https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124165
--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> --- This loop is an O(N*M) loop now. Rather than O(N). One thing that could speed it up slightly is cache the result based on the previous use which might be have the same BB. and bypass it that way. Will that always help here; I don't think so.
