[FFmpeg-devel-irc] IRC log for 2010-04-17

[irc at mansr.com]

[00:07:40] <bcoudurier> well
[00:07:50] <bcoudurier> I'm concerned about the code license
[00:07:58] <bcoudurier> it was inspired from the ffmpeg muxer
[00:15:36] <ramiro> the license in the zip file reads: Added by Clavain Technologies Ltd. with approval from CodeShop B.V.: The FFmpeg Output Patch will be released under the LGPL after a code-rework is done.
[00:16:17] <bcoudurier> that is impossible I'm afraid
[00:16:33] <bcoudurier> the file contains code from ffmpeg licensed under LGPL
[00:16:34] <ramiro> even "with approval from CodeShop B.V."?
[00:16:46] <bcoudurier> you cannot ship as creative common
[00:17:05] <bcoudurier> there was a change forbidden in the first place
[00:17:16] <ramiro> i think codeshop's code has no ffmpeg stuff in it, but I didn't check. clavain technologies added ffmpeg stuff.
[00:17:54] <bcoudurier> ah
[00:18:42] <SmkMnstr> im hugely respected in dev community, once i start using ffmpeg its going to catch on fast
[00:19:10] <ramiro> SmkMnstr: what dev community?
[00:19:14] <Dark_Shikari> lol
[00:19:19] <SmkMnstr> jk
[00:20:10] <ramiro> oh, good =)
[00:20:32] <ramiro> I was about to unleash a huge congratu-fucking-lations =)
[00:22:12] <ramiro> bcoudurier: what parts specifically are from ffmpeg on that patch?
[00:23:00] <bcoudurier> have a look
[00:24:09] <bcoudurier> there is some copy-pasted code from utils.c
[00:24:39] <bcoudurier> and movenc.c was copied in the first place then they replaced the code in the functions
[00:25:09] <bcoudurier> anyway I sent a mail to code-shop, if they are not responsible for the patch, I hope they will point me to the right person :>
[00:27:19] <ramiro> ok
[00:27:32] <ramiro> anyways, I'm off again...
[00:28:00] <bcoudurier> bye
[00:39:15] <bcoudurier> hummmm
[00:41:34] <bcoudurier> code ripers
[00:43:25] <bcoudurier> they ripped the mpeg ts muxer code
[00:44:23] <bcoudurier> that's not the most beautiful piece of code in the codebase for sure ;)
[02:05:24] <CIA-81> ffmpeg: conrad * r22896 /trunk/libavcodec/ (9 files in 3 dirs):
[02:05:24] <CIA-81> ffmpeg: vp3: DC-only IDCT
[02:05:24] <CIA-81> ffmpeg: 2-4% faster overall decode
[02:05:25] <CIA-81> ffmpeg: conrad * r22897 /trunk/libavcodec/vp3.c:
[02:05:25] <CIA-81> ffmpeg: vp3: Don't crop if there's a left/top offset, it's wrong
[02:05:25] <CIA-81> ffmpeg: Fixes issue1834
[02:05:25] <CIA-81> ffmpeg: conrad * r22898 /trunk/libavcodec/vp3.c: vp3: Read fps and aspect ratio in the decoder
[02:16:35] <Yuvi> hm, ffmpeg is a bit slower than libtheora on atom
[02:18:22] <Dark_Shikari> probably function selection
[02:18:31] <Dark_Shikari> libtheora probably has less sse code
[02:18:34] <Dark_Shikari> and some sse code is terrible on atom
[02:18:41] <Dark_Shikari> also, multiplies are sloooow
[02:18:49] <Yuvi> well, the sse2 idct is still a lot faster than mmx on atom
[02:19:05] <Dark_Shikari> also, out of order execution issues
[02:19:12] <saintdev> does gcc reorder instructions when you use march=atom?
[02:19:16] <Dark_Shikari> maybe the libtheora stuff is better-scheduled
[02:22:23] <Yuvi> loop filter maybe, sse2 idct seems to have been written assuming no oooe
[02:22:33] <Yuvi> the mmx idct is pretty much identical
[10:38:33] <CIA-81> ffmpeg: conrad * r22899 /trunk/libavcodec/vp3.c: vp3: Use avctx pointer directly
[10:48:38] <mru> lol @ vp8 is 5 more better than vp3
[10:48:41] <mru> mike's blog
[10:54:35] <kierank> I lol'd at this link from the blog: http://www.stickergiant.com/charts-and-graphs_d49.html
[11:01:44] <CIA-81> ffmpeg: cehoyos * r22900 /trunk/libavcodec/huffyuv.c:
[11:01:45] <CIA-81> ffmpeg: Add const to some pointer parameters.
[11:01:45] <CIA-81> ffmpeg: Patch by Michael Kaufmann, hallo michael-kaufmann ch
[11:29:12] <Kovensky> http://27.media.tumblr.com/tumblr_l0z3o57qT31qzou68o1_500.jpg
[11:29:13] <Kovensky> mru: lol
[11:30:00] <mru> lol
[17:25:30] <DonDiego> Yuvi: you around?
[18:08:52] <siretart> hi DonDiego!
[18:45:59] <Yuvi> DonDiego: yo
[18:48:15] <siretart> hi Yuvi
[18:48:24] <Yuvi> hi
[19:42:12] <DonDiego> Yuvi: what's the status with ogg and theora?
[19:44:27] <Yuvi> DonDiego: the crashers are fixed, there's one issue I introduced last night with aspect ratio (encoding from theora gives the wrong one) that I need to look at
[19:47:48] <Yuvi> But thtS pretty minor
[19:48:02] <Yuvi> Other thAn that it should be good
[19:48:25] <DonDiego> ok
[19:48:28] <DonDiego> what about feature work?
[19:48:35] <DonDiego> i kinda lost track
[19:48:43] <DonDiego> do you have some speedups in the queue still?
[19:48:50] <DonDiego> some other features?
[19:48:56] <DonDiego> chained ogg or whatnot?
[19:49:38] <Yuvi> Yes to the speedups, but those are fairly involved (mAybe regression prone)
[19:49:49] <DonDiego> ok
[19:49:52] <DonDiego> anything else?
[19:50:28] <DonDiego> btw, we should look into nicking all the speedups we can get from libtheora and theorarm
[19:50:35] <Yuvi> No features in the works, the chained off patch needed some big reworking to not fail sometimes
[19:51:00] <DonDiego> though i'm only aware of mmx asm in libtheora
[19:51:09] <DonDiego> Yuvi: so nothing is pending?
[19:51:39] <DonDiego> and what about the chained off patch (i haven't followed the ml)
[19:51:43] <DonDiego> ?
[19:52:30] <Yuvi> I have some armv6 optimizations in progress, most of theorarm is for older arms /bad compiler
[19:53:02] <Yuvi> It failed badly if the streams
[19:53:17] <Yuvi> If the chains had different extradata
[19:53:53] <DonDiego> will you get that fixed (soon)?
[19:59:26] <Yuvi> Not soon, I'll probabavly do more optims first
[20:02:51] <DonDiego> ok
[20:02:59] <DonDiego> we're good to go for the release then
[20:03:44] <siretart> \o/
[22:42:21] <Dark_Shikari> mru: you know anything about prefetch on ppc?  does it crash/die on invalid address, or ignore it?
[22:50:24] <astrange> prefetch (dcbt) ignores errors
[22:57:18] <Dark_Shikari> ok