<gediz0x539>
messpl: okay. i do not know too much about this domain but you can tell me if you need an extra pair of hands to help, if you feel like it. i'd be happy to see gst and qt utilizing cedrus.
<PPA>
sunshavi: Yeah, release 2019.03. Though I didn't get to building it since I read that the /include/h264-ctrls.h and hevc-ctrls.h files needed to be replaced with the target kernel's
<PPA>
and then the source code elsewhere modified accordingly to reflect any changes in those files
<PPA>
So I diffed h264-ctrls.h from the libva-v4-l2-request include folder against the one from kernel 5.10 and found that so much had changed that I lacked the understanding to account for all of this
<PPA>
(same with hevc-ctrls.h)
apritzel has joined #linux-sunxi
hlauer has joined #linux-sunxi
matthias_bgg has joined #linux-sunxi
gediz0x539 has quit [Quit: Leaving]
victhor has joined #linux-sunxi
kaspter has quit [Ping timeout: 245 seconds]
kaspter has joined #linux-sunxi
random_yanek has quit [Ping timeout: 272 seconds]
random_yanek has joined #linux-sunxi
<messpl>
ndufresne: jernej: gediz0x539: I stucked on https://ctxt.io/2/AACgeXp1EA. Should I add aditional plugin to gstreamer? I thought the "bunny" video should work without any additional plugins, but maybe I am wrong.
tnovotny has joined #linux-sunxi
Mangy_Dog has joined #linux-sunxi
warpme_ has joined #linux-sunxi
camus has joined #linux-sunxi
kaspter has quit [Remote host closed the connection]
camus is now known as kaspter
pgreco has quit [Ping timeout: 272 seconds]
pgreco has joined #linux-sunxi
bshah_ is now known as bshah
gaston1980 has joined #linux-sunxi
<ndufresne>
messpl: gstreamer whise my recommendation would be to move away from VA-API, and make sure you have v4l2codecs plugin installed, it depends on gudev (a udev glib wrapper) to find the devices
<ndufresne>
of course, not sure if VA was for gst or qt
<ndufresne>
wasn't QT multi backend, aka it would do ffmpeg, qt, or some OS native multimedia stack like mediafoundation and the osx thing
<ndufresne>
* ffmpeg, gst I meant
<ndufresne>
messpl: that being said, I've used qmlsink element in gst, works quite well, it's a gstreamer element that provide a QML widget
mmarc__ has quit [Remote host closed the connection]
<messpl>
ndufresne: VA-API means gst's VA-API plugin? If yes I don't have it installed. According to gst-inspect I have v4l2codecs installed.
<ndufresne>
totally, miss-read, I see the line v4l2codecs: v4l2slh264dec: V4L2 Stateless H.264 Video Decoder now
<messpl>
Then it comes to Qt, it uses only gst1 in linux based platforms
<ndufresne>
make sense
<ndufresne>
messpl: are you expecting audio ?
<messpl>
no
<ndufresne>
ok, then I see your are missing the codec parsers from -bad
<ndufresne>
even though in staging (unstable API), these are required
sunshavi has joined #linux-sunxi
<messpl>
ndufresne: audio codecs?
faruk has joined #linux-sunxi
<ndufresne>
messpl: in this case you are missing video parser, h264parse
<ndufresne>
which will prevent playback to work, since let's ay ISOMP4 uses AVCc packing, while v4l2slh264dec current wants byte-stream Annex B
kaspter has quit [Ping timeout: 246 seconds]
gediz0x539 has joined #linux-sunxi
<ndufresne>
h264parse is used to convert, it also used to complete the caps, which otherwise might not be a subset of the decoder caps (e.g. missing profile, level, etc.)
JohnDoe_71Rus has joined #linux-sunxi
<faruk>
Hello all. I have a question about network. I use uClibc on A13. I enabled the network settings I thought necessary and compiled.
mmarc__ has joined #linux-sunxi
<faruk>
I used usb-ethernet for network. My device does not get IP automatically.
<faruk>
I manually enter configurations such as ip address and gateway one by one each time.
<faruk>
How to automate this job?
kaspter has joined #linux-sunxi
mmarc__ has quit [Ping timeout: 240 seconds]
wigyori has joined #linux-sunxi
mmarc__ has joined #linux-sunxi
<messpl>
ndufresne: same effect. should i remove playback plugin or isomp4?
kaspter has quit [Ping timeout: 276 seconds]
<messpl>
it is weird, on kernel 5.10 I had at least information that v4l2_requests cannot allocate memory, now it seems it does not reach driver at all
kaspter has joined #linux-sunxi
<ndufresne>
messpl: let's start from the bottom, since you are I guess trying to get a smaller gst build
<ndufresne>
so start with testing the decoder along
<ndufresne>
something partially static like this ^
camus has joined #linux-sunxi
<ndufresne>
then slowly replace bits, e.g. change the decoder with decodebin then move on to paybin
<ndufresne>
that's how we debug pipelines when developing GStreamer apps, test a small pipeline subset, and then grow it
kaspter has quit [Ping timeout: 246 seconds]
camus is now known as kaspter
mr_rcollins has left #linux-sunxi ["WeeChat 2.8"]
faruk has quit [Quit: Leaving]
kaspter has quit [Read error: Connection reset by peer]
cnxsoft1 has quit [Quit: cnxsoft1]
kaspter has joined #linux-sunxi
gediz0x539 has quit [Ping timeout: 256 seconds]
<messpl>
ndufresne: i got some results. I could not decode the bunny video, but I stared to play with another test video (downloaded from http://demo.nimius.net/video_test, the first one). The command you sent worked on this video so I tried to open it with Qt, In case of Qt I had problems with audio so I removed audio from this file and tried to open it from Qt. it looks like gst is trying to decode frames abut i see only artifacts not correct i
<messpl>
mage. I tried to open another video, and this another video returned me this followed error from kernel: https://ctxt.io/2/AACgtSLEFw
<messpl>
now I see the second video's resolution is 3k x 2k, so maybe this is the problem
<ndufresne>
messpl: got a tritium up and running, I'll test that sample, is that close enough ?
<messpl>
ndufresne: if you are asking about artifacts then no, is see random shapes, colors, lines
<messpl>
I*
gediz0x539 has joined #linux-sunxi
reinforce has quit [Quit: Leaving.]
kaspter has quit [Quit: kaspter]
buzzmarshall has joined #linux-sunxi
<ndufresne>
messpl: yeah, you need a lot of cma for codec without iommu
<ndufresne>
I pass this to my kernel atm, cma=320M
<messpl>
ndufresne: what does the "without this patch cedrus can also oops" ? :)
<messpl>
mean*
<ndufresne>
the kernel can crash
<ndufresne>
I which we were couple of month in the future, gst 1.20 would be out, folks would be on linux 5.11 or 5.12, and it would just work (well despite perhaps a slight overhead ;-P)
<ndufresne>
* overheat
<messpl>
ndufresne: oh crash, seen worst oops
<messpl>
ndufresne: nothing has changed still artifacts in Qt, and still decodebin crashes
<messpl>
linux 5.9 sources seems to be patched
willmore has quit [Ping timeout: 260 seconds]
willmore has joined #linux-sunxi
prefixcactus has quit [Ping timeout: 264 seconds]
<ndufresne>
messpl: just check if the reported format plus video sizes matches your stream
<messpl>
ndufresne: this is so random, now it works with devodebin, but I had a case when it crashed when I was using v4l2slh264dec
<ndufresne>
patch is not applied, Selected format NV12_32L32 16x32
<ndufresne>
would have reported NV12_32L32 640x480 otherwise
<messpl>
i double checked linux sources and they looked patched, let me recompile everything again
<messpl>
ndufresne: f me, it works!
<messpl>
ndufresne: how can I send you a beer?
<asdf28>
PPA nice, lol. i once had a graphics card without fan, had to blow on it every two minutes so the pc didn't hang up
<PPA>
Haha, nice
<ndufresne>
messpl: you know what I miss the most, having a beer in a conference ...
<ndufresne>
but eh, we can't control virus
<ndufresne>
messpl: btw, I still got on my todo to validate on the Lime2, it feels like you have a similar board
<ndufresne>
my goals was to get kmssink to pass the tiled data to the HW directly
<ndufresne>
thought it was a cool thing to support
<ndufresne>
and perhaps get weston to do that for me too
<ndufresne>
(that'S more work, need modifiers)
<ndufresne>
on H3/H5, it just work, cause the decode does linea NV12
<messpl>
I have A33.
<messpl>
custom LVDS
<ndufresne>
the lime2 is an A20 it seems, not sure if smaller means older, we never know
<ndufresne>
but yeah, the display controller can cope with the tiled format, but I don't think the GPU driver does (and its not implement in glupload anyway)
<ndufresne>
it's lack of time the problem, but I want to evolve kmssink to a point we can do app like cody, video player on an HW layer, and GUI overlayed
<ndufresne>
like Khodi (what a weird typo, cody)
matthias_bgg has joined #linux-sunxi
<messpl>
I hope you know you are talking to me in different language ? :D
<wens>
montjoie: did you want "crypto: sun4i-ss - IV register does not work on A10 and A13" backported? gregkh sent notices saying backports didn't apply directly for < 5.4
eduardas has joined #linux-sunxi
vagrantc has joined #linux-sunxi
<montjoie>
yes I see it, lot of patch dont apply on backport
<montjoie>
I need to check them
gediz539 has joined #linux-sunxi
freemangordon has quit [Ping timeout: 260 seconds]
gediz0x539 has quit [Ping timeout: 265 seconds]
freemangordon has joined #linux-sunxi
kaspter has joined #linux-sunxi
kaspter has quit [Remote host closed the connection]
<PPA>
ndufresne: if I've built and installed from source now, and get errors like:
<PPA>
<v4l2decoder1> MEDIA_REQUEST_IOC_QUEUE, failed: No such file or directory
<PPA>
does that mean I need to enable these during the build process (and obviously haven't done that)?
<ndufresne>
PPA: it's a weird error, since we blinding translate the errno to string
apritzel has quit [Ping timeout: 245 seconds]
<ndufresne>
PPA: what is you GStreamer / Kernel combo %
<ndufresne>
?
<ndufresne>
I've hit that one few days ago, but it was my fault, I forgot though
eduardas has quit [Quit: Konversation terminated!]
<PPA>
with no idea what I was doing really, beyond downloading some missing dependencies that meson complained about
<PPA>
The full error message is:
<PPA>
0:00:00.853180533 5208 0xb5a28688 ERROR v4l2codecs-decoder gstv4l2decoder.c:1071:gst_v4l2_request_queue:<v4l2decoder1> MEDIA_REQUEST_IOC_QUEUE, failed: No such file or directory
<PPA>
(running “sudo gst-play-1.0 ...” after doing “sudo meson install” on the completed build directory. Very new to this so I'm not sure if that's how you're supposed to do it.)
<PPA>
oh, one possible mistake I made; going to try something
<PPA>
nevermind
<PPA>
(“nevermind” as in what I thought was a mistake doesn't seem to be one so I'm still in the dark)
kaspter has joined #linux-sunxi
<ndufresne>
need to remember ...
<PPA>
double-oops, sorry for spamming: I *did* make a mistake it seems
<PPA>
gst-play is in gst-examples, right?
<ndufresne>
arg seriously
<PPA>
by the looks of it I didn't build gst-play, so the command earlier invoked the outdated 1.18.3 one—which would explain the errors
<ndufresne>
media: cedrus: Remove checking for required controls
kaspter has quit [Client Quit]
<ndufresne>
PPA: you are missing that change ^ I thought it was in 5.11
<PPA>
oh
* ndufresne
had to read the media controller doc to find out what ENOENT means
<jernej>
ndufresne: so, this commit is must have for gstreamer?
<ndufresne>
it meant that some controls was missing the request, but this is false claim from the driver, since the control framework will cache controls, so we don't update that unless they have changed
<ndufresne>
jernej: well, if you want to avoid sending spurious controls, yes
<ndufresne>
otherwise you have to send SPS/PPS/DECODE_PARAMS for every slices
<jernej>
I know that, I'm just wondering, if gstreamer sends only changed controls or all of them
<ndufresne>
only the changed one
<ndufresne>
it was part of the update for the stable API
<ndufresne>
jernej: Any idea how to request that to be put in stable ?
<jernej>
you mean that commit?
<ndufresne>
yes
<ndufresne>
do I simply ask Mauro by email ?
<ndufresne>
or do we have to ask with a special formating to a specific list ?
<jernej>
there was a discussion on #v4l about that and hverkuil thought that it's not for -rc because it's just relaxing requirements
<jernej>
and it's a bit big
<ndufresne>
well, it's against the spec to do that, will ping Hans
<jernej>
I didn't object too much, though
<PPA>
(Just to be clear so as to not cause unnecessary work: I was using an outdated gst-play probably incompatible with the updated everything else, if therein lay the problem)
<ndufresne>
no, this is a driver issue
<PPA>
oh, ok
<ndufresne>
rkvdec has the same issue though, as it was copy pasted from cedrus I supposed (or perhaps all this was all copy pasted from hantro, can't say)
<ndufresne>
it's from Paul's miss-understanding of what is stateless in all this
<ndufresne>
PPA: but the patch applies cleanly, so if armbian accepts upstream patches, that would be nice
<ndufresne>
obviously we need CI for this
<PPA>
It does. Its build system even has a folder you can just drop patches into, so I can test it myself