1. InferShape should be a method for `OpDescBind` not `Operator`, since
  `OpDescBind` is a compile-time concept.
2. Pre-create operators and store them into a map. Make InferShape
   faster

Use PROTO_LITE when refactoring Paddle

Add unit test of attributes in backward

* Adding implementation for copying a vector to tensor
* Changing Tensor test to access gpu memory indirectly

It will significantly reduce binary size. It is useful for mobile
deployment.

Fix compile error in develop branch

Fix bug of foward default attribute not passed to backward

Merge infershapecontext and ExecutionContext

Fix flushing vars

* Implementing the Adamax optimizer step operator
* Adding unit tests for adamax_op

* Changing learning rate and time step to inputs from attributes

* Changing learning rate and time step to input(tensors)

* Making the Adamax operator conform to naming convention

* Removing Tensor<float> from comments

* Rectifying the Adamax implementation

* Changing Unit Test values and adding comments

* Changing Unit Test to test multiple steps

Unify Reduce functions and simplify register code

Build mobile inference library for minimum size

Backward for op desc

* Update program.md
* Update
* Update

* fix block syntax

* add builder

* fix argument names

* uniform all Var to var

* fix c demo

* factor set_outputs

* update block

* delete as_ifelse_input interface from block design

* simplify set_outputs and output_num

* make the c++ codes compatible with demo

* fix compatible

* fix syntax

* update ie

* update syntax

Implementing tanhshrink operator

Update if_else_op.md 

Complete register gradient for compile time

Add compile time infershape