Follow the Installing Tensorflow from Source instructions to install all the dependens. Clone the source code from Github, and ckeckout to the specific version you want. Then follow the instructions below
cd tensorflow
./configure
bazel build -c opt //tensorflow:libtensorflow_cc.so
Copy libtensorflow_cc.so to your installation directories, then you can link libtensorflow_cc.so in g++ as the other normal shared libs. As for the include files, I do not know which are they exactly, I copy the following directories the include directories.
Supporse you are in the root directory of tensorflow source code, and your include directory is "INCLUDE_DIR". Copy the following to "INCLUDE_DIR" should work.
mkdir INCLUDE_DIR/tensorflow
cp -r tensorflow INCLUDE_DIR/tensorflow
cp -r bazel-genfiles/tensorflow INCLUDE_DIR/tensorflow
mkdir -p INCLUDE_DIR/tensorflow/third_party
cp -r third_party/eigen3 INCLUDE_DIR/tensorflow/third_party
And now, when you use libtensroflow_cc.so, you just need to add "INCULDE_DIR/tensorflow" to the include search path.
To use libtensorflow_cc.so, you need to have the right version Eigen and protobuf installed. If you notice the workspace.bzl under "tensorflow/tensorflow", the building tools downloaded Eigen and protobuf. For Eigen, you just need to download and uncompress it. As for the protobuf, you can download the right pre-compiled version, or download the source code and compile it yourself, see Compile protobuf from source for more information.
Jemalloc
The main benefit is scalability in multi-processor and multi-threaded systems achieved, in part, by using multiple arenas (the chunks of raw memory from which allocations are made). See jemalloc for more details.
XLA
XLA currently supports JIT compilation on x86-64 and NVIDIA GPUs; and AOT compilation for x86-64 and ARM. See tensorflow/performance for more details.
JIT
CUDA computing power
See Nvidia for your GPU computing power.
Eable instristic instructions support
bazle build -c opt --copt=-msse4.1 --copt=-msse4.2 --copt=-mavx