Current Server Time: 12:52 (Central Europe)

#apertus IRC Channel Logs

2018/05/07

Timezone: UTC


02:00
rton
left the channel
03:39
sups
joined the channel
03:46
sups
left the channel
08:49
se6astian|away
changed nick to: se6astian
09:01
rton
joined the channel
09:37
sebix
joined the channel
09:37
sebix
left the channel
09:37
sebix
joined the channel
10:07
se6astian
https://www.apertus.org/axiom-beta-roadmap updated
10:18
Bertl_zZ
changed nick to: Bertl
10:18
Bertl
morning folks!
11:15
seaman
joined the channel
13:01
TofuLynx
left the channel
13:55
Bertl
off for now ... bbl
13:55
Bertl
changed nick to: Bertl_oO
14:06
Kjetil_
joined the channel
14:06
alexML_
joined the channel
14:11
anuejn2
left the channel
14:11
Kjetil
left the channel
14:11
alexML
left the channel
14:11
anuejn2
joined the channel
14:45
RexOrCine|away
changed nick to: RexOrCine
15:00
comradekingu
left the channel
16:26
se6astian
changed nick to: se6astian|away
16:28
TofuLynx
joined the channel
16:30
TofuLynx
left the channel
16:43
nmdis1999
joined the channel
17:09
sebix
left the channel
17:48
nmdis1999
Good evening everyone! :D
17:48
BAndiT1983
hi nmdis1999
17:48
Bertl_oO
changed nick to: Bertl
17:48
Bertl
evening nmdis1999!
17:48
nmdis1999
Hello BAndiT1983!
17:49
nmdis1999
How was the fare Bertl?
17:49
Bertl
great! thanks for asking!
17:50
nmdis1999
BTW I did some research on how to benefit from cache utilization as you asked :)
17:50
se6astian|away
changed nick to: se6astian
17:50
nmdis1999
Here are the point : 1. Using small data types
17:51
nmdis1999
2. Organizing data to avoid alignment holes 3. Problems caused by standard dynamic memory allocator (we already discussed as I believe)
17:52
se6astian
hi nmdis1999
17:52
alexML_
hello
17:52
nmdis1999
Also, I found a great article on how to optimize data cache access by using transformation
17:52
nmdis1999
Hi alexM_ and sebastian :D
17:57
Bertl
excellent
17:58
nmdis1999
One more point, which I am not sure about though : instead of reading one item per cache line in the inner loop (in case of matrix for example) we can use all of the items
17:58
nmdis1999
for example : itemsPerCacheLine = CacheLineSize/sizeof(elementType)
17:58
nmdis1999
is it a good idea?
17:59
Bertl
there is no problem with cache access, as long as it doesn't trigger writeback or unwanted fetches
17:59
BAndiT1983
what is it about currently?
18:00
Bertl
but this is something you have to test on the actual hardware (of course it helps to be aware of this)
18:01
nmdis1999
Yeah, I had a slight guess about it.
18:01
Kjetil_
I guess this is supposed to run on a PC. So if you want high speed gain from cache optimizations multiple cores should, if possible, work on the same L3/L2 cache lines
18:02
nmdis1999
Kjetil_ : but isn't optimization for cache kind of different L2/L3 cache?
18:03
nmdis1999
Bertl : you asked me to find out which data is used for which visualization, isn't the data same (pixel data?) I have some doubt in it
18:04
Bertl
the source for all will always be the sensel data from the sensor
18:05
Bertl
but the access pattern will be different I guess
18:05
Kjetil_
nmdis1999: hm?
18:05
nmdis1999
Okay :)
18:06
Kjetil_
changed nick to: Kjetil
18:06
Bertl
Kjetil: it is supposed to run on the Axiom Beta
18:06
nmdis1999
I was wondering if the optimization for cache works differently for L2/L3
18:06
Kjetil
Bertl: ah ok
18:06
alexML_
regarding cache, here's a quick exercise anyone can try
18:07
alexML_
remember the qualification task, with lodepng?
18:07
nmdis1999
Yup
18:07
alexML_
take a 4K image (4096x3072) and do some processing on it using two for loops
18:08
Bertl
note that the Cortex A9 in the Zynq doesn't have L2/L3 cache it has an I-cache and a D-cache
18:09
alexML_
let's say for (int y = 0; y < 3072; y++) { for (int x = 0; x < 4096; x++) { do something with im[x + y*pitch] } }
18:09
nmdis1999
Bertl : I didn't know that , thanks!
18:09
Bertl
https://developer.arm.com/products/processors/cortex-a/cortex-a9
18:09
alexML_
write down the execution time, then reverse the order of the two for's
18:09
alexML_
first do the "for x" in the outer loop, then the "for y" in the inner loop
18:09
Kjetil
nmdis1999: I guess is is less relevant as this should run on the Cortex. But, L2/L3 is usually shared between multiple cores. So if multiple cores require the same piece of data, the first core to request it will suffer cache miss penalty. But if the other cores shortly after need the same data they will then hit in the cache.
18:10
nmdis1999
The execution time will be larger when we'll reverse the loops ?
18:11
Bertl
https://mgc-images.imgix.net/esl/Zynq7000_credited-9D62B659.png?q=80&w=1600&fit=max
18:11
Kjetil
(One example is multithreaded matrix multiplication were all cores multiply separate rows with the same column)
18:13
alexML_
yeah; didn't try this on ARM, but on a low-end x86-64 you can easily get a 20-30x slowdown after swapping the loops
18:13
nmdis1999
6.2.1 : https://lwn.net/Articles/255364/
18:14
nmdis1999
It did explained how and why the execution time will be greater, also how we can use itemsPerCacheLine = CacheLineSize/sizeof(elementType) in our benefit :)
18:17
Kjetil
Remember that you should also consider alignment to cache lines when using itemsPerCacheLine trickery
18:19
alexML_
I'd say we shouldn't worry too much about cache at first; these optimizations are best done after getting a working proof of concept, IMO
18:21
Kjetil
If you make sure that the processing parts also can be used on a x86 you can use tools like cachegrind to view cacheperformance. (The caches won't be the same, different sizes, different types, hw prefechers will be different) But it might give you are clue where to start optimising
18:34
nmdis1999
Thanks, Kjetil that helps :D
18:44
nmdis1999
off for now, good night :D
18:45
nmdis1999
left the channel
19:50
BAndiT1983
changed nick to: BAndiT1983|away
20:09
XDjackieXD
left the channel
20:12
XDjackieXD
joined the channel
20:58
BAndiT1983|away
changed nick to: BAndiT1983
21:29
TofuLynx
joined the channel
21:29
TofuLynx
Good Evening!
21:30
TofuLynx
Hello BAndiT1983!
21:30
TofuLynx
Are you on?
21:41
RexOrCine
changed nick to: RexOrCine|away
21:53
parasew[m]
left the channel
21:55
parasew[m]
joined the channel
22:00
Bertl
off for now ... bbl
22:00
Bertl
changed nick to: Bertl_oO
22:10
BAndiT1983
hey TofuLynx, just for a bit
22:10
BAndiT1983
what's the latest state?
22:21
TofuLynx
Hello!
22:21
TofuLynx
I fixed that accident with the class name, I renamed it back to BilinearDebayer.
22:22
TofuLynx
Also, I'm implementing a pattern system, however, I have some questions regarding this
22:23
TofuLynx
From what I saw, it seems that the bilinear debayer functions do also depend on the image pattern, so it isnt enough to have pattern offsets
22:23
TofuLynx
is this true?
22:24
BAndiT1983
offsets are always equal, so it should be enough in fact
22:24
TofuLynx
but, for example , imagine this situation:
22:25
se6astian
changed nick to: se6astian|away
22:25
BAndiT1983
for example RGGB, you wil have R at offset 0, G1 at offset 1, G2 at offset width + 1 and B at width + 2
22:25
TofuLynx
you need to generate a red pixel value for a green0 pixel offset in a RGGB pattern
22:25
TofuLynx
G2 at offset width + 0 and B at width + 1 :)
22:25
BAndiT1983
of course, but you know, that your start point is at index 1, bet ween 2 reds, for example
22:26
TofuLynx
can you explain it better?
22:26
TofuLynx
wait
22:26
BAndiT1983
just like in the linear interpolation, you would start at index 1 of the data, then you wil get red at 0 and at 2
22:26
TofuLynx
I will paste it on lab chat
22:26
BAndiT1983
this is also what my implementation does
22:27
TofuLynx
pastebin*
22:27
BAndiT1983
what about trello, supragy is using it, also nmdis1999
22:27
TofuLynx
https://pastebin.com/Nfvtp8FG
22:27
TofuLynx
check it
22:27
TofuLynx
well I suggested you and g3ggo to use trello, but I think g3ggo couldnt use it or something like that
22:28
BAndiT1983
why do you do the shift as index?
22:28
BAndiT1983
g3gg0 is commenting sometimes on supragyas board, so no problem there
22:28
TofuLynx
shift as index?
22:28
BAndiT1983
by the way, he is absent today, but you were also off for 2 days, hope he will be online tomorrow
22:29
TofuLynx
okk!
22:29
BAndiT1983
width << 1, this looks a bit awkward
22:29
TofuLynx
well, how do you suggest to change to the next row?
22:30
BAndiT1983
borders should be done in linear way, as you just have 2 values there
22:30
BAndiT1983
i would use index = 1 as starting point for red, or whatever color is first in the sensor
22:31
TofuLynx
yep, I am not doing bilinear on borders, except on the down border, for now. I have to avoid it in the four borders
22:31
BAndiT1983
maybe 2 steps, first most data, starting at x 1 and y 1, so you can get it diagonally
22:31
BAndiT1983
afterwards the borders
22:31
TofuLynx
yeah
22:31
TofuLynx
why do you suggest index = 1 as starting point?
22:32
BAndiT1983
it was for general linear interpolation, so you are between 2 reds
22:32
TofuLynx
hmm
22:32
BAndiT1983
but for bilinear you should start at 1,1 or so, so you are in the middle of diagonal cross of pixels, can't remember the patterns curretnly
22:32
BAndiT1983
*currently
22:33
TofuLynx
ah yeah it does start at 1,1 :)
22:33
BAndiT1983
nice examples there -> https://www.semanticscholar.org/paper/Low-cost-Bayer-to-RGB-bilinear-interpolation-with-Pérez-Espeso/9baca257d1f737fc50ba250e0c7fdcf3d7e81f2c
22:33
TofuLynx
ok, andrej, I have a query
22:33
TofuLynx
basically
22:33
TofuLynx
imagine a RGGB pattern
22:33
BAndiT1983
give the loop constant values, without shifting and such, also no calculations, otherwise the loop will execute it at every iteration
22:34
TofuLynx
and you need the red value for the Green0 offset
22:34
TofuLynx
you basically grab the red value at the left and the red value at the right, correct?
22:34
BAndiT1983
yep
22:34
TofuLynx
ok
22:34
TofuLynx
now imagine a GBGR pattern
22:35
TofuLynx
we need a red value for the first green offset
22:35
TofuLynx
you basically grab the red value at down and red value at up, correct?
22:35
BAndiT1983
there is not gbgr, gbrg would it be
22:35
TofuLynx
oops, my bad
22:35
BAndiT1983
otherwise you would have columsn of green
22:36
BAndiT1983
https://github.com/codeplaysoftware/visioncpp/wiki/Example:-Bayer-Filter-Demosaic
22:36
TofuLynx
imagine
22:36
TofuLynx
GRBG
22:36
TofuLynx
we need the red value at the left and the red value at the right, for the first green offset, correct?
22:37
BAndiT1983
right one will get interpolated value
22:37
BAndiT1983
but left one will get half of red on index 1
22:37
TofuLynx
hmm?
22:37
TofuLynx
can you explain it?
22:38
BAndiT1983
my suggestion is to have some sort of balancing of values for now, not an expert, to be honest
22:38
BAndiT1983
you have a value for red on index 1 and required red values for index 0 and 2
22:38
TofuLynx
correct
22:38
BAndiT1983
index 2 is simple, just interpolate between reds on index 1 and 3
22:39
TofuLynx
wait
22:39
BAndiT1983
but as we have a border case here, so index 0 would get half of red on index 1
22:39
TofuLynx
where is the red value at index 3?
22:39
BAndiT1983
ah you have red in the next line, then width + index ;)
22:39
TofuLynx
hmm can you repeat it? xD
22:40
BAndiT1983
it's image d in my link
22:40
BAndiT1983
there are also the cases and according interpolation patterns
22:41
TofuLynx
Hmm I see
22:41
TofuLynx
bookmarked!
22:43
TofuLynx
also
22:43
TofuLynx
how do I replace the (_width << 1) ?
22:45
BAndiT1983
with red offset + 1
22:45
BAndiT1983
also width << 1 shouldn't be used in the loop, but calculated beforehand, it accelerates the processing
22:46
BAndiT1983
https://link.springer.com/content/pdf/10.1186%2Fs13640-017-0196-z.pdf
22:48
TofuLynx
Ok! :)
22:49
BAndiT1983
interpolation in outer areas, like in bggr pattern, requires some sort of weighting, to get the border right
22:49
BAndiT1983
don't know if it visible if you would replicate red value to the missing pixels on the left and top
22:49
danieel
joined the channel
22:51
TofuLynx
Is it noticeable the change from bilinear to linear in this case?
22:52
BAndiT1983
don't think so, but you have to process top and bottom horizontally, left and right vertically
22:52
TofuLynx
yes?
22:53
BAndiT1983
do you want to know it for border areas or for main image?
22:54
TofuLynx
I am trying to understand if is it neccessary some aditional processing for the linear interpolation at the borders
22:54
TofuLynx
also, what do you think about the SetPatternOffsets function? >> https://pastebin.com/v57QrEE6
22:54
BAndiT1983
no, just known pixels / 2 between them
22:55
BAndiT1983
this method should be ok, at first glance
22:55
TofuLynx
ah ok, but you are talking about weighting, what do you mean?
22:55
BAndiT1983
i've had a task for you this days
22:56
BAndiT1983
bggr, if we are looking at the first red pixel, which has 3 unknown pixels on top, top-left and left
22:56
BAndiT1983
then red pixle would have weight of 1, but others would get something like 0.5 or 0.75 of the value
22:57
BAndiT1983
my task would be, as excercise for you, to implement a unit test for downscaler, an example for bayerpreprocessor is already in the repo and also executing succesfully now
22:57
TofuLynx
It's a weighted variation of bilinear interpolation?
22:58
TofuLynx
Hmmm, and do you think we need to test the downscaler for my project?
22:59
BAndiT1983
yes, on one side as excercise, on the other to ensure it's still working if we would do adjustments
22:59
BAndiT1983
also a unit test is faster to execute, instead of an application
23:00
BAndiT1983
bilinear interpolation is also using weighting
23:00
TofuLynx
I thought bilinear interpolation didnt use weighting, as it does simple division
23:01
BAndiT1983
https://en.wikipedia.org/wiki/Bilinear_interpolation
23:01
TofuLynx
hmm I will do the unit test after I finish the patternoffsets
23:02
BAndiT1983
you should try unit tests while working on it, found several places which need adjustments in OC, after implementing bigger test
23:03
TofuLynx
But isnt downscaler in a different pipeline?
23:03
BAndiT1983
http://www.lira.dist.unige.it/teaching/SINA_08-09/papers/demosaicking-JEI-02.pdf
23:04
BAndiT1983
shouldn't be, as every class should be callable without many dependencies
23:04
BAndiT1983
https://pdfs.semanticscholar.org/presentation/a899/335359e089be010bd1c2fc369a1b3e215fe9.pdf
23:04
BAndiT1983
http://www.dmi.unict.it/~battiato/EI_MOBILE0708/Color%20interpolation%20(Guarnera).pdf
23:05
TofuLynx
the lira link doesnt work for me
23:05
BAndiT1983
http://www.cs.tut.fi/kurssit/SGN-16006/kuva/SGN-16006_labwork.pdf
23:05
TofuLynx
nvm
23:05
TofuLynx
now it loade
23:05
TofuLynx
loaded*
23:06
BAndiT1983
alright, off for today, write here or in lab chat, monitoring multiple sources through the day
23:06
TofuLynx
Goodbye!
23:06
TofuLynx
Sleep Well!
23:06
BAndiT1983
have fun and try to start writing a unit test
23:06
BAndiT1983
you too
23:06
BAndiT1983
changed nick to: BAndiT1983|away
23:33
TofuLynx
left the channel
23:36
RexOrCine|away
changed nick to: RexOrCine
00:21
comradekingu
joined the channel