[Mlir-commits] [mlir] [mlir][arith] Add more canonicalization and integration tests coverage (PR #92272)

[Andrzej Warzyński]

banach-space wrote:

Just a couple of clarifications.

> Maybe we could work together on the integration testing guidelines proposal, as we seem to want changes in similar directions :)

What I had in mind in my earlier post was merely updating [the MLIR docs](https://mlir.llvm.org/getting_started/TestingGuide/#integration-tests) to clarify what the expected scope of e2e should be.

I think that you want to expand the scope of such tests. I am not convinced that that's the right approach.

> I don't quite agree with this in the context of general testing approaches for MLIR. We can't expect to always have a well-known, or verified formula for all rewrite patterns.

I was referring specifically to `floordivsi` as your motivating example and [this formula/expansion](https://github.com/llvm/llvm-project/blob/e61d6066e2678b1244b1bcbd59a154a82acf1eb8/mlir/lib/Dialect/Arith/Transforms/ExpandOps.cpp#L113-L119). 

https://github.com/llvm/llvm-project/pull/92272