ϑiSSKrSSKrSSKJr SSKJr SSKJr SSKJr SSK J r SSK J r /r SSjrS rS rS r"S S 5rSrSrSrSSjrg)N)_C_ops)core)check_variable_and_dtype)EagerParamBase) LayerHelper)in_dynamic_or_pir_modec[5(a[R"X5$[US/SQS5 [ S0[ 5D6nUR URURS9nURSSU/0SU/0SU0S9 U$) a  This OP takes in the Tensor :attr:`x` and change it to output with aclFormat with int value. This API is only used for Ascend NPU. Args: x(Tensor): An input N-D Tensor with data type bool, float16, float32, float64, int32, int64, int16, int8, uint8. format(int): Storage data format of the output in aclFormat, default value is -1. Returns: Tensor: A Tensor with acl storage format on Ascend NPU. Examples: .. code-block:: pycon >>> # doctest: +REQUIRES(env:NPU) >>> import paddle >>> paddle.device.set_device('npu') >>> x = paddle.ones(shape=[6]) >>> y = paddle.incubate._npu_identity(x, 3) # ACL_FORMAT_NC1HWC0 = 3 >>> print(y.shape) paddle.Size([1, 1, 1, 1, 16]) x) boolint8uint8int16int32int64float16float32float64 npu_identity)dtype stop_gradientoutformat)typeinputsoutputsattrs)r) rrrrrlocals"create_variable_for_type_inferencerr append_op)r rhelperrs c/var/www/html/banglarbhumi/venv/lib/python3.13/site-packages/paddle/incubate/tensor/manipulation.py _npu_identityr"s6""1--    "8vx877''8  !:SENV$   cV[U[5(aL[R"UR5n[UR UR 40UD6nU"X05nX44$[U[R5(a)[RR5nU"XP5nXT4$g)zh Helper to create a new destination tensor and call 'func(dst, src)' which is either offload or reload. N) isinstancercopydeepcopy__dict__shaperpaddleTensorreager) src_tensorfuncstate new_paramtask new_varbases r!_load_reload_implr3Zs *n-- j112":#3#3Z5E5EOO I* J . .jj'') K,   /r#c[RR5n[R"5(a[R "5$[ SU55(ag[R"5$)zZ Constructs a new AsyncLoad object. It is used to load/reload data asynchronously on GPU. c3N# UHn[R"U5v M g7fNr*is_compiled_with_custom_device.0devs r! $create_async_load..rs! >Ls--c22n#%N)r*deviceget_all_custom_device_typeis_compiled_with_xpur XpuAsyncLoadany AsyncLoad)custom_devicess r!create_async_loadrFjs` ]]==?N ""$$  ""  >L   ~~r#c,[R"5$)zZ Constructs a new AsyncLoad object. It is used to load/reload data asynchronously on XPU. )rrBr#r!create_xpu_async_loadrIzs    r#c*\rSrSrSrSrSrSrSrg)_NoopAsyncTasku(A dummy Task for sync‐fallback on XPU.cg)NTrHselfs r! is_completed_NoopAsyncTask.is_completedsr#cgr6rHrNs r!cpu_wait_NoopAsyncTask.cpu_wait r#cgr6rHrNs r!xpu_wait_NoopAsyncTask.xpu_waitrUr#rHN) __name__ __module__ __qualname____firstlineno____doc__rPrSrW__static_attributes__rHr#r!rKrKs2  r#rKcL[R"5=(a- [US5=(a URR 5n[R R 5n[SU55=(a0 [US5=(a URR5U;nU(dU(aRUR5n[R"UR5[R"5S9nU[54$[XR5$)a Loads the source tensor into the destination tensor asynchronously. Args: src_tensor (EagerParamBase|paddle.Tensor): The source tensor. async_load (core.AsyncLoad): The AsyncLoad object. Returns: tuple: A tuple containing two elements: - dest_tensor (EagerParamBase|paddle.Tensor): The destination tensor. - task (Task): The task that loads the source tensor into the destination tensor. placec3N# UHn[R"U5v M g7fr6r7r9s r!r< async_offload..s! BP3F 1 1# 6 6.r>r`)r*rAhasattrr` is_xpu_placer?r@rCcustom_device_typecpu to_tensornumpyCPUPlacerKr3offload)r- async_load is_xpu_tensorrEis_custom_tensor host_tensorrs r! async_offloadrps ##% , J ( ,    ) ) +]]==?N  BP   D J ( D    / / 1^ C ( nn& {002&//:KLN$$$ Z);); <rsX <003  9x !     #=L<44 :r#