Index: llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp
===================================================================
--- llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -185,7 +185,7 @@
   // size. If so, check to see whether we will end up actually reducing the
   // number of stores used.
   unsigned Bytes = unsigned(End-Start);
-  unsigned MaxIntSize = DL.getLargestLegalIntTypeSize();
+  unsigned MaxIntSize = DL.getLargestLegalIntTypeSize() / 8;
   if (MaxIntSize == 0)
     MaxIntSize = 1;
   unsigned NumPointerStores = Bytes / MaxIntSize;
Index: llvm/trunk/test/Transforms/MemCpyOpt/profitable-memset.ll
===================================================================
--- llvm/trunk/test/Transforms/MemCpyOpt/profitable-memset.ll
+++ llvm/trunk/test/Transforms/MemCpyOpt/profitable-memset.ll
@@ -0,0 +1,20 @@
+; RUN: opt < %s -memcpyopt -S | FileCheck %s
+
+target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
+
+; CHECK-LABEL: @foo(
+; CHECK-NOT: store
+; CHECK: call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 8, i32 2, i1 false)
+
+define void @foo(i64* nocapture %P) {
+entry:
+  %0 = bitcast i64* %P to i16*
+  %arrayidx = getelementptr inbounds i16, i16* %0, i64 1
+  %1 = bitcast i16* %arrayidx to i32*
+  %arrayidx1 = getelementptr inbounds i16, i16* %0, i64 3
+  store i16 0, i16* %0, align 2
+  store i32 0, i32* %1, align 4
+  store i16 0, i16* %arrayidx1, align 2
+  ret void
+}
+