* support set master_grad

* move register_hook to auto_cast

* update unittest

* fix fp16 test

* update for review comments

* [Opt Performance] Optimize custom operator performance, reconstruct python API auto-gen, add cache and use const inference

* opt AutoGradMeta implementation

* remove profiler codes

* fix unit test

* change year, 2021->2023

* fix int64_t parse bug

* add strategy force_sequential_run

* remove flag

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* mv WITH_ASCEND_CL

* mv WITH_ASCEND

* rollback

* remove WITH_ASCEND

* remove WITH_ASCEND

* add strategy force_sequential_run

* fix

* fix

* fix

* fix

* fix

* mv WITH_ASCEND_CL

* mv WITH_ASCEND

* rollback

* rem is_compiled_with_npu

* rem nup related code

* make lint happy

* rem test

* remove some tests

* Update grad_scaler.py

* fix an error

* add gloo gather

* add gloo_tools

* fix CI bug

* use gloo gather

* remove redundant code

* fix process_group_gloo.py

* rename send_recv

* fix conflict

* fix conflict

* fix codestyle

* fix CI bug

* add PADDLE_ENFORCE_NE

* rem is_compiled_with_mlu

* fix some mlu_place and mlu_device_coount

* make lint happy

* Delete the /paddle/fluid/platform/device/npu directory

* clear Cmakelists

* Try removing npu in the header file

* gather with doc

* resolve comment

* polish

* polish

* code style

* polish doc

* add_test

* polish

* polish

* add test check

* add test check

* polish

* polish

* polish

* polish

* fix_time_out

* polish

* fix timeout

* fix_timeout

* polish

* polish

* polish

* polish

* polish

* [CustomOP Inplace] Automap inplace dtype and shape, prepare for vector<Tensor> output

* delete custom_inplace_setup.py

* [CustomOP Optional Inplace] Custom operator supports inplace optional Tensor input

* fix bug for vector<Tensor> inplace test

1. add type caster for paddle's complex type, to allow pybind to automatically cast it with python's complex type;
2. add complex64 and complex128 data type for `libpaddle.Tensor`'s element get and set(which is required to perturb an element to get the numerical derivative)
3. add support for cuda pinned place in `libpaddle.Tensor` element get and set

---
4. fix a bug in op code generation.(Creation of output folder in concurrent with parsing op yamls.)

* [AMP] Add python API for collecting operator stats.

* Fix import and polish codes.

* Add more unittest.

* Add doc for the new APIs.

* [CustomOP Inplace] Automap inplace dtype and shape, prepare for vector<Tensor> output

* delete dtype,shape func of multi_inplace op

* [CustomOP Inplace] Automap inplace dtype and shape, support vector<Tensor> output

* [CustomOP Inplace] Auto-generate python API for inplace vector<Tensor> output

* add fuse adamw pass

* fix some bugs

* fix CIbug

* change chunk_size

* fix CI bug

* rm test_fused_adam_op.py

* fix CI bugs

* fix fuse_adamw_op_pass.cc

* change code style

* fix CI bug

* fix ut bug and use_adamw_op_pass.cc

* fix test_fuse_adamw_pass.py

* fix CI bug

* remove fluid

* fix ci bug

* fix CI bug

Add basic functionalities to support Scalar & Scalars in operator attribute.

1. extend allowed types in operator's attribute type, add `paddle::experimental::Scalar`, add corresponding protobuf Message types;
2. Scalar enhancement, add formatting, equality;
3. add code to handle Scalar & Scalars in opmaker, conversion from  paddle operator to phi kernel, opdesc construction and manipulation,  tensorrt converter, tracer, operator construction, etc;
4. bind `paddle::experimental::Scalar` to python, as `libpaddle.Scalar`;
5. add functionality to canonicalize attribute map according to OpProto(if the op the attribute map used for has an OpProto);
6. add code to manipulate Scalar proto message via protobuffer python API;

Add unittests.

1. add test cases for formatting, equality for Scalars, and WrapAsScalars;
2. add test cases for 'casting' between different morphs of attributes;
3. add test cases for extracting scalar & scalars from attribute;
4. add test cases for CanonicalizeScalarAttrs(and fix a bug in type index offset);
5. fix gmock's library filename on windows platform.
6. clean code: use canonicalize_attrs instead of inlining the function;
7. add test cases for libpaddle.Scalar in python code.
8. add test cases for `make_scalar_proto`, which manipulate proto message `Scalar` via protobuffer python API.

* delete old dygraph op test

* [CustomOP Inplace] Automap inplace dtype and shape, prepare for vector<Tensor> output

* delete dtype,shape func of multi_inplace op

* [CustomOP Inplace] Automap inplace dtype and shape, support vector<Tensor> output

fix bug

remove useless

fix bug

add pybind

remove log

fix style

fix style

change api

* Add fused_feed_forward pass for semi-automatic static graph training.

* Add fused_feedforward property in parallel_executor.cc

* Polish code.

* Polish fused feed_forward pass code. Support use_dropout1 and
use_dropout2 option.

* Support model parallel in fused_feedforward pass.

* [CustomOP Optional] CustomOP supports optional Tensor

* fix test_custom_concat, add pytest to CMakeLists