https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84490
--- Comment #13 from Martin Jambor <jamborm at gcc dot gnu.org> --- (In reply to Richard Biener from comment #12) > Wonder if we can have an update on this? TL;DR: there still seems to be a regression, but smaller and difficult to pin down. The benchmark often goes up and down a bit: https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=134.100.0&plot.1=37.100.0&plot.2=69.100.0&plot.3=248.100.0&plot.4=27.100.0&; Nevertheless, I recently got the following numbers at Ofast and generic march (times, so lower is better): | GCC 7.5 | 100.00% | | GCC 8.3 | 102.07% | | GCC 9.2 | 104.15% | | trunk 17 Feb | 109.33% | | trunk 24 Feb | 104.66% | Zen2 based CPU | GCC 7.5 | 100.00% | | GCC 8.3 | 100.00% | | GCC 9.2 | 110.06% | | trunk 17 Feb | 111.32% | | trunk 24 Feb | 105.66% | OTOH, on zen2 GCC 9 and trunk are 23% better at -Ofast and native over GCC 8 or 7 (but those do not know znver2 and I did not try forcing 256bit vectors).
