libcudnn (R5) not found in library path

环境：Ubuntu 18.04 +  Torch7 + cuda10

在运行使用cudnn的lua程序的时候产生错误：

/home/majiabiao/torch/install/bin/luajit: /home/majiabiao/torch/install/share/lua/5.1/trepl/init.lua:389: /home/majiabiao/torch/install/share/lua/5.1/trepl/init.lua:389: /home/majiabiao/torch/install/share/lua/5.1/cudnn/ffi.lua:1603: 'libcudnn (R5) not found in library path.
Please install CuDNN from https://developer.nvidia.com/cuDNN
Then make sure files named as libcudnn.so.5 or libcudnn.5.dylib are placed in
your library load path (for example /usr/local/lib , or manually add a path to LD_LIBRARY_PATH)

Alternatively, set the path to libcudnn.so.5 or libcudnn.5.dylib
to the environment variable CUDNN_PATH and rerun torch.
For example: export CUDNN_PATH="/usr/local/cuda/lib64/libcudnn.so.5"

stack traceback:
    [C]: in function 'error'
    /home/majiabiao/torch/install/share/lua/5.1/trepl/init.lua:389: in function 'require'
    test.lua:41: in main chunk
    [C]: in function 'dofile'
    ...biao/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
    [C]: at 0x555bad5c4570

去我安装的cuda文件夹下查看：/usr/local/cuda-10.1/lib64，发现没有 libcudnn.so.5 or libcudnn.5.dylib而是libcudnn.so.7 libcudnn.so.7.5.0

根据错误信息查看文件/home/majiabiao/torch/install/share/lua/5.1/cudnn/ffi.lua

local libnames = {'libcudnn.so.5', 'libcudnn.5.dylib', 'cudnn64_5.dll'}
    local ok = false
    for i=1,#libnames do
        ok = pcall(function () cudnn.C = ffi.load(libnames[i]) end)
        if ok then break; end
    end

    if not ok then

将'libcudnn.so.5', 'libcudnn.5.dylib'修改为'libcudnn.so.7', 'libcudnn.7.dylib'结果变为以下错误

/home/majiabiao/torch/install/bin/luajit: /home/majiabiao/torch/install/share/lua/5.1/trepl/init.lua:389: /home/majiabiao/torch/install/share/lua/5.1/trepl/init.lua:389: /home/majiabiao/torch/install/share/lua/5.1/cudnn/ffi.lua:1619: These bindings are for CUDNN 5.x (5005 <= cudnn.version > 6000) , while the loaded CuDNN is version: 7500  
Are you using an older or newer version of CuDNN?
stack traceback:
    [C]: in function 'error'
    /home/majiabiao/torch/install/share/lua/5.1/trepl/init.lua:389: in function 'require'
    test.lua:41: in main chunk
    [C]: in function 'dofile'
    ...biao/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
    [C]: at 0x561e50b7e570

那么之前那个找不到libcudnn.so.5 or libcudnn.5.dylib的错误的原因是不兼容某些版本的cudnn，只兼容5005到6000之间的版本

那么如果使用的是不兼容版本的cudnn，那么就在现有cuda的基础上更改cudnn的版本，只需要替换一些文件

去nvidia官网下载将要使用的版本的cudnn，解压出来的文件夹cuda中有两个文件夹include和lib64

#删除原有版本的cudnn（注意你的文件夹是否是这个路径）
sudo rm -rf /usr/local/cuda/include/cudnn.h
sudo rm -rf /usr/local/cuda/lib64/libcudnn*
#将下载好的新的版本的cudnn文件放到刚才删除的文件所在位置：
sudo cp include/cudnn.h /usr/local/cuda/include/
sudo cp lib64/lib* /usr/local/cuda/lib64/
#cd到/usr/local/cuda/lib64/文件夹下，建立软链接（注意版本号换成你自己的）
sudo chmod +r libcudnn.so.5.0.5  
sudo ln -sf libcudnn.so.5.0.5 libcudnn.so.5  
sudo ln -sf libcudnn.so.5 libcudnn.so  
sudo ldconfig  

之后就可以了

转载于:https://www.cnblogs.com/biaoJM/p/10810263.html