Re: [xine-devel] Re: xine profiling results

[Andreas H. <and...@gm...>]

Hello,

Am Die, 2003-04-15 um 23.52 schrieb Thibaut Mattern:
> Hi,
> 
> On Tuesday 15 April 2003 23:01, Miguel Freitas wrote:
> [...]
> > you have basically two factors that make you feel the seek slower:
> >
> > - waiting audio driver to finish playing before seeking. some drivers
> > (oss/alsa) have commands to flush card's buffers and they are used.
> >
> > - prebuffering of the metronom. this adds 0.33 sec delay to every seek
> > in order to give a time to the pipeline to get full before start playing
> > again. you can change it with XINE_PARAM_METRONOM_PREBUFFER.
> >
> > there might be room for some minor further optimization here. guenter
> > once added a code to xine_stop to discard audio/video buffers before
> > actually trying to stop the engine and that ideia might also be used on
> > seeking case. i would have to check because i'm not sure about it.
> 
> I've a patch that really improves seeking while moving the slider with 
> xine-ui.
> Basically it does 2 things :
> 1) do not wait 0.33 s, and allow a new seek as soon as a frame is decoded. 
> This first frame is detected, marked as first frame, and the video_loop is 
> modified to not discard first frames and display them immediately instead.
> 2) send audio data in smaller chunks to the audio driver
> 
> http://xine.sourceforge.net/tmattern/seeking/xine_lib.patch

I've tested you patch, fine work. I also set
XINE_PARAM_METRONOM_PREBUFFER to 9000. Both improves seeking a lot.

I have a question/idea. Is it necessary to play audio during seeking in
videostreams? I think, when the user pressed the mousebutton and moves
the slider, should we really play sound during that time ?

IMHO the seeking would be much faster and nicer to handle, if xine would
just draw keyframes during seeking.

Andreas Heinchen