[PATCH] D152458: AMDGPU: Move AMDGPUAttributor run earlier
Matt Arsenault via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Fri Jun 9 18:15:45 PDT 2023
arsenm added inline comments.
================
Comment at: llvm/test/CodeGen/AMDGPU/simple-indirect-call.ll:46-48
+; GFX9-NEXT: s_add_u32 flat_scratch_lo, s12, s17
+; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s13, 0
+; GFX9-NEXT: s_add_u32 s0, s0, s17
----------------
Pierre-vh wrote:
> nit: Do you know why there's more registers being used?
> We go up to s31 for the call at the end so I guess the SGPR usage is the same overall, I'm just wondering.
This is what the change description referenced. Since InferAddressSpaces/SROA no longer eliminate the cast and alloca, it doesn't figure out it can turn off all the inputs
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D152458/new/
https://reviews.llvm.org/D152458
More information about the llvm-commits
mailing list