Latest build: crash while building kernels

[epilectrolytics]

Would like to test, but latest BlendLuxCore build 41da08d is crashing on my WinPC (exit while building kernels).

[B.Y.O.B.]

Are you compiling Cuda or OpenCL kernels?
What's your GPU model?

I am getting the following error when compiling Cuda kernels, I'll look into it:

Code: Select all

[LuxRays][22.906] [PathOCL kernel] CUDA program compilation error:
materialdefs_funcs_disney.cl(309): error: no suitable constructor exists to convert from "float" to "float3"

1 error detected in the compilation of "PathOCL kernel".

[epilectrolytics]

Win10PC with Ryzen 3900X and 2x RTX 2070, cuda kernel compilation.
I'm getting exits without error message with PathOCL cuda.

CPU rendering works, but light tracing is way slower than before and sky background is broken (too dark).

EDIT:
OCL compilation works.
CUDA compilation with only one device works without crash, but same "float3" error than posted above.

[Dade]

Are you compiling Cuda or OpenCL kernels?
What's your GPU model?

I am getting the following error when compiling Cuda kernels, I'll look into it:

Code: Select all

[LuxRays][22.906] [PathOCL kernel] CUDA program compilation error:
materialdefs_funcs_disney.cl(309): error: no suitable constructor exists to convert from "float" to "float3"

1 error detected in the compilation of "PathOCL kernel".

In CUDA you have to explicit convert between float3 <=> float (not in OpenCL). I assume it is the new thin film code, just compile with CUDA add MAKE_FLOAT3() where required.

[B.Y.O.B.]

I have already pushed a fix: https://github.com/LuxCoreRender/LuxCor ... 907869fe55
I hope the float3 + float addition in the following line works in both OpenCL and Cuda.

[B.Y.O.B.]

@epilectrolytics check the new build.

[epilectrolytics]

@epilectrolytics check the new build.

Yep, works now, thanks for instant fixing!

BTW there is still a crash with the default state of the CUDA version which has all devices activated in both modes, CUDA and OCL.
Only when one branch is deactivated it works.

This leads to constant crashing with old OCL files if not the devices are deactivated partly before (viewport)rendering.
If possible, a CUDA-only activation in the CUDA version of BlendLuxCore would be more adequate.