This wraps an nvcc, allowing it to be treated as a real C++ compiler with all the usual flags.
Simple run cmake as usual to configure the compiler. The only two flags that need to be set are:
-DCMAKE_CXX_COMPILER=
-DCMAKE_CUDA_COMPILER=
so that the wrapper knows which underlying nvcc and which underlying g++ to use.
This is configured as both a CXX and CUDA project with CMake, which means a valid
CUDA installation must be available.