[Mlir-commits] [mlir] [mlir][affine] emit `in_bounds` on `transfer_read`/`write` when statically provable in `affine-super-vectorize` (PR #201180)

[Federico Bruzzone]

FedericoBruzzone wrote:

> > I noticed that this PR has not been assigned to anyone, so since I simply saw that you were assigned to previous PRs on [mlir][affine] I kindly decided to ping you 🫶
> > @ftynse @krzysz00 @bondhugula @AGindinson @banach-space
> 
> I've added some reviewers. @ftynse is our maintainer for Affine, but I can take a look as well if he is busy.
> 
> As a side question, since we don't really get any PRs for this area, what's your use case? I am just interested in learning how SuperVectorize is used by people. Thanks!

Thanks! :D 
I don't have any preference, of course, over who does this review.

My use case is mostly exploratory rather than a single production compiler pipeline. 
I'd like to do a paper (an in-depth empirical study) on how different primitive tensor operations (e.g., matmul) behave across different compilers and architectures. While starting this research, I ran into this problem.
Additionally, I'd like to start contributing systematically to MLIR and LLVM. So I took the opportunity to make this PR.


https://github.com/llvm/llvm-project/pull/201180