* fix_paddle_numel_check

* fix_paddle_numel_check

add a inner loop for index_select_grad_init() in index_select op when dealing with large-shape data (#41563)

* replace for with CUDA_KERNEL_LOOP for index_select_grad_init() in index_select op

* use CUDA_KERNEL_LOOP_TYPE

* fix code style

* replace index_select_grad_init with SetConstant

* add new method for custom double grad

* add tanh double grad unittest

* change year

* revert tensor init method

* support setting vector out size in yaml

* support setting size of vector<tensor> for out in yaml

* add no need buffer

* add no need buffer

* remove determinant no need buffer

* add multi_dot,maxout,multiplex yaml

* add code converage

* Modify op-benchamrk script

* fix

* add assign yaml

* add assign api

* add assign backward api

* add assign

* add assign yaml

* add assign

* assign yaml

* add assign raw kernel and use assign_raw in yaml

* merge develop branch

* add missing python_api

* gather op

* add mod

* [Yaml] final state for uniform and uniform_random

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

* delete useless comment

* delete useless comment

* full api fix

* when out is None, go old dygraph mode

* fix

* add name for buffer

* fix by code review

* fix

* by static check

* add mkldnn int8 pass

* add mkldnn int8 pass

* update pass

* update name

* update name

* fix test

* fix fleet bind

* update name

* update name

* fix test

* fix gpups wrapper

* remove Push/Pull/Load/Save with context in client and wrapper base class

* fix

* fix

* remove some interface

* fix

* remove

* code style

* recover

* fix

* remove code unused

* remove some unused table & accessor & CommonDenseTable => MemoryDenseTable

* fix

* fix

* fix

* recover

* remove unused code

* recover unittest

* fix

* remove

* fix

* remove code unuseful

* remove

* fix

* recover

* remove
Co-authored-by: Nesythan <esythan@126.com>

* 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