[FFmpeg-devel] Adding AMD OpenVideo Decode acceleration

[Leith Bade]

Hi

I am new to FFmpeg.

I would like to use FFmpeg to drive AMD's new OpenVideo Decode API (which
interops with OpenCL and OpenGL) with H264.

I have been learning the API, but the documentation is very poor (lots of
typos, mistakes, vague or missing descriptions).

As far as I can tell the API's structures are very similar to DXVA, but
simplified a bit. (I have used DXVA docs to fill in the gaps in the OVD
docs).

I tried decoding only I-frames by using h264bistream library, but I am not
very familiar with H264 and the spec is incredibly complex and hard to
follow.

I have not managed to get anything out of OVD except zero filled NV12
buffers :-( (despite the API returning success code for the decode call).

I read that FFmpeg has DXVA support now, and the H264 parser works
correctly.

How much effort would it be to add a custom hardware acceleration module for
H264? I want it to interop with my OpenCL context, and output to a OpenCL
memory buffer rather than a CPU memory buffer (like the default API does).

What files should I look at?

I hope to be one of the first people to get OVD working, and I would be
happy to contribute the module back to FFmpeg.

Just imagine what VLC could do with native OpenCL/OpenGL filters and
effects! (One application would be to port those extreme CPU intensive
deinterlacing algos VLC has to OpenCL to lessen CPU load...)

Thanks,
Leith Bade
leith at leithalweapon.geek.nz