Skip to content

[0029] [Main] For OuterProductAccumulate, matrix layout must be outerproductoptimal and matrix stride must be zero #7417

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
damyanp merged 17 commits into from
May 13, 2025
Merged

[0029] [Main] For OuterProductAccumulate, matrix layout must be outerproductoptimal and matrix stride must be zero #7417

Show file tree
Hide file tree
Changes from 13 commits
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
0fa5058
Check if MatrixLayout parameter in OuterPrduct Accumulate is OuterPro…
anupamachandra May 1, 2025
7549a90
Clang Format
anupamachandra May 1, 2025
01aecfe
Simlify matrix layout check and fix typo in validation message matrix…
anupamachandra May 1, 2025
6e6d542
Remove stray CHECK directive
anupamachandra May 1, 2025
84eac73
Fix test with new requirement for outer product accumulate layout and…
anupamachandra May 1, 2025
15c77e1
Fix more tests
anupamachandra May 1, 2025
aeb3754
Fix Test, OPA stride = 0
anupamachandra May 1, 2025
390fdf8
Fix typo in test
anupamachandra May 2, 2025
766ecf5
Updates per review feedback
anupamachandra May 5, 2025
8b2170b
Remove old file
anupamachandra May 5, 2025
baa4376
Add Check for invalid matrix stride, add source correlation strings
anupamachandra May 6, 2025
b6dde46
Fix path
anupamachandra May 6, 2025
c33d080
Move outer-product test location, fix test cmd
anupamachandra May 12, 2025
f69c97b
Update test comments per review feedback
anupamachandra May 12, 2025
f915958
Update error message per review feedback
anupamachandra May 12, 2025
c7aea41
Merge remote-tracking branch 'origin/master' into anupamac/outer-prod…
anupamachandra May 12, 2025
30525f5
Merge from main
anupamachandra May 12, 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
582 changes: 292 additions & 290 deletions docs/DXIL.rst
View file
Open in desktop
Comment thread
damyanp marked this conversation as resolved.

Large diffs are not rendered by default.

1 change: 1 addition & 0 deletions include/dxc/DXIL/DxilConstants.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1609,6 +1609,7 @@ const unsigned kMatVecMulAddIsOutputUnsignedIdx = 15;
// Outer Product Accumulate
const unsigned kOuterProdAccMatrixInterpretation = 5;
const unsigned kOuterProdAccMatrixLayout = 6;
const unsigned kOuterProdAccMatrixStride = 7;

// TODO: add operand index for all the OpCodeClass.
} // namespace OperandIndex
Expand Down
26 changes: 26 additions & 0 deletions lib/DxilValidation/DxilValidation.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1229,6 +1229,32 @@ static void ValidateImmOperandsForOuterProdAcc(CallInst *CI,
{"MatrixLayout"});
return;
}
ConstantInt *ML = cast<ConstantInt>(MatrixLayout);
uint64_t MLValue = ML->getLimitedValue();
if (MLValue !=
static_cast<unsigned>(DXIL::LinalgMatrixLayout::OuterProductOptimal))
ValCtx.EmitInstrFormatError(
CI,
ValidationRule::
InstrLinalgInvalidMatrixLayoutValueForOuterProductAccumulate,
{GetMatrixLayoutStr(MLValue),
GetMatrixLayoutStr(static_cast<unsigned>(
DXIL::LinalgMatrixLayout::OuterProductOptimal))});

llvm::Value *MatrixStride =
CI->getOperand(DXIL::OperandIndex::kOuterProdAccMatrixStride);
if (!llvm::isa<llvm::Constant>(MatrixStride)) {
ValCtx.EmitInstrError(
CI, ValidationRule::InstrLinalgMatrixStrideZeroForOptimalLayouts);
Comment thread
anupamachandra marked this conversation as resolved.
return;
}
ConstantInt *MS = cast<ConstantInt>(MatrixStride);
uint64_t MSValue = MS->getLimitedValue();
if (MSValue != 0) {
ValCtx.EmitInstrError(
CI, ValidationRule::InstrLinalgMatrixStrideZeroForOptimalLayouts);
return;
}
}

// Validate the type-defined mask compared to the store value mask which
Expand Down
2 changes: 1 addition & 1 deletion tools/clang/test/CodeGenDXIL/hlsl/intrinsics/linalg_builtins/check-shader-stages.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,7 +43,7 @@ void UseCoopVec() {
const uint opa_matrix_offset = 0;
const uint opa_matrix_interpretation = 5; /*U32*/
const uint opa_matrix_layout = 3; /*OuterProductOptimal*/
const uint opa_matrix_stride = 64;
const uint opa_matrix_stride = 0;

__builtin_OuterProductAccumulate(input_vector1, input_vector2,
rw_matrix_buffer, opa_matrix_offset, opa_matrix_interpretation,
Expand Down
4 changes: 2 additions & 2 deletions tools/clang/test/CodeGenDXIL/hlsl/intrinsics/linalg_builtins/linalg-builtins.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,12 +58,12 @@ void cs_main()
const uint opa_matrix_offset = 0;
const uint opa_matrix_interpretation = 5; /*U32*/
const uint opa_matrix_layout = 3; /*OuterProductOptimal*/
const uint opa_matrix_stride = 64;
const uint opa_matrix_stride = 0;

// CHECK: %[[MLD2:[^ ]+]] = load %struct.RWByteAddressBuffer, %struct.RWByteAddressBuffer* @"\01?rw_matrix_buffer@@3URWByteAddressBuffer@@A"
// CHECK: %[[MCH2:[^ ]+]] = call %dx.types.Handle @"dx.hl.createhandle..%dx.types.Handle (i32, %struct.RWByteAddressBuffer)"(i32 0, %struct.RWByteAddressBuffer %[[MLD2]])
// 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)
// 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)
// 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)
__builtin_OuterProductAccumulate(input_vector1, input_vector2,
rw_matrix_buffer, opa_matrix_offset, opa_matrix_interpretation,
opa_matrix_layout, opa_matrix_stride);
Expand Down
18 changes: 9 additions & 9 deletions ...t/CodeGenDXIL/hlsl/intrinsics/linalg_builtins/outer-product-accumulate-multioverload.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,8 @@
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=float16_t -DMI=F16 -DML=RowMajor | FileCheck %s --check-prefixes COMMON,DXIL-0
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=float16_t -DMI=F16 -DML=OuterProductOptimal | FileCheck %s --check-prefixes COMMON,DXIL-0
// 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
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=uint -DMI=U8 -DML=OuterProductOptimal | FileCheck %s --check-prefixes COMMON,DXIL-2

// 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
// 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
// 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
// RUN: %dxc -T cs_6_9 %s -enable-16bit-types -DITY=uint -DMI=U8 -DML=OuterProductOptimal -fcgl | FileCheck %s --check-prefixes COMMON,HLOP-2

Expand All @@ -11,12 +11,12 @@ ByteAddressBuffer input_vector_buffer2;
RWByteAddressBuffer matrix_buffer;

// COMMON: define void @main()
// 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)
// 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)
// 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)
// 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)
// 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)
// 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)
// 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)
// 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)
// 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)
// 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)
// 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)
// 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)

enum CompType {
Invalid = 0,
Expand Down Expand Up @@ -63,7 +63,7 @@ void main()
const uint matrix_interpretation = MI;
const uint matrix_layout = ML;
const uint matrix_offset = 0;
const uint matrix_stride = 64;
const uint matrix_stride = 0;

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

Expand Down
28 changes: 28 additions & 0 deletions tools/clang/test/CodeGenDXIL/hlsl/linalg/outer-product-accumulate-matrix-layout.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
// 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

//Source file for the IR in \tools\clang\test\LitDXILValidation\outer-product-accumulate-matrix-layout-failing.ll
//Source file for the IR in \tools\clang\test\LitDXILValidation\outer-product-accumulate-matrix-layout-passing.ll

ByteAddressBuffer input_vector_buffer;
ByteAddressBuffer input_vector_buffer2;
RWByteAddressBuffer matrix_buffer;

#include <dx/linalg.h>

// 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)
using namespace dx::linalg;

[Numthreads(1,1,1)]
[shader("compute")]
void main()
{
vector<half, 8> input_vector1 = input_vector_buffer.Load<vector<half, 8> >(0);
vector<half, 8> input_vector2 = input_vector_buffer2.Load<vector<half, 8> >(0);

const uint matrix_interpretation = DATA_TYPE_FLOAT16;
const uint matrix_layout = ML;
const uint matrix_offset = 0;
const uint matrix_stride = STRIDE;

__builtin_OuterProductAccumulate(input_vector1, input_vector2, matrix_buffer, matrix_offset, matrix_interpretation, matrix_layout, matrix_stride);
}
4 changes: 2 additions & 2 deletions tools/clang/test/DXC/Passes/DxilGen/linalg-builtins.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,8 +76,8 @@ entry:

;CHECK: %[[RWMCH0:[^ ]+]] = call %dx.types.Handle @dx.op.createHandleForLib.struct.RWByteAddressBuffer(i32 160, %struct.RWByteAddressBuffer %[[RWMLD0]]
;CHECK: %[[RWMAH0:[^ ]+]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle %[[RWMCH0]]
;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)
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
;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)
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


%tmp33 = load %struct.RWByteAddressBuffer, %struct.RWByteAddressBuffer* @"\01?rw_matrix_buffer@@3URWByteAddressBuffer@@A", !dbg !38 ; line:77 col:5
Expand Down
84 changes: 84 additions & 0 deletions tools/clang/test/LitDXILValidation/outer-product-accumulate-matrix-layout-failing.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,84 @@
; REQUIRES: dxil-1-9
Comment thread
bob80905 marked this conversation as resolved.
; RUN: not %dxv %s 2>&1 | FileCheck %s

;Original Source: \tools\clang\test\CodeGenHLSL\linalg\outer-product-accumulate-matrix-layout.hlsl

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"
target triple = "dxil-ms-dx"

%dx.types.Handle = type { i8* }
%dx.types.ResBind = type { i32, i32, i32, i8 }
%dx.types.ResourceProperties = type { i32, i32 }
%dx.types.ResRet.v8f16 = type { <8 x half>, i32 }
%struct.ByteAddressBuffer = type { i32 }
%struct.RWByteAddressBuffer = type { i32 }

; As noted in other tests, the validation errors come out in
; an order different from the IR. So listed them here in the
; order they appear and added comments for correlation

;CHECK: error: matrix stride must be zero for optimal layouts
;CHECK: error: matrix stride must be zero for optimal layouts
;CHECK-NOT: error: matrix layout value 'OuterProductOptimal' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
;CHECK: error: matrix layout value 'MulOptimal' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
;CHECK: error: matrix layout value 'ColumnMajor' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
;CHECK: error: matrix layout value 'RowMajor' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
; CHECK: Validation failed.

define void @main() {
%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)
%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)
%3 = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind zeroinitializer, i32 0, i1 false) ; CreateHandleFromBinding(bind,index,nonUniformIndex)
%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
%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)
%6 = extractvalue %dx.types.ResRet.v8f16 %5, 0
%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
%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)
%9 = extractvalue %dx.types.ResRet.v8f16 %8, 0
%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
; error: matrix layout value 'RowMajor' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
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)
; error: matrix layout value 'ColumnMajor' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
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)
; matrix layout value 'MulOptimal' is not valid for outerproductaccumulate, must be 'OuterProductOptimal'
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)
; error: matrix stride must be zero for optimal layouts
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)
; error: matrix stride must be zero for optimal layouts
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)
ret void
}

; Function Attrs: nounwind readonly
declare %dx.types.ResRet.v8f16 @dx.op.rawBufferVectorLoad.v8f16(i32, %dx.types.Handle, i32, i32, i32) #0

; Function Attrs: nounwind
declare void @dx.op.outerProductAccumulate.v8f16.v8f16(i32, <8 x half>, <8 x half>, %dx.types.Handle, i32, i32, i32, i32) #1

; Function Attrs: nounwind readnone
declare %dx.types.Handle @dx.op.annotateHandle(i32, %dx.types.Handle, %dx.types.ResourceProperties) #2

; Function Attrs: nounwind readnone
declare %dx.types.Handle @dx.op.createHandleFromBinding(i32, %dx.types.ResBind, i32, i1) #2

attributes #0 = { nounwind readonly }
attributes #1 = { nounwind }
attributes #2 = { nounwind readnone }

!dx.version = !{!0}
!dx.valver = !{!0}
!dx.shaderModel = !{!1}
!dx.resources = !{!2}
!dx.entryPoints = !{!8}

!0 = !{i32 1, i32 9}
!1 = !{!"cs", i32 6, i32 9}
!2 = !{!3, !6, null, null}
!3 = !{!4, !5}
!4 = !{i32 0, %struct.ByteAddressBuffer* undef, !"", i32 0, i32 0, i32 1, i32 11, i32 0, null}
!5 = !{i32 1, %struct.ByteAddressBuffer* undef, !"", i32 0, i32 1, i32 1, i32 11, i32 0, null}
!6 = !{!7}
!7 = !{i32 0, %struct.RWByteAddressBuffer* undef, !"", i32 0, i32 0, i32 1, i32 11, i1 false, i1 false, i1 false, null}
!8 = !{void ()* @main, !"main", null, !2, !9}
!9 = !{i32 0, i64 8598323216, i32 4, !10}
!10 = !{i32 1, i32 1, i32 1}
Loading