* add roformer pass&&plugin（novarlen）

* XPU support eager mode

* add unittest for XPU eager mode

* minor bugfix

* minor bugfix, test=kunlun

* correct copyright info

* 1. remove unsed vars/funcs
2. ProcessGroupBKCL inherit from ProcessGroupStream

* bugfix for fp16 in eager mode multi-card, test=kunlun

* rebase & fix a few issues

* use new processgroup interface, test=kunlun

* fix compile issue, test=kunlun

* skip_merge_layernorm

* add UT

* modify comments

* add ci check for code-gen script

* update

* fix paddle with cinn cannot link relu op bug

* change cmake activation_op to generator_op

* add fuse_multi_transformer_layer_pass

* refactor: send, recv, send_partial, recv_partial

* refactor: rm useless const ref

* remove locked allocator

* fix ut

* add heafer file

* Get params and grads in cpp to avoid gpu idel time

* Using python param instead of cpp return param to fix test_asp_optimize_dynamic.py

* Get grads from cpp and construct params_grads on python

* Check meta and remove comments

* Analysis API interface for disabling fc passes

* Unit tests corrected

* Python API added

* test runs only when PADDLE_WITH_MKLDNN

* Fc op changed to relu in matmul_op_test

* Disable fc passes in tests where acc drops

* code formating

* Unit test for analysisConf added

* Unit test gpu added

* fc passes disabled when iterations=0 in gru test

* style

* passes disabled when fp32 in gru test

* fc passes disabled in lstm test

* Import from inference, not fluid in doc

* fix flake8 CodeStyle E266

* fix comments

* first commit

- more fixes

- compilation fix

- compilation fix

- fix

- another fix

- yet another fix

- Fix

- fix to fused ops

- compilation fix

- compilation fix

- another compilation fix

- another fix

- fix

- fix

- fix

- fix

- yet another fix

- fix

- fix

- cosmetic fix

:- lint

- Revert some changes (to be brought back later)

- fix to build

- Added prototype of slice

- fix

compilation fix

- compilation fix

- fix

- fix

- Fix

- fix

 fix
	modified:   cmake/flags.cmake

* lint

* rerun of CI

* - Fix

* - lint

* - lint2

* generate static graph code of some op

* polish code

* fix bug

* update default value

* move fluid op generator into fluid

* remove parsed op

* resolve sig undef error

* append python interp find logic

* remove dup code

* upgrade scale and slice op convert for Paddle-TensorRT

* refine python call error report

* softplus+activation

* fc + elementwise_add test refactored

* rename MKLDNN to OneDNN

* fc+activation tests refactored

* remove softplus ut

* whitespace

* whitespace

* codestyle

* codestyle

* add more cases to fc+act

* remove softplus+hard_sigmoid pass

* remove softplus + hard_sigmoid UT

* add approximate for gelu

* swish beta range

* new codestyle

* reduce number of tests

* allow fold fill_constant && allow nms3 into trt in int8 model
* use unordered_map
* fix CI failing

* Split kernel registered, tests for uint/int added

* Split quantized

* Split output scales calculated only once

* NearestInterp test fix reversed

* DequantizeOutputs corrected

* removing dependent to fluid/framework/eigen.h in phi

* more fix according to PR-CI-Py3 fail

* support pow_double_grad op

* add unit test for pow double grad

* fix pow double grad

* optimize pow double grad kernel

* fix pow double grad kernel

* fix cast bug

* Fix undefined symbol: shm_open

* Fix for Windows

* Exclude APLLE

* suqeeze2 transpose2 fuse onednn

* format

* fix output shape

* fix conflict

* format

* format

* remove useless

* remove log

* simply pass

* fix comment

* fix

* fix msg

* fix error msg

* format