vendor/llvm/llvm-trunk-r351319

1 files changed, 816 insertions, 246 deletions

diff --git a/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp b/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
index 4b537540046f..be291b127301 100644
--- a/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
+++ b/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
@@ -20,6 +20,26 @@
 // ==>
 //  s_buffer_load_dwordx2 s[4:5], s[0:3], 4
 //
+// This pass also tries to promote constant offset to the immediate by
+// adjusting the base. It tries to use a base from the nearby instructions that
+// allows it to have a 13bit constant offset and then promotes the 13bit offset
+// to the immediate.
+// E.g.
+//  s_movk_i32 s0, 0x1800
+//  v_add_co_u32_e32 v0, vcc, s0, v2
+//  v_addc_co_u32_e32 v1, vcc, 0, v6, vcc
+//
+//  s_movk_i32 s0, 0x1000
+//  v_add_co_u32_e32 v5, vcc, s0, v2
+//  v_addc_co_u32_e32 v6, vcc, 0, v6, vcc
+//  global_load_dwordx2 v[5:6], v[5:6], off
+//  global_load_dwordx2 v[0:1], v[0:1], off
+// =>
+//  s_movk_i32 s0, 0x1000
+//  v_add_co_u32_e32 v5, vcc, s0, v2
+//  v_addc_co_u32_e32 v6, vcc, 0, v6, vcc
+//  global_load_dwordx2 v[5:6], v[5:6], off
+//  global_load_dwordx2 v[0:1], v[5:6], off offset:2048
 //
 // Future improvements:
 //
@@ -43,9 +63,9 @@
 
 #include "AMDGPU.h"
 #include "AMDGPUSubtarget.h"
+#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "SIInstrInfo.h"
 #include "SIRegisterInfo.h"
-#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "Utils/AMDGPUBaseInfo.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/SmallVector.h"
@@ -74,23 +94,38 @@ using namespace llvm;
 #define DEBUG_TYPE "si-load-store-opt"
 
 namespace {
+enum InstClassEnum {
+  UNKNOWN,
+  DS_READ,
+  DS_WRITE,
+  S_BUFFER_LOAD_IMM,
+  BUFFER_LOAD_OFFEN = AMDGPU::BUFFER_LOAD_DWORD_OFFEN,
+  BUFFER_LOAD_OFFSET = AMDGPU::BUFFER_LOAD_DWORD_OFFSET,
+  BUFFER_STORE_OFFEN = AMDGPU::BUFFER_STORE_DWORD_OFFEN,
+  BUFFER_STORE_OFFSET = AMDGPU::BUFFER_STORE_DWORD_OFFSET,
+  BUFFER_LOAD_OFFEN_exact = AMDGPU::BUFFER_LOAD_DWORD_OFFEN_exact,
+  BUFFER_LOAD_OFFSET_exact = AMDGPU::BUFFER_LOAD_DWORD_OFFSET_exact,
+  BUFFER_STORE_OFFEN_exact = AMDGPU::BUFFER_STORE_DWORD_OFFEN_exact,
+  BUFFER_STORE_OFFSET_exact = AMDGPU::BUFFER_STORE_DWORD_OFFSET_exact,
+};
 
-class SILoadStoreOptimizer : public MachineFunctionPass {
-  enum InstClassEnum {
-    DS_READ_WRITE,
-    S_BUFFER_LOAD_IMM,
-    BUFFER_LOAD_OFFEN,
-    BUFFER_LOAD_OFFSET,
-    BUFFER_STORE_OFFEN,
-    BUFFER_STORE_OFFSET,
-  };
+enum RegisterEnum {
+  SBASE = 0x1,
+  SRSRC = 0x2,
+  SOFFSET = 0x4,
+  VADDR = 0x8,
+  ADDR = 0x10,
+};
 
+class SILoadStoreOptimizer : public MachineFunctionPass {
   struct CombineInfo {
     MachineBasicBlock::iterator I;
     MachineBasicBlock::iterator Paired;
     unsigned EltSize;
     unsigned Offset0;
     unsigned Offset1;
+    unsigned Width0;
+    unsigned Width1;
     unsigned BaseOff;
     InstClassEnum InstClass;
     bool GLC0;
@@ -98,9 +133,23 @@ class SILoadStoreOptimizer : public MachineFunctionPass {
     bool SLC0;
     bool SLC1;
     bool UseST64;
-    bool IsX2;
-    SmallVector<MachineInstr*, 8> InstsToMove;
-   };
+    SmallVector<MachineInstr *, 8> InstsToMove;
+  };
+
+  struct BaseRegisters {
+    unsigned LoReg = 0;
+    unsigned HiReg = 0;
+
+    unsigned LoSubReg = 0;
+    unsigned HiSubReg = 0;
+  };
+
+  struct MemAddress {
+    BaseRegisters Base;
+    int64_t Offset = 0;
+  };
+
+  using MemInfoMap = DenseMap<MachineInstr *, MemAddress>;
 
 private:
   const GCNSubtarget *STM = nullptr;
@@ -108,9 +157,16 @@ private:
   const SIRegisterInfo *TRI = nullptr;
   MachineRegisterInfo *MRI = nullptr;
   AliasAnalysis *AA = nullptr;
-  unsigned CreatedX2;
+  bool OptimizeAgain;
 
   static bool offsetsCanBeCombined(CombineInfo &CI);
+  static bool widthsFit(const GCNSubtarget &STM, const CombineInfo &CI);
+  static unsigned getNewOpcode(const CombineInfo &CI);
+  static std::pair<unsigned, unsigned> getSubRegIdxs(const CombineInfo &CI);
+  const TargetRegisterClass *getTargetRegisterClass(const CombineInfo &CI);
+  unsigned getOpcodeWidth(const MachineInstr &MI);
+  InstClassEnum getInstClass(unsigned Opc);
+  unsigned getRegs(unsigned Opc);
 
   bool findMatchingInst(CombineInfo &CI);
 
@@ -123,10 +179,21 @@ private:
   MachineBasicBlock::iterator mergeWrite2Pair(CombineInfo &CI);
   MachineBasicBlock::iterator mergeSBufferLoadImmPair(CombineInfo &CI);
   MachineBasicBlock::iterator mergeBufferLoadPair(CombineInfo &CI);
-  unsigned promoteBufferStoreOpcode(const MachineInstr &I, bool &IsX2,
-                                    bool &IsOffen) const;
   MachineBasicBlock::iterator mergeBufferStorePair(CombineInfo &CI);
 
+  void updateBaseAndOffset(MachineInstr &I, unsigned NewBase,
+                           int32_t NewOffset);
+  unsigned computeBase(MachineInstr &MI, const MemAddress &Addr);
+  MachineOperand createRegOrImm(int32_t Val, MachineInstr &MI);
+  Optional<int32_t> extractConstOffset(const MachineOperand &Op);
+  void processBaseWithConstOffset(const MachineOperand &Base, MemAddress &Addr);
+  /// Promotes constant offset to the immediate by adjusting the base. It
+  /// tries to use a base from the nearby instructions that allows it to have
+  /// a 13bit constant offset which gets promoted to the immediate.
+  bool promoteConstantOffsetToImm(MachineInstr &CI,
+                                  MemInfoMap &Visited,
+                                  SmallPtrSet<MachineInstr *, 4> &Promoted);
+
 public:
   static char ID;
 
@@ -153,8 +220,8 @@ public:
 INITIALIZE_PASS_BEGIN(SILoadStoreOptimizer, DEBUG_TYPE,
                       "SI Load Store Optimizer", false, false)
 INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)
-INITIALIZE_PASS_END(SILoadStoreOptimizer, DEBUG_TYPE,
-                    "SI Load Store Optimizer", false, false)
+INITIALIZE_PASS_END(SILoadStoreOptimizer, DEBUG_TYPE, "SI Load Store Optimizer",
+                    false, false)
 
 char SILoadStoreOptimizer::ID = 0;
 
@@ -165,7 +232,7 @@ FunctionPass *llvm::createSILoadStoreOptimizerPass() {
 }
 
 static void moveInstsAfter(MachineBasicBlock::iterator I,
-                           ArrayRef<MachineInstr*> InstsToMove) {
+                           ArrayRef<MachineInstr *> InstsToMove) {
   MachineBasicBlock *MBB = I->getParent();
   ++I;
   for (MachineInstr *MI : InstsToMove) {
@@ -191,21 +258,19 @@ static void addDefsUsesToList(const MachineInstr &MI,
 static bool memAccessesCanBeReordered(MachineBasicBlock::iterator A,
                                       MachineBasicBlock::iterator B,
                                       const SIInstrInfo *TII,
-                                      AliasAnalysis * AA) {
+                                      AliasAnalysis *AA) {
   // RAW or WAR - cannot reorder
   // WAW - cannot reorder
   // RAR - safe to reorder
   return !(A->mayStore() || B->mayStore()) ||
-    TII->areMemAccessesTriviallyDisjoint(*A, *B, AA);
+         TII->areMemAccessesTriviallyDisjoint(*A, *B, AA);
 }
 
 // Add MI and its defs to the lists if MI reads one of the defs that are
 // already in the list. Returns true in that case.
-static bool
-addToListsIfDependent(MachineInstr &MI,
-                      DenseSet<unsigned> &RegDefs,
-                      DenseSet<unsigned> &PhysRegUses,
-                      SmallVectorImpl<MachineInstr*> &Insts) {
+static bool addToListsIfDependent(MachineInstr &MI, DenseSet<unsigned> &RegDefs,
+                                  DenseSet<unsigned> &PhysRegUses,
+                                  SmallVectorImpl<MachineInstr *> &Insts) {
   for (MachineOperand &Use : MI.operands()) {
     // If one of the defs is read, then there is a use of Def between I and the
     // instruction that I will potentially be merged with. We will need to move
@@ -228,18 +293,16 @@ addToListsIfDependent(MachineInstr &MI,
   return false;
 }
 
-static bool
-canMoveInstsAcrossMemOp(MachineInstr &MemOp,
-                        ArrayRef<MachineInstr*> InstsToMove,
-                        const SIInstrInfo *TII,
-                        AliasAnalysis *AA) {
+static bool canMoveInstsAcrossMemOp(MachineInstr &MemOp,
+                                    ArrayRef<MachineInstr *> InstsToMove,
+                                    const SIInstrInfo *TII, AliasAnalysis *AA) {
   assert(MemOp.mayLoadOrStore());
 
   for (MachineInstr *InstToMove : InstsToMove) {
     if (!InstToMove->mayLoadOrStore())
       continue;
     if (!memAccessesCanBeReordered(MemOp, *InstToMove, TII, AA))
-        return false;
+      return false;
   }
   return true;
 }
@@ -260,10 +323,9 @@ bool SILoadStoreOptimizer::offsetsCanBeCombined(CombineInfo &CI) {
   CI.BaseOff = 0;
 
   // Handle SMEM and VMEM instructions.
-  if (CI.InstClass != DS_READ_WRITE) {
-    unsigned Diff = CI.IsX2 ? 2 : 1;
-    return (EltOffset0 + Diff == EltOffset1 ||
-            EltOffset1 + Diff == EltOffset0) &&
+  if ((CI.InstClass != DS_READ) && (CI.InstClass != DS_WRITE)) {
+    return (EltOffset0 + CI.Width0 == EltOffset1 ||
+            EltOffset1 + CI.Width1 == EltOffset0) &&
            CI.GLC0 == CI.GLC1 &&
            (CI.InstClass == S_BUFFER_LOAD_IMM || CI.SLC0 == CI.SLC1);
   }
@@ -305,42 +367,176 @@ bool SILoadStoreOptimizer::offsetsCanBeCombined(CombineInfo &CI) {
   return false;
 }
 
+bool SILoadStoreOptimizer::widthsFit(const GCNSubtarget &STM,
+                                     const CombineInfo &CI) {
+  const unsigned Width = (CI.Width0 + CI.Width1);
+  switch (CI.InstClass) {
+  default:
+    return (Width <= 4) && (STM.hasDwordx3LoadStores() || (Width != 3));
+  case S_BUFFER_LOAD_IMM:
+    switch (Width) {
+    default:
+      return false;
+    case 2:
+    case 4:
+      return true;
+    }
+  }
+}
+
+unsigned SILoadStoreOptimizer::getOpcodeWidth(const MachineInstr &MI) {
+  const unsigned Opc = MI.getOpcode();
+
+  if (TII->isMUBUF(MI)) {
+    return AMDGPU::getMUBUFDwords(Opc);
+  }
+
+  switch (Opc) {
+  default:
+    return 0;
+  case AMDGPU::S_BUFFER_LOAD_DWORD_IMM:
+    return 1;
+  case AMDGPU::S_BUFFER_LOAD_DWORDX2_IMM:
+    return 2;
+  case AMDGPU::S_BUFFER_LOAD_DWORDX4_IMM:
+    return 4;
+  }
+}
+
+InstClassEnum SILoadStoreOptimizer::getInstClass(unsigned Opc) {
+  if (TII->isMUBUF(Opc)) {
+    const int baseOpcode = AMDGPU::getMUBUFBaseOpcode(Opc);
+
+    // If we couldn't identify the opcode, bail out.
+    if (baseOpcode == -1) {
+      return UNKNOWN;
+    }
+
+    switch (baseOpcode) {
+    default:
+      return UNKNOWN;
+    case AMDGPU::BUFFER_LOAD_DWORD_OFFEN:
+      return BUFFER_LOAD_OFFEN;
+    case AMDGPU::BUFFER_LOAD_DWORD_OFFSET:
+      return BUFFER_LOAD_OFFSET;
+    case AMDGPU::BUFFER_STORE_DWORD_OFFEN:
+      return BUFFER_STORE_OFFEN;
+    case AMDGPU::BUFFER_STORE_DWORD_OFFSET:
+      return BUFFER_STORE_OFFSET;
+    case AMDGPU::BUFFER_LOAD_DWORD_OFFEN_exact:
+      return BUFFER_LOAD_OFFEN_exact;
+    case AMDGPU::BUFFER_LOAD_DWORD_OFFSET_exact:
+      return BUFFER_LOAD_OFFSET_exact;
+    case AMDGPU::BUFFER_STORE_DWORD_OFFEN_exact:
+      return BUFFER_STORE_OFFEN_exact;
+    case AMDGPU::BUFFER_STORE_DWORD_OFFSET_exact:
+      return BUFFER_STORE_OFFSET_exact;
+    }
+  }
+
+  switch (Opc) {
+  default:
+    return UNKNOWN;
+  case AMDGPU::S_BUFFER_LOAD_DWORD_IMM:
+  case AMDGPU::S_BUFFER_LOAD_DWORDX2_IMM:
+  case AMDGPU::S_BUFFER_LOAD_DWORDX4_IMM:
+    return S_BUFFER_LOAD_IMM;
+  case AMDGPU::DS_READ_B32:
+  case AMDGPU::DS_READ_B64:
+  case AMDGPU::DS_READ_B32_gfx9:
+  case AMDGPU::DS_READ_B64_gfx9:
+    return DS_READ;
+  case AMDGPU::DS_WRITE_B32:
+  case AMDGPU::DS_WRITE_B64:
+  case AMDGPU::DS_WRITE_B32_gfx9:
+  case AMDGPU::DS_WRITE_B64_gfx9:
+    return DS_WRITE;
+  }
+}
+
+unsigned SILoadStoreOptimizer::getRegs(unsigned Opc) {
+  if (TII->isMUBUF(Opc)) {
+    unsigned result = 0;
+
+    if (AMDGPU::getMUBUFHasVAddr(Opc)) {
+      result |= VADDR;
+    }
+
+    if (AMDGPU::getMUBUFHasSrsrc(Opc)) {
+      result |= SRSRC;
+    }
+
+    if (AMDGPU::getMUBUFHasSoffset(Opc)) {
+      result |= SOFFSET;
+    }
+
+    return result;
+  }
+
+  switch (Opc) {
+  default:
+    return 0;
+  case AMDGPU::S_BUFFER_LOAD_DWORD_IMM:
+  case AMDGPU::S_BUFFER_LOAD_DWORDX2_IMM:
+  case AMDGPU::S_BUFFER_LOAD_DWORDX4_IMM:
+    return SBASE;
+  case AMDGPU::DS_READ_B32:
+  case AMDGPU::DS_READ_B64:
+  case AMDGPU::DS_READ_B32_gfx9:
+  case AMDGPU::DS_READ_B64_gfx9:
+  case AMDGPU::DS_WRITE_B32:
+  case AMDGPU::DS_WRITE_B64:
+  case AMDGPU::DS_WRITE_B32_gfx9:
+  case AMDGPU::DS_WRITE_B64_gfx9:
+    return ADDR;
+  }
+}
+
 bool SILoadStoreOptimizer::findMatchingInst(CombineInfo &CI) {
   MachineBasicBlock *MBB = CI.I->getParent();
   MachineBasicBlock::iterator E = MBB->end();
   MachineBasicBlock::iterator MBBI = CI.I;
 
-  unsigned AddrOpName[3] = {0};
-  int AddrIdx[3];
-  const MachineOperand *AddrReg[3];
+  const unsigned Opc = CI.I->getOpcode();
+  const InstClassEnum InstClass = getInstClass(Opc);
+
+  if (InstClass == UNKNOWN) {
+    return false;
+  }
+
+  const unsigned Regs = getRegs(Opc);
+
+  unsigned AddrOpName[5] = {0};
+  int AddrIdx[5];
+  const MachineOperand *AddrReg[5];
   unsigned NumAddresses = 0;
 
-  switch (CI.InstClass) {
-  case DS_READ_WRITE:
+  if (Regs & ADDR) {
     AddrOpName[NumAddresses++] = AMDGPU::OpName::addr;
-    break;
-  case S_BUFFER_LOAD_IMM:
+  }
+
+  if (Regs & SBASE) {
     AddrOpName[NumAddresses++] = AMDGPU::OpName::sbase;
-    break;
-  case BUFFER_LOAD_OFFEN:
-  case BUFFER_STORE_OFFEN:
-    AddrOpName[NumAddresses++] = AMDGPU::OpName::srsrc;
-    AddrOpName[NumAddresses++] = AMDGPU::OpName::vaddr;
-    AddrOpName[NumAddresses++] = AMDGPU::OpName::soffset;
-    break;
-  case BUFFER_LOAD_OFFSET:
-  case BUFFER_STORE_OFFSET:
+  }
+
+  if (Regs & SRSRC) {
     AddrOpName[NumAddresses++] = AMDGPU::OpName::srsrc;
+  }
+
+  if (Regs & SOFFSET) {
     AddrOpName[NumAddresses++] = AMDGPU::OpName::soffset;
-    break;
+  }
+
+  if (Regs & VADDR) {
+    AddrOpName[NumAddresses++] = AMDGPU::OpName::vaddr;
   }
 
   for (unsigned i = 0; i < NumAddresses; i++) {
     AddrIdx[i] = AMDGPU::getNamedOperandIdx(CI.I->getOpcode(), AddrOpName[i]);
     AddrReg[i] = &CI.I->getOperand(AddrIdx[i]);
 
-    // We only ever merge operations with the same base address register, so don't
-    // bother scanning forward if there are no other uses.
+    // We only ever merge operations with the same base address register, so
+    // don't bother scanning forward if there are no other uses.
     if (AddrReg[i]->isReg() &&
         (TargetRegisterInfo::isPhysicalRegister(AddrReg[i]->getReg()) ||
          MRI->hasOneNonDBGUse(AddrReg[i]->getReg())))
@@ -353,8 +549,11 @@ bool SILoadStoreOptimizer::findMatchingInst(CombineInfo &CI) {
   DenseSet<unsigned> PhysRegUsesToMove;
   addDefsUsesToList(*CI.I, RegDefsToMove, PhysRegUsesToMove);
 
-  for ( ; MBBI != E; ++MBBI) {
-    if (MBBI->getOpcode() != CI.I->getOpcode()) {
+  for (; MBBI != E; ++MBBI) {
+    const bool IsDS = (InstClass == DS_READ) || (InstClass == DS_WRITE);
+
+    if ((getInstClass(MBBI->getOpcode()) != InstClass) ||
+        (IsDS && (MBBI->getOpcode() != Opc))) {
       // This is not a matching DS instruction, but we can keep looking as
       // long as one of these conditions are met:
       // 1. It is safe to move I down past MBBI.
@@ -368,8 +567,8 @@ bool SILoadStoreOptimizer::findMatchingInst(CombineInfo &CI) {
       }
 
       if (MBBI->mayLoadOrStore() &&
-        (!memAccessesCanBeReordered(*CI.I, *MBBI, TII, AA) ||
-         !canMoveInstsAcrossMemOp(*MBBI, CI.InstsToMove, TII, AA))) {
+          (!memAccessesCanBeReordered(*CI.I, *MBBI, TII, AA) ||
+           !canMoveInstsAcrossMemOp(*MBBI, CI.InstsToMove, TII, AA))) {
         // We fail condition #1, but we may still be able to satisfy condition
         // #2.  Add this instruction to the move list and then we will check
         // if condition #2 holds once we have selected the matching instruction.
@@ -413,8 +612,8 @@ bool SILoadStoreOptimizer::findMatchingInst(CombineInfo &CI) {
         continue;
       }
 
-      // Check same base pointer. Be careful of subregisters, which can occur with
-      // vectors of pointers.
+      // Check same base pointer. Be careful of subregisters, which can occur
+      // with vectors of pointers.
       if (AddrReg[i]->getReg() != AddrRegNext.getReg() ||
           AddrReg[i]->getSubReg() != AddrRegNext.getSubReg()) {
         Match = false;
@@ -423,13 +622,15 @@ bool SILoadStoreOptimizer::findMatchingInst(CombineInfo &CI) {
     }
 
     if (Match) {
-      int OffsetIdx = AMDGPU::getNamedOperandIdx(CI.I->getOpcode(),
-                                                 AMDGPU::OpName::offset);
+      int OffsetIdx =
+          AMDGPU::getNamedOperandIdx(CI.I->getOpcode(), AMDGPU::OpName::offset);
       CI.Offset0 = CI.I->getOperand(OffsetIdx).getImm();
+      CI.Width0 = getOpcodeWidth(*CI.I);
       CI.Offset1 = MBBI->getOperand(OffsetIdx).getImm();
+      CI.Width1 = getOpcodeWidth(*MBBI);
       CI.Paired = MBBI;
 
-      if (CI.InstClass == DS_READ_WRITE) {
+      if ((CI.InstClass == DS_READ) || (CI.InstClass == DS_WRITE)) {
         CI.Offset0 &= 0xffff;
         CI.Offset1 &= 0xffff;
       } else {
@@ -445,7 +646,7 @@ bool SILoadStoreOptimizer::findMatchingInst(CombineInfo &CI) {
       // We also need to go through the list of instructions that we plan to
       // move and make sure they are all safe to move down past the merged
       // instruction.
-      if (offsetsCanBeCombined(CI))
+      if (widthsFit(*STM, CI) && offsetsCanBeCombined(CI))
         if (canMoveInstsAcrossMemOp(*MBBI, CI.InstsToMove, TII, AA))
           return true;
     }
@@ -472,12 +673,12 @@ unsigned SILoadStoreOptimizer::read2ST64Opcode(unsigned EltSize) const {
   if (STM->ldsRequiresM0Init())
     return (EltSize == 4) ? AMDGPU::DS_READ2ST64_B32 : AMDGPU::DS_READ2ST64_B64;
 
-  return (EltSize == 4) ?
-    AMDGPU::DS_READ2ST64_B32_gfx9 : AMDGPU::DS_READ2ST64_B64_gfx9;
+  return (EltSize == 4) ? AMDGPU::DS_READ2ST64_B32_gfx9
+                        : AMDGPU::DS_READ2ST64_B64_gfx9;
 }
 
-MachineBasicBlock::iterator  SILoadStoreOptimizer::mergeRead2Pair(
-  CombineInfo &CI) {
+MachineBasicBlock::iterator
+SILoadStoreOptimizer::mergeRead2Pair(CombineInfo &CI) {
   MachineBasicBlock *MBB = CI.I->getParent();
 
   // Be careful, since the addresses could be subregisters themselves in weird
@@ -489,8 +690,8 @@ MachineBasicBlock::iterator  SILoadStoreOptimizer::mergeRead2Pair(
 
   unsigned NewOffset0 = CI.Offset0;
   unsigned NewOffset1 = CI.Offset1;
-  unsigned Opc = CI.UseST64 ?
-    read2ST64Opcode(CI.EltSize) : read2Opcode(CI.EltSize);
+  unsigned Opc =
+      CI.UseST64 ? read2ST64Opcode(CI.EltSize) : read2Opcode(CI.EltSize);
 
   unsigned SubRegIdx0 = (CI.EltSize == 4) ? AMDGPU::sub0 : AMDGPU::sub0_sub1;
   unsigned SubRegIdx1 = (CI.EltSize == 4) ? AMDGPU::sub1 : AMDGPU::sub2_sub3;
@@ -502,39 +703,40 @@ MachineBasicBlock::iterator  SILoadStoreOptimizer::mergeRead2Pair(
   }
 
   assert((isUInt<8>(NewOffset0) && isUInt<8>(NewOffset1)) &&
-         (NewOffset0 != NewOffset1) &&
-         "Computed offset doesn't fit");
+         (NewOffset0 != NewOffset1) && "Computed offset doesn't fit");
 
   const MCInstrDesc &Read2Desc = TII->get(Opc);
 
-  const TargetRegisterClass *SuperRC
-    = (CI.EltSize == 4) ? &AMDGPU::VReg_64RegClass : &AMDGPU::VReg_128RegClass;
+  const TargetRegisterClass *SuperRC =
+      (CI.EltSize == 4) ? &AMDGPU::VReg_64RegClass : &AMDGPU::VReg_128RegClass;
   unsigned DestReg = MRI->createVirtualRegister(SuperRC);
 
   DebugLoc DL = CI.I->getDebugLoc();
 
   unsigned BaseReg = AddrReg->getReg();
+  unsigned BaseSubReg = AddrReg->getSubReg();
   unsigned BaseRegFlags = 0;
   if (CI.BaseOff) {
     unsigned ImmReg = MRI->createVirtualRegister(&AMDGPU::SGPR_32RegClass);
     BuildMI(*MBB, CI.Paired, DL, TII->get(AMDGPU::S_MOV_B32), ImmReg)
-      .addImm(CI.BaseOff);
+        .addImm(CI.BaseOff);
 
     BaseReg = MRI->createVirtualRegister(&AMDGPU::VGPR_32RegClass);
     BaseRegFlags = RegState::Kill;
 
     TII->getAddNoCarry(*MBB, CI.Paired, DL, BaseReg)
-      .addReg(ImmReg)
-      .addReg(AddrReg->getReg());
+        .addReg(ImmReg)
+        .addReg(AddrReg->getReg(), 0, BaseSubReg);
+    BaseSubReg = 0;
   }
 
   MachineInstrBuilder Read2 =
-    BuildMI(*MBB, CI.Paired, DL, Read2Desc, DestReg)
-      .addReg(BaseReg, BaseRegFlags) // addr
-      .addImm(NewOffset0)            // offset0
-      .addImm(NewOffset1)            // offset1
-      .addImm(0)                     // gds
-      .setMemRefs(CI.I->mergeMemRefsWith(*CI.Paired));
+      BuildMI(*MBB, CI.Paired, DL, Read2Desc, DestReg)
+          .addReg(BaseReg, BaseRegFlags, BaseSubReg) // addr
+          .addImm(NewOffset0)                        // offset0
+          .addImm(NewOffset1)                        // offset1
+          .addImm(0)                                 // gds
+          .cloneMergedMemRefs({&*CI.I, &*CI.Paired});
 
   (void)Read2;
 
@@ -561,32 +763,36 @@ MachineBasicBlock::iterator  SILoadStoreOptimizer::mergeRead2Pair(
 unsigned SILoadStoreOptimizer::write2Opcode(unsigned EltSize) const {
   if (STM->ldsRequiresM0Init())
     return (EltSize == 4) ? AMDGPU::DS_WRITE2_B32 : AMDGPU::DS_WRITE2_B64;
-  return (EltSize == 4) ? AMDGPU::DS_WRITE2_B32_gfx9 : AMDGPU::DS_WRITE2_B64_gfx9;
+  return (EltSize == 4) ? AMDGPU::DS_WRITE2_B32_gfx9
+                        : AMDGPU::DS_WRITE2_B64_gfx9;
 }
 
 unsigned SILoadStoreOptimizer::write2ST64Opcode(unsigned EltSize) const {
   if (STM->ldsRequiresM0Init())
-    return (EltSize == 4) ? AMDGPU::DS_WRITE2ST64_B32 : AMDGPU::DS_WRITE2ST64_B64;
+    return (EltSize == 4) ? AMDGPU::DS_WRITE2ST64_B32
+                          : AMDGPU::DS_WRITE2ST64_B64;
 
-  return (EltSize == 4) ?
-    AMDGPU::DS_WRITE2ST64_B32_gfx9 : AMDGPU::DS_WRITE2ST64_B64_gfx9;
+  return (EltSize == 4) ? AMDGPU::DS_WRITE2ST64_B32_gfx9
+                        : AMDGPU::DS_WRITE2ST64_B64_gfx9;
 }
 
-MachineBasicBlock::iterator SILoadStoreOptimizer::mergeWrite2Pair(
-  CombineInfo &CI) {
+MachineBasicBlock::iterator
+SILoadStoreOptimizer::mergeWrite2Pair(CombineInfo &CI) {
   MachineBasicBlock *MBB = CI.I->getParent();
 
   // Be sure to use .addOperand(), and not .addReg() with these. We want to be
   // sure we preserve the subregister index and any register flags set on them.
-  const MachineOperand *AddrReg = TII->getNamedOperand(*CI.I, AMDGPU::OpName::addr);
-  const MachineOperand *Data0 = TII->getNamedOperand(*CI.I, AMDGPU::OpName::data0);
-  const MachineOperand *Data1
-    = TII->getNamedOperand(*CI.Paired, AMDGPU::OpName::data0);
+  const MachineOperand *AddrReg =
+      TII->getNamedOperand(*CI.I, AMDGPU::OpName::addr);
+  const MachineOperand *Data0 =
+      TII->getNamedOperand(*CI.I, AMDGPU::OpName::data0);
+  const MachineOperand *Data1 =
+      TII->getNamedOperand(*CI.Paired, AMDGPU::OpName::data0);
 
   unsigned NewOffset0 = CI.Offset0;
   unsigned NewOffset1 = CI.Offset1;
-  unsigned Opc = CI.UseST64 ?
-    write2ST64Opcode(CI.EltSize) : write2Opcode(CI.EltSize);
+  unsigned Opc =
+      CI.UseST64 ? write2ST64Opcode(CI.EltSize) : write2Opcode(CI.EltSize);
 
   if (NewOffset0 > NewOffset1) {
     // Canonicalize the merged instruction so the smaller offset comes first.
@@ -595,36 +801,37 @@ MachineBasicBlock::iterator SILoadStoreOptimizer::mergeWrite2Pair(
   }
 
   assert((isUInt<8>(NewOffset0) && isUInt<8>(NewOffset1)) &&
-         (NewOffset0 != NewOffset1) &&
-         "Computed offset doesn't fit");
+         (NewOffset0 != NewOffset1) && "Computed offset doesn't fit");
 
   const MCInstrDesc &Write2Desc = TII->get(Opc);
   DebugLoc DL = CI.I->getDebugLoc();
 
   unsigned BaseReg = AddrReg->getReg();
+  unsigned BaseSubReg = AddrReg->getSubReg();
   unsigned BaseRegFlags = 0;
   if (CI.BaseOff) {
     unsigned ImmReg = MRI->createVirtualRegister(&AMDGPU::SGPR_32RegClass);
     BuildMI(*MBB, CI.Paired, DL, TII->get(AMDGPU::S_MOV_B32), ImmReg)
-      .addImm(CI.BaseOff);
+        .addImm(CI.BaseOff);
 
     BaseReg = MRI->createVirtualRegister(&AMDGPU::VGPR_32RegClass);
     BaseRegFlags = RegState::Kill;
 
     TII->getAddNoCarry(*MBB, CI.Paired, DL, BaseReg)
-      .addReg(ImmReg)
-      .addReg(AddrReg->getReg());
+        .addReg(ImmReg)
+        .addReg(AddrReg->getReg(), 0, BaseSubReg);
+    BaseSubReg = 0;
   }
 
   MachineInstrBuilder Write2 =
-    BuildMI(*MBB, CI.Paired, DL, Write2Desc)
-      .addReg(BaseReg, BaseRegFlags) // addr
-      .add(*Data0)                   // data0
-      .add(*Data1)                   // data1
-      .addImm(NewOffset0)            // offset0
-      .addImm(NewOffset1)            // offset1
-      .addImm(0)                     // gds
-      .setMemRefs(CI.I->mergeMemRefsWith(*CI.Paired));
+      BuildMI(*MBB, CI.Paired, DL, Write2Desc)
+          .addReg(BaseReg, BaseRegFlags, BaseSubReg) // addr
+          .add(*Data0)                               // data0
+          .add(*Data1)                               // data1
+          .addImm(NewOffset0)                        // offset0
+          .addImm(NewOffset1)                        // offset1
+          .addImm(0)                                 // gds
+          .cloneMergedMemRefs({&*CI.I, &*CI.Paired});
 
   moveInstsAfter(Write2, CI.InstsToMove);
 
@@ -636,15 +843,14 @@ MachineBasicBlock::iterator SILoadStoreOptimizer::mergeWrite2Pair(
   return Next;
 }
 
-MachineBasicBlock::iterator SILoadStoreOptimizer::mergeSBufferLoadImmPair(
-  CombineInfo &CI) {
+MachineBasicBlock::iterator
+SILoadStoreOptimizer::mergeSBufferLoadImmPair(CombineInfo &CI) {
   MachineBasicBlock *MBB = CI.I->getParent();
   DebugLoc DL = CI.I->getDebugLoc();
-  unsigned Opcode = CI.IsX2 ? AMDGPU::S_BUFFER_LOAD_DWORDX4_IMM :
-                              AMDGPU::S_BUFFER_LOAD_DWORDX2_IMM;
+  const unsigned Opcode = getNewOpcode(CI);
+
+  const TargetRegisterClass *SuperRC = getTargetRegisterClass(CI);
 
-  const TargetRegisterClass *SuperRC =
-    CI.IsX2 ? &AMDGPU::SReg_128RegClass : &AMDGPU::SReg_64_XEXECRegClass;
   unsigned DestReg = MRI->createVirtualRegister(SuperRC);
   unsigned MergedOffset = std::min(CI.Offset0, CI.Offset1);
 
@@ -652,14 +858,11 @@ MachineBasicBlock::iterator SILoadStoreOptimizer::mergeSBufferLoadImmPair(
       .add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::sbase))
       .addImm(MergedOffset) // offset
       .addImm(CI.GLC0)      // glc
-      .setMemRefs(CI.I->mergeMemRefsWith(*CI.Paired));
-
-  unsigned SubRegIdx0 = CI.IsX2 ? AMDGPU::sub0_sub1 : AMDGPU::sub0;
-  unsigned SubRegIdx1 = CI.IsX2 ? AMDGPU::sub2_sub3 : AMDGPU::sub1;
+      .cloneMergedMemRefs({&*CI.I, &*CI.Paired});
 
-  // Handle descending offsets
-  if (CI.Offset0 > CI.Offset1)
-    std::swap(SubRegIdx0, SubRegIdx1);
+  std::pair<unsigned, unsigned> SubRegIdx = getSubRegIdxs(CI);
+  const unsigned SubRegIdx0 = std::get<0>(SubRegIdx);
+  const unsigned SubRegIdx1 = std::get<1>(SubRegIdx);
 
   // Copy to the old destination registers.
   const MCInstrDesc &CopyDesc = TII->get(TargetOpcode::COPY);
@@ -681,29 +884,25 @@ MachineBasicBlock::iterator SILoadStoreOptimizer::mergeSBufferLoadImmPair(
   return Next;
 }
 
-MachineBasicBlock::iterator SILoadStoreOptimizer::mergeBufferLoadPair(
-  CombineInfo &CI) {
+MachineBasicBlock::iterator
+SILoadStoreOptimizer::mergeBufferLoadPair(CombineInfo &CI) {
   MachineBasicBlock *MBB = CI.I->getParent();
   DebugLoc DL = CI.I->getDebugLoc();
-  unsigned Opcode;
 
-  if (CI.InstClass == BUFFER_LOAD_OFFEN) {
-    Opcode = CI.IsX2 ? AMDGPU::BUFFER_LOAD_DWORDX4_OFFEN :
-                       AMDGPU::BUFFER_LOAD_DWORDX2_OFFEN;
-  } else {
-    Opcode = CI.IsX2 ? AMDGPU::BUFFER_LOAD_DWORDX4_OFFSET :
-                       AMDGPU::BUFFER_LOAD_DWORDX2_OFFSET;
-  }
+  const unsigned Opcode = getNewOpcode(CI);
 
-  const TargetRegisterClass *SuperRC =
-    CI.IsX2 ? &AMDGPU::VReg_128RegClass : &AMDGPU::VReg_64RegClass;
+  const TargetRegisterClass *SuperRC = getTargetRegisterClass(CI);
+
+  // Copy to the new source register.
   unsigned DestReg = MRI->createVirtualRegister(SuperRC);
   unsigned MergedOffset = std::min(CI.Offset0, CI.Offset1);
 
   auto MIB = BuildMI(*MBB, CI.Paired, DL, TII->get(Opcode), DestReg);
 
-  if (CI.InstClass == BUFFER_LOAD_OFFEN)
-      MIB.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::vaddr));
+  const unsigned Regs = getRegs(Opcode);
+
+  if (Regs & VADDR)
+    MIB.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::vaddr));
 
   MIB.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::srsrc))
       .add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::soffset))
@@ -711,14 +910,11 @@ MachineBasicBlock::iterator SILoadStoreOptimizer::mergeBufferLoadPair(
       .addImm(CI.GLC0)      // glc
       .addImm(CI.SLC0)      // slc
       .addImm(0)            // tfe
-      .setMemRefs(CI.I->mergeMemRefsWith(*CI.Paired));
-
-  unsigned SubRegIdx0 = CI.IsX2 ? AMDGPU::sub0_sub1 : AMDGPU::sub0;
-  unsigned SubRegIdx1 = CI.IsX2 ? AMDGPU::sub2_sub3 : AMDGPU::sub1;
+      .cloneMergedMemRefs({&*CI.I, &*CI.Paired});
 
-  // Handle descending offsets
-  if (CI.Offset0 > CI.Offset1)
-    std::swap(SubRegIdx0, SubRegIdx1);
+  std::pair<unsigned, unsigned> SubRegIdx = getSubRegIdxs(CI);
+  const unsigned SubRegIdx0 = std::get<0>(SubRegIdx);
+  const unsigned SubRegIdx1 = std::get<1>(SubRegIdx);
 
   // Copy to the old destination registers.
   const MCInstrDesc &CopyDesc = TII->get(TargetOpcode::COPY);
@@ -740,57 +936,137 @@ MachineBasicBlock::iterator SILoadStoreOptimizer::mergeBufferLoadPair(
   return Next;
 }
 
-unsigned SILoadStoreOptimizer::promoteBufferStoreOpcode(
-  const MachineInstr &I, bool &IsX2, bool &IsOffen) const {
-  IsX2 = false;
-  IsOffen = false;
-
-  switch (I.getOpcode()) {
-  case AMDGPU::BUFFER_STORE_DWORD_OFFEN:
-    IsOffen = true;
-    return AMDGPU::BUFFER_STORE_DWORDX2_OFFEN;
-  case AMDGPU::BUFFER_STORE_DWORD_OFFEN_exact:
-    IsOffen = true;
-    return AMDGPU::BUFFER_STORE_DWORDX2_OFFEN_exact;
-  case AMDGPU::BUFFER_STORE_DWORDX2_OFFEN:
-    IsX2 = true;
-    IsOffen = true;
-    return AMDGPU::BUFFER_STORE_DWORDX4_OFFEN;
-  case AMDGPU::BUFFER_STORE_DWORDX2_OFFEN_exact:
-    IsX2 = true;
-    IsOffen = true;
-    return AMDGPU::BUFFER_STORE_DWORDX4_OFFEN_exact;
-  case AMDGPU::BUFFER_STORE_DWORD_OFFSET:
-    return AMDGPU::BUFFER_STORE_DWORDX2_OFFSET;
-  case AMDGPU::BUFFER_STORE_DWORD_OFFSET_exact:
-    return AMDGPU::BUFFER_STORE_DWORDX2_OFFSET_exact;
-  case AMDGPU::BUFFER_STORE_DWORDX2_OFFSET:
-    IsX2 = true;
-    return AMDGPU::BUFFER_STORE_DWORDX4_OFFSET;
-  case AMDGPU::BUFFER_STORE_DWORDX2_OFFSET_exact:
-    IsX2 = true;
-    return AMDGPU::BUFFER_STORE_DWORDX4_OFFSET_exact;
-  }
-  return 0;
-}
-
-MachineBasicBlock::iterator SILoadStoreOptimizer::mergeBufferStorePair(
-  CombineInfo &CI) {
+unsigned SILoadStoreOptimizer::getNewOpcode(const CombineInfo &CI) {
+  const unsigned Width = CI.Width0 + CI.Width1;
+
+  switch (CI.InstClass) {
+  default:
+    return AMDGPU::getMUBUFOpcode(CI.InstClass, Width);
+  case UNKNOWN:
+    llvm_unreachable("Unknown instruction class");
+  case S_BUFFER_LOAD_IMM:
+    switch (Width) {
+    default:
+      return 0;
+    case 2:
+      return AMDGPU::S_BUFFER_LOAD_DWORDX2_IMM;
+    case 4:
+      return AMDGPU::S_BUFFER_LOAD_DWORDX4_IMM;
+    }
+  }
+}
+
+std::pair<unsigned, unsigned>
+SILoadStoreOptimizer::getSubRegIdxs(const CombineInfo &CI) {
+  if (CI.Offset0 > CI.Offset1) {
+    switch (CI.Width0) {
+    default:
+      return std::make_pair(0, 0);
+    case 1:
+      switch (CI.Width1) {
+      default:
+        return std::make_pair(0, 0);
+      case 1:
+        return std::make_pair(AMDGPU::sub1, AMDGPU::sub0);
+      case 2:
+        return std::make_pair(AMDGPU::sub2, AMDGPU::sub0_sub1);
+      case 3:
+        return std::make_pair(AMDGPU::sub3, AMDGPU::sub0_sub1_sub2);
+      }
+    case 2:
+      switch (CI.Width1) {
+      default:
+        return std::make_pair(0, 0);
+      case 1:
+        return std::make_pair(AMDGPU::sub1_sub2, AMDGPU::sub0);
+      case 2:
+        return std::make_pair(AMDGPU::sub2_sub3, AMDGPU::sub0_sub1);
+      }
+    case 3:
+      switch (CI.Width1) {
+      default:
+        return std::make_pair(0, 0);
+      case 1:
+        return std::make_pair(AMDGPU::sub1_sub2_sub3, AMDGPU::sub0);
+      }
+    }
+  } else {
+    switch (CI.Width0) {
+    default:
+      return std::make_pair(0, 0);
+    case 1:
+      switch (CI.Width1) {
+      default:
+        return std::make_pair(0, 0);
+      case 1:
+        return std::make_pair(AMDGPU::sub0, AMDGPU::sub1);
+      case 2:
+        return std::make_pair(AMDGPU::sub0, AMDGPU::sub1_sub2);
+      case 3:
+        return std::make_pair(AMDGPU::sub0, AMDGPU::sub1_sub2_sub3);
+      }
+    case 2:
+      switch (CI.Width1) {
+      default:
+        return std::make_pair(0, 0);
+      case 1:
+        return std::make_pair(AMDGPU::sub0_sub1, AMDGPU::sub2);
+      case 2:
+        return std::make_pair(AMDGPU::sub0_sub1, AMDGPU::sub2_sub3);
+      }
+    case 3:
+      switch (CI.Width1) {
+      default:
+        return std::make_pair(0, 0);
+      case 1:
+        return std::make_pair(AMDGPU::sub0_sub1_sub2, AMDGPU::sub3);
+      }
+    }
+  }
+}
+
+const TargetRegisterClass *
+SILoadStoreOptimizer::getTargetRegisterClass(const CombineInfo &CI) {
+  if (CI.InstClass == S_BUFFER_LOAD_IMM) {
+    switch (CI.Width0 + CI.Width1) {
+    default:
+      return nullptr;
+    case 2:
+      return &AMDGPU::SReg_64_XEXECRegClass;
+    case 4:
+      return &AMDGPU::SReg_128RegClass;
+    case 8:
+      return &AMDGPU::SReg_256RegClass;
+    case 16:
+      return &AMDGPU::SReg_512RegClass;
+    }
+  } else {
+    switch (CI.Width0 + CI.Width1) {
+    default:
+      return nullptr;
+    case 2:
+      return &AMDGPU::VReg_64RegClass;
+    case 3:
+      return &AMDGPU::VReg_96RegClass;
+    case 4:
+      return &AMDGPU::VReg_128RegClass;
+    }
+  }
+}
+
+MachineBasicBlock::iterator
+SILoadStoreOptimizer::mergeBufferStorePair(CombineInfo &CI) {
   MachineBasicBlock *MBB = CI.I->getParent();
   DebugLoc DL = CI.I->getDebugLoc();
-  bool Unused1, Unused2;
-  unsigned Opcode = promoteBufferStoreOpcode(*CI.I, Unused1, Unused2);
 
-  unsigned SubRegIdx0 = CI.IsX2 ? AMDGPU::sub0_sub1 : AMDGPU::sub0;
-  unsigned SubRegIdx1 = CI.IsX2 ? AMDGPU::sub2_sub3 : AMDGPU::sub1;
+  const unsigned Opcode = getNewOpcode(CI);
 
-  // Handle descending offsets
-  if (CI.Offset0 > CI.Offset1)
-    std::swap(SubRegIdx0, SubRegIdx1);
+  std::pair<unsigned, unsigned> SubRegIdx = getSubRegIdxs(CI);
+  const unsigned SubRegIdx0 = std::get<0>(SubRegIdx);
+  const unsigned SubRegIdx1 = std::get<1>(SubRegIdx);
 
   // Copy to the new source register.
-  const TargetRegisterClass *SuperRC =
-    CI.IsX2 ? &AMDGPU::VReg_128RegClass : &AMDGPU::VReg_64RegClass;
+  const TargetRegisterClass *SuperRC = getTargetRegisterClass(CI);
   unsigned SrcReg = MRI->createVirtualRegister(SuperRC);
 
   const auto *Src0 = TII->getNamedOperand(*CI.I, AMDGPU::OpName::vdata);
@@ -803,18 +1079,20 @@ MachineBasicBlock::iterator SILoadStoreOptimizer::mergeBufferStorePair(
       .addImm(SubRegIdx1);
 
   auto MIB = BuildMI(*MBB, CI.Paired, DL, TII->get(Opcode))
-      .addReg(SrcReg, RegState::Kill);
+                 .addReg(SrcReg, RegState::Kill);
 
-  if (CI.InstClass == BUFFER_STORE_OFFEN)
-      MIB.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::vaddr));
+  const unsigned Regs = getRegs(Opcode);
+
+  if (Regs & VADDR)
+    MIB.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::vaddr));
 
   MIB.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::srsrc))
       .add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::soffset))
       .addImm(std::min(CI.Offset0, CI.Offset1)) // offset
-      .addImm(CI.GLC0)      // glc
-      .addImm(CI.SLC0)      // slc
-      .addImm(0)            // tfe
-      .setMemRefs(CI.I->mergeMemRefsWith(*CI.Paired));
+      .addImm(CI.GLC0)                          // glc
+      .addImm(CI.SLC0)                          // slc
+      .addImm(0)                                // tfe
+      .cloneMergedMemRefs({&*CI.I, &*CI.Paired});
 
   moveInstsAfter(MIB, CI.InstsToMove);
 
@@ -824,105 +1102,399 @@ MachineBasicBlock::iterator SILoadStoreOptimizer::mergeBufferStorePair(
   return Next;
 }
 
+MachineOperand
+SILoadStoreOptimizer::createRegOrImm(int32_t Val, MachineInstr &MI) {
+  APInt V(32, Val, true);
+  if (TII->isInlineConstant(V))
+    return MachineOperand::CreateImm(Val);
+
+  unsigned Reg = MRI->createVirtualRegister(&AMDGPU::SReg_32RegClass);
+  MachineInstr *Mov =
+  BuildMI(*MI.getParent(), MI.getIterator(), MI.getDebugLoc(),
+          TII->get(AMDGPU::S_MOV_B32), Reg)
+    .addImm(Val);
+  (void)Mov;
+  LLVM_DEBUG(dbgs() << "    "; Mov->dump());
+  return MachineOperand::CreateReg(Reg, false);
+}
+
+// Compute base address using Addr and return the final register.
+unsigned SILoadStoreOptimizer::computeBase(MachineInstr &MI,
+                                           const MemAddress &Addr) {
+  MachineBasicBlock *MBB = MI.getParent();
+  MachineBasicBlock::iterator MBBI = MI.getIterator();
+  DebugLoc DL = MI.getDebugLoc();
+
+  assert((TRI->getRegSizeInBits(Addr.Base.LoReg, *MRI) == 32 ||
+          Addr.Base.LoSubReg) &&
+         "Expected 32-bit Base-Register-Low!!");
+
+  assert((TRI->getRegSizeInBits(Addr.Base.HiReg, *MRI) == 32 ||
+          Addr.Base.HiSubReg) &&
+         "Expected 32-bit Base-Register-Hi!!");
+
+  LLVM_DEBUG(dbgs() << "  Re-Computed Anchor-Base:\n");
+  MachineOperand OffsetLo = createRegOrImm(static_cast<int32_t>(Addr.Offset), MI);
+  MachineOperand OffsetHi =
+    createRegOrImm(static_cast<int32_t>(Addr.Offset >> 32), MI);
+  unsigned CarryReg = MRI->createVirtualRegister(&AMDGPU::SReg_64_XEXECRegClass);
+  unsigned DeadCarryReg =
+    MRI->createVirtualRegister(&AMDGPU::SReg_64_XEXECRegClass);
+
+  unsigned DestSub0 = MRI->createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+  unsigned DestSub1 = MRI->createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+  MachineInstr *LoHalf =
+    BuildMI(*MBB, MBBI, DL, TII->get(AMDGPU::V_ADD_I32_e64), DestSub0)
+      .addReg(CarryReg, RegState::Define)
+      .addReg(Addr.Base.LoReg, 0, Addr.Base.LoSubReg)
+    .add(OffsetLo);
+  (void)LoHalf;
+  LLVM_DEBUG(dbgs() << "    "; LoHalf->dump(););
+
+  MachineInstr *HiHalf =
+  BuildMI(*MBB, MBBI, DL, TII->get(AMDGPU::V_ADDC_U32_e64), DestSub1)
+    .addReg(DeadCarryReg, RegState::Define | RegState::Dead)
+    .addReg(Addr.Base.HiReg, 0, Addr.Base.HiSubReg)
+    .add(OffsetHi)
+    .addReg(CarryReg, RegState::Kill);
+  (void)HiHalf;
+  LLVM_DEBUG(dbgs() << "    "; HiHalf->dump(););
+
+  unsigned FullDestReg = MRI->createVirtualRegister(&AMDGPU::VReg_64RegClass);
+  MachineInstr *FullBase =
+    BuildMI(*MBB, MBBI, DL, TII->get(TargetOpcode::REG_SEQUENCE), FullDestReg)
+      .addReg(DestSub0)
+      .addImm(AMDGPU::sub0)
+      .addReg(DestSub1)
+      .addImm(AMDGPU::sub1);
+  (void)FullBase;
+  LLVM_DEBUG(dbgs() << "    "; FullBase->dump(); dbgs() << "\n";);
+
+  return FullDestReg;
+}
+
+// Update base and offset with the NewBase and NewOffset in MI.
+void SILoadStoreOptimizer::updateBaseAndOffset(MachineInstr &MI,
+                                               unsigned NewBase,
+                                               int32_t NewOffset) {
+  TII->getNamedOperand(MI, AMDGPU::OpName::vaddr)->setReg(NewBase);
+  TII->getNamedOperand(MI, AMDGPU::OpName::offset)->setImm(NewOffset);
+}
+
+Optional<int32_t>
+SILoadStoreOptimizer::extractConstOffset(const MachineOperand &Op) {
+  if (Op.isImm())
+    return Op.getImm();
+
+  if (!Op.isReg())
+    return None;
+
+  MachineInstr *Def = MRI->getUniqueVRegDef(Op.getReg());
+  if (!Def || Def->getOpcode() != AMDGPU::S_MOV_B32 ||
+      !Def->getOperand(1).isImm())
+    return None;
+
+  return Def->getOperand(1).getImm();
+}
+
+// Analyze Base and extracts:
+//  - 32bit base registers, subregisters
+//  - 64bit constant offset
+// Expecting base computation as:
+//   %OFFSET0:sgpr_32 = S_MOV_B32 8000
+//   %LO:vgpr_32, %c:sreg_64_xexec =
+//       V_ADD_I32_e64 %BASE_LO:vgpr_32, %103:sgpr_32,
+//   %HI:vgpr_32, = V_ADDC_U32_e64 %BASE_HI:vgpr_32, 0, killed %c:sreg_64_xexec
+//   %Base:vreg_64 =
+//       REG_SEQUENCE %LO:vgpr_32, %subreg.sub0, %HI:vgpr_32, %subreg.sub1
+void SILoadStoreOptimizer::processBaseWithConstOffset(const MachineOperand &Base,
+                                                      MemAddress &Addr) {
+  if (!Base.isReg())
+    return;
+
+  MachineInstr *Def = MRI->getUniqueVRegDef(Base.getReg());
+  if (!Def || Def->getOpcode() != AMDGPU::REG_SEQUENCE
+      || Def->getNumOperands() != 5)
+    return;
+
+  MachineOperand BaseLo = Def->getOperand(1);
+  MachineOperand BaseHi = Def->getOperand(3);
+  if (!BaseLo.isReg() || !BaseHi.isReg())
+    return;
+
+  MachineInstr *BaseLoDef = MRI->getUniqueVRegDef(BaseLo.getReg());
+  MachineInstr *BaseHiDef = MRI->getUniqueVRegDef(BaseHi.getReg());
+
+  if (!BaseLoDef || BaseLoDef->getOpcode() != AMDGPU::V_ADD_I32_e64 ||
+      !BaseHiDef || BaseHiDef->getOpcode() != AMDGPU::V_ADDC_U32_e64)
+    return;
+
+  const auto *Src0 = TII->getNamedOperand(*BaseLoDef, AMDGPU::OpName::src0);
+  const auto *Src1 = TII->getNamedOperand(*BaseLoDef, AMDGPU::OpName::src1);
+
+  auto Offset0P = extractConstOffset(*Src0);
+  if (Offset0P)
+    BaseLo = *Src1;
+  else {
+    if (!(Offset0P = extractConstOffset(*Src1)))
+      return;
+    BaseLo = *Src0;
+  }
+
+  Src0 = TII->getNamedOperand(*BaseHiDef, AMDGPU::OpName::src0);
+  Src1 = TII->getNamedOperand(*BaseHiDef, AMDGPU::OpName::src1);
+
+  if (Src0->isImm())
+    std::swap(Src0, Src1);
+
+  if (!Src1->isImm())
+    return;
+
+  uint64_t Offset1 = Src1->getImm();
+  BaseHi = *Src0;
+
+  Addr.Base.LoReg = BaseLo.getReg();
+  Addr.Base.HiReg = BaseHi.getReg();
+  Addr.Base.LoSubReg = BaseLo.getSubReg();
+  Addr.Base.HiSubReg = BaseHi.getSubReg();
+  Addr.Offset = (*Offset0P & 0x00000000ffffffff) | (Offset1 << 32);
+}
+
+bool SILoadStoreOptimizer::promoteConstantOffsetToImm(
+    MachineInstr &MI,
+    MemInfoMap &Visited,
+    SmallPtrSet<MachineInstr *, 4> &AnchorList) {
+
+  // TODO: Support flat and scratch.
+  if (AMDGPU::getGlobalSaddrOp(MI.getOpcode()) < 0 ||
+      TII->getNamedOperand(MI, AMDGPU::OpName::vdata) != NULL)
+    return false;
+
+  // TODO: Support Store.
+  if (!MI.mayLoad())
+    return false;
+
+  if (AnchorList.count(&MI))
+    return false;
+
+  LLVM_DEBUG(dbgs() << "\nTryToPromoteConstantOffsetToImmFor "; MI.dump());
+
+  if (TII->getNamedOperand(MI, AMDGPU::OpName::offset)->getImm()) {
+    LLVM_DEBUG(dbgs() << "  Const-offset is already promoted.\n";);
+    return false;
+  }
+
+  // Step1: Find the base-registers and a 64bit constant offset.
+  MachineOperand &Base = *TII->getNamedOperand(MI, AMDGPU::OpName::vaddr);
+  MemAddress MAddr;
+  if (Visited.find(&MI) == Visited.end()) {
+    processBaseWithConstOffset(Base, MAddr);
+    Visited[&MI] = MAddr;
+  } else
+    MAddr = Visited[&MI];
+
+  if (MAddr.Offset == 0) {
+    LLVM_DEBUG(dbgs() << "  Failed to extract constant-offset or there are no"
+                         " constant offsets that can be promoted.\n";);
+    return false;
+  }
+
+  LLVM_DEBUG(dbgs() << "  BASE: {" << MAddr.Base.HiReg << ", "
+             << MAddr.Base.LoReg << "} Offset: " << MAddr.Offset << "\n\n";);
+
+  // Step2: Traverse through MI's basic block and find an anchor(that has the
+  // same base-registers) with the highest 13bit distance from MI's offset.
+  // E.g. (64bit loads)
+  // bb:
+  //   addr1 = &a + 4096;   load1 = load(addr1,  0)
+  //   addr2 = &a + 6144;   load2 = load(addr2,  0)
+  //   addr3 = &a + 8192;   load3 = load(addr3,  0)
+  //   addr4 = &a + 10240;  load4 = load(addr4,  0)
+  //   addr5 = &a + 12288;  load5 = load(addr5,  0)
+  //
+  // Starting from the first load, the optimization will try to find a new base
+  // from which (&a + 4096) has 13 bit distance. Both &a + 6144 and &a + 8192
+  // has 13bit distance from &a + 4096. The heuristic considers &a + 8192
+  // as the new-base(anchor) because of the maximum distance which can
+  // accomodate more intermediate bases presumeably.
+  //
+  // Step3: move (&a + 8192) above load1. Compute and promote offsets from
+  // (&a + 8192) for load1, load2, load4.
+  //   addr = &a + 8192
+  //   load1 = load(addr,       -4096)
+  //   load2 = load(addr,       -2048)
+  //   load3 = load(addr,       0)
+  //   load4 = load(addr,       2048)
+  //   addr5 = &a + 12288;  load5 = load(addr5,  0)
+  //
+  MachineInstr *AnchorInst = nullptr;
+  MemAddress AnchorAddr;
+  uint32_t MaxDist = std::numeric_limits<uint32_t>::min();
+  SmallVector<std::pair<MachineInstr *, int64_t>, 4> InstsWCommonBase;
+
+  MachineBasicBlock *MBB = MI.getParent();
+  MachineBasicBlock::iterator E = MBB->end();
+  MachineBasicBlock::iterator MBBI = MI.getIterator();
+  ++MBBI;
+  const SITargetLowering *TLI =
+    static_cast<const SITargetLowering *>(STM->getTargetLowering());
+
+  for ( ; MBBI != E; ++MBBI) {
+    MachineInstr &MINext = *MBBI;
+    // TODO: Support finding an anchor(with same base) from store addresses or
+    // any other load addresses where the opcodes are different.
+    if (MINext.getOpcode() != MI.getOpcode() ||
+        TII->getNamedOperand(MINext, AMDGPU::OpName::offset)->getImm())
+      continue;
+
+    const MachineOperand &BaseNext =
+      *TII->getNamedOperand(MINext, AMDGPU::OpName::vaddr);
+    MemAddress MAddrNext;
+    if (Visited.find(&MINext) == Visited.end()) {
+      processBaseWithConstOffset(BaseNext, MAddrNext);
+      Visited[&MINext] = MAddrNext;
+    } else
+      MAddrNext = Visited[&MINext];
+
+    if (MAddrNext.Base.LoReg != MAddr.Base.LoReg ||
+        MAddrNext.Base.HiReg != MAddr.Base.HiReg ||
+        MAddrNext.Base.LoSubReg != MAddr.Base.LoSubReg ||
+        MAddrNext.Base.HiSubReg != MAddr.Base.HiSubReg)
+      continue;
+
+    InstsWCommonBase.push_back(std::make_pair(&MINext, MAddrNext.Offset));
+
+    int64_t Dist = MAddr.Offset - MAddrNext.Offset;
+    TargetLoweringBase::AddrMode AM;
+    AM.HasBaseReg = true;
+    AM.BaseOffs = Dist;
+    if (TLI->isLegalGlobalAddressingMode(AM) &&
+        (uint32_t)std::abs(Dist) > MaxDist) {
+      MaxDist = std::abs(Dist);
+
+      AnchorAddr = MAddrNext;
+      AnchorInst = &MINext;
+    }
+  }
+
+  if (AnchorInst) {
+    LLVM_DEBUG(dbgs() << "  Anchor-Inst(with max-distance from Offset): ";
+               AnchorInst->dump());
+    LLVM_DEBUG(dbgs() << "  Anchor-Offset from BASE: "
+               <<  AnchorAddr.Offset << "\n\n");
+
+    // Instead of moving up, just re-compute anchor-instruction's base address.
+    unsigned Base = computeBase(MI, AnchorAddr);
+
+    updateBaseAndOffset(MI, Base, MAddr.Offset - AnchorAddr.Offset);
+    LLVM_DEBUG(dbgs() << "  After promotion: "; MI.dump(););
+
+    for (auto P : InstsWCommonBase) {
+      TargetLoweringBase::AddrMode AM;
+      AM.HasBaseReg = true;
+      AM.BaseOffs = P.second - AnchorAddr.Offset;
+
+      if (TLI->isLegalGlobalAddressingMode(AM)) {
+        LLVM_DEBUG(dbgs() << "  Promote Offset(" << P.second;
+                   dbgs() << ")"; P.first->dump());
+        updateBaseAndOffset(*P.first, Base, P.second - AnchorAddr.Offset);
+        LLVM_DEBUG(dbgs() << "     After promotion: "; P.first->dump());
+      }
+    }
+    AnchorList.insert(AnchorInst);
+    return true;
+  }
+
+  return false;
+}
+
 // Scan through looking for adjacent LDS operations with constant offsets from
 // the same base register. We rely on the scheduler to do the hard work of
 // clustering nearby loads, and assume these are all adjacent.
 bool SILoadStoreOptimizer::optimizeBlock(MachineBasicBlock &MBB) {
   bool Modified = false;
 
+  // Contain the list
+  MemInfoMap Visited;
+  // Contains the list of instructions for which constant offsets are being
+  // promoted to the IMM.
+  SmallPtrSet<MachineInstr *, 4> AnchorList;
+
   for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end(); I != E;) {
     MachineInstr &MI = *I;
 
+    if (promoteConstantOffsetToImm(MI, Visited, AnchorList))
+      Modified = true;
+
     // Don't combine if volatile.
     if (MI.hasOrderedMemoryRef()) {
       ++I;
       continue;
     }
 
+    const unsigned Opc = MI.getOpcode();
+
     CombineInfo CI;
     CI.I = I;
-    unsigned Opc = MI.getOpcode();
-    if (Opc == AMDGPU::DS_READ_B32 || Opc == AMDGPU::DS_READ_B64 ||
-        Opc == AMDGPU::DS_READ_B32_gfx9 || Opc == AMDGPU::DS_READ_B64_gfx9) {
+    CI.InstClass = getInstClass(Opc);
 
-      CI.InstClass = DS_READ_WRITE;
+    switch (CI.InstClass) {
+    default:
+      break;
+    case DS_READ:
       CI.EltSize =
-        (Opc == AMDGPU::DS_READ_B64 || Opc == AMDGPU::DS_READ_B64_gfx9) ? 8 : 4;
-
+          (Opc == AMDGPU::DS_READ_B64 || Opc == AMDGPU::DS_READ_B64_gfx9) ? 8
+                                                                          : 4;
       if (findMatchingInst(CI)) {
         Modified = true;
         I = mergeRead2Pair(CI);
       } else {
         ++I;
       }
-
       continue;
-    } else if (Opc == AMDGPU::DS_WRITE_B32 || Opc == AMDGPU::DS_WRITE_B64 ||
-               Opc == AMDGPU::DS_WRITE_B32_gfx9 ||
-               Opc == AMDGPU::DS_WRITE_B64_gfx9) {
-      CI.InstClass = DS_READ_WRITE;
-      CI.EltSize
-        = (Opc == AMDGPU::DS_WRITE_B64 || Opc == AMDGPU::DS_WRITE_B64_gfx9) ? 8 : 4;
-
+    case DS_WRITE:
+      CI.EltSize =
+          (Opc == AMDGPU::DS_WRITE_B64 || Opc == AMDGPU::DS_WRITE_B64_gfx9) ? 8
+                                                                            : 4;
       if (findMatchingInst(CI)) {
         Modified = true;
         I = mergeWrite2Pair(CI);
       } else {
         ++I;
       }
-
       continue;
-    }
-    if (Opc == AMDGPU::S_BUFFER_LOAD_DWORD_IMM ||
-        Opc == AMDGPU::S_BUFFER_LOAD_DWORDX2_IMM) {
-      // EltSize is in units of the offset encoding.
-      CI.InstClass = S_BUFFER_LOAD_IMM;
+    case S_BUFFER_LOAD_IMM:
       CI.EltSize = AMDGPU::getSMRDEncodedOffset(*STM, 4);
-      CI.IsX2 = Opc == AMDGPU::S_BUFFER_LOAD_DWORDX2_IMM;
       if (findMatchingInst(CI)) {
         Modified = true;
         I = mergeSBufferLoadImmPair(CI);
-        if (!CI.IsX2)
-          CreatedX2++;
+        OptimizeAgain |= (CI.Width0 + CI.Width1) < 16;
       } else {
         ++I;
       }
       continue;
-    }
-    if (Opc == AMDGPU::BUFFER_LOAD_DWORD_OFFEN ||
-        Opc == AMDGPU::BUFFER_LOAD_DWORDX2_OFFEN ||
-        Opc == AMDGPU::BUFFER_LOAD_DWORD_OFFSET ||
-        Opc == AMDGPU::BUFFER_LOAD_DWORDX2_OFFSET) {
-      if (Opc == AMDGPU::BUFFER_LOAD_DWORD_OFFEN ||
-          Opc == AMDGPU::BUFFER_LOAD_DWORDX2_OFFEN)
-        CI.InstClass = BUFFER_LOAD_OFFEN;
-      else
-        CI.InstClass = BUFFER_LOAD_OFFSET;
-
+    case BUFFER_LOAD_OFFEN:
+    case BUFFER_LOAD_OFFSET:
+    case BUFFER_LOAD_OFFEN_exact:
+    case BUFFER_LOAD_OFFSET_exact:
       CI.EltSize = 4;
-      CI.IsX2 = Opc == AMDGPU::BUFFER_LOAD_DWORDX2_OFFEN ||
-                Opc == AMDGPU::BUFFER_LOAD_DWORDX2_OFFSET;
       if (findMatchingInst(CI)) {
         Modified = true;
         I = mergeBufferLoadPair(CI);
-        if (!CI.IsX2)
-          CreatedX2++;
+        OptimizeAgain |= (CI.Width0 + CI.Width1) < 4;
       } else {
         ++I;
       }
       continue;
-    }
-
-    bool StoreIsX2, IsOffen;
-    if (promoteBufferStoreOpcode(*I, StoreIsX2, IsOffen)) {
-      CI.InstClass = IsOffen ? BUFFER_STORE_OFFEN : BUFFER_STORE_OFFSET;
+    case BUFFER_STORE_OFFEN:
+    case BUFFER_STORE_OFFSET:
+    case BUFFER_STORE_OFFEN_exact:
+    case BUFFER_STORE_OFFSET_exact:
       CI.EltSize = 4;
-      CI.IsX2 = StoreIsX2;
       if (findMatchingInst(CI)) {
         Modified = true;
         I = mergeBufferStorePair(CI);
-        if (!CI.IsX2)
-          CreatedX2++;
+        OptimizeAgain |= (CI.Width0 + CI.Width1) < 4;
       } else {
         ++I;
       }
@@ -956,12 +1528,10 @@ bool SILoadStoreOptimizer::runOnMachineFunction(MachineFunction &MF) {
   bool Modified = false;
 
   for (MachineBasicBlock &MBB : MF) {
-    CreatedX2 = 0;
-    Modified |= optimizeBlock(MBB);
-
-    // Run again to convert x2 to x4.
-    if (CreatedX2 >= 1)
+    do {
+      OptimizeAgain = false;
       Modified |= optimizeBlock(MBB);
+    } while (OptimizeAgain);
   }
 
   return Modified;