iKUdZddlZddlZddlmZddlmZddlmZm Z m Z ddl Z ddl Z ddl m ZddlmZmZdd lmZdd lmZmZd aej0Zej4Zgaeeegdfeefe d <e!e jDd dZ#e eee$dfZ%ea&dZ'ee jDjPe d<de)fdZ*e*rCe jDjVZ+e jDjXZ-e jDj\Z/nedZ+de$de$fdZ-de$de$fdZ/edde$fdZ0de)fdZ1dBde)de)fdZ2de)fdZ3dZ4dCdZ5dCdZ6dCd Z7Gd!d"Z8Gd#dZ Gd$d%e Z9de%ddfd&Z:dDde%dzdefd'Z;eddDde%dzdede$fd*Z?de$eze jzde jfd+Z@de%d,e%de)fd-ZAGd.d/ZBd0e d1deBfd2ZCdCd3ZDd0eddfd4ZEdDde%dzdefd5ZFdDd6e$de%dzdefd7ZGdDde%ddfd8ZHdeefd9ZIdefd:ZJde jde jDjPfd;ZK dEdZMdd?lNmOZOmPZPmQZQmRZRmSZSmTZTmUZUmVZVmWZWmXZXmYZYmZZZm[Z[m\Z\dd@l]m^Z^m_Z_m`Z`maZambZbmcZcmdZdmeZemfZfgdAZgy)Fz This package introduces support for the XPU backend, specifically tailored for Intel GPU optimization. This package is lazily initialized, so you can always import it, and use :func:`is_available()` to determine if your system supports XPU. N)Callable) lru_cache)AnyOptionalUniondevice) _dummy_type_LazySeedTracker)_get_device_index)EventStreamF _queued_calls_xpu_isInBadForkcyNFrL/var/www/html/engine/venv/lib/python3.12/site-packages/torch/xpu/__init__.pyrsrrdefault_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$2!"LMMrctdr!r"rs r_maybe_exchange_devicer'5r%r)maxsizecTtsytjjS)z*Return the number of XPU device available.r)rrr_xpu_getDeviceCountrrr device_countr+9s > 88 ' ' ))rctdkDS)z7Return a bool indicating if XPU is currently available.r)r+rrr is_availabler-As >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_supportedr3Gs. > F 99 * * , E ErchtsytjjjS)zGReturn a bool indicating if the current XPU device supports dtype tf32.F)r-rr0r1'has_subgroup_matrix_multiply_accumulaterrris_tf32_supportedr6Qs% > 99 * * , T TTrc(txr t S)z8Return whether PyTorch's XPU state has been initialized.) _initialized_is_in_bad_forkrrris_initializedr:\s  1 111rc ftr|y|jddr)tj|t j y|jddr)tj |t j ytj|t j fy)Nseed_allFseed) r:get_lazy_seed_trackerqueue_seed_all traceback format_stack queue_seedrappend)callablekwargss r _lazy_callrGas{  ::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_initrrrinitrJos 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: )r:hasattr_tls_initialization_lockr9 RuntimeErrorrAssertionErrorrr _xpu_initrLrextendr? get_calls Exceptionstrjoindelattrr8) queued_callorig_tracebackemsgs rrIrIxsA74):; !    !!  I ~ !FG G  $X0B0L0L0NXX -/< 0+ ^0M 0 D+ , C!!2!0OPSTUPVxXBBD''.BYAZ\$C.a/ 0 D+ ,A!!sM D=BD=9 D(C+ D(D=+ D%4,D  D%%D((D::D==Ec6eZdZdeddfdZdZdededefd Zy) _DeviceGuardindexrNc ||_d|_yN)idxprev_idx)selfres r__init__z_DeviceGuard.__init__s rc`tjj|j|_yrNrr0r$rirjrks r __enter__z_DeviceGuard.__enter__ 22488< rtypevaluerAc`tjj|j|_yrrr0r'rjrirkrrrsrAs r__exit__z_DeviceGuard.__exit__9933DMMBr)__name__ __module__ __qualname__intrlrprrwrrrrdrds4cd=Srrdc:eZdZdZdeddfdZdZdededefd Zy) 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``. rNc6t|d|_d|_y)NToptionalrh)r rirj)rkr s rrlzdevice.__init__s$Vd; rc`tjj|j|_yrNrnros rrpzdevice.__enter__rqrrrrsrAc`tjj|j|_yrrurvs rrwzdevice.__exit__rxr)ryrzr{__doc__rrlrprwrrrr r s9st=Src$eZdZdZdfd ZxZS) 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| |yrg)is_xpu get_devicesuperrl)rkobjri __class__s rrlzdevice_of.__init__s$"%**cnn" rrN)ryrzr{rrl __classcell__)rs@rrrsrrcxtt|}|dk\r tjj |yy)zSet the current device. Args: device (torch.device or int or str): selected device. This function is a no-op if this argument is negative. rN)rIr rr_xpu_setDevicers r set_devicers1L v &F { 'rc,t|jS)aGet the name of a device. Args: device (torch.device or int or str, optional): device for which to return the name. This function is a no-op if this argument is a negative integer. It uses the current device, given by :func:`~torch.xpu.current_device`, if :attr:`device` is ``None`` (default). Returns: str: the name of the device )r1namers rget_device_namers ! ( - --rc t|}ttttt dt ttf}t|Dcic].}|jdstt||x}|r||0c}Scc}w)aGet the xpu capability of a device. Args: device (torch.device or int or str, optional): device for which to return the device capability. This function is a no-op if this argument is a negative integer. It uses the current device, given by :func:`~torch.xpu.current_device`, if :attr:`device` is ``None`` (default). Returns: dict[str, Any]: the xpu capability dictionary of the device N__) r1r|floatboolr\rrlisttupledictdir startswith isinstancegetattr)r propsserializable_typeskeyrss rget_device_capabilityrsv "& )EudCdT5$Ou:  ~~d# !447I J U   s3A>cFtt|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)rIr _get_device_propertiesrs rr1r1s L v 5F !& ))rcRttjjS)z0Return the index of a currently selected device.)rIrr_xpu_getDevicerrrcurrent_devicersL 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)rr\rr r|rs r _get_devicersD &#f% M FC eV, Mrpeerctt|d}t|d}tjj ||S)a/Query whether a device can access a peer device's memory. Args: device (torch.device or int or str): selected device. peer (torch.device or int or str): peer device to query access to. Returns: bool: ``True`` if ``device`` can access ``peer``, ``False`` otherwise. Tr)rIr rr_xpu_canDeviceAccessPeer)r rs rcan_device_access_peerr)s8L v 5F TD 1D 88 , ,VT ::rcReZdZUdZeded<dedddfdZdZd ed ed 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_streamstreamrNc\||_tdd|_|jd|_yy)NTrh)rr ri)rkrs rrlzStreamContext.__init__Gs- $T40 88 DH rc|j}||jdk(rytjj d|_|j j |j k7rLt |j 5tjj |j |_dddtjj|y#1swY)xYwrg) rrirr0current_streamsrc_prev_streamr dst_prev_stream set_stream)rkrs rrpzStreamContext.__enter__Ms[[  R $yy77=    & &**;*; ; ))* S',yy'?'? @Q@Q'R$ S Z( S Ss ;/CCrrrsrAc*|j}||jdk(ry|jj|jk7r)tj j |jtj j |jyrg)rrirr rr0rr)rkrrrsrArs rrwzStreamContext.__exit__Ysj[[  R     & &**;*; ; II !5!5 6 T112r) ryrzr{rr__annotations__rlrprrwrrrrr9sM +,,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 rrrds   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_idrms$ 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)rIrrrrrs 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). Trrr r)rIrr_xpu_getCurrentStreamr r)r streamdatas rrrsJL//&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. Trrr rr)rIrr_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)rIr 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.rrRz-device ,c3 K|]}|ywrNr)rOarchs rrQz$get_gencode_flags..s:t:s )rlenr]) arch_lists rget_gencode_flagsrs5I 9~ chh: ::; <t}|jyrN)r set_offset)default_generator final_devicers rcbz!_set_rng_state_offset..cbs*<8$$V,rr)rrG)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. )rIrr get_offset)r rrs r_get_rng_state_offsetrs-Lv&L&|4  ' ' ))r)change_current_allocator empty_cacheget_per_process_memory_fractionmax_memory_allocatedmax_memory_reserved mem_get_infomemory_allocatedmemory_reserved memory_statsmemory_stats_as_nested_dictreset_accumulated_memory_statsreset_peak_memory_statsset_per_process_memory_fractionXPUPluggableAllocator) get_rng_stateget_rng_state_all initial_seed manual_seedmanual_seed_allr=r< set_rng_stateset_rng_state_all)1rrrrrrrrrr rr+rrrrr1rrrrrrJrr-r3r:r6rrrrrrrrrrrr=r<rrrrrrstreamsr)TrrN)r0)hr threadingrAcollections.abcr functoolsrtypingrrrrtorch._Cr _device torch._utilsr r _utilsr rrrr8localrTLockrUrrrr\rrrr9r| _device_tr?r Generatorrrr_xpu_exchangeDevicer$_xpu_maybeExchangeDevicer'r+r-r3r6r:rGrJrIrdrrrrrr1rrrrrrrrrrrrrrrmemoryrrrrrrrrrrrrrrrandomrrrrrr=r<rr__all__rrrrs$'' #6%" y%y~~'t (2t8 d3i '(%(($6 F '3T) * %'02E%((,,-2d > 8888xx33"XX>>''=>NNNNsNsN 1*c**d 44U4U2 G%P  (   (y (T ( .I, . . 4)d"2d38n2 $*  **"%% c ELL0 U\\  ;9 ;I ;$ ; (3(3V!8./!M! v$&9t+v$sI4DPV2 - -T -tCy=3= -5<< -EHH,>,> -5: sU\\1 & *#)ell": *s *"   2 r