* Add LoDRankTable

LoD Rank Table stores the `level` of `lod` which is ordered by sequence
length in descending order. It is useful when implement dynamic RNN and
is shared by dynamic RNN memory, dynamic RNN slice input and dynamic
RNN slice output operators.

* Add skeleton for array_to_lod_tensor and lod_tensor_to_array

* Add VarType::LoDTensorArray

* Add PyBind of LoDTensorArray

* Add InferVarType

* Add first unittest

* Add ut

* Add unittest

* Add unittest

* Add unittests

* update

* init

* add infershape for lod_tensor_to_array_op

* compelete array_to_lod_tensor_op

* copy data

* clean code

* clean code

* Fix unittest data

* fix bugs

* fix compile error

* Refine TensorToArrayOp

* refactor array_to_lod_tensor

* Unittest

* fix bugs

* Fix unittest

* Fix unittest

* debug

* Debug

* Fix unittest

* Add grad for ops

* Debug

* Fix a bug

* fix a bug

* fix a bug

* add fill_constant_batch_size_like_op to rnn h_boot

* first commit

* merge develop; fix conflict

* update to main_program

Follow comments

* Add LoDRankTable

LoD Rank Table stores the `level` of `lod` which is ordered by sequence
length in descending order. It is useful when implement dynamic RNN and
is shared by dynamic RNN memory, dynamic RNN slice input and dynamic
RNN slice output operators.

* Add skeleton for array_to_lod_tensor and lod_tensor_to_array

* Add VarType::LoDTensorArray

* Add PyBind of LoDTensorArray

* Add InferVarType

* Add first unittest

* Add ut

* Add unittest

* Add unittest

* Add unittests

* update

* init

* add infershape for lod_tensor_to_array_op

* compelete array_to_lod_tensor_op

* copy data

* clean code

* clean code

* Fix unittest data

* fix bugs

* fix compile error

* Refine TensorToArrayOp

* refactor array_to_lod_tensor

* Unittest

* fix bugs

* Fix unittest

* Fix unittest

* debug

* Debug

* Fix unittest

* clean code

* refactor

* use ostream

* update test

* fix gpu build error

* make gpu test pass

Used for shrink memories state in DyRNN. The height of state could
be shrinked after running a step block.

* Compare Operator

* Follow comments

1. user can disable peephole connections.
2. not calculate some gradients.

* Use stable_sort in lod_rank_table

It is easy to debug and test when use `stable_sort`and the time
complexity is not changed.

* Add LoDTensorArray

* Stash

* Better debug message for IsInitialized

* Stash

* Better debug message for IsInitialized

* Complete array read/write op unittests

* Add unittest, Gradient of array read/write

* Follow comments

* Use stable_sort in lod_rank_table

It is easy to debug and test when use `stable_sort`and the time
complexity is not changed.

* Add LoDTensorArray

* Stash

* Better debug message for IsInitialized

* Stash

* Better debug message for IsInitialized

* Complete array read/write op unittests

* Use stable_sort in lod_rank_table

It is easy to debug and test when use `stable_sort`and the time
complexity is not changed.

* Add LoDTensorArray

* proj init.

* add unittest and implementation.

* add acc layer
* memory log level change from 3 to 10
* use gaussian random to init conv parameters
* use initializer
* fix import
* batch_norm use helper to create persistable var
* refine code
* train only 2 batches for test
* use g_program and g_init_program
* use XavierInitializer to init fc parameter

* Add LoDRankTable

LoD Rank Table stores the `level` of `lod` which is ordered by sequence
length in descending order. It is useful when implement dynamic RNN and
is shared by dynamic RNN memory, dynamic RNN slice input and dynamic
RNN slice output operators.

* Add InferVarType

* init

* Fix bug

* rename test_filw

* refine test

* Fix bug in lookup_table_op & layers

* Missing Act in layers

* Should += in CPU

* Remove check in python

* Fix bug in sequence_conv_pool()

* Fix a bug in test_recommender_system.py

* Just skip test_evaluator

* Adding the Xavier Initializer
* Addressing code review feedback

* Init commit

* Make executor use ProgramDescBind

* Change Attribute from BlockDesc to BlockDescBind

* Since we will get the program desc in RNN, just BlockDesc is not
  enough.

* Add DeviceContext to Executor API

* Rewrite RNN

* Pass Python

* AddBiasOp does not care num_flatten_dims

* Stash

* Fix MacOS Compile

* Pass RNN forward

* add python test

* refactor test

* Make compile pass

* add gradopmaker

* First draft done

* Polish code

* add grad op maker and grad infershape

* Polish code

* Fix backward.cc bug

* Fix infershape

* Rename function

* add backward test

* simplify recurrent test

* Update

* Pass unittest

* Add comments & refine test

* Add comments

* refactor test

* Complete Unittest

* fix StepScopes enforce

* Remove unused unittest

* no type error

* Update

* Make RNN Pass unittest

* "add sequence conv layer"

* "add book recommender_system testing"

* "add training loop"

* "add sequence layer"

* "add recommender system training data"

* "fix conv2d layer bug"

* add sequence_conv_pool

* "fix input is Null"

* add networks

* "fix based comment"

* "add sum op layer"

* "merge layers"

* Update layers.py

* "fix input is NULL bug"

* "debug embedding table"

* "modify layers.py"

* "fix pool interface"

* "add export type to layers"

* "fix based on comment"

* "need lod info support in all operator"

* "remove accuracy layer"

* "tuning learning rate"

* "add sparse test"

* "add gpu test"

* Update test_recommender_system.py

* optimizer use init_program

* create persistable variable

* add create_persistable_var to block

* optimizer use create_persistable_var

* fix prefix

* move create_global_persistable_var from Block to LayerHelper

* Polish Optimizer initialization code.

* Using the LayerHelper to create initialize operator and variables

* add_accumulator should use an independent data type

* default use param data type for accumulator