cuda shared library linking: undefined reference to cudaRegisterLinkedBinary

Question

Goal:

1. create a shared library containing my CUDA kernels that has a CUDA-free wrapper/header.
2. create a test executable for the shared l

Answer 1

The other answers did not work for me (maybe because I’m using cuda 10). The solution that worked for me was compiling the cuda files as:

nvcc -dc -o cuda_file.o cuda_file.cu

Than compiling the c++ file as:

g++ -c -o cpp_file.o cpp_file.cpp

And finally linking all using nvcc:

nvcc -o my_prog cpp_file.o cuda_file.o -lcudart -lcuda -L

Don’t take this code literally. But the core of the solution to the error was using nvcc instead of g++ in the final linking step.