[Bf-cycles] cycles standalone strange rendering...or code / build bug?

[Wilhelm Landt [The Soulcage Department]]

hi,

managed to compile the .cubin, as in rays 
advise...cmake->cuda_build_cubin=on, setting proper cuda architecture 
(in my case sm_35)
but how do i link/implement the generated cubin directly to the exe?
at the moment the compiler generates the cubins somewhere in .../kernel 
but the exe wants always to start nvcc to generate also the cubin.
sorry for my noob questions...i have no experience with cuda 
compilation/implementation...i just want to get it work with cuda support...

--wil

---
Wilhelm Landt
The Soulcage Department

Herbststr. 31
D-28215 Bremen

Fon +49 (0)421 379 42 87
Fax +49 (0)421 379 42 96
willi at soulcage-department.de

http://www.soulcage-department.de

Am 28.08.2017 um 18:49 schrieb Ray Molenkamp:
> While dynamic compile is somewhat iffy on windows, regular build with pre-made cubins should be no issue.
>
> just be sure that in your cmake config CUDA_BUILD_CUBIN=On and CYCLES_CUDA_BINARIES_ARCH  contains the architectures you require.
>
> --Ray
>
>
> On 8/28/2017 10:34 AM, Wilhelm Landt [The Soulcage Department] wrote:
>> yes, very hard to get a headless compile with cuda. didnt managed yet.
>> with dynload the cuda compiler throws errors like "host compiler target not matching" or something...
>> i compiled everything with vc14 with cuda version 8.
>>
>> ---
>> Wilhelm Landt
>> The Soulcage Department
>>
>> Herbststr. 31
>> D-28215 Bremen
>>
>> Fon +49 (0)421 379 42 87
>> Fax +49 (0)421 379 42 96
>> willi at soulcage-department.de
>>
>> http://www.soulcage-department.de
>>
>> Am 28.08.2017 um 18:26 schrieb Ray Molenkamp:
>>> On 8/28/2017 8:15 AM, Stefan Werner wrote:
>>>> Cuda-binaries instructs the build to compile the CUDA binaries with the rest of cycles, as opposed to doing it on-demand at runtime. I'm not sure if the latter works on Windows at the moment.
>>> Currently broken on windows, D2019 fixes some of it, but even with that applied it'll only work if the environment is setup just properly and both a proper version of msvc and cuda are found in the path.
>>>
>>> --Ray
>>>
>>> _______________________________________________
>>> Bf-cycles mailing list
>>> Bf-cycles at blender.org
>>> https://lists.blender.org/mailman/listinfo/bf-cycles
>>>
>> _______________________________________________
>> Bf-cycles mailing list
>> Bf-cycles at blender.org
>> https://lists.blender.org/mailman/listinfo/bf-cycles
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> https://lists.blender.org/mailman/listinfo/bf-cycles