[Bf-cycles] problem with Cycles Multi Device Buffer

Mohamed Sakr 3dsakr at gmail.com
Fri Apr 22 02:00:35 CEST 2016


I'm trying to manually handle the OpenGL drawing, so what I've done:
in session.cpp I changed:
//device_use_gl = ((params.device.type != DEVICE_CPU) &&
device_use_gl = false; //forcing it to run the cpu code instead of gpu
code, to avoid a potential crash with gl map/unmap buffer"

when I use a CUDA device, everything renders fine, writes fine, etc..
when I use a multi CUDA device "GTX 780 + 970" , the device copies data
from device_pointer to data_pointer wrong!

session->background = false;
device->background = true; "because it failed in CUDA GL allocate buffer,
so it set it to true and went to the other path".

image results attached of good and bad image of the same render: "it
renders completely fine with any single device, CPU, CUDA"

it also renders fine ONLY in one case for Multi device:
let render size = 800 * 600 , when tile is large, and a factor of 800 * 600
"so small number of total tiles" , like tile size = 400 * 300, it renders
fine, if the tile size is somehow off, like 399 * 300, black lines begin to

any ideas?

Mohamed Sakr
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20160422/e856a793/attachment.htm 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test_image_multi - bad.jpg
Type: image/jpeg
Size: 17376 bytes
Desc: not available
Url : http://lists.blender.org/pipermail/bf-cycles/attachments/20160422/e856a793/attachment.jpg 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test_image_multi - good.jpg
Type: image/jpeg
Size: 31460 bytes
Desc: not available
Url : http://lists.blender.org/pipermail/bf-cycles/attachments/20160422/e856a793/attachment-0001.jpg 

More information about the Bf-cycles mailing list