[FFmpeg-user] enable the vaapi hardware acceleration decoding with ffmpeg and libx264

[Vinxxe]

On Wed, Oct 17, 2012 at 10:24 AM, Carl Eugen Hoyos <cehoyos at ag.or.at> wrote:
> Vinxxe <vinxxe <at> gmail.com> writes:
>
>> For now I've successfully managed to use a piece of
>> code that play a h264 encoded movie using the vaapi
>> hw acceleration. The problem is that this code put
>> the decoded frame directly to the hw video surface
>> but I want to put the decoded frame to the pixel
>> buffer object.
>
> (Some people claim that this will cost all performance
> gain that you are searching for.)

why? perhaps I'm loosing something in terms of GPU architecture but
let me explain:
my GPU has not local memory, all its memory is shared with the CPU. I
have, CPU side, the pointer to the pixel buffer object, which is on
the GPU side (in terms of virtual address spaces, not in terms of hw
memory, unless I misunderstood the architecture of shared memory). If
I manage to decode the video frame directly to the pixel buffer object
(GPU side) using the dedicated HW through the VAAPI I leverage the CPU
and don't need to copy back the decoded frame because it's already
where I need it (i.e. in the pixel buffer object).
In other words I would be happy (very very happy in fact) if the
avcodec_decode_video2 were able to decode the frame using the vaapi.
Am I right or am I completely wrong, in your opinion?

thanks a lot for your time

V

P.s. I'm downloading the 1.0 release, thank you for the advice