[x264-devel] Accelerate motion estimation with nVidia's CUDA

[Roger Dahl]

Hello everyone,

This is my first post. I signed up to the mailing list because I want to 
have a go at accelerating x264's motion estimation system using nVidia's 
CUDA GPU development platform.

I have a couple of smaller CUDA/PTX projects behind me and would like to 
tackle a bigger project. I think ME would be a perfect fit. I'm 
fascinated by the process of "massive parallelizing", that is, 
transforming a problem to run on thousands of threads using an 
architecture such as CUDA that is based on light weight threads and 
program managed L1 cache.

Depending on how successful the ME part is, I would then like to go on 
to tackle the other parts of the x264 encoder that would benefit from 
massive parallelization. My guess is that most or all of the parts of 
x264 that have x86 assembly versions are good candidates. Eventually I 
would like to end up with functionality equivalent to the Badaboom GPU 
encoder, hopefully beating it on speed and quality in the process.

However, though I have been reading about H.264 ME lately, it would 
greatly speed up the development process if I could find someone that is 
very familiar with x264's ME implementation, has an interest in massive 
parallelization and that has the patience to answer lots of silly 
questions over instant messaging. Any takers?

Thanks,

Roger Dahl