<mszyprow>
memeka: it is common to 5410, 5420, 5422 and 5800 SoCs
<memeka>
mszyprow: it's not part of soc node, right?
<memeka>
oh i think it is
<mszyprow>
memeka: good question
<mszyprow>
memeka: I would put it in soc
<mszyprow>
as there is nothing outside it
<memeka>
yes :)
<mszyprow>
and exynos3250 also have it under soc
<mszyprow>
memeka: once you generate a patch (a diff with subject, commit message and sign-offs), check it with "./scripts/checkpatch.pl PATCH_FILE" and send to all reported by "./scripts/get_maintainer.pl PATCH_FILE"
<mszyprow>
memeka: better :) maybe extent first commit messge line to "Enable support for ARM Performance Monitoring Units available in Cortex-A7 and Cortex-A15 CPU cores for Exynos54xx SoCs (5410, 5420 and 5422/5800)."
<memeka>
damn sent it already :P
<mszyprow>
memeka: and add your "Signed-off-by: Marian Mihailescu <mihailescu2m@gmail.com>"
<memeka>
time for V2
<mszyprow>
maybe I can also get "Suggested-by: Marek Szyprowski <m.szyprowski@samsung.com>" ;)
<wwilly__>
by the way, 5 counters and 7 counters, is it actually 4 and 6 programmable, + cycles at all time?
<memeka>
ok V2 sent thanks mszyprow
<memeka>
mszyprow: what's the "Scaler" ?
<mszyprow>
memeka: Memory-to-Memory scaler
<memeka>
it's not GScaler, issit?
<mszyprow>
memeka: a bit more powerful comparing to GScaler
<memeka>
There are 3 Scaler devices in Exynos5420 SoCs
<mszyprow>
memeka: has almost no limits related to processed formats
<memeka>
never heard of them until now
<mszyprow>
memeka: while GScaler requires buffers to be aligned to 8 pixels
<memeka>
i mean not even in the 3.x kernels were any reference to them, from what i can remember
<mszyprow>
maybe noone advertized them ;)
<memeka>
are there gonna be v4l2 drivers for them as well? :D
<mszyprow>
I plan to make a generic v4l2 mem2mem driver on top of Exynos DRM IPP
<memeka>
(or are there already patches ?:D)
<mszyprow>
to avoid this dual-world mess
<memeka>
tri-world :))
<memeka>
v4l2, ipp, ippv2 :))
<memeka>
crazy
<mszyprow>
yea...
<mszyprow>
:(
<mszyprow>
ipp v1 is so broken that we shouldn't count it ;)
<memeka>
is there any benefit in the v2 patches?
<mszyprow>
you mean ipp v2?
<memeka>
yeah
<mszyprow>
it is working compared to ipp v1
<memeka>
dunno what actually uses it :)
<mszyprow>
state-less api
<memeka>
as in, what was using it and had issues?
<mszyprow>
nothing besides some Samsung internal things with additional patches
<mszyprow>
memeka: ipp v2 has very simple userspace api
<mszyprow>
memeka: all processing done by a single ioctl
<memeka>
yeah but still it will be used mostly by the internal samsung stuff :)
<memeka>
without it, there is not pts coming out of the encoder :)
<mszyprow>
hmm, there have been some discussion about those time stamps, but I must admit that I didn't track it
<mszyprow>
hmmm, just checked that dmabuf export is already there for both encoder and decoder
<memeka>
i just know that i had some issues with gst encoding (the output had no pts), found a discussion also about this on a chromebook forum, and adapted a patch from the 3.x kernel for 4.x
<memeka>
yeah export is there - it was only encoder import that was important :)
<memeka>
btw, latest ffmpeg released 3 days ago has MFC support (v4l2) :)
<memeka>
but has no dmasupport yet, so it's pretty slow :D
<mszyprow>
okay, so the last thing is to check the iommu related issues
<memeka>
oh yeah iommu on drm dmabuf import
<memeka>
i think i saw something about this on a page you maintain
<memeka>
mainline to-do
<memeka>
yes
<memeka>
"with IOMMU, importing a dma-buf in exynos-drm leads to a system crash as well:"
<memeka>
not sure how long transfer.sh keeps files
<memeka>
if you checkout tag 1.12.3 in all those repos
<memeka>
it should be ok
<memeka>
you just need to ignore the first and last patch in there - they're for the debian packaging files. just make sure to compile using the flag "--without-libv4l2" (the first patch adds that to debian/rules)
<memeka>
or even better, you can install gst on debian unstable, and just apt-get source for gst-plugins-good
<memeka>
apply the patches (now all should work), then rebuild just the gst-plugins-good
<memeka>
anw
<memeka>
mszyprow: finally, had to install several CPAN modules :|
<memeka>
is the email ok now? :)
<mszyprow>
memeka: I got the email
<mszyprow>
memeka: what are CPAN modules?
<mszyprow>
memeka: the only thing that has to be fixed is that patch should not be sent as attachment
<mszyprow>
memeka: for git send-email you need to setup a few entries in .gitconfig (like smtpserver, smtpuser)
<javier___>
memeka: exactly, I meant with IOMMU enabled. With CMA it worked well for me
<memeka>
yeah me too, 1080p movie with 5-8% cpu load on kms output :|
<javier___>
and I also remember Nicolas Dufresne (v4l2 gst maintainer) also reported the same dma-buf import issue on exynos4
<javier___>
which may imply that the problem is in the exynos drm driver, since for exynos5 is mfc -> gsc -> exynos drm but for exynos4 is mfc -> fimc -> exynos drm
<memeka>
javier___: yes i agree
<memeka>
with exynos_iommu=y, mfc -> gsc dmabuf-input -> other sink works fine
<memeka>
it's so annoying samsung dropped Arktik 1020 :(
<memeka>
for the exynos4 arktiks it provides a nice wayland mali drivers, including dmabuf-import
<memeka>
for the exynos5422 1020 - they dropped it before a nice wayland driver was released
<memeka>
damn!
mszyprow has quit [Ping timeout: 248 seconds]
LiquidAcid has joined #linux-exynos
nighty- has joined #linux-exynos
nighty- has quit [Ping timeout: 260 seconds]
kxkamil has quit [Quit: Leaving]
LiquidAcid has quit [Remote host closed the connection]
LiquidAcid has joined #linux-exynos
LiquidAcid has quit [Remote host closed the connection]