r, (outs GR32:$dst), (ins GR32:$src1), 
+def INC64_32r : I<0xFF, MRM0r, (outs GR32:$dst), (ins GR32:$src1),
                   "inc{l}\t$dst",
                   [(set GR32:$dst, EFLAGS, (X86inc_flag GR32:$src1))],
                   IIC_UNARY_REG>,
                 Requires<[In64BitMode]>;
-def DEC64_16r : I<0xFF, MRM1r, (outs GR16:$dst), (ins GR16:$src1), 
+def DEC64_16r : I<0xFF, MRM1r, (outs GR16:$dst), (ins GR16:$src1),
                   "dec{w}\t$dst",
                   [(set GR16:$dst, EFLAGS, (X86dec_flag GR16:$src1))],
                   IIC_UNARY_REG>,
                 OpSize, Requires<[In64BitMode]>;
-def DEC64_32r : I<0xFF, MRM1r, (outs GR32:$dst), (ins GR32:$src1), 
+def DEC64_32r : I<0xFF, MRM1r, (outs GR32:$dst), (ins GR32:$src1),
                   "dec{l}\t$dst",
                   [(set GR32:$dst, EFLAGS, (X86dec_flag GR32:$src1))],
                   IIC_UNARY_REG>,
                 Requires<[In64BitMode]>;
 } // isConvertibleToThreeAddress = 1, CodeSize = 2
 
-} // Constraints = "$src1 = $dst"
+} // Constraints = "$src1 = $dst", SchedRW
 
-let CodeSize = 2 in {
+let CodeSize = 2, SchedRW = [WriteALULd, WriteRMW] in {
   def INC8m  : I<0xFE, MRM0m, (outs), (ins i8mem :$dst), "inc{b}\t$dst",
                [(store (add (loadi8 addr:$dst), 1), addr:$dst),
                 (implicit EFLAGS)], IIC_UNARY_MEM>;
@@ -469,7 +514,7 @@ let CodeSize = 2 in {
   def INC64m : RI<0xFF, MRM0m, (outs), (ins i64mem:$dst), "inc{q}\t$dst",
                   [(store (add (loadi64 addr:$dst), 1), addr:$dst),
                    (implicit EFLAGS)], IIC_UNARY_MEM>;
-                   
+
 // These are duplicates of their 32-bit counterparts. Only needed so X86 knows
 // how to unfold them.
 // FIXME: What is this for??
@@ -489,21 +534,21 @@ def DEC64_32m : I<0xFF, MRM1m, (outs), (ins i32mem:$dst), "dec{l}\t$dst",
                   [(store (add (loadi32 addr:$dst), -1), addr:$dst),
                     (implicit EFLAGS)], IIC_UNARY_MEM>,
                 Requires<[In64BitMode]>;
-} // CodeSize = 2
+} // CodeSize = 2, SchedRW
 
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteALU] in {
 let CodeSize = 2 in
 def DEC8r  : I<0xFE, MRM1r, (outs GR8 :$dst), (ins GR8 :$src1),
                "dec{b}\t$dst",
                [(set GR8:$dst, EFLAGS, (X86dec_flag GR8:$src1))],
                IIC_UNARY_REG>;
 let isConvertibleToThreeAddress = 1, CodeSize = 1 in {   // Can xform into LEA.
-def DEC16r : I<0x48, AddRegFrm, (outs GR16:$dst), (ins GR16:$src1), 
+def DEC16r : I<0x48, AddRegFrm, (outs GR16:$dst), (ins GR16:$src1),
                "dec{w}\t$dst",
                [(set GR16:$dst, EFLAGS, (X86dec_flag GR16:$src1))],
                IIC_UNARY_REG>,
              OpSize, Requires<[In32BitMode]>;
-def DEC32r : I<0x48, AddRegFrm, (outs GR32:$dst), (ins GR32:$src1), 
+def DEC32r : I<0x48, AddRegFrm, (outs GR32:$dst), (ins GR32:$src1),
                "dec{l}\t$dst",
                [(set GR32:$dst, EFLAGS, (X86dec_flag GR32:$src1))],
                IIC_UNARY_REG>,
@@ -512,10 +557,10 @@ def DEC64r : RI<0xFF, MRM1r, (outs GR64:$dst), (ins GR64:$src1), "dec{q}\t$dst",
                 [(set GR64:$dst, EFLAGS, (X86dec_flag GR64:$src1))],
                 IIC_UNARY_REG>;
 } // CodeSize = 2
-} // Constraints = "$src1 = $dst"
+} // Constraints = "$src1 = $dst", SchedRW
 
 
-let CodeSize = 2 in {
+let CodeSize = 2, SchedRW = [WriteALULd, WriteRMW] in {
   def DEC8m  : I<0xFE, MRM1m, (outs), (ins i8mem :$dst), "dec{b}\t$dst",
                [(store (add (loadi8 addr:$dst), -1), addr:$dst),
                 (implicit EFLAGS)], IIC_UNARY_MEM>;
@@ -530,7 +575,7 @@ let CodeSize = 2 in {
   def DEC64m : RI<0xFF, MRM1m, (outs), (ins i64mem:$dst), "dec{q}\t$dst",
                   [(store (add (loadi64 addr:$dst), -1), addr:$dst),
                    (implicit EFLAGS)], IIC_UNARY_MEM>;
-} // CodeSize = 2
+} // CodeSize = 2, SchedRW
 } // Defs = [EFLAGS]
 
 
@@ -544,57 +589,57 @@ class X86TypeInfo<ValueType vt, string instrsuffix, RegisterClass regclass,
                   bit hasOddOpcode, bit hasOpSizePrefix, bit hasREX_WPrefix> {
   /// VT - This is the value type itself.
   ValueType VT = vt;
-  
+
   /// InstrSuffix - This is the suffix used on instructions with this type.  For
   /// example, i8 -> "b", i16 -> "w", i32 -> "l", i64 -> "q".
   string InstrSuffix = instrsuffix;
-  
+
   /// RegClass - This is the register class associated with this type.  For
   /// example, i8 -> GR8, i16 -> GR16, i32 -> GR32, i64 -> GR64.
   RegisterClass RegClass = regclass;
-  
+
   /// LoadNode - This is the load node associated with this type.  For
   /// example, i8 -> loadi8, i16 -> loadi16, i32 -> loadi32, i64 -> loadi64.
   PatFrag LoadNode = loadnode;
-  
+
   /// MemOperand - This is the memory operand associated with this type.  For
   /// example, i8 -> i8mem, i16 -> i16mem, i32 -> i32mem, i64 -> i64mem.
   X86MemOperand MemOperand = memoperand;
-  
+
   /// ImmEncoding - This is the encoding of an immediate of this type.  For
   /// example, i8 -> Imm8, i16 -> Imm16, i32 -> Imm32.  Note that i64 -> Imm32
   /// since the immediate fields of i64 instructions is a 32-bit sign extended
   /// value.
   ImmType ImmEncoding = immkind;
-  
+
   /// ImmOperand - This is the operand kind of an immediate of this type.  For
   /// example, i8 -> i8imm, i16 -> i16imm, i32 -> i32imm.  Note that i64 ->
   /// i64i32imm since the immediate fields of i64 instructions is a 32-bit sign
   /// extended value.
   Operand ImmOperand = immoperand;
-  
+
   /// ImmOperator - This is the operator that should be used to match an
   /// immediate of this kind in a pattern (e.g. imm, or i64immSExt32).
   SDPatternOperator ImmOperator = immoperator;
-  
+
   /// Imm8Operand - This is the operand kind to use for an imm8 of this type.
   /// For example, i8 -> <invalid>, i16 -> i16i8imm, i32 -> i32i8imm.  This is
   /// only used for instructions that have a sign-extended imm8 field form.
   Operand Imm8Operand = imm8operand;
-  
+
   /// Imm8Operator - This is the operator that should be used to match an 8-bit
   /// sign extended immediate of this kind in a pattern (e.g. imm16immSExt8).
   SDPatternOperator Imm8Operator = imm8operator;
-  
+
   /// HasOddOpcode - This bit is true if the instruction should have an odd (as
   /// opposed to even) opcode.  Operations on i8 are usually even, operations on
   /// other datatypes are odd.
   bit HasOddOpcode = hasOddOpcode;
-  
+
   /// HasOpSizePrefix - This bit is set to true if the instruction should have
   /// the 0x66 operand size prefix.  This is set for i16 types.
   bit HasOpSizePrefix = hasOpSizePrefix;
-  
+
   /// HasREX_WPrefix - This bit is set to true if the instruction should have
   /// the 0x40 REX prefix.  This is set for i64 types.
   bit HasREX_WPrefix = hasREX_WPrefix;
@@ -624,12 +669,12 @@ def Xi64 : X86TypeInfo<i64, "q", GR64, loadi64, i64mem,
 /// 3. Infers whether the instruction should have a 0x40 REX_W prefix.
 /// 4. Infers whether the low bit of the opcode should be 0 (for i8 operations)
 ///    or 1 (for i16,i32,i64 operations).
-class ITy<bits<8> opcode, Format f, X86TypeInfo typeinfo, dag outs, dag ins, 
+class ITy<bits<8> opcode, Format f, X86TypeInfo typeinfo, dag outs, dag ins,
           string mnemonic, string args, list<dag> pattern,
           InstrItinClass itin = IIC_BIN_NONMEM>
   : I<{opcode{7}, opcode{6}, opcode{5}, opcode{4},
        opcode{3}, opcode{2}, opcode{1}, typeinfo.HasOddOpcode },
-      f, outs, ins, 
+      f, outs, ins,
       !strconcat(mnemonic, "{", typeinfo.InstrSuffix, "}\t", args), pattern,
       itin> {
 
@@ -644,7 +689,8 @@ class BinOpRR<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
               Format f = MRMDestReg>
   : ITy<opcode, f, typeinfo, outlist,
         (ins typeinfo.RegClass:$src1, typeinfo.RegClass:$src2),
-        mnemonic, "{$src2, $src1|$src1, $src2}", pattern, itin>;
+        mnemonic, "{$src2, $src1|$src1, $src2}", pattern, itin>,
+    Sched<[WriteALU]>;
 
 // BinOpRR_R - Instructions like "add reg, reg, reg", where the pattern has
 // just a regclass (no eflags) as a result.
@@ -687,18 +733,22 @@ class BinOpRR_Rev<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo>
   : ITy<opcode, MRMSrcReg, typeinfo,
         (outs typeinfo.RegClass:$dst),
         (ins typeinfo.RegClass:$src1, typeinfo.RegClass:$src2),
-        mnemonic, "{$src2, $dst|$dst, $src2}", [], IIC_BIN_NONMEM> {
+        mnemonic, "{$src2, $dst|$dst, $src2}", [], IIC_BIN_NONMEM>,
+    Sched<[WriteALU]> {
   // The disassembler should know about this, but not the asmparser.
   let isCodeGenOnly = 1;
+  let hasSideEffects = 0;
 }
 
 // BinOpRR_F_Rev - Instructions like "cmp reg, reg" (reversed encoding).
 class BinOpRR_F_Rev<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo>
   : ITy<opcode, MRMSrcReg, typeinfo, (outs),
         (ins typeinfo.RegClass:$src1, typeinfo.RegClass:$src2),
-        mnemonic, "{$src2, $src1|$src1, $src2}", [], IIC_BIN_NONMEM> {
+        mnemonic, "{$src2, $src1|$src1, $src2}", [], IIC_BIN_NONMEM>,
+    Sched<[WriteALU]> {
   // The disassembler should know about this, but not the asmparser.
   let isCodeGenOnly = 1;
+  let hasSideEffects = 0;
 }
 
 // BinOpRM - Instructions like "add reg, reg, [mem]".
@@ -706,7 +756,8 @@ class BinOpRM<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
               dag outlist, list<dag> pattern>
   : ITy<opcode, MRMSrcMem, typeinfo, outlist,
         (ins typeinfo.RegClass:$src1, typeinfo.MemOperand:$src2),
-        mnemonic, "{$src2, $src1|$src1, $src2}", pattern, IIC_BIN_NONMEM>;
+        mnemonic, "{$src2, $src1|$src1, $src2}", pattern, IIC_BIN_NONMEM>,
+    Sched<[WriteALULd, ReadAfterLd]>;
 
 // BinOpRM_R - Instructions like "add reg, reg, [mem]".
 class BinOpRM_R<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
@@ -742,7 +793,8 @@ class BinOpRI<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
               Format f, dag outlist, list<dag> pattern>
   : ITy<opcode, f, typeinfo, outlist,
         (ins typeinfo.RegClass:$src1, typeinfo.ImmOperand:$src2),
-        mnemonic, "{$src2, $src1|$src1, $src2}", pattern, IIC_BIN_NONMEM> {
+        mnemonic, "{$src2, $src1|$src1, $src2}", pattern, IIC_BIN_NONMEM>,
+    Sched<[WriteALU]> {
   let ImmT = typeinfo.ImmEncoding;
 }
 
@@ -764,13 +816,13 @@ class BinOpRI_F<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
 class BinOpRI_RF<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
                  SDNode opnode, Format f>
   : BinOpRI<opcode, mnemonic, typeinfo, f, (outs typeinfo.RegClass:$dst),
-            [(set typeinfo.RegClass:$dst, EFLAGS, 
+            [(set typeinfo.RegClass:$dst, EFLAGS,
                 (opnode typeinfo.RegClass:$src1, typeinfo.ImmOperator:$src2))]>;
 // BinOpRI_RFF - Instructions like "adc reg, reg, imm".
 class BinOpRI_RFF<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
                  SDNode opnode, Format f>
   : BinOpRI<opcode, mnemonic, typeinfo, f, (outs typeinfo.RegClass:$dst),
-            [(set typeinfo.RegClass:$dst, EFLAGS, 
+            [(set typeinfo.RegClass:$dst, EFLAGS,
                 (opnode typeinfo.RegClass:$src1, typeinfo.ImmOperator:$src2,
                         EFLAGS))]>;
 
@@ -779,7 +831,8 @@ class BinOpRI8<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
                Format f, dag outlist, list<dag> pattern>
   : ITy<opcode, f, typeinfo, outlist,
         (ins typeinfo.RegClass:$src1, typeinfo.Imm8Operand:$src2),
-        mnemonic, "{$src2, $src1|$src1, $src2}", pattern, IIC_BIN_NONMEM> {
+        mnemonic, "{$src2, $src1|$src1, $src2}", pattern, IIC_BIN_NONMEM>,
+    Sched<[WriteALU]> {
   let ImmT = Imm8; // Always 8-bit immediate.
 }
 
@@ -789,7 +842,7 @@ class BinOpRI8_R<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
   : BinOpRI8<opcode, mnemonic, typeinfo, f, (outs typeinfo.RegClass:$dst),
              [(set typeinfo.RegClass:$dst,
                (opnode typeinfo.RegClass:$src1, typeinfo.Imm8Operator:$src2))]>;
-               
+
 // BinOpRI8_F - Instructions like "cmp reg, imm8".
 class BinOpRI8_F<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
                   SDNode opnode, Format f>
@@ -817,7 +870,8 @@ class BinOpMR<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
               list<dag> pattern>
   : ITy<opcode, MRMDestMem, typeinfo,
         (outs), (ins typeinfo.MemOperand:$dst, typeinfo.RegClass:$src),
-        mnemonic, "{$src, $dst|$dst, $src}", pattern, IIC_BIN_MEM>;
+        mnemonic, "{$src, $dst|$dst, $src}", pattern, IIC_BIN_MEM>,
+    Sched<[WriteALULd, WriteRMW]>;
 
 // BinOpMR_RMW - Instructions like "add [mem], reg".
 class BinOpMR_RMW<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
@@ -845,21 +899,22 @@ class BinOpMI<string mnemonic, X86TypeInfo typeinfo,
               Format f, list<dag> pattern, bits<8> opcode = 0x80>
   : ITy<opcode, f, typeinfo,
         (outs), (ins typeinfo.MemOperand:$dst, typeinfo.ImmOperand:$src),
-        mnemonic, "{$src, $dst|$dst, $src}", pattern, IIC_BIN_MEM> {
+        mnemonic, "{$src, $dst|$dst, $src}", pattern, IIC_BIN_MEM>,
+    Sched<[WriteALULd, WriteRMW]> {
   let ImmT = typeinfo.ImmEncoding;
 }
 
 // BinOpMI_RMW - Instructions like "add [mem], imm".
 class BinOpMI_RMW<string mnemonic, X86TypeInfo typeinfo,
                   SDNode opnode, Format f>
-  : BinOpMI<mnemonic, typeinfo, f, 
+  : BinOpMI<mnemonic, typeinfo, f,
             [(store (opnode (typeinfo.VT (load addr:$dst)),
                             typeinfo.ImmOperator:$src), addr:$dst),
              (implicit EFLAGS)]>;
 // BinOpMI_RMW_FF - Instructions like "adc [mem], imm".
 class BinOpMI_RMW_FF<string mnemonic, X86TypeInfo typeinfo,
                   SDNode opnode, Format f>
-  : BinOpMI<mnemonic, typeinfo, f, 
+  : BinOpMI<mnemonic, typeinfo, f,
             [(store (opnode (typeinfo.VT (load addr:$dst)),
                             typeinfo.ImmOperator:$src, EFLAGS), addr:$dst),
              (implicit EFLAGS)]>;
@@ -867,7 +922,7 @@ class BinOpMI_RMW_FF<string mnemonic, X86TypeInfo typeinfo,
 // BinOpMI_F - Instructions like "cmp [mem], imm".
 class BinOpMI_F<string mnemonic, X86TypeInfo typeinfo,
                 SDPatternOperator opnode, Format f, bits<8> opcode = 0x80>
-  : BinOpMI<mnemonic, typeinfo, f, 
+  : BinOpMI<mnemonic, typeinfo, f,
             [(set EFLAGS, (opnode (typeinfo.VT (load addr:$dst)),
                                                typeinfo.ImmOperator:$src))],
             opcode>;
@@ -877,7 +932,8 @@ class BinOpMI8<string mnemonic, X86TypeInfo typeinfo,
                Format f, list<dag> pattern>
   : ITy<0x82, f, typeinfo,
         (outs), (ins typeinfo.MemOperand:$dst, typeinfo.Imm8Operand:$src),
-        mnemonic, "{$src, $dst|$dst, $src}", pattern, IIC_BIN_MEM> {
+        mnemonic, "{$src, $dst|$dst, $src}", pattern, IIC_BIN_MEM>,
+    Sched<[WriteALULd, WriteRMW]> {
   let ImmT = Imm8; // Always 8-bit immediate.
 }
 
@@ -909,10 +965,11 @@ class BinOpAI<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo,
               Register areg, string operands>
   : ITy<opcode, RawFrm, typeinfo,
         (outs), (ins typeinfo.ImmOperand:$src),
-        mnemonic, operands, []> {
+        mnemonic, operands, []>, Sched<[WriteALU]> {
   let ImmT = typeinfo.ImmEncoding;
   let Uses = [areg];
   let Defs = [areg];
+  let hasSideEffects = 0;
 }
 
 /// ArithBinOp_RF - This is an arithmetic binary operator where the pattern is
@@ -928,61 +985,61 @@ multiclass ArithBinOp_RF<bits<8> BaseOpc, bits<8> BaseOpc2, bits<8> BaseOpc4,
     let Constraints = "$src1 = $dst" in {
       let isCommutable = CommutableRR,
           isConvertibleToThreeAddress = ConvertibleToThreeAddress in {
-        def #NAME#8rr  : BinOpRR_RF<BaseOpc, mnemonic, Xi8 , opnodeflag>;
-        def #NAME#16rr : BinOpRR_RF<BaseOpc, mnemonic, Xi16, opnodeflag>;
-        def #NAME#32rr : BinOpRR_RF<BaseOpc, mnemonic, Xi32, opnodeflag>;
-        def #NAME#64rr : BinOpRR_RF<BaseOpc, mnemonic, Xi64, opnodeflag>;
+        def NAME#8rr  : BinOpRR_RF<BaseOpc, mnemonic, Xi8 , opnodeflag>;
+        def NAME#16rr : BinOpRR_RF<BaseOpc, mnemonic, Xi16, opnodeflag>;
+        def NAME#32rr : BinOpRR_RF<BaseOpc, mnemonic, Xi32, opnodeflag>;
+        def NAME#64rr : BinOpRR_RF<BaseOpc, mnemonic, Xi64, opnodeflag>;
       } // isCommutable
 
-      def #NAME#8rr_REV  : BinOpRR_Rev<BaseOpc2, mnemonic, Xi8>;
-      def #NAME#16rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi16>;
-      def #NAME#32rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi32>;
-      def #NAME#64rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi64>;
+      def NAME#8rr_REV  : BinOpRR_Rev<BaseOpc2, mnemonic, Xi8>;
+      def NAME#16rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi16>;
+      def NAME#32rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi32>;
+      def NAME#64rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi64>;
 
-      def #NAME#8rm   : BinOpRM_RF<BaseOpc2, mnemonic, Xi8 , opnodeflag>;
-      def #NAME#16rm  : BinOpRM_RF<BaseOpc2, mnemonic, Xi16, opnodeflag>;
-      def #NAME#32rm  : BinOpRM_RF<BaseOpc2, mnemonic, Xi32, opnodeflag>;
-      def #NAME#64rm  : BinOpRM_RF<BaseOpc2, mnemonic, Xi64, opnodeflag>;
+      def NAME#8rm   : BinOpRM_RF<BaseOpc2, mnemonic, Xi8 , opnodeflag>;
+      def NAME#16rm  : BinOpRM_RF<BaseOpc2, mnemonic, Xi16, opnodeflag>;
+      def NAME#32rm  : BinOpRM_RF<BaseOpc2, mnemonic, Xi32, opnodeflag>;
+      def NAME#64rm  : BinOpRM_RF<BaseOpc2, mnemonic, Xi64, opnodeflag>;
 
       let isConvertibleToThreeAddress = ConvertibleToThreeAddress in {
         // NOTE: These are order specific, we want the ri8 forms to be listed
         // first so that they are slightly preferred to the ri forms.
-        def #NAME#16ri8 : BinOpRI8_RF<0x82, mnemonic, Xi16, opnodeflag, RegMRM>;
-        def #NAME#32ri8 : BinOpRI8_RF<0x82, mnemonic, Xi32, opnodeflag, RegMRM>;
-        def #NAME#64ri8 : BinOpRI8_RF<0x82, mnemonic, Xi64, opnodeflag, RegMRM>;
-
-        def #NAME#8ri   : BinOpRI_RF<0x80, mnemonic, Xi8 , opnodeflag, RegMRM>;
-        def #NAME#16ri  : BinOpRI_RF<0x80, mnemonic, Xi16, opnodeflag, RegMRM>;
-        def #NAME#32ri  : BinOpRI_RF<0x80, mnemonic, Xi32, opnodeflag, RegMRM>;
-        def #NAME#64ri32: BinOpRI_RF<0x80, mnemonic, Xi64, opnodeflag, RegMRM>;
+        def NAME#16ri8 : BinOpRI8_RF<0x82, mnemonic, Xi16, opnodeflag, RegMRM>;
+        def NAME#32ri8 : BinOpRI8_RF<0x82, mnemonic, Xi32, opnodeflag, RegMRM>;
+        def NAME#64ri8 : BinOpRI8_RF<0x82, mnemonic, Xi64, opnodeflag, RegMRM>;
+
+        def NAME#8ri   : BinOpRI_RF<0x80, mnemonic, Xi8 , opnodeflag, RegMRM>;
+        def NAME#16ri  : BinOpRI_RF<0x80, mnemonic, Xi16, opnodeflag, RegMRM>;
+        def NAME#32ri  : BinOpRI_RF<0x80, mnemonic, Xi32, opnodeflag, RegMRM>;
+        def NAME#64ri32: BinOpRI_RF<0x80, mnemonic, Xi64, opnodeflag, RegMRM>;
       }
     } // Constraints = "$src1 = $dst"
 
-    def #NAME#8mr    : BinOpMR_RMW<BaseOpc, mnemonic, Xi8 , opnode>;
-    def #NAME#16mr   : BinOpMR_RMW<BaseOpc, mnemonic, Xi16, opnode>;
-    def #NAME#32mr   : BinOpMR_RMW<BaseOpc, mnemonic, Xi32, opnode>;
-    def #NAME#64mr   : BinOpMR_RMW<BaseOpc, mnemonic, Xi64, opnode>;
+    def NAME#8mr    : BinOpMR_RMW<BaseOpc, mnemonic, Xi8 , opnode>;
+    def NAME#16mr   : BinOpMR_RMW<BaseOpc, mnemonic, Xi16, opnode>;
+    def NAME#32mr   : BinOpMR_RMW<BaseOpc, mnemonic, Xi32, opnode>;
+    def NAME#64mr   : BinOpMR_RMW<BaseOpc, mnemonic, Xi64, opnode>;
 
     // NOTE: These are order specific, we want the mi8 forms to be listed
     // first so that they are slightly preferred to the mi forms.
-    def #NAME#16mi8  : BinOpMI8_RMW<mnemonic, Xi16, opnode, MemMRM>;
-    def #NAME#32mi8  : BinOpMI8_RMW<mnemonic, Xi32, opnode, MemMRM>;
-    def #NAME#64mi8  : BinOpMI8_RMW<mnemonic, Xi64, opnode, MemMRM>;
-                       
-    def #NAME#8mi    : BinOpMI_RMW<mnemonic, Xi8 , opnode, MemMRM>;
-    def #NAME#16mi   : BinOpMI_RMW<mnemonic, Xi16, opnode, MemMRM>;
-    def #NAME#32mi   : BinOpMI_RMW<mnemonic, Xi32, opnode, MemMRM>;
-    def #NAME#64mi32 : BinOpMI_RMW<mnemonic, Xi64, opnode, MemMRM>;
-
-    def #NAME#8i8   : BinOpAI<BaseOpc4, mnemonic, Xi8 , AL,
-                              "{$src, %al|AL, $src}">;
-    def #NAME#16i16 : BinOpAI<BaseOpc4, mnemonic, Xi16, AX,
-                              "{$src, %ax|AX, $src}">;
-    def #NAME#32i32 : BinOpAI<BaseOpc4, mnemonic, Xi32, EAX,
-                              "{$src, %eax|EAX, $src}">;
-    def #NAME#64i32 : BinOpAI<BaseOpc4, mnemonic, Xi64, RAX,
-                              "{$src, %rax|RAX, $src}">;
-  }                          
+    def NAME#16mi8  : BinOpMI8_RMW<mnemonic, Xi16, opnode, MemMRM>;
+    def NAME#32mi8  : BinOpMI8_RMW<mnemonic, Xi32, opnode, MemMRM>;
+    def NAME#64mi8  : BinOpMI8_RMW<mnemonic, Xi64, opnode, MemMRM>;
+
+    def NAME#8mi    : BinOpMI_RMW<mnemonic, Xi8 , opnode, MemMRM>;
+    def NAME#16mi   : BinOpMI_RMW<mnemonic, Xi16, opnode, MemMRM>;
+    def NAME#32mi   : BinOpMI_RMW<mnemonic, Xi32, opnode, MemMRM>;
+    def NAME#64mi32 : BinOpMI_RMW<mnemonic, Xi64, opnode, MemMRM>;
+
+    def NAME#8i8   : BinOpAI<BaseOpc4, mnemonic, Xi8 , AL,
+                             "{$src, %al|AL, $src}">;
+    def NAME#16i16 : BinOpAI<BaseOpc4, mnemonic, Xi16, AX,
+                             "{$src, %ax|AX, $src}">;
+    def NAME#32i32 : BinOpAI<BaseOpc4, mnemonic, Xi32, EAX,
+                             "{$src, %eax|EAX, $src}">;
+    def NAME#64i32 : BinOpAI<BaseOpc4, mnemonic, Xi64, RAX,
+                             "{$src, %rax|RAX, $src}">;
+  }
 }
 
 /// ArithBinOp_RFF - This is an arithmetic binary operator where the pattern is
@@ -999,61 +1056,61 @@ multiclass ArithBinOp_RFF<bits<8> BaseOpc, bits<8> BaseOpc2, bits<8> BaseOpc4,
     let Constraints = "$src1 = $dst" in {
       let isCommutable = CommutableRR,
           isConvertibleToThreeAddress = ConvertibleToThreeAddress in {
-        def #NAME#8rr  : BinOpRR_RFF<BaseOpc, mnemonic, Xi8 , opnode>;
-        def #NAME#16rr : BinOpRR_RFF<BaseOpc, mnemonic, Xi16, opnode>;
-        def #NAME#32rr : BinOpRR_RFF<BaseOpc, mnemonic, Xi32, opnode>;
-        def #NAME#64rr : BinOpRR_RFF<BaseOpc, mnemonic, Xi64, opnode>;
+        def NAME#8rr  : BinOpRR_RFF<BaseOpc, mnemonic, Xi8 , opnode>;
+        def NAME#16rr : BinOpRR_RFF<BaseOpc, mnemonic, Xi16, opnode>;
+        def NAME#32rr : BinOpRR_RFF<BaseOpc, mnemonic, Xi32, opnode>;
+        def NAME#64rr : BinOpRR_RFF<BaseOpc, mnemonic, Xi64, opnode>;
       } // isCommutable
 
-      def #NAME#8rr_REV  : BinOpRR_Rev<BaseOpc2, mnemonic, Xi8>;
-      def #NAME#16rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi16>;
-      def #NAME#32rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi32>;
-      def #NAME#64rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi64>;
+      def NAME#8rr_REV  : BinOpRR_Rev<BaseOpc2, mnemonic, Xi8>;
+      def NAME#16rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi16>;
+      def NAME#32rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi32>;
+      def NAME#64rr_REV : BinOpRR_Rev<BaseOpc2, mnemonic, Xi64>;
 
-      def #NAME#8rm   : BinOpRM_RFF<BaseOpc2, mnemonic, Xi8 , opnode>;
-      def #NAME#16rm  : BinOpRM_RFF<BaseOpc2, mnemonic, Xi16, opnode>;
-      def #NAME#32rm  : BinOpRM_RFF<BaseOpc2, mnemonic, Xi32, opnode>;
-      def #NAME#64rm  : BinOpRM_RFF<BaseOpc2, mnemonic, Xi64, opnode>;
+      def NAME#8rm   : BinOpRM_RFF<BaseOpc2, mnemonic, Xi8 , opnode>;
+      def NAME#16rm  : BinOpRM_RFF<BaseOpc2, mnemonic, Xi16, opnode>;
+      def NAME#32rm  : BinOpRM_RFF<BaseOpc2, mnemonic, Xi32, opnode>;
+      def NAME#64rm  : BinOpRM_RFF<BaseOpc2, mnemonic, Xi64, opnode>;
 
       let isConvertibleToThreeAddress = ConvertibleToThreeAddress in {
         // NOTE: These are order specific, we want the ri8 forms to be listed
         // first so that they are slightly preferred to the ri forms.
-        def #NAME#16ri8 : BinOpRI8_RFF<0x82, mnemonic, Xi16, opnode, RegMRM>;
-        def #NAME#32ri8 : BinOpRI8_RFF<0x82, mnemonic, Xi32, opnode, RegMRM>;
-        def #NAME#64ri8 : BinOpRI8_RFF<0x82, mnemonic, Xi64, opnode, RegMRM>;
-
-        def #NAME#8ri   : BinOpRI_RFF<0x80, mnemonic, Xi8 , opnode, RegMRM>;
-        def #NAME#16ri  : BinOpRI_RFF<0x80, mnemonic, Xi16, opnode, RegMRM>;
-        def #NAME#32ri  : BinOpRI_RFF<0x80, mnemonic, Xi32, opnode, RegMRM>;
-        def #NAME#64ri32: BinOpRI_RFF<0x80, mnemonic, Xi64, opnode, RegMRM>;
+        def NAME#16ri8 : BinOpRI8_RFF<0x82, mnemonic, Xi16, opnode, RegMRM>;
+        def NAME#32ri8 : BinOpRI8_RFF<0x82, mnemonic, Xi32, opnode, RegMRM>;
+        def NAME#64ri8 : BinOpRI8_RFF<0x82, mnemonic, Xi64, opnode, RegMRM>;
+
+        def NAME#8ri   : BinOpRI_RFF<0x80, mnemonic, Xi8 , opnode, RegMRM>;
+        def NAME#16ri  : BinOpRI_RFF<0x80, mnemonic, Xi16, opnode, RegMRM>;
+        def NAME#32ri  : BinOpRI_RFF<0x80, mnemonic, Xi32, opnode, RegMRM>;
+        def NAME#64ri32: BinOpRI_RFF<0x80, mnemonic, Xi64, opnode, RegMRM>;
       }
     } // Constraints = "$src1 = $dst"
 
-    def #NAME#8mr    : BinOpMR_RMW_FF<BaseOpc, mnemonic, Xi8 , opnode>;
-    def #NAME#16mr   : BinOpMR_RMW_FF<BaseOpc, mnemonic, Xi16, opnode>;
-    def #NAME#32mr   : BinOpMR_RMW_FF<BaseOpc, mnemonic, Xi32, opnode>;
-    def #NAME#64mr   : BinOpMR_RMW_FF<BaseOpc, mnemonic, Xi64, opnode>;
+    def NAME#8mr    : BinOpMR_RMW_FF<BaseOpc, mnemonic, Xi8 , opnode>;
+    def NAME#16mr   : BinOpMR_RMW_FF<BaseOpc, mnemonic, Xi16, opnode>;
+    def NAME#32mr   : BinOpMR_RMW_FF<BaseOpc, mnemonic, Xi32, opnode>;
+    def NAME#64mr   : BinOpMR_RMW_FF<BaseOpc, mnemonic, Xi64, opnode>;
 
     // NOTE: These are order specific, we want the mi8 forms to be listed
     // first so that they are slightly preferred to the mi forms.
-    def #NAME#16mi8  : BinOpMI8_RMW_FF<mnemonic, Xi16, opnode, MemMRM>;
-    def #NAME#32mi8  : BinOpMI8_RMW_FF<mnemonic, Xi32, opnode, MemMRM>;
-    def #NAME#64mi8  : BinOpMI8_RMW_FF<mnemonic, Xi64, opnode, MemMRM>;
-                       
-    def #NAME#8mi    : BinOpMI_RMW_FF<mnemonic, Xi8 , opnode, MemMRM>;
-    def #NAME#16mi   : BinOpMI_RMW_FF<mnemonic, Xi16, opnode, MemMRM>;
-    def #NAME#32mi   : BinOpMI_RMW_FF<mnemonic, Xi32, opnode, MemMRM>;
-    def #NAME#64mi32 : BinOpMI_RMW_FF<mnemonic, Xi64, opnode, MemMRM>;
-
-    def #NAME#8i8   : BinOpAI<BaseOpc4, mnemonic, Xi8 , AL,
-                              "{$src, %al|AL, $src}">;
-    def #NAME#16i16 : BinOpAI<BaseOpc4, mnemonic, Xi16, AX,
-                              "{$src, %ax|AX, $src}">;
-    def #NAME#32i32 : BinOpAI<BaseOpc4, mnemonic, Xi32, EAX,
-                              "{$src, %eax|EAX, $src}">;
-    def #NAME#64i32 : BinOpAI<BaseOpc4, mnemonic, Xi64, RAX, 
-                              "{$src, %rax|RAX, $src}">;
-  }                          
+    def NAME#16mi8  : BinOpMI8_RMW_FF<mnemonic, Xi16, opnode, MemMRM>;
+    def NAME#32mi8  : BinOpMI8_RMW_FF<mnemonic, Xi32, opnode, MemMRM>;
+    def NAME#64mi8  : BinOpMI8_RMW_FF<mnemonic, Xi64, opnode, MemMRM>;
+
+    def NAME#8mi    : BinOpMI_RMW_FF<mnemonic, Xi8 , opnode, MemMRM>;
+    def NAME#16mi   : BinOpMI_RMW_FF<mnemonic, Xi16, opnode, MemMRM>;
+    def NAME#32mi   : BinOpMI_RMW_FF<mnemonic, Xi32, opnode, MemMRM>;
+    def NAME#64mi32 : BinOpMI_RMW_FF<mnemonic, Xi64, opnode, MemMRM>;
+
+    def NAME#8i8   : BinOpAI<BaseOpc4, mnemonic, Xi8 , AL,
+                             "{$src, %al|AL, $src}">;
+    def NAME#16i16 : BinOpAI<BaseOpc4, mnemonic, Xi16, AX,
+                             "{$src, %ax|AX, $src}">;
+    def NAME#32i32 : BinOpAI<BaseOpc4, mnemonic, Xi32, EAX,
+                             "{$src, %eax|EAX, $src}">;
+    def NAME#64i32 : BinOpAI<BaseOpc4, mnemonic, Xi64, RAX,
+                             "{$src, %rax|RAX, $src}">;
+  }
 }
 
 /// ArithBinOp_F - This is an arithmetic binary operator where the pattern is
@@ -1067,60 +1124,60 @@ multiclass ArithBinOp_F<bits<8> BaseOpc, bits<8> BaseOpc2, bits<8> BaseOpc4,
   let Defs = [EFLAGS] in {
     let isCommutable = CommutableRR,
         isConvertibleToThreeAddress = ConvertibleToThreeAddress in {
-      def #NAME#8rr  : BinOpRR_F<BaseOpc, mnemonic, Xi8 , opnode>;
-      def #NAME#16rr : BinOpRR_F<BaseOpc, mnemonic, Xi16, opnode>;
-      def #NAME#32rr : BinOpRR_F<BaseOpc, mnemonic, Xi32, opnode>;
-      def #NAME#64rr : BinOpRR_F<BaseOpc, mnemonic, Xi64, opnode>;
+      def NAME#8rr  : BinOpRR_F<BaseOpc, mnemonic, Xi8 , opnode>;
+      def NAME#16rr : BinOpRR_F<BaseOpc, mnemonic, Xi16, opnode>;
+      def NAME#32rr : BinOpRR_F<BaseOpc, mnemonic, Xi32, opnode>;
+      def NAME#64rr : BinOpRR_F<BaseOpc, mnemonic, Xi64, opnode>;
     } // isCommutable
 
-    def #NAME#8rr_REV  : BinOpRR_F_Rev<BaseOpc2, mnemonic, Xi8>;
-    def #NAME#16rr_REV : BinOpRR_F_Rev<BaseOpc2, mnemonic, Xi16>;
-    def #NAME#32rr_REV : BinOpRR_F_Rev<BaseOpc2, mnemonic, Xi32>;
-    def #NAME#64rr_REV : BinOpRR_F_Rev<BaseOpc2, mnemonic, Xi64>;
+    def NAME#8rr_REV  : BinOpRR_F_Rev<BaseOpc2, mnemonic, Xi8>;
+    def NAME#16rr_REV : BinOpRR_F_Rev<BaseOpc2, mnemonic, Xi16>;
+    def NAME#32rr_REV : BinOpRR_F_Rev<BaseOpc2, mnemonic, Xi32>;
+    def NAME#64rr_REV : BinOpRR_F_Rev<BaseOpc2, mnemonic, Xi64>;
 
-    def #NAME#8rm   : BinOpRM_F<BaseOpc2, mnemonic, Xi8 , opnode>;
-    def #NAME#16rm  : BinOpRM_F<BaseOpc2, mnemonic, Xi16, opnode>;
-    def #NAME#32rm  : BinOpRM_F<BaseOpc2, mnemonic, Xi32, opnode>;
-    def #NAME#64rm  : BinOpRM_F<BaseOpc2, mnemonic, Xi64, opnode>;
+    def NAME#8rm   : BinOpRM_F<BaseOpc2, mnemonic, Xi8 , opnode>;
+    def NAME#16rm  : BinOpRM_F<BaseOpc2, mnemonic, Xi16, opnode>;
+    def NAME#32rm  : BinOpRM_F<BaseOpc2, mnemonic, Xi32, opnode>;
+    def NAME#64rm  : BinOpRM_F<BaseOpc2, mnemonic, Xi64, opnode>;
 
     let isConvertibleToThreeAddress = ConvertibleToThreeAddress in {
       // NOTE: These are order specific, we want the ri8 forms to be listed
       // first so that they are slightly preferred to the ri forms.
-      def #NAME#16ri8 : BinOpRI8_F<0x82, mnemonic, Xi16, opnode, RegMRM>;
-      def #NAME#32ri8 : BinOpRI8_F<0x82, mnemonic, Xi32, opnode, RegMRM>;
-      def #NAME#64ri8 : BinOpRI8_F<0x82, mnemonic, Xi64, opnode, RegMRM>;
-      
-      def #NAME#8ri   : BinOpRI_F<0x80, mnemonic, Xi8 , opnode, RegMRM>;
-      def #NAME#16ri  : BinOpRI_F<0x80, mnemonic, Xi16, opnode, RegMRM>;
-      def #NAME#32ri  : BinOpRI_F<0x80, mnemonic, Xi32, opnode, RegMRM>;
-      def #NAME#64ri32: BinOpRI_F<0x80, mnemonic, Xi64, opnode, RegMRM>;
+      def NAME#16ri8 : BinOpRI8_F<0x82, mnemonic, Xi16, opnode, RegMRM>;
+      def NAME#32ri8 : BinOpRI8_F<0x82, mnemonic, Xi32, opnode, RegMRM>;
+      def NAME#64ri8 : BinOpRI8_F<0x82, mnemonic, Xi64, opnode, RegMRM>;
+
+      def NAME#8ri   : BinOpRI_F<0x80, mnemonic, Xi8 , opnode, RegMRM>;
+      def NAME#16ri  : BinOpRI_F<0x80, mnemonic, Xi16, opnode, RegMRM>;
+      def NAME#32ri  : BinOpRI_F<0x80, mnemonic, Xi32, opnode, RegMRM>;
+      def NAME#64ri32: BinOpRI_F<0x80, mnemonic, Xi64, opnode, RegMRM>;
     }
 
-    def #NAME#8mr    : BinOpMR_F<BaseOpc, mnemonic, Xi8 , opnode>;
-    def #NAME#16mr   : BinOpMR_F<BaseOpc, mnemonic, Xi16, opnode>;
-    def #NAME#32mr   : BinOpMR_F<BaseOpc, mnemonic, Xi32, opnode>;
-    def #NAME#64mr   : BinOpMR_F<BaseOpc, mnemonic, Xi64, opnode>;
+    def NAME#8mr    : BinOpMR_F<BaseOpc, mnemonic, Xi8 , opnode>;
+    def NAME#16mr   : BinOpMR_F<BaseOpc, mnemonic, Xi16, opnode>;
+    def NAME#32mr   : BinOpMR_F<BaseOpc, mnemonic, Xi32, opnode>;
+    def NAME#64mr   : BinOpMR_F<BaseOpc, mnemonic, Xi64, opnode>;
 
     // NOTE: These are order specific, we want the mi8 forms to be listed
     // first so that they are slightly preferred to the mi forms.
-    def #NAME#16mi8  : BinOpMI8_F<mnemonic, Xi16, opnode, MemMRM>;
-    def #NAME#32mi8  : BinOpMI8_F<mnemonic, Xi32, opnode, MemMRM>;
-    def #NAME#64mi8  : BinOpMI8_F<mnemonic, Xi64, opnode, MemMRM>;
-                       
-    def #NAME#8mi    : BinOpMI_F<mnemonic, Xi8 , opnode, MemMRM>;
-    def #NAME#16mi   : BinOpMI_F<mnemonic, Xi16, opnode, MemMRM>;
-    def #NAME#32mi   : BinOpMI_F<mnemonic, Xi32, opnode, MemMRM>;
-    def #NAME#64mi32 : BinOpMI_F<mnemonic, Xi64, opnode, MemMRM>;
-
-    def #NAME#8i8   : BinOpAI<BaseOpc4, mnemonic, Xi8 , AL,
-                              "{$src, %al|AL, $src}">;
-    def #NAME#16i16 : BinOpAI<BaseOpc4, mnemonic, Xi16, AX,
-                              "{$src, %ax|AX, $src}">;
-    def #NAME#32i32 : BinOpAI<BaseOpc4, mnemonic, Xi32, EAX,
-                              "{$src, %eax|EAX, $src}">;
-    def #NAME#64i32 : BinOpAI<BaseOpc4, mnemonic, Xi64, RAX,
-                              "{$src, %rax|RAX, $src}">;
-  }                          
+    def NAME#16mi8  : BinOpMI8_F<mnemonic, Xi16, opnode, MemMRM>;
+    def NAME#32mi8  : BinOpMI8_F<mnemonic, Xi32, opnode, MemMRM>;
+    def NAME#64mi8  : BinOpMI8_F<mnemonic, Xi64, opnode, MemMRM>;
+
+    def NAME#8mi    : BinOpMI_F<mnemonic, Xi8 , opnode, MemMRM>;
+    def NAME#16mi   : BinOpMI_F<mnemonic, Xi16, opnode, MemMRM>;
+    def NAME#32mi   : BinOpMI_F<mnemonic, Xi32, opnode, MemMRM>;
+    def NAME#64mi32 : BinOpMI_F<mnemonic, Xi64, opnode, MemMRM>;
+
+    def NAME#8i8   : BinOpAI<BaseOpc4, mnemonic, Xi8 , AL,
+                             "{$src, %al|AL, $src}">;
+    def NAME#16i16 : BinOpAI<BaseOpc4, mnemonic, Xi16, AX,
+                             "{$src, %ax|AX, $src}">;
+    def NAME#32i32 : BinOpAI<BaseOpc4, mnemonic, Xi32, EAX,
+                             "{$src, %eax|EAX, $src}">;
+    def NAME#64i32 : BinOpAI<BaseOpc4, mnemonic, Xi64, RAX,
+                             "{$src, %rax|RAX, $src}">;
+  }
 }
 
 
@@ -1180,7 +1237,7 @@ let isCompare = 1, Defs = [EFLAGS] in {
   def TEST16mi   : BinOpMI_F<"test", Xi16, X86testpat, MRM0m, 0xF6>;
   def TEST32mi   : BinOpMI_F<"test", Xi32, X86testpat, MRM0m, 0xF6>;
   def TEST64mi32 : BinOpMI_F<"test", Xi64, X86testpat, MRM0m, 0xF6>;
-                     
+
   def TEST8i8    : BinOpAI<0xA8, "test", Xi8 , AL,
                            "{$src, %al|AL, $src}">;
   def TEST16i16  : BinOpAI<0xA8, "test", Xi16, AX,
@@ -1194,7 +1251,7 @@ let isCompare = 1, Defs = [EFLAGS] in {
   // register class is constrained to GR8_NOREX.
   let isPseudo = 1 in
   def TEST8ri_NOREX : I<0, Pseudo, (outs), (ins GR8_NOREX:$src, i8imm:$mask),
-                        "", [], IIC_BIN_NONMEM>;
+                        "", [], IIC_BIN_NONMEM>, Sched<[WriteALU]>;
 }
 
 //===----------------------------------------------------------------------===//
@@ -1204,12 +1261,13 @@ multiclass bmi_andn<string mnemonic, RegisterClass RC, X86MemOperand x86memop,
                     PatFrag ld_frag> {
   def rr : I<0xF2, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
             !strconcat(mnemonic, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-            [(set RC:$dst, EFLAGS, (X86andn_flag RC:$src1, RC:$src2))],
-            IIC_BIN_NONMEM>;
+            [(set RC:$dst, EFLAGS, (X86and_flag (not RC:$src1), RC:$src2))],
+            IIC_BIN_NONMEM>, Sched<[WriteALU]>;
   def rm : I<0xF2, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
             !strconcat(mnemonic, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
             [(set RC:$dst, EFLAGS,
-             (X86andn_flag RC:$src1, (ld_frag addr:$src2)))], IIC_BIN_MEM>;
+             (X86and_flag (not RC:$src1), (ld_frag addr:$src2)))], IIC_BIN_MEM>,
+           Sched<[WriteALULd, ReadAfterLd]>;
 }
 
 let Predicates = [HasBMI], Defs = [EFLAGS] in {
@@ -1217,6 +1275,17 @@ let Predicates = [HasBMI], Defs = [EFLAGS] in {
   defm ANDN64 : bmi_andn<"andn{q}", GR64, i64mem, loadi64>, T8, VEX_4V, VEX_W;
 }
 
+let Predicates = [HasBMI] in {
+  def : Pat<(and (not GR32:$src1), GR32:$src2),
+            (ANDN32rr GR32:$src1, GR32:$src2)>;
+  def : Pat<(and (not GR64:$src1), GR64:$src2),
+            (ANDN64rr GR64:$src1, GR64:$src2)>;
+  def : Pat<(and (not GR32:$src1), (loadi32 addr:$src2)),
+            (ANDN32rm GR32:$src1, addr:$src2)>;
+  def : Pat<(and (not GR64:$src1), (loadi64 addr:$src2)),
+            (ANDN64rm GR64:$src1, addr:$src2)>;
+}
+
 //===----------------------------------------------------------------------===//
 // MULX Instruction
 //
@@ -1225,12 +1294,12 @@ let neverHasSideEffects = 1 in {
   let isCommutable = 1 in
   def rr : I<0xF6, MRMSrcReg, (outs RC:$dst1, RC:$dst2), (ins RC:$src),
              !strconcat(mnemonic, "\t{$src, $dst2, $dst1|$dst1, $dst2, $src}"),
-             [], IIC_MUL8>, T8XD, VEX_4V;
+             [], IIC_MUL8>, T8XD, VEX_4V, Sched<[WriteIMul]>;
 
   let mayLoad = 1 in
   def rm : I<0xF6, MRMSrcMem, (outs RC:$dst1, RC:$dst2), (ins x86memop:$src),
              !strconcat(mnemonic, "\t{$src, $dst2, $dst1|$dst1, $dst2, $src}"),
-             [], IIC_MUL8>, T8XD, VEX_4V;
+             [], IIC_MUL8>, T8XD, VEX_4V, Sched<[WriteIMulLd]>;
 }
 }
 
@@ -1240,3 +1309,53 @@ let Predicates = [HasBMI2] in {
   let Uses = [RDX] in
     defm MULX64 : bmi_mulx<"mulx{q}", GR64, i64mem>, VEX_W;
 }
+
+//===----------------------------------------------------------------------===//
+// ADCX Instruction
+//
+let hasSideEffects = 0, Predicates = [HasADX], Defs = [EFLAGS] in {
+  let SchedRW = [WriteALU] in {
+  def ADCX32rr : I<0xF6, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
+             "adcx{l}\t{$src, $dst|$dst, $src}",
+             [], IIC_BIN_NONMEM>, T8, OpSize;
+
+  def ADCX64rr : I<0xF6, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
+             "adcx{q}\t{$src, $dst|$dst, $src}",
+             [], IIC_BIN_NONMEM>, T8, OpSize, REX_W, Requires<[In64BitMode]>;
+  } // SchedRW
+
+  let mayLoad = 1, SchedRW = [WriteALULd] in {
+  def ADCX32rm : I<0xF6, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
+             "adcx{l}\t{$src, $dst|$dst, $src}",
+             [], IIC_BIN_MEM>, T8, OpSize;
+ 
+  def ADCX64rm : I<0xF6, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
+             "adcx{q}\t{$src, $dst|$dst, $src}",
+             [], IIC_BIN_MEM>, T8, OpSize, REX_W, Requires<[In64BitMode]>;
+  }
+}
+
+//===----------------------------------------------------------------------===//
+// ADOX Instruction
+//
+let hasSideEffects = 0, Predicates = [HasADX], Defs = [EFLAGS] in {
+  let SchedRW = [WriteALU] in {
+  def ADOX32rr : I<0xF6, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
+             "adox{l}\t{$src, $dst|$dst, $src}",
+             [], IIC_BIN_NONMEM>, T8XS;
+
+  def ADOX64rr : I<0xF6, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
+             "adox{q}\t{$src, $dst|$dst, $src}",
+             [], IIC_BIN_NONMEM>, T8XS, REX_W, Requires<[In64BitMode]>;
+  } // SchedRW
+
+  let mayLoad = 1, SchedRW = [WriteALULd] in {
+  def ADOX32rm : I<0xF6, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
+             "adox{l}\t{$src, $dst|$dst, $src}",
+             [], IIC_BIN_MEM>, T8XS;
+ 
+  def ADOX64rm : I<0xF6, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
+             "adox{q}\t{$src, $dst|$dst, $src}",
+             [], IIC_BIN_MEM>, T8XS, REX_W, Requires<[In64BitMode]>;
+  }
+}
diff --git a/lib/Target/X86/X86InstrCMovSetCC.td b/lib/Target/X86/X86InstrCMovSetCC.td
index adeaf5410dcc..a967a4da5cf7 100644
--- a/lib/Target/X86/X86InstrCMovSetCC.td
+++ b/lib/Target/X86/X86InstrCMovSetCC.td
@@ -16,20 +16,20 @@
 // SetCC instructions.
 multiclass CMOV<bits<8> opc, string Mnemonic, PatLeaf CondNode> {
   let Uses = [EFLAGS], Predicates = [HasCMov], Constraints = "$src1 = $dst",
-      isCommutable = 1 in {
-    def #NAME#16rr
+      isCommutable = 1, SchedRW = [WriteALU] in {
+    def NAME#16rr
       : I<opc, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src1, GR16:$src2),
           !strconcat(Mnemonic, "{w}\t{$src2, $dst|$dst, $src2}"),
           [(set GR16:$dst,
                 (X86cmov GR16:$src1, GR16:$src2, CondNode, EFLAGS))],
                 IIC_CMOV16_RR>,TB,OpSize;
-    def #NAME#32rr
+    def NAME#32rr
       : I<opc, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src1, GR32:$src2),
           !strconcat(Mnemonic, "{l}\t{$src2, $dst|$dst, $src2}"),
           [(set GR32:$dst,
                 (X86cmov GR32:$src1, GR32:$src2, CondNode, EFLAGS))],
                 IIC_CMOV32_RR>, TB;
-    def #NAME#64rr
+    def NAME#64rr
       :RI<opc, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
           !strconcat(Mnemonic, "{q}\t{$src2, $dst|$dst, $src2}"),
           [(set GR64:$dst,
@@ -37,19 +37,20 @@ multiclass CMOV<bits<8> opc, string Mnemonic, PatLeaf CondNode> {
                 IIC_CMOV32_RR>, TB;
   }
 
-  let Uses = [EFLAGS], Predicates = [HasCMov], Constraints = "$src1 = $dst" in {
-    def #NAME#16rm
+  let Uses = [EFLAGS], Predicates = [HasCMov], Constraints = "$src1 = $dst",
+      SchedRW = [WriteALULd, ReadAfterLd] in {
+    def NAME#16rm
       : I<opc, MRMSrcMem, (outs GR16:$dst), (ins GR16:$src1, i16mem:$src2),
           !strconcat(Mnemonic, "{w}\t{$src2, $dst|$dst, $src2}"),
           [(set GR16:$dst, (X86cmov GR16:$src1, (loadi16 addr:$src2),
                                     CondNode, EFLAGS))], IIC_CMOV16_RM>,
                                     TB, OpSize;
-    def #NAME#32rm
+    def NAME#32rm
       : I<opc, MRMSrcMem, (outs GR32:$dst), (ins GR32:$src1, i32mem:$src2),
           !strconcat(Mnemonic, "{l}\t{$src2, $dst|$dst, $src2}"),
           [(set GR32:$dst, (X86cmov GR32:$src1, (loadi32 addr:$src2),
                                     CondNode, EFLAGS))], IIC_CMOV32_RM>, TB;
-    def #NAME#64rm
+    def NAME#64rm
       :RI<opc, MRMSrcMem, (outs GR64:$dst), (ins GR64:$src1, i64mem:$src2),
           !strconcat(Mnemonic, "{q}\t{$src2, $dst|$dst, $src2}"),
           [(set GR64:$dst, (X86cmov GR64:$src1, (loadi64 addr:$src2),
@@ -83,11 +84,11 @@ multiclass SETCC<bits<8> opc, string Mnemonic, PatLeaf OpNode> {
     def r    : I<opc, MRM0r,  (outs GR8:$dst), (ins),
                      !strconcat(Mnemonic, "\t$dst"),
                      [(set GR8:$dst, (X86setcc OpNode, EFLAGS))],
-                     IIC_SET_R>, TB;
+                     IIC_SET_R>, TB, Sched<[WriteALU]>;
     def m    : I<opc, MRM0m,  (outs), (ins i8mem:$dst),
                      !strconcat(Mnemonic, "\t$dst"),
                      [(store (X86setcc OpNode, EFLAGS), addr:$dst)],
-                     IIC_SET_M>, TB;
+                     IIC_SET_M>, TB, Sched<[WriteALU, WriteStore]>;
   } // Uses = [EFLAGS]
 }
 
diff --git a/lib/Target/X86/X86InstrCompiler.td b/lib/Target/X86/X86InstrCompiler.td
index 9e6f27988f71..d9ff0c63c55f 100644
--- a/lib/Target/X86/X86InstrCompiler.td
+++ b/lib/Target/X86/X86InstrCompiler.td
@@ -149,11 +149,12 @@ let Defs = [EAX, EDX, EFLAGS], FPForm = SpecialFP in {
 //===----------------------------------------------------------------------===//
 // EH Pseudo Instructions
 //
+let SchedRW = [WriteSystem] in {
 let isTerminator = 1, isReturn = 1, isBarrier = 1,
     hasCtrlDep = 1, isCodeGenOnly = 1 in {
 def EH_RETURN   : I<0xC3, RawFrm, (outs), (ins GR32:$addr),
                     "ret\t#eh_return, addr: $addr",
-                    [(X86ehret GR32:$addr)], IIC_RET>;
+                    [(X86ehret GR32:$addr)], IIC_RET>, Sched<[WriteJumpLd]>;
 
 }
 
@@ -161,7 +162,7 @@ let isTerminator = 1, isReturn = 1, isBarrier = 1,
     hasCtrlDep = 1, isCodeGenOnly = 1 in {
 def EH_RETURN64   : I<0xC3, RawFrm, (outs), (ins GR64:$addr),
                      "ret\t#eh_return, addr: $addr",
-                     [(X86ehret GR64:$addr)], IIC_RET>;
+                     [(X86ehret GR64:$addr)], IIC_RET>, Sched<[WriteJumpLd]>;
 
 }
 
@@ -186,6 +187,7 @@ let hasSideEffects = 1, isBarrier = 1, isCodeGenOnly = 1,
                           Requires<[In64BitMode]>;
   }
 }
+} // SchedRW
 
 let isBranch = 1, isTerminator = 1, isCodeGenOnly = 1 in {
   def EH_SjLj_Setup : I<0, Pseudo, (outs), (ins brtarget:$dst),
@@ -220,7 +222,7 @@ def MORESTACK_RET_RESTORE_R10 : I<0, Pseudo, (outs), (ins),
 let Defs = [EFLAGS], isReMaterializable = 1, isAsCheapAsAMove = 1,
     isCodeGenOnly = 1 in {
 def MOV8r0   : I<0x30, MRMInitReg, (outs GR8 :$dst), (ins), "",
-                 [(set GR8:$dst, 0)], IIC_ALU_NONMEM>;
+                 [(set GR8:$dst, 0)], IIC_ALU_NONMEM>, Sched<[WriteZero]>;
 
 // We want to rewrite MOV16r0 in terms of MOV32r0, because it's a smaller
 // encoding and avoids a partial-register update sometimes, but doing so
@@ -229,11 +231,12 @@ def MOV8r0   : I<0x30, MRMInitReg, (outs GR8 :$dst), (ins), "",
 // to an MCInst.
 def MOV16r0   : I<0x31, MRMInitReg, (outs GR16:$dst), (ins),
                  "",
-                 [(set GR16:$dst, 0)], IIC_ALU_NONMEM>, OpSize;
+                 [(set GR16:$dst, 0)], IIC_ALU_NONMEM>, OpSize,
+                 Sched<[WriteZero]>;
 
 // FIXME: Set encoding to pseudo.
 def MOV32r0  : I<0x31, MRMInitReg, (outs GR32:$dst), (ins), "",
-                 [(set GR32:$dst, 0)], IIC_ALU_NONMEM>;
+                 [(set GR32:$dst, 0)], IIC_ALU_NONMEM>, Sched<[WriteZero]>;
 }
 
 // We want to rewrite MOV64r0 in terms of MOV32r0, because it's sometimes a
@@ -245,7 +248,7 @@ def MOV32r0  : I<0x31, MRMInitReg, (outs GR32:$dst), (ins), "",
 let Defs = [EFLAGS], isCodeGenOnly=1,
     AddedComplexity = 1, isReMaterializable = 1, isAsCheapAsAMove = 1 in
 def MOV64r0   : I<0x31, MRMInitReg, (outs GR64:$dst), (ins), "",
-                 [(set GR64:$dst, 0)], IIC_ALU_NONMEM>;
+                 [(set GR64:$dst, 0)], IIC_ALU_NONMEM>, Sched<[WriteZero]>;
 
 // Materialize i64 constant where top 32-bits are zero. This could theoretically
 // use MOV32ri with a SUBREG_TO_REG to represent the zero-extension, however
@@ -254,10 +257,10 @@ let AddedComplexity = 1, isReMaterializable = 1, isAsCheapAsAMove = 1,
     isCodeGenOnly = 1 in
 def MOV64ri64i32 : Ii32<0xB8, AddRegFrm, (outs GR64:$dst), (ins i64i32imm:$src),
                         "", [(set GR64:$dst, i64immZExt32:$src)],
-                        IIC_ALU_NONMEM>;
+                        IIC_ALU_NONMEM>, Sched<[WriteALU]>;
 
 // Use sbb to materialize carry bit.
-let Uses = [EFLAGS], Defs = [EFLAGS], isPseudo = 1 in {
+let Uses = [EFLAGS], Defs = [EFLAGS], isPseudo = 1, SchedRW = [WriteALU] in {
 // FIXME: These are pseudo ops that should be replaced with Pat<> patterns.
 // However, Pat<> can't replicate the destination reg into the inputs of the
 // result.
@@ -320,6 +323,7 @@ def : Pat<(sub GR64:$op, (i64 (X86setcc_c X86_COND_B, EFLAGS))),
 //===----------------------------------------------------------------------===//
 // String Pseudo Instructions
 //
+let SchedRW = [WriteMicrocoded] in {
 let Defs = [ECX,EDI,ESI], Uses = [ECX,EDI,ESI], isCodeGenOnly = 1 in {
 def REP_MOVSB_32 : I<0xA4, RawFrm, (outs), (ins), "{rep;movsb|rep movsb}",
                     [(X86rep_movs i8)], IIC_REP_MOVS>, REP,
@@ -382,6 +386,7 @@ let Defs = [RCX,RDI], isCodeGenOnly = 1 in {
                       [(X86rep_stos i64)], IIC_REP_STOS>, REP,
                      Requires<[In64BitMode]>;
 }
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // Thread Local Storage Instructions
@@ -513,18 +518,22 @@ def CMOV_RFP80 : I<0, Pseudo,
 
 multiclass PSEUDO_ATOMIC_LOAD_BINOP<string mnemonic> {
   let usesCustomInserter = 1, mayLoad = 1, mayStore = 1 in {
-    def #NAME#8  : I<0, Pseudo, (outs GR8:$dst),
-                     (ins i8mem:$ptr, GR8:$val),
-                     !strconcat(mnemonic, "8 PSEUDO!"), []>;
-    def #NAME#16 : I<0, Pseudo,(outs GR16:$dst),
-                     (ins i16mem:$ptr, GR16:$val),
-                     !strconcat(mnemonic, "16 PSEUDO!"), []>;
-    def #NAME#32 : I<0, Pseudo, (outs GR32:$dst),
-                     (ins i32mem:$ptr, GR32:$val),
-                     !strconcat(mnemonic, "32 PSEUDO!"), []>;
-    def #NAME#64 : I<0, Pseudo, (outs GR64:$dst),
-                     (ins i64mem:$ptr, GR64:$val),
-                     !strconcat(mnemonic, "64 PSEUDO!"), []>;
+    let Defs = [EFLAGS, AL] in
+    def NAME#8  : I<0, Pseudo, (outs GR8:$dst),
+                    (ins i8mem:$ptr, GR8:$val),
+                    !strconcat(mnemonic, "8 PSEUDO!"), []>;
+    let Defs = [EFLAGS, AX] in
+    def NAME#16 : I<0, Pseudo,(outs GR16:$dst),
+                    (ins i16mem:$ptr, GR16:$val),
+                    !strconcat(mnemonic, "16 PSEUDO!"), []>;
+    let Defs = [EFLAGS, EAX] in
+    def NAME#32 : I<0, Pseudo, (outs GR32:$dst),
+                    (ins i32mem:$ptr, GR32:$val),
+                    !strconcat(mnemonic, "32 PSEUDO!"), []>;
+    let Defs = [EFLAGS, RAX] in
+    def NAME#64 : I<0, Pseudo, (outs GR64:$dst),
+                    (ins i64mem:$ptr, GR64:$val),
+                    !strconcat(mnemonic, "64 PSEUDO!"), []>;
   }
 }
 
@@ -559,10 +568,11 @@ defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMUMAX", "atomic_load_umax">;
 defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMUMIN", "atomic_load_umin">;
 
 multiclass PSEUDO_ATOMIC_LOAD_BINOP6432<string mnemonic> {
-  let usesCustomInserter = 1, mayLoad = 1, mayStore = 1 in
-    def #NAME#6432 : I<0, Pseudo, (outs GR32:$dst1, GR32:$dst2),
-                       (ins i64mem:$ptr, GR32:$val1, GR32:$val2),
-                       !strconcat(mnemonic, "6432 PSEUDO!"), []>;
+  let usesCustomInserter = 1, Defs = [EFLAGS, EAX, EDX],
+      mayLoad = 1, mayStore = 1, hasSideEffects = 0 in
+    def NAME#6432 : I<0, Pseudo, (outs GR32:$dst1, GR32:$dst2),
+                      (ins i64mem:$ptr, GR32:$val1, GR32:$val2),
+                      !strconcat(mnemonic, "6432 PSEUDO!"), []>;
 }
 
 defm ATOMAND  : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMAND">;
@@ -589,12 +599,13 @@ defm ATOMSWAP : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMSWAP">;
 let isCodeGenOnly = 1, Defs = [EFLAGS] in
 def OR32mrLocked  : I<0x09, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$zero),
                       "or{l}\t{$zero, $dst|$dst, $zero}",
-                      [], IIC_ALU_MEM>, Requires<[In32BitMode]>, LOCK;
+                      [], IIC_ALU_MEM>, Requires<[In32BitMode]>, LOCK,
+                    Sched<[WriteALULd, WriteRMW]>;
 
 let hasSideEffects = 1 in
 def Int_MemBarrier : I<0, Pseudo, (outs), (ins),
                      "#MEMBARRIER",
-                     [(X86MemBarrier)]>;
+                     [(X86MemBarrier)]>, Sched<[WriteLoad]>;
 
 // RegOpc corresponds to the mr version of the instruction
 // ImmOpc corresponds to the mi version of the instruction
@@ -602,79 +613,80 @@ def Int_MemBarrier : I<0, Pseudo, (outs), (ins),
 // ImmMod corresponds to the instruction format of the mi and mi8 versions
 multiclass LOCK_ArithBinOp<bits<8> RegOpc, bits<8> ImmOpc, bits<8> ImmOpc8,
                            Format ImmMod, string mnemonic> {
-let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1 in {
-
-def #NAME#8mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
-                   RegOpc{3}, RegOpc{2}, RegOpc{1}, 0 },
-                   MRMDestMem, (outs), (ins i8mem:$dst, GR8:$src2),
-                   !strconcat(mnemonic, "{b}\t",
+let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1,
+    SchedRW = [WriteALULd, WriteRMW] in {
+
+def NAME#8mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
+                  RegOpc{3}, RegOpc{2}, RegOpc{1}, 0 },
+                  MRMDestMem, (outs), (ins i8mem:$dst, GR8:$src2),
+                  !strconcat(mnemonic, "{b}\t",
+                             "{$src2, $dst|$dst, $src2}"),
+                  [], IIC_ALU_NONMEM>, LOCK;
+def NAME#16mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
+                   RegOpc{3}, RegOpc{2}, RegOpc{1}, 1 },
+                   MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src2),
+                   !strconcat(mnemonic, "{w}\t",
+                              "{$src2, $dst|$dst, $src2}"),
+                   [], IIC_ALU_NONMEM>, OpSize, LOCK;
+def NAME#32mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
+                   RegOpc{3}, RegOpc{2}, RegOpc{1}, 1 },
+                   MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src2),
+                   !strconcat(mnemonic, "{l}\t",
                               "{$src2, $dst|$dst, $src2}"),
                    [], IIC_ALU_NONMEM>, LOCK;
-def #NAME#16mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
+def NAME#64mr : RI<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
                     RegOpc{3}, RegOpc{2}, RegOpc{1}, 1 },
-                    MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src2),
-                    !strconcat(mnemonic, "{w}\t",
-                               "{$src2, $dst|$dst, $src2}"),
-                    [], IIC_ALU_NONMEM>, OpSize, LOCK;
-def #NAME#32mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
-                    RegOpc{3}, RegOpc{2}, RegOpc{1}, 1 },
-                    MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src2),
-                    !strconcat(mnemonic, "{l}\t",
+                    MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
+                    !strconcat(mnemonic, "{q}\t",
                                "{$src2, $dst|$dst, $src2}"),
                     [], IIC_ALU_NONMEM>, LOCK;
-def #NAME#64mr : RI<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4},
-                     RegOpc{3}, RegOpc{2}, RegOpc{1}, 1 },
-                     MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
-                     !strconcat(mnemonic, "{q}\t",
-                                "{$src2, $dst|$dst, $src2}"),
-                     [], IIC_ALU_NONMEM>, LOCK;
-
-def #NAME#8mi : Ii8<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4},
-                     ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 0 },
-                     ImmMod, (outs), (ins i8mem :$dst, i8imm :$src2),
-                     !strconcat(mnemonic, "{b}\t",
-                                "{$src2, $dst|$dst, $src2}"),
-                     [], IIC_ALU_MEM>, LOCK;
-
-def #NAME#16mi : Ii16<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4},
-                       ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 },
-                       ImmMod, (outs), (ins i16mem :$dst, i16imm :$src2),
-                       !strconcat(mnemonic, "{w}\t",
-                                  "{$src2, $dst|$dst, $src2}"),
-                       [], IIC_ALU_MEM>, OpSize, LOCK;
-
-def #NAME#32mi : Ii32<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4},
-                       ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 },
-                       ImmMod, (outs), (ins i32mem :$dst, i32imm :$src2),
-                       !strconcat(mnemonic, "{l}\t",
-                                  "{$src2, $dst|$dst, $src2}"),
-                       [], IIC_ALU_MEM>, LOCK;
-
-def #NAME#64mi32 : RIi32<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4},
-                          ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 },
-                          ImmMod, (outs), (ins i64mem :$dst, i64i32imm :$src2),
-                          !strconcat(mnemonic, "{q}\t",
-                                     "{$src2, $dst|$dst, $src2}"),
-                          [], IIC_ALU_MEM>, LOCK;
 
-def #NAME#16mi8 : Ii8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4},
-                       ImmOpc8{3}, ImmOpc8{2}, ImmOpc8{1}, 1 },
-                       ImmMod, (outs), (ins i16mem :$dst, i16i8imm :$src2),
-                       !strconcat(mnemonic, "{w}\t",
-                                  "{$src2, $dst|$dst, $src2}"),
-                       [], IIC_ALU_MEM>, OpSize, LOCK;
-def #NAME#32mi8 : Ii8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4},
+def NAME#8mi : Ii8<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4},
+                    ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 0 },
+                    ImmMod, (outs), (ins i8mem :$dst, i8imm :$src2),
+                    !strconcat(mnemonic, "{b}\t",
+                               "{$src2, $dst|$dst, $src2}"),
+                    [], IIC_ALU_MEM>, LOCK;
+
+def NAME#16mi : Ii16<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4},
+                      ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 },
+                      ImmMod, (outs), (ins i16mem :$dst, i16imm :$src2),
+                      !strconcat(mnemonic, "{w}\t",
+                                 "{$src2, $dst|$dst, $src2}"),
+                      [], IIC_ALU_MEM>, OpSize, LOCK;
+
+def NAME#32mi : Ii32<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4},
+                      ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 },
+                      ImmMod, (outs), (ins i32mem :$dst, i32imm :$src2),
+                      !strconcat(mnemonic, "{l}\t",
+                                 "{$src2, $dst|$dst, $src2}"),
+                      [], IIC_ALU_MEM>, LOCK;
+
+def NAME#64mi32 : RIi32<{ImmOpc{7}, ImmOpc{6}, ImmOpc{5}, ImmOpc{4},
+                         ImmOpc{3}, ImmOpc{2}, ImmOpc{1}, 1 },
+                         ImmMod, (outs), (ins i64mem :$dst, i64i32imm :$src2),
+                         !strconcat(mnemonic, "{q}\t",
+                                    "{$src2, $dst|$dst, $src2}"),
+                         [], IIC_ALU_MEM>, LOCK;
+
+def NAME#16mi8 : Ii8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4},
+                      ImmOpc8{3}, ImmOpc8{2}, ImmOpc8{1}, 1 },
+                      ImmMod, (outs), (ins i16mem :$dst, i16i8imm :$src2),
+                      !strconcat(mnemonic, "{w}\t",
+                                 "{$src2, $dst|$dst, $src2}"),
+                      [], IIC_ALU_MEM>, OpSize, LOCK;
+def NAME#32mi8 : Ii8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4},
+                      ImmOpc8{3}, ImmOpc8{2}, ImmOpc8{1}, 1 },
+                      ImmMod, (outs), (ins i32mem :$dst, i32i8imm :$src2),
+                      !strconcat(mnemonic, "{l}\t",
+                                 "{$src2, $dst|$dst, $src2}"),
+                      [], IIC_ALU_MEM>, LOCK;
+def NAME#64mi8 : RIi8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4},
                        ImmOpc8{3}, ImmOpc8{2}, ImmOpc8{1}, 1 },
-                       ImmMod, (outs), (ins i32mem :$dst, i32i8imm :$src2),
-                       !strconcat(mnemonic, "{l}\t",
+                       ImmMod, (outs), (ins i64mem :$dst, i64i8imm :$src2),
+                       !strconcat(mnemonic, "{q}\t",
                                   "{$src2, $dst|$dst, $src2}"),
                        [], IIC_ALU_MEM>, LOCK;
-def #NAME#64mi8 : RIi8<{ImmOpc8{7}, ImmOpc8{6}, ImmOpc8{5}, ImmOpc8{4},
-                        ImmOpc8{3}, ImmOpc8{2}, ImmOpc8{1}, 1 },
-                        ImmMod, (outs), (ins i64mem :$dst, i64i8imm :$src2),
-                        !strconcat(mnemonic, "{q}\t",
-                                   "{$src2, $dst|$dst, $src2}"),
-                        [], IIC_ALU_MEM>, LOCK;
 
 }
 
@@ -689,20 +701,21 @@ defm LOCK_XOR : LOCK_ArithBinOp<0x30, 0x80, 0x83, MRM6m, "xor">;
 // Optimized codegen when the non-memory output is not used.
 multiclass LOCK_ArithUnOp<bits<8> Opc8, bits<8> Opc, Format Form,
                           string mnemonic> {
-let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1 in {
-
-def #NAME#8m  : I<Opc8, Form, (outs), (ins i8mem :$dst),
-                  !strconcat(mnemonic, "{b}\t$dst"),
+let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1,
+    SchedRW = [WriteALULd, WriteRMW] in {
+
+def NAME#8m  : I<Opc8, Form, (outs), (ins i8mem :$dst),
+                 !strconcat(mnemonic, "{b}\t$dst"),
+                 [], IIC_UNARY_MEM>, LOCK;
+def NAME#16m : I<Opc, Form, (outs), (ins i16mem:$dst),
+                 !strconcat(mnemonic, "{w}\t$dst"),
+                 [], IIC_UNARY_MEM>, OpSize, LOCK;
+def NAME#32m : I<Opc, Form, (outs), (ins i32mem:$dst),
+                 !strconcat(mnemonic, "{l}\t$dst"),
+                 [], IIC_UNARY_MEM>, LOCK;
+def NAME#64m : RI<Opc, Form, (outs), (ins i64mem:$dst),
+                  !strconcat(mnemonic, "{q}\t$dst"),
                   [], IIC_UNARY_MEM>, LOCK;
-def #NAME#16m : I<Opc, Form, (outs), (ins i16mem:$dst),
-                  !strconcat(mnemonic, "{w}\t$dst"),
-                  [], IIC_UNARY_MEM>, OpSize, LOCK;
-def #NAME#32m : I<Opc, Form, (outs), (ins i32mem:$dst),
-                  !strconcat(mnemonic, "{l}\t$dst"),
-                  [], IIC_UNARY_MEM>, LOCK;
-def #NAME#64m : RI<Opc, Form, (outs), (ins i64mem:$dst),
-                   !strconcat(mnemonic, "{q}\t$dst"),
-                   [], IIC_UNARY_MEM>, LOCK;
 }
 }
 
@@ -714,43 +727,44 @@ multiclass LCMPXCHG_UnOp<bits<8> Opc, Format Form, string mnemonic,
                          SDPatternOperator frag, X86MemOperand x86memop,
                          InstrItinClass itin> {
 let isCodeGenOnly = 1 in {
-  def #NAME# : I<Opc, Form, (outs), (ins x86memop:$ptr),
-                 !strconcat(mnemonic, "\t$ptr"),
-                 [(frag addr:$ptr)], itin>, TB, LOCK;
+  def NAME : I<Opc, Form, (outs), (ins x86memop:$ptr),
+               !strconcat(mnemonic, "\t$ptr"),
+               [(frag addr:$ptr)], itin>, TB, LOCK;
 }
 }
 
 multiclass LCMPXCHG_BinOp<bits<8> Opc8, bits<8> Opc, Format Form,
                           string mnemonic, SDPatternOperator frag,
                           InstrItinClass itin8, InstrItinClass itin> {
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, SchedRW = [WriteALULd, WriteRMW] in {
   let Defs = [AL, EFLAGS], Uses = [AL] in
-  def #NAME#8  : I<Opc8, Form, (outs), (ins i8mem:$ptr, GR8:$swap),
-                   !strconcat(mnemonic, "{b}\t{$swap, $ptr|$ptr, $swap}"),
-                   [(frag addr:$ptr, GR8:$swap, 1)], itin8>, TB, LOCK;
+  def NAME#8  : I<Opc8, Form, (outs), (ins i8mem:$ptr, GR8:$swap),
+                  !strconcat(mnemonic, "{b}\t{$swap, $ptr|$ptr, $swap}"),
+                  [(frag addr:$ptr, GR8:$swap, 1)], itin8>, TB, LOCK;
   let Defs = [AX, EFLAGS], Uses = [AX] in
-  def #NAME#16 : I<Opc, Form, (outs), (ins i16mem:$ptr, GR16:$swap),
-                   !strconcat(mnemonic, "{w}\t{$swap, $ptr|$ptr, $swap}"),
-                   [(frag addr:$ptr, GR16:$swap, 2)], itin>, TB, OpSize, LOCK;
+  def NAME#16 : I<Opc, Form, (outs), (ins i16mem:$ptr, GR16:$swap),
+                  !strconcat(mnemonic, "{w}\t{$swap, $ptr|$ptr, $swap}"),
+                  [(frag addr:$ptr, GR16:$swap, 2)], itin>, TB, OpSize, LOCK;
   let Defs = [EAX, EFLAGS], Uses = [EAX] in
-  def #NAME#32 : I<Opc, Form, (outs), (ins i32mem:$ptr, GR32:$swap),
-                   !strconcat(mnemonic, "{l}\t{$swap, $ptr|$ptr, $swap}"),
-                   [(frag addr:$ptr, GR32:$swap, 4)], itin>, TB, LOCK;
+  def NAME#32 : I<Opc, Form, (outs), (ins i32mem:$ptr, GR32:$swap),
+                  !strconcat(mnemonic, "{l}\t{$swap, $ptr|$ptr, $swap}"),
+                  [(frag addr:$ptr, GR32:$swap, 4)], itin>, TB, LOCK;
   let Defs = [RAX, EFLAGS], Uses = [RAX] in
-  def #NAME#64 : RI<Opc, Form, (outs), (ins i64mem:$ptr, GR64:$swap),
-                    !strconcat(mnemonic, "{q}\t{$swap, $ptr|$ptr, $swap}"),
-                    [(frag addr:$ptr, GR64:$swap, 8)], itin>, TB, LOCK;
+  def NAME#64 : RI<Opc, Form, (outs), (ins i64mem:$ptr, GR64:$swap),
+                   !strconcat(mnemonic, "{q}\t{$swap, $ptr|$ptr, $swap}"),
+                   [(frag addr:$ptr, GR64:$swap, 8)], itin>, TB, LOCK;
 }
 }
 
-let Defs = [EAX, EDX, EFLAGS], Uses = [EAX, EBX, ECX, EDX] in {
+let Defs = [EAX, EDX, EFLAGS], Uses = [EAX, EBX, ECX, EDX],
+    SchedRW = [WriteALULd, WriteRMW] in {
 defm LCMPXCHG8B : LCMPXCHG_UnOp<0xC7, MRM1m, "cmpxchg8b",
                                 X86cas8, i64mem,
                                 IIC_CMPX_LOCK_8B>;
 }
 
 let Defs = [RAX, RDX, EFLAGS], Uses = [RAX, RBX, RCX, RDX],
-    Predicates = [HasCmpxchg16b] in {
+    Predicates = [HasCmpxchg16b], SchedRW = [WriteALULd, WriteRMW] in {
 defm LCMPXCHG16B : LCMPXCHG_UnOp<0xC7, MRM1m, "cmpxchg16b",
                                  X86cas16, i128mem,
                                  IIC_CMPX_LOCK_16B>, REX_W;
@@ -763,34 +777,35 @@ defm LCMPXCHG : LCMPXCHG_BinOp<0xB0, 0xB1, MRMDestMem, "cmpxchg",
 multiclass ATOMIC_LOAD_BINOP<bits<8> opc8, bits<8> opc, string mnemonic,
                              string frag,
                              InstrItinClass itin8, InstrItinClass itin> {
-  let Constraints = "$val = $dst", Defs = [EFLAGS], isCodeGenOnly = 1 in {
-    def #NAME#8  : I<opc8, MRMSrcMem, (outs GR8:$dst),
-                     (ins GR8:$val, i8mem:$ptr),
-                     !strconcat(mnemonic, "{b}\t{$val, $ptr|$ptr, $val}"),
-                     [(set GR8:$dst,
-                           (!cast<PatFrag>(frag # "_8") addr:$ptr, GR8:$val))],
-                     itin8>;
-    def #NAME#16 : I<opc, MRMSrcMem, (outs GR16:$dst),
-                     (ins GR16:$val, i16mem:$ptr),
-                     !strconcat(mnemonic, "{w}\t{$val, $ptr|$ptr, $val}"),
+  let Constraints = "$val = $dst", Defs = [EFLAGS], isCodeGenOnly = 1,
+      SchedRW = [WriteALULd, WriteRMW] in {
+    def NAME#8  : I<opc8, MRMSrcMem, (outs GR8:$dst),
+                    (ins GR8:$val, i8mem:$ptr),
+                    !strconcat(mnemonic, "{b}\t{$val, $ptr|$ptr, $val}"),
+                    [(set GR8:$dst,
+                          (!cast<PatFrag>(frag # "_8") addr:$ptr, GR8:$val))],
+                    itin8>;
+    def NAME#16 : I<opc, MRMSrcMem, (outs GR16:$dst),
+                    (ins GR16:$val, i16mem:$ptr),
+                    !strconcat(mnemonic, "{w}\t{$val, $ptr|$ptr, $val}"),
+                    [(set
+                       GR16:$dst,
+                       (!cast<PatFrag>(frag # "_16") addr:$ptr, GR16:$val))],
+                    itin>, OpSize;
+    def NAME#32 : I<opc, MRMSrcMem, (outs GR32:$dst),
+                    (ins GR32:$val, i32mem:$ptr),
+                    !strconcat(mnemonic, "{l}\t{$val, $ptr|$ptr, $val}"),
+                    [(set
+                       GR32:$dst,
+                       (!cast<PatFrag>(frag # "_32") addr:$ptr, GR32:$val))],
+                    itin>;
+    def NAME#64 : RI<opc, MRMSrcMem, (outs GR64:$dst),
+                     (ins GR64:$val, i64mem:$ptr),
+                     !strconcat(mnemonic, "{q}\t{$val, $ptr|$ptr, $val}"),
                      [(set
-                        GR16:$dst,
-                        (!cast<PatFrag>(frag # "_16") addr:$ptr, GR16:$val))],
-                     itin>, OpSize;
-    def #NAME#32 : I<opc, MRMSrcMem, (outs GR32:$dst),
-                     (ins GR32:$val, i32mem:$ptr),
-                     !strconcat(mnemonic, "{l}\t{$val, $ptr|$ptr, $val}"),
-                     [(set
-                        GR32:$dst,
-                        (!cast<PatFrag>(frag # "_32") addr:$ptr, GR32:$val))],
+                        GR64:$dst,
+                        (!cast<PatFrag>(frag # "_64") addr:$ptr, GR64:$val))],
                      itin>;
-    def #NAME#64 : RI<opc, MRMSrcMem, (outs GR64:$dst),
-                      (ins GR64:$val, i64mem:$ptr),
-                      !strconcat(mnemonic, "{q}\t{$val, $ptr|$ptr, $val}"),
-                      [(set
-                         GR64:$dst,
-                         (!cast<PatFrag>(frag # "_64") addr:$ptr, GR64:$val))],
-                      itin>;
   }
 }
 
@@ -985,9 +1000,6 @@ def : Pat<(i64 (X86Wrapper tglobaltlsaddr :$dst)),
 // This corresponds to add $foo@tpoff, %rax
 def : Pat<(add GR64:$src1, (X86Wrapper tglobaltlsaddr :$dst)),
           (ADD64ri32 GR64:$src1, tglobaltlsaddr :$dst)>;
-// This corresponds to mov foo@tpoff(%rbx), %eax
-def : Pat<(load (i64 (X86Wrapper tglobaltlsaddr :$dst))),
-          (MOV64rm tglobaltlsaddr :$dst)>;
 
 
 // Direct PC relative function call for small code model. 32-bit displacement
@@ -1076,12 +1088,14 @@ def : Pat<(X86cmp GR64:$src1, 0),
 // inverted.
 multiclass CMOVmr<PatLeaf InvertedCond, Instruction Inst16, Instruction Inst32,
                   Instruction Inst64> {
-  def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, InvertedCond, EFLAGS),
-            (Inst16 GR16:$src2, addr:$src1)>;
-  def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, InvertedCond, EFLAGS),
-            (Inst32 GR32:$src2, addr:$src1)>;
-  def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, InvertedCond, EFLAGS),
-            (Inst64 GR64:$src2, addr:$src1)>;
+  let Predicates = [HasCMov] in {
+    def : Pat<(X86cmov (loadi16 addr:$src1), GR16:$src2, InvertedCond, EFLAGS),
+              (Inst16 GR16:$src2, addr:$src1)>;
+    def : Pat<(X86cmov (loadi32 addr:$src1), GR32:$src2, InvertedCond, EFLAGS),
+              (Inst32 GR32:$src2, addr:$src1)>;
+    def : Pat<(X86cmov (loadi64 addr:$src1), GR64:$src2, InvertedCond, EFLAGS),
+              (Inst64 GR64:$src2, addr:$src1)>;
+  }
 }
 
 defm : CMOVmr<X86_COND_B , CMOVAE16rm, CMOVAE32rm, CMOVAE64rm>;
@@ -1185,7 +1199,8 @@ def or_is_add : PatFrag<(ops node:$lhs, node:$rhs), (or node:$lhs, node:$rhs),[{
 
 
 // (or x1, x2) -> (add x1, x2) if two operands are known not to share bits.
-let AddedComplexity = 5 in { // Try this before the selecting to OR
+// Try this before the selecting to OR.
+let AddedComplexity = 5, SchedRW = [WriteALU] in {
 
 let isConvertibleToThreeAddress = 1,
     Constraints = "$src1 = $dst", Defs = [EFLAGS] in {
@@ -1232,7 +1247,7 @@ def ADD64ri32_DB : I<0, Pseudo,
                       [(set GR64:$dst, (or_is_add GR64:$src1,
                                                   i64immSExt32:$src2))]>;
 }
-} // AddedComplexity
+} // AddedComplexity, SchedRW
 
 
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/X86/X86InstrControl.td b/lib/Target/X86/X86InstrControl.td
index bfe954114c55..0e696513d47c 100644
--- a/lib/Target/X86/X86InstrControl.td
+++ b/lib/Target/X86/X86InstrControl.td
@@ -20,7 +20,7 @@
 // The X86retflag return instructions are variadic because we may add ST0 and
 // ST1 arguments when returning values on the x87 stack.
 let isTerminator = 1, isReturn = 1, isBarrier = 1,
-    hasCtrlDep = 1, FPForm = SpecialFP in {
+    hasCtrlDep = 1, FPForm = SpecialFP, SchedRW = [WriteJumpLd] in {
   def RET    : I   <0xC3, RawFrm, (outs), (ins variable_ops),
                     "ret",
                     [(X86retflag 0)], IIC_RET>;
@@ -46,7 +46,7 @@ let isTerminator = 1, isReturn = 1, isBarrier = 1,
 }
 
 // Unconditional branches.
-let isBarrier = 1, isBranch = 1, isTerminator = 1 in {
+let isBarrier = 1, isBranch = 1, isTerminator = 1, SchedRW = [WriteJump] in {
   def JMP_4 : Ii32PCRel<0xE9, RawFrm, (outs), (ins brtarget:$dst),
                         "jmp\t$dst", [(br bb:$dst)], IIC_JMP_REL>;
   def JMP_1 : Ii8PCRel<0xEB, RawFrm, (outs), (ins brtarget8:$dst),
@@ -58,7 +58,7 @@ let isBarrier = 1, isBranch = 1, isTerminator = 1 in {
 }
 
 // Conditional Branches.
-let isBranch = 1, isTerminator = 1, Uses = [EFLAGS] in {
+let isBranch = 1, isTerminator = 1, Uses = [EFLAGS], SchedRW = [WriteJump] in {
   multiclass ICBr<bits<8> opc1, bits<8> opc4, string asm, PatFrag Cond> {
     def _1 : Ii8PCRel <opc1, RawFrm, (outs), (ins brtarget8:$dst), asm, [],
                        IIC_Jcc>;
@@ -85,7 +85,7 @@ defm JLE : ICBr<0x7E, 0x8E, "jle\t$dst", X86_COND_LE>;
 defm JG  : ICBr<0x7F, 0x8F, "jg\t$dst" , X86_COND_G>;
 
 // jcx/jecx/jrcx instructions.
-let isBranch = 1, isTerminator = 1 in {
+let isBranch = 1, isTerminator = 1, SchedRW = [WriteJump] in {
   // These are the 32-bit versions of this instruction for the asmparser.  In
   // 32-bit mode, the address size prefix is jcxz and the unprefixed version is
   // jecxz.
@@ -110,36 +110,46 @@ let isBranch = 1, isTerminator = 1 in {
 // Indirect branches
 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
   def JMP32r     : I<0xFF, MRM4r, (outs), (ins GR32:$dst), "jmp{l}\t{*}$dst",
-                     [(brind GR32:$dst)], IIC_JMP_REG>, Requires<[In32BitMode]>;
+                     [(brind GR32:$dst)], IIC_JMP_REG>, Requires<[In32BitMode]>,
+                   Sched<[WriteJump]>;
   def JMP32m     : I<0xFF, MRM4m, (outs), (ins i32mem:$dst), "jmp{l}\t{*}$dst",
-                     [(brind (loadi32 addr:$dst))], IIC_JMP_MEM>, Requires<[In32BitMode]>;
+                     [(brind (loadi32 addr:$dst))], IIC_JMP_MEM>,
+                   Requires<[In32BitMode]>, Sched<[WriteJumpLd]>;
 
   def JMP64r     : I<0xFF, MRM4r, (outs), (ins GR64:$dst), "jmp{q}\t{*}$dst",
-                     [(brind GR64:$dst)], IIC_JMP_REG>, Requires<[In64BitMode]>;
+                     [(brind GR64:$dst)], IIC_JMP_REG>, Requires<[In64BitMode]>,
+                   Sched<[WriteJump]>;
   def JMP64m     : I<0xFF, MRM4m, (outs), (ins i64mem:$dst), "jmp{q}\t{*}$dst",
-                     [(brind (loadi64 addr:$dst))], IIC_JMP_MEM>, Requires<[In64BitMode]>;
+                     [(brind (loadi64 addr:$dst))], IIC_JMP_MEM>,
+                   Requires<[In64BitMode]>, Sched<[WriteJumpLd]>;
 
   def FARJMP16i  : Iseg16<0xEA, RawFrmImm16, (outs),
                           (ins i16imm:$off, i16imm:$seg),
-                          "ljmp{w}\t{$seg, $off|$off, $seg}", [], IIC_JMP_FAR_PTR>, OpSize;
+                          "ljmp{w}\t{$seg, $off|$off, $seg}", [],
+                          IIC_JMP_FAR_PTR>, OpSize, Sched<[WriteJump]>;
   def FARJMP32i  : Iseg32<0xEA, RawFrmImm16, (outs),
                           (ins i32imm:$off, i16imm:$seg),
-                          "ljmp{l}\t{$seg, $off|$off, $seg}", [], IIC_JMP_FAR_PTR>;
+                          "ljmp{l}\t{$seg, $off|$off, $seg}", [],
+                          IIC_JMP_FAR_PTR>, Sched<[WriteJump]>;
   def FARJMP64   : RI<0xFF, MRM5m, (outs), (ins opaque80mem:$dst),
-                      "ljmp{q}\t{*}$dst", [], IIC_JMP_FAR_MEM>;
+                      "ljmp{q}\t{*}$dst", [], IIC_JMP_FAR_MEM>,
+                   Sched<[WriteJump]>;
 
   def FARJMP16m  : I<0xFF, MRM5m, (outs), (ins opaque32mem:$dst),
-                     "ljmp{w}\t{*}$dst", [], IIC_JMP_FAR_MEM>, OpSize;
+                     "ljmp{w}\t{*}$dst", [], IIC_JMP_FAR_MEM>, OpSize,
+                   Sched<[WriteJumpLd]>;
   def FARJMP32m  : I<0xFF, MRM5m, (outs), (ins opaque48mem:$dst),
-                     "ljmp{l}\t{*}$dst", [], IIC_JMP_FAR_MEM>;
+                     "ljmp{l}\t{*}$dst", [], IIC_JMP_FAR_MEM>,
+                   Sched<[WriteJumpLd]>;
 }
 
 
 // Loop instructions
-
+let SchedRW = [WriteJump] in {
 def LOOP   : Ii8PCRel<0xE2, RawFrm, (outs), (ins brtarget8:$dst), "loop\t$dst", [], IIC_LOOP>;
 def LOOPE  : Ii8PCRel<0xE1, RawFrm, (outs), (ins brtarget8:$dst), "loope\t$dst", [], IIC_LOOPE>;
 def LOOPNE : Ii8PCRel<0xE0, RawFrm, (outs), (ins brtarget8:$dst), "loopne\t$dst", [], IIC_LOOPNE>;
+}
 
 //===----------------------------------------------------------------------===//
 //  Call Instructions...
@@ -152,27 +162,32 @@ let isCall = 1 in
   let Uses = [ESP] in {
     def CALLpcrel32 : Ii32PCRel<0xE8, RawFrm,
                            (outs), (ins i32imm_pcrel:$dst),
-                           "call{l}\t$dst", [], IIC_CALL_RI>, Requires<[In32BitMode]>;
+                           "call{l}\t$dst", [], IIC_CALL_RI>,
+                      Requires<[In32BitMode]>, Sched<[WriteJump]>;
     def CALL32r     : I<0xFF, MRM2r, (outs), (ins GR32:$dst),
                         "call{l}\t{*}$dst", [(X86call GR32:$dst)], IIC_CALL_RI>,
-                         Requires<[In32BitMode]>;
+                      Requires<[In32BitMode]>, Sched<[WriteJump]>;
     def CALL32m     : I<0xFF, MRM2m, (outs), (ins i32mem:$dst),
-                        "call{l}\t{*}$dst", [(X86call (loadi32 addr:$dst))], IIC_CALL_MEM>,
-                        Requires<[In32BitMode]>;
+                        "call{l}\t{*}$dst", [(X86call (loadi32 addr:$dst))],
+                        IIC_CALL_MEM>,
+                      Requires<[In32BitMode,FavorMemIndirectCall]>,
+                      Sched<[WriteJumpLd]>;
 
     def FARCALL16i  : Iseg16<0x9A, RawFrmImm16, (outs),
                              (ins i16imm:$off, i16imm:$seg),
                              "lcall{w}\t{$seg, $off|$off, $seg}", [],
-                             IIC_CALL_FAR_PTR>, OpSize;
+                             IIC_CALL_FAR_PTR>, OpSize, Sched<[WriteJump]>;
     def FARCALL32i  : Iseg32<0x9A, RawFrmImm16, (outs),
                              (ins i32imm:$off, i16imm:$seg),
                              "lcall{l}\t{$seg, $off|$off, $seg}", [],
-                             IIC_CALL_FAR_PTR>;
+                             IIC_CALL_FAR_PTR>, Sched<[WriteJump]>;
 
     def FARCALL16m  : I<0xFF, MRM3m, (outs), (ins opaque32mem:$dst),
-                        "lcall{w}\t{*}$dst", [], IIC_CALL_FAR_MEM>, OpSize;
+                        "lcall{w}\t{*}$dst", [], IIC_CALL_FAR_MEM>, OpSize,
+                      Sched<[WriteJumpLd]>;
     def FARCALL32m  : I<0xFF, MRM3m, (outs), (ins opaque48mem:$dst),
-                        "lcall{l}\t{*}$dst", [], IIC_CALL_FAR_MEM>;
+                        "lcall{l}\t{*}$dst", [], IIC_CALL_FAR_MEM>,
+                      Sched<[WriteJumpLd]>;
 
     // callw for 16 bit code for the assembler.
     let isAsmParserOnly = 1 in
@@ -185,7 +200,7 @@ let isCall = 1 in
 // Tail call stuff.
 
 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1,
-    isCodeGenOnly = 1 in
+    isCodeGenOnly = 1, SchedRW = [WriteJumpLd] in
   let Uses = [ESP] in {
   def TCRETURNdi : PseudoI<(outs),
                      (ins i32imm_pcrel:$dst, i32imm:$offset), []>;
@@ -216,7 +231,7 @@ let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1,
 // RSP is marked as a use to prevent stack-pointer assignments that appear
 // immediately before calls from potentially appearing dead. Uses for argument
 // registers are added manually.
-let isCall = 1, Uses = [RSP] in {
+let isCall = 1, Uses = [RSP], SchedRW = [WriteJump] in {
   // NOTE: this pattern doesn't match "X86call imm", because we do not know
   // that the offset between an arbitrary immediate and the call will fit in
   // the 32-bit pcrel field that we have.
@@ -231,7 +246,7 @@ let isCall = 1, Uses = [RSP] in {
   def CALL64m       : I<0xFF, MRM2m, (outs), (ins i64mem:$dst),
                         "call{q}\t{*}$dst", [(X86call (loadi64 addr:$dst))],
                         IIC_CALL_MEM>,
-                      Requires<[In64BitMode]>;
+                      Requires<[In64BitMode,FavorMemIndirectCall]>;
 
   def FARCALL64   : RI<0xFF, MRM3m, (outs), (ins opaque80mem:$dst),
                        "lcall{q}\t{*}$dst", [], IIC_CALL_FAR_MEM>;
@@ -245,13 +260,12 @@ let isCall = 1, isCodeGenOnly = 1 in
     def W64ALLOCA : Ii32PCRel<0xE8, RawFrm,
                       (outs), (ins i64i32imm_pcrel:$dst),
                       "call{q}\t$dst", [], IIC_CALL_RI>,
-                    Requires<[IsWin64]>;
+                    Requires<[IsWin64]>, Sched<[WriteJump]>;
   }
 
 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1,
-    isCodeGenOnly = 1 in
-  let Uses = [RSP],
-      usesCustomInserter = 1 in {
+    isCodeGenOnly = 1, Uses = [RSP], usesCustomInserter = 1,
+    SchedRW = [WriteJump] in {
   def TCRETURNdi64 : PseudoI<(outs),
                       (ins i64i32imm_pcrel:$dst, i32imm:$offset),
                       []>;
diff --git a/lib/Target/X86/X86InstrExtension.td b/lib/Target/X86/X86InstrExtension.td
index 2eb454ded21b..6dc7175357b3 100644
--- a/lib/Target/X86/X86InstrExtension.td
+++ b/lib/Target/X86/X86InstrExtension.td
@@ -42,48 +42,54 @@ let neverHasSideEffects = 1 in {
 let neverHasSideEffects = 1 in {
 def MOVSX16rr8 : I<0xBE, MRMSrcReg, (outs GR16:$dst), (ins GR8:$src),
                    "movs{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVSX_R16_R8>,
-                   TB, OpSize;
+                   TB, OpSize, Sched<[WriteALU]>;
 let mayLoad = 1 in
 def MOVSX16rm8 : I<0xBE, MRMSrcMem, (outs GR16:$dst), (ins i8mem:$src),
                    "movs{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVSX_R16_M8>,
-                   TB, OpSize;
+                   TB, OpSize, Sched<[WriteALULd]>;
 } // neverHasSideEffects = 1
 def MOVSX32rr8 : I<0xBE, MRMSrcReg, (outs GR32:$dst), (ins GR8:$src),
                    "movs{bl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (sext GR8:$src))], IIC_MOVSX>, TB;
+                   [(set GR32:$dst, (sext GR8:$src))], IIC_MOVSX>, TB,
+                   Sched<[WriteALU]>;
 def MOVSX32rm8 : I<0xBE, MRMSrcMem, (outs GR32:$dst), (ins i8mem :$src),
                    "movs{bl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (sextloadi32i8 addr:$src))], IIC_MOVSX>, TB;
+                   [(set GR32:$dst, (sextloadi32i8 addr:$src))], IIC_MOVSX>, TB,
+                   Sched<[WriteALULd]>;
 def MOVSX32rr16: I<0xBF, MRMSrcReg, (outs GR32:$dst), (ins GR16:$src),
                    "movs{wl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (sext GR16:$src))], IIC_MOVSX>, TB;
+                   [(set GR32:$dst, (sext GR16:$src))], IIC_MOVSX>, TB,
+                   Sched<[WriteALU]>;
 def MOVSX32rm16: I<0xBF, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src),
                    "movs{wl|x}\t{$src, $dst|$dst, $src}",
                    [(set GR32:$dst, (sextloadi32i16 addr:$src))], IIC_MOVSX>,
-                   TB;
+                   TB, Sched<[WriteALULd]>;
 
 let neverHasSideEffects = 1 in {
 def MOVZX16rr8 : I<0xB6, MRMSrcReg, (outs GR16:$dst), (ins GR8:$src),
                    "movz{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX_R16_R8>,
-                   TB, OpSize;
+                   TB, OpSize, Sched<[WriteALU]>;
 let mayLoad = 1 in
 def MOVZX16rm8 : I<0xB6, MRMSrcMem, (outs GR16:$dst), (ins i8mem:$src),
                    "movz{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX_R16_M8>,
-                   TB, OpSize;
+                   TB, OpSize, Sched<[WriteALULd]>;
 } // neverHasSideEffects = 1
 def MOVZX32rr8 : I<0xB6, MRMSrcReg, (outs GR32:$dst), (ins GR8 :$src),
                    "movz{bl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (zext GR8:$src))], IIC_MOVZX>, TB;
+                   [(set GR32:$dst, (zext GR8:$src))], IIC_MOVZX>, TB,
+                   Sched<[WriteALU]>;
 def MOVZX32rm8 : I<0xB6, MRMSrcMem, (outs GR32:$dst), (ins i8mem :$src),
                    "movz{bl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (zextloadi32i8 addr:$src))], IIC_MOVZX>, TB;
+                   [(set GR32:$dst, (zextloadi32i8 addr:$src))], IIC_MOVZX>, TB,
+                   Sched<[WriteALULd]>;
 def MOVZX32rr16: I<0xB7, MRMSrcReg, (outs GR32:$dst), (ins GR16:$src),
                    "movz{wl|x}\t{$src, $dst|$dst, $src}",
-                   [(set GR32:$dst, (zext GR16:$src))], IIC_MOVZX>, TB;
+                   [(set GR32:$dst, (zext GR16:$src))], IIC_MOVZX>, TB,
+                   Sched<[WriteALU]>;
 def MOVZX32rm16: I<0xB7, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src),
                    "movz{wl|x}\t{$src, $dst|$dst, $src}",
                    [(set GR32:$dst, (zextloadi32i16 addr:$src))], IIC_MOVZX>,
-                   TB;
+                   TB, Sched<[WriteALULd]>;
 
 // These are the same as the regular MOVZX32rr8 and MOVZX32rm8
 // except that they use GR32_NOREX for the output operand register class
@@ -92,12 +98,12 @@ let neverHasSideEffects = 1, isCodeGenOnly = 1 in {
 def MOVZX32_NOREXrr8 : I<0xB6, MRMSrcReg,
                          (outs GR32_NOREX:$dst), (ins GR8_NOREX:$src),
                          "movz{bl|x}\t{$src, $dst|$dst, $src}",
-                         [], IIC_MOVZX>, TB;
+                         [], IIC_MOVZX>, TB, Sched<[WriteALU]>;
 let mayLoad = 1 in
 def MOVZX32_NOREXrm8 : I<0xB6, MRMSrcMem,
                          (outs GR32_NOREX:$dst), (ins i8mem_NOREX:$src),
                          "movz{bl|x}\t{$src, $dst|$dst, $src}",
-                         [], IIC_MOVZX>, TB;
+                         [], IIC_MOVZX>, TB, Sched<[WriteALULd]>;
 }
 
 // MOVSX64rr8 always has a REX prefix and it has an 8-bit register
@@ -106,38 +112,42 @@ def MOVZX32_NOREXrm8 : I<0xB6, MRMSrcMem,
 // were generalized, this would require a special register class.
 def MOVSX64rr8 : RI<0xBE, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src),
                     "movs{bq|x}\t{$src, $dst|$dst, $src}",
-                    [(set GR64:$dst, (sext GR8:$src))], IIC_MOVSX>, TB;
+                    [(set GR64:$dst, (sext GR8:$src))], IIC_MOVSX>, TB,
+                    Sched<[WriteALU]>;
 def MOVSX64rm8 : RI<0xBE, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src),
                     "movs{bq|x}\t{$src, $dst|$dst, $src}",
                     [(set GR64:$dst, (sextloadi64i8 addr:$src))], IIC_MOVSX>,
-                    TB;
+                    TB, Sched<[WriteALULd]>;
 def MOVSX64rr16: RI<0xBF, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
                     "movs{wq|x}\t{$src, $dst|$dst, $src}",
-                    [(set GR64:$dst, (sext GR16:$src))], IIC_MOVSX>, TB;
+                    [(set GR64:$dst, (sext GR16:$src))], IIC_MOVSX>, TB,
+                    Sched<[WriteALU]>;
 def MOVSX64rm16: RI<0xBF, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
                     "movs{wq|x}\t{$src, $dst|$dst, $src}",
                     [(set GR64:$dst, (sextloadi64i16 addr:$src))], IIC_MOVSX>,
-                    TB;
+                    TB, Sched<[WriteALULd]>;
 def MOVSX64rr32: RI<0x63, MRMSrcReg, (outs GR64:$dst), (ins GR32:$src),
                     "movs{lq|xd}\t{$src, $dst|$dst, $src}",
-                    [(set GR64:$dst, (sext GR32:$src))], IIC_MOVSX>;
+                    [(set GR64:$dst, (sext GR32:$src))], IIC_MOVSX>,
+                    Sched<[WriteALU]>;
 def MOVSX64rm32: RI<0x63, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src),
                     "movs{lq|xd}\t{$src, $dst|$dst, $src}",
-                    [(set GR64:$dst, (sextloadi64i32 addr:$src))], IIC_MOVSX>;
+                    [(set GR64:$dst, (sextloadi64i32 addr:$src))], IIC_MOVSX>,
+                    Sched<[WriteALULd]>;
 
 // movzbq and movzwq encodings for the disassembler
 def MOVZX64rr8_Q : RI<0xB6, MRMSrcReg, (outs GR64:$dst), (ins GR8:$src),
                        "movz{bq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
-                       TB;
+                       TB, Sched<[WriteALU]>;
 def MOVZX64rm8_Q : RI<0xB6, MRMSrcMem, (outs GR64:$dst), (ins i8mem:$src),
                        "movz{bq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
-                       TB;
+                       TB, Sched<[WriteALULd]>;
 def MOVZX64rr16_Q : RI<0xB7, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
                        "movz{wq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
-                       TB;
+                       TB, Sched<[WriteALU]>;
 def MOVZX64rm16_Q : RI<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
                        "movz{wq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>,
-                       TB;
+                       TB, Sched<[WriteALULd]>;
 
 // FIXME: These should be Pat patterns.
 let isCodeGenOnly = 1 in {
@@ -145,17 +155,19 @@ let isCodeGenOnly = 1 in {
 // Use movzbl instead of movzbq when the destination is a register; it's
 // equivalent due to implicit zero-extending, and it has a smaller encoding.
 def MOVZX64rr8 : I<0xB6, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src),
-                   "", [(set GR64:$dst, (zext GR8:$src))], IIC_MOVZX>, TB;
+                   "", [(set GR64:$dst, (zext GR8:$src))], IIC_MOVZX>, TB,
+                   Sched<[WriteALU]>;
 def MOVZX64rm8 : I<0xB6, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src),
                    "", [(set GR64:$dst, (zextloadi64i8 addr:$src))], IIC_MOVZX>,
-                   TB;
+                   TB, Sched<[WriteALULd]>;
 // Use movzwl instead of movzwq when the destination is a register; it's
 // equivalent due to implicit zero-extending, and it has a smaller encoding.
 def MOVZX64rr16: I<0xB7, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
-                   "", [(set GR64:$dst, (zext GR16:$src))], IIC_MOVZX>, TB;
+                   "", [(set GR64:$dst, (zext GR16:$src))], IIC_MOVZX>, TB,
+                   Sched<[WriteALU]>;
 def MOVZX64rm16: I<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
                    "", [(set GR64:$dst, (zextloadi64i16 addr:$src))],
-                   IIC_MOVZX>, TB;
+                   IIC_MOVZX>, TB, Sched<[WriteALULd]>;
 
 // There's no movzlq instruction, but movl can be used for this purpose, using
 // implicit zero-extension. The preferred way to do 32-bit-to-64-bit zero
@@ -165,9 +177,10 @@ def MOVZX64rm16: I<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
 // necessarily all zero. In such cases, we fall back to these explicit zext
 // instructions.
 def MOVZX64rr32 : I<0x89, MRMDestReg, (outs GR64:$dst), (ins GR32:$src),
-                    "", [(set GR64:$dst, (zext GR32:$src))], IIC_MOVZX>;
+                    "", [(set GR64:$dst, (zext GR32:$src))], IIC_MOVZX>,
+                    Sched<[WriteALU]>;
 def MOVZX64rm32 : I<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src),
                     "", [(set GR64:$dst, (zextloadi64i32 addr:$src))],
-                    IIC_MOVZX>;
+                    IIC_MOVZX>, Sched<[WriteALULd]>;
 }
 
diff --git a/lib/Target/X86/X86InstrFMA.td b/lib/Target/X86/X86InstrFMA.td
index 959d91a9ab6b..7759a8a2dabb 100644
--- a/lib/Target/X86/X86InstrFMA.td
+++ b/lib/Target/X86/X86InstrFMA.td
@@ -60,14 +60,14 @@ multiclass fma3p_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
                        PatFrag MemFrag128, PatFrag MemFrag256,
                        SDNode Op, ValueType OpTy128, ValueType OpTy256> {
   defm r213 : fma3p_rm<opc213,
-                       !strconcat(OpcodeStr, !strconcat("213", PackTy)),
+                       !strconcat(OpcodeStr, "213", PackTy),
                        MemFrag128, MemFrag256, OpTy128, OpTy256, Op>;
 let neverHasSideEffects = 1 in {
   defm r132 : fma3p_rm<opc132,
-                       !strconcat(OpcodeStr, !strconcat("132", PackTy)),
+                       !strconcat(OpcodeStr, "132", PackTy),
                        MemFrag128, MemFrag256, OpTy128, OpTy256>;
   defm r231 : fma3p_rm<opc231,
-                       !strconcat(OpcodeStr, !strconcat("231", PackTy)),
+                       !strconcat(OpcodeStr, "231", PackTy),
                        MemFrag128, MemFrag256, OpTy128, OpTy256>;
 } // neverHasSideEffects = 1
 }
@@ -160,15 +160,15 @@ multiclass fma3s_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
                        X86MemOperand x86memop, Operand memop, PatFrag mem_frag,
                        ComplexPattern mem_cpat> {
 let neverHasSideEffects = 1 in {
-  defm r132 : fma3s_rm<opc132, !strconcat(OpStr, !strconcat("132", PackTy)),
+  defm r132 : fma3s_rm<opc132, !strconcat(OpStr, "132", PackTy),
                        x86memop, RC, OpVT, mem_frag>;
-  defm r231 : fma3s_rm<opc231, !strconcat(OpStr, !strconcat("231", PackTy)),
+  defm r231 : fma3s_rm<opc231, !strconcat(OpStr, "231", PackTy),
                        x86memop, RC, OpVT, mem_frag>;
 }
 
-defm r213 : fma3s_rm<opc213, !strconcat(OpStr, !strconcat("213", PackTy)),
+defm r213 : fma3s_rm<opc213, !strconcat(OpStr, "213", PackTy),
                      x86memop, RC, OpVT, mem_frag, OpNode>,
-            fma3s_rm_int<opc213, !strconcat(OpStr, !strconcat("213", PackTy)),
+            fma3s_rm_int<opc213, !strconcat(OpStr, "213", PackTy),
                          memop, mem_cpat, Int, RC>;
 }
 
@@ -220,7 +220,7 @@ multiclass fma4s<bits<8> opc, string OpcodeStr, RegisterClass RC,
            [(set RC:$dst,
              (OpNode RC:$src1, (mem_frag addr:$src2), RC:$src3))]>;
 // For disassembler
-let isCodeGenOnly = 1 in
+let isCodeGenOnly = 1, hasSideEffects = 0 in
   def rr_REV : FMA4<opc, MRMSrcReg, (outs RC:$dst),
                (ins RC:$src1, RC:$src2, RC:$src3),
                !strconcat(OpcodeStr,
@@ -294,7 +294,7 @@ multiclass fma4p<bits<8> opc, string OpcodeStr, SDNode OpNode,
            [(set VR256:$dst, (OpNode VR256:$src1,
                               (ld_frag256 addr:$src2), VR256:$src3))]>, VEX_L;
 // For disassembler
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, hasSideEffects = 0 in {
   def rr_REV : FMA4<opc, MRMSrcReg, (outs VR128:$dst),
                (ins VR128:$src1, VR128:$src2, VR128:$src3),
                !strconcat(OpcodeStr,
@@ -307,8 +307,6 @@ let isCodeGenOnly = 1 in {
 } // isCodeGenOnly = 1
 }
 
-let Predicates = [HasFMA4] in {
-
 defm VFMADDSS4  : fma4s<0x6A, "vfmaddss", FR32, f32mem, f32, X86Fmadd, loadf32>,
                   fma4s_int<0x6A, "vfmaddss", ssmem, sse_load_f32,
                             int_x86_fma_vfmadd_ss>;
@@ -338,29 +336,33 @@ defm VFNMSUBSD4 : fma4s<0x7F, "vfnmsubsd", FR64, f64mem, f64,
                   fma4s_int<0x7F, "vfnmsubsd", sdmem, sse_load_f64,
                             int_x86_fma_vfnmsub_sd>;
 
-defm VFMADDPS4    : fma4p<0x68, "vfmaddps", X86Fmadd, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFMADDPD4    : fma4p<0x69, "vfmaddpd", X86Fmadd, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-defm VFMSUBPS4    : fma4p<0x6C, "vfmsubps", X86Fmsub, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFMSUBPD4    : fma4p<0x6D, "vfmsubpd", X86Fmsub, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-defm VFNMADDPS4   : fma4p<0x78, "vfnmaddps", X86Fnmadd, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFNMADDPD4   : fma4p<0x79, "vfnmaddpd", X86Fnmadd, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-defm VFNMSUBPS4   : fma4p<0x7C, "vfnmsubps", X86Fnmsub, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFNMSUBPD4   : fma4p<0x7D, "vfnmsubpd", X86Fnmsub, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-defm VFMADDSUBPS4 : fma4p<0x5C, "vfmaddsubps", X86Fmaddsub, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFMADDSUBPD4 : fma4p<0x5D, "vfmaddsubpd", X86Fmaddsub, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-defm VFMSUBADDPS4 : fma4p<0x5E, "vfmsubaddps", X86Fmsubadd, v4f32, v8f32,
-                          memopv4f32, memopv8f32>;
-defm VFMSUBADDPD4 : fma4p<0x5F, "vfmsubaddpd", X86Fmsubadd, v2f64, v4f64,
-                          memopv2f64, memopv4f64>;
-} // HasFMA4
+let ExeDomain = SSEPackedSingle in {
+  defm VFMADDPS4    : fma4p<0x68, "vfmaddps", X86Fmadd, v4f32, v8f32,
+                            memopv4f32, memopv8f32>;
+  defm VFMSUBPS4    : fma4p<0x6C, "vfmsubps", X86Fmsub, v4f32, v8f32,
+                            memopv4f32, memopv8f32>;
+  defm VFNMADDPS4   : fma4p<0x78, "vfnmaddps", X86Fnmadd, v4f32, v8f32,
+                            memopv4f32, memopv8f32>;
+  defm VFNMSUBPS4   : fma4p<0x7C, "vfnmsubps", X86Fnmsub, v4f32, v8f32,
+                            memopv4f32, memopv8f32>;
+  defm VFMADDSUBPS4 : fma4p<0x5C, "vfmaddsubps", X86Fmaddsub, v4f32, v8f32,
+                            memopv4f32, memopv8f32>;
+  defm VFMSUBADDPS4 : fma4p<0x5E, "vfmsubaddps", X86Fmsubadd, v4f32, v8f32,
+                            memopv4f32, memopv8f32>;
+}
+
+let ExeDomain = SSEPackedDouble in {
+  defm VFMADDPD4    : fma4p<0x69, "vfmaddpd", X86Fmadd, v2f64, v4f64,
+                            memopv2f64, memopv4f64>;
+  defm VFMSUBPD4    : fma4p<0x6D, "vfmsubpd", X86Fmsub, v2f64, v4f64,
+                            memopv2f64, memopv4f64>;
+  defm VFNMADDPD4   : fma4p<0x79, "vfnmaddpd", X86Fnmadd, v2f64, v4f64,
+                            memopv2f64, memopv4f64>;
+  defm VFNMSUBPD4   : fma4p<0x7D, "vfnmsubpd", X86Fnmsub, v2f64, v4f64,
+                            memopv2f64, memopv4f64>;
+  defm VFMADDSUBPD4 : fma4p<0x5D, "vfmaddsubpd", X86Fmaddsub, v2f64, v4f64,
+                            memopv2f64, memopv4f64>;
+  defm VFMSUBADDPD4 : fma4p<0x5F, "vfmsubaddpd", X86Fmsubadd, v2f64, v4f64,
+                            memopv2f64, memopv4f64>;
+}
 
diff --git a/lib/Target/X86/X86InstrFPStack.td b/lib/Target/X86/X86InstrFPStack.td
index 568726e08ece..2224a08d59f4 100644
--- a/lib/Target/X86/X86InstrFPStack.td
+++ b/lib/Target/X86/X86InstrFPStack.td
@@ -422,7 +422,7 @@ def IST_Fp32m80  : FpI_<(outs), (ins i32mem:$op, RFP80:$src), OneArgFP, []>;
 def IST_Fp64m80  : FpI_<(outs), (ins i64mem:$op, RFP80:$src), OneArgFP, []>;
 }
 
-let mayLoad = 1 in {
+let mayLoad = 1, SchedRW = [WriteLoad] in {
 def LD_F32m   : FPI<0xD9, MRM0m, (outs), (ins f32mem:$src), "fld{s}\t$src",
                     IIC_FLD>;
 def LD_F64m   : FPI<0xDD, MRM0m, (outs), (ins f64mem:$src), "fld{l}\t$src",
@@ -436,7 +436,7 @@ def ILD_F32m  : FPI<0xDB, MRM0m, (outs), (ins i32mem:$src), "fild{l}\t$src",
 def ILD_F64m  : FPI<0xDF, MRM5m, (outs), (ins i64mem:$src), "fild{ll}\t$src",
                     IIC_FILD>;
 }
-let mayStore = 1 in {
+let mayStore = 1, SchedRW = [WriteStore] in {
 def ST_F32m   : FPI<0xD9, MRM2m, (outs), (ins f32mem:$dst), "fst{s}\t$dst",
                     IIC_FST>;
 def ST_F64m   : FPI<0xDD, MRM2m, (outs), (ins f64mem:$dst), "fst{l}\t$dst",
@@ -481,7 +481,7 @@ def ISTT_Fp64m80 : FpI_<(outs), (ins i64mem:$op, RFP80:$src), OneArgFP,
                     [(X86fp_to_i64mem RFP80:$src, addr:$op)]>;
 } // Predicates = [HasSSE3]
 
-let mayStore = 1 in {
+let mayStore = 1, SchedRW = [WriteStore] in {
 def ISTT_FP16m : FPI<0xDF, MRM1m, (outs), (ins i16mem:$dst), "fisttp{s}\t$dst",
   IIC_FST>;
 def ISTT_FP32m : FPI<0xDB, MRM1m, (outs), (ins i32mem:$dst), "fisttp{l}\t$dst",
@@ -491,6 +491,7 @@ def ISTT_FP64m : FPI<0xDD, MRM1m, (outs), (ins i64mem:$dst),
 }
 
 // FP Stack manipulation instructions.
+let SchedRW = [WriteMove] in {
 def LD_Frr   : FPI<0xC0, AddRegFrm, (outs), (ins RST:$op), "fld\t$op",
                    IIC_FLD>, D9;
 def ST_Frr   : FPI<0xD0, AddRegFrm, (outs), (ins RST:$op), "fst\t$op",
@@ -499,6 +500,7 @@ def ST_FPrr  : FPI<0xD8, AddRegFrm, (outs), (ins RST:$op), "fstp\t$op",
                    IIC_FST>, DD;
 def XCH_F    : FPI<0xC8, AddRegFrm, (outs), (ins RST:$op), "fxch\t$op",
                    IIC_FXCH>, D9;
+}
 
 // Floating point constant loads.
 let isReMaterializable = 1 in {
@@ -516,19 +518,23 @@ def LD_Fp180 : FpI_<(outs RFP80:$dst), (ins), ZeroArgFP,
                 [(set RFP80:$dst, fpimm1)]>;
 }
 
+let SchedRW = [WriteZero] in {
 def LD_F0 : FPI<0xEE, RawFrm, (outs), (ins), "fldz", IIC_FLDZ>, D9;
 def LD_F1 : FPI<0xE8, RawFrm, (outs), (ins), "fld1", IIC_FIST>, D9;
-
+}
 
 // Floating point compares.
+let SchedRW = [WriteFAdd] in {
 def UCOM_Fpr32 : FpIf32<(outs), (ins RFP32:$lhs, RFP32:$rhs), CompareFP,
                         [(set FPSW, (trunc (X86cmp RFP32:$lhs, RFP32:$rhs)))]>;
 def UCOM_Fpr64 : FpIf64<(outs), (ins RFP64:$lhs, RFP64:$rhs), CompareFP,
                         [(set FPSW, (trunc (X86cmp RFP64:$lhs, RFP64:$rhs)))]>;
 def UCOM_Fpr80 : FpI_  <(outs), (ins RFP80:$lhs, RFP80:$rhs), CompareFP,
                         [(set FPSW, (trunc (X86cmp RFP80:$lhs, RFP80:$rhs)))]>;
+} // SchedRW
 } // Defs = [FPSW]
 
+let SchedRW = [WriteFAdd] in {
 // CC = ST(0) cmp ST(i)
 let Defs = [EFLAGS, FPSW] in {
 def UCOM_FpIr32: FpIf32<(outs), (ins RFP32:$lhs, RFP32:$rhs), CompareFP,
@@ -566,8 +572,10 @@ def COM_FIr : FPI<0xF0, AddRegFrm, (outs), (ins RST:$reg),
 def COM_FIPr : FPI<0xF0, AddRegFrm, (outs), (ins RST:$reg),
                    "fcompi\t$reg", IIC_FCOMI>, DF;
 }
+} // SchedRW
 
 // Floating point flag ops.
+let SchedRW = [WriteALU] in {
 let Defs = [AX], Uses = [FPSW] in
 def FNSTSW16r : I<0xE0, RawFrm,                  // AX = fp flags
                   (outs), (ins), "fnstsw %ax",
@@ -576,23 +584,26 @@ def FNSTSW16r : I<0xE0, RawFrm,                  // AX = fp flags
 def FNSTCW16m : I<0xD9, MRM7m,                   // [mem16] = X87 control world
                   (outs), (ins i16mem:$dst), "fnstcw\t$dst",
                   [(X86fp_cwd_get16 addr:$dst)], IIC_FNSTCW>;
-                  
+} // SchedRW
 let mayLoad = 1 in
 def FLDCW16m  : I<0xD9, MRM5m,                   // X87 control world = [mem16]
-                  (outs), (ins i16mem:$dst), "fldcw\t$dst", [], IIC_FLDCW>;
+                  (outs), (ins i16mem:$dst), "fldcw\t$dst", [], IIC_FLDCW>,
+                Sched<[WriteLoad]>;
 
 // FPU control instructions
+let SchedRW = [WriteMicrocoded] in {
 let Defs = [FPSW] in
 def FNINIT : I<0xE3, RawFrm, (outs), (ins), "fninit", [], IIC_FNINIT>, DB;
 def FFREE : FPI<0xC0, AddRegFrm, (outs), (ins RST:$reg),
                 "ffree\t$reg", IIC_FFREE>, DD;
-
 // Clear exceptions
 
 let Defs = [FPSW] in
 def FNCLEX : I<0xE2, RawFrm, (outs), (ins), "fnclex", [], IIC_FNCLEX>, DB;
+} // SchedRW
 
 // Operandless floating-point instructions for the disassembler.
+let SchedRW = [WriteMicrocoded] in {
 def WAIT : I<0x9B, RawFrm, (outs), (ins), "wait", [], IIC_WAIT>;
 
 def FNOP : I<0xD0, RawFrm, (outs), (ins), "fnop", [], IIC_FNOP>, D9;
@@ -627,6 +638,7 @@ def FXRSTOR : I<0xAE, MRM1m, (outs), (ins opaque512mem:$src),
 def FXRSTOR64 : I<0xAE, MRM1m, (outs), (ins opaque512mem:$src),
                   "fxrstorq\t$src", [], IIC_FXRSTOR>, TB, REX_W,
                   Requires<[In64BitMode]>;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // Non-Instruction Patterns
diff --git a/lib/Target/X86/X86InstrFormats.td b/lib/Target/X86/X86InstrFormats.td
index 268e9fc9c017..0ef9491eb7fc 100644
--- a/lib/Target/X86/X86InstrFormats.td
+++ b/lib/Target/X86/X86InstrFormats.td
@@ -45,14 +45,15 @@ def MRM_D0 : Format<45>;
 def MRM_D1 : Format<46>;
 def MRM_D4 : Format<47>;
 def MRM_D5 : Format<48>;
-def MRM_D8 : Format<49>;
-def MRM_D9 : Format<50>;
-def MRM_DA : Format<51>;
-def MRM_DB : Format<52>;
-def MRM_DC : Format<53>;
-def MRM_DD : Format<54>;
-def MRM_DE : Format<55>;
-def MRM_DF : Format<56>;
+def MRM_D6 : Format<49>;
+def MRM_D8 : Format<50>;
+def MRM_D9 : Format<51>;
+def MRM_DA : Format<52>;
+def MRM_DB : Format<53>;
+def MRM_DC : Format<54>;
+def MRM_DD : Format<55>;
+def MRM_DE : Format<56>;
+def MRM_DF : Format<57>;
 
 // ImmType - This specifies the immediate type used by an instruction. This is
 // part of the ad-hoc solution used to emit machine instruction encodings by our
@@ -208,47 +209,47 @@ class PseudoI<dag oops, dag iops, list<dag> pattern>
 }
 
 class I<bits<8> o, Format f, dag outs, dag ins, string asm,
-        list<dag> pattern, InstrItinClass itin = IIC_DEFAULT,
+        list<dag> pattern, InstrItinClass itin = NoItinerary,
         Domain d = GenericDomain>
   : X86Inst<o, f, NoImm, outs, ins, asm, itin, d> {
   let Pattern = pattern;
   let CodeSize = 3;
 }
 class Ii8 <bits<8> o, Format f, dag outs, dag ins, string asm, 
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT,
+           list<dag> pattern, InstrItinClass itin = NoItinerary,
            Domain d = GenericDomain>
   : X86Inst<o, f, Imm8, outs, ins, asm, itin, d> {
   let Pattern = pattern;
   let CodeSize = 3;
 }
 class Ii8PCRel<bits<8> o, Format f, dag outs, dag ins, string asm, 
-               list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+               list<dag> pattern, InstrItinClass itin = NoItinerary>
   : X86Inst<o, f, Imm8PCRel, outs, ins, asm, itin> {
   let Pattern = pattern;
   let CodeSize = 3;
 }
 class Ii16<bits<8> o, Format f, dag outs, dag ins, string asm, 
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
   : X86Inst<o, f, Imm16, outs, ins, asm, itin> {
   let Pattern = pattern;
   let CodeSize = 3;
 }
 class Ii32<bits<8> o, Format f, dag outs, dag ins, string asm, 
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
   : X86Inst<o, f, Imm32, outs, ins, asm, itin> {
   let Pattern = pattern;
   let CodeSize = 3;
 }
 
 class Ii16PCRel<bits<8> o, Format f, dag outs, dag ins, string asm, 
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
            : X86Inst<o, f, Imm16PCRel, outs, ins, asm, itin> {
   let Pattern = pattern;
   let CodeSize = 3;
 }
 
 class Ii32PCRel<bits<8> o, Format f, dag outs, dag ins, string asm, 
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
   : X86Inst<o, f, Imm32PCRel, outs, ins, asm, itin> {
   let Pattern = pattern;
   let CodeSize = 3;
@@ -257,12 +258,12 @@ class Ii32PCRel<bits<8> o, Format f, dag outs, dag ins, string asm,
 // FPStack Instruction Templates:
 // FPI - Floating Point Instruction template.
 class FPI<bits<8> o, Format F, dag outs, dag ins, string asm,
-          InstrItinClass itin = IIC_DEFAULT>
+          InstrItinClass itin = NoItinerary>
   : I<o, F, outs, ins, asm, [], itin> {}
 
 // FpI_ - Floating Point Pseudo Instruction template. Not Predicated.
 class FpI_<dag outs, dag ins, FPFormat fp, list<dag> pattern,
-           InstrItinClass itin = IIC_DEFAULT>
+           InstrItinClass itin = NoItinerary>
   : X86Inst<0, Pseudo, NoImm, outs, ins, "", itin> {
   let FPForm = fp;
   let Pattern = pattern;
@@ -275,14 +276,14 @@ class FpI_<dag outs, dag ins, FPFormat fp, list<dag> pattern,
 //   Iseg32 - 16-bit segment selector, 32-bit offset
 
 class Iseg16 <bits<8> o, Format f, dag outs, dag ins, string asm, 
-              list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+              list<dag> pattern, InstrItinClass itin = NoItinerary>
       : X86Inst<o, f, Imm16, outs, ins, asm, itin> {
   let Pattern = pattern;
   let CodeSize = 3;
 }
 
 class Iseg32 <bits<8> o, Format f, dag outs, dag ins, string asm, 
-              list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+              list<dag> pattern, InstrItinClass itin = NoItinerary>
       : X86Inst<o, f, Imm32, outs, ins, asm, itin> {
   let Pattern = pattern;
   let CodeSize = 3;
@@ -292,7 +293,7 @@ def __xs : XS;
 
 // SI - SSE 1 & 2 scalar instructions
 class SI<bits<8> o, Format F, dag outs, dag ins, string asm,
-         list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+         list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin> {
   let Predicates = !if(hasVEXPrefix /* VEX */, [HasAVX],
             !if(!eq(Prefix, __xs.Prefix), [UseSSE1], [UseSSE2]));
@@ -303,7 +304,7 @@ class SI<bits<8> o, Format F, dag outs, dag ins, string asm,
 
 // SIi8 - SSE 1 & 2 scalar instructions
 class SIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin> {
   let Predicates = !if(hasVEXPrefix /* VEX */, [HasAVX],
             !if(!eq(Prefix, __xs.Prefix), [UseSSE1], [UseSSE2]));
@@ -350,25 +351,25 @@ class PIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
 //   VPSI  - SSE1 instructions with TB prefix in AVX form.
 
 class SSI<bits<8> o, Format F, dag outs, dag ins, string asm,
-          list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+          list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, XS, Requires<[UseSSE1]>;
 class SSIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin>, XS, Requires<[UseSSE1]>;
 class PSI<bits<8> o, Format F, dag outs, dag ins, string asm,
-          list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+          list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedSingle>, TB,
         Requires<[UseSSE1]>;
 class PSIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedSingle>, TB,
         Requires<[UseSSE1]>;
 class VSSI<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin>, XS,
         Requires<[HasAVX]>;
 class VPSI<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin, SSEPackedSingle>, TB,
         Requires<[HasAVX]>;
 
@@ -388,42 +389,42 @@ class VPSI<bits<8> o, Format F, dag outs, dag ins, string asm,
 //               MMX operands.
 
 class SDI<bits<8> o, Format F, dag outs, dag ins, string asm,
-          list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+          list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, XD, Requires<[UseSSE2]>;
 class SDIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin>, XD, Requires<[UseSSE2]>;
 class S2SI<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, XS, Requires<[UseSSE2]>;
 class S2SIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-             list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+             list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern>, XS, Requires<[UseSSE2]>;
 class PDI<bits<8> o, Format F, dag outs, dag ins, string asm,
-          list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+          list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>, TB, OpSize,
         Requires<[UseSSE2]>;
 class PDIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>, TB, OpSize,
         Requires<[UseSSE2]>;
 class VSDI<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin>, XD,
         Requires<[HasAVX]>;
 class VS2SI<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin>, XS,
         Requires<[HasAVX]>;
 class VPDI<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin, SSEPackedDouble>, TB,
         OpSize, Requires<[HasAVX]>;
 class MMXSDIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-               list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+               list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin>, XD, Requires<[HasSSE2]>;
 class MMXS2SIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-                list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+                list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern>, XS, Requires<[HasSSE2]>;
 
 // SSE3 Instruction Templates:
@@ -433,15 +434,15 @@ class MMXS2SIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
 //   S3DI  - SSE3 instructions with XD prefix.
 
 class S3SI<bits<8> o, Format F, dag outs, dag ins, string asm, 
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedSingle>, XS,
         Requires<[UseSSE3]>;
 class S3DI<bits<8> o, Format F, dag outs, dag ins, string asm, 
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>, XD,
         Requires<[UseSSE3]>;
 class S3I<bits<8> o, Format F, dag outs, dag ins, string asm,
-          list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+          list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>, TB, OpSize,
         Requires<[UseSSE3]>;
 
@@ -458,19 +459,19 @@ class S3I<bits<8> o, Format F, dag outs, dag ins, string asm,
 // classes. They need to be enabled even if AVX is enabled.
 
 class SS38I<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8,
         Requires<[UseSSSE3]>;
 class SS3AI<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA,
         Requires<[UseSSSE3]>;
 class MMXSS38I<bits<8> o, Format F, dag outs, dag ins, string asm,
-               list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+               list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8,
         Requires<[HasSSSE3]>;
 class MMXSS3AI<bits<8> o, Format F, dag outs, dag ins, string asm,
-               list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+               list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA,
         Requires<[HasSSSE3]>;
 
@@ -480,11 +481,11 @@ class MMXSS3AI<bits<8> o, Format F, dag outs, dag ins, string asm,
 //   SS41AIi8 - SSE 4.1 instructions with TA prefix and ImmT == Imm8.
 //
 class SS48I<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8,
         Requires<[UseSSE41]>;
 class SS4AIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA,
         Requires<[UseSSE41]>;
 
@@ -492,19 +493,19 @@ class SS4AIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
 // 
 //   SS428I - SSE 4.2 instructions with T8 prefix.
 class SS428I<bits<8> o, Format F, dag outs, dag ins, string asm,
-             list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+             list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8,
         Requires<[UseSSE42]>;
 
 //   SS42FI - SSE 4.2 instructions with T8XD prefix.
 // NOTE: 'HasSSE42' is used as SS42FI is only used for CRC32 insns.
 class SS42FI<bits<8> o, Format F, dag outs, dag ins, string asm,
-             list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+             list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, T8XD, Requires<[HasSSE42]>;
 
 //   SS42AI = SSE 4.2 instructions with TA prefix
 class SS42AI<bits<8> o, Format F, dag outs, dag ins, string asm,
-             list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+             list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA,
         Requires<[UseSSE42]>;
 
@@ -514,11 +515,11 @@ class SS42AI<bits<8> o, Format F, dag outs, dag ins, string asm,
 //   AVX8I - AVX instructions with T8 and OpSize prefix.
 //   AVXAIi8 - AVX instructions with TA, OpSize prefix and ImmT = Imm8.
 class AVX8I<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8, OpSize,
         Requires<[HasAVX]>;
 class AVXAIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-              list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+              list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, OpSize,
         Requires<[HasAVX]>;
 
@@ -528,11 +529,11 @@ class AVXAIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
 //   AVX28I - AVX2 instructions with T8 and OpSize prefix.
 //   AVX2AIi8 - AVX2 instructions with TA, OpSize prefix and ImmT = Imm8.
 class AVX28I<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8, OpSize,
         Requires<[HasAVX2]>;
 class AVX2AIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-              list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+              list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, OpSize,
         Requires<[HasAVX2]>;
 
@@ -541,53 +542,53 @@ class AVX2AIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
 // AES8I
 // These use the same encoding as the SSE4.2 T8 and TA encodings.
 class AES8I<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag>pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag>pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8,
         Requires<[HasAES]>;
 
 class AESAI<bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA,
         Requires<[HasAES]>;
 
 // PCLMUL Instruction Templates
 class PCLMULIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-               list<dag>pattern, InstrItinClass itin = IIC_DEFAULT>
+               list<dag>pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA,
         OpSize, Requires<[HasPCLMUL]>;
 
 class AVXPCLMULIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-                  list<dag>pattern, InstrItinClass itin = IIC_DEFAULT>
+                  list<dag>pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA,
         OpSize, VEX_4V, Requires<[HasAVX, HasPCLMUL]>;
 
 // FMA3 Instruction Templates
 class FMA3<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag>pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag>pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, T8,
-        OpSize, VEX_4V, Requires<[HasFMA]>;
+        OpSize, VEX_4V, FMASC, Requires<[HasFMA]>;
 
 // FMA4 Instruction Templates
 class FMA4<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag>pattern, InstrItinClass itin = IIC_DEFAULT>
-      : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA,
-        OpSize, VEX_4V, VEX_I8IMM, Requires<[HasFMA4]>;
+           list<dag>pattern, InstrItinClass itin = NoItinerary>
+      : Ii8<o, F, outs, ins, asm, pattern, itin>, TA,
+        OpSize, VEX_4V, VEX_I8IMM, FMASC, Requires<[HasFMA4]>;
 
 // XOP 2, 3 and 4 Operand Instruction Template
 class IXOP<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>,
          XOP, XOP9, Requires<[HasXOP]>;
 
 // XOP 2, 3 and 4 Operand Instruction Templates with imm byte
 class IXOPi8<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>,
          XOP, XOP8, Requires<[HasXOP]>;
 
 //  XOP 5 operand instruction (VEX encoding!)
 class IXOP5<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag>pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag>pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA,
         OpSize, VEX_4V, VEX_I8IMM, Requires<[HasXOP]>;
 
@@ -595,33 +596,33 @@ class IXOP5<bits<8> o, Format F, dag outs, dag ins, string asm,
 //
 
 class RI<bits<8> o, Format F, dag outs, dag ins, string asm,
-         list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+         list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, REX_W;
 class RIi8 <bits<8> o, Format F, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin>, REX_W;
 class RIi32 <bits<8> o, Format F, dag outs, dag ins, string asm,
-             list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+             list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii32<o, F, outs, ins, asm, pattern, itin>, REX_W;
 
 class RIi64<bits<8> o, Format f, dag outs, dag ins, string asm,
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
   : X86Inst<o, f, Imm64, outs, ins, asm, itin>, REX_W {
   let Pattern = pattern;
   let CodeSize = 3;
 }
 
 class RSSI<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : SSI<o, F, outs, ins, asm, pattern, itin>, REX_W;
 class RSDI<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : SDI<o, F, outs, ins, asm, pattern, itin>, REX_W;
 class RPDI<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : PDI<o, F, outs, ins, asm, pattern, itin>, REX_W;
 class VRPDI<bits<8> o, Format F, dag outs, dag ins, string asm,
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : VPDI<o, F, outs, ins, asm, pattern, itin>, VEX_W;
 
 // MMX Instruction templates
@@ -635,23 +636,23 @@ class VRPDI<bits<8> o, Format F, dag outs, dag ins, string asm,
 // MMXID  - MMX instructions with XD prefix.
 // MMXIS  - MMX instructions with XS prefix.
 class MMXI<bits<8> o, Format F, dag outs, dag ins, string asm, 
-           list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+           list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, TB, Requires<[HasMMX]>;
 class MMXI64<bits<8> o, Format F, dag outs, dag ins, string asm, 
-             list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+             list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, TB, Requires<[HasMMX,In64BitMode]>;
 class MMXRI<bits<8> o, Format F, dag outs, dag ins, string asm, 
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, TB, REX_W, Requires<[HasMMX]>;
 class MMX2I<bits<8> o, Format F, dag outs, dag ins, string asm, 
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, TB, OpSize, Requires<[HasMMX]>;
 class MMXIi8<bits<8> o, Format F, dag outs, dag ins, string asm, 
-             list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+             list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin>, TB, Requires<[HasMMX]>;
 class MMXID<bits<8> o, Format F, dag outs, dag ins, string asm, 
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin>, XD, Requires<[HasMMX]>;
 class MMXIS<bits<8> o, Format F, dag outs, dag ins, string asm, 
-            list<dag> pattern, InstrItinClass itin = IIC_DEFAULT>
+            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin>, XS, Requires<[HasMMX]>;
diff --git a/lib/Target/X86/X86InstrFragmentsSIMD.td b/lib/Target/X86/X86InstrFragmentsSIMD.td
index 73ba0011df1b..2a72fb6f7b2a 100644
--- a/lib/Target/X86/X86InstrFragmentsSIMD.td
+++ b/lib/Target/X86/X86InstrFragmentsSIMD.td
@@ -27,6 +27,11 @@ def SDTX86FPShiftOp : SDTypeProfile<1, 2, [ SDTCisSameAs<0, 1>,
 def SDTX86VFCMP : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<1, 2>,
                                        SDTCisFP<1>, SDTCisVT<3, i8>]>;
 
+def X86umin    : SDNode<"X86ISD::UMIN",      SDTIntBinOp>;
+def X86umax    : SDNode<"X86ISD::UMAX",      SDTIntBinOp>;
+def X86smin    : SDNode<"X86ISD::SMIN",      SDTIntBinOp>;
+def X86smax    : SDNode<"X86ISD::SMAX",      SDTIntBinOp>;
+
 def X86fmin    : SDNode<"X86ISD::FMIN",      SDTFPBinOp>;
 def X86fmax    : SDNode<"X86ISD::FMAX",      SDTFPBinOp>;
 
@@ -128,6 +133,7 @@ def X86vsrai   : SDNode<"X86ISD::VSRAI", SDTIntShiftOp>;
 def SDTX86CmpPTest : SDTypeProfile<1, 2, [SDTCisVT<0, i32>,
                                           SDTCisVec<1>,
                                           SDTCisSameAs<2, 1>]>;
+def X86subus   : SDNode<"X86ISD::SUBUS", SDTIntBinOp>;
 def X86ptest   : SDNode<"X86ISD::PTEST", SDTX86CmpPTest>;
 def X86testp   : SDNode<"X86ISD::TESTP", SDTX86CmpPTest>;
 
@@ -154,7 +160,7 @@ def SDTBlend : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
 def SDTFma : SDTypeProfile<1, 3, [SDTCisSameAs<0,1>,
                            SDTCisSameAs<1,2>, SDTCisSameAs<1,3>]>;
 
-def X86PAlign : SDNode<"X86ISD::PALIGN", SDTShuff3OpI>;
+def X86PAlignr : SDNode<"X86ISD::PALIGNR", SDTShuff3OpI>;
 
 def X86PShufd  : SDNode<"X86ISD::PSHUFD", SDTShuff2OpI>;
 def X86PShufhw : SDNode<"X86ISD::PSHUFHW", SDTShuff2OpI>;
@@ -187,9 +193,7 @@ def X86VPerm2x128 : SDNode<"X86ISD::VPERM2X128", SDTShuff3OpI>;
 
 def X86VBroadcast : SDNode<"X86ISD::VBROADCAST", SDTVBroadcast>;
 
-def X86Blendpw   : SDNode<"X86ISD::BLENDPW",   SDTBlend>;
-def X86Blendps   : SDNode<"X86ISD::BLENDPS",   SDTBlend>;
-def X86Blendpd   : SDNode<"X86ISD::BLENDPD",   SDTBlend>;
+def X86Blendi    : SDNode<"X86ISD::BLENDI",   SDTBlend>;
 def X86Fmadd     : SDNode<"X86ISD::FMADD",     SDTFma>;
 def X86Fnmadd    : SDNode<"X86ISD::FNMADD",    SDTFma>;
 def X86Fmsub     : SDNode<"X86ISD::FMSUB",     SDTFma>;
diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 5a99ff004d48..7ba542c87520 100644
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -17,15 +17,15 @@
 #include "X86MachineFunctionInfo.h"
 #include "X86Subtarget.h"
 #include "X86TargetMachine.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/LLVMContext.h"
 #include "llvm/ADT/STLExtras.h"
+#include "llvm/CodeGen/LiveVariables.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/LiveVariables.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/LLVMContext.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/Support/CommandLine.h"
@@ -297,7 +297,7 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::DIV32r,      X86::DIV32m,        TB_FOLDED_LOAD },
     { X86::DIV64r,      X86::DIV64m,        TB_FOLDED_LOAD },
     { X86::DIV8r,       X86::DIV8m,         TB_FOLDED_LOAD },
-    { X86::EXTRACTPSrr, X86::EXTRACTPSmr,   TB_FOLDED_STORE | TB_ALIGN_16 },
+    { X86::EXTRACTPSrr, X86::EXTRACTPSmr,   TB_FOLDED_STORE },
     { X86::FsMOVAPDrr,  X86::MOVSDmr,       TB_FOLDED_STORE | TB_NO_REVERSE },
     { X86::FsMOVAPSrr,  X86::MOVSSmr,       TB_FOLDED_STORE | TB_NO_REVERSE },
     { X86::IDIV16r,     X86::IDIV16m,       TB_FOLDED_LOAD },
@@ -355,7 +355,7 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::TEST64ri32,  X86::TEST64mi32,    TB_FOLDED_LOAD },
     { X86::TEST8ri,     X86::TEST8mi,       TB_FOLDED_LOAD },
     // AVX 128-bit versions of foldable instructions
-    { X86::VEXTRACTPSrr,X86::VEXTRACTPSmr,  TB_FOLDED_STORE | TB_ALIGN_16 },
+    { X86::VEXTRACTPSrr,X86::VEXTRACTPSmr,  TB_FOLDED_STORE  },
     { X86::FsVMOVAPDrr, X86::VMOVSDmr,      TB_FOLDED_STORE | TB_NO_REVERSE },
     { X86::FsVMOVAPSrr, X86::VMOVSSmr,      TB_FOLDED_STORE | TB_NO_REVERSE },
     { X86::VEXTRACTF128rr, X86::VEXTRACTF128mr, TB_FOLDED_STORE | TB_ALIGN_16 },
@@ -467,9 +467,7 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::RSQRTSSr,        X86::RSQRTSSm,            0 },
     { X86::RSQRTSSr_Int,    X86::RSQRTSSm_Int,        0 },
     { X86::SQRTPDr,         X86::SQRTPDm,             TB_ALIGN_16 },
-    { X86::SQRTPDr_Int,     X86::SQRTPDm_Int,         TB_ALIGN_16 },
     { X86::SQRTPSr,         X86::SQRTPSm,             TB_ALIGN_16 },
-    { X86::SQRTPSr_Int,     X86::SQRTPSm_Int,         TB_ALIGN_16 },
     { X86::SQRTSDr,         X86::SQRTSDm,             0 },
     { X86::SQRTSDr_Int,     X86::SQRTSDm_Int,         0 },
     { X86::SQRTSSr,         X86::SQRTSSm,             0 },
@@ -510,27 +508,25 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::VMOVDQArr,       X86::VMOVDQArm,           TB_ALIGN_16 },
     { X86::VMOVSLDUPrr,     X86::VMOVSLDUPrm,         TB_ALIGN_16 },
     { X86::VMOVSHDUPrr,     X86::VMOVSHDUPrm,         TB_ALIGN_16 },
-    { X86::VMOVUPDrr,       X86::VMOVUPDrm,           TB_ALIGN_16 },
+    { X86::VMOVUPDrr,       X86::VMOVUPDrm,           0 },
     { X86::VMOVUPSrr,       X86::VMOVUPSrm,           0 },
     { X86::VMOVZDI2PDIrr,   X86::VMOVZDI2PDIrm,       0 },
     { X86::VMOVZQI2PQIrr,   X86::VMOVZQI2PQIrm,       0 },
     { X86::VMOVZPQILo2PQIrr,X86::VMOVZPQILo2PQIrm,    TB_ALIGN_16 },
-    { X86::VPABSBrr128,     X86::VPABSBrm128,         TB_ALIGN_16 },
-    { X86::VPABSDrr128,     X86::VPABSDrm128,         TB_ALIGN_16 },
-    { X86::VPABSWrr128,     X86::VPABSWrm128,         TB_ALIGN_16 },
-    { X86::VPERMILPDri,     X86::VPERMILPDmi,         TB_ALIGN_16 },
-    { X86::VPERMILPSri,     X86::VPERMILPSmi,         TB_ALIGN_16 },
-    { X86::VPSHUFDri,       X86::VPSHUFDmi,           TB_ALIGN_16 },
-    { X86::VPSHUFHWri,      X86::VPSHUFHWmi,          TB_ALIGN_16 },
-    { X86::VPSHUFLWri,      X86::VPSHUFLWmi,          TB_ALIGN_16 },
-    { X86::VRCPPSr,         X86::VRCPPSm,             TB_ALIGN_16 },
-    { X86::VRCPPSr_Int,     X86::VRCPPSm_Int,         TB_ALIGN_16 },
-    { X86::VRSQRTPSr,       X86::VRSQRTPSm,           TB_ALIGN_16 },
-    { X86::VRSQRTPSr_Int,   X86::VRSQRTPSm_Int,       TB_ALIGN_16 },
-    { X86::VSQRTPDr,        X86::VSQRTPDm,            TB_ALIGN_16 },
-    { X86::VSQRTPDr_Int,    X86::VSQRTPDm_Int,        TB_ALIGN_16 },
-    { X86::VSQRTPSr,        X86::VSQRTPSm,            TB_ALIGN_16 },
-    { X86::VSQRTPSr_Int,    X86::VSQRTPSm_Int,        TB_ALIGN_16 },
+    { X86::VPABSBrr128,     X86::VPABSBrm128,         0 },
+    { X86::VPABSDrr128,     X86::VPABSDrm128,         0 },
+    { X86::VPABSWrr128,     X86::VPABSWrm128,         0 },
+    { X86::VPERMILPDri,     X86::VPERMILPDmi,         0 },
+    { X86::VPERMILPSri,     X86::VPERMILPSmi,         0 },
+    { X86::VPSHUFDri,       X86::VPSHUFDmi,           0 },
+    { X86::VPSHUFHWri,      X86::VPSHUFHWmi,          0 },
+    { X86::VPSHUFLWri,      X86::VPSHUFLWmi,          0 },
+    { X86::VRCPPSr,         X86::VRCPPSm,             0 },
+    { X86::VRCPPSr_Int,     X86::VRCPPSm_Int,         0 },
+    { X86::VRSQRTPSr,       X86::VRSQRTPSm,           0 },
+    { X86::VRSQRTPSr_Int,   X86::VRSQRTPSm_Int,       0 },
+    { X86::VSQRTPDr,        X86::VSQRTPDm,            0 },
+    { X86::VSQRTPSr,        X86::VSQRTPSm,            0 },
     { X86::VUCOMISDrr,      X86::VUCOMISDrm,          0 },
     { X86::VUCOMISSrr,      X86::VUCOMISSrm,          0 },
     { X86::VBROADCASTSSrr,  X86::VBROADCASTSSrm,      TB_NO_REVERSE },
@@ -541,28 +537,41 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::VMOVDQAYrr,      X86::VMOVDQAYrm,          TB_ALIGN_32 },
     { X86::VMOVUPDYrr,      X86::VMOVUPDYrm,          0 },
     { X86::VMOVUPSYrr,      X86::VMOVUPSYrm,          0 },
-    { X86::VPERMILPDYri,    X86::VPERMILPDYmi,        TB_ALIGN_32 },
-    { X86::VPERMILPSYri,    X86::VPERMILPSYmi,        TB_ALIGN_32 },
+    { X86::VPERMILPDYri,    X86::VPERMILPDYmi,        0 },
+    { X86::VPERMILPSYri,    X86::VPERMILPSYmi,        0 },
 
     // AVX2 foldable instructions
-    { X86::VPABSBrr256,     X86::VPABSBrm256,         TB_ALIGN_32 },
-    { X86::VPABSDrr256,     X86::VPABSDrm256,         TB_ALIGN_32 },
-    { X86::VPABSWrr256,     X86::VPABSWrm256,         TB_ALIGN_32 },
-    { X86::VPSHUFDYri,      X86::VPSHUFDYmi,          TB_ALIGN_32 },
-    { X86::VPSHUFHWYri,     X86::VPSHUFHWYmi,         TB_ALIGN_32 },
-    { X86::VPSHUFLWYri,     X86::VPSHUFLWYmi,         TB_ALIGN_32 },
-    { X86::VRCPPSYr,        X86::VRCPPSYm,            TB_ALIGN_32 },
-    { X86::VRCPPSYr_Int,    X86::VRCPPSYm_Int,        TB_ALIGN_32 },
-    { X86::VRSQRTPSYr,      X86::VRSQRTPSYm,          TB_ALIGN_32 },
-    { X86::VRSQRTPSYr_Int,  X86::VRSQRTPSYm_Int,      TB_ALIGN_32 },
-    { X86::VSQRTPDYr,       X86::VSQRTPDYm,           TB_ALIGN_32 },
-    { X86::VSQRTPDYr_Int,   X86::VSQRTPDYm_Int,       TB_ALIGN_32 },
-    { X86::VSQRTPSYr,       X86::VSQRTPSYm,           TB_ALIGN_32 },
-    { X86::VSQRTPSYr_Int,   X86::VSQRTPSYm_Int,       TB_ALIGN_32 },
+    { X86::VPABSBrr256,     X86::VPABSBrm256,         0 },
+    { X86::VPABSDrr256,     X86::VPABSDrm256,         0 },
+    { X86::VPABSWrr256,     X86::VPABSWrm256,         0 },
+    { X86::VPSHUFDYri,      X86::VPSHUFDYmi,          0 },
+    { X86::VPSHUFHWYri,     X86::VPSHUFHWYmi,         0 },
+    { X86::VPSHUFLWYri,     X86::VPSHUFLWYmi,         0 },
+    { X86::VRCPPSYr,        X86::VRCPPSYm,            0 },
+    { X86::VRCPPSYr_Int,    X86::VRCPPSYm_Int,        0 },
+    { X86::VRSQRTPSYr,      X86::VRSQRTPSYm,          0 },
+    { X86::VSQRTPDYr,       X86::VSQRTPDYm,           0 },
+    { X86::VSQRTPSYr,       X86::VSQRTPSYm,           0 },
     { X86::VBROADCASTSSYrr, X86::VBROADCASTSSYrm,     TB_NO_REVERSE },
     { X86::VBROADCASTSDYrr, X86::VBROADCASTSDYrm,     TB_NO_REVERSE },
 
-    // BMI/BMI2 foldable instructions
+    // BMI/BMI2/LZCNT/POPCNT foldable instructions
+    { X86::BEXTR32rr,       X86::BEXTR32rm,           0 },
+    { X86::BEXTR64rr,       X86::BEXTR64rm,           0 },
+    { X86::BLSI32rr,        X86::BLSI32rm,            0 },
+    { X86::BLSI64rr,        X86::BLSI64rm,            0 },
+    { X86::BLSMSK32rr,      X86::BLSMSK32rm,          0 },
+    { X86::BLSMSK64rr,      X86::BLSMSK64rm,          0 },
+    { X86::BLSR32rr,        X86::BLSR32rm,            0 },
+    { X86::BLSR64rr,        X86::BLSR64rm,            0 },
+    { X86::BZHI32rr,        X86::BZHI32rm,            0 },
+    { X86::BZHI64rr,        X86::BZHI64rm,            0 },
+    { X86::LZCNT16rr,       X86::LZCNT16rm,           0 },
+    { X86::LZCNT32rr,       X86::LZCNT32rm,           0 },
+    { X86::LZCNT64rr,       X86::LZCNT64rm,           0 },
+    { X86::POPCNT16rr,      X86::POPCNT16rm,          0 },
+    { X86::POPCNT32rr,      X86::POPCNT32rm,          0 },
+    { X86::POPCNT64rr,      X86::POPCNT64rm,          0 },
     { X86::RORX32ri,        X86::RORX32mi,            0 },
     { X86::RORX64ri,        X86::RORX64mi,            0 },
     { X86::SARX32rr,        X86::SARX32rm,            0 },
@@ -571,6 +580,9 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::SHRX64rr,        X86::SHRX64rm,            0 },
     { X86::SHLX32rr,        X86::SHLX32rm,            0 },
     { X86::SHLX64rr,        X86::SHLX64rm,            0 },
+    { X86::TZCNT16rr,       X86::TZCNT16rm,           0 },
+    { X86::TZCNT32rr,       X86::TZCNT32rm,           0 },
+    { X86::TZCNT64rr,       X86::TZCNT64rm,           0 },
   };
 
   for (unsigned i = 0, e = array_lengthof(OpTbl1); i != e; ++i) {
@@ -691,21 +703,13 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::Int_CVTSI2SSrr,  X86::Int_CVTSI2SSrm,      0 },
     { X86::Int_CVTSS2SDrr,  X86::Int_CVTSS2SDrm,      0 },
     { X86::MAXPDrr,         X86::MAXPDrm,       TB_ALIGN_16 },
-    { X86::MAXPDrr_Int,     X86::MAXPDrm_Int,   TB_ALIGN_16 },
     { X86::MAXPSrr,         X86::MAXPSrm,       TB_ALIGN_16 },
-    { X86::MAXPSrr_Int,     X86::MAXPSrm_Int,   TB_ALIGN_16 },
     { X86::MAXSDrr,         X86::MAXSDrm,       0 },
-    { X86::MAXSDrr_Int,     X86::MAXSDrm_Int,   0 },
     { X86::MAXSSrr,         X86::MAXSSrm,       0 },
-    { X86::MAXSSrr_Int,     X86::MAXSSrm_Int,   0 },
     { X86::MINPDrr,         X86::MINPDrm,       TB_ALIGN_16 },
-    { X86::MINPDrr_Int,     X86::MINPDrm_Int,   TB_ALIGN_16 },
     { X86::MINPSrr,         X86::MINPSrm,       TB_ALIGN_16 },
-    { X86::MINPSrr_Int,     X86::MINPSrm_Int,   TB_ALIGN_16 },
     { X86::MINSDrr,         X86::MINSDrm,       0 },
-    { X86::MINSDrr_Int,     X86::MINSDrm_Int,   0 },
     { X86::MINSSrr,         X86::MINSSrm,       0 },
-    { X86::MINSSrr_Int,     X86::MINSSrm_Int,   0 },
     { X86::MPSADBWrri,      X86::MPSADBWrmi,    TB_ALIGN_16 },
     { X86::MULPDrr,         X86::MULPDrm,       TB_ALIGN_16 },
     { X86::MULPSrr,         X86::MULPSrm,       TB_ALIGN_16 },
@@ -756,6 +760,14 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::PMAXUBrr,        X86::PMAXUBrm,      TB_ALIGN_16 },
     { X86::PMINSWrr,        X86::PMINSWrm,      TB_ALIGN_16 },
     { X86::PMINUBrr,        X86::PMINUBrm,      TB_ALIGN_16 },
+    { X86::PMINSBrr,        X86::PMINSBrm,      TB_ALIGN_16 },
+    { X86::PMINSDrr,        X86::PMINSDrm,      TB_ALIGN_16 },
+    { X86::PMINUDrr,        X86::PMINUDrm,      TB_ALIGN_16 },
+    { X86::PMINUWrr,        X86::PMINUWrm,      TB_ALIGN_16 },
+    { X86::PMAXSBrr,        X86::PMAXSBrm,      TB_ALIGN_16 },
+    { X86::PMAXSDrr,        X86::PMAXSDrm,      TB_ALIGN_16 },
+    { X86::PMAXUDrr,        X86::PMAXUDrm,      TB_ALIGN_16 },
+    { X86::PMAXUWrr,        X86::PMAXUWrm,      TB_ALIGN_16 },
     { X86::PMULDQrr,        X86::PMULDQrm,      TB_ALIGN_16 },
     { X86::PMULHRSWrr128,   X86::PMULHRSWrm128, TB_ALIGN_16 },
     { X86::PMULHUWrr,       X86::PMULHUWrm,     TB_ALIGN_16 },
@@ -827,31 +839,31 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::Int_VCVTSI2SSrr,   X86::Int_VCVTSI2SSrm,    0 },
     { X86::VCVTSS2SDrr,       X86::VCVTSS2SDrm,        0 },
     { X86::Int_VCVTSS2SDrr,   X86::Int_VCVTSS2SDrm,    0 },
-    { X86::VCVTTPD2DQrr,      X86::VCVTTPD2DQXrm,      TB_ALIGN_16 },
-    { X86::VCVTTPS2DQrr,      X86::VCVTTPS2DQrm,       TB_ALIGN_16 },
+    { X86::VCVTTPD2DQrr,      X86::VCVTTPD2DQXrm,      0 },
+    { X86::VCVTTPS2DQrr,      X86::VCVTTPS2DQrm,       0 },
     { X86::VRSQRTSSr,         X86::VRSQRTSSm,          0 },
     { X86::VSQRTSDr,          X86::VSQRTSDm,           0 },
     { X86::VSQRTSSr,          X86::VSQRTSSm,           0 },
-    { X86::VADDPDrr,          X86::VADDPDrm,           TB_ALIGN_16 },
-    { X86::VADDPSrr,          X86::VADDPSrm,           TB_ALIGN_16 },
+    { X86::VADDPDrr,          X86::VADDPDrm,           0 },
+    { X86::VADDPSrr,          X86::VADDPSrm,           0 },
     { X86::VADDSDrr,          X86::VADDSDrm,           0 },
     { X86::VADDSSrr,          X86::VADDSSrm,           0 },
-    { X86::VADDSUBPDrr,       X86::VADDSUBPDrm,        TB_ALIGN_16 },
-    { X86::VADDSUBPSrr,       X86::VADDSUBPSrm,        TB_ALIGN_16 },
-    { X86::VANDNPDrr,         X86::VANDNPDrm,          TB_ALIGN_16 },
-    { X86::VANDNPSrr,         X86::VANDNPSrm,          TB_ALIGN_16 },
-    { X86::VANDPDrr,          X86::VANDPDrm,           TB_ALIGN_16 },
-    { X86::VANDPSrr,          X86::VANDPSrm,           TB_ALIGN_16 },
-    { X86::VBLENDPDrri,       X86::VBLENDPDrmi,        TB_ALIGN_16 },
-    { X86::VBLENDPSrri,       X86::VBLENDPSrmi,        TB_ALIGN_16 },
-    { X86::VBLENDVPDrr,       X86::VBLENDVPDrm,        TB_ALIGN_16 },
-    { X86::VBLENDVPSrr,       X86::VBLENDVPSrm,        TB_ALIGN_16 },
-    { X86::VCMPPDrri,         X86::VCMPPDrmi,          TB_ALIGN_16 },
-    { X86::VCMPPSrri,         X86::VCMPPSrmi,          TB_ALIGN_16 },
+    { X86::VADDSUBPDrr,       X86::VADDSUBPDrm,        0 },
+    { X86::VADDSUBPSrr,       X86::VADDSUBPSrm,        0 },
+    { X86::VANDNPDrr,         X86::VANDNPDrm,          0 },
+    { X86::VANDNPSrr,         X86::VANDNPSrm,          0 },
+    { X86::VANDPDrr,          X86::VANDPDrm,           0 },
+    { X86::VANDPSrr,          X86::VANDPSrm,           0 },
+    { X86::VBLENDPDrri,       X86::VBLENDPDrmi,        0 },
+    { X86::VBLENDPSrri,       X86::VBLENDPSrmi,        0 },
+    { X86::VBLENDVPDrr,       X86::VBLENDVPDrm,        0 },
+    { X86::VBLENDVPSrr,       X86::VBLENDVPSrm,        0 },
+    { X86::VCMPPDrri,         X86::VCMPPDrmi,          0 },
+    { X86::VCMPPSrri,         X86::VCMPPSrmi,          0 },
     { X86::VCMPSDrr,          X86::VCMPSDrm,           0 },
     { X86::VCMPSSrr,          X86::VCMPSSrm,           0 },
-    { X86::VDIVPDrr,          X86::VDIVPDrm,           TB_ALIGN_16 },
-    { X86::VDIVPSrr,          X86::VDIVPSrm,           TB_ALIGN_16 },
+    { X86::VDIVPDrr,          X86::VDIVPDrm,           0 },
+    { X86::VDIVPSrr,          X86::VDIVPSrm,           0 },
     { X86::VDIVSDrr,          X86::VDIVSDrm,           0 },
     { X86::VDIVSSrr,          X86::VDIVSSrm,           0 },
     { X86::VFsANDNPDrr,       X86::VFsANDNPDrm,        TB_ALIGN_16 },
@@ -862,263 +874,267 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::VFsORPSrr,         X86::VFsORPSrm,          TB_ALIGN_16 },
     { X86::VFsXORPDrr,        X86::VFsXORPDrm,         TB_ALIGN_16 },
     { X86::VFsXORPSrr,        X86::VFsXORPSrm,         TB_ALIGN_16 },
-    { X86::VHADDPDrr,         X86::VHADDPDrm,          TB_ALIGN_16 },
-    { X86::VHADDPSrr,         X86::VHADDPSrm,          TB_ALIGN_16 },
-    { X86::VHSUBPDrr,         X86::VHSUBPDrm,          TB_ALIGN_16 },
-    { X86::VHSUBPSrr,         X86::VHSUBPSrm,          TB_ALIGN_16 },
+    { X86::VHADDPDrr,         X86::VHADDPDrm,          0 },
+    { X86::VHADDPSrr,         X86::VHADDPSrm,          0 },
+    { X86::VHSUBPDrr,         X86::VHSUBPDrm,          0 },
+    { X86::VHSUBPSrr,         X86::VHSUBPSrm,          0 },
     { X86::Int_VCMPSDrr,      X86::Int_VCMPSDrm,       0 },
     { X86::Int_VCMPSSrr,      X86::Int_VCMPSSrm,       0 },
-    { X86::VMAXPDrr,          X86::VMAXPDrm,           TB_ALIGN_16 },
-    { X86::VMAXPDrr_Int,      X86::VMAXPDrm_Int,       TB_ALIGN_16 },
-    { X86::VMAXPSrr,          X86::VMAXPSrm,           TB_ALIGN_16 },
-    { X86::VMAXPSrr_Int,      X86::VMAXPSrm_Int,       TB_ALIGN_16 },
+    { X86::VMAXPDrr,          X86::VMAXPDrm,           0 },
+    { X86::VMAXPSrr,          X86::VMAXPSrm,           0 },
     { X86::VMAXSDrr,          X86::VMAXSDrm,           0 },
-    { X86::VMAXSDrr_Int,      X86::VMAXSDrm_Int,       0 },
     { X86::VMAXSSrr,          X86::VMAXSSrm,           0 },
-    { X86::VMAXSSrr_Int,      X86::VMAXSSrm_Int,       0 },
-    { X86::VMINPDrr,          X86::VMINPDrm,           TB_ALIGN_16 },
-    { X86::VMINPDrr_Int,      X86::VMINPDrm_Int,       TB_ALIGN_16 },
-    { X86::VMINPSrr,          X86::VMINPSrm,           TB_ALIGN_16 },
-    { X86::VMINPSrr_Int,      X86::VMINPSrm_Int,       TB_ALIGN_16 },
+    { X86::VMINPDrr,          X86::VMINPDrm,           0 },
+    { X86::VMINPSrr,          X86::VMINPSrm,           0 },
     { X86::VMINSDrr,          X86::VMINSDrm,           0 },
-    { X86::VMINSDrr_Int,      X86::VMINSDrm_Int,       0 },
     { X86::VMINSSrr,          X86::VMINSSrm,           0 },
-    { X86::VMINSSrr_Int,      X86::VMINSSrm_Int,       0 },
-    { X86::VMPSADBWrri,       X86::VMPSADBWrmi,        TB_ALIGN_16 },
-    { X86::VMULPDrr,          X86::VMULPDrm,           TB_ALIGN_16 },
-    { X86::VMULPSrr,          X86::VMULPSrm,           TB_ALIGN_16 },
+    { X86::VMPSADBWrri,       X86::VMPSADBWrmi,        0 },
+    { X86::VMULPDrr,          X86::VMULPDrm,           0 },
+    { X86::VMULPSrr,          X86::VMULPSrm,           0 },
     { X86::VMULSDrr,          X86::VMULSDrm,           0 },
     { X86::VMULSSrr,          X86::VMULSSrm,           0 },
-    { X86::VORPDrr,           X86::VORPDrm,            TB_ALIGN_16 },
-    { X86::VORPSrr,           X86::VORPSrm,            TB_ALIGN_16 },
-    { X86::VPACKSSDWrr,       X86::VPACKSSDWrm,        TB_ALIGN_16 },
-    { X86::VPACKSSWBrr,       X86::VPACKSSWBrm,        TB_ALIGN_16 },
-    { X86::VPACKUSDWrr,       X86::VPACKUSDWrm,        TB_ALIGN_16 },
-    { X86::VPACKUSWBrr,       X86::VPACKUSWBrm,        TB_ALIGN_16 },
-    { X86::VPADDBrr,          X86::VPADDBrm,           TB_ALIGN_16 },
-    { X86::VPADDDrr,          X86::VPADDDrm,           TB_ALIGN_16 },
-    { X86::VPADDQrr,          X86::VPADDQrm,           TB_ALIGN_16 },
-    { X86::VPADDSBrr,         X86::VPADDSBrm,          TB_ALIGN_16 },
-    { X86::VPADDSWrr,         X86::VPADDSWrm,          TB_ALIGN_16 },
-    { X86::VPADDUSBrr,        X86::VPADDUSBrm,         TB_ALIGN_16 },
-    { X86::VPADDUSWrr,        X86::VPADDUSWrm,         TB_ALIGN_16 },
-    { X86::VPADDWrr,          X86::VPADDWrm,           TB_ALIGN_16 },
-    { X86::VPALIGNR128rr,     X86::VPALIGNR128rm,      TB_ALIGN_16 },
-    { X86::VPANDNrr,          X86::VPANDNrm,           TB_ALIGN_16 },
-    { X86::VPANDrr,           X86::VPANDrm,            TB_ALIGN_16 },
-    { X86::VPAVGBrr,          X86::VPAVGBrm,           TB_ALIGN_16 },
-    { X86::VPAVGWrr,          X86::VPAVGWrm,           TB_ALIGN_16 },
-    { X86::VPBLENDWrri,       X86::VPBLENDWrmi,        TB_ALIGN_16 },
-    { X86::VPCMPEQBrr,        X86::VPCMPEQBrm,         TB_ALIGN_16 },
-    { X86::VPCMPEQDrr,        X86::VPCMPEQDrm,         TB_ALIGN_16 },
-    { X86::VPCMPEQQrr,        X86::VPCMPEQQrm,         TB_ALIGN_16 },
-    { X86::VPCMPEQWrr,        X86::VPCMPEQWrm,         TB_ALIGN_16 },
-    { X86::VPCMPGTBrr,        X86::VPCMPGTBrm,         TB_ALIGN_16 },
-    { X86::VPCMPGTDrr,        X86::VPCMPGTDrm,         TB_ALIGN_16 },
-    { X86::VPCMPGTQrr,        X86::VPCMPGTQrm,         TB_ALIGN_16 },
-    { X86::VPCMPGTWrr,        X86::VPCMPGTWrm,         TB_ALIGN_16 },
-    { X86::VPHADDDrr,         X86::VPHADDDrm,          TB_ALIGN_16 },
-    { X86::VPHADDSWrr128,     X86::VPHADDSWrm128,      TB_ALIGN_16 },
-    { X86::VPHADDWrr,         X86::VPHADDWrm,          TB_ALIGN_16 },
-    { X86::VPHSUBDrr,         X86::VPHSUBDrm,          TB_ALIGN_16 },
-    { X86::VPHSUBSWrr128,     X86::VPHSUBSWrm128,      TB_ALIGN_16 },
-    { X86::VPHSUBWrr,         X86::VPHSUBWrm,          TB_ALIGN_16 },
-    { X86::VPERMILPDrr,       X86::VPERMILPDrm,        TB_ALIGN_16 },
-    { X86::VPERMILPSrr,       X86::VPERMILPSrm,        TB_ALIGN_16 },
-    { X86::VPINSRWrri,        X86::VPINSRWrmi,         TB_ALIGN_16 },
-    { X86::VPMADDUBSWrr128,   X86::VPMADDUBSWrm128,    TB_ALIGN_16 },
-    { X86::VPMADDWDrr,        X86::VPMADDWDrm,         TB_ALIGN_16 },
-    { X86::VPMAXSWrr,         X86::VPMAXSWrm,          TB_ALIGN_16 },
-    { X86::VPMAXUBrr,         X86::VPMAXUBrm,          TB_ALIGN_16 },
-    { X86::VPMINSWrr,         X86::VPMINSWrm,          TB_ALIGN_16 },
-    { X86::VPMINUBrr,         X86::VPMINUBrm,          TB_ALIGN_16 },
-    { X86::VPMULDQrr,         X86::VPMULDQrm,          TB_ALIGN_16 },
-    { X86::VPMULHRSWrr128,    X86::VPMULHRSWrm128,     TB_ALIGN_16 },
-    { X86::VPMULHUWrr,        X86::VPMULHUWrm,         TB_ALIGN_16 },
-    { X86::VPMULHWrr,         X86::VPMULHWrm,          TB_ALIGN_16 },
-    { X86::VPMULLDrr,         X86::VPMULLDrm,          TB_ALIGN_16 },
-    { X86::VPMULLWrr,         X86::VPMULLWrm,          TB_ALIGN_16 },
-    { X86::VPMULUDQrr,        X86::VPMULUDQrm,         TB_ALIGN_16 },
-    { X86::VPORrr,            X86::VPORrm,             TB_ALIGN_16 },
-    { X86::VPSADBWrr,         X86::VPSADBWrm,          TB_ALIGN_16 },
-    { X86::VPSHUFBrr,         X86::VPSHUFBrm,          TB_ALIGN_16 },
-    { X86::VPSIGNBrr,         X86::VPSIGNBrm,          TB_ALIGN_16 },
-    { X86::VPSIGNWrr,         X86::VPSIGNWrm,          TB_ALIGN_16 },
-    { X86::VPSIGNDrr,         X86::VPSIGNDrm,          TB_ALIGN_16 },
-    { X86::VPSLLDrr,          X86::VPSLLDrm,           TB_ALIGN_16 },
-    { X86::VPSLLQrr,          X86::VPSLLQrm,           TB_ALIGN_16 },
-    { X86::VPSLLWrr,          X86::VPSLLWrm,           TB_ALIGN_16 },
-    { X86::VPSRADrr,          X86::VPSRADrm,           TB_ALIGN_16 },
-    { X86::VPSRAWrr,          X86::VPSRAWrm,           TB_ALIGN_16 },
-    { X86::VPSRLDrr,          X86::VPSRLDrm,           TB_ALIGN_16 },
-    { X86::VPSRLQrr,          X86::VPSRLQrm,           TB_ALIGN_16 },
-    { X86::VPSRLWrr,          X86::VPSRLWrm,           TB_ALIGN_16 },
-    { X86::VPSUBBrr,          X86::VPSUBBrm,           TB_ALIGN_16 },
-    { X86::VPSUBDrr,          X86::VPSUBDrm,           TB_ALIGN_16 },
-    { X86::VPSUBSBrr,         X86::VPSUBSBrm,          TB_ALIGN_16 },
-    { X86::VPSUBSWrr,         X86::VPSUBSWrm,          TB_ALIGN_16 },
-    { X86::VPSUBWrr,          X86::VPSUBWrm,           TB_ALIGN_16 },
-    { X86::VPUNPCKHBWrr,      X86::VPUNPCKHBWrm,       TB_ALIGN_16 },
-    { X86::VPUNPCKHDQrr,      X86::VPUNPCKHDQrm,       TB_ALIGN_16 },
-    { X86::VPUNPCKHQDQrr,     X86::VPUNPCKHQDQrm,      TB_ALIGN_16 },
-    { X86::VPUNPCKHWDrr,      X86::VPUNPCKHWDrm,       TB_ALIGN_16 },
-    { X86::VPUNPCKLBWrr,      X86::VPUNPCKLBWrm,       TB_ALIGN_16 },
-    { X86::VPUNPCKLDQrr,      X86::VPUNPCKLDQrm,       TB_ALIGN_16 },
-    { X86::VPUNPCKLQDQrr,     X86::VPUNPCKLQDQrm,      TB_ALIGN_16 },
-    { X86::VPUNPCKLWDrr,      X86::VPUNPCKLWDrm,       TB_ALIGN_16 },
-    { X86::VPXORrr,           X86::VPXORrm,            TB_ALIGN_16 },
-    { X86::VSHUFPDrri,        X86::VSHUFPDrmi,         TB_ALIGN_16 },
-    { X86::VSHUFPSrri,        X86::VSHUFPSrmi,         TB_ALIGN_16 },
-    { X86::VSUBPDrr,          X86::VSUBPDrm,           TB_ALIGN_16 },
-    { X86::VSUBPSrr,          X86::VSUBPSrm,           TB_ALIGN_16 },
+    { X86::VORPDrr,           X86::VORPDrm,            0 },
+    { X86::VORPSrr,           X86::VORPSrm,            0 },
+    { X86::VPACKSSDWrr,       X86::VPACKSSDWrm,        0 },
+    { X86::VPACKSSWBrr,       X86::VPACKSSWBrm,        0 },
+    { X86::VPACKUSDWrr,       X86::VPACKUSDWrm,        0 },
+    { X86::VPACKUSWBrr,       X86::VPACKUSWBrm,        0 },
+    { X86::VPADDBrr,          X86::VPADDBrm,           0 },
+    { X86::VPADDDrr,          X86::VPADDDrm,           0 },
+    { X86::VPADDQrr,          X86::VPADDQrm,           0 },
+    { X86::VPADDSBrr,         X86::VPADDSBrm,          0 },
+    { X86::VPADDSWrr,         X86::VPADDSWrm,          0 },
+    { X86::VPADDUSBrr,        X86::VPADDUSBrm,         0 },
+    { X86::VPADDUSWrr,        X86::VPADDUSWrm,         0 },
+    { X86::VPADDWrr,          X86::VPADDWrm,           0 },
+    { X86::VPALIGNR128rr,     X86::VPALIGNR128rm,      0 },
+    { X86::VPANDNrr,          X86::VPANDNrm,           0 },
+    { X86::VPANDrr,           X86::VPANDrm,            0 },
+    { X86::VPAVGBrr,          X86::VPAVGBrm,           0 },
+    { X86::VPAVGWrr,          X86::VPAVGWrm,           0 },
+    { X86::VPBLENDWrri,       X86::VPBLENDWrmi,        0 },
+    { X86::VPCMPEQBrr,        X86::VPCMPEQBrm,         0 },
+    { X86::VPCMPEQDrr,        X86::VPCMPEQDrm,         0 },
+    { X86::VPCMPEQQrr,        X86::VPCMPEQQrm,         0 },
+    { X86::VPCMPEQWrr,        X86::VPCMPEQWrm,         0 },
+    { X86::VPCMPGTBrr,        X86::VPCMPGTBrm,         0 },
+    { X86::VPCMPGTDrr,        X86::VPCMPGTDrm,         0 },
+    { X86::VPCMPGTQrr,        X86::VPCMPGTQrm,         0 },
+    { X86::VPCMPGTWrr,        X86::VPCMPGTWrm,         0 },
+    { X86::VPHADDDrr,         X86::VPHADDDrm,          0 },
+    { X86::VPHADDSWrr128,     X86::VPHADDSWrm128,      0 },
+    { X86::VPHADDWrr,         X86::VPHADDWrm,          0 },
+    { X86::VPHSUBDrr,         X86::VPHSUBDrm,          0 },
+    { X86::VPHSUBSWrr128,     X86::VPHSUBSWrm128,      0 },
+    { X86::VPHSUBWrr,         X86::VPHSUBWrm,          0 },
+    { X86::VPERMILPDrr,       X86::VPERMILPDrm,        0 },
+    { X86::VPERMILPSrr,       X86::VPERMILPSrm,        0 },
+    { X86::VPINSRWrri,        X86::VPINSRWrmi,         0 },
+    { X86::VPMADDUBSWrr128,   X86::VPMADDUBSWrm128,    0 },
+    { X86::VPMADDWDrr,        X86::VPMADDWDrm,         0 },
+    { X86::VPMAXSWrr,         X86::VPMAXSWrm,          0 },
+    { X86::VPMAXUBrr,         X86::VPMAXUBrm,          0 },
+    { X86::VPMINSWrr,         X86::VPMINSWrm,          0 },
+    { X86::VPMINUBrr,         X86::VPMINUBrm,          0 },
+    { X86::VPMINSBrr,         X86::VPMINSBrm,          0 },
+    { X86::VPMINSDrr,         X86::VPMINSDrm,          0 },
+    { X86::VPMINUDrr,         X86::VPMINUDrm,          0 },
+    { X86::VPMINUWrr,         X86::VPMINUWrm,          0 },
+    { X86::VPMAXSBrr,         X86::VPMAXSBrm,          0 },
+    { X86::VPMAXSDrr,         X86::VPMAXSDrm,          0 },
+    { X86::VPMAXUDrr,         X86::VPMAXUDrm,          0 },
+    { X86::VPMAXUWrr,         X86::VPMAXUWrm,          0 },
+    { X86::VPMULDQrr,         X86::VPMULDQrm,          0 },
+    { X86::VPMULHRSWrr128,    X86::VPMULHRSWrm128,     0 },
+    { X86::VPMULHUWrr,        X86::VPMULHUWrm,         0 },
+    { X86::VPMULHWrr,         X86::VPMULHWrm,          0 },
+    { X86::VPMULLDrr,         X86::VPMULLDrm,          0 },
+    { X86::VPMULLWrr,         X86::VPMULLWrm,          0 },
+    { X86::VPMULUDQrr,        X86::VPMULUDQrm,         0 },
+    { X86::VPORrr,            X86::VPORrm,             0 },
+    { X86::VPSADBWrr,         X86::VPSADBWrm,          0 },
+    { X86::VPSHUFBrr,         X86::VPSHUFBrm,          0 },
+    { X86::VPSIGNBrr,         X86::VPSIGNBrm,          0 },
+    { X86::VPSIGNWrr,         X86::VPSIGNWrm,          0 },
+    { X86::VPSIGNDrr,         X86::VPSIGNDrm,          0 },
+    { X86::VPSLLDrr,          X86::VPSLLDrm,           0 },
+    { X86::VPSLLQrr,          X86::VPSLLQrm,           0 },
+    { X86::VPSLLWrr,          X86::VPSLLWrm,           0 },
+    { X86::VPSRADrr,          X86::VPSRADrm,           0 },
+    { X86::VPSRAWrr,          X86::VPSRAWrm,           0 },
+    { X86::VPSRLDrr,          X86::VPSRLDrm,           0 },
+    { X86::VPSRLQrr,          X86::VPSRLQrm,           0 },
+    { X86::VPSRLWrr,          X86::VPSRLWrm,           0 },
+    { X86::VPSUBBrr,          X86::VPSUBBrm,           0 },
+    { X86::VPSUBDrr,          X86::VPSUBDrm,           0 },
+    { X86::VPSUBSBrr,         X86::VPSUBSBrm,          0 },
+    { X86::VPSUBSWrr,         X86::VPSUBSWrm,          0 },
+    { X86::VPSUBWrr,          X86::VPSUBWrm,           0 },
+    { X86::VPUNPCKHBWrr,      X86::VPUNPCKHBWrm,       0 },
+    { X86::VPUNPCKHDQrr,      X86::VPUNPCKHDQrm,       0 },
+    { X86::VPUNPCKHQDQrr,     X86::VPUNPCKHQDQrm,      0 },
+    { X86::VPUNPCKHWDrr,      X86::VPUNPCKHWDrm,       0 },
+    { X86::VPUNPCKLBWrr,      X86::VPUNPCKLBWrm,       0 },
+    { X86::VPUNPCKLDQrr,      X86::VPUNPCKLDQrm,       0 },
+    { X86::VPUNPCKLQDQrr,     X86::VPUNPCKLQDQrm,      0 },
+    { X86::VPUNPCKLWDrr,      X86::VPUNPCKLWDrm,       0 },
+    { X86::VPXORrr,           X86::VPXORrm,            0 },
+    { X86::VSHUFPDrri,        X86::VSHUFPDrmi,         0 },
+    { X86::VSHUFPSrri,        X86::VSHUFPSrmi,         0 },
+    { X86::VSUBPDrr,          X86::VSUBPDrm,           0 },
+    { X86::VSUBPSrr,          X86::VSUBPSrm,           0 },
     { X86::VSUBSDrr,          X86::VSUBSDrm,           0 },
     { X86::VSUBSSrr,          X86::VSUBSSrm,           0 },
-    { X86::VUNPCKHPDrr,       X86::VUNPCKHPDrm,        TB_ALIGN_16 },
-    { X86::VUNPCKHPSrr,       X86::VUNPCKHPSrm,        TB_ALIGN_16 },
-    { X86::VUNPCKLPDrr,       X86::VUNPCKLPDrm,        TB_ALIGN_16 },
-    { X86::VUNPCKLPSrr,       X86::VUNPCKLPSrm,        TB_ALIGN_16 },
-    { X86::VXORPDrr,          X86::VXORPDrm,           TB_ALIGN_16 },
-    { X86::VXORPSrr,          X86::VXORPSrm,           TB_ALIGN_16 },
+    { X86::VUNPCKHPDrr,       X86::VUNPCKHPDrm,        0 },
+    { X86::VUNPCKHPSrr,       X86::VUNPCKHPSrm,        0 },
+    { X86::VUNPCKLPDrr,       X86::VUNPCKLPDrm,        0 },
+    { X86::VUNPCKLPSrr,       X86::VUNPCKLPSrm,        0 },
+    { X86::VXORPDrr,          X86::VXORPDrm,           0 },
+    { X86::VXORPSrr,          X86::VXORPSrm,           0 },
     // AVX 256-bit foldable instructions
-    { X86::VADDPDYrr,         X86::VADDPDYrm,          TB_ALIGN_32 },
-    { X86::VADDPSYrr,         X86::VADDPSYrm,          TB_ALIGN_32 },
-    { X86::VADDSUBPDYrr,      X86::VADDSUBPDYrm,       TB_ALIGN_32 },
-    { X86::VADDSUBPSYrr,      X86::VADDSUBPSYrm,       TB_ALIGN_32 },
-    { X86::VANDNPDYrr,        X86::VANDNPDYrm,         TB_ALIGN_32 },
-    { X86::VANDNPSYrr,        X86::VANDNPSYrm,         TB_ALIGN_32 },
-    { X86::VANDPDYrr,         X86::VANDPDYrm,          TB_ALIGN_32 },
-    { X86::VANDPSYrr,         X86::VANDPSYrm,          TB_ALIGN_32 },
-    { X86::VBLENDPDYrri,      X86::VBLENDPDYrmi,       TB_ALIGN_32 },
-    { X86::VBLENDPSYrri,      X86::VBLENDPSYrmi,       TB_ALIGN_32 },
-    { X86::VBLENDVPDYrr,      X86::VBLENDVPDYrm,       TB_ALIGN_32 },
-    { X86::VBLENDVPSYrr,      X86::VBLENDVPSYrm,       TB_ALIGN_32 },
-    { X86::VCMPPDYrri,        X86::VCMPPDYrmi,         TB_ALIGN_32 },
-    { X86::VCMPPSYrri,        X86::VCMPPSYrmi,         TB_ALIGN_32 },
-    { X86::VDIVPDYrr,         X86::VDIVPDYrm,          TB_ALIGN_32 },
-    { X86::VDIVPSYrr,         X86::VDIVPSYrm,          TB_ALIGN_32 },
-    { X86::VHADDPDYrr,        X86::VHADDPDYrm,         TB_ALIGN_32 },
-    { X86::VHADDPSYrr,        X86::VHADDPSYrm,         TB_ALIGN_32 },
-    { X86::VHSUBPDYrr,        X86::VHSUBPDYrm,         TB_ALIGN_32 },
-    { X86::VHSUBPSYrr,        X86::VHSUBPSYrm,         TB_ALIGN_32 },
-    { X86::VINSERTF128rr,     X86::VINSERTF128rm,      TB_ALIGN_32 },
-    { X86::VMAXPDYrr,         X86::VMAXPDYrm,          TB_ALIGN_32 },
-    { X86::VMAXPDYrr_Int,     X86::VMAXPDYrm_Int,      TB_ALIGN_32 },
-    { X86::VMAXPSYrr,         X86::VMAXPSYrm,          TB_ALIGN_32 },
-    { X86::VMAXPSYrr_Int,     X86::VMAXPSYrm_Int,      TB_ALIGN_32 },
-    { X86::VMINPDYrr,         X86::VMINPDYrm,          TB_ALIGN_32 },
-    { X86::VMINPDYrr_Int,     X86::VMINPDYrm_Int,      TB_ALIGN_32 },
-    { X86::VMINPSYrr,         X86::VMINPSYrm,          TB_ALIGN_32 },
-    { X86::VMINPSYrr_Int,     X86::VMINPSYrm_Int,      TB_ALIGN_32 },
-    { X86::VMULPDYrr,         X86::VMULPDYrm,          TB_ALIGN_32 },
-    { X86::VMULPSYrr,         X86::VMULPSYrm,          TB_ALIGN_32 },
-    { X86::VORPDYrr,          X86::VORPDYrm,           TB_ALIGN_32 },
-    { X86::VORPSYrr,          X86::VORPSYrm,           TB_ALIGN_32 },
-    { X86::VPERM2F128rr,      X86::VPERM2F128rm,       TB_ALIGN_32 },
-    { X86::VPERMILPDYrr,      X86::VPERMILPDYrm,       TB_ALIGN_32 },
-    { X86::VPERMILPSYrr,      X86::VPERMILPSYrm,       TB_ALIGN_32 },
-    { X86::VSHUFPDYrri,       X86::VSHUFPDYrmi,        TB_ALIGN_32 },
-    { X86::VSHUFPSYrri,       X86::VSHUFPSYrmi,        TB_ALIGN_32 },
-    { X86::VSUBPDYrr,         X86::VSUBPDYrm,          TB_ALIGN_32 },
-    { X86::VSUBPSYrr,         X86::VSUBPSYrm,          TB_ALIGN_32 },
-    { X86::VUNPCKHPDYrr,      X86::VUNPCKHPDYrm,       TB_ALIGN_32 },
-    { X86::VUNPCKHPSYrr,      X86::VUNPCKHPSYrm,       TB_ALIGN_32 },
-    { X86::VUNPCKLPDYrr,      X86::VUNPCKLPDYrm,       TB_ALIGN_32 },
-    { X86::VUNPCKLPSYrr,      X86::VUNPCKLPSYrm,       TB_ALIGN_32 },
-    { X86::VXORPDYrr,         X86::VXORPDYrm,          TB_ALIGN_32 },
-    { X86::VXORPSYrr,         X86::VXORPSYrm,          TB_ALIGN_32 },
+    { X86::VADDPDYrr,         X86::VADDPDYrm,          0 },
+    { X86::VADDPSYrr,         X86::VADDPSYrm,          0 },
+    { X86::VADDSUBPDYrr,      X86::VADDSUBPDYrm,       0 },
+    { X86::VADDSUBPSYrr,      X86::VADDSUBPSYrm,       0 },
+    { X86::VANDNPDYrr,        X86::VANDNPDYrm,         0 },
+    { X86::VANDNPSYrr,        X86::VANDNPSYrm,         0 },
+    { X86::VANDPDYrr,         X86::VANDPDYrm,          0 },
+    { X86::VANDPSYrr,         X86::VANDPSYrm,          0 },
+    { X86::VBLENDPDYrri,      X86::VBLENDPDYrmi,       0 },
+    { X86::VBLENDPSYrri,      X86::VBLENDPSYrmi,       0 },
+    { X86::VBLENDVPDYrr,      X86::VBLENDVPDYrm,       0 },
+    { X86::VBLENDVPSYrr,      X86::VBLENDVPSYrm,       0 },
+    { X86::VCMPPDYrri,        X86::VCMPPDYrmi,         0 },
+    { X86::VCMPPSYrri,        X86::VCMPPSYrmi,         0 },
+    { X86::VDIVPDYrr,         X86::VDIVPDYrm,          0 },
+    { X86::VDIVPSYrr,         X86::VDIVPSYrm,          0 },
+    { X86::VHADDPDYrr,        X86::VHADDPDYrm,         0 },
+    { X86::VHADDPSYrr,        X86::VHADDPSYrm,         0 },
+    { X86::VHSUBPDYrr,        X86::VHSUBPDYrm,         0 },
+    { X86::VHSUBPSYrr,        X86::VHSUBPSYrm,         0 },
+    { X86::VINSERTF128rr,     X86::VINSERTF128rm,      0 },
+    { X86::VMAXPDYrr,         X86::VMAXPDYrm,          0 },
+    { X86::VMAXPSYrr,         X86::VMAXPSYrm,          0 },
+    { X86::VMINPDYrr,         X86::VMINPDYrm,          0 },
+    { X86::VMINPSYrr,         X86::VMINPSYrm,          0 },
+    { X86::VMULPDYrr,         X86::VMULPDYrm,          0 },
+    { X86::VMULPSYrr,         X86::VMULPSYrm,          0 },
+    { X86::VORPDYrr,          X86::VORPDYrm,           0 },
+    { X86::VORPSYrr,          X86::VORPSYrm,           0 },
+    { X86::VPERM2F128rr,      X86::VPERM2F128rm,       0 },
+    { X86::VPERMILPDYrr,      X86::VPERMILPDYrm,       0 },
+    { X86::VPERMILPSYrr,      X86::VPERMILPSYrm,       0 },
+    { X86::VSHUFPDYrri,       X86::VSHUFPDYrmi,        0 },
+    { X86::VSHUFPSYrri,       X86::VSHUFPSYrmi,        0 },
+    { X86::VSUBPDYrr,         X86::VSUBPDYrm,          0 },
+    { X86::VSUBPSYrr,         X86::VSUBPSYrm,          0 },
+    { X86::VUNPCKHPDYrr,      X86::VUNPCKHPDYrm,       0 },
+    { X86::VUNPCKHPSYrr,      X86::VUNPCKHPSYrm,       0 },
+    { X86::VUNPCKLPDYrr,      X86::VUNPCKLPDYrm,       0 },
+    { X86::VUNPCKLPSYrr,      X86::VUNPCKLPSYrm,       0 },
+    { X86::VXORPDYrr,         X86::VXORPDYrm,          0 },
+    { X86::VXORPSYrr,         X86::VXORPSYrm,          0 },
     // AVX2 foldable instructions
-    { X86::VINSERTI128rr,     X86::VINSERTI128rm,      TB_ALIGN_16 },
-    { X86::VPACKSSDWYrr,      X86::VPACKSSDWYrm,       TB_ALIGN_32 },
-    { X86::VPACKSSWBYrr,      X86::VPACKSSWBYrm,       TB_ALIGN_32 },
-    { X86::VPACKUSDWYrr,      X86::VPACKUSDWYrm,       TB_ALIGN_32 },
-    { X86::VPACKUSWBYrr,      X86::VPACKUSWBYrm,       TB_ALIGN_32 },
-    { X86::VPADDBYrr,         X86::VPADDBYrm,          TB_ALIGN_32 },
-    { X86::VPADDDYrr,         X86::VPADDDYrm,          TB_ALIGN_32 },
-    { X86::VPADDQYrr,         X86::VPADDQYrm,          TB_ALIGN_32 },
-    { X86::VPADDSBYrr,        X86::VPADDSBYrm,         TB_ALIGN_32 },
-    { X86::VPADDSWYrr,        X86::VPADDSWYrm,         TB_ALIGN_32 },
-    { X86::VPADDUSBYrr,       X86::VPADDUSBYrm,        TB_ALIGN_32 },
-    { X86::VPADDUSWYrr,       X86::VPADDUSWYrm,        TB_ALIGN_32 },
-    { X86::VPADDWYrr,         X86::VPADDWYrm,          TB_ALIGN_32 },
-    { X86::VPALIGNR256rr,     X86::VPALIGNR256rm,      TB_ALIGN_32 },
-    { X86::VPANDNYrr,         X86::VPANDNYrm,          TB_ALIGN_32 },
-    { X86::VPANDYrr,          X86::VPANDYrm,           TB_ALIGN_32 },
-    { X86::VPAVGBYrr,         X86::VPAVGBYrm,          TB_ALIGN_32 },
-    { X86::VPAVGWYrr,         X86::VPAVGWYrm,          TB_ALIGN_32 },
-    { X86::VPBLENDDrri,       X86::VPBLENDDrmi,        TB_ALIGN_32 },
-    { X86::VPBLENDDYrri,      X86::VPBLENDDYrmi,       TB_ALIGN_32 },
-    { X86::VPBLENDWYrri,      X86::VPBLENDWYrmi,       TB_ALIGN_32 },
-    { X86::VPCMPEQBYrr,       X86::VPCMPEQBYrm,        TB_ALIGN_32 },
-    { X86::VPCMPEQDYrr,       X86::VPCMPEQDYrm,        TB_ALIGN_32 },
-    { X86::VPCMPEQQYrr,       X86::VPCMPEQQYrm,        TB_ALIGN_32 },
-    { X86::VPCMPEQWYrr,       X86::VPCMPEQWYrm,        TB_ALIGN_32 },
-    { X86::VPCMPGTBYrr,       X86::VPCMPGTBYrm,        TB_ALIGN_32 },
-    { X86::VPCMPGTDYrr,       X86::VPCMPGTDYrm,        TB_ALIGN_32 },
-    { X86::VPCMPGTQYrr,       X86::VPCMPGTQYrm,        TB_ALIGN_32 },
-    { X86::VPCMPGTWYrr,       X86::VPCMPGTWYrm,        TB_ALIGN_32 },
-    { X86::VPERM2I128rr,      X86::VPERM2I128rm,       TB_ALIGN_32 },
-    { X86::VPERMDYrr,         X86::VPERMDYrm,          TB_ALIGN_32 },
-    { X86::VPERMPDYri,        X86::VPERMPDYmi,         TB_ALIGN_32 },
-    { X86::VPERMPSYrr,        X86::VPERMPSYrm,         TB_ALIGN_32 },
-    { X86::VPERMQYri,         X86::VPERMQYmi,          TB_ALIGN_32 },
-    { X86::VPHADDDYrr,        X86::VPHADDDYrm,         TB_ALIGN_32 },
-    { X86::VPHADDSWrr256,     X86::VPHADDSWrm256,      TB_ALIGN_32 },
-    { X86::VPHADDWYrr,        X86::VPHADDWYrm,         TB_ALIGN_32 },
-    { X86::VPHSUBDYrr,        X86::VPHSUBDYrm,         TB_ALIGN_32 },
-    { X86::VPHSUBSWrr256,     X86::VPHSUBSWrm256,      TB_ALIGN_32 },
-    { X86::VPHSUBWYrr,        X86::VPHSUBWYrm,         TB_ALIGN_32 },
-    { X86::VPMADDUBSWrr256,   X86::VPMADDUBSWrm256,    TB_ALIGN_32 },
-    { X86::VPMADDWDYrr,       X86::VPMADDWDYrm,        TB_ALIGN_32 },
-    { X86::VPMAXSWYrr,        X86::VPMAXSWYrm,         TB_ALIGN_32 },
-    { X86::VPMAXUBYrr,        X86::VPMAXUBYrm,         TB_ALIGN_32 },
-    { X86::VPMINSWYrr,        X86::VPMINSWYrm,         TB_ALIGN_32 },
-    { X86::VPMINUBYrr,        X86::VPMINUBYrm,         TB_ALIGN_32 },
-    { X86::VMPSADBWYrri,      X86::VMPSADBWYrmi,       TB_ALIGN_32 },
-    { X86::VPMULDQYrr,        X86::VPMULDQYrm,         TB_ALIGN_32 },
-    { X86::VPMULHRSWrr256,    X86::VPMULHRSWrm256,     TB_ALIGN_32 },
-    { X86::VPMULHUWYrr,       X86::VPMULHUWYrm,        TB_ALIGN_32 },
-    { X86::VPMULHWYrr,        X86::VPMULHWYrm,         TB_ALIGN_32 },
-    { X86::VPMULLDYrr,        X86::VPMULLDYrm,         TB_ALIGN_32 },
-    { X86::VPMULLWYrr,        X86::VPMULLWYrm,         TB_ALIGN_32 },
-    { X86::VPMULUDQYrr,       X86::VPMULUDQYrm,        TB_ALIGN_32 },
-    { X86::VPORYrr,           X86::VPORYrm,            TB_ALIGN_32 },
-    { X86::VPSADBWYrr,        X86::VPSADBWYrm,         TB_ALIGN_32 },
-    { X86::VPSHUFBYrr,        X86::VPSHUFBYrm,         TB_ALIGN_32 },
-    { X86::VPSIGNBYrr,        X86::VPSIGNBYrm,         TB_ALIGN_32 },
-    { X86::VPSIGNWYrr,        X86::VPSIGNWYrm,         TB_ALIGN_32 },
-    { X86::VPSIGNDYrr,        X86::VPSIGNDYrm,         TB_ALIGN_32 },
-    { X86::VPSLLDYrr,         X86::VPSLLDYrm,          TB_ALIGN_16 },
-    { X86::VPSLLQYrr,         X86::VPSLLQYrm,          TB_ALIGN_16 },
-    { X86::VPSLLWYrr,         X86::VPSLLWYrm,          TB_ALIGN_16 },
-    { X86::VPSLLVDrr,         X86::VPSLLVDrm,          TB_ALIGN_16 },
-    { X86::VPSLLVDYrr,        X86::VPSLLVDYrm,         TB_ALIGN_32 },
-    { X86::VPSLLVQrr,         X86::VPSLLVQrm,          TB_ALIGN_16 },
-    { X86::VPSLLVQYrr,        X86::VPSLLVQYrm,         TB_ALIGN_32 },
-    { X86::VPSRADYrr,         X86::VPSRADYrm,          TB_ALIGN_16 },
-    { X86::VPSRAWYrr,         X86::VPSRAWYrm,          TB_ALIGN_16 },
-    { X86::VPSRAVDrr,         X86::VPSRAVDrm,          TB_ALIGN_16 },
-    { X86::VPSRAVDYrr,        X86::VPSRAVDYrm,         TB_ALIGN_32 },
-    { X86::VPSRLDYrr,         X86::VPSRLDYrm,          TB_ALIGN_16 },
-    { X86::VPSRLQYrr,         X86::VPSRLQYrm,          TB_ALIGN_16 },
-    { X86::VPSRLWYrr,         X86::VPSRLWYrm,          TB_ALIGN_16 },
-    { X86::VPSRLVDrr,         X86::VPSRLVDrm,          TB_ALIGN_16 },
-    { X86::VPSRLVDYrr,        X86::VPSRLVDYrm,         TB_ALIGN_32 },
-    { X86::VPSRLVQrr,         X86::VPSRLVQrm,          TB_ALIGN_16 },
-    { X86::VPSRLVQYrr,        X86::VPSRLVQYrm,         TB_ALIGN_32 },
-    { X86::VPSUBBYrr,         X86::VPSUBBYrm,          TB_ALIGN_32 },
-    { X86::VPSUBDYrr,         X86::VPSUBDYrm,          TB_ALIGN_32 },
-    { X86::VPSUBSBYrr,        X86::VPSUBSBYrm,         TB_ALIGN_32 },
-    { X86::VPSUBSWYrr,        X86::VPSUBSWYrm,         TB_ALIGN_32 },
-    { X86::VPSUBWYrr,         X86::VPSUBWYrm,          TB_ALIGN_32 },
-    { X86::VPUNPCKHBWYrr,     X86::VPUNPCKHBWYrm,      TB_ALIGN_32 },
-    { X86::VPUNPCKHDQYrr,     X86::VPUNPCKHDQYrm,      TB_ALIGN_32 },
-    { X86::VPUNPCKHQDQYrr,    X86::VPUNPCKHQDQYrm,     TB_ALIGN_16 },
-    { X86::VPUNPCKHWDYrr,     X86::VPUNPCKHWDYrm,      TB_ALIGN_32 },
-    { X86::VPUNPCKLBWYrr,     X86::VPUNPCKLBWYrm,      TB_ALIGN_32 },
-    { X86::VPUNPCKLDQYrr,     X86::VPUNPCKLDQYrm,      TB_ALIGN_32 },
-    { X86::VPUNPCKLQDQYrr,    X86::VPUNPCKLQDQYrm,     TB_ALIGN_32 },
-    { X86::VPUNPCKLWDYrr,     X86::VPUNPCKLWDYrm,      TB_ALIGN_32 },
-    { X86::VPXORYrr,          X86::VPXORYrm,           TB_ALIGN_32 },
+    { X86::VINSERTI128rr,     X86::VINSERTI128rm,      0 },
+    { X86::VPACKSSDWYrr,      X86::VPACKSSDWYrm,       0 },
+    { X86::VPACKSSWBYrr,      X86::VPACKSSWBYrm,       0 },
+    { X86::VPACKUSDWYrr,      X86::VPACKUSDWYrm,       0 },
+    { X86::VPACKUSWBYrr,      X86::VPACKUSWBYrm,       0 },
+    { X86::VPADDBYrr,         X86::VPADDBYrm,          0 },
+    { X86::VPADDDYrr,         X86::VPADDDYrm,          0 },
+    { X86::VPADDQYrr,         X86::VPADDQYrm,          0 },
+    { X86::VPADDSBYrr,        X86::VPADDSBYrm,         0 },
+    { X86::VPADDSWYrr,        X86::VPADDSWYrm,         0 },
+    { X86::VPADDUSBYrr,       X86::VPADDUSBYrm,        0 },
+    { X86::VPADDUSWYrr,       X86::VPADDUSWYrm,        0 },
+    { X86::VPADDWYrr,         X86::VPADDWYrm,          0 },
+    { X86::VPALIGNR256rr,     X86::VPALIGNR256rm,      0 },
+    { X86::VPANDNYrr,         X86::VPANDNYrm,          0 },
+    { X86::VPANDYrr,          X86::VPANDYrm,           0 },
+    { X86::VPAVGBYrr,         X86::VPAVGBYrm,          0 },
+    { X86::VPAVGWYrr,         X86::VPAVGWYrm,          0 },
+    { X86::VPBLENDDrri,       X86::VPBLENDDrmi,        0 },
+    { X86::VPBLENDDYrri,      X86::VPBLENDDYrmi,       0 },
+    { X86::VPBLENDWYrri,      X86::VPBLENDWYrmi,       0 },
+    { X86::VPCMPEQBYrr,       X86::VPCMPEQBYrm,        0 },
+    { X86::VPCMPEQDYrr,       X86::VPCMPEQDYrm,        0 },
+    { X86::VPCMPEQQYrr,       X86::VPCMPEQQYrm,        0 },
+    { X86::VPCMPEQWYrr,       X86::VPCMPEQWYrm,        0 },
+    { X86::VPCMPGTBYrr,       X86::VPCMPGTBYrm,        0 },
+    { X86::VPCMPGTDYrr,       X86::VPCMPGTDYrm,        0 },
+    { X86::VPCMPGTQYrr,       X86::VPCMPGTQYrm,        0 },
+    { X86::VPCMPGTWYrr,       X86::VPCMPGTWYrm,        0 },
+    { X86::VPERM2I128rr,      X86::VPERM2I128rm,       0 },
+    { X86::VPERMDYrr,         X86::VPERMDYrm,          0 },
+    { X86::VPERMPDYri,        X86::VPERMPDYmi,         0 },
+    { X86::VPERMPSYrr,        X86::VPERMPSYrm,         0 },
+    { X86::VPERMQYri,         X86::VPERMQYmi,          0 },
+    { X86::VPHADDDYrr,        X86::VPHADDDYrm,         0 },
+    { X86::VPHADDSWrr256,     X86::VPHADDSWrm256,      0 },
+    { X86::VPHADDWYrr,        X86::VPHADDWYrm,         0 },
+    { X86::VPHSUBDYrr,        X86::VPHSUBDYrm,         0 },
+    { X86::VPHSUBSWrr256,     X86::VPHSUBSWrm256,      0 },
+    { X86::VPHSUBWYrr,        X86::VPHSUBWYrm,         0 },
+    { X86::VPMADDUBSWrr256,   X86::VPMADDUBSWrm256,    0 },
+    { X86::VPMADDWDYrr,       X86::VPMADDWDYrm,        0 },
+    { X86::VPMAXSWYrr,        X86::VPMAXSWYrm,         0 },
+    { X86::VPMAXUBYrr,        X86::VPMAXUBYrm,         0 },
+    { X86::VPMINSWYrr,        X86::VPMINSWYrm,         0 },
+    { X86::VPMINUBYrr,        X86::VPMINUBYrm,         0 },
+    { X86::VPMINSBYrr,        X86::VPMINSBYrm,         0 },
+    { X86::VPMINSDYrr,        X86::VPMINSDYrm,         0 },
+    { X86::VPMINUDYrr,        X86::VPMINUDYrm,         0 },
+    { X86::VPMINUWYrr,        X86::VPMINUWYrm,         0 },
+    { X86::VPMAXSBYrr,        X86::VPMAXSBYrm,         0 },
+    { X86::VPMAXSDYrr,        X86::VPMAXSDYrm,         0 },
+    { X86::VPMAXUDYrr,        X86::VPMAXUDYrm,         0 },
+    { X86::VPMAXUWYrr,        X86::VPMAXUWYrm,         0 },
+    { X86::VMPSADBWYrri,      X86::VMPSADBWYrmi,       0 },
+    { X86::VPMULDQYrr,        X86::VPMULDQYrm,         0 },
+    { X86::VPMULHRSWrr256,    X86::VPMULHRSWrm256,     0 },
+    { X86::VPMULHUWYrr,       X86::VPMULHUWYrm,        0 },
+    { X86::VPMULHWYrr,        X86::VPMULHWYrm,         0 },
+    { X86::VPMULLDYrr,        X86::VPMULLDYrm,         0 },
+    { X86::VPMULLWYrr,        X86::VPMULLWYrm,         0 },
+    { X86::VPMULUDQYrr,       X86::VPMULUDQYrm,        0 },
+    { X86::VPORYrr,           X86::VPORYrm,            0 },
+    { X86::VPSADBWYrr,        X86::VPSADBWYrm,         0 },
+    { X86::VPSHUFBYrr,        X86::VPSHUFBYrm,         0 },
+    { X86::VPSIGNBYrr,        X86::VPSIGNBYrm,         0 },
+    { X86::VPSIGNWYrr,        X86::VPSIGNWYrm,         0 },
+    { X86::VPSIGNDYrr,        X86::VPSIGNDYrm,         0 },
+    { X86::VPSLLDYrr,         X86::VPSLLDYrm,          0 },
+    { X86::VPSLLQYrr,         X86::VPSLLQYrm,          0 },
+    { X86::VPSLLWYrr,         X86::VPSLLWYrm,          0 },
+    { X86::VPSLLVDrr,         X86::VPSLLVDrm,          0 },
+    { X86::VPSLLVDYrr,        X86::VPSLLVDYrm,         0 },
+    { X86::VPSLLVQrr,         X86::VPSLLVQrm,          0 },
+    { X86::VPSLLVQYrr,        X86::VPSLLVQYrm,         0 },
+    { X86::VPSRADYrr,         X86::VPSRADYrm,          0 },
+    { X86::VPSRAWYrr,         X86::VPSRAWYrm,          0 },
+    { X86::VPSRAVDrr,         X86::VPSRAVDrm,          0 },
+    { X86::VPSRAVDYrr,        X86::VPSRAVDYrm,         0 },
+    { X86::VPSRLDYrr,         X86::VPSRLDYrm,          0 },
+    { X86::VPSRLQYrr,         X86::VPSRLQYrm,          0 },
+    { X86::VPSRLWYrr,         X86::VPSRLWYrm,          0 },
+    { X86::VPSRLVDrr,         X86::VPSRLVDrm,          0 },
+    { X86::VPSRLVDYrr,        X86::VPSRLVDYrm,         0 },
+    { X86::VPSRLVQrr,         X86::VPSRLVQrm,          0 },
+    { X86::VPSRLVQYrr,        X86::VPSRLVQYrm,         0 },
+    { X86::VPSUBBYrr,         X86::VPSUBBYrm,          0 },
+    { X86::VPSUBDYrr,         X86::VPSUBDYrm,          0 },
+    { X86::VPSUBSBYrr,        X86::VPSUBSBYrm,         0 },
+    { X86::VPSUBSWYrr,        X86::VPSUBSWYrm,         0 },
+    { X86::VPSUBWYrr,         X86::VPSUBWYrm,          0 },
+    { X86::VPUNPCKHBWYrr,     X86::VPUNPCKHBWYrm,      0 },
+    { X86::VPUNPCKHDQYrr,     X86::VPUNPCKHDQYrm,      0 },
+    { X86::VPUNPCKHQDQYrr,    X86::VPUNPCKHQDQYrm,     0 },
+    { X86::VPUNPCKHWDYrr,     X86::VPUNPCKHWDYrm,      0 },
+    { X86::VPUNPCKLBWYrr,     X86::VPUNPCKLBWYrm,      0 },
+    { X86::VPUNPCKLDQYrr,     X86::VPUNPCKLDQYrm,      0 },
+    { X86::VPUNPCKLQDQYrr,    X86::VPUNPCKLQDQYrm,     0 },
+    { X86::VPUNPCKLWDYrr,     X86::VPUNPCKLWDYrm,      0 },
+    { X86::VPXORYrr,          X86::VPXORYrm,           0 },
     // FIXME: add AVX 256-bit foldable instructions
 
     // FMA4 foldable patterns
@@ -1156,8 +1172,14 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::VFMSUBADDPD4rrY,   X86::VFMSUBADDPD4mrY,    TB_ALIGN_32 },
 
     // BMI/BMI2 foldable instructions
+    { X86::ANDN32rr,          X86::ANDN32rm,            0 },
+    { X86::ANDN64rr,          X86::ANDN64rm,            0 },
     { X86::MULX32rr,          X86::MULX32rm,            0 },
     { X86::MULX64rr,          X86::MULX64rm,            0 },
+    { X86::PDEP32rr,          X86::PDEP32rm,            0 },
+    { X86::PDEP64rr,          X86::PDEP64rm,            0 },
+    { X86::PEXT32rr,          X86::PEXT32rm,            0 },
+    { X86::PEXT64rr,          X86::PEXT64rm,            0 },
   };
 
   for (unsigned i = 0, e = array_lengthof(OpTbl2); i != e; ++i) {
@@ -1551,16 +1573,19 @@ X86InstrInfo::isReallyTriviallyReMaterializable(const MachineInstr *MI,
   case X86::MOVUPSrm:
   case X86::MOVAPDrm:
   case X86::MOVDQArm:
+  case X86::MOVDQUrm:
   case X86::VMOVSSrm:
   case X86::VMOVSDrm:
   case X86::VMOVAPSrm:
   case X86::VMOVUPSrm:
   case X86::VMOVAPDrm:
   case X86::VMOVDQArm:
+  case X86::VMOVDQUrm:
   case X86::VMOVAPSYrm:
   case X86::VMOVUPSYrm:
   case X86::VMOVAPDYrm:
   case X86::VMOVDQAYrm:
+  case X86::VMOVDQUYrm:
   case X86::MMX_MOVD64rm:
   case X86::MMX_MOVQ64rm:
   case X86::FsVMOVAPSrm:
@@ -2159,7 +2184,7 @@ X86InstrInfo::commuteInstruction(MachineInstr *MI, bool NewMI) const {
     }
     MI->setDesc(get(Opc));
     MI->getOperand(3).setImm(Size-Amt);
-    return TargetInstrInfoImpl::commuteInstruction(MI, NewMI);
+    return TargetInstrInfo::commuteInstruction(MI, NewMI);
   }
   case X86::CMOVB16rr:  case X86::CMOVB32rr:  case X86::CMOVB64rr:
   case X86::CMOVAE16rr: case X86::CMOVAE32rr: case X86::CMOVAE64rr:
@@ -2238,7 +2263,7 @@ X86InstrInfo::commuteInstruction(MachineInstr *MI, bool NewMI) const {
     // Fallthrough intended.
   }
   default:
-    return TargetInstrInfoImpl::commuteInstruction(MI, NewMI);
+    return TargetInstrInfo::commuteInstruction(MI, NewMI);
   }
 }
 
@@ -2840,6 +2865,8 @@ void X86InstrInfo::copyPhysReg(MachineBasicBlock &MBB,
   }
 
   // Moving EFLAGS to / from another register requires a push and a pop.
+  // Notice that we have to adjust the stack if we don't want to clobber the
+  // first frame index. See X86FrameLowering.cpp - colobbersTheStack.
   if (SrcReg == X86::EFLAGS) {
     if (X86::GR64RegClass.contains(DestReg)) {
       BuildMI(MBB, MI, DL, get(X86::PUSHF64));
@@ -3149,19 +3176,15 @@ inline static bool isDefConvertible(MachineInstr *MI) {
   case X86::SUB8ri:    case X86::SUB64rr:  case X86::SUB32rr:
   case X86::SUB16rr:   case X86::SUB8rr:   case X86::SUB64rm:
   case X86::SUB32rm:   case X86::SUB16rm:  case X86::SUB8rm:
-  case X86::DEC64r:  case X86::DEC32r:  case X86::DEC16r: case X86::DEC8r:
-  case X86::DEC64m:  case X86::DEC32m:  case X86::DEC16m: case X86::DEC8m:
+  case X86::DEC64r:    case X86::DEC32r:   case X86::DEC16r: case X86::DEC8r:
   case X86::DEC64_32r: case X86::DEC64_16r:
-  case X86::DEC64_32m: case X86::DEC64_16m:
   case X86::ADD64ri32: case X86::ADD64ri8: case X86::ADD32ri:
   case X86::ADD32ri8:  case X86::ADD16ri:  case X86::ADD16ri8:
   case X86::ADD8ri:    case X86::ADD64rr:  case X86::ADD32rr:
   case X86::ADD16rr:   case X86::ADD8rr:   case X86::ADD64rm:
   case X86::ADD32rm:   case X86::ADD16rm:  case X86::ADD8rm:
-  case X86::INC64r:  case X86::INC32r:  case X86::INC16r: case X86::INC8r:
-  case X86::INC64m:  case X86::INC32m:  case X86::INC16m: case X86::INC8m:
+  case X86::INC64r:    case X86::INC32r:   case X86::INC16r: case X86::INC8r:
   case X86::INC64_32r: case X86::INC64_16r:
-  case X86::INC64_32m: case X86::INC64_16m:
   case X86::AND64ri32: case X86::AND64ri8: case X86::AND32ri:
   case X86::AND32ri8:  case X86::AND16ri:  case X86::AND16ri8:
   case X86::AND8ri:    case X86::AND64rr:  case X86::AND32rr:
@@ -3177,6 +3200,8 @@ inline static bool isDefConvertible(MachineInstr *MI) {
   case X86::OR8ri:     case X86::OR64rr:   case X86::OR32rr:
   case X86::OR16rr:    case X86::OR8rr:    case X86::OR64rm:
   case X86::OR32rm:    case X86::OR16rm:   case X86::OR8rm:
+  case X86::ANDN32rr:  case X86::ANDN32rm:
+  case X86::ANDN64rr:  case X86::ANDN64rm:
     return true;
   }
 }
@@ -3499,43 +3524,44 @@ optimizeLoadInstr(MachineInstr *MI, const MachineRegisterInfo *MRI,
 /// to:
 ///   %xmm4 = PXORrr %xmm4<undef>, %xmm4<undef>
 ///
-static bool Expand2AddrUndef(MachineInstr *MI, const MCInstrDesc &Desc) {
+static bool Expand2AddrUndef(MachineInstrBuilder &MIB,
+                             const MCInstrDesc &Desc) {
   assert(Desc.getNumOperands() == 3 && "Expected two-addr instruction.");
-  unsigned Reg = MI->getOperand(0).getReg();
-  MI->setDesc(Desc);
+  unsigned Reg = MIB->getOperand(0).getReg();
+  MIB->setDesc(Desc);
 
   // MachineInstr::addOperand() will insert explicit operands before any
   // implicit operands.
-  MachineInstrBuilder(MI).addReg(Reg, RegState::Undef)
-                         .addReg(Reg, RegState::Undef);
+  MIB.addReg(Reg, RegState::Undef).addReg(Reg, RegState::Undef);
   // But we don't trust that.
-  assert(MI->getOperand(1).getReg() == Reg &&
-         MI->getOperand(2).getReg() == Reg && "Misplaced operand");
+  assert(MIB->getOperand(1).getReg() == Reg &&
+         MIB->getOperand(2).getReg() == Reg && "Misplaced operand");
   return true;
 }
 
 bool X86InstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
   bool HasAVX = TM.getSubtarget<X86Subtarget>().hasAVX();
+  MachineInstrBuilder MIB(*MI->getParent()->getParent(), MI);
   switch (MI->getOpcode()) {
   case X86::SETB_C8r:
-    return Expand2AddrUndef(MI, get(X86::SBB8rr));
+    return Expand2AddrUndef(MIB, get(X86::SBB8rr));
   case X86::SETB_C16r:
-    return Expand2AddrUndef(MI, get(X86::SBB16rr));
+    return Expand2AddrUndef(MIB, get(X86::SBB16rr));
   case X86::SETB_C32r:
-    return Expand2AddrUndef(MI, get(X86::SBB32rr));
+    return Expand2AddrUndef(MIB, get(X86::SBB32rr));
   case X86::SETB_C64r:
-    return Expand2AddrUndef(MI, get(X86::SBB64rr));
+    return Expand2AddrUndef(MIB, get(X86::SBB64rr));
   case X86::V_SET0:
   case X86::FsFLD0SS:
   case X86::FsFLD0SD:
-    return Expand2AddrUndef(MI, get(HasAVX ? X86::VXORPSrr : X86::XORPSrr));
+    return Expand2AddrUndef(MIB, get(HasAVX ? X86::VXORPSrr : X86::XORPSrr));
   case X86::AVX_SET0:
     assert(HasAVX && "AVX not supported");
-    return Expand2AddrUndef(MI, get(X86::VXORPSYrr));
+    return Expand2AddrUndef(MIB, get(X86::VXORPSYrr));
   case X86::V_SETALLONES:
-    return Expand2AddrUndef(MI, get(HasAVX ? X86::VPCMPEQDrr : X86::PCMPEQDrr));
+    return Expand2AddrUndef(MIB, get(HasAVX ? X86::VPCMPEQDrr : X86::PCMPEQDrr));
   case X86::AVX2_SETALLONES:
-    return Expand2AddrUndef(MI, get(X86::VPCMPEQDYrr));
+    return Expand2AddrUndef(MIB, get(X86::VPCMPEQDYrr));
   case X86::TEST8ri_NOREX:
     MI->setDesc(get(X86::TEST8ri));
     return true;
@@ -3561,9 +3587,10 @@ static MachineInstr *FuseTwoAddrInst(MachineFunction &MF, unsigned Opcode,
                                      MachineInstr *MI,
                                      const TargetInstrInfo &TII) {
   // Create the base instruction with the memory operand as the first part.
+  // Omit the implicit operands, something BuildMI can't do.
   MachineInstr *NewMI = MF.CreateMachineInstr(TII.get(Opcode),
                                               MI->getDebugLoc(), true);
-  MachineInstrBuilder MIB(NewMI);
+  MachineInstrBuilder MIB(MF, NewMI);
   unsigned NumAddrOps = MOs.size();
   for (unsigned i = 0; i != NumAddrOps; ++i)
     MIB.addOperand(MOs[i]);
@@ -3587,9 +3614,10 @@ static MachineInstr *FuseInst(MachineFunction &MF,
                               unsigned Opcode, unsigned OpNo,
                               const SmallVectorImpl<MachineOperand> &MOs,
                               MachineInstr *MI, const TargetInstrInfo &TII) {
+  // Omit the implicit operands, something BuildMI can't do.
   MachineInstr *NewMI = MF.CreateMachineInstr(TII.get(Opcode),
                                               MI->getDebugLoc(), true);
-  MachineInstrBuilder MIB(NewMI);
+  MachineInstrBuilder MIB(MF, NewMI);
 
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     MachineOperand &MO = MI->getOperand(i);
@@ -3627,7 +3655,16 @@ X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
                                     const SmallVectorImpl<MachineOperand> &MOs,
                                     unsigned Size, unsigned Align) const {
   const DenseMap<unsigned, std::pair<unsigned,unsigned> > *OpcodeTablePtr = 0;
+  bool isCallRegIndirect = TM.getSubtarget<X86Subtarget>().callRegIndirect();
   bool isTwoAddrFold = false;
+
+  // Atom favors register form of call. So, we do not fold loads into calls
+  // when X86Subtarget is Atom.
+  if (isCallRegIndirect &&
+    (MI->getOpcode() == X86::CALL32r || MI->getOpcode() == X86::CALL64r)) {
+    return NULL;
+  }
+
   unsigned NumOps = MI->getDesc().getNumOperands();
   bool isTwoAddr = NumOps > 1 &&
     MI->getDesc().getOperandConstraint(1, MCOI::TIED_TO) != -1;
@@ -3836,8 +3873,8 @@ MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
 
   // Unless optimizing for size, don't fold to avoid partial
   // register update stalls
-  if (!MF.getFunction()->getFnAttributes().
-        hasAttribute(Attributes::OptimizeForSize) &&
+  if (!MF.getFunction()->getAttributes().
+        hasAttribute(AttributeSet::FunctionIndex, Attribute::OptimizeForSize) &&
       hasPartialRegUpdate(MI->getOpcode()))
     return 0;
 
@@ -3878,8 +3915,8 @@ MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
 
   // Unless optimizing for size, don't fold to avoid partial
   // register update stalls
-  if (!MF.getFunction()->getFnAttributes().
-        hasAttribute(Attributes::OptimizeForSize) &&
+  if (!MF.getFunction()->getAttributes().
+        hasAttribute(AttributeSet::FunctionIndex, Attribute::OptimizeForSize) &&
       hasPartialRegUpdate(MI->getOpcode()))
     return 0;
 
@@ -3982,6 +4019,21 @@ MachineInstr* X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
     break;
   }
   default: {
+    if ((LoadMI->getOpcode() == X86::MOVSSrm ||
+         LoadMI->getOpcode() == X86::VMOVSSrm) &&
+        MF.getRegInfo().getRegClass(LoadMI->getOperand(0).getReg())->getSize()
+          > 4)
+      // These instructions only load 32 bits, we can't fold them if the
+      // destination register is wider than 32 bits (4 bytes).
+      return NULL;
+    if ((LoadMI->getOpcode() == X86::MOVSDrm ||
+         LoadMI->getOpcode() == X86::VMOVSDrm) &&
+        MF.getRegInfo().getRegClass(LoadMI->getOperand(0).getReg())->getSize()
+          > 8)
+      // These instructions only load 64 bits, we can't fold them if the
+      // destination register is wider than 64 bits (8 bytes).
+      return NULL;
+
     // Folding a normal load. Just copy the load's address operands.
     unsigned NumOps = LoadMI->getDesc().getNumOperands();
     for (unsigned i = NumOps - X86::AddrNumOperands; i != NumOps; ++i)
@@ -4049,7 +4101,7 @@ bool X86InstrInfo::canFoldMemoryOperand(const MachineInstr *MI,
 
   if (OpcodeTablePtr && OpcodeTablePtr->count(Opc))
     return true;
-  return TargetInstrInfoImpl::canFoldMemoryOperand(MI, Ops);
+  return TargetInstrInfo::canFoldMemoryOperand(MI, Ops);
 }
 
 bool X86InstrInfo::unfoldMemoryOperand(MachineFunction &MF, MachineInstr *MI,
@@ -4114,7 +4166,7 @@ bool X86InstrInfo::unfoldMemoryOperand(MachineFunction &MF, MachineInstr *MI,
 
   // Emit the data processing instruction.
   MachineInstr *DataMI = MF.CreateMachineInstr(MCID, MI->getDebugLoc(), true);
-  MachineInstrBuilder MIB(DataMI);
+  MachineInstrBuilder MIB(MF, DataMI);
 
   if (FoldedStore)
     MIB.addReg(Reg, RegState::Define);
@@ -4620,13 +4672,9 @@ bool X86InstrInfo::isHighLatencyDef(int opc) const {
   case X86::DIVSSrr:
   case X86::DIVSSrr_Int:
   case X86::SQRTPDm:
-  case X86::SQRTPDm_Int:
   case X86::SQRTPDr:
-  case X86::SQRTPDr_Int:
   case X86::SQRTPSm:
-  case X86::SQRTPSm_Int:
   case X86::SQRTPSr:
-  case X86::SQRTPSr_Int:
   case X86::SQRTSDm:
   case X86::SQRTSDm_Int:
   case X86::SQRTSDr:
@@ -4645,13 +4693,9 @@ bool X86InstrInfo::isHighLatencyDef(int opc) const {
   case X86::VDIVSSrr:
   case X86::VDIVSSrr_Int:
   case X86::VSQRTPDm:
-  case X86::VSQRTPDm_Int:
   case X86::VSQRTPDr:
-  case X86::VSQRTPDr_Int:
   case X86::VSQRTPSm:
-  case X86::VSQRTPSm_Int:
   case X86::VSQRTPSr:
-  case X86::VSQRTPSr_Int:
   case X86::VSQRTSDm:
   case X86::VSQRTSDm_Int:
   case X86::VSQRTSDr:
diff --git a/lib/Target/X86/X86InstrInfo.td b/lib/Target/X86/X86InstrInfo.td
index 650fa95d7f23..ccc1aa2e35a5 100644
--- a/lib/Target/X86/X86InstrInfo.td
+++ b/lib/Target/X86/X86InstrInfo.td
@@ -142,6 +142,9 @@ def X86sahf    : SDNode<"X86ISD::SAHF",     SDTX86sahf>;
 def X86rdrand  : SDNode<"X86ISD::RDRAND",   SDTX86rdrand,
                         [SDNPHasChain, SDNPSideEffect]>;
 
+def X86rdseed  : SDNode<"X86ISD::RDSEED",   SDTX86rdrand,
+                        [SDNPHasChain, SDNPSideEffect]>;
+
 def X86cas : SDNode<"X86ISD::LCMPXCHG_DAG", SDTX86cas,
                         [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPMayStore,
                          SDNPMayLoad, SDNPMemOperand]>;
@@ -247,9 +250,9 @@ def X86and_flag  : SDNode<"X86ISD::AND",  SDTBinaryArithWithFlags,
                           [SDNPCommutative]>;
 def X86andn_flag : SDNode<"X86ISD::ANDN", SDTBinaryArithWithFlags>;
 
-def X86blsi_flag : SDNode<"X86ISD::BLSI",  SDTUnaryArithWithFlags>;
-def X86blsmsk_flag : SDNode<"X86ISD::BLSMSK",  SDTUnaryArithWithFlags>;
-def X86blsr_flag : SDNode<"X86ISD::BLSR",  SDTUnaryArithWithFlags>;
+def X86blsi   : SDNode<"X86ISD::BLSI",   SDTIntUnaryOp>;
+def X86blsmsk : SDNode<"X86ISD::BLSMSK", SDTIntUnaryOp>;
+def X86blsr   : SDNode<"X86ISD::BLSR",   SDTIntUnaryOp>;
 
 def X86mul_imm : SDNode<"X86ISD::MUL_IMM", SDTIntBinOp>;
 
@@ -525,6 +528,13 @@ def lea64_32mem : Operand<i32> {
   let ParserMatchClass = X86MemAsmOperand;
 }
 
+// Memory operands that use 64-bit pointers in both ILP32 and LP64.
+def lea64mem : Operand<i64> {
+  let PrintMethod = "printi64mem";
+  let MIOperandInfo = (ops GR64, i8imm, GR64_NOSP, i32imm, i8imm);
+  let ParserMatchClass = X86MemAsmOperand;
+}
+
 
 //===----------------------------------------------------------------------===//
 // X86 Complex Pattern Definitions.
@@ -535,6 +545,12 @@ def addr      : ComplexPattern<iPTR, 5, "SelectAddr", [], [SDNPWantParent]>;
 def lea32addr : ComplexPattern<i32, 5, "SelectLEAAddr",
                                [add, sub, mul, X86mul_imm, shl, or, frameindex],
                                []>;
+// In 64-bit mode 32-bit LEAs can use RIP-relative addressing.
+def lea64_32addr : ComplexPattern<i32, 5, "SelectLEAAddr",
+                                  [add, sub, mul, X86mul_imm, shl, or,
+                                   frameindex, X86WrapperRIP],
+                                  []>;
+
 def tls32addr : ComplexPattern<i32, 5, "SelectTLSADDRAddr",
                                [tglobaltlsaddr], []>;
 
@@ -590,13 +606,19 @@ def HasLZCNT     : Predicate<"Subtarget->hasLZCNT()">;
 def HasBMI       : Predicate<"Subtarget->hasBMI()">;
 def HasBMI2      : Predicate<"Subtarget->hasBMI2()">;
 def HasRTM       : Predicate<"Subtarget->hasRTM()">;
+def HasHLE       : Predicate<"Subtarget->hasHLE()">;
+def HasTSX       : Predicate<"Subtarget->hasRTM() || Subtarget->hasHLE()">;
+def HasADX       : Predicate<"Subtarget->hasADX()">;
+def HasPRFCHW    : Predicate<"Subtarget->hasPRFCHW()">;
+def HasRDSEED    : Predicate<"Subtarget->hasRDSEED()">;
+def HasPrefetchW : Predicate<"Subtarget->has3DNow() || Subtarget->hasPRFCHW()">;
 def FPStackf32   : Predicate<"!Subtarget->hasSSE1()">;
 def FPStackf64   : Predicate<"!Subtarget->hasSSE2()">;
 def HasCmpxchg16b: Predicate<"Subtarget->hasCmpxchg16b()">;
 def In32BitMode  : Predicate<"!Subtarget->is64Bit()">,
-                             AssemblerPredicate<"!Mode64Bit">;
+                             AssemblerPredicate<"!Mode64Bit", "32-bit mode">;
 def In64BitMode  : Predicate<"Subtarget->is64Bit()">,
-                             AssemblerPredicate<"Mode64Bit">;
+                             AssemblerPredicate<"Mode64Bit", "64-bit mode">;
 def IsWin64      : Predicate<"Subtarget->isTargetWin64()">;
 def IsNaCl       : Predicate<"Subtarget->isTargetNaCl()">;
 def NotNaCl      : Predicate<"!Subtarget->isTargetNaCl()">;
@@ -612,6 +634,7 @@ def OptForSize   : Predicate<"OptForSize">;
 def OptForSpeed  : Predicate<"!OptForSize">;
 def FastBTMem    : Predicate<"!Subtarget->isBTMemSlow()">;
 def CallImmAddr  : Predicate<"Subtarget->IsLegalToCallImmediateAddr(TM)">;
+def FavorMemIndirectCall  : Predicate<"!Subtarget->callRegIndirect()">;
 
 //===----------------------------------------------------------------------===//
 // X86 Instruction Format Definitions.
@@ -744,7 +767,7 @@ def trunc_su : PatFrag<(ops node:$src), (trunc node:$src), [{
 //
 
 // Nop
-let neverHasSideEffects = 1 in {
+let neverHasSideEffects = 1, SchedRW = [WriteZero] in {
   def NOOP : I<0x90, RawFrm, (outs), (ins), "nop", [], IIC_NOP>;
   def NOOPW : I<0x1f, MRM0m, (outs), (ins i16mem:$zero),
                 "nop{w}\t$zero", [], IIC_NOP>, TB, OpSize;
@@ -755,8 +778,9 @@ let neverHasSideEffects = 1 in {
 
 // Constructing a stack frame.
 def ENTER : Ii16<0xC8, RawFrmImm8, (outs), (ins i16imm:$len, i8imm:$lvl),
-                 "enter\t$len, $lvl", [], IIC_ENTER>;
+                 "enter\t$len, $lvl", [], IIC_ENTER>, Sched<[WriteMicrocoded]>;
 
+let SchedRW = [WriteALU] in {
 let Defs = [EBP, ESP], Uses = [EBP, ESP], mayLoad = 1, neverHasSideEffects=1 in
 def LEAVE    : I<0xC9, RawFrm,
                  (outs), (ins), "leave", [], IIC_LEAVE>,
@@ -766,13 +790,14 @@ let Defs = [RBP,RSP], Uses = [RBP,RSP], mayLoad = 1, neverHasSideEffects = 1 in
 def LEAVE64  : I<0xC9, RawFrm,
                  (outs), (ins), "leave", [], IIC_LEAVE>,
                  Requires<[In64BitMode]>;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 //  Miscellaneous Instructions.
 //
 
 let Defs = [ESP], Uses = [ESP], neverHasSideEffects=1 in {
-let mayLoad = 1 in {
+let mayLoad = 1, SchedRW = [WriteLoad] in {
 def POP16r  : I<0x58, AddRegFrm, (outs GR16:$reg), (ins), "pop{w}\t$reg", [],
                 IIC_POP_REG16>, OpSize;
 def POP32r  : I<0x58, AddRegFrm, (outs GR32:$reg), (ins), "pop{l}\t$reg", [],
@@ -789,9 +814,9 @@ def POP32rmm: I<0x8F, MRM0m, (outs i32mem:$dst), (ins), "pop{l}\t$dst", [],
 def POPF16   : I<0x9D, RawFrm, (outs), (ins), "popf{w}", [], IIC_POP_F>, OpSize;
 def POPF32   : I<0x9D, RawFrm, (outs), (ins), "popf{l|d}", [], IIC_POP_FD>,
                Requires<[In32BitMode]>;
-}
+} // mayLoad, SchedRW
 
-let mayStore = 1 in {
+let mayStore = 1, SchedRW = [WriteStore] in {
 def PUSH16r  : I<0x50, AddRegFrm, (outs), (ins GR16:$reg), "push{w}\t$reg",[],
                  IIC_PUSH_REG>, OpSize;
 def PUSH32r  : I<0x50, AddRegFrm, (outs), (ins GR32:$reg), "push{l}\t$reg",[],
@@ -818,29 +843,30 @@ def PUSHF16  : I<0x9C, RawFrm, (outs), (ins), "pushf{w}", [], IIC_PUSH_F>,
 def PUSHF32  : I<0x9C, RawFrm, (outs), (ins), "pushf{l|d}", [], IIC_PUSH_F>,
                Requires<[In32BitMode]>;
 
-}
+} // mayStore, SchedRW
 }
 
 let Defs = [RSP], Uses = [RSP], neverHasSideEffects=1 in {
-let mayLoad = 1 in {
+let mayLoad = 1, SchedRW = [WriteLoad] in {
 def POP64r   : I<0x58, AddRegFrm,
                  (outs GR64:$reg), (ins), "pop{q}\t$reg", [], IIC_POP_REG>;
 def POP64rmr: I<0x8F, MRM0r, (outs GR64:$reg), (ins), "pop{q}\t$reg", [],
                 IIC_POP_REG>;
 def POP64rmm: I<0x8F, MRM0m, (outs i64mem:$dst), (ins), "pop{q}\t$dst", [],
                 IIC_POP_MEM>;
-}
-let mayStore = 1 in {
+} // mayLoad, SchedRW
+let mayStore = 1, SchedRW = [WriteStore] in {
 def PUSH64r  : I<0x50, AddRegFrm,
                  (outs), (ins GR64:$reg), "push{q}\t$reg", [], IIC_PUSH_REG>;
 def PUSH64rmr: I<0xFF, MRM6r, (outs), (ins GR64:$reg), "push{q}\t$reg", [],
                  IIC_PUSH_REG>;
 def PUSH64rmm: I<0xFF, MRM6m, (outs), (ins i64mem:$src), "push{q}\t$src", [],
                  IIC_PUSH_MEM>;
-}
+} // mayStore, SchedRW
 }
 
-let Defs = [RSP], Uses = [RSP], neverHasSideEffects = 1, mayStore = 1 in {
+let Defs = [RSP], Uses = [RSP], neverHasSideEffects = 1, mayStore = 1,
+    SchedRW = [WriteStore] in {
 def PUSH64i8   : Ii8<0x6a, RawFrm, (outs), (ins i64i8imm:$imm),
                      "push{q}\t$imm", [], IIC_PUSH_IMM>;
 def PUSH64i16  : Ii16<0x68, RawFrm, (outs), (ins i16imm:$imm),
@@ -851,25 +877,24 @@ def PUSH64i32  : Ii32<0x68, RawFrm, (outs), (ins i64i32imm:$imm),
 
 let Defs = [RSP, EFLAGS], Uses = [RSP], mayLoad = 1, neverHasSideEffects=1 in
 def POPF64   : I<0x9D, RawFrm, (outs), (ins), "popfq", [], IIC_POP_FD>,
-               Requires<[In64BitMode]>;
+               Requires<[In64BitMode]>, Sched<[WriteLoad]>;
 let Defs = [RSP], Uses = [RSP, EFLAGS], mayStore = 1, neverHasSideEffects=1 in
 def PUSHF64    : I<0x9C, RawFrm, (outs), (ins), "pushfq", [], IIC_PUSH_F>,
-                 Requires<[In64BitMode]>;
-
-
+                 Requires<[In64BitMode]>, Sched<[WriteStore]>;
 
 let Defs = [EDI, ESI, EBP, EBX, EDX, ECX, EAX, ESP], Uses = [ESP],
-    mayLoad=1, neverHasSideEffects=1 in {
-def POPA32   : I<0x61, RawFrm, (outs), (ins), "popa{l}", [], IIC_POP_A>,
+    mayLoad = 1, neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
+def POPA32   : I<0x61, RawFrm, (outs), (ins), "popa{l|d}", [], IIC_POP_A>,
                Requires<[In32BitMode]>;
 }
 let Defs = [ESP], Uses = [EDI, ESI, EBP, EBX, EDX, ECX, EAX, ESP],
-    mayStore=1, neverHasSideEffects=1 in {
-def PUSHA32  : I<0x60, RawFrm, (outs), (ins), "pusha{l}", [], IIC_PUSH_A>,
+    mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
+def PUSHA32  : I<0x60, RawFrm, (outs), (ins), "pusha{l|d}", [], IIC_PUSH_A>,
                Requires<[In32BitMode]>;
 }
 
-let Constraints = "$src = $dst" in {    // GR32 = bswap GR32
+let Constraints = "$src = $dst", SchedRW = [WriteALU] in {
+// GR32 = bswap GR32
 def BSWAP32r : I<0xC8, AddRegFrm,
                  (outs GR32:$dst), (ins GR32:$src),
                  "bswap{l}\t$dst",
@@ -878,60 +903,63 @@ def BSWAP32r : I<0xC8, AddRegFrm,
 def BSWAP64r : RI<0xC8, AddRegFrm, (outs GR64:$dst), (ins GR64:$src),
                   "bswap{q}\t$dst",
                   [(set GR64:$dst, (bswap GR64:$src))], IIC_BSWAP>, TB;
-} // Constraints = "$src = $dst"
+} // Constraints = "$src = $dst", SchedRW
 
 // Bit scan instructions.
 let Defs = [EFLAGS] in {
 def BSF16rr  : I<0xBC, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
                  "bsf{w}\t{$src, $dst|$dst, $src}",
                  [(set GR16:$dst, EFLAGS, (X86bsf GR16:$src))],
-                  IIC_BSF>, TB, OpSize;
+                  IIC_BSF>, TB, OpSize, Sched<[WriteShift]>;
 def BSF16rm  : I<0xBC, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
                  "bsf{w}\t{$src, $dst|$dst, $src}",
                  [(set GR16:$dst, EFLAGS, (X86bsf (loadi16 addr:$src)))],
-                  IIC_BSF>, TB, OpSize;
+                  IIC_BSF>, TB, OpSize, Sched<[WriteShiftLd]>;
 def BSF32rr  : I<0xBC, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
                  "bsf{l}\t{$src, $dst|$dst, $src}",
-                 [(set GR32:$dst, EFLAGS, (X86bsf GR32:$src))], IIC_BSF>, TB;
+                 [(set GR32:$dst, EFLAGS, (X86bsf GR32:$src))], IIC_BSF>, TB,
+               Sched<[WriteShift]>;
 def BSF32rm  : I<0xBC, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
                  "bsf{l}\t{$src, $dst|$dst, $src}",
                  [(set GR32:$dst, EFLAGS, (X86bsf (loadi32 addr:$src)))],
-                 IIC_BSF>, TB;
+                 IIC_BSF>, TB, Sched<[WriteShiftLd]>;
 def BSF64rr  : RI<0xBC, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
                   "bsf{q}\t{$src, $dst|$dst, $src}",
                   [(set GR64:$dst, EFLAGS, (X86bsf GR64:$src))],
-                  IIC_BSF>, TB;
+                  IIC_BSF>, TB, Sched<[WriteShift]>;
 def BSF64rm  : RI<0xBC, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
                   "bsf{q}\t{$src, $dst|$dst, $src}",
                   [(set GR64:$dst, EFLAGS, (X86bsf (loadi64 addr:$src)))],
-                  IIC_BSF>, TB;
+                  IIC_BSF>, TB, Sched<[WriteShiftLd]>;
 
 def BSR16rr  : I<0xBD, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
                  "bsr{w}\t{$src, $dst|$dst, $src}",
                  [(set GR16:$dst, EFLAGS, (X86bsr GR16:$src))], IIC_BSR>,
-                 TB, OpSize;
+                 TB, OpSize, Sched<[WriteShift]>;
 def BSR16rm  : I<0xBD, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
                  "bsr{w}\t{$src, $dst|$dst, $src}",
                  [(set GR16:$dst, EFLAGS, (X86bsr (loadi16 addr:$src)))],
                  IIC_BSR>, TB,
-                 OpSize;
+                 OpSize, Sched<[WriteShiftLd]>;
 def BSR32rr  : I<0xBD, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
                  "bsr{l}\t{$src, $dst|$dst, $src}",
-                 [(set GR32:$dst, EFLAGS, (X86bsr GR32:$src))], IIC_BSR>, TB;
+                 [(set GR32:$dst, EFLAGS, (X86bsr GR32:$src))], IIC_BSR>, TB,
+               Sched<[WriteShift]>;
 def BSR32rm  : I<0xBD, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
                  "bsr{l}\t{$src, $dst|$dst, $src}",
                  [(set GR32:$dst, EFLAGS, (X86bsr (loadi32 addr:$src)))],
-                 IIC_BSR>, TB;
+                 IIC_BSR>, TB, Sched<[WriteShiftLd]>;
 def BSR64rr  : RI<0xBD, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
                   "bsr{q}\t{$src, $dst|$dst, $src}",
-                  [(set GR64:$dst, EFLAGS, (X86bsr GR64:$src))], IIC_BSR>, TB;
+                  [(set GR64:$dst, EFLAGS, (X86bsr GR64:$src))], IIC_BSR>, TB,
+               Sched<[WriteShift]>;
 def BSR64rm  : RI<0xBD, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
                   "bsr{q}\t{$src, $dst|$dst, $src}",
                   [(set GR64:$dst, EFLAGS, (X86bsr (loadi64 addr:$src)))],
-                  IIC_BSR>, TB;
+                  IIC_BSR>, TB, Sched<[WriteShiftLd]>;
 } // Defs = [EFLAGS]
 
-
+let SchedRW = [WriteMicrocoded] in {
 // These uses the DF flag in the EFLAGS register to inc or dec EDI and ESI
 let Defs = [EDI,ESI], Uses = [EDI,ESI,EFLAGS] in {
 def MOVSB : I<0xA4, RawFrm, (outs), (ins), "movsb", [], IIC_MOVS>;
@@ -959,12 +987,12 @@ def CMPS8 : I<0xA6, RawFrm, (outs), (ins), "cmpsb", [], IIC_CMPS>;
 def CMPS16 : I<0xA7, RawFrm, (outs), (ins), "cmpsw", [], IIC_CMPS>, OpSize;
 def CMPS32 : I<0xA7, RawFrm, (outs), (ins), "cmps{l|d}", [], IIC_CMPS>;
 def CMPS64 : RI<0xA7, RawFrm, (outs), (ins), "cmpsq", [], IIC_CMPS>;
-
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 //  Move Instructions.
 //
-
+let SchedRW = [WriteMove] in {
 let neverHasSideEffects = 1 in {
 def MOV8rr  : I<0x88, MRMDestReg, (outs GR8 :$dst), (ins GR8 :$src),
                 "mov{b}\t{$src, $dst|$dst, $src}", [], IIC_MOV>;
@@ -975,6 +1003,7 @@ def MOV32rr : I<0x89, MRMDestReg, (outs GR32:$dst), (ins GR32:$src),
 def MOV64rr : RI<0x89, MRMDestReg, (outs GR64:$dst), (ins GR64:$src),
                  "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV>;
 }
+
 let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
 def MOV8ri  : Ii8 <0xB0, AddRegFrm, (outs GR8 :$dst), (ins i8imm :$src),
                    "mov{b}\t{$src, $dst|$dst, $src}",
@@ -992,7 +1021,9 @@ def MOV64ri32 : RIi32<0xC7, MRM0r, (outs GR64:$dst), (ins i64i32imm:$src),
                       "mov{q}\t{$src, $dst|$dst, $src}",
                       [(set GR64:$dst, i64immSExt32:$src)], IIC_MOV>;
 }
+} // SchedRW
 
+let SchedRW = [WriteStore] in {
 def MOV8mi  : Ii8 <0xC6, MRM0m, (outs), (ins i8mem :$dst, i8imm :$src),
                    "mov{b}\t{$src, $dst|$dst, $src}",
                    [(store (i8 imm:$src), addr:$dst)], IIC_MOV_MEM>;
@@ -1005,9 +1036,11 @@ def MOV32mi : Ii32<0xC7, MRM0m, (outs), (ins i32mem:$dst, i32imm:$src),
 def MOV64mi32 : RIi32<0xC7, MRM0m, (outs), (ins i64mem:$dst, i64i32imm:$src),
                       "mov{q}\t{$src, $dst|$dst, $src}",
                       [(store i64immSExt32:$src, addr:$dst)], IIC_MOV_MEM>;
+} // SchedRW
 
 /// moffs8, moffs16 and moffs32 versions of moves.  The immediate is a
 /// 32-bit offset from the PC.  These are only valid in x86-32 mode.
+let SchedRW = [WriteALU] in {
 def MOV8o8a : Ii32 <0xA0, RawFrm, (outs), (ins offset8:$src),
                    "mov{b}\t{$src, %al|AL, $src}", [], IIC_MOV_MEM>,
                    Requires<[In32BitMode]>;
@@ -1026,6 +1059,7 @@ def MOV16ao16 : Ii32 <0xA3, RawFrm, (outs offset16:$dst), (ins),
 def MOV32ao32 : Ii32 <0xA3, RawFrm, (outs offset32:$dst), (ins),
                       "mov{l}\t{%eax, $dst|$dst, EAX}", [], IIC_MOV_MEM>,
                      Requires<[In32BitMode]>;
+}
 
 // FIXME: These definitions are utterly broken
 // Just leave them commented out for now because they're useless outside
@@ -1043,7 +1077,7 @@ def MOV64ao64 : RIi32<0xA3, RawFrm, (outs offset64:$dst), (ins),
 */
 
 
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
 def MOV8rr_REV : I<0x8A, MRMSrcReg, (outs GR8:$dst), (ins GR8:$src),
                    "mov{b}\t{$src, $dst|$dst, $src}", [], IIC_MOV>;
 def MOV16rr_REV : I<0x8B, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
@@ -1054,7 +1088,7 @@ def MOV64rr_REV : RI<0x8B, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
                      "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV>;
 }
 
-let canFoldAsLoad = 1, isReMaterializable = 1 in {
+let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in {
 def MOV8rm  : I<0x8A, MRMSrcMem, (outs GR8 :$dst), (ins i8mem :$src),
                 "mov{b}\t{$src, $dst|$dst, $src}",
                 [(set GR8:$dst, (loadi8 addr:$src))], IIC_MOV_MEM>;
@@ -1069,6 +1103,7 @@ def MOV64rm : RI<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
                  [(set GR64:$dst, (load addr:$src))], IIC_MOV_MEM>;
 }
 
+let SchedRW = [WriteStore] in {
 def MOV8mr  : I<0x88, MRMDestMem, (outs), (ins i8mem :$dst, GR8 :$src),
                 "mov{b}\t{$src, $dst|$dst, $src}",
                 [(store GR8:$src, addr:$dst)], IIC_MOV_MEM>;
@@ -1081,6 +1116,7 @@ def MOV32mr : I<0x89, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
 def MOV64mr : RI<0x89, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
                  "mov{q}\t{$src, $dst|$dst, $src}",
                  [(store GR64:$src, addr:$dst)], IIC_MOV_MEM>;
+} // SchedRW
 
 // Versions of MOV8rr, MOV8mr, and MOV8rm that use i8mem_NOREX and GR8_NOREX so
 // that they can be used for copying and storing h registers, which can't be
@@ -1089,34 +1125,37 @@ let isCodeGenOnly = 1 in {
 let neverHasSideEffects = 1 in
 def MOV8rr_NOREX : I<0x88, MRMDestReg,
                      (outs GR8_NOREX:$dst), (ins GR8_NOREX:$src),
-                     "mov{b}\t{$src, $dst|$dst, $src}  # NOREX", [], IIC_MOV>;
+                     "mov{b}\t{$src, $dst|$dst, $src}  # NOREX", [], IIC_MOV>,
+                   Sched<[WriteMove]>;
 let mayStore = 1 in
 def MOV8mr_NOREX : I<0x88, MRMDestMem,
                      (outs), (ins i8mem_NOREX:$dst, GR8_NOREX:$src),
                      "mov{b}\t{$src, $dst|$dst, $src}  # NOREX", [],
-                     IIC_MOV_MEM>;
+                     IIC_MOV_MEM>, Sched<[WriteStore]>;
 let mayLoad = 1, neverHasSideEffects = 1,
     canFoldAsLoad = 1, isReMaterializable = 1 in
 def MOV8rm_NOREX : I<0x8A, MRMSrcMem,
                      (outs GR8_NOREX:$dst), (ins i8mem_NOREX:$src),
                      "mov{b}\t{$src, $dst|$dst, $src}  # NOREX", [],
-                     IIC_MOV_MEM>;
+                     IIC_MOV_MEM>, Sched<[WriteLoad]>;
 }
 
 
 // Condition code ops, incl. set if equal/not equal/...
+let SchedRW = [WriteALU] in {
 let Defs = [EFLAGS], Uses = [AH] in
 def SAHF     : I<0x9E, RawFrm, (outs),  (ins), "sahf",
                  [(set EFLAGS, (X86sahf AH))], IIC_AHF>;
 let Defs = [AH], Uses = [EFLAGS], neverHasSideEffects = 1 in
 def LAHF     : I<0x9F, RawFrm, (outs),  (ins), "lahf", [],
                 IIC_AHF>;  // AH = flags
-
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // Bit tests instructions: BT, BTS, BTR, BTC.
 
 let Defs = [EFLAGS] in {
+let SchedRW = [WriteALU] in {
 def BT16rr : I<0xA3, MRMDestReg, (outs), (ins GR16:$src1, GR16:$src2),
                "bt{w}\t{$src2, $src1|$src1, $src2}",
                [(set EFLAGS, (X86bt GR16:$src1, GR16:$src2))], IIC_BT_RR>,
@@ -1127,31 +1166,35 @@ def BT32rr : I<0xA3, MRMDestReg, (outs), (ins GR32:$src1, GR32:$src2),
 def BT64rr : RI<0xA3, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2),
                "bt{q}\t{$src2, $src1|$src1, $src2}",
                [(set EFLAGS, (X86bt GR64:$src1, GR64:$src2))], IIC_BT_RR>, TB;
+} // SchedRW
 
 // Unlike with the register+register form, the memory+register form of the
 // bt instruction does not ignore the high bits of the index. From ISel's
 // perspective, this is pretty bizarre. Make these instructions disassembly
 // only for now.
 
-def BT16mr : I<0xA3, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2),
-               "bt{w}\t{$src2, $src1|$src1, $src2}",
-//               [(X86bt (loadi16 addr:$src1), GR16:$src2),
-//                (implicit EFLAGS)]
-               [], IIC_BT_MR
-               >, OpSize, TB, Requires<[FastBTMem]>;
-def BT32mr : I<0xA3, MRMDestMem, (outs), (ins i32mem:$src1, GR32:$src2),
-               "bt{l}\t{$src2, $src1|$src1, $src2}",
-//               [(X86bt (loadi32 addr:$src1), GR32:$src2),
-//                (implicit EFLAGS)]
-               [], IIC_BT_MR
-               >, TB, Requires<[FastBTMem]>;
-def BT64mr : RI<0xA3, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2),
-               "bt{q}\t{$src2, $src1|$src1, $src2}",
-//               [(X86bt (loadi64 addr:$src1), GR64:$src2),
-//                (implicit EFLAGS)]
-                [], IIC_BT_MR
-                >, TB;
-
+let mayLoad = 1, hasSideEffects = 0, SchedRW = [WriteALULd] in {
+  def BT16mr : I<0xA3, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2),
+                 "bt{w}\t{$src2, $src1|$src1, $src2}",
+  //               [(X86bt (loadi16 addr:$src1), GR16:$src2),
+  //                (implicit EFLAGS)]
+                 [], IIC_BT_MR
+                 >, OpSize, TB, Requires<[FastBTMem]>;
+  def BT32mr : I<0xA3, MRMDestMem, (outs), (ins i32mem:$src1, GR32:$src2),
+                 "bt{l}\t{$src2, $src1|$src1, $src2}",
+  //               [(X86bt (loadi32 addr:$src1), GR32:$src2),
+  //                (implicit EFLAGS)]
+                 [], IIC_BT_MR
+                 >, TB, Requires<[FastBTMem]>;
+  def BT64mr : RI<0xA3, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2),
+                 "bt{q}\t{$src2, $src1|$src1, $src2}",
+  //               [(X86bt (loadi64 addr:$src1), GR64:$src2),
+  //                (implicit EFLAGS)]
+                  [], IIC_BT_MR
+                  >, TB;
+}
+
+let SchedRW = [WriteALU] in {
 def BT16ri8 : Ii8<0xBA, MRM4r, (outs), (ins GR16:$src1, i16i8imm:$src2),
                 "bt{w}\t{$src2, $src1|$src1, $src2}",
                 [(set EFLAGS, (X86bt GR16:$src1, i16immSExt8:$src2))],
@@ -1164,10 +1207,12 @@ def BT64ri8 : RIi8<0xBA, MRM4r, (outs), (ins GR64:$src1, i64i8imm:$src2),
                 "bt{q}\t{$src2, $src1|$src1, $src2}",
                 [(set EFLAGS, (X86bt GR64:$src1, i64immSExt8:$src2))],
                 IIC_BT_RI>, TB;
+} // SchedRW
 
 // Note that these instructions don't need FastBTMem because that
 // only applies when the other operand is in a register. When it's
 // an immediate, bt is still fast.
+let SchedRW = [WriteALU] in {
 def BT16mi8 : Ii8<0xBA, MRM4m, (outs), (ins i16mem:$src1, i16i8imm:$src2),
                 "bt{w}\t{$src2, $src1|$src1, $src2}",
                 [(set EFLAGS, (X86bt (loadi16 addr:$src1), i16immSExt8:$src2))
@@ -1180,8 +1225,10 @@ def BT64mi8 : RIi8<0xBA, MRM4m, (outs), (ins i64mem:$src1, i64i8imm:$src2),
                 "bt{q}\t{$src2, $src1|$src1, $src2}",
                 [(set EFLAGS, (X86bt (loadi64 addr:$src1),
                                      i64immSExt8:$src2))], IIC_BT_MI>, TB;
+} // SchedRW
 
-
+let hasSideEffects = 0 in {
+let SchedRW = [WriteALU] in {
 def BTC16rr : I<0xBB, MRMDestReg, (outs), (ins GR16:$src1, GR16:$src2),
                 "btc{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>,
                 OpSize, TB;
@@ -1189,6 +1236,9 @@ def BTC32rr : I<0xBB, MRMDestReg, (outs), (ins GR32:$src1, GR32:$src2),
                 "btc{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, TB;
 def BTC64rr : RI<0xBB, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2),
                  "btc{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, TB;
+} // SchedRW
+
+let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in {
 def BTC16mr : I<0xBB, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2),
                 "btc{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>,
                 OpSize, TB;
@@ -1196,6 +1246,9 @@ def BTC32mr : I<0xBB, MRMDestMem, (outs), (ins i32mem:$src1, GR32:$src2),
                 "btc{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, TB;
 def BTC64mr : RI<0xBB, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2),
                  "btc{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, TB;
+}
+
+let SchedRW = [WriteALU] in {
 def BTC16ri8 : Ii8<0xBA, MRM7r, (outs), (ins GR16:$src1, i16i8imm:$src2),
                     "btc{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>,
                     OpSize, TB;
@@ -1203,6 +1256,9 @@ def BTC32ri8 : Ii8<0xBA, MRM7r, (outs), (ins GR32:$src1, i32i8imm:$src2),
                     "btc{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB;
 def BTC64ri8 : RIi8<0xBA, MRM7r, (outs), (ins GR64:$src1, i64i8imm:$src2),
                     "btc{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB;
+} // SchedRW
+
+let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in {
 def BTC16mi8 : Ii8<0xBA, MRM7m, (outs), (ins i16mem:$src1, i16i8imm:$src2),
                     "btc{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>,
                     OpSize, TB;
@@ -1210,7 +1266,9 @@ def BTC32mi8 : Ii8<0xBA, MRM7m, (outs), (ins i32mem:$src1, i32i8imm:$src2),
                     "btc{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, TB;
 def BTC64mi8 : RIi8<0xBA, MRM7m, (outs), (ins i64mem:$src1, i64i8imm:$src2),
                     "btc{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, TB;
+}
 
+let SchedRW = [WriteALU] in {
 def BTR16rr : I<0xB3, MRMDestReg, (outs), (ins GR16:$src1, GR16:$src2),
                 "btr{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>,
                 OpSize, TB;
@@ -1218,6 +1276,9 @@ def BTR32rr : I<0xB3, MRMDestReg, (outs), (ins GR32:$src1, GR32:$src2),
                 "btr{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, TB;
 def BTR64rr : RI<0xB3, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2),
                  "btr{q}\t{$src2, $src1|$src1, $src2}", []>, TB;
+} // SchedRW
+
+let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in {
 def BTR16mr : I<0xB3, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2),
                 "btr{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>,
                 OpSize, TB;
@@ -1225,6 +1286,9 @@ def BTR32mr : I<0xB3, MRMDestMem, (outs), (ins i32mem:$src1, GR32:$src2),
                 "btr{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, TB;
 def BTR64mr : RI<0xB3, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2),
                  "btr{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, TB;
+}
+
+let SchedRW = [WriteALU] in {
 def BTR16ri8 : Ii8<0xBA, MRM6r, (outs), (ins GR16:$src1, i16i8imm:$src2),
                     "btr{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>,
                     OpSize, TB;
@@ -1232,6 +1296,9 @@ def BTR32ri8 : Ii8<0xBA, MRM6r, (outs), (ins GR32:$src1, i32i8imm:$src2),
                     "btr{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB;
 def BTR64ri8 : RIi8<0xBA, MRM6r, (outs), (ins GR64:$src1, i64i8imm:$src2),
                     "btr{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB;
+} // SchedRW
+
+let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in {
 def BTR16mi8 : Ii8<0xBA, MRM6m, (outs), (ins i16mem:$src1, i16i8imm:$src2),
                     "btr{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>,
                     OpSize, TB;
@@ -1239,7 +1306,9 @@ def BTR32mi8 : Ii8<0xBA, MRM6m, (outs), (ins i32mem:$src1, i32i8imm:$src2),
                     "btr{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, TB;
 def BTR64mi8 : RIi8<0xBA, MRM6m, (outs), (ins i64mem:$src1, i64i8imm:$src2),
                     "btr{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, TB;
+}
 
+let SchedRW = [WriteALU] in {
 def BTS16rr : I<0xAB, MRMDestReg, (outs), (ins GR16:$src1, GR16:$src2),
                 "bts{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>,
                 OpSize, TB;
@@ -1247,6 +1316,9 @@ def BTS32rr : I<0xAB, MRMDestReg, (outs), (ins GR32:$src1, GR32:$src2),
                 "bts{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, TB;
 def BTS64rr : RI<0xAB, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2),
                  "bts{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, TB;
+} // SchedRW
+
+let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in {
 def BTS16mr : I<0xAB, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2),
                 "bts{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>,
                 OpSize, TB;
@@ -1254,6 +1326,9 @@ def BTS32mr : I<0xAB, MRMDestMem, (outs), (ins i32mem:$src1, GR32:$src2),
                 "bts{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, TB;
 def BTS64mr : RI<0xAB, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2),
                  "bts{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, TB;
+}
+
+let SchedRW = [WriteALU] in {
 def BTS16ri8 : Ii8<0xBA, MRM5r, (outs), (ins GR16:$src1, i16i8imm:$src2),
                     "bts{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>,
                     OpSize, TB;
@@ -1261,6 +1336,9 @@ def BTS32ri8 : Ii8<0xBA, MRM5r, (outs), (ins GR32:$src1, i32i8imm:$src2),
                     "bts{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB;
 def BTS64ri8 : RIi8<0xBA, MRM5r, (outs), (ins GR64:$src1, i64i8imm:$src2),
                     "bts{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB;
+} // SchedRW
+
+let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in {
 def BTS16mi8 : Ii8<0xBA, MRM5m, (outs), (ins i16mem:$src1, i16i8imm:$src2),
                     "bts{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>,
                     OpSize, TB;
@@ -1268,6 +1346,8 @@ def BTS32mi8 : Ii8<0xBA, MRM5m, (outs), (ins i32mem:$src1, i32i8imm:$src2),
                     "bts{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, TB;
 def BTS64mi8 : RIi8<0xBA, MRM5m, (outs), (ins i64mem:$src1, i64i8imm:$src2),
                     "bts{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, TB;
+}
+} // hasSideEffects = 0
 } // Defs = [EFLAGS]
 
 
@@ -1279,41 +1359,42 @@ def BTS64mi8 : RIi8<0xBA, MRM5m, (outs), (ins i64mem:$src1, i64i8imm:$src2),
 // operand is referenced, the atomicity is ensured.
 multiclass ATOMIC_SWAP<bits<8> opc8, bits<8> opc, string mnemonic, string frag,
                        InstrItinClass itin> {
-  let Constraints = "$val = $dst" in {
-    def #NAME#8rm  : I<opc8, MRMSrcMem, (outs GR8:$dst),
-                       (ins GR8:$val, i8mem:$ptr),
-                       !strconcat(mnemonic, "{b}\t{$val, $ptr|$ptr, $val}"),
-                       [(set
-                          GR8:$dst,
-                          (!cast<PatFrag>(frag # "_8") addr:$ptr, GR8:$val))],
-                       itin>;
-    def #NAME#16rm : I<opc, MRMSrcMem, (outs GR16:$dst),
-                       (ins GR16:$val, i16mem:$ptr),
-                       !strconcat(mnemonic, "{w}\t{$val, $ptr|$ptr, $val}"),
+  let Constraints = "$val = $dst", SchedRW = [WriteALULd, WriteRMW] in {
+    def NAME#8rm  : I<opc8, MRMSrcMem, (outs GR8:$dst),
+                      (ins GR8:$val, i8mem:$ptr),
+                      !strconcat(mnemonic, "{b}\t{$val, $ptr|$ptr, $val}"),
+                      [(set
+                         GR8:$dst,
+                         (!cast<PatFrag>(frag # "_8") addr:$ptr, GR8:$val))],
+                      itin>;
+    def NAME#16rm : I<opc, MRMSrcMem, (outs GR16:$dst),
+                      (ins GR16:$val, i16mem:$ptr),
+                      !strconcat(mnemonic, "{w}\t{$val, $ptr|$ptr, $val}"),
+                      [(set
+                         GR16:$dst,
+                         (!cast<PatFrag>(frag # "_16") addr:$ptr, GR16:$val))],
+                      itin>, OpSize;
+    def NAME#32rm : I<opc, MRMSrcMem, (outs GR32:$dst),
+                      (ins GR32:$val, i32mem:$ptr),
+                      !strconcat(mnemonic, "{l}\t{$val, $ptr|$ptr, $val}"),
+                      [(set
+                         GR32:$dst,
+                         (!cast<PatFrag>(frag # "_32") addr:$ptr, GR32:$val))],
+                      itin>;
+    def NAME#64rm : RI<opc, MRMSrcMem, (outs GR64:$dst),
+                       (ins GR64:$val, i64mem:$ptr),
+                       !strconcat(mnemonic, "{q}\t{$val, $ptr|$ptr, $val}"),
                        [(set
-                          GR16:$dst,
-                          (!cast<PatFrag>(frag # "_16") addr:$ptr, GR16:$val))],
-                       itin>, OpSize;
-    def #NAME#32rm : I<opc, MRMSrcMem, (outs GR32:$dst),
-                       (ins GR32:$val, i32mem:$ptr),
-                       !strconcat(mnemonic, "{l}\t{$val, $ptr|$ptr, $val}"),
-                       [(set
-                          GR32:$dst,
-                          (!cast<PatFrag>(frag # "_32") addr:$ptr, GR32:$val))],
+                         GR64:$dst,
+                         (!cast<PatFrag>(frag # "_64") addr:$ptr, GR64:$val))],
                        itin>;
-    def #NAME#64rm : RI<opc, MRMSrcMem, (outs GR64:$dst),
-                        (ins GR64:$val, i64mem:$ptr),
-                        !strconcat(mnemonic, "{q}\t{$val, $ptr|$ptr, $val}"),
-                        [(set
-                          GR64:$dst,
-                          (!cast<PatFrag>(frag # "_64") addr:$ptr, GR64:$val))],
-                        itin>;
   }
 }
 
 defm XCHG    : ATOMIC_SWAP<0x86, 0x87, "xchg", "atomic_swap", IIC_XCHG_MEM>;
 
 // Swap between registers.
+let SchedRW = [WriteALU] in {
 let Constraints = "$val = $dst" in {
 def XCHG8rr : I<0x86, MRMSrcReg, (outs GR8:$dst), (ins GR8:$val, GR8:$src),
                 "xchg{b}\t{$val, $src|$src, $val}", [], IIC_XCHG_REG>;
@@ -1338,9 +1419,9 @@ def XCHG32ar64 : I<0x90, AddRegFrm, (outs), (ins GR32_NOAX:$src),
                    Requires<[In64BitMode]>;
 def XCHG64ar : RI<0x90, AddRegFrm, (outs), (ins GR64:$src),
                   "xchg{q}\t{$src, %rax|RAX, $src}", [], IIC_XCHG_REG>;
+} // SchedRW
 
-
-
+let SchedRW = [WriteALU] in {
 def XADD8rr : I<0xC0, MRMDestReg, (outs GR8:$dst), (ins GR8:$src),
                 "xadd{b}\t{$src, $dst|$dst, $src}", [], IIC_XADD_REG>, TB;
 def XADD16rr : I<0xC1, MRMDestReg, (outs GR16:$dst), (ins GR16:$src),
@@ -1350,8 +1431,9 @@ def XADD32rr  : I<0xC1, MRMDestReg, (outs GR32:$dst), (ins GR32:$src),
                  "xadd{l}\t{$src, $dst|$dst, $src}", [], IIC_XADD_REG>, TB;
 def XADD64rr  : RI<0xC1, MRMDestReg, (outs GR64:$dst), (ins GR64:$src),
                    "xadd{q}\t{$src, $dst|$dst, $src}", [], IIC_XADD_REG>, TB;
+} // SchedRW
 
-let mayLoad = 1, mayStore = 1 in {
+let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in {
 def XADD8rm   : I<0xC0, MRMDestMem, (outs), (ins i8mem:$dst, GR8:$src),
                  "xadd{b}\t{$src, $dst|$dst, $src}", [], IIC_XADD_MEM>, TB;
 def XADD16rm  : I<0xC1, MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src),
@@ -1364,6 +1446,7 @@ def XADD64rm  : RI<0xC1, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
 
 }
 
+let SchedRW = [WriteALU] in {
 def CMPXCHG8rr : I<0xB0, MRMDestReg, (outs GR8:$dst), (ins GR8:$src),
                    "cmpxchg{b}\t{$src, $dst|$dst, $src}", [],
                    IIC_CMPXCHG_REG8>, TB;
@@ -1376,7 +1459,9 @@ def CMPXCHG32rr  : I<0xB1, MRMDestReg, (outs GR32:$dst), (ins GR32:$src),
 def CMPXCHG64rr  : RI<0xB1, MRMDestReg, (outs GR64:$dst), (ins GR64:$src),
                       "cmpxchg{q}\t{$src, $dst|$dst, $src}", [],
                       IIC_CMPXCHG_REG>, TB;
+} // SchedRW
 
+let SchedRW = [WriteALULd, WriteRMW] in {
 let mayLoad = 1, mayStore = 1 in {
 def CMPXCHG8rm   : I<0xB0, MRMDestMem, (outs), (ins i8mem:$dst, GR8:$src),
                      "cmpxchg{b}\t{$src, $dst|$dst, $src}", [],
@@ -1400,7 +1485,7 @@ let Defs = [RAX, RDX, EFLAGS], Uses = [RAX, RBX, RCX, RDX] in
 def CMPXCHG16B : RI<0xC7, MRM1m, (outs), (ins i128mem:$dst),
                     "cmpxchg16b\t$dst", [], IIC_CMPXCHG_16B>,
                     TB, Requires<[HasCmpxchg16b]>;
-
+} // SchedRW
 
 
 // Lock instruction prefix
@@ -1423,17 +1508,21 @@ def REPNE_PREFIX : I<0xF2, RawFrm, (outs),  (ins), "repne", []>;
 
 
 // String manipulation instructions
+let SchedRW = [WriteMicrocoded] in {
 def LODSB : I<0xAC, RawFrm, (outs), (ins), "lodsb", [], IIC_LODS>;
 def LODSW : I<0xAD, RawFrm, (outs), (ins), "lodsw", [], IIC_LODS>, OpSize;
 def LODSD : I<0xAD, RawFrm, (outs), (ins), "lods{l|d}", [], IIC_LODS>;
 def LODSQ : RI<0xAD, RawFrm, (outs), (ins), "lodsq", [], IIC_LODS>;
+}
 
+let SchedRW = [WriteSystem] in {
 def OUTSB : I<0x6E, RawFrm, (outs), (ins), "outsb", [], IIC_OUTS>;
 def OUTSW : I<0x6F, RawFrm, (outs), (ins), "outsw", [], IIC_OUTS>, OpSize;
 def OUTSD : I<0x6F, RawFrm, (outs), (ins), "outs{l|d}", [], IIC_OUTS>;
-
+}
 
 // Flag instructions
+let SchedRW = [WriteALU] in {
 def CLC : I<0xF8, RawFrm, (outs), (ins), "clc", [], IIC_CLC>;
 def STC : I<0xF9, RawFrm, (outs), (ins), "stc", [], IIC_STC>;
 def CLI : I<0xFA, RawFrm, (outs), (ins), "cli", [], IIC_CLI>;
@@ -1443,10 +1532,13 @@ def STD : I<0xFD, RawFrm, (outs), (ins), "std", [], IIC_STD>;
 def CMC : I<0xF5, RawFrm, (outs), (ins), "cmc", [], IIC_CMC>;
 
 def CLTS : I<0x06, RawFrm, (outs), (ins), "clts", [], IIC_CLTS>, TB;
+}
 
 // Table lookup instructions
-def XLAT : I<0xD7, RawFrm, (outs), (ins), "xlatb", [], IIC_XLAT>;
+def XLAT : I<0xD7, RawFrm, (outs), (ins), "xlatb", [], IIC_XLAT>,
+           Sched<[WriteLoad]>;
 
+let SchedRW = [WriteMicrocoded] in {
 // ASCII Adjust After Addition
 // sets AL, AH and CF and AF of EFLAGS and uses AL and AF of EFLAGS
 def AAA : I<0x37, RawFrm, (outs), (ins), "aaa", [], IIC_AAA>,
@@ -1476,7 +1568,9 @@ def DAA : I<0x27, RawFrm, (outs), (ins), "daa", [], IIC_DAA>,
 // sets AL, CF and AF of EFLAGS and uses AL, CF and AF of EFLAGS
 def DAS : I<0x2F, RawFrm, (outs), (ins), "das", [], IIC_DAS>,
             Requires<[In32BitMode]>;
+} // SchedRW
 
+let SchedRW = [WriteSystem] in {
 // Check Array Index Against Bounds
 def BOUNDS16rm : I<0x62, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
                    "bound\t{$src, $dst|$dst, $src}", [], IIC_BOUND>, OpSize,
@@ -1486,17 +1580,19 @@ def BOUNDS32rm : I<0x62, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
                    Requires<[In32BitMode]>;
 
 // Adjust RPL Field of Segment Selector
-def ARPL16rr : I<0x63, MRMDestReg, (outs GR16:$src), (ins GR16:$dst),
+def ARPL16rr : I<0x63, MRMDestReg, (outs GR16:$dst), (ins GR16:$src),
                  "arpl\t{$src, $dst|$dst, $src}", [], IIC_ARPL_REG>,
                  Requires<[In32BitMode]>;
-def ARPL16mr : I<0x63, MRMSrcMem, (outs GR16:$src), (ins i16mem:$dst),
+def ARPL16mr : I<0x63, MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src),
                  "arpl\t{$src, $dst|$dst, $src}", [], IIC_ARPL_MEM>,
                  Requires<[In32BitMode]>;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // MOVBE Instructions
 //
 let Predicates = [HasMOVBE] in {
+  let SchedRW = [WriteALULd] in {
   def MOVBE16rm : I<0xF0, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
                     "movbe{w}\t{$src, $dst|$dst, $src}",
                     [(set GR16:$dst, (bswap (loadi16 addr:$src)))], IIC_MOVBE>,
@@ -1509,6 +1605,8 @@ let Predicates = [HasMOVBE] in {
                      "movbe{q}\t{$src, $dst|$dst, $src}",
                      [(set GR64:$dst, (bswap (loadi64 addr:$src)))], IIC_MOVBE>,
                      T8;
+  }
+  let SchedRW = [WriteStore] in {
   def MOVBE16mr : I<0xF1, MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src),
                     "movbe{w}\t{$src, $dst|$dst, $src}",
                     [(store (bswap GR16:$src), addr:$dst)], IIC_MOVBE>,
@@ -1521,6 +1619,7 @@ let Predicates = [HasMOVBE] in {
                      "movbe{q}\t{$src, $dst|$dst, $src}",
                      [(store (bswap GR64:$src), addr:$dst)], IIC_MOVBE>,
                      T8;
+  }
 }
 
 //===----------------------------------------------------------------------===//
@@ -1539,6 +1638,21 @@ let Predicates = [HasRDRAND], Defs = [EFLAGS] in {
 }
 
 //===----------------------------------------------------------------------===//
+// RDSEED Instruction
+//
+let Predicates = [HasRDSEED], Defs = [EFLAGS] in {
+  def RDSEED16r : I<0xC7, MRM7r, (outs GR16:$dst), (ins),
+                    "rdseed{w}\t$dst",
+                    [(set GR16:$dst, EFLAGS, (X86rdseed))]>, OpSize, TB;
+  def RDSEED32r : I<0xC7, MRM7r, (outs GR32:$dst), (ins),
+                    "rdseed{l}\t$dst",
+                    [(set GR32:$dst, EFLAGS, (X86rdseed))]>, TB;
+  def RDSEED64r : RI<0xC7, MRM7r, (outs GR64:$dst), (ins),
+                     "rdseed{q}\t$dst",
+                     [(set GR64:$dst, EFLAGS, (X86rdseed))]>, TB;
+}
+
+//===----------------------------------------------------------------------===//
 // LZCNT Instruction
 //
 let Predicates = [HasLZCNT], Defs = [EFLAGS] in {
@@ -1605,26 +1719,26 @@ multiclass bmi_bls<string mnemonic, Format RegMRM, Format MemMRM,
                   PatFrag ld_frag> {
   def rr : I<0xF3, RegMRM, (outs RC:$dst), (ins RC:$src),
              !strconcat(mnemonic, "\t{$src, $dst|$dst, $src}"),
-             [(set RC:$dst, EFLAGS, (OpNode RC:$src))]>, T8, VEX_4V;
+             [(set RC:$dst, (OpNode RC:$src)), (implicit EFLAGS)]>, T8, VEX_4V;
   def rm : I<0xF3, MemMRM, (outs RC:$dst), (ins x86memop:$src),
              !strconcat(mnemonic, "\t{$src, $dst|$dst, $src}"),
-             [(set RC:$dst, EFLAGS, (OpNode (ld_frag addr:$src)))]>,
+             [(set RC:$dst, (OpNode (ld_frag addr:$src))), (implicit EFLAGS)]>,
              T8, VEX_4V;
 }
 
 let Predicates = [HasBMI], Defs = [EFLAGS] in {
   defm BLSR32 : bmi_bls<"blsr{l}", MRM1r, MRM1m, GR32, i32mem,
-                        X86blsr_flag, loadi32>;
+                        X86blsr, loadi32>;
   defm BLSR64 : bmi_bls<"blsr{q}", MRM1r, MRM1m, GR64, i64mem,
-                        X86blsr_flag, loadi64>, VEX_W;
+                        X86blsr, loadi64>, VEX_W;
   defm BLSMSK32 : bmi_bls<"blsmsk{l}", MRM2r, MRM2m, GR32, i32mem,
-                          X86blsmsk_flag, loadi32>;
+                          X86blsmsk, loadi32>;
   defm BLSMSK64 : bmi_bls<"blsmsk{q}", MRM2r, MRM2m, GR64, i64mem,
-                          X86blsmsk_flag, loadi64>, VEX_W;
+                          X86blsmsk, loadi64>, VEX_W;
   defm BLSI32 : bmi_bls<"blsi{l}", MRM3r, MRM3m, GR32, i32mem,
-                        X86blsi_flag, loadi32>;
+                        X86blsi, loadi32>;
   defm BLSI64 : bmi_bls<"blsi{q}", MRM3r, MRM3m, GR64, i64mem,
-                        X86blsi_flag, loadi64>, VEX_W;
+                        X86blsi, loadi64>, VEX_W;
 }
 
 multiclass bmi_bextr_bzhi<bits<8> opc, string mnemonic, RegisterClass RC,
@@ -1886,6 +2000,8 @@ def : InstAlias<"fmulp",        (MUL_FPrST0  ST1)>;
 def : InstAlias<"fdivp",        (DIVR_FPrST0 ST1)>;
 def : InstAlias<"fdivrp",       (DIV_FPrST0  ST1)>;
 def : InstAlias<"fxch",         (XCH_F       ST1)>;
+def : InstAlias<"fcom",         (COM_FST0r   ST1)>;
+def : InstAlias<"fcomp",        (COMP_FST0r  ST1)>;
 def : InstAlias<"fcomi",        (COM_FIr     ST1)>;
 def : InstAlias<"fcompi",       (COM_FIPr    ST1)>;
 def : InstAlias<"fucom",        (UCOM_Fr     ST1)>;
diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td
index 127af6f7f93a..49721df7c118 100644
--- a/lib/Target/X86/X86InstrMMX.td
+++ b/lib/Target/X86/X86InstrMMX.td
@@ -20,6 +20,7 @@
 // MMX Multiclasses
 //===----------------------------------------------------------------------===//
 
+let Sched = WriteVecALU in {
 def MMX_INTALU_ITINS : OpndItins<
   IIC_MMX_ALU_RR, IIC_MMX_ALU_RM
 >;
@@ -35,11 +36,14 @@ def MMX_PHADDSUBW : OpndItins<
 def MMX_PHADDSUBD : OpndItins<
   IIC_MMX_PHADDSUBD_RR, IIC_MMX_PHADDSUBD_RM
 >;
+}
 
+let Sched = WriteVecIMul in
 def MMX_PMUL_ITINS : OpndItins<
   IIC_MMX_PMUL, IIC_MMX_PMUL
 >;
 
+let Sched = WriteVecALU in {
 def MMX_PSADBW_ITINS : OpndItins<
   IIC_MMX_PSADBW, IIC_MMX_PSADBW
 >;
@@ -47,11 +51,13 @@ def MMX_PSADBW_ITINS : OpndItins<
 def MMX_MISC_FUNC_ITINS : OpndItins<
   IIC_MMX_MISC_FUNC_MEM, IIC_MMX_MISC_FUNC_REG
 >;
+}
 
 def MMX_SHIFT_ITINS : ShiftOpndItins<
   IIC_MMX_SHIFT_RR, IIC_MMX_SHIFT_RM, IIC_MMX_SHIFT_RI
 >;
 
+let Sched = WriteShuffle in {
 def MMX_UNPCK_H_ITINS : OpndItins<
   IIC_MMX_UNPCK_H_RR, IIC_MMX_UNPCK_H_RM
 >;
@@ -67,7 +73,9 @@ def MMX_PCK_ITINS : OpndItins<
 def MMX_PSHUF_ITINS : OpndItins<
   IIC_MMX_PSHUF, IIC_MMX_PSHUF
 >;
+} // Sched
 
+let Sched = WriteCvtF2I in {
 def MMX_CVT_PD_ITINS : OpndItins<
   IIC_MMX_CVT_PD_RR, IIC_MMX_CVT_PD_RM
 >;
@@ -75,6 +83,7 @@ def MMX_CVT_PD_ITINS : OpndItins<
 def MMX_CVT_PS_ITINS : OpndItins<
   IIC_MMX_CVT_PS_RR, IIC_MMX_CVT_PS_RM
 >;
+}
 
 let Constraints = "$src1 = $dst" in {
   // MMXI_binop_rm_int - Simple MMX binary operator based on intrinsic.
@@ -84,7 +93,8 @@ let Constraints = "$src1 = $dst" in {
     def irr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
                  (ins VR64:$src1, VR64:$src2),
                  !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
-                 [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr> {
+                 [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>,
+              Sched<[itins.Sched]> {
       let isCommutable = Commutable;
     }
     def irm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
@@ -92,7 +102,7 @@ let Constraints = "$src1 = $dst" in {
                  !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
                  [(set VR64:$dst, (IntId VR64:$src1,
                                    (bitconvert (load_mmx addr:$src2))))],
-                 itins.rm>;
+                 itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
   }
 
   multiclass MMXI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
@@ -101,17 +111,19 @@ let Constraints = "$src1 = $dst" in {
     def rr : MMXI<opc, MRMSrcReg, (outs VR64:$dst),
                                   (ins VR64:$src1, VR64:$src2),
                   !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
-                  [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>;
+                  [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>,
+             Sched<[WriteVecShift]>;
     def rm : MMXI<opc, MRMSrcMem, (outs VR64:$dst),
                                   (ins VR64:$src1, i64mem:$src2),
                   !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
                   [(set VR64:$dst, (IntId VR64:$src1,
                                     (bitconvert (load_mmx addr:$src2))))],
-                  itins.rm>;
+                  itins.rm>, Sched<[WriteVecShiftLd, ReadAfterLd]>;
     def ri : MMXIi8<opc2, ImmForm, (outs VR64:$dst),
                                    (ins VR64:$src1, i32i8imm:$src2),
                     !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
-           [(set VR64:$dst, (IntId2 VR64:$src1, (i32 imm:$src2)))], itins.ri>;
+           [(set VR64:$dst, (IntId2 VR64:$src1, (i32 imm:$src2)))], itins.ri>,
+           Sched<[WriteVecShift]>;
   }
 }
 
@@ -120,13 +132,14 @@ multiclass SS3I_unop_rm_int_mm<bits<8> opc, string OpcodeStr,
                                Intrinsic IntId64, OpndItins itins> {
   def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
                    !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                   [(set VR64:$dst, (IntId64 VR64:$src))], itins.rr>;
+                   [(set VR64:$dst, (IntId64 VR64:$src))], itins.rr>,
+             Sched<[itins.Sched]>;
 
   def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
                    !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                    [(set VR64:$dst,
                      (IntId64 (bitconvert (memopmmx addr:$src))))],
-                   itins.rm>;
+                   itins.rm>, Sched<[itins.Sched.Folded]>;
 }
 
 /// Binary MMX instructions requiring SSSE3.
@@ -137,13 +150,15 @@ multiclass SS3I_binop_rm_int_mm<bits<8> opc, string OpcodeStr,
   def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst),
        (ins VR64:$src1, VR64:$src2),
         !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
-       [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))], itins.rr>;
+       [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))], itins.rr>,
+      Sched<[itins.Sched]>;
   def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst),
        (ins VR64:$src1, i64mem:$src2),
         !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
        [(set VR64:$dst,
          (IntId64 VR64:$src1,
-          (bitconvert (memopmmx addr:$src2))))], itins.rm>;
+          (bitconvert (memopmmx addr:$src2))))], itins.rm>,
+      Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 }
 
@@ -164,9 +179,11 @@ multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
                          Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
                          string asm, OpndItins itins, Domain d> {
   def irr : MMXPI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
-                  [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr, d>;
+                  [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr, d>,
+            Sched<[itins.Sched]>;
   def irm : MMXPI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
-                  [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm, d>;
+                  [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm, d>,
+            Sched<[itins.Sched.Folded]>;
 }
 
 multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
@@ -174,11 +191,11 @@ multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
                     PatFrag ld_frag, string asm, Domain d> {
   def irr : PI<opc, MRMSrcReg, (outs DstRC:$dst),(ins DstRC:$src1, SrcRC:$src2),
               asm, [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))], 
-              IIC_DEFAULT, d>;
+              NoItinerary, d>;
   def irm : PI<opc, MRMSrcMem, (outs DstRC:$dst),
                    (ins DstRC:$src1, x86memop:$src2), asm,
               [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))], 
-              IIC_DEFAULT, d>;
+              NoItinerary, d>;
 }
 
 //===----------------------------------------------------------------------===//
@@ -197,16 +214,17 @@ def MMX_MOVD64rr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
                         "movd\t{$src, $dst|$dst, $src}",
                         [(set VR64:$dst, 
                          (x86mmx (scalar_to_vector GR32:$src)))],
-                        IIC_MMX_MOV_MM_RM>;
+                        IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>;
 let canFoldAsLoad = 1 in
 def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src),
                         "movd\t{$src, $dst|$dst, $src}",
                         [(set VR64:$dst,
                         (x86mmx (scalar_to_vector (loadi32 addr:$src))))],
-                        IIC_MMX_MOV_MM_RM>;
+                        IIC_MMX_MOV_MM_RM>, Sched<[WriteLoad]>;
 let mayStore = 1 in
 def MMX_MOVD64mr : MMXI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR64:$src),
-                        "movd\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOV_MM_RM>;
+                        "movd\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOV_MM_RM>,
+                   Sched<[WriteStore]>;
 
 // Low word of MMX to GPR.
 def MMX_X86movd2w : SDNode<"X86ISD::MMX_MOVD2W", SDTypeProfile<1, 1,
@@ -214,16 +232,18 @@ def MMX_X86movd2w : SDNode<"X86ISD::MMX_MOVD2W", SDTypeProfile<1, 1,
 def MMX_MOVD64grr : MMXI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR64:$src),
                          "movd\t{$src, $dst|$dst, $src}",
                          [(set GR32:$dst,
-                          (MMX_X86movd2w (x86mmx VR64:$src)))], IIC_MMX_MOV_REG_MM>;
+                          (MMX_X86movd2w (x86mmx VR64:$src)))],
+                          IIC_MMX_MOV_REG_MM>, Sched<[WriteMove]>;
 
 let neverHasSideEffects = 1 in
 def MMX_MOVD64to64rr : MMXRI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src),
                              "movd\t{$src, $dst|$dst, $src}",
-                             [], IIC_MMX_MOV_MM_RM>;
+                             [], IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>;
 
 // These are 64 bit moves, but since the OS X assembler doesn't
 // recognize a register-register movq, we write them as
 // movd.
+let SchedRW = [WriteMove] in {
 def MMX_MOVD64from64rr : MMXRI<0x7E, MRMDestReg,
                                (outs GR64:$dst), (ins VR64:$src),
                                "movd\t{$src, $dst|$dst, $src}", 
@@ -237,6 +257,9 @@ let neverHasSideEffects = 1 in
 def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
                         "movq\t{$src, $dst|$dst, $src}", [],
                         IIC_MMX_MOVQ_RR>;
+} // SchedRW
+
+let SchedRW = [WriteLoad] in {
 let canFoldAsLoad = 1 in
 def MMX_MOVQ64rm : MMXI<0x6F, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
                         "movq\t{$src, $dst|$dst, $src}",
@@ -246,7 +269,9 @@ def MMX_MOVQ64mr : MMXI<0x7F, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
                         "movq\t{$src, $dst|$dst, $src}",
                         [(store (x86mmx VR64:$src), addr:$dst)],
                         IIC_MMX_MOVQ_RM>;
+} // SchedRW
 
+let SchedRW = [WriteMove] in {
 def MMX_MOVDQ2Qrr : MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
                              (ins VR128:$src), "movdq2q\t{$src, $dst|$dst, $src}",
                              [(set VR64:$dst,
@@ -271,11 +296,12 @@ def MMX_MOVQ2FR64rr: MMXS2SIi8<0xD6, MRMSrcReg, (outs FR64:$dst),
 def MMX_MOVFR642Qrr: MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst),
                               (ins FR64:$src), "movdq2q\t{$src, $dst|$dst, $src}",
                               [], IIC_MMX_MOVQ_RR>;
+} // SchedRW
 
 def MMX_MOVNTQmr  : MMXI<0xE7, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src),
                          "movntq\t{$src, $dst|$dst, $src}",
                          [(int_x86_mmx_movnt_dq addr:$dst, VR64:$src)],
-                         IIC_MMX_MOVQ_RM>;
+                         IIC_MMX_MOVQ_RM>, Sched<[WriteStore]>;
 
 let AddedComplexity = 15 in
 // movd to MMX register zero-extends
@@ -283,7 +309,7 @@ def MMX_MOVZDI2PDIrr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
                              "movd\t{$src, $dst|$dst, $src}",
               [(set VR64:$dst,
                     (x86mmx (X86vzmovl (x86mmx (scalar_to_vector GR32:$src)))))],
-                            IIC_MMX_MOV_MM_RM>;
+                            IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>;
 let AddedComplexity = 20 in
 def MMX_MOVZDI2PDIrm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst),
                            (ins i32mem:$src),
@@ -291,7 +317,7 @@ def MMX_MOVZDI2PDIrm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst),
           [(set VR64:$dst,
                 (x86mmx (X86vzmovl (x86mmx
                                    (scalar_to_vector (loadi32 addr:$src))))))],
-                            IIC_MMX_MOV_MM_RM>;
+                            IIC_MMX_MOV_MM_RM>, Sched<[WriteLoad]>;
 
 // Arithmetic Instructions
 defm MMX_PABSB : SS3I_unop_rm_int_mm<0x1C, "pabsb", int_x86_ssse3_pabs_b,
@@ -491,14 +517,14 @@ def MMX_PSHUFWri : MMXIi8<0x70, MRMSrcReg,
                           "pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                           [(set VR64:$dst,
                              (int_x86_sse_pshuf_w VR64:$src1, imm:$src2))],
-                          IIC_MMX_PSHUF>;
+                          IIC_MMX_PSHUF>, Sched<[WriteShuffle]>;
 def MMX_PSHUFWmi : MMXIi8<0x70, MRMSrcMem,
                           (outs VR64:$dst), (ins i64mem:$src1, i8imm:$src2),
                           "pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                           [(set VR64:$dst,
                              (int_x86_sse_pshuf_w (load_mmx addr:$src1),
                                                    imm:$src2))],
-                          IIC_MMX_PSHUF>;
+                          IIC_MMX_PSHUF>, Sched<[WriteShuffleLd]>;
 
 
 
@@ -532,7 +558,7 @@ def MMX_PEXTRWirri: MMXIi8<0xC5, MRMSrcReg,
                            "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                            [(set GR32:$dst, (int_x86_mmx_pextr_w VR64:$src1,
                                              (iPTR imm:$src2)))],
-                           IIC_MMX_PEXTR>;
+                           IIC_MMX_PEXTR>, Sched<[WriteShuffle]>;
 let Constraints = "$src1 = $dst" in {
   def MMX_PINSRWirri : MMXIi8<0xC4, MRMSrcReg,
                       (outs VR64:$dst), 
@@ -540,7 +566,7 @@ let Constraints = "$src1 = $dst" in {
                       "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
                       [(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1,
                                         GR32:$src2, (iPTR imm:$src3)))],
-                      IIC_MMX_PINSRW>;
+                      IIC_MMX_PINSRW>, Sched<[WriteShuffle]>;
 
   def MMX_PINSRWirmi : MMXIi8<0xC4, MRMSrcMem,
                      (outs VR64:$dst),
@@ -549,7 +575,7 @@ let Constraints = "$src1 = $dst" in {
                      [(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1,
                                          (i32 (anyext (loadi16 addr:$src2))),
                                        (iPTR imm:$src3)))],
-                     IIC_MMX_PINSRW>;
+                     IIC_MMX_PINSRW>, Sched<[WriteShuffleLd, ReadAfterLd]>;
 }
 
 // Mask creation
@@ -570,6 +596,7 @@ def : Pat<(x86mmx (MMX_X86movdq2q (loadv2i64 addr:$src))),
           (x86mmx (MMX_MOVQ64rm addr:$src))>;
 
 // Misc.
+let SchedRW = [WriteShuffle] in {
 let Uses = [EDI] in
 def MMX_MASKMOVQ : MMXI<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
                         "maskmovq\t{$mask, $src|$src, $mask}",
@@ -580,6 +607,7 @@ def MMX_MASKMOVQ64: MMXI64<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
                            "maskmovq\t{$mask, $src|$src, $mask}",
                            [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, RDI)],
                            IIC_MMX_MASKMOV>;
+}
 
 // 64-bit bit convert.
 let Predicates = [HasSSE2] in {
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 6f48d7ed7fe1..384238741b18 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -16,6 +16,8 @@
 class OpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm> {
   InstrItinClass rr = arg_rr;
   InstrItinClass rm = arg_rm;
+  // InstrSchedModel info.
+  X86FoldableSchedWrite Sched = WriteFAdd;
 }
 
 class SizeItins<OpndItins arg_s, OpndItins arg_d> {
@@ -33,6 +35,7 @@ class ShiftOpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm,
 
 
 // scalar
+let Sched = WriteFAdd in {
 def SSE_ALU_F32S : OpndItins<
   IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM
 >;
@@ -40,11 +43,13 @@ def SSE_ALU_F32S : OpndItins<
 def SSE_ALU_F64S : OpndItins<
   IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM
 >;
+}
 
 def SSE_ALU_ITINS_S : SizeItins<
   SSE_ALU_F32S, SSE_ALU_F64S
 >;
 
+let Sched = WriteFMul in {
 def SSE_MUL_F32S : OpndItins<
   IIC_SSE_MUL_F32S_RR, IIC_SSE_MUL_F64S_RM
 >;
@@ -52,11 +57,13 @@ def SSE_MUL_F32S : OpndItins<
 def SSE_MUL_F64S : OpndItins<
   IIC_SSE_MUL_F64S_RR, IIC_SSE_MUL_F64S_RM
 >;
+}
 
 def SSE_MUL_ITINS_S : SizeItins<
   SSE_MUL_F32S, SSE_MUL_F64S
 >;
 
+let Sched = WriteFDiv in {
 def SSE_DIV_F32S : OpndItins<
   IIC_SSE_DIV_F32S_RR, IIC_SSE_DIV_F64S_RM
 >;
@@ -64,12 +71,14 @@ def SSE_DIV_F32S : OpndItins<
 def SSE_DIV_F64S : OpndItins<
   IIC_SSE_DIV_F64S_RR, IIC_SSE_DIV_F64S_RM
 >;
+}
 
 def SSE_DIV_ITINS_S : SizeItins<
   SSE_DIV_F32S, SSE_DIV_F64S
 >;
 
 // parallel
+let Sched = WriteFAdd in {
 def SSE_ALU_F32P : OpndItins<
   IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM
 >;
@@ -77,11 +86,13 @@ def SSE_ALU_F32P : OpndItins<
 def SSE_ALU_F64P : OpndItins<
   IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM
 >;
+}
 
 def SSE_ALU_ITINS_P : SizeItins<
   SSE_ALU_F32P, SSE_ALU_F64P
 >;
 
+let Sched = WriteFMul in {
 def SSE_MUL_F32P : OpndItins<
   IIC_SSE_MUL_F32P_RR, IIC_SSE_MUL_F64P_RM
 >;
@@ -89,11 +100,13 @@ def SSE_MUL_F32P : OpndItins<
 def SSE_MUL_F64P : OpndItins<
   IIC_SSE_MUL_F64P_RR, IIC_SSE_MUL_F64P_RM
 >;
+}
 
 def SSE_MUL_ITINS_P : SizeItins<
   SSE_MUL_F32P, SSE_MUL_F64P
 >;
 
+let Sched = WriteFDiv in {
 def SSE_DIV_F32P : OpndItins<
   IIC_SSE_DIV_F32P_RR, IIC_SSE_DIV_F64P_RM
 >;
@@ -101,6 +114,7 @@ def SSE_DIV_F32P : OpndItins<
 def SSE_DIV_F64P : OpndItins<
   IIC_SSE_DIV_F64P_RR, IIC_SSE_DIV_F64P_RM
 >;
+}
 
 def SSE_DIV_ITINS_P : SizeItins<
   SSE_DIV_F32P, SSE_DIV_F64P
@@ -110,6 +124,7 @@ def SSE_BIT_ITINS_P : OpndItins<
   IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
 >;
 
+let Sched = WriteVecALU in {
 def SSE_INTALU_ITINS_P : OpndItins<
   IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
 >;
@@ -117,7 +132,9 @@ def SSE_INTALU_ITINS_P : OpndItins<
 def SSE_INTALUQ_ITINS_P : OpndItins<
   IIC_SSE_INTALUQ_P_RR, IIC_SSE_INTALUQ_P_RM
 >;
+}
 
+let Sched = WriteVecIMul in
 def SSE_INTMUL_ITINS_P : OpndItins<
   IIC_SSE_INTMUL_P_RR, IIC_SSE_INTMUL_P_RM
 >;
@@ -148,13 +165,15 @@ multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>;
+       [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>,
+       Sched<[itins.Sched]>;
   }
   def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>;
+       [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>,
+       Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
@@ -169,14 +188,16 @@ multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
            !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst, (!cast<Intrinsic>(
                  !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
-             RC:$src1, RC:$src2))], itins.rr>;
+             RC:$src1, RC:$src2))], itins.rr>,
+       Sched<[itins.Sched]>;
   def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
        !if(Is2Addr,
            !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
                                           SSEVer, "_", OpcodeStr, FPSizeStr))
-             RC:$src1, mem_cpat:$src2))], itins.rm>;
+             RC:$src1, mem_cpat:$src2))], itins.rm>,
+       Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
@@ -189,54 +210,36 @@ multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>;
+       [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>,
+       Sched<[itins.Sched]>;
   let mayLoad = 1 in
     def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
-          itins.rm, d>;
+          itins.rm, d>,
+       Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
                                       string OpcodeStr, X86MemOperand x86memop,
                                       list<dag> pat_rr, list<dag> pat_rm,
-                                      bit Is2Addr = 1,
-                                      bit rr_hasSideEffects = 0> {
-  let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
+                                      bit Is2Addr = 1> {
+  let isCommutable = 1, hasSideEffects = 0 in
     def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       pat_rr, IIC_DEFAULT, d>;
+       pat_rr, NoItinerary, d>,
+       Sched<[WriteVecLogic]>;
   def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       pat_rm, IIC_DEFAULT, d>;
-}
-
-/// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
-multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
-                           string asm, string SSEVer, string FPSizeStr,
-                           X86MemOperand x86memop, PatFrag mem_frag,
-                           Domain d, OpndItins itins, bit Is2Addr = 1> {
-  def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
-       !if(Is2Addr,
-           !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
-           !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-           [(set RC:$dst, (!cast<Intrinsic>(
-                     !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
-                 RC:$src1, RC:$src2))], IIC_DEFAULT, d>;
-  def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
-       !if(Is2Addr,
-           !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
-           !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       [(set RC:$dst, (!cast<Intrinsic>(
-                     !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
-             RC:$src1, (mem_frag addr:$src2)))], IIC_DEFAULT, d>;
+       pat_rm, NoItinerary, d>,
+       Sched<[WriteVecLogicLd, ReadAfterLd]>;
 }
 
 //===----------------------------------------------------------------------===//
@@ -367,7 +370,7 @@ let Predicates = [HasAVX] in {
 // Alias instructions that map fld0 to xorps for sse or vxorps for avx.
 // This is expanded by ExpandPostRAPseudos.
 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
-    isPseudo = 1 in {
+    isPseudo = 1, SchedRW = [WriteZero] in {
   def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
                    [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
   def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
@@ -384,7 +387,7 @@ let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
 // We set canFoldAsLoad because this can be converted to a constant-pool
 // load of an all-zeros value if folding it would be beneficial.
 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
-    isPseudo = 1 in {
+    isPseudo = 1, SchedRW = [WriteZero] in {
 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
                [(set VR128:$dst, (v4f32 immAllZerosV))]>;
 }
@@ -401,7 +404,7 @@ def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
 // at the rename stage without using any execution unit, so SET0PSY
 // and SET0PDY can be used for vector int instructions without penalty
 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
-    isPseudo = 1, Predicates = [HasAVX] in {
+    isPseudo = 1, Predicates = [HasAVX], SchedRW = [WriteZero] in {
 def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
                  [(set VR256:$dst, (v8f32 immAllZerosV))]>;
 }
@@ -439,7 +442,7 @@ def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
 // We set canFoldAsLoad because this can be converted to a constant-pool
 // load of an all-ones value if folding it would be beneficial.
 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
-    isPseudo = 1 in {
+    isPseudo = 1, SchedRW = [WriteZero] in {
   def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
                        [(set VR128:$dst, (v4i32 immAllOnesV))]>;
   let Predicates = [HasAVX2] in
@@ -458,93 +461,70 @@ let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
 //===----------------------------------------------------------------------===//
 
-class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
-      SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
-      [(set VR128:$dst, (vt (OpNode VR128:$src1,
-                             (scalar_to_vector RC:$src2))))],
-      IIC_SSE_MOV_S_RR>;
+multiclass sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt,
+                         X86MemOperand x86memop, string base_opc,
+                         string asm_opr> {
+  def rr : SI<0x10, MRMSrcReg, (outs VR128:$dst),
+              (ins VR128:$src1, RC:$src2),
+              !strconcat(base_opc, asm_opr),
+              [(set VR128:$dst, (vt (OpNode VR128:$src1,
+                                 (scalar_to_vector RC:$src2))))],
+              IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
 
-// Loading from memory automatically zeroing upper bits.
-class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
-                    PatFrag mem_pat, string OpcodeStr> :
-      SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
-         !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                        [(set RC:$dst, (mem_pat addr:$src))],
-                        IIC_SSE_MOV_S_RM>;
-
-// AVX
-def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
-                "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
-                VEX_LIG;
-def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
-                "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
-                VEX_LIG;
-
-// For the disassembler
-let isCodeGenOnly = 1 in {
-  def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
-                        (ins VR128:$src1, FR32:$src2),
-                        "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
-                        IIC_SSE_MOV_S_RR>,
-                        XS, VEX_4V, VEX_LIG;
-  def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
-                        (ins VR128:$src1, FR64:$src2),
-                        "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
-                        IIC_SSE_MOV_S_RR>,
-                        XD, VEX_4V, VEX_LIG;
+  // For the disassembler
+  let isCodeGenOnly = 1, hasSideEffects = 0 in
+  def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
+                  (ins VR128:$src1, RC:$src2),
+                  !strconcat(base_opc, asm_opr),
+                  [], IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
 }
 
-let canFoldAsLoad = 1, isReMaterializable = 1 in {
-  def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
-                 VEX_LIG;
-  let AddedComplexity = 20 in
-    def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
-                   VEX_LIG;
-}
+multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt,
+                      X86MemOperand x86memop, string OpcodeStr> {
+  // AVX
+  defm V#NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
+                              "\t{$src2, $src1, $dst|$dst, $src1, $src2}">,
+                              VEX_4V, VEX_LIG;
 
-def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
-                  "movss\t{$src, $dst|$dst, $src}",
-                  [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
-                  XS, VEX, VEX_LIG;
-def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
-                  "movsd\t{$src, $dst|$dst, $src}",
-                  [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
-                  XD, VEX, VEX_LIG;
+  def V#NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
+                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+                     [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
+                     VEX, VEX_LIG, Sched<[WriteStore]>;
+  // SSE1 & 2
+  let Constraints = "$src1 = $dst" in {
+    defm NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
+                              "\t{$src2, $dst|$dst, $src2}">;
+  }
 
-// SSE1 & 2
-let Constraints = "$src1 = $dst" in {
-  def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
-                          "movss\t{$src2, $dst|$dst, $src2}">, XS;
-  def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
-                          "movsd\t{$src2, $dst|$dst, $src2}">, XD;
+  def NAME#mr   : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
+                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+                     [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
+                  Sched<[WriteStore]>;
+}
 
-  // For the disassembler
-  let isCodeGenOnly = 1 in {
-    def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
-                         (ins VR128:$src1, FR32:$src2),
-                         "movss\t{$src2, $dst|$dst, $src2}", [],
-                         IIC_SSE_MOV_S_RR>, XS;
-    def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
-                         (ins VR128:$src1, FR64:$src2),
-                         "movsd\t{$src2, $dst|$dst, $src2}", [],
-                         IIC_SSE_MOV_S_RR>, XD;
-  }
+// Loading from memory automatically zeroing upper bits.
+multiclass sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
+                         PatFrag mem_pat, string OpcodeStr> {
+  def V#NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
+                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+                     [(set RC:$dst, (mem_pat addr:$src))],
+                     IIC_SSE_MOV_S_RM>, VEX, VEX_LIG, Sched<[WriteLoad]>;
+  def NAME#rm   : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
+                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+                     [(set RC:$dst, (mem_pat addr:$src))],
+                     IIC_SSE_MOV_S_RM>, Sched<[WriteLoad]>;
 }
 
+defm MOVSS : sse12_move<FR32, X86Movss, v4f32, f32mem, "movss">, XS;
+defm MOVSD : sse12_move<FR64, X86Movsd, v2f64, f64mem, "movsd">, XD;
+
 let canFoldAsLoad = 1, isReMaterializable = 1 in {
-  def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
+  defm MOVSS : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
 
   let AddedComplexity = 20 in
-    def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
+    defm MOVSD : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
 }
 
-def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
-                  "movss\t{$src, $dst|$dst, $src}",
-                  [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
-def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
-                  "movsd\t{$src, $dst|$dst, $src}",
-                  [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
-
 // Patterns
 let Predicates = [HasAVX] in {
   let AddedComplexity = 15 in {
@@ -791,11 +771,13 @@ multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
                             bit IsReMaterializable = 1> {
 let neverHasSideEffects = 1 in
   def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
-              !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>;
+              !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>,
+           Sched<[WriteMove]>;
 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
   def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
               !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
-                   [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>;
+                   [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>,
+           Sched<[WriteLoad]>;
 }
 
 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
@@ -836,6 +818,7 @@ defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
                               "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
                               TB, OpSize;
 
+let SchedRW = [WriteStore] in {
 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
                    "movaps\t{$src, $dst|$dst, $src}",
                    [(alignedstore (v4f32 VR128:$src), addr:$dst)],
@@ -868,9 +851,10 @@ def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
                    "movupd\t{$src, $dst|$dst, $src}",
                    [(store (v4f64 VR256:$src), addr:$dst)],
                    IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
+} // SchedRW
 
 // For disassembler
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
   def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
                           (ins VR128:$src),
                           "movaps\t{$src, $dst|$dst, $src}", [],
@@ -926,6 +910,7 @@ def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
           (VMOVUPDYmr addr:$dst, VR256:$src)>;
 
+let SchedRW = [WriteStore] in {
 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
                    "movaps\t{$src, $dst|$dst, $src}",
                    [(alignedstore (v4f32 VR128:$src), addr:$dst)],
@@ -942,9 +927,10 @@ def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
                    "movupd\t{$src, $dst|$dst, $src}",
                    [(store (v2f64 VR128:$src), addr:$dst)],
                    IIC_SSE_MOVU_P_MR>;
+} // SchedRW
 
 // For disassembler
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
   def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
                          "movaps\t{$src, $dst|$dst, $src}", [],
                          IIC_SSE_MOVA_P_RR>;
@@ -1055,7 +1041,7 @@ let Predicates = [HasAVX] in {
             (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
   def : Pat<(store (v8i16 (extract_subvector
                            (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
-            (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+            (VMOVUPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
   def : Pat<(store (v16i8 (extract_subvector
                            (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
             (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
@@ -1090,7 +1076,7 @@ let Predicates = [UseSSE1] in {
 
 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
 // bits are disregarded. FIXME: Set encoding to pseudo!
-let neverHasSideEffects = 1 in {
+let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
                        "movaps\t{$src, $dst|$dst, $src}", [],
                        IIC_SSE_MOVA_P_RR>, VEX;
@@ -1107,7 +1093,7 @@ def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
 
 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
 // bits are disregarded. FIXME: Set encoding to pseudo!
-let canFoldAsLoad = 1, isReMaterializable = 1 in {
+let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in {
 let isCodeGenOnly = 1 in {
   def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
                          "movaps\t{$src, $dst|$dst, $src}",
@@ -1132,36 +1118,46 @@ def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
 // SSE 1 & 2 - Move Low packed FP Instructions
 //===----------------------------------------------------------------------===//
 
-multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
-                                 SDNode psnode, SDNode pdnode, string base_opc,
-                                 string asm_opr, InstrItinClass itin> {
+multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
+                                      string base_opc, string asm_opr,
+                                      InstrItinClass itin> {
   def PSrm : PI<opc, MRMSrcMem,
          (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
          !strconcat(base_opc, "s", asm_opr),
-     [(set RC:$dst,
-       (psnode RC:$src1,
+     [(set VR128:$dst,
+       (psnode VR128:$src1,
               (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
-              itin, SSEPackedSingle>, TB;
+              itin, SSEPackedSingle>, TB,
+     Sched<[WriteShuffleLd, ReadAfterLd]>;
 
   def PDrm : PI<opc, MRMSrcMem,
-         (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
+         (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
          !strconcat(base_opc, "d", asm_opr),
-     [(set RC:$dst, (v2f64 (pdnode RC:$src1,
+     [(set VR128:$dst, (v2f64 (pdnode VR128:$src1,
                               (scalar_to_vector (loadf64 addr:$src2)))))],
-              itin, SSEPackedDouble>, TB, OpSize;
+              itin, SSEPackedDouble>, TB, OpSize,
+     Sched<[WriteShuffleLd, ReadAfterLd]>;
+
 }
 
-let AddedComplexity = 20 in {
-  defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
-                     "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                     IIC_SSE_MOV_LH>, VEX_4V;
+multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
+                                 string base_opc, InstrItinClass itin> {
+  defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
+                                    "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+                                    itin>, VEX_4V;
+
+let Constraints = "$src1 = $dst" in
+  defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
+                                    "\t{$src2, $dst|$dst, $src2}",
+                                    itin>;
 }
-let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
-  defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
-                                   "\t{$src2, $dst|$dst, $src2}",
-                                   IIC_SSE_MOV_LH>;
+
+let AddedComplexity = 20 in {
+  defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp",
+                                    IIC_SSE_MOV_LH>;
 }
 
+let SchedRW = [WriteStore] in {
 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    "movlps\t{$src, $dst|$dst, $src}",
                    [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
@@ -1182,6 +1178,7 @@ def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    [(store (f64 (vector_extract (v2f64 VR128:$src),
                                  (iPTR 0))), addr:$dst)],
                                  IIC_SSE_MOV_LH>;
+} // SchedRW
 
 let Predicates = [HasAVX] in {
   // Shuffle with VMOVLPS
@@ -1257,16 +1254,11 @@ let Predicates = [UseSSE2] in {
 //===----------------------------------------------------------------------===//
 
 let AddedComplexity = 20 in {
-  defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
-                     "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                     IIC_SSE_MOV_LH>, VEX_4V;
-}
-let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
-  defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
-                                   "\t{$src2, $dst|$dst, $src2}",
-                                   IIC_SSE_MOV_LH>;
+  defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp",
+                                    IIC_SSE_MOV_LH>;
 }
 
+let SchedRW = [WriteStore] in {
 // v2f64 extract element 1 is always custom lowered to unpack high to low
 // and extract element 0 so the non-store version isn't too horrible.
 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
@@ -1291,6 +1283,7 @@ def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    [(store (f64 (vector_extract
                                  (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
                                  (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
+} // SchedRW
 
 let Predicates = [HasAVX] in {
   // VMOVHPS patterns
@@ -1341,14 +1334,14 @@ let AddedComplexity = 20 in {
                       [(set VR128:$dst,
                         (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
                         IIC_SSE_MOV_LH>,
-                      VEX_4V;
+                      VEX_4V, Sched<[WriteShuffle]>;
   def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
                                        (ins VR128:$src1, VR128:$src2),
                       "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                       [(set VR128:$dst,
                         (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
                         IIC_SSE_MOV_LH>,
-                      VEX_4V;
+                      VEX_4V, Sched<[WriteShuffle]>;
 }
 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
   def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
@@ -1356,13 +1349,13 @@ let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
                       "movlhps\t{$src2, $dst|$dst, $src2}",
                       [(set VR128:$dst,
                         (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
-                        IIC_SSE_MOV_LH>;
+                        IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
   def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
                                        (ins VR128:$src1, VR128:$src2),
                       "movhlps\t{$src2, $dst|$dst, $src2}",
                       [(set VR128:$dst,
                         (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
-                        IIC_SSE_MOV_LH>;
+                        IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
 }
 
 let Predicates = [HasAVX] in {
@@ -1397,22 +1390,27 @@ def SSE_CVT_PD : OpndItins<
   IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
 >;
 
+let Sched = WriteCvtI2F in
 def SSE_CVT_PS : OpndItins<
   IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
 >;
 
+let Sched = WriteCvtI2F in
 def SSE_CVT_Scalar : OpndItins<
   IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
 >;
 
+let Sched = WriteCvtF2I in
 def SSE_CVT_SS2SI_32 : OpndItins<
   IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
 >;
 
+let Sched = WriteCvtF2I in
 def SSE_CVT_SS2SI_64 : OpndItins<
   IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
 >;
 
+let Sched = WriteCvtF2I in
 def SSE_CVT_SD2SI : OpndItins<
   IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
 >;
@@ -1422,10 +1420,10 @@ multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
                      string asm, OpndItins itins> {
   def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
                         [(set DstRC:$dst, (OpNode SrcRC:$src))],
-                        itins.rr>;
+                        itins.rr>, Sched<[itins.Sched]>;
   def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
                         [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
-                        itins.rm>;
+                        itins.rm>, Sched<[itins.Sched.Folded]>;
 }
 
 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
@@ -1433,10 +1431,10 @@ multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
                        OpndItins itins> {
 let neverHasSideEffects = 1 in {
   def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
-             [], itins.rr, d>;
+             [], itins.rr, d>, Sched<[itins.Sched]>;
   let mayLoad = 1 in
   def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
-             [], itins.rm, d>;
+             [], itins.rm, d>, Sched<[itins.Sched.Folded]>;
 }
 }
 
@@ -1444,11 +1442,13 @@ multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
                           X86MemOperand x86memop, string asm> {
 let neverHasSideEffects = 1 in {
   def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
-              !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
+              !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
+           Sched<[WriteCvtI2F]>;
   let mayLoad = 1 in
   def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
               (ins DstRC:$src1, x86memop:$src),
-              !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
+              !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
+           Sched<[WriteCvtI2FLd, ReadAfterLd]>;
 } // neverHasSideEffects = 1
 }
 
@@ -1457,7 +1457,7 @@ defm VCVTTSS2SI   : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
                                 SSE_CVT_SS2SI_32>,
                                 XS, VEX, VEX_LIG;
 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
-                                "cvttss2si{q}\t{$src, $dst|$dst, $src}",
+                                "cvttss2si\t{$src, $dst|$dst, $src}",
                                 SSE_CVT_SS2SI_64>,
                                 XS, VEX, VEX_W, VEX_LIG;
 defm VCVTTSD2SI   : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
@@ -1465,26 +1465,43 @@ defm VCVTTSD2SI   : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
                                 SSE_CVT_SD2SI>,
                                 XD, VEX, VEX_LIG;
 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
-                                "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+                                "cvttsd2si\t{$src, $dst|$dst, $src}",
                                 SSE_CVT_SD2SI>,
                                 XD, VEX, VEX_W, VEX_LIG;
 
+def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
+                (VCVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
+def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
+                (VCVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
+def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
+                (VCVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
+def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
+                (VCVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
+def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
+                (VCVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
+def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
+                (VCVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
+def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
+                (VCVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
+def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
+                (VCVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
+
 // The assembler can recognize rr 64-bit instructions by seeing a rxx
 // register, but the same isn't true when only using memory operands,
 // provide other assembly "l" and "q" forms to address this explicitly
 // where appropriate to do so.
-defm VCVTSI2SS   : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
+defm VCVTSI2SS   : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss{l}">,
                                   XS, VEX_4V, VEX_LIG;
 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
                                   XS, VEX_4V, VEX_W, VEX_LIG;
-defm VCVTSI2SD   : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
+defm VCVTSI2SD   : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
                                   XD, VEX_4V, VEX_LIG;
 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
                                   XD, VEX_4V, VEX_W, VEX_LIG;
 
-def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
-                (VCVTSI2SDrr FR64:$dst, FR64:$src1, GR32:$src)>;
-def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
+def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}",
+                (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src)>;
+def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}",
                 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
 
 let Predicates = [HasAVX] in {
@@ -1511,27 +1528,49 @@ defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
                       "cvttss2si\t{$src, $dst|$dst, $src}",
                       SSE_CVT_SS2SI_32>, XS;
 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
-                      "cvttss2si{q}\t{$src, $dst|$dst, $src}",
+                      "cvttss2si\t{$src, $dst|$dst, $src}",
                       SSE_CVT_SS2SI_64>, XS, REX_W;
 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
                       "cvttsd2si\t{$src, $dst|$dst, $src}",
                       SSE_CVT_SD2SI>, XD;
 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
-                      "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+                      "cvttsd2si\t{$src, $dst|$dst, $src}",
                       SSE_CVT_SD2SI>, XD, REX_W;
 defm CVTSI2SS  : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
-                      "cvtsi2ss\t{$src, $dst|$dst, $src}",
+                      "cvtsi2ss{l}\t{$src, $dst|$dst, $src}",
                       SSE_CVT_Scalar>, XS;
 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
                       "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
                       SSE_CVT_Scalar>, XS, REX_W;
 defm CVTSI2SD  : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
-                      "cvtsi2sd\t{$src, $dst|$dst, $src}",
+                      "cvtsi2sd{l}\t{$src, $dst|$dst, $src}",
                       SSE_CVT_Scalar>, XD;
 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
                       "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
                       SSE_CVT_Scalar>, XD, REX_W;
 
+def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
+                (CVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
+def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
+                (CVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
+def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
+                (CVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
+def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
+                (CVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
+def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
+                (CVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
+def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
+                (CVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
+def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+                (CVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
+def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+                (CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
+
+def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}",
+                (CVTSI2SSrm FR64:$dst, i32mem:$src)>;
+def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}",
+                (CVTSI2SDrm FR64:$dst, i32mem:$src)>;
+
 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
 // and/or XMM operand(s).
 
@@ -1540,10 +1579,12 @@ multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
                          string asm, OpndItins itins> {
   def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
               !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
-              [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
+              [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>,
+           Sched<[itins.Sched]>;
   def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
               !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
-              [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>;
+              [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>,
+           Sched<[itins.Sched.Folded]>;
 }
 
 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
@@ -1555,38 +1596,38 @@ multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
                   !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
                   !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
               [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
-              itins.rr>;
+              itins.rr>, Sched<[itins.Sched]>;
   def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
               (ins DstRC:$src1, x86memop:$src2),
               !if(Is2Addr,
                   !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
                   !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
               [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
-              itins.rm>;
+              itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
-                  int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si{l}",
+                  int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si",
                   SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
-                    int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si{q}",
+                    int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si",
                     SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
 
 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
-                 sdmem, sse_load_f64, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
+                 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD;
 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
-                   sdmem, sse_load_f64, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
+                   sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, REX_W;
 
 
 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
-          int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
+          int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}",
           SSE_CVT_Scalar, 0>, XS, VEX_4V;
 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
           int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
           SSE_CVT_Scalar, 0>, XS, VEX_4V,
           VEX_W;
 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
-          int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
+          int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}",
           SSE_CVT_Scalar, 0>, XD, VEX_4V;
 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
           int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
@@ -1596,13 +1637,13 @@ defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
 let Constraints = "$src1 = $dst" in {
   defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
                         int_x86_sse_cvtsi2ss, i32mem, loadi32,
-                        "cvtsi2ss", SSE_CVT_Scalar>, XS;
+                        "cvtsi2ss{l}", SSE_CVT_Scalar>, XS;
   defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
                         int_x86_sse_cvtsi642ss, i64mem, loadi64,
                         "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
   defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
                         int_x86_sse2_cvtsi2sd, i32mem, loadi32,
-                        "cvtsi2sd", SSE_CVT_Scalar>, XD;
+                        "cvtsi2sd{l}", SSE_CVT_Scalar>, XD;
   defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
                         int_x86_sse2_cvtsi642sd, i64mem, loadi64,
                         "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
@@ -1616,40 +1657,40 @@ defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
                                     SSE_CVT_SS2SI_32>, XS, VEX;
 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
                                    int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
-                                   "cvttss2si{q}", SSE_CVT_SS2SI_64>,
+                                   "cvttss2si", SSE_CVT_SS2SI_64>,
                                    XS, VEX, VEX_W;
 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
                                     sdmem, sse_load_f64, "cvttsd2si",
                                     SSE_CVT_SD2SI>, XD, VEX;
 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
                                   int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
-                                  "cvttsd2si{q}", SSE_CVT_SD2SI>,
+                                  "cvttsd2si", SSE_CVT_SD2SI>,
                                   XD, VEX, VEX_W;
 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
                                     ssmem, sse_load_f32, "cvttss2si",
                                     SSE_CVT_SS2SI_32>, XS;
 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
                                    int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
-                                   "cvttss2si{q}", SSE_CVT_SS2SI_64>, XS, REX_W;
+                                   "cvttss2si", SSE_CVT_SS2SI_64>, XS, REX_W;
 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
                                     sdmem, sse_load_f64, "cvttsd2si",
                                     SSE_CVT_SD2SI>, XD;
 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
                                   int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
-                                  "cvttsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
+                                  "cvttsd2si", SSE_CVT_SD2SI>, XD, REX_W;
 
 defm VCVTSS2SI   : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
-                                  ssmem, sse_load_f32, "cvtss2si{l}",
+                                  ssmem, sse_load_f32, "cvtss2si",
                                   SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
-                                  ssmem, sse_load_f32, "cvtss2si{q}",
+                                  ssmem, sse_load_f32, "cvtss2si",
                                   SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
 
 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
-                               ssmem, sse_load_f32, "cvtss2si{l}",
+                               ssmem, sse_load_f32, "cvtss2si",
                                SSE_CVT_SS2SI_32>, XS;
 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
-                                 ssmem, sse_load_f32, "cvtss2si{q}",
+                                 ssmem, sse_load_f32, "cvtss2si",
                                  SSE_CVT_SS2SI_64>, XS, REX_W;
 
 defm VCVTDQ2PS   : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
@@ -1666,6 +1707,40 @@ defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
                             SSEPackedSingle, SSE_CVT_PS>,
                             TB, Requires<[UseSSE2]>;
 
+def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
+                (VCVTSS2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
+                (VCVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
+def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
+                (VCVTSD2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
+                (VCVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
+def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
+                (VCVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
+                (VCVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
+def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
+                (VCVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
+                (VCVTSD2SI64rm GR64:$dst, sdmem:$src), 0>;
+
+def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
+                (CVTSS2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
+                (CVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
+def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
+                (CVTSD2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
+                (CVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
+def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
+                (CVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
+                (CVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
+def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
+                (CVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
+                (CVTSD2SI64rm GR64:$dst, sdmem:$src)>;
+
 /// SSE 2 Only
 
 // Convert scalar double to scalar single
@@ -1673,13 +1748,15 @@ let neverHasSideEffects = 1 in {
 def VCVTSD2SSrr  : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
                        (ins FR64:$src1, FR64:$src2),
                       "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
-                      IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
+                      IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG,
+                      Sched<[WriteCvtF2F]>;
 let mayLoad = 1 in
 def VCVTSD2SSrm  : I<0x5A, MRMSrcMem, (outs FR32:$dst),
                        (ins FR64:$src1, f64mem:$src2),
                       "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                       [], IIC_SSE_CVT_Scalar_RM>,
-                      XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
+                      XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG,
+                      Sched<[WriteCvtF2FLd, ReadAfterLd]>;
 }
 
 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
@@ -1688,26 +1765,28 @@ def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
 def CVTSD2SSrr  : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
                       "cvtsd2ss\t{$src, $dst|$dst, $src}",
                       [(set FR32:$dst, (fround FR64:$src))],
-                      IIC_SSE_CVT_Scalar_RR>;
+                      IIC_SSE_CVT_Scalar_RR>, Sched<[WriteCvtF2F]>;
 def CVTSD2SSrm  : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
                       "cvtsd2ss\t{$src, $dst|$dst, $src}",
                       [(set FR32:$dst, (fround (loadf64 addr:$src)))],
                       IIC_SSE_CVT_Scalar_RM>,
                       XD,
-                  Requires<[UseSSE2, OptForSize]>;
+                  Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
 
 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
                        (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
                        "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                        [(set VR128:$dst,
                          (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
-                       IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>;
+                       IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>,
+                       Sched<[WriteCvtF2F]>;
 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
                        (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
                        "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                        [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
                                           VR128:$src1, sse_load_f64:$src2))],
-                       IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>;
+                       IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>,
+                       Sched<[WriteCvtF2FLd, ReadAfterLd]>;
 
 let Constraints = "$src1 = $dst" in {
 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
@@ -1715,13 +1794,15 @@ def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
                        "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                        [(set VR128:$dst,
                          (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
-                       IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>;
+                       IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>,
+                       Sched<[WriteCvtF2F]>;
 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
                        (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
                        "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                        [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
                                           VR128:$src1, sse_load_f64:$src2))],
-                       IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>;
+                       IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>,
+                       Sched<[WriteCvtF2FLd, ReadAfterLd]>;
 }
 
 // Convert scalar single to scalar double
@@ -1731,13 +1812,15 @@ def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
                     (ins FR32:$src1, FR32:$src2),
                     "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                     [], IIC_SSE_CVT_Scalar_RR>,
-                    XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
+                    XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG,
+                    Sched<[WriteCvtF2F]>;
 let mayLoad = 1 in
 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
                     (ins FR32:$src1, f32mem:$src2),
                     "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                     [], IIC_SSE_CVT_Scalar_RM>,
-                    XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
+                    XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>,
+                    Sched<[WriteCvtF2FLd, ReadAfterLd]>;
 }
 
 def : Pat<(f64 (fextend FR32:$src)),
@@ -1756,12 +1839,12 @@ def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
                    "cvtss2sd\t{$src, $dst|$dst, $src}",
                    [(set FR64:$dst, (fextend FR32:$src))],
                    IIC_SSE_CVT_Scalar_RR>, XS,
-                 Requires<[UseSSE2]>;
+                 Requires<[UseSSE2]>, Sched<[WriteCvtF2F]>;
 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
                    "cvtss2sd\t{$src, $dst|$dst, $src}",
                    [(set FR64:$dst, (extloadf32 addr:$src))],
                    IIC_SSE_CVT_Scalar_RM>, XS,
-                 Requires<[UseSSE2, OptForSize]>;
+                 Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
 
 // extload f32 -> f64.  This matches load+fextend because we have a hack in
 // the isel (PreprocessForFPConvert) that can introduce loads after dag
@@ -1778,57 +1861,61 @@ def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
                     "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                     [(set VR128:$dst,
                       (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
-                    IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>;
+                    IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>,
+                    Sched<[WriteCvtF2F]>;
 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
                       (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
                     "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                     [(set VR128:$dst,
                       (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
-                    IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>;
+                    IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>,
+                    Sched<[WriteCvtF2FLd, ReadAfterLd]>;
 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
                       (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
                     "cvtss2sd\t{$src2, $dst|$dst, $src2}",
                     [(set VR128:$dst,
                       (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
-                    IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>;
+                    IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>,
+                    Sched<[WriteCvtF2F]>;
 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
                       (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
                     "cvtss2sd\t{$src2, $dst|$dst, $src2}",
                     [(set VR128:$dst,
                       (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
-                    IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>;
+                    IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>,
+                    Sched<[WriteCvtF2FLd, ReadAfterLd]>;
 }
 
 // Convert packed single/double fp to doubleword
 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                        "cvtps2dq\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
-                       IIC_SSE_CVT_PS_RR>, VEX;
+                       IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                        "cvtps2dq\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst,
                          (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
-                       IIC_SSE_CVT_PS_RM>, VEX;
+                       IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
                         "cvtps2dq\t{$src, $dst|$dst, $src}",
                         [(set VR256:$dst,
                           (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
-                        IIC_SSE_CVT_PS_RR>, VEX, VEX_L;
+                        IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
                         "cvtps2dq\t{$src, $dst|$dst, $src}",
                         [(set VR256:$dst,
                           (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
-                        IIC_SSE_CVT_PS_RM>, VEX, VEX_L;
+                        IIC_SSE_CVT_PS_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                      "cvtps2dq\t{$src, $dst|$dst, $src}",
                      [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
-                     IIC_SSE_CVT_PS_RR>;
+                     IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                      "cvtps2dq\t{$src, $dst|$dst, $src}",
                      [(set VR128:$dst,
                        (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
-                     IIC_SSE_CVT_PS_RM>;
+                     IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
 
 
 // Convert Packed Double FP to Packed DW Integers
@@ -1839,7 +1926,7 @@ let Predicates = [HasAVX] in {
 def VCVTPD2DQrr  : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                        "vcvtpd2dq\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
-                       VEX;
+                       VEX, Sched<[WriteCvtF2I]>;
 
 // XMM only
 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
@@ -1847,18 +1934,20 @@ def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                        "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst,
-                         (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX;
+                         (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX,
+                       Sched<[WriteCvtF2ILd]>;
 
 // YMM only
 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
                        "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst,
-                         (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L;
+                         (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L,
+                       Sched<[WriteCvtF2I]>;
 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
                        "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst,
                          (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
-                       VEX, VEX_L;
+                       VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
                 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
 }
@@ -1867,11 +1956,11 @@ def CVTPD2DQrm  : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                       "cvtpd2dq\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst,
                         (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
-                      IIC_SSE_CVT_PD_RM>;
+                      IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2ILd]>;
 def CVTPD2DQrr  : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                       "cvtpd2dq\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
-                      IIC_SSE_CVT_PD_RR>;
+                      IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
 
 // Convert with truncation packed single/double fp to doubleword
 // SSE2 packed instructions with XS prefix
@@ -1879,32 +1968,33 @@ def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                          "cvttps2dq\t{$src, $dst|$dst, $src}",
                          [(set VR128:$dst,
                            (int_x86_sse2_cvttps2dq VR128:$src))],
-                         IIC_SSE_CVT_PS_RR>, VEX;
+                         IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                          "cvttps2dq\t{$src, $dst|$dst, $src}",
                          [(set VR128:$dst, (int_x86_sse2_cvttps2dq
                                             (memopv4f32 addr:$src)))],
-                         IIC_SSE_CVT_PS_RM>, VEX;
+                         IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
                           "cvttps2dq\t{$src, $dst|$dst, $src}",
                           [(set VR256:$dst,
                             (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
-                          IIC_SSE_CVT_PS_RR>, VEX, VEX_L;
+                          IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
                           "cvttps2dq\t{$src, $dst|$dst, $src}",
                           [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
                                              (memopv8f32 addr:$src)))],
-                          IIC_SSE_CVT_PS_RM>, VEX, VEX_L;
+                          IIC_SSE_CVT_PS_RM>, VEX, VEX_L,
+                          Sched<[WriteCvtF2ILd]>;
 
 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                        "cvttps2dq\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
-                       IIC_SSE_CVT_PS_RR>;
+                       IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                        "cvttps2dq\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst,
                          (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
-                       IIC_SSE_CVT_PS_RM>;
+                       IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
 
 let Predicates = [HasAVX] in {
   def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
@@ -1954,7 +2044,7 @@ def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                         "cvttpd2dq\t{$src, $dst|$dst, $src}",
                         [(set VR128:$dst,
                               (int_x86_sse2_cvttpd2dq VR128:$src))],
-                              IIC_SSE_CVT_PD_RR>, VEX;
+                              IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2I]>;
 
 // The assembler can recognize rr 256-bit instructions by seeing a ymm
 // register, but the same isn't true when using memory operands instead.
@@ -1967,19 +2057,19 @@ def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                          "cvttpd2dqx\t{$src, $dst|$dst, $src}",
                          [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
                                             (memopv2f64 addr:$src)))],
-                         IIC_SSE_CVT_PD_RM>, VEX;
+                         IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2ILd]>;
 
 // YMM only
 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
                          "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
                          [(set VR128:$dst,
                            (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
-                         IIC_SSE_CVT_PD_RR>, VEX, VEX_L;
+                         IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
                          "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
                          [(set VR128:$dst,
                           (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
-                         IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
+                         IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
                 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
 
@@ -1993,12 +2083,13 @@ let Predicates = [HasAVX] in {
 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                       "cvttpd2dq\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
-                      IIC_SSE_CVT_PD_RR>;
+                      IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
                       "cvttpd2dq\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
                                         (memopv2f64 addr:$src)))],
-                                        IIC_SSE_CVT_PD_RM>;
+                                        IIC_SSE_CVT_PD_RM>,
+                      Sched<[WriteCvtF2ILd]>;
 
 // Convert packed single to packed double
 let Predicates = [HasAVX] in {
@@ -2006,32 +2097,32 @@ let Predicates = [HasAVX] in {
 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                      "vcvtps2pd\t{$src, $dst|$dst, $src}",
                      [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
-                     IIC_SSE_CVT_PD_RR>, TB, VEX;
+                     IIC_SSE_CVT_PD_RR>, TB, VEX, Sched<[WriteCvtF2F]>;
 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
                     "vcvtps2pd\t{$src, $dst|$dst, $src}",
                     [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
-                    IIC_SSE_CVT_PD_RM>, TB, VEX;
+                    IIC_SSE_CVT_PD_RM>, TB, VEX, Sched<[WriteCvtF2FLd]>;
 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
                      "vcvtps2pd\t{$src, $dst|$dst, $src}",
                      [(set VR256:$dst,
                        (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
-                     IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L;
+                     IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L, Sched<[WriteCvtF2F]>;
 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
                      "vcvtps2pd\t{$src, $dst|$dst, $src}",
                      [(set VR256:$dst,
                        (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
-                     IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L;
+                     IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
 }
 
 let Predicates = [UseSSE2] in {
 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                        "cvtps2pd\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
-                       IIC_SSE_CVT_PD_RR>, TB;
+                       IIC_SSE_CVT_PD_RR>, TB, Sched<[WriteCvtF2F]>;
 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
                    "cvtps2pd\t{$src, $dst|$dst, $src}",
                    [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
-                   IIC_SSE_CVT_PD_RM>, TB;
+                   IIC_SSE_CVT_PD_RM>, TB, Sched<[WriteCvtF2FLd]>;
 }
 
 // Convert Packed DW Integers to Packed Double FP
@@ -2039,30 +2130,33 @@ let Predicates = [HasAVX] in {
 let neverHasSideEffects = 1, mayLoad = 1 in
 def VCVTDQ2PDrm  : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
                      "vcvtdq2pd\t{$src, $dst|$dst, $src}",
-                     []>, VEX;
+                     []>, VEX, Sched<[WriteCvtI2FLd]>;
 def VCVTDQ2PDrr  : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                      "vcvtdq2pd\t{$src, $dst|$dst, $src}",
                      [(set VR128:$dst,
-                       (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX;
+                       (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX,
+                   Sched<[WriteCvtI2F]>;
 def VCVTDQ2PDYrm  : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
                      "vcvtdq2pd\t{$src, $dst|$dst, $src}",
                      [(set VR256:$dst,
                        (int_x86_avx_cvtdq2_pd_256
-                        (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L;
+                        (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L,
+                    Sched<[WriteCvtI2FLd]>;
 def VCVTDQ2PDYrr  : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
                      "vcvtdq2pd\t{$src, $dst|$dst, $src}",
                      [(set VR256:$dst,
-                       (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L;
+                       (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L,
+                    Sched<[WriteCvtI2F]>;
 }
 
 let neverHasSideEffects = 1, mayLoad = 1 in
 def CVTDQ2PDrm  : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
                        "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
-                       IIC_SSE_CVT_PD_RR>;
+                       IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtI2FLd]>;
 def CVTDQ2PDrr  : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                        "cvtdq2pd\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
-                       IIC_SSE_CVT_PD_RM>;
+                       IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtI2F]>;
 
 // AVX 256-bit register conversion intrinsics
 let Predicates = [HasAVX] in {
@@ -2079,7 +2173,7 @@ let Predicates = [HasAVX] in {
 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                        "cvtpd2ps\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
-                       IIC_SSE_CVT_PD_RR>, VEX;
+                       IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2F]>;
 
 // XMM only
 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
@@ -2088,31 +2182,31 @@ def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                         "cvtpd2psx\t{$src, $dst|$dst, $src}",
                         [(set VR128:$dst,
                           (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
-                        IIC_SSE_CVT_PD_RM>, VEX;
+                        IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2FLd]>;
 
 // YMM only
 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
                         "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
                         [(set VR128:$dst,
                           (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
-                        IIC_SSE_CVT_PD_RR>, VEX, VEX_L;
+                        IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2F]>;
 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
                         "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
                         [(set VR128:$dst,
                           (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
-                        IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
+                        IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
                 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
 
 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                      "cvtpd2ps\t{$src, $dst|$dst, $src}",
                      [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
-                     IIC_SSE_CVT_PD_RR>;
+                     IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2F]>;
 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                      "cvtpd2ps\t{$src, $dst|$dst, $src}",
                      [(set VR128:$dst,
                        (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
-                     IIC_SSE_CVT_PD_RM>;
+                     IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2FLd]>;
 
 
 // AVX 256-bit register conversion intrinsics
@@ -2165,22 +2259,24 @@ multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
   def rr : SIi8<0xC2, MRMSrcReg,
                 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
                 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
-                itins.rr>;
+                itins.rr>, Sched<[itins.Sched]>;
   def rm : SIi8<0xC2, MRMSrcMem,
                 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
                 [(set RC:$dst, (OpNode (VT RC:$src1),
                                          (ld_frag addr:$src2), imm:$cc))],
-                                         itins.rm>;
+                                         itins.rm>,
+           Sched<[itins.Sched.Folded, ReadAfterLd]>;
 
   // Accept explicit immediate argument form instead of comparison code.
   let neverHasSideEffects = 1 in {
     def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
                       (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
-                      IIC_SSE_ALU_F32S_RR>;
+                      IIC_SSE_ALU_F32S_RR>, Sched<[itins.Sched]>;
     let mayLoad = 1 in
     def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
                       (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
-                      IIC_SSE_ALU_F32S_RM>;
+                      IIC_SSE_ALU_F32S_RM>,
+                      Sched<[itins.Sched.Folded, ReadAfterLd]>;
   }
 }
 
@@ -2213,12 +2309,14 @@ multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
                       (ins VR128:$src1, VR128:$src, CC:$cc), asm,
                         [(set VR128:$dst, (Int VR128:$src1,
                                                VR128:$src, imm:$cc))],
-                                               itins.rr>;
+                                               itins.rr>,
+           Sched<[itins.Sched]>;
   def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
                       (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
                         [(set VR128:$dst, (Int VR128:$src1,
                                                (load addr:$src), imm:$cc))],
-                                               itins.rm>;
+                                               itins.rm>,
+           Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 // Aliases to match intrinsics which expect XMM operand(s).
@@ -2248,12 +2346,14 @@ multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
   def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
                      !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
                      [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
-                     IIC_SSE_COMIS_RR, d>;
+                     IIC_SSE_COMIS_RR, d>,
+          Sched<[WriteFAdd]>;
   def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
                      !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
                      [(set EFLAGS, (OpNode (vt RC:$src1),
                                            (ld_frag addr:$src2)))],
-                                           IIC_SSE_COMIS_RM, d>;
+                                           IIC_SSE_COMIS_RM, d>,
+          Sched<[WriteFAddLd, ReadAfterLd]>;
 }
 
 let Defs = [EFLAGS] in {
@@ -2310,20 +2410,23 @@ multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
   def rri : PIi8<0xC2, MRMSrcReg,
              (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
              [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
-             IIC_SSE_CMPP_RR, d>;
+             IIC_SSE_CMPP_RR, d>,
+            Sched<[WriteFAdd]>;
   def rmi : PIi8<0xC2, MRMSrcMem,
              (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
              [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
-             IIC_SSE_CMPP_RM, d>;
+             IIC_SSE_CMPP_RM, d>,
+            Sched<[WriteFAddLd, ReadAfterLd]>;
 
   // Accept explicit immediate argument form instead of comparison code.
   let neverHasSideEffects = 1 in {
     def rri_alt : PIi8<0xC2, MRMSrcReg,
                (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
-               asm_alt, [], IIC_SSE_CMPP_RR, d>;
+               asm_alt, [], IIC_SSE_CMPP_RR, d>, Sched<[WriteFAdd]>;
     def rmi_alt : PIi8<0xC2, MRMSrcMem,
                (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
-               asm_alt, [], IIC_SSE_CMPP_RM, d>;
+               asm_alt, [], IIC_SSE_CMPP_RM, d>,
+               Sched<[WriteFAddLd, ReadAfterLd]>;
   }
 }
 
@@ -2399,12 +2502,14 @@ multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
   def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
                    (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
                    [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
-                                       (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
+                                       (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
+            Sched<[WriteShuffleLd, ReadAfterLd]>;
   let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
     def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
                    (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
                    [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
-                                       (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
+                                       (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
+              Sched<[WriteShuffle]>;
 }
 
 defm VSHUFPS  : sse12_shuffle<VR128, f128mem, v4f32,
@@ -2488,13 +2593,14 @@ multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
                 (outs RC:$dst), (ins RC:$src1, RC:$src2),
                 asm, [(set RC:$dst,
                            (vt (OpNode RC:$src1, RC:$src2)))],
-                           IIC_SSE_UNPCK, d>;
+                           IIC_SSE_UNPCK, d>, Sched<[WriteShuffle]>;
     def rm : PI<opc, MRMSrcMem,
                 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
                 asm, [(set RC:$dst,
                            (vt (OpNode RC:$src1,
                                        (mem_frag addr:$src2))))],
-                                       IIC_SSE_UNPCK, d>;
+                                       IIC_SSE_UNPCK, d>,
+             Sched<[WriteShuffleLd, ReadAfterLd]>;
 }
 
 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
@@ -2585,10 +2691,11 @@ multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
                                 Domain d> {
   def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
                 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
-                     [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
+                     [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>,
+             Sched<[WriteVecLogic]>;
   def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
                 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
-                IIC_SSE_MOVMSK, d>, REX_W;
+                IIC_SSE_MOVMSK, d>, REX_W, Sched<[WriteVecLogic]>;
 }
 
 let Predicates = [HasAVX] in {
@@ -2616,18 +2723,18 @@ let Predicates = [HasAVX] in {
   // Assembler Only
   def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
              "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
-             SSEPackedSingle>, TB, VEX;
+             SSEPackedSingle>, TB, VEX, Sched<[WriteVecLogic]>;
   def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
              "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
              SSEPackedDouble>, TB,
-             OpSize, VEX;
+             OpSize, VEX, Sched<[WriteVecLogic]>;
   def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
              "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
-             SSEPackedSingle>, TB, VEX, VEX_L;
+             SSEPackedSingle>, TB, VEX, VEX_L, Sched<[WriteVecLogic]>;
   def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
              "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
              SSEPackedDouble>, TB,
-             OpSize, VEX, VEX_L;
+             OpSize, VEX, VEX_L, Sched<[WriteVecLogic]>;
 }
 
 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
@@ -2657,17 +2764,16 @@ let ExeDomain = SSEPackedInt in { // SSE integer instructions
 /// PDI_binop_rm - Simple SSE2 binary operator.
 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
                         ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
-                        X86MemOperand x86memop,
-                        OpndItins itins,
-                        bit IsCommutable = 0,
-                        bit Is2Addr = 1> {
+                        X86MemOperand x86memop, OpndItins itins,
+                        bit IsCommutable, bit Is2Addr> {
   let isCommutable = IsCommutable in
   def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
        (ins RC:$src1, RC:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
+       [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
+       Sched<[itins.Sched]>;
   def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
        (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
@@ -2675,44 +2781,35 @@ multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst, (OpVT (OpNode RC:$src1,
                                      (bitconvert (memop_frag addr:$src2)))))],
-                                     itins.rm>;
+                                     itins.rm>,
+       Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 } // ExeDomain = SSEPackedInt
 
-// These are ordered here for pattern ordering requirements with the fp versions
+multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
+                         ValueType OpVT128, ValueType OpVT256,
+                         OpndItins itins, bit IsCommutable = 0> {
+let Predicates = [HasAVX] in
+  defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
+                    VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
 
-let Predicates = [HasAVX] in {
-defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
-                          i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
-defm VPOR  : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
-                          i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
-defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
-                          i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
-defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
-                          i128mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
+let Constraints = "$src1 = $dst" in
+  defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
+                           memopv2i64, i128mem, itins, IsCommutable, 1>;
+
+let Predicates = [HasAVX2] in
+  defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
+                               OpVT256, VR256, memopv4i64, i256mem, itins,
+                               IsCommutable, 0>, VEX_4V, VEX_L;
 }
 
-let Constraints = "$src1 = $dst" in {
-defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
-                         i128mem, SSE_BIT_ITINS_P, 1>;
-defm POR  : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
-                         i128mem, SSE_BIT_ITINS_P, 1>;
-defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
-                         i128mem, SSE_BIT_ITINS_P, 1>;
-defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
-                          i128mem, SSE_BIT_ITINS_P, 0>;
-} // Constraints = "$src1 = $dst"
+// These are ordered here for pattern ordering requirements with the fp versions
 
-let Predicates = [HasAVX2] in {
-defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
-                           i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPORY  : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
-                           i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
-                           i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
-                            i256mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-}
+defm PAND  : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
+defm POR   : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
+defm PXOR  : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
+defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
+                           SSE_BIT_ITINS_P, 0>;
 
 //===----------------------------------------------------------------------===//
 // SSE 1 & 2 - Logical Instructions
@@ -2757,6 +2854,20 @@ let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
 ///
 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
                                    SDNode OpNode> {
+  defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
+        !strconcat(OpcodeStr, "ps"), f256mem,
+        [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
+        [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
+                           (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
+
+  defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
+        !strconcat(OpcodeStr, "pd"), f256mem,
+        [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
+                                  (bc_v4i64 (v4f64 VR256:$src2))))],
+        [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
+                                  (memopv4i64 addr:$src2)))], 0>,
+                                  TB, OpSize, VEX_4V, VEX_L;
+
   // In AVX no need to add a pattern for 128-bit logical rr ps, because they
   // are all promoted to v2i64, and the patterns are covered by the int
   // version. This is needed in SSE only, because v2i64 isn't supported on
@@ -2764,7 +2875,7 @@ multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
   defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
        !strconcat(OpcodeStr, "ps"), f128mem, [],
        [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
-                                 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
+                                 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
 
   defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
        !strconcat(OpcodeStr, "pd"), f128mem,
@@ -2773,6 +2884,7 @@ multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
        [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
                                  (memopv2i64 addr:$src2)))], 0>,
                                                  TB, OpSize, VEX_4V;
+
   let Constraints = "$src1 = $dst" in {
     defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
          !strconcat(OpcodeStr, "ps"), f128mem,
@@ -2789,31 +2901,6 @@ multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
   }
 }
 
-/// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
-///
-multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
-                                     SDNode OpNode> {
-    defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
-          !strconcat(OpcodeStr, "ps"), f256mem,
-          [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
-          [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
-                             (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
-
-    defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
-          !strconcat(OpcodeStr, "pd"), f256mem,
-          [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
-                                    (bc_v4i64 (v4f64 VR256:$src2))))],
-          [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
-                                    (memopv4i64 addr:$src2)))], 0>,
-                                    TB, OpSize, VEX_4V, VEX_L;
-}
-
-// AVX 256-bit packed logical ops forms
-defm VAND  : sse12_fp_packed_logical_y<0x54, "and", and>;
-defm VOR   : sse12_fp_packed_logical_y<0x56, "or", or>;
-defm VXOR  : sse12_fp_packed_logical_y<0x57, "xor", xor>;
-defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
-
 defm AND  : sse12_fp_packed_logical<0x54, "and", and>;
 defm OR   : sse12_fp_packed_logical<0x56, "or", or>;
 defm XOR  : sse12_fp_packed_logical<0x57, "xor", xor>;
@@ -2848,26 +2935,32 @@ multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
                             itins.d, Is2Addr>, XD;
 }
 
-multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
-                                   SizeItins itins,
-                                   bit Is2Addr = 1> {
+multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
+                                  SDNode OpNode, SizeItins itins> {
+let Predicates = [HasAVX] in {
+  defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
+                               VR128, v4f32, f128mem, memopv4f32,
+                               SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
+  defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
+                               VR128, v2f64, f128mem, memopv2f64,
+                               SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V;
+
+  defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
+                        OpNode, VR256, v8f32, f256mem, memopv8f32,
+                        SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
+  defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
+                        OpNode, VR256, v4f64, f256mem, memopv4f64,
+                        SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L;
+}
+
+let Constraints = "$src1 = $dst" in {
   defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
-              v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>,
-              TB;
+                            v4f32, f128mem, memopv4f32, SSEPackedSingle,
+                            itins.s, 1>, TB;
   defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
-              v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>,
-              TB, OpSize;
+                            v2f64, f128mem, memopv2f64, SSEPackedDouble,
+                            itins.d, 1>, TB, OpSize;
 }
-
-multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
-                                    SDNode OpNode,
-                                    SizeItins itins> {
-  defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
-                v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>,
-                TB, VEX_L;
-  defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
-                v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>,
-                TB, OpSize, VEX_L;
 }
 
 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
@@ -2881,116 +2974,69 @@ multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
      itins.d, Is2Addr>, XD;
 }
 
-multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
-                                      SizeItins itins,
-                                      bit Is2Addr = 1> {
-  defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
-     !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
-                              SSEPackedSingle, itins.s, Is2Addr>,
-                              TB;
-
-  defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
-     !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
-                              SSEPackedDouble, itins.d, Is2Addr>,
-                              TB, OpSize;
+// Binary Arithmetic instructions
+defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>;
+defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>;
+let isCommutable = 0 in {
+  defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>;
+  defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>;
+  defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>;
+  defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>;
 }
 
-multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr,
-                                        SizeItins itins> {
-  defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
-     !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
-      SSEPackedSingle, itins.s, 0>, TB, VEX_L;
-
-  defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
-     !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
-      SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_L;
+let isCodeGenOnly = 1 in {
+  defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>;
+  defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>;
 }
 
-// Binary Arithmetic instructions
 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
             basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
               VEX_4V, VEX_LIG;
-defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>,
-            basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>,
-              VEX_4V;
 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
             basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
               VEX_4V, VEX_LIG;
-defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>,
-            basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
-              VEX_4V;
 
 let isCommutable = 0 in {
   defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
               basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
                 VEX_4V, VEX_LIG;
-  defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>,
-              basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
-                VEX_4V;
   defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
               basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
                 VEX_4V, VEX_LIG;
-  defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>,
-              basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
-                VEX_4V;
   defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
               basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
                 VEX_4V, VEX_LIG;
-  defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>,
-              basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>,
-              basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
-              basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>,
-                VEX_4V;
   defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
               basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
                 VEX_4V, VEX_LIG;
-  defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>,
-              basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>,
-              basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>,
-              basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
-                VEX_4V;
 }
 
 let Constraints = "$src1 = $dst" in {
   defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
-             basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
              basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
   defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
-             basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
              basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
 
   let isCommutable = 0 in {
     defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
-               basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
                basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
     defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
-               basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
                basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
     defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
-               basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
-               basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>,
-               basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>;
+               basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
     defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
-               basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
-               basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>,
-               basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>;
+               basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
   }
 }
 
 let isCodeGenOnly = 1 in {
   defm VMAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S, 0>,
        VEX_4V, VEX_LIG;
-  defm VMAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P, 0>,
-       basic_sse12_fp_binop_p_y<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>, VEX_4V;
   defm VMINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S, 0>,
        VEX_4V, VEX_LIG;
-  defm VMINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P, 0>,
-       basic_sse12_fp_binop_p_y<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>, VEX_4V;
   let Constraints = "$src1 = $dst" in {
-    defm MAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>,
-         basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>;
-    defm MINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>,
-         basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>;
+    defm MAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
+    defm MINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
   }
 }
 
@@ -3002,6 +3048,7 @@ let isCodeGenOnly = 1 in {
 ///
 /// And, we have a special variant form for a full-vector intrinsic form.
 
+let Sched = WriteFSqrt in {
 def SSE_SQRTP : OpndItins<
   IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
 >;
@@ -3009,7 +3056,9 @@ def SSE_SQRTP : OpndItins<
 def SSE_SQRTS : OpndItins<
   IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
 >;
+}
 
+let Sched = WriteFRcp in {
 def SSE_RCPP : OpndItins<
   IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
 >;
@@ -3017,13 +3066,36 @@ def SSE_RCPP : OpndItins<
 def SSE_RCPS : OpndItins<
   IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
 >;
+}
 
 /// sse1_fp_unop_s - SSE1 unops in scalar form.
 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
                           SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
+let Predicates = [HasAVX], hasSideEffects = 0 in {
+  def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
+                      (ins FR32:$src1, FR32:$src2),
+                      !strconcat("v", OpcodeStr,
+                                 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                      []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
+  let mayLoad = 1 in {
+  def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
+                      (ins FR32:$src1,f32mem:$src2),
+                      !strconcat("v", OpcodeStr,
+                                 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                      []>, VEX_4V, VEX_LIG,
+                   Sched<[itins.Sched.Folded, ReadAfterLd]>;
+  def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+                      (ins VR128:$src1, ssmem:$src2),
+                      !strconcat("v", OpcodeStr,
+                                 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                      []>, VEX_4V, VEX_LIG,
+                      Sched<[itins.Sched.Folded, ReadAfterLd]>;
+  }
+}
+
   def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
                 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
-                [(set FR32:$dst, (OpNode FR32:$src))]>;
+                [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
   // For scalar unary operations, fold a load into the operation
   // only in OptForSize mode. It eliminates an instruction, but it also
   // eliminates a whole-register clobber (the load), so it introduces a
@@ -3031,204 +3103,238 @@ multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
   def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
                 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
                 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
-            Requires<[UseSSE1, OptForSize]>;
+            Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
   def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                     !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
-                    [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
+                    [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>,
+                Sched<[itins.Sched]>;
   def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
                     !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
-                    [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
+                    [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>,
+                Sched<[itins.Sched.Folded]>;
 }
 
-/// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
-multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
-  def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
-                !strconcat(OpcodeStr,
-                           "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
+/// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
+multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
+                           OpndItins itins> {
+let Predicates = [HasAVX], hasSideEffects = 0 in {
+  def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
+                       (ins FR32:$src1, FR32:$src2),
+                       !strconcat("v", OpcodeStr,
+                           "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
   let mayLoad = 1 in {
-  def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
-                !strconcat(OpcodeStr,
-                           "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
-  def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
-                (ins VR128:$src1, ssmem:$src2),
-                !strconcat(OpcodeStr,
-                           "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
+  def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
+                      (ins FR32:$src1,f32mem:$src2),
+                      !strconcat("v", OpcodeStr,
+                                 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                      []>, VEX_4V, VEX_LIG,
+                   Sched<[itins.Sched.Folded, ReadAfterLd]>;
+  def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+                      (ins VR128:$src1, ssmem:$src2),
+                      !strconcat("v", OpcodeStr,
+                                 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                      []>, VEX_4V, VEX_LIG,
+                      Sched<[itins.Sched.Folded, ReadAfterLd]>;
+  }
+}
+
+  def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
+                !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
+                [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
+  // For scalar unary operations, fold a load into the operation
+  // only in OptForSize mode. It eliminates an instruction, but it also
+  // eliminates a whole-register clobber (the load), so it introduces a
+  // partial register update condition.
+  def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
+                !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
+                [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
+            Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
+  let Constraints = "$src1 = $dst" in {
+    def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
+                      (ins VR128:$src1, VR128:$src2),
+                      !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+                      [], itins.rr>, Sched<[itins.Sched]>;
+    let mayLoad = 1, hasSideEffects = 0 in
+    def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+                      (ins VR128:$src1, ssmem:$src2),
+                      !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+                      [], itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
   }
 }
 
 /// sse1_fp_unop_p - SSE1 unops in packed form.
 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
                           OpndItins itins> {
-  def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
-              !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
-              [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
-  def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
-                !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
-                [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
+let Predicates = [HasAVX] in {
+  def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+                       !strconcat("v", OpcodeStr,
+                                  "ps\t{$src, $dst|$dst, $src}"),
+                       [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
+                       itins.rr>, VEX, Sched<[itins.Sched]>;
+  def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+                       !strconcat("v", OpcodeStr,
+                                  "ps\t{$src, $dst|$dst, $src}"),
+                       [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))],
+                       itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
+  def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+                        !strconcat("v", OpcodeStr,
+                                   "ps\t{$src, $dst|$dst, $src}"),
+                        [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
+                        itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
+  def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
+                        !strconcat("v", OpcodeStr,
+                                   "ps\t{$src, $dst|$dst, $src}"),
+                        [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
+                        itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
 }
 
-/// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
-multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
-                            OpndItins itins> {
-  def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
-              !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
-              [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
-              itins.rr>, VEX_L;
-  def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
+  def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+                !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
+                [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>,
+            Sched<[itins.Sched]>;
+  def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
-                [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
-                itins.rm>, VEX_L;
+                [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>,
+            Sched<[itins.Sched.Folded]>;
 }
 
 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
-                              Intrinsic V4F32Int, OpndItins itins> {
+                              Intrinsic V4F32Int, Intrinsic V8F32Int,
+                              OpndItins itins> {
+let Predicates = [HasAVX] in {
+  def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+                           !strconcat("v", OpcodeStr,
+                                      "ps\t{$src, $dst|$dst, $src}"),
+                           [(set VR128:$dst, (V4F32Int VR128:$src))],
+                           itins.rr>, VEX, Sched<[itins.Sched]>;
+  def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+                          !strconcat("v", OpcodeStr,
+                          "ps\t{$src, $dst|$dst, $src}"),
+                          [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
+                          itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
+  def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+                            !strconcat("v", OpcodeStr,
+                                       "ps\t{$src, $dst|$dst, $src}"),
+                            [(set VR256:$dst, (V8F32Int VR256:$src))],
+                            itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
+  def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
+                          (ins f256mem:$src),
+                          !strconcat("v", OpcodeStr,
+                                    "ps\t{$src, $dst|$dst, $src}"),
+                          [(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
+                          itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
+}
+
   def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                     !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
                     [(set VR128:$dst, (V4F32Int VR128:$src))],
-                    itins.rr>;
+                    itins.rr>, Sched<[itins.Sched]>;
   def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                     !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
                     [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
-                    itins.rm>;
-}
-
-/// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
-multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
-                                Intrinsic V4F32Int, OpndItins itins> {
-  def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
-                    !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
-                    [(set VR256:$dst, (V4F32Int VR256:$src))],
-                    itins.rr>, VEX_L;
-  def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
-                    !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
-                    [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
-                    itins.rm>, VEX_L;
+                    itins.rm>, Sched<[itins.Sched.Folded]>;
 }
 
 /// sse2_fp_unop_s - SSE2 unops in scalar form.
 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
                           SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
+let Predicates = [HasAVX], hasSideEffects = 0 in {
+  def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
+                      (ins FR64:$src1, FR64:$src2),
+                      !strconcat("v", OpcodeStr,
+                                 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                      []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
+  let mayLoad = 1 in {
+  def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
+                      (ins FR64:$src1,f64mem:$src2),
+                      !strconcat("v", OpcodeStr,
+                                 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                      []>, VEX_4V, VEX_LIG,
+                   Sched<[itins.Sched.Folded, ReadAfterLd]>;
+  def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
+                      (ins VR128:$src1, sdmem:$src2),
+                      !strconcat("v", OpcodeStr,
+                                 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                      []>, VEX_4V, VEX_LIG,
+                      Sched<[itins.Sched.Folded, ReadAfterLd]>;
+  }
+}
+
   def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
                 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
-                [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
+                [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>,
+            Sched<[itins.Sched]>;
   // See the comments in sse1_fp_unop_s for why this is OptForSize.
   def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
                 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
                 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
-            Requires<[UseSSE2, OptForSize]>;
+            Requires<[UseSSE2, OptForSize]>, Sched<[itins.Sched.Folded]>;
   def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                     !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
-                    [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
+                    [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>,
+                Sched<[itins.Sched]>;
   def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
                     !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
-                    [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
-}
-
-/// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
-let hasSideEffects = 0 in
-multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
-  def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
-               !strconcat(OpcodeStr,
-                          "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
-  let mayLoad = 1 in {
-  def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
-               !strconcat(OpcodeStr,
-                          "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
-  def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
-               (ins VR128:$src1, sdmem:$src2),
-               !strconcat(OpcodeStr,
-                          "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
-  }
+                    [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>,
+                Sched<[itins.Sched.Folded]>;
 }
 
 /// sse2_fp_unop_p - SSE2 unops in vector forms.
 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
                           SDNode OpNode, OpndItins itins> {
+let Predicates = [HasAVX] in {
+  def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+                       !strconcat("v", OpcodeStr,
+                                  "pd\t{$src, $dst|$dst, $src}"),
+                       [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
+                       itins.rr>, VEX, Sched<[itins.Sched]>;
+  def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+                       !strconcat("v", OpcodeStr,
+                                  "pd\t{$src, $dst|$dst, $src}"),
+                       [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))],
+                       itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
+  def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+                        !strconcat("v", OpcodeStr,
+                                   "pd\t{$src, $dst|$dst, $src}"),
+                        [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
+                        itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
+  def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
+                        !strconcat("v", OpcodeStr,
+                                   "pd\t{$src, $dst|$dst, $src}"),
+                        [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
+                        itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
+}
+
   def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
               !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
-              [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
+              [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>,
+            Sched<[itins.Sched]>;
   def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
-                [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
+                [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>,
+            Sched<[itins.Sched.Folded]>;
 }
 
-/// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
-multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
-                          OpndItins itins> {
-  def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
-              !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
-              [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
-              itins.rr>, VEX_L;
-  def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
-                !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
-                [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
-                itins.rm>, VEX_L;
-}
-
-/// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
-multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
-                              Intrinsic V2F64Int, OpndItins itins> {
-  def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
-                    !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
-                    [(set VR128:$dst, (V2F64Int VR128:$src))],
-                    itins.rr>;
-  def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
-                    !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
-                    [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
-                    itins.rm>;
-}
-
-/// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
-multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
-                                Intrinsic V2F64Int, OpndItins itins> {
-  def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
-                    !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
-                    [(set VR256:$dst, (V2F64Int VR256:$src))],
-                    itins.rr>, VEX_L;
-  def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
-                    !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
-                    [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
-                    itins.rm>, VEX_L;
-}
+// Square root.
+defm SQRT  : sse1_fp_unop_s<0x51, "sqrt",  fsqrt, int_x86_sse_sqrt_ss,
+                            SSE_SQRTS>,
+             sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>,
+             sse2_fp_unop_s<0x51, "sqrt",  fsqrt, int_x86_sse2_sqrt_sd,
+                            SSE_SQRTS>,
+             sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>;
 
-let Predicates = [HasAVX] in {
-  // Square root.
-  defm VSQRT  : sse1_fp_unop_s_avx<0x51, "vsqrt">,
-                sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
-
-  defm VSQRT  : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
-                sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
-                sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
-                sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
-                sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
-                                   SSE_SQRTP>,
-                sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
-                                    SSE_SQRTP>,
-                sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
-                                    SSE_SQRTP>,
-                sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
-                                    SSE_SQRTP>,
-                VEX;
-
-  // Reciprocal approximations. Note that these typically require refinement
-  // in order to obtain suitable precision.
-  defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
-  defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
-                sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
-                sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
-                                    SSE_SQRTP>,
-                sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
-                                    SSE_SQRTP>, VEX;
-
-  defm VRCP   : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
-  defm VRCP   : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
-                sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
-                sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
-                                    SSE_RCPP>,
-                sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
-                                    SSE_RCPP>, VEX;
-}
+// Reciprocal approximations. Note that these typically require refinement
+// in order to obtain suitable precision.
+defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
+             sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTP>,
+             sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
+                                int_x86_avx_rsqrt_ps_256, SSE_SQRTP>;
+defm RCP   : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
+             sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
+             sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
+                                int_x86_avx_rcp_ps_256, SSE_RCPP>;
 
 def : Pat<(f32 (fsqrt FR32:$src)),
           (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
@@ -3283,59 +3389,11 @@ let Predicates = [HasAVX] in {
             (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
 }
 
-// Square root.
-defm SQRT  : sse1_fp_unop_s<0x51, "sqrt",  fsqrt, int_x86_sse_sqrt_ss,
-                            SSE_SQRTS>,
-             sse1_fp_unop_p<0x51, "sqrt",  fsqrt, SSE_SQRTS>,
-             sse1_fp_unop_p_int<0x51, "sqrt",  int_x86_sse_sqrt_ps, SSE_SQRTS>,
-             sse2_fp_unop_s<0x51, "sqrt",  fsqrt, int_x86_sse2_sqrt_sd,
-                            SSE_SQRTS>,
-             sse2_fp_unop_p<0x51, "sqrt",  fsqrt, SSE_SQRTS>,
-             sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
-
-/// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
-multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
-                               Intrinsic F32Int, OpndItins itins> {
-  def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
-                !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
-                [(set FR32:$dst, (OpNode FR32:$src))]>;
-  // For scalar unary operations, fold a load into the operation
-  // only in OptForSize mode. It eliminates an instruction, but it also
-  // eliminates a whole-register clobber (the load), so it introduces a
-  // partial register update condition.
-  def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
-                !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
-                [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
-            Requires<[UseSSE1, OptForSize]>;
-  let Constraints = "$src1 = $dst" in {
-    def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
-                      (ins VR128:$src1, VR128:$src2),
-                      !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
-                      [], itins.rr>;
-    def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
-                      (ins VR128:$src1, ssmem:$src2),
-                      !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
-                      [], itins.rm>;
-  }
-}
-
 // Reciprocal approximations. Note that these typically require refinement
 // in order to obtain suitable precision.
-defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
-                             SSE_SQRTS>,
-             sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
-             sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
-                            SSE_SQRTS>;
 let Predicates = [UseSSE1] in {
   def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
             (RSQRTSSr_Int VR128:$src, VR128:$src)>;
-}
-
-defm RCP   : sse1_fp_unop_rw<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
-                             SSE_RCPS>,
-             sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
-             sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
-let Predicates = [UseSSE1] in {
   def : Pat<(int_x86_sse_rcp_ss VR128:$src),
             (RCPSSr_Int VR128:$src, VR128:$src)>;
 }
@@ -3347,52 +3405,48 @@ let Predicates = [UseSSE1] in {
 //===----------------------------------------------------------------------===//
 
 let AddedComplexity = 400 in { // Prefer non-temporal versions
-  def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
-                       (ins f128mem:$dst, VR128:$src),
-                       "movntps\t{$src, $dst|$dst, $src}",
-                       [(alignednontemporalstore (v4f32 VR128:$src),
-                                                 addr:$dst)],
-                                                 IIC_SSE_MOVNT>, VEX;
-  def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
-                       (ins f128mem:$dst, VR128:$src),
-                       "movntpd\t{$src, $dst|$dst, $src}",
-                       [(alignednontemporalstore (v2f64 VR128:$src),
-                                                 addr:$dst)],
-                                                 IIC_SSE_MOVNT>, VEX;
-
-  let ExeDomain = SSEPackedInt in
-  def VMOVNTDQmr    : VPDI<0xE7, MRMDestMem, (outs),
-                           (ins f128mem:$dst, VR128:$src),
-                           "movntdq\t{$src, $dst|$dst, $src}",
-                           [(alignednontemporalstore (v2i64 VR128:$src),
-                                                     addr:$dst)],
-                                                     IIC_SSE_MOVNT>, VEX;
-
-  def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
-            (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
-
-  def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
-                       (ins f256mem:$dst, VR256:$src),
-                       "movntps\t{$src, $dst|$dst, $src}",
-                       [(alignednontemporalstore (v8f32 VR256:$src),
-                                                 addr:$dst)],
-                                                 IIC_SSE_MOVNT>, VEX, VEX_L;
-  def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
-                       (ins f256mem:$dst, VR256:$src),
-                       "movntpd\t{$src, $dst|$dst, $src}",
-                       [(alignednontemporalstore (v4f64 VR256:$src),
-                                                 addr:$dst)],
-                                                 IIC_SSE_MOVNT>, VEX, VEX_L;
-  let ExeDomain = SSEPackedInt in
-  def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
-                      (ins f256mem:$dst, VR256:$src),
-                      "movntdq\t{$src, $dst|$dst, $src}",
-                      [(alignednontemporalstore (v4i64 VR256:$src),
-                                                addr:$dst)],
-                                                IIC_SSE_MOVNT>, VEX, VEX_L;
-}
+let SchedRW = [WriteStore] in {
+def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
+                     (ins f128mem:$dst, VR128:$src),
+                     "movntps\t{$src, $dst|$dst, $src}",
+                     [(alignednontemporalstore (v4f32 VR128:$src),
+                                               addr:$dst)],
+                                               IIC_SSE_MOVNT>, VEX;
+def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
+                     (ins f128mem:$dst, VR128:$src),
+                     "movntpd\t{$src, $dst|$dst, $src}",
+                     [(alignednontemporalstore (v2f64 VR128:$src),
+                                               addr:$dst)],
+                                               IIC_SSE_MOVNT>, VEX;
+
+let ExeDomain = SSEPackedInt in
+def VMOVNTDQmr    : VPDI<0xE7, MRMDestMem, (outs),
+                         (ins f128mem:$dst, VR128:$src),
+                         "movntdq\t{$src, $dst|$dst, $src}",
+                         [(alignednontemporalstore (v2i64 VR128:$src),
+                                                   addr:$dst)],
+                                                   IIC_SSE_MOVNT>, VEX;
+
+def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
+                     (ins f256mem:$dst, VR256:$src),
+                     "movntps\t{$src, $dst|$dst, $src}",
+                     [(alignednontemporalstore (v8f32 VR256:$src),
+                                               addr:$dst)],
+                                               IIC_SSE_MOVNT>, VEX, VEX_L;
+def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
+                     (ins f256mem:$dst, VR256:$src),
+                     "movntpd\t{$src, $dst|$dst, $src}",
+                     [(alignednontemporalstore (v4f64 VR256:$src),
+                                               addr:$dst)],
+                                               IIC_SSE_MOVNT>, VEX, VEX_L;
+let ExeDomain = SSEPackedInt in
+def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
+                    (ins f256mem:$dst, VR256:$src),
+                    "movntdq\t{$src, $dst|$dst, $src}",
+                    [(alignednontemporalstore (v4i64 VR256:$src),
+                                              addr:$dst)],
+                                              IIC_SSE_MOVNT>, VEX, VEX_L;
 
-let AddedComplexity = 400 in { // Prefer non-temporal versions
 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
                     "movntps\t{$src, $dst|$dst, $src}",
                     [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
@@ -3408,9 +3462,6 @@ def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
                     [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
                     IIC_SSE_MOVNT>;
 
-def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
-          (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
-
 // There is no AVX form for instructions below this point
 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
                  "movnti{l}\t{$src, $dst|$dst, $src}",
@@ -3422,14 +3473,21 @@ def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
                      [(nontemporalstore (i64 GR64:$src), addr:$dst)],
                      IIC_SSE_MOVNT>,
                   TB, Requires<[HasSSE2]>;
-}
+} // SchedRW = [WriteStore]
+
+def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
+          (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
+
+def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
+          (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
+} // AddedComplexity
 
 //===----------------------------------------------------------------------===//
 // SSE 1 & 2 - Prefetch and memory fence
 //===----------------------------------------------------------------------===//
 
 // Prefetch intrinsic.
-let Predicates = [HasSSE1] in {
+let Predicates = [HasSSE1], SchedRW = [WriteLoad] in {
 def PREFETCHT0   : I<0x18, MRM1m, (outs), (ins i8mem:$src),
     "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
     IIC_SSE_PREFETCH>, TB;
@@ -3444,6 +3502,8 @@ def PREFETCHNTA  : I<0x18, MRM0m, (outs), (ins i8mem:$src),
     IIC_SSE_PREFETCH>, TB;
 }
 
+// FIXME: How should these memory instructions be modeled?
+let SchedRW = [WriteLoad] in {
 // Flush cache
 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
                "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
@@ -3463,6 +3523,7 @@ def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
                "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
                TB, Requires<[HasSSE2]>;
+} // SchedRW
 
 def : Pat<(X86SFence), (SFENCE)>;
 def : Pat<(X86LFence), (LFENCE)>;
@@ -3474,17 +3535,17 @@ def : Pat<(X86MFence), (MFENCE)>;
 
 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
                   "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
-                  IIC_SSE_LDMXCSR>, VEX;
+                  IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>;
 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
                   "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
-                  IIC_SSE_STMXCSR>, VEX;
+                  IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>;
 
 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
                   "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
-                  IIC_SSE_LDMXCSR>;
+                  IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>;
 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
                   "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
-                  IIC_SSE_STMXCSR>;
+                  IIC_SSE_STMXCSR>, Sched<[WriteStore]>;
 
 //===---------------------------------------------------------------------===//
 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
@@ -3492,23 +3553,23 @@ def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
 
 let ExeDomain = SSEPackedInt in { // SSE integer instructions
 
-let neverHasSideEffects = 1 in {
+let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
 def VMOVDQArr  : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                     "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
                     VEX;
 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
                     "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
                     VEX, VEX_L;
-}
 def VMOVDQUrr  : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                     "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
                     VEX;
 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
                     "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
                     VEX, VEX_L;
+}
 
 // For Disassembler
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in {
 def VMOVDQArr_REV  : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
                         "movdqa\t{$src, $dst|$dst, $src}", [],
                         IIC_SSE_MOVA_P_RR>,
@@ -3525,7 +3586,8 @@ def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
                         IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
 }
 
-let canFoldAsLoad = 1, mayLoad = 1 in {
+let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
+    neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
 def VMOVDQArm  : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
                    "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
                    VEX;
@@ -3542,7 +3604,7 @@ let Predicates = [HasAVX] in {
 }
 }
 
-let mayStore = 1 in {
+let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
 def VMOVDQAmr  : VPDI<0x7F, MRMDestMem, (outs),
                      (ins i128mem:$dst, VR128:$src),
                      "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
@@ -3561,6 +3623,7 @@ def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
 }
 }
 
+let SchedRW = [WriteMove] in {
 let neverHasSideEffects = 1 in
 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                    "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
@@ -3570,7 +3633,7 @@ def MOVDQUrr :   I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                    [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
 
 // For Disassembler
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, hasSideEffects = 0 in {
 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
                        "movdqa\t{$src, $dst|$dst, $src}", [],
                        IIC_SSE_MOVA_P_RR>;
@@ -3579,8 +3642,10 @@ def MOVDQUrr_REV :   I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
                        "movdqu\t{$src, $dst|$dst, $src}",
                        [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
 }
+} // SchedRW
 
-let canFoldAsLoad = 1, mayLoad = 1 in {
+let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
+    neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
                    "movdqa\t{$src, $dst|$dst, $src}",
                    [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
@@ -3592,7 +3657,7 @@ def MOVDQUrm :   I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
                  XS, Requires<[UseSSE2]>;
 }
 
-let mayStore = 1 in {
+let mayStore = 1, SchedRW = [WriteStore] in {
 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
                    "movdqa\t{$src, $dst|$dst, $src}",
                    [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
@@ -3604,30 +3669,23 @@ def MOVDQUmr :   I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
                  XS, Requires<[UseSSE2]>;
 }
 
-// Intrinsic forms of MOVDQU load and store
-def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
-                       "vmovdqu\t{$src, $dst|$dst, $src}",
-                       [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
-                       IIC_SSE_MOVU_P_MR>,
-                     XS, VEX, Requires<[HasAVX]>;
-
-def MOVDQUmr_Int :   I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
-                       "movdqu\t{$src, $dst|$dst, $src}",
-                       [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
-                       IIC_SSE_MOVU_P_MR>,
-                     XS, Requires<[UseSSE2]>;
-
 } // ExeDomain = SSEPackedInt
 
 let Predicates = [HasAVX] in {
+  def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
+            (VMOVDQUmr addr:$dst, VR128:$src)>;
   def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
             (VMOVDQUYmr addr:$dst, VR256:$src)>;
 }
+let Predicates = [UseSSE2] in
+def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
+          (MOVDQUmr addr:$dst, VR128:$src)>;
 
 //===---------------------------------------------------------------------===//
 // SSE2 - Packed Integer Arithmetic Instructions
 //===---------------------------------------------------------------------===//
 
+let Sched = WriteVecIMul in
 def SSE_PMADD : OpndItins<
   IIC_SSE_PMADD, IIC_SSE_PMADD
 >;
@@ -3646,14 +3704,33 @@ multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
+       [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>,
+      Sched<[itins.Sched]>;
   def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
        (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
-       itins.rm>;
+       itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
+}
+
+multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
+                             Intrinsic IntId256, OpndItins itins,
+                             bit IsCommutable = 0> {
+let Predicates = [HasAVX] in
+  defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
+                                 VR128, memopv2i64, i128mem, itins,
+                                 IsCommutable, 0>, VEX_4V;
+
+let Constraints = "$src1 = $dst" in
+  defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
+                               i128mem, itins, IsCommutable, 1>;
+
+let Predicates = [HasAVX2] in
+  defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
+                                   VR256, memopv4i64, i256mem, itins,
+                                   IsCommutable, 0>, VEX_4V, VEX_L;
 }
 
 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
@@ -3669,23 +3746,25 @@ multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
-        itins.rr>;
+        itins.rr>, Sched<[WriteVecShift]>;
   def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
        (ins RC:$src1, i128mem:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst, (DstVT (OpNode RC:$src1,
-                       (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
+                       (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>,
+      Sched<[WriteVecShiftLd, ReadAfterLd]>;
   def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
        (ins RC:$src1, i32i8imm:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
+       [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>,
+       Sched<[WriteVecShift]>;
 }
 
-/// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
+/// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
                          ValueType DstVT, ValueType SrcVT, RegisterClass RC,
                          PatFrag memop_frag, X86MemOperand x86memop,
@@ -3697,260 +3776,88 @@ multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
+       [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
+       Sched<[itins.Sched]>;
   def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
        (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
-                                     (bitconvert (memop_frag addr:$src2)))))]>;
+                                     (bitconvert (memop_frag addr:$src2)))))]>,
+       Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 } // ExeDomain = SSEPackedInt
 
-// 128-bit Integer Arithmetic
+defm PADDB   : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
+                             SSE_INTALU_ITINS_P, 1>;
+defm PADDW   : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
+                             SSE_INTALU_ITINS_P, 1>;
+defm PADDD   : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
+                             SSE_INTALU_ITINS_P, 1>;
+defm PADDQ   : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
+                             SSE_INTALUQ_ITINS_P, 1>;
+defm PMULLW  : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
+                             SSE_INTMUL_ITINS_P, 1>;
+defm PSUBB   : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
+                             SSE_INTALU_ITINS_P, 0>;
+defm PSUBW   : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
+                             SSE_INTALU_ITINS_P, 0>;
+defm PSUBD   : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
+                             SSE_INTALU_ITINS_P, 0>;
+defm PSUBQ   : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
+                             SSE_INTALUQ_ITINS_P, 0>;
+defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
+                             SSE_INTALU_ITINS_P, 0>;
+defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
+                             SSE_INTALU_ITINS_P, 0>;
+defm PMINUB  : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
+                             SSE_INTALU_ITINS_P, 1>;
+defm PMINSW  : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
+                             SSE_INTALU_ITINS_P, 1>;
+defm PMAXUB  : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
+                             SSE_INTALU_ITINS_P, 1>;
+defm PMAXSW  : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
+                             SSE_INTALU_ITINS_P, 1>;
 
-let Predicates = [HasAVX] in {
-defm VPADDB  : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
-                            i128mem, SSE_INTALU_ITINS_P, 1, 0 /*3addr*/>,
-                            VEX_4V;
-defm VPADDW  : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
-                            i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDD  : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
-                            i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDQ  : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
-                            i128mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
-defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
-                            i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
-defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
-                            i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
-                            i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
-                            i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
-                            i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
+// Intrinsic forms
+defm PSUBSB  : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
+                                 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
+defm PSUBSW  : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
+                                 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
+defm PADDSB  : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
+                                 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
+defm PADDSW  : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
+                                 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
+defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
+                                 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
+defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
+                                 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
+defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
+                                 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
+defm PMULHW  : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
+                                 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
+defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
+                                 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
+defm PAVGB   : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
+                                 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
+defm PAVGW   : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
+                                 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
+defm PSADBW  : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
+                                 int_x86_avx2_psad_bw, SSE_INTALU_ITINS_P, 1>;
+
+let Predicates = [HasAVX] in
 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
                               memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
                               VEX_4V;
-
-// Intrinsic forms
-defm VPSUBSB  : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBSW  : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPADDSB  : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDSW  : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
-defm VPMULHW  : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
-defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_PMADD, 1, 0>, VEX_4V;
-defm VPAVGB   : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPAVGW   : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMINUB  : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMINSW  : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMAXUB  : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMAXSW  : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPSADBW  : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-}
-
-let Predicates = [HasAVX2] in {
-defm VPADDBY  : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
-                             i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPADDWY  : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
-                             i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPADDDY  : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
-                             i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPADDQY  : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
-                             i256mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
-                             i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPSUBBY  : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
-                             i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPSUBWY  : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
-                             i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPSUBDY  : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
-                             i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPSUBQY  : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
-                             i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V, VEX_L;
+let Predicates = [HasAVX2] in
 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
                                VR256, memopv4i64, i256mem,
                                SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-
-// Intrinsic forms
-defm VPSUBSBY  : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPSUBSWY  : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPADDSBY  : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPADDSWY  : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMULHWY  : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_PMADD, 1, 0>, VEX_4V, VEX_L;
-defm VPAVGBY   : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPAVGWY   : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMINUBY  : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMINSWY  : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMAXUBY  : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPMAXSWY  : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-defm VPSADBWY  : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
-                                  VR256, memopv4i64, i256mem,
-                                  SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-}
-
-let Constraints = "$src1 = $dst" in {
-defm PADDB  : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
-                           i128mem, SSE_INTALU_ITINS_P, 1>;
-defm PADDW  : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
-                           i128mem, SSE_INTALU_ITINS_P, 1>;
-defm PADDD  : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
-                           i128mem, SSE_INTALU_ITINS_P, 1>;
-defm PADDQ  : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
-                           i128mem, SSE_INTALUQ_ITINS_P, 1>;
-defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
-                           i128mem, SSE_INTMUL_ITINS_P, 1>;
-defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
-                          i128mem, SSE_INTALU_ITINS_P>;
-defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
-                          i128mem, SSE_INTALU_ITINS_P>;
-defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
-                          i128mem, SSE_INTALU_ITINS_P>;
-defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
-                          i128mem, SSE_INTALUQ_ITINS_P>;
+let Constraints = "$src1 = $dst" in
 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
                              memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
 
-// Intrinsic forms
-defm PSUBSB  : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P>;
-defm PSUBSW  : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P>;
-defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P>;
-defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P>;
-defm PADDSB  : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-defm PADDSW  : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTMUL_ITINS_P, 1>;
-defm PMULHW  : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTMUL_ITINS_P, 1>;
-defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
-                                VR128, memopv2i64, i128mem,
-                                SSE_PMADD, 1>;
-defm PAVGB   : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-defm PAVGW   : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-defm PMINUB  : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-defm PMINSW  : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-defm PMAXUB  : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-defm PMAXSW  : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-defm PSADBW  : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1>;
-
-} // Constraints = "$src1 = $dst"
-
 //===---------------------------------------------------------------------===//
 // SSE2 - Packed Integer Logical Instructions
 //===---------------------------------------------------------------------===//
@@ -3983,7 +3890,7 @@ defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
                             VR128, v4i32, v4i32, bc_v4i32,
                             SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
 
-let ExeDomain = SSEPackedInt in {
+let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
   // 128-bit logical shifts.
   def VPSLLDQri : PDIi8<0x73, MRM7r,
                     (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
@@ -4029,7 +3936,7 @@ defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
                              VR256, v8i32, v4i32, bc_v4i32,
                              SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
 
-let ExeDomain = SSEPackedInt in {
+let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
   // 256-bit logical shifts.
   def VPSLLDQYri : PDIi8<0x73, MRM7r,
                     (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
@@ -4075,7 +3982,7 @@ defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
                            VR128, v4i32, v4i32, bc_v4i32,
                            SSE_INTSHIFT_ITINS_P>;
 
-let ExeDomain = SSEPackedInt in {
+let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
   // 128-bit logical shifts.
   def PSLLDQri : PDIi8<0x73, MRM7r,
                        (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
@@ -4132,186 +4039,109 @@ let Predicates = [UseSSE2] in {
 // SSE2 - Packed Integer Comparison Instructions
 //===---------------------------------------------------------------------===//
 
-let Predicates = [HasAVX] in {
-  defm VPCMPEQB  : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-  defm VPCMPEQW  : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-  defm VPCMPEQD  : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-  defm VPCMPGTB  : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-  defm VPCMPGTW  : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-  defm VPCMPGTD  : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
-                                VR128, memopv2i64, i128mem,
-                                SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-}
-
-let Predicates = [HasAVX2] in {
-  defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
-                                VR256, memopv4i64, i256mem,
-                                SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-  defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
-                                VR256, memopv4i64, i256mem,
-                                SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-  defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
-                                VR256, memopv4i64, i256mem,
-                                SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L;
-  defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
-                                VR256, memopv4i64, i256mem,
-                                SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-  defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
-                                VR256, memopv4i64, i256mem,
-                                SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-  defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
-                                VR256, memopv4i64, i256mem,
-                                SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-}
-
-let Constraints = "$src1 = $dst" in {
-  defm PCMPEQB  : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
-                               VR128, memopv2i64, i128mem,
-                               SSE_INTALU_ITINS_P, 1>;
-  defm PCMPEQW  : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
-                               VR128, memopv2i64, i128mem,
-                               SSE_INTALU_ITINS_P, 1>;
-  defm PCMPEQD  : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
-                               VR128, memopv2i64, i128mem,
-                               SSE_INTALU_ITINS_P, 1>;
-  defm PCMPGTB  : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
-                               VR128, memopv2i64, i128mem,
-                               SSE_INTALU_ITINS_P>;
-  defm PCMPGTW  : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
-                               VR128, memopv2i64, i128mem,
-                               SSE_INTALU_ITINS_P>;
-  defm PCMPGTD  : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
-                               VR128, memopv2i64, i128mem,
-                               SSE_INTALU_ITINS_P>;
-} // Constraints = "$src1 = $dst"
+defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
+                             SSE_INTALU_ITINS_P, 1>;
+defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
+                             SSE_INTALU_ITINS_P, 1>;
+defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
+                             SSE_INTALU_ITINS_P, 1>;
+defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
+                             SSE_INTALU_ITINS_P, 0>;
+defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
+                             SSE_INTALU_ITINS_P, 0>;
+defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
+                             SSE_INTALU_ITINS_P, 0>;
 
 //===---------------------------------------------------------------------===//
 // SSE2 - Packed Integer Pack Instructions
 //===---------------------------------------------------------------------===//
 
-let Predicates = [HasAVX] in {
-defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
-                                  VR128, memopv2i64, i128mem,
-                                  SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
-                                  VR128, memopv2i64, i128mem,
-                                  SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
-                                  VR128, memopv2i64, i128mem,
-                                  SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-}
-
-let Predicates = [HasAVX2] in {
-defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
-                                   VR256, memopv4i64, i256mem,
-                                   SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
-                                   VR256, memopv4i64, i256mem,
-                                   SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
-                                   VR256, memopv4i64, i256mem,
-                                   SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
-}
-
-let Constraints = "$src1 = $dst" in {
-defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P>;
-defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P>;
-defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
-                                 VR128, memopv2i64, i128mem,
-                                 SSE_INTALU_ITINS_P>;
-} // Constraints = "$src1 = $dst"
+defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
+                                  int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
+defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
+                                  int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
+defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
+                                  int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
 
 //===---------------------------------------------------------------------===//
 // SSE2 - Packed Integer Shuffle Instructions
 //===---------------------------------------------------------------------===//
 
 let ExeDomain = SSEPackedInt in {
-multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
-def ri : Ii8<0x70, MRMSrcReg,
-             (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
-             !strconcat(OpcodeStr,
-                        "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-              [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
-              IIC_SSE_PSHUF>;
-def mi : Ii8<0x70, MRMSrcMem,
-             (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
-             !strconcat(OpcodeStr,
-                        "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-              [(set VR128:$dst,
-                (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
-                             (i8 imm:$src2))))],
-                             IIC_SSE_PSHUF>;
-}
-
-multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
-def Yri : Ii8<0x70, MRMSrcReg,
-              (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
-              !strconcat(OpcodeStr,
-                         "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-              [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
-def Ymi : Ii8<0x70, MRMSrcMem,
-              (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
-              !strconcat(OpcodeStr,
-                         "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-              [(set VR256:$dst,
-                (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
-                             (i8 imm:$src2))))]>;
-}
-} // ExeDomain = SSEPackedInt
-
+multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
+                         SDNode OpNode> {
 let Predicates = [HasAVX] in {
- let AddedComplexity = 5 in
-  defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
-
- // SSE2 with ImmT == Imm8 and XS prefix.
-  defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
-
- // SSE2 with ImmT == Imm8 and XD prefix.
-  defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
-
- def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
-           (VPSHUFDmi addr:$src1, imm:$imm)>;
- def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
-           (VPSHUFDri VR128:$src1, imm:$imm)>;
+  def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
+                      (ins VR128:$src1, i8imm:$src2),
+                      !strconcat("v", OpcodeStr,
+                                 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                      [(set VR128:$dst,
+                        (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
+                      IIC_SSE_PSHUF>, VEX, Sched<[WriteShuffle]>;
+  def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
+                      (ins i128mem:$src1, i8imm:$src2),
+                      !strconcat("v", OpcodeStr,
+                                 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                     [(set VR128:$dst,
+                       (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
+                        (i8 imm:$src2))))], IIC_SSE_PSHUF>, VEX,
+                  Sched<[WriteShuffleLd]>;
 }
 
 let Predicates = [HasAVX2] in {
-  defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>,
-                                TB, OpSize, VEX,VEX_L;
-  defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>,
-                                  XS, VEX, VEX_L;
-  defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>,
-                                  XD, VEX, VEX_L;
+  def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
+                       (ins VR256:$src1, i8imm:$src2),
+                       !strconcat("v", OpcodeStr,
+                                  "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                       [(set VR256:$dst,
+                         (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
+                       IIC_SSE_PSHUF>, VEX, VEX_L, Sched<[WriteShuffle]>;
+  def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
+                       (ins i256mem:$src1, i8imm:$src2),
+                       !strconcat("v", OpcodeStr,
+                                  "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                      [(set VR256:$dst,
+                        (vt256 (OpNode (bitconvert (memopv4i64 addr:$src1)),
+                         (i8 imm:$src2))))], IIC_SSE_PSHUF>, VEX, VEX_L,
+                   Sched<[WriteShuffleLd]>;
 }
 
 let Predicates = [UseSSE2] in {
- let AddedComplexity = 5 in
-  defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
+  def ri : Ii8<0x70, MRMSrcReg,
+               (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
+               !strconcat(OpcodeStr,
+                          "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                [(set VR128:$dst,
+                  (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
+                IIC_SSE_PSHUF>, Sched<[WriteShuffle]>;
+  def mi : Ii8<0x70, MRMSrcMem,
+               (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
+               !strconcat(OpcodeStr,
+                          "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                [(set VR128:$dst,
+                  (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
+                          (i8 imm:$src2))))], IIC_SSE_PSHUF>,
+           Sched<[WriteShuffleLd]>;
+}
+}
+} // ExeDomain = SSEPackedInt
 
- // SSE2 with ImmT == Imm8 and XS prefix.
-  defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
+defm PSHUFD  : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, TB, OpSize;
+defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS;
+defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD;
 
- // SSE2 with ImmT == Imm8 and XD prefix.
-  defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
+let Predicates = [HasAVX] in {
+  def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
+            (VPSHUFDmi addr:$src1, imm:$imm)>;
+  def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+            (VPSHUFDri VR128:$src1, imm:$imm)>;
+}
 
- def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
-           (PSHUFDmi addr:$src1, imm:$imm)>;
- def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
-           (PSHUFDri VR128:$src1, imm:$imm)>;
+let Predicates = [UseSSE2] in {
+  def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
+            (PSHUFDmi addr:$src1, imm:$imm)>;
+  def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+            (PSHUFDri VR128:$src1, imm:$imm)>;
 }
 
 //===---------------------------------------------------------------------===//
@@ -4327,7 +4157,7 @@ multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
           !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
           !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
       [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
-      IIC_SSE_UNPCK>;
+      IIC_SSE_UNPCK>, Sched<[WriteShuffle]>;
   def rm : PDI<opc, MRMSrcMem,
       (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
       !if(Is2Addr,
@@ -4336,7 +4166,8 @@ multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
       [(set VR128:$dst, (OpNode VR128:$src1,
                                   (bc_frag (memopv2i64
                                                addr:$src2))))],
-                                               IIC_SSE_UNPCK>;
+                                               IIC_SSE_UNPCK>,
+      Sched<[WriteShuffleLd, ReadAfterLd]>;
 }
 
 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
@@ -4344,12 +4175,14 @@ multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
   def Yrr : PDI<opc, MRMSrcReg,
       (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
       !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-      [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
+      [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>,
+      Sched<[WriteShuffle]>;
   def Yrm : PDI<opc, MRMSrcMem,
       (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
       !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
       [(set VR256:$dst, (OpNode VR256:$src1,
-                                  (bc_frag (memopv4i64 addr:$src2))))]>;
+                                  (bc_frag (memopv4i64 addr:$src2))))]>,
+      Sched<[WriteShuffleLd, ReadAfterLd]>;
 }
 
 let Predicates = [HasAVX] in {
@@ -4426,7 +4259,8 @@ multiclass sse2_pinsrw<bit Is2Addr = 1> {
            "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
            "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
        [(set VR128:$dst,
-         (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
+         (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>,
+       Sched<[WriteShuffle]>;
   def rmi : Ii8<0xC4, MRMSrcMem,
                        (outs VR128:$dst), (ins VR128:$src1,
                         i16mem:$src2, i32i8imm:$src3),
@@ -4435,7 +4269,8 @@ multiclass sse2_pinsrw<bit Is2Addr = 1> {
            "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
        [(set VR128:$dst,
          (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
-                    imm:$src3))], IIC_SSE_PINSRW>;
+                    imm:$src3))], IIC_SSE_PINSRW>,
+       Sched<[WriteShuffleLd, ReadAfterLd]>;
 }
 
 // Extract
@@ -4444,12 +4279,14 @@ def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
                     (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
                     "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                     [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
-                                                imm:$src2))]>, TB, OpSize, VEX;
+                                                imm:$src2))]>, TB, OpSize, VEX,
+                Sched<[WriteShuffle]>;
 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
                     (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
                     "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                     [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
-                                                imm:$src2))], IIC_SSE_PEXTRW>;
+                                                imm:$src2))], IIC_SSE_PEXTRW>,
+               Sched<[WriteShuffleLd, ReadAfterLd]>;
 
 // Insert
 let Predicates = [HasAVX] in {
@@ -4457,7 +4294,7 @@ let Predicates = [HasAVX] in {
   def  VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
        (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
        "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
-       []>, TB, OpSize, VEX_4V;
+       []>, TB, OpSize, VEX_4V, Sched<[WriteShuffle]>;
 }
 
 let Constraints = "$src1 = $dst" in
@@ -4469,7 +4306,7 @@ let Constraints = "$src1 = $dst" in
 // SSE2 - Packed Mask Creation
 //===---------------------------------------------------------------------===//
 
-let ExeDomain = SSEPackedInt in {
+let ExeDomain = SSEPackedInt, SchedRW = [WriteVecLogic] in {
 
 def VPMOVMSKBrr  : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
            "pmovmskb\t{$src, $dst|$dst, $src}",
@@ -4497,7 +4334,7 @@ def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
 // SSE2 - Conditional Store
 //===---------------------------------------------------------------------===//
 
-let ExeDomain = SSEPackedInt in {
+let ExeDomain = SSEPackedInt, SchedRW = [WriteStore] in {
 
 let Uses = [EDI] in
 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
@@ -4536,41 +4373,42 @@ def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst,
                         (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
-                        VEX;
+                        VEX, Sched<[WriteMove]>;
 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst,
                         (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
                         IIC_SSE_MOVDQ>,
-                      VEX;
+                      VEX, Sched<[WriteLoad]>;
 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
                         "mov{d|q}\t{$src, $dst|$dst, $src}",
                         [(set VR128:$dst,
                           (v2i64 (scalar_to_vector GR64:$src)))],
-                          IIC_SSE_MOVDQ>, VEX;
+                          IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
                        "mov{d|q}\t{$src, $dst|$dst, $src}",
                        [(set FR64:$dst, (bitconvert GR64:$src))],
-                       IIC_SSE_MOVDQ>, VEX;
+                       IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
 
 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst,
-                        (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
+                        (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
+                  Sched<[WriteMove]>;
 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst,
                         (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
-                        IIC_SSE_MOVDQ>;
+                        IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
                         "mov{d|q}\t{$src, $dst|$dst, $src}",
                         [(set VR128:$dst,
                           (v2i64 (scalar_to_vector GR64:$src)))],
-                          IIC_SSE_MOVDQ>;
+                          IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
                        "mov{d|q}\t{$src, $dst|$dst, $src}",
                        [(set FR64:$dst, (bitconvert GR64:$src))],
-                       IIC_SSE_MOVDQ>;
+                       IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
 
 //===---------------------------------------------------------------------===//
 // Move Int Doubleword to Single Scalar
@@ -4578,22 +4416,22 @@ def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
 def VMOVDI2SSrr  : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(set FR32:$dst, (bitconvert GR32:$src))],
-                      IIC_SSE_MOVDQ>, VEX;
+                      IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
 
 def VMOVDI2SSrm  : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
                       IIC_SSE_MOVDQ>,
-                      VEX;
+                      VEX, Sched<[WriteLoad]>;
 def MOVDI2SSrr  : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(set FR32:$dst, (bitconvert GR32:$src))],
-                      IIC_SSE_MOVDQ>;
+                      IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
 
 def MOVDI2SSrm  : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
-                      IIC_SSE_MOVDQ>;
+                      IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
 
 //===---------------------------------------------------------------------===//
 // Move Packed Doubleword Int to Packed Double Int
@@ -4601,26 +4439,29 @@ def MOVDI2SSrm  : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
 def VMOVPDI2DIrr  : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
                        "movd\t{$src, $dst|$dst, $src}",
                        [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
-                                        (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
+                                        (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX,
+                    Sched<[WriteMove]>;
 def VMOVPDI2DImr  : VPDI<0x7E, MRMDestMem, (outs),
                        (ins i32mem:$dst, VR128:$src),
                        "movd\t{$src, $dst|$dst, $src}",
                        [(store (i32 (vector_extract (v4i32 VR128:$src),
                                      (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
-                                     VEX;
+                                     VEX, Sched<[WriteLoad]>;
 def MOVPDI2DIrr  : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
                        "movd\t{$src, $dst|$dst, $src}",
                        [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
-                                        (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
+                                        (iPTR 0)))], IIC_SSE_MOVD_ToGP>,
+                   Sched<[WriteMove]>;
 def MOVPDI2DImr  : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
                        "movd\t{$src, $dst|$dst, $src}",
                        [(store (i32 (vector_extract (v4i32 VR128:$src),
                                      (iPTR 0))), addr:$dst)],
-                                     IIC_SSE_MOVDQ>;
+                                     IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
 
 //===---------------------------------------------------------------------===//
 // Move Packed Doubleword Int first element to Doubleword Int
 //
+let SchedRW = [WriteMove] in {
 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
                           "vmov{d|q}\t{$src, $dst|$dst, $src}",
                           [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
@@ -4633,6 +4474,7 @@ def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
                         [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
                                                          (iPTR 0)))],
                                                          IIC_SSE_MOVD_ToGP>;
+} //SchedRW
 
 //===---------------------------------------------------------------------===//
 // Bitcast FR64 <-> GR64
@@ -4641,28 +4483,28 @@ let Predicates = [HasAVX] in
 def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
                         "vmovq\t{$src, $dst|$dst, $src}",
                         [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
-                        VEX;
+                        VEX, Sched<[WriteLoad]>;
 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
                          "mov{d|q}\t{$src, $dst|$dst, $src}",
                          [(set GR64:$dst, (bitconvert FR64:$src))],
-                         IIC_SSE_MOVDQ>, VEX;
+                         IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
                          "movq\t{$src, $dst|$dst, $src}",
                          [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
-                         IIC_SSE_MOVDQ>, VEX;
+                         IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
 
 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
                        "movq\t{$src, $dst|$dst, $src}",
                        [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
-                       IIC_SSE_MOVDQ>;
+                       IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
                        "mov{d|q}\t{$src, $dst|$dst, $src}",
                        [(set GR64:$dst, (bitconvert FR64:$src))],
-                       IIC_SSE_MOVD_ToGP>;
+                       IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
                        "movq\t{$src, $dst|$dst, $src}",
                        [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
-                       IIC_SSE_MOVDQ>;
+                       IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
 
 //===---------------------------------------------------------------------===//
 // Move Scalar Single to Double Int
@@ -4670,23 +4512,24 @@ def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
 def VMOVSS2DIrr  : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(set GR32:$dst, (bitconvert FR32:$src))],
-                      IIC_SSE_MOVD_ToGP>, VEX;
+                      IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>;
 def VMOVSS2DImr  : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
-                      IIC_SSE_MOVDQ>, VEX;
+                      IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
 def MOVSS2DIrr  : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(set GR32:$dst, (bitconvert FR32:$src))],
-                      IIC_SSE_MOVD_ToGP>;
+                      IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
 def MOVSS2DImr  : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
                       "movd\t{$src, $dst|$dst, $src}",
                       [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
-                      IIC_SSE_MOVDQ>;
+                      IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
 
 //===---------------------------------------------------------------------===//
 // Patterns and instructions to describe movd/movq to XMM register zero-extends
 //
+let SchedRW = [WriteMove] in {
 let AddedComplexity = 15 in {
 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
                        "movd\t{$src, $dst|$dst, $src}",
@@ -4712,8 +4555,9 @@ def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
                                       (v2i64 (scalar_to_vector GR64:$src)))))],
                                       IIC_SSE_MOVDQ>;
 }
+} // SchedRW
 
-let AddedComplexity = 20 in {
+let AddedComplexity = 20, SchedRW = [WriteLoad] in {
 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
                        "movd\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst,
@@ -4726,7 +4570,7 @@ def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
                          (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
                                                    (loadi32 addr:$src))))))],
                                                    IIC_SSE_MOVDQ>;
-}
+} // AddedComplexity, SchedRW
 
 let Predicates = [HasAVX] in {
   // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
@@ -4775,6 +4619,8 @@ def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
 //===---------------------------------------------------------------------===//
 // Move Quadword Int to Packed Quadword Int
 //
+
+let SchedRW = [WriteLoad] in {
 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
                     "vmovq\t{$src, $dst|$dst, $src}",
                     [(set VR128:$dst,
@@ -4786,10 +4632,12 @@ def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
                       (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
                       IIC_SSE_MOVDQ>, XS,
                     Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
+} // SchedRW
 
 //===---------------------------------------------------------------------===//
 // Move Packed Quadword Int to Quadword Int
 //
+let SchedRW = [WriteStore] in {
 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
                       "movq\t{$src, $dst|$dst, $src}",
                       [(store (i64 (vector_extract (v2i64 VR128:$src),
@@ -4800,17 +4648,19 @@ def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
                       [(store (i64 (vector_extract (v2i64 VR128:$src),
                                     (iPTR 0))), addr:$dst)],
                                     IIC_SSE_MOVDQ>;
+} // SchedRW
 
 //===---------------------------------------------------------------------===//
 // Store / copy lower 64-bits of a XMM register.
 //
 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
                      "movq\t{$src, $dst|$dst, $src}",
-                     [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
+                     [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX,
+                  Sched<[WriteStore]>;
 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
                      "movq\t{$src, $dst|$dst, $src}",
                      [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
-                     IIC_SSE_MOVDQ>;
+                     IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
 
 let AddedComplexity = 20 in
 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
@@ -4819,7 +4669,7 @@ def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
                        (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
                                                  (loadi64 addr:$src))))))],
                                                  IIC_SSE_MOVDQ>,
-                     XS, VEX, Requires<[HasAVX]>;
+                     XS, VEX, Requires<[HasAVX]>, Sched<[WriteLoad]>;
 
 let AddedComplexity = 20 in
 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
@@ -4828,7 +4678,7 @@ def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
                        (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
                                                  (loadi64 addr:$src))))))],
                                                  IIC_SSE_MOVDQ>,
-                     XS, Requires<[UseSSE2]>;
+                     XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>;
 
 let Predicates = [HasAVX], AddedComplexity = 20 in {
   def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
@@ -4858,6 +4708,7 @@ def : Pat<(v4i64 (X86vzload addr:$src)),
 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
 // IA32 document. movq xmm1, xmm2 does clear the high bits.
 //
+let SchedRW = [WriteVecLogic] in {
 let AddedComplexity = 15 in
 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                         "vmovq\t{$src, $dst|$dst, $src}",
@@ -4870,7 +4721,9 @@ def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                     [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
                     IIC_SSE_MOVQ_RR>,
                       XS, Requires<[UseSSE2]>;
+} // SchedRW
 
+let SchedRW = [WriteVecLogicLd] in {
 let AddedComplexity = 20 in
 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
                         "vmovq\t{$src, $dst|$dst, $src}",
@@ -4886,6 +4739,7 @@ def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
                                              IIC_SSE_MOVDQ>,
                       XS, Requires<[UseSSE2]>;
 }
+} // SchedRW
 
 let AddedComplexity = 20 in {
   let Predicates = [HasAVX] in {
@@ -4903,6 +4757,7 @@ let AddedComplexity = 20 in {
 }
 
 // Instructions to match in the assembler
+let SchedRW = [WriteMove] in {
 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
                       "movq\t{$src, $dst|$dst, $src}", [],
                       IIC_SSE_MOVDQ>, VEX, VEX_W;
@@ -4913,16 +4768,19 @@ def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
                           "movd\t{$src, $dst|$dst, $src}", [],
                           IIC_SSE_MOVDQ>, VEX, VEX_W;
+} // SchedRW
 
 // Instructions for the disassembler
 // xr = XMM register
 // xm = mem64
 
+let SchedRW = [WriteMove] in {
 let Predicates = [HasAVX] in
 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                  "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                  "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
+} // SchedRW
 
 //===---------------------------------------------------------------------===//
 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
@@ -4933,11 +4791,11 @@ multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                       [(set RC:$dst, (vt (OpNode RC:$src)))],
-                      IIC_SSE_MOV_LH>;
+                      IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                       [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
-                      IIC_SSE_MOV_LH>;
+                      IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
 }
 
 let Predicates = [HasAVX] in {
@@ -4993,25 +4851,27 @@ multiclass sse3_replicate_dfp<string OpcodeStr> {
 let neverHasSideEffects = 1 in
 def rr  : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                    [], IIC_SSE_MOV_LH>;
+                    [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
 def rm  : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                     [(set VR128:$dst,
                       (v2f64 (X86Movddup
                               (scalar_to_vector (loadf64 addr:$src)))))],
-                              IIC_SSE_MOV_LH>;
+                              IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
 }
 
 // FIXME: Merge with above classe when there're patterns for the ymm version
 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
 def rr  : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                    [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
+                    [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>,
+                    Sched<[WriteShuffle]>;
 def rm  : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                     [(set VR256:$dst,
                       (v4f64 (X86Movddup
-                              (scalar_to_vector (loadf64 addr:$src)))))]>;
+                              (scalar_to_vector (loadf64 addr:$src)))))]>,
+                    Sched<[WriteShuffleLd]>;
 }
 
 let Predicates = [HasAVX] in {
@@ -5059,6 +4919,7 @@ let Predicates = [UseSSE3] in {
 // SSE3 - Move Unaligned Integer
 //===---------------------------------------------------------------------===//
 
+let SchedRW = [WriteLoad] in {
 let Predicates = [HasAVX] in {
   def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
                    "vlddqu\t{$src, $dst|$dst, $src}",
@@ -5072,6 +4933,7 @@ def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
                    "lddqu\t{$src, $dst|$dst, $src}",
                    [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
                    IIC_SSE_LDDQU>;
+}
 
 //===---------------------------------------------------------------------===//
 // SSE3 - Arithmetic
@@ -5085,13 +4947,15 @@ multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
+       [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>,
+       Sched<[itins.Sched]>;
   def rm : I<0xD0, MRMSrcMem,
        (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
            !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
            !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-       [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
+       [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>,
+       Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 let Predicates = [HasAVX] in {
@@ -5128,14 +4992,15 @@ multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
        !if(Is2Addr,
          !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
          !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-      [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
+      [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
+      Sched<[WriteFAdd]>;
 
   def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
          !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
          !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
       [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
-        IIC_SSE_HADDSUB_RM>;
+        IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
 }
 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
                   X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
@@ -5143,14 +5008,15 @@ multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
        !if(Is2Addr,
          !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
          !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
-      [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
+      [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
+      Sched<[WriteFAdd]>;
 
   def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
          !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
          !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
       [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
-        IIC_SSE_HADDSUB_RM>;
+        IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
 }
 
 let Predicates = [HasAVX] in {
@@ -5199,7 +5065,7 @@ multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
                     (ins VR128:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                     [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
-                    OpSize;
+                    OpSize, Sched<[WriteVecALU]>;
 
   def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
                     (ins i128mem:$src),
@@ -5207,7 +5073,7 @@ multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
                     [(set VR128:$dst,
                       (IntId128
                        (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
-                    OpSize;
+                    OpSize, Sched<[WriteVecALULd]>;
 }
 
 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
@@ -5217,14 +5083,15 @@ multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
                     (ins VR256:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                     [(set VR256:$dst, (IntId256 VR256:$src))]>,
-                    OpSize;
+                    OpSize, Sched<[WriteVecALU]>;
 
   def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
                     (ins i256mem:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                     [(set VR256:$dst,
                       (IntId256
-                       (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
+                       (bitconvert (memopv4i64 addr:$src))))]>, OpSize,
+                    Sched<[WriteVecALULd]>;
 }
 
 let Predicates = [HasAVX] in {
@@ -5256,6 +5123,7 @@ defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
 // SSSE3 - Packed Binary Operator Instructions
 //===---------------------------------------------------------------------===//
 
+let Sched = WriteVecALU in {
 def SSE_PHADDSUBD : OpndItins<
   IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
 >;
@@ -5265,12 +5133,16 @@ def SSE_PHADDSUBSW : OpndItins<
 def SSE_PHADDSUBW : OpndItins<
   IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
 >;
+}
+let Sched = WriteShuffle in
 def SSE_PSHUFB : OpndItins<
   IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
 >;
+let Sched = WriteVecALU in
 def SSE_PSIGN : OpndItins<
   IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
 >;
+let Sched = WriteVecIMul in
 def SSE_PMULHRSW : OpndItins<
   IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
 >;
@@ -5287,7 +5159,7 @@ multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
          !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
          !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
-       OpSize;
+       OpSize, Sched<[itins.Sched]>;
   def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
        (ins RC:$src1, x86memop:$src2),
        !if(Is2Addr,
@@ -5295,7 +5167,8 @@ multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
          !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set RC:$dst,
          (OpVT (OpNode RC:$src1,
-          (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
+          (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize,
+       Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
@@ -5309,7 +5182,7 @@ multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
          !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
          !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
-       OpSize;
+       OpSize, Sched<[itins.Sched]>;
   def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
        (ins VR128:$src1, i128mem:$src2),
        !if(Is2Addr,
@@ -5317,7 +5190,8 @@ multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
          !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
        [(set VR128:$dst,
          (IntId128 VR128:$src1,
-          (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
+          (bitconvert (memopv2i64 addr:$src2))))]>, OpSize,
+       Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
@@ -5451,7 +5325,7 @@ defm PMULHRSW    : SS3I_binop_rm_int<0x0B, "pmulhrsw",
 // SSSE3 - Packed Align Instruction Patterns
 //===---------------------------------------------------------------------===//
 
-multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
+multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
   let neverHasSideEffects = 1 in {
   def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
       (ins VR128:$src1, VR128:$src2, i8imm:$src3),
@@ -5459,7 +5333,7 @@ multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
         !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
         !strconcat(asm,
                   "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
-      [], IIC_SSE_PALIGNR>, OpSize;
+      [], IIC_SSE_PALIGNR>, OpSize, Sched<[WriteShuffle]>;
   let mayLoad = 1 in
   def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
       (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
@@ -5467,63 +5341,63 @@ multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
         !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
         !strconcat(asm,
                   "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
-      [], IIC_SSE_PALIGNR>, OpSize;
+      [], IIC_SSE_PALIGNR>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
   }
 }
 
-multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
+multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
   let neverHasSideEffects = 1 in {
   def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
       (ins VR256:$src1, VR256:$src2, i8imm:$src3),
       !strconcat(asm,
                  "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
-      []>, OpSize;
+      []>, OpSize, Sched<[WriteShuffle]>;
   let mayLoad = 1 in
   def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
       (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
       !strconcat(asm,
                  "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
-      []>, OpSize;
+      []>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>;
   }
 }
 
 let Predicates = [HasAVX] in
-  defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
+  defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V;
 let Predicates = [HasAVX2] in
-  defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V, VEX_L;
+  defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L;
 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
-  defm PALIGN : ssse3_palign<"palignr">;
+  defm PALIGN : ssse3_palignr<"palignr">;
 
 let Predicates = [HasAVX2] in {
-def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
           (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
-def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
           (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
-def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
           (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
-def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
           (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
 }
 
 let Predicates = [HasAVX] in {
-def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
           (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
           (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
           (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
           (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
 }
 
 let Predicates = [UseSSSE3] in {
-def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
           (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
           (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
           (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
           (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
 }
 
@@ -5531,6 +5405,7 @@ def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
 // SSSE3 - Thread synchronization
 //===---------------------------------------------------------------------===//
 
+let SchedRW = [WriteSystem] in {
 let usesCustomInserter = 1 in {
 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
                 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
@@ -5544,6 +5419,7 @@ let Uses = [ECX, EAX] in
 def MWAITrr   : I<0x01, MRM_C9, (outs), (ins), "mwait",
                 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
                 TB, Requires<[HasSSE3]>;
+} // SchedRW
 
 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
@@ -5850,6 +5726,55 @@ defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
 defm PMOVSXBQ   : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
 defm PMOVZXBQ   : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
 
+let Predicates = [HasAVX2] in {
+  def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
+  def : Pat<(v8i32  (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
+  def : Pat<(v4i64  (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
+
+  def : Pat<(v8i32  (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
+  def : Pat<(v4i64  (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
+
+  def : Pat<(v4i64  (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
+
+  def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
+            (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+  def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
+            (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+  def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
+            (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+  def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
+            (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+  def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
+            (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+  def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
+            (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+  def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
+            (VPMOVSXWDYrm addr:$src)>;
+  def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
+            (VPMOVSXDQYrm addr:$src)>;
+
+  def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64 
+                    (scalar_to_vector (loadi64 addr:$src))))))),
+            (VPMOVSXBDYrm addr:$src)>;
+  def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64 
+                    (scalar_to_vector (loadf64 addr:$src))))))),
+            (VPMOVSXBDYrm addr:$src)>;
+
+  def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64 
+                    (scalar_to_vector (loadi64 addr:$src))))))),
+            (VPMOVSXWQYrm addr:$src)>;
+  def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64 
+                    (scalar_to_vector (loadf64 addr:$src))))))),
+            (VPMOVSXWQYrm addr:$src)>;
+
+  def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32 
+                    (scalar_to_vector (loadi32 addr:$src))))))),
+            (VPMOVSXBQYrm addr:$src)>;
+}
+
 let Predicates = [HasAVX] in {
   // Common patterns involving scalar load
   def : Pat<(int_x86_sse41_pmovsxbq
@@ -5864,6 +5789,15 @@ let Predicates = [HasAVX] in {
 }
 
 let Predicates = [UseSSE41] in {
+  def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
+  def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
+  def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
+
+  def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
+  def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
+
+  def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
+
   // Common patterns involving scalar load
   def : Pat<(int_x86_sse41_pmovsxbq
               (bitconvert (v4i32 (X86vzmovl
@@ -5874,6 +5808,34 @@ let Predicates = [UseSSE41] in {
               (bitconvert (v4i32 (X86vzmovl
                             (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
             (PMOVZXBQrm addr:$src)>;
+
+  def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
+                    (scalar_to_vector (loadi64 addr:$src))))))),
+            (PMOVSXWDrm addr:$src)>;
+  def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
+                    (scalar_to_vector (loadf64 addr:$src))))))),
+            (PMOVSXWDrm addr:$src)>;
+  def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
+                    (scalar_to_vector (loadi32 addr:$src))))))),
+            (PMOVSXBDrm addr:$src)>;
+  def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
+                    (scalar_to_vector (loadi32 addr:$src))))))),
+            (PMOVSXWQrm addr:$src)>;
+  def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
+                    (scalar_to_vector (extloadi32i16 addr:$src))))))),
+            (PMOVSXBQrm addr:$src)>;
+  def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
+                    (scalar_to_vector (loadi64 addr:$src))))))),
+            (PMOVSXDQrm addr:$src)>;
+  def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
+                    (scalar_to_vector (loadf64 addr:$src))))))),
+            (PMOVSXDQrm addr:$src)>;
+  def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
+                    (scalar_to_vector (loadi64 addr:$src))))))),
+            (PMOVSXBWrm addr:$src)>;
+  def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
+                    (scalar_to_vector (loadf64 addr:$src))))))),
+            (PMOVSXBWrm addr:$src)>;
 }
 
 let Predicates = [HasAVX2] in {
@@ -5934,6 +5896,44 @@ let Predicates = [HasAVX] in {
             (VPMOVZXDQrm addr:$src)>;
   def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
             (VPMOVZXDQrm addr:$src)>;
+
+  def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
+  def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>;
+  def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>;
+
+  def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
+  def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>;
+
+  def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
+
+  def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
+                    (scalar_to_vector (loadi64 addr:$src))))))),
+            (VPMOVSXWDrm addr:$src)>;
+  def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
+                    (scalar_to_vector (loadi64 addr:$src))))))),
+            (VPMOVSXDQrm addr:$src)>;
+  def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
+                    (scalar_to_vector (loadf64 addr:$src))))))),
+            (VPMOVSXWDrm addr:$src)>;
+  def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
+                    (scalar_to_vector (loadf64 addr:$src))))))),
+            (VPMOVSXDQrm addr:$src)>;
+  def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
+                    (scalar_to_vector (loadi64 addr:$src))))))),
+            (VPMOVSXBWrm addr:$src)>;
+  def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
+                    (scalar_to_vector (loadf64 addr:$src))))))),
+            (VPMOVSXBWrm addr:$src)>;
+
+  def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
+                    (scalar_to_vector (loadi32 addr:$src))))))),
+            (VPMOVSXBDrm addr:$src)>;
+  def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
+                    (scalar_to_vector (loadi32 addr:$src))))))),
+            (VPMOVSXWQrm addr:$src)>;
+  def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
+                    (scalar_to_vector (extloadi32i16 addr:$src))))))),
+            (VPMOVSXBQrm addr:$src)>;
 }
 
 let Predicates = [UseSSE41] in {
@@ -6273,6 +6273,7 @@ multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
                             Intrinsic F64Int, bit Is2Addr = 1> {
 let ExeDomain = GenericDomain in {
   // Operation, reg.
+  let hasSideEffects = 0 in
   def SSr : SS4AIi8<opcss, MRMSrcReg,
       (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
       !if(Is2Addr,
@@ -6306,6 +6307,7 @@ let ExeDomain = GenericDomain in {
         OpSize;
 
   // Operation, reg.
+  let hasSideEffects = 0 in
   def SDr : SS4AIi8<opcsd, MRMSrcReg,
         (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
         !if(Is2Addr,
@@ -6378,12 +6380,47 @@ let Predicates = [HasAVX] in {
 
   def : Pat<(v4f32 (ffloor VR128:$src)),
             (VROUNDPSr VR128:$src, (i32 0x1))>;
+  def : Pat<(v4f32 (fnearbyint VR128:$src)),
+            (VROUNDPSr VR128:$src, (i32 0xC))>;
+  def : Pat<(v4f32 (fceil VR128:$src)),
+            (VROUNDPSr VR128:$src, (i32 0x2))>;
+  def : Pat<(v4f32 (frint VR128:$src)),
+            (VROUNDPSr VR128:$src, (i32 0x4))>;
+  def : Pat<(v4f32 (ftrunc VR128:$src)),
+            (VROUNDPSr VR128:$src, (i32 0x3))>;
+
   def : Pat<(v2f64 (ffloor VR128:$src)),
             (VROUNDPDr VR128:$src, (i32 0x1))>;
+  def : Pat<(v2f64 (fnearbyint VR128:$src)),
+            (VROUNDPDr VR128:$src, (i32 0xC))>;
+  def : Pat<(v2f64 (fceil VR128:$src)),
+            (VROUNDPDr VR128:$src, (i32 0x2))>;
+  def : Pat<(v2f64 (frint VR128:$src)),
+            (VROUNDPDr VR128:$src, (i32 0x4))>;
+  def : Pat<(v2f64 (ftrunc VR128:$src)),
+            (VROUNDPDr VR128:$src, (i32 0x3))>;
+
   def : Pat<(v8f32 (ffloor VR256:$src)),
             (VROUNDYPSr VR256:$src, (i32 0x1))>;
+  def : Pat<(v8f32 (fnearbyint VR256:$src)),
+            (VROUNDYPSr VR256:$src, (i32 0xC))>;
+  def : Pat<(v8f32 (fceil VR256:$src)),
+            (VROUNDYPSr VR256:$src, (i32 0x2))>;
+  def : Pat<(v8f32 (frint VR256:$src)),
+            (VROUNDYPSr VR256:$src, (i32 0x4))>;
+  def : Pat<(v8f32 (ftrunc VR256:$src)),
+            (VROUNDYPSr VR256:$src, (i32 0x3))>;
+
   def : Pat<(v4f64 (ffloor VR256:$src)),
             (VROUNDYPDr VR256:$src, (i32 0x1))>;
+  def : Pat<(v4f64 (fnearbyint VR256:$src)),
+            (VROUNDYPDr VR256:$src, (i32 0xC))>;
+  def : Pat<(v4f64 (fceil VR256:$src)),
+            (VROUNDYPDr VR256:$src, (i32 0x2))>;
+  def : Pat<(v4f64 (frint VR256:$src)),
+            (VROUNDYPDr VR256:$src, (i32 0x4))>;
+  def : Pat<(v4f64 (ftrunc VR256:$src)),
+            (VROUNDYPDr VR256:$src, (i32 0x3))>;
 }
 
 defm ROUND  : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
@@ -6417,8 +6454,25 @@ let Predicates = [UseSSE41] in {
 
   def : Pat<(v4f32 (ffloor VR128:$src)),
             (ROUNDPSr VR128:$src, (i32 0x1))>;
+  def : Pat<(v4f32 (fnearbyint VR128:$src)),
+            (ROUNDPSr VR128:$src, (i32 0xC))>;
+  def : Pat<(v4f32 (fceil VR128:$src)),
+            (ROUNDPSr VR128:$src, (i32 0x2))>;
+  def : Pat<(v4f32 (frint VR128:$src)),
+            (ROUNDPSr VR128:$src, (i32 0x4))>;
+  def : Pat<(v4f32 (ftrunc VR128:$src)),
+            (ROUNDPSr VR128:$src, (i32 0x3))>;
+
   def : Pat<(v2f64 (ffloor VR128:$src)),
             (ROUNDPDr VR128:$src, (i32 0x1))>;
+  def : Pat<(v2f64 (fnearbyint VR128:$src)),
+            (ROUNDPDr VR128:$src, (i32 0xC))>;
+  def : Pat<(v2f64 (fceil VR128:$src)),
+            (ROUNDPDr VR128:$src, (i32 0x2))>;
+  def : Pat<(v2f64 (frint VR128:$src)),
+            (ROUNDPDr VR128:$src, (i32 0x4))>;
+  def : Pat<(v2f64 (ftrunc VR128:$src)),
+            (ROUNDPDr VR128:$src, (i32 0x3))>;
 }
 
 //===----------------------------------------------------------------------===//
@@ -6575,67 +6629,6 @@ multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
           (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
 }
 
-let Predicates = [HasAVX] in {
-  let isCommutable = 0 in
-  defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
-                                                         0>, VEX_4V;
-  defm VPMINSB   : SS41I_binop_rm_int<0x38, "vpminsb",   int_x86_sse41_pminsb,
-                                                         0>, VEX_4V;
-  defm VPMINSD   : SS41I_binop_rm_int<0x39, "vpminsd",   int_x86_sse41_pminsd,
-                                                         0>, VEX_4V;
-  defm VPMINUD   : SS41I_binop_rm_int<0x3B, "vpminud",   int_x86_sse41_pminud,
-                                                         0>, VEX_4V;
-  defm VPMINUW   : SS41I_binop_rm_int<0x3A, "vpminuw",   int_x86_sse41_pminuw,
-                                                         0>, VEX_4V;
-  defm VPMAXSB   : SS41I_binop_rm_int<0x3C, "vpmaxsb",   int_x86_sse41_pmaxsb,
-                                                         0>, VEX_4V;
-  defm VPMAXSD   : SS41I_binop_rm_int<0x3D, "vpmaxsd",   int_x86_sse41_pmaxsd,
-                                                         0>, VEX_4V;
-  defm VPMAXUD   : SS41I_binop_rm_int<0x3F, "vpmaxud",   int_x86_sse41_pmaxud,
-                                                         0>, VEX_4V;
-  defm VPMAXUW   : SS41I_binop_rm_int<0x3E, "vpmaxuw",   int_x86_sse41_pmaxuw,
-                                                         0>, VEX_4V;
-  defm VPMULDQ   : SS41I_binop_rm_int<0x28, "vpmuldq",   int_x86_sse41_pmuldq,
-                                                         0>, VEX_4V;
-}
-
-let Predicates = [HasAVX2] in {
-  let isCommutable = 0 in
-  defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
-                                        int_x86_avx2_packusdw>, VEX_4V, VEX_L;
-  defm VPMINSB   : SS41I_binop_rm_int_y<0x38, "vpminsb",
-                                        int_x86_avx2_pmins_b>, VEX_4V, VEX_L;
-  defm VPMINSD   : SS41I_binop_rm_int_y<0x39, "vpminsd",
-                                        int_x86_avx2_pmins_d>, VEX_4V, VEX_L;
-  defm VPMINUD   : SS41I_binop_rm_int_y<0x3B, "vpminud",
-                                        int_x86_avx2_pminu_d>, VEX_4V, VEX_L;
-  defm VPMINUW   : SS41I_binop_rm_int_y<0x3A, "vpminuw",
-                                        int_x86_avx2_pminu_w>, VEX_4V, VEX_L;
-  defm VPMAXSB   : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
-                                        int_x86_avx2_pmaxs_b>, VEX_4V, VEX_L;
-  defm VPMAXSD   : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
-                                        int_x86_avx2_pmaxs_d>, VEX_4V, VEX_L;
-  defm VPMAXUD   : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
-                                        int_x86_avx2_pmaxu_d>, VEX_4V, VEX_L;
-  defm VPMAXUW   : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
-                                        int_x86_avx2_pmaxu_w>, VEX_4V, VEX_L;
-  defm VPMULDQ   : SS41I_binop_rm_int_y<0x28, "vpmuldq",
-                                        int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
-}
-
-let Constraints = "$src1 = $dst" in {
-  let isCommutable = 0 in
-  defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
-  defm PMINSB   : SS41I_binop_rm_int<0x38, "pminsb",   int_x86_sse41_pminsb>;
-  defm PMINSD   : SS41I_binop_rm_int<0x39, "pminsd",   int_x86_sse41_pminsd>;
-  defm PMINUD   : SS41I_binop_rm_int<0x3B, "pminud",   int_x86_sse41_pminud>;
-  defm PMINUW   : SS41I_binop_rm_int<0x3A, "pminuw",   int_x86_sse41_pminuw>;
-  defm PMAXSB   : SS41I_binop_rm_int<0x3C, "pmaxsb",   int_x86_sse41_pmaxsb>;
-  defm PMAXSD   : SS41I_binop_rm_int<0x3D, "pmaxsd",   int_x86_sse41_pmaxsd>;
-  defm PMAXUD   : SS41I_binop_rm_int<0x3F, "pmaxud",   int_x86_sse41_pmaxud>;
-  defm PMAXUW   : SS41I_binop_rm_int<0x3E, "pmaxuw",   int_x86_sse41_pmaxuw>;
-  defm PMULDQ   : SS41I_binop_rm_int<0x28, "pmuldq",   int_x86_sse41_pmuldq>;
-}
 
 /// SS48I_binop_rm - Simple SSE41 binary operator.
 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
@@ -6659,6 +6652,76 @@ multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
 }
 
 let Predicates = [HasAVX] in {
+  let isCommutable = 0 in
+  defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
+                                                         0>, VEX_4V;
+  defm VPMINSB   : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
+                                  memopv2i64, i128mem, 0>, VEX_4V;
+  defm VPMINSD   : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
+                                  memopv2i64, i128mem, 0>, VEX_4V;
+  defm VPMINUD   : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
+                                  memopv2i64, i128mem, 0>, VEX_4V;
+  defm VPMINUW   : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
+                                  memopv2i64, i128mem, 0>, VEX_4V;
+  defm VPMAXSB   : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
+                                  memopv2i64, i128mem, 0>, VEX_4V;
+  defm VPMAXSD   : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
+                                  memopv2i64, i128mem, 0>, VEX_4V;
+  defm VPMAXUD   : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
+                                  memopv2i64, i128mem, 0>, VEX_4V;
+  defm VPMAXUW   : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
+                                  memopv2i64, i128mem, 0>, VEX_4V;
+  defm VPMULDQ   : SS41I_binop_rm_int<0x28, "vpmuldq",   int_x86_sse41_pmuldq,
+                                                         0>, VEX_4V;
+}
+
+let Predicates = [HasAVX2] in {
+  let isCommutable = 0 in
+  defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
+                                        int_x86_avx2_packusdw>, VEX_4V, VEX_L;
+  defm VPMINSBY  : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
+                                  memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+  defm VPMINSDY  : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
+                                  memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+  defm VPMINUDY  : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
+                                  memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+  defm VPMINUWY  : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
+                                  memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+  defm VPMAXSBY  : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
+                                  memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+  defm VPMAXSDY  : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
+                                  memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+  defm VPMAXUDY  : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
+                                  memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+  defm VPMAXUWY  : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
+                                  memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+  defm VPMULDQ   : SS41I_binop_rm_int_y<0x28, "vpmuldq",
+                                        int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
+}
+
+let Constraints = "$src1 = $dst" in {
+  let isCommutable = 0 in
+  defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
+  defm PMINSB   : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
+                                 memopv2i64, i128mem>;
+  defm PMINSD   : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
+                                 memopv2i64, i128mem>;
+  defm PMINUD   : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
+                                 memopv2i64, i128mem>;
+  defm PMINUW   : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
+                                 memopv2i64, i128mem>;
+  defm PMAXSB   : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
+                                 memopv2i64, i128mem>;
+  defm PMAXSD   : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
+                                 memopv2i64, i128mem>;
+  defm PMAXUD   : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
+                                 memopv2i64, i128mem>;
+  defm PMAXUW   : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
+                                 memopv2i64, i128mem>;
+  defm PMULDQ   : SS41I_binop_rm_int<0x28, "pmuldq",   int_x86_sse41_pmuldq>;
+}
+
+let Predicates = [HasAVX] in {
   defm VPMULLD  : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
                                 memopv2i64, i128mem, 0>, VEX_4V;
   defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
@@ -6776,7 +6839,7 @@ multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
                   !strconcat(OpcodeStr,
                     "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
                   [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
-                  IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
+                  NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
 
   def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
                   (ins RC:$src1, x86memop:$src2, RC:$src3),
@@ -6785,7 +6848,7 @@ multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
                   [(set RC:$dst,
                         (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
                                RC:$src3))],
-                  IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
+                  NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
 }
 
 let Predicates = [HasAVX] in {
@@ -6839,31 +6902,31 @@ let Predicates = [HasAVX] in {
                             (v4f64 VR256:$src2))),
             (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
 
-  def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2),
+  def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
                                (imm:$mask))),
-            (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>;
-  def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2),
+            (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
+  def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
                                (imm:$mask))),
-            (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>;
+            (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
 
-  def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
+  def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
                                (imm:$mask))),
-            (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
-  def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
+            (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
+  def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
                                (imm:$mask))),
-            (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
-  def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
+            (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
+  def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
                                (imm:$mask))),
-            (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
+            (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
 }
 
 let Predicates = [HasAVX2] in {
   def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
                             (v32i8 VR256:$src2))),
-            (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
-  def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2),
+            (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
+  def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
                                (imm:$mask))),
-            (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>;
+            (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
 }
 
 /// SS41I_ternary_int - SSE 4.1 ternary operator
@@ -6927,15 +6990,15 @@ let Predicates = [UseSSE41] in {
                             (v2f64 VR128:$src2))),
             (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
 
-  def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
+  def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
                                (imm:$mask))),
-            (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
-  def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
+            (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
+  def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
                                (imm:$mask))),
-            (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
-  def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
+            (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
+  def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
                                (imm:$mask))),
-            (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
+            (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
 
 }
 
@@ -7821,6 +7884,13 @@ defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
                                     VR256, memopv4i64, i256mem>, VEX_L;
 }
 
+def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
+                  imm:$mask)),
+          (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
+def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
+                  imm:$mask)),
+          (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
+
 //===----------------------------------------------------------------------===//
 // VPBROADCAST - Load from memory and broadcast to all elements of the
 //               destination operand
diff --git a/lib/Target/X86/X86InstrShiftRotate.td b/lib/Target/X86/X86InstrShiftRotate.td
index 893488c159ea..5b6298b541bc 100644
--- a/lib/Target/X86/X86InstrShiftRotate.td
+++ b/lib/Target/X86/X86InstrShiftRotate.td
@@ -15,7 +15,7 @@
 
 let Defs = [EFLAGS] in {
 
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in {
 let Uses = [CL] in {
 def SHL8rCL  : I<0xD2, MRM4r, (outs GR8 :$dst), (ins GR8 :$src1),
                  "shl{b}\t{%cl, $dst|$dst, CL}",
@@ -51,6 +51,7 @@ def SHL64ri  : RIi8<0xC1, MRM4r, (outs GR64:$dst),
 
 // NOTE: We don't include patterns for shifts of a register by one, because
 // 'add reg,reg' is cheaper (and we have a Pat pattern for shift-by-one).
+let hasSideEffects = 0 in {
 def SHL8r1   : I<0xD0, MRM4r, (outs GR8:$dst), (ins GR8:$src1),
                  "shl{b}\t$dst", [], IIC_SR>;
 def SHL16r1  : I<0xD1, MRM4r, (outs GR16:$dst), (ins GR16:$src1),
@@ -59,10 +60,12 @@ def SHL32r1  : I<0xD1, MRM4r, (outs GR32:$dst), (ins GR32:$src1),
                  "shl{l}\t$dst", [], IIC_SR>;
 def SHL64r1  : RI<0xD1, MRM4r, (outs GR64:$dst), (ins GR64:$src1),
                  "shl{q}\t$dst", [], IIC_SR>;
+} // hasSideEffects = 0
 } // isConvertibleToThreeAddress = 1
-} // Constraints = "$src = $dst" 
+} // Constraints = "$src = $dst", SchedRW
 
 
+let SchedRW = [WriteShiftLd, WriteRMW] in {
 // FIXME: Why do we need an explicit "Uses = [CL]" when the instr has a pattern
 // using CL?
 let Uses = [CL] in {
@@ -116,8 +119,9 @@ def SHL64m1 : RI<0xD1, MRM4m, (outs), (ins i64mem:$dst),
                   "shl{q}\t$dst",
                  [(store (shl (loadi64 addr:$dst), (i8 1)), addr:$dst)],
                  IIC_SR>;
+} // SchedRW
 
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in {
 let Uses = [CL] in {
 def SHR8rCL  : I<0xD2, MRM5r, (outs GR8 :$dst), (ins GR8 :$src1),
                  "shr{b}\t{%cl, $dst|$dst, CL}",
@@ -161,9 +165,10 @@ def SHR32r1  : I<0xD1, MRM5r, (outs GR32:$dst), (ins GR32:$src1),
 def SHR64r1  : RI<0xD1, MRM5r, (outs GR64:$dst), (ins GR64:$src1),
                  "shr{q}\t$dst",
                  [(set GR64:$dst, (srl GR64:$src1, (i8 1)))], IIC_SR>;
-} // Constraints = "$src = $dst"
+} // Constraints = "$src = $dst", SchedRW
 
 
+let SchedRW = [WriteShiftLd, WriteRMW] in {
 let Uses = [CL] in {
 def SHR8mCL  : I<0xD2, MRM5m, (outs), (ins i8mem :$dst),
                  "shr{b}\t{%cl, $dst|$dst, CL}",
@@ -214,8 +219,9 @@ def SHR64m1 : RI<0xD1, MRM5m, (outs), (ins i64mem:$dst),
                   "shr{q}\t$dst",
                  [(store (srl (loadi64 addr:$dst), (i8 1)), addr:$dst)],
                  IIC_SR>;
+} // SchedRW
 
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in {
 let Uses = [CL] in {
 def SAR8rCL  : I<0xD2, MRM7r, (outs GR8 :$dst), (ins GR8 :$src1),
                  "sar{b}\t{%cl, $dst|$dst, CL}",
@@ -271,9 +277,10 @@ def SAR64r1  : RI<0xD1, MRM7r, (outs GR64:$dst), (ins GR64:$src1),
                  "sar{q}\t$dst",
                  [(set GR64:$dst, (sra GR64:$src1, (i8 1)))],
                  IIC_SR>;
-} // Constraints = "$src = $dst"
+} // Constraints = "$src = $dst", SchedRW
 
 
+let SchedRW = [WriteShiftLd, WriteRMW] in {
 let Uses = [CL] in {
 def SAR8mCL  : I<0xD2, MRM7m, (outs), (ins i8mem :$dst),
                  "sar{b}\t{%cl, $dst|$dst, CL}",
@@ -328,12 +335,14 @@ def SAR64m1 : RI<0xD1, MRM7m, (outs), (ins i64mem:$dst),
                   "sar{q}\t$dst",
                  [(store (sra (loadi64 addr:$dst), (i8 1)), addr:$dst)],
                  IIC_SR>;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // Rotate instructions
 //===----------------------------------------------------------------------===//
 
-let Constraints = "$src1 = $dst" in {
+let hasSideEffects = 0 in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in {
 def RCL8r1 : I<0xD0, MRM2r, (outs GR8:$dst), (ins GR8:$src1),
                "rcl{b}\t$dst", [], IIC_SR>;
 def RCL8ri : Ii8<0xC0, MRM2r, (outs GR8:$dst), (ins GR8:$src1, i8imm:$cnt),
@@ -402,6 +411,7 @@ def RCR64rCL : RI<0xD3, MRM3r, (outs GR64:$dst), (ins GR64:$src1),
 
 } // Constraints = "$src = $dst"
 
+let SchedRW = [WriteShiftLd, WriteRMW] in {
 def RCL8m1 : I<0xD0, MRM2m, (outs), (ins i8mem:$dst),
                "rcl{b}\t$dst", [], IIC_SR>;
 def RCL8mi : Ii8<0xC0, MRM2m, (outs), (ins i8mem:$dst, i8imm:$cnt),
@@ -455,8 +465,10 @@ def RCR32mCL : I<0xD3, MRM3m, (outs), (ins i32mem:$dst),
 def RCR64mCL : RI<0xD3, MRM3m, (outs), (ins i64mem:$dst),
                   "rcr{q}\t{%cl, $dst|$dst, CL}", [], IIC_SR>;
 }
+} // SchedRW
+} // hasSideEffects = 0
 
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in {
 // FIXME: provide shorter instructions when imm8 == 1
 let Uses = [CL] in {
 def ROL8rCL  : I<0xD2, MRM0r, (outs GR8 :$dst), (ins GR8 :$src1),
@@ -508,8 +520,9 @@ def ROL64r1  : RI<0xD1, MRM0r, (outs GR64:$dst), (ins GR64:$src1),
                   "rol{q}\t$dst",
                   [(set GR64:$dst, (rotl GR64:$src1, (i8 1)))],
                   IIC_SR>;
-} // Constraints = "$src = $dst"
+} // Constraints = "$src = $dst", SchedRW
 
+let SchedRW = [WriteShiftLd, WriteRMW] in {
 let Uses = [CL] in {
 def ROL8mCL  : I<0xD2, MRM0m, (outs), (ins i8mem :$dst),
                  "rol{b}\t{%cl, $dst|$dst, CL}",
@@ -564,8 +577,9 @@ def ROL64m1  : RI<0xD1, MRM0m, (outs), (ins i64mem:$dst),
                  "rol{q}\t$dst",
                [(store (rotl (loadi64 addr:$dst), (i8 1)), addr:$dst)],
                IIC_SR>;
+} // SchedRW
 
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in {
 let Uses = [CL] in {
 def ROR8rCL  : I<0xD2, MRM1r, (outs GR8 :$dst), (ins GR8 :$src1),
                  "ror{b}\t{%cl, $dst|$dst, CL}",
@@ -616,8 +630,9 @@ def ROR64r1  : RI<0xD1, MRM1r, (outs GR64:$dst), (ins GR64:$src1),
                   "ror{q}\t$dst",
                   [(set GR64:$dst, (rotr GR64:$src1, (i8 1)))],
                   IIC_SR>;
-} // Constraints = "$src = $dst"
+} // Constraints = "$src = $dst", SchedRW
 
+let SchedRW = [WriteShiftLd, WriteRMW] in {
 let Uses = [CL] in {
 def ROR8mCL  : I<0xD2, MRM1m, (outs), (ins i8mem :$dst),
                  "ror{b}\t{%cl, $dst|$dst, CL}",
@@ -672,13 +687,14 @@ def ROR64m1  : RI<0xD1, MRM1m, (outs), (ins i64mem:$dst),
                  "ror{q}\t$dst",
                [(store (rotr (loadi64 addr:$dst), (i8 1)), addr:$dst)],
                IIC_SR>;
+} // SchedRW
 
 
 //===----------------------------------------------------------------------===//
 // Double shift instructions (generalizations of rotate)
 //===----------------------------------------------------------------------===//
 
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in {
 
 let Uses = [CL] in {
 def SHLD16rrCL : I<0xA5, MRMDestReg, (outs GR16:$dst), 
@@ -761,8 +777,9 @@ def SHRD64rri8 : RIi8<0xAC, MRMDestReg,
                                        (i8 imm:$src3)))], IIC_SHD64_REG_IM>,
                  TB;
 }
-} // Constraints = "$src = $dst"
+} // Constraints = "$src = $dst", SchedRW
 
+let SchedRW = [WriteShiftLd, WriteRMW] in {
 let Uses = [CL] in {
 def SHLD16mrCL : I<0xA5, MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src2),
                    "shld{w}\t{%cl, $src2, $dst|$dst, $src2, CL}",
@@ -836,6 +853,7 @@ def SHRD64mri8 : RIi8<0xAC, MRMDestMem,
                                        (i8 imm:$src3)), addr:$dst)],
                                        IIC_SHD64_MEM_IM>,
                  TB;
+} // SchedRW
 
 } // Defs = [EFLAGS]
 
@@ -853,12 +871,12 @@ multiclass bmi_rotate<string asm, RegisterClass RC, X86MemOperand x86memop> {
 let neverHasSideEffects = 1 in {
   def ri : Ii8<0xF0, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, i8imm:$src2),
                !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-               []>, TAXD, VEX;
+               []>, TAXD, VEX, Sched<[WriteShift]>;
   let mayLoad = 1 in
   def mi : Ii8<0xF0, MRMSrcMem, (outs RC:$dst),
                (ins x86memop:$src1, i8imm:$src2),
                !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-               []>, TAXD, VEX;
+               []>, TAXD, VEX, Sched<[WriteShiftLd]>;
 }
 }
 
@@ -866,11 +884,17 @@ multiclass bmi_shift<string asm, RegisterClass RC, X86MemOperand x86memop> {
 let neverHasSideEffects = 1 in {
   def rr : I<0xF7, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
              !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>,
-             VEX_4VOp3;
+             VEX_4VOp3, Sched<[WriteShift]>;
   let mayLoad = 1 in
   def rm : I<0xF7, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src1, RC:$src2),
              !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>,
-             VEX_4VOp3;
+             VEX_4VOp3,
+             Sched<[WriteShiftLd,
+                    // x86memop:$src1
+                    ReadDefault, ReadDefault, ReadDefault, ReadDefault,
+                    ReadDefault,
+                    // RC:$src1
+                    ReadAfterLd]>;
 }
 }
 
diff --git a/lib/Target/X86/X86InstrSystem.td b/lib/Target/X86/X86InstrSystem.td
index ea716bfd6bd8..053417ccde63 100644
--- a/lib/Target/X86/X86InstrSystem.td
+++ b/lib/Target/X86/X86InstrSystem.td
@@ -13,6 +13,7 @@
 //
 //===----------------------------------------------------------------------===//
 
+let SchedRW = [WriteSystem] in {
 let Defs = [RAX, RDX] in
   def RDTSC : I<0x31, RawFrm, (outs), (ins), "rdtsc", [(X86rdtsc)], IIC_RDTSC>,
               TB;
@@ -35,6 +36,7 @@ let Uses = [EFLAGS] in
   def INTO : I<0xce, RawFrm, (outs), (ins), "into", []>;
 def INT3 : I<0xcc, RawFrm, (outs), (ins), "int3",
               [(int_x86_int (i8 3))], IIC_INT3>;
+} // SchedRW
 
 def : Pat<(debugtrap),
           (INT3)>;
@@ -43,6 +45,7 @@ def : Pat<(debugtrap),
 // FIXME: This doesn't work because InstAlias can't match immediate constants.
 //def : InstAlias<"int\t$3", (INT3)>;
 
+let SchedRW = [WriteSystem] in {
 
 def INT : Ii8<0xcd, RawFrm, (outs), (ins i8imm:$trap), "int\t$trap",
               [(int_x86_int imm:$trap)], IIC_INT>;
@@ -65,11 +68,13 @@ def IRET16 : I<0xcf, RawFrm, (outs), (ins), "iret{w}", [], IIC_IRET>, OpSize;
 def IRET32 : I<0xcf, RawFrm, (outs), (ins), "iret{l|d}", [], IIC_IRET>;
 def IRET64 : RI<0xcf, RawFrm, (outs), (ins), "iretq", [], IIC_IRET>,
              Requires<[In64BitMode]>;
+} // SchedRW
 
 
 //===----------------------------------------------------------------------===//
 //  Input/Output Instructions.
 //
+let SchedRW = [WriteSystem] in {
 let Defs = [AL], Uses = [DX] in
 def IN8rr  : I<0xEC, RawFrm, (outs), (ins),
                "in{b}\t{%dx, %al|AL, DX}", [], IIC_IN_RR>;
@@ -113,10 +118,12 @@ def OUT32ir : Ii8<0xE7, RawFrm, (outs), (ins i8imm:$port),
 def IN8  : I<0x6C, RawFrm, (outs), (ins), "ins{b}", [], IIC_INS>;
 def IN16 : I<0x6D, RawFrm, (outs), (ins), "ins{w}", [], IIC_INS>,  OpSize;
 def IN32 : I<0x6D, RawFrm, (outs), (ins), "ins{l}", [], IIC_INS>;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // Moves to and from debug registers
 
+let SchedRW = [WriteSystem] in {
 def MOV32rd : I<0x21, MRMDestReg, (outs GR32:$dst), (ins DEBUG_REG:$src),
                 "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV_REG_DR>, TB;
 def MOV64rd : I<0x21, MRMDestReg, (outs GR64:$dst), (ins DEBUG_REG:$src),
@@ -126,10 +133,12 @@ def MOV32dr : I<0x23, MRMSrcReg, (outs DEBUG_REG:$dst), (ins GR32:$src),
                 "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV_DR_REG>, TB;
 def MOV64dr : I<0x23, MRMSrcReg, (outs DEBUG_REG:$dst), (ins GR64:$src),
                 "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV_DR_REG>, TB;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // Moves to and from control registers
 
+let SchedRW = [WriteSystem] in {
 def MOV32rc : I<0x20, MRMDestReg, (outs GR32:$dst), (ins CONTROL_REG:$src),
                 "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV_REG_CR>, TB;
 def MOV64rc : I<0x20, MRMDestReg, (outs GR64:$dst), (ins CONTROL_REG:$src),
@@ -139,6 +148,7 @@ def MOV32cr : I<0x22, MRMSrcReg, (outs CONTROL_REG:$dst), (ins GR32:$src),
                 "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV_CR_REG>, TB;
 def MOV64cr : I<0x22, MRMSrcReg, (outs CONTROL_REG:$dst), (ins GR64:$src),
                 "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV_CR_REG>, TB;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // Segment override instruction prefixes
@@ -155,6 +165,7 @@ def GS_PREFIX : I<0x65, RawFrm, (outs), (ins), "gs", []>;
 // Moves to and from segment registers.
 //
 
+let SchedRW = [WriteMove] in {
 def MOV16rs : I<0x8C, MRMDestReg, (outs GR16:$dst), (ins SEGMENT_REG:$src),
                 "mov{w}\t{$src, $dst|$dst, $src}", [], IIC_MOV_REG_SR>, OpSize;
 def MOV32rs : I<0x8C, MRMDestReg, (outs GR32:$dst), (ins SEGMENT_REG:$src),
@@ -182,10 +193,12 @@ def MOV32sm : I<0x8E, MRMSrcMem, (outs SEGMENT_REG:$dst), (ins i32mem:$src),
                 "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV_SR_MEM>;
 def MOV64sm : RI<0x8E, MRMSrcMem, (outs SEGMENT_REG:$dst), (ins i64mem:$src),
                  "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV_SR_MEM>;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // Segmentation support instructions.
 
+let SchedRW = [WriteSystem] in {
 def SWAPGS : I<0x01, MRM_F8, (outs), (ins), "swapgs", [], IIC_SWAPGS>, TB;
 
 def LAR16rm : I<0x02, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src), 
@@ -347,16 +360,18 @@ def VERWr : I<0x00, MRM5r, (outs), (ins GR16:$seg),
               "verw\t$seg", [], IIC_VERW_MEM>, TB;
 def VERWm : I<0x00, MRM5m, (outs), (ins i16mem:$seg),
               "verw\t$seg", [], IIC_VERW_REG>, TB;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // Descriptor-table support instructions
 
+let SchedRW = [WriteSystem] in {
 def SGDT16m : I<0x01, MRM0m, (outs opaque48mem:$dst), (ins),
-              "sgdtw\t$dst", [], IIC_SGDT>, TB, OpSize, Requires<[In32BitMode]>;
+              "sgdt{w}\t$dst", [], IIC_SGDT>, TB, OpSize, Requires<[In32BitMode]>;
 def SGDTm : I<0x01, MRM0m, (outs opaque48mem:$dst), (ins),
               "sgdt\t$dst", [], IIC_SGDT>, TB;
 def SIDT16m : I<0x01, MRM1m, (outs opaque48mem:$dst), (ins),
-              "sidtw\t$dst", [], IIC_SIDT>, TB, OpSize, Requires<[In32BitMode]>;
+              "sidt{w}\t$dst", [], IIC_SIDT>, TB, OpSize, Requires<[In32BitMode]>;
 def SIDTm : I<0x01, MRM1m, (outs opaque48mem:$dst), (ins),
               "sidt\t$dst", []>, TB;
 def SLDT16r : I<0x00, MRM0r, (outs GR16:$dst), (ins),
@@ -374,20 +389,22 @@ def SLDT64m : RI<0x00, MRM0m, (outs i16mem:$dst), (ins),
                  "sldt{q}\t$dst", [], IIC_SLDT>, TB;
 
 def LGDT16m : I<0x01, MRM2m, (outs), (ins opaque48mem:$src),
-              "lgdtw\t$src", [], IIC_LGDT>, TB, OpSize, Requires<[In32BitMode]>;
+              "lgdt{w}\t$src", [], IIC_LGDT>, TB, OpSize, Requires<[In32BitMode]>;
 def LGDTm : I<0x01, MRM2m, (outs), (ins opaque48mem:$src),
               "lgdt\t$src", [], IIC_LGDT>, TB;
 def LIDT16m : I<0x01, MRM3m, (outs), (ins opaque48mem:$src),
-              "lidtw\t$src", [], IIC_LIDT>, TB, OpSize, Requires<[In32BitMode]>;
+              "lidt{w}\t$src", [], IIC_LIDT>, TB, OpSize, Requires<[In32BitMode]>;
 def LIDTm : I<0x01, MRM3m, (outs), (ins opaque48mem:$src),
               "lidt\t$src", [], IIC_LIDT>, TB;
 def LLDT16r : I<0x00, MRM2r, (outs), (ins GR16:$src),
                 "lldt{w}\t$src", [], IIC_LLDT_REG>, TB;
 def LLDT16m : I<0x00, MRM2m, (outs), (ins i16mem:$src),
                 "lldt{w}\t$src", [], IIC_LLDT_MEM>, TB;
-                
+} // SchedRW
+
 //===----------------------------------------------------------------------===//
 // Specialized register support
+let SchedRW = [WriteSystem] in {
 def WRMSR : I<0x30, RawFrm, (outs), (ins), "wrmsr", [], IIC_WRMSR>, TB;
 def RDMSR : I<0x32, RawFrm, (outs), (ins), "rdmsr", [], IIC_RDMSR>, TB;
 def RDPMC : I<0x33, RawFrm, (outs), (ins), "rdpmc", [], IIC_RDPMC>, TB;
@@ -410,14 +427,18 @@ def LMSW16m : I<0x01, MRM6m, (outs), (ins i16mem:$src),
                 "lmsw{w}\t$src", [], IIC_LMSW_REG>, TB;
                 
 def CPUID : I<0xA2, RawFrm, (outs), (ins), "cpuid", [], IIC_CPUID>, TB;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // Cache instructions
+let SchedRW = [WriteSystem] in {
 def INVD : I<0x08, RawFrm, (outs), (ins), "invd", [], IIC_INVD>, TB;
 def WBINVD : I<0x09, RawFrm, (outs), (ins), "wbinvd", [], IIC_INVD>, TB;
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // XSAVE instructions
+let SchedRW = [WriteSystem] in {
 let Defs = [RDX, RAX], Uses = [RCX] in
   def XGETBV : I<0x01, MRM_D0, (outs), (ins), "xgetbv", []>, TB;
 
@@ -438,6 +459,7 @@ let Uses = [RDX, RAX] in {
   def XSAVEOPT64 : I<0xAE, MRM6m, (outs opaque512mem:$dst), (ins),
                     "xsaveoptq\t$dst", []>, TB, REX_W, Requires<[In64BitMode]>;
 }
+} // SchedRW
 
 //===----------------------------------------------------------------------===//
 // VIA PadLock crypto instructions
diff --git a/lib/Target/X86/X86InstrTSX.td b/lib/Target/X86/X86InstrTSX.td
index ad55058ede6c..363a190aa854 100644
--- a/lib/Target/X86/X86InstrTSX.td
+++ b/lib/Target/X86/X86InstrTSX.td
@@ -15,6 +15,9 @@
 //===----------------------------------------------------------------------===//
 // TSX instructions
 
+def X86xtest: SDNode<"X86ISD::XTEST", SDTypeProfile<1, 0, [SDTCisVT<0, i32>]>,
+                     [SDNPHasChain, SDNPSideEffect]>;
+
 let usesCustomInserter = 1 in
 def XBEGIN : I<0, Pseudo, (outs GR32:$dst), (ins),
                "# XBEGIN", [(set GR32:$dst, (int_x86_xbegin))]>,
@@ -22,11 +25,15 @@ def XBEGIN : I<0, Pseudo, (outs GR32:$dst), (ins),
 
 let isBranch = 1, isTerminator = 1, Defs = [EAX] in
 def XBEGIN_4 : Ii32PCRel<0xc7, MRM_F8, (outs), (ins brtarget:$dst),
-                         "xbegin\t$dst", []>;
+                         "xbegin\t$dst", []>, Requires<[HasRTM]>;
 
 def XEND : I<0x01, MRM_D5, (outs), (ins),
              "xend", [(int_x86_xend)]>, TB, Requires<[HasRTM]>;
 
+let Defs = [EFLAGS] in
+def XTEST : I<0x01, MRM_D6, (outs), (ins),
+              "xtest", [(set EFLAGS, (X86xtest))]>, TB, Requires<[HasTSX]>;
+
 def XABORT : Ii8<0xc6, MRM_F8, (outs), (ins i8imm:$imm),
                  "xabort\t$imm",
                  [(int_x86_xabort imm:$imm)]>, Requires<[HasRTM]>;
diff --git a/lib/Target/X86/X86JITInfo.cpp b/lib/Target/X86/X86JITInfo.cpp
index 764aa5d4f236..44d8cce05413 100644
--- a/lib/Target/X86/X86JITInfo.cpp
+++ b/lib/Target/X86/X86JITInfo.cpp
@@ -16,7 +16,7 @@
 #include "X86Relocations.h"
 #include "X86Subtarget.h"
 #include "X86TargetMachine.h"
-#include "llvm/Function.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/Valgrind.h"
@@ -79,7 +79,7 @@ static TargetJITInfo::JITCompilerFn JITCompilerFunction;
 # define CFI(x)
 #endif
 
-// Provide a wrapper for X86CompilationCallback2 that saves non-traditional
+// Provide a wrapper for LLVMX86CompilationCallback2 that saves non-traditional
 // callee saved registers, for the fastcc calling convention.
 extern "C" {
 #if defined(X86_64_JIT)
@@ -131,12 +131,12 @@ extern "C" {
     "subq    $32, %rsp\n"
     "movq    %rbp, %rcx\n"    // Pass prev frame and return address
     "movq    8(%rbp), %rdx\n"
-    "call    " ASMPREFIX "X86CompilationCallback2\n"
+    "call    " ASMPREFIX "LLVMX86CompilationCallback2\n"
     "addq    $32, %rsp\n"
 #else
     "movq    %rbp, %rdi\n"    // Pass prev frame and return address
     "movq    8(%rbp), %rsi\n"
-    "call    " ASMPREFIX "X86CompilationCallback2\n"
+    "call    " ASMPREFIX "LLVMX86CompilationCallback2\n"
 #endif
     // Restore all XMM arg registers
     "movaps  112(%rsp), %xmm7\n"
@@ -213,7 +213,7 @@ extern "C" {
     "movl    4(%ebp), %eax\n" // Pass prev frame and return address
     "movl    %eax, 4(%esp)\n"
     "movl    %ebp, (%esp)\n"
-    "call    " ASMPREFIX "X86CompilationCallback2\n"
+    "call    " ASMPREFIX "LLVMX86CompilationCallback2\n"
     "movl    %ebp, %esp\n"    // Restore ESP
     CFI(".cfi_def_cfa_register %esp\n")
     "subl    $12, %esp\n"
@@ -269,7 +269,7 @@ extern "C" {
     "movl    4(%ebp), %eax\n" // Pass prev frame and return address
     "movl    %eax, 4(%esp)\n"
     "movl    %ebp, (%esp)\n"
-    "call    " ASMPREFIX "X86CompilationCallback2\n"
+    "call    " ASMPREFIX "LLVMX86CompilationCallback2\n"
     "addl    $16, %esp\n"
     "movaps  48(%esp), %xmm3\n"
     CFI(".cfi_restore %xmm3\n")
@@ -300,10 +300,7 @@ extern "C" {
     SIZE(X86CompilationCallback_SSE)
   );
 # else
-  // the following function is called only from this translation unit,
-  // unless we are under 64bit Windows with MSC, where there is
-  // no support for inline assembly
-  static void X86CompilationCallback2(intptr_t *StackPtr, intptr_t RetAddr);
+  void LLVMX86CompilationCallback2(intptr_t *StackPtr, intptr_t RetAddr);
 
   _declspec(naked) void X86CompilationCallback(void) {
     __asm {
@@ -317,7 +314,7 @@ extern "C" {
       mov   eax, dword ptr [ebp+4]
       mov   dword ptr [esp+4], eax
       mov   dword ptr [esp], ebp
-      call  X86CompilationCallback2
+      call  LLVMX86CompilationCallback2
       mov   esp, ebp
       sub   esp, 12
       pop   ecx
@@ -337,20 +334,17 @@ extern "C" {
 #endif
 }
 
-/// X86CompilationCallback2 - This is the target-specific function invoked by the
+/// This is the target-specific function invoked by the
 /// function stub when we did not know the real target of a call.  This function
 /// must locate the start of the stub or call site and pass it into the JIT
 /// compiler function.
 extern "C" {
-#if !(defined (X86_64_JIT) && defined(_MSC_VER))
- // the following function is called only from this translation unit,
- // unless we are under 64bit Windows with MSC, where there is
- // no support for inline assembly
-static
-#endif
-void LLVM_ATTRIBUTE_USED
-X86CompilationCallback2(intptr_t *StackPtr, intptr_t RetAddr) {
+LLVM_LIBRARY_VISIBILITY void LLVMX86CompilationCallback2(intptr_t *StackPtr,
+                                                         intptr_t RetAddr) {
   intptr_t *RetAddrLoc = &StackPtr[1];
+  // We are reading raw stack data here. Tell MemorySanitizer that it is
+  // sufficiently initialized.
+  __msan_unpoison(RetAddrLoc, sizeof(*RetAddrLoc));
   assert(*RetAddrLoc == RetAddr &&
          "Could not find return address on the stack!");
 
@@ -517,7 +511,7 @@ void *X86JITInfo::emitFunctionStub(const Function* F, void *Target,
 
   // This used to use 0xCD, but that value is used by JITMemoryManager to
   // initialize the buffer with garbage, which means it may follow a
-  // noreturn function call, confusing X86CompilationCallback2.  PR 4929.
+  // noreturn function call, confusing LLVMX86CompilationCallback2.  PR 4929.
   JCE.emitByte(0xCE);   // Interrupt - Just a marker identifying the stub!
   return Result;
 }
diff --git a/lib/Target/X86/X86JITInfo.h b/lib/Target/X86/X86JITInfo.h
index d7c08dfb0fdf..f916327378a9 100644
--- a/lib/Target/X86/X86JITInfo.h
+++ b/lib/Target/X86/X86JITInfo.h
@@ -14,8 +14,8 @@
 #ifndef X86JITINFO_H
 #define X86JITINFO_H
 
-#include "llvm/Function.h"
 #include "llvm/CodeGen/JITCodeEmitter.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Target/TargetJITInfo.h"
 
 namespace llvm {
diff --git a/lib/Target/X86/X86MCInstLower.cpp b/lib/Target/X86/X86MCInstLower.cpp
index cfd68f74b7b2..a8a9fd8accde 100644
--- a/lib/Target/X86/X86MCInstLower.cpp
+++ b/lib/Target/X86/X86MCInstLower.cpp
@@ -13,19 +13,20 @@
 //===----------------------------------------------------------------------===//
 
 #include "X86AsmPrinter.h"
-#include "X86COFFMachineModuleInfo.h"
 #include "InstPrinter/X86ATTInstPrinter.h"
-#include "llvm/Type.h"
+#include "X86COFFMachineModuleInfo.h"
+#include "llvm/ADT/SmallString.h"
 #include "llvm/CodeGen/MachineModuleInfoImpls.h"
+#include "llvm/IR/Type.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCInstBuilder.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/Target/Mangler.h"
 #include "llvm/Support/FormattedStream.h"
-#include "llvm/ADT/SmallString.h"
+#include "llvm/Target/Mangler.h"
 using namespace llvm;
 
 namespace {
@@ -238,7 +239,8 @@ static void lower_lea64_32mem(MCInst *MI, unsigned OpNo) {
     if (!MI->getOperand(OpNo+i).isReg()) continue;
 
     unsigned Reg = MI->getOperand(OpNo+i).getReg();
-    if (Reg == 0) continue;
+    // LEAs can use RIP-relative addressing, and RIP has no sub/super register.
+    if (Reg == 0 || Reg == X86::RIP) continue;
 
     MI->getOperand(OpNo+i).setReg(getX86SubSuperRegister(Reg, MVT::i64));
   }
@@ -405,6 +407,57 @@ ReSimplify:
     LowerUnaryToTwoAddr(OutMI, X86::XOR32rr); // MOV32r0 -> XOR32rr
     break;
 
+  // Commute operands to get a smaller encoding by using VEX.R instead of VEX.B
+  // if one of the registers is extended, but other isn't.
+  case X86::VMOVAPDrr:
+  case X86::VMOVAPDYrr:
+  case X86::VMOVAPSrr:
+  case X86::VMOVAPSYrr:
+  case X86::VMOVDQArr:
+  case X86::VMOVDQAYrr:
+  case X86::VMOVDQUrr:
+  case X86::VMOVDQUYrr:
+  case X86::VMOVUPDrr:
+  case X86::VMOVUPDYrr:
+  case X86::VMOVUPSrr:
+  case X86::VMOVUPSYrr: {
+    if (!X86II::isX86_64ExtendedReg(OutMI.getOperand(0).getReg()) &&
+        X86II::isX86_64ExtendedReg(OutMI.getOperand(1).getReg())) {
+      unsigned NewOpc;
+      switch (OutMI.getOpcode()) {
+      default: llvm_unreachable("Invalid opcode");
+      case X86::VMOVAPDrr:  NewOpc = X86::VMOVAPDrr_REV;  break;
+      case X86::VMOVAPDYrr: NewOpc = X86::VMOVAPDYrr_REV; break;
+      case X86::VMOVAPSrr:  NewOpc = X86::VMOVAPSrr_REV;  break;
+      case X86::VMOVAPSYrr: NewOpc = X86::VMOVAPSYrr_REV; break;
+      case X86::VMOVDQArr:  NewOpc = X86::VMOVDQArr_REV;  break;
+      case X86::VMOVDQAYrr: NewOpc = X86::VMOVDQAYrr_REV; break;
+      case X86::VMOVDQUrr:  NewOpc = X86::VMOVDQUrr_REV;  break;
+      case X86::VMOVDQUYrr: NewOpc = X86::VMOVDQUYrr_REV; break;
+      case X86::VMOVUPDrr:  NewOpc = X86::VMOVUPDrr_REV;  break;
+      case X86::VMOVUPDYrr: NewOpc = X86::VMOVUPDYrr_REV; break;
+      case X86::VMOVUPSrr:  NewOpc = X86::VMOVUPSrr_REV;  break;
+      case X86::VMOVUPSYrr: NewOpc = X86::VMOVUPSYrr_REV; break;
+      }
+      OutMI.setOpcode(NewOpc);
+    }
+    break;
+  }
+  case X86::VMOVSDrr:
+  case X86::VMOVSSrr: {
+    if (!X86II::isX86_64ExtendedReg(OutMI.getOperand(0).getReg()) &&
+        X86II::isX86_64ExtendedReg(OutMI.getOperand(2).getReg())) {
+      unsigned NewOpc;
+      switch (OutMI.getOpcode()) {
+      default: llvm_unreachable("Invalid opcode");
+      case X86::VMOVSDrr:   NewOpc = X86::VMOVSDrr_REV;   break;
+      case X86::VMOVSSrr:   NewOpc = X86::VMOVSSrr_REV;   break;
+      }
+      OutMI.setOpcode(NewOpc);
+    }
+    break;
+  }
+
   // TAILJMPr64, CALL64r, CALL64pcrel32 - These instructions have register
   // inputs modeled as normal uses instead of implicit uses.  As such, truncate
   // off all but the first operand (the callee).  FIXME: Change isel.
@@ -549,18 +602,14 @@ ReSimplify:
     OutMI.setOpcode(X86::RET);
     break;
 
-  case X86::MORESTACK_RET_RESTORE_R10: {
-    MCInst retInst;
-
+  case X86::MORESTACK_RET_RESTORE_R10:
     OutMI.setOpcode(X86::MOV64rr);
     OutMI.addOperand(MCOperand::CreateReg(X86::R10));
     OutMI.addOperand(MCOperand::CreateReg(X86::RAX));
 
-    retInst.setOpcode(X86::RET);
-    AsmPrinter.OutStreamer.EmitInstruction(retInst);
+    AsmPrinter.OutStreamer.EmitInstruction(MCInstBuilder(X86::RET));
     break;
   }
-  }
 }
 
 static void LowerTlsAddr(MCStreamer &OutStreamer,
@@ -574,11 +623,8 @@ static void LowerTlsAddr(MCStreamer &OutStreamer,
 
   MCContext &context = OutStreamer.getContext();
 
-  if (needsPadding) {
-    MCInst prefix;
-    prefix.setOpcode(X86::DATA16_PREFIX);
-    OutStreamer.EmitInstruction(prefix);
-  }
+  if (needsPadding)
+    OutStreamer.EmitInstruction(MCInstBuilder(X86::DATA16_PREFIX));
 
   MCSymbolRefExpr::VariantKind SRVK;
   switch (MI.getOpcode()) {
@@ -628,20 +674,11 @@ static void LowerTlsAddr(MCStreamer &OutStreamer,
   OutStreamer.EmitInstruction(LEA);
 
   if (needsPadding) {
-    MCInst prefix;
-    prefix.setOpcode(X86::DATA16_PREFIX);
-    OutStreamer.EmitInstruction(prefix);
-    prefix.setOpcode(X86::DATA16_PREFIX);
-    OutStreamer.EmitInstruction(prefix);
-    prefix.setOpcode(X86::REX64_PREFIX);
-    OutStreamer.EmitInstruction(prefix);
+    OutStreamer.EmitInstruction(MCInstBuilder(X86::DATA16_PREFIX));
+    OutStreamer.EmitInstruction(MCInstBuilder(X86::DATA16_PREFIX));
+    OutStreamer.EmitInstruction(MCInstBuilder(X86::REX64_PREFIX));
   }
 
-  MCInst call;
-  if (is64Bits)
-    call.setOpcode(X86::CALL64pcrel32);
-  else
-    call.setOpcode(X86::CALLpcrel32);
   StringRef name = is64Bits ? "__tls_get_addr" : "___tls_get_addr";
   MCSymbol *tlsGetAddr = context.GetOrCreateSymbol(name);
   const MCSymbolRefExpr *tlsRef =
@@ -649,8 +686,9 @@ static void LowerTlsAddr(MCStreamer &OutStreamer,
                             MCSymbolRefExpr::VK_PLT,
                             context);
 
-  call.addOperand(MCOperand::CreateExpr(tlsRef));
-  OutStreamer.EmitInstruction(call);
+  OutStreamer.EmitInstruction(MCInstBuilder(is64Bits ? X86::CALL64pcrel32
+                                                     : X86::CALLpcrel32)
+    .addExpr(tlsRef));
 }
 
 void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
@@ -694,7 +732,6 @@ void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
     return LowerTlsAddr(OutStreamer, MCInstLowering, *MI);
 
   case X86::MOVPC32r: {
-    MCInst TmpInst;
     // This is a pseudo op for a two instruction sequence with a label, which
     // looks like:
     //     call "L1$pb"
@@ -703,20 +740,17 @@ void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
 
     // Emit the call.
     MCSymbol *PICBase = MF->getPICBaseSymbol();
-    TmpInst.setOpcode(X86::CALLpcrel32);
     // FIXME: We would like an efficient form for this, so we don't have to do a
     // lot of extra uniquing.
-    TmpInst.addOperand(MCOperand::CreateExpr(MCSymbolRefExpr::Create(PICBase,
-                                                                 OutContext)));
-    OutStreamer.EmitInstruction(TmpInst);
+    OutStreamer.EmitInstruction(MCInstBuilder(X86::CALLpcrel32)
+      .addExpr(MCSymbolRefExpr::Create(PICBase, OutContext)));
 
     // Emit the label.
     OutStreamer.EmitLabel(PICBase);
 
     // popl $reg
-    TmpInst.setOpcode(X86::POP32r);
-    TmpInst.getOperand(0) = MCOperand::CreateReg(MI->getOperand(0).getReg());
-    OutStreamer.EmitInstruction(TmpInst);
+    OutStreamer.EmitInstruction(MCInstBuilder(X86::POP32r)
+      .addReg(MI->getOperand(0).getReg()));
     return;
   }
 
@@ -746,12 +780,10 @@ void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
     DotExpr = MCBinaryExpr::CreateAdd(MCSymbolRefExpr::Create(OpSym,OutContext),
                                       DotExpr, OutContext);
 
-    MCInst TmpInst;
-    TmpInst.setOpcode(X86::ADD32ri);
-    TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
-    TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(1).getReg()));
-    TmpInst.addOperand(MCOperand::CreateExpr(DotExpr));
-    OutStreamer.EmitInstruction(TmpInst);
+    OutStreamer.EmitInstruction(MCInstBuilder(X86::ADD32ri)
+      .addReg(MI->getOperand(0).getReg())
+      .addReg(MI->getOperand(1).getReg())
+      .addExpr(DotExpr));
     return;
   }
   }
diff --git a/lib/Target/X86/X86PadShortFunction.cpp b/lib/Target/X86/X86PadShortFunction.cpp
new file mode 100644
index 000000000000..83e75ea994ca
--- /dev/null
+++ b/lib/Target/X86/X86PadShortFunction.cpp
@@ -0,0 +1,212 @@
+//===-------- X86PadShortFunction.cpp - pad short functions -----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the pass which will pad short functions to prevent
+// a stall if a function returns before the return address is ready. This
+// is needed for some Intel Atom processors.
+//
+//===----------------------------------------------------------------------===//
+
+#include <algorithm>
+
+#define DEBUG_TYPE "x86-pad-short-functions"
+#include "X86.h"
+#include "X86InstrInfo.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetInstrInfo.h"
+
+using namespace llvm;
+
+STATISTIC(NumBBsPadded, "Number of basic blocks padded");
+
+namespace {
+  struct VisitedBBInfo {
+    // HasReturn - Whether the BB contains a return instruction
+    bool HasReturn;
+
+    // Cycles - Number of cycles until return if HasReturn is true, otherwise
+    // number of cycles until end of the BB
+    unsigned int Cycles;
+
+    VisitedBBInfo() : HasReturn(false), Cycles(0) {}
+    VisitedBBInfo(bool HasReturn, unsigned int Cycles)
+      : HasReturn(HasReturn), Cycles(Cycles) {}
+  };
+
+  struct PadShortFunc : public MachineFunctionPass {
+    static char ID;
+    PadShortFunc() : MachineFunctionPass(ID)
+                   , Threshold(4), TM(0), TII(0) {}
+
+    virtual bool runOnMachineFunction(MachineFunction &MF);
+
+    virtual const char *getPassName() const {
+      return "X86 Atom pad short functions";
+    }
+
+  private:
+    void findReturns(MachineBasicBlock *MBB,
+                     unsigned int Cycles = 0);
+
+    bool cyclesUntilReturn(MachineBasicBlock *MBB,
+                           unsigned int &Cycles);
+
+    void addPadding(MachineBasicBlock *MBB,
+                    MachineBasicBlock::iterator &MBBI,
+                    unsigned int NOOPsToAdd);
+
+    const unsigned int Threshold;
+
+    // ReturnBBs - Maps basic blocks that return to the minimum number of
+    // cycles until the return, starting from the entry block.
+    DenseMap<MachineBasicBlock*, unsigned int> ReturnBBs;
+
+    // VisitedBBs - Cache of previously visited BBs.
+    DenseMap<MachineBasicBlock*, VisitedBBInfo> VisitedBBs;
+
+    const TargetMachine *TM;
+    const TargetInstrInfo *TII;
+  };
+
+  char PadShortFunc::ID = 0;
+}
+
+FunctionPass *llvm::createX86PadShortFunctions() {
+  return new PadShortFunc();
+}
+
+/// runOnMachineFunction - Loop over all of the basic blocks, inserting
+/// NOOP instructions before early exits.
+bool PadShortFunc::runOnMachineFunction(MachineFunction &MF) {
+  const AttributeSet &FnAttrs = MF.getFunction()->getAttributes();
+  if (FnAttrs.hasAttribute(AttributeSet::FunctionIndex,
+                           Attribute::OptimizeForSize) ||
+      FnAttrs.hasAttribute(AttributeSet::FunctionIndex,
+                           Attribute::MinSize)) {
+    return false;
+  }
+
+  TM = &MF.getTarget();
+  TII = TM->getInstrInfo();
+
+  // Search through basic blocks and mark the ones that have early returns
+  ReturnBBs.clear();
+  VisitedBBs.clear();
+  findReturns(MF.begin());
+
+  bool MadeChange = false;
+
+  MachineBasicBlock *MBB;
+  unsigned int Cycles = 0;
+
+  // Pad the identified basic blocks with NOOPs
+  for (DenseMap<MachineBasicBlock*, unsigned int>::iterator I = ReturnBBs.begin();
+       I != ReturnBBs.end(); ++I) {
+    MBB = I->first;
+    Cycles = I->second;
+
+    if (Cycles < Threshold) {
+      // BB ends in a return. Skip over any DBG_VALUE instructions
+      // trailing the terminator.
+      assert(MBB->size() > 0 &&
+             "Basic block should contain at least a RET but is empty");
+      MachineBasicBlock::iterator ReturnLoc = --MBB->end();
+
+      while (ReturnLoc->isDebugValue())
+        --ReturnLoc;
+      assert(ReturnLoc->isReturn() && !ReturnLoc->isCall() &&
+             "Basic block does not end with RET");
+
+      addPadding(MBB, ReturnLoc, Threshold - Cycles);
+      NumBBsPadded++;
+      MadeChange = true;
+    }
+  }
+
+  return MadeChange;
+}
+
+/// findReturn - Starting at MBB, follow control flow and add all
+/// basic blocks that contain a return to ReturnBBs.
+void PadShortFunc::findReturns(MachineBasicBlock *MBB, unsigned int Cycles) {
+  // If this BB has a return, note how many cycles it takes to get there.
+  bool hasReturn = cyclesUntilReturn(MBB, Cycles);
+  if (Cycles >= Threshold)
+    return;
+
+  if (hasReturn) {
+    ReturnBBs[MBB] = std::max(ReturnBBs[MBB], Cycles);
+    return;
+  }
+
+  // Follow branches in BB and look for returns
+  for (MachineBasicBlock::succ_iterator I = MBB->succ_begin();
+       I != MBB->succ_end(); ++I) {
+    if (*I == MBB)
+      continue;
+    findReturns(*I, Cycles);
+  }
+}
+
+/// cyclesUntilReturn - return true if the MBB has a return instruction,
+/// and return false otherwise.
+/// Cycles will be incremented by the number of cycles taken to reach the
+/// return or the end of the BB, whichever occurs first.
+bool PadShortFunc::cyclesUntilReturn(MachineBasicBlock *MBB,
+                                     unsigned int &Cycles) {
+  // Return cached result if BB was previously visited
+  DenseMap<MachineBasicBlock*, VisitedBBInfo>::iterator it
+    = VisitedBBs.find(MBB);
+  if (it != VisitedBBs.end()) {
+    VisitedBBInfo BBInfo = it->second;
+    Cycles += BBInfo.Cycles;
+    return BBInfo.HasReturn;
+  }
+
+  unsigned int CyclesToEnd = 0;
+
+  for (MachineBasicBlock::iterator MBBI = MBB->begin();
+        MBBI != MBB->end(); ++MBBI) {
+    MachineInstr *MI = MBBI;
+    // Mark basic blocks with a return instruction. Calls to other
+    // functions do not count because the called function will be padded,
+    // if necessary.
+    if (MI->isReturn() && !MI->isCall()) {
+      VisitedBBs[MBB] = VisitedBBInfo(true, CyclesToEnd);
+      Cycles += CyclesToEnd;
+      return true;
+    }
+
+    CyclesToEnd += TII->getInstrLatency(TM->getInstrItineraryData(), MI);
+  }
+
+  VisitedBBs[MBB] = VisitedBBInfo(false, CyclesToEnd);
+  Cycles += CyclesToEnd;
+  return false;
+}
+
+/// addPadding - Add the given number of NOOP instructions to the function
+/// just prior to the return at MBBI
+void PadShortFunc::addPadding(MachineBasicBlock *MBB,
+                              MachineBasicBlock::iterator &MBBI,
+                              unsigned int NOOPsToAdd) {
+  DebugLoc DL = MBBI->getDebugLoc();
+
+  while (NOOPsToAdd-- > 0) {
+    BuildMI(*MBB, MBBI, DL, TII->get(X86::NOOP));
+    BuildMI(*MBB, MBBI, DL, TII->get(X86::NOOP));
+  }
+}
diff --git a/lib/Target/X86/X86RegisterInfo.cpp b/lib/Target/X86/X86RegisterInfo.cpp
index 73ac7477427f..16886e432d19 100644
--- a/lib/Target/X86/X86RegisterInfo.cpp
+++ b/lib/Target/X86/X86RegisterInfo.cpp
@@ -19,25 +19,25 @@
 #include "X86MachineFunctionInfo.h"
 #include "X86Subtarget.h"
 #include "X86TargetMachine.h"
-#include "llvm/Constants.h"
-#include "llvm/Function.h"
-#include "llvm/Type.h"
-#include "llvm/CodeGen/ValueTypes.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Type.h"
 #include "llvm/MC/MCAsmInfo.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Target/TargetFrameLowering.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetOptions.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/STLExtras.h"
-#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/CommandLine.h"
 
 #define GET_REGINFO_TARGET_DESC
 #include "X86GenRegisterInfo.inc"
@@ -50,16 +50,18 @@ ForceStackAlign("force-align-stack",
                            " needed for the function."),
                  cl::init(false), cl::Hidden);
 
-cl::opt<bool>
+static cl::opt<bool>
 EnableBasePointer("x86-use-base-pointer", cl::Hidden, cl::init(true),
           cl::desc("Enable use of a base pointer for complex stack frames"));
 
 X86RegisterInfo::X86RegisterInfo(X86TargetMachine &tm,
                                  const TargetInstrInfo &tii)
-  : X86GenRegisterInfo(tm.getSubtarget<X86Subtarget>().is64Bit()
-                         ? X86::RIP : X86::EIP,
+  : X86GenRegisterInfo((tm.getSubtarget<X86Subtarget>().is64Bit()
+                         ? X86::RIP : X86::EIP),
                        X86_MC::getDwarfRegFlavour(tm.getTargetTriple(), false),
-                       X86_MC::getDwarfRegFlavour(tm.getTargetTriple(), true)),
+                       X86_MC::getDwarfRegFlavour(tm.getTargetTriple(), true),
+                       (tm.getSubtarget<X86Subtarget>().is64Bit()
+                         ? X86::RIP : X86::EIP)),
                        TM(tm), TII(tii) {
   X86_MC::InitLLVM2SEHRegisterMapping(this);
 
@@ -175,21 +177,27 @@ X86RegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC) const{
 const TargetRegisterClass *
 X86RegisterInfo::getPointerRegClass(const MachineFunction &MF, unsigned Kind)
                                                                          const {
+  const X86Subtarget &Subtarget = TM.getSubtarget<X86Subtarget>();
   switch (Kind) {
   default: llvm_unreachable("Unexpected Kind in getPointerRegClass!");
   case 0: // Normal GPRs.
-    if (TM.getSubtarget<X86Subtarget>().is64Bit())
+    if (Subtarget.isTarget64BitLP64())
       return &X86::GR64RegClass;
     return &X86::GR32RegClass;
   case 1: // Normal GPRs except the stack pointer (for encoding reasons).
-    if (TM.getSubtarget<X86Subtarget>().is64Bit())
+    if (Subtarget.isTarget64BitLP64())
       return &X86::GR64_NOSPRegClass;
     return &X86::GR32_NOSPRegClass;
   case 2: // Available for tailcall (not callee-saved GPRs).
-    if (TM.getSubtarget<X86Subtarget>().isTargetWin64())
+    if (Subtarget.isTargetWin64())
       return &X86::GR64_TCW64RegClass;
-    if (TM.getSubtarget<X86Subtarget>().is64Bit())
+    else if (Subtarget.is64Bit())
       return &X86::GR64_TCRegClass;
+
+    const Function *F = MF.getFunction();
+    bool hasHipeCC = (F ? F->getCallingConv() == CallingConv::HiPE : false);
+    if (hasHipeCC)
+      return &X86::GR32RegClass;
     return &X86::GR32_TCRegClass;
   }
 }
@@ -227,36 +235,40 @@ X86RegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC,
 
 const uint16_t *
 X86RegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
-  bool callsEHReturn = false;
-  bool ghcCall = false;
-  bool oclBiCall = false;
-  bool HasAVX = TM.getSubtarget<X86Subtarget>().hasAVX();
-
-  if (MF) {
-    callsEHReturn = MF->getMMI().callsEHReturn();
-    const Function *F = MF->getFunction();
-    ghcCall = (F ? F->getCallingConv() == CallingConv::GHC : false);
-    oclBiCall = (F ? F->getCallingConv() == CallingConv::Intel_OCL_BI : false);
-  }
-
-  if (ghcCall)
+  switch (MF->getFunction()->getCallingConv()) {
+  case CallingConv::GHC:
+  case CallingConv::HiPE:
     return CSR_NoRegs_SaveList;
-  if (oclBiCall) {
+
+  case CallingConv::Intel_OCL_BI: {
+    bool HasAVX = TM.getSubtarget<X86Subtarget>().hasAVX();
     if (HasAVX && IsWin64)
-        return CSR_Win64_Intel_OCL_BI_AVX_SaveList;
+      return CSR_Win64_Intel_OCL_BI_AVX_SaveList;
     if (HasAVX && Is64Bit)
-        return CSR_64_Intel_OCL_BI_AVX_SaveList;
+      return CSR_64_Intel_OCL_BI_AVX_SaveList;
     if (!HasAVX && !IsWin64 && Is64Bit)
-        return CSR_64_Intel_OCL_BI_SaveList;
+      return CSR_64_Intel_OCL_BI_SaveList;
+    break;
+  }
+
+  case CallingConv::Cold:
+    if (Is64Bit)
+      return CSR_MostRegs_64_SaveList;
+    break;
+
+  default:
+    break;
   }
+
+  bool CallsEHReturn = MF->getMMI().callsEHReturn();
   if (Is64Bit) {
     if (IsWin64)
       return CSR_Win64_SaveList;
-    if (callsEHReturn)
+    if (CallsEHReturn)
       return CSR_64EHRet_SaveList;
     return CSR_64_SaveList;
   }
-  if (callsEHReturn)
+  if (CallsEHReturn)
     return CSR_32EHRet_SaveList;
   return CSR_32_SaveList;
 }
@@ -273,10 +285,12 @@ X86RegisterInfo::getCallPreservedMask(CallingConv::ID CC) const {
     if (!HasAVX && !IsWin64 && Is64Bit)
       return CSR_64_Intel_OCL_BI_RegMask;
   }
-  if (CC == CallingConv::GHC)
+  if (CC == CallingConv::GHC || CC == CallingConv::HiPE)
     return CSR_NoRegs_RegMask;
   if (!Is64Bit)
     return CSR_32_RegMask;
+  if (CC == CallingConv::Cold)
+    return CSR_MostRegs_64_RegMask;
   if (IsWin64)
     return CSR_Win64_RegMask;
   return CSR_64_RegMask;
@@ -380,7 +394,13 @@ bool X86RegisterInfo::hasBasePointer(const MachineFunction &MF) const {
 
    // When we need stack realignment and there are dynamic allocas, we can't
    // reference off of the stack pointer, so we reserve a base pointer.
-   if (needsStackRealignment(MF) && MFI->hasVarSizedObjects())
+   //
+   // This is also true if the function contain MS-style inline assembly.  We
+   // do this because if any stack changes occur in the inline assembly, e.g.,
+   // "pusha", then any C local variable or C argument references in the
+   // inline assembly will be wrong because the SP is not properly tracked.
+   if ((needsStackRealignment(MF) && MFI->hasVarSizedObjects()) ||
+       MF.hasMSInlineAsm())
      return true;
 
    return false;
@@ -410,7 +430,8 @@ bool X86RegisterInfo::needsStackRealignment(const MachineFunction &MF) const {
   unsigned StackAlign = TM.getFrameLowering()->getStackAlignment();
   bool requiresRealignment =
     ((MFI->getMaxAlignment() > StackAlign) ||
-     F->getFnAttributes().hasAttribute(Attributes::StackAlignment));
+     F->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
+                                     Attribute::StackAlignment));
 
   // If we've requested that we force align the stack do so now.
   if (ForceStackAlign)
@@ -430,123 +451,16 @@ bool X86RegisterInfo::hasReservedSpillSlot(const MachineFunction &MF,
   return false;
 }
 
-static unsigned getSUBriOpcode(unsigned is64Bit, int64_t Imm) {
-  if (is64Bit) {
-    if (isInt<8>(Imm))
-      return X86::SUB64ri8;
-    return X86::SUB64ri32;
-  } else {
-    if (isInt<8>(Imm))
-      return X86::SUB32ri8;
-    return X86::SUB32ri;
-  }
-}
-
-static unsigned getADDriOpcode(unsigned is64Bit, int64_t Imm) {
-  if (is64Bit) {
-    if (isInt<8>(Imm))
-      return X86::ADD64ri8;
-    return X86::ADD64ri32;
-  } else {
-    if (isInt<8>(Imm))
-      return X86::ADD32ri8;
-    return X86::ADD32ri;
-  }
-}
-
-void X86RegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-  bool reseveCallFrame = TFI->hasReservedCallFrame(MF);
-  int Opcode = I->getOpcode();
-  bool isDestroy = Opcode == TII.getCallFrameDestroyOpcode();
-  DebugLoc DL = I->getDebugLoc();
-  uint64_t Amount = !reseveCallFrame ? I->getOperand(0).getImm() : 0;
-  uint64_t CalleeAmt = isDestroy ? I->getOperand(1).getImm() : 0;
-  I = MBB.erase(I);
-
-  if (!reseveCallFrame) {
-    // If the stack pointer can be changed after prologue, turn the
-    // adjcallstackup instruction into a 'sub ESP, <amt>' and the
-    // adjcallstackdown instruction into 'add ESP, <amt>'
-    // TODO: consider using push / pop instead of sub + store / add
-    if (Amount == 0)
-      return;
-
-    // We need to keep the stack aligned properly.  To do this, we round the
-    // amount of space needed for the outgoing arguments up to the next
-    // alignment boundary.
-    unsigned StackAlign = TM.getFrameLowering()->getStackAlignment();
-    Amount = (Amount + StackAlign - 1) / StackAlign * StackAlign;
-
-    MachineInstr *New = 0;
-    if (Opcode == TII.getCallFrameSetupOpcode()) {
-      New = BuildMI(MF, DL, TII.get(getSUBriOpcode(Is64Bit, Amount)),
-                    StackPtr)
-        .addReg(StackPtr)
-        .addImm(Amount);
-    } else {
-      assert(Opcode == TII.getCallFrameDestroyOpcode());
-
-      // Factor out the amount the callee already popped.
-      Amount -= CalleeAmt;
-
-      if (Amount) {
-        unsigned Opc = getADDriOpcode(Is64Bit, Amount);
-        New = BuildMI(MF, DL, TII.get(Opc), StackPtr)
-          .addReg(StackPtr).addImm(Amount);
-      }
-    }
-
-    if (New) {
-      // The EFLAGS implicit def is dead.
-      New->getOperand(3).setIsDead();
-
-      // Replace the pseudo instruction with a new instruction.
-      MBB.insert(I, New);
-    }
-
-    return;
-  }
-
-  if (Opcode == TII.getCallFrameDestroyOpcode() && CalleeAmt) {
-    // If we are performing frame pointer elimination and if the callee pops
-    // something off the stack pointer, add it back.  We do this until we have
-    // more advanced stack pointer tracking ability.
-    unsigned Opc = getSUBriOpcode(Is64Bit, CalleeAmt);
-    MachineInstr *New = BuildMI(MF, DL, TII.get(Opc), StackPtr)
-      .addReg(StackPtr).addImm(CalleeAmt);
-
-    // The EFLAGS implicit def is dead.
-    New->getOperand(3).setIsDead();
-
-    // We are not tracking the stack pointer adjustment by the callee, so make
-    // sure we restore the stack pointer immediately after the call, there may
-    // be spill code inserted between the CALL and ADJCALLSTACKUP instructions.
-    MachineBasicBlock::iterator B = MBB.begin();
-    while (I != B && !llvm::prior(I)->isCall())
-      --I;
-    MBB.insert(I, New);
-  }
-}
-
 void
 X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
-                                     int SPAdj, RegScavenger *RS) const {
+                                     int SPAdj, unsigned FIOperandNum,
+                                     RegScavenger *RS) const {
   assert(SPAdj == 0 && "Unexpected");
 
-  unsigned i = 0;
   MachineInstr &MI = *II;
   MachineFunction &MF = *MI.getParent()->getParent();
   const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
-  }
-
-  int FrameIndex = MI.getOperand(i).getIndex();
+  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
   unsigned BasePtr;
 
   unsigned Opc = MI.getOpcode();
@@ -562,7 +476,7 @@ X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
 
   // This must be part of a four operand memory reference.  Replace the
   // FrameIndex with base register with EBP.  Add an offset to the offset.
-  MI.getOperand(i).ChangeToRegister(BasePtr, false);
+  MI.getOperand(FIOperandNum).ChangeToRegister(BasePtr, false);
 
   // Now add the frame object offset to the offset from EBP.
   int FIOffset;
@@ -573,17 +487,18 @@ X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   } else
     FIOffset = TFI->getFrameIndexOffset(MF, FrameIndex);
 
-  if (MI.getOperand(i+3).isImm()) {
+  if (MI.getOperand(FIOperandNum+3).isImm()) {
     // Offset is a 32-bit integer.
-    int Imm = (int)(MI.getOperand(i + 3).getImm());
+    int Imm = (int)(MI.getOperand(FIOperandNum + 3).getImm());
     int Offset = FIOffset + Imm;
     assert((!Is64Bit || isInt<32>((long long)FIOffset + Imm)) &&
            "Requesting 64-bit offset in 32-bit immediate!");
-    MI.getOperand(i + 3).ChangeToImmediate(Offset);
+    MI.getOperand(FIOperandNum + 3).ChangeToImmediate(Offset);
   } else {
     // Offset is symbolic. This is extremely rare.
-    uint64_t Offset = FIOffset + (uint64_t)MI.getOperand(i+3).getOffset();
-    MI.getOperand(i+3).setOffset(Offset);
+    uint64_t Offset = FIOffset +
+      (uint64_t)MI.getOperand(FIOperandNum+3).getOffset();
+    MI.getOperand(FIOperandNum + 3).setOffset(Offset);
   }
 }
 
@@ -608,7 +523,15 @@ unsigned getX86SubSuperRegister(unsigned Reg, MVT::SimpleValueType VT,
   case MVT::i8:
     if (High) {
       switch (Reg) {
-      default: return getX86SubSuperRegister(Reg, MVT::i64, High);
+      default: return getX86SubSuperRegister(Reg, MVT::i64);
+      case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
+        return X86::SI;
+      case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
+        return X86::DI;
+      case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
+        return X86::BP;
+      case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
+        return X86::SP;
       case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
         return X86::AH;
       case X86::DH: case X86::DL: case X86::DX: case X86::EDX: case X86::RDX:
@@ -728,22 +651,6 @@ unsigned getX86SubSuperRegister(unsigned Reg, MVT::SimpleValueType VT,
       return X86::R15D;
     }
   case MVT::i64:
-    // For 64-bit mode if we've requested a "high" register and the
-    // Q or r constraints we want one of these high registers or
-    // just the register name otherwise.
-    if (High) {
-      switch (Reg) {
-      case X86::SIL: case X86::SI: case X86::ESI: case X86::RSI:
-        return X86::SI;
-      case X86::DIL: case X86::DI: case X86::EDI: case X86::RDI:
-        return X86::DI;
-      case X86::BPL: case X86::BP: case X86::EBP: case X86::RBP:
-        return X86::BP;
-      case X86::SPL: case X86::SP: case X86::ESP: case X86::RSP:
-        return X86::SP;
-      // Fallthrough.
-      }
-    }
     switch (Reg) {
     default: llvm_unreachable("Unexpected register");
     case X86::AH: case X86::AL: case X86::AX: case X86::EAX: case X86::RAX:
@@ -782,46 +689,3 @@ unsigned getX86SubSuperRegister(unsigned Reg, MVT::SimpleValueType VT,
   }
 }
 }
-
-namespace {
-  struct MSAH : public MachineFunctionPass {
-    static char ID;
-    MSAH() : MachineFunctionPass(ID) {}
-
-    virtual bool runOnMachineFunction(MachineFunction &MF) {
-      const X86TargetMachine *TM =
-        static_cast<const X86TargetMachine *>(&MF.getTarget());
-      const TargetFrameLowering *TFI = TM->getFrameLowering();
-      MachineRegisterInfo &RI = MF.getRegInfo();
-      X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
-      unsigned StackAlignment = TFI->getStackAlignment();
-
-      // Be over-conservative: scan over all vreg defs and find whether vector
-      // registers are used. If yes, there is a possibility that vector register
-      // will be spilled and thus require dynamic stack realignment.
-      for (unsigned i = 0, e = RI.getNumVirtRegs(); i != e; ++i) {
-        unsigned Reg = TargetRegisterInfo::index2VirtReg(i);
-        if (RI.getRegClass(Reg)->getAlignment() > StackAlignment) {
-          FuncInfo->setForceFramePointer(true);
-          return true;
-        }
-      }
-      // Nothing to do
-      return false;
-    }
-
-    virtual const char *getPassName() const {
-      return "X86 Maximal Stack Alignment Check";
-    }
-
-    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
-      AU.setPreservesCFG();
-      MachineFunctionPass::getAnalysisUsage(AU);
-    }
-  };
-
-  char MSAH::ID = 0;
-}
-
-FunctionPass*
-llvm::createX86MaxStackAlignmentHeuristicPass() { return new MSAH(); }
diff --git a/lib/Target/X86/X86RegisterInfo.h b/lib/Target/X86/X86RegisterInfo.h
index 7932ede8dd65..b9d7b8cf8b9a 100644
--- a/lib/Target/X86/X86RegisterInfo.h
+++ b/lib/Target/X86/X86RegisterInfo.h
@@ -117,12 +117,9 @@ public:
   bool hasReservedSpillSlot(const MachineFunction &MF, unsigned Reg,
                             int &FrameIdx) const;
 
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator MI) const;
-
   void eliminateFrameIndex(MachineBasicBlock::iterator MI,
-                           int SPAdj, RegScavenger *RS = NULL) const;
+                           int SPAdj, unsigned FIOperandNum,
+                           RegScavenger *RS = NULL) const;
 
   // Debug information queries.
   unsigned getFrameRegister(const MachineFunction &MF) const;
diff --git a/lib/Target/X86/X86SchedHaswell.td b/lib/Target/X86/X86SchedHaswell.td
new file mode 100644
index 000000000000..7de6791f2e48
--- /dev/null
+++ b/lib/Target/X86/X86SchedHaswell.td
@@ -0,0 +1,126 @@
+//=- X86SchedHaswell.td - X86 Haswell Scheduling -------------*- tablegen -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the machine model for Haswell to support instruction
+// scheduling and other instruction cost heuristics.
+//
+//===----------------------------------------------------------------------===//
+
+def HaswellModel : SchedMachineModel {
+  // All x86 instructions are modeled as a single micro-op, and HW can decode 4
+  // instructions per cycle.
+  let IssueWidth = 4;
+  let MinLatency = 0; // 0 = Out-of-order execution.
+  let LoadLatency = 4;
+  let ILPWindow = 40;
+  let MispredictPenalty = 16;
+}
+
+let SchedModel = HaswellModel in {
+
+// Haswell can issue micro-ops to 8 different ports in one cycle.
+
+// Ports 0, 1, 5, 6 and 7 handle all computation.
+// Port 4 gets the data half of stores. Store data can be available later than
+// the store address, but since we don't model the latency of stores, we can
+// ignore that.
+// Ports 2 and 3 are identical. They handle loads and the address half of
+// stores. Port 7 can handle address calculations.
+def HWPort0 : ProcResource<1>;
+def HWPort1 : ProcResource<1>;
+def HWPort2 : ProcResource<1>;
+def HWPort3 : ProcResource<1>;
+def HWPort4 : ProcResource<1>;
+def HWPort5 : ProcResource<1>;
+def HWPort6 : ProcResource<1>;
+def HWPort7 : ProcResource<1>;
+
+// Many micro-ops are capable of issuing on multiple ports.
+def HWPort23  : ProcResGroup<[HWPort2, HWPort3]>;
+def HWPort237 : ProcResGroup<[HWPort2, HWPort3, HWPort7]>;
+def HWPort05  : ProcResGroup<[HWPort0, HWPort5]>;
+def HWPort056 : ProcResGroup<[HWPort0, HWPort5, HWPort6]>;
+def HWPort15  : ProcResGroup<[HWPort1, HWPort5]>;
+def HWPort015 : ProcResGroup<[HWPort0, HWPort1, HWPort5]>;
+def HWPort0156: ProcResGroup<[HWPort0, HWPort1, HWPort5, HWPort6]>;
+
+// Integer division issued on port 0.
+def HWDivider : ProcResource<1>;
+
+// Loads are 4 cycles, so ReadAfterLd registers needn't be available until 4
+// cycles after the memory operand.
+def : ReadAdvance<ReadAfterLd, 4>;
+
+// Many SchedWrites are defined in pairs with and without a folded load.
+// Instructions with folded loads are usually micro-fused, so they only appear
+// as two micro-ops when queued in the reservation station.
+// This multiclass defines the resource usage for variants with and without
+// folded loads.
+multiclass HWWriteResPair<X86FoldableSchedWrite SchedRW,
+                          ProcResourceKind ExePort,
+                          int Lat> {
+  // Register variant is using a single cycle on ExePort.
+  def : WriteRes<SchedRW, [ExePort]> { let Latency = Lat; }
+
+  // Memory variant also uses a cycle on port 2/3 and adds 4 cycles to the
+  // latency.
+  def : WriteRes<SchedRW.Folded, [HWPort23, ExePort]> {
+     let Latency = !add(Lat, 4);
+  }
+}
+
+// A folded store needs a cycle on port 4 for the store data, but it does not
+// need an extra port 2/3 cycle to recompute the address.
+def : WriteRes<WriteRMW, [HWPort4]>;
+
+def : WriteRes<WriteStore, [HWPort237, HWPort4]>;
+def : WriteRes<WriteLoad,  [HWPort23]> { let Latency = 4; }
+def : WriteRes<WriteMove,  [HWPort0156]>;
+def : WriteRes<WriteZero,  []>;
+
+defm : HWWriteResPair<WriteALU,   HWPort0156, 1>;
+defm : HWWriteResPair<WriteIMul,  HWPort1,   3>;
+defm : HWWriteResPair<WriteShift, HWPort056,  1>;
+defm : HWWriteResPair<WriteJump,  HWPort5,   1>;
+
+// This is for simple LEAs with one or two input operands.
+// The complex ones can only execute on port 1, and they require two cycles on
+// the port to read all inputs. We don't model that.
+def : WriteRes<WriteLEA, [HWPort15]>;
+
+// This is quite rough, latency depends on the dividend.
+def : WriteRes<WriteIDiv, [HWPort0, HWDivider]> {
+  let Latency = 25;
+  let ResourceCycles = [1, 10];
+}
+def : WriteRes<WriteIDivLd, [HWPort23, HWPort0, HWDivider]> {
+  let Latency = 29;
+  let ResourceCycles = [1, 1, 10];
+}
+
+// Scalar and vector floating point.
+defm : HWWriteResPair<WriteFAdd,   HWPort1, 3>;
+defm : HWWriteResPair<WriteFMul,   HWPort0, 5>;
+defm : HWWriteResPair<WriteFDiv,   HWPort0, 12>; // 10-14 cycles.
+defm : HWWriteResPair<WriteFRcp,   HWPort0, 5>;
+defm : HWWriteResPair<WriteFSqrt,  HWPort0, 15>;
+defm : HWWriteResPair<WriteCvtF2I, HWPort1, 3>;
+defm : HWWriteResPair<WriteCvtI2F, HWPort1, 4>;
+defm : HWWriteResPair<WriteCvtF2F, HWPort1, 3>;
+
+// Vector integer operations.
+defm : HWWriteResPair<WriteVecShift, HWPort05,  1>;
+defm : HWWriteResPair<WriteVecLogic, HWPort015, 1>;
+defm : HWWriteResPair<WriteVecALU,   HWPort15,  1>;
+defm : HWWriteResPair<WriteVecIMul,  HWPort0,   5>;
+defm : HWWriteResPair<WriteShuffle,  HWPort15,  1>;
+
+def : WriteRes<WriteSystem,     [HWPort0156]> { let Latency = 100; }
+def : WriteRes<WriteMicrocoded, [HWPort0156]> { let Latency = 100; }
+} // SchedModel
diff --git a/lib/Target/X86/X86SchedSandyBridge.td b/lib/Target/X86/X86SchedSandyBridge.td
new file mode 100644
index 000000000000..74d5f1b6eba8
--- /dev/null
+++ b/lib/Target/X86/X86SchedSandyBridge.td
@@ -0,0 +1,122 @@
+//=- X86SchedSandyBridge.td - X86 Sandy Bridge Scheduling ----*- tablegen -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the machine model for Sandy Bridge to support instruction
+// scheduling and other instruction cost heuristics.
+//
+//===----------------------------------------------------------------------===//
+
+def SandyBridgeModel : SchedMachineModel {
+  // All x86 instructions are modeled as a single micro-op, and SB can decode 4
+  // instructions per cycle.
+  // FIXME: Identify instructions that aren't a single fused micro-op.
+  let IssueWidth = 4;
+  let MinLatency = 0; // 0 = Out-of-order execution.
+  let LoadLatency = 4;
+  let ILPWindow = 30;
+  let MispredictPenalty = 16;
+}
+
+let SchedModel = SandyBridgeModel in {
+
+// Sandy Bridge can issue micro-ops to 6 different ports in one cycle.
+
+// Ports 0, 1, and 5 handle all computation.
+def SBPort0 : ProcResource<1>;
+def SBPort1 : ProcResource<1>;
+def SBPort5 : ProcResource<1>;
+
+// Ports 2 and 3 are identical. They handle loads and the address half of
+// stores.
+def SBPort23 : ProcResource<2>;
+
+// Port 4 gets the data half of stores. Store data can be available later than
+// the store address, but since we don't model the latency of stores, we can
+// ignore that.
+def SBPort4 : ProcResource<1>;
+
+// Many micro-ops are capable of issuing on multiple ports.
+def SBPort05  : ProcResGroup<[SBPort0, SBPort5]>;
+def SBPort15  : ProcResGroup<[SBPort1, SBPort5]>;
+def SBPort015 : ProcResGroup<[SBPort0, SBPort1, SBPort5]>;
+
+// Integer division issued on port 0.
+def SBDivider : ProcResource<1>;
+
+// Loads are 4 cycles, so ReadAfterLd registers needn't be available until 4
+// cycles after the memory operand.
+def : ReadAdvance<ReadAfterLd, 4>;
+
+// Many SchedWrites are defined in pairs with and without a folded load.
+// Instructions with folded loads are usually micro-fused, so they only appear
+// as two micro-ops when queued in the reservation station.
+// This multiclass defines the resource usage for variants with and without
+// folded loads.
+multiclass SBWriteResPair<X86FoldableSchedWrite SchedRW,
+                          ProcResourceKind ExePort,
+                          int Lat> {
+  // Register variant is using a single cycle on ExePort.
+  def : WriteRes<SchedRW, [ExePort]> { let Latency = Lat; }
+
+  // Memory variant also uses a cycle on port 2/3 and adds 4 cycles to the
+  // latency.
+  def : WriteRes<SchedRW.Folded, [SBPort23, ExePort]> {
+     let Latency = !add(Lat, 4);
+  }
+}
+
+// A folded store needs a cycle on port 4 for the store data, but it does not
+// need an extra port 2/3 cycle to recompute the address.
+def : WriteRes<WriteRMW, [SBPort4]>;
+
+def : WriteRes<WriteStore, [SBPort23, SBPort4]>;
+def : WriteRes<WriteLoad,  [SBPort23]> { let Latency = 4; }
+def : WriteRes<WriteMove,  [SBPort015]>;
+def : WriteRes<WriteZero,  []>;
+
+defm : SBWriteResPair<WriteALU,   SBPort015, 1>;
+defm : SBWriteResPair<WriteIMul,  SBPort1,   3>;
+defm : SBWriteResPair<WriteShift, SBPort05,  1>;
+defm : SBWriteResPair<WriteJump,  SBPort5,   1>;
+
+// This is for simple LEAs with one or two input operands.
+// The complex ones can only execute on port 1, and they require two cycles on
+// the port to read all inputs. We don't model that.
+def : WriteRes<WriteLEA, [SBPort15]>;
+
+// This is quite rough, latency depends on the dividend.
+def : WriteRes<WriteIDiv, [SBPort0, SBDivider]> {
+  let Latency = 25;
+  let ResourceCycles = [1, 10];
+}
+def : WriteRes<WriteIDivLd, [SBPort23, SBPort0, SBDivider]> {
+  let Latency = 29;
+  let ResourceCycles = [1, 1, 10];
+}
+
+// Scalar and vector floating point.
+defm : SBWriteResPair<WriteFAdd,   SBPort1, 3>;
+defm : SBWriteResPair<WriteFMul,   SBPort0, 5>;
+defm : SBWriteResPair<WriteFDiv,   SBPort0, 12>; // 10-14 cycles.
+defm : SBWriteResPair<WriteFRcp,   SBPort0, 5>;
+defm : SBWriteResPair<WriteFSqrt,  SBPort0, 15>;
+defm : SBWriteResPair<WriteCvtF2I, SBPort1, 3>;
+defm : SBWriteResPair<WriteCvtI2F, SBPort1, 4>;
+defm : SBWriteResPair<WriteCvtF2F, SBPort1, 3>;
+
+// Vector integer operations.
+defm : SBWriteResPair<WriteVecShift, SBPort05,  1>;
+defm : SBWriteResPair<WriteVecLogic, SBPort015, 1>;
+defm : SBWriteResPair<WriteVecALU,   SBPort15,  1>;
+defm : SBWriteResPair<WriteVecIMul,  SBPort0,   5>;
+defm : SBWriteResPair<WriteShuffle,  SBPort15,  1>;
+
+def : WriteRes<WriteSystem,     [SBPort015]> { let Latency = 100; }
+def : WriteRes<WriteMicrocoded, [SBPort015]> { let Latency = 100; }
+} // SchedModel
diff --git a/lib/Target/X86/X86Schedule.td b/lib/Target/X86/X86Schedule.td
index c14407f9ac1b..9fbde88b7100 100644
--- a/lib/Target/X86/X86Schedule.td
+++ b/lib/Target/X86/X86Schedule.td
@@ -7,9 +7,94 @@
 //
 //===----------------------------------------------------------------------===//
 
+// InstrSchedModel annotations for out-of-order CPUs.
+//
+// These annotations are independent of the itinerary classes defined below.
+
+// Instructions with folded loads need to read the memory operand immediately,
+// but other register operands don't have to be read until the load is ready.
+// These operands are marked with ReadAfterLd.
+def ReadAfterLd : SchedRead;
+
+// Instructions with both a load and a store folded are modeled as a folded
+// load + WriteRMW.
+def WriteRMW : SchedWrite;
+
+// Most instructions can fold loads, so almost every SchedWrite comes in two
+// variants: With and without a folded load.
+// An X86FoldableSchedWrite holds a reference to the corresponding SchedWrite
+// with a folded load.
+class X86FoldableSchedWrite : SchedWrite {
+  // The SchedWrite to use when a load is folded into the instruction.
+  SchedWrite Folded;
+}
+
+// Multiclass that produces a linked pair of SchedWrites.
+multiclass X86SchedWritePair {
+  // Register-Memory operation.
+  def Ld : SchedWrite;
+  // Register-Register operation.
+  def NAME : X86FoldableSchedWrite {
+    let Folded = !cast<SchedWrite>(NAME#"Ld");
+  }
+}
+
+// Arithmetic.
+defm WriteALU  : X86SchedWritePair; // Simple integer ALU op.
+defm WriteIMul : X86SchedWritePair; // Integer multiplication.
+defm WriteIDiv : X86SchedWritePair; // Integer division.
+def  WriteLEA  : SchedWrite;        // LEA instructions can't fold loads.
+
+// Integer shifts and rotates.
+defm WriteShift : X86SchedWritePair;
+
+// Loads, stores, and moves, not folded with other operations.
+def WriteLoad  : SchedWrite;
+def WriteStore : SchedWrite;
+def WriteMove  : SchedWrite;
+
+// Idioms that clear a register, like xorps %xmm0, %xmm0.
+// These can often bypass execution ports completely.
+def WriteZero : SchedWrite;
+
+// Branches don't produce values, so they have no latency, but they still
+// consume resources. Indirect branches can fold loads.
+defm WriteJump : X86SchedWritePair;
+
+// Floating point. This covers both scalar and vector operations.
+defm WriteFAdd  : X86SchedWritePair; // Floating point add/sub/compare.
+defm WriteFMul  : X86SchedWritePair; // Floating point multiplication.
+defm WriteFDiv  : X86SchedWritePair; // Floating point division.
+defm WriteFSqrt : X86SchedWritePair; // Floating point square root.
+defm WriteFRcp  : X86SchedWritePair; // Floating point reciprocal.
+defm WriteFMA   : X86SchedWritePair; // Fused Multiply Add.
+
+// FMA Scheduling helper class.
+class FMASC { X86FoldableSchedWrite Sched = WriteFAdd; }
+
+// Vector integer operations.
+defm WriteVecALU   : X86SchedWritePair; // Vector integer ALU op, no logicals.
+defm WriteVecShift : X86SchedWritePair; // Vector integer shifts.
+defm WriteVecIMul  : X86SchedWritePair; // Vector integer multiply.
+
+// Vector bitwise operations.
+// These are often used on both floating point and integer vectors.
+defm WriteVecLogic : X86SchedWritePair; // Vector and/or/xor.
+defm WriteShuffle  : X86SchedWritePair; // Vector shuffles and blends.
+
+// Conversion between integer and float.
+defm WriteCvtF2I : X86SchedWritePair; // Float -> Integer.
+defm WriteCvtI2F : X86SchedWritePair; // Integer -> Float.
+defm WriteCvtF2F : X86SchedWritePair; // Float -> Float size conversion.
+
+// Catch-all for expensive system instructions.
+def WriteSystem : SchedWrite;
+
+// Old microcoded instructions that nobody use.
+def WriteMicrocoded : SchedWrite;
+
 //===----------------------------------------------------------------------===//
 // Instruction Itinerary classes used for X86
-def IIC_DEFAULT     : InstrItinClass;
 def IIC_ALU_MEM     : InstrItinClass;
 def IIC_ALU_NONMEM  : InstrItinClass;
 def IIC_LEA         : InstrItinClass;
@@ -470,12 +555,19 @@ def IIC_NOP : InstrItinClass;
 // latencies. Since these latencies are not used for pipeline hazards,
 // they do not need to be exact.
 //
+// ILPWindow=10 is an arbitrary threshold that approximates cycles of
+// latency hidden by instruction buffers. The actual value is not very
+// important but should be zero for inorder and nonzero for OOO processors.
+//
 // The GenericModel contains no instruciton itineraries.
 def GenericModel : SchedMachineModel {
   let IssueWidth = 4;
   let MinLatency = 0;
   let LoadLatency = 4;
   let HighLatency = 10;
+  let ILPWindow = 10;
 }
 
 include "X86ScheduleAtom.td"
+include "X86SchedSandyBridge.td"
+include "X86SchedHaswell.td"
diff --git a/lib/Target/X86/X86ScheduleAtom.td b/lib/Target/X86/X86ScheduleAtom.td
index 87102614cc8b..cce8f1b11436 100644
--- a/lib/Target/X86/X86ScheduleAtom.td
+++ b/lib/Target/X86/X86ScheduleAtom.td
@@ -33,7 +33,6 @@ def AtomItineraries : ProcessorItineraries<
   // InstrItinData<class, [InstrStage<N, [P0], 0>,  InstrStage<N, [P1]>] >,
   //
   // Default is 1 cycle, port0 or port1
-  InstrItinData<IIC_DEFAULT, [InstrStage<1, [Port0, Port1]>] >,
   InstrItinData<IIC_ALU_MEM, [InstrStage<1, [Port0]>] >,
   InstrItinData<IIC_ALU_NONMEM, [InstrStage<1, [Port0, Port1]>] >,
   InstrItinData<IIC_LEA, [InstrStage<1, [Port1]>] >,
@@ -525,6 +524,7 @@ def AtomModel : SchedMachineModel {
                        // OperandCycles may be used for expected latency.
   let LoadLatency = 3; // Expected cycles, may be overriden by OperandCycles.
   let HighLatency = 30;// Expected, may be overriden by OperandCycles.
+  let ILPWindow = 0; // Always try to hide expected latency.
 
   let Itineraries = AtomItineraries;
 }
diff --git a/lib/Target/X86/X86SelectionDAGInfo.cpp b/lib/Target/X86/X86SelectionDAGInfo.cpp
index 723e50cc1886..f934fdd85914 100644
--- a/lib/Target/X86/X86SelectionDAGInfo.cpp
+++ b/lib/Target/X86/X86SelectionDAGInfo.cpp
@@ -13,8 +13,8 @@
 
 #define DEBUG_TYPE "x86-selectiondag-info"
 #include "X86TargetMachine.h"
-#include "llvm/DerivedTypes.h"
 #include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/IR/DerivedTypes.h"
 using namespace llvm;
 
 X86SelectionDAGInfo::X86SelectionDAGInfo(const X86TargetMachine &TM) :
@@ -202,6 +202,14 @@ X86SelectionDAGInfo::EmitTargetCodeForMemcpy(SelectionDAG &DAG, DebugLoc dl,
       SrcPtrInfo.getAddrSpace() >= 256)
     return SDValue();
 
+  // ESI might be used as a base pointer, in that case we can't simply overwrite
+  // the register.  Fall back to generic code.
+  const X86RegisterInfo *TRI =
+      static_cast<const X86RegisterInfo *>(DAG.getTarget().getRegisterInfo());
+  if (TRI->hasBasePointer(DAG.getMachineFunction()) &&
+      TRI->getBaseRegister() == X86::ESI)
+    return SDValue();
+
   MVT AVT;
   if (Align & 1)
     AVT = MVT::i8;
diff --git a/lib/Target/X86/X86Subtarget.cpp b/lib/Target/X86/X86Subtarget.cpp
index d1ed68028771..14619b63927b 100644
--- a/lib/Target/X86/X86Subtarget.cpp
+++ b/lib/Target/X86/X86Subtarget.cpp
@@ -14,11 +14,13 @@
 #define DEBUG_TYPE "subtarget"
 #include "X86Subtarget.h"
 #include "X86InstrInfo.h"
-#include "llvm/GlobalValue.h"
+#include "llvm/IR/Attributes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalValue.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/raw_ostream.h"
 #include "llvm/Support/Host.h"
+#include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetOptions.h"
 
@@ -35,8 +37,7 @@ using namespace llvm;
 /// ClassifyBlockAddressReference - Classify a blockaddress reference for the
 /// current subtarget according to how we should reference it in a non-pcrel
 /// context.
-unsigned char X86Subtarget::
-ClassifyBlockAddressReference() const {
+unsigned char X86Subtarget::ClassifyBlockAddressReference() const {
   if (isPICStyleGOT())    // 32-bit ELF targets.
     return X86II::MO_GOTOFF;
 
@@ -155,6 +156,12 @@ const char *X86Subtarget::getBZeroEntry() const {
   return 0;
 }
 
+bool X86Subtarget::hasSinCos() const {
+  return getTargetTriple().isMacOSX() &&
+    !getTargetTriple().isMacOSXVersionLT(10, 9) &&
+    is64Bit();
+}
+
 /// IsLegalToCallImmediateAddr - Return true if the subtarget allows calls
 /// to immediate address.
 bool X86Subtarget::IsLegalToCallImmediateAddr(const TargetMachine &TM) const {
@@ -234,12 +241,20 @@ void X86Subtarget::AutoDetectSubtargetFeatures() {
       ToggleFeature(X86::FeatureSlowBTMem);
     }
 
-    // If it's Nehalem, unaligned memory access is fast.
-    // Include Westmere and Sandy Bridge as well.
-    // FIXME: add later processors.
-    if (IsIntel && ((Family == 6 && Model == 26) ||
-        (Family == 6 && Model == 44) ||
-        (Family == 6 && Model == 42))) {
+    // If it's an Intel chip since Nehalem and not an Atom chip, unaligned
+    // memory access is fast. We hard code model numbers here because they
+    // aren't strictly increasing for Intel chips it seems.
+    if (IsIntel &&
+        ((Family == 6 && Model == 0x1E) || // Nehalem: Clarksfield, Lynnfield,
+                                           //          Jasper Froest
+         (Family == 6 && Model == 0x1A) || // Nehalem: Bloomfield, Nehalem-EP
+         (Family == 6 && Model == 0x2E) || // Nehalem: Nehalem-EX
+         (Family == 6 && Model == 0x25) || // Westmere: Arrandale, Clarksdale
+         (Family == 6 && Model == 0x2C) || // Westmere: Gulftown, Westmere-EP
+         (Family == 6 && Model == 0x2F) || // Westmere: Westmere-EX
+         (Family == 6 && Model == 0x2A) || // SandyBridge
+         (Family == 6 && Model == 0x2D) || // SandyBridge: SandyBridge-E*
+         (Family == 6 && Model == 0x3A))) {// IvyBridge
       IsUAMemFast = true;
       ToggleFeature(X86::FeatureFastUAMem);
     }
@@ -267,6 +282,10 @@ void X86Subtarget::AutoDetectSubtargetFeatures() {
         HasLZCNT = true;
         ToggleFeature(X86::FeatureLZCNT);
       }
+      if (IsIntel && ((ECX >> 8) & 0x1)) {
+        HasPRFCHW = true;
+        ToggleFeature(X86::FeaturePRFCHW);
+      }
       if (IsAMD) {
         if ((ECX >> 6) & 0x1) {
           HasSSE4A = true;
@@ -294,6 +313,10 @@ void X86Subtarget::AutoDetectSubtargetFeatures() {
         HasBMI = true;
         ToggleFeature(X86::FeatureBMI);
       }
+      if ((EBX >> 4) & 0x1) {
+        HasHLE = true;
+        ToggleFeature(X86::FeatureHLE);
+      }
       if (IsIntel && ((EBX >> 5) & 0x1)) {
         X86SSELevel = AVX2;
         ToggleFeature(X86::FeatureAVX2);
@@ -306,48 +329,35 @@ void X86Subtarget::AutoDetectSubtargetFeatures() {
         HasRTM = true;
         ToggleFeature(X86::FeatureRTM);
       }
+      if (IsIntel && ((EBX >> 19) & 0x1)) {
+        HasADX = true;
+        ToggleFeature(X86::FeatureADX);
+      }
+      if (IsIntel && ((EBX >> 18) & 0x1)) {
+        HasRDSEED = true;
+        ToggleFeature(X86::FeatureRDSEED);
+      }
     }
   }
 }
 
-X86Subtarget::X86Subtarget(const std::string &TT, const std::string &CPU,
-                           const std::string &FS,
-                           unsigned StackAlignOverride, bool is64Bit)
-  : X86GenSubtargetInfo(TT, CPU, FS)
-  , X86ProcFamily(Others)
-  , PICStyle(PICStyles::None)
-  , X86SSELevel(NoMMXSSE)
-  , X863DNowLevel(NoThreeDNow)
-  , HasCMov(false)
-  , HasX86_64(false)
-  , HasPOPCNT(false)
-  , HasSSE4A(false)
-  , HasAES(false)
-  , HasPCLMUL(false)
-  , HasFMA(false)
-  , HasFMA4(false)
-  , HasXOP(false)
-  , HasMOVBE(false)
-  , HasRDRAND(false)
-  , HasF16C(false)
-  , HasFSGSBase(false)
-  , HasLZCNT(false)
-  , HasBMI(false)
-  , HasBMI2(false)
-  , HasRTM(false)
-  , IsBTMemSlow(false)
-  , IsUAMemFast(false)
-  , HasVectorUAMem(false)
-  , HasCmpxchg16b(false)
-  , UseLeaForSP(false)
-  , HasSlowDivide(false)
-  , PostRAScheduler(false)
-  , stackAlignment(4)
-  // FIXME: this is a known good value for Yonah. How about others?
-  , MaxInlineSizeThreshold(128)
-  , TargetTriple(TT)
-  , In64BitMode(is64Bit) {
-  // Determine default and user specified characteristics
+void X86Subtarget::resetSubtargetFeatures(const MachineFunction *MF) {
+  AttributeSet FnAttrs = MF->getFunction()->getAttributes();
+  Attribute CPUAttr = FnAttrs.getAttribute(AttributeSet::FunctionIndex,
+                                           "target-cpu");
+  Attribute FSAttr = FnAttrs.getAttribute(AttributeSet::FunctionIndex,
+                                          "target-features");
+  std::string CPU =
+    !CPUAttr.hasAttribute(Attribute::None) ?CPUAttr.getValueAsString() : "";
+  std::string FS =
+    !FSAttr.hasAttribute(Attribute::None) ? FSAttr.getValueAsString() : "";
+  if (!FS.empty()) {
+    initializeEnvironment();
+    resetSubtargetFeatures(CPU, FS);
+  }
+}
+
+void X86Subtarget::resetSubtargetFeatures(StringRef CPU, StringRef FS) {
   std::string CPUName = CPU;
   if (!FS.empty() || !CPU.empty()) {
     if (CPUName.empty()) {
@@ -424,6 +434,57 @@ X86Subtarget::X86Subtarget(const std::string &TT, const std::string &CPU,
     stackAlignment = 16;
 }
 
+void X86Subtarget::initializeEnvironment() {
+  X86SSELevel = NoMMXSSE;
+  X863DNowLevel = NoThreeDNow;
+  HasCMov = false;
+  HasX86_64 = false;
+  HasPOPCNT = false;
+  HasSSE4A = false;
+  HasAES = false;
+  HasPCLMUL = false;
+  HasFMA = false;
+  HasFMA4 = false;
+  HasXOP = false;
+  HasMOVBE = false;
+  HasRDRAND = false;
+  HasF16C = false;
+  HasFSGSBase = false;
+  HasLZCNT = false;
+  HasBMI = false;
+  HasBMI2 = false;
+  HasRTM = false;
+  HasHLE = false;
+  HasADX = false;
+  HasPRFCHW = false;
+  HasRDSEED = false;
+  IsBTMemSlow = false;
+  IsUAMemFast = false;
+  HasVectorUAMem = false;
+  HasCmpxchg16b = false;
+  UseLeaForSP = false;
+  HasSlowDivide = false;
+  PostRAScheduler = false;
+  PadShortFunctions = false;
+  CallRegIndirect = false;
+  stackAlignment = 4;
+  // FIXME: this is a known good value for Yonah. How about others?
+  MaxInlineSizeThreshold = 128;
+}
+
+X86Subtarget::X86Subtarget(const std::string &TT, const std::string &CPU,
+                           const std::string &FS,
+                           unsigned StackAlignOverride, bool is64Bit)
+  : X86GenSubtargetInfo(TT, CPU, FS)
+  , X86ProcFamily(Others)
+  , PICStyle(PICStyles::None)
+  , TargetTriple(TT)
+  , StackAlignOverride(StackAlignOverride)
+  , In64BitMode(is64Bit) {
+  initializeEnvironment();
+  resetSubtargetFeatures(CPU, FS);
+}
+
 bool X86Subtarget::enablePostRAScheduler(
            CodeGenOpt::Level OptLevel,
            TargetSubtargetInfo::AntiDepBreakMode& Mode,
diff --git a/lib/Target/X86/X86Subtarget.h b/lib/Target/X86/X86Subtarget.h
index 8bf4cc77f762..6fbdb1d5f00f 100644
--- a/lib/Target/X86/X86Subtarget.h
+++ b/lib/Target/X86/X86Subtarget.h
@@ -14,8 +14,8 @@
 #ifndef X86SUBTARGET_H
 #define X86SUBTARGET_H
 
-#include "llvm/CallingConv.h"
 #include "llvm/ADT/Triple.h"
+#include "llvm/IR/CallingConv.h"
 #include "llvm/Target/TargetSubtargetInfo.h"
 #include <string>
 
@@ -121,6 +121,18 @@ protected:
   /// HasRTM - Processor has RTM instructions.
   bool HasRTM;
 
+  /// HasHLE - Processor has HLE.
+  bool HasHLE;
+
+  /// HasADX - Processor has ADX instructions.
+  bool HasADX;
+
+  /// HasPRFCHW - Processor has PRFCHW instructions.
+  bool HasPRFCHW;
+
+  /// HasRDSEED - Processor has RDSEED instructions.
+  bool HasRDSEED;
+
   /// IsBTMemSlow - True if BT (bit test) of memory instructions are slow.
   bool IsBTMemSlow;
 
@@ -146,6 +158,14 @@ protected:
   /// PostRAScheduler - True if using post-register-allocation scheduler.
   bool PostRAScheduler;
 
+  /// PadShortFunctions - True if the short functions should be padded to prevent
+  /// a stall when returning too early.
+  bool PadShortFunctions;
+
+  /// CallRegIndirect - True if the Calls with memory reference should be converted
+  /// to a register-based indirect call.
+  bool CallRegIndirect;
+
   /// stackAlignment - The minimum alignment known to hold of the stack frame on
   /// entry to the function and which must be maintained by every function.
   unsigned stackAlignment;
@@ -161,11 +181,13 @@ protected:
   InstrItineraryData InstrItins;
 
 private:
+  /// StackAlignOverride - Override the stack alignment.
+  unsigned StackAlignOverride;
+
   /// In64BitMode - True if compiling for 64-bit, false for 32-bit.
   bool In64BitMode;
 
 public:
-
   /// This constructor initializes the data members to match that
   /// of the specified triple.
   ///
@@ -190,7 +212,26 @@ public:
   /// instruction.
   void AutoDetectSubtargetFeatures();
 
-  bool is64Bit() const { return In64BitMode; }
+  /// \brief Reset the features for the X86 target.
+  virtual void resetSubtargetFeatures(const MachineFunction *MF);
+private:
+  void initializeEnvironment();
+  void resetSubtargetFeatures(StringRef CPU, StringRef FS);
+public:
+  /// Is this x86_64? (disregarding specific ABI / programming model)
+  bool is64Bit() const {
+    return In64BitMode;
+  }
+
+  /// Is this x86_64 with the ILP32 programming model (x32 ABI)?
+  bool isTarget64BitILP32() const {
+    return In64BitMode && (TargetTriple.getEnvironment() == Triple::GNUX32);
+  }
+
+  /// Is this x86_64 with the LP64 programming model (standard AMD64, no x32)?
+  bool isTarget64BitLP64() const {
+    return In64BitMode && (TargetTriple.getEnvironment() != Triple::GNUX32);
+  }
 
   PICStyles::Style getPICStyle() const { return PICStyle; }
   void setPICStyle(PICStyles::Style Style)  { PICStyle = Style; }
@@ -205,6 +246,8 @@ public:
   bool hasSSE42() const { return X86SSELevel >= SSE42; }
   bool hasAVX() const { return X86SSELevel >= AVX; }
   bool hasAVX2() const { return X86SSELevel >= AVX2; }
+  bool hasFp256() const { return hasAVX(); }
+  bool hasInt256() const { return hasAVX2(); }
   bool hasSSE4A() const { return HasSSE4A; }
   bool has3DNow() const { return X863DNowLevel >= ThreeDNow; }
   bool has3DNowA() const { return X863DNowLevel >= ThreeDNowA; }
@@ -223,12 +266,18 @@ public:
   bool hasBMI() const { return HasBMI; }
   bool hasBMI2() const { return HasBMI2; }
   bool hasRTM() const { return HasRTM; }
+  bool hasHLE() const { return HasHLE; }
+  bool hasADX() const { return HasADX; }
+  bool hasPRFCHW() const { return HasPRFCHW; }
+  bool hasRDSEED() const { return HasRDSEED; }
   bool isBTMemSlow() const { return IsBTMemSlow; }
   bool isUnalignedMemAccessFast() const { return IsUAMemFast; }
   bool hasVectorUAMem() const { return HasVectorUAMem; }
   bool hasCmpxchg16b() const { return HasCmpxchg16b; }
   bool useLeaForSP() const { return UseLeaForSP; }
   bool hasSlowDivide() const { return HasSlowDivide; }
+  bool padShortFunctions() const { return PadShortFunctions; }
+  bool callRegIndirect() const { return CallRegIndirect; }
 
   bool isAtom() const { return X86ProcFamily == IntelAtom; }
 
@@ -247,7 +296,7 @@ public:
   }
   bool isTargetLinux() const { return TargetTriple.getOS() == Triple::Linux; }
   bool isTargetNaCl() const {
-    return TargetTriple.getOS() == Triple::NativeClient;
+    return TargetTriple.getOS() == Triple::NaCl;
   }
   bool isTargetNaCl32() const { return isTargetNaCl() && !is64Bit(); }
   bool isTargetNaCl64() const { return isTargetNaCl() && is64Bit(); }
@@ -308,6 +357,10 @@ public:
   /// memset with zero passed as the second argument. Otherwise it
   /// returns null.
   const char *getBZeroEntry() const;
+  
+  /// This function returns true if the target has sincos() routine in its
+  /// compiler runtime or math libraries.
+  bool hasSinCos() const;
 
   /// enablePostRAScheduler - run for Atom optimization.
   bool enablePostRAScheduler(CodeGenOpt::Level OptLevel,
diff --git a/lib/Target/X86/X86TargetMachine.cpp b/lib/Target/X86/X86TargetMachine.cpp
index 158f9dc06693..8aa58a204260 100644
--- a/lib/Target/X86/X86TargetMachine.cpp
+++ b/lib/Target/X86/X86TargetMachine.cpp
@@ -13,13 +13,13 @@
 
 #include "X86TargetMachine.h"
 #include "X86.h"
-#include "llvm/PassManager.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/Passes.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/FormattedStream.h"
-#include "llvm/Target/TargetOptions.h"
 #include "llvm/Support/TargetRegistry.h"
+#include "llvm/Target/TargetOptions.h"
 using namespace llvm;
 
 extern "C" void LLVMInitializeX86Target() {
@@ -46,10 +46,9 @@ X86_32TargetMachine::X86_32TargetMachine(const Target &T, StringRef TT,
                "e-p:32:32-f64:32:64-i64:32:64-f80:32:32-f128:128:128-"
                "n8:16:32-S128"),
     InstrInfo(*this),
-    TSInfo(*this),
     TLInfo(*this),
-    JITInfo(*this),
-    STTI(&TLInfo), VTTI(&TLInfo) {
+    TSInfo(*this),
+    JITInfo(*this) {
 }
 
 void X86_64TargetMachine::anchor() { }
@@ -60,13 +59,16 @@ X86_64TargetMachine::X86_64TargetMachine(const Target &T, StringRef TT,
                                          Reloc::Model RM, CodeModel::Model CM,
                                          CodeGenOpt::Level OL)
   : X86TargetMachine(T, TT, CPU, FS, Options, RM, CM, OL, true),
-    DL("e-p:64:64-s:64-f64:64:64-i64:64:64-f80:128:128-f128:128:128-"
-               "n8:16:32:64-S128"),
+    // The x32 ABI dictates the ILP32 programming model for x64.
+    DL(getSubtargetImpl()->isTarget64BitILP32() ?
+        "e-p:32:32-s:64-f64:64:64-i64:64:64-f80:128:128-f128:128:128-"
+        "n8:16:32:64-S128" :
+        "e-p:64:64-s:64-f64:64:64-i64:64:64-f80:128:128-f128:128:128-"
+        "n8:16:32:64-S128"),
     InstrInfo(*this),
-    TSInfo(*this),
     TLInfo(*this),
-    JITInfo(*this),
-    STTI(&TLInfo), VTTI(&TLInfo){
+    TSInfo(*this),
+    JITInfo(*this) {
 }
 
 /// X86TargetMachine ctor - Create an X86 target.
@@ -121,6 +123,19 @@ X86EarlyIfConv("x86-early-ifcvt",
 	       cl::desc("Enable early if-conversion on X86"));
 
 //===----------------------------------------------------------------------===//
+// X86 Analysis Pass Setup
+//===----------------------------------------------------------------------===//
+
+void X86TargetMachine::addAnalysisPasses(PassManagerBase &PM) {
+  // Add first the target-independent BasicTTI pass, then our X86 pass. This
+  // allows the X86 pass to delegate to the target independent layer when
+  // appropriate.
+  PM.add(createBasicTargetTransformInfoPass(getTargetLowering()));
+  PM.add(createX86TargetTransformInfoPass(this));
+}
+
+
+//===----------------------------------------------------------------------===//
 // Pass Pipeline Configuration
 //===----------------------------------------------------------------------===//
 
@@ -140,6 +155,7 @@ public:
   }
 
   virtual bool addInstSelector();
+  virtual bool addILPOpts();
   virtual bool addPreRegAlloc();
   virtual bool addPostRegAlloc();
   virtual bool addPreEmitPass();
@@ -147,12 +163,7 @@ public:
 } // namespace
 
 TargetPassConfig *X86TargetMachine::createPassConfig(PassManagerBase &PM) {
-  X86PassConfig *PC = new X86PassConfig(this, PM);
-
-  if (X86EarlyIfConv && Subtarget.hasCMov())
-    PC->enablePass(&EarlyIfConverterID);
-
-  return PC;
+  return new X86PassConfig(this, PM);
 }
 
 bool X86PassConfig::addInstSelector() {
@@ -170,8 +181,15 @@ bool X86PassConfig::addInstSelector() {
   return false;
 }
 
+bool X86PassConfig::addILPOpts() {
+  if (X86EarlyIfConv && getX86Subtarget().hasCMov()) {
+    addPass(&EarlyIfConverterID);
+    return true;
+  }
+  return false;
+}
+
 bool X86PassConfig::addPreRegAlloc() {
-  addPass(createX86MaxStackAlignmentHeuristicPass());
   return false;  // -print-machineinstr shouldn't print after this.
 }
 
@@ -192,6 +210,12 @@ bool X86PassConfig::addPreEmitPass() {
     ShouldPrint = true;
   }
 
+  if (getOptLevel() != CodeGenOpt::None &&
+      getX86Subtarget().padShortFunctions()) {
+    addPass(createX86PadShortFunctions());
+    ShouldPrint = true;
+  }
+
   return ShouldPrint;
 }
 
diff --git a/lib/Target/X86/X86TargetMachine.h b/lib/Target/X86/X86TargetMachine.h
index 12311a1abfbd..174d3918318d 100644
--- a/lib/Target/X86/X86TargetMachine.h
+++ b/lib/Target/X86/X86TargetMachine.h
@@ -15,16 +15,15 @@
 #define X86TARGETMACHINE_H
 
 #include "X86.h"
-#include "X86InstrInfo.h"
-#include "X86ISelLowering.h"
 #include "X86FrameLowering.h"
+#include "X86ISelLowering.h"
+#include "X86InstrInfo.h"
 #include "X86JITInfo.h"
 #include "X86SelectionDAGInfo.h"
 #include "X86Subtarget.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/DataLayout.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetTransformImpl.h"
+#include "llvm/Target/TargetMachine.h"
 
 namespace llvm {
 
@@ -65,6 +64,9 @@ public:
     return &InstrItins;
   }
 
+  /// \brief Register X86 analysis passes with a pass manager.
+  virtual void addAnalysisPasses(PassManagerBase &PM);
+
   // Set up the pass pipeline.
   virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
 
@@ -78,11 +80,9 @@ class X86_32TargetMachine : public X86TargetMachine {
   virtual void anchor();
   const DataLayout  DL; // Calculates type size & alignment
   X86InstrInfo      InstrInfo;
-  X86SelectionDAGInfo TSInfo;
   X86TargetLowering TLInfo;
+  X86SelectionDAGInfo TSInfo;
   X86JITInfo        JITInfo;
-  ScalarTargetTransformImpl STTI;
-  X86VectorTargetTransformInfo VTTI;
 public:
   X86_32TargetMachine(const Target &T, StringRef TT,
                       StringRef CPU, StringRef FS, const TargetOptions &Options,
@@ -101,12 +101,6 @@ public:
   virtual       X86JITInfo       *getJITInfo()         {
     return &JITInfo;
   }
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
 };
 
 /// X86_64TargetMachine - X86 64-bit target machine.
@@ -115,11 +109,9 @@ class X86_64TargetMachine : public X86TargetMachine {
   virtual void anchor();
   const DataLayout  DL; // Calculates type size & alignment
   X86InstrInfo      InstrInfo;
-  X86SelectionDAGInfo TSInfo;
   X86TargetLowering TLInfo;
+  X86SelectionDAGInfo TSInfo;
   X86JITInfo        JITInfo;
-  ScalarTargetTransformImpl STTI;
-  X86VectorTargetTransformInfo VTTI;
 public:
   X86_64TargetMachine(const Target &T, StringRef TT,
                       StringRef CPU, StringRef FS, const TargetOptions &Options,
@@ -138,12 +130,6 @@ public:
   virtual       X86JITInfo       *getJITInfo()         {
     return &JITInfo;
   }
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
 };
 
 } // End llvm namespace
diff --git a/lib/Target/X86/X86TargetObjectFile.cpp b/lib/Target/X86/X86TargetObjectFile.cpp
index 92aee0dd3fcf..871dacd6a1c1 100644
--- a/lib/Target/X86/X86TargetObjectFile.cpp
+++ b/lib/Target/X86/X86TargetObjectFile.cpp
@@ -8,23 +8,19 @@
 //===----------------------------------------------------------------------===//
 
 #include "X86TargetObjectFile.h"
-#include "X86TargetMachine.h"
-#include "llvm/ADT/StringExtras.h"
-#include "llvm/CodeGen/MachineModuleInfoImpls.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCSectionELF.h"
-#include "llvm/MC/MCSectionMachO.h"
-#include "llvm/Target/Mangler.h"
 #include "llvm/Support/Dwarf.h"
-#include "llvm/Support/ELF.h"
+#include "llvm/Target/Mangler.h"
+
 using namespace llvm;
 using namespace dwarf;
 
 const MCExpr *X86_64MachoTargetObjectFile::
-getExprForDwarfGlobalReference(const GlobalValue *GV, Mangler *Mang,
-                               MachineModuleInfo *MMI, unsigned Encoding,
-                               MCStreamer &Streamer) const {
+getTTypeGlobalReference(const GlobalValue *GV, Mangler *Mang,
+                        MachineModuleInfo *MMI, unsigned Encoding,
+                        MCStreamer &Streamer) const {
 
   // On Darwin/X86-64, we can reference dwarf symbols with foo@GOTPCREL+4, which
   // is an indirect pc-relative reference.
@@ -37,7 +33,7 @@ getExprForDwarfGlobalReference(const GlobalValue *GV, Mangler *Mang,
   }
 
   return TargetLoweringObjectFileMachO::
-    getExprForDwarfGlobalReference(GV, Mang, MMI, Encoding, Streamer);
+    getTTypeGlobalReference(GV, Mang, MMI, Encoding, Streamer);
 }
 
 MCSymbol *X86_64MachoTargetObjectFile::
diff --git a/lib/Target/X86/X86TargetObjectFile.h b/lib/Target/X86/X86TargetObjectFile.h
index 2d320c594cb9..9d26d389d4de 100644
--- a/lib/Target/X86/X86TargetObjectFile.h
+++ b/lib/Target/X86/X86TargetObjectFile.h
@@ -11,8 +11,8 @@
 #define LLVM_TARGET_X86_TARGETOBJECTFILE_H
 
 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetLoweringObjectFile.h"
+#include "llvm/Target/TargetMachine.h"
 
 namespace llvm {
 
@@ -21,9 +21,9 @@ namespace llvm {
   class X86_64MachoTargetObjectFile : public TargetLoweringObjectFileMachO {
   public:
     virtual const MCExpr *
-    getExprForDwarfGlobalReference(const GlobalValue *GV, Mangler *Mang,
-                                   MachineModuleInfo *MMI, unsigned Encoding,
-                                   MCStreamer &Streamer) const;
+    getTTypeGlobalReference(const GlobalValue *GV, Mangler *Mang,
+                            MachineModuleInfo *MMI, unsigned Encoding,
+                            MCStreamer &Streamer) const;
 
     // getCFIPersonalitySymbol - The symbol that gets passed to
     // .cfi_personality.
diff --git a/lib/Target/X86/X86TargetTransformInfo.cpp b/lib/Target/X86/X86TargetTransformInfo.cpp
new file mode 100644
index 000000000000..a98c6991192c
--- /dev/null
+++ b/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -0,0 +1,495 @@
+//===-- X86TargetTransformInfo.cpp - X86 specific TTI pass ----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+/// \file
+/// This file implements a TargetTransformInfo analysis pass specific to the
+/// X86 target machine. It uses the target's detailed information to provide
+/// more precise answers to certain TTI queries, while letting the target
+/// independent and default TTI implementations handle the rest.
+///
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "x86tti"
+#include "X86.h"
+#include "X86TargetMachine.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/CostTable.h"
+using namespace llvm;
+
+// Declare the pass initialization routine locally as target-specific passes
+// don't havve a target-wide initialization entry point, and so we rely on the
+// pass constructor initialization.
+namespace llvm {
+void initializeX86TTIPass(PassRegistry &);
+}
+
+namespace {
+
+class X86TTI : public ImmutablePass, public TargetTransformInfo {
+  const X86TargetMachine *TM;
+  const X86Subtarget *ST;
+  const X86TargetLowering *TLI;
+
+  /// Estimate the overhead of scalarizing an instruction. Insert and Extract
+  /// are set if the result needs to be inserted and/or extracted from vectors.
+  unsigned getScalarizationOverhead(Type *Ty, bool Insert, bool Extract) const;
+
+public:
+  X86TTI() : ImmutablePass(ID), TM(0), ST(0), TLI(0) {
+    llvm_unreachable("This pass cannot be directly constructed");
+  }
+
+  X86TTI(const X86TargetMachine *TM)
+      : ImmutablePass(ID), TM(TM), ST(TM->getSubtargetImpl()),
+        TLI(TM->getTargetLowering()) {
+    initializeX86TTIPass(*PassRegistry::getPassRegistry());
+  }
+
+  virtual void initializePass() {
+    pushTTIStack(this);
+  }
+
+  virtual void finalizePass() {
+    popTTIStack();
+  }
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+    TargetTransformInfo::getAnalysisUsage(AU);
+  }
+
+  /// Pass identification.
+  static char ID;
+
+  /// Provide necessary pointer adjustments for the two base classes.
+  virtual void *getAdjustedAnalysisPointer(const void *ID) {
+    if (ID == &TargetTransformInfo::ID)
+      return (TargetTransformInfo*)this;
+    return this;
+  }
+
+  /// \name Scalar TTI Implementations
+  /// @{
+  virtual PopcntSupportKind getPopcntSupport(unsigned TyWidth) const;
+
+  /// @}
+
+  /// \name Vector TTI Implementations
+  /// @{
+
+  virtual unsigned getNumberOfRegisters(bool Vector) const;
+  virtual unsigned getRegisterBitWidth(bool Vector) const;
+  virtual unsigned getMaximumUnrollFactor() const;
+  virtual unsigned getArithmeticInstrCost(unsigned Opcode, Type *Ty,
+                                          OperandValueKind,
+                                          OperandValueKind) const;
+  virtual unsigned getShuffleCost(ShuffleKind Kind, Type *Tp,
+                                  int Index, Type *SubTp) const;
+  virtual unsigned getCastInstrCost(unsigned Opcode, Type *Dst,
+                                    Type *Src) const;
+  virtual unsigned getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
+                                      Type *CondTy) const;
+  virtual unsigned getVectorInstrCost(unsigned Opcode, Type *Val,
+                                      unsigned Index) const;
+  virtual unsigned getMemoryOpCost(unsigned Opcode, Type *Src,
+                                   unsigned Alignment,
+                                   unsigned AddressSpace) const;
+
+  /// @}
+};
+
+} // end anonymous namespace
+
+INITIALIZE_AG_PASS(X86TTI, TargetTransformInfo, "x86tti",
+                   "X86 Target Transform Info", true, true, false)
+char X86TTI::ID = 0;
+
+ImmutablePass *
+llvm::createX86TargetTransformInfoPass(const X86TargetMachine *TM) {
+  return new X86TTI(TM);
+}
+
+
+//===----------------------------------------------------------------------===//
+//
+// X86 cost model.
+//
+//===----------------------------------------------------------------------===//
+
+X86TTI::PopcntSupportKind X86TTI::getPopcntSupport(unsigned TyWidth) const {
+  assert(isPowerOf2_32(TyWidth) && "Ty width must be power of 2");
+  // TODO: Currently the __builtin_popcount() implementation using SSE3
+  //   instructions is inefficient. Once the problem is fixed, we should
+  //   call ST->hasSSE3() instead of ST->hasSSE4().
+  return ST->hasSSE41() ? PSK_FastHardware : PSK_Software;
+}
+
+unsigned X86TTI::getNumberOfRegisters(bool Vector) const {
+  if (Vector && !ST->hasSSE1())
+    return 0;
+
+  if (ST->is64Bit())
+    return 16;
+  return 8;
+}
+
+unsigned X86TTI::getRegisterBitWidth(bool Vector) const {
+  if (Vector) {
+    if (ST->hasAVX()) return 256;
+    if (ST->hasSSE1()) return 128;
+    return 0;
+  }
+
+  if (ST->is64Bit())
+    return 64;
+  return 32;
+
+}
+
+unsigned X86TTI::getMaximumUnrollFactor() const {
+  if (ST->isAtom())
+    return 1;
+
+  // Sandybridge and Haswell have multiple execution ports and pipelined
+  // vector units.
+  if (ST->hasAVX())
+    return 4;
+
+  return 2;
+}
+
+unsigned X86TTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty,
+                                        OperandValueKind Op1Info,
+                                        OperandValueKind Op2Info) const {
+  // Legalize the type.
+  std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Ty);
+
+  int ISD = TLI->InstructionOpcodeToISD(Opcode);
+  assert(ISD && "Invalid opcode");
+
+  static const CostTblEntry<MVT> AVX2CostTable[] = {
+    // Shifts on v4i64/v8i32 on AVX2 is legal even though we declare to
+    // customize them to detect the cases where shift amount is a scalar one.
+    { ISD::SHL,     MVT::v4i32,    1 },
+    { ISD::SRL,     MVT::v4i32,    1 },
+    { ISD::SRA,     MVT::v4i32,    1 },
+    { ISD::SHL,     MVT::v8i32,    1 },
+    { ISD::SRL,     MVT::v8i32,    1 },
+    { ISD::SRA,     MVT::v8i32,    1 },
+    { ISD::SHL,     MVT::v2i64,    1 },
+    { ISD::SRL,     MVT::v2i64,    1 },
+    { ISD::SHL,     MVT::v4i64,    1 },
+    { ISD::SRL,     MVT::v4i64,    1 },
+
+    { ISD::SHL,  MVT::v32i8,  42 }, // cmpeqb sequence.
+    { ISD::SHL,  MVT::v16i16,  16*10 }, // Scalarized.
+
+    { ISD::SRL,  MVT::v32i8,  32*10 }, // Scalarized.
+    { ISD::SRL,  MVT::v16i16,  8*10 }, // Scalarized.
+
+    { ISD::SRA,  MVT::v32i8,  32*10 }, // Scalarized.
+    { ISD::SRA,  MVT::v16i16,  16*10 }, // Scalarized.
+    { ISD::SRA,  MVT::v4i64,  4*10 }, // Scalarized.
+  };
+
+  // Look for AVX2 lowering tricks.
+  if (ST->hasAVX2()) {
+    int Idx = CostTableLookup<MVT>(AVX2CostTable, array_lengthof(AVX2CostTable),
+                                   ISD, LT.second);
+    if (Idx != -1)
+      return LT.first * AVX2CostTable[Idx].Cost;
+  }
+
+  static const CostTblEntry<MVT> SSE2UniformConstCostTable[] = {
+    // We don't correctly identify costs of casts because they are marked as
+    // custom.
+    // Constant splats are cheaper for the following instructions.
+    { ISD::SHL,  MVT::v16i8,  1 }, // psllw.
+    { ISD::SHL,  MVT::v8i16,  1 }, // psllw.
+    { ISD::SHL,  MVT::v4i32,  1 }, // pslld
+    { ISD::SHL,  MVT::v2i64,  1 }, // psllq.
+
+    { ISD::SRL,  MVT::v16i8,  1 }, // psrlw.
+    { ISD::SRL,  MVT::v8i16,  1 }, // psrlw.
+    { ISD::SRL,  MVT::v4i32,  1 }, // psrld.
+    { ISD::SRL,  MVT::v2i64,  1 }, // psrlq.
+
+    { ISD::SRA,  MVT::v16i8,  4 }, // psrlw, pand, pxor, psubb.
+    { ISD::SRA,  MVT::v8i16,  1 }, // psraw.
+    { ISD::SRA,  MVT::v4i32,  1 }, // psrad.
+  };
+
+  if (Op2Info == TargetTransformInfo::OK_UniformConstantValue &&
+      ST->hasSSE2()) {
+    int Idx = CostTableLookup<MVT>(SSE2UniformConstCostTable,
+                                   array_lengthof(SSE2UniformConstCostTable),
+                                   ISD, LT.second);
+    if (Idx != -1)
+      return LT.first * SSE2UniformConstCostTable[Idx].Cost;
+  }
+
+
+  static const CostTblEntry<MVT> SSE2CostTable[] = {
+    // We don't correctly identify costs of casts because they are marked as
+    // custom.
+    // For some cases, where the shift amount is a scalar we would be able
+    // to generate better code. Unfortunately, when this is the case the value
+    // (the splat) will get hoisted out of the loop, thereby making it invisible
+    // to ISel. The cost model must return worst case assumptions because it is
+    // used for vectorization and we don't want to make vectorized code worse
+    // than scalar code.
+    { ISD::SHL,  MVT::v16i8,  30 }, // cmpeqb sequence.
+    { ISD::SHL,  MVT::v8i16,  8*10 }, // Scalarized.
+    { ISD::SHL,  MVT::v4i32,  2*5 }, // We optimized this using mul.
+    { ISD::SHL,  MVT::v2i64,  2*10 }, // Scalarized.
+
+    { ISD::SRL,  MVT::v16i8,  16*10 }, // Scalarized.
+    { ISD::SRL,  MVT::v8i16,  8*10 }, // Scalarized.
+    { ISD::SRL,  MVT::v4i32,  4*10 }, // Scalarized.
+    { ISD::SRL,  MVT::v2i64,  2*10 }, // Scalarized.
+
+    { ISD::SRA,  MVT::v16i8,  16*10 }, // Scalarized.
+    { ISD::SRA,  MVT::v8i16,  8*10 }, // Scalarized.
+    { ISD::SRA,  MVT::v4i32,  4*10 }, // Scalarized.
+    { ISD::SRA,  MVT::v2i64,  2*10 }, // Scalarized.
+  };
+
+  if (ST->hasSSE2()) {
+    int Idx = CostTableLookup<MVT>(SSE2CostTable, array_lengthof(SSE2CostTable),
+                                   ISD, LT.second);
+    if (Idx != -1)
+      return LT.first * SSE2CostTable[Idx].Cost;
+  }
+
+  static const CostTblEntry<MVT> AVX1CostTable[] = {
+    // We don't have to scalarize unsupported ops. We can issue two half-sized
+    // operations and we only need to extract the upper YMM half.
+    // Two ops + 1 extract + 1 insert = 4.
+    { ISD::MUL,     MVT::v8i32,    4 },
+    { ISD::SUB,     MVT::v8i32,    4 },
+    { ISD::ADD,     MVT::v8i32,    4 },
+    { ISD::SUB,     MVT::v4i64,    4 },
+    { ISD::ADD,     MVT::v4i64,    4 },
+    // A v4i64 multiply is custom lowered as two split v2i64 vectors that then
+    // are lowered as a series of long multiplies(3), shifts(4) and adds(2)
+    // Because we believe v4i64 to be a legal type, we must also include the
+    // split factor of two in the cost table. Therefore, the cost here is 18
+    // instead of 9.
+    { ISD::MUL,     MVT::v4i64,    18 },
+  };
+
+  // Look for AVX1 lowering tricks.
+  if (ST->hasAVX() && !ST->hasAVX2()) {
+    int Idx = CostTableLookup<MVT>(AVX1CostTable, array_lengthof(AVX1CostTable),
+                                   ISD, LT.second);
+    if (Idx != -1)
+      return LT.first * AVX1CostTable[Idx].Cost;
+  }
+
+  // Custom lowering of vectors.
+  static const CostTblEntry<MVT> CustomLowered[] = {
+    // A v2i64/v4i64 and multiply is custom lowered as a series of long
+    // multiplies(3), shifts(4) and adds(2).
+    { ISD::MUL,     MVT::v2i64,    9 },
+    { ISD::MUL,     MVT::v4i64,    9 },
+  };
+  int Idx = CostTableLookup<MVT>(CustomLowered, array_lengthof(CustomLowered),
+                                 ISD, LT.second);
+  if (Idx != -1)
+    return LT.first * CustomLowered[Idx].Cost;
+
+  // Special lowering of v4i32 mul on sse2, sse3: Lower v4i32 mul as 2x shuffle,
+  // 2x pmuludq, 2x shuffle.
+  if (ISD == ISD::MUL && LT.second == MVT::v4i32 && ST->hasSSE2() &&
+      !ST->hasSSE41())
+    return 6;
+
+  // Fallback to the default implementation.
+  return TargetTransformInfo::getArithmeticInstrCost(Opcode, Ty, Op1Info,
+                                                     Op2Info);
+}
+
+unsigned X86TTI::getShuffleCost(ShuffleKind Kind, Type *Tp, int Index,
+                                Type *SubTp) const {
+  // We only estimate the cost of reverse shuffles.
+  if (Kind != SK_Reverse)
+    return TargetTransformInfo::getShuffleCost(Kind, Tp, Index, SubTp);
+
+  std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Tp);
+  unsigned Cost = 1;
+  if (LT.second.getSizeInBits() > 128)
+    Cost = 3; // Extract + insert + copy.
+
+  // Multiple by the number of parts.
+  return Cost * LT.first;
+}
+
+unsigned X86TTI::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) const {
+  int ISD = TLI->InstructionOpcodeToISD(Opcode);
+  assert(ISD && "Invalid opcode");
+
+  EVT SrcTy = TLI->getValueType(Src);
+  EVT DstTy = TLI->getValueType(Dst);
+
+  if (!SrcTy.isSimple() || !DstTy.isSimple())
+    return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
+
+  static const TypeConversionCostTblEntry<MVT> AVXConversionTbl[] = {
+    { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i16, 1 },
+    { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i16, 1 },
+    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i32, 1 },
+    { ISD::ZERO_EXTEND, MVT::v4i64, MVT::v4i32, 1 },
+    { ISD::TRUNCATE,    MVT::v4i32, MVT::v4i64, 1 },
+    { ISD::TRUNCATE,    MVT::v8i16, MVT::v8i32, 1 },
+
+    { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i1,  8 },
+    { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i8,  8 },
+    { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i16, 5 },
+    { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i32, 1 },
+    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i1,  3 },
+    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i8,  3 },
+    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i16, 3 },
+    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i32, 1 },
+    { ISD::SINT_TO_FP,  MVT::v4f64, MVT::v4i1,  3 },
+    { ISD::SINT_TO_FP,  MVT::v4f64, MVT::v4i8,  3 },
+    { ISD::SINT_TO_FP,  MVT::v4f64, MVT::v4i16, 3 },
+    { ISD::SINT_TO_FP,  MVT::v4f64, MVT::v4i32, 1 },
+
+    { ISD::UINT_TO_FP,  MVT::v8f32, MVT::v8i1,  6 },
+    { ISD::UINT_TO_FP,  MVT::v8f32, MVT::v8i8,  5 },
+    { ISD::UINT_TO_FP,  MVT::v8f32, MVT::v8i16, 5 },
+    { ISD::UINT_TO_FP,  MVT::v8f32, MVT::v8i32, 9 },
+    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i1,  7 },
+    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i8,  2 },
+    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i16, 2 },
+    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i32, 6 },
+    { ISD::UINT_TO_FP,  MVT::v4f64, MVT::v4i1,  7 },
+    { ISD::UINT_TO_FP,  MVT::v4f64, MVT::v4i8,  2 },
+    { ISD::UINT_TO_FP,  MVT::v4f64, MVT::v4i16, 2 },
+    { ISD::UINT_TO_FP,  MVT::v4f64, MVT::v4i32, 6 },
+
+    { ISD::FP_TO_SINT,  MVT::v8i8,  MVT::v8f32, 1 },
+    { ISD::FP_TO_SINT,  MVT::v4i8,  MVT::v4f32, 1 },
+    { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i1,  6 },
+    { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i1,  9 },
+    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i1,  8 },
+    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i8,  6 },
+    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i16, 6 },
+    { ISD::TRUNCATE,    MVT::v8i32, MVT::v8i64, 3 },
+  };
+
+  if (ST->hasAVX()) {
+    int Idx = ConvertCostTableLookup<MVT>(AVXConversionTbl,
+                                 array_lengthof(AVXConversionTbl),
+                                 ISD, DstTy.getSimpleVT(), SrcTy.getSimpleVT());
+    if (Idx != -1)
+      return AVXConversionTbl[Idx].Cost;
+  }
+
+  return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
+}
+
+unsigned X86TTI::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
+                                    Type *CondTy) const {
+  // Legalize the type.
+  std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(ValTy);
+
+  MVT MTy = LT.second;
+
+  int ISD = TLI->InstructionOpcodeToISD(Opcode);
+  assert(ISD && "Invalid opcode");
+
+  static const CostTblEntry<MVT> SSE42CostTbl[] = {
+    { ISD::SETCC,   MVT::v2f64,   1 },
+    { ISD::SETCC,   MVT::v4f32,   1 },
+    { ISD::SETCC,   MVT::v2i64,   1 },
+    { ISD::SETCC,   MVT::v4i32,   1 },
+    { ISD::SETCC,   MVT::v8i16,   1 },
+    { ISD::SETCC,   MVT::v16i8,   1 },
+  };
+
+  static const CostTblEntry<MVT> AVX1CostTbl[] = {
+    { ISD::SETCC,   MVT::v4f64,   1 },
+    { ISD::SETCC,   MVT::v8f32,   1 },
+    // AVX1 does not support 8-wide integer compare.
+    { ISD::SETCC,   MVT::v4i64,   4 },
+    { ISD::SETCC,   MVT::v8i32,   4 },
+    { ISD::SETCC,   MVT::v16i16,  4 },
+    { ISD::SETCC,   MVT::v32i8,   4 },
+  };
+
+  static const CostTblEntry<MVT> AVX2CostTbl[] = {
+    { ISD::SETCC,   MVT::v4i64,   1 },
+    { ISD::SETCC,   MVT::v8i32,   1 },
+    { ISD::SETCC,   MVT::v16i16,  1 },
+    { ISD::SETCC,   MVT::v32i8,   1 },
+  };
+
+  if (ST->hasAVX2()) {
+    int Idx = CostTableLookup<MVT>(AVX2CostTbl, array_lengthof(AVX2CostTbl), ISD, MTy);
+    if (Idx != -1)
+      return LT.first * AVX2CostTbl[Idx].Cost;
+  }
+
+  if (ST->hasAVX()) {
+    int Idx = CostTableLookup<MVT>(AVX1CostTbl, array_lengthof(AVX1CostTbl), ISD, MTy);
+    if (Idx != -1)
+      return LT.first * AVX1CostTbl[Idx].Cost;
+  }
+
+  if (ST->hasSSE42()) {
+    int Idx = CostTableLookup<MVT>(SSE42CostTbl, array_lengthof(SSE42CostTbl), ISD, MTy);
+    if (Idx != -1)
+      return LT.first * SSE42CostTbl[Idx].Cost;
+  }
+
+  return TargetTransformInfo::getCmpSelInstrCost(Opcode, ValTy, CondTy);
+}
+
+unsigned X86TTI::getVectorInstrCost(unsigned Opcode, Type *Val,
+                                    unsigned Index) const {
+  assert(Val->isVectorTy() && "This must be a vector type");
+
+  if (Index != -1U) {
+    // Legalize the type.
+    std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Val);
+
+    // This type is legalized to a scalar type.
+    if (!LT.second.isVector())
+      return 0;
+
+    // The type may be split. Normalize the index to the new type.
+    unsigned Width = LT.second.getVectorNumElements();
+    Index = Index % Width;
+
+    // Floating point scalars are already located in index #0.
+    if (Val->getScalarType()->isFloatingPointTy() && Index == 0)
+      return 0;
+  }
+
+  return TargetTransformInfo::getVectorInstrCost(Opcode, Val, Index);
+}
+
+unsigned X86TTI::getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
+                                 unsigned AddressSpace) const {
+  // Legalize the type.
+  std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Src);
+  assert((Opcode == Instruction::Load || Opcode == Instruction::Store) &&
+         "Invalid Opcode");
+
+  // Each load/store unit costs 1.
+  unsigned Cost = LT.first * 1;
+
+  // On Sandybridge 256bit load/stores are double pumped
+  // (but not on Haswell).
+  if (LT.second.getSizeInBits() > 128 && !ST->hasAVX2())
+    Cost*=2;
+
+  return Cost;
+}
diff --git a/lib/Target/X86/X86VZeroUpper.cpp b/lib/Target/X86/X86VZeroUpper.cpp
index c4a58874a414..0f77948c0eff 100644
--- a/lib/Target/X86/X86VZeroUpper.cpp
+++ b/lib/Target/X86/X86VZeroUpper.cpp
@@ -120,9 +120,19 @@ static bool checkFnHasLiveInYmm(MachineRegisterInfo &MRI) {
   return false;
 }
 
+static bool clobbersAllYmmRegs(const MachineOperand &MO) {
+  for (unsigned reg = X86::YMM0; reg < X86::YMM15; ++reg) {
+    if (!MO.clobbersPhysReg(reg))
+      return false;
+  }
+  return true;
+}
+
 static bool hasYmmReg(MachineInstr *MI) {
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     const MachineOperand &MO = MI->getOperand(i);
+    if (MI->isCall() && MO.isRegMask() && !clobbersAllYmmRegs(MO))
+      return true;
     if (!MO.isReg())
       continue;
     if (MO.isDebug())
diff --git a/lib/Target/XCore/CMakeLists.txt b/lib/Target/XCore/CMakeLists.txt
index ca94f03a6496..099ad390d2a7 100644
--- a/lib/Target/XCore/CMakeLists.txt
+++ b/lib/Target/XCore/CMakeLists.txt
@@ -2,6 +2,7 @@ set(LLVM_TARGET_DEFINITIONS XCore.td)
 
 tablegen(LLVM XCoreGenRegisterInfo.inc -gen-register-info)
 tablegen(LLVM XCoreGenInstrInfo.inc -gen-instr-info)
+tablegen(LLVM XCoreGenDisassemblerTables.inc -gen-disassembler)
 tablegen(LLVM XCoreGenAsmWriter.inc -gen-asm-writer)
 tablegen(LLVM XCoreGenDAGISel.inc -gen-dag-isel)
 tablegen(LLVM XCoreGenCallingConv.inc -gen-callingconv)
@@ -15,6 +16,7 @@ add_llvm_target(XCoreCodeGen
   XCoreISelDAGToDAG.cpp
   XCoreISelLowering.cpp
   XCoreMachineFunctionInfo.cpp
+  XCoreMCInstLower.cpp
   XCoreRegisterInfo.cpp
   XCoreSubtarget.cpp
   XCoreTargetMachine.cpp
@@ -24,5 +26,7 @@ add_llvm_target(XCoreCodeGen
 
 add_dependencies(LLVMXCoreCodeGen intrinsics_gen)
 
+add_subdirectory(Disassembler)
+add_subdirectory(InstPrinter)
 add_subdirectory(TargetInfo)
 add_subdirectory(MCTargetDesc)
diff --git a/lib/Target/XCore/Disassembler/CMakeLists.txt b/lib/Target/XCore/Disassembler/CMakeLists.txt
new file mode 100644
index 000000000000..cdc5d993b8bf
--- /dev/null
+++ b/lib/Target/XCore/Disassembler/CMakeLists.txt
@@ -0,0 +1,5 @@
+add_llvm_library(LLVMXCoreDisassembler
+  XCoreDisassembler.cpp
+  )
+
+add_dependencies(LLVMXCoreDisassembler XCoreCommonTableGen)
diff --git a/lib/Target/XCore/Disassembler/LLVMBuild.txt b/lib/Target/XCore/Disassembler/LLVMBuild.txt
new file mode 100644
index 000000000000..028de2cb3433
--- /dev/null
+++ b/lib/Target/XCore/Disassembler/LLVMBuild.txt
@@ -0,0 +1,23 @@
+;===- ./lib/Target/XCore/Disassembler/LLVMBuild.txt ------------*- Conf -*--===;
+;
+;                     The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+;   http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[component_0]
+type = Library
+name = XCoreDisassembler
+parent = XCore
+required_libraries = MC Support XCoreInfo
+add_to_library_groups = XCore
diff --git a/lib/Target/XCore/Disassembler/Makefile b/lib/Target/XCore/Disassembler/Makefile
new file mode 100644
index 000000000000..4caffdd1da6a
--- /dev/null
+++ b/lib/Target/XCore/Disassembler/Makefile
@@ -0,0 +1,16 @@
+##===- lib/Target/XCore/Disassembler/Makefile --------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+
+LEVEL = ../../../..
+LIBRARYNAME = LLVMXCoreDisassembler
+
+# Hack: we need to include 'main' XCore target directory to grab private headers
+CPP.Flags += -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/Target/XCore/Disassembler/XCoreDisassembler.cpp b/lib/Target/XCore/Disassembler/XCoreDisassembler.cpp
new file mode 100644
index 000000000000..7b99967c4f32
--- /dev/null
+++ b/lib/Target/XCore/Disassembler/XCoreDisassembler.cpp
@@ -0,0 +1,800 @@
+//===- XCoreDisassembler.cpp - Disassembler for XCore -----------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// \brief This file is part of the XCore Disassembler.
+///
+//===----------------------------------------------------------------------===//
+
+#include "XCore.h"
+#include "XCoreRegisterInfo.h"
+#include "llvm/MC/MCDisassembler.h"
+#include "llvm/MC/MCFixedLenDisassembler.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/Support/MemoryObject.h"
+#include "llvm/Support/TargetRegistry.h"
+
+using namespace llvm;
+
+typedef MCDisassembler::DecodeStatus DecodeStatus;
+
+namespace {
+
+/// \brief A disassembler class for XCore.
+class XCoreDisassembler : public MCDisassembler {
+  const MCRegisterInfo *RegInfo;
+public:
+  XCoreDisassembler(const MCSubtargetInfo &STI, const MCRegisterInfo *Info) :
+    MCDisassembler(STI), RegInfo(Info) {}
+
+  /// \brief See MCDisassembler.
+  virtual DecodeStatus getInstruction(MCInst &instr,
+                                      uint64_t &size,
+                                      const MemoryObject &region,
+                                      uint64_t address,
+                                      raw_ostream &vStream,
+                                      raw_ostream &cStream) const;
+
+  const MCRegisterInfo *getRegInfo() const { return RegInfo; }
+};
+}
+
+static bool readInstruction16(const MemoryObject &region,
+                              uint64_t address,
+                              uint64_t &size,
+                              uint16_t &insn) {
+  uint8_t Bytes[4];
+
+  // We want to read exactly 2 Bytes of data.
+  if (region.readBytes(address, 2, Bytes, NULL) == -1) {
+    size = 0;
+    return false;
+  }
+  // Encoded as a little-endian 16-bit word in the stream.
+  insn = (Bytes[0] <<  0) | (Bytes[1] <<  8);
+  return true;
+}
+
+static bool readInstruction32(const MemoryObject &region,
+                              uint64_t address,
+                              uint64_t &size,
+                              uint32_t &insn) {
+  uint8_t Bytes[4];
+
+  // We want to read exactly 4 Bytes of data.
+  if (region.readBytes(address, 4, Bytes, NULL) == -1) {
+    size = 0;
+    return false;
+  }
+  // Encoded as a little-endian 32-bit word in the stream.
+  insn = (Bytes[0] << 0) | (Bytes[1] << 8) | (Bytes[2] << 16) |
+         (Bytes[3] << 24);
+  return true;
+}
+
+static unsigned getReg(const void *D, unsigned RC, unsigned RegNo) {
+  const XCoreDisassembler *Dis = static_cast<const XCoreDisassembler*>(D);
+  return *(Dis->getRegInfo()->getRegClass(RC).begin() + RegNo);
+}
+
+static DecodeStatus DecodeGRRegsRegisterClass(MCInst &Inst,
+                                              unsigned RegNo,
+                                              uint64_t Address,
+                                              const void *Decoder);
+
+static DecodeStatus DecodeRRegsRegisterClass(MCInst &Inst,
+                                             unsigned RegNo,
+                                             uint64_t Address,
+                                             const void *Decoder);
+
+static DecodeStatus DecodeBitpOperand(MCInst &Inst, unsigned Val,
+                                      uint64_t Address, const void *Decoder);
+
+static DecodeStatus DecodeMEMiiOperand(MCInst &Inst, unsigned Val,
+                                       uint64_t Address, const void *Decoder);
+
+static DecodeStatus Decode2RInstruction(MCInst &Inst,
+                                        unsigned Insn,
+                                        uint64_t Address,
+                                        const void *Decoder);
+
+static DecodeStatus Decode2RImmInstruction(MCInst &Inst,
+                                           unsigned Insn,
+                                           uint64_t Address,
+                                           const void *Decoder);
+
+static DecodeStatus DecodeR2RInstruction(MCInst &Inst,
+                                         unsigned Insn,
+                                         uint64_t Address,
+                                         const void *Decoder);
+
+static DecodeStatus Decode2RSrcDstInstruction(MCInst &Inst,
+                                              unsigned Insn,
+                                              uint64_t Address,
+                                              const void *Decoder);
+
+static DecodeStatus DecodeRUSInstruction(MCInst &Inst,
+                                         unsigned Insn,
+                                         uint64_t Address,
+                                         const void *Decoder);
+
+static DecodeStatus DecodeRUSBitpInstruction(MCInst &Inst,
+                                             unsigned Insn,
+                                             uint64_t Address,
+                                             const void *Decoder);
+
+static DecodeStatus DecodeRUSSrcDstBitpInstruction(MCInst &Inst,
+                                                   unsigned Insn,
+                                                   uint64_t Address,
+                                                   const void *Decoder);
+
+static DecodeStatus DecodeL2RInstruction(MCInst &Inst,
+                                         unsigned Insn,
+                                         uint64_t Address,
+                                         const void *Decoder);
+
+static DecodeStatus DecodeLR2RInstruction(MCInst &Inst,
+                                          unsigned Insn,
+                                          uint64_t Address,
+                                          const void *Decoder);
+
+static DecodeStatus Decode3RInstruction(MCInst &Inst,
+                                        unsigned Insn,
+                                        uint64_t Address,
+                                        const void *Decoder);
+
+static DecodeStatus Decode3RImmInstruction(MCInst &Inst,
+                                           unsigned Insn,
+                                           uint64_t Address,
+                                           const void *Decoder);
+
+static DecodeStatus Decode2RUSInstruction(MCInst &Inst,
+                                          unsigned Insn,
+                                          uint64_t Address,
+                                          const void *Decoder);
+
+static DecodeStatus Decode2RUSBitpInstruction(MCInst &Inst,
+                                              unsigned Insn,
+                                              uint64_t Address,
+                                              const void *Decoder);
+
+static DecodeStatus DecodeL3RInstruction(MCInst &Inst,
+                                         unsigned Insn,
+                                         uint64_t Address,
+                                         const void *Decoder);
+
+static DecodeStatus DecodeL3RSrcDstInstruction(MCInst &Inst,
+                                               unsigned Insn,
+                                               uint64_t Address,
+                                               const void *Decoder);
+
+static DecodeStatus DecodeL2RUSInstruction(MCInst &Inst,
+                                           unsigned Insn,
+                                           uint64_t Address,
+                                           const void *Decoder);
+
+static DecodeStatus DecodeL2RUSBitpInstruction(MCInst &Inst,
+                                               unsigned Insn,
+                                               uint64_t Address,
+                                               const void *Decoder);
+
+static DecodeStatus DecodeL6RInstruction(MCInst &Inst,
+                                         unsigned Insn,
+                                         uint64_t Address,
+                                         const void *Decoder);
+
+static DecodeStatus DecodeL5RInstruction(MCInst &Inst,
+                                         unsigned Insn,
+                                         uint64_t Address,
+                                         const void *Decoder);
+
+static DecodeStatus DecodeL4RSrcDstInstruction(MCInst &Inst,
+                                               unsigned Insn,
+                                               uint64_t Address,
+                                               const void *Decoder);
+
+static DecodeStatus DecodeL4RSrcDstSrcDstInstruction(MCInst &Inst,
+                                                     unsigned Insn,
+                                                     uint64_t Address,
+                                                     const void *Decoder);
+
+#include "XCoreGenDisassemblerTables.inc"
+
+static DecodeStatus DecodeGRRegsRegisterClass(MCInst &Inst,
+                                              unsigned RegNo,
+                                              uint64_t Address,
+                                              const void *Decoder)
+{
+  if (RegNo > 11)
+    return MCDisassembler::Fail;
+  unsigned Reg = getReg(Decoder, XCore::GRRegsRegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Reg));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus DecodeRRegsRegisterClass(MCInst &Inst,
+                                             unsigned RegNo,
+                                             uint64_t Address,
+                                             const void *Decoder)
+{
+  if (RegNo > 15)
+    return MCDisassembler::Fail;
+  unsigned Reg = getReg(Decoder, XCore::RRegsRegClassID, RegNo);
+  Inst.addOperand(MCOperand::CreateReg(Reg));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus DecodeBitpOperand(MCInst &Inst, unsigned Val,
+                                      uint64_t Address, const void *Decoder) {
+  if (Val > 11)
+    return MCDisassembler::Fail;
+  static unsigned Values[] = {
+    32 /*bpw*/, 1, 2, 3, 4, 5, 6, 7, 8, 16, 24, 32
+  };
+  Inst.addOperand(MCOperand::CreateImm(Values[Val]));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus DecodeMEMiiOperand(MCInst &Inst, unsigned Val,
+                                       uint64_t Address, const void *Decoder) {
+  Inst.addOperand(MCOperand::CreateImm(Val));
+  Inst.addOperand(MCOperand::CreateImm(0));
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus
+Decode2OpInstruction(unsigned Insn, unsigned &Op1, unsigned &Op2) {
+  unsigned Combined = fieldFromInstruction(Insn, 6, 5);
+  if (Combined < 27)
+    return MCDisassembler::Fail;
+  if (fieldFromInstruction(Insn, 5, 1)) {
+    if (Combined == 31)
+      return MCDisassembler::Fail;
+    Combined += 5;
+  }
+  Combined -= 27;
+  unsigned Op1High = Combined % 3;
+  unsigned Op2High = Combined / 3;
+  Op1 = (Op1High << 2) | fieldFromInstruction(Insn, 2, 2);
+  Op2 = (Op2High << 2) | fieldFromInstruction(Insn, 0, 2);
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus
+Decode3OpInstruction(unsigned Insn, unsigned &Op1, unsigned &Op2,
+                     unsigned &Op3) {
+  unsigned Combined = fieldFromInstruction(Insn, 6, 5);
+  if (Combined >= 27)
+    return MCDisassembler::Fail;
+
+  unsigned Op1High = Combined % 3;
+  unsigned Op2High = (Combined / 3) % 3;
+  unsigned Op3High = Combined / 9;
+  Op1 = (Op1High << 2) | fieldFromInstruction(Insn, 4, 2);
+  Op2 = (Op2High << 2) | fieldFromInstruction(Insn, 2, 2);
+  Op3 = (Op3High << 2) | fieldFromInstruction(Insn, 0, 2);
+  return MCDisassembler::Success;
+}
+
+static DecodeStatus
+Decode2OpInstructionFail(MCInst &Inst, unsigned Insn, uint64_t Address,
+                         const void *Decoder) {
+  // Try and decode as a 3R instruction.
+  unsigned Opcode = fieldFromInstruction(Insn, 11, 5);
+  switch (Opcode) {
+  case 0x0:
+    Inst.setOpcode(XCore::STW_2rus);
+    return Decode2RUSInstruction(Inst, Insn, Address, Decoder);
+  case 0x1:
+    Inst.setOpcode(XCore::LDW_2rus);
+    return Decode2RUSInstruction(Inst, Insn, Address, Decoder);
+  case 0x2:
+    Inst.setOpcode(XCore::ADD_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x3:
+    Inst.setOpcode(XCore::SUB_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x4:
+    Inst.setOpcode(XCore::SHL_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x5:
+    Inst.setOpcode(XCore::SHR_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x6:
+    Inst.setOpcode(XCore::EQ_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x7:
+    Inst.setOpcode(XCore::AND_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x8:
+    Inst.setOpcode(XCore::OR_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x9:
+    Inst.setOpcode(XCore::LDW_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x10:
+    Inst.setOpcode(XCore::LD16S_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x11:
+    Inst.setOpcode(XCore::LD8U_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x12:
+    Inst.setOpcode(XCore::ADD_2rus);
+    return Decode2RUSInstruction(Inst, Insn, Address, Decoder);
+  case 0x13:
+    Inst.setOpcode(XCore::SUB_2rus);
+    return Decode2RUSInstruction(Inst, Insn, Address, Decoder);
+  case 0x14:
+    Inst.setOpcode(XCore::SHL_2rus);
+    return Decode2RUSBitpInstruction(Inst, Insn, Address, Decoder);
+  case 0x15:
+    Inst.setOpcode(XCore::SHR_2rus);
+    return Decode2RUSBitpInstruction(Inst, Insn, Address, Decoder);
+  case 0x16:
+    Inst.setOpcode(XCore::EQ_2rus);
+    return Decode2RUSInstruction(Inst, Insn, Address, Decoder);
+  case 0x17:
+    Inst.setOpcode(XCore::TSETR_3r);
+    return Decode3RImmInstruction(Inst, Insn, Address, Decoder);
+  case 0x18:
+    Inst.setOpcode(XCore::LSS_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x19:
+    Inst.setOpcode(XCore::LSU_3r);
+    return Decode3RInstruction(Inst, Insn, Address, Decoder);
+  }
+  return MCDisassembler::Fail;
+}
+
+static DecodeStatus
+Decode2RInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                    const void *Decoder) {
+  unsigned Op1, Op2;
+  DecodeStatus S = Decode2OpInstruction(Insn, Op1, Op2);
+  if (S != MCDisassembler::Success)
+    return Decode2OpInstructionFail(Inst, Insn, Address, Decoder);
+
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+  return S;
+}
+
+static DecodeStatus
+Decode2RImmInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                       const void *Decoder) {
+  unsigned Op1, Op2;
+  DecodeStatus S = Decode2OpInstruction(Insn, Op1, Op2);
+  if (S != MCDisassembler::Success)
+    return Decode2OpInstructionFail(Inst, Insn, Address, Decoder);
+
+  Inst.addOperand(MCOperand::CreateImm(Op1));
+  DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+  return S;
+}
+
+static DecodeStatus
+DecodeR2RInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                     const void *Decoder) {
+  unsigned Op1, Op2;
+  DecodeStatus S = Decode2OpInstruction(Insn, Op2, Op1);
+  if (S != MCDisassembler::Success)
+    return Decode2OpInstructionFail(Inst, Insn, Address, Decoder);
+
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+  return S;
+}
+
+static DecodeStatus
+Decode2RSrcDstInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                          const void *Decoder) {
+  unsigned Op1, Op2;
+  DecodeStatus S = Decode2OpInstruction(Insn, Op1, Op2);
+  if (S != MCDisassembler::Success)
+    return Decode2OpInstructionFail(Inst, Insn, Address, Decoder);
+
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+  return S;
+}
+
+static DecodeStatus
+DecodeRUSInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                     const void *Decoder) {
+  unsigned Op1, Op2;
+  DecodeStatus S = Decode2OpInstruction(Insn, Op1, Op2);
+  if (S != MCDisassembler::Success)
+    return Decode2OpInstructionFail(Inst, Insn, Address, Decoder);
+
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  Inst.addOperand(MCOperand::CreateImm(Op2));
+  return S;
+}
+
+static DecodeStatus
+DecodeRUSBitpInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                         const void *Decoder) {
+  unsigned Op1, Op2;
+  DecodeStatus S = Decode2OpInstruction(Insn, Op1, Op2);
+  if (S != MCDisassembler::Success)
+    return Decode2OpInstructionFail(Inst, Insn, Address, Decoder);
+
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  DecodeBitpOperand(Inst, Op2, Address, Decoder);
+  return S;
+}
+
+static DecodeStatus
+DecodeRUSSrcDstBitpInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                               const void *Decoder) {
+  unsigned Op1, Op2;
+  DecodeStatus S = Decode2OpInstruction(Insn, Op1, Op2);
+  if (S != MCDisassembler::Success)
+    return Decode2OpInstructionFail(Inst, Insn, Address, Decoder);
+
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  DecodeBitpOperand(Inst, Op2, Address, Decoder);
+  return S;
+}
+
+static DecodeStatus
+DecodeL2OpInstructionFail(MCInst &Inst, unsigned Insn, uint64_t Address,
+                          const void *Decoder) {
+  // Try and decode as a L3R / L2RUS instruction.
+  unsigned Opcode = fieldFromInstruction(Insn, 16, 4) |
+                    fieldFromInstruction(Insn, 27, 5) << 4;
+  switch (Opcode) {
+  case 0x0c:
+    Inst.setOpcode(XCore::STW_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x1c:
+    Inst.setOpcode(XCore::XOR_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x2c:
+    Inst.setOpcode(XCore::ASHR_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x3c:
+    Inst.setOpcode(XCore::LDAWF_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x4c:
+    Inst.setOpcode(XCore::LDAWB_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x5c:
+    Inst.setOpcode(XCore::LDA16F_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x6c:
+    Inst.setOpcode(XCore::LDA16B_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x7c:
+    Inst.setOpcode(XCore::MUL_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x8c:
+    Inst.setOpcode(XCore::DIVS_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x9c:
+    Inst.setOpcode(XCore::DIVU_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x10c:
+    Inst.setOpcode(XCore::ST16_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x11c:
+    Inst.setOpcode(XCore::ST8_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x12c:
+    Inst.setOpcode(XCore::ASHR_l2rus);
+    return DecodeL2RUSBitpInstruction(Inst, Insn, Address, Decoder);
+  case 0x12d:
+    Inst.setOpcode(XCore::OUTPW_l2rus);
+    return DecodeL2RUSBitpInstruction(Inst, Insn, Address, Decoder);
+  case 0x12e:
+    Inst.setOpcode(XCore::INPW_l2rus);
+    return DecodeL2RUSBitpInstruction(Inst, Insn, Address, Decoder);
+  case 0x13c:
+    Inst.setOpcode(XCore::LDAWF_l2rus);
+    return DecodeL2RUSInstruction(Inst, Insn, Address, Decoder);
+  case 0x14c:
+    Inst.setOpcode(XCore::LDAWB_l2rus);
+    return DecodeL2RUSInstruction(Inst, Insn, Address, Decoder);
+  case 0x15c:
+    Inst.setOpcode(XCore::CRC_l3r);
+    return DecodeL3RSrcDstInstruction(Inst, Insn, Address, Decoder);
+  case 0x18c:
+    Inst.setOpcode(XCore::REMS_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  case 0x19c:
+    Inst.setOpcode(XCore::REMU_l3r);
+    return DecodeL3RInstruction(Inst, Insn, Address, Decoder);
+  }
+  return MCDisassembler::Fail;
+}
+
+static DecodeStatus
+DecodeL2RInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                               const void *Decoder) {
+  unsigned Op1, Op2;
+  DecodeStatus S = Decode2OpInstruction(fieldFromInstruction(Insn, 0, 16),
+                                        Op1, Op2);
+  if (S != MCDisassembler::Success)
+    return DecodeL2OpInstructionFail(Inst, Insn, Address, Decoder);
+
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+  return S;
+}
+
+static DecodeStatus
+DecodeLR2RInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                               const void *Decoder) {
+  unsigned Op1, Op2;
+  DecodeStatus S = Decode2OpInstruction(fieldFromInstruction(Insn, 0, 16),
+                                        Op1, Op2);
+  if (S != MCDisassembler::Success)
+    return DecodeL2OpInstructionFail(Inst, Insn, Address, Decoder);
+
+  DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  return S;
+}
+
+static DecodeStatus
+Decode3RInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                    const void *Decoder) {
+  unsigned Op1, Op2, Op3;
+  DecodeStatus S = Decode3OpInstruction(Insn, Op1, Op2, Op3);
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op3, Address, Decoder);
+  }
+  return S;
+}
+
+static DecodeStatus
+Decode3RImmInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                       const void *Decoder) {
+  unsigned Op1, Op2, Op3;
+  DecodeStatus S = Decode3OpInstruction(Insn, Op1, Op2, Op3);
+  if (S == MCDisassembler::Success) {
+    Inst.addOperand(MCOperand::CreateImm(Op1));
+    DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op3, Address, Decoder);
+  }
+  return S;
+}
+
+static DecodeStatus
+Decode2RUSInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                      const void *Decoder) {
+  unsigned Op1, Op2, Op3;
+  DecodeStatus S = Decode3OpInstruction(Insn, Op1, Op2, Op3);
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+    Inst.addOperand(MCOperand::CreateImm(Op3));
+  }
+  return S;
+}
+
+static DecodeStatus
+Decode2RUSBitpInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                      const void *Decoder) {
+  unsigned Op1, Op2, Op3;
+  DecodeStatus S = Decode3OpInstruction(Insn, Op1, Op2, Op3);
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+    DecodeBitpOperand(Inst, Op3, Address, Decoder);
+  }
+  return S;
+}
+
+static DecodeStatus
+DecodeL3RInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                     const void *Decoder) {
+  unsigned Op1, Op2, Op3;
+  DecodeStatus S =
+    Decode3OpInstruction(fieldFromInstruction(Insn, 0, 16), Op1, Op2, Op3);
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op3, Address, Decoder);
+  }
+  return S;
+}
+
+static DecodeStatus
+DecodeL3RSrcDstInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                           const void *Decoder) {
+  unsigned Op1, Op2, Op3;
+  DecodeStatus S =
+  Decode3OpInstruction(fieldFromInstruction(Insn, 0, 16), Op1, Op2, Op3);
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op3, Address, Decoder);
+  }
+  return S;
+}
+
+static DecodeStatus
+DecodeL2RUSInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                       const void *Decoder) {
+  unsigned Op1, Op2, Op3;
+  DecodeStatus S =
+  Decode3OpInstruction(fieldFromInstruction(Insn, 0, 16), Op1, Op2, Op3);
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+    Inst.addOperand(MCOperand::CreateImm(Op3));
+  }
+  return S;
+}
+
+static DecodeStatus
+DecodeL2RUSBitpInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                           const void *Decoder) {
+  unsigned Op1, Op2, Op3;
+  DecodeStatus S =
+  Decode3OpInstruction(fieldFromInstruction(Insn, 0, 16), Op1, Op2, Op3);
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+    DecodeBitpOperand(Inst, Op3, Address, Decoder);
+  }
+  return S;
+}
+
+static DecodeStatus
+DecodeL6RInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                     const void *Decoder) {
+  unsigned Op1, Op2, Op3, Op4, Op5, Op6;
+  DecodeStatus S =
+    Decode3OpInstruction(fieldFromInstruction(Insn, 0, 16), Op1, Op2, Op3);
+  if (S != MCDisassembler::Success)
+    return S;
+  S = Decode3OpInstruction(fieldFromInstruction(Insn, 16, 16), Op4, Op5, Op6);
+  if (S != MCDisassembler::Success)
+    return S;
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op4, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op3, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op5, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op6, Address, Decoder);
+  return S;
+}
+
+static DecodeStatus
+DecodeL5RInstructionFail(MCInst &Inst, unsigned Insn, uint64_t Address,
+                     const void *Decoder) {
+  // Try and decode as a L6R instruction.
+  Inst.clear();
+  unsigned Opcode = fieldFromInstruction(Insn, 27, 5);
+  switch (Opcode) {
+  case 0x00:
+    Inst.setOpcode(XCore::LMUL_l6r);
+    return DecodeL6RInstruction(Inst, Insn, Address, Decoder);
+  }
+  return MCDisassembler::Fail;
+}
+
+static DecodeStatus
+DecodeL5RInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                     const void *Decoder) {
+  unsigned Op1, Op2, Op3, Op4, Op5;
+  DecodeStatus S =
+    Decode3OpInstruction(fieldFromInstruction(Insn, 0, 16), Op1, Op2, Op3);
+  if (S != MCDisassembler::Success)
+    return DecodeL5RInstructionFail(Inst, Insn, Address, Decoder);
+  S = Decode2OpInstruction(fieldFromInstruction(Insn, 16, 16), Op4, Op5);
+  if (S != MCDisassembler::Success)
+    return DecodeL5RInstructionFail(Inst, Insn, Address, Decoder);
+
+  DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op4, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op3, Address, Decoder);
+  DecodeGRRegsRegisterClass(Inst, Op5, Address, Decoder);
+  return S;
+}
+
+static DecodeStatus
+DecodeL4RSrcDstInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                           const void *Decoder) {
+  unsigned Op1, Op2, Op3;
+  unsigned Op4 = fieldFromInstruction(Insn, 16, 4);
+  DecodeStatus S =
+    Decode3OpInstruction(fieldFromInstruction(Insn, 0, 16), Op1, Op2, Op3);
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    S = DecodeGRRegsRegisterClass(Inst, Op4, Address, Decoder);
+  }
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op4, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op3, Address, Decoder);
+  }
+  return S;
+}
+
+static DecodeStatus
+DecodeL4RSrcDstSrcDstInstruction(MCInst &Inst, unsigned Insn, uint64_t Address,
+                                 const void *Decoder) {
+  unsigned Op1, Op2, Op3;
+  unsigned Op4 = fieldFromInstruction(Insn, 16, 4);
+  DecodeStatus S =
+  Decode3OpInstruction(fieldFromInstruction(Insn, 0, 16), Op1, Op2, Op3);
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    S = DecodeGRRegsRegisterClass(Inst, Op4, Address, Decoder);
+  }
+  if (S == MCDisassembler::Success) {
+    DecodeGRRegsRegisterClass(Inst, Op1, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op4, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op2, Address, Decoder);
+    DecodeGRRegsRegisterClass(Inst, Op3, Address, Decoder);
+  }
+  return S;
+}
+
+MCDisassembler::DecodeStatus
+XCoreDisassembler::getInstruction(MCInst &instr,
+                                  uint64_t &Size,
+                                  const MemoryObject &Region,
+                                  uint64_t Address,
+                                  raw_ostream &vStream,
+                                  raw_ostream &cStream) const {
+  uint16_t insn16;
+
+  if (!readInstruction16(Region, Address, Size, insn16)) {
+    return Fail;
+  }
+
+  // Calling the auto-generated decoder function.
+  DecodeStatus Result = decodeInstruction(DecoderTable16, instr, insn16,
+                                          Address, this, STI);
+  if (Result != Fail) {
+    Size = 2;
+    return Result;
+  }
+
+  uint32_t insn32;
+
+  if (!readInstruction32(Region, Address, Size, insn32)) {
+    return Fail;
+  }
+
+  // Calling the auto-generated decoder function.
+  Result = decodeInstruction(DecoderTable32, instr, insn32, Address, this, STI);
+  if (Result != Fail) {
+    Size = 4;
+    return Result;
+  }
+
+  return Fail;
+}
+
+namespace llvm {
+  extern Target TheXCoreTarget;
+}
+
+static MCDisassembler *createXCoreDisassembler(const Target &T,
+                                               const MCSubtargetInfo &STI) {
+  return new XCoreDisassembler(STI, T.createMCRegInfo(""));
+}
+
+extern "C" void LLVMInitializeXCoreDisassembler() {
+  // Register the disassembler.
+  TargetRegistry::RegisterMCDisassembler(TheXCoreTarget,
+                                         createXCoreDisassembler);
+}
diff --git a/lib/Target/XCore/InstPrinter/CMakeLists.txt b/lib/Target/XCore/InstPrinter/CMakeLists.txt
new file mode 100644
index 000000000000..930e733cd7f1
--- /dev/null
+++ b/lib/Target/XCore/InstPrinter/CMakeLists.txt
@@ -0,0 +1,7 @@
+include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
+
+add_llvm_library(LLVMXCoreAsmPrinter
+  XCoreInstPrinter.cpp
+  )
+
+add_dependencies(LLVMXCoreAsmPrinter XCoreCommonTableGen)
diff --git a/lib/Target/XCore/InstPrinter/LLVMBuild.txt b/lib/Target/XCore/InstPrinter/LLVMBuild.txt
new file mode 100644
index 000000000000..8750bc7acedc
--- /dev/null
+++ b/lib/Target/XCore/InstPrinter/LLVMBuild.txt
@@ -0,0 +1,23 @@
+;===- ./lib/Target/XCore/InstPrinter/LLVMBuild.txt -------------*- Conf -*--===;
+;
+;                     The LLVM Compiler Infrastructure
+;
+; This file is distributed under the University of Illinois Open Source
+; License. See LICENSE.TXT for details.
+;
+;===------------------------------------------------------------------------===;
+;
+; This is an LLVMBuild description file for the components in this subdirectory.
+;
+; For more information on the LLVMBuild system, please see:
+;
+;   http://llvm.org/docs/LLVMBuild.html
+;
+;===------------------------------------------------------------------------===;
+
+[component_0]
+type = Library
+name = XCoreAsmPrinter
+parent = XCore
+required_libraries = MC Support
+add_to_library_groups = XCore
diff --git a/lib/Target/XCore/InstPrinter/Makefile b/lib/Target/XCore/InstPrinter/Makefile
new file mode 100644
index 000000000000..1c1c61299c39
--- /dev/null
+++ b/lib/Target/XCore/InstPrinter/Makefile
@@ -0,0 +1,16 @@
+##===- lib/Target/XCore/AsmPrinter/Makefile ----------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+
+LEVEL = ../../../..
+LIBRARYNAME = LLVMXCoreAsmPrinter
+
+# Hack: we need to include 'main' xcore target directory to grab private headers
+CPP.Flags += -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/..
+
+include $(LEVEL)/Makefile.common
diff --git a/lib/Target/XCore/InstPrinter/XCoreInstPrinter.cpp b/lib/Target/XCore/InstPrinter/XCoreInstPrinter.cpp
new file mode 100644
index 000000000000..1592351c3861
--- /dev/null
+++ b/lib/Target/XCore/InstPrinter/XCoreInstPrinter.cpp
@@ -0,0 +1,97 @@
+//===-- XCoreInstPrinter.cpp - Convert XCore MCInst to assembly syntax ----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class prints an XCore MCInst to a .s file.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "asm-printer"
+#include "XCoreInstPrinter.h"
+#include "llvm/ADT/StringExtras.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCSymbol.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+using namespace llvm;
+
+#include "XCoreGenAsmWriter.inc"
+
+void XCoreInstPrinter::printRegName(raw_ostream &OS, unsigned RegNo) const {
+  OS << StringRef(getRegisterName(RegNo)).lower();
+}
+
+void XCoreInstPrinter::printInst(const MCInst *MI, raw_ostream &O,
+                                 StringRef Annot) {
+  printInstruction(MI, O);
+  printAnnotation(O, Annot);
+}
+
+void XCoreInstPrinter::
+printInlineJT(const MCInst *MI, int opNum, raw_ostream &O) {
+  report_fatal_error("can't handle InlineJT");
+}
+
+void XCoreInstPrinter::
+printInlineJT32(const MCInst *MI, int opNum, raw_ostream &O) {
+  report_fatal_error("can't handle InlineJT32");
+}
+
+static void printExpr(const MCExpr *Expr, raw_ostream &OS) {
+  int Offset = 0;
+  const MCSymbolRefExpr *SRE;
+
+  if (const MCBinaryExpr *BE = dyn_cast<MCBinaryExpr>(Expr)) {
+    SRE = dyn_cast<MCSymbolRefExpr>(BE->getLHS());
+    const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(BE->getRHS());
+    assert(SRE && CE && "Binary expression must be sym+const.");
+    Offset = CE->getValue();
+  } else {
+    SRE = dyn_cast<MCSymbolRefExpr>(Expr);
+    assert(SRE && "Unexpected MCExpr type.");
+  }
+  assert(SRE->getKind() == MCSymbolRefExpr::VK_None);
+
+  OS << SRE->getSymbol();
+
+  if (Offset) {
+    if (Offset > 0)
+      OS << '+';
+    OS << Offset;
+  }
+}
+
+void XCoreInstPrinter::
+printOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O) {
+  const MCOperand &Op = MI->getOperand(OpNo);
+  if (Op.isReg()) {
+    printRegName(O, Op.getReg());
+    return;
+  }
+
+  if (Op.isImm()) {
+    O << Op.getImm();
+    return;
+  }
+
+  assert(Op.isExpr() && "unknown operand kind in printOperand");
+  printExpr(Op.getExpr(), O);
+}
+
+void XCoreInstPrinter::
+printMemOperand(const MCInst *MI, int opNum, raw_ostream &O) {
+  printOperand(MI, opNum, O);
+
+  if (MI->getOperand(opNum+1).isImm() && MI->getOperand(opNum+1).getImm() == 0)
+    return;
+
+  O << "+";
+  printOperand(MI, opNum+1, O);
+}
diff --git a/lib/Target/XCore/InstPrinter/XCoreInstPrinter.h b/lib/Target/XCore/InstPrinter/XCoreInstPrinter.h
new file mode 100644
index 000000000000..772c515b5c9e
--- /dev/null
+++ b/lib/Target/XCore/InstPrinter/XCoreInstPrinter.h
@@ -0,0 +1,44 @@
+//== XCoreInstPrinter.h - Convert XCore MCInst to assembly syntax -*- C++ -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// \brief This file contains the declaration of the XCoreInstPrinter class,
+/// which is used to print XCore MCInst to a .s file.
+///
+//===----------------------------------------------------------------------===//
+
+#ifndef XCOREINSTPRINTER_H
+#define XCOREINSTPRINTER_H
+#include "llvm/MC/MCInstPrinter.h"
+
+namespace llvm {
+
+class TargetMachine;
+
+class XCoreInstPrinter : public MCInstPrinter {
+public:
+  XCoreInstPrinter(const MCAsmInfo &MAI, const MCInstrInfo &MII,
+                  const MCRegisterInfo &MRI)
+    : MCInstPrinter(MAI, MII, MRI) {}
+
+  // Autogenerated by tblgen.
+  void printInstruction(const MCInst *MI, raw_ostream &O);
+  static const char *getRegisterName(unsigned RegNo);
+
+  virtual void printRegName(raw_ostream &OS, unsigned RegNo) const;
+  virtual void printInst(const MCInst *MI, raw_ostream &O, StringRef Annot);
+private:
+  void printInlineJT(const MCInst *MI, int opNum, raw_ostream &O);
+  void printInlineJT32(const MCInst *MI, int opNum, raw_ostream &O);
+  void printOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printMemOperand(const MCInst *MI, int opNum, raw_ostream &O);
+};
+} // end namespace llvm
+
+#endif
diff --git a/lib/Target/XCore/LLVMBuild.txt b/lib/Target/XCore/LLVMBuild.txt
index 53b4a9e3f5f7..59e64ad0855c 100644
--- a/lib/Target/XCore/LLVMBuild.txt
+++ b/lib/Target/XCore/LLVMBuild.txt
@@ -16,13 +16,14 @@
 ;===------------------------------------------------------------------------===;
 
 [common]
-subdirectories = MCTargetDesc TargetInfo
+subdirectories = Disassembler InstPrinter MCTargetDesc TargetInfo
 
 [component_0]
 type = TargetGroup
 name = XCore
 parent = Target
 has_asmprinter = 1
+has_disassembler = 1
 
 [component_1]
 type = Library
diff --git a/lib/Target/XCore/MCTargetDesc/LLVMBuild.txt b/lib/Target/XCore/MCTargetDesc/LLVMBuild.txt
index a80c939b4372..8213f9e42883 100644
--- a/lib/Target/XCore/MCTargetDesc/LLVMBuild.txt
+++ b/lib/Target/XCore/MCTargetDesc/LLVMBuild.txt
@@ -19,5 +19,5 @@
 type = Library
 name = XCoreDesc
 parent = XCore
-required_libraries = MC XCoreInfo
+required_libraries = MC XCoreAsmPrinter XCoreInfo
 add_to_library_groups = XCore
diff --git a/lib/Target/XCore/MCTargetDesc/XCoreMCTargetDesc.cpp b/lib/Target/XCore/MCTargetDesc/XCoreMCTargetDesc.cpp
index bbfdd4356f2a..b5b072dcbda6 100644
--- a/lib/Target/XCore/MCTargetDesc/XCoreMCTargetDesc.cpp
+++ b/lib/Target/XCore/MCTargetDesc/XCoreMCTargetDesc.cpp
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "XCoreMCTargetDesc.h"
+#include "InstPrinter/XCoreInstPrinter.h"
 #include "XCoreMCAsmInfo.h"
 #include "llvm/MC/MCCodeGenInfo.h"
 #include "llvm/MC/MCInstrInfo.h"
@@ -69,6 +70,15 @@ static MCCodeGenInfo *createXCoreMCCodeGenInfo(StringRef TT, Reloc::Model RM,
   return X;
 }
 
+static MCInstPrinter *createXCoreMCInstPrinter(const Target &T,
+                                               unsigned SyntaxVariant,
+                                               const MCAsmInfo &MAI,
+                                               const MCInstrInfo &MII,
+                                               const MCRegisterInfo &MRI,
+                                               const MCSubtargetInfo &STI) {
+  return new XCoreInstPrinter(MAI, MII, MRI);
+}
+
 // Force static initialization.
 extern "C" void LLVMInitializeXCoreTargetMC() {
   // Register the MC asm info.
@@ -87,4 +97,8 @@ extern "C" void LLVMInitializeXCoreTargetMC() {
   // Register the MC subtarget info.
   TargetRegistry::RegisterMCSubtargetInfo(TheXCoreTarget,
                                           createXCoreMCSubtargetInfo);
+
+  // Register the MCInstPrinter
+  TargetRegistry::RegisterMCInstPrinter(TheXCoreTarget,
+                                        createXCoreMCInstPrinter);
 }
diff --git a/lib/Target/XCore/Makefile b/lib/Target/XCore/Makefile
index b823c4ed37e9..92ddc8860876 100644
--- a/lib/Target/XCore/Makefile
+++ b/lib/Target/XCore/Makefile
@@ -14,10 +14,10 @@ TARGET = XCore
 # Make sure that tblgen is run, first thing.
 BUILT_SOURCES = XCoreGenRegisterInfo.inc XCoreGenInstrInfo.inc \
 		XCoreGenAsmWriter.inc \
-                XCoreGenDAGISel.inc XCoreGenCallingConv.inc \
-		XCoreGenSubtargetInfo.inc
+		XCoreGenDAGISel.inc XCoreGenCallingConv.inc \
+		XCoreGenDisassemblerTables.inc XCoreGenSubtargetInfo.inc
 
-DIRS = TargetInfo MCTargetDesc
+DIRS = Disassembler InstPrinter TargetInfo MCTargetDesc
 
 include $(LEVEL)/Makefile.common
 
diff --git a/lib/Target/XCore/TargetInfo/XCoreTargetInfo.cpp b/lib/Target/XCore/TargetInfo/XCoreTargetInfo.cpp
index 9a0971d1e45f..00e34e04fbe5 100644
--- a/lib/Target/XCore/TargetInfo/XCoreTargetInfo.cpp
+++ b/lib/Target/XCore/TargetInfo/XCoreTargetInfo.cpp
@@ -8,7 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "XCore.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
diff --git a/lib/Target/XCore/XCore.td b/lib/Target/XCore/XCore.td
index 04a1dd5e95be..e9a6d88fd68e 100644
--- a/lib/Target/XCore/XCore.td
+++ b/lib/Target/XCore/XCore.td
@@ -41,7 +41,13 @@ def : Proc<"xs1b-generic", []>;
 // Declare the target which we are implementing
 //===----------------------------------------------------------------------===//
 
+def XCoreAsmWriter : AsmWriter {
+  string AsmWriterClassName  = "InstPrinter";
+  bit isMCAsmWriter = 1;
+}
+
 def XCore : Target {
   // Pull in Instruction Info:
   let InstructionSet = XCoreInstrInfo;
+  let AssemblyWriters = [XCoreAsmWriter];
 }
diff --git a/lib/Target/XCore/XCoreAsmPrinter.cpp b/lib/Target/XCore/XCoreAsmPrinter.cpp
index caae56227214..0d146ba4d98d 100644
--- a/lib/Target/XCore/XCoreAsmPrinter.cpp
+++ b/lib/Target/XCore/XCoreAsmPrinter.cpp
@@ -14,31 +14,34 @@
 
 #define DEBUG_TYPE "asm-printer"
 #include "XCore.h"
+#include "InstPrinter/XCoreInstPrinter.h"
 #include "XCoreInstrInfo.h"
+#include "XCoreMCInstLower.h"
 #include "XCoreSubtarget.h"
 #include "XCoreTargetMachine.h"
-#include "llvm/Constants.h"
-#include "llvm/DebugInfo.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Module.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/CodeGen/AsmPrinter.h"
-#include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
+#include "llvm/DebugInfo.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Module.h"
 #include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
-#include "llvm/Target/Mangler.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetLoweringObjectFile.h"
-#include "llvm/ADT/SmallString.h"
-#include "llvm/ADT/StringExtras.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/Mangler.h"
+#include "llvm/Target/TargetLoweringObjectFile.h"
 #include <algorithm>
 #include <cctype>
 using namespace llvm;
@@ -52,16 +55,17 @@ static cl::opt<unsigned> MaxThreads("xcore-max-threads", cl::Optional,
 namespace {
   class XCoreAsmPrinter : public AsmPrinter {
     const XCoreSubtarget &Subtarget;
+    XCoreMCInstLower MCInstLowering;
     void PrintDebugValueComment(const MachineInstr *MI, raw_ostream &OS);
   public:
     explicit XCoreAsmPrinter(TargetMachine &TM, MCStreamer &Streamer)
-      : AsmPrinter(TM, Streamer), Subtarget(TM.getSubtarget<XCoreSubtarget>()){}
+      : AsmPrinter(TM, Streamer), Subtarget(TM.getSubtarget<XCoreSubtarget>()),
+        MCInstLowering(*this) {}
 
     virtual const char *getPassName() const {
       return "XCore Assembly Printer";
     }
 
-    void printMemOperand(const MachineInstr *MI, int opNum, raw_ostream &O);
     void printInlineJT(const MachineInstr *MI, int opNum, raw_ostream &O,
                        const std::string &directive = ".jmptable");
     void printInlineJT32(const MachineInstr *MI, int opNum, raw_ostream &O) {
@@ -75,18 +79,14 @@ namespace {
     void emitArrayBound(MCSymbol *Sym, const GlobalVariable *GV);
     virtual void EmitGlobalVariable(const GlobalVariable *GV);
 
-    void printInstruction(const MachineInstr *MI, raw_ostream &O); // autogen'd.
-    static const char *getRegisterName(unsigned RegNo);
-
     void EmitFunctionEntryLabel();
     void EmitInstruction(const MachineInstr *MI);
+    void EmitFunctionBodyStart();
     void EmitFunctionBodyEnd();
     virtual MachineLocation getDebugValueLocation(const MachineInstr *MI) const;
   };
 } // end of anonymous namespace
 
-#include "XCoreGenAsmWriter.inc"
-
 void XCoreAsmPrinter::emitArrayBound(MCSymbol *Sym, const GlobalVariable *GV) {
   assert(((GV->hasExternalLinkage() ||
     GV->hasWeakLinkage()) ||
@@ -171,12 +171,16 @@ void XCoreAsmPrinter::EmitGlobalVariable(const GlobalVariable *GV) {
   // The ABI requires that unsigned scalar types smaller than 32 bits
   // are padded to 32 bits.
   if (Size < 4)
-    OutStreamer.EmitZeros(4 - Size, 0);
+    OutStreamer.EmitZeros(4 - Size);
   
   // Mark the end of the global
   OutStreamer.EmitRawText("\t.cc_bottom " + Twine(GVSym->getName()) + ".data");
 }
 
+void XCoreAsmPrinter::EmitFunctionBodyStart() {
+  MCInstLowering.Initialize(Mang, &MF->getContext());
+}
+
 /// EmitFunctionBodyEnd - Targets can override this to emit stuff after
 /// the last basic block in the function.
 void XCoreAsmPrinter::EmitFunctionBodyEnd() {
@@ -192,17 +196,6 @@ void XCoreAsmPrinter::EmitFunctionEntryLabel() {
   OutStreamer.EmitLabel(CurrentFnSym);
 }
 
-void XCoreAsmPrinter::printMemOperand(const MachineInstr *MI, int opNum,
-                                      raw_ostream &O) {
-  printOperand(MI, opNum, O);
-  
-  if (MI->getOperand(opNum+1).isImm() && MI->getOperand(opNum+1).getImm() == 0)
-    return;
-  
-  O << "+";
-  printOperand(MI, opNum+1, O);
-}
-
 void XCoreAsmPrinter::
 printInlineJT(const MachineInstr *MI, int opNum, raw_ostream &O,
               const std::string &directive) {
@@ -225,7 +218,7 @@ void XCoreAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
   const MachineOperand &MO = MI->getOperand(opNum);
   switch (MO.getType()) {
   case MachineOperand::MO_Register:
-    O << getRegisterName(MO.getReg());
+    O << XCoreInstPrinter::getRegisterName(MO.getReg());
     break;
   case MachineOperand::MO_Immediate:
     O << MO.getImm();
@@ -270,7 +263,7 @@ bool XCoreAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
       return AsmPrinter::PrintAsmOperand(MI, OpNo, AsmVariant, ExtraCode, O);
     }
 
-printOperand(MI, OpNo, O);
+  printOperand(MI, OpNo, O);
   return false;
 }
 
@@ -317,15 +310,30 @@ void XCoreAsmPrinter::EmitInstruction(const MachineInstr *MI) {
   }
   case XCore::ADD_2rus:
     if (MI->getOperand(2).getImm() == 0) {
-      O << "\tmov " << getRegisterName(MI->getOperand(0).getReg()) << ", "
-        << getRegisterName(MI->getOperand(1).getReg());
+      O << "\tmov "
+        << XCoreInstPrinter::getRegisterName(MI->getOperand(0).getReg()) << ", "
+        << XCoreInstPrinter::getRegisterName(MI->getOperand(1).getReg());
       OutStreamer.EmitRawText(O.str());
       return;
     }
     break;
+  case XCore::BR_JT:
+  case XCore::BR_JT32:
+    O << "\tbru "
+      << XCoreInstPrinter::getRegisterName(MI->getOperand(1).getReg()) << '\n';
+    if (MI->getOpcode() == XCore::BR_JT)
+      printInlineJT(MI, 0, O);
+    else
+      printInlineJT32(MI, 0, O);
+    O << '\n';
+    OutStreamer.EmitRawText(O.str());
+    return;
   }
-  printInstruction(MI, O);
-  OutStreamer.EmitRawText(O.str());
+
+  MCInst TmpInst;
+  MCInstLowering.Lower(MI, TmpInst);
+
+  OutStreamer.EmitInstruction(TmpInst);
 }
 
 // Force static initialization.
diff --git a/lib/Target/XCore/XCoreFrameLowering.cpp b/lib/Target/XCore/XCoreFrameLowering.cpp
index e18d97384d3d..beeb07f831c6 100644
--- a/lib/Target/XCore/XCoreFrameLowering.cpp
+++ b/lib/Target/XCore/XCoreFrameLowering.cpp
@@ -16,16 +16,16 @@
 #include "XCore.h"
 #include "XCoreInstrInfo.h"
 #include "XCoreMachineFunctionInfo.h"
-#include "llvm/Function.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Target/TargetOptions.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
 
@@ -98,13 +98,10 @@ void XCoreFrameLowering::emitPrologue(MachineFunction &MF) const {
   DebugLoc dl = MBBI != MBB.end() ? MBBI->getDebugLoc() : DebugLoc();
 
   bool FP = hasFP(MF);
-  const AttrListPtr &PAL = MF.getFunction()->getAttributes();
+  const AttributeSet &PAL = MF.getFunction()->getAttributes();
 
-  for (unsigned I = 0, E = PAL.getNumAttrs(); I != E; ++I)
-    if (PAL.getAttributesAtIndex(I).hasAttribute(Attributes::Nest)) {
-      loadFromStack(MBB, MBBI, XCore::R11, 0, dl, TII);
-      break;
-    }
+  if (PAL.hasAttrSomewhere(Attribute::Nest))
+    loadFromStack(MBB, MBBI, XCore::R11, 0, dl, TII);
 
   // Work out frame sizes.
   int FrameSize = MFI->getStackSize();
@@ -264,7 +261,7 @@ void XCoreFrameLowering::emitEpilogue(MachineFunction &MF,
       BuildMI(MBB, MBBI, dl, TII.get(Opcode)).addImm(FrameSize);
       MBB.erase(MBBI);
     } else {
-      int Opcode = (isU6) ? XCore::LDAWSP_ru6_RRegs : XCore::LDAWSP_lru6_RRegs;
+      int Opcode = (isU6) ? XCore::LDAWSP_ru6 : XCore::LDAWSP_lru6;
       BuildMI(MBB, MBBI, dl, TII.get(Opcode), XCore::SP).addImm(FrameSize);
     }
   }
@@ -335,6 +332,58 @@ bool XCoreFrameLowering::restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
   return true;
 }
 
+// This function eliminates ADJCALLSTACKDOWN,
+// ADJCALLSTACKUP pseudo instructions
+void XCoreFrameLowering::
+eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
+                              MachineBasicBlock::iterator I) const {
+  const XCoreInstrInfo &TII =
+    *static_cast<const XCoreInstrInfo*>(MF.getTarget().getInstrInfo());
+  if (!hasReservedCallFrame(MF)) {
+    // Turn the adjcallstackdown instruction into 'extsp <amt>' and the
+    // adjcallstackup instruction into 'ldaw sp, sp[<amt>]'
+    MachineInstr *Old = I;
+    uint64_t Amount = Old->getOperand(0).getImm();
+    if (Amount != 0) {
+      // We need to keep the stack aligned properly.  To do this, we round the
+      // amount of space needed for the outgoing arguments up to the next
+      // alignment boundary.
+      unsigned Align = getStackAlignment();
+      Amount = (Amount+Align-1)/Align*Align;
+
+      assert(Amount%4 == 0);
+      Amount /= 4;
+
+      bool isU6 = isImmU6(Amount);
+      if (!isU6 && !isImmU16(Amount)) {
+        // FIX could emit multiple instructions in this case.
+#ifndef NDEBUG
+        errs() << "eliminateCallFramePseudoInstr size too big: "
+               << Amount << "\n";
+#endif
+        llvm_unreachable(0);
+      }
+
+      MachineInstr *New;
+      if (Old->getOpcode() == XCore::ADJCALLSTACKDOWN) {
+        int Opcode = isU6 ? XCore::EXTSP_u6 : XCore::EXTSP_lu6;
+        New=BuildMI(MF, Old->getDebugLoc(), TII.get(Opcode))
+          .addImm(Amount);
+      } else {
+        assert(Old->getOpcode() == XCore::ADJCALLSTACKUP);
+        int Opcode = isU6 ? XCore::LDAWSP_ru6 : XCore::LDAWSP_lru6;
+        New=BuildMI(MF, Old->getDebugLoc(), TII.get(Opcode), XCore::SP)
+          .addImm(Amount);
+      }
+
+      // Replace the pseudo instruction with a new instruction...
+      MBB.insert(I, New);
+    }
+  }
+  
+  MBB.erase(I);
+}
+
 void
 XCoreFrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
                                                      RegScavenger *RS) const {
@@ -360,7 +409,7 @@ XCoreFrameLowering::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
   }
   if (RegInfo->requiresRegisterScavenging(MF)) {
     // Reserve a slot close to SP or frame pointer.
-    RS->setScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
+    RS->addScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
                                                        RC->getAlignment(),
                                                        false));
   }
diff --git a/lib/Target/XCore/XCoreFrameLowering.h b/lib/Target/XCore/XCoreFrameLowering.h
index db1bbb60d968..ebad62f2fa53 100644
--- a/lib/Target/XCore/XCoreFrameLowering.h
+++ b/lib/Target/XCore/XCoreFrameLowering.h
@@ -39,6 +39,10 @@ namespace llvm {
                                      const std::vector<CalleeSavedInfo> &CSI,
                                      const TargetRegisterInfo *TRI) const;
 
+    void eliminateCallFramePseudoInstr(MachineFunction &MF,
+                                       MachineBasicBlock &MBB,
+                                       MachineBasicBlock::iterator I) const;
+
     bool hasFP(const MachineFunction &MF) const;
 
     void processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
diff --git a/lib/Target/XCore/XCoreISelDAGToDAG.cpp b/lib/Target/XCore/XCoreISelDAGToDAG.cpp
index 7564fbad7d45..fbf86c523054 100644
--- a/lib/Target/XCore/XCoreISelDAGToDAG.cpp
+++ b/lib/Target/XCore/XCoreISelDAGToDAG.cpp
@@ -13,23 +13,23 @@
 
 #include "XCore.h"
 #include "XCoreTargetMachine.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/CallingConv.h"
-#include "llvm/Constants.h"
-#include "llvm/LLVMContext.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
-#include "llvm/Target/TargetLowering.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/LLVMContext.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetLowering.h"
 using namespace llvm;
 
 /// XCoreDAGToDAGISel - XCore specific code to select XCore machine
@@ -211,15 +211,10 @@ SDNode *XCoreDAGToDAGISel::Select(SDNode *N) {
     return CurDAG->getMachineNode(XCore::LMUL_l6r, dl, MVT::i32, MVT::i32,
                                   Ops, 4);
   }
-  case ISD::INTRINSIC_WO_CHAIN: {
-    unsigned IntNo = cast<ConstantSDNode>(N->getOperand(0))->getZExtValue();
-    switch (IntNo) {
-    case Intrinsic::xcore_crc8:
-      SDValue Ops[] = { N->getOperand(1), N->getOperand(2), N->getOperand(3) };
-      return CurDAG->getMachineNode(XCore::CRC8_l4r, dl, MVT::i32, MVT::i32,
-                                    Ops, 3);
-    }
-    break;
+  case XCoreISD::CRC8: {
+    SDValue Ops[] = { N->getOperand(0), N->getOperand(1), N->getOperand(2) };
+    return CurDAG->getMachineNode(XCore::CRC8_l4r, dl, MVT::i32, MVT::i32,
+                                  Ops, 3);
   }
   case ISD::BRIND:
     if (SDNode *ResNode = SelectBRIND(N))
diff --git a/lib/Target/XCore/XCoreISelLowering.cpp b/lib/Target/XCore/XCoreISelLowering.cpp
index 9e7816e21f80..a5d2be88db7d 100644
--- a/lib/Target/XCore/XCoreISelLowering.cpp
+++ b/lib/Target/XCore/XCoreISelLowering.cpp
@@ -14,17 +14,11 @@
 #define DEBUG_TYPE "xcore-lower"
 
 #include "XCoreISelLowering.h"
-#include "XCoreMachineFunctionInfo.h"
 #include "XCore.h"
-#include "XCoreTargetObjectFile.h"
-#include "XCoreTargetMachine.h"
+#include "XCoreMachineFunctionInfo.h"
 #include "XCoreSubtarget.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/CallingConv.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/GlobalAlias.h"
+#include "XCoreTargetMachine.h"
+#include "XCoreTargetObjectFile.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -33,6 +27,12 @@
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/CodeGen/ValueTypes.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalAlias.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
@@ -54,6 +54,7 @@ getTargetNodeName(unsigned Opcode) const
     case XCoreISD::LMUL              : return "XCoreISD::LMUL";
     case XCoreISD::MACCU             : return "XCoreISD::MACCU";
     case XCoreISD::MACCS             : return "XCoreISD::MACCS";
+    case XCoreISD::CRC8              : return "XCoreISD::CRC8";
     case XCoreISD::BR_JT             : return "XCoreISD::BR_JT";
     case XCoreISD::BR_JT32           : return "XCoreISD::BR_JT32";
     default                          : return NULL;
@@ -83,7 +84,7 @@ XCoreTargetLowering::XCoreTargetLowering(XCoreTargetMachine &XTM)
   setBooleanVectorContents(ZeroOrOneBooleanContent); // FIXME: Is this correct?
 
   // XCore does not have the NodeTypes below.
-  setOperationAction(ISD::BR_CC,     MVT::Other, Expand);
+  setOperationAction(ISD::BR_CC,     MVT::i32,   Expand);
   setOperationAction(ISD::SELECT_CC, MVT::i32,   Custom);
   setOperationAction(ISD::ADDC, MVT::i32, Expand);
   setOperationAction(ISD::ADDE, MVT::i32, Expand);
@@ -152,9 +153,12 @@ XCoreTargetLowering::XCoreTargetLowering(XCoreTargetMachine &XTM)
   setOperationAction(ISD::INIT_TRAMPOLINE, MVT::Other, Custom);
   setOperationAction(ISD::ADJUST_TRAMPOLINE, MVT::Other, Custom);
 
-  maxStoresPerMemset = maxStoresPerMemsetOptSize = 4;
-  maxStoresPerMemmove = maxStoresPerMemmoveOptSize
-    = maxStoresPerMemcpy = maxStoresPerMemcpyOptSize = 2;
+  // We want to custom lower some of our intrinsics.
+  setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
+
+  MaxStoresPerMemset = MaxStoresPerMemsetOptSize = 4;
+  MaxStoresPerMemmove = MaxStoresPerMemmoveOptSize
+    = MaxStoresPerMemcpy = MaxStoresPerMemcpyOptSize = 2;
 
   // We have target-specific dag combine patterns for the following nodes:
   setTargetDAGCombine(ISD::STORE);
@@ -167,24 +171,25 @@ SDValue XCoreTargetLowering::
 LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   switch (Op.getOpcode())
   {
-  case ISD::GlobalAddress:    return LowerGlobalAddress(Op, DAG);
-  case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
-  case ISD::BlockAddress:     return LowerBlockAddress(Op, DAG);
-  case ISD::ConstantPool:     return LowerConstantPool(Op, DAG);
-  case ISD::BR_JT:            return LowerBR_JT(Op, DAG);
-  case ISD::LOAD:             return LowerLOAD(Op, DAG);
-  case ISD::STORE:            return LowerSTORE(Op, DAG);
-  case ISD::SELECT_CC:        return LowerSELECT_CC(Op, DAG);
-  case ISD::VAARG:            return LowerVAARG(Op, DAG);
-  case ISD::VASTART:          return LowerVASTART(Op, DAG);
-  case ISD::SMUL_LOHI:        return LowerSMUL_LOHI(Op, DAG);
-  case ISD::UMUL_LOHI:        return LowerUMUL_LOHI(Op, DAG);
+  case ISD::GlobalAddress:      return LowerGlobalAddress(Op, DAG);
+  case ISD::GlobalTLSAddress:   return LowerGlobalTLSAddress(Op, DAG);
+  case ISD::BlockAddress:       return LowerBlockAddress(Op, DAG);
+  case ISD::ConstantPool:       return LowerConstantPool(Op, DAG);
+  case ISD::BR_JT:              return LowerBR_JT(Op, DAG);
+  case ISD::LOAD:               return LowerLOAD(Op, DAG);
+  case ISD::STORE:              return LowerSTORE(Op, DAG);
+  case ISD::SELECT_CC:          return LowerSELECT_CC(Op, DAG);
+  case ISD::VAARG:              return LowerVAARG(Op, DAG);
+  case ISD::VASTART:            return LowerVASTART(Op, DAG);
+  case ISD::SMUL_LOHI:          return LowerSMUL_LOHI(Op, DAG);
+  case ISD::UMUL_LOHI:          return LowerUMUL_LOHI(Op, DAG);
   // FIXME: Remove these when LegalizeDAGTypes lands.
   case ISD::ADD:
-  case ISD::SUB:              return ExpandADDSUB(Op.getNode(), DAG);
-  case ISD::FRAMEADDR:        return LowerFRAMEADDR(Op, DAG);
-  case ISD::INIT_TRAMPOLINE:  return LowerINIT_TRAMPOLINE(Op, DAG);
-  case ISD::ADJUST_TRAMPOLINE: return LowerADJUST_TRAMPOLINE(Op, DAG);
+  case ISD::SUB:                return ExpandADDSUB(Op.getNode(), DAG);
+  case ISD::FRAMEADDR:          return LowerFRAMEADDR(Op, DAG);
+  case ISD::INIT_TRAMPOLINE:    return LowerINIT_TRAMPOLINE(Op, DAG);
+  case ISD::ADJUST_TRAMPOLINE:  return LowerADJUST_TRAMPOLINE(Op, DAG);
+  case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
   default:
     llvm_unreachable("unimplemented operand");
   }
@@ -225,20 +230,16 @@ getGlobalAddressWrapper(SDValue GA, const GlobalValue *GV,
 {
   // FIXME there is no actual debug info here
   DebugLoc dl = GA.getDebugLoc();
-  if (isa<Function>(GV)) {
-    return DAG.getNode(XCoreISD::PCRelativeWrapper, dl, MVT::i32, GA);
+  const GlobalValue *UnderlyingGV = GV;
+  // If GV is an alias then use the aliasee to determine the wrapper type
+  if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(GV))
+    UnderlyingGV = GA->resolveAliasedGlobal();
+  if (const GlobalVariable *GVar = dyn_cast<GlobalVariable>(UnderlyingGV)) {
+    if (GVar->isConstant())
+      return DAG.getNode(XCoreISD::CPRelativeWrapper, dl, MVT::i32, GA);
+    return DAG.getNode(XCoreISD::DPRelativeWrapper, dl, MVT::i32, GA);
   }
-  const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV);
-  if (!GVar) {
-    // If GV is an alias then use the aliasee to determine constness
-    if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(GV))
-      GVar = dyn_cast_or_null<GlobalVariable>(GA->resolveAliasedGlobal());
-  }
-  bool isConst = GVar && GVar->isConstant();
-  if (isConst) {
-    return DAG.getNode(XCoreISD::CPRelativeWrapper, dl, MVT::i32, GA);
-  }
-  return DAG.getNode(XCoreISD::DPRelativeWrapper, dl, MVT::i32, GA);
+  return DAG.getNode(XCoreISD::PCRelativeWrapper, dl, MVT::i32, GA);
 }
 
 SDValue XCoreTargetLowering::
@@ -740,13 +741,13 @@ ExpandADDSUB(SDNode *N, SelectionDAG &DAG) const
   unsigned Opcode = (N->getOpcode() == ISD::ADD) ? XCoreISD::LADD :
                                                    XCoreISD::LSUB;
   SDValue Zero = DAG.getConstant(0, MVT::i32);
-  SDValue Carry = DAG.getNode(Opcode, dl, DAG.getVTList(MVT::i32, MVT::i32),
-                                  LHSL, RHSL, Zero);
-  SDValue Lo(Carry.getNode(), 1);
+  SDValue Lo = DAG.getNode(Opcode, dl, DAG.getVTList(MVT::i32, MVT::i32),
+                           LHSL, RHSL, Zero);
+  SDValue Carry(Lo.getNode(), 1);
 
-  SDValue Ignored = DAG.getNode(Opcode, dl, DAG.getVTList(MVT::i32, MVT::i32),
-                                  LHSH, RHSH, Carry);
-  SDValue Hi(Ignored.getNode(), 1);
+  SDValue Hi = DAG.getNode(Opcode, dl, DAG.getVTList(MVT::i32, MVT::i32),
+                           LHSH, RHSH, Carry);
+  SDValue Ignored(Hi.getNode(), 1);
   // Merge the pieces
   return DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, Lo, Hi);
 }
@@ -862,6 +863,23 @@ LowerINIT_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const {
   return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains, 5);
 }
 
+SDValue XCoreTargetLowering::
+LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc DL = Op.getDebugLoc();
+  unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
+  switch (IntNo) {
+    case Intrinsic::xcore_crc8:
+      EVT VT = Op.getValueType();
+      SDValue Data =
+        DAG.getNode(XCoreISD::CRC8, DL, DAG.getVTList(VT, VT),
+                    Op.getOperand(1), Op.getOperand(2) , Op.getOperand(3));
+      SDValue Crc(Data.getNode(), 1);
+      SDValue Results[] = { Crc, Data };
+      return DAG.getMergeValues(Results, 2, DL);
+  }
+  return SDValue();
+}
+
 //===----------------------------------------------------------------------===//
 //                      Calling Convention Implementation
 //===----------------------------------------------------------------------===//
@@ -1231,15 +1249,11 @@ XCoreTargetLowering::LowerReturn(SDValue Chain,
   // Analyze return values.
   CCInfo.AnalyzeReturn(Outs, RetCC_XCore);
 
-  // If this is the first return lowered for this function, add
-  // the regs to the liveout set for the function.
-  if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
-    for (unsigned i = 0; i != RVLocs.size(); ++i)
-      if (RVLocs[i].isRegLoc())
-        DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
-  }
-
   SDValue Flag;
+  SmallVector<SDValue, 4> RetOps(1, Chain);
+
+  // Return on XCore is always a "retsp 0"
+  RetOps.push_back(DAG.getConstant(0, MVT::i32));
 
   // Copy the result values into the output registers.
   for (unsigned i = 0; i != RVLocs.size(); ++i) {
@@ -1252,15 +1266,17 @@ XCoreTargetLowering::LowerReturn(SDValue Chain,
     // guarantee that all emitted copies are
     // stuck together, avoiding something bad
     Flag = Chain.getValue(1);
+    RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
   }
 
-  // Return on XCore is always a "retsp 0"
+  RetOps[0] = Chain;  // Update chain.
+
+  // Add the flag if we have it.
   if (Flag.getNode())
-    return DAG.getNode(XCoreISD::RETSP, dl, MVT::Other,
-                       Chain, DAG.getConstant(0, MVT::i32), Flag);
-  else // Return Void
-    return DAG.getNode(XCoreISD::RETSP, dl, MVT::Other,
-                       Chain, DAG.getConstant(0, MVT::i32));
+    RetOps.push_back(Flag);
+
+  return DAG.getNode(XCoreISD::RETSP, dl, MVT::Other,
+                     &RetOps[0], RetOps.size());
 }
 
 //===----------------------------------------------------------------------===//
@@ -1357,13 +1373,13 @@ SDValue XCoreTargetLowering::PerformDAGCombine(SDNode *N,
       SDValue Carry = DAG.getConstant(0, VT);
       SDValue Result = DAG.getNode(ISD::AND, dl, VT, N2,
                                    DAG.getConstant(1, VT));
-      SDValue Ops [] = { Carry, Result };
+      SDValue Ops[] = { Result, Carry };
       return DAG.getMergeValues(Ops, 2, dl);
     }
 
     // fold (ladd x, 0, y) -> 0, add x, y iff carry is unused and y has only the
     // low bit set
-    if (N1C && N1C->isNullValue() && N->hasNUsesOfValue(0, 0)) {
+    if (N1C && N1C->isNullValue() && N->hasNUsesOfValue(0, 1)) {
       APInt KnownZero, KnownOne;
       APInt Mask = APInt::getHighBitsSet(VT.getSizeInBits(),
                                          VT.getSizeInBits() - 1);
@@ -1371,7 +1387,7 @@ SDValue XCoreTargetLowering::PerformDAGCombine(SDNode *N,
       if ((KnownZero & Mask) == Mask) {
         SDValue Carry = DAG.getConstant(0, VT);
         SDValue Result = DAG.getNode(ISD::ADD, dl, VT, N0, N2);
-        SDValue Ops [] = { Carry, Result };
+        SDValue Ops[] = { Result, Carry };
         return DAG.getMergeValues(Ops, 2, dl);
       }
     }
@@ -1395,14 +1411,14 @@ SDValue XCoreTargetLowering::PerformDAGCombine(SDNode *N,
         SDValue Borrow = N2;
         SDValue Result = DAG.getNode(ISD::SUB, dl, VT,
                                      DAG.getConstant(0, VT), N2);
-        SDValue Ops [] = { Borrow, Result };
+        SDValue Ops[] = { Result, Borrow };
         return DAG.getMergeValues(Ops, 2, dl);
       }
     }
 
     // fold (lsub x, 0, y) -> 0, sub x, y iff borrow is unused and y has only the
     // low bit set
-    if (N1C && N1C->isNullValue() && N->hasNUsesOfValue(0, 0)) {
+    if (N1C && N1C->isNullValue() && N->hasNUsesOfValue(0, 1)) {
       APInt KnownZero, KnownOne;
       APInt Mask = APInt::getHighBitsSet(VT.getSizeInBits(),
                                          VT.getSizeInBits() - 1);
@@ -1410,7 +1426,7 @@ SDValue XCoreTargetLowering::PerformDAGCombine(SDNode *N,
       if ((KnownZero & Mask) == Mask) {
         SDValue Borrow = DAG.getConstant(0, VT);
         SDValue Result = DAG.getNode(ISD::SUB, dl, VT, N0, N2);
-        SDValue Ops [] = { Borrow, Result };
+        SDValue Ops[] = { Result, Borrow };
         return DAG.getMergeValues(Ops, 2, dl);
       }
     }
@@ -1436,11 +1452,15 @@ SDValue XCoreTargetLowering::PerformDAGCombine(SDNode *N,
       // If the high result is unused fold to add(a, b)
       if (N->hasNUsesOfValue(0, 0)) {
         SDValue Lo = DAG.getNode(ISD::ADD, dl, VT, N2, N3);
-        SDValue Ops [] = { Lo, Lo };
+        SDValue Ops[] = { Lo, Lo };
         return DAG.getMergeValues(Ops, 2, dl);
       }
       // Otherwise fold to ladd(a, b, 0)
-      return DAG.getNode(XCoreISD::LADD, dl, DAG.getVTList(VT, VT), N2, N3, N1);
+      SDValue Result =
+        DAG.getNode(XCoreISD::LADD, dl, DAG.getVTList(VT, VT), N2, N3, N1);
+      SDValue Carry(Result.getNode(), 1);
+      SDValue Ops[] = { Carry, Result };
+      return DAG.getMergeValues(Ops, 2, dl);
     }
   }
   break;
@@ -1534,7 +1554,7 @@ void XCoreTargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
   default: break;
   case XCoreISD::LADD:
   case XCoreISD::LSUB:
-    if (Op.getResNo() == 0) {
+    if (Op.getResNo() == 1) {
       // Top bits of carry / borrow are clear.
       KnownZero = APInt::getHighBitsSet(KnownZero.getBitWidth(),
                                         KnownZero.getBitWidth() - 1);
diff --git a/lib/Target/XCore/XCoreISelLowering.h b/lib/Target/XCore/XCoreISelLowering.h
index 2874f00e4763..8d258f5054c1 100644
--- a/lib/Target/XCore/XCoreISelLowering.h
+++ b/lib/Target/XCore/XCoreISelLowering.h
@@ -63,6 +63,9 @@ namespace llvm {
       // Corresponds to MACCS instruction
       MACCS,
 
+      // Corresponds to CRC8 instruction
+      CRC8,
+
       // Jumptable branch.
       BR_JT,
 
@@ -81,7 +84,7 @@ namespace llvm {
     explicit XCoreTargetLowering(XCoreTargetMachine &TM);
 
     virtual unsigned getJumpTableEncoding() const;
-    virtual MVT getShiftAmountTy(EVT LHSTy) const { return MVT::i32; }
+    virtual MVT getScalarShiftAmountTy(EVT LHSTy) const { return MVT::i32; }
 
     /// LowerOperation - Provide custom lowering hooks for some operations.
     virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
@@ -147,6 +150,7 @@ namespace llvm {
     SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerINIT_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerADJUST_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
 
     // Inline asm support
     std::pair<unsigned, const TargetRegisterClass*>
diff --git a/lib/Target/XCore/XCoreInstrFormats.td b/lib/Target/XCore/XCoreInstrFormats.td
index 1963a70fb30d..379cc39aa617 100644
--- a/lib/Target/XCore/XCoreInstrFormats.td
+++ b/lib/Target/XCore/XCoreInstrFormats.td
@@ -10,7 +10,7 @@
 //===----------------------------------------------------------------------===//
 // Instruction format superclass
 //===----------------------------------------------------------------------===//
-class InstXCore<dag outs, dag ins, string asmstr, list<dag> pattern>
+class InstXCore<int sz, dag outs, dag ins, string asmstr, list<dag> pattern>
     : Instruction {
   field bits<32> Inst;
 
@@ -19,102 +19,259 @@ class InstXCore<dag outs, dag ins, string asmstr, list<dag> pattern>
   dag InOperandList = ins;
   let AsmString   = asmstr;
   let Pattern = pattern;
+  let Size = sz;
+  field bits<32> SoftFail = 0;
 }
 
 // XCore pseudo instructions format
 class PseudoInstXCore<dag outs, dag ins, string asmstr, list<dag> pattern>
-   : InstXCore<outs, ins, asmstr, pattern>;
+   : InstXCore<0, outs, ins, asmstr, pattern> {
+  let isPseudo = 1;
+}
 
 //===----------------------------------------------------------------------===//
 // Instruction formats
 //===----------------------------------------------------------------------===//
 
-class _F3R<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _F3R<bits<5> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<2, outs, ins, asmstr, pattern> {
+  let Inst{15-11} = opc;
+  let DecoderMethod = "Decode3RInstruction";
+}
+
+// 3R with first operand as an immediate. Used for TSETR where the first
+// operand is treated as an immediate since it refers to a register number in
+// another thread.
+class _F3RImm<bits<5> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : _F3R<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "Decode3RImmInstruction";
+}
+
+class _FL3R<bits<9> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<4, outs, ins, asmstr, pattern> {
+  let Inst{31-27} = opc{8-4};
+  let Inst{26-20} = 0b1111110;
+  let Inst{19-16} = opc{3-0};
+
+  let Inst{15-11} = 0b11111;
+  let DecoderMethod = "DecodeL3RInstruction";
+}
+
+// L3R with first operand as both a source and a destination.
+class _FL3RSrcDst<bits<9> opc, dag outs, dag ins, string asmstr,
+                  list<dag> pattern> : _FL3R<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "DecodeL3RSrcDstInstruction";
+}
+
+class _F2RUS<bits<5> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<2, outs, ins, asmstr, pattern> {
+  let Inst{15-11} = opc;
+  let DecoderMethod = "Decode2RUSInstruction";
+}
+
+// 2RUS with bitp operand
+class _F2RUSBitp<bits<5> opc, dag outs, dag ins, string asmstr,
+                 list<dag> pattern>
+    : _F2RUS<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "Decode2RUSBitpInstruction";
+}
+
+class _FL2RUS<bits<9> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<4, outs, ins, asmstr, pattern> {
+  let Inst{31-27} = opc{8-4};
+  let Inst{26-20} = 0b1111110;
+  let Inst{19-16} = opc{3-0};
+
+  let Inst{15-11} = 0b11111;
+  let DecoderMethod = "DecodeL2RUSInstruction";
+}
+
+// L2RUS with bitp operand
+class _FL2RUSBitp<bits<9> opc, dag outs, dag ins, string asmstr,
+                  list<dag> pattern>
+    : _FL2RUS<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "DecodeL2RUSBitpInstruction";
+}
+
+class _FRU6<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<2, outs, ins, asmstr, pattern> {
+  bits<4> a;
+  bits<6> b;
+
+  let Inst{15-10} = opc;
+  let Inst{9-6} = a;
+  let Inst{5-0} = b;
 }
 
-class _FL3R<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _FLRU6<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<4, outs, ins, asmstr, pattern> {
+  bits<4> a;
+  bits<16> b;
+
+  let Inst{31-26} = opc;
+  let Inst{25-22} = a;
+  let Inst{21-16} = b{5-0};
+  let Inst{15-10} = 0b111100;
+  let Inst{9-0} = b{15-6};
 }
 
-class _F2RUS<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _FU6<bits<10> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<2, outs, ins, asmstr, pattern> {
+  bits<6> a;
+
+  let Inst{15-6} = opc;
+  let Inst{5-0} = a;
 }
 
-class _FL2RUS<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _FLU6<bits<10> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<4, outs, ins, asmstr, pattern> {
+  bits<16> a;
+
+  let Inst{31-22} = opc;
+  let Inst{21-16} = a{5-0};
+  let Inst{15-10} = 0b111100;
+  let Inst{9-0} = a{15-6};
 }
 
-class _FRU6<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _FU10<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<2, outs, ins, asmstr, pattern> {
+  bits<10> a;
+
+  let Inst{15-10} = opc;
+  let Inst{9-0} = a;
 }
 
-class _FLRU6<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _FLU10<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<4, outs, ins, asmstr, pattern> {
+  bits<20> a;
+
+  let Inst{31-26} = opc;
+  let Inst{25-16} = a{9-0};
+  let Inst{15-10} = 0b111100;
+  let Inst{9-0} = a{19-10};
 }
 
-class _FU6<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _F2R<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<2, outs, ins, asmstr, pattern> {
+  let Inst{15-11} = opc{5-1};
+  let Inst{4} = opc{0};
+  let DecoderMethod = "Decode2RInstruction";
 }
 
-class _FLU6<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+// 2R with first operand as an immediate. Used for TSETMR where the first
+// operand is treated as an immediate since it refers to a register number in
+// another thread.
+class _F2RImm<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : _F2R<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "Decode2RImmInstruction";
 }
 
-class _FU10<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+// 2R with first operand as both a source and a destination.
+class _F2RSrcDst<bits<6> opc, dag outs, dag ins, string asmstr,
+                 list<dag> pattern> : _F2R<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "Decode2RSrcDstInstruction";
 }
 
-class _FLU10<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+// Same as 2R with last two operands swapped
+class _FR2R<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : _F2R<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "DecodeR2RInstruction";
 }
 
-class _F2R<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _FRUS<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<2, outs, ins, asmstr, pattern> {
+  let Inst{15-11} = opc{5-1};
+  let Inst{4} = opc{0};
+  let DecoderMethod = "DecodeRUSInstruction";
 }
 
-class _FRUS<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+// RUS with bitp operand
+class _FRUSBitp<bits<6> opc, dag outs, dag ins, string asmstr,
+                list<dag> pattern>
+    : _FRUS<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "DecodeRUSBitpInstruction";
 }
 
-class _FL2R<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+// RUS with first operand as both a source and a destination and a bitp second
+// operand
+class _FRUSSrcDstBitp<bits<6> opc, dag outs, dag ins, string asmstr,
+                      list<dag> pattern>
+    : _FRUS<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "DecodeRUSSrcDstBitpInstruction";
 }
 
-class _F1R<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _FL2R<bits<10> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<4, outs, ins, asmstr, pattern> {
+  let Inst{31-27} = opc{9-5};
+  let Inst{26-20} = 0b1111110;
+  let Inst{19-16} = opc{4-1};
+
+  let Inst{15-11} = 0b11111;
+  let Inst{4} = opc{0};
+  let DecoderMethod = "DecodeL2RInstruction";
 }
 
-class _F0R<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+// Same as L2R with last two operands swapped
+class _FLR2R<bits<10> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : _FL2R<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "DecodeLR2RInstruction";
 }
 
-class _L4R<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _F1R<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<2, outs, ins, asmstr, pattern> {
+  bits<4> a;
+
+  let Inst{15-11} = opc{5-1};
+  let Inst{10-5} = 0b111111;
+  let Inst{4} = opc{0};
+  let Inst{3-0} = a;
 }
 
-class _L5R<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _F0R<bits<10> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<2, outs, ins, asmstr, pattern> {
+  let Inst{15-11} = opc{9-5};
+  let Inst{10-5} = 0b111111;
+  let Inst{4-0} = opc{4-0};
 }
 
-class _L6R<dag outs, dag ins, string asmstr, list<dag> pattern>
-    : InstXCore<outs, ins, asmstr, pattern> {
-  let Inst{31-0} = 0;
+class _FL4R<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<4, outs, ins, asmstr, pattern> {
+  bits<4> d;
+
+  let Inst{31-27} = opc{5-1};
+  let Inst{26-21} = 0b111111;
+  let Inst{20} = opc{0};
+  let Inst{19-16} = d;
+  let Inst{15-11} = 0b11111;
+}
+
+// L4R with 4th operand as both a source and a destination.
+class _FL4RSrcDst<bits<6> opc, dag outs, dag ins, string asmstr,
+                  list<dag> pattern>
+    : _FL4R<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "DecodeL4RSrcDstInstruction";
+}
+
+// L4R with 1st and 4th operand as both a source and a destination.
+class _FL4RSrcDstSrcDst<bits<6> opc, dag outs, dag ins, string asmstr,
+                        list<dag> pattern>
+    : _FL4R<opc, outs, ins, asmstr, pattern> {
+  let DecoderMethod = "DecodeL4RSrcDstSrcDstInstruction";
+}
+
+class _FL5R<bits<6> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<4, outs, ins, asmstr, pattern> {
+  let Inst{31-27} = opc{5-1};
+  let Inst{20} = opc{0};
+  let Inst{15-11} = 0b11111;
+
+  let DecoderMethod = "DecodeL5RInstruction";
+}
+
+class _FL6R<bits<5> opc, dag outs, dag ins, string asmstr, list<dag> pattern>
+    : InstXCore<4, outs, ins, asmstr, pattern> {
+  let Inst{31-27} = opc;
+  let Inst{15-11} = 0b11111;
+
+  let DecoderMethod = "DecodeL6RInstruction";
 }
diff --git a/lib/Target/XCore/XCoreInstrInfo.cpp b/lib/Target/XCore/XCoreInstrInfo.cpp
index 0a3008d7ab33..e457e0dbf027 100644
--- a/lib/Target/XCore/XCoreInstrInfo.cpp
+++ b/lib/Target/XCore/XCoreInstrInfo.cpp
@@ -12,12 +12,12 @@
 //===----------------------------------------------------------------------===//
 
 #include "XCoreInstrInfo.h"
-#include "XCoreMachineFunctionInfo.h"
 #include "XCore.h"
-#include "llvm/MC/MCContext.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "XCoreMachineFunctionInfo.h"
 #include "llvm/ADT/STLExtras.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/MC/MCContext.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
diff --git a/lib/Target/XCore/XCoreInstrInfo.td b/lib/Target/XCore/XCoreInstrInfo.td
index 3e7666bdb936..03653cb2b3de 100644
--- a/lib/Target/XCore/XCoreInstrInfo.td
+++ b/lib/Target/XCore/XCoreInstrInfo.td
@@ -32,8 +32,8 @@ def XCoreBranchLink     : SDNode<"XCoreISD::BL",SDT_XCoreBranchLink,
                             [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
                              SDNPVariadic]>;
 
-def XCoreRetsp       : SDNode<"XCoreISD::RETSP", SDTBrind,
-                         [SDNPHasChain, SDNPOptInGlue, SDNPMayLoad]>;
+def XCoreRetsp : SDNode<"XCoreISD::RETSP", SDTBrind,
+                      [SDNPHasChain, SDNPOptInGlue, SDNPMayLoad, SDNPVariadic]>;
 
 def SDT_XCoreBR_JT    : SDTypeProfile<0, 2,
                                       [SDTCisVT<0, i32>, SDTCisVT<1, i32>]>;
@@ -182,6 +182,7 @@ def ADDRcpii : ComplexPattern<i32, 2, "SelectADDRcpii", [add, cprelwrapper],
 // Address operands
 def MEMii : Operand<i32> {
   let PrintMethod = "printMemOperand";
+  let DecoderMethod = "DecodeMEMiiOperand";
   let MIOperandInfo = (ops i32imm, i32imm);
 }
 
@@ -200,154 +201,117 @@ def InlineJT32 : Operand<i32> {
 
 // Three operand short
 
-multiclass F3R_2RUS<string OpcStr, SDNode OpNode> {
-  def _3r: _F3R<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
-  def _2rus : _F2RUS<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, i32imm:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 [(set GRRegs:$dst, (OpNode GRRegs:$b, immUs:$c))]>;
+multiclass F3R_2RUS<bits<5> opc1, bits<5> opc2, string OpcStr, SDNode OpNode> {
+  def _3r: _F3R<opc1, (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
+                !strconcat(OpcStr, " $dst, $b, $c"),
+                [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
+  def _2rus : _F2RUS<opc2, (outs GRRegs:$dst), (ins GRRegs:$b, i32imm:$c),
+                     !strconcat(OpcStr, " $dst, $b, $c"),
+                     [(set GRRegs:$dst, (OpNode GRRegs:$b, immUs:$c))]>;
 }
 
-multiclass F3R_2RUS_np<string OpcStr> {
-  def _3r: _F3R<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 []>;
-  def _2rus : _F2RUS<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, i32imm:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 []>;
+multiclass F3R_2RUS_np<bits<5> opc1, bits<5> opc2, string OpcStr> {
+  def _3r: _F3R<opc1, (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
+                !strconcat(OpcStr, " $dst, $b, $c"), []>;
+  def _2rus : _F2RUS<opc2, (outs GRRegs:$dst), (ins GRRegs:$b, i32imm:$c),
+                     !strconcat(OpcStr, " $dst, $b, $c"), []>;
 }
 
-multiclass F3R_2RBITP<string OpcStr, SDNode OpNode> {
-  def _3r: _F3R<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
-  def _2rus : _F2RUS<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, i32imm:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 [(set GRRegs:$dst, (OpNode GRRegs:$b, immBitp:$c))]>;
+multiclass F3R_2RBITP<bits<5> opc1, bits<5> opc2, string OpcStr,
+                      SDNode OpNode> {
+  def _3r: _F3R<opc1, (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
+                !strconcat(OpcStr, " $dst, $b, $c"),
+                [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
+  def _2rus : _F2RUSBitp<opc2, (outs GRRegs:$dst), (ins GRRegs:$b, i32imm:$c),
+                         !strconcat(OpcStr, " $dst, $b, $c"),
+                         [(set GRRegs:$dst, (OpNode GRRegs:$b, immBitp:$c))]>;
 }
 
-class F3R<string OpcStr, SDNode OpNode> : _F3R<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
+class F3R<bits<5> opc, string OpcStr, SDNode OpNode> :
+  _F3R<opc, (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
+       !strconcat(OpcStr, " $dst, $b, $c"),
+       [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
 
-class F3R_np<string OpcStr> : _F3R<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 []>;
+class F3R_np<bits<5> opc, string OpcStr> :
+  _F3R<opc, (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
+       !strconcat(OpcStr, " $dst, $b, $c"), []>;
 // Three operand long
 
 /// FL3R_L2RUS multiclass - Define a normal FL3R/FL2RUS pattern in one shot.
-multiclass FL3R_L2RUS<string OpcStr, SDNode OpNode> {
-  def _l3r: _FL3R<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
-  def _l2rus : _FL2RUS<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, i32imm:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 [(set GRRegs:$dst, (OpNode GRRegs:$b, immUs:$c))]>;
+multiclass FL3R_L2RUS<bits<9> opc1, bits<9> opc2, string OpcStr,
+                      SDNode OpNode> {
+  def _l3r: _FL3R<opc1, (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
+                  !strconcat(OpcStr, " $dst, $b, $c"),
+                  [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
+  def _l2rus : _FL2RUS<opc2, (outs GRRegs:$dst), (ins GRRegs:$b, i32imm:$c),
+                       !strconcat(OpcStr, " $dst, $b, $c"),
+                       [(set GRRegs:$dst, (OpNode GRRegs:$b, immUs:$c))]>;
 }
 
 /// FL3R_L2RUS multiclass - Define a normal FL3R/FL2RUS pattern in one shot.
-multiclass FL3R_L2RBITP<string OpcStr, SDNode OpNode> {
-  def _l3r: _FL3R<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
-  def _l2rus : _FL2RUS<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, i32imm:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 [(set GRRegs:$dst, (OpNode GRRegs:$b, immBitp:$c))]>;
+multiclass FL3R_L2RBITP<bits<9> opc1, bits<9> opc2, string OpcStr,
+                        SDNode OpNode> {
+  def _l3r: _FL3R<opc1, (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
+                  !strconcat(OpcStr, " $dst, $b, $c"),
+                  [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
+  def _l2rus : _FL2RUSBitp<opc2, (outs GRRegs:$dst), (ins GRRegs:$b, i32imm:$c),
+                           !strconcat(OpcStr, " $dst, $b, $c"),
+                           [(set GRRegs:$dst, (OpNode GRRegs:$b, immBitp:$c))]>;
 }
 
-class FL3R<string OpcStr, SDNode OpNode> : _FL3R<
-                 (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
-                 !strconcat(OpcStr, " $dst, $b, $c"),
-                 [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
+class FL3R<bits<9> opc, string OpcStr, SDNode OpNode> :
+  _FL3R<opc, (outs GRRegs:$dst), (ins GRRegs:$b, GRRegs:$c),
+        !strconcat(OpcStr, " $dst, $b, $c"),
+        [(set GRRegs:$dst, (OpNode GRRegs:$b, GRRegs:$c))]>;
 
 // Register - U6
 // Operand register - U6
-multiclass FRU6_LRU6_branch<string OpcStr> {
-  def _ru6: _FRU6<
-                 (outs), (ins GRRegs:$cond, brtarget:$dest),
-                 !strconcat(OpcStr, " $cond, $dest"),
-                 []>;
-  def _lru6: _FLRU6<
-                 (outs), (ins GRRegs:$cond, brtarget:$dest),
-                 !strconcat(OpcStr, " $cond, $dest"),
-                 []>;
+multiclass FRU6_LRU6_branch<bits<6> opc, string OpcStr> {
+  def _ru6: _FRU6<opc, (outs), (ins GRRegs:$a, brtarget:$b),
+                  !strconcat(OpcStr, " $a, $b"), []>;
+  def _lru6: _FLRU6<opc, (outs), (ins GRRegs:$a, brtarget:$b),
+                    !strconcat(OpcStr, " $a, $b"), []>;
 }
 
-multiclass FRU6_LRU6_cp<string OpcStr> {
-  def _ru6: _FRU6<
-                 (outs GRRegs:$dst), (ins i32imm:$a),
-                 !strconcat(OpcStr, " $dst, cp[$a]"),
-                 []>;
-  def _lru6: _FLRU6<
-                 (outs GRRegs:$dst), (ins i32imm:$a),
-                 !strconcat(OpcStr, " $dst, cp[$a]"),
-                 []>;
+multiclass FRU6_LRU6_backwards_branch<bits<6> opc, string OpcStr> {
+  def _ru6: _FRU6<opc, (outs), (ins GRRegs:$a, brtarget:$b),
+                  !strconcat(OpcStr, " $a, -$b"), []>;
+  def _lru6: _FLRU6<opc, (outs), (ins GRRegs:$a, brtarget:$b),
+                    !strconcat(OpcStr, " $a, -$b"), []>;
 }
 
-// U6
-multiclass FU6_LU6<string OpcStr, SDNode OpNode> {
-  def _u6: _FU6<
-                 (outs), (ins i32imm:$b),
-                 !strconcat(OpcStr, " $b"),
-                 [(OpNode immU6:$b)]>;
-  def _lu6: _FLU6<
-                 (outs), (ins i32imm:$b),
-                 !strconcat(OpcStr, " $b"),
-                 [(OpNode immU16:$b)]>;
+multiclass FRU6_LRU6_cp<bits<6> opc, string OpcStr> {
+  def _ru6: _FRU6<opc, (outs RRegs:$a), (ins i32imm:$b),
+                  !strconcat(OpcStr, " $a, cp[$b]"), []>;
+  def _lru6: _FLRU6<opc, (outs RRegs:$a), (ins i32imm:$b),
+                    !strconcat(OpcStr, " $a, cp[$b]"), []>;
 }
-multiclass FU6_LU6_int<string OpcStr, Intrinsic Int> {
-  def _u6: _FU6<
-                 (outs), (ins i32imm:$b),
-                 !strconcat(OpcStr, " $b"),
-                 [(Int immU6:$b)]>;
-  def _lu6: _FLU6<
-                 (outs), (ins i32imm:$b),
-                 !strconcat(OpcStr, " $b"),
-                 [(Int immU16:$b)]>;
+
+// U6
+multiclass FU6_LU6<bits<10> opc, string OpcStr, SDNode OpNode> {
+  def _u6: _FU6<opc, (outs), (ins i32imm:$a), !strconcat(OpcStr, " $a"),
+                [(OpNode immU6:$a)]>;
+  def _lu6: _FLU6<opc, (outs), (ins i32imm:$a), !strconcat(OpcStr, " $a"),
+                  [(OpNode immU16:$a)]>;
 }
 
-multiclass FU6_LU6_np<string OpcStr> {
-  def _u6: _FU6<
-                 (outs), (ins i32imm:$b),
-                 !strconcat(OpcStr, " $b"),
-                 []>;
-  def _lu6: _FLU6<
-                 (outs), (ins i32imm:$b),
-                 !strconcat(OpcStr, " $b"),
-                 []>;
+multiclass FU6_LU6_int<bits<10> opc, string OpcStr, Intrinsic Int> {
+  def _u6: _FU6<opc, (outs), (ins i32imm:$a), !strconcat(OpcStr, " $a"),
+                [(Int immU6:$a)]>;
+  def _lu6: _FLU6<opc, (outs), (ins i32imm:$a), !strconcat(OpcStr, " $a"),
+                  [(Int immU16:$a)]>;
 }
 
-// U10
-multiclass FU10_LU10_np<string OpcStr> {
-  def _u10: _FU10<
-                 (outs), (ins i32imm:$b),
-                 !strconcat(OpcStr, " $b"),
-                 []>;
-  def _lu10: _FLU10<
-                 (outs), (ins i32imm:$b),
-                 !strconcat(OpcStr, " $b"),
-                 []>;
+multiclass FU6_LU6_np<bits<10> opc, string OpcStr> {
+  def _u6: _FU6<opc, (outs), (ins i32imm:$a), !strconcat(OpcStr, " $a"), []>;
+  def _lu6: _FLU6<opc, (outs), (ins i32imm:$a), !strconcat(OpcStr, " $a"), []>;
 }
 
 // Two operand short
 
-class F2R_np<string OpcStr> : _F2R<
-                 (outs GRRegs:$dst), (ins GRRegs:$b),
-                 !strconcat(OpcStr, " $dst, $b"),
-                 []>;
+class F2R_np<bits<6> opc, string OpcStr> :
+  _F2R<opc, (outs GRRegs:$dst), (ins GRRegs:$b),
+       !strconcat(OpcStr, " $dst, $b"), []>;
 
 // Two operand long
 
@@ -357,23 +321,23 @@ class F2R_np<string OpcStr> : _F2R<
 
 let Defs = [SP], Uses = [SP] in {
 def ADJCALLSTACKDOWN : PseudoInstXCore<(outs), (ins i32imm:$amt),
-                               "${:comment} ADJCALLSTACKDOWN $amt",
+                               "# ADJCALLSTACKDOWN $amt",
                                [(callseq_start timm:$amt)]>;
 def ADJCALLSTACKUP : PseudoInstXCore<(outs), (ins i32imm:$amt1, i32imm:$amt2),
-                            "${:comment} ADJCALLSTACKUP $amt1",
+                            "# ADJCALLSTACKUP $amt1",
                             [(callseq_end timm:$amt1, timm:$amt2)]>;
 }
 
 def LDWFI : PseudoInstXCore<(outs GRRegs:$dst), (ins MEMii:$addr),
-                             "${:comment} LDWFI $dst, $addr",
+                             "# LDWFI $dst, $addr",
                              [(set GRRegs:$dst, (load ADDRspii:$addr))]>;
 
 def LDAWFI : PseudoInstXCore<(outs GRRegs:$dst), (ins MEMii:$addr),
-                             "${:comment} LDAWFI $dst, $addr",
+                             "# LDAWFI $dst, $addr",
                              [(set GRRegs:$dst, ADDRspii:$addr)]>;
 
 def STWFI : PseudoInstXCore<(outs), (ins GRRegs:$src, MEMii:$addr),
-                            "${:comment} STWFI $src, $addr",
+                            "# STWFI $src, $addr",
                             [(store GRRegs:$src, ADDRspii:$addr)]>;
 
 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation.  Expanded after
@@ -381,7 +345,7 @@ def STWFI : PseudoInstXCore<(outs), (ins GRRegs:$src, MEMii:$addr),
 let usesCustomInserter = 1 in {
   def SELECT_CC : PseudoInstXCore<(outs GRRegs:$dst),
                               (ins GRRegs:$cond, GRRegs:$T, GRRegs:$F),
-                              "${:comment} SELECT_CC PSEUDO!",
+                              "# SELECT_CC PSEUDO!",
                               [(set GRRegs:$dst,
                                  (select GRRegs:$cond, GRRegs:$T, GRRegs:$F))]>;
 }
@@ -391,572 +355,564 @@ let usesCustomInserter = 1 in {
 //===----------------------------------------------------------------------===//
 
 // Three operand short
-defm ADD : F3R_2RUS<"add", add>;
-defm SUB : F3R_2RUS<"sub", sub>;
+defm ADD : F3R_2RUS<0b00010, 0b10010, "add", add>;
+defm SUB : F3R_2RUS<0b00011, 0b10011, "sub", sub>;
 let neverHasSideEffects = 1 in {
-defm EQ : F3R_2RUS_np<"eq">;
-def LSS_3r : F3R_np<"lss">;
-def LSU_3r : F3R_np<"lsu">;
+defm EQ : F3R_2RUS_np<0b00110, 0b10110, "eq">;
+def LSS_3r : F3R_np<0b11000, "lss">;
+def LSU_3r : F3R_np<0b11001, "lsu">;
 }
-def AND_3r : F3R<"and", and>;
-def OR_3r : F3R<"or", or>;
+def AND_3r : F3R<0b00111, "and", and>;
+def OR_3r : F3R<0b01000, "or", or>;
 
 let mayLoad=1 in {
-def LDW_3r : _F3R<(outs GRRegs:$dst), (ins GRRegs:$addr, GRRegs:$offset),
-                  "ldw $dst, $addr[$offset]",
-                  []>;
+def LDW_3r : _F3R<0b01001, (outs GRRegs:$dst),
+                  (ins GRRegs:$addr, GRRegs:$offset),
+                  "ldw $dst, $addr[$offset]", []>;
 
-def LDW_2rus : _F2RUS<(outs GRRegs:$dst), (ins GRRegs:$addr, i32imm:$offset),
-                  "ldw $dst, $addr[$offset]",
-                  []>;
+def LDW_2rus : _F2RUS<0b00001, (outs GRRegs:$dst),
+                      (ins GRRegs:$addr, i32imm:$offset),
+                      "ldw $dst, $addr[$offset]", []>;
 
-def LD16S_3r :  _F3R<(outs GRRegs:$dst), (ins GRRegs:$addr, GRRegs:$offset),
-                  "ld16s $dst, $addr[$offset]",
-                  []>;
+def LD16S_3r :  _F3R<0b10000, (outs GRRegs:$dst),
+                     (ins GRRegs:$addr, GRRegs:$offset),
+                     "ld16s $dst, $addr[$offset]", []>;
 
-def LD8U_3r :  _F3R<(outs GRRegs:$dst), (ins GRRegs:$addr, GRRegs:$offset),
-                  "ld8u $dst, $addr[$offset]",
-                  []>;
+def LD8U_3r :  _F3R<0b10001, (outs GRRegs:$dst),
+                    (ins GRRegs:$addr, GRRegs:$offset),
+                    "ld8u $dst, $addr[$offset]", []>;
 }
 
 let mayStore=1 in {
-def STW_3r : _F3R<(outs), (ins GRRegs:$val, GRRegs:$addr, GRRegs:$offset),
-                  "stw $val, $addr[$offset]",
-                  []>;
+def STW_l3r : _FL3R<0b000001100, (outs),
+                    (ins GRRegs:$val, GRRegs:$addr, GRRegs:$offset),
+                    "stw $val, $addr[$offset]", []>;
 
-def STW_2rus : _F2RUS<(outs), (ins GRRegs:$val, GRRegs:$addr, i32imm:$offset),
-                  "stw $val, $addr[$offset]",
-                  []>;
+def STW_2rus : _F2RUS<0b0000, (outs),
+                      (ins GRRegs:$val, GRRegs:$addr, i32imm:$offset),
+                      "stw $val, $addr[$offset]", []>;
 }
 
-defm SHL : F3R_2RBITP<"shl", shl>;
-defm SHR : F3R_2RBITP<"shr", srl>;
-// TODO tsetr
+defm SHL : F3R_2RBITP<0b00100, 0b10100, "shl", shl>;
+defm SHR : F3R_2RBITP<0b00101, 0b10101, "shr", srl>;
+
+// The first operand is treated as an immediate since it refers to a register
+// number in another thread.
+def TSETR_3r : _F3RImm<0b10111, (outs), (ins i32imm:$a, GRRegs:$b, GRRegs:$c),
+                       "set t[$c]:r$a, $b", []>;
 
 // Three operand long
-def LDAWF_l3r : _FL3R<(outs GRRegs:$dst), (ins GRRegs:$addr, GRRegs:$offset),
-                  "ldaw $dst, $addr[$offset]",
-                  [(set GRRegs:$dst, (ldawf GRRegs:$addr, GRRegs:$offset))]>;
+def LDAWF_l3r : _FL3R<0b000111100, (outs GRRegs:$dst),
+                      (ins GRRegs:$addr, GRRegs:$offset),
+                      "ldaw $dst, $addr[$offset]",
+                      [(set GRRegs:$dst,
+                         (ldawf GRRegs:$addr, GRRegs:$offset))]>;
 
 let neverHasSideEffects = 1 in
-def LDAWF_l2rus : _FL2RUS<(outs GRRegs:$dst),
-                    (ins GRRegs:$addr, i32imm:$offset),
-                    "ldaw $dst, $addr[$offset]",
-                    []>;
+def LDAWF_l2rus : _FL2RUS<0b100111100, (outs GRRegs:$dst),
+                          (ins GRRegs:$addr, i32imm:$offset),
+                          "ldaw $dst, $addr[$offset]", []>;
 
-def LDAWB_l3r : _FL3R<(outs GRRegs:$dst), (ins GRRegs:$addr, GRRegs:$offset),
-                  "ldaw $dst, $addr[-$offset]",
-                  [(set GRRegs:$dst, (ldawb GRRegs:$addr, GRRegs:$offset))]>;
+def LDAWB_l3r : _FL3R<0b001001100, (outs GRRegs:$dst),
+                      (ins GRRegs:$addr, GRRegs:$offset),
+                      "ldaw $dst, $addr[-$offset]",
+                      [(set GRRegs:$dst,
+                         (ldawb GRRegs:$addr, GRRegs:$offset))]>;
 
 let neverHasSideEffects = 1 in
-def LDAWB_l2rus : _FL2RUS<(outs GRRegs:$dst),
-                    (ins GRRegs:$addr, i32imm:$offset),
-                    "ldaw $dst, $addr[-$offset]",
-                    []>;
-
-def LDA16F_l3r : _FL3R<(outs GRRegs:$dst), (ins GRRegs:$addr, GRRegs:$offset),
-                  "lda16 $dst, $addr[$offset]",
-                  [(set GRRegs:$dst, (lda16f GRRegs:$addr, GRRegs:$offset))]>;
-
-def LDA16B_l3r : _FL3R<(outs GRRegs:$dst), (ins GRRegs:$addr, GRRegs:$offset),
-                  "lda16 $dst, $addr[-$offset]",
-                  [(set GRRegs:$dst, (lda16b GRRegs:$addr, GRRegs:$offset))]>;
-
-def MUL_l3r : FL3R<"mul", mul>;
+def LDAWB_l2rus : _FL2RUS<0b101001100, (outs GRRegs:$dst),
+                         (ins GRRegs:$addr, i32imm:$offset),
+                         "ldaw $dst, $addr[-$offset]", []>;
+
+def LDA16F_l3r : _FL3R<0b001011100, (outs GRRegs:$dst),
+                       (ins GRRegs:$addr, GRRegs:$offset),
+                       "lda16 $dst, $addr[$offset]",
+                       [(set GRRegs:$dst,
+                          (lda16f GRRegs:$addr, GRRegs:$offset))]>;
+
+def LDA16B_l3r : _FL3R<0b001101100, (outs GRRegs:$dst),
+                       (ins GRRegs:$addr, GRRegs:$offset),
+                       "lda16 $dst, $addr[-$offset]",
+                       [(set GRRegs:$dst,
+                          (lda16b GRRegs:$addr, GRRegs:$offset))]>;
+
+def MUL_l3r : FL3R<0b001111100, "mul", mul>;
 // Instructions which may trap are marked as side effecting.
 let hasSideEffects = 1 in {
-def DIVS_l3r : FL3R<"divs", sdiv>;
-def DIVU_l3r : FL3R<"divu", udiv>;
-def REMS_l3r : FL3R<"rems", srem>;
-def REMU_l3r : FL3R<"remu", urem>;
+def DIVS_l3r : FL3R<0b010001100, "divs", sdiv>;
+def DIVU_l3r : FL3R<0b010011100, "divu", udiv>;
+def REMS_l3r : FL3R<0b110001100, "rems", srem>;
+def REMU_l3r : FL3R<0b110011100, "remu", urem>;
 }
-def XOR_l3r : FL3R<"xor", xor>;
-defm ASHR : FL3R_L2RBITP<"ashr", sra>;
+def XOR_l3r : FL3R<0b000011100, "xor", xor>;
+defm ASHR : FL3R_L2RBITP<0b000101100, 0b100101100, "ashr", sra>;
 
 let Constraints = "$src1 = $dst" in
-def CRC_l3r : _FL3R<(outs GRRegs:$dst),
-                     (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3),
-                     "crc32 $dst, $src2, $src3",
-                     [(set GRRegs:$dst,
-                        (int_xcore_crc32 GRRegs:$src1, GRRegs:$src2,
-                                         GRRegs:$src3))]>;
+def CRC_l3r : _FL3RSrcDst<0b101011100, (outs GRRegs:$dst),
+                          (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3),
+                          "crc32 $dst, $src2, $src3",
+                          [(set GRRegs:$dst,
+                             (int_xcore_crc32 GRRegs:$src1, GRRegs:$src2,
+                                              GRRegs:$src3))]>;
 
-// TODO inpw, outpw
 let mayStore=1 in {
-def ST16_l3r : _FL3R<(outs), (ins GRRegs:$val, GRRegs:$addr, GRRegs:$offset),
-                "st16 $val, $addr[$offset]",
-                []>;
+def ST16_l3r : _FL3R<0b100001100, (outs),
+                     (ins GRRegs:$val, GRRegs:$addr, GRRegs:$offset),
+                     "st16 $val, $addr[$offset]", []>;
 
-def ST8_l3r : _FL3R<(outs), (ins GRRegs:$val, GRRegs:$addr, GRRegs:$offset),
-                "st8 $val, $addr[$offset]",
-                []>;
+def ST8_l3r : _FL3R<0b100011100, (outs),
+                    (ins GRRegs:$val, GRRegs:$addr, GRRegs:$offset),
+                    "st8 $val, $addr[$offset]", []>;
 }
 
-// Four operand long
-let Constraints = "$src1 = $dst1,$src2 = $dst2" in {
-def MACCU_l4r : _L4R<(outs GRRegs:$dst1, GRRegs:$dst2),
-                    (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3,
-                      GRRegs:$src4),
-                    "maccu $dst1, $dst2, $src3, $src4",
-                    []>;
+def INPW_l2rus : _FL2RUSBitp<0b100101110, (outs GRRegs:$a),
+                             (ins GRRegs:$b, i32imm:$c), "inpw $a, res[$b], $c",
+                             []>;
 
-def MACCS_l4r : _L4R<(outs GRRegs:$dst1, GRRegs:$dst2),
-                    (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3,
-                      GRRegs:$src4),
-                    "maccs $dst1, $dst2, $src3, $src4",
-                    []>;
+def OUTPW_l2rus : _FL2RUSBitp<0b100101101, (outs),
+                              (ins GRRegs:$a, GRRegs:$b, i32imm:$c),
+                              "outpw res[$b], $a, $c", []>;
+
+// Four operand long
+let Constraints = "$e = $a,$f = $b" in {
+def MACCU_l4r : _FL4RSrcDstSrcDst<
+  0b000001, (outs GRRegs:$a, GRRegs:$b),
+  (ins GRRegs:$e, GRRegs:$f, GRRegs:$c, GRRegs:$d), "maccu $a, $b, $c, $d", []>;
+
+def MACCS_l4r : _FL4RSrcDstSrcDst<
+  0b000010, (outs GRRegs:$a, GRRegs:$b),
+  (ins GRRegs:$e, GRRegs:$f, GRRegs:$c, GRRegs:$d), "maccs $a, $b, $c, $d", []>;
 }
 
-let Constraints = "$src1 = $dst1" in
-def CRC8_l4r : _L4R<(outs GRRegs:$dst1, GRRegs:$dst2),
-                    (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3),
-                    "crc8 $dst1, $dst2, $src2, $src3",
-                    []>;
+let Constraints = "$e = $b" in
+def CRC8_l4r : _FL4RSrcDst<0b000000, (outs GRRegs:$a, GRRegs:$b),
+                           (ins GRRegs:$e, GRRegs:$c, GRRegs:$d),
+                           "crc8 $b, $a, $c, $d", []>;
 
 // Five operand long
 
-def LADD_l5r : _L5R<(outs GRRegs:$dst1, GRRegs:$dst2),
-                    (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3),
-                    "ladd $dst1, $dst2, $src1, $src2, $src3",
-                    []>;
+def LADD_l5r : _FL5R<0b000001, (outs GRRegs:$dst1, GRRegs:$dst2),
+                     (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3),
+                     "ladd $dst2, $dst1, $src1, $src2, $src3",
+                     []>;
 
-def LSUB_l5r : _L5R<(outs GRRegs:$dst1, GRRegs:$dst2),
-                    (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3),
-                    "lsub $dst1, $dst2, $src1, $src2, $src3",
-                    []>;
+def LSUB_l5r : _FL5R<0b000010, (outs GRRegs:$dst1, GRRegs:$dst2),
+                     (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3),
+                     "lsub $dst2, $dst1, $src1, $src2, $src3", []>;
 
-def LDIV_l5r : _L5R<(outs GRRegs:$dst1, GRRegs:$dst2),
-                    (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3),
-                    "ldiv $dst1, $dst2, $src1, $src2, $src3",
-                    []>;
+def LDIVU_l5r : _FL5R<0b000000, (outs GRRegs:$dst1, GRRegs:$dst2),
+                      (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3),
+                      "ldivu $dst1, $dst2, $src3, $src1, $src2", []>;
 
 // Six operand long
 
-def LMUL_l6r : _L6R<(outs GRRegs:$dst1, GRRegs:$dst2),
-                    (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3,
-                      GRRegs:$src4),
-                    "lmul $dst1, $dst2, $src1, $src2, $src3, $src4",
-                    []>;
+def LMUL_l6r : _FL6R<
+  0b00000, (outs GRRegs:$dst1, GRRegs:$dst2),
+  (ins GRRegs:$src1, GRRegs:$src2, GRRegs:$src3, GRRegs:$src4),
+  "lmul $dst1, $dst2, $src1, $src2, $src3, $src4", []>;
 
 // Register - U6
 
 //let Uses = [DP] in ...
 let neverHasSideEffects = 1, isReMaterializable = 1 in
-def LDAWDP_ru6: _FRU6<(outs GRRegs:$dst), (ins MEMii:$a),
-                    "ldaw $dst, dp[$a]",
-                    []>;
+def LDAWDP_ru6: _FRU6<0b011000, (outs RRegs:$a), (ins MEMii:$b),
+                      "ldaw $a, dp[$b]", []>;
 
 let isReMaterializable = 1 in                    
-def LDAWDP_lru6: _FLRU6<
-                    (outs GRRegs:$dst), (ins MEMii:$a),
-                    "ldaw $dst, dp[$a]",
-                    [(set GRRegs:$dst, ADDRdpii:$a)]>;
+def LDAWDP_lru6: _FLRU6<0b011000, (outs RRegs:$a), (ins MEMii:$b),
+                        "ldaw $a, dp[$b]",
+                        [(set RRegs:$a, ADDRdpii:$b)]>;
 
 let mayLoad=1 in
-def LDWDP_ru6: _FRU6<(outs GRRegs:$dst), (ins MEMii:$a),
-                    "ldw $dst, dp[$a]",
-                    []>;
-                    
-def LDWDP_lru6: _FLRU6<
-                    (outs GRRegs:$dst), (ins MEMii:$a),
-                    "ldw $dst, dp[$a]",
-                    [(set GRRegs:$dst, (load ADDRdpii:$a))]>;
+def LDWDP_ru6: _FRU6<0b010110, (outs RRegs:$a), (ins MEMii:$b),
+                     "ldw $a, dp[$b]", []>;
+
+def LDWDP_lru6: _FLRU6<0b010110, (outs RRegs:$a), (ins MEMii:$b),
+                       "ldw $a, dp[$b]",
+                       [(set RRegs:$a, (load ADDRdpii:$b))]>;
 
 let mayStore=1 in
-def STWDP_ru6 : _FRU6<(outs), (ins GRRegs:$val, MEMii:$addr),
-                  "stw $val, dp[$addr]",
-                  []>;
+def STWDP_ru6 : _FRU6<0b010100, (outs), (ins RRegs:$a, MEMii:$b),
+                      "stw $a, dp[$b]", []>;
 
-def STWDP_lru6 : _FLRU6<(outs), (ins GRRegs:$val, MEMii:$addr),
-                  "stw $val, dp[$addr]",
-                  [(store GRRegs:$val, ADDRdpii:$addr)]>;
+def STWDP_lru6 : _FLRU6<0b010100, (outs), (ins RRegs:$a, MEMii:$b),
+                        "stw $a, dp[$b]",
+                        [(store RRegs:$a, ADDRdpii:$b)]>;
 
 //let Uses = [CP] in ..
 let mayLoad = 1, isReMaterializable = 1, neverHasSideEffects = 1 in
-defm LDWCP : FRU6_LRU6_cp<"ldw">;
+defm LDWCP : FRU6_LRU6_cp<0b011011, "ldw">;
 
 let Uses = [SP] in {
 let mayStore=1 in {
-def STWSP_ru6 : _FRU6<
-                 (outs), (ins GRRegs:$val, i32imm:$index),
-                 "stw $val, sp[$index]",
-                 [(XCoreStwsp GRRegs:$val, immU6:$index)]>;
-
-def STWSP_lru6 : _FLRU6<
-                 (outs), (ins GRRegs:$val, i32imm:$index),
-                 "stw $val, sp[$index]",
-                 [(XCoreStwsp GRRegs:$val, immU16:$index)]>;
+def STWSP_ru6 : _FRU6<0b010101, (outs), (ins RRegs:$a, i32imm:$b),
+                      "stw $a, sp[$b]",
+                      [(XCoreStwsp RRegs:$a, immU6:$b)]>;
+
+def STWSP_lru6 : _FLRU6<0b010101, (outs), (ins RRegs:$a, i32imm:$b),
+                        "stw $a, sp[$b]",
+                        [(XCoreStwsp RRegs:$a, immU16:$b)]>;
 }
 
 let mayLoad=1 in {
-def LDWSP_ru6 : _FRU6<
-                 (outs GRRegs:$dst), (ins i32imm:$b),
-                 "ldw $dst, sp[$b]",
-                 []>;
+def LDWSP_ru6 : _FRU6<0b010111, (outs RRegs:$a), (ins i32imm:$b),
+                      "ldw $a, sp[$b]", []>;
 
-def LDWSP_lru6 : _FLRU6<
-                 (outs GRRegs:$dst), (ins i32imm:$b),
-                 "ldw $dst, sp[$b]",
-                 []>;
+def LDWSP_lru6 : _FLRU6<0b010111, (outs RRegs:$a), (ins i32imm:$b),
+                        "ldw $a, sp[$b]", []>;
 }
 
 let neverHasSideEffects = 1 in {
-def LDAWSP_ru6 : _FRU6<
-                 (outs GRRegs:$dst), (ins i32imm:$b),
-                 "ldaw $dst, sp[$b]",
-                 []>;
+def LDAWSP_ru6 : _FRU6<0b011001, (outs RRegs:$a), (ins i32imm:$b),
+                       "ldaw $a, sp[$b]", []>;
 
-def LDAWSP_lru6 : _FLRU6<
-                 (outs GRRegs:$dst), (ins i32imm:$b),
-                 "ldaw $dst, sp[$b]",
-                 []>;
-
-def LDAWSP_ru6_RRegs : _FRU6<
-                 (outs RRegs:$dst), (ins i32imm:$b),
-                 "ldaw $dst, sp[$b]",
-                 []>;
-
-def LDAWSP_lru6_RRegs : _FLRU6<
-                 (outs RRegs:$dst), (ins i32imm:$b),
-                 "ldaw $dst, sp[$b]",
-                 []>;
+def LDAWSP_lru6 : _FLRU6<0b011001, (outs RRegs:$a), (ins i32imm:$b),
+                         "ldaw $a, sp[$b]", []>;
 }
 }
 
 let isReMaterializable = 1 in {
-def LDC_ru6 : _FRU6<
-                 (outs GRRegs:$dst), (ins i32imm:$b),
-                 "ldc $dst, $b",
-                 [(set GRRegs:$dst, immU6:$b)]>;
-
-def LDC_lru6 : _FLRU6<
-                 (outs GRRegs:$dst), (ins i32imm:$b),
-                 "ldc $dst, $b",
-                 [(set GRRegs:$dst, immU16:$b)]>;
+def LDC_ru6 : _FRU6<0b011010, (outs RRegs:$a), (ins i32imm:$b),
+                    "ldc $a, $b", [(set RRegs:$a, immU6:$b)]>;
+
+def LDC_lru6 : _FLRU6<0b011010, (outs RRegs:$a), (ins i32imm:$b),
+                      "ldc $a, $b", [(set RRegs:$a, immU16:$b)]>;
 }
 
-def SETC_ru6 : _FRU6<(outs), (ins GRRegs:$r, i32imm:$val),
-                  "setc res[$r], $val",
-                  [(int_xcore_setc GRRegs:$r, immU6:$val)]>;
+def SETC_ru6 : _FRU6<0b111010, (outs), (ins GRRegs:$a, i32imm:$b),
+                     "setc res[$a], $b",
+                     [(int_xcore_setc GRRegs:$a, immU6:$b)]>;
 
-def SETC_lru6 : _FLRU6<(outs), (ins GRRegs:$r, i32imm:$val),
-                  "setc res[$r], $val",
-                  [(int_xcore_setc GRRegs:$r, immU16:$val)]>;
+def SETC_lru6 : _FLRU6<0b111010, (outs), (ins GRRegs:$a, i32imm:$b),
+                       "setc res[$a], $b",
+                       [(int_xcore_setc GRRegs:$a, immU16:$b)]>;
 
 // Operand register - U6
 let isBranch = 1, isTerminator = 1 in {
-defm BRFT: FRU6_LRU6_branch<"bt">;
-defm BRBT: FRU6_LRU6_branch<"bt">;
-defm BRFF: FRU6_LRU6_branch<"bf">;
-defm BRBF: FRU6_LRU6_branch<"bf">;
+defm BRFT: FRU6_LRU6_branch<0b011100, "bt">;
+defm BRBT: FRU6_LRU6_backwards_branch<0b011101, "bt">;
+defm BRFF: FRU6_LRU6_branch<0b011110, "bf">;
+defm BRBF: FRU6_LRU6_backwards_branch<0b011111, "bf">;
 }
 
 // U6
 let Defs = [SP], Uses = [SP] in {
 let neverHasSideEffects = 1 in
-defm EXTSP : FU6_LU6_np<"extsp">;
+defm EXTSP : FU6_LU6_np<0b0111011110, "extsp">;
+
 let mayStore = 1 in
-defm ENTSP : FU6_LU6_np<"entsp">;
+defm ENTSP : FU6_LU6_np<0b0111011101, "entsp">;
 
 let isReturn = 1, isTerminator = 1, mayLoad = 1, isBarrier = 1 in {
-defm RETSP : FU6_LU6<"retsp", XCoreRetsp>;
+defm RETSP : FU6_LU6<0b0111011111, "retsp", XCoreRetsp>;
 }
 }
 
-// TODO extdp, kentsp, krestsp, blat
-// getsr, kalli
+let neverHasSideEffects = 1 in
+defm EXTDP : FU6_LU6_np<0b0111001110, "extdp">;
+
+let Uses = [R11], isCall=1 in
+defm BLAT : FU6_LU6_np<0b0111001101, "blat">;
+
 let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
-def BRBU_u6 : _FU6<
-                 (outs),
-                 (ins brtarget:$target),
-                 "bu $target",
-                 []>;
+def BRBU_u6 : _FU6<0b0111011100, (outs), (ins brtarget:$a), "bu -$a", []>;
 
-def BRBU_lu6 : _FLU6<
-                 (outs),
-                 (ins brtarget:$target),
-                 "bu $target",
-                 []>;
+def BRBU_lu6 : _FLU6<0b0111011100, (outs), (ins brtarget:$a), "bu -$a", []>;
 
-def BRFU_u6 : _FU6<
-                 (outs),
-                 (ins brtarget:$target),
-                 "bu $target",
-                 []>;
+def BRFU_u6 : _FU6<0b0111001100, (outs), (ins brtarget:$a), "bu $a", []>;
 
-def BRFU_lu6 : _FLU6<
-                 (outs),
-                 (ins brtarget:$target),
-                 "bu $target",
-                 []>;
+def BRFU_lu6 : _FLU6<0b0111001100, (outs), (ins brtarget:$a), "bu $a", []>;
 }
 
 //let Uses = [CP] in ...
 let Defs = [R11], neverHasSideEffects = 1, isReMaterializable = 1 in
-def LDAWCP_u6: _FRU6<(outs), (ins MEMii:$a),
-                    "ldaw r11, cp[$a]",
+def LDAWCP_u6: _FU6<0b0111111101, (outs), (ins MEMii:$a), "ldaw r11, cp[$a]",
                     []>;
 
 let Defs = [R11], isReMaterializable = 1 in
-def LDAWCP_lu6: _FLRU6<
-                    (outs), (ins MEMii:$a),
-                    "ldaw r11, cp[$a]",
-                    [(set R11, ADDRcpii:$a)]>;
+def LDAWCP_lu6: _FLU6<0b0111111101, (outs), (ins MEMii:$a), "ldaw r11, cp[$a]",
+                      [(set R11, ADDRcpii:$a)]>;
+
+let Defs = [R11] in
+defm GETSR : FU6_LU6_np<0b0111111100, "getsr r11,">;
 
-defm SETSR : FU6_LU6_int<"setsr", int_xcore_setsr>;
+defm SETSR : FU6_LU6_int<0b0111101101, "setsr", int_xcore_setsr>;
 
-defm CLRSR : FU6_LU6_int<"clrsr", int_xcore_clrsr>;
+defm CLRSR : FU6_LU6_int<0b0111101100, "clrsr", int_xcore_clrsr>;
 
 // setsr may cause a branch if it is used to enable events. clrsr may
 // branch if it is executed while events are enabled.
-let isBranch=1, isIndirectBranch=1, isTerminator=1, isBarrier = 1 in {
-defm SETSR_branch : FU6_LU6_np<"setsr">;
-defm CLRSR_branch : FU6_LU6_np<"clrsr">;
+let isBranch=1, isIndirectBranch=1, isTerminator=1, isBarrier = 1,
+    isCodeGenOnly = 1 in {
+defm SETSR_branch : FU6_LU6_np<0b0111101101, "setsr">;
+defm CLRSR_branch : FU6_LU6_np<0b0111101100, "clrsr">;
 }
 
+defm KCALL : FU6_LU6_np<0b0111001111, "kcall">;
+
+let Uses = [SP], Defs = [SP], mayStore = 1 in
+defm KENTSP : FU6_LU6_np<0b0111101110, "kentsp">;
+
+let Uses = [SP], Defs = [SP], mayLoad = 1 in
+defm KRESTSP : FU6_LU6_np<0b0111101111, "krestsp">;
+
 // U10
-// TODO ldwcpl, blacp
 
 let Defs = [R11], isReMaterializable = 1, neverHasSideEffects = 1 in
-def LDAP_u10 : _FU10<
-                  (outs),
-                  (ins i32imm:$addr),
-                  "ldap r11, $addr",
-                  []>;
+def LDAPF_u10 : _FU10<0b110110, (outs), (ins i32imm:$a), "ldap r11, $a", []>;
 
 let Defs = [R11], isReMaterializable = 1 in
-def LDAP_lu10 : _FLU10<
-                  (outs),
-                  (ins i32imm:$addr),
-                  "ldap r11, $addr",
-                  [(set R11, (pcrelwrapper tglobaladdr:$addr))]>;
+def LDAPF_lu10 : _FLU10<0b110110, (outs), (ins i32imm:$a), "ldap r11, $a",
+                        [(set R11, (pcrelwrapper tglobaladdr:$a))]>;
 
-let Defs = [R11], isReMaterializable = 1 in
-def LDAP_lu10_ba : _FLU10<(outs),
-                          (ins i32imm:$addr),
-                          "ldap r11, $addr",
-                          [(set R11, (pcrelwrapper tblockaddress:$addr))]>;
+let Defs = [R11], isReMaterializable = 1, isCodeGenOnly = 1 in
+def LDAPF_lu10_ba : _FLU10<0b110110, (outs), (ins i32imm:$a), "ldap r11, $a",
+                           [(set R11, (pcrelwrapper tblockaddress:$a))]>;
 
 let isCall=1,
 // All calls clobber the link register and the non-callee-saved registers:
 Defs = [R0, R1, R2, R3, R11, LR], Uses = [SP] in {
-def BL_u10 : _FU10<
-                  (outs), (ins calltarget:$target),
-                  "bl $target",
-                  [(XCoreBranchLink immU10:$target)]>;
-
-def BL_lu10 : _FLU10<
-                  (outs), (ins calltarget:$target),
-                  "bl $target",
-                  [(XCoreBranchLink immU20:$target)]>;
+def BLACP_u10 : _FU10<0b111000, (outs), (ins i32imm:$a), "bla cp[$a]", []>;
+
+def BLACP_lu10 : _FLU10<0b111000, (outs), (ins i32imm:$a), "bla cp[$a]", []>;
+
+def BLRF_u10 : _FU10<0b110100, (outs), (ins calltarget:$a), "bl $a",
+                     [(XCoreBranchLink immU10:$a)]>;
+
+def BLRF_lu10 : _FLU10<0b110100, (outs), (ins calltarget:$a), "bl $a",
+                       [(XCoreBranchLink immU20:$a)]>;
+}
+
+let Defs = [R11], mayLoad = 1, isReMaterializable = 1,
+    neverHasSideEffects = 1 in {
+def LDWCP_u10 : _FU10<0b111001, (outs), (ins i32imm:$a), "ldw r11, cp[$a]", []>;
+
+def LDWCP_lu10 : _FLU10<0b111001, (outs), (ins i32imm:$a), "ldw r11, cp[$a]",
+                        []>;
 }
 
 // Two operand short
-// TODO eet, eef, tsetmr
-def NOT : _F2R<(outs GRRegs:$dst), (ins GRRegs:$b),
-                 "not $dst, $b",
-                 [(set GRRegs:$dst, (not GRRegs:$b))]>;
+def NOT : _F2R<0b100010, (outs GRRegs:$dst), (ins GRRegs:$b),
+                "not $dst, $b", [(set GRRegs:$dst, (not GRRegs:$b))]>;
 
-def NEG : _F2R<(outs GRRegs:$dst), (ins GRRegs:$b),
-                 "neg $dst, $b",
-                 [(set GRRegs:$dst, (ineg GRRegs:$b))]>;
+def NEG : _F2R<0b100100, (outs GRRegs:$dst), (ins GRRegs:$b),
+                "neg $dst, $b", [(set GRRegs:$dst, (ineg GRRegs:$b))]>;
 
 let Constraints = "$src1 = $dst" in {
-def SEXT_rus : _FRUS<(outs GRRegs:$dst), (ins GRRegs:$src1, i32imm:$src2),
-                      "sext $dst, $src2",
-                      [(set GRRegs:$dst, (int_xcore_sext GRRegs:$src1,
-                                                         immBitp:$src2))]>;
-
-def SEXT_2r : _FRUS<(outs GRRegs:$dst), (ins GRRegs:$src1, GRRegs:$src2),
-                     "sext $dst, $src2",
-                     [(set GRRegs:$dst, (int_xcore_sext GRRegs:$src1,
-                                                        GRRegs:$src2))]>;
-
-def ZEXT_rus : _FRUS<(outs GRRegs:$dst), (ins GRRegs:$src1, i32imm:$src2),
-                      "zext $dst, $src2",
-                      [(set GRRegs:$dst, (int_xcore_zext GRRegs:$src1,
-                                                         immBitp:$src2))]>;
-
-def ZEXT_2r : _FRUS<(outs GRRegs:$dst), (ins GRRegs:$src1, GRRegs:$src2),
-                     "zext $dst, $src2",
-                     [(set GRRegs:$dst, (int_xcore_zext GRRegs:$src1,
-                                                        GRRegs:$src2))]>;
-
-def ANDNOT_2r : _F2R<(outs GRRegs:$dst), (ins GRRegs:$src1, GRRegs:$src2),
-                 "andnot $dst, $src2",
-                 [(set GRRegs:$dst, (and GRRegs:$src1, (not GRRegs:$src2)))]>;
+def SEXT_rus :
+  _FRUSSrcDstBitp<0b001101, (outs GRRegs:$dst), (ins GRRegs:$src1, i32imm:$src2),
+                  "sext $dst, $src2",
+                  [(set GRRegs:$dst, (int_xcore_sext GRRegs:$src1,
+                                                     immBitp:$src2))]>;
+
+def SEXT_2r :
+  _F2RSrcDst<0b001100, (outs GRRegs:$dst), (ins GRRegs:$src1, GRRegs:$src2),
+             "sext $dst, $src2",
+             [(set GRRegs:$dst, (int_xcore_sext GRRegs:$src1, GRRegs:$src2))]>;
+
+def ZEXT_rus :
+  _FRUSSrcDstBitp<0b010001, (outs GRRegs:$dst), (ins GRRegs:$src1, i32imm:$src2),
+                  "zext $dst, $src2",
+                  [(set GRRegs:$dst, (int_xcore_zext GRRegs:$src1,
+                                                     immBitp:$src2))]>;
+
+def ZEXT_2r :
+  _F2RSrcDst<0b010000, (outs GRRegs:$dst), (ins GRRegs:$src1, GRRegs:$src2),
+             "zext $dst, $src2",
+             [(set GRRegs:$dst, (int_xcore_zext GRRegs:$src1, GRRegs:$src2))]>;
+
+def ANDNOT_2r :
+  _F2RSrcDst<0b001010, (outs GRRegs:$dst), (ins GRRegs:$src1, GRRegs:$src2),
+             "andnot $dst, $src2",
+             [(set GRRegs:$dst, (and GRRegs:$src1, (not GRRegs:$src2)))]>;
 }
 
 let isReMaterializable = 1, neverHasSideEffects = 1 in
-def MKMSK_rus : _FRUS<(outs GRRegs:$dst), (ins i32imm:$size),
-                 "mkmsk $dst, $size",
-                 []>;
+def MKMSK_rus : _FRUSBitp<0b101001, (outs GRRegs:$dst), (ins i32imm:$size),
+                          "mkmsk $dst, $size", []>;
 
-def MKMSK_2r : _FRUS<(outs GRRegs:$dst), (ins GRRegs:$size),
-                 "mkmsk $dst, $size",
-                 [(set GRRegs:$dst, (add (shl 1, GRRegs:$size), -1))]>;
+def MKMSK_2r : _F2R<0b101000, (outs GRRegs:$dst), (ins GRRegs:$size),
+                    "mkmsk $dst, $size",
+                    [(set GRRegs:$dst, (add (shl 1, GRRegs:$size), -1))]>;
 
-def GETR_rus : _FRUS<(outs GRRegs:$dst), (ins i32imm:$type),
-                 "getr $dst, $type",
-                 [(set GRRegs:$dst, (int_xcore_getr immUs:$type))]>;
+def GETR_rus : _FRUS<0b100000, (outs GRRegs:$dst), (ins i32imm:$type),
+                     "getr $dst, $type",
+                     [(set GRRegs:$dst, (int_xcore_getr immUs:$type))]>;
 
-def GETTS_2r : _F2R<(outs GRRegs:$dst), (ins GRRegs:$r),
-                 "getts $dst, res[$r]",
-                 [(set GRRegs:$dst, (int_xcore_getts GRRegs:$r))]>;
+def GETTS_2r : _F2R<0b001110, (outs GRRegs:$dst), (ins GRRegs:$r),
+                    "getts $dst, res[$r]",
+                    [(set GRRegs:$dst, (int_xcore_getts GRRegs:$r))]>;
 
-def SETPT_2r : _F2R<(outs), (ins GRRegs:$r, GRRegs:$val),
-                 "setpt res[$r], $val",
-                 [(int_xcore_setpt GRRegs:$r, GRRegs:$val)]>;
+def SETPT_2r : _FR2R<0b001111, (outs), (ins GRRegs:$r, GRRegs:$val),
+                     "setpt res[$r], $val",
+                     [(int_xcore_setpt GRRegs:$r, GRRegs:$val)]>;
 
-def OUTCT_2r : _F2R<(outs), (ins GRRegs:$r, GRRegs:$val),
-                 "outct res[$r], $val",
-                 [(int_xcore_outct GRRegs:$r, GRRegs:$val)]>;
+def OUTCT_2r : _F2R<0b010010, (outs), (ins GRRegs:$r, GRRegs:$val),
+                    "outct res[$r], $val",
+                    [(int_xcore_outct GRRegs:$r, GRRegs:$val)]>;
 
-def OUTCT_rus : _F2R<(outs), (ins GRRegs:$r, i32imm:$val),
-                 "outct res[$r], $val",
-                 [(int_xcore_outct GRRegs:$r, immUs:$val)]>;
+def OUTCT_rus : _FRUS<0b010011, (outs), (ins GRRegs:$r, i32imm:$val),
+                       "outct res[$r], $val",
+                       [(int_xcore_outct GRRegs:$r, immUs:$val)]>;
 
-def OUTT_2r : _F2R<(outs), (ins GRRegs:$r, GRRegs:$val),
-                 "outt res[$r], $val",
-                 [(int_xcore_outt GRRegs:$r, GRRegs:$val)]>;
+def OUTT_2r : _FR2R<0b000011, (outs), (ins GRRegs:$r, GRRegs:$val),
+                    "outt res[$r], $val",
+                    [(int_xcore_outt GRRegs:$r, GRRegs:$val)]>;
 
-def OUT_2r : _F2R<(outs), (ins GRRegs:$r, GRRegs:$val),
-                 "out res[$r], $val",
-                 [(int_xcore_out GRRegs:$r, GRRegs:$val)]>;
+def OUT_2r : _FR2R<0b101010, (outs), (ins GRRegs:$r, GRRegs:$val),
+                   "out res[$r], $val",
+                   [(int_xcore_out GRRegs:$r, GRRegs:$val)]>;
 
 let Constraints = "$src = $dst" in
-def OUTSHR_2r : _F2R<(outs GRRegs:$dst), (ins GRRegs:$r, GRRegs:$src),
-                 "outshr res[$r], $src",
-                 [(set GRRegs:$dst, (int_xcore_outshr GRRegs:$r,
-                                                      GRRegs:$src))]>;
+def OUTSHR_2r :
+  _F2RSrcDst<0b101011, (outs GRRegs:$dst), (ins GRRegs:$src, GRRegs:$r),
+             "outshr res[$r], $src",
+             [(set GRRegs:$dst, (int_xcore_outshr GRRegs:$r, GRRegs:$src))]>;
 
-def INCT_2r : _F2R<(outs GRRegs:$dst), (ins GRRegs:$r),
-                 "inct $dst, res[$r]",
-                 [(set GRRegs:$dst, (int_xcore_inct GRRegs:$r))]>;
+def INCT_2r : _F2R<0b100001, (outs GRRegs:$dst), (ins GRRegs:$r),
+                   "inct $dst, res[$r]",
+                   [(set GRRegs:$dst, (int_xcore_inct GRRegs:$r))]>;
 
-def INT_2r : _F2R<(outs GRRegs:$dst), (ins GRRegs:$r),
-                 "int $dst, res[$r]",
-                 [(set GRRegs:$dst, (int_xcore_int GRRegs:$r))]>;
+def INT_2r : _F2R<0b100011, (outs GRRegs:$dst), (ins GRRegs:$r),
+                  "int $dst, res[$r]",
+                  [(set GRRegs:$dst, (int_xcore_int GRRegs:$r))]>;
 
-def IN_2r : _F2R<(outs GRRegs:$dst), (ins GRRegs:$r),
+def IN_2r : _F2R<0b101100, (outs GRRegs:$dst), (ins GRRegs:$r),
                  "in $dst, res[$r]",
                  [(set GRRegs:$dst, (int_xcore_in GRRegs:$r))]>;
 
 let Constraints = "$src = $dst" in
-def INSHR_2r : _F2R<(outs GRRegs:$dst), (ins GRRegs:$r, GRRegs:$src),
-                 "inshr $dst, res[$r]",
-                 [(set GRRegs:$dst, (int_xcore_inshr GRRegs:$r,
-                                                     GRRegs:$src))]>;
+def INSHR_2r :
+  _F2RSrcDst<0b101101, (outs GRRegs:$dst), (ins GRRegs:$src, GRRegs:$r),
+             "inshr $dst, res[$r]",
+             [(set GRRegs:$dst, (int_xcore_inshr GRRegs:$r, GRRegs:$src))]>;
 
-def CHKCT_2r : _F2R<(outs), (ins GRRegs:$r, GRRegs:$val),
-                 "chkct res[$r], $val",
-                 [(int_xcore_chkct GRRegs:$r, GRRegs:$val)]>;
+def CHKCT_2r : _F2R<0b110010, (outs), (ins GRRegs:$r, GRRegs:$val),
+                    "chkct res[$r], $val",
+                    [(int_xcore_chkct GRRegs:$r, GRRegs:$val)]>;
 
-def CHKCT_rus : _F2R<(outs), (ins GRRegs:$r, i32imm:$val),
-                 "chkct res[$r], $val",
-                 [(int_xcore_chkct GRRegs:$r, immUs:$val)]>;
+def CHKCT_rus : _FRUSBitp<0b110011, (outs), (ins GRRegs:$r, i32imm:$val),
+                          "chkct res[$r], $val",
+                          [(int_xcore_chkct GRRegs:$r, immUs:$val)]>;
 
-def TESTCT_2r : _F2R<(outs GRRegs:$dst), (ins GRRegs:$src),
+def TESTCT_2r : _F2R<0b101111, (outs GRRegs:$dst), (ins GRRegs:$src),
                      "testct $dst, res[$src]",
                      [(set GRRegs:$dst, (int_xcore_testct GRRegs:$src))]>;
 
-def TESTWCT_2r : _F2R<(outs GRRegs:$dst), (ins GRRegs:$src),
+def TESTWCT_2r : _F2R<0b110001, (outs GRRegs:$dst), (ins GRRegs:$src),
                       "testwct $dst, res[$src]",
                       [(set GRRegs:$dst, (int_xcore_testwct GRRegs:$src))]>;
 
-def SETD_2r : _F2R<(outs), (ins GRRegs:$r, GRRegs:$val),
-                 "setd res[$r], $val",
-                 [(int_xcore_setd GRRegs:$r, GRRegs:$val)]>;
+def SETD_2r : _FR2R<0b000101, (outs), (ins GRRegs:$r, GRRegs:$val),
+                    "setd res[$r], $val",
+                    [(int_xcore_setd GRRegs:$r, GRRegs:$val)]>;
 
-def GETST_2r : _F2R<(outs GRRegs:$dst), (ins GRRegs:$r),
+def SETPSC_2r : _FR2R<0b110000, (outs), (ins GRRegs:$src1, GRRegs:$src2),
+                      "setpsc res[$src1], $src2",
+                      [(int_xcore_setpsc GRRegs:$src1, GRRegs:$src2)]>;
+
+def GETST_2r : _F2R<0b000001, (outs GRRegs:$dst), (ins GRRegs:$r),
                     "getst $dst, res[$r]",
                     [(set GRRegs:$dst, (int_xcore_getst GRRegs:$r))]>;
 
-def INITSP_2r : _F2R<(outs), (ins GRRegs:$t, GRRegs:$src),
+def INITSP_2r : _F2R<0b000100, (outs), (ins GRRegs:$src, GRRegs:$t),
                      "init t[$t]:sp, $src",
                      [(int_xcore_initsp GRRegs:$t, GRRegs:$src)]>;
 
-def INITPC_2r : _F2R<(outs), (ins GRRegs:$t, GRRegs:$src),
+def INITPC_2r : _F2R<0b000000, (outs), (ins GRRegs:$src, GRRegs:$t),
                      "init t[$t]:pc, $src",
                      [(int_xcore_initpc GRRegs:$t, GRRegs:$src)]>;
 
-def INITCP_2r : _F2R<(outs), (ins GRRegs:$t, GRRegs:$src),
+def INITCP_2r : _F2R<0b000110, (outs), (ins GRRegs:$src, GRRegs:$t),
                      "init t[$t]:cp, $src",
                      [(int_xcore_initcp GRRegs:$t, GRRegs:$src)]>;
 
-def INITDP_2r : _F2R<(outs), (ins GRRegs:$t, GRRegs:$src),
+def INITDP_2r : _F2R<0b000010, (outs), (ins GRRegs:$src, GRRegs:$t),
                      "init t[$t]:dp, $src",
                      [(int_xcore_initdp GRRegs:$t, GRRegs:$src)]>;
 
+def PEEK_2r : _F2R<0b101110, (outs GRRegs:$dst), (ins GRRegs:$src),
+                    "peek $dst, res[$src]",
+                    [(set GRRegs:$dst, (int_xcore_peek GRRegs:$src))]>;
+
+def ENDIN_2r : _F2R<0b100101, (outs GRRegs:$dst), (ins GRRegs:$src),
+                     "endin $dst, res[$src]",
+                     [(set GRRegs:$dst, (int_xcore_endin GRRegs:$src))]>;
+
+def EEF_2r : _F2R<0b001011, (outs), (ins GRRegs:$a, GRRegs:$b),
+                  "eef $a, res[$b]", []>;
+
+def EET_2r : _F2R<0b001001, (outs), (ins GRRegs:$a, GRRegs:$b),
+                  "eet $a, res[$b]", []>;
+
+def TSETMR_2r : _F2RImm<0b000111, (outs), (ins i32imm:$a, GRRegs:$b),
+                        "tsetmr r$a, $b", []>;
+
 // Two operand long
-// getd, testlcl
-def BITREV_l2r : _FL2R<(outs GRRegs:$dst), (ins GRRegs:$src),
-                 "bitrev $dst, $src",
-                 [(set GRRegs:$dst, (int_xcore_bitrev GRRegs:$src))]>;
+def BITREV_l2r : _FL2R<0b0000011000, (outs GRRegs:$dst), (ins GRRegs:$src),
+                       "bitrev $dst, $src",
+                       [(set GRRegs:$dst, (int_xcore_bitrev GRRegs:$src))]>;
+
+def BYTEREV_l2r : _FL2R<0b0000011001, (outs GRRegs:$dst), (ins GRRegs:$src),
+                        "byterev $dst, $src",
+                        [(set GRRegs:$dst, (bswap GRRegs:$src))]>;
 
-def BYTEREV_l2r : _FL2R<(outs GRRegs:$dst), (ins GRRegs:$src),
-                 "byterev $dst, $src",
-                 [(set GRRegs:$dst, (bswap GRRegs:$src))]>;
+def CLZ_l2r : _FL2R<0b000111000, (outs GRRegs:$dst), (ins GRRegs:$src),
+                    "clz $dst, $src",
+                    [(set GRRegs:$dst, (ctlz GRRegs:$src))]>;
 
-def CLZ_l2r : _FL2R<(outs GRRegs:$dst), (ins GRRegs:$src),
-                 "clz $dst, $src",
-                 [(set GRRegs:$dst, (ctlz GRRegs:$src))]>;
+def GETD_l2r : _FL2R<0b0001111001, (outs GRRegs:$dst), (ins GRRegs:$src),
+                     "getd $dst, res[$src]", []>;
 
-def SETC_l2r : _FL2R<(outs), (ins GRRegs:$r, GRRegs:$val),
-                  "setc res[$r], $val",
-                  [(int_xcore_setc GRRegs:$r, GRRegs:$val)]>;
+def GETN_l2r : _FL2R<0b0011011001, (outs GRRegs:$dst), (ins GRRegs:$src),
+                     "getn $dst, res[$src]", []>;
 
-def SETTW_l2r : _FL2R<(outs), (ins GRRegs:$r, GRRegs:$val),
-                  "settw res[$r], $val",
-                  [(int_xcore_settw GRRegs:$r, GRRegs:$val)]>;
+def SETC_l2r : _FL2R<0b0010111001, (outs), (ins GRRegs:$r, GRRegs:$val),
+                     "setc res[$r], $val",
+                     [(int_xcore_setc GRRegs:$r, GRRegs:$val)]>;
 
-def GETPS_l2r : _FL2R<(outs GRRegs:$dst), (ins GRRegs:$src),
-                 "get $dst, ps[$src]",
-                 [(set GRRegs:$dst, (int_xcore_getps GRRegs:$src))]>;
+def SETTW_l2r : _FLR2R<0b0010011001, (outs), (ins GRRegs:$r, GRRegs:$val),
+                       "settw res[$r], $val",
+                       [(int_xcore_settw GRRegs:$r, GRRegs:$val)]>;
 
-def SETPS_l2r : _FL2R<(outs), (ins GRRegs:$src1, GRRegs:$src2),
-                 "set ps[$src1], $src2",
-                 [(int_xcore_setps GRRegs:$src1, GRRegs:$src2)]>;
+def GETPS_l2r : _FL2R<0b0001011001, (outs GRRegs:$dst), (ins GRRegs:$src),
+                      "get $dst, ps[$src]",
+                      [(set GRRegs:$dst, (int_xcore_getps GRRegs:$src))]>;
 
-def INITLR_l2r : _FL2R<(outs), (ins GRRegs:$t, GRRegs:$src),
+def SETPS_l2r : _FLR2R<0b0001111000, (outs), (ins GRRegs:$src1, GRRegs:$src2),
+                       "set ps[$src1], $src2",
+                       [(int_xcore_setps GRRegs:$src1, GRRegs:$src2)]>;
+
+def INITLR_l2r : _FL2R<0b0001011000, (outs), (ins GRRegs:$src, GRRegs:$t),
                        "init t[$t]:lr, $src",
                        [(int_xcore_initlr GRRegs:$t, GRRegs:$src)]>;
 
-def SETCLK_l2r : _FL2R<(outs), (ins GRRegs:$src1, GRRegs:$src2),
-                       "setclk res[$src1], $src2",
-                       [(int_xcore_setclk GRRegs:$src1, GRRegs:$src2)]>;
-
-def SETRDY_l2r : _FL2R<(outs), (ins GRRegs:$src1, GRRegs:$src2),
-                       "setrdy res[$src1], $src2",
-                       [(int_xcore_setrdy GRRegs:$src1, GRRegs:$src2)]>;
+def SETCLK_l2r : _FLR2R<0b0000111001, (outs), (ins GRRegs:$src1, GRRegs:$src2),
+                        "setclk res[$src1], $src2",
+                        [(int_xcore_setclk GRRegs:$src1, GRRegs:$src2)]>;
 
-def SETPSC_l2r : _FL2R<(outs), (ins GRRegs:$src1, GRRegs:$src2),
-                       "setpsc res[$src1], $src2",
-                       [(int_xcore_setpsc GRRegs:$src1, GRRegs:$src2)]>;
+def SETN_l2r : _FLR2R<0b0011011000, (outs), (ins GRRegs:$src1, GRRegs:$src2),
+                      "setn res[$src1], $src2", []>;
 
-def PEEK_l2r : _FL2R<(outs GRRegs:$dst), (ins GRRegs:$src),
-                      "peek $dst, res[$src]",
-                      [(set GRRegs:$dst, (int_xcore_peek GRRegs:$src))]>;
+def SETRDY_l2r : _FLR2R<0b0010111000, (outs), (ins GRRegs:$src1, GRRegs:$src2),
+                        "setrdy res[$src1], $src2",
+                        [(int_xcore_setrdy GRRegs:$src1, GRRegs:$src2)]>;
 
-def ENDIN_l2r : _FL2R<(outs GRRegs:$dst), (ins GRRegs:$src),
-                       "endin $dst, res[$src]",
-                       [(set GRRegs:$dst, (int_xcore_endin GRRegs:$src))]>;
+def TESTLCL_l2r : _FL2R<0b0010011000, (outs GRRegs:$dst), (ins GRRegs:$src),
+                        "testlcl $dst, res[$src]", []>;
 
 // One operand short
-// TODO edu, eeu, waitet, waitef, tstart, clrtp
-// setdp, setcp, setev, kcall
-// dgetreg
-def MSYNC_1r : _F1R<(outs), (ins GRRegs:$i),
-                    "msync res[$i]",
-                    [(int_xcore_msync GRRegs:$i)]>;
-def MJOIN_1r : _F1R<(outs), (ins GRRegs:$i),
-                    "mjoin res[$i]",
-                    [(int_xcore_mjoin GRRegs:$i)]>;
+def MSYNC_1r : _F1R<0b000111, (outs), (ins GRRegs:$a),
+                    "msync res[$a]",
+                    [(int_xcore_msync GRRegs:$a)]>;
+def MJOIN_1r : _F1R<0b000101, (outs), (ins GRRegs:$a),
+                    "mjoin res[$a]",
+                    [(int_xcore_mjoin GRRegs:$a)]>;
 
 let isBranch=1, isIndirectBranch=1, isTerminator=1, isBarrier = 1 in
-def BAU_1r : _F1R<(outs), (ins GRRegs:$addr),
-                 "bau $addr",
-                 [(brind GRRegs:$addr)]>;
+def BAU_1r : _F1R<0b001001, (outs), (ins GRRegs:$a),
+                 "bau $a",
+                 [(brind GRRegs:$a)]>;
 
 let isBranch=1, isIndirectBranch=1, isTerminator=1, isBarrier = 1 in
 def BR_JT : PseudoInstXCore<(outs), (ins InlineJT:$t, GRRegs:$i),
@@ -968,88 +924,150 @@ def BR_JT32 : PseudoInstXCore<(outs), (ins InlineJT32:$t, GRRegs:$i),
                               "bru $i\n$t",
                               [(XCoreBR_JT32 tjumptable:$t, GRRegs:$i)]>;
 
+let isBranch=1, isIndirectBranch=1, isTerminator=1, isBarrier = 1 in
+def BRU_1r : _F1R<0b001010, (outs), (ins GRRegs:$a), "bru $a", []>;
+
 let Defs=[SP], neverHasSideEffects=1 in
-def SETSP_1r : _F1R<(outs), (ins GRRegs:$src),
-                 "set sp, $src",
-                 []>;
+def SETSP_1r : _F1R<0b001011, (outs), (ins GRRegs:$a), "set sp, $a", []>;
+
+let neverHasSideEffects=1 in
+def SETDP_1r : _F1R<0b001100, (outs), (ins GRRegs:$a), "set dp, $a", []>;
+
+let neverHasSideEffects=1 in
+def SETCP_1r : _F1R<0b001101, (outs), (ins GRRegs:$a), "set cp, $a", []>;
 
 let hasCtrlDep = 1 in 
-def ECALLT_1r : _F1R<(outs), (ins GRRegs:$src),
-                 "ecallt $src",
+def ECALLT_1r : _F1R<0b010011, (outs), (ins GRRegs:$a),
+                 "ecallt $a",
                  []>;
 
 let hasCtrlDep = 1 in 
-def ECALLF_1r : _F1R<(outs), (ins GRRegs:$src),
-                 "ecallf $src",
+def ECALLF_1r : _F1R<0b010010, (outs), (ins GRRegs:$a),
+                 "ecallf $a",
                  []>;
 
 let isCall=1, 
 // All calls clobber the link register and the non-callee-saved registers:
 Defs = [R0, R1, R2, R3, R11, LR], Uses = [SP] in {
-def BLA_1r : _F1R<(outs), (ins GRRegs:$addr),
-                 "bla $addr",
-                 [(XCoreBranchLink GRRegs:$addr)]>;
+def BLA_1r : _F1R<0b001000, (outs), (ins GRRegs:$a),
+                 "bla $a",
+                 [(XCoreBranchLink GRRegs:$a)]>;
 }
 
-def SYNCR_1r : _F1R<(outs), (ins GRRegs:$r),
-                 "syncr res[$r]",
-                 [(int_xcore_syncr GRRegs:$r)]>;
+def SYNCR_1r : _F1R<0b100001, (outs), (ins GRRegs:$a),
+                 "syncr res[$a]",
+                 [(int_xcore_syncr GRRegs:$a)]>;
 
-def FREER_1r : _F1R<(outs), (ins GRRegs:$r),
-               "freer res[$r]",
-               [(int_xcore_freer GRRegs:$r)]>;
+def FREER_1r : _F1R<0b000100, (outs), (ins GRRegs:$a),
+               "freer res[$a]",
+               [(int_xcore_freer GRRegs:$a)]>;
 
 let Uses=[R11] in {
-def SETV_1r : _F1R<(outs), (ins GRRegs:$r),
-                   "setv res[$r], r11",
-                   [(int_xcore_setv GRRegs:$r, R11)]>;
+def SETV_1r : _F1R<0b010001, (outs), (ins GRRegs:$a),
+                   "setv res[$a], r11",
+                   [(int_xcore_setv GRRegs:$a, R11)]>;
 
-def SETEV_1r : _F1R<(outs), (ins GRRegs:$r),
-                    "setev res[$r], r11",
-                    [(int_xcore_setev GRRegs:$r, R11)]>;
+def SETEV_1r : _F1R<0b001111, (outs), (ins GRRegs:$a),
+                    "setev res[$a], r11",
+                    [(int_xcore_setev GRRegs:$a, R11)]>;
 }
 
-def EEU_1r : _F1R<(outs), (ins GRRegs:$r),
-               "eeu res[$r]",
-               [(int_xcore_eeu GRRegs:$r)]>;
+def DGETREG_1r : _F1R<0b001110, (outs GRRegs:$a), (ins), "dgetreg $a", []>;
+
+def EDU_1r : _F1R<0b000000, (outs), (ins GRRegs:$a), "edu res[$a]", []>;
+
+def EEU_1r : _F1R<0b000001, (outs), (ins GRRegs:$a),
+               "eeu res[$a]",
+               [(int_xcore_eeu GRRegs:$a)]>;
+
+def KCALL_1r : _F1R<0b010000, (outs), (ins GRRegs:$a), "kcall $a", []>;
+
+def WAITEF_1R : _F1R<0b000011, (outs), (ins GRRegs:$a), "waitef $a", []>;
+
+def WAITET_1R : _F1R<0b000010, (outs), (ins GRRegs:$a), "waitet $a", []>;
+
+def TSTART_1R : _F1R<0b000110, (outs), (ins GRRegs:$a), "start t[$a]", []>;
+
+def CLRPT_1R : _F1R<0b100000, (outs), (ins GRRegs:$a), "clrpt res[$a]", []>;
 
 // Zero operand short
-// TODO freet, ldspc, stspc, ldssr, stssr, ldsed, stsed,
-// stet, getkep, getksp, setkep, getid, kret, dcall, dret,
-// dentsp, drestsp
 
-def CLRE_0R : _F0R<(outs), (ins), "clre", [(int_xcore_clre)]>;
+def CLRE_0R : _F0R<0b0000001101, (outs), (ins), "clre", [(int_xcore_clre)]>;
+
+def DCALL_0R : _F0R<0b0000011100, (outs), (ins), "dcall", []>;
+
+let Defs = [SP], Uses = [SP] in
+def DENTSP_0R : _F0R<0b0001001100, (outs), (ins), "dentsp", []>;
+
+let Defs = [SP] in
+def DRESTSP_0R : _F0R<0b0001001101, (outs), (ins), "drestsp", []>;
+
+def DRET_0R : _F0R<0b0000011110, (outs), (ins), "dret", []>;
+
+def FREET_0R : _F0R<0b0000001111, (outs), (ins), "freet", []>;
 
 let Defs = [R11] in {
-def GETID_0R : _F0R<(outs), (ins),
+def GETID_0R : _F0R<0b0001001110, (outs), (ins),
                     "get r11, id",
                     [(set R11, (int_xcore_getid))]>;
 
-def GETED_0R : _F0R<(outs), (ins),
+def GETED_0R : _F0R<0b0000111110, (outs), (ins),
                     "get r11, ed",
                     [(set R11, (int_xcore_geted))]>;
 
-def GETET_0R : _F0R<(outs), (ins),
+def GETET_0R : _F0R<0b0000111111, (outs), (ins),
                     "get r11, et",
                     [(set R11, (int_xcore_getet))]>;
+
+def GETKEP_0R : _F0R<0b0001001111, (outs), (ins),
+                     "get r11, kep", []>;
+
+def GETKSP_0R : _F0R<0b0001011100, (outs), (ins),
+                     "get r11, ksp", []>;
+}
+
+let Defs = [SP] in
+def KRET_0R : _F0R<0b0000011101, (outs), (ins), "kret", []>;
+
+let Uses = [SP], mayLoad = 1 in {
+def LDET_0R : _F0R<0b0001011110, (outs), (ins), "ldw et, sp[4]", []>;
+
+def LDSED_0R : _F0R<0b0001011101, (outs), (ins), "ldw sed, sp[3]", []>;
+
+def LDSPC_0R : _F0R<0b0000101100, (outs), (ins), "ldw spc, sp[1]", []>;
+
+def LDSSR_0R : _F0R<0b0000101110, (outs), (ins), "ldw ssr, sp[2]", []>;
 }
 
-def SSYNC_0r : _F0R<(outs), (ins),
+let Uses=[R11] in
+def SETKEP_0R : _F0R<0b0000011111, (outs), (ins), "set kep, r11", []>;
+
+def SSYNC_0r : _F0R<0b0000001110, (outs), (ins),
                     "ssync",
                     [(int_xcore_ssync)]>;
 
+let Uses = [SP], mayStore = 1 in {
+def STET_0R : _F0R<0b0000111101, (outs), (ins), "stw et, sp[4]", []>;
+
+def STSED_0R : _F0R<0b0000111100, (outs), (ins), "stw sed, sp[3]", []>;
+
+def STSPC_0R : _F0R<0b0000101101, (outs), (ins), "stw spc, sp[1]", []>;
+
+def STSSR_0R : _F0R<0b0000101111, (outs), (ins), "stw ssr, sp[2]", []>;
+}
+
 let isBranch=1, isIndirectBranch=1, isTerminator=1, isBarrier = 1,
     hasSideEffects = 1 in
-def WAITEU_0R : _F0R<(outs), (ins),
-                 "waiteu",
-                 [(brind (int_xcore_waitevent))]>;
+def WAITEU_0R : _F0R<0b0000001100, (outs), (ins),
+                     "waiteu",
+                     [(brind (int_xcore_waitevent))]>;
 
 //===----------------------------------------------------------------------===//
 // Non-Instruction Patterns
 //===----------------------------------------------------------------------===//
 
-def : Pat<(XCoreBranchLink tglobaladdr:$addr), (BL_lu10 tglobaladdr:$addr)>;
-def : Pat<(XCoreBranchLink texternalsym:$addr), (BL_lu10 texternalsym:$addr)>;
+def : Pat<(XCoreBranchLink tglobaladdr:$addr), (BLRF_lu10 tglobaladdr:$addr)>;
+def : Pat<(XCoreBranchLink texternalsym:$addr), (BLRF_lu10 texternalsym:$addr)>;
 
 /// sext_inreg
 def : Pat<(sext_inreg GRRegs:$b, i1), (SEXT_rus GRRegs:$b, 1)>;
@@ -1091,7 +1109,7 @@ def : Pat<(truncstorei16 GRRegs:$val, GRRegs:$addr),
           (ST16_l3r GRRegs:$val, GRRegs:$addr, (LDC_ru6 0))>;
 
 def : Pat<(store GRRegs:$val, (ldawf GRRegs:$addr, GRRegs:$offset)),
-          (STW_3r GRRegs:$val, GRRegs:$addr, GRRegs:$offset)>;
+          (STW_l3r GRRegs:$val, GRRegs:$addr, GRRegs:$offset)>;
 def : Pat<(store GRRegs:$val, (add GRRegs:$addr, immUs4:$offset)),
           (STW_2rus GRRegs:$val, GRRegs:$addr, (div4_xform immUs4:$offset))>;
 def : Pat<(store GRRegs:$val, GRRegs:$addr),
diff --git a/lib/Target/XCore/XCoreMCInstLower.cpp b/lib/Target/XCore/XCoreMCInstLower.cpp
new file mode 100644
index 000000000000..f96eda9fcb9f
--- /dev/null
+++ b/lib/Target/XCore/XCoreMCInstLower.cpp
@@ -0,0 +1,117 @@
+//===-- XCoreMCInstLower.cpp - Convert XCore MachineInstr to MCInst -------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// \brief This file contains code to lower XCore MachineInstrs to their
+/// corresponding MCInst records.
+///
+//===----------------------------------------------------------------------===//
+#include "XCoreMCInstLower.h"
+#include "llvm/CodeGen/AsmPrinter.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/CodeGen/MachineOperand.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/Target/Mangler.h"
+
+using namespace llvm;
+
+XCoreMCInstLower::XCoreMCInstLower(class AsmPrinter &asmprinter)
+: Printer(asmprinter) {}
+
+void XCoreMCInstLower::Initialize(Mangler *M, MCContext *C) {
+  Mang = M;
+  Ctx = C;
+}
+
+MCOperand XCoreMCInstLower::LowerSymbolOperand(const MachineOperand &MO,
+                                               MachineOperandType MOTy,
+                                               unsigned Offset) const {
+  MCSymbolRefExpr::VariantKind Kind = MCSymbolRefExpr::VK_None;
+  const MCSymbol *Symbol;
+
+  switch (MOTy) {
+    case MachineOperand::MO_MachineBasicBlock:
+      Symbol = MO.getMBB()->getSymbol();
+      break;
+    case MachineOperand::MO_GlobalAddress:
+      Symbol = Mang->getSymbol(MO.getGlobal());
+      Offset += MO.getOffset();
+      break;
+    case MachineOperand::MO_BlockAddress:
+      Symbol = Printer.GetBlockAddressSymbol(MO.getBlockAddress());
+      Offset += MO.getOffset();
+      break;
+    case MachineOperand::MO_ExternalSymbol:
+      Symbol = Printer.GetExternalSymbolSymbol(MO.getSymbolName());
+      Offset += MO.getOffset();
+      break;
+    case MachineOperand::MO_JumpTableIndex:
+      Symbol = Printer.GetJTISymbol(MO.getIndex());
+      break;
+    case MachineOperand::MO_ConstantPoolIndex:
+      Symbol = Printer.GetCPISymbol(MO.getIndex());
+      Offset += MO.getOffset();
+      break;
+    default:
+      llvm_unreachable("<unknown operand type>");
+  }
+
+  const MCSymbolRefExpr *MCSym = MCSymbolRefExpr::Create(Symbol, Kind, *Ctx);
+
+  if (!Offset)
+    return MCOperand::CreateExpr(MCSym);
+
+  // Assume offset is never negative.
+  assert(Offset > 0);
+
+  const MCConstantExpr *OffsetExpr =  MCConstantExpr::Create(Offset, *Ctx);
+  const MCBinaryExpr *Add = MCBinaryExpr::CreateAdd(MCSym, OffsetExpr, *Ctx);
+  return MCOperand::CreateExpr(Add);
+}
+
+MCOperand XCoreMCInstLower::LowerOperand(const MachineOperand &MO,
+                                         unsigned offset) const {
+  MachineOperandType MOTy = MO.getType();
+
+  switch (MOTy) {
+    default: llvm_unreachable("unknown operand type");
+    case MachineOperand::MO_Register:
+      // Ignore all implicit register operands.
+      if (MO.isImplicit()) break;
+      return MCOperand::CreateReg(MO.getReg());
+    case MachineOperand::MO_Immediate:
+      return MCOperand::CreateImm(MO.getImm() + offset);
+    case MachineOperand::MO_MachineBasicBlock:
+    case MachineOperand::MO_GlobalAddress:
+    case MachineOperand::MO_ExternalSymbol:
+    case MachineOperand::MO_JumpTableIndex:
+    case MachineOperand::MO_ConstantPoolIndex:
+    case MachineOperand::MO_BlockAddress:
+      return LowerSymbolOperand(MO, MOTy, offset);
+    case MachineOperand::MO_RegisterMask:
+      break;
+  }
+
+  return MCOperand();
+}
+
+void XCoreMCInstLower::Lower(const MachineInstr *MI, MCInst &OutMI) const {
+  OutMI.setOpcode(MI->getOpcode());
+
+  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+    const MachineOperand &MO = MI->getOperand(i);
+    MCOperand MCOp = LowerOperand(MO);
+
+    if (MCOp.isValid())
+      OutMI.addOperand(MCOp);
+  }
+}
diff --git a/lib/Target/XCore/XCoreMCInstLower.h b/lib/Target/XCore/XCoreMCInstLower.h
new file mode 100644
index 000000000000..28e702bb9884
--- /dev/null
+++ b/lib/Target/XCore/XCoreMCInstLower.h
@@ -0,0 +1,42 @@
+//===-- XCoreMCInstLower.h - Lower MachineInstr to MCInst ------*- C++ -*--===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef XCOREMCINSTLOWER_H
+#define XCOREMCINSTLOWER_H
+#include "llvm/CodeGen/MachineOperand.h"
+#include "llvm/Support/Compiler.h"
+
+namespace llvm {
+  class MCContext;
+  class MCInst;
+  class MCOperand;
+  class MachineInstr;
+  class MachineFunction;
+  class Mangler;
+  class AsmPrinter;
+
+/// \brief This class is used to lower an MachineInstr into an MCInst.
+class LLVM_LIBRARY_VISIBILITY XCoreMCInstLower {
+  typedef MachineOperand::MachineOperandType MachineOperandType;
+  MCContext *Ctx;
+  Mangler *Mang;
+  AsmPrinter &Printer;
+public:
+  XCoreMCInstLower(class AsmPrinter &asmprinter);
+  void Initialize(Mangler *mang, MCContext *C);
+  void Lower(const MachineInstr *MI, MCInst &OutMI) const;
+  MCOperand LowerOperand(const MachineOperand& MO, unsigned offset = 0) const;
+
+private:
+  MCOperand LowerSymbolOperand(const MachineOperand &MO,
+                               MachineOperandType MOTy, unsigned Offset) const;
+};
+}
+
+#endif
diff --git a/lib/Target/XCore/XCoreMachineFunctionInfo.h b/lib/Target/XCore/XCoreMachineFunctionInfo.h
index f869fcf26de3..69d5de3e03ad 100644
--- a/lib/Target/XCore/XCoreMachineFunctionInfo.h
+++ b/lib/Target/XCore/XCoreMachineFunctionInfo.h
@@ -14,8 +14,8 @@
 #ifndef XCOREMACHINEFUNCTIONINFO_H
 #define XCOREMACHINEFUNCTIONINFO_H
 
-#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
 #include <vector>
 
 namespace llvm {
diff --git a/lib/Target/XCore/XCoreRegisterInfo.cpp b/lib/Target/XCore/XCoreRegisterInfo.cpp
index be5855abcd0b..49b563497c0b 100644
--- a/lib/Target/XCore/XCoreRegisterInfo.cpp
+++ b/lib/Target/XCore/XCoreRegisterInfo.cpp
@@ -12,25 +12,25 @@
 //===----------------------------------------------------------------------===//
 
 #include "XCoreRegisterInfo.h"
-#include "XCoreMachineFunctionInfo.h"
 #include "XCore.h"
-#include "llvm/Type.h"
-#include "llvm/Function.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
-#include "llvm/CodeGen/MachineFunction.h"
+#include "XCoreMachineFunctionInfo.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/Target/TargetFrameLowering.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/ADT/BitVector.h"
-#include "llvm/ADT/STLExtras.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
 
 #define GET_REGINFO_TARGET_DESC
 #include "XCoreGenRegisterInfo.inc"
@@ -101,72 +101,14 @@ XCoreRegisterInfo::useFPForScavengingIndex(const MachineFunction &MF) const {
   return false;
 }
 
-// This function eliminates ADJCALLSTACKDOWN,
-// ADJCALLSTACKUP pseudo instructions
-void XCoreRegisterInfo::
-eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
-                              MachineBasicBlock::iterator I) const {
-  const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
-
-  if (!TFI->hasReservedCallFrame(MF)) {
-    // Turn the adjcallstackdown instruction into 'extsp <amt>' and the
-    // adjcallstackup instruction into 'ldaw sp, sp[<amt>]'
-    MachineInstr *Old = I;
-    uint64_t Amount = Old->getOperand(0).getImm();
-    if (Amount != 0) {
-      // We need to keep the stack aligned properly.  To do this, we round the
-      // amount of space needed for the outgoing arguments up to the next
-      // alignment boundary.
-      unsigned Align = TFI->getStackAlignment();
-      Amount = (Amount+Align-1)/Align*Align;
-
-      assert(Amount%4 == 0);
-      Amount /= 4;
-
-      bool isU6 = isImmU6(Amount);
-      if (!isU6 && !isImmU16(Amount)) {
-        // FIX could emit multiple instructions in this case.
-#ifndef NDEBUG
-        errs() << "eliminateCallFramePseudoInstr size too big: "
-               << Amount << "\n";
-#endif
-        llvm_unreachable(0);
-      }
-
-      MachineInstr *New;
-      if (Old->getOpcode() == XCore::ADJCALLSTACKDOWN) {
-        int Opcode = isU6 ? XCore::EXTSP_u6 : XCore::EXTSP_lu6;
-        New=BuildMI(MF, Old->getDebugLoc(), TII.get(Opcode))
-          .addImm(Amount);
-      } else {
-        assert(Old->getOpcode() == XCore::ADJCALLSTACKUP);
-        int Opcode = isU6 ? XCore::LDAWSP_ru6_RRegs : XCore::LDAWSP_lru6_RRegs;
-        New=BuildMI(MF, Old->getDebugLoc(), TII.get(Opcode), XCore::SP)
-          .addImm(Amount);
-      }
-
-      // Replace the pseudo instruction with a new instruction...
-      MBB.insert(I, New);
-    }
-  }
-  
-  MBB.erase(I);
-}
-
 void
 XCoreRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
-                                       int SPAdj, RegScavenger *RS) const {
+                                       int SPAdj, unsigned FIOperandNum,
+                                       RegScavenger *RS) const {
   assert(SPAdj == 0 && "Unexpected");
   MachineInstr &MI = *II;
   DebugLoc dl = MI.getDebugLoc();
-  unsigned i = 0;
-
-  while (!MI.getOperand(i).isFI()) {
-    ++i;
-    assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
-  }
-
-  MachineOperand &FrameOp = MI.getOperand(i);
+  MachineOperand &FrameOp = MI.getOperand(FIOperandNum);
   int FrameIndex = FrameOp.getIndex();
 
   MachineFunction &MF = *MI.getParent()->getParent();
@@ -190,14 +132,14 @@ XCoreRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
 
   // Special handling of DBG_VALUE instructions.
   if (MI.isDebugValue()) {
-    MI.getOperand(i).ChangeToRegister(FrameReg, false /*isDef*/);
-    MI.getOperand(i+1).ChangeToImmediate(Offset);
+    MI.getOperand(FIOperandNum).ChangeToRegister(FrameReg, false /*isDef*/);
+    MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset);
     return;
   }
 
   // fold constant into offset.
-  Offset += MI.getOperand(i + 1).getImm();
-  MI.getOperand(i + 1).ChangeToImmediate(0);
+  Offset += MI.getOperand(FIOperandNum + 1).getImm();
+  MI.getOperand(FIOperandNum + 1).ChangeToImmediate(0);
   
   assert(Offset%4 == 0 && "Misaligned stack offset");
 
@@ -231,7 +173,7 @@ XCoreRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
               .addReg(ScratchReg, RegState::Kill);
         break;
       case XCore::STWFI:
-        BuildMI(MBB, II, dl, TII.get(XCore::STW_3r))
+        BuildMI(MBB, II, dl, TII.get(XCore::STW_l3r))
               .addReg(Reg, getKillRegState(isKill))
               .addReg(FrameReg)
               .addReg(ScratchReg, RegState::Kill);
diff --git a/lib/Target/XCore/XCoreRegisterInfo.h b/lib/Target/XCore/XCoreRegisterInfo.h
index c4dcb6b533c2..1db32489cf8d 100644
--- a/lib/Target/XCore/XCoreRegisterInfo.h
+++ b/lib/Target/XCore/XCoreRegisterInfo.h
@@ -54,12 +54,9 @@ public:
 
   bool useFPForScavengingIndex(const MachineFunction &MF) const;
 
-  void eliminateCallFramePseudoInstr(MachineFunction &MF,
-                                     MachineBasicBlock &MBB,
-                                     MachineBasicBlock::iterator I) const;
-
   void eliminateFrameIndex(MachineBasicBlock::iterator II,
-                           int SPAdj, RegScavenger *RS = NULL) const;
+                           int SPAdj, unsigned FIOperandNum,
+                           RegScavenger *RS = NULL) const;
 
   // Debug information queries.
   unsigned getFrameRegister(const MachineFunction &MF) const;
diff --git a/lib/Target/XCore/XCoreRegisterInfo.td b/lib/Target/XCore/XCoreRegisterInfo.td
index 9edfda1f5007..6694b2882aca 100644
--- a/lib/Target/XCore/XCoreRegisterInfo.td
+++ b/lib/Target/XCore/XCoreRegisterInfo.td
@@ -45,12 +45,15 @@ def LR : Ri<15, "lr">, DwarfRegNum<[15]>;
 def GRRegs : RegisterClass<"XCore", [i32], 32,
   // Return values and arguments
   (add R0, R1, R2, R3,
-  // Not preserved across procedure calls
-  R11,
   // Callee save
-  R4, R5, R6, R7, R8, R9, R10)>;
+  R4, R5, R6, R7, R8, R9, R10,
+  // Not preserved across procedure calls
+  R11)>;
 
 // Reserved
-def RRegs : RegisterClass<"XCore", [i32], 32, (add CP, DP, SP, LR)> {
+def RRegs : RegisterClass<"XCore", [i32], 32,
+  (add R0, R1, R2, R3,
+   R4, R5, R6, R7, R8, R9, R10,
+   R11, CP, DP, SP, LR)> {
   let isAllocatable = 0;
 }
diff --git a/lib/Target/XCore/XCoreSubtarget.h b/lib/Target/XCore/XCoreSubtarget.h
index 8d0f254e087a..5ac4dbc4bc07 100644
--- a/lib/Target/XCore/XCoreSubtarget.h
+++ b/lib/Target/XCore/XCoreSubtarget.h
@@ -14,8 +14,8 @@
 #ifndef XCORESUBTARGET_H
 #define XCORESUBTARGET_H
 
-#include "llvm/Target/TargetSubtargetInfo.h"
 #include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
 #include <string>
 
 #define GET_SUBTARGETINFO_HEADER
diff --git a/lib/Target/XCore/XCoreTargetMachine.cpp b/lib/Target/XCore/XCoreTargetMachine.cpp
index d5a932c5189d..28c3d12c05fe 100644
--- a/lib/Target/XCore/XCoreTargetMachine.cpp
+++ b/lib/Target/XCore/XCoreTargetMachine.cpp
@@ -12,9 +12,9 @@
 
 #include "XCoreTargetMachine.h"
 #include "XCore.h"
-#include "llvm/Module.h"
-#include "llvm/PassManager.h"
 #include "llvm/CodeGen/Passes.h"
+#include "llvm/IR/Module.h"
+#include "llvm/PassManager.h"
 #include "llvm/Support/TargetRegistry.h"
 using namespace llvm;
 
@@ -32,7 +32,7 @@ XCoreTargetMachine::XCoreTargetMachine(const Target &T, StringRef TT,
     InstrInfo(),
     FrameLowering(Subtarget),
     TLInfo(*this),
-    TSInfo(*this), STTI(&TLInfo), VTTI(&TLInfo) {
+    TSInfo(*this) {
 }
 
 namespace {
diff --git a/lib/Target/XCore/XCoreTargetMachine.h b/lib/Target/XCore/XCoreTargetMachine.h
index c60c6a37f95b..eb9a1aa420eb 100644
--- a/lib/Target/XCore/XCoreTargetMachine.h
+++ b/lib/Target/XCore/XCoreTargetMachine.h
@@ -15,13 +15,12 @@
 #define XCORETARGETMACHINE_H
 
 #include "XCoreFrameLowering.h"
-#include "XCoreSubtarget.h"
-#include "XCoreInstrInfo.h"
 #include "XCoreISelLowering.h"
+#include "XCoreInstrInfo.h"
 #include "XCoreSelectionDAGInfo.h"
+#include "XCoreSubtarget.h"
+#include "llvm/IR/DataLayout.h"
 #include "llvm/Target/TargetMachine.h"
-#include "llvm/Target/TargetTransformImpl.h"
-#include "llvm/DataLayout.h"
 
 namespace llvm {
 
@@ -32,8 +31,6 @@ class XCoreTargetMachine : public LLVMTargetMachine {
   XCoreFrameLowering FrameLowering;
   XCoreTargetLowering TLInfo;
   XCoreSelectionDAGInfo TSInfo;
-  ScalarTargetTransformImpl STTI;
-  VectorTargetTransformImpl VTTI;
 public:
   XCoreTargetMachine(const Target &T, StringRef TT,
                      StringRef CPU, StringRef FS, const TargetOptions &Options,
@@ -56,12 +53,6 @@ public:
   virtual const TargetRegisterInfo *getRegisterInfo() const {
     return &InstrInfo.getRegisterInfo();
   }
-  virtual const ScalarTargetTransformInfo *getScalarTargetTransformInfo()const {
-    return &STTI;
-  }
-  virtual const VectorTargetTransformInfo *getVectorTargetTransformInfo()const {
-    return &VTTI;
-  }
   virtual const DataLayout       *getDataLayout() const { return &DL; }
 
   // Pass Pipeline Configuration
diff --git a/lib/Target/XCore/XCoreTargetObjectFile.cpp b/lib/Target/XCore/XCoreTargetObjectFile.cpp
index 7f4e1c1b4fd7..820389935b38 100644
--- a/lib/Target/XCore/XCoreTargetObjectFile.cpp
+++ b/lib/Target/XCore/XCoreTargetObjectFile.cpp
@@ -11,8 +11,8 @@
 #include "XCoreSubtarget.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCSectionELF.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/Support/ELF.h"
+#include "llvm/Target/TargetMachine.h"
 using namespace llvm;