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     for(i = 0; i < 16 - shift; i++)
 209         d->B(i) = d->B(i + shift);
 210     for(i = 16 - shift; i < 16; i++)
 211         d->B(i) = 0;
 212 }
 213
 214 void glue(helper_pslldq, SUFFIX)(Reg *d, Reg *s)
 215 {
 216     int shift, i;
 217
 218     shift = s->L(0);
 219     if (shift > 16)
 220         shift = 16;
 221     for(i = 15; i >= shift; i--)
 222         d->B(i) = d->B(i - shift);
 223     for(i = 0; i < shift; i++)
 224         d->B(i) = 0;
 225 }
 226 #endif
 227
 228 #define SSE_HELPER_B(name, F)\
 229 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 230 {\
 231     d->B(0) = F(d->B(0), s->B(0));\
 232     d->B(1) = F(d->B(1), s->B(1));\
 233     d->B(2) = F(d->B(2), s->B(2));\
 234     d->B(3) = F(d->B(3), s->B(3));\
 235     d->B(4) = F(d->B(4), s->B(4));\
 236     d->B(5) = F(d->B(5), s->B(5));\
 237     d->B(6) = F(d->B(6), s->B(6));\
 238     d->B(7) = F(d->B(7), s->B(7));\
 239     XMM_ONLY(\
 240     d->B(8) = F(d->B(8), s->B(8));\
 241     d->B(9) = F(d->B(9), s->B(9));\
 242     d->B(10) = F(d->B(10), s->B(10));\
 243     d->B(11) = F(d->B(11), s->B(11));\
 244     d->B(12) = F(d->B(12), s->B(12));\
 245     d->B(13) = F(d->B(13), s->B(13));\
 246     d->B(14) = F(d->B(14), s->B(14));\
 247     d->B(15) = F(d->B(15), s->B(15));\
 248     )\
 249 }
 250
 251 #define SSE_HELPER_W(name, F)\
 252 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 253 {\
 254     d->W(0) = F(d->W(0), s->W(0));\
 255     d->W(1) = F(d->W(1), s->W(1));\
 256     d->W(2) = F(d->W(2), s->W(2));\
 257     d->W(3) = F(d->W(3), s->W(3));\
 258     XMM_ONLY(\
 259     d->W(4) = F(d->W(4), s->W(4));\
 260     d->W(5) = F(d->W(5), s->W(5));\
 261     d->W(6) = F(d->W(6), s->W(6));\
 262     d->W(7) = F(d->W(7), s->W(7));\
 263     )\
 264 }
 265
 266 #define SSE_HELPER_L(name, F)\
 267 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 268 {\
 269     d->L(0) = F(d->L(0), s->L(0));\
 270     d->L(1) = F(d->L(1), s->L(1));\
 271     XMM_ONLY(\
 272     d->L(2) = F(d->L(2), s->L(2));\
 273     d->L(3) = F(d->L(3), s->L(3));\
 274     )\
 275 }
 276
 277 #define SSE_HELPER_Q(name, F)\
 278 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 279 {\
 280     d->Q(0) = F(d->Q(0), s->Q(0));\
 281     XMM_ONLY(\
 282     d->Q(1) = F(d->Q(1), s->Q(1));\
 283     )\
 284 }
 285
 286 #if SHIFT == 0
 287 static inline int satub(int x)
 288 {
 289     if (x < 0)
 290         return 0;
 291     else if (x > 255)
 292         return 255;
 293     else
 294         return x;
 295 }
 296
 297 static inline int satuw(int x)
 298 {
 299     if (x < 0)
 300         return 0;
 301     else if (x > 65535)
 302         return 65535;
 303     else
 304         return x;
 305 }
 306
 307 static inline int satsb(int x)
 308 {
 309     if (x < -128)
 310         return -128;
 311     else if (x > 127)
 312         return 127;
 313     else
 314         return x;
 315 }
 316
 317 static inline int satsw(int x)
 318 {
 319     if (x < -32768)
 320         return -32768;
 321     else if (x > 32767)
 322         return 32767;
 323     else
 324         return x;
 325 }
 326
 327 #define FADD(a, b) ((a) + (b))
 328 #define FADDUB(a, b) satub((a) + (b))
 329 #define FADDUW(a, b) satuw((a) + (b))
 330 #define FADDSB(a, b) satsb((int8_t)(a) + (int8_t)(b))
 331 #define FADDSW(a, b) satsw((int16_t)(a) + (int16_t)(b))
 332
 333 #define FSUB(a, b) ((a) - (b))
 334 #define FSUBUB(a, b) satub((a) - (b))
 335 #define FSUBUW(a, b) satuw((a) - (b))
 336 #define FSUBSB(a, b) satsb((int8_t)(a) - (int8_t)(b))
 337 #define FSUBSW(a, b) satsw((int16_t)(a) - (int16_t)(b))
 338 #define FMINUB(a, b) ((a) < (b)) ? (a) : (b)
 339 #define FMINSW(a, b) ((int16_t)(a) < (int16_t)(b)) ? (a) : (b)
 340 #define FMAXUB(a, b) ((a) > (b)) ? (a) : (b)
 341 #define FMAXSW(a, b) ((int16_t)(a) > (int16_t)(b)) ? (a) : (b)
 342
 343 #define FAND(a, b) (a) & (b)
 344 #define FANDN(a, b) ((~(a)) & (b))
 345 #define FOR(a, b) (a) | (b)
 346 #define FXOR(a, b) (a) ^ (b)
 347
 348 #define FCMPGTB(a, b) (int8_t)(a) > (int8_t)(b) ? -1 : 0
 349 #define FCMPGTW(a, b) (int16_t)(a) > (int16_t)(b) ? -1 : 0
 350 #define FCMPGTL(a, b) (int32_t)(a) > (int32_t)(b) ? -1 : 0
 351 #define FCMPEQ(a, b) (a) == (b) ? -1 : 0
 352
 353 #define FMULLW(a, b) (a) * (b)
 354 #define FMULHRW(a, b) ((int16_t)(a) * (int16_t)(b) + 0x8000) >> 16
 355 #define FMULHUW(a, b) (a) * (b) >> 16
 356 #define FMULHW(a, b) (int16_t)(a) * (int16_t)(b) >> 16
 357
 358 #define FAVG(a, b) ((a) + (b) + 1) >> 1
 359 #endif
 360
 361 SSE_HELPER_B(helper_paddb, FADD)
 362 SSE_HELPER_W(helper_paddw, FADD)
 363 SSE_HELPER_L(helper_paddl, FADD)
 364 SSE_HELPER_Q(helper_paddq, FADD)
 365
 366 SSE_HELPER_B(helper_psubb, FSUB)
 367 SSE_HELPER_W(helper_psubw, FSUB)
 368 SSE_HELPER_L(helper_psubl, FSUB)
 369 SSE_HELPER_Q(helper_psubq, FSUB)
 370
 371 SSE_HELPER_B(helper_paddusb, FADDUB)
 372 SSE_HELPER_B(helper_paddsb, FADDSB)
 373 SSE_HELPER_B(helper_psubusb, FSUBUB)
 374 SSE_HELPER_B(helper_psubsb, FSUBSB)
 375
 376 SSE_HELPER_W(helper_paddusw, FADDUW)
 377 SSE_HELPER_W(helper_paddsw, FADDSW)
 378 SSE_HELPER_W(helper_psubusw, FSUBUW)
 379 SSE_HELPER_W(helper_psubsw, FSUBSW)
 380
 381 SSE_HELPER_B(helper_pminub, FMINUB)
 382 SSE_HELPER_B(helper_pmaxub, FMAXUB)
 383
 384 SSE_HELPER_W(helper_pminsw, FMINSW)
 385 SSE_HELPER_W(helper_pmaxsw, FMAXSW)
 386
 387 SSE_HELPER_Q(helper_pand, FAND)
 388 SSE_HELPER_Q(helper_pandn, FANDN)
 389 SSE_HELPER_Q(helper_por, FOR)
 390 SSE_HELPER_Q(helper_pxor, FXOR)
 391
 392 SSE_HELPER_B(helper_pcmpgtb, FCMPGTB)
 393 SSE_HELPER_W(helper_pcmpgtw, FCMPGTW)
 394 SSE_HELPER_L(helper_pcmpgtl, FCMPGTL)
 395
 396 SSE_HELPER_B(helper_pcmpeqb, FCMPEQ)
 397 SSE_HELPER_W(helper_pcmpeqw, FCMPEQ)
 398 SSE_HELPER_L(helper_pcmpeql, FCMPEQ)
 399
 400 SSE_HELPER_W(helper_pmullw, FMULLW)
 401 #if SHIFT == 0
 402 SSE_HELPER_W(helper_pmulhrw, FMULHRW)
 403 #endif
 404 SSE_HELPER_W(helper_pmulhuw, FMULHUW)
 405 SSE_HELPER_W(helper_pmulhw, FMULHW)
 406
 407 SSE_HELPER_B(helper_pavgb, FAVG)
 408 SSE_HELPER_W(helper_pavgw, FAVG)
 409
 410 void glue(helper_pmuludq, SUFFIX) (Reg *d, Reg *s)
 411 {
 412     d->Q(0) = (uint64_t)s->L(0) * (uint64_t)d->L(0);
 413 #if SHIFT == 1
 414     d->Q(1) = (uint64_t)s->L(2) * (uint64_t)d->L(2);
 415 #endif
 416 }
 417
 418 void glue(helper_pmaddwd, SUFFIX) (Reg *d, Reg *s)
 419 {
 420     int i;
 421
 422     for(i = 0; i < (2 << SHIFT); i++) {
 423         d->L(i) = (int16_t)s->W(2*i) * (int16_t)d->W(2*i) +
 424             (int16_t)s->W(2*i+1) * (int16_t)d->W(2*i+1);
 425     }
 426 }
 427
 428 #if SHIFT == 0
 429 static inline int abs1(int a)
 430 {
 431     if (a < 0)
 432         return -a;
 433     else
 434         return a;
 435 }
 436 #endif
 437 void glue(helper_psadbw, SUFFIX) (Reg *d, Reg *s)
 438 {
 439     unsigned int val;
 440
 441     val = 0;
 442     val += abs1(d->B(0) - s->B(0));
 443     val += abs1(d->B(1) - s->B(1));
 444     val += abs1(d->B(2) - s->B(2));
 445     val += abs1(d->B(3) - s->B(3));
 446     val += abs1(d->B(4) - s->B(4));
 447     val += abs1(d->B(5) - s->B(5));
 448     val += abs1(d->B(6) - s->B(6));
 449     val += abs1(d->B(7) - s->B(7));
 450     d->Q(0) = val;
 451 #if SHIFT == 1
 452     val = 0;
 453     val += abs1(d->B(8) - s->B(8));
 454     val += abs1(d->B(9) - s->B(9));
 455     val += abs1(d->B(10) - s->B(10));
 456     val += abs1(d->B(11) - s->B(11));
 457     val += abs1(d->B(12) - s->B(12));
 458     val += abs1(d->B(13) - s->B(13));
 459     val += abs1(d->B(14) - s->B(14));
 460     val += abs1(d->B(15) - s->B(15));
 461     d->Q(1) = val;
 462 #endif
 463 }
 464
 465 void glue(helper_maskmov, SUFFIX) (Reg *d, Reg *s, target_ulong a0)
 466 {
 467     int i;
 468     for(i = 0; i < (8 << SHIFT); i++) {
 469         if (s->B(i) & 0x80)
 470             stb(a0 + i, d->B(i));
 471     }
 472 }
 473
 474 void glue(helper_movl_mm_T0, SUFFIX) (Reg *d, uint32_t val)
 475 {
 476     d->L(0) = val;
 477     d->L(1) = 0;
 478 #if SHIFT == 1
 479     d->Q(1) = 0;
 480 #endif
 481 }
 482
 483 #ifdef TARGET_X86_64
 484 void glue(helper_movq_mm_T0, SUFFIX) (Reg *d, uint64_t val)
 485 {
 486     d->Q(0) = val;
 487 #if SHIFT == 1
 488     d->Q(1) = 0;
 489 #endif
 490 }
 491 #endif
 492
 493 #if SHIFT == 0
 494 void glue(helper_pshufw, SUFFIX) (Reg *d, Reg *s, int order)
 495 {
 496     Reg r;
 497     r.W(0) = s->W(order & 3);
 498     r.W(1) = s->W((order >> 2) & 3);
 499     r.W(2) = s->W((order >> 4) & 3);
 500     r.W(3) = s->W((order >> 6) & 3);
 501     *d = r;
 502 }
 503 #else
 504 void helper_shufps(Reg *d, Reg *s, int order)
 505 {
 506     Reg r;
 507     r.L(0) = d->L(order & 3);
 508     r.L(1) = d->L((order >> 2) & 3);
 509     r.L(2) = s->L((order >> 4) & 3);
 510     r.L(3) = s->L((order >> 6) & 3);
 511     *d = r;
 512 }
 513
 514 void helper_shufpd(Reg *d, Reg *s, int order)
 515 {
 516     Reg r;
 517     r.Q(0) = d->Q(order & 1);
 518     r.Q(1) = s->Q((order >> 1) & 1);
 519     *d = r;
 520 }
 521
 522 void glue(helper_pshufd, SUFFIX) (Reg *d, Reg *s, int order)
 523 {
 524     Reg r;
 525     r.L(0) = s->L(order & 3);
 526     r.L(1) = s->L((order >> 2) & 3);
 527     r.L(2) = s->L((order >> 4) & 3);
 528     r.L(3) = s->L((order >> 6) & 3);
 529     *d = r;
 530 }
 531
 532 void glue(helper_pshuflw, SUFFIX) (Reg *d, Reg *s, int order)
 533 {
 534     Reg r;
 535     r.W(0) = s->W(order & 3);
 536     r.W(1) = s->W((order >> 2) & 3);
 537     r.W(2) = s->W((order >> 4) & 3);
 538     r.W(3) = s->W((order >> 6) & 3);
 539     r.Q(1) = s->Q(1);
 540     *d = r;
 541 }
 542
 543 void glue(helper_pshufhw, SUFFIX) (Reg *d, Reg *s, int order)
 544 {
 545     Reg r;
 546     r.Q(0) = s->Q(0);
 547     r.W(4) = s->W(4 + (order & 3));
 548     r.W(5) = s->W(4 + ((order >> 2) & 3));
 549     r.W(6) = s->W(4 + ((order >> 4) & 3));
 550     r.W(7) = s->W(4 + ((order >> 6) & 3));
 551     *d = r;
 552 }
 553 #endif
 554
 555 #if SHIFT == 1
 556 /* FPU ops */
 557 /* XXX: not accurate */
 558
 559 #define SSE_HELPER_S(name, F)\
 560 void helper_ ## name ## ps (Reg *d, Reg *s)\
 561 {\
 562     d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 563     d->XMM_S(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
 564     d->XMM_S(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
 565     d->XMM_S(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
 566 }\
 567 \
 568 void helper_ ## name ## ss (Reg *d, Reg *s)\
 569 {\
 570     d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 571 }\
 572 void helper_ ## name ## pd (Reg *d, Reg *s)\
 573 {\
 574     d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 575     d->XMM_D(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
 576 }\
 577 \
 578 void helper_ ## name ## sd (Reg *d, Reg *s)\
 579 {\
 580     d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 581 }
 582
 583 #define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
 584 #define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
 585 #define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
 586 #define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
 587 #define FPU_MIN(size, a, b) (a) < (b) ? (a) : (b)
 588 #define FPU_MAX(size, a, b) (a) > (b) ? (a) : (b)
 589 #define FPU_SQRT(size, a, b) float ## size ## _sqrt(b, &env->sse_status)
 590
 591 SSE_HELPER_S(add, FPU_ADD)
 592 SSE_HELPER_S(sub, FPU_SUB)
 593 SSE_HELPER_S(mul, FPU_MUL)
 594 SSE_HELPER_S(div, FPU_DIV)
 595 SSE_HELPER_S(min, FPU_MIN)
 596 SSE_HELPER_S(max, FPU_MAX)
 597 SSE_HELPER_S(sqrt, FPU_SQRT)
 598
 599
 600 /* float to float conversions */
 601 void helper_cvtps2pd(Reg *d, Reg *s)
 602 {
 603     float32 s0, s1;
 604     s0 = s->XMM_S(0);
 605     s1 = s->XMM_S(1);
 606     d->XMM_D(0) = float32_to_float64(s0, &env->sse_status);
 607     d->XMM_D(1) = float32_to_float64(s1, &env->sse_status);
 608 }
 609
 610 void helper_cvtpd2ps(Reg *d, Reg *s)
 611 {
 612     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 613     d->XMM_S(1) = float64_to_float32(s->XMM_D(1), &env->sse_status);
 614     d->Q(1) = 0;
 615 }
 616
 617 void helper_cvtss2sd(Reg *d, Reg *s)
 618 {
 619     d->XMM_D(0) = float32_to_float64(s->XMM_S(0), &env->sse_status);
 620 }
 621
 622 void helper_cvtsd2ss(Reg *d, Reg *s)
 623 {
 624     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 625 }
 626
 627 /* integer to float */
 628 void helper_cvtdq2ps(Reg *d, Reg *s)
 629 {
 630     d->XMM_S(0) = int32_to_float32(s->XMM_L(0), &env->sse_status);
 631     d->XMM_S(1) = int32_to_float32(s->XMM_L(1), &env->sse_status);
 632     d->XMM_S(2) = int32_to_float32(s->XMM_L(2), &env->sse_status);
 633     d->XMM_S(3) = int32_to_float32(s->XMM_L(3), &env->sse_status);
 634 }
 635
 636 void helper_cvtdq2pd(Reg *d, Reg *s)
 637 {
 638     int32_t l0, l1;
 639     l0 = (int32_t)s->XMM_L(0);
 640     l1 = (int32_t)s->XMM_L(1);
 641     d->XMM_D(0) = int32_to_float64(l0, &env->sse_status);
 642     d->XMM_D(1) = int32_to_float64(l1, &env->sse_status);
 643 }
 644
 645 void helper_cvtpi2ps(XMMReg *d, MMXReg *s)
 646 {
 647     d->XMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
 648     d->XMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
 649 }
 650
 651 void helper_cvtpi2pd(XMMReg *d, MMXReg *s)
 652 {
 653     d->XMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
 654     d->XMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
 655 }
 656
 657 void helper_cvtsi2ss(XMMReg *d, uint32_t val)
 658 {
 659     d->XMM_S(0) = int32_to_float32(val, &env->sse_status);
 660 }
 661
 662 void helper_cvtsi2sd(XMMReg *d, uint32_t val)
 663 {
 664     d->XMM_D(0) = int32_to_float64(val, &env->sse_status);
 665 }
 666
 667 #ifdef TARGET_X86_64
 668 void helper_cvtsq2ss(XMMReg *d, uint64_t val)
 669 {
 670     d->XMM_S(0) = int64_to_float32(val, &env->sse_status);
 671 }
 672
 673 void helper_cvtsq2sd(XMMReg *d, uint64_t val)
 674 {
 675     d->XMM_D(0) = int64_to_float64(val, &env->sse_status);
 676 }
 677 #endif
 678
 679 /* float to integer */
 680 void helper_cvtps2dq(XMMReg *d, XMMReg *s)
 681 {
 682     d->XMM_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 683     d->XMM_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 684     d->XMM_L(2) = float32_to_int32(s->XMM_S(2), &env->sse_status);
 685     d->XMM_L(3) = float32_to_int32(s->XMM_S(3), &env->sse_status);
 686 }
 687
 688 void helper_cvtpd2dq(XMMReg *d, XMMReg *s)
 689 {
 690     d->XMM_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 691     d->XMM_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 692     d->XMM_Q(1) = 0;
 693 }
 694
 695 void helper_cvtps2pi(MMXReg *d, XMMReg *s)
 696 {
 697     d->MMX_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 698     d->MMX_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 699 }
 700
 701 void helper_cvtpd2pi(MMXReg *d, XMMReg *s)
 702 {
 703     d->MMX_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 704     d->MMX_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 705 }
 706
 707 int32_t helper_cvtss2si(XMMReg *s)
 708 {
 709     return float32_to_int32(s->XMM_S(0), &env->sse_status);
 710 }
 711
 712 int32_t helper_cvtsd2si(XMMReg *s)
 713 {
 714     return float64_to_int32(s->XMM_D(0), &env->sse_status);
 715 }
 716
 717 #ifdef TARGET_X86_64
 718 int64_t helper_cvtss2sq(XMMReg *s)
 719 {
 720     return float32_to_int64(s->XMM_S(0), &env->sse_status);
 721 }
 722
 723 int64_t helper_cvtsd2sq(XMMReg *s)
 724 {
 725     return float64_to_int64(s->XMM_D(0), &env->sse_status);
 726 }
 727 #endif
 728
 729 /* float to integer truncated */
 730 void helper_cvttps2dq(XMMReg *d, XMMReg *s)
 731 {
 732     d->XMM_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 733     d->XMM_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 734     d->XMM_L(2) = float32_to_int32_round_to_zero(s->XMM_S(2), &env->sse_status);
 735     d->XMM_L(3) = float32_to_int32_round_to_zero(s->XMM_S(3), &env->sse_status);
 736 }
 737
 738 void helper_cvttpd2dq(XMMReg *d, XMMReg *s)
 739 {
 740     d->XMM_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 741     d->XMM_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 742     d->XMM_Q(1) = 0;
 743 }
 744
 745 void helper_cvttps2pi(MMXReg *d, XMMReg *s)
 746 {
 747     d->MMX_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 748     d->MMX_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 749 }
 750
 751 void helper_cvttpd2pi(MMXReg *d, XMMReg *s)
 752 {
 753     d->MMX_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 754     d->MMX_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 755 }
 756
 757 int32_t helper_cvttss2si(XMMReg *s)
 758 {
 759     return float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 760 }
 761
 762 int32_t helper_cvttsd2si(XMMReg *s)
 763 {
 764     return float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 765 }
 766
 767 #ifdef TARGET_X86_64
 768 int64_t helper_cvttss2sq(XMMReg *s)
 769 {
 770     return float32_to_int64_round_to_zero(s->XMM_S(0), &env->sse_status);
 771 }
 772
 773 int64_t helper_cvttsd2sq(XMMReg *s)
 774 {
 775     return float64_to_int64_round_to_zero(s->XMM_D(0), &env->sse_status);
 776 }
 777 #endif
 778
 779 void helper_rsqrtps(XMMReg *d, XMMReg *s)
 780 {
 781     d->XMM_S(0) = approx_rsqrt(s->XMM_S(0));
 782     d->XMM_S(1) = approx_rsqrt(s->XMM_S(1));
 783     d->XMM_S(2) = approx_rsqrt(s->XMM_S(2));
 784     d->XMM_S(3) = approx_rsqrt(s->XMM_S(3));
 785 }
 786
 787 void helper_rsqrtss(XMMReg *d, XMMReg *s)
 788 {
 789     d->XMM_S(0) = approx_rsqrt(s->XMM_S(0));
 790 }
 791
 792 void helper_rcpps(XMMReg *d, XMMReg *s)
 793 {
 794     d->XMM_S(0) = approx_rcp(s->XMM_S(0));
 795     d->XMM_S(1) = approx_rcp(s->XMM_S(1));
 796     d->XMM_S(2) = approx_rcp(s->XMM_S(2));
 797     d->XMM_S(3) = approx_rcp(s->XMM_S(3));
 798 }
 799
 800 void helper_rcpss(XMMReg *d, XMMReg *s)
 801 {
 802     d->XMM_S(0) = approx_rcp(s->XMM_S(0));
 803 }
 804
 805 void helper_haddps(XMMReg *d, XMMReg *s)
 806 {
 807     XMMReg r;
 808     r.XMM_S(0) = d->XMM_S(0) + d->XMM_S(1);
 809     r.XMM_S(1) = d->XMM_S(2) + d->XMM_S(3);
 810     r.XMM_S(2) = s->XMM_S(0) + s->XMM_S(1);
 811     r.XMM_S(3) = s->XMM_S(2) + s->XMM_S(3);
 812     *d = r;
 813 }
 814
 815 void helper_haddpd(XMMReg *d, XMMReg *s)
 816 {
 817     XMMReg r;
 818     r.XMM_D(0) = d->XMM_D(0) + d->XMM_D(1);
 819     r.XMM_D(1) = s->XMM_D(0) + s->XMM_D(1);
 820     *d = r;
 821 }
 822
 823 void helper_hsubps(XMMReg *d, XMMReg *s)
 824 {
 825     XMMReg r;
 826     r.XMM_S(0) = d->XMM_S(0) - d->XMM_S(1);
 827     r.XMM_S(1) = d->XMM_S(2) - d->XMM_S(3);
 828     r.XMM_S(2) = s->XMM_S(0) - s->XMM_S(1);
 829     r.XMM_S(3) = s->XMM_S(2) - s->XMM_S(3);
 830     *d = r;
 831 }
 832
 833 void helper_hsubpd(XMMReg *d, XMMReg *s)
 834 {
 835     XMMReg r;
 836     r.XMM_D(0) = d->XMM_D(0) - d->XMM_D(1);
 837     r.XMM_D(1) = s->XMM_D(0) - s->XMM_D(1);
 838     *d = r;
 839 }
 840
 841 void helper_addsubps(XMMReg *d, XMMReg *s)
 842 {
 843     d->XMM_S(0) = d->XMM_S(0) - s->XMM_S(0);
 844     d->XMM_S(1) = d->XMM_S(1) + s->XMM_S(1);
 845     d->XMM_S(2) = d->XMM_S(2) - s->XMM_S(2);
 846     d->XMM_S(3) = d->XMM_S(3) + s->XMM_S(3);
 847 }
 848
 849 void helper_addsubpd(XMMReg *d, XMMReg *s)
 850 {
 851     d->XMM_D(0) = d->XMM_D(0) - s->XMM_D(0);
 852     d->XMM_D(1) = d->XMM_D(1) + s->XMM_D(1);
 853 }
 854
 855 /* XXX: unordered */
 856 #define SSE_HELPER_CMP(name, F)\
 857 void helper_ ## name ## ps (Reg *d, Reg *s)\
 858 {\
 859     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 860     d->XMM_L(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
 861     d->XMM_L(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
 862     d->XMM_L(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
 863 }\
 864 \
 865 void helper_ ## name ## ss (Reg *d, Reg *s)\
 866 {\
 867     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 868 }\
 869 void helper_ ## name ## pd (Reg *d, Reg *s)\
 870 {\
 871     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 872     d->XMM_Q(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
 873 }\
 874 \
 875 void helper_ ## name ## sd (Reg *d, Reg *s)\
 876 {\
 877     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 878 }
 879
 880 #define FPU_CMPEQ(size, a, b) float ## size ## _eq(a, b, &env->sse_status) ? -1 : 0
 881 #define FPU_CMPLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? -1 : 0
 882 #define FPU_CMPLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? -1 : 0
 883 #define FPU_CMPUNORD(size, a, b) float ## size ## _unordered(a, b, &env->sse_status) ? - 1 : 0
 884 #define FPU_CMPNEQ(size, a, b) float ## size ## _eq(a, b, &env->sse_status) ? 0 : -1
 885 #define FPU_CMPNLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? 0 : -1
 886 #define FPU_CMPNLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? 0 : -1
 887 #define FPU_CMPORD(size, a, b) float ## size ## _unordered(a, b, &env->sse_status) ? 0 : -1
 888
 889 SSE_HELPER_CMP(cmpeq, FPU_CMPEQ)
 890 SSE_HELPER_CMP(cmplt, FPU_CMPLT)
 891 SSE_HELPER_CMP(cmple, FPU_CMPLE)
 892 SSE_HELPER_CMP(cmpunord, FPU_CMPUNORD)
 893 SSE_HELPER_CMP(cmpneq, FPU_CMPNEQ)
 894 SSE_HELPER_CMP(cmpnlt, FPU_CMPNLT)
 895 SSE_HELPER_CMP(cmpnle, FPU_CMPNLE)
 896 SSE_HELPER_CMP(cmpord, FPU_CMPORD)
 897
 898 const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
 899
 900 void helper_ucomiss(Reg *d, Reg *s)
 901 {
 902     int ret;
 903     float32 s0, s1;
 904
 905     s0 = d->XMM_S(0);
 906     s1 = s->XMM_S(0);
 907     ret = float32_compare_quiet(s0, s1, &env->sse_status);
 908     CC_SRC = comis_eflags[ret + 1];
 909 }
 910
 911 void helper_comiss(Reg *d, Reg *s)
 912 {
 913     int ret;
 914     float32 s0, s1;
 915
 916     s0 = d->XMM_S(0);
 917     s1 = s->XMM_S(0);
 918     ret = float32_compare(s0, s1, &env->sse_status);
 919     CC_SRC = comis_eflags[ret + 1];
 920 }
 921
 922 void helper_ucomisd(Reg *d, Reg *s)
 923 {
 924     int ret;
 925     float64 d0, d1;
 926
 927     d0 = d->XMM_D(0);
 928     d1 = s->XMM_D(0);
 929     ret = float64_compare_quiet(d0, d1, &env->sse_status);
 930     CC_SRC = comis_eflags[ret + 1];
 931 }
 932
 933 void helper_comisd(Reg *d, Reg *s)
 934 {
 935     int ret;
 936     float64 d0, d1;
 937
 938     d0 = d->XMM_D(0);
 939     d1 = s->XMM_D(0);
 940     ret = float64_compare(d0, d1, &env->sse_status);
 941     CC_SRC = comis_eflags[ret + 1];
 942 }
 943
 944 uint32_t helper_movmskps(Reg *s)
 945 {
 946     int b0, b1, b2, b3;
 947     b0 = s->XMM_L(0) >> 31;
 948     b1 = s->XMM_L(1) >> 31;
 949     b2 = s->XMM_L(2) >> 31;
 950     b3 = s->XMM_L(3) >> 31;
 951     return b0 | (b1 << 1) | (b2 << 2) | (b3 << 3);
 952 }
 953
 954 uint32_t helper_movmskpd(Reg *s)
 955 {
 956     int b0, b1;
 957     b0 = s->XMM_L(1) >> 31;
 958     b1 = s->XMM_L(3) >> 31;
 959     return b0 | (b1 << 1);
 960 }
 961
 962 #endif
 963
 964 uint32_t glue(helper_pmovmskb, SUFFIX)(Reg *s)
 965 {
 966     uint32_t val;
 967     val = 0;
 968     val |= (s->B(0) >> 7);
 969     val |= (s->B(1) >> 6) & 0x02;
 970     val |= (s->B(2) >> 5) & 0x04;
 971     val |= (s->B(3) >> 4) & 0x08;
 972     val |= (s->B(4) >> 3) & 0x10;
 973     val |= (s->B(5) >> 2) & 0x20;
 974     val |= (s->B(6) >> 1) & 0x40;
 975     val |= (s->B(7)) & 0x80;
 976 #if SHIFT == 1
 977     val |= (s->B(8) << 1) & 0x0100;
 978     val |= (s->B(9) << 2) & 0x0200;
 979     val |= (s->B(10) << 3) & 0x0400;
 980     val |= (s->B(11) << 4) & 0x0800;
 981     val |= (s->B(12) << 5) & 0x1000;
 982     val |= (s->B(13) << 6) & 0x2000;
 983     val |= (s->B(14) << 7) & 0x4000;
 984     val |= (s->B(15) << 8) & 0x8000;
 985 #endif
 986     return val;
 987 }
 988
 989 void glue(helper_packsswb, SUFFIX) (Reg *d, Reg *s)
 990 {
 991     Reg r;
 992
 993     r.B(0) = satsb((int16_t)d->W(0));
 994     r.B(1) = satsb((int16_t)d->W(1));
 995     r.B(2) = satsb((int16_t)d->W(2));
 996     r.B(3) = satsb((int16_t)d->W(3));
 997 #if SHIFT == 1
 998     r.B(4) = satsb((int16_t)d->W(4));
 999     r.B(5) = satsb((int16_t)d->W(5));
1000     r.B(6) = satsb((int16_t)d->W(6));
1001     r.B(7) = satsb((int16_t)d->W(7));
1002 #endif
1003     r.B((4 << SHIFT) + 0) = satsb((int16_t)s->W(0));
1004     r.B((4 << SHIFT) + 1) = satsb((int16_t)s->W(1));
1005     r.B((4 << SHIFT) + 2) = satsb((int16_t)s->W(2));
1006     r.B((4 << SHIFT) + 3) = satsb((int16_t)s->W(3));
1007 #if SHIFT == 1
1008     r.B(12) = satsb((int16_t)s->W(4));
1009     r.B(13) = satsb((int16_t)s->W(5));
1010     r.B(14) = satsb((int16_t)s->W(6));
1011     r.B(15) = satsb((int16_t)s->W(7));
1012 #endif
1013     *d = r;
1014 }
1015
1016 void glue(helper_packuswb, SUFFIX) (Reg *d, Reg *s)
1017 {
1018     Reg r;
1019
1020     r.B(0) = satub((int16_t)d->W(0));
1021     r.B(1) = satub((int16_t)d->W(1));
1022     r.B(2) = satub((int16_t)d->W(2));
1023     r.B(3) = satub((int16_t)d->W(3));
1024 #if SHIFT == 1
1025     r.B(4) = satub((int16_t)d->W(4));
1026     r.B(5) = satub((int16_t)d->W(5));
1027     r.B(6) = satub((int16_t)d->W(6));
1028     r.B(7) = satub((int16_t)d->W(7));
1029 #endif
1030     r.B((4 << SHIFT) + 0) = satub((int16_t)s->W(0));
1031     r.B((4 << SHIFT) + 1) = satub((int16_t)s->W(1));
1032     r.B((4 << SHIFT) + 2) = satub((int16_t)s->W(2));
1033     r.B((4 << SHIFT) + 3) = satub((int16_t)s->W(3));
1034 #if SHIFT == 1
1035     r.B(12) = satub((int16_t)s->W(4));
1036     r.B(13) = satub((int16_t)s->W(5));
1037     r.B(14) = satub((int16_t)s->W(6));
1038     r.B(15) = satub((int16_t)s->W(7));
1039 #endif
1040     *d = r;
1041 }
1042
1043 void glue(helper_packssdw, SUFFIX) (Reg *d, Reg *s)
1044 {
1045     Reg r;
1046
1047     r.W(0) = satsw(d->L(0));
1048     r.W(1) = satsw(d->L(1));
1049 #if SHIFT == 1
1050     r.W(2) = satsw(d->L(2));
1051     r.W(3) = satsw(d->L(3));
1052 #endif
1053     r.W((2 << SHIFT) + 0) = satsw(s->L(0));
1054     r.W((2 << SHIFT) + 1) = satsw(s->L(1));
1055 #if SHIFT == 1
1056     r.W(6) = satsw(s->L(2));
1057     r.W(7) = satsw(s->L(3));
1058 #endif
1059     *d = r;
1060 }
1061
1062 #define UNPCK_OP(base_name, base)                               \
1063                                                                 \
1064 void glue(helper_punpck ## base_name ## bw, SUFFIX) (Reg *d, Reg *s)   \
1065 {                                                               \
1066     Reg r;                                              \
1067                                                                 \
1068     r.B(0) = d->B((base << (SHIFT + 2)) + 0);                   \
1069     r.B(1) = s->B((base << (SHIFT + 2)) + 0);                   \
1070     r.B(2) = d->B((base << (SHIFT + 2)) + 1);                   \
1071     r.B(3) = s->B((base << (SHIFT + 2)) + 1);                   \
1072     r.B(4) = d->B((base << (SHIFT + 2)) + 2);                   \
1073     r.B(5) = s->B((base << (SHIFT + 2)) + 2);                   \
1074     r.B(6) = d->B((base << (SHIFT + 2)) + 3);                   \
1075     r.B(7) = s->B((base << (SHIFT + 2)) + 3);                   \
1076 XMM_ONLY(                                                       \
1077     r.B(8) = d->B((base << (SHIFT + 2)) + 4);                   \
1078     r.B(9) = s->B((base << (SHIFT + 2)) + 4);                   \
1079     r.B(10) = d->B((base << (SHIFT + 2)) + 5);                  \
1080     r.B(11) = s->B((base << (SHIFT + 2)) + 5);                  \
1081     r.B(12) = d->B((base << (SHIFT + 2)) + 6);                  \
1082     r.B(13) = s->B((base << (SHIFT + 2)) + 6);                  \
1083     r.B(14) = d->B((base << (SHIFT + 2)) + 7);                  \
1084     r.B(15) = s->B((base << (SHIFT + 2)) + 7);                  \
1085 )                                                               \
1086     *d = r;                                                     \
1087 }                                                               \
1088                                                                 \
1089 void glue(helper_punpck ## base_name ## wd, SUFFIX) (Reg *d, Reg *s)   \
1090 {                                                               \
1091     Reg r;                                              \
1092                                                                 \
1093     r.W(0) = d->W((base << (SHIFT + 1)) + 0);                   \
1094     r.W(1) = s->W((base << (SHIFT + 1)) + 0);                   \
1095     r.W(2) = d->W((base << (SHIFT + 1)) + 1);                   \
1096     r.W(3) = s->W((base << (SHIFT + 1)) + 1);                   \
1097 XMM_ONLY(                                                       \
1098     r.W(4) = d->W((base << (SHIFT + 1)) + 2);                   \
1099     r.W(5) = s->W((base << (SHIFT + 1)) + 2);                   \
1100     r.W(6) = d->W((base << (SHIFT + 1)) + 3);                   \
1101     r.W(7) = s->W((base << (SHIFT + 1)) + 3);                   \
1102 )                                                               \
1103     *d = r;                                                     \
1104 }                                                               \
1105                                                                 \
1106 void glue(helper_punpck ## base_name ## dq, SUFFIX) (Reg *d, Reg *s)   \
1107 {                                                               \
1108     Reg r;                                              \
1109                                                                 \
1110     r.L(0) = d->L((base << SHIFT) + 0);                         \
1111     r.L(1) = s->L((base << SHIFT) + 0);                         \
1112 XMM_ONLY(                                                       \
1113     r.L(2) = d->L((base << SHIFT) + 1);                         \
1114     r.L(3) = s->L((base << SHIFT) + 1);                         \
1115 )                                                               \
1116     *d = r;                                                     \
1117 }                                                               \
1118                                                                 \
1119 XMM_ONLY(                                                       \
1120 void glue(helper_punpck ## base_name ## qdq, SUFFIX) (Reg *d, Reg *s)  \
1121 {                                                               \
1122     Reg r;                                              \
1123                                                                 \
1124     r.Q(0) = d->Q(base);                                        \
1125     r.Q(1) = s->Q(base);                                        \
1126     *d = r;                                                     \
1127 }                                                               \
1128 )
1129
1130 UNPCK_OP(l, 0)
1131 UNPCK_OP(h, 1)
1132
1133 /* 3DNow! float ops */
1134 #if SHIFT == 0
1135 void helper_pi2fd(MMXReg *d, MMXReg *s)
1136 {
1137     d->MMX_S(0) = int32_to_float32(s->MMX_L(0), &env->mmx_status);
1138     d->MMX_S(1) = int32_to_float32(s->MMX_L(1), &env->mmx_status);
1139 }
1140
1141 void helper_pi2fw(MMXReg *d, MMXReg *s)
1142 {
1143     d->MMX_S(0) = int32_to_float32((int16_t)s->MMX_W(0), &env->mmx_status);
1144     d->MMX_S(1) = int32_to_float32((int16_t)s->MMX_W(2), &env->mmx_status);
1145 }
1146
1147 void helper_pf2id(MMXReg *d, MMXReg *s)
1148 {
1149     d->MMX_L(0) = float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status);
1150     d->MMX_L(1) = float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status);
1151 }
1152
1153 void helper_pf2iw(MMXReg *d, MMXReg *s)
1154 {
1155     d->MMX_L(0) = satsw(float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status));
1156     d->MMX_L(1) = satsw(float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status));
1157 }
1158
1159 void helper_pfacc(MMXReg *d, MMXReg *s)
1160 {
1161     MMXReg r;
1162     r.MMX_S(0) = float32_add(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1163     r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1164     *d = r;
1165 }
1166
1167 void helper_pfadd(MMXReg *d, MMXReg *s)
1168 {
1169     d->MMX_S(0) = float32_add(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1170     d->MMX_S(1) = float32_add(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1171 }
1172
1173 void helper_pfcmpeq(MMXReg *d, MMXReg *s)
1174 {
1175     d->MMX_L(0) = float32_eq(d->MMX_S(0), s->MMX_S(0), &env->mmx_status) ? -1 : 0;
1176     d->MMX_L(1) = float32_eq(d->MMX_S(1), s->MMX_S(1), &env->mmx_status) ? -1 : 0;
1177 }
1178
1179 void helper_pfcmpge(MMXReg *d, MMXReg *s)
1180 {
1181     d->MMX_L(0) = float32_le(s->MMX_S(0), d->MMX_S(0), &env->mmx_status) ? -1 : 0;
1182     d->MMX_L(1) = float32_le(s->MMX_S(1), d->MMX_S(1), &env->mmx_status) ? -1 : 0;
1183 }
1184
1185 void helper_pfcmpgt(MMXReg *d, MMXReg *s)
1186 {
1187     d->MMX_L(0) = float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status) ? -1 : 0;
1188     d->MMX_L(1) = float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status) ? -1 : 0;
1189 }
1190
1191 void helper_pfmax(MMXReg *d, MMXReg *s)
1192 {
1193     if (float32_lt(d->MMX_S(0), s->MMX_S(0), &env->mmx_status))
1194         d->MMX_S(0) = s->MMX_S(0);
1195     if (float32_lt(d->MMX_S(1), s->MMX_S(1), &env->mmx_status))
1196         d->MMX_S(1) = s->MMX_S(1);
1197 }
1198
1199 void helper_pfmin(MMXReg *d, MMXReg *s)
1200 {
1201     if (float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status))
1202         d->MMX_S(0) = s->MMX_S(0);
1203     if (float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status))
1204         d->MMX_S(1) = s->MMX_S(1);
1205 }
1206
1207 void helper_pfmul(MMXReg *d, MMXReg *s)
1208 {
1209     d->MMX_S(0) = float32_mul(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1210     d->MMX_S(1) = float32_mul(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1211 }
1212
1213 void helper_pfnacc(MMXReg *d, MMXReg *s)
1214 {
1215     MMXReg r;
1216     r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1217     r.MMX_S(1) = float32_sub(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1218     *d = r;
1219 }
1220
1221 void helper_pfpnacc(MMXReg *d, MMXReg *s)
1222 {
1223     MMXReg r;
1224     r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1225     r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1226     *d = r;
1227 }
1228
1229 void helper_pfrcp(MMXReg *d, MMXReg *s)
1230 {
1231     d->MMX_S(0) = approx_rcp(s->MMX_S(0));
1232     d->MMX_S(1) = d->MMX_S(0);
1233 }
1234
1235 void helper_pfrsqrt(MMXReg *d, MMXReg *s)
1236 {
1237     d->MMX_L(1) = s->MMX_L(0) & 0x7fffffff;
1238     d->MMX_S(1) = approx_rsqrt(d->MMX_S(1));
1239     d->MMX_L(1) |= s->MMX_L(0) & 0x80000000;
1240     d->MMX_L(0) = d->MMX_L(1);
1241 }
1242
1243 void helper_pfsub(MMXReg *d, MMXReg *s)
1244 {
1245     d->MMX_S(0) = float32_sub(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1246     d->MMX_S(1) = float32_sub(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1247 }
1248
1249 void helper_pfsubr(MMXReg *d, MMXReg *s)
1250 {
1251     d->MMX_S(0) = float32_sub(s->MMX_S(0), d->MMX_S(0), &env->mmx_status);
1252     d->MMX_S(1) = float32_sub(s->MMX_S(1), d->MMX_S(1), &env->mmx_status);
1253 }
1254
1255 void helper_pswapd(MMXReg *d, MMXReg *s)
1256 {
1257     MMXReg r;
1258     r.MMX_L(0) = s->MMX_L(1);
1259     r.MMX_L(1) = s->MMX_L(0);
1260     *d = r;
1261 }
1262 #endif
1263
1264 /* SSSE3 op helpers */
1265 void glue(helper_pshufb, SUFFIX) (Reg *d, Reg *s)
1266 {
1267     int i;
1268     Reg r;
1269
1270     for (i = 0; i < (8 << SHIFT); i++)
1271         r.B(i) = (s->B(i) & 0x80) ? 0 : (d->B(s->B(i) & ((8 << SHIFT) - 1)));
1272
1273     *d = r;
1274 }
1275
1276 void glue(helper_phaddw, SUFFIX) (Reg *d, Reg *s)
1277 {
1278     d->W(0) = (int16_t)d->W(0) + (int16_t)d->W(1);
1279     d->W(1) = (int16_t)d->W(2) + (int16_t)d->W(3);
1280     XMM_ONLY(d->W(2) = (int16_t)d->W(4) + (int16_t)d->W(5));
1281     XMM_ONLY(d->W(3) = (int16_t)d->W(6) + (int16_t)d->W(7));
1282     d->W((2 << SHIFT) + 0) = (int16_t)s->W(0) + (int16_t)s->W(1);
1283     d->W((2 << SHIFT) + 1) = (int16_t)s->W(2) + (int16_t)s->W(3);
1284     XMM_ONLY(d->W(6) = (int16_t)s->W(4) + (int16_t)s->W(5));
1285     XMM_ONLY(d->W(7) = (int16_t)s->W(6) + (int16_t)s->W(7));
1286 }
1287
1288 void glue(helper_phaddd, SUFFIX) (Reg *d, Reg *s)
1289 {
1290     d->L(0) = (int32_t)d->L(0) + (int32_t)d->L(1);
1291     XMM_ONLY(d->L(1) = (int32_t)d->L(2) + (int32_t)d->L(3));
1292     d->L((1 << SHIFT) + 0) = (int32_t)s->L(0) + (int32_t)s->L(1);
1293     XMM_ONLY(d->L(3) = (int32_t)s->L(2) + (int32_t)s->L(3));
1294 }
1295
1296 void glue(helper_phaddsw, SUFFIX) (Reg *d, Reg *s)
1297 {
1298     d->W(0) = satsw((int16_t)d->W(0) + (int16_t)d->W(1));
1299     d->W(1) = satsw((int16_t)d->W(2) + (int16_t)d->W(3));
1300     XMM_ONLY(d->W(2) = satsw((int16_t)d->W(4) + (int16_t)d->W(5)));
1301     XMM_ONLY(d->W(3) = satsw((int16_t)d->W(6) + (int16_t)d->W(7)));
1302     d->W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) + (int16_t)s->W(1));
1303     d->W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) + (int16_t)s->W(3));
1304     XMM_ONLY(d->W(6) = satsw((int16_t)s->W(4) + (int16_t)s->W(5)));
1305     XMM_ONLY(d->W(7) = satsw((int16_t)s->W(6) + (int16_t)s->W(7)));
1306 }
1307
1308 void glue(helper_pmaddubsw, SUFFIX) (Reg *d, Reg *s)
1309 {
1310     d->W(0) = satsw((int8_t)s->B( 0) * (uint8_t)d->B( 0) +
1311                     (int8_t)s->B( 1) * (uint8_t)d->B( 1));
1312     d->W(1) = satsw((int8_t)s->B( 2) * (uint8_t)d->B( 2) +
1313                     (int8_t)s->B( 3) * (uint8_t)d->B( 3));
1314     d->W(2) = satsw((int8_t)s->B( 4) * (uint8_t)d->B( 4) +
1315                     (int8_t)s->B( 5) * (uint8_t)d->B( 5));
1316     d->W(3) = satsw((int8_t)s->B( 6) * (uint8_t)d->B( 6) +
1317                     (int8_t)s->B( 7) * (uint8_t)d->B( 7));
1318 #if SHIFT == 1
1319     d->W(4) = satsw((int8_t)s->B( 8) * (uint8_t)d->B( 8) +
1320                     (int8_t)s->B( 9) * (uint8_t)d->B( 9));
1321     d->W(5) = satsw((int8_t)s->B(10) * (uint8_t)d->B(10) +
1322                     (int8_t)s->B(11) * (uint8_t)d->B(11));
1323     d->W(6) = satsw((int8_t)s->B(12) * (uint8_t)d->B(12) +
1324                     (int8_t)s->B(13) * (uint8_t)d->B(13));
1325     d->W(7) = satsw((int8_t)s->B(14) * (uint8_t)d->B(14) +
1326                     (int8_t)s->B(15) * (uint8_t)d->B(15));
1327 #endif
1328 }
1329
1330 void glue(helper_phsubw, SUFFIX) (Reg *d, Reg *s)
1331 {
1332     d->W(0) = (int16_t)d->W(0) - (int16_t)d->W(1);
1333     d->W(1) = (int16_t)d->W(2) - (int16_t)d->W(3);
1334     XMM_ONLY(d->W(2) = (int16_t)d->W(4) - (int16_t)d->W(5));
1335     XMM_ONLY(d->W(3) = (int16_t)d->W(6) - (int16_t)d->W(7));
1336     d->W((2 << SHIFT) + 0) = (int16_t)s->W(0) - (int16_t)s->W(1);
1337     d->W((2 << SHIFT) + 1) = (int16_t)s->W(2) - (int16_t)s->W(3);
1338     XMM_ONLY(d->W(6) = (int16_t)s->W(4) - (int16_t)s->W(5));
1339     XMM_ONLY(d->W(7) = (int16_t)s->W(6) - (int16_t)s->W(7));
1340 }
1341
1342 void glue(helper_phsubd, SUFFIX) (Reg *d, Reg *s)
1343 {
1344     d->L(0) = (int32_t)d->L(0) - (int32_t)d->L(1);
1345     XMM_ONLY(d->L(1) = (int32_t)d->L(2) - (int32_t)d->L(3));
1346     d->L((1 << SHIFT) + 0) = (int32_t)s->L(0) - (int32_t)s->L(1);
1347     XMM_ONLY(d->L(3) = (int32_t)s->L(2) - (int32_t)s->L(3));
1348 }
1349
1350 void glue(helper_phsubsw, SUFFIX) (Reg *d, Reg *s)
1351 {
1352     d->W(0) = satsw((int16_t)d->W(0) - (int16_t)d->W(1));
1353     d->W(1) = satsw((int16_t)d->W(2) - (int16_t)d->W(3));
1354     XMM_ONLY(d->W(2) = satsw((int16_t)d->W(4) - (int16_t)d->W(5)));
1355     XMM_ONLY(d->W(3) = satsw((int16_t)d->W(6) - (int16_t)d->W(7)));
1356     d->W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) - (int16_t)s->W(1));
1357     d->W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) - (int16_t)s->W(3));
1358     XMM_ONLY(d->W(6) = satsw((int16_t)s->W(4) - (int16_t)s->W(5)));
1359     XMM_ONLY(d->W(7) = satsw((int16_t)s->W(6) - (int16_t)s->W(7)));
1360 }
1361
1362 #define FABSB(_, x) x > INT8_MAX  ? -(int8_t ) x : x
1363 #define FABSW(_, x) x > INT16_MAX ? -(int16_t) x : x
1364 #define FABSL(_, x) x > INT32_MAX ? -(int32_t) x : x
1365 SSE_HELPER_B(helper_pabsb, FABSB)
1366 SSE_HELPER_W(helper_pabsw, FABSW)
1367 SSE_HELPER_L(helper_pabsd, FABSL)
1368
1369 #define FMULHRSW(d, s) ((int16_t) d * (int16_t) s + 0x4000) >> 15
1370 SSE_HELPER_W(helper_pmulhrsw, FMULHRSW)
1371
1372 #define FSIGNB(d, s) s <= INT8_MAX  ? s ? d : 0 : -(int8_t ) d
1373 #define FSIGNW(d, s) s <= INT16_MAX ? s ? d : 0 : -(int16_t) d
1374 #define FSIGNL(d, s) s <= INT32_MAX ? s ? d : 0 : -(int32_t) d
1375 SSE_HELPER_B(helper_psignb, FSIGNB)
1376 SSE_HELPER_W(helper_psignw, FSIGNW)
1377 SSE_HELPER_L(helper_psignd, FSIGNL)
1378
1379 void glue(helper_palignr, SUFFIX) (Reg *d, Reg *s, int32_t shift)
1380 {
1381     Reg r;
1382
1383     /* XXX could be checked during translation */
1384     if (shift >= (16 << SHIFT)) {
1385         r.Q(0) = 0;
1386         XMM_ONLY(r.Q(1) = 0);
1387     } else {
1388         shift <<= 3;
1389 #define SHR(v, i) (i < 64 && i > -64 ? i > 0 ? v >> (i) : (v << -(i)) : 0)
1390 #if SHIFT == 0
1391         r.Q(0) = SHR(s->Q(0), shift -   0) |
1392                  SHR(d->Q(0), shift -  64);
1393 #else
1394         r.Q(0) = SHR(s->Q(0), shift -   0) |
1395                  SHR(s->Q(1), shift -  64) |
1396                  SHR(d->Q(0), shift - 128) |
1397                  SHR(d->Q(1), shift - 192);
1398         r.Q(1) = SHR(s->Q(0), shift +  64) |
1399                  SHR(s->Q(1), shift -   0) |
1400                  SHR(d->Q(0), shift -  64) |
1401                  SHR(d->Q(1), shift - 128);
1402 #endif
1403 #undef SHR
1404     }
1405
1406     *d = r;
1407 }
1408
1409 #define XMM0 env->xmm_regs[0]
1410
1411 #if SHIFT == 1
1412 #define SSE_HELPER_V(name, elem, num, F)\
1413 void glue(name, SUFFIX) (Reg *d, Reg *s)\
1414 {\
1415     d->elem(0) = F(d->elem(0), s->elem(0), XMM0.elem(0));\
1416     d->elem(1) = F(d->elem(1), s->elem(1), XMM0.elem(1));\
1417     if (num > 2) {\
1418         d->elem(2) = F(d->elem(2), s->elem(2), XMM0.elem(2));\
1419         d->elem(3) = F(d->elem(3), s->elem(3), XMM0.elem(3));\
1420         if (num > 4) {\
1421             d->elem(4) = F(d->elem(4), s->elem(4), XMM0.elem(4));\
1422             d->elem(5) = F(d->elem(5), s->elem(5), XMM0.elem(5));\
1423             d->elem(6) = F(d->elem(6), s->elem(6), XMM0.elem(6));\
1424             d->elem(7) = F(d->elem(7), s->elem(7), XMM0.elem(7));\
1425             if (num > 8) {\
1426                 d->elem(8) = F(d->elem(8), s->elem(8), XMM0.elem(8));\
1427                 d->elem(9) = F(d->elem(9), s->elem(9), XMM0.elem(9));\
1428                 d->elem(10) = F(d->elem(10), s->elem(10), XMM0.elem(10));\
1429                 d->elem(11) = F(d->elem(11), s->elem(11), XMM0.elem(11));\
1430                 d->elem(12) = F(d->elem(12), s->elem(12), XMM0.elem(12));\
1431                 d->elem(13) = F(d->elem(13), s->elem(13), XMM0.elem(13));\
1432                 d->elem(14) = F(d->elem(14), s->elem(14), XMM0.elem(14));\
1433                 d->elem(15) = F(d->elem(15), s->elem(15), XMM0.elem(15));\
1434             }\
1435         }\
1436     }\
1437 }
1438
1439 #define SSE_HELPER_I(name, elem, num, F)\
1440 void glue(name, SUFFIX) (Reg *d, Reg *s, uint32_t imm)\
1441 {\
1442     d->elem(0) = F(d->elem(0), s->elem(0), ((imm >> 0) & 1));\
1443     d->elem(1) = F(d->elem(1), s->elem(1), ((imm >> 1) & 1));\
1444     if (num > 2) {\
1445         d->elem(2) = F(d->elem(2), s->elem(2), ((imm >> 2) & 1));\
1446         d->elem(3) = F(d->elem(3), s->elem(3), ((imm >> 3) & 1));\
1447         if (num > 4) {\
1448             d->elem(4) = F(d->elem(4), s->elem(4), ((imm >> 4) & 1));\
1449             d->elem(5) = F(d->elem(5), s->elem(5), ((imm >> 5) & 1));\
1450             d->elem(6) = F(d->elem(6), s->elem(6), ((imm >> 6) & 1));\
1451             d->elem(7) = F(d->elem(7), s->elem(7), ((imm >> 7) & 1));\
1452             if (num > 8) {\
1453                 d->elem(8) = F(d->elem(8), s->elem(8), ((imm >> 8) & 1));\
1454                 d->elem(9) = F(d->elem(9), s->elem(9), ((imm >> 9) & 1));\
1455                 d->elem(10) = F(d->elem(10), s->elem(10), ((imm >> 10) & 1));\
1456                 d->elem(11) = F(d->elem(11), s->elem(11), ((imm >> 11) & 1));\
1457                 d->elem(12) = F(d->elem(12), s->elem(12), ((imm >> 12) & 1));\
1458                 d->elem(13) = F(d->elem(13), s->elem(13), ((imm >> 13) & 1));\
1459                 d->elem(14) = F(d->elem(14), s->elem(14), ((imm >> 14) & 1));\
1460                 d->elem(15) = F(d->elem(15), s->elem(15), ((imm >> 15) & 1));\
1461             }\
1462         }\
1463     }\
1464 }
1465
1466 /* SSE4.1 op helpers */
1467 #define FBLENDVB(d, s, m) (m & 0x80) ? s : d
1468 #define FBLENDVPS(d, s, m) (m & 0x80000000) ? s : d
1469 #define FBLENDVPD(d, s, m) (m & 0x8000000000000000LL) ? s : d
1470 SSE_HELPER_V(helper_pblendvb, B, 16, FBLENDVB)
1471 SSE_HELPER_V(helper_blendvps, L, 4, FBLENDVPS)
1472 SSE_HELPER_V(helper_blendvpd, Q, 2, FBLENDVPD)
1473
1474 void glue(helper_ptest, SUFFIX) (Reg *d, Reg *s)
1475 {
1476     uint64_t zf = (s->Q(0) &  d->Q(0)) | (s->Q(1) &  d->Q(1));
1477     uint64_t cf = (s->Q(0) & ~d->Q(0)) | (s->Q(1) & ~d->Q(1));
1478
1479     CC_SRC = (zf ? 0 : CC_Z) | (cf ? 0 : CC_C);
1480 }
1481
1482 #define SSE_HELPER_F(name, elem, num, F)\
1483 void glue(name, SUFFIX) (Reg *d, Reg *s)\
1484 {\
1485     d->elem(0) = F(0);\
1486     d->elem(1) = F(1);\
1487     if (num > 2) {\
1488         d->elem(2) = F(2);\
1489         d->elem(3) = F(3);\
1490         if (num > 4) {\
1491             d->elem(4) = F(4);\
1492             d->elem(5) = F(5);\
1493             d->elem(6) = F(6);\
1494             d->elem(7) = F(7);\
1495         }\
1496     }\
1497 }
1498
1499 SSE_HELPER_F(helper_pmovsxbw, W, 8, (int8_t) s->B)
1500 SSE_HELPER_F(helper_pmovsxbd, L, 4, (int8_t) s->B)
1501 SSE_HELPER_F(helper_pmovsxbq, Q, 2, (int8_t) s->B)
1502 SSE_HELPER_F(helper_pmovsxwd, L, 4, (int16_t) s->W)
1503 SSE_HELPER_F(helper_pmovsxwq, Q, 2, (int16_t) s->W)
1504 SSE_HELPER_F(helper_pmovsxdq, Q, 2, (int32_t) s->L)
1505 SSE_HELPER_F(helper_pmovzxbw, W, 8, s->B)
1506 SSE_HELPER_F(helper_pmovzxbd, L, 4, s->B)
1507 SSE_HELPER_F(helper_pmovzxbq, Q, 2, s->B)
1508 SSE_HELPER_F(helper_pmovzxwd, L, 4, s->W)
1509 SSE_HELPER_F(helper_pmovzxwq, Q, 2, s->W)
1510 SSE_HELPER_F(helper_pmovzxdq, Q, 2, s->L)
1511
1512 void glue(helper_pmuldq, SUFFIX) (Reg *d, Reg *s)
1513 {
1514     d->Q(0) = (int64_t) (int32_t) d->L(0) * (int32_t) s->L(0);
1515     d->Q(1) = (int64_t) (int32_t) d->L(2) * (int32_t) s->L(2);
1516 }
1517
1518 #define FCMPEQQ(d, s) d == s ? -1 : 0
1519 SSE_HELPER_Q(helper_pcmpeqq, FCMPEQQ)
1520
1521 void glue(helper_packusdw, SUFFIX) (Reg *d, Reg *s)
1522 {
1523     d->W(0) = satuw((int32_t) d->L(0));
1524     d->W(1) = satuw((int32_t) d->L(1));
1525     d->W(2) = satuw((int32_t) d->L(2));
1526     d->W(3) = satuw((int32_t) d->L(3));
1527     d->W(4) = satuw((int32_t) s->L(0));
1528     d->W(5) = satuw((int32_t) s->L(1));
1529     d->W(6) = satuw((int32_t) s->L(2));
1530     d->W(7) = satuw((int32_t) s->L(3));
1531 }
1532
1533 #define FMINSB(d, s) MIN((int8_t) d, (int8_t) s)
1534 #define FMINSD(d, s) MIN((int32_t) d, (int32_t) s)
1535 #define FMAXSB(d, s) MAX((int8_t) d, (int8_t) s)
1536 #define FMAXSD(d, s) MAX((int32_t) d, (int32_t) s)
1537 SSE_HELPER_B(helper_pminsb, FMINSB)
1538 SSE_HELPER_L(helper_pminsd, FMINSD)
1539 SSE_HELPER_W(helper_pminuw, MIN)
1540 SSE_HELPER_L(helper_pminud, MIN)
1541 SSE_HELPER_B(helper_pmaxsb, FMAXSB)
1542 SSE_HELPER_L(helper_pmaxsd, FMAXSD)
1543 SSE_HELPER_W(helper_pmaxuw, MAX)
1544 SSE_HELPER_L(helper_pmaxud, MAX)
1545
1546 #define FMULLD(d, s) (int32_t) d * (int32_t) s
1547 SSE_HELPER_L(helper_pmulld, FMULLD)
1548
1549 void glue(helper_phminposuw, SUFFIX) (Reg *d, Reg *s)
1550 {
1551     int idx = 0;
1552
1553     if (s->W(1) < s->W(idx))
1554         idx = 1;
1555     if (s->W(2) < s->W(idx))
1556         idx = 2;
1557     if (s->W(3) < s->W(idx))
1558         idx = 3;
1559     if (s->W(4) < s->W(idx))
1560         idx = 4;
1561     if (s->W(5) < s->W(idx))
1562         idx = 5;
1563     if (s->W(6) < s->W(idx))
1564         idx = 6;
1565     if (s->W(7) < s->W(idx))
1566         idx = 7;
1567
1568     d->Q(1) = 0;
1569     d->L(1) = 0;
1570     d->W(1) = idx;
1571     d->W(0) = s->W(idx);
1572 }
1573
1574 void glue(helper_roundps, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1575 {
1576     signed char prev_rounding_mode;
1577
1578     prev_rounding_mode = env->sse_status.float_rounding_mode;
1579     if (!(mode & (1 << 2)))
1580         switch (mode & 3) {
1581         case 0:
1582             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1583             break;
1584         case 1:
1585             set_float_rounding_mode(float_round_down, &env->sse_status);
1586             break;
1587         case 2:
1588             set_float_rounding_mode(float_round_up, &env->sse_status);
1589             break;
1590         case 3:
1591             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1592             break;
1593         }
1594
1595     d->L(0) = float64_round_to_int(s->L(0), &env->sse_status);
1596     d->L(1) = float64_round_to_int(s->L(1), &env->sse_status);
1597     d->L(2) = float64_round_to_int(s->L(2), &env->sse_status);
1598     d->L(3) = float64_round_to_int(s->L(3), &env->sse_status);
1599
1600 #if 0 /* TODO */
1601     if (mode & (1 << 3))
1602         set_float_exception_flags(
1603                         get_float_exception_flags(&env->sse_status) &
1604                         ~float_flag_inexact,
1605                         &env->sse_status);
1606 #endif
1607     env->sse_status.float_rounding_mode = prev_rounding_mode;
1608 }
1609
1610 void glue(helper_roundpd, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1611 {
1612     signed char prev_rounding_mode;
1613
1614     prev_rounding_mode = env->sse_status.float_rounding_mode;
1615     if (!(mode & (1 << 2)))
1616         switch (mode & 3) {
1617         case 0:
1618             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1619             break;
1620         case 1:
1621             set_float_rounding_mode(float_round_down, &env->sse_status);
1622             break;
1623         case 2:
1624             set_float_rounding_mode(float_round_up, &env->sse_status);
1625             break;
1626         case 3:
1627             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1628             break;
1629         }
1630
1631     d->Q(0) = float64_round_to_int(s->Q(0), &env->sse_status);
1632     d->Q(1) = float64_round_to_int(s->Q(1), &env->sse_status);
1633
1634 #if 0 /* TODO */
1635     if (mode & (1 << 3))
1636         set_float_exception_flags(
1637                         get_float_exception_flags(&env->sse_status) &
1638                         ~float_flag_inexact,
1639                         &env->sse_status);
1640 #endif
1641     env->sse_status.float_rounding_mode = prev_rounding_mode;
1642 }
1643
1644 void glue(helper_roundss, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1645 {
1646     signed char prev_rounding_mode;
1647
1648     prev_rounding_mode = env->sse_status.float_rounding_mode;
1649     if (!(mode & (1 << 2)))
1650         switch (mode & 3) {
1651         case 0:
1652             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1653             break;
1654         case 1:
1655             set_float_rounding_mode(float_round_down, &env->sse_status);
1656             break;
1657         case 2:
1658             set_float_rounding_mode(float_round_up, &env->sse_status);
1659             break;
1660         case 3:
1661             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1662             break;
1663         }
1664
1665     d->L(0) = float64_round_to_int(s->L(0), &env->sse_status);
1666
1667 #if 0 /* TODO */
1668     if (mode & (1 << 3))
1669         set_float_exception_flags(
1670                         get_float_exception_flags(&env->sse_status) &
1671                         ~float_flag_inexact,
1672                         &env->sse_status);
1673 #endif
1674     env->sse_status.float_rounding_mode = prev_rounding_mode;
1675 }
1676
1677 void glue(helper_roundsd, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1678 {
1679     signed char prev_rounding_mode;
1680
1681     prev_rounding_mode = env->sse_status.float_rounding_mode;
1682     if (!(mode & (1 << 2)))
1683         switch (mode & 3) {
1684         case 0:
1685             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1686             break;
1687         case 1:
1688             set_float_rounding_mode(float_round_down, &env->sse_status);
1689             break;
1690         case 2:
1691             set_float_rounding_mode(float_round_up, &env->sse_status);
1692             break;
1693         case 3:
1694             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1695             break;
1696         }
1697
1698     d->Q(0) = float64_round_to_int(s->Q(0), &env->sse_status);
1699
1700 #if 0 /* TODO */
1701     if (mode & (1 << 3))
1702         set_float_exception_flags(
1703                         get_float_exception_flags(&env->sse_status) &
1704                         ~float_flag_inexact,
1705                         &env->sse_status);
1706 #endif
1707     env->sse_status.float_rounding_mode = prev_rounding_mode;
1708 }
1709
1710 #define FBLENDP(d, s, m) m ? s : d
1711 SSE_HELPER_I(helper_blendps, L, 4, FBLENDP)
1712 SSE_HELPER_I(helper_blendpd, Q, 2, FBLENDP)
1713 SSE_HELPER_I(helper_pblendw, W, 8, FBLENDP)
1714
1715 void glue(helper_dpps, SUFFIX) (Reg *d, Reg *s, uint32_t mask)
1716 {
1717     float32 iresult = 0 /*float32_zero*/;
1718
1719     if (mask & (1 << 4))
1720         iresult = float32_add(iresult,
1721                         float32_mul(d->L(0), s->L(0), &env->sse_status),
1722                         &env->sse_status);
1723     if (mask & (1 << 5))
1724         iresult = float32_add(iresult,
1725                         float32_mul(d->L(1), s->L(1), &env->sse_status),
1726                         &env->sse_status);
1727     if (mask & (1 << 6))
1728         iresult = float32_add(iresult,
1729                         float32_mul(d->L(2), s->L(2), &env->sse_status),
1730                         &env->sse_status);
1731     if (mask & (1 << 7))
1732         iresult = float32_add(iresult,
1733                         float32_mul(d->L(3), s->L(3), &env->sse_status),
1734                         &env->sse_status);
1735     d->L(0) = (mask & (1 << 0)) ? iresult : 0 /*float32_zero*/;
1736     d->L(1) = (mask & (1 << 1)) ? iresult : 0 /*float32_zero*/;
1737     d->L(2) = (mask & (1 << 2)) ? iresult : 0 /*float32_zero*/;
1738     d->L(3) = (mask & (1 << 3)) ? iresult : 0 /*float32_zero*/;
1739 }
1740
1741 void glue(helper_dppd, SUFFIX) (Reg *d, Reg *s, uint32_t mask)
1742 {
1743     float64 iresult = 0 /*float64_zero*/;
1744
1745     if (mask & (1 << 4))
1746         iresult = float64_add(iresult,
1747                         float64_mul(d->Q(0), s->Q(0), &env->sse_status),
1748                         &env->sse_status);
1749     if (mask & (1 << 5))
1750         iresult = float64_add(iresult,
1751                         float64_mul(d->Q(1), s->Q(1), &env->sse_status),
1752                         &env->sse_status);
1753     d->Q(0) = (mask & (1 << 0)) ? iresult : 0 /*float64_zero*/;
1754     d->Q(1) = (mask & (1 << 1)) ? iresult : 0 /*float64_zero*/;
1755 }
1756
1757 void glue(helper_mpsadbw, SUFFIX) (Reg *d, Reg *s, uint32_t offset)
1758 {
1759     int s0 = (offset & 3) << 2;
1760     int d0 = (offset & 4) << 0;
1761     int i;
1762     Reg r;
1763
1764     for (i = 0; i < 8; i++, d0++) {
1765         r.W(i) = 0;
1766         r.W(i) += abs1(d->B(d0 + 0) - s->B(s0 + 0));
1767         r.W(i) += abs1(d->B(d0 + 1) - s->B(s0 + 1));
1768         r.W(i) += abs1(d->B(d0 + 2) - s->B(s0 + 2));
1769         r.W(i) += abs1(d->B(d0 + 3) - s->B(s0 + 3));
1770     }
1771
1772     *d = r;
1773 }
1774
1775 /* SSE4.2 op helpers */
1776 /* it's unclear whether signed or unsigned */
1777 #define FCMPGTQ(d, s) d > s ? -1 : 0
1778 SSE_HELPER_Q(helper_pcmpgtq, FCMPGTQ)
1779
1780 static inline int pcmp_elen(int reg, uint32_t ctrl)
1781 {
1782     int val;
1783
1784     /* Presence of REX.W is indicated by a bit higher than 7 set */
1785     if (ctrl >> 8)
1786         val = abs1((int64_t) env->regs[reg]);
1787     else
1788         val = abs1((int32_t) env->regs[reg]);
1789
1790     if (ctrl & 1) {
1791         if (val > 8)
1792             return 8;
1793     } else
1794         if (val > 16)
1795             return 16;
1796
1797     return val;
1798 }
1799
1800 static inline int pcmp_ilen(Reg *r, uint8_t ctrl)
1801 {
1802     int val = 0;
1803
1804     if (ctrl & 1) {
1805         while (val < 8 && r->W(val))
1806             val++;
1807     } else
1808         while (val < 16 && r->B(val))
1809             val++;
1810
1811     return val;
1812 }
1813
1814 static inline int pcmp_val(Reg *r, uint8_t ctrl, int i)
1815 {
1816     switch ((ctrl >> 0) & 3) {
1817     case 0:
1818         return r->B(i);
1819     case 1:
1820         return r->W(i);
1821     case 2:
1822         return (int8_t) r->B(i);
1823     case 3:
1824     default:
1825         return (int16_t) r->W(i);
1826     }
1827 }
1828
1829 static inline unsigned pcmpxstrx(Reg *d, Reg *s,
1830                 int8_t ctrl, int valids, int validd)
1831 {
1832     unsigned int res = 0;
1833     int v;
1834     int j, i;
1835     int upper = (ctrl & 1) ? 7 : 15;
1836
1837     valids--;
1838     validd--;
1839
1840     CC_SRC = (valids < upper ? CC_Z : 0) | (validd < upper ? CC_S : 0);
1841
1842     switch ((ctrl >> 2) & 3) {
1843     case 0:
1844         for (j = valids; j >= 0; j--) {
1845             res <<= 1;
1846             v = pcmp_val(s, ctrl, j);
1847             for (i = validd; i >= 0; i--)
1848                 res |= (v == pcmp_val(d, ctrl, i));
1849         }
1850         break;
1851     case 1:
1852         for (j = valids; j >= 0; j--) {
1853             res <<= 1;
1854             v = pcmp_val(s, ctrl, j);
1855             for (i = ((validd - 1) | 1); i >= 0; i -= 2)
1856                 res |= (pcmp_val(d, ctrl, i - 0) <= v &&
1857                         pcmp_val(d, ctrl, i - 1) >= v);
1858         }
1859         break;
1860     case 2:
1861         res = (2 << (upper - MAX(valids, validd))) - 1;
1862         res <<= MAX(valids, validd) - MIN(valids, validd);
1863         for (i = MIN(valids, validd); i >= 0; i--) {
1864             res <<= 1;
1865             v = pcmp_val(s, ctrl, i);
1866             res |= (v == pcmp_val(d, ctrl, i));
1867         }
1868         break;
1869     case 3:
1870         for (j = valids - validd; j >= 0; j--) {
1871             res <<= 1;
1872             res |= 1;
1873             for (i = MIN(upper - j, validd); i >= 0; i--)
1874                 res &= (pcmp_val(s, ctrl, i + j) == pcmp_val(d, ctrl, i));
1875         }
1876         break;
1877     }
1878
1879     switch ((ctrl >> 4) & 3) {
1880     case 1:
1881         res ^= (2 << upper) - 1;
1882         break;
1883     case 3:
1884         res ^= (2 << valids) - 1;
1885         break;
1886     }
1887
1888     if (res)
1889        CC_SRC |= CC_C;
1890     if (res & 1)
1891        CC_SRC |= CC_O;
1892
1893     return res;
1894 }
1895
1896 static inline int rffs1(unsigned int val)
1897 {
1898     int ret = 1, hi;
1899
1900     for (hi = sizeof(val) * 4; hi; hi /= 2)
1901         if (val >> hi) {
1902             val >>= hi;
1903             ret += hi;
1904         }
1905
1906     return ret;
1907 }
1908
1909 static inline int ffs1(unsigned int val)
1910 {
1911     int ret = 1, hi;
1912
1913     for (hi = sizeof(val) * 4; hi; hi /= 2)
1914         if (val << hi) {
1915             val <<= hi;
1916             ret += hi;
1917         }
1918
1919     return ret;
1920 }
1921
1922 void glue(helper_pcmpestri, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1923 {
1924     unsigned int res = pcmpxstrx(d, s, ctrl,
1925                     pcmp_elen(R_EDX, ctrl),
1926                     pcmp_elen(R_EAX, ctrl));
1927
1928     if (res)
1929         env->regs[R_ECX] = ((ctrl & (1 << 6)) ? rffs1 : ffs1)(res) - 1;
1930     else
1931         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
1932 }
1933
1934 void glue(helper_pcmpestrm, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1935 {
1936     int i;
1937     unsigned int res = pcmpxstrx(d, s, ctrl,
1938                     pcmp_elen(R_EDX, ctrl),
1939                     pcmp_elen(R_EAX, ctrl));
1940
1941     if ((ctrl >> 6) & 1) {
1942         if (ctrl & 1)
1943             for (i = 0; i <= 8; i--, res >>= 1)
1944                 d->W(i) = (res & 1) ? ~0 : 0;
1945         else
1946             for (i = 0; i <= 16; i--, res >>= 1)
1947                 d->B(i) = (res & 1) ? ~0 : 0;
1948     } else {
1949         d->Q(1) = 0;
1950         d->Q(0) = res;
1951     }
1952 }
1953
1954 void glue(helper_pcmpistri, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1955 {
1956     unsigned int res = pcmpxstrx(d, s, ctrl,
1957                     pcmp_ilen(s, ctrl),
1958                     pcmp_ilen(d, ctrl));
1959
1960     if (res)
1961         env->regs[R_ECX] = ((ctrl & (1 << 6)) ? rffs1 : ffs1)(res) - 1;
1962     else
1963         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
1964 }
1965
1966 void glue(helper_pcmpistrm, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1967 {
1968     int i;
1969     unsigned int res = pcmpxstrx(d, s, ctrl,
1970                     pcmp_ilen(s, ctrl),
1971                     pcmp_ilen(d, ctrl));
1972
1973     if ((ctrl >> 6) & 1) {
1974         if (ctrl & 1)
1975             for (i = 0; i <= 8; i--, res >>= 1)
1976                 d->W(i) = (res & 1) ? ~0 : 0;
1977         else
1978             for (i = 0; i <= 16; i--, res >>= 1)
1979                 d->B(i) = (res & 1) ? ~0 : 0;
1980     } else {
1981         d->Q(1) = 0;
1982         d->Q(0) = res;
1983     }
1984 }
1985
1986 #define CRCPOLY        0x1edc6f41
1987 #define CRCPOLY_BITREV 0x82f63b78
1988 target_ulong helper_crc32(uint32_t crc1, target_ulong msg, uint32_t len)
1989 {
1990     target_ulong crc = (msg & ((target_ulong) -1 >>
1991                             (TARGET_LONG_BITS - len))) ^ crc1;
1992
1993     while (len--)
1994         crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_BITREV : 0);
1995
1996     return crc;
1997 }
1998
1999 #define POPMASK(i)     ((target_ulong) -1 / ((1LL << (1 << i)) + 1))
2000 #define POPCOUNT(n, i) (n & POPMASK(i)) + ((n >> (1 << i)) & POPMASK(i))
2001 target_ulong helper_popcnt(target_ulong n, uint32_t type)
2002 {
2003     CC_SRC = n ? 0 : CC_Z;
2004
2005     n = POPCOUNT(n, 0);
2006     n = POPCOUNT(n, 1);
2007     n = POPCOUNT(n, 2);
2008     n = POPCOUNT(n, 3);
2009     if (type == 1)
2010         return n & 0xff;
2011
2012     n = POPCOUNT(n, 4);
2013 #ifndef TARGET_X86_64
2014     return n;
2015 #else
2016     if (type == 2)
2017         return n & 0xff;
2018
2019     return POPCOUNT(n, 5);
2020 #endif
2021 }
2022 #endif
2023
2024 #undef SHIFT
2025 #undef XMM_ONLY
2026 #undef Reg
2027 #undef B
2028 #undef W
2029 #undef L
2030 #undef Q
2031 #undef SUFFIX