isend¶

paddle.distributed. isend ( tensor, dst, group=None ) [source]

Sends a tensor asynchronously

Parameters

• tensor (Tensor) – The Tensor to send. Its data type should be float16, float32, float64, int32, int64, int8, uint8, bool or bfloat16.

• dst (int) – The destination rank.

• group (Group, optional) – The group instance return by new_group or None for global default group. Default: None.

Returns

A distributed task object.

Warning

This API only supports the dygraph mode.

Examples

# required: distributed
import paddle
import paddle.distributed as dist

dist.init_parallel_env()
if dist.get_rank() == 0:
    data = paddle.to_tensor([7, 8, 9])
    task = dist.isend(data, dst=1)
else:
    data = paddle.to_tensor([1, 2, 3])
    task = dist.irecv(data, src=0)
task.wait()
print(data)
# [7, 8, 9] (2 GPUs)