Skip to content

[AMDGPU] Add builtins for wave reduction intrinsics #127013

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

Closed
wants to merge 5 commits into from
Closed

[AMDGPU] Add builtins for wave reduction intrinsics #127013

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
29bb614
[AMDGPU] Wave Reduce Intrinsics for i32 type
easyonaadit Feb 5, 2025
48783dc
Renamed function to getIdentityValueForWaveReduction
easyonaadit Feb 12, 2025
38642e4
[AMDGPU] Add builtins for wave reduction intrinsics
easyonaadit Feb 13, 2025
ad1dd4e
add strategy argument in builtin
easyonaadit Feb 19, 2025
054aa39
test for stratergy
easyonaadit Feb 19, 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
18 changes: 18 additions & 0 deletions clang/include/clang/Basic/BuiltinsAMDGPU.def
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -346,6 +346,24 @@ BUILTIN(__builtin_amdgcn_endpgm, "v", "nr")
BUILTIN(__builtin_amdgcn_get_fpenv, "WUi", "n")
BUILTIN(__builtin_amdgcn_set_fpenv, "vWUi", "n")

//===----------------------------------------------------------------------===//

// Wave Reduction builtins.

//===----------------------------------------------------------------------===//

BUILTIN(__builtin_amdgcn_wave_reduce_add_i32, "ii", "nc")
BUILTIN(__builtin_amdgcn_wave_reduce_uadd_i32, "ii", "nc")
BUILTIN(__builtin_amdgcn_wave_reduce_sub_i32, "ii", "nc")
BUILTIN(__builtin_amdgcn_wave_reduce_usub_i32, "ii", "nc")
BUILTIN(__builtin_amdgcn_wave_reduce_min_i32, "ii", "nc")
BUILTIN(__builtin_amdgcn_wave_reduce_umin_i32, "ii", "nc")
BUILTIN(__builtin_amdgcn_wave_reduce_max_i32, "ii", "nc")
BUILTIN(__builtin_amdgcn_wave_reduce_umax_i32, "ii", "nc")
BUILTIN(__builtin_amdgcn_wave_reduce_and_i32, "ii", "nc")
BUILTIN(__builtin_amdgcn_wave_reduce_or_i32, "ii", "nc")
BUILTIN(__builtin_amdgcn_wave_reduce_xor_i32, "ii", "nc")

//===----------------------------------------------------------------------===//
// R600-NI only builtins.
//===----------------------------------------------------------------------===//
Expand Down
53 changes: 53 additions & 0 deletions clang/lib/CodeGen/CGBuiltin.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20212,6 +20212,59 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
llvm::Value *Env = EmitScalarExpr(E->getArg(0));
return Builder.CreateCall(F, {Env});
}
case AMDGPU::BI__builtin_amdgcn_wave_reduce_add_i32:
case AMDGPU::BI__builtin_amdgcn_wave_reduce_uadd_i32:
case AMDGPU::BI__builtin_amdgcn_wave_reduce_sub_i32:
case AMDGPU::BI__builtin_amdgcn_wave_reduce_usub_i32:
case AMDGPU::BI__builtin_amdgcn_wave_reduce_min_i32:
case AMDGPU::BI__builtin_amdgcn_wave_reduce_umin_i32:
case AMDGPU::BI__builtin_amdgcn_wave_reduce_max_i32:
case AMDGPU::BI__builtin_amdgcn_wave_reduce_umax_i32:
case AMDGPU::BI__builtin_amdgcn_wave_reduce_and_i32:
case AMDGPU::BI__builtin_amdgcn_wave_reduce_or_i32:
case AMDGPU::BI__builtin_amdgcn_wave_reduce_xor_i32: {
Intrinsic::ID IID;
switch (BuiltinID) {
case AMDGPU::BI__builtin_amdgcn_wave_reduce_add_i32:
IID = Intrinsic::amdgcn_wave_reduce_add;
break;
case AMDGPU::BI__builtin_amdgcn_wave_reduce_uadd_i32:
IID = Intrinsic::amdgcn_wave_reduce_uadd;
break;
case AMDGPU::BI__builtin_amdgcn_wave_reduce_sub_i32:
IID = Intrinsic::amdgcn_wave_reduce_sub;
break;
case AMDGPU::BI__builtin_amdgcn_wave_reduce_usub_i32:
IID = Intrinsic::amdgcn_wave_reduce_usub;
break;
case AMDGPU::BI__builtin_amdgcn_wave_reduce_min_i32:
IID = Intrinsic::amdgcn_wave_reduce_min;
break;
case AMDGPU::BI__builtin_amdgcn_wave_reduce_umin_i32:
IID = Intrinsic::amdgcn_wave_reduce_umin;
break;
case AMDGPU::BI__builtin_amdgcn_wave_reduce_max_i32:
IID = Intrinsic::amdgcn_wave_reduce_max;
break;
case AMDGPU::BI__builtin_amdgcn_wave_reduce_umax_i32:
IID = Intrinsic::amdgcn_wave_reduce_umax;
break;
case AMDGPU::BI__builtin_amdgcn_wave_reduce_and_i32:
IID = Intrinsic::amdgcn_wave_reduce_and;
break;
case AMDGPU::BI__builtin_amdgcn_wave_reduce_or_i32:
IID = Intrinsic::amdgcn_wave_reduce_or;
break;
case AMDGPU::BI__builtin_amdgcn_wave_reduce_xor_i32:
IID = Intrinsic::amdgcn_wave_reduce_xor;
break;
}
llvm::Value *Src0 = EmitScalarExpr(E->getArg(0));
llvm::Function *F = CGM.getIntrinsic(IID, {Src0->getType()});
llvm::Value *Strategy =
llvm::ConstantInt::get(llvm::Type::getInt32Ty(getLLVMContext()), 0);
return Builder.CreateCall(F, {Src0, Strategy});
}
case AMDGPU::BI__builtin_amdgcn_read_exec:
return EmitAMDGCNBallotForExec(*this, E, Int64Ty, Int64Ty, false);
case AMDGPU::BI__builtin_amdgcn_read_exec_lo:
Expand Down
77 changes: 77 additions & 0 deletions clang/test/CodeGenOpenCL/builtins-amdgcn.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -398,6 +398,83 @@ void test_s_sendmsghalt_var(int in)
__builtin_amdgcn_s_sendmsghalt(1, in);
}

// CHECK-LABEL: @test_wave_reduce_add_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.add.i32(
void test_wave_reduce_add_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_add_i32(in);
}

// CHECK-LABEL: @test_wave_reduce_uadd_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.uadd.i32(
void test_wave_reduce_uadd_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_uadd_i32(in);
}

// CHECK-LABEL: @test_wave_reduce_sub_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.sub.i32(
void test_wave_reduce_sub_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_sub_i32(in);
}

// CHECK-LABEL: @test_wave_reduce_usub_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.usub.i32(
void test_wave_reduce_usub_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_usub_i32(in);
}

// CHECK-LABEL: @test_wave_reduce_min_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.min.i32(
void test_wave_reduce_min_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_min_i32(in);
}

// CHECK-LABEL: @test_wave_reduce_umin_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.umin.i32(
void test_wave_reduce_umin_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_umin_i32(in);
}

// CHECK-LABEL: @test_wave_reduce_max_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.max.i32(
void test_wave_reduce_max_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_max_i32(in);
}

// CHECK-LABEL: @test_wave_reduce_umax_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.umax.i32(
void test_wave_reduce_umax_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_umax_i32(in);
}

// CHECK-LABEL: @test_wave_reduce_and_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.and.i32(
void test_wave_reduce_and_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_and_i32(in);
}

// CHECK-LABEL: @test_wave_reduce_or_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.or.i32(
void test_wave_reduce_or_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_or_i32(in);
}

// CHECK-LABEL: @test_wave_reduce_xor_i32
// CHECK: {{.*}}call{{.*}} i32 @llvm.amdgcn.wave.reduce.xor.i32(
void test_wave_reduce_xor_i32(global int* out, int in)
{
*out = __builtin_amdgcn_wave_reduce_xor_i32(in);
}

// CHECK-LABEL: @test_s_barrier
// CHECK: {{.*}}call{{.*}} void @llvm.amdgcn.s.barrier(
void test_s_barrier()
Expand Down