* trt support 0 dim

* trt support 0 dim

* update activation ut

To make it synchronized at the first recv operator.
If warping all send and recv operators with group start and end, the received tensor will be not complete.

* clang format

* add cumsum_grad op to xpu2_op_list

* 'fmin'

* 'fix'

* 'fix'

* test py3.8

* fix

* test gcc12

* test gcc12

* test gcc12

* test py3.8

* test py3

* fix

* fix

* add ubutnu20

* add ubutnu20

* add ubutnu20

* test py3

* fix error

* fix error

* fix error

* fix error

* fix error

* update py version

* fix

* test docker

* test docker

* test docker

* fix pip version,use pip3.9

* log error

* log add

* test ci

* add test log

* del

* fix

* test ci

* test ci

* update dockerfile

* update ci_dockerfile

* update ci_dockerfile

* update ci_dockerfile

* fix

* add git

* fix

* add mul doubel grad

* add sub_double_grad

* add add sub high test

* add mutiply test

* modify other unsqueeze

* delete api.yaml

* only for make ci run

* midify unsqueeze

* modify unsqueeze

* tmp

* modify operants gen

* change judgement for DropoutGradGPUKernelDriver

* add UnrollerWithoutVecSize and after this Loaddata to be refined

* pass unittest

* use same unroller with XPU

* BroadcastWithInt64Index

* BroadcastDataLoader template partial specialization

* fix compile errs in ROCms

* PR comment

* [phi] move sequence_pool kernel to phi

* mv kernels impl

* fix parameter error

* clean include

* fix compat filename

* [phi] move fluid sequence_pool_grad to phi

* [phi][compat] sig rm GradVarName

* [phi] fix sequence_pool out type

* [phi] rm impl, add const string

* [phi] fix const str

* fix sequence_pooling cmake

* [phi] mv sequence_pooling_test

* [phi] fix grad sig

* [phi] fix sequence_pool is_test error

* [phi] fix sequence_pooling gpu include

* [phi] mv to impl

* [phi] fix SequencePoolFunctor cu include

* [phi] modify out max_index int32_t

* [phi] add pooltype mapping determine

* [phi] fix sequence_pool_sig

* [phi] fix sequence_pool_sig sum

* [phi] try ci

* [phi] fix max_index optional

[Dy2St]Get grad names when call append backward to fix high order gradient (#53250)

* Update slim approve list

* Fix id, test=document_fix

* [static op generation] triangular_solve

* [phi] mv triangular_solve_grad to static_backward

* [phi] fix import

* [phi] mv to ops.yaml、 backward.yaml

* fix forward attr

* [phi] fix triangular_solve_grad args

* [CINN Support 0D-Tensor] CINN supports 0D-Tensor with trick temporarily

* Add unittest

* support fp16 for maxout op

* format code

* change api

* add test for static float16

* format code

* formatting code

* atol alignment

* experiment—1

* experiment-2

* experiment-3

* format code

* trans fused_feedward Compute function to phi

* add register info

* remove maxfunctor

* move fused feedward to phi

* remove sig file

* remove fliud include

* add include

* add include

* add sig file

* add output register info

* fix sig file

* Update fused_feedforward_sig.cc

* fix grad kernel

* update output register info

* fix

* open fused_feedforward static build

* add optional and fix code style

* fix output info for fused attention

* add optional param

* merge

* support OD level and skip dynamic loss scaling for bf16

* [Fix CppExtension Unittest] Change CUDAExtension to CppExtension if necessary

* Temporarily test cpp_extension under GPU

* Split mixed_extension unittest

* add jacobian and hessian in paddle.autograd

* disable unitest 'func_multi_input' for bug in high-order gradient of multiply

* add dimension checks

* add support for 0-D tensor

* change return type from Jacobian to Hessian in hessian function

* refine Jacobian _flatten function for single xs

* refine support for 0-D tensor

* 1. add 'func_multi_input' unitest for multiply_grad_kernel bug fixed
already.
2. support non-inplace math operation via magical method overwriting.

* add unitest for math operation and raise error when 0-D tensor is indexed

* add ndim check on ys and xs according to is_batched, and add one unitest

* refine docstring of jacobian and hessian

* move paddle.incubate.autograd.Jacobian/Hessian to paddle.incubate.autograd.functional.Jacobian/Hessian

* remove single_input unitest case because numerical differentiation is wrong

* remove 3 unitest for numerical result(reference result) is wrong

* 1. rename autodiff.py to autograd.py
2. increase TIMEOUT to 100

* cancel modification for functional Jacobian/Hessian

* 1. use tuple as return type instead of list
2. refine docstring

* add more unitest case to improve coverage

* remove 2 unitest of Hessian for numerical result is wrong

* remove 1 unitest of Hessian for numerical result is wrong

* remove 1 unitest of Hessian for numerical result is wrong

* change unit test to shape check

* correct doc and replace incubate API to stable API in _grad

* modify concat_grad add sum comp rule

* modify opcompat

* updata Adamw.py

out.backward()  -> loss.backward()

* Update adamw.py

* test,test=develop

* test,test=develop

* test,test=develop

* test,test=develop

* test,test=develop

* test,test=develop

* test,test=develop

* test,test=develop

* [XPU] remove scale_loss in parallel.py

* [XPU] throw Unimplemented when using Reducer

* update

* fix bug