maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[OPENMP] Fixed bug in atomic update/capture/write constructs. 

Fixed a bug with codegen for destination atomic l-value with padding and junk in this padding bytes.

llvm-svn: 237422
This commit is contained in:
Alexey Bataev 2015-05-15 08:36:34 +00:00
parent cdad63b33d
commit f0ab553fea
8 changed files with 662 additions and 799 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

409
clang/lib/CodeGen/CGAtomic.cpp
View File

@ -215,6 +215,17 @@ namespace {
llvm::AtomicOrdering Failure = llvm::SequentiallyConsistent,
bool IsWeak = false);
/// \brief Emits atomic update.
/// \brief AO Atomic ordering.
/// \brief UpdateOp Update operation for the current lvalue.
void EmitAtomicUpdate(llvm::AtomicOrdering AO,
const llvm::function_ref<RValue(RValue)> &UpdateOp,
bool IsVolatile);
/// \brief Emits atomic update.
/// \brief AO Atomic ordering.
void EmitAtomicUpdate(llvm::AtomicOrdering AO, RValue UpdateRVal,
bool IsVolatile);
/// Materialize an atomic r-value in atomic-layout memory.
llvm::Value *materializeRValue(RValue rvalue) const;
@ -235,16 +246,31 @@ namespace {
/// \brief Emits atomic load as LLVM instruction.
llvm::Value *EmitAtomicLoadOp(llvm::AtomicOrdering AO, bool IsVolatile);
/// \brief Emits atomic compare-and-exchange op as a libcall.
std::pair<RValue, llvm::Value *> EmitAtomicCompareExchangeLibcall(
RValue Expected, RValue DesiredAddr,
llvm::Value *EmitAtomicCompareExchangeLibcall(
llvm::Value *ExpectedAddr, llvm::Value *DesiredAddr,
llvm::AtomicOrdering Success = llvm::SequentiallyConsistent,
llvm::AtomicOrdering Failure = llvm::SequentiallyConsistent);
/// \brief Emits atomic compare-and-exchange op as LLVM instruction.
std::pair<RValue, llvm::Value *> EmitAtomicCompareExchangeOp(
RValue Expected, RValue Desired,
std::pair<llvm::Value *, llvm::Value *> EmitAtomicCompareExchangeOp(
llvm::Value *ExpectedVal, llvm::Value *DesiredVal,
llvm::AtomicOrdering Success = llvm::SequentiallyConsistent,
llvm::AtomicOrdering Failure = llvm::SequentiallyConsistent,
bool IsWeak = false);
/// \brief Emit atomic update as libcalls.
void
EmitAtomicUpdateLibcall(llvm::AtomicOrdering AO,
const llvm::function_ref<RValue(RValue)> &UpdateOp,
bool IsVolatile);
/// \brief Emit atomic update as LLVM instructions.
void EmitAtomicUpdateOp(llvm::AtomicOrdering AO,
const llvm::function_ref<RValue(RValue)> &UpdateOp,
bool IsVolatile);
/// \brief Emit atomic update as libcalls.
void EmitAtomicUpdateLibcall(llvm::AtomicOrdering AO, RValue UpdateRVal,
bool IsVolatile);
/// \brief Emit atomic update as LLVM instructions.
void EmitAtomicUpdateOp(llvm::AtomicOrdering AO, RValue UpdateRal,
bool IsVolatile);
};
}
@ -1313,12 +1339,10 @@ llvm::Value *AtomicInfo::convertRValueToInt(RValue RVal) const {
getAtomicAlignment().getQuantity());
}
std::pair<RValue, llvm::Value *> AtomicInfo::EmitAtomicCompareExchangeOp(
RValue Expected, RValue Desired, llvm::AtomicOrdering Success,
llvm::AtomicOrdering Failure, bool IsWeak) {
std::pair<llvm::Value *, llvm::Value *> AtomicInfo::EmitAtomicCompareExchangeOp(
llvm::Value *ExpectedVal, llvm::Value *DesiredVal,
llvm::AtomicOrdering Success, llvm::AtomicOrdering Failure, bool IsWeak) {
// Do the atomic store.
auto *ExpectedVal = convertRValueToInt(Expected);
auto *DesiredVal = convertRValueToInt(Desired);
auto *Addr = emitCastToAtomicIntPointer(getAtomicAddress());
auto *Inst = CGF.Builder.CreateAtomicCmpXchg(Addr, ExpectedVal, DesiredVal,
Success, Failure);
@ -1329,20 +1353,16 @@ std::pair<RValue, llvm::Value *> AtomicInfo::EmitAtomicCompareExchangeOp(
// Okay, turn that back into the original value type.
auto *PreviousVal = CGF.Builder.CreateExtractValue(Inst, /*Idxs=*/0);
auto *SuccessFailureVal = CGF.Builder.CreateExtractValue(Inst, /*Idxs=*/1);
return std::make_pair(
ConvertIntToValueOrAtomic(PreviousVal, AggValueSlot::ignored(),
SourceLocation(), /*AsValue=*/false),
SuccessFailureVal);
return std::make_pair(PreviousVal, SuccessFailureVal);
}
std::pair<RValue, llvm::Value *>
AtomicInfo::EmitAtomicCompareExchangeLibcall(RValue Expected, RValue Desired,
llvm::Value *
AtomicInfo::EmitAtomicCompareExchangeLibcall(llvm::Value *ExpectedAddr,
llvm::Value *DesiredAddr,
llvm::AtomicOrdering Success,
llvm::AtomicOrdering Failure) {
// bool __atomic_compare_exchange(size_t size, void *obj, void *expected,
// void *desired, int success, int failure);
auto *ExpectedAddr = materializeRValue(Expected);
auto *DesiredAddr = materializeRValue(Desired);
CallArgList Args;
Args.add(RValue::get(getAtomicSizeValue()), CGF.getContext().getSizeType());
Args.add(RValue::get(CGF.EmitCastToVoidPtr(getAtomicAddress())),
@ -1360,10 +1380,7 @@ AtomicInfo::EmitAtomicCompareExchangeLibcall(RValue Expected, RValue Desired,
auto SuccessFailureRVal = emitAtomicLibcall(CGF, "__atomic_compare_exchange",
CGF.getContext().BoolTy, Args);
return std::make_pair(
convertTempToRValue(ExpectedAddr, AggValueSlot::ignored(),
SourceLocation(), /*AsValue=*/false),
SuccessFailureRVal.getScalarVal());
return SuccessFailureRVal.getScalarVal();
}
std::pair<RValue, llvm::Value *> AtomicInfo::EmitAtomicCompareExchange(
@ -1376,14 +1393,247 @@ std::pair<RValue, llvm::Value *> AtomicInfo::EmitAtomicCompareExchange(
// Check whether we should use a library call.
if (shouldUseLibcall()) {
// Produce a source address.
return EmitAtomicCompareExchangeLibcall(Expected, Desired, Success,
Failure);
auto *ExpectedAddr = materializeRValue(Expected);
auto *DesiredAddr = materializeRValue(Desired);
auto *Res = EmitAtomicCompareExchangeLibcall(ExpectedAddr, DesiredAddr,
Success, Failure);
return std::make_pair(
convertTempToRValue(ExpectedAddr, AggValueSlot::ignored(),
SourceLocation(), /*AsValue=*/false),
Res);
}
// If we've got a scalar value of the right size, try to avoid going
// through memory.
return EmitAtomicCompareExchangeOp(Expected, Desired, Success, Failure,
IsWeak);
auto *ExpectedVal = convertRValueToInt(Expected);
auto *DesiredVal = convertRValueToInt(Desired);
auto Res = EmitAtomicCompareExchangeOp(ExpectedVal, DesiredVal, Success,
Failure, IsWeak);
return std::make_pair(
ConvertIntToValueOrAtomic(Res.first, AggValueSlot::ignored(),
SourceLocation(), /*AsValue=*/false),
Res.second);
}
static void
EmitAtomicUpdateValue(CodeGenFunction &CGF, AtomicInfo &Atomics, RValue OldRVal,
const llvm::function_ref<RValue(RValue)> &UpdateOp,
llvm::Value *DesiredAddr) {
llvm::Value *Ptr = nullptr;
LValue UpdateLVal;
RValue UpRVal;
LValue AtomicLVal = Atomics.getAtomicLValue();
LValue DesiredLVal;
if (AtomicLVal.isSimple()) {
UpRVal = OldRVal;
DesiredLVal =
LValue::MakeAddr(DesiredAddr, AtomicLVal.getType(),
AtomicLVal.getAlignment(), CGF.CGM.getContext());
} else {
// Build new lvalue for temp address
Ptr = Atomics.materializeRValue(OldRVal);
if (AtomicLVal.isBitField()) {
UpdateLVal =
LValue::MakeBitfield(Ptr, AtomicLVal.getBitFieldInfo(),
AtomicLVal.getType(), AtomicLVal.getAlignment());
DesiredLVal =
LValue::MakeBitfield(DesiredAddr, AtomicLVal.getBitFieldInfo(),
AtomicLVal.getType(), AtomicLVal.getAlignment());
} else if (AtomicLVal.isVectorElt()) {
UpdateLVal = LValue::MakeVectorElt(Ptr, AtomicLVal.getVectorIdx(),
AtomicLVal.getType(),
AtomicLVal.getAlignment());
DesiredLVal = LValue::MakeVectorElt(
DesiredAddr, AtomicLVal.getVectorIdx(), AtomicLVal.getType(),
AtomicLVal.getAlignment());
} else {
assert(AtomicLVal.isExtVectorElt());
UpdateLVal = LValue::MakeExtVectorElt(Ptr, AtomicLVal.getExtVectorElts(),
AtomicLVal.getType(),
AtomicLVal.getAlignment());
DesiredLVal = LValue::MakeExtVectorElt(
DesiredAddr, AtomicLVal.getExtVectorElts(), AtomicLVal.getType(),
AtomicLVal.getAlignment());
}
UpdateLVal.setTBAAInfo(AtomicLVal.getTBAAInfo());
DesiredLVal.setTBAAInfo(AtomicLVal.getTBAAInfo());
UpRVal = CGF.EmitLoadOfLValue(UpdateLVal, SourceLocation());
}
// Store new value in the corresponding memory area
RValue NewRVal = UpdateOp(UpRVal);
if (NewRVal.isScalar()) {
CGF.EmitStoreThroughLValue(NewRVal, DesiredLVal);
} else {
assert(NewRVal.isComplex());
CGF.EmitStoreOfComplex(NewRVal.getComplexVal(), DesiredLVal,
/*isInit=*/false);
}
}
void AtomicInfo::EmitAtomicUpdateLibcall(
llvm::AtomicOrdering AO, const llvm::function_ref<RValue(RValue)> &UpdateOp,
bool IsVolatile) {
auto Failure = llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(AO);
llvm::Value *ExpectedAddr = CreateTempAlloca();
EmitAtomicLoadLibcall(ExpectedAddr, AO, IsVolatile);
auto *ContBB = CGF.createBasicBlock("atomic_cont");
auto *ExitBB = CGF.createBasicBlock("atomic_exit");
CGF.EmitBlock(ContBB);
auto *DesiredAddr = CreateTempAlloca();
if ((LVal.isBitField() && BFI.Size != ValueSizeInBits) ||
requiresMemSetZero(
getAtomicAddress()->getType()->getPointerElementType())) {
auto *OldVal = CGF.Builder.CreateAlignedLoad(
ExpectedAddr, getAtomicAlignment().getQuantity());
CGF.Builder.CreateAlignedStore(OldVal, DesiredAddr,
getAtomicAlignment().getQuantity());
}
auto OldRVal = convertTempToRValue(ExpectedAddr, AggValueSlot::ignored(),
SourceLocation(), /*AsValue=*/false);
EmitAtomicUpdateValue(CGF, *this, OldRVal, UpdateOp, DesiredAddr);
auto *Res =
EmitAtomicCompareExchangeLibcall(ExpectedAddr, DesiredAddr, AO, Failure);
CGF.Builder.CreateCondBr(Res, ExitBB, ContBB);
CGF.EmitBlock(ExitBB, /*IsFinished=*/true);
}
void AtomicInfo::EmitAtomicUpdateOp(
llvm::AtomicOrdering AO, const llvm::function_ref<RValue(RValue)> &UpdateOp,
bool IsVolatile) {
auto Failure = llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(AO);
// Do the atomic load.
auto *OldVal = EmitAtomicLoadOp(AO, IsVolatile);
// For non-simple lvalues perform compare-and-swap procedure.
auto *ContBB = CGF.createBasicBlock("atomic_cont");
auto *ExitBB = CGF.createBasicBlock("atomic_exit");
auto *CurBB = CGF.Builder.GetInsertBlock();
CGF.EmitBlock(ContBB);
llvm::PHINode *PHI = CGF.Builder.CreatePHI(OldVal->getType(),
/*NumReservedValues=*/2);
PHI->addIncoming(OldVal, CurBB);
auto *NewAtomicAddr = CreateTempAlloca();
auto *NewAtomicIntAddr = emitCastToAtomicIntPointer(NewAtomicAddr);
if ((LVal.isBitField() && BFI.Size != ValueSizeInBits) ||
requiresMemSetZero(
getAtomicAddress()->getType()->getPointerElementType())) {
CGF.Builder.CreateAlignedStore(PHI, NewAtomicIntAddr,
getAtomicAlignment().getQuantity());
}
auto OldRVal = ConvertIntToValueOrAtomic(PHI, AggValueSlot::ignored(),
SourceLocation(), /*AsValue=*/false);
EmitAtomicUpdateValue(CGF, *this, OldRVal, UpdateOp, NewAtomicAddr);
auto *DesiredVal = CGF.Builder.CreateAlignedLoad(
NewAtomicIntAddr, getAtomicAlignment().getQuantity());
// Try to write new value using cmpxchg operation
auto Res = EmitAtomicCompareExchangeOp(PHI, DesiredVal, AO, Failure);
PHI->addIncoming(Res.first, CGF.Builder.GetInsertBlock());
CGF.Builder.CreateCondBr(Res.second, ExitBB, ContBB);
CGF.EmitBlock(ExitBB, /*IsFinished=*/true);
}
static void EmitAtomicUpdateValue(CodeGenFunction &CGF, AtomicInfo &Atomics,
RValue UpdateRVal, llvm::Value *DesiredAddr) {
LValue AtomicLVal = Atomics.getAtomicLValue();
LValue DesiredLVal;
// Build new lvalue for temp address
if (AtomicLVal.isBitField()) {
DesiredLVal =
LValue::MakeBitfield(DesiredAddr, AtomicLVal.getBitFieldInfo(),
AtomicLVal.getType(), AtomicLVal.getAlignment());
} else if (AtomicLVal.isVectorElt()) {
DesiredLVal =
LValue::MakeVectorElt(DesiredAddr, AtomicLVal.getVectorIdx(),
AtomicLVal.getType(), AtomicLVal.getAlignment());
} else {
assert(AtomicLVal.isExtVectorElt());
DesiredLVal = LValue::MakeExtVectorElt(
DesiredAddr, AtomicLVal.getExtVectorElts(), AtomicLVal.getType(),
AtomicLVal.getAlignment());
}
DesiredLVal.setTBAAInfo(AtomicLVal.getTBAAInfo());
// Store new value in the corresponding memory area
assert(UpdateRVal.isScalar());
CGF.EmitStoreThroughLValue(UpdateRVal, DesiredLVal);
}
void AtomicInfo::EmitAtomicUpdateLibcall(llvm::AtomicOrdering AO,
RValue UpdateRVal, bool IsVolatile) {
auto Failure = llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(AO);
llvm::Value *ExpectedAddr = CreateTempAlloca();
EmitAtomicLoadLibcall(ExpectedAddr, AO, IsVolatile);
auto *ContBB = CGF.createBasicBlock("atomic_cont");
auto *ExitBB = CGF.createBasicBlock("atomic_exit");
CGF.EmitBlock(ContBB);
auto *DesiredAddr = CreateTempAlloca();
if ((LVal.isBitField() && BFI.Size != ValueSizeInBits) ||
requiresMemSetZero(
getAtomicAddress()->getType()->getPointerElementType())) {
auto *OldVal = CGF.Builder.CreateAlignedLoad(
ExpectedAddr, getAtomicAlignment().getQuantity());
CGF.Builder.CreateAlignedStore(OldVal, DesiredAddr,
getAtomicAlignment().getQuantity());
}
EmitAtomicUpdateValue(CGF, *this, UpdateRVal, DesiredAddr);
auto *Res =
EmitAtomicCompareExchangeLibcall(ExpectedAddr, DesiredAddr, AO, Failure);
CGF.Builder.CreateCondBr(Res, ExitBB, ContBB);
CGF.EmitBlock(ExitBB, /*IsFinished=*/true);
}
void AtomicInfo::EmitAtomicUpdateOp(llvm::AtomicOrdering AO, RValue UpdateRVal,
bool IsVolatile) {
auto Failure = llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(AO);
// Do the atomic load.
auto *OldVal = EmitAtomicLoadOp(AO, IsVolatile);
// For non-simple lvalues perform compare-and-swap procedure.
auto *ContBB = CGF.createBasicBlock("atomic_cont");
auto *ExitBB = CGF.createBasicBlock("atomic_exit");
auto *CurBB = CGF.Builder.GetInsertBlock();
CGF.EmitBlock(ContBB);
llvm::PHINode *PHI = CGF.Builder.CreatePHI(OldVal->getType(),
/*NumReservedValues=*/2);
PHI->addIncoming(OldVal, CurBB);
auto *NewAtomicAddr = CreateTempAlloca();
auto *NewAtomicIntAddr = emitCastToAtomicIntPointer(NewAtomicAddr);
if ((LVal.isBitField() && BFI.Size != ValueSizeInBits) ||
requiresMemSetZero(
getAtomicAddress()->getType()->getPointerElementType())) {
CGF.Builder.CreateAlignedStore(PHI, NewAtomicIntAddr,
getAtomicAlignment().getQuantity());
}
EmitAtomicUpdateValue(CGF, *this, UpdateRVal, NewAtomicAddr);
auto *DesiredVal = CGF.Builder.CreateAlignedLoad(
NewAtomicIntAddr, getAtomicAlignment().getQuantity());
// Try to write new value using cmpxchg operation
auto Res = EmitAtomicCompareExchangeOp(PHI, DesiredVal, AO, Failure);
PHI->addIncoming(Res.first, CGF.Builder.GetInsertBlock());
CGF.Builder.CreateCondBr(Res.second, ExitBB, ContBB);
CGF.EmitBlock(ExitBB, /*IsFinished=*/true);
}
void AtomicInfo::EmitAtomicUpdate(
llvm::AtomicOrdering AO, const llvm::function_ref<RValue(RValue)> &UpdateOp,
bool IsVolatile) {
if (shouldUseLibcall()) {
EmitAtomicUpdateLibcall(AO, UpdateOp, IsVolatile);
} else {
EmitAtomicUpdateOp(AO, UpdateOp, IsVolatile);
}
}
void AtomicInfo::EmitAtomicUpdate(llvm::AtomicOrdering AO, RValue UpdateRVal,
bool IsVolatile) {
if (shouldUseLibcall()) {
EmitAtomicUpdateLibcall(AO, UpdateRVal, IsVolatile);
} else {
EmitAtomicUpdateOp(AO, UpdateRVal, IsVolatile);
}
}
void CodeGenFunction::EmitAtomicStore(RValue rvalue, LValue lvalue,
@ -1465,46 +1715,8 @@ void CodeGenFunction::EmitAtomicStore(RValue rvalue, LValue dest,
return;
}
// Atomic load of prev value.
RValue OldRVal =
atomics.EmitAtomicLoad(AggValueSlot::ignored(), SourceLocation(),
/*AsValue=*/false, AO, IsVolatile);
// For non-simple lvalues perform compare-and-swap procedure.
auto *ContBB = createBasicBlock("atomic_cont");
auto *ExitBB = createBasicBlock("atomic_exit");
auto *CurBB = Builder.GetInsertBlock();
EmitBlock(ContBB);
llvm::PHINode *PHI = Builder.CreatePHI(OldRVal.getScalarVal()->getType(),
/*NumReservedValues=*/2);
PHI->addIncoming(OldRVal.getScalarVal(), CurBB);
RValue OriginalRValue = RValue::get(PHI);
// Build new lvalue for temp address
auto *Ptr = atomics.materializeRValue(OriginalRValue);
// Build new lvalue for temp address
LValue UpdateLVal;
if (LVal.isBitField())
UpdateLVal = LValue::MakeBitfield(Ptr, LVal.getBitFieldInfo(),
LVal.getType(), LVal.getAlignment());
else if (LVal.isVectorElt())
UpdateLVal = LValue::MakeVectorElt(Ptr, LVal.getVectorIdx(), LVal.getType(),
LVal.getAlignment());
else {
assert(LVal.isExtVectorElt());
UpdateLVal = LValue::MakeExtVectorElt(Ptr, LVal.getExtVectorElts(),
LVal.getType(), LVal.getAlignment());
}
UpdateLVal.setTBAAInfo(LVal.getTBAAInfo());
// Store new value in the corresponding memory area
EmitStoreThroughLValue(rvalue, UpdateLVal);
// Load new value
RValue NewRValue = RValue::get(EmitLoadOfScalar(
Ptr, LVal.isVolatile(), atomics.getAtomicAlignment().getQuantity(),
atomics.getAtomicType(), SourceLocation()));
// Try to write new value using cmpxchg operation
auto Pair = atomics.EmitAtomicCompareExchange(OriginalRValue, NewRValue, AO);
PHI->addIncoming(Pair.first.getScalarVal(), Builder.GetInsertBlock());
Builder.CreateCondBr(Pair.second, ExitBB, ContBB);
EmitBlock(ExitBB, /*IsFinished=*/true);
// Emit simple atomic update operation.
atomics.EmitAtomicUpdate(AO, rvalue, IsVolatile);
}
/// Emit a compare-and-exchange op for atomic type.
@ -1529,72 +1741,9 @@ std::pair<RValue, llvm::Value *> CodeGenFunction::EmitAtomicCompareExchange(
void CodeGenFunction::EmitAtomicUpdate(
LValue LVal, llvm::AtomicOrdering AO,
const std::function<RValue(RValue)> &UpdateOp, bool IsVolatile) {
const llvm::function_ref<RValue(RValue)> &UpdateOp, bool IsVolatile) {
AtomicInfo Atomics(*this, LVal);
LValue AtomicLVal = Atomics.getAtomicLValue();
// Atomic load of prev value.
RValue OldRVal =
Atomics.EmitAtomicLoad(AggValueSlot::ignored(), SourceLocation(),
/*AsValue=*/false, AO, IsVolatile);
bool IsScalar = OldRVal.isScalar();
auto *OldVal =
IsScalar ? OldRVal.getScalarVal() : Atomics.convertRValueToInt(OldRVal);
// For non-simple lvalues perform compare-and-swap procedure.
auto *ContBB = createBasicBlock("atomic_cont");
auto *ExitBB = createBasicBlock("atomic_exit");
auto *CurBB = Builder.GetInsertBlock();
EmitBlock(ContBB);
llvm::PHINode *PHI = Builder.CreatePHI(OldVal->getType(),
/*NumReservedValues=*/2);
PHI->addIncoming(OldVal, CurBB);
RValue OriginalRValue =
IsScalar ? RValue::get(PHI) : Atomics.ConvertIntToValueOrAtomic(
PHI, AggValueSlot::ignored(),
SourceLocation(), /*AsValue=*/false);
// Build new lvalue for temp address
LValue UpdateLVal;
llvm::Value *Ptr = nullptr;
RValue UpRVal;
if (AtomicLVal.isSimple()) {
UpRVal = OriginalRValue;
} else {
// Build new lvalue for temp address
Ptr = Atomics.materializeRValue(OriginalRValue);
if (AtomicLVal.isBitField())
UpdateLVal =
LValue::MakeBitfield(Ptr, AtomicLVal.getBitFieldInfo(),
AtomicLVal.getType(), AtomicLVal.getAlignment());
else if (AtomicLVal.isVectorElt())
UpdateLVal = LValue::MakeVectorElt(Ptr, AtomicLVal.getVectorIdx(),
AtomicLVal.getType(),
AtomicLVal.getAlignment());
else {
assert(AtomicLVal.isExtVectorElt());
UpdateLVal = LValue::MakeExtVectorElt(Ptr, AtomicLVal.getExtVectorElts(),
AtomicLVal.getType(),
AtomicLVal.getAlignment());
}
UpdateLVal.setTBAAInfo(LVal.getTBAAInfo());
UpRVal = EmitLoadOfLValue(UpdateLVal, SourceLocation());
}
// Store new value in the corresponding memory area
RValue NewRVal = UpdateOp(UpRVal);
if (!AtomicLVal.isSimple()) {
EmitStoreThroughLValue(NewRVal, UpdateLVal);
// Load new value
NewRVal = RValue::get(
EmitLoadOfScalar(Ptr, AtomicLVal.isVolatile(),
Atomics.getAtomicAlignment().getQuantity(),
Atomics.getAtomicType(), SourceLocation()));
}
// Try to write new value using cmpxchg operation
auto Pair = Atomics.EmitAtomicCompareExchange(OriginalRValue, NewRVal, AO);
OldVal = IsScalar ? Pair.first.getScalarVal()
: Atomics.convertRValueToInt(Pair.first);
PHI->addIncoming(OldVal, Builder.GetInsertBlock());
Builder.CreateCondBr(Pair.second, ExitBB, ContBB);
EmitBlock(ExitBB, /*IsFinished=*/true);
Atomics.EmitAtomicUpdate(AO, UpdateOp, IsVolatile);
}
void CodeGenFunction::EmitAtomicInit(Expr *init, LValue dest) {

2
clang/lib/CodeGen/CodeGenFunction.h
View File

@ -2272,7 +2272,7 @@ public:
bool IsWeak = false, AggValueSlot Slot = AggValueSlot::ignored());
void EmitAtomicUpdate(LValue LVal, llvm::AtomicOrdering AO,
const std::function<RValue(RValue)> &UpdateOp,
const llvm::function_ref<RValue(RValue)> &UpdateOp,
bool IsVolatile);
/// EmitToMemory - Change a scalar value from its value

507
clang/test/OpenMP/atomic_capture_codegen.cpp
View File

File diff suppressed because it is too large Load Diff

8
clang/test/OpenMP/atomic_codegen.cpp
View File

@ -37,6 +37,8 @@ void parallel_atomic_ewc() {
// CHECK: [[OMP_UPDATE]]
// CHECK: [[OLD_PHI_VAL:%.+]] = phi i32 [ [[OLD_VAL]], %{{.+}} ], [ [[NEW_OLD_VAL:%.+]], %[[OMP_UPDATE]] ]
// CHECK: [[NEW_VAL:%.+]] = srem i32 [[OLD_PHI_VAL]], [[B_VAL]]
// CHECK: store i32 [[NEW_VAL]], i32* [[TEMP:%.+]],
// CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i32* [[SCALAR_ADDR]], i32 [[OLD_PHI_VAL]], i32 [[NEW_VAL]] monotonic monotonic
// CHECK: [[NEW_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[COND:%.+]] = extractvalue { i32, i1 } [[RES]], 1
@ -52,13 +54,15 @@ void parallel_atomic_ewc() {
// CHECK: br label %[[OMP_UPDATE:.+]]
// CHECK: [[OMP_UPDATE]]
// CHECK: [[OLD_PHI_VAL:%.+]] = phi i32 [ [[OLD_VAL]], %{{.+}} ], [ [[NEW_OLD_VAL:%.+]], %[[OMP_UPDATE]] ]
// CHECK: [[NEW_VAL:%.+]] = srem i32 [[OLD_PHI_VAL]], [[B_VAL]]
// CHECK: [[NEW_CALC_VAL:%.+]] = srem i32 [[OLD_PHI_VAL]], [[B_VAL]]
// CHECK: store i32 [[NEW_CALC_VAL]], i32* [[TEMP:%.+]],
// CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i32* [[SCALAR_ADDR]], i32 [[OLD_PHI_VAL]], i32 [[NEW_VAL]] monotonic monotonic
// CHECK: [[NEW_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[COND:%.+]] = extractvalue { i32, i1 } [[RES]], 1
// CHECK: br i1 [[COND]], label %[[OMP_DONE:.+]], label %[[OMP_UPDATE]]
// CHECK: [[OMP_DONE]]
// CHECK: store i32 [[NEW_VAL]], i32* @a,
// CHECK: store i32 [[NEW_CALC_VAL]], i32* @a,
// CHECK: invoke void @_ZN2StD1Ev(%struct.St* [[TEMP_ST_ADDR]])
#pragma omp atomic capture
a = St().get() %= b;

430
clang/test/OpenMP/atomic_update_codegen.cpp
View File

@ -99,6 +99,8 @@ int main() {
// CHECK: [[CONV:%.+]] = zext i16 [[EXPECTED]] to i32
// CHECK: [[ADD:%.+]] = add nsw i32 [[CONV]], [[EXPR]]
// CHECK: [[DESIRED:%.+]] = trunc i32 [[ADD]] to i16
// CHECK: store i16 [[DESIRED]], i16* [[TEMP:%.+]],
// CHECK: [[DESIRED:%.+]] = load i16, i16* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i16* [[X_ADDR]], i16 [[EXPECTED]], i16 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_X]] = extractvalue { i16, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i16, i1 } [[RES]], 1
@ -112,6 +114,8 @@ int main() {
// CHECK: [[CONT]]
// CHECK: [[EXPECTED:%.+]] = phi i32 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[DESIRED:%.+]] = mul nsw i32 [[EXPECTED]], [[EXPR]]
// CHECK: store i32 [[DESIRED]], i32* [[TEMP:%.+]],
// CHECK: [[DESIRED:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i32* [[X_ADDR]], i32 [[EXPECTED]], i32 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_X]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
@ -129,6 +133,8 @@ int main() {
// CHECK: [[CONT]]
// CHECK: [[EXPECTED:%.+]] = phi i32 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[DESIRED:%.+]] = shl i32 [[EXPECTED]], [[EXPR]]
// CHECK: store i32 [[DESIRED]], i32* [[TEMP:%.+]],
// CHECK: [[DESIRED:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i32* [[X_ADDR]], i32 [[EXPECTED]], i32 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_X]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
@ -142,6 +148,8 @@ int main() {
// CHECK: [[CONT]]
// CHECK: [[EXPECTED:%.+]] = phi i32 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[DESIRED:%.+]] = lshr i32 [[EXPECTED]], [[EXPR]]
// CHECK: store i32 [[DESIRED]], i32* [[TEMP:%.+]],
// CHECK: [[DESIRED:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i32* [[X_ADDR]], i32 [[EXPECTED]], i32 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_X]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
@ -155,6 +163,8 @@ int main() {
// CHECK: [[CONT]]
// CHECK: [[EXPECTED:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[DESIRED:%.+]] = sdiv i64 [[EXPECTED]], [[EXPR]]
// CHECK: store i64 [[DESIRED]], i64* [[TEMP:%.+]],
// CHECK: [[DESIRED:%.+]] = load i64, i64* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i64* [[X_ADDR]], i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_X]] = extractvalue { i64, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
@ -176,95 +186,69 @@ int main() {
ullx |= ullv;
// CHECK: [[EXPR:%.+]] = load float, float* @{{.+}},
// CHECK: [[OLD:%.+]] = load atomic i32, i32* bitcast (float* [[X_ADDR:@.+]] to i32*) monotonic
// CHECK: [[X:%.+]] = bitcast i32 [[OLD]] to float
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi float [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[EXPECTED:%.+]] = phi i32 [ [[OLD]], %{{.+}} ], [ [[PREV:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast float* [[TEMP:%.+]] to i32*
// CHECK: [[OLD:%.+]] = bitcast i32 [[EXPECTED]] to float
// CHECK: [[ADD:%.+]] = fadd float [[OLD]], [[EXPR]]
// CHECK: [[EXPECTED:%.+]] = bitcast float [[OLD]] to i32
// CHECK: [[DESIRED:%.+]] = bitcast float [[ADD]] to i32
// CHECK: store float [[ADD]], float* [[TEMP]],
// CHECK: [[DESIRED:%.+]] = load i32, i32* [[BITCAST]],
// CHECK: [[RES:%.+]] = cmpxchg i32* bitcast (float* [[X_ADDR]] to i32*), i32 [[EXPECTED]], i32 [[DESIRED]] monotonic monotonic
// CHECK: [[PREV:%.+]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
// CHECK: [[OLD_X]] = bitcast i32 [[PREV]] to float
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic update
fx = fx + fv;
// CHECK: [[EXPR:%.+]] = load double, double* @{{.+}},
// CHECK: [[OLD:%.+]] = load atomic i64, i64* bitcast (double* [[X_ADDR:@.+]] to i64*) monotonic
// CHECK: [[X:%.+]] = bitcast i64 [[OLD]] to double
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi double [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[EXPECTED:%.+]] = phi i64 [ [[OLD]], %{{.+}} ], [ [[PREV:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast double* [[TEMP:%.+]] to i64*
// CHECK: [[OLD:%.+]] = bitcast i64 [[EXPECTED]] to double
// CHECK: [[SUB:%.+]] = fsub double [[EXPR]], [[OLD]]
// CHECK: [[EXPECTED:%.+]] = bitcast double [[OLD]] to i64
// CHECK: [[DESIRED:%.+]] = bitcast double [[SUB]] to i64
// CHECK: store double [[SUB]], double* [[TEMP]],
// CHECK: [[DESIRED:%.+]] = load i64, i64* [[BITCAST]],
// CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (double* [[X_ADDR]] to i64*), i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
// CHECK: [[PREV:%.+]] = extractvalue { i64, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
// CHECK: [[OLD_X]] = bitcast i64 [[PREV]] to double
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic
dx = dv - dx;
// CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}},
// CHECK: [[OLD:%.+]] = load atomic i128, i128* bitcast (x86_fp80* [[X_ADDR:@.+]] to i128*) monotonic
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i128*
// CHECK: store i128 [[OLD]], i128* [[BITCAST]]
// CHECK: [[X:%.+]] = load x86_fp80, x86_fp80* [[TEMP]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi x86_fp80 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[EXPECTED:%.+]] = phi i128 [ [[OLD]], %{{.+}} ], [ [[PREV:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i128*
// CHECK: store i128 [[EXPECTED]], i128* [[BITCAST]],
// CHECK: [[BITCAST1:%.+]] = bitcast x86_fp80* [[TEMP1:%.+]] to i128*
// CHECK: store i128 [[EXPECTED]], i128* [[BITCAST1]],
// CHECK: [[OLD:%.+]] = load x86_fp80, x86_fp80* [[TEMP1]]
// CHECK: [[MUL:%.+]] = fmul x86_fp80 [[OLD]], [[EXPR]]
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i8*
// CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false)
// CHECK: store x86_fp80 [[OLD]], x86_fp80* [[TEMP]]
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP]] to i128*
// CHECK: [[EXPECTED:%.+]] = load i128, i128* [[BITCAST]]
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i8*
// CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false)
// CHECK: store x86_fp80 [[MUL]], x86_fp80* [[TEMP]]
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP]] to i128*
// CHECK: [[DESIRED:%.+]] = load i128, i128* [[BITCAST]]
// CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (x86_fp80* [[X_ADDR]] to i128*), i128 [[EXPECTED]], i128 [[DESIRED]] monotonic monotonic
// CHECK: [[PREV:%.+]] = extractvalue { i128, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i128, i1 } [[RES]], 1
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i128*
// CHECK: store i128 [[PREV]], i128* [[BITCAST]]
// CHECK: [[OLD_X]] = load x86_fp80, x86_fp80* [[TEMP]],
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic update
ldx = ldx * ldv;
// CHECK: [[EXPR_RE:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 0)
// CHECK: [[EXPR_IM:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 1)
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP:%.+]] to i8*
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR:%.+]] to i8*
// CHECK: call void @__atomic_load(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR:@.+]] to i8*), i8* [[BITCAST]], i32 0)
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 0
// CHECK: [[LD_RE:%.+]] = load i32, i32* [[LD_RE_ADDR]]
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: [[LD_IM:%.+]] = load i32, i32* [[LD_IM_ADDR]]
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: store i32 [[LD_RE]], i32* [[LD_RE_ADDR]]
// CHECK: store i32 [[LD_IM]], i32* [[LD_IM_ADDR]]
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i64*
// CHECK: [[X:%.+]] = load i64, i64* [[BITCAST]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP:%.+]] to i64*
// CHECK: store i64 [[OLD]], i64* [[BITCAST]]
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 0
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 0
// CHECK: [[X_RE:%.+]] = load i32, i32* [[X_RE_ADDR]]
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load i32, i32* [[X_IM_ADDR]]
// <Skip checks for complex calculations>
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR:%.+]], i32 0, i32 0
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 1
// CHECK: store i32 [[X_RE]], i32* [[X_RE_ADDR]]
// CHECK: store i32 [[X_IM]], i32* [[X_IM_ADDR]]
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR:%.+]], i32 0, i32 0
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR]], i32 0, i32 1
// CHECK: store i32 %{{.+}}, i32* [[X_RE_ADDR]]
@ -272,48 +256,21 @@ int main() {
// CHECK: [[EXPECTED:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR]] to i8*
// CHECK: [[DESIRED:%.+]] = bitcast { i32, i32 }* [[DESIRED_ADDR]] to i8*
// CHECK: [[SUCCESS_FAIL:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR]] to i8*), i8* [[EXPECTED]], i8* [[DESIRED]], i32 0, i32 0)
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[X_RE:%.+]] = load i32, i32* [[X_RE_ADDR]]
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load i32, i32* [[X_IM_ADDR]]
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: store i32 [[X_RE]], i32* [[LD_RE_ADDR]]
// CHECK: store i32 [[X_IM]], i32* [[LD_IM_ADDR]]
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i64*
// CHECK: [[OLD_X]] = load i64, i64* [[BITCAST]]
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic
cix = civ / cix;
// CHECK: [[EXPR_RE:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.+}}, i32 0, i32 0)
// CHECK: [[EXPR_IM:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.+}}, i32 0, i32 1)
// CHECK: [[BITCAST:%.+]] = bitcast { float, float }* [[TEMP:%.+]] to i8*
// CHECK: [[BITCAST:%.+]] = bitcast { float, float }* [[EXPECTED_ADDR:%.+]] to i8*
// CHECK: call void @__atomic_load(i64 8, i8* bitcast ({ float, float }* [[X_ADDR:@.+]] to i8*), i8* [[BITCAST]], i32 0)
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 0
// CHECK: [[LD_RE:%.+]] = load float, float* [[LD_RE_ADDR]]
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1
// CHECK: [[LD_IM:%.+]] = load float, float* [[LD_IM_ADDR]]
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1
// CHECK: store float [[LD_RE]], float* [[LD_RE_ADDR]]
// CHECK: store float [[LD_IM]], float* [[LD_IM_ADDR]]
// CHECK: [[BITCAST:%.+]] = bitcast { float, float }* [[TEMP]] to i64*
// CHECK: [[X:%.+]] = load i64, i64* [[BITCAST]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast { float, float }* [[TEMP:%.+]] to i64*
// CHECK: store i64 [[OLD]], i64* [[BITCAST]]
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 0
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[EXPECTED_ADDR]], i32 0, i32 0
// CHECK: [[X_RE:%.+]] = load float, float* [[X_RE_ADDR]]
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[EXPECTED_ADDR]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load float, float* [[X_IM_ADDR]]
// <Skip checks for complex calculations>
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[EXPECTED_ADDR:%.+]], i32 0, i32 0
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[EXPECTED_ADDR]], i32 0, i32 1
// CHECK: store float [[X_RE]], float* [[X_RE_ADDR]]
// CHECK: store float [[X_IM]], float* [[X_IM_ADDR]]
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[DESIRED_ADDR:%.+]], i32 0, i32 0
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[DESIRED_ADDR]], i32 0, i32 1
// CHECK: store float %{{.+}}, float* [[X_RE_ADDR]]
@ -321,48 +278,21 @@ int main() {
// CHECK: [[EXPECTED:%.+]] = bitcast { float, float }* [[EXPECTED_ADDR]] to i8*
// CHECK: [[DESIRED:%.+]] = bitcast { float, float }* [[DESIRED_ADDR]] to i8*
// CHECK: [[SUCCESS_FAIL:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 8, i8* bitcast ({ float, float }* [[X_ADDR]] to i8*), i8* [[EXPECTED]], i8* [[DESIRED]], i32 0, i32 0)
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[X_RE:%.+]] = load float, float* [[X_RE_ADDR]]
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load float, float* [[X_IM_ADDR]]
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1
// CHECK: store float [[X_RE]], float* [[LD_RE_ADDR]]
// CHECK: store float [[X_IM]], float* [[LD_IM_ADDR]]
// CHECK: [[BITCAST:%.+]] = bitcast { float, float }* [[TEMP]] to i64*
// CHECK: [[OLD_X]] = load i64, i64* [[BITCAST]]
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic update
cfx = cfv + cfx;
// CHECK: [[EXPR_RE:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.+}}, i32 0, i32 0)
// CHECK: [[EXPR_IM:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.+}}, i32 0, i32 1)
// CHECK: [[BITCAST:%.+]] = bitcast { double, double }* [[TEMP:%.+]] to i8*
// CHECK: [[BITCAST:%.+]] = bitcast { double, double }* [[EXPECTED_ADDR:%.+]] to i8*
// CHECK: call void @__atomic_load(i64 16, i8* bitcast ({ double, double }* [[X_ADDR:@.+]] to i8*), i8* [[BITCAST]], i32 5)
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 0
// CHECK: [[LD_RE:%.+]] = load double, double* [[LD_RE_ADDR]]
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 1
// CHECK: [[LD_IM:%.+]] = load double, double* [[LD_IM_ADDR]]
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 1
// CHECK: store double [[LD_RE]], double* [[LD_RE_ADDR]]
// CHECK: store double [[LD_IM]], double* [[LD_IM_ADDR]]
// CHECK: [[BITCAST:%.+]] = bitcast { double, double }* [[TEMP]] to i128*
// CHECK: [[X:%.+]] = load i128, i128* [[BITCAST]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi i128 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast { double, double }* [[TEMP:%.+]] to i128*
// CHECK: store i128 [[OLD]], i128* [[BITCAST]]
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 0
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[EXPECTED_ADDR]], i32 0, i32 0
// CHECK: [[X_RE:%.+]] = load double, double* [[X_RE_ADDR]]
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[EXPECTED_ADDR]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load double, double* [[X_IM_ADDR]]
// <Skip checks for complex calculations>
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[EXPECTED_ADDR:%.+]], i32 0, i32 0
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[EXPECTED_ADDR]], i32 0, i32 1
// CHECK: store double [[X_RE]], double* [[X_RE_ADDR]]
// CHECK: store double [[X_IM]], double* [[X_IM_ADDR]]
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[DESIRED_ADDR:%.+]], i32 0, i32 0
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[DESIRED_ADDR]], i32 0, i32 1
// CHECK: store double %{{.+}}, double* [[X_RE_ADDR]]
@ -370,16 +300,6 @@ int main() {
// CHECK: [[EXPECTED:%.+]] = bitcast { double, double }* [[EXPECTED_ADDR]] to i8*
// CHECK: [[DESIRED:%.+]] = bitcast { double, double }* [[DESIRED_ADDR]] to i8*
// CHECK: [[SUCCESS_FAIL:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 16, i8* bitcast ({ double, double }* [[X_ADDR]] to i8*), i8* [[EXPECTED]], i8* [[DESIRED]], i32 5, i32 5)
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[X_RE:%.+]] = load double, double* [[X_RE_ADDR]]
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load double, double* [[X_IM_ADDR]]
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 1
// CHECK: store double [[X_RE]], double* [[LD_RE_ADDR]]
// CHECK: store double [[X_IM]], double* [[LD_IM_ADDR]]
// CHECK: [[BITCAST:%.+]] = bitcast { double, double }* [[TEMP]] to i128*
// CHECK: [[OLD_X]] = load i128, i128* [[BITCAST]]
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
// CHECK: call{{.*}} @__kmpc_flush(
@ -394,19 +314,19 @@ int main() {
// CHECK: [[CV:%.+]] = load i8, i8* @{{.+}}, align 1
// CHECK: [[EXPR:%.+]] = sext i8 [[CV]] to i32
// CHECK: [[BX:%.+]] = load atomic i8, i8* [[BX_ADDR:@.+]] monotonic
// CHECK: [[X:%.+]] = trunc i8 [[BX]] to i1
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi i1 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[EXPECTED:%.+]] = phi i8 [ [[BX]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[OLD:%.+]] = trunc i8 [[EXPECTED]] to i1
// CHECK: [[X_RVAL:%.+]] = zext i1 [[OLD]] to i32
// CHECK: [[AND:%.+]] = and i32 [[EXPR]], [[X_RVAL]]
// CHECK: [[CAST:%.+]] = icmp ne i32 [[AND]], 0
// CHECK: [[EXPECTED:%.+]] = zext i1 [[OLD]] to i8
// CHECK: [[DESIRED:%.+]] = zext i1 [[CAST]] to i8
// CHECK: store i8 [[DESIRED]], i8* [[TEMP:%.+]],
// CHECK: [[DESIRED:%.+]] = load i8, i8* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i8* [[BX_ADDR]], i8 [[EXPECTED]], i8 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD:%.+]] = extractvalue { i8, i1 } [[RES]], 0
// CHECK: [[OLD_X:%.+]] = extractvalue { i8, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i8, i1 } [[RES]], 1
// CHECK: [[OLD_X]] = trunc i8 [[OLD]] to i1
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic
@ -420,6 +340,8 @@ int main() {
// CHECK: [[X_RVAL:%.+]] = sext i8 [[EXPECTED]] to i32
// CHECK: [[ASHR:%.+]] = ashr i32 [[X_RVAL]], [[EXPR]]
// CHECK: [[DESIRED:%.+]] = trunc i32 [[ASHR]] to i8
// CHECK: store i8 [[DESIRED]], i8* [[TEMP:%.+]],
// CHECK: [[DESIRED:%.+]] = load i8, i8* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i8* [[CX_ADDR]], i8 [[EXPECTED]], i8 [[DESIRED]] seq_cst seq_cst
// CHECK: [[OLD_X:%.+]] = extractvalue { i8, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i8, i1 } [[RES]], 1
@ -437,6 +359,8 @@ int main() {
// CHECK: [[X_RVAL:%.+]] = trunc i64 [[EXPECTED]] to i32
// CHECK: [[SHL:%.+]] = shl i32 [[EXPR]], [[X_RVAL]]
// CHECK: [[DESIRED:%.+]] = sext i32 [[SHL]] to i64
// CHECK: store i64 [[DESIRED]], i64* [[TEMP:%.+]],
// CHECK: [[DESIRED:%.+]] = load i64, i64* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i64* [[ULX_ADDR]], i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_X:%.+]] = extractvalue { i64, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
@ -451,6 +375,8 @@ int main() {
// CHECK: [[CONT]]
// CHECK: [[EXPECTED:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[DESIRED:%.+]] = srem i64 [[EXPECTED]], [[EXPR]]
// CHECK: store i64 [[DESIRED]], i64* [[TEMP:%.+]],
// CHECK: [[DESIRED:%.+]] = load i64, i64* [[TEMP]],
// CHECK: [[RES:%.+]] = cmpxchg i64* [[LX_ADDR]], i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_X:%.+]] = extractvalue { i64, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
@ -468,32 +394,15 @@ int main() {
#pragma omp atomic
ix = ix & uiv;
// CHECK: [[EXPR:%.+]] = load i64, i64* @{{.+}},
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP:%.+]] to i8*
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR:%.+]] to i8*
// CHECK: call void @__atomic_load(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR:@.+]] to i8*), i8* [[BITCAST]], i32 0)
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 0
// CHECK: [[LD_RE:%.+]] = load i32, i32* [[LD_RE_ADDR]]
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: [[LD_IM:%.+]] = load i32, i32* [[LD_IM_ADDR]]
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: store i32 [[LD_RE]], i32* [[LD_RE_ADDR]]
// CHECK: store i32 [[LD_IM]], i32* [[LD_IM_ADDR]]
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i64*
// CHECK: [[X:%.+]] = load i64, i64* [[BITCAST]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP:%.+]] to i64*
// CHECK: store i64 [[OLD]], i64* [[BITCAST]]
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 0
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 0
// CHECK: [[X_RE:%.+]] = load i32, i32* [[X_RE_ADDR]]
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load i32, i32* [[X_IM_ADDR]]
// <Skip checks for complex calculations>
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR:%.+]], i32 0, i32 0
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 1
// CHECK: store i32 [[X_RE]], i32* [[X_RE_ADDR]]
// CHECK: store i32 [[X_IM]], i32* [[X_IM_ADDR]]
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR:%.+]], i32 0, i32 0
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR]], i32 0, i32 1
// CHECK: store i32 %{{.+}}, i32* [[X_RE_ADDR]]
@ -501,16 +410,6 @@ int main() {
// CHECK: [[EXPECTED:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR]] to i8*
// CHECK: [[DESIRED:%.+]] = bitcast { i32, i32 }* [[DESIRED_ADDR]] to i8*
// CHECK: [[SUCCESS_FAIL:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR]] to i8*), i8* [[EXPECTED]], i8* [[DESIRED]], i32 0, i32 0)
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[X_RE:%.+]] = load i32, i32* [[X_RE_ADDR]]
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load i32, i32* [[X_IM_ADDR]]
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: store i32 [[X_RE]], i32* [[LD_RE_ADDR]]
// CHECK: store i32 [[X_IM]], i32* [[LD_IM_ADDR]]
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i64*
// CHECK: [[OLD_X]] = load i64, i64* [[BITCAST]]
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic update
@ -518,17 +417,17 @@ int main() {
// CHECK: [[ULV:%.+]] = load i64, i64* @{{.+}},
// CHECK: [[EXPR:%.+]] = uitofp i64 [[ULV]] to float
// CHECK: [[OLD:%.+]] = load atomic i32, i32* bitcast (float* [[X_ADDR:@.+]] to i32*) monotonic
// CHECK: [[X:%.+]] = bitcast i32 [[OLD]] to float
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi float [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[EXPECTED:%.+]] = phi i32 [ [[OLD]], %{{.+}} ], [ [[PREV:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast float* [[TEMP:%.+]] to i32*
// CHECK: [[OLD:%.+]] = bitcast i32 [[EXPECTED]] to float
// CHECK: [[MUL:%.+]] = fmul float [[OLD]], [[EXPR]]
// CHECK: [[EXPECTED:%.+]] = bitcast float [[OLD]] to i32
// CHECK: [[DESIRED:%.+]] = bitcast float [[MUL]] to i32
// CHECK: store float [[MUL]], float* [[TEMP]],
// CHECK: [[DESIRED:%.+]] = load i32, i32* [[BITCAST]],
// CHECK: [[RES:%.+]] = cmpxchg i32* bitcast (float* [[X_ADDR]] to i32*), i32 [[EXPECTED]], i32 [[DESIRED]] monotonic monotonic
// CHECK: [[PREV:%.+]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
// CHECK: [[OLD_X]] = bitcast i32 [[PREV]] to float
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic
@ -536,17 +435,17 @@ int main() {
// CHECK: [[LLV:%.+]] = load i64, i64* @{{.+}},
// CHECK: [[EXPR:%.+]] = sitofp i64 [[LLV]] to double
// CHECK: [[OLD:%.+]] = load atomic i64, i64* bitcast (double* [[X_ADDR:@.+]] to i64*) monotonic
// CHECK: [[X:%.+]] = bitcast i64 [[OLD]] to double
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi double [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[EXPECTED:%.+]] = phi i64 [ [[OLD]], %{{.+}} ], [ [[PREV:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast double* [[TEMP:%.+]] to i64*
// CHECK: [[OLD:%.+]] = bitcast i64 [[EXPECTED]] to double
// CHECK: [[DIV:%.+]] = fdiv double [[OLD]], [[EXPR]]
// CHECK: [[EXPECTED:%.+]] = bitcast double [[OLD]] to i64
// CHECK: [[DESIRED:%.+]] = bitcast double [[DIV]] to i64
// CHECK: store double [[DIV]], double* [[TEMP]],
// CHECK: [[DESIRED:%.+]] = load i64, i64* [[BITCAST]],
// CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (double* [[X_ADDR]] to i64*), i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
// CHECK: [[PREV:%.+]] = extractvalue { i64, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
// CHECK: [[OLD_X]] = bitcast i64 [[PREV]] to double
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic update
@ -554,60 +453,33 @@ int main() {
// CHECK: [[ULLV:%.+]] = load i64, i64* @{{.+}},
// CHECK: [[EXPR:%.+]] = uitofp i64 [[ULLV]] to x86_fp80
// CHECK: [[OLD:%.+]] = load atomic i128, i128* bitcast (x86_fp80* [[X_ADDR:@.+]] to i128*) monotonic
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i128*
// CHECK: store i128 [[OLD]], i128* [[BITCAST]]
// CHECK: [[X:%.+]] = load x86_fp80, x86_fp80* [[TEMP]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi x86_fp80 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[EXPECTED:%.+]] = phi i128 [ [[OLD]], %{{.+}} ], [ [[PREV:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST1:%.+]] = bitcast x86_fp80* [[TEMP1:%.+]] to i128*
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i128*
// CHECK: store i128 [[EXPECTED]], i128* [[BITCAST]]
// CHECK: [[OLD:%.+]] = load x86_fp80, x86_fp80* [[TEMP]]
// CHECK: [[SUB:%.+]] = fsub x86_fp80 [[OLD]], [[EXPR]]
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i8*
// CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false)
// CHECK: store x86_fp80 [[OLD]], x86_fp80* [[TEMP]]
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP]] to i128*
// CHECK: [[EXPECTED:%.+]] = load i128, i128* [[BITCAST]]
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i8*
// CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false)
// CHECK: store x86_fp80 [[SUB]], x86_fp80* [[TEMP]]
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP]] to i128*
// CHECK: [[DESIRED:%.+]] = load i128, i128* [[BITCAST]]
// CHECK: store x86_fp80 [[SUB]], x86_fp80* [[TEMP1]]
// CHECK: [[DESIRED:%.+]] = load i128, i128* [[BITCAST1]]
// CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (x86_fp80* [[X_ADDR]] to i128*), i128 [[EXPECTED]], i128 [[DESIRED]] monotonic monotonic
// CHECK: [[PREV:%.+]] = extractvalue { i128, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i128, i1 } [[RES]], 1
// CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i128*
// CHECK: store i128 [[PREV]], i128* [[BITCAST]]
// CHECK: [[OLD_X]] = load x86_fp80, x86_fp80* [[TEMP]],
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic
ldx -= ullv;
// CHECK: [[EXPR:%.+]] = load float, float* @{{.+}},
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP:%.+]] to i8*
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR:%.+]] to i8*
// CHECK: call void @__atomic_load(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR:@.+]] to i8*), i8* [[BITCAST]], i32 0)
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 0
// CHECK: [[LD_RE:%.+]] = load i32, i32* [[LD_RE_ADDR]]
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: [[LD_IM:%.+]] = load i32, i32* [[LD_IM_ADDR]]
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: store i32 [[LD_RE]], i32* [[LD_RE_ADDR]]
// CHECK: store i32 [[LD_IM]], i32* [[LD_IM_ADDR]]
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i64*
// CHECK: [[X:%.+]] = load i64, i64* [[BITCAST]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP:%.+]] to i64*
// CHECK: store i64 [[OLD]], i64* [[BITCAST]]
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 0
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 0
// CHECK: [[X_RE:%.+]] = load i32, i32* [[X_RE_ADDR]]
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load i32, i32* [[X_IM_ADDR]]
// <Skip checks for complex calculations>
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR:%.+]], i32 0, i32 0
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 1
// CHECK: store i32 [[X_RE]], i32* [[X_RE_ADDR]]
// CHECK: store i32 [[X_IM]], i32* [[X_IM_ADDR]]
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR:%.+]], i32 0, i32 0
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR]], i32 0, i32 1
// CHECK: store i32 %{{.+}}, i32* [[X_RE_ADDR]]
@ -615,16 +487,6 @@ int main() {
// CHECK: [[EXPECTED:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR]] to i8*
// CHECK: [[DESIRED:%.+]] = bitcast { i32, i32 }* [[DESIRED_ADDR]] to i8*
// CHECK: [[SUCCESS_FAIL:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR]] to i8*), i8* [[EXPECTED]], i8* [[DESIRED]], i32 0, i32 0)
// CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[X_RE:%.+]] = load i32, i32* [[X_RE_ADDR]]
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load i32, i32* [[X_IM_ADDR]]
// CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
// CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
// CHECK: store i32 [[X_RE]], i32* [[LD_RE_ADDR]]
// CHECK: store i32 [[X_IM]], i32* [[LD_IM_ADDR]]
// CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i64*
// CHECK: [[OLD_X]] = load i64, i64* [[BITCAST]]
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic update
@ -638,6 +500,8 @@ int main() {
// CHECK: [[X_RVAL:%.+]] = sitofp i32 [[CONV]] to double
// CHECK: [[ADD:%.+]] = fadd double [[X_RVAL]], [[EXPR]]
// CHECK: [[DESIRED:%.+]] = fptosi double [[ADD]] to i16
// CHECK: store i16 [[DESIRED]], i16* [[TEMP:%.+]]
// CHECK: [[DESIRED:%.+]] = load i16, i16* [[TEMP]]
// CHECK: [[RES:%.+]] = cmpxchg i16* [[X_ADDR]], i16 [[EXPECTED]], i16 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_X]] = extractvalue { i16, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i16, i1 } [[RES]], 1
@ -647,20 +511,20 @@ int main() {
sx = sx + dv;
// CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}},
// CHECK: [[XI8:%.+]] = load atomic i8, i8* [[X_ADDR:@.+]] monotonic
// CHECK: [[X:%.+]] = trunc i8 [[XI8]] to i1
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[BOOL_EXPECTED:%.+]] = phi i1 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[EXPECTED:%.+]] = phi i8 [ [[XI8]], %{{.+}} ], [ [[OLD_XI8:%.+]], %[[CONT]] ]
// CHECK: [[BOOL_EXPECTED:%.+]] = trunc i8 [[EXPECTED]] to i1
// CHECK: [[CONV:%.+]] = zext i1 [[BOOL_EXPECTED]] to i32
// CHECK: [[X_RVAL:%.+]] = sitofp i32 [[CONV]] to x86_fp80
// CHECK: [[MUL:%.+]] = fmul x86_fp80 [[EXPR]], [[X_RVAL]]
// CHECK: [[BOOL_DESIRED:%.+]] = fcmp une x86_fp80 [[MUL]], 0xK00000000000000000000
// CHECK: [[EXPECTED:%.+]] = zext i1 [[BOOL_EXPECTED]] to i8
// CHECK: [[DESIRED:%.+]] = zext i1 [[BOOL_DESIRED]] to i8
// CHECK: store i8 [[DESIRED]], i8* [[TEMP:%.+]]
// CHECK: [[DESIRED:%.+]] = load i8, i8* [[TEMP]]
// CHECK: [[RES:%.+]] = cmpxchg i8* [[X_ADDR]], i8 [[EXPECTED]], i8 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_XI8:%.+]] = extractvalue { i8, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i8, i1 } [[RES]], 1
// CHECK: [[OLD_X]] = trunc i8 [[OLD_XI8]] to i1
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic update
@ -668,22 +532,22 @@ int main() {
// CHECK: [[EXPR_RE:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* [[CIV_ADDR:@.+]], i32 0, i32 0),
// CHECK: [[EXPR_IM:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* [[CIV_ADDR]], i32 0, i32 1),
// CHECK: [[XI8:%.+]] = load atomic i8, i8* [[X_ADDR:@.+]] monotonic
// CHECK: [[X:%.+]] = trunc i8 [[XI8]] to i1
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[BOOL_EXPECTED:%.+]] = phi i1 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
// CHECK: [[EXPECTED:%.+]] = phi i8 [ [[XI8]], %{{.+}} ], [ [[OLD_XI8:%.+]], %[[CONT]] ]
// CHECK: [[BOOL_EXPECTED:%.+]] = trunc i8 [[EXPECTED]] to i1
// CHECK: [[X_RVAL:%.+]] = zext i1 [[BOOL_EXPECTED]] to i32
// CHECK: [[SUB_RE:%.+]] = sub i32 [[EXPR_RE:%.+]], [[X_RVAL]]
// CHECK: [[SUB_IM:%.+]] = sub i32 [[EXPR_IM:%.+]], 0
// CHECK: icmp ne i32 [[SUB_RE]], 0
// CHECK: icmp ne i32 [[SUB_IM]], 0
// CHECK: [[BOOL_DESIRED:%.+]] = or i1
// CHECK: [[EXPECTED:%.+]] = zext i1 [[BOOL_EXPECTED]] to i8
// CHECK: [[DESIRED:%.+]] = zext i1 [[BOOL_DESIRED]] to i8
// CHECK: store i8 [[DESIRED]], i8* [[TEMP:%.+]]
// CHECK: [[DESIRED:%.+]] = load i8, i8* [[TEMP]]
// CHECK: [[RES:%.+]] = cmpxchg i8* [[X_ADDR]], i8 [[EXPECTED]], i8 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_XI8:%.+]] = extractvalue { i8, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i8, i1 } [[RES]], 1
// CHECK: [[OLD_X]] = trunc i8 [[OLD_XI8]] to i1
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic
@ -702,6 +566,8 @@ int main() {
// CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1
// CHECK: [[X_IM:%.+]] = load float, float* [[X_IM_ADDR]]
// CHECK: [[DESIRED:%.+]] = fptoui float [[X_RE]] to i16
// CHECK: store i16 [[DESIRED]], i16* [[TEMP:%.+]]
// CHECK: [[DESIRED:%.+]] = load i16, i16* [[TEMP]]
// CHECK: [[RES:%.+]] = cmpxchg i16* [[X_ADDR]], i16 [[EXPECTED]], i16 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_X]] = extractvalue { i16, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i16, i1 } [[RES]], 1
@ -719,6 +585,8 @@ int main() {
// CHECK: [[ADD_RE:%.+]] = fadd double [[X_RVAL]], [[EXPR_RE]]
// CHECK: [[ADD_IM:%.+]] = fadd double 0.000000e+00, [[EXPR_IM]]
// CHECK: [[DESIRED:%.+]] = fptosi double [[ADD_RE]] to i64
// CHECK: store i64 [[DESIRED]], i64* [[TEMP:%.+]]
// CHECK: [[DESIRED:%.+]] = load i64, i64* [[TEMP]]
// CHECK: [[RES:%.+]] = cmpxchg i64* [[X_ADDR]], i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
// CHECK: [[OLD_X]] = extractvalue { i64, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
@ -730,24 +598,23 @@ int main() {
// CHECK: load i8, i8*
// CHECK: [[VEC_ITEM_VAL:%.+]] = zext i1 %{{.+}} to i32
// CHECK: [[I128VAL:%.+]] = load atomic i128, i128* bitcast (<4 x i32>* [[DEST:@.+]] to i128*) monotonic
// CHECK: [[LD:%.+]] = bitcast i128 [[I128VAL]] to <4 x i32>
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_VEC_VAL:%.+]] = phi <4 x i32> [ [[LD]], %{{.+}} ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[OLD_I128:%.+]] = phi i128 [ [[I128VAL]], %{{.+}} ], [ [[FAILED_I128_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast <4 x i32>* [[TEMP:%.+]] to i128*
// CHECK: store i128 [[OLD_I128]], i128* [[BITCAST]],
// CHECK: [[OLD_VEC_VAL:%.+]] = bitcast i128 [[OLD_I128]] to <4 x i32>
// CHECK: store <4 x i32> [[OLD_VEC_VAL]], <4 x i32>* [[LDTEMP:%.+]],
// CHECK: [[VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[LDTEMP]]
// CHECK: [[ITEM:%.+]] = extractelement <4 x i32> [[VEC_VAL]], i16 [[IDX]]
// CHECK: [[OR:%.+]] = or i32 [[ITEM]], [[VEC_ITEM_VAL]]
// CHECK: [[VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[LDTEMP]]
// CHECK: [[VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[TEMP]]
// CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <4 x i32> [[VEC_VAL]], i32 [[OR]], i16 [[IDX]]
// CHECK: store <4 x i32> [[NEW_VEC_VAL]], <4 x i32>* [[LDTEMP]]
// CHECK: [[NEW_VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[LDTEMP]]
// CHECK: [[OLD_I128:%.+]] = bitcast <4 x i32> [[OLD_VEC_VAL]] to i128
// CHECK: [[NEW_I128:%.+]] = bitcast <4 x i32> [[NEW_VEC_VAL]] to i128
// CHECK: store <4 x i32> [[NEW_VEC_VAL]], <4 x i32>* [[TEMP]]
// CHECK: [[NEW_I128:%.+]] = load i128, i128* [[BITCAST]]
// CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (<4 x i32>* [[DEST]] to i128*), i128 [[OLD_I128]], i128 [[NEW_I128]] monotonic monotonic
// CHECK: [[FAILED_I128_OLD_VAL:%.+]] = extractvalue { i128, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i128, i1 } [[RES]], 1
// CHECK: [[FAILED_OLD_VAL]] = bitcast i128 [[FAILED_I128_OLD_VAL]] to <4 x i32>
// CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic update
@ -757,6 +624,7 @@ int main() {
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP1:%.+]],
// CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP:%.+]],
// CHECK: [[A_LD:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[A_SHL:%.+]] = shl i32 [[A_LD]], 1
@ -764,12 +632,12 @@ int main() {
// CHECK: [[X_RVAL:%.+]] = sitofp i32 [[A_ASHR]] to x86_fp80
// CHECK: [[SUB:%.+]] = fsub x86_fp80 [[X_RVAL]], [[EXPR]]
// CHECK: [[CONV:%.+]] = fptosi x86_fp80 [[SUB]] to i32
// CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP1]],
// CHECK: [[BF_VALUE:%.+]] = and i32 [[CONV]], 2147483647
// CHECK: [[BF_CLEAR:%.+]] = and i32 [[NEW_VAL]], -2147483648
// CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
// CHECK: store i32 %{{.+}}, i32* [[TEMP1]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[TEMP1]]
// CHECK: [[RES:%.+]] = cmpxchg i32* bitcast (i8* getelementptr (i8, i8* bitcast (%struct.BitFields* @{{.+}} to i8*), i64 4) to i32*), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
// CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
@ -780,29 +648,26 @@ int main() {
// CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
// CHECK: [[BITCAST:%.+]] = bitcast i32* [[LDTEMP:%.+]] to i8*
// CHECK: call void @__atomic_load(i64 4, i8* getelementptr (i8, i8* bitcast (%struct.BitFields_packed* @{{.+}} to i8*), i64 4), i8* [[BITCAST]], i32 0)
// CHECK: [[PREV_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP:%.+]],
// CHECK: [[PREV_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
// CHECK: store i32 [[PREV_VALUE]], i32* [[TEMP1:%.+]],
// CHECK: [[PREV_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
// CHECK: store i32 [[PREV_VALUE]], i32* [[TEMP:%.+]],
// CHECK: [[A_LD:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[A_SHL:%.+]] = shl i32 [[A_LD]], 1
// CHECK: [[A_ASHR:%.+]] = ashr i32 [[A_SHL]], 1
// CHECK: [[X_RVAL:%.+]] = sitofp i32 [[A_ASHR]] to x86_fp80
// CHECK: [[MUL:%.+]] = fmul x86_fp80 [[X_RVAL]], [[EXPR]]
// CHECK: [[CONV:%.+]] = fptosi x86_fp80 [[MUL]] to i32
// CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP1]],
// CHECK: [[BF_VALUE:%.+]] = and i32 [[CONV]], 2147483647
// CHECK: [[BF_CLEAR:%.+]] = and i32 [[NEW_VAL]], -2147483648
// CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
// CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP_OLD_BF_ADDR:%.+]],
// CHECK: store i32 [[NEW_BF_VALUE]], i32* [[TEMP_NEW_BF_ADDR:%.+]],
// CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* [[TEMP_OLD_BF_ADDR]] to i8*
// CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* [[TEMP_NEW_BF_ADDR]] to i8*
// CHECK: store i32 %{{.+}}, i32* [[TEMP1]]
// CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* [[LDTEMP]] to i8*
// CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* [[TEMP1]] to i8*
// CHECK: [[FAIL_SUCCESS:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 4, i8* getelementptr (i8, i8* bitcast (%struct.BitFields_packed* @{{.+}} to i8*), i64 4), i8* [[BITCAST_TEMP_OLD_BF_ADDR]], i8* [[BITCAST_TEMP_NEW_BF_ADDR]], i32 0, i32 0)
// CHECK: [[FAILED_OLD_VAL]] = load i32, i32* [[TEMP_OLD_BF_ADDR]]
// CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic update
@ -812,19 +677,20 @@ int main() {
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP1:%.+]],
// CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP:%.+]],
// CHECK: [[A_LD:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[A_ASHR:%.+]] = ashr i32 [[A_LD]], 31
// CHECK: [[X_RVAL:%.+]] = sitofp i32 [[A_ASHR]] to x86_fp80
// CHECK: [[SUB:%.+]] = fsub x86_fp80 [[X_RVAL]], [[EXPR]]
// CHECK: [[CONV:%.+]] = fptosi x86_fp80 [[SUB]] to i32
// CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP1]],
// CHECK: [[BF_AND:%.+]] = and i32 [[CONV]], 1
// CHECK: [[BF_VALUE:%.+]] = shl i32 [[BF_AND]], 31
// CHECK: [[BF_CLEAR:%.+]] = and i32 [[NEW_VAL]], 2147483647
// CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
// CHECK: store i32 %{{.+}}, i32* [[TEMP1]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[TEMP1]]
// CHECK: [[RES:%.+]] = cmpxchg i32* getelementptr inbounds (%struct.BitFields2, %struct.BitFields2* @{{.+}}, i32 0, i32 0), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
// CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
@ -837,6 +703,8 @@ int main() {
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST1:%.+]] = bitcast i32* %{{.+}} to i8*
// CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST1]],
// CHECK: [[BITCAST:%.+]] = bitcast i32* %{{.+}} to i8*
// CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST]],
// CHECK: [[A_LD:%.+]] = load i8, i8* [[BITCAST]],
@ -846,13 +714,13 @@ int main() {
// CHECK: [[DIV:%.+]] = fdiv x86_fp80 [[EXPR]], [[X_RVAL]]
// CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 [[DIV]] to i32
// CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i8
// CHECK: [[BF_LD:%.+]] = load i8, i8* [[BITCAST]],
// CHECK: [[BF_LD:%.+]] = load i8, i8* [[BITCAST1]],
// CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 1
// CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 7
// CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, 127
// CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
// CHECK: store i8 %{{.+}}, i8* [[BITCAST1]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[BITCAST1]]
// CHECK: [[RES:%.+]] = cmpxchg i8* getelementptr (i8, i8* bitcast (%struct.BitFields2_packed* @{{.+}} to i8*), i64 3), i8 [[OLD_BF_VALUE]], i8 [[NEW_BF_VALUE]] monotonic monotonic
// CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
@ -865,6 +733,7 @@ int main() {
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP1:%.+]],
// CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP:%.+]],
// CHECK: [[A_LD:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[A_SHL:%.+]] = shl i32 [[A_LD]], 7
@ -872,13 +741,13 @@ int main() {
// CHECK: [[X_RVAL:%.+]] = sitofp i32 [[A_ASHR]] to x86_fp80
// CHECK: [[DIV:%.+]] = fdiv x86_fp80 [[X_RVAL]], [[EXPR]]
// CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 [[DIV]] to i32
// CHECK: [[BF_LD:%.+]] = load i32, i32* [[TEMP]],
// CHECK: [[BF_LD:%.+]] = load i32, i32* [[TEMP1]],
// CHECK: [[BF_AND:%.+]] = and i32 [[NEW_VAL]], 16383
// CHECK: [[BF_VALUE:%.+]] = shl i32 [[BF_AND]], 11
// CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -33552385
// CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
// CHECK: store i32 %{{.+}}, i32* [[TEMP1]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[TEMP1]]
// CHECK: [[RES:%.+]] = cmpxchg i32* getelementptr inbounds (%struct.BitFields3, %struct.BitFields3* @{{.+}}, i32 0, i32 0), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
// CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
@ -890,13 +759,13 @@ int main() {
// CHECK: [[LDTEMP:%.+]] = bitcast i32* %{{.+}} to i24*
// CHECK: [[BITCAST:%.+]] = bitcast i24* %{{.+}} to i8*
// CHECK: call void @__atomic_load(i64 3, i8* getelementptr (i8, i8* bitcast (%struct.BitFields3_packed* @{{.+}} to i8*), i64 1), i8* [[BITCAST]], i32 0)
// CHECK: [[PREV_VALUE:%.+]] = load i24, i24* [[LDTEMP]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i24 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast i32* %{{.+}} to i24*
// CHECK: store i24 [[OLD_BF_VALUE]], i24* [[BITCAST]],
// CHECK: [[A_LD:%.+]] = load i24, i24* [[BITCAST]],
// CHECK: [[PREV_VALUE:%.+]] = load i24, i24* [[LDTEMP]]
// CHECK: store i24 [[PREV_VALUE]], i24* [[TEMP1:%.+]],
// CHECK: [[PREV_VALUE:%.+]] = load i24, i24* [[LDTEMP]]
// CHECK: store i24 [[PREV_VALUE]], i24* [[TEMP:%.+]],
// CHECK: [[A_LD:%.+]] = load i24, i24* [[TEMP]],
// CHECK: [[A_SHL:%.+]] = shl i24 [[A_LD]], 7
// CHECK: [[A_ASHR:%.+]] = ashr i24 [[A_SHL]], 10
// CHECK: [[CAST:%.+]] = sext i24 [[A_ASHR]] to i32
@ -904,21 +773,15 @@ int main() {
// CHECK: [[ADD:%.+]] = fadd x86_fp80 [[X_RVAL]], [[EXPR]]
// CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 [[ADD]] to i32
// CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i24
// CHECK: [[BF_LD:%.+]] = load i24, i24* [[BITCAST]],
// CHECK: [[BF_LD:%.+]] = load i24, i24* [[TEMP1]],
// CHECK: [[BF_AND:%.+]] = and i24 [[TRUNC]], 16383
// CHECK: [[BF_VALUE:%.+]] = shl i24 [[BF_AND]], 3
// CHECK: [[BF_CLEAR:%.+]] = and i24 [[BF_LD]], -131065
// CHECK: or i24 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i24 %{{.+}}, i24* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i24, i24* [[LDTEMP]]
// CHECK: [[TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* %{{.+}} to i24*
// CHECK: store i24 [[OLD_BF_VALUE]], i24* [[TEMP_OLD_BF_ADDR]]
// CHECK: [[TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* %{{.+}} to i24*
// CHECK: store i24 [[NEW_BF_VALUE]], i24* [[TEMP_NEW_BF_ADDR]]
// CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i24* [[TEMP_OLD_BF_ADDR]] to i8*
// CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i24* [[TEMP_NEW_BF_ADDR]] to i8*
// CHECK: store i24 %{{.+}}, i24* [[TEMP1]]
// CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i24* [[LDTEMP]] to i8*
// CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i24* [[TEMP1]] to i8*
// CHECK: [[FAIL_SUCCESS:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 3, i8* getelementptr (i8, i8* bitcast (%struct.BitFields3_packed* @{{.+}} to i8*), i64 1), i8* [[BITCAST_TEMP_OLD_BF_ADDR]], i8* [[BITCAST_TEMP_NEW_BF_ADDR]], i32 0, i32 0)
// CHECK: [[FAILED_OLD_VAL]] = load i24, i24* [[TEMP_OLD_BF_ADDR]]
// CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic update
@ -928,6 +791,7 @@ int main() {
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: store i64 [[OLD_BF_VALUE]], i64* [[TEMP1:%.+]],
// CHECK: store i64 [[OLD_BF_VALUE]], i64* [[TEMP:%.+]],
// CHECK: [[A_LD:%.+]] = load i64, i64* [[TEMP]],
// CHECK: [[A_SHL:%.+]] = shl i64 [[A_LD]], 47
@ -937,13 +801,13 @@ int main() {
// CHECK: [[MUL:%.+]] = fmul x86_fp80 [[X_RVAL]], [[EXPR]]
// CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 [[MUL]] to i32
// CHECK: [[ZEXT:%.+]] = zext i32 [[NEW_VAL]] to i64
// CHECK: [[BF_LD:%.+]] = load i64, i64* [[TEMP]],
// CHECK: [[BF_LD:%.+]] = load i64, i64* [[TEMP1]],
// CHECK: [[BF_AND:%.+]] = and i64 [[ZEXT]], 1
// CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND]], 16
// CHECK: [[BF_CLEAR:%.+]] = and i64 [[BF_LD]], -65537
// CHECK: or i64 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i64 %{{.+}}, i64* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[LDTEMP]]
// CHECK: store i64 %{{.+}}, i64* [[TEMP1]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[TEMP1]]
// CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic
// CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
@ -956,6 +820,8 @@ int main() {
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %{{.+}} ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST1:%.+]] = bitcast i32* %{{.+}} to i8*
// CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST1]],
// CHECK: [[BITCAST:%.+]] = bitcast i32* %{{.+}} to i8*
// CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST]],
// CHECK: [[A_LD:%.+]] = load i8, i8* [[BITCAST]],
@ -966,12 +832,12 @@ int main() {
// CHECK: [[SUB: %.+]] = fsub x86_fp80 [[CONV]], [[EXPR]]
// CHECK: [[CONV:%.+]] = fptosi x86_fp80 [[SUB:%.+]] to i32
// CHECK: [[NEW_VAL:%.+]] = trunc i32 [[CONV]] to i8
// CHECK: [[BF_LD:%.+]] = load i8, i8* [[BITCAST]],
// CHECK: [[BF_LD:%.+]] = load i8, i8* [[BITCAST1]],
// CHECK: [[BF_VALUE:%.+]] = and i8 [[NEW_VAL]], 1
// CHECK: [[BF_CLEAR:%.+]] = and i8 [[BF_LD]], -2
// CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
// CHECK: store i8 %{{.+}}, i8* [[BITCAST1]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[BITCAST1]]
// CHECK: [[RES:%.+]] = cmpxchg i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2), i8 [[OLD_BF_VALUE]], i8 [[NEW_BF_VALUE]] monotonic monotonic
// CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
@ -984,6 +850,7 @@ int main() {
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: store i64 [[OLD_BF_VALUE]], i64* [[TEMP1:%.+]],
// CHECK: store i64 [[OLD_BF_VALUE]], i64* [[TEMP:%.+]],
// CHECK: [[A_LD:%.+]] = load i64, i64* [[TEMP]],
// CHECK: [[A_SHL:%.+]] = shl i64 [[A_LD]], 40
@ -991,13 +858,13 @@ int main() {
// CHECK: [[CONV:%.+]] = sitofp i64 [[A_ASHR]] to x86_fp80
// CHECK: [[DIV:%.+]] = fdiv x86_fp80 [[CONV]], [[EXPR]]
// CHECK: [[CONV:%.+]] = fptosi x86_fp80 [[DIV]] to i64
// CHECK: [[BF_LD:%.+]] = load i64, i64* [[TEMP]],
// CHECK: [[BF_LD:%.+]] = load i64, i64* [[TEMP1]],
// CHECK: [[BF_AND:%.+]] = and i64 [[CONV]], 127
// CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND:%.+]], 17
// CHECK: [[BF_CLEAR:%.+]] = and i64 [[BF_LD]], -16646145
// CHECK: [[VAL:%.+]] = or i64 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i64 [[VAL]], i64* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[LDTEMP]]
// CHECK: store i64 [[VAL]], i64* [[TEMP1]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[TEMP1]]
// CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic
// CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
@ -1010,6 +877,8 @@ int main() {
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST1:%.+]] = bitcast i64* %{{.+}} to i8*
// CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST1]],
// CHECK: [[BITCAST:%.+]] = bitcast i64* %{{.+}} to i8*
// CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST]],
// CHECK: [[A_LD:%.+]] = load i8, i8* [[BITCAST]],
@ -1019,13 +888,13 @@ int main() {
// CHECK: [[ADD:%.+]] = fadd x86_fp80 [[CONV]], [[EXPR]]
// CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 [[ADD]] to i64
// CHECK: [[TRUNC:%.+]] = trunc i64 [[NEW_VAL]] to i8
// CHECK: [[BF_LD:%.+]] = load i8, i8* [[BITCAST]],
// CHECK: [[BF_LD:%.+]] = load i8, i8* [[BITCAST1]],
// CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 127
// CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 1
// CHECK: [[BF_CLEAR:%.+]] = and i8 [[BF_LD]], 1
// CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
// CHECK: store i8 %{{.+}}, i8* [[BITCAST1]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[BITCAST1]]
// CHECK: [[RES:%.+]] = cmpxchg i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2), i8 [[OLD_BF_VALUE]], i8 [[NEW_BF_VALUE]] monotonic monotonic
// CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
@ -1036,24 +905,23 @@ int main() {
// CHECK: load i64, i64*
// CHECK: [[EXPR:%.+]] = uitofp i64 %{{.+}} to float
// CHECK: [[I64VAL:%.+]] = load atomic i64, i64* bitcast (<2 x float>* [[DEST:@.+]] to i64*) monotonic
// CHECK: [[LD:%.+]] = bitcast i64 [[I64VAL]] to <2 x float>
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_VEC_VAL:%.+]] = phi <2 x float> [ [[LD]], %{{.+}} ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[OLD_I64:%.+]] = phi i64 [ [[I64VAL]], %{{.+}} ], [ [[FAILED_I64_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast <2 x float>* [[TEMP:%.+]] to i64*
// CHECK: store i64 [[OLD_I64]], i64* [[BITCAST]],
// CHECK: [[OLD_VEC_VAL:%.+]] = bitcast i64 [[OLD_I64]] to <2 x float>
// CHECK: store <2 x float> [[OLD_VEC_VAL]], <2 x float>* [[LDTEMP:%.+]],
// CHECK: [[VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]]
// CHECK: [[X:%.+]] = extractelement <2 x float> [[VEC_VAL]], i64 0
// CHECK: [[VEC_ITEM_VAL:%.+]] = fsub float [[EXPR]], [[X]]
// CHECK: [[VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]],
// CHECK: [[VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[TEMP]],
// CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <2 x float> [[VEC_VAL]], float [[VEC_ITEM_VAL]], i64 0
// CHECK: store <2 x float> [[NEW_VEC_VAL]], <2 x float>* [[LDTEMP]]
// CHECK: [[NEW_VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]]
// CHECK: [[OLD_I64:%.+]] = bitcast <2 x float> [[OLD_VEC_VAL]] to i64
// CHECK: [[NEW_I64:%.+]] = bitcast <2 x float> [[NEW_VEC_VAL]] to i64
// CHECK: store <2 x float> [[NEW_VEC_VAL]], <2 x float>* [[TEMP]]
// CHECK: [[NEW_I64:%.+]] = load i64, i64* [[BITCAST]]
// CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (<2 x float>* [[DEST]] to i64*), i64 [[OLD_I64]], i64 [[NEW_I64]] monotonic monotonic
// CHECK: [[FAILED_I64_OLD_VAL:%.+]] = extractvalue { i64, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
// CHECK: [[FAILED_OLD_VAL]] = bitcast i64 [[FAILED_I64_OLD_VAL]] to <2 x float>
// CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic

55
clang/test/OpenMP/atomic_write_codegen.c
View File

@ -265,21 +265,18 @@ int main() {
// CHECK-DAG: load i8, i8*
// CHECK-DAG: [[VEC_ITEM_VAL:%.+]] = zext i1 %{{.+}} to i32
// CHECK: [[I128VAL:%.+]] = load atomic i128, i128* bitcast (<4 x i32>* [[DEST:@.+]] to i128*) monotonic
// CHECK: [[LD:%.+]] = bitcast i128 [[I128VAL]] to <4 x i32>
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_VEC_VAL:%.+]] = phi <4 x i32> [ [[LD]], %{{.+}} ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: store <4 x i32> [[OLD_VEC_VAL]], <4 x i32>* [[LDTEMP:%.+]],
// CHECK: [[OLD_I128:%.+]] = phi i128 [ [[I128VAL]], %{{.+}} ], [ [[FAILED_I128_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast <4 x i32>* [[LDTEMP:%.+]] to i128*
// CHECK: store i128 [[OLD_I128]], i128* [[BITCAST]],
// CHECK: [[VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[LDTEMP]]
// CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <4 x i32> [[VEC_VAL]], i32 [[VEC_ITEM_VAL]], i16 [[IDX]]
// CHECK: store <4 x i32> [[NEW_VEC_VAL]], <4 x i32>* [[LDTEMP]]
// CHECK: [[NEW_VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[LDTEMP]]
// CHECK: [[OLD_I128:%.+]] = bitcast <4 x i32> [[OLD_VEC_VAL]] to i128
// CHECK: [[NEW_I128:%.+]] = bitcast <4 x i32> [[NEW_VEC_VAL]] to i128
// CHECK: [[NEW_I128:%.+]] = load i128, i128* [[BITCAST]]
// CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (<4 x i32>* [[DEST]] to i128*), i128 [[OLD_I128]], i128 [[NEW_I128]] monotonic monotonic
// CHECK: [[FAILED_I128_OLD_VAL:%.+]] = extractvalue { i128, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i128, i1 } [[RES]], 1
// CHECK: [[FAILED_OLD_VAL]] = bitcast i128 [[FAILED_I128_OLD_VAL]] to <4 x i32>
// CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic write
@ -306,21 +303,18 @@ int main() {
// CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
// CHECK: [[BITCAST:%.+]] = bitcast i32* [[LDTEMP:%.+]] to i8*
// CHECK: call void @__atomic_load(i64 4, i8* getelementptr (i8, i8* bitcast (%struct.BitFields_packed* @{{.+}} to i8*), i64 4), i8* [[BITCAST]], i32 0)
// CHECK: [[PREV_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[OLD_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]],
// CHECK: store i32 [[OLD_BF_VALUE]], i32* [[LDTEMP1:%.+]],
// CHECK: [[OLD_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP1]],
// CHECK: [[BF_VALUE:%.+]] = and i32 [[NEW_VAL]], 2147483647
// CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -2147483648
// CHECK: [[BF_CLEAR:%.+]] = and i32 [[OLD_BF_VALUE]], -2147483648
// CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
// CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP_OLD_BF_ADDR:%.+]],
// CHECK: store i32 [[NEW_BF_VALUE]], i32* [[TEMP_NEW_BF_ADDR:%.+]],
// CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* [[TEMP_OLD_BF_ADDR]] to i8*
// CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* [[TEMP_NEW_BF_ADDR]] to i8*
// CHECK: store i32 %{{.+}}, i32* [[LDTEMP1]]
// CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* [[LDTEMP]] to i8*
// CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* [[LDTEMP1]] to i8*
// CHECK: [[FAIL_SUCCESS:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 4, i8* getelementptr (i8, i8* bitcast (%struct.BitFields_packed* @{{.+}} to i8*), i64 4), i8* [[BITCAST_TEMP_OLD_BF_ADDR]], i8* [[BITCAST_TEMP_NEW_BF_ADDR]], i32 0, i32 0)
// CHECK: [[FAILED_OLD_VAL]] = load i32, i32* [[TEMP_OLD_BF_ADDR]]
// CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic write
@ -388,25 +382,19 @@ int main() {
// CHECK: [[LDTEMP:%.+]] = bitcast i32* %{{.+}} to i24*
// CHECK: [[BITCAST:%.+]] = bitcast i24* %{{.+}} to i8*
// CHECK: call void @__atomic_load(i64 3, i8* getelementptr (i8, i8* bitcast (%struct.BitFields3_packed* @{{.+}} to i8*), i64 1), i8* [[BITCAST]], i32 0)
// CHECK: [[PREV_VALUE:%.+]] = load i24, i24* [[LDTEMP]]
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_BF_VALUE:%.+]] = phi i24 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[OLD_VAL:%.+]] = load i24, i24* %{{.+}},
// CHECK: store i24 [[OLD_VAL]], i24* [[TEMP:%.+]],
// CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i24
// CHECK: [[BF_AND:%.+]] = and i24 [[TRUNC]], 16383
// CHECK: [[BF_VALUE:%.+]] = shl i24 [[BF_AND]], 3
// CHECK: [[BF_CLEAR:%.+]] = and i24 %{{.+}}, -131065
// CHECK: or i24 [[BF_CLEAR]], [[BF_VALUE]]
// CHECK: store i24 %{{.+}}, i24* [[LDTEMP:%.+]]
// CHECK: [[NEW_BF_VALUE:%.+]] = load i24, i24* [[LDTEMP]]
// CHECK: [[TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* %{{.+}} to i24*
// CHECK: store i24 [[OLD_BF_VALUE]], i24* [[TEMP_OLD_BF_ADDR]]
// CHECK: [[TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* %{{.+}} to i24*
// CHECK: store i24 [[NEW_BF_VALUE]], i24* [[TEMP_NEW_BF_ADDR]]
// CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i24* [[TEMP_OLD_BF_ADDR]] to i8*
// CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i24* [[TEMP_NEW_BF_ADDR]] to i8*
// CHECK: store i24 %{{.+}}, i24* [[TEMP]]
// CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i24* [[LDTEMP]] to i8*
// CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i24* [[TEMP]] to i8*
// CHECK: [[FAIL_SUCCESS:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 3, i8* getelementptr (i8, i8* bitcast (%struct.BitFields3_packed* @{{.+}} to i8*), i64 1), i8* [[BITCAST_TEMP_OLD_BF_ADDR]], i8* [[BITCAST_TEMP_NEW_BF_ADDR]], i32 0, i32 0)
// CHECK: [[FAILED_OLD_VAL]] = load i24, i24* [[TEMP_OLD_BF_ADDR]]
// CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic write
@ -492,21 +480,18 @@ int main() {
// CHECK: load i64, i64*
// CHECK: [[VEC_ITEM_VAL:%.+]] = uitofp i64 %{{.+}} to float
// CHECK: [[I64VAL:%.+]] = load atomic i64, i64* bitcast (<2 x float>* [[DEST:@.+]] to i64*) monotonic
// CHECK: [[LD:%.+]] = bitcast i64 [[I64VAL]] to <2 x float>
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[OLD_VEC_VAL:%.+]] = phi <2 x float> [ [[LD]], %{{.+}} ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: store <2 x float> [[OLD_VEC_VAL]], <2 x float>* [[LDTEMP:%.+]],
// CHECK: [[OLD_I64:%.+]] = phi i64 [ [[I64VAL]], %{{.+}} ], [ [[FAILED_I64_OLD_VAL:%.+]], %[[CONT]] ]
// CHECK: [[BITCAST:%.+]] = bitcast <2 x float>* [[LDTEMP:%.+]] to i64*
// CHECK: store i64 [[OLD_I64]], i64* [[BITCAST]],
// CHECK: [[VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]]
// CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <2 x float> [[VEC_VAL]], float [[VEC_ITEM_VAL]], i64 0
// CHECK: store <2 x float> [[NEW_VEC_VAL]], <2 x float>* [[LDTEMP]]
// CHECK: [[NEW_VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]]
// CHECK: [[OLD_I64:%.+]] = bitcast <2 x float> [[OLD_VEC_VAL]] to i64
// CHECK: [[NEW_I64:%.+]] = bitcast <2 x float> [[NEW_VEC_VAL]] to i64
// CHECK: [[NEW_I64:%.+]] = load i64, i64* [[BITCAST]]
// CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (<2 x float>* [[DEST]] to i64*), i64 [[OLD_I64]], i64 [[NEW_I64]] monotonic monotonic
// CHECK: [[FAILED_I64_OLD_VAL:%.+]] = extractvalue { i64, i1 } [[RES]], 0
// CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
// CHECK: [[FAILED_OLD_VAL]] = bitcast i64 [[FAILED_I64_OLD_VAL]] to <2 x float>
// CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
// CHECK: [[EXIT]]
#pragma omp atomic write

26
clang/test/OpenMP/for_reduction_codegen.cpp
View File

@ -306,19 +306,16 @@ int main() {
// t_var += t_var_reduction;
// CHECK: load float, float* [[T_VAR_PRIV]]
// CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32*
// CHECK: load atomic i32, i32* [[T_VAR_REF_INT]] monotonic,
// CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float
// CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR_REF_INT]] monotonic,
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ]
// CHECK: [[UP:%.+]] = fadd float
// CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32
// CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32
// CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ]
// CHECK: fadd float
// CHECK: [[UP_INT:%.+]] = load i32, i32*
// CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32*
// CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic
// CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
// CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]]
// CHECK: [[ATOMIC_DONE]]
@ -351,21 +348,18 @@ int main() {
// t_var1 = min(t_var1, t_var1_reduction);
// CHECK: load float, float* [[T_VAR1_PRIV]]
// CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32*
// CHECK: load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic,
// CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float
// CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic,
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %{{.+}} ]
// CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %{{.+}} ]
// CHECK: [[CMP:%.+]] = fcmp olt float
// CHECK: br i1 [[CMP]]
// CHECK: [[UP:%.+]] = phi float
// CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32
// CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32
// CHECK: phi float
// CHECK: [[UP_INT:%.+]] = load i32
// CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32*
// CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR1_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic
// CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
// CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]]
// CHECK: [[ATOMIC_DONE]]

24
clang/test/OpenMP/parallel_reduction_codegen.cpp
View File

@ -298,19 +298,16 @@ int main() {
// t_var += t_var_reduction;
// CHECK: load float, float* [[T_VAR_PRIV]]
// CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32*
// CHECK: load atomic i32, i32* [[T_VAR_REF_INT]] monotonic,
// CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float
// CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR_REF_INT]] monotonic,
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ]
// CHECK: [[UP:%.+]] = fadd float
// CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32
// CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32
// CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %[[CONT]] ]
// CHECK: fadd float
// CHECK: [[UP_INT:%.+]] = load i32
// CHECK: [[T_VAR_REF_INT:%.+]] = bitcast float* [[T_VAR_REF]] to i32*
// CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic
// CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
// CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]]
// CHECK: [[ATOMIC_DONE]]
@ -343,21 +340,18 @@ int main() {
// t_var1 = min(t_var1, t_var1_reduction);
// CHECK: load float, float* [[T_VAR1_PRIV]]
// CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32*
// CHECK: load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic,
// CHECK: [[OLD1:%.+]] = bitcast i32 %{{.+}} to float
// CHECK: [[OLD1:%.+]] = load atomic i32, i32* [[T_VAR1_REF_INT]] monotonic,
// CHECK: br label %[[CONT:.+]]
// CHECK: [[CONT]]
// CHECK: [[ORIG_OLD:%.+]] = phi float [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %{{.+}} ]
// CHECK: [[ORIG_OLD_INT:%.+]] = phi i32 [ [[OLD1]], %{{.+}} ], [ [[OLD2:%.+]], %{{.+}} ]
// CHECK: [[CMP:%.+]] = fcmp olt float
// CHECK: br i1 [[CMP]]
// CHECK: [[UP:%.+]] = phi float
// CHECK: [[ORIG_OLD_INT:%.+]] = bitcast float [[ORIG_OLD]] to i32
// CHECK: [[UP_INT:%.+]] = bitcast float [[UP]] to i32
// CHECK: [[UP_INT:%.+]] = load i32
// CHECK: [[T_VAR1_REF_INT:%.+]] = bitcast float* [[T_VAR1_REF]] to i32*
// CHECK: [[RES:%.+]] = cmpxchg i32* [[T_VAR1_REF_INT]], i32 [[ORIG_OLD_INT]], i32 [[UP_INT]] monotonic monotonic
// CHECK: [[OLD_VAL_INT:%.+]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[OLD2:%.+]] = extractvalue { i32, i1 } [[RES]], 0
// CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
// CHECK: [[OLD2]] = bitcast i32 [[OLD_VAL_INT]] to float
// CHECK: br i1 [[SUCCESS_FAIL]], label %[[ATOMIC_DONE:.+]], label %[[CONT]]
// CHECK: [[ATOMIC_DONE]]