rhFnUdZddlZddlZddlmZddlmZmZmZm Z ddl Z ddl Z ddl m Z ddlmZmZddlmZdd lmZmZd aej.Zej2Zgaeeegdfeefed <e e jBd d Z"e e ee#dfZ$ea%dZ&ee jBjNed<de(fdZ)e)rCe jBjTZ*e jBjVZ,e jBjZZ.nedZ*de#de#fdZ,de#de#fdZ.edde#fdZ/de(fdZ0d>de(de(fdZ1dZ2dZ3dZ4dZ5Gdd Z6Gd!dZ Gd"d#e Z7de$ddfd$Z8d?dee$defd%Z9edd?dee$de:eeffd&Z;d?dee$de*fd'ZGd*d+Z?d,ed-de?fd.Z@d/ZAd,efd0ZBd?dee$defd1ZC d?d2e#dee$defd3ZDd?de$ddfd4ZEdeefd5ZFdefd6ZGde jde jBjNfd7ZH d@d8e#de e#ee jfddfd9ZId@de e#ee jfde#fd:ZJdd;lKmLZLmMZMmNZNmOZOmPZPmQZQmRZRmSZSmTZTmUZUddrsrrdefault_generatorsreturnc6tjjS)z(Return true if compile with XPU support.)torch_C_has_xpurrr _is_compiledr#s 88  r_XpuDevicePropertiesr ctdNz(PyTorch was compiled without XPU supportNotImplementedErrorrs r_exchange_devicer$0!"LMMrctdr!r"rs r_maybe_exchange_devicer'3r%r)maxsizecTtsytjjS)z*Return the number of XPU device available.r)rrr_xpu_getDeviceCountrrr device_countr+7s > 88 ' ' ))rctdkDS)z7Return a bool indicating if XPU is currently available.r)r+rrr is_availabler-?s >A rincluding_emulationcptsy|xs(tjjjS)zKReturn a bool indicating if the current XPU device supports dtype bfloat16.F)r-rxpuget_device_propertieshas_bfloat16_conversions)r.s ris_bf16_supportedr3Es. > F 99 * * , E Erc(txr t S)z8Return whether PyTorch's XPU state has been initialized.) _initialized_is_in_bad_forkrrris_initializedr7Os  1 111rc ftr|y|jddr)tj|t j y|jddr)tj |t j ytj|t j fy)Nseed_allFseed) r7get_lazy_seed_trackerqueue_seed_all traceback format_stack queue_seedrappend)callablekwargss r _lazy_callrDTs{  ::j% (  - -h 8N8N8P Q ZZ &  ) )(I4J4J4L M  (I,B,B,D!E Frcty)zInitialize PyTorch's XPU state. This is a Python API about lazy initialization that avoids initializing XPU until the first time it is accessed. Does nothing if the XPU state is already initialized. N) _lazy_initrrrinitrGbs Lrc tsttdryt5tr dddyt r t dt s tdtjjdt_ tjdtjD tD] \}} | t'tddadddy#t $r1}dt#|ddj%|}t!||d}~wwxYw#t'tdwxYw#1swYyxYw) Nis_initializingzuCannot re-initialize XPU in forked subprocess. To use XPU with multiprocessing, you must use the 'spawn' start methodz#Torch not compiled with XPU enabledTc3&K|] }|s| ywNr).0callss r z_lazy_init..sXuRWUXsz5XPU call failed lazily at initialization with error: z' XPU call was originally invoked at: )r7hasattr_tls_initialization_lockr6 RuntimeErrorrAssertionErrorrr _xpu_initrIrextendr< get_calls Exceptionstrjoindelattrr5) queued_callorig_tracebackemsgs rrFrFksA74):; !    !!  I ~ !FG G  $X0B0L0L0NXX -/< 0+ ^0M 0 D+ , C!!2!0OPSTUPVxXBBD''.BYAZ\$C.a/ 0 D+ ,A!!sM D=BD=9 D(C+ D(D=+ D%4,D  D%%D((D::D==Ec2eZdZdefdZdZdededefdZy) _DeviceGuardindexc ||_d|_yN)idxprev_idx)selfrbs r__init__z_DeviceGuard.__init__s rc`tjj|j|_yrKrr0r$rfrgrhs r __enter__z_DeviceGuard.__enter__ 22488< rtypevaluer>c`tjj|j|_yrrr0r'rgrfrhrorpr>s r__exit__z_DeviceGuard.__exit__9933DMMBrN)__name__ __module__ __qualname__intrirmrrtrrrraras-c=Srrac6eZdZdZdefdZdZdededefdZy) r zContext-manager that changes the selected device. Args: device (torch.device or int or str): device index to select. It's a no-op if this argument is a negative integer or ``None``. c6t|d|_d|_y)NToptionalre)r rfrg)rhr s rrizdevice.__init__s$Vd; rc`tjj|j|_yrKrkrls rrmzdevice.__enter__rnrrorpr>c`tjj|j|_yrrrrss rrtzdevice.__exit__rurN)rvrwrx__doc__rrirmrtrrrr r s2s=Src"eZdZdZfdZxZS) device_ofaContext-manager that changes the current device to that of given object. You can use both tensors and storages as arguments. If a given object is not allocated on a XPU, this is a no-op. Args: obj (Tensor or Storage): object allocated on the selected device. c`|jr|jnd}t| |yrd)is_xpu get_devicesuperri)rhobjrf __class__s rrizdevice_of.__init__s$"%**cnn" r)rvrwrxrri __classcell__)rs@rrrsrrcxtt|}|dk\r tjj |yy)zSet the current device. Args: device (torch.device or int or str): selected device. This function is a no-op if this argument is negative. rN)rFr rr_xpu_setDevicers r set_devicers1L v &F { 'rc,t|jS)aGet the name of a device. Args: device (torch.device or int or str, optional): device for which to return the name. This function is a no-op if this argument is a negative integer. It uses the current device, given by :func:`~torch.xpu.current_device`, if :attr:`device` is ``None`` (default). Returns: str: the name of the device )r1namers rget_device_namers ! ( - --rct|}t|Dcic] }|jds |t||"c}Scc}w)aGet the xpu capability of a device. Args: device (torch.device or int or str, optional): device for which to return the device capability. This function is a no-op if this argument is a negative integer. It uses the current device, given by :func:`~torch.xpu.current_device`, if :attr:`device` is ``None`` (default). Returns: Dict[str, Any]: the xpu capability dictionary of the device )__ _pybind11_)r1dir startswithgetattr)r propsprops rget_device_capabilityrsM "& )EJ  34 geT""  s%AcFtt|d}t|S)zGet the properties of a device. Args: device (torch.device or int or str): device for which to return the properties of the device. Returns: _XpuDeviceProperties: the properties of the device Tr|)rFr _get_device_propertiesrs rr1r1s L v 5F !& ))rcRttjjS)z0Return the index of a currently selected device.)rFrr_xpu_getDevicerrrcurrent_devicer sL 88 " " $$rct|trtj|}|St|trtjd|}|S)zReturn the torch.device type object from the passed in device. Args: device (torch.device or int or str): selected device. r0) isinstancerYrr ryrs r _get_devicersD &#f% M FC eV, MrcNeZdZUdZeded<dedfdZdZdeded efd Z y ) StreamContextaContext-manager that selects a given stream. All XPU kernels queued within its context will be enqueued on a selected stream. Args: Stream (Stream): selected stream. This manager is a no-op if it's ``None``. .. note:: Streams are per-device. torch.xpu.Stream cur_streamstreamc\||_tdd|_|jd|_yy)NTre)rr rf)rhrs rrizStreamContext.__init__)s- $T40 88 DH rc|j}||jdk(rytjj d|_|j j |j k7rLt |j 5tjj |j |_dddtjj|y#1swY)xYwrd) rrfrr0current_streamsrc_prev_streamr dst_prev_stream set_stream)rhrs rrmzStreamContext.__enter__/s[[  R $yy77=    & &**;*; ; ))* S',yy'?'? @Q@Q'R$ S Z( S Ss ;/CCrorpr>c*|j}||jdk(ry|jj|jk7r)tj j |jtj j |jyrd)rrfrr rr0rr)rhrorpr>rs rrtzStreamContext.__exit__;sj[[  R     & &**;*; ; II !5!5 6 T112rN) rvrwrxrr__annotations__rirmrrtrrrrrsF +,,x(:; )3S333rrrrct|S)zWrap around the Context-manager StreamContext that selects a given stream. Arguments: stream (Stream): selected stream. This manager is a no-op if it's ``None``. )rrs rrrFs   rcHtjj|||y)aset stream specified by the stream id, device index and device type Args: stream_id (int): not visible to the user, used to assigned to the specific stream. device_index (int): selected device index. device_type (int): selected device type.  stream_id device_index device_typeN)rr_xpu_setStreamrs r_set_stream_by_idrOs$ HH!rcv|ytt|j|j|jy)aSet the current stream.This is a wrapper API to set the stream. Usage of this function is discouraged in favor of the ``stream`` context manager. Args: stream (Stream): selected stream. This function is a no-op if this argument is ``None``. Nr)rFrrrrrs rrr]s4~L""((&&rcttjjt |d}t |d|d|dS)aRReturn the currently selected :class:`Stream` for a given device. Args: device (torch.device or int, optional): selected device. Returns the currently selected :class:`Stream` for the current device, given by :func:`~torch.xpu.current_device`, if :attr:`device` is ``None`` (default). Tr|rr r)rFrr_xpu_getCurrentStreamr r)r streamdatas rrrpsJL//&40J Q-jmTU rdata_ptrcttjj|t |d}t |d|d|dS)a;Return a :class:`Stream` from an external SYCL queue. This function is used to wrap SYCL queue created in other libraries in order to facilitate data exchange and multi-library interactions. .. note:: This function doesn't manage the queue life-cycle, it is the user responsibility to keep the referenced queue alive while this returned stream is being used. The different SYCL queue pointers will result in distinct :class:`Stream` objects, even if the SYCL queues they dereference are equivalent. Args: data_ptr(int): Integer representation of the `sycl::queue*` value passed externally. device(torch.device or int, optional): the device where the queue was originally created. It is the user responsibility to ensure the device is specified correctly. Tr|rr rr)rFrr_xpu_getStreamFromExternalr r)rr rs rget_stream_from_externalrsM$L44#FT:J Q-jmTU rcntt|d}tjj |S)a*Wait for all kernels in all streams on a XPU device to complete. Args: device (torch.device or int, optional): device for which to synchronize. It uses the current device, given by :func:`~torch.xpu.current_device`, if :attr:`device` is ``None`` (default). Tr|)rFr rr_xpu_synchronizers r synchronizers*L v 5F 88 $ $V ,,rc~tsgStjj}|gS|j S)z ++-J    rcjt}t|dk(ryddjd|DS)zIReturn XPU AOT(ahead-of-time) build flags this library was compiled with.rrOz-device ,c3 K|]}|ywrKr)rLarchs rrNz$get_gencode_flags..s:t:s )rlenrZ) arch_lists rget_gencode_flagsrs5I 9~ chh: ::; <t}|jyrK)r set_offset)default_generator final_devicers rcbz!_set_rng_state_offset..cbs*<8$$V,rN)rrD)rr rrs` @r_set_rng_state_offsetrsv&L-rNrcbtt|}t|}|jS)aLReturn the random number generator state offset of the specified GPU. Args: device (torch.device or int, optional): The device to return the RNG state offset of. Default: ``'xpu'`` (i.e., ``torch.device('xpu')``, the current XPU device). .. warning:: This function eagerly initializes XPU. )rFrr get_offset)r rrs r_get_rng_state_offsetrs-Lv&L&|4  ' ' ))r) empty_cachemax_memory_allocatedmax_memory_reserved mem_get_infomemory_allocatedmemory_reserved memory_statsmemory_stats_as_nested_dictreset_accumulated_memory_statsreset_peak_memory_stats) get_rng_stateget_rng_state_all initial_seed manual_seedmanual_seed_allr:r9 set_rng_stateset_rng_state_all)+rrrrrrr rr+rrrrr1rrrrrGrr-r3r7rrrrrrrrrrrr:r9rrrrrstreamsr)TrK)r0)ar threadingr> functoolsrtypingrrrrrtorch._Cr _device torch._utilsr r _utilsr rrrr5localrQLockrRrlisttuplerYrrrr6ry _device_tr<r Generatorboolrr_xpu_exchangeDevicer$_xpu_maybeExchangeDevicer'r+r-r3r7rDrGrFrarrrdictrr1rrrrrrrrrrrrrrmemoryrrrrrrrrrrrandomrrrrrr:r9rr__all__rrrrs11 #6%" y%y~~'t (2t8 d3i '(%(($6 F '3T) * %'02E%((,,-2d > 8888xx33"XX>>''=>NNNNsNsN 1*c**d 442 G%P  (   (y (T ( .HY/ .3 . 4(9"5c3h4 *(9"5 *AU *%% c3 45 %,, '3'3T!8./!M! v&8I.&&26#I. 6 - -T -tCy=3= -5<< -EHH,>,> -:? sC56 & *%S%,,(>"? *C *"      , r