#asahi-dev · 2026-07-20 14:05
sofus(mostly related to slice sizes)#asahi-dev · 2026-07-20 14:05
sofusand either gstreamer or the pr has bugs because our driver only works with gstreamer#asahi-dev · 2026-07-20 14:04
sofusim still not sure firefox actually supports the v4l2 stateless#asahi-dev · 2026-07-20 14:03
ndufesneI suppose you are already familiar with https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20847 (needed for Firefox support), that you can enable support for Chromium directly and that gstreamer supports that too#asahi-dev · 2026-07-20 14:03
ndufesneOk, that's good to know#asahi-dev · 2026-07-20 14:02
sofusthe driver is very simple and should be easy to port to whatever#asahi-dev · 2026-07-20 14:02
ndufesneOk, so to me you should continue that direction#asahi-dev · 2026-07-20 14:01
sofusyes, im only missing av1#asahi-dev · 2026-07-20 14:00
ndufesne(using request)#asahi-dev · 2026-07-20 14:00
sofusyes#asahi-dev · 2026-07-20 14:00
ndufesneyou mean V4L2 stateless ?#asahi-dev · 2026-07-20 14:00
sofusndufesne: fwiw i have made kernel side for h26{4,5} and vp9#asahi-dev · 2026-07-20 13:59
ndufesneThe general long term idea, is that with HW specific driver + user space driver (aka in mesa), through VV standard API, you can move the hard work into userspace#asahi-dev · 2026-07-20 13:59
ndufesneWith V4L2, the kernel API you need here is that stateless codec one, and its more of less the final API implemented by the kernel, so you can already guess that the kernel side is bigger#asahi-dev · 2026-07-20 13:58
ndufesne(of course you can do VV on top of V4L2, its not ideal/optimal)#asahi-dev · 2026-07-20 13:57
ndufesneSo, in both VV and V4L2 case, on Linux, you will need a kernel side#asahi-dev · 2026-07-20 13:57
ndufesne(sorry, back)#asahi-dev · 2026-07-20 13:53
sofuswith VV (right now at least) we would also be "responsible" for parsing bitstream stuff#asahi-dev · 2026-07-20 13:52
sofusif we do v4l2, its only submitting parameters to the HW we are "responsible for"#asahi-dev · 2026-07-20 13:52
sofusto me the question of VV vs v4l2 is kinda more where we put code, and who is responsible#asahi-dev · 2026-07-20 13:50
sofusndufresne: nice, thanks! It looks our driver will be the "most slicebased one"#asahi-dev · 2026-07-20 13:49
ndufresnestill think VV is a long term direction to take#asahi-dev · 2026-07-20 13:49
ndufresneThis does not help the dilema of VV vs V4L2 ... though VV is frame based in its flow#asahi-dev · 2026-07-20 13:48
ndufresneBut its not really a thing today, I talks to about it to the Linux Media group already, have lined up a talk at Linux Plumber in Prague this year, and we have a encoder (kernel side in Rust as a bonus) in the development for the RK3588 chips#asahi-dev · 2026-07-20 13:47
ndufresneWe are working on introduce VV specific (HW specific) drivers in the long term to slim down the software, and more some bits to userspace in the long run#asahi-dev · 2026-07-20 13:46
ndufresneOn HEVC we have a third mode, frame based with slices, this allow for batching in slice based HEVC decoder, or for less capable frame based decoders#asahi-dev · 2026-07-20 13:45
ndufresnehi sofus, at linux kernel level, the V4L2 stateless API supports both slice and frame based decoders, this is because Allwinner CPU access codecs are slice based, but most other are frame based#asahi-dev · 2026-07-20 13:44
leioHey!#asahi-dev · 2026-07-20 13:43
ndufresneo/#asahi-dev · 2026-07-20 13:33
sofusbut it would be slower and more error prone#asahi-dev · 2026-07-20 13:32
sofusyeah, in theory we could let the cm3 do everything#asahi-dev · 2026-07-20 13:29
leiosofus: btw, could the bitstream parsing be something that the CM3 might be capable to handle?#asahi-dev · 2026-07-20 13:13
sofusi think we should do VV if we dont do v4l2#asahi-dev · 2026-07-20 13:03
leioVAAPI would give the most widespread support in apps now, while VV is the future#asahi-dev · 2026-07-20 13:02
sofusbut i think the work required would be the same#asahi-dev · 2026-07-20 13:01
sofusvaapi is also missing a few things, but its not as bad#asahi-dev · 2026-07-20 12:58
leiothat sounds like that would make a straight up VAAPI driver most convenient instead?#asahi-dev · 2026-07-20 12:56
sofusno that makes sense, avd is very much slice based#asahi-dev · 2026-07-20 12:56
leiobut I'm on work call rn, so I'm probably talking nonsense while I can't look#asahi-dev · 2026-07-20 12:55
leiovaapi is slice based, VV is frame based, or so I've been told by people that know both#asahi-dev · 2026-07-20 12:54
sofusi looked at the headers here: https://gitlab.freedesktop.org/mesa/mesa/-/tree/main/include/vk_video#asahi-dev · 2026-07-20 12:52
sofusleio: unless i missed something its not enough (its the same as vaapi)#asahi-dev · 2026-07-20 12:46
leioerr, VV#asahi-dev · 2026-07-20 12:46
leioin VK#asahi-dev · 2026-07-20 12:45
leiosofus: I knew the application is expected to be parsing the bitstream and passes you lower level stuff already?#asahi-dev · 2026-07-20 12:41
leioso there was no bindings to C stuff needed? That works too, I guess#asahi-dev · 2026-07-20 12:39
JamesCalligeros[m]Its all nvk specific stuff#asahi-dev · 2026-07-20 12:39
leioI haven't looked, I assumed most rust stuff you need from mesa generic stuff would be in that PR at least#asahi-dev · 2026-07-20 12:38
sofusleio: am i missing something? It looks like the bindings are only for nouveau?#asahi-dev · 2026-07-20 12:32
sofusmaybe im missing something, this is mainly for the slices/tiles