* fix bug

* refine code

* refine code

* refine code

* refine code

* refine code

* refine code

* fix bug

* add reference of lbfgs

* add reference of lbfgs

* modify concat kernel choose

* modify ci

* rename fluid to base

* fix setup and rename paddle.fluid

* fix protobuf generated paddle.fluid

* fix some errors

* fix cmake depends

* fix some error

* try to fix cache error

* fix codestyle && some doc error

* remove build change

* fix some error

---------
Co-authored-by: NXiangGao <jeff41404@gmail.com>

* complex op logsigmoid

* fix 2023-08-31

* [IR] add control flow dialect

* [NewIR]Update send recv infermeta and add unittest

* rm new ir flag

* rm fluid api

* skip runing startup prog

* update flag name

* update recv_v2 yaml

* fix conflict

* unittest only for pp

* fix cmakelist

* unittest check precision

* control random

* fix cmakelist

* add spmd and reshard code gen

* add backward reshard code gen

* test matmul forward success

* polish test impl

* add unsafe mutable value

* polish details and add test

* fix unittest time out

* fix typo

* refactor reshard input generate impl

* resolve conflict with develop

* fix compile error

* add <with_grad_semantic> in OpInputInfo to represent whether an input of OP has grad semantics

* add support for check OP's input grad semantic by comparing fwd_op inputs and bwd_op outputs

* add pybind interface to support checking OP's inputs grad semantic in python-level

* add test

* fix bugs

* fix bugs in op_gen

* fix bugs in op_gen

* add test for multiply_op

* fix bugs in codestype

* fix bugs in codestype

* [XPU] add squeeze_excitation_block_xpu op&pass to optimize ppocr_v3_det model test=kunlun

* fix

* fix Codestype

* remove xpu name

* fix setvalue dtype error when using dy2st and amp O2

* add one test

* remove test share_buffer since win/linux have different number

* fix bug

* fix bug

* div passed v1.0

* IrChange->IrGuard & optimize static input dtype

* clean

* remove IrChange and optimize IrGuard

* support new ir primitive operator in static operants

* support more vjp code gen

* support more vjp code gen

* support more vjp code gen

* use code gen

* fix operants codegen

* support more vjp code gen

* Fix ci build error

* set FLAGS_tensor_operants_mode to static in generated_vjp for testing

* fix bugs

* change the order of ops_name of divide_grad

* replace FLAGS_enable_new_ir_in_executor by FLAGS_enable_new_ir_api in codegen and test_vjp_prim

---------
Co-authored-by: NCharles-hit <wanghao107@baidu.com>
Co-authored-by: N0x45f <wangzhen45@baidu.com>

* add symbolicDimProduct & symbolicDimMgr without method shape_constraint related.

* split ddim in phi, add a target ddim, used by pd_type.

* add pd_type.cc to ir_shape CMakeLists.

* add dimOp, tieProductEqualOp. access constraint_func in SymbolTable.

* put DenseTensorType into builtin_type.

* add constraint related Mgr func.

* move to out assert.

* add_n and combine support selected rows

* fix bug

* fix bug

* [NewIR]Support Build(GroupOps) in NewIRCompiler and Add UT

* fix unittest

* sep group

* add test

* test ok

* polish

* test cmake script generated

* add sep group

* format

* polish

* polish

* fix static_build for pp

* add mannual_event to support streams across progs

* revert static_build.sh

* fix coverage-ci

* modify the method to name events

* change code according to review

* fix addn infermeta

* fix rule bug

* [WIP] Support std::vector<phi::Tensor> input and output for DistTensor.
Concat forward and backward are verified.

* Polish code for new dist tensor implementation.

* Fix bug of DistTensor upgrade. Add support functions for std::vector<Tensor> -> std::vector<Tensor>.

* Add support for DistTensor type of std::vector<phi::Tensor> as input or output of operators.
Following testcases are passed.
1. concat: std::vector<phi::Tensor> -> phi::Tensor
2. unbind: phi::Tensor -> std::vector<phi::Tensor>
3. broadcast_tensors: std::vector<phi::Tensor> -> std::vector<phi::Tensor>

* Polish code. Remove useless comments.

* Add update_loss_scaling in skip_op_lists.

* Polish code.

* replace amp.use_pure_fp16 with amp.dtype and amp.level

* old api still use use_pure_fp16

* test_fuse_adamw_pass still use use_pure_fp16

* add test case tuning recompute with amp(float16,o2)

* reset new test case properties TIMEOUT 60

* set smaller value of batch_size and batch_num

* deepcopy dist_context fix _rename_input problem

* fix loss name after cast

* set tuning.enable=True and use engine._tune()

* restore some changes in _rename_input()/_rename_output()

* add self.amp_dtype for _cast_loss() in auto_parallel_amp.py

* fix insert op index in _cast_loss()

* add rotate_half in fused_rope

* add position_ids in fused_rope

* modified examples about fused_rope

* add set_device in examples

* remove has_structed_kerenl and has_fluid_kernel

* add test_fused_layernorm_op to STATIC_BUILD_TESTS list

* open static_build flag

* remove distributed_fused_lamb_init from StaticBuildBlackList

* use initialized replacing IsInitialized

* recover codes

* delete useless codes

* close the flag

* add num_splist for flash_attn_bwd and FlashAttnUnpaddedGradKernel

* Add assertTrue

* Update submodule to a specific commit

* disable strided split

* add code

* add code

* refine code

* add code

* fix bug

* fix bug

* fix bug

* add code

* add ut

* polish code

* fix bug

* refine code

* fix bug

* refine code

* fix bug

* refine code

* fix bug

* refine code

* fix bug

* refine code

* add code

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* refine code

* Modify MarkTrtEngineOutputs API

Change FlattenLoops in the elementwise schedule to Fuse