X86: rework expansion of atomic instructions

[t.p.northover]

Summary

Hi all,

This is a patch very much along the lines of the recent ARM & AArch64 refactoring of atomics: it moves the complex control-flow (& naive instruction selection) logic out of the MachineInstr level and into IR proper. It has many of the same benefits:

+ More efficient code since the ISel phase can see when constants are used (for example).
+ Neater code

That said, the primary purpose is to support 128-bit atomic operations on x86_64 (hence the added test). These benefits mostly come about because it's the easiest way to do so.

It would have been good to completely reuse the existing LoadLinked pass, instead of copy/paste/adapt, but unfortunately the constraints and optimal forms don't *quite* match up between x86 and LL/SC architectures, so I've added it as an x86-specific pass.

The other detail that may need explaining is the removal of NOCMOV checks in atomic-minmax-i6432.ll: it turns out that any CPU with cmpxchg8b will have cmov, and they can't be controlled independently in LLVM. Since this test is fundamentally about cmpxchg8b, I decided to simply remove those tests.

Do the changes look OK, and are people who more frequently fiddle around with x86 happy with the direction?

Cheers.

Tim.

[t.p.northover]

Ping?

Cheers.

Tim.

[t.p.northover]

Re-ping?

Cheers.

Tim.

[andreadb]

Hi Tim,

I really like this patch and in general it looks very good to me!

However, since I am not the code owner, you might want to ask for more feedback from Nadav or Jim.

Cheers,
Andrea

lib/Target/X86/X86AtomicExpandPass.cpp

108

Please remove the 'else' after return.

147

Please remove the 'else' after return.

[echristo]

One comment on the caching and use of the target machine, but otherwise looks ok here. I haven't reviewed the actual atomics aspect or the testcase changes yet.

lib/Target/X86/X86AtomicExpandPass.cpp

88

Is there any way to get this information from TTI instead? Relying on the subtarget here is going to go badly soon.

[t.p.northover]

Hi Eric,

+bool X86AtomicExpandPass::needsCmpXchgNb(llvm::Type *MemType) {
+ const X86Subtarget &Subtarget = TM->getSubtarget<X86Subtarget>();

+ if (!Subtarget.hasCmpxchg16b())

Is there any way to get this information from TTI instead? Relying on the subtarget here is going to go badly soon.

Should be fairly trivial, as long as a TTI can query the subtarget.
I'm fairly sure they can at the moment, but is that planned to go away
too? I'll get on it anyway.

Cheers.

Tim.

[chandlerc]

I have some small nits and one big question about ABI. Provided the answer to the question is "no" (IE, the ABI is exactly the same), then LGTM.

lib/Target/X86/X86AtomicExpandPass.cpp

46–47

I'd like some comment about why only AtomicRMW and Store need to be handled here...

83–86

This has the potential to be an ABI-breaking change. Does your new pass fall back to libcalls in exactly the same places that the old code did? Otherwise, depending on how the libctalls are implemented, this could cause a break with code compiled previously.

196–200

I feel like this big switch would be a touch cleaner hoisted into a static function returning the NewVal. Thoughts?

[t.p.northover]

Hi Chandler,

Thanks for looking at the patch!

This has the potential to be an ABI-breaking change. Does your new pass fall
back to libcalls in exactly the same places that the old code did? Otherwise,
depending on how the libctalls are implemented, this could cause a break with
code compiled previously.

That's a difficult one. It *does* change when we fall back, but
because of how it happens I think it's probably OK for a variety of
reasons:

1. The helpers aren't actually implemented on either Linux or OS X.

GCC already inlines __int128 atomic ops so it's arguably *fixing* an
ABI bug.

2. We already inline atomic_compare_exchange (on 128-bit types) to

cmpxchg16b when available; this expands the inlining to other
atomicrmw operations. Since these can be mixed, I think any ABI
implementation we're compatible with already has to implement these
calls in terms of cmpxchg16b rather than any global lock. So it
probably doesn't break currently unbroken ABIs.

3. The change is in the backend, and Clang has its own

MaxAtomicInlineWidth override. It seems a reasonable position that if
you're using LLVM's atomic instructions, you're requesting inlining if
possible.

So I think it's a positive change regardless. And if we do want to go
back to helpers, we should probably do it in Clang (I'd be happy to
get the patch together to do that if I've not managed to convince
you).

Meanwhile, I'll update the patch for your other comments.

Cheers.

Tim.

lib/Target/X86/CMakeLists.txt

 set(LLVM_TARGET_DEFINITIONS X86.td)
 
 tablegen(LLVM X86GenRegisterInfo.inc -gen-register-info)
 tablegen(LLVM X86GenDisassemblerTables.inc -gen-disassembler)
 tablegen(LLVM X86GenInstrInfo.inc -gen-instr-info)
 tablegen(LLVM X86GenAsmWriter.inc -gen-asm-writer)
 tablegen(LLVM X86GenAsmWriter1.inc -gen-asm-writer -asmwriternum=1)
 tablegen(LLVM X86GenAsmMatcher.inc -gen-asm-matcher)
 tablegen(LLVM X86GenDAGISel.inc -gen-dag-isel)
 tablegen(LLVM X86GenFastISel.inc -gen-fast-isel)
 tablegen(LLVM X86GenCallingConv.inc -gen-callingconv)
 tablegen(LLVM X86GenSubtargetInfo.inc -gen-subtarget)
 add_public_tablegen_target(X86CommonTableGen)
 
 set(sources
   X86AsmPrinter.cpp
+  X86AtomicExpandPass.cpp
   X86CodeEmitter.cpp
   X86FastISel.cpp
   X86FloatingPoint.cpp
   X86FrameLowering.cpp
   X86ISelDAGToDAG.cpp
   X86ISelLowering.cpp
   X86InstrInfo.cpp
   X86JITInfo.cpp

lib/Target/X86/X86.h

 
 namespace llvm {
 
 class FunctionPass;
 class ImmutablePass;
 class JITCodeEmitter;
 class X86TargetMachine;
 
+/// createX86AtomicExpandPass - This pass expands atomic operations that cannot
+/// be handled natively in terms of a loop using cmpxchg.
+FunctionPass *createX86AtomicExpandPass(const X86TargetMachine *TM);
+
 /// createX86ISelDag - This pass converts a legalized DAG into a
 /// X86-specific DAG, ready for instruction scheduling.
 ///
 FunctionPass *createX86ISelDag(X86TargetMachine &TM,
                                CodeGenOpt::Level OptLevel);
 
 /// createX86GlobalBaseRegPass - This pass initializes a global base
 /// register for PIC on x86-32.

lib/Target/X86/X86AtomicExpandPass.cpp

+//===-- X86AtomicExpandPass.cpp - Expand illegal atomic instructions --0---===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains a pass (at IR level) to replace atomic instructions which
+// cannot be implemented as a single instruction with cmpxchg-based loops.
+//
+//===----------------------------------------------------------------------===//
+
+#include "X86.h"
+#include "X86TargetMachine.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Module.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/TargetMachine.h"
+using namespace llvm;
+
+#define DEBUG_TYPE "x86-atomic-expand"
+
+namespace {
+  class X86AtomicExpandPass : public FunctionPass {
+    const X86TargetMachine *TM;
+  public:
+    static char ID; // Pass identification, replacement for typeid
+    explicit X86AtomicExpandPass(const X86TargetMachine *TM)
+      : FunctionPass(ID), TM(TM) {}
+
+    bool runOnFunction(Function &F) override;
+    bool expandAtomicInsts(Function &F);
+
+    bool needsCmpXchgNb(Type *MemType);
+    bool shouldExpandStore(StoreInst *SI);
+    bool shouldExpandAtomicRMW(AtomicRMWInst *AI);
+    bool shouldExpand(Instruction *Inst);
+
+    bool expandAtomicRMW(AtomicRMWInst *AI);
+    bool expandAtomicStore(StoreInst *SI);
+  };
+}
+
+char X86AtomicExpandPass::ID = 0;
+
+FunctionPass *llvm::createX86AtomicExpandPass(const X86TargetMachine *TM) {
+  return new X86AtomicExpandPass(TM);
+}
+
+bool X86AtomicExpandPass::runOnFunction(Function &F) {
+  SmallVector<Instruction *, 1> AtomicInsts;
+
+  // Changing control-flow while iterating through it is a bad idea, so gather a
+  // list of all atomic instructions before we start.
+  for (BasicBlock &BB : F)
+    for (Instruction &Inst : BB) {
+      if (isa<AtomicRMWInst>(&Inst) ||
+          (isa<StoreInst>(&Inst) && cast<StoreInst>(&Inst)->isAtomic()))
+        AtomicInsts.push_back(&Inst);
+    }
+
+  bool MadeChange = false;
+  for (Instruction *Inst : AtomicInsts) {
+    if (!shouldExpand(Inst))
+      continue;
+
+    if (AtomicRMWInst *AI = dyn_cast<AtomicRMWInst>(Inst))
+      MadeChange |= expandAtomicRMW(AI);
+    if (StoreInst *SI = dyn_cast<StoreInst>(Inst))
+      MadeChange |= expandAtomicStore(SI);
+  }
+
+  return MadeChange;
+}
+
+/// Returns true if operations on the given type will need to use either
+/// cmpxchg8b or cmpxchg16b. This occurs if the type is 1 step up from the
+/// native width, and the instructions are available (otherwise we leave them
+/// alone to become __sync_fetch_and_... calls).
+bool X86AtomicExpandPass::needsCmpXchgNb(llvm::Type *MemType) {
+  const X86Subtarget &Subtarget = TM->getSubtarget<X86Subtarget>();
+  if (!Subtarget.hasCmpxchg16b())
+    return false;
+
+  unsigned CmpXchgNbWidth = Subtarget.is64Bit() ? 128 : 64;
+
+  unsigned OpWidth = MemType->getPrimitiveSizeInBits();
+  if (OpWidth == CmpXchgNbWidth)
+    return true;
+
+  return false;
+}
+
+
+bool X86AtomicExpandPass::shouldExpandAtomicRMW(AtomicRMWInst *AI) {
+  const X86Subtarget &Subtarget = TM->getSubtarget<X86Subtarget>();
+  unsigned NativeWidth = Subtarget.is64Bit() ? 64 : 32;
+
+  if (needsCmpXchgNb(AI->getType()))
+    return true;
+  else if (AI->getType()->getPrimitiveSizeInBits() > NativeWidth)
+    return false;
+
+  AtomicRMWInst::BinOp Op = AI->getOperation();
+  switch (Op) {
+  default:
+    llvm_unreachable("Unknown atomic operation");
+  case AtomicRMWInst::Xchg:
+  case AtomicRMWInst::Add:
+  case AtomicRMWInst::Sub:
+    // It's better to use xadd, xsub or xchg for these in all cases.
+    return false;
+  case AtomicRMWInst::Or:
+  case AtomicRMWInst::And:
+  case AtomicRMWInst::Xor:
+    // If the atomicrmw's result isn't actually used, we can just add a "lock"
+    // prefix to a normal instruction for these operations.
+    return !AI->use_empty();
+  case AtomicRMWInst::Nand:
+  case AtomicRMWInst::Max:
+  case AtomicRMWInst::Min:
+  case AtomicRMWInst::UMax:
+  case AtomicRMWInst::UMin:
+    // These always require a non-trivial set of data operations on x86. We must
+    // use a cmpxchg loop.
+    return true;
+  }
+}
+
+bool X86AtomicExpandPass::shouldExpandStore(StoreInst *SI) {
+  if (needsCmpXchgNb(SI->getValueOperand()->getType()))
+    return true;
+
+  return false;
+}
+
+bool X86AtomicExpandPass::shouldExpand(Instruction *Inst) {
+  if (AtomicRMWInst *AI = dyn_cast<AtomicRMWInst>(Inst))
+    return shouldExpandAtomicRMW(AI);
+  else if (StoreInst *SI = dyn_cast<StoreInst>(Inst))
+    return shouldExpandStore(SI);
+  return false;
+}
+
+
+bool X86AtomicExpandPass::expandAtomicRMW(AtomicRMWInst *AI) {
+  AtomicOrdering Order =
+      AI->getOrdering() == Unordered ? Monotonic : AI->getOrdering();
+  Value *Addr = AI->getPointerOperand();
+  BasicBlock *BB = AI->getParent();
+  Function *F = BB->getParent();
+  LLVMContext &Ctx = F->getContext();
+
+  // Given: atomicrmw some_op iN* %addr, iN %incr ordering
+  //
+  // The standard expansion we produce is:
+  //     [...]
+  //     %init_loaded = load atomic iN* %addr
+  //     br label %loop
+  // loop:
+  //     %loaded = phi iN [ %init_loaded, %entry ], [ %new_loaded, %loop ]
+  //     %new = some_op iN %loaded, %incr
+  //     %pair = cmpxchg iN* %addr, iN %loaded, iN %new
+  //     %new_loaded = extractvalue { iN, i1 } %pair, 0
+  //     %success = extractvalue { iN, i1 } %pair, 1
+  //     br i1 %success, label %atomicrmw.end, label %loop
+  // atomicrmw.end:
+  //     [...]
+  BasicBlock *ExitBB = BB->splitBasicBlock(AI, "atomicrmw.end");
+  BasicBlock *LoopBB =  BasicBlock::Create(Ctx, "atomicrmw.start", F, ExitBB);
+
+  // This grabs the DebugLoc from AI.
+  IRBuilder<> Builder(AI);
+
+  // The split call above "helpfully" added a branch at the end of BB (to the
+  // wrong place), but we want a load. It's easiest to just remove
+  // the branch entirely.
+  std::prev(BB->end())->eraseFromParent();
+  Builder.SetInsertPoint(BB);
+  LoadInst *InitLoaded = Builder.CreateLoad(Addr);
+  InitLoaded->setAlignment(AI->getType()->getPrimitiveSizeInBits());
+  Builder.CreateBr(LoopBB);
+
+  // Start the main loop block now that we've taken care of the preliminaries.
+  Builder.SetInsertPoint(LoopBB);
+  PHINode *Loaded = Builder.CreatePHI(AI->getType(), 2, "loaded");
+  Loaded->addIncoming(InitLoaded, BB);
+
+  Value *NewVal;
+  switch (AI->getOperation()) {
+  case AtomicRMWInst::Xchg:
+    NewVal = AI->getValOperand();
+    break;
+  case AtomicRMWInst::Add:
+    NewVal = Builder.CreateAdd(Loaded, AI->getValOperand(), "new");
+    break;
+  case AtomicRMWInst::Sub:
+    NewVal = Builder.CreateSub(Loaded, AI->getValOperand(), "new");
+    break;
+  case AtomicRMWInst::And:
+    NewVal = Builder.CreateAnd(Loaded, AI->getValOperand(), "new");
+    break;
+  case AtomicRMWInst::Nand:
+    NewVal = Builder.CreateNot(Builder.CreateAnd(Loaded, AI->getValOperand()),
+                               "new");
+    break;
+  case AtomicRMWInst::Or:
+    NewVal = Builder.CreateOr(Loaded, AI->getValOperand(), "new");
+    break;
+  case AtomicRMWInst::Xor:
+    NewVal = Builder.CreateXor(Loaded, AI->getValOperand(), "new");
+    break;
+  case AtomicRMWInst::Max:
+    NewVal = Builder.CreateICmpSGT(Loaded, AI->getValOperand());
+    NewVal = Builder.CreateSelect(NewVal, Loaded, AI->getValOperand(), "new");
+    break;
+  case AtomicRMWInst::Min:
+    NewVal = Builder.CreateICmpSLE(Loaded, AI->getValOperand());
+    NewVal = Builder.CreateSelect(NewVal, Loaded, AI->getValOperand(), "new");
+    break;
+  case AtomicRMWInst::UMax:
+    NewVal = Builder.CreateICmpUGT(Loaded, AI->getValOperand());
+    NewVal = Builder.CreateSelect(NewVal, Loaded, AI->getValOperand(), "new");
+    break;
+  case AtomicRMWInst::UMin:
+    NewVal = Builder.CreateICmpULE(Loaded, AI->getValOperand());
+    NewVal = Builder.CreateSelect(NewVal, Loaded, AI->getValOperand(), "new");
+    break;
+  default:
+    llvm_unreachable("Unknown atomic op");
+  }
+
+  Value *Pair = Builder.CreateAtomicCmpXchg(
+      Addr, Loaded, NewVal, Order,
+      AtomicCmpXchgInst::getStrongestFailureOrdering(Order));
+  Value *NewLoaded = Builder.CreateExtractValue(Pair, 0, "newloaded");
+  Loaded->addIncoming(NewLoaded, LoopBB);
+
+  Value *Success = Builder.CreateExtractValue(Pair, 1, "success");
+  Builder.CreateCondBr(Success, ExitBB, LoopBB);
+
+  AI->replaceAllUsesWith(NewLoaded);
+  AI->eraseFromParent();
+
+  return true;
+}
+
+bool X86AtomicExpandPass::expandAtomicStore(StoreInst *SI) {
+  // An atomic store might need cmpxchg16b (or 8b on x86) to execute. Express
+  // this in terms of the usual expansion to "atomicrmw xchg".
+  IRBuilder<> Builder(SI);
+  AtomicRMWInst *AI =
+      Builder.CreateAtomicRMW(AtomicRMWInst::Xchg, SI->getPointerOperand(),
+                              SI->getValueOperand(), SI->getOrdering());
+  SI->eraseFromParent();
+
+  // Now we have an appropriate swap instruction, lower it as usual.
+  if (shouldExpandAtomicRMW(AI))
+    return expandAtomicRMW(AI);
+
+  return AI;
+}

lib/Target/X86/X86ISelDAGToDAG.cpp

case ISD::INTRINSIC_W_CHAIN: {
     }
     }
     break;
   }
   case X86ISD::GlobalBaseReg:
     return getGlobalBaseReg();
 
 
-  case X86ISD::ATOMOR64_DAG:
-  case X86ISD::ATOMXOR64_DAG:
-  case X86ISD::ATOMADD64_DAG:
-  case X86ISD::ATOMSUB64_DAG:
-  case X86ISD::ATOMNAND64_DAG:
-  case X86ISD::ATOMAND64_DAG:
-  case X86ISD::ATOMMAX64_DAG:
-  case X86ISD::ATOMMIN64_DAG:
-  case X86ISD::ATOMUMAX64_DAG:
-  case X86ISD::ATOMUMIN64_DAG:
-  case X86ISD::ATOMSWAP64_DAG: {
-    unsigned Opc;
-    switch (Opcode) {
-    default: llvm_unreachable("Impossible opcode");
-    case X86ISD::ATOMOR64_DAG:   Opc = X86::ATOMOR6432;   break;
-    case X86ISD::ATOMXOR64_DAG:  Opc = X86::ATOMXOR6432;  break;
-    case X86ISD::ATOMADD64_DAG:  Opc = X86::ATOMADD6432;  break;
-    case X86ISD::ATOMSUB64_DAG:  Opc = X86::ATOMSUB6432;  break;
-    case X86ISD::ATOMNAND64_DAG: Opc = X86::ATOMNAND6432; break;
-    case X86ISD::ATOMAND64_DAG:  Opc = X86::ATOMAND6432;  break;
-    case X86ISD::ATOMMAX64_DAG:  Opc = X86::ATOMMAX6432;  break;
-    case X86ISD::ATOMMIN64_DAG:  Opc = X86::ATOMMIN6432;  break;
-    case X86ISD::ATOMUMAX64_DAG: Opc = X86::ATOMUMAX6432; break;
-    case X86ISD::ATOMUMIN64_DAG: Opc = X86::ATOMUMIN6432; break;
-    case X86ISD::ATOMSWAP64_DAG: Opc = X86::ATOMSWAP6432; break;
-    }
-    SDNode *RetVal = SelectAtomic64(Node, Opc);
-    if (RetVal)
-      return RetVal;
-    break;
-  }
-
   case ISD::ATOMIC_LOAD_XOR:
   case ISD::ATOMIC_LOAD_AND:
   case ISD::ATOMIC_LOAD_OR:
   case ISD::ATOMIC_LOAD_ADD: {
     SDNode *RetVal = SelectAtomicLoadArith(Node, NVT);
     if (RetVal)
       return RetVal;
     break;

lib/Target/X86/X86ISelLowering.h

enum NodeType {
 
       // PCMP*STRI
       PCMPISTRI,
       PCMPESTRI,
 
       // XTEST - Test if in transactional execution.
       XTEST,
 
-      // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG,
-      // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG -
-      // Atomic 64-bit binary operations.
-      ATOMADD64_DAG = ISD::FIRST_TARGET_MEMORY_OPCODE,
-      ATOMSUB64_DAG,
-      ATOMOR64_DAG,
-      ATOMXOR64_DAG,
-      ATOMAND64_DAG,
-      ATOMNAND64_DAG,
-      ATOMMAX64_DAG,
-      ATOMMIN64_DAG,
-      ATOMUMAX64_DAG,
-      ATOMUMIN64_DAG,
-      ATOMSWAP64_DAG,
-
       // LCMPXCHG_DAG, LCMPXCHG8_DAG, LCMPXCHG16_DAG - Compare and swap.
-      LCMPXCHG_DAG,
+      LCMPXCHG_DAG = ISD::FIRST_TARGET_MEMORY_OPCODE,
       LCMPXCHG8_DAG,
       LCMPXCHG16_DAG,
 
       // VZEXT_LOAD - Load, scalar_to_vector, and zero extend.
       VZEXT_LOAD,
 
       // FNSTCW16m - Store FP control world into i16 memory.
       FNSTCW16m,

lib/Target/X86/X86ISelLowering.cpp

void X86TargetLowering::resetOperationActions() {
   // Expand certain atomics
   for (unsigned i = 0; i != array_lengthof(IntVTs); ++i) {
     MVT VT = IntVTs[i];
     setOperationAction(ISD::ATOMIC_CMP_SWAP_WITH_SUCCESS, VT, Custom);
     setOperationAction(ISD::ATOMIC_LOAD_SUB, VT, Custom);
     setOperationAction(ISD::ATOMIC_STORE, VT, Custom);
   }
 
-  if (!Subtarget->is64Bit()) {
-    setOperationAction(ISD::ATOMIC_LOAD, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_LOAD_ADD, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_LOAD_SUB, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_LOAD_AND, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_LOAD_OR, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_LOAD_XOR, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_LOAD_NAND, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_SWAP, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_LOAD_MAX, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_LOAD_MIN, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i64, Custom);
-    setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i64, Custom);
-  }
-
   if (Subtarget->hasCmpxchg16b()) {
     setOperationAction(ISD::ATOMIC_CMP_SWAP_WITH_SUCCESS, MVT::i128, Custom);
   }
 
   // FIXME - use subtarget debug flags
   if (!Subtarget->isTargetDarwin() &&
       !Subtarget->isTargetELF() &&
       !Subtarget->isTargetCygMing()) {
SDValue Swap =
                            cast<AtomicSDNode>(Node)->getMemOperand(),
                            cast<AtomicSDNode>(Node)->getOrdering(),
                            cast<AtomicSDNode>(Node)->getOrdering(),
                            cast<AtomicSDNode>(Node)->getSynchScope());
   Results.push_back(Swap.getValue(0));
   Results.push_back(Swap.getValue(2));
 }
 
-static void
-ReplaceATOMIC_BINARY_64(SDNode *Node, SmallVectorImpl<SDValue>&Results,
-                        SelectionDAG &DAG, unsigned NewOp) {
-  SDLoc dl(Node);
-  assert (Node->getValueType(0) == MVT::i64 &&
-          "Only know how to expand i64 atomics");
-
-  SDValue Chain = Node->getOperand(0);
-  SDValue In1 = Node->getOperand(1);
-  SDValue In2L = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32,
-                             Node->getOperand(2), DAG.getIntPtrConstant(0));
-  SDValue In2H = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32,
-                             Node->getOperand(2), DAG.getIntPtrConstant(1));
-  SDValue Ops[] = { Chain, In1, In2L, In2H };
-  SDVTList Tys = DAG.getVTList(MVT::i32, MVT::i32, MVT::Other);
-  SDValue Result =
-    DAG.getMemIntrinsicNode(NewOp, dl, Tys, Ops, MVT::i64,
-                            cast<MemSDNode>(Node)->getMemOperand());
-  SDValue OpsF[] = { Result.getValue(0), Result.getValue(1)};
-  Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, OpsF));
-  Results.push_back(Result.getValue(2));
-}
-
 /// ReplaceNodeResults - Replace a node with an illegal result type
 /// with a new node built out of custom code.
 void X86TargetLowering::ReplaceNodeResults(SDNode *N,
                                            SmallVectorImpl<SDValue>&Results,
                                            SelectionDAG &DAG) const {
   SDLoc dl(N);
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   switch (N->getOpcode()) {
SDValue Success =
                     DAG.getConstant(X86::COND_E, MVT::i8), EFLAGS);
     Success = DAG.getZExtOrTrunc(Success, dl, N->getValueType(1));
 
     Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, T, OpsF));
     Results.push_back(Success);
     Results.push_back(EFLAGS.getValue(1));
     return;
   }
-  case ISD::ATOMIC_LOAD_ADD:
-  case ISD::ATOMIC_LOAD_AND:
-  case ISD::ATOMIC_LOAD_NAND:
-  case ISD::ATOMIC_LOAD_OR:
-  case ISD::ATOMIC_LOAD_SUB:
-  case ISD::ATOMIC_LOAD_XOR:
-  case ISD::ATOMIC_LOAD_MAX:
-  case ISD::ATOMIC_LOAD_MIN:
-  case ISD::ATOMIC_LOAD_UMAX:
-  case ISD::ATOMIC_LOAD_UMIN:
-  case ISD::ATOMIC_SWAP: {
-    unsigned Opc;
-    switch (N->getOpcode()) {
-    default: llvm_unreachable("Unexpected opcode");
-    case ISD::ATOMIC_LOAD_ADD:
-      Opc = X86ISD::ATOMADD64_DAG;
-      break;
-    case ISD::ATOMIC_LOAD_AND:
-      Opc = X86ISD::ATOMAND64_DAG;
-      break;
-    case ISD::ATOMIC_LOAD_NAND:
-      Opc = X86ISD::ATOMNAND64_DAG;
-      break;
-    case ISD::ATOMIC_LOAD_OR:
-      Opc = X86ISD::ATOMOR64_DAG;
-      break;
-    case ISD::ATOMIC_LOAD_SUB:
-      Opc = X86ISD::ATOMSUB64_DAG;
-      break;
-    case ISD::ATOMIC_LOAD_XOR:
-      Opc = X86ISD::ATOMXOR64_DAG;
-      break;
-    case ISD::ATOMIC_LOAD_MAX:
-      Opc = X86ISD::ATOMMAX64_DAG;
-      break;
-    case ISD::ATOMIC_LOAD_MIN:
-      Opc = X86ISD::ATOMMIN64_DAG;
-      break;
-    case ISD::ATOMIC_LOAD_UMAX:
-      Opc = X86ISD::ATOMUMAX64_DAG;
-      break;
-    case ISD::ATOMIC_LOAD_UMIN:
-      Opc = X86ISD::ATOMUMIN64_DAG;
-      break;
-    case ISD::ATOMIC_SWAP:
-      Opc = X86ISD::ATOMSWAP64_DAG;
-      break;
-    }
-    ReplaceATOMIC_BINARY_64(N, Results, DAG, Opc);
-    return;
-  }
   case ISD::ATOMIC_LOAD: {
     ReplaceATOMIC_LOAD(N, Results, DAG);
     return;
   }
   case ISD::BITCAST: {
     assert(Subtarget->hasSSE2() && "Requires at least SSE2!");
     EVT DstVT = N->getValueType(0);
     EVT SrcVT = N->getOperand(0)->getValueType(0);
const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::EH_SJLJ_LONGJMP:    return "X86ISD::EH_SJLJ_LONGJMP";
   case X86ISD::EH_RETURN:          return "X86ISD::EH_RETURN";
   case X86ISD::TC_RETURN:          return "X86ISD::TC_RETURN";
   case X86ISD::FNSTCW16m:          return "X86ISD::FNSTCW16m";
   case X86ISD::FNSTSW16r:          return "X86ISD::FNSTSW16r";
   case X86ISD::LCMPXCHG_DAG:       return "X86ISD::LCMPXCHG_DAG";
   case X86ISD::LCMPXCHG8_DAG:      return "X86ISD::LCMPXCHG8_DAG";
   case X86ISD::LCMPXCHG16_DAG:     return "X86ISD::LCMPXCHG16_DAG";
-  case X86ISD::ATOMADD64_DAG:      return "X86ISD::ATOMADD64_DAG";
-  case X86ISD::ATOMSUB64_DAG:      return "X86ISD::ATOMSUB64_DAG";
-  case X86ISD::ATOMOR64_DAG:       return "X86ISD::ATOMOR64_DAG";
-  case X86ISD::ATOMXOR64_DAG:      return "X86ISD::ATOMXOR64_DAG";
-  case X86ISD::ATOMAND64_DAG:      return "X86ISD::ATOMAND64_DAG";
-  case X86ISD::ATOMNAND64_DAG:     return "X86ISD::ATOMNAND64_DAG";
   case X86ISD::VZEXT_MOVL:         return "X86ISD::VZEXT_MOVL";
   case X86ISD::VZEXT_LOAD:         return "X86ISD::VZEXT_LOAD";
   case X86ISD::VZEXT:              return "X86ISD::VZEXT";
   case X86ISD::VSEXT:              return "X86ISD::VSEXT";
   case X86ISD::VTRUNC:             return "X86ISD::VTRUNC";
   case X86ISD::VTRUNCM:            return "X86ISD::VTRUNCM";
   case X86ISD::VINSERT:            return "X86ISD::VINSERT";
   case X86ISD::VFPEXT:             return "X86ISD::VFPEXT";
static MachineBasicBlock *EmitXBegin(MachineInstr *MI, MachineBasicBlock *MBB,
   BuildMI(*sinkMBB, sinkMBB->begin(), DL,
           TII->get(TargetOpcode::COPY), MI->getOperand(0).getReg())
     .addReg(X86::EAX);
 
   MI->eraseFromParent();
   return sinkMBB;
 }
 
-// Get CMPXCHG opcode for the specified data type.
-static unsigned getCmpXChgOpcode(EVT VT) {
-  switch (VT.getSimpleVT().SimpleTy) {
-  case MVT::i8:  return X86::LCMPXCHG8;
-  case MVT::i16: return X86::LCMPXCHG16;
-  case MVT::i32: return X86::LCMPXCHG32;
-  case MVT::i64: return X86::LCMPXCHG64;
-  default:
-    break;
-  }
-  llvm_unreachable("Invalid operand size!");
-}
-
-// Get LOAD opcode for the specified data type.
-static unsigned getLoadOpcode(EVT VT) {
-  switch (VT.getSimpleVT().SimpleTy) {
-  case MVT::i8:  return X86::MOV8rm;
-  case MVT::i16: return X86::MOV16rm;
-  case MVT::i32: return X86::MOV32rm;
-  case MVT::i64: return X86::MOV64rm;
-  default:
-    break;
-  }
-  llvm_unreachable("Invalid operand size!");
-}
-
-// Get opcode of the non-atomic one from the specified atomic instruction.
-static unsigned getNonAtomicOpcode(unsigned Opc) {
-  switch (Opc) {
-  case X86::ATOMAND8:  return X86::AND8rr;
-  case X86::ATOMAND16: return X86::AND16rr;
-  case X86::ATOMAND32: return X86::AND32rr;
-  case X86::ATOMAND64: return X86::AND64rr;
-  case X86::ATOMOR8:   return X86::OR8rr;
-  case X86::ATOMOR16:  return X86::OR16rr;
-  case X86::ATOMOR32:  return X86::OR32rr;
-  case X86::ATOMOR64:  return X86::OR64rr;
-  case X86::ATOMXOR8:  return X86::XOR8rr;
-  case X86::ATOMXOR16: return X86::XOR16rr;
-  case X86::ATOMXOR32: return X86::XOR32rr;
-  case X86::ATOMXOR64: return X86::XOR64rr;
-  }
-  llvm_unreachable("Unhandled atomic-load-op opcode!");
-}
-
-// Get opcode of the non-atomic one from the specified atomic instruction with
-// extra opcode.
-static unsigned getNonAtomicOpcodeWithExtraOpc(unsigned Opc,
-                                               unsigned &ExtraOpc) {
-  switch (Opc) {
-  case X86::ATOMNAND8:  ExtraOpc = X86::NOT8r;   return X86::AND8rr;
-  case X86::ATOMNAND16: ExtraOpc = X86::NOT16r;  return X86::AND16rr;
-  case X86::ATOMNAND32: ExtraOpc = X86::NOT32r;  return X86::AND32rr;
-  case X86::ATOMNAND64: ExtraOpc = X86::NOT64r;  return X86::AND64rr;
-  case X86::ATOMMAX8:   ExtraOpc = X86::CMP8rr;  return X86::CMOVL32rr;
-  case X86::ATOMMAX16:  ExtraOpc = X86::CMP16rr; return X86::CMOVL16rr;
-  case X86::ATOMMAX32:  ExtraOpc = X86::CMP32rr; return X86::CMOVL32rr;
-  case X86::ATOMMAX64:  ExtraOpc = X86::CMP64rr; return X86::CMOVL64rr;
-  case X86::ATOMMIN8:   ExtraOpc = X86::CMP8rr;  return X86::CMOVG32rr;
-  case X86::ATOMMIN16:  ExtraOpc = X86::CMP16rr; return X86::CMOVG16rr;
-  case X86::ATOMMIN32:  ExtraOpc = X86::CMP32rr; return X86::CMOVG32rr;
-  case X86::ATOMMIN64:  ExtraOpc = X86::CMP64rr; return X86::CMOVG64rr;
-  case X86::ATOMUMAX8:  ExtraOpc = X86::CMP8rr;  return X86::CMOVB32rr;
-  case X86::ATOMUMAX16: ExtraOpc = X86::CMP16rr; return X86::CMOVB16rr;
-  case X86::ATOMUMAX32: ExtraOpc = X86::CMP32rr; return X86::CMOVB32rr;
-  case X86::ATOMUMAX64: ExtraOpc = X86::CMP64rr; return X86::CMOVB64rr;
-  case X86::ATOMUMIN8:  ExtraOpc = X86::CMP8rr;  return X86::CMOVA32rr;
-  case X86::ATOMUMIN16: ExtraOpc = X86::CMP16rr; return X86::CMOVA16rr;
-  case X86::ATOMUMIN32: ExtraOpc = X86::CMP32rr; return X86::CMOVA32rr;
-  case X86::ATOMUMIN64: ExtraOpc = X86::CMP64rr; return X86::CMOVA64rr;
-  }
-  llvm_unreachable("Unhandled atomic-load-op opcode!");
-}
-
-// Get opcode of the non-atomic one from the specified atomic instruction for
-// 64-bit data type on 32-bit target.
-static unsigned getNonAtomic6432Opcode(unsigned Opc, unsigned &HiOpc) {
-  switch (Opc) {
-  case X86::ATOMAND6432:  HiOpc = X86::AND32rr; return X86::AND32rr;
-  case X86::ATOMOR6432:   HiOpc = X86::OR32rr;  return X86::OR32rr;
-  case X86::ATOMXOR6432:  HiOpc = X86::XOR32rr; return X86::XOR32rr;
-  case X86::ATOMADD6432:  HiOpc = X86::ADC32rr; return X86::ADD32rr;
-  case X86::ATOMSUB6432:  HiOpc = X86::SBB32rr; return X86::SUB32rr;
-  case X86::ATOMSWAP6432: HiOpc = X86::MOV32rr; return X86::MOV32rr;
-  case X86::ATOMMAX6432:  HiOpc = X86::SETLr;   return X86::SETLr;
-  case X86::ATOMMIN6432:  HiOpc = X86::SETGr;   return X86::SETGr;
-  case X86::ATOMUMAX6432: HiOpc = X86::SETBr;   return X86::SETBr;
-  case X86::ATOMUMIN6432: HiOpc = X86::SETAr;   return X86::SETAr;
-  }
-  llvm_unreachable("Unhandled atomic-load-op opcode!");
-}
-
-// Get opcode of the non-atomic one from the specified atomic instruction for
-// 64-bit data type on 32-bit target with extra opcode.
-static unsigned getNonAtomic6432OpcodeWithExtraOpc(unsigned Opc,
-                                                   unsigned &HiOpc,
-                                                   unsigned &ExtraOpc) {
-  switch (Opc) {
-  case X86::ATOMNAND6432:
-    ExtraOpc = X86::NOT32r;
-    HiOpc = X86::AND32rr;
-    return X86::AND32rr;
-  }
-  llvm_unreachable("Unhandled atomic-load-op opcode!");
-}
-
-// Get pseudo CMOV opcode from the specified data type.
-static unsigned getPseudoCMOVOpc(EVT VT) {
-  switch (VT.getSimpleVT().SimpleTy) {
-  case MVT::i8:  return X86::CMOV_GR8;
-  case MVT::i16: return X86::CMOV_GR16;
-  case MVT::i32: return X86::CMOV_GR32;
-  default:
-    break;
-  }
-  llvm_unreachable("Unknown CMOV opcode!");
-}
-
-// EmitAtomicLoadArith - emit the code sequence for pseudo atomic instructions.
-// They will be translated into a spin-loop or compare-exchange loop from
-//
-//    ...
-//    dst = atomic-fetch-op MI.addr, MI.val
-//    ...
-//
-// to
-//
-//    ...
-//    t1 = LOAD MI.addr
-// loop:
-//    t4 = phi(t1, t3 / loop)
-//    t2 = OP MI.val, t4
-//    EAX = t4
-//    LCMPXCHG [MI.addr], t2, [EAX is implicitly used & defined]
-//    t3 = EAX
-//    JNE loop
-// sink:
-//    dst = t3
-//    ...
-MachineBasicBlock *
-X86TargetLowering::EmitAtomicLoadArith(MachineInstr *MI,
-                                       MachineBasicBlock *MBB) const {
-  MachineFunction *MF = MBB->getParent();
-  const TargetInstrInfo *TII = MF->getTarget().getInstrInfo();
-  DebugLoc DL = MI->getDebugLoc();
-
-  MachineRegisterInfo &MRI = MF->getRegInfo();
-
-  const BasicBlock *BB = MBB->getBasicBlock();
-  MachineFunction::iterator I = MBB;
-  ++I;
-
-  assert(MI->getNumOperands() <= X86::AddrNumOperands + 4 &&
-         "Unexpected number of operands");
-
-  assert(MI->hasOneMemOperand() &&
-         "Expected atomic-load-op to have one memoperand");
-
-  // Memory Reference
-  MachineInstr::mmo_iterator MMOBegin = MI->memoperands_begin();
-  MachineInstr::mmo_iterator MMOEnd = MI->memoperands_end();
-
-  unsigned DstReg, SrcReg;
-  unsigned MemOpndSlot;
-
-  unsigned CurOp = 0;
-
-  DstReg = MI->getOperand(CurOp++).getReg();
-  MemOpndSlot = CurOp;
-  CurOp += X86::AddrNumOperands;
-  SrcReg = MI->getOperand(CurOp++).getReg();
-
-  const TargetRegisterClass *RC = MRI.getRegClass(DstReg);
-  MVT::SimpleValueType VT = *RC->vt_begin();
-  unsigned t1 = MRI.createVirtualRegister(RC);
-  unsigned t2 = MRI.createVirtualRegister(RC);
-  unsigned t3 = MRI.createVirtualRegister(RC);
-  unsigned t4 = MRI.createVirtualRegister(RC);
-  unsigned PhyReg = getX86SubSuperRegister(X86::EAX, VT);
-
-  unsigned LCMPXCHGOpc = getCmpXChgOpcode(VT);
-  unsigned LOADOpc = getLoadOpcode(VT);
-
-  // For the atomic load-arith operator, we generate
-  //
-  //  thisMBB:
-  //    t1 = LOAD [MI.addr]
-  //  mainMBB:
-  //    t4 = phi(t1 / thisMBB, t3 / mainMBB)
-  //    t1 = OP MI.val, EAX
-  //    EAX = t4
-  //    LCMPXCHG [MI.addr], t1, [EAX is implicitly used & defined]
-  //    t3 = EAX
-  //    JNE mainMBB
-  //  sinkMBB:
-  //    dst = t3
-
-  MachineBasicBlock *thisMBB = MBB;
-  MachineBasicBlock *mainMBB = MF->CreateMachineBasicBlock(BB);
-  MachineBasicBlock *sinkMBB = MF->CreateMachineBasicBlock(BB);
-  MF->insert(I, mainMBB);
-  MF->insert(I, sinkMBB);
-
-  MachineInstrBuilder MIB;
-
-  // Transfer the remainder of BB and its successor edges to sinkMBB.
-  sinkMBB->splice(sinkMBB->begin(), MBB,
-                  std::next(MachineBasicBlock::iterator(MI)), MBB->end());
-  sinkMBB->transferSuccessorsAndUpdatePHIs(MBB);
-
-  // thisMBB:
-  MIB = BuildMI(thisMBB, DL, TII->get(LOADOpc), t1);
-  for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
-    MachineOperand NewMO = MI->getOperand(MemOpndSlot + i);
-    if (NewMO.isReg())
-      NewMO.setIsKill(false);
-    MIB.addOperand(NewMO);
-  }
-  for (MachineInstr::mmo_iterator MMOI = MMOBegin; MMOI != MMOEnd; ++MMOI) {
-    unsigned flags = (*MMOI)->getFlags();
-    flags = (flags & ~MachineMemOperand::MOStore) | MachineMemOperand::MOLoad;
-    MachineMemOperand *MMO =
-      MF->getMachineMemOperand((*MMOI)->getPointerInfo(), flags,
-                               (*MMOI)->getSize(),
-                               (*MMOI)->getBaseAlignment(),
-                               (*MMOI)->getTBAAInfo(),
-                               (*MMOI)->getRanges());
-    MIB.addMemOperand(MMO);
-  }
-
-  thisMBB->addSuccessor(mainMBB);
-
-  // mainMBB:
-  MachineBasicBlock *origMainMBB = mainMBB;
-
-  // Add a PHI.
-  MachineInstr *Phi = BuildMI(mainMBB, DL, TII->get(X86::PHI), t4)
-                        .addReg(t1).addMBB(thisMBB).addReg(t3).addMBB(mainMBB);
-
-  unsigned Opc = MI->getOpcode();
-  switch (Opc) {
-  default:
-    llvm_unreachable("Unhandled atomic-load-op opcode!");
-  case X86::ATOMAND8:
-  case X86::ATOMAND16:
-  case X86::ATOMAND32:
-  case X86::ATOMAND64:
-  case X86::ATOMOR8:
-  case X86::ATOMOR16:
-  case X86::ATOMOR32:
-  case X86::ATOMOR64:
-  case X86::ATOMXOR8:
-  case X86::ATOMXOR16:
-  case X86::ATOMXOR32:
-  case X86::ATOMXOR64: {
-    unsigned ARITHOpc = getNonAtomicOpcode(Opc);
-    BuildMI(mainMBB, DL, TII->get(ARITHOpc), t2).addReg(SrcReg)
-      .addReg(t4);
-    break;
-  }
-  case X86::ATOMNAND8:
-  case X86::ATOMNAND16:
-  case X86::ATOMNAND32:
-  case X86::ATOMNAND64: {
-    unsigned Tmp = MRI.createVirtualRegister(RC);
-    unsigned NOTOpc;
-    unsigned ANDOpc = getNonAtomicOpcodeWithExtraOpc(Opc, NOTOpc);
-    BuildMI(mainMBB, DL, TII->get(ANDOpc), Tmp).addReg(SrcReg)
-      .addReg(t4);
-    BuildMI(mainMBB, DL, TII->get(NOTOpc), t2).addReg(Tmp);
-    break;
-  }
-  case X86::ATOMMAX8:
-  case X86::ATOMMAX16:
-  case X86::ATOMMAX32:
-  case X86::ATOMMAX64:
-  case X86::ATOMMIN8:
-  case X86::ATOMMIN16:
-  case X86::ATOMMIN32:
-  case X86::ATOMMIN64:
-  case X86::ATOMUMAX8:
-  case X86::ATOMUMAX16:
-  case X86::ATOMUMAX32:
-  case X86::ATOMUMAX64:
-  case X86::ATOMUMIN8:
-  case X86::ATOMUMIN16:
-  case X86::ATOMUMIN32:
-  case X86::ATOMUMIN64: {
-    unsigned CMPOpc;
-    unsigned CMOVOpc = getNonAtomicOpcodeWithExtraOpc(Opc, CMPOpc);
-
-    BuildMI(mainMBB, DL, TII->get(CMPOpc))
-      .addReg(SrcReg)
-      .addReg(t4);
-
-    if (Subtarget->hasCMov()) {
-      if (VT != MVT::i8) {
-        // Native support
-        BuildMI(mainMBB, DL, TII->get(CMOVOpc), t2)
-          .addReg(SrcReg)
-          .addReg(t4);
-      } else {
-        // Promote i8 to i32 to use CMOV32
-        const TargetRegisterInfo* TRI = MF->getTarget().getRegisterInfo();
-        const TargetRegisterClass *RC32 =
-          TRI->getSubClassWithSubReg(getRegClassFor(MVT::i32), X86::sub_8bit);
-        unsigned SrcReg32 = MRI.createVirtualRegister(RC32);
-        unsigned AccReg32 = MRI.createVirtualRegister(RC32);
-        unsigned Tmp = MRI.createVirtualRegister(RC32);
-
-        unsigned Undef = MRI.createVirtualRegister(RC32);
-        BuildMI(mainMBB, DL, TII->get(TargetOpcode::IMPLICIT_DEF), Undef);
-
-        BuildMI(mainMBB, DL, TII->get(TargetOpcode::INSERT_SUBREG), SrcReg32)
-          .addReg(Undef)
-          .addReg(SrcReg)
-          .addImm(X86::sub_8bit);
-        BuildMI(mainMBB, DL, TII->get(TargetOpcode::INSERT_SUBREG), AccReg32)
-          .addReg(Undef)
-          .addReg(t4)
-          .addImm(X86::sub_8bit);
-
-        BuildMI(mainMBB, DL, TII->get(CMOVOpc), Tmp)
-          .addReg(SrcReg32)
-          .addReg(AccReg32);
-
-        BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), t2)
-          .addReg(Tmp, 0, X86::sub_8bit);
-      }
-    } else {
-      // Use pseudo select and lower them.
-      assert((VT == MVT::i8 || VT == MVT::i16 || VT == MVT::i32) &&
-             "Invalid atomic-load-op transformation!");
-      unsigned SelOpc = getPseudoCMOVOpc(VT);
-      X86::CondCode CC = X86::getCondFromCMovOpc(CMOVOpc);
-      assert(CC != X86::COND_INVALID && "Invalid atomic-load-op transformation!");
-      MIB = BuildMI(mainMBB, DL, TII->get(SelOpc), t2)
-              .addReg(SrcReg).addReg(t4)
-              .addImm(CC);
-      mainMBB = EmitLoweredSelect(MIB, mainMBB);
-      // Replace the original PHI node as mainMBB is changed after CMOV
-      // lowering.
-      BuildMI(*origMainMBB, Phi, DL, TII->get(X86::PHI), t4)
-        .addReg(t1).addMBB(thisMBB).addReg(t3).addMBB(mainMBB);
-      Phi->eraseFromParent();
-    }
-    break;
-  }
-  }
-
-  // Copy PhyReg back from virtual register.
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), PhyReg)
-    .addReg(t4);
-
-  MIB = BuildMI(mainMBB, DL, TII->get(LCMPXCHGOpc));
-  for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
-    MachineOperand NewMO = MI->getOperand(MemOpndSlot + i);
-    if (NewMO.isReg())
-      NewMO.setIsKill(false);
-    MIB.addOperand(NewMO);
-  }
-  MIB.addReg(t2);
-  MIB.setMemRefs(MMOBegin, MMOEnd);
-
-  // Copy PhyReg back to virtual register.
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), t3)
-    .addReg(PhyReg);
-
-  BuildMI(mainMBB, DL, TII->get(X86::JNE_4)).addMBB(origMainMBB);
-
-  mainMBB->addSuccessor(origMainMBB);
-  mainMBB->addSuccessor(sinkMBB);
-
-  // sinkMBB:
-  BuildMI(*sinkMBB, sinkMBB->begin(), DL,
-          TII->get(TargetOpcode::COPY), DstReg)
-    .addReg(t3);
-
-  MI->eraseFromParent();
-  return sinkMBB;
-}
-
-// EmitAtomicLoadArith6432 - emit the code sequence for pseudo atomic
-// instructions. They will be translated into a spin-loop or compare-exchange
-// loop from
-//
-//    ...
-//    dst = atomic-fetch-op MI.addr, MI.val
-//    ...
-//
-// to
-//
-//    ...
-//    t1L = LOAD [MI.addr + 0]
-//    t1H = LOAD [MI.addr + 4]
-// loop:
-//    t4L = phi(t1L, t3L / loop)
-//    t4H = phi(t1H, t3H / loop)
-//    t2L = OP MI.val.lo, t4L
-//    t2H = OP MI.val.hi, t4H
-//    EAX = t4L
-//    EDX = t4H
-//    EBX = t2L
-//    ECX = t2H
-//    LCMPXCHG8B [MI.addr], [ECX:EBX & EDX:EAX are implicitly used and EDX:EAX is implicitly defined]
-//    t3L = EAX
-//    t3H = EDX
-//    JNE loop
-// sink:
-//    dstL = t3L
-//    dstH = t3H
-//    ...
-MachineBasicBlock *
-X86TargetLowering::EmitAtomicLoadArith6432(MachineInstr *MI,
-                                           MachineBasicBlock *MBB) const {
-  MachineFunction *MF = MBB->getParent();
-  const TargetInstrInfo *TII = MF->getTarget().getInstrInfo();
-  DebugLoc DL = MI->getDebugLoc();
-
-  MachineRegisterInfo &MRI = MF->getRegInfo();
-
-  const BasicBlock *BB = MBB->getBasicBlock();
-  MachineFunction::iterator I = MBB;
-  ++I;
-
-  assert(MI->getNumOperands() <= X86::AddrNumOperands + 7 &&
-         "Unexpected number of operands");
-
-  assert(MI->hasOneMemOperand() &&
-         "Expected atomic-load-op32 to have one memoperand");
-
-  // Memory Reference
-  MachineInstr::mmo_iterator MMOBegin = MI->memoperands_begin();
-  MachineInstr::mmo_iterator MMOEnd = MI->memoperands_end();
-
-  unsigned DstLoReg, DstHiReg;
-  unsigned SrcLoReg, SrcHiReg;
-  unsigned MemOpndSlot;
-
-  unsigned CurOp = 0;
-
-  DstLoReg = MI->getOperand(CurOp++).getReg();
-  DstHiReg = MI->getOperand(CurOp++).getReg();
-  MemOpndSlot = CurOp;
-  CurOp += X86::AddrNumOperands;
-  SrcLoReg = MI->getOperand(CurOp++).getReg();
-  SrcHiReg = MI->getOperand(CurOp++).getReg();
-
-  const TargetRegisterClass *RC = &X86::GR32RegClass;
-  const TargetRegisterClass *RC8 = &X86::GR8RegClass;
-
-  unsigned t1L = MRI.createVirtualRegister(RC);
-  unsigned t1H = MRI.createVirtualRegister(RC);
-  unsigned t2L = MRI.createVirtualRegister(RC);
-  unsigned t2H = MRI.createVirtualRegister(RC);
-  unsigned t3L = MRI.createVirtualRegister(RC);
-  unsigned t3H = MRI.createVirtualRegister(RC);
-  unsigned t4L = MRI.createVirtualRegister(RC);
-  unsigned t4H = MRI.createVirtualRegister(RC);
-
-  unsigned LCMPXCHGOpc = X86::LCMPXCHG8B;
-  unsigned LOADOpc = X86::MOV32rm;
-
-  // For the atomic load-arith operator, we generate
-  //
-  //  thisMBB:
-  //    t1L = LOAD [MI.addr + 0]
-  //    t1H = LOAD [MI.addr + 4]
-  //  mainMBB:
-  //    t4L = phi(t1L / thisMBB, t3L / mainMBB)
-  //    t4H = phi(t1H / thisMBB, t3H / mainMBB)
-  //    t2L = OP MI.val.lo, t4L
-  //    t2H = OP MI.val.hi, t4H
-  //    EBX = t2L
-  //    ECX = t2H
-  //    LCMPXCHG8B [MI.addr], [ECX:EBX & EDX:EAX are implicitly used and EDX:EAX is implicitly defined]
-  //    t3L = EAX
-  //    t3H = EDX
-  //    JNE loop
-  //  sinkMBB:
-  //    dstL = t3L
-  //    dstH = t3H
-
-  MachineBasicBlock *thisMBB = MBB;
-  MachineBasicBlock *mainMBB = MF->CreateMachineBasicBlock(BB);
-  MachineBasicBlock *sinkMBB = MF->CreateMachineBasicBlock(BB);
-  MF->insert(I, mainMBB);
-  MF->insert(I, sinkMBB);
-
-  MachineInstrBuilder MIB;
-
-  // Transfer the remainder of BB and its successor edges to sinkMBB.
-  sinkMBB->splice(sinkMBB->begin(), MBB,
-                  std::next(MachineBasicBlock::iterator(MI)), MBB->end());
-  sinkMBB->transferSuccessorsAndUpdatePHIs(MBB);
-
-  // thisMBB:
-  // Lo
-  MIB = BuildMI(thisMBB, DL, TII->get(LOADOpc), t1L);
-  for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
-    MachineOperand NewMO = MI->getOperand(MemOpndSlot + i);
-    if (NewMO.isReg())
-      NewMO.setIsKill(false);
-    MIB.addOperand(NewMO);
-  }
-  for (MachineInstr::mmo_iterator MMOI = MMOBegin; MMOI != MMOEnd; ++MMOI) {
-    unsigned flags = (*MMOI)->getFlags();
-    flags = (flags & ~MachineMemOperand::MOStore) | MachineMemOperand::MOLoad;
-    MachineMemOperand *MMO =
-      MF->getMachineMemOperand((*MMOI)->getPointerInfo(), flags,
-                               (*MMOI)->getSize(),
-                               (*MMOI)->getBaseAlignment(),
-                               (*MMOI)->getTBAAInfo(),
-                               (*MMOI)->getRanges());
-    MIB.addMemOperand(MMO);
-  };
-  MachineInstr *LowMI = MIB;
-
-  // Hi
-  MIB = BuildMI(thisMBB, DL, TII->get(LOADOpc), t1H);
-  for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
-    if (i == X86::AddrDisp) {
-      MIB.addDisp(MI->getOperand(MemOpndSlot + i), 4); // 4 == sizeof(i32)
-    } else {
-      MachineOperand NewMO = MI->getOperand(MemOpndSlot + i);
-      if (NewMO.isReg())
-        NewMO.setIsKill(false);
-      MIB.addOperand(NewMO);
-    }
-  }
-  MIB.setMemRefs(LowMI->memoperands_begin(), LowMI->memoperands_end());
-
-  thisMBB->addSuccessor(mainMBB);
-
-  // mainMBB:
-  MachineBasicBlock *origMainMBB = mainMBB;
-
-  // Add PHIs.
-  MachineInstr *PhiL = BuildMI(mainMBB, DL, TII->get(X86::PHI), t4L)
-                        .addReg(t1L).addMBB(thisMBB).addReg(t3L).addMBB(mainMBB);
-  MachineInstr *PhiH = BuildMI(mainMBB, DL, TII->get(X86::PHI), t4H)
-                        .addReg(t1H).addMBB(thisMBB).addReg(t3H).addMBB(mainMBB);
-
-  unsigned Opc = MI->getOpcode();
-  switch (Opc) {
-  default:
-    llvm_unreachable("Unhandled atomic-load-op6432 opcode!");
-  case X86::ATOMAND6432:
-  case X86::ATOMOR6432:
-  case X86::ATOMXOR6432:
-  case X86::ATOMADD6432:
-  case X86::ATOMSUB6432: {
-    unsigned HiOpc;
-    unsigned LoOpc = getNonAtomic6432Opcode(Opc, HiOpc);
-    BuildMI(mainMBB, DL, TII->get(LoOpc), t2L).addReg(t4L)
-      .addReg(SrcLoReg);
-    BuildMI(mainMBB, DL, TII->get(HiOpc), t2H).addReg(t4H)
-      .addReg(SrcHiReg);
-    break;
-  }
-  case X86::ATOMNAND6432: {
-    unsigned HiOpc, NOTOpc;
-    unsigned LoOpc = getNonAtomic6432OpcodeWithExtraOpc(Opc, HiOpc, NOTOpc);
-    unsigned TmpL = MRI.createVirtualRegister(RC);
-    unsigned TmpH = MRI.createVirtualRegister(RC);
-    BuildMI(mainMBB, DL, TII->get(LoOpc), TmpL).addReg(SrcLoReg)
-      .addReg(t4L);
-    BuildMI(mainMBB, DL, TII->get(HiOpc), TmpH).addReg(SrcHiReg)
-      .addReg(t4H);
-    BuildMI(mainMBB, DL, TII->get(NOTOpc), t2L).addReg(TmpL);
-    BuildMI(mainMBB, DL, TII->get(NOTOpc), t2H).addReg(TmpH);
-    break;
-  }
-  case X86::ATOMMAX6432:
-  case X86::ATOMMIN6432:
-  case X86::ATOMUMAX6432:
-  case X86::ATOMUMIN6432: {
-    unsigned HiOpc;
-    unsigned LoOpc = getNonAtomic6432Opcode(Opc, HiOpc);
-    unsigned cL = MRI.createVirtualRegister(RC8);
-    unsigned cH = MRI.createVirtualRegister(RC8);
-    unsigned cL32 = MRI.createVirtualRegister(RC);
-    unsigned cH32 = MRI.createVirtualRegister(RC);
-    unsigned cc = MRI.createVirtualRegister(RC);
-    // cl := cmp src_lo, lo
-    BuildMI(mainMBB, DL, TII->get(X86::CMP32rr))
-      .addReg(SrcLoReg).addReg(t4L);
-    BuildMI(mainMBB, DL, TII->get(LoOpc), cL);
-    BuildMI(mainMBB, DL, TII->get(X86::MOVZX32rr8), cL32).addReg(cL);
-    // ch := cmp src_hi, hi
-    BuildMI(mainMBB, DL, TII->get(X86::CMP32rr))
-      .addReg(SrcHiReg).addReg(t4H);
-    BuildMI(mainMBB, DL, TII->get(HiOpc), cH);
-    BuildMI(mainMBB, DL, TII->get(X86::MOVZX32rr8), cH32).addReg(cH);
-    // cc := if (src_hi == hi) ? cl : ch;
-    if (Subtarget->hasCMov()) {
-      BuildMI(mainMBB, DL, TII->get(X86::CMOVE32rr), cc)
-        .addReg(cH32).addReg(cL32);
-    } else {
-      MIB = BuildMI(mainMBB, DL, TII->get(X86::CMOV_GR32), cc)
-              .addReg(cH32).addReg(cL32)
-              .addImm(X86::COND_E);
-      mainMBB = EmitLoweredSelect(MIB, mainMBB);
-    }
-    BuildMI(mainMBB, DL, TII->get(X86::TEST32rr)).addReg(cc).addReg(cc);
-    if (Subtarget->hasCMov()) {
-      BuildMI(mainMBB, DL, TII->get(X86::CMOVNE32rr), t2L)
-        .addReg(SrcLoReg).addReg(t4L);
-      BuildMI(mainMBB, DL, TII->get(X86::CMOVNE32rr), t2H)
-        .addReg(SrcHiReg).addReg(t4H);
-    } else {
-      MIB = BuildMI(mainMBB, DL, TII->get(X86::CMOV_GR32), t2L)
-              .addReg(SrcLoReg).addReg(t4L)
-              .addImm(X86::COND_NE);
-      mainMBB = EmitLoweredSelect(MIB, mainMBB);
-      // As the lowered CMOV won't clobber EFLAGS, we could reuse it for the
-      // 2nd CMOV lowering.
-      mainMBB->addLiveIn(X86::EFLAGS);
-      MIB = BuildMI(mainMBB, DL, TII->get(X86::CMOV_GR32), t2H)
-              .addReg(SrcHiReg).addReg(t4H)
-              .addImm(X86::COND_NE);
-      mainMBB = EmitLoweredSelect(MIB, mainMBB);
-      // Replace the original PHI node as mainMBB is changed after CMOV
-      // lowering.
-      BuildMI(*origMainMBB, PhiL, DL, TII->get(X86::PHI), t4L)
-        .addReg(t1L).addMBB(thisMBB).addReg(t3L).addMBB(mainMBB);
-      BuildMI(*origMainMBB, PhiH, DL, TII->get(X86::PHI), t4H)
-        .addReg(t1H).addMBB(thisMBB).addReg(t3H).addMBB(mainMBB);
-      PhiL->eraseFromParent();
-      PhiH->eraseFromParent();
-    }
-    break;
-  }
-  case X86::ATOMSWAP6432: {
-    unsigned HiOpc;
-    unsigned LoOpc = getNonAtomic6432Opcode(Opc, HiOpc);
-    BuildMI(mainMBB, DL, TII->get(LoOpc), t2L).addReg(SrcLoReg);
-    BuildMI(mainMBB, DL, TII->get(HiOpc), t2H).addReg(SrcHiReg);
-    break;
-  }
-  }
-
-  // Copy EDX:EAX back from HiReg:LoReg
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::EAX).addReg(t4L);
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::EDX).addReg(t4H);
-  // Copy ECX:EBX from t1H:t1L
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::EBX).addReg(t2L);
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), X86::ECX).addReg(t2H);
-
-  MIB = BuildMI(mainMBB, DL, TII->get(LCMPXCHGOpc));
-  for (unsigned i = 0; i < X86::AddrNumOperands; ++i) {
-    MachineOperand NewMO = MI->getOperand(MemOpndSlot + i);
-    if (NewMO.isReg())
-      NewMO.setIsKill(false);
-    MIB.addOperand(NewMO);
-  }
-  MIB.setMemRefs(MMOBegin, MMOEnd);
-
-  // Copy EDX:EAX back to t3H:t3L
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), t3L).addReg(X86::EAX);
-  BuildMI(mainMBB, DL, TII->get(TargetOpcode::COPY), t3H).addReg(X86::EDX);
-
-  BuildMI(mainMBB, DL, TII->get(X86::JNE_4)).addMBB(origMainMBB);
-
-  mainMBB->addSuccessor(origMainMBB);
-  mainMBB->addSuccessor(sinkMBB);
-
-  // sinkMBB:
-  BuildMI(*sinkMBB, sinkMBB->begin(), DL,
-          TII->get(TargetOpcode::COPY), DstLoReg)
-    .addReg(t3L);
-  BuildMI(*sinkMBB, sinkMBB->begin(), DL,
-          TII->get(TargetOpcode::COPY), DstHiReg)
-    .addReg(t3H);
-
-  MI->eraseFromParent();
-  return sinkMBB;
-}
-
 // FIXME: When we get size specific XMM0 registers, i.e. XMM0_V16I8
 // or XMM0_V32I8 in AVX all of this code can be replaced with that
 // in the .td file.
 static MachineBasicBlock *EmitPCMPSTRM(MachineInstr *MI, MachineBasicBlock *BB,
                                        const TargetInstrInfo *TII) {
   unsigned Opc;
   switch (MI->getOpcode()) {
   default: llvm_unreachable("illegal opcode!");
X86TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
   // Thread synchronization.
   case X86::MONITOR:
     return EmitMonitor(MI, BB, BB->getParent()->getTarget().getInstrInfo(), Subtarget);
 
   // xbegin
   case X86::XBEGIN:
     return EmitXBegin(MI, BB, BB->getParent()->getTarget().getInstrInfo());
 
-  // Atomic Lowering.
-  case X86::ATOMAND8:
-  case X86::ATOMAND16:
-  case X86::ATOMAND32:
-  case X86::ATOMAND64:
-    // Fall through
-  case X86::ATOMOR8:
-  case X86::ATOMOR16:
-  case X86::ATOMOR32:
-  case X86::ATOMOR64:
-    // Fall through
-  case X86::ATOMXOR16:
-  case X86::ATOMXOR8:
-  case X86::ATOMXOR32:
-  case X86::ATOMXOR64:
-    // Fall through
-  case X86::ATOMNAND8:
-  case X86::ATOMNAND16:
-  case X86::ATOMNAND32:
-  case X86::ATOMNAND64:
-    // Fall through
-  case X86::ATOMMAX8:
-  case X86::ATOMMAX16:
-  case X86::ATOMMAX32:
-  case X86::ATOMMAX64:
-    // Fall through
-  case X86::ATOMMIN8:
-  case X86::ATOMMIN16:
-  case X86::ATOMMIN32:
-  case X86::ATOMMIN64:
-    // Fall through
-  case X86::ATOMUMAX8:
-  case X86::ATOMUMAX16:
-  case X86::ATOMUMAX32:
-  case X86::ATOMUMAX64:
-    // Fall through
-  case X86::ATOMUMIN8:
-  case X86::ATOMUMIN16:
-  case X86::ATOMUMIN32:
-  case X86::ATOMUMIN64:
-    return EmitAtomicLoadArith(MI, BB);
-
-  // This group does 64-bit operations on a 32-bit host.
-  case X86::ATOMAND6432:
-  case X86::ATOMOR6432:
-  case X86::ATOMXOR6432:
-  case X86::ATOMNAND6432:
-  case X86::ATOMADD6432:
-  case X86::ATOMSUB6432:
-  case X86::ATOMMAX6432:
-  case X86::ATOMMIN6432:
-  case X86::ATOMUMAX6432:
-  case X86::ATOMUMIN6432:
-  case X86::ATOMSWAP6432:
-    return EmitAtomicLoadArith6432(MI, BB);
-
   case X86::VASTART_SAVE_XMM_REGS:
     return EmitVAStartSaveXMMRegsWithCustomInserter(MI, BB);
 
   case X86::VAARG_64:
     return EmitVAARG64WithCustomInserter(MI, BB);
 
   case X86::EH_SjLj_SetJmp32:
   case X86::EH_SjLj_SetJmp64:

lib/Target/X86/X86InstrCompiler.td

def CMOV_RFP80 : I<0, Pseudo,
                     "#CMOV_RFP80 PSEUDO!",
                     [(set RFP80:$dst,
                       (X86cmov RFP80:$src1, RFP80:$src2, imm:$cond,
                                                   EFLAGS))]>;
 } // UsesCustomInserter = 1, Uses = [EFLAGS]
 
 
 //===----------------------------------------------------------------------===//
-// Atomic Instruction Pseudo Instructions
-//===----------------------------------------------------------------------===//
-
-// Pseudo atomic instructions
-
-multiclass PSEUDO_ATOMIC_LOAD_BINOP<string mnemonic> {
-  let usesCustomInserter = 1, mayLoad = 1, mayStore = 1 in {
-    let Defs = [EFLAGS, AL] in
-    def NAME#8  : I<0, Pseudo, (outs GR8:$dst),
-                    (ins i8mem:$ptr, GR8:$val),
-                    !strconcat(mnemonic, "8 PSEUDO!"), []>;
-    let Defs = [EFLAGS, AX] in
-    def NAME#16 : I<0, Pseudo,(outs GR16:$dst),
-                    (ins i16mem:$ptr, GR16:$val),
-                    !strconcat(mnemonic, "16 PSEUDO!"), []>;
-    let Defs = [EFLAGS, EAX] in
-    def NAME#32 : I<0, Pseudo, (outs GR32:$dst),
-                    (ins i32mem:$ptr, GR32:$val),
-                    !strconcat(mnemonic, "32 PSEUDO!"), []>;
-    let Defs = [EFLAGS, RAX] in
-    def NAME#64 : I<0, Pseudo, (outs GR64:$dst),
-                    (ins i64mem:$ptr, GR64:$val),
-                    !strconcat(mnemonic, "64 PSEUDO!"), []>;
-  }
-}
-
-multiclass PSEUDO_ATOMIC_LOAD_BINOP_PATS<string name, string frag> {
-  def : Pat<(!cast<PatFrag>(frag # "_8") addr:$ptr, GR8:$val),
-            (!cast<Instruction>(name # "8") addr:$ptr, GR8:$val)>;
-  def : Pat<(!cast<PatFrag>(frag # "_16") addr:$ptr, GR16:$val),
-            (!cast<Instruction>(name # "16") addr:$ptr, GR16:$val)>;
-  def : Pat<(!cast<PatFrag>(frag # "_32") addr:$ptr, GR32:$val),
-            (!cast<Instruction>(name # "32") addr:$ptr, GR32:$val)>;
-  def : Pat<(!cast<PatFrag>(frag # "_64") addr:$ptr, GR64:$val),
-            (!cast<Instruction>(name # "64") addr:$ptr, GR64:$val)>;
-}
-
-// Atomic exchange, and, or, xor
-defm ATOMAND  : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMAND">;
-defm ATOMOR   : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMOR">;
-defm ATOMXOR  : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMXOR">;
-defm ATOMNAND : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMNAND">;
-defm ATOMMAX  : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMMAX">;
-defm ATOMMIN  : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMMIN">;
-defm ATOMUMAX : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMUMAX">;
-defm ATOMUMIN : PSEUDO_ATOMIC_LOAD_BINOP<"#ATOMUMIN">;
-
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMAND",  "atomic_load_and">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMOR",   "atomic_load_or">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMXOR",  "atomic_load_xor">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMNAND", "atomic_load_nand">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMMAX",  "atomic_load_max">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMMIN",  "atomic_load_min">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMUMAX", "atomic_load_umax">;
-defm : PSEUDO_ATOMIC_LOAD_BINOP_PATS<"ATOMUMIN", "atomic_load_umin">;
-
-multiclass PSEUDO_ATOMIC_LOAD_BINOP6432<string mnemonic> {
-  let usesCustomInserter = 1, Defs = [EFLAGS, EAX, EDX],
-      mayLoad = 1, mayStore = 1, hasSideEffects = 0 in
-    def NAME#6432 : I<0, Pseudo, (outs GR32:$dst1, GR32:$dst2),
-                      (ins i64mem:$ptr, GR32:$val1, GR32:$val2),
-                      !strconcat(mnemonic, "6432 PSEUDO!"), []>;
-}
-
-defm ATOMAND  : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMAND">;
-defm ATOMOR   : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMOR">;
-defm ATOMXOR  : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMXOR">;
-defm ATOMNAND : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMNAND">;
-defm ATOMADD  : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMADD">;
-defm ATOMSUB  : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMSUB">;
-defm ATOMMAX  : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMMAX">;
-defm ATOMMIN  : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMMIN">;
-defm ATOMUMAX : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMUMAX">;
-defm ATOMUMIN : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMUMIN">;
-defm ATOMSWAP : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMSWAP">;
-
-//===----------------------------------------------------------------------===//
 // Normal-Instructions-With-Lock-Prefix Pseudo Instructions
 //===----------------------------------------------------------------------===//
 
 // FIXME: Use normal instructions and add lock prefix dynamically.
 
 // Memory barriers
 
 // TODO: Get this to fold the constant into the instruction.

lib/Target/X86/X86InstrInfo.td

def X86cas : SDNode<"X86ISD::LCMPXCHG_DAG", SDTX86cas,
                          SDNPMayLoad, SDNPMemOperand]>;
 def X86cas8 : SDNode<"X86ISD::LCMPXCHG8_DAG", SDTX86caspair,
                         [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPMayStore,
                          SDNPMayLoad, SDNPMemOperand]>;
 def X86cas16 : SDNode<"X86ISD::LCMPXCHG16_DAG", SDTX86caspair,
                         [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPMayStore,
                          SDNPMayLoad, SDNPMemOperand]>;
 
-def X86AtomAdd64 : SDNode<"X86ISD::ATOMADD64_DAG", SDTX86atomicBinary,
-                        [SDNPHasChain, SDNPMayStore,
-                         SDNPMayLoad, SDNPMemOperand]>;
-def X86AtomSub64 : SDNode<"X86ISD::ATOMSUB64_DAG", SDTX86atomicBinary,
-                        [SDNPHasChain, SDNPMayStore,
-                         SDNPMayLoad, SDNPMemOperand]>;
-def X86AtomOr64 : SDNode<"X86ISD::ATOMOR64_DAG", SDTX86atomicBinary,
-                        [SDNPHasChain, SDNPMayStore,
-                         SDNPMayLoad, SDNPMemOperand]>;
-def X86AtomXor64 : SDNode<"X86ISD::ATOMXOR64_DAG", SDTX86atomicBinary,
-                        [SDNPHasChain, SDNPMayStore,
-                         SDNPMayLoad, SDNPMemOperand]>;
-def X86AtomAnd64 : SDNode<"X86ISD::ATOMAND64_DAG", SDTX86atomicBinary,
-                        [SDNPHasChain, SDNPMayStore,
-                         SDNPMayLoad, SDNPMemOperand]>;
-def X86AtomNand64 : SDNode<"X86ISD::ATOMNAND64_DAG", SDTX86atomicBinary,
-                        [SDNPHasChain, SDNPMayStore,
-                         SDNPMayLoad, SDNPMemOperand]>;
-def X86AtomSwap64 : SDNode<"X86ISD::ATOMSWAP64_DAG", SDTX86atomicBinary,
-                        [SDNPHasChain, SDNPMayStore,
-                         SDNPMayLoad, SDNPMemOperand]>;
 def X86retflag : SDNode<"X86ISD::RET_FLAG", SDTX86Ret,
                         [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
 
 def X86vastart_save_xmm_regs :
                  SDNode<"X86ISD::VASTART_SAVE_XMM_REGS",
                         SDT_X86VASTART_SAVE_XMM_REGS,
                         [SDNPHasChain, SDNPVariadic]>;
 def X86vaarg64 :

lib/Target/X86/X86TargetMachine.cpp

public:
   X86TargetMachine &getX86TargetMachine() const {
     return getTM<X86TargetMachine>();
   }
 
   const X86Subtarget &getX86Subtarget() const {
     return *getX86TargetMachine().getSubtargetImpl();
   }
 
+  void addIRPasses() override;
   bool addInstSelector() override;
   bool addILPOpts() override;
   bool addPreRegAlloc() override;
   bool addPostRegAlloc() override;
   bool addPreEmitPass() override;
 };
 } // namespace
 
 TargetPassConfig *X86TargetMachine::createPassConfig(PassManagerBase &PM) {
   return new X86PassConfig(this, PM);
 }
 
+void X86PassConfig::addIRPasses() {
+  addPass(createX86AtomicExpandPass(&getX86TargetMachine()));
+
+  TargetPassConfig::addIRPasses();
+}
+
 bool X86PassConfig::addInstSelector() {
   // Install an instruction selector.
   addPass(createX86ISelDag(getX86TargetMachine(), getOptLevel()));
 
   // For ELF, cleanup any local-dynamic TLS accesses.
   if (getX86Subtarget().isTargetELF() && getOptLevel() != CodeGenOpt::None)
     addPass(createCleanupLocalDynamicTLSPass());
 

test/CodeGen/X86/2010-01-08-Atomic64Bug.ll

 ; RUN: llc < %s -mtriple=i386-apple-darwin -mcpu=corei7 | FileCheck %s
 ; rdar://r7512579
 
 ; PHI defs in the atomic loop should be used by the add / adc
 ; instructions. They should not be dead.
 
 define void @t(i64* nocapture %p) nounwind ssp {
 entry:
 ; CHECK-LABEL: t:
 ; CHECK: movl ([[REG:%[a-z]+]]), %eax
 ; CHECK: movl 4([[REG]]), %edx
 ; CHECK: LBB0_1:
 ; CHECK: movl %eax, %ebx
-; CHECK: addl {{%[a-z]+}}, %ebx
+; CHECK: addl $1, %ebx
 ; CHECK: movl %edx, %ecx
-; CHECK: adcl {{%[a-z]+}}, %ecx
+; CHECK: adcl $0, %ecx
 ; CHECK: lock
 ; CHECK-NEXT: cmpxchg8b ([[REG]])
 ; CHECK-NEXT: jne
   %0 = atomicrmw add i64* %p, i64 1 seq_cst
   ret void
 }

test/CodeGen/X86/Atomics-64.ll

 ; RUN: llc < %s -march=x86-64 > %t.x86-64
-; RUN: llc < %s -march=x86 > %t.x86
+; RUN: llc < %s -march=x86 -mattr=cx16 > %t.x86
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
 target triple = "x86_64-apple-darwin8"
 
 @sc = common global i8 0
 @uc = common global i8 0
 @ss = common global i16 0
 @us = common global i16 0
 @si = common global i32 0

test/CodeGen/X86/atomic-load-store-wide.ll

-; RUN: llc < %s -march=x86 -verify-machineinstrs | FileCheck %s
+; RUN: llc < %s -mcpu=corei7 -march=x86 -verify-machineinstrs | FileCheck %s
 
 ; 64-bit load/store on x86-32
 ; FIXME: The generated code can be substantially improved.
 
 define void @test1(i64* %ptr, i64 %val1) {
 ; CHECK: test1
 ; CHECK: cmpxchg8b
 ; CHECK-NEXT: jne

test/CodeGen/X86/atomic-minmax-i6432.ll

-; RUN: llc -march=x86 -mattr=+cmov -mtriple=i386-pc-linux -verify-machineinstrs < %s | FileCheck %s -check-prefix=LINUX
-; RUN: llc -march=x86 -mattr=-cmov -mtriple=i386-pc-linux -verify-machineinstrs < %s | FileCheck %s -check-prefix=NOCMOV
-; RUN: llc -march=x86 -mtriple=i386-macosx -relocation-model=pic -verify-machineinstrs < %s | FileCheck %s -check-prefix=PIC
+; RUN: llc -march=x86 -mattr=+cmov,cx16 -mtriple=i386-pc-linux -verify-machineinstrs < %s | FileCheck %s -check-prefix=LINUX
+; RUN: llc -march=x86 -mattr=cx16 -mtriple=i386-macosx -relocation-model=pic -verify-machineinstrs < %s | FileCheck %s -check-prefix=PIC
 
 @sc64 = external global i64
 
 define void @atomic_maxmin_i6432() {
 ; LINUX: atomic_maxmin_i6432
   %1 = atomicrmw max  i64* @sc64, i64 5 acquire
 ; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
 ; LINUX: cmpl
-; LINUX: setl
-; LINUX: cmpl
-; LINUX: setl
+; LINUX: seta
 ; LINUX: cmovne
 ; LINUX: cmovne
 ; LINUX: lock
 ; LINUX-NEXT: cmpxchg8b
 ; LINUX: jne [[LABEL]]
-; NOCMOV: [[LABEL:.LBB[0-9]+_[0-9]+]]
-; NOCMOV: cmpl
-; NOCMOV: setl
-; NOCMOV: cmpl
-; NOCMOV: setl
-; NOCMOV: jne
-; NOCMOV: jne
-; NOCMOV: lock
-; NOCMOV-NEXT: cmpxchg8b
-; NOCMOV: jne [[LABEL]]
   %2 = atomicrmw min  i64* @sc64, i64 6 acquire
 ; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
 ; LINUX: cmpl
-; LINUX: setg
-; LINUX: cmpl
-; LINUX: setg
+; LINUX: setb
 ; LINUX: cmovne
 ; LINUX: cmovne
 ; LINUX: lock
 ; LINUX-NEXT: cmpxchg8b
 ; LINUX: jne [[LABEL]]
-; NOCMOV: [[LABEL:.LBB[0-9]+_[0-9]+]]
-; NOCMOV: cmpl
-; NOCMOV: setg
-; NOCMOV: cmpl
-; NOCMOV: setg
-; NOCMOV: jne
-; NOCMOV: jne
-; NOCMOV: lock
-; NOCMOV-NEXT: cmpxchg8b
-; NOCMOV: jne [[LABEL]]
   %3 = atomicrmw umax i64* @sc64, i64 7 acquire
 ; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
 ; LINUX: cmpl
-; LINUX: setb
-; LINUX: cmpl
-; LINUX: setb
+; LINUX: seta
 ; LINUX: cmovne
 ; LINUX: cmovne
 ; LINUX: lock
 ; LINUX-NEXT: cmpxchg8b
 ; LINUX: jne [[LABEL]]
-; NOCMOV: [[LABEL:.LBB[0-9]+_[0-9]+]]
-; NOCMOV: cmpl
-; NOCMOV: setb
-; NOCMOV: cmpl
-; NOCMOV: setb
-; NOCMOV: jne
-; NOCMOV: jne
-; NOCMOV: lock
-; NOCMOV-NEXT: cmpxchg8b
-; NOCMOV: jne [[LABEL]]
   %4 = atomicrmw umin i64* @sc64, i64 8 acquire
 ; LINUX: [[LABEL:.LBB[0-9]+_[0-9]+]]
 ; LINUX: cmpl
-; LINUX: seta
-; LINUX: cmpl
-; LINUX: seta
+; LINUX: setb
 ; LINUX: cmovne
 ; LINUX: cmovne
 ; LINUX: lock
 ; LINUX-NEXT: cmpxchg8b
 ; LINUX: jne [[LABEL]]
-; NOCMOV: [[LABEL:.LBB[0-9]+_[0-9]+]]
-; NOCMOV: cmpl
-; NOCMOV: seta
-; NOCMOV: cmpl
-; NOCMOV: seta
-; NOCMOV: jne
-; NOCMOV: jne
-; NOCMOV: lock
-; NOCMOV-NEXT: cmpxchg8b
-; NOCMOV: jne [[LABEL]]
   ret void
 }
 
 ; rdar://12453106
 @id = internal global i64 0, align 8
 
 define void @tf_bug(i8* %ptr) nounwind {
 ; PIC-LABEL: tf_bug:
-; PIC: movl _id-L1$pb(
-; PIC: movl (_id-L1$pb)+4(
+; PIC-DAG: movl _id-L1$pb(
+; PIC-DAG: movl (_id-L1$pb)+4(
   %tmp1 = atomicrmw add i64* @id, i64 1 seq_cst
   %tmp2 = add i64 %tmp1, 1
   %tmp3 = bitcast i8* %ptr to i64*
   store i64 %tmp2, i64* %tmp3, align 4
   ret void
 }

test/CodeGen/X86/atomic128.ll

+; RUN: llc < %s -march=x86-64 -verify-machineinstrs -mattr=cx16 | FileCheck %s
+
+@var = global i128 0
+
+define i128 @val_compare_and_swap(i128* %p, i128 %oldval, i128 %newval) {
+; CHECK-LABEL: val_compare_and_swap:
+; CHECK: movq %rsi, %rax
+; CHECK: movq %rcx, %rbx
+; CHECK: movq %r8, %rcx
+; CHECK: lock
+; CHECK: cmpxchg16b (%rdi)
+
+  %pair = cmpxchg i128* %p, i128 %oldval, i128 %newval acquire acquire
+  %val = extractvalue { i128, i1 } %pair, 0
+  ret i128 %val
+}
+
+define void @fetch_and_nand(i128* %p, i128 %bits) {
+; CHECK-LABEL: fetch_and_nand:
+; CHECK-DAG:     movq %rdx, [[INCHI:%[a-z0-9]+]]
+; CHECK-DAG:     movq (%rdi), %rax
+; CHECK-DAG:     movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         movq %rdx, %rcx
+; CHECK:         andq [[INCHI]], %rcx
+; CHECK:         movq %rax, %rbx
+  ; INCLO equivalent comes in in %rsi, so it makes sense it stays there.
+; CHECK:         andq %rsi, %rbx
+; CHECK:         notq %rbx
+; CHECK:         notq %rcx
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+; CHECK:         movq %rax, _var
+; CHECK:         movq %rdx, _var+8
+  %val = atomicrmw nand i128* %p, i128 %bits release
+  store i128 %val, i128* @var, align 16
+  ret void
+}
+
+define void @fetch_and_or(i128* %p, i128 %bits) {
+; CHECK-LABEL: fetch_and_or:
+; CHECK-DAG:     movq %rdx, [[INCHI:%[a-z0-9]+]]
+; CHECK-DAG:     movq (%rdi), %rax
+; CHECK-DAG:     movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         movq %rax, %rbx
+  ; INCLO equivalent comes in in %rsi, so it makes sense it stays there.
+; CHECK:         orq %rsi, %rbx
+; CHECK:         movq %rdx, %rcx
+; CHECK:         orq [[INCHI]], %rcx
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+; CHECK:         movq %rax, _var
+; CHECK:         movq %rdx, _var+8
+
+  %val = atomicrmw or i128* %p, i128 %bits seq_cst
+  store i128 %val, i128* @var, align 16
+  ret void
+}
+
+define void @fetch_and_add(i128* %p, i128 %bits) {
+; CHECK-LABEL: fetch_and_add:
+; CHECK-DAG:     movq %rdx, [[INCHI:%[a-z0-9]+]]
+; CHECK-DAG:     movq (%rdi), %rax
+; CHECK-DAG:     movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         movq %rax, %rbx
+  ; INCLO equivalent comes in in %rsi, so it makes sense it stays there.
+; CHECK:         addq %rsi, %rbx
+; CHECK:         movq %rdx, %rcx
+; CHECK:         adcq [[INCHI]], %rcx
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+; CHECK:         movq %rax, _var
+; CHECK:         movq %rdx, _var+8
+
+  %val = atomicrmw add i128* %p, i128 %bits seq_cst
+  store i128 %val, i128* @var, align 16
+  ret void
+}
+
+define void @fetch_and_sub(i128* %p, i128 %bits) {
+; CHECK-LABEL: fetch_and_sub:
+; CHECK-DAG:     movq %rdx, [[INCHI:%[a-z0-9]+]]
+; CHECK-DAG:     movq (%rdi), %rax
+; CHECK-DAG:     movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         movq %rax, %rbx
+  ; INCLO equivalent comes in in %rsi, so it makes sense it stays there.
+; CHECK:         subq %rsi, %rbx
+; CHECK:         movq %rdx, %rcx
+; CHECK:         sbbq [[INCHI]], %rcx
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+; CHECK:         movq %rax, _var
+; CHECK:         movq %rdx, _var+8
+
+  %val = atomicrmw sub i128* %p, i128 %bits seq_cst
+  store i128 %val, i128* @var, align 16
+  ret void
+}
+
+define void @fetch_and_min(i128* %p, i128 %bits) {
+; CHECK-LABEL: fetch_and_min:
+; CHECK-DAG:     movq %rdx, [[INCHI:%[a-z0-9]+]]
+; CHECK-DAG:     movq (%rdi), %rax
+; CHECK-DAG:     movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         cmpq %rsi, %rax
+; CHECK:         setbe [[CMP:%[a-z0-9]+]]
+; CHECK:         cmpq [[INCHI]], %rdx
+; CHECK:         setle [[HICMP:%[a-z0-9]+]]
+; CHECK:         je [[USE_LO:.?LBB[0-9]+_[0-9]+]]
+
+; CHECK:         movb [[HICMP]], [[CMP]]
+; CHECK: [[USE_LO]]:
+; CHECK:         testb [[CMP]], [[CMP]]
+; CHECK:         movq %rsi, %rbx
+; CHECK:         cmovneq %rax, %rbx
+; CHECK:         movq [[INCHI]], %rcx
+; CHECK:         cmovneq %rdx, %rcx
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+; CHECK:         movq %rax, _var
+; CHECK:         movq %rdx, _var+8
+
+  %val = atomicrmw min i128* %p, i128 %bits seq_cst
+  store i128 %val, i128* @var, align 16
+  ret void
+}
+
+define void @fetch_and_max(i128* %p, i128 %bits) {
+; CHECK-LABEL: fetch_and_max:
+; CHECK-DAG:     movq %rdx, [[INCHI:%[a-z0-9]+]]
+; CHECK-DAG:     movq (%rdi), %rax
+; CHECK-DAG:     movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         cmpq %rsi, %rax
+; CHECK:         setae [[CMP:%[a-z0-9]+]]
+; CHECK:         cmpq [[INCHI]], %rdx
+; CHECK:         setge [[HICMP:%[a-z0-9]+]]
+; CHECK:         je [[USE_LO:.?LBB[0-9]+_[0-9]+]]
+
+; CHECK:         movb [[HICMP]], [[CMP]]
+; CHECK: [[USE_LO]]:
+; CHECK:         testb [[CMP]], [[CMP]]
+; CHECK:         movq %rsi, %rbx
+; CHECK:         cmovneq %rax, %rbx
+; CHECK:         movq [[INCHI]], %rcx
+; CHECK:         cmovneq %rdx, %rcx
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+; CHECK:         movq %rax, _var
+; CHECK:         movq %rdx, _var+8
+
+  %val = atomicrmw max i128* %p, i128 %bits seq_cst
+  store i128 %val, i128* @var, align 16
+  ret void
+}
+
+define void @fetch_and_umin(i128* %p, i128 %bits) {
+; CHECK-LABEL: fetch_and_umin:
+; CHECK-DAG:     movq %rdx, [[INCHI:%[a-z0-9]+]]
+; CHECK-DAG:     movq (%rdi), %rax
+; CHECK-DAG:     movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         cmpq %rsi, %rax
+; CHECK:         setbe [[CMP:%[a-z0-9]+]]
+; CHECK:         cmpq [[INCHI]], %rdx
+; CHECK:         setbe [[HICMP:%[a-z0-9]+]]
+; CHECK:         je [[USE_LO:.?LBB[0-9]+_[0-9]+]]
+
+; CHECK:         movb [[HICMP]], [[CMP]]
+; CHECK: [[USE_LO]]:
+; CHECK:         testb [[CMP]], [[CMP]]
+; CHECK:         movq %rsi, %rbx
+; CHECK:         cmovneq %rax, %rbx
+; CHECK:         movq [[INCHI]], %rcx
+; CHECK:         cmovneq %rdx, %rcx
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+; CHECK:         movq %rax, _var
+; CHECK:         movq %rdx, _var+8
+
+  %val = atomicrmw umin i128* %p, i128 %bits seq_cst
+  store i128 %val, i128* @var, align 16
+  ret void
+}
+
+define void @fetch_and_umax(i128* %p, i128 %bits) {
+; CHECK-LABEL: fetch_and_umax:
+; CHECK-DAG:     movq %rdx, [[INCHI:%[a-z0-9]+]]
+; CHECK-DAG:     movq (%rdi), %rax
+; CHECK-DAG:     movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         cmpq %rax, %rsi
+; CHECK:         setb [[CMP:%[a-z0-9]+]]
+; CHECK:         cmpq [[INCHI]], %rdx
+; CHECK:         seta [[HICMP:%[a-z0-9]+]]
+; CHECK:         je [[USE_LO:.?LBB[0-9]+_[0-9]+]]
+
+; CHECK:         movb [[HICMP]], [[CMP]]
+; CHECK: [[USE_LO]]:
+; CHECK:         testb [[CMP]], [[CMP]]
+; CHECK:         movq %rsi, %rbx
+; CHECK:         cmovneq %rax, %rbx
+; CHECK:         movq [[INCHI]], %rcx
+; CHECK:         cmovneq %rdx, %rcx
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+; CHECK:         movq %rax, _var
+; CHECK:         movq %rdx, _var+8
+
+  %val = atomicrmw umax i128* %p, i128 %bits seq_cst
+  store i128 %val, i128* @var, align 16
+  ret void
+}
+
+define i128 @atomic_load_seq_cst(i128* %p) {
+; CHECK-LABEL: atomic_load_seq_cst:
+; CHECK: xorl %eax, %eax
+; CHECK: xorl %edx, %edx
+; CHECK: xorl %ebx, %ebx
+; CHECK: xorl %ecx, %ecx
+; CHECK: lock
+; CHECK: cmpxchg16b (%rdi)
+
+   %r = load atomic i128* %p seq_cst, align 16
+   ret i128 %r
+}
+
+define i128 @atomic_load_relaxed(i128* %p) {
+; CHECK: atomic_load_relaxed:
+; CHECK: xorl %eax, %eax
+; CHECK: xorl %edx, %edx
+; CHECK: xorl %ebx, %ebx
+; CHECK: xorl %ecx, %ecx
+; CHECK: lock
+; CHECK: cmpxchg16b (%rdi)
+
+   %r = load atomic i128* %p monotonic, align 16
+   ret i128 %r
+}
+
+define void @atomic_store_seq_cst(i128* %p, i128 %in) {
+; CHECK-LABEL: atomic_store_seq_cst:
+; CHECK:         movq %rdx, %rcx
+; CHECK:         movq %rsi, %rbx
+; CHECK:         movq (%rdi), %rax
+; CHECK:         movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+   store atomic i128 %in, i128* %p seq_cst, align 16
+   ret void
+}
+
+define void @atomic_store_release(i128* %p, i128 %in) {
+; CHECK-LABEL: atomic_store_release:
+; CHECK:         movq %rdx, %rcx
+; CHECK:         movq %rsi, %rbx
+; CHECK:         movq (%rdi), %rax
+; CHECK:         movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+   store atomic i128 %in, i128* %p release, align 16
+   ret void
+}
+
+define void @atomic_store_relaxed(i128* %p, i128 %in) {
+; CHECK-LABEL: atomic_store_relaxed:
+; CHECK:         movq %rdx, %rcx
+; CHECK:         movq %rsi, %rbx
+; CHECK:         movq (%rdi), %rax
+; CHECK:         movq 8(%rdi), %rdx
+
+; CHECK: [[LOOP:.?LBB[0-9]+_[0-9]+]]:
+; CHECK:         lock
+; CHECK:         cmpxchg16b (%rdi)
+; CHECK:         jne [[LOOP]]
+
+   store atomic i128 %in, i128* %p unordered, align 16
+   ret void
+}

test/CodeGen/X86/atomic16.ll

 ; RUN: llc < %s -O0 -mtriple=x86_64-unknown-unknown -mcpu=corei7 -verify-machineinstrs -show-mc-encoding | FileCheck %s --check-prefix X64
 ; RUN: llc < %s -O0 -mtriple=i386-unknown-unknown -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X32
 
 @sc16 = external global i16
 
 define void @atomic_fetch_add16() nounwind {
-; X64:   atomic_fetch_add16
-; X32:   atomic_fetch_add16
+; X64-LABEL:   atomic_fetch_add16
+; X32-LABEL:   atomic_fetch_add16
 entry:
 ; 32-bit
   %t1 = atomicrmw add  i16* @sc16, i16 1 acquire
 ; X64:       lock
 ; X64:       incw
 ; X32:       lock
 ; X32:       incw
   %t2 = atomicrmw add  i16* @sc16, i16 3 acquire

 ; X32:       lock
 ; X32:       addw
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_sub16() nounwind {
-; X64:   atomic_fetch_sub16
-; X32:   atomic_fetch_sub16
+; X64-LABEL:   atomic_fetch_sub16
+; X32-LABEL:   atomic_fetch_sub16
   %t1 = atomicrmw sub  i16* @sc16, i16 1 acquire
 ; X64:       lock
 ; X64:       decw
 ; X32:       lock
 ; X32:       decw
   %t2 = atomicrmw sub  i16* @sc16, i16 3 acquire
 ; X64:       lock
 ; X64:       subw $3, {{.*}} # encoding: [0xf0,0x66

 ; X32:       lock
 ; X32:       subw
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_and16() nounwind {
-; X64:   atomic_fetch_and16
-; X32:   atomic_fetch_and16
+; X64-LABEL:   atomic_fetch_and16
+; X32-LABEL:   atomic_fetch_and16
   %t1 = atomicrmw and  i16* @sc16, i16 3 acquire
 ; X64:       lock
 ; X64:       andw $3, {{.*}} # encoding: [0xf0,0x66
 ; X32:       lock
 ; X32:       andw $3
   %t2 = atomicrmw and  i16* @sc16, i16 5 acquire
-; X64:       andw
+; X64:       andl
 ; X64:       lock
 ; X64:       cmpxchgw
-; X32:       andw
+; X32:       andl
 ; X32:       lock
 ; X32:       cmpxchgw
   %t3 = atomicrmw and  i16* @sc16, i16 %t2 acquire
 ; X64:       lock
 ; X64:       andw {{.*}} # encoding: [0xf0,0x66
 ; X32:       lock
 ; X32:       andw
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_or16() nounwind {
-; X64:   atomic_fetch_or16
-; X32:   atomic_fetch_or16
+; X64-LABEL:   atomic_fetch_or16
+; X32-LABEL:   atomic_fetch_or16
   %t1 = atomicrmw or   i16* @sc16, i16 3 acquire
 ; X64:       lock
 ; X64:       orw $3, {{.*}} # encoding: [0xf0,0x66
 ; X32:       lock
 ; X32:       orw $3
   %t2 = atomicrmw or   i16* @sc16, i16 5 acquire
-; X64:       orw
+; X64:       orl
 ; X64:       lock
 ; X64:       cmpxchgw
-; X32:       orw
+; X32:       orl
 ; X32:       lock
 ; X32:       cmpxchgw
   %t3 = atomicrmw or   i16* @sc16, i16 %t2 acquire
 ; X64:       lock
 ; X64:       orw {{.*}} # encoding: [0xf0,0x66
 ; X32:       lock
 ; X32:       orw
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_xor16() nounwind {
-; X64:   atomic_fetch_xor16
-; X32:   atomic_fetch_xor16
+; X64-LABEL:   atomic_fetch_xor16
+; X32-LABEL:   atomic_fetch_xor16
   %t1 = atomicrmw xor  i16* @sc16, i16 3 acquire
 ; X64:       lock
 ; X64:       xorw $3, {{.*}} # encoding: [0xf0,0x66
 ; X32:       lock
 ; X32:       xorw $3
   %t2 = atomicrmw xor  i16* @sc16, i16 5 acquire
-; X64:       xorw
+; X64:       xorl
 ; X64:       lock
 ; X64:       cmpxchgw
-; X32:       xorw
+; X32:       xorl
 ; X32:       lock
 ; X32:       cmpxchgw
   %t3 = atomicrmw xor  i16* @sc16, i16 %t2 acquire
 ; X64:       lock
 ; X64:       xorw {{.*}} # encoding: [0xf0,0x66
 ; X32:       lock
 ; X32:       xorw
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_nand16(i16 %x) nounwind {
-; X64:   atomic_fetch_nand16
-; X32:   atomic_fetch_nand16
+; X64-LABEL:   atomic_fetch_nand16
+; X32-LABEL:   atomic_fetch_nand16
   %t1 = atomicrmw nand i16* @sc16, i16 %x acquire
-; X64:       andw
-; X64:       notw
+; X64:       andl
+; X64:       notl
 ; X64:       lock
 ; X64:       cmpxchgw
-; X32:       andw
-; X32:       notw
+; X32:       andl
+; X32:       notl
 ; X32:       lock
 ; X32:       cmpxchgw
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_max16(i16 %x) nounwind {
   %t1 = atomicrmw max  i16* @sc16, i16 %x acquire
-; X64:       cmpw
+; X64:       movswl
+; X64:       movswl
+; X64:       subl
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgw
 
-; X32:       cmpw
+; X32:       movswl
+; X32:       movswl
+; X32:       subl
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgw
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_min16(i16 %x) nounwind {
   %t1 = atomicrmw min  i16* @sc16, i16 %x acquire
-; X64:       cmpw
+; X64:       movswl
+; X64:       movswl
+; X64:       subl
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgw
 
-; X32:       cmpw
+; X32:       movswl
+; X32:       movswl
+; X32:       subl
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgw
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_umax16(i16 %x) nounwind {
   %t1 = atomicrmw umax i16* @sc16, i16 %x acquire
-; X64:       cmpw
+; X64:       movzwl
+; X64:       movzwl
+; X64:       subl
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgw
 
-; X32:       cmpw
+; X32:       movzwl
+; X32:       movzwl
+; X32:       subl
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgw
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_umin16(i16 %x) nounwind {
   %t1 = atomicrmw umin i16* @sc16, i16 %x acquire
-; X64:       cmpw
+; X64:       movzwl
+; X64:       movzwl
+; X64:       subl
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgw
-; X32:       cmpw
+
+; X32:       movzwl
+; X32:       movzwl
+; X32:       subl
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgw
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 

test/CodeGen/X86/atomic32.ll

 ; RUN: llc < %s -O0 -march=x86-64 -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X64
 ; RUN: llc < %s -O0 -march=x86 -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X32
 ; RUN: llc < %s -O0 -march=x86 -mcpu=corei7 -mattr=-cmov -verify-machineinstrs | FileCheck %s --check-prefix NOCMOV
 
 @sc32 = external global i32
 
 define void @atomic_fetch_add32() nounwind {
-; X64:   atomic_fetch_add32
-; X32:   atomic_fetch_add32
+; X64-LABEL:   atomic_fetch_add32:
+; X32-LABEL:   atomic_fetch_add32:
 entry:
 ; 32-bit
   %t1 = atomicrmw add  i32* @sc32, i32 1 acquire
 ; X64:       lock
 ; X64:       incl
 ; X32:       lock
 ; X32:       incl
   %t2 = atomicrmw add  i32* @sc32, i32 3 acquire

 ; X32:       lock
 ; X32:       addl
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_sub32() nounwind {
-; X64:   atomic_fetch_sub32
-; X32:   atomic_fetch_sub32
+; X64-LABEL:   atomic_fetch_sub32:
+; X32-LABEL:   atomic_fetch_sub32:
   %t1 = atomicrmw sub  i32* @sc32, i32 1 acquire
 ; X64:       lock
 ; X64:       decl
 ; X32:       lock
 ; X32:       decl
   %t2 = atomicrmw sub  i32* @sc32, i32 3 acquire
 ; X64:       lock
 ; X64:       subl $3

 ; X32:       lock
 ; X32:       subl
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_and32() nounwind {
-; X64:   atomic_fetch_and32
-; X32:   atomic_fetch_and32
+; X64-LABEL:   atomic_fetch_and32:
+; X32-LABEL:   atomic_fetch_and32:
   %t1 = atomicrmw and  i32* @sc32, i32 3 acquire
 ; X64:       lock
 ; X64:       andl $3
 ; X32:       lock
 ; X32:       andl $3
   %t2 = atomicrmw and  i32* @sc32, i32 5 acquire
 ; X64:       andl
 ; X64:       lock
 ; X64:       cmpxchgl
 ; X32:       andl
 ; X32:       lock
 ; X32:       cmpxchgl
   %t3 = atomicrmw and  i32* @sc32, i32 %t2 acquire
 ; X64:       lock
 ; X64:       andl
 ; X32:       lock
 ; X32:       andl
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_or32() nounwind {
-; X64:   atomic_fetch_or32
-; X32:   atomic_fetch_or32
+; X64-LABEL:   atomic_fetch_or32:
+; X32-LABEL:   atomic_fetch_or32:
   %t1 = atomicrmw or   i32* @sc32, i32 3 acquire
 ; X64:       lock
 ; X64:       orl $3
 ; X32:       lock
 ; X32:       orl $3
   %t2 = atomicrmw or   i32* @sc32, i32 5 acquire
 ; X64:       orl
 ; X64:       lock
 ; X64:       cmpxchgl
 ; X32:       orl
 ; X32:       lock
 ; X32:       cmpxchgl
   %t3 = atomicrmw or   i32* @sc32, i32 %t2 acquire
 ; X64:       lock
 ; X64:       orl
 ; X32:       lock
 ; X32:       orl
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_xor32() nounwind {
-; X64:   atomic_fetch_xor32
-; X32:   atomic_fetch_xor32
+; X64-LABEL:   atomic_fetch_xor32:
+; X32-LABEL:   atomic_fetch_xor32:
   %t1 = atomicrmw xor  i32* @sc32, i32 3 acquire
 ; X64:       lock
 ; X64:       xorl $3
 ; X32:       lock
 ; X32:       xorl $3
   %t2 = atomicrmw xor  i32* @sc32, i32 5 acquire
 ; X64:       xorl
 ; X64:       lock
 ; X64:       cmpxchgl
 ; X32:       xorl
 ; X32:       lock
 ; X32:       cmpxchgl
   %t3 = atomicrmw xor  i32* @sc32, i32 %t2 acquire
 ; X64:       lock
 ; X64:       xorl
 ; X32:       lock
 ; X32:       xorl
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_nand32(i32 %x) nounwind {
-; X64:   atomic_fetch_nand32
-; X32:   atomic_fetch_nand32
+; X64-LABEL:   atomic_fetch_nand32:
+; X32-LABEL:   atomic_fetch_nand32:
   %t1 = atomicrmw nand i32* @sc32, i32 %x acquire
 ; X64:       andl
 ; X64:       notl
 ; X64:       lock
 ; X64:       cmpxchgl
 ; X32:       andl
 ; X32:       notl
 ; X32:       lock
 ; X32:       cmpxchgl
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_max32(i32 %x) nounwind {
+; X64-LABEL: atomic_fetch_max32:
+; X32-LABEL: atomic_fetch_max32:
+
   %t1 = atomicrmw max  i32* @sc32, i32 %x acquire
-; X64:       cmpl
+; X64:       subl
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgl
 
-; X32:       cmpl
+; X32:       subl
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgl
 
-; NOCMOV:    cmpl
-; NOCMOV:    jl
+; NOCMOV:    subl
+; NOCMOV:    jge
 ; NOCMOV:    lock
 ; NOCMOV:    cmpxchgl
   ret void
 ; X64:       ret
 ; X32:       ret
 ; NOCMOV:    ret
 }
 
 define void @atomic_fetch_min32(i32 %x) nounwind {
+; X64-LABEL: atomic_fetch_min32:
+; X32-LABEL: atomic_fetch_min32:
+; NOCMOV-LABEL: atomic_fetch_min32:
+
   %t1 = atomicrmw min  i32* @sc32, i32 %x acquire
-; X64:       cmpl
+; X64:       subl
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgl
 
-; X32:       cmpl
+; X32:       subl
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgl
 
-; NOCMOV:    cmpl
-; NOCMOV:    jg
+; NOCMOV:    subl
+; NOCMOV:    jle
 ; NOCMOV:    lock
 ; NOCMOV:    cmpxchgl
   ret void
 ; X64:       ret
 ; X32:       ret
 ; NOCMOV:    ret
 }
 
 define void @atomic_fetch_umax32(i32 %x) nounwind {
+; X64-LABEL: atomic_fetch_umax32:
+; X32-LABEL: atomic_fetch_umax32:
+; NOCMOV-LABEL: atomic_fetch_umax32:
+
   %t1 = atomicrmw umax i32* @sc32, i32 %x acquire
-; X64:       cmpl
+; X64:       subl
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgl
 
-; X32:       cmpl
+; X32:       subl
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgl
 
-; NOCMOV:    cmpl
-; NOCMOV:    jb
+; NOCMOV:    subl
+; NOCMOV:    ja
 ; NOCMOV:    lock
 ; NOCMOV:    cmpxchgl
   ret void
 ; X64:       ret
 ; X32:       ret
 ; NOCMOV:    ret
 }
 
 define void @atomic_fetch_umin32(i32 %x) nounwind {
+; X64-LABEL: atomic_fetch_umin32:
+; X32-LABEL: atomic_fetch_umin32:
+; NOCMOV-LABEL: atomic_fetch_umin32:
+
   %t1 = atomicrmw umin i32* @sc32, i32 %x acquire
-; X64:       cmpl
+; X64:       subl
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgl
 
-; X32:       cmpl
+; X32:       subl
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchgl
 
-; NOCMOV:    cmpl
-; NOCMOV:    ja
+; NOCMOV:    subl
+; NOCMOV:    jb
 ; NOCMOV:    lock
 ; NOCMOV:    cmpxchgl
   ret void
 ; X64:       ret
 ; X32:       ret
 ; NOCMOV:    ret
 }
 
 define void @atomic_fetch_cmpxchg32() nounwind {
+; X64-LABEL: atomic_fetch_cmpxchg32:
+; X32-LABEL: atomic_fetch_cmpxchg32:
+
   %t1 = cmpxchg i32* @sc32, i32 0, i32 1 acquire acquire
 ; X64:       lock
 ; X64:       cmpxchgl
 ; X32:       lock
 ; X32:       cmpxchgl
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_store32(i32 %x) nounwind {
+; X64-LABEL: atomic_fetch_store32:
+; X32-LABEL: atomic_fetch_store32:
+
   store atomic i32 %x, i32* @sc32 release, align 4
 ; X64-NOT:   lock
 ; X64:       movl
 ; X32-NOT:   lock
 ; X32:       movl
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_swap32(i32 %x) nounwind {
+; X64-LABEL: atomic_fetch_swap32:
+; X32-LABEL: atomic_fetch_swap32:
+
   %t1 = atomicrmw xchg i32* @sc32, i32 %x acquire
 ; X64-NOT:   lock
 ; X64:       xchgl
 ; X32-NOT:   lock
 ; X32:       xchgl
   ret void
 ; X64:       ret
 ; X32:       ret
 }

test/CodeGen/X86/atomic64.ll

 ; RUN: llc < %s -O0 -march=x86-64 -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X64
 
 @sc64 = external global i64
 
 define void @atomic_fetch_add64() nounwind {
-; X64:   atomic_fetch_add64
+; X64-LABEL:   atomic_fetch_add64:
+; X32-LABEL:   atomic_fetch_add64:
 entry:
   %t1 = atomicrmw add  i64* @sc64, i64 1 acquire
 ; X64:       lock
 ; X64:       incq
   %t2 = atomicrmw add  i64* @sc64, i64 3 acquire
 ; X64:       lock
 ; X64:       addq $3
   %t3 = atomicrmw add  i64* @sc64, i64 5 acquire
 ; X64:       lock
 ; X64:       xaddq
   %t4 = atomicrmw add  i64* @sc64, i64 %t3 acquire
 ; X64:       lock
 ; X64:       addq
   ret void
 ; X64:       ret
 }
 
 define void @atomic_fetch_sub64() nounwind {
-; X64:   atomic_fetch_sub64
+; X64-LABEL:   atomic_fetch_sub64:
+; X32-LABEL:   atomic_fetch_sub64:
   %t1 = atomicrmw sub  i64* @sc64, i64 1 acquire
 ; X64:       lock
 ; X64:       decq
   %t2 = atomicrmw sub  i64* @sc64, i64 3 acquire
 ; X64:       lock
 ; X64:       subq $3
   %t3 = atomicrmw sub  i64* @sc64, i64 5 acquire
 ; X64:       lock
 ; X64:       xaddq
   %t4 = atomicrmw sub  i64* @sc64, i64 %t3 acquire
 ; X64:       lock
 ; X64:       subq
   ret void
 ; X64:       ret
 }
 
 define void @atomic_fetch_and64() nounwind {
-; X64:   atomic_fetch_and64
+; X64-LABEL:   atomic_fetch_and64:
+; X32-LABEL:   atomic_fetch_and64:
   %t1 = atomicrmw and  i64* @sc64, i64 3 acquire
 ; X64:       lock
 ; X64:       andq $3
   %t2 = atomicrmw and  i64* @sc64, i64 5 acquire
 ; X64:       andq
 ; X64:       lock
 ; X64:       cmpxchgq
   %t3 = atomicrmw and  i64* @sc64, i64 %t2 acquire
 ; X64:       lock
 ; X64:       andq
   ret void
 ; X64:       ret
 }
 
 define void @atomic_fetch_or64() nounwind {
-; X64:   atomic_fetch_or64
+; X64-LABEL:   atomic_fetch_or64:
+; X32-LABEL:   atomic_fetch_or64:
   %t1 = atomicrmw or   i64* @sc64, i64 3 acquire
 ; X64:       lock
 ; X64:       orq $3
   %t2 = atomicrmw or   i64* @sc64, i64 5 acquire
 ; X64:       orq
 ; X64:       lock
 ; X64:       cmpxchgq
   %t3 = atomicrmw or   i64* @sc64, i64 %t2 acquire
 ; X64:       lock
 ; X64:       orq
   ret void
 ; X64:       ret
 }
 
 define void @atomic_fetch_xor64() nounwind {
-; X64:   atomic_fetch_xor64
+; X64-LABEL:   atomic_fetch_xor64:
+; X32-LABEL:   atomic_fetch_xor64:
   %t1 = atomicrmw xor  i64* @sc64, i64 3 acquire
 ; X64:       lock
 ; X64:       xorq $3
   %t2 = atomicrmw xor  i64* @sc64, i64 5 acquire
 ; X64:       xorq
 ; X64:       lock
 ; X64:       cmpxchgq
   %t3 = atomicrmw xor  i64* @sc64, i64 %t2 acquire
 ; X64:       lock
 ; X64:       xorq
   ret void
 ; X64:       ret
 }
 
 define void @atomic_fetch_nand64(i64 %x) nounwind {
-; X64:   atomic_fetch_nand64
-; X32:   atomic_fetch_nand64
+; X64-LABEL:   atomic_fetch_nand64:
+; X32-LABEL:   atomic_fetch_nand64:
   %t1 = atomicrmw nand i64* @sc64, i64 %x acquire
 ; X64:       andq
 ; X64:       notq
 ; X64:       lock
 ; X64:       cmpxchgq
 ; X32:       andl
 ; X32:       andl
 ; X32:       notl
 ; X32:       notl
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_max64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_max64:
+; X32-LABEL:   atomic_fetch_max64:
   %t1 = atomicrmw max  i64* @sc64, i64 %x acquire
-; X64:       cmpq
+; X64:       subq
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgq
 
 ; X32:       cmpl
 ; X32:       cmpl
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_min64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_min64:
+; X32-LABEL:   atomic_fetch_min64:
   %t1 = atomicrmw min  i64* @sc64, i64 %x acquire
-; X64:       cmpq
+; X64:       subq
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgq
 
 ; X32:       cmpl
 ; X32:       cmpl
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_umax64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_umax64:
+; X32-LABEL:   atomic_fetch_umax64:
   %t1 = atomicrmw umax i64* @sc64, i64 %x acquire
-; X64:       cmpq
+; X64:       subq
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgq
 
 ; X32:       cmpl
 ; X32:       cmpl
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_umin64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_umin64:
+; X32-LABEL:   atomic_fetch_umin64:
   %t1 = atomicrmw umin i64* @sc64, i64 %x acquire
-; X64:       cmpq
+; X64:       subq
 ; X64:       cmov
 ; X64:       lock
 ; X64:       cmpxchgq
 
 ; X32:       cmpl
 ; X32:       cmpl
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_cmpxchg64() nounwind {
+; X64-LABEL:   atomic_fetch_cmpxchg64:
+; X32-LABEL:   atomic_fetch_cmpxchg64:
   %t1 = cmpxchg i64* @sc64, i64 0, i64 1 acquire acquire
 ; X64:       lock
 ; X64:       cmpxchgq
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_store64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_store64:
+; X32-LABEL:   atomic_fetch_store64:
   store atomic i64 %x, i64* @sc64 release, align 8
 ; X64-NOT:   lock
 ; X64:       movq
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_swap64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_swap64:
+; X32-LABEL:   atomic_fetch_swap64:
   %t1 = atomicrmw xchg i64* @sc64, i64 %x acquire
 ; X64-NOT:   lock
 ; X64:       xchgq
 ; X32:       lock
 ; X32:       xchg8b
   ret void
 ; X64:       ret
 ; X32:       ret
 }

test/CodeGen/X86/atomic6432.ll

 ; RUN: llc < %s -O0 -march=x86 -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X32
 
 @sc64 = external global i64
 
 define void @atomic_fetch_add64() nounwind {
-; X32:   atomic_fetch_add64
+; X64-LABEL:   atomic_fetch_add64:
+; X32-LABEL:   atomic_fetch_add64:
 entry:
   %t1 = atomicrmw add  i64* @sc64, i64 1 acquire
 ; X32:       addl
 ; X32:       adcl
 ; X32:       lock
 ; X32:       cmpxchg8b
   %t2 = atomicrmw add  i64* @sc64, i64 3 acquire
 ; X32:       addl

 ; X32:       adcl
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_sub64() nounwind {
-; X32:   atomic_fetch_sub64
+; X64-LABEL:   atomic_fetch_sub64:
+; X32-LABEL:   atomic_fetch_sub64:
   %t1 = atomicrmw sub  i64* @sc64, i64 1 acquire
-; X32:       subl
-; X32:       sbbl
+; X32:       addl $-1
+; X32:       adcl $-1
 ; X32:       lock
 ; X32:       cmpxchg8b
   %t2 = atomicrmw sub  i64* @sc64, i64 3 acquire
-; X32:       subl
-; X32:       sbbl
+; X32:       addl $-3
+; X32:       adcl $-1
 ; X32:       lock
 ; X32:       cmpxchg8b
   %t3 = atomicrmw sub  i64* @sc64, i64 5 acquire
-; X32:       subl
-; X32:       sbbl
+; X32:       addl $-5
+; X32:       adcl $-1
 ; X32:       lock
 ; X32:       cmpxchg8b
   %t4 = atomicrmw sub  i64* @sc64, i64 %t3 acquire
 ; X32:       subl
 ; X32:       sbbl
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_and64() nounwind {
-; X32:   atomic_fetch_and64
+; X64-LABEL:   atomic_fetch_and:64
+; X32-LABEL:   atomic_fetch_and64:
   %t1 = atomicrmw and  i64* @sc64, i64 3 acquire
-; X32:       andl
-; X32:       andl
+; X32:       andl $3
+; X32-NOT:       andl
 ; X32:       lock
 ; X32:       cmpxchg8b
-  %t2 = atomicrmw and  i64* @sc64, i64 5 acquire
-; X32:       andl
-; X32:       andl
+  %t2 = atomicrmw and  i64* @sc64, i64 4294967297 acquire
+; X32:       andl $1
+; X32:       andl $1
 ; X32:       lock
 ; X32:       cmpxchg8b
   %t3 = atomicrmw and  i64* @sc64, i64 %t2 acquire
 ; X32:       andl
 ; X32:       andl
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_or64() nounwind {
-; X32:   atomic_fetch_or64
+; X64-LABEL:   atomic_fetch_or64:
+; X32-LABEL:   atomic_fetch_or64:
   %t1 = atomicrmw or   i64* @sc64, i64 3 acquire
-; X32:       orl
-; X32:       orl
+; X32:       orl $3
+; X32-NOT:       orl
 ; X32:       lock
 ; X32:       cmpxchg8b
-  %t2 = atomicrmw or   i64* @sc64, i64 5 acquire
-; X32:       orl
-; X32:       orl
+  %t2 = atomicrmw or   i64* @sc64, i64 4294967297 acquire
+; X32:       orl $1
+; X32:       orl $1
 ; X32:       lock
 ; X32:       cmpxchg8b
   %t3 = atomicrmw or   i64* @sc64, i64 %t2 acquire
 ; X32:       orl
 ; X32:       orl
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_xor64() nounwind {
-; X32:   atomic_fetch_xor64
+; X64-LABEL:   atomic_fetch_xor:64
+; X32-LABEL:   atomic_fetch_xor64:
   %t1 = atomicrmw xor  i64* @sc64, i64 3 acquire
 ; X32:       xorl
-; X32:       xorl
+; X32-NOT:       xorl
 ; X32:       lock
 ; X32:       cmpxchg8b
-  %t2 = atomicrmw xor  i64* @sc64, i64 5 acquire
-; X32:       xorl
-; X32:       xorl
+  %t2 = atomicrmw xor  i64* @sc64, i64 4294967297 acquire
+; X32:       xorl $1
+; X32:       xorl $1
 ; X32:       lock
 ; X32:       cmpxchg8b
   %t3 = atomicrmw xor  i64* @sc64, i64 %t2 acquire
 ; X32:       xorl
 ; X32:       xorl
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_nand64(i64 %x) nounwind {
-; X32:   atomic_fetch_nand64
+; X64-LABEL:   atomic_fetch_nand64:
+; X32-LABEL:   atomic_fetch_nand64:
   %t1 = atomicrmw nand i64* @sc64, i64 %x acquire
 ; X32:       andl
 ; X32:       andl
 ; X32:       notl
 ; X32:       notl
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_max64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_max:64
+; X32-LABEL:   atomic_fetch_max64:
   %t1 = atomicrmw max  i64* @sc64, i64 %x acquire
-; X32:       cmpl
-; X32:       cmpl
-; X32:       cmov
+; X32:       subl
+; X32:       subl
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_min64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_min64:
+; X32-LABEL:   atomic_fetch_min64:
   %t1 = atomicrmw min  i64* @sc64, i64 %x acquire
-; X32:       cmpl
-; X32:       cmpl
-; X32:       cmov
+; X32:       subl
+; X32:       subl
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_umax64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_umax:64
+; X32-LABEL:   atomic_fetch_umax64:
   %t1 = atomicrmw umax i64* @sc64, i64 %x acquire
-; X32:       cmpl
-; X32:       cmpl
-; X32:       cmov
+; X32:       subl
+; X32:       subl
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_umin64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_umin64:
+; X32-LABEL:   atomic_fetch_umin64:
   %t1 = atomicrmw umin i64* @sc64, i64 %x acquire
-; X32:       cmpl
-; X32:       cmpl
-; X32:       cmov
+; X32:       subl
+; X32:       subl
 ; X32:       cmov
 ; X32:       cmov
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_cmpxchg64() nounwind {
+; X64-LABEL:   atomic_fetch_cmpxchg:64
+; X32-LABEL:   atomic_fetch_cmpxchg64:
   %t1 = cmpxchg i64* @sc64, i64 0, i64 1 acquire acquire
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_store64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_store64:
+; X32-LABEL:   atomic_fetch_store64:
   store atomic i64 %x, i64* @sc64 release, align 8
 ; X32:       lock
 ; X32:       cmpxchg8b
   ret void
 ; X32:       ret
 }
 
 define void @atomic_fetch_swap64(i64 %x) nounwind {
+; X64-LABEL:   atomic_fetch_swap64:
+; X32-LABEL:   atomic_fetch_swap64:
   %t1 = atomicrmw xchg i64* @sc64, i64 %x acquire
 ; X32:       lock
 ; X32:       xchg8b
   ret void
 ; X32:       ret
 }

test/CodeGen/X86/atomic8.ll

 ; RUN: llc < %s -O0 -march=x86-64 -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X64
 ; RUN: llc < %s -O0 -march=x86 -mcpu=corei7 -verify-machineinstrs | FileCheck %s --check-prefix X32
 
 @sc8 = external global i8
 
 define void @atomic_fetch_add8() nounwind {
-; X64:   atomic_fetch_add8
-; X32:   atomic_fetch_add8
+; X64-LABEL:   atomic_fetch_add8:
+; X32-LABEL:   atomic_fetch_add8:
 entry:
 ; 32-bit
   %t1 = atomicrmw add  i8* @sc8, i8 1 acquire
 ; X64:       lock
 ; X64:       incb
 ; X32:       lock
 ; X32:       incb
   %t2 = atomicrmw add  i8* @sc8, i8 3 acquire

 ; X32:       lock
 ; X32:       addb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_sub8() nounwind {
-; X64:   atomic_fetch_sub8
-; X32:   atomic_fetch_sub8
+; X64-LABEL:   atomic_fetch_sub8:
+; X32-LABEL:   atomic_fetch_sub8:
   %t1 = atomicrmw sub  i8* @sc8, i8 1 acquire
 ; X64:       lock
 ; X64:       decb
 ; X32:       lock
 ; X32:       decb
   %t2 = atomicrmw sub  i8* @sc8, i8 3 acquire
 ; X64:       lock
 ; X64:       subb $3

 ; X32:       lock
 ; X32:       subb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_and8() nounwind {
-; X64:   atomic_fetch_and8
-; X32:   atomic_fetch_and8
+; X64-LABEL:   atomic_fetch_and8:
+; X32-LABEL:   atomic_fetch_and8:
   %t1 = atomicrmw and  i8* @sc8, i8 3 acquire
 ; X64:       lock
 ; X64:       andb $3
 ; X32:       lock
 ; X32:       andb $3
   %t2 = atomicrmw and  i8* @sc8, i8 5 acquire
 ; X64:       andb
 ; X64:       lock
 ; X64:       cmpxchgb
 ; X32:       andb
 ; X32:       lock
 ; X32:       cmpxchgb
   %t3 = atomicrmw and  i8* @sc8, i8 %t2 acquire
 ; X64:       lock
 ; X64:       andb
 ; X32:       lock
 ; X32:       andb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_or8() nounwind {
-; X64:   atomic_fetch_or8
-; X32:   atomic_fetch_or8
+; X64-LABEL:   atomic_fetch_or8:
+; X32-LABEL:   atomic_fetch_or8:
   %t1 = atomicrmw or   i8* @sc8, i8 3 acquire
 ; X64:       lock
 ; X64:       orb $3
 ; X32:       lock
 ; X32:       orb $3
   %t2 = atomicrmw or   i8* @sc8, i8 5 acquire
 ; X64:       orb
 ; X64:       lock
 ; X64:       cmpxchgb
 ; X32:       orb
 ; X32:       lock
 ; X32:       cmpxchgb
   %t3 = atomicrmw or   i8* @sc8, i8 %t2 acquire
 ; X64:       lock
 ; X64:       orb
 ; X32:       lock
 ; X32:       orb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_xor8() nounwind {
-; X64:   atomic_fetch_xor8
-; X32:   atomic_fetch_xor8
+; X64-LABEL:   atomic_fetch_xor8:
+; X32-LABEL:   atomic_fetch_xor8:
   %t1 = atomicrmw xor  i8* @sc8, i8 3 acquire
 ; X64:       lock
 ; X64:       xorb $3
 ; X32:       lock
 ; X32:       xorb $3
   %t2 = atomicrmw xor  i8* @sc8, i8 5 acquire
 ; X64:       xorb
 ; X64:       lock
 ; X64:       cmpxchgb
 ; X32:       xorb
 ; X32:       lock
 ; X32:       cmpxchgb
   %t3 = atomicrmw xor  i8* @sc8, i8 %t2 acquire
 ; X64:       lock
 ; X64:       xorb
 ; X32:       lock
 ; X32:       xorb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_nand8(i8 %x) nounwind {
-; X64:   atomic_fetch_nand8
-; X32:   atomic_fetch_nand8
+; X64-LABEL:   atomic_fetch_nand8:
+; X32-LABEL:   atomic_fetch_nand8:
   %t1 = atomicrmw nand i8* @sc8, i8 %x acquire
 ; X64:       andb
 ; X64:       notb
 ; X64:       lock
 ; X64:       cmpxchgb
 ; X32:       andb
 ; X32:       notb
 ; X32:       lock
 ; X32:       cmpxchgb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_max8(i8 %x) nounwind {
+; X64-LABEL:   atomic_fetch_max8:
+; X32-LABEL:   atomic_fetch_max8:
   %t1 = atomicrmw max  i8* @sc8, i8 %x acquire
-; X64:       cmpb
-; X64:       cmov
+; X64:       movsbl
+; X64:       movsbl
+; X64:       subl
 ; X64:       lock
 ; X64:       cmpxchgb
 
-; X32:       cmpb
-; X32:       cmov
+; X32:       movsbl
+; X32:       movsbl
+; X32:       subl
 ; X32:       lock
 ; X32:       cmpxchgb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_min8(i8 %x) nounwind {
+; X64-LABEL:   atomic_fetch_min8:
+; X32-LABEL:   atomic_fetch_min8:
   %t1 = atomicrmw min  i8* @sc8, i8 %x acquire
-; X64:       cmpb
-; X64:       cmov
+; X64:       movsbl
+; X64:       movsbl
+; X64:       subl
 ; X64:       lock
 ; X64:       cmpxchgb
 
-; X32:       cmpb
-; X32:       cmov
+; X32:       movsbl
+; X32:       movsbl
+; X32:       subl
 ; X32:       lock
 ; X32:       cmpxchgb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_umax8(i8 %x) nounwind {
+; X64-LABEL:   atomic_fetch_umax8:
+; X32-LABEL:   atomic_fetch_umax8:
   %t1 = atomicrmw umax i8* @sc8, i8 %x acquire
-; X64:       cmpb
-; X64:       cmov
+; X64:       movzbl
+; X64:       movzbl
+; X64:       subl
 ; X64:       lock
 ; X64:       cmpxchgb
 
-; X32:       cmpb
-; X32:       cmov
+; X32:       movzbl
+; X32:       movzbl
+; X32:       subl
 ; X32:       lock
 ; X32:       cmpxchgb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_umin8(i8 %x) nounwind {
+; X64-LABEL:   atomic_fetch_umin8:
+; X32-LABEL:   atomic_fetch_umin8:
   %t1 = atomicrmw umin i8* @sc8, i8 %x acquire
-; X64:       cmpb
-; X64:       cmov
+; X64:       movzbl
+; X64:       movzbl
+; X64:       subl
 ; X64:       lock
 ; X64:       cmpxchgb
-; X32:       cmpb
-; X32:       cmov
+
+; X32:       movzbl
+; X32:       movzbl
+; X32:       subl
 ; X32:       lock
 ; X32:       cmpxchgb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_cmpxchg8() nounwind {
+; X64-LABEL:   atomic_fetch_cmpxchg8:
+; X32-LABEL:   atomic_fetch_cmpxchg8:
   %t1 = cmpxchg i8* @sc8, i8 0, i8 1 acquire acquire
 ; X64:       lock
 ; X64:       cmpxchgb
 ; X32:       lock
 ; X32:       cmpxchgb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_store8(i8 %x) nounwind {
+; X64-LABEL:   atomic_fetch_store8:
+; X32-LABEL:   atomic_fetch_store8:
   store atomic i8 %x, i8* @sc8 release, align 4
 ; X64-NOT:   lock
 ; X64:       movb
 ; X32-NOT:   lock
 ; X32:       movb
   ret void
 ; X64:       ret
 ; X32:       ret
 }
 
 define void @atomic_fetch_swap8(i8 %x) nounwind {
+; X64-LABEL:   atomic_fetch_swap8:
+; X32-LABEL:   atomic_fetch_swap8:
   %t1 = atomicrmw xchg i8* @sc8, i8 %x acquire
 ; X64-NOT:   lock
 ; X64:       xchgb
 ; X32-NOT:   lock
 ; X32:       xchgb
   ret void
 ; X64:       ret
 ; X32:       ret
 }

test/CodeGen/X86/atomic_op.ll

-; RUN: llc < %s -mcpu=generic -march=x86 -mattr=+cmov -verify-machineinstrs | FileCheck %s
+; RUN: llc < %s -mcpu=generic -march=x86 -mattr=+cmov,cx16 -verify-machineinstrs | FileCheck %s
 
 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"
 
 define void @func(i32 %argc, i8** %argv) nounwind {
 entry:
   %argc.addr = alloca i32   ; <i32*> [#uses=1]
   %argv.addr = alloca i8**    ; <i8***> [#uses=1]
   %val1 = alloca i32    ; <i32*> [#uses=2]
entry:
   %16 = extractvalue { i32, i1 } %pair16, 0
   store i32 %16, i32* %old
         ; CHECK: lock
         ; CHECK: cmpxchgl
   %pair17 = cmpxchg i32* %val2, i32 1976, i32 1 monotonic monotonic
   %17 = extractvalue { i32, i1 } %pair17, 0
   store i32 %17, i32* %old
         ; CHECK: movl  [[R17atomic:.*]], %eax
-        ; CHECK: movl $1401, %[[R17mask:[a-z]*]]
-        ; CHECK: andl %eax, %[[R17mask]]
-        ; CHECK: notl %[[R17mask]]
+        ; CHECK: movl %eax, %[[R17mask:[a-z]*]]
+        ; CHECK: notl %[[R17mask]]
+        ; CHECK: orl $-1402, %[[R17mask]]
         ; CHECK: lock
         ; CHECK: cmpxchgl %[[R17mask]], [[R17atomic]]
         ; CHECK: jne
         ; CHECK: movl %eax,
   %18 = atomicrmw nand i32* %val2, i32 1401 monotonic
   store i32 %18, i32* %old
-        ; CHECK: andl
-        ; CHECK: andl
         ; CHECK: notl
         ; CHECK: notl
+        ; CHECK: orl $252645135
+        ; CHECK: orl $252645135
         ; CHECK: lock
         ; CHECK: cmpxchg8b
   %19 = atomicrmw nand i64* %temp64, i64 17361641481138401520 monotonic
   store i64 %19, i64* %temp64
   ret void
 }
 
 define void @test2(i32 addrspace(256)* nocapture %P) nounwind {
 entry:
 ; CHECK: lock
 ; CHECK:  cmpxchgl  %{{.*}}, %gs:(%{{.*}})
 
   %pair0 = cmpxchg i32 addrspace(256)* %P, i32 0, i32 1 monotonic monotonic
   %0 = extractvalue { i32, i1 } %pair0, 0
   ret void
 }

test/CodeGen/X86/pr5145.ll

 ; RUN: llc -march=x86-64 < %s | FileCheck %s
 @sc8 = external global i8
 
 define void @atomic_maxmin_i8() {
 ; CHECK: atomic_maxmin_i8
   %1 = atomicrmw max  i8* @sc8, i8 5 acquire
 ; CHECK: [[LABEL1:\.?LBB[0-9]+_[0-9]+]]:
-; CHECK: cmpb
-; CHECK: cmovl
+; CHECK: movsbl
+; CHECK: cmpl
 ; CHECK: lock
 ; CHECK-NEXT: cmpxchgb
 ; CHECK: jne [[LABEL1]]
   %2 = atomicrmw min  i8* @sc8, i8 6 acquire
 ; CHECK: [[LABEL3:\.?LBB[0-9]+_[0-9]+]]:
-; CHECK: cmpb
-; CHECK: cmovg
+; CHECK: movsbl
+; CHECK: cmpl
 ; CHECK: lock
 ; CHECK-NEXT: cmpxchgb
 ; CHECK: jne [[LABEL3]]
   %3 = atomicrmw umax i8* @sc8, i8 7 acquire
 ; CHECK: [[LABEL5:\.?LBB[0-9]+_[0-9]+]]:
-; CHECK: cmpb
-; CHECK: cmovb
+; CHECK: movzbl
+; CHECK: cmpl
 ; CHECK: lock
 ; CHECK-NEXT: cmpxchgb
 ; CHECK: jne [[LABEL5]]
   %4 = atomicrmw umin i8* @sc8, i8 8 acquire
 ; CHECK: [[LABEL7:\.?LBB[0-9]+_[0-9]+]]:
-; CHECK: cmpb
-; CHECK: cmova
+; CHECK: movzbl
+; CHECK: cmpl
 ; CHECK: lock
 ; CHECK-NEXT: cmpxchgb
 ; CHECK: jne [[LABEL7]]
   ret void
 }