[FFmpeg-user] [FFmpeg-devel] next Outreachy & GSoC ideas braindump [RFC]

Ronald S. Bultje rsbultje at gmail.com
Thu Aug 4 03:37:17 EEST 2016


Hi,

On Wed, Aug 3, 2016 at 7:48 PM, Ronald S. Bultje <rsbultje at gmail.com> wrote:

> Hi,
>
> On Wed, Aug 3, 2016 at 1:57 PM, Michael Niedermayer <
> michael at niedermayer.cc> wrote:
>
>> Hi all
>>
>> Sun, Jul 31, 2016 at 10:01:00PM +0200, Michael Niedermayer wrote:
>> > Hi all
>> >
>> > you have a great idea for a Outreachy task or a GSoC task ?
>> > Or something you always wanted to do but never had the time and
>> > it would fit in the time for GSoC/Outreachy ?
>> > Or some feature you always wanted which would fit as a task?
>> > Also keep in mind Outreachy is more flexible and not limited to
>> > coding tasks !
>> > reply and dump it here below:
>>
>> ping ?
>>
>> also cc-ing ffmpeg-user
>>
>> or are all feature requests on trac fake and noone has anything they
>> think should be implemented ?
>>
>> we need ideas for Outreachy, the page is EMPTY:
>> https://trac.ffmpeg.org/wiki/SponsoringPrograms/Outreachy/2016-12
>
>
> I have some ideas. Most are TODOs / extensions to stuff I wrote that I
> never ended up doing myself. Some might be hard.
>
> vp9:
> - avx2 loopfilter functions. Qualification task: explain to me without any
> guidance why it's an issue for AVX2 (or SSE2) that the loopfilter in vp9.c
> works on 8px blocks, and explain how I solved that to work on 16px blocks
> to use SSE2 instructions, and likewise explain how you would extend that to
> work on 32px blocks.
> - coefficient reading optimizations or arithmetic symbol reading
> optimizations. I think this is too hard but it would be an amazing speed
> optimization in vp9 decoding, since this is where pretty much all runtime
> is spent after other optimizations.
> - tile threading. This may actually be a nice qual task for the next block
> since this is quite easy.
>
> frame/slice-mt:
> - implement a mode that allows combining those. Qual task could be to
> implement vp9 tile threading or something like that. Final deliverable
> would implement this for at least vpN codec and one h.26x codec.
>
> I'll think about others. I'm willing to mentor the above maybe.
>

Another one, also for vp8/9:
- alpha plane support for native decoders. (Thanks to James for remembering
about this, I had forgotten.)

Ronald


More information about the ffmpeg-user mailing list