[Ffmpeg-devel-irc] ffmpeg-devel.log.20131128

burek burek021 at gmail.com
Fri Nov 29 02:05:02 CET 2013


[00:28] <cone-203> ffmpeg.git 03Michael Niedermayer 07master:0075a22f415f: Revert "configure: add #include "version.h" to config.h"
[01:18] <Timothy_Gu> Anymore ideas on the version.h patch?
[01:19] <Timothy_Gu> Also, Michaelni: you forgot to sign-off on the revert commit
[01:26] Action: Timothy_Gu starts to be inpatient with the IRC channel.
[01:35] <llogan> i don't think we have a "sign-off" policy
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.0:8f7d839e15c9: lavd/lavfi: support unknown channel layouts. (cherry picked from commit 863fb11f63f7f60feec390f3c54dd13606e07d05)
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.0:62baf22ec085: lavfi/avfiltergraph: suggest a solution when format selection fails.
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.0:c2d37e736428: lavfi/avfiltergraph: do not reduce incompatible lists.
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.0:9a4acedf3177: lavfi: parsing helper for unknown channel layouts.
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.0:02c8c064ea28: lswr: fix assert failure on unknown layouts. (cherry picked from commit 4a640a6ac89099bfb02d6d3d3ada04e321a37476)
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.0:cf8462ce00a7: lavfi/af_pan: support unknown layouts on output. (cherry picked from commit 4e9adc9b7363cc336e3d47c98455e1508902fd29)
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.0:2005887707fe: lavfi/af_pan: support unknown layouts on input.
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.1:31647c5a4693: lavd/lavfi: support unknown channel layouts. (cherry picked from commit 863fb11f63f7f60feec390f3c54dd13606e07d05)
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.1:c2ae9f75d731: lavfi/avfiltergraph: suggest a solution when format selection fails.
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.1:1a676881caad: lavfi/avfiltergraph: do not reduce incompatible lists.
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.1:838a453e3962: lavfi: parsing helper for unknown channel layouts.
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.1:cfcb22a77bd4: lswr: fix assert failure on unknown layouts. (cherry picked from commit 4a640a6ac89099bfb02d6d3d3ada04e321a37476)
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.1:bc04a3a48951: lavfi/af_pan: support unknown layouts on output. (cherry picked from commit 4e9adc9b7363cc336e3d47c98455e1508902fd29)
[01:36] <cone-203> ffmpeg.git 03Nicolas George 07release/2.1:ad19cb3ca70c: lavfi/af_pan: support unknown layouts on input.
[01:37] <llogan> Timothy_Gu: patience. and don't forgot the time zone differences (although some are vampires)
[01:39] <llogan> Timothy_Gu: are you going to make any more changes to "web: move RSS generation to a seperate shell script"?
[01:39] <Timothy_Gu> Llogan: I know, but it seems that most people here are vampires 😨
[01:39] <llogan> not me, but my timezone puts me in vampire time
[01:40] <Timothy_Gu> Llogan: no, I'm done with it
[01:40] <Timothy_Gu> Whats your timezone?
[01:40] <llogan> -9
[01:40] <Timothy_Gu> (Weird)
[01:41] <llogan> It's only 3:41 pm now
[01:41] <Timothy_Gu> Do you live in Yukon or something?
[01:41] <Timothy_Gu> Mine is -8
[01:41] <llogan> im in alaska
[01:41] <Timothy_Gu> Ah
[01:42] <Timothy_Gu> Mountain man? ;-) 
[01:42] <llogan> they are very near by. i can walk to a trailhead in 5 mins
[01:43] <Timothy_Gu> Do you like AL?
[01:43] <llogan> Alabama is fine I guess.
[01:43] <Timothy_Gu> I mean AK
[01:43] <llogan> yes, but I try to get out often. this isn't a fun part of the year.
[01:44] <Timothy_Gu> I'm better off in sunny socal
[01:44] <Timothy_Gu> Southern CA
[01:44] <llogan> i was in sequoia park right before they closed it. nice place.
[01:44] <llogan> big trees. small bears.
[01:45] <Timothy_Gu> Ive never been there (ashamed)
[01:45] <Timothy_Gu> My history teacher was at one point a ranger there
[09:52] <wm4> michaelni: AVFrame.pkt_dts is incorrect for avi/ogm/vfw-mkv, it returns the dts of the second packet instead of the first, so e.g. starting to decode a file will never yield an AVFrame with pkt_dts=0
[09:52] <wm4> michaelni: or generally, the codec delay selects which packet's dts to use
[09:52] <wm4> michaelni: is there a way to get the exact timestamp without additional hacks in the user code?
[09:53] <wm4> also, in other words: effectively this means that the best_effort_timestamp will be incorrectly delayed by the codec delay
[10:23] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:c4a03ac3b8d4: doc/ffserver: document configuration syntax
[10:23] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:f7fbb7ac9906: ffserver: clarify feedback in case of wrong AVOptionAudio/Video error
[10:23] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:15f2cac01819: doc/ffserver: remove outdated troubleshoot entry about audio
[10:28] <cone-399> ffmpeg.git 03Timothy Gu 07master:a212d66c2acf: doc/default.css: sync with ffmpeg-web
[10:58] <cone-399> ffmpeg.git 03Anton Khirnov 07master:4d388c0cd05d: h264_refs: make sure not to write over the bounds of the default ref list
[10:58] <cone-399> ffmpeg.git 03Michael Niedermayer 07master:0a5ca63c24b8: Merge commit '4d388c0cd05dd4de545e8ea333ab4de7d67ad12d'
[11:23] <cone-399> ffmpeg.git 03Anton Khirnov 07master:873379939299: h263dec: sanitize a condition.
[11:23] <cone-399> ffmpeg.git 03Michael Niedermayer 07master:03beb5792a3c: Merge remote-tracking branch 'qatar/master'
[11:43] <wm4> Warning: Ticket description is too long (must be less than 262144 characters)
[11:43] <wm4> trac doesn't allow me to post full uncut console output
[11:45] <ubitux> stop playing dumb :p
[11:46] <wm4> michaelni: I reported that issue as bug
[11:52] <michaelni> wm4, you could attach it, i also could increase the limit, if people want that
[11:52] <wm4> well, I don't think the remainder of the output will be interesting anyway
[11:53] <wm4> also, people like cehoyos could just TRY TO REPRODUCE IT
[11:53] <wm4> he has to do this anyway if he wants to handle bug reports properly
[11:54] <michaelni> iam not a big fan of having pages filled with unneeded information then again few users know what is needed, asking fir full output is the easy solution
[12:03] <ubitux> wm4: it's not bullshit, all our fields are randomly shifted after every addition from libav (which cause a minor bump for them but would require a major for us)
[12:03] <wm4> it's a big load of endless bullshit
[12:04] <ubitux> ...
[12:04] <wm4> can trac do fixed formatting?
[12:04] <ubitux> yes
[12:04] <ubitux> {{{ }}}
[12:04] <wm4> oh
[12:04] <ubitux> that's in the default description
[12:05] <wm4> can't seem to edit the ticket anyway
[12:05] <ubitux> updated
[12:06] <ubitux> that's one of the thing i hate the most with trac :(
[12:06] <wm4> thanks
[12:06] <ubitux> the !fixed default formatting
[12:08] <ubitux> wm4: missing command btw
[12:08] <wm4> ./ffprobe /tmp/\[CCCP\]_Mega_Lossless_Audio_Test.mkv  -select_streams 0 -show_frames
[12:57] <saste> we got a 1K$ donation today
[12:57] <ubitux> :)
[13:01] <Compn> if only it was 1000 btc donation
[13:01] <Compn> :P
[13:06] <ubitux> Compn: you didn't set up a btc donation system
[13:06] <saste> Compn, how much 1BTC is?
[13:06] <ubitux> ~1k
[13:06] <saste> BTW now SPI is discussing how to accept paypal donations (for debian)
[13:06] <saste> there is a lenghty discussion on SPI mailing list
[13:07] <saste> i don't know if/when this will be extended to other SPI associate projects
[13:07] <Compn> ubitux : i guess i should do that eh ? :)
[13:07] <Compn> saste : 1btc = $1000
[13:09] <wm4> BTC is changing in value all the time
[13:10] <nevcairiel> yeah last week it was still like $500
[13:10] <ubitux> it's crazy since the last days
[13:10] <ubitux> wm4: yeah but mostly raising, a lot.
[13:11] <wm4> just wait for the big crash
[13:11] <nevcairiel> i should've gotten a few BTCs like 1-2 years ago, i would be rich now
[13:13] <JEEB> I remember playing around them 2009-2010, almost had a single full bitcoin... but then I let my account on a group mining thing around 2011 or so
[13:13] <Compn> wm4 : you dont see $2k btc in 1 year ?
[13:13] <Compn> pfft
[13:13] <nevcairiel> simply buying $100 worth of bitcoins a few month ago would possibly get you 100x results today
[13:13] <wm4> is mining a bitcoin harder now than 2 years ago?
[13:14] <Compn> yes, complexity went skyrocketing
[13:14] <JEEB> yes
[13:14] <Compn> thats why price is jumping...
[13:14] <wm4> weird stuff
[13:14] <JEEB> I think it was 2011 or 2012 when they halved the amount of bitcoins you get per one full "mine"
[13:14] <JEEB> from 50 to 25
[13:20] <cone-399> ffmpeg.git 03Clément BSsch 07master:e5102feb6b5e: doc/examples: add transcode_aac to local Makefile.
[13:20] <cone-399> ffmpeg.git 03Clément BSsch 07master:245931885f26: Add transcode_aac example to .gitignore.
[13:25] <saste> can we extend tags to the wiki as well?
[13:29] <ubitux> i'm going to integrate the other examples into the configure as well
[13:29] <ubitux> any objection?
[13:32] <ubitux> i have no idea why avutil is in the dependencies of the muxing example though
[13:32] <ubitux> it should be a subdep
[13:32] <ubitux> unless i'm missing something
[13:34] <ubitux> we can fate test them now btw, which is a good thing
[13:46] <ubitux> rhaa how much duplication of the list is there.
[13:59] <BBB> so ubitux 
[13:59] <BBB> ffvp9-mt?
[14:00] <ubitux> i didn't have time to test, i've commented what i could in a first pass
[14:00] <BBB> I'll make a parallelmode sample (highres for testing and lowres for fate)
[14:09] <cone-399> ffmpeg.git 03Clément BSsch 07master:a8ac2a153030: doc/examples: make fill_samples static.
[14:09] <cone-399> ffmpeg.git 03Clément BSsch 07master:1a88e84fd54f: build: integrate remaining examples.
[14:09] <cone-399> ffmpeg.git 03Clément BSsch 07master:fe318e143b6f: build: split examples cleaning out of docclean.
[14:09] <cone-399> ffmpeg.git 03Clément BSsch 07master:1d751768f3d4: build: also clean the examples with debug symbols
[14:09] <cone-399> ffmpeg.git 03Clément BSsch 07master:0d2c832445e1: doc/examples: update README.
[14:10] <cone-399> ffmpeg.git 03Clément BSsch 07master:01c9d8094af9: doc/Makefile: cosmetics after previous commits.
[14:10] <ubitux> ok, will add a FATE instance building those examples now
[14:17] <ubitux> x86_64-archlinux-gcc-examples added.
[14:18] <ubitux> back to work :(
[14:19] <plepere> my ASM doesn't seem to work when cross-compiling for windows. is there some obscure thing to do in case of windows compiling ?
[14:20] <plepere> I see in the deblock a if WIN64 adding a shadow space on a register, but I don't get what it's supposed to do
[14:20] <ubitux> redzone, calling convention, ... ?
[14:21] <ubitux> (show your code?)
[14:22] <plepere> https://github.com/OpenHEVC/FFmpeg/blob/hevc/libavcodec/x86/hevc_mc.asm
[14:22] <plepere> but I was wondering in general
[14:23] <plepere> like in x64, we need to do     movsxd   on registers having an int parameter
[14:23] <ubitux> can't you replace the call + RET with a simple jump?
[14:23] <ubitux> you should also use local labels btw for non exported syms
[14:24] <plepere> so you'd have me remove all the functions, and have only labels except for "master" functions ?
[14:24] <ubitux> i'd macrotize all of this personnally
[14:24] <ubitux> to avoid a jump :p
[14:24] <plepere> I've never macrotized. :p
[14:25] <ubitux> %macro NAME numargs ... %endmacro
[14:25] <plepere> yes, I've seen it used, but I don't know when to use them.
[14:26] <ubitux> numargs=N ’ %1 %2 ... %N
[14:26] <ubitux> it's really straightforward
[14:26] <ubitux> at least for the simple cases
[14:27] <BBB> plepere: don't use int for stuff used in pointers
[14:27] <plepere> I believe you, but then, when should I use macros, and when should I stay "classic" ? that's something I have no experience with
[14:27] <BBB> plepere: use intptr_t or ptrdiff_t
[14:28] <plepere> BBB : so all strides to be set to ptrdiff_t ?
[14:29] <BBB> yes
[14:29] <BBB> check vp9, or h264, or any other codec
[14:29] <BBB> hevc is epel right?
[14:29] <plepere> I'm working on epel
[14:29] <BBB> why not share the epel asm with vp9?
[14:29] <BBB> it's all epel and 8tap
[14:30] <plepere> I don't know if it's got the same coefficients and all
[14:30] <BBB> so make split coefficient tables
[14:30] <BBB> the tables are an input argument to vp9's asm anyway
[14:30] <smarter> hevc epel is 4-tap
[14:34] <BBB> o_o
[14:35] <smarter> epel is only used for chroma
[14:36] <BBB> oh right well qpel is still 8tap then
[14:36] <BBB> so share that with vp9
[14:37] <smarter> depends, some of the qpel filters are 8taps and some are 7taps :p
[14:37] <BBB> that's the case for vp9 as well
[14:37] <BBB> just make them all 8tap, easier
[14:37] <BBB> it's caller zero-padding
[14:39] <smarter> that means an additional copy
[14:45] <BBB> smarter: ?
[14:45] <BBB> why would zero-padding need an additional copy?
[14:45] <BBB> the _coefficient table_ is zero-padded
[14:46] <smarter> oh right
[14:46] <BBB> we don't care what's in the source data, as long as buffer is big enough (which other pieces of libavcodec take care of)
[14:46] <BBB> zero times N is zero for any N
[14:46] <BBB> except out of memory Ns
[14:46] <smarter> yes :)
[14:46] <BBB> then it's crash
[14:55] <plepere> I've had crashes like that. :)
[14:55] <plepere> in qpel
[15:05] <BBB> plepere: show me crashes, I can help; sharing code is a good idea, it typically leads to at least one of the two, possibly both, having much faster code
[15:05] <BBB> plus smaller binary size
[15:05] <BBB> there's only advantages
[15:05] <plepere> well all the code is in the openHEVC git
[15:05] <plepere> but the crash was when I was optimising with SSE intrinsics
[15:06] <BBB> don't use intrinsics :)
[15:06] <plepere> I'm putting everything in ASM now. :)
[15:06] <BBB> use asm, point out the instruction where it crashes, and we'll help you fix it
[15:06] <plepere> that's what I'm doing now
[15:06] <BBB> \o/
[15:07] <plepere> I've done epel h, v, and now I'm on hv
[15:07] <BBB> see if you can copy stuff from vp9; again, if we can share anything at all, and 8tap looks ike an obvious candidate, then we should try
[15:07] <BBB> epel won't be sharable unfortunately
[15:07] <BBB> but qpel may be
[15:07] <BBB> vp8 has a 4tap filter ;)
[15:07] <BBB> anyway, just keep going for now
[15:08] <plepere> I'd be glad to be able to do this. it's also the opportunity for me to discover a bit of vp9
[15:08] <plepere> but I think that the intrinsics "step" was good for me. It's noob-friendly. :)
[15:09] <BBB> right
[15:10] <BBB> it's good in some cases, like large transforms etc. where you'll need stack and it does register allocation/management for you
[15:11] <wm4> is this the same reason it's bad in other cases?
[15:12] <plepere> we have an implementation of the idct transforms in intrinsics. up to 32x32 too
[15:13] <smarter> plepere: if you want to learn about VP9, there's a nice overview here: http://forum.doom9.org/showthread.php?t=168947
[15:18] <plepere> added to favorites
[15:19] <plepere> thank you
[15:28] <BBB> wm4: the reason in the docs is kind of old, related to bad compilers and so on
[15:28] <wm4> just old or outdated?
[15:32] Action: {V} guesses 2.95.3
[15:40] <wm4> after all there must be a reason to go through all this yasm and x86inc trouble, instead of just using intrinsics
[15:40] <wm4> I understand why not to use inline asm, though... compiler specific, and literally breaks all the time
[15:41] <JEEB> with asm you generally get better performance with a greater probability
[15:41] <JEEB> you can read up on avery lee's adventures into intrinsicsworld on his blog
[15:43] <JEEB> that said, a lot of his texts are somewhat old
[15:43] <iive> i don't think 2.95.3 even have intrinsics.
[15:43] <JEEB> http://virtualdub.org/blog/pivot/entry.php?id=310 http://www.virtualdub.org/blog/pivot/entry.php?id=354
[15:43] <Daemon404> does anyone except the truly perverted care about 2.95 at all?
[15:44] <JEEB> he did review intrinsics with VS2010 and 2011 tho
[15:44] <wm4> maybe some BSDs</sarcasm>
[15:44] <iive> but last time I've tried them myself (1-2 years ago) they still had tendency of spontaneously creating braind-damaged code.
[15:53] <BBB> wm4: it's like languages, they have different uses
[15:53] <BBB> wm4: nobody would say java is useless, just not quite right for multimedia decoding
[15:53] <BBB> likewise, inline asm and yasm have different uses
[15:53] <BBB> for things where you can use both, we typically use yasm
[15:53] <BBB> but inline asm can do some things yasm cannot, like being inlined, and then it's useful
[15:54] <Daemon404> [14:53] <@BBB> wm4: nobody would say java is useless, just not quite right for multimedia decoding <-- must have missed the 1990s fad of java applet decoders ;)
[15:55] <wm4> also, whatever happened to this fabled js video decoder
[15:57] <BBB> so as for intrinsics vs asm, asm is typically better if you know what you're doing
[15:58] <BBB> but when you get to stuff where you need to move a ton of stuff to the stack and back, it can get hard, esp. from a portability perspective (32bit vs 64bit)
[15:58] <BBB> (or win64 vs others)
[15:58] <BBB> and then intrinsics will be easier
[15:58] <BBB> but asm will still give you better perf, _if you know what you're doing_
[15:58] <BBB> but that can get really hard in some cases
[15:59] <{V}> wm4, which one? orbx.js, broadway.js, jsmpeg, ... ?
[16:00] <wm4> I was thinking of orbx.js
[16:00] <iive> javascript is not java.
[16:01] <wm4> but js is even more dynamic and more sandboxed
[16:04] <clever> BBB: what if you #ifdef 64BIT out the 64bit code, and leave a c code fallback in the #else block
[16:04] <Daemon404> then you get spaghetti.
[16:04] <clever> so it will always work on every platform, but just not perform optimaly on some
[16:05] <Daemon404> unmaintainable crap
[16:05] <BBB> clever: that's typically done through vfunc assignments instead
[16:05] <BBB> better maintainability
[16:05] <BBB> if (64bit) funcptr = function_64bit_sse7;
[16:05] <clever> ah
[16:05] <BBB> or, rather, if (64bit && have_sse7)
[16:05] <BBB> so yes we do that and it works
[16:05] <clever> only bonus to #ifdef would be inlining things
[16:05] <BBB> but it sucks for 32bit people (which I don't care about, but still)
[16:05] <clever> save the call and ret instructions
[16:06] <BBB> they are vfunc ptrs anyway
[16:06] <BBB> so we don't care
[16:06] <BBB> (that's how our dsp function system works)
[16:06] <Daemon404> BBB, i would think you used to be paid to care abotu 32bit
[16:06] <Daemon404> since chrome is 32bit only
[16:06] <BBB> but I don't anymore
[16:06] <Daemon404> ;)
[16:06] <ubitux> :)
[16:06] <clever> BBB: the rpi decoding i was working on is on hold right now
[16:07] <clever> i cant figure out how to read yuv420 out of the decoder
[16:07] <ubitux> ?
[16:07] <ubitux> what do you mean?
[16:09] <Daemon404> perhaps he means the hw decoder
[16:09] <Daemon404> via its api
[16:09] <nevcairiel> usually hw decoders want to throw nv12 at you, though, not yuv420
[16:10] <Daemon404> good thing swscale can handle it now!
[16:11] <wm4> can it?
[16:12] <Daemon404> ef627bf9ecddfa888d4c6e4fd633852798c3b4a9
[16:13] <BBB> oh right I was going to make samples
[16:13] <wm4> I thought it was still slow after fixing it... but maybe that was still before that commit
[16:13] <BBB> ubitux: after that I'll probably refactor the idct function tables to not be retarded
[16:13] <ubitux> ok :)
[16:13] <nevcairiel> when anton claims its slow, he usually refers to libav, which didnt merge it yet at that point =P
[16:13] <nevcairiel> although its probably not as fast as it can be
[16:13] <Daemon404> nothing in swscale is
[16:13] <Daemon404> because swscale
[16:13] <BBB> pfff colorspace conversion
[16:13] <BBB> nobody uses that anymore
[16:14] <kierank> lol
[16:14] <Daemon404> except everybody
[16:14] <Daemon404> ... i fell victim to poe's law
[16:14] Action: Daemon404 deskfaces
[16:14] <BBB> well you're doing conversions, what can I say, stop doing that
[16:14] <BBB> I just play videos
[16:14] <ubitux> BBB: sure, i'm not on idct anyway, but lf
[16:14] <BBB> ok cool
[16:14] <BBB> is lf causing brain damage yet?
[16:15] <ubitux> no progress since last time, will do this week end
[16:15] <BBB> ok
[16:16] <plepere> since I've done the qpel, epel, (un)weighted prediction...etc in intrinsics, it's impossible to validate them while I work to translate them to asm ? It breaks my heart to see reports that our decoder is slow on VLC.
[16:17] <BBB> validate?
[16:17] <Daemon404> perhaps next time you write code for a project you should check its standards first?
[16:17] <Daemon404> >_>
[16:17] <BBB> that too :)
[16:17] <plepere> I'm new to the whole thing. I was told to do intrinsics. :p
[16:17] <BBB> by who
[16:18] <plepere> and we weren't part of ffmpeg yet
[16:18] <plepere> mraulet
[16:18] <BBB> mraulet?
[16:18] Action: BBB gives mraulet a look
[16:18] <Daemon404> ant remember who employed them
[16:18] <Daemon404> maybe theme
[16:18] <Daemon404> er atheme
[16:18] <plepere> anyways, we have great results with SSE
[16:19] <plepere> intrinsics
[16:21] <mraulet> Daemon404: ?
[16:21] <Daemon404> i vaguely remebered openhevc being worked on my some corp
[16:21] <Daemon404> maybe im mistaken
[16:21] <Daemon404> s/my/by/
[16:21] <mraulet> we are part of a french consortium working on UHD
[16:22] <Daemon404> i have no idea why i thought of atheme then
[16:22] <mraulet> where ATEME is one of the member 
[16:22] <Daemon404> oh
[16:22] <Daemon404> one day ill remember how to spell t.
[16:22] <Daemon404> it.
[16:22] <mraulet> yes
[16:22] <mraulet> :p
[16:23] <mraulet> for those interested in
[16:23] <mraulet> http://www.4ever-project.com/index.php?rub=index&srub=&langue=en
[16:24] <ubitux> i can't hold a smile looking at the domain
[16:24] <Daemon404> lawl
[16:24] <Daemon404> i like how technicolor is listed
[16:24] <Daemon404> theyre nothing but a patent troll^Wentity now
[16:28] <mraulet> so basically the decoder and the player is done opensource in this project
[16:29] <mraulet> by uni
[16:39] <mraulet> BBB: any specs for VP9
[16:39] <mraulet> any pointers?
[16:40] <nevcairiel> read the code :)
[16:40] <mraulet> thanks ... that was not my question :p
[16:40] <nevcairiel> but thats the answer to vp9 specs
[16:40] <mraulet> pfiou
[16:53] <smarter> mraulet: there's a nice overview here: http://forum.doom9.org/showthread.php?t=168947 but no real spec
[16:53] <mraulet> I saw it on linkedin
[16:54] <mraulet> same thread thanjs
[16:54] <mraulet> *ks
[17:03] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:39192b76f800: doc/ffserver: remove more pointless/outdated entries
[17:03] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:80193a12e017: doc/ffserver: use @command{} for specifying ffserver
[17:03] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:6dc90334e3ce: doc/ffserver: remove "What can this do?" section
[17:03] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:000337992667: doc/ffserver: improve rendering for -n option text
[17:03] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:2989281d4d97: doc/ffserver: document NoDaemon deprecated option
[17:14] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:398479050e56: doc/ffserver: specify default value for Preroll
[17:14] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:e79f299cafd2: doc/ffserver: remove tip about MaxBandwidth
[17:14] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:bb097d037cd4: doc/ffserver: add notice about case-insensitiveness
[17:19] <ubitux> saste: grep RTPS doc/ffserver.texi
[17:19] <ubitux> i think you mean RTSP
[17:20] <Daemon404> real time protocol streaming!
[17:20] <Daemon404> new protocols every minute
[17:20] <saste> ubitux, feel free to fix with no patch
[17:20] <ubitux> lazy
[17:20] <ubitux> :)
[17:29] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:b89983c4ead6: doc/ffserver: clearly specify whence ffserver reads the configuration file
[17:29] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:2121945770dd: doc/ffserver: fix RTPS -> RTSP ttypo
[17:29] <ubitux> :D
[17:44] <BBB> mraulet: what would you like to know about vp9?
[17:44] <BBB> mraulet: I can answer practical questions about it; as for specs, ask google, not me
[17:56] <mraulet> see MC without having to read VP9 code but it seems to have no alternative solution
[18:15] <BBB> mraulet: MC in vp9 is simply 8tap, epel (luma) or 16thpel, 8tap (chroma), so luma/chroma mc code can be shared with hevc luma mc (since it's both 8tap)
[18:16] <BBB> mraulet: and mx/my are just indices in a table that is provided as an argument into the mc function
[18:16] <BBB> unless rounding/precision is different
[19:00] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:1ee9341a70ad: doc/ffserver: extend documentation for CustomLog option
[19:00] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:07315da1497c: doc/ffserver: fix documentation for -n option
[19:02] <ubitux> saste: so you are our new ffserver maintainer?
[19:02] <ubitux> congratulation
[19:03] <saste> ubitux, no
[19:03] <saste> just fixing documentation and doing some minor fixes
[19:03] <ubitux> oh yes you are.
[19:03] <saste> it is not as bad as I thought
[19:03] <ubitux> :)
[19:03] <ubitux> yes there are a few things to understand
[19:04] <ubitux> my favorite thing is Command
[19:04] <saste> ??
[19:04] <ubitux> or well don't remember the name
[19:04] <ubitux> Launch*
[19:04] <saste> Launch?
[19:04] <ubitux> yes
[19:04] <saste> yes, so review my related patches
[19:04] <ubitux> :(
[19:04] <saste> the logic is quite hackish
[19:07] <saste> all the code does a lot of assumptions, and can be exploited in various funny ways
[19:07] <saste> never let $random_user write your ffserver configuration file :)
[19:08] <ubitux> well, we even had xss injection in ffserver so&
[19:08] <ubitux> :)
[19:08] <ubitux> which was hardly exploitable in a meaningful way though
[19:08] <saste> what was this LoadModule thing?
[19:09] <ubitux> 885739f3b4ca3fb60abf417120845e3fcfb99b53
[19:10] <saste> nice
[19:11] <saste> now I know why I had those weird "?" all around
[19:12] <saste> http_escaping would had been better...
[19:24] <ubitux> saste: hf writing one honoring charset
[19:24] <saste> ubitux, hf??
[19:24] <ubitux> have fun
[19:25] <ubitux> which reminds be a vuln in php (or was it the mysql escape?) with chinese charset
[19:28] <ubitux> http://shiflett.org/blog/2006/jan/addslashes-versus-mysql-real-escape-string
[19:28] <ubitux> saste: check this ^
[19:28] <ubitux> it's a funny vuln in that spirit :)
[19:29] <ubitux> though, it's for sql, not xss
[19:29] <ubitux> http://shiflett.org/blog/2005/dec/google-xss-example
[19:29] <ubitux> here is one with charset :)
[19:30] <ubitux> and xss
[19:30] <ubitux> so anyway, current solution sounds like a safer solution for now
[19:42] <saste> ubitux but we use quotes
[19:43] <saste> we should at least replace quotes, and replace all the rest
[19:43] <saste> *replace with ?
[19:54] <ubitux> mmh?
[20:03] <saste> ubitux, if char == ' => substitute with "e; else accept only ASCII chars
[20:17] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:4e39962f0c39: doc/ffserver: document "Launch" feed option
[20:17] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:312d075e5de9: doc/ffserver: move ACL syntax section before the global option sections
[20:17] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:bcbf8920ede1: doc/ffserver: clarify/extend documentation about status streams
[20:17] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:805b607484e1: doc/ffserver: extend documentation for VideoBitRateRange
[20:17] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:58010e57b2b9: doc/ffserver: fix documentation for CustomLog
[21:24] <ubitux> saste: wait before pushing the error feedback patch please
[21:44] <burek> man, debian is so broken..
[21:44] <burek> i tried to build libquvi for ffmpeg (youtube)
[21:44] <burek> and it requires like 160 dependencies.. :S
[21:44] <burek> f.. it
[21:52] <wm4> I don't get why ffmpeg would need libquvi
[21:52] <wm4> it just causes us problems, like random crashes due to library clashes
[21:55] <ubitux> just don't enable it?
[21:59] <cone-399> ffmpeg.git 03Vittorio Giovara 07master:0673ede985a6: configure: add strip flags checks
[21:59] <cone-399> ffmpeg.git 03Michael Niedermayer 07master:71b95f2ab603: Merge commit '0673ede985a6560e7efb86dab1c58fb7f95ce587'
[22:00] <cone-399> ffmpeg.git 03Michael Niedermayer 07master:e283c26c35c4: build sys: rename STRIPFLAGS to ASMSTRIPFLAGS
[22:00] <cone-399> ffmpeg.git 03Michael Niedermayer 07master:191454f26e5a: configure: remove code that disables striping in the absence of some flags
[22:00] <cone-399> ffmpeg.git 03Michael Niedermayer 07master:e975c147e180: library.mak: only run asm strip if ASMSTRIP flags are set
[22:05] <wm4> ubitux: distros will enable it, and then crashy crashy
[22:05] <ubitux> what crash are you talking about?
[22:06] <ubitux> API misusage?
[22:07] <wm4> distros like to package both lua 5.1 and 5.2 (and there's also luajit), and they all share the same symbol names, so static linking and mixing the libs lead to misery
[22:08] <ubitux> oO
[22:08] <ubitux> sounds like a distro or libquvi issue
[22:09] <ubitux> i find libquvi useful in ffmpeg
[22:09] <ubitux> maybe just for the fancy ffplay 'http://youtube...' :)
[22:23] <BBB> ubitux: can you upload fate samples?
[22:23] <BBB> I have a new one for parallelmode
[22:23] <BBB> it turns out we decode it correctly (\o/)
[22:23] <BBB> so now we can do real speedtests
[22:24] <ubitux> no i can't, i don't have permissions
[22:24] <ubitux> ask michaelni 
[22:25] <ubitux> BBB: we already have 92% of coverage in vp9
[22:25] <ubitux> http://coverage.ffmpeg.org/ffmpeg/libavcodec/vp9.c.gcov.html
[22:26] <ubitux> is this going to test the part that was replaced with a memcpy?
[22:26] <burek> it shouldn't crash just because it's linked in ffmpeg (if its not used) right?
[22:26] <BBB> yes
[22:26] <ubitux> nice :)
[22:26] <BBB> oh cool we have graphs
[22:27] <ubitux> graphs?
[22:27] <burek> quvi :)
[22:27] <BBB> coverage
[22:27] <ubitux> there are graphs? @_@
[22:27] <burek> btw, i've worked with lua sockets and they were so buggy at the time
[22:27] <BBB> direct frame referencing is not exposed in the encoder
[22:27] <BBB> ubitux: no, but this is pretty
[22:27] <BBB> your link
[22:27] <ubitux> BBB: ok :D
[22:27] <burek> and quvi uses them, so i'm pretty much skeptic about it :/
[22:27] <ubitux> burek: see with wm4, i have no idea what he's talking about :)
[22:27] <wm4> burek: AFAIK only quvi 0.9 uses them
[22:28] <wm4> 0.9 is the beta version (with different core and api from 0.4 and all)
[22:28] <BBB> hey
[22:28] <BBB> how come segmentation is not tested?
[22:28] <BBB> I thought we had a segmentation sample
[22:28] <BBB> is it broken?
[22:28] <wm4> burek: it's true that 0.9 is a PITA in many ways
[22:28] <ubitux> BBB: no idea :)
[22:28] <BBB> hm
[22:29] <ubitux> flushing is also not called
[22:29] <burek> sh..t... that makes sense :( now i see why i had to make so many symlinks in order for ffmpeg to detect it :(
[22:29] <burek> back to the lab :s
[22:29] <burek> thanks wm4 :)
[22:30] <wm4> 0.9 is also AGPL3
[22:37] <burek> This series is licensed under the GNU Lesser General Public License (LGPLv2.1+)
[22:37] <burek> 0.4
[22:37] <burek> can this go into static build?
[22:40] <BBB> ubitux: well we need to figure out whether segmentation works or not, I was pretty sure that the segmentation code was called
[22:42] <burek> hm, there might be a bug in the configure script of ffmpeg.. there is no -lcurl when testing for libquvi
[22:42] <burek> and -llua
[22:43] <BBB> hm yeah the segmentation sample does not test segmentation
[22:43] <BBB> that utterly sucks
[22:43] <ubitux> lol
[22:43] <BBB> I'll try to generate a new one
[22:43] <ubitux> BBB: is the code triggered by etv.webm?
[22:43] <BBB> unlikely
[22:43] <BBB> you need to hack the code to enable segmentation
[22:43] <BBB> (libvpx encoder)
[22:43] <ubitux> ah&
[22:43] <BBB> I thought I did that, but apparently I screwed up
[22:43] <BBB> I'll fix it and confirm it triggers that code
[22:44] <BBB> (and gives identical output to libvpx)
[22:46] <cone-399> ffmpeg.git 03Michael Niedermayer 07master:d1916d13e28b: dsputil/pngdsp: fix signed/unsigned type in end comparison
[22:46] <cone-399> ffmpeg.git 03Michael Niedermayer 07master:f21ca39e8156: Merge commit 'd1916d13e28b87f4b1b214231149e12e1d536b4b'
[22:54] <smarter> BBB: --aq-mode 1 or 2 should test segmentation
[22:59] <BBB> \o/
[23:00] <BBB> which one do I use?
[23:00] <BBB> or what's the difference?
[23:00] <smarter> aq-mode 1 is variance-based, 2 is to help hit the bitrate I think, it's a recent addition
[23:01] <smarter> also your mbgraph stuff should still work
[23:01] <smarter> it probably tests more code paths since it enables/disables segmentation
[23:01] <BBB> I'll look at mbgraph
[23:02] <BBB> or well no let's do aq-mode
[23:02] <cone-399> ffmpeg.git 03Diego Biurrun 07master:cf7860db608d: x86: dsputil: Suppress deprecation warnings for XvMC bits
[23:02] <cone-399> ffmpeg.git 03Michael Niedermayer 07master:3adb82565096: Merge commit 'cf7860db608df7c76471d8b61f07abbd5aad8dd5'
[23:04] <BBB> does aq-mode work on keyframes?
[23:05] <BBB> smarter: ^
[23:05] <smarter> aq-mode 1 enables segmentation on all frames
[23:06] <BBB> cool
[23:06] <smarter> not sure what aq-mode 2 does
[23:06] <BBB> that's better then
[23:06] <BBB> I'll test aq-mode 1
[23:08] <BBB> hm libvpx fails to build for me now
[23:09] <BBB> good job!
[23:13] <BBB> yeah it doesn't build at all on a mac
[23:13] <BBB> I don't care enough to fix it ATM
[23:15] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:04702a0d3d73: ffserver: add some error handling to main()
[23:16] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:7cbbc4f7e7ff: ffserver: extend error handling when parsing the configuration file
[23:16] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:4f60d53b1867: ffserver: add comment about how Launch path program is created
[23:16] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:ddab41995fe1: ffserver: drop stream AudioQuality option
[23:16] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:0e1e5d005329: ffserver: improve error feedback in open_input_stream()
[23:16] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:77f5cb92aff4: ffserver: factorize code setting ReadOnlyFile/File options
[23:16] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:2a0349786a58: doc/ffserver: document InputFormat stream option
[23:16] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:6aea3d597e36: doc/ffserver: add notice about re-launch behavior
[23:16] <cone-399> ffmpeg.git 03Stefano Sabatini 07master:ab89d2dd2a6e: ffserver: fix typo in comment: defaut -> default
[00:00] --- Fri Nov 29 2013


More information about the Ffmpeg-devel-irc mailing list