target-i386/ops_sse.h

   1 /*
   2  *  MMX/3DNow!/SSE/SSE2/SSE3/SSSE3/SSE4/PNI support
   3  *
   4  *  Copyright (c) 2005 Fabrice Bellard
   5  *  Copyright (c) 2008 Intel Corporation  <andrew.zaborowski@intel.com>
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  19  */
  20 #if SHIFT == 0
  21 #define Reg MMXReg
  22 #define XMM_ONLY(...)
  23 #define B(n) MMX_B(n)
  24 #define W(n) MMX_W(n)
  25 #define L(n) MMX_L(n)
  26 #define Q(n) q
  27 #define SUFFIX _mmx
  28 #else
  29 #define Reg XMMReg
  30 #define XMM_ONLY(...) __VA_ARGS__
  31 #define B(n) XMM_B(n)
  32 #define W(n) XMM_W(n)
  33 #define L(n) XMM_L(n)
  34 #define Q(n) XMM_Q(n)
  35 #define SUFFIX _xmm
  36 #endif
  37
  38 void glue(helper_psrlw, SUFFIX)(Reg *d, Reg *s)
  39 {
  40     int shift;
  41
  42     if (s->Q(0) > 15) {
  43         d->Q(0) = 0;
  44 #if SHIFT == 1
  45         d->Q(1) = 0;
  46 #endif
  47     } else {
  48         shift = s->B(0);
  49         d->W(0) >>= shift;
  50         d->W(1) >>= shift;
  51         d->W(2) >>= shift;
  52         d->W(3) >>= shift;
  53 #if SHIFT == 1
  54         d->W(4) >>= shift;
  55         d->W(5) >>= shift;
  56         d->W(6) >>= shift;
  57         d->W(7) >>= shift;
  58 #endif
  59     }
  60 }
  61
  62 void glue(helper_psraw, SUFFIX)(Reg *d, Reg *s)
  63 {
  64     int shift;
  65
  66     if (s->Q(0) > 15) {
  67         shift = 15;
  68     } else {
  69         shift = s->B(0);
  70     }
  71     d->W(0) = (int16_t)d->W(0) >> shift;
  72     d->W(1) = (int16_t)d->W(1) >> shift;
  73     d->W(2) = (int16_t)d->W(2) >> shift;
  74     d->W(3) = (int16_t)d->W(3) >> shift;
  75 #if SHIFT == 1
  76     d->W(4) = (int16_t)d->W(4) >> shift;
  77     d->W(5) = (int16_t)d->W(5) >> shift;
  78     d->W(6) = (int16_t)d->W(6) >> shift;
  79     d->W(7) = (int16_t)d->W(7) >> shift;
  80 #endif
  81 }
  82
  83 void glue(helper_psllw, SUFFIX)(Reg *d, Reg *s)
  84 {
  85     int shift;
  86
  87     if (s->Q(0) > 15) {
  88         d->Q(0) = 0;
  89 #if SHIFT == 1
  90         d->Q(1) = 0;
  91 #endif
  92     } else {
  93         shift = s->B(0);
  94         d->W(0) <<= shift;
  95         d->W(1) <<= shift;
  96         d->W(2) <<= shift;
  97         d->W(3) <<= shift;
  98 #if SHIFT == 1
  99         d->W(4) <<= shift;
 100         d->W(5) <<= shift;
 101         d->W(6) <<= shift;
 102         d->W(7) <<= shift;
 103 #endif
 104     }
 105 }
 106
 107 void glue(helper_psrld, SUFFIX)(Reg *d, Reg *s)
 108 {
 109     int shift;
 110
 111     if (s->Q(0) > 31) {
 112         d->Q(0) = 0;
 113 #if SHIFT == 1
 114         d->Q(1) = 0;
 115 #endif
 116     } else {
 117         shift = s->B(0);
 118         d->L(0) >>= shift;
 119         d->L(1) >>= shift;
 120 #if SHIFT == 1
 121         d->L(2) >>= shift;
 122         d->L(3) >>= shift;
 123 #endif
 124     }
 125 }
 126
 127 void glue(helper_psrad, SUFFIX)(Reg *d, Reg *s)
 128 {
 129     int shift;
 130
 131     if (s->Q(0) > 31) {
 132         shift = 31;
 133     } else {
 134         shift = s->B(0);
 135     }
 136     d->L(0) = (int32_t)d->L(0) >> shift;
 137     d->L(1) = (int32_t)d->L(1) >> shift;
 138 #if SHIFT == 1
 139     d->L(2) = (int32_t)d->L(2) >> shift;
 140     d->L(3) = (int32_t)d->L(3) >> shift;
 141 #endif
 142 }
 143
 144 void glue(helper_pslld, SUFFIX)(Reg *d, Reg *s)
 145 {
 146     int shift;
 147
 148     if (s->Q(0) > 31) {
 149         d->Q(0) = 0;
 150 #if SHIFT == 1
 151         d->Q(1) = 0;
 152 #endif
 153     } else {
 154         shift = s->B(0);
 155         d->L(0) <<= shift;
 156         d->L(1) <<= shift;
 157 #if SHIFT == 1
 158         d->L(2) <<= shift;
 159         d->L(3) <<= shift;
 160 #endif
 161     }
 162 }
 163
 164 void glue(helper_psrlq, SUFFIX)(Reg *d, Reg *s)
 165 {
 166     int shift;
 167
 168     if (s->Q(0) > 63) {
 169         d->Q(0) = 0;
 170 #if SHIFT == 1
 171         d->Q(1) = 0;
 172 #endif
 173     } else {
 174         shift = s->B(0);
 175         d->Q(0) >>= shift;
 176 #if SHIFT == 1
 177         d->Q(1) >>= shift;
 178 #endif
 179     }
 180 }
 181
 182 void glue(helper_psllq, SUFFIX)(Reg *d, Reg *s)
 183 {
 184     int shift;
 185
 186     if (s->Q(0) > 63) {
 187         d->Q(0) = 0;
 188 #if SHIFT == 1
 189         d->Q(1) = 0;
 190 #endif
 191     } else {
 192         shift = s->B(0);
 193         d->Q(0) <<= shift;
 194 #if SHIFT == 1
 195         d->Q(1) <<= shift;
 196 #endif
 197     }
 198 }
 199
 200 #if SHIFT == 1
 201 void glue(helper_psrldq, SUFFIX)(Reg *d, Reg *s)
 202 {
 203     int shift, i;
 204
 205     shift = s->L(0);
 206     if (shift > 16) {
 207         shift = 16;
 208     }
 209     for (i = 0; i < 16 - shift; i++) {
 210         d->B(i) = d->B(i + shift);
 211     }
 212     for (i = 16 - shift; i < 16; i++) {
 213         d->B(i) = 0;
 214     }
 215 }
 216
 217 void glue(helper_pslldq, SUFFIX)(Reg *d, Reg *s)
 218 {
 219     int shift, i;
 220
 221     shift = s->L(0);
 222     if (shift > 16) {
 223         shift = 16;
 224     }
 225     for (i = 15; i >= shift; i--) {
 226         d->B(i) = d->B(i - shift);
 227     }
 228     for (i = 0; i < shift; i++) {
 229         d->B(i) = 0;
 230     }
 231 }
 232 #endif
 233
 234 #define SSE_HELPER_B(name, F)                                   \
 235     void glue(name, SUFFIX)(Reg *d, Reg *s)                     \
 236     {                                                           \
 237         d->B(0) = F(d->B(0), s->B(0));                          \
 238         d->B(1) = F(d->B(1), s->B(1));                          \
 239         d->B(2) = F(d->B(2), s->B(2));                          \
 240         d->B(3) = F(d->B(3), s->B(3));                          \
 241         d->B(4) = F(d->B(4), s->B(4));                          \
 242         d->B(5) = F(d->B(5), s->B(5));                          \
 243         d->B(6) = F(d->B(6), s->B(6));                          \
 244         d->B(7) = F(d->B(7), s->B(7));                          \
 245         XMM_ONLY(                                               \
 246                  d->B(8) = F(d->B(8), s->B(8));                 \
 247                  d->B(9) = F(d->B(9), s->B(9));                 \
 248                  d->B(10) = F(d->B(10), s->B(10));              \
 249                  d->B(11) = F(d->B(11), s->B(11));              \
 250                  d->B(12) = F(d->B(12), s->B(12));              \
 251                  d->B(13) = F(d->B(13), s->B(13));              \
 252                  d->B(14) = F(d->B(14), s->B(14));              \
 253                  d->B(15) = F(d->B(15), s->B(15));              \
 254                                                         )       \
 255             }
 256
 257 #define SSE_HELPER_W(name, F)                                   \
 258     void glue(name, SUFFIX)(Reg *d, Reg *s)                     \
 259     {                                                           \
 260         d->W(0) = F(d->W(0), s->W(0));                          \
 261         d->W(1) = F(d->W(1), s->W(1));                          \
 262         d->W(2) = F(d->W(2), s->W(2));                          \
 263         d->W(3) = F(d->W(3), s->W(3));                          \
 264         XMM_ONLY(                                               \
 265                  d->W(4) = F(d->W(4), s->W(4));                 \
 266                  d->W(5) = F(d->W(5), s->W(5));                 \
 267                  d->W(6) = F(d->W(6), s->W(6));                 \
 268                  d->W(7) = F(d->W(7), s->W(7));                 \
 269                                                         )       \
 270             }
 271
 272 #define SSE_HELPER_L(name, F)                                   \
 273     void glue(name, SUFFIX)(Reg *d, Reg *s)                     \
 274     {                                                           \
 275         d->L(0) = F(d->L(0), s->L(0));                          \
 276         d->L(1) = F(d->L(1), s->L(1));                          \
 277         XMM_ONLY(                                               \
 278                  d->L(2) = F(d->L(2), s->L(2));                 \
 279                  d->L(3) = F(d->L(3), s->L(3));                 \
 280                                                         )       \
 281             }
 282
 283 #define SSE_HELPER_Q(name, F)                                   \
 284     void glue(name, SUFFIX)(Reg *d, Reg *s)                     \
 285     {                                                           \
 286         d->Q(0) = F(d->Q(0), s->Q(0));                          \
 287         XMM_ONLY(                                               \
 288                  d->Q(1) = F(d->Q(1), s->Q(1));                 \
 289                                                         )       \
 290             }
 291
 292 #if SHIFT == 0
 293 static inline int satub(int x)
 294 {
 295     if (x < 0) {
 296         return 0;
 297     } else if (x > 255) {
 298         return 255;
 299     } else {
 300         return x;
 301     }
 302 }
 303
 304 static inline int satuw(int x)
 305 {
 306     if (x < 0) {
 307         return 0;
 308     } else if (x > 65535) {
 309         return 65535;
 310     } else {
 311         return x;
 312     }
 313 }
 314
 315 static inline int satsb(int x)
 316 {
 317     if (x < -128) {
 318         return -128;
 319     } else if (x > 127) {
 320         return 127;
 321     } else {
 322         return x;
 323     }
 324 }
 325
 326 static inline int satsw(int x)
 327 {
 328     if (x < -32768) {
 329         return -32768;
 330     } else if (x > 32767) {
 331         return 32767;
 332     } else {
 333         return x;
 334     }
 335 }
 336
 337 #define FADD(a, b) ((a) + (b))
 338 #define FADDUB(a, b) satub((a) + (b))
 339 #define FADDUW(a, b) satuw((a) + (b))
 340 #define FADDSB(a, b) satsb((int8_t)(a) + (int8_t)(b))
 341 #define FADDSW(a, b) satsw((int16_t)(a) + (int16_t)(b))
 342
 343 #define FSUB(a, b) ((a) - (b))
 344 #define FSUBUB(a, b) satub((a) - (b))
 345 #define FSUBUW(a, b) satuw((a) - (b))
 346 #define FSUBSB(a, b) satsb((int8_t)(a) - (int8_t)(b))
 347 #define FSUBSW(a, b) satsw((int16_t)(a) - (int16_t)(b))
 348 #define FMINUB(a, b) ((a) < (b)) ? (a) : (b)
 349 #define FMINSW(a, b) ((int16_t)(a) < (int16_t)(b)) ? (a) : (b)
 350 #define FMAXUB(a, b) ((a) > (b)) ? (a) : (b)
 351 #define FMAXSW(a, b) ((int16_t)(a) > (int16_t)(b)) ? (a) : (b)
 352
 353 #define FAND(a, b) ((a) & (b))
 354 #define FANDN(a, b) ((~(a)) & (b))
 355 #define FOR(a, b) ((a) | (b))
 356 #define FXOR(a, b) ((a) ^ (b))
 357
 358 #define FCMPGTB(a, b) ((int8_t)(a) > (int8_t)(b) ? -1 : 0)
 359 #define FCMPGTW(a, b) ((int16_t)(a) > (int16_t)(b) ? -1 : 0)
 360 #define FCMPGTL(a, b) ((int32_t)(a) > (int32_t)(b) ? -1 : 0)
 361 #define FCMPEQ(a, b) ((a) == (b) ? -1 : 0)
 362
 363 #define FMULLW(a, b) ((a) * (b))
 364 #define FMULHRW(a, b) (((int16_t)(a) * (int16_t)(b) + 0x8000) >> 16)
 365 #define FMULHUW(a, b) ((a) * (b) >> 16)
 366 #define FMULHW(a, b) ((int16_t)(a) * (int16_t)(b) >> 16)
 367
 368 #define FAVG(a, b) (((a) + (b) + 1) >> 1)
 369 #endif
 370
 371 SSE_HELPER_B(helper_paddb, FADD)
 372 SSE_HELPER_W(helper_paddw, FADD)
 373 SSE_HELPER_L(helper_paddl, FADD)
 374 SSE_HELPER_Q(helper_paddq, FADD)
 375
 376 SSE_HELPER_B(helper_psubb, FSUB)
 377 SSE_HELPER_W(helper_psubw, FSUB)
 378 SSE_HELPER_L(helper_psubl, FSUB)
 379 SSE_HELPER_Q(helper_psubq, FSUB)
 380
 381 SSE_HELPER_B(helper_paddusb, FADDUB)
 382 SSE_HELPER_B(helper_paddsb, FADDSB)
 383 SSE_HELPER_B(helper_psubusb, FSUBUB)
 384 SSE_HELPER_B(helper_psubsb, FSUBSB)
 385
 386 SSE_HELPER_W(helper_paddusw, FADDUW)
 387 SSE_HELPER_W(helper_paddsw, FADDSW)
 388 SSE_HELPER_W(helper_psubusw, FSUBUW)
 389 SSE_HELPER_W(helper_psubsw, FSUBSW)
 390
 391 SSE_HELPER_B(helper_pminub, FMINUB)
 392 SSE_HELPER_B(helper_pmaxub, FMAXUB)
 393
 394 SSE_HELPER_W(helper_pminsw, FMINSW)
 395 SSE_HELPER_W(helper_pmaxsw, FMAXSW)
 396
 397 SSE_HELPER_Q(helper_pand, FAND)
 398 SSE_HELPER_Q(helper_pandn, FANDN)
 399 SSE_HELPER_Q(helper_por, FOR)
 400 SSE_HELPER_Q(helper_pxor, FXOR)
 401
 402 SSE_HELPER_B(helper_pcmpgtb, FCMPGTB)
 403 SSE_HELPER_W(helper_pcmpgtw, FCMPGTW)
 404 SSE_HELPER_L(helper_pcmpgtl, FCMPGTL)
 405
 406 SSE_HELPER_B(helper_pcmpeqb, FCMPEQ)
 407 SSE_HELPER_W(helper_pcmpeqw, FCMPEQ)
 408 SSE_HELPER_L(helper_pcmpeql, FCMPEQ)
 409
 410 SSE_HELPER_W(helper_pmullw, FMULLW)
 411 #if SHIFT == 0
 412 SSE_HELPER_W(helper_pmulhrw, FMULHRW)
 413 #endif
 414 SSE_HELPER_W(helper_pmulhuw, FMULHUW)
 415 SSE_HELPER_W(helper_pmulhw, FMULHW)
 416
 417 SSE_HELPER_B(helper_pavgb, FAVG)
 418 SSE_HELPER_W(helper_pavgw, FAVG)
 419
 420 void glue(helper_pmuludq, SUFFIX)(Reg *d, Reg *s)
 421 {
 422     d->Q(0) = (uint64_t)s->L(0) * (uint64_t)d->L(0);
 423 #if SHIFT == 1
 424     d->Q(1) = (uint64_t)s->L(2) * (uint64_t)d->L(2);
 425 #endif
 426 }
 427
 428 void glue(helper_pmaddwd, SUFFIX)(Reg *d, Reg *s)
 429 {
 430     int i;
 431
 432     for (i = 0; i < (2 << SHIFT); i++) {
 433         d->L(i) = (int16_t)s->W(2 * i) * (int16_t)d->W(2 * i) +
 434             (int16_t)s->W(2 * i + 1) * (int16_t)d->W(2 * i + 1);
 435     }
 436 }
 437
 438 #if SHIFT == 0
 439 static inline int abs1(int a)
 440 {
 441     if (a < 0) {
 442         return -a;
 443     } else {
 444         return a;
 445     }
 446 }
 447 #endif
 448 void glue(helper_psadbw, SUFFIX)(Reg *d, Reg *s)
 449 {
 450     unsigned int val;
 451
 452     val = 0;
 453     val += abs1(d->B(0) - s->B(0));
 454     val += abs1(d->B(1) - s->B(1));
 455     val += abs1(d->B(2) - s->B(2));
 456     val += abs1(d->B(3) - s->B(3));
 457     val += abs1(d->B(4) - s->B(4));
 458     val += abs1(d->B(5) - s->B(5));
 459     val += abs1(d->B(6) - s->B(6));
 460     val += abs1(d->B(7) - s->B(7));
 461     d->Q(0) = val;
 462 #if SHIFT == 1
 463     val = 0;
 464     val += abs1(d->B(8) - s->B(8));
 465     val += abs1(d->B(9) - s->B(9));
 466     val += abs1(d->B(10) - s->B(10));
 467     val += abs1(d->B(11) - s->B(11));
 468     val += abs1(d->B(12) - s->B(12));
 469     val += abs1(d->B(13) - s->B(13));
 470     val += abs1(d->B(14) - s->B(14));
 471     val += abs1(d->B(15) - s->B(15));
 472     d->Q(1) = val;
 473 #endif
 474 }
 475
 476 void glue(helper_maskmov, SUFFIX)(Reg *d, Reg *s, target_ulong a0)
 477 {
 478     int i;
 479
 480     for (i = 0; i < (8 << SHIFT); i++) {
 481         if (s->B(i) & 0x80) {
 482             stb(a0 + i, d->B(i));
 483         }
 484     }
 485 }
 486
 487 void glue(helper_movl_mm_T0, SUFFIX)(Reg *d, uint32_t val)
 488 {
 489     d->L(0) = val;
 490     d->L(1) = 0;
 491 #if SHIFT == 1
 492     d->Q(1) = 0;
 493 #endif
 494 }
 495
 496 #ifdef TARGET_X86_64
 497 void glue(helper_movq_mm_T0, SUFFIX)(Reg *d, uint64_t val)
 498 {
 499     d->Q(0) = val;
 500 #if SHIFT == 1
 501     d->Q(1) = 0;
 502 #endif
 503 }
 504 #endif
 505
 506 #if SHIFT == 0
 507 void glue(helper_pshufw, SUFFIX)(Reg *d, Reg *s, int order)
 508 {
 509     Reg r;
 510
 511     r.W(0) = s->W(order & 3);
 512     r.W(1) = s->W((order >> 2) & 3);
 513     r.W(2) = s->W((order >> 4) & 3);
 514     r.W(3) = s->W((order >> 6) & 3);
 515     *d = r;
 516 }
 517 #else
 518 void helper_shufps(Reg *d, Reg *s, int order)
 519 {
 520     Reg r;
 521
 522     r.L(0) = d->L(order & 3);
 523     r.L(1) = d->L((order >> 2) & 3);
 524     r.L(2) = s->L((order >> 4) & 3);
 525     r.L(3) = s->L((order >> 6) & 3);
 526     *d = r;
 527 }
 528
 529 void helper_shufpd(Reg *d, Reg *s, int order)
 530 {
 531     Reg r;
 532
 533     r.Q(0) = d->Q(order & 1);
 534     r.Q(1) = s->Q((order >> 1) & 1);
 535     *d = r;
 536 }
 537
 538 void glue(helper_pshufd, SUFFIX)(Reg *d, Reg *s, int order)
 539 {
 540     Reg r;
 541
 542     r.L(0) = s->L(order & 3);
 543     r.L(1) = s->L((order >> 2) & 3);
 544     r.L(2) = s->L((order >> 4) & 3);
 545     r.L(3) = s->L((order >> 6) & 3);
 546     *d = r;
 547 }
 548
 549 void glue(helper_pshuflw, SUFFIX)(Reg *d, Reg *s, int order)
 550 {
 551     Reg r;
 552
 553     r.W(0) = s->W(order & 3);
 554     r.W(1) = s->W((order >> 2) & 3);
 555     r.W(2) = s->W((order >> 4) & 3);
 556     r.W(3) = s->W((order >> 6) & 3);
 557     r.Q(1) = s->Q(1);
 558     *d = r;
 559 }
 560
 561 void glue(helper_pshufhw, SUFFIX)(Reg *d, Reg *s, int order)
 562 {
 563     Reg r;
 564
 565     r.Q(0) = s->Q(0);
 566     r.W(4) = s->W(4 + (order & 3));
 567     r.W(5) = s->W(4 + ((order >> 2) & 3));
 568     r.W(6) = s->W(4 + ((order >> 4) & 3));
 569     r.W(7) = s->W(4 + ((order >> 6) & 3));
 570     *d = r;
 571 }
 572 #endif
 573
 574 #if SHIFT == 1
 575 /* FPU ops */
 576 /* XXX: not accurate */
 577
 578 #define SSE_HELPER_S(name, F)                           \
 579     void helper_ ## name ## ps(Reg *d, Reg *s)          \
 580     {                                                   \
 581         d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));  \
 582         d->XMM_S(1) = F(32, d->XMM_S(1), s->XMM_S(1));  \
 583         d->XMM_S(2) = F(32, d->XMM_S(2), s->XMM_S(2));  \
 584         d->XMM_S(3) = F(32, d->XMM_S(3), s->XMM_S(3));  \
 585     }                                                   \
 586                                                         \
 587     void helper_ ## name ## ss(Reg *d, Reg *s)          \
 588     {                                                   \
 589         d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));  \
 590     }                                                   \
 591                                                         \
 592     void helper_ ## name ## pd(Reg *d, Reg *s)          \
 593     {                                                   \
 594         d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));  \
 595         d->XMM_D(1) = F(64, d->XMM_D(1), s->XMM_D(1));  \
 596     }                                                   \
 597                                                         \
 598     void helper_ ## name ## sd(Reg *d, Reg *s)          \
 599     {                                                   \
 600         d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));  \
 601     }
 602
 603 #define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
 604 #define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
 605 #define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
 606 #define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
 607 #define FPU_SQRT(size, a, b) float ## size ## _sqrt(b, &env->sse_status)
 608
 609 /* Note that the choice of comparison op here is important to get the
 610  * special cases right: for min and max Intel specifies that (-0,0),
 611  * (NaN, anything) and (anything, NaN) return the second argument.
 612  */
 613 #define FPU_MIN(size, a, b)                                     \
 614     (float ## size ## _lt(a, b, &env->sse_status) ? (a) : (b))
 615 #define FPU_MAX(size, a, b)                                     \
 616     (float ## size ## _lt(b, a, &env->sse_status) ? (a) : (b))
 617
 618 SSE_HELPER_S(add, FPU_ADD)
 619 SSE_HELPER_S(sub, FPU_SUB)
 620 SSE_HELPER_S(mul, FPU_MUL)
 621 SSE_HELPER_S(div, FPU_DIV)
 622 SSE_HELPER_S(min, FPU_MIN)
 623 SSE_HELPER_S(max, FPU_MAX)
 624 SSE_HELPER_S(sqrt, FPU_SQRT)
 625
 626
 627 /* float to float conversions */
 628 void helper_cvtps2pd(Reg *d, Reg *s)
 629 {
 630     float32 s0, s1;
 631
 632     s0 = s->XMM_S(0);
 633     s1 = s->XMM_S(1);
 634     d->XMM_D(0) = float32_to_float64(s0, &env->sse_status);
 635     d->XMM_D(1) = float32_to_float64(s1, &env->sse_status);
 636 }
 637
 638 void helper_cvtpd2ps(Reg *d, Reg *s)
 639 {
 640     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 641     d->XMM_S(1) = float64_to_float32(s->XMM_D(1), &env->sse_status);
 642     d->Q(1) = 0;
 643 }
 644
 645 void helper_cvtss2sd(Reg *d, Reg *s)
 646 {
 647     d->XMM_D(0) = float32_to_float64(s->XMM_S(0), &env->sse_status);
 648 }
 649
 650 void helper_cvtsd2ss(Reg *d, Reg *s)
 651 {
 652     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 653 }
 654
 655 /* integer to float */
 656 void helper_cvtdq2ps(Reg *d, Reg *s)
 657 {
 658     d->XMM_S(0) = int32_to_float32(s->XMM_L(0), &env->sse_status);
 659     d->XMM_S(1) = int32_to_float32(s->XMM_L(1), &env->sse_status);
 660     d->XMM_S(2) = int32_to_float32(s->XMM_L(2), &env->sse_status);
 661     d->XMM_S(3) = int32_to_float32(s->XMM_L(3), &env->sse_status);
 662 }
 663
 664 void helper_cvtdq2pd(Reg *d, Reg *s)
 665 {
 666     int32_t l0, l1;
 667
 668     l0 = (int32_t)s->XMM_L(0);
 669     l1 = (int32_t)s->XMM_L(1);
 670     d->XMM_D(0) = int32_to_float64(l0, &env->sse_status);
 671     d->XMM_D(1) = int32_to_float64(l1, &env->sse_status);
 672 }
 673
 674 void helper_cvtpi2ps(XMMReg *d, MMXReg *s)
 675 {
 676     d->XMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
 677     d->XMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
 678 }
 679
 680 void helper_cvtpi2pd(XMMReg *d, MMXReg *s)
 681 {
 682     d->XMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
 683     d->XMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
 684 }
 685
 686 void helper_cvtsi2ss(XMMReg *d, uint32_t val)
 687 {
 688     d->XMM_S(0) = int32_to_float32(val, &env->sse_status);
 689 }
 690
 691 void helper_cvtsi2sd(XMMReg *d, uint32_t val)
 692 {
 693     d->XMM_D(0) = int32_to_float64(val, &env->sse_status);
 694 }
 695
 696 #ifdef TARGET_X86_64
 697 void helper_cvtsq2ss(XMMReg *d, uint64_t val)
 698 {
 699     d->XMM_S(0) = int64_to_float32(val, &env->sse_status);
 700 }
 701
 702 void helper_cvtsq2sd(XMMReg *d, uint64_t val)
 703 {
 704     d->XMM_D(0) = int64_to_float64(val, &env->sse_status);
 705 }
 706 #endif
 707
 708 /* float to integer */
 709 void helper_cvtps2dq(XMMReg *d, XMMReg *s)
 710 {
 711     d->XMM_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 712     d->XMM_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 713     d->XMM_L(2) = float32_to_int32(s->XMM_S(2), &env->sse_status);
 714     d->XMM_L(3) = float32_to_int32(s->XMM_S(3), &env->sse_status);
 715 }
 716
 717 void helper_cvtpd2dq(XMMReg *d, XMMReg *s)
 718 {
 719     d->XMM_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 720     d->XMM_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 721     d->XMM_Q(1) = 0;
 722 }
 723
 724 void helper_cvtps2pi(MMXReg *d, XMMReg *s)
 725 {
 726     d->MMX_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 727     d->MMX_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 728 }
 729
 730 void helper_cvtpd2pi(MMXReg *d, XMMReg *s)
 731 {
 732     d->MMX_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 733     d->MMX_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 734 }
 735
 736 int32_t helper_cvtss2si(XMMReg *s)
 737 {
 738     return float32_to_int32(s->XMM_S(0), &env->sse_status);
 739 }
 740
 741 int32_t helper_cvtsd2si(XMMReg *s)
 742 {
 743     return float64_to_int32(s->XMM_D(0), &env->sse_status);
 744 }
 745
 746 #ifdef TARGET_X86_64
 747 int64_t helper_cvtss2sq(XMMReg *s)
 748 {
 749     return float32_to_int64(s->XMM_S(0), &env->sse_status);
 750 }
 751
 752 int64_t helper_cvtsd2sq(XMMReg *s)
 753 {
 754     return float64_to_int64(s->XMM_D(0), &env->sse_status);
 755 }
 756 #endif
 757
 758 /* float to integer truncated */
 759 void helper_cvttps2dq(XMMReg *d, XMMReg *s)
 760 {
 761     d->XMM_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 762     d->XMM_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 763     d->XMM_L(2) = float32_to_int32_round_to_zero(s->XMM_S(2), &env->sse_status);
 764     d->XMM_L(3) = float32_to_int32_round_to_zero(s->XMM_S(3), &env->sse_status);
 765 }
 766
 767 void helper_cvttpd2dq(XMMReg *d, XMMReg *s)
 768 {
 769     d->XMM_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 770     d->XMM_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 771     d->XMM_Q(1) = 0;
 772 }
 773
 774 void helper_cvttps2pi(MMXReg *d, XMMReg *s)
 775 {
 776     d->MMX_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 777     d->MMX_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 778 }
 779
 780 void helper_cvttpd2pi(MMXReg *d, XMMReg *s)
 781 {
 782     d->MMX_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 783     d->MMX_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 784 }
 785
 786 int32_t helper_cvttss2si(XMMReg *s)
 787 {
 788     return float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 789 }
 790
 791 int32_t helper_cvttsd2si(XMMReg *s)
 792 {
 793     return float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 794 }
 795
 796 #ifdef TARGET_X86_64
 797 int64_t helper_cvttss2sq(XMMReg *s)
 798 {
 799     return float32_to_int64_round_to_zero(s->XMM_S(0), &env->sse_status);
 800 }
 801
 802 int64_t helper_cvttsd2sq(XMMReg *s)
 803 {
 804     return float64_to_int64_round_to_zero(s->XMM_D(0), &env->sse_status);
 805 }
 806 #endif
 807
 808 void helper_rsqrtps(XMMReg *d, XMMReg *s)
 809 {
 810     d->XMM_S(0) = float32_div(float32_one,
 811                               float32_sqrt(s->XMM_S(0), &env->sse_status),
 812                               &env->sse_status);
 813     d->XMM_S(1) = float32_div(float32_one,
 814                               float32_sqrt(s->XMM_S(1), &env->sse_status),
 815                               &env->sse_status);
 816     d->XMM_S(2) = float32_div(float32_one,
 817                               float32_sqrt(s->XMM_S(2), &env->sse_status),
 818                               &env->sse_status);
 819     d->XMM_S(3) = float32_div(float32_one,
 820                               float32_sqrt(s->XMM_S(3), &env->sse_status),
 821                               &env->sse_status);
 822 }
 823
 824 void helper_rsqrtss(XMMReg *d, XMMReg *s)
 825 {
 826     d->XMM_S(0) = float32_div(float32_one,
 827                               float32_sqrt(s->XMM_S(0), &env->sse_status),
 828                               &env->sse_status);
 829 }
 830
 831 void helper_rcpps(XMMReg *d, XMMReg *s)
 832 {
 833     d->XMM_S(0) = float32_div(float32_one, s->XMM_S(0), &env->sse_status);
 834     d->XMM_S(1) = float32_div(float32_one, s->XMM_S(1), &env->sse_status);
 835     d->XMM_S(2) = float32_div(float32_one, s->XMM_S(2), &env->sse_status);
 836     d->XMM_S(3) = float32_div(float32_one, s->XMM_S(3), &env->sse_status);
 837 }
 838
 839 void helper_rcpss(XMMReg *d, XMMReg *s)
 840 {
 841     d->XMM_S(0) = float32_div(float32_one, s->XMM_S(0), &env->sse_status);
 842 }
 843
 844 static inline uint64_t helper_extrq(uint64_t src, int shift, int len)
 845 {
 846     uint64_t mask;
 847
 848     if (len == 0) {
 849         mask = ~0LL;
 850     } else {
 851         mask = (1ULL << len) - 1;
 852     }
 853     return (src >> shift) & mask;
 854 }
 855
 856 void helper_extrq_r(XMMReg *d, XMMReg *s)
 857 {
 858     d->XMM_Q(0) = helper_extrq(d->XMM_Q(0), s->XMM_B(1), s->XMM_B(0));
 859 }
 860
 861 void helper_extrq_i(XMMReg *d, int index, int length)
 862 {
 863     d->XMM_Q(0) = helper_extrq(d->XMM_Q(0), index, length);
 864 }
 865
 866 static inline uint64_t helper_insertq(uint64_t src, int shift, int len)
 867 {
 868     uint64_t mask;
 869
 870     if (len == 0) {
 871         mask = ~0ULL;
 872     } else {
 873         mask = (1ULL << len) - 1;
 874     }
 875     return (src & ~(mask << shift)) | ((src & mask) << shift);
 876 }
 877
 878 void helper_insertq_r(XMMReg *d, XMMReg *s)
 879 {
 880     d->XMM_Q(0) = helper_insertq(s->XMM_Q(0), s->XMM_B(9), s->XMM_B(8));
 881 }
 882
 883 void helper_insertq_i(XMMReg *d, int index, int length)
 884 {
 885     d->XMM_Q(0) = helper_insertq(d->XMM_Q(0), index, length);
 886 }
 887
 888 void helper_haddps(XMMReg *d, XMMReg *s)
 889 {
 890     XMMReg r;
 891
 892     r.XMM_S(0) = float32_add(d->XMM_S(0), d->XMM_S(1), &env->sse_status);
 893     r.XMM_S(1) = float32_add(d->XMM_S(2), d->XMM_S(3), &env->sse_status);
 894     r.XMM_S(2) = float32_add(s->XMM_S(0), s->XMM_S(1), &env->sse_status);
 895     r.XMM_S(3) = float32_add(s->XMM_S(2), s->XMM_S(3), &env->sse_status);
 896     *d = r;
 897 }
 898
 899 void helper_haddpd(XMMReg *d, XMMReg *s)
 900 {
 901     XMMReg r;
 902
 903     r.XMM_D(0) = float64_add(d->XMM_D(0), d->XMM_D(1), &env->sse_status);
 904     r.XMM_D(1) = float64_add(s->XMM_D(0), s->XMM_D(1), &env->sse_status);
 905     *d = r;
 906 }
 907
 908 void helper_hsubps(XMMReg *d, XMMReg *s)
 909 {
 910     XMMReg r;
 911
 912     r.XMM_S(0) = float32_sub(d->XMM_S(0), d->XMM_S(1), &env->sse_status);
 913     r.XMM_S(1) = float32_sub(d->XMM_S(2), d->XMM_S(3), &env->sse_status);
 914     r.XMM_S(2) = float32_sub(s->XMM_S(0), s->XMM_S(1), &env->sse_status);
 915     r.XMM_S(3) = float32_sub(s->XMM_S(2), s->XMM_S(3), &env->sse_status);
 916     *d = r;
 917 }
 918
 919 void helper_hsubpd(XMMReg *d, XMMReg *s)
 920 {
 921     XMMReg r;
 922
 923     r.XMM_D(0) = float64_sub(d->XMM_D(0), d->XMM_D(1), &env->sse_status);
 924     r.XMM_D(1) = float64_sub(s->XMM_D(0), s->XMM_D(1), &env->sse_status);
 925     *d = r;
 926 }
 927
 928 void helper_addsubps(XMMReg *d, XMMReg *s)
 929 {
 930     d->XMM_S(0) = float32_sub(d->XMM_S(0), s->XMM_S(0), &env->sse_status);
 931     d->XMM_S(1) = float32_add(d->XMM_S(1), s->XMM_S(1), &env->sse_status);
 932     d->XMM_S(2) = float32_sub(d->XMM_S(2), s->XMM_S(2), &env->sse_status);
 933     d->XMM_S(3) = float32_add(d->XMM_S(3), s->XMM_S(3), &env->sse_status);
 934 }
 935
 936 void helper_addsubpd(XMMReg *d, XMMReg *s)
 937 {
 938     d->XMM_D(0) = float64_sub(d->XMM_D(0), s->XMM_D(0), &env->sse_status);
 939     d->XMM_D(1) = float64_add(d->XMM_D(1), s->XMM_D(1), &env->sse_status);
 940 }
 941
 942 /* XXX: unordered */
 943 #define SSE_HELPER_CMP(name, F)                         \
 944     void helper_ ## name ## ps(Reg *d, Reg *s)          \
 945     {                                                   \
 946         d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));  \
 947         d->XMM_L(1) = F(32, d->XMM_S(1), s->XMM_S(1));  \
 948         d->XMM_L(2) = F(32, d->XMM_S(2), s->XMM_S(2));  \
 949         d->XMM_L(3) = F(32, d->XMM_S(3), s->XMM_S(3));  \
 950     }                                                   \
 951                                                         \
 952     void helper_ ## name ## ss(Reg *d, Reg *s)          \
 953     {                                                   \
 954         d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));  \
 955     }                                                   \
 956                                                         \
 957     void helper_ ## name ## pd(Reg *d, Reg *s)          \
 958     {                                                   \
 959         d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));  \
 960         d->XMM_Q(1) = F(64, d->XMM_D(1), s->XMM_D(1));  \
 961     }                                                   \
 962                                                         \
 963     void helper_ ## name ## sd(Reg *d, Reg *s)          \
 964     {                                                   \
 965         d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));  \
 966     }
 967
 968 #define FPU_CMPEQ(size, a, b)                                           \
 969     (float ## size ## _eq_quiet(a, b, &env->sse_status) ? -1 : 0)
 970 #define FPU_CMPLT(size, a, b)                                           \
 971     (float ## size ## _lt(a, b, &env->sse_status) ? -1 : 0)
 972 #define FPU_CMPLE(size, a, b)                                           \
 973     (float ## size ## _le(a, b, &env->sse_status) ? -1 : 0)
 974 #define FPU_CMPUNORD(size, a, b)                                        \
 975     (float ## size ## _unordered_quiet(a, b, &env->sse_status) ? -1 : 0)
 976 #define FPU_CMPNEQ(size, a, b)                                          \
 977     (float ## size ## _eq_quiet(a, b, &env->sse_status) ? 0 : -1)
 978 #define FPU_CMPNLT(size, a, b)                                          \
 979     (float ## size ## _lt(a, b, &env->sse_status) ? 0 : -1)
 980 #define FPU_CMPNLE(size, a, b)                                          \
 981     (float ## size ## _le(a, b, &env->sse_status) ? 0 : -1)
 982 #define FPU_CMPORD(size, a, b)                                          \
 983     (float ## size ## _unordered_quiet(a, b, &env->sse_status) ? 0 : -1)
 984
 985 SSE_HELPER_CMP(cmpeq, FPU_CMPEQ)
 986 SSE_HELPER_CMP(cmplt, FPU_CMPLT)
 987 SSE_HELPER_CMP(cmple, FPU_CMPLE)
 988 SSE_HELPER_CMP(cmpunord, FPU_CMPUNORD)
 989 SSE_HELPER_CMP(cmpneq, FPU_CMPNEQ)
 990 SSE_HELPER_CMP(cmpnlt, FPU_CMPNLT)
 991 SSE_HELPER_CMP(cmpnle, FPU_CMPNLE)
 992 SSE_HELPER_CMP(cmpord, FPU_CMPORD)
 993
 994 static const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
 995
 996 void helper_ucomiss(Reg *d, Reg *s)
 997 {
 998     int ret;
 999     float32 s0, s1;
1000
1001     s0 = d->XMM_S(0);
1002     s1 = s->XMM_S(0);
1003     ret = float32_compare_quiet(s0, s1, &env->sse_status);
1004     CC_SRC = comis_eflags[ret + 1];
1005 }
1006
1007 void helper_comiss(Reg *d, Reg *s)
1008 {
1009     int ret;
1010     float32 s0, s1;
1011
1012     s0 = d->XMM_S(0);
1013     s1 = s->XMM_S(0);
1014     ret = float32_compare(s0, s1, &env->sse_status);
1015     CC_SRC = comis_eflags[ret + 1];
1016 }
1017
1018 void helper_ucomisd(Reg *d, Reg *s)
1019 {
1020     int ret;
1021     float64 d0, d1;
1022
1023     d0 = d->XMM_D(0);
1024     d1 = s->XMM_D(0);
1025     ret = float64_compare_quiet(d0, d1, &env->sse_status);
1026     CC_SRC = comis_eflags[ret + 1];
1027 }
1028
1029 void helper_comisd(Reg *d, Reg *s)
1030 {
1031     int ret;
1032     float64 d0, d1;
1033
1034     d0 = d->XMM_D(0);
1035     d1 = s->XMM_D(0);
1036     ret = float64_compare(d0, d1, &env->sse_status);
1037     CC_SRC = comis_eflags[ret + 1];
1038 }
1039
1040 uint32_t helper_movmskps(Reg *s)
1041 {
1042     int b0, b1, b2, b3;
1043
1044     b0 = s->XMM_L(0) >> 31;
1045     b1 = s->XMM_L(1) >> 31;
1046     b2 = s->XMM_L(2) >> 31;
1047     b3 = s->XMM_L(3) >> 31;
1048     return b0 | (b1 << 1) | (b2 << 2) | (b3 << 3);
1049 }
1050
1051 uint32_t helper_movmskpd(Reg *s)
1052 {
1053     int b0, b1;
1054
1055     b0 = s->XMM_L(1) >> 31;
1056     b1 = s->XMM_L(3) >> 31;
1057     return b0 | (b1 << 1);
1058 }
1059
1060 #endif
1061
1062 uint32_t glue(helper_pmovmskb, SUFFIX)(Reg *s)
1063 {
1064     uint32_t val;
1065
1066     val = 0;
1067     val |= (s->B(0) >> 7);
1068     val |= (s->B(1) >> 6) & 0x02;
1069     val |= (s->B(2) >> 5) & 0x04;
1070     val |= (s->B(3) >> 4) & 0x08;
1071     val |= (s->B(4) >> 3) & 0x10;
1072     val |= (s->B(5) >> 2) & 0x20;
1073     val |= (s->B(6) >> 1) & 0x40;
1074     val |= (s->B(7)) & 0x80;
1075 #if SHIFT == 1
1076     val |= (s->B(8) << 1) & 0x0100;
1077     val |= (s->B(9) << 2) & 0x0200;
1078     val |= (s->B(10) << 3) & 0x0400;
1079     val |= (s->B(11) << 4) & 0x0800;
1080     val |= (s->B(12) << 5) & 0x1000;
1081     val |= (s->B(13) << 6) & 0x2000;
1082     val |= (s->B(14) << 7) & 0x4000;
1083     val |= (s->B(15) << 8) & 0x8000;
1084 #endif
1085     return val;
1086 }
1087
1088 void glue(helper_packsswb, SUFFIX)(Reg *d, Reg *s)
1089 {
1090     Reg r;
1091
1092     r.B(0) = satsb((int16_t)d->W(0));
1093     r.B(1) = satsb((int16_t)d->W(1));
1094     r.B(2) = satsb((int16_t)d->W(2));
1095     r.B(3) = satsb((int16_t)d->W(3));
1096 #if SHIFT == 1
1097     r.B(4) = satsb((int16_t)d->W(4));
1098     r.B(5) = satsb((int16_t)d->W(5));
1099     r.B(6) = satsb((int16_t)d->W(6));
1100     r.B(7) = satsb((int16_t)d->W(7));
1101 #endif
1102     r.B((4 << SHIFT) + 0) = satsb((int16_t)s->W(0));
1103     r.B((4 << SHIFT) + 1) = satsb((int16_t)s->W(1));
1104     r.B((4 << SHIFT) + 2) = satsb((int16_t)s->W(2));
1105     r.B((4 << SHIFT) + 3) = satsb((int16_t)s->W(3));
1106 #if SHIFT == 1
1107     r.B(12) = satsb((int16_t)s->W(4));
1108     r.B(13) = satsb((int16_t)s->W(5));
1109     r.B(14) = satsb((int16_t)s->W(6));
1110     r.B(15) = satsb((int16_t)s->W(7));
1111 #endif
1112     *d = r;
1113 }
1114
1115 void glue(helper_packuswb, SUFFIX)(Reg *d, Reg *s)
1116 {
1117     Reg r;
1118
1119     r.B(0) = satub((int16_t)d->W(0));
1120     r.B(1) = satub((int16_t)d->W(1));
1121     r.B(2) = satub((int16_t)d->W(2));
1122     r.B(3) = satub((int16_t)d->W(3));
1123 #if SHIFT == 1
1124     r.B(4) = satub((int16_t)d->W(4));
1125     r.B(5) = satub((int16_t)d->W(5));
1126     r.B(6) = satub((int16_t)d->W(6));
1127     r.B(7) = satub((int16_t)d->W(7));
1128 #endif
1129     r.B((4 << SHIFT) + 0) = satub((int16_t)s->W(0));
1130     r.B((4 << SHIFT) + 1) = satub((int16_t)s->W(1));
1131     r.B((4 << SHIFT) + 2) = satub((int16_t)s->W(2));
1132     r.B((4 << SHIFT) + 3) = satub((int16_t)s->W(3));
1133 #if SHIFT == 1
1134     r.B(12) = satub((int16_t)s->W(4));
1135     r.B(13) = satub((int16_t)s->W(5));
1136     r.B(14) = satub((int16_t)s->W(6));
1137     r.B(15) = satub((int16_t)s->W(7));
1138 #endif
1139     *d = r;
1140 }
1141
1142 void glue(helper_packssdw, SUFFIX)(Reg *d, Reg *s)
1143 {
1144     Reg r;
1145
1146     r.W(0) = satsw(d->L(0));
1147     r.W(1) = satsw(d->L(1));
1148 #if SHIFT == 1
1149     r.W(2) = satsw(d->L(2));
1150     r.W(3) = satsw(d->L(3));
1151 #endif
1152     r.W((2 << SHIFT) + 0) = satsw(s->L(0));
1153     r.W((2 << SHIFT) + 1) = satsw(s->L(1));
1154 #if SHIFT == 1
1155     r.W(6) = satsw(s->L(2));
1156     r.W(7) = satsw(s->L(3));
1157 #endif
1158     *d = r;
1159 }
1160
1161 #define UNPCK_OP(base_name, base)                                       \
1162                                                                         \
1163     void glue(helper_punpck ## base_name ## bw, SUFFIX)(Reg *d, Reg *s) \
1164     {                                                                   \
1165         Reg r;                                                          \
1166                                                                         \
1167         r.B(0) = d->B((base << (SHIFT + 2)) + 0);                       \
1168         r.B(1) = s->B((base << (SHIFT + 2)) + 0);                       \
1169         r.B(2) = d->B((base << (SHIFT + 2)) + 1);                       \
1170         r.B(3) = s->B((base << (SHIFT + 2)) + 1);                       \
1171         r.B(4) = d->B((base << (SHIFT + 2)) + 2);                       \
1172         r.B(5) = s->B((base << (SHIFT + 2)) + 2);                       \
1173         r.B(6) = d->B((base << (SHIFT + 2)) + 3);                       \
1174         r.B(7) = s->B((base << (SHIFT + 2)) + 3);                       \
1175         XMM_ONLY(                                                       \
1176                  r.B(8) = d->B((base << (SHIFT + 2)) + 4);              \
1177                  r.B(9) = s->B((base << (SHIFT + 2)) + 4);              \
1178                  r.B(10) = d->B((base << (SHIFT + 2)) + 5);             \
1179                  r.B(11) = s->B((base << (SHIFT + 2)) + 5);             \
1180                  r.B(12) = d->B((base << (SHIFT + 2)) + 6);             \
1181                  r.B(13) = s->B((base << (SHIFT + 2)) + 6);             \
1182                  r.B(14) = d->B((base << (SHIFT + 2)) + 7);             \
1183                  r.B(15) = s->B((base << (SHIFT + 2)) + 7);             \
1184                                                                         ) \
1185             *d = r;                                                     \
1186     }                                                                   \
1187                                                                         \
1188     void glue(helper_punpck ## base_name ## wd, SUFFIX)(Reg *d, Reg *s) \
1189     {                                                                   \
1190         Reg r;                                                          \
1191                                                                         \
1192         r.W(0) = d->W((base << (SHIFT + 1)) + 0);                       \
1193         r.W(1) = s->W((base << (SHIFT + 1)) + 0);                       \
1194         r.W(2) = d->W((base << (SHIFT + 1)) + 1);                       \
1195         r.W(3) = s->W((base << (SHIFT + 1)) + 1);                       \
1196         XMM_ONLY(                                                       \
1197                  r.W(4) = d->W((base << (SHIFT + 1)) + 2);              \
1198                  r.W(5) = s->W((base << (SHIFT + 1)) + 2);              \
1199                  r.W(6) = d->W((base << (SHIFT + 1)) + 3);              \
1200                  r.W(7) = s->W((base << (SHIFT + 1)) + 3);              \
1201                                                                         ) \
1202             *d = r;                                                     \
1203     }                                                                   \
1204                                                                         \
1205     void glue(helper_punpck ## base_name ## dq, SUFFIX)(Reg *d, Reg *s) \
1206     {                                                                   \
1207         Reg r;                                                          \
1208                                                                         \
1209         r.L(0) = d->L((base << SHIFT) + 0);                             \
1210         r.L(1) = s->L((base << SHIFT) + 0);                             \
1211         XMM_ONLY(                                                       \
1212                  r.L(2) = d->L((base << SHIFT) + 1);                    \
1213                  r.L(3) = s->L((base << SHIFT) + 1);                    \
1214                                                                         ) \
1215             *d = r;                                                     \
1216     }                                                                   \
1217                                                                         \
1218     XMM_ONLY(                                                           \
1219              void glue(helper_punpck ## base_name ## qdq, SUFFIX)(Reg *d, \
1220                                                                   Reg *s) \
1221              {                                                          \
1222                  Reg r;                                                 \
1223                                                                         \
1224                  r.Q(0) = d->Q(base);                                   \
1225                  r.Q(1) = s->Q(base);                                   \
1226                  *d = r;                                                \
1227              }                                                          \
1228                                                                         )
1229
1230 UNPCK_OP(l, 0)
1231 UNPCK_OP(h, 1)
1232
1233 /* 3DNow! float ops */
1234 #if SHIFT == 0
1235 void helper_pi2fd(MMXReg *d, MMXReg *s)
1236 {
1237     d->MMX_S(0) = int32_to_float32(s->MMX_L(0), &env->mmx_status);
1238     d->MMX_S(1) = int32_to_float32(s->MMX_L(1), &env->mmx_status);
1239 }
1240
1241 void helper_pi2fw(MMXReg *d, MMXReg *s)
1242 {
1243     d->MMX_S(0) = int32_to_float32((int16_t)s->MMX_W(0), &env->mmx_status);
1244     d->MMX_S(1) = int32_to_float32((int16_t)s->MMX_W(2), &env->mmx_status);
1245 }
1246
1247 void helper_pf2id(MMXReg *d, MMXReg *s)
1248 {
1249     d->MMX_L(0) = float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status);
1250     d->MMX_L(1) = float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status);
1251 }
1252
1253 void helper_pf2iw(MMXReg *d, MMXReg *s)
1254 {
1255     d->MMX_L(0) = satsw(float32_to_int32_round_to_zero(s->MMX_S(0),
1256                                                        &env->mmx_status));
1257     d->MMX_L(1) = satsw(float32_to_int32_round_to_zero(s->MMX_S(1),
1258                                                        &env->mmx_status));
1259 }
1260
1261 void helper_pfacc(MMXReg *d, MMXReg *s)
1262 {
1263     MMXReg r;
1264
1265     r.MMX_S(0) = float32_add(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1266     r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1267     *d = r;
1268 }
1269
1270 void helper_pfadd(MMXReg *d, MMXReg *s)
1271 {
1272     d->MMX_S(0) = float32_add(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1273     d->MMX_S(1) = float32_add(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1274 }
1275
1276 void helper_pfcmpeq(MMXReg *d, MMXReg *s)
1277 {
1278     d->MMX_L(0) = float32_eq_quiet(d->MMX_S(0), s->MMX_S(0),
1279                                    &env->mmx_status) ? -1 : 0;
1280     d->MMX_L(1) = float32_eq_quiet(d->MMX_S(1), s->MMX_S(1),
1281                                    &env->mmx_status) ? -1 : 0;
1282 }
1283
1284 void helper_pfcmpge(MMXReg *d, MMXReg *s)
1285 {
1286     d->MMX_L(0) = float32_le(s->MMX_S(0), d->MMX_S(0),
1287                              &env->mmx_status) ? -1 : 0;
1288     d->MMX_L(1) = float32_le(s->MMX_S(1), d->MMX_S(1),
1289                              &env->mmx_status) ? -1 : 0;
1290 }
1291
1292 void helper_pfcmpgt(MMXReg *d, MMXReg *s)
1293 {
1294     d->MMX_L(0) = float32_lt(s->MMX_S(0), d->MMX_S(0),
1295                              &env->mmx_status) ? -1 : 0;
1296     d->MMX_L(1) = float32_lt(s->MMX_S(1), d->MMX_S(1),
1297                              &env->mmx_status) ? -1 : 0;
1298 }
1299
1300 void helper_pfmax(MMXReg *d, MMXReg *s)
1301 {
1302     if (float32_lt(d->MMX_S(0), s->MMX_S(0), &env->mmx_status)) {
1303         d->MMX_S(0) = s->MMX_S(0);
1304     }
1305     if (float32_lt(d->MMX_S(1), s->MMX_S(1), &env->mmx_status)) {
1306         d->MMX_S(1) = s->MMX_S(1);
1307     }
1308 }
1309
1310 void helper_pfmin(MMXReg *d, MMXReg *s)
1311 {
1312     if (float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status)) {
1313         d->MMX_S(0) = s->MMX_S(0);
1314     }
1315     if (float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status)) {
1316         d->MMX_S(1) = s->MMX_S(1);
1317     }
1318 }
1319
1320 void helper_pfmul(MMXReg *d, MMXReg *s)
1321 {
1322     d->MMX_S(0) = float32_mul(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1323     d->MMX_S(1) = float32_mul(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1324 }
1325
1326 void helper_pfnacc(MMXReg *d, MMXReg *s)
1327 {
1328     MMXReg r;
1329
1330     r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1331     r.MMX_S(1) = float32_sub(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1332     *d = r;
1333 }
1334
1335 void helper_pfpnacc(MMXReg *d, MMXReg *s)
1336 {
1337     MMXReg r;
1338
1339     r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1340     r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1341     *d = r;
1342 }
1343
1344 void helper_pfrcp(MMXReg *d, MMXReg *s)
1345 {
1346     d->MMX_S(0) = float32_div(float32_one, s->MMX_S(0), &env->mmx_status);
1347     d->MMX_S(1) = d->MMX_S(0);
1348 }
1349
1350 void helper_pfrsqrt(MMXReg *d, MMXReg *s)
1351 {
1352     d->MMX_L(1) = s->MMX_L(0) & 0x7fffffff;
1353     d->MMX_S(1) = float32_div(float32_one,
1354                               float32_sqrt(d->MMX_S(1), &env->mmx_status),
1355                               &env->mmx_status);
1356     d->MMX_L(1) |= s->MMX_L(0) & 0x80000000;
1357     d->MMX_L(0) = d->MMX_L(1);
1358 }
1359
1360 void helper_pfsub(MMXReg *d, MMXReg *s)
1361 {
1362     d->MMX_S(0) = float32_sub(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1363     d->MMX_S(1) = float32_sub(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1364 }
1365
1366 void helper_pfsubr(MMXReg *d, MMXReg *s)
1367 {
1368     d->MMX_S(0) = float32_sub(s->MMX_S(0), d->MMX_S(0), &env->mmx_status);
1369     d->MMX_S(1) = float32_sub(s->MMX_S(1), d->MMX_S(1), &env->mmx_status);
1370 }
1371
1372 void helper_pswapd(MMXReg *d, MMXReg *s)
1373 {
1374     MMXReg r;
1375
1376     r.MMX_L(0) = s->MMX_L(1);
1377     r.MMX_L(1) = s->MMX_L(0);
1378     *d = r;
1379 }
1380 #endif
1381
1382 /* SSSE3 op helpers */
1383 void glue(helper_pshufb, SUFFIX)(Reg *d, Reg *s)
1384 {
1385     int i;
1386     Reg r;
1387
1388     for (i = 0; i < (8 << SHIFT); i++) {
1389         r.B(i) = (s->B(i) & 0x80) ? 0 : (d->B(s->B(i) & ((8 << SHIFT) - 1)));
1390     }
1391
1392     *d = r;
1393 }
1394
1395 void glue(helper_phaddw, SUFFIX)(Reg *d, Reg *s)
1396 {
1397     d->W(0) = (int16_t)d->W(0) + (int16_t)d->W(1);
1398     d->W(1) = (int16_t)d->W(2) + (int16_t)d->W(3);
1399     XMM_ONLY(d->W(2) = (int16_t)d->W(4) + (int16_t)d->W(5));
1400     XMM_ONLY(d->W(3) = (int16_t)d->W(6) + (int16_t)d->W(7));
1401     d->W((2 << SHIFT) + 0) = (int16_t)s->W(0) + (int16_t)s->W(1);
1402     d->W((2 << SHIFT) + 1) = (int16_t)s->W(2) + (int16_t)s->W(3);
1403     XMM_ONLY(d->W(6) = (int16_t)s->W(4) + (int16_t)s->W(5));
1404     XMM_ONLY(d->W(7) = (int16_t)s->W(6) + (int16_t)s->W(7));
1405 }
1406
1407 void glue(helper_phaddd, SUFFIX)(Reg *d, Reg *s)
1408 {
1409     d->L(0) = (int32_t)d->L(0) + (int32_t)d->L(1);
1410     XMM_ONLY(d->L(1) = (int32_t)d->L(2) + (int32_t)d->L(3));
1411     d->L((1 << SHIFT) + 0) = (int32_t)s->L(0) + (int32_t)s->L(1);
1412     XMM_ONLY(d->L(3) = (int32_t)s->L(2) + (int32_t)s->L(3));
1413 }
1414
1415 void glue(helper_phaddsw, SUFFIX)(Reg *d, Reg *s)
1416 {
1417     d->W(0) = satsw((int16_t)d->W(0) + (int16_t)d->W(1));
1418     d->W(1) = satsw((int16_t)d->W(2) + (int16_t)d->W(3));
1419     XMM_ONLY(d->W(2) = satsw((int16_t)d->W(4) + (int16_t)d->W(5)));
1420     XMM_ONLY(d->W(3) = satsw((int16_t)d->W(6) + (int16_t)d->W(7)));
1421     d->W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) + (int16_t)s->W(1));
1422     d->W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) + (int16_t)s->W(3));
1423     XMM_ONLY(d->W(6) = satsw((int16_t)s->W(4) + (int16_t)s->W(5)));
1424     XMM_ONLY(d->W(7) = satsw((int16_t)s->W(6) + (int16_t)s->W(7)));
1425 }
1426
1427 void glue(helper_pmaddubsw, SUFFIX)(Reg *d, Reg *s)
1428 {
1429     d->W(0) = satsw((int8_t)s->B(0) * (uint8_t)d->B(0) +
1430                     (int8_t)s->B(1) * (uint8_t)d->B(1));
1431     d->W(1) = satsw((int8_t)s->B(2) * (uint8_t)d->B(2) +
1432                     (int8_t)s->B(3) * (uint8_t)d->B(3));
1433     d->W(2) = satsw((int8_t)s->B(4) * (uint8_t)d->B(4) +
1434                     (int8_t)s->B(5) * (uint8_t)d->B(5));
1435     d->W(3) = satsw((int8_t)s->B(6) * (uint8_t)d->B(6) +
1436                     (int8_t)s->B(7) * (uint8_t)d->B(7));
1437 #if SHIFT == 1
1438     d->W(4) = satsw((int8_t)s->B(8) * (uint8_t)d->B(8) +
1439                     (int8_t)s->B(9) * (uint8_t)d->B(9));
1440     d->W(5) = satsw((int8_t)s->B(10) * (uint8_t)d->B(10) +
1441                     (int8_t)s->B(11) * (uint8_t)d->B(11));
1442     d->W(6) = satsw((int8_t)s->B(12) * (uint8_t)d->B(12) +
1443                     (int8_t)s->B(13) * (uint8_t)d->B(13));
1444     d->W(7) = satsw((int8_t)s->B(14) * (uint8_t)d->B(14) +
1445                     (int8_t)s->B(15) * (uint8_t)d->B(15));
1446 #endif
1447 }
1448
1449 void glue(helper_phsubw, SUFFIX)(Reg *d, Reg *s)
1450 {
1451     d->W(0) = (int16_t)d->W(0) - (int16_t)d->W(1);
1452     d->W(1) = (int16_t)d->W(2) - (int16_t)d->W(3);
1453     XMM_ONLY(d->W(2) = (int16_t)d->W(4) - (int16_t)d->W(5));
1454     XMM_ONLY(d->W(3) = (int16_t)d->W(6) - (int16_t)d->W(7));
1455     d->W((2 << SHIFT) + 0) = (int16_t)s->W(0) - (int16_t)s->W(1);
1456     d->W((2 << SHIFT) + 1) = (int16_t)s->W(2) - (int16_t)s->W(3);
1457     XMM_ONLY(d->W(6) = (int16_t)s->W(4) - (int16_t)s->W(5));
1458     XMM_ONLY(d->W(7) = (int16_t)s->W(6) - (int16_t)s->W(7));
1459 }
1460
1461 void glue(helper_phsubd, SUFFIX)(Reg *d, Reg *s)
1462 {
1463     d->L(0) = (int32_t)d->L(0) - (int32_t)d->L(1);
1464     XMM_ONLY(d->L(1) = (int32_t)d->L(2) - (int32_t)d->L(3));
1465     d->L((1 << SHIFT) + 0) = (int32_t)s->L(0) - (int32_t)s->L(1);
1466     XMM_ONLY(d->L(3) = (int32_t)s->L(2) - (int32_t)s->L(3));
1467 }
1468
1469 void glue(helper_phsubsw, SUFFIX)(Reg *d, Reg *s)
1470 {
1471     d->W(0) = satsw((int16_t)d->W(0) - (int16_t)d->W(1));
1472     d->W(1) = satsw((int16_t)d->W(2) - (int16_t)d->W(3));
1473     XMM_ONLY(d->W(2) = satsw((int16_t)d->W(4) - (int16_t)d->W(5)));
1474     XMM_ONLY(d->W(3) = satsw((int16_t)d->W(6) - (int16_t)d->W(7)));
1475     d->W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) - (int16_t)s->W(1));
1476     d->W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) - (int16_t)s->W(3));
1477     XMM_ONLY(d->W(6) = satsw((int16_t)s->W(4) - (int16_t)s->W(5)));
1478     XMM_ONLY(d->W(7) = satsw((int16_t)s->W(6) - (int16_t)s->W(7)));
1479 }
1480
1481 #define FABSB(_, x) (x > INT8_MAX  ? -(int8_t)x : x)
1482 #define FABSW(_, x) (x > INT16_MAX ? -(int16_t)x : x)
1483 #define FABSL(_, x) (x > INT32_MAX ? -(int32_t)x : x)
1484 SSE_HELPER_B(helper_pabsb, FABSB)
1485 SSE_HELPER_W(helper_pabsw, FABSW)
1486 SSE_HELPER_L(helper_pabsd, FABSL)
1487
1488 #define FMULHRSW(d, s) (((int16_t) d * (int16_t)s + 0x4000) >> 15)
1489 SSE_HELPER_W(helper_pmulhrsw, FMULHRSW)
1490
1491 #define FSIGNB(d, s) (s <= INT8_MAX  ? s ? d : 0 : -(int8_t)d)
1492 #define FSIGNW(d, s) (s <= INT16_MAX ? s ? d : 0 : -(int16_t)d)
1493 #define FSIGNL(d, s) (s <= INT32_MAX ? s ? d : 0 : -(int32_t)d)
1494 SSE_HELPER_B(helper_psignb, FSIGNB)
1495 SSE_HELPER_W(helper_psignw, FSIGNW)
1496 SSE_HELPER_L(helper_psignd, FSIGNL)
1497
1498 void glue(helper_palignr, SUFFIX)(Reg *d, Reg *s, int32_t shift)
1499 {
1500     Reg r;
1501
1502     /* XXX could be checked during translation */
1503     if (shift >= (16 << SHIFT)) {
1504         r.Q(0) = 0;
1505         XMM_ONLY(r.Q(1) = 0);
1506     } else {
1507         shift <<= 3;
1508 #define SHR(v, i) (i < 64 && i > -64 ? i > 0 ? v >> (i) : (v << -(i)) : 0)
1509 #if SHIFT == 0
1510         r.Q(0) = SHR(s->Q(0), shift - 0) |
1511             SHR(d->Q(0), shift -  64);
1512 #else
1513         r.Q(0) = SHR(s->Q(0), shift - 0) |
1514             SHR(s->Q(1), shift -  64) |
1515             SHR(d->Q(0), shift - 128) |
1516             SHR(d->Q(1), shift - 192);
1517         r.Q(1) = SHR(s->Q(0), shift + 64) |
1518             SHR(s->Q(1), shift -   0) |
1519             SHR(d->Q(0), shift -  64) |
1520             SHR(d->Q(1), shift - 128);
1521 #endif
1522 #undef SHR
1523     }
1524
1525     *d = r;
1526 }
1527
1528 #define XMM0 (env->xmm_regs[0])
1529
1530 #if SHIFT == 1
1531 #define SSE_HELPER_V(name, elem, num, F)                                \
1532     void glue(name, SUFFIX)(Reg *d, Reg *s)                             \
1533     {                                                                   \
1534         d->elem(0) = F(d->elem(0), s->elem(0), XMM0.elem(0));           \
1535         d->elem(1) = F(d->elem(1), s->elem(1), XMM0.elem(1));           \
1536         if (num > 2) {                                                  \
1537             d->elem(2) = F(d->elem(2), s->elem(2), XMM0.elem(2));       \
1538             d->elem(3) = F(d->elem(3), s->elem(3), XMM0.elem(3));       \
1539             if (num > 4) {                                              \
1540                 d->elem(4) = F(d->elem(4), s->elem(4), XMM0.elem(4));   \
1541                 d->elem(5) = F(d->elem(5), s->elem(5), XMM0.elem(5));   \
1542                 d->elem(6) = F(d->elem(6), s->elem(6), XMM0.elem(6));   \
1543                 d->elem(7) = F(d->elem(7), s->elem(7), XMM0.elem(7));   \
1544                 if (num > 8) {                                          \
1545                     d->elem(8) = F(d->elem(8), s->elem(8), XMM0.elem(8)); \
1546                     d->elem(9) = F(d->elem(9), s->elem(9), XMM0.elem(9)); \
1547                     d->elem(10) = F(d->elem(10), s->elem(10), XMM0.elem(10)); \
1548                     d->elem(11) = F(d->elem(11), s->elem(11), XMM0.elem(11)); \
1549                     d->elem(12) = F(d->elem(12), s->elem(12), XMM0.elem(12)); \
1550                     d->elem(13) = F(d->elem(13), s->elem(13), XMM0.elem(13)); \
1551                     d->elem(14) = F(d->elem(14), s->elem(14), XMM0.elem(14)); \
1552                     d->elem(15) = F(d->elem(15), s->elem(15), XMM0.elem(15)); \
1553                 }                                                       \
1554             }                                                           \
1555         }                                                               \
1556     }
1557
1558 #define SSE_HELPER_I(name, elem, num, F)                                \
1559     void glue(name, SUFFIX)(Reg *d, Reg *s, uint32_t imm)               \
1560     {                                                                   \
1561         d->elem(0) = F(d->elem(0), s->elem(0), ((imm >> 0) & 1));       \
1562         d->elem(1) = F(d->elem(1), s->elem(1), ((imm >> 1) & 1));       \
1563         if (num > 2) {                                                  \
1564             d->elem(2) = F(d->elem(2), s->elem(2), ((imm >> 2) & 1));   \
1565             d->elem(3) = F(d->elem(3), s->elem(3), ((imm >> 3) & 1));   \
1566             if (num > 4) {                                              \
1567                 d->elem(4) = F(d->elem(4), s->elem(4), ((imm >> 4) & 1)); \
1568                 d->elem(5) = F(d->elem(5), s->elem(5), ((imm >> 5) & 1)); \
1569                 d->elem(6) = F(d->elem(6), s->elem(6), ((imm >> 6) & 1)); \
1570                 d->elem(7) = F(d->elem(7), s->elem(7), ((imm >> 7) & 1)); \
1571                 if (num > 8) {                                          \
1572                     d->elem(8) = F(d->elem(8), s->elem(8), ((imm >> 8) & 1)); \
1573                     d->elem(9) = F(d->elem(9), s->elem(9), ((imm >> 9) & 1)); \
1574                     d->elem(10) = F(d->elem(10), s->elem(10),           \
1575                                     ((imm >> 10) & 1));                 \
1576                     d->elem(11) = F(d->elem(11), s->elem(11),           \
1577                                     ((imm >> 11) & 1));                 \
1578                     d->elem(12) = F(d->elem(12), s->elem(12),           \
1579                                     ((imm >> 12) & 1));                 \
1580                     d->elem(13) = F(d->elem(13), s->elem(13),           \
1581                                     ((imm >> 13) & 1));                 \
1582                     d->elem(14) = F(d->elem(14), s->elem(14),           \
1583                                     ((imm >> 14) & 1));                 \
1584                     d->elem(15) = F(d->elem(15), s->elem(15),           \
1585                                     ((imm >> 15) & 1));                 \
1586                 }                                                       \
1587             }                                                           \
1588         }                                                               \
1589     }
1590
1591 /* SSE4.1 op helpers */
1592 #define FBLENDVB(d, s, m) ((m & 0x80) ? s : d)
1593 #define FBLENDVPS(d, s, m) ((m & 0x80000000) ? s : d)
1594 #define FBLENDVPD(d, s, m) ((m & 0x8000000000000000LL) ? s : d)
1595 SSE_HELPER_V(helper_pblendvb, B, 16, FBLENDVB)
1596 SSE_HELPER_V(helper_blendvps, L, 4, FBLENDVPS)
1597 SSE_HELPER_V(helper_blendvpd, Q, 2, FBLENDVPD)
1598
1599 void glue(helper_ptest, SUFFIX)(Reg *d, Reg *s)
1600 {
1601     uint64_t zf = (s->Q(0) &  d->Q(0)) | (s->Q(1) &  d->Q(1));
1602     uint64_t cf = (s->Q(0) & ~d->Q(0)) | (s->Q(1) & ~d->Q(1));
1603
1604     CC_SRC = (zf ? 0 : CC_Z) | (cf ? 0 : CC_C);
1605 }
1606
1607 #define SSE_HELPER_F(name, elem, num, F)        \
1608     void glue(name, SUFFIX)(Reg *d, Reg *s)     \
1609     {                                           \
1610         d->elem(0) = F(0);                      \
1611         d->elem(1) = F(1);                      \
1612         if (num > 2) {                          \
1613             d->elem(2) = F(2);                  \
1614             d->elem(3) = F(3);                  \
1615             if (num > 4) {                      \
1616                 d->elem(4) = F(4);              \
1617                 d->elem(5) = F(5);              \
1618                 d->elem(6) = F(6);              \
1619                 d->elem(7) = F(7);              \
1620             }                                   \
1621         }                                       \
1622     }
1623
1624 SSE_HELPER_F(helper_pmovsxbw, W, 8, (int8_t) s->B)
1625 SSE_HELPER_F(helper_pmovsxbd, L, 4, (int8_t) s->B)
1626 SSE_HELPER_F(helper_pmovsxbq, Q, 2, (int8_t) s->B)
1627 SSE_HELPER_F(helper_pmovsxwd, L, 4, (int16_t) s->W)
1628 SSE_HELPER_F(helper_pmovsxwq, Q, 2, (int16_t) s->W)
1629 SSE_HELPER_F(helper_pmovsxdq, Q, 2, (int32_t) s->L)
1630 SSE_HELPER_F(helper_pmovzxbw, W, 8, s->B)
1631 SSE_HELPER_F(helper_pmovzxbd, L, 4, s->B)
1632 SSE_HELPER_F(helper_pmovzxbq, Q, 2, s->B)
1633 SSE_HELPER_F(helper_pmovzxwd, L, 4, s->W)
1634 SSE_HELPER_F(helper_pmovzxwq, Q, 2, s->W)
1635 SSE_HELPER_F(helper_pmovzxdq, Q, 2, s->L)
1636
1637 void glue(helper_pmuldq, SUFFIX)(Reg *d, Reg *s)
1638 {
1639     d->Q(0) = (int64_t)(int32_t) d->L(0) * (int32_t) s->L(0);
1640     d->Q(1) = (int64_t)(int32_t) d->L(2) * (int32_t) s->L(2);
1641 }
1642
1643 #define FCMPEQQ(d, s) (d == s ? -1 : 0)
1644 SSE_HELPER_Q(helper_pcmpeqq, FCMPEQQ)
1645
1646 void glue(helper_packusdw, SUFFIX)(Reg *d, Reg *s)
1647 {
1648     d->W(0) = satuw((int32_t) d->L(0));
1649     d->W(1) = satuw((int32_t) d->L(1));
1650     d->W(2) = satuw((int32_t) d->L(2));
1651     d->W(3) = satuw((int32_t) d->L(3));
1652     d->W(4) = satuw((int32_t) s->L(0));
1653     d->W(5) = satuw((int32_t) s->L(1));
1654     d->W(6) = satuw((int32_t) s->L(2));
1655     d->W(7) = satuw((int32_t) s->L(3));
1656 }
1657
1658 #define FMINSB(d, s) MIN((int8_t)d, (int8_t)s)
1659 #define FMINSD(d, s) MIN((int32_t)d, (int32_t)s)
1660 #define FMAXSB(d, s) MAX((int8_t)d, (int8_t)s)
1661 #define FMAXSD(d, s) MAX((int32_t)d, (int32_t)s)
1662 SSE_HELPER_B(helper_pminsb, FMINSB)
1663 SSE_HELPER_L(helper_pminsd, FMINSD)
1664 SSE_HELPER_W(helper_pminuw, MIN)
1665 SSE_HELPER_L(helper_pminud, MIN)
1666 SSE_HELPER_B(helper_pmaxsb, FMAXSB)
1667 SSE_HELPER_L(helper_pmaxsd, FMAXSD)
1668 SSE_HELPER_W(helper_pmaxuw, MAX)
1669 SSE_HELPER_L(helper_pmaxud, MAX)
1670
1671 #define FMULLD(d, s) ((int32_t)d * (int32_t)s)
1672 SSE_HELPER_L(helper_pmulld, FMULLD)
1673
1674 void glue(helper_phminposuw, SUFFIX)(Reg *d, Reg *s)
1675 {
1676     int idx = 0;
1677
1678     if (s->W(1) < s->W(idx)) {
1679         idx = 1;
1680     }
1681     if (s->W(2) < s->W(idx)) {
1682         idx = 2;
1683     }
1684     if (s->W(3) < s->W(idx)) {
1685         idx = 3;
1686     }
1687     if (s->W(4) < s->W(idx)) {
1688         idx = 4;
1689     }
1690     if (s->W(5) < s->W(idx)) {
1691         idx = 5;
1692     }
1693     if (s->W(6) < s->W(idx)) {
1694         idx = 6;
1695     }
1696     if (s->W(7) < s->W(idx)) {
1697         idx = 7;
1698     }
1699
1700     d->Q(1) = 0;
1701     d->L(1) = 0;
1702     d->W(1) = idx;
1703     d->W(0) = s->W(idx);
1704 }
1705
1706 void glue(helper_roundps, SUFFIX)(Reg *d, Reg *s, uint32_t mode)
1707 {
1708     signed char prev_rounding_mode;
1709
1710     prev_rounding_mode = env->sse_status.float_rounding_mode;
1711     if (!(mode & (1 << 2))) {
1712         switch (mode & 3) {
1713         case 0:
1714             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1715             break;
1716         case 1:
1717             set_float_rounding_mode(float_round_down, &env->sse_status);
1718             break;
1719         case 2:
1720             set_float_rounding_mode(float_round_up, &env->sse_status);
1721             break;
1722         case 3:
1723             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1724             break;
1725         }
1726     }
1727
1728     d->XMM_S(0) = float32_round_to_int(s->XMM_S(0), &env->sse_status);
1729     d->XMM_S(1) = float32_round_to_int(s->XMM_S(1), &env->sse_status);
1730     d->XMM_S(2) = float32_round_to_int(s->XMM_S(2), &env->sse_status);
1731     d->XMM_S(3) = float32_round_to_int(s->XMM_S(3), &env->sse_status);
1732
1733 #if 0 /* TODO */
1734     if (mode & (1 << 3)) {
1735         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1736                                   ~float_flag_inexact,
1737                                   &env->sse_status);
1738     }
1739 #endif
1740     env->sse_status.float_rounding_mode = prev_rounding_mode;
1741 }
1742
1743 void glue(helper_roundpd, SUFFIX)(Reg *d, Reg *s, uint32_t mode)
1744 {
1745     signed char prev_rounding_mode;
1746
1747     prev_rounding_mode = env->sse_status.float_rounding_mode;
1748     if (!(mode & (1 << 2))) {
1749         switch (mode & 3) {
1750         case 0:
1751             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1752             break;
1753         case 1:
1754             set_float_rounding_mode(float_round_down, &env->sse_status);
1755             break;
1756         case 2:
1757             set_float_rounding_mode(float_round_up, &env->sse_status);
1758             break;
1759         case 3:
1760             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1761             break;
1762         }
1763     }
1764
1765     d->XMM_D(0) = float64_round_to_int(s->XMM_D(0), &env->sse_status);
1766     d->XMM_D(1) = float64_round_to_int(s->XMM_D(1), &env->sse_status);
1767
1768 #if 0 /* TODO */
1769     if (mode & (1 << 3)) {
1770         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1771                                   ~float_flag_inexact,
1772                                   &env->sse_status);
1773     }
1774 #endif
1775     env->sse_status.float_rounding_mode = prev_rounding_mode;
1776 }
1777
1778 void glue(helper_roundss, SUFFIX)(Reg *d, Reg *s, uint32_t mode)
1779 {
1780     signed char prev_rounding_mode;
1781
1782     prev_rounding_mode = env->sse_status.float_rounding_mode;
1783     if (!(mode & (1 << 2))) {
1784         switch (mode & 3) {
1785         case 0:
1786             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1787             break;
1788         case 1:
1789             set_float_rounding_mode(float_round_down, &env->sse_status);
1790             break;
1791         case 2:
1792             set_float_rounding_mode(float_round_up, &env->sse_status);
1793             break;
1794         case 3:
1795             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1796             break;
1797         }
1798     }
1799
1800     d->XMM_S(0) = float32_round_to_int(s->XMM_S(0), &env->sse_status);
1801
1802 #if 0 /* TODO */
1803     if (mode & (1 << 3)) {
1804         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1805                                   ~float_flag_inexact,
1806                                   &env->sse_status);
1807     }
1808 #endif
1809     env->sse_status.float_rounding_mode = prev_rounding_mode;
1810 }
1811
1812 void glue(helper_roundsd, SUFFIX)(Reg *d, Reg *s, uint32_t mode)
1813 {
1814     signed char prev_rounding_mode;
1815
1816     prev_rounding_mode = env->sse_status.float_rounding_mode;
1817     if (!(mode & (1 << 2))) {
1818         switch (mode & 3) {
1819         case 0:
1820             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1821             break;
1822         case 1:
1823             set_float_rounding_mode(float_round_down, &env->sse_status);
1824             break;
1825         case 2:
1826             set_float_rounding_mode(float_round_up, &env->sse_status);
1827             break;
1828         case 3:
1829             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1830             break;
1831         }
1832     }
1833
1834     d->XMM_D(0) = float64_round_to_int(s->XMM_D(0), &env->sse_status);
1835
1836 #if 0 /* TODO */
1837     if (mode & (1 << 3)) {
1838         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1839                                   ~float_flag_inexact,
1840                                   &env->sse_status);
1841     }
1842 #endif
1843     env->sse_status.float_rounding_mode = prev_rounding_mode;
1844 }
1845
1846 #define FBLENDP(d, s, m) (m ? s : d)
1847 SSE_HELPER_I(helper_blendps, L, 4, FBLENDP)
1848 SSE_HELPER_I(helper_blendpd, Q, 2, FBLENDP)
1849 SSE_HELPER_I(helper_pblendw, W, 8, FBLENDP)
1850
1851 void glue(helper_dpps, SUFFIX)(Reg *d, Reg *s, uint32_t mask)
1852 {
1853     float32 iresult = float32_zero;
1854
1855     if (mask & (1 << 4)) {
1856         iresult = float32_add(iresult,
1857                               float32_mul(d->XMM_S(0), s->XMM_S(0),
1858                                           &env->sse_status),
1859                               &env->sse_status);
1860     }
1861     if (mask & (1 << 5)) {
1862         iresult = float32_add(iresult,
1863                               float32_mul(d->XMM_S(1), s->XMM_S(1),
1864                                           &env->sse_status),
1865                               &env->sse_status);
1866     }
1867     if (mask & (1 << 6)) {
1868         iresult = float32_add(iresult,
1869                               float32_mul(d->XMM_S(2), s->XMM_S(2),
1870                                           &env->sse_status),
1871                               &env->sse_status);
1872     }
1873     if (mask & (1 << 7)) {
1874         iresult = float32_add(iresult,
1875                               float32_mul(d->XMM_S(3), s->XMM_S(3),
1876                                           &env->sse_status),
1877                               &env->sse_status);
1878     }
1879     d->XMM_S(0) = (mask & (1 << 0)) ? iresult : float32_zero;
1880     d->XMM_S(1) = (mask & (1 << 1)) ? iresult : float32_zero;
1881     d->XMM_S(2) = (mask & (1 << 2)) ? iresult : float32_zero;
1882     d->XMM_S(3) = (mask & (1 << 3)) ? iresult : float32_zero;
1883 }
1884
1885 void glue(helper_dppd, SUFFIX)(Reg *d, Reg *s, uint32_t mask)
1886 {
1887     float64 iresult = float64_zero;
1888
1889     if (mask & (1 << 4)) {
1890         iresult = float64_add(iresult,
1891                               float64_mul(d->XMM_D(0), s->XMM_D(0),
1892                                           &env->sse_status),
1893                               &env->sse_status);
1894     }
1895     if (mask & (1 << 5)) {
1896         iresult = float64_add(iresult,
1897                               float64_mul(d->XMM_D(1), s->XMM_D(1),
1898                                           &env->sse_status),
1899                               &env->sse_status);
1900     }
1901     d->XMM_D(0) = (mask & (1 << 0)) ? iresult : float64_zero;
1902     d->XMM_D(1) = (mask & (1 << 1)) ? iresult : float64_zero;
1903 }
1904
1905 void glue(helper_mpsadbw, SUFFIX)(Reg *d, Reg *s, uint32_t offset)
1906 {
1907     int s0 = (offset & 3) << 2;
1908     int d0 = (offset & 4) << 0;
1909     int i;
1910     Reg r;
1911
1912     for (i = 0; i < 8; i++, d0++) {
1913         r.W(i) = 0;
1914         r.W(i) += abs1(d->B(d0 + 0) - s->B(s0 + 0));
1915         r.W(i) += abs1(d->B(d0 + 1) - s->B(s0 + 1));
1916         r.W(i) += abs1(d->B(d0 + 2) - s->B(s0 + 2));
1917         r.W(i) += abs1(d->B(d0 + 3) - s->B(s0 + 3));
1918     }
1919
1920     *d = r;
1921 }
1922
1923 /* SSE4.2 op helpers */
1924 /* it's unclear whether signed or unsigned */
1925 #define FCMPGTQ(d, s) (d > s ? -1 : 0)
1926 SSE_HELPER_Q(helper_pcmpgtq, FCMPGTQ)
1927
1928 static inline int pcmp_elen(int reg, uint32_t ctrl)
1929 {
1930     int val;
1931
1932     /* Presence of REX.W is indicated by a bit higher than 7 set */
1933     if (ctrl >> 8) {
1934         val = abs1((int64_t)env->regs[reg]);
1935     } else {
1936         val = abs1((int32_t)env->regs[reg]);
1937     }
1938
1939     if (ctrl & 1) {
1940         if (val > 8) {
1941             return 8;
1942         }
1943     } else {
1944         if (val > 16) {
1945             return 16;
1946         }
1947     }
1948     return val;
1949 }
1950
1951 static inline int pcmp_ilen(Reg *r, uint8_t ctrl)
1952 {
1953     int val = 0;
1954
1955     if (ctrl & 1) {
1956         while (val < 8 && r->W(val)) {
1957             val++;
1958         }
1959     } else {
1960         while (val < 16 && r->B(val)) {
1961             val++;
1962         }
1963     }
1964
1965     return val;
1966 }
1967
1968 static inline int pcmp_val(Reg *r, uint8_t ctrl, int i)
1969 {
1970     switch ((ctrl >> 0) & 3) {
1971     case 0:
1972         return r->B(i);
1973     case 1:
1974         return r->W(i);
1975     case 2:
1976         return (int8_t)r->B(i);
1977     case 3:
1978     default:
1979         return (int16_t)r->W(i);
1980     }
1981 }
1982
1983 static inline unsigned pcmpxstrx(Reg *d, Reg *s,
1984                                  int8_t ctrl, int valids, int validd)
1985 {
1986     unsigned int res = 0;
1987     int v;
1988     int j, i;
1989     int upper = (ctrl & 1) ? 7 : 15;
1990
1991     valids--;
1992     validd--;
1993
1994     CC_SRC = (valids < upper ? CC_Z : 0) | (validd < upper ? CC_S : 0);
1995
1996     switch ((ctrl >> 2) & 3) {
1997     case 0:
1998         for (j = valids; j >= 0; j--) {
1999             res <<= 1;
2000             v = pcmp_val(s, ctrl, j);
2001             for (i = validd; i >= 0; i--) {
2002                 res |= (v == pcmp_val(d, ctrl, i));
2003             }
2004         }
2005         break;
2006     case 1:
2007         for (j = valids; j >= 0; j--) {
2008             res <<= 1;
2009             v = pcmp_val(s, ctrl, j);
2010             for (i = ((validd - 1) | 1); i >= 0; i -= 2) {
2011                 res |= (pcmp_val(d, ctrl, i - 0) <= v &&
2012                         pcmp_val(d, ctrl, i - 1) >= v);
2013             }
2014         }
2015         break;
2016     case 2:
2017         res = (2 << (upper - MAX(valids, validd))) - 1;
2018         res <<= MAX(valids, validd) - MIN(valids, validd);
2019         for (i = MIN(valids, validd); i >= 0; i--) {
2020             res <<= 1;
2021             v = pcmp_val(s, ctrl, i);
2022             res |= (v == pcmp_val(d, ctrl, i));
2023         }
2024         break;
2025     case 3:
2026         for (j = valids - validd; j >= 0; j--) {
2027             res <<= 1;
2028             res |= 1;
2029             for (i = MIN(upper - j, validd); i >= 0; i--) {
2030                 res &= (pcmp_val(s, ctrl, i + j) == pcmp_val(d, ctrl, i));
2031             }
2032         }
2033         break;
2034     }
2035
2036     switch ((ctrl >> 4) & 3) {
2037     case 1:
2038         res ^= (2 << upper) - 1;
2039         break;
2040     case 3:
2041         res ^= (2 << valids) - 1;
2042         break;
2043     }
2044
2045     if (res) {
2046         CC_SRC |= CC_C;
2047     }
2048     if (res & 1) {
2049         CC_SRC |= CC_O;
2050     }
2051
2052     return res;
2053 }
2054
2055 static inline int rffs1(unsigned int val)
2056 {
2057     int ret = 1, hi;
2058
2059     for (hi = sizeof(val) * 4; hi; hi /= 2) {
2060         if (val >> hi) {
2061             val >>= hi;
2062             ret += hi;
2063         }
2064     }
2065
2066     return ret;
2067 }
2068
2069 static inline int ffs1(unsigned int val)
2070 {
2071     int ret = 1, hi;
2072
2073     for (hi = sizeof(val) * 4; hi; hi /= 2) {
2074         if (val << hi) {
2075             val <<= hi;
2076             ret += hi;
2077         }
2078     }
2079
2080     return ret;
2081 }
2082
2083 void glue(helper_pcmpestri, SUFFIX)(Reg *d, Reg *s, uint32_t ctrl)
2084 {
2085     unsigned int res = pcmpxstrx(d, s, ctrl,
2086                                  pcmp_elen(R_EDX, ctrl),
2087                                  pcmp_elen(R_EAX, ctrl));
2088
2089     if (res) {
2090         env->regs[R_ECX] = ((ctrl & (1 << 6)) ? rffs1 : ffs1)(res) - 1;
2091     } else {
2092         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2093     }
2094 }
2095
2096 void glue(helper_pcmpestrm, SUFFIX)(Reg *d, Reg *s, uint32_t ctrl)
2097 {
2098     int i;
2099     unsigned int res = pcmpxstrx(d, s, ctrl,
2100                                  pcmp_elen(R_EDX, ctrl),
2101                                  pcmp_elen(R_EAX, ctrl));
2102
2103     if ((ctrl >> 6) & 1) {
2104         if (ctrl & 1) {
2105             for (i = 0; i < 8; i++, res >>= 1) {
2106                 d->W(i) = (res & 1) ? ~0 : 0;
2107             }
2108         } else {
2109             for (i = 0; i < 16; i++, res >>= 1) {
2110                 d->B(i) = (res & 1) ? ~0 : 0;
2111             }
2112         }
2113     } else {
2114         d->Q(1) = 0;
2115         d->Q(0) = res;
2116     }
2117 }
2118
2119 void glue(helper_pcmpistri, SUFFIX)(Reg *d, Reg *s, uint32_t ctrl)
2120 {
2121     unsigned int res = pcmpxstrx(d, s, ctrl,
2122                                  pcmp_ilen(s, ctrl),
2123                                  pcmp_ilen(d, ctrl));
2124
2125     if (res) {
2126         env->regs[R_ECX] = ((ctrl & (1 << 6)) ? rffs1 : ffs1)(res) - 1;
2127     } else {
2128         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2129     }
2130 }
2131
2132 void glue(helper_pcmpistrm, SUFFIX)(Reg *d, Reg *s, uint32_t ctrl)
2133 {
2134     int i;
2135     unsigned int res = pcmpxstrx(d, s, ctrl,
2136                                  pcmp_ilen(s, ctrl),
2137                                  pcmp_ilen(d, ctrl));
2138
2139     if ((ctrl >> 6) & 1) {
2140         if (ctrl & 1) {
2141             for (i = 0; i < 8; i++, res >>= 1) {
2142                 d->W(i) = (res & 1) ? ~0 : 0;
2143             }
2144         } else {
2145             for (i = 0; i < 16; i++, res >>= 1) {
2146                 d->B(i) = (res & 1) ? ~0 : 0;
2147             }
2148         }
2149     } else {
2150         d->Q(1) = 0;
2151         d->Q(0) = res;
2152     }
2153 }
2154
2155 #define CRCPOLY        0x1edc6f41
2156 #define CRCPOLY_BITREV 0x82f63b78
2157 target_ulong helper_crc32(uint32_t crc1, target_ulong msg, uint32_t len)
2158 {
2159     target_ulong crc = (msg & ((target_ulong) -1 >>
2160                                (TARGET_LONG_BITS - len))) ^ crc1;
2161
2162     while (len--) {
2163         crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_BITREV : 0);
2164     }
2165
2166     return crc;
2167 }
2168
2169 #define POPMASK(i)     ((target_ulong) -1 / ((1LL << (1 << i)) + 1))
2170 #define POPCOUNT(n, i) ((n & POPMASK(i)) + ((n >> (1 << i)) & POPMASK(i)))
2171 target_ulong helper_popcnt(target_ulong n, uint32_t type)
2172 {
2173     CC_SRC = n ? 0 : CC_Z;
2174
2175     n = POPCOUNT(n, 0);
2176     n = POPCOUNT(n, 1);
2177     n = POPCOUNT(n, 2);
2178     n = POPCOUNT(n, 3);
2179     if (type == 1) {
2180         return n & 0xff;
2181     }
2182
2183     n = POPCOUNT(n, 4);
2184 #ifndef TARGET_X86_64
2185     return n;
2186 #else
2187     if (type == 2) {
2188         return n & 0xff;
2189     }
2190
2191     return POPCOUNT(n, 5);
2192 #endif
2193 }
2194 #endif
2195
2196 #undef SHIFT
2197 #undef XMM_ONLY
2198 #undef Reg
2199 #undef B
2200 #undef W
2201 #undef L
2202 #undef Q
2203 #undef SUFFIX