[llvm] [AMDGPU] Fold saddsat(x, -C) to ssubsat(x, C) when C is an inline constant (PR #216592)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Mon Aug 17 03:04:52 PDT 2026
arsenm wrote:
> I tested it with v2i16/v3i16 with constant (-32, -16, -65). The output is already optimal - packed type uses v_pk_add_i16 with clamp, the scalar v_sub_i16 pattern doesnt fire for these packed type.
I mean commit it to the test, not just check that it works
https://github.com/llvm/llvm-project/pull/216592
More information about the llvm-commits
mailing list