[XviD-devel] PATCH: Per slices rendering

[Alban Bedel]

Hi Christoph Lampert,

on Sun, 7 Jul 2002 18:48:41 +0200 (CEST) you wrote:

> On Sun, 7 Jul 2002, Alban Bedel wrote:
> 
> > 
> > Hi all,
> > 
> > I'm working on MPlayer and recently found that your decoder was now as fast as
> > libavcodec (even faster in some case). So I wanted to add somekind of direct
> > rendering in xvid. I implemented what we call "slices" rendering wich alredy give
> > a little speedup. Comments are welcome.
> 
> Are you sure there is a real difference to XVID_CSP_USER format which was
> inspired by Mplayer, too? Both seem to simply return the pointers to
> internal buffer instead of copying mem.

With XVID_CSP_USER you can access the internal buffer of xvid. Then you need to
copy it to video memory. In this version xvid only copy the modified MB to a
buffer provided by the host app (ussally in video memory) thus avoiding a few
unneded copy. And as the copy append during the decoding the data might be
copied from the L2 cache wich give more speedup. A few benchmaks :

VIDEO:  [DIVX]  640x272  16bpp  29.97 fps  537.4 kbps (65.6 kbyte/s)

./mplayer ~/download/duality.divx -vc xvid -frames 2500 -benchmark -nosound
-dr -noslices : (XVID_CSP_USER)

BENCHMARKs: VC:  22.591s VO:  28.408s A:   0.000s Sys:   1.237s =   52.236s
BENCHMARK%: VC: 43.2472% VO: 54.3840% A:  0.0000% Sys:  2.3688% = 100.0000%

./mplayer ~/download/duality.divx -vc xvid -frames 2500 -benchmark -nosound -dr
(XVID_CSP_EXTERN)

BENCHMARKs: VC:  26.455s VO:  18.898s A:   0.000s Sys:   1.394s =   46.747s
BENCHMARK%: VC: 56.5920% VO: 40.4263% A:  0.0000% Sys:  2.9817% = 100.0000%

./mplayer ~/download/duality.divx -vc ffodivx -frames 2500 -benchmark -nosound -dr
(ffmpeg decder)

BENCHMARKs: VC:  39.825s VO:  18.670s A:   0.000s Sys:   1.293s =   59.788s
BENCHMARK%: VC: 66.6104% VO: 31.2268% A:  0.0000% Sys:  2.1628% = 100.0000%


VIDEO:  [DIVX]  352x240  24bpp  33.00 fps  841.4 kbps (102.7 kbyte/s)

./mplayer test.avi -vc xvid -benchmark -nosound -dr -noslices
(XVID_CSP_USER)

BENCHMARKs: VC:   4.923s VO:   3.901s A:   0.000s Sys:   0.476s =    9.299s
BENCHMARK%: VC: 52.9376% VO: 41.9488% A:  0.0000% Sys:  5.1136% = 100.0000%

./mplayer test.avi -vc xvid -benchmark -nosound -dr
(XVID_CSP_EXTERN)

BENCHMARKs: VC:   5.849s VO:   2.694s A:   0.000s Sys:   0.307s =    8.850s
BENCHMARK%: VC: 66.0886% VO: 30.4439% A:  0.0000% Sys:  3.4675% = 100.0000%

./mplayer test.avi -vc ffodivx -benchmark -nosound -dr
(ffmpeg decder)

BENCHMARKs: VC:   5.459s VO:   2.641s A:   0.000s Sys:   0.330s =    8.429s
BENCHMARK%: VC: 64.7617% VO: 31.3265% A:  0.0000% Sys:  3.9117% = 100.0000%

PS: Note that the ffmeg decoder use a smthg simillar to my XVID_CSP_EXTERN

	Albeu