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 bias_term = 0;
26

27
#define shape_constant_id_offset 1
28
layout (constant_id = shape_constant_id_offset + 0) const int dims = 0;
29
layout (constant_id = shape_constant_id_offset + 1) const int w = 0;
30
layout (constant_id = shape_constant_id_offset + 2) const int h = 0;
31
layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
32
layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
33

34
#if NCNN_image_shader
35
layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
36
layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
37
layout (binding = 2) uniform unfp sampler3D scale_blob;
38
layout (binding = 3) uniform unfp sampler3D bias_blob;
39
#else
40
layout (binding = 0) buffer bottom_top_blob { sfpvec8 bottom_top_blob_data[]; };
41
layout (binding = 1) readonly buffer scale_blob { sfpvec8 scale_blob_data[]; };
42
layout (binding = 2) readonly buffer bias_blob { sfpvec8 bias_blob_data[]; };
43
#endif
44

45
layout (push_constant) uniform parameter
46
{
47
    int dims;
48
    int w;
49
    int h;
50
    int c;
51
    int cstep;
52
} p;
53

54
void main()
55
{
56
    int gx = int(gl_GlobalInvocationID.x);
57
    int gy = int(gl_GlobalInvocationID.y);
58
    int gz = int(gl_GlobalInvocationID.z);
59

60
    if (gx >= psc(w) || gy >= psc(h) || gz >= psc(c))
61
        return;
62

63
#if NCNN_image_shader
64
    afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
65
#else
66
    const int gi = gz * psc(cstep) + gy * psc(w) + gx;
67

68
    afpvec8 v = buffer_ld8(bottom_top_blob_data, gi);
69
#endif
70

71
    const ivec3 gxyz = ivec3(gx, gy, gz);
72
    const int abi = gxyz[psc(dims) - 1];
73

74
#if NCNN_image_shader
75
    afpvec8 scale = image3d_ld8(scale_blob, ivec3(abi, 0, 0));
76
#else
77
    afpvec8 scale = buffer_ld8(scale_blob_data, abi);
78
#endif
79

80
    if (bias_term == 1)
81
    {
82
#if NCNN_image_shader
83
        afpvec8 bias = image3d_ld8(bias_blob, ivec3(abi, 0, 0));
84
#else
85
        afpvec8 bias = buffer_ld8(bias_blob_data, abi);
86
#endif
87
        v[0] = scale[0] * v[0] + bias[0];
88
        v[1] = scale[1] * v[1] + bias[1];
89
    }
90
    else
91
    {
92
        v[0] = scale[0] * v[0];
93
        v[1] = scale[1] * v[1];
94
    }
95

96
#if NCNN_image_shader
97
    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
98
#else
99
    buffer_st8(bottom_top_blob_data, gi, v);
100
#endif
101
}
102