cnxsoft has joined #imx6-dev
cnxsoft has quit [Ping timeout: 252 seconds]
akaizen has quit [Remote host closed the connection]
akaizen has joined #imx6-dev
akaizen has quit [Remote host closed the connection]
akaizen has joined #imx6-dev
akaizen has quit [Remote host closed the connection]
akaizen has joined #imx6-dev
diego_r has joined #imx6-dev
akaizen has quit [Remote host closed the connection]
akaizen has joined #imx6-dev
fossxplorer has quit [Remote host closed the connection]
akaizen has quit [Remote host closed the connection]
rz2k has joined #imx6-dev
uRandomMM has joined #imx6-dev
rz2k has quit [Ping timeout: 252 seconds]
rz2k has joined #imx6-dev
datagutt has joined #imx6-dev
steeve has quit [Remote host closed the connection]
codinho has joined #imx6-dev
codinho has joined #imx6-dev
codinho has quit [Changing host]
projectg1s has joined #imx6-dev
projectgus has quit [Ping timeout: 252 seconds]
rz2k has joined #imx6-dev
akaizen has joined #imx6-dev
akaizen has quit [Ping timeout: 240 seconds]
codinho has quit [Ping timeout: 240 seconds]
diego_r has quit [Quit: Konversation terminated!]
jas-hacks has joined #imx6-dev
<
jas-hacks>
dv_: hi, regarding 'zero copy'. I'm guessing in this pipe line the buffers are copied ?
<
jas-hacks>
dv_: gst-launch-1.0 v4l2src ! 'image/jpeg,width=1280,height=720,framerate=10/1' ! jpegdec ! autovideosink
akaizen has joined #imx6-dev
<
dv_>
this is the regular v4l2src , isnt it
<
jas-hacks>
dv_: yep
<
dv_>
I can see at least one copy: inside the videosink
<
dv_>
on the imx6, the autovideosink will select my eglvivsink if installed. otherwise the ximagesink I guess
<
dv_>
the ximagesink is super slow. eglvivsink will probably show something smooth, but with higher CPU usage
<
jas-hacks>
it picks ups eglvivsink
<
dv_>
also note that the vpu decoder can decode motion JPEG
<
jas-hacks>
yes at the moment around 70% cpu
<
dv_>
thats partially because of jpegdec, partially because of the CPU frame copy
<
dv_>
in theory eglvivsink would propose a dma buffer pool to the upstream elements, so they would decode directly into a dma buffer
akaizen has quit [Ping timeout: 240 seconds]
<
dv_>
in practice though this is *slower*, because the existing allocators do not allow me to allocate with writecombine flags
<
dv_>
hopefully, with jnettlets kernel , this will change
<
jas-hacks>
so I could try 'imxvpudec' instead of jpegdec?
<
jas-hacks>
dv_: I was expecting this to work but there's no output
<
jas-hacks>
gst-launch-1.0 v4l2src ! 'image/jpeg,width=1280,height=720,framerate=10/1' ! imxvpudec ! autovideosink
<
dv_>
run this with GST_DEBUG=*vpu*:6
<
dv_>
hm it is showing the frames.
<
dv_>
oh Y42B as output!
<
dv_>
try replacing autovideosink with imxeglvivsink
<
dv_>
it will probably not run then
<
dv_>
thats interesting. so far I havent encountered a motion JPEG input which produced Y42B output
<
jas-hacks>
sets output as Y42B
<
jas-hacks>
what's deciding the output as Y42B?
<
jas-hacks>
temp fix is to put a 'videoconvert' between imxvpudec & imxeglvivsink
<
dv_>
try imxipuvideoconvert
<
jas-hacks>
'no element'
<
dv_>
err imxipuvideotransform
<
jas-hacks>
yes found it, great now it works better
<
dv_>
you see an image?
<
dv_>
not corrupted?
<
jas-hacks>
imxeglvivsink reports "alot of buffers were dropped"
<
jas-hacks>
video is not corrupted
<
dv_>
this is normal at the beginning
<
dv_>
test if the video lags a lot
<
dv_>
I am testing with an old webcam here, and I get corrupted frames
<
jas-hacks>
I'm testing with a microsoft HD-3000, the frames are clear
<
jas-hacks>
CPU is < 20 %
<
dv_>
hmm interesting. I get the same corruptions if I use the fsl plugins
<
dv_>
perhaps this is a VPU problem, and the cam is sending frames it cannot handle
<
dv_>
anyway, the eglvivsink should report an error if you try to show Y42B frames.
<
dv_>
I'll fix that
<
jas-hacks>
I noticed I'm getting a pause periodically
<
dv_>
might be a timestamp problem
<
jas-hacks>
dv_: what's the issue with the timestamps?
<
dv_>
I dont know. but I do know that v4l sources sometimes do have timestamp issues
<
dv_>
try if you get the pauses with jpegdec
<
jas-hacks>
does seem to happen with jpegdec
<
jas-hacks>
typo - it's not happening with jpegdec
<
jas-hacks>
I left the imxiputransform element in the pipeline and just replaced imxvpudec
<
dv_>
oh I think I now what this is
<
dv_>
yet another special case
<
dv_>
yeah. something is special about mjpeg inside the VPU wrapper. gha.
<
dv_>
try adding sync=false to the sink
<
jas-hacks>
behaves better :)
<
jas-hacks>
dv_: i'll do some more testing, thanks for your help today !
uRandomMM has quit [Quit: Page closed]
jas-hacks has left #imx6-dev [#imx6-dev]
bfederau has quit [Read error: Connection reset by peer]
bfederau has joined #imx6-dev
akaizen has joined #imx6-dev
unsolo has quit [Ping timeout: 240 seconds]
akaizen has quit [Ping timeout: 240 seconds]
akaizen has joined #imx6-dev
akaizen has quit [Read error: Connection reset by peer]
akaizen has joined #imx6-dev
akaizen has quit [Read error: Connection reset by peer]
akaizen has joined #imx6-dev