notesum.ai
Published at November 25Optimizing Winograd Convolution on ARMv8 manycore processors
cs.PF
Released Date: November 25, 2024
Authors: Haoyuan Gui1, Xiaoyu Zhang2, Chong Zhang2, Zitong Su2, Huiyuan Li3
Aff.: 1Institute of Software, Chinese Academy of Sciences, China; 2Institute of Software, Chinese Academy of Sciences, China; Also at University of Chinese Academy of Sciences, China; 3Institute of Software, Chinese Academy of Sciences; Also at State Key Laboratory of Computer Science, Chinese Academy of Sciences, China

| Layer | C | K | H & W | R & S |
| VggNet_1.2 | 64 | 64 | 224 | 3 |
| VggNet_2.2 | 128 | 128 | 112 | 3 |
| VggNet_3.2 | 256 | 256 | 56 | 3 |
| VggNet_4.2 | 512 | 512 | 28 | 3 |
| VggNet_5.2 | 512 | 512 | 14 | 3 |
| FusionNet_1.2 | 64 | 64 | 640 | 3 |
| FusionNet_2.2 | 128 | 128 | 320 | 3 |
| FusionNet_3.2 | 256 | 256 | 160 | 3 |
| FusionNet_4.2 | 512 | 512 | 80 | 3 |
| FusionNet_5.2 | 1024 | 1024 | 40 | 3 |
| ResNet_2.1 | 64 | 64 | 112 | 3 |
| ResNet_3.1 | 128 | 128 | 56 | 3 |
| ResNet_4.1 | 256 | 256 | 28 | 3 |
| ResNet_5.1 | 512 | 512 | 14 | 3 |