https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92549
--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> --- E.g. Agner Fog has in the tables for Atom mov r,r 1uops, latency 1, rec. throughput 1/2, while for xchg r,r 3uops, latency 6, rec. throughput 6. It doesn't look beneficial speed wise then. Though, even say on Skylake-X the tables say mov r32,r32 is 1uops, latency 0-1, rec. thr. 0.25, while xchg r,r is 3uops, latency 2, rec. thr. 1.
