Note on the compilation difference compared to other samples:

• In this sample we need to add -Xcompiler /MD directive to nvcc compile (see build.bat), otherwise the linking fails.
• The reason of the error is that the CUDA code is compiled with static C Runtime Library, whereas Cython code is compiled with dynamic C Runtime Library causing the conflict during linking.
• The conflict is caused by Numpy dependency. If such dependency is removed, the sample builds just fine.