scatter

paddle.distributed. scatter ( tensor, tensor_list=None, src=0, group=None, use_calc_stream=True ) [source]

Scatter a tensor to all participators. As shown below, 4 GPUs each start 4 processes and the source of the scatter is GPU0. Through scatter operator, the data in GPU0 will be sent to all GPUs averagely.

scatter
Parameters
  • tensor (Tensor) – The output Tensor. Its data type should be float16, float32, float64, int32 or int64.

  • tensor_list (list|tuple) – A list/tuple of Tensors to scatter. Every element in the list must be a Tensor whose data type should be float16, float32, float64, int32 or int64. Default value is None.

  • src (int) – The source rank id. Default value is 0.

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

  • use_calc_stream (bool) – Wether to use calculation stream (True) or communication stream (False). Default to True.

Returns

None.

Examples

# required: distributed
import numpy as np
import paddle
from paddle.distributed import init_parallel_env

paddle.set_device('gpu:%d'%paddle.distributed.ParallelEnv().dev_id)
init_parallel_env()
if paddle.distributed.ParallelEnv().local_rank == 0:
    np_data1 = np.array([7, 8, 9])
    np_data2 = np.array([10, 11, 12])
else:
    np_data1 = np.array([1, 2, 3])
    np_data2 = np.array([4, 5, 6])
data1 = paddle.to_tensor(np_data1)
data2 = paddle.to_tensor(np_data2)
if paddle.distributed.ParallelEnv().local_rank == 0:
    paddle.distributed.scatter(data1, src=1)
else:
    paddle.distributed.scatter(data1, tensor_list=[data1, data2], src=1)
out = data1.numpy()