https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88922
--- Comment #4 from rguenther at suse dot de <rguenther at suse dot de> --- On Mon, 21 Jan 2019, ebotcazou at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88922 > > --- Comment #3 from Eric Botcazou <ebotcazou at gcc dot gnu.org> --- > > IIRC I had patches to do this. My main target was vector float constants vs > > scalar float constants. > > OK, this particular case (scalar vs vector) would probably be worth the > hassle. Yeah, saves some overhead from vectorization where also both variants are usually accessed at the same time. IIRC my implementation was kind of a hack simply inserting additional constant descriptors for element zero of a vector constant (and thus dependet on getting the vector variant into the descriptor hash first...)
