[Mlir-commits] [mlir] [mlir][test] Test conversion of TOSA to EmitC via LinAlg (PR #94640)

[Paul Kirth]

ilovepi wrote:

> I pushed a fix for the crash in #94936. I would suggest that you test the script on an actual model, as it should have fixed shapes. That will not run into most of unsupported features seen in the current tests. Feel free to share the input model in the TOSA dialect if things don't work directly.

Thanks for the suggestion. Since this was testing the conversion, from ToSa -> LinAlg -> EmitC, I assumed using the tests for ToSa -> LinAlg would be a good starting point. 

I'm not all that familiar w/ how the models are derived. Do you have a suggestion as to how we should generate a minimal case? On the LLVM side, I'd either start w/ some minimal C code and generate IR or write it by hand. I'm a little out of my depth on the MLIR side of things, and its never been clear to me how the MLIR used for MLGO is derived in the first place.

https://github.com/llvm/llvm-project/pull/94640