1# Copyright 2019 Huawei Technologies Co., Ltd 2# 3# Licensed under the Apache License, Version 2.0 (the "License"); 4# you may not use this file except in compliance with the License. 5# You may obtain a copy of the License at 6# 7# http://www.apache.org/licenses/LICENSE-2.0 8# 9# Unless required by applicable law or agreed to in writing, software 10# distributed under the License is distributed on an "AS IS" BASIS, 11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12# See the License for the specific language governing permissions and 13# limitations under the License. 14 15import numpy as np 16 17import mindspore as ms 18import mindspore.nn as nn 19from mindspore import Tensor 20from mindspore import context 21from mindspore.common.api import _cell_graph_executor 22from mindspore.ops import composite as C 23from mindspore.ops import operations as P 24from mindspore.parallel._utils import _reset_op_id as reset_op_id 25from tests.ut.python.ops.test_math_ops import VirtualLoss 26 27 28grad_all = C.GradOperation(get_all=True) 29 30 31class NetWithLoss(nn.Cell): 32 def __init__(self, network): 33 super(NetWithLoss, self).__init__() 34 self.loss = VirtualLoss() 35 self.network = network 36 37 def construct(self, x, y, b): 38 predict = self.network(x, y, b) 39 return self.loss(predict) 40 41 42class GradWrap(nn.Cell): 43 def __init__(self, network): 44 super(GradWrap, self).__init__() 45 self.network = network 46 47 def construct(self, x, y, b): 48 return grad_all(self.network)(x, y, b) 49 50 51# core dump, step_auto_parallel should SetInputs for transpose axis 52def test_two_matmul_transpose(): 53 class Net(nn.Cell): 54 def __init__(self): 55 super().__init__() 56 self.matmul1 = P.MatMul() 57 self.matmul2 = P.MatMul() 58 self.transpose1 = P.Transpose() 59 self.transpose2 = P.Transpose() 60 61 def construct(self, x, y, b): 62 out = self.matmul1(x, y) 63 out = self.matmul2(out, b) 64 out = self.transpose1(out, (1, 0)) 65 out = self.transpose2(out, (1, 0)) 66 return out 67 68 size = 16 69 context.set_auto_parallel_context(device_num=size, global_rank=0) 70 x = Tensor(np.ones([128, 32]), dtype=ms.float32) 71 y = Tensor(np.ones([32, 64]), dtype=ms.float32) 72 b = Tensor(np.ones([64, 64]), dtype=ms.float32) 73 74 net = NetWithLoss(Net()) 75 context.set_auto_parallel_context(parallel_mode="auto_parallel") 76 net.set_auto_parallel() 77 reset_op_id() 78 79 net.set_train() 80 _cell_graph_executor.compile(net, x, y, b, phase='train') 81 strategies = _cell_graph_executor._get_shard_strategy(net) 82 print(strategies) 83 expected_strategies = {'Default/network-Net/Transpose-op0': [[1, 16]], 84 'Default/network-Net/Transpose-op1': [[16, 1]], 85 'Default/network-Net/MatMul-op3': [[16, 1], [1, 1]], 86 'Default/network-Net/MatMul-op2': [[16, 1], [1, 1]]} 87 assert strategies == expected_strategies 88