> and everything got about 20 times faster. We have never seen such an behaviour before, so it would be great if you could investigate it in more detail.
Maybe a strange bug? Did you use gcc10 or clang, or maybe the microsoft or intel compiler as backend? -d:release is fine, -d:danger can be a bit faster as it turns off all runtime checks. Both is O3 for gcc by default. But specifying --opt:size at the same time may be not that good, do you use -O3 and -Os for gcc at the same time? Most people do not.| ---|--- Try --passC:-flto to for lto, that give you inlining for all procs. I enable generally -march=native also. And you may try PGO, there was a post about it recently on the forum.