vendor/llvm/llvm-release_32-r168974

14 files changed, 754 insertions, 195 deletions

diff --git a/lib/Transforms/InstCombine/InstCombine.h b/lib/Transforms/InstCombine/InstCombine.h
index 0d5ef904ee472..7467eca7ab1fc 100644
--- a/lib/Transforms/InstCombine/InstCombine.h
+++ b/lib/Transforms/InstCombine/InstCombine.h
@@ -18,10 +18,11 @@
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/Support/InstVisitor.h"
 #include "llvm/Support/TargetFolder.h"
+#include "llvm/Transforms/Utils/SimplifyLibCalls.h"
 
 namespace llvm {
   class CallSite;
-  class TargetData;
+  class DataLayout;
   class TargetLibraryInfo;
   class DbgDeclareInst;
   class MemIntrinsic;
@@ -71,9 +72,10 @@ public:
 class LLVM_LIBRARY_VISIBILITY InstCombiner
                              : public FunctionPass,
                                public InstVisitor<InstCombiner, Instruction*> {
-  TargetData *TD;
+  DataLayout *TD;
   TargetLibraryInfo *TLI;
   bool MadeIRChange;
+  LibCallSimplifier *Simplifier;
 public:
   /// Worklist - All of the instructions that need to be simplified.
   InstCombineWorklist Worklist;
@@ -95,7 +97,7 @@ public:
 
   virtual void getAnalysisUsage(AnalysisUsage &AU) const;
 
-  TargetData *getTargetData() const { return TD; }
+  DataLayout *getDataLayout() const { return TD; }
 
   TargetLibraryInfo *getTargetLibraryInfo() const { return TLI; }
 
@@ -218,7 +220,7 @@ private:
                           Type *Ty);
 
   Instruction *visitCallSite(CallSite CS);
-  Instruction *tryOptimizeCall(CallInst *CI, const TargetData *TD);
+  Instruction *tryOptimizeCall(CallInst *CI, const DataLayout *TD);
   bool transformConstExprCastCall(CallSite CS);
   Instruction *transformCallThroughTrampoline(CallSite CS,
                                               IntrinsicInst *Tramp);
@@ -365,6 +367,10 @@ private:
 
 
   Value *EvaluateInDifferentType(Value *V, Type *Ty, bool isSigned);
+
+  /// Descale - Return a value X such that Val = X * Scale, or null if none.  If
+  /// the multiplication is known not to overflow then NoSignedWrap is set.
+  Value *Descale(Value *Val, APInt Scale, bool &NoSignedWrap);
 };
 
       
diff --git a/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/lib/Transforms/InstCombine/InstCombineAddSub.cpp
index 99b62f8d05a75..d8257e64d8370 100644
--- a/lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ b/lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -13,7 +13,7 @@
 
 #include "InstCombine.h"
 #include "llvm/Analysis/InstructionSimplify.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Support/GetElementPtrTypeIterator.h"
 #include "llvm/Support/PatternMatch.h"
 using namespace llvm;
diff --git a/lib/Transforms/InstCombine/InstCombineCalls.cpp b/lib/Transforms/InstCombine/InstCombineCalls.cpp
index cbe1ca4ddcec0..48f270429e5a6 100644
--- a/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -13,7 +13,7 @@
 
 #include "InstCombine.h"
 #include "llvm/Support/CallSite.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Analysis/MemoryBuiltins.h"
 #include "llvm/Transforms/Utils/BuildLibCalls.h"
 #include "llvm/Transforms/Utils/Local.h"
@@ -29,6 +29,26 @@ static Type *getPromotedType(Type *Ty) {
   return Ty;
 }
 
+/// reduceToSingleValueType - Given an aggregate type which ultimately holds a
+/// single scalar element, like {{{type}}} or [1 x type], return type.
+static Type *reduceToSingleValueType(Type *T) {
+  while (!T->isSingleValueType()) {
+    if (StructType *STy = dyn_cast<StructType>(T)) {
+      if (STy->getNumElements() == 1)
+        T = STy->getElementType(0);
+      else
+        break;
+    } else if (ArrayType *ATy = dyn_cast<ArrayType>(T)) {
+      if (ATy->getNumElements() == 1)
+        T = ATy->getElementType();
+      else
+        break;
+    } else
+      break;
+  }
+
+  return T;
+}
 
 Instruction *InstCombiner::SimplifyMemTransfer(MemIntrinsic *MI) {
   unsigned DstAlign = getKnownAlignment(MI->getArgOperand(0), TD);
@@ -74,35 +94,37 @@ Instruction *InstCombiner::SimplifyMemTransfer(MemIntrinsic *MI) {
   // dest address will be promotable.  See if we can find a better type than the
   // integer datatype.
   Value *StrippedDest = MI->getArgOperand(0)->stripPointerCasts();
+  MDNode *CopyMD = 0;
   if (StrippedDest != MI->getArgOperand(0)) {
     Type *SrcETy = cast<PointerType>(StrippedDest->getType())
                                     ->getElementType();
     if (TD && SrcETy->isSized() && TD->getTypeStoreSize(SrcETy) == Size) {
       // The SrcETy might be something like {{{double}}} or [1 x double].  Rip
       // down through these levels if so.
-      while (!SrcETy->isSingleValueType()) {
-        if (StructType *STy = dyn_cast<StructType>(SrcETy)) {
-          if (STy->getNumElements() == 1)
-            SrcETy = STy->getElementType(0);
-          else
-            break;
-        } else if (ArrayType *ATy = dyn_cast<ArrayType>(SrcETy)) {
-          if (ATy->getNumElements() == 1)
-            SrcETy = ATy->getElementType();
-          else
-            break;
-        } else
-          break;
-      }
+      SrcETy = reduceToSingleValueType(SrcETy);
 
       if (SrcETy->isSingleValueType()) {
         NewSrcPtrTy = PointerType::get(SrcETy, SrcAddrSp);
         NewDstPtrTy = PointerType::get(SrcETy, DstAddrSp);
+
+        // If the memcpy has metadata describing the members, see if we can
+        // get the TBAA tag describing our copy.
+        if (MDNode *M = MI->getMetadata(LLVMContext::MD_tbaa_struct)) {
+          if (M->getNumOperands() == 3 &&
+              M->getOperand(0) &&
+              isa<ConstantInt>(M->getOperand(0)) &&
+              cast<ConstantInt>(M->getOperand(0))->isNullValue() &&
+              M->getOperand(1) &&
+              isa<ConstantInt>(M->getOperand(1)) &&
+              cast<ConstantInt>(M->getOperand(1))->getValue() == Size &&
+              M->getOperand(2) &&
+              isa<MDNode>(M->getOperand(2)))
+            CopyMD = cast<MDNode>(M->getOperand(2));
+        }
       }
     }
   }
 
-
   // If the memcpy/memmove provides better alignment info than we can
   // infer, use it.
   SrcAlign = std::max(SrcAlign, CopyAlign);
@@ -112,8 +134,12 @@ Instruction *InstCombiner::SimplifyMemTransfer(MemIntrinsic *MI) {
   Value *Dest = Builder->CreateBitCast(MI->getArgOperand(0), NewDstPtrTy);
   LoadInst *L = Builder->CreateLoad(Src, MI->isVolatile());
   L->setAlignment(SrcAlign);
+  if (CopyMD)
+    L->setMetadata(LLVMContext::MD_tbaa, CopyMD);
   StoreInst *S = Builder->CreateStore(L, Dest, MI->isVolatile());
   S->setAlignment(DstAlign);
+  if (CopyMD)
+    S->setMetadata(LLVMContext::MD_tbaa, CopyMD);
 
   // Set the size of the copy to 0, it will be deleted on the next iteration.
   MI->setArgOperand(2, Constant::getNullValue(MemOpLength->getType()));
@@ -168,7 +194,7 @@ Instruction *InstCombiner::SimplifyMemSet(MemSetInst *MI) {
 /// the heavy lifting.
 ///
 Instruction *InstCombiner::visitCallInst(CallInst &CI) {
-  if (isFreeCall(&CI))
+  if (isFreeCall(&CI, TLI))
     return visitFree(CI);
 
   // If the caller function is nounwind, mark the call as nounwind, even if the
@@ -243,7 +269,7 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
   default: break;
   case Intrinsic::objectsize: {
     uint64_t Size;
-    if (getObjectSize(II->getArgOperand(0), Size, TD))
+    if (getObjectSize(II->getArgOperand(0), Size, TD, TLI))
       return ReplaceInstUsesWith(CI, ConstantInt::get(CI.getType(), Size));
     return 0;
   }
@@ -731,7 +757,7 @@ Instruction *InstCombiner::visitInvokeInst(InvokeInst &II) {
 /// passed through the varargs area, we can eliminate the use of the cast.
 static bool isSafeToEliminateVarargsCast(const CallSite CS,
                                          const CastInst * const CI,
-                                         const TargetData * const TD,
+                                         const DataLayout * const TD,
                                          const int ix) {
   if (!CI->isLosslessCast())
     return false;
@@ -752,49 +778,17 @@ static bool isSafeToEliminateVarargsCast(const CallSite CS,
   return true;
 }
 
-namespace {
-class InstCombineFortifiedLibCalls : public SimplifyFortifiedLibCalls {
-  InstCombiner *IC;
-protected:
-  void replaceCall(Value *With) {
-    NewInstruction = IC->ReplaceInstUsesWith(*CI, With);
-  }
-  bool isFoldable(unsigned SizeCIOp, unsigned SizeArgOp, bool isString) const {
-    if (CI->getArgOperand(SizeCIOp) == CI->getArgOperand(SizeArgOp))
-      return true;
-    if (ConstantInt *SizeCI =
-                           dyn_cast<ConstantInt>(CI->getArgOperand(SizeCIOp))) {
-      if (SizeCI->isAllOnesValue())
-        return true;
-      if (isString) {
-        uint64_t Len = GetStringLength(CI->getArgOperand(SizeArgOp));
-        // If the length is 0 we don't know how long it is and so we can't
-        // remove the check.
-        if (Len == 0) return false;
-        return SizeCI->getZExtValue() >= Len;
-      }
-      if (ConstantInt *Arg = dyn_cast<ConstantInt>(
-                                                  CI->getArgOperand(SizeArgOp)))
-        return SizeCI->getZExtValue() >= Arg->getZExtValue();
-    }
-    return false;
-  }
-public:
-  InstCombineFortifiedLibCalls(InstCombiner *IC) : IC(IC), NewInstruction(0) { }
-  Instruction *NewInstruction;
-};
-} // end anonymous namespace
-
 // Try to fold some different type of calls here.
 // Currently we're only working with the checking functions, memcpy_chk,
 // mempcpy_chk, memmove_chk, memset_chk, strcpy_chk, stpcpy_chk, strncpy_chk,
 // strcat_chk and strncat_chk.
-Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const TargetData *TD) {
+Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const DataLayout *TD) {
   if (CI->getCalledFunction() == 0) return 0;
 
-  InstCombineFortifiedLibCalls Simplifier(this);
-  Simplifier.fold(CI, TD, TLI);
-  return Simplifier.NewInstruction;
+  if (Value *With = Simplifier->optimizeCall(CI))
+    return ReplaceInstUsesWith(*CI, With);
+
+  return 0;
 }
 
 static IntrinsicInst *FindInitTrampolineFromAlloca(Value *TrampMem) {
@@ -877,7 +871,7 @@ static IntrinsicInst *FindInitTrampoline(Value *Callee) {
 // visitCallSite - Improvements for call and invoke instructions.
 //
 Instruction *InstCombiner::visitCallSite(CallSite CS) {
-  if (isAllocLikeFn(CS.getInstruction()))
+  if (isAllocLikeFn(CS.getInstruction(), TLI))
     return visitAllocSite(*CS.getInstruction());
 
   bool Changed = false;
@@ -961,7 +955,7 @@ Instruction *InstCombiner::visitCallSite(CallSite CS) {
     Changed = true;
   }
 
-  // Try to optimize the call if possible, we require TargetData for most of
+  // Try to optimize the call if possible, we require DataLayout for most of
   // this.  None of these calls are seen as possibly dead so go ahead and
   // delete the instruction now.
   if (CallInst *CI = dyn_cast<CallInst>(CS.getInstruction())) {
@@ -1013,8 +1007,8 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
       return false;   // Cannot transform this return value.
 
     if (!CallerPAL.isEmpty() && !Caller->use_empty()) {
-      Attributes RAttrs = CallerPAL.getRetAttributes();
-      if (RAttrs & Attribute::typeIncompatible(NewRetTy))
+      AttrBuilder RAttrs = CallerPAL.getRetAttributes();
+      if (RAttrs.hasAttributes(Attributes::typeIncompatible(NewRetTy)))
         return false;   // Attribute not compatible with transformed value.
     }
 
@@ -1044,12 +1038,13 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
       return false;   // Cannot transform this parameter value.
 
     Attributes Attrs = CallerPAL.getParamAttributes(i + 1);
-    if (Attrs & Attribute::typeIncompatible(ParamTy))
+    if (AttrBuilder(Attrs).
+          hasAttributes(Attributes::typeIncompatible(ParamTy)))
       return false;   // Attribute not compatible with transformed value.
 
     // If the parameter is passed as a byval argument, then we have to have a
     // sized type and the sized type has to have the same size as the old type.
-    if (ParamTy != ActTy && (Attrs & Attribute::ByVal)) {
+    if (ParamTy != ActTy && Attrs.hasAttribute(Attributes::ByVal)) {
       PointerType *ParamPTy = dyn_cast<PointerType>(ParamTy);
       if (ParamPTy == 0 || !ParamPTy->getElementType()->isSized() || TD == 0)
         return false;
@@ -1101,7 +1096,7 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
       if (CallerPAL.getSlot(i - 1).Index <= FT->getNumParams())
         break;
       Attributes PAttrs = CallerPAL.getSlot(i - 1).Attrs;
-      if (PAttrs & Attribute::VarArgsIncompatible)
+      if (PAttrs.hasIncompatibleWithVarArgsAttrs())
         return false;
     }
 
@@ -1114,15 +1109,17 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
   attrVec.reserve(NumCommonArgs);
 
   // Get any return attributes.
-  Attributes RAttrs = CallerPAL.getRetAttributes();
+  AttrBuilder RAttrs = CallerPAL.getRetAttributes();
 
   // If the return value is not being used, the type may not be compatible
   // with the existing attributes.  Wipe out any problematic attributes.
-  RAttrs &= ~Attribute::typeIncompatible(NewRetTy);
+  RAttrs.removeAttributes(Attributes::typeIncompatible(NewRetTy));
 
   // Add the new return attributes.
-  if (RAttrs)
-    attrVec.push_back(AttributeWithIndex::get(0, RAttrs));
+  if (RAttrs.hasAttributes())
+    attrVec.push_back(
+      AttributeWithIndex::get(AttrListPtr::ReturnIndex,
+                              Attributes::get(FT->getContext(), RAttrs)));
 
   AI = CS.arg_begin();
   for (unsigned i = 0; i != NumCommonArgs; ++i, ++AI) {
@@ -1136,7 +1133,8 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
     }
 
     // Add any parameter attributes.
-    if (Attributes PAttrs = CallerPAL.getParamAttributes(i + 1))
+    Attributes PAttrs = CallerPAL.getParamAttributes(i + 1);
+    if (PAttrs.hasAttributes())
       attrVec.push_back(AttributeWithIndex::get(i + 1, PAttrs));
   }
 
@@ -1164,19 +1162,23 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
         }
 
         // Add any parameter attributes.
-        if (Attributes PAttrs = CallerPAL.getParamAttributes(i + 1))
+        Attributes PAttrs = CallerPAL.getParamAttributes(i + 1);
+        if (PAttrs.hasAttributes())
           attrVec.push_back(AttributeWithIndex::get(i + 1, PAttrs));
       }
     }
   }
 
-  if (Attributes FnAttrs =  CallerPAL.getFnAttributes())
-    attrVec.push_back(AttributeWithIndex::get(~0, FnAttrs));
+  Attributes FnAttrs = CallerPAL.getFnAttributes();
+  if (FnAttrs.hasAttributes())
+    attrVec.push_back(AttributeWithIndex::get(AttrListPtr::FunctionIndex,
+                                              FnAttrs));
 
   if (NewRetTy->isVoidTy())
     Caller->setName("");   // Void type should not have a name.
 
-  const AttrListPtr &NewCallerPAL = AttrListPtr::get(attrVec);
+  const AttrListPtr &NewCallerPAL = AttrListPtr::get(Callee->getContext(),
+                                                     attrVec);
 
   Instruction *NC;
   if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
@@ -1240,8 +1242,9 @@ InstCombiner::transformCallThroughTrampoline(CallSite CS,
 
   // If the call already has the 'nest' attribute somewhere then give up -
   // otherwise 'nest' would occur twice after splicing in the chain.
-  if (Attrs.hasAttrSomewhere(Attribute::Nest))
-    return 0;
+  for (unsigned I = 0, E = Attrs.getNumAttrs(); I != E; ++I)
+    if (Attrs.getAttributesAtIndex(I).hasAttribute(Attributes::Nest))
+      return 0;
 
   assert(Tramp &&
          "transformCallThroughTrampoline called with incorrect CallSite.");
@@ -1254,12 +1257,12 @@ InstCombiner::transformCallThroughTrampoline(CallSite CS,
   if (!NestAttrs.isEmpty()) {
     unsigned NestIdx = 1;
     Type *NestTy = 0;
-    Attributes NestAttr = Attribute::None;
+    Attributes NestAttr;
 
     // Look for a parameter marked with the 'nest' attribute.
     for (FunctionType::param_iterator I = NestFTy->param_begin(),
          E = NestFTy->param_end(); I != E; ++NestIdx, ++I)
-      if (NestAttrs.paramHasAttr(NestIdx, Attribute::Nest)) {
+      if (NestAttrs.getParamAttributes(NestIdx).hasAttribute(Attributes::Nest)){
         // Record the parameter type and any other attributes.
         NestTy = *I;
         NestAttr = NestAttrs.getParamAttributes(NestIdx);
@@ -1278,8 +1281,10 @@ InstCombiner::transformCallThroughTrampoline(CallSite CS,
       // mean appending it.  Likewise for attributes.
 
       // Add any result attributes.
-      if (Attributes Attr = Attrs.getRetAttributes())
-        NewAttrs.push_back(AttributeWithIndex::get(0, Attr));
+      Attributes Attr = Attrs.getRetAttributes();
+      if (Attr.hasAttributes())
+        NewAttrs.push_back(AttributeWithIndex::get(AttrListPtr::ReturnIndex,
+                                                   Attr));
 
       {
         unsigned Idx = 1;
@@ -1299,7 +1304,8 @@ InstCombiner::transformCallThroughTrampoline(CallSite CS,
 
           // Add the original argument and attributes.
           NewArgs.push_back(*I);
-          if (Attributes Attr = Attrs.getParamAttributes(Idx))
+          Attr = Attrs.getParamAttributes(Idx);
+          if (Attr.hasAttributes())
             NewAttrs.push_back
               (AttributeWithIndex::get(Idx + (Idx >= NestIdx), Attr));
 
@@ -1308,8 +1314,10 @@ InstCombiner::transformCallThroughTrampoline(CallSite CS,
       }
 
       // Add any function attributes.
-      if (Attributes Attr = Attrs.getFnAttributes())
-        NewAttrs.push_back(AttributeWithIndex::get(~0, Attr));
+      Attr = Attrs.getFnAttributes();
+      if (Attr.hasAttributes())
+        NewAttrs.push_back(AttributeWithIndex::get(AttrListPtr::FunctionIndex,
+                                                   Attr));
 
       // The trampoline may have been bitcast to a bogus type (FTy).
       // Handle this by synthesizing a new function type, equal to FTy
@@ -1348,7 +1356,7 @@ InstCombiner::transformCallThroughTrampoline(CallSite CS,
         NestF->getType() == PointerType::getUnqual(NewFTy) ?
         NestF : ConstantExpr::getBitCast(NestF,
                                          PointerType::getUnqual(NewFTy));
-      const AttrListPtr &NewPAL = AttrListPtr::get(NewAttrs);
+      const AttrListPtr &NewPAL = AttrListPtr::get(FTy->getContext(), NewAttrs);
 
       Instruction *NewCaller;
       if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
diff --git a/lib/Transforms/InstCombine/InstCombineCasts.cpp b/lib/Transforms/InstCombine/InstCombineCasts.cpp
index 555b4428d2e8f..bb59db8e7ba17 100644
--- a/lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -13,7 +13,7 @@
 
 #include "InstCombine.h"
 #include "llvm/Analysis/ConstantFolding.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Support/PatternMatch.h"
 using namespace llvm;
@@ -78,7 +78,7 @@ static Value *DecomposeSimpleLinearExpr(Value *Val, unsigned &Scale,
 /// try to eliminate the cast by moving the type information into the alloc.
 Instruction *InstCombiner::PromoteCastOfAllocation(BitCastInst &CI,
                                                    AllocaInst &AI) {
-  // This requires TargetData to get the alloca alignment and size information.
+  // This requires DataLayout to get the alloca alignment and size information.
   if (!TD) return 0;
 
   PointerType *PTy = cast<PointerType>(CI.getType());
@@ -229,7 +229,7 @@ isEliminableCastPair(
   const CastInst *CI, ///< The first cast instruction
   unsigned opcode,       ///< The opcode of the second cast instruction
   Type *DstTy,     ///< The target type for the second cast instruction
-  TargetData *TD         ///< The target data for pointer size
+  DataLayout *TD         ///< The target data for pointer size
 ) {
 
   Type *SrcTy = CI->getOperand(0)->getType();   // A from above
@@ -238,17 +238,20 @@ isEliminableCastPair(
   // Get the opcodes of the two Cast instructions
   Instruction::CastOps firstOp = Instruction::CastOps(CI->getOpcode());
   Instruction::CastOps secondOp = Instruction::CastOps(opcode);
-
+  Type *SrcIntPtrTy = TD && SrcTy->isPtrOrPtrVectorTy() ?
+    TD->getIntPtrType(SrcTy) : 0;
+  Type *MidIntPtrTy = TD && MidTy->isPtrOrPtrVectorTy() ?
+    TD->getIntPtrType(MidTy) : 0;
+  Type *DstIntPtrTy = TD && DstTy->isPtrOrPtrVectorTy() ?
+    TD->getIntPtrType(DstTy) : 0;
   unsigned Res = CastInst::isEliminableCastPair(firstOp, secondOp, SrcTy, MidTy,
-                                                DstTy,
-                                  TD ? TD->getIntPtrType(CI->getContext()) : 0);
-  
+                                                DstTy, SrcIntPtrTy, MidIntPtrTy,
+                                                DstIntPtrTy);
+
   // We don't want to form an inttoptr or ptrtoint that converts to an integer
   // type that differs from the pointer size.
-  if ((Res == Instruction::IntToPtr &&
-          (!TD || SrcTy != TD->getIntPtrType(CI->getContext()))) ||
-      (Res == Instruction::PtrToInt &&
-          (!TD || DstTy != TD->getIntPtrType(CI->getContext()))))
+  if ((Res == Instruction::IntToPtr && SrcTy != DstIntPtrTy) ||
+      (Res == Instruction::PtrToInt && DstTy != SrcIntPtrTy))
     Res = 0;
   
   return Instruction::CastOps(Res);
diff --git a/lib/Transforms/InstCombine/InstCombineCompares.cpp b/lib/Transforms/InstCombine/InstCombineCompares.cpp
index bdd310e97f6c2..7c3f8fe15d307 100644
--- a/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -16,7 +16,8 @@
 #include "llvm/Analysis/ConstantFolding.h"
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/Analysis/MemoryBuiltins.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
+#include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Support/ConstantRange.h"
 #include "llvm/Support/GetElementPtrTypeIterator.h"
 #include "llvm/Support/PatternMatch.h"
@@ -473,7 +474,7 @@ FoldCmpLoadFromIndexedGlobal(GetElementPtrInst *GEP, GlobalVariable *GV,
 /// If we can't emit an optimized form for this expression, this returns null.
 ///
 static Value *EvaluateGEPOffsetExpression(User *GEP, InstCombiner &IC) {
-  TargetData &TD = *IC.getTargetData();
+  DataLayout &TD = *IC.getDataLayout();
   gep_type_iterator GTI = gep_type_begin(GEP);
 
   // Check to see if this gep only has a single variable index.  If so, and if
@@ -2355,8 +2356,25 @@ Instruction *InstCombiner::visitICmpInst(ICmpInst &I) {
         // Try not to increase register pressure.
         BO0->hasOneUse() && BO1->hasOneUse()) {
       // Determine Y and Z in the form icmp (X+Y), (X+Z).
-      Value *Y = (A == C || A == D) ? B : A;
-      Value *Z = (C == A || C == B) ? D : C;
+      Value *Y, *Z;
+      if (A == C) {
+        // C + B == C + D  ->  B == D
+        Y = B;
+        Z = D;
+      } else if (A == D) {
+        // D + B == C + D  ->  B == C
+        Y = B;
+        Z = C;
+      } else if (B == C) {
+        // A + C == C + D  ->  A == D
+        Y = A;
+        Z = D;
+      } else {
+        assert(B == D);
+        // A + D == C + D  ->  A == C
+        Y = A;
+        Z = C;
+      }
       return new ICmpInst(Pred, Y, Z);
     }
 
@@ -2894,10 +2912,6 @@ Instruction *InstCombiner::visitFCmpInst(FCmpInst &I) {
         if (!RHSF)
           break;
 
-        // We can't convert a PPC double double.
-        if (RHSF->getType()->isPPC_FP128Ty())
-          break;
-
         const fltSemantics *Sem;
         // FIXME: This shouldn't be here.
         if (LHSExt->getSrcTy()->isHalfTy())
@@ -2910,6 +2924,8 @@ Instruction *InstCombiner::visitFCmpInst(FCmpInst &I) {
           Sem = &APFloat::IEEEquad;
         else if (LHSExt->getSrcTy()->isX86_FP80Ty())
           Sem = &APFloat::x87DoubleExtended;
+        else if (LHSExt->getSrcTy()->isPPC_FP128Ty())
+          Sem = &APFloat::PPCDoubleDouble;
         else
           break;
 
@@ -2985,6 +3001,44 @@ Instruction *InstCombiner::visitFCmpInst(FCmpInst &I) {
                 return Res;
         }
         break;
+      case Instruction::Call: {
+        CallInst *CI = cast<CallInst>(LHSI);
+        LibFunc::Func Func;
+        // Various optimization for fabs compared with zero.
+        if (RHSC->isNullValue() && CI->getCalledFunction() &&
+            TLI->getLibFunc(CI->getCalledFunction()->getName(), Func) &&
+            TLI->has(Func)) {
+          if (Func == LibFunc::fabs || Func == LibFunc::fabsf ||
+              Func == LibFunc::fabsl) {
+            switch (I.getPredicate()) {
+            default: break;
+            // fabs(x) < 0 --> false
+            case FCmpInst::FCMP_OLT:
+              return ReplaceInstUsesWith(I, Builder->getFalse());
+            // fabs(x) > 0 --> x != 0
+            case FCmpInst::FCMP_OGT:
+              return new FCmpInst(FCmpInst::FCMP_ONE, CI->getArgOperand(0),
+                                  RHSC);
+            // fabs(x) <= 0 --> x == 0
+            case FCmpInst::FCMP_OLE:
+              return new FCmpInst(FCmpInst::FCMP_OEQ, CI->getArgOperand(0),
+                                  RHSC);
+            // fabs(x) >= 0 --> !isnan(x)
+            case FCmpInst::FCMP_OGE:
+              return new FCmpInst(FCmpInst::FCMP_ORD, CI->getArgOperand(0),
+                                  RHSC);
+            // fabs(x) == 0 --> x == 0
+            // fabs(x) != 0 --> x != 0
+            case FCmpInst::FCMP_OEQ:
+            case FCmpInst::FCMP_UEQ:
+            case FCmpInst::FCMP_ONE:
+            case FCmpInst::FCMP_UNE:
+              return new FCmpInst(I.getPredicate(), CI->getArgOperand(0),
+                                  RHSC);
+            }
+          }
+        }
+      }
       }
   }
 
diff --git a/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp b/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
index c485844aaeb45..4d106fc188534 100644
--- a/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
+++ b/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
@@ -14,13 +14,161 @@
 #include "InstCombine.h"
 #include "llvm/IntrinsicInst.h"
 #include "llvm/Analysis/Loads.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/ADT/Statistic.h"
 using namespace llvm;
 
-STATISTIC(NumDeadStore, "Number of dead stores eliminated");
+STATISTIC(NumDeadStore,    "Number of dead stores eliminated");
+STATISTIC(NumGlobalCopies, "Number of allocas copied from constant global");
+
+/// pointsToConstantGlobal - Return true if V (possibly indirectly) points to
+/// some part of a constant global variable.  This intentionally only accepts
+/// constant expressions because we can't rewrite arbitrary instructions.
+static bool pointsToConstantGlobal(Value *V) {
+  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
+    return GV->isConstant();
+  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
+    if (CE->getOpcode() == Instruction::BitCast ||
+        CE->getOpcode() == Instruction::GetElementPtr)
+      return pointsToConstantGlobal(CE->getOperand(0));
+  return false;
+}
+
+/// isOnlyCopiedFromConstantGlobal - Recursively walk the uses of a (derived)
+/// pointer to an alloca.  Ignore any reads of the pointer, return false if we
+/// see any stores or other unknown uses.  If we see pointer arithmetic, keep
+/// track of whether it moves the pointer (with IsOffset) but otherwise traverse
+/// the uses.  If we see a memcpy/memmove that targets an unoffseted pointer to
+/// the alloca, and if the source pointer is a pointer to a constant global, we
+/// can optimize this.
+static bool
+isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
+                               SmallVectorImpl<Instruction *> &ToDelete,
+                               bool IsOffset = false) {
+  // We track lifetime intrinsics as we encounter them.  If we decide to go
+  // ahead and replace the value with the global, this lets the caller quickly
+  // eliminate the markers.
+
+  for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI!=E; ++UI) {
+    User *U = cast<Instruction>(*UI);
+
+    if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
+      // Ignore non-volatile loads, they are always ok.
+      if (!LI->isSimple()) return false;
+      continue;
+    }
+
+    if (BitCastInst *BCI = dyn_cast<BitCastInst>(U)) {
+      // If uses of the bitcast are ok, we are ok.
+      if (!isOnlyCopiedFromConstantGlobal(BCI, TheCopy, ToDelete, IsOffset))
+        return false;
+      continue;
+    }
+    if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(U)) {
+      // If the GEP has all zero indices, it doesn't offset the pointer.  If it
+      // doesn't, it does.
+      if (!isOnlyCopiedFromConstantGlobal(GEP, TheCopy, ToDelete,
+                                          IsOffset || !GEP->hasAllZeroIndices()))
+        return false;
+      continue;
+    }
+
+    if (CallSite CS = U) {
+      // If this is the function being called then we treat it like a load and
+      // ignore it.
+      if (CS.isCallee(UI))
+        continue;
+
+      // If this is a readonly/readnone call site, then we know it is just a
+      // load (but one that potentially returns the value itself), so we can
+      // ignore it if we know that the value isn't captured.
+      unsigned ArgNo = CS.getArgumentNo(UI);
+      if (CS.onlyReadsMemory() &&
+          (CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
+        continue;
+
+      // If this is being passed as a byval argument, the caller is making a
+      // copy, so it is only a read of the alloca.
+      if (CS.isByValArgument(ArgNo))
+        continue;
+    }
+
+    // Lifetime intrinsics can be handled by the caller.
+    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U)) {
+      if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
+          II->getIntrinsicID() == Intrinsic::lifetime_end) {
+        assert(II->use_empty() && "Lifetime markers have no result to use!");
+        ToDelete.push_back(II);
+        continue;
+      }
+    }
+
+    // If this is isn't our memcpy/memmove, reject it as something we can't
+    // handle.
+    MemTransferInst *MI = dyn_cast<MemTransferInst>(U);
+    if (MI == 0)
+      return false;
+
+    // If the transfer is using the alloca as a source of the transfer, then
+    // ignore it since it is a load (unless the transfer is volatile).
+    if (UI.getOperandNo() == 1) {
+      if (MI->isVolatile()) return false;
+      continue;
+    }
+
+    // If we already have seen a copy, reject the second one.
+    if (TheCopy) return false;
+
+    // If the pointer has been offset from the start of the alloca, we can't
+    // safely handle this.
+    if (IsOffset) return false;
+
+    // If the memintrinsic isn't using the alloca as the dest, reject it.
+    if (UI.getOperandNo() != 0) return false;
+
+    // If the source of the memcpy/move is not a constant global, reject it.
+    if (!pointsToConstantGlobal(MI->getSource()))
+      return false;
+
+    // Otherwise, the transform is safe.  Remember the copy instruction.
+    TheCopy = MI;
+  }
+  return true;
+}
+
+/// isOnlyCopiedFromConstantGlobal - Return true if the specified alloca is only
+/// modified by a copy from a constant global.  If we can prove this, we can
+/// replace any uses of the alloca with uses of the global directly.
+static MemTransferInst *
+isOnlyCopiedFromConstantGlobal(AllocaInst *AI,
+                               SmallVectorImpl<Instruction *> &ToDelete) {
+  MemTransferInst *TheCopy = 0;
+  if (isOnlyCopiedFromConstantGlobal(AI, TheCopy, ToDelete))
+    return TheCopy;
+  return 0;
+}
+
+/// getPointeeAlignment - Compute the minimum alignment of the value pointed
+/// to by the given pointer.
+static unsigned getPointeeAlignment(Value *V, const DataLayout &TD) {
+  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
+    if (CE->getOpcode() == Instruction::BitCast ||
+        (CE->getOpcode() == Instruction::GetElementPtr &&
+         cast<GEPOperator>(CE)->hasAllZeroIndices()))
+      return getPointeeAlignment(CE->getOperand(0), TD);
+
+  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
+    if (!GV->isDeclaration())
+      return TD.getPreferredAlignment(GV);
+
+  if (PointerType *PT = dyn_cast<PointerType>(V->getType()))
+    if (PT->getElementType()->isSized())
+      return TD.getABITypeAlignment(PT->getElementType());
+
+  return 0;
+}
 
 Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
   // Ensure that the alloca array size argument has type intptr_t, so that
@@ -99,12 +247,16 @@ Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
           return &AI;
         }
 
+        // If the alignment of the entry block alloca is 0 (unspecified),
+        // assign it the preferred alignment.
+        if (EntryAI->getAlignment() == 0)
+          EntryAI->setAlignment(
+            TD->getPrefTypeAlignment(EntryAI->getAllocatedType()));
         // Replace this zero-sized alloca with the one at the start of the entry
         // block after ensuring that the address will be aligned enough for both
         // types.
-        unsigned MaxAlign =
-          std::max(TD->getPrefTypeAlignment(EntryAI->getAllocatedType()),
-                   TD->getPrefTypeAlignment(AI.getAllocatedType()));
+        unsigned MaxAlign = std::max(EntryAI->getAlignment(),
+                                     AI.getAlignment());
         EntryAI->setAlignment(MaxAlign);
         if (AI.getType() != EntryAI->getType())
           return new BitCastInst(EntryAI, AI.getType());
@@ -113,6 +265,31 @@ Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
     }
   }
 
+  if (TD) {
+    // Check to see if this allocation is only modified by a memcpy/memmove from
+    // a constant global whose alignment is equal to or exceeds that of the
+    // allocation.  If this is the case, we can change all users to use
+    // the constant global instead.  This is commonly produced by the CFE by
+    // constructs like "void foo() { int A[] = {1,2,3,4,5,6,7,8,9...}; }" if 'A'
+    // is only subsequently read.
+    SmallVector<Instruction *, 4> ToDelete;
+    if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
+      if (AI.getAlignment() <= getPointeeAlignment(Copy->getSource(), *TD)) {
+        DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n');
+        DEBUG(dbgs() << "  memcpy = " << *Copy << '\n');
+        for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)
+          EraseInstFromFunction(*ToDelete[i]);
+        Constant *TheSrc = cast<Constant>(Copy->getSource());
+        Instruction *NewI
+          = ReplaceInstUsesWith(AI, ConstantExpr::getBitCast(TheSrc,
+                                                             AI.getType()));
+        EraseInstFromFunction(*Copy);
+        ++NumGlobalCopies;
+        return NewI;
+      }
+    }
+  }
+
   // At last, use the generic allocation site handler to aggressively remove
   // unused allocas.
   return visitAllocSite(AI);
@@ -121,7 +298,7 @@ Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
 
 /// InstCombineLoadCast - Fold 'load (cast P)' -> cast (load P)' when possible.
 static Instruction *InstCombineLoadCast(InstCombiner &IC, LoadInst &LI,
-                                        const TargetData *TD) {
+                                        const DataLayout *TD) {
   User *CI = cast<User>(LI.getOperand(0));
   Value *CastOp = CI->getOperand(0);
 
@@ -151,14 +328,14 @@ static Instruction *InstCombineLoadCast(InstCombiner &IC, LoadInst &LI,
             SrcPTy = SrcTy->getElementType();
           }
 
-      if (IC.getTargetData() &&
+      if (IC.getDataLayout() &&
           (SrcPTy->isIntegerTy() || SrcPTy->isPointerTy() || 
             SrcPTy->isVectorTy()) &&
           // Do not allow turning this into a load of an integer, which is then
           // casted to a pointer, this pessimizes pointer analysis a lot.
           (SrcPTy->isPointerTy() == LI.getType()->isPointerTy()) &&
-          IC.getTargetData()->getTypeSizeInBits(SrcPTy) ==
-               IC.getTargetData()->getTypeSizeInBits(DestPTy)) {
+          IC.getDataLayout()->getTypeSizeInBits(SrcPTy) ==
+               IC.getDataLayout()->getTypeSizeInBits(DestPTy)) {
 
         // Okay, we are casting from one integer or pointer type to another of
         // the same size.  Instead of casting the pointer before the load, cast
@@ -336,11 +513,11 @@ static Instruction *InstCombineStoreToCast(InstCombiner &IC, StoreInst &SI) {
   
   // If the pointers point into different address spaces or if they point to
   // values with different sizes, we can't do the transformation.
-  if (!IC.getTargetData() ||
+  if (!IC.getDataLayout() ||
       SrcTy->getAddressSpace() != 
         cast<PointerType>(CI->getType())->getAddressSpace() ||
-      IC.getTargetData()->getTypeSizeInBits(SrcPTy) !=
-      IC.getTargetData()->getTypeSizeInBits(DestPTy))
+      IC.getDataLayout()->getTypeSizeInBits(SrcPTy) !=
+      IC.getDataLayout()->getTypeSizeInBits(DestPTy))
     return 0;
 
   // Okay, we are casting from one integer or pointer type to another of
diff --git a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
index 35a0bbb76146d..cefe45ec862ce 100644
--- a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
+++ b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
@@ -37,7 +37,7 @@ static Value *simplifyValueKnownNonZero(Value *V, InstCombiner &IC) {
   if (match(V, m_LShr(m_OneUse(m_Shl(m_Value(PowerOf2), m_Value(A))),
                       m_Value(B))) &&
       // The "1" can be any value known to be a power of 2.
-      isPowerOfTwo(PowerOf2, IC.getTargetData())) {
+      isPowerOfTwo(PowerOf2, IC.getDataLayout())) {
     A = IC.Builder->CreateSub(A, B);
     return IC.Builder->CreateShl(PowerOf2, A);
   }
@@ -46,7 +46,7 @@ static Value *simplifyValueKnownNonZero(Value *V, InstCombiner &IC) {
   // inexact.  Similarly for <<.
   if (BinaryOperator *I = dyn_cast<BinaryOperator>(V))
     if (I->isLogicalShift() &&
-        isPowerOfTwo(I->getOperand(0), IC.getTargetData())) {
+        isPowerOfTwo(I->getOperand(0), IC.getDataLayout())) {
       // We know that this is an exact/nuw shift and that the input is a
       // non-zero context as well.
       if (Value *V2 = simplifyValueKnownNonZero(I->getOperand(0), IC)) {
@@ -462,12 +462,23 @@ Instruction *InstCombiner::visitUDiv(BinaryOperator &I) {
     }
   }
 
+  // (x lshr C1) udiv C2 --> x udiv (C2 << C1)
+  if (ConstantInt *C2 = dyn_cast<ConstantInt>(Op1)) {
+    Value *X;
+    ConstantInt *C1;
+    if (match(Op0, m_LShr(m_Value(X), m_ConstantInt(C1)))) {
+      APInt NC = C2->getValue().shl(C1->getLimitedValue(C1->getBitWidth()-1));
+      return BinaryOperator::CreateUDiv(X, Builder->getInt(NC));
+    }
+  }
+
   // X udiv (C1 << N), where C1 is "1<<C2"  -->  X >> (N+C2)
   { const APInt *CI; Value *N;
     if (match(Op1, m_Shl(m_Power2(CI), m_Value(N))) ||
         match(Op1, m_ZExt(m_Shl(m_Power2(CI), m_Value(N))))) {
       if (*CI != 1)
-        N = Builder->CreateAdd(N, ConstantInt::get(I.getType(),CI->logBase2()));
+        N = Builder->CreateAdd(N,
+                               ConstantInt::get(N->getType(), CI->logBase2()));
       if (ZExtInst *Z = dyn_cast<ZExtInst>(Op1))
         N = Builder->CreateZExt(N, Z->getDestTy());
       if (I.isExact())
diff --git a/lib/Transforms/InstCombine/InstCombinePHI.cpp b/lib/Transforms/InstCombine/InstCombinePHI.cpp
index 664546c165511..de9c77e6005a4 100644
--- a/lib/Transforms/InstCombine/InstCombinePHI.cpp
+++ b/lib/Transforms/InstCombine/InstCombinePHI.cpp
@@ -13,7 +13,7 @@
 
 #include "InstCombine.h"
 #include "llvm/Analysis/InstructionSimplify.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/STLExtras.h"
 using namespace llvm;
diff --git a/lib/Transforms/InstCombine/InstCombineSelect.cpp b/lib/Transforms/InstCombine/InstCombineSelect.cpp
index 291e80019e8df..a2d4c888f2cf9 100644
--- a/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -287,7 +287,7 @@ Instruction *InstCombiner::FoldSelectIntoOp(SelectInst &SI, Value *TrueVal,
 /// SimplifyWithOpReplaced - See if V simplifies when its operand Op is
 /// replaced with RepOp.
 static Value *SimplifyWithOpReplaced(Value *V, Value *Op, Value *RepOp,
-                                     const TargetData *TD,
+                                     const DataLayout *TD,
                                      const TargetLibraryInfo *TLI) {
   // Trivial replacement.
   if (V == Op)
@@ -333,6 +333,10 @@ static Value *SimplifyWithOpReplaced(Value *V, Value *Op, Value *RepOp,
 
     // All operands were constants, fold it.
     if (ConstOps.size() == I->getNumOperands()) {
+      if (CmpInst *C = dyn_cast<CmpInst>(I))
+        return ConstantFoldCompareInstOperands(C->getPredicate(), ConstOps[0],
+                                               ConstOps[1], TD, TLI);
+
       if (LoadInst *LI = dyn_cast<LoadInst>(I))
         if (!LI->isVolatile())
           return ConstantFoldLoadFromConstPtr(ConstOps[0], TD);
@@ -903,7 +907,7 @@ Instruction *InstCombiner::visitSelectInst(SelectInst &SI) {
     return &SI;
   }
 
-  if (VectorType* VecTy = dyn_cast<VectorType>(SI.getType())) {
+  if (VectorType *VecTy = dyn_cast<VectorType>(SI.getType())) {
     unsigned VWidth = VecTy->getNumElements();
     APInt UndefElts(VWidth, 0);
     APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
@@ -912,6 +916,28 @@ Instruction *InstCombiner::visitSelectInst(SelectInst &SI) {
         return ReplaceInstUsesWith(SI, V);
       return &SI;
     }
+
+    if (ConstantVector *CV = dyn_cast<ConstantVector>(CondVal)) {
+      // Form a shufflevector instruction.
+      SmallVector<Constant *, 8> Mask(VWidth);
+      Type *Int32Ty = Type::getInt32Ty(CV->getContext());
+      for (unsigned i = 0; i != VWidth; ++i) {
+        Constant *Elem = cast<Constant>(CV->getOperand(i));
+        if (ConstantInt *E = dyn_cast<ConstantInt>(Elem))
+          Mask[i] = ConstantInt::get(Int32Ty, i + (E->isZero() ? VWidth : 0));
+        else if (isa<UndefValue>(Elem))
+          Mask[i] = UndefValue::get(Int32Ty);
+        else
+          return 0;
+      }
+      Constant *MaskVal = ConstantVector::get(Mask);
+      Value *V = Builder->CreateShuffleVector(TrueVal, FalseVal, MaskVal);
+      return ReplaceInstUsesWith(SI, V);
+    }
+
+    if (isa<ConstantAggregateZero>(CondVal)) {
+      return ReplaceInstUsesWith(SI, FalseVal);
+    }
   }
 
   return 0;
diff --git a/lib/Transforms/InstCombine/InstCombineShifts.cpp b/lib/Transforms/InstCombine/InstCombineShifts.cpp
index 4bb2403299ce8..57021f1bef84b 100644
--- a/lib/Transforms/InstCombine/InstCombineShifts.cpp
+++ b/lib/Transforms/InstCombine/InstCombineShifts.cpp
@@ -190,7 +190,7 @@ static Value *GetShiftedValue(Value *V, unsigned NumBits, bool isLeftShift,
       V = IC.Builder->CreateLShr(C, NumBits);
     // If we got a constantexpr back, try to simplify it with TD info.
     if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
-      V = ConstantFoldConstantExpression(CE, IC.getTargetData(),
+      V = ConstantFoldConstantExpression(CE, IC.getDataLayout(),
                                          IC.getTargetLibraryInfo());
     return V;
   }
diff --git a/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp b/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
index 54be8ed3fa90f..602b20337144e 100644
--- a/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
+++ b/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
@@ -14,7 +14,7 @@
 
 
 #include "InstCombine.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/IntrinsicInst.h"
 
 using namespace llvm;
diff --git a/lib/Transforms/InstCombine/InstCombineVectorOps.cpp b/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
index cf60f0f426dcb..dd7ea14e8a898 100644
--- a/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
+++ b/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
@@ -636,8 +636,11 @@ Instruction *InstCombiner::visitShuffleVectorInst(ShuffleVectorInst &SVI) {
 
       // If LHS's width is changed, shift the mask value accordingly.
       // If newRHS == NULL, i.e. LHSOp0 == RHSOp0, we want to remap any
-      // references to RHSOp0 to LHSOp0, so we don't need to shift the mask.
-      if (eltMask >= 0 && newRHS != NULL)
+      // references from RHSOp0 to LHSOp0, so we don't need to shift the mask.
+      // If newRHS == newLHS, we want to remap any references from newRHS to
+      // newLHS so that we can properly identify splats that may occur due to
+      // obfuscation accross the two vectors.
+      if (eltMask >= 0 && newRHS != NULL && newLHS != newRHS)
         eltMask += newLHSWidth;
     }
 
diff --git a/lib/Transforms/InstCombine/InstCombineWorklist.h b/lib/Transforms/InstCombine/InstCombineWorklist.h
index 99a02fc0df3f6..ea654ae9ed0a8 100644
--- a/lib/Transforms/InstCombine/InstCombineWorklist.h
+++ b/lib/Transforms/InstCombine/InstCombineWorklist.h
@@ -26,8 +26,8 @@ class LLVM_LIBRARY_VISIBILITY InstCombineWorklist {
   SmallVector<Instruction*, 256> Worklist;
   DenseMap<Instruction*, unsigned> WorklistMap;
   
-  void operator=(const InstCombineWorklist&RHS);   // DO NOT IMPLEMENT
-  InstCombineWorklist(const InstCombineWorklist&); // DO NOT IMPLEMENT
+  void operator=(const InstCombineWorklist&RHS) LLVM_DELETED_FUNCTION;
+  InstCombineWorklist(const InstCombineWorklist&) LLVM_DELETED_FUNCTION;
 public:
   InstCombineWorklist() {}
   
diff --git a/lib/Transforms/InstCombine/InstructionCombining.cpp b/lib/Transforms/InstCombine/InstructionCombining.cpp
index 68ecd516049de..9a46f25e66ff8 100644
--- a/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ b/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -40,7 +40,7 @@
 #include "llvm/Analysis/ConstantFolding.h"
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/Analysis/MemoryBuiltins.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Support/CFG.h"
@@ -88,7 +88,7 @@ void InstCombiner::getAnalysisUsage(AnalysisUsage &AU) const {
 
 
 Value *InstCombiner::EmitGEPOffset(User *GEP) {
-  return llvm::EmitGEPOffset(Builder, *getTargetData(), GEP);
+  return llvm::EmitGEPOffset(Builder, *getDataLayout(), GEP);
 }
 
 /// ShouldChangeType - Return true if it is desirable to convert a computation
@@ -805,6 +805,244 @@ static bool shouldMergeGEPs(GEPOperator &GEP, GEPOperator &Src) {
   return true;
 }
 
+/// Descale - Return a value X such that Val = X * Scale, or null if none.  If
+/// the multiplication is known not to overflow then NoSignedWrap is set.
+Value *InstCombiner::Descale(Value *Val, APInt Scale, bool &NoSignedWrap) {
+  assert(isa<IntegerType>(Val->getType()) && "Can only descale integers!");
+  assert(cast<IntegerType>(Val->getType())->getBitWidth() ==
+         Scale.getBitWidth() && "Scale not compatible with value!");
+
+  // If Val is zero or Scale is one then Val = Val * Scale.
+  if (match(Val, m_Zero()) || Scale == 1) {
+    NoSignedWrap = true;
+    return Val;
+  }
+
+  // If Scale is zero then it does not divide Val.
+  if (Scale.isMinValue())
+    return 0;
+
+  // Look through chains of multiplications, searching for a constant that is
+  // divisible by Scale.  For example, descaling X*(Y*(Z*4)) by a factor of 4
+  // will find the constant factor 4 and produce X*(Y*Z).  Descaling X*(Y*8) by
+  // a factor of 4 will produce X*(Y*2).  The principle of operation is to bore
+  // down from Val:
+  //
+  //     Val = M1 * X          ||   Analysis starts here and works down
+  //      M1 = M2 * Y          ||   Doesn't descend into terms with more
+  //      M2 =  Z * 4          \/   than one use
+  //
+  // Then to modify a term at the bottom:
+  //
+  //     Val = M1 * X
+  //      M1 =  Z * Y          ||   Replaced M2 with Z
+  //
+  // Then to work back up correcting nsw flags.
+
+  // Op - the term we are currently analyzing.  Starts at Val then drills down.
+  // Replaced with its descaled value before exiting from the drill down loop.
+  Value *Op = Val;
+
+  // Parent - initially null, but after drilling down notes where Op came from.
+  // In the example above, Parent is (Val, 0) when Op is M1, because M1 is the
+  // 0'th operand of Val.
+  std::pair<Instruction*, unsigned> Parent;
+
+  // RequireNoSignedWrap - Set if the transform requires a descaling at deeper
+  // levels that doesn't overflow.
+  bool RequireNoSignedWrap = false;
+
+  // logScale - log base 2 of the scale.  Negative if not a power of 2.
+  int32_t logScale = Scale.exactLogBase2();
+
+  for (;; Op = Parent.first->getOperand(Parent.second)) { // Drill down
+
+    if (ConstantInt *CI = dyn_cast<ConstantInt>(Op)) {
+      // If Op is a constant divisible by Scale then descale to the quotient.
+      APInt Quotient(Scale), Remainder(Scale); // Init ensures right bitwidth.
+      APInt::sdivrem(CI->getValue(), Scale, Quotient, Remainder);
+      if (!Remainder.isMinValue())
+        // Not divisible by Scale.
+        return 0;
+      // Replace with the quotient in the parent.
+      Op = ConstantInt::get(CI->getType(), Quotient);
+      NoSignedWrap = true;
+      break;
+    }
+
+    if (BinaryOperator *BO = dyn_cast<BinaryOperator>(Op)) {
+
+      if (BO->getOpcode() == Instruction::Mul) {
+        // Multiplication.
+        NoSignedWrap = BO->hasNoSignedWrap();
+        if (RequireNoSignedWrap && !NoSignedWrap)
+          return 0;
+
+        // There are three cases for multiplication: multiplication by exactly
+        // the scale, multiplication by a constant different to the scale, and
+        // multiplication by something else.
+        Value *LHS = BO->getOperand(0);
+        Value *RHS = BO->getOperand(1);
+
+        if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
+          // Multiplication by a constant.
+          if (CI->getValue() == Scale) {
+            // Multiplication by exactly the scale, replace the multiplication
+            // by its left-hand side in the parent.
+            Op = LHS;
+            break;
+          }
+
+          // Otherwise drill down into the constant.
+          if (!Op->hasOneUse())
+            return 0;
+
+          Parent = std::make_pair(BO, 1);
+          continue;
+        }
+
+        // Multiplication by something else. Drill down into the left-hand side
+        // since that's where the reassociate pass puts the good stuff.
+        if (!Op->hasOneUse())
+          return 0;
+
+        Parent = std::make_pair(BO, 0);
+        continue;
+      }
+
+      if (logScale > 0 && BO->getOpcode() == Instruction::Shl &&
+          isa<ConstantInt>(BO->getOperand(1))) {
+        // Multiplication by a power of 2.
+        NoSignedWrap = BO->hasNoSignedWrap();
+        if (RequireNoSignedWrap && !NoSignedWrap)
+          return 0;
+
+        Value *LHS = BO->getOperand(0);
+        int32_t Amt = cast<ConstantInt>(BO->getOperand(1))->
+          getLimitedValue(Scale.getBitWidth());
+        // Op = LHS << Amt.
+
+        if (Amt == logScale) {
+          // Multiplication by exactly the scale, replace the multiplication
+          // by its left-hand side in the parent.
+          Op = LHS;
+          break;
+        }
+        if (Amt < logScale || !Op->hasOneUse())
+          return 0;
+
+        // Multiplication by more than the scale.  Reduce the multiplying amount
+        // by the scale in the parent.
+        Parent = std::make_pair(BO, 1);
+        Op = ConstantInt::get(BO->getType(), Amt - logScale);
+        break;
+      }
+    }
+
+    if (!Op->hasOneUse())
+      return 0;
+
+    if (CastInst *Cast = dyn_cast<CastInst>(Op)) {
+      if (Cast->getOpcode() == Instruction::SExt) {
+        // Op is sign-extended from a smaller type, descale in the smaller type.
+        unsigned SmallSize = Cast->getSrcTy()->getPrimitiveSizeInBits();
+        APInt SmallScale = Scale.trunc(SmallSize);
+        // Suppose Op = sext X, and we descale X as Y * SmallScale.  We want to
+        // descale Op as (sext Y) * Scale.  In order to have
+        //   sext (Y * SmallScale) = (sext Y) * Scale
+        // some conditions need to hold however: SmallScale must sign-extend to
+        // Scale and the multiplication Y * SmallScale should not overflow.
+        if (SmallScale.sext(Scale.getBitWidth()) != Scale)
+          // SmallScale does not sign-extend to Scale.
+          return 0;
+        assert(SmallScale.exactLogBase2() == logScale);
+        // Require that Y * SmallScale must not overflow.
+        RequireNoSignedWrap = true;
+
+        // Drill down through the cast.
+        Parent = std::make_pair(Cast, 0);
+        Scale = SmallScale;
+        continue;
+      }
+
+      if (Cast->getOpcode() == Instruction::Trunc) {
+        // Op is truncated from a larger type, descale in the larger type.
+        // Suppose Op = trunc X, and we descale X as Y * sext Scale.  Then
+        //   trunc (Y * sext Scale) = (trunc Y) * Scale
+        // always holds.  However (trunc Y) * Scale may overflow even if
+        // trunc (Y * sext Scale) does not, so nsw flags need to be cleared
+        // from this point up in the expression (see later).
+        if (RequireNoSignedWrap)
+          return 0;
+
+        // Drill down through the cast.
+        unsigned LargeSize = Cast->getSrcTy()->getPrimitiveSizeInBits();
+        Parent = std::make_pair(Cast, 0);
+        Scale = Scale.sext(LargeSize);
+        if (logScale + 1 == (int32_t)Cast->getType()->getPrimitiveSizeInBits())
+          logScale = -1;
+        assert(Scale.exactLogBase2() == logScale);
+        continue;
+      }
+    }
+
+    // Unsupported expression, bail out.
+    return 0;
+  }
+
+  // We know that we can successfully descale, so from here on we can safely
+  // modify the IR.  Op holds the descaled version of the deepest term in the
+  // expression.  NoSignedWrap is 'true' if multiplying Op by Scale is known
+  // not to overflow.
+
+  if (!Parent.first)
+    // The expression only had one term.
+    return Op;
+
+  // Rewrite the parent using the descaled version of its operand.
+  assert(Parent.first->hasOneUse() && "Drilled down when more than one use!");
+  assert(Op != Parent.first->getOperand(Parent.second) &&
+         "Descaling was a no-op?");
+  Parent.first->setOperand(Parent.second, Op);
+  Worklist.Add(Parent.first);
+
+  // Now work back up the expression correcting nsw flags.  The logic is based
+  // on the following observation: if X * Y is known not to overflow as a signed
+  // multiplication, and Y is replaced by a value Z with smaller absolute value,
+  // then X * Z will not overflow as a signed multiplication either.  As we work
+  // our way up, having NoSignedWrap 'true' means that the descaled value at the
+  // current level has strictly smaller absolute value than the original.
+  Instruction *Ancestor = Parent.first;
+  do {
+    if (BinaryOperator *BO = dyn_cast<BinaryOperator>(Ancestor)) {
+      // If the multiplication wasn't nsw then we can't say anything about the
+      // value of the descaled multiplication, and we have to clear nsw flags
+      // from this point on up.
+      bool OpNoSignedWrap = BO->hasNoSignedWrap();
+      NoSignedWrap &= OpNoSignedWrap;
+      if (NoSignedWrap != OpNoSignedWrap) {
+        BO->setHasNoSignedWrap(NoSignedWrap);
+        Worklist.Add(Ancestor);
+      }
+    } else if (Ancestor->getOpcode() == Instruction::Trunc) {
+      // The fact that the descaled input to the trunc has smaller absolute
+      // value than the original input doesn't tell us anything useful about
+      // the absolute values of the truncations.
+      NoSignedWrap = false;
+    }
+    assert((Ancestor->getOpcode() != Instruction::SExt || NoSignedWrap) &&
+           "Failed to keep proper track of nsw flags while drilling down?");
+
+    if (Ancestor == Val)
+      // Got to the top, all done!
+      return Val;
+
+    // Move up one level in the expression.
+    assert(Ancestor->hasOneUse() && "Drilled down when more than one use!");
+    Ancestor = Ancestor->use_back();
+  } while (1);
+}
+
 Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
   SmallVector<Value*, 8> Ops(GEP.op_begin(), GEP.op_end());
 
@@ -817,7 +1055,7 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
   // by multiples of a zero size type with zero.
   if (TD) {
     bool MadeChange = false;
-    Type *IntPtrTy = TD->getIntPtrType(GEP.getContext());
+    Type *IntPtrTy = TD->getIntPtrType(GEP.getPointerOperandType());
 
     gep_type_iterator GTI = gep_type_begin(GEP);
     for (User::op_iterator I = GEP.op_begin() + 1, E = GEP.op_end();
@@ -836,7 +1074,7 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
         }
 
       Type *IndexTy = (*I)->getType();
-      if (IndexTy != IntPtrTy && !IndexTy->isVectorTy()) {
+      if (IndexTy != IntPtrTy) {
         // If we are using a wider index than needed for this platform, shrink
         // it to what we need.  If narrower, sign-extend it to what we need.
         // This explicit cast can make subsequent optimizations more obvious.
@@ -855,7 +1093,7 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
     if (!shouldMergeGEPs(*cast<GEPOperator>(&GEP), *Src))
       return 0;
 
-    // Note that if our source is a gep chain itself that we wait for that
+    // Note that if our source is a gep chain itself then we wait for that
     // chain to be resolved before we perform this transformation.  This
     // avoids us creating a TON of code in some cases.
     if (GEPOperator *SrcGEP =
@@ -987,63 +1225,74 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
       }
 
       // Transform things like:
+      // %V = mul i64 %N, 4
+      // %t = getelementptr i8* bitcast (i32* %arr to i8*), i32 %V
+      // into:  %t1 = getelementptr i32* %arr, i32 %N; bitcast
+      if (TD && ResElTy->isSized() && SrcElTy->isSized()) {
+        // Check that changing the type amounts to dividing the index by a scale
+        // factor.
+        uint64_t ResSize = TD->getTypeAllocSize(ResElTy);
+        uint64_t SrcSize = TD->getTypeAllocSize(SrcElTy);
+        if (ResSize && SrcSize % ResSize == 0) {
+          Value *Idx = GEP.getOperand(1);
+          unsigned BitWidth = Idx->getType()->getPrimitiveSizeInBits();
+          uint64_t Scale = SrcSize / ResSize;
+
+          // Earlier transforms ensure that the index has type IntPtrType, which
+          // considerably simplifies the logic by eliminating implicit casts.
+          assert(Idx->getType() == TD->getIntPtrType(GEP.getContext()) &&
+                 "Index not cast to pointer width?");
+
+          bool NSW;
+          if (Value *NewIdx = Descale(Idx, APInt(BitWidth, Scale), NSW)) {
+            // Successfully decomposed Idx as NewIdx * Scale, form a new GEP.
+            // If the multiplication NewIdx * Scale may overflow then the new
+            // GEP may not be "inbounds".
+            Value *NewGEP = GEP.isInBounds() && NSW ?
+              Builder->CreateInBoundsGEP(StrippedPtr, NewIdx, GEP.getName()) :
+              Builder->CreateGEP(StrippedPtr, NewIdx, GEP.getName());
+            // The NewGEP must be pointer typed, so must the old one -> BitCast
+            return new BitCastInst(NewGEP, GEP.getType());
+          }
+        }
+      }
+
+      // Similarly, transform things like:
       // getelementptr i8* bitcast ([100 x double]* X to i8*), i32 %tmp
       //   (where tmp = 8*tmp2) into:
       // getelementptr [100 x double]* %arr, i32 0, i32 %tmp2; bitcast
-
-      if (TD && SrcElTy->isArrayTy() && ResElTy->isIntegerTy(8)) {
+      if (TD && ResElTy->isSized() && SrcElTy->isSized() &&
+          SrcElTy->isArrayTy()) {
+        // Check that changing to the array element type amounts to dividing the
+        // index by a scale factor.
+        uint64_t ResSize = TD->getTypeAllocSize(ResElTy);
         uint64_t ArrayEltSize =
-            TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType());
-
-        // Check to see if "tmp" is a scale by a multiple of ArrayEltSize.  We
-        // allow either a mul, shift, or constant here.
-        Value *NewIdx = 0;
-        ConstantInt *Scale = 0;
-        if (ArrayEltSize == 1) {
-          NewIdx = GEP.getOperand(1);
-          Scale = ConstantInt::get(cast<IntegerType>(NewIdx->getType()), 1);
-        } else if (ConstantInt *CI = dyn_cast<ConstantInt>(GEP.getOperand(1))) {
-          NewIdx = ConstantInt::get(CI->getType(), 1);
-          Scale = CI;
-        } else if (Instruction *Inst =dyn_cast<Instruction>(GEP.getOperand(1))){
-          if (Inst->getOpcode() == Instruction::Shl &&
-              isa<ConstantInt>(Inst->getOperand(1))) {
-            ConstantInt *ShAmt = cast<ConstantInt>(Inst->getOperand(1));
-            uint32_t ShAmtVal = ShAmt->getLimitedValue(64);
-            Scale = ConstantInt::get(cast<IntegerType>(Inst->getType()),
-                                     1ULL << ShAmtVal);
-            NewIdx = Inst->getOperand(0);
-          } else if (Inst->getOpcode() == Instruction::Mul &&
-                     isa<ConstantInt>(Inst->getOperand(1))) {
-            Scale = cast<ConstantInt>(Inst->getOperand(1));
-            NewIdx = Inst->getOperand(0);
+          TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType());
+        if (ResSize && ArrayEltSize % ResSize == 0) {
+          Value *Idx = GEP.getOperand(1);
+          unsigned BitWidth = Idx->getType()->getPrimitiveSizeInBits();
+          uint64_t Scale = ArrayEltSize / ResSize;
+
+          // Earlier transforms ensure that the index has type IntPtrType, which
+          // considerably simplifies the logic by eliminating implicit casts.
+          assert(Idx->getType() == TD->getIntPtrType(GEP.getContext()) &&
+                 "Index not cast to pointer width?");
+
+          bool NSW;
+          if (Value *NewIdx = Descale(Idx, APInt(BitWidth, Scale), NSW)) {
+            // Successfully decomposed Idx as NewIdx * Scale, form a new GEP.
+            // If the multiplication NewIdx * Scale may overflow then the new
+            // GEP may not be "inbounds".
+            Value *Off[2];
+            Off[0] = Constant::getNullValue(Type::getInt32Ty(GEP.getContext()));
+            Off[1] = NewIdx;
+            Value *NewGEP = GEP.isInBounds() && NSW ?
+              Builder->CreateInBoundsGEP(StrippedPtr, Off, GEP.getName()) :
+              Builder->CreateGEP(StrippedPtr, Off, GEP.getName());
+            // The NewGEP must be pointer typed, so must the old one -> BitCast
+            return new BitCastInst(NewGEP, GEP.getType());
           }
         }
-
-        // If the index will be to exactly the right offset with the scale taken
-        // out, perform the transformation. Note, we don't know whether Scale is
-        // signed or not. We'll use unsigned version of division/modulo
-        // operation after making sure Scale doesn't have the sign bit set.
-        if (ArrayEltSize && Scale && Scale->getSExtValue() >= 0LL &&
-            Scale->getZExtValue() % ArrayEltSize == 0) {
-          Scale = ConstantInt::get(Scale->getType(),
-                                   Scale->getZExtValue() / ArrayEltSize);
-          if (Scale->getZExtValue() != 1) {
-            Constant *C = ConstantExpr::getIntegerCast(Scale, NewIdx->getType(),
-                                                       false /*ZExt*/);
-            NewIdx = Builder->CreateMul(NewIdx, C, "idxscale");
-          }
-
-          // Insert the new GEP instruction.
-          Value *Idx[2];
-          Idx[0] = Constant::getNullValue(Type::getInt32Ty(GEP.getContext()));
-          Idx[1] = NewIdx;
-          Value *NewGEP = GEP.isInBounds() ?
-            Builder->CreateInBoundsGEP(StrippedPtr, Idx, GEP.getName()):
-            Builder->CreateGEP(StrippedPtr, Idx, GEP.getName());
-          // The NewGEP must be pointer typed, so must the old one -> BitCast
-          return new BitCastInst(NewGEP, GEP.getType());
-        }
       }
     }
   }
@@ -1068,7 +1317,7 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
         // If the bitcast is of an allocation, and the allocation will be
         // converted to match the type of the cast, don't touch this.
         if (isa<AllocaInst>(BCI->getOperand(0)) ||
-            isAllocationFn(BCI->getOperand(0))) {
+            isAllocationFn(BCI->getOperand(0), TLI)) {
           // See if the bitcast simplifies, if so, don't nuke this GEP yet.
           if (Instruction *I = visitBitCast(*BCI)) {
             if (I != BCI) {
@@ -1107,7 +1356,8 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
 
 
 static bool
-isAllocSiteRemovable(Instruction *AI, SmallVectorImpl<WeakVH> &Users) {
+isAllocSiteRemovable(Instruction *AI, SmallVectorImpl<WeakVH> &Users,
+                     const TargetLibraryInfo *TLI) {
   SmallVector<Instruction*, 4> Worklist;
   Worklist.push_back(AI);
 
@@ -1163,7 +1413,7 @@ isAllocSiteRemovable(Instruction *AI, SmallVectorImpl<WeakVH> &Users) {
           }
         }
 
-        if (isFreeCall(I)) {
+        if (isFreeCall(I, TLI)) {
           Users.push_back(I);
           continue;
         }
@@ -1188,7 +1438,7 @@ Instruction *InstCombiner::visitAllocSite(Instruction &MI) {
   // to null and free calls, delete the calls and replace the comparisons with
   // true or false as appropriate.
   SmallVector<WeakVH, 64> Users;
-  if (isAllocSiteRemovable(&MI, Users)) {
+  if (isAllocSiteRemovable(&MI, Users, TLI)) {
     for (unsigned i = 0, e = Users.size(); i != e; ++i) {
       Instruction *I = cast_or_null<Instruction>(&*Users[i]);
       if (!I) continue;
@@ -1853,7 +2103,7 @@ static bool TryToSinkInstruction(Instruction *I, BasicBlock *DestBlock) {
 static bool AddReachableCodeToWorklist(BasicBlock *BB,
                                        SmallPtrSet<BasicBlock*, 64> &Visited,
                                        InstCombiner &IC,
-                                       const TargetData *TD,
+                                       const DataLayout *TD,
                                        const TargetLibraryInfo *TLI) {
   bool MadeIRChange = false;
   SmallVector<BasicBlock*, 256> Worklist;
@@ -1872,7 +2122,7 @@ static bool AddReachableCodeToWorklist(BasicBlock *BB,
       Instruction *Inst = BBI++;
 
       // DCE instruction if trivially dead.
-      if (isInstructionTriviallyDead(Inst)) {
+      if (isInstructionTriviallyDead(Inst, TLI)) {
         ++NumDeadInst;
         DEBUG(errs() << "IC: DCE: " << *Inst << '\n');
         Inst->eraseFromParent();
@@ -2002,7 +2252,7 @@ bool InstCombiner::DoOneIteration(Function &F, unsigned Iteration) {
     if (I == 0) continue;  // skip null values.
 
     // Check to see if we can DCE the instruction.
-    if (isInstructionTriviallyDead(I)) {
+    if (isInstructionTriviallyDead(I, TLI)) {
       DEBUG(errs() << "IC: DCE: " << *I << '\n');
       EraseInstFromFunction(*I);
       ++NumDeadInst;
@@ -2102,7 +2352,7 @@ bool InstCombiner::DoOneIteration(Function &F, unsigned Iteration) {
 
         // If the instruction was modified, it's possible that it is now dead.
         // if so, remove it.
-        if (isInstructionTriviallyDead(I)) {
+        if (isInstructionTriviallyDead(I, TLI)) {
           EraseInstFromFunction(*I);
         } else {
           Worklist.Add(I);
@@ -2117,9 +2367,27 @@ bool InstCombiner::DoOneIteration(Function &F, unsigned Iteration) {
   return MadeIRChange;
 }
 
+namespace {
+class InstCombinerLibCallSimplifier : public LibCallSimplifier {
+  InstCombiner *IC;
+public:
+  InstCombinerLibCallSimplifier(const DataLayout *TD,
+                                const TargetLibraryInfo *TLI,
+                                InstCombiner *IC)
+    : LibCallSimplifier(TD, TLI) {
+    this->IC = IC;
+  }
+
+  /// replaceAllUsesWith - override so that instruction replacement
+  /// can be defined in terms of the instruction combiner framework.
+  virtual void replaceAllUsesWith(Instruction *I, Value *With) const {
+    IC->ReplaceInstUsesWith(*I, With);
+  }
+};
+}
 
 bool InstCombiner::runOnFunction(Function &F) {
-  TD = getAnalysisIfAvailable<TargetData>();
+  TD = getAnalysisIfAvailable<DataLayout>();
   TLI = &getAnalysis<TargetLibraryInfo>();
 
   /// Builder - This is an IRBuilder that automatically inserts new
@@ -2129,6 +2397,9 @@ bool InstCombiner::runOnFunction(Function &F) {
                InstCombineIRInserter(Worklist));
   Builder = &TheBuilder;
 
+  InstCombinerLibCallSimplifier TheSimplifier(TD, TLI, this);
+  Simplifier = &TheSimplifier;
+
   bool EverMadeChange = false;
 
   // Lower dbg.declare intrinsics otherwise their value may be clobbered