/usr/bin/ld: cannot find -lcuda

Hi all,

I don’t know if it is the issue but I don’t have root access so I install CUDA in my home directory. Then I added those paths to PATH and LD_LIBRARY_PATH:

export PATH="/my_local_path/cuda/bin:$PATH"
export LD_LIBRARY_PATH="/my_local_path/cuda/lib64:/my_local_path/cuda/lib:$LD_LIBRARY_PATH"
export CFLAGS=-I/my_local_path/cudnn/include
export LDFLAGS=-L/my_local_path/cudnn/lib64

I tried to install chainer from source and by pip install chainer --no-cache-dir -vvvv but they all failed to build with cuda:

Bellow are logs when I built chainer from source. Pip install outputs similar logs.

Options: {'profile': False, 'annotate': False, 'linetrace': False, 'no_cuda': False}
Include directories: ['/users/ud2017/hoavt/cuda/include']
Library directories: ['/users/ud2017/hoavt/cuda/lib64', '/users/ud2017/hoavt/cuda/lib']
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++ [enabled by default]
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++ [enabled by default]
/usr/bin/ld: cannot find -lcuda
collect2: error: ld returned 1 exit status
Cannot build a stub file.
Original error: command 'c++' failed with exit status 1
**************************************************
*** WARNING: Cannot link libraries: ['cublas', 'cuda', 'cudart', 'curand', 'nvToolsExt']
*** WARNING: Skip installing cuda support
*** WARNING: Check your LDFLAGS environment variable
**************************************************
Include directories: ['/users/ud2017/hoavt/cuda/include']
Library directories: ['/users/ud2017/hoavt/cuda/lib64', '/users/ud2017/hoavt/cuda/lib']
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++ [enabled by default]
/tmp/tmpHcwk2p/a.cpp:1:19: fatal error: cudnn.h: No such file or directory
 #include <cudnn.h>
                   ^
compilation terminated.
command 'x86_64-linux-gnu-gcc' failed with exit status 1
**************************************************
*** WARNING: Include files not found: ['cudnn.h']
*** WARNING: Skip installing cudnn support
*** WARNING: Check your CFLAGS environment variable
**************************************************
Include directories: ['/users/ud2017/hoavt/cuda/include']
Library directories: ['/users/ud2017/hoavt/cuda/lib64', '/users/ud2017/hoavt/cuda/lib']
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++ [enabled by default]
/tmp/tmpAvadWz/a.cpp:1:18: fatal error: nccl.h: No such file or directory
 #include <nccl.h>
                  ^
compilation terminated.
command 'x86_64-linux-gnu-gcc' failed with exit status 1
**************************************************
*** WARNING: Include files not found: ['nccl.h']
*** WARNING: Skip installing nccl support
*** WARNING: Check your CFLAGS environment variable
**************************************************
Include directories: ['/users/ud2017/hoavt/cuda/include']
Library directories: ['/users/ud2017/hoavt/cuda/lib64', '/users/ud2017/hoavt/cuda/lib']
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++ [enabled by default]
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++ [enabled by default]
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++ [enabled by default]
running instal

It took me two days to fix this issue without any success. I would really appreciate any suggestion. I want to use chainer for my project but without cuda support, it is quite slow.

OS: Ubuntu 14.04 gcc 4.8.4 python 2.7.6 Cuda compilation tools, release 7.0, V7.0.27 cudnn:

define CUDNN_MAJOR 4 define CUDNN_MINOR 0 define CUDNN_PATCHLEVEL 7

define CUDNN_VERSION (CUDNN_MAJOR * 1000 + CUDNN_MINOR * 100 + CUDNN_PATCHLEVEL)

[Paosder]

@hoavt-54, did you put -I or -L characters in front of your cudnn environment variables? (LD_LIBRARY_PATH, CFLAGS…)

i.g.

export CFLAGS=-I/Developer/NVIDIA/CUDA-8.0/include
export LDFLAGS=-L/Developer/NVIDIA/CUDA-8.0/lib

Thank you @Paosder yes I did. I updated my first comment.

And you should also check that your cudnn path has -L characters(for lib directory). LD_LIBRARY_PATH=-L/Developer/NVIDIA/CUDA-8.0/lib:$LD_LIBRARY_PATH And you don’t have to put cuda path into LDFLAGS.

In my case, in MacOS, I put these characters so that I could fixed these link errors.

[unnonouno]

No, usually a liker command uses $LIBRARY_PATH. $LD_LIBRARY_PATH is used only on runtime by dlopen system call. See http://man7.org/linux/man-pages/man3/dlopen.3.html