[llvm] [X86] checkSignTestSetCCCombine - remove early return in CMP(X,0) case (PR #178710)
Phoebe Wang via llvm-commits
llvm-commits at lists.llvm.org
Fri Jan 30 01:05:27 PST 2026
phoebewang wrote:
> @RKSimon I've updated to only peek through SRA per your feedback, but removing the early return is causing test regressions (vec_int_to_fp.ll, masked_load.ll, bmi.ll, etc.).
>
> It seems the existing code was specifically designed to only handle CMP(SRA(...), 0) patterns. Could you provide guidance on the correct approach to enable SHL/SIGN_EXTEND_INREG handling in CMP(X,0) without breaking existing optimizations?
By regressions, are they just test failures or more instructions get generated? You can update these tests to show the problem.
https://github.com/llvm/llvm-project/pull/178710
More information about the llvm-commits
mailing list