diff --git a/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index d2723ab342b..9a39c8fd29e 100644
--- a/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -615,10 +615,12 @@ bool MemCpyOpt::processMemCpy(MemCpyInst* M) {
   if (dep == MemoryDependenceAnalysis::None ||
       dep == MemoryDependenceAnalysis::NonLocal)
     return false;
-  else if (CallInst* C = dyn_cast<CallInst>(dep))
-    return performCallSlotOptzn(M, C);
-  else if (!isa<MemCpyInst>(dep))
-    return false;
+  else if (!isa<MemCpyInst>(dep)) {
+    if (CallInst* C = dyn_cast<CallInst>(dep))
+      return performCallSlotOptzn(M, C);
+    else
+      return false;
+  }
   
   MemCpyInst* MDep = cast<MemCpyInst>(dep);
   
diff --git a/test/Transforms/MemCpyOpt/2008-04-29-SRetRemoval.ll b/test/Transforms/MemCpyOpt/2008-04-29-SRetRemoval.ll
new file mode 100644
index 00000000000..1d9a35d63f2
--- /dev/null
+++ b/test/Transforms/MemCpyOpt/2008-04-29-SRetRemoval.ll
@@ -0,0 +1,17 @@
+; RUN: llvm-as < %s | opt -memcpyopt | llvm-dis | grep {call.*memcpy.*agg.result}
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
+target triple = "i386-apple-darwin8"
+@x = external global { x86_fp80, x86_fp80 }		; <{ x86_fp80, x86_fp80 }*> [#uses=1]
+
+define void @foo({ x86_fp80, x86_fp80 }* noalias sret %agg.result) nounwind  {
+entry:
+	%x.0 = alloca { x86_fp80, x86_fp80 }		; <{ x86_fp80, x86_fp80 }*> [#uses=1]
+	%x.01 = bitcast { x86_fp80, x86_fp80 }* %x.0 to i8*		; <i8*> [#uses=2]
+	call void @llvm.memcpy.i32( i8* %x.01, i8* bitcast ({ x86_fp80, x86_fp80 }* @x to i8*), i32 32, i32 16 )
+	%agg.result2 = bitcast { x86_fp80, x86_fp80 }* %agg.result to i8*		; <i8*> [#uses=1]
+	call void @llvm.memcpy.i32( i8* %agg.result2, i8* %x.01, i32 32, i32 16 )
+	ret void
+}
+
+declare void @llvm.memcpy.i32(i8*, i8*, i32, i32) nounwind