cherry-pick #53659

Fix bug in log_softmax kernel when lastdim is larger than 100000

There is an unexpected log in the calculation

Cherry-Pick: #53654

Revert argsort to the version without full sort algorithm implemented

* Support different dtypes of inputs for broadcast for dropout optimization  (#52093)

* 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

* dropout_nd_optimization (#51479)

* with printf

* add DropOutNdForwardKernel

* PR comment

* Dropout optimize & clean broadcast inT and ElementwiseType (#52969)

* 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

* clean ElementwiseT and InT for BroadcastKernel

* default axis and clean inT

* remove redundant fast divmod computation

* optimize drop_nd & drop_nd_grad

* optimize BroadcastDataLoader bf16 fp16

* rm InT etc. after merge develop

* delete constexpr for windows ci

* fix conflict

* fix conflic with develop

* fix conflic

* new clean

* clean

* Fix xpu2 kp compile error (#53548)

* fix conflict

* conflict

Cherry pick fused linear

* add complex support for  optest

* add complex grad test

* append one

* move some debug info

* move some debug info

* move some debug info

* move some debug info

* add more complex test

* Fix naming ambiguity

* Revert "add more complex test"

This reverts commit dbcb0516b8e53ba42e2d6089878a39b395345969.

* change backward gradient, add TODO

* [Zero-Dim] fix functool.reduce more safe with intial value, to support empty list (#53182)

* [Zero-Dim] support 0d tensor for shape and squeeze onednn kernel (#52832)

* support 0d tensor for shape and squeeze onednn kernel

* set python api for shape op ut

* [Zero-Dim] distributed scatter/all_to_all support input 0D tensor (#53186)

* [Zero-Dim] Support paddle.sum/mean/loss api output 0D,test=allcase (#52739)

* [CINN Support 0D-Tensor] CINN supports 0D-Tensor with trick temporarily (#53382)

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

* Add unittest

* [CINN Support 0D-Tensor] CINN hack squeeze2 with trick temporarily (#53454)

* fix test_autograd_dynamic (#53473)
Co-authored-by: Nzhwesky2010 <zhouwei25@baidu.com>

---------
Co-authored-by: NYangQun <qun.yang@intel.com>
Co-authored-by: NHongyuJia <jiahongyu@baidu.com>
Co-authored-by: NHydrogenSulfate <490868991@qq.com>

* support 0-D output and 0-D as indice in __getitem__

* fix tests

* fix inference and UT

* add unittest for setitem

* fix xpu test

* fix xpu 0-d

* fix right value is 0d and index is List/Tensor

* Hack__getitem__ from 0-d to 1-d with FLAGS_set_to_1d

* change PHI_DECLARE_xxx to DECLARE_xxx since the change not merged to 2.5

* hack 1-D tensor to Scalar

* throw warning at __getitem__, not slice_utils

Cherry-Pick: #53582
修改内容：在除法out = x / y中，将y的反向公式由dy = -dout * out / y 改为 dy = -dout * ((x / y) / y)
修改原因：使用result作为反向的输入，在低精度的时候本身cast之后就会存在一些精度损失，所以重新计算后才是更准确的结果
修改影响：此改动可以使结果更精确且对性能影响忽略不计

* Add fused_gate_attention API. (#53432)
* Add PADDLE_THROW in take_along_axis kernel when the datatype of index is wrong. (#53556)

* add 0D output support for inalg.slogdet,test=allcase

* fix zerom dime test error test=allcase

* fix test error test=allcase

* add static backward test, test=allcase

* support_0D_output_for_matrix_rank_multi_dot, test=allcase

* add 0D output test for matrox_rank and mutli_dot test=allcase

* fix assert error ,test=allcase

* fix test error, test=allcase

* fix other test error, test=allcase

* fix other test error, test=allcase

* fix test error, test=allcase

* fix matrix_rank and multi dot test err test=allcase

* fix test error test=allcase

* fix test zero dim test, test=allcase

* add static backward test for multi_dot, test=allcase

* add tol 2d broadcast test case, test=allcase

* fix test error test=allcase

* fix test error test=allcase

* test=allcase

* support_0d_output_for_linalg.norm

* fix test error test=allcase

* fix 0D test

* fix test error test=allcase

* fix test error test=allcase

* fix tets,test=allcase

* fix error,test=allcase

* fix errors ,test=allcase

* add static backward , test=allcase

* add static backwward test, test=allcase

* slogdet_support_0D_output

* add new case

* fix tests, test=allcase

* cherry-pick

* cherry-pick

* fix trace gpu kernel 0d error, test=allcase

* fix windows error, test=allcase

* add matrixrank cherry-pick

att, cherry-pick: #52902 #53113

低精度算子支持和单测补充，合并 cherry pick 17个Hackathon PR，共覆盖25个OP的低精度支持及完善

[cherry-pick2.5] [Zero-Dim] Support all/any/min/max/prod/logsumexp/amax/amin/some loss output 0D (#53192)

[Cherry-Pick]Support output 0D for is_empty/as_complex/inner/dot/rank/tensordot/squeeze_/static.accuracy/static.auc/metric.accuracy (#53199)

* support output 0D for is_empty/as_complex/inner/dot/rank/tensordot/squeeze_/static.accuracy/static.auc/metric.accuracy

* test_dot_py

* test_dot_py

* Reduce inference library size and compilation time
* fix docstring

This reverts commit a79c04f3.

* support 0-D output and 0-D as indice in __getitem__

* fix tests

* fix inference and UT

* add unittest for setitem

* fix xpu test

* fix xpu 0-d

* fix the set_value error in cpu

* add a unitest for set_value OP

* fix platform::is_gpu_place

* add todo note for set_value

* fix test

* add 0D output support for inalg.slogdet,test=allcase

* fix zerom dime test error test=allcase

* fix test error test=allcase

* add static backward test, test=allcase

unique支持float16和bfloat16数据类型，并完善相关单测。

* add random control for fused dropout add

* add __init__

* fix multinomial

* fix test_elementwise

* fix convert_float_to_uint16

* aadd test_multimial_op

* fix code style

* untracked files

* bce_loss_fp16

* remove unused files

* back max_rel_erro still big

* simplify code

* upd

* fix max_relative_error

* restart ci

* Update test_bce_loss.py

* Update test_bce_loss.py

* Update test_bce_loss.py

* Update test_bce_loss.py

* try to pass test

* restore file

* remove error value

* fix bug

---------
Co-authored-by: NZhang Ting <Douyaer2020@qq.com>

* fix multiply double grad error

* fix multiply dy only kenrel

* Implement optimized kernel for OP-expand_as.

* Support fp16.
Co-authored-by: Timber-Ye <ye_hanqiao@163.com>
Co-authored-by: NBrianQian1999 <brianqianhitsz@gmail.com>

* remove fp16 support

* remove MAX_RANK_SUPPORTED

---------
Co-authored-by: NBrianQian1999 <brianqianhitsz@gmail.com>

* add register info for eigh and eig_gard

* add sync_batch_norm_op.cu register info

* add lamb output register info

* add unique register info

* change type name

* change type name

* add output register info for check_finite_and_unscale

* update cmake and config file

* add register info for adagrad

* fix build error

* add sync to run_unittests.sh

* add register info for unique_consecutive

* fix build error

* add eigh to STATIC_BUILD_TESTS

* update eig_kernel.cc

* update eig_kernel.cc

* fix infer mate error

* fix unique register error

* fix lamb register info error

* fix lamb register info

* update lamb register info

* fix lamb

* remove one Output Register

* update static build file

* add eigh op to disable_wingpu_test

* update run_unittests

* Add digamma, dirichlet tests

* Fix code

* add erf FP16 test

[Zero-Dim] support 0-D tensor for reduce/reshape/stack/prelu/expand_v2/gaussion onednn kernels (#52185)

* support 0-D tensor for reduce/reshape/stack/prelu/expand_v2/gaussion ops

* fix gaussian random mkldnn op ut

* [phi] move sequence_pool kernel to phi

* [phi] mv sequence_pooling to phi funcs

* [phi] mv sequence_pooling_test

* [phi] RollBACK `paddle/fluid/operators/sequence_ops/sequence_pool_op.cc`

* [phi][funcs] fix mutable_data

* [phi][funcs] fix mutable_data