[Mlir-commits] [mlir] [mlir][ArmSVE] Lower predicate-sized vector.create_masks to whilelt (PR #95531)

[Benjamin Maxwell]

MacDue wrote:

> Left some comments mostly minor, also occurred to me this is the first Vector -> ArmSVE conversion but we're doing it during LLVM conversion. 

It's Vector -> LLVM SVE intrinsics (not high-level ArmSVE operations). I'm not sure if a general Vector -> ArmSVE conversion would make much sense, as I don't think it'd be much more expressive than the vector dialect (and there's no problems needing solving there, e.g. like tile allocation for SME). Also, with some later patches in mind the ArmSVE ops I use have specific requirements (like `+sme` or `+sve2`), so a general conversion pass would be fairly limited in what it could do. 

https://github.com/llvm/llvm-project/pull/95531