How to avoid nvcc redeclaration issues?


I have some source files produced automatically from a tool. Basically, every file contains one CUDA kernel. I want to compile these kernels as fatbins and offload them from a host application to the GPU, using the driver API. However, every source file also contains any required standard declarations, besides the kernel function. These declarations cause conflicts during nvcc compilation.

How can I overcome this issue? Does anyone know a way to skip all standard includes during compilation with nvcc?

Thank you in advance.