[Ffmpeg-devel-irc] ffmpeg-devel.log.20171201

Sat Dec 2 03:05:03 EET 2017

[00:07:45 CET] <durandal_1707> jamrial: that is something else,  for alternating stuff
[00:08:42 CET] <JEEB> yhttp://up-cat.net/p/d1a1da67
[00:08:46 CET] <JEEB> http://up-cat.net/p/d1a1da67
[00:08:51 CET] <JEEB> anyone else seeing this stuff?
[00:09:39 CET] <cone-225> ffmpeg 03John Stebbins 07master:2d015d3bf9fe: lavf/mov: fix huge alloc in mov_read_ctts
[00:09:40 CET] <jamrial> durandal_1707: can't you use it to choose between ALTERNATING_LR and ALTERNATING_RL when the side data is AV_STEREO3D_FRAMESEQUENCE?
[00:10:11 CET] <JEEB> http://up-cat.net/p/7b1c1582
[00:10:12 CET] <JEEB> oh fun
[00:10:22 CET] <jamrial> maybe also others
[00:10:26 CET] <JEEB> now failure with gcc
[00:11:15 CET] <durandal_1707> jamrial: yes, but there is yet another field iirc
[00:11:59 CET] <jamrial> JEEB: the android fate client in fate.ffmpeg.org is broken for a different reason, so guess that's why nobody noticed it
[00:13:25 CET] <JEEB> yea, the mediacodec one is definitely breakage in general, the NEON asm is clang-specific (with gas-preprocessor), as GOOG has recommended clang for quite a while now
[00:14:55 CET] <jamrial> JEEB: try including hwaccel.h in that mediacodec file
[00:16:27 CET] <JEEB> seems to have fixed it
[00:16:56 CET] <JEEB> I guess the moving of the hwaccel stuff derp'd it
[00:18:57 CET] <JEEB> jamrial: yes, that links
[00:20:04 CET] <jcdutton> tmm1, just submitted it. How do I find out if it will be accepted or not?
[00:23:47 CET] <tmm1> jcdutton: you should get some replies to the patch with any review and then someone (the maintainer of that codec probably) will apply and reply
[00:25:09 CET] <jcdutton> Ok. It is only something small. It adds to the segment muxer so that files are put in subdirectories by date.
[00:25:48 CET] <cone-225> ffmpeg 03Jacob Trimble 07master:83ecdc9a920d: configure: Fix dependencies of aac_at decoder.
[00:34:51 CET] <jamrial> fuck that december email messing my inbox
[00:36:34 CET] <wm4> heh
[00:38:51 CET] <StevenLi_> jamrial: why the libavutil/reverse.c don't move to the libavutil/reverse.h? 
[00:40:03 CET] <StevenLi_> i see there is an array. and just extern it in the libavutil/reverse.h, but the commit http://git.videolan.org/?p=ffmpeg.git;a=commitdiff;h=3cd616a0c95853411f6ab23ce797f696dc38abe7 there have a c source include the other c source.
[00:40:53 CET] <jamrial> StevenLi_: shared builds needed that
[00:41:15 CET] <jamrial> mmh, i think we could replace the uglyness with av_export_avutil
[00:42:53 CET] <StevenLi_> ok, Thanks :D
[00:58:50 CET] <SortaCore> tmm1: do you use the same hwa context for both decoder + encoder when transcoding?
[00:58:55 CET] <SortaCore> for hw_device_ctx
[01:37:06 CET] <atomnuker> daddesio: huh, send a patch and I'll review it
[01:57:09 CET] <daddesio> atomnuker: Will do. I'll try to get my patches posted on the ML by tonight.
[03:08:47 CET] <daddesio> https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/opus_silk.c#L65
[03:11:08 CET] <daddesio>  ^ That loop simply determines if there are any NLSFs with a "negative min_diff" (i.e. is too close to the previous NLSF) and if so, sets k to the one that has the worst min_diff--otherwise, the function returns.
[03:11:38 CET] <daddesio> However, the "if (pass == 20) break;" is never true and can be removed. (I double-checked against the original libopus code.)
[03:13:08 CET] <atomnuker> go ahead and remove it then
[03:13:28 CET] <daddesio> set k to 0 as well to silence the warning?
[03:13:39 CET] <daddesio> I should add a comment too about what the loop is doing.
[03:15:32 CET] <Compn> wm4 : question, why not fork chromiums' fork of ffmpeg ? since its just decoders and just for the top 10 formats or whatever...
[03:15:45 CET] <wm4> ?
[03:15:55 CET] <Compn> instead of the full ffmpeg with encoders
[03:16:55 CET] <Compn> seems easier to maintain a smaller codebase
[03:16:58 CET] <Compn> than 1 million LOC
[03:17:22 CET] <Compn> especially for mpv, which is player only?
[03:17:46 CET] <wm4> I have no idea what you're trying to do
[03:17:55 CET] <Compn> im just asking a qquestion 
[03:18:11 CET] <Compn> maybe i misunderstand your situation
[03:29:26 CET] <atomnuker> daddesio: if you want to, I don't mind
[03:39:42 CET] <SortaCore> do you use the same hw_device_ctx in both decoder/encoder when transcoding?
[05:16:49 CET] <atomnuker> michaelni: gcc 4.9 compiles the first patch just fine
[05:22:16 CET] <jamrial> atomnuker: that's because 4.9 is the first version with c11 atomics support
[05:22:38 CET] <jamrial> 4.8 and older use the compat code
[05:30:50 CET] <wm4> amazingly late
[05:30:57 CET] <wm4> how's clang support?
[05:31:02 CET] <wm4> and msvc? (ok, lol)
[05:31:34 CET] <cone-508> ffmpeg 03Dale Curtis 07release/3.4:ceed79323cb0: Don't manipulate duration when it's AV_NOPTS_VALUE.
[05:31:34 CET] <cone-508> ffmpeg 03Jacob Trimble 07release/3.4:5ab992cd3875: configure: Fix dependencies of aac_at decoder.
[05:31:34 CET] <cone-508> ffmpeg 03James Almer 07release/3.4:8d51090dcb8b: avformat/matroskaenc: actually enforce the stream limit
[05:31:34 CET] <cone-508> ffmpeg 03James Almer 07release/3.4:2940b3e17cd3: avformat/matroskaenc: add missing allocation failure checks for stream durations
[05:31:34 CET] <cone-508> ffmpeg 03James Almer 07release/3.4:bcfbcbec481c: avcodec/proresdec: align dequantization matrix buffers
[05:31:47 CET] <atomnuker> clang took longer, wasn't until 4.0 or 5.0 that it got stdatomic.h
[05:37:49 CET] <cone-508> ffmpeg 03James Almer 07master:5e9b39b37321: avcodec/h264_sei: remove redundant prefix to H264SEIFramePacking fields
[05:37:50 CET] <cone-508> ffmpeg 03James Almer 07master:e01d2c00ae07: avcodec/h264_slice: use H264_SEI_FpaType enum constants
[10:43:12 CET] <nevcairiel> wm4: msvc has the c++ stdatomic, which basically follow the same syntax just with c++isms (and afaik was the template for the C implementation in the first place), i always wondered why they dont just provide it in C as well, but its not a key focus for them
[10:57:56 CET] <cone-502> ffmpeg 03Paul B Mahol 07master:2cfc8b172ce7: avfilter/vf_tile: add init_padding option
[12:40:45 CET] <durandal_1707> kierank: what do you think about filter that detects telecine patterns in video?
[12:42:14 CET] <nevcairiel> dont we have that
[12:42:21 CET] <JEEB> I would think that generally is something the fieldmatch filter should be doing
[12:46:18 CET] <nevcairiel> yeah thats the one
[12:46:28 CET] <nevcairiel> paired with decimate to actually get rid of the repeated fields if wanted
[12:50:52 CET] <durandal_1707> fieldmatch changes video
[12:55:48 CET] <JEEB> durandal_1707: well you in the end want to do the IVTC
[12:56:02 CET] <JEEB> otherwise why would you be doing the telecine pattern searches?
[12:56:10 CET] <JEEB> what is the use case for *only* doing the pattern search
[12:57:51 CET] <nevcairiel> you would basically re-write fieldmatch anyway, with all the detection just without the field reconstruction
[12:58:44 CET] <durandal_1707> you guys dont get it
[13:00:52 CET] <nevcairiel> thats because you dont say more then two words =p
[15:17:02 CET] <ubitux> is the matroska muxer braindead wrt time base?
[15:17:53 CET] <ubitux> ffmpeg -f lavfi -i testsrc2=r=60:d=1 -y /tmp/media.mp4
[15:17:56 CET] <ubitux> vs
[15:17:58 CET] <ubitux> ffmpeg -f lavfi -i testsrc2=r=60:d=1 -y /tmp/media.mkv
[15:18:44 CET] <iive> afair mkv has timebase in ms
[15:18:56 CET] <ubitux> pts have a waaay too low resolution in mkv
[15:20:20 CET] <ubitux> ffprobe -of flat -show_entries packet=pts /tmp/media.mp4
[15:20:32 CET] <ubitux> ffprobe -of flat -show_entries packet=pts /tmp/media.mkv
[15:20:55 CET] <ubitux> that's a really low resolution
[15:21:09 CET] <ubitux> iive: ms as in millisecond, not microsecond, right?
[15:21:21 CET] <iive> 1/1000 s
[15:21:27 CET] <ubitux> yeah, milli
[15:21:37 CET] <ubitux> dammit, i shouldn't use mkv
[15:21:47 CET] <iive> :) use nut :D
[15:22:49 CET] <ubitux> i guess it's fine for animes with their 15 fps
[15:59:33 CET] <StevenLi_> Hello ?
[15:59:37 CET] <StevenLi_> How should i find the git-2014-11-19-6a054a4 source commit?
[16:00:16 CET] <StevenLi_> i cannot find the 6a054a4 commit id at the newset git clone
[16:02:13 CET] <StevenLi_> or that comiit have been removed?
[16:02:30 CET] <StevenLi_> s/comiit/commit/g
[16:02:32 CET] <jdarnley> I don't think we have ever altered history in the official repository
[16:04:25 CET] <StevenLi_> Yes, i don't think so, but there really have no this commit
[16:05:19 CET] <StevenLi_> https://www.pconlife.com/viewfileinfo/swresample-1-dll/
[16:05:29 CET] <jdarnley> I assume that hash is part of someone's private tree.
[16:06:08 CET] <StevenLi_> Oh, i see, thanks, let me find it on other private tree :D
[16:06:36 CET] <jdarnley> Why that has and not any of the others that look more like how FFmpeg formats its version numbers?
[16:06:45 CET] <jdarnley> git log --before=2014-11-20
[16:06:58 CET] <jdarnley> (related but I meant these...)
[16:07:00 CET] <jdarnley> N-71262-g511585c
[16:07:05 CET] <jdarnley> N-65738-g3c197442
[16:07:14 CET] <jdarnley> N-67999-gd0879a9
[16:07:20 CET] <jdarnley> N-65810-g1b5ec6a
[16:07:27 CET] <jdarnley> N-70239-g111d79a
[16:07:45 CET] <StevenLi_> Thanks
[16:08:05 CET] <jdarnley> commit 511585ce7f
[16:08:05 CET] <jdarnley> Author: wm4 <nfxjfg at googlemail.com>
[16:08:05 CET] <jdarnley> Date:   Fri Apr 3 16:11:53 2015 +0200
[16:08:52 CET] <StevenLi_> yes, i know this, on ffbuild/version.sh
[16:10:42 CET] <iive> jdarnley: we've done once or twice. the first time was shortly after the project moved to git..
[16:11:13 CET] <jdarnley> Okay.  I stand corrected.
[16:11:38 CET] <jdarnley> I guess I also had VideoLAN delete some branch I accidentally pushed to the repo.
[16:12:01 CET] <jdarnley> So I guess I was Fake News
[16:12:18 CET] <iive> but since the git is hosted on videlan, ffmpeg admins should ask them for any changes... so it is huge pain to mess with the repo
[16:12:59 CET] <iive> there is differene between been wrong and been fake news.
[16:13:16 CET] <StevenLi_> iive: do you mean , that maybe on old version control repo?
[16:15:48 CET] <iive> i don't remember the details. I do remember that the commit made a huge mess, something like duplicating the history of the svn/cvs repo. 
[16:16:26 CET] <nevcairiel> that was long before 2014 tho
[16:17:19 CET] <StevenLi_> haha, what a pitty
[16:17:37 CET] <StevenLi_> haha, what a pity 
[16:18:34 CET] <iive> more like 2011
[17:39:07 CET] <BtbN> huh, the coverity build fails because zimg is too old
[17:39:11 CET] <BtbN> it's litterally using git master
[17:46:32 CET] <cone-963> ffmpeg 03Dale Curtis 07master:95bacb521af8: avcodec/vorbis: Fix another 1 << 31 > int32_t::max() with 1u.
[17:51:57 CET] <jamrial> BtbN: 002db7d49a maybe
[17:52:26 CET] <BtbN> yeah, but I don't understand why the build image is too old
[17:52:31 CET] <BtbN> it's master from a few days ago
[17:53:33 CET] <wm4> jdarnley: something wrong with that commit?
[17:53:45 CET] <jdarnley> No
[17:53:49 CET] <jdarnley> wghat?
[17:54:00 CET] <jdarnley> Oh, the copy-pasted log from earlier?
[17:54:33 CET] <jdarnley> That was just to show that one hash listed was in the git history.
[17:55:05 CET] <jdarnley> Sorry about the highlight.
[18:00:00 CET] <wm4> ok
[18:00:43 CET] <wm4> jkqxz: what's with the intel driver and easily triggered assertions
[18:03:17 CET] <omerjerk> Hi, I was looking at this bug - http://trac.ffmpeg.org/ticket/6216
[18:03:44 CET] <omerjerk> I think I'll need some spec file to tackle it. 
[18:03:58 CET] <omerjerk> Anyone has truehd spec?
[18:06:25 CET] <durandal_1707> not here
[18:07:21 CET] <omerjerk> How can I fix the bug without understanding the format? :/
[18:07:37 CET] <omerjerk> any tips? 
[18:07:46 CET] <jdarnley> Extreme Reverse Engineering
[18:07:50 CET] <durandal_1707> omerjerk: your iq must be above 200
[18:09:30 CET] <omerjerk> didn't really understand the reverse engineering part. 
[18:09:32 CET] <nevcairiel> the same way the rest of the encoder was created
[18:09:44 CET] <omerjerk> Neither got Paul
[18:09:51 CET] <omerjerk> *Paul's sarcasm 
[18:10:19 CET] <nevcairiel> various methods, find a binary encoder and de-compile it, or look at what the decoder expects and try to reverse from there whats wrong
[18:11:39 CET] <durandal_1707> our mlp enoder implements subset of possibilties that are allowed by decoder
[18:14:52 CET] <omerjerk> Cool. got it. Will start with it. 
[18:59:40 CET] <jkqxz> wm4:  Well, I don't feel guilty for not testing an ancient version which only works on hardware at least five generations old.
[18:59:42 CET] <cone-963> ffmpeg 03Paul B Mahol 07master:e1dd97bd4c03: avfilter: add fillborders filter
[19:01:01 CET] <durandal_170> how much to expect from adding simd to hflip filter?
[19:10:09 CET] <wm4> why does av_get_pix_fmt_string() even exist
[19:10:22 CET] <wm4> its only purpose is literally printing a formatted list of pixfmts
[19:12:52 CET] <ubitux> to run ffmpeg -pixfmts iirc
[19:13:01 CET] <ubitux> -pix_fmts*
[19:13:05 CET] <wm4> so why doesn't ffmpeg.c implement this
[19:13:12 CET] <ubitux> probably a bad reason
[19:13:39 CET] <ubitux> ah it seems that's not even used
[19:13:52 CET] <ubitux> 2010
[19:14:36 CET] <ubitux> which comes from avcodec_pix_fmt_string() from even earlier
[19:15:16 CET] <ubitux> (2007)
[19:15:37 CET] <ubitux> it's probably safe to deprecate t
[19:15:39 CET] <ubitux> it*
[19:42:02 CET] <SortaCore> does libx264 still use the AVCodeContext?
[19:42:17 CET] <SortaCore> I have it in my code comments notes
[20:02:33 CET] <jamrial> durandal_170: probably a decent boost. looks like a job for pshufb
[20:03:39 CET] <durandal_170> but why it havent be added already?
[20:04:11 CET] <jamrial> because nobody wrote it i suppose :p
[20:20:14 CET] <durandal_170> jamrial: what if width is not multiple of 16?
[20:21:46 CET] <atomnuker> pad it to the left and adjust the dst data pointer such that the last pixel lines up with the start of the frame
[20:31:59 CET] <jdarnley> I love seeing users post the lame error message I put in configure
[21:09:08 CET] <cone-963> ffmpeg 03Martin Vignali 07master:9bed17cd0f02: checkasm/utvideo : be more explicit to the WIDTH_PADDED define
[21:31:28 CET] <JEEB> jamrial: posted the patch to fix the mediacodecdec build since you didn't :)
[21:38:45 CET] <jamrial> JEEB: ah cool, thanks
[21:39:03 CET] <JEEB> https://ffmpeg.org/pipermail/ffmpeg-devel/2017-December/221515.html
[21:40:16 CET] <jkqxz> JEEB:  IMO just apply that.  (Though, if you care, those headers are usually in alphabetical order.)
[21:41:14 CET] <jkqxz> And nice to know that lots of people are building mediacodec, I guess :/
[21:43:00 CET] <JEEB> jkqxz: I just checked in which order hevc had them and it seemed like "add to the end" more or less
[21:43:06 CET] <JEEB> if it's alphabetic that's fine by me too
[21:49:23 CET] <JEEB> also reported the HEVC SIMD build failure with clang + gas-preprocessor finally
[21:53:02 CET] <JEEB> alright moved the header to its spot
[21:56:31 CET] <cone-963> ffmpeg 03Jan Ekström 07master:6939b3cb9d60: mediacodecdec: fix build by including hwaccel.h
[22:12:52 CET] <cone-963> ffmpeg 03Mark Thompson 07master:a763d278274c: examples/vaapi_encode: Fix leak on hwframe init failure
[22:18:11 CET] <sfan5> michaelni: i think you're correct about ff_hevc_frame_nb_refs(), any idea where I can get the number of frames that reference the current one?
[23:05:47 CET] <durandal_170> yaj, i can replace compiler now
[23:11:36 CET] <atomnuker> durandal_170: what did you use?
[23:12:41 CET] <cone-963> ffmpeg 03Michael Niedermayer 07master:0409d333115e: avcodec/hevcdsp_template: Fix undefined shift in put_hevc_epel_bi_w_h()
[23:12:42 CET] <cone-963> ffmpeg 03Michael Niedermayer 07master:067408700453: avcodec/j2kenc: Fix out of array access in encode_cblk()
[23:12:43 CET] <cone-963> ffmpeg 03Michael Niedermayer 07master:b5587fd2c6ce: avcodec/jpeg2000: Only allocate Jpeg2000Pass for the encoder
[23:13:52 CET] <durandal_170> atomnuker: for remainder of mmsize i use non simd code
[00:00:00 CET] --- Sat Dec  2 2017