[llvm-branch-commits] [llvm] [AArch64] Split large loop dependence masks (PR #153187)

[Benjamin Maxwell via llvm-branch-commits]

MacDue wrote:

Hey, I was going over this patch again and tried to simplify the flow of `LowerLOOP_DEPENDENCE_MASK`, so in a single step it only does one thing (widen, convert to scalable, or split). I have something working here, https://github.com/MacDue/llvm-project/commits/alias-intrinsic-backend-splitting/ (split across two patches), that I think is a little simpler. It also seems to have improved the codegen for some fixed-width cases (but please double-check they're okay).  

Feel free to take or modify the commits :slightly_smiling_face: 

https://github.com/llvm/llvm-project/pull/153187