Skip to content

[HLSL] Don't use CreateRuntimeFunction for intrinsics #145334

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 1 commit into from
Jun 23, 2025
Merged

[HLSL] Don't use CreateRuntimeFunction for intrinsics #145334

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
37 changes: 9 additions & 28 deletions clang/lib/CodeGen/CGHLSLBuiltins.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -676,35 +676,23 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
case Builtin::BI__builtin_hlsl_wave_active_sum: {
// Due to the use of variadic arguments, explicitly retreive argument
Value *OpExpr = EmitScalarExpr(E->getArg(0));
llvm::FunctionType *FT = llvm::FunctionType::get(
OpExpr->getType(), ArrayRef{OpExpr->getType()}, false);
Intrinsic::ID IID = getWaveActiveSumIntrinsic(
getTarget().getTriple().getArch(), CGM.getHLSLRuntime(),
E->getArg(0)->getType());

// Get overloaded name
std::string Name =
Intrinsic::getName(IID, ArrayRef{OpExpr->getType()}, &CGM.getModule());
return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
/*Local=*/false,
/*AssumeConvergent=*/true),
return EmitRuntimeCall(Intrinsic::getOrInsertDeclaration(
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is correct. @Keenuts did EmitRuntimeCall(Intrinsic::getOrInsertDeclaration( in https://github.com/llvm/llvm-project/pull/143127/files Thats what we expect to use for these convergence intrinsics.

Looks like this was just an accident introduced in this pr: https://github.com/llvm/llvm-project/pull/118580/files#diff-202c36399fe94363f22692e534129341972137a24721c385b1f1f05fb239dd79R19523 and then the patttern may have been copied by others.

&CGM.getModule(), IID, {OpExpr->getType()}),
ArrayRef{OpExpr}, "hlsl.wave.active.sum");
}
case Builtin::BI__builtin_hlsl_wave_active_max: {
// Due to the use of variadic arguments, explicitly retreive argument
Value *OpExpr = EmitScalarExpr(E->getArg(0));
llvm::FunctionType *FT = llvm::FunctionType::get(
OpExpr->getType(), ArrayRef{OpExpr->getType()}, false);
Intrinsic::ID IID = getWaveActiveMaxIntrinsic(
getTarget().getTriple().getArch(), CGM.getHLSLRuntime(),
E->getArg(0)->getType());

// Get overloaded name
std::string Name =
Intrinsic::getName(IID, ArrayRef{OpExpr->getType()}, &CGM.getModule());
return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
/*Local=*/false,
/*AssumeConvergent=*/true),
return EmitRuntimeCall(Intrinsic::getOrInsertDeclaration(
&CGM.getModule(), IID, {OpExpr->getType()}),
ArrayRef{OpExpr}, "hlsl.wave.active.max");
}
case Builtin::BI__builtin_hlsl_wave_get_lane_index: {
Expand Down Expand Up @@ -739,18 +727,11 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
// create our function type.
Value *OpExpr = EmitScalarExpr(E->getArg(0));
Value *OpIndex = EmitScalarExpr(E->getArg(1));
llvm::FunctionType *FT = llvm::FunctionType::get(
OpExpr->getType(), ArrayRef{OpExpr->getType(), OpIndex->getType()},
false);

// Get overloaded name
std::string Name =
Intrinsic::getName(CGM.getHLSLRuntime().getWaveReadLaneAtIntrinsic(),
ArrayRef{OpExpr->getType()}, &CGM.getModule());
return EmitRuntimeCall(CGM.CreateRuntimeFunction(FT, Name, {},
/*Local=*/false,
/*AssumeConvergent=*/true),
ArrayRef{OpExpr, OpIndex}, "hlsl.wave.readlane");
return EmitRuntimeCall(
Intrinsic::getOrInsertDeclaration(
&CGM.getModule(), CGM.getHLSLRuntime().getWaveReadLaneAtIntrinsic(),
{OpExpr->getType()}),
ArrayRef{OpExpr, OpIndex}, "hlsl.wave.readlane");
}
case Builtin::BI__builtin_hlsl_elementwise_sign: {
auto *Arg0 = E->getArg(0);
Expand Down
6 changes: 3 additions & 3 deletions clang/test/CodeGenHLSL/builtins/WaveActiveMax.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ int test_int(int expr) {
}

// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.reduce.max.i32([[TY]]) #[[#attr:]]
// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.reduce.max.i32([[TY]]) #[[#attr:]]
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.reduce.max.i32([[TY]]) #[[#attr:]]

// CHECK-LABEL: test_uint64_t
uint64_t test_uint64_t(uint64_t expr) {
Expand All @@ -27,7 +27,7 @@ uint64_t test_uint64_t(uint64_t expr) {
}

// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.reduce.umax.i64([[TY]]) #[[#attr:]]
// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.reduce.umax.i64([[TY]]) #[[#attr:]]
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.reduce.umax.i64([[TY]]) #[[#attr:]]

// Test basic lowering to runtime function call with array and float value.

Expand All @@ -40,7 +40,7 @@ float4 test_floatv4(float4 expr) {
}

// CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.reduce.max.v4f32([[TY1]]) #[[#attr]]
// CHECK-SPIRV: declare spir_func [[TY1]] @llvm.spv.wave.reduce.max.v4f32([[TY1]]) #[[#attr]]
// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.reduce.max.v4f32([[TY1]]) #[[#attr]]

// CHECK: attributes #[[#attr]] = {{{.*}} convergent {{.*}}}

6 changes: 3 additions & 3 deletions clang/test/CodeGenHLSL/builtins/WaveActiveSum.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ int test_int(int expr) {
}

// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.reduce.sum.i32([[TY]]) #[[#attr:]]
// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.reduce.sum.i32([[TY]]) #[[#attr:]]
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.reduce.sum.i32([[TY]]) #[[#attr:]]

// CHECK-LABEL: test_uint64_t
uint64_t test_uint64_t(uint64_t expr) {
Expand All @@ -27,7 +27,7 @@ uint64_t test_uint64_t(uint64_t expr) {
}

// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.reduce.usum.i64([[TY]]) #[[#attr:]]
// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.reduce.sum.i64([[TY]]) #[[#attr:]]
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.reduce.sum.i64([[TY]]) #[[#attr:]]

// Test basic lowering to runtime function call with array and float value.

Expand All @@ -40,6 +40,6 @@ float4 test_floatv4(float4 expr) {
}

// CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.reduce.sum.v4f32([[TY1]]) #[[#attr]]
// CHECK-SPIRV: declare spir_func [[TY1]] @llvm.spv.wave.reduce.sum.v4f32([[TY1]]) #[[#attr]]
// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.reduce.sum.v4f32([[TY1]]) #[[#attr]]

// CHECK: attributes #[[#attr]] = {{{.*}} convergent {{.*}}}
12 changes: 6 additions & 6 deletions clang/test/CodeGenHLSL/builtins/WaveReadLaneAt.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@ int test_int(int expr, uint idx) {
}

// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.readlane.i32([[TY]], i32) #[[#attr:]]
// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.readlane.i32([[TY]], i32) #[[#attr:]]
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.readlane.i32([[TY]], i32) #[[#attr:]]

// CHECK-LABEL: test_uint
uint test_uint(uint expr, uint idx) {
Expand All @@ -38,7 +38,7 @@ int64_t test_int64_t(int64_t expr, uint idx) {
}

// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.readlane.i64([[TY]], i32) #[[#attr:]]
// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.readlane.i64([[TY]], i32) #[[#attr:]]
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.readlane.i64([[TY]], i32) #[[#attr:]]

// CHECK-LABEL: test_uint64_t
uint64_t test_uint64_t(uint64_t expr, uint idx) {
Expand All @@ -60,7 +60,7 @@ int16_t test_int16(int16_t expr, uint idx) {
}

// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.readlane.i16([[TY]], i32) #[[#attr:]]
// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.readlane.i16([[TY]], i32) #[[#attr:]]
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.readlane.i16([[TY]], i32) #[[#attr:]]

// CHECK-LABEL: test_uint16
uint16_t test_uint16(uint16_t expr, uint idx) {
Expand All @@ -84,7 +84,7 @@ half test_half(half expr, uint idx) {
}

// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.readlane.f16([[TY]], i32) #[[#attr:]]
// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.readlane.f16([[TY]], i32) #[[#attr:]]
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.readlane.f16([[TY]], i32) #[[#attr:]]

// CHECK-LABEL: test_double
double test_double(double expr, uint idx) {
Expand All @@ -96,7 +96,7 @@ double test_double(double expr, uint idx) {
}

// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.readlane.f64([[TY]], i32) #[[#attr:]]
// CHECK-SPIRV: declare spir_func [[TY]] @llvm.spv.wave.readlane.f64([[TY]], i32) #[[#attr:]]
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.readlane.f64([[TY]], i32) #[[#attr:]]

// CHECK-LABEL: test_floatv4
float4 test_floatv4(float4 expr, uint idx) {
Expand All @@ -108,6 +108,6 @@ float4 test_floatv4(float4 expr, uint idx) {
}

// CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.readlane.v4f32([[TY1]], i32) #[[#attr]]
// CHECK-SPIRV: declare spir_func [[TY1]] @llvm.spv.wave.readlane.v4f32([[TY1]], i32) #[[#attr]]
// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.readlane.v4f32([[TY1]], i32) #[[#attr]]

// CHECK: attributes #[[#attr]] = {{{.*}} convergent {{.*}}}
Loading