https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125388
--- Comment #5 from Matthias Kretz (Vir) <mkretz at gcc dot gnu.org> --- (In reply to Drea Pinski from comment #4) > Yes it is. > Use generic vectors here instead. Let the compiler do its job and let it > decide if to use scalars or vectors here. IOW, resolve this via PR90993? That PR got stuck on the question whether the auto-vectorizer would be allowed to do this (because of FE_INEXACT). It is not clear to me whether the optimization can only be done by a library implementation.
