Skip to content

[Clang] Fix clang to emit llvm-ir for fadd/fsub atomics #162679

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

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

[Clang] Fix clang to emit llvm-ir for fadd/fsub atomics #162679

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e654aa7
[LLVM] Fix clang to emit llvm-ir for fadd/fsub atomics
Amichaxx Oct 6, 2025
653b974
clang format
Amichaxx Oct 9, 2025
52d5596
Comments addressing
Amichaxx Oct 22, 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
21 changes: 16 additions & 5 deletions clang/lib/CodeGen/CGExprScalar.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3847,7 +3847,16 @@ LValue ScalarExprEmitter::EmitCompoundAssignLValue(
llvm::PHINode *atomicPHI = nullptr;
if (const AtomicType *atomicTy = LHSTy->getAs<AtomicType>()) {
QualType type = atomicTy->getValueType();
if (!type->isBooleanType() && type->isIntegerType() &&
const bool isFloat = type->isFloatingType();
const bool isInteger = type->isIntegerType();

bool isPowerOfTwo = false;
if (isFloat || isInteger) {
llvm::Type *IRTy = CGF.ConvertType(type);
uint64_t StoreBits = CGF.CGM.getDataLayout().getTypeStoreSizeInBits(IRTy);
isPowerOfTwo = llvm::isPowerOf2_64(StoreBits);
}
if (!type->isBooleanType() && (isInteger || (isPowerOfTwo && isFloat)) &&
!(type->isUnsignedIntegerType() &&
CGF.SanOpts.has(SanitizerKind::UnsignedIntegerOverflow)) &&
CGF.getLangOpts().getSignedOverflowBehavior() !=
Expand All @@ -3862,12 +3871,14 @@ LValue ScalarExprEmitter::EmitCompoundAssignLValue(
case BO_ShrAssign:
break;
case BO_AddAssign:
AtomicOp = llvm::AtomicRMWInst::Add;
Op = llvm::Instruction::Add;
AtomicOp =
isFloat ? llvm::AtomicRMWInst::FAdd : llvm::AtomicRMWInst::Add;
Op = isFloat ? llvm::Instruction::FAdd : llvm::Instruction::Add;
break;
case BO_SubAssign:
AtomicOp = llvm::AtomicRMWInst::Sub;
Op = llvm::Instruction::Sub;
AtomicOp =
isFloat ? llvm::AtomicRMWInst::FSub : llvm::AtomicRMWInst::Sub;
Op = isFloat ? llvm::Instruction::FSub : llvm::Instruction::Sub;
break;
case BO_AndAssign:
AtomicOp = llvm::AtomicRMWInst::And;
Expand Down
123 changes: 123 additions & 0 deletions clang/test/CodeGen/aarch64-lsfe-atomics.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,123 @@
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 6
// RUN: %clang_cc1 %s -emit-llvm -o - -triple=aarch64-linux-gnu | FileCheck %s --check-prefix=CHECK-LLVM

_Atomic(float) f;
_Atomic(double) d;
_Atomic(__bf16) bf;
_Atomic(_Float16) h;

// CHECK-LLVM-LABEL: define dso_local void @test_float_add(
// CHECK-LLVM-SAME: float noundef [[VAL:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK-LLVM-NEXT: [[ENTRY:.*:]]
// CHECK-LLVM-NEXT: [[VAL_ADDR:%.*]] = alloca float, align 4
// CHECK-LLVM-NEXT: store float [[VAL]], ptr [[VAL_ADDR]], align 4
// CHECK-LLVM-NEXT: [[TMP0:%.*]] = load float, ptr [[VAL_ADDR]], align 4
// CHECK-LLVM-NEXT: [[TMP1:%.*]] = atomicrmw fadd ptr @f, float [[TMP0]] seq_cst, align 4
// CHECK-LLVM-NEXT: [[TMP2:%.*]] = fadd float [[TMP1]], [[TMP0]]
// CHECK-LLVM-NEXT: ret void
//
void test_float_add(float val) {
f += val;
}

// CHECK-LLVM-LABEL: define dso_local void @test_double_add(
// CHECK-LLVM-SAME: double noundef [[VAL:%.*]]) #[[ATTR0]] {
// CHECK-LLVM-NEXT: [[ENTRY:.*:]]
// CHECK-LLVM-NEXT: [[VAL_ADDR:%.*]] = alloca double, align 8
// CHECK-LLVM-NEXT: store double [[VAL]], ptr [[VAL_ADDR]], align 8
// CHECK-LLVM-NEXT: [[TMP0:%.*]] = load double, ptr [[VAL_ADDR]], align 8
// CHECK-LLVM-NEXT: [[TMP1:%.*]] = atomicrmw fadd ptr @d, double [[TMP0]] seq_cst, align 8
// CHECK-LLVM-NEXT: [[TMP2:%.*]] = fadd double [[TMP1]], [[TMP0]]
// CHECK-LLVM-NEXT: ret void
//
void test_double_add(double val) {
d += val;
}

// CHECK-LLVM-LABEL: define dso_local void @test_bf16_add(
// CHECK-LLVM-SAME: bfloat noundef [[VAL:%.*]]) #[[ATTR0]] {
// CHECK-LLVM-NEXT: [[ENTRY:.*:]]
// CHECK-LLVM-NEXT: [[VAL_ADDR:%.*]] = alloca bfloat, align 2
// CHECK-LLVM-NEXT: store bfloat [[VAL]], ptr [[VAL_ADDR]], align 2
// CHECK-LLVM-NEXT: [[TMP0:%.*]] = load bfloat, ptr [[VAL_ADDR]], align 2
// CHECK-LLVM-NEXT: [[EXT:%.*]] = fpext bfloat [[TMP0]] to float
// CHECK-LLVM-NEXT: [[CONV:%.*]] = fptrunc float [[EXT]] to bfloat
// CHECK-LLVM-NEXT: [[TMP1:%.*]] = atomicrmw fadd ptr @bf, bfloat [[CONV]] seq_cst, align 2
// CHECK-LLVM-NEXT: [[TMP2:%.*]] = fadd bfloat [[TMP1]], [[CONV]]
// CHECK-LLVM-NEXT: ret void
//
void test_bf16_add(__bf16 val){
bf += val;
}

// CHECK-LLVM-LABEL: define dso_local void @test_f16_add(
// CHECK-LLVM-SAME: half noundef [[VAL:%.*]]) #[[ATTR0]] {
// CHECK-LLVM-NEXT: [[ENTRY:.*:]]
// CHECK-LLVM-NEXT: [[VAL_ADDR:%.*]] = alloca half, align 2
// CHECK-LLVM-NEXT: store half [[VAL]], ptr [[VAL_ADDR]], align 2
// CHECK-LLVM-NEXT: [[TMP0:%.*]] = load half, ptr [[VAL_ADDR]], align 2
// CHECK-LLVM-NEXT: [[TMP1:%.*]] = atomicrmw fadd ptr @h, half [[TMP0]] seq_cst, align 2
// CHECK-LLVM-NEXT: [[TMP2:%.*]] = fadd half [[TMP1]], [[TMP0]]
// CHECK-LLVM-NEXT: ret void
//
void test_f16_add(_Float16 val){
h += val;
}

// CHECK-LLVM-LABEL: define dso_local void @test_float_sub(
// CHECK-LLVM-SAME: float noundef [[VAL:%.*]]) #[[ATTR0]] {
// CHECK-LLVM-NEXT: [[ENTRY:.*:]]
// CHECK-LLVM-NEXT: [[VAL_ADDR:%.*]] = alloca float, align 4
// CHECK-LLVM-NEXT: store float [[VAL]], ptr [[VAL_ADDR]], align 4
// CHECK-LLVM-NEXT: [[TMP0:%.*]] = load float, ptr [[VAL_ADDR]], align 4
// CHECK-LLVM-NEXT: [[TMP1:%.*]] = atomicrmw fsub ptr @f, float [[TMP0]] seq_cst, align 4
// CHECK-LLVM-NEXT: [[TMP2:%.*]] = fsub float [[TMP1]], [[TMP0]]
// CHECK-LLVM-NEXT: ret void
//
void test_float_sub(float val) {
f -= val;
}

// CHECK-LLVM-LABEL: define dso_local void @test_double_sub(
// CHECK-LLVM-SAME: double noundef [[VAL:%.*]]) #[[ATTR0]] {
// CHECK-LLVM-NEXT: [[ENTRY:.*:]]
// CHECK-LLVM-NEXT: [[VAL_ADDR:%.*]] = alloca double, align 8
// CHECK-LLVM-NEXT: store double [[VAL]], ptr [[VAL_ADDR]], align 8
// CHECK-LLVM-NEXT: [[TMP0:%.*]] = load double, ptr [[VAL_ADDR]], align 8
// CHECK-LLVM-NEXT: [[TMP1:%.*]] = atomicrmw fsub ptr @d, double [[TMP0]] seq_cst, align 8
// CHECK-LLVM-NEXT: [[TMP2:%.*]] = fsub double [[TMP1]], [[TMP0]]
// CHECK-LLVM-NEXT: ret void
//
void test_double_sub(double val){
d -= val;
}

// CHECK-LLVM-LABEL: define dso_local void @test_bf16_sub(
// CHECK-LLVM-SAME: bfloat noundef [[VAL:%.*]]) #[[ATTR0]] {
// CHECK-LLVM-NEXT: [[ENTRY:.*:]]
// CHECK-LLVM-NEXT: [[VAL_ADDR:%.*]] = alloca bfloat, align 2
// CHECK-LLVM-NEXT: store bfloat [[VAL]], ptr [[VAL_ADDR]], align 2
// CHECK-LLVM-NEXT: [[TMP0:%.*]] = load bfloat, ptr [[VAL_ADDR]], align 2
// CHECK-LLVM-NEXT: [[EXT:%.*]] = fpext bfloat [[TMP0]] to float
// CHECK-LLVM-NEXT: [[CONV:%.*]] = fptrunc float [[EXT]] to bfloat
// CHECK-LLVM-NEXT: [[TMP1:%.*]] = atomicrmw fsub ptr @bf, bfloat [[CONV]] seq_cst, align 2
// CHECK-LLVM-NEXT: [[TMP2:%.*]] = fsub bfloat [[TMP1]], [[CONV]]
// CHECK-LLVM-NEXT: ret void
//
void test_bf16_sub(__bf16 val){
bf -= val;
}

// CHECK-LLVM-LABEL: define dso_local void @test_f16_sub(
// CHECK-LLVM-SAME: half noundef [[VAL:%.*]]) #[[ATTR0]] {
// CHECK-LLVM-NEXT: [[ENTRY:.*:]]
// CHECK-LLVM-NEXT: [[VAL_ADDR:%.*]] = alloca half, align 2
// CHECK-LLVM-NEXT: store half [[VAL]], ptr [[VAL_ADDR]], align 2
// CHECK-LLVM-NEXT: [[TMP0:%.*]] = load half, ptr [[VAL_ADDR]], align 2
// CHECK-LLVM-NEXT: [[TMP1:%.*]] = atomicrmw fsub ptr @h, half [[TMP0]] seq_cst, align 2
// CHECK-LLVM-NEXT: [[TMP2:%.*]] = fsub half [[TMP1]], [[TMP0]]
// CHECK-LLVM-NEXT: ret void
//
void test_f16_sub(_Float16 val){
h -= val;
}