* fix distributed_fused_lamb nan

* remove CUDA_ASSERT

* add reduce_mean,reduce_sum primitive ops

* add ne_p gt_p primitive operators

* add ge_p abs_p primitive oparators

* clear extra attr of activation op in opmaker

* fix syntax bug

* fix mkldnn kernel

* fix merge conflict

* fix bug

* make memcpy op to support custom device

* fix bug

* add operator<< for BuildStrategy

* add fake_coalesce

* fit allreduce mode for new_exe

* remove dubeg code

* follow comments

* migrate load kernel

* remove load op

* fix test failed

* [CustomDevice] add dy2static support

* update

* add fusion dir and fuse_softmax_mask kernel

* remove fusion kernel dir

* migrate infershape

* fix code errror

* gaussian random

* mkldnn to onednn renaming

* fix merge conflicts

* remove fluid code

* onednn renaming

* Move classes from mkldnn_reuse.h to onednn_reuse.h

* Migrate pool+grad, clip+grad and cast oneDNN kernels to PHI

* Refactor grad kernels into separate files

* Fix CI failures

* Fix Codestyle

* Implement reviewer suggestions

* Add new lines after includes for readability
Co-authored-by: NSilv3S <slawomir.siwek@intel.com>

* add gemm_epilogue

* xpu-paddlepaddle-40 [任务] fused_gemm_epilogue 支持 test=kunlun

* stop pass filter int32/int16/int64/bool inputs in cond_op

* fix bugs: except block 0, the backward vars and forward vars exist in different blocks.

* fix code by review

* add save kernel

* add save_sr_kernel

* remove original save_op

* add save gpu kernel

* remove combine kernel

* add port.h include

* add save selected rows test

* remove useless kernel.h

* Construct exec and ctx only once in cond op to speed up

* Fix construct function error

* gaussian random

* mkldnn to onednn renaming

* fix merge conflicts

* Migrate reduce_op oneDNN kernels to phi

* Remove unnecessary header

* remove fluid code

* onednn renaming

* Change std::vector<int64_t> to IntArray

* Fix code style

* Move classes from mkldnn_reuse.h to onednn_reuse.h

* Move more functions from mkldnn_helper.h to onednn_helpper.h

* Change MKLDNN to OneDNN in VLOG message

* Implement reviewer suggestions
Co-authored-by: NSilv3S <slawomir.siwek@intel.com>

Adapt tensor output_size for conv2d_transpose and depthwise_conv2d_transpose

* clear extra attrs of reduce op in opmaker

* fix reduce_mean

* [MLU] fix sync_bn of mlu and add unittests

* [MLU] remove redunant code of pytest

* scale kernel

* endline

* add inplace

* fix merge conflicts

* Merge conflicts

* add tensor array

* fix ci bugs

* fix ci bugs

* fix ci bugs

* fix ci bugs

* update by comment

* update code

* concat workaround

* CI rerun

* set use_cudnn=true for conv2d

* clear opmaker of matmul_v2

* fix bug of set_attr

* add extra attr checker in infer_shape