[00:16:58] <kierank> just to double check, the width in av_image_fill_linesizes includes the edges, right? [00:18:20] <BBB> caller includes edges [00:18:47] <kierank> BBB: yeah that's what i meant [00:20:09] <in3xes> svn.ffmpeg.org/soc/ is up? [01:32:59] <BBB> astrange: looking at it more, I think the bigger issue is that p->state is used for two things - to indicate that buffers are needed/available, and to indicate that the thread is waiting for input [01:33:14] <BBB> astrange: the two conflict and overwrite each other, I think they should either be separate or they should be flags [01:33:41] <BBB> separate makes sense since you can then indicate that multiple threads need buffers, which would require it to be a counter (i.e. integer) instead of a flag anyway [01:35:50] <astrange> p-> is per-thread, so the state machine is per-thread too [01:36:06] <astrange> i don't think multiple threads should be stomping on each others' state [01:37:11] <BBB> so I'm supposed to scan neighbouring thread's p-> to see if they need a buffer? [01:38:33] <astrange> intra codecs (the problem here) only call get_buffer once [01:38:44] <astrange> so submit_frame just waits for it after starting the thread [01:39:07] <astrange> that always satisfies the request before returning, so there's never any outstanding [01:39:32] <BBB> for -threads 2, I get at least 3 calls to get_buffer [01:39:39] <mru> what, they insist on using the same buffer for every frame? [01:39:47] <BBB> not surprisingly, the third one blocks [01:39:50] <astrange> once per decode() call [01:39:52] <astrange> not once ever [01:39:55] <mru> oh [01:40:31] <dalias> lovely.. [01:41:34] <astrange> or rather, that's the idea, but obviously it breaks somewhere [01:41:43] <astrange> i'm getting called away again :/ [01:42:04] <BBB> window, latch, open, throw out phone [01:42:05] <BBB> done [01:42:32] <gnafu> s/phone/baby? [01:42:38] <gnafu> (Or was that too far?) [01:42:46] <gnafu> (If so, I apologize.) [01:42:51] <gnafu> (If not, I do not.) [01:43:21] <gnafu> (Define far, I guess.) [01:43:41] <mru> don't throw it too far [01:43:49] <BBB> baby is cute and undisturbing [01:43:54] <BBB> apparently, astrange's phone is not [01:44:11] <gnafu> Most phones are definitely not cute. [01:44:28] <gnafu> What some people do with their phones is most disturbing. [01:44:41] <mru> gnafu clearly has the wrong background image on his phone [01:45:17] * gnafu has a phone that is so old that his background image is actually just a random doodle he created with the built-in drawing application. [01:46:33] <gnafu> But soon, I'll have an Android phone, and perhaps I will locate a suitably cute background image. [01:46:42] <gnafu> But preferably not disturbing. [01:48:15] * BBB has a babypicture as background picture [01:49:45] <gnafu> I suppose I should have a picture from my wedding or something until I have a baby. [01:50:01] <gnafu> I think I look cute in my purple bowtie. [01:51:39] <gnafu> But perhaps a bit disturbing if I have a picture of /just/ me for a background. [01:52:26] <mru> that might seem a little narcissistic [02:13:46] <dalias> haha [02:30:38] <BBB> astrange: if submit_packet() is supposed to ensure that a buffer is provided to each thread, then it's not doing that for any thread. problem is that (even though I think your patch 1/2 is supposed to prevent that), ff_thread_finish_setup() sets p->state to STATE_FINISHED_SETUP, causing the thread to pass the get_buffer stage. or in other words, get_buffer is never actually called, ever [02:30:59] <BBB> astrange: you probably know when ff_thread_finish_setup is supposed to be called and when not, so please have a look there [02:42:14] <BBB> astrange: in fact, if I remove the ! (so && avctx->thread_safe_callbacks instead of && !avctx->thread_safe_callbacks) in your first hunk of patch 1/2, it works correctly [02:42:46] <BBB> astrange: looks like a typo to me anyway, I don't see why you'd want to run ff_thread_finish_setup() there if the get_buffer() callback is _not_ threadsafe [02:44:31] <BBB> astrange: that said, my wife wants me to sleep, so please submit a new patch and let's keep this stuff rolling :) [02:48:29] <in3xes> Where is the SoC repo which has JPEG2000 ? [02:48:54] <in3xes> pengvado: Any idea? [02:53:20] <gnafu> Would it be the one mentioned here?: [02:53:23] <gnafu> http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/2011-February/107863.html [02:54:16] <kierank> lol jpeg2000 [02:56:21] <gnafu> kierank: I've been getting that impression lately. [02:57:12] <kierank> we really need an xjpeg2000 [02:57:17] <kierank> all the encoders are awful [02:57:47] <mru> why do we need jpeg2k at all? [02:57:52] <Dark_Shikari> what mru said [02:58:10] <kierank> http://code.google.com/p/opencinematools/ [02:58:17] <kierank> that's why [03:03:19] <in3xes> How difficult is writing a demuxer? I mean, considering SoC qualification task. [03:04:13] <Dark_Shikari> depends on how complicated the format is [03:04:39] <in3xes> Okay [03:05:28] <kierank> in3xes: there's a guide on the wiki [03:05:56] <in3xes> kierank: I am already following that [03:06:57] <in3xes> I just started with VIVO. I'll be lucky if I am able to finished by weekend. [03:07:13] <in3xes> finish* [03:08:06] <in3xes> kierank: Thanks for pointing, anyway :) [06:01:49] * elenril pokes BBB [07:48:55] <Tjoppen> low pressure on the ml. only 20 new mail in a day [07:52:03] <kierank> someone didn't learn what top-posting is... [07:54:03] <spaam> kierank: tell him about it [07:54:18] <Tjoppen> yeah, I saw that. hilarious [07:54:21] <kierank> I think it's a she spaam [07:56:00] <spaam> kierank: its tru.. [07:56:04] <spaam> true [07:56:11] <spaam> i found her on facebook.. [07:57:13] <kierank> .... [07:57:28] <kierank> people like you are why i don't do social networking [07:59:03] <spaam> come on.. you can change the privacy settings so only your friends can see your profile.. [08:00:15] <spaam> i changed my things to that after i found my mother on facebook... [08:00:29] <Dark_Shikari> quite easily [08:00:33] <Dark_Shikari> click "delete account" [08:00:33] <Dark_Shikari> bam [08:00:36] <Dark_Shikari> your profile is now private. [08:01:08] <kierank> AND NOTHING OF VALUE WAS LOST [08:01:13] <Dark_Shikari> ^ [08:01:32] <kierank> except maybe the x264 facebook group [08:01:39] <Dark_Shikari> >facebook group [08:01:39] <spaam> awww [08:01:40] <Dark_Shikari> >value [08:02:01] <spaam> facebook is nice.. [08:02:07] <Dark_Shikari> nice... [08:02:10] <Dark_Shikari> FOR ME TO POOP ON. [08:02:25] <peloverde> didn't you used to work for facebook? [08:02:39] <spaam> Dark_Shikari: you know.. even KotH have facebook ;) [08:03:06] <Dark_Shikari> peloverde: yes [08:04:00] <cartman> moin [08:04:02] <spaam> cartman: BBB did miss you last night.. [08:05:55] <cartman> spaam: did he found the bug? [08:06:30] <spaam> he was asking you "any progress". [08:06:46] <cartman> spaam: you should have handed him a void* [08:13:29] <TheFluff> I have facebook because it's a convenient way to get invited to parties [08:13:39] <TheFluff> if you set it to the most antisocial settings possible it's not too bad [08:14:18] <Dark_Shikari> Er, why would I want to get invited to parties? [08:14:30] <TheFluff> becau- [08:14:33] <TheFluff> oh wait, right, nerd. [08:14:35] <TheFluff> never mind. [08:14:46] <Dark_Shikari> Parties are just a place for a bunch of stupid frat boys to get drunk and do stupid things. [08:14:52] <Dark_Shikari> While listening to very bad music. [08:15:10] <peloverde> i think in the movie zuck said that they are fun and lead to a better life [08:15:23] <Dark_Shikari> He's obviously never been to a party [08:15:25] <TheFluff> either you'll growp to be a unix beard or you'll understand when you get older, young man [08:15:26] <saintdev> you actually watched that? [08:15:34] <TheFluff> grow up* [08:15:50] * thresh likes parties [08:15:53] <Dark_Shikari> sorry, I don't subscribe to the "grown-ups aren't allowed to have fun" rule [08:15:57] <spaam> Dark_Shikari: come on.. you can have a party with your friends and play with your things.. [08:16:03] <Dark_Shikari> LAN parties aren't really parties. [08:16:36] <kshishkov> thresh: вечеринки, светские рауты, корпоративы или пьянки? [08:16:37] <peloverde> it was an awesome movie [08:16:38] <TheFluff> have you ever been to breakpoint or any of the other big european hacker cons? [08:16:43] <Dark_Shikari> Those aren't parties, those are cons [08:16:55] <thresh> kshishkov: вечеринки, плавно переходящие в пьянки [08:17:03] <Dark_Shikari> a party has ~10-200 people, usually at a college dorm, drinking while listening to loud rap music and bad techno. [08:17:06] <TheFluff> call them whatever you want but there's a lot of very loud music and alcohol at them [08:17:16] <TheFluff> that's a pretty narrow definition of a party [08:17:20] <saintdev> <Dark_Shikari> Parties are just a place for a bunch of stupid frat boys to get drunk and do stupid things. <-- Those aren't parties, those are just frat houses [08:17:24] <Dark_Shikari> that seems to be about 90% of the parties out there [08:17:27] <thresh> hmm, that's american definition of parties [08:17:35] <Dark_Shikari> Yes, ok, american parties are crap. [08:17:42] <Dark_Shikari> Can't judge european ones. [08:17:50] <Dark_Shikari> (We don't seem to have those awesome hacker cons here for some reason) [08:17:50] <kshishkov> try Russian [08:17:53] <TheFluff> college dorm parties suck here too [08:18:03] <TheFluff> no surprise there [08:18:11] <kshishkov> Dark_Shikari: because everybody knows what happens to hackers coming to USA [08:18:15] <Dark_Shikari> `-` true [08:18:22] <TheFluff> if you've only been to those it's not that strange you hate parties [08:18:27] <TheFluff> europe parties harder than the US [08:18:43] <TheFluff> lasers <3 [08:19:04] <TheFluff> anyway go to breakpoint sometime [08:19:32] <TheFluff> or, well, breakpoint doesn't really exist anymore but there are successors [08:19:39] <TheFluff> lots of cool people, awesome demos, awesome music, etc etc [08:19:57] <cartman> Does code.google.com works for anyone? Gives 500 here. [08:20:08] <TheFluff> dreamhack in sweden is p cool too but there are too many gamers and not enough hackers [08:20:11] <andoma> cartman: works fine for me [08:20:14] <TheFluff> also you can't drink at dreamhack [08:20:19] <cartman> andoma: thanks :/ [08:20:25] <Dark_Shikari> TheFluff: sounds like a minus and a plus [08:20:33] <kshishkov> TheFluff: that's what I like in Sweden [08:20:35] <Dark_Shikari> probably the minus is worse though. [08:20:49] <votz> TheFluff: they didnt let people drink at CPL events either [08:21:10] <votz> everyone just went to the hotel bars just outside, though :) [08:22:03] <TheFluff> at last breakpoint you could just go to the shop inside and ask for "apple juice" and you'd get a beer [08:22:12] <TheFluff> iirc [08:22:24] <votz> some 'upgraded juice' [08:22:31] <TheFluff> :V [08:22:35] * thresh asks for a beer and gets beer usually [08:24:01] <andoma> mm beer [08:24:06] <kshishkov> thresh: well, _you_ call it beer [08:24:15] <Tjoppen> TheFluff: unfortunately dreamhack treats the sceners like shit [08:24:19] <TheFluff> yeah [08:24:21] <TheFluff> I know [08:24:23] <kshishkov> thresh: I'm not sure that even Danish would drink that [08:24:31] <thresh> kshishkov: I don't drink Russian beer, if you mean that :) [08:24:36] <TheFluff> they've been working on making creative better these last few years though [08:24:37] <thresh> well Russian "beer" [08:24:41] <TheFluff> or at least trying [08:24:45] <Tjoppen> "here, you can sit right where the people walk, right next to the nacho cheese guy" [08:24:51] <andoma> hahaha [08:25:12] <andoma> are there any decent demo partys in the nordics anymore? [08:25:20] <TheFluff> assembly in finland [08:25:23] <andoma> ah [08:25:27] <kshishkov> thresh: well, have tried comparing the same beer brand bought in Russia and somewhere in Andorra for example? [08:25:36] <TheFluff> is said to be good, I've never been there though [08:25:47] <thresh> kshishkov: yeah, imported bottles == same stuff [08:25:49] <andoma> i only went to The Part and The Gathering back in the days (early 90ties) [08:25:55] <andoma> The Party... even [08:26:06] <TheFluff> Tjoppen: I was at the creative section at DH, uh, last DH winter? [08:26:07] <TheFluff> I think [08:26:10] <Tjoppen> yes, assembly was good. going there again this summer probably [08:26:14] <TheFluff> I got third in the useless utility competition [08:26:28] <Tjoppen> we were there.. hm [08:26:42] <Tjoppen> we won the useless utility when we were there [08:26:55] <TheFluff> were you the guys with the windows achievements? [08:26:59] <Tjoppen> yes :) [08:27:01] <TheFluff> :D [08:27:07] <TheFluff> mine was the cuckoo clock [08:27:15] <TheFluff> the video was so shitty you couldn't see what it was doing though [08:27:16] <Tjoppen> also placed third with our partyprod [08:28:23] <kshishkov> TheFluff: http://www.oldskool.org/pc/8088_Corruption [08:28:29] <Tjoppen> for assembly I'm trying to cobble together bits and pieces for an atari 2600 prod, and an entry in the game competition on the same platform :) [08:28:34] <TheFluff> (it announced whole hours by ejecting the cd sled, going "cuckoo" and retracting it again) [08:29:24] <TheFluff> Tjoppen: you don't happen to know a guy known as kalaspuff, do you? [08:29:31] <Tjoppen> nope [08:29:32] <TheFluff> (his last name's aaro, if it helps) [08:29:35] <TheFluff> mmk [08:30:01] <TheFluff> he seems to know almost everyone, figured he might know you too [08:30:05] <TheFluff> (we work together) [08:30:51] <Tjoppen> we haven't been doing demos for long. well, on and off a bit over the years. probably should at least get an account on pouet [08:31:28] <TheFluff> I don't really do demos at all [08:31:33] <TheFluff> he does, though [08:31:42] <TheFluff> I just like watching :V [08:31:46] <Tjoppen> hangaround? :) [08:32:01] <Tjoppen> brb, meeting [08:32:33] <TheFluff> yep [08:44:39] <pross-au> kshishkov: https://github.com/clone2727/scummvm/compare/ra-smush [08:46:14] <kshishkov> pross-au: yep, seen it [08:46:26] <kshishkov> pross-au: I even lol'd at http://www.xvid.org/News.64.0.html?&cHash=d69f350538&tx_ttnews[backPid]=64&tx_ttnews[tt_news]=9 [08:47:45] <pJok> god morgon [08:47:56] <kshishkov> goda morgnar, pJok [08:49:08] <pross-au> kshishkov: Art is never finished, only abandoned [08:49:47] <saintdev> pross-au: unless you're george lucas [08:50:10] <kshishkov> saintdev: well, jar-jar was a masterpiece [08:51:46] <saintdev> well things no longer sound horrible, yipee [08:52:33] <kshishkov> even with ffaacend? [08:52:38] <kshishkov> *ffaacenc [08:53:06] <pJok> ffaacenc got fixed? o_O [08:53:26] <saintdev> pJok: of course not [08:53:34] <saintdev> just not as horrible as it sounded last night [08:53:38] <pJok> hehe [08:53:46] <pJok> well, thare are various degrees of fixed [08:54:22] <saintdev> ok, then fixed my stupid mistakes. [08:54:32] <saintdev> or wait, kshishkov said i could blame him. [08:54:48] <saintdev> s/my/kshishkov's/ [08:55:09] <kshishkov> exactly [08:57:41] <pJok> there* [08:57:46] <pJok> time to do some hp [08:57:48] <pJok> php* [09:09:59] <saintdev> hmm, my RAM seems to have asploded :/ [09:10:45] <elenril> http://tvtropes.org/pmwiki/pmwiki.php/Main/MadeOfExplodium ? [09:11:07] <saintdev> quite possibly [10:03:14] <cartman> BBB: I am bisecting atm. Should be over today. [11:58:21] <cartman> BBB: bisect finished, bug updated [12:00:03] <kshishkov> i.e. "this patch will make clangs demonstrate this bug more reliably and harder to fix"? [12:00:31] <kshishkov> cartman: also judging from last av500's photos he's on vacation [12:00:33] <cartman> well bisect shows that so called "TBAA" optimization pass broke code [12:00:36] <cartman> kshishkov: yeah :D [12:03:35] <Dark_Shikari> TBAA? [12:03:56] <kshishkov> "take bytes and alter" [12:04:21] <cartman> Type Based Alias Analysis [12:14:55] <Dark_Shikari> ALIASING! [12:14:56] <Dark_Shikari> I knew it. [12:15:08] <cartman> you better tell us next time... [12:15:36] <cartman> 1 hour 55 minutes of computing power lost while bisecting :P [12:17:05] <iive> cartman: i told you. but nobody ever listens to me. [12:17:47] <cartman> iive: I somehow missed it then, sorry! :) [12:22:10] <Dark_Shikari> I was told that Clang didn't do aliasing optimizations yet [12:22:31] <cartman> Dark_Shikari: yeah its newly enabled, in December 2010 [12:22:40] <cartman> and looks like still buggy [12:23:34] <kshishkov> yep, old-fashioned GCC bugs are better [12:23:53] <cartman> gcc 4.6 has the lowest regression count ever, its good news :> [12:24:08] <cartman> >> Still 119 regressions in 4.3, 116 in 4.4, 122 in 4.5 and 93 in 4.6 [12:24:54] <iive> cartman: give it some time ;) [12:28:14] <mru> cartman: of course the regressions go down [12:28:27] <mru> with each version, there are less things left working [12:28:42] <mru> break the same percentage each time, and the total count will go down [12:29:09] <cartman> mru: lol :) [12:35:53] <BBB> elenril: sprry, forgot, hopefully today, past 3 days were busy at work [12:36:41] <kshishkov> BBB: so that's why nobody committed bink-b audio patches? [12:37:24] <BBB> am I nobody? [12:37:42] <cartman> at least, nobody is perfect [12:37:55] * BBB awaits astrange [12:38:03] <BBB> btw thanks cartman :) [12:38:20] <cartman> ah np, lets see what happens now [12:40:20] <kshishkov> BBB: you belong there too, it seems - http://patches.ffmpeg.org/patch/1218/ for instance [12:41:20] <BBB> ok ok I get the hint [12:41:30] <BBB> were they all OK? [12:41:53] <kshishkov> yep [12:43:17] <BBB> kshishkov: ok will look at it at work [13:56:27] <siretart> kshishkov: sorry, I didn't get to your patches yesterday, I'm basically drowning in day-job work :-( - and the weekend doesn't look much better [13:57:04] <siretart> kshishkov: moreover, my emacs crashed here, so I lost your link(s) from yesteday :-( [14:02:02] <Tjoppen> why doesn't lavc adjust qscale by default for cbr mpeg2video output? [14:02:19] <Tjoppen> I was trying to simplify http://www.itbroadcastanddigitalcinema.com/ffmpeg_howto.html#Encoding_D10 [14:02:35] <Tjoppen> especially having to set qmax=3 to not get crap output seems.. arbitrary [14:03:56] <kshishkov> siretart: okay :( [14:04:38] <Dark_Shikari> setting a qmax like that will cause vbv underflow [14:04:52] <Dark_Shikari> you might as well not use maxrate/bufsize if you're going to do that [14:05:30] <Dark_Shikari> your problem is likely that lavc's vbv ratecontrol is a pile of ass and terrible [14:05:41] <Dark_Shikari> I wouldn't recommend using lavc mpeg2 for anything serious that requires vbv [14:06:12] <mru> how hard would it be to make x264 encode mpeg2? [14:06:13] <Tjoppen> I see [14:06:16] <Dark_Shikari> mru: it's being done [14:06:22] <Dark_Shikari> x262 already does I and P-frames [14:06:23] <Tjoppen> x262= [14:06:29] <Tjoppen> :) [14:06:29] <mru> nice [14:06:32] <mru> didn't know that [14:06:39] <mru> who's doing that? [14:06:44] <Dark_Shikari> ifb and kierank [14:06:47] <Dark_Shikari> for the open broadcast encoder [14:06:57] <Dark_Shikari> it's going to be done faster than xvp8 which will make me laugh. [14:07:16] <elenril> ETA until we remove all encoders, rename lavc into lavd and x264 into lave [14:07:17] <mru> mpeg2 is a relatively simple coded [14:08:47] <Dark_Shikari> the interlaced mode will still be messy to do though [14:15:09] <Tjoppen> feels like getting the current encoder not to act completely retarded "shouldn't be that hard" [14:16:11] <Tjoppen> I did have a look at the mpeg internals to try and figure out what it does, but.. my eyes [14:16:21] <superdump> :) [14:16:55] * mru experiments with the iar compiler [14:17:24] <mru> only one ICE building ffmpeg [14:17:27] <mru> I'm impressed [14:17:30] <kshishkov> which arch? [14:17:35] <mru> arm [14:18:07] <superdump> Dark_Shikari: is x262 going to get merged into the x264 tree when it's done? [14:24:15] <pJok> Dark_Shikari, will it actually get an mpeg audio layer 2 encoder with predictive output maximum peak? [14:46:48] <Kovensky> 11:07.16 elenril: ETA until we remove all encoders, rename lavc into lavd and x264 into lave <-- "lave" is the imperative form of "to wash" in portuguese [14:47:34] <kshishkov> Kovensky: we have some time till xbink [14:47:51] <kshishkov> Kovensky: and that's definitely not the worst scenario [14:48:00] <Kovensky> lul [14:53:54] <mru> crappy compiler can't align struct members [14:59:39] <kshishkov> s/crappy/professional(or enterprise)/ [15:00:09] <mru> quite a few can do it [15:00:37] <mru> the amusing ones are the ones that can align data objects but not struct members [15:00:40] <mru> iar is one of those [15:01:37] <kshishkov> aligning is always funny [15:01:49] <kshishkov> especially when it works selctively [15:02:07] * cartman notes that iar is pricey too [15:02:07] <kshishkov> i.e. not on stack or not in structures [15:02:16] <mru> the TI compilers are a bit obscure [15:02:28] <kshishkov> GCC too [15:02:32] <mru> they can do all kinds of alignment but the syntax differs wildly [15:03:54] <Kovensky> binutils can't align rodata [15:03:57] <Kovensky> windows binutils that is [15:04:58] * mru builds w/o asm and watches fate errors instead [15:05:14] <cartman> you are lucky that binutils run on Windows [15:05:23] <mru> I'm not on windows [15:05:32] <mru> iar compiler runs fine under wine [15:05:34] <cartman> mru: that was for Kovensky [15:05:53] <cartman> iar under wine eeeeeeeeek [15:06:05] <mru> what's wrong with that? [15:06:12] <cartman> mru: I never trust wine :P [15:06:15] <bilboed-pi> works perfectly fine [15:06:16] <KotH> mru: apropos iar, do you have experience with iar for arm? [15:06:26] <mru> KotH: getting some now [15:06:30] <mru> and it ain't good [15:06:33] <cartman> just use a VM [15:06:35] <KotH> :-) [15:07:05] <KotH> mru: iar got dropped here because it miscompiled stuff a couple of years ago... and support didnt respond within a month.. not even an ack [15:07:20] <cartman> not nice [15:07:30] <KotH> mru: now we've customer who wants to have some stuff done with iar [15:07:39] <mru> change customer [15:07:45] <KotH> lol [15:07:49] <kshishkov> do the same to him as iar [15:07:55] <cartman> pluggable customers ftw [15:08:01] <mru> works for me [15:08:29] <cartman> mru: whats the preffered working arm compiler? Arm's own? [15:08:48] <mru> depends [15:09:00] <mru> for random open source stuff I'd recommend gcc [15:09:31] <cartman> for $$$serious work? [15:09:41] <mru> depends [15:09:53] <mru> gcc is usually ok [15:10:08] <cartman> codesourcery or self built? [15:10:09] <mru> armcc gives more control over some low-level aspects [15:10:23] <mru> for linux apps I'd start with gcc [15:10:30] <mru> less hassle [15:10:34] <kshishkov> cartman: there's also Linaro toolchain these days [15:10:43] <mru> vanilla or linaro would be my choice [15:10:45] <cartman> kshishkov: yeah I wonder how it performs [15:10:57] <kshishkov> cartman: what do you expect from GCC? [15:11:02] * mru should do another compiler benchmark run [15:11:11] <cartman> kshishkov: miscompilations [15:11:18] <mru> cartman: check fate [15:11:32] <spaam> cartman: did you fix the bugs yet? [15:11:33] <cartman> mru: my fate is not green, stop making me sad [15:11:43] <cartman> spaam: found the faulty commit [15:11:48] <spaam> Nice [15:11:54] <spaam> where is that report? [15:11:57] <KotH> cartman: we're working with gcc here and have not had any serious problems yet [15:12:00] <mru> fate/arm is has errors with latest vanilla gcc and latest armcc [15:12:04] <mru> linaro gcc passes [15:12:07] <cartman> KotH: cool [15:12:21] <KotH> cartman: sometimes it's a bit akward, but that goes with the field of embedded systems [15:12:22] <cartman> spaam: http://llvm.org/bugs/show_bug.cgi?id=9307 [15:12:39] <cartman> KotH: Hopefully one day I'll get my hands on some <Foo>Board [15:12:54] <KotH> cartman: we build those <foo>boards ourself :) [15:13:01] <cartman> KotH: awesome :( [15:13:02] <cartman> :) [15:13:12] <KotH> cartman: having the right job in the right company helps ;) [15:13:35] <cartman> KotH: yeah well my embedded job is over, had fun :P Maybe in the future [15:13:45] <KotH> but having to review code i've written a year ago sucks .... [15:13:55] <KotH> cartman: you didnt had an embedded job [15:13:57] <mru> that's why you switch jobs often [15:14:06] <KotH> cartman: you had a job working with portable PCs [15:14:07] <mru> make it somebody else's problem [15:14:09] <cartman> KotH: I had access to some tablets :) [15:14:14] <cartman> mru: lol [15:14:40] <cartman> KotH: overriding malloc on WinCE was fun :-P [15:14:47] <mru> unfortunately it works both ways [15:14:53] <KotH> cartman: embedded is when you dont have a full fledged OS with networking and everything in your toaster, but run your code on bare metal [15:15:06] <cartman> KotH: I had an offer for a job like that [15:15:11] <KotH> mru: somebody made it my problem ^^' [15:15:21] <mru> everyone is someone's someone else [15:15:27] * cartman has a friend whose company produces 2-3$ chips [15:15:51] <KotH> cartman: you should do it like me: tell your friend whos company is looking for an engineer that youre looking for a new job :)$ [15:16:03] <cartman> KotH: but its in Turkey [15:16:06] <cartman> I wanted EU [15:16:19] * KotH holds cartman to the light [15:16:27] * KotH thinks that something is missing [15:16:28] * cartman sees Ataturk [15:16:30] <cartman> :P [15:16:42] <cartman> I have a masterpiece at $HOME [15:16:43] <kshishkov> cartman: well, .ch is not that bad [15:16:57] <cartman> Mustafa Kemal Ataturk watching Sabiha Gökçen's first flight [15:17:00] <cartman> now beat that kshishkov [15:17:02] <cartman> KotH* [15:17:02] <cartman> :D [15:17:16] <KotH> cartman: i know where you live! ;) [15:17:20] <cartman> lol [15:17:28] <kshishkov> KotH: he's moving! [15:17:36] <cartman> kshishkov: sssssssssh! [15:17:41] <kshishkov> KotH: so it will be easier for you to reach him! [15:17:49] * cartman welcomes KotH [15:18:03] <kshishkov> KotH: I think there should be rather direct train Zürich-Nürnberg [15:18:03] <mru> something in the mpeg2/4 decoders miscompiles with iar... [15:19:03] <cartman> kshishkov: good for me! [15:36:56] <cartman> trains are too expensive though [15:37:03] * cartman would like to see his sister in Amsterdam [15:37:33] <thresh> it's not easy to get out of amsterdam in a sane state of mind [15:37:44] <cartman> thresh: I do manage fine :) [15:39:47] <cartman> have a nice weekend! [15:44:26] <mru> wow, ffvp8 is 63% faster than libvpx on one clip [15:44:33] <mru> on cortex-a8 [15:45:44] <kshishkov> that means other pathways are not that optimised [15:46:10] <mru> I don't know what's magical about that clip [15:46:17] <mru> maybe profiling will tell [16:18:12] <BBB> mru: 63% is good right? [16:18:20] <mru> very [16:18:25] <BBB> or were you expecting at least 500% faster per clip? [16:18:26] <BBB> :) [16:18:31] <mru> no [16:18:58] <BBB> thank dark_shikari, he did really crazy stuff [16:19:06] <mru> btw, that number means (ffvp8 fps)/(libvpx fps) = 1.63 [16:19:15] <BBB> I know :) [16:19:42] <BBB> also, I'm thinking about adding edge writing to the buffer after frame decoding completes, so we can do away with MC edge emulation [16:19:52] <BBB> dark_shikari has been pestering me to do that since, like, well, forever [16:21:17] <BBB> might make it another 10% faster [16:21:22] <BBB> er, not 10% [16:21:24] <BBB> 0.1% [16:21:25] <BBB> sorry [16:24:25] <lu_zero> still something =P [16:54:06] <BBB> lu_zero: 0.1% is pretty significant [16:54:16] <BBB> I'd say a couple of 0.1% improvements can easily make a noticeable difference [16:54:38] <BBB> optimized codecs are difficult to optimize by 10% "just like that" [17:11:03] <lu_zero> mans mind if I add a ${SDL_CONFIG:-${cross-prefix}sdl-config}? [17:11:29] <mru> if that's how it's meant to be.. [17:13:11] <lu_zero> right now I'm installing sdl to /usr/CHOST/usr/ [17:13:24] <lu_zero> and the sdl-config resides in /usr/CHOST/usr/bin/ [17:13:28] <lu_zero> w/out prefixes [17:14:00] <mru> of course it'll still fail miserably with a sysroot setup [17:14:31] <lu_zero> right now I workarounded this way [17:15:34] <lu_zero> I'd propose to check for sdl-config OR use pkg-config [17:16:00] <mru> pkgconfig is perhaps slightly more predictable [17:16:59] <lu_zero> yup [17:17:13] <lu_zero> I'll check what's working better [17:19:38] <Tjoppen> http://www.youtube.com/watch?v=aYBkDxao3wg Regular Ordinary Swedish Meal Time - Superior Smörgåscake [17:36:18] <shahriman> i am trying to compile ffmpeg in windows 7 (MSYS) [17:36:23] <shahriman> i got this error [17:36:24] <lu_zero> shahriman: hi [17:36:24] <shahriman> libavformat/avidec.c: In function 'avi_metadata_creation_time': [17:36:25] <shahriman> libavformat/avidec.c:288:13: error: implicit declaration of function 'strcasecmp [17:36:25] <shahriman> ' [17:36:36] <lu_zero> your string.h is bogus [17:36:43] <shahriman> lu_zero:hi [17:36:59] <lu_zero> edit it to not ifdef or ifndef on __STRICT_ANSI__ [17:37:11] <lu_zero> (yes apparently errors are infective) [17:37:46] <shahriman> thanks lu_zero, trying it right now [17:38:05] <lu_zero> shahriman: ping me on #ffmpeg if you need more help [17:40:08] <shahriman> lu_zero: where exactly is string.h.. i can't find it [17:50:31] <BBB> hm [17:50:34] <Kovensky> 14:36.36 lu_zero: your string.h is bogus <-- strcasecmp is defined in strings.h on mingw [17:50:36] <BBB> apple guys think it's a bug in our code [17:50:39] <BBB> I wonder how to debug that [17:50:41] <Kovensky> it's wrong, but that's where it's defined [17:55:51] <mru> hmm, iar can't multiply [17:58:20] <BBB> also, did we have a valgrind fate box? [17:58:25] <mru> yes [17:58:37] <mru> x86 and ppc [17:58:39] <mru> both green [17:59:14] <BBB> awesome [17:59:26] * BBB tries to find people to do a cygwin and minw64 x86-64 fate box [18:01:16] * elenril yawns [18:01:56] <thresh> I think we have some buildbot with mingw64 on videolan servers [18:01:58] <BBB> elenril: interested? [18:02:12] <BBB> thresh: that'd be awesome also [18:02:13] <thresh> I'm not in a mood of configuring that python abomination though [18:02:17] <BBB> thresh: but it needs to run fate [18:02:25] <BBB> elenril: yes your patch I know, going through it [18:02:29] <BBB> (after lunch) [18:02:40] <elenril> where are all the reviewers who aren't you [18:04:45] <elenril> oh well, i guess i could do some work instead [18:06:27] <elenril> btw did i mention that QFT is the best thing ever? [18:07:12] <shahriman> lu_zero: thanks a lot, i just finished compiling ffmpeg [18:07:41] <shahriman> got loads of warnings ... is it normal? [18:08:13] <iive> yep [18:12:39] <mru> no matter what you do, you'll always get warnings with some compilers [18:13:05] <mru> and even if targeting only a single compiler, some warnings should be considered acceptable [18:13:21] <mru> the alternative is to turn off entire classes of warnings [19:23:40] <Kovensky> <@elenril> btw did i mention that QFT is the best thing ever? <-- getting over your QFT trauma by being traumatized by solid state physics instead? [19:30:41] <elenril> i got over solid state too [19:30:45] <elenril> now it's group theory [19:31:21] <kshishkov> huh? [19:31:44] <kshishkov> so now you'll learn the math you used in physics before that? [19:31:55] <elenril> exactly! [19:32:09] <kshishkov> ah, so it's traditional education [19:32:10] <BBB> mru: could you apply http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/2011-February/107476.html ? [19:33:19] <kshishkov> BBB: you're not pross-au but still I have to nag you about bink-b [19:33:26] * elenril stabs kshishkov [19:33:40] <elenril> no, my patches are more important =p [19:33:44] <BBB> elenril: you first [19:33:48] <BBB> elenril: title of patch threads? [19:34:35] <elenril> [PATCH 1/4] lavf: use a new ffio_wfourcc macro instead of put_tag() where possible [19:35:39] <elenril> (and 2/4, 3/4 and 4/4) [19:35:46] <elenril> didn't you review them already? [19:36:13] <mru> BBB: dxva patch queued, will push out later [19:36:18] <BBB> thank you [19:36:26] <BBB> elenril: I did, I'm queueing them and testing them only [19:39:08] <BBB> fate is running, seems good so far [19:40:59] <BBB> then I'll get to kostya's bink-b-audio patches [19:42:45] * BBB loves make -j8 fate [19:43:33] <CIA-15> ffmpeg: Anton Khirnov <anton@khirnov.net> master * r0abdb29317 ffmpeg/libavformat/ (12 files): [19:43:34] <CIA-15> ffmpeg: lavf: use a new ffio_wfourcc macro instead of put_tag() where possible [19:43:34] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [19:43:38] <CIA-15> ffmpeg: Anton Khirnov <anton@khirnov.net> master * r99f42c27ab ffmpeg/libavformat/avienc.c: [19:43:38] <CIA-15> ffmpeg: avienc: replace &tag[0] with tag. [19:43:38] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [19:43:40] <CIA-15> ffmpeg: Anton Khirnov <anton@khirnov.net> master * rbbc413f943 ffmpeg/libavformat/ (8 files): [19:43:40] <CIA-15> ffmpeg: lavf: replace remaining uses of put_tag with avio_write [19:43:40] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [19:43:42] <CIA-15> ffmpeg: Anton Khirnov <anton@khirnov.net> master * r61840b4360 ffmpeg/libavformat/ (avio.h aviobuf.c): [19:43:42] <CIA-15> ffmpeg: avio: deprecate put_tag [19:43:42] <CIA-15> ffmpeg: it's not used internally anymore and shouldn't be public. [19:43:42] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [19:45:08] <BBB> elenril: anything else? [19:45:39] <BBB> if not, let's move on to kshishkov [19:45:48] <BBB> kshishkov: list of patch thread subjects? [19:47:29] <kshishkov> BBB: http://patches.ffmpeg.org/patch/1214/ till 1219 [19:47:46] <elenril> BBB: avio_get_str [19:47:52] <BBB> elenril: oh right [19:48:48] <BBB> needs a rebase [19:49:03] <CIA-15> ffmpeg: Dave Yeo <daveryeo@telus.net> master * rcc4e9d2a24 ffmpeg/configure: [19:49:03] <CIA-15> ffmpeg: OS/2: lxlite should use stdout [19:49:03] <CIA-15> ffmpeg: This causes lxlite to use stdout instead of vioXXX [19:49:03] <CIA-15> ffmpeg: functions. This improves fate and build logs readability. [19:49:03] <CIA-15> ffmpeg: Affects OS/2 only. [19:49:04] <CIA-15> ffmpeg: Signed-off-by: Mans Rullgard <mans@mansr.com> [19:49:13] <CIA-15> ffmpeg: Kyle <kshawkeye@gmail.com> master * r04973f8082 ffmpeg/libavcodec/dxva2_internal.h: [19:49:13] <CIA-15> ffmpeg: dxva2: define required feature selection macros [19:49:13] <CIA-15> ffmpeg: Signed-off-by: Mans Rullgard <mans@mansr.com> [19:49:41] <BBB> you know, reimar has a point [19:49:54] <BBB> stream copy destroys the codec id, are we sure that's a good way to do this? [19:50:00] <BBB> if you think it is, it's fine with me... [19:50:36] <CIA-15> ffmpeg: Mans Rullgard <mans@mansr.com> master * r8997bb8807 ffmpeg/libavcodec/bink.c: [19:50:36] <CIA-15> ffmpeg: bink: use LOCAL_ALIGNED for aligned stack data [19:50:36] <CIA-15> ffmpeg: Signed-off-by: Mans Rullgard <mans@mansr.com> [19:50:40] <kshishkov> it doesn't matter here [19:52:11] <BBB> ok, then it's fine I guess [19:54:03] <BBB> it doesn't wan to apply [19:54:03] <BBB> odd [19:56:05] <kshishkov> get_le16 in the way? [19:56:25] <BBB> indeed [19:56:28] <BBB> fixed, queued [19:56:34] <kshishkov> thanks [19:56:51] <kshishkov> five more left ;) [20:06:30] <BBB> kshishkov: 5? 4 more [20:06:32] <BBB> it's 5 in total [20:07:27] <kshishkov> and http://patches.ffmpeg.org/patch/1219/ [20:08:56] <BBB> got it [20:09:03] <BBB> were the idct prevent overflows patches ever applied? [20:11:01] <kshishkov> BBB: it seems not :( [20:14:10] <BBB> kshishkov: you should push us more :-p [20:14:26] <kshishkov> it's you who should patches, not me [20:15:12] <BBB> you push me, I push patches [20:15:15] <BBB> all 7 queued [20:15:21] <kshishkov> thank you [20:17:30] <BBB> anyone have an opinion on wbs' av_pkt_dump_log()? should it take an AVStream stream or a AVRational time_base? [20:18:02] <BBB> I essentially don't care a lot, but AVRational makes a little more sense since, well, it really only uses that much [20:19:43] <kshishkov> and it's a bit more independent then [20:21:48] <mru> this iar compiler is funny [20:21:59] <kshishkov> obvious from name [20:22:52] <mru> guess how it compiles int x; /* ... */ int s = ~x >> 31; [20:23:07] <kshishkov> 32-bit? [20:23:11] <mru> yes [20:24:00] <kshishkov> not;shl ? [20:24:09] <mru> this is arm [20:24:21] <kshishkov> should be one instruction then [20:24:57] <CIA-15> ffmpeg: Peter Ross <pross@xvid.org> master * r582ac86d19 ffmpeg/libavcodec/binkaudio.c: [20:24:58] <CIA-15> ffmpeg: binkaudio: perform band table scaling in decode_init [20:24:58] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [20:25:07] <BBB> kshishkov, pross-au: all pushed [20:25:08] <CIA-15> ffmpeg: Peter Ross <pross@xvid.org> master * rf0ca29eb5f ffmpeg/libavformat/bink.c: [20:25:08] <CIA-15> ffmpeg: bink: set audio stream codec_tag such that binkaudio decoder can identify bitstream version [20:25:08] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [20:25:09] <CIA-15> ffmpeg: Peter Ross <pross@xvid.org> master * rccfcddb3f2 ffmpeg/ (Changelog libavcodec/avcodec.h libavcodec/binkaudio.c): [20:25:09] <CIA-15> ffmpeg: Bink version 'b' audio decoder [20:25:09] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [20:25:10] <BBB> let me know if any are left [20:25:14] <CIA-15> ffmpeg: Peter Ross <pross@xvid.org> master * ra304def1dc ffmpeg/libavcodec/binkaudio.c: [20:25:14] <CIA-15> ffmpeg: binkaudio: remove unnecessary loop [20:25:14] <CIA-15> ffmpeg: decode_init sets bands[0] == 2, so this loop always sets the band table [20:25:14] <CIA-15> ffmpeg: index (k) to zero. [20:25:14] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [20:25:16] <CIA-15> ffmpeg: Peter Ross <pross@xvid.org> master * r8a8c283edd ffmpeg/libavcodec/binkaudio.c: [20:25:16] <CIA-15> ffmpeg: binkaudio: simplify frame_len_bits and frame_len calculation [20:25:16] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [20:25:18] <kshishkov> BBB: thanks, man [20:25:18] <CIA-15> ffmpeg: Peter Ross <pross@xvid.org> master * r588a3ffd96 ffmpeg/libavformat/bink.c: [20:25:19] <CIA-15> ffmpeg: bink: decode audio track identifiers into AVStream.id [20:25:19] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [20:25:25] <CIA-15> ffmpeg: Peter Ross <pross@xvid.org> master * re211e255aa ffmpeg/ (libavcodec/binkidct.c tests/ref/fate/bink-demux-video): [20:25:25] <CIA-15> ffmpeg: bink: prevent overflows within binkidct by using int-sized intermediate array [20:25:25] <CIA-15> ffmpeg: Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com> [20:25:41] <kshishkov> BBB: and you have http://patchwork.libav.org/project/FFmpeg-devel/list/ for monitoring [20:25:45] <mru> kshishkov: yes, it _can_ be done in one instruction [20:26:32] <kshishkov> mru: and, obviously, this I-or compiler manages to use three instructions? [20:26:44] <mru> it manages to get the wrong result [20:26:53] <kshishkov> even better [20:27:09] <pJok> nice [20:27:13] <pJok> more bink audio [20:27:31] <kshishkov> pJok: now we support more Bink variants than RAD tools :P [20:27:43] <pJok> kshishkov, frightening :P [20:27:50] <kierank> kshishkov: have bink ever contacted you? [20:27:59] <pJok> now when is ffbikenc ready? [20:28:25] <kshishkov> kierank: http://codecs.multimedia.cx/?p=196#comments [20:29:09] <mru> kshishkov: of the compilers I have at hand, only TI does it with one instruction [20:30:27] <kshishkov> mru: troll ARMcc devs with it ;) [20:30:45] <mru> armcc at least gets the right result [20:31:13] <kshishkov> even GCC sometimes can do that, big deal [20:31:35] <mru> armcc and gcc both do (~x)>>31 [20:31:41] <mru> TI does ~(x>>31) [20:31:45] * kshishkov still remembers that wonderful optimisation GCC performs on float arrays on ARM [20:31:54] <mru> that one is fixed [20:32:05] <pJok> mru, reroll gcc random code generator? [20:32:10] <mru> iar does (x >> 31) ^ 1 [20:32:28] <kshishkov> lol [20:33:01] <kshishkov> why cannot it be called i-or instead? [20:35:39] <pasteeater> are inserted or attached patches preferred on the devel ML? i see both being used. [20:35:53] <mru> pasteeater: either is fine as long as the patch isn't mangled [20:36:27] <pasteeater> i see. thanks. [20:36:34] <mru> if you're unsure, attach [21:09:05] <astrange> BBB: agreed with the diagnosis. let me finish my homework... [21:19:02] <lu_zero> merbanan: you around? [21:21:07] <lu_zero> BBB: I'm about to push the "mov tkhd' width and height usage" patchset [21:32:18] <CIA-15> ffmpeg: Maksym Veremeyenko <verem@m1stereo.tv> master * r77d207cbe6 ffmpeg/libavformat/movenc.c: [21:32:18] <CIA-15> ffmpeg: reindent after tapt patch [21:32:18] <CIA-15> ffmpeg: Signed-off-by: Luca Barbato <lu_zero@gentoo.org> [21:32:22] <CIA-15> ffmpeg: Maksym Veremeyenko <verem@m1stereo.tv> master * rea1afa124c ffmpeg/libavformat/movenc.c: [21:32:22] <CIA-15> ffmpeg: use tapt atom for sample aspect ratio [21:32:22] <CIA-15> ffmpeg: Signed-off-by: Luca Barbato <lu_zero@gentoo.org> [21:32:27] <CIA-15> ffmpeg: Maksym Veremeyenko <verem@m1stereo.tv> master * rd184c86cd3 ffmpeg/libavformat/movenc.c: [21:32:27] <CIA-15> ffmpeg: store pasp atom for all types of quicktime movie [21:32:27] <CIA-15> ffmpeg: Signed-off-by: Luca Barbato <lu_zero@gentoo.org> [22:22:01] <mru> BBB, elenril: you added some warnings: http://fate.ffmpeg.org/arm-linux-gcc-4.3/20110225221948/compile/201102242219... [22:31:18] <BBB> lu_zero: I trust you fully, so push it if you feel confident about the patch [22:31:20] <BBB> mru: on it [22:31:44] <lu_zero> BBB: just in case =) [22:32:39] <BBB> mru: grep put_tag libavformat/* returns nothing except the old function definition [22:32:44] <BBB> mru: so I think that warning is old [22:33:06] <BBB> oh, it's maksym's patches [22:33:15] <BBB> somebody forgot to update them to the new API [22:33:17] <mru> BBB: fate says it showed up between those two dates in the url [22:42:25] <BBB> mru: fixed, see 2nd email [22:42:29] <BBB> first one is incomplete, I forgot one [23:01:51] <Dark_Shikari> mru: ffvp8 is like, generally 50% faster or so. [23:01:57] <Dark_Shikari> it's really kinda embarassing [23:02:17] <Dark_Shikari> "lol, this decoder that got ~3 man-weeks put into it is 50% faster than something with a dozen full-time maintainers" [23:05:07] <superdump> that kind of discounts the number of people and amount of time put into the shared code used in the vp8 decoder, no/ [23:05:17] <superdump> ?* [23:05:33] <superdump> (the vp8 decoder == ffvp8) [23:09:33] <Dark_Shikari> ? [23:12:20] <superdump> i'm saying the 3 man-weeks is probably not a fair total estimate if it is leveraging shared functions from ffh264 or so [23:15:49] <Dark_Shikari> It didn't leverage much shared stuff [23:15:54] <Dark_Shikari> things it leveraged: [23:16:01] <Dark_Shikari> some predict functions (trivial) [23:16:08] <Dark_Shikari> emulated edge mc (a few minutes of work) [23:16:18] <Dark_Shikari> the bitstream reader/etc [23:16:25] <Dark_Shikari> vp56 arith decoding (which we rewrote 3 times anyways...) [23:19:50] <superdump> fair enough then [23:43:45] <mru> Dark_Shikari: it obviously also shares all the buffer management and other high-level things [23:50:38] <Dark_Shikari> True, but that stuff isn't very complicated. [23:51:17] <mru> doing it all from scratch still takes some time [23:52:16] <Dark_Shikari> not 6+ years [23:53:29] <mru> of course not [23:53:41] <mru> but a few weeks perhaps [23:53:46] <kierank> on2's cto doesn't come on #x264 any more [23:53:56] <kierank> you could have told him [23:54:05] <kierank> probably run away with his $$$