[X86] Transform AtomicRMW logic operations to BT{R|C|S} if only changing/testing a single bit.

[goldstein.w.n]

Summary

This is essentially expanding on the optimizations added on: D120199
but applies the optimization to cases where the bit being changed /
tested is not am IMM but is a provable power of 2.

The only case currently added for cases like:
__atomic_fetch_xor(p, 1 << c, __ATOMIC_RELAXED) & (1 << c)

Which instead of using a cmpxchg loop can be done with btcl; setcc; shl.

There are still a variety of missed cases that could/should be
addressed in the future. This commit documents many of those
cases with Todos.

[goldstein.w.n]

This is patch [2/2].

Note: This is a follow up to https://reviews.llvm.org/D140645 and is splitting up the reviews for the tests and code.

[goldstein.w.n]

Propegate BMI test removal

[pengfei]

llvm/lib/Target/X86/X86ISelLowering.cpp

5653

Should the Size be same with operand 1 rather than result?

31434

or

31435

No parentheses needed for single line scope.

31453

Should it be a canonical form that we can assume it is always X ^ -1?

31454

We can use match(....) for better readability?

31527

Is enough to only check BitChange.second == UndefBit?

31559

Missing the last period .

31564

ditto.

llvm/test/CodeGen/X86/atomic-rm-bit-test.ll

4622–4628

The branch code doesn't look necessary. Can we necessary it?

[goldstein.w.n]

Use match instead of hand written logic.
Fix some typos.
Fix some code style nits.

llvm/lib/Target/X86/X86ISelLowering.cpp

5653

Should the Size be same with operand 1 rather than result?

If I change it I run into assertion errors.

I.getArgOperand(0)->getType()->getScalarSizeInBits() the Size can be zero (bitwidth too small)

or I.getArgOperand(1)->getType()->getScalarSizeInBits() I run into:

llc: /home/noah/programs/opensource/llvm-dev/src/llvm-project/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp:11108: llvm::MemSDNode::MemSDNode(unsigned int, unsigned int, const llvm::DebugLoc&, llvm::SDVTList, llvm::EVT, llvm::MachineMemOperand*): Assertion `memvt.getStoreSize().getKnownMinSize() <= MMO->getSize() && "Size mismatch!"' failed.

Maybe that means there is a bug in the patch elsewhere?

Running the test here are the sizes I see:

Sizes: 8 / 0 / 16
Sizes: 8 / 0 / 16
Sizes: 8 / 0 / 16
Sizes: 8 / 0 / 16
Sizes: 16 / 0 / 8
Sizes: 16 / 0 / 8
Sizes: 16 / 0 / 8
Sizes: 16 / 0 / 8
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 8 / 0 / 32
Sizes: 32 / 0 / 8
Sizes: 32 / 0 / 8
Sizes: 32 / 0 / 8
Sizes: 32 / 0 / 8
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 8 / 0 / 64
Sizes: 64 / 0 / 8
Sizes: 64 / 0 / 8
Sizes: 64 / 0 / 8

31453

Should it be a canonical form that we can assume it is always X ^ -1?

Probably defacto fixed by using match (didn't know match existed when I wrote this patch :/ alot to learn).

llvm/test/CodeGen/X86/atomic-rm-bit-test.ll

4622–4628

The branch code doesn't look necessary. Can we necessary it?

I think it is b.c we don't cmovcc loads.

For

if.then:                                          ; preds = %entry
  %idxprom = zext i32 %c to i64
  %arrayidx = getelementptr inbounds i32, ptr %v, i64 %idxprom
  %1 = load i32, ptr %arrayidx, align 4
  br label %return

And

return:                                           ; preds = %entry, %if.then
  %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
  ret i32 %retval.0

a branch seems correct.

[pengfei]

llvm/test/CodeGen/X86/atomic-rm-bit-test.ll

4622–4628

Sorry for the wrong words. I mean can we eliminate the branch by modifying the IR code, e.g.,

entry:
  %shl = shl nuw i32 1, %c
  %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
  %and = and i32 %0, %shl
  %tobool.not = icmp eq i32 %and, 0
  %ret = zext i1 %tobool.not to i32
  ret i32 %ret

This will help to reduce the nosie in reviewing the code and pay more attention to the change we expected.

[pengfei]

llvm/lib/Target/X86/X86ISelLowering.cpp

5653

I mean split x86_atomic_bt*_rm from it and then change it to I.getArgOperand(1)->getType()->getScalarSizeInBits()
Other intrinsics are correct to use I.getType()->getScalarSizeInBits().

[goldstein.w.n]

llvm/lib/Target/X86/X86ISelLowering.cpp

5653

I mean split x86_atomic_bt*_rm from it and then change it to I.getArgOperand(1)->getType()->getScalarSizeInBits()
Other intrinsics are correct to use I.getType()->getScalarSizeInBits().

What does "split x86_atomic_bt*_rm` from it" mean?

llvm/test/CodeGen/X86/atomic-rm-bit-test.ll

4622–4628

Sorry for the wrong words. I mean can we eliminate the branch by modifying the IR code, e.g.,

entry:
  %shl = shl nuw i32 1, %c
  %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
  %and = and i32 %0, %shl
  %tobool.not = icmp eq i32 %and, 0
  %ret = zext i1 %tobool.not to i32
  ret i32 %ret

This will help to reduce the nosie in reviewing the code and pay more attention to the change we expected.

I see, so there are 6 difference test types:
_br -> branch on value
_brz -> branch on !value
_brnz -> branch on !!value
_val -> return value
_valz -> return !value
_valnz -> return !!value

Imo they are all worth testing. For example we have logic that searches the uses to see if its only for a truth value to see if we can optimize out the shift.
IIRC when writing the code, there where some edge cases where br behaved differently than setc so think its worth keeping.

[pengfei]

llvm/lib/Target/X86/X86ISelLowering.cpp

5653

case Intrinsic::x86_cmpccxadd32:
case Intrinsic::x86_cmpccxadd64:
case Intrinsic::x86_atomic_bts:
case Intrinsic::x86_atomic_btc:
case Intrinsic::x86_atomic_btr: {
  ... ...
}
case Intrinsic::x86_atomic_bts_rm:
case Intrinsic::x86_atomic_btc_rm:
case Intrinsic::x86_atomic_btr_rm: {
  ... ...
}

[goldstein.w.n]

Propegate test changes.
Use proper size

[goldstein.w.n]

llvm/lib/Target/X86/X86ISelLowering.cpp

5653

case Intrinsic::x86_cmpccxadd32:
case Intrinsic::x86_cmpccxadd64:
case Intrinsic::x86_atomic_bts:
case Intrinsic::x86_atomic_btc:
case Intrinsic::x86_atomic_btr: {
  ... ...
}
case Intrinsic::x86_atomic_bts_rm:
case Intrinsic::x86_atomic_btc_rm:
case Intrinsic::x86_atomic_btr_rm: {
  ... ...
}

Done, for some reason had thought they where in different cases.

[goldstein.w.n]

ping.

[pengfei]

LGTM.

[mysterymath]

Hey, it looks like this change triggered an assertion failure in the Fuchsia continous build: https://luci-milo.appspot.com/ui/p/fuchsia/builders/ci/clang_toolchain.ci.core.x64-release/b8791734713064712625/overview

clang: llvm/lib/Target/X86/X86ISelLowering.cpp:31525: TargetLowering::AtomicExpansionKind llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR(AtomicRMWInst *) const: Assertion `I->getOperand(0)...
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.	Program arguments: ../../../recipe_cleanup/clangs6i4kjx7/bin/clang -MD -MF obj/src/connectivity/wlan/drivers/third_party/intel/iwlwifi/mvm/mvm.sta.c.o.d -D_LIBCPP_DISABLE_VISIBILITY_ANNOTATIONS ...
1.	<eof> parser at end of file
2.	Code generation
3.	Running pass 'Function Pass Manager' on module '../../src/connectivity/wlan/drivers/third_party/intel/iw...

[goldstein.w.n]

In D140939#4067362, @mysterymath wrote:

Hey, it looks like this change triggered an assertion failure in the Fuchsia continous build: https://luci-milo.appspot.com/ui/p/fuchsia/builders/ci/clang_toolchain.ci.core.x64-release/b8791734713064712625/overview

clang: llvm/lib/Target/X86/X86ISelLowering.cpp:31525: TargetLowering::AtomicExpansionKind llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR(AtomicRMWInst *) const: Assertion `I->getOperand(0)...
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.	Program arguments: ../../../recipe_cleanup/clangs6i4kjx7/bin/clang -MD -MF obj/src/connectivity/wlan/drivers/third_party/intel/iwlwifi/mvm/mvm.sta.c.o.d -D_LIBCPP_DISABLE_VISIBILITY_ANNOTATIONS ...
1.	<eof> parser at end of file
2.	Code generation
3.	Running pass 'Function Pass Manager' on module '../../src/connectivity/wlan/drivers/third_party/intel/iw...

The easiest thing to do now would be:

-  assert(I->getOperand(0) == AI);
+  if (I->getOperand(0) != AI)
+    return AtomicExpansionKind::CmpXChg;

or revert.

Although I think if that assertion is being hit the old code was buggy:

// The following instruction must be a AND single bit.
auto *C2 = dyn_cast<ConstantInt>(I->getOperand(1));
unsigned Bits = AI->getType()->getPrimitiveSizeInBits();
if (!C2 || Bits == 8 || !isPowerOf2_64(C2->getZExtValue()))
  return AtomicExpansionKind::CmpXChg;

if (AI->getOperation() == AtomicRMWInst::And)
  return ~C1->getValue() == C2->getValue()
             ? AtomicExpansionKind::BitTestIntrinsic
             : AtomicExpansionKind::CmpXChg;

return C1 == C2 ? AtomicExpansionKind::BitTestIntrinsic
                : AtomicExpansionKind::CmpXChg;

Was assuming the above assert AFAICT.

[goldstein.w.n]

In D140939#4067362, @mysterymath wrote:

Hey, it looks like this change triggered an assertion failure in the Fuchsia continous build: https://luci-milo.appspot.com/ui/p/fuchsia/builders/ci/clang_toolchain.ci.core.x64-release/b8791734713064712625/overview

clang: llvm/lib/Target/X86/X86ISelLowering.cpp:31525: TargetLowering::AtomicExpansionKind llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR(AtomicRMWInst *) const: Assertion `I->getOperand(0)...
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.	Program arguments: ../../../recipe_cleanup/clangs6i4kjx7/bin/clang -MD -MF obj/src/connectivity/wlan/drivers/third_party/intel/iwlwifi/mvm/mvm.sta.c.o.d -D_LIBCPP_DISABLE_VISIBILITY_ANNOTATIONS ...
1.	<eof> parser at end of file
2.	Code generation
3.	Running pass 'Function Pass Manager' on module '../../src/connectivity/wlan/drivers/third_party/intel/iw...

Created: https://reviews.llvm.org/D142166 Its a proper fix and tested. Otherwise revert is probably best for now

[goldstein.w.n]

In D140939#4067387, @goldstein.w.n wrote:

In D140939#4067362, @mysterymath wrote:

Hey, it looks like this change triggered an assertion failure in the Fuchsia continous build: https://luci-milo.appspot.com/ui/p/fuchsia/builders/ci/clang_toolchain.ci.core.x64-release/b8791734713064712625/overview

clang: llvm/lib/Target/X86/X86ISelLowering.cpp:31525: TargetLowering::AtomicExpansionKind llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR(AtomicRMWInst *) const: Assertion `I->getOperand(0)...
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.	Program arguments: ../../../recipe_cleanup/clangs6i4kjx7/bin/clang -MD -MF obj/src/connectivity/wlan/drivers/third_party/intel/iwlwifi/mvm/mvm.sta.c.o.d -D_LIBCPP_DISABLE_VISIBILITY_ANNOTATIONS ...
1.	<eof> parser at end of file
2.	Code generation
3.	Running pass 'Function Pass Manager' on module '../../src/connectivity/wlan/drivers/third_party/intel/iw...

Created: https://reviews.llvm.org/D142166 Its a proper fix and tested. Otherwise revert is probably best for now

D142166 has been accepted with some nits. Going to wait ~24hours to any remaining comments, then push.

[alexfh]

Unfortunately, there's another problem, which doesn't get fixed by https://reviews.llvm.org/D142166 / 2e25204779e5b972d668bf66a0014c1325813b35:

assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x56257abc9524  __assert_fail
    @     0x562578c7fd1c  FindSingleBitChange()
    @     0x562578c7f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x562578c8098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x5625790a6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x5625790a59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x56257a86e87d  llvm::FPPassManager::runOnFunction()
    @     0x56257a876104  llvm::FPPassManager::runOnModule()
    @     0x56257a86ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x562575bad581  clang::EmitBackendOutput()
    @     0x562575baabe9  clang::BackendConsumer::HandleTranslationUnit()
    @     0x562576a85e5c  clang::ParseAST()
    @     0x5625767cbf23  clang::FrontendAction::Execute()
    @     0x562576741cad  clang::CompilerInstance::ExecuteAction()
    @     0x5625757871e8  clang::ExecuteCompilerInvocation()
    @     0x56257577ae41  cc1_main()
    @     0x562575776ec8  ExecuteCC1Tool()
    @     0x5625768ed317  llvm::function_ref<>::callback_fn<>()
    @     0x56257aa23d20  llvm::CrashRecoveryContext::RunSafely()
    @     0x5625768ecb63  clang::driver::CC1Command::Execute()
    @     0x5625768ab2e6  clang::driver::Compilation::ExecuteCommand()
    @     0x5625768ab60f  clang::driver::Compilation::ExecuteJobs()
    @     0x5625768caf70  clang::driver::Driver::ExecuteCompilation()
    @     0x562575776027  clang_main()
    @     0x7f8f04be3633  __libc_start_main
    @     0x562575772bea  _start

[alexfh]

In D140939#4072473, @alexfh wrote:

Unfortunately, there's another problem, which doesn't get fixed by https://reviews.llvm.org/D142166 / 2e25204779e5b972d668bf66a0014c1325813b35:

assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x56257abc9524  __assert_fail
    @     0x562578c7fd1c  FindSingleBitChange()
    @     0x562578c7f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x562578c8098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x5625790a6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x5625790a59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x56257a86e87d  llvm::FPPassManager::runOnFunction()
    @     0x56257a876104  llvm::FPPassManager::runOnModule()
    @     0x56257a86ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x562575bad581  clang::EmitBackendOutput()
    @     0x562575baabe9  clang::BackendConsumer::HandleTranslationUnit()
    @     0x562576a85e5c  clang::ParseAST()
    @     0x5625767cbf23  clang::FrontendAction::Execute()
    @     0x562576741cad  clang::CompilerInstance::ExecuteAction()
    @     0x5625757871e8  clang::ExecuteCompilerInvocation()
    @     0x56257577ae41  cc1_main()
    @     0x562575776ec8  ExecuteCC1Tool()
    @     0x5625768ed317  llvm::function_ref<>::callback_fn<>()
    @     0x56257aa23d20  llvm::CrashRecoveryContext::RunSafely()
    @     0x5625768ecb63  clang::driver::CC1Command::Execute()
    @     0x5625768ab2e6  clang::driver::Compilation::ExecuteCommand()
    @     0x5625768ab60f  clang::driver::Compilation::ExecuteJobs()
    @     0x5625768caf70  clang::driver::Driver::ExecuteCompilation()
    @     0x562575776027  clang_main()
    @     0x7f8f04be3633  __libc_start_main
    @     0x562575772bea  _start

I'd suggest to revert while investigating. I'm working on an isolated test case.

[alexfh]

In D140939#4072474, @alexfh wrote:

In D140939#4072473, @alexfh wrote:

Unfortunately, there's another problem, which doesn't get fixed by https://reviews.llvm.org/D142166 / 2e25204779e5b972d668bf66a0014c1325813b35:

assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x56257abc9524  __assert_fail
    @     0x562578c7fd1c  FindSingleBitChange()
    @     0x562578c7f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x562578c8098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x5625790a6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x5625790a59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x56257a86e87d  llvm::FPPassManager::runOnFunction()
    @     0x56257a876104  llvm::FPPassManager::runOnModule()
    @     0x56257a86ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x562575bad581  clang::EmitBackendOutput()
    @     0x562575baabe9  clang::BackendConsumer::HandleTranslationUnit()
    @     0x562576a85e5c  clang::ParseAST()
    @     0x5625767cbf23  clang::FrontendAction::Execute()
    @     0x562576741cad  clang::CompilerInstance::ExecuteAction()
    @     0x5625757871e8  clang::ExecuteCompilerInvocation()
    @     0x56257577ae41  cc1_main()
    @     0x562575776ec8  ExecuteCC1Tool()
    @     0x5625768ed317  llvm::function_ref<>::callback_fn<>()
    @     0x56257aa23d20  llvm::CrashRecoveryContext::RunSafely()
    @     0x5625768ecb63  clang::driver::CC1Command::Execute()
    @     0x5625768ab2e6  clang::driver::Compilation::ExecuteCommand()
    @     0x5625768ab60f  clang::driver::Compilation::ExecuteJobs()
    @     0x5625768caf70  clang::driver::Driver::ExecuteCompilation()
    @     0x562575776027  clang_main()
    @     0x7f8f04be3633  __libc_start_main
    @     0x562575772bea  _start

I'd suggest to revert while investigating. I'm working on an isolated test case.

$ cat reduced.ll
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

define weak_odr void @f() {
entry:
  br label %if.end19

cont11:                                           ; No predecessors!
  %not = xor i32 0, -1
  %0 = atomicrmw and ptr null, i32 %not monotonic, align 4
  %and13 = and i32 %0, 0
  br label %if.end19

if.end19:                                         ; preds = %cont11, %entry
  ret void
}
$ ./good-clang -c reduced.ll -o good.o
$ ./bad-clang -c reduced.ll -o bad.o
assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x559d015c9524  __assert_fail
    @     0x559cff67fd1c  FindSingleBitChange()
    @     0x559cff67f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x559cff68098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x559cffaa6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x559cffaa59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x559d0126e87d  llvm::FPPassManager::runOnFunction()
    @     0x559d01276104  llvm::FPPassManager::runOnModule()
    @     0x559d0126ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x559cfc5ad581  clang::EmitBackendOutput()
    @     0x559cfc5a94f8  clang::CodeGenAction::ExecuteAction()
    @     0x559cfd1cbf23  clang::FrontendAction::Execute()
    @     0x559cfd141cad  clang::CompilerInstance::ExecuteAction()
    @     0x559cfc1871e8  clang::ExecuteCompilerInvocation()
    @     0x559cfc17ae41  cc1_main()

[goldstein.w.n]

In D140939#4072488, @alexfh wrote:

In D140939#4072474, @alexfh wrote:

In D140939#4072473, @alexfh wrote:

Unfortunately, there's another problem, which doesn't get fixed by https://reviews.llvm.org/D142166 / 2e25204779e5b972d668bf66a0014c1325813b35:

assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x56257abc9524  __assert_fail
    @     0x562578c7fd1c  FindSingleBitChange()
    @     0x562578c7f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x562578c8098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x5625790a6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x5625790a59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x56257a86e87d  llvm::FPPassManager::runOnFunction()
    @     0x56257a876104  llvm::FPPassManager::runOnModule()
    @     0x56257a86ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x562575bad581  clang::EmitBackendOutput()
    @     0x562575baabe9  clang::BackendConsumer::HandleTranslationUnit()
    @     0x562576a85e5c  clang::ParseAST()
    @     0x5625767cbf23  clang::FrontendAction::Execute()
    @     0x562576741cad  clang::CompilerInstance::ExecuteAction()
    @     0x5625757871e8  clang::ExecuteCompilerInvocation()
    @     0x56257577ae41  cc1_main()
    @     0x562575776ec8  ExecuteCC1Tool()
    @     0x5625768ed317  llvm::function_ref<>::callback_fn<>()
    @     0x56257aa23d20  llvm::CrashRecoveryContext::RunSafely()
    @     0x5625768ecb63  clang::driver::CC1Command::Execute()
    @     0x5625768ab2e6  clang::driver::Compilation::ExecuteCommand()
    @     0x5625768ab60f  clang::driver::Compilation::ExecuteJobs()
    @     0x5625768caf70  clang::driver::Driver::ExecuteCompilation()
    @     0x562575776027  clang_main()
    @     0x7f8f04be3633  __libc_start_main
    @     0x562575772bea  _start

I'd suggest to revert while investigating. I'm working on an isolated test case.

$ cat reduced.ll
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

define weak_odr void @f() {
entry:
  br label %if.end19

cont11:                                           ; No predecessors!
  %not = xor i32 0, -1
  %0 = atomicrmw and ptr null, i32 %not monotonic, align 4
  %and13 = and i32 %0, 0
  br label %if.end19

if.end19:                                         ; preds = %cont11, %entry
  ret void
}
$ ./good-clang -c reduced.ll -o good.o
$ ./bad-clang -c reduced.ll -o bad.o
assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x559d015c9524  __assert_fail
    @     0x559cff67fd1c  FindSingleBitChange()
    @     0x559cff67f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x559cff68098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x559cffaa6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x559cffaa59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x559d0126e87d  llvm::FPPassManager::runOnFunction()
    @     0x559d01276104  llvm::FPPassManager::runOnModule()
    @     0x559d0126ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x559cfc5ad581  clang::EmitBackendOutput()
    @     0x559cfc5a94f8  clang::CodeGenAction::ExecuteAction()
    @     0x559cfd1cbf23  clang::FrontendAction::Execute()
    @     0x559cfd141cad  clang::CompilerInstance::ExecuteAction()
    @     0x559cfc1871e8  clang::ExecuteCompilerInvocation()
    @     0x559cfc17ae41  cc1_main()

Thank you for finding the case. A another one (and probably more realistic one is):

define weak_odr void @f(i32 %0) {
entry:
  br label %if.end19

cont11:                                           ; No predecessors!
  %not = xor i32 %0, -1
  %1 = atomicrmw and ptr null, i32 %not monotonic, align 4
  %and13 = and i32 %1, 0
  br label %if.end19

if.end19:                                         ; preds = %cont11, %entry
  ret void
}

So the fix is:

-      assert(I != nullptr);
+
+      // If constant it will fold and we can evaluate later. If its an argument
+      // or something of that nature, we can't analyze.
+      if (I == nullptr)
+        return {nullptr, UndefBit};

I'm happy to post that as a patch.

If you still want to revert, however, happy to do so as well and then repost the full version.

Whichever you think is more prudent.

[goldstein.w.n]

In D140939#4072672, @goldstein.w.n wrote:

In D140939#4072488, @alexfh wrote:

In D140939#4072474, @alexfh wrote:

In D140939#4072473, @alexfh wrote:

Unfortunately, there's another problem, which doesn't get fixed by https://reviews.llvm.org/D142166 / 2e25204779e5b972d668bf66a0014c1325813b35:

assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x56257abc9524  __assert_fail
    @     0x562578c7fd1c  FindSingleBitChange()
    @     0x562578c7f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x562578c8098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x5625790a6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x5625790a59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x56257a86e87d  llvm::FPPassManager::runOnFunction()
    @     0x56257a876104  llvm::FPPassManager::runOnModule()
    @     0x56257a86ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x562575bad581  clang::EmitBackendOutput()
    @     0x562575baabe9  clang::BackendConsumer::HandleTranslationUnit()
    @     0x562576a85e5c  clang::ParseAST()
    @     0x5625767cbf23  clang::FrontendAction::Execute()
    @     0x562576741cad  clang::CompilerInstance::ExecuteAction()
    @     0x5625757871e8  clang::ExecuteCompilerInvocation()
    @     0x56257577ae41  cc1_main()
    @     0x562575776ec8  ExecuteCC1Tool()
    @     0x5625768ed317  llvm::function_ref<>::callback_fn<>()
    @     0x56257aa23d20  llvm::CrashRecoveryContext::RunSafely()
    @     0x5625768ecb63  clang::driver::CC1Command::Execute()
    @     0x5625768ab2e6  clang::driver::Compilation::ExecuteCommand()
    @     0x5625768ab60f  clang::driver::Compilation::ExecuteJobs()
    @     0x5625768caf70  clang::driver::Driver::ExecuteCompilation()
    @     0x562575776027  clang_main()
    @     0x7f8f04be3633  __libc_start_main
    @     0x562575772bea  _start

I'd suggest to revert while investigating. I'm working on an isolated test case.

$ cat reduced.ll
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

define weak_odr void @f() {
entry:
  br label %if.end19

cont11:                                           ; No predecessors!
  %not = xor i32 0, -1
  %0 = atomicrmw and ptr null, i32 %not monotonic, align 4
  %and13 = and i32 %0, 0
  br label %if.end19

if.end19:                                         ; preds = %cont11, %entry
  ret void
}
$ ./good-clang -c reduced.ll -o good.o
$ ./bad-clang -c reduced.ll -o bad.o
assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x559d015c9524  __assert_fail
    @     0x559cff67fd1c  FindSingleBitChange()
    @     0x559cff67f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x559cff68098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x559cffaa6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x559cffaa59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x559d0126e87d  llvm::FPPassManager::runOnFunction()
    @     0x559d01276104  llvm::FPPassManager::runOnModule()
    @     0x559d0126ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x559cfc5ad581  clang::EmitBackendOutput()
    @     0x559cfc5a94f8  clang::CodeGenAction::ExecuteAction()
    @     0x559cfd1cbf23  clang::FrontendAction::Execute()
    @     0x559cfd141cad  clang::CompilerInstance::ExecuteAction()
    @     0x559cfc1871e8  clang::ExecuteCompilerInvocation()
    @     0x559cfc17ae41  cc1_main()

Thank you for finding the case. A another one (and probably more realistic one is):

define weak_odr void @f(i32 %0) {
entry:
  br label %if.end19

cont11:                                           ; No predecessors!
  %not = xor i32 %0, -1
  %1 = atomicrmw and ptr null, i32 %not monotonic, align 4
  %and13 = and i32 %1, 0
  br label %if.end19

if.end19:                                         ; preds = %cont11, %entry
  ret void
}

So the fix is:

-      assert(I != nullptr);
+
+      // If constant it will fold and we can evaluate later. If its an argument
+      // or something of that nature, we can't analyze.
+      if (I == nullptr)
+        return {nullptr, UndefBit};

I'm happy to post that as a patch.

Created: https://reviews.llvm.org/D142339

But let me know if revert is still preferable (I say as much there).

If you still want to revert, however, happy to do so as well and then repost the full version.

Whichever you think is more prudent.

[alexfh]

In D140939#4072755, @goldstein.w.n wrote:

In D140939#4072672, @goldstein.w.n wrote:

In D140939#4072488, @alexfh wrote:

In D140939#4072474, @alexfh wrote:

In D140939#4072473, @alexfh wrote:

Unfortunately, there's another problem, which doesn't get fixed by https://reviews.llvm.org/D142166 / 2e25204779e5b972d668bf66a0014c1325813b35:

assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x56257abc9524  __assert_fail
    @     0x562578c7fd1c  FindSingleBitChange()
    @     0x562578c7f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x562578c8098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x5625790a6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x5625790a59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x56257a86e87d  llvm::FPPassManager::runOnFunction()
    @     0x56257a876104  llvm::FPPassManager::runOnModule()
    @     0x56257a86ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x562575bad581  clang::EmitBackendOutput()
    @     0x562575baabe9  clang::BackendConsumer::HandleTranslationUnit()
    @     0x562576a85e5c  clang::ParseAST()
    @     0x5625767cbf23  clang::FrontendAction::Execute()
    @     0x562576741cad  clang::CompilerInstance::ExecuteAction()
    @     0x5625757871e8  clang::ExecuteCompilerInvocation()
    @     0x56257577ae41  cc1_main()
    @     0x562575776ec8  ExecuteCC1Tool()
    @     0x5625768ed317  llvm::function_ref<>::callback_fn<>()
    @     0x56257aa23d20  llvm::CrashRecoveryContext::RunSafely()
    @     0x5625768ecb63  clang::driver::CC1Command::Execute()
    @     0x5625768ab2e6  clang::driver::Compilation::ExecuteCommand()
    @     0x5625768ab60f  clang::driver::Compilation::ExecuteJobs()
    @     0x5625768caf70  clang::driver::Driver::ExecuteCompilation()
    @     0x562575776027  clang_main()
    @     0x7f8f04be3633  __libc_start_main
    @     0x562575772bea  _start

I'd suggest to revert while investigating. I'm working on an isolated test case.

$ cat reduced.ll
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

define weak_odr void @f() {
entry:
  br label %if.end19

cont11:                                           ; No predecessors!
  %not = xor i32 0, -1
  %0 = atomicrmw and ptr null, i32 %not monotonic, align 4
  %and13 = and i32 %0, 0
  br label %if.end19

if.end19:                                         ; preds = %cont11, %entry
  ret void
}
$ ./good-clang -c reduced.ll -o good.o
$ ./bad-clang -c reduced.ll -o bad.o
assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x559d015c9524  __assert_fail
    @     0x559cff67fd1c  FindSingleBitChange()
    @     0x559cff67f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x559cff68098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x559cffaa6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x559cffaa59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x559d0126e87d  llvm::FPPassManager::runOnFunction()
    @     0x559d01276104  llvm::FPPassManager::runOnModule()
    @     0x559d0126ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x559cfc5ad581  clang::EmitBackendOutput()
    @     0x559cfc5a94f8  clang::CodeGenAction::ExecuteAction()
    @     0x559cfd1cbf23  clang::FrontendAction::Execute()
    @     0x559cfd141cad  clang::CompilerInstance::ExecuteAction()
    @     0x559cfc1871e8  clang::ExecuteCompilerInvocation()
    @     0x559cfc17ae41  cc1_main()

Thank you for finding the case. A another one (and probably more realistic one is):

define weak_odr void @f(i32 %0) {
entry:
  br label %if.end19

cont11:                                           ; No predecessors!
  %not = xor i32 %0, -1
  %1 = atomicrmw and ptr null, i32 %not monotonic, align 4
  %and13 = and i32 %1, 0
  br label %if.end19

if.end19:                                         ; preds = %cont11, %entry
  ret void
}

So the fix is:

-      assert(I != nullptr);
+
+      // If constant it will fold and we can evaluate later. If its an argument
+      // or something of that nature, we can't analyze.
+      if (I == nullptr)
+        return {nullptr, UndefBit};

I'm happy to post that as a patch.

Created: https://reviews.llvm.org/D142339

But let me know if revert is still preferable (I say as much there).

If you still want to revert, however, happy to do so as well and then repost the full version.

Whichever you think is more prudent.

We're seeing a ton of fallout due to this, and given that there are already two different patterns that cause crashes, I'd like to have a known good version in the tree before we play whack-a-mole ;). Thus, I'd prefer that you reverted this and then recommitted with your proposed fix.

[goldstein.w.n]

In D140939#4072961, @alexfh wrote:

In D140939#4072755, @goldstein.w.n wrote:

In D140939#4072672, @goldstein.w.n wrote:

In D140939#4072488, @alexfh wrote:

In D140939#4072474, @alexfh wrote:

In D140939#4072473, @alexfh wrote:

Unfortunately, there's another problem, which doesn't get fixed by https://reviews.llvm.org/D142166 / 2e25204779e5b972d668bf66a0014c1325813b35:

assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x56257abc9524  __assert_fail
    @     0x562578c7fd1c  FindSingleBitChange()
    @     0x562578c7f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x562578c8098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x5625790a6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x5625790a59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x56257a86e87d  llvm::FPPassManager::runOnFunction()
    @     0x56257a876104  llvm::FPPassManager::runOnModule()
    @     0x56257a86ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x562575bad581  clang::EmitBackendOutput()
    @     0x562575baabe9  clang::BackendConsumer::HandleTranslationUnit()
    @     0x562576a85e5c  clang::ParseAST()
    @     0x5625767cbf23  clang::FrontendAction::Execute()
    @     0x562576741cad  clang::CompilerInstance::ExecuteAction()
    @     0x5625757871e8  clang::ExecuteCompilerInvocation()
    @     0x56257577ae41  cc1_main()
    @     0x562575776ec8  ExecuteCC1Tool()
    @     0x5625768ed317  llvm::function_ref<>::callback_fn<>()
    @     0x56257aa23d20  llvm::CrashRecoveryContext::RunSafely()
    @     0x5625768ecb63  clang::driver::CC1Command::Execute()
    @     0x5625768ab2e6  clang::driver::Compilation::ExecuteCommand()
    @     0x5625768ab60f  clang::driver::Compilation::ExecuteJobs()
    @     0x5625768caf70  clang::driver::Driver::ExecuteCompilation()
    @     0x562575776027  clang_main()
    @     0x7f8f04be3633  __libc_start_main
    @     0x562575772bea  _start

I'd suggest to revert while investigating. I'm working on an isolated test case.

$ cat reduced.ll
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

define weak_odr void @f() {
entry:
  br label %if.end19

cont11:                                           ; No predecessors!
  %not = xor i32 0, -1
  %0 = atomicrmw and ptr null, i32 %not monotonic, align 4
  %and13 = and i32 %0, 0
  br label %if.end19

if.end19:                                         ; preds = %cont11, %entry
  ret void
}
$ ./good-clang -c reduced.ll -o good.o
$ ./bad-clang -c reduced.ll -o bad.o
assert.h assertion failed at llvm/lib/Target/X86/X86ISelLowering.cpp:31466 in std::pair<Value *, BitTestKind> FindSingleBitChange(Value *): I != nullptr
    @     0x559d015c9524  __assert_fail
    @     0x559cff67fd1c  FindSingleBitChange()
    @     0x559cff67f6dd  llvm::X86TargetLowering::shouldExpandLogicAtomicRMWInIR()
    @     0x559cff68098b  llvm::X86TargetLowering::shouldExpandAtomicRMWInIR()
    @     0x559cffaa6099  (anonymous namespace)::AtomicExpand::tryExpandAtomicRMW()
    @     0x559cffaa59dd  (anonymous namespace)::AtomicExpand::runOnFunction()
    @     0x559d0126e87d  llvm::FPPassManager::runOnFunction()
    @     0x559d01276104  llvm::FPPassManager::runOnModule()
    @     0x559d0126ef7c  llvm::legacy::PassManagerImpl::run()
    @     0x559cfc5ad581  clang::EmitBackendOutput()
    @     0x559cfc5a94f8  clang::CodeGenAction::ExecuteAction()
    @     0x559cfd1cbf23  clang::FrontendAction::Execute()
    @     0x559cfd141cad  clang::CompilerInstance::ExecuteAction()
    @     0x559cfc1871e8  clang::ExecuteCompilerInvocation()
    @     0x559cfc17ae41  cc1_main()

Thank you for finding the case. A another one (and probably more realistic one is):

define weak_odr void @f(i32 %0) {
entry:
  br label %if.end19

cont11:                                           ; No predecessors!
  %not = xor i32 %0, -1
  %1 = atomicrmw and ptr null, i32 %not monotonic, align 4
  %and13 = and i32 %1, 0
  br label %if.end19

if.end19:                                         ; preds = %cont11, %entry
  ret void
}

So the fix is:

-      assert(I != nullptr);
+
+      // If constant it will fold and we can evaluate later. If its an argument
+      // or something of that nature, we can't analyze.
+      if (I == nullptr)
+        return {nullptr, UndefBit};

I'm happy to post that as a patch.

Created: https://reviews.llvm.org/D142339

But let me know if revert is still preferable (I say as much there).

If you still want to revert, however, happy to do so as well and then repost the full version.

Whichever you think is more prudent.

We're seeing a ton of fallout due to this, and given that there are already two different patterns that cause crashes, I'd like to have a known good version in the tree before we play whack-a-mole ;). Thus, I'd prefer that you reverted this and then recommitted with your proposed fix.

Hi, So as you probably saw, moments before this the change was pushed. The opinion in the other PR seemed to be its best to push the fix.
I don't see anything in the build logs this morning that looks related to the AtomicRMW code (seem to be blocked by something else though),
so my feeling is if things are fine with the 3rd fix, leave up, if we see ANY issue that may be related to the code, immediately reverting all
three is the way to go. Ping back here if you see any issue and I'll revert or you can just do so.

Sorry for the hassle this has caused :(

llvm/include/llvm/IR/IntrinsicsX86.td

 // Lock bit test.
 let TargetPrefix = "x86" in {
   def int_x86_atomic_bts : Intrinsic<[llvm_anyint_ty], [llvm_ptr_ty, llvm_i8_ty],
                                      [ImmArg<ArgIndex<1>>]>;
   def int_x86_atomic_btc : Intrinsic<[llvm_anyint_ty], [llvm_ptr_ty, llvm_i8_ty],
                                      [ImmArg<ArgIndex<1>>]>;
   def int_x86_atomic_btr : Intrinsic<[llvm_anyint_ty], [llvm_ptr_ty, llvm_i8_ty],
                                      [ImmArg<ArgIndex<1>>]>;
+  def int_x86_atomic_bts_rm  : Intrinsic<[llvm_i8_ty], [llvm_ptr_ty, llvm_anyint_ty],
+                                         []>;
+  def int_x86_atomic_btc_rm  : Intrinsic<[llvm_i8_ty], [llvm_ptr_ty, llvm_anyint_ty],
+                                         []>;
+  def int_x86_atomic_btr_rm  : Intrinsic<[llvm_i8_ty], [llvm_ptr_ty, llvm_anyint_ty],
+                                         []>;
+
+
 }
 
 // Lock binary arith with CC.
 let TargetPrefix = "x86" in {
   def int_x86_atomic_add_cc  : Intrinsic<[llvm_i8_ty], [llvm_ptr_ty, llvm_anyint_ty, llvm_i32_ty],
                                          [ImmArg<ArgIndex<2>>]>;
   def int_x86_atomic_sub_cc  : Intrinsic<[llvm_i8_ty], [llvm_ptr_ty, llvm_anyint_ty, llvm_i32_ty],
                                          [ImmArg<ArgIndex<2>>]>;

llvm/lib/Target/X86/X86ISelLowering.h

enum NodeType : unsigned {
     LADD,
     LSUB,
     LOR,
     LXOR,
     LAND,
     LBTS,
     LBTC,
     LBTR,
+    LBTS_RM,
+    LBTC_RM,
+    LBTR_RM,
 
     /// RAO arithmetic instructions.
     /// OUTCHAIN = AADD(INCHAIN, PTR, RHS)
     AADD,
     AOR,
     AXOR,
     AAND,
 

llvm/lib/Target/X86/X86ISelLowering.cpp

case Intrinsic::x86_aesdecwide256kl:
       Info.opc = ISD::INTRINSIC_W_CHAIN;
       Info.ptrVal = I.getArgOperand(0);
       Info.memVT = EVT::getIntegerVT(I.getType()->getContext(), 64);
       Info.align = Align(1);
       Info.flags |= MachineMemOperand::MOLoad;
       return true;
     case Intrinsic::x86_cmpccxadd32:
     case Intrinsic::x86_cmpccxadd64:
+    case Intrinsic::x86_atomic_bts_rm:
+    case Intrinsic::x86_atomic_btc_rm:
+    case Intrinsic::x86_atomic_btr_rm:
     case Intrinsic::x86_atomic_bts:
     case Intrinsic::x86_atomic_btc:
     case Intrinsic::x86_atomic_btr: {
       Info.opc = ISD::INTRINSIC_W_CHAIN;
       Info.ptrVal = I.getArgOperand(0);
       unsigned Size = I.getType()->getScalarSizeInBits();
       Info.memVT = EVT::getIntegerVT(I.getType()->getContext(), Size);
       Info.align = Align(Size);
       Info.flags |= MachineMemOperand::MOLoad | MachineMemOperand::MOStore |
                     MachineMemOperand::MOVolatile;
       return true;
     }
+
     case Intrinsic::x86_aadd32:
     case Intrinsic::x86_aadd64:
     case Intrinsic::x86_aand32:
     case Intrinsic::x86_aand64:
     case Intrinsic::x86_aor32:
     case Intrinsic::x86_aor64:
     case Intrinsic::x86_axor32:
     case Intrinsic::x86_axor64:
case Intrinsic::x86_testui: {
       SDLoc dl(Op);
       SDValue Chain = Op.getOperand(0);
       SDVTList VTs = DAG.getVTList(MVT::i32, MVT::Other);
       SDValue Operation = DAG.getNode(X86ISD::TESTUI, dl, VTs, Chain);
       SDValue SetCC = getSETCC(X86::COND_B, Operation.getValue(0), dl, DAG);
       return DAG.getNode(ISD::MERGE_VALUES, dl, Op->getVTList(), SetCC,
                          Operation.getValue(1));
     }
+    case Intrinsic::x86_atomic_bts_rm:
+    case Intrinsic::x86_atomic_btc_rm:
+    case Intrinsic::x86_atomic_btr_rm: {
+      SDLoc DL(Op);
+      MVT VT = Op.getSimpleValueType();
+      SDValue Chain = Op.getOperand(0);
+      SDValue Op1 = Op.getOperand(2);
+      SDValue Op2 = Op.getOperand(3);
+      unsigned Opc = IntNo == Intrinsic::x86_atomic_bts_rm   ? X86ISD::LBTS_RM
+                     : IntNo == Intrinsic::x86_atomic_btc_rm ? X86ISD::LBTC_RM
+                                                             : X86ISD::LBTR_RM;
+      MachineMemOperand *MMO = cast<MemIntrinsicSDNode>(Op)->getMemOperand();
+      SDValue Res =
+          DAG.getMemIntrinsicNode(Opc, DL, DAG.getVTList(MVT::i32, MVT::Other),
+                                  {Chain, Op1, Op2}, VT, MMO);
+      Chain = Res.getValue(1);
+      Res = DAG.getZExtOrTrunc(getSETCC(X86::COND_B, Res, DL, DAG), DL, VT);
+      return DAG.getNode(ISD::MERGE_VALUES, DL, Op->getVTList(), Res, Chain);
+    }
     case Intrinsic::x86_atomic_bts:
     case Intrinsic::x86_atomic_btc:
     case Intrinsic::x86_atomic_btr: {
       SDLoc DL(Op);
       MVT VT = Op.getSimpleValueType();
       SDValue Chain = Op.getOperand(0);
       SDValue Op1 = Op.getOperand(2);
       SDValue Op2 = Op.getOperand(3);
       unsigned Opc = IntNo == Intrinsic::x86_atomic_bts   ? X86ISD::LBTS
                      : IntNo == Intrinsic::x86_atomic_btc ? X86ISD::LBTC
                                                           : X86ISD::LBTR;
+
       SDValue Size = DAG.getConstant(VT.getScalarSizeInBits(), DL, MVT::i32);
       MachineMemOperand *MMO = cast<MemIntrinsicSDNode>(Op)->getMemOperand();
       SDValue Res =
           DAG.getMemIntrinsicNode(Opc, DL, DAG.getVTList(MVT::i32, MVT::Other),
                                   {Chain, Op1, Op2, Size}, VT, MMO);
       Chain = Res.getValue(1);
       Res = DAG.getZExtOrTrunc(getSETCC(X86::COND_B, Res, DL, DAG), DL, VT);
       unsigned Imm = cast<ConstantSDNode>(Op2)->getZExtValue();
if (MemType->getPrimitiveSizeInBits() == 64 && !Subtarget.is64Bit() &&
       !Subtarget.useSoftFloat() && !NoImplicitFloatOps &&
       (Subtarget.hasSSE1() || Subtarget.hasX87()))
     return AtomicExpansionKind::None;
 
   return needsCmpXchgNb(MemType) ? AtomicExpansionKind::CmpXChg
                                  : AtomicExpansionKind::None;
 }
 
+enum BitTestKind : unsigned {
+  UndefBit,
+  ConstantBit,
+  NotConstantBit,
+  ShiftBit,
+  NotShiftBit
+};
+
+static std::pair<Value *, BitTestKind> FindSingleBitChange(Value *V) {
+  BitTestKind BTK = UndefBit;
+  auto *C = dyn_cast<ConstantInt>(V);
+  if (C) {
+    // Check if V is a power of 2 or or NOT power of 2.
+    if (isPowerOf2_64(C->getZExtValue())) {
+      BTK = ConstantBit;
+    } else if (isPowerOf2_64((~C->getValue()).getZExtValue())) {
+      BTK = NotConstantBit;
+    }
+    return {V, BTK};
+  }
+
+  // Check if V is some power of 2 pattern known to be non-zero
+  auto *I = dyn_cast<Instruction>(V);
+  if (I) {
+    bool Not = false;
+    // Check if we have a NOT
+    if (I->getOpcode() == Instruction::Sub ||
+        I->getOpcode() == Instruction::Xor) {
+
+      auto *OpC0 = dyn_cast<ConstantInt>(I->getOperand(0));
+      auto *OpC1 = dyn_cast<ConstantInt>(I->getOperand(1));
+      // Check if this is a NOT instruction: -1 - X or X/-1 ^ -1/X
+      if (!OpC0 && (!OpC1 || I->getOpcode() == Instruction::Sub))
+        return {nullptr, UndefBit};
+
+      auto *MaybeNeg1 = OpC0 ? OpC0 : OpC1;
+      if (!MaybeNeg1->isMinusOne())
+        return {nullptr, UndefBit};
+
+      auto *OpI0 = dyn_cast<Instruction>(I->getOperand(0));
+      auto *OpI1 = dyn_cast<Instruction>(I->getOperand(1));
+
+      assert(OpI0 != nullptr || OpI1 != nullptr);
+      assert(OpI0 == nullptr || OpI1 == nullptr);
+
+      I = OpI0 ? OpI0 : OpI1;
+      Not = true;
+    }
+    // We can only use 1 << X without more sophisticated analysis. C << X where
+    // C is a power of 2 but not 1 can result in zero which cannot be translated
+    // to bittest. Likewise any C >> X (either arith or logical) can be zero.
+    if (I->getOpcode() == Instruction::Shl) {
+      // Todo(1): The cmpxchg case is pretty costly so matching `BLSI(X)`, `X &
+      // -X` and some other provable power of 2 patterns that we can use CTZ on
+      // may be profitable.
+      // Todo(2): It may be possible in some cases to prove that Shl(C, X) is
+      // non-zero even where C != 1. Likewise LShr(C, X) and AShr(C, X) may also
+      // be provably a non-zero power of 2.
+      // Todo(3): ROTL and ROTR patterns on a power of 2 C should also be
+      // transformable to bittest.
+      auto *ShiftVal = dyn_cast<ConstantInt>(I->getOperand(0));
+      if (!ShiftVal)
+        return {nullptr, UndefBit};
+      if (ShiftVal->equalsInt(1))
+        BTK = Not ? NotShiftBit : ShiftBit;
+
+      if (BTK == UndefBit)
+        return {nullptr, UndefBit};
+
+      Value *BitV = I->getOperand(1);
+      if (auto *I1 = dyn_cast<Instruction>(BitV)) {
+        // Read past a shiftmask instruction to find count
+        if (I1->getOpcode() == Instruction::And) {
+          auto *OpC0 = dyn_cast<ConstantInt>(I1->getOperand(0));
+          auto *OpC1 = dyn_cast<ConstantInt>(I1->getOperand(1));
+          if (OpC0 || OpC1) {
+            assert(OpC0 == nullptr || OpC1 == nullptr);
+            auto *C1 = OpC0 ? OpC0 : OpC1;
+            if (C1->equalsInt(I->getType()->getPrimitiveSizeInBits() - 1))
+              BitV = OpC0 ? I1->getOperand(1) : I1->getOperand(0);
+          }
+        }
+      }
+
+      return {BitV, BTK};
+    }
+  }
+  return {nullptr, UndefBit};
+}
+
 TargetLowering::AtomicExpansionKind
 X86TargetLowering::shouldExpandLogicAtomicRMWInIR(AtomicRMWInst *AI) const {
   // If the atomicrmw's result isn't actually used, we can just add a "lock"
   // prefix to a normal instruction for these operations.
   if (AI->use_empty())
     return AtomicExpansionKind::None;
 
   // If the atomicrmw's result is used by a single bit AND, we may use
   // bts/btr/btc instruction for these operations.
-  auto *C1 = dyn_cast<ConstantInt>(AI->getValOperand());
+  // Note: InstCombinePass can cause a de-optimization here. It replaces the
+  // SETCC(And(AtomicRMW(P, power_of_2), power_of_2)) with LShr and Xor
+  // (depending on CC). This pattern can only use bts/btr/btc but we don't
+  // detect it.
   Instruction *I = AI->user_back();
-  if (!C1 || !AI->hasOneUse() || I->getOpcode() != Instruction::And ||
+  auto BitChange = FindSingleBitChange(AI->getValOperand());
+  if (!BitChange.first || BitChange.second == UndefBit || !AI->hasOneUse() ||
+      I->getOpcode() != Instruction::And ||
+      AI->getType()->getPrimitiveSizeInBits() == 8 ||
       AI->getParent() != I->getParent())
     return AtomicExpansionKind::CmpXChg;
+
+  assert(I->getOperand(0) == AI);
   // The following instruction must be a AND single bit.
+  if (BitChange.second == ConstantBit || BitChange.second == NotConstantBit) {
+    auto *C1 = dyn_cast<ConstantInt>(AI->getValOperand());
+    assert(C1 != nullptr);
     auto *C2 = dyn_cast<ConstantInt>(I->getOperand(1));
-  unsigned Bits = AI->getType()->getPrimitiveSizeInBits();
-  if (!C2 || Bits == 8 || !isPowerOf2_64(C2->getZExtValue()))
+    if (!C2 || !isPowerOf2_64(C2->getZExtValue())) {
+      return AtomicExpansionKind::CmpXChg;
+    }
+    if (AI->getOperation() == AtomicRMWInst::And) {
+      return ~C1->getValue() == C2->getValue()
+                 ? AtomicExpansionKind::BitTestIntrinsic
+                 : AtomicExpansionKind::CmpXChg;
+    }
+    return C1 == C2 ? AtomicExpansionKind::BitTestIntrinsic
+                    : AtomicExpansionKind::CmpXChg;
+  }
+
+  assert(BitChange.second == ShiftBit || BitChange.second == NotShiftBit);
+
+  auto BitTested = FindSingleBitChange(I->getOperand(1));
+  if (BitTested.second != ShiftBit && BitTested.second != NotShiftBit)
+    return AtomicExpansionKind::CmpXChg;
+
+  assert(BitTested.first != nullptr);
+
+  // If shift amounts are not the same we can't use BitTestIntrinsic
+  if (BitChange.first != BitTested.first)
     return AtomicExpansionKind::CmpXChg;
 
+  // If atomic AND need to be masking all be one bit and testing the one bit
+  // unset in the mask
   if (AI->getOperation() == AtomicRMWInst::And)
-    return ~C1->getValue() == C2->getValue()
+    return (BitChange.second == NotShiftBit && BitTested.second == ShiftBit)
                ? AtomicExpansionKind::BitTestIntrinsic
                : AtomicExpansionKind::CmpXChg;
 
-  return C1 == C2 ? AtomicExpansionKind::BitTestIntrinsic
-                  : AtomicExpansionKind::CmpXChg;
+  // If atomic XOR/OR need to be setting and testing the same bit.
+  return (BitChange.second == ShiftBit && BitTested.second == ShiftBit)
+             ? AtomicExpansionKind::BitTestIntrinsic
 }
 
 void X86TargetLowering::emitBitTestAtomicRMWIntrinsic(AtomicRMWInst *AI) const {
   IRBuilder<> Builder(AI);
-  Intrinsic::ID IID = Intrinsic::not_intrinsic;
+  Intrinsic::ID IID_C = Intrinsic::not_intrinsic;
+  Intrinsic::ID IID_I = Intrinsic::not_intrinsic;
   switch (AI->getOperation()) {
   default:
     llvm_unreachable("Unknown atomic operation");
   case AtomicRMWInst::Or:
-    IID = Intrinsic::x86_atomic_bts;
+    IID_C = Intrinsic::x86_atomic_bts;
+    IID_I = Intrinsic::x86_atomic_bts_rm;
     break;
   case AtomicRMWInst::Xor:
-    IID = Intrinsic::x86_atomic_btc;
+    IID_C = Intrinsic::x86_atomic_btc;
+    IID_I = Intrinsic::x86_atomic_btc_rm;
     break;
   case AtomicRMWInst::And:
-    IID = Intrinsic::x86_atomic_btr;
+    IID_C = Intrinsic::x86_atomic_btr;
+    IID_I = Intrinsic::x86_atomic_btr_rm;
     break;
   }
   Instruction *I = AI->user_back();
   LLVMContext &Ctx = AI->getContext();
-  unsigned Imm =
-      countTrailingZeros(cast<ConstantInt>(I->getOperand(1))->getZExtValue());
-  Function *BitTest =
-      Intrinsic::getDeclaration(AI->getModule(), IID, AI->getType());
   Value *Addr = Builder.CreatePointerCast(AI->getPointerOperand(),
                                           Type::getInt8PtrTy(Ctx));
-  Value *Result = Builder.CreateCall(BitTest, {Addr, Builder.getInt8(Imm)});
+  Function *BitTest = nullptr;
+  Value *Result = nullptr;
+  auto BitTested = FindSingleBitChange(AI->getValOperand());
+  assert(BitTested.first != nullptr);
+  if (BitTested.second == ConstantBit || BitTested.second == NotConstantBit) {
+    auto *C = dyn_cast<ConstantInt>(I->getOperand(1));
+    assert(C != nullptr);
+
+    BitTest = Intrinsic::getDeclaration(AI->getModule(), IID_C, AI->getType());
+
+    unsigned Imm = countTrailingZeros(C->getZExtValue());
+    Result = Builder.CreateCall(BitTest, {Addr, Builder.getInt8(Imm)});
+  } else {
+    BitTest = Intrinsic::getDeclaration(AI->getModule(), IID_I, AI->getType());
+
+    assert(BitTested.second == ShiftBit || BitTested.second == NotShiftBit);
+
+    Value *SI = BitTested.first;
+    assert(SI != nullptr);
+
+    // BT{S|R|C} on memory operand don't modulo bit position so we need to
+    // mask it.
+    unsigned ShiftBits = SI->getType()->getPrimitiveSizeInBits();
+    Value *BitPos =
+        Builder.CreateAnd(SI, Builder.getIntN(ShiftBits, ShiftBits - 1));
+    // Todo(1): In many cases it may be provable that SI is less than
+    // ShiftBits in which case this mask is unnecessary
+    // Todo(2): In the fairly idiomatic case of P[X / sizeof_bits(X)] OP 1
+    // << (X % sizeof_bits(X)) we can drop the shift mask and AGEN in
+    // favor of just a raw BT{S|R|C}.
+
+    Result = Builder.CreateCall(BitTest, {Addr, BitPos});
+    Result = Builder.CreateZExtOrTrunc(Result, AI->getType());
+
+    // If the result is only used for zero/non-zero status then we don't need to
+    // shift value back. Otherwise do so.
+    for (auto It = I->user_begin(); It != I->user_end(); ++It) {
+      if (auto *ICmp = dyn_cast<ICmpInst>(*It)) {
+        if (ICmp->isEquality()) {
+          auto *C0 = dyn_cast<ConstantInt>(ICmp->getOperand(0));
+          auto *C1 = dyn_cast<ConstantInt>(ICmp->getOperand(1));
+          if (C0 || C1) {
+            assert(C0 == nullptr || C1 == nullptr);
+            if ((C0 ? C0 : C1)->isZero())
+              continue;
+          }
+        }
+      }
+      Result = Builder.CreateShl(Result, BitPos);
+      break;
+    }
+  }
+
   I->replaceAllUsesWith(Result);
   I->eraseFromParent();
   AI->eraseFromParent();
 }
 
 static bool shouldExpandCmpArithRMWInIR(AtomicRMWInst *AI) {
   using namespace llvm::PatternMatch;
   if (!AI->hasOneUse())
#define NODE_NAME_CASE(NODE) case X86ISD::NODE: return "X86ISD::" #NODE;
   NODE_NAME_CASE(LADD)
   NODE_NAME_CASE(LSUB)
   NODE_NAME_CASE(LOR)
   NODE_NAME_CASE(LXOR)
   NODE_NAME_CASE(LAND)
   NODE_NAME_CASE(LBTS)
   NODE_NAME_CASE(LBTC)
   NODE_NAME_CASE(LBTR)
+  NODE_NAME_CASE(LBTS_RM)
+  NODE_NAME_CASE(LBTC_RM)
+  NODE_NAME_CASE(LBTR_RM)
   NODE_NAME_CASE(AADD)
   NODE_NAME_CASE(AOR)
   NODE_NAME_CASE(AXOR)
   NODE_NAME_CASE(AAND)
   NODE_NAME_CASE(VZEXT_MOVL)
   NODE_NAME_CASE(VZEXT_LOAD)
   NODE_NAME_CASE(VEXTRACT_STORE)
   NODE_NAME_CASE(VTRUNC)

llvm/lib/Target/X86/X86InstrCompiler.td

def X86LBTest : SDTypeProfile<1, 3, [SDTCisVT<0, i32>, SDTCisPtrTy<1>,
                                      SDTCisVT<2, i8>, SDTCisVT<3, i32>]>;
 def x86bts : SDNode<"X86ISD::LBTS", X86LBTest,
                     [SDNPHasChain, SDNPMayLoad, SDNPMayStore, SDNPMemOperand]>;
 def x86btc : SDNode<"X86ISD::LBTC", X86LBTest,
                     [SDNPHasChain, SDNPMayLoad, SDNPMayStore, SDNPMemOperand]>;
 def x86btr : SDNode<"X86ISD::LBTR", X86LBTest,
                     [SDNPHasChain, SDNPMayLoad, SDNPMayStore, SDNPMemOperand]>;
 
+def X86LBTestRM : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<1>,
+                                       SDTCisInt<2>]>;
+
+def x86_rm_bts : SDNode<"X86ISD::LBTS_RM", X86LBTestRM,
+                        [SDNPHasChain, SDNPMayLoad, SDNPMayStore, SDNPMemOperand]>;
+def x86_rm_btc : SDNode<"X86ISD::LBTC_RM", X86LBTestRM,
+                        [SDNPHasChain, SDNPMayLoad, SDNPMayStore, SDNPMemOperand]>;
+def x86_rm_btr : SDNode<"X86ISD::LBTR_RM", X86LBTestRM,
+                        [SDNPHasChain, SDNPMayLoad, SDNPMayStore, SDNPMemOperand]>;
+
+
 multiclass ATOMIC_LOGIC_OP<Format Form, string s> {
   let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1,
       SchedRW = [WriteBitTestSetRegRMW]  in {
     def 16m : Ii8<0xBA, Form, (outs), (ins i16mem:$src1, i8imm:$src2),
                   !strconcat(s, "{w}\t{$src2, $src1|$src1, $src2}"),
                   [(set EFLAGS, (!cast<SDNode>("x86" # s) addr:$src1, timm:$src2, (i32 16)))]>,
               OpSize16, TB, LOCK;
     def 32m : Ii8<0xBA, Form, (outs), (ins i32mem:$src1, i8imm:$src2),
                   !strconcat(s, "{l}\t{$src2, $src1|$src1, $src2}"),
                   [(set EFLAGS, (!cast<SDNode>("x86" # s) addr:$src1, timm:$src2, (i32 32)))]>,
               OpSize32, TB, LOCK;
     def 64m : RIi8<0xBA, Form, (outs), (ins i64mem:$src1, i8imm:$src2),
                    !strconcat(s, "{q}\t{$src2, $src1|$src1, $src2}"),
                    [(set EFLAGS, (!cast<SDNode>("x86" # s) addr:$src1, timm:$src2, (i32 64)))]>,
               TB, LOCK;
   }
 }
 
+multiclass ATOMIC_LOGIC_OP_RM<bits<8> Opc8, string s> {
+  let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1,
+      SchedRW = [WriteBitTestSetRegRMW]  in {
+    def 16rm : Ii8<Opc8, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2),
+                  !strconcat(s, "{w}\t{$src2, $src1|$src1, $src2}"),
+                  [(set EFLAGS, (!cast<SDNode>("x86_rm_" # s) addr:$src1, GR16:$src2))]>,
+               OpSize16, TB, LOCK;
+    def 32rm : Ii8<Opc8, MRMDestMem, (outs), (ins i32mem:$src1, GR32:$src2),
+                  !strconcat(s, "{l}\t{$src2, $src1|$src1, $src2}"),
+                  [(set EFLAGS, (!cast<SDNode>("x86_rm_" # s) addr:$src1, GR32:$src2))]>,
+               OpSize32, TB, LOCK;
+    def 64rm : RIi8<Opc8, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2),
+                   !strconcat(s, "{q}\t{$src2, $src1|$src1, $src2}"),
+                   [(set EFLAGS, (!cast<SDNode>("x86_rm_" # s) addr:$src1, GR64:$src2))]>,
+               TB, LOCK;
+  }
+}
+
+
 defm LOCK_BTS : ATOMIC_LOGIC_OP<MRM5m, "bts">;
 defm LOCK_BTC : ATOMIC_LOGIC_OP<MRM7m, "btc">;
 defm LOCK_BTR : ATOMIC_LOGIC_OP<MRM6m, "btr">;
 
+defm LOCK_BTS_RM : ATOMIC_LOGIC_OP_RM<0xAB, "bts">;
+defm LOCK_BTC_RM : ATOMIC_LOGIC_OP_RM<0xBB, "btc">;
+defm LOCK_BTR_RM : ATOMIC_LOGIC_OP_RM<0xB3, "btr">;
+
 // Atomic compare and swap.
 multiclass LCMPXCHG_BinOp<bits<8> Opc8, bits<8> Opc, Format Form,
                           string mnemonic, SDPatternOperator frag> {
 let isCodeGenOnly = 1, SchedRW = [WriteCMPXCHGRMW] in {
   let Defs = [AL, EFLAGS], Uses = [AL] in
   def NAME#8  : I<Opc8, Form, (outs), (ins i8mem:$ptr, GR8:$swap),
                   !strconcat(mnemonic, "{b}\t{$swap, $ptr|$ptr, $swap}"),
                   [(frag addr:$ptr, GR8:$swap, 1)]>, TB, LOCK;

llvm/test/CodeGen/X86/atomic-rm-bit-test.ll

entry:
   %0 = atomicrmw xor ptr %v, i16 %conv1 monotonic, align 2
   %conv5 = and i16 %0, %conv1
   ret i16 %conv5
 }
 
 define zeroext i16 @atomic_shl1_small_mask_xor_16_gpr_val(ptr %v, i16 zeroext %c) nounwind {
 ; X86-LABEL: atomic_shl1_small_mask_xor_16_gpr_val:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    andb $7, %cl
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movzwl (%edx), %eax
-; X86-NEXT:    movzwl %si, %ecx
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB13_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %esi
-; X86-NEXT:    xorl %ecx, %esi
-; X86-NEXT:    # kill: def $ax killed $ax killed $eax
-; X86-NEXT:    lock cmpxchgw %si, (%edx)
-; X86-NEXT:    # kill: def $ax killed $ax def $eax
-; X86-NEXT:    jne .LBB13_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    andl %ecx, %eax
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    andl $7, %ecx
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    lock btcw %cx, (%edx)
+; X86-NEXT:    setb %al
+; X86-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NEXT:    shll %cl, %eax
 ; X86-NEXT:    # kill: def $ax killed $ax killed $eax
-; X86-NEXT:    popl %esi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_small_mask_xor_16_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    andb $7, %cl
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $7, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btcw %cx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movzwl (%rdi), %eax
-; X64-NEXT:    movzwl %dx, %ecx
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB13_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %edx
-; X64-NEXT:    xorl %ecx, %edx
-; X64-NEXT:    # kill: def $ax killed $ax killed $eax
-; X64-NEXT:    lock cmpxchgw %dx, (%rdi)
-; X64-NEXT:    # kill: def $ax killed $ax def $eax
-; X64-NEXT:    jne .LBB13_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andl %ecx, %eax
+; X64-NEXT:    shll %cl, %eax
 ; X64-NEXT:    # kill: def $ax killed $ax killed $eax
 ; X64-NEXT:    retq
 entry:
   %0 = and i16 %c, 7
   %shl = shl nuw nsw i16 1, %0
   %1 = atomicrmw xor ptr %v, i16 %shl monotonic, align 2
   %and = and i16 %1, %shl
   ret i16 %and
entry:
   %shl4 = shl nuw i16 1, %1
   %and = and i16 %0, %shl4
   ret i16 %and
 }
 
 define zeroext i16 @atomic_shl1_mask01_xor_16_gpr_val(ptr %v, i16 zeroext %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask01_xor_16_gpr_val:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    andb $15, %cl
-; X86-NEXT:    movl $1, %edx
-; X86-NEXT:    shll %cl, %edx
-; X86-NEXT:    movzwl (%esi), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB16_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    xorl %edx, %ecx
-; X86-NEXT:    # kill: def $ax killed $ax killed $eax
-; X86-NEXT:    lock cmpxchgw %cx, (%esi)
-; X86-NEXT:    # kill: def $ax killed $ax def $eax
-; X86-NEXT:    jne .LBB16_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    andl %edx, %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    andl $15, %ecx
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    lock btcw %cx, (%edx)
+; X86-NEXT:    setb %al
+; X86-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NEXT:    shll %cl, %eax
 ; X86-NEXT:    # kill: def $ax killed $ax killed $eax
-; X86-NEXT:    popl %esi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask01_xor_16_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    andb $15, %cl
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $15, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btcw %cx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movzwl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB16_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    xorl %edx, %ecx
-; X64-NEXT:    # kill: def $ax killed $ax killed $eax
-; X64-NEXT:    lock cmpxchgw %cx, (%rdi)
-; X64-NEXT:    # kill: def $ax killed $ax def $eax
-; X64-NEXT:    jne .LBB16_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andl %edx, %eax
+; X64-NEXT:    shll %cl, %eax
 ; X64-NEXT:    # kill: def $ax killed $ax killed $eax
 ; X64-NEXT:    retq
 entry:
   %0 = and i16 %c, 15
   %shl = shl nuw i16 1, %0
   %1 = atomicrmw xor ptr %v, i16 %shl monotonic, align 2
   %conv7 = and i16 %1, %shl
   ret i16 %conv7
entry:
   %1 = atomicrmw and ptr %v, i16 %conv1 monotonic, align 2
   %conv5 = and i16 %1, %0
   ret i16 %conv5
 }
 
 define zeroext i16 @atomic_shl1_small_mask_and_16_gpr_val(ptr %v, i16 zeroext %c) nounwind {
 ; X86-LABEL: atomic_shl1_small_mask_and_16_gpr_val:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    andb $7, %cl
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movw $-2, %di
-; X86-NEXT:    rolw %cl, %di
-; X86-NEXT:    movzwl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB37_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    andl %edi, %ecx
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    andl $7, %ecx
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    lock btrw %cx, (%edx)
+; X86-NEXT:    setb %al
+; X86-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NEXT:    shll %cl, %eax
 ; X86-NEXT:    # kill: def $ax killed $ax killed $eax
-; X86-NEXT:    lock cmpxchgw %cx, (%edx)
-; X86-NEXT:    # kill: def $ax killed $ax def $eax
-; X86-NEXT:    jne .LBB37_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    movzwl %si, %ecx
-; X86-NEXT:    andl %eax, %ecx
-; X86-NEXT:    movl %ecx, %eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_small_mask_and_16_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    andb $7, %cl
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movw $-2, %si
+; X64-NEXT:    andl $7, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btrw %cx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-NEXT:    rolw %cl, %si
-; X64-NEXT:    movzwl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB37_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    andl %esi, %ecx
+; X64-NEXT:    shll %cl, %eax
 ; X64-NEXT:    # kill: def $ax killed $ax killed $eax
-; X64-NEXT:    lock cmpxchgw %cx, (%rdi)
-; X64-NEXT:    # kill: def $ax killed $ax def $eax
-; X64-NEXT:    jne .LBB37_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movzwl %dx, %ecx
-; X64-NEXT:    andl %eax, %ecx
-; X64-NEXT:    movl %ecx, %eax
 ; X64-NEXT:    retq
 entry:
   %0 = and i16 %c, 7
   %shl = shl nuw nsw i16 1, %0
   %not = xor i16 %shl, -1
   %1 = atomicrmw and ptr %v, i16 %not monotonic, align 2
   %and = and i16 %1, %shl
   ret i16 %and
entry:
   %shl4 = shl nuw i16 1, %2
   %and = and i16 %1, %shl4
   ret i16 %and
 }
 
 define zeroext i16 @atomic_shl1_mask01_and_16_gpr_val(ptr %v, i16 zeroext %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask01_and_16_gpr_val:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    andb $15, %cl
-; X86-NEXT:    movl $1, %edx
-; X86-NEXT:    shll %cl, %edx
-; X86-NEXT:    movw $-2, %di
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    rolw %cl, %di
-; X86-NEXT:    movzwl (%esi), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB40_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    andl %edi, %ecx
-; X86-NEXT:    # kill: def $ax killed $ax killed $eax
-; X86-NEXT:    lock cmpxchgw %cx, (%esi)
-; X86-NEXT:    # kill: def $ax killed $ax def $eax
-; X86-NEXT:    jne .LBB40_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    andl %edx, %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    andl $15, %ecx
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    lock btrw %cx, (%edx)
+; X86-NEXT:    setb %al
+; X86-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NEXT:    shll %cl, %eax
 ; X86-NEXT:    # kill: def $ax killed $ax killed $eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask01_and_16_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    andb $15, %cl
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movw $-2, %r8w
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    rolw %cl, %r8w
-; X64-NEXT:    movzwl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB40_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    andl %r8d, %ecx
-; X64-NEXT:    # kill: def $ax killed $ax killed $eax
-; X64-NEXT:    lock cmpxchgw %cx, (%rdi)
-; X64-NEXT:    # kill: def $ax killed $ax def $eax
-; X64-NEXT:    jne .LBB40_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andl %edx, %eax
+; X64-NEXT:    andl $15, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btrw %cx, (%rdi)
+; X64-NEXT:    setb %al
+; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NEXT:    shll %cl, %eax
 ; X64-NEXT:    # kill: def $ax killed $ax killed $eax
 ; X64-NEXT:    retq
 entry:
   %0 = and i16 %c, 15
   %shl = shl nuw i16 1, %0
   %conv1 = xor i16 %shl, -1
   %1 = atomicrmw and ptr %v, i16 %conv1 monotonic, align 2
   %conv7 = and i16 %1, %shl

 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i16 [ %2, %if.then ], [ 123, %entry ]
   ret i16 %retval.0
 }
 
 define i32 @atomic_shl1_or_32_gpr_val(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_or_32_gpr_val:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl $1, %edx
-; X86-NEXT:    shll %cl, %edx
-; X86-NEXT:    movl (%esi), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB60_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    orl %edx, %ecx
-; X86-NEXT:    lock cmpxchgl %ecx, (%esi)
-; X86-NEXT:    jne .LBB60_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    andl %edx, %eax
-; X86-NEXT:    popl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    andl $31, %ecx
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    lock btsl %ecx, (%edx)
+; X86-NEXT:    setb %al
+; X86-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NEXT:    shll %cl, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_or_32_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $31, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btsl %ecx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB60_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    orl %edx, %ecx
-; X64-NEXT:    lock cmpxchgl %ecx, (%rdi)
-; X64-NEXT:    jne .LBB60_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andl %edx, %eax
+; X64-NEXT:    shll %cl, %eax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i32 1, %c
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %0, %shl
   ret i32 %and
 }
 
 define i32 @atomic_shl1_small_mask_or_32_gpr_val(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_small_mask_or_32_gpr_val:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    andb $15, %cl
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB61_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    orl %esi, %ecx
-; X86-NEXT:    lock cmpxchgl %ecx, (%edx)
-; X86-NEXT:    jne .LBB61_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    andl %esi, %eax
-; X86-NEXT:    popl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    andl $15, %ecx
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    lock btsl %ecx, (%edx)
+; X86-NEXT:    setb %al
+; X86-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NEXT:    shll %cl, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_small_mask_or_32_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    andb $15, %cl
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $15, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btsl %ecx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB61_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    orl %edx, %ecx
-; X64-NEXT:    lock cmpxchgl %ecx, (%rdi)
-; X64-NEXT:    jne .LBB61_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andl %edx, %eax
+; X64-NEXT:    shll %cl, %eax
 ; X64-NEXT:    retq
 entry:
   %0 = and i32 %c, 15
   %shl = shl nuw nsw i32 1, %0
   %1 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %1, %shl
   ret i32 %and
 }
 
 define i32 @atomic_shl1_mask0_or_32_gpr_val(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask0_or_32_gpr_val:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB62_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB62_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    movl $1, %edx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    andl $31, %ecx
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    lock btsl %ecx, (%edx)
+; X86-NEXT:    setb %al
 ; X86-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X86-NEXT:    shll %cl, %edx
-; X86-NEXT:    andl %edx, %eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
+; X86-NEXT:    shll %cl, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask0_or_32_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB62_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB62_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $31, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btsl %ecx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    andl %edx, %eax
+; X64-NEXT:    shll %cl, %eax
 ; X64-NEXT:    retq
 entry:
   %0 = and i32 %c, 31
   %shl = shl nuw i32 1, %0
   %1 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %shl1 = shl nuw i32 1, %c
   %and = and i32 %1, %shl1
   ret i32 %and
 }
 
 define i32 @atomic_shl1_mask1_or_32_gpr_val(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask1_or_32_gpr_val:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB63_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB63_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    movl $1, %edx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    andl $31, %ecx
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    lock btsl %ecx, (%edx)
+; X86-NEXT:    setb %al
 ; X86-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X86-NEXT:    shll %cl, %edx
-; X86-NEXT:    andl %edx, %eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
+; X86-NEXT:    shll %cl, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask1_or_32_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB63_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB63_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $31, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btsl %ecx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    andl %edx, %eax
+; X64-NEXT:    shll %cl, %eax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i32 1, %c
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %1 = and i32 %c, 31
   %shl1 = shl nuw i32 1, %1
   %and = and i32 %0, %shl1
   ret i32 %and
 }
 
 define i32 @atomic_shl1_mask01_or_32_gpr_val(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask01_or_32_gpr_val:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl $1, %edx
-; X86-NEXT:    shll %cl, %edx
-; X86-NEXT:    movl (%esi), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB64_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    orl %edx, %ecx
-; X86-NEXT:    lock cmpxchgl %ecx, (%esi)
-; X86-NEXT:    jne .LBB64_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    andl %edx, %eax
-; X86-NEXT:    popl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    andl $31, %ecx
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:    lock btsl %ecx, (%edx)
+; X86-NEXT:    setb %al
+; X86-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NEXT:    shll %cl, %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask01_or_32_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $31, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btsl %ecx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB64_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    orl %edx, %ecx
-; X64-NEXT:    lock cmpxchgl %ecx, (%rdi)
-; X64-NEXT:    jne .LBB64_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andl %edx, %eax
+; X64-NEXT:    shll %cl, %eax
 ; X64-NEXT:    retq
 entry:
   %0 = and i32 %c, 31
   %shl = shl nuw i32 1, %0
   %1 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %1, %shl
   ret i32 %and
 }
entry:
   %tobool = icmp ne i32 %and3, 0
   %lnot.ext = zext i1 %tobool to i32
   ret i32 %lnot.ext
 }
 
 define i32 @atomic_shl1_or_32_gpr_br(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_or_32_gpr_br:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB78_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB78_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    testl %esi, %eax
-; X86-NEXT:    je .LBB78_3
-; X86-NEXT:  # %bb.4: # %if.then
-; X86-NEXT:    movl (%edx,%ecx,4), %eax
-; X86-NEXT:    jmp .LBB78_5
-; X86-NEXT:  .LBB78_3:
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl $31, %edx
+; X86-NEXT:    lock btsl %edx, (%ecx)
+; X86-NEXT:    jae .LBB78_1
+; X86-NEXT:  # %bb.2: # %if.then
+; X86-NEXT:    movl (%ecx,%eax,4), %eax
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB78_1:
 ; X86-NEXT:    movl $123, %eax
-; X86-NEXT:  .LBB78_5: # %return
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_or_32_gpr_br:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB78_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB78_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    testl %edx, %eax
-; X64-NEXT:    je .LBB78_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    jae .LBB78_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB78_3:
+; X64-NEXT:  .LBB78_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i32 1, %c
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %0, %shl
   %tobool.not = icmp eq i32 %and, 0
   br i1 %tobool.not, label %return, label %if.then
 
 if.then:                                          ; preds = %entry
   %idxprom = zext i32 %c to i64
   %arrayidx = getelementptr inbounds i32, ptr %v, i64 %idxprom
   %1 = load i32, ptr %arrayidx, align 4
   br label %return
 
 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_small_mask_or_32_gpr_br(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_small_mask_or_32_gpr_br:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    andl $15, %ecx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB79_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB79_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    testl %esi, %eax
-; X86-NEXT:    je .LBB79_3
-; X86-NEXT:  # %bb.4: # %if.then
-; X86-NEXT:    movl (%edx,%ecx,4), %eax
-; X86-NEXT:    jmp .LBB79_5
-; X86-NEXT:  .LBB79_3:
+; X86-NEXT:    lock btsl %ecx, (%eax)
+; X86-NEXT:    jae .LBB79_1
+; X86-NEXT:  # %bb.2: # %if.then
+; X86-NEXT:    movl (%eax,%ecx,4), %eax
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB79_1:
 ; X86-NEXT:    movl $123, %eax
-; X86-NEXT:  .LBB79_5: # %return
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_small_mask_or_32_gpr_br:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    andl $15, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB79_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB79_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    testl %edx, %eax
-; X64-NEXT:    je .LBB79_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    andl $15, %esi
+; X64-NEXT:    lock btsl %esi, (%rdi)
+; X64-NEXT:    jae .LBB79_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB79_3:
+; X64-NEXT:  .LBB79_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %0 = and i32 %c, 15
   %shl = shl nuw nsw i32 1, %0
   %1 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %1, %shl
   %tobool.not = icmp eq i32 %and, 0
   br i1 %tobool.not, label %return, label %if.then
 
 if.then:                                          ; preds = %entry
   %conv2 = zext i32 %0 to i64
   %arrayidx = getelementptr inbounds i32, ptr %v, i64 %conv2
   %2 = load i32, ptr %arrayidx, align 4
   br label %return
 
 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %2, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_mask0_or_32_gpr_br(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask0_or_32_gpr_br:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB80_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB80_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    btl %ecx, %eax
-; X86-NEXT:    jae .LBB80_3
-; X86-NEXT:  # %bb.4: # %if.then
-; X86-NEXT:    movl (%edx,%ecx,4), %eax
-; X86-NEXT:    jmp .LBB80_5
-; X86-NEXT:  .LBB80_3:
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl $31, %edx
+; X86-NEXT:    lock btsl %edx, (%ecx)
+; X86-NEXT:    jae .LBB80_1
+; X86-NEXT:  # %bb.2: # %if.then
+; X86-NEXT:    movl (%ecx,%eax,4), %eax
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB80_1:
 ; X86-NEXT:    movl $123, %eax
-; X86-NEXT:  .LBB80_5: # %return
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask0_or_32_gpr_br:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB80_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB80_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    btl %ecx, %eax
-; X64-NEXT:    jae .LBB80_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    jae .LBB80_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB80_3:
+; X64-NEXT:  .LBB80_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %rem = and i32 %c, 31
   %shl = shl nuw i32 1, %rem
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %shl1 = shl nuw i32 1, %c
   %and = and i32 %0, %shl1

 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_mask1_or_32_gpr_br(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask1_or_32_gpr_br:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB81_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB81_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    btl %ecx, %eax
-; X86-NEXT:    jae .LBB81_3
-; X86-NEXT:  # %bb.4: # %if.then
-; X86-NEXT:    movl (%edx,%ecx,4), %eax
-; X86-NEXT:    jmp .LBB81_5
-; X86-NEXT:  .LBB81_3:
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl $31, %edx
+; X86-NEXT:    lock btsl %edx, (%ecx)
+; X86-NEXT:    jae .LBB81_1
+; X86-NEXT:  # %bb.2: # %if.then
+; X86-NEXT:    movl (%ecx,%eax,4), %eax
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB81_1:
 ; X86-NEXT:    movl $123, %eax
-; X86-NEXT:  .LBB81_5: # %return
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask1_or_32_gpr_br:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB81_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB81_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    btl %ecx, %eax
-; X64-NEXT:    jae .LBB81_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    jae .LBB81_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB81_3:
+; X64-NEXT:  .LBB81_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i32 1, %c
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %rem = and i32 %c, 31
   %shl1 = shl nuw i32 1, %rem
   %and = and i32 %0, %shl1

 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_mask01_or_32_gpr_br(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask01_or_32_gpr_br:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB82_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB82_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    testl %esi, %eax
-; X86-NEXT:    je .LBB82_3
-; X86-NEXT:  # %bb.4: # %if.then
-; X86-NEXT:    movl (%edx,%ecx,4), %eax
-; X86-NEXT:    jmp .LBB82_5
-; X86-NEXT:  .LBB82_3:
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl $31, %edx
+; X86-NEXT:    lock btsl %edx, (%ecx)
+; X86-NEXT:    jae .LBB82_1
+; X86-NEXT:  # %bb.2: # %if.then
+; X86-NEXT:    movl (%ecx,%eax,4), %eax
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB82_1:
 ; X86-NEXT:    movl $123, %eax
-; X86-NEXT:  .LBB82_5: # %return
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask01_or_32_gpr_br:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB82_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB82_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    testl %edx, %eax
-; X64-NEXT:    je .LBB82_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    jae .LBB82_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB82_3:
+; X64-NEXT:  .LBB82_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %rem = and i32 %c, 31
   %shl = shl nuw i32 1, %rem
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %0, %shl
   %tobool.not = icmp eq i32 %and, 0

 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_or_32_gpr_brz(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_or_32_gpr_brz:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movl $1, %edi
-; X86-NEXT:    shll %cl, %edi
-; X86-NEXT:    movl (%esi), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB84_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    orl %edi, %edx
-; X86-NEXT:    lock cmpxchgl %edx, (%esi)
-; X86-NEXT:    jne .LBB84_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    movl $123, %edx
-; X86-NEXT:    testl %edi, %eax
-; X86-NEXT:    jne .LBB84_4
-; X86-NEXT:  # %bb.3: # %if.then
-; X86-NEXT:    movl (%esi,%ecx,4), %edx
-; X86-NEXT:  .LBB84_4: # %return
-; X86-NEXT:    movl %edx, %eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    andl $31, %eax
+; X86-NEXT:    lock btsl %eax, (%edx)
+; X86-NEXT:    movl $123, %eax
+; X86-NEXT:    jae .LBB84_1
+; X86-NEXT:  # %bb.2: # %return
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB84_1: # %if.then
+; X86-NEXT:    movl (%edx,%ecx,4), %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_or_32_gpr_brz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %esi
-; X64-NEXT:    shll %cl, %esi
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB84_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %edx
-; X64-NEXT:    orl %esi, %edx
-; X64-NEXT:    lock cmpxchgl %edx, (%rdi)
-; X64-NEXT:    jne .LBB84_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $123, %edx
-; X64-NEXT:    testl %esi, %eax
-; X64-NEXT:    je .LBB84_3
-; X64-NEXT:  # %bb.4: # %return
-; X64-NEXT:    movl %edx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    movl $123, %eax
+; X64-NEXT:    jae .LBB84_1
+; X64-NEXT:  # %bb.2: # %return
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB84_3: # %if.then
-; X64-NEXT:    movl %ecx, %eax
-; X64-NEXT:    movl (%rdi,%rax,4), %edx
-; X64-NEXT:    movl %edx, %eax
+; X64-NEXT:  .LBB84_1: # %if.then
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i32 1, %c
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %0, %shl
   %tobool.not = icmp eq i32 %and, 0
   br i1 %tobool.not, label %if.then, label %return
 
 if.then:                                          ; preds = %entry
   %idxprom = zext i32 %c to i64
   %arrayidx = getelementptr inbounds i32, ptr %v, i64 %idxprom
   %1 = load i32, ptr %arrayidx, align 4
   br label %return
 
 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_small_mask_or_32_gpr_brz(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_small_mask_or_32_gpr_brz:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    andl $15, %ecx
-; X86-NEXT:    movl $1, %edi
-; X86-NEXT:    shll %cl, %edi
-; X86-NEXT:    movl (%esi), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB85_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    orl %edi, %edx
-; X86-NEXT:    lock cmpxchgl %edx, (%esi)
-; X86-NEXT:    jne .LBB85_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    movl $123, %edx
-; X86-NEXT:    testl %edi, %eax
-; X86-NEXT:    jne .LBB85_4
-; X86-NEXT:  # %bb.3: # %if.then
-; X86-NEXT:    movl (%esi,%ecx,4), %edx
-; X86-NEXT:  .LBB85_4: # %return
-; X86-NEXT:    movl %edx, %eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    andl $15, %edx
+; X86-NEXT:    lock btsl %edx, (%ecx)
+; X86-NEXT:    movl $123, %eax
+; X86-NEXT:    jae .LBB85_1
+; X86-NEXT:  # %bb.2: # %return
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB85_1: # %if.then
+; X86-NEXT:    movl (%ecx,%edx,4), %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_small_mask_or_32_gpr_brz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    andl $15, %ecx
-; X64-NEXT:    movl $1, %esi
-; X64-NEXT:    shll %cl, %esi
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB85_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %edx
-; X64-NEXT:    orl %esi, %edx
-; X64-NEXT:    lock cmpxchgl %edx, (%rdi)
-; X64-NEXT:    jne .LBB85_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $123, %edx
-; X64-NEXT:    testl %esi, %eax
-; X64-NEXT:    je .LBB85_3
-; X64-NEXT:  # %bb.4: # %return
-; X64-NEXT:    movl %edx, %eax
+; X64-NEXT:    andl $15, %esi
+; X64-NEXT:    lock btsl %esi, (%rdi)
+; X64-NEXT:    movl $123, %eax
+; X64-NEXT:    jae .LBB85_1
+; X64-NEXT:  # %bb.2: # %return
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB85_3: # %if.then
-; X64-NEXT:    movl %ecx, %eax
-; X64-NEXT:    movl (%rdi,%rax,4), %edx
-; X64-NEXT:    movl %edx, %eax
+; X64-NEXT:  .LBB85_1: # %if.then
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
 entry:
   %0 = and i32 %c, 15
   %shl = shl nuw nsw i32 1, %0
   %1 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %1, %shl
   %tobool.not = icmp eq i32 %and, 0
   br i1 %tobool.not, label %if.then, label %return
 
 if.then:                                          ; preds = %entry
   %conv2 = zext i32 %0 to i64
   %arrayidx = getelementptr inbounds i32, ptr %v, i64 %conv2
   %2 = load i32, ptr %arrayidx, align 4
   br label %return
 
 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %2, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_mask0_or_32_gpr_brz(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask0_or_32_gpr_brz:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movl $1, %edx
-; X86-NEXT:    shll %cl, %edx
-; X86-NEXT:    movl (%esi), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB86_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %edx, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%esi)
-; X86-NEXT:    jne .LBB86_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    movl $123, %edx
-; X86-NEXT:    btl %ecx, %eax
-; X86-NEXT:    jb .LBB86_4
-; X86-NEXT:  # %bb.3: # %if.then
-; X86-NEXT:    movl (%esi,%ecx,4), %edx
-; X86-NEXT:  .LBB86_4: # %return
-; X86-NEXT:    movl %edx, %eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    andl $31, %eax
+; X86-NEXT:    lock btsl %eax, (%edx)
+; X86-NEXT:    movl $123, %eax
+; X86-NEXT:    jae .LBB86_1
+; X86-NEXT:  # %bb.2: # %return
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB86_1: # %if.then
+; X86-NEXT:    movl (%edx,%ecx,4), %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask0_or_32_gpr_brz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB86_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB86_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $123, %edx
-; X64-NEXT:    btl %ecx, %eax
-; X64-NEXT:    jae .LBB86_3
-; X64-NEXT:  # %bb.4: # %return
-; X64-NEXT:    movl %edx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    movl $123, %eax
+; X64-NEXT:    jae .LBB86_1
+; X64-NEXT:  # %bb.2: # %return
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB86_3: # %if.then
-; X64-NEXT:    movl %ecx, %eax
-; X64-NEXT:    movl (%rdi,%rax,4), %edx
-; X64-NEXT:    movl %edx, %eax
+; X64-NEXT:  .LBB86_1: # %if.then
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
 entry:
   %rem = and i32 %c, 31
   %shl = shl nuw i32 1, %rem
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %shl1 = shl nuw i32 1, %c
   %and = and i32 %0, %shl1
   %tobool.not = icmp eq i32 %and, 0
   br i1 %tobool.not, label %if.then, label %return
 
 if.then:                                          ; preds = %entry
   %conv = zext i32 %c to i64
   %arrayidx = getelementptr inbounds i32, ptr %v, i64 %conv
   %1 = load i32, ptr %arrayidx, align 4
   br label %return
 
 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_mask1_or_32_gpr_brz(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask1_or_32_gpr_brz:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movl $1, %edx
-; X86-NEXT:    shll %cl, %edx
-; X86-NEXT:    movl (%esi), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB87_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %edx, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%esi)
-; X86-NEXT:    jne .LBB87_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    movl $123, %edx
-; X86-NEXT:    btl %ecx, %eax
-; X86-NEXT:    jb .LBB87_4
-; X86-NEXT:  # %bb.3: # %if.then
-; X86-NEXT:    movl (%esi,%ecx,4), %edx
-; X86-NEXT:  .LBB87_4: # %return
-; X86-NEXT:    movl %edx, %eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    andl $31, %eax
+; X86-NEXT:    lock btsl %eax, (%edx)
+; X86-NEXT:    movl $123, %eax
+; X86-NEXT:    jae .LBB87_1
+; X86-NEXT:  # %bb.2: # %return
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB87_1: # %if.then
+; X86-NEXT:    movl (%edx,%ecx,4), %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask1_or_32_gpr_brz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB87_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB87_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $123, %edx
-; X64-NEXT:    btl %ecx, %eax
-; X64-NEXT:    jae .LBB87_3
-; X64-NEXT:  # %bb.4: # %return
-; X64-NEXT:    movl %edx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    movl $123, %eax
+; X64-NEXT:    jae .LBB87_1
+; X64-NEXT:  # %bb.2: # %return
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB87_3: # %if.then
-; X64-NEXT:    movl %ecx, %eax
-; X64-NEXT:    movl (%rdi,%rax,4), %edx
-; X64-NEXT:    movl %edx, %eax
+; X64-NEXT:  .LBB87_1: # %if.then
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i32 1, %c
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %rem = and i32 %c, 31
   %shl1 = shl nuw i32 1, %rem
   %and = and i32 %0, %shl1
   %tobool.not = icmp eq i32 %and, 0
   br i1 %tobool.not, label %if.then, label %return
 
 if.then:                                          ; preds = %entry
   %conv = zext i32 %c to i64
   %arrayidx = getelementptr inbounds i32, ptr %v, i64 %conv
   %1 = load i32, ptr %arrayidx, align 4
   br label %return
 
 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_mask01_or_32_gpr_brz(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask01_or_32_gpr_brz:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movl $1, %edi
-; X86-NEXT:    shll %cl, %edi
-; X86-NEXT:    movl (%esi), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB88_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    orl %edi, %edx
-; X86-NEXT:    lock cmpxchgl %edx, (%esi)
-; X86-NEXT:    jne .LBB88_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    movl $123, %edx
-; X86-NEXT:    testl %edi, %eax
-; X86-NEXT:    jne .LBB88_4
-; X86-NEXT:  # %bb.3: # %if.then
-; X86-NEXT:    movl (%esi,%ecx,4), %edx
-; X86-NEXT:  .LBB88_4: # %return
-; X86-NEXT:    movl %edx, %eax
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    andl $31, %eax
+; X86-NEXT:    lock btsl %eax, (%edx)
+; X86-NEXT:    movl $123, %eax
+; X86-NEXT:    jae .LBB88_1
+; X86-NEXT:  # %bb.2: # %return
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB88_1: # %if.then
+; X86-NEXT:    movl (%edx,%ecx,4), %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask01_or_32_gpr_brz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %esi
-; X64-NEXT:    shll %cl, %esi
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB88_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %edx
-; X64-NEXT:    orl %esi, %edx
-; X64-NEXT:    lock cmpxchgl %edx, (%rdi)
-; X64-NEXT:    jne .LBB88_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $123, %edx
-; X64-NEXT:    testl %esi, %eax
-; X64-NEXT:    je .LBB88_3
-; X64-NEXT:  # %bb.4: # %return
-; X64-NEXT:    movl %edx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    movl $123, %eax
+; X64-NEXT:    jae .LBB88_1
+; X64-NEXT:  # %bb.2: # %return
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB88_3: # %if.then
-; X64-NEXT:    movl %ecx, %eax
-; X64-NEXT:    movl (%rdi,%rax,4), %edx
-; X64-NEXT:    movl %edx, %eax
+; X64-NEXT:  .LBB88_1: # %if.then
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
 entry:
   %rem = and i32 %c, 31
   %shl = shl nuw i32 1, %rem
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %0, %shl
   %tobool.not = icmp eq i32 %and, 0
   br i1 %tobool.not, label %if.then, label %return

 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_or_32_gpr_brnz(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_or_32_gpr_brnz:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB90_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB90_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    testl %esi, %eax
-; X86-NEXT:    je .LBB90_3
-; X86-NEXT:  # %bb.4: # %if.then
-; X86-NEXT:    movl (%edx,%ecx,4), %eax
-; X86-NEXT:    jmp .LBB90_5
-; X86-NEXT:  .LBB90_3:
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl $31, %edx
+; X86-NEXT:    lock btsl %edx, (%ecx)
+; X86-NEXT:    jae .LBB90_1
+; X86-NEXT:  # %bb.2: # %if.then
+; X86-NEXT:    movl (%ecx,%eax,4), %eax
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB90_1:
 ; X86-NEXT:    movl $123, %eax
-; X86-NEXT:  .LBB90_5: # %return
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_or_32_gpr_brnz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB90_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB90_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    testl %edx, %eax
-; X64-NEXT:    je .LBB90_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    jae .LBB90_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB90_3:
+; X64-NEXT:  .LBB90_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i32 1, %c
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %0, %shl
   %tobool.not = icmp eq i32 %and, 0
   br i1 %tobool.not, label %return, label %if.then
 
 if.then:                                          ; preds = %entry
   %idxprom = zext i32 %c to i64
   %arrayidx = getelementptr inbounds i32, ptr %v, i64 %idxprom
   %1 = load i32, ptr %arrayidx, align 4
   br label %return
 
 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_small_mask_or_32_gpr_brnz(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_small_mask_or_32_gpr_brnz:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    andl $15, %ecx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB91_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB91_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    testl %esi, %eax
-; X86-NEXT:    je .LBB91_3
-; X86-NEXT:  # %bb.4: # %if.then
-; X86-NEXT:    movl (%edx,%ecx,4), %eax
-; X86-NEXT:    jmp .LBB91_5
-; X86-NEXT:  .LBB91_3:
+; X86-NEXT:    lock btsl %ecx, (%eax)
+; X86-NEXT:    jae .LBB91_1
+; X86-NEXT:  # %bb.2: # %if.then
+; X86-NEXT:    movl (%eax,%ecx,4), %eax
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB91_1:
 ; X86-NEXT:    movl $123, %eax
-; X86-NEXT:  .LBB91_5: # %return
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_small_mask_or_32_gpr_brnz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    andl $15, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB91_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB91_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    testl %edx, %eax
-; X64-NEXT:    je .LBB91_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    andl $15, %esi
+; X64-NEXT:    lock btsl %esi, (%rdi)
+; X64-NEXT:    jae .LBB91_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB91_3:
+; X64-NEXT:  .LBB91_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %0 = and i32 %c, 15
   %shl = shl nuw nsw i32 1, %0
   %1 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %1, %shl
   %tobool.not = icmp eq i32 %and, 0
   br i1 %tobool.not, label %return, label %if.then
 
 if.then:                                          ; preds = %entry
   %conv2 = zext i32 %0 to i64
   %arrayidx = getelementptr inbounds i32, ptr %v, i64 %conv2
   %2 = load i32, ptr %arrayidx, align 4
   br label %return
 
 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %2, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_mask0_or_32_gpr_brnz(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask0_or_32_gpr_brnz:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB92_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB92_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    btl %ecx, %eax
-; X86-NEXT:    jae .LBB92_3
-; X86-NEXT:  # %bb.4: # %if.then
-; X86-NEXT:    movl (%edx,%ecx,4), %eax
-; X86-NEXT:    jmp .LBB92_5
-; X86-NEXT:  .LBB92_3:
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl $31, %edx
+; X86-NEXT:    lock btsl %edx, (%ecx)
+; X86-NEXT:    jae .LBB92_1
+; X86-NEXT:  # %bb.2: # %if.then
+; X86-NEXT:    movl (%ecx,%eax,4), %eax
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB92_1:
 ; X86-NEXT:    movl $123, %eax
-; X86-NEXT:  .LBB92_5: # %return
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask0_or_32_gpr_brnz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB92_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB92_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    btl %ecx, %eax
-; X64-NEXT:    jae .LBB92_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    jae .LBB92_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB92_3:
+; X64-NEXT:  .LBB92_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %rem = and i32 %c, 31
   %shl = shl nuw i32 1, %rem
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %shl1 = shl nuw i32 1, %c
   %and = and i32 %0, %shl1

 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_mask1_or_32_gpr_brnz(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask1_or_32_gpr_brnz:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB93_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB93_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    btl %ecx, %eax
-; X86-NEXT:    jae .LBB93_3
-; X86-NEXT:  # %bb.4: # %if.then
-; X86-NEXT:    movl (%edx,%ecx,4), %eax
-; X86-NEXT:    jmp .LBB93_5
-; X86-NEXT:  .LBB93_3:
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl $31, %edx
+; X86-NEXT:    lock btsl %edx, (%ecx)
+; X86-NEXT:    jae .LBB93_1
+; X86-NEXT:  # %bb.2: # %if.then
+; X86-NEXT:    movl (%ecx,%eax,4), %eax
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB93_1:
 ; X86-NEXT:    movl $123, %eax
-; X86-NEXT:  .LBB93_5: # %return
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask1_or_32_gpr_brnz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB93_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB93_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    btl %ecx, %eax
-; X64-NEXT:    jae .LBB93_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    jae .LBB93_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB93_3:
+; X64-NEXT:  .LBB93_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i32 1, %c
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %rem = and i32 %c, 31
   %shl1 = shl nuw i32 1, %rem
   %and = and i32 %0, %shl1

 return:                                           ; preds = %entry, %if.then
   %retval.0 = phi i32 [ %1, %if.then ], [ 123, %entry ]
   ret i32 %retval.0
 }
 
 define i32 @atomic_shl1_mask01_or_32_gpr_brnz(ptr %v, i32 %c) nounwind {
 ; X86-LABEL: atomic_shl1_mask01_or_32_gpr_brnz:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl $1, %esi
-; X86-NEXT:    shll %cl, %esi
-; X86-NEXT:    movl (%edx), %eax
-; X86-NEXT:    .p2align 4, 0x90
-; X86-NEXT:  .LBB94_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    lock cmpxchgl %edi, (%edx)
-; X86-NEXT:    jne .LBB94_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    testl %esi, %eax
-; X86-NEXT:    je .LBB94_3
-; X86-NEXT:  # %bb.4: # %if.then
-; X86-NEXT:    movl (%edx,%ecx,4), %eax
-; X86-NEXT:    jmp .LBB94_5
-; X86-NEXT:  .LBB94_3:
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    andl $31, %edx
+; X86-NEXT:    lock btsl %edx, (%ecx)
+; X86-NEXT:    jae .LBB94_1
+; X86-NEXT:  # %bb.2: # %if.then
+; X86-NEXT:    movl (%ecx,%eax,4), %eax
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB94_1:
 ; X86-NEXT:    movl $123, %eax
-; X86-NEXT:  .LBB94_5: # %return
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask01_or_32_gpr_brnz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movl %esi, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shll %cl, %edx
-; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB94_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movl %eax, %esi
-; X64-NEXT:    orl %edx, %esi
-; X64-NEXT:    lock cmpxchgl %esi, (%rdi)
-; X64-NEXT:    jne .LBB94_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    testl %edx, %eax
-; X64-NEXT:    je .LBB94_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $31, %eax
+; X64-NEXT:    lock btsl %eax, (%rdi)
+; X64-NEXT:    jae .LBB94_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl (%rdi,%rax,4), %eax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB94_3:
+; X64-NEXT:  .LBB94_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %rem = and i32 %c, 31
   %shl = shl nuw i32 1, %rem
   %0 = atomicrmw or ptr %v, i32 %shl monotonic, align 4
   %and = and i32 %0, %shl
   %tobool.not = icmp eq i32 %and, 0

 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_xor_64_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $63, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btcq %rcx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB122_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    xorq %rdx, %rcx
-; X64-NEXT:    lock cmpxchgq %rcx, (%rdi)
-; X64-NEXT:    jne .LBB122_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andq %rdx, %rax
+; X64-NEXT:    shlq %cl, %rax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i64 1, %c
   %0 = atomicrmw xor ptr %v, i64 %shl monotonic, align 8
   %and = and i64 %0, %shl
   ret i64 %and
 }
 

 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_small_mask_xor_64_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    andb $31, %cl
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $31, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btcq %rcx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB125_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    xorq %rdx, %rcx
-; X64-NEXT:    lock cmpxchgq %rcx, (%rdi)
-; X64-NEXT:    jne .LBB125_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andl %edx, %eax
+; X64-NEXT:    shlq %cl, %rax
 ; X64-NEXT:    retq
 entry:
   %rem = and i64 %c, 31
   %shl = shl nuw nsw i64 1, %rem
   %0 = atomicrmw xor ptr %v, i64 %shl monotonic, align 8
   %and = and i64 %0, %shl
   ret i64 %and
 }

 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask0_xor_64_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB126_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    xorq %rdx, %rsi
-; X64-NEXT:    lock cmpxchgq %rsi, (%rdi)
-; X64-NEXT:    jne .LBB126_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $63, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btcq %rcx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    andq %rdx, %rax
+; X64-NEXT:    shlq %cl, %rax
 ; X64-NEXT:    retq
 entry:
   %rem = and i64 %c, 63
   %shl = shl nuw i64 1, %rem
   %0 = atomicrmw xor ptr %v, i64 %shl monotonic, align 8
   %shl1 = shl nuw i64 1, %c
   %and = and i64 %0, %shl1
   ret i64 %and

 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask1_xor_64_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB127_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    xorq %rdx, %rsi
-; X64-NEXT:    lock cmpxchgq %rsi, (%rdi)
-; X64-NEXT:    jne .LBB127_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $63, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btcq %rcx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    andq %rdx, %rax
+; X64-NEXT:    shlq %cl, %rax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i64 1, %c
   %0 = atomicrmw xor ptr %v, i64 %shl monotonic, align 8
   %rem = and i64 %c, 63
   %shl1 = shl nuw i64 1, %rem
   %and = and i64 %0, %shl1
   ret i64 %and

 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask01_xor_64_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $63, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btcq %rcx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB128_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    xorq %rdx, %rcx
-; X64-NEXT:    lock cmpxchgq %rcx, (%rdi)
-; X64-NEXT:    jne .LBB128_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andq %rdx, %rax
+; X64-NEXT:    shlq %cl, %rax
 ; X64-NEXT:    retq
 entry:
   %rem = and i64 %c, 63
   %shl = shl nuw i64 1, %rem
   %0 = atomicrmw xor ptr %v, i64 %shl monotonic, align 8
   %and = and i64 %0, %shl
   ret i64 %and
 }

 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_and_64_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq $-2, %rsi
+; X64-NEXT:    andl $63, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btrq %rcx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-NEXT:    rolq %cl, %rsi
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB146_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    andq %rsi, %rcx
-; X64-NEXT:    lock cmpxchgq %rcx, (%rdi)
-; X64-NEXT:    jne .LBB146_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andq %rdx, %rax
+; X64-NEXT:    shlq %cl, %rax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i64 1, %c
   %not = xor i64 %shl, -1
   %0 = atomicrmw and ptr %v, i64 %not monotonic, align 8
   %and = and i64 %0, %shl
   ret i64 %and
 }

 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_small_mask_and_64_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    andb $31, %cl
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq $-2, %rsi
+; X64-NEXT:    andl $31, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btrq %rcx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-NEXT:    rolq %cl, %rsi
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB149_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    andq %rsi, %rcx
-; X64-NEXT:    lock cmpxchgq %rcx, (%rdi)
-; X64-NEXT:    jne .LBB149_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andl %edx, %eax
+; X64-NEXT:    shlq %cl, %rax
 ; X64-NEXT:    retq
 entry:
   %rem = and i64 %c, 31
   %shl = shl nuw nsw i64 1, %rem
   %not = xor i64 %shl, -1
   %0 = atomicrmw and ptr %v, i64 %not monotonic, align 8
   %and = and i64 %0, %shl
   ret i64 %and

 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask0_and_64_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movq $-2, %rdx
-; X64-NEXT:    rolq %cl, %rdx
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB150_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    andq %rdx, %rsi
-; X64-NEXT:    lock cmpxchgq %rsi, (%rdi)
-; X64-NEXT:    jne .LBB150_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $63, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btrq %rcx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    andq %rdx, %rax
+; X64-NEXT:    shlq %cl, %rax
 ; X64-NEXT:    retq
 entry:
   %rem = and i64 %c, 63
   %shl = shl nuw i64 1, %rem
   %not = xor i64 %shl, -1
   %0 = atomicrmw and ptr %v, i64 %not monotonic, align 8
   %shl1 = shl nuw i64 1, %c
   %and = and i64 %0, %shl1

 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask1_and_64_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movq $-2, %rdx
-; X64-NEXT:    rolq %cl, %rdx
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB151_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    andq %rdx, %rsi
-; X64-NEXT:    lock cmpxchgq %rsi, (%rdi)
-; X64-NEXT:    jne .LBB151_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    movl $1, %edx
+; X64-NEXT:    andl $63, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btrq %rcx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    andq %rdx, %rax
+; X64-NEXT:    shlq %cl, %rax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i64 1, %c
   %not = xor i64 %shl, -1
   %0 = atomicrmw and ptr %v, i64 %not monotonic, align 8
   %rem = and i64 %c, 63
   %shl1 = shl nuw i64 1, %rem
   %and = and i64 %0, %shl1

 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask01_and_64_gpr_val:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq $-2, %rsi
+; X64-NEXT:    andl $63, %ecx
+; X64-NEXT:    xorl %eax, %eax
+; X64-NEXT:    lock btrq %rcx, (%rdi)
+; X64-NEXT:    setb %al
 ; X64-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-NEXT:    rolq %cl, %rsi
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB152_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    andq %rsi, %rcx
-; X64-NEXT:    lock cmpxchgq %rcx, (%rdi)
-; X64-NEXT:    jne .LBB152_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    andq %rdx, %rax
+; X64-NEXT:    shlq %cl, %rax
 ; X64-NEXT:    retq
 entry:
   %rem = and i64 %c, 63
   %shl = shl nuw i64 1, %rem
   %not = xor i64 %shl, -1
   %0 = atomicrmw and ptr %v, i64 %not monotonic, align 8
   %and = and i64 %0, %shl
   ret i64 %and

 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_and_64_gpr_brnz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq $-2, %rsi
-; X64-NEXT:    rolq %cl, %rsi
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB162_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    andq %rsi, %r8
-; X64-NEXT:    lock cmpxchgq %r8, (%rdi)
-; X64-NEXT:    jne .LBB162_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    testq %rdx, %rax
-; X64-NEXT:    je .LBB162_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movq (%rdi,%rcx,8), %rax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $63, %eax
+; X64-NEXT:    lock btrq %rax, (%rdi)
+; X64-NEXT:    jae .LBB162_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movq (%rdi,%rsi,8), %rax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB162_3:
+; X64-NEXT:  .LBB162_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i64 1, %c
   %not = xor i64 %shl, -1
   %0 = atomicrmw and ptr %v, i64 %not monotonic, align 8
   %and = and i64 %0, %shl
   %tobool.not = icmp eq i64 %and, 0

 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_small_mask_and_64_gpr_brnz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    andl $31, %ecx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq $-2, %rsi
-; X64-NEXT:    rolq %cl, %rsi
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB165_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    andq %rsi, %r8
-; X64-NEXT:    lock cmpxchgq %r8, (%rdi)
-; X64-NEXT:    jne .LBB165_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    testl %edx, %eax
-; X64-NEXT:    je .LBB165_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movq (%rdi,%rcx,8), %rax
+; X64-NEXT:    andl $31, %esi
+; X64-NEXT:    lock btrq %rsi, (%rdi)
+; X64-NEXT:    jae .LBB165_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movq (%rdi,%rsi,8), %rax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB165_3:
+; X64-NEXT:  .LBB165_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %rem = and i64 %c, 31
   %shl = shl nuw nsw i64 1, %rem
   %not = xor i64 %shl, -1
   %0 = atomicrmw and ptr %v, i64 %not monotonic, align 8
   %and = and i64 %0, %shl

 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask0_and_64_gpr_brnz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movq $-2, %rdx
-; X64-NEXT:    rolq %cl, %rdx
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB166_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    andq %rdx, %rsi
-; X64-NEXT:    lock cmpxchgq %rsi, (%rdi)
-; X64-NEXT:    jne .LBB166_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    btq %rcx, %rax
-; X64-NEXT:    jae .LBB166_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movq (%rdi,%rcx,8), %rax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $63, %eax
+; X64-NEXT:    lock btrq %rax, (%rdi)
+; X64-NEXT:    jae .LBB166_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movq (%rdi,%rsi,8), %rax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB166_3:
+; X64-NEXT:  .LBB166_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %rem = and i64 %c, 63
   %shl = shl nuw i64 1, %rem
   %not = xor i64 %shl, -1
   %0 = atomicrmw and ptr %v, i64 %not monotonic, align 8
   %shl1 = shl nuw i64 1, %c

 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask1_and_64_gpr_brnz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movq $-2, %rdx
-; X64-NEXT:    rolq %cl, %rdx
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB167_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    andq %rdx, %rsi
-; X64-NEXT:    lock cmpxchgq %rsi, (%rdi)
-; X64-NEXT:    jne .LBB167_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    btq %rcx, %rax
-; X64-NEXT:    jae .LBB167_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movq (%rdi,%rcx,8), %rax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $63, %eax
+; X64-NEXT:    lock btrq %rax, (%rdi)
+; X64-NEXT:    jae .LBB167_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movq (%rdi,%rsi,8), %rax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB167_3:
+; X64-NEXT:  .LBB167_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %shl = shl nuw i64 1, %c
   %not = xor i64 %shl, -1
   %0 = atomicrmw and ptr %v, i64 %not monotonic, align 8
   %rem = and i64 %c, 63
   %shl1 = shl nuw i64 1, %rem

 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: atomic_shl1_mask01_and_64_gpr_brnz:
 ; X64:       # %bb.0: # %entry
-; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movl $1, %edx
-; X64-NEXT:    shlq %cl, %rdx
-; X64-NEXT:    movq $-2, %rsi
-; X64-NEXT:    rolq %cl, %rsi
-; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    .p2align 4, 0x90
-; X64-NEXT:  .LBB168_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    andq %rsi, %r8
-; X64-NEXT:    lock cmpxchgq %r8, (%rdi)
-; X64-NEXT:    jne .LBB168_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
-; X64-NEXT:    testq %rdx, %rax
-; X64-NEXT:    je .LBB168_3
-; X64-NEXT:  # %bb.4: # %if.then
-; X64-NEXT:    movq (%rdi,%rcx,8), %rax
+; X64-NEXT:    movl %esi, %eax
+; X64-NEXT:    andl $63, %eax
+; X64-NEXT:    lock btrq %rax, (%rdi)
+; X64-NEXT:    jae .LBB168_1
+; X64-NEXT:  # %bb.2: # %if.then
+; X64-NEXT:    movq (%rdi,%rsi,8), %rax
 ; X64-NEXT:    retq
-; X64-NEXT:  .LBB168_3:
+; X64-NEXT:  .LBB168_1:
 ; X64-NEXT:    movl $123, %eax
 ; X64-NEXT:    retq
 entry:
   %rem = and i64 %c, 63
   %shl = shl nuw i64 1, %rem
   %not = xor i64 %shl, -1
   %0 = atomicrmw and ptr %v, i64 %not monotonic, align 8
   %and = and i64 %0, %shl