scatter.py

# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import numpy as np

import paddle
import paddle.distributed as dist
from paddle import framework
from paddle.distributed.communication import stream

from .serialization_utils import (
    convert_object_to_tensor,
    convert_tensor_to_object,
)


def scatter(tensor, tensor_list=None, src=0, group=None, sync_op=True):
    """

    Scatter a tensor to all participators. As shown below, one process is started with a GPU and the source of the scatter
    is GPU0. Through scatter operator, the data in GPU0 will be sent to all GPUs averagely.

    .. image:: https://githubraw.cdn.bcebos.com/PaddlePaddle/docs/develop/docs/api/paddle/distributed/img/scatter.png
        :width: 800
        :alt: scatter
        :align: center

    Args:
        tensor (Tensor): The output Tensor. Its data type
            should be float16, float32, float64, int32, int64, int8, uint8, bool or bfloat16.
        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, int64, int8, uint8, bool or bfloat16. Default value is None.
        src (int): The source rank id. Default value is 0.
        group (Group, optional): The group instance return by new_group or None for global default group.
        sync_op (bool, optional): Whether this op is a sync op. The default value is True.

    Returns:
        None.

    Examples:
        .. code-block:: python

            >>> # doctest: +REQUIRES(env: DISTRIBUTED)
            >>> import paddle
            >>> import paddle.distributed as dist

            >>> dist.init_parallel_env()
            >>> if dist.get_rank() == 0:
            ...     data1 = paddle.to_tensor([7, 8, 9])
            ...     data2 = paddle.to_tensor([10, 11, 12])
            ...     dist.scatter(data1, src=1)
            >>> else:
            ...     data1 = paddle.to_tensor([1, 2, 3])
            ...     data2 = paddle.to_tensor([4, 5, 6])
            ...     dist.scatter(data1, tensor_list=[data1, data2], src=1)
            >>> print(data1, data2)
            >>> # [1, 2, 3] [10, 11, 12] (2 GPUs, out for rank 0)
            >>> # [4, 5, 6] [4, 5, 6] (2 GPUs, out for rank 1)
    """
    return stream.scatter(tensor, tensor_list, src, group, sync_op)


def scatter_object_list(
    out_object_list, in_object_list=None, src=0, group=None
):
    """

    Scatter picklable objects from the source to all others. Similiar to scatter(), but python object can be passed in.

    Args:
        out_object_list (list): The list of objects to store the scattered objects.
        in_object_list (list): The list of objects to scatter. Only objects on the src rank will be scattered.
        src (int): The source rank in global view.
        group (Group): The group instance return by new_group or None for global default group.

    Returns:
        None.

    Warning:
        This API only supports the dygraph mode.

    Examples:
        .. code-block:: python

            >>> # doctest: +REQUIRES(env: DISTRIBUTED)
            >>> import paddle.distributed as dist

            >>> dist.init_parallel_env()
            >>> out_object_list = []
            >>> if dist.get_rank() == 0:
            ...     in_object_list = [{'foo': [1, 2, 3]}, {'foo': [4, 5, 6]}]
            >>> else:
            ...     in_object_list = [{'bar': [1, 2, 3]}, {'bar': [4, 5, 6]}]
            >>> dist.scatter_object_list(out_object_list, in_object_list, src=1)
            >>> print(out_object_list)
            >>> # [{'bar': [1, 2, 3]}] (2 GPUs, out for rank 0)
            >>> # [{'bar': [4, 5, 6]}] (2 GPUs, out for rank 1)
    """
    assert (
        framework.in_dynamic_mode()
    ), "scatter_object_list doesn't support static graph mode."

    rank = dist.get_rank()
    in_obj_tensors = []
    in_obj_sizes = []

    if rank == src:
        for obj in in_object_list:
            obj_tensor, obj_size = convert_object_to_tensor(obj)
            in_obj_tensors.append(obj_tensor)
            in_obj_sizes.append(obj_size)
        max_obj_size_tensor = max(in_obj_sizes)
    else:
        max_obj_size_tensor = paddle.empty([], dtype="int64")
    stream.broadcast(max_obj_size_tensor, src)
    max_obj_size = int(max_obj_size_tensor.item())

    # resize to the same size
    in_tensor_list = []
    for tensor in in_obj_tensors:
        numpy_data = tensor.numpy()
        numpy_data = np.resize(numpy_data, [max_obj_size])
        in_tensor = paddle.to_tensor(numpy_data)
        in_tensor_list.append(in_tensor)
    out_tensor = paddle.empty([max_obj_size], dtype="uint8")
    scatter(out_tensor, in_tensor_list if rank == src else None, src, group)

    out_tensor_size = paddle.empty([], dtype="int64")
    scatter(out_tensor_size, in_obj_sizes if rank == src else None, src, group)

    out_object_list.clear()
    out_object_list.append(
        convert_tensor_to_object(out_tensor, out_tensor_size.item())
    )