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 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
Everything has quit [Quit: leaving]
cone-173 has quit [Quit: transmission timeout]
derpydoo has joined #ffmpeg-devel
Guest40_ has joined #ffmpeg-devel
markh has quit [Remote host closed the connection]
thilo has quit [Ping timeout: 268 seconds]
thilo has joined #ffmpeg-devel
thilo has quit [Changing host]
thilo has joined #ffmpeg-devel
abdu60 has joined #ffmpeg-devel
markh has joined #ffmpeg-devel
\\Mr_C\\ has quit [Remote host closed the connection]
abdu60 has quit [Ping timeout: 240 seconds]
abdu60 has joined #ffmpeg-devel
derpydoo has quit [Quit: derpydoo]
cone-401 has joined #ffmpeg-devel
<cone-401>
ffmpeg softworkz master:ace9f03a6c0a: avformat/hls: Partially revert "reduce default max reload to 3"
<cone-401>
ffmpeg Michael Niedermayer master:469b7a0ee444: doc/developer: Better {} style rule
<cone-451>
ffmpeg Lynne master:629e8a2425ff: vulkan: add support for AV_PIX_FMT_GRAY32
Guest42 has joined #ffmpeg-devel
Guest42 has quit [Client Quit]
<jamrial>
Lynne: version bump and apichanges entry
<Lynne>
thanks, will post a patch in a minute
<Lynne>
oof, we didn't have apichanges for any recent pixfmts
<Lynne>
should I add patches for those?
SohamK has quit [Quit: Client closed]
SohamK has joined #ffmpeg-devel
<jamrial>
Lynne: yeah
<Lynne>
hunted down the dates when pushed, sent patches for all of them
SohamK has quit [Quit: Client closed]
^Neo has joined #ffmpeg-devel
^Neo has joined #ffmpeg-devel
SohamK has joined #ffmpeg-devel
<haasn>
Lynne: in your patch you say "planar gray and gray"
<haasn>
but should be planar gbr and gray
SohamK63 has joined #ffmpeg-devel
<Lynne>
it says planar gray and float gray?
SohamK has quit [Ping timeout: 240 seconds]
SohamK63 has quit [Client Quit]
Mirarora has quit [Quit: Mirarora encountered a fatal error and needs to close]
abdu82 has joined #ffmpeg-devel
abdu85 has quit [Ping timeout: 240 seconds]
Mirarora has joined #ffmpeg-devel
^Neo has quit [Ping timeout: 245 seconds]
<haasn>
commit title
<haasn>
or well, your email title
<Lynne>
"add entries for new planar gray and float gray pixfmts"?
SohamK has joined #ffmpeg-devel
SohamK has quit [Client Quit]
SohamK has joined #ffmpeg-devel
<haasn>
new planar gray -> new planar gbr
<haasn>
or rather, "new float gray and planar gbr" because the gbrp formats are also float
SohamK has quit [Client Quit]
cone-451 has quit [Quit: transmission timeout]
SohamK has joined #ffmpeg-devel
Traneptora has quit [Quit: Quit]
Traneptora has joined #ffmpeg-devel
Kei_N has joined #ffmpeg-devel
Kei_N_ has quit [Read error: Connection reset by peer]
<haasn>
Lynne: -vf hwupload,format=vulkan,hwdowload seems to leak memory, the first ff_vk_exec is never cleaned up
<haasn>
by any component
<haasn>
if I force a ff_vk_exec_wait(&p->vkctx, exec); at the bottom of vulkan_transfer_frame() the issue disappears
Guest40 has joined #ffmpeg-devel
<haasn>
it seems there's a fundamental issue here: the exec_ctx belonging to a hwfc will only be cleaned up if the hwfc is free'd
<haasn>
but the hwfc is only free'd after ff_vk_exec_pool_free() cleans up the frame refs keeping the hwfc alive
<haasn>
so we have two things referencing each other and no obvious place to break the loop
elvis_a_presley has quit [Quit: smoke-bomb ; grapple-hook]
elvis_a_presley has joined #ffmpeg-devel
<haasn>
I think that the correct solution would be to add some sort of explicit (*sync)() command on the vulkan hwfc which can be used to forcibly flush the internal exec pools
<haasn>
Lynne: I think a wholly different solution would be to have the vk exec ctx not take refs to AVFrames it depends on
<haasn>
but rather ref the underlying vulkan images directly
<haasn>
so that the AVFrame can still be freed, thus triggering a free of the hwfc
<haasn>
but that seems much more invasive and error prone
Sean_McG has joined #ffmpeg-devel
* Sean_McG
waves
<Sean_McG>
jamrial: I have to be honest I have not been paying the closest attention to the mailing list, did you ever get any responses here or elsewhere regarding your library version bump patchset? If yes I'll need to hold off on the work I am doing for Trac #11120.
<jamrial>
i got at least one request to wait a bit before pushing it
<Sean_McG>
OK.
<Sean_McG>
michaelni: the new CVE shellscript currently uses wget but perhaps it should probe if curl is available incase wget is not... I don't remember if it is shipped as part of Debian base install. VLC does this to fetch their contrib set, perhaps it could be borrowed/adapted from there?
iive has joined #ffmpeg-devel
<Lynne>
haasn: ah, that's a tricky issue to fix
<haasn>
to be honest I only noticed it because the nvidia vulkan drivers like to randomly segfault when you exit() without destroying the VkDevice
<Lynne>
I think before introducing a new API, I'd like to try reffing the buffer ref of the AVVkFrame rather than the AVFrame as a whole
<haasn>
but hanging on to a VKDevice ref indefinitely is always a bad idea
<haasn>
since you don't know how many times a user is going to init/uninit a filter graph per process
<Lynne>
yeah, weird that the validation layers don't complain
<haasn>
oh, we could try that
<Lynne>
I'm not a fan of the way we store refs either, I'd like to try using the refstruct mechanism (if its usable outside of lavc) to avoid even needing to malloc memory per-transfer
<haasn>
could you write up a quick patch for it? currently busy with sth else
<Lynne>
yup, sure
abdu32 has quit [Ping timeout: 240 seconds]
mkver has joined #ffmpeg-devel
minahermina has joined #ffmpeg-devel
<minahermina>
Hey there! I hope you're doing well!
odrling has quit [Remote host closed the connection]
odrling has joined #ffmpeg-devel
SohamK11 has joined #ffmpeg-devel
minahermina has quit [Quit: Client closed]
SohamK11 has quit [Quit: Client closed]
<Lynne>
mkver: any tips on how to use the refstruc api?
<mkver>
For what?
<Lynne>
I'd like to store AVVkFrame hardware frame structs (AVBufferRefs) from AVFrame->data
<Lynne>
since reffing the avframe as a whole refs hw_frames_ctx, which makes it a circular dep
<Lynne>
and I'd like to avoid mallocing every single ref, since it happens per-frame
<mkver>
So you want to create a ref for every AVBufferRef set in AVFrame.buf. Do you intend to create multiple refs from these refs or just one?
<Lynne>
there's only ever a single avbufferref per avframe I need
<Lynne>
nope, just one, once
<Lynne>
I need to keep the data ref'd just so it doesn't go poof whilst its being processed on the GPU
<mkver>
I don't see how the refstruct API could help you here. After all, AVFrame.buf uses the AVBuffer API, not the RefStruct API.
<Lynne>
I could store pointers to the atomic int counters of the buffer and decrement them atomically once I no longer need the data to be around, and it wouldn't be completely haram since it's libavutil code doing it, but it feels naughty
abdu66 has quit [Quit: Client closed]
abdu66 has joined #ffmpeg-devel
<mkver>
Everything stays in libavutil? Then you could treat sizeof(AVBufferRef) as known and add an ff_buffer_unref(AVBufferRef *ref) that is like av_buffer_unref(), but it doesn't free the AVBufferRef (because it is not separately allocated).
<mkver>
Don't know if everything would be ok with this though.
<mkver>
I regard it as bullshit that sizeof(AVBufferRef) is not public.
<Lynne>
oh, okay, I'll do that then, thanks
ramiro has quit [Ping timeout: 268 seconds]
<Lynne>
ah... it's not all within lavu
<Lynne>
the code is templated between lavu, lavc and lavfi
<Lynne>
did we support using libraries with different versions? pretty sure we didn't
<Lynne>
if not, or if it's too ugly, I can make the code lavu-only and internal, since I only need this specific handling in lavu where it's problematic
ramiro has joined #ffmpeg-devel
SohamK55 has joined #ffmpeg-devel
Guest40 has joined #ffmpeg-devel
abdu66 has quit [Ping timeout: 240 seconds]
c1480_ has quit [Quit: WeeChat 4.1.1]
minahermina has joined #ffmpeg-devel
cone-775 has quit [Quit: transmission timeout]
mina has quit [Quit: Lost terminal]
sudden has quit [Ping timeout: 252 seconds]
minahermina has quit [Ping timeout: 240 seconds]
sudden has joined #ffmpeg-devel
SohamK55 has quit [Quit: Client closed]
<fflogger>
[editedticket] Legimet: Ticket #11432 ([undetermined] Empty output file after conversion if source is 5.1 audio (ac3>opus)) updated https://trac.ffmpeg.org/ticket/11432#comment:2
\\Mr_C\\ has joined #ffmpeg-devel
<Lynne>
haasn: patch sent
<Lynne>
ended up being rather simple
cone-868 has joined #ffmpeg-devel
<cone-868>
ffmpeg Adam Lackorzynski master:76b181001778: libswscale/arm/swscale_unscaled: Fix function prototype