ncnn

1
// Tencent is pleased to support the open source community by making ncnn available.
2
//
3
// Copyright (C) 2019 THL A29 Limited, a Tencent company. All rights reserved.
4
//
5
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
6
// in compliance with the License. You may obtain a copy of the License at
7
//
8
// https://opensource.org/licenses/BSD-3-Clause
9
//
10
// Unless required by applicable law or agreed to in writing, software distributed
11
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
12
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
13
// specific language governing permissions and limitations under the License.
14

15
#include "testutil.h"
16

17
static int test_convolution(int w, int h, int c, int outch, int kernel, int dilation, int stride, int pad, int bias)
18
{
19
    ncnn::Mat a = RandomMat(w, h, c);
20

21
    ncnn::ParamDict pd;
22
    pd.set(0, outch);
23
    pd.set(1, kernel);
24
    pd.set(2, dilation);
25
    pd.set(3, stride);
26
    pd.set(4, pad);
27
    pd.set(5, bias);
28
    pd.set(6, outch * c * kernel * kernel);
29

30
    int activation_type = RAND() % 7; // 0 1 2 3 4 5 6
31
    ncnn::Mat activation_params(2);
32
    activation_params[0] = (activation_type == 6) ? RandomFloat(0, 1) : RandomFloat(-1, 0); // alpha
33
    activation_params[1] = RandomFloat(0, 1);                                               // beta
34
    pd.set(9, activation_type);
35
    pd.set(10, activation_params);
36

37
    std::vector<ncnn::Mat> weights(bias ? 2 : 1);
38
    weights[0] = RandomMat(outch * c * kernel * kernel);
39
    if (bias)
40
        weights[1] = RandomMat(outch);
41

42
    float epsilon = 0.001;
43

44
    int ret = test_layer("Convolution", pd, weights, a, epsilon);
45
    if (ret != 0)
46
    {
47
        fprintf(stderr, "test_convolution failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d act=%d actparams=[%f,%f]\n", w, h, c, outch, kernel, dilation, stride, pad, bias, activation_type, activation_params[0], activation_params[1]);
48
        return ret;
49
    }
50

51
    {
52
        ncnn::Option opt;
53
        opt.num_threads = 1;
54
        opt.use_packing_layout = true;
55
        opt.use_fp16_packed = false;
56
        opt.use_fp16_storage = false;
57
        opt.use_fp16_arithmetic = false;
58
        opt.use_bf16_storage = false;
59
        opt.use_shader_pack8 = false;
60
        opt.use_image_storage = false;
61
        opt.use_sgemm_convolution = false;
62
        opt.use_winograd_convolution = false;
63

64
        ret = test_layer_opt("Convolution", pd, weights, opt, a, epsilon);
65
        if (ret != 0)
66
        {
67
            fprintf(stderr, "test_convolution failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d act=%d actparams=[%f,%f]\n", w, h, c, outch, kernel, dilation, stride, pad, bias, activation_type, activation_params[0], activation_params[1]);
68
            return ret;
69
        }
70
    }
71

72
    {
73
        ncnn::Option opt;
74
        opt.num_threads = 1;
75
        opt.use_packing_layout = true;
76
        opt.use_fp16_packed = true;
77
        opt.use_fp16_storage = true;
78
        opt.use_fp16_arithmetic = true;
79
        opt.use_bf16_storage = true;
80
        opt.use_shader_pack8 = true;
81
        opt.use_image_storage = true;
82
        opt.use_sgemm_convolution = false;
83
        opt.use_winograd_convolution = false;
84

85
        ret = test_layer_opt("Convolution", pd, weights, opt, a, epsilon);
86
        if (ret != 0)
87
        {
88
            fprintf(stderr, "test_convolution failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d act=%d actparams=[%f,%f]\n", w, h, c, outch, kernel, dilation, stride, pad, bias, activation_type, activation_params[0], activation_params[1]);
89
            return ret;
90
        }
91
    }
92

93
#if __aarch64__
94
    {
95
        ncnn::Option opt;
96
        opt.num_threads = 1;
97
        opt.use_a53_a55_optimized_kernel = true;
98

99
        ret = test_layer_opt("Convolution", pd, weights, opt, a, epsilon);
100
        if (ret != 0)
101
        {
102
            fprintf(stderr, "test_convolution failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d act=%d actparams=[%f,%f]\n", w, h, c, outch, kernel, dilation, stride, pad, bias, activation_type, activation_params[0], activation_params[1]);
103
            return ret;
104
        }
105
    }
106
#endif // __aarch64__
107

108
    return ret;
109
}
110

111
static int test_convolution_0()
112
{
113
    static const int kdsp[16][4] = {
114
        {1, 1, 1, 0},
115
        {1, 1, 2, 0},
116
        {2, 1, 1, 1},
117
        {2, 1, 2, -233},
118
        {3, 1, 1, 1},
119
        {3, 1, 2, 1},
120
        {3, 2, 1, 1},
121
        {4, 1, 1, 2},
122
        {4, 1, 2, -233},
123
        {4, 2, 1, -234},
124
        {5, 1, 1, -234},
125
        {5, 1, 2, 2},
126
        {5, 2, 2, 2},
127
        {7, 1, 1, 3},
128
        {7, 1, 2, 3},
129
        {7, 2, 1, -233},
130
    };
131

132
    for (int i = 12; i < 16; i++)
133
    {
134
        const int k = kdsp[i][0];
135
        const int d = kdsp[i][1];
136
        const int s = kdsp[i][2];
137
        const int p = kdsp[i][3];
138

139
        int ret = 0
140
                  || test_convolution(9, 7, 1, 1, k, d, s, p, 1)
141
                  || test_convolution(9, 7, 4, 13, k, d, s, p, 0)
142
                  || test_convolution(9, 7, 13, 4, k, d, s, p, 1)
143
                  || test_convolution(9, 7, 12, 12, k, d, s, p, 0)
144
                  || test_convolution(9, 7, 8, 12, k, d, s, p, 1)
145
                  || test_convolution(9, 7, 8, 13, k, d, s, p, 0)
146
                  || test_convolution(9, 7, 13, 8, k, d, s, p, 1)
147
                  || test_convolution(9, 7, 12, 16, k, d, s, p, 0)
148
                  || test_convolution(9, 7, 15, 15, k, d, s, p, 0)
149
                  || test_convolution(9, 7, 16, 16, k, d, s, p, 0)
150
                  || test_convolution(18, 17, 1, 1, k, d, s, p, 1)
151
                  || test_convolution(18, 17, 4, 13, k, d, s, p, 0)
152
                  || test_convolution(18, 17, 13, 4, k, d, s, p, 1)
153
                  || test_convolution(18, 17, 12, 12, k, d, s, p, 0)
154
                  || test_convolution(18, 17, 8, 12, k, d, s, p, 1)
155
                  || test_convolution(18, 17, 8, 13, k, d, s, p, 0)
156
                  || test_convolution(18, 17, 13, 8, k, d, s, p, 1)
157
                  || test_convolution(18, 17, 12, 16, k, d, s, p, 0)
158
                  || test_convolution(18, 17, 15, 15, k, d, s, p, 0)
159
                  || test_convolution(18, 17, 16, 16, k, d, s, p, 0)
160
                  || test_convolution(25, 33, 1, 1, k, d, s, p, 1)
161
                  || test_convolution(25, 33, 4, 13, k, d, s, p, 0)
162
                  || test_convolution(25, 33, 13, 4, k, d, s, p, 1)
163
                  || test_convolution(25, 33, 12, 12, k, d, s, p, 0)
164
                  || test_convolution(25, 33, 8, 12, k, d, s, p, 1)
165
                  || test_convolution(25, 33, 8, 13, k, d, s, p, 0)
166
                  || test_convolution(25, 33, 13, 8, k, d, s, p, 1)
167
                  || test_convolution(25, 33, 12, 16, k, d, s, p, 0)
168
                  || test_convolution(25, 33, 15, 15, k, d, s, p, 0)
169
                  || test_convolution(25, 33, 16, 16, k, d, s, p, 0);
170

171
        if (ret != 0)
172
            return -1;
173
    }
174

175
    return 0;
176
}
177

178
int main()
179
{
180
    SRAND(7767517);
181

182
    return test_convolution_0();
183
}
184