* fix gather op and add logsumexp op on kunlun

* update xpu depence

* update tests and fix elementwise_add

* Fix gather infer shape using axis (#33413)

* fix gather shape bug

* fix None

* fix topo

* Fix hang of hybrid parallel in new_group  (#33141)

* fix hang of hybrid parallel

* fix new_group for hang problem

* fix hang

修复pylayer 返回to_tensor时触发段错误的bug。
原因：

如果在Python端修改了stop_gradient属性，c++ 端InnerSetOverridedStopGradient 无法修改stop_gradient属性，在c++端调用SetOverridedStopGradient修改stop_gradient属性。
to_tensor产生的tensor的grad var的DataType为默认值（-1），在backward的过程中grad var的DataType不能为默认值（-1），因此在调用ForwardDataType设置grad var的DataType。

原始PR：#33303

* Eliminate numerical differences of LayerNorm; fix LayerNorm Nan Bug while large data input

* fix bug while large shape of data input

* add unit8 for concat (#32850)

* add bool type for tril api (#33402)

Support diff dataset tensor place in single process dataloader

cherry-pick of #33470

使用op benchmark时发现，当输入数据量小于某个值时，python 端 log_softmax 接口的输入值经过计算过后 会被改变为nan。输出正常。

cherry-pick自 #32937

* Fix two english api documents, transpose and strided_slice

* OP:strided_slice_op supports bool type inputs

* fix compare op in for in the cuda device

* fix the paddle compare op for the broadcast

* fix compile error on jetson platform

* remove unused head file

* rm decode_jpeg op on jetson platform

修复了py_layer_op由于没有析构PyLayerContext造成内存(显存)泄露的问题。

原始pr：#32707

* clear 'BasicEngine' when an exception occurs in the backward. (#32546)

* clear 'BasicEngine' when an exception occurs in the backward.

* deal with conflict.

* deal with conflict.

* forward return any type. (#32661)

* base changes for fix

* minor change

* fix for bwd kernel

* removed unnecessary import

* implemented reviewers suggestions

* CI fix

cherry-pick:change softmax_with_cross_entropy_op's parameter name from softmax_switch to use_softmax (#32750)

* change parameter name from softmax_switch to use_softmax, test=develop

* cherry-pick:change parameter name from softmax_switch to use_softmax, test=develop

* add relu6_ hardsigmoid_ leaky_relu_ Inplace APIs

* add softmax_with_cross_entropy_ Inplace API

* add clip_ scale_ add_ subtract_ Inplace APIs

* add wlist

* fix parameter of scale api

* add add_n_ Inplace API and remove log_ Inplace API

* fix elementwise_add_ and elementwise_sub_ broadcast problem

* elementwise inplace api give error message before run the op

* use broadcast_shape in elementwise inplace op

* add 8 inplace apis that is auto generated

* add unittest for all inplace apis

* add decorator for inplace apis in static mode

* fix windows blas fail of exp inplace api, change array_equal to allclose

* add flatten inplace api

* add flatten unittest

* fix flatten unittest

* add decorator

* fix grad.numpy in test_pylayer_op

* unsupport softmax_with_cross_entropy_

* add test_inplace_softmax_with_cross_entropy to static_mode_white_list

* delete __all__ in inplace_utils

* delete activation inplace function and add Tensor.inplace_func

* change paddle.inplace_ to Tensor.inplace_

* fix little problem

* add paddle in inplace_utils

* add op read_file and decode_jpeg

* Add bf16 uniform random initializer

* Remove duplicated section

* Change UT to CPU place only

* Put detail functions into anonymous namespace

This is cherry-pick of #32281

- Executor is nt always having FLAGS_use_mkldnn set to true

* optimize update_loss_scaling_op by fused for loop to one kernel, test=develop

* remove useless while loop and optimize variable name, test=develop

* optimize variable name from out_addrs_tensor to out_addrs_mem, test=develop

* optimize variable name for readable by change prefix identifier from t_ to local_

* [OPs] Bug fix, fix the segment mean for illegal syncthreads usage.

* new optimize for where_index_op with prefix sum version.

* write a scan prefix sum kernel with stream for where index op.

* optimize where_index by using cub::DeviceScan::InclusiveSum instead of imperfect self-kernel.

* remove CheckTrue struct and rename stide_array for readable.

* optimize variable name for readable.

* optimize function name and annotation.

* fix model parallel

* rm parallel_help.py

* add embedding

* support backward return None.

* edit unittest.

* edit code according to CI

* Improve error information

* optimize slice op and slice grad op, test=develop

* optimize variable name and annotation information, test=develop