* Using the maximum workspace_size of all alogirhms to limit the workspace size in exhaustive search mode.

* Use the system cudaMalloc and cudaFree to allocate workspace during searching.

* Enable switch of two kind of workspace setting methods.
Co-authored-by: NLiu Yiqun <liuyiqun01@baidu.com>

* no

* maintain old profiler

* add get profiler from serialization config

* add unit test

* improve coverage

* fix

* Revert "improve coverage"

This reverts commit 4a980bfda48adadee551d0e1c5740bc5b7389200.

* fix unit

* fix

* fix

* fix_ci_problem3

* support windows no default error

* fix bug that no thread is waked up when adding task to threadpool

* fix typo

fix group_norm vectorized address misalignment

* remove elementwise add in conv

* use reshape

* modify unittest of lstm forward, *test=kunlun

* modify unittest of lstm forward, *test=kunlun

* [Eager]Fix segment_pool/allclose/isclose/scale API bug

* fix kernel register problem

* [ROCm] fix dcu error in device event base, test=develop

* fix, test=develop

* fix image cv2 import

* fix lamb

* ad conj flip yaml

* add flip conj pixel shuffle

* add swish yaml and final state api

* skip mkldnn test

* fix grad mkldnn test

* afs wrapper

* format

* format

* macro

* add dist erfinv gumbel softmax

* fix test gumbel softmax op bug

* try to fix gumbel softmax error

* add label smooth backlist

* add yaml

* perfect converage

* modify matrix_rank

* add matrix_rank shape

* add matrix_rank shape

* Add yaml for matrix_rank OP

* Add UT
Co-authored-by: Nzhoujianqian <15205085056@163.com>

* add unbind yaml

* fix unittest

* Sitch some dy2st UT to eager mode

* Fix test_lstm and remove test_transformer

* Run test_resnet_v2 in old dy mode

* Profile Executors

* update

* fix ut

* fix names

* update

* update