File size: 7,098 Bytes
be903e2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 | // Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2023 THL A29 Limited, a Tencent company. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.
#version 450
#if NCNN_fp16_storage
#extension GL_EXT_shader_16bit_storage: require
struct sfpvec8 { f16vec4 abcd; f16vec4 efgh; };
#endif
#if NCNN_fp16_arithmetic
#extension GL_EXT_shader_explicit_arithmetic_types_float16: require
#endif
layout (constant_id = 0) const int op_type = 0;
#define shape_constant_id_offset 1
layout (constant_id = shape_constant_id_offset + 0) const int adims = 0;
layout (constant_id = shape_constant_id_offset + 1) const int aw = 0;
layout (constant_id = shape_constant_id_offset + 2) const int ah = 0;
layout (constant_id = shape_constant_id_offset + 3) const int ad = 0;
layout (constant_id = shape_constant_id_offset + 4) const int ac = 0;
layout (constant_id = shape_constant_id_offset + 5) const int acstep = 0;
layout (constant_id = shape_constant_id_offset + 6) const int bdims = 0;
layout (constant_id = shape_constant_id_offset + 7) const int bw = 0;
layout (constant_id = shape_constant_id_offset + 8) const int bh = 0;
layout (constant_id = shape_constant_id_offset + 9) const int bd = 0;
layout (constant_id = shape_constant_id_offset + 10) const int bc = 0;
layout (constant_id = shape_constant_id_offset + 11) const int bcstep = 0;
layout (constant_id = shape_constant_id_offset + 12) const int outdims = 0;
layout (constant_id = shape_constant_id_offset + 13) const int outw = 0;
layout (constant_id = shape_constant_id_offset + 14) const int outh = 0;
layout (constant_id = shape_constant_id_offset + 15) const int outd = 0;
layout (constant_id = shape_constant_id_offset + 16) const int outc = 0;
layout (constant_id = shape_constant_id_offset + 17) const int outcstep = 0;
#if NCNN_image_shader
layout (binding = 0) uniform unfp sampler3D a_blob_3d;
layout (binding = 1) uniform unfp sampler3D b_blob_3d;
layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
#else
layout (binding = 0) readonly buffer a_blob { sfpvec8 a_blob_data[]; };
layout (binding = 1) readonly buffer b_blob { sfpvec8 b_blob_data[]; };
layout (binding = 2) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
#endif
layout (push_constant) uniform parameter
{
int adims;
int aw;
int ah;
int ad;
int ac;
int acstep;
int bdims;
int bw;
int bh;
int bd;
int bc;
int bcstep;
int outdims;
int outw;
int outh;
int outd;
int outc;
int outcstep;
} p;
void main()
{
int gx = int(gl_GlobalInvocationID.x);
int gy = int(gl_GlobalInvocationID.y);
int gz = int(gl_GlobalInvocationID.z);
if (gx >= psc(outw) || gy >= psc(outh) * psc(outd) || gz >= psc(outc))
return;
int yd = gy / psc(outh);
int yh = gy % psc(outh);
int ax = gx;
int ay = gy;
int az = gz;
int bx = gx;
int by = gy;
int bz = gz;
if (psc(adims) == psc(bdims))
{
// explicit broadcast
ax = min(gx, psc(aw) - 1);
ay = min(yd, psc(ad) - 1) * psc(ah) + min(yh, psc(ah) - 1);
az = min(gz, psc(ac) - 1);
bx = min(gx, psc(bw) - 1);
by = min(yd, psc(bd) - 1) * psc(bh) + min(yh, psc(bh) - 1);
bz = min(gz, psc(bc) - 1);
}
else
{
// implicit broadcast
if (psc(adims) == 4)
{
if (psc(bdims) == 3)
{
// type 13
bx = yh;
by = yd;
bz = gz;
}
if (psc(bdims) == 2)
{
// type 12
bx = yd;
by = gz;
bz = 0;
}
if (psc(bdims) == 1)
{
// type 11
bx = gz;
by = 0;
bz = 0;
}
}
else if (psc(adims) == 3)
{
if (psc(bdims) == 2)
{
// type 10
bx = gy;
by = gz;
bz = 0;
}
if (psc(bdims) == 1)
{
// type 9
bx = gz;
by = 0;
bz = 0;
}
}
else // if (psc(adims) == 2)
{
// if (psc(bdims) == 1)
{
// type 8
bx = gy;
by = 0;
bz = 0;
}
}
}
#if NCNN_image_shader
afpvec8 v1 = image3d_ld8(a_blob_3d, ivec3(ax, ay, az));
afpvec8 v2 = image3d_ld8(b_blob_3d, ivec3(bx, by, bz));
#else
int ai = az * psc(acstep) + ay * psc(aw) + ax;
int bi = bz * psc(bcstep) + by * psc(bw) + bx;
afpvec8 v1 = buffer_ld8(a_blob_data, ai);
afpvec8 v2 = buffer_ld8(b_blob_data, bi);
#endif
afpvec8 res;
if (op_type == 0)
{
res[0] = v1[0] + v2[0];
res[1] = v1[1] + v2[1];
}
if (op_type == 1)
{
res[0] = v1[0] - v2[0];
res[1] = v1[1] - v2[1];
}
if (op_type == 2)
{
res[0] = v1[0] * v2[0];
res[1] = v1[1] * v2[1];
}
if (op_type == 3)
{
res[0] = v1[0] / v2[0];
res[1] = v1[1] / v2[1];
}
if (op_type == 4)
{
res[0] = max(v1[0], v2[0]);
res[1] = max(v1[1], v2[1]);
}
if (op_type == 5)
{
res[0] = min(v1[0], v2[0]);
res[1] = min(v1[1], v2[1]);
}
if (op_type == 6)
{
res[0] = pow(v1[0], v2[0]);
res[1] = pow(v1[1], v2[1]);
}
if (op_type == 7)
{
res[0] = v2[0] - v1[0];
res[1] = v2[1] - v1[1];
}
if (op_type == 8)
{
res[0] = v2[0] / v1[0];
res[1] = v2[1] / v1[1];
}
if (op_type == 9)
{
res[0] = pow(v2[0], v1[0]);
res[1] = pow(v2[1], v1[1]);
}
if (op_type == 10)
{
#if NCNN_moltenvk
res[0] = afpvec4(atan(vec4(v1[0]), vec4(v2[0])));
res[1] = afpvec4(atan(vec4(v1[1]), vec4(v2[1])));
#else
res[0] = atan(v1[0], v2[0]);
res[1] = atan(v1[1], v2[1]);
#endif
}
if (op_type == 11)
{
#if NCNN_moltenvk
res[0] = afpvec4(atan(vec4(v2[0]), vec4(v1[0])));
res[1] = afpvec4(atan(vec4(v2[1]), vec4(v1[1])));
#else
res[0] = atan(v2[0], v1[0]);
res[1] = atan(v2[1], v1[1]);
#endif
}
#if NCNN_image_shader
image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
#else
int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
buffer_st8(top_blob_data, gi, res);
#endif
}
|