Open
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
This PR adds support for the com.microsoft::FusedMatMul ONNX contrib operator in MIGraphX, along with parser/verification tests and generated ONNX fixtures to validate transpose/batch-transpose/alpha behaviors.
Changes:
- Implemented a new ONNX parser for
FusedMatMulthat lowers to MIGraphX ops (transpose,dot, optionalmul) with 1-D promotion/squeeze behavior. - Added ONNX parse tests covering
transA,transB,transBatchA,transBatchB,alpha, and an invalid-rank error case. - Added a numerical verify test and updated
gen_onnx.py+ embedded.onnxmodels for the new operator.
Reviewed changes
Copilot reviewed 22 out of 25 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| src/onnx/parse_fused_matmul.cpp | Adds the FusedMatMul parser/lowering logic. |
| test/onnx/gen_onnx.py | Generates ONNX models for FusedMatMul test cases. |
| test/onnx/parse/fused_matmul_*.cpp | Adds parser equivalence tests for various attribute combinations and an error case. |
| test/onnx/verify/fused_matmul_verify_test.cpp | Adds numeric verification of FusedMatMul output vs gold values. |
| test/onnx/fused_matmul_*.onnx | Adds embedded ONNX fixtures used by the new tests. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+119
to
+123
| if(s0.dynamic() or s1.dynamic()) | ||
| { | ||
| MIGRAPHX_THROW("PARSE_FUSEDMATMUL: dynamic inputs not supported"); | ||
| } | ||
|
|
There was a problem hiding this comment.
FusedMatMul currently throws on any dynamic input shapes, even though the underlying ops used here (transpose, unsqueeze, and the op-builder dot which already handles dynamic broadcasting via broadcast_for_dot) support dynamic shapes. This makes FusedMatMul unnecessarily less capable than the existing MatMul parser. Consider removing this restriction and letting dot/transpose handle dynamic shapes (or only rejecting specific unsupported dynamic cases, if any).
Suggested change
| if(s0.dynamic() or s1.dynamic()) | |
| { | |
| MIGRAPHX_THROW("PARSE_FUSEDMATMUL: dynamic inputs not supported"); | |
| } |
Contributor
There was a problem hiding this comment.
+1
Is there an issue you run into if you just apply the same transposes, etc. for the dynamic case?
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #4806 +/- ##
===========================================
- Coverage 92.49% 92.48% -0.01%
===========================================
Files 583 584 +1
Lines 29562 29628 +66
===========================================
+ Hits 27343 27400 +57
- Misses 2219 2228 +9
🚀 New features to boost your workflow:
|
Collaborator
Regressions detected 🔴 |
Collaborator
|
CharlieL7
requested changes
Apr 21, 2026
Comment on lines
+45
to
+46
| // transBatchA : int, for rank-R A, permute [1, 2, ..., R-2, 0, R-1] (default 0) | ||
| // transBatchB : int, for rank-R B, permute [1, 2, ..., R-2, 0, R-1] (default 0) |
Collaborator
There was a problem hiding this comment.
Are we sure that this the transpose that is intended? Reading the spec I would think it's
[0, 1, 2, ..., R-1] -> [R-2, 0, 1, 2, ..., R-3, R-1]. The spec is really imprecise about it however.
| static instruction_ref apply_trans_last_two(const onnx_parser::node_info& info, | ||
| instruction_ref x) | ||
| { | ||
| auto r = x->get_shape().ndim(); |
Contributor
There was a problem hiding this comment.
might be good to assert r >= 2 to help debug if this parser is modified in the future that causes a rank 1 x to sneak in here
Comment on lines
+119
to
+123
| if(s0.dynamic() or s1.dynamic()) | ||
| { | ||
| MIGRAPHX_THROW("PARSE_FUSEDMATMUL: dynamic inputs not supported"); | ||
| } | ||
|
|
Contributor
There was a problem hiding this comment.
+1
Is there an issue you run into if you just apply the same transposes, etc. for the dynamic case?
|
|
||
| return ([node], [m1, m2], [y]) | ||
|
|
||
|
|
Contributor
There was a problem hiding this comment.
There seem to be a few missing scenarios for test case completeness:
- 1D tensors and tensors that need batch broadcasting
- Verify with more variations of attributes (setting transA, the transBatch attrs, etc.)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Motivation
Add fused MatMul operator
Technical Details
Changelog Category
Add a
CHANGELOG.mdentry for any option other thanNot Applicable