* support ScaleTensor for scale npu kernel

* add more tests for adam npu

* fix compile

* fix unittest

* refine adam optimizer

* optimize OP's compilation time, test=develop

* add more op and run ci test, test=develop

* CUDA Kernel register in cc file, test=develop

* fix macros, test=develop

* fix undefined symbol error, test=develop

* fix compilation error and undefined symbol, test=develop

* fix compilation error on Windows, test=develop

* fix compilation error on Windows, test=develop

add the error message check for the some operator

* update ops's unittest of elementwise_pow, elementwise_max, elementwise_min, scale and sqrt
1. update elementwise_pow, elementwise_max and scale's unitests with input data type (float32 -> float64)
2. fix bug that the elementwise_pow doesn't meet threshold requirements with tackling float64 data
3. remove sqrt from op_accuracy_white_list.py
4. update the unittests of elementwise_pow, elementwise_max and elementwise_min ops that their input data shape over 100
5. test=develop

* modify the writing style according suggestions
test=develop

* add Adam beta1/beta2 support Variable. test=develop

* fix input<tensor>
test=develop

* fix split_ids
test=develop

* ElementwiseMul should not support SelectedRows

* fix scale op
test=develop

* change GetTensorFromVar() method to GetTensorOrSelectedRowsFromVar()

* fix operator

* refine MultiOutput

* fix MultiOutput
test=develop

* disable test_dist_save_load
test=develop

* fix elementwise_op
test=develop

* add get_sparse_as_op
test=develop

* add info for check
test=develop

* rename get_sparse_as_op with extract_rows_as_op.
test=develop

* elementwise doesn't support selected_rows

* fix regularizer

* remove extract_rows_as
test=develop

* fix ci
test=develop

* add test for sum_op

* fix regularizer
test=develop

*  test=develop

* fix pserver weight decay multi inputs test=develop

* add ScaleOpVarTypeInference for scale op

* scale op support scale selected rows

* optimize code

* use FindVar

* use FindVarRecursive in ScaleOpVarTypeInference

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`

* 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

Fix #3902

* Make `PADDLE_ENFORCE_EQ` supports custom class, like DDim

* add NDEBUG switch to PADDLE_ENFORCE

* Free will be added soon