<rafael2k>
It would be nice if we could send a N900 running Maemo (and with libcamera set) in case any developer there picks the task
<sicelo>
omap3isp
<rafael2k>
right
<sicelo>
Laurent definitely would help
<rafael2k>
Laurent is listed as mentor
<rafael2k>
for the GSOC project
<sicelo>
gawd! it has 350 hours there, compared to <200 for some of the others :p
<rafael2k>
eheheheh
<rafael2k>
students have plenty of time!
<sicelo>
hehe
<sicelo>
if they won't feel that these devices are dinosaurs
<sicelo>
luckily, at least for N900, student wouldn't be starting from scratch. pavel has taken 'green' pics with it ... i think even documented it on our mailing list
<sicelo>
we have broken dts atm for the camera, afaict, but that should be easy (i'll actually do it)
<freemangordon>
sicelo: knowing c/c++ is the easiest part
<freemangordon>
c has, how many, 19 reserved words? /me check
<sicelo>
thanks, i already had two of the docs, besides the smia functional specifications. tyvm
<freemangordon>
hmm, 32, this seems to have increased back from k&r days :)
<sicelo>
hehe, yes, the languages themselves are easy ... it's groking them that's the problem :p
<sicelo>
structs and pointers
mro has joined #maemo-leste
Twig has quit [Remote host closed the connection]
<sicelo>
and other stuff that you guys seem to magically know. recent example - by looking at the N900 dts, Wizzup realized the modem used PIO. i've looked at it a couple of times, and even today i can't tell how that conclusion was reached
* sicelo
should get a mentor
<sicelo>
< freemangordon> sicelo: knowing c/c++ is the easiest part <<== what's the harder part?
<uvos__>
knowing how everyones elses software, frameworks work/ fit together
<uvos__>
a (c) libarary/framework is kinda its own language but with 100 to 1000 of keywords instead of just 32
<rafael2k>
greenish... need to play with while balance controls in pinhole
<uvos__>
green istent really a white balance thing
<uvos__>
its mostly about the sensetivity curve of cmos sensors
<rafael2k>
I'm guessing
<uvos__>
you have to callibrate it away
<rafael2k>
how?
<rafael2k>
change the gamma curve?
<rafael2k>
or do a red / blue balance?
<uvos__>
so for raw (DNG) conversion
<uvos__>
eatch camera has a "base" curve for eatch channel
<uvos__>
that is applied to convert the cameras lopsided colorspace to (s)RGB
<uvos__>
idk how this relates/ works in libcamera ofc
<rafael2k>
MegaPixels set some registers in the driver and trigger some controls...
<rafael2k>
it is not the green the image there
<uvos__>
the software has to do the covnersion
<uvos__>
megapixels might be doing it
<uvos__>
and assuming the pp sensor
<rafael2k>
the camera needs appropriate calibration controls to be set
<rafael2k>
and megapixels seem to do it right... I'll take a look
<rafael2k>
uvos__, can be, we can even have an option to save the DNG
rafael2k has quit [Remote host closed the connection]
rafael2k has joined #maemo-leste
<rafael2k>
sicelo, tks!
<rafael2k>
Can I say in libcamera that we can donate a N900 if a student gets a GSOC to do this work?
<rafael2k>
(of the N900 camera support)
<sicelo>
yes :-D
<rafael2k>
ok!
<uvos__>
we also can also donate a mapphone as a streach goal, its front camera is pretty mutch in the same state as n900 (omap4s isp is in stageing and out of tree drivers for the sensor exist)
<uvos__>
presumeably most of the work can be shared here
<uvos__>
note "front"
pere has quit [Ping timeout: 268 seconds]
uvos__ has quit [Ping timeout: 252 seconds]
uvos__ has joined #maemo-leste
mro has quit [Quit: Leaving...]
pere has joined #maemo-leste
<rafael2k>
wrote
<Wizzup>
sicelo: re: pio, I just looked at it prior and probably concluded that probably dma just never worked, and probably someone here confirmed that and I just went with it
<Wizzup>
I don't think I deduced it from the dts
<Wizzup>
and yeah we can definitely donate mapphone(s)
<rafael2k>
uvos__, then if someone picks the job, you became a mentor! :P
<rafael2k>
( :P but seriously )
<freemangordon>
Wizzup: yes, it was me (pio)
<Wizzup>
right
<freemangordon>
sicelo: no, you got it wrong - 'we' don't know it magically, it is all hard work looking in
<freemangordon>
the source code of others
<freemangordon>
reading through docs, etc.
<freemangordon>
that's it
<Wizzup>
there is some experience factor I think
<freemangordon>
when you do that for years, you got experience and usually know what to look for and where
<freemangordon>
Wizzup: yes
<freemangordon>
so, it is the same as with other things - hard work
<freemangordon>
and dedication
<freemangordon>
sure, some of 'us' are better than others, but that's to be expected
<Wizzup>
mhm
<freemangordon>
Wizzup: speaking of pio, I don;t remember seeing a patch sent upstream, did I miss something
<Wizzup>
sicelo is in charge of this atm I think :)
<Wizzup>
or rather, I'm didn't send anything :D
<freemangordon>
hehe
<freemangordon>
be careful what you wish for :p
<freemangordon>
ttyl
<sicelo>
mmm, i guess it should just be a question/discussion? i suppose we're not at patch stage, since we're just forcing pio on a driver that could possibly be used by something else with dma
<Wizzup>
the question I guess is what changes that made it now take a dma path when it's not supported
<Wizzup>
changed*
<norayr>
having fcamera and blessn900 working on all maemo phones would be amazing.
<Wizzup>
blessn900 isn't foss jus so you know
<Wizzup>
afaik
<bencoh>
fcamera would be cool indeed :)
<bencoh>
but looks like rafael2k already has an app of his own anyway
<sicelo>
yeah, seems good to me
<bencoh>
still, I'm not quite sure how the fcamera kernel driver(s) worked, but having a look at it might ease development on mainline
<bencoh>
iirc they bypassed the ISP in every aspect (?)
<sicelo>
the little i read seems to say - forget about fcamera :-P
<sicelo>
it needed its own special kernel module, and a lot of other hacks
<bencoh>
sicelo: well, fcam aimed at providing frame-accurate metadatas
<bencoh>
nobody did anything close to that on linux back then
<uvos__>
well besides all the other phoens running linux presumably
<bencoh>
back then?
<bencoh>
and no, frame-accurate wasn't (and still isn't?) a common thing on linux
<bencoh>
v4l2 had no proper metadata handling mechanism back then
<uvos__>
android probubly, motorola linux (dont remember what it was called, most late motorola featire phones ran linux)
<bencoh>
(it's hardly better nowadays)
<bencoh>
uvos__: in 2009?
<uvos__>
yes
<uvos__>
in 2007 even
<bencoh>
samsung implemented there own framework, but afaict it came a bit later
<uvos__>
anyhow
<bencoh>
their*
<uvos__>
these devices where locked down linux
<uvos__>
so probubly terrible hacks
<uvos__>
(besides android idk how the state was on android that early)
<uvos__>
android is a good bit older than n900 after all
<bencoh>
only 2y ?
<uvos__>
a good bit in this space
<bencoh>
not even 2y, it was released in sept 2008 apparently
<sicelo>
bencoh: i completely agree with what you said, re-fcam. so yes, it's hard to use it with regular linux because of exactly that same reason
<uvos__>
i used to use this ^^^
<Wizzup>
looks like google doubt bought it in 2005
<Wizzup>
s/doubt//
<uvos__>
sue i mean post htc dream ofc
<uvos__>
and post d1 it was the second largest platform
<uvos__>
this is same time as n900
<uvos__>
anyhow
<uvos__>
this is hardly relevant
<sicelo>
yes, I was still trying to understand how it's relevant. to be clear, fcam that's being discussed is not the stock camera
<bencoh>
what's the current camera state on n900/mainline btw?
<bencoh>
according to elinux it worked on v4.14-rc1
<sicelo>
drivers and isp work
<sicelo>
but i think something's wrong with some clocks ... probably just a simple dts update (that i will look into)
<bencoh>
ah
<sicelo>
actual capturing is a completely different problem ... but maybe libcamera helps there
<sicelo>
see complex cameras talk on YouTube, by Pavel Machek
<sicelo>
at least he could take green photos of his horses :-p
<bencoh>
haha
<bencoh>
yeah, whitebalance/focus is yet another story, but once you have a way to capture frames and control the sensor/actuator, it's already a great start