* optimize performance of dygraph

* optimize performance of dygraph and elementwise_add

* optimize the trace op

* fix bug

* fix bug

* fix unittest bug

* fix code format

* unify utils naming style

* polish details

* [Performance]Remove redundant op_type  in RecordEvent

* [Performance]Remove redundant op_type  in RecordEvent

* [Performance]Remove redundant op_type  in RecordEvent

* optimize the performanece of sum api

* optimize IsDenseTensorInput

* remove debug log

* polish some impl

* add lost attr type

* polish details

* fix error type

* polish in name lists

* add double attr

* adapt infrt attr parse

* init commit

* remove two hash impl

* fix bug

* polish details

* fix compile failed

* fix compile failed

* fix compile failed

* add default kernel sig cache

* fix get kernel arg defs error

* remove kernel arg defs cache

* fix origin op execute

* merge all phi lib to several big static lib

* merge all phi lib to several big static lib

* opt dygraph scheduling

* revert part impl

* add variant and replace any

* split attribute

* optimiaze performance of PreparePhiData

* dygraph performance optimization

[Phi]Separate AddKernel/DivideKernel/SubtractKernel/MultiplyKernel from ElementwiseKernel（Part1） (#41806)

* seperate add/div/sub/mul from elementwise

* delete code

* fix compile bugs

* deal with conflict

* fix bugs when compile

* fix windows unit test bug

* fix ci converage bugs

* split phi and fluid infermeta context

* resolve conflict

* fix type error

* optimize scheduling perf

* spec small vector size

* replace all grad var name

* fix test failed

* move init defalut signature

* polish details

* polish details

* fix no init bug

* init sig for tests

* add init sig for infer

* fix infrt error

* fix infrt failed

* fix kunlun error

* fix infrt failed

* change cudnn helper for auto-tune

* Add FLAGS_use_autotune to set the global status of autotune and change the order of choosing algorithm.

* Fix the bug in calculating and printing current step cache hit rate.

* Improve the autotune cache and fix unittest.

* Change the key from AlgorithmType to int64_t.

* Fix unittest for cpu-only env.

* change ChooseAlgoByWorkspace for heuristic mode
Co-authored-by: NLiu Yiqun <liuyiqun01@baidu.com>

* remove stack_grad infershape

* fix bug of output with null

* fix bug

* [KP]fix bug when TruncatedNormal cannot fall back in cpu

* delete useless comment

* delete useless comment

* no

* maintain old profiler

* fix old dygraph record event

* switch autotune

* implement AutoTuneCache

* implement AutoTuneCache class

* add pybind api

* add dygraph test

* support static mode and eager mode and improve unittests

* rename the SwitchAutoTune Class and improve tests

* improve AutoTuneStatus and reduce the cost of tests

* [KP] fix bug in phi static graph mode

* modify the useless code

* add cross_entropy_with_softmax phi kernel

* remove softmax_with_cross_entropy kernel

* add softmax_with_cross_entropy grad kernel

* remove original op kernel

* refine cross entropy impl

* fix pointer error

* revert kernel cu change

* fix xpu failed

* fix cinn failed

* fix npu failed

* add forward sig

* add check_nan_inf for pt kernel

* remove repeat cmake item

* fix unittest error

* add interploate cpu kernel

* fix nullptr bug

* add interpolate gpu kernel

* fix unit test error

* remove raw kernels

* add cuda kernel impl

* add infermeta

* recover accidentally deleted kernels in interpolate op

* fix grad x_grad name error

* remove interpolate_v2_op.h

* rm unused codes

* fix xpu build error

* fix build error

* fix namespace error

* add register header for nup

* fix infermeta error

* modify by review

* add the missing args in test_trt_convert_nearest_interp_v2

* fix bug in activation xpu kp kernel

* delete useless comment

* rename scalar_array to int_array

* update cmake

* fix conflict

* remove useless log

* [KP] fix bug in phi kp

* delete useless comment

* update

* update

* choose the xpu kp kernel in phi

* remove in_dtype, out_dtype in redcue grad

* set the dtype and layout in noneedbufferInputs func

* add mul phi kernel

* remove mul op kernel

* remove original mul grad op

* fix cinn test

* fix dygraph test failed

* fix some slice bug, test=develop

* eager slice, test=develop

* eager slice, test=develop

* refine, test=develop

* refine, test=develop

* fix bug, test=develop

* refine, test=develop

* rename function name, test=develop

* Removed redundant use of declarations.h

* Fixed minor bug

* first commit

* fix compile error

* support std::vector<std::srting>

* fix

* fix op support on GPU by chenweihang

* pass test

* infershape

* add set_dtype

* fix order

* fix

* unify the impl of dt and sr

* fix

* move assign kernel init commit

* change vec<tensor> to vec<tensor*>

* support tensor array

* support api declare

* fix test_list failed

* fix npu and xpu failed

* fix infrt failed

* remove assign array size in operator

* move assign sr header into sr dir

* add infermeta for assign

* test op success

* fix test_list failed

* fix kunlun failed

* add set host allocator in tests

* support tensor array in arg ctx

* open set layout in share_meta

* fix meta tensor layout error

* fix test failed

* move roi_align_grad kernel

* move roi_align grad kernel and infershape to phi

* remove roi_align infershape

* [kp]fix bug that cannot fallback to CPU normally in XPU KP

* fix bug in static graph

* run python api in eager model and filter the out in argument list

* fix code

* [NPU] add AMP O1 support

* [NPU] fix NOTE and warnings