Improve training performance - batch-norm using cuDNN.

AlexeyAB

2018-03-19 537d135feba179636e9bbfe296e078d51f59914c

 src/network_kernels.cu

@@ -121,7 +121,7 @@
   }
#endif
    forward_network_gpu(net, state);
   cudaStreamSynchronize(get_cuda_stream());
   //cudaStreamSynchronize(get_cuda_stream());
    backward_network_gpu(net, state);
}