It will be used for LoD information in LoDTensor since LoD is a copy
on write field.

It is pretty slow for copying LoD information between operators. For
resnet it will cost roughly 10% time of whole time, including reading
data.

Add the concurrent programming design doc

* send_recv variable

* delete unused logs

* fix ci failed

* update

* resize tensor before tensor copy

* add selectedrows unit test

* check rows

Add a simple C++ inference example for fluid

Correctly handle lod information of image operators

* Add squared error layers doc

* Add doc for gru_unit

* Remove cdot which isn't supported

* Update layers.rst

* Update layers.rst (minor)

Async to drop kid

Default use one thread in fluid

Update the version of openblas for Android and iOS

Add flag use_mkl_packed

* "remove cudnndevicecontext"

* "remove unused init code"

* "fix hash functions"

Expose some activations

* "fix typo"

* "fix based on comments"

* "follow gogle style"

* "fix based on comemnts"

Tiny enhance of while_op

Doc fix and enhancement for lstm_unit python wrapper.

* "fix frigled test gradient of rnn"

* "fix based on comments"

refine pybind when *_op.cc contains several operators

* fix bug in SetAttrDescVisitor

* add comments

add norm op（cross_channel_norm）

enable MKL Packed Recurrent Layer