* Add a interface to enable cudnn for inference.

* Add cudnn_placement_pass.
test=develop

* Set the default value of cudnn_enabled_op_types to null.
test=develop

* Write the common basic class, placement_pass_base, to refine the codes.
test=develop

* Call EnableCUDNN in unittest.
test=develop

* Refine cudnn_placement_pass tester.

* Enable the testing of cudnn_placement_pass in inference's unittest.
test=develop

* Add the check of op kernels.
test=develop

* modify the interface of Pass::Allay
test=develop

* Polish code
test=develop

* Fix Travis CI
test=develop

* fix Pass::Apply interface
test=develop

* Fix Travis CI
test=develop

test=develop

test=develop

* Support Sync Batch Norm.
* Note, do not enable it in one device.

Usage:

build_strategy = fluid.BuildStrategy()
build_strategy.sync_batch_norm = True
binary = fluid.compiler.CompiledProgram(tp).with_data_parallel(
        loss_name=loss_mean.name,
        build_strategy=build_strategy)

test=develop

* add MKL-DNN placement pass

This patch also refactors conv+bn (includes changes from PR
https://github.com/PaddlePaddle/Paddle/pull/13926)
updated to use the mkldnn-placement-pass.

test=develop

* remove redundant pass list

* add comment on the default first pass

* fix test for conv+relu mkldnn fuse

fix a few other things

* dist transpiler add control dependency var between send and recv

* fix async deps

* follow comments and refine

* fix deps connect for rpc ops

* make enforce a target and dependent on nccl when gpu is enabled

* add some more dependency

ISSUE=4586495

git-svn-id: https://svn.baidu.com/idl/trunk/paddle@1408 1ad973e4-5ce8-4261-8a94-b56d1f490c56