1# Copyright 2021 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# ============================================================================ 15 16import numpy as np 17import pytest 18 19import mindspore.context as context 20import mindspore.nn as nn 21from mindspore import Tensor 22from mindspore.ops import operations as P 23from mindspore.ops.composite import GradOperation 24 25 26class Grad(nn.Cell): 27 def __init__(self, network): 28 super(Grad, self).__init__() 29 self.grad = GradOperation(get_all=True, sens_param=True) 30 self.network = network 31 32 def construct(self, input_x, dout): 33 return self.grad(self.network)(input_x, dout) 34 35 36class Net(nn.Cell): 37 def __init__(self): 38 super(Net, self).__init__() 39 self.hswish = P.HSwish() 40 41 def construct(self, x): 42 return self.hswish(x) 43 44 45def expect_hswish_forward_result(x): 46 return np.where(x <= -3, 0, np.where(x >= 3, x, x * (x + 3) / 6)) 47 48 49def expect_hswish_backward_result(x, dout): 50 return np.where(x <= -3, 0, np.where(x >= 3, 1, x / 3 + 0.5)) * dout 51 52 53def judge_result_correct(result, expect): 54 assert result.dtype == expect.dtype 55 assert result.shape == expect.shape 56 assert np.allclose(result, expect) 57 58 59def generate_test_cases(np_type, mode): 60 context.set_context(mode=mode, device_target="GPU") 61 x = np.array([-1, -2, 0, 4, 5]).astype(np_type) 62 net = Net() 63 output = net(Tensor(x)) 64 expect = expect_hswish_forward_result(x) 65 judge_result_correct(output.asnumpy(), expect) 66 67 sens = np.array([-1.45, 0.63, 0.34, 6.43, 34.6]).astype(np_type) 68 backward_net = Grad(Net()) 69 output = backward_net(Tensor(x), Tensor(sens)) 70 expect = expect_hswish_backward_result(x, sens) 71 judge_result_correct(output[0].asnumpy(), expect) 72 73 74@pytest.mark.level0 75@pytest.mark.platform_x86_gpu_training 76@pytest.mark.env_onecard 77def test_hswish_forward_and_backward(): 78 modes = (context.GRAPH_MODE, context.PYNATIVE_MODE) 79 dtypes = (np.float32, np.float16) 80 for mode in modes: 81 for dtype in dtypes: 82 generate_test_cases(dtype, mode) 83