These are chat archives for buddhi1980/mandelbulber2

10th
Sep 2017
Krzysztof Marczak
@buddhi1980
Sep 10 2017 15:36
I have solved problem qith missing qDebug() output. I was system configuration problem. It returned working when I deleted /etc/xdg/QtProject/qtlogging.ini
Sebastian Jennen
@zebastian
Sep 10 2017 19:23
Thats good news. The debug channel is probably deactivated by default, since it is only to meant to be used by developers, i guess.
Krzysztof Marczak
@buddhi1980
Sep 10 2017 19:24
Probably you are right. qWarning() and qCritical() worked properly.
Sebastian Jennen
@zebastian
Sep 10 2017 19:25
hmm, what about qInfo()?
Krzysztof Marczak
@buddhi1980
Sep 10 2017 19:25
We need to change all temporary qDebug() outputs to WriteLog()
I haven't tested qInfo
Sebastian Jennen
@zebastian
Sep 10 2017 19:25
The problem about qInfo was that in old versions of qt it did not work properly
At least that is what i remember
Anyway it would be rasonable to have one "consistent" logging mechanism, best to use our own WriteLog
BTW: What are the next steps you want to do in the project?
Krzysztof Marczak
@buddhi1980
Sep 10 2017 19:28
I think it's good time to start cleaning up the code. DOF effects are almost done.
We need to add more error checking to the code to make all more robust
Sebastian Jennen
@zebastian
Sep 10 2017 19:30
Yes, i also encountered some crashes and shutdowns, mostly some race conditions about new opencl code.
I will try to find a way to reproduce and investigate with debugger whats wrong.
Krzysztof Marczak
@buddhi1980
Sep 10 2017 19:31
One actual issue is to use GPU for rendering DOF and SAAO using GPU. On most of GFX cards it's only possible to allocate 512MB per memory block. When image resolution is high (like 6400x4800) rendering of DOF fails.
Sebastian Jennen
@zebastian
Sep 10 2017 19:31
I also could reproduce the problem with julia vector copy: buddhi1980/mandelbulber2#352 but i could not identify the error
Krzysztof Marczak
@buddhi1980
Sep 10 2017 19:32
I'm going to predict memory usage and switch to CPU rendering when would be not possible with GPU
Sebastian Jennen
@zebastian
Sep 10 2017 19:33
how much of a pixel range do you need to calculate these effects? maybe its possible to make these effects tile based to save memory. That would maybe only require little overhead of copying overlapping areas
Krzysztof Marczak
@buddhi1980
Sep 10 2017 19:33
i will look at issue #352. It's also reproducible for me
these effects have to have access to full image, so it's not possible to divide image
Sebastian Jennen
@zebastian
Sep 10 2017 19:34
#352 is probably a use after free, maybe the parameter stuff interferes with the preview below the julia parameters...
Krzysztof Marczak
@buddhi1980
Sep 10 2017 19:37
So as you see now we should focus on debugging and cleaning up. Actual state of development is enough for new release. Most of effects and fractal calculations already works. More will came in v.2.13
Preparation for release will take probably more than month, because there was a lot of major changes. We need to check all old functions (like calculation of fractal formulas, which are now 4D), bring back compatibility with older versions, and make application more stable.
Sebastian Jennen
@zebastian
Sep 10 2017 19:46
we will also need to test that queue / cli / netrender is working as expected with and without opencl (at least not crash, when opencl in not yet supported).
I will also add stuff to the test.cpp so we can check this more frequently, though this will need a lot of manual testing
Sebastian Jennen
@zebastian
Sep 10 2017 19:55
There are two points which are a little controversial:
  • installation maybe do with make install (needs to be added to pro file)
  • recombine mandelbulber.pro and mandelbulber-opencl.pro
    see also here: buddhi1980/mandelbulber2#345
qmake and cmake take also instructions for installation, which could enable putting the shared data to the target location. but this will confuse existing package managing.
Also IMHO we can remove mandelbulber-opencl.pro again, and make the opencl inclusion based on a check, like the other optional packages. In the end its still not harmful for someone who buillds mandelbulber with opencl support and with no proper opencl setup, as long he does not enable the opencl in his settings.
Krzysztof Marczak
@buddhi1980
Sep 10 2017 19:59
I want to have install option for make. If we need to do any chages in directory structure, we have to do them.
Sebastian Jennen
@zebastian
Sep 10 2017 20:01
enabling make install with qmake should not be to hard and should work with existing structure, see here for more info: http://doc.qt.io/qt-5/qmake-advanced-usage.html
for cmake it should be possible as well: https://stackoverflow.com/a/11097012