https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64896
--- Comment #7 from Jan Hubicka <hubicka at gcc dot gnu.org> --- Hmm, OK, lets go for extra temporary and hope that copy propagation/NRV will do the trick. I wonder if we don't want to drop the extra alignment on x86_64 local vars. Vectorizer knows how to upgrade and I doubdt the memcpy/memset expanders care too much given that the offset in frame can often be determined.
