[MPlayer-users] DIRECT RENDERING WITH VIDIX

Nick Kurshev nickols_k at mail.ru
Sat Feb 2 09:26:06 CET 2002


Hello, Arpi!

On Fri, 1 Feb 2002 13:41:04 +0200 (CEST) you wrote:

> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> Hi,
> 
> > My benchmarks (for -benchmark -frames 1000)
> > 
> > with direct rendering:
> > BENCHMARKs: V:   8.274s VO:   0.005s A:   0.544s Sys:  21.109s =   29.932s
> > BENCHMARK%: V: 27.6434% VO:  0.0158% A:  1.8187% Sys: 70.5221% = 100.0000%
> > 
> > before (without direct rendering):
> > BENCHMARKs: V:   7.413s VO:   2.244s A:   0.578s Sys:  20.028s =   30.264s
> > BENCHMARK%: V: 24.4960% VO:  7.4144% A:  1.9106% Sys: 66.1790% = 100.0000%
> > 
> > As you can see VO time is ~0 for direct rendering.
> it's not 0, but its value is moved to V: from VO: ...
> 
Not all.
> > (I don't know why it's not 0 - probably it's error in benchmark stuff)
> or scheduler - don't forget it's 10ms accurate only.
> 
O'k


I've done some benchmarks also.

K7 - misconfigured MTRR:
~~~~~~~~~~~~~~~~~~~~~~~~
old stuff (without direct rendering):
BENCHMARKs: V:   6.954s VO:   2.973s A:   0.483s Sys:  19.567s =   29.976s
BENCHMARK%: V: 23.1972% VO:  9.9166% A:  1.6114% Sys: 65.2748% = 100.0000%
new stuff (with direct rendering):
BENCHMARKs: V:  10.975s VO:   0.004s A:   0.485s Sys:  18.533s =   29.996s
BENCHMARK%: V: 36.5868% VO:  0.0121% A:  1.6174% Sys: 61.7837% = 100.0000%


K7 - configured MTRR:
~~~~~~~~~~~~~~~~~~~~~
old stuff (without direct rendering):
BENCHMARKs: V:   6.957s VO:   2.103s A:   0.483s Sys:  20.388s =   29.931s
BENCHMARK%: V: 23.2436% VO:  7.0276% A:  1.6131% Sys: 68.1157% = 100.0000%
new stuff (with direct rendering):
BENCHMARKs: V:   7.794s VO:   0.004s A:   0.490s Sys:  21.645s =   29.932s
BENCHMARK%: V: 26.0385% VO:  0.0117% A:  1.6359% Sys: 72.3139% = 100.0000%

Resume:
As you can see with misconfigured MTRRs old stuff is even faster of new stuff
(I'll add MTRR autoconfiguring into VIDIX).
but with configured MTRR new stuff outperforms old stuff on 20% (include accurate).
(as 6.957+2.103=9.060s vs 7.794s)

Well this result is for K7.
I'm not sure about old cpus (K6, K6-2, PentMMX). There are no MTRR but
there is no L2. Anyway it would be better if someone will test this feature.
(It seems that currently only rage128-radeon owners can do that).

(as "old stuff" can be taken any snapshot from 25 upto 31 jan 2002 as new -
current CVS).

Anyway - this test shows results only for dshow codec (which might be unoptimized
for direct rendering). After hacking of libavcodec will could get better digits.

As for MGA owners - I'm not sure but maybe Arpi or someone other will be able to
adapt mga_vid for direct rendering no dshow codecs?

> 
> A'rpi / Astral & ESP-team
> 
> --
> mailto:arpi at thot.banki.hu
> http://esp-team.scene.hu
> 
> _______________________________________________
> RTFM!!!  http://www.MPlayerHQ.hu/DOCS
> Search:  http://www.MPlayerHQ.hu/cgi-bin/htsearch
> http://mplayerhq.hu/mailman/listinfo/mplayer-users
> 


Best regards! Nick




More information about the MPlayer-users mailing list