[MPlayer-dev-eng] [PATCH] Runtime optimization march for x86_64

[Loren Merritt]

On Tue, 17 Oct 2006, Zuxy Meng wrote:

> At least gcc 4.1 has an undocumented march option x86-64 that is an
> intersection of k8 and nocona, hence suitable for those who are
> preparing a binary release.
>
> BTW: Do we really need to check CPU's capability at run time for
> x86-64? It architecturally supports MMX, SSE and SSE2 anyway and the
> only advantage for a run time check is the possibility to make use of
> 3DNow!.

Some SSE2 code is faster than MMX on Intel cpus, but slower on AMD. I 
don't know if there's any of that in MPlayer/lavc yet, but if it's ever 
added we'll want run-time detection even if we know SSE2 is supported.

--Loren Merritt