Ignacio Vazquez-Abrams (ivazqueznet(a)gmail.com) said:
On Wed, 2009-02-04 at 09:26 -0500, Bill Nottingham wrote:
> Also, -march=i686 was a win over -march=i586, in general, in testing
> across Atom, Core2Duo, and a Athlon64, although not a particularly
> significant one.
How do they compare versus -march=i386?
Didn't test, on the assumption that:
- keeping around support for i386/i486 is pointless
- if emitting i586/i686 instructions was slower... gcc wouldn't do it
Bill