Skip to content

Commit fb4d7d1

Browse files
anupamachandraanupamachandra
authored
[0029] [Main] For OuterProductAccumulate, matrix layout must be outerproductoptimal and matrix stride must be zero (#7417)
Implements the DXIL portion of microsoft/hlsl-specs#494, The HLSL checks will be a part of the HLSL validation checks, did not add it to this PR due to shared infrastructure.
1 parent 377c4ca commit fb4d7d1

File tree

11 files changed

+525
-305
lines changed

11 files changed

+525
-305
lines changed

‎docs/DXIL.rst

Lines changed: 293 additions & 291 deletions
Large diffs are not rendered by default.

‎include/dxc/DXIL/DxilConstants.h

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1609,6 +1609,7 @@ const unsigned kMatVecMulAddIsOutputUnsignedIdx = 15;
16091609
// Outer Product Accumulate
16101610
const unsigned kOuterProdAccMatrixInterpretation = 5;
16111611
const unsigned kOuterProdAccMatrixLayout = 6;
1612+
const unsigned kOuterProdAccMatrixStride = 7;
16121613

16131614
// TODO: add operand index for all the OpCodeClass.
16141615
} // namespace OperandIndex

‎lib/DxilValidation/DxilValidation.cpp

Lines changed: 26 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1233,6 +1233,32 @@ static void ValidateImmOperandsForOuterProdAcc(CallInst *CI,
12331233
{"MatrixLayout"});
12341234
return;
12351235
}
1236+
ConstantInt *ML = cast<ConstantInt>(MatrixLayout);
1237+
uint64_t MLValue = ML->getLimitedValue();
1238+
if (MLValue !=
1239+
static_cast<unsigned>(DXIL::LinalgMatrixLayout::OuterProductOptimal))
1240+
ValCtx.EmitInstrFormatError(
1241+
CI,
1242+
ValidationRule::
1243+
InstrLinalgInvalidMatrixLayoutValueForOuterProductAccumulate,
1244+
{GetMatrixLayoutStr(MLValue),
1245+
GetMatrixLayoutStr(static_cast<unsigned>(
1246+
DXIL::LinalgMatrixLayout::OuterProductOptimal))});
1247+
1248+
llvm::Value *MatrixStride =
1249+
CI->getOperand(DXIL::OperandIndex::kOuterProdAccMatrixStride);
1250+
if (!llvm::isa<llvm::Constant>(MatrixStride)) {
1251+
ValCtx.EmitInstrError(
1252+
CI, ValidationRule::InstrLinalgMatrixStrideZeroForOptimalLayouts);
1253+
return;
1254+
}
1255+
ConstantInt *MS = cast<ConstantInt>(MatrixStride);
1256+
uint64_t MSValue = MS->getLimitedValue();
1257+
if (MSValue != 0) {
1258+
ValCtx.EmitInstrError(
1259+
CI, ValidationRule::InstrLinalgMatrixStrideZeroForOptimalLayouts);
1260+
return;
1261+
}
12361262
}
12371263

12381264
// Validate the type-defined mask compared to the store value mask which

‎tools/clang/test/CodeGenDXIL/hlsl/intrinsics/linalg_builtins/check-shader-stages.hlsl

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -43,7 +43,7 @@ void UseCoopVec() {
4343
const uint opa_matrix_offset = 0;
4444
const uint opa_matrix_interpretation = 5; /*U32*/
4545
const uint opa_matrix_layout = 3; /*OuterProductOptimal*/
46-
const uint opa_matrix_stride = 64;
46+
const uint opa_matrix_stride = 0;
4747

4848
__builtin_OuterProductAccumulate(input_vector1, input_vector2,
4949
rw_matrix_buffer, opa_matrix_offset, opa_matrix_interpretation,

‎tools/clang/test/CodeGenDXIL/hlsl/intrinsics/linalg_builtins/linalg-builtins.hlsl

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -58,12 +58,12 @@ void cs_main()
5858
const uint opa_matrix_offset = 0;
5959
const uint opa_matrix_interpretation = 5; /*U32*/
6060
const uint opa_matrix_layout = 3; /*OuterProductOptimal*/
61-
const uint opa_matrix_stride = 64;
61+
const uint opa_matrix_stride = 0;
6262

6363
// CHECK: %[[MLD2:[^ ]+]] = load %struct.RWByteAddressBuffer, %struct.RWByteAddressBuffer* @"\01?rw_matrix_buffer@@3URWByteAddressBuffer@@A"
6464
// CHECK: %[[MCH2:[^ ]+]] = call %dx.types.Handle @"dx.hl.createhandle..%dx.types.Handle (i32, %struct.RWByteAddressBuffer)"(i32 0, %struct.RWByteAddressBuffer %[[MLD2]])
6565
// CHECK: %[[MAH2:[^ ]+]] = call %dx.types.Handle @"dx.hl.annotatehandle..%dx.types.Handle (i32, %dx.types.Handle, %dx.types.ResourceProperties, %struct.RWByteAddressBuffer)"(i32 14, %dx.types.Handle %[[MCH2]], %dx.types.ResourceProperties { i32 4107, i32 0 }, %struct.RWByteAddressBuffer undef)
66-
// CHECK: call void @"dx.hl.op..void (i32, <8 x i32>, <8 x i32>, %dx.types.Handle, i32, i32, i32, i32)"(i32 392, <8 x i32> %{{[^ ]+}}, <8 x i32> %{{[^ ]+}}, %dx.types.Handle %[[MAH2]], i32 0, i32 5, i32 3, i32 64)
66+
// CHECK: call void @"dx.hl.op..void (i32, <8 x i32>, <8 x i32>, %dx.types.Handle, i32, i32, i32, i32)"(i32 392, <8 x i32> %{{[^ ]+}}, <8 x i32> %{{[^ ]+}}, %dx.types.Handle %[[MAH2]], i32 0, i32 5, i32 3, i32 0)
6767
__builtin_OuterProductAccumulate(input_vector1, input_vector2,
6868
rw_matrix_buffer, opa_matrix_offset, opa_matrix_interpretation,
6969
opa_matrix_layout, opa_matrix_stride);

‎tools/clang/test/CodeGenDXIL/hlsl/intrinsics/linalg_builtins/outer-product-accumulate-multioverload.hlsl

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -1,8 +1,8 @@
1-
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=float16_t -DMI=F16 -DML=RowMajor | FileCheck %s --check-prefixes COMMON,DXIL-0
1+
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=float16_t -DMI=F16 -DML=OuterProductOptimal | FileCheck %s --check-prefixes COMMON,DXIL-0
22
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=float16_t -DMI=F8_E4M3 -DML=OuterProductOptimal | FileCheck %s --check-prefixes COMMON,DXIL-1
33
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=uint -DMI=U8 -DML=OuterProductOptimal | FileCheck %s --check-prefixes COMMON,DXIL-2
44

5-
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=float16_t -DMI=F16 -DML=RowMajor -fcgl | FileCheck %s --check-prefixes COMMON,HLOP-0
5+
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=float16_t -DMI=F16 -DML=OuterProductOptimal -fcgl | FileCheck %s --check-prefixes COMMON,HLOP-0
66
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=float16_t -DMI=F8_E4M3 -DML=OuterProductOptimal -fcgl | FileCheck %s --check-prefixes COMMON,HLOP-1
77
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=uint -DMI=U8 -DML=OuterProductOptimal -fcgl | FileCheck %s --check-prefixes COMMON,HLOP-2
88

@@ -11,12 +11,12 @@ ByteAddressBuffer input_vector_buffer2;
1111
RWByteAddressBuffer matrix_buffer;
1212

1313
// COMMON: define void @main()
14-
// DXIL-0: call void @dx.op.outerProductAccumulate.v8f16.v8f16(i32 307, <8 x half> %{{[^ ]+}}, <8 x half> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 8, i32 0, i32 64) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
15-
// HLOP-0: call void @"dx.hl.op..void (i32, <8 x half>, <8 x half>, %dx.types.Handle, i32, i32, i32, i32)"(i32 392, <8 x half> %{{[^ ]+}}, <8 x half> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 8, i32 0, i32 64)
16-
// DXIL-1: call void @dx.op.outerProductAccumulate.v8f16.v8f16(i32 307, <8 x half> %{{[^ ]+}}, <8 x half> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 21, i32 3, i32 64) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
17-
// HLOP-1: call void @"dx.hl.op..void (i32, <8 x half>, <8 x half>, %dx.types.Handle, i32, i32, i32, i32)"(i32 392, <8 x half> %{{[^ ]+}}, <8 x half> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 21, i32 3, i32 64)
18-
// DXIL-2: call void @dx.op.outerProductAccumulate.v8i32.v8i32(i32 307, <8 x i32> %{{[^ ]+}}, <8 x i32> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 19, i32 3, i32 64) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
19-
// HLOP-2: call void @"dx.hl.op..void (i32, <8 x i32>, <8 x i32>, %dx.types.Handle, i32, i32, i32, i32)"(i32 392, <8 x i32> %{{[^ ]+}}, <8 x i32> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 19, i32 3, i32 64)
14+
// DXIL-0: call void @dx.op.outerProductAccumulate.v8f16.v8f16(i32 307, <8 x half> %{{[^ ]+}}, <8 x half> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 8, i32 3, i32 0) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
15+
// HLOP-0: call void @"dx.hl.op..void (i32, <8 x half>, <8 x half>, %dx.types.Handle, i32, i32, i32, i32)"(i32 392, <8 x half> %{{[^ ]+}}, <8 x half> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 8, i32 3, i32 0)
16+
// DXIL-1: call void @dx.op.outerProductAccumulate.v8f16.v8f16(i32 307, <8 x half> %{{[^ ]+}}, <8 x half> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 21, i32 3, i32 0) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
17+
// HLOP-1: call void @"dx.hl.op..void (i32, <8 x half>, <8 x half>, %dx.types.Handle, i32, i32, i32, i32)"(i32 392, <8 x half> %{{[^ ]+}}, <8 x half> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 21, i32 3, i32 0)
18+
// DXIL-2: call void @dx.op.outerProductAccumulate.v8i32.v8i32(i32 307, <8 x i32> %{{[^ ]+}}, <8 x i32> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 19, i32 3, i32 0) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
19+
// HLOP-2: call void @"dx.hl.op..void (i32, <8 x i32>, <8 x i32>, %dx.types.Handle, i32, i32, i32, i32)"(i32 392, <8 x i32> %{{[^ ]+}}, <8 x i32> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 19, i32 3, i32 0)
2020

2121
enum CompType {
2222
Invalid = 0,
@@ -63,7 +63,7 @@ void main()
6363
const uint matrix_interpretation = MI;
6464
const uint matrix_layout = ML;
6565
const uint matrix_offset = 0;
66-
const uint matrix_stride = 64;
66+
const uint matrix_stride = 0;
6767

6868
__builtin_OuterProductAccumulate(input_vector1, input_vector2, matrix_buffer, matrix_offset, matrix_interpretation, matrix_layout, matrix_stride);
6969

‎tools/clang/test/CodeGenDXIL/hlsl/linalg/outer-product-accumulate-matrix-layout.hlsl

Lines changed: 28 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,28 @@
1+
// RUN: %dxc -I %hlsl_headers -T cs_6_9 %s -enable-16bit-types -DML=MATRIX_LAYOUT_OUTER_PRODUCT_OPTIMAL -DSTRIDE=0 2>&1 | FileCheck %s
2+
3+
//Source file for the IR in \tools\clang\test\LitDXILValidation\outer-product-accumulate-matrix-layout-failing.ll
4+
//Source file for the IR in \tools\clang\test\LitDXILValidation\outer-product-accumulate-matrix-layout-passing.ll
5+
6+
ByteAddressBuffer input_vector_buffer;
7+
ByteAddressBuffer input_vector_buffer2;
8+
RWByteAddressBuffer matrix_buffer;
9+
10+
#include <dx/linalg.h>
11+
12+
// CHECK: call void @dx.op.outerProductAccumulate.v8f16.v8f16(i32 307, <8 x half> %{{[^ ]+}}, <8 x half> %{{[^ ]+}}, %dx.types.Handle %{{[^ ]+}}, i32 0, i32 8, i32 3, i32 0)
13+
using namespace dx::linalg;
14+
15+
[Numthreads(1,1,1)]
16+
[shader("compute")]
17+
void main()
18+
{
19+
vector<half, 8> input_vector1 = input_vector_buffer.Load<vector<half, 8> >(0);
20+
vector<half, 8> input_vector2 = input_vector_buffer2.Load<vector<half, 8> >(0);
21+
22+
const uint matrix_interpretation = DATA_TYPE_FLOAT16;
23+
const uint matrix_layout = ML;
24+
const uint matrix_offset = 0;
25+
const uint matrix_stride = STRIDE;
26+
27+
__builtin_OuterProductAccumulate(input_vector1, input_vector2, matrix_buffer, matrix_offset, matrix_interpretation, matrix_layout, matrix_stride);
28+
}

‎tools/clang/test/DXC/Passes/DxilGen/linalg-builtins.ll

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -76,8 +76,8 @@ entry:
7676

7777
;CHECK: %[[RWMCH0:[^ ]+]] = call %dx.types.Handle @dx.op.createHandleForLib.struct.RWByteAddressBuffer(i32 160, %struct.RWByteAddressBuffer %[[RWMLD0]]
7878
;CHECK: %[[RWMAH0:[^ ]+]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle %[[RWMCH0]]
79-
;CHECK: call void @dx.op.outerProductAccumulate.v8i32.v8i32(i32 307, <8 x i32> %{{[^ ]+}}, <8 x i32> %{{[^ ]+}}, %dx.types.Handle %[[RWMAH0]], i32 0, i32 5, i32 3, i32 64)
80-
call void @"dx.hl.op..void (i32, <8 x i32>, <8 x i32>, %dx.types.Handle, i32, i32, i32, i32)"(i32 392, <8 x i32> %tmp25, <8 x i32> %tmp29, %dx.types.Handle %tmp32, i32 0, i32 5, i32 3, i32 64), !dbg !37 ; line:67 col:5
79+
;CHECK: call void @dx.op.outerProductAccumulate.v8i32.v8i32(i32 307, <8 x i32> %{{[^ ]+}}, <8 x i32> %{{[^ ]+}}, %dx.types.Handle %[[RWMAH0]], i32 0, i32 5, i32 3, i32 0)
80+
call void @"dx.hl.op..void (i32, <8 x i32>, <8 x i32>, %dx.types.Handle, i32, i32, i32, i32)"(i32 392, <8 x i32> %tmp25, <8 x i32> %tmp29, %dx.types.Handle %tmp32, i32 0, i32 5, i32 3, i32 0), !dbg !37 ; line:67 col:5
8181

8282

8383
%tmp33 = load %struct.RWByteAddressBuffer, %struct.RWByteAddressBuffer* @"\01?rw_matrix_buffer@@3URWByteAddressBuffer@@A", !dbg !38 ; line:77 col:5

‎tools/clang/test/LitDXILValidation/outer-product-accumulate-matrix-layout-failing.ll

Lines changed: 86 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,86 @@
1+
; REQUIRES: dxil-1-9
2+
; RUN: not %dxv %s 2>&1 | FileCheck %s
3+
4+
; Original Source: \tools\clang\test\CodeGenHLSL\linalg\outer-product-accumulate-matrix-layout.hlsl
5+
; The failing tests were generated by manually editing the IR produced from the IR from the passing
6+
; case generated by running the hlsl above (Original Source)
7+
8+
target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64"
9+
target triple = "dxil-ms-dx"
10+
11+
%dx.types.Handle = type { i8* }
12+
%dx.types.ResBind = type { i32, i32, i32, i8 }
13+
%dx.types.ResourceProperties = type { i32, i32 }
14+
%dx.types.ResRet.v8f16 = type { <8 x half>, i32 }
15+
%struct.ByteAddressBuffer = type { i32 }
16+
%struct.RWByteAddressBuffer = type { i32 }
17+
18+
; As noted in other tests, the validation errors come out in
19+
; an order different from the IR. So listed them here in the
20+
; order they appear and added comments for correlation
21+
22+
;CHECK: error: matrix stride must be a constant zero for optimal layouts
23+
;CHECK: error: matrix stride must be a constant zero for optimal layouts
24+
;CHECK-NOT: error: matrix layout value 'OuterProductOptimal' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
25+
;CHECK: error: matrix layout value 'MulOptimal' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
26+
;CHECK: error: matrix layout value 'ColumnMajor' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
27+
;CHECK: error: matrix layout value 'RowMajor' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
28+
; CHECK: Validation failed.
29+
30+
define void @main() {
31+
%1 = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 0, i32 0, i32 0, i8 1 }, i32 0, i1 false) ; CreateHandleFromBinding(bind,index,nonUniformIndex)
32+
%2 = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 1, i32 1, i32 0, i8 0 }, i32 1, i1 false) ; CreateHandleFromBinding(bind,index,nonUniformIndex)
33+
%3 = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind zeroinitializer, i32 0, i1 false) ; CreateHandleFromBinding(bind,index,nonUniformIndex)
34+
%4 = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle %3, %dx.types.ResourceProperties { i32 11, i32 0 }) ; AnnotateHandle(res,props) resource: ByteAddressBuffer
35+
%5 = call %dx.types.ResRet.v8f16 @dx.op.rawBufferVectorLoad.v8f16(i32 303, %dx.types.Handle %4, i32 0, i32 undef, i32 2) ; RawBufferVectorLoad(buf,index,elementOffset,alignment)
36+
%6 = extractvalue %dx.types.ResRet.v8f16 %5, 0
37+
%7 = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle %2, %dx.types.ResourceProperties { i32 11, i32 0 }) ; AnnotateHandle(res,props) resource: ByteAddressBuffer
38+
%8 = call %dx.types.ResRet.v8f16 @dx.op.rawBufferVectorLoad.v8f16(i32 303, %dx.types.Handle %7, i32 0, i32 undef, i32 2) ; RawBufferVectorLoad(buf,index,elementOffset,alignment)
39+
%9 = extractvalue %dx.types.ResRet.v8f16 %8, 0
40+
%10 = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle %1, %dx.types.ResourceProperties { i32 4107, i32 0 }) ; AnnotateHandle(res,props) resource: RWByteAddressBuffer
41+
; error: matrix layout value 'RowMajor' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
42+
call void @dx.op.outerProductAccumulate.v8f16.v8f16(i32 307, <8 x half> %6, <8 x half> %9, %dx.types.Handle %10, i32 0, i32 8, i32 0, i32 0) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
43+
; error: matrix layout value 'ColumnMajor' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
44+
call void @dx.op.outerProductAccumulate.v8f16.v8f16(i32 307, <8 x half> %6, <8 x half> %9, %dx.types.Handle %10, i32 0, i32 8, i32 1, i32 0) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
45+
; matrix layout value 'MulOptimal' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
46+
call void @dx.op.outerProductAccumulate.v8f16.v8f16(i32 307, <8 x half> %6, <8 x half> %9, %dx.types.Handle %10, i32 0, i32 8, i32 2, i32 0) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
47+
; error: matrix stride must be a constant zero for optimal layouts
48+
call void @dx.op.outerProductAccumulate.v8f16.v8f16(i32 307, <8 x half> %6, <8 x half> %9, %dx.types.Handle %10, i32 0, i32 8, i32 3, i32 64) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
49+
; error: matrix stride must be a constant zero for optimal layouts
50+
call void @dx.op.outerProductAccumulate.v8f16.v8f16(i32 307, <8 x half> %6, <8 x half> %9, %dx.types.Handle %10, i32 0, i32 8, i32 3, i32 63) ; OuterProductAccumulate(inputVector1,inputVector2,matrixBuffer,matrixOffset,matrixIntepretation,matrixLayout,matrixStride)
51+
ret void
52+
}
53+
54+
; Function Attrs: nounwind readonly
55+
declare %dx.types.ResRet.v8f16 @dx.op.rawBufferVectorLoad.v8f16(i32, %dx.types.Handle, i32, i32, i32) #0
56+
57+
; Function Attrs: nounwind
58+
declare void @dx.op.outerProductAccumulate.v8f16.v8f16(i32, <8 x half>, <8 x half>, %dx.types.Handle, i32, i32, i32, i32) #1
59+
60+
; Function Attrs: nounwind readnone
61+
declare %dx.types.Handle @dx.op.annotateHandle(i32, %dx.types.Handle, %dx.types.ResourceProperties) #2
62+
63+
; Function Attrs: nounwind readnone
64+
declare %dx.types.Handle @dx.op.createHandleFromBinding(i32, %dx.types.ResBind, i32, i1) #2
65+
66+
attributes #0 = { nounwind readonly }
67+
attributes #1 = { nounwind }
68+
attributes #2 = { nounwind readnone }
69+
70+
!dx.version = !{!0}
71+
!dx.valver = !{!0}
72+
!dx.shaderModel = !{!1}
73+
!dx.resources = !{!2}
74+
!dx.entryPoints = !{!8}
75+
76+
!0 = !{i32 1, i32 9}
77+
!1 = !{!"cs", i32 6, i32 9}
78+
!2 = !{!3, !6, null, null}
79+
!3 = !{!4, !5}
80+
!4 = !{i32 0, %struct.ByteAddressBuffer* undef, !"", i32 0, i32 0, i32 1, i32 11, i32 0, null}
81+
!5 = !{i32 1, %struct.ByteAddressBuffer* undef, !"", i32 0, i32 1, i32 1, i32 11, i32 0, null}
82+
!6 = !{!7}
83+
!7 = !{i32 0, %struct.RWByteAddressBuffer* undef, !"", i32 0, i32 0, i32 1, i32 11, i1 false, i1 false, i1 false, null}
84+
!8 = !{void ()* @main, !"main", null, !2, !9}
85+
!9 = !{i32 0, i64 8598323216, i32 4, !10}
86+
!10 = !{i32 1, i32 1, i32 1}

0 commit comments

Comments
 (0)