ubuntu16.04 从源码安装tensorflow-gpu r1.4版本---2017-12-07

声明：以下文章基于https://www.cnblogs.com/lindaxin/p/7797477.html修改，感谢原作者

参考文章：https://www.tensorflow.org/install/install_sources，可以首先按照其中的说明安装依赖项，具体步骤参考下面

本文假设已安装好cuda 8.0和cudnn 6.0,安装参考文章：https://zhuanlan.zhihu.com/p/27890924

1、安装bazel
参考官网这里https://docs.bazel.build/versions/master/install-ubuntu.html#install-on-ubuntu,只有三步，很简单

 2、安装第三方库

在terminal中输入以下命令，务必参考官方教程：https://www.tensorflow.org/install/install_sources

1 2	sudo apt- get  install python-numpy swig python-dev python-wheel #安装第三方库 sudo apt- get  install git

 3、安装tensorflow

在terminal中输入以下命令，下载tensorflow源码

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

 在terminal中输入以下命令：执行配置文件，对安装步骤进行配置

1 2	cd ~/tensorflow #切换到tensorflow文件夹 ./configure #执行configure文件

由于一直在更新， 下面的配置步骤是2个月之前的，不一定完整，具体可搜索相关信息，一般遇到不太认识的就选n

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Do you wish to use jemalloc  as  the malloc implementation? [Y/n] y jemalloc enabled Do you wish to build TensorFlow with Google Cloud Platform support? [y/N] n No Google Cloud Platform support will be enabled  for  TensorFlow Do you wish to build TensorFlow with Hadoop File System support? [y/N] n No Hadoop File System support will be enabled  for  TensorFlow Do you wish to build TensorFlow with the XLA just- in -time compiler (experimental)? [y/N] n No XLA JIT support will be enabled  for  TensorFlow Found possible Python library paths:    /usr/lib/python2.7/site-packages    /usr/lib64/python2.7/site-packages Please input the desired Python library path to use.  Default  is  [/usr/lib/python2.7/site-packages]   Using python library path: /usr/lib/python2.7/site-packages Do you wish to build TensorFlow with OpenCL support? [y/N] n No OpenCL support will be enabled  for  TensorFlow Do you wish to build TensorFlow with CUDA support? [y/N] y CUDA support will be enabled  for  TensorFlow Please specify which gcc should be used  by  nvcc  as  the host compiler. [Default  is  /usr/bin/gcc]: Please specify the CUDA SDK version you want to use, e.g. 7.0. [Leave empty to use system  default ]: 8.0 Please specify the location  where  CUDA 8.0 toolkit  is  installed. Refer to README.md  for  more details. [Default  is  /usr/local/cuda]: /usr/local/cuda-8.0

 4、创建pip,此步骤是从源码编译生成tensorflow的whl文件，可选

在tensorflow的根目录下，在terminal中输入以下命令：

1 2 3 4

bazel build -c opt  //tensorflow/tools/pip_package:build_pip_package bazel build -c opt --config=cuda  //tensorflow/tools/pip_package:build_pip_package bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg   可以在/tmp/tensorflow_pkg目录中找到生成的whl文件

5、从源码安装tensorflow环境

1 2 3 4 5 6 7 8

bazel build -c opt  //tensorflow/tools/pip_package:build_pip_package   # To build with GPU support: bazel build -c opt --config=cuda  //tensorflow/tools/pip_package:build_pip_package mkdir _python_build cd _python_build ln -s ../bazel-bin/tensorflow/tools/pip_package/build_pip_package.runfiles/org_tensorflow/* . ln -s ../tensorflow/tools/pip_package/* . python setup.py develop

 6、tensorflow测试

 

1 2 3 4 5 6 7 8 9 10 11 12

$ python   >>> import tensorflow  as  tf >>> hello = tf.constant( 'Hello, TensorFlow!' ) >>> sess = tf.Session() >>> print sess.run(hello) Hello, TensorFlow! >>> a = tf.constant(10) >>> b = tf.constant(32) >>> print sess.run(a+b) 42 >>>

 大功告成

出现的错误
操作

1	bazel build -c opt --config=cuda  //tensorflow/cc:tutorials_example_trainer

 报错

1 2 3 4 5 6 7 8 9 10

ERROR: /home/yaroslavvb/tensorflow.git/tensorflow/tensorflow/core/kernels/BUILD:1080:1: undeclared inclusion(s)  in  rule  '//tensorflow/core/kernels:cwise_op_gpu' : this  is  missing dependency dependency  for  following files included  by  'tensorflow/core/kernels/cwise_op_gpu_floor.cu.cc' :    '/usr/local/cuda-8.0/include/cuda_runtime.h'    '/usr/local/cuda-8.0/include/host_config.h'    '/usr/local/cuda-8.0/include/builtin_types.h'    '/usr/local/cuda-8.0/include/device_types.h'    '/usr/local/cuda-8.0/include/host_defines.h'    '/usr/local/cuda-8.0/include/driver_types.h'    '/usr/local/cuda-8.0/include/surface_types.h'    '/usr/local/cuda-8.0/include/texture_types.h'

 可以进入tensorflow/third_party/gpus/crosstool/目录，打开CROSSTOOL文件，搜索cxx_builtin_include_directory，应该有三行，在下面添加行如下 
cxx_builtin_include_directory: "/usr/local/cuda-8.0/include" 

如果出现的错误是类似的，只要将cxx_builtin_include_directory: "/usr/local/cuda-8.0/include"的文件路径改一下就可以了，亲测有效

再次运行上一步的命令，应该就没问题了。