1# Copyright 2021 Google LLC 2# 3# This source code is licensed under the BSD-style license found in the 4# LICENSE file in the root directory of this source tree. 5 6# AArch32 assembly 7- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8__aarch32_neon_mlal_lane_cortex_a7 8 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 9 k-block: 8 10 assembly: true 11- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8__aarch32_neon_mlal_lane_prfm_cortex_a7 12 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 13 k-block: 8 14 assembly: true 15- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_cortex_a53 16 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 17 k-block: 8 18 assembly: true 19- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_ld64 20 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 21 k-block: 8 22 assembly: true 23- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_prfm_cortex_a53 24 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 25 k-block: 8 26 assembly: true 27- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_prfm_ld64 28 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 29 k-block: 8 30 assembly: true 31- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_cortex_a55 32 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 33 k-block: 8 34 assembly: true 35- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__aarch32_neondot_ld64 36 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 37 k-block: 8 38 assembly: true 39# AArch64 assembly 40- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c8__aarch64_neon_mlal 41 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 42 k-block: 16 43 assembly: true 44- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c8__aarch64_neon_mlal_cortex_a53 45 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 46 k-block: 16 47 assembly: true 48- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c8__aarch64_neon_mlal_prfm 49 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 50 k-block: 16 51 assembly: true 52- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c8__aarch64_neon_mlal_prfm_cortex_a53 53 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 54 k-block: 16 55 assembly: true 56- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c8__aarch64_neon_mlal 57 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 58 k-block: 16 59 assembly: true 60- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c8__aarch64_neon_mlal_cortex_a53 61 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 62 k-block: 16 63 assembly: true 64- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c8__aarch64_neon_mlal_prfm 65 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 66 k-block: 16 67 assembly: true 68- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c8__aarch64_neon_mlal_prfm_cortex_a53 69 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 70 k-block: 16 71 assembly: true 72- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c16__aarch64_neon_mlal 73 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 74 k-block: 16 75 assembly: true 76- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8__aarch64_neon_mlal_lane_ld64 77 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 78 k-block: 8 79 assembly: true 80- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8__aarch64_neon_mlal_lane_prfm_ld64 81 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 82 k-block: 8 83 assembly: true 84- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16__aarch64_neon_mlal_lane_cortex_a53 85 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 86 k-block: 8 87 assembly: true 88- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16__aarch64_neon_mlal_lane_ld64 89 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 90 k-block: 8 91 assembly: true 92- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16__aarch64_neon_mlal_lane_prfm_cortex_a53 93 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 94 k-block: 8 95 assembly: true 96- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16__aarch64_neon_mlal_lane_prfm_ld64 97 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 98 k-block: 8 99 assembly: true 100- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__aarch64_neondot_cortex_a55 101 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 102 k-block: 16 103 assembly: true 104- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__aarch64_neondot_ld64 105 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 106 k-block: 8 107 assembly: true 108- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__aarch64_neondot_ld128 109 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 110 k-block: 16 111 assembly: true 112# ARM NEON 113- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8__neon_mlal_lane 114 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 115 k-block: 8 116- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8__neon_mlal_lane_prfm 117 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 118 k-block: 8 119- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8__neon_mull_addw_dup 120 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 121 k-block: 8 122- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2__neon_mlal_dup 123 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 124 k-block: 16 125- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2__neon_mlal_ld1r 126 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 127 k-block: 16 128- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2__neon_mlal_ld2r 129 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 130 k-block: 16 131- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2__neon_mlal_ld4r 132 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 133 k-block: 16 134- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2__neon_mull_dup 135 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 136 k-block: 8 137- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2__neon_mull_ld1r 138 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 139 k-block: 8 140- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2__neon_mull_ld2r 141 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 142 k-block: 8 143- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2__neon_mull_ld4r 144 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 145 k-block: 8 146- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mlal 147 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 148 k-block: 16 149- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c2s4__neon_mull 150 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 151 k-block: 8 152- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4__neon_mlal_dup 153 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 154 k-block: 16 155- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4__neon_mlal_ld1r 156 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 157 k-block: 16 158- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4__neon_mlal_ld2r 159 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 160 k-block: 16 161- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4__neon_mull_dup 162 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 163 k-block: 8 164- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4__neon_mull_ld1r 165 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 166 k-block: 8 167- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4__neon_mull_ld2r 168 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 169 k-block: 8 170- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4__neondot 171 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 172 k-block: 8 173- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mlal 174 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 175 k-block: 16 176- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c4s2__neon_mull 177 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 178 k-block: 8 179- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c8__neon_mlal 180 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 181 k-block: 16 182- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c8__neon_mull 183 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 184 k-block: 8 185- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x8c16__neon_mlal 186 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 187 k-block: 16 188- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16__neon_mlal_lane 189 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 190 k-block: 8 191- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16__neon_mlal_lane_prfm 192 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 193 k-block: 8 194- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16__neon_mull_addw_dup 195 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 196 k-block: 8 197- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2__neon_mlal_dup 198 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 199 k-block: 16 200- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2__neon_mlal_ld1r 201 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 202 k-block: 16 203- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2__neon_mlal_ld2r 204 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 205 k-block: 16 206- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2__neon_mlal_ld4r 207 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 208 k-block: 16 209- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2__neon_mull_dup 210 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 211 k-block: 8 212- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2__neon_mull_ld1r 213 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 214 k-block: 8 215- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2__neon_mull_ld2r 216 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 217 k-block: 8 218- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2__neon_mull_ld4r 219 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 220 k-block: 8 221- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mlal 222 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 223 k-block: 16 224- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c2s4__neon_mull 225 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 226 k-block: 8 227- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4__neon_mlal_dup 228 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 229 k-block: 16 230- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4__neon_mlal_ld1r 231 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 232 k-block: 16 233- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4__neon_mlal_ld2r 234 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 235 k-block: 16 236- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4__neon_mull_dup 237 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 238 k-block: 8 239- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4__neon_mull_ld1r 240 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 241 k-block: 8 242- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4__neon_mull_ld2r 243 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 244 k-block: 8 245- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4__neondot 246 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 247 k-block: 8 248- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mlal 249 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 250 k-block: 16 251- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c4s2__neon_mull 252 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 253 k-block: 8 254- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c8__neon_mlal 255 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 256 k-block: 16 257- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c8__neon_mull 258 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 259 k-block: 8 260- name: xnn_qs8_igemm_minmax_rndnu_ukernel_1x16c16__neon_mlal 261 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 262 k-block: 16 263- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8__neon_mlal_lane 264 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 265 k-block: 8 266- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8__neon_mlal_lane_prfm 267 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 268 k-block: 8 269- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8__neon_mull_addw_dup 270 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 271 k-block: 8 272- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2__neon_mlal_dup 273 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 274 k-block: 16 275- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2__neon_mlal_ld1r 276 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 277 k-block: 16 278- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2__neon_mlal_ld2r 279 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 280 k-block: 16 281- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2__neon_mlal_ld4r 282 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 283 k-block: 16 284- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2__neon_mull_dup 285 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 286 k-block: 8 287- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2__neon_mull_ld1r 288 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 289 k-block: 8 290- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2__neon_mull_ld2r 291 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 292 k-block: 8 293- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2__neon_mull_ld4r 294 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 295 k-block: 8 296- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mlal 297 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 298 k-block: 16 299- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c2s4__neon_mull 300 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 301 k-block: 8 302- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4__neon_mlal_dup 303 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 304 k-block: 16 305- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4__neon_mlal_ld1r 306 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 307 k-block: 16 308- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4__neon_mlal_ld2r 309 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 310 k-block: 16 311- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4__neon_mull_dup 312 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 313 k-block: 8 314- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4__neon_mull_ld1r 315 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 316 k-block: 8 317- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4__neon_mull_ld2r 318 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 319 k-block: 8 320- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mlal 321 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 322 k-block: 16 323- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c4s2__neon_mull 324 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 325 k-block: 8 326- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c8__neon_mlal 327 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 328 k-block: 16 329- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c8__neon_mull 330 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 331 k-block: 8 332- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x8c16__neon_mlal 333 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 334 k-block: 16 335- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16__neon_mlal_lane 336 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 337 k-block: 8 338- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16__neon_mlal_lane_prfm 339 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 340 k-block: 8 341- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16__neon_mull_addw_dup 342 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 343 k-block: 8 344- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2__neon_mlal_dup 345 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 346 k-block: 16 347- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2__neon_mlal_ld1r 348 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 349 k-block: 16 350- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2__neon_mlal_ld2r 351 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 352 k-block: 16 353- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2__neon_mlal_ld4r 354 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 355 k-block: 16 356- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2__neon_mull_dup 357 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 358 k-block: 8 359- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2__neon_mull_ld1r 360 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 361 k-block: 8 362- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2__neon_mull_ld2r 363 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 364 k-block: 8 365- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2__neon_mull_ld4r 366 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 367 k-block: 8 368- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mlal 369 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 370 k-block: 16 371- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c2s4__neon_mull 372 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 373 k-block: 8 374- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4__neon_mlal_dup 375 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 376 k-block: 16 377- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4__neon_mlal_ld1r 378 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 379 k-block: 16 380- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4__neon_mlal_ld2r 381 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 382 k-block: 16 383- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4__neon_mull_dup 384 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 385 k-block: 8 386- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4__neon_mull_ld1r 387 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 388 k-block: 8 389- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4__neon_mull_ld2r 390 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 391 k-block: 8 392- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mlal 393 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 394 k-block: 16 395- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c4s2__neon_mull 396 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 397 k-block: 8 398- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c8__neon_mlal 399 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 400 k-block: 16 401- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c8__neon_mull 402 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 403 k-block: 8 404- name: xnn_qs8_igemm_minmax_rndnu_ukernel_2x16c16__neon_mlal 405 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 406 k-block: 16 407- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8__neon_mlal_lane 408 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 409 k-block: 8 410- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8__neon_mlal_lane_prfm 411 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 412 k-block: 8 413- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8__neon_mull_addw_dup 414 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 415 k-block: 8 416- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2__neon_mlal_dup 417 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 418 k-block: 16 419- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2__neon_mlal_ld1r 420 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 421 k-block: 16 422- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2__neon_mlal_ld2r 423 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 424 k-block: 16 425- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2__neon_mlal_ld4r 426 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 427 k-block: 16 428- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2__neon_mull_dup 429 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 430 k-block: 8 431- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2__neon_mull_ld1r 432 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 433 k-block: 8 434- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2__neon_mull_ld2r 435 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 436 k-block: 8 437- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2__neon_mull_ld4r 438 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 439 k-block: 8 440- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mlal 441 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 442 k-block: 16 443- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c2s4__neon_mull 444 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 445 k-block: 8 446- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4__neon_mlal_dup 447 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 448 k-block: 16 449- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4__neon_mlal_ld1r 450 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 451 k-block: 16 452- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4__neon_mlal_ld2r 453 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 454 k-block: 16 455- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4__neon_mull_dup 456 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 457 k-block: 8 458- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4__neon_mull_ld1r 459 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 460 k-block: 8 461- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4__neon_mull_ld2r 462 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 463 k-block: 8 464- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mlal 465 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 466 k-block: 16 467- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c4s2__neon_mull 468 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 469 k-block: 8 470- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c8__neon_mlal 471 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 472 k-block: 16 473- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c8__neon_mull 474 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 475 k-block: 8 476- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x8c16__neon_mlal 477 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 478 k-block: 16 479- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16__neon_mlal_lane 480 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 481 k-block: 8 482- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16__neon_mlal_lane_prfm 483 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 484 k-block: 8 485- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16__neon_mull_addw_dup 486 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 487 k-block: 8 488- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2__neon_mlal_dup 489 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 490 k-block: 16 491- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2__neon_mlal_ld1r 492 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 493 k-block: 16 494- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2__neon_mlal_ld2r 495 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 496 k-block: 16 497- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2__neon_mlal_ld4r 498 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 499 k-block: 16 500- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2__neon_mull_dup 501 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 502 k-block: 8 503- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2__neon_mull_ld1r 504 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 505 k-block: 8 506- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2__neon_mull_ld2r 507 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 508 k-block: 8 509- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2__neon_mull_ld4r 510 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 511 k-block: 8 512- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal 513 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 514 k-block: 16 515- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mull 516 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 517 k-block: 8 518- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4__neon_mlal_dup 519 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 520 k-block: 16 521- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4__neon_mlal_ld1r 522 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 523 k-block: 16 524- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4__neon_mlal_ld2r 525 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 526 k-block: 16 527- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4__neon_mull_dup 528 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 529 k-block: 8 530- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4__neon_mull_ld1r 531 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 532 k-block: 8 533- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4__neon_mull_ld2r 534 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 535 k-block: 8 536- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal 537 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 538 k-block: 16 539- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mull 540 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 541 k-block: 8 542- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c8__neon_mlal 543 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 544 k-block: 16 545- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c8__neon_mull 546 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 547 k-block: 8 548- name: xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c16__neon_mlal 549 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 550 k-block: 16 551- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8__neon_mlal_lane 552 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 553 k-block: 8 554- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8__neon_mlal_lane_prfm 555 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 556 k-block: 8 557- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8__neon_mull_addw_dup 558 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 559 k-block: 8 560- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2__neon_mlal_dup 561 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 562 k-block: 16 563- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2__neon_mlal_ld1r 564 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 565 k-block: 16 566- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2__neon_mlal_ld2r 567 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 568 k-block: 16 569- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2__neon_mlal_ld4r 570 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 571 k-block: 16 572- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2__neon_mull_dup 573 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 574 k-block: 8 575- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2__neon_mull_ld1r 576 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 577 k-block: 8 578- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2__neon_mull_ld2r 579 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 580 k-block: 8 581- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2__neon_mull_ld4r 582 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 583 k-block: 8 584- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mlal 585 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 586 k-block: 16 587- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c2s4__neon_mull 588 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 589 k-block: 8 590- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__neon_mlal_dup 591 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 592 k-block: 16 593- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__neon_mlal_ld1r 594 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 595 k-block: 16 596- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__neon_mlal_ld2r 597 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 598 k-block: 16 599- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__neon_mull_dup 600 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 601 k-block: 8 602- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__neon_mull_ld1r 603 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 604 k-block: 8 605- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__neon_mull_ld2r 606 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 607 k-block: 8 608- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4__neondot 609 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 610 k-block: 8 611- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mlal 612 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 613 k-block: 16 614- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c4s2__neon_mull 615 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 616 k-block: 8 617- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c8__neon_mlal 618 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 619 k-block: 16 620- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c8__neon_mull 621 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 622 k-block: 8 623- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x8c16__neon_mlal 624 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 625 k-block: 16 626- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16__neon_mlal_lane 627 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 628 k-block: 8 629- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16__neon_mlal_lane_prfm 630 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 631 k-block: 8 632- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16__neon_mull_addw_dup 633 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 634 k-block: 8 635- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2__neon_mlal_dup 636 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 637 k-block: 16 638- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2__neon_mlal_ld1r 639 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 640 k-block: 16 641- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2__neon_mlal_ld2r 642 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 643 k-block: 16 644- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2__neon_mlal_ld4r 645 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 646 k-block: 16 647- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2__neon_mull_dup 648 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 649 k-block: 8 650- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2__neon_mull_ld1r 651 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 652 k-block: 8 653- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2__neon_mull_ld2r 654 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 655 k-block: 8 656- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2__neon_mull_ld4r 657 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 658 k-block: 8 659- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mlal 660 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 661 k-block: 16 662- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c2s4__neon_mull 663 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 664 k-block: 8 665- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__neon_mlal_dup 666 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 667 k-block: 16 668- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__neon_mlal_ld1r 669 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 670 k-block: 16 671- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__neon_mlal_ld2r 672 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 673 k-block: 16 674- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__neon_mull_dup 675 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 676 k-block: 8 677- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__neon_mull_ld1r 678 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 679 k-block: 8 680- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__neon_mull_ld2r 681 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 682 k-block: 8 683- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4__neondot 684 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 685 k-block: 8 686- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mlal 687 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 688 k-block: 16 689- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull 690 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 691 k-block: 8 692- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c8__neon_mlal 693 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 694 k-block: 16 695- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c8__neon_mull 696 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 697 k-block: 8 698- name: xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c16__neon_mlal 699 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 700 k-block: 16 701- name: xnn_qs8_igemm_minmax_rndnu_ukernel_6x8__neon_mlal_lane 702 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 703 k-block: 8 704- name: xnn_qs8_igemm_minmax_rndnu_ukernel_6x8__neon_mlal_lane_prfm 705 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 706 k-block: 8 707- name: xnn_qs8_igemm_minmax_rndnu_ukernel_6x8c4__neondot 708 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 709 k-block: 8 710- name: xnn_qs8_igemm_minmax_rndnu_ukernel_6x16__neon_mlal_lane 711 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 712 k-block: 8 713- name: xnn_qs8_igemm_minmax_rndnu_ukernel_6x16__neon_mlal_lane_prfm 714 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 715 k-block: 8 716- name: xnn_qs8_igemm_minmax_rndnu_ukernel_6x16c4__neondot 717 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 718 k-block: 8 719- name: xnn_qs8_igemm_minmax_rndnu_ukernel_8x8c4__neondot 720 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 721 k-block: 8 722- name: xnn_qs8_igemm_minmax_rndnu_ukernel_8x16c4__neondot 723 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 724 k-block: 8 725# AArch32 JIT assembly 726- name: xnn_generate_qs8_igemm_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_ld64 727 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 728 k-block: 8 729 assembly: true 730- name: xnn_generate_qs8_igemm_rndnu_ukernel_4x8__aarch32_neon_mlal_lane_prfm_ld64 731 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 732 k-block: 8 733 assembly: true 734- name: xnn_generate_qs8_igemm_rndnu_ukernel_4x8c4__aarch32_neondot_ld64 735 init: xnn_init_qs8_conv_minmax_rndnu_neon_params 736 k-block: 8 737 assembly: true 738