https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118211
--- Comment #3 from Tamar Christina <tnfchris at gcc dot gnu.org> --- (In reply to Richard Biener from comment #2) > Looks like the alignment peeling enablement for early break didn't go in in > time? That should have fixed this by peeling scalar iterations until 's' is > appropriately aligned (though we should probably use a very low VF to make > that not excessive). The patches are approved but we've held off applying the last one as we noticed the broken multi exit loop profiles would cause performance regressions due to incorrect basic block rotations and we wanted to be sure we could fix them. We have patches for those now that pass regressions and fixes the incorrect bb rotations and will be posting them next week. Since it triggers a lot we did it this way to avoid people reporting lots of regression as with peeling for alignment early break triggers a lot. After it goes in I'll fix the misalignment issue as well and make a gcc 14 back port for that alone.
