Added the float lowerings for avg pool with corresponding tests.
Details
Details
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Comment Actions
LGTM, but just a few questions/clarifications.
mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp | ||
---|---|---|
1758 | Why is there a tosa.mul op here? I am assuming the added mulop is expected to be converted to a generic op by the dialect conversion framework. |
Comment Actions
I spoke to Hanhan and it sounded like we will be migrating from TC ops to YML for pooling ops, similar to what we have already done for matmul. I was going to wait for this part of the migration before implementing the integer variants. Is there any plan for when that may happen?
Why is there a tosa.mul op here? I am assuming the added mulop is expected to be converted to a generic op by the dialect conversion framework.
Also if the existing pooling op defs doesnt support what the TOSA pooling op needs, is it worth adding a new op. I dont have a strong preference, just a question.