[FFmpeg-devel] Adding AMD OpenVideo Decode acceleration

Leith Bade leith
Sun Jan 9 12:21:04 CET 2011


On 9 January 2011 23:47, Vitor Sessak <vitor1001 at gmail.com> wrote:

> On 01/09/2011 10:48 AM, Leith Bade wrote:
>
>> Thanks for that.
>>
>
> You're welcome, but just a detail: please try to follow the ML conventions
> and avoid top-posting.


Sorry about that, but Gmail seems to default to top post...

>
>
>  I copied the DXVA code and modified it to use OpenVideo Decode.
>>
>> I now want to test it but I cannot figure out how you are supposed to use
>> get/reget/release_buffer with hwaccel contexts from the application side.
>>
>> Is there some existing open source application that uses FFmpeg's DXVA
>> hwaccel so I can see how to do this?
>>
>
>
> There is MPlayer VDPAU support at
> http://lists.mplayerhq.hu/pipermail/mplayer-cvslog/2009-February/036464.html
>
> and VLC DXVA support at
>
>
>
> http://git.videolan.org/?p=vlc.git;a=commitdiff;h=670ef981c72d124788cf5862562de6541428e7e6
>
> and
>
>
> http://git.videolan.org/?p=vlc.git;a=commitdiff;h=e0028dbfcdf1671fb4c503336fe10684d3724d51


VLC seems to explain it very well.

One thing I was wondering is why mpegvideo.c:233 alloc_frame_buffer()
expects get_buffer to put a value in pic->data[0], yet
dxva2.c:27 ff_dxva2_get_surface() expects a value in picture->data[3].

If you look in VLC's dxva2.c:508 Get() you see they are forced to set both
data[0] and data[3] to the same surface pointer to keep both ffmpeg
functions happy.

Would it not be better, more consistent, and less confusing for application
developers to change dxva.c to use data[0]?

That way you only need to set data[0].


>
> -Vitor
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at mplayerhq.hu
> https://lists.mplayerhq.hu/mailman/listinfo/ffmpeg-devel
>



More information about the ffmpeg-devel mailing list