版本對應關係：
bazel 0.15
tensorflow 1.10
cuda 9.2
cudnn 7.3.1
protobuf 3.6.0
nvidia driver
<https://docs.bazel.build/versions/master/install-ubuntu.html>
1. download the Bazel binary installer named bazel-<version>-installer-linux-x86_64.sh from the Bazel releases page on GitHub（https://github.com/bazelbuild/bazel/releases）.
1) wget https://github.com/bazelbuild/bazel/releases/download/0.15.0/bazel-0.15.0-installer-linux-x86_64.sh
2) bash bazel-0.15.0-installer-linux-x86_64.sh
<https://developer.nvidia.com/cuda-downloads>
2. download cuda
1)wget
2)sudo sh cuda_9.2.148_396.37_linux.run 需要安裝driver和cuda
<https://developer.nvidia.com/rdp/cudnn-archive#a-collapse731-92>
3. download cudnn (libcudnn7_7.3.1.20-1+cuda9.2_amd64.deb)
1)註冊賬號
2)下載“Runtime Library”和“Developer Library”連個deb安裝包
3)將上述連個安裝包先後安裝 sudo dpkg -i libcudnn7-dev_7.3.1.20-1+cuda9.2_amd64.deb
4. download TensorFlow
1)git clone https://github.com/tensorflow/tensorflow.git -b r1.10 --depth=1
2)在文件“tensorflow/tensorflow/workspace.bzl”中查看eigen、nsync、protobuf的下載地址，並直接下載這些項目
5. download and install eigen
1)wget https://mirror.bazel.build/bitbucket.org/eigen/eigen/get/fd6845384b86.tar.gz
2)解壓 tar -xvf fd6845384b86.tar.gz
3)mkdir build && cd build && cmake ..
4)make
5)make install
6)將生成目錄添加到CPP路徑
export CPLUS_INCLUDE_PATH="/usr/local/include/eigen3/:$CPLUS_INCLUDE_PATH"
6. download and install nsync
1)wget https://mirror.bazel.build/github.com/google/nsync/archive/1.20.0.tar.gz
2)解壓 tar -xvf 1.20.0.tar.gz
3)將目錄“nsync-1.20.0/public”添加到CPP路徑
export CPLUS_INCLUDE_PATH="$HOME/nsync-1.20.0/public/:$CPLUS_INCLUDE_PATH"
7. download and install protobuf
1)wget https://mirror.bazel.build/github.com/google/protobuf/archive/v3.6.0.tar.gz
2)sudo apt-get install autoconf automake libtool #安裝相關工具
3)./autogen.sh
4)./configure --prefix=$HOME/output/protobuf
5)make
6)make install prefix=$HOME/output/protobuf
export CPLUS_INCLUDE_PATH="$HOME/output/protobuf/include/google:$CPLUS_INCLUDE_PATH"
8. install tensorflow
1)./configure #指定cuda和cudnn的版本，注意gcc指定gcc-5版本，不能太高
2)bazel build //tensorflow:libtensorflow_cc.so #可能需要科學聯網
9. test tensorflow
1)將上個步驟“在bazel-bin/tensorflow下libtensorflow_cc.so和libtensorflow_framework.so”兩個編譯好的動態庫提取出來；
2)下載git代碼“https://github.com/zhangcliff/tensorflow-c-mnist”
3)修改CMakLists.txt中的頭文件目錄，將步驟1的兩個動態庫添加到lib文件夾；
4)在build目錄下，直接cmake ..，然後make，根據錯誤提示進行tf.cpp代碼調整
10. install grpc
1)git clone https://github.com/grpc/grpc.git --depth=1
2)vim .gitmodules (將protobuf的branch指定爲3.6.0)
3)git submodule update --init
4)make
5)sudo make install prefix=$HOME/output/grpc
備註：注：gRpc 源碼中的 Makefile 文件中會自動檢測當前系統是否已經安裝了 protoc，如果沒有安裝，那麼就會自動從其項目中的第三方庫源碼目錄中進行安裝。
11. download tensorflow_serving
1)git clone https://github.com/tensorflow/serving.git --depth=1
12. get 'grpc.pb.h' and 'pb.h' file
protoc -I=$SRC_DIR --cpp_out=$DST_DIR /path/to/file.proto
# 上面的命令會生成xxx.pb.h 和 xxx.pb.cc兩個C++文件。 (數據定義文件)
protoc -I=$SRC_DIR --grpc_out=$DST_DIR --plugin=protoc-gen-grpc=/path/to/grpc_cpp_plugin /path/to/file.proto
# 上面的命令會生成xxx.grpc.pb.h 和 xxx.grpc.pb.cc兩個C++文件。 (接口文件)
舉例：
xupeng@ubantu16:/home/alpha_cpp/Output_linux/include/proto_use$ protoc -I=/home/alpha_cpp/Output_linux/include/ --cpp_out=protoc_output /home/alpha_cpp/Output_linux/include/tensorflow_serving/apis/predict.proto
xupeng@ubantu16:/home/alpha_cpp/Output_linux/include/proto_use$ protoc -I=/home/alpha_cpp/Output_linux/include/ --grpc_out=protoc_output --plugin=protoc-gen-grpc=/home/xupeng/output/bin/grpc_cpp_plugin /home/alpha_cpp/Output_linux/include/tensorflow_serving/apis/predict.proto
參考：
https://www.jianshu.com/p/725c45353c9d
https://tensorflow.google.cn/install/source#docker_linux_builds