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
#version 450
16

17
#if NCNN_fp16_storage
18
#extension GL_EXT_shader_16bit_storage: require
19
#endif
20
#if NCNN_fp16_arithmetic
21
#extension GL_EXT_shader_explicit_arithmetic_types_float16: require
22
#endif
23

24
layout (constant_id = 0) const int op_type = 0;
25
layout (constant_id = 1) const int with_scalar = 0;
26
layout (constant_id = 2) const float const_b = 0;
27

28
#define shape_constant_id_offset 3
29
layout (constant_id = shape_constant_id_offset + 0) const int adims = 0;
30
layout (constant_id = shape_constant_id_offset + 1) const int aw = 0;
31
layout (constant_id = shape_constant_id_offset + 2) const int ah = 0;
32
layout (constant_id = shape_constant_id_offset + 3) const int ac = 0;
33
layout (constant_id = shape_constant_id_offset + 4) const int acstep = 0;
34

35
layout (constant_id = shape_constant_id_offset + 5) const int bdims = 0;
36
layout (constant_id = shape_constant_id_offset + 6) const int bw = 0;
37
layout (constant_id = shape_constant_id_offset + 7) const int bh = 0;
38
layout (constant_id = shape_constant_id_offset + 8) const int bc = 0;
39
layout (constant_id = shape_constant_id_offset + 9) const int bcstep = 0;
40

41
layout (constant_id = shape_constant_id_offset + 10) const int outdims = 0;
42
layout (constant_id = shape_constant_id_offset + 11) const int outw = 0;
43
layout (constant_id = shape_constant_id_offset + 12) const int outh = 0;
44
layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
45
layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
46

47
#if NCNN_image_shader
48
layout (binding = 0) uniform unfp sampler3D a_blob_3d;
49
layout (binding = 1) uniform unfp sampler3D b_blob_3d;
50
layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
51
#else
52
layout (binding = 0) buffer a_blob { sfpvec4 a_blob_data[]; };
53
layout (binding = 1) readonly buffer b_blob { sfpvec4 b_blob_data[]; };
54
layout (binding = 2) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
55
#endif
56

57
layout (push_constant) uniform parameter
58
{
59
    int adims;
60
    int aw;
61
    int ah;
62
    int ac;
63
    int acstep;
64

65
    int bdims;
66
    int bw;
67
    int bh;
68
    int bc;
69
    int bcstep;
70

71
    int outdims;
72
    int outw;
73
    int outh;
74
    int outc;
75
    int outcstep;
76
} p;
77

78
void main()
79
{
80
    int gx = int(gl_GlobalInvocationID.x);
81
    int gy = int(gl_GlobalInvocationID.y);
82
    int gz = int(gl_GlobalInvocationID.z);
83

84
    if (gx >= psc(outw) || gy >= psc(outh) || gz >= psc(outc))
85
        return;
86

87
#if NCNN_image_shader
88
    afpvec4 v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
89
#else
90
    const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
91

92
    afpvec4 v1 = buffer_ld4(a_blob_data, gi);
93
#endif
94

95
    afpvec4 v2;
96

97
    if (with_scalar == 1)
98
    {
99
        // scalar
100
        v2 = afpvec4(afp(const_b));
101
    }
102
    else
103
    {
104
        // no broadcast
105
#if NCNN_image_shader
106
        v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
107
#else
108
        v2 = buffer_ld4(b_blob_data, gi);
109
#endif
110
    }
111

112
    afpvec4 res;
113

114
    if (op_type == 0) res = v1 + v2;
115
    if (op_type == 1) res = v1 - v2;
116
    if (op_type == 2) res = v1 * v2;
117
    if (op_type == 3) res = v1 / v2;
118
    if (op_type == 4) res = max(v1, v2);
119
    if (op_type == 5) res = min(v1, v2);
120
    if (op_type == 6) res = pow(v1, v2);
121
    if (op_type == 7) res = v2 - v1;
122
    if (op_type == 8) res = v2 / v1;
123
    if (op_type == 9) res = pow(v2, v1);
124
#if NCNN_moltenvk
125
    if (op_type == 10) res = afpvec4(atan(vec4(v1), vec4(v2)));
126
    if (op_type == 11) res = afpvec4(atan(vec4(v2), vec4(v1)));
127
#else
128
    if (op_type == 10) res = atan(v1, v2);
129
    if (op_type == 11) res = atan(v2, v1);
130
#endif
131

132
#if NCNN_image_shader
133
    image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
134
#else
135
    buffer_st4(top_blob_data, gi, res);
136
#endif
137
}
138