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
DauntlessOne4 has joined #ffmpeg-devel
System_Error has quit [Ping timeout: 264 seconds]
System_Error has joined #ffmpeg-devel
c1480_ has quit [Read error: Connection reset by peer]
c1480 has joined #ffmpeg-devel
minimal has quit [Quit: Leaving]
c1480 has quit [Read error: Connection reset by peer]
c1480_ has joined #ffmpeg-devel
Gramner has joined #ffmpeg-devel
thilo has quit [Ping timeout: 276 seconds]
thilo has joined #ffmpeg-devel
c1480_ has quit [Read error: Connection reset by peer]
c1480 has joined #ffmpeg-devel
System_Error has quit [Ping timeout: 264 seconds]
delewis has joined #ffmpeg-devel
jamrial has quit []
Martchus_ has joined #ffmpeg-devel
Martchus has quit [Ping timeout: 260 seconds]
philipl has quit [Quit: leaving]
philipl has joined #ffmpeg-devel
HarshK23 has quit [Quit: Connection closed for inactivity]
System_Error has joined #ffmpeg-devel
Teukka has quit [Read error: Connection reset by peer]
Teukka has joined #ffmpeg-devel
Teukka has quit [Changing host]
Teukka has joined #ffmpeg-devel
c1480 has quit [Read error: Connection reset by peer]
c1480 has joined #ffmpeg-devel
quietvoid has quit [Ping timeout: 248 seconds]
HarshK23 has joined #ffmpeg-devel
quietvoid has joined #ffmpeg-devel
bhaskar has quit [Ping timeout: 252 seconds]
bhaskar has joined #ffmpeg-devel
ngaullier has joined #ffmpeg-devel
eslam has joined #ffmpeg-devel
rvalue has quit [Read error: Connection reset by peer]
rvalue has joined #ffmpeg-devel
quietvoid has quit [Ping timeout: 245 seconds]
^Neo has quit [Ping timeout: 244 seconds]
cone-196 has joined #ffmpeg-devel
<cone-196>
ffmpeg Lynne master:779763181f85: bwdif_vulkan: convert to storage images
<cone-196>
ffmpeg Lynne master:389fb36f924d: avgblur_vulkan: port to imageLoad()
<cone-196>
ffmpeg Lynne master:1029f51285ec: vulkan: fix crash in ff_vk_shader_free
<cone-196>
ffmpeg Lynne master:042ed96d0a2d: vulkan_filter: use GENERAL image layout when no sampler is given
<cone-196>
ffmpeg Lynne master:a5356396200f: nlmeans_vulkan: switch to imageLoad()
<cone-196>
ffmpeg Lynne master:4d3e540fa4a6: flip_vulkan: port to imageLoad()
<cone-196>
ffmpeg Lynne master:b02f9157b8c8: gblur_vulkan: port to imageLoad()
<cone-196>
ffmpeg Lynne master:379cfd185556: transpose_vulkan: port to imageLoad()
<cone-196>
ffmpeg Lynne master:2b8d38cbc198: blend_vulkan: port to imageLoad()
<cone-196>
ffmpeg Lynne master:08e37fa0820c: overlay_vulkan: port to imageLoad()
<fflogger>
[editedticket] Gyan: Ticket #11469 ([ffmpeg] ffmpeg_demux: readrate plays "catch up" if output is blocked, then later resumed) updated https://trac.ffmpeg.org/ticket/11469#comment:4
odrling has quit [Remote host closed the connection]
odrling has joined #ffmpeg-devel
odrling has quit [Remote host closed the connection]
odrling has joined #ffmpeg-devel
Guest86 has joined #ffmpeg-devel
Guest86 has quit [Client Quit]
quietvoid has joined #ffmpeg-devel
quietvoid has quit [Ping timeout: 244 seconds]
System_Error has quit [Ping timeout: 264 seconds]
System_Error has joined #ffmpeg-devel
tortoise has quit [Ping timeout: 246 seconds]
quietvoid has joined #ffmpeg-devel
System_Error has quit [Remote host closed the connection]
tortoise has joined #ffmpeg-devel
<Lynne>
I think at least a few of the PRs in the test instance are not tests anymore
<Lynne>
BtbN: how would the instances be synchronized during the period where PRs are accepted?
<Lynne>
whoever merges anything from either pushes to the other immediately?
jkkm has quit [Ping timeout: 246 seconds]
quietvoid has quit [Ping timeout: 248 seconds]
quietvoid has joined #ffmpeg-devel
jkkm has joined #ffmpeg-devel
<haasn>
is it sane to call av_set_options_string on a stack object?
<haasn>
to parse options passed as a const char *
eslam has quit [Ping timeout: 260 seconds]
eslam has joined #ffmpeg-devel
tanay-man has quit [Remote host closed the connection]
<BtbN>
Lynne: push it to the actual repo, and mark the PR as merged
<BtbN>
that's the only somewhat sane way
cone-196 has quit [Quit: transmission timeout]
<Lynne>
I think github marks merged PRs automatically, not sure about forgejo
<Lynne>
but yeah, that works
<Lynne>
did you figure out the registration issues?
<Lynne>
also, I haven't seen any new members on this page in a while https://code.ffmpeg.org/explore/users and afaik at least a few have registered that aren't on the list
<BtbN>
There literally is a button on the PRs to "Merge Manually"
<BtbN>
which just marks it as merged, but does nothing otherwise
<BtbN>
For Members, I need to manually add them, and I haven't bothered so far
abdu has joined #ffmpeg-devel
<Lynne>
cool
<haasn>
Lynne: what's the best way to handle writing vulkan equivalents of existing filters without having to duplicate all options and wrapping code?
<Lynne>
if you think the instance is ready, you could submit a patch to merge the tests into the repo and a note on the web page that PRs are accepted for now
<haasn>
do we have a precedent for this that is not just copy/pasting the filter
System_Error has quit [Remote host closed the connection]
System_Error has joined #ffmpeg-devel
<haasn>
maybe setting the original filter context as a subclass?
abdu has quit [Quit: Client closed]
abdu has joined #ffmpeg-devel
Tanay has joined #ffmpeg-devel
Compn has quit [Ping timeout: 248 seconds]
<haasn>
Lynne: unrelated, do we have a helper for vulkan filters with inputs but no outputs? (for example, a scene change detection filter)
abdu has quit [Quit: Client closed]
abdu has joined #ffmpeg-devel
Compn has joined #ffmpeg-devel
Compn has quit [Read error: Connection reset by peer]
<haasn>
I guess I can just modify ff_vk_filter_process_simple to make out_f optional
<Lynne>
we used to have a helper for inplace vulkan filters (e.g. where the input frame was also the output), which got removed due to it being unused and incompatible
Compn has joined #ffmpeg-devel
<Lynne>
but yeah, that works
<Lynne>
in_f is already optional, no reason not to extend that to the output frame too
<Lynne>
are you writing a scene detect filter?
<haasn>
yes
<haasn>
actually, if I want to attach an SSBO for gathering output
<JEEB>
scxvid :3
<haasn>
that means I need to write my own dispatch code in any case
<Lynne>
feel free to refer to vf_nlmeans_vulkan.c, it does its own dispatching and buffer alloc
<Lynne>
same with the ffv1 encoder, buffer pools are great
<haasn>
interesting that you use a separate buffer for each weight array instead of one buffer for all
System_Error has quit [Remote host closed the connection]
<haasn>
and yeah I'll use that as a reference, thanks
<Lynne>
which buffer?
<haasn>
in nlmeans, all your float weights etc are separate storage buffers
<Lynne>
yeah, you basically compute thounsands of integral images with different offsets, then you analyze each and put it into a single shared weights buffer
<Lynne>
we could do multiple integral images per invoc, but the issue is parallelization
<Lynne>
each barrier completely and utterly destroys performance, so we waste gigabytes on integral image buffers just so we can avoid barriers by not sharing as many buffers
<haasn>
also unrelated but how hard would it be to make planeextract work on vulkan frames? I guess i's not possible atm because you can't put a ref on individual planes of a vulkan frame
eslam has quit [Ping timeout: 252 seconds]
System_Error has joined #ffmpeg-devel
<haasn>
it would require tearing apart AvVkFrame and moving all per-plane stuff into a per-plane allocation struct that is then attached to the planes as buf[] references
<Lynne>
not that hard, you can just ref the entire frame, modify it by aliasing a wanted plane, and outputting it
<Lynne>
btw keep in mind the nlmeans filter in ffmpeg is kinda bogus for anything but monochrome images, since its performed on each plane individually, and according to prunedtree, plane separable denoisers have been out of date for 20 years now, even the nlmeans paper defined the algorithm on top of RGB
<haasn>
not sure what you mean, wouldn't that create data hazards with other users of the frame?
<Lynne>
it still works great though
<Lynne>
no, you can mark buffers as read-only
<Lynne>
in general, if you hold a ref, you can assume the frame will not change
<Lynne>
and in the case of vulkan, where semaphores are accessed and layout fields are changed, we have frame locking functions
<haasn>
wait, vf_planeextract doesn't even ref planes
<haasn>
it always does a memcpy
<haasn>
in that case nvm my question, we can obviously just do the same in vulkan land
Tanay has quit [Read error: Connection reset by peer]
tanay-man has joined #ffmpeg-devel
abdu78 has joined #ffmpeg-devel
abdu has quit [Ping timeout: 240 seconds]
ccawley2011 has joined #ffmpeg-devel
ccawley2011 has quit [Ping timeout: 268 seconds]
ccawley2011 has joined #ffmpeg-devel
abdu71 has joined #ffmpeg-devel
abdu78 has quit [Ping timeout: 240 seconds]
abdu41 has joined #ffmpeg-devel
abdu71 has quit [Ping timeout: 240 seconds]
abdu38 has joined #ffmpeg-devel
abdu41 has quit [Ping timeout: 240 seconds]
cone-951 has joined #ffmpeg-devel
<cone-951>
ffmpeg Shreesh Adiga master:e18f87ed9f9f: swscale/x86/rgb2rgb: add AVX512ICL version of uyvytoyuv422
abdu38 has quit [Quit: Client closed]
ColdLeader has joined #ffmpeg-devel
<ColdLeader>
16 bit sonarc decoding completed, mission accomplished!
minimal has joined #ffmpeg-devel
Everything has joined #ffmpeg-devel
<ePirat>
haasn, had the same issue for xfade, where I even had to copy stuff like the activate function
<ePirat>
and then manually keeping them in sync
<ePirat>
less than ideal
<ePirat>
I just realised the opencl version of the filter still has the old activate function…
<ColdLeader>
the ffmpeg normal xfade incarnation usage is not going to work reliably as it consumes huge amount of memory for each xfade inistance in filtergraph, the ffmpeg is dead
ccawley2011_ has joined #ffmpeg-devel
ccawley2011__ has joined #ffmpeg-devel
<ePirat>
haasn, imho having graphdump option for ffmpeg cli itself is more useful than the separate tool
ccawley2011 has quit [Ping timeout: 260 seconds]
<ePirat>
makes it easier to see all the inserted filters without having to stare at the verbose log output
ccawley2011 has joined #ffmpeg-devel
ccawley2011_ has quit [Ping timeout: 272 seconds]
<ColdLeader>
think about lib usage, not only about cli limited tool
ccawley2011__ has quit [Ping timeout: 272 seconds]
ColdLeader has quit [Quit: Client closed]
ColdLeader has joined #ffmpeg-devel
<beastd>
ColdLeader: best would be both lib and cli with minimal effort to create the cli via lib apis
<JEEB>
the whole dumping was already an API, I think the only thing that isn't integrated are the additional formats posted on the ML ages ago
<JEEB>
but yea, some dumping formats are more useful as random logs into stderr in ffmpeg, and some are more useful as proper stdout output from a separate tool
<JEEB>
although I think the ML patch might have added output to a file
ccawley2011 has quit [Ping timeout: 252 seconds]
ColdLeader has quit [Quit: Client closed]
ColdLeader has joined #ffmpeg-devel
abdu has joined #ffmpeg-devel
<devinheitmuell-1>
@JEEB At least with the patch I did, I defined an environment variable named export FFMPEG_DEBUG_DUMP_DOT_DIR=/tmp, which would then write files to that directory. This syntax is essentially borrowed from gStreamer. But yeah, it’s definitely useful to have the info written directly to a file rather than being made available via av_log().
<beastd>
+1
eslam has joined #ffmpeg-devel
<ColdLeader>
that is very bad for lib usage
<ColdLeader>
libs do not need file output
<ColdLeader>
libs only need actual data
<devinheitmuell-1>
Yeah, I was not suggesting we actually use an environment variable. Just that it’s nice to have some way to put out structured data that isn’t to the logfile.
<jamrial>
output to av_log() can't be versioned and can't be expected to be parseable
<devinheitmuell-1>
The fact that most filters for examples just log their stats with no way to propagate them back to the calling application drives me nuts.
<jamrial>
i don't know how avfilter could do to write to some output. lavf does it with aviocontext
Guest98 has joined #ffmpeg-devel
<ColdLeader>
so you want structured data ala xml, ffmpeg libs are not for that
Guest17_new has joined #ffmpeg-devel
<JEEB>
at one point I thought it could be attached to the filter output frames
<JEEB>
so you have your side data of type X, which you can then write out as whatever, in a similar way to ffprobe if you want structured textual output
<ColdLeader>
filter have stats via log and/or exported options and/or file and/or frame metadata
<JEEB>
yes, logs are widely (ab)used
<JEEB>
then yes, there are some pre-side data metadata things
<ColdLeader>
ideally you want events
<ColdLeader>
buy ffmpeg is so 90ish product
<ColdLeader>
*but
<devinheitmuell-1>
Yeah, I would be perfectly happy if you could register an event handler which gets called by filters with feedback information.
<JEEB>
yea, if you don't need it to be synchronized at all with the actual frame receipt then something like event callbacks
<JEEB>
since things that can be in synch with received stuff can just be attached to the AVFrame as structs
<ColdLeader>
do not abuse AVFrame for kitchen-sink
<JEEB>
I am not abusing anyone
<JEEB>
also if you did not notice I actually mentioned event callbacks
<JEEB>
but you conveniently ignore that I actually take differnet things into account
<ColdLeader>
AVFrame is not a person
abdu26 has joined #ffmpeg-devel
Guest17_new has quit [Quit: Client closed]
abdu has quit [Ping timeout: 240 seconds]
MyNetAz has quit [Remote host closed the connection]
ngaullier has quit [Remote host closed the connection]
MyNetAz has joined #ffmpeg-devel
ColdLeader has quit [Quit: Client closed]
abdu26 has quit [Quit: Client closed]
abdu26 has joined #ffmpeg-devel
ccawley2011 has joined #ffmpeg-devel
abdu14 has joined #ffmpeg-devel
abdu39 has joined #ffmpeg-devel
delewis_ has joined #ffmpeg-devel
delewis has quit [Remote host closed the connection]
abdu26 has quit [Ping timeout: 240 seconds]
delewis has joined #ffmpeg-devel
delewis_ has quit [Remote host closed the connection]
abdu14 has quit [Ping timeout: 240 seconds]
ccawley2011 has quit [Ping timeout: 252 seconds]
cone-951 has quit [Quit: transmission timeout]
ccawley2011 has joined #ffmpeg-devel
ccawley2011_ has joined #ffmpeg-devel
ccawley2011 has quit [Ping timeout: 268 seconds]
System_Error has quit [Remote host closed the connection]
System_Error has joined #ffmpeg-devel
ccawley2011_ has quit [Ping timeout: 245 seconds]
System_Error has quit [Remote host closed the connection]
System_Error has joined #ffmpeg-devel
Guest59 has joined #ffmpeg-devel
Guest59 has quit [Client Quit]
Guest59 has joined #ffmpeg-devel
Guest59 has quit [Client Quit]
ccawley2011_ has joined #ffmpeg-devel
Guest59 has joined #ffmpeg-devel
Guest59 has quit [Client Quit]
MyNetAz has quit [Remote host closed the connection]
MyNetAz has joined #ffmpeg-devel
Everything has quit [Quit: leaving]
abdu39 has quit [Quit: Client closed]
abdu39 has joined #ffmpeg-devel
<BtbN>
hm, fun. I'm locked out from my own mail server, cause bots tried to many wrong passwords, and faillock activated
<jamrial>
BtbN: shouldn't that just block the offending ips after a certain amount of tries and not just shut down everything?
<Lynne>
I wish I could put smtp and imap under wireguard, but clients hate it
<BtbN>
jamrial: no, it's a PAM module that locks out PAM logins after too many failures
<BtbN>
luckily SSH via pubkey does not go via PAM, or I'd be fully locked out
System_Error has quit [Remote host closed the connection]
System_Error has joined #ffmpeg-devel
abdu39 has quit [Quit: Client closed]
abdu39 has joined #ffmpeg-devel
Traneptora has joined #ffmpeg-devel
<ePirat>
jamrial, why did you port sbcenc to get_supported_config?
<jamrial>
ePirat: to remove a deprecated warning
<ePirat>
IMO we should not do that, I discussed this with haasn before
<ePirat>
those fields should be moved, but obviously we can not do that without a major bump
<ePirat>
for which we need to declare them deprecated before which caused the warnings you see
<ePirat>
I tried to fix this by adding the fields in the private struct and copy them to the public ones for the time being but that does not work either because we declare FFCodec as const…