Fix an instcombine abort on a scalar-to-vector bitcast. This fixes PR4487.

llvm-svn: 74646

2 files changed, 29 insertions, 1 deletions

diff --git a/llvm/lib/Transforms/Scalar/InstructionCombining.cpp b/llvm/lib/Transforms/Scalar/InstructionCombining.cpp
index 5bd17e0737b..16e5ce07c34 100644
--- a/llvm/lib/Transforms/Scalar/InstructionCombining.cpp
+++ b/llvm/lib/Transforms/Scalar/InstructionCombining.cpp
@@ -1085,8 +1085,22 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
     break;
   }
   case Instruction::BitCast:
-    if (!I->getOperand(0)->getType()->isInteger())
+    if (!I->getOperand(0)->getType()->isIntOrIntVector())
       return false;  // vector->int or fp->int?
+
+    if (const VectorType *DstVTy = dyn_cast<VectorType>(I->getType())) {
+      if (const VectorType *SrcVTy =
+            dyn_cast<VectorType>(I->getOperand(0)->getType())) {
+        if (DstVTy->getNumElements() != SrcVTy->getNumElements())
+          // Don't touch a bitcast between vectors of different element counts.
+          return false;
+      } else
+        // Don't touch a scalar-to-vector bitcast.
+        return false;
+    } else if (isa<VectorType>(I->getOperand(0)->getType()))
+      // Don't touch a vector-to-scalar bitcast.
+      return false;
+
     if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMask,
                              RHSKnownZero, RHSKnownOne, Depth+1))
       return I;
diff --git a/llvm/test/Transforms/InstCombine/bitcast-scalar-to-vector.ll b/llvm/test/Transforms/InstCombine/bitcast-scalar-to-vector.ll
new file mode 100644
index 00000000000..8695d1edbc1
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/bitcast-scalar-to-vector.ll
@@ -0,0 +1,14 @@
+; RUN: llvm-as < %s | opt -instcombine | llvm-dis | grep {ret i32 0}
+; PR4487
+
+; Bitcasts between vectors and scalars are valid, despite being ill-advised.
+
+define i32 @test(i64 %a) {
+bb20:
+        %t1 = bitcast i64 %a to <2 x i32>
+        %t2 = bitcast i64 %a to <2 x i32>
+        %t3 = xor <2 x i32> %t1, %t2
+        %t4 = extractelement <2 x i32> %t3, i32 0
+        ret i32 %t4
+}
+