* add batch norm double grad, test=release/1.8

* update, test=develop

* fix, test=develop

* update err info, test=develop

* update, test=release/1.8

* Add trainable_statistics in attr for batch_norm
* Unifying behavior of dynamic graph and static graph

* add inplace_abn_op. test=develop

* update error information about batch_norm_grad

* update bn,test=develop

* batch_norm momentum support variable. test=develop

* fix format. test=develop

* add batch_norm momentum variable example. test=develop

* move MomentumTensor to training branch. test=develop

* split example. test=develop

* fix doc. test=develop

* fix PADDLE_ENFORCE ci. test=develop

* fix format. test=develop

* fix_bn

* revert unittest,test=develop

* fix bn

* fix spelling errors; test=develop

* Update API.spec

update md5

* Update API.spec

* change the order of api;test=develop

* fix cudnn batch norm accuracy test=develop

* fix cudnn batch norm accuracy test=develop

* disable failed test for later fix 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)

* Enable BatchNorm to use global mean and variane during training
* Update doc and follow comments.

test=develop

Fix code to support cpplint syntax check

test=develop

* add fp16 backward support
test=develop

* add sum_op fp16 test

* disable test_dist_save_load
test=develop

* add check_grad for sum

* add unit test for softmax_grad fp16
test=develop

* add scale_op unit test

* add mul_grad_op unit test for fp16

* add cross_entropy_grad and eman_grad unit test for fp16
test=develop

* fix cross_entropy unit test

* add pool2d fp16 unit test

* refine conv2d fp16 unit test
test=develop

* refine activation unit test
test=develop

* fix ci
test=develop

* follow zhihong's comment, copy from https://github.com/PaddlePaddle/Paddle/pull/12796
test=develop

* Fix batch norm when only 1 elements in normzalize dimension during training.

* "fix double type error"

* "fix ci"

* add data layout

* fix ci

There are mainly following fixes:

- take `DeviceContext` as the template parameter of math functors and OpKernel instead of `Place`
- remove `eigen_device` interface in base class  `DeviceContext`
- remove `GetEigenDevice` interface in `ExecutionContext` and base class `DeviceContext`
- remove unused `platform::EigenDeviceConverter`
- rename `REGISTER_OP_GPU_KERNEL` to `REGISTER_OP_CUDA_KERNEL`
- rename `USE_GPU_ONLY_OP` to `USE_CUDA_ONLY_OP`

* batch norm support matrix input

* update gpu code

* format code