[MPlayer-dev-eng] [PATCH] Optimize vo_macosx

[Nicolas Plourde]

On 2-Jul-06, at 3:40 PM, Alexander Strange wrote:

>
> On Jul 2, 2006, at 3:27 PM, Guillaume POIRIER wrote:
>
>> Hi,
>>
>> On 7/2/06, Alexander Strange <astrange at ithinksw.com> wrote:
>>> This patch simply removes the memcpy calls on the picture data in
>>> vo_macosx, making it 2x faster.
>>> As a side effect, it also disables the OSD rendering; I removed that
>>> code and VFCAP_OSD, but that means this depends on the other patch
>>> (mentioned in a previous thread) that automatically inserts -vf
>>> expand=::::1. If you try it now, it should work normally but you'll
>>> have to put that filter in by hand.
>>
>> Is the two folds speed-up with or without "expand" filter inserted?
>>
>> Guillaume
>
> With. Enabling expand has no noticeable speed difference, which is  
> why I used it instead of adding OpenGL blending.
>
> Old mplayer (-frames 1000 -nosound -benchmark, subtitles off)
> BENCHMARKs: VC:  26.990s VO:   7.770s A:   0.000s Sys:   1.789s =    
> 36.549s
> BENCHMARK%: VC: 73.8463% VO: 21.2585% A:  0.0000% Sys:  4.8952% =  
> 100.0000%
>
> New (same options):
> BENCHMARKs: VC:  27.926s VO:   5.140s A:   0.000s Sys:   1.817s =    
> 34.882s
> BENCHMARK%: VC: 80.0575% VO: 14.7341% A:  0.0000% Sys:  5.2084% =  
> 100.0000%
>
> New (subtitles on and vf expand):
> BENCHMARKs: VC:  28.009s VO:   5.374s A:   0.000s Sys:   1.499s =    
> 34.882s
> BENCHMARK%: VC: 80.2974% VO: 15.4060% A:  0.0000% Sys:  4.2966% =  
> 100.0000%
>
> The numbers look better when you have more than 30 seconds of  
> video, of course.
>

My test result on a ~45 min video. Im not convinced of the gain.

BENCHMARKs: VC: 329.183s VO: 483.574s A:   0.000s Sys:  56.945s =   
869.702s
BENCHMARK%: VC: 37.8501% VO: 55.6022% A:  0.0000% Sys:  6.5477% =  
100.0000%

BENCHMARKs: VC: 324.940s VO: 488.833s A:   0.000s Sys:  55.542s =   
869.315s
BENCHMARK%: VC: 37.3789% VO: 56.2320% A:  0.0000% Sys:  6.3891% =  
100.0000%


====
Nicolas Plourde <nicolas.plourde at gmail.com>