Jakub Jelinek (jakub(a)redhat.com) said:
> Actually, in some testing I did, the fasted was
'-march=i686
> -mtune=generic' (which was ~1% faster than '-march=i586
-mtune=generic'),
> tested across Core2Duo, Athlon64, and Atom.
>
> Twiddling the instruction scheduling (-mtune) invariably made things
> worse, across the board.
Even with -mtune=atom on Atom CPUs?
At the time I tested (early February), I don't think there was a
-mtune=atom.
Bill