Build CUDAExtension with '-03'. Maybe this helps?

setup.py

@@ -27,7 +27,7 @@ if __name__ == "__main__":
               CUDAExtension(module_name,
                             [generated_file] + cuda_sources,
                             extra_compile_args={'cxx': [],
-                                                'nvcc': ['-arch=sm_35']})
+                                                'nvcc': ['-arch=sm_35', '-O3']})
           ],
           cmdclass={
               'build_ext': BuildExtension