1
// Tencent is pleased to support the open source community by making ncnn available.
3
// Copyright (C) 2019 THL A29 Limited, a Tencent company. All rights reserved.
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
8
// https://opensource.org/licenses/BSD-3-Clause
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.
18
#extension GL_EXT_shader_16bit_storage: require
20
#if NCNN_fp16_arithmetic
21
#extension GL_EXT_shader_explicit_arithmetic_types_float16: require
24
#extension GL_GOOGLE_include_directive: enable
25
#include "vulkan_activation.comp"
27
layout (constant_id = 0) const int kernel_w = 1;
28
layout (constant_id = 1) const int kernel_h = 1;
29
layout (constant_id = 2) const int dilation_w = 1;
30
layout (constant_id = 3) const int dilation_h = 1;
31
layout (constant_id = 4) const int stride_w = 1;
32
layout (constant_id = 5) const int stride_h = 1;
33
layout (constant_id = 6) const int bias_term = 0;
34
layout (constant_id = 7) const int activation_type = 0;
35
layout (constant_id = 8) const float activation_param_0 = 0;
36
layout (constant_id = 9) const float activation_param_1 = 0;
38
#define shape_constant_id_offset 10
39
layout (constant_id = shape_constant_id_offset + 0) const int dims = 0;
40
layout (constant_id = shape_constant_id_offset + 1) const int w = 0;
41
layout (constant_id = shape_constant_id_offset + 2) const int h = 0;
42
layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
43
layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
45
layout (constant_id = shape_constant_id_offset + 5) const int outdims = 0;
46
layout (constant_id = shape_constant_id_offset + 6) const int outw = 0;
47
layout (constant_id = shape_constant_id_offset + 7) const int outh = 0;
48
layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
49
layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
52
layout (binding = 0) uniform unfp sampler3D bottom_blob;
53
layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
54
layout (binding = 2) uniform unfp sampler3D weight_blob;
55
layout (binding = 3) uniform unfp sampler3D bias_blob;
57
layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
58
layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
59
#if NCNN_fp16_packed || (NCNN_fp16_storage && !NCNN_fp16_arithmetic)
60
// GL_EXT_shader_16bit_storage does not define f16mat4 type :(
61
layout (binding = 2) readonly buffer weight_blob { sfpvec4 weight_data[]; };
63
layout (binding = 2) readonly buffer weight_blob { sfpmat4 weight_data[]; };
65
layout (binding = 3) readonly buffer bias_blob { sfpvec4 bias_data[]; };
68
layout (push_constant) uniform parameter
85
int gx = int(gl_GlobalInvocationID.x);
86
int gy = int(gl_GlobalInvocationID.y);
87
int gz = int(gl_GlobalInvocationID.z);
89
if (gx >= psc(outw) || gy >= psc(outh) || gz >= psc(outc))
97
sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
99
sum = buffer_ld4(bias_data, gz);
107
const int kernel_extent_w = dilation_w * (kernel_w - 1) + 1;
108
const int kernel_extent_h = dilation_h * (kernel_h - 1) + 1;
111
for (int y = 0; y < kernel_h; y++)
113
int sys = (gy + y * dilation_h - (kernel_extent_h - 1));
114
if (sys < 0 || sys % stride_h != 0)
117
int sy = sys / stride_h;
121
for (int x = 0; x < kernel_w; x++)
123
int sxs = (gx + x * dilation_w - (kernel_extent_w - 1));
124
if (sxs < 0 || sxs % stride_w != 0)
127
int sx = sxs / stride_w;
131
int wx = (y * kernel_w + x) * 4;
133
for (int z = 0; z < psc(c); z++)
135
afpvec4 v = image3d_ld4(bottom_blob, ivec3(sx, sy, z));
138
image3d_ld4(weight_blob, ivec3(wx + 0, z, gz)),
139
image3d_ld4(weight_blob, ivec3(wx + 1, z, gz)),
140
image3d_ld4(weight_blob, ivec3(wx + 2, z, gz)),
141
image3d_ld4(weight_blob, ivec3(wx + 3, z, gz))
149
int w_offset_0 = gz * psc(c) * kernel_w * kernel_h;
151
for (int y = 0; y < kernel_h; y++)
153
int sys = (gy + y * dilation_h - (kernel_extent_h - 1));
154
if (sys < 0 || sys % stride_h != 0)
157
int sy = sys / stride_h;
161
for (int x = 0; x < kernel_w; x++)
163
int sxs = (gx + x * dilation_w - (kernel_extent_w - 1));
164
if (sxs < 0 || sxs % stride_w != 0)
167
int sx = sxs / stride_w;
171
int v_offset = sy * psc(w) + sx;
172
int w_offset = w_offset_0 + y * kernel_w + x;
174
for (int z = 0; z < psc(c); z++)
176
afpvec4 v = buffer_ld4(bottom_blob_data, v_offset);
178
#if NCNN_fp16_packed || (NCNN_fp16_storage && !NCNN_fp16_arithmetic)
179
// GL_EXT_shader_16bit_storage does not define f16mat4 type :(
181
buffer_ld4(weight_data, w_offset * 4 + 0),
182
buffer_ld4(weight_data, w_offset * 4 + 1),
183
buffer_ld4(weight_data, w_offset * 4 + 2),
184
buffer_ld4(weight_data, w_offset * 4 + 3)
187
afpmat4 k = afpmat4(weight_data[w_offset]);
192
v_offset += psc(cstep);
193
w_offset += kernel_w * kernel_h;
199
sum = activation_afpvec4(sum, activation_type, activation_param_0, activation_param_1);
202
image3d_st4(top_blob, ivec3(gx, gy, gz), sum);
204
const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
206
buffer_st4(top_blob_data, gi, sum);