send.py 4.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
# 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.

15
import paddle.fluid.data_feeder as data_feeder
16
import paddle.framework as framework
17 18 19
from paddle.distributed.communication.group import (
    _get_global_group,
    _get_or_throw_group_rank,
20
    _warn_cur_rank_not_in_group,
21 22 23 24 25 26
)


def _send_in_dygraph(
    tensor, dst_rank_in_group, group, sync_op, use_calc_stream
):
27
    if use_calc_stream:
28 29 30
        return group.process_group.send_on_calc_stream(
            tensor, dst_rank_in_group
        )
31

32
    task = group.process_group.send(tensor, dst_rank_in_group, sync_op)
33 34 35 36 37 38
    if sync_op:
        task.wait()

    return task


39 40 41 42 43 44 45 46 47 48 49 50
def _send_in_static_mode(
    tensor, dst_rank_in_group, group, sync_op, use_calc_stream
):
    op_type = 'send_v2'
    data_feeder.check_variable_and_dtype(
        tensor,
        'tensor',
        ['float16', 'float32', 'float64', 'int32', 'int64'],
        'send',
    )

    ring_id = 0 if group is None else group.id
51
    helper = framework.LayerHelper(op_type, **locals())
52 53 54 55 56 57 58 59 60 61 62 63
    helper.append_op(
        type=op_type,
        inputs={'X': [tensor]},
        attrs={
            'ring_id': ring_id,
            'peer': dst_rank_in_group,
            'use_calc_stream': sync_op,
        },
    )
    return None


64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96
def send(tensor, dst=0, group=None, sync_op=True, use_calc_stream=False):
    """

    Send a tensor to the destination device.

    Args:
        tensor (Tensor): The tensor to send. Support float16, float32, float64, int32, int64, int8, uint8 or bool as its data type.
        dst (int, optional): Rank of the destination device. If none is given, use `0` as default.
        group (Group, optional): Communicate in which group. If none is given, use the global group as default.
        sync_op (bool, optional): Indicate whether the communication is sync or not. If none is given, use true as default.
        use_calc_stream (bool, optional): Indicate whether the communication is done on calculation stream. If none is given, use false as default. This
            option is designed for high performance demand, be careful to turn it on except you are clearly know its meaning.

    Returns:
        Return a task object.

    Examples:
        .. code-block:: python

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

            dist.init_parallel_env()
            local_rank = dist.get_rank()
            if local_rank == 0:
                data = paddle.to_tensor([[4, 5, 6], [4, 5, 6]])
                task = dist.stream.send(data, dst=1, sync_op=False)
            else:
                data = paddle.to_tensor([[1, 2, 3], [1, 2, 3]])
                task = dist.stream.recv(data, src=0, sync_op=False)
            task.wait()
            out = data.numpy()
97
            # [[4, 5, 6], [4, 5, 6]] (2 GPUs)
98
    """
99 100
    if _warn_cur_rank_not_in_group(group):
        return
101 102 103

    if not sync_op and use_calc_stream:
        raise RuntimeError(
104 105
            "use_calc_stream can only be True in sync op behavior."
        )
106 107

    if framework.in_dygraph_mode():
108 109
        group = _get_global_group() if group is None else group
        dst_rank_in_group = _get_or_throw_group_rank(dst, group)
110

111 112 113 114
        return _send_in_dygraph(
            tensor, dst_rank_in_group, group, sync_op, use_calc_stream
        )
    else:
115 116 117
        assert (
            group is None
        ), "Group can not be used in static graph mode for now."
118 119 120
        return _send_in_static_mode(
            tensor, dst, group, sync_op, use_calc_stream
        )