[Mplayer-users] What to do before version 1.0
ak at cave.hop.stu.neva.ru
Tue Jun 12 20:51:15 CEST 2001
On Tue, 12 Jun 2001, Arpi wrote:
> > "none, MMX, SSE, 3DNOW, SSE2" just as I said.
> there is *big* speed difference between code compiled for 586 and 686.
> the pipelineing is totally different of these CPUs. same for k5/k6.
This all sounds good in theory, but when I compare two binaries, one
compiled with -march=i586 -mcpu=i586 and the other one with -march=i686
-mcpu=i686, I don't see any speed difference. I'm playing 320x240 mpeg-1
file via Xv on my Celeron-266 system, and here's what I get with both
A: 112.8 ( 112.5) V: 112.5 A-V: 0.004 ct: 0.028 3363 26% 0% 5.8% 52
'top' also shows about 30% load for mplayer, 17% for X in both cases.
Of course extensive tests on different systems are needed (586 vs athlon
binaries on AMD processors etc). But I think all difference will be lost
in the noise, so to say. So Christoph is probably right. And
none/MMX/SSE/3DNOW optimized routines should be decided at runtime (imho).
So, an universal binary might be possible, after all :-)
Mplayer-users mailing list
Mplayer-users at lists.sourceforge.net
More information about the MPlayer-users