https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126351
--- Comment #2 from ktkachov at gcc dot gnu.org --- (In reply to Drea Pinski from comment #1) > The non aes flavor has a load. So it depends on that being in the data cache. I suppose so. FWIW this particular variant appears in a popular embedded benchmark that (partially) motivated the GCC CRC pass and I did measure a total 8% lower score from the PMULL sequence vs the table variant.
