mindspore.ops.Send

class mindspore.ops.Send(sr_tag, dest_rank, group=GlobalComm.WORLD_COMM_GROUP, group_back=GlobalComm.WORLD_COMM_GROUP)

Send tensors to the specified dest_rank.

Note

Send and Receive must be used in combination and have same sr_tag.

Parameters

• sr_tag (int) – The tag to identify the send/recv message. The message sent by this operator will be received by the Receive op with the same "sr_tag".

• dest_rank (int) – A required integer identifying the destination rank.

• group (str, optional) – The communication group to work on. Default: GlobalComm.WORLD_COMM_GROUP.

• group_back (str, optional) – The communication group for backpropagation. Default: GlobalComm.WORLD_COMM_GROUP.

Inputs:

• input_x (Tensor) - The shape of tensor is $(x_1,x_2,...,x_R)$.

Raises

• TypeError – If group is not a str.

• RuntimeError – If device target is invalid, or backend is invalid, or distributed initialization fails.

• ValueError – If the local rank id of the calling process in the group is larger than the group's rank size.

Supported Platforms:

Ascend GPU

Examples

Note

Before running the following examples, you need to configure the communication environment variables.

For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method without any third-party or configuration file dependencies. Please see the msrun start up for more details.

This example should be run with 2 devices.

>>> import os
>>> import numpy as np
>>> import mindspore.ops as ops
>>> import mindspore.nn as nn
>>> import mindspore as ms
>>> from mindspore.communication import init
>>> from mindspore import Tensor
>>>
>>> ms.set_context(mode=ms.GRAPH_MODE, jit_level="O2")
>>> init()
>>>
>>> class SendNet(nn.Cell):
...     def __init__(self):
...         super(SendNet, self).__init__()
...         self.depend = ops.Depend()
...         self.send = ops.Send(sr_tag=0, dest_rank=1, group="hccl_world_group")
...
...     def construct(self, x):
...         out = self.depend(x, self.send(x))
...         return out
>>>
>>> class ReceiveNet(nn.Cell):
...     def __init__(self):
...         super(ReceiveNet, self).__init__()
...         self.recv = ops.Receive(sr_tag=0, src_rank=0, shape=[2, 8], dtype=ms.float32,
...                                 group="hccl_world_group")
...
...     def construct(self):
...         out = self.recv()
...         return out
>>>
>>> if __name__ == "__main__":
...     rank_id = os.environ["RANK_ID"]
...     rank_size = os.environ["RANK_SIZE"]
...     if rank_id == "0":
...         input_ = Tensor(np.ones([2, 8]).astype(np.float32))
...         send_net = SendNet()
...         output = send_net(input_)
...     else:
...         recv_net = ReceiveNet()
...         output = recv_net()
...         print(output.asnumpy())

Tutorial Examples:

• Distributed Set Communication Primitives - Send