https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82981
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Actually, it might be better to verify if the can_widen_mult_without_libcall fails that hmode exists and is exactly half the size of prec, otherwise we could end up with the worst case fallback that can't do overflow. And/or, the PR71289 change could be guarded by precision equal to TYPE_MODE precision and umulv4_optab present for that mode, otherwise MUL_OVERFLOW might be more expensive than the division.
