These are chat archives for buddhi1980/mandelbulber2

29th
Oct 2018
Krzysztof Marczak
@buddhi1980
Oct 29 2018 05:26
@Starmute What doesn't work on macOS?
Starmute
@Starmute
Oct 29 2018 14:10
@buddhi1980 OpenCL with discrete GPU is completely broken on my MacBook Pro. Doesn’t even start rendering before the program crashes.
Starmute
@Starmute
Oct 29 2018 14:15
On bleeding edge build it crashes when open preferences and also crashes with integrated GPU OpenCL
Krzysztof Marczak
@buddhi1980
Oct 29 2018 16:52
Which MacBook Pro do you have?
Starmute
@Starmute
Oct 29 2018 16:53
15 inch, 2018, AMD RX 560X
Starmute
@Starmute
Oct 29 2018 16:55
It’s deprecated but you can still use I’m pretty sure
I’m on 10.13
Krzysztof Marczak
@buddhi1980
Oct 29 2018 16:56
But when some years ago it was planned to kill OpenCL under MacOS, developers of graphics drivers were not interested to maintain OpenCL compiler and could be a lot of bugs
Even older MacOS versios didn't work well with OpenCL
Can you see any errors in mandelbulber_log.txt when it crashes?
Starmute
@Starmute
Oct 29 2018 17:00
Will check in few minutes.
luchansky
@luchansky
Oct 29 2018 18:05
there is the log
Krzysztof Marczak
@buddhi1980
Oct 29 2018 18:09
Does it happen for every settings? By the way can you attach full log file?
This is strange: CL_KERNEL_WORK_GROUP_SIZE, value = 0
luchansky
@luchansky
Oct 29 2018 18:12
Yep all settings, that is full log
Krzysztof Marczak
@buddhi1980
Oct 29 2018 18:14
Set log verbosity level to 2. Close the program. Start again. Press render. Send the log file.
Log started form 22nd second
I need date reported from OpenCL at initialization of the program.
*data
Have you tried to delete mandelbulber.ini file and set OpenCL settings once again?
luchansky
@luchansky
Oct 29 2018 18:25
Yes I've deleted it and tried
Krzysztof Marczak
@buddhi1980
Oct 29 2018 18:30
Thx. I will analyze it.
Krzysztof Marczak
@buddhi1980
Oct 29 2018 18:44
Can you generate log from basic mandelbulb? Just start the program and press render.
luchansky
@luchansky
Oct 29 2018 19:19
new log
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:18
You forgot to enable OpenCL when you generated last log file
Starmute
@Starmute
Oct 29 2018 20:19
Oh, thought you wanted no OCL as control
Will try with OCL. This one uses integrated GPU, which works. Was that what you wanted?
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:20
Send log file with rendering basic mandelbulb (not any sophisticated settings) made on device which crashes the program. You can also send log file from working device, just to compare,
luchansky
@luchansky
Oct 29 2018 20:23
This one hangs (not usual crash)
I actually can't get it to even start rendering using iGPU with normal mbulb
CVMcompiler uses 100% cpu and hangs there
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:28
which one is iGPU? Intel or AMD?
Starmute
@Starmute
Oct 29 2018 20:29
Intel
UHD Graphics 630
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:29
What if you try with AMD?
Starmute
@Starmute
Oct 29 2018 20:29
Same thing. With my more sophisticated settings the I GPU will render
dGPU crashes on those
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:29
Intel GPU is mostly not capable to run such a big rendering kernels
Starmute
@Starmute
Oct 29 2018 20:30
I know
Only reason I used it is because AMD would not work.
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:30
What is in the log when you run basic mandelbulb on AMD GPU?
By the way when you did full restart of the system?
luchansky
@luchansky
Oct 29 2018 20:32
Basic mbulb / AMD GPU
Last full restart was few minutes ago :)
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:33
OK. Thx
It takes 36s to compile the OpenCL kernel? !!!
luchansky
@luchansky
Oct 29 2018 20:33
No idea why.
On Windows (Bootcamp) it's much faster to compile.
But I think not as efficient as it would be running native on macOS
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:38
Have you tried "Fast" OpenCl mode on AMD? Does it work?
luchansky
@luchansky
Oct 29 2018 20:39
Waits few seconds then crashes.
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:41
In general it looks like a bug in AMD OpenCL compiler which is in the system
PID: 2287, time: 51.295352570, OpenCl kernel program successfully compiled
PID: 2287, time: 51.295486855, cOpenClEngineRenderFractal: Opencl build time [s], value = 36.5591
Program is compiled succesfuly, but it takes horrible time
luchansky
@luchansky
Oct 29 2018 20:42
But then it crashes after compiling
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:42
Then it reports:
PID: 2287, time: 51.295670912, CL_KERNEL_WORK_GROUP_SIZE, value = 0
PID: 2287, time: 51.295706271, CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, value = 0
luchansky
@luchansky
Oct 29 2018 20:42
wtf
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:43
it means that maximum OpenCL workgroup size is ZERO!
It menas that it can render zero pixels in one single job
This is reason why it crashes. It tries to allocate zero bytes for output buffer, because it expects zero rendered pixels.
luchansky
@luchansky
Oct 29 2018 20:45
Why would it allocate zero bytes?
Can we hardcode value for testing purposes ?
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:46
But I don't know how much memory should be allocated, when is reported wrong workgroup size
luchansky
@luchansky
Oct 29 2018 20:47
On windows on same machine it recommends 32XX MB
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:47
It is completely different thing
luchansky
@luchansky
Oct 29 2018 20:47
hm, ok
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:49
I see that it reports more wrong numbers:
PID: 2287, time: 6.916951825, CL_DEVICE_GLOBAL_MEM_SIZE, value = 0
luchansky
@luchansky
Oct 29 2018 20:51
wtf !
Krzysztof Marczak
@buddhi1980
Oct 29 2018 20:51
PID: 2287, time: 6.923492254, CL_DEVICE_MAX_WORK_GROUP_SIZE, value = 256
This is also quite low. Normally it is 512 or 1024
luchansky
@luchansky
Oct 29 2018 20:51
On windows side it's generally 256 so maybe oddity of this gpu?