Richard Henderson | bd277fa | 2012-09-18 21:59:44 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Loongson Multimedia Instruction emulation helpers for QEMU. |
| 3 | * |
| 4 | * Copyright (c) 2011 Richard Henderson <rth@twiddle.net> |
| 5 | * |
| 6 | * This library is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU Lesser General Public |
| 8 | * License as published by the Free Software Foundation; either |
| 9 | * version 2 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * This library is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * Lesser General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU Lesser General Public |
| 17 | * License along with this library; if not, see <http://www.gnu.org/licenses/>. |
| 18 | */ |
| 19 | |
| 20 | #include "cpu.h" |
Richard Henderson | 2ef6175 | 2014-04-07 22:31:41 -0700 | [diff] [blame] | 21 | #include "exec/helper-proto.h" |
Richard Henderson | bd277fa | 2012-09-18 21:59:44 -0700 | [diff] [blame] | 22 | |
| 23 | /* If the byte ordering doesn't matter, i.e. all columns are treated |
| 24 | identically, then this union can be used directly. If byte ordering |
| 25 | does matter, we generally ignore dumping to memory. */ |
| 26 | typedef union { |
| 27 | uint8_t ub[8]; |
| 28 | int8_t sb[8]; |
| 29 | uint16_t uh[4]; |
| 30 | int16_t sh[4]; |
| 31 | uint32_t uw[2]; |
| 32 | int32_t sw[2]; |
| 33 | uint64_t d; |
| 34 | } LMIValue; |
| 35 | |
| 36 | /* Some byte ordering issues can be mitigated by XORing in the following. */ |
| 37 | #ifdef HOST_WORDS_BIGENDIAN |
| 38 | # define BYTE_ORDER_XOR(N) N |
| 39 | #else |
| 40 | # define BYTE_ORDER_XOR(N) 0 |
| 41 | #endif |
| 42 | |
| 43 | #define SATSB(x) (x < -0x80 ? -0x80 : x > 0x7f ? 0x7f : x) |
| 44 | #define SATUB(x) (x > 0xff ? 0xff : x) |
| 45 | |
| 46 | #define SATSH(x) (x < -0x8000 ? -0x8000 : x > 0x7fff ? 0x7fff : x) |
| 47 | #define SATUH(x) (x > 0xffff ? 0xffff : x) |
| 48 | |
| 49 | #define SATSW(x) \ |
| 50 | (x < -0x80000000ll ? -0x80000000ll : x > 0x7fffffff ? 0x7fffffff : x) |
| 51 | #define SATUW(x) (x > 0xffffffffull ? 0xffffffffull : x) |
| 52 | |
| 53 | uint64_t helper_paddsb(uint64_t fs, uint64_t ft) |
| 54 | { |
| 55 | LMIValue vs, vt; |
| 56 | unsigned int i; |
| 57 | |
| 58 | vs.d = fs; |
| 59 | vt.d = ft; |
| 60 | for (i = 0; i < 8; ++i) { |
| 61 | int r = vs.sb[i] + vt.sb[i]; |
| 62 | vs.sb[i] = SATSB(r); |
| 63 | } |
| 64 | return vs.d; |
| 65 | } |
| 66 | |
| 67 | uint64_t helper_paddusb(uint64_t fs, uint64_t ft) |
| 68 | { |
| 69 | LMIValue vs, vt; |
| 70 | unsigned int i; |
| 71 | |
| 72 | vs.d = fs; |
| 73 | vt.d = ft; |
| 74 | for (i = 0; i < 8; ++i) { |
| 75 | int r = vs.ub[i] + vt.ub[i]; |
| 76 | vs.ub[i] = SATUB(r); |
| 77 | } |
| 78 | return vs.d; |
| 79 | } |
| 80 | |
| 81 | uint64_t helper_paddsh(uint64_t fs, uint64_t ft) |
| 82 | { |
| 83 | LMIValue vs, vt; |
| 84 | unsigned int i; |
| 85 | |
| 86 | vs.d = fs; |
| 87 | vt.d = ft; |
| 88 | for (i = 0; i < 4; ++i) { |
| 89 | int r = vs.sh[i] + vt.sh[i]; |
| 90 | vs.sh[i] = SATSH(r); |
| 91 | } |
| 92 | return vs.d; |
| 93 | } |
| 94 | |
| 95 | uint64_t helper_paddush(uint64_t fs, uint64_t ft) |
| 96 | { |
| 97 | LMIValue vs, vt; |
| 98 | unsigned int i; |
| 99 | |
| 100 | vs.d = fs; |
| 101 | vt.d = ft; |
| 102 | for (i = 0; i < 4; ++i) { |
| 103 | int r = vs.uh[i] + vt.uh[i]; |
| 104 | vs.uh[i] = SATUH(r); |
| 105 | } |
| 106 | return vs.d; |
| 107 | } |
| 108 | |
| 109 | uint64_t helper_paddb(uint64_t fs, uint64_t ft) |
| 110 | { |
| 111 | LMIValue vs, vt; |
| 112 | unsigned int i; |
| 113 | |
| 114 | vs.d = fs; |
| 115 | vt.d = ft; |
| 116 | for (i = 0; i < 8; ++i) { |
| 117 | vs.ub[i] += vt.ub[i]; |
| 118 | } |
| 119 | return vs.d; |
| 120 | } |
| 121 | |
| 122 | uint64_t helper_paddh(uint64_t fs, uint64_t ft) |
| 123 | { |
| 124 | LMIValue vs, vt; |
| 125 | unsigned int i; |
| 126 | |
| 127 | vs.d = fs; |
| 128 | vt.d = ft; |
| 129 | for (i = 0; i < 4; ++i) { |
| 130 | vs.uh[i] += vt.uh[i]; |
| 131 | } |
| 132 | return vs.d; |
| 133 | } |
| 134 | |
| 135 | uint64_t helper_paddw(uint64_t fs, uint64_t ft) |
| 136 | { |
| 137 | LMIValue vs, vt; |
| 138 | unsigned int i; |
| 139 | |
| 140 | vs.d = fs; |
| 141 | vt.d = ft; |
| 142 | for (i = 0; i < 2; ++i) { |
| 143 | vs.uw[i] += vt.uw[i]; |
| 144 | } |
| 145 | return vs.d; |
| 146 | } |
| 147 | |
| 148 | uint64_t helper_psubsb(uint64_t fs, uint64_t ft) |
| 149 | { |
| 150 | LMIValue vs, vt; |
| 151 | unsigned int i; |
| 152 | |
| 153 | vs.d = fs; |
| 154 | vt.d = ft; |
| 155 | for (i = 0; i < 8; ++i) { |
| 156 | int r = vs.sb[i] - vt.sb[i]; |
| 157 | vs.sb[i] = SATSB(r); |
| 158 | } |
| 159 | return vs.d; |
| 160 | } |
| 161 | |
| 162 | uint64_t helper_psubusb(uint64_t fs, uint64_t ft) |
| 163 | { |
| 164 | LMIValue vs, vt; |
| 165 | unsigned int i; |
| 166 | |
| 167 | vs.d = fs; |
| 168 | vt.d = ft; |
| 169 | for (i = 0; i < 8; ++i) { |
| 170 | int r = vs.ub[i] - vt.ub[i]; |
| 171 | vs.ub[i] = SATUB(r); |
| 172 | } |
| 173 | return vs.d; |
| 174 | } |
| 175 | |
| 176 | uint64_t helper_psubsh(uint64_t fs, uint64_t ft) |
| 177 | { |
| 178 | LMIValue vs, vt; |
| 179 | unsigned int i; |
| 180 | |
| 181 | vs.d = fs; |
| 182 | vt.d = ft; |
| 183 | for (i = 0; i < 4; ++i) { |
| 184 | int r = vs.sh[i] - vt.sh[i]; |
| 185 | vs.sh[i] = SATSH(r); |
| 186 | } |
| 187 | return vs.d; |
| 188 | } |
| 189 | |
| 190 | uint64_t helper_psubush(uint64_t fs, uint64_t ft) |
| 191 | { |
| 192 | LMIValue vs, vt; |
| 193 | unsigned int i; |
| 194 | |
| 195 | vs.d = fs; |
| 196 | vt.d = ft; |
| 197 | for (i = 0; i < 4; ++i) { |
| 198 | int r = vs.uh[i] - vt.uh[i]; |
| 199 | vs.uh[i] = SATUH(r); |
| 200 | } |
| 201 | return vs.d; |
| 202 | } |
| 203 | |
| 204 | uint64_t helper_psubb(uint64_t fs, uint64_t ft) |
| 205 | { |
| 206 | LMIValue vs, vt; |
| 207 | unsigned int i; |
| 208 | |
| 209 | vs.d = fs; |
| 210 | vt.d = ft; |
| 211 | for (i = 0; i < 8; ++i) { |
| 212 | vs.ub[i] -= vt.ub[i]; |
| 213 | } |
| 214 | return vs.d; |
| 215 | } |
| 216 | |
| 217 | uint64_t helper_psubh(uint64_t fs, uint64_t ft) |
| 218 | { |
| 219 | LMIValue vs, vt; |
| 220 | unsigned int i; |
| 221 | |
| 222 | vs.d = fs; |
| 223 | vt.d = ft; |
| 224 | for (i = 0; i < 4; ++i) { |
| 225 | vs.uh[i] -= vt.uh[i]; |
| 226 | } |
| 227 | return vs.d; |
| 228 | } |
| 229 | |
| 230 | uint64_t helper_psubw(uint64_t fs, uint64_t ft) |
| 231 | { |
| 232 | LMIValue vs, vt; |
| 233 | unsigned int i; |
| 234 | |
| 235 | vs.d = fs; |
| 236 | vt.d = ft; |
| 237 | for (i = 0; i < 2; ++i) { |
| 238 | vs.uw[i] -= vt.uw[i]; |
| 239 | } |
| 240 | return vs.d; |
| 241 | } |
| 242 | |
| 243 | uint64_t helper_pshufh(uint64_t fs, uint64_t ft) |
| 244 | { |
| 245 | unsigned host = BYTE_ORDER_XOR(3); |
| 246 | LMIValue vd, vs; |
| 247 | unsigned i; |
| 248 | |
| 249 | vs.d = fs; |
| 250 | vd.d = 0; |
| 251 | for (i = 0; i < 4; i++, ft >>= 2) { |
| 252 | vd.uh[i ^ host] = vs.uh[(ft & 3) ^ host]; |
| 253 | } |
| 254 | return vd.d; |
| 255 | } |
| 256 | |
| 257 | uint64_t helper_packsswh(uint64_t fs, uint64_t ft) |
| 258 | { |
| 259 | uint64_t fd = 0; |
| 260 | int64_t tmp; |
| 261 | |
| 262 | tmp = (int32_t)(fs >> 0); |
| 263 | tmp = SATSH(tmp); |
| 264 | fd |= (tmp & 0xffff) << 0; |
| 265 | |
| 266 | tmp = (int32_t)(fs >> 32); |
| 267 | tmp = SATSH(tmp); |
| 268 | fd |= (tmp & 0xffff) << 16; |
| 269 | |
| 270 | tmp = (int32_t)(ft >> 0); |
| 271 | tmp = SATSH(tmp); |
| 272 | fd |= (tmp & 0xffff) << 32; |
| 273 | |
| 274 | tmp = (int32_t)(ft >> 32); |
| 275 | tmp = SATSH(tmp); |
| 276 | fd |= (tmp & 0xffff) << 48; |
| 277 | |
| 278 | return fd; |
| 279 | } |
| 280 | |
| 281 | uint64_t helper_packsshb(uint64_t fs, uint64_t ft) |
| 282 | { |
| 283 | uint64_t fd = 0; |
| 284 | unsigned int i; |
| 285 | |
| 286 | for (i = 0; i < 4; ++i) { |
| 287 | int16_t tmp = fs >> (i * 16); |
| 288 | tmp = SATSB(tmp); |
| 289 | fd |= (uint64_t)(tmp & 0xff) << (i * 8); |
| 290 | } |
| 291 | for (i = 0; i < 4; ++i) { |
| 292 | int16_t tmp = ft >> (i * 16); |
| 293 | tmp = SATSB(tmp); |
| 294 | fd |= (uint64_t)(tmp & 0xff) << (i * 8 + 32); |
| 295 | } |
| 296 | |
| 297 | return fd; |
| 298 | } |
| 299 | |
| 300 | uint64_t helper_packushb(uint64_t fs, uint64_t ft) |
| 301 | { |
| 302 | uint64_t fd = 0; |
| 303 | unsigned int i; |
| 304 | |
| 305 | for (i = 0; i < 4; ++i) { |
| 306 | int16_t tmp = fs >> (i * 16); |
| 307 | tmp = SATUB(tmp); |
| 308 | fd |= (uint64_t)(tmp & 0xff) << (i * 8); |
| 309 | } |
| 310 | for (i = 0; i < 4; ++i) { |
| 311 | int16_t tmp = ft >> (i * 16); |
| 312 | tmp = SATUB(tmp); |
| 313 | fd |= (uint64_t)(tmp & 0xff) << (i * 8 + 32); |
| 314 | } |
| 315 | |
| 316 | return fd; |
| 317 | } |
| 318 | |
| 319 | uint64_t helper_punpcklwd(uint64_t fs, uint64_t ft) |
| 320 | { |
| 321 | return (fs & 0xffffffff) | (ft << 32); |
| 322 | } |
| 323 | |
| 324 | uint64_t helper_punpckhwd(uint64_t fs, uint64_t ft) |
| 325 | { |
| 326 | return (fs >> 32) | (ft & ~0xffffffffull); |
| 327 | } |
| 328 | |
| 329 | uint64_t helper_punpcklhw(uint64_t fs, uint64_t ft) |
| 330 | { |
| 331 | unsigned host = BYTE_ORDER_XOR(3); |
| 332 | LMIValue vd, vs, vt; |
| 333 | |
| 334 | vs.d = fs; |
| 335 | vt.d = ft; |
| 336 | vd.uh[0 ^ host] = vs.uh[0 ^ host]; |
| 337 | vd.uh[1 ^ host] = vt.uh[0 ^ host]; |
| 338 | vd.uh[2 ^ host] = vs.uh[1 ^ host]; |
| 339 | vd.uh[3 ^ host] = vt.uh[1 ^ host]; |
| 340 | |
| 341 | return vd.d; |
| 342 | } |
| 343 | |
| 344 | uint64_t helper_punpckhhw(uint64_t fs, uint64_t ft) |
| 345 | { |
| 346 | unsigned host = BYTE_ORDER_XOR(3); |
| 347 | LMIValue vd, vs, vt; |
| 348 | |
| 349 | vs.d = fs; |
| 350 | vt.d = ft; |
| 351 | vd.uh[0 ^ host] = vs.uh[2 ^ host]; |
| 352 | vd.uh[1 ^ host] = vt.uh[2 ^ host]; |
| 353 | vd.uh[2 ^ host] = vs.uh[3 ^ host]; |
| 354 | vd.uh[3 ^ host] = vt.uh[3 ^ host]; |
| 355 | |
| 356 | return vd.d; |
| 357 | } |
| 358 | |
| 359 | uint64_t helper_punpcklbh(uint64_t fs, uint64_t ft) |
| 360 | { |
| 361 | unsigned host = BYTE_ORDER_XOR(7); |
| 362 | LMIValue vd, vs, vt; |
| 363 | |
| 364 | vs.d = fs; |
| 365 | vt.d = ft; |
| 366 | vd.ub[0 ^ host] = vs.ub[0 ^ host]; |
| 367 | vd.ub[1 ^ host] = vt.ub[0 ^ host]; |
| 368 | vd.ub[2 ^ host] = vs.ub[1 ^ host]; |
| 369 | vd.ub[3 ^ host] = vt.ub[1 ^ host]; |
| 370 | vd.ub[4 ^ host] = vs.ub[2 ^ host]; |
| 371 | vd.ub[5 ^ host] = vt.ub[2 ^ host]; |
| 372 | vd.ub[6 ^ host] = vs.ub[3 ^ host]; |
| 373 | vd.ub[7 ^ host] = vt.ub[3 ^ host]; |
| 374 | |
| 375 | return vd.d; |
| 376 | } |
| 377 | |
| 378 | uint64_t helper_punpckhbh(uint64_t fs, uint64_t ft) |
| 379 | { |
| 380 | unsigned host = BYTE_ORDER_XOR(7); |
| 381 | LMIValue vd, vs, vt; |
| 382 | |
| 383 | vs.d = fs; |
| 384 | vt.d = ft; |
| 385 | vd.ub[0 ^ host] = vs.ub[4 ^ host]; |
| 386 | vd.ub[1 ^ host] = vt.ub[4 ^ host]; |
| 387 | vd.ub[2 ^ host] = vs.ub[5 ^ host]; |
| 388 | vd.ub[3 ^ host] = vt.ub[5 ^ host]; |
| 389 | vd.ub[4 ^ host] = vs.ub[6 ^ host]; |
| 390 | vd.ub[5 ^ host] = vt.ub[6 ^ host]; |
| 391 | vd.ub[6 ^ host] = vs.ub[7 ^ host]; |
| 392 | vd.ub[7 ^ host] = vt.ub[7 ^ host]; |
| 393 | |
| 394 | return vd.d; |
| 395 | } |
| 396 | |
| 397 | uint64_t helper_pavgh(uint64_t fs, uint64_t ft) |
| 398 | { |
| 399 | LMIValue vs, vt; |
| 400 | unsigned i; |
| 401 | |
| 402 | vs.d = fs; |
| 403 | vt.d = ft; |
| 404 | for (i = 0; i < 4; i++) { |
| 405 | vs.uh[i] = (vs.uh[i] + vt.uh[i] + 1) >> 1; |
| 406 | } |
| 407 | return vs.d; |
| 408 | } |
| 409 | |
| 410 | uint64_t helper_pavgb(uint64_t fs, uint64_t ft) |
| 411 | { |
| 412 | LMIValue vs, vt; |
| 413 | unsigned i; |
| 414 | |
| 415 | vs.d = fs; |
| 416 | vt.d = ft; |
| 417 | for (i = 0; i < 8; i++) { |
| 418 | vs.ub[i] = (vs.ub[i] + vt.ub[i] + 1) >> 1; |
| 419 | } |
| 420 | return vs.d; |
| 421 | } |
| 422 | |
| 423 | uint64_t helper_pmaxsh(uint64_t fs, uint64_t ft) |
| 424 | { |
| 425 | LMIValue vs, vt; |
| 426 | unsigned i; |
| 427 | |
| 428 | vs.d = fs; |
| 429 | vt.d = ft; |
| 430 | for (i = 0; i < 4; i++) { |
| 431 | vs.sh[i] = (vs.sh[i] >= vt.sh[i] ? vs.sh[i] : vt.sh[i]); |
| 432 | } |
| 433 | return vs.d; |
| 434 | } |
| 435 | |
| 436 | uint64_t helper_pminsh(uint64_t fs, uint64_t ft) |
| 437 | { |
| 438 | LMIValue vs, vt; |
| 439 | unsigned i; |
| 440 | |
| 441 | vs.d = fs; |
| 442 | vt.d = ft; |
| 443 | for (i = 0; i < 4; i++) { |
| 444 | vs.sh[i] = (vs.sh[i] <= vt.sh[i] ? vs.sh[i] : vt.sh[i]); |
| 445 | } |
| 446 | return vs.d; |
| 447 | } |
| 448 | |
| 449 | uint64_t helper_pmaxub(uint64_t fs, uint64_t ft) |
| 450 | { |
| 451 | LMIValue vs, vt; |
| 452 | unsigned i; |
| 453 | |
| 454 | vs.d = fs; |
| 455 | vt.d = ft; |
| 456 | for (i = 0; i < 4; i++) { |
| 457 | vs.ub[i] = (vs.ub[i] >= vt.ub[i] ? vs.ub[i] : vt.ub[i]); |
| 458 | } |
| 459 | return vs.d; |
| 460 | } |
| 461 | |
| 462 | uint64_t helper_pminub(uint64_t fs, uint64_t ft) |
| 463 | { |
| 464 | LMIValue vs, vt; |
| 465 | unsigned i; |
| 466 | |
| 467 | vs.d = fs; |
| 468 | vt.d = ft; |
| 469 | for (i = 0; i < 4; i++) { |
| 470 | vs.ub[i] = (vs.ub[i] <= vt.ub[i] ? vs.ub[i] : vt.ub[i]); |
| 471 | } |
| 472 | return vs.d; |
| 473 | } |
| 474 | |
| 475 | uint64_t helper_pcmpeqw(uint64_t fs, uint64_t ft) |
| 476 | { |
| 477 | LMIValue vs, vt; |
| 478 | unsigned i; |
| 479 | |
| 480 | vs.d = fs; |
| 481 | vt.d = ft; |
| 482 | for (i = 0; i < 2; i++) { |
| 483 | vs.uw[i] = -(vs.uw[i] == vt.uw[i]); |
| 484 | } |
| 485 | return vs.d; |
| 486 | } |
| 487 | |
| 488 | uint64_t helper_pcmpgtw(uint64_t fs, uint64_t ft) |
| 489 | { |
| 490 | LMIValue vs, vt; |
| 491 | unsigned i; |
| 492 | |
| 493 | vs.d = fs; |
| 494 | vt.d = ft; |
| 495 | for (i = 0; i < 2; i++) { |
| 496 | vs.uw[i] = -(vs.uw[i] > vt.uw[i]); |
| 497 | } |
| 498 | return vs.d; |
| 499 | } |
| 500 | |
| 501 | uint64_t helper_pcmpeqh(uint64_t fs, uint64_t ft) |
| 502 | { |
| 503 | LMIValue vs, vt; |
| 504 | unsigned i; |
| 505 | |
| 506 | vs.d = fs; |
| 507 | vt.d = ft; |
| 508 | for (i = 0; i < 4; i++) { |
| 509 | vs.uh[i] = -(vs.uh[i] == vt.uh[i]); |
| 510 | } |
| 511 | return vs.d; |
| 512 | } |
| 513 | |
| 514 | uint64_t helper_pcmpgth(uint64_t fs, uint64_t ft) |
| 515 | { |
| 516 | LMIValue vs, vt; |
| 517 | unsigned i; |
| 518 | |
| 519 | vs.d = fs; |
| 520 | vt.d = ft; |
| 521 | for (i = 0; i < 4; i++) { |
| 522 | vs.uh[i] = -(vs.uh[i] > vt.uh[i]); |
| 523 | } |
| 524 | return vs.d; |
| 525 | } |
| 526 | |
| 527 | uint64_t helper_pcmpeqb(uint64_t fs, uint64_t ft) |
| 528 | { |
| 529 | LMIValue vs, vt; |
| 530 | unsigned i; |
| 531 | |
| 532 | vs.d = fs; |
| 533 | vt.d = ft; |
| 534 | for (i = 0; i < 8; i++) { |
| 535 | vs.ub[i] = -(vs.ub[i] == vt.ub[i]); |
| 536 | } |
| 537 | return vs.d; |
| 538 | } |
| 539 | |
| 540 | uint64_t helper_pcmpgtb(uint64_t fs, uint64_t ft) |
| 541 | { |
| 542 | LMIValue vs, vt; |
| 543 | unsigned i; |
| 544 | |
| 545 | vs.d = fs; |
| 546 | vt.d = ft; |
| 547 | for (i = 0; i < 8; i++) { |
| 548 | vs.ub[i] = -(vs.ub[i] > vt.ub[i]); |
| 549 | } |
| 550 | return vs.d; |
| 551 | } |
| 552 | |
| 553 | uint64_t helper_psllw(uint64_t fs, uint64_t ft) |
| 554 | { |
| 555 | LMIValue vs; |
| 556 | unsigned i; |
| 557 | |
| 558 | ft &= 0x7f; |
| 559 | if (ft > 31) { |
| 560 | return 0; |
| 561 | } |
| 562 | vs.d = fs; |
| 563 | for (i = 0; i < 2; ++i) { |
| 564 | vs.uw[i] <<= ft; |
| 565 | } |
| 566 | return vs.d; |
| 567 | } |
| 568 | |
| 569 | uint64_t helper_psrlw(uint64_t fs, uint64_t ft) |
| 570 | { |
| 571 | LMIValue vs; |
| 572 | unsigned i; |
| 573 | |
| 574 | ft &= 0x7f; |
| 575 | if (ft > 31) { |
| 576 | return 0; |
| 577 | } |
| 578 | vs.d = fs; |
| 579 | for (i = 0; i < 2; ++i) { |
| 580 | vs.uw[i] >>= ft; |
| 581 | } |
| 582 | return vs.d; |
| 583 | } |
| 584 | |
| 585 | uint64_t helper_psraw(uint64_t fs, uint64_t ft) |
| 586 | { |
| 587 | LMIValue vs; |
| 588 | unsigned i; |
| 589 | |
| 590 | ft &= 0x7f; |
| 591 | if (ft > 31) { |
| 592 | ft = 31; |
| 593 | } |
| 594 | vs.d = fs; |
| 595 | for (i = 0; i < 2; ++i) { |
| 596 | vs.sw[i] >>= ft; |
| 597 | } |
| 598 | return vs.d; |
| 599 | } |
| 600 | |
| 601 | uint64_t helper_psllh(uint64_t fs, uint64_t ft) |
| 602 | { |
| 603 | LMIValue vs; |
| 604 | unsigned i; |
| 605 | |
| 606 | ft &= 0x7f; |
| 607 | if (ft > 15) { |
| 608 | return 0; |
| 609 | } |
| 610 | vs.d = fs; |
| 611 | for (i = 0; i < 4; ++i) { |
| 612 | vs.uh[i] <<= ft; |
| 613 | } |
| 614 | return vs.d; |
| 615 | } |
| 616 | |
| 617 | uint64_t helper_psrlh(uint64_t fs, uint64_t ft) |
| 618 | { |
| 619 | LMIValue vs; |
| 620 | unsigned i; |
| 621 | |
| 622 | ft &= 0x7f; |
| 623 | if (ft > 15) { |
| 624 | return 0; |
| 625 | } |
| 626 | vs.d = fs; |
| 627 | for (i = 0; i < 4; ++i) { |
| 628 | vs.uh[i] >>= ft; |
| 629 | } |
| 630 | return vs.d; |
| 631 | } |
| 632 | |
| 633 | uint64_t helper_psrah(uint64_t fs, uint64_t ft) |
| 634 | { |
| 635 | LMIValue vs; |
| 636 | unsigned i; |
| 637 | |
| 638 | ft &= 0x7f; |
| 639 | if (ft > 15) { |
| 640 | ft = 15; |
| 641 | } |
| 642 | vs.d = fs; |
| 643 | for (i = 0; i < 4; ++i) { |
| 644 | vs.sh[i] >>= ft; |
| 645 | } |
| 646 | return vs.d; |
| 647 | } |
| 648 | |
| 649 | uint64_t helper_pmullh(uint64_t fs, uint64_t ft) |
| 650 | { |
| 651 | LMIValue vs, vt; |
| 652 | unsigned i; |
| 653 | |
| 654 | vs.d = fs; |
| 655 | vt.d = ft; |
| 656 | for (i = 0; i < 4; ++i) { |
| 657 | vs.sh[i] *= vt.sh[i]; |
| 658 | } |
| 659 | return vs.d; |
| 660 | } |
| 661 | |
| 662 | uint64_t helper_pmulhh(uint64_t fs, uint64_t ft) |
| 663 | { |
| 664 | LMIValue vs, vt; |
| 665 | unsigned i; |
| 666 | |
| 667 | vs.d = fs; |
| 668 | vt.d = ft; |
| 669 | for (i = 0; i < 4; ++i) { |
| 670 | int32_t r = vs.sh[i] * vt.sh[i]; |
| 671 | vs.sh[i] = r >> 16; |
| 672 | } |
| 673 | return vs.d; |
| 674 | } |
| 675 | |
| 676 | uint64_t helper_pmulhuh(uint64_t fs, uint64_t ft) |
| 677 | { |
| 678 | LMIValue vs, vt; |
| 679 | unsigned i; |
| 680 | |
| 681 | vs.d = fs; |
| 682 | vt.d = ft; |
| 683 | for (i = 0; i < 4; ++i) { |
| 684 | uint32_t r = vs.uh[i] * vt.uh[i]; |
| 685 | vs.uh[i] = r >> 16; |
| 686 | } |
| 687 | return vs.d; |
| 688 | } |
| 689 | |
| 690 | uint64_t helper_pmaddhw(uint64_t fs, uint64_t ft) |
| 691 | { |
| 692 | unsigned host = BYTE_ORDER_XOR(3); |
| 693 | LMIValue vs, vt; |
| 694 | uint32_t p0, p1; |
| 695 | |
| 696 | vs.d = fs; |
| 697 | vt.d = ft; |
| 698 | p0 = vs.sh[0 ^ host] * vt.sh[0 ^ host]; |
| 699 | p0 += vs.sh[1 ^ host] * vt.sh[1 ^ host]; |
| 700 | p1 = vs.sh[2 ^ host] * vt.sh[2 ^ host]; |
| 701 | p1 += vs.sh[3 ^ host] * vt.sh[3 ^ host]; |
| 702 | |
| 703 | return ((uint64_t)p1 << 32) | p0; |
| 704 | } |
| 705 | |
| 706 | uint64_t helper_pasubub(uint64_t fs, uint64_t ft) |
| 707 | { |
| 708 | LMIValue vs, vt; |
| 709 | unsigned i; |
| 710 | |
| 711 | vs.d = fs; |
| 712 | vt.d = ft; |
| 713 | for (i = 0; i < 8; ++i) { |
| 714 | int r = vs.ub[i] - vt.ub[i]; |
| 715 | vs.ub[i] = (r < 0 ? -r : r); |
| 716 | } |
| 717 | return vs.d; |
| 718 | } |
| 719 | |
| 720 | uint64_t helper_biadd(uint64_t fs) |
| 721 | { |
| 722 | unsigned i, fd; |
| 723 | |
| 724 | for (i = fd = 0; i < 8; ++i) { |
| 725 | fd += (fs >> (i * 8)) & 0xff; |
| 726 | } |
| 727 | return fd & 0xffff; |
| 728 | } |
| 729 | |
| 730 | uint64_t helper_pmovmskb(uint64_t fs) |
| 731 | { |
| 732 | unsigned fd = 0; |
| 733 | |
| 734 | fd |= ((fs >> 7) & 1) << 0; |
| 735 | fd |= ((fs >> 15) & 1) << 1; |
| 736 | fd |= ((fs >> 23) & 1) << 2; |
| 737 | fd |= ((fs >> 31) & 1) << 3; |
| 738 | fd |= ((fs >> 39) & 1) << 4; |
| 739 | fd |= ((fs >> 47) & 1) << 5; |
| 740 | fd |= ((fs >> 55) & 1) << 6; |
| 741 | fd |= ((fs >> 63) & 1) << 7; |
| 742 | |
| 743 | return fd & 0xff; |
| 744 | } |