* 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

Reduce one level of inheritence.

add doc

* Add learning rate decay test

* fix test name

* doesn't share @LR_DECAY_COUNTER@

* refine multi-thread CPU Parallel exe

* refine multi thread CPU Parallel exe

* Refine CPU version for ParallelExecutor

* add share_parameter_between_cards_

* Fix ParallelExecutor bug

* Fix unit test

* Fix parameter opt balance

* Fix with opti (param->grad)

* Add grad to op var

* Remove shard_param_between_cards

* Fix tensorcopy bug

* follow comment

* Refine TensorCopy

* 1. Create buddy allocator in each places before NcclBcast the variables
2. Check the memory usage of ALL gpus rather than the first one

* 1. Make NCCLGroupGuard guards only the ncclBcast part, which avoid ncclGroupEnd blocking the exception throwing
2. NOTE the usage of NCCLGroupGuard

* Remove the memory usage check of gpus

* Fix code style

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