3 files changed, 25 insertions, 0 deletions
diff --git a/lib/Analysis/ConstantFolding.cpp b/lib/Analysis/ConstantFolding.cpp
index df79849c3c..613804d6d2 100644
--- a/lib/Analysis/ConstantFolding.cpp
+++ b/lib/Analysis/ConstantFolding.cpp
@@ -51,6 +51,12 @@ static Constant *FoldBitCast(Constant *C, Type *DestTy,
   if (C->isAllOnesValue() && !DestTy->isX86_MMXTy())
     return Constant::getAllOnesValue(DestTy);
 
+  // Bitcast of Bitcast can be done using a single cast.
+  ConstantExpr *CE = dyn_cast<ConstantExpr>(C);
+  if (CE && CE->getOpcode() == Instruction::BitCast) {
+    return ConstantExpr::getBitCast(CE->getOperand(0), DestTy);
+  }
+
   // The code below only handles casts to vectors currently.
   VectorType *DestVTy = dyn_cast<VectorType>(DestTy);
   if (DestVTy == 0)
diff --git a/lib/Transforms/InstCombine/InstCombineCasts.cpp b/lib/Transforms/InstCombine/InstCombineCasts.cpp
index ba90bf6b5c..6c42c0c660 100644
--- a/lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -1659,6 +1659,11 @@ Instruction *InstCombiner::visitBitCast(BitCastInst &CI) {
   if (DestTy == Src->getType())
     return ReplaceInstUsesWith(CI, Src);
 
+  // Bitcasts are transitive.
+  if (BitCastInst* BSrc = dyn_cast<BitCastInst>(Src)) {
+    return CastInst::Create(Instruction::BitCast, BSrc->getOperand(0), DestTy);
+  }
+
   if (PointerType *DstPTy = dyn_cast<PointerType>(DestTy)) {
     PointerType *SrcPTy = cast<PointerType>(SrcTy);
     Type *DstElTy = DstPTy->getElementType();
diff --git a/test/Transforms/InstCombine/cast.ll b/test/Transforms/InstCombine/cast.ll
index f85636f8df..06cdae24b4 100644
--- a/test/Transforms/InstCombine/cast.ll
+++ b/test/Transforms/InstCombine/cast.ll
@@ -630,3 +630,17 @@ entry:
 ; CHECK: uitofp
 }
 
+define <4 x float> @test64(<4 x float> %c) nounwind {
+  %t0 = bitcast <4 x float> %c to <4 x i32>
+  %t1 = bitcast <4 x i32> %t0 to <2 x double>
+  %t2 = bitcast <2 x double> %t1 to <4 x float>
+  ret <4 x float> %t2
+; CHECK: @test64
+; CHECK-NEXT: ret <4 x float> %c
+}
+
+define float @test2c() {
+  ret float extractelement (<2 x float> bitcast (double bitcast (<2 x float> <float -1.000000e+00, float -1.000000e+00> to double) to <2 x float>), i32 0)
+; CHECK: @test2c
+; CHECK-NOT: extractelement
+}