cherry-pick PR #42777

* Reduce time variation for cuda_managed_memory_test (#42458)

* Disable standalone executor for test_tensordot (#42476)

[cherry-pick 2.3] Add fused_multi_transformer op to optimize transformer generation performance (#42311)

* Add fused_multi_transformer op to optimize transformer generation performance (#41814)

* fix fused_multi_transformer compile failed in cuda arch < sm53 (#42315)

* fix ci timeout

* full api fix

* when out is None, go old dygraph mode

* by static check

* first version: support 2-inputs forwards. TODO: 1. backward  2. BroadCast  3. MultiVariable

* time out -> 120

* sharding_for_eager_tensor (#41415)

* fix_sharding_copy_right (#41849)

* fix utest

* fix time

* Add yaml for matrix rank op (#41466)

* modify matrix_rank

* add matrix_rank shape

* add matrix_rank shape

* Add yaml for matrix_rank OP

* Add UT
Co-authored-by: Nzhoujianqian <15205085056@163.com>

* Add yaml for eye OP (#41476)

* [cherry-pick] Add yaml config for matrix_rank, eye, deformable_conv and
deformable_conv_v1 OPs
* Add yaml for deformable_conv and deformable_conv_v1 OPs

* Add UT

* Add to skipped_phi_api list for infrt
Co-authored-by: Nzhoujianqian <15205085056@163.com>

* Use `self`as a parameter of _hash_with_id function to avoid error caused by hash_id reuse (#41200)

* Add fill_constant_batch_size YAML and UT (#41474)

* Switch some dy2st UT to eager mode (#41382)

* Sitch some dy2st UT to eager mode

* Fix test_lstm and remove test_transformer

* Run test_resnet_v2 in old dy mode

* [Eager] Support test_layers's test cases switch to eager mode

* Update batch_norm _C_ops action to fix CI

* Use None instead of new EmptyTensor

* Updated var name

* Make sure to switch eager mode, Fix Coverage_CI

* Remove _non_static_mode statement

* Remove batch_norm dispensable input statement

* Polish batch_norm code

* Fix CI issue

* support process group in dp with fleet api

* update

* fix uts

* update

* add nms op and batched_nms api

* enable new executor by default

* enable stream safe allocator

* test=document_fix;test=coverage

* do not use scope in op kernel

* fit empty program for new executor

* fix communication depend

* fix test_sync_batch_norm

* skip unsupported place

* refine datatransfer

* fit for dirtributed program

* fix dependencpy

* fix some ut

* [Dygraph] Support sparse tensor in refactored reducer

* add uts

* refactor

* update

* fix bugs

* add infershape and forward yaml

* add final_state call

* add base unittests

* add backward yaml and test

* fix without softmax test error

* add cross_entropy test

* Add yaml for reduce_sum OP

* Fix CI errors

* Fix CI errors

* Fix CI errors

* Fix CI errors

Add new APIs for GPU memory monitoring (max_memory_allocated, max_memory_reserved, memory_allocated, memory_reserved) (#38657)

* Add new API memory_reserved

* Add memory_allocated, max_memory_reserved and max_memory_allocater

* Fix CI error

* Fix CI error

* Enhance UT

* Add FLAGS_memory_stats_opt

* Add STATS macro functions

* Add StatAllocator

* Fix CI errors

* Add UT

* Fix CI errors

* suppor inplace in tensor_method_setitem

* delete bump_inplace_version

* optimize inplace unittest

* fix

* fix setitem bug

* update eager_generator

* optimize inplace unittest

* little change

* add uts for EagerReducer

* add more uts

* fix bugs

* fix bugs

* modify

* modify uts

* fix bugs

* update

* update

* update

* solve conflicts and merge

* add some other uts

* modify time of uts

* update

* update

* update

* remove uts of resnet

* refactor eager flags

* fix flags error when we switch from eager to dygraph

* fix ci problem

* fix ci

* fix ci

* merge develop and fix code style

* merge develop and fix code style

* fix op test error

* fix op test error

* fix op test error

* fix op test error

* fix op test error

* merge develop

* [Eager] Support eager grad interface, draft version

* Support eager grad interface with allow_unused and multi startup_op

* Fix code format

* Fix allow_unused case, return PyNone if tensor not initialize

* Support output's stop_gradient related to create_graph

* Support grad exception case in eager mode, fix coverage CI

* Update ToPyObject, return PyNone if not initialize

* AccumulationNode add FLAGS_retain_grad_for_all_tensor

* Fix ci issue

* Fix CI issue

* fix, use core.eager.Tensor

* Add func SetBufferSlotRankZeros for GradTensorHolder

* Support retain_graph by using ClearTensorWrappers

* Support retain_graph by using ClearTensorWrappers

* Update retain_graph and no_grad_vars related test case

* Update code gen logic for ClearTensorWrappers

* Fix by override statement

* fix override func args

* Support retain_graph, update unit tests

* Updated ClearTensorWrappers logic

* fix grad python interface

* Use deep copy and update unit tests

* Polish code

* Polish code

* Fix CI issue, Deep copy only use when user set grad_tensors

* Fix CI, use Backward instead RunBackward

* Fix CI, Declare kernel explicitly in test file

* Polish, remove vector of TensorWrapper

* Refactor the logic of grad/backward, polish codes

* Update code after merge upstream develop

* Polish after merge upstream develop

* Update to adapt new GradNodeBase superclass

* Fix error introduced during conflict resolution

* support inplace strategy in eager_fluid state

* solve conflict

* nothing

* Update purify potential_startup_nodes logic

* Fix errors

* Polish code

* Remove useless args for ToPyObject

* Remove useless TensorWrappersSet

* fix record conflict

* Fix code-format, re-install pre-commit

* fix tensor_wrapper bug

* Fix pre-process logic for potential_startup_ops

* Update unit tests, use eager mode

* Fix conflicts

* fix unittest timeout

* little change
Co-authored-by: NWeilong Wu <veyron_wu@163.com>

[multiprocessing] Add paddle.incubate.multiprocessing for sharing tensors  between python processes. (#37302)

* Add support for paddle.multiprocessing
* move multiprocessing to incubate.

* fix time of utest

* Added cuBlasLtHandle_t to device context.

* Added fused_gemm_epilogue op.

1. Added fused_gemm_epilogue op to leverage cuBlastLt Epilogue.
2. Support fusion Act(X*Y + bias), X'dims >=2 and Y'dims shoule be 2.
2. Act currently only be supported ReLU. (Will add GeLU in the future).

* Added UT to fused_gemm_epilogue op.

* Added LinearAct Pattern

1. Added LinearAct into graph_pattern_detector.* to define (2.)'s
pattern.
2. LinearAct is used to detect act(element_add(matmul_v2(x, w), bias)).
3. act currently only support ReLU (Will support GeLU in the future).

* Added FuseGemmEpiloguePass

1, Added FuseGemmEpiloguePass to handle nn.Linear + Act{ReLU}
fusion (GeLU will be supported in the future).
2. Only support matmul_v2 from nn.Linear.

* Added pybind to BuildStrageter.fuse_gemm_epilogue_.

* Added UT for fuse_gemm_epilogue_pass.

* GeLU support and EpilogueSingleton

1. Added GeLU support to fused_gemm_epilogue op.
2. Added EpilogueSingleton to cache auxiliary pointer.
3. Added related UTs.

* Rename cublaslt_epilogue_opto gemm_epilogue_op.*.

* Added both train and infer pattern to LinearAct.

1. Added support of fwd graph with grap_ops linking to LinearAct.
2. Added related changes to fuse_gemm_epilogue_pass for above
modification.

* Changed CUDA requirement from 11.4 to 11.6 for fuse_gemm_epilogue_pass.

* Added identity activation support to gemm_epilogue_op.

* Added Linear Fusion (matmul_v2 + ele_add)

1. Added matmul_v2 + ele_add pattern to LinearActPattern.
2. Added matmul_v2 + ele_add support to fuse_gemm_epilogue_pass.

* Rename gemm_epilogue_op.* to fused_gemm_epilogue_op.*

* Add fused_gemm_epilogue_grad op.

1. Added fused_gemm_epilogue_grad to support backward epilogue fusion.

* Add UTs to fused_gemm_epilogue_grad_op.

* Change attribute name in fused_gemm_epilogue_grad_op for clearing.

* Allow DX and DBias be dispensable to fused_gemm_epilogue_grad op.

* Added ElementwiseAdd+Matmul+Act graph pattern detection.

* Fuse backward of Linear( Act(x))

1. Added backward fusion pass to Linear( Act(x)).
2. Added backward fusion pass to Linear(x).

* Added UTs to backward fusion of Linear(Act(x)).

* Complete document of arguments to fused_gemm_epilogue_op.

* Made arguments of some functions pass by reference.

* Modify code with review comments.

1. Made arguments of some function pass by reference.
2. Removed redundant code.
3. Followed Google code style to change code.

* Made 'const' code style be consistent

* Fixed random seed of python UTs.

* Set Compiling constrains to cuBlasLt

1. Require CUDA 11.6+
2. Remove fuse_gemm_epilogue related tests when CUDA < 11.6.

* Code Reivew from Paddle

1. Changed arguments name is_first_gemm to without_x_gradient for
clearing.
2. Applied PADDLE_THROW in fused_gemm_epilogue_op.

* Remove EpilogueSingleton

1. Applied ReserveSpace to replace Epilogue for passing auxiliary
pointers between FWD and BWD.

* Fix a logical error and enhance UTs.

1. Added act op count checking in UTs.
2. Fix issue to fuse backward or ReLU(Linear(X)).
3. TODO: solve GELU fusion issues.

* Fix Linear and GeLU fusion issues.

1. Modified graph_detech_pattern to fit with both linear wiht gelu or
relu.
2. Modified data range in Uts to allow negative values.

* Removed fused_gemm_epilogue_op.h.

* Rename namespace pten to phi.

* Rename name of arguments in fused_gemm_epilogue_op

1. bias -> Bias.
2. out -> Out.
3. reserve_space -> ReserveSpace.

* Change EpiloguePassActivationCache as local variable.

1. Removed singleton in EpiloguePassActivationCache.
2. Made EpiloguePassActivationCache as an argument to each pass
functions.

* delete gloo connect retry

* the_one_ps dirs reconstruct

* .

* .

* create the_one_ps dirs

* create the_one_ps dirs

* create the_one_ps dirs

* create the_one_ps dirs

* create the_one_ps dirs

* create the_one_ps dirs

* the one ps dirs modify

* the one ps dirs modify

* the one ps dirs modify

* the one ps dirs modify

* refactor ps optimize

* refactor ps optimize

* refactor ps optimize

* .

* .

* .

* .

* .

* .

* refactor theoneps

* the_one_ps

* add ps pass unittest

* add ps pass unittest

* ps unitest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* ps unittest frame

* add cpu_async_ps_mode test

* add cpu_async_ps_mode test

* add cpu_async_ps_mode test

* ps unittest ready

* ps unittest ready

* solve dist_pass init conflict

* solve import CommContext error

* unittest ok

* implement AllocateFrom

* solve setup.py.in conflict

* solve conflict

* solve conflict

* solve conflict

* .

* .

* cpu-async-ps minimize test ok & gpu minimize test ok

* add heter 2stage unittest

* add heter 2stage unittest

* add heter 2stage unittest

* sync/geo test ok & fix heter_worker program ok

* .

* new fleet desc generator

* new fleet_desc builder

* new fleet_desc builder

* .

* .

* correct ps.proto compile

* .
Co-authored-by: Nzkh2016 <zhangkaihuo@baidu.com>

* open eager when WITH_PYTHON, test=develop

* refine, test=develop

* refine, test=develop

* add DWITH_PYTHON for gen_fluid_lib, test=develop

* update;test=cpu-py3

* add processgroup_nccl

* disable some distribute test case when in CPU test env, test=develop

* refine, test=develop

* refine, test=develop

* refine, test=develop

* add DistributedFusedLamb op

* polish code

* fix compile error

* compatible with pten changement

* fix rocm compile error

* improve converage

* update upstream/develop

* fix cast_with_ptr.h

* add FLAGS_distributed_lamb_divide_nranks_when_allreduce=1

* fix clip before allreduce

* add use_master_param_norm

* code polish

* fix bug

* fix ROCM ci