[FFmpeg-devel-irc] IRC log for 2011-02-18

irc at mansr.com irc at mansr.com
Sat Feb 19 01:00:15 CET 2011


[00:01:14] <Jumpyshoes> is there a benefit to ffmpeg's config style vs. x264's?
[00:01:33] <mru> ffmpeg's is better :)
[00:02:33] <Sean_McG> mru: out of nothing more than idle curiosity, can I ask what kind of G4 saracen is?
[00:02:43] <mru> windtunnel
[00:02:46] <Sean_McG> ah
[00:02:58] <Sean_McG> yeah I  used to have a Quicksilver
[00:03:09] <mru> I keep it behind a closed door :)
[00:03:13] <Sean_McG> buying a used G4 Mini from an acquaintance
[00:03:24] <Sean_McG> I wanna learn POWER more.
[00:03:48] <mru> get ready for the 5-operand shift/rotate
[00:06:19] <drv> sounds like barrels of fun
[00:07:12] <Sean_McG> I need something to take my mind off my job before it devours what's left of my soul :\
[00:08:42] <mru> Sean_McG: why don't you change jobs?
[00:09:00] <Sean_McG> there's not a lot available for my skillset
[00:09:02] <mru> few things are more satisfying than quitting a shitty job
[00:09:24] <Sean_McG> I do IBM Tivoli stuff, and I hate it
[00:09:30] <mru> so learn something new while pretending to work
[00:09:47] <Sean_McG> that only works when you have time, and a team leader who isn't f**king micromanaging
[00:10:24] <mru> oh, your job really sucks...
[00:11:26] <Sean_McG> in the beginning it wasn't so bad, but the past 6 months have been really shitty
[00:11:37] <mru> it's always like that
[00:11:49] <mru> in the beginning it's at least different from what you did before
[00:12:00] <mru> and all companies get worse over time
[00:12:06] <Sean_McG> aye... in the beginning I was a severely underpaid web developer
[00:12:37] <Sean_McG> well, I work for the government so really it should be a given... the whole "do more, with less money" thing
[00:12:49] <mru> if you don't have to lower your demands, you're not asking for enough
[00:12:58] <mru> so I've learned
[00:14:59] <Sean_McG> past 2 weeks have been really bad... I feel like I'm on a sinking ship :|
[00:16:02] <mru> have the rats left the building?
[00:16:17] <Jumpyshoes> where do you work at?
[00:16:23] <Sean_McG> Revenue Canada
[00:16:52] <Jumpyshoes> oh, canada
[00:17:00] <Sean_McG> heh
[00:17:58] <Jumpyshoes> don't know anything about canada
[00:18:43] <mru> it's that empty place in the north
[00:19:06] <mru> but it's not actually as big as it looks on the maps
[00:19:14] <mru> they're cheating, you see
[00:20:53] <kierank> [00:17] Jumpyshoes: oh, canada --> *rimshot*
[00:21:02] <Jumpyshoes> what
[00:21:11] <Jumpyshoes> i thought Sean_McG was in the US
[00:21:21] <mru> it's pretty obvious he's not
[00:21:32] <mru> from the rogers.com IP if nothing else
[00:21:46] <Jumpyshoes> i don't whois people
[00:21:50] <Jumpyshoes> generally
[00:21:55] <mru> my irc client displays it when he comes on
[00:22:07] <Sean_McG> wow, why so much red on fate today?
[00:22:37] <mru> Sean_McG: yeah, BBB broke ppc
[00:22:46] <Sean_McG> ahhh the VC1 stuff
[00:22:58] <BBB> orking on it
[00:23:51] <BBB> baby cries unstoppingly
[00:23:52] <BBB> hard to work
[00:23:55] <BBB> I mostly got it fixed
[00:23:57] <Sean_McG> ouch
[00:25:11] <mru> BBB: maybe they're related: baby cries when fate is red
[00:26:08] <Sean_McG> hmmm well, I think I'mma watch some Doctor Who
[00:26:15] <Sean_McG> later folks
[00:32:54] <BBB> ok works now
[00:32:59] <BBB> will send a patch
[00:33:10] <mru> \o/
[00:34:28] <BBB> patch sent
[00:34:30] <BBB> have a look
[00:35:30] * BBB isrocking baby with one foot typing with two hands and using the other foot to remain stable on his chair
[00:35:40] <BBB> bouncy chair \o/
[00:35:44] <BBB> anyway
[00:36:00] <BBB> now let's go back to fixing vc1 performance for real
[00:39:38] <mru> BBB: I'm not entirely happy with the amount of added ifdeffery
[00:39:54] <mru> can't the vc1 bits simply be moved to another file?
[01:02:39] <BBB> mru: they depend on stuff shared in that template file
[01:02:51] <BBB> mru: so I'd have to split the template file in two and add the shared stuff in a third file
[01:02:57] <BBB> mru: which looked messy to me
[01:03:17] <mru> CHROMA_MC8_ALTIVEC_CORE?
[01:04:07] <mru> btw that file looks like whoever wrote it was trying to make altivec as messy as mmx
[01:04:35] <Sean_McG> yeah the propensity for macros in the altivec files make the baby jesus cry
[01:04:37] <BBB> hehe :)
[01:07:44] <BBB> mru: so what do you suggest I do? move the CORE into a header and split the template file?
[01:10:50] <mru> is that the only common part?
[01:11:04] <BBB> I think so
[01:14:03] * BBB pokes mru
[01:14:31] <mru> gah, can't make up my mind
[01:14:39] <mru> it's all ugly
[01:14:41] <BBB> I can just keep ppc broken
[01:14:45] <BBB> and yes it's ugly
[01:14:53] <BBB> how do you call this type of asm again?
[01:14:59] <BBB> it's not yasm, it's not inline asm
[01:15:05] <mru> intrinsics
[01:15:08] <mru> worst kind
[01:15:10] <BBB> oh yeah, intrinsics
[01:15:15] <BBB> I remember the mmx intrinsics
[01:15:19] <BBB> wasn't that called mmx.h?
[01:15:22] <BBB> that was fabulous
[01:15:27] <BBB> like Sex and the City III
[01:15:34] <BBB> absolutely horrid
[01:16:58] <kierank> women and gay people would beg to differ
[01:17:10] <mru> some find gay people horrid too
[01:17:26] <mru> not that I'd ever endorse such opinions
[01:17:45] <kierank> what's that got to do with women and gay people liking sex and the city
[01:17:53] <mru> nothing
[01:18:34] <Jumpyshoes> i thought the women in sex and the city 3 were old
[01:18:59] <mru> Jumpyshoes: you're young...
[01:20:16] <Jumpyshoes> one of the main actresses is currently 45 and the show ended in 2004
[01:20:21] <mru> not that I'd know how old aforementioned women are, not having seen the film
[01:20:40] <mru> nor the show
[01:20:48] <mru> some things are better left unseen
[01:21:31] <Jumpyshoes> well i haven't watched it either, i just read an article about old women in the movie
[01:21:49] <mru> don't believe everything you read
[01:21:55] <mru> unless it's on irc
[01:22:00] <mru> everything on irc is true
[01:23:03] <CIA-15> ffmpeg: Ronald S. Bultje <rsbultje at gmail.com> master * red040f35f2 ffmpeg/libavcodec/ppc/ (h264_altivec.c h264_template_altivec.c vc1dsp_altivec.c): Fix PPC build.
[01:24:54] <kierank> Jumpyshoes: ask BBB
[01:24:58] <kierank> he's the satc expert
[01:25:32] <Jumpyshoes> i did not know that
[01:26:52] <BBB> huh?
[01:26:53] <BBB> dude
[01:26:56] <BBB> you gotta be kidding me
[01:26:59] <BBB> my wife made me go
[01:27:01] <BBB> or actually her sister
[01:27:13] <BBB> my wife's sister's boyfriend had to go to, he hated it also
[01:27:19] <Jumpyshoes> sometimes you need to step up and say no`
[01:27:23] <BBB> and we went to watch the worldcup matches afterwards :)
[01:28:31] <kierank> you mean watch holland get beaten by spain?
[01:31:19] <bcoudurier> I didn't know the 3rd was out yet
[03:38:35] <Dark_Shikari> Anyone know what the heck qscale_table is used for in the vp3 decoder?
[03:38:39] <Dark_Shikari> It's written to everywhere, and used nowhere.
[03:40:15] <astrange> libpostproc
[03:40:24] <Dark_Shikari>     memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
[03:40:27] <Dark_Shikari> what is with this?
[03:40:29] <Dark_Shikari> why are only 512 bytes set?
[03:40:32] <Dark_Shikari> why is it [2048]?
[03:40:36] <Dark_Shikari> this makes no sense
[03:40:43] <astrange> and/or mplayer -lavdopts vstats
[03:41:01] <Dark_Shikari> I'm fixing vp3 to work with widths > 2048
[03:41:50] <astrange> qscale_table assumes MPEG1-4, unfortunately
[03:42:01] <Dark_Shikari> What does h264 set it to?
[03:42:01] <Dark_Shikari> etc
[03:42:11] <astrange> so the values are the QP for every 16 pixels
[03:42:29] <Dark_Shikari> But what's the memset for
[03:43:07] <astrange> beats me. maybe the qps in vp3 aren't the same scale as mpeg
[03:43:15] <Dark_Shikari> Why 512 bytes?
[03:44:46] <Dark_Shikari> what if qscale_table isn't set?
[03:44:49] <Dark_Shikari> what if it's null?
[03:44:49] <Dark_Shikari> etc
[03:45:26] <astrange> it's always set if it's a static array
[03:45:27] <astrange> b928ec64 (Michael Niedermayer 2003-08-20 07:35:23 +0000  707)     memset(s->qscale_table, (FFMAX(s->intra_y_dequant[1], s->intra_c_dequant[1])+8)/16, 512); //FIXME finetune
[03:46:16] <astrange> 512 should be picture width in 16x16 macroblocks
[03:46:18] <BBB>  omg slashdot alert, D_S h264 expert switches to theora, dumps vp8?!?one
[03:47:17] <Dark_Shikari> astrange: why is it just one row?
[03:47:49] <Jumpyshoes> lol BBB
[03:48:14] <astrange> well it shouldn't be that either, but that's more correct at least
[03:48:26] <astrange> in that it doesn't set things that aren't read
[03:48:45] <astrange> but it should be something kind of like a macroblock qp, or else nothing
[03:49:07] <Dark_Shikari> how does postproc use it?
[03:49:09] <Dark_Shikari> what if it isn't set?
[03:49:25] <astrange> increases deblock/dering strength based on qp as if it were mpeg4
[03:49:34] <astrange> if it's not set, no pp
[03:49:51] <Dark_Shikari> what should we do then?
[03:50:03] <astrange> well pp doesn't even make sense for vp3 so that doesn't matter
[03:50:12] <Dark_Shikari> so we should just kill it?
[03:50:14] <Dark_Shikari> leave it null?
[03:50:29] <astrange> i'd delete it. qscale_table doesn't have to be set in AVFrame
[03:50:31] <Dark_Shikari> oh I see
[03:50:33] <Dark_Shikari> it set qstride to 0
[03:50:36] <Dark_Shikari> to repeat the same row over and over
[03:51:03] <astrange> only useful for the stats files mplayer prints, but if they aren't real numbers in the bitstream i don't think it's good for research
[03:54:28] <bcoudurier> some people use it to retrieve qps
[03:54:58] <Dark_Shikari> .... oh god, this is more embarassing than I thought
[03:55:00] <Dark_Shikari> 14:49 < gmaxwell> https://people.xiph.org/~greg/vp8/primer-a.00000001.png
[03:55:00] <Dark_Shikari> 14:49 < gmaxwell> https://people.xiph.org/~greg/vp8/primer-b.00000001.png
[03:55:00] <Dark_Shikari> 14:49 < gmaxwell> https://people.xiph.org/~greg/vp8/primer-c.00000001.png
[03:55:08] <Dark_Shikari> Guess which codec each of these is... then read the spoilers.
[03:55:14] <Dark_Shikari> You'll laugh your head off
[03:55:21] <astrange> theora, vp8, x264 or newer theora?
[03:55:25] <Dark_Shikari> A: VP8 encode from many months ago.  B: VP8 encode from today.  C: Theora
[03:55:32] <astrange> ...
[03:56:51] <bcoudurier> theora is shifted
[03:57:02] <Dark_Shikari> Yes, it's not exactly the same frame, due to ffmpeg not being frame-exact.
[03:57:11] <Dark_Shikari> A/B are enough to be hilarious though.
[03:57:27] <bcoudurier> frame exact ? you mean seeking ?
[03:57:33] <Dark_Shikari> yes, i.e. using -ss to grab a frame
[03:57:37] <Dark_Shikari> for a screenshot
[03:57:43] <Dark_Shikari> ffmpeg -i input -ss whatever -vframes 1
[03:57:43] <bcoudurier> ah
[03:57:56] <bcoudurier> yep
[03:58:00] <Dark_Shikari> anyways, it's hilarious.
[03:58:06] <Dark_Shikari> libvpx has actually gotten WORSE.
[03:58:11] <BBB> 1 is better than 2
[03:58:15] <Dark_Shikari> That's the point
[03:58:15] <BBB> what the hell
[03:58:22] <Dark_Shikari> This is why x264 and xvp8 will win
[03:58:25] <Dark_Shikari> because libvpx is not just bad
[03:58:29] <Dark_Shikari> it gets worse every month they develop it for
[03:58:35] <BBB> e.g. look at the detail of his beard, it's completely fuzzy
[03:58:37] <BBB> how did they do that
[03:58:43] <Dark_Shikari> Turning on ARNR by default
[03:58:53] <Dark_Shikari> `-`
[03:59:14] <Dark_Shikari> Bitrate of 2 is apparently a bit lower, a few %
[03:59:17] <Dark_Shikari> due to ratecontrol fuckery
[03:59:19] <Dark_Shikari> but not massively
[03:59:32] <Dark_Shikari> 14:59 < gmaxwell> I think it's even worse in motion, the new one is obviously splotchy.
[04:00:27] <BBB> kshishkov: ping, do you have some kind of a vc1 testsuite? I'd like a variety of samples that use different vc1 features, e.g. where v->rangeredfrm is enabled or where v->pq has >9 values, since that apparently enables different decoder features
[04:00:53] <BBB> Dark_Shikari: nice future for libvpx :(
[04:02:03] * BBB sleep
[04:02:18] <kierank> BBB: there is an official one iirc
[04:04:19] <kierank> meh it costs $450
[04:05:22] <saintdev> Dark_Shikari: I'm guessing my choices are vp8, theora (ptalarbvorm), and x264?
[04:06:29] <Dark_Shikari> saintdev: there's no x264 btw
[04:06:33] <kierank> hmm there's a torrent
[04:06:48] <saintdev> so theora (current?), and ptalarbvorm?
[04:07:23] <saintdev> just trying to get my choices, or is that part of the fun?
[04:08:03] <gnafu[away]> saintdev: Are you referring to the PNGs Dark_Shikari linked to?  He already said what they were.
[04:08:21] <saintdev> i know, i want to play along, so i'm not reading anything below those lines
[04:08:30] <saintdev> or rather i assumed he did
[04:08:35] <Dark_Shikari> Just guess for yourself, and then read it
[04:08:36] <Dark_Shikari> and thne laugh
[04:09:15] <saintdev> well i'm guessing B is vp8 (lolblur)
[04:09:42] <saintdev> oh they're both vp8, hahahaha
[04:10:49] <saintdev> Dark_Shikari: but isn't video _supposed_ to be blurry? ^_^
[04:17:26] <saintdev> peloverde: ping
[04:35:13] <Sean_McG> so IBM actually has a Watson commercial
[04:35:18] <Sean_McG> like anyone could afford one
[04:36:03] <gnafu> Watson commissioned it... as a warning.
[04:36:49] <Sean_McG> now that Watson has beaten Ken Jennings, how much closer are we to booting Skynet?
[04:40:40] <ohsix> watson just cross correlated a few data sources & found a fitness function that'd find the most likely from its large set of inferences
[04:42:26] <Sean_McG> thanks for killing the joke
[04:43:26] <gnafu> ohsix: That's just what they want you to think.
[04:44:17] <ohsix> i'm saddled with the knowledge of how to do it, i blame henry winston
[04:45:03] <Jumpyshoes> stupid question. how do i add a directory for ffmpeg to search for yasm? and prefix it, so it will stop after finding that dir
[04:45:22] <Sean_McG> I think they call it $PATH
[04:45:50] <Jumpyshoes> without actually modifying path, isn't there an option in configure?
[04:48:49] <saintdev> --as ?
[04:49:41] <saintdev> although that probably gcc's as also
[04:49:46] <Sean_McG> I've got yasm in /usr/local/bin so I just add that to the end of $PATH, which seems to follow standards(5) just fine, at least on Solaris
[04:49:47] <saintdev> *overrides
[04:53:01] <Jumpyshoes> i also have yasm, but it doesn't work. so i need to build another version that's unix compliant
[05:18:32] <j0sh> Jumpyshoes: cflags/ldflags?
[05:31:03] <peloverde> saintdev: pong
[05:32:34] <saintdev> peloverde: did you get a chance to look that over?
[05:37:18] <peloverde> sections of it don't seem to match the text
[05:38:57] <peloverde> but some expressions may be folded
[05:39:40] <peloverde> I do see the 115% stuff, i thought you said that part was missing
[05:44:42] <saintdev> not that it's missing, but I guess that was a bad example.
[05:46:25] <peloverde> what you sent does look like a pretty good pseudo code explanation of what the pe code should look like
[05:47:06] <saintdev> you think it would be safe to work off of that?
[05:47:45] <saintdev> and it does leave a few things out, and a few things are (intentionally) under-specified.
[05:50:26] <peloverde> yeah it looks ok
[06:07:11] <cartman> moin
[06:10:42] <pJok> ohayou gozaimasu
[06:10:54] <lu_zero> god morgon
[06:13:09] <cartman> you guys are earlier than me :)
[06:17:57] * elenril glares at BBB 
[06:24:32] <thresh> moroning
[06:25:50] <kshishkov> a wonderful MKAD morning to you
[06:26:22] <thresh> are you inside of MKAD?
[06:27:43] <kshishkov> of course not
[06:27:51] <kshishkov> not even close
[06:28:05] <thresh> I'm not either
[06:28:16] <thresh> so technically I'm in Russia
[06:29:28] <kshishkov> by you're close to its internal border
[06:30:27] <kshishkov> а в Москве небось и утро блатное, не то что у вас в Королёве
[06:30:40] <thresh> что тут что там -28 градусов сегодня было
[06:32:29] <j0sh> how do you guys type cyrillic on the same keyboard
[06:32:32] <j0sh> sticky modifiers?
[06:32:54] <kshishkov> keyboard layout switcher
[06:33:28] <kshishkov> 日本語も
[06:33:34] <j0sh> heh cool
[06:33:43] <saintdev> IME
[06:34:05] <astrange> 語 looks really odd in this font
[06:34:15] * cartman hates when people are late to the early morning meetings
[06:34:23] <astrange> read it as 望
[06:34:41] <kshishkov> cartman: you have left the company, so no problem for you
[06:34:57] <cartman> kshishkov: the meeting is about my resignation :D
[06:35:04] <cartman> and I am the only one here atm. :D
[06:35:30] <lu_zero> makes sense
[06:36:33] <thresh> j0sh: I don't look at the keyboard at all
[06:37:11] <kshishkov> I look at mine, it has only english anyway
[06:37:26] * cartman looks at the synergy daemon
[06:38:27] <spaam> did you find anything?
[06:38:36] <cartman> :P
[07:24:38] <lu_zero> bcoudu: hi
[07:26:29] <bcoudu> hi
[07:41:05] <Tjoppen> morrn
[07:49:53] <spaam> God morgon
[07:54:20] <peloverde> happy thursday
[07:54:27] <cartman> you mean friday!
[07:55:02] <peloverde> it's still thursday here
[07:55:20] <spaam> friday here :D
[07:55:28] <cartman> peloverde: http://nelsonhaha.com
[07:56:44] <superdump> ahoy
[07:57:27] <cartman> ehlo superdump
[08:38:48] <astrange> lu_zero: post ts samples please
[08:39:05] <astrange> lu_zero: in the meantime i'm explaining how to fix lavf
[08:39:15] <lu_zero> I asked for a stable url to the reported
[08:39:18] <lu_zero> reporter
[08:39:34] <lu_zero> I fixed his problem hacking the heuristic to behave differently
[08:41:08] <lu_zero> astrange: still
[08:41:23] <lu_zero> what's particularly wrong in what I said?
[08:42:11] <astrange> well i don't really see the difference in adding a field and adding a utility function. either way i want to be able to ask people "change your code this way" so they fix some a/v sync bugs from mt
[08:42:21] <astrange> but i think "read this field instead of the other one" is a bit easier
[08:42:46] <astrange> i'm trying to remember the problem with using dts timing entirely... it's easier to generate...
[08:43:13] <astrange> oh. with field samples, if you use pkt_dts, the frame combined from the 2 fields gets the time assigned to the second field
[08:43:41] <astrange> and the same for other combined-packet things, it gets the time of the last packet
[08:43:48] <astrange> (other = just mpeg-4 N-VOPs)
[08:44:55] <lu_zero> the utility function makes easier to opt out
[08:58:12] <iive> lu_zero: may I ask you something technical, and please don't get offending.
[08:58:23] <iive> I'm starting to suspect that you may not know what b-frame lag is, and how reordering is done.
[08:58:34] <iive> do you want explanation of that?
[09:23:31] <peloverde> Mike is up yo more crazy http://groups.google.com/a/webmproject.org/group/webm-discuss/browse_thread/thread/80eedd7d4b94a789?
[09:24:03] <kshishkov> peloverde: repost this link when lyakh is around
[09:24:07] <astrange> discovering many odd things about mpeg files...
[09:24:36] * lyakh is around
[09:25:04] <kshishkov> okay, then the news have found its target
[09:29:08] <lyakh> yeah... nice, thanks... do I remember it right, that libvpx wasn't particularly faster on vp8 decoding, than ffmpeg?
[09:30:54] <Dark_Shikari> "wasn't particularly faster"
[09:30:56] <Dark_Shikari> more like 50% slower
[09:31:55] <lyakh> Dark_Shikari: yes, I didn't want to be impolite, in case I remembered it wrongly;)
[10:10:48] <j-b> mike?
[10:11:28] <kshishkov> Mike Melanson, ex-Xinde dev
[10:11:31] <kshishkov> *Xine
[10:11:37] <kshishkov> so you don't know him
[10:11:59] <j-b> I know mike
[10:12:19] <j-b> I was just wondering if he was on IRC
[10:12:46] <thresh> he was
[10:14:50] <j-b> kshishkov: I've met him twice in the USA
[10:15:38] <Dark_Shikari>  je
[10:15:40] <Dark_Shikari> er
[10:15:42] <Dark_Shikari> he's on irc rarely
[10:15:46] <Dark_Shikari> like once every 3 years
[10:16:24] <peloverde> I think we saw him quite recently so the Mike comet will return sometime in 2014
[10:19:15] <superdump> his nick is MultimediaMike when he comes on here
[10:19:19] <kshishkov> maybe he'll re some game codecs
[10:19:24] <Dark_Shikari> peloverde: Hah~
[10:19:40] <superdump> he mostly only comes on when asked though
[10:49:39] <kshishkov> mate!
[10:50:01] <kshishkov> I don't have anything to bug you with, no worries
[10:50:22] <pross-au> A first
[10:50:34] <pross-au> Lets flip the table
[10:50:53] <pross-au> Smush! Smush!
[10:51:09] <kshishkov> ohkay, the patch is on the list
[10:51:36] <pross-au> its needs to get off the list and into the tree
[10:51:41] <kshishkov> indeed
[10:51:52] <kshishkov> but it involves complete rewrite
[11:08:40] <ubitux> oh great, draw_text filter at least :)
[11:11:03] <kshishkov> once I even made "running line" filter for MPlayer
[11:15:00] <j-b> lu_zero: ping
[12:02:34] <Compn> kshishkov : why didnt you post patch?
[12:02:35] <Compn> :)
[12:04:08] <kshishkov> Compn: was too lazy and considered it to be too insignificant
[12:05:00] * elenril kicks committers
[12:05:08] <elenril> wake up lazy british people
[12:05:15] <elenril> patches are waiting
[12:05:18] * Compn needs a video stabilization filter
[12:05:32] <Compn> e.g. entire video jumps up/down 4 pixels or so due to bad vcr/vhs tape
[12:06:36] <kshishkov> there was one somewhere
[12:20:14] <Compn> i should use it :)
[12:20:21] <Compn> i think i searched for them before
[12:30:05] <cartman> mru: do you remember Beagleboard 2010 student, "maltanar" ?
[12:30:37] <cartman> GSoC that is
[12:30:53] <mru> name sounds familiar
[12:31:00] <mru> forgot what he was working on
[12:31:01] <cartman> Yaman Umuroglu
[12:31:07] <kshishkov> POSIX for DSPEasy
[12:31:08] <cartman> dsp something
[12:31:10] <cartman> yeah
[12:31:36] <cartman> He is my colleague from my current $COMPANY
[12:31:49] <cartman> taught me how to pronounce your name :p
[12:32:00] <thresh> 'kostya' is easy
[12:32:11] <cartman> thresh: try mru's name :)
[12:32:15] <kshishkov> thresh: you won't believe how many people get it wrong
[12:32:21] <thresh> kshishkov: wouldnt I?..
[12:32:21] <mru> cartman: nothing hard about my name
[12:32:22] <thresh> :)
[12:32:29] <mru> most languages have the required sounds
[12:32:31] <cartman> mru: the funny a is problematic
[12:32:36] <spaam> cartman: how hard is it? :P
[12:32:40] <cartman> sounds like an ou
[12:32:48] <kshishkov> cartman: it $COMPANY==SuSE or some Turkish?
[12:32:56] <cartman> kshishkov: Turkish one
[12:33:04] <cartman> I told him to stay away from av500 of course
[12:33:20] * kshishkov sees no problems pronouncing MÃ¥ns, it's not like it has "skj" in it
[12:33:20] <thresh> because he cant pronounce his name?
[12:34:01] <cartman> for some other reasons :P
[12:34:07] <spaam> kshishkov: can you say this: Sju sjösjuka sjömän sköttes av sju sköna sjuksköterskor
[12:34:46] <mru> spaam: that's a good one
[12:35:07] <kshishkov> spaam: I used this - http://slayradio.com/mastering_swedish_lesson_3.php
[12:35:13] <cartman> lol @ google translate
[12:35:26] <cartman> Seven seasick seamen were handled by seven beautiful nurses
[12:35:28] <spaam> cartman: http://upload.wikimedia.org/wikipedia/commons/f/f9/Sv-sju_sj%C3%B6sjuka_sj%C3%B6m%C3%A4n.ogg if you want to listen to it ;)
[12:35:57] <cartman> whoa
[12:35:58] <cartman> tough one
[12:37:16] <lu_zero> j-b: pong
[12:37:38] <cartman> spaam: try "Bir berber bir berbere gel beraber bir berber dükkanı açalım demiş." in Google Translate, Turkish
[12:38:07] <thresh> both those languages look like someone accidently smashed his keyboard.
[12:38:52] <lu_zero> unpong
[12:39:01] * lu_zero goes back to the meeting
[12:39:03] <spaam> cartman: haha
[12:39:18] <cartman> spaam: sounds weird ha :D
[12:39:22] <spaam> yeah
[12:40:55] <kshishkov> thresh: ну здравствуйте!
[12:42:19] <kshishkov> thresh: ваши скороговорки вида "шел Лукашенко по шоссе" ничуть не лучше
[12:52:56] <BBB> elenril: is the patch fine now? no more second thoughts?
[12:56:06] <thresh> kshishkov: ты так говоришь "ваши"...
[12:56:26] <kshishkov> thresh: а еще я сало ем
[12:56:51] <thresh> kshishkov: в шоколаде?
[12:57:05] <kshishkov> thresh: было дело
[13:06:22] <elenril> BBB: i think so
[13:06:39] <elenril> at least it compiles without any visible warnings :)
[13:07:06] <mru> did you check for invisible warnings?
[13:07:41] <mru> maybe they were printed in ultraviolet
[13:07:58] * elenril tries grep --ultraviolet
[13:11:07] <kshishkov> try grep --contractclauses next if warnings were printed in usual 0.000005pt font
[13:25:36] <j-b> lu_zero: 1) did not got your mail. 2) libnemesis patch, YOU HAZ?
[13:38:01] <Flameeyes> j-b: I think we're going to declare libnemesi completely dead in favour of libavformat's rtsp support ^^;;
[13:38:41] <kshishkov> so FFmpeg is libnemesi's nemesis?
[13:39:25] <j-b> Flameeyes: arf
[13:41:40] <thresh> rofl
[13:46:56] <Flameeyes> j-b: we're so swamped out that I'm now writing reports... and I hate writing reports
[13:48:01] <j-b> Flameeyes: :'(
[14:08:24] <Tjoppen> it seems someone broke muxing ts to a pipe
[14:13:42] <Tjoppen> somewhere between today and two months ago.. git-bisect shall deduce whom
[14:19:53] <Tjoppen> my fault, forgot -y
[14:21:05] <spaam> Tjoppen: fail
[14:29:40] <spaam> mru: what do you say about http://www.linaro.org/snowball-announcement/ ? :D
[14:30:24] <spaam> good shit?
[14:31:14] <mru> I'm not that impressed
[14:31:30] <kshishkov> it looks smaller than Panda
[14:31:47] <mru> cheap board is good, but it doesn't end there
[14:31:49] <BBB> elenril: gotta go now, will apply patches
[14:32:07] <mru> I expect they'll release a single, old, ugly kernel patch that barely works
[14:32:11] <mru> and that'll be it
[14:32:21] <mru> or worse, and sdk
[14:32:24] <mru> *an
[14:32:54] <mru> and of course no trm
[14:35:42] <cartman> trm?
[14:35:53] <mru> technical reference manual
[14:35:56] <cartman> ah :)
[14:56:24] <lu_zero> http://bbrv.blogspot.com/2011/02/95-smarttop.html
[14:56:39] <lu_zero> cheap and more or less well supported kernel wise =P
[14:56:43] <lu_zero> </plug>
[14:58:15] <kshishkov> and I'd say partially Luca-supported too!
[14:59:57] <lu_zero> that means that you can blame me if something breaks ^^;
[15:00:08] <kshishkov> nope, I'll just run to you for help
[15:00:17] * kshishkov mostly blames Freescale
[15:00:48] <kshishkov> though it's still A8, A9-based system would be much nicer
[15:01:44] <mru> A8 is faster for some things
[15:01:52] <lu_zero> kshishkov: imx53 looks nicer but still a8
[15:03:44] <kshishkov> mru: it's just more about being dual-core
[15:04:21] <mru> better make sure your apps are multi-threaded then
[15:05:01] <kshishkov> but you can run two of them!
[15:09:33] <CIA-15> ffmpeg: Jean-Daniel Dupas <jd.dupas at ninsight.com> master * r351423ae1f ffmpeg/libavcodec/targa.c:
[15:09:34] <CIA-15> ffmpeg: targa: fix potential buffer overreads
[15:09:34] <CIA-15> ffmpeg: Signed-off-by: Mans Rullgard <mans at mansr.com>
[15:11:33] <mru> elenril: ping
[15:20:25] <lu_zero> j-b: ping
[15:22:19] <elenril> mru: pong
[15:23:12] <mru> elenril: would you mind taking a look at patchwork to see if any of your patches there need to have their status updated?
[15:24:26] <elenril> http://patches.ffmpeg.org/patch/1133/ and http://patches.ffmpeg.org/patch/1130/ are obsolete
[15:24:48] <mru> mind updating them?
[15:25:19] <elenril> i have to register for that, right?
[15:25:46] <mru> yeah
[15:25:59] <mru> and have jannau give the right permissions
[15:26:43] <jannau> you should be able to change the status of your own patches if you register with the same mail address
[15:32:56] <elenril> what does archiving do?
[15:33:30] <mru> sets the archive flag, whatever that means
[16:20:54] <uau> another issue with the lavf avi demuxer: seeks without index won't go to video keyframes
[16:22:30] <bilboed-tp> anyone has a list of rtmp urls to test ?
[16:23:15] <Compn> bilboed-tp : you dont like the ones on the wiki ?
[16:23:24] * bilboed-tp facepalms
[16:23:26] <bilboed-tp> thanks :)
[16:23:34] <mru> uau: how would you find key frames in avi without index?
[16:25:47] <uau> mru: i've never looked at the details of avi, but mplayer's old internal demuxer seems to be able to build a working index
[16:25:52] <uau> i'll see what it does
[16:25:59] <mru> probably parses the video data
[16:26:04] <mru> there's no other way
[16:26:22] <mru> the key frame flag is only in the index
[16:26:41] <mru> my personal favourite is avi files where packets and index disagree
[16:27:36] <uau> it does have this comment at least // Fix keyframes for DivX files:
[16:27:42] <uau> so could be specific to mpeg4
[16:27:58] <mru> the packet header in avi has only stream number and size
[16:30:44] <mru> "unreasonable forms of switch (e.g. Duff's device)"
[16:32:32] <uau> the relatively simple special case (selecting mode based on fourcc, and a couple of lines to parse data) seems to work pretty well in practice though
[16:32:44] <uau> as mpeg4 is the most common codec for avi
[16:34:19] <mru> of course it does
[17:20:54] <j-b> lu_zero: ?
[17:23:02] <lu_zero> diego told me to ping you
[17:58:00] <kierank> why is av_url_split not in libavutil?
[17:59:27] <elenril> because you didn't move it there
[17:59:40] <kierank> k
[18:01:35] * kierank moves it there
[18:03:35] <elenril> inb4 i get a gazillion conflicts
[18:03:42] * elenril pokes BBB 
[18:03:58] <BBB> kkk
[18:04:06] <BBB> let me have lunch for a sec
[18:04:10] <BBB> I will commit it, patch is fine
[18:04:13] <BBB> just need time to test :-p
[18:04:38] <elenril> lunch is a good idea
[18:04:45] * elenril greps house for food
[18:08:45] <kierank> elenril: remember internet connected fridges
[18:08:54] <kierank> dotcom bubble ftw
[18:11:16] <kierank> erm what do I do with FF_API_URL_SPLIT
[18:12:46] <elenril> ?
[18:13:10] <kierank> there's some old attribute_deprecated void ff_url_split lying around
[18:14:11] <elenril> make FF_API_URL_SPLIT_2?
[18:14:12] <elenril> or something
[18:14:19] <elenril> why do you want it in lavu anyway?
[18:15:19] <kierank> so I can use it for url parsing without having to include lavf
[18:19:36] * kierank is drinking the lavu kool-aid
[18:23:48] <elenril> anybody against removing url_fgets()?
[18:23:52] <elenril> it's not used anywhere
[18:24:18] <mru> is it considered public?
[18:24:30] <elenril> it's in avio, so i guess yes
[18:30:00] <elenril> err....why is url_fprintf under CONFIG_MUXERS?
[18:37:02] <kierank> dammit yahoo mail, hurry up and send my mail
[18:46:09] <_av500_> yahoo still exists?
[18:46:30] <kshishkov> you mean M$ Bing frontend?
[18:46:40] <kierank> yeah, for $1.99 domain names they still exist
[18:47:26] <_av500_> per year?
[18:47:33] <kierank> yes
[18:47:47] <kierank> or maybe for one year
[18:47:54] <kierank> still a good deal nonetheless
[19:30:17] <BBB> lu_zero: related to the URLContext.slave thing, is that a good idea?
[19:30:40] <BBB> lu_zero: we used to have this get_file_handle() thing, which didn't work at all with rtp, because rtp has two file handles, and rtsp has even more
[19:30:50] <BBB> lu_zero: when desining this, we want to make sure it fits in with rtsp also
[19:31:31] <BBB> maybe it works
[19:31:35] <BBB> haven't quite checked it out yet
[19:31:40] <BBB> just mentioning we should keep it in mind
[19:35:58] <Flameeyes> mpeg4 part 12 is iso media, right?
[19:36:12] <mru> yes
[19:36:24] <Flameeyes> thanks
[19:36:27] <lu_zero> BBB: uhm
[19:36:36] <lu_zero> right now I'm a funny issue with the protocols
[19:36:43] <lu_zero> I'm trying to complete sctp
[19:36:49] <lu_zero> to add it to our rtp/rtsp
[19:37:10] <kierank> lu_zero: people use sctp?
[19:37:15] <Flameeyes> kierank: no.
[19:37:28] <Flameeyes> lu_zero: I'd sincerely redesign it from scratch if you gave me a bit more time :)
[19:37:46] <lu_zero> kierank: some
[19:37:48] <Flameeyes> unless of course you _need_ it ...
[19:37:53] <lu_zero> Flameeyes: redesign how?
[19:37:59] <lu_zero> and what?
[19:38:06] <Flameeyes> lu_zero: let it be usable as seqpacket rather than only as a stream
[19:38:09] <lu_zero> the protocol layer or the rtsp-sctp?
[19:38:14] <Flameeyes> rtsp-sctp
[19:38:16] <lu_zero> it _should_ =_=
[19:38:31] <Flameeyes> no it cannot be used a seqpacket, we already discussed that
[19:38:35] <mru> kierank: there's a patch from you sitting in the moderation queue...
[19:38:44] <Flameeyes> for seqpacket you need to use it like it was udp, not like it was tcp
[19:38:52] <lu_zero> yup
[19:39:06] <Flameeyes> our current implementation in feng will only work as a stream; sure it keeps the multiqueue, sure it keeps sack, sure it makes a lot of sense already but it doesn't use seqpacket
[19:39:36] <kierank> mru: yeah, had to send it from another email address and forgot that it wasn't subscribed
[19:39:53] <mru> so it didn't get through yet?
[19:40:25] <kierank> mru: dunno what happened to the first one i sent from my normal email address
[19:40:40] <mru> ok, approved now
[19:40:58] <kierank> thanks
[19:41:02] <lu_zero> Flameeyes: so it's our implementation mimiking too much tcp interleaved...
[19:41:11] <Flameeyes> yah
[19:41:27] <Flameeyes> right now we have a tcp interleaved on steroid.. not bad, but not great either
[19:41:27] <lu_zero> (not that the way I'm trying to implement sctp in ffmpeg is that different...
[19:42:01] <lu_zero> and even there we have some stream-even-when-we-have-packets issue ^^;
[19:42:07] <kierank> lu_zero: can feng do rtsp vod? someone was asking me the other day
[19:43:45] <lu_zero> kierank: yup
[21:20:59] <saintd3v> BBB: ping
[21:31:14] <BBB> pong
[21:32:50] <BBB> saintd3v: privmsg me, gotta run
[21:32:59] <saintd3v> was going to anyway =P
[21:43:50] <ruggles> hmm. it seems there is a file in the fate suite that isn't actually tested.
[21:43:59] <ruggles>  /iff/8svx_fib.iff
[21:46:35] <kierank> lol iff
[22:41:32] <Sean_McG> it's Friday!
[22:41:38] <Sean_McG> and almost beer o'clock
[22:42:02] <CIA-15> ffmpeg: Jason Garrett-Glaser <jason at x264.com> master * r902685b8ab ffmpeg/libavcodec/vp3.c:
[22:42:02] <CIA-15> ffmpeg: VP3: fix decoding of videos with stride > 2048
[22:42:02] <CIA-15> ffmpeg: Also remove qscale_table code; this didn't make sense anyways as VP3 doesn't
[22:42:02] <CIA-15> ffmpeg: use an MPEG-like quantizer scale.
[22:57:20] <kierank> mru: what happened to that tomi cpu?
[22:57:56] <ruggles> in case this is useful to anyone: http://ffmpeg.pastebin.com/fGaRYVLF
[23:28:34] <Sean_McG> mru: those odd characters in FATE reports can be fixed if you export LANG=C and LC_*=C in fate.sh
[23:29:36] <Sean_McG> mru: gcc doesn't seem to like 'en_CA.UTF8'
[23:34:58] <BBB> Dark_Shikari: let's discuss here then
[23:35:17] <BBB> Dark_Shikari: so I want to add a fixed value to each dc/ac coeff, like (for int i=0;i<64;i++)block[i]+=1;
[23:35:29] <BBB> Dark_Shikari: that should be easy, like adding 1 to the block[0] before idct
[23:35:33] <Dark_Shikari> yes
[23:35:38] <BBB> Dark_Shikari: but that doesn't add exactly 1 to each
[23:35:47] <BBB> rather, it usually adds one, sometimes zero and sometimes two
[23:35:48] <BBB> why?
[23:36:00] <BBB> (vc1 idct, not h264 idct, in case you're wondering)
[23:36:42] <bcoudurier> init last dc to some value
[23:36:48] <Dark_Shikari> BBB: dither?
[23:36:57] <BBB> (also, it's not actually 1:1, it's more like adding 0x100 to each after idct == adding 0x380 before idct, I guess that's scaling?
[23:37:24] <iive> 2048->256 ?
[23:38:06] <bcoudurier> Init last dc to 1024 for i += 128
[23:39:57] <BBB> I thought block[0] is the dc and the rest is ac?
[23:40:20] <BBB> so which is the "last" dc?
[23:40:36] <astrange> last dc = previous dc
[23:40:47] <astrange> used in dc prediction
[23:41:21] <BBB> oh
[23:43:39] <BBB> but doesn't do the same thing as (just before, or even in the idct) doing block[0] += somevalue;?
[23:45:26] <bcoudurier> well it's faster, even if it's the same
[23:46:07] <BBB> right, but my problem is that it dithers (or something like that) :(
[23:46:09] <bcoudurier> ie dc is init at the beginning of each row, although it might be different for vc1
[23:46:59] <bcoudurier> anyway if the vc1 decoder do it this way, there might be a reason
[23:49:49] <iive> BBB: i didn't quite get what are you doing, why, what result do you expect instead.
[23:50:46] <BBB> vc1 decoder does this: read DCT'ed coeffs from bitstream, run idct, then for (i=0;i<64;i++) block[i] += 128; (or 64, or whatever, sometimes with an if, depending on which part of the decoder you look at)
[23:51:07] <BBB> I was wondering if you could do this without the loop by adding 128 or 64 under that "whatever if" to block[0] before idct
[23:51:10] <BBB> so you get rid of the loop
[23:51:21] <BBB> but that changes the post-idct coeffs slightly, i.e. the dithering
[23:51:33] <BBB> most have 128 added, but some 127, some 129
[23:51:38] <bcoudurier> well the dnxhd decoder does it like I told you
[23:51:40] <bcoudurier> and it works
[23:51:48] <bcoudurier> dnxhd is mpeg2 basically
[23:51:51] <BBB> vc1 is probably weird ...
[23:51:58] <bcoudurier> I don't know what vc1 does, but something funky is happening
[23:52:03] <BBB> yeah
[23:53:15] <Yuvi> add/put_pixels_signed vs. unsigned?
[23:54:48] <BBB> Yuvi: that's what it looks like, yes
[23:57:05] <Yuvi> do that then I think, either merge it w/ the idct as idct_add/idct_add_signed or keep it separate and change a local put_pixels pointer
[23:57:36] <Dark_Shikari> not merging -> slower


More information about the FFmpeg-devel-irc mailing list