* add dla

* add python api
Co-authored-by: Nshangzhizhou <root@szth-rp-fanyi-opera49.szth.baidu.com>
Co-authored-by: Nshangzhizhou <root@szth-rp-fanyi-opera49.szth.baidu.com>

Co-authored-by: NJacek Czaja <jacek.czaja@intel.com>

* Add tf32 support for A100 tensor core acceleration for cuBLAS (#28732)

* Fixed an error

* Fixed an error

This PR is cherry-picked from PR: #29192
Function: Added TF32 switch for cuDNN. Turned on as default, turned off when users set the switch as False

Cherry pick PR #30520 .
Fix error message of Inplace strategy.

[cherry-pick] support layer_norm fp16 in dygraph amp (#30430)

cherry-pick #30553
fix bug of multicard grad ncclAllReduce, the gradient accumulater of parameters should be keep order, otherwsie, it will influence multicard ncclAllReduce of grad.

cherry-pick #30536

Fix the compiling error of update_loss_scaling when using cuda9.

* add trainers for pserver

Change-Id: I99c0ab1cc427318f1f9bf8f8f5faff2b8890645d

* add trainers for pserver

Change-Id: I1a75793ec81ce126d07f4c47cae09b95d530bbc8

* support transformer v2.0

* fix range op crash in dygraph xpu place

Co-authored-by: NWojciech Uss <wojciech.uss@intel.com>

* Modify the calculation logic of LambOptimizer (#29313)

* Modify the calculation logic of LambOptimizer

* Modify the calculation logic of LambOptimizer

* Modify the calculation logic of LambOptimizer

* add pad and concat double grad

* resolve conflict

* add fp16 support for tril_triu op (#30186)

* add VecCastCUDAKernel (#30296)
Co-authored-by: Nfurnace <34057289+windstamp@users.noreply.github.com>

Cherry-pick PR 30103. Add Inplace strategy (Output reuse Input Varbase) in dygraph (#30103) (#30496)

* add view strategy on squeeze,unsqueeze,reshape,flatten

* add squeeze unittest

* add unittests

* use View strategy as name rather than Reuse Allacation

* fix view api doc

* fix format

* use core.ops when input of reshape2 is Tensor

* fix test_cross_entropy_loss error because of reshape2

* fix test_cross_entropy_loss error because of reshape2

* add inplace strategy

* add elementwise_add sub

* let backward op not use inplace

* grad op do not use inplace

* fix memory increase error and add leaf error message

* delete selected_rows

* change op_function

* little change

* solve HandleViewBetweenInputAndOutput

* add unittest and leaf error message

* merge view error

* optimize op_function_generator format and support sum inplace op

* fix format of basic_engine

* fix format for framework

* little change of variable wrapper

* add reshape, squeeze, unsqueeze, scatter api

* add relu elu tanh softmax inplace api

* fix test_squeeze_op unittest

* fix test_relu_op unittest

* fix comment problems

* delete sample code of inplace api

* add reference of grad_pending_nodes in basic_engine

* fix unittest name

* add inplace apis into wlist

* fix error message

* add PADDLE_ENFORCE for set grad op twice

* fix head file error

* Cherry-pick 30072, add dispenable input for core.ops.reshape2/expand/slice (#30072)

* add dispenable input 'shape' for core.ops.reshape2

* add dispenable inputs for core.ops.reshape2/expand/slice

* add ut

* save reshape update in pr 30180

* save reshape update v2 in pr 30180
Co-authored-by: NLeo Chen <chenqiuliang@baidu.com>

built-in `rsqrt` is shadowed

* add transpose double grad test=develop (#29600)

* add transpose double grad test=develop

* cherry-pick test=develop

* test=develop, add distributed_infer (#30300)

* test=develop, add distributed_infer

* test=develop, fix unittest cmakefile conflict

* test=develop, fix test_dist_fleet_base

* fix the rnn mask memory bug for out of read

* update the code for the rnn

* optimize memcpy perf for kunlun (#30291)

* optimize memcpy perf for kunlun

* remove useless unitest for kunlun mean

* minor

* fix bug that cann't find mkldnn(kunlun) (#30394)

* Add double grad for conv_transpose (#29706)

* add double grad for conv_transpose

* register cudnn conv double grad for depthwise conv (#29807)

* softmax backward optimize

* set expected place in child thread for dataloader

* set device id when set tensor from numpy

* revert tensor_py change

* add compile guard

* fix ci

* fix bug