Index: llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
===================================================================
--- llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -945,6 +945,13 @@
             C->getArgOperand(ArgI)->getType()->getPointerAddressSpace())
       return false;
 
+  // If alias scopes get expanded, bail
+  auto callScope = C->getMetadata(LLVMContext::MD_alias_scope);
+  auto cpyScope = C->getMetadata(LLVMContext::MD_alias_scope);
+  if (MDNode::getMostGenericAliasScope(callScope, cpyScope) != nullptr) {
+    return false;
+  }
+
   // All the checks have passed, so do the transformation.
   bool changedArgument = false;
   for (unsigned ArgI = 0; ArgI < C->arg_size(); ++ArgI)
Index: llvm/test/Transforms/MemCpyOpt/callslot_badaa.ll
===================================================================
--- /dev/null
+++ llvm/test/Transforms/MemCpyOpt/callslot_badaa.ll
@@ -0,0 +1,34 @@
+; RUN: opt < %s -S -memcpyopt | FileCheck %s
+
+; Make sure callslot optimization does not allow merging here because of alias.scope.
+; Merging here generates:
+;   call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 %dst, i8* align 8 %src, i64 1, i1 false), !alias.scope !1
+;   call void @llvm.lifetime.end.p0i8(i64 8, i8* nonnull %src), !noalias !0
+;   %ret_value = load i8, i8* %dst, align 1
+;   ...
+;   !0 = distinct !{!0}
+;   !1 = !{!0, !2}
+;   !2 = distinct !{!2}
+; Which is incorrect because the lifetime.end of %src will now "noalias" the above memcpy
+; allowing future optimizations to incorrectly utilize %src after its lifetime has ended.
+define i8 @test(i8 %input) {
+  %tmp = alloca i8
+  %dst = alloca i8
+  %src = alloca i8
+; CHECK:   call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 %tmp, i8* align 8 %src, i64 1, i1 false)
+; CHECK:   call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 %dst, i8* align 8 %tmp, i64 1, i1 false)
+  call void @llvm.lifetime.start.p0i8(i64 8, i8* nonnull %src), !noalias !1
+  store i8 %input, i8* %src
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 %tmp, i8* align 8 %src, i64 1, i1 false), !alias.scope !0
+  call void @llvm.lifetime.end.p0i8(i64 8, i8* nonnull %src), !noalias !1
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 %dst, i8* align 8 %tmp, i64 1, i1 false), !alias.scope !1
+  %ret_value = load i8, i8* %dst
+  ret i8 %ret_value
+}
+
+declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture)
+declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture)
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8*, i8*, i64, i1)
+
+!0 = !{!0}
+!1 = !{!1}