Skip to content

Commit 0e77b1a

Browse files
kcloudy0717kcloudy0717
committed
[HLSL][DXIL][SPRIV] WavePrefixSum intrinsic support
This commit adds WavePrefixSum intrinsic support to HLSL.
1 parent 140e07c commit 0e77b1a

File tree

15 files changed

+384
-13
lines changed

15 files changed

+384
-13
lines changed

‎clang/include/clang/Basic/Builtins.td‎

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5072,6 +5072,12 @@ def HLSLWaveGetLaneCount : LangBuiltin<"HLSL_LANG"> {
50725072
let Prototype = "unsigned int()";
50735073
}
50745074

5075+
def HLSLWavePrefixSum : LangBuiltin<"HLSL_LANG"> {
5076+
let Spellings = ["__builtin_hlsl_wave_prefix_sum"];
5077+
let Attributes = [NoThrow, Const];
5078+
let Prototype = "void(...)";
5079+
}
5080+
50755081
def HLSLClamp : LangBuiltin<"HLSL_LANG"> {
50765082
let Spellings = ["__builtin_hlsl_elementwise_clamp"];
50775083
let Attributes = [NoThrow, Const, CustomTypeChecking];

‎clang/lib/CodeGen/CGHLSLBuiltins.cpp‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -864,6 +864,13 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
864864
{OpExpr->getType()}),
865865
ArrayRef{OpExpr, OpIndex}, "hlsl.wave.readlane");
866866
}
867+
case Builtin::BI__builtin_hlsl_wave_prefix_sum: {
868+
Value *OpExpr = EmitScalarExpr(E->getArg(0));
869+
Intrinsic::ID IID = CGM.getHLSLRuntime().getWavePrefixSumIntrinsic();
870+
return EmitRuntimeCall(Intrinsic::getOrInsertDeclaration(
871+
&CGM.getModule(), IID, {OpExpr->getType()}),
872+
ArrayRef{OpExpr}, "hlsl.wave.prefix.sum");
873+
}
867874
case Builtin::BI__builtin_hlsl_elementwise_sign: {
868875
auto *Arg0 = E->getArg(0);
869876
Value *Op0 = EmitScalarExpr(Arg0);

‎clang/lib/CodeGen/CGHLSLRuntime.h‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -144,6 +144,7 @@ class CGHLSLRuntime {
144144
GENERATE_HLSL_INTRINSIC_FUNCTION(WaveIsFirstLane, wave_is_first_lane)
145145
GENERATE_HLSL_INTRINSIC_FUNCTION(WaveGetLaneCount, wave_get_lane_count)
146146
GENERATE_HLSL_INTRINSIC_FUNCTION(WaveReadLaneAt, wave_readlane)
147+
GENERATE_HLSL_INTRINSIC_FUNCTION(WavePrefixSum, wave_prefix_sum)
147148
GENERATE_HLSL_INTRINSIC_FUNCTION(FirstBitUHigh, firstbituhigh)
148149
GENERATE_HLSL_INTRINSIC_FUNCTION(FirstBitSHigh, firstbitshigh)
149150
GENERATE_HLSL_INTRINSIC_FUNCTION(FirstBitLow, firstbitlow)

‎clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h‎

Lines changed: 99 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2802,6 +2802,105 @@ __attribute__((convergent)) double3 WaveActiveSum(double3);
28022802
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_active_sum)
28032803
__attribute__((convergent)) double4 WaveActiveSum(double4);
28042804

2805+
//===----------------------------------------------------------------------===//
2806+
// WavePrefixSum builtins
2807+
//===----------------------------------------------------------------------===//
2808+
2809+
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.0)
2810+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2811+
__attribute__((convergent)) half WavePrefixSum(half);
2812+
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.0)
2813+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2814+
__attribute__((convergent)) half2 WavePrefixSum(half2);
2815+
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.0)
2816+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2817+
__attribute__((convergent)) half3 WavePrefixSum(half3);
2818+
_HLSL_16BIT_AVAILABILITY(shadermodel, 6.0)
2819+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2820+
__attribute__((convergent)) half4 WavePrefixSum(half4);
2821+
2822+
#ifdef __HLSL_ENABLE_16_BIT
2823+
_HLSL_AVAILABILITY(shadermodel, 6.0)
2824+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2825+
__attribute__((convergent)) int16_t WavePrefixSum(int16_t);
2826+
_HLSL_AVAILABILITY(shadermodel, 6.0)
2827+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2828+
__attribute__((convergent)) int16_t2 WavePrefixSum(int16_t2);
2829+
_HLSL_AVAILABILITY(shadermodel, 6.0)
2830+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2831+
__attribute__((convergent)) int16_t3 WavePrefixSum(int16_t3);
2832+
_HLSL_AVAILABILITY(shadermodel, 6.0)
2833+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2834+
__attribute__((convergent)) int16_t4 WavePrefixSum(int16_t4);
2835+
2836+
_HLSL_AVAILABILITY(shadermodel, 6.0)
2837+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2838+
__attribute__((convergent)) uint16_t WavePrefixSum(uint16_t);
2839+
_HLSL_AVAILABILITY(shadermodel, 6.0)
2840+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2841+
__attribute__((convergent)) uint16_t2 WavePrefixSum(uint16_t2);
2842+
_HLSL_AVAILABILITY(shadermodel, 6.0)
2843+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2844+
__attribute__((convergent)) uint16_t3 WavePrefixSum(uint16_t3);
2845+
_HLSL_AVAILABILITY(shadermodel, 6.0)
2846+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2847+
__attribute__((convergent)) uint16_t4 WavePrefixSum(uint16_t4);
2848+
#endif
2849+
2850+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2851+
__attribute__((convergent)) int WavePrefixSum(int);
2852+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2853+
__attribute__((convergent)) int2 WavePrefixSum(int2);
2854+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2855+
__attribute__((convergent)) int3 WavePrefixSum(int3);
2856+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2857+
__attribute__((convergent)) int4 WavePrefixSum(int4);
2858+
2859+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2860+
__attribute__((convergent)) uint WavePrefixSum(uint);
2861+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2862+
__attribute__((convergent)) uint2 WavePrefixSum(uint2);
2863+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2864+
__attribute__((convergent)) uint3 WavePrefixSum(uint3);
2865+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2866+
__attribute__((convergent)) uint4 WavePrefixSum(uint4);
2867+
2868+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2869+
__attribute__((convergent)) int64_t WavePrefixSum(int64_t);
2870+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2871+
__attribute__((convergent)) int64_t2 WavePrefixSum(int64_t2);
2872+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2873+
__attribute__((convergent)) int64_t3 WavePrefixSum(int64_t3);
2874+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2875+
__attribute__((convergent)) int64_t4 WavePrefixSum(int64_t4);
2876+
2877+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2878+
__attribute__((convergent)) uint64_t WavePrefixSum(uint64_t);
2879+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2880+
__attribute__((convergent)) uint64_t2 WavePrefixSum(uint64_t2);
2881+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2882+
__attribute__((convergent)) uint64_t3 WavePrefixSum(uint64_t3);
2883+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2884+
__attribute__((convergent)) uint64_t4 WavePrefixSum(uint64_t4);
2885+
2886+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2887+
__attribute__((convergent)) float WavePrefixSum(float);
2888+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2889+
__attribute__((convergent)) float2 WavePrefixSum(float2);
2890+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2891+
__attribute__((convergent)) float3 WavePrefixSum(float3);
2892+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2893+
__attribute__((convergent)) float4 WavePrefixSum(float4);
2894+
2895+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2896+
__attribute__((convergent)) double WavePrefixSum(double);
2897+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2898+
__attribute__((convergent)) double2 WavePrefixSum(double2);
2899+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2900+
__attribute__((convergent)) double3 WavePrefixSum(double3);
2901+
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_wave_prefix_sum)
2902+
__attribute__((convergent)) double4 WavePrefixSum(double4);
2903+
28052904
//===----------------------------------------------------------------------===//
28062905
// sign builtins
28072906
//===----------------------------------------------------------------------===//

‎clang/lib/Sema/SemaHLSL.cpp‎

Lines changed: 32 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -2892,10 +2892,13 @@ static bool CheckAnyScalarOrVector(Sema *S, CallExpr *TheCall,
28922892
return false;
28932893
}
28942894

2895-
static bool CheckWaveActive(Sema *S, CallExpr *TheCall) {
2895+
// Check that the argument is not a bool or vector<bool>
2896+
// Returns true on error
2897+
static bool CheckNotBoolScalarOrVector(Sema *S, CallExpr *TheCall,
2898+
unsigned ArgIndex) {
28962899
QualType BoolType = S->getASTContext().BoolTy;
2897-
assert(TheCall->getNumArgs() >= 1);
2898-
QualType ArgType = TheCall->getArg(0)->getType();
2900+
assert(ArgIndex < TheCall->getNumArgs());
2901+
QualType ArgType = TheCall->getArg(ArgIndex)->getType();
28992902
auto *VTy = ArgType->getAs<VectorType>();
29002903
// is the bool or vector<bool>
29012904
if (S->Context.hasSameUnqualifiedType(ArgType, BoolType) ||
@@ -2909,6 +2912,18 @@ static bool CheckWaveActive(Sema *S, CallExpr *TheCall) {
29092912
return false;
29102913
}
29112914

2915+
static bool CheckWaveActive(Sema *S, CallExpr *TheCall) {
2916+
if (CheckNotBoolScalarOrVector(S, TheCall, 0))
2917+
return true;
2918+
return false;
2919+
}
2920+
2921+
static bool CheckWavePrefix(Sema *S, CallExpr *TheCall) {
2922+
if (CheckNotBoolScalarOrVector(S, TheCall, 0))
2923+
return true;
2924+
return false;
2925+
}
2926+
29122927
static bool CheckBoolSelect(Sema *S, CallExpr *TheCall) {
29132928
assert(TheCall->getNumArgs() == 3);
29142929
Expr *Arg1 = TheCall->getArg(1);
@@ -3371,6 +3386,20 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
33713386
return true;
33723387
break;
33733388
}
3389+
case Builtin::BI__builtin_hlsl_wave_prefix_sum: {
3390+
if (SemaRef.checkArgCount(TheCall, 1))
3391+
return true;
3392+
3393+
// Ensure input expr type is a scalar/vector and the same as the return type
3394+
if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0))
3395+
return true;
3396+
if (CheckWavePrefix(&SemaRef, TheCall))
3397+
return true;
3398+
ExprResult Expr = TheCall->getArg(0);
3399+
QualType ArgTyExpr = Expr.get()->getType();
3400+
TheCall->setType(ArgTyExpr);
3401+
break;
3402+
}
33743403
case Builtin::BI__builtin_hlsl_elementwise_splitdouble: {
33753404
if (SemaRef.checkArgCount(TheCall, 3))
33763405
return true;

‎clang/test/CodeGenHLSL/builtins/WavePrefixSum.hlsl‎

Lines changed: 46 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,46 @@
1+
// RUN: %clang_cc1 -std=hlsl2021 -finclude-default-header -triple \
2+
// RUN: dxil-pc-shadermodel6.3-compute %s -emit-llvm -disable-llvm-passes -o - | \
3+
// RUN: FileCheck %s --check-prefixes=CHECK,CHECK-DXIL
4+
// RUN: %clang_cc1 -std=hlsl2021 -finclude-default-header -triple \
5+
// RUN: spirv-pc-vulkan-compute %s -emit-llvm -disable-llvm-passes -o - | \
6+
// RUN: FileCheck %s --check-prefixes=CHECK,CHECK-SPIRV
7+
8+
// Test basic lowering to runtime function call.
9+
10+
// CHECK-LABEL: test_int
11+
int test_int(int expr) {
12+
// CHECK-SPIRV: %[[RET:.*]] = call spir_func [[TY:.*]] @llvm.spv.wave.prefix.sum.i32([[TY]] %[[#]])
13+
// CHECK-DXIL: %[[RET:.*]] = call [[TY:.*]] @llvm.dx.wave.prefix.sum.i32([[TY]] %[[#]])
14+
// CHECK: ret [[TY]] %[[RET]]
15+
return WavePrefixSum(expr);
16+
}
17+
18+
// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.prefix.sum.i32([[TY]]) #[[#attr:]]
19+
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.prefix.sum.i32([[TY]]) #[[#attr:]]
20+
21+
// CHECK-LABEL: test_uint64_t
22+
uint64_t test_uint64_t(uint64_t expr) {
23+
// CHECK-SPIRV: %[[RET:.*]] = call spir_func [[TY:.*]] @llvm.spv.wave.prefix.sum.i64([[TY]] %[[#]])
24+
// CHECK-DXIL: %[[RET:.*]] = call [[TY:.*]] @llvm.dx.wave.prefix.sum.i64([[TY]] %[[#]])
25+
// CHECK: ret [[TY]] %[[RET]]
26+
return WavePrefixSum(expr);
27+
}
28+
29+
// CHECK-DXIL: declare [[TY]] @llvm.dx.wave.prefix.sum.i64([[TY]]) #[[#attr:]]
30+
// CHECK-SPIRV: declare [[TY]] @llvm.spv.wave.prefix.sum.i64([[TY]]) #[[#attr:]]
31+
32+
// Test basic lowering to runtime function call with array and float value.
33+
34+
// CHECK-LABEL: test_floatv4
35+
float4 test_floatv4(float4 expr) {
36+
// CHECK-SPIRV: %[[RET1:.*]] = call reassoc nnan ninf nsz arcp afn spir_func [[TY1:.*]] @llvm.spv.wave.prefix.sum.v4f32([[TY1]] %[[#]]
37+
// CHECK-DXIL: %[[RET1:.*]] = call reassoc nnan ninf nsz arcp afn [[TY1:.*]] @llvm.dx.wave.prefix.sum.v4f32([[TY1]] %[[#]])
38+
// CHECK: ret [[TY1]] %[[RET1]]
39+
return WavePrefixSum(expr);
40+
}
41+
42+
// CHECK-DXIL: declare [[TY1]] @llvm.dx.wave.prefix.sum.v4f32([[TY1]]) #[[#attr]]
43+
// CHECK-SPIRV: declare [[TY1]] @llvm.spv.wave.prefix.sum.v4f32([[TY1]]) #[[#attr]]
44+
45+
// CHECK: attributes #[[#attr]] = {{{.*}} convergent {{.*}}}
46+

‎clang/test/SemaHLSL/BuiltIns/WavePrefixSum-errors.hlsl‎

Lines changed: 28 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,28 @@
1+
// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -emit-llvm-only -disable-llvm-passes -verify
2+
3+
int test_too_few_arg() {
4+
return __builtin_hlsl_wave_prefix_sum();
5+
// expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
6+
}
7+
8+
float2 test_too_many_arg(float2 p0) {
9+
return __builtin_hlsl_wave_prefix_sum(p0, p0);
10+
// expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
11+
}
12+
13+
bool test_expr_bool_type_check(bool p0) {
14+
return __builtin_hlsl_wave_prefix_sum(p0);
15+
// expected-error@-1 {{invalid operand of type 'bool'}}
16+
}
17+
18+
bool2 test_expr_bool_vec_type_check(bool2 p0) {
19+
return __builtin_hlsl_wave_prefix_sum(p0);
20+
// expected-error@-1 {{invalid operand of type 'bool2' (aka 'vector<bool, 2>')}}
21+
}
22+
23+
struct S { float f; };
24+
25+
S test_expr_struct_type_check(S p0) {
26+
return __builtin_hlsl_wave_prefix_sum(p0);
27+
// expected-error@-1 {{invalid operand of type 'S' where a scalar or vector is required}}
28+
}

‎llvm/include/llvm/IR/IntrinsicsDirectX.td‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -164,6 +164,7 @@ def int_dx_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrCon
164164
def int_dx_wave_readlane : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
165165
def int_dx_wave_get_lane_count
166166
: DefaultAttrsIntrinsic<[llvm_i32_ty], [], [IntrConvergent]>;
167+
def int_dx_wave_prefix_sum : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>], [IntrConvergent, IntrNoMem]>;
167168
def int_dx_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
168169
def int_dx_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
169170
def int_dx_splitdouble : DefaultAttrsIntrinsic<[llvm_anyint_ty, LLVMMatchType<0>],

‎llvm/include/llvm/IR/IntrinsicsSPIRV.td‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -129,6 +129,7 @@ def int_spv_rsqrt : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty]
129129
def int_spv_wave_readlane : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
130130
def int_spv_wave_get_lane_count
131131
: DefaultAttrsIntrinsic<[llvm_i32_ty], [], [IntrConvergent]>;
132+
def int_spv_wave_prefix_sum : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>], [IntrConvergent, IntrNoMem]>;
132133
def int_spv_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
133134
def int_spv_radians : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty], [IntrNoMem]>;
134135
def int_spv_group_memory_barrier_with_group_sync

‎llvm/lib/Target/DirectX/DXIL.td‎

Lines changed: 18 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1079,6 +1079,24 @@ def WaveActiveOp : DXILOp<119, waveActiveOp> {
10791079
let attributes = [Attributes<DXIL1_0, []>];
10801080
}
10811081

1082+
def WavePrefixOp : DXILOp<121, wavePrefixOp> {
1083+
let Doc = "returns partial result of the computation in the corresponding lane";
1084+
let intrinsics = [
1085+
IntrinSelect<int_dx_wave_prefix_sum,
1086+
[
1087+
IntrinArgIndex<0>, IntrinArgI8<WaveOpKind_Sum>
1088+
]>
1089+
];
1090+
1091+
let arguments = [OverloadTy, Int8Ty];
1092+
let result = OverloadTy;
1093+
let overloads = [
1094+
Overloads<DXIL1_0, [HalfTy, FloatTy, DoubleTy, Int16Ty, Int32Ty, Int64Ty]>
1095+
];
1096+
let stages = [Stages<DXIL1_0, [all_stages]>];
1097+
let attributes = [Attributes<DXIL1_0, []>];
1098+
}
1099+
10821100
def LegacyF16ToF32 : DXILOp<131, legacyF16ToF32> {
10831101
let Doc = "returns the float16 stored in the low-half of the uint converted "
10841102
"to a float";

0 commit comments

Comments
 (0)