[tensorflow gpu] build tensorflow shared library for C & apply it

environment

• ubuntu 18.04
• nvidia 418
• cuda 9.0
• cudnn 7.3.1
• bazel 0.18.0
• GeForce GTX 1080

2019-03-12 20:10:40.178166: E tensorflow/stream_executor/cuda/cuda_dnn.cc:332] could not create cudnn handle: CUDNN_STATUS_NOT_INITIALIZED
2019-03-12 20:10:40.178233: E tensorflow/stream_executor/cuda/cuda_dnn.cc:340] possibly insufficient driver version: 390.116.0

To solve above errors, I upgraded nvidia driver version to 418 from 390.(guide) No input method(KB, mouse) is working after 418

$ sudo apt-get install xserver-xorg-input-all

build tensorflow shared library for C++

tensorflow$ bazel build //tensorflow:libtensorflow_cc.so

below 2 .so files are generated

libtensorflow_cc.so
libtensorflow_framework.so

how to apply

It's simply applied with cmake for instance

link_libraries(
  ${PATH_TO_LIB}/libtensorflow_cc.so
  ${PATH_TO_LIB}/libtensorflow_framework.so
  )