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
<haasn>
ramiro: 16 doesn’t divide by 3, so you’re always reading or writing some extra unused elements
<haasn>
With each iter
<haasn>
I mean I guess we can unroll two iters together where one is aligned to the start and the second is aligned to the end
<haasn>
And the parts in the middle overlap
<haasn>
But unless unrolling brings performance advantages you’re not gaining anything from doing so?
<haasn>
Except sacrificing flexibility
witchymary has quit [Quit: Leaving]
<cone-178>
ffmpeg Andreas Rheinhardt master:bd2dcfaed478: tests/fate/matroska: Add container cropping test
Poundednote has quit [Remote host closed the connection]
Poundednote has joined #ffmpeg-devel
Poundednote has quit [Remote host closed the connection]
<fflogger>
[newticket] liangliu: Ticket #11596 ([undetermined] convert mp4 to hls failed for certain files, -err_detect ignore_err does not work. The same command works on version 4.4.2, but failed on latest version.) created https://trac.ffmpeg.org/ticket/11596
Poundednote has joined #ffmpeg-devel
<fflogger>
[editedticket] liangliu: Ticket #11596 ([undetermined] convert mp4 to hls failed for certain files, -err_detect ignore_err does not work. The same command works on version 4.4.2, but failed on latest version.) updated https://trac.ffmpeg.org/ticket/11596#comment:1
cone-725 has quit [Quit: transmission timeout]
mkver has joined #ffmpeg-devel
Hassan01 has joined #ffmpeg-devel
cone-239 has joined #ffmpeg-devel
<cone-239>
ffmpeg Gyan Doshi master:c55d65ac0a78: configure: correct liboapv feature support
<Hassan01>
Hey, I'm trying to port a project from gstreamer to ffmpeg and I have a problem where zoompan interpolation doesn't seem to be as smooth as the gstreamer controllable crop filter
<Hassan01>
the expressions are correct in the sense that both videos reach the correct values at the requested keyframes (3,6,9) but ffmpeg is much choppier in the intervals
<Hassan01>
should I make a ticket for this?
<michaelni>
Hassan01, opening a ticket is always a good idea
<michaelni>
but if gstreamer has a filter thats better (maybe zoompan ? i dont know at all) then you can port that too of course
<michaelni>
Hassan01, also check libavfilter/vf_perspective.c this should have high quality interpolation IIRC
Poundednote has quit [Remote host closed the connection]
Poundednote has joined #ffmpeg-devel
Poundednote has quit [Remote host closed the connection]
<Hassan01>
michaelni, gstreamer has a whole system for interpolating parameters of filters generically as long as the filter is written in such a way that the parameter is "controllable" so don't think it's portable to gstreamer
<Hassan01>
to ffmpeg
<Hassan01>
hopefully someone has a better solution than diving into the sources
<michaelni>
Hassan01, try vf_perspective, might work without changing source
Poundednote has joined #ffmpeg-devel
Poundednote has quit [Remote host closed the connection]
<Hassan01>
michaelni, you mean 39.185 perspective ? doesn't seem it can do what zoompan can do
<Hassan01>
zoompan does what I need it just doesn't do it.. smoothly
Poundednote has joined #ffmpeg-devel
Anthony_ZO has joined #ffmpeg-devel
<michaelni>
Hassan01, perspective can do full perspective correction so its not just zoom and pan but also rotate, and general perspective correction
<haasn>
ramiro: I'm not sure I see the point in the memops backend tbh, I think at that point we should cover it with the generic ops code
<haasn>
IMO the normal ops path should be fast enough in this case also
<haasn>
especially with asmjit I struggle to see how a separate backend is improving things over your generated asm loop
<ramiro>
haasn: working on each plane individually is faster than working with multiple planes simultaneously
<haasn>
How about we split separable ops lists and compile/dispatch each plane separately then?
<haasn>
I will begin working on a general solution for that once I’m done with dither
<haasn>
Ideally we can kill backend_mureer
<haasn>
If splitting planes gets us noop=memcpy for free
<ramiro>
haasn: that would definitely help. there are other conversions that would benefit from being independent per plane (I can't remember them atm though, I'm running a new batch of benchmarks)
<ramiro>
I would like to keep the memops backend though. it takes care of a bunch of conversions
cone-923 has joined #ffmpeg-devel
<cone-923>
ffmpeg Henrik Gramner master:fd18ae88ae73: avcodec/x86/vp9: Add AVX-512ICL for 16x16 and 32x32 8bpc inverse transforms
microchip_ has quit [Ping timeout: 252 seconds]
<haasn>
ramiro: for 1-element packed ops I moved the check to op_match in ops_chain.c instead
<haasn>
(allow matching 1-element packed rw as 1-element planar instead)
ChichiMsdk1 has joined #ffmpeg-devel
ChichiMsdk1 has quit [Client Quit]
<IndecisiveTurtle>
jamrial Sorry for the ping, when you get a chance could you do a quick re-review of the new VC-2 patchset
compn has quit [Read error: Connection reset by peer]
compn has joined #ffmpeg-devel
secondcreek has quit [Ping timeout: 252 seconds]
zsoltiv_ has quit [Ping timeout: 248 seconds]
zsoltiv_ has joined #ffmpeg-devel
ngaullier has quit [Remote host closed the connection]
<TheVibeCoder>
Lynne: any ETA when you will start back working on speeding up the TX code performance?
<ePirat>
BtbN, I dont care particularly but I can't imagine its good for performance/cache-ability :D
<ePirat>
TheVibeCoder, solved the issue with my activate function btw. Had to track EOF state of inputs as else it could get stuck in a loop notifying EOF for the same stream over and over again without doing any other work…
<BtbN>
that's just gonna be in whatever order the DB throws it out
<TheVibeCoder>
ePirat: yes, forgot to mention that case that day...
Traneptora has joined #ffmpeg-devel
Traneptora has quit [Remote host closed the connection]
Poundednote has quit [Remote host closed the connection]