Optimization: bitcast (<1 x ...> insertelement ..., X, ...) to ... ==> bitcast X...

llvm-svn: 174905

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
index 98fd05a..fbc259b 100644 (file)
--- a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -1738,11 +1738,22 @@ Instruction *InstCombiner::visitBitCast(BitCastInst &CI) {
   }
 
   if (VectorType *SrcVTy = dyn_cast<VectorType>(SrcTy)) {
-    if (SrcVTy->getNumElements() == 1 && !DestTy->isVectorTy()) {
-      Value *Elem =
-        Builder->CreateExtractElement(Src,
-                   Constant::getNullValue(Type::getInt32Ty(CI.getContext())));
-      return CastInst::Create(Instruction::BitCast, Elem, DestTy);
+    if (SrcVTy->getNumElements() == 1) {
+      // If our destination is not a vector, then make this a straight
+      // scalar-scalar cast.
+      if (!DestTy->isVectorTy()) {
+        Value *Elem =
+          Builder->CreateExtractElement(Src,
+                     Constant::getNullValue(Type::getInt32Ty(CI.getContext())));
+        return CastInst::Create(Instruction::BitCast, Elem, DestTy);
+      }
+
+      // Otherwise, see if our source is an insert. If so, then use the scalar
+      // component directly.
+      if (InsertElementInst *IEI =
+            dyn_cast<InsertElementInst>(CI.getOperand(0)))
+        return CastInst::Create(Instruction::BitCast, IEI->getOperand(1),
+                                DestTy);
     }
   }
 

diff --git a/llvm/test/Transforms/InstCombine/bitcast.ll b/llvm/test/Transforms/InstCombine/bitcast.ll
index c980b89..1e61132 100644 (file)
--- a/llvm/test/Transforms/InstCombine/bitcast.ll
+++ b/llvm/test/Transforms/InstCombine/bitcast.ll
@@ -137,4 +137,10 @@ define i32 @All111(i32 %in) {
 ; CHECK: ret i32 0
 }
 
-
+define <2 x i16> @BitcastInsert(i32 %a) {
+  %v = insertelement <1 x i32> undef, i32 %a, i32 0
+  %r = bitcast <1 x i32> %v to <2 x i16>
+  ret <2 x i16> %r
+; CHECK: @BitcastInsert
+; CHECK: bitcast i32 %a to <2 x i16>
+}