OK, that must be the problem. I think when installed from a repo all the cuda include files required for compiling (cufft.h, cuda_runtime.h) appear in a standard directory and they're probably specified in an environment variable so g++ and nvcc know where they are; same for the cuda libraries required for linking. See
https://docs.nvidia.com/cuda/cufft/index.html section 2.1.
The easiest way is to reinstall cuda from repo.
Another way (probably better) is to edit the makefile, but could take some digging to get right.
See the makefile, last two clauses: these are for g++ and nvcc compilation respectively. You need to specify for both the include path (similar to how BorisLib and BorisCUDALib are included there), e.g. as:
-I/usr/local/cuda-10.0/targets/x86_64-linux/include
Now you also need to edit the line for the install clause in the makefile where the linking is done. Before the -lcudart and -lcufft flags you need to tell it where to find these libraries, e.g. as (I assume this is where the library directory would be, but you'll need to check):
-L/usr/local/cuda-10.0/targets/x86_64-linux/lib
Regards,
Serban