Page 6 of 13

Re: Windows Build FAILED

Posted: Mon Apr 27, 2020 12:57 pm
by Dade
acasta69 wrote: Mon Apr 27, 2020 7:22 am At the moment I am on my laptop and situation is a bit different. This is what I get in Blender console when trying to render the default cube scene with PathOCL + CUDA
I tried and I get the same B.Y.O.B. error: your laptop could have some system wide custom setting, like heaving the nvrtc-builtins64_101.dll path in DLL search list or something like that.

Check: https://docs.microsoft.com/en-us/window ... arch-order

We could use https://docs.microsoft.com/en-us/window ... ldirectory to add the nvrtc-builtins64_101.dll ? May be, it is doable from Python ?

Re: Windows Build FAILED

Posted: Mon Apr 27, 2020 1:43 pm
by Dade
acasta69 wrote: Mon Apr 27, 2020 7:22 am At the moment I am on my laptop and situation is a bit different. This is what I get in Blender console when trying to render the default cube scene with PathOCL + CUDA:
I fixed a bug causing CUDA to use 8 more threads (and related GPU ram) than OpenCL (!). Now it should work fine.

Re: Windows Build FAILED

Posted: Mon Apr 27, 2020 1:49 pm
by acasta69
Dade wrote: Mon Apr 27, 2020 12:57 pm I tried and I get the same B.Y.O.B. error: your laptop could have some system wide custom setting, like heaving the nvrtc-builtins64_101.dll path in DLL search list or something like that.
That can be one problem for sure: if I modify PATH I get different behaviour. However, the other DLL, nvrtc64_101_0.dll, is found without problems by just being in the same directory...
I'll try as soon as possible on the desktop.

Re: Windows Build FAILED

Posted: Mon Apr 27, 2020 1:54 pm
by Dade
acasta69 wrote: Mon Apr 27, 2020 1:49 pm That can be one problem for sure: if I modify PATH I get different behaviour. However, the other DLL, nvrtc64_101_0.dll, is found without problems by just being in the same directory...
This is probably because we are the one to load the main nvrtc64_101_0.dll while it is the nvrtc64_101_0.dll to load nvrtc-builtins64_101.dll. Different settings may apply to the two different DLL loads. The first load probably uses the Python settings while the second may only use the System one :?:

P.S. yes, NVIDIA splitting that DLL in 2 ... /facepalm.

Re: Windows Build FAILED

Posted: Mon Apr 27, 2020 1:57 pm
by B.Y.O.B.
I have pushed a fix: https://github.com/LuxCoreRender/BlendL ... edae1a088f
Does it work on your end?

Re: Windows Build FAILED

Posted: Mon Apr 27, 2020 1:58 pm
by Dade
B.Y.O.B. wrote: Mon Apr 27, 2020 1:57 pm I have pushed a fix: https://github.com/LuxCoreRender/BlendL ... edae1a088f
Does it work on your end?
I will check as soon as the build is ready.

Re: Windows Build FAILED

Posted: Mon Apr 27, 2020 2:23 pm
by Dade
Dade wrote: Mon Apr 27, 2020 1:58 pm
B.Y.O.B. wrote: Mon Apr 27, 2020 1:57 pm I have pushed a fix: https://github.com/LuxCoreRender/BlendL ... edae1a088f
Does it work on your end?
I will check as soon as the build is ready.
It is working fine here.

Re: Windows Build FAILED

Posted: Mon Apr 27, 2020 4:32 pm
by epilectrolytics
Tested latest BlendLuxCore build 70e9438 with LuxCore2.1Benchmark scene:
Works with Cuda, but with OCL during kernel compilation exit without error.

Win10 PC with 2xRTX2070

Re: Windows Build FAILED

Posted: Mon Apr 27, 2020 5:38 pm
by B.Y.O.B.
I'm also getting crashes with regular OpenCL with the latest builds, for example in the danish mood scene. In simple cube-type scenes it works.

Re: Windows Build FAILED

Posted: Mon Apr 27, 2020 6:21 pm
by neo2068
I get errors with latest build and OpenCL and the food demo scene, too, e.g. PATHOCL:clFinish(CL_INVALID_COMMAND_QUEUE) or TILEPATHOCL: clEnqueueReadBuffer(CL_OUT_OF_RESOURCES). With cuda my GTX 780 is 1.5 times faster than the GTX 1080. Is it normal behaviour?