* Enable cross_entropy operator for a ngraph engine
test=develop

* Update tests
test=develop

* Added PADDLE_ENFORCE for the batch_norm operator
test=develop

* Update the message about which format are supported right now
test=develop

* Correct the comment in control_flow.py.

* Correct the argument list of ops.
test=develop

* Update API.spec.
test=develop

* Skip op_callstack attr for all op apis.
test=develop

* Remove use_mkldnn and is_test from python api.
test=develop

* Remove use_mkldnn and is_test from op_proto_maker and hard-coding them in python when generating doc string.
test=develop

test=release/1.3

test=release/1.3

* Add the CRFDecoding and LayerNorm's test case

test=develop

* Fix the size checking issue

test=develop

* Remove the remnant code

test=develop

* Add TestAllImpls and double support

test=develop

* Clean Code

test=develop

* Add benchmark test for LayerNorm & CRFDecoding

test=develop

test=develop

test=develop

test=develop

* inplace group

* test=develop

* fix && test=develop

* fix && test=develop

* test=develop

* Fix debug mode in prior_box_op
* Refine code

test=develop

test=develop

test=develop

test=develop

* Fix debug mode in fake_quantize_op
* Remove template specialization

This also fixes linking to libpaddle_fluid.so built in debug mode

test=develop

test=develop

test=develop

* Added ngraph sum op test=develop

* Added sigmoid, relu_grad and tanh_grad test=develop

* remove duplicates test=develop

test=develop

test=develop