https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122861
--- Comment #1 from Tamar Christina <tnfchris at gcc dot gnu.org> --- Ok, so the compares themselves are all correct, however there's a suspicious change in Perl_pp_split where the codegen changes to do an SVE load instead of an Adv. SIMD load ldr z29, [x5] and to do this the stack is now adjusted VLA adding padding bits for the load. It creates VL + offset, and then loads from offset, so it just loads from inside the padding bits. So something is throwing off the stack adjustments when an SVE subreg is used.
