ncnn

1
// Tencent is pleased to support the open source community by making ncnn available.
2
//
3
// Copyright (C) 2020 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
#version 450
16

17
#if NCNN_fp16_storage
18
#extension GL_EXT_shader_16bit_storage: require
19
struct sfpvec8 { f16vec4 abcd; f16vec4 efgh; };
20
#endif
21
#if NCNN_fp16_arithmetic
22
#extension GL_EXT_shader_explicit_arithmetic_types_float16: require
23
#endif
24

25
layout (constant_id = 0) const int region_type = 0;
26
layout (constant_id = 1) const int local_size = 0;
27
layout (constant_id = 2) const float alpha = 0;
28
layout (constant_id = 3) const float beta = 0;
29
layout (constant_id = 4) const float bias_constant = 0;
30

31
#define shape_constant_id_offset 5
32
layout (constant_id = shape_constant_id_offset + 0) const int dims = 0;
33
layout (constant_id = shape_constant_id_offset + 1) const int w = 0;
34
layout (constant_id = shape_constant_id_offset + 2) const int h = 0;
35
layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
36
layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
37

38
layout (constant_id = shape_constant_id_offset + 5) const int outdims = 0;
39
layout (constant_id = shape_constant_id_offset + 6) const int outw = 0;
40
layout (constant_id = shape_constant_id_offset + 7) const int outh = 0;
41
layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
42
layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
43

44
#if NCNN_image_shader
45
layout (binding = 0) uniform highp sampler3D square_workspace;
46
layout (binding = 1) uniform unfp sampler3D bottom_blob;
47
layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob;
48
#else
49
layout (binding = 0) readonly buffer square_workspace { mat2x4 square_workspace_data[]; };
50
layout (binding = 1) buffer bottom_top_blob { sfpvec8 bottom_top_blob_data[]; };
51
#endif
52

53
layout (push_constant) uniform parameter
54
{
55
    int dims;
56
    int w;
57
    int h;
58
    int c;
59
    int cstep;
60

61
    int outdims;
62
    int outw;
63
    int outh;
64
    int outc;
65
    int outcstep;
66
} p;
67

68
void main()
69
{
70
    int gx = int(gl_GlobalInvocationID.x);
71
    int gy = int(gl_GlobalInvocationID.y);
72
    int gz = int(gl_GlobalInvocationID.z);
73

74
    if (gx >= psc(outw) || gy >= psc(outh) || gz >= psc(outc))
75
        return;
76

77
    // support region_type == 1 only
78

79
    mat2x4 sum = mat2x4(0.f);
80

81
#if NCNN_image_shader
82
    for (int y = 0; y < local_size; y++)
83
    {
84
        for (int x = 0; x < local_size; x++)
85
        {
86
            sum[0] += texelFetch(square_workspace, ivec3((gx + x) * 2, gy + y, gz), 0);
87
            sum[1] += texelFetch(square_workspace, ivec3((gx + x) * 2 + 1, gy + y, gz), 0);
88
        }
89
    }
90
#else
91
    int v_offset = gz * psc(cstep) + gy * psc(w) + gx;
92

93
    for (int y = 0; y < local_size; y++)
94
    {
95
        for (int x = 0; x < local_size; x++)
96
        {
97
            sum += square_workspace_data[v_offset + x];
98
        }
99

100
        v_offset += psc(w);
101
    }
102
#endif
103

104
    const float alpha_div_size = alpha / (local_size * local_size);
105
    afpvec8 scale;
106
    scale[0] = afpvec4(pow(bias_constant + alpha_div_size * sum[0], vec4(-beta)));
107
    scale[1] = afpvec4(pow(bias_constant + alpha_div_size * sum[1], vec4(-beta)));
108

109
#if NCNN_image_shader
110
    afpvec8 v = image3d_ld8(bottom_blob, ivec3(gx, gy, gz));
111
#else
112
    int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
113

114
    afpvec8 v = buffer_ld8(bottom_top_blob_data, gi);
115
#endif
116

117
    v[0] *= scale[0];
118
    v[1] *= scale[1];
119

120
#if NCNN_image_shader
121
    image3d_st8(top_blob, ivec3(gx, gy, gz), v);
122
#else
123
    buffer_st8(bottom_top_blob_data, gi, v);
124
#endif
125
}
126