[FFmpeg-devel] [PATCH] Add VDA support to ffmpeg

Reimar Döffinger Reimar.Doeffinger at gmx.de
Sat Jul 21 12:04:02 CEST 2012


On Thu, Jul 19, 2012 at 07:47:51AM +0200, Hendrik Leppkes wrote:
> On Thu, Jul 19, 2012 at 1:08 AM, Xunzhen Quan <quanxunzhen at gmail.com> wrote:
> 
> >
> > I think that HWAccels are just used to do some complex mathematical
> > computing acceleration, but VDA is an all-in-one hardware acceleration
> > api, which supply only four functions: create decoder, destroy decoder,
> > decode and flush. So I think that it should be wrapped as a Decoder
> > rather than a HWAccel. And if you think it makes code bload, I suggest
> > to just remove the corresponding HWAccel sinceno one seems to be
> > using it.
> >
> >
> I'm sorry, but this is just totally wrong.
> First of all, the HWAccels are working just fine. I know first-hand that
> the DXVA and VAAPI HWAccels work perfectly for decoding H264 (and other
> codecs), and i've also been told that the VDA HWAccel does also work fine.
> All of these you can see in use in for example VLC (including VDA,
> according to j-b). They are also not doing "complex mathematical
> computing", but high-level bitstream decoding. The codec feeds it full
> frames, and not only some math tasks to do.
> 
> If mplayer does not support using the HWAccel infrastructure, thats
> something you should address with the mplayer developers. Adding a codec
> wrapper just to accommodate players which do not care to implement proper
> HWAccel support is just pure bloat.

While I half-agree, I also strongly disagree in parts: The way HWAccel
is implemented currently leads to massive bloat by every single player
having to re-implement a huge part of the acceleration code (and in
cases when CPU-based filters are used to no benefit at all).
As a secondary effect it leads to FFmpeg not supporting any HWAccel
based acceleration at all.
I very much do believe this is an issue that very much should be
addressed inside FFmpeg (not saying that this patch is the right way
for that though, most likely not). If it leads to less work for MPlayer that
is even more welcome of course, but is not the main point.


More information about the ffmpeg-devel mailing list