CUDA compilation

Hi all,

I have a question on CUDA compilation.

Generally, we put both host and device code in a cu file, and then compile it using nvcc, which is the case in the NVIDIA samples. However, I am now trying to compile host and device code separately. Specifically, I have cpp file where the host code is saved and ptx file where the device code is saved. Does anyone know how to compile them to an executable? Many thanks.