* add all_reduce, reduce kernel and api

* fix all_reduce reduce ut

fix reduce op maker conflict

fix merge conflicts

* fix conflicts, rename ReduceOp->ReduceBaseOp in reduce_ops

rename allreduce op, to remove

* fix code format

fix comments

* modify test_collective_reduce_api ut timeout

* fix PR-CI-Build

fix comments: format phi operator

* decouple memory copy

* fix ci bugs

* fix ci compile bugs

* fix rocm compile

* fix ci bugs

* decouple memory

* deal with conflict

* fix xpu compile bugs

* fix xpu bugs

* deal with xpu bugs

* fix cmake bugs

* fix windows bugs

* fix ci bugs

* fix ci bugs

* delete redundance code

* add code for pybind

* fix py3 bugs

* fix ci bugs

* [CUSTOM DEVICE]analysis predictor custom device support

* del debug log

* fix momentum dtype infer

* fix momentum datatype

* fix on cpu

* add momentum

* untracked files

* kthvalue perf

* remove unused files

* fix isnan

* fix isnan2

* fix bug

* try to fix rocm error

* first version, notest

* return final rst, notest

* use infinity() instead of max

* ut structure

* start up of ut

* generate lse

* update

* add depense

* reconstruct cmake

* move file

* add memory efficient attention and fix blasimpl

* update

* update cmake

* add namespace

* update cmake

* use .cu

* update for pad3d

* bug fix

* bug fix

* update

* bug fix

* update enforce

* add test case

* merge the lse pad

* fix kernel_fn of backward

* fix PADDLE_ENFORCE_EQ and phi_api

* fix PADDLE_ENFORCE

* fix PADDLE_ENFORCE

* rerun coverage

* fix memory efficient attention test

* rerun ci

* add cuda version condition

* add cuda version condition

* delete WIP test

* replace PADDLE_ENFORCE

* edit the namespace of datatype in multiple.cc

* rerun

* rerun

---------
Co-authored-by: Nliuyuang <liuyuang@baidu.com>

[Polish Log] Polish Tensor operants' log: 'OperantsManager reusing XXX mode API {func_name}' (#51991)

* [Polish Log] Polish Tensor operants' log: 'OperantsManager reusing XXX mode API {func_name}'

* Make API name more precise

* delete prim flag for matmul_2_grad

* delete prim flag for matmul_2_grad

* add new setgradoutmeta for matmul_double_grad_node

* modify test and delete log

* deal with review

* add meshgrid composite rule

* add meshgrid composite rule

* update

* add into CMakeLists

* fix

* update

* update

* optimize code

* fix meshgrid op

* update test

* remove fluid deps in fused_linear_param_grad_add_kernel

* fix compile error

* fix ut error

* follow comments

* first commit

* fix bugs

* remove_useless sync

* unify add_position_encoding

* unify affine_channel

* unify alloc_float_status

* unify allreduce

* unify alltoall

* unify anchor_generator

* unify ascend_trigger

* fix bug

* fix test

* update

* update

* update

* update

* update

* fix test

* to support cuda12, pybind need to upgrade to v2.10.0

* add DEPS of pybind in test_custom_plugin_creater.cc

* only change the tag

* please let CI pass

* try pybind v2.10/3

* modify the include header in test

* code check

* Add bf16 support for elementwise_pow

* Update ut

* add bf16 and fp16 tests

* fix dtype check

* fix distributed graph engine

* add fp16 and bfp16 for temporalshift

* add fp16 and bfp16 for complex

* fix bug

* fix bug

* add fp16 and bf16 for conj

* fix bug

* fix bug

* Update complex_kernel.h

fix bug

* Update temporal_shift_grad_kernel.h

fix bug

* Update temporal_shift_kernel.h

fix bug

* add nanmedian output defs

* remove the multiclass_nms3 momentum

[CodeStyle][C408][C409][C410] Fix unnecessary <dict/list/tuple> call and unnecessary <list/tuple> passed to <list/tupule>() (#51928)

* autofix

* add select config

* autofix C410

* add C410 select

* finish pr

* skip cpu test for logical

* change test style

* fix error.

* allow return none when stop_gradient=True

* remove useless code

* refine code

* refine code

* fix test cast

* change more test

* add more tests

* fix error msg in pylayer

* test_get_kernel

* add invoke signature

* change reduce_max

* change frobenius_norm

* reset reduce_max according to composite and change reduce_all

* fix the bug when Scalar(*)

* fix 'scalar when support_tensor'

* change code according to review

* change 'keep_signature' to 'manual_signature' and add some erro info

* support optimizers autogen

* change sgd yaml

* change generate signature

* fix test/cpp/new_executor/CM

* reset signature generated function

* change signature funciton

* change signature funciton

* support 0-d tensor for element wise unary ops

* fix python code style check

* fix approval check

* support 0-d tensor for onednn softmax and logsoftmax kernels

* fix commnets

* fix some unittests

* fix distribute_fpn_proposals

* fix bug

* Add fused_feed_forward pass for semi-automatic static graph training.

* Add fused_feedforward property in parallel_executor.cc

* Polish code.

* Polish fused feed_forward pass code. Support use_dropout1 and
use_dropout2 option.

* Support model parallel in fused_feedforward pass.