diff --git a/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index 68be3b7c77f..c02d4cec4cf 100644
--- a/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -692,15 +692,15 @@ bool MemCpyOpt::performCallSlotOptzn(Instruction *cpy,
   bool changedArgument = false;
   for (unsigned i = 0; i < CS.arg_size(); ++i)
     if (CS.getArgument(i)->stripPointerCasts() == cpySrc) {
-      if (cpySrc->getType() != cpyDest->getType())
-        cpyDest = CastInst::CreatePointerCast(cpyDest, cpySrc->getType(),
-                                              cpyDest->getName(), C);
+      Value *Dest = cpySrc->getType() == cpyDest->getType() ?  cpyDest
+        : CastInst::CreatePointerCast(cpyDest, cpySrc->getType(),
+                                      cpyDest->getName(), C);
       changedArgument = true;
-      if (CS.getArgument(i)->getType() == cpyDest->getType())
-        CS.setArgument(i, cpyDest);
+      if (CS.getArgument(i)->getType() == Dest->getType())
+        CS.setArgument(i, Dest);
       else
-        CS.setArgument(i, CastInst::CreatePointerCast(cpyDest,
-                          CS.getArgument(i)->getType(), cpyDest->getName(), C));
+        CS.setArgument(i, CastInst::CreatePointerCast(Dest,
+                          CS.getArgument(i)->getType(), Dest->getName(), C));
     }
 
   if (!changedArgument)
diff --git a/test/Transforms/MemCpyOpt/2008-03-13-ReturnSlotBitcast.ll b/test/Transforms/MemCpyOpt/2008-03-13-ReturnSlotBitcast.ll
index 61ba3c7e6cc..597b69dee3d 100644
--- a/test/Transforms/MemCpyOpt/2008-03-13-ReturnSlotBitcast.ll
+++ b/test/Transforms/MemCpyOpt/2008-03-13-ReturnSlotBitcast.ll
@@ -9,11 +9,11 @@ declare void @g(%a*)
 define float @f() {
 entry:
   %a_var = alloca %a
-  %b_var = alloca %b
+  %b_var = alloca %b, align 1
   call void @g(%a* %a_var)
   %a_i8 = bitcast %a* %a_var to i8*
   %b_i8 = bitcast %b* %b_var to i8*
-  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %b_i8, i8* %a_i8, i32 4, i32 4, i1 false)
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %b_i8, i8* %a_i8, i32 4, i32 1, i1 false)
   %tmp1 = getelementptr %b* %b_var, i32 0, i32 0
   %tmp2 = load float* %tmp1
   ret float %tmp2