[Ffmpeg-devel-irc] ffmpeg-devel.log.20200202

[burek]

[00:00:00 CET] --- Sun Feb  2 2020
[04:03:16 CET] <taliho> ah the trac logger is back... what was the problem? 
[09:40:44 CET] <darkapex> can someone push my mlpenc patchset if it looks ok?
[09:55:39 CET] <durandal_1707> darkapex: concentrate on fixing actual bugs in code, when all bugs have been resolved i'm sure it will be pushed at once
[09:57:29 CET] <darkapex> fair enough. but current patchset does fix some if not all the lossless failures.
[10:11:52 CET] <durandal_1707> i believe Lynne is against that :(
[10:21:08 CET] <cone-663> ffmpeg 03Paul B Mahol 07master:69477e10f061: doc/filters: make filters order more consistent
[10:21:08 CET] <cone-663> ffmpeg 03Paul B Mahol 07master:84286789e638: doc/filters: fix tonemap_vaapi filter name in documentation
[10:21:53 CET] <Lynne> darkapex: have you integrated all the changes durandal_1707 made?
[10:29:08 CET] <kurosu> slide 14 from the Grok author on JPEG2000 could have be titled "why jp2k is slow", "what have you done" or "oh, that's why jp2k-ht, ok"
[10:30:33 CET] <darkapex> Lynne: yes. https://patchwork.ffmpeg.org/project/ffmpeg/patch/20171128195544.5283-2-onemda@gmail.com/ is covered in my patchset
[10:30:53 CET] <darkapex> i also merged https://patchwork.ffmpeg.org/project/ffmpeg/patch/20171128195544.5283-1-onemda@gmail.com/ on my github branch but did not send a patch. i can send if you want.
[10:54:29 CET] <durandal_1707> darkapex: do you have valgrind?
[11:09:48 CET] <darkapex> durandal_1707: never used it before but i can try
[11:12:34 CET] <durandal_1707> darkapex: mlpenc in typical encoding process dumps bunch of issues with integers, so compile mlpenc with -fsanitize=integer and/or -fsanitize=undefined
[11:24:06 CET] <darkapex> durandal_1707: tried that rn with gcc after adding -fsantize=undefined to CFLAGS. did not get any issues on mlpenc.
[11:24:38 CET] <darkapex> i think i have a rough idea on where the issue is happening with Lynne's sample ill share
[11:25:15 CET] <durandal_1707> darkapex: you need to recompile mlpenc.c
[11:25:33 CET] <durandal_1707> with both CFLAGS and LDFLAGS added with -fsantitze=integer
[11:25:47 CET] <durandal_1707> just edit ffbuild/config.mak
[11:25:59 CET] <darkapex> yeah did that. didn't do LDFLAGS will try again wait.
[11:26:13 CET] <durandal_1707> to recompile mlpenc.c just resave it
[11:26:28 CET] <darkapex> -fsanitize integer did not work with gcc btw
[11:26:34 CET] <darkapex> should i test with clang?
[11:27:04 CET] <durandal_1707> yea, i use clang
[11:29:24 CET] <durandal_1707> https://pastebin.com/jLSyGhtq
[11:30:45 CET] <durandal_1707> https://pastebin.com/ajmJENYP
[11:31:53 CET] <darkapex> i see
[11:32:05 CET] <durandal_1707> try encoding Lynne sample in this mode, if it shows new lines, it may point to source of problem
[11:32:43 CET] <darkapex> yeah so for lynne sample problem is definitely in huff lsb calculation. wait ill send a paste.
[11:32:53 CET] <durandal_1707> important is to try both integer and undefined variant
[11:35:35 CET] <darkapex> yeah im recompiling with clang will try and let you know
[11:35:50 CET] <darkapex> try encdoing lynne's sample with this btw https://pastebin.com/w5X1PMT8
[11:36:13 CET] <darkapex> the lsb_bits are being set to > 24 because the entropy coder shift + codebook shift > 24
[11:41:25 CET] <durandal_1707> darkapex: i do not have lynne sample i think
[11:41:42 CET] <darkapex> https://lynne.ee/files/sample.flac
[11:43:57 CET] <darkapex> https://pastebin.com/VgviWQQv these are the errors. huff_lsbs are being stored as 17+8=25 which messes the decoder.
[11:44:47 CET] <darkapex> btw i tried it with -fsanitize=integer and good catch i see many conversions. will try to fix em.
[14:16:46 CET] <cone-510> ffmpeg 03Paul B Mahol 07master:cd823dadf9d1: avfilter: add xfade opencl filter
[17:09:40 CET] <cone-510> ffmpeg 03Zane van Iperen 07master:b49404ea30bf: avcodec/adpcm_argo: simplify and move duplicated logic into a function
[17:09:41 CET] <cone-510> ffmpeg 03Michael Niedermayer 07master:fcc9f13717c8: avcodec/mlpdsp: Fix a invalid shift in ff_mlp_rematrix_channel()
[18:47:45 CET] <darkapex> durandal_1707: i figured it out!!!! the error was in the decoder :o
[18:48:50 CET] <darkapex> it reset huff_lsbs to 0 when they were greater than 24. which should only be done if a codebook is used, not when all bits are being bypassed.
[18:50:12 CET] <darkapex> apply this before decoding the mlp generated for lynne's sample https://pastebin.com/78yNwLvQ
[18:50:26 CET] <darkapex> bit exact output (ofc leaving the last few samples which dont get encoded at all)
[18:59:47 CET] <durandal_1707> is there reference binary decoder for mlp somewhere ?
[19:03:58 CET] <darkapex> surcode has one but its proprietary. ramiro revengg. both surcode encoder and decoder so he should be able to confirm if this change is ok or not.
[19:39:49 CET] <darkapex> durandal_1707: I was able to play the encoded bitstream via surcode, it did not throw any errors but it usually doesn't. 
[19:41:01 CET] <durandal_1707> does it have way to decode to wav ?
[19:41:32 CET] <darkapex> nope it just plays it back. it is extremely shit :/
[19:41:58 CET] <darkapex> ill try to figure out a way to create a dummy audio device in window, so that it throws the samples on the dummy device which saves it to a file
[19:42:29 CET] <darkapex> but if you wanna be sure about my change this is what made it click for me https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/mlpdec.c#L908
[19:42:55 CET] <darkapex> windows*
[19:44:47 CET] <durandal_1707> darkapex: write patch and send it ml
[19:44:53 CET] <darkapex> cool thx
[20:01:44 CET] <cone-510> ffmpeg 03Marton Balint 07master:c2b6493bf7cc: avformat/udp: remove setting cancel state from the TX thread
[20:01:45 CET] <cone-510> ffmpeg 03Marton Balint 07master:53aa76686e7f: avformat/udp: cancel pending IO on win32 manually
[00:00:00 CET] --- Mon Feb  3 2020