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 pad_head = 0;
27
layout (constant_id = 2) const int pad_tail = 0;
28

29
#define shape_constant_id_offset 3
30
layout (constant_id = shape_constant_id_offset + 0) const int dims = 0;
31
layout (constant_id = shape_constant_id_offset + 1) const int w = 0;
32
layout (constant_id = shape_constant_id_offset + 2) const int h = 0;
33
layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
34
layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
35

36
layout (constant_id = shape_constant_id_offset + 5) const int outdims = 0;
37
layout (constant_id = shape_constant_id_offset + 6) const int outw = 0;
38
layout (constant_id = shape_constant_id_offset + 7) const int outh = 0;
39
layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
40
layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
41

42
#if NCNN_image_shader
43
layout (binding = 0) uniform unfp sampler3D bottom_blob;
44
layout (binding = 1, rgba32f) writeonly uniform highp image3D square_workspace;
45
#else
46
layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
47
layout (binding = 1) writeonly buffer square_workspace { mat2x4 square_workspace_data[]; };
48
#endif
49

50
layout (push_constant) uniform parameter
51
{
52
    int dims;
53
    int w;
54
    int h;
55
    int c;
56
    int cstep;
57

58
    int outdims;
59
    int outw;
60
    int outh;
61
    int outc;
62
    int outcstep;
63
} p;
64

65
void main()
66
{
67
    int gx = int(gl_GlobalInvocationID.x);
68
    int gy = int(gl_GlobalInvocationID.y);
69
    int gz = int(gl_GlobalInvocationID.z);
70

71
    if (gx >= psc(outw) || gy >= psc(outh) || gz >= psc(outc))
72
        return;
73

74
    // support region_type == 1 only
75

76
    afpvec8 res;
77

78
    int x = gx - pad_head;
79
    int y = gy - pad_head;
80

81
    if (x >= 0 && x < psc(w) && y >= 0 && y < psc(h))
82
    {
83
#if NCNN_image_shader
84
        afpvec8 v = image3d_ld8(bottom_blob, ivec3(x, y, gz));
85
#else
86
        int v_offset = gz * psc(cstep) + y * psc(w) + x;
87
        afpvec8 v = buffer_ld8(bottom_blob_data, v_offset);
88
#endif
89
        res[0] = v[0] * v[0];
90
        res[1] = v[1] * v[1];
91
    }
92
    else
93
    {
94
        res = afpvec8(afpvec4(0.f), afpvec4(0.f));
95
    }
96

97
#if NCNN_image_shader
98
    imageStore(square_workspace, ivec3(gx * 2, gy, gz), res[0]);
99
    imageStore(square_workspace, ivec3(gx * 2 + 1, gy, gz), res[1]);
100
#else
101
    const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
102

103
    square_workspace_data[gi] = mat2x4(res);
104
#endif
105
}
106