1# Copyright 2020 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 pytest 17import numpy as np 18 19import mindspore 20import mindspore.nn as nn 21import mindspore.context as context 22 23from mindspore import Tensor 24from mindspore.ops.composite import GradOperation 25 26@pytest.mark.level0 27@pytest.mark.platform_x86_cpu 28@pytest.mark.env_onecard 29def test_mirror_pad(): 30 context.set_context(mode=context.GRAPH_MODE, device_target="CPU") 31 32 test1_arr_in = [[[[1, 2, 3], [4, 5, 6], [7, 8, 9]]]] 33 test_1_paddings = ((0, 0), (0, 0), (1, 1), (2, 2)) 34 test1_arr_exp = [[[[6, 5, 4, 5, 6, 5, 4], [3, 2, 1, 2, 3, 2, 1], [6, 5, 4, 5, 6, 5, 4], 35 [9, 8, 7, 8, 9, 8, 7], [6, 5, 4, 5, 6, 5, 4]]]] 36 37 test2_arr_in = [[[[1, 2, 3], [4, 5, 6], [7, 8, 9]]]] 38 test_2_paddings = ((0, 0), (0, 0), (1, 1), (2, 2)) 39 test2_arr_exp = [[[[2, 1, 1, 2, 3, 3, 2], [2, 1, 1, 2, 3, 3, 2], [5, 4, 4, 5, 6, 6, 5], 40 [8, 7, 7, 8, 9, 9, 8], [8, 7, 7, 8, 9, 9, 8]]]] 41 42 reflectOp = nn.Pad(mode='REFLECT', paddings=test_1_paddings) 43 symmOp = nn.Pad(mode='SYMMETRIC', paddings=test_2_paddings) 44 45 x_test_1 = Tensor(np.array(test1_arr_in), dtype=mindspore.float32) 46 x_test_2 = Tensor(np.array(test2_arr_in), dtype=mindspore.float32) 47 48 y_test_1 = reflectOp(x_test_1).asnumpy() 49 y_test_2 = symmOp(x_test_2).asnumpy() 50 51 print(np.array(test1_arr_in)) 52 print(y_test_1) 53 54 np.testing.assert_equal(np.array(test1_arr_exp), y_test_1) 55 np.testing.assert_equal(np.array(test2_arr_exp), y_test_2) 56 57 58class Grad(nn.Cell): 59 def __init__(self, network): 60 super(Grad, self).__init__() 61 self.grad = GradOperation(get_all=True, sens_param=True) 62 self.network = network 63 def construct(self, input_, output_grad): 64 return self.grad(self.network)(input_, output_grad) 65 66class Net(nn.Cell): 67 def __init__(self, pads, mode_): 68 super(Net, self).__init__() 69 self.pad = nn.Pad(mode=mode_, paddings=pads) 70 def construct(self, x): 71 return self.pad(x) 72 73 74@pytest.mark.level0 75@pytest.mark.platform_x86_cpu 76@pytest.mark.env_onecard 77def test_mirror_pad_backprop(): 78 context.set_context(mode=context.GRAPH_MODE, device_target="CPU") 79 test_arr_in = [[[[1, 2, 3], [4, 5, 6], [7, 8, 9]]]] # size -> 3*3 80 test_arr_in = Tensor(test_arr_in, dtype=mindspore.float32) 81 dy = (np.ones((1, 1, 4, 5)) * 0.1).astype(np.float32) 82 expected_dx = np.array([[[[0.2, 0.2, 0.1], 83 [0.4, 0.4, 0.2], 84 [0.2, 0.2, 0.1]]]]) 85 net = Grad(Net(((0, 0), (0, 0), (1, 0), (0, 2)), "REFLECT")) 86 dx = net(test_arr_in, Tensor(dy)) 87 dx = dx[0].asnumpy() 88 np.testing.assert_array_almost_equal(dx, expected_dx) 89 90@pytest.mark.level0 91@pytest.mark.platform_x86_cpu 92@pytest.mark.env_onecard 93def test_mirror_pad_fwd_back_4d_int32_reflect(): 94 context.set_context(mode=context.GRAPH_MODE, device_target="CPU") 95 # set constants 96 shape = (2, 3, 3, 5) 97 pads = ((1, 0), (2, 0), (1, 2), (3, 4)) 98 total_val = np.prod(shape) 99 test_arr_np = np.arange(total_val).reshape(shape) + 1 100 test_arr_ms = Tensor(test_arr_np, dtype=mindspore.int32) 101 # fwd_pass_check 102 op = nn.Pad(mode="REFLECT", paddings=pads) 103 expected_np_result = np.pad(test_arr_np, pads, 'reflect') 104 obtained_ms_res = op(test_arr_ms).asnumpy() 105 np.testing.assert_array_equal(expected_np_result, obtained_ms_res) 106 # backwards pass check 107 GradNet = Grad(Net(pads, "REFLECT")) 108 dy_value = Tensor(np.ones(obtained_ms_res.shape), dtype=mindspore.int32) 109 dx_value_obtained = GradNet(test_arr_ms, dy_value)[0].asnumpy() 110 dx_value_expected = np.array([[[[4, 6, 6, 6, 2], 111 [6, 9, 9, 9, 3], 112 [2, 3, 3, 3, 1]], 113 [[8, 12, 12, 12, 4], 114 [12, 18, 18, 18, 6], 115 [4, 6, 6, 6, 2]], 116 [[8, 12, 12, 12, 4], 117 [12, 18, 18, 18, 6], 118 [4, 6, 6, 6, 2]]], 119 [[[8, 12, 12, 12, 4], 120 [12, 18, 18, 18, 6], 121 [4, 6, 6, 6, 2]], 122 [[16, 24, 24, 24, 8], 123 [24, 36, 36, 36, 12], 124 [8, 12, 12, 12, 4]], 125 [[16, 24, 24, 24, 8], 126 [24, 36, 36, 36, 12], 127 [8, 12, 12, 12, 4]]]], dtype=np.int32) 128 np.testing.assert_array_equal(dx_value_expected, dx_value_obtained) 129 130 131@pytest.mark.level0 132@pytest.mark.platform_x86_cpu 133@pytest.mark.env_onecard 134def test_mirror_pad_fwd_back_4d_int32_symm(): 135 context.set_context(mode=context.GRAPH_MODE, device_target="CPU") 136 # set constants 137 shape = (2, 3, 3, 5) 138 pads = ((1, 0), (2, 0), (1, 2), (3, 4)) 139 total_val = np.prod(shape) 140 test_arr_np = np.arange(total_val).reshape(shape) + 1 141 test_arr_ms = Tensor(test_arr_np, dtype=mindspore.int32) 142 # fwd_pass_check 143 op = nn.Pad(mode="SYMMETRIC", paddings=pads) 144 expected_np_result = np.pad(test_arr_np, pads, 'symmetric') 145 obtained_ms_res = op(test_arr_ms).asnumpy() 146 np.testing.assert_array_equal(expected_np_result, obtained_ms_res) 147 # backwards pass check 148 GradNet = Grad(Net(pads, "SYMMETRIC")) 149 dy_value = Tensor(np.ones(obtained_ms_res.shape), dtype=mindspore.int32) 150 dx_value_obtained = GradNet(test_arr_ms, dy_value)[0].asnumpy() 151 dx_value_expected = np.array([[[[16, 24, 24, 16, 16], 152 [16, 24, 24, 16, 16], 153 [16, 24, 24, 16, 16]], 154 [[16, 24, 24, 16, 16], 155 [16, 24, 24, 16, 16], 156 [16, 24, 24, 16, 16]], 157 [[8, 12, 12, 8, 8], 158 [8, 12, 12, 8, 8], 159 [8, 12, 12, 8, 8]]], 160 [[[8, 12, 12, 8, 8], 161 [8, 12, 12, 8, 8], 162 [8, 12, 12, 8, 8]], 163 [[8, 12, 12, 8, 8], 164 [8, 12, 12, 8, 8], 165 [8, 12, 12, 8, 8]], 166 [[4, 6, 6, 4, 4], 167 [4, 6, 6, 4, 4], 168 [4, 6, 6, 4, 4]]]], dtype=np.int32) 169 np.testing.assert_array_equal(dx_value_expected, dx_value_obtained) 170