The element insertion code in scalar replacement doesn't handle incorrect

element types, even though the element extraction code does. It is surprising
that this bug has been here for so long. Fixes <rdar://problem/10318778>.

llvm-svn: 142740

2 files changed, 23 insertions, 2 deletions

diff --git a/llvm/lib/Transforms/Scalar/ScalarReplAggregates.cpp b/llvm/lib/Transforms/Scalar/ScalarReplAggregates.cpp
index c6d9123d661..b89f7306a28 100644
--- a/llvm/lib/Transforms/Scalar/ScalarReplAggregates.cpp
+++ b/llvm/lib/Transforms/Scalar/ScalarReplAggregates.cpp
@@ -806,8 +806,10 @@ ConvertScalar_InsertValue(Value *SV, Value *Old,
         return Builder.CreateBitCast(SV, AllocaType);
 
     // Must be an element insertion.
-    assert(SV->getType() == VTy->getElementType());
-    uint64_t EltSize = TD.getTypeAllocSizeInBits(VTy->getElementType());
+    Type *EltTy = VTy->getElementType();
+    if (SV->getType() != EltTy)
+      SV = Builder.CreateBitCast(SV, EltTy);
+    uint64_t EltSize = TD.getTypeAllocSizeInBits(EltTy);
     unsigned Elt = Offset/EltSize;
     return Builder.CreateInsertElement(Old, SV, Builder.getInt32(Elt));
   }
diff --git a/llvm/test/Transforms/ScalarRepl/2011-10-22-VectorCrash.ll b/llvm/test/Transforms/ScalarRepl/2011-10-22-VectorCrash.ll
new file mode 100644
index 00000000000..cd21ff5f0b5
--- /dev/null
+++ b/llvm/test/Transforms/ScalarRepl/2011-10-22-VectorCrash.ll
@@ -0,0 +1,19 @@
+; RUN: opt < %s -S -scalarrepl | FileCheck %s
+target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
+target triple = "thumbv7-apple-ios5.0.0"
+
+%union.anon = type { <4 x float> }
+
+; CHECK: @test
+; CHECK-NOT: alloca
+
+define void @test() nounwind {
+entry:
+  %u = alloca %union.anon, align 16
+  %u164 = bitcast %union.anon* %u to [4 x i32]*
+  %arrayidx165 = getelementptr inbounds [4 x i32]* %u164, i32 0, i32 0
+  store i32 undef, i32* %arrayidx165, align 4
+  %v186 = bitcast %union.anon* %u to <4 x float>*
+  store <4 x float> undef, <4 x float>* %v186, align 16
+  ret void
+}