ncnn

1
// Tencent is pleased to support the open source community by making ncnn available.
2
//
3
// Copyright (C) 2022 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
#extension GL_GOOGLE_include_directive: enable
25
#include "vulkan_activation.comp"
26

27
#define LOCAL_MEMORY_UNROLL_INCH 8
28

29
layout (constant_id = 0) const int bias_term = 0;
30
layout (constant_id = 1) const int activation_type = 0;
31
layout (constant_id = 2) const float activation_param_0 = 0;
32
layout (constant_id = 3) const float activation_param_1 = 0;
33

34
#define shape_constant_id_offset 4
35
layout (constant_id = shape_constant_id_offset + 0) const int w = 0;
36
layout (constant_id = shape_constant_id_offset + 1) const int h = 0;
37
layout (constant_id = shape_constant_id_offset + 2) const int c = 0;
38
layout (constant_id = shape_constant_id_offset + 3) const int cstep = 0;
39

40
layout (constant_id = shape_constant_id_offset + 4) const int outw = 0;
41
layout (constant_id = shape_constant_id_offset + 5) const int outh = 0;
42
layout (constant_id = shape_constant_id_offset + 6) const int outc = 0;
43
layout (constant_id = shape_constant_id_offset + 7) const int outcstep = 0;
44

45
#if NCNN_image_shader
46
layout (binding = 0) uniform unfp sampler3D bottom_blob;
47
layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
48
layout (binding = 2) uniform unfp sampler3D weight_blob;
49
layout (binding = 3) uniform unfp sampler3D bias_blob;
50
#else
51
layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
52
layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
53
layout (binding = 2) readonly buffer weight_blob { sfpvec4 weight_data[]; };
54
layout (binding = 3) readonly buffer bias_blob { sfpvec4 bias_data[]; };
55
#endif
56

57
layout (push_constant) uniform parameter
58
{
59
    int w;
60
    int h;
61
    int c;
62
    int cstep;
63

64
    int outw;
65
    int outh;
66
    int outc;
67
    int outcstep;
68
} p;
69

70
#if NCNN_shader_local_memory
71
shared lfp tmp_v[8][LOCAL_MEMORY_UNROLL_INCH][4];
72
shared lfpvec4 tmp_k[8][LOCAL_MEMORY_UNROLL_INCH];
73
#endif
74

75
void main()
76
{
77
    int gx = int(gl_GlobalInvocationID.x) * 4;
78
    int gy = int(gl_GlobalInvocationID.y);
79

80
#if NCNN_image_shader
81
    if (gx >= psc(outw) * psc(outh) || gy >= psc(outc))
82
        return;
83
#else
84
#if !NCNN_shader_local_memory
85
    if (gx >= psc(outcstep) || gy >= psc(outc))
86
        return;
87
#endif
88
#endif
89

90
    afpvec4 sum0;
91
    afpvec4 sum1;
92
    afpvec4 sum2;
93
    afpvec4 sum3;
94

95
    if (bias_term == 1)
96
    {
97
#if NCNN_image_shader
98
        afpvec4 b = image3d_ld4(bias_blob, ivec3(gy, 0, 0));
99
#else
100
        afpvec4 b = buffer_ld4(bias_data, gy);
101
#endif
102
        sum0 = b;
103
        sum1 = b;
104
        sum2 = b;
105
        sum3 = b;
106
    }
107
    else
108
    {
109
        sum0 = afpvec4(0.f);
110
        sum1 = afpvec4(0.f);
111
        sum2 = afpvec4(0.f);
112
        sum3 = afpvec4(0.f);
113
    }
114

115
#if NCNN_image_shader
116
    ivec4 gx4 = gx + ivec4(0, 1, 2, 3);
117

118
    ivec4 sy4 = gx4 / psc(w);
119
    ivec4 sx4 = gx4 % psc(w);
120

121
    for (int z = 0; z < psc(c); z++)
122
    {
123
        afp v0 = image3d_ld1(bottom_blob, ivec3(sx4.r, sy4.r, z));
124
        afp v1 = image3d_ld1(bottom_blob, ivec3(sx4.g, sy4.g, z));
125
        afp v2 = image3d_ld1(bottom_blob, ivec3(sx4.b, sy4.b, z));
126
        afp v3 = image3d_ld1(bottom_blob, ivec3(sx4.a, sy4.a, z));
127

128
        afpvec4 k = image3d_ld4(weight_blob, ivec3(0, z, gy));
129

130
        sum0 += v0 * k;
131
        sum1 += v1 * k;
132
        sum2 += v2 * k;
133
        sum3 += v3 * k;
134
    }
135
#else
136
    int w_offset = gy * psc(c);
137
    int v_offset = gx;
138

139
#if NCNN_shader_local_memory
140
    const int lx = int(gl_LocalInvocationID.x);
141
    const int ly = int(gl_LocalInvocationID.y);
142

143
    int z = 0;
144
    for (; z + (LOCAL_MEMORY_UNROLL_INCH - 1) < psc(c); z += LOCAL_MEMORY_UNROLL_INCH)
145
    {
146
        if (ly < 4)
147
        {
148
            for (int z4 = 0; z4 < LOCAL_MEMORY_UNROLL_INCH; z4++)
149
            {
150
                tmp_v[lx][z4][ly] = sfp2lfp(bottom_blob_data[v_offset + z4 * psc(cstep) + ly]);
151
            }
152
        }
153

154
        if (lx == 0)
155
        {
156
            for (int z4 = 0; z4 < LOCAL_MEMORY_UNROLL_INCH; z4++)
157
            {
158
                tmp_k[ly][z4] = sfp2lfpvec4(weight_data[w_offset + z4]);
159
            }
160
        }
161

162
        barrier();
163

164
        for (int z4 = 0; z4 < LOCAL_MEMORY_UNROLL_INCH; z4++)
165
        {
166
            afp v0 = lfp2afp(tmp_v[lx][z4][0]);
167
            afp v1 = lfp2afp(tmp_v[lx][z4][1]);
168
            afp v2 = lfp2afp(tmp_v[lx][z4][2]);
169
            afp v3 = lfp2afp(tmp_v[lx][z4][3]);
170

171
            afpvec4 k = lfp2afpvec4(tmp_k[ly][z4]);
172

173
            sum0 += v0 * k;
174
            sum1 += v1 * k;
175
            sum2 += v2 * k;
176
            sum3 += v3 * k;
177
        }
178

179
        v_offset += LOCAL_MEMORY_UNROLL_INCH * psc(cstep);
180
        w_offset += LOCAL_MEMORY_UNROLL_INCH;
181

182
        barrier();
183
    }
184

185
    if (z < psc(c))
186
    {
187
        const int remain = psc(c) - z;
188

189
        if (ly < 4)
190
        {
191
            for (int z4 = 0; z4 < remain; z4++)
192
            {
193
                tmp_v[lx][z4][ly] = sfp2lfp(bottom_blob_data[v_offset + z4 * psc(cstep) + ly]);
194
            }
195
        }
196

197
        if (lx == 0)
198
        {
199
            for (int z4 = 0; z4 < remain; z4++)
200
            {
201
                tmp_k[ly][z4] = sfp2lfpvec4(weight_data[w_offset + z4]);
202
            }
203
        }
204

205
        barrier();
206

207
        for (int z4 = 0; z4 < remain; z4++)
208
        {
209
            afp v0 = lfp2afp(tmp_v[lx][z4][0]);
210
            afp v1 = lfp2afp(tmp_v[lx][z4][1]);
211
            afp v2 = lfp2afp(tmp_v[lx][z4][2]);
212
            afp v3 = lfp2afp(tmp_v[lx][z4][3]);
213

214
            afpvec4 k = lfp2afpvec4(tmp_k[ly][z4]);
215

216
            sum0 += v0 * k;
217
            sum1 += v1 * k;
218
            sum2 += v2 * k;
219
            sum3 += v3 * k;
220
        }
221
    }
222
#else
223
    for (int z = 0; z < psc(c); z++)
224
    {
225
        afp v0 = buffer_ld1(bottom_blob_data, v_offset + 0);
226
        afp v1 = buffer_ld1(bottom_blob_data, v_offset + 1);
227
        afp v2 = buffer_ld1(bottom_blob_data, v_offset + 2);
228
        afp v3 = buffer_ld1(bottom_blob_data, v_offset + 3);
229

230
        afpvec4 k = buffer_ld4(weight_data, w_offset);
231

232
        sum0 += v0 * k;
233
        sum1 += v1 * k;
234
        sum2 += v2 * k;
235
        sum3 += v3 * k;
236

237
        w_offset += 1;
238
        v_offset += psc(cstep);
239
    }
240
#endif
241
#endif
242

243
    sum0 = activation_afpvec4(sum0, activation_type, activation_param_0, activation_param_1);
244
    sum1 = activation_afpvec4(sum1, activation_type, activation_param_0, activation_param_1);
245
    sum2 = activation_afpvec4(sum2, activation_type, activation_param_0, activation_param_1);
246
    sum3 = activation_afpvec4(sum3, activation_type, activation_param_0, activation_param_1);
247

248
#if NCNN_image_shader
249
    image3d_st4(top_blob, ivec3(sx4.r, sy4.r, gy), sum0);
250
    image3d_st4(top_blob, ivec3(sx4.g, sy4.g, gy), sum1);
251
    image3d_st4(top_blob, ivec3(sx4.b, sy4.b, gy), sum2);
252
    image3d_st4(top_blob, ivec3(sx4.a, sy4.a, gy), sum3);
253
#else
254
#if NCNN_shader_local_memory
255
    if (gx >= psc(outcstep) || gy >= psc(outc))
256
        return;
257
#endif
258

259
    int gi = gy * psc(outcstep) + gx;
260

261
    buffer_st4(top_blob_data, gi + 0, sum0);
262
    if (gx + 1 < psc(outcstep)) buffer_st4(top_blob_data, gi + 1, sum1);
263
    if (gx + 2 < psc(outcstep)) buffer_st4(top_blob_data, gi + 2, sum2);
264
    if (gx + 3 < psc(outcstep)) buffer_st4(top_blob_data, gi + 3, sum3);
265
#endif
266
}
267