* add default_ctx in backward.py

* record grad_var_to_var with grad_times

* fix backward

* update annotation

* add complete_high_order_grad in complete_forward

* add dist slice op

* update grad_var_to_var type

* update partition_block init mapping before loss op

* update compatible for 'XShape' & update 'allreduce_vars'

* add dist reshape op when input dim equal to output dim

* update 'set_grad_var_shape' with grad_var_to_var

* fix dist slice

* fix set_grad_var_shape

* add dist pnorm op

* fix dist pnorm dist_attr

* fix engine startprogram & adapt highorder grad

* fix set_grad_var_shape when mp

* update unittest

* update cmakelist

* default strategy in engine: dp

* bug fix

* tiny fix

* flatten outputs

* fix default strategy

* init default ctx

* tiny fix

* test=allcase

* merge recent changes

* fix setting pipline

This PR hotfixed the `test_cond.py` in CUDA 11.2

The reason of the bug is that the `fill_constant` op returns wrong value in the modified test case `test_extremely_simple_net_with_op_in_condition`, SWEs can use `layers.Print(a)` and `layers.Print(b)` in the test case to reproduce it and they can see the `fill_constant` returns something `e-50` instead of `1.23` and `1.25`

This PR hotfixed the bug by comparing `b` value instead of actual number, which makes sure the `cond` logic is right. **However, the PR didn't fix `fill_constant`**. We would let the SWEs who are working here to find the op bug and fix it.

* Extend python einsum interface to make einsum_v2 support multi-operands and switch it to default.

* add opt_einsum dependence

* add yaml and support eager model

* fix by code review

* [Eager] Remove enable_legacy_dygraph setting

* Add more tests

* add double yaml

* add inline func

* [Dy2Stat]Fix losting pre/post hook from outermost layer while jit.save

* fix kwargs

* fix unittest

* add gradient merge for DistributedFusedLamb

* use master acc gradient

* fix CI ut

* polish

* remove math_function_impl.h change

* fix test_update_loss_scaling_op.py

* try to fix XPU/NPU CI

* add gm ut

* test op_test test=allcases

* fix

* avoid copy many same file

* fix for win

* test PYTHONPATH

* change path adding way

* fix win

* use old way

* use old way test=allcase

* use old way test=allcase

* fix collections.Sequence in python3.10

* fix format

* added test for shuffle_channel_mkldnn_detect_pass

* added UT using new framework

* CI fix

* [DO NOT MERGE] test op_test

* update with more related modifications

* split op_test.py to use test=allcases for testing

* split op_test.py to use test=allcases for testing

* 增加PixelUnshuffle的形状推断

* 增加PixelUnshuffle的算子注册

* 增加PixelUnshuffle及其梯度的核函数

* 增加PixelUnshuffle算子的描述

* 增加PixelUnshuffle算子的签名

* 在Python层面增加PixelUnshuffle

* 增加PixelUnshuffle的单测

* Update test_pixel_unshuffle.py

* test=document_fix

* Update test_pixel_unshuffle.py

增加对extra_repr的测试

* 修正代码格式

* Update test_pixel_unshuffle.py

修正对extra_repr的测试

* 修改pixel_unshuffle核函数的实现位置

* 修正代码格式

* 完善对输入的检查

* Update test_pixel_unshuffle.py

* 完善pixel_unshuffle的输入检查

* Update pixel_unshuffle_op.cc

* Update unary.cc

* add pixel_unshuffle

* Update test_pixel_unshuffle.py

* Update vision.py

* 调整代码格式

* Update vision.py

* Delete extra spaces

* Update pixel_unshuffle_sig.cc

* Update vision.py

* Update vision.py

* add PixelUnshuffleGradInferMeta

* remove PixelUnshuffleOpArgumentMapping

* Update pixel_unshuffle_op.cc

* 调整pixel_unshuffle及其梯度的核函数的实现位置

* Update pixel_unshuffle_op.cc

* [Eager] Support final_state_adam when argument grad (position 1) is selected_rows

* Remove needless code

* Add adam_dense_param_sparse_grad kernel

* full api fix

* when out is None, go old dygraph mode

* by static check

* first version: support 2-inputs forwards. TODO: 1. backward  2. BroadCast  3. MultiVariable

* time out -> 120

* Add infermeta for ChannelShuffle

* Create channel_shuffle_grad_kernel.h

* Create channel_shuffle_kernel.h

* Create channel_shuffle_sig.cc

* Create channel_shuffle_op.cc

ChannelShuffle算子的描述

* Create channel_shuffle_kernel_impl.h

ChannelShuffle核函数的实现

* Create channel_shuffle_grad_kernel_impl.h

ChannelShuffle反向核函数的实现

* Add kernel register of channel shuffle and grad

注册ChannelShuffle及其反向的核函数

* add nn.functional.channel_shuffle

* add nn.ChannelShuffle

* Create test_channel_shuffle.py

* Update example of ChannelShuffle in vision.py

* Update test_channel_shuffle.py

* 修改channel_shuffle核函数的实现位置

* 修正代码格式

* 删除多余空格

* 完善channel_shuffle的错误检查

* Update unary.cc

* Update channel_shuffle_op.cc

* Update test_channel_shuffle.py

* Update unary.cc

* add channel_shuffle

* Update test_channel_shuffle.py

* Update vision.py

* 调整代码格式

* Update channel_shuffle_sig.cc

* 更新ChannelShuffle的文档

* 更新channel_shuffle的文档

* remove ChannelShuffleOpArgumentMapping

* add ChannelShuffleGradInferMeta

* Update channel_shuffle_op.cc

* 调整channel_shuffle及其梯度的核函数的位置

* [Eager] Support div scalar in eager mode

* Updated and remove debug logs

* Remove list, use 'or' directly

* Remove useless statement

* test=py3-eager

* test=py3-eager

* test=py3-eager