* refine reduce by cub
* optimize KernelDepthwiseConvFilterGrad
* optimize depthwise conv and reduce mean and reduce sum
* fix bug: dilation
* cuda arch and cuda 8 compatible

* show detail error log on ci

* test

* fix memopt and dist

* update apispec

* will fix different batch issue test=develop

- add naive executor
- fix concurrency performance issue

* add GraphNum

test=develop

* add graph number check in parallelExecutor

test=develop

* fix transformer_model bug

test=develop

* fix graph num

* Add CUDA implementation for generate_proposals_op.
* Clean code.
* Update code.

* add dist ut for text_classification

* add dist ut for text_classification

* add simnet bow unittest

* add dist ut for simnet bow

* add trainning data url for simnet bow

* add trainning data url for simnet bow

* modify simnet test_reader to train reader

* add test_dist_ctr

* test_dist_ctr can run now

* dense update is good

* add unit test for selected rows

* debug unit test

* fix dist sparse update problem

* Constant args at init

* optimize code

* simnet optimize

* fix DebugStringEx

* optimize sum_op.h

* add ScaleOpVarTypeInference

* clean code

* fix test_dist_transpiler.py

* code optimize

* modify delta

* fix sparse update bug

* dist test use one cpu

* update some data

* remove unused code

* add use cuda config

* unit test fix

* unit test fix

* unit test fix

* unit test fix

* dist_word2vec use CPU

* unit test fix

* unit test fix

* code clean

* code clean

* merge develop

* api spec update

* Revert: api spec update

* replace simnet data with fake

* replace simnet data with fake

* update dim

* add batch auc

* code clean

* code clean

* modify print to stderr

* update simnet delta -> 1e-5

* update RUN_STEP

* add use_reader_alloc

* add use_reader_alloc

* add use_reader_alloc

* modify delta

* add use_reader_alloc

* fix stderr write

* python3 compatibility

test=develop

* python3 compatibility, test=develop

* Update dist_text_classification.py

* test=develop

* add distributed auc

* add attr "is distributed" and config it

* add distributed auc

* add batch auc and code format

* code format

* auc optimize

* metric_op optimize

* code clean

* bug fix and code clean

* bug fix and code clean

* code optimize

* code optimize

* api spec update

* Comments optimized

* add mutex

* Revert: add mutex

* remove distribute metric

* remove distribute metric

* spec modifyed

* add annotation, test=develop

* keep API compatibility
test=develop

* hide operator API

* update

* update api.spec

* fix merge

* fix test

only mode

* add nccl2 dist api

* update apispec

* update

* update api spec

* wip

* clean up

* should fix running with memopt

* add ut

* mark lr schedule op role

* hide lr_schedule_guard

* use op_role_var instead of ufind

* unify dist test name

* wip for py3 support

* fix var deref

* fix python3 mem_opt order

* remove comments

* Add roi perspective transform.

* Add roi_perspective_transform_op.

* Fix code style.

* Add python api and fix doc.

* Fix API.spec

* Fix python api.

* Fix API.spec

* Move src to detection.

This reverts commit a6c8d6b9.

This reverts commit bd79e046, reversing
changes made to 6b4d290c.

Add warning when disable test_fuse_elewise_add_act_pass on Mac

* Optimize GRU with AVX instruction

* Clean code

* Add the Unitest and fix the align issue

* Remove the remanent part of the unitest part

* Code clean

* Fix the parameters length issue for fusion_gru to pass CI

* Change the default type as float32

* Add Preface

* Add demo code

* Save file

* Refine code

* seems can work

* use elementwise strategy

* Use ElementwiseComputeEx

* Add comments

* extract functions from operator

* Refine code

* Follow comment

* code refine

* add op_fuse  pass

* add backward

* code refine

* use TopologySortOperations

* follow comments

* refine IsFusible

* code enhance

* fix op_fusion_pass

* refine code

* refine fuse_elemwise_act_op

* adjust the input and output

* refine logic

* add intermediate_edge

* disable inplace

* follow comments

* refine logic

* follow comments

* Remove the removable IntermediateOut

* change strategy

* code refine

* enable fuse backward

* code refine

* code refine

* rename unit test

* follow comments

Sequence Scatter Op

In the future, we'd like to encourage user to save
everything during training.
This allows us to
1. Do more flexible optimization passes
2. Re-train and fune-tune