diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index 15c19177056..cc8db35262f 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -23,6 +23,7 @@
 #include "llvm/Intrinsics.h"
 #include "llvm/Support/CFG.h"
 #include "llvm/Type.h"
+#include "llvm/CodeGen/FunctionLoweringInfo.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
@@ -1351,7 +1352,7 @@ SDNode *X86DAGToDAGISel::SelectAtomicLoadAdd(SDNode *Node, EVT NVT) {
 
   bool isInc = false, isDec = false, isSub = false, isCN = false;
   ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Val);
-  if (CN) {
+  if (CN && CN->getSExtValue() == (int32_t)CN->getSExtValue()) {
     isCN = true;
     int64_t CNVal = CN->getSExtValue();
     if (CNVal == 1)
@@ -1371,6 +1372,7 @@ SDNode *X86DAGToDAGISel::SelectAtomicLoadAdd(SDNode *Node, EVT NVT) {
     Val = Val.getOperand(1);
   }
 
+  DebugLoc dl = Node->getDebugLoc();
   unsigned Opc = 0;
   switch (NVT.getSimpleVT().SimpleTy) {
   default: return 0;
@@ -1462,7 +1464,6 @@ SDNode *X86DAGToDAGISel::SelectAtomicLoadAdd(SDNode *Node, EVT NVT) {
     break;
   }
 
-  DebugLoc dl = Node->getDebugLoc();
   SDValue Undef = SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF,
                                                  dl, NVT), 0);
   MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
@@ -1579,7 +1580,7 @@ SDNode *X86DAGToDAGISel::SelectAtomicLoadArith(SDNode *Node, EVT NVT) {
   
   bool isCN = false;
   ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Val);
-  if (CN) {
+  if (CN && (int32_t)CN->getSExtValue() == CN->getSExtValue()) {
     isCN = true;
     Val = CurDAG->getTargetConstant(CN->getSExtValue(), NVT);
   }
diff --git a/test/CodeGen/X86/atomic-or.ll b/test/CodeGen/X86/atomic-or.ll
index cd622908aca..9db6f6f06dd 100644
--- a/test/CodeGen/X86/atomic-or.ll
+++ b/test/CodeGen/X86/atomic-or.ll
@@ -2,17 +2,35 @@
 
 ; rdar://9692967
 
-define void @do_the_sync(i64* %p, i32 %b) nounwind {
+define void @t1(i64* %p, i32 %b) nounwind {
 entry:
   %p.addr = alloca i64*, align 8
   store i64* %p, i64** %p.addr, align 8
   %tmp = load i64** %p.addr, align 8
   call void @llvm.memory.barrier(i1 true, i1 true, i1 true, i1 true, i1 true)
+; CHECK: t1:
+; CHECK: movl    $2147483648, %eax
 ; CHECK: lock
-; CHECK-NEXT: orq     $2147483648
+; CHECK-NEXT: orq %rax, (%rdi)
   %0 = call i64 @llvm.atomic.load.or.i64.p0i64(i64* %tmp, i64 2147483648)
   call void @llvm.memory.barrier(i1 true, i1 true, i1 true, i1 true, i1 true)
   ret void
 }
+
+define void @t2(i64* %p, i32 %b) nounwind {
+entry:
+  %p.addr = alloca i64*, align 8
+  store i64* %p, i64** %p.addr, align 8
+  %tmp = load i64** %p.addr, align 8
+  call void @llvm.memory.barrier(i1 true, i1 true, i1 true, i1 true, i1 true)
+; CHECK: t2:
+; CHECK-NOT: movl
+; CHECK: lock
+; CHECK-NEXT: orq $2147483644, (%rdi)
+  %0 = call i64 @llvm.atomic.load.or.i64.p0i64(i64* %tmp, i64 2147483644)
+  call void @llvm.memory.barrier(i1 true, i1 true, i1 true, i1 true, i1 true)
+  ret void
+}
+
 declare i64 @llvm.atomic.load.or.i64.p0i64(i64* nocapture, i64) nounwind
 declare void @llvm.memory.barrier(i1, i1, i1, i1, i1) nounwind