* add DistributedFusedLamb op

* polish code

* fix compile error

* compatible with pten changement

* fix rocm compile error

* improve converage

* update upstream/develop

* fix cast_with_ptr.h

* add FLAGS_distributed_lamb_divide_nranks_when_allreduce=1

* fix clip before allreduce

* add use_master_param_norm

* code polish

* fix bug

* fix ROCM ci

* [CustomRuntime] Add DeviceManager

* [CustomRuntime] Add DeviceInterface

* [CustomRuntime] Add Stream, Event, DeviceGuard, CallbackManager

* [CustomRuntime] Add plug-in device

* [CustomRuntime] Memory module support PluggableDevice

* [CustomRuntime] Add WITH_PLUGGABLE_DEVICE cmake option

* update

* [API] update API doc based on comments, test=develop
Co-authored-by: Nqili93 <qili93@qq.com>

* Rough implementation for experiment

* Support allocate cuda managed memory

* Fix CI error

* Modify UT

* Check whether support memory oversubscription

* Fix ROCM Compile error

* Fix ROCM Compile error

* Fix UT cuda_managed_memory_test

* Set UT timeout to 40

* Add UT OOMExceptionTest

* Set UT timeout to 50

* Support allocate_from in Tensor and allocate_data in Context

* fix #ifdef CUDA

* fix cycle depends

* fix test_xxx_dev_api failed

* fix windows compiling error

* fix unittest

* modify into PImpl

* fix selected rows

* add TODO comment

* refine interface according reviewer

* sync misc changes

* apply comments 01

* fix compile error

* remove is_ipu_place check

* add authors
Co-authored-by: NXiaobing Wang <xiaobingw@graphcore.ai>
Co-authored-by: NAllen Guo <alleng@graphcore.ai>
Co-authored-by: NZhixin Yao <zhixiny@graphcore.ai>
Co-authored-by: NHaicheng Jiang <haichengj@graphcore.ai>
Co-authored-by: NHan Zhao <hanzhao@graphcore.ai>

* sync changes

* restore cmake

* update ir cmake and setup.py

* update inference_lib cmake

* split PR
Co-authored-by: NXiaobing Wang <xiaobingw@graphcore.ai>
Co-authored-by: NZhixin Yao <zhixiny@graphcore.ai>
Co-authored-by: NHaicheng Jiang <haichengj@graphcore.ai>
Co-authored-by: NHan Zhao <hanzhao@graphcore.ai>

* [MLU]add mlu ci dockerfile

* fix comment

* add cncl

* add cpu_context.

* update

* update

* update

* update

* update

* fix ci problem

* fix npu ci problem

* update

* fix ci compile

* add pten::Place data structure.

* update ci problem

* fix ci problem

* update

* using platform::Place=pten::Place

* remove BOOST_GET_CONST for CPUPlace and GPUPlace

* compile pass 25%.

* compile pass 45%

* compile pass 60%

* remove boost_get for xpu npu mlu and ipu

* compile pass on cpu and gpu.

* fix compile problem

* fix compile error.

* update

* fix ci problem

* update

* ci approve

* fix ci problem

* fix ci eager test problem

* remove BOOST_GET_CONST

* fix npu compile

* [MLU]: add mean and reduce mean op

* [MLU]add mlu pytest dir in CMakeLists.txt

* [MLU]fix tensor data

* [MLU]fix TensorToPyArray and license

* Fix bug of GetAllocatorInterfaceTest

* Replace some shared_ptr with unique_ptr

* Change Alloc call

* [MLU]support calling mlu op from python interface

* [MLU]fix

* fix

* [mlu]fix mlu_places

* [mlu]fix required mlu

* fix

* [MLU]fix tensor copy

* [mlu] fix MLUPlace call path

Fix Buddy Allocator random CI failure due to machine environment.

* fix reshape move storage error

* remove needless set type

* alloc tensor by shared storage

* Utilize StreamSafeCUDAAllocator to support fast GC in new executor

* Fix compile error for Windows and ROCm

* Fix compile error for Windows

* Modify UT stream_safe_cuda_alloc_test

* Modify UT stream_safe_cuda_alloc_test

* Rewrite fast GC

* Rewrite fast GC

* Fix compile error for BOOST_GET_CONST

* Fix compile error for BOOST_GET_CONST

* Changes default stream for StreamSafeCUDAAllocator

* Fix a small CI error

* Remove some redundant code

* Fix conflict

* Fix compile error for ROCm

* Fix Windoes CI error

* Fix CI error

* Remove some unnecessary code

* Fix CI error

* Add UT for fast GC

* Fix CI error

* add device-agnostic stream class

* add stream.h

* fix ut

* fix cpu compile

* Use RWLock in GetAllocator

* Fix CI error
Co-authored-by: NChen Weihang <chenweihang@baidu.com>
Co-authored-by: Nzhiqiu <chenqiuliang@baidu.com>

* add device-agnostic stream class

* add stream.h

* fix ut

* fix cpu compile

* Get GPU BasePtr from CUDA allocation

* Fix compile error for ROCm

* Add BasePtr function for IPUPlace in naive_best_fit_allocator.cc

* Add alignment for BuddyAllocator

* Set address alignment of BuddyAllocator to 32 bytes

* Fix CI error

* Remove code for naive_best_fit strategy

* Fix CUDAGraph bug for StreamSafeCUDAAllocator

* Add CUDAGrapthAllocator check in multi-stream interface

* Set FLAGS_use_stream_safe_cuda_allocator defaulted to false

* Fix environment error for cmake

* Fix cmake error

* Add UT of GetAllocatorInterfaceTest

* Add UT of CUDAGraphExceptionTest

* Enhance CUDAGraphExceptionTest

* refine structure for cuda and rocm

* update

* update

* update

* update

* [NPU] reorganization for device API abstraction

* [NPU] delete old files

* [NPU] fix npu_collective_helper

* [NPU] fix collective_helper

* [NPU] fix ut

* [NPU] mod memory allocation and hccl_helper

* [NPU] fix place_type

* [NPU] split enfoce.h

* move acl* call into npu_info

* merge conflict

* fix merge

* merge conflict

* merge conflict

* Support multi-stream allocation for CUDA place

* Do not notify the retrying from other streams when free CUDA allocation

* Fix compile error for CPU

* Fix compile error for HIP

* Release memory for StreamSafeCUDAAllocaRetry in malloc_test

* Add FLAGS_use_stream_safe_cuda_allocator

* Fix CI error for 'set_tests_properties'

* Invalidate stream safe CUDA allocator for naive_best_fit and thread_local strategy

* Performance improvement: insert allocation pair to outstanding_events_map when free but not alloc; replace recursive_mutex with SpinLock

* FLAGS priority changes: FLAGS_use_system_allocator > FLAGS_use_stream_safe_cuda_allocator

* Performance improvement: directly delete allocation when the recorded_streams is empty in FreeImpl of StreamSafeCUDAAllocator

* Add UT for alloc interface

* Changes multi-stream interface; move retry code from AllocatorFacadePrivate to StreamSafeCUDAAllocator

* [XPU] Reorganize xpu device codes in platform, test=develop

* fix xpu_header.h, test=develop