* moe ref

* ref commit; test=document_fix

* update; test=document_fix

* update test=document_fix

As the title

* refactor mp in eager mode

* update

* update

* add uts

* change cudnn helper for auto-tune

* Add FLAGS_use_autotune to set the global status of autotune and change the order of choosing algorithm.

* Fix the bug in calculating and printing current step cache hit rate.

* Improve the autotune cache and fix unittest.

* Change the key from AlgorithmType to int64_t.

* Fix unittest for cpu-only env.

* change ChooseAlgoByWorkspace for heuristic mode
Co-authored-by: NLiu Yiqun <liuyiqun01@baidu.com>

* regist elementwise_xxx

* Change tensor name to match activation

* declare fc_eltwise_add pass

* merge conv_eltwise refactor PR

* first compilable draft

* unittest feedback tools

* Fuse pass tester

* Move IsReachable() to shared file

* 100% coverage of fuse_pass_tester.cc

* register pass

* Add bias node

* Improve unit tests / remove bias node from pattern

* improve fc_eltwiseadd_unittest

* cancel eltwise_add fuse if act is already fused

* Add elementwise_input scale

* Residual MVP

* Add new FC attrs

* Add more test cases

* Add missing op attrs

* Adapt code to new Elementwise pattern

* reuse existing fcpattern

* improve code style

* remove unused arguments

* fix typo

* remove whitespace

* remove int8 related code

* Remove attributes from base ops

* style

* style check

* Remove input from base op

* Set attribute during fuse

* ut timeout

* download and test model

* DRY

* apply feedback from review

* Style check

* fix typo

* cosmetic changes

* explicitly set residual as output

* VIT-OCR accuracy check

* trigger CI

* remove whitespaces

* fix missing data file

* support multi layer and bidirection of lstm_grad, *test=kunlun

* support multi layer and bidirection of lstm_grad, *test=kunlun

* added shuffle_channel bf16/fp32 fwd kernel

* added missing files

* CI fix

* changed from pten to phi

* tmp save

* added reviewers suggestions

* fix for test

* native commit for triple grad of sigmod

* Updated unittests files

* init functional jacobian api

* Updated trible_test func

* Updated gradient_checker & test_script

* finish test with dtype float32

* add float64 test case

* polish code

* use atol=1e-5 with dtype float64

* fix for ci

* set timeout for test_jacobian

* fix dygraph grad to support high differential

* polish API docstring

* Updated gradient checker and some related files

* fix double grad strip error for high differential

* fix double grad strip error for high differential

* Add Sigmoid triple grad tests

* fix dygraph double grad dtype error when calling for high differential senario

* Updated triple grad teses func

* Use np.random to initialize ddx

* Updated triple_grad_check func

* add todo for gradient checker and refine some comments

* remove additional code

* add test for warnging in backward.py

* format python code

* support multi input in triple gradient checker

* Add matmul triple grad kernel

* Updated comments of TODO

* Supported some special tests

* Change code-format to follow CI std

* Updated gradient_checker.py

* Fix conflicts

* Removed unnecessary printing log

* Change code style to follow CI std

* merge upstream

* add_p

* rm useless files

* add sub_p mul_p div_p

* add sqrt_p and tanh_p

* add reshape_p

* add broadcast_p

* add broadcast_p fill_constant_p matmul_p reduce_p reshape_p transpose_p

* add split_p and concat_p

* add gather_p and scatter_add_p

* add slice_select_p and slice_assign_p

* add multi input check for add_p, sub_p, mul_p, div_p

* update concat_p

* refine gather_p and scatter_add_p

* refine slice_assign_p and slice_select_p

* add 9 test for prim ops

* add more test and fix some bug

* add more test

* register proto

* add shape valid check for broadcast_p op, and add keepdim attr into reduce_p op proto

* support multi input and multi output for split_p and concat_p

* fix slice bug for slice_select_p and slice_assign_p

* dtype for axis attr should be long int

* update dtype for axis attr int64_t

* update for iscan CI

* add more shape and dtype check

* change IndexTensor into int32 dtype

* add sgd yaml

* change python api

* open eager mode in sgd

* fix bug

* add expand, poisson

* add poison grad

* add expand equal_all poisson triangular solve yaml

* remove stack_grad infershape

* fix bug of output with null

* fix bug

* fix split error when sections has 0 size and add test case

* fix test case

* update sign op on xpu. test=kunlun

* fix typo. test=kunlun

* gather op

* add mod

* [Yaml] final state for uniform and uniform_random

* 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>

fix group_norm vectorized address misalignment

* 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

* ad conj flip yaml

* add flip conj pixel shuffle

* add yaml

* perfect converage

* add Output(Step) to distributed fused lamb op

* add _set_step

* ignore some failed test for KL2
*test=kunlun

* minor
*test=kunlun

* minor
*test=kunlun

* momentum support l2decay for xpu. test=kunlun

* fix include file. test=kunlun

* fix cmake for device_worker. test=kunlun