* modified SetInputArgModifyFn

* Delete the CHECK changes in the assign_op.cpp file

* Format

* Modified the OutputArgModifyFn interface

* add return

* maybe error stack from CheckAndConstructOp to OutputArgModifier callback function

* maybe error stack from CheckAndConstructOp to OutputArgModifier callback function

* OutputArgModifier return maybe part_1

* maybe error stack from CheckAndConstructOp to OutputArgModifier callback function

* input_arg_modifier return maybe

* gen_bw_fn return maybe

* bw_gen_fn return maybe

* registry_callback_fn return maybe

* fix bug after merge master

* fix bug

Co-authored-by: aishangjj <702572275@qq.com>

* refactor message OperatorConf, change device_type to device_tag

* subsititute HobDeviceType with HobDeviceTag in user_op kernel registration

* remove c_api_util.DeviceType4DeviceTag

* fix error when buil with cuda off

* fix can not use CHECK_JUST macro in another macro

* cmake dont panic when build cuda

* naive changes

* fix cudaMemcpyKind

* fix acc actor

* fix actor

* fix gdb

* fix vm

* fix work type

* fix cuda type

* fix cuda type

* fix collective backend

* fix global scope

* amp

* rm PackKernelUtil gpu

* fix log

* fix rand

* fix sync size

* fix allocator

* fix vm

* fix kernel

* fix kernel

* fix kernels

* fix kernel

* fix softmax

* fix kernels

* fix reshape kernels

* add workaround

* try fix symbol not found

* fix vm

* fix vm

* fix jpeg

* fix broadcast gpu

* fix broadcast like

* fix transpose

* fix matmul

* fix CopyElemOnGpu

* fix sigmoid

* fix sigmoid and softmax

* fix relu

* fix sparse cross entropy

* fix kernels

* fix tanh

* fix same padding

* fix softmax

* fix undefined symbol: gzgets

* fix CopyField

* fix scalar add

* fix CopyNDGpuImpl

* copier

* fix slice boxing

* fix mem copier

* fix zero like

* fix acc actor

* fix dev pure cpu (#3410)

* add pure cpu message

* add default for src dir in ci

* add cpu workflow

* fix extra_oneflow_cmake_args

* fix link problem add update readme

* fix _GetDefaultConfigProto

* use gpu

* rm default value for gpu dev num

* rm cpu ci

* check in skip code

* refine ci and add back

* add arg for appendix

* fix env arg

* fix env arg

* rm make nccl

* add back to pass gpu ci

* get with cuda in py

* fix fmt

* fix test kernels

* fix activation

* add skips

* fmt

* skip all reduce

* fix assign

* fix bn

* fix bn

* rm tmp_wheel

* rm tmp_wheel

* move clean up

* for CPU-only OneFlow make gpu_device_num equivalent to cpu_device_num

* change warning to info

* only run 3 iters for cpu

* skip bert for cpu

* fix check

* add warning

* add cpu Integration test

* Dev pure cpu test cases (#3422)

* fix test cpu cases

* change os.getenv('ONEFLOW_TEST_CPU_ONLY') == 'True' to os.getenv('ONEFLOW_TEST_CPU_ONLY')

* print traceback for info and warning

* fix test cpu cases

* add more info on why skip check of resnet

* skip gan in cpu

Co-authored-by: tsai <caishenghang@1f-dev.kbaeegfb1x0ubnoznzequyxzve.bx.internal.cloudapp.net>
Co-authored-by: tsai <caishenghang@oneflow.org>
Co-authored-by: oneflow-bot <69100618+oneflow-bot@users.noreply.github.com>
Co-authored-by: JackieWu <wkcn@live.cn>
Co-authored-by: OuYang Yu <xuanjiuye@gmail.com>

* conv pool padding

* same padding op

* nn ops

* test case

* rm same_padding num_spatial_dims constant_value

* conv pads len ndims+2 -> ndims

* pool pad len ndims+2 -> ndims

* polling dynamic use samepaddingop -> cudnn padding

* add test case

* rm useless

* conv pads -> padding_before

* format

* pool pads_before -> padding_before

* add conv dynamic test case

* pool add test case

* address review

* fix conv cpu dynamic

* fix pool cpu dynamic

* UserKernelOpInferContext tensordesc return nullptr when arg can not find

* fix annotations

* memcopier add CopyElem

* same padding use copyElem

* refine

* add float16

* add gpu fp16

* deconv use padding_before

* same_padding add sbp

* NCHW->NHWC

* test case

* rm useless and format

* ResetIfDynamic->Reset

* compat_conv2d

* set mirrored view

* update pad conversion in onnx (#3327)

* fix tensorrt conv pooling padding

* address review

* address review

* fix

Co-authored-by: daquexian <daquexian566@gmail.com>
Co-authored-by: Shenghang Tsai <jackalcooper@gmail.com>
Co-authored-by: cheng cheng <472491134@qq.com>