Skip to content

[mlir][llvm][OpenMP] Hoist __atomic_load alloca #132888

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Apr 9, 2025
Merged

[mlir][llvm][OpenMP] Hoist __atomic_load alloca #132888

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
629450c
[mlir][llvm][OpenMP] Hoist __atomic_load alloca
NimishMishra Apr 1, 2025
2507e02
Fix unittest
NimishMishra Apr 7, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion flang/test/Integration/OpenMP/atomic-capture-complex.f90
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,14 +9,14 @@
!RUN: %if x86-registered-target %{ %flang_fc1 -triple x86_64-unknown-linux-gnu -emit-llvm -fopenmp %s -o - | FileCheck --check-prefixes=CHECK,X86 %s %}
!RUN: %if aarch64-registerd-target %{ %flang_fc1 -triple aarch64-unknown-linux-gnu -emit-llvm -fopenmp %s -o - | FileCheck --check-prefixes=CHECK,AARCH64 %s %}

!CHECK: %[[ATOMIC_TEMP_LOAD:.*]] = alloca { float, float }, align 8
!CHECK: %[[X_NEW_VAL:.*]] = alloca { float, float }, align 8
!CHECK: %[[VAL_1:.*]] = alloca { float, float }, i64 1, align 8
!CHECK: %[[ORIG_VAL:.*]] = alloca { float, float }, i64 1, align 8
!CHECK: store { float, float } { float 2.000000e+00, float 2.000000e+00 }, ptr %[[ORIG_VAL]], align 4
!CHECK: br label %entry

!CHECK: entry:
!CHECK: %[[ATOMIC_TEMP_LOAD:.*]] = alloca { float, float }, align 8
!CHECK: call void @__atomic_load(i64 8, ptr %[[ORIG_VAL]], ptr %[[ATOMIC_TEMP_LOAD]], i32 0)
!CHECK: %[[PHI_NODE_ENTRY_1:.*]] = load { float, float }, ptr %[[ATOMIC_TEMP_LOAD]], align 8
!CHECK: br label %.atomic.cont
Expand Down
5 changes: 3 additions & 2 deletions llvm/include/llvm/Frontend/Atomic/Atomic.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,14 +22,15 @@ class AtomicInfo {
Align AtomicAlign;
Align ValueAlign;
bool UseLibcall;
IRBuilderBase::InsertPoint AllocaIP;

public:
AtomicInfo(IRBuilderBase *Builder, Type *Ty, uint64_t AtomicSizeInBits,
uint64_t ValueSizeInBits, Align AtomicAlign, Align ValueAlign,
bool UseLibcall)
bool UseLibcall, IRBuilderBase::InsertPoint AllocaIP)
: Builder(Builder), Ty(Ty), AtomicSizeInBits(AtomicSizeInBits),
ValueSizeInBits(ValueSizeInBits), AtomicAlign(AtomicAlign),
ValueAlign(ValueAlign), UseLibcall(UseLibcall) {}
ValueAlign(ValueAlign), UseLibcall(UseLibcall), AllocaIP(AllocaIP) {}

virtual ~AtomicInfo() = default;

Expand Down
10 changes: 6 additions & 4 deletions llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -489,9 +489,10 @@ class OpenMPIRBuilder {
public:
AtomicInfo(IRBuilder<> *Builder, llvm::Type *Ty, uint64_t AtomicSizeInBits,
uint64_t ValueSizeInBits, llvm::Align AtomicAlign,
llvm::Align ValueAlign, bool UseLibcall, llvm::Value *AtomicVar)
llvm::Align ValueAlign, bool UseLibcall,
IRBuilderBase::InsertPoint AllocaIP, llvm::Value *AtomicVar)
: llvm::AtomicInfo(Builder, Ty, AtomicSizeInBits, ValueSizeInBits,
AtomicAlign, ValueAlign, UseLibcall),
AtomicAlign, ValueAlign, UseLibcall, AllocaIP),
AtomicVar(AtomicVar) {}

llvm::Value *getAtomicPointer() const override { return AtomicVar; }
Expand Down Expand Up @@ -3268,11 +3269,12 @@ class OpenMPIRBuilder {
/// value
/// \param AO Atomic ordering of the generated atomic
/// instructions.
///
/// \param AllocaIP Insert point for allocas
//
/// \return Insertion point after generated atomic read IR.
InsertPointTy createAtomicRead(const LocationDescription &Loc,
AtomicOpValue &X, AtomicOpValue &V,
AtomicOrdering AO);
AtomicOrdering AO, InsertPointTy AllocaIP);

/// Emit atomic write for : X = Expr --- Only Scalar data types.
///
Expand Down
4 changes: 4 additions & 0 deletions llvm/lib/Frontend/Atomic/Atomic.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -118,8 +118,12 @@ AtomicInfo::EmitAtomicLoadLibcall(AtomicOrdering AO) {
Value *PtrVal = getAtomicPointer();
PtrVal = Builder->CreateAddrSpaceCast(PtrVal, PointerType::getUnqual(Ctx));
Args.push_back(PtrVal);

auto CurrentIP = Builder->saveIP();
Builder->restoreIP(AllocaIP);
AllocaInst *AllocaResult =
CreateAlloca(Ty, getAtomicPointer()->getName() + "atomic.temp.load");
Builder->restoreIP(CurrentIP);
const Align AllocaAlignment = DL.getPrefTypeAlign(SizedIntTy);
AllocaResult->setAlignment(AllocaAlignment);
Args.push_back(AllocaResult);
Expand Down
6 changes: 3 additions & 3 deletions llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8565,7 +8565,7 @@ bool OpenMPIRBuilder::checkAndEmitFlushAfterAtomic(
OpenMPIRBuilder::InsertPointTy
OpenMPIRBuilder::createAtomicRead(const LocationDescription &Loc,
AtomicOpValue &X, AtomicOpValue &V,
AtomicOrdering AO) {
AtomicOrdering AO, InsertPointTy AllocaIP) {
if (!updateToLocation(Loc))
return Loc.IP;

Expand Down Expand Up @@ -8593,7 +8593,7 @@ OpenMPIRBuilder::createAtomicRead(const LocationDescription &Loc,
LoadDL.getTypeStoreSize(OldVal->getPointerOperand()->getType());
OpenMPIRBuilder::AtomicInfo atomicInfo(
&Builder, XElemTy, LoadSize * 8, LoadSize * 8, OldVal->getAlign(),
OldVal->getAlign(), true /* UseLibcall */, X.Var);
OldVal->getAlign(), true /* UseLibcall */, AllocaIP, X.Var);
auto AtomicLoadRes = atomicInfo.EmitAtomicLoadLibcall(AO);
XRead = AtomicLoadRes.first;
OldVal->eraseFromParent();
Expand Down Expand Up @@ -8758,7 +8758,7 @@ Expected<std::pair<Value *, Value *>> OpenMPIRBuilder::emitAtomicUpdate(

OpenMPIRBuilder::AtomicInfo atomicInfo(
&Builder, XElemTy, LoadSize * 8, LoadSize * 8, OldVal->getAlign(),
OldVal->getAlign(), true /* UseLibcall */, X);
OldVal->getAlign(), true /* UseLibcall */, AllocaIP, X);
auto AtomicLoadRes = atomicInfo.EmitAtomicLoadLibcall(AO);
BasicBlock *CurBB = Builder.GetInsertBlock();
Instruction *CurBBTI = CurBB->getTerminator();
Expand Down
4 changes: 3 additions & 1 deletion mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2675,6 +2675,8 @@ convertOmpAtomicRead(Operation &opInst, llvm::IRBuilderBase &builder,
return failure();

llvm::OpenMPIRBuilder *ompBuilder = moduleTranslation.getOpenMPBuilder();
llvm::OpenMPIRBuilder::InsertPointTy allocaIP =
findAllocaInsertPoint(builder, moduleTranslation);

llvm::OpenMPIRBuilder::LocationDescription ompLoc(builder);

Expand All @@ -2687,7 +2689,7 @@ convertOmpAtomicRead(Operation &opInst, llvm::IRBuilderBase &builder,

llvm::OpenMPIRBuilder::AtomicOpValue V = {v, elementType, false, false};
llvm::OpenMPIRBuilder::AtomicOpValue X = {x, elementType, false, false};
builder.restoreIP(ompBuilder->createAtomicRead(ompLoc, X, V, AO));
builder.restoreIP(ompBuilder->createAtomicRead(ompLoc, X, V, AO, allocaIP));
return success();
}

Expand Down
9 changes: 5 additions & 4 deletions mlir/test/Target/LLVMIR/openmp-llvm.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1368,6 +1368,7 @@ llvm.func @omp_atomic_read(%arg0 : !llvm.ptr, %arg1 : !llvm.ptr) -> () {

// CHECK-LABEL: @omp_atomic_read_implicit_cast
llvm.func @omp_atomic_read_implicit_cast () {
//CHECK: %[[ATOMIC_LOAD_TEMP:.*]] = alloca { float, float }, align 8
//CHECK: %[[Z:.*]] = alloca float, i64 1, align 4
//CHECK: %[[Y:.*]] = alloca double, i64 1, align 8
//CHECK: %[[X:.*]] = alloca [2 x { float, float }], i64 1, align 8
Expand All @@ -1392,7 +1393,7 @@ llvm.func @omp_atomic_read_implicit_cast () {
%16 = llvm.mul %10, %9 overflow<nsw> : i64
%17 = llvm.getelementptr %5[%15] : (!llvm.ptr, i64) -> !llvm.ptr, !llvm.struct<(f32, f32)>

//CHECK: %[[ATOMIC_LOAD_TEMP:.*]] = alloca { float, float }, align 8

//CHECK: call void @__atomic_load(i64 8, ptr %[[X_ELEMENT]], ptr %[[ATOMIC_LOAD_TEMP]], i32 0)
//CHECK: %[[LOAD:.*]] = load { float, float }, ptr %[[ATOMIC_LOAD_TEMP]], align 8
//CHECK: %[[EXT:.*]] = extractvalue { float, float } %[[LOAD]], 0
Expand Down Expand Up @@ -1480,14 +1481,14 @@ llvm.func @omp_atomic_update(%x:!llvm.ptr, %expr: i32, %xbool: !llvm.ptr, %exprb

// -----

//CHECK: %[[ATOMIC_TEMP_LOAD:.*]] = alloca { float, float }, align 8
//CHECK: %[[X_NEW_VAL:.*]] = alloca { float, float }, align 8
//CHECK: {{.*}} = alloca { float, float }, i64 1, align 8
//CHECK: %[[ORIG_VAL:.*]] = alloca { float, float }, i64 1, align 8

//CHECK: br label %entry

//CHECK: entry:
//CHECK: %[[ATOMIC_TEMP_LOAD:.*]] = alloca { float, float }, align 8
//CHECK: call void @__atomic_load(i64 8, ptr %[[ORIG_VAL]], ptr %[[ATOMIC_TEMP_LOAD]], i32 0)
//CHECK: %[[PHI_NODE_ENTRY_1:.*]] = load { float, float }, ptr %[[ATOMIC_TEMP_LOAD]], align 8
//CHECK: br label %.atomic.cont
Expand Down Expand Up @@ -1532,14 +1533,14 @@ llvm.func @_QPomp_atomic_update_complex() {

// -----

//CHECK: %[[ATOMIC_TEMP_LOAD:.*]] = alloca { float, float }, align 8
//CHECK: %[[X_NEW_VAL:.*]] = alloca { float, float }, align 8
//CHECK: %[[VAL_1:.*]] = alloca { float, float }, i64 1, align 8
//CHECK: %[[ORIG_VAL:.*]] = alloca { float, float }, i64 1, align 8
//CHECK: store { float, float } { float 2.000000e+00, float 2.000000e+00 }, ptr %[[ORIG_VAL]], align 4
//CHECK: br label %entry

//CHECK: entry: ; preds = %0
//CHECK: %[[ATOMIC_TEMP_LOAD:.*]] = alloca { float, float }, align 8
//CHECK: call void @__atomic_load(i64 8, ptr %[[ORIG_VAL]], ptr %[[ATOMIC_TEMP_LOAD]], i32 0)
//CHECK: %[[PHI_NODE_ENTRY_1:.*]] = load { float, float }, ptr %[[ATOMIC_TEMP_LOAD]], align 8
//CHECK: br label %.atomic.cont
Expand Down Expand Up @@ -1597,9 +1598,9 @@ llvm.func @_QPomp_atomic_capture_complex() {
// CHECK-LABEL: define void @omp_atomic_read_complex() {
llvm.func @omp_atomic_read_complex(){

// CHECK: %[[ATOMIC_TEMP_LOAD:.*]] = alloca { float, float }, align 8
// CHECK: %[[a:.*]] = alloca { float, float }, i64 1, align 8
// CHECK: %[[b:.*]] = alloca { float, float }, i64 1, align 8
// CHECK: %[[ATOMIC_TEMP_LOAD:.*]] = alloca { float, float }, align 8
// CHECK: call void @__atomic_load(i64 8, ptr %[[b]], ptr %[[ATOMIC_TEMP_LOAD]], i32 0)
// CHECK: %[[LOADED_VAL:.*]] = load { float, float }, ptr %[[ATOMIC_TEMP_LOAD]], align 8
// CHECK: store { float, float } %[[LOADED_VAL]], ptr %[[a]], align 4
Expand Down
Loading