add option to compile inference demo

* bnorm+relu fuse for mkldnn

* separate fuse_relu function

* bug fix

* proper while range in inference_transpiler

* description fix

* review fix

* review fix

* unit test for fwd batch norm+relu MKLDNN fuse

small clean

* optimize bipartite_match_op in large scale input

* Speed depthwise transposed conv2d.

* Set zero outside functor

Add Python Reader Op (CPP side)

Add design doc of Python data feeding

create about_us.rst

1.PaddlePaddle是百度提供的开源深度学习框架->PaddlePaddle是百度自主研发并开源的深度学习框架
2.框架本身具有易学,去掉本身
3.在兼具性能的同时->在保证性能的同时

Fix and enhance beam_search_op and beam_searc_decode_op

Fix dist train broadcasting bug

增加【关于我们】文案

* change return_numpy[PE] default value

* Remove convert to numpy in unit test

enable dynamic load mklml lib on fluid

Checkpoint M2: lookup table checkpoint

Fix async update failed