* Enable performance measurement in INT8 calibration unit test

test=develop

* Some improvements to support bert mixed precision training

test=develop

* Revert the cast in layer_norm

test=develop

* Refine beam_search_op to output an extra parent_idx tensor.
test=develop

* Fix the unittest test_beam_search_op.
test=develop

* Fix the merging mistake.
test=develop

test=develop

test=develop

* fix default create_parameter dtype maching initializers test=develop

* update type check test=develop

* update test=develop

test=develop

test=develop

test=develop

test=develop

test=develop

* Enable mobilenet UT in separate test class; use download cache by paddle download utility and cache unzip; and fix typo; test=develop

* Extract cache_unzipping function for reuse; format code style; test=develop

* Simplify the test code by define a combined function for both downloading and unzipping; test=develop

test=develop

test=develop

test=develop

* Refine the beam_search op and test.

* A basic CUDA implementation of beam_search for small batch_size.

* Implement CUDA kernel for beam_search_op.

* Use multiple CUDA threads in the same block to select the top beam.

* Update the python api of beam_search op.

* Enable extend function in CPU kernel of beam_search op.

* Unify the CUDA codes.
test=develop

* Unify the CPU kernel of beam_search op.

* Ensure the seletced items of beam_search_op's CPU kernel sorted by scores.

* Update the description of beam_search in API.spec.

* Enable the use of CUDA kernel in beam_search op.

* Exclude the beam_search's CUDA unittest when there is no CUDA gpu, and delete some debuging statements.
test=develop

* Follow comments.
test=develop

* Call the CPU kernel for beam_search op when batch_size > 4.
test=develop

* Remove the except of is_empty op in PrepareData.
test=develop

test=develop

* nce add check sample lables, test=develop

test=develop

test=develop

test=develop

test=develop