Fixed multi-GPU training for Tensor Cores

AlexeyAB

2018-03-09 880cf187d87c904f5fe574802ecff99118643f2d

 src/cuda.h

@@ -35,6 +35,7 @@

#ifdef CUDNN
cudnnHandle_t cudnn_handle();
enum {cudnn_fastest, cudnn_smallest};
#endif

#endif