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 FLT_MAX 3.402823466e+38
26

27
#define shape_constant_id_offset 0
28
layout (constant_id = shape_constant_id_offset + 0) const int w = 0;
29
layout (constant_id = shape_constant_id_offset + 1) const int c = 0;
30
layout (constant_id = shape_constant_id_offset + 2) const int cstep = 0;
31

32
layout (constant_id = shape_constant_id_offset + 3) const int outw = 0;
33
layout (constant_id = shape_constant_id_offset + 4) const int outcstep = 0;
34

35
#if NCNN_image_shader
36
layout (binding = 0) uniform unfp sampler3D bottom_blob;
37
layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
38
#else
39
layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
40
layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
41
#endif
42

43
layout (push_constant) uniform parameter
44
{
45
    int w;
46
    int c;
47
    int cstep;
48

49
    int outw;
50
    int outcstep;
51
} p;
52

53
void main()
54
{
55
    int gx = int(gl_GlobalInvocationID.x);
56
    int gy = int(gl_GlobalInvocationID.y);
57
    int gz = int(gl_GlobalInvocationID.z);
58

59
    if (gx >= psc(outw) || gy >= 1 || gz >= psc(c))
60
        return;
61

62
    const int size_1 = psc(w) - 1;
63

64
    const int v_offset = gz * psc(cstep);
65

66
    afpvec8 res = afpvec8(afpvec4(-FLT_MAX), afpvec4(-FLT_MAX));
67

68
    for (int ii = 0; ii < 8; ii++)
69
    {
70
        int i = min(gx + ii * psc(outw), size_1);
71

72
#if NCNN_image_shader
73
        afpvec8 v = image3d_ld8(bottom_blob, ivec3(i, 0, gz));
74
#else
75
        afpvec8 v = buffer_ld8(bottom_blob_data, v_offset + i);
76
#endif
77
        res[0] = max(res[0], v[0]);
78
        res[1] = max(res[1], v[1]);
79
    }
80

81
#if NCNN_image_shader
82
    image3d_st8(top_blob, ivec3(gx, 0, gz), res);
83
#else
84
    buffer_st8(top_blob_data, gz * psc(outcstep) + gx, res);
85
#endif
86
}
87