michaelni changed the topic of #ffmpeg-devel to: Welcome to the FFmpeg development channel | Questions about using FFmpeg or developing with libav* libs should be asked in #ffmpeg | This channel is publicly logged | FFmpeg 7.1.1 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
HarshK23 has quit [Quit: Connection closed for inactivity]
_whitelogger has joined #ffmpeg-devel
<TheVibeCoder>
how to clone AVIOContext * so if it is used in such a way that when seeiking within such context one get always fixed +offset from original context ?
<JEEB>
Lynne: btw have you seen AMD's vulkan driver randomly deciding to allocate "shared" memory instead of actual VRAM, even if looking at the stats it seems like things would fit?
<Lynne>
this is solved by the mechanism to increase the visible VRAM addressing, right?
<JEEB>
no idea, I may look into the code of this thing later
<JEEB>
just wondered if you had seen something like that, esp. with compute
<Lynne>
not really? hard to notice on 24GiB of VRAM
<JEEB>
I'm at 10.9 or 11 gig VRAM usage, out of 16 (80MiB reserved), dedicated VRAM usage goes from 10.9 to 9.5. shared VRAM usage goes from 0.4 to 2.2. previous runs of that app had allocated dedicated memory fine, but then it went pröööt
<Lynne>
not entirely, I know the code I need to port to fix it, but I needed a few more hours without worrying about whether it was going to land in 8.0 or not, so I'll fix it tomorrow or so
<Sean_McG>
In file included from src/libavcodec/vulkan_av1.c:21:
<Sean_McG>
src/libavcodec/vulkan_decode.h:41:5: error: unknown type name ‘VkVideoDecodeVP9ProfileInfoKHR’; did you mean ‘VkVideoDecodeAV1ProfileInfoKHR’?
<BtbN>
I've manually launched 20 runners, so the backlog should clear quickly now, will need to check why the automatic mechanism didn't work
<Lynne>
Sean_McG: is CONFIG_VP9_VULKAN_HWACCEL defined in config.h?
<Sean_McG>
this was from a FATE run so that config.h doesn't exist -- I'll have to run it manually
<jianhuaw>
Hi there. Could someone with admin access to the Fate server help review this PR (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20161)? The small VVC fix has been reviewed by Nuomi and Frank, which is needed for 8.0.
mkver has quit [Ping timeout: 245 seconds]
<frankplow>
jianhuaw: Have you sent an email to samples-request@ffmpeg.org?
<jianhuaw>
No. I didn't know about it. Should I send the links to the samples to the samples-request@ffmpeg.org?
<frankplow>
jianhuaw: Yeah exactly, plus the paths you want them to be stored at in the FATE samples dir, and it might also be useful to send a link to the relevant PR
<jianhuaw>
Okay. I'll try. Thanks!
<BtbN>
I've just added them. I guess I as supposed to unzip the files, and add the *.bit files, but rename them to remove the manufacturer part?
<JEEB>
reference suite files are usually kept as-is? at least looking at the ISO/ITU formats
<jianhuaw>
Yes. The manufacturer is removed as convention.
<jianhuaw>
Most of them in fate-suite/vvc-conformance have remove the manufacturer.
<jamrial>
BtbN: could disabling cone also made the cvslog ml not get any new email?
<BtbN>
I don't think so? iirc the cvslog ml is on our server
<BtbN>
as in, the thing that sends to it
<BtbN>
But I have never come into contact with that
<jamrial>
well, something happened because as nicolas reported it hasn't gotten new mails in three days :/
<BtbN>
Hm, that does coincide with the videolan hooks being removed
<wbs>
jamrial: it didn't get mails for pushes to master before that either; it was silent for many days, but then there was some pushes to release branches, which did generate mails there
<wbs>
(possibly that commit syncs for release branches worked, while syncs for master didn't)
<BtbN>
yeah, must have come from videolan then
<BtbN>
but there is ffmpeg-web commits on there, which isn't on videolan
<BtbN>
so we must have it set up as well, and I just need to turn it on. But I have no clue where the mails come from
<wbs>
the mails for ffmpeg-web look slightly different than the ones from ffmpeg as well, so they're probably different scripts on different hosts
<BtbN>
I've just run git config hooks.mailinglist ffmpeg-cvslog@ffmpeg.org on the ffmpeg.git repo
<BtbN>
that's the only difference I see compared to ffmpeg-web. So maybe that's all it takes
<jianhuaw>
Thanks BtbN The fate works. Feel free to merge the PR if you have permission as Michael is going to make the 8.0 branch. frankplow Thanks.
<Yalda>
kasper93: the good news is, based on some initial tests, this demuxer does not seem to have the issue. the bad news is, I don't know why the issue does not manifest. so I'm running with some more tests/analysis first to be truly sure.
<kasper93>
Yalda: have you tried with some of those most problematic examples?
<Yalda>
kasper93: not yet but I will test thoroughly
<Yalda>
I agree it should happen based on what I am reading
<Yalda>
(as in, the issue should happen)
<kasper93>
I think the issue was manifested mostly on discs with 1000+ segments, where audio tracks had overlapping parts and if you don't correct them, the audio went out of sync because the small offset adds up
<Yalda>
appreciate you flagging this, I would have missed it completely
<Yalda>
beside accounting for this issue and seeking still being janky the demuxer is basically done
<kasper93>
Yalda: thanks, I don't have blu-ray drive in my pc anymore, can't really test much. without connecting it
<kasper93>
my current pc case, don't even have bay for 3.5" xd
<Yalda>
no worries. I have an army of drives
<Yalda>
mostly DVD but have some BD in there
<kasper93>
5.25" I should say, bad memory to numbers
sepro has quit [Quit: Bye!]
<Yalda>
it was sad building my last machine. local store only had one case with 5.25" bay, off in the corner, unwanted by anyone (except for me lol)
<JEEB>
I have a buffalo USB3 blu-ray drive for laptop or new stuff usage. and then I have a "UHD friendly" Asus SATA drive for UHD BD usage
<Yalda>
the ASUS/LG drives are good. whatever you do, do not trust them for DVD
<Yalda>
they will return junk data on scratched or bad sectors and simply say "block read ok"
<Yalda>
this is called "Jamless Play" feature. good for playback, bad for archival
<kasper93>
Yalda: you should have powers to add a label, I think for those things you can just add ad ask for forgiveness later if someone goes off that it cannot be like that :)
<beastd>
true, but we should eventually come to strong rules on how to label what. otherwise labels are just free form tags and don't provide much universal utility. or the opposite if developers get confused.
<Yalda>
i have been cloning some ancient IR remotes and using it to help with pronto codes
<beastd>
Yalda: yes, the checking the work is what makes the whole thing not really helpful at all. and that makes the use case to use LLMs for coding / generating programs inefficient.
<beastd>
unless you are yolo and reckless, which seems to be almost our complete industrie as it seems :-/
<Yalda>
beastd: agreed
<Yalda>
i now ask it more along, "make me the regex" or "generate the command to do it myself"
<beastd>
you can get sth out of LLMs, but to me it looks like it is no where near as advertised and portrayed in the press. and i am convinced you have to be very careful.
<Yalda>
they are a tool in the box, like anything else, cannot depend on for everything
<jkqxz>
The encoder also wants them before CBS write can run. I think I moved the derive function out of CBS for that, but splitting it would probably be better.
<jkqxz>
I'll have a look tomorrow. (And also maybe finish the encoder at some point.)
<fjlogger>
[FFmpeg/FFmpeg] Pull request #20187 opened: avcodec/apv_decode: make apv_format_table consistent with the code and check it (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20187) by michaelni
<kasper93>
there are also cbs_vp9_syntax_template.c:374:9: warning: result of comparison of constant -1 with expression of type 'uint8_t' (aka 'unsigned char') is always true
<kasper93>
not sure what you wanted to check there tho
uau has quit [Server closed connection]
ccawley2011 has quit [Read error: Connection reset by peer]
uau has joined #ffmpeg-devel
_whitelogger has joined #ffmpeg-devel
jianhuaw has quit [Quit: Client closed]
<desmond-netint>
I want to add hwaccel video codecs to libavcodec. Is it better to add them in one go or add them piece-meal in separate PRs?