[Ffmpeg-devel-irc] ffmpeg-devel.log.20160928

burek burek021 at gmail.com
Thu Sep 29 03:05:02 EEST 2016


[00:10:14 CEST] <llogan> the maintainer for qdm2 is not active, but you can give it a day or two for others to have a chance to comment since it was submitted only ~3.5 hours ago
[00:16:27 CEST] <Chloe> Patchwork's patch button should be named diff
[01:25:34 CEST] <cone-103> ffmpeg 03Sophia Wang 07master:8c83062acb22: avformat/matroskadec: retain error codes in matroska_resync() and matroska_read_packet()
[01:31:37 CEST] <cone-103> ffmpeg 03James Almer 07master:29b6c2be291e: avformat/matroskadec: set AVCodecParameters.field_order on progressive video
[02:23:30 CEST] <cone-103> ffmpeg 03James Almer 07master:92de2c23a61e: ffprobe: don't use AVStream.codec to set decoder framerate
[05:27:18 CEST] <cone-146> ffmpeg 03Sasi Inguva 07master:7e9e1b707024: lavc/movtextdec.c: Avoid infinite loop on invalid data.
[11:06:45 CEST] Action: wm4 is waiting for google to classify these j2eeguys.com as spam
[11:08:10 CEST] <BtbN> I already did just that.
[11:23:17 CEST] <nevcairiel> me too
[11:23:53 CEST] <BtbN> if that continues, can't we just unsubscribe him? It's not like he's getting any mails from the list anyway.
[13:19:12 CEST] <atomnuker> does checkasm not work without a make install?
[13:23:45 CEST] <Chloe> michaelni__: is there a specific type of qdm2 sample we need to reach that code?
[13:27:15 CEST] <michaelni__> Chloe, yes
[13:30:02 CEST] <Chloe> I should have been more specific, but looking at it, I doubt I could find an appropriate sample, whatever the special feature it requires is.
[13:34:36 CEST] <Chloe> michaelni__: I'm going to push that with a diff commit message: 'lavc/qdm2: increase code clarity' unless you have a better commit message
[13:35:30 CEST] <michaelni__> ok
[13:36:20 CEST] <BtbN> that guy who submited patches for nvenc got them merged in libav instead of fixing the issues that were pointed out at ffmpeg-devel...
[13:37:40 CEST] <Chloe> Isn't it normally the other way round? xD
[13:37:48 CEST] <nevcairiel> i thought you had a deal with libav to try to keep things unified
[13:37:52 CEST] <nevcairiel> guess that didnt work out?
[13:38:14 CEST] <BtbN> The ffmpeg nvenc is basically identical to the libav one
[13:38:33 CEST] <BtbN> And it looks like Luca just merged those commits, and I never talked to him
[13:39:09 CEST] <nevcairiel> if the changes are crap we can just not merge them when we reach that point (some day)
[13:39:11 CEST] <BtbN> Strangely enough, the issues I pointed out were fixed. So I'm just going to cherry-pick them.
[13:39:13 CEST] <cone-480> ffmpeg 03Adriano Pallavicino 07master:25866680fd6b: lavc/qdm2: increase code clarity
[13:39:33 CEST] <Chloe> odd that they didn't resubmit
[13:39:43 CEST] <BtbN> No, the changes were fine. Adding new SDK 7 features. Just some issues like unused variables and parameters
[13:42:16 CEST] <JEEB> any comments here regarding the smooth streaming patches?
[13:43:44 CEST] <BtbN> hm, can't really cherry-pick the commits, as they bring the libav encoder on par with the ffmpeg one. So I'll just make a new commit with the diferences, and reference the libav commits.
[13:44:09 CEST] <BtbN> https://git.libav.org/?p=libav.git;a=blob;f=libavcodec/nvenc.c;h=7f8737e0bf3224f3f88849fc8849f50252f484f6;hb=e02e2515b24bfc37ede6ca1744696230be55e50b#l467 wow, a typo!
[13:46:42 CEST] <BtbN> that function is super weird in libav anyway
[13:46:52 CEST] <BtbN> it returns an int, which is allways an error code, which is then just never checked
[13:57:00 CEST] <BtbN> https://git.libav.org/?p=libav.git;a=commitdiff;h=352741b5ead1543d775ccf6040f33023e4491186
[13:57:05 CEST] <BtbN> is that syntax even standard C?
[13:57:39 CEST] <nevcairiel> yes
[13:58:42 CEST] <BtbN> what will it do if there are gaps in the indices? Just fill it with 0?
[13:58:54 CEST] <BBB> yes
[13:59:04 CEST] <BBB> static const data is initialized to zero if unset
[14:02:37 CEST] <flux> the syntax is actually more versatile than I thought: http://www.drdobbs.com/the-new-c-declarations-initializations/184401377
[14:04:28 CEST] <BtbN> I expected it to be a GNU extension
[14:05:14 CEST] <wm4> just C99
[14:05:22 CEST] <wm4> we already use it in some other places
[14:05:55 CEST] <wm4> one case in a public header: http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavutil/error.h;h=71df4da353b9cd95e45e86c14acc0c285854f0af;hb=HEAD#l119
[14:06:06 CEST] <wm4> (pretty trippy shit, but very convenient)
[14:50:17 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:ba642f031906: avformat/utils: End probing if the expected codec surpasses AVPROBE_SCORE_STREAM_RETRY
[14:50:18 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:ed38046c5c2e: avformat/avidec: Fix infinite loop in avi_read_nikon()
[14:50:19 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:5aaf7e31829d: swscale/swscale_unscaled: Fix packed_16bpc_bswap() with slices
[14:50:20 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:e91b7852dfae: swscale/swscale_unscaled: Try to fix Rgb16ToPlanarRgb16Wrapper() with slices
[14:50:21 CEST] <cone-480> ffmpeg 03Sergey Volk 07release/3.1:7a3dc2f7b6c2: avformat/mov: Fix potential integer overflow in mov_read_keys
[14:50:22 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:ed1c6f701a78: avcodec/svq3: Reintroduce slice_type
[14:50:23 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:65c10f0f5c52: avcodec/ccaption_dec: Use simple array instead of AVBuffer
[14:50:24 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:21a979773783: avformat/movenc: Check first DTS similar to dts difference
[14:50:25 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:c2087fc48b46: avcodec/avpacket: clear side_data_elems
[14:50:26 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:ac8ac46641ad: avcodec/g726: Add missing ADDB output mask
[14:50:27 CEST] <cone-480> ffmpeg 03Xinzheng Zhang 07release/3.1:c68ce48260cf: avformat/utils: fix timebase error in avformat_seek_file()
[14:50:28 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:03f996d18346: avformat/movenc: Factor check_pkt() out
[14:50:29 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:77c9c3509309: avformat/movenc: Check packet in mov_write_single_packet() too
[14:50:30 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:9d738e696875: avcodec/cavsdsp: use av_clip_uint8() for idct
[14:50:31 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:496267f8e9ec: avcodec/ansi: Check dimensions
[14:50:32 CEST] <cone-480> ffmpeg 03Sasi Inguva 07release/3.1:39dc26f0c104: lavc/movtextdec.c: Avoid infinite loop on invalid data.
[14:50:33 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:8834e080c20d: avformat/avidec: Fix memleak with dv in avi
[14:50:34 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.1:e6351504dc54: Update for 3.1.4
[15:05:17 CEST] <wm4> michaelni__: do you know about HLS regressions in the current ffmpeg release?
[15:07:02 CEST] <nevcairiel> reading or writing?
[15:12:41 CEST] <michaelni__> wm4, no
[15:13:36 CEST] <wm4> demuxing
[15:13:47 CEST] <wm4> they say the initial segment is repeated
[15:23:47 CEST] <BtbN> So that's what's happening. I was blaming Kodi for it.
[15:24:01 CEST] <BtbN> Yes, that's definitely happening.
[15:28:01 CEST] <wm4> here's the mpv bug report: https://github.com/mpv-player/mpv/issues/3480
[15:28:40 CEST] <wm4> according to this 2.8.7 works, 3.1.3 doesn't work, ffmpeg git works
[15:33:44 CEST] <anrao> durandal_1707: ping
[16:03:02 CEST] <michaelni__> should i wait with releasing 3.1.4 ? (i had planned to release later today but i can wait if someone wants to backport a hls fix, i can also cherry pick one if i know what needs to be picked)
[16:06:26 CEST] <BtbN> I have a super minor fix for nvenc.
[16:06:27 CEST] <BtbN> http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavcodec/nvenc_hevc.c;h=1ce7c89a4b2d5bb12d4affe1d5a5c9b2f2db909c;hb=refs/heads/release/3.1#l79
[16:06:42 CEST] <BtbN> it's missing the "device", so the const options don't do anything.
[16:07:22 CEST] <BtbN> I'll just push a fix for that to release/3.1 in a few minutes.
[16:08:37 CEST] <michaelni__> sure, no hurry, i wont release before evening/night for sure and am happy to wait long if people want
[16:09:12 CEST] <BtbN> if there is a fix for the hls issue, I'd definitely like to have it. I don't particulary care about it getting into 3.1.4 though.
[16:11:29 CEST] <cone-480> ffmpeg 03Timo Rothenpieler 07release/3.1:e60a00e0cc48: avcodec/nvenc: fix const options for hevc gpu setting
[16:12:39 CEST] <michaelni__> s/long/longer/
[16:18:41 CEST] <mateo`__> michaelni__: I think I need to backport a MediaCodec fix for 3.1.4
[16:23:07 CEST] <BtbN> What is up with x86/vp9itxfm.asm, just noticed that it takes by far the longest to compile out of all c/asm files in ffmpeg.
[16:25:11 CEST] <cone-480> ffmpeg 03Matthieu Bouron 07release/3.1:d0590d93493a: lavc/mediacodecdec_h264: fix SODB escaping
[16:25:15 CEST] <mateo`__> michaelni__: done
[16:26:45 CEST] <atomnuker> BtbN: make sure you use a modern yasm version or better yet nasm, it's fast on my machine
[16:27:00 CEST] <BtbN> nasm symlinked to yasm?
[16:27:06 CEST] <wm4> michaelni__: unfortunately I don't know what causes the HLS issue
[16:27:18 CEST] <BtbN> atomnuker, I'm using yasm 1.2.0
[16:27:39 CEST] <Gramner> BtbN: that file is just large. also yasm 1.3.0 is much faster than 1.2.0
[16:27:44 CEST] <atomnuker> 1.3.0 has been out for years
[16:27:55 CEST] <BtbN> 1.2 is stable on gentoo though, 1.3 marked as testing
[16:28:06 CEST] <Gramner> https://github.com/yasm/yasm/commit/e32ff2c2abc22532a58c9b687d922500d23fd709
[16:28:35 CEST] <BtbN> Also, i have nasm installed as well. Is it better in general, or just another option?
[16:28:40 CEST] <Gramner> best diffsize:speedincrease ratio ever?
[16:29:33 CEST] <kierank> ah ubuntu has 1.2
[16:29:36 CEST] <atomnuker> BtbN: yep, it's developed and updated more often
[16:29:36 CEST] <Gramner> yasm is essentially dead, it doesn support new instruction sets so migration to nasm is going to happen eventually
[16:29:48 CEST] <Gramner> nasm is activly developed
[16:29:49 CEST] <BtbN> atomnuker, but ffmpeg seems to prefer yasm if both are available, hm
[16:30:01 CEST] <BBB> because nasm did stupid things back when
[16:30:02 CEST] <jamrial> yasm devs didnt want to deal with avx512?
[16:30:06 CEST] <kierank> it used to be that nasm was the one that wasn't developed
[16:30:36 CEST] <Gramner> jamrial: there aren't any active yasm devs any more, nothing about avx-512 in particular
[16:30:44 CEST] <jamrial> ah, i see
[16:31:31 CEST] <Gramner> although the fact that avx-512 make big changes and requires large part of the assembler to be rewritten probably doesn't help
[16:31:41 CEST] <BBB> Gramner: cant you become a yasm dev?
[16:32:00 CEST] <Gramner> I have more patches in nasm than in yasm
[16:32:15 CEST] <wm4> huh, why would new instructions require a rewrite?
[16:32:16 CEST] <BtbN> some configure option to select nasm would be nice
[16:32:28 CEST] <BtbN> wm4, because the size of various registers changed i guess?
[16:32:40 CEST] <BBB> BtbN: its there already?
[16:32:54 CEST] <BtbN> BBB, hm?
[16:33:08 CEST] <BBB> --yasmexe=nasm
[16:33:11 CEST] <BBB> I think
[16:33:14 CEST] <Gramner> gonna switch x264 to nasm when hpa is done with his backend nasm rewrite, because nasm currently can't handle some of the clever PIC coding in x264
[16:33:18 CEST] <BtbN> BBB, ah, hmm
[16:33:25 CEST] <jamrial> it also uses nasm if yasm isn't available
[16:33:33 CEST] <jamrial> we'll have to make configure favor nasm eventually
[16:34:19 CEST] <Gramner> wm4: avx-512 adds completely new features like opmasks, embedded bradcasts, embedded floating-point rounding-modes etc.
[16:34:41 CEST] <Gramner> and a whole new instruction encoding scheme (evex)
[16:35:02 CEST] <Gramner> probably the biggest change to the x86 instruction set ever tbh
[16:47:27 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07master:14bac7e00d72: avformat/avidec: Remove ancient assert
[16:47:28 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07master:2679ad4773aa: avformat/avidec: Check nb_streams in read_gab2_sub()
[16:57:15 CEST] <cone-480> ffmpeg 03Timo Rothenpieler 07master:a81b000a392e: avcodec/nvenc: Make sure that enum and array index match
[16:57:16 CEST] <cone-480> ffmpeg 03Timo Rothenpieler 07master:033f98c902f5: avcodec/nvenc: add HEVC REXT profile
[16:57:17 CEST] <cone-480> ffmpeg 03Yogender Gupta 07master:facc19ef06a7: avcodec/nvenc: Extended rate-control support as provided by SDK 7
[16:57:18 CEST] <cone-480> ffmpeg 03Timo Rothenpieler 07master:c03b9d6a62db: avcodec/nvenc: nicely align AVOptions
[16:58:13 CEST] <nevcairiel> man when did the videolan gitweb break, it now shows all merge commits for all files even if they didnt touch that file
[16:58:28 CEST] <nevcairiel> reading history for single files is impossible
[16:58:45 CEST] <BtbN> I'm using github for that since a long time.
[16:59:08 CEST] <nevcairiel> githubs web interface is just so sluggish and slow
[16:59:15 CEST] <BtbN> Also, 4 less commits in the libav merge queue! Well, almost.
[16:59:26 CEST] <nevcairiel> cant even browse avcodec folder properly because it'll scream at you for there being too many files
[17:00:28 CEST] <wm4> lately I've even seen it hide some issues on long threads (you have to click to expand them)
[17:03:44 CEST] <jamrial> nevcairiel: that started since the server upgrade, which included a new gitweb version
[17:03:59 CEST] <nevcairiel> figures
[17:04:07 CEST] <jamrial> j-b or whoever handles this may be able to solve it
[17:04:27 CEST] <nevcairiel> i faintly remember having that on my personal gitweb as well, but i forgot how i fixed it
[17:10:14 CEST] <philipl> BtbN: cuda 8 final has the ancient garbage cuvid headers in it. Clearly they want people to get them from the video sdk. Not a surprise in the end.
[17:11:41 CEST] <nevcairiel> not like the headers really changed for a long time, other then some cleaning here and there
[17:11:48 CEST] <nevcairiel> and the odd new codec id they added
[17:12:59 CEST] <j-b> jamrial: see with thresh
[17:36:01 CEST] <cone-480> ffmpeg 03Sylvain Laurent 07master:58776ccbdb4d: lavf/mpegtsenc: Set min PID for data pkt to 0x0010.
[17:50:02 CEST] <cone-480> ffmpeg 03Carl Eugen Hoyos 07master:83bf40f3cfcf: lavc/8bps: Fix 32bit output of 24bit video.
[18:02:59 CEST] <cone-480> ffmpeg 03Anton Khirnov 07master:5fa255b65c78: avconv: initialize output framerate earlier
[18:03:00 CEST] <cone-480> ffmpeg 03Anton Khirnov 07master:6ed0f70f97c8: avconv: factor out initializing stream parameters for streamcopy
[18:03:01 CEST] <cone-480> ffmpeg 03James Almer 07master:df2ae8f3f0a8: Merge commit '5fa255b65c7887cc913f097aed1b581fbf1a8745'
[18:03:02 CEST] <cone-480> ffmpeg 03James Almer 07master:3597d32e88c8: Merge commit '6ed0f70f97c882813199b3bafd724ceeb43659de'
[18:04:12 CEST] <ubitux> any comment on "[PATCH] doc/libav-merge: complete TODO section"?
[18:05:30 CEST] <jamrial> ubitux: it needs one more line about the first of the merges i just pushed
[18:05:42 CEST] <jamrial> but it's fine otherwise
[18:06:25 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:6744d3f6b953: swscale/swscale_unscaled: Fix packed_16bpc_bswap() with slices
[18:06:26 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:cb8a29f87290: swscale/swscale_unscaled: Try to fix Rgb16ToPlanarRgb16Wrapper() with slices
[18:06:27 CEST] <cone-480> ffmpeg 03Sergey Volk 07release/3.0:9259b7f38e00: avformat/mov: Fix potential integer overflow in mov_read_keys
[18:06:28 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:d669b7f4f6e8: avcodec/ccaption_dec: Use simple array instead of AVBuffer
[18:06:29 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:b7b3b0086d4e: avformat/movenc: Check first DTS similar to dts difference
[18:06:30 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:26a8fc1c00d5: avcodec/avpacket: clear side_data_elems
[18:06:31 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:9357aa67572c: avcodec/g726: Add missing ADDB output mask
[18:06:32 CEST] <cone-480> ffmpeg 03Xinzheng Zhang 07release/3.0:f2f7d49f41b1: avformat/utils: fix timebase error in avformat_seek_file()
[18:06:33 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:a1f77124c876: avformat/movenc: Factor check_pkt() out
[18:06:34 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:7fefa4138ddd: avformat/movenc: Check packet in mov_write_single_packet() too
[18:06:35 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:63f951601e75: avcodec/cavsdsp: use av_clip_uint8() for idct
[18:06:36 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:e5bf7ab3e7c6: avcodec/ansi: Check dimensions
[18:06:37 CEST] <cone-480> ffmpeg 03Sasi Inguva 07release/3.0:82b58841c9d0: lavc/movtextdec.c: Avoid infinite loop on invalid data.
[18:06:38 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:fb7617df4eb1: avformat/avidec: Fix memleak with dv in avi
[18:06:39 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:77d5a237ef68: avformat/avidec: Remove ancient assert
[18:06:40 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:8c43f320574d: avformat/avidec: Check nb_streams in read_gab2_sub()
[18:06:41 CEST] <cone-480> ffmpeg 03Michael Niedermayer 07release/3.0:b9a1d389b2a0: Update for 3.0.4
[18:06:49 CEST] <jamrial> ubitux: i'm going to merge a few more commits
[18:06:53 CEST] <ubitux> jamrial: ok ok
[18:06:58 CEST] <ubitux> i'll delay the patch
[18:11:00 CEST] <cone-480> ffmpeg 03Neil Birkbeck 07master:5d560d38deca: pixfmt: Add ARIB STD-B76 color transfer characteristic
[18:11:01 CEST] <cone-480> ffmpeg 03James Almer 07master:f013ba475b0d: Merge commit '5d560d38deca1e4705e6d3784d737363b9c830fe'
[18:12:54 CEST] <nevcairiel> there should be a bunch of easy merges for now, i think
[18:15:04 CEST] <cone-480> ffmpeg 03Vittorio Giovara 07master:a8164323374e: pixdesc: Add new SMPTE 431, 432, and 2085 color properties
[18:15:05 CEST] <cone-480> ffmpeg 03James Almer 07master:6e76c9c45018: Merge commit 'a8164323374e86ce5f93759230868c98356833a2'
[18:20:19 CEST] <cone-480> ffmpeg 03Vittorio Giovara 07master:40dd5166d2ba: truemotion2rt: Use ff_set_dimensions
[18:20:20 CEST] <cone-480> ffmpeg 03James Almer 07master:0153cc49b20a: Merge commit '40dd5166d2ba4f9035b93748840e408cd8be40e5'
[18:25:39 CEST] <cone-480> ffmpeg 03Anton Khirnov 07master:e85f6f7f8d03: lavc: allow using AVCodecContext.hw_frames_ctx for decoding
[18:25:39 CEST] <cone-480> ffmpeg 03James Almer 07master:32c25f06b79f: Merge commit 'e85f6f7f8d037c0af0f294000718d9ba22753baa'
[18:27:18 CEST] <cone-480> ffmpeg 03Anton Khirnov 07master:6f19bbcf8532: qsvdec: move reading the user-provided session to qsv_decode_init()
[18:27:19 CEST] <cone-480> ffmpeg 03James Almer 07master:e9a5fc967867: Merge commit '6f19bbcf8532d018d8d6d82e000738d0ac2385c9'
[18:32:09 CEST] <cone-480> ffmpeg 03Anton Khirnov 07master:59e7361cc791: hwcontext: add a QSV implementation
[18:32:10 CEST] <cone-480> ffmpeg 03James Almer 07master:eba041476879: Merge commit '59e7361cc791e5103be1712dc59a2055f118d0da'
[18:32:23 CEST] <jamrial> ok, done
[18:32:25 CEST] <wm4> is it fine to add artificial hacks to ffmpeg.c just to make them pass fate under corner cases?
[18:32:36 CEST] <wm4> s/them/it/
[18:32:56 CEST] <durandal_1707> never enough hacks in ffmpeg.c
[18:37:25 CEST] <durandal_1707> nevcairiel: why you added magicyuv support to lav but not sheervideo?
[18:38:10 CEST] <nevcairiel> because i dont even know what that is
[18:39:01 CEST] <durandal_1707> nevcairiel: loosless codec just like magicyuv
[18:54:28 CEST] <BtbN> wm4, a hack that can't be put in any of the libraries instead?
[18:55:37 CEST] <jamrial> ubitux: so yeah, add a line about 5fa255b65c78 needing to be applied before further avconv commits and push that doc patch
[18:56:58 CEST] <wm4> BtbN: well, it's about ffmpeg.c wanting to send dts timestamps over flush packets (I've mentioned that before)
[19:46:05 CEST] <philipl> BtbN: https://github.com/philipl/FFmpeg/commit/d6573275a71554cef25a78e462e688f58d585423
[19:46:32 CEST] <BtbN> how else does it get included?
[19:46:57 CEST] <philipl> ffmpeg_cuvid.c doesn't make any cuvid calls. just base cuda ones.
[19:47:06 CEST] <BtbN> ah, right.
[19:47:13 CEST] <BtbN> And I think it can be even further slimmed down?
[19:47:31 CEST] <philipl> If we break out the dlopen stuff from nvenc and share it, yes.
[19:47:43 CEST] <BtbN> hm?
[19:47:55 CEST] <BtbN> I meant using the new hwctx allocation
[19:48:04 CEST] <philipl> cuInit/cuDeviceGet/cuCtxCreate
[19:48:05 CEST] <philipl> oh.
[19:48:16 CEST] <BtbN> https://github.com/FFmpeg/FFmpeg/blob/master/libavutil/hwcontext_cuda.c#L262
[19:48:38 CEST] <philipl> I see.
[19:48:38 CEST] <BtbN> that makes most of ffmpeg_cuvid.c obsolete. I just haven't gotten around to making use of it.
[19:48:41 CEST] <philipl> yeah, new merge.
[19:50:48 CEST] <philipl> BtbN: do you think it's worth moving to dlopen everything? Given the nvenc precedence - redeclaring the cuda prototypes and dlopening should qualify cuvid to stop being nonfree.
[19:51:20 CEST] <BtbN> cuvid needs the cuda.h header. And the license in that is too obscure for anything.
[19:51:39 CEST] <BtbN> And I'm not in favor of duplicating more and more stuff of cuda into ffmpeg
[19:51:52 CEST] <philipl> we edit the header not to include cuda.h and redeclare. nvenc redeclares a bunch of cuda.h contents.
[19:52:12 CEST] <BtbN> At some point it becomes questionable if copying large parts of the header warants the license
[19:52:24 CEST] <nevcairiel> and then you have to proof you figured this out without reading cuda.h to not kill the license
[19:53:14 CEST] <philipl> memcpy2d is the only function that we don't already redeclare for nvenc.
[19:53:32 CEST] <philipl> which is unfortunately non trivial due to the struct arg
[19:53:39 CEST] <BtbN> which nvenc doesn't need without global cuda support.
[19:54:28 CEST] <philipl> nevcairiel: what's your basis for lavf being lgpl with cuda support?
[19:55:15 CEST] <philipl> (or is it also a nonfree build?)
[19:55:29 CEST] <BtbN> --enable-cuda needs nonfree, yes.
[19:55:49 CEST] <BtbN> There is no way to include the cuda.h header without making it a nonfree build.
[19:55:57 CEST] <BtbN> The license in there is too broken.
[19:56:04 CEST] <philipl> I agree - it's a useless licence.
[19:56:56 CEST] <philipl> It's just my point here is that the difference between nvenc which we think can be free and cuvid which is not, is one function prototype from cuda.h
[19:58:05 CEST] <RiCON> ffmpeg configure needs an update to support openjpeg 2.2
[19:58:23 CEST] <BtbN> I also don't want to cater nvidias horrible licensing practise by weaseling our way around it.
[19:58:44 CEST] <BtbN> If they want wider adoption of their stuff, they should fix the license situation.
[19:58:52 CEST] <philipl> This is true.
[19:59:14 CEST] <BtbN> Put a MIT header on the cynload_cuda.h, and everything is fine.
[19:59:19 CEST] <BtbN> Except for libnpp*
[19:59:25 CEST] <BtbN> That is nonfree forever.
[19:59:41 CEST] <philipl> you just need to merge nvresize to replace it :-P
[20:00:23 CEST] <philipl> We got them to relicence the video headers by asking. Maybe it's not a waste of time to ask about cuda.h
[20:09:39 CEST] <cone-480> ffmpeg 03Philip Langdale 07master:d6573275a715: ffmpeg_cuvid: Don't unnecessarily include nvcuvid.h
[20:13:35 CEST] <cone-480> ffmpeg 03James Almer 07master:13dd5edb88c9: avutil/hwcontext: use CONFIG_QSV instead of CONFIG_LIBMFX for qsv
[20:36:04 CEST] <philipl> BtbN: After the latest set of merges, I needed to duplicate your hwframe format hack from ffmpeg_cuvid.c into mpv.
[20:36:11 CEST] <philipl> I'm not sure what changed to require that.
[20:36:38 CEST] <BtbN> which hack?
[20:37:49 CEST] <cone-480> ffmpeg 03raymondzheng1412 at gmail.com 07master:2366efce3cd5: avformat/concatdec: don't call open_file when seek position within a file
[20:39:36 CEST] <philipl> BtbN: /* This is a bit hacky, av_hwframe_ctx_init is called by the cuvid decoder... etc
[20:40:04 CEST] <BtbN> well, of course? You need to init the context.
[20:40:24 CEST] <JEEB> anyone else has any interest in the smooth streaming movenc patches I posted?
[20:40:27 CEST] <JEEB> "Makes sense to me but I'd prefer if somebody else (looks again and) applies"
[20:40:31 CEST] <JEEB> from cehoyos
[20:40:34 CEST] <JEEB> :D
[20:44:23 CEST] <philipl> I wasn't doing it before, and it worked fine
[21:36:18 CEST] <BtbN> philipl, you didn't specify a hwframe_ctx at all?
[21:36:31 CEST] <BtbN> Or just a completely unititialized oned?
[21:52:33 CEST] <rcombs> JEEB: people still have a use for smoothstreaming?
[21:52:39 CEST] <JEEB> yes, unfortunately
[21:52:44 CEST] <JEEB> some streaming servers take it as input
[21:52:57 CEST] <JEEB> instead of proper segmented ISOBMFF
[21:53:07 CEST] <rcombs> kak
[21:53:14 CEST] <rcombs> well good luck
[21:53:54 CEST] <JEEB> yeah, the pre-codecpar version almost got merged but it had a timestamp-modifying patch included that I have since removed
[21:54:04 CEST] <JEEB> and I was :effort: to update FATE samples
[21:54:18 CEST] <JEEB> so as long as my post-codecpar changes are OK it should get merged :V
[21:54:34 CEST] <JEEB> unless there's some issues in general in them)
[21:56:07 CEST] <jamrial> michaelni__: where in the docs? a quick grep doesn't show anything for "colorspace_type" or "color_primaries_type"
[21:57:15 CEST] <jamrial> michaelni__: nevermind, codecs.texi it seems
[22:01:56 CEST] <jamrial> this file is a mess...
[22:28:34 CEST] <jamrial> BBB: see my latest patch
[22:30:30 CEST] <Chloe> What's PRI and SPC?
[22:32:27 CEST] <BBB> duh :(
[22:32:30 CEST] <BBB> ok
[22:32:35 CEST] <BBB> ty
[22:32:37 CEST] <BBB> jamrial: ^
[22:34:34 CEST] <cone-480> ffmpeg 03James Almer 07master:e4bfc9ecf73d: avfilter/vf_colorspace: fix range for output colorspace option
[22:34:35 CEST] <BBB> Chloe: primaries and space
[22:35:59 CEST] <BBB> Chloe: primaries define what color intensities (photon count) belong to linearized float values for r/g/b (or: xyz/rgb coefficients); colorspace defines the relationship between yuv and (exponential/gamma-corrected) rgb values (or: rgb/yuv coefficients)
[22:37:03 CEST] <BBB> Chloe: theres also the trc, i.e. linearization/gamma-correction (rgb-gamma/exp to rgb-linear) and range (jpeg/mpeg, i.e. do values in float 0.0-1.0 start at integer 0 or hav ean integer offset like in mpeg, e.g. 16)
[22:37:25 CEST] <BBB> these 4 (trc, spc, prm, rng) together define how yuv integer values are to be displayed on-screen in terms of photon count
[22:39:59 CEST] <JEEB> this reminds me that I should probably try to add support for non-gamme to zscale
[22:40:15 CEST] <JEEB> since the original zimg library added support for PQ and ARIB STD-B67
[22:44:26 CEST] <cone-480> ffmpeg 03James Almer 07release/3.1:2303cea5be08: avfilter/vf_colorspace: fix range for output colorspace option
[22:44:27 CEST] <cone-480> ffmpeg 03James Almer 07release/3.1:bc6174d4af76: Changelog: update after the last few commits
[22:52:43 CEST] <Chloe> BBB: is it specific to YUV?
[22:53:16 CEST] <BBB> trc/gamma are rgb (but since yuv is just a transformation of rgb, they indirectly apply to yuv also); spc/rng apply to yuv only
[22:53:28 CEST] <BBB> trc/prm, I mean
[22:54:30 CEST] <wm4> so what is libswscale's state
[22:54:39 CEST] <wm4> still oblivious to modernity?
[22:55:53 CEST] <Chloe> if you're talking about the inline asm, then yes, it's still there
[22:56:54 CEST] <wm4> I mean support for colorspaces
[22:57:16 CEST] <wm4> vf_colorspace shouldn't exist
[22:57:24 CEST] <BBB> we wrote a wiki page on that
[22:57:36 CEST] <BBB> I can do swscale colorspace integration long-term, but its harder than it sounds
[22:58:18 CEST] <wm4> maybe it would be easier to turn it around and make libswscale a frontend to libavfilter?
[22:58:31 CEST] <BBB> that would be inefficient
[22:58:38 CEST] <BBB> I mean sure it could be done, but itd be faster not to
[22:59:01 CEST] <BBB> (as in: it would take more time to code it but it would run faster in fps/cycle count)
[22:59:15 CEST] <Compn> wm4 : could always write a wrapper for ffmpeg to use imagemagick or other scaler implementations... ;p
[22:59:22 CEST] Action: Compn ducks
[22:59:28 CEST] <wm4> ...
[23:00:00 CEST] <wm4> BBB: why would it run faster? better cache locality by filtering slices?
[23:00:14 CEST] <BBB> also being able to combine filter passes
[23:00:35 CEST] <BBB> also allows for higher precision between all steps
[23:00:50 CEST] <BBB> so youre only clipping and truncating in the output phase
[23:00:57 CEST] <BBB> every clip/trunc affects quality
[23:01:03 CEST] <BBB> so there
[23:01:04 CEST] <BBB> oops
[23:01:21 CEST] <BBB> so theres lots of reasons wed want to do all of this in swscale, but like I Said, lots of work
[23:01:40 CEST] <BBB> I wrote a prototype scaler once called avscale which did this
[23:01:45 CEST] <BBB> but never got very far with it and lost interest
[23:01:54 CEST] <BBB> and it should probably be done inside swscale for compat reasons
[23:24:04 CEST] <philipl> BtbN: I created one using the same set of calls as ffmpeg_cuvid.c but didnt' set the format field.
[23:53:48 CEST] <cone-480> ffmpeg 03Jan Ekström 07master:6c10f8fe7658: movenc: use similar logic to DASH when writing bit rate to ISML
[00:00:00 CEST] --- Thu Sep 29 2016


More information about the Ffmpeg-devel-irc mailing list