Ubuntu16.04 +cuda8.0+cudnn+opencv+caffe+theano+tensorfl

安装英伟达显卡驱动

首先去官网上查看适合你GPU的驱动

（http://www.nvidia.com/Download/index.aspx?lang=en-us）

[python] view plain copy

sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt-get update
sudo apt-get install nvidia-375（375是你查到的版本号）
sudo apt-get install mesa-common-dev
sudo apt-get install freeglut3-dev

执行完上述后，重启（reboot）。

重启后输入

[python] view plain copy

nvidia-smi

如果出现了你的GPU列表，则说明驱动安装成功了。另外也可以通过，或者输入

[python] view plain copy

nvidia-settings

出现

配置cuda

https://developer.nvidia.com/cuda-downloads

在cuda所在目录打开terminal依次输入以下指令：

[python] view plain copy

sudo dpkg -i cuda-repo-ubuntu1604-8-0-rc_8.0.27-1_amd64.deb
sudo apt-get update
sudo apt-get install cuda

ubuntu的gcc编译器是5.4.0，然而cuda8.0不支持5.0以上的编译器，因此需要降级，把编译器版本降到4.9：

在terminal中执行：

[python] view plain copy

sudo apt-get install gcc -4.9 gcc-5 g++-4.9 g++-5
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.9 20
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-5 10
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-4.9 20
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-5 10
sudo update-alternatives --install /usr/bin/cc cc /usr/bin/gcc 30
sudo update-alternatives --set cc /usr/bin/gcc
sudo update-alternatives --install /usr/bin/c++ c++ /usr/bin/g++ 30
sudo update-alternatives --set c++ /usr/bin/g++

配置cuda8.0之后主要加上的一个环境变量声明，在文件~/.bashrc之后加上

[python] view plain copy

gedit ~/.bashrc
export PATH=/usr/local/cuda-8.0/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

然后设置环境变量和动态链接库，在命令行输入

[python] view plain copy

sudo gedit /etc/profile

在打开的文件里面加上（注意等号两边不能有空格）

[python] view plain copy

export PATH=/usr/local/cuda/bin:$PATH

保存之后，创建链接文件

[python] view plain copy

sudo gedit /etc/ld.so.conf.d/cuda.conf

在打开的文件中添加如下语句：

[python] view plain copy

/usr/local/cuda/lib64

保存退出执行命令行：

[python] view plain copy

sudo ldconfig

使链接立即生效。

2、测试cuda的Samples

命令行输入（注意cuda-8.0是要相对应自己的cuda版本）

[python] view plain copy

cd /usr/local/cuda-8.0/samples/1_Utilities/deviceQuery
make
sudo ./deviceQuery

返回GPU的信息则表示配置成功

3、使用cudnn

上官网下载对应的cudnn

https://developer.nvidia.com/cudnn

下载完cudnn后，命令行输入文件所在的文件夹 (ubuntu为本机用户名)

[python] view plain copy

cd home/ubuntu/Downloads/
tar zxvf cudnn-8.0-linux-x64-v5.1.tgz #解压文件

cd进入cudnn5.1解压之后的include目录，在命令行进行如下操作：

[python] view plain copy

sudo cp cudnn.h /usr/local/cuda/include/ #复制头文件

再cd进入lib64目录下的动态文件进行复制和链接：（5.1.5为对应版本具体可修改）

[python] view plain copy

sudo cp lib* /usr/local/cuda/lib64/ #复制动态链接库
cd /usr/local/cuda/lib64/
sudo rm -rf libcudnn.so libcudnn.so.5 #删除原有动态文件
sudo ln -s libcudnn.so.5.1.5 libcudnn.so.5 #生成软衔接
sudo ln -s libcudnn.so.5 libcudnn.so #生成软链接

4、安装opencv3.1.0

从官网上下载opencv3.1.0

http://OpenCV.org/downloads.html

并将其解压到你要安装的位置，（下载的位置还是在home/ubuntu、Downloads文件夹下）

首先安装Ubuntu系统需要的依赖项，虽然我也不知道有些依赖项是干啥的，但是只管装就行，也不会占据很多空间的。

[python] view plain copy

sudo apt-get install --assume-yes libopencv-dev build-essential cmake git libgtk2.0-dev pkg-config python-dev python-numpy libdc1394-22 libdc1394-22-dev libjpeg-dev libpng12-dev libtiff5-dev libjasper-dev libavcodec-dev libavformat-dev libswscale-dev libxine2-dev libgstreamer0.10-dev libgstreamer-plugins-base0.10-dev libv4l-dev libtbb-dev libqt4-dev libfaac-dev libmp3lame-dev libopencore-amrnb-dev libopencore-amrwb-dev libtheora-dev libvorbis-dev libxvidcore-dev x264 v4l-utils unzip

然后安装opencv需要的一些依赖项，一些文件编码解码之类的东东。

[python] view plain copy

sudo apt-get install build-essential cmake git
sudo apt-get install ffmpeg libopencv-dev libgtk-3-dev python-numpy python3-numpy libdc1394-22 libdc1394-22-dev libjpeg-dev libpng12-dev libtiff5-dev libjasper-dev libavcodec-dev libavformat-dev libswscale-dev libxine2-dev libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev libv4l-dev libtbb-dev qtbase5-dev libfaac-dev libmp3lame-dev libopencore-amrnb-dev libopencore-amrwb-dev libtheora-dev libvorbis-dev libxvidcore-dev x264 v4l-utils unzip

在终端中cd到opencv文件夹下（解压的那个文件夹），然后

[python] view plain copy

mkdir build #新建一个build文件夹，编译的工程都在这个文件夹里
cd build/
cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D WITH_TBB=ON -D WITH_V4L=ON -D WITH_QT=ON -D WITH_OPENGL=ON -DCUDA_NVCC_FLAGS="-D_FORCE_INLINES" ..（后面两点不要忘记）

cmake成功后，会出现如下结果，提示配置和生成成功：

-- Configuring done

-- Generating done

-- Build files have been written to: /home/ise/software/opencv-3.1.0/build

由于CUDA 8.0不支持OpenCV的 GraphCut 算法，可能出现以下错误：

/home/usrname/opencv-3.1.0/modules/cudalegacy/src/graphcuts.cpp:120:54: error: 'NppiGraphcutState' has not been declared

typedef NppStatus (*init_func_t)(NppiSize oSize, NppiGraphcutState** ppStat

/home/usrname/opencv-3.1.0/modules/cudalegacy/src/graphcuts.cpp:135:18: error: 'NppiGraphcutState' does not name a type

operator NppiGraphcutState*()

/home/usrname/opencv-3.1.0/modules/cudalegacy/src/graphcuts.cpp:141:9: error: 'NppiGraphcutState' does not name a type

NppiGraphcutState* pState;

.......

进入opencv-3.1.0/modules/cudalegacy/src/目录，修改graphcuts.cpp文件，将：

#include "precomp.hpp"

#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)

改为

#include "precomp.hpp"

#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER) || (CUDART_VERSION >= 8000)

然后make编译就可以了

[python] view plain copy

make -j8

上面是将opencv编译成功，但是并没有安装到我们的系统中，有很多的设置都没有写入到系统中，因此还要进行install。

[python] view plain copy

sudo apt-get install checkinstall
sudo checkinstall

[python] view plain copy

sudo /bin/bash -c 'echo "/usr/local/lib" > /etc/ld.so.conf.d/opencv.conf'
sudo ldconfig

这里感谢这位朋友的提醒

然后按照提示安装就可以了。使用checkinstall的目的是为了更好的管理我安装的opencv，因为opencv的安装很麻烦，卸载更麻烦，其安装的时候修改了一大堆的文件，当我想使用别的版本的opencv时，将当前版本的opencv卸载就是一件头疼的事情，因此需要使用checkinstall来管理我的安装。执行了checkinstall后，会在build文件下生成一个以backup开头的.tgz的备份文件和一个以build开头的.deb安装文件，当你想卸载当前的opencv时，直接执行dpkg -r build即可。

5、配置caffe环境

切换编译器

选择g++ 5.0以上的对应编号

[python] view plain copy

sudo update-alternatives --config g++

[python] view plain copy

sudo update-alternatives --config gcc

安装依赖库

[python] view plain copy

sudo add-apt-repository universe
sudo apt-get update -y
sudo apt-get install cmake -y

# General Dependencies

[python] view plain copy

sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev \
libhdf5-serial-dev protobuf-compiler -y
sudo apt-get install --no-install-recommends libboost-all-dev -y

# BLAS

[python] view plain copy

sudo apt-get install libatlas-base-dev -y

# Remaining Dependencies

[python] view plain copy

sudo apt-get install libgflags-dev libgoogle-glog-dev liblmdb-dev -y
sudo apt-get install python-dev python-numpy –y
sudo apt-get install -y python-pip
sudo apt-get install -y python-dev
sudo apt-get install -y python-numpy python-scipy

编译 Caffe，cd到要安装caffe的位置

[python] view plain copy

git clone https://github.com/BVLC/caffe.git
cd caffe
cp Makefile.config.example Makefile.config

修改Makefile.config:

[python] view plain copy

gedit Makefile.config

对打开的文件编辑

# cuDNN acceleration switch (uncomment to build with cuDNN).

USE_CUDNN := 1

# Uncomment if you're using OpenCV 3 如果用的是opencv3版本

OPENCV_VERSION := 3

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

WITH_PYTHON_LAYER := 1

在问件里面添加文本由于hdf5库目录更改，所以需要单独添加：

INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/

LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-Linux-gnu/hdf5/serial/

打开makefile文件

[python] view plain copy

gedit Makefile

将

NVCCFLAGS +=-ccbin=$(CXX) -Xcompiler-fPIC $(COMMON_FLAGS)

替换

NVCCFLAGS += -D_FORCE_INLINES -ccbin=$(CXX) -Xcompiler -fPIC $(COMMON_FLAGS)

编辑/usr/local/cuda/include/host_config.h，将其中的第115行注释掉：

[python] view plain copy

sudo gedit /usr/local/cuda/include/host_config.h

将

#error-- unsupported GNU version! gcc versions later than 4.9 are not supported!

改为

//#error-- unsupported GNU version! gcc versions later than 4.9 are not supported!

之后编辑即可

[python] view plain copy

make -j4 all
make -j4 runtest

为了更好地使用pycaffe ，建议安装：

[python] view plain copy

sudo apt-get install python-numpy python-setuptools python-pip cython python-skimage python-protobuf
make pycaffe
cd python
python
import caffe #测试安装成功

到这里Caffe开发环境就配置好了！

可以测试一下，输出AlexNet的时间测试结果：

[python] view plain copy

cd ~/caffe
./build/tools/caffe time --gpu 0 --model ./models/bvlc_alexnet/deploy.prototxt

6、theano安装

1、直接输入命令：

[python] view plain copy

sudo pip install theano

2、配置参数文件：.theanorc

[python] view plain copy

sudo gedit ~/.theanorc

对打开的文件进行编辑

[global]

floatX=float32

device=gpu

base_compiledir=~/external/.theano/

allow_gc=False

warn_float64=warn

[mode]=FAST_RUN

[nvcc]

fastmath=True

[cuda]

root=/usr/local/cuda

3、运行测试例子：

[python] view plain copy

sudo Vim test.py

from theano import function, config, shared, sandbox

import theano.tensor as T

import numpy

import time

vlen = 10 * 30 * 768 # 10 x #cores x # threads per core

iters = 1000

rng = numpy.random.RandomState(22)

x = shared(numpy.asarray(rng.rand(vlen), config.floatX))

f = function([], T.exp(x))

print(f.maker.fgraph.toposort())

t0 = time.time()

for i in range(iters):

r = f()

t1 = time.time()

print("Looping %d times took %f seconds" % (iters, t1 - t0))

print("Result is %s" % (r,))

if numpy.any([isinstance(x.op, T.Elemwise) for x in f.maker.fgraph.toposort()]):

print('Used the cpu')

else:

print('Used the gpu')

可以看到结果：

/usr/bin/python2.7 /home/hjimce/PycharmProjects/untitled/.idea/temp.py

Using gpu device 0: GeForce GTX 960 (CNMeM is disabled, cuDNN not available)

[GpuElemwise{exp,no_inplace}(<CudaNdarrayType(float32, vector)>), HostFromGpu(GpuElemwise{exp,no_inplace}.0)]

Looping 1000 times took 0.302778 seconds

Result is [ 1.23178029 1.61879349 1.52278066 ..., 2.20771813 2.29967761

1.62323296]

Used the gpu

说明安装成功

7、tensorflow 安装

https://github.com/tensorflow/tensorflow/blob/master/tensorflow/g3doc/get_started/os_setup.md

先安装anaconda

https://repo.continuum.io/archive/Anaconda2-4.2.0-Windows-x86_64.exe

上面的地址下载该包默认在downloads里面

[python] view plain copy

cd /home/username/Downloads
sudo bash Anaconda2-4.2.0-Linux-x86_64.sh

配置环境变量

[python] view plain copy

gedit /etc/profile

末尾添上,我是一路yes下来，所以安在了root下，你可以自己选路径，这时候的环境变量要改

export PATH=/root/anaconda2/bin:$PATH

重启

打开终端

[python] view plain copy

python

安装成功

2、创建conda环境名字叫tensorflow

[python] view plain copy

conda create -n tensorflow python=2.7
source activate tensorflow #使能该环境

#下面这句话只能下载给CPU用的tensorflow

[python] view plain copy

conda install -c conda-forge tensorflow

利用pip来下载给GPU用的tensorflow

[python] view plain copy

export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-0.11.0-cp27-none-linux_x86_64.whl

下载安装

[python] view plain copy

pip install --ignore-installed --upgrade $TF_BINARY_URL

安装IPython

[python] view plain copy

conda install ipython

关掉该环境

[python] view plain copy

source deactivate

测试安装是否正确

[python] view plain copy

source activate tensorflow
python

输入

[python] view plain copy

<div ><span >import tensorflow as tf</span></div><div ><span >import numpy as np</span></div><div ><span > </span></div><div ><span ># Create 100 phony x, y data points in NumPy, y = x * 0.1 + 0.3</span></div><div ><span >x_data = np.random.rand(100).astype(np.float32)</span></div><div ><span ># Try to find values for W and b that compute y_data = W * x_data + b</span></div><div ><span >y_data = x_data * 0.1 + 0.3</span></div>
<div ><span ># figure that out for us.)</span></div># (We know that W should be 0.1 and b 0.3, but TensorFlow will
<div ><span >y = W * x_data + b</span></div>W = tf.Variable(tf.random_uniform([1], -1.0, 1.0))
b = tf.Variable(tf.zeros([1]))
<div ><span >optimizer = tf.train.GradientDescentOptimizer(0.5)</span></div># Minimize the mean squared errors.
loss = tf.reduce_mean(tf.square(y - y_data))
train = optimizer.minimize(loss)
<div ><span ># Launch the graph.</span></div># Before starting, initialize the variables. We will 'run' this first.
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
# Fit the line.
for step in range(201):
<div ><span ># Learns best fit is W: [0.1], b: [0.3]</span></div>sess.run(train)
if step % 20 == 0:
print(step, sess.run(W), sess.run(b))
<div ><span > </span></div>