Installing v2024.3 but cmake is generating an erroneous option --c++17

User discussions
1

GROMACS version: 2024.3
GROMACS modification: No

I am attempting to install cuda-enabled gmx@2024.3 to our shared cluster:

OS: Rocky8
CUDA: NVHPC/24.5 (CUDA RT version 12.4)
Host compiler: GCC/12.3.0
CMake version: 3.30.5
FFTW version: 3.3.10 built with GCC/12.3.0

I am able to configure/cmake, but the generated build system is adding an erroneous flag ‘–c++17’ for NVCC. This would be correct for nvc++, and the flag ‘–std=c++17’ is accepted by nvcc, however this is not what is generated.

I’ve edited the CMakeCache to set the latter ‘–std=c++17’ option manually, and it appears in the list of options, as does the offending ‘–c++17’ which is breaking the build. I observe that CMAKE_CUDA_COMPILER points to (correct path to) nvcc and CMAKE_CXX_COMPILER points to (correct path to) nvc++

The error messages start after ~10% compilation, and look like:

nvcc fatal   : Unknown option '--c++17'
CMake Error at libgromacs_generated_gpuhaloexchange_impl_gpu.cpp.o.Release.cmake:216 (message):
  Error generating
  /apps/fwillmor/gromacs/gromacs-2024.3.gpu/build/src/gromacs/CMakeFiles/libgromacs.dir/domdec/./libgromacs_generated_gpuhaloexchange_impl_gpu.cpp.o

Any insights here would be greatly appreciated!

2

Update: I was able to compile using a spack build of the cuda runtime instead of the nvhpc bundle.