CUDA Project Structure

Question

The template and cppIntegration examples in the CUDA SDK (version 3.1) use Externs to link function calls from the host code to the device code.

However, Tom's comment at http://stackoverflow.com/questions/2090974/how-to-separate-cuda-code-into-multiple-files#comment-2024913 indicates that the usage of extern is deprecated.

If this the case, what's the correct structure for a CUDA project such as the template example or cppIntegration example?

Answer 1

Depends what your host code is. If you end up mixing C and C++ you still need the externs. Fore details see this guide.