[MPlayer-users] mplayer benchmark interpretation

[Reimar Döffinger]

On 28.06.2013, at 20:47, Umar Qureshey <umar at janteq.com> wrote:
> BENCHMARKs: VC:   1.590s VO:  48.035s A:   0.000s Sys:   0.567s =   50.193s
> BENCHMARK%: VC:  3.1679% VO: 95.7016% A:  0.0000% Sys:  1.1304% = 100.0000%
> 
> It looks like VO is taking 95% of the processing.

With threads, the vc value will be far too small, since MPlayer only calculates the time waiting for this processing step, not the amount of CPU time used for it.
However it is correct the 48 seconds were used for filtering (including colour space conversion and possibly scaling) and copying the image fir display.

> Would it be correct to assume that colorspace conversion is taking most of these cycles since it's being done in software?

It's likely, but as said use a profiling tool to get a reliable answer.

> Also, would compiling mplayer with openGL or directfb allow for this colorspace conversion to be offloaded to hardware assuming there is a hw colorspace converter?

I doubt directfb can do it.
Some hardware (like TI platforms) support YUV framebuffers, which -vo fbdev2 (if I remember right) can use, but you will have to configure them manually before starting MPlayer.