Yolo loss darknet

use embedding=128 bert model for test

* Add INT8 calibration support in Paddle package; test=develop

* Enable pool2d operator for a ngraph engine
test=develop

* Update
test=develop

test=develop, polish code and fix some wrong change

* move var strusted to vars_distributed.py, add optimizer's block name, test=develop

* rename optimzier's seems complex, revert it, test=develop

* replace * with details, test=develop

* Enable performance measurement in INT8 calibration unit test

test=develop

Make separate folders for mkldnn codes

* Some improvements to support bert mixed precision training

test=develop

* Revert the cast in layer_norm

test=develop

* Refine beam_search_op to output an extra parent_idx tensor.
test=develop

* Fix the unittest test_beam_search_op.
test=develop

* Fix the merging mistake.
test=develop

Add axis for boxcoder

add align_corners and align_mode for image_resize

fix trt models utest failed.

test=develop

test=develop

test=develop

test=develop, Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into feature/imperative

refine softmax kernel