[FFmpeg-devel] VP8 decoder optimization status

[David Conrad]

On Jun 29, 2010, at 12:43 AM, Frank Barchard wrote:

> On Mon, Jun 28, 2010 at 7:09 PM, Jason Garrett-Glaser <darkshikari at gmail.com
>> wrote:
> 
>> 
>> The current top priority for x86 speed is by far and away the Normal
>> loopfilter -- it's something like 60-70%+ of the total time, since
>> we've SIMD-optimized nearly everything else of note.
>> 
> 
> Agreed. For libvpx version its the bottleneck, and its sse2.
> 
> Any thoughts about multithreaded decode?
> 'token partitions' only apply to token decoding, not loop filtering.
> Could it be handled like mpeg slices and/or ffmpeg-mt frame level
> multithreading?

MPEG slices, no: there's unavoidable serial dependencies for intra prediction as well.
Frame level multithreading will work fine, though the discardable probability updates might make it a little tricky.

Random thought for such an implementation: since all the MVs and block modes can be read independently of coeff data, it might be worth it to read them a bit ahead to reduce the locking overhead for MC (if it's significant.)