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

[Paul Kirth]

ilovepi wrote:

> What's happening is that the `--split-input-file` argument splices the file on the `// -----` lines, runs the passes on each chunk independently and joins the results back together. So multiple chunks generate this attribute independently with the same name. This then fails to parse in the next run line.
> 
> TL;DR: remove the `--split-input-file`arguments.
> 

Thanks. I'll update the patch in a bit. 

> Interestingly the test then runs to line 15 where it crashes. The input to that invocation is [this](https://gist.github.com/simon-camp/8a4bc05eda7d65a6b136944dd032f732). I expected the lowering to fail there as we support neither dynamic shapes nor 0d memrefs in the conversion to EmitC, but the passes should fail with an error message instead of an assertion. I can take a look at this next week.

I guess that may require some of the post processing you mentioned? I'll see if I can add some of those python clean ups you provided in the meantime. We can add some TODO's and remove those as the functionality improves.

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