python/parallel/test_alltoall.py

# Copyright 2019 Huawei Technologies Co., Ltd
#
# 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 re
import pytest
import numpy as np

import mindspore as ms
import mindspore.nn as nn
from mindspore import Tensor
from mindspore import context
from mindspore.common.api import _cell_graph_executor
from mindspore.common.parameter import Parameter
from mindspore.nn.loss import SoftmaxCrossEntropyWithLogits
from mindspore.nn.optim.momentum import Momentum
from mindspore.ops import operations as P
from mindspore.ops.operations.comm_ops import AlltoAll
from mindspore.parallel._utils import _reset_op_id
from mindspore.train import Model
from mindspore.context import ParallelMode
from mindspore.communication.management import GlobalComm, init
from tests.dataset_mock import MindData

context.set_context(device_target="Ascend")
GlobalComm.CHECK_ENVS = False
init("hccl")
GlobalComm.CHECK_ENVS = True

_x1 = Tensor(np.ones([64, 3, 224, 224]), dtype=ms.float32)


class Dataset(MindData):
    def __init__(self, predict, label, length=3):
        super(Dataset, self).__init__(size=length)
        self.predict = predict
        self.label = label
        self.index = 0
        self.length = length

    def __iter__(self):
        return self

    def __next__(self):
        if self.index >= self.length:
            raise StopIteration
        self.index += 1
        return self.predict, self.label

    def reset(self):
        self.index = 0


class AllToAllNet(nn.Cell):
    def __init__(self, strategy1):
        super(AllToAllNet, self).__init__()
        self.matmul = P.MatMul().shard(((1, 1), (1, 8)))
        self.matmul_weight = Parameter(Tensor(np.ones([128, 256]), dtype=ms.float32), name="weight")
        self.transpose1 = P.Transpose().shard(strategy1)

    def construct(self, x):
        x = self.matmul(x, self.matmul_weight)
        x = self.transpose1(x, (1, 0))
        return x


def all_to_all_net(strategy1):
    return AllToAllNet(strategy1=strategy1)


def all_to_all_common(strategy1):
    learning_rate = 0.1
    momentum = 0.9
    epoch_size = 2

    context.reset_auto_parallel_context()
    context.set_auto_parallel_context(parallel_mode=ParallelMode.SEMI_AUTO_PARALLEL, device_num=8)
    predict = Tensor(np.ones([32, 128]), dtype=ms.float32)
    label = Tensor(np.ones([32]), dtype=ms.int32)
    dataset = Dataset(predict, label, 2)
    net = all_to_all_net(strategy1)

    loss = SoftmaxCrossEntropyWithLogits(sparse=True)
    loss.softmax_cross_entropy.shard(((8, 1), (8, 1)))
    loss.one_hot.shard(((8, 1), (), ()))
    opt = Momentum(net.trainable_params(), learning_rate, momentum)
    model = Model(net, loss, opt)

    model.train(epoch_size, dataset, dataset_sink_mode=False)
    strategys = _cell_graph_executor._get_shard_strategy(model._train_network)
    return strategys


def test_all_to_all():
    strategy1 = ((8, 1),)
    context.set_context(mode=context.GRAPH_MODE)
    _reset_op_id()
    strategys = all_to_all_common(strategy1)
    print(strategys)
    for (k, v) in strategys.items():
        if re.search('SoftmaxCrossEntropyWithLogits-op', k) is not None:
            assert v == [[8, 1], [8, 1]]
        elif re.search('OneHot-op', k) is not None:
            assert v == [[8, 1], [], []]
        elif re.search('Transpose-op', k) is not None:
            assert v == [[8, 1]]
        elif re.search('MatMul-op', k) is not None:
            assert v == [[1, 1], [1, 8]]


def test_all_to_all_success():
    """
    Feature: AlltoAll
    Description: on 8p, a 4d tensor split at dim 2 and concat at dim 3
    Expectation: success
    """
    context.set_auto_parallel_context(device_num=8, global_rank=0)

    class Net(nn.Cell):
        def __init__(self):
            super(Net, self).__init__()
            self.alltoallv = AlltoAll(split_count=8, split_dim=2, concat_dim=3)

        def construct(self, x1):
            out = self.alltoallv(x1)
            return out

    net = Net()
    _cell_graph_executor.compile(net, _x1)


def test_all_to_all_invalid_split_count_value_failed():
    """
    Feature: AlltoAll
    Description: split_count should be equal to rank size, but not
    Expectation: throw ValueError
    """
    context.set_auto_parallel_context(device_num=8, global_rank=0)

    class Net(nn.Cell):
        def __init__(self):
            super(Net, self).__init__()
            self.alltoallv = AlltoAll(split_count=7, split_dim=2, concat_dim=3)

        def construct(self, x1):
            out = self.alltoallv(x1)
            return out

    with pytest.raises(ValueError):
        net = Net()
        _cell_graph_executor.compile(net, _x1)


def test_all_to_all_invalid_split_count_type_failed():
    """
    Feature: AlltoAll
    Description: split_count should be int, but a list is given
    Expectation: throw TypeError
    """
    context.set_auto_parallel_context(device_num=8, global_rank=0)

    class Net(nn.Cell):
        def __init__(self):
            super(Net, self).__init__()
            self.alltoallv = AlltoAll(split_count=[8], split_dim=2, concat_dim=3)

        def construct(self, x1):
            out = self.alltoallv(x1)
            return out

    with pytest.raises(TypeError):
        net = Net()
        _cell_graph_executor.compile(net, _x1)


def test_all_to_all_invalid_split_dim_value_failed():
    """
    Feature: AlltoAll
    Description: split_dim over input shape
    Expectation: throw IndexError
    """
    context.set_auto_parallel_context(device_num=8, global_rank=0)

    class Net(nn.Cell):
        def __init__(self):
            super(Net, self).__init__()
            self.alltoallv = AlltoAll(split_count=8, split_dim=4, concat_dim=3)

        def construct(self, x1):
            out = self.alltoallv(x1)
            return out

    with pytest.raises(IndexError):
        net = Net()
        _cell_graph_executor.compile(net, _x1)


def test_all_to_all_invalid_split_dim_type_failed():
    """
    Feature: AlltoAll
    Description: split_dim should be int, but a tuple is given
    Expectation: throw TypeError
    """
    context.set_auto_parallel_context(device_num=8, global_rank=0)

    class Net(nn.Cell):
        def __init__(self):
            super(Net, self).__init__()
            self.alltoallv = AlltoAll(split_count=8, split_dim=(3,), concat_dim=3)

        def construct(self, x1):
            out = self.alltoallv(x1)
            return out

    with pytest.raises(TypeError):
        net = Net()
        _cell_graph_executor.compile(net, _x1)


def test_all_to_all_invalid_concat_dim_value_failed():
    """
    Feature: AlltoAll
    Description: concat_dim over input shape
    Expectation: throw IndexError
    """
    context.set_auto_parallel_context(device_num=8, global_rank=0)

    class Net(nn.Cell):
        def __init__(self):
            super(Net, self).__init__()
            self.alltoallv = AlltoAll(split_count=8, split_dim=3, concat_dim=4)

        def construct(self, x1):
            out = self.alltoallv(x1)
            return out

    with pytest.raises(IndexError):
        net = Net()
        _cell_graph_executor.compile(net, _x1)


def test_all_to_all_invalid_concat_dim_type_failed():
    """
    Feature: AlltoAll
    Description: concat_dim should be int, but a tuple is given
    Expectation: throw TypeError
    """
    context.set_auto_parallel_context(device_num=8, global_rank=0)

    class Net(nn.Cell):
        def __init__(self):
            super(Net, self).__init__()
            self.alltoallv = AlltoAll(split_count=8, split_dim=3, concat_dim=([3],))

        def construct(self, x1):
            out = self.alltoallv(x1)
            return out

    with pytest.raises(TypeError):
        net = Net()
        _cell_graph_executor.compile(net, _x1)


def test_all_to_all_invalid_split_count_cannot_be_divisible_failed():
    """
    Feature: AlltoAll
    Description: shape at split_dim should be divisible by split_count, but not
    Expectation: throw ValueError
    """
    context.set_auto_parallel_context(device_num=3, global_rank=0)

    class Net(nn.Cell):
        def __init__(self):
            super(Net, self).__init__()
            self.alltoallv = AlltoAll(split_count=3, split_dim=3, concat_dim=3)

        def construct(self, x1):
            out = self.alltoallv(x1)
            return out

    with pytest.raises(ValueError):
        net = Net()
        _cell_graph_executor.compile(net, _x1)


def test_all_to_all_invalid_group_type_failed():
    """
    Feature: AlltoAll
    Description: group should be str, but a tuple is given
    Expectation: throw TypeError
    """
    context.set_auto_parallel_context(device_num=8, global_rank=0)

    class Net(nn.Cell):
        def __init__(self):
            super(Net, self).__init__()
            self.alltoallv = AlltoAll(split_count=8, split_dim=3, concat_dim=3, group=3)

        def construct(self, x1):
            out = self.alltoallv(x1)
            return out

    with pytest.raises(TypeError):
        net = Net()
        _cell_graph_executor.compile(net, _x1)


if __name__ == '__main__':
    test_all_to_all()