[Ffmpeg-devel-irc] ffmpeg-devel.log.20141123

[burek]

[00:00] <ubitux> yeah, it's useful
[00:00] <ubitux> though i thought we could do it somehow...
[00:00] <saste> I got the feeling it could be generalized though
[00:00] <ubitux> blend filters or something
[00:00] <ubitux> check vf_blend
[00:00] <saste> ubitux, yes, but that would imply some tweaking with timestamps
[00:01] <ubitux> ah yeah indeed, i remember, it takes 2 streams
[00:01] <jamrial> kierank: try adding a vsynth test then
[00:01] <saste> at the moment we probably don't have filters dealing with consecutive frames from the same stream
[00:02] <jamrial> those are used to test the bitexact output of video encoders
[00:02] <saste> we could extend the filter and take a queue of frames, and compute a function of the previous frames
[00:02] <kierank> ok the asm matches the c version
[00:03] <kierank> doesn't match remuxing an existing file for some reason
[00:10] <kierank> on some samples it can be 4x faster
[00:16] <kurosu> kierank, btw, you're doing 6 or 12 pixels at a time, right ?
[00:16] <kierank> 6 pixels at a time
[00:17] <kurosu> does the code properly handle the tail (remaining pixels) ?
[00:17] <kierank> that part is in c
[00:17] <kurosu> ok I haven't noticed that
[00:17] <kierank> i.e the asm does one line (apart from the end bit)
[00:17] <kierank> 1280 isn't a multiple of 6 so there are real world uses for non mod-6
[00:18] <kurosu> sure
[00:18] <kurosu> and now I have seen the remainder of the code
[01:26] <cone-26> ffmpeg.git 03Michael Niedermayer 07master:4b68edd08ac3: flvenc: Remove an unused variable
[01:26] <cone-26> ffmpeg.git 03Martin Storsjö 07master:eec7f032a903: lavf: Remove a redundant include of sys/stat.h
[01:26] <cone-26> ffmpeg.git 03Michael Niedermayer 07master:d899ea2017ef: Merge commit '4b68edd08ac352e314ae3fc701f90b081e549324'
[01:26] <cone-26> ffmpeg.git 03Michael Niedermayer 07master:6d8dda169384: Merge commit 'eec7f032a903e06d249d1e8aa6630b65292bf40f'
[01:40] <cone-26> ffmpeg.git 03Martin Storsjö 07master:f856d9c2f314: dashenc: Don't require the stream bitrate to be known
[01:40] <cone-26> ffmpeg.git 03Michael Niedermayer 07master:9f0fd17c614f: Merge commit 'f856d9c2f314c493c672dfb9c876da182525da3d'
[02:03] <cone-26> ffmpeg.git 03Michael Niedermayer 07master:65ce8f889553: avcodec/x86/Makefile: fix order
[03:35] <cone-26> ffmpeg.git 03Lukasz Marek 07master:691f9be622c7: lavc/anm: fix mem leak in case of init failure
[03:35] <cone-26> ffmpeg.git 03Lukasz Marek 07master:969382162f18: lavc/smacker: fix mem leak in case of init failure
[03:35] <cone-26> ffmpeg.git 03Lukasz Marek 07master:bceabbdabab3: lavc/libvorbisdec: fix mem leak in case of init failure
[03:35] <cone-26> ffmpeg.git 03Lukasz Marek 07master:f87a34486af0: lavc/libvorbisdec: use better error codes
[06:53] <cone-390> ffmpeg.git 03Reimar Döffinger 07master:458aadf8627c: lpc: Reduce stack usage by allocating LLSModel in context.
[06:53] <cone-390> ffmpeg.git 03Reimar Döffinger 07master:d0682b5eb0d9: svq1enc: reduce stack usage of recursively-called function.
[06:53] <cone-390> ffmpeg.git 03Reimar Döffinger 07master:6369a7b742bd: xface: Fix encoder crashes due to too small on-stack array.
[06:53] <cone-390> ffmpeg.git 03Reimar Döffinger 07master:dfc6f56c5a6a: xface: reduce table sizes.
[06:53] <cone-390> ffmpeg.git 03Reimar Döffinger 07master:cad3148ea3da: xface: reduce stack usage by directly storing 2 bytes data instead of pointers.
[11:56] <wm4> ubitux: just noting: https://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/2014-November/072772.html
[12:13] <ubitux> wm4: yeah i just saw that
[12:18] <cone-390> ffmpeg.git 03Clément BSsch 07master:8ef46f4a0cd3: avcodec/microdvddec: support various broken form of color tags
[12:19] <wm4> that was quick
[12:21] <ubitux> about the italic thing with / it's weird..
[12:21] <ubitux> should i support that as well?
[12:21] <ubitux> i mean there is already a system for italic
[12:21] <ubitux> it looks like another random markup convert from another format
[12:21] <ubitux> "convert"
[12:22] <wm4> actually I've seen this often
[12:23] <ubitux> do you have samples?
[12:24] <wm4> {3113}{3186}/W Londynie jest pewien reporter.|/Nazywa si<EA> Simon Ross.
[12:24] <ubitux> thx
[12:24] <ubitux> does the 'W' means anything?
[12:25] <wm4> no idea... maybe it's part of the text
[12:25] <ubitux> is "Nazywa" a word or it's /N + "azywa"?
[12:25] <wm4> another one: {3618}{3669}/Facet pisze do Guardiana.
[12:25] <wm4> {4549}{4619}Je<9C>li za to bekn<EA>, Mark,|je<9C>li sprawa si<EA> pogorszy...
[12:25] <wm4> oops the <9C> etc. are from less
[12:25] <ubitux> yeah i guessed so
[12:26] <wm4> so it seems / is strictly standalone
[12:26] <ubitux> k
[12:52] <ubitux> wm4: i support it only add the beginning of a line, right?
[12:54] <wm4> it can happen after | too (apparently)
[12:56] <ubitux> yeah it's the beginning of a line
[12:56] <wm4> yes then it should be fine
[12:58] <ubitux> mmh i need to support if it's written after the real tags too i suppose
[12:58] <ubitux> like, {3618}{3669}{y:b}/fuck you sanity
[12:58] <ubitux> and {3618}{3669}/{y:b}fuck you sanity
[12:58] <wm4> I have no example of such a thing
[13:14] <cone-390> ffmpeg.git 03Clément BSsch 07master:6e411d9cc7ec: avcodec/microdvddec: support non persistent italic marker '/'
[13:14] <ubitux> wm4: i supported it anyway
[13:15] <wm4> nice
[14:31] <ubitux> kierank: random example: http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavfilter/vf_lut3d.c;h=862dddeacf55bb0f934692eac52fde8e1361cf0c;hb=HEAD#l201
[14:31] <kierank> was hoping it wouldn't look like that but ok
[14:31] <ubitux> another solution is to make a function with if (bitdepth == 8) { ... } else { ... }
[14:32] <ubitux> with av_always_inline
[14:32] <ubitux> and then create wrapper function above
[14:32] <ubitux> where did i do that recently...
[14:32] <ubitux> ah, xbr
[14:32] <ubitux> kierank: http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavfilter/vf_xbr.c;h=47e4b769ca233a0489967d01428a8f8ba07730a5;hb=HEAD#l325
[14:32] <ubitux> it's not bitdepth here
[14:32] <ubitux> but you get the idea
[14:33] <ubitux> the if (n == 2) { ... } else if (n == 3) { ... } ... aren't evaluated in the inner loop that way
[14:42] <ubitux> kierank: and you could mix both too
[14:43] <ubitux> last solution is #define BIT_DEPTH 8 and #include "v210enc_template.c"
[14:43] <ubitux> but overkill for your case
[15:19] <cone-390> ffmpeg.git 03Carl Eugen Hoyos 07master:600e38f5632e: Fix standalone compilation of the apng decoder on x86.
[15:19] <cone-390> ffmpeg.git 03Carl Eugen Hoyos 07master:78093cf849b6: Print a warning if vp6 is muxed into flv: The output is flipped.
[17:03] <cone-390> ffmpeg.git 03James Almer 07master:305b03097db3: lavf/apngdec: print currently unsupported in-stream tags in a more readable form
[17:15] <mrec> does encoding mpeg2audio/video work with ffmpeg at the moment?
[17:15] <mrec> seems like our application fails with the latest ffmpeg library also some green frames come up now randomly
[17:26] <mrec> just terrible this unstable API
[17:31] <jamrial> afaik it's working, but if you think there's a bug open a ticket in trac.ffmpeg.org and provide a way to reproduce it (preferably with the ffmpeg command line tool)
[18:18] <kierank> is it possible to do a byte equivalent of CLIPW?
[18:53] <jamrial> pmaxsb pminsb?
[18:54] <jamrial> they are sse4.1 instructions, though
[18:56] <jamrial> you may be able to emulate them using sse2 instructions
[19:09] <kierank> jamrial: is it available unsigned?
[19:09] <kierank> oh it is
[20:54] <cone-348> ffmpeg.git 03Lukasz Marek 07master:331fae80a1fb: lavc/mss1: fix mem leak in case of init failure
[20:54] <cone-348> ffmpeg.git 03Lukasz Marek 07master:4e9745fbff51: lavc/rv30: fix mem leak in case of init failure
[20:59] <kierank> if anyone is interested in figuring out why my 8-bit asm is broken please look: http://pastebin.com/sK5VDd6U
[20:59] <kierank> most of the luma is fine, just not the last 4 bytes out of 12 each time
[21:07] <jamrial> you're using a shuf10 constant in the 8bit version. is that intended?
[21:08] <kierank> yes because the final shuffle should be the same
[21:08] <kierank> in both versions
[21:25] <BoRiS> Hi guys
[21:27] <BoRiS> I have a patch for the latest ffmpeg via git has a problem with libxcb and shape causing ffmpeg not to compile properly. (xcbgrab.c:(.text.unlikely+0x52c): undefined reference to `xcb_shape_rectangles')
[21:31] <J_Darnley> typically we ask people to use gt to send a patch to the mailing list.
[21:31] <J_Darnley> uh
[21:31] <J_Darnley> that should be git
[21:31] <BoRiS> http://pastebin.com/egGkbj09
[21:31] <JEEBcz> BoRiS, commit and then use git format-patch to create a patch that can be applied as-is including the commit message etc
[21:32] <JEEBcz> that can then be posted on the ffmpeg-devel mailing list for review
[21:32] <JEEBcz> you can either use git send-email or otherwise make the content of the e-mail be the format-patch
[21:33] <JEEBcz> git send-email is generally recommended since it doesn't do weird stuff to the content when sending
[21:33] <JEEBcz> format-patch and copy-pasting into a mail client can lead to lolfails
[21:43] <kierank> jamrial: turns out it was the clipping mask
[21:50] <BoRiS> Submitted my patch to the ffmpeg-devel mailing list.
[21:50] <JEEBcz> coal
[21:51] <JEEBcz> ok, it's still not format-patch, but I guess someone might fix that for you when applying it :P
[22:06] <Daemon404> "fix:
[22:15] <kierank> Daemon404: 24-bit was claimed to work
[22:16] <Daemon404> it used to print a message saying it was unimpletented
[22:17] <kierank> http://git.videolan.org/?p=ffmpeg.git;a=commitdiff;h=ce6141259bb910aa7f580f55cf71e3c503f9a4fb
[22:17] <kierank> lol
[22:18] <J_Darnley> Well, at least that isn't WMA lossless _encoding_ support!
[22:18] <Daemon404> lolwut
[22:25] <jamrial> not even the 16bit version is truly lossless last i checked
[22:41] <rcombs> huzzah, I got `git send-email` working!
[00:00] --- Mon Nov 24 2014