* "optimizer op support float64"

* "fix ci"

* "fix ftrl op"

* Load/save the params from the params_path

* Switch to use load_persistables and save_persistables

* Instaed of setup the executor to run program and scope. Pass the program to the load_persistables

Fluid new API: dist train without modifying code

Wrap shfl_x_sync

* fix remove op

* update

Polish MatMul, clean copy & paste code

Distributed survey

tensorrt convert init

* Added auto transform to beam_search_decode_op

* Added some comment

* Added unittest for beam_search_decode_op on GPU

* Fix CPPLint issyes in gru_kernel.h

* Fix CPPLint issyes in gru_kernel.h

* Fix Compile error

fix delete_ops

* add test float16 inference accuracy example

* complete the test

* clean code

* add argument parse and refine tests

* add shell script

* add float16 benchmark code

* refine code

* prepare for contrib/float16

* put things in contrib float16 folder

* update benchmark result

* further update benchmark report

* add float16 inference report

* update report

Works with 1 trainer 1 pserver. 2 trainer 1 pserver will stuck at the
end of first step, still investigating.

The user only need to set envrionment variables to enable distributed
training.

run pserver:

PADDLE_TRAINING_ROLE=PSERVER PADDLE_PSERVER_IPS=127.0.0.1 PADDLE_TRAINERS=2 PADDLE_CURRENT_IP=127.0.0.1 python no_test_word2vec_new_api.py

run trainer:

PADDLE_TRAINING_ROLE=TRAINER PADDLE_PSERVER_IPS=127.0.0.1 PADDLE_TRAINERS=2 PADDLE_TRAINER_ID=0 python no_test_word2vec_new_api.py

* update README.md

* change to .html

* use v2 api

fix pool with mask layer bug

Feature/clean blas

Wrap __shfl

* Add understand apiu with stacked lstm for new API

* Complete exam

Fix __shfl and __shfl_down for CUDA9.0