ncnn

1
// Tencent is pleased to support the open source community by making ncnn available.
2
//
3
// Copyright (C) 2023 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
#define shape_constant_id_offset 0
26
layout (constant_id = shape_constant_id_offset + 0) const int w = 0;
27
layout (constant_id = shape_constant_id_offset + 1) const int c = 0;
28
layout (constant_id = shape_constant_id_offset + 2) const int cstep = 0;
29

30
#if NCNN_image_shader
31
layout (binding = 0) uniform highp sampler3D bottom_blob;
32
layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
33
#else
34
layout (binding = 0) readonly buffer bottom_blob { mat2x4 bottom_blob_data[]; };
35
layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
36
#endif
37

38
layout (push_constant) uniform parameter
39
{
40
    int w;
41
    int c;
42
    int cstep;
43
    int size;
44
} p;
45

46
void main()
47
{
48
    int gx = int(gl_GlobalInvocationID.x);
49
    int gy = int(gl_GlobalInvocationID.y);
50
    int gz = int(gl_GlobalInvocationID.z);
51

52
    if (gx >= 1 || gy >= 1 || gz >= psc(c))
53
        return;
54

55
    const int v_offset = gz * psc(cstep);
56

57
    mat2x4 sum = mat2x4(0.f);
58

59
    for (int i = 0; i < psc(w); i++)
60
    {
61
#if NCNN_image_shader
62
        mat2x4 v;
63
        v[0] = texelFetch(bottom_blob, ivec3(i * 2, 0, gz), 0);
64
        v[1] = texelFetch(bottom_blob, ivec3(i * 2 + 1, 0, gz), 0);
65
#else
66
        mat2x4 v = bottom_blob_data[v_offset + i];
67
#endif
68
        sum[0] += v[0];
69
        sum[1] += v[1];
70
    }
71

72
    afpvec8 res;
73
    res[0] = afpvec4(sum[0] / p.size);
74
    res[1] = afpvec4(sum[1] / p.size);
75

76
#if NCNN_image_shader
77
    image3d_st8(top_blob, ivec3(gz, 0, 0), res);
78
#else
79
    buffer_st8(top_blob_data, gz, res);
80
#endif
81
}
82