Add a test to ensure profiler works on multi-gpu

Move back operator's event to RunImpl()

Fixed some outdated contents in Contribute Documentations

Feature/shuffle reader

Add product reduction for reduce op.

Enhance regularizer.py

Fix distributed train doc deadlink

Add api doc std

Merge branch develop

* "Refine nccl op"

* "refine code "

* "refine nccl code"

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into feature/Enhance_regularizer_py

Enhance look_up_table op 

move fluid dist design to fluid folder

Adjust some contents and create a new folder src to store pictures.

Feature/recordio file reader