[Mlir-commits] [mlir] [mlir] [arith] Combine op for select (PR #234818)

[Renato Golin]

https://github.com/rengolin requested changes to this pull request.

There's too much AI slop in this PR, I'll stop reviewing. There are many unrelated tests to the feature being introduced, but not enough tests about the feature. Comments are random, code assumes too much. Also it's not clear that the code is safe, but it's clear that the LLM generated code that fit a particular pattern, which is worrying.

Note, this isn't about the author (you have done good contributions so far) or the feature (I think it makes sense and would be a good addition to the project).

It's about the PR itself that is just not good enough. I recommend you start over, write it yourself and only use LLMs to make sure you're doing the right thing and writing enough tests.

https://github.com/llvm/llvm-project/pull/234818