1
// Tencent is pleased to support the open source community by making ncnn available.
3
// Copyright (C) 2020 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
layout (constant_id = 0) const int order_type = 0;
25
layout (constant_id = 1) const int bugihfa = 0;
27
#define shape_constant_id_offset 2
28
layout (constant_id = shape_constant_id_offset + 0) const int dims = 0;
29
layout (constant_id = shape_constant_id_offset + 1) const int w = 0;
30
layout (constant_id = shape_constant_id_offset + 2) const int h = 0;
31
layout (constant_id = shape_constant_id_offset + 3) const int d = 0;
32
layout (constant_id = shape_constant_id_offset + 4) const int c = 0;
33
layout (constant_id = shape_constant_id_offset + 5) const int cstep = 0;
35
layout (constant_id = shape_constant_id_offset + 6) const int outdims = 0;
36
layout (constant_id = shape_constant_id_offset + 7) const int outw = 0;
37
layout (constant_id = shape_constant_id_offset + 8) const int outh = 0;
38
layout (constant_id = shape_constant_id_offset + 9) const int outd = 0;
39
layout (constant_id = shape_constant_id_offset + 10) const int outc = 0;
40
layout (constant_id = shape_constant_id_offset + 11) const int outcstep = 0;
43
layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
44
layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
46
layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
47
layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
50
layout (push_constant) uniform parameter
69
int gx = int(gl_GlobalInvocationID.x);
70
int gy = int(gl_GlobalInvocationID.y);
71
int gz = int(gl_GlobalInvocationID.z);
73
if (gx >= psc(outw) || gy >= psc(outh) * psc(outd) || gz >= psc(outc))
92
y4 = gy * 4 + ivec4(0, 1, 2, 3);
96
x4 = gy * 4 + ivec4(0, 1, 2, 3);
100
else if (psc(dims) == 3)
114
z4 = gz * 4 + ivec4(0, 1, 2, 3);
120
z4 = gz * 4 + ivec4(0, 1, 2, 3);
125
y4 = gz * 4 + ivec4(0, 1, 2, 3);
131
y4 = gz * 4 + ivec4(0, 1, 2, 3);
136
x4 = gz * 4 + ivec4(0, 1, 2, 3);
142
x4 = gz * 4 + ivec4(0, 1, 2, 3);
147
else // if (psc(dims) == 4)
175
int yd = gy / psc(outh);
176
int yh = gy % psc(outh);
181
y4 = ivec4(yd * psc(h) + yh);
182
z4 = gz * 4 + ivec4(0, 1, 2, 3);
187
y4 = ivec4(yd * psc(h) + gx);
188
z4 = gz * 4 + ivec4(0, 1, 2, 3);
193
y4 = ivec4(yh * psc(h) + yd);
194
z4 = gz * 4 + ivec4(0, 1, 2, 3);
199
y4 = ivec4(gx * psc(h) + yd);
200
z4 = gz * 4 + ivec4(0, 1, 2, 3);
205
y4 = ivec4(yh * psc(h) + gx);
206
z4 = gz * 4 + ivec4(0, 1, 2, 3);
211
y4 = ivec4(gx * psc(h) + yh);
212
z4 = gz * 4 + ivec4(0, 1, 2, 3);
217
y4 = (gz * 4 + ivec4(0, 1, 2, 3)) * psc(h) + yh;
223
y4 = (gz * 4 + ivec4(0, 1, 2, 3)) * psc(h) + gx;
229
y4 = (gz * 4 + ivec4(0, 1, 2, 3)) * psc(h) + yd;
235
y4 = (gz * 4 + ivec4(0, 1, 2, 3)) * psc(h) + yd;
238
if (order_type == 10)
241
y4 = (gz * 4 + ivec4(0, 1, 2, 3)) * psc(h) + gx;
244
if (order_type == 11)
247
y4 = (gz * 4 + ivec4(0, 1, 2, 3)) * psc(h) + yh;
250
if (order_type == 12)
253
y4 = yh * psc(h) + gz * 4 + ivec4(0, 1, 2, 3);
256
if (order_type == 13)
259
y4 = gx * psc(h) + gz * 4 + ivec4(0, 1, 2, 3);
262
if (order_type == 14)
265
y4 = yd * psc(h) + gz * 4 + ivec4(0, 1, 2, 3);
268
if (order_type == 15)
271
y4 = yd * psc(h) + gz * 4 + ivec4(0, 1, 2, 3);
274
if (order_type == 16)
277
y4 = gx * psc(h) + gz * 4 + ivec4(0, 1, 2, 3);
280
if (order_type == 17)
283
y4 = yh * psc(h) + gz * 4 + ivec4(0, 1, 2, 3);
286
if (order_type == 18)
288
x4 = gz * 4 + ivec4(0, 1, 2, 3);
289
y4 = ivec4(yh * psc(h) + gx);
292
if (order_type == 19)
294
x4 = gz * 4 + ivec4(0, 1, 2, 3);
295
y4 = ivec4(gx * psc(h) + yh);
298
if (order_type == 20)
300
x4 = gz * 4 + ivec4(0, 1, 2, 3);
301
y4 = ivec4(yd * psc(h) + gx);
304
if (order_type == 21)
306
x4 = gz * 4 + ivec4(0, 1, 2, 3);
307
y4 = ivec4(yd * psc(h) + yh);
310
if (order_type == 22)
312
x4 = gz * 4 + ivec4(0, 1, 2, 3);
313
y4 = ivec4(gx * psc(h) + yd);
316
if (order_type == 23)
318
x4 = gz * 4 + ivec4(0, 1, 2, 3);
319
y4 = ivec4(yh * psc(h) + yd);
326
v.r = image3d_ld1(bottom_blob_3d, ivec3(x4.r, y4.r, z4.r));
327
v.g = image3d_ld1(bottom_blob_3d, ivec3(x4.g, y4.g, z4.g));
328
v.b = image3d_ld1(bottom_blob_3d, ivec3(x4.b, y4.b, z4.b));
329
v.a = image3d_ld1(bottom_blob_3d, ivec3(x4.a, y4.a, z4.a));
331
image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
333
ivec4 v_offset = z4 * psc(cstep) + y4 * psc(w) + x4;
335
const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
337
buffer_cp1to4(top_blob_data, gi, bottom_blob_data, v_offset);