CUDNN版本太低出错+编译PSPNet用Cuda-8.0出现错误。

1 CUDNN版本太低出错的解决办法：

1）将./include/caffe/util/cudnn.hpp 换成最新版的caffe里的cudnn的实现，即相应的cudnn.hpp.

2） 将./include/caffe/layers里的，所有以cudnn开头的文件，例如cudnn_conv_layer.hpp。 都替换成最新版的caffe里的相应的同名文件。

3）将./src/caffe/layer里的，所有以cudnn开头的文件，例如cudnn_lrn_layer.cu，cudnn_pooling_layer.cpp，cudnn_sigmoid_layer.cu。都替换成最新版的caffe里的相应的同名文件。

4）将./src/caffe/util里的，所有以cudnn开头的文件，都替换成最新版的caffe里的相应的同名文件。

2 在编译PSPNet的时候，用Cuda-8.0，出现错误。

错误原因

CUDA 8.0 provides a definition of atomicAdd on double quantities that was not present in previous CUDA toolkits. The code you are working with also apparently provides its own definition/implementation, and this is the source of the error message. The correct fix is to make source code changes to the software in question to make it compatible with CUDA 8.

1）I had to modify the file common.cuh from the pspNet master branch in the following way:

#ifndef CAFFE_COMMON_CUH_
#define CAFFE_COMMON_CUH_
#include <cuda.h>
#if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 600
#else
static __inline__ __device__ double atomicAdd(double *address, double val)         {
unsigned long long int* address_as_ull = (unsigned long long int*)address;
unsigned long long int old = *address_as_ull, assumed;
if (val==0.0)
  return __longlong_as_double(old);
do {
  assumed = old;
  old = atomicCAS(address_as_ull, assumed, __double_as_longlong(val +__longlong_as_double(assumed)));
} while (assumed != old);
return __longlong_as_double(old);
}
#endif
#endif