linux-caffe-gpu 编译注意

caffe支持python2.x

所以这里，重新安装了Ancona2，将环境变量改为

export PATH=/home/devops/anaconda2/bin:$PATH

 

另外 linux自带的python 安装第三模块要使用下面命令来安装

pip install python-numpy

而anconda下则可以使用conda或者pip来安装python的第三方模块

 

1.按照官网的安装依赖库

sudo yum install protobuf-devel leveldb-devel snappy-devel opencv-devel boost-devel hdf5-devel

sudo yum install gflags-devel glog-devel lmdb-devel

yum install atlas-devel

yum install python-devel

sudo yum install openblas-devel

 

2.安装caffe

wget -c https://github.com/BVLC/caffe/archive/1.0.tar.gz

tar zxvf 1.0.tar.gz

cd caffe-1.0

cp Makefile.config.example Makefile.config

 

3.修改文件

编辑 Makefile.config 文件，

修改前

BLAS := atlas

修改后

BLAS := open

 

 

4.编译：

make all

make test

make runtest

make pycaffe

 

编译 pycaffe时报错：fatal error: numpy/arrayobject.h没有那个文件或目录

其实numpy已经是安装的，anaconda2里面有，python中import numpy也没有问题，

这是因为找不到numpy的路径，通过以下python脚本找到numpy的路径，

import numpy as np np.get_include()

解决方法：

sudo vi Makefile.config

添加

\

/home/devops/anaconda2/lib/python2.7/site-packages/numpy/core/include

 

 

5.使用caffe

添加环境变量

sudo vi /etc/profile

 

export PYTHONPATH=/home/devops/caffe-1.0/python:$PYTHONPATH

 

python

import caffe

 

 

6.报错：

ImportError: No module named google.protobuf.internal

解决方法：

sudo chmod 777 -R anaconda2

conda install protobuf

 

 

7.linux本机自带的python要使用caffe

要安装

sudo pip install numpy

sudo pip install scikit-image numpy

 

（1）python2.7和anaconda3的python3.5共存

在配置文件中添加 黑色的部分，即可让linux自带的python2.7和anaconda3的python3.5共存

export PATH=/usr/local/bin:/bin:/usr/bin:/usr/local/sbin:/usr/sbin

export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64\${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}$

export PATH=/home/devops/anaconda3/bin:$PATH

export CUDA_HOME=/usr/local/cuda

export PYTHONPATH=/home/devops/caffe-1.0/python:$PYTHONPATH

alias py27="/usr/bin/python2.7"

alias pyana3="/home/myname/anaconda3/bin/python3.6"

 

这样

使用python和pyana3进入的都是python3.6的环境，,使用py27进入的是是python2.7的环境

 

（2）让python2.7的pip小工具和python3.5 的pip小工具共存

cd /home/devops/anaconda3/bin

mv pip pipconda3

 

cd /home/devops/anaconda2/bin

mv pip pipconda2

 

这样

使用pip list看的就是python2.7的第三方包，使用pipconda3 list 看的就是python3.5的第三方包

 

 

 

 

 

安装caffe-kd

https://github.com/yjxiong/caffe/tree/kd

 

1.cd /home/devops/caffe-kd

2.mkdir build && cd build

yum install cmake 3.cmake .. -DUSE_MPI=ON

 

（1）报错：

CMake Error: The following variables are used in this project, but they are set to NOTFOUND.

Please set them or make sure they are set and tested correctly in the CMake files:

MPI_CXX_LIBRARIES (ADVANCED)

linked by target "caffe" in directory /home/devops/caffe-kd/src/caffe

MPI_HEADER_PATH

解决：

sudo yum install openmpi openmpi-devel

sudo vi /etc/profile

export PATH=/usr/lib64/openmpi/bin/:${PATH}

source /etc/profile

 

（2）报错：Could NOT find Atlas (missing: Atlas_CBLAS_INCLUDE_DIR

  Atlas_CLAPACK_INCLUDE_DIR Atlas_CBLAS_LIBRARY Atlas_BLAS_LIBRARY

  Atlas_LAPACK_LIBRARY)

 

解决方式1：

这种解决方式是使用atlas来编译的，相当的不友好，在之后make install的时候会继续报错

cmake .. -DUSE_MPI=ON .. -DAtlas=/usr/lib64/atlas

 

解决方式2：

优点：这种解决方式是使用openbals来编译的，相比atlas友好太多了，在之后make install的时候可以编译成功

cmake .. -DUSE_MPI=ON -DBLAS=open

修改sudo vi Makefile.config ，将默认的atlas改为open即可

/home/devops/software/OpenBLAS/usr/local/include

/home/devops/software/OpenBLAS/usr/local/lib

 

缺点：是要下载Openblas的安装包，然后编译，并设置环境变量

查看安装的atlas

sudo yum list installed | grep atlas

sudo yum list installed | grep blas

卸载安装的atlas

sudo yum remove atlas.x86_64 atlas-devel.x86_64

OpenBLAS安装OpenBLAS下载链接

下载后，解压，然后安装。 

或者可以使用我用的版本，OpenBLAS–楼主使用的版本 

然后执行下面的指令 

tar -xzvf OpenBLAS.tar.gz 

make

此处会报错：

undefined reference to `LAPACKE_sgesvd'……

undefined reference to `LAPACKE_zlagsy_work'…… 

请忽略报错即可，继续安装即可

make install PREFIX='usr/local/'

配置环境变量

sudo vi /etc/profile

export OpenBLAS_HOME=/home/devops/software/OpenBLAS/usr/local

source /etc/profile

 

 

 

 

4.make && make install

（3）报错：

[ 1%] Building CXX object src/caffe/CMakeFiles/proto.dir/__/__/include/caffe/proto/caffe.pb.cc.o

In file included from /home/devops/caffe-kd/build/include/caffe/proto/caffe.pb.cc:4:0:

/home/devops/caffe-kd/build/include/caffe/proto/caffe.pb.h:12:2: error: #error This file was generated by a newer version of protoc which is

#error This file was generated by a newer version of protoc which is

^

/home/devops/caffe-kd/build/include/caffe/proto/caffe.pb.h:13:2: error: #error incompatible with your Protocol Buffer headers. Please update

#error incompatible with your Protocol Buffer headers. Please update

^

/home/devops/caffe-kd/build/include/caffe/proto/caffe.pb.h:14:2: error: #error your headers.

#error your headers.

^

In file included from /home/devops/caffe-kd/build/include/caffe/proto/caffe.pb.cc:4:0:

/home/devops/caffe-kd/build/include/caffe/proto/caffe.pb.h:23:35: fatal error: google/protobuf/arena.h: No such file or directory

#include <google/protobuf/arena.h>

^

compilation terminated.

make[2]: *** [src/caffe/CMakeFiles/proto.dir/__/__/include/caffe/proto/caffe.pb.cc.o] Error 1

make[1]: *** [src/caffe/CMakeFiles/proto.dir/all] Error 2

make: *** [all] Error 2

 

主要原因：是protobuf 的版本有问题，估计是装的protobuf 的版本太多了，导致不知道找那个了,其实是因为anconda中的protobuf和这里的冲突了

解决方法：把环境变量中的anaconda的路径给注释掉，让其不再排斥

 

 

查看当前protoc版本：protoc --version

查看protoc安装位置：which protoc

查找protoc相关文件：sudo find / -name protoc

找到所有protoc相关文件都删除

 

 

（4）报错：

[1%]BuildingNVCC(Device)object src/caffe/CMakeFiles/cuda_compile.dir/layers/./cuda_compile_generated_normalize_layer.cu.o

/home/devops/caffe-kd/include/caffe/util/cudnn.hpp(107): error: too few arguments in function call

 

1 error detected in the compilation of "/tmp/tmpxft_000069cb_00000000-5_normalize_layer.cpp4.ii".

CMake Error at cuda_compile_generated_normalize_layer.cu.o.cmake:264 (message):

Error generating file

/home/devops/caffe-kd/build/src/caffe/CMakeFiles/cuda_compile.dir/layers/./cuda_compile_generated_normalize_layer.cu.o

 

make[2]: *** [src/caffe/CMakeFiles/cuda_compile.dir/layers/./cuda_compile_generated_normalize_layer.cu.o] Error 1

make[1]: *** [src/caffe/CMakeFiles/caffe.dir/all] Error 2

make: *** [all] Error 2

 

原因：cdunn的v6比v5多一个参数，参照win10-caffe-gpu编译注意

解决方法：但是这种方法会有各种警告，最终编译还是失败

cd /home/devops/caffe-kd/include/caffe/util

vi cudnn.hpp

修改第 107行

原来：

CUDNN_CHECK(cudnnSetConvolution2dDescriptor(*conv,

pad_h, pad_w, stride_h, stride_w, 1, 1, CUDNN_CROSS_CORRELATION));

现在改成：

CUDNN_CHECK(cudnnSetConvolution2dDescriptor(*conv,

pad_h, pad_w, stride_h, stride_w, 1, 1, CUDNN_CROSS_CORRELATION,CUDNN_DATA_FLOAT));

 

然后再重新make && make install即可

 

（5）报错（解决方式1导致的错误）：

[ 91%] Building CXX object tools/CMakeFiles/caffe.bin.dir/caffe.cpp.o

Linking CXX executable caffe

../lib/libcaffe.so: undefined reference to `cblas_sgemv'

../lib/libcaffe.so: undefined reference to `cblas_dgemm'

../lib/libcaffe.so: undefined reference to `cblas_sscal'

../lib/libcaffe.so: undefined reference to `cblas_dgemv'

../lib/libcaffe.so: undefined reference to `cblas_saxpy'

../lib/libcaffe.so: undefined reference to `cblas_ddot'

../lib/libcaffe.so: undefined reference to `cblas_dasum'

../lib/libcaffe.so: undefined reference to `cblas_sgemm'

../lib/libcaffe.so: undefined reference to `cblas_dscal'

../lib/libcaffe.so: undefined reference to `cblas_scopy'

../lib/libcaffe.so: undefined reference to `cblas_sasum'

../lib/libcaffe.so: undefined reference to `cblas_daxpy'

../lib/libcaffe.so: undefined reference to `cblas_dcopy'

../lib/libcaffe.so: undefined reference to `cblas_sdot'

collect2: error: ld returned 1 exit status

make[2]: *** [tools/caffe] Error 1

make[1]: *** [tools/CMakeFiles/caffe.bin.dir/all] Error 2

make: *** [all] Error 2

解决：改错没找到解决方案，坑了，建议放弃atlas方式安装

 

 

 

（6）报错（解决方式2导致的错误）：

-- Could not find OpenBLAS include. Turning OpenBLAS_FOUND off

CMake Error at cmake/Modules/FindOpenBLAS.cmake:53 (MESSAGE):

Could not find OpenBLAS

Call Stack (most recent call first):

cmake/Dependencies.cmake:78 (find_package)

CMakeLists.txt:30 (include)

解决：下载并编译openblas，同时设置openblas的环境变量，使用yum的方式安装openblas是不行的，因为安装的路径无法被caffe找到，从报错信息中查看源码可以看到，只有在以下路径才能被检测到openblas

 

 

（7）报错：

[devops@data-mining-gpu-1 caffe-kd]$ make pycaffe

CXX/LD -o python/caffe/_caffe.so python/caffe/_caffe.cpp

python/caffe/_caffe.cpp:10:31: fatal error: numpy/arrayobject.h: No such file or directory

#include <numpy/arrayobject.h>

 

解决：

查找 numpy所在的目录

sudo find / -name numpy

可以得到路径/usr/lib64/python2.7/site-packages/numpy

和/usr/lib64/python2.7/site-packages/numpy/core/include/numpy

修改配置文件Makefile.config，加入找到的路径

sudo vi Makefile.config

 

报错：

_caffe.so: undefined symbol: _ZN5caffe4mtx_E

解决：

修改 sudo vi Makefile.config

# Uncomment to support layers written in Python (will link against Python libs)

WITH_PYTHON_LAYER := 1

但是要把整个项目删掉，重新编译pycaffe才行，否则还是报这个错