[prim] modify multiply_double_grad composite rules to solve paddlescience poiseuille_flow.py problem (#54908)

* modify eular_beam

* modify matmul infermeta

* add test

* modify timeout

* modify mutiply_double nullptr

* modify tanh_triple_gradnode create

* COPY FROM issue of paddle nn

* 移除额外添加的示例，并修改Examplex->Examples

* remove example name of transformer

Co-authored-by: NSigureMo <sigure.qaq@gmail.com>

* fix vpp mp init prob

* triger ci

* update dygraph collective

fix ut

* remove debug log

* style

* more

* update ctest

* Update legacy_backward.yaml

* Update legacy_ops.yaml

* Update legacy_ops.yaml

* update

* update

* update for move

* add skip_gc_vars for 1f1b schedule mode

* add pp_degree and pp_stage

* remove function sequence_conv_pool

* fix test_glu.py

* remove function scaled_dot_product_attention

* remove function img_conv_group

* remove function simple_img_conv_pool

* delete the nets.py

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* Update test_image_classification.py

* Update test_recognize_digits.py

* Update test_recommender_system.py

* Update test_image_classification_fp16.py

* fix bug

* fix bug

* fix bug

* remove the nets.py in fluid

* fix bug

* fix the codestyle

* fix conflict

* fix bug

* Fix TODO

* add clone for inpalce

* fix name

* add inplace pow

* fix typro

* add note

* fix typro

* fix typro

* fix bug

* fix test error

* add type error test

* adjust indentation

* add gc for multi jobs

* fix job.h

* update OpInfo to OpInOutInfo

* update get_skip_gc_vars algo order

* demo1

* add test cases

* modify the usage of StepDecay

* refine

* replace PiecewiseDecay(LearningRateDecay) with PiecewiseDecay(LRScheduler)

* fix bug

* fix bug

* replace the StepDecay,MultiStepDecay,LambdaDecay with 2.0 version

* add pos_weight in kernel

* fix unittest

* fix xpu

* fix bce unittest, change infermeta order

* [ROCM] fix cupti, hipcub

* update

* update

* [kunlun] fix the dataloader problem in RDMA env

When running multi-machine training with Paddle DataLoader, an
unexpected segmentfault will be raised in DataLoader Process,
where the traceback goes all back to a runtime error that dataloader
workers exit unexpectedly. Similar problems have been discussed
that lead to a misbehavior of OpenCV working in multiprocessing
environment.
See
https://stackoverflow.com/questions/54013846/pytorch-dataloader-stucked-if-using-opencv-resize-method

* code style

* fix 'RuntimeError: context has already been set'

* Update dataloader_iter.py

spawn method raise error 'Can't pickle local object' in some situations

* code format check

* code style

* modify eular_beam

* modify matmul infermeta

* add test

* modify timeout

* base rule

* add sharidng merge

* add sharidng axis merge

* define unified data class for inferencing dist_attr

* test wrap DistTensorSpec in dygraph mode

* matmul main logic done

* define unified data class for inferencing dist_attr

---------
Co-authored-by: NYichen Zhang <zhangyichen03@baidu.com>

* remove the CosineDecay in fluid

* Update test_basic_api_transformation.py

* remove the NaturalExpDecay in fluid

* fix bug

* remove the ExponentialDecay in fluid

* remove the InverseTimeDecay in fluid

* remove the InverseTimeDecay class

* fix bug

* feat(cmake): add cmake of cinn

* feat(cmake): add cmake of cinn python test

* feat(cmake): add jit

* feat(cmake): test/CMakeList.txt

* feat(cmake): rebase to develop

* feat(cmake): remove some flags

* fix(cmake): fix cinn's gflags depends

* feat(cmake): add ci scripts of cinn

* feat(cmake): copy code of cinn

* fix(cmake): fix cinn third_party model path

* gflags dynamic dependce

* fix ci build_demo

* tmp update to c++17 of cinn-only test

* fix cinn only with c++17

* add fetch kernel

* support fetch var in new ir

* fix bug

* polish code

* change array equal to np.testing

* support feed in new ir

* fix bug

* try to hack combine op

* add scope guard

* revert atan2 op

* polish code

* [XPU] support xpu runtime profiler: follow up

* fix compile issue

* auto parallel support pipeline scheduler with standalone executor

* rm check_fetch

* update cmakelist and flags env

* rm set micro batch id

* rm import

* update utils func

* raise error when merge tensor for return_numpy is False

* fix _pipeline_opt

* fix unittest

* new_ir to shared

* refine code

* add ir lib path to env

* refine type

* refine code

* fix bug

* fix bug

* refine code

* refine code

* close win

* refine code

* refine code

* refine code

* add win share

* refine code

* refie code

* refine code

* refine code

* refien code

* fix bug

* fix bug

* fix bug

* solve conflict

* solve conflict

* fix bug

* refine code

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* refine code

* fix interpretercore program bug

* delete unuse code

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* fix cinn bug

* fix cinn bug

* debug

* fix cinn bug

* delete unused code

* fix cinn bug

* fix cinn bug

* fix  ug

* test win openblas

* test win openblas

* fix win openblas bug

* polish code

* fix win open blas bug

* close win dll

* fix flag bug

* test for windows

* fix compile bug

* add auto tuner

* compare and record module

* revert launch main

* add prune rule

* add unit test

* add auto tuner

* revert launch main

* add prune rule

* modify unit test script

* fix bug for dump nodes; fix bug for checking log file

* fix bug

---------
Co-authored-by: Ncaozhou <caozhou@radi.ac.cn>

* no endpoints in dy mode

* fix fleet api inconsistent

* Run plan in standalone executor

* Update codes

* Update atol and rtol for py3-CI

* Add scope to cache key

* Fix CI errors

* Fix code style

* Update codes

* Remove fetch_name in standalone executor

* Fix UT

* Update codes

* Fix new IR bug

* fix lamb always_adapt

* fix optest

* fix all optests