https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119092
--- Comment #9 from Alexander Monakov <amonakov at gcc dot gnu.org> --- I don't think it's good to drag ABI into this. It can be up to each backend, like it is today with -mveclibabi. > IMO _iff_ we want to support those the only practical way at the moment is to lower them in the frontend to scalar operations on vector extracts and build up a vector from the results. Why in the front-end specifically, and not at veclower time? Regardless, I doubt anybody will heavily rely on scalar fallback of element-wise built-ins. The important part is whether stuff like __builtin_elementwise_sqrt reliably produces native sqrt instructions without overhead.
