caffe升级到cuda10.2 + cudnn 8方法

将caffe的环境升级到cuda10.2 + cudnn8时，需进行下列修改，否则无法完成编译。

1）include/caffe/util/device_alternate.hpp

    将头文件#include <driver_types.h>放置在前面，如下： 

#include <driver_types.h>  // cuda driver types

#include <cublas_v2.h>

#include <cuda.h>

#include <cuda_runtime.h>

#include <curand.h>

#ifdef USE_CUDNN  // cuDNN acceleration library.

#include "caffe/util/cudnn.hpp"

2）__shfl_up和__shfl_up修改

    1> src/caffe/layers/reduce.cu

       Line 44 to : shuff = __shfl_down_sync(0xFFFFFFFF, x, offset);

    2> include/contrib/moderngpu/include/device/intrinsics.cuh

       Line 115 to : var = __shfl_up_sync(0xFFFFFFFF, var, delta, width);

       Line 125 to : p.x = __shfl_up_sync(0xFFFFFFFF, p.x, delta, width);

       Line 126 to : p.y = __shfl_up_sync(0xFFFFFFFF, p.y, delta, width)