[PATCH] D95432: AMDGPU/GlobalISel: Combine uniform icmp with one use

[Jay Foad via Phabricator via llvm-commits]

foad added a comment.

I don't understand why this needs to be AMDGPU-specific, and why it only works for uniform values, and why it only works inside G_SELECT and G_BRCOND.

Can't you have a generic combine that simplifies (zext (trunc x)) -> x if the types match and the high bits of x are known to be zero?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D95432/new/

https://reviews.llvm.org/D95432