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
<ramiro>
haasn: I'm having an issue with 0rgb -> xv48be. I think it's due to more aggressive optimization of the operation list. the convert operation wants to convert alpha as well, but the vectors haven't been initialized to 0, nor have they been populated by the swizzle: https://pastebin.com/TtggQEQr
thilo has quit [Ping timeout: 260 seconds]
thilo has joined #ffmpeg-devel
thilo has quit [Changing host]
thilo has joined #ffmpeg-devel
Kei_N has quit [Read error: Connection reset by peer]
Kei_N_ has joined #ffmpeg-devel
Kei_N_ has quit [Ping timeout: 252 seconds]
Kei_N has joined #ffmpeg-devel
<fflogger>
[newticket] Martmists: Ticket #11574 ([ffplay] FFmpeg fails to handle symlinks to video devide) created https://trac.ffmpeg.org/ticket/11574
<haasn>
ramiro: good catch; when we optimize away a redundant clear we also need to reset all downstream metadata
<haasn>
in general, maybe we should just always reset all metadata between each optimization pass
<haasn>
since it's free to reinfer at any time
<haasn>
maybe we shouldn't even store inferred metadata alongside the op itself
<haasn>
and just make it a separate call to get the metadata; though that would require extra allocation so I'm not too excited about it
<haasn>
it would certainly be the more robust approach to always reinfer from scratch
abdu has quit [Ping timeout: 240 seconds]
<haasn>
ramiro: pushed a fix and other potential related issues; simplifies the optimizer loop a bit as well
<haasn>
now we no longer make any assumptions about component flags and when they are safe to remove or not and just always fully recompute them after every single optimization pass
<haasn>
on a side note, SWS_COMPS_ZERO is technically redundant with min=max=0 + SWS_COMP_EXACT, but I guess it doesn't hurt to keep it around if only because it makes the code a bit simpler
<jkqxz>
jamrial: Is the reason why you want the new cbs function there purely the buffer referencing for the tiles?
<jamrial>
jkqxz: parsers are not reference counted, so i want to avoid a copy
<jkqxz>
Is removing that referencing a cleaner answer? The fake buffer being made in the parser code is rather clumsy.
<jamrial>
ff_cbs_read() will copy the whole au
<jamrial>
yes, it's not ideal, but until the whole avparser API is replaced by something reference counted (like bsfs), then there are not many options
<jkqxz>
Hmm. AV1 and VVC both do the same thing.
<jamrial>
yeah
<jamrial>
seeing vvc does packetization using avbufferef, it may benefit from this new function without using fake buffers
<jkqxz>
Are others going to want a buffer containing exactly the required data with no offsett?
<jkqxz>
Wondering whether it would be better to just add a buffer argument to ff_cbs_read().
<jkqxz>
(Which can be null, with same behaviour as currently.)
<jkqxz>
4:2:0 from the reference encoder just works with small edits. Worth sending, or would it be better to make a bug against openapv to stop making 4:2:0 files?
ccawley2011 has quit [Ping timeout: 244 seconds]
<jamrial>
jkqxz: they have definitions for 420 in the public headers. might be worth asking what's the plan for the format
abdu has joined #ffmpeg-devel
abdu has quit [Ping timeout: 240 seconds]
abdu92 has joined #ffmpeg-devel
Teukka has quit [Read error: Connection reset by peer]
<jkqxz>
Yay, multisymbol decode does win. First implementation is something like +~50% overall decoder perf, and there is plenty more to tune.
rvalue- has joined #ffmpeg-devel
zsoltiv__ has joined #ffmpeg-devel
zsoltiv_ has quit [Ping timeout: 252 seconds]
rvalue has quit [Ping timeout: 260 seconds]
<jkqxz>
It doesn't fit into the existing multisymbol VLC stuff because you need extra stuff for the adaptive state and to deal with the block being able to end halfway through your multisymbol lookup entry.
abdu has quit [Ping timeout: 240 seconds]
abdu has joined #ffmpeg-devel
rvalue- is now known as rvalue
Coinflipper has quit [Quit: ]
Coinflipper has joined #ffmpeg-devel
___nick___ has quit [Remote host closed the connection]
mkver has joined #ffmpeg-devel
Gramner has quit [Ping timeout: 252 seconds]
<jamrial>
jkqxz: for that matter, i opened a ticket on apv's repo because the library returns a raw bitstream AU from section 12.1 (so including au_size)
<jamrial>
still no replies, but ideally they'd make it configurable
<jamrial>
for now, i made the encoder wrapper strip the au_size