[PATCH] D141127: AMDGPU/GlobalISel: Widen s1 SGPR constants during regbankselect

Matt Arsenault via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Fri Jan 6 06:48:47 PST 2023


arsenm added inline comments.


================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/bool-legalization.ll:71
 ; WAVE64-NEXT:    s_waitcnt lgkmcnt(0)
-; WAVE64-NEXT:    s_xor_b32 s0, s0, -1
+; WAVE64-NEXT:    s_xor_b32 s0, s0, 1
 ; WAVE64-NEXT:    s_and_b32 s0, s0, 1
----------------
Pierre-vh wrote:
> If I understand correctly, it was interpreted as 1 on all lanes before (so -1 as a 32 bit value) but now we're widening the constant so it's just 1, and the end result is the same?
The next instruction is an s_and_b32 1 to clear the high bits. We're missing simplify known bits anyway


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D141127/new/

https://reviews.llvm.org/D141127



More information about the llvm-commits mailing list