[llvm-branch-commits] [llvm] [AArch64][llvm] Ensure svnot_u32_x() intrinsic is optimised to subr (PR #237251)
via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Mon Nov 2 06:54:19 PST 2026
================
@@ -853,6 +853,17 @@ let Predicates = [HasSVE_or_SME] in {
foreach VT = [nxv16i8, nxv8i16, nxv4i32, nxv2i64] in
def : Pat<(VT (vnot VT:$op)), (SUBR_ZI_B_PSEUDO ZPR8:$op, (i32 255), (i32 0))>;
+ // An all-active NOT intrinsic also produces ~op in every lane.
----------------
CarolineConcatto wrote:
Maybe we should use the foreach, like the solution proposed in the line above.
https://github.com/llvm/llvm-project/pull/237251
More information about the llvm-branch-commits
mailing list