* optimizer conv alog speed

* code polish

* remove useless code

* fix compile error

* fix cpu compile error

* not use cudnn alog t

* add search cache max number

* polish code

* fix cache test bug

* add groups data format to conv args

* fix cache test bug

* fix cudnn_deterministic bug

* fix test switch auto tune bug

* fix test swith autotune bug;

* fix conv cache bug

* fix cache test error

* fix cache test bug

* fix windows mac compile error

* fix workspace search error

* update cudnn cache

* fix cache test bug; test=develop

* fix autotune swith test error

* polish code

* oplish code

* fix unique_kernel bugs

* fix unique kernel cu bugs

* add python final state

* fix bug

* fix bugs

* fix bug

* fix bug

* revert impl, final state mul not support selected rows

* fix softmax use cudnn error

* add softlable false unitest

* revert loss.py

* fix bug fix

* fix utest

* fix utest

* fix utest

* fix log

* fix random utest

* make tensor_util contains no cuda code

* refine isfinite

* revert ut

* move isfinite function to its op

* fix test

* fix compile

* std::isnan is not defined for int type on windows

* fix windows compile

* fix fp16

* fix rocm compile

* revert gradient node

* fix arm compile problem

* fix format

* fix ci error

* add legacy_api.yaml

* set merged_momentum inplace only

* support inplace optional<vector<tensor>>

* add dygraph_mode api

* optimize TensorToConstDenseTensorPtr

* 【Hackathon No.34】优化 poisson op

* [poisson] code style fix

* modify code style

* prevent from big number

* modify code style

* modify code style

* modify import

* modify import

* modify code style

* support beam_search operator on xpu. test=kunlun

* support beam_search operator on xpu. test=kunlun

* support beam_search operator on xpu. test=kunlun

* support beam_search operator on xpu. test=kunlun

* support beam_search operator on xpu. test=kunlun

* support fp16 of adam operator in xpu environment. test=kunlun

* support fp16 of adam operator in xpu environment. test=kunlun

* support fp16 of adam operator in xpu environment. test=kunlun

* Adapt minlength attr for bincount

* add kernel dtypes check

* polish details

* fix print info, test=document_fix

* skip repeated kernel, test=document_fix

* remove test kernel, test=document_fix

* fix

* optimize

* while support for python container.
It is convenient to convert more dynamic graph codes into static graphs.

* cond support python container

* add quant pass

* rename the member function of SparseTensor

* use shorter function names

* erfinv

* fix some tiny issues

* Update scope.h

* typo

* Update dense_tensor.inl

* move distribute_fpn_proposals

* fix some code

* fix yaml bugs

* add set dtype

* move proposal_impl to funcs

* fix compile bugs

* [CustomDevice] add profiler apis

* migrate CalculateEstOccupancy into cuda_tracer

* update

* add ut