mindspore.ops.communication.scatter_tensor

View Source On AtomGit
mindspore.ops.communication.scatter_tensor(output_tensor, input_tensor, src=0, group=None, async_op=False)[source]

Scatter tensor evenly across the processes in the specified communication group.

Note

  • The interface behavior only support Tensor input and scatter evenly, which is different from that of pytorch.distributed.scatter.

  • Only the tensor in process src (global rank) will do scatter.

  • Only support PyNative mode, Graph mode is not currently supported.

Parameters

  • output_tensor (Tensor) – Output tensor. It should have the same size across all ranks.

  • input_tensor (Tensor) – The input tensor to be scattered. The shape of tensor is \((x_1, x_2, ..., x_R)\).

  • src (int, optional) – Specifies the rank(global rank) of the process that send the tensor. And only process src will send the tensor. Default is 0.

  • group (str, optional) – The communication group to work on. Default: None, which means "hccl_world_group" in Ascend.

  • async_op (bool, optional) – Whether this operator should be an async operator. Default: False.

Returns

CommHandle. If async_op is set to True, CommHandle is an async work handle. If async_op is set to False, CommHandle will be None.

Raises

  • TypeError – If the type of output_tensor or input_tensor parameter is not Tensor, src is not an int, group is not a str, or async_op is not bool.

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

Supported Platforms:

Ascend

Examples

Note

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

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

This example should be run with 2 devices.

>>> import mindspore as ms
>>> from mindspore.ops.communication import init_process_group
>>> from mindspore.communication.comm_func import scatter_tensor
>>> import numpy as np
>>> # Launch 2 processes.
>>>
>>> init_process_group()
>>> input = ms.Tensor(np.arange(8).reshape([4, 2]).astype(np.float32))
>>> output = ms.Tensor(np.zeros([2, 2]).astype(np.float32))
>>> out = scatter_tensor(output, input, src=0)
>>> print(output)
# rank_0
[[0. 1.]
 [2. 3.]]
# rank_1
[[4. 5.]
 [6. 7.]]