File size: 23,669 Bytes
9422383 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 | // Early backbone partitions used by the split-projection optimization. Each
// partition stops at an output-channel split that projection_optimization.cpp
// executes directly through XNNPACK's packed GEMM operator. Reserved values
// preserve XNNPACK's performance-sensitive declaration inventory.
// Normalizes the input layout, runs models 0 and 1, and prepares the two
// model.2 input projections at P2.
bool BuildStemToP2(const GraphContext& context) {
FixedGraphBuilder graph(context, GraphId::kStemToP2, StemToP2Bindings::kCount);
if (!graph.Begin()) return false;
[[maybe_unused]] const GraphValue reserved_value_0 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), Constant(3), ResolutionDivided(1), ResolutionDivided(1)});
const GraphValue input = graph.DefineInput(
ValueType::kFp16, {Constant(1), ResolutionDivided(1), ResolutionDivided(1), Constant(3)},
StemToP2Bindings::kInput);
const GraphValue model0_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(16), Constant(3), Constant(3), Constant(3)},
WeightSlot::kModel0ConvWeight);
[[maybe_unused]] const GraphValue reserved_value_3 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(2), ResolutionDivided(2), Constant(16)});
const GraphValue model0_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(16)}, WeightSlot::kModel0ConvBias);
[[maybe_unused]] const GraphValue reserved_value_5 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(2), ResolutionDivided(2), Constant(16)});
const GraphValue model0_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(2), ResolutionDivided(2), Constant(16)});
const GraphValue model1_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(32), Constant(3), Constant(3), Constant(16)},
WeightSlot::kModel1ConvWeight);
[[maybe_unused]] const GraphValue reserved_value_8 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(32)});
const GraphValue model1_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(32)}, WeightSlot::kModel1ConvBias);
[[maybe_unused]] const GraphValue reserved_value_10 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(32)});
const GraphValue model1_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(32)});
const GraphValue model2_cv1_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(32), Constant(1), Constant(1), Constant(32)},
WeightSlot::kModel2Cv1ConvWeight);
[[maybe_unused]] const GraphValue reserved_value_13 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(32)});
const GraphValue model2_cv1_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(32)}, WeightSlot::kModel2Cv1ConvBias);
[[maybe_unused]] const GraphValue reserved_value_15 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(32)});
const GraphValue model2_cv1_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(32)});
const GraphValue p2_projection_output = graph.DefineOutput(
ValueType::kFp16, {Constant(1), Constant(32), ResolutionDivided(4), ResolutionDivided(4)},
StemToP2Bindings::kP2ProjectionOutput);
if (!graph.DefinitionsValid()) return false;
if (!graph.Convolution(input, model0_conv_weight, model0_conv_bias, model0_conv_output,
Padding{Constant(1), Constant(1), Constant(1), Constant(1)},
Kernel{Constant(3), Constant(3)}, Stride{Constant(2), Constant(2)},
Dilation{Constant(1), Constant(1)},
Groups{Constant(1), Constant(3), Constant(16)},
XNN_FLAG_YOLO26_IGEMM_SILU_FINAL_STORE))
return false;
if (!graph.Convolution(
model0_conv_output, model1_conv_weight, model1_conv_bias, model1_conv_output,
Padding{Constant(1), Constant(1), Constant(1), Constant(1)},
Kernel{Constant(3), Constant(3)}, Stride{Constant(2), Constant(2)},
Dilation{Constant(1), Constant(1)}, Groups{Constant(1), Constant(16), Constant(32)},
XNN_FLAG_YOLO26_IGEMM_SILU_FINAL_STORE))
return false;
if (!graph.Convolution(
model1_conv_output, model2_cv1_conv_weight, model2_cv1_conv_bias, model2_cv1_conv_output,
Padding{Constant(0), Constant(0), Constant(0), Constant(0)},
Kernel{Constant(1), Constant(1)}, Stride{Constant(1), Constant(1)},
Dilation{Constant(1), Constant(1)}, Groups{Constant(1), Constant(32), Constant(32)},
XNN_FLAG_YOLO26_GEMM_SILU_FINAL_STORE))
return false;
if (!graph.Transpose(model2_cv1_conv_output, p2_projection_output,
{Constant(0), Constant(3), Constant(1), Constant(2)}, 0))
return false;
return graph.Finish();
}
// Completes model.2, downsamples through model.3, and prepares the two model.4
// input projections at P3.
bool BuildP2ToP3(const GraphContext& context) {
FixedGraphBuilder graph(context, GraphId::kP2ToP3, P2ToP3Bindings::kCount);
if (!graph.Begin()) return false;
const GraphValue projection_right_input = graph.DefineInput(
ValueType::kFp16, {Constant(1), Constant(16), ResolutionDivided(4), ResolutionDivided(4)},
P2ToP3Bindings::kProjectionRightInput);
const GraphValue transpose_1_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(16)});
const GraphValue projection_left_input = graph.DefineInput(
ValueType::kFp16, {Constant(1), Constant(16), ResolutionDivided(4), ResolutionDivided(4)},
P2ToP3Bindings::kProjectionLeftInput);
const GraphValue transpose_2_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(16)});
const GraphValue model2_m0_cv1_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(8), Constant(3), Constant(3), Constant(16)},
WeightSlot::kModel2M0Cv1ConvWeight);
const GraphValue model2_m0_cv1_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(8)});
const GraphValue model2_m0_cv1_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(8)}, WeightSlot::kModel2M0Cv1ConvBias);
[[maybe_unused]] const GraphValue reserved_value_7 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(8)});
const GraphValue model2_m0_cv1_conv_activation = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(8)});
const GraphValue model2_m0_cv2_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(16), Constant(3), Constant(3), Constant(8)},
WeightSlot::kModel2M0Cv2ConvWeight);
const GraphValue model2_m0_cv2_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(16)});
const GraphValue model2_m0_cv2_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(16)}, WeightSlot::kModel2M0Cv2ConvBias);
[[maybe_unused]] const GraphValue reserved_value_12 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(16)});
const GraphValue model2_m0_cv2_conv_activation = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(16)});
const GraphValue residual_sum_1 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(16)});
const GraphValue concatenation_1 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(48)});
const GraphValue model2_cv2_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(64), Constant(1), Constant(1), Constant(48)},
WeightSlot::kModel2Cv2ConvWeight);
[[maybe_unused]] const GraphValue reserved_value_17 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(64)});
const GraphValue model2_cv2_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(64)}, WeightSlot::kModel2Cv2ConvBias);
[[maybe_unused]] const GraphValue reserved_value_19 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(64)});
const GraphValue model2_cv2_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(4), ResolutionDivided(4), Constant(64)});
const GraphValue model3_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(64), Constant(3), Constant(3), Constant(64)},
WeightSlot::kModel3ConvWeight);
const GraphValue model3_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(64)});
const GraphValue model3_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(64)}, WeightSlot::kModel3ConvBias);
[[maybe_unused]] const GraphValue reserved_value_24 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(64)});
const GraphValue model3_conv_activation = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(64)});
const GraphValue model4_cv1_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(64), Constant(1), Constant(1), Constant(64)},
WeightSlot::kModel4Cv1ConvWeight);
[[maybe_unused]] const GraphValue reserved_value_27 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(64)});
const GraphValue model4_cv1_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(64)}, WeightSlot::kModel4Cv1ConvBias);
[[maybe_unused]] const GraphValue reserved_value_29 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(64)});
const GraphValue model4_cv1_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(64)});
const GraphValue p3_output = graph.DefineOutput(
ValueType::kFp16, {Constant(1), Constant(64), ResolutionDivided(8), ResolutionDivided(8)},
P2ToP3Bindings::kP3Output);
if (!graph.DefinitionsValid()) return false;
if (!graph.Transpose(projection_right_input, transpose_1_output,
{Constant(0), Constant(2), Constant(3), Constant(1)}, 0))
return false;
if (!graph.Transpose(projection_left_input, transpose_2_output,
{Constant(0), Constant(2), Constant(3), Constant(1)}, 0))
return false;
if (!graph.Convolution(
transpose_1_output, model2_m0_cv1_conv_weight, model2_m0_cv1_conv_bias,
model2_m0_cv1_conv_output, Padding{Constant(1), Constant(1), Constant(1), Constant(1)},
Kernel{Constant(3), Constant(3)}, Stride{Constant(1), Constant(1)},
Dilation{Constant(1), Constant(1)}, Groups{Constant(1), Constant(16), Constant(8)}, 0))
return false;
if (!graph.Silu(model2_m0_cv1_conv_output, model2_m0_cv1_conv_activation, 0)) return false;
if (!graph.Convolution(
model2_m0_cv1_conv_activation, model2_m0_cv2_conv_weight, model2_m0_cv2_conv_bias,
model2_m0_cv2_conv_output, Padding{Constant(1), Constant(1), Constant(1), Constant(1)},
Kernel{Constant(3), Constant(3)}, Stride{Constant(1), Constant(1)},
Dilation{Constant(1), Constant(1)}, Groups{Constant(1), Constant(8), Constant(16)}, 0))
return false;
if (!graph.Silu(model2_m0_cv2_conv_output, model2_m0_cv2_conv_activation, 0)) return false;
if (!graph.Add(transpose_1_output, model2_m0_cv2_conv_activation, residual_sum_1, 0))
return false;
if (!graph.Concatenate({transpose_2_output, transpose_1_output, residual_sum_1}, concatenation_1,
Constant(3), XNN_FLAG_YOLO26_FUSED_C2F_CONCATENATE))
return false;
if (!graph.Convolution(
concatenation_1, model2_cv2_conv_weight, model2_cv2_conv_bias, model2_cv2_conv_output,
Padding{Constant(0), Constant(0), Constant(0), Constant(0)},
Kernel{Constant(1), Constant(1)}, Stride{Constant(1), Constant(1)},
Dilation{Constant(1), Constant(1)}, Groups{Constant(1), Constant(48), Constant(64)},
XNN_FLAG_YOLO26_GEMM_SILU_FINAL_STORE))
return false;
if (!graph.Convolution(
model2_cv2_conv_output, model3_conv_weight, model3_conv_bias, model3_conv_output,
Padding{Constant(1), Constant(1), Constant(1), Constant(1)},
Kernel{Constant(3), Constant(3)}, Stride{Constant(2), Constant(2)},
Dilation{Constant(1), Constant(1)}, Groups{Constant(1), Constant(64), Constant(64)}, 0))
return false;
if (!graph.Silu(model3_conv_output, model3_conv_activation, 0)) return false;
if (!graph.Convolution(
model3_conv_activation, model4_cv1_conv_weight, model4_cv1_conv_bias,
model4_cv1_conv_output, Padding{Constant(0), Constant(0), Constant(0), Constant(0)},
Kernel{Constant(1), Constant(1)}, Stride{Constant(1), Constant(1)},
Dilation{Constant(1), Constant(1)}, Groups{Constant(1), Constant(64), Constant(64)},
XNN_FLAG_YOLO26_GEMM_SILU_FINAL_STORE))
return false;
if (!graph.Transpose(model4_cv1_conv_output, p3_output,
{Constant(0), Constant(3), Constant(1), Constant(2)}, 0))
return false;
return graph.Finish();
}
// Completes model.4, downsamples through model.5, and prepares the two model.6
// input projections at P4.
bool BuildP3ToP4(const GraphContext& context) {
FixedGraphBuilder graph(context, GraphId::kP3ToP4, P3ToP4Bindings::kCount);
if (!graph.Begin()) return false;
const GraphValue projection_right_input = graph.DefineInput(
ValueType::kFp16, {Constant(1), Constant(32), ResolutionDivided(8), ResolutionDivided(8)},
P3ToP4Bindings::kProjectionRightInput);
const GraphValue transpose_1_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(32)});
const GraphValue projection_left_input = graph.DefineInput(
ValueType::kFp16, {Constant(1), Constant(32), ResolutionDivided(8), ResolutionDivided(8)},
P3ToP4Bindings::kProjectionLeftInput);
const GraphValue transpose_2_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(32)});
const GraphValue model4_m0_cv1_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(16), Constant(3), Constant(3), Constant(32)},
WeightSlot::kModel4M0Cv1ConvWeight);
const GraphValue model4_m0_cv1_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(16)});
const GraphValue model4_m0_cv1_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(16)}, WeightSlot::kModel4M0Cv1ConvBias);
[[maybe_unused]] const GraphValue reserved_value_7 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(16)});
const GraphValue model4_m0_cv1_conv_activation = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(16)});
const GraphValue model4_m0_cv2_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(32), Constant(3), Constant(3), Constant(16)},
WeightSlot::kModel4M0Cv2ConvWeight);
const GraphValue model4_m0_cv2_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(32)});
const GraphValue model4_m0_cv2_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(32)}, WeightSlot::kModel4M0Cv2ConvBias);
[[maybe_unused]] const GraphValue reserved_value_12 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(32)});
const GraphValue model4_m0_cv2_conv_activation = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(32)});
const GraphValue residual_sum_1 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(32)});
const GraphValue concatenation_1 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(96)});
const GraphValue model4_cv2_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(128), Constant(1), Constant(1), Constant(96)},
WeightSlot::kModel4Cv2ConvWeight);
[[maybe_unused]] const GraphValue reserved_value_17 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(128)});
const GraphValue model4_cv2_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(128)}, WeightSlot::kModel4Cv2ConvBias);
[[maybe_unused]] const GraphValue reserved_value_19 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(128)});
const GraphValue p3_skip_input = graph.DefineOutput(
ValueType::kFp16, {Constant(1), ResolutionDivided(8), ResolutionDivided(8), Constant(128)},
P3ToP4Bindings::kP3SkipInput);
[[maybe_unused]] const GraphValue reserved_value_21 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), Constant(128), ResolutionDivided(8), ResolutionDivided(8)});
const GraphValue model5_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(128), Constant(3), Constant(3), Constant(128)},
WeightSlot::kModel5ConvWeight);
const GraphValue model5_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(16), ResolutionDivided(16), Constant(128)});
const GraphValue model5_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(128)}, WeightSlot::kModel5ConvBias);
[[maybe_unused]] const GraphValue reserved_value_25 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(16), ResolutionDivided(16), Constant(128)});
const GraphValue model5_conv_activation = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(16), ResolutionDivided(16), Constant(128)});
const GraphValue model6_cv1_conv_weight =
graph.DefineWeight(ValueType::kFp32, {Constant(128), Constant(1), Constant(1), Constant(128)},
WeightSlot::kModel6Cv1ConvWeight);
[[maybe_unused]] const GraphValue reserved_value_28 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(16), ResolutionDivided(16), Constant(128)});
const GraphValue model6_cv1_conv_bias =
graph.DefineWeight(ValueType::kFp32, {Constant(128)}, WeightSlot::kModel6Cv1ConvBias);
[[maybe_unused]] const GraphValue reserved_value_30 = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(16), ResolutionDivided(16), Constant(128)});
const GraphValue model6_cv1_conv_output = graph.DefineIntermediate(
ValueType::kFp16, {Constant(1), ResolutionDivided(16), ResolutionDivided(16), Constant(128)});
const GraphValue p4_entry_output = graph.DefineOutput(
ValueType::kFp16, {Constant(1), Constant(128), ResolutionDivided(16), ResolutionDivided(16)},
P3ToP4Bindings::kP4EntryOutput);
if (!graph.DefinitionsValid()) return false;
if (!graph.Transpose(projection_right_input, transpose_1_output,
{Constant(0), Constant(2), Constant(3), Constant(1)}, 0))
return false;
if (!graph.Transpose(projection_left_input, transpose_2_output,
{Constant(0), Constant(2), Constant(3), Constant(1)}, 0))
return false;
if (!graph.Convolution(
transpose_1_output, model4_m0_cv1_conv_weight, model4_m0_cv1_conv_bias,
model4_m0_cv1_conv_output, Padding{Constant(1), Constant(1), Constant(1), Constant(1)},
Kernel{Constant(3), Constant(3)}, Stride{Constant(1), Constant(1)},
Dilation{Constant(1), Constant(1)}, Groups{Constant(1), Constant(32), Constant(16)}, 0))
return false;
if (!graph.Silu(model4_m0_cv1_conv_output, model4_m0_cv1_conv_activation, 0)) return false;
if (!graph.Convolution(
model4_m0_cv1_conv_activation, model4_m0_cv2_conv_weight, model4_m0_cv2_conv_bias,
model4_m0_cv2_conv_output, Padding{Constant(1), Constant(1), Constant(1), Constant(1)},
Kernel{Constant(3), Constant(3)}, Stride{Constant(1), Constant(1)},
Dilation{Constant(1), Constant(1)}, Groups{Constant(1), Constant(16), Constant(32)}, 0))
return false;
if (!graph.Silu(model4_m0_cv2_conv_output, model4_m0_cv2_conv_activation, 0)) return false;
if (!graph.Add(transpose_1_output, model4_m0_cv2_conv_activation, residual_sum_1, 0))
return false;
if (!graph.Concatenate({transpose_2_output, transpose_1_output, residual_sum_1}, concatenation_1,
Constant(3), 0))
return false;
if (!graph.Convolution(concatenation_1, model4_cv2_conv_weight, model4_cv2_conv_bias,
p3_skip_input, Padding{Constant(0), Constant(0), Constant(0), Constant(0)},
Kernel{Constant(1), Constant(1)}, Stride{Constant(1), Constant(1)},
Dilation{Constant(1), Constant(1)},
Groups{Constant(1), Constant(96), Constant(128)},
XNN_FLAG_YOLO26_GEMM_SILU_FINAL_STORE))
return false;
if (!graph.Convolution(p3_skip_input, model5_conv_weight, model5_conv_bias, model5_conv_output,
Padding{Constant(1), Constant(1), Constant(1), Constant(1)},
Kernel{Constant(3), Constant(3)}, Stride{Constant(2), Constant(2)},
Dilation{Constant(1), Constant(1)},
Groups{Constant(1), Constant(128), Constant(128)}, 0))
return false;
if (!graph.Silu(model5_conv_output, model5_conv_activation, 0)) return false;
if (!graph.Convolution(
model5_conv_activation, model6_cv1_conv_weight, model6_cv1_conv_bias,
model6_cv1_conv_output, Padding{Constant(0), Constant(0), Constant(0), Constant(0)},
Kernel{Constant(1), Constant(1)}, Stride{Constant(1), Constant(1)},
Dilation{Constant(1), Constant(1)}, Groups{Constant(1), Constant(128), Constant(128)},
XNN_FLAG_YOLO26_GEMM_SILU_FINAL_STORE))
return false;
if (!graph.Transpose(model6_cv1_conv_output, p4_entry_output,
{Constant(0), Constant(3), Constant(1), Constant(2)}, 0))
return false;
return graph.Finish();
}
|