Fix fused matmul check/rewrite functions

[bmehta001]

• Patterns now declare _outputs filters to bind intermediate values
• Rewrites use fused.producer() or transposed.producer() instead of scanning .uses() which may pick up other nodes that use x or y
• For ir.Value parameters, use a default of None in case the parameter does not exist
• Attribute extraction updated to use as_float() / as_ints() for type safety
• Since rewrite/check functions will have all ir.Value variables passed in, but they may not use all variables, use **_ to read in unused variables
• Updated docstrings from "by" to "with" for clarity and changed fusedmatmul to matmul where appropriate
• Add more patterns:

1. If Transpose.perm indices are [1:-1, 0, -1] and transBatchA is 0, we can change transBatchA to 1
2. If Transpose.perm indices are [-2, 0:-2, -1] and transBatchA is 1, we can change transBatchA to 0.
3. If Transpose.perm indices are [1:, 0] and transBatchA is 0, we can change transBatchA to 1 and transA to 1- transA
4. If Transpose.perm indices are [-1, 0:-1] and transBatchA is 1, we can change transBatchA to 0 and transA to 1- transA
5. If Transpose.perm indices are [-1, 1:-1, 0] and transBatchA is 1, we can change transA to 1- transA
6. And also do all of 1-5 for transBatchB

• Add tests to make sure above changes work for .producer() and the added conditions related to transBatch
  All tests pass

[codecov]

Codecov Report

Attention: Patch coverage is 70.72368% with 89 lines in your changes missing coverage. Please review.

Project coverage is 70.14%. Comparing base (af452c7) to head (514649f).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
...ewriter/ort_fusions/fused_matmul_rule_sets_test.py	59.67%	74 Missing and 1 partial ⚠️
...ipt/rewriter/ort_fusions/fused_matmul_rule_sets.py	88.13%	6 Missing and 8 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2331      +/-   ##
==========================================
- Coverage   70.23%   70.14%   -0.09%     
==========================================
  Files         197      197              
  Lines       24748    24964     +216     
  Branches     2652     2667      +15     
==========================================
+ Hits        17381    17512     +131     
- Misses       6446     6524      +78     
- Partials      921      928       +7     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Copilot]

Pull Request Overview

This PR fixes the fused matmul check/rewrite functions by updating argument handling, error messaging, and test coverage while extending support for batch transpose operations. Key changes include:

• Updating checks for batch dimensions and transpose operations in the FusedMatMul operator.
• Refactoring rewrite rules to use explicit attribute extraction via fused.producer() and adding new rules for flipping transBatch and trans positions.
• Enhancing the test suite with new models to verify the behavior when fused matmul is combined with intermediate Transpose or MatMul nodes.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
onnxscript/rewriter/ort_fusions/fused_matmul_rule_sets_test.py	Updated tests to cover additional fused matmul cases and introduced new models for batch handling
onnxscript/rewriter/ort_fusions/fused_matmul_rule_sets.py	Refactored rewrite rules with clearer attribute extraction and added new rules for handling transBatch and transpose flipping

Comments suppressed due to low confidence (1)

onnxscript/rewriter/ort_fusions/fused_matmul_rule_sets.py:195

• [nitpick] The variable name 'property' is ambiguous; consider renaming it (e.g., 'transBatchProperty') to improve clarity.

if self._pos == 1:
            property = "transBatchA"
        else:
            property = "transBatchB"

[justinchuby]

Thanks - also feel free to point out things that are unintuitive or make suggestions as you use the onnx ir and rewriter apis. Would love to hear about them!

[Copilot]

Pull Request Overview

This PR enhances fused MatMul rewrite rules by leveraging producer() calls, strong-typed attribute extraction, and adds comprehensive batch-transpose patterns along with corresponding unit tests.

• Introduced get_node/get_kwargs helpers and updated rewrite/check logic to use .producer() and as_float/as_ints.
• Extended transpose‐fusion rules to support flipping batch and matrix transpose flags across six new scenarios.
• Updated tests to pass explicit transA, transB, transBatchA, transBatchB defaults and added models for in-middle and batch-transpose cases.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
fused_matmul_rule_sets.py	Added helper functions; updated pattern checks and rewrites to use producer-based lookup and strong-typed attributes; introduced new batch-transpose rules
fused_matmul_rule_sets_test.py	Expanded _run helper to handle batch‐transpose; added tests for fusion in middle of graph and all batch-transpose scenarios

[Copilot]

Pull Request Overview

This PR fixes the fused matmul check and rewrite functions by updating attribute extraction for type safety, modifying rewrite patterns to use the new fused producer functions, and enhancing unit tests to cover various transposition and batching cases.

• Updated attribute extraction using as_float() and as_ints()
• Revised patterns to explicitly tag intermediate outputs with _outputs
• Added new test cases for handling transposition in models with batch dimensions

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
onnxscript/rewriter/ort_fusions/fused_matmul_rule_sets_test.py	Updates tests to validate batch and transpose conditions with fused matmul rewrites
onnxscript/rewriter/ort_fusions/fused_matmul_rule_sets.py	Adjustments to check/rewrite rules, improved helper functions, and new rewrite patterns for batch and transpose handling