Centos 编译TensorFlow C/C++库 libtensorflow.so libtensorflow_framework.so

叁歲伎倆 2022-11-22 12:42 199阅读 0赞

*  安装依赖

yum -y install automake libtool libsysfs libxslt-devel libffi libffi-devel python-devel gcc openssl openssl-devel sqlite-devel

pip install six numpy wheel coverage mock

*  安装CMake

从https://cmake.org/download/选择最新的版本

wget https://github.com/Kitware/CMake/releases/download/v3.14.0-rc4/cmake-3.14.0-rc4-Linux-x86\_64.tar.gz

tar -xvf cmake-3.14.0-rc4-Linux-x86\_64.tar.gz

mv cmake-3.14.0-rc4-Linux-x86\_64 /usr/local/share/cmake

ln -s /usr/local/share/cmake/bin/ccmake /usr/bin/ccmake

ln -s /usr/local/share/cmake/bin/cmake /usr/bin/cmake

ln -s /usr/local/share/cmake/bin/ctest /usr/bin/ctest

*  安装java

yum -y install java-1.8.0-openjdk\*

*  安装Bazel

从https://github.com/bazelbuild/bazel/releases选择合适的版本，具体版本对应关系在https://tensorflow.google.cn/install/source，tensorflow r1.12对应的应是bazel 0.15.0

wget https://github.com/bazelbuild/bazel/releases/download/0.15.0/bazel-0.15.0-installer-linux-x86\_64.sh

sh bazel-0.15.0-installer-linux-x86\_64.sh --prefix=/usr/local/share/bazel

ln -s /usr/local/share/bazel/bin/bazel /usr/bin/bazel

*  下载TensorFlow

git clone https://github.com/tensorflow/tensorflow

或者

git clone https://jhc888007%40163.com:\[password\]@github.com/tensorflow/tensorflow

cd tensorflow

git checkout r1.12

./configure

接下来有一系列设置（这步好像不是必须的？），不懂的可以用默认值

*  编译C库

cd tensorflow

bazel build :libtensorflow.so

在../bazel-bin/tensorflow/生成libtensorflow.so和libtensorflow\_framework.so

\# 问题：

\#  **bazel clean --expunge**

*  编译C++库

cd tensorflow

bazel build :libtensorflow\_cc.so

在../bazel-bin/tensorflow/生成libtensorflow\_cc.so和libtensorflow\_framework.so

*  安装依赖

cd contrib/makefile

./build\_all\_linux.sh

*  安装Eigen

cd downloads/eigen

mkdir build

cd build

cmake -DCMAKE\_INSTALL\_PREFIX=/usr/local/share/eigen ..

make install

cd /usr/local/include

ln -s ../share/eigen/include/eigen3/Eigen

ln -s ../share/eigen/include/eigen3/unsupported

*  安装Protobuf（实践证明对于1.12版本不兼容，行不通）

cd downloads/protobuf

./autogen.sh

./configure --prefix=/usr/local/share/protobuf

make && make install

如果不成功则make clean，再重新make

vim /etc/profile

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/share/protobuf/lib
    export LIBRARY_PATH=$LIBRARY_PATH:/usr/local/share/protobuf/lib
    export PATH=$PATH:/usr/local/share/protobuf/bin
    export PKG_CONFIG_PATH=/usr/local/share/protobuf/lib/pkgconfig/

*  安装Protobuf（手动）

在https://github.com/protocolbuffers/protobuf/releases找到合适版本

wget https://github.com/protocolbuffers/protobuf/releases/download/v3.6.0/protobuf-all-3.6.0.tar.gz

tar -xvf protobuf-all-3.6.0.tar.gz

cd protobuf-3.6.0/

./configure --prefix=/usr/local/share/protobuf

make && make check && make install

vim /etc/profile

*  如果已经有相同系统环境编译好的tensorflow包，则只需要安装依赖环境，然后将已安装的Eigen拷贝到新机器对应位置，然后将编译好的tensorflow包拷贝到python安装目录lib目录下

[Centos 安装TensorFlow][Centos _TensorFlow]

[Centos 编译TensorFlow C/C++库 libtensorflow.so libtensorflow\_framework.so][Centos _TensorFlow C_C_ libtensorflow.so libtensorflow_framework.so]

[Centos C/C++库 libtensorflow.so libtensorflow\_framework.so 测试][Centos C_C_ libtensorflow.so libtensorflow_framework.so]

参考文献：

https://www.cnblogs.com/hrlnw/p/7007648.html

https://www.jianshu.com/p/d46596558640

https://blog.csdn.net/qq\_37674858/article/details/81095101

http://www.cnblogs.com/zjutzz/p/10182099.html

https://blog.csdn.net/u010700335/article/details/69422282/

https://www.cnblogs.com/hrlnw/p/7383951.html

https://blog.csdn.net/qq\_29936933/article/details/82107861

https://blog.csdn.net/laowulong350/article/details/79001511

https://tensorflow.google.cn/install/source\_windows

转载于:https://www.cnblogs.com/jhc888007/p/10518425.html

[Centos _TensorFlow]: https://www.cnblogs.com/jhc888007/p/10524405.html
[Centos _TensorFlow C_C_ libtensorflow.so libtensorflow_framework.so]: https://www.cnblogs.com/jhc888007/p/10518425.html
[Centos C_C_ libtensorflow.so libtensorflow_framework.so]: https://www.cnblogs.com/jhc888007/p/10524922.html