* c++ property deserialization

* fix for comment

* more error info

* fix exception info

* fix ci

* fix compile

* fix layer test ci

* phi_fill_diagonal

* remove old kernels

* update

* update attr args

* refix

* update

* [CustomDevice] add custom ccl 2/2

* update

* update

* update launch

* add sequence_unpad for xpu,*test=kunlun

* add sequence_unpad, *test=kunlun

* fix bug in testcase,should not be sequence_pad,*test=kunlun

* save change

* save change by YSL

* save change by YSL

* change by YSL

* test pre commit

* Revert "test pre commit"

This reverts commit eee5e116331186cc544de871b4a5174a6431f17c.

* fix code style

* fix ctest

* temp save

* save change

* change by YSL

* final change by ysl

* fix ci

* fix code style

* delete unuse code

* change by ysl

* fix dy2staic: cpu, cuda, assign([Var, Var, ])

* fix1

* fix2
Co-authored-by: Nxiongkun <xiongkun03@baidu.com>

* move repeat interleave

* fix api name

* recover op registration

* fix arguments order

* fix

* fix infermeta

* fix infermeta

* fix header

* fix infermeta

* fix

* fix

* fix dtype

* log&test

* test

* remove logs

* fix

* remove logs

* combine files

* combine

* combine files

* fix cuda place

* [Dy2St]Support generate whole program in ProgramHelper for Engine

* support to(mode)

* fix word typo

* fix unittest

* fix_auc in instag task

* fix_auc in instag task

* fix_auc in instag task

* fix opmaker error

* optmize default value of ins_tag

* fix multiary

* fix multiary

* reback ctr_metric_bundle

* reback ctr_metric_bundle

* for ci auc_test

* code style

* code style fix

* suit paddle

* suit paddle

* update code example

* complete Auc unittest

* set asdispensable

* set optional

* compatable and code style

* fix cu

* fix cu

* for ci win

* update unittests of auc

* update code doc by reviews

* change code doc by reviews

* mv kernel

* mv infershape

* mv yaml

* update some

* maintain decoe_jpeg in old dygraph

* fix as review

* rm decode_jpeg_op.cu

* update for rocm

* release the convert_len()

* fix bugs

* fix bugs

* fix start error.

* when shape==0, we should also call paddle.shape

* mv eigvalsh op

* add cmake enforce

* transfer multiclass_nms3  to phi

* phi_matrix_nms

* remove old kernels and add optest check_eager

* reoder args

* reoder args in infermate

* update

* get back legacy dygraph

* add use_hierarchical_allreduce

* support hierarchical allreduce for more cases

* phi_edit_distance

* fix

* move renorm op

* change python api

* change op class func

* alloc data

* remove comments

* fix grad arguments

* fix python argument

* fix python argument

* change unit-test

* remove shape func registration

* recover extra-arguments

* recover shape functor

* [PFCC] SeluKernel Optimization

* selu kernel optimization

* add private

Co-authored-by: carryyu <>

* multihead matmul add fp16

* fix windows error

* fix rocm error

* fix rocm error

* just a test

* remove split assign test

* remove other useless code related to split assign