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 7.0.2 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
iive has quit [Remote host closed the connection]
thilo has quit [Ping timeout: 245 seconds]
witchymary has quit [Remote host closed the connection]
<JEEB>
ok, is this where I start preparing champagne when one of the many orgs poking at v4l2-requests finally started upstreaming
<Lynne>
its too late to the party
<JEEB>
sure, but it was becoming a joke
<JEEB>
that it would forever be in just XYZ forks
<Lynne>
doesn't it need the annoying proprietary sand pixel format to even be useful on raspberry pi 4s?
<JEEB>
yea the output is interestingly packed, not sure if the tiling is visible when you try to access it
Krowl has quit [Read error: Connection reset by peer]
Curid has joined #ffmpeg-devel
rvalue- has joined #ffmpeg-devel
rvalue has quit [Ping timeout: 245 seconds]
rvalue- is now known as rvalue
<Kwiboo>
JEEB: there is still no org behind upstreaming efforts, just me being tired at waiting on RPi folks, instead I rewrote/adjusted our poc variant to also work efficiently with RPi ;-)
<JEEB>
coolio :)
<Kwiboo>
Lynne: correct, on RPi the buffer format use a special sand128 memory layout, I still do not understand the mem layout, but for displaying decoded video we do not need to understand it, just pass it along to the display system
<Lynne>
you have download code in your patchset, what happens if you try to download frames on an rpi?
<Kwiboo>
the rpi fork has lots of code to deal with the format to be able to convert it into other formats, not something I am planing on digging into
<Kwiboo>
sw_format is set to NONE and the transfer_from op return EINVAL
<Lynne>
any reason to have that code then?
<Lynne>
from the sound of it, on platforms like rockchip, where it should be possible, it wouldn't work
<JEEB>
P030 which is apparently the drm name for the 10bit format it's like 2:10:10:10 for one plane in luma (bitbanging), and then chroma is similar but 2x [2:10:10:10]
<Kwiboo>
some decoders can decode video into linear NV12, those can use hwdownload and convert into other formats, mostly to get MD5 for conformance testing
<JEEB>
yea I think some should have NV12, P010
<JEEB>
although with 10bit I wonder how many vendors chose to bitbang
SystemError has quit [Remote host closed the connection]
<Lynne>
Kwiboo: so sw_format is only set to none on a rpi4?
<JEEB>
but given that this thread I see only talk about 10bit, I guess even rpi has relatively sane 8bit?
Krowl has joined #ffmpeg-devel
<JEEB>
oh and the SAND128 modifier, apparently 128 byte columnar?
<Kwiboo>
when working on kernel drivers we must supply fluster testing scores to see that we are note breaking (or fixing) decoding, with https://github.com/fluendo/fluster/pull/179 we can test decoding that produce a format that can be converted by ffmpeg
<JEEB>
really weird that they are exposing that tiling or whatever to user space without alternatives?
<Kwiboo>
Lynne: any format that is currently not known to ffmpeg is set to none, there is a mapping table in one of the files
<JEEB>
at least either in mesa or somewhere there is some shader code for this stuff
<Kwiboo>
JEEB: I think both 8- and 10-bit formats on rpi use the sand128 format
<Kwiboo>
if I understand correctly it is the native format of the hw decoder and display pipeline, and if you just want to display the decoded video as efficiently as possible userspace does not need to understand the memory layout at all
<JEEB>
sure
<JEEB>
it's just that there's so much hardware that does NV12/P010 etc
<JEEB>
of course embedded is embedded
<Kwiboo>
hehe, yeah, it is possible the hw could support other formats, I have not digged deep enough into what rpivid fully can do
<JEEB>
I would not be surprised if this is just broadcom being broadcom, that's why > embedded is embedded
<Kwiboo>
there is also other decoders on e.g. allwinner that will produce frames using arm framebuffer compression to save on memory bandwith for 4k 10-bit 60fps use cases
zsoltiv_ has quit [Ping timeout: 252 seconds]
<Kwiboo>
and mediatek are also using special/custom formats with modifiers, have not tested on mediatek yet but had a board delivered yesterday to play around with
johge has joined #ffmpeg-devel
Krowl has quit [Read error: Connection reset by peer]
johge has quit [Remote host closed the connection]