* Merged LoDTensor with Tensor,test=allcases

* Patched python level LoDTensor

* Patched python level LoDTensor

* Merge Tensor into DenseTensor

* Fixed namespace issues,test=allcases

* Fixed merge issues

* Fixed inference issues

* Fixed NPU test issues

* Fixed merge issues

* add pten::Place data structure.

* update ci problem

* fix ci problem

* update

* using platform::Place=pten::Place

* remove BOOST_GET_CONST for CPUPlace and GPUPlace

* compile pass 25%.

* compile pass 45%

* compile pass 60%

* remove boost_get for xpu npu mlu and ipu

* compile pass on cpu and gpu.

* fix compile problem

* fix compile error.

* update

* fix ci problem

* update

* ci approve

* fix ci problem

* fix ci eager test problem

* remove BOOST_GET_CONST

* fix npu compile

* first commit

* fix wrong filename

* fix the wrong spell name

* fix gpu config warper

* modify according to pr advices

* fix GpuLauchConfig1D api bugs

* change the config for dropout grad

* fix bugs

* modification according to pr advices

* modification according to pr advices

* Add fp16 support for scale/bias for fused_layernnorm_residual_dropout_bias op.

* combine reduce_cuda codes

* support float16 in pten redcue_mean

* replace ReduceCudaKernel impl with pten reduce impl

* mv reduce funcs into reduce_cuda_impl

* rm unsed codes and headers

* mv GetReduceDim into reduce_cuda_impl

* recover GetReduceDim in reduce_op.h

* add new dispatch macro

* fix pool op output not inited and cause transform to pten::denseTensor error

* fix output tensor not initialized error

* rename new dispatch macro and format code style

* rm reduce_functor_op.h file

* Add the transformop parameter in TensorReduceFunctorImpl

* refine structure for cuda and rocm

* update

* update

* update

* update

Add support for bias is none for fused_attention op.

* GeneratePass support attr condition and mapping, test=develop

* fix coverage, test=develop

* Add fuse_resnet_unit pass, test=develop

* fix CI errors, test=develop

* fix CI errors, test=develop

* fix unittest error when compiling without CUDA, test=develop

* fix static ci error, test=develop

* limit kernel size must equal 1, test=develop

* Change first batch of mkldnn headers and namespace names to dnnl

* Revert changes to tensor.h, which require approval

* Format changes with pre-commit

* Add int32 tests

* Fix int32 tests and call GetDataFromTensor for int32

* Fix test

fused_attention_op的实现中，使用了bias_add，且其实现是通过使用kernel primitive来实现的，之后kernel primitive的WriteData api接口及函数内部实现发生了更改，将判断越界的逻辑移到了template的参数中，使得调用的分支有错误，产生了越界赋值操作，污染了别的显存空间的内容。具体表现为：test_fused_attention_op_api.py 单次执行基本上不会报错，多次循环执行不同shape的输入，结果计算不对，具有偶发性，bug不易察觉。

* fix bug:
1. atten: set the default value of attn_dropout_rate to None
2. ffn: add activation parameter

att, bug fix

目前的fused_attention_op不支持attn_mask=None的输入，本PR对此进行了补充，并补充了相应的单测逻辑。

* - WIP

- compilation fix

- fix

- fixes

- fix

- fix

- fix again

- fix

- another fix

- another compilation fix

- fix

- fix

- fix

- lint

* - pool2d partially stripped from cache

- pool2d partially stripped of caching

* - compilation fix

* - compilation fix

* - Fix to UT of caching

* - Enabling test_conv3d_mkldnn

* - conv_transpose stripped of cache

* - compilation fix

* - fix

* - fix

* - compilation fix

* - fix

* Reverted disabling caching of conv2d

* - compilation fix

* - ut reverted

* Fix bug when pre_layer_norm is false.

功能：本PR的目标是提高attention模块的计算性能。
为了减少框架层对op的调度开销，本PR通过在C++层手动实现attention模块，对外提供attention 大op；
为了减少防存开销，本PR采取了两种优化方法：
（1）在q,k,v计算时通过共享输入X，将该处的gemm，transpose和bias add从三次调用减少为一次；
（2）使用kernel融合优化技术，在不同cuda kernel之间通过寄存器传输数据；

将 #35905 和 #35843 PR中新增的的python api接口移到incubate目录下。

这个PR是fused_feedforward反向的代码

相关kernel实现：fused_dropout_act_bias, fused_residual_dropout_bias, fused_layernorm_residual_dropout_bias

fused_feedforward是一个融合算子，该算子对transformer模型的feed forward层的算子进行融合和封装，使得前端只呈现一个接口，通过融合减少部分访存和kernel launch的时间，以此提升性能。

这个PR只包含fused_feedforward前向的代码。

相关kernel实现：fused_dropout_act_bias, fused_residual_dropout_bias, fused_layernorm_residual_dropout_bias

fused_feedforward是一个融合算子，该算子对transformer模型的feed forward层的算子进行融合和封装，使得前端只呈现一个接口，通过融合减少部分访存和kernel launch的时间，以此提升性能。

功能：本PR的目标是提高attention模块的计算性能。
为了减少框架层对op的调度开销，本PR通过在C++层手动实现attention模块，对外提供attention 大op；
为了减少防存开销，本PR采取了两种优化方法：
（1）在q,k,v计算时通过共享输入X，将该处的gemm，transpose和bias add从三次调用减少为一次；
（2）使用kernel融合优化技术，在不同cuda kernel之间通过寄存器传输数据；

* Update the implement of reduceAnyKernel according to kernel primitive api
* Fix a bug in ReadData, ReadDataBc and ReadDataReduce when NX != 1

* broadcast qkv_op

* use PADDLE_ENFORCE_GT to replace assert

Fused elementwise_add, dropout, elementwise_add and layer_norm into one operator, only support Forward. 
No Python API changed.

Implement FunctionTraits to support two kinds of elementwise functor and remove some old codes for broadcast. (#35688)

Revert "Implement FunctionTraits to support two kinds of elementwise functor and remove some old codes for broadcast. (#35487)" (#35686)

Implement FunctionTraits to support two kinds of elementwise functor and remove some old codes for broadcast. (#35487)