TobiasTh1Viking has quit [Read error: Connection reset by peer]
TobiasTheViking has joined #linux-amlogic
Barada has joined #linux-amlogic
Elpaulo_m has joined #linux-amlogic
jakogut_ has joined #linux-amlogic
jakogut has quit [Ping timeout: 240 seconds]
jakogut_ is now known as jakogut
tingoose has joined #linux-amlogic
Elpaulo_m has quit [Ping timeout: 264 seconds]
Elpaulo_m has joined #linux-amlogic
a5m has joined #linux-amlogic
tingoose has quit [Ping timeout: 240 seconds]
cthugha is now known as ldevulder
mchappellier has joined #linux-amlogic
mchappel_ has joined #linux-amlogic
mchappellier has quit [Read error: Connection reset by peer]
mchappellier has joined #linux-amlogic
mchappel_ has quit [Read error: Connection reset by peer]
mchappel_ has joined #linux-amlogic
mchappel_ has quit [Read error: Connection reset by peer]
mchappellier has quit [Read error: Connection reset by peer]
mchappellier has joined #linux-amlogic
mchappel_ has joined #linux-amlogic
mchappellier has quit [Read error: Connection reset by peer]
mchappellier has joined #linux-amlogic
mchappel_ has quit [Read error: Connection reset by peer]
afaerber has joined #linux-amlogic
xdarklight has quit [Ping timeout: 240 seconds]
libv has quit [Remote host closed the connection]
xdarklight has joined #linux-amlogic
libv has joined #linux-amlogic
fixmer has joined #linux-amlogic
<fixmer>
hi
afaerber has quit [Quit: Leaving]
mchappel_ has joined #linux-amlogic
shadeslayer_ has joined #linux-amlogic
bengal_ has joined #linux-amlogic
mag- has joined #linux-amlogic
mchappellier has quit [*.net *.split]
shadeslayer has quit [*.net *.split]
indy has quit [*.net *.split]
naobsd has quit [*.net *.split]
libv has quit [*.net *.split]
The_Coolest has quit [*.net *.split]
bengal has quit [*.net *.split]
mag has quit [*.net *.split]
nashpa has quit [*.net *.split]
bengal_ is now known as bengal
bengal has quit [Changing host]
bengal has joined #linux-amlogic
nashpa has joined #linux-amlogic
naobsd has joined #linux-amlogic
libv has joined #linux-amlogic
indy_ has joined #linux-amlogic
Elpaulo_m has quit [Ping timeout: 256 seconds]
shadeslayer_ is now known as shadeslayer
afaerber has joined #linux-amlogic
indy_ is now known as indy
afaerber has quit [Remote host closed the connection]
indy has quit [Read error: Connection reset by peer]
indy has joined #linux-amlogic
<narmstrong>
Ely: is there a lot missing in the HEVC driver ?
yann has joined #linux-amlogic
<Ely>
hi fixmer
<Ely>
narmstrong: yes! Right now I can't get the esparser to work with hevc, and I only implemented the hevc init sequence, no ISR or anything else yet.
afaerber has joined #linux-amlogic
<Ely>
In fact I'm so stuck on this esparser thing that I'm wondering using the "frame based" mode of the hevc decoder where you can just feed it packets from memory instead of relying on the esparser
<Ely>
But it's still very unclear between all the SoCs versions if it would be easily backportable to previous SoCs
nobe has quit [Quit: Leaving]
tingoose has joined #linux-amlogic
dsd_ has joined #linux-amlogic
afaerber has quit [Ping timeout: 240 seconds]
afaerber has joined #linux-amlogic
a5m has quit [Remote host closed the connection]
Elpaulo_m has joined #linux-amlogic
Elpaulo_m has quit [Client Quit]
afaerber has quit [Client Quit]
Barada has quit [Quit: Barada]
<Ely>
the H.264 decoder still gives me sparse corruption on some bitstreams.. Like there are some gops it really doesn't like and where the output will have artifacts for a few secs. It's always the same files at the same moment.
<Ely>
Have you noticed that as well ?
<narmstrong>
yes
<narmstrong>
it's reproductible with the bbb streams
<Ely>
ah yup got it..
<narmstrong>
Ely: it was a stupid 1 character error ?
<Ely>
oh I mean I reproduced it with the bbb 1080p30 stream, I didn't fix it :D
<Ely>
It's encoded with bframes=16 so maybe we hit too many consecutive bframes
<Ely>
It seems to be failing at decoding a keyframe... But I can't put my finger on the why
yann has quit [Read error: Connection reset by peer]
<ndufresne>
Ely, have you trying growing some of the internal buffer size ?
<Ely>
Yeah I tried growing the ref size, the v4l2 buffers (canvas number), and parameters like "max_reference_size" and "max_dpb_size" but none fixes it
<ndufresne>
Ely, might sound stupid, but could it be that some key frames are larger then 2M ?
<ndufresne>
(encoded)
<Ely>
I checked, the packet size is like 400kiB
<ndufresne>
GStreamer naïvely request 2M
<ndufresne>
ok
<Ely>
but the error specifically triggers on the following chain: PPPPI
<ndufresne>
coda have some code to increase that over what we request, but I think it's rare that driver have any kind of logic/heuristic
<Ely>
and then the I (ref) frame gets corrupt, which corrups all the following frames
<ndufresne>
hmm ....
<ndufresne>
well, stupid question, but is it clean on amlogic driver / kernel ?
<Ely>
must admit I haven't tried
<Ely>
I need another SD card because it's a pain otherwise
<Ely>
I wonder how many video engineers got crazy watching bbb over and over again
<dsd_>
have you actually watched it in full?
<Ely>
I don't believe I have
<dsd_>
i'm one of those people but i think i never got past about 20 seconds
<Ely>
lmao
<ndufresne>
haha, I know the end, because I keep debugging EOS problems (dropped frames) with it ;-P
<ndufresne>
8 years of bbb ! should add that to my resume
dsd_ has quit [Quit: Lost terminal]
dsd_ has joined #linux-amlogic
yann has joined #linux-amlogic
vagrantc has joined #linux-amlogic
tingoose has quit [Ping timeout: 252 seconds]
<Ely>
ndufresne: I think the problem lies within the esparser and the speed at which we throw buffers at it. Like, right before the error there are like 10+ buffers queued. It amounts for a ridiculously low amount of data (476KB), but maybe we do it too fast.
<Ely>
This problem is exacerbated when using ffmpeg -f null -, the buffers are queued in so fast that every output frame has errors.
<ndufresne>
maybe we need to drain on IDR ?
<ndufresne>
I remember there was some old example they would inject some special SEI, decoder commands
<ndufresne>
(that's how we found out how to drain at EOS)
<Ely>
you mean their userspace injects packets ?
<narmstrong>
Ely: can't you throttle the es inject with the buffers capture rate ?
<narmstrong>
seems amlogic does something similar in their code,
<Ely>
I added a quick msleep(20) in between every esparser queue but the error still happens at the same spot
<Ely>
"seems amlogic does something similar in their code" => Where did you see this ?
<Ely>
The only stalling I saw was when the vififo (streambuf 32MiB) is full, but it never happens on my setup.
<narmstrong>
their housekeeping loop seems less aggressive than yours
ldevulder has quit [Ping timeout: 276 seconds]
ldevulder has joined #linux-amlogic
<dsd_>
the decoder needs previous frames in order to decode new ones; are you freeing previous ones too soon?
<Ely>
The problem here is that it fails on an I-frame which shouldn't require previous frames anyway
<Ely>
But no in any case I don't think I'm freeing them too soon, I give them to userspace when the decoder tells me they're ready, and only "free" them when userspace gives them back to me.
<ndufresne>
Ely, I-frame vs IDR Frames
<jakogut>
I see on linux-meson.com that eMMC and SD cards are supported, but NAND is not. Is that currently the blocker for onboard storage working?
trem has joined #linux-amlogic
tingoose has joined #linux-amlogic
<Ely>
ndufresne: right!
<Ely>
In this case it's an IDR
ldevulder has quit [Ping timeout: 265 seconds]
sputnik_ has quit [Remote host closed the connection]
sputnik_ has joined #linux-amlogic
<Ely>
ndufresne: I can see the EOS corner case in EM kernel but it's only used for EOS.
dsd_ has quit [Quit: Lost terminal]
mchappel_ has quit [Read error: Connection reset by peer]
mchappellier has joined #linux-amlogic
ldevulder has joined #linux-amlogic
trem has quit [Quit: Leaving]
tingoose has quit [Ping timeout: 265 seconds]
mchappellier has quit []
<Ely>
Huh. I have a lot less errors when increasing the vififo size from 32MiB to 64MiB. Maybe the errors happen when it has to wrap around, like it writes half the packet at the end, the other half at the start.
bengal has quit [Ping timeout: 248 seconds]
<Ely>
yup! Happens when it wraps around
<Ely>
On to find out how AML works around that now