Log in

View Full Version : c't tests alternative patent free codecs

Doom9
29th July 2005, 20:01
Only in the printed edition though.

Basically it comes down to this: Snow seems to play in the same league as today's mainstream codecs when two pass is used (but not really beating anything serious yet), dirac and theora are still far away, and dirac is really slow (though theora is also plenty slow in the high quality mode).
Tommy Carrot
29th July 2005, 21:31
What did they use for snow encoding, mencoder or ffdshow? This is not irrelevant, because ffdshow's implementation is somehow flawed, it outputs clearly inferior quality compared to mencoder.
Doom9
29th July 2005, 21:50
considering that ffdshow doesn't support 2 pass snow encoding, guess what they used ;)
.
.
.
.
tada.. they even provided the mencoder commandlines.
Tommy Carrot
29th July 2005, 22:30
Ahh, you're right, i totally forgot about the lack of 2-pass option in ffdshow. Anyway, i just wanted to spread the word that don't use ffdshow for snow encoding, or you'll get false impression about the codec. :)
hellfred
30th July 2005, 08:46
Hi
I have missed taking a note, in which issue of cīt the last codec test was. Can anybody name the issue, please? I want to read that articel and have to roam some libraries.

Hellfred
Doom9
30th July 2005, 11:23
it's issue 16 (the latest one)

if I finally get all the conditional compilation stuff running the way I want to, I'll have a nice GUI for encoding Snow using mencoder - of course MeGUI already can do it..
hellfred
30th July 2005, 15:47
it's issue 16 (the latest one)

if I finally get all the conditional compilation stuff running the way I want to, I'll have a nice GUI for encoding Snow using mencoder - of course MeGUI already can do it..
I was not asking for the current test of OS codecs, but the one before. The one featuring e.g. x264 and Neros/Athemas h264 implementation. Sorry for not expressing me clearly.

Hellfred

EDIT
Found the issue (http://www.heise.de/ct/05/10/146/) by using google. Lazy me, could have done that right away. Interesting, that one is able to find a cīt magazin article using google, but not by using the search page of the magazinīs web page itself, using the same terms. :rolleyes:
bond
31st July 2005, 16:11
afaik snow is not meant to be patent free
dirac
25th August 2005, 08:22
FYI, Dirac 0.5.3 has just been released, which has better compression quality - I've junked the 'constant quality' encoder control for the moment as it caused a lot of quality instability. (You still set a single number for encoding though).

I know Dirac encoding is really slow at the moment. We've just started looking at speeding the encoder up. I think we should be able to get 3x quite easily, and maybe another factor 1.5-2 by MMX after that. Hopefully that should happen over the next couple of months.

Thomas
PatchWorKs
25th August 2005, 08:28
and maybe another factor 1.5-2 by MMX after that
...as Theora. But why not SSE optimization first ?
dirac
25th August 2005, 09:51
...as Theora. But why not SSE optimization first ?

I'm *very* far from an expert, but I think it's a bit harder with arbitrary motion compensation block sizes, which is what we support at the moment. Doesn't SSE work with 4x32 bit blocks of data? That's quite hard to work with.

As the spec matures we're likely to fix block sizes and then we can provide optimised routines for all specific block sizes rather than optimising generic code. I guess this will be when SSE will really help.

But I could easily be talking rubbish here! I've been focused on making the algorithm and source code work faster, and MMX versus SSE isn't an issue yet. When I upload the faster algorithm we'd really love some help on the assembler side ...

Thomas
Leak
25th August 2005, 11:13
I've been focused on making the algorithm and source code work faster, and MMX versus SSE isn't an issue yet. When I upload the faster algorithm we'd really love some help on the assembler side ...
This page (http://www.agner.org/assem/#0) has a very good document (pentopt.pdf) on the subject...

Makes a very good read IMHO... :)
dirac
25th August 2005, 12:08
Thanks!
PatchWorKs
26th August 2005, 09:46
I think that you can read this too: Optimizing for SSE: A Case Study (http://www.cortstratton.org/articles/OptimizingForSSE.php)

HP's Architectural and Algorithmic Optimizations for Down-Scale Transcoding of Compressed Video (http://www.hpl.hp.com/techreports/2001/HPL-2001-258R1.pdf) seems interesting too.

Another interesting test (already done for Theora) is to try LIBOIL (http://www.schleef.org/liboil/), Library of Optimized Inner Loops...

...and finally keep in touch with the Ogg Vorbis acceleration project (http://homepage3.nifty.com/blacksword/index_e.htm) (aka Lancer, an SSE optimized vorbis[aotuv] encoder) author, Blacksword --> blacksword8192 _at_ hotmail _dot_ com