Hugging Face's logo

camenduru
/
ncnn

Model card Files
xet
Community
ncnn
File size: 5,091 Bytes
be903e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
 
// Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2020 THL A29 Limited, a Tencent company. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.

#version 450

#if NCNN_fp16_storage
#extension GL_EXT_shader_16bit_storage: require
struct sfpvec8 { f16vec4 abcd; f16vec4 efgh; };
#endif
#if NCNN_fp16_arithmetic
#extension GL_EXT_shader_explicit_arithmetic_types_float16: require
#endif

layout (constant_id = 0) const int axis = 0;

#define shape_constant_id_offset 1
layout (constant_id = shape_constant_id_offset + 0) const int dims = 0;
layout (constant_id = shape_constant_id_offset + 1) const int w = 0;
layout (constant_id = shape_constant_id_offset + 2) const int h = 0;
layout (constant_id = shape_constant_id_offset + 3) const int d = 0;
layout (constant_id = shape_constant_id_offset + 4) const int c = 0;
layout (constant_id = shape_constant_id_offset + 5) const int cstep = 0;

layout (constant_id = shape_constant_id_offset + 6) const int outdims = 0;
layout (constant_id = shape_constant_id_offset + 7) const int outw = 0;
layout (constant_id = shape_constant_id_offset + 8) const int outh = 0;
layout (constant_id = shape_constant_id_offset + 9) const int outd = 0;
layout (constant_id = shape_constant_id_offset + 10) const int outc = 0;
layout (constant_id = shape_constant_id_offset + 11) const int outcstep = 0;

#if NCNN_image_shader
layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
#else
layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
#endif

layout (push_constant) uniform parameter
{
    int dims;
    int w;
    int h;
    int d;
    int c;
    int cstep;

    int outdims;
    int outw;
    int outh;
    int outd;
    int outc;
    int outcstep;

    int offset;
} p;

void main()
{
    int gx = int(gl_GlobalInvocationID.x);
    int gy = int(gl_GlobalInvocationID.y);
    int gz = int(gl_GlobalInvocationID.z);

    if (gx >= psc(w) || gy >= psc(h) * psc(d) || gz >= psc(c))
        return;

    int positive_axis = axis < 0 ? psc(dims) + axis : axis;

    ivec3 gxyz;

    if (psc(dims) == 4)
    {
        int yd = gy / psc(h);
        int yh = gy % psc(h);

        ivec4 gxydz = ivec4(gx, yh, yd, gz);
        gxydz[psc(dims) - 1] *= 2;
        gxydz[psc(dims) - 1 - positive_axis] += p.offset;

        gxyz = ivec3(gxydz.r, gxydz.g + gxydz.b * psc(outh), gxydz.a);
    }
    else
    {
        gxyz = ivec3(gx, gy, gz);
        gxyz[psc(dims) - 1] *= 2;
        gxyz[psc(dims) - 1 - positive_axis] += p.offset;
    }

#if NCNN_image_shader
    afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));

    if (psc(dims) == 1)
    {
        image3d_st4(top_blob_3d, ivec3(gxyz.x + 0, 0, 0), v[0]);
        image3d_st4(top_blob_3d, ivec3(gxyz.x + 1, 0, 0), v[1]);

    }
    else if (psc(dims) == 2)
    {
        if (positive_axis == 0)
        {
            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y + 0, 0), v[0]);
            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y + 1, 0), v[1]);
        }
        if (positive_axis == 1)
        {
            image3d_st4(top_blob_3d, ivec3(gxyz.x + 0, gxyz.y, 0), v[0]);
            image3d_st4(top_blob_3d, ivec3(gxyz.x + 1, gxyz.y, 0), v[1]);
        }
    }
    else // if (psc(dims) == 3)
    {
        if (positive_axis == 0)
        {
            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 0), v[0]);
            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 1), v[1]);
        }
        if (positive_axis == 1)
        {
            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y + 0, gxyz.z), v[0]);
            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y + 1, gxyz.z), v[1]);
        }
        if (positive_axis == 2)
        {
            image3d_st4(top_blob_3d, ivec3(gxyz.x + 0, gxyz.y, gxyz.z), v[0]);
            image3d_st4(top_blob_3d, ivec3(gxyz.x + 1, gxyz.y, gxyz.z), v[1]);
        }
    }
#else
    const int gi = gz * psc(cstep) + gy * psc(w) + gx;

    int v_offset_0 = gxyz.z * psc(outcstep) + gxyz.y * psc(outw) + gxyz.x;

    ivec4 gxydz4;

    if (psc(dims) == 4)
    {
        gxydz4 = ivec4(1, psc(outw), psc(outw) * psc(outh), psc(outcstep));
    }
    else
    {
        gxydz4 = ivec4(1, psc(outw), psc(outcstep), 0);
    }

    ivec2 v_offset = v_offset_0 + ivec2(0, 1) * gxydz4[psc(dims) - 1 - positive_axis];

    buffer_cp8to4(top_blob_data, v_offset, bottom_blob_data, gi);
#endif
}