I’m currently writing a C# wrapper for the driver API.
When I tried to call cuCtxGetDevice(…) my system told me that it is not able to find this entry point in nvcuda.dll. I am using version 6.14.11.6921.
When I used the Dependency Walker from VS I couldn’t find the method as well.
So, what’s wrong?
Thanks for any useful reply in advance!
i am currently writting a java wrapper for the driver api and can confirm this as well. entry point can not be found.
btw, i plan on making the wrapper completely self contained, that is include nvopencc to compile the kernel code directly. however, i couldn’t get nvopencc to compile. is there any way to skip all the cudafe processing and just compile a device kernel directly to a cubin with nvcc? i read the documentation and it’s not entirely clear for me. testing it with nvcc it always needs cl/gcc installed which i want to circumnavigate.
Yes, it is possible:
“$(CUDA_BIN_PATH)\nvcc.exe” -cubin “$(ProjectDir)my_first_kernel.cu” -o “$(ProjectDir)my_first_kernel.cubin”
Ups, did just read your post again. Now that I realized what your real problem is, I can only say: I don’t know, but would also like to know.