[FFmpeg-devel-irc] IRC log for 2011-02-26

[irc at mansr.com]

[00:02:15] <skal> ;5D/win goto 2
[00:02:18] <skal> blah
[00:29:45] <pross-au> fast decoder != on2 profit
[00:29:59] <kierank> of course
[00:30:17] <kierank> but a quality product is conducive towards more profit
[00:30:48] <pross-au> agree totally
[00:31:06] <mru> for that it has to be actually good
[00:31:16] <pross-au> besides, you gotta leave scope for vp9
[00:31:16] <mru> not merely good enough to sell
[00:31:22] <pross-au> otherwise be out of a job
[00:31:56] <mru> the profit/effort is often higher in churning out a series of mediocre products and spending some money on marketing hype
[00:32:29] <pross-au> precisely
[00:32:43] <pross-au> theres something inherently wrong with that, the world
[00:35:13] <mru> solution: shoot the PHBs
[00:40:47] <CIA-15> ffmpeg: Ronald S. Bultje <rsbultje at gmail.com> master * r13ff92d197 ffmpeg/libavformat/movenc.c:
[00:40:47] <CIA-15> ffmpeg: movenc: remove uses of deprecated API.
[00:40:47] <CIA-15> ffmpeg: Replace put_tag() with ffio_wfourcc() and ByteIOContext with AVIOContext.
[00:41:24] * Terminating due to: TERM
[00:41:37] * /join #ffmpeg-devel ...
[00:41:41] *** TOPIC: Welcome to the FFmpeg development channel. | Discussions about the development of FFmpeg itself are ontopic here. | Questions about using FFmpeg or developing with the libav* libraries should be asked in #ffmpeg. | Channel is logged.
[00:41:41] *** TOPICINFO: merbzt!~benjamin at m83-186-120-172.cust.tele2.se, 1295456661
[01:26:18] <BBB> Dark_Shikari: don't forget that "sharing code between ffvp8 and rest of ffmpeg" is our attempt at a PR thing
[01:26:36] <BBB> Dark_Shikari: it doesn't have to be super-significant, I mean, "goldenframes, woohoo!!!", it's all about how it sounds to the public
[01:27:02] <BBB> Dark_Shikari: and sharing code sounds fantastic to the public :)
[01:27:37] <kierank> PR thing and a troll at the same time
[01:27:45] <kierank> bbbut vp8 uses h.264 prediction
[01:27:49] <kierank> what about the patents!
[01:29:12] <gnafu> OMGNO!
[01:29:22] <gnafu> :-P
[03:29:36] <Dark_Shikari> http://pic.phyrefile.com/s/sf/sft/2011/02/24/keepern_1.png
[03:29:47] <Dark_Shikari> screenshots from the first commercial fully-x264-encoded bluray.
[03:30:11] <mru> not bad
[03:30:17] <mru> what film?
[03:30:17] <Dark_Shikari> http://pic.phyrefile.com/s/sf/sft/2011/02/24/keepern_2.png
[03:30:19] <Dark_Shikari> http://pic.phyrefile.com/s/sf/sft/2011/02/24/keepern_3.png
[03:30:28] <Dark_Shikari> http://www.imdb.com/title/tt1488574/
[03:30:36] <Dark_Shikari> Norwegian comedy film.
[03:33:25] <Dark_Shikari> Note the error-diffusion dither from 10-bit input.
[03:34:41] <mru> so those are the good frames, what are the bad ones like?
[03:35:02] <Dark_Shikari> Dunno, those were just the ones someone posted.
[03:35:23] <Dark_Shikari> There's one bit of information those frames give you: it wasn't lowpassed.
[03:35:39] <mru> of course, bluray has enough bitrate that bad frames aren't really excusable
[03:36:28] * saintd3v waits for the blug post
[03:36:30] <saintd3v> =P
[03:36:45] <Dark_Shikari> I have more important things to do, like play more civ4
[03:37:01] <saintd3v> blog about civ4!
[03:38:00] <mru> when will they include blogs as a game element?
[03:39:16] <j0sh> isn't civ5 out already
[03:39:20] <Dark_Shikari> yes.  it's crap.
[03:39:21] <saintd3v> even better, little awards that are automatically posted to your facebook prof....oh, wait...
[03:39:46] <j0sh> i swore off civ after i started college. wasted too much time in high school on civ3
[03:39:48] <mru> saintd3v: no, I mean virtual in-game-only blogs
[03:40:05] <saintd3v> mru: virtual in-game facebook, lol
[03:40:34] <mru> http://www.youtube.com/watch?v=Rw8gE3lnpLQ
[03:41:26] <saintd3v> onion \o/
[03:41:49] <drv> there's already an in-game facebook in the newest need for speed
[03:42:14] <mru> I thought need for speed was a racing game
[03:42:16] <kierank> there's also those random microsoft things in arkham asylum
[03:42:18] <drv> it is
[03:42:20] <mru> clearly I have not been paying attention
[03:42:23] <drv> when you beat your friends, it posts that to your "wall"
[03:42:32] <mru> oh, that's something else
[03:42:40] <saintd3v> kierank: you mean the achievements?
[03:42:50] <kierank> yes
[03:43:04] <saintd3v> yeah, achievements are old..
[03:43:16] <kierank> yeah but it posts them to your hotmail account
[03:43:24] <kierank> and iirc you couldn't start playing without logging in
[03:43:34] <saintd3v> yeah you can
[03:43:52] <saintd3v> you have to setup an offline games for windows account
[03:44:12] <saintd3v> and you can't carry your game progress over to an actual games for windows account
[03:44:58] <saintd3v> found that out _after_ i had beaten the game, and wanted to go back and get some of the achievements :/
[03:45:07] <Dark_Shikari> "achievements" lololol
[03:45:48] <saintd3v> Dark_Shikari: well they tell more of the story
[03:45:54] <kierank> they do?
[03:45:55] <Sho_> The only achievement I'm perpetually unlocking is not to care about the achievements metagame
[03:46:43] <mru> some game I was playing once crashed in a curious fashion
[03:46:45] <saintd3v> kierank: all of arkham's writings
[03:47:15] <kierank> oh those. don't care about them
[03:47:16] <mru> after restarting the ps3, it listed all the extras as completed
[03:47:17] <saintd3v> needless to say i didn't feel like beating the game again just to do it, so i gave up
[07:40:36] <rukhsana> Hi
[07:41:16] <rukhsana> 'I want to participate in GSoC this year'
[07:42:06] <rukhsana> 'I want to work on JPEG 2000 project which was abandoed last year'
[07:43:31] <kierank> did you complete your qualification task
[07:43:43] <rukhsana> 'I already have created an account in gitorious repository in order to create a clone to work on ffmpeg'
[07:44:20] <rukhsana> 'Actually I just have started, I did not pass the qualification task yet'
[07:44:34] <in3xes_> how to debug demuxer_probe()
[07:44:48] <kierank> rukhsana: I mean your other qualification task
[07:45:16] <kierank> top-posting one
[07:45:35] <rukhsana> 'yes, I have passed that probably'
[07:46:11] <rukhsana> 'right now, I know what is top-posting and will avoid it from now on'
[07:46:45] <kierank> why do you write in single quotes?
[07:47:04] <rukhsana> I did not user IRC before
[07:47:20] <rukhsana> this is the first time i am using ITC chat
[07:47:40] <rukhsana> that's why I was probably making mistake
[07:48:01] <rukhsana> Thanks for the correction
[07:48:49] <kierank> do you have the jpeg2000 spec?
[07:48:51] <in3xes_> calling av_log() in xxx_probe() and `ffprobe input_sample` should work?
[07:50:09] <rukhsana> I just have the JPEG 2000 code downloaded from svn. however, i did not see spec for JPEG 2000
[07:52:27] <kierank> lemme see if i can find it
[07:53:53] <rukhsana> Thank you kierank
[07:59:34] <kierank> http://dl.dropbox.com/u/2701213/Specs/JPEG2000/15444-1annexi.pdf
[07:59:47] <kierank> seems to be on the mplayerhq ftp
[08:00:07] <rukhsana> Thank you very much
[08:00:18] <rukhsana> I will go through it
[08:00:47] <kierank> probably easier to try understanding the code
[08:00:59] <kierank> or finding a book that summarises jpeg2000
[08:01:41] <rukhsana> ok, I will go through the code too
[08:02:05] <kierank> i would strongly recommend finding a book that explains the basics otherwise you will get nowhere
[08:02:19] <rukhsana> I will get back to you guys soon.
[08:03:09] <rukhsana> I will try to find a book too
[08:04:04] <kierank> something like http://www.amazon.co.uk/JPEG2000-Compression-Fundamentals-International-Engineering/dp/079237519X/ref=sr_1_1?ie=UTF8&qid=1298707406&sr=8-1
[08:04:50] <rukhsana> ok, thank you very much
[08:05:55] <in3xes> Suggest me simplest demuxer to look at the code
[08:12:40] <kierank> in3xes: wav
[08:12:52] <saintdev> peloverde: ping
[08:13:44] <in3xes> Ok
[08:14:04] <peloverde> saintdev: pong
[08:14:50] <saintdev> peloverde: would you mind doing a review on my current tree?
[08:15:46] <peloverde> I should have some time for it tomorrow, it's kind of late here right now, and I'm knee deep in minecraft
[08:16:13] <saintdev> minecraft, hehe
[08:19:02] <peloverde> now that I've gotten the hang of it, it has become fun
[08:19:48] <saintdev> built yourself a computer yet?
[08:22:07] <peloverde> yes, the next goal is to port ffmpeg to it :)
[08:23:29] <saintdev> new arch on fate?
[08:24:38] <pJok> ffminecraft?
[08:25:07] <saintdev> minecraft-ffmpeg-minecraftcc-1.0
[08:25:15] <kierank> cuda edition (tm)
[08:27:34] <pJok> well
[08:27:37] <pJok> its all java
[08:27:44] <pJok> so it will take ages to get anywhere ;)
[08:57:47] <saintdev> grr, how come every command line pastebin script seems to be broken :(
[09:10:33] <saintdev> peloverde: whenever you get a chance. I've still got more work to do, so no rush. http://pastebin.com/2ENEGzxQ
[09:13:24] <saintdev> oops, forget the top change =P
[10:15:35] <cryptw> saintdev: use sprunge
[11:32:45] <BBB> mru: please update clang, 126540 should fix it
[11:33:02] <mru> updating
[11:33:25] <mru> so it was a clang bug
[11:33:30] <Dark_Shikari> What'd they change?
[11:35:00] <kshishkov> switched to pure C?
[11:45:16] <mru> hmm, there's still a fixme in the code
[11:45:22] <mru> let's break it!
[11:45:27] <Dark_Shikari> link to the diff?
[11:45:29] <Dark_Shikari> or the commit?
[11:45:33] <mru> http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGExpr.cpp?r1=126540&r2=126539&pathrev=126540
[11:46:28] <Dark_Shikari> oh ho ho
[11:49:28] <mru> updated
[11:49:38] <Dark_Shikari> ?
[11:49:46] <Dark_Shikari> oh, on the fate box.
[11:52:43] <CIA-15> ffmpeg: Mans Rullgard <mans at mansr.com> master * raa3805a486 ffmpeg/configure:
[11:52:43] <CIA-15> ffmpeg: fate: get samples location from env var if not explicitly set
[11:52:43] <CIA-15> ffmpeg: Use the FATE_SAMPLES environment variable if samples location
[11:52:43] <CIA-15> ffmpeg: is not set with the --samples configure option or on the make
[11:52:43] <CIA-15> ffmpeg: command line.
[11:52:44] <CIA-15> ffmpeg: Signed-off-by: Mans Rullgard <mans at mansr.com>
[11:56:10] <wbs> mru: I'm getting errors building the vp8 neon assembler code with the android NDK, these errors: http://pastebin.com/2fGRcamE
[11:56:28] <mru> buggy assembler
[11:56:34] <mru> and why are you building PIC?
[11:57:30] <wbs> good question
[11:57:40] <kurosu> ndk mandates it
[11:57:48] <mru> madness
[11:57:57] <kurosu> check toolchain/<version>/setup.mk
[11:58:05] <MrNaz> ok i'm getting this:     No such filter: 'pixelaspect'    <--  didn't the syntax for setting the PAR *just* change recently?    http://paste.pocoo.org/show/344767/   <--- that's the command i'm using... ffmpeg and libx264 were built using current svn pull (an hour ago) and latest x264 tarball (also an hour ago)... (yes i know svn is deprecated i'll rebuild from git once i solve this)
[11:58:06] <wbs> normally I do build it as shared libraries, to ease LGPL compliance in a commercial app
[11:58:09] <mru> they don't allow sharing libs between processes anyway, do they?
[11:58:15] <MrNaz> woops that was meant for #ffmpeg
[11:58:16] <Dark_Shikari> mru: ffmpeg doesn't use svn anymore
[11:58:25] <kurosu> TARGET_CFLAGS :=     -fpic     -mthumb-interwork     -ffunction-sections     -funwind-tables     -fstack-protector    -fno-short-enums     -D__ARM_ARCH_5__ -D__ARM_ARCH_5T__     -D__ARM_ARCH_5E__ -D__ARM_ARCH_5TE__
[11:58:25] <mru> MrNaz: ^^
[11:58:49] <mru> omg what a broken command line
[11:59:10] <MrNaz> mru mine?
[11:59:12] <kurosu> yeah, the shared libs are specific to an application, nobody except the system is allowed to access them
[11:59:23] <mru> MrNaz: no, kurosu's
[11:59:33] <kurosu> ndk's, please :)
[11:59:40] <kurosu> I'm not responsible for it :)
[11:59:40] <mru> kurosu: so there's nothing to be gained from pic
[12:00:01] <kurosu> there's a lot of weird stuff anyway, maybe making more sense to you than me
[12:00:30] <kurosu> the only arch that can be targeted are armv5 or armv7, no armv6 and its FPU
[12:00:55] <kurosu> (i may not make sense or use proper words, but i guess you should get what i mean)
[12:01:30] <mru> wbs: does this fix it? http://pastie.org/1609516
[12:01:56] <mru> ok, here's what's wrong with those flags:
[12:02:09] <mru> -fpic makes code slow for no benefit (since sharing is not allowed)
[12:02:26] <mru> thumb interworking is only required on armv4t, which is not supported
[12:02:50] <mru> -ffunction-sections can help remove dead code, but can also impact performance
[12:03:09] <wbs> mru: yes, that helps, thanks :-)
[12:03:19] <kurosu> how much slower for fpic? i thought this was mostly true for x86 due to register pressure
[12:03:30] <mru> -funwind-tables is automatically enabled as needed
[12:03:54] <mru> -fstack-protector is rather pointless in a sandbox environment
[12:04:17] <kurosu> for thumb interworks, i guess it's dangerous as other parts of the system may depend on it (you do have access to system libs compiled with that)
[12:04:26] <mru> -fno-short-enums is default
[12:04:57] <mru> the __ARM_FOO__ symbols are defined automatically based on -march
[12:05:20] <mru> kurosu: thumb interworking is simply not necessary on v5 and up
[12:05:43] <kurosu> ok
[12:05:45] <mru> v5t has the bx and blx instructions instead
[12:06:02] <BBB> kurosu: you were right btw, vc1 idct must be done in 24 bits (and thus in 32 bits/int)
[12:06:04] <kurosu> other cflags: -march=armv5te -msoft-float
[12:06:30] <mru> see, that -march already defines all the __ARM_FOO__ things
[12:06:34] <kurosu> BBB: i was wrong in fact, hoping we could get away with it
[12:06:50] <mru> at least they don't conflict...
[12:06:56] <kurosu> mru: yeah they add counter-effect cflags
[12:07:09] <kurosu> not conflict, but break what the other implies
[12:07:18] <BBB> kurosu: oh, well, join the club then, we were both wrong
[12:07:28] <BBB> kurosu: I think you can almost get away with it, but it'll fail in corner cases
[12:07:35] <BBB> kurosu: so needs 32bit I'm affraid
[12:07:46] <kurosu> BBB: maybe branch based on qp ? (not sure at all)
[12:07:59] <mru> apparently you can get away with it in most real videos
[12:08:02] <kurosu> that's a lot of efforts probably
[12:08:29] <BBB> I was actually thinking that also
[12:08:49] <BBB> not just qp, but still
[12:08:52] <BBB> something like that
[12:09:09] <BBB> need to hook up the vc1 testsuite and see if it fails
[12:09:13] <BBB> if it fails, then that's good
[12:09:20] <BBB> if it doesn't, we don't care :)
[12:09:22] <mru> kurosu: it's not as bad as what a hardware vendor insisted we use once: -mbig-endian ... -mlittle-endian
[12:09:28] <kurosu> assuming 16bits arith will... at most double the throughput of the function, right? but would that really improve much? vc1 subpel interpolation still uses mmx, there'd be probably much more to be gained to add ssse3 versions
[12:09:28] <mru> of course it broke
[12:09:49] <BBB> kurosu: idct8x8 takes 15% here
[12:09:52] <BBB> kurosu: so that first
[12:09:57] <BBB> kurosu: but I'll do the others also
[12:09:59] <Dark_Shikari> subpel takes less than idct?
[12:10:06] <BBB> currently yes
[12:10:08] <Dark_Shikari> have you summed up all the subpel functions?
[12:10:12] <BBB> no
[12:10:18] <kurosu> BBB: i mean having 16bits and 32b arith dct functions rather than ssse3 subpel
[12:10:22] <Dark_Shikari> They probably use more when summed up
[12:10:28] <BBB> possibly
[12:10:32] <kurosu> mru: a *hardware* vendor ? do they not understand what they build?
[12:11:58] * kurosu goes on removing cflags from its ndk command lines
[12:12:30] <kurosu> it feels like cranking up to -funroll-loops=11!
[12:13:14] <mru> kurosu: no, they do not
[12:13:35] <mru> they take a chip and a ref board, slap their logo on it, and call it a product
[12:14:16] <kurosu> yeah, i thought of writing rather manufacturate, and vendor says it all anyway
[12:14:35] <mru> I don't remember which one it was
[12:15:10] <kurosu> i'm just doing this as a hobby anyway, so the non-recommandation will not be of much use to me :)
[12:15:21] <kurosu> and they are probably already bankrupted now :D
[12:15:25] <mru> I doubt it
[12:15:47] <kshishkov> kurosu: hw and sw vendors offering silliest stuff usually survive
[12:15:47] <mru> likely candidates are thomson and humax
[12:16:03] <kshishkov> mru: there's no thomson anymore
[12:16:08] <mru> who bought them?
[12:16:12] <wbs> I argued with an indian once... there was an api for setting/getting key-value pairs, like set(uint32_t key, uint32_t value); set(uint32_t keu, float value); (c++ with overloading etc)... for the keys, there were definitions like uint32_t keyFoo = 0x01; uint32_t keyBar = 0x02; etc. then he added float keyBaz = 0x04;, and I was unable to convince him that the key still should be uint32_t even if the _value_ stored with the ...
[12:16:18] <wbs> ... key was float
[12:16:31] <kshishkov> mru: they bought Technicolor and renamed themselves to it
[12:16:42] <mru> so they still exist
[12:16:45] * kierank knows of someone at humax
[12:17:03] <mru> humax are a real pain to work with
[12:17:08] <kshishkov> mru: of course, along with MP3
[12:17:28] <kierank> mru: the guy who I spoke to seemed quite helpful...
[12:17:43] <kurosu> thomson *is* bankrupted essentially, their debt was pushed back again and again
[12:17:55] <mru> kurosu: good
[12:18:02] <kurosu> i think it's the other way around: technicolor bought them, afaik
[12:18:03] <kshishkov> excellent!
[12:18:05] <mru> kierank: they're friendly but mostly clueless in my experience
[12:18:14] <mru> at least the people they'll let you talk to directly
[12:18:27] <kurosu> i doubt banks would like an almost-dead company buying companies
[12:18:32] <kshishkov> kurosu: next thing yu'll say that SCO bought Caldera
[12:18:47] <kurosu> kshishkov: i'm indeed unsure, but that didn't make sense to me
[12:18:54] <kierank> kurosu: the thomson h.264 encoder is quite good
[12:19:06] <kurosu> is that grassvalley rather ?
[12:19:33] <kierank> well I think it's all under the technicolor umbrella now
[12:19:41] <kierank> but yes grassvalley
[12:20:30] <mru> thomson gazelles grazing?
[12:20:35] * mru sends in the lions
[12:20:39] <kshishkov> kurosu: http://en.wikipedia.org/wiki/Technicolor_SA
[12:21:04] <kurosu> yeah was checking that
[12:21:17] <kurosu> didn't know it was already a subsidiary
[12:23:04] <kshishkov> and they have shitloads of ideas about adding AAC extensions to MP3 too. Luckily none of them really catched up
[12:24:38] <kurosu> kshishkov: though "they bought Technicolor and renamed themselves to it" was a bit misleading to me, as the 2 events seems quite separated in time (and having no link of causality); not an important matter anyway
[12:54:52] <CIA-15> ffmpeg: Reimar Döffinger <Reimar.Doeffinger at gmx.de> master * r52b3cc6047 ffmpeg/configure:
[12:54:52] <CIA-15> ffmpeg: configure: document FATE_SAMPLES env var in --help text
[12:54:52] <CIA-15> ffmpeg: Signed-off-by: Mans Rullgard <mans at mansr.com>
[13:07:21] <mru> clang is back to green \o/
[13:26:42] <mmu_man> plop
[13:26:46] <mmu_man> Support the nomination of the International FLOSS Community to Prince of Asturias Award 2011 : http://www.cenatic.es/swlppa/en
[13:26:49] <mmu_man> \o/
[13:28:38] <BBB> mru: awesome :)
[13:31:57] <mmu_man> clang... reminds me the name of the bad ET in TMNT
[13:32:08] <mmu_man> ah no that was Crang :D
[13:33:02] * kshishkov should remind mmu_man what most people outside Japan think about haikus
[13:33:34] <mmu_man> that it's a fine OS that rulz d4 w0rlD
[13:34:08] <kshishkov> no, that's rather obscure short pieces nobody except its author cares about
[13:35:26] <mru> so it's an apt name for that OS
[13:38:09] <mmu_man> hmm that deserves a FATE run to slap you :p
[13:46:23] <mru> how the fuck does one build clang as a cross-compiler?
[13:56:32] <BBB> I guess I have to wait for kostylev to update his systems? I told him it's fixed
[13:56:53] <mru> which ones?
[13:57:04] <mru> ah, the bsds
[14:00:28] <BBB> yeah
[14:00:39] <BBB> so, most x86-like failures are now on icc
[14:00:50] <BBB> is anyone on top of helping the icc devs figuring out their mistakes?
[14:01:01] <BBB> I think I tried and it made no difference
[14:01:06] <BBB> I need a contact in their dev team maybe
[14:01:15] <mru> each release seems to fix some bugs and add new ones
[14:01:22] <kshishkov> try somebody in America with cluebat
[14:01:39] <BBB> mru: that's my impression also
[14:01:57] <kshishkov> mru: it's called M$ release
[14:12:34] * pross-au working on: s/get_bits/ffbits_r/g
[14:17:41] * mru curses clang
[14:31:27] * elenril summons some reviewers for avio_get_str
[14:35:21] <elenril> any reason to use url_feof() instead of checking AVIOContext.eof_reached directly?
[14:35:54] <mru> for that authentic java feeling (tm)
[14:36:07] <elenril> i'm deprecating it then
[14:36:20] <mru> I have no idea why it's there
[14:36:32] <elenril> because it's always been there?
[14:38:30] <kshishkov> it's shorter and reflects stdio
[14:45:43] <elenril> avio > stdio =p
[14:58:03] <Kovensky> why not rename eof_reached to eof
[14:58:10] <Kovensky> solves the length issue :>
[15:02:02] <kshishkov> well, I feel a bit uneasy accessing structure contents not created by me
[15:08:07] <elenril> you should switch to java then =p
[15:08:50] <Yuvi> mru: clang doesn't really support cross compiling except from darwin -> darwin
[15:09:03] <mru> Yuvi: so I've noticed
[15:09:12] * kshishkov translated elentril into JVM bytecode and runs garbage collection over him
[15:09:31] <Yuvi> using the hidden options -ccc-host-triple and -ccc-gcc-name somewhat works for cross compiling, but isn't really supported
[15:09:33] <mru> I see a lot of talk about how great a cross-compiler clang is and how they're going to improve cross-compilation
[15:10:21] <kshishkov> mru: so you've been to FOSDEM
[15:10:42] <mru> that talk has been going since at least 2009
[15:10:47] * elenril doesn't stab kshishkov 
[15:10:52] <mru> and nothing ever materialises
[15:11:17] <kshishkov> mru: Apple toolchain for ARM is not completely nothing
[15:11:18] <mru> Yuvi: great, even more hidden flags
[15:11:24] <mru> I already found about 4 sets of hidden flags
[15:11:41] <Yuvi> mru: well, in this case you really aren't supposed to be using them ;)
[15:14:44] <Yuvi> I'm pretty sure they want something like -ccc-host-triple as the only option you have to specify for cross compile, but I dunno what they'll do about linking
[16:02:17] <mru> Yuvi: I've found a magic incantation to make clang spit out arm code
[16:02:31] <mru> but I can't make it do hard-float abi
[16:02:46] <mru> none of the 8 or so different flags for that seem to make a difference
[16:03:06] <kshishkov> mru: maybe hadfp is not glamorous enough to be supported
[16:03:18] <mru> the help output claims it is
[16:03:23] <mru> but the argument is ignored
[16:03:35] <mru> sometimes with a warning, sometimes not
[16:08:20] <Yuvi> mru: Does clang -v say anything about the float abi?
[16:09:08] <mru> if I use flags like -mllvm -float-abi=hard they show up there
[16:09:13] <mru> but then seem to go nowhere
[16:10:51] <mru> -Xclang -msoft-float results in pure softfloat code
[16:11:00] <mru> otherwise I get vfp code with softfloat abi
[16:11:39] <mru> -mllvm -soft-float has no effect
[16:11:58] <Yuvi> -Xclang?
[16:12:14] <Yuvi> It's possible llvm doesn't implement it I forget
[16:12:19] <mru> don't ask me what it is
[16:12:33] <mru> -mllvm -help lists it as an option
[16:12:51] <mru> and the source code suggests it does something with it during code gen
[16:13:02] <mru> but the option is lost somewhere along the way
[16:17:25] <mru> let's patch the default and see what happens
[16:28:46] <lu_zero> mru: what are you doing?
[16:28:58] <mru> messing with compilers
[16:33:17] <mru> that worked
[16:36:47] <spaam> mru: going to use clang for arm ?
[16:37:21] <mru> more like going to watch it fail
[16:40:21] <spaam> can it compile some stuff for arm?
[16:45:11] <mru> it hates arm inline asm apparently
[16:46:28] <mru> at least some instructions
[16:46:56] <spaam> neon stuff?
[16:47:04] <mru> some armv6 stuff
[16:48:19] <mru> hey, it built
[16:48:36] <spaam> :)
[16:48:54] <mru> but running is another matter
[17:25:11] <mru> found one bug
[17:25:17] <mru> in clang that is
[17:25:24] <mru> it creates bad relocations
[17:26:08] <mru> enabling pic prevents that particular relocation type, so the code runs
[17:27:07] <mru> now it even passes some tests
[17:27:11] <mru> *some*
[17:27:28] <lu_zero> nice
[18:15:22] <_av500_> gm
[18:15:56] <mru> morning
[18:16:19] <_av500_> we are out of beer
[18:16:35] <mru> sounds awful
[18:17:01] <kshishkov> so you use Internet as cheap beer substitute?
[18:17:03] <_av500_> well, red wine and lot of spirits are still there
[18:17:16] <mru> it ain't the same
[18:17:35] <_av500_> mru: too much beer makes me not board well
[18:18:22] <_av500_> so i prefer the high % stuff atm
[18:19:00] <kshishkov> what mountains you are at?
[18:19:40] <_av500_> .at
[18:20:06] <lu_zero> oh
[18:20:11] <kshishkov> Harz? Voralrberg?
[18:20:21] <lu_zero> red wine in .at?
[18:20:40] * lu_zero knows some white wines
[18:21:28] <_av500_> lu_zero: imported
[18:21:38] <_av500_> kshishkov: zell am ziller
[18:21:55] <lu_zero> ah
[18:22:11] <lu_zero> going to detour a bit?
[18:22:31] <_av500_> detour?
[18:22:47] <mru> detour de france?
[18:23:22] <kshishkov> _av500_: big deal, I have at least two Zells in < 2hrs of train
[18:23:49] <lu_zero> _av500_: to taste some troll beer =P
[18:24:06] <kshishkov> lu_zero: Austrian trolls are tasteless indeed
[18:24:33] <lu_zero> kshishkov: =)
[18:24:48] <mru> next clang bug: misaligning static data
[18:25:01] <kshishkov> lu_zero: ask our local troll expert if you don't believe me
[18:26:35] <lu_zero> mru: by the end of the day you'll be the top reporter...
[18:27:08] <mru> they'll probably just blame it on my build methods
[18:27:33] <kshishkov> mru: well, there are many compilers out there, some will listen
[18:27:50] <lu_zero> kshishkov: s/many/2/ ?
[18:28:13] <kshishkov> lu_zero: armcc, ticc too
[18:28:39] <mru> when I downloaded that iar compiler I got an automatic "personal" email "from" some support person
[18:28:59] <mru> offering help with my evaluation
[18:29:14] * kshishkov thought it would be threats
[18:29:22] <mru> I'm thinking of writing something snarky about the quality
[18:29:43] <kshishkov> "usable only on enterprise code"
[18:38:10] <lu_zero> ticc?
[18:38:41] <Sean_McG> hey mru how the heck do you get gcc 4.0.4 on Gentoo PPC, it's like, not even marked unstable
[18:39:33] <mru> Sean_McG: "sys-devel/gcc **" in package.keywords
[18:39:39] <Sean_McG> ahhh
[18:39:40] <Sean_McG> cheers
[18:39:58] <lu_zero> ^^;
[18:42:20] <mru> I even run gentoo on arches it doesn't know about
[18:42:40] <Sean_McG> wow
[18:42:49] <Sean_McG> that blackfin box?
[18:42:55] <mru> takes a little coercion but it works
[18:42:59] <mru> avr32
[18:43:01] <spaam> mr cool award  goes do.. *drums* mru
[18:43:14] <mru> the blackfin too
[18:43:21] <mru> but there's barely anything on it
[18:43:33] <mru> uclibc, busybox, and dropbear only
[18:43:43] <mru> a 5-line init script
[18:43:46] <Sean_McG> how much RAM on it?
[18:48:41] <mru> 64MB
[18:49:03] <mru> I'm doing some rather evil things with that though
[18:49:17] <mru> to avoid fragmentation
[18:49:21] <kshishkov> fax-g3 test for FATE?
[18:49:47] <mru> that's trying to allocate more memory than is available
[18:49:58] <spaam> mru: like what?
[18:50:56] * Sean_McG has a chai latte 
[18:51:09] <Sean_McG> of the Tassimo variety, sadly
[18:52:34] <mru> spaam: like setting aside a portion of the ram and with kernel patches and linker flags placing ffmpeg's bss+heap there
[18:53:05] <Sean_McG> ah so a nonstandard config then
[18:53:12] <mru> spares the kernel the effort of trying to allocate all those megabytes contiguously every time
[18:53:20] <mru> which fails pretty quickly
[18:53:55] <mru> it's still not stable though
[18:54:04] <mru> I've no idea what's wrong with it
[18:56:13] <Sean_McG> hey what fixed the clang build?
[18:56:25] <mru> clang
[18:56:29] <Sean_McG> ah
[19:18:46] <BBB> mru: have you ever been able to look into writing emu_edge versions of the frext h264 conformance tests for fate?
[19:19:40] <mru> I forgot all about that actually
[19:37:04] <_av500_> j-b: http://www.flickr.com/photos/av500/5471488290/
[20:06:08] <BBB> mru: where is h264.mak included in soem other makefile?
[20:06:19] <mru> grep?
[20:06:43] <BBB> oh nm, toplevel Makefile
[20:06:48] <BBB> I grepped all over tests/ and couldn't find it
[20:08:37] <BBB> http://ffmpeg.pastebin.com/Srkjszfd why doesn't that work?
[20:08:58] <BBB> tests/fate/vc1.mak:1: *** Recursive variable `FATE_VC1_AP_L0' references itself (eventually).  Stop.
[20:09:07] <BBB> h264.mak does that also, appears to work there
[20:09:26] <mru> that's very repetitive
[20:09:55] <BBB> sorry dude I suck at makefiles and this won't go upstream anyway
[20:09:59] <BBB> :)
[20:10:05] <BBB> (I don't own copyrights over the testsuite)
[20:11:48] <mru> remove the \ on line 4
[20:14:22] <BBB> ah
[20:14:37] <BBB> \o/
[20:14:41] <BBB> stupid me, sorry :)
[20:14:52] <BBB> now I gotta generate all output and make sure it's ok
[20:15:16] <BBB> I guess there's no quick way to do that?
[20:15:16] <mru> does the reference decoder work?
[20:15:46] <BBB> I can look at it, but I was really just planning to use this to validate our C agianst our ASM
[20:16:02] <BBB> or, well, the other way around
[20:16:05] <BBB> validate our ASM against our C
[20:16:13] <mru> then you can simply run it once and copy the output files to the ref location
[20:16:19] <BBB> ok
[20:16:39] <mru> it will of course "fail" on that run
[20:17:00] <BBB> of course
[20:17:01] <BBB> that's ok
[20:17:11] <BBB> let's try
[21:11:21] <j-b> av500: cool
[21:16:47] <BBB> kshishkov: ping
[21:16:58] <BBB> [vc1 @ 0x102803200] Luma scaling is not supported, expect wrong picture
[21:16:58] <BBB> [vc1 @ 0x102803200] Chroma scaling is not supported, expect wrong picture
[21:17:05] <BBB> kshishkov: what is that? the picture actually looks good
[21:17:37] <mru> good can still be wrong
[21:17:47] <mru> if you order pizza and get icecream, that's wrong, right?
[21:17:53] <Sean_McG> heheheheh
[21:17:57] <Sean_McG> nice analogy there
[21:20:17] <BBB> mru: well, that's what kshishkov is for
[21:20:27] <BBB> this is SA00048.vc1 btw
[21:20:28] <kurosu> delivering pizzas?
[21:20:45] <BBB> [vc1 @ 0x101828c00] Sliced decoding is not implemented (yet)
[21:20:45] <BBB> [vc1 @ 0x101828c00] warning: first frame is no keyframe
[21:20:50] <BBB> SA00049.vc1
[21:20:53] <BBB> kshishkov: ^^
[21:21:30] <BBB> SA00054.vc1 gives a combination of the two above
[21:22:29] <Kovensky> mru: arguably, the result is improved in that case :>
[21:23:05] <mru> no good pizza in brazil?
[21:23:14] <Kovensky> there is
[21:23:20] <Kovensky> but arguably ice cream is better than pizza!
[21:23:28] * Kovensky has a sweet tooth :3
[21:23:42] <mru> that's worse than comparing apples and oranges
[21:23:54] * mru sends Kovensky some garlic icecream
[21:24:04] <Kovensky> nuu D:
[21:24:18] <Sean_McG> actually that would be interesting to try sometime
[21:25:48] <elenril> obviously apples > oranges
[21:30:35] <lu_zero> but Orange > apple
[21:30:45] <lu_zero> (or not?)
[21:30:57] <mru> apples > oranges
[21:31:00] <mru> oranges > pears
[21:31:04] <mru> pears > apples
[21:31:50] <lu_zero> might make sense
[21:39:46] <saintd3v> * mru sends Kovensky some garlic icecream <-- we made guinness ice cream once
[21:40:57] <BBB> I had guiness icecream once
[21:40:59] <BBB> it's delicious
[21:41:14] <BBB> how come vc1 duplicates so many frames even when using -vsync 0?
[21:41:19] <BBB> are there other sync options?
[21:41:27] <Sean_McG> does it get you wasted as much as drinking a Guinness?
[21:41:44] <BBB> Sean_McG: not really
[21:42:40] <saintd3v> BBB: it wasn't as bad as i expected, but i don't drink guinness or beer for that matter.
[21:42:57] <Sean_McG> it's the beer that eats like a meal
[21:43:49] <mru> guinness is good for you
[21:44:02] <thresh> +1
[21:44:12] <thresh> morning
[21:44:23] <Sean_McG> happy $TIMEZONE
[21:44:32] * lu_zero had menabrea ice cream more than once
[21:45:46] <Sean_McG> myself I don't have a sweet tooth really
[22:28:17] <BBB> kshishkov: ping a ping
[22:34:12] <j0sh> merbanan: do you have neon for yuv->rgb conversion?
[22:34:58] <kierank> [22:28] BBB: kshishkov: ping a ping --> ping a ping a roses, a pocket full of posies...
[23:04:29] <saintd3v> mru: good for me, like pre-dinner mayo?
[23:10:31] <MrNaz> google returns quite a few hits for   "libavfilter yadif"  am i to understand that yadif has been implemented in libavfilter? if so, that would totally rock my underpantaloons
[23:12:34] <drv> ffmpeg -filters | grep yadif
[23:12:59] <MrNaz> drv i just saw it
[23:13:05] <drv> :)
[23:13:17] <MrNaz> my underpantaloons are currently rocking
[23:58:56] <merbanan> j0sh: not that I can share :/