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 6.1 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
derpydoo has quit [Ping timeout: 240 seconds]
iive has quit [Quit: They came for me...]
jamrial has quit [Read error: Connection reset by peer]
<Lynne>
I think VLC may be misusing the API and not setting avctx->hw_frames_ctx
jkkm has quit [Server closed connection]
jkkm has joined #ffmpeg-devel
Manouchehri has quit [Server closed connection]
Manouchehri has joined #ffmpeg-devel
mindfreeze has quit [Server closed connection]
mindfreeze has joined #ffmpeg-devel
Flat_ has joined #ffmpeg-devel
Flat has quit [Ping timeout: 255 seconds]
AbleBacon has quit [Read error: Connection reset by peer]
BBB has quit [Server closed connection]
BBB has joined #ffmpeg-devel
Sebastinas has quit [Server closed connection]
Sebastinas has joined #ffmpeg-devel
daddesio has quit [Server closed connection]
daddesio has joined #ffmpeg-devel
Arsen has quit [Server closed connection]
Arsen has joined #ffmpeg-devel
kurosu has joined #ffmpeg-devel
Guest7535 has quit [Server closed connection]
root has joined #ffmpeg-devel
root is now known as Guest2064
tmatth has quit [Server closed connection]
tmatth has joined #ffmpeg-devel
bossjones has quit [Server closed connection]
bossjones has joined #ffmpeg-devel
<JEEB>
kierank: can you remind me how to connect to the M1 machine in a pm? I think I had my key in there from earlier years?
emersion has quit [Server closed connection]
emersion has joined #ffmpeg-devel
av500 has quit [Remote host closed the connection]
av500 has joined #ffmpeg-devel
<wbs>
JEEB: I don't see your username on the box, but I'm sure kierank can add you
<JEEB>
ok, then it got reset at some point since I recall trying to poke at VP9 in VT there :)
<JEEB>
(or I completely misremember)
<wbs>
I don't think that machine has been reset at any point, you might be thinking of some other machine somewhere else
<JEEB>
so the second case :)
linjie has quit [Server closed connection]
linjie has joined #ffmpeg-devel
<nevcairiel>
Lynne: the API does not demand that hw_frames_ctx is set for all hwaccels, its called ad-hoc mode and is still supported for some of the older hwaccels, namely dxva2 and vaapi
<nevcairiel>
(and the old d3d11va interface, which noone should be using anymore, but i guess vlc might)
Krowl has joined #ffmpeg-devel
<Lynne>
thanks, I'll send a patch then
<jkqxz>
The user doesn't need to set hw_frames_ctx in device mode either. What mode is being used in the case which goes wrong?
<Lynne>
device mode?
<nevcairiel>
in device mode avctx derives a frames ctx from the device
<nevcairiel>
only in ad-hoc mode you have neither
<nevcairiel>
i think
<Lynne>
right, that mode where frame_params is called by decode.c
<nevcairiel>
but the fix seems simple, just check that a frames ctx is set when free_frame_priv is set, error otherwise, and if not just allocate a plain buffer for the frame data
<nevcairiel>
only vulkan uses that callback i think, and it wont work without a frame ctx
<elenril>
dear past self, thank you for documenting it at least
nto has quit [Server closed connection]
nto has joined #ffmpeg-devel
<courmisch>
alsa.conf eww
<elenril>
that is modprobe
<courmisch>
well I guess that's actually modprobe conf
<courmisch>
so what happens if you make a new kernel with HDA statically linked
<elenril>
and I'll take alsa conf over pulse conf any day
<courmisch>
pulse is so 2000s
<elenril>
it baffles the mind how something can be even less documented than alsa
<elenril>
pipewire documentation was next to useless when I looked at it last
<elenril>
granted, there was more of it than for pulse
<elenril>
but it started with assuming that you already know how everything works and fits together
<courmisch>
ALSA conf is incomprehensible
<courmisch>
you need to be a ALSAlib dev to know how it works
galad has quit [Remote host closed the connection]
<courmisch>
and the UCM stuff is trying to reimplement PA badly
galad has joined #ffmpeg-devel
<courmisch>
well PA is dead anyway
<elenril>
the best time to start using it
<courmisch>
maintenance-wise it's been all downhill since Lennart stopped caring
<elenril>
(I actually did on one of the machines I'm managing)
<courmisch>
I don't even? are you still using ALSA dmix??
<elenril>
yes
<elenril>
it works well enough for my purposes
Traneptora has quit [Quit: Quit]
<courmisch>
yikes
<elenril>
it works well enough for my purposes
<courmisch>
I recall setting it up literally 20 years ago
<courmisch>
I should check if it's still used on my T40
<courmisch>
running oldoldoldoldstable or somethint
<elenril>
I see no reason to study the unreadable pipewire documentation to fix something that isn't broken
Coinflipper has quit [Server closed connection]
<courmisch>
ALSA device management is a bad joke
<courmisch>
I want a list of output ports, not channel layouts
Coinflipper has joined #ffmpeg-devel
<elenril>
all the options suck as far as I'm concerned
<elenril>
didn't try JACK I guess
galad has quit [Quit: ZNC 1.9.x-git-230-b1f873df-frankenznc - https://znc.in]
galad has joined #ffmpeg-devel
dellas has quit [Remote host closed the connection]
<courmisch>
elenril: MMdevice
courmisch has left #ffmpeg-devel [#ffmpeg-devel]
courmisch has joined #ffmpeg-devel
<elenril>
...he said and quit in shame
<courmisch>
it could be worse. there's sndio
<courmisch>
multimedia dev's worst nightmare
<elenril>
is it?
<elenril>
we're talking about the user's experience though
<elenril>
I hear PA has quite decent API documentation
<courmisch>
the API is crap
<courmisch>
the functionality is crap
<courmisch>
Yeah PA doxygen is good. the api is also surprisingly sane
<courmisch>
anybody else had their FFmpeg patches sent to by a Kuaishou.com email?
<courmisch>
apparently it's a major thing in China but never heard thereof
<elenril>
"sent to by"?
<courmisch>
sent to thyself
<elenril>
bounced?
<courmisch>
you know ,when somebody git-send-email your patches and doesn't remove you fromCc
<elenril>
ah, no
lexano has joined #ffmpeg-devel
dellas has joined #ffmpeg-devel
chainik has quit [Server closed connection]
chainik has joined #ffmpeg-devel
<kurosu>
Ouf, that was quite the dissonance fix for me
<courmisch>
I'd seen it before for vendor kernel fix backports on "my" subsystem
<courmisch>
but not on a meaningless feature patch for ffmpeg
<kurosu>
kuaishou is kind of the next tiktok, but I never noticed it was the same as something that kind of become of a bigwig in MPEG, eg large contributor to MPEG. kwai the name is
<kurosu>
also tencent-backed apparently, so...
<courmisch>
I didn't catch the meaning of "dissonance fix " here
<courmisch>
and your message got cropped at the name is
<kurosu>
The name is kwai
<kurosu>
I didn't relate that company to that social app, and Tencent
<kurosu>
much like Bytedance contributed a lot to VVC
<courmisch>
this was an AAC RVV patch though :shrug:
<kurosu>
Yeah, sorry, off-topic and unrelated - I was reacting to "it's a major thing in China but never heard thereof" only
nitroxis has quit [Server closed connection]
nitroxis has joined #ffmpeg-devel
AbleBacon has joined #ffmpeg-devel
ccawley2011 has quit [Read error: Connection reset by peer]
ccawley2011 has joined #ffmpeg-devel
durandal_1707 has joined #ffmpeg-devel
<ePirat>
elenril, regarding AVDictionary, using the same key for multiple entries, while certainly makes sense, would be tricky given that insertion can be case-insensitive per insertion, not for the whole dict, same for lookup. So you could only group actually identical keys, which I guess might be even more confusing than just not doing it…
Mikhail___AMD has joined #ffmpeg-devel
Mikhail_AMD has quit [Ping timeout: 246 seconds]
Gramner has quit [Server closed connection]
Gramner has joined #ffmpeg-devel
<quietvoid>
jamrial: about #10541, x265 supports signaling metadata on keyframes/changes only fwiw
<jamrial>
quietvoid: itu-t t-35 metadata is per frame, right? it's not meant to remain active until cancelled, like other sei types
<courmisch>
what is this?! 23hours after I merge something, nobody else has merged anything
<courmisch>
those lazy opensoars devs
<quietvoid>
yes i'm not sure if the method is compliant but it's a thing
<jamrial>
courmisch: i'm watching the Nintendo direct
<jamrial>
will push something once it's over
<quietvoid>
most retail hdr10+ files always have metadata per frame all the time
<typological>
hey! I'm new to ffmpeg's codebase and I'm trying to tackle https://trac.ffmpeg.org/ticket/10659 , pretty sure that patch I sent is bad, but can someone shed some light into what should I pay attention to?
<courmisch>
jamrial: how do you dare refuse to work for free
cone-184 has joined #ffmpeg-devel
<cone-184>
ffmpeg Rémi Denis-Courmont master:20e6195c5410: checkasm: test the noise case of sbrdsp.hf_apply_noise
<cone-184>
ffmpeg Rémi Denis-Courmont master:c536e9220702: lavc/sbrdsp: R-V V hf_apply_noise functions
<courmisch>
what next? unionisation of FFmpeg developers?
CoreX has quit [Server closed connection]
<elenril>
grmbl, right as i was about to push
CoreX has joined #ffmpeg-devel
<courmisch>
I'm going to tell Elon that he needs to stop using FFmpeg
<jamrial>
elenril: better than a race, like it happened to me a couple times. the patches get pushed, but the branch is not ff'd, so the commits are left dangling until garbage collection is run
<courmisch>
elenril: you had 4 minutes to toast me, and you failed?
<cone-184>
ffmpeg Anton Khirnov master:23de85d1ec17: tests/fate/ffmpeg: replace deprecated -vbsf with -bsf:v
<cone-184>
ffmpeg Anton Khirnov master:a8d9d6b08d63: tests/fate: replace deprecated -vsync with -fps_mode
<cone-184>
ffmpeg Anton Khirnov master:436b972fc82e: doc/ffmpeg: expand -bsf documentation
<cone-184>
ffmpeg Anton Khirnov master:de85815bfaa3: lavf/mux: do not apply max_interleave_delta to subtitles
<cone-184>
ffmpeg Paul B Mahol master:7282137f48c3: lavfi/af_amix: make sure the output does not depend on input ordering
<cone-184>
ffmpeg Anton Khirnov master:6dbde68cb546: lavc/8bps: fix exporting palette after 63767b79a570404628b2521b83104108b7b6884c
<cone-184>
ffmpeg Anton Khirnov master:7c97a0c63f89: fftools/ffmpeg: move a few inline function into a new header
<cone-184>
ffmpeg Anton Khirnov master:4f7b91a6980d: fftools/thread_queue: do not return elements for receive-finished streams
<cone-184>
ffmpeg Anton Khirnov master:87016e031ffa: fftools/thread_queue: count receive-finished streams as finished
<cone-184>
ffmpeg Anton Khirnov master:889a022cceea: fftools/ffmpeg: rework keeping track of file duration for -stream_loop
<elenril>
courmisch: proper procedure > being a 1337 dude on the internet
<courmisch>
elenril: ridiculing a courmisch > proper procedure, if you ask me
<elenril>
that is why you're not a TC members
<cone-184>
ffmpeg Zhao Zhili master:a1a6a328f0f5: fftools/ffplay: add hwaccel decoding support
<elenril>
s/s$//
<elenril>
!!!
<elenril>
got tired of waiting for review, huh?
<courmisch>
elenril: I don't think becoming a TC member would help me ridicule courmisches very much
<durandal_1707>
TC members are fflabes employees
<elenril>
durandal_1707: I am not an fflabs employee though
<courmisch>
and besides, I can make courmisches ridicule just fine without being a TC member
<elenril>
courmisch: are you trying to bait people on the internet into humiliating you publicly?
<elenril>
I mean whatever rocks your boat, but I wonder if this is the best place
<durandal_1707>
ffmpeg is hummiliiated every day on ml
<courmisch>
elenril: you may be extrapolating a little bit too much
<cone-184>
ffmpeg James Almer master:b360c917523e: avcodec/codecpar: mention how to allocate coded_side_data
ocrete has quit [Server closed connection]
<courmisch>
I have heard it said that FFmpeg is no fun. Well, soon it will be twice as entertaining as it ever was, and half as boring too! That's four times the amusement for everyone!
ocrete has joined #ffmpeg-devel
<courmisch>
So voute Courmisch for the future that you deserve!
<courmisch>
vote*
<cone-184>
ffmpeg Cosmin Stejerean via ffmpeg-devel master:575efc040686: tools/general_assembly.pl - add options to print names, emails or both
<elenril>
grmbl
<elenril>
mistakes=made
<JEEB>
such is life, and fine as long as they're noticed so they can be possibly learned from
HarshK23 has quit [Quit: Connection closed for inactivity]
<kierank>
elenril: wow h264dec changes
<j-b>
elenril: git push -f?
<j-b>
since it's top-of-tree?
<kierank>
18:24:45 <durandal_1707> ffmpeg is hummiliiated every day on ml
Krowl has quit [Read error: Connection reset by peer]
<ePirat>
elenril, do you have any idea of the "worst case" item count for avdictionary? it seems mostly used for options so I would guess usually it would rarely exceed 100 items…
<elenril>
yeah, mostly it should
<courmisch>
inb4 hash table vs AST
<courmisch>
Round 1
<elenril>
but it'd be nice to not have it perform too horribly for large numbers of entries
<courmisch>
-- FIGHT! --
<elenril>
why would it need to be an AS T specifically
<ePirat>
I think the only sensible optimization might have it sorted and do binary search then?
<courmisch>
elenril: red-black tree, whatever rocks your boat
Mikhail___AMD has quit [Read error: Connection reset by peer]
zwayn has joined #ffmpeg-devel
odrling has quit [Server closed connection]
odrling has joined #ffmpeg-devel
lexano has quit [Ping timeout: 268 seconds]
<ePirat>
why on earth are we discussing if votes should be public now…
<ePirat>
it makes absolutely no sense to even bring that up now
<courmisch>
ePirat: because NG can not miss an opportunity to contradict
TheSashmo has quit [Read error: Connection reset by peer]
TheSashmo has joined #ffmpeg-devel
ccawley2011 has quit [Read error: Connection reset by peer]
<BBB>
I really think some people need to chime in with a +1 here if they think a particular discussion should end on the ML
<BBB>
until we have a CC where the CC members have ban/block access on the mailinglist and a protocol where they are permitted to use it, this nonsense will go on forever
<BBB>
this is your project, folks. make it what you'd like it to be
Daemon404 has joined #ffmpeg-devel
<durandal_1707>
you can not censor NG on ML
<j-b>
elenril: it's better than something ugly like that, sorry.
lexano has joined #ffmpeg-devel
<j-b>
Advocating against voting privately? wow
<Daemon404>
ng already said he thinks democracy is wrong for ffmpeg over a month ag
<BtbN>
I feel like anything would be wrong that dares disagree with him
balling has joined #ffmpeg-devel
Daemon404 has quit [Server closed connection]
Daemon404 has joined #ffmpeg-devel
sm2n has quit [Server closed connection]
sm2n has joined #ffmpeg-devel
balling79 has joined #ffmpeg-devel
balling79 has quit [Client Quit]
HarshK23 has joined #ffmpeg-devel
balling is now known as YourNick
YourNick is now known as balling
balling has quit [Quit: Quit]
balling has joined #ffmpeg-devel
<ePirat>
BBB +1
<BBB>
I meant on the ML ;)
<ePirat>
yes
<ePirat>
:D
<ePirat>
I guess ML takes a while to do its thing
<BBB>
I think many of us want some of the ML discussions to stop. you can make that happen
<BBB>
until the CC is alive again, that's the only option we have
<BBB>
ah there it is
<BBB>
:)
<BtbN>
Thunderbird has long given up even sowing the subjects of the E-Mails, and I won't read through ever long paragraphs.
<courmisch>
err, the matrix has bugged
typological has quit [Quit: Connection closed]
<balling>
It is not a bug in Gmail API (though RAW send api would be nice) or Thunderbird. This is bug in patchwork code and it has bugs on top of bugs, like RE: patches are ignored.
<courmisch>
no but it should be physically impossible for NG to thank me
Daemon404 has quit [Changing host]
Daemon404 has joined #ffmpeg-devel
<durandal_1707>
NG is universe
<courmisch>
I think he just might have realised that public ballots were likely against his interests
<courmisch>
because everybody else would feel pressured to vote for El Presidente
<courmisch>
and he does not like El Presidente
<durandal_1707>
El Presidente is little siren
typological has joined #ffmpeg-devel
<courmisch>
El Presidente is the player character in Tropico
<durandal_1707>
what is Tropico?
<courmisch>
a Banana Republic simulation game
<JEEB>
good game
<courmisch>
series
<durandal_1707>
i'm becoming more and more ffmpegophobic
<balling>
LoL, just managed to REGISTER my real gmail by exploting and changing my email to valzapodvz@googlemail.com, libera does not know gmail has aliases.
<durandal_1707>
why are you here? you are officially banned on all FFmpeg channels.
<balling>
Maybe you should look up my gmail account in git log. I have the rights.
<durandal_1707>
and probably on other, non-FFmpeg places...
<balling>
100 of places in fact. So what?
<durandal_1707>
100 places? so you are all the time on net?
balling has quit [Read error: Connection reset by peer]
<Daemon404>
^ this guy was banned from the ML btw
cone-184 has quit [Quit: transmission timeout]
<ePirat>
the name rings a bell
<elenril>
not from trac though, afaik
balling has joined #ffmpeg-devel
balling has quit [Client Quit]
balling1 has joined #ffmpeg-devel
balling1 has quit [Client Quit]
balling has joined #ffmpeg-devel
<balling>
Strange after REGISTER I cannot see your messages. Hmmm
<balling>
Maybe I should threaten to revoke my copyright?? Wow, what a nerve. I helped ffmpeg more than any of you except maybe Paul. Even he does not spend 3 years of his time every day every hour
<balling>
Remember how you had completly broken eac3 encoder for high bitrate and I was bothering Andreas to fix it? And yep, it was fixed. That alone made EAC3 useless. See 968c158abde36ebb7520706a69eebe3e8eacbd81
<balling>
I am him on trac.
<durandal_1707>
ban this citizen!
balling has quit [Ping timeout: 268 seconds]
<durandal_1707>
did someone made git push -F ?
<ePirat>
does not seem like it
<JEEB>
at least I would be surprised if such was done, since there were multiple such mailing list changed patches applied
<JEEB>
already in the git history
<durandal_1707>
Updating 84e400ae37..575efc0406
<durandal_1707>
error: The following untracked working tree files would be overwritten by merge: fftools/ffmpeg_utils.h
<JEEB>
that just means that for some reason you had uncommitted changes there?
<JEEB>
> ffmpeg Cosmin Stejerean via ffmpeg-devel master:575efc040686
<durandal_1707>
i never worked on ffmpeg.c stuff
<JEEB>
> f084e9b0be..575efc0406
<JEEB>
so 575efc is the tip
<JEEB>
and was not changed
<durandal_1707>
so i dunno from where that file came from
<JEEB>
dunno, but it normally updates for me. and 84e400ae37..575efc0406 is a normal straightforward update looking at gitk
<durandal_1707>
strange...
mkver has joined #ffmpeg-devel
<durandal_1707>
ah, this is remaining of my usage of elenril wip patch
pengvado has quit [Server closed connection]
<JEEB>
yea I expected something like that
pengvado has joined #ffmpeg-devel
<JEEB>
esp. since it was untracked
<JEEB>
so it was not in your history
<durandal_1707>
yea
<elenril>
BtbN: why is ddagrab not outputting absolute times as pts?
<elenril>
wouldn't that make more sense?
<BtbN>
A filter can't do that
<BtbN>
I originally made it do that, and it plain did not work at all
<elenril>
why not?
<BtbN>
remove the line that subtracts the start timestamp and see
<BtbN>
it seems like something sits there and wants to wait out the 50 years from 1970 to now
<BtbN>
dropping each and every frame that comes out
<BtbN>
I couldn't figure out what part does that, so I just made the timestamps start at 0
<elenril>
fun
<JEEB>
is it lavfi or ffmpeg cli that's doing the dropping?
<BtbN>
no idea. My guess is ffmpeg.c
<elenril>
i wonder if it runs in wine
<JEEB>
I think I actually at one point poked at it and I somehow had it rolling with some arguments is my memory
<JEEB>
but my memory is at this point *very* hazy
<BtbN>
ddagrab is deeply involved with the graphics driver. I'd be surprised if it works in wine
<kepstin>
probably the ffmpeg cli, there's some weird timestamp stuff it does sometimes on frames it gets from the filters. could also be between the encoder and muxer tho? :/
balling has joined #ffmpeg-devel
<kepstin>
I developed gdigrab while working mostly in wine, hilariously enough.
<JEEB>
yea -vsync passthrough -copyts generally makes it pass things through
<JEEB>
kepstin: :D
<JEEB>
<3
<kepstin>
(was faster in wine on my haswell laptop than in actual windows 7)
<BtbN>
Probably cause the VRAM roundtrip does not exist :D
<JEEB>
then on the output side you might need something like the discontinuity movflag so it doesn't add a packet starting at 0 pts to the beginning of the first actual packet)
<BtbN>
You should observe the exact same effect with any vsrc filter though, if you make it start at a non-zero timestamp
<JEEB>
but just `-f framehash pipe:` with `-vsync passthrough -copyts` or so should give hints?
<durandal_1707>
who uses loudnorm filter in single-pass, dynamic mode ?
<BtbN>
so looking at that specific issue should be easy enough on any OS
<JEEB>
yea
<kepstin>
durandal_1707 i do, occasionally, mostly on live content
<j-b>
good night folks
<BtbN>
There is actually one more capture API that would be cool to have in ffmpeg, but it's only accessible via some insane managed .NET API for some reason
<durandal_1707>
kepstin: i have patch that rewrites filter, and changes dynamic mode algorithm, as I think current dynamic algortithm is suboptimal
<durandal_1707>
rewrites so it uses ebur128 native code, not copy-paste r128 header file.
<JEEB>
if I use loudnorm in mpv, I would guess that is single pass too?
<courmisch>
checkasm: all 95 tests passed
<courmisch>
soon time for the 100th party
<durandal_1707>
JEEB: yes, unless mpv can overwrite rules of physics
<JEEB>
you never know xD but yea
<kepstin>
(yeah, i use it via mpv as well, to fix people livestreaming on twitch or youtube who don't know how to manage their audio levels correctly)
<JEEB>
courmisch: all that delicious SIMD
<courmisch>
DON'T YOU DARE INSULT MY SCALABLE VECTOR CODE AS "SIMD"
<JEEB>
8)
<durandal_1707>
kepstin: the dynamic mode code have faulty limiter and ugly states and complex code flow
<balling>
Meanwhile autovectoriser gcc patch still not applied
<balling>
And BTW, I do use Thorium AVX2, it is so much faster.
<durandal_1707>
it may also over-kill dynamics too much, expecially expander part
<JEEB>
last distribution that I saw enable autovectorization had to manually then disable optimizations for IIRC CABAC or so. and that sounds like fun.
<balling>
Indeed. CABAC is not a problem to fix.
<courmisch>
AFAIU, compilers never went the last mile of auto-ifunc'ing *when* able to autovectorise
<courmisch>
so autovectorise doesn't happen because you still have one guy running a Pentium III and you can't enable SSE2
<balling>
GCC has a lot of bugs, Andreas reported some. In this case the bug was only in CABAC and simple to fix. Also can just disable that file.
<courmisch>
(or on x86_64, you get SSE enabled by default, but not AVX)
<JEEB>
yea I think they're trying to brand new subsets which end with AVX2
<courmisch>
now on RVV, autovectorisation is already working surprisingly well, since the ISA was designed to be easy to autovectorise, but again without automatic ifunc it's mostly useless...
<JEEB>
but most likely not by default outside of distributions that gladly cut compatibility
<balling>
AVX2 is so very nice to have. In Chrome it makes a difference of 20% or more. The fact is x86_64-v3+haswell target is being done for mpv automatically in that repo and Fedora/gentoo start providing v3
<JEEB>
I don't expect you will get that sort of numbers from FFmpeg, since even HEVC decoding in the last flamegraph I made had so much stuff being run by SIMD
<durandal_1707>
i having some very credible info that coup is being prepared for FFmpeg in shadows.
<balling>
Except that *the* patch has numbers. And shows it does make it better as a whole.
<balling>
I have galaxy S22 Ultra (Chinese region model, so has 8 gen 1) and I compile mpv for 845 snapdragon. Also makes a big difference.
<durandal_1707>
whenever power is concentrated into small group it will make FFmpeg future worse.
<durandal_1707>
the compiler SIMD is mostly always suboptimal, better write SIMD for that architecture directly.
<JEEB>
the patch ([PATCH] enable auto vectorization for gcc 7 and higher) does not have numbers. of course it specifically mentions aarch64, so I'm not surprised there might be an effect there as long as no proper SIMD is written.
<balling>
There was also configure: reenable tree vectorization for GCC and some others
balling has quit [Read error: Connection reset by peer]
staceee has quit [Server closed connection]
staceee has joined #ffmpeg-devel
balling19 has joined #ffmpeg-devel
balling19 has quit [Client Quit]
balling has joined #ffmpeg-devel
balling was kicked from #ffmpeg-devel by kierank [balling]
<kierank>
durandal_1707: happy?
<kierank>
Paul asked me to do that
<durandal_1707>
kierank: now just repeat again but now with your nick : kierank
kierank was kicked from #ffmpeg-devel by kierank [kierank]
balling has joined #ffmpeg-devel
<balling>
LOL, seriously?
balling has quit [Client Quit]
kierank has joined #ffmpeg-devel
<kierank>
Miss me?
balling has joined #ffmpeg-devel
balling83 has joined #ffmpeg-devel
<balling83>
Kierank, did not you still not merge the MPEG-TS code for ideal buffer managment?
<kierank>
sure, it talks to x264 for the correct buffer management
<kierank>
how do you plan to make it work for non x264 encoders?
<balling83>
>how do you plan to make it work with rest of ffmpeg
<balling83>
There are other patches still not applied for "interlaced" overlapping of streams in MPEG-TS
balling has quit [Read error: Connection reset by peer]
<balling83>
libavformat/mpegtsenc: new interlaced mux mode
balling83 has quit [Quit: Connection closed]
balling83 has joined #ffmpeg-devel
balling has joined #ffmpeg-devel
<kierank>
still probaly just a finger in the air mux
<balling>
How hard can it be? H.222 spec if finally public, you just need to insert null packets in correct places, no?
<balling>
The patch is libavformat/mpegtsenc: new interlaced mux mode
<kierank>
lol
<kierank>
you have to decided on audio/video scheduling based on vbv
<balling>
Is video only working correctly in HEAD then?
<durandal_1707>
ffmpeg is broken and obsolete
<balling>
1707: yeahBTW, the whole LG Chess.ts file being broken because of 3 tiles and seek is bad in it. I think there is a patch, would be nice to apply it.
<balling>
And quite a lot of othe ts files break horribly on seek. -c copy to mp4 fixes it.
<balling>
avformat/avformat: Introduced `AVInputFormat.read_timestamp2` to fix keyframe seeking for formats that rely on `read_timestamp` for seeking
balling83 has quit [Quit: Connection closed]
balling83 has joined #ffmpeg-devel
* kierank
puts person on ignore
balling83 has quit [Client Quit]
balling83 has joined #ffmpeg-devel
balling has quit [Read error: Connection reset by peer]
balling83 has quit [Client Quit]
balling has joined #ffmpeg-devel
balling has quit [Read error: Connection reset by peer]
<typological>
should I wait for people to review the "hacky" patch I suggested in https://trac.ffmpeg.org/ticket/10659#comment:1 or should I send it to the mailing list and receive review their? I'm only asking because this is my first time dealing with ffmpeg codebase
dellas has quit [Remote host closed the connection]
<balling>
Your patch assumes that subtitles lines cannot overlap. OF COURSE they can.
<balling>
typological: Submit the patch of course, I imagine if the actual underlying PTS/DTS is smaller yet you get rounding error overlap that should be fixed. It is simpler to fix though. If the subtitles underlying timestamps are not overlapping BUT rounded are — it is a bug.
<kierank>
durandal_1707: should we start SpeedAndPowereg
<balling>
They think their FFmpeg fork will be functional. Ahahahaha
<durandal_1707>
BoringAV
<kierank>
SpAceV
<kierank>
AV Exploration Technologies
<durandal_1707>
AV Space Exploration Technologies Limited
<balling>
*Incorporated
<pross>
funmpeg
<durandal_1707>
propeg
<pross>
lgtmpeg
* JEEB
gives the parrot stamp to pross for fun
<Daemon404>
ffmpreg
<durandal_1707>
BikeshedAV
<balling>
Neverwasonmarsmpeg
<durandal_1707>
FlatAV
<JEEB>
or I guess the other thing is "improving your own experience" :P
<JEEB>
like the reason why I'm poking at the side data to encoder config patch set is because I'm annoyed that it's not plugged in yet
<JEEB>
"the info is available, yet not there"
<balling>
Is chroma siting plugged in yet?
<JEEB>
it has been for a good while
<JEEB>
if some wrapper doesn't utilize it, that's a separate thing :P plugged it most definitely is
<balling>
Then why is going to JPEG 420 does not select center automatically?
<JEEB>
dunno, I'm talking of ffmpeg cli, that takes the AVFrame received and utilizes its values
averne has quit [Server closed connection]
<balling>
And AVIF does not set chroma siting in the bitstream.
averne has joined #ffmpeg-devel
<JEEB>
the information is available for the output modules (encoder, muxer)
<JEEB>
that's the plugging I did, if something explicitly ignores that information that's completely separate
<balling>
When you losslessly extract HEVC frame that is tagged as top-left thr md5 will be the same in AVIF 420 samples, but top-left meta is not set. It will not decode tonsame 444
<balling>
*to same 444
dellas has joined #ffmpeg-devel
<typological>
balling: I will submit the patch, thanks. although I don't get your first point, how am I assuming that lines cannot overlap?
<gnafu>
ffaom
<gnafu>
durandal_1707: ^
<durandal_1707>
ffame
<balling>
typological: Lines can overlap. You can have one line that has a modifier to be presented on the top left and another bottom right. Those can be overlapping in time, will not be in space of the frame.
dellas has quit [Remote host closed the connection]
<typological>
balling: I'm aware of that, but my only change is to round the duration after adding the start time, then subtract the rounded start time, this won't affect lines overlapping time, right?
<balling>
typological: what does ccextractor does?