[FFmpeg-user] Hi, I want to add opencl to ffmpeg, but I don't know how to modify the make file to make the ffmpeg link OpenCL.lib.I add the OpenCL in to libavutil.

[Wei Gao]

Hi Roger and James

Thank you for your reply.

We want to do some optimization on ffmpeg and contribute our code to ffmpeg.

1.Add hwaccel to ffmpeg.exe to optimize the decode process using dxva2,we
have done some work on it,and it has been testing now.It has made some
effort on h264,mpeg2,vc1,wmv3.
2.Add opencl code to optimize filter process for libs,such as
deshake,unsharp,scale.
3.Optimize pipline to make it can switch between cpu and gpu.

So when we complete coding and testing,how can we submit it to ffmpeg?Will
you accept it?

Btw,I want ask a question.I found that the hwaccel is based on dx9,we want
to upgrade it to dx11,will you accept it?

Thanks

Best regards




2012/7/27 James Darnley <james.darnley at gmail.com>

> On 2012-07-27 02:41, Wei Gao wrote:
> > Hi Roger,
> >
> > Thanks.I have fix the problem and run my Opencl kernel.
> >
> > I want to ask a question again.I read the ffmpeg code but I didn't find
> > any  CUDA foot print to optimize ffmpeg.Why don't you use CUDA to
> optimize?
>
> That is because nobody has written any yet.  If the same reasoning from
> x264 also applies to FFmpeg, that is because:
>
> 1 - It is not very useful.  You can't just magically speed up the code
> you run on a CPU by running it on a GPU.  Code designed specifically for
> the GPU may not be any faster or produce any better quality.
>
> 2 - The complexity of writing it makes people run off into the night
> screaming.  Only a team of mighty warriors has managed to produce some
> code to offload x264's lookahead to the GPU.  It has been (at least) 8
> months and it is still a work in progress.
> _______________________________________________
> ffmpeg-user mailing list
> ffmpeg-user at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-user
>