MKLDNN layout: Support for pool operator

replace use_event with use_cuda, because use_event means the program running with CUDA, so use_cuda maybe more intuitive.

Check SSA Graph

* 1. add weight decay feature into fluid benchmark test
2. add learning rate decay feature into fluid benchmark test
3. add L1&L2 regularization feature into fluid benchmark test
4. add error clipping feature into fluid benchmark test
5. add gradient clipping feature into fluid benchmark test

* Add some document to README.md under benchmark/fluid/ repo

* Add model_base.py

* Fix bugs in test_listen_and_serv_op

* 1. remove args out of fluid_benchmark.py
2. remove lr_decay, regularization, clipping out of fluid_benchmark.py

* add async_mode description to doc and remove the clipping description out

* for restart build

* to restart build

* remove optimization args from args.py

* 1. remove optimization from models
2. fix bug in test_listen_and_serv_op

* change the name retry_times to left_time

* change retry_times to the pserver start left time

Add lock to record_event.

Fix identifier error of 'dshape' in 'Paddle/benchmark/fluid/models/vgg.py'

Add argmin and argmax op

add FLAGS_use_mkldnn to global control use_mkldnn

* Remove unused arg_max/min_op.h
* Remove reference parameter. Use pointer insteaded.
* undef macro
* Always set OutT as int64_t.

Simplize API Reference Documentation

fix deadlink in api_doc_std_*.md

* Expose maxout API.

* Fix code style.

fix distribute_transpiler

New api about checkpoint and models

polish sparse update logic

Fix bad code in c plus and python

* Use for_test=True in the Fluid Trainer to clone the test program

* fix typo

* Should do the samething to the inferencer