[FFmpeg-devel] Implementing VDPAU

Gwenole Beauchesne gbeauchesne
Mon Jan 5 14:11:27 CET 2009


On Mon, 5 Jan 2009, compn wrote:

> you will still need a video output to use vdpau. ffmpeg's vdpau is
> limited to the decoder side. similar to xvmc support.

Talking of which, is a model where ffmpeg actually decodes the whole 
stream itself desired? i.e. only have presentation (Vdp queue, 
vaPutSurface(), etc.) in VO?

In the worst case, it could also be possible to get the rendered images 
back. In case someone wants to implement a real decode pipeline 
(gstreamer) in a cheap [programmatically, not at all performance-wise] 
way. Besides, this would also reduce the work needed for other video 

However, the client application shall still have to create the render 
states and other configuration stuff (e.g. profile setup) itself prior to 
calling ffmpeg decode.

In other words, and more specifically for VA API though, I am still 
thinking about a way to split the library further. i.e. have a library 
without any X11 dependency, and another one with the X11 bits. The client 
application will have to link against the X11 (or other display-dependent) 
lib and ffmpeg against the "normal" lib though. That is, have ffmpeg do as 
much as it can with the HW decode API.


More information about the ffmpeg-devel mailing list