[Mlir-commits] [mlir] [mlir][tosa] Support variadic batch dimensions in matmul ops (PR #226179)

[llvmlistbot at llvm.org]

llvmorg-github-actions[bot] wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-mlir

Author: Deeptanshu Sekhri (deeptanshusekhri)

<details>
<summary>Changes</summary>

Allow tosa.matmul and tosa.matmul_t operands and results to have rank two or greater. Infer and verify right aligned broadcasting across their leading batch dimensions.

Preserve TOSA 1.0 compatibility through target validation and restrict the 1.1-to-1.0 downgrade to supported rank 3 cases. Extend level, shape inference, verifier, validation, and downgrade coverage.

---
Full diff: https://github.com/llvm/llvm-project/pull/226179.diff


13 Files Affected:

- (modified) mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td (+22-10) 
- (modified) mlir/include/mlir/Dialect/Tosa/IR/TosaTypesBase.td (+8) 
- (modified) mlir/lib/Dialect/Tosa/IR/TosaOps.cpp (+141-112) 
- (modified) mlir/lib/Dialect/Tosa/Transforms/TosaDowngrade1p1To1p0.cpp (+11) 
- (modified) mlir/lib/Dialect/Tosa/Transforms/TosaValidation.cpp (+39-2) 
- (modified) mlir/test/Dialect/Tosa/error_if_check.mlir (+20) 
- (modified) mlir/test/Dialect/Tosa/level_check.mlir (+36) 
- (modified) mlir/test/Dialect/Tosa/tosa-downgrade-1-1-to-1-0.mlir (+44) 
- (modified) mlir/test/Dialect/Tosa/tosa-infer-shapes.mlir (+88) 
- (modified) mlir/test/Dialect/Tosa/tosa-validation-version-1p0-invalid.mlir (+30) 
- (modified) mlir/test/Dialect/Tosa/tosa-validation-version-1p0-valid.mlir (+10) 
- (modified) mlir/test/Dialect/Tosa/tosa-validation-version-1p1-valid.mlir (+40) 
- (modified) mlir/test/Dialect/Tosa/verifier.mlir (+77-4) 


``````````diff
The server is unavailable at this time. Please wait a few minutes before you try again.
``````````

</details>


https://github.com/llvm/llvm-project/pull/226179