[SOLVED] NVCC defaults to 32-bit compilation

Win10, CMake 3.10.1, VS2017, attempting to build gFluidSurface.
CMake generates sln/prj files ok but defaults to “WIN32” on all build types.
Selected “x64” within VS but still fails with:

Generating fluid_system_cuda.ptx
nvcc fatal : 32 bit compilation is only supported for Microsoft Visual Studio 2013 and earlier.

I don’t know where to change this to 64-bit.

Thank you.

UPDATE: Instead of trying to build libgvdb, I have reinstalled the binaries via the Windows Installer. The only issue now is that the executable samples require cudart64_80.dll. All well and good, but this particular dll is a legacy and not included in CUDA 9.1. Thus to obtain said legacy dll one must download the 1.3 GB CUDA 8.0.61 and Express Install for a 360kb dll. This seems inefficient. :)

When using cmake, be sure that you specify the compiler and use the 64-bit version (in the compiler list).
That should automatically generate a build for that platform config, including nvcc.

Note that the Windows Installer does not have the very latest updates, and it provided for only a single build configuration (CUDA 8, VS 2015). If you need more variation in build setup use the github source and cmakes.


Ohhh…THAT 64-bit button! Oops. Sorry. I never even looked at that part of CMake. Thank you!

–Michael Prostka