properly set up dep of concat and fetch_bar

Fix inference api impl deps

add distributed training deps.

Fix grpc destroy bug

fix fluid_benchmark.py

unify libpaddle_inference_api and libpaddle_fluid

* add flatten op

* enable dist se resnext

* small batch size

* refine regularization for selected_rows

* clean lookup_table

* refine rpc_server_test

* temporally disable rpc_server_test

* fix rpc_server_test

* add unit test

Add tensorrt pooling converter

refine num_threads control

Add self to author

update inference_optimize() to support program with readers

fix cudnn enforce

fix manylinux1 Failed to publish artifacts

add flag to prevent unnessary memory free

* refine conv cudnn enforce

* update

* update all cudnn ops

* fix

distribute lookup table work with parallel executor

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into parallel-executor-support-prefetch

construct a SSAGraph at the beginning.