[llvm] [X86][GlobalISel] Enable scalar versions of G_UITOFP and G_FPTOUI (PR #100079)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Wed Sep 11 04:34:34 PDT 2024
arsenm wrote:
> Yes, this is true, thank you for noticing this! Updated version uses the lowering from the helper. The problem with `X86cvtts2Int` is that there are patterns only for vector form in contrast to `fp_to_sint`. We currently struggle with it in GlobalISel.
I think this is fine as-is to make progress, but I think this needs to be cleaned up in a follow up. I would suggest either emitting explicit clamping code and having an x86 combine match the instruction behavior, or introduce another G_* node with the clamping behavior and let other targets lower that with the clamp code
https://github.com/llvm/llvm-project/pull/100079
More information about the llvm-commits
mailing list