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
#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 upscale_factor = 0;
25
layout (constant_id = 1) const int mode = 0;
26

27
#define shape_constant_id_offset 2
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
layout (constant_id = shape_constant_id_offset + 5) const int outdims = 0;
35
layout (constant_id = shape_constant_id_offset + 6) const int outw = 0;
36
layout (constant_id = shape_constant_id_offset + 7) const int outh = 0;
37
layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
38
layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
39

40
#if NCNN_image_shader
41
layout (binding = 0) uniform unfp sampler3D bottom_blob;
42
layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
43
#else
44
#if NCNN_fp16_packed
45
layout (binding = 0) readonly buffer bottom_blob { sfpvec2 bottom_blob_data[]; };
46
#else
47
layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
48
#endif
49
layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
50
#endif
51

52
layout (push_constant) uniform parameter
53
{
54
    int dims;
55
    int w;
56
    int h;
57
    int c;
58
    int cstep;
59

60
    int outdims;
61
    int outw;
62
    int outh;
63
    int outc;
64
    int outcstep;
65
} p;
66

67
void main()
68
{
69
    int gx = int(gl_GlobalInvocationID.x);
70
    int gy = int(gl_GlobalInvocationID.y);
71
    int gz = int(gl_GlobalInvocationID.z);
72

73
    if (gx >= psc(outw) || gy >= psc(outh) || gz >= psc(outc))
74
        return;
75

76
    ivec4 gz4 = gz * 4 + ivec4(0, 1, 2, 3);
77

78
    int x = gx / upscale_factor;
79
    int y = gy / upscale_factor;
80
    ivec4 z4;
81
    if (mode == 0)
82
    {
83
        z4 = gz4 * upscale_factor * upscale_factor + (gy % upscale_factor) * upscale_factor + gx % upscale_factor;
84
    }
85
    else // if (mode == 1)
86
    {
87
        z4 = ((gy % upscale_factor) * upscale_factor + gx % upscale_factor) * psc(outc) * 4 + gz4;
88
    }
89

90
#if NCNN_image_shader
91
    afpvec4 vr = image3d_ld4(bottom_blob, ivec3(x, y, z4.r / 4));
92
    afpvec4 vg = image3d_ld4(bottom_blob, ivec3(x, y, z4.g / 4));
93
    afpvec4 vb = image3d_ld4(bottom_blob, ivec3(x, y, z4.b / 4));
94
    afpvec4 va = image3d_ld4(bottom_blob, ivec3(x, y, z4.a / 4));
95

96
    ivec4 lane4 = z4 % 4;
97

98
    afpvec4 v = afpvec4(vr[lane4.r], vg[lane4.g], vb[lane4.b], va[lane4.a]);
99

100
    image3d_st4(top_blob, ivec3(gx, gy, gz), v);
101
#else
102
    int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
103

104
#if NCNN_fp16_packed
105
    ivec4 v_offset = ((z4 / 4) * psc(cstep) + y * psc(w) + x) * 2 + (z4 % 4) / 2;
106
    ivec4 lane2 = z4 % 2;
107

108
    afpvec2 vr = buffer_ld2(bottom_blob_data, v_offset.r);
109
    afpvec2 vg = buffer_ld2(bottom_blob_data, v_offset.g);
110
    afpvec2 vb = buffer_ld2(bottom_blob_data, v_offset.b);
111
    afpvec2 va = buffer_ld2(bottom_blob_data, v_offset.a);
112

113
    afpvec4 v = afpvec4(vr[lane2.r], vg[lane2.g], vb[lane2.b], va[lane2.a]);
114

115
    buffer_st4(top_blob_data, gi, v);
116
#else
117
    ivec4 v_offset = ((z4 / 4) * psc(cstep) + y * psc(w) + x) * 4 + z4 % 4;
118

119
    buffer_cp1to4(top_blob_data, gi, bottom_blob_data, v_offset);
120
#endif
121
#endif
122
}
123