saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-27 20:02:15 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Set Transpose Attribute instead for manipulating MatMul Strides (#21927)

Browse source 
### Description
Update DML EP for `FusedMatMul` ORT graph node have TransA/B attribute
set instead of updating the strides.



### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
This commit is contained in:
raoanag 2024-09-19 16:26:20 -07:00 committed by GitHub
parent bd60add8ce
commit 73b5c3354c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 6 additions and 13 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
onnxruntime/core/providers/dml/DmlExecutionProvider/src/Operators/DmlOperatorFusedMatMul.cpp
View file

@ -36,9 +36,9 @@ public:
"Two inputs should have same rank and rank >= 3 if transBatchA or transBatchB is true");
}
auto [sizesA, stridesA] = OperatorHelper::GetFusedMatMulSizesAndStrides(inputShape0, transBatchA, transA);
auto [sizesA, stridesA] = OperatorHelper::GetFusedMatMulSizesAndStrides(inputShape0, transBatchA);
auto [sizesB, stridesB] = OperatorHelper::GetFusedMatMulSizesAndStrides(inputShape1, transBatchB, transB);
auto [sizesB, stridesB] = OperatorHelper::GetFusedMatMulSizesAndStrides(inputShape1, transBatchB);
OperatorHelper::FusedMatMulShapeMapping(sizesA, stridesA, sizesB, stridesB, outputShape);
@ -67,8 +67,8 @@ public:
gemmDesc.BTensor = &inputDescs[1];
gemmDesc.CTensor = nullptr;
gemmDesc.OutputTensor = &outputDescs[0];
gemmDesc.TransA = DML_MATRIX_TRANSFORM_NONE;
gemmDesc.TransB = DML_MATRIX_TRANSFORM_NONE;
gemmDesc.TransA = (transA && inputShape0.size() != 1 ? DML_MATRIX_TRANSFORM_TRANSPOSE : DML_MATRIX_TRANSFORM_NONE);
gemmDesc.TransB = (transB && inputShape1.size() != 1 ? DML_MATRIX_TRANSFORM_TRANSPOSE : DML_MATRIX_TRANSFORM_NONE);
gemmDesc.Alpha = alpha;
gemmDesc.Beta = 0.0f;
gemmDesc.FusedActivation = fusedActivation ? &fusedActivationDmlDesc : nullptr;

8
onnxruntime/core/providers/dml/OperatorAuthorHelper/OperatorHelper.cpp
View file

@ -606,8 +606,7 @@ namespace OperatorHelper
std::pair<std::vector<uint32_t>, std::vector<uint32_t>> GetFusedMatMulSizesAndStrides(
gsl::span<const uint32_t> sizes,
int32_t transBatch,
int32_t transpose)
int32_t transBatch)
{
const uint32_t dimensionCount = gsl::narrow_cast<uint32_t>(sizes.size());
std::vector<uint32_t> newStrides(dimensionCount);
@ -633,11 +632,6 @@ namespace OperatorHelper
std::rotate(newStrides.begin(), newStrides.begin() + 1, newStrides.end() - 1);
}
if (transpose && dimensionCount > 1)
{
std::swap(newStrides[dimensionCount - 2], newStrides[dimensionCount - 1]);
std::swap(newSizes[dimensionCount - 2], newSizes[dimensionCount - 1]);
}
return std::make_pair(newSizes, newStrides);
}

3
onnxruntime/core/providers/dml/OperatorAuthorHelper/OperatorHelper.h
View file

@ -285,8 +285,7 @@ void FusedMatMulShapeMapping(
std::pair<std::vector<uint32_t>, std::vector<uint32_t>> GetFusedMatMulSizesAndStrides(
gsl::span<const uint32_t> sizes,
int32_t transBatch = 0,
int32_t transpose = 0);
int32_t transBatch = 0);
class GetOutputShapeAsInputShapeHelper
{