[Ffmpeg-devel-irc] ffmpeg-devel.log.20180712

[burek]

[00:06:34 CEST] <cone-553> ffmpeg 03Ruiling Song 07master:48a1abed133d: lavfi/opencl: add macro for opencl error handling.
[00:06:35 CEST] <cone-553> ffmpeg 03Ruiling Song 07master:923ee63d6cd1: lavfi/vf_avgblur_opencl: remove useless clFinish().
[00:18:35 CEST] <cone-553> ffmpeg 03Danil Iashchenko 07master:2bdb6b64969f: lavfi/avgblur_opencl: fix using uninitialized value
[01:29:06 CEST] <cone-553> ffmpeg 03Michael Niedermayer 07master:c995e01b1e01: avformat/mov: remove modulo operations from mov_estimate_video_delay()
[01:29:07 CEST] <cone-553> ffmpeg 03Michael Niedermayer 07master:3ce4034308a3: avformat/mov: Eliminate variable buf_size from mov_estimate_video_delay()
[02:56:56 CEST] <gnafu> Someone in another channel asked about AV1 in Ogg, and it got me thinking: Anyone working on AV1 in NUT yet?
[09:31:33 CEST] <cone-780> ffmpeg 03Zhong Li 07master:3c26ce464435: lavc/qsv: handle MFX_FRAMETYPE_UNKNOWN case
[10:50:50 CEST] <atomnuker> gnafu: its just a matter of adding a fourcc code to libavformat/nut.c
[11:33:24 CEST] <jkqxz> ZhongLi:  For the patch above, what is the case where you get UNKNOWN?  That seems pretty weird.
[12:42:59 CEST] <gagandeep> kierank: i am using update_thread_context to pass the buffer with coefficients to 2nd thread and then once all the buffers are allocated ff_thread_finish_setup
[12:43:15 CEST] <gagandeep> pass the buffer pointers i mean
[12:43:18 CEST] <kierank> Ok
[12:44:38 CEST] <gagandeep> the rest of the context is copied right, or i have to copy the context myself on using update context
[14:13:32 CEST] <gnafu> atomnuker: Is it really that simple?
[14:17:22 CEST] <atomnuker> yeah, its not complicated, its the same as adding support for av1 in matroska
[14:17:56 CEST] <atomnuker> though some people seem bent on overcomplicating that and having a special, non-standard mapping of av1 in matroska for no good reason
[14:17:59 CEST] <nevcairiel> you should check how much the matroska people are d iscussing this to death
[14:18:37 CEST] <atomnuker> I haven't heard anything since the original proposal which insisted on annexb, not on the mailing list anyway
[14:18:53 CEST] <nevcairiel> you must not be on the right mailing list
[14:19:01 CEST] <nevcairiel> but the annexb thing was killed anyway
[14:19:17 CEST] <atomnuker> I am on the main av1 one, but not on the cellar one
[14:19:54 CEST] <nevcairiel> latest version can be read here anyway https://github.com/Matroska-Org/matroska-specification/blob/av1-mappin/codec/av1.md, plus changes in the commits of course
[14:19:59 CEST] <atomnuker> yeah, just saw
[14:20:06 CEST] <atomnuker> 39 commits since I first saw it
[14:20:10 CEST] <j-b> atomnuker: the spec has been updated with remarks from cellar, the main ML, the IRC logs
[14:20:20 CEST] <j-b> atomnuker: we plan to make it official at the same time as the MP4 one.
[14:24:09 CEST] <atomnuker> j-b: do you know if there's been any development on any list/irc about having more accurate timestamps?
[14:25:05 CEST] <j-b> atomnuker: I have seen no discussion about timestamps since a long time.
[14:25:14 CEST] <j-b> (for AV1)
[14:25:20 CEST] <j-b> or are you talking about MKV?
[14:25:54 CEST] <atomnuker> mkv
[14:26:03 CEST] <nevcairiel> i mentioned this the last time, you can have like nanosecond timestamps, that should be plenty? =p
[14:26:25 CEST] <j-b> atomnuker: yes, of course.
[14:26:33 CEST] <j-b> atomnuker: but noone cared much, IIRC.
[14:26:41 CEST] <j-b> atomnuker: it's trivial to make, if you want.
[14:27:07 CEST] <atomnuker> nevcairiel: wat? matroska still requires millisecond precision timestamps, no?
[14:27:14 CEST] <nevcairiel> not at all
[14:27:50 CEST] <atomnuker> you can't use a timebase which isn't 1/1000 when muxing, else everything breaks
[14:28:03 CEST] <nevcairiel> if thats even true, then its not matroskas problem
[14:28:22 CEST] <nevcairiel> it allows up to full nanosecond precision
[14:28:28 CEST] <nevcairiel> although thats probably a bit overkill
[14:29:23 CEST] <nevcairiel> you can set the TimecodeScale value in the mkv header to indicate the scale of the timestamps relative to nanoseconds, ie. the default scale of 1000000 means every timestamp increment is 1ms
[14:29:31 CEST] <nevcairiel> so if you used a TimecodeScale of 1, it would only be 1ns
[14:31:42 CEST] <atomnuker> put_ebml_uint(pb, MATROSKA_ID_TIMECODESCALE, 1000000);
[14:31:55 CEST] <atomnuker> grr, this blows
[14:31:58 CEST] <nevcairiel> blame the muxer, not the format :)
[14:33:36 CEST] <atomnuker> j-b: anyway, what's the difference currently between webm and matroska av1 mapping? shouldn't both be basically identical
[14:33:48 CEST] <nevcairiel> they will be
[14:33:55 CEST] <nevcairiel> there is no official webm mapping yet
[14:34:09 CEST] <nevcairiel> what aomenc currently produces is just a random file with no official menaing
[14:34:11 CEST] <j-b> atomnuker: they will be 100% the same.
[14:34:32 CEST] <j-b> but the issue is that currently both avformat and aomenc are not respecting the spec when producing webm
[14:35:58 CEST] <nevcairiel> i dont think avformat can even produce webm/mkv with av1
[14:37:01 CEST] <nevcairiel> not webm at the very least since there are codec checks
[14:38:20 CEST] <j-b> ok, I might have misunderstood.
[14:38:32 CEST] <j-b> I was waiting for james av1/mp4 patches to be merged :)
[14:45:52 CEST] <durandal_1707> atomnuker: i dont see noise reduction code in main opus, only in experimental branch, also who is working on porting that code?
[14:55:59 CEST] <atomnuker> not me right now, I'm working on ffv2 and the lavu mdcts, you can go ahead and port that to a filter
[14:56:17 CEST] <atomnuker> I thought it was merged in libopus git master, I guess it hasn't
[14:56:29 CEST] <atomnuker> I'll ask jmspeex if he intends to
[14:57:18 CEST] <atomnuker> j-b: oh, ok, I thought webm mapping was already defined (since we've been using it as a container since the codec was started)
[14:59:29 CEST] <j-b> atomnuker: it is invalid and wrong, so no.
[20:44:47 CEST] <gagandeep> guys, when using update_thread_context, do i also manually need to write code inside this funcition to copy the avctx or just the part i want to be changed?
[20:52:15 CEST] <gagandeep_> nevermind, in pngdec.c, it isn't done so i am assuming this function is only to change relevant avctx parts
[22:59:38 CEST] <jkqxz> Are there any official AV1 test streams yet?
[23:01:03 CEST] <TD-Linux> no, not yet.
[23:03:43 CEST] <jamrial> libaom downloads some
[23:03:50 CEST] <jamrial> when you run the tests
[23:04:05 CEST] <jamrial> https://aomedia.googlesource.com/aom/+/master/test/test_vectors.cc
[23:05:01 CEST] <jamrial> jkqxz: ^
[23:05:03 CEST] <gnafu> See also: https://aomedia.googlesource.com/aom/+/master/test/test_vectors.cc
[23:05:05 CEST] <jamrial> feature coverage is not too great, though. for example, none seem to use delayed random access points
[23:05:17 CEST] <gnafu> Oops
[23:05:24 CEST] <gnafu> Meant: http://storage.googleapis.com/aom-test-data
[23:05:54 CEST] <gnafu> I believe running 'make testdata' downloads all of the current test files.
[23:05:55 CEST] <jamrial> no 12bit samples, no still picture (with or without reduced header)
[23:05:57 CEST] <TD-Linux> those are the same files. and yeah they don't cover much of the important stuff to test in ffmpeg (seeking)
[23:06:02 CEST] <TD-Linux> their codec coverage isn't good either
[23:09:04 CEST] <kierank> GOOG should just pay argon to publish them
[23:09:41 CEST] <atomnuker> I think argon is basically google
[23:09:42 CEST] <jkqxz> Hmm, yeah.  I guess they're coverage for size and quantisation edge cases, but not much else?
[23:12:35 CEST] <jamrial> one intra-only sample and that's it, yeah
[23:17:21 CEST] <kierank> 10:09:39 PM <"atomnuker> I think argon is basically google
[23:17:22 CEST] <kierank> atomnuker: ?
[23:18:29 CEST] <gnafu> Je suis Google?
[23:34:01 CEST] <atomnuker> kierank: well they have a weird relationship, I think google invested quite a bit in them
[23:34:49 CEST] <atomnuker> and afaik hired them and even reduced the entry requirements to get into aom to do work on av1
[23:36:14 CEST] <kierank> atomnuker: hired maybe but the public record doesn't show google investment
[23:37:22 CEST] <atomnuker> dunno then, I always have had the impression they had some relation to google
[23:37:52 CEST] <kierank> maybe spent a lot
[23:37:59 CEST] <kierank> the public record is a bit old though
[23:49:52 CEST] <kierank> nevcairiel: not sure what carl is asking for
[23:49:55 CEST] <kierank> he's asking for 264
[23:51:22 CEST] <jkqxz> jamrial:  <http://ixia.jkqxz.net/~mrt/ffmpeg/av1/0002-lavc-Add-coded-bitstream-read-write-support-for-AV1.patch>
[23:52:44 CEST] <jkqxz> Not complete but does implement a working av1_metadata filter.  (I think it needs a full set of streams and traces from something else to compare with to finish with confidence that it's actually right.)
[00:00:00 CEST] --- Fri Jul 13 2018