https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116078
--- Comment #5 from Filip Kastl <pheeck at gcc dot gnu.org> --- I've double checked that the slowdown really happens on this commit. It realy does. I've also double checked that the resulting binary is different. I've seen this slowdown on 2 separate Zen 2 machines. I've run perf on binaries compiled with GCCs with and without the commit. The dynamic instruction count is actually higher without the commit but only by ~0.0001%. The static instruction count is almost exactly the same (+-3 instructions). Unless I'm doing something wrong, this is weird. I guess this slowdown could be explained by instructions randomly being organized in way that the Zen 2 microarchitecture likes more? I've heard that this sometimes happens.
