Split the init.trampoline intrinsic, which currently combines GCC's

[oota-llvm.git] / lib / Transforms / InstCombine / InstCombineVectorOps.cpp

diff --git a/lib/Transforms/InstCombine/InstCombineVectorOps.cpp b/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
index 4aa1328f134511dbdbd0357fcfba7848a6d38124..154267c03465cd310b63a1ef9a94ead5a04a28f8 100644 (file)
--- a/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
+++ b/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
@@ -77,7 +77,7 @@ static std::vector<int> getShuffleMask(const ShuffleVectorInst *SVI) {
 /// extracted from the vector.
 static Value *FindScalarElement(Value *V, unsigned EltNo) {
   assert(V->getType()->isVectorTy() && "Not looking at a vector?");
-  const VectorType *PTy = cast<VectorType>(V->getType());
+  VectorType *PTy = cast<VectorType>(V->getType());
   unsigned Width = PTy->getNumElements();
   if (EltNo >= Width)  // Out of range access.
     return UndefValue::get(PTy->getElementType());
@@ -175,7 +175,7 @@ Instruction *InstCombiner::visitExtractElementInst(ExtractElementInst &EI) {
     // the same number of elements, see if we can find the source element from
     // it.  In this case, we will end up needing to bitcast the scalars.
     if (BitCastInst *BCI = dyn_cast<BitCastInst>(EI.getOperand(0))) {
-      if (const VectorType *VT =
+      if (VectorType *VT =
           dyn_cast<VectorType>(BCI->getOperand(0)->getType()))
         if (VT->getNumElements() == VectorWidth)
           if (Value *Elt = FindScalarElement(BCI->getOperand(0), IndexVal))
@@ -225,13 +225,21 @@ Instruction *InstCombiner::visitExtractElementInst(ExtractElementInst &EI) {
           SrcIdx -= LHSWidth;
           Src = SVI->getOperand(1);
         }
-        const Type *Int32Ty = Type::getInt32Ty(EI.getContext());
+        Type *Int32Ty = Type::getInt32Ty(EI.getContext());
         return ExtractElementInst::Create(Src,
                                           ConstantInt::get(Int32Ty,
                                                            SrcIdx, false));
       }
+    } else if (CastInst *CI = dyn_cast<CastInst>(I)) {
+      // Canonicalize extractelement(cast) -> cast(extractelement)
+      // bitcasts can change the number of vector elements and they cost nothing
+      if (CI->hasOneUse() && EI.hasOneUse() &&
+          (CI->getOpcode() != Instruction::BitCast)) {
+        Value *EE = Builder->CreateExtractElement(CI->getOperand(0),
+                                                  EI.getIndexOperand());
+        return CastInst::Create(CI->getOpcode(), EE, EI.getType());
+      }
     }
-    // FIXME: Canonicalize extractelement(bitcast) -> bitcast(extractelement)
   }
   return 0;
 }
@@ -426,8 +434,11 @@ Instruction *InstCombiner::visitInsertElementInst(InsertElementInst &IE) {
   unsigned VWidth = cast<VectorType>(VecOp->getType())->getNumElements();
   APInt UndefElts(VWidth, 0);
   APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
-  if (SimplifyDemandedVectorElts(&IE, AllOnesEltMask, UndefElts))
+  if (Value *V = SimplifyDemandedVectorElts(&IE, AllOnesEltMask, UndefElts)) {
+    if (V != &IE)
+      return ReplaceInstUsesWith(IE, V);
     return &IE;
+  }
 
   return 0;
 }
@@ -451,7 +462,9 @@ Instruction *InstCombiner::visitShuffleVectorInst(ShuffleVectorInst &SVI) {
 
   APInt UndefElts(VWidth, 0);
   APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
-  if (SimplifyDemandedVectorElts(&SVI, AllOnesEltMask, UndefElts)) {
+  if (Value *V = SimplifyDemandedVectorElts(&SVI, AllOnesEltMask, UndefElts)) {
+    if (V != &SVI)
+      return ReplaceInstUsesWith(SVI, V);
     LHS = SVI.getOperand(0);
     RHS = SVI.getOperand(1);
     MadeChange = true;
@@ -542,7 +555,7 @@ Instruction *InstCombiner::visitShuffleVectorInst(ShuffleVectorInst &SVI) {
         // shuffle mask, do the replacement.
         if (isSplat || NewMask == LHSMask || NewMask == Mask) {
           std::vector<Constant*> Elts;
-          const Type *Int32Ty = Type::getInt32Ty(SVI.getContext());
+          Type *Int32Ty = Type::getInt32Ty(SVI.getContext());
           for (unsigned i = 0, e = NewMask.size(); i != e; ++i) {
             if (NewMask[i] < 0) {
               Elts.push_back(UndefValue::get(Int32Ty));