Build Tensorflow native shared library

.mk

# How to build Tensroflow native shared library

1. Download Tensorflow's source code to dicrectory ${TENSORFLOW_HOME}.

2. Prepare the build environemnt following the instructions from https://www.tensorflow.org/install/install_sources

3. Adding following code into file ${TENSORFLOW_HOME}/tensorflow/core/distributed_runtime/rpc/BUILD

  ```
   cc_binary(
       name = "libgrpc_tensorflow_server.so",
       linkshared = 1,
       srcs = [
           "grpc_tensorflow_server.cc",
       ],
       deps = [
           ":grpc_server_lib",
           "//tensorflow/core:all_kernels",
           "//tensorflow/core:core_cpu",
           "//tensorflow/core:framework_internal",                                                                                                                 
           "//tensorflow/core:lib",
           "//tensorflow/core:protos_all_cc",
           "//tensorflow/core/distributed_runtime:server_lib",
           "@grpc//:grpc++_unsecure",
       ],
   )
  ```

4. Enter dircetory ${TENSORFLOW_HOME}, then run following command

  ```
  bazel build -c opt //tensorflow/core/distributed_runtime/rpc:libgrpc_tensorflow_server.so
  ```
  
5. Then we will get a shared library `libgrpc_tensorflow_server.so` located at ${TENSORFLOW_HOME}/bazel-bin/tensorflow/core/distributed_runtime/rpc

# How to build Tensorflow's JNI shared library

1. Get the source code of Tensorflow on Yarn project from https://github.com/Intel-bigdata/HDL 

2. Build the library using command:

  ```
  g++ -std=c++11 -shared -o libbridge.so -fPIC -I${JDK_HOME}/include -I${JDK_HOME}/include/linux/ -I${TENSORFLOW_HOME}/bazel-genfiles/ -I${TENSORFLOW_HOME} -L${TENSORFLOW_HOME}/bazel-bin/tensorflow/core/distributed_runtime/rpc -lgrpc_tensorflow_server ${HDL_HOME}/hadoop-deeplearning-project/YARN-TensorFlow/tensorflow-bridge/src/main/native/org_tensorflow_bridge_TFServer.cpp ${HDL_HOME}/hadoop-deeplearning-project/YARN-TensorFlow/tensorflow-bridge/src/main/native/exception_jni.cc
  ```
  
  Please note when building against the latest Tensorflow 1.0, protoc 3.2 may be needed in your build environent.