* 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

* fix m_ops

* fix activation op

* polish p ops

* fix precision_recall

* fix linear_chain_crf_op

* small fix

* polish r operator docs

* fix on naming convention

* Adding the doc format for AdaDelta

* Updating the documentation for Adagrad, Adam and Adamax

* Updating the auc op

* Fix review comments

* Updating doc for Batch Norm

* Updating the cast op

* Updating the clip op

* Fixing review comment

* Fixing review comment:

* Small change to restart PR_CI

* Adding documentation for seq_expand

* Adding documentation for seq_concat_op

* Adding documentation for sequence_conv

* Adding sequence_pool

* Fixing review comment

* Adding sequence_softmax

* Updating doc for sigmoid_cross_entropy_with_logits

* Doc fix for smooth L1 loss

* Adding doc for softmax_op

* Added doc for softmax_with_cross_entropy

* Adding documentation for transpose_op

* small change to restart TeamCity CI

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

After discussion with Helin and Yi, this change adds "print_operators_doc" executable to the Paddle docker nightly image.  This docker image will be pulled by PaddlePaddle.org nightly job and will generate the operator documentation to be put on PaddlePaddle.org website.

* polish from concat to conv_shift op doc

* small fix

* small fix

* proj init.

* add unittest and implementation.

* Polish the documentation for uniform_random and top_k ops

* Polishing more operators

* 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

* 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

* refine evaluator op types

* update

* follow comments

* update

* fix v2 mnist case

* fix v2 mnist case

* update

* update

* hack auc evaluator for dense vec

* follow comments

* 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 documentation of cross-compiling for iOS.

* Correst the typo in documentation of cross-compiling for raspberry pi.

* Set ANDROID_API to 21 when it is specified < 21 for arm64-v8a in build_android.sh.

* Check the input and print the usage in MergeModel.cpp.

* "replace enum with string"

* "fix layers"

Set gradient's data type based on its forward variable