Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 14 | pmbaty | 1 | /*===---- arm_fp16.h - ARM FP16 intrinsics ---------------------------------=== |
| 2 | * |
||
| 3 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
||
| 4 | * of this software and associated documentation files (the "Software"), to deal |
||
| 5 | * in the Software without restriction, including without limitation the rights |
||
| 6 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
||
| 7 | * copies of the Software, and to permit persons to whom the Software is |
||
| 8 | * furnished to do so, subject to the following conditions: |
||
| 9 | * |
||
| 10 | * The above copyright notice and this permission notice shall be included in |
||
| 11 | * all copies or substantial portions of the Software. |
||
| 12 | * |
||
| 13 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
||
| 14 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
| 15 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
||
| 16 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
||
| 17 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
||
| 18 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
||
| 19 | * THE SOFTWARE. |
||
| 20 | * |
||
| 21 | *===-----------------------------------------------------------------------=== |
||
| 22 | */ |
||
| 23 | |||
| 24 | #ifndef __ARM_FP16_H |
||
| 25 | #define __ARM_FP16_H |
||
| 26 | |||
| 27 | #include <stdint.h> |
||
| 28 | |||
| 29 | typedef __fp16 float16_t; |
||
| 30 | #define __ai static __inline__ __attribute__((__always_inline__, __nodebug__)) |
||
| 31 | |||
| 32 | #if defined(__aarch64__) |
||
| 33 | #define vabdh_f16(__p0, __p1) __extension__ ({ \ |
||
| 34 | float16_t __ret; \ |
||
| 35 | float16_t __s0 = __p0; \ |
||
| 36 | float16_t __s1 = __p1; \ |
||
| 37 | __ret = (float16_t) __builtin_neon_vabdh_f16(__s0, __s1); \ |
||
| 38 | __ret; \ |
||
| 39 | }) |
||
| 40 | #define vabsh_f16(__p0) __extension__ ({ \ |
||
| 41 | float16_t __ret; \ |
||
| 42 | float16_t __s0 = __p0; \ |
||
| 43 | __ret = (float16_t) __builtin_neon_vabsh_f16(__s0); \ |
||
| 44 | __ret; \ |
||
| 45 | }) |
||
| 46 | #define vaddh_f16(__p0, __p1) __extension__ ({ \ |
||
| 47 | float16_t __ret; \ |
||
| 48 | float16_t __s0 = __p0; \ |
||
| 49 | float16_t __s1 = __p1; \ |
||
| 50 | __ret = (float16_t) __builtin_neon_vaddh_f16(__s0, __s1); \ |
||
| 51 | __ret; \ |
||
| 52 | }) |
||
| 53 | #define vcageh_f16(__p0, __p1) __extension__ ({ \ |
||
| 54 | uint16_t __ret; \ |
||
| 55 | float16_t __s0 = __p0; \ |
||
| 56 | float16_t __s1 = __p1; \ |
||
| 57 | __ret = (uint16_t) __builtin_neon_vcageh_f16(__s0, __s1); \ |
||
| 58 | __ret; \ |
||
| 59 | }) |
||
| 60 | #define vcagth_f16(__p0, __p1) __extension__ ({ \ |
||
| 61 | uint16_t __ret; \ |
||
| 62 | float16_t __s0 = __p0; \ |
||
| 63 | float16_t __s1 = __p1; \ |
||
| 64 | __ret = (uint16_t) __builtin_neon_vcagth_f16(__s0, __s1); \ |
||
| 65 | __ret; \ |
||
| 66 | }) |
||
| 67 | #define vcaleh_f16(__p0, __p1) __extension__ ({ \ |
||
| 68 | uint16_t __ret; \ |
||
| 69 | float16_t __s0 = __p0; \ |
||
| 70 | float16_t __s1 = __p1; \ |
||
| 71 | __ret = (uint16_t) __builtin_neon_vcaleh_f16(__s0, __s1); \ |
||
| 72 | __ret; \ |
||
| 73 | }) |
||
| 74 | #define vcalth_f16(__p0, __p1) __extension__ ({ \ |
||
| 75 | uint16_t __ret; \ |
||
| 76 | float16_t __s0 = __p0; \ |
||
| 77 | float16_t __s1 = __p1; \ |
||
| 78 | __ret = (uint16_t) __builtin_neon_vcalth_f16(__s0, __s1); \ |
||
| 79 | __ret; \ |
||
| 80 | }) |
||
| 81 | #define vceqh_f16(__p0, __p1) __extension__ ({ \ |
||
| 82 | uint16_t __ret; \ |
||
| 83 | float16_t __s0 = __p0; \ |
||
| 84 | float16_t __s1 = __p1; \ |
||
| 85 | __ret = (uint16_t) __builtin_neon_vceqh_f16(__s0, __s1); \ |
||
| 86 | __ret; \ |
||
| 87 | }) |
||
| 88 | #define vceqzh_f16(__p0) __extension__ ({ \ |
||
| 89 | uint16_t __ret; \ |
||
| 90 | float16_t __s0 = __p0; \ |
||
| 91 | __ret = (uint16_t) __builtin_neon_vceqzh_f16(__s0); \ |
||
| 92 | __ret; \ |
||
| 93 | }) |
||
| 94 | #define vcgeh_f16(__p0, __p1) __extension__ ({ \ |
||
| 95 | uint16_t __ret; \ |
||
| 96 | float16_t __s0 = __p0; \ |
||
| 97 | float16_t __s1 = __p1; \ |
||
| 98 | __ret = (uint16_t) __builtin_neon_vcgeh_f16(__s0, __s1); \ |
||
| 99 | __ret; \ |
||
| 100 | }) |
||
| 101 | #define vcgezh_f16(__p0) __extension__ ({ \ |
||
| 102 | uint16_t __ret; \ |
||
| 103 | float16_t __s0 = __p0; \ |
||
| 104 | __ret = (uint16_t) __builtin_neon_vcgezh_f16(__s0); \ |
||
| 105 | __ret; \ |
||
| 106 | }) |
||
| 107 | #define vcgth_f16(__p0, __p1) __extension__ ({ \ |
||
| 108 | uint16_t __ret; \ |
||
| 109 | float16_t __s0 = __p0; \ |
||
| 110 | float16_t __s1 = __p1; \ |
||
| 111 | __ret = (uint16_t) __builtin_neon_vcgth_f16(__s0, __s1); \ |
||
| 112 | __ret; \ |
||
| 113 | }) |
||
| 114 | #define vcgtzh_f16(__p0) __extension__ ({ \ |
||
| 115 | uint16_t __ret; \ |
||
| 116 | float16_t __s0 = __p0; \ |
||
| 117 | __ret = (uint16_t) __builtin_neon_vcgtzh_f16(__s0); \ |
||
| 118 | __ret; \ |
||
| 119 | }) |
||
| 120 | #define vcleh_f16(__p0, __p1) __extension__ ({ \ |
||
| 121 | uint16_t __ret; \ |
||
| 122 | float16_t __s0 = __p0; \ |
||
| 123 | float16_t __s1 = __p1; \ |
||
| 124 | __ret = (uint16_t) __builtin_neon_vcleh_f16(__s0, __s1); \ |
||
| 125 | __ret; \ |
||
| 126 | }) |
||
| 127 | #define vclezh_f16(__p0) __extension__ ({ \ |
||
| 128 | uint16_t __ret; \ |
||
| 129 | float16_t __s0 = __p0; \ |
||
| 130 | __ret = (uint16_t) __builtin_neon_vclezh_f16(__s0); \ |
||
| 131 | __ret; \ |
||
| 132 | }) |
||
| 133 | #define vclth_f16(__p0, __p1) __extension__ ({ \ |
||
| 134 | uint16_t __ret; \ |
||
| 135 | float16_t __s0 = __p0; \ |
||
| 136 | float16_t __s1 = __p1; \ |
||
| 137 | __ret = (uint16_t) __builtin_neon_vclth_f16(__s0, __s1); \ |
||
| 138 | __ret; \ |
||
| 139 | }) |
||
| 140 | #define vcltzh_f16(__p0) __extension__ ({ \ |
||
| 141 | uint16_t __ret; \ |
||
| 142 | float16_t __s0 = __p0; \ |
||
| 143 | __ret = (uint16_t) __builtin_neon_vcltzh_f16(__s0); \ |
||
| 144 | __ret; \ |
||
| 145 | }) |
||
| 146 | #define vcvth_n_s16_f16(__p0, __p1) __extension__ ({ \ |
||
| 147 | int16_t __ret; \ |
||
| 148 | float16_t __s0 = __p0; \ |
||
| 149 | __ret = (int16_t) __builtin_neon_vcvth_n_s16_f16(__s0, __p1); \ |
||
| 150 | __ret; \ |
||
| 151 | }) |
||
| 152 | #define vcvth_n_s32_f16(__p0, __p1) __extension__ ({ \ |
||
| 153 | int32_t __ret; \ |
||
| 154 | float16_t __s0 = __p0; \ |
||
| 155 | __ret = (int32_t) __builtin_neon_vcvth_n_s32_f16(__s0, __p1); \ |
||
| 156 | __ret; \ |
||
| 157 | }) |
||
| 158 | #define vcvth_n_s64_f16(__p0, __p1) __extension__ ({ \ |
||
| 159 | int64_t __ret; \ |
||
| 160 | float16_t __s0 = __p0; \ |
||
| 161 | __ret = (int64_t) __builtin_neon_vcvth_n_s64_f16(__s0, __p1); \ |
||
| 162 | __ret; \ |
||
| 163 | }) |
||
| 164 | #define vcvth_n_u16_f16(__p0, __p1) __extension__ ({ \ |
||
| 165 | uint16_t __ret; \ |
||
| 166 | float16_t __s0 = __p0; \ |
||
| 167 | __ret = (uint16_t) __builtin_neon_vcvth_n_u16_f16(__s0, __p1); \ |
||
| 168 | __ret; \ |
||
| 169 | }) |
||
| 170 | #define vcvth_n_u32_f16(__p0, __p1) __extension__ ({ \ |
||
| 171 | uint32_t __ret; \ |
||
| 172 | float16_t __s0 = __p0; \ |
||
| 173 | __ret = (uint32_t) __builtin_neon_vcvth_n_u32_f16(__s0, __p1); \ |
||
| 174 | __ret; \ |
||
| 175 | }) |
||
| 176 | #define vcvth_n_u64_f16(__p0, __p1) __extension__ ({ \ |
||
| 177 | uint64_t __ret; \ |
||
| 178 | float16_t __s0 = __p0; \ |
||
| 179 | __ret = (uint64_t) __builtin_neon_vcvth_n_u64_f16(__s0, __p1); \ |
||
| 180 | __ret; \ |
||
| 181 | }) |
||
| 182 | #define vcvth_s16_f16(__p0) __extension__ ({ \ |
||
| 183 | int16_t __ret; \ |
||
| 184 | float16_t __s0 = __p0; \ |
||
| 185 | __ret = (int16_t) __builtin_neon_vcvth_s16_f16(__s0); \ |
||
| 186 | __ret; \ |
||
| 187 | }) |
||
| 188 | #define vcvth_s32_f16(__p0) __extension__ ({ \ |
||
| 189 | int32_t __ret; \ |
||
| 190 | float16_t __s0 = __p0; \ |
||
| 191 | __ret = (int32_t) __builtin_neon_vcvth_s32_f16(__s0); \ |
||
| 192 | __ret; \ |
||
| 193 | }) |
||
| 194 | #define vcvth_s64_f16(__p0) __extension__ ({ \ |
||
| 195 | int64_t __ret; \ |
||
| 196 | float16_t __s0 = __p0; \ |
||
| 197 | __ret = (int64_t) __builtin_neon_vcvth_s64_f16(__s0); \ |
||
| 198 | __ret; \ |
||
| 199 | }) |
||
| 200 | #define vcvth_u16_f16(__p0) __extension__ ({ \ |
||
| 201 | uint16_t __ret; \ |
||
| 202 | float16_t __s0 = __p0; \ |
||
| 203 | __ret = (uint16_t) __builtin_neon_vcvth_u16_f16(__s0); \ |
||
| 204 | __ret; \ |
||
| 205 | }) |
||
| 206 | #define vcvth_u32_f16(__p0) __extension__ ({ \ |
||
| 207 | uint32_t __ret; \ |
||
| 208 | float16_t __s0 = __p0; \ |
||
| 209 | __ret = (uint32_t) __builtin_neon_vcvth_u32_f16(__s0); \ |
||
| 210 | __ret; \ |
||
| 211 | }) |
||
| 212 | #define vcvth_u64_f16(__p0) __extension__ ({ \ |
||
| 213 | uint64_t __ret; \ |
||
| 214 | float16_t __s0 = __p0; \ |
||
| 215 | __ret = (uint64_t) __builtin_neon_vcvth_u64_f16(__s0); \ |
||
| 216 | __ret; \ |
||
| 217 | }) |
||
| 218 | #define vcvtah_s16_f16(__p0) __extension__ ({ \ |
||
| 219 | int16_t __ret; \ |
||
| 220 | float16_t __s0 = __p0; \ |
||
| 221 | __ret = (int16_t) __builtin_neon_vcvtah_s16_f16(__s0); \ |
||
| 222 | __ret; \ |
||
| 223 | }) |
||
| 224 | #define vcvtah_s32_f16(__p0) __extension__ ({ \ |
||
| 225 | int32_t __ret; \ |
||
| 226 | float16_t __s0 = __p0; \ |
||
| 227 | __ret = (int32_t) __builtin_neon_vcvtah_s32_f16(__s0); \ |
||
| 228 | __ret; \ |
||
| 229 | }) |
||
| 230 | #define vcvtah_s64_f16(__p0) __extension__ ({ \ |
||
| 231 | int64_t __ret; \ |
||
| 232 | float16_t __s0 = __p0; \ |
||
| 233 | __ret = (int64_t) __builtin_neon_vcvtah_s64_f16(__s0); \ |
||
| 234 | __ret; \ |
||
| 235 | }) |
||
| 236 | #define vcvtah_u16_f16(__p0) __extension__ ({ \ |
||
| 237 | uint16_t __ret; \ |
||
| 238 | float16_t __s0 = __p0; \ |
||
| 239 | __ret = (uint16_t) __builtin_neon_vcvtah_u16_f16(__s0); \ |
||
| 240 | __ret; \ |
||
| 241 | }) |
||
| 242 | #define vcvtah_u32_f16(__p0) __extension__ ({ \ |
||
| 243 | uint32_t __ret; \ |
||
| 244 | float16_t __s0 = __p0; \ |
||
| 245 | __ret = (uint32_t) __builtin_neon_vcvtah_u32_f16(__s0); \ |
||
| 246 | __ret; \ |
||
| 247 | }) |
||
| 248 | #define vcvtah_u64_f16(__p0) __extension__ ({ \ |
||
| 249 | uint64_t __ret; \ |
||
| 250 | float16_t __s0 = __p0; \ |
||
| 251 | __ret = (uint64_t) __builtin_neon_vcvtah_u64_f16(__s0); \ |
||
| 252 | __ret; \ |
||
| 253 | }) |
||
| 254 | #define vcvth_f16_u16(__p0) __extension__ ({ \ |
||
| 255 | float16_t __ret; \ |
||
| 256 | uint16_t __s0 = __p0; \ |
||
| 257 | __ret = (float16_t) __builtin_neon_vcvth_f16_u16(__s0); \ |
||
| 258 | __ret; \ |
||
| 259 | }) |
||
| 260 | #define vcvth_f16_s16(__p0) __extension__ ({ \ |
||
| 261 | float16_t __ret; \ |
||
| 262 | int16_t __s0 = __p0; \ |
||
| 263 | __ret = (float16_t) __builtin_neon_vcvth_f16_s16(__s0); \ |
||
| 264 | __ret; \ |
||
| 265 | }) |
||
| 266 | #define vcvth_f16_u32(__p0) __extension__ ({ \ |
||
| 267 | float16_t __ret; \ |
||
| 268 | uint32_t __s0 = __p0; \ |
||
| 269 | __ret = (float16_t) __builtin_neon_vcvth_f16_u32(__s0); \ |
||
| 270 | __ret; \ |
||
| 271 | }) |
||
| 272 | #define vcvth_f16_s32(__p0) __extension__ ({ \ |
||
| 273 | float16_t __ret; \ |
||
| 274 | int32_t __s0 = __p0; \ |
||
| 275 | __ret = (float16_t) __builtin_neon_vcvth_f16_s32(__s0); \ |
||
| 276 | __ret; \ |
||
| 277 | }) |
||
| 278 | #define vcvth_f16_u64(__p0) __extension__ ({ \ |
||
| 279 | float16_t __ret; \ |
||
| 280 | uint64_t __s0 = __p0; \ |
||
| 281 | __ret = (float16_t) __builtin_neon_vcvth_f16_u64(__s0); \ |
||
| 282 | __ret; \ |
||
| 283 | }) |
||
| 284 | #define vcvth_f16_s64(__p0) __extension__ ({ \ |
||
| 285 | float16_t __ret; \ |
||
| 286 | int64_t __s0 = __p0; \ |
||
| 287 | __ret = (float16_t) __builtin_neon_vcvth_f16_s64(__s0); \ |
||
| 288 | __ret; \ |
||
| 289 | }) |
||
| 290 | #define vcvth_n_f16_u32(__p0, __p1) __extension__ ({ \ |
||
| 291 | float16_t __ret; \ |
||
| 292 | uint32_t __s0 = __p0; \ |
||
| 293 | __ret = (float16_t) __builtin_neon_vcvth_n_f16_u32(__s0, __p1); \ |
||
| 294 | __ret; \ |
||
| 295 | }) |
||
| 296 | #define vcvth_n_f16_s32(__p0, __p1) __extension__ ({ \ |
||
| 297 | float16_t __ret; \ |
||
| 298 | int32_t __s0 = __p0; \ |
||
| 299 | __ret = (float16_t) __builtin_neon_vcvth_n_f16_s32(__s0, __p1); \ |
||
| 300 | __ret; \ |
||
| 301 | }) |
||
| 302 | #define vcvth_n_f16_u64(__p0, __p1) __extension__ ({ \ |
||
| 303 | float16_t __ret; \ |
||
| 304 | uint64_t __s0 = __p0; \ |
||
| 305 | __ret = (float16_t) __builtin_neon_vcvth_n_f16_u64(__s0, __p1); \ |
||
| 306 | __ret; \ |
||
| 307 | }) |
||
| 308 | #define vcvth_n_f16_s64(__p0, __p1) __extension__ ({ \ |
||
| 309 | float16_t __ret; \ |
||
| 310 | int64_t __s0 = __p0; \ |
||
| 311 | __ret = (float16_t) __builtin_neon_vcvth_n_f16_s64(__s0, __p1); \ |
||
| 312 | __ret; \ |
||
| 313 | }) |
||
| 314 | #define vcvth_n_f16_u16(__p0, __p1) __extension__ ({ \ |
||
| 315 | float16_t __ret; \ |
||
| 316 | uint16_t __s0 = __p0; \ |
||
| 317 | __ret = (float16_t) __builtin_neon_vcvth_n_f16_u16(__s0, __p1); \ |
||
| 318 | __ret; \ |
||
| 319 | }) |
||
| 320 | #define vcvth_n_f16_s16(__p0, __p1) __extension__ ({ \ |
||
| 321 | float16_t __ret; \ |
||
| 322 | int16_t __s0 = __p0; \ |
||
| 323 | __ret = (float16_t) __builtin_neon_vcvth_n_f16_s16(__s0, __p1); \ |
||
| 324 | __ret; \ |
||
| 325 | }) |
||
| 326 | #define vcvtmh_s16_f16(__p0) __extension__ ({ \ |
||
| 327 | int16_t __ret; \ |
||
| 328 | float16_t __s0 = __p0; \ |
||
| 329 | __ret = (int16_t) __builtin_neon_vcvtmh_s16_f16(__s0); \ |
||
| 330 | __ret; \ |
||
| 331 | }) |
||
| 332 | #define vcvtmh_s32_f16(__p0) __extension__ ({ \ |
||
| 333 | int32_t __ret; \ |
||
| 334 | float16_t __s0 = __p0; \ |
||
| 335 | __ret = (int32_t) __builtin_neon_vcvtmh_s32_f16(__s0); \ |
||
| 336 | __ret; \ |
||
| 337 | }) |
||
| 338 | #define vcvtmh_s64_f16(__p0) __extension__ ({ \ |
||
| 339 | int64_t __ret; \ |
||
| 340 | float16_t __s0 = __p0; \ |
||
| 341 | __ret = (int64_t) __builtin_neon_vcvtmh_s64_f16(__s0); \ |
||
| 342 | __ret; \ |
||
| 343 | }) |
||
| 344 | #define vcvtmh_u16_f16(__p0) __extension__ ({ \ |
||
| 345 | uint16_t __ret; \ |
||
| 346 | float16_t __s0 = __p0; \ |
||
| 347 | __ret = (uint16_t) __builtin_neon_vcvtmh_u16_f16(__s0); \ |
||
| 348 | __ret; \ |
||
| 349 | }) |
||
| 350 | #define vcvtmh_u32_f16(__p0) __extension__ ({ \ |
||
| 351 | uint32_t __ret; \ |
||
| 352 | float16_t __s0 = __p0; \ |
||
| 353 | __ret = (uint32_t) __builtin_neon_vcvtmh_u32_f16(__s0); \ |
||
| 354 | __ret; \ |
||
| 355 | }) |
||
| 356 | #define vcvtmh_u64_f16(__p0) __extension__ ({ \ |
||
| 357 | uint64_t __ret; \ |
||
| 358 | float16_t __s0 = __p0; \ |
||
| 359 | __ret = (uint64_t) __builtin_neon_vcvtmh_u64_f16(__s0); \ |
||
| 360 | __ret; \ |
||
| 361 | }) |
||
| 362 | #define vcvtnh_s16_f16(__p0) __extension__ ({ \ |
||
| 363 | int16_t __ret; \ |
||
| 364 | float16_t __s0 = __p0; \ |
||
| 365 | __ret = (int16_t) __builtin_neon_vcvtnh_s16_f16(__s0); \ |
||
| 366 | __ret; \ |
||
| 367 | }) |
||
| 368 | #define vcvtnh_s32_f16(__p0) __extension__ ({ \ |
||
| 369 | int32_t __ret; \ |
||
| 370 | float16_t __s0 = __p0; \ |
||
| 371 | __ret = (int32_t) __builtin_neon_vcvtnh_s32_f16(__s0); \ |
||
| 372 | __ret; \ |
||
| 373 | }) |
||
| 374 | #define vcvtnh_s64_f16(__p0) __extension__ ({ \ |
||
| 375 | int64_t __ret; \ |
||
| 376 | float16_t __s0 = __p0; \ |
||
| 377 | __ret = (int64_t) __builtin_neon_vcvtnh_s64_f16(__s0); \ |
||
| 378 | __ret; \ |
||
| 379 | }) |
||
| 380 | #define vcvtnh_u16_f16(__p0) __extension__ ({ \ |
||
| 381 | uint16_t __ret; \ |
||
| 382 | float16_t __s0 = __p0; \ |
||
| 383 | __ret = (uint16_t) __builtin_neon_vcvtnh_u16_f16(__s0); \ |
||
| 384 | __ret; \ |
||
| 385 | }) |
||
| 386 | #define vcvtnh_u32_f16(__p0) __extension__ ({ \ |
||
| 387 | uint32_t __ret; \ |
||
| 388 | float16_t __s0 = __p0; \ |
||
| 389 | __ret = (uint32_t) __builtin_neon_vcvtnh_u32_f16(__s0); \ |
||
| 390 | __ret; \ |
||
| 391 | }) |
||
| 392 | #define vcvtnh_u64_f16(__p0) __extension__ ({ \ |
||
| 393 | uint64_t __ret; \ |
||
| 394 | float16_t __s0 = __p0; \ |
||
| 395 | __ret = (uint64_t) __builtin_neon_vcvtnh_u64_f16(__s0); \ |
||
| 396 | __ret; \ |
||
| 397 | }) |
||
| 398 | #define vcvtph_s16_f16(__p0) __extension__ ({ \ |
||
| 399 | int16_t __ret; \ |
||
| 400 | float16_t __s0 = __p0; \ |
||
| 401 | __ret = (int16_t) __builtin_neon_vcvtph_s16_f16(__s0); \ |
||
| 402 | __ret; \ |
||
| 403 | }) |
||
| 404 | #define vcvtph_s32_f16(__p0) __extension__ ({ \ |
||
| 405 | int32_t __ret; \ |
||
| 406 | float16_t __s0 = __p0; \ |
||
| 407 | __ret = (int32_t) __builtin_neon_vcvtph_s32_f16(__s0); \ |
||
| 408 | __ret; \ |
||
| 409 | }) |
||
| 410 | #define vcvtph_s64_f16(__p0) __extension__ ({ \ |
||
| 411 | int64_t __ret; \ |
||
| 412 | float16_t __s0 = __p0; \ |
||
| 413 | __ret = (int64_t) __builtin_neon_vcvtph_s64_f16(__s0); \ |
||
| 414 | __ret; \ |
||
| 415 | }) |
||
| 416 | #define vcvtph_u16_f16(__p0) __extension__ ({ \ |
||
| 417 | uint16_t __ret; \ |
||
| 418 | float16_t __s0 = __p0; \ |
||
| 419 | __ret = (uint16_t) __builtin_neon_vcvtph_u16_f16(__s0); \ |
||
| 420 | __ret; \ |
||
| 421 | }) |
||
| 422 | #define vcvtph_u32_f16(__p0) __extension__ ({ \ |
||
| 423 | uint32_t __ret; \ |
||
| 424 | float16_t __s0 = __p0; \ |
||
| 425 | __ret = (uint32_t) __builtin_neon_vcvtph_u32_f16(__s0); \ |
||
| 426 | __ret; \ |
||
| 427 | }) |
||
| 428 | #define vcvtph_u64_f16(__p0) __extension__ ({ \ |
||
| 429 | uint64_t __ret; \ |
||
| 430 | float16_t __s0 = __p0; \ |
||
| 431 | __ret = (uint64_t) __builtin_neon_vcvtph_u64_f16(__s0); \ |
||
| 432 | __ret; \ |
||
| 433 | }) |
||
| 434 | #define vdivh_f16(__p0, __p1) __extension__ ({ \ |
||
| 435 | float16_t __ret; \ |
||
| 436 | float16_t __s0 = __p0; \ |
||
| 437 | float16_t __s1 = __p1; \ |
||
| 438 | __ret = (float16_t) __builtin_neon_vdivh_f16(__s0, __s1); \ |
||
| 439 | __ret; \ |
||
| 440 | }) |
||
| 441 | #define vfmah_f16(__p0, __p1, __p2) __extension__ ({ \ |
||
| 442 | float16_t __ret; \ |
||
| 443 | float16_t __s0 = __p0; \ |
||
| 444 | float16_t __s1 = __p1; \ |
||
| 445 | float16_t __s2 = __p2; \ |
||
| 446 | __ret = (float16_t) __builtin_neon_vfmah_f16(__s0, __s1, __s2); \ |
||
| 447 | __ret; \ |
||
| 448 | }) |
||
| 449 | #define vfmsh_f16(__p0, __p1, __p2) __extension__ ({ \ |
||
| 450 | float16_t __ret; \ |
||
| 451 | float16_t __s0 = __p0; \ |
||
| 452 | float16_t __s1 = __p1; \ |
||
| 453 | float16_t __s2 = __p2; \ |
||
| 454 | __ret = (float16_t) __builtin_neon_vfmsh_f16(__s0, __s1, __s2); \ |
||
| 455 | __ret; \ |
||
| 456 | }) |
||
| 457 | #define vmaxh_f16(__p0, __p1) __extension__ ({ \ |
||
| 458 | float16_t __ret; \ |
||
| 459 | float16_t __s0 = __p0; \ |
||
| 460 | float16_t __s1 = __p1; \ |
||
| 461 | __ret = (float16_t) __builtin_neon_vmaxh_f16(__s0, __s1); \ |
||
| 462 | __ret; \ |
||
| 463 | }) |
||
| 464 | #define vmaxnmh_f16(__p0, __p1) __extension__ ({ \ |
||
| 465 | float16_t __ret; \ |
||
| 466 | float16_t __s0 = __p0; \ |
||
| 467 | float16_t __s1 = __p1; \ |
||
| 468 | __ret = (float16_t) __builtin_neon_vmaxnmh_f16(__s0, __s1); \ |
||
| 469 | __ret; \ |
||
| 470 | }) |
||
| 471 | #define vminh_f16(__p0, __p1) __extension__ ({ \ |
||
| 472 | float16_t __ret; \ |
||
| 473 | float16_t __s0 = __p0; \ |
||
| 474 | float16_t __s1 = __p1; \ |
||
| 475 | __ret = (float16_t) __builtin_neon_vminh_f16(__s0, __s1); \ |
||
| 476 | __ret; \ |
||
| 477 | }) |
||
| 478 | #define vminnmh_f16(__p0, __p1) __extension__ ({ \ |
||
| 479 | float16_t __ret; \ |
||
| 480 | float16_t __s0 = __p0; \ |
||
| 481 | float16_t __s1 = __p1; \ |
||
| 482 | __ret = (float16_t) __builtin_neon_vminnmh_f16(__s0, __s1); \ |
||
| 483 | __ret; \ |
||
| 484 | }) |
||
| 485 | #define vmulh_f16(__p0, __p1) __extension__ ({ \ |
||
| 486 | float16_t __ret; \ |
||
| 487 | float16_t __s0 = __p0; \ |
||
| 488 | float16_t __s1 = __p1; \ |
||
| 489 | __ret = (float16_t) __builtin_neon_vmulh_f16(__s0, __s1); \ |
||
| 490 | __ret; \ |
||
| 491 | }) |
||
| 492 | #define vmulxh_f16(__p0, __p1) __extension__ ({ \ |
||
| 493 | float16_t __ret; \ |
||
| 494 | float16_t __s0 = __p0; \ |
||
| 495 | float16_t __s1 = __p1; \ |
||
| 496 | __ret = (float16_t) __builtin_neon_vmulxh_f16(__s0, __s1); \ |
||
| 497 | __ret; \ |
||
| 498 | }) |
||
| 499 | #define vnegh_f16(__p0) __extension__ ({ \ |
||
| 500 | float16_t __ret; \ |
||
| 501 | float16_t __s0 = __p0; \ |
||
| 502 | __ret = (float16_t) __builtin_neon_vnegh_f16(__s0); \ |
||
| 503 | __ret; \ |
||
| 504 | }) |
||
| 505 | #define vrecpeh_f16(__p0) __extension__ ({ \ |
||
| 506 | float16_t __ret; \ |
||
| 507 | float16_t __s0 = __p0; \ |
||
| 508 | __ret = (float16_t) __builtin_neon_vrecpeh_f16(__s0); \ |
||
| 509 | __ret; \ |
||
| 510 | }) |
||
| 511 | #define vrecpsh_f16(__p0, __p1) __extension__ ({ \ |
||
| 512 | float16_t __ret; \ |
||
| 513 | float16_t __s0 = __p0; \ |
||
| 514 | float16_t __s1 = __p1; \ |
||
| 515 | __ret = (float16_t) __builtin_neon_vrecpsh_f16(__s0, __s1); \ |
||
| 516 | __ret; \ |
||
| 517 | }) |
||
| 518 | #define vrecpxh_f16(__p0) __extension__ ({ \ |
||
| 519 | float16_t __ret; \ |
||
| 520 | float16_t __s0 = __p0; \ |
||
| 521 | __ret = (float16_t) __builtin_neon_vrecpxh_f16(__s0); \ |
||
| 522 | __ret; \ |
||
| 523 | }) |
||
| 524 | #define vrndh_f16(__p0) __extension__ ({ \ |
||
| 525 | float16_t __ret; \ |
||
| 526 | float16_t __s0 = __p0; \ |
||
| 527 | __ret = (float16_t) __builtin_neon_vrndh_f16(__s0); \ |
||
| 528 | __ret; \ |
||
| 529 | }) |
||
| 530 | #define vrndah_f16(__p0) __extension__ ({ \ |
||
| 531 | float16_t __ret; \ |
||
| 532 | float16_t __s0 = __p0; \ |
||
| 533 | __ret = (float16_t) __builtin_neon_vrndah_f16(__s0); \ |
||
| 534 | __ret; \ |
||
| 535 | }) |
||
| 536 | #define vrndih_f16(__p0) __extension__ ({ \ |
||
| 537 | float16_t __ret; \ |
||
| 538 | float16_t __s0 = __p0; \ |
||
| 539 | __ret = (float16_t) __builtin_neon_vrndih_f16(__s0); \ |
||
| 540 | __ret; \ |
||
| 541 | }) |
||
| 542 | #define vrndmh_f16(__p0) __extension__ ({ \ |
||
| 543 | float16_t __ret; \ |
||
| 544 | float16_t __s0 = __p0; \ |
||
| 545 | __ret = (float16_t) __builtin_neon_vrndmh_f16(__s0); \ |
||
| 546 | __ret; \ |
||
| 547 | }) |
||
| 548 | #define vrndnh_f16(__p0) __extension__ ({ \ |
||
| 549 | float16_t __ret; \ |
||
| 550 | float16_t __s0 = __p0; \ |
||
| 551 | __ret = (float16_t) __builtin_neon_vrndnh_f16(__s0); \ |
||
| 552 | __ret; \ |
||
| 553 | }) |
||
| 554 | #define vrndph_f16(__p0) __extension__ ({ \ |
||
| 555 | float16_t __ret; \ |
||
| 556 | float16_t __s0 = __p0; \ |
||
| 557 | __ret = (float16_t) __builtin_neon_vrndph_f16(__s0); \ |
||
| 558 | __ret; \ |
||
| 559 | }) |
||
| 560 | #define vrndxh_f16(__p0) __extension__ ({ \ |
||
| 561 | float16_t __ret; \ |
||
| 562 | float16_t __s0 = __p0; \ |
||
| 563 | __ret = (float16_t) __builtin_neon_vrndxh_f16(__s0); \ |
||
| 564 | __ret; \ |
||
| 565 | }) |
||
| 566 | #define vrsqrteh_f16(__p0) __extension__ ({ \ |
||
| 567 | float16_t __ret; \ |
||
| 568 | float16_t __s0 = __p0; \ |
||
| 569 | __ret = (float16_t) __builtin_neon_vrsqrteh_f16(__s0); \ |
||
| 570 | __ret; \ |
||
| 571 | }) |
||
| 572 | #define vrsqrtsh_f16(__p0, __p1) __extension__ ({ \ |
||
| 573 | float16_t __ret; \ |
||
| 574 | float16_t __s0 = __p0; \ |
||
| 575 | float16_t __s1 = __p1; \ |
||
| 576 | __ret = (float16_t) __builtin_neon_vrsqrtsh_f16(__s0, __s1); \ |
||
| 577 | __ret; \ |
||
| 578 | }) |
||
| 579 | #define vsqrth_f16(__p0) __extension__ ({ \ |
||
| 580 | float16_t __ret; \ |
||
| 581 | float16_t __s0 = __p0; \ |
||
| 582 | __ret = (float16_t) __builtin_neon_vsqrth_f16(__s0); \ |
||
| 583 | __ret; \ |
||
| 584 | }) |
||
| 585 | #define vsubh_f16(__p0, __p1) __extension__ ({ \ |
||
| 586 | float16_t __ret; \ |
||
| 587 | float16_t __s0 = __p0; \ |
||
| 588 | float16_t __s1 = __p1; \ |
||
| 589 | __ret = (float16_t) __builtin_neon_vsubh_f16(__s0, __s1); \ |
||
| 590 | __ret; \ |
||
| 591 | }) |
||
| 592 | #endif |
||
| 593 | |||
| 594 | #undef __ai |
||
| 595 | |||
| 596 | #endif /* __ARM_FP16_H */ |