* [CustomDevice] register Copy for custom device

* [CustomDevice] register Copy for custom device

* [CustomDevice] register Copy for custom device

* merge and add uts

* merge and add uts

* fix for blocking and unittests coverage

* fc support fp16

* add a ‘,’ on paddle_pass_builder.cc

* fc support fp16 on non-cuda.

* fix the outputs of net is x,x

* add unittest for duplicate output

* fix

* fix _infer_program use the original program not the amp program.

* get _***program_id back and avoid duplicate cache
ing

* fix

* add property.proto, can compiled

* property get and deserilize

* support get float

* format code

* format code

* add unittest

* add more set method

* fix grammar error

* Update paddle/fluid/jit/property.h
Co-authored-by: NAurelius84 <zhangliujie@baidu.com>

* Update paddle/fluid/jit/property.cc
Co-authored-by: NAurelius84 <zhangliujie@baidu.com>

* Update paddle/fluid/jit/property.cc
Co-authored-by: NAurelius84 <zhangliujie@baidu.com>

* Update paddle/fluid/jit/property.cc
Co-authored-by: NAurelius84 <zhangliujie@baidu.com>

* fix comment

* fix error throw

* fix property save unit test

* fix error info

* fix copyright and header import

* reorder jit property tensor datatype
Co-authored-by: NAurelius84 <zhangliujie@baidu.com>

* fix_fill_constant

* fix_fill_constant

* fix_ernie

* svd cpu forward

* svd gpu forward

* transfer the backward of svd

* remove cusolver in svd_grad

* svd kernel bug fix

* fix bugs

* fix bugs.

* fix bug

* bilinear_tensor_product yaml

* Support predictor function in JitLayer

* Pybind PEFunction

* Pybind PEFunction and call phi api in layer_test

* Call sqrt phi API

* Polish flags

* Fix comments

* Replace with dygraph op calling method.

* [Dy2Sta]Fix Segment Fault while training multi-card if params have no grad

* fix unittest

* add slot attr for push sparse op

* add pybind

* remove fleet

* add unittest

* fix

* add xpu pnorm op and fix pool op, *test=kunlun

* add adaptive pool, and softmax with cross entropy supports different axis, *test=kunlun

* fix bugs of random

* fix unittest error

* fix unittest bugs

* fix some error found in tipc.

* update

* convfusionfp16

* convfusionfp16

* convfusionfp16

* add temporal_shift yaml and unittest

* fix unittest

* fix log_dir

* _enable_legacy_dygraph

* seperate xshape kernel from normal kernel

* fix bugs in infermeta

* fix compile bugs

* fix compile bugs

* update api approve members, test=document_fix

* add qingqnig into list, test=document_fix

* move diag_embed to phi.

* [Phi] migrate exponential kernel to phi

* fix comment

* fix CI

* Update ps_gpu_wrapper.h

* Update ps_gpu_wrapper.h

* Update ps_gpu_wrapper.cc

* add adam/sharedadam optimzier for gpups;edit optimizer struct;test=develop

* fix CreateVarNode in multi_devices_graph_pass

* Revert "Fix var duplication bug for graph_to_program_pass (#44278)"

This reverts commit a2c4c86b.

* device_guard support xpu. test=kunlun

* sum op of xpu support LoDTensorArray. add test for while op of xpu. test=kunlun.