[RFC 0/2] VP8 stateless V4L2 encoding uAPI + driver

[Nicolas Frattaroli]

On Monday, 20 March 2023 11:07:19 CET Andrzej Pietrasiewicz wrote:
> Hi Nicolas,
> 
> 
> W dniu 18.03.2023 o 10:20, Nicolas Frattaroli pisze:
> > On a related side note, since I let this run all night with different
> > parameters I can happily report that it seems to be quite stable, no
> > problems encountered at all.
> 
> Thank you for reporting. In the (expected) case this turns into a
> patchset proper I would kindly ask for your Tested-by then.

Will do, I'll be closely tracking this patchset and might also throw
a patch your way to enable it on RK356x at some point, since that has
the same Hantro encode IP as well as far as I know.

> Can you share what you used for the nightly tests, both in terms of
> testing harness and unencoded video material?

The source material I used is the "Original" quality of the short film
"Wanderers" by Erik Wernquist: https://vimeo.com/108650530

You can click on "Download" and choose "Original" from there, which
gives you a 4:2:2 10-bit Apple ProRes .mov file. It's quite high
quality and includes some interesting segments, such as the asteroid
field shot, as well as plenty of grain, both of which really stress an 
encoder.

My testing harness is a little primitive, the precise gst pipeline I
used is in this command I ran:

for i in {0..63}; do echo "q $i"; \
gst-launch-1.0 filesrc location=~/Wanderers.mov ! \
qtdemux name=demux demux.video_0 ! decodebin ! videoconvert ! \
v4l2slvp8enc min-quality=$i max-quality=$i ! queue ! matroskamux ! \
filesink location="/mnt/usb/w2/vp8_wanderers_q_$i.mkv"; done

I figured I'd try out all the quantiser levels this way. It is worth
noting that the resultant mkv files are somewhat odd, I had to remux
them with ffmpeg (with -c:v copy to copy the vp8 bitstream over) to
get mpv to seek properly in them and show bitrate information in the
stats overlay (shift+i by default). There's probably a gstreamer
thing I'm unaware of to make it properly generate the matroska
container as well. Either way, not a problem with the encoder, just
the muxer.

I'm pretty sure videoconvert gets rid of the 4:2:2-ness and 10-bit-ness
of it, since I don't think this hardware encoder is capable of handling
that, which is fine by me.

However, since this does nicely transcode to a 10-bit video with a
software encoder of your choice, Collabora might also be interested
in using this footage for any upcoming 10-bit patches to the video
decode side of things. :)

> Regards,
> 
> Andrzej

Kind regards,
Nicolas Frattaroli