Ok guys this is a call for meeting in
#spu-medialib wednesday the 2nd of april at 20:00 CEST Duration 2-3 hours tops
Agenda:
-What needs to be done to mplayer / ffmpeg in order to get distributed decoding general or do we need to make it all from scratch ?
-How can mpeg 1-2 be split into multiple tasks
-How can H264(x264) be split into multiple tasks
-What tools do we need
-Who will do what Project structure
In order to have success i think we need to adapt a structured organization to also recruit more developers and inform the masses
My idea for a team layout withing the project is:
-ADM Administrators (responsible for arranging releases , make sure we have tools servers, handle financial if any )
-TRAIN Trainers ( members of this team should be able to train people in how the cell/ps3 works and how to write code for it)
-PR public relations (web,news etc)
-REC recruiters (just get more people on the project and make sure developers have tasks they work on)
-SPU_Team (port known algos to the spu's)
-PPU_Team (modify existing programs to be spu friendly)
Members can participate in more than one team.
I have already payed for spu-medialib.org for the next 1 year and 8 moths so all we need are some redundant servers
If you are interested in joining this project and therby also spu-medialib in general i suggest you sign up by replying to this post
Not sure if you know this, or if this is of any help... but there is already some work done on H264 encoding on PS3, which won some competition IBM once held: http://sourceforge.net/projects/cell-h264/
I've been watching your progress for months with top interest. Shit... this is the reason I purchased a PS3. I've even purchased a book on C and been browsing MIT's multicore programming primer, but, I'm afraid I'll be late to the programming party. This is my first experiance with a compiled language.
Anyways, I'm capable of helping promote things, help with sysadmin tasks, and donate resources on a dedicated server or anything else I could get my hands on.
I will try to join the chat and voice some of my opinions about the topic, but I don't want to get involved with the spu-medialib project yet because I'm simply not in a position to be able to guarantee only the smallest level of commitment on the long-term at the moment. I do have some idea's though from my experience with the MPEG1-decoder I'm writing.
In short, it is my opinion that starting from scratch will yield the best results, but porting small pieces of ffmpeg will yield a result soonest. I honestly do not believe in the ffmpeg source code as a good base to start from in the first place. I can elaborate on that in the IRC channel, if I make it. My own decoder is specifically set-up to cater to heterogeneous multi-core chips like Cell (or e.g. GPGPU processing, which is one of the things I'll be implementing even *before* cell optimization), and it might even be a better starting point *for someone who already knows all about H264*, and can actually *write* a codec instead of *porting it*. Problem is, my code is still in early development (I have MPEG1 I-frame & P-frame decoding working 100%, but no other codecs and no optimization yet), and I'm also unsure about the license I want to attach to it when it is presentable.
I can help people out with some general coding theory or interpretation of the ISO or ITU specs though.
@alanine
I also found that project, but from a quick scan it appears to be dead, and it only implements a very tiny subset of a full decoder. But it might be useful.
yep... unfortunately there are not enough people willing to put in a lot of time into this... cell programming and video codecs is not easy stuff... you need a lot of experience... ideally pros.
unsolo wrote:it is not dead..
working on new more "mature" approach to scaling and csc atm..
What's the best way to keep up with your development progress, other than monitoring IRC 24x7?
I can't help program it, but I'd like to keep up with it, testing with bug reports if welcome. I'm running Ubuntu on a PS3/60, with the spu-medialib version from January.