[FFmpeg-devel] Piece of code to parallelize

[Reimar Döffinger]

On 20.08.2013, at 03:24, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Mon, Aug 19, 2013 at 10:08:26PM -0300, LuizGuilherme CM wrote:
>> actually I'm little lost too, but, I assume for all the things we talked,
>> I'll use threads with loops restructuring and data dependences techniques.
>> We talked about FFmpeg were already highly parallelized, and because of
>> that I'm here, I'm spend some time looking into the code and didn't have a
>> good result with it.
>> In case of I can't work out with FFmpeg, I have to find something else.
>> 
>>> Did you check
>>> http://git.videolan.org/?p=ffmpeg.git;a=blob;f=doc/multithreading.txt
>>> ;hb=HEAD<
>> http://git.videolan.org/?p=ffmpeg.git;a=blob;f=doc/multithreading.txt;h=2b992fcbc53325577ad838dcf62519ff38493121;hb=HEAD
>>> 
>>> ?
>>> 
>> 
>> I hadn't checked but I will check it out. Tks
> 
> if you want to add multithreading support to some parts, also take a
> look at
> https://github.com/FFmpeg/FFmpeg/commit/a75529e81ea14c6d7c449e205aef97a0acb2645f
> and
> https://github.com/FFmpeg/FFmpeg/commit/83f97355927f8e5f343fc04039e658cadc5c8b39
> 
> these show the changes that simple codecs and filters needed for
> multithreading support

There is also libswscale that might have some tight loops that might be interesting to optimize stand-alone for practicing, even though I am not sure they are as interesting for us as a project since they can be parallelized through slices in the filter and are rarely the bottleneck (only ARM SIMD would be a major improvement I'd expect...).