* Fix the gelu backward to avoid nan

test=develop

* Remove unnecessary calls

test=develop

* expose square
test=develop

* fix activation
test=develop

* Add square API
test=develop

* add necessary op

* code refine

* fix API.spec
test=develop

* fix unit test
test=develop

* add unit test sparse_grad_clip
test=develop

* fix API.spec
test=develop

* remove mac test for test_gradient_clip
test=develop

* remove selectedrows_mul_tensor
test=develop

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

This reverts commit bf3c3496.

* "cherry picked operators changes"

* "remove duplicated code"

* "add constant setter"

* "add get expected kernel"

* "fix ci"

* "add fill constant"

This reverts commit 1eeb11ef.

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`

* "add floor, ceil, round op"

* "reuse zero gradient"

* "fix divide zero"

* "fix numpy floor error"

* reimplement pow operator

* add pow_grad operator

* fix code style

* fix build error

* fix op_test bug

* revert pow operator

* add FIXME comment

* Simplize Gradient Check

* Stash

* Extract apply_backward_pass to backward.py

Rename apply_backward_pass to append_backward_ops

* Use graph API to check gradient

* Fix ci

* Fix CI

* Fix backward for double precision

* Stash

* Fix CI

* Fix ci

* Ignore GRU test

* Ignore xe op

* Fix CI

* Fix softmax with xe gradient

The correct equation should be IG = OG * (d_softmax_with_xe())

* Fix typo

* Fix merge error

* Disable LRN

* Adding Hard Sigmoid Activation

* Adding a comment for slope to be only positive

* Fixing grammatical mistake in comment

* Adding thresholded_relu op
* Adding test for thresholded relu op

* implementing softplus

* small fix

* small fix

* small fix

* small fix

* Implemented the hardShrink activation

* Fixing the unit test

* Add numerically-stable logsigmoid activation

* Add softshrink operator

* Adjust relative tolerance for grad-check

* Address review comments