[MPlayer-users] Re: G400/ATI Xpert200 and G200 performance was: Re: [MPlayer-users] files in ftp/incoming

[Alexander Werth (gmx)]

Am Don, 2001-11-01 um 20.44 schrieb Dariusz Pietrzak:
> and while we're offtopic - what happened to the idea that faq would
> contain some hints about video hardware .. like 'nvidia baad, matrox
> good, mkay?' ? - i.e. i'm going to buy 16M g400, but I could get for the
> same money 32M Ati Xpert2000, which purchase I'm going to regret less?

I would go for the matrox G400. The mga_vid video sink for mplayer is
awsome and matrox in general is a lot of fun on linux. Btw. I can only
get superhigh performance out of mga_vid with YUY2 but not with YV12.
It's about 100% slower with YV12.
-vc divx -vo xv
BENCHMARKs: V:   1.050s VO:  11.213s A:   0.000s Sys:   0.048s =   12.311s
-vc divx -vo xmga
VO: [xmga] 576x432 => 576x432 Packed YUY2 fs vm 
BENCHMARKs: V:   1.205s VO:   5.709s A:   0.000s Sys:   0.057s =    6.972s
-vc divx4 -vo xmga
VO: [xmga] 576x432 => 576x432 Packed YUY2 fs vm 
BENCHMARKs: V:   0.963s VO:   5.742s A:   0.000s Sys:   0.070s =    6.775s
-vc odivx -vx xmga
VO: [xmga] 576x432 => 576x432 Planar YV12 fs vm 
BENCHMARKs: V:   0.678s VO:  12.001s A:   0.000s Sys:   0.042s =   12.722s
-vc odivx -vo xv
VO: [xv] 576x432 => 576x432 Planar YV12 fs vm 
BENCHMARKs: V:   0.549s VO:   7.857s A:   0.000s Sys:   0.042s =    8.447s
Of course I would like to combine the perforemance of mga_vid with
odivx. Is this even possible? Or is this unsupported by the matrox g200?
I have a K6-2 450Mhz and on fast moving scenes only the odivx can keep
up with decoding but is unable to also display a satisfying framerate.
Something about mtrr: this is my /proc/mtrr:
reg00: base=0xe5000000 (3664MB), size=   4MB: write-combining, count=1
reg01: base=0xe5400000 (3668MB), size=   1MB: write-combining, count=1
Is this enough (I can give X 5MB of videoram without disturbing the
marvel drivers or what amount of memory is needed by mga_vid?
There is something in the mga_vid code about YV12 but I can't find
anything that does pixel format conversion. What's the bottleneck there?
Is it possible to render directly into video ram? Why are these memory
copy operations necessary at all?
Alexander Werth