windows10下如何进行源码编译安装tensorflow？

1.获取python3.5.x

https://www.python.org/ftp/python/3.5.4/python-3.5.4-amd64.exe

2.安装python3.5.x,默认安装即可，安装后将路径添加到环境变量PATH中

笔者安装后python的路径为:C:\Users\jello\AppData\Local\Programs\Python\Python35

pip路径为:C:\Users\jello\AppData\Local\Programs\Python\Python35\Scripts

3.安装tensorflow的依赖包

pip3 install six numpy wheel

pip3 install keras_applications==1.0.6 --no-deps

pip3 install keras_preprocessing==1.0.5 --no-deps

4.获取MSYS2

http://repo.msys2.org/distrib/x86_64/msys2-x86_64-20180531.exe

5.安装MSYS2

5.1添加路径C:\msys64\usr\bin到PATH环境变量

5.2安装git patch unzip等软件

pacman -S git patch unzip

6.获取Visual C++ Build Tools 2015

https://www.microsoft.com/en-us/download/details.aspx?id=48159

7.安装Visual C++ Build Tools 2015

8.获取Microsoft Visual C++ 2015 Redistributable Update 3

https://www.microsoft.com/en-us/download/details.aspx?id=53587

9.安装Microsoft Visual C++ 2015 Redistributable Update 3 (如果安装无法进行，可能有其它版本，卸载最新版本即可,如安装2013 和2017,那么卸载2017这个版本即可)

10.获取Visual Studio 2015

https://download.microsoft.com/download/b/e/d/bedddfc4-55f4-4748-90a8-ffe38a40e89f/vs2015.3.com_enu.iso

11.安装Visual Studio 2015

12.获取bazel

https://github.com/bazelbuild/bazel/releases (从此链接选择合适的版本即可)

13.安装bazel (依赖MSYS2和Visual C++ Build Tools 2015),安装方法如下

将文件bazel-0.21.0-windows-x86_64重命名为bazel,然后将文件bazel放到C:\msys64\usr\bin目录下

14.获取tensorflow源码

git clone https://github.com/tensorflow/tensorflow.git (如果使用git无法获取源码，提示fatal: BUG: disabling cancellation: Invalid argument，那么直接去github上下载源码压缩包即可)
cd tensorflow
git checkout remotes/origin/r1.13

15.配置tensorflow

python ./configure.py

笔者的配置如下:

C:\Users\jello\tensorflow>python ./configure.py

WARNING: The following rc files are no longer being read, please transfer their contents or import their path into one of the standard rc files:

nul

WARNING: --batch mode is deprecated. Please instead explicitly shut down your Bazel server using the command "bazel shutdown".

INFO: Invocation ID: 89e98bd0-d528-40ac-ac67-87aa0682b890

You have bazel 0.21.0 installed.

Please specify the location of python. [Default is C:\Users\jello\AppData\Local\Programs\Python\Python35\python.exe]:

Found possible Python library paths:

C:\Users\jello\AppData\Local\Programs\Python\Python35\lib\site-packages

Please input the desired Python library path to use. Default is [C:\Users\jello\AppData\Local\Programs\Python\Python35\lib\site-packages]

Do you wish to build TensorFlow with XLA JIT support? [y/N]: n

No XLA JIT support will be enabled for TensorFlow.

Do you wish to build TensorFlow with ROCm support? [y/N]: n

No ROCm support will be enabled for TensorFlow.

Do you wish to build TensorFlow with CUDA support? [y/N]: n

No CUDA support will be enabled for TensorFlow.

Please specify optimization flags to use during compilation when bazel option "--config=opt" is specified [Default is /arch:AVX]:

Preconfigured Bazel build configs. You can use any of the below by adding "--config=<>" to your build command. See .bazelrc for more details.

16.构建tensorflow pip包

方法一:在cmd中构建

16.1 设置变量

set BAZEL_VS=C:\Program Files (x86)\Microsoft Visual Studio 14.0  (这个变量是用来告知bazel，Visual Studio所安装的目录)

16.2 bazel build -c opt --copt=-msse4.1 --copt=-msse4.2 --copt=-mavx --copt=-mavx2 --copt=-mfma --copt=-O3 //tensorflow/tools/pip_package:build_pip_package

16.3 bazel-bin\tensorflow\tools\pip_package\build_pip_package C:/tmp/tensorflow_pkg

方法二:在MSYS2中构建

16.1 启动MSYS2

16.2 禁用MSYS2路径转换

export MSYS_NO_PATHCONV=1
export MSYS2_ARG_CONV_EXCL="*"


16.3 设置环境变量

export PATH="/c/Users/jello/AppData/Local/Programs/Python/Python35:$PATH" (将python的路径加入环境变量PATH中)


16.4 获取tensorflow源码

16.5 bazel build -c opt --copt=-msse4.1 --copt=-msse4.2 --copt=-mavx --copt=-mavx2 --copt=-mfma --copt=-O3 //tensorflow/tools/pip_package:build_pip_package

16.6 bazel-bin\tensorflow\tools\pip_package\build_pip_package C:/tmp/tensorflow_pkg

17.安装tensorflow

pip3 install C:/tmp/tensorflow_pkg/tensorflow-version-cp35-cp35m-win_amd64.whl

18.参考资料

https://www.tensorflow.org/install/source_windows