* fix fleet for multi-stream

* fix memcpy for ncclid

* use sync to solve move operation

* support deepcopy for Layer/Tensor/Paramerbase

* fix some code

* add tensor_indices in AssignGroupBySize

* add rebuild group in reducer

* use has_grad instead of train_mode

* add vlog for debug

* fix ut

* fix ut

* The leaf tensor concept is exposed and the gradient accumulation of leaf tensor

* The leaf tensor concept is exposed and the gradient accumulation of leaf tensor

* fix coverage

* fix api doc

* fix CI unittest

* fix CI unittest

* fix unitest

* empty tensor does’t need inner_var_

* fix some error message

* Add a class TensorInplaceVersion to count the inplace version and put it in framework::Tensor instead of Allocation or Variable.

* Add a new attribute `_inplace_version` for VarBase.

* Raise exception if an inplace operation can result in incorrect gradient computation.

* Add a new interface _bump_inplace_version() for VarBase to bump the version whenever the Tensor is modified through an inplace operation.

* For api assign, call _bump_inplace_version() when it's an inplace operation inn dynamic mode.

* Use original var_wrapper if the inplace_version is not changed.

* Replace SnapshotVarWrapperList with SnapshotVarWrapper to optimize performane.

* add reducer

* refine envent for memorycopy

* add concat&split for allreduce

* apply concat & split for fuse tensor

* fix nccl dep

* fix the untest, compile problem and ddp initialize problem

* fix untest for mac & add some comments & solve the repeated param in sublayers

* fix untest for windows & fix document

* split train_mode and has_grad

* fix format

* fix ci problems

* fix sample code

* add base hook classes and reduce hook impl

* fix constructor typo

* polish comment format

* refactor baisc hook class design

* polish design details

* remove slelected rows all reduce over height check

* polish unittest

* add retry for dygraph parallel socket bind

* change to loop always

* fix writing error

* nccl init support hostname and ip; test=develop

* add sumary feature

* refine printting tensor

* add sci_mode

* add sample code

* fix indent error

* fix _format_item

* polish code

* support item indent

* add ut

* set place for ut

* fix py2 issue

* fix ut

* Add support for mkldnn ops types selection with FLAGS in dygraph

* use regex to match DNNL verbose

* python3 encoding fix

* refine err msg

* follow comments

* use iwyu clean include, test=develop, test=win

* compilation error, test=develop

* fix compilation error2, test=develop

* fix compilation error3, test=develop

* fix compilation error4, test=develop

* fix compilation error5, test=develop

* fix compilation error6, test=develop

* fix compilation error7, test=develop

* fix compilation error8, test=develop

* fix compilation error8, test=develop

* fix compilation error10, test=develop

* fix compilation error11, test=develop

* get use of global 'use_mkldnn' in layer_helper

* update for CI

* update for CI, relu test

* update for CI, relu test added, make FLAGS_use_mkldnn a public flag

* added more strict tests, fixes after review

* fixes after review

* fixes after review, CI stuff

Use `Tensor` instead of `Variable`  in the doc of paddle.grad.

* update the demo code and the doc of varbase.backward.

* update the doc of the fake interface `paddle.fluid.Variable`.

* remove BackwardStrategy.

* support Baidu AI Accelerator
  * test=kunlun

* minor
 * test=kunlun

* support xpu op in separate file
 * test=kunlun

* update XPU error message and remove duplicated code

 * test=kunlun

* minor
 * test=kunlun

* minor
 * test=kunlun

* enable mkldnn layout conversion

* review fix: remove tmp_place

* fix test mkldnn swish

* add UT for PrepareData CPU->MKLDNN

* add #ifdef PADDLE_WITH_MKLDNN

* Force-push commit
Co-authored-by: Ngrygielski <adam.grygielski@gmail.com>

* add unchaged infershape function

* add broadcast infershape function

* fix bug

* rename infershape functions

* add UnaryOpUnchangedInferShapeCheckAxis

* add error message

* add test for common infer shape functions

* dont update existed ops

* dont update op_desc.h

* add more test

* add error check, refine error message

* add auto_cast, test=develop

* add loss scaler, test=develop

* add comments, test=develop

* refine code, test=develop

* refine code, test=develop

* do not set flags automatically, test=develop

* fix custom op bug, test=develop

* add more test, test=develop

* refine enable logic, test=develop

* enable amp test with GPU, test=develop

* add unittest

* add test for found_inf

* follow comments

* follow comments

* remove global variable, use singleton

* add some notes

* update comments

* update comments

* update comments

* add use_dynamic_loss_scaling argument

* refine found_inf

* refine found_inf

* Add some error messages for the op without double grads.

* fix the test_imperative_double_grad UT.

* fix double grad visitid unit; test=develop

* change name hash_pair to HashPair; test=develop

* follow comment; test=develop

* Added DNNL cache management for DyGraph

* move FLAGS_use_mkldnn to more general CMakeLists, getu use of the flag in ClearGradients

* missing file

* Fixes after review

* Bringing back original idea of place for 'use_mkldnn' flag to be accessible from platform nad imperative.

* Removed duplicate and added docs

* Fixes for CI

* fix the double grad bug for the star gan. test=develop

* update the retain_graph parameter doc. test=develop

* add the unit test for the retain_graph parameter. test=develop

Add Support for SelectedRows for Transpose OP and Fix a Bug That SelectedRows Cannot be Supported in SimNet (#25536)

This PR fixes a bug that SelectedRows cannot be supported in SimNet. The reason of this bug is that dygraph basic_engine didn't copy var's type when the var needs to be accumulated during backward. So when a var is SelectedRows and needs to be accumulated, like SimNet which calls net for two times, the var's type will be changed to default LoDTensor thus bug happens. To fix it, we just also copy the type.

Without this PR, the accumulated SelectedRows parameters in dygraph will be changed into LoDTensor. So when we fixed the bug of supporting SelectedRows in SimNet, we found `test_imperative_lod_tensor_to_selected_rows` failed and threw the error that SelectedRows was not supported for Transpose OP. To fix it, too, this PR also added support for SelectedRows for Transpose OP.

* Modify tmp var name prefix in dygraph test=develop

* refine comment test=develop

* support selectedrows allreduce in multi-cards dygraph, test=develop

* remove useless import modules in unittests, test=develop

* add nccl cmake to get nccl version, test=develop

* add if-condition to compiled correctly, test=develop

* add detail version parseing for old nccl, test=develop

* polish camke details, test=develop

* fix remove test cmake error, test=develop

* fix cmake condition, test=develop

* change unittest camke list, test=develop

* fix unittest cmake rule, test=develop, test=framep0

* support gradient accumulation of fp16 in imperative mode, test=develop

* enhance coverage test, test=develop

* follow comments, test=develop

* append error op hint for grad op maker, test=develop

* add unittests for coverage, test=develop