[Mlir-commits] [mlir] [mlir][vector] Set InBound for vector read write after peeling (PR #89108)

[llvmlistbot at llvm.org]

ShivaChen wrote:

> > The loop is generated by affine vectorizer and then peeling.
> 
> Let me suggest something else... Could you apply peeling before vectorization and then run the vectorizer on the main loop and optionally on the remainder loop? In that way, masks won't be generated for the main loop and you can decide how to go about the remainder loop (i.e., vectorize it or not...). WDYT?

Yes, that makes sense to me. The issue is that it seems the affine dialect didn't have upstream support for peeling. Is it worth to introduce peeling in affine dialect? Could I ask how to determine which loop to apply vectorization by vectorizer? Sorry for many questions and really appreciate your comments.

https://github.com/llvm/llvm-project/pull/89108