ΑiSSKJr SSKrSSKJrJr SSKJr SSK J r SSK J r J r \(a$SSKJrJr SSK Jr SSKJr SS KJr \\\\/\4r"S S 5r\R0SSS jj5rSS jrSSjrg)) annotationsN) TYPE_CHECKINGCallable) framework)_get_global_group_warn_cur_rank_not_in_group) GeneratorSequence)Tensor)task)Groupcd\rSrSr%SrS\S'S\S'S\S'S \S 'SSS jjrS rg )P2POp$a? A class that makes point-to-point operations for "batch_isend_irecv". This class creates the type of P2P operation, communication buffer, peer rank, Group. Instances of this class will be passed to ``paddle.distributed.batch_isend_irecv`` for point-to-point communication. Args: op (callable): A function to send data to or receive data from a peer process. The type of ``op`` is either ``paddle.distributed.isend`` or ``paddle.distributed.irecv``. tensor (Tensor): Tensor to send or receive. peer (int): The destination or source rank. group (Group, optional): The group instance return by new_group or None for global default group. Default: None. Examples: .. code-block:: python >>> # doctest: +REQUIRES(env: DISTRIBUTED) >>> import paddle >>> import paddle.distributed as dist >>> dist.init_parallel_env() >>> rank = dist.get_rank() >>> world_size = dist.get_world_size() >>> send_t = paddle.arange(2) + rank >>> # paddle.tensor([0, 1]) # Rank-0 >>> # paddle.tensor([1, 2]) # Rank-1 >>> recv_t = paddle.empty(shape=[2], dtype=send_t.dtype) >>> send_op = dist.P2POp(dist.isend, send_t, (rank + 1) % world_size) >>> recv_op = dist.P2POp(dist.irecv, recv_t, (rank - 1 + world_size) % world_size) _P2POpTypeopr tensorintpeer Group | NonegroupNcU[R[R4;a [S5eXlX lX0lUc[5UlgUUlg)NztInvalid ``op`` function. Expected ``op`` to be of type ``paddle.distributed.isend`` or ``paddle.distributed.irecv``.) distisendirecv RuntimeErrorrrrrr)selfrrrrs r/var/www/html/banglarbhumi/venv/lib/python3.13/site-packages/paddle/distributed/communication/batch_isend_irecv.py__init__P2POp.__init__PsQ djj$**- -0    ,1M&( u )rrrrN) rrrr rrrrreturnNone)__name__ __module__ __qualname____firstlineno____doc____annotations__r__static_attributes__r!rrr$se$L N N I # E EE E  E  EEr!rc#F# Uc [5OUnURnUR5 Sv Ub[U5S:XaUR 5 gUR U5 g!Ub[U5S:XaUR 5 fUR U5 f=f7f)Nr)r process_group_start_coalescinglen_end_coalescing)rtaskspgs r_coalescing_managerr4ds$)=  eE   B& =CJ!O      u % =CJ!O      u %s,B!A(5B!(6BB!c^[U[5(a[SU55(d [S5eUSRR m[U4SjU55(d [S5eg)zi Helper to check that the ``p2p_op_list`` is a list of P2POp instances and all ops use the same backend. c3B# UHn[U[5v M g7fr") isinstancer).0p2p_ops r %_check_p2p_op_list..ys40;f 65!! sz[Invalid ``p2p_op_list``. Each op is expected to to be of type ``paddle.distributed.P2POp``.rc3V># UHnTURR:Hv M g7fr")rbackend)r8r9r=s rr:r;s I[6w&,,...[s&)z(All groups need to use the same backend.N)r7listallrrr=) p2p_op_listr=s @r_check_p2p_op_listrAtsy k4 ( (40;411 :  !n""**G I[I I IEFF Jr!c[U5 USRn[U5(ag[R"5(aUc [ 5OUnUR n/n[X5 UHRnURnURnURnURnU"XgU5n U cMAURU 5 MT SSS5 U$[S5e!,(df  U$=f)a Send or Receive a batch of tensors asynchronously and return a list of requests. Process each of the point-to-point operations in ``p2p_op_list`` and return the corresponding tasks. NCCL are currently supported. Args: p2p_op_list (List[P2POp]): A list of point-to-point operations(type of each operator is ``paddle.distributed.P2POp``). The order of the isend/irecv in the list matters and it needs to match with corresponding isend/irecv on the remote end. Returns: A list of distributed tasks returned by calling the corresponding op in the op_list. Warning: This API only supports the dygraph mode. Examples: .. code-block:: python >>> # doctest: +REQUIRES(env: DISTRIBUTED) >>> import paddle >>> import paddle.distributed as dist >>> dist.init_parallel_env() >>> rank = dist.get_rank() >>> world_size = dist.get_world_size() >>> send_t = paddle.arange(2) + rank >>> # paddle.tensor([0, 1]) # Rank-0 >>> # paddle.tensor([1, 2]) # Rank-1 >>> recv_t = paddle.empty(shape=[2], dtype=send_t.dtype) >>> send_op = dist.P2POp(dist.isend, send_t, (rank + 1) % world_size) >>> recv_op = dist.P2POp(dist.irecv, recv_t, (rank - 1 + world_size) % world_size) >>> tasks = dist.batch_isend_irecv([send_op, recv_op]) >>> for task in tasks: ... task.wait() >>> print(recv_t) >>> # paddle.tensor([1, 2]) # Rank-0 >>> # paddle.tensor([0, 1]) # Rank-1 rNz*Don't support static graph mode currently.) rArrrin_dynamic_moderr=r4rrrappendr) r@rr=r2r9rrr comm_groupr s rbatch_isend_irecvrFsd{# N E"5))  ""',}!#%--  .%YY{{#\\ & 3#LL&&/ GHH/ . s.AC2C C+r")rr r2z task | Noner#zGenerator[None, None, None])r@zSequence[P2POp]r#r$)r@z list[P2POp]r#z list[task]) __future__r contextlibtypingrrpaddle.distributed distributedrpaddler&paddle.distributed.communication.grouprrcollections.abcr r r paddle.base.corer r rrrcontextmanagerr4rArFr,r!rrQs#*! 3%(63.45J=E=E@ '+ &  &$ &  & &G$FIr!