[Ffmpeg-devel-irc] ffmpeg-devel.log.20140617

[burek]

[00:02] <cone-497> ffmpeg.git 03Michael Niedermayer 07master:2fbdfba0f2d1: avformat/flvdec: Support live flv / NGINX RTMP streams
[02:37] <cone-497> ffmpeg.git 03Michael Niedermayer 07master:61df0819d49b: avcodec/cavsdec: fix low delay decoding
[03:27] <jamrial> mmh, plepere is not here
[03:27] <jamrial> just read the logs and found the avx2 discussion
[03:27] <jamrial> he wants vpbroadcastw to do a SPLATW with ymm registeres
[03:27] <jamrial> i guess we should expand the macro to use that
[09:14] <plepere> got SDE working ! I can check my AVX2 code. :D
[09:17] <jamrial> plepere: you need to use vpbroadcastw if you want to do the equivalent of SPLATW for ymm regs with avx2
[09:17] <plepere> jamarial : ok thanks
[09:18] <jamrial> no prob
[09:19] <jamrial> we probably should expand SPLATW to use vpbroadcastw when cpuflags == avx2 (it works for both xmm and ymm regs)
[09:20] <jamrial> that way you can avoid some ifdeffery
[09:20] <plepere> jamrial, it would be better if someone could patch SPLATW to avoid self modification.
[09:46] <durandal_1707> i can't get expansion like metadata in drawtext filter to work
[09:46] <durandal_1707> this worked when i added that feature
[09:57] <ubitux> durandal_1707: we have no test for drawtext :(
[09:58] <ubitux> we need a "bitexact" mode with freetype
[09:58] <ubitux> it might even exists
[09:59] <durandal_1707> how you otherwise could display values from signalstats filter...
[09:59] <ubitux> ffprobe
[10:01] <durandal_1707> how?
[10:04] <ubitux> with movie source
[10:05] <ubitux> ffprobe -f lavfi movie=bla,signalstats -of flat -show_entries frame=tags
[10:05] <ubitux> or something like that
[10:05] <ubitux> don't remember exactly
[10:07] <ubitux> -show_entries frame_tags
[10:09] <plepere> ok, that's a new one. Ubitux can you take a look at my code quickly ?
[10:09] <ubitux> maybe :p
[10:09] <plepere> please !!!
[10:10] <ubitux> ;)
[10:10] <plepere> http://pastebin.com/p382ppvA 
[10:10] <plepere> there's the SSE2 idct16 that(s working, but the AVX2 isn't
[10:11] <plepere> I get a very strange error
[10:11] <plepere> if I pass the %rep to 5, it's just an error.
[10:12] <ubitux> what error?
[10:12] <plepere> but if I go beyond, I get munmap_chunk(): invalid pointer or double free or corruption (out)
[10:12] <plepere> md5 errors
[10:13] <ubitux> did you look at the generated load/store instructions?
[10:13] <plepere> (which is logical, since by doing rep 5, I'm doing a 24x32 block
[10:14] <plepere> no. :/
[10:14] <plepere> but even by forcing vmovdqu I have these errors
[10:16] <ubitux> prototype is weird
[10:16] <ubitux> isn't it missing a digit?
[10:17] <plepere> owait
[10:17] <plepere> the 8 got me confused, indeed.
[10:18] <plepere> but it doesn't change the bug.
[10:18] <plepere> do you think it's code size that is bugging things ?
[10:19] <plepere> since just changing the %rep makes or breaks the build
[10:23] <ubitux> plepere: i odn't think so
[10:23] <ubitux> mmh "breaks the build"? how?
[10:25] <plepere> when I have %rep 5, there is no real error. it's just the decoding that is incorrect. When %rep is 6 or 7, I get the errors.
[10:27] <plepere> munmap_chunk(): invalid pointer or double free or corruption (out)
[11:13] <plepere> ubitux, I've found the problem. Next time I'm coming for help, remind me that I'm an idiot and that I missed something really obvious
[11:17] <plepere> in this case, I was calling the idct32 instead of the idct16. so it's logical that increasing the %rep was the problem since it was bringing me out of bounds
[11:32] <plepere> michaelni, is my patch OK ? I've added idct_dc_add avx2 functions (conformance takes a long time since I'm on an emularor) do you thinkI should do a 2nd patch with only them or re-submit the whole thing ?
[13:02] <michaelni> plepere, 2nd patch with only them is probably better as it keeps the patch smaller 
[13:04] <michaelni> plepere, is the latest patch on the ML or there are some changes that wherent submited (except the avx2 idct stuff)
[13:05] <plepere> michaelni, the last fix was sent on monday
[13:05] <michaelni> ok
[13:07] <cone-605> ffmpeg.git 03Anton Khirnov 07master:71d6551e674c: avconv: use the correct variable in comparison
[13:07] <cone-605> ffmpeg.git 03Michael Niedermayer 07master:fa0d0fb42ecd: Merge commit '71d6551e674c86d07c165439c2bf6613b3892307'
[13:27] <cone-605> ffmpeg.git 03plepere 07master:92cccb7bcd79: avcodec/hevc: new idct + asm
[15:55] <ubitux> > Support Ogg/Vorbis in avi
[15:55] <ubitux> ._.
[15:57] <wm4> ubitux: mpv is not using the srt converter, because AFAIK there are fixes to the mpv one that are not in the ff srt one
[15:57] <wm4> these were added back in mplayer2 times
[15:57] <wm4> I'm not sure if ff has them
[15:57] <wm4> also, the ff converter doesn't set the subtitle resolution, which is annoying
[16:00] <ubitux> i didn't write the srt decoder in ffmpeg, Aurelien did
[16:00] <ubitux> and he did that after i wrote it in mplayer iirc
[16:00] <wm4> oh, annoying
[16:00] <ubitux> but afaik it's pretty complete
[16:01] <ubitux> what's this subtitle resolution thing?
[16:01] <ubitux> ah, yeah
[16:01] <ubitux> mmh that's an issue with all the decoders right?
[16:01] <ubitux> should we add an option for that?
[16:02] <wm4> why an option?
[16:03] <ubitux> how else would you control it?
[16:05] <wm4> why control it?
[16:07] <wm4> the resolution controls font size and how \pos tags are used
[16:08] <wm4> if you don't specify a resolution, libass will pick a static fallback value that is consistent with scripts from 10 years ago that lacked the resolution
[16:09] <ubitux> yes but what resolution should be set?
[16:10] <wm4> the fallback resolution, since apparently the code was designed for it
[16:10] <wm4> 384x288 AFAIK
[16:12] <ubitux> what would be the difference with the static fallback?
[16:13] <ubitux> http://pastie.org/9299198 ?
[16:13] <ubitux> sth like that would be fine for you?
[16:14] <ubitux> i'm not sure how that would help
[16:15] <ubitux> avoid a warning maybe? :P
[16:16] <iive> in theory, the pos should have been in the [0;1) range, or % for convenience
[16:19] <wm4> yes avoid a warning
[16:20] <wm4> I'm not sure if headers are order dependent (maybe not), but normally the playres comes after the script type
[16:20] <wm4> iive: that wouldn't change much
[16:28] <wm4> is there a function that can split avoptions out of a flat string, set them on some context, _and_ put missing options into a AVDictionary?
[16:29] <wm4> it's what you typically need for opening some things (like avio contexts or demuxers)
[17:09] <Mista-D> anyway to boost volume with a limiter?
[17:09] <Mista-D> sorry wrong window.
[17:38] <cone-460> ffmpeg.git 03Michael Niedermayer 07master:a50c6f99b44e: ffplay: remove -ec handling code
[19:35] <llogan> Timothy_Gu: you forgot to send your patch for #3616
[19:36] <llogan> *3613
[21:36] <cone-460> ffmpeg.git 03Diego Biurrun 07master:fe4d5fe93611: jpeg2000: Mark static data init functions as av_cold
[21:36] <cone-460> ffmpeg.git 03Michael Niedermayer 07master:f9cd317223cb: Merge commit 'fe4d5fe9361162f9033ff1bd84bfc1b2091ba785'
[23:07] <ubitux> oh, git didn't like the unicode 7 character :(
[23:11] <ubitux> i guess i'll go for 
[23:19] <wm4> what does git reject and why?
[23:20] <ubitux> no idea, didn't look much further
[23:20] <ubitux> https://ffmpeg.org/pipermail/ffmpeg-devel/2014-June/158969.html
[23:20] <ubitux> 🖔 was the character
[23:21] <wm4> also git normally doesn't reject anything, so this is some ffmpeg specific hook?
[23:21] <ubitux> dunno where the failure(s) occur, maybe that's not even in git
[23:22] <wm4> well, what part blows up?
[23:22] <nevcairiel> my font doesnt have your character anyway
[23:23] <ubitux> wm4: well at least on the mailing list it's broken
[23:23] <nevcairiel> but it does the star
[23:23] <jamrial> why did you delete all the entries from older releases?
[23:23] <ubitux> because it doesn't matter in the release itself
[23:33] <Youka> ah, cmon: http://pastebin.com/Rs4mpD7r
[23:34] <Youka> Building ffmpeg 2.2.3 on windows with mingw (tdm-gcc 32bit) results in error above.
[23:34] <Youka> btw.: with msvc it worked.
[23:48] <Youka> Could it be that it expects gcc only features which mingw doesn't have? Even TDM-GCC with GCC v4.8.1 hasn't nanosleep?
[23:50] <wm4> that depends on the headers, not gcc
[23:50] <wm4> I suspect there's something wrong with the configure script though
[23:50] <Youka> Ok, i found out that MinGW has no timespec or nanosleep, it remains unique for unix gcc.
[23:51] <wm4> again this has nothing to do with gcc
[23:51] <Youka> so ffmpeg configure should switch to another HAVE_*** for the an available sleep function with mingw
[23:51] <wm4> hm the timespec use is guarded with #if HAVE_CLOCK_GETTIME && defined(CLOCK_MONOTONIC)
[23:52] <wm4> and <time.h> is included
[23:53] <wm4> so, the mingw headers must somehow be inconsistent
[23:53] <wm4> though there are more correct ways to check for the presence of clock_gettime
[23:54] <jamrial> Youka: what version of mingw are you using. mingw32? mingw-w64 v2, v3, trunk?
[23:54] <wm4> in my own code I'm using: #if defined(_POSIX_TIMERS) && _POSIX_TIMERS > 0
[23:54] <Youka> TDM-GCC (MinGW) 32bit with GCC 4.8.1
[23:55] <jamrial> that doesn't say much. it could be using either mingw32 or mingw-w64 compiled for x86_32
[23:56] <Youka> mingw32
[23:57] <wm4> and the nanosleep thing shows that the configure script did something wrong
[23:57] <wm4> or the one running the configure script
[23:57] <Youka> I don't think, i reloaded ffmpeg 2.2.3 to be sure all is clean, then run configure and make like shown in pastebin
[23:58] <Daemon404> ... you dont happen to have messed with include / lib dirs or the env have you
[23:58] <Youka> no, i haven't
[23:58] <Youka> all like on install
[00:00] --- Wed Jun 18 2014