I’ve seen around a lot of talks about porting h264 encoders to CUDA. I need to build some ‘test’ programs to demonstrate (or not) the efficiency of GPUs for such applications. I guess I should focus on a specific part of the encoder. However, I’m not a H264 expert. Which part of the encoder do you think the gpu would be the most efficient ?
I would appreciate if someone could give me some clues, going through H264 could be very long :wacko:
I think the motion prediction has been worked on, try searching google scholar by "H.264/AVC MOTION ESTIMATION IMPLEMENTATION ON
COMPUTE UNIFIED DEVICE ARCHITECTURE (CUDA)" in “ICME 2008”. I’m not sure if the actual encoding has been, though. Range coding on GPUs looks difficult, and if I have time I’m probably going to be reading P G Howards’ papers on fast arithmetic encoding; one of them mentions parallelism…