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_SQRT(size, a, b) float ## size ## _sqrt(b, &env->sse_status)
 588
 589 /* Note that the choice of comparison op here is important to get the
 590  * special cases right: for min and max Intel specifies that (-0,0),
 591  * (NaN, anything) and (anything, NaN) return the second argument.
 592  */
 593 #define FPU_MIN(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? (a) : (b)
 594 #define FPU_MAX(size, a, b) float ## size ## _lt(b, a, &env->sse_status) ? (a) : (b)
 595
 596 SSE_HELPER_S(add, FPU_ADD)
 597 SSE_HELPER_S(sub, FPU_SUB)
 598 SSE_HELPER_S(mul, FPU_MUL)
 599 SSE_HELPER_S(div, FPU_DIV)
 600 SSE_HELPER_S(min, FPU_MIN)
 601 SSE_HELPER_S(max, FPU_MAX)
 602 SSE_HELPER_S(sqrt, FPU_SQRT)
 603
 604
 605 /* float to float conversions */
 606 void helper_cvtps2pd(Reg *d, Reg *s)
 607 {
 608     float32 s0, s1;
 609     s0 = s->XMM_S(0);
 610     s1 = s->XMM_S(1);
 611     d->XMM_D(0) = float32_to_float64(s0, &env->sse_status);
 612     d->XMM_D(1) = float32_to_float64(s1, &env->sse_status);
 613 }
 614
 615 void helper_cvtpd2ps(Reg *d, Reg *s)
 616 {
 617     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 618     d->XMM_S(1) = float64_to_float32(s->XMM_D(1), &env->sse_status);
 619     d->Q(1) = 0;
 620 }
 621
 622 void helper_cvtss2sd(Reg *d, Reg *s)
 623 {
 624     d->XMM_D(0) = float32_to_float64(s->XMM_S(0), &env->sse_status);
 625 }
 626
 627 void helper_cvtsd2ss(Reg *d, Reg *s)
 628 {
 629     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 630 }
 631
 632 /* integer to float */
 633 void helper_cvtdq2ps(Reg *d, Reg *s)
 634 {
 635     d->XMM_S(0) = int32_to_float32(s->XMM_L(0), &env->sse_status);
 636     d->XMM_S(1) = int32_to_float32(s->XMM_L(1), &env->sse_status);
 637     d->XMM_S(2) = int32_to_float32(s->XMM_L(2), &env->sse_status);
 638     d->XMM_S(3) = int32_to_float32(s->XMM_L(3), &env->sse_status);
 639 }
 640
 641 void helper_cvtdq2pd(Reg *d, Reg *s)
 642 {
 643     int32_t l0, l1;
 644     l0 = (int32_t)s->XMM_L(0);
 645     l1 = (int32_t)s->XMM_L(1);
 646     d->XMM_D(0) = int32_to_float64(l0, &env->sse_status);
 647     d->XMM_D(1) = int32_to_float64(l1, &env->sse_status);
 648 }
 649
 650 void helper_cvtpi2ps(XMMReg *d, MMXReg *s)
 651 {
 652     d->XMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
 653     d->XMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
 654 }
 655
 656 void helper_cvtpi2pd(XMMReg *d, MMXReg *s)
 657 {
 658     d->XMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
 659     d->XMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
 660 }
 661
 662 void helper_cvtsi2ss(XMMReg *d, uint32_t val)
 663 {
 664     d->XMM_S(0) = int32_to_float32(val, &env->sse_status);
 665 }
 666
 667 void helper_cvtsi2sd(XMMReg *d, uint32_t val)
 668 {
 669     d->XMM_D(0) = int32_to_float64(val, &env->sse_status);
 670 }
 671
 672 #ifdef TARGET_X86_64
 673 void helper_cvtsq2ss(XMMReg *d, uint64_t val)
 674 {
 675     d->XMM_S(0) = int64_to_float32(val, &env->sse_status);
 676 }
 677
 678 void helper_cvtsq2sd(XMMReg *d, uint64_t val)
 679 {
 680     d->XMM_D(0) = int64_to_float64(val, &env->sse_status);
 681 }
 682 #endif
 683
 684 /* float to integer */
 685 void helper_cvtps2dq(XMMReg *d, XMMReg *s)
 686 {
 687     d->XMM_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 688     d->XMM_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 689     d->XMM_L(2) = float32_to_int32(s->XMM_S(2), &env->sse_status);
 690     d->XMM_L(3) = float32_to_int32(s->XMM_S(3), &env->sse_status);
 691 }
 692
 693 void helper_cvtpd2dq(XMMReg *d, XMMReg *s)
 694 {
 695     d->XMM_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 696     d->XMM_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 697     d->XMM_Q(1) = 0;
 698 }
 699
 700 void helper_cvtps2pi(MMXReg *d, XMMReg *s)
 701 {
 702     d->MMX_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 703     d->MMX_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 704 }
 705
 706 void helper_cvtpd2pi(MMXReg *d, XMMReg *s)
 707 {
 708     d->MMX_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 709     d->MMX_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 710 }
 711
 712 int32_t helper_cvtss2si(XMMReg *s)
 713 {
 714     return float32_to_int32(s->XMM_S(0), &env->sse_status);
 715 }
 716
 717 int32_t helper_cvtsd2si(XMMReg *s)
 718 {
 719     return float64_to_int32(s->XMM_D(0), &env->sse_status);
 720 }
 721
 722 #ifdef TARGET_X86_64
 723 int64_t helper_cvtss2sq(XMMReg *s)
 724 {
 725     return float32_to_int64(s->XMM_S(0), &env->sse_status);
 726 }
 727
 728 int64_t helper_cvtsd2sq(XMMReg *s)
 729 {
 730     return float64_to_int64(s->XMM_D(0), &env->sse_status);
 731 }
 732 #endif
 733
 734 /* float to integer truncated */
 735 void helper_cvttps2dq(XMMReg *d, XMMReg *s)
 736 {
 737     d->XMM_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 738     d->XMM_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 739     d->XMM_L(2) = float32_to_int32_round_to_zero(s->XMM_S(2), &env->sse_status);
 740     d->XMM_L(3) = float32_to_int32_round_to_zero(s->XMM_S(3), &env->sse_status);
 741 }
 742
 743 void helper_cvttpd2dq(XMMReg *d, XMMReg *s)
 744 {
 745     d->XMM_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 746     d->XMM_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 747     d->XMM_Q(1) = 0;
 748 }
 749
 750 void helper_cvttps2pi(MMXReg *d, XMMReg *s)
 751 {
 752     d->MMX_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 753     d->MMX_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 754 }
 755
 756 void helper_cvttpd2pi(MMXReg *d, XMMReg *s)
 757 {
 758     d->MMX_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 759     d->MMX_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 760 }
 761
 762 int32_t helper_cvttss2si(XMMReg *s)
 763 {
 764     return float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 765 }
 766
 767 int32_t helper_cvttsd2si(XMMReg *s)
 768 {
 769     return float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 770 }
 771
 772 #ifdef TARGET_X86_64
 773 int64_t helper_cvttss2sq(XMMReg *s)
 774 {
 775     return float32_to_int64_round_to_zero(s->XMM_S(0), &env->sse_status);
 776 }
 777
 778 int64_t helper_cvttsd2sq(XMMReg *s)
 779 {
 780     return float64_to_int64_round_to_zero(s->XMM_D(0), &env->sse_status);
 781 }
 782 #endif
 783
 784 void helper_rsqrtps(XMMReg *d, XMMReg *s)
 785 {
 786     d->XMM_S(0) = float32_div(float32_one,
 787                               float32_sqrt(s->XMM_S(0), &env->sse_status),
 788                               &env->sse_status);
 789     d->XMM_S(1) = float32_div(float32_one,
 790                               float32_sqrt(s->XMM_S(1), &env->sse_status),
 791                               &env->sse_status);
 792     d->XMM_S(2) = float32_div(float32_one,
 793                               float32_sqrt(s->XMM_S(2), &env->sse_status),
 794                               &env->sse_status);
 795     d->XMM_S(3) = float32_div(float32_one,
 796                               float32_sqrt(s->XMM_S(3), &env->sse_status),
 797                               &env->sse_status);
 798 }
 799
 800 void helper_rsqrtss(XMMReg *d, XMMReg *s)
 801 {
 802     d->XMM_S(0) = float32_div(float32_one,
 803                               float32_sqrt(s->XMM_S(0), &env->sse_status),
 804                               &env->sse_status);
 805 }
 806
 807 void helper_rcpps(XMMReg *d, XMMReg *s)
 808 {
 809     d->XMM_S(0) = float32_div(float32_one, s->XMM_S(0), &env->sse_status);
 810     d->XMM_S(1) = float32_div(float32_one, s->XMM_S(1), &env->sse_status);
 811     d->XMM_S(2) = float32_div(float32_one, s->XMM_S(2), &env->sse_status);
 812     d->XMM_S(3) = float32_div(float32_one, s->XMM_S(3), &env->sse_status);
 813 }
 814
 815 void helper_rcpss(XMMReg *d, XMMReg *s)
 816 {
 817     d->XMM_S(0) = float32_div(float32_one, s->XMM_S(0), &env->sse_status);
 818 }
 819
 820 static inline uint64_t helper_extrq(uint64_t src, int shift, int len)
 821 {
 822     uint64_t mask;
 823
 824     if (len == 0) {
 825         mask = ~0LL;
 826     } else {
 827         mask = (1ULL << len) - 1;
 828     }
 829     return (src >> shift) & mask;
 830 }
 831
 832 void helper_extrq_r(XMMReg *d, XMMReg *s)
 833 {
 834     d->XMM_Q(0) = helper_extrq(d->XMM_Q(0), s->XMM_B(1), s->XMM_B(0));
 835 }
 836
 837 void helper_extrq_i(XMMReg *d, int index, int length)
 838 {
 839     d->XMM_Q(0) = helper_extrq(d->XMM_Q(0), index, length);
 840 }
 841
 842 static inline uint64_t helper_insertq(uint64_t src, int shift, int len)
 843 {
 844     uint64_t mask;
 845
 846     if (len == 0) {
 847         mask = ~0ULL;
 848     } else {
 849         mask = (1ULL << len) - 1;
 850     }
 851     return (src & ~(mask << shift)) | ((src & mask) << shift);
 852 }
 853
 854 void helper_insertq_r(XMMReg *d, XMMReg *s)
 855 {
 856     d->XMM_Q(0) = helper_insertq(s->XMM_Q(0), s->XMM_B(9), s->XMM_B(8));
 857 }
 858
 859 void helper_insertq_i(XMMReg *d, int index, int length)
 860 {
 861     d->XMM_Q(0) = helper_insertq(d->XMM_Q(0), index, length);
 862 }
 863
 864 void helper_haddps(XMMReg *d, XMMReg *s)
 865 {
 866     XMMReg r;
 867     r.XMM_S(0) = float32_add(d->XMM_S(0), d->XMM_S(1), &env->sse_status);
 868     r.XMM_S(1) = float32_add(d->XMM_S(2), d->XMM_S(3), &env->sse_status);
 869     r.XMM_S(2) = float32_add(s->XMM_S(0), s->XMM_S(1), &env->sse_status);
 870     r.XMM_S(3) = float32_add(s->XMM_S(2), s->XMM_S(3), &env->sse_status);
 871     *d = r;
 872 }
 873
 874 void helper_haddpd(XMMReg *d, XMMReg *s)
 875 {
 876     XMMReg r;
 877     r.XMM_D(0) = float64_add(d->XMM_D(0), d->XMM_D(1), &env->sse_status);
 878     r.XMM_D(1) = float64_add(s->XMM_D(0), s->XMM_D(1), &env->sse_status);
 879     *d = r;
 880 }
 881
 882 void helper_hsubps(XMMReg *d, XMMReg *s)
 883 {
 884     XMMReg r;
 885     r.XMM_S(0) = float32_sub(d->XMM_S(0), d->XMM_S(1), &env->sse_status);
 886     r.XMM_S(1) = float32_sub(d->XMM_S(2), d->XMM_S(3), &env->sse_status);
 887     r.XMM_S(2) = float32_sub(s->XMM_S(0), s->XMM_S(1), &env->sse_status);
 888     r.XMM_S(3) = float32_sub(s->XMM_S(2), s->XMM_S(3), &env->sse_status);
 889     *d = r;
 890 }
 891
 892 void helper_hsubpd(XMMReg *d, XMMReg *s)
 893 {
 894     XMMReg r;
 895     r.XMM_D(0) = float64_sub(d->XMM_D(0), d->XMM_D(1), &env->sse_status);
 896     r.XMM_D(1) = float64_sub(s->XMM_D(0), s->XMM_D(1), &env->sse_status);
 897     *d = r;
 898 }
 899
 900 void helper_addsubps(XMMReg *d, XMMReg *s)
 901 {
 902     d->XMM_S(0) = float32_sub(d->XMM_S(0), s->XMM_S(0), &env->sse_status);
 903     d->XMM_S(1) = float32_add(d->XMM_S(1), s->XMM_S(1), &env->sse_status);
 904     d->XMM_S(2) = float32_sub(d->XMM_S(2), s->XMM_S(2), &env->sse_status);
 905     d->XMM_S(3) = float32_add(d->XMM_S(3), s->XMM_S(3), &env->sse_status);
 906 }
 907
 908 void helper_addsubpd(XMMReg *d, XMMReg *s)
 909 {
 910     d->XMM_D(0) = float64_sub(d->XMM_D(0), s->XMM_D(0), &env->sse_status);
 911     d->XMM_D(1) = float64_add(d->XMM_D(1), s->XMM_D(1), &env->sse_status);
 912 }
 913
 914 /* XXX: unordered */
 915 #define SSE_HELPER_CMP(name, F)\
 916 void helper_ ## name ## ps (Reg *d, Reg *s)\
 917 {\
 918     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 919     d->XMM_L(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
 920     d->XMM_L(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
 921     d->XMM_L(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
 922 }\
 923 \
 924 void helper_ ## name ## ss (Reg *d, Reg *s)\
 925 {\
 926     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 927 }\
 928 void helper_ ## name ## pd (Reg *d, Reg *s)\
 929 {\
 930     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 931     d->XMM_Q(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
 932 }\
 933 \
 934 void helper_ ## name ## sd (Reg *d, Reg *s)\
 935 {\
 936     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 937 }
 938
 939 #define FPU_CMPEQ(size, a, b) float ## size ## _eq_quiet(a, b, &env->sse_status) ? -1 : 0
 940 #define FPU_CMPLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? -1 : 0
 941 #define FPU_CMPLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? -1 : 0
 942 #define FPU_CMPUNORD(size, a, b) float ## size ## _unordered_quiet(a, b, &env->sse_status) ? - 1 : 0
 943 #define FPU_CMPNEQ(size, a, b) float ## size ## _eq_quiet(a, b, &env->sse_status) ? 0 : -1
 944 #define FPU_CMPNLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? 0 : -1
 945 #define FPU_CMPNLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? 0 : -1
 946 #define FPU_CMPORD(size, a, b) float ## size ## _unordered_quiet(a, b, &env->sse_status) ? 0 : -1
 947
 948 SSE_HELPER_CMP(cmpeq, FPU_CMPEQ)
 949 SSE_HELPER_CMP(cmplt, FPU_CMPLT)
 950 SSE_HELPER_CMP(cmple, FPU_CMPLE)
 951 SSE_HELPER_CMP(cmpunord, FPU_CMPUNORD)
 952 SSE_HELPER_CMP(cmpneq, FPU_CMPNEQ)
 953 SSE_HELPER_CMP(cmpnlt, FPU_CMPNLT)
 954 SSE_HELPER_CMP(cmpnle, FPU_CMPNLE)
 955 SSE_HELPER_CMP(cmpord, FPU_CMPORD)
 956
 957 static const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
 958
 959 void helper_ucomiss(Reg *d, Reg *s)
 960 {
 961     int ret;
 962     float32 s0, s1;
 963
 964     s0 = d->XMM_S(0);
 965     s1 = s->XMM_S(0);
 966     ret = float32_compare_quiet(s0, s1, &env->sse_status);
 967     CC_SRC = comis_eflags[ret + 1];
 968 }
 969
 970 void helper_comiss(Reg *d, Reg *s)
 971 {
 972     int ret;
 973     float32 s0, s1;
 974
 975     s0 = d->XMM_S(0);
 976     s1 = s->XMM_S(0);
 977     ret = float32_compare(s0, s1, &env->sse_status);
 978     CC_SRC = comis_eflags[ret + 1];
 979 }
 980
 981 void helper_ucomisd(Reg *d, Reg *s)
 982 {
 983     int ret;
 984     float64 d0, d1;
 985
 986     d0 = d->XMM_D(0);
 987     d1 = s->XMM_D(0);
 988     ret = float64_compare_quiet(d0, d1, &env->sse_status);
 989     CC_SRC = comis_eflags[ret + 1];
 990 }
 991
 992 void helper_comisd(Reg *d, Reg *s)
 993 {
 994     int ret;
 995     float64 d0, d1;
 996
 997     d0 = d->XMM_D(0);
 998     d1 = s->XMM_D(0);
 999     ret = float64_compare(d0, d1, &env->sse_status);
1000     CC_SRC = comis_eflags[ret + 1];
1001 }
1002
1003 uint32_t helper_movmskps(Reg *s)
1004 {
1005     int b0, b1, b2, b3;
1006     b0 = s->XMM_L(0) >> 31;
1007     b1 = s->XMM_L(1) >> 31;
1008     b2 = s->XMM_L(2) >> 31;
1009     b3 = s->XMM_L(3) >> 31;
1010     return b0 | (b1 << 1) | (b2 << 2) | (b3 << 3);
1011 }
1012
1013 uint32_t helper_movmskpd(Reg *s)
1014 {
1015     int b0, b1;
1016     b0 = s->XMM_L(1) >> 31;
1017     b1 = s->XMM_L(3) >> 31;
1018     return b0 | (b1 << 1);
1019 }
1020
1021 #endif
1022
1023 uint32_t glue(helper_pmovmskb, SUFFIX)(Reg *s)
1024 {
1025     uint32_t val;
1026     val = 0;
1027     val |= (s->B(0) >> 7);
1028     val |= (s->B(1) >> 6) & 0x02;
1029     val |= (s->B(2) >> 5) & 0x04;
1030     val |= (s->B(3) >> 4) & 0x08;
1031     val |= (s->B(4) >> 3) & 0x10;
1032     val |= (s->B(5) >> 2) & 0x20;
1033     val |= (s->B(6) >> 1) & 0x40;
1034     val |= (s->B(7)) & 0x80;
1035 #if SHIFT == 1
1036     val |= (s->B(8) << 1) & 0x0100;
1037     val |= (s->B(9) << 2) & 0x0200;
1038     val |= (s->B(10) << 3) & 0x0400;
1039     val |= (s->B(11) << 4) & 0x0800;
1040     val |= (s->B(12) << 5) & 0x1000;
1041     val |= (s->B(13) << 6) & 0x2000;
1042     val |= (s->B(14) << 7) & 0x4000;
1043     val |= (s->B(15) << 8) & 0x8000;
1044 #endif
1045     return val;
1046 }
1047
1048 void glue(helper_packsswb, SUFFIX) (Reg *d, Reg *s)
1049 {
1050     Reg r;
1051
1052     r.B(0) = satsb((int16_t)d->W(0));
1053     r.B(1) = satsb((int16_t)d->W(1));
1054     r.B(2) = satsb((int16_t)d->W(2));
1055     r.B(3) = satsb((int16_t)d->W(3));
1056 #if SHIFT == 1
1057     r.B(4) = satsb((int16_t)d->W(4));
1058     r.B(5) = satsb((int16_t)d->W(5));
1059     r.B(6) = satsb((int16_t)d->W(6));
1060     r.B(7) = satsb((int16_t)d->W(7));
1061 #endif
1062     r.B((4 << SHIFT) + 0) = satsb((int16_t)s->W(0));
1063     r.B((4 << SHIFT) + 1) = satsb((int16_t)s->W(1));
1064     r.B((4 << SHIFT) + 2) = satsb((int16_t)s->W(2));
1065     r.B((4 << SHIFT) + 3) = satsb((int16_t)s->W(3));
1066 #if SHIFT == 1
1067     r.B(12) = satsb((int16_t)s->W(4));
1068     r.B(13) = satsb((int16_t)s->W(5));
1069     r.B(14) = satsb((int16_t)s->W(6));
1070     r.B(15) = satsb((int16_t)s->W(7));
1071 #endif
1072     *d = r;
1073 }
1074
1075 void glue(helper_packuswb, SUFFIX) (Reg *d, Reg *s)
1076 {
1077     Reg r;
1078
1079     r.B(0) = satub((int16_t)d->W(0));
1080     r.B(1) = satub((int16_t)d->W(1));
1081     r.B(2) = satub((int16_t)d->W(2));
1082     r.B(3) = satub((int16_t)d->W(3));
1083 #if SHIFT == 1
1084     r.B(4) = satub((int16_t)d->W(4));
1085     r.B(5) = satub((int16_t)d->W(5));
1086     r.B(6) = satub((int16_t)d->W(6));
1087     r.B(7) = satub((int16_t)d->W(7));
1088 #endif
1089     r.B((4 << SHIFT) + 0) = satub((int16_t)s->W(0));
1090     r.B((4 << SHIFT) + 1) = satub((int16_t)s->W(1));
1091     r.B((4 << SHIFT) + 2) = satub((int16_t)s->W(2));
1092     r.B((4 << SHIFT) + 3) = satub((int16_t)s->W(3));
1093 #if SHIFT == 1
1094     r.B(12) = satub((int16_t)s->W(4));
1095     r.B(13) = satub((int16_t)s->W(5));
1096     r.B(14) = satub((int16_t)s->W(6));
1097     r.B(15) = satub((int16_t)s->W(7));
1098 #endif
1099     *d = r;
1100 }
1101
1102 void glue(helper_packssdw, SUFFIX) (Reg *d, Reg *s)
1103 {
1104     Reg r;
1105
1106     r.W(0) = satsw(d->L(0));
1107     r.W(1) = satsw(d->L(1));
1108 #if SHIFT == 1
1109     r.W(2) = satsw(d->L(2));
1110     r.W(3) = satsw(d->L(3));
1111 #endif
1112     r.W((2 << SHIFT) + 0) = satsw(s->L(0));
1113     r.W((2 << SHIFT) + 1) = satsw(s->L(1));
1114 #if SHIFT == 1
1115     r.W(6) = satsw(s->L(2));
1116     r.W(7) = satsw(s->L(3));
1117 #endif
1118     *d = r;
1119 }
1120
1121 #define UNPCK_OP(base_name, base)                               \
1122                                                                 \
1123 void glue(helper_punpck ## base_name ## bw, SUFFIX) (Reg *d, Reg *s)   \
1124 {                                                               \
1125     Reg r;                                              \
1126                                                                 \
1127     r.B(0) = d->B((base << (SHIFT + 2)) + 0);                   \
1128     r.B(1) = s->B((base << (SHIFT + 2)) + 0);                   \
1129     r.B(2) = d->B((base << (SHIFT + 2)) + 1);                   \
1130     r.B(3) = s->B((base << (SHIFT + 2)) + 1);                   \
1131     r.B(4) = d->B((base << (SHIFT + 2)) + 2);                   \
1132     r.B(5) = s->B((base << (SHIFT + 2)) + 2);                   \
1133     r.B(6) = d->B((base << (SHIFT + 2)) + 3);                   \
1134     r.B(7) = s->B((base << (SHIFT + 2)) + 3);                   \
1135 XMM_ONLY(                                                       \
1136     r.B(8) = d->B((base << (SHIFT + 2)) + 4);                   \
1137     r.B(9) = s->B((base << (SHIFT + 2)) + 4);                   \
1138     r.B(10) = d->B((base << (SHIFT + 2)) + 5);                  \
1139     r.B(11) = s->B((base << (SHIFT + 2)) + 5);                  \
1140     r.B(12) = d->B((base << (SHIFT + 2)) + 6);                  \
1141     r.B(13) = s->B((base << (SHIFT + 2)) + 6);                  \
1142     r.B(14) = d->B((base << (SHIFT + 2)) + 7);                  \
1143     r.B(15) = s->B((base << (SHIFT + 2)) + 7);                  \
1144 )                                                               \
1145     *d = r;                                                     \
1146 }                                                               \
1147                                                                 \
1148 void glue(helper_punpck ## base_name ## wd, SUFFIX) (Reg *d, Reg *s)   \
1149 {                                                               \
1150     Reg r;                                              \
1151                                                                 \
1152     r.W(0) = d->W((base << (SHIFT + 1)) + 0);                   \
1153     r.W(1) = s->W((base << (SHIFT + 1)) + 0);                   \
1154     r.W(2) = d->W((base << (SHIFT + 1)) + 1);                   \
1155     r.W(3) = s->W((base << (SHIFT + 1)) + 1);                   \
1156 XMM_ONLY(                                                       \
1157     r.W(4) = d->W((base << (SHIFT + 1)) + 2);                   \
1158     r.W(5) = s->W((base << (SHIFT + 1)) + 2);                   \
1159     r.W(6) = d->W((base << (SHIFT + 1)) + 3);                   \
1160     r.W(7) = s->W((base << (SHIFT + 1)) + 3);                   \
1161 )                                                               \
1162     *d = r;                                                     \
1163 }                                                               \
1164                                                                 \
1165 void glue(helper_punpck ## base_name ## dq, SUFFIX) (Reg *d, Reg *s)   \
1166 {                                                               \
1167     Reg r;                                              \
1168                                                                 \
1169     r.L(0) = d->L((base << SHIFT) + 0);                         \
1170     r.L(1) = s->L((base << SHIFT) + 0);                         \
1171 XMM_ONLY(                                                       \
1172     r.L(2) = d->L((base << SHIFT) + 1);                         \
1173     r.L(3) = s->L((base << SHIFT) + 1);                         \
1174 )                                                               \
1175     *d = r;                                                     \
1176 }                                                               \
1177                                                                 \
1178 XMM_ONLY(                                                       \
1179 void glue(helper_punpck ## base_name ## qdq, SUFFIX) (Reg *d, Reg *s)  \
1180 {                                                               \
1181     Reg r;                                              \
1182                                                                 \
1183     r.Q(0) = d->Q(base);                                        \
1184     r.Q(1) = s->Q(base);                                        \
1185     *d = r;                                                     \
1186 }                                                               \
1187 )
1188
1189 UNPCK_OP(l, 0)
1190 UNPCK_OP(h, 1)
1191
1192 /* 3DNow! float ops */
1193 #if SHIFT == 0
1194 void helper_pi2fd(MMXReg *d, MMXReg *s)
1195 {
1196     d->MMX_S(0) = int32_to_float32(s->MMX_L(0), &env->mmx_status);
1197     d->MMX_S(1) = int32_to_float32(s->MMX_L(1), &env->mmx_status);
1198 }
1199
1200 void helper_pi2fw(MMXReg *d, MMXReg *s)
1201 {
1202     d->MMX_S(0) = int32_to_float32((int16_t)s->MMX_W(0), &env->mmx_status);
1203     d->MMX_S(1) = int32_to_float32((int16_t)s->MMX_W(2), &env->mmx_status);
1204 }
1205
1206 void helper_pf2id(MMXReg *d, MMXReg *s)
1207 {
1208     d->MMX_L(0) = float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status);
1209     d->MMX_L(1) = float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status);
1210 }
1211
1212 void helper_pf2iw(MMXReg *d, MMXReg *s)
1213 {
1214     d->MMX_L(0) = satsw(float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status));
1215     d->MMX_L(1) = satsw(float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status));
1216 }
1217
1218 void helper_pfacc(MMXReg *d, MMXReg *s)
1219 {
1220     MMXReg r;
1221     r.MMX_S(0) = float32_add(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1222     r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1223     *d = r;
1224 }
1225
1226 void helper_pfadd(MMXReg *d, MMXReg *s)
1227 {
1228     d->MMX_S(0) = float32_add(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1229     d->MMX_S(1) = float32_add(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1230 }
1231
1232 void helper_pfcmpeq(MMXReg *d, MMXReg *s)
1233 {
1234     d->MMX_L(0) = float32_eq_quiet(d->MMX_S(0), s->MMX_S(0), &env->mmx_status) ? -1 : 0;
1235     d->MMX_L(1) = float32_eq_quiet(d->MMX_S(1), s->MMX_S(1), &env->mmx_status) ? -1 : 0;
1236 }
1237
1238 void helper_pfcmpge(MMXReg *d, MMXReg *s)
1239 {
1240     d->MMX_L(0) = float32_le(s->MMX_S(0), d->MMX_S(0), &env->mmx_status) ? -1 : 0;
1241     d->MMX_L(1) = float32_le(s->MMX_S(1), d->MMX_S(1), &env->mmx_status) ? -1 : 0;
1242 }
1243
1244 void helper_pfcmpgt(MMXReg *d, MMXReg *s)
1245 {
1246     d->MMX_L(0) = float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status) ? -1 : 0;
1247     d->MMX_L(1) = float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status) ? -1 : 0;
1248 }
1249
1250 void helper_pfmax(MMXReg *d, MMXReg *s)
1251 {
1252     if (float32_lt(d->MMX_S(0), s->MMX_S(0), &env->mmx_status))
1253         d->MMX_S(0) = s->MMX_S(0);
1254     if (float32_lt(d->MMX_S(1), s->MMX_S(1), &env->mmx_status))
1255         d->MMX_S(1) = s->MMX_S(1);
1256 }
1257
1258 void helper_pfmin(MMXReg *d, MMXReg *s)
1259 {
1260     if (float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status))
1261         d->MMX_S(0) = s->MMX_S(0);
1262     if (float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status))
1263         d->MMX_S(1) = s->MMX_S(1);
1264 }
1265
1266 void helper_pfmul(MMXReg *d, MMXReg *s)
1267 {
1268     d->MMX_S(0) = float32_mul(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1269     d->MMX_S(1) = float32_mul(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1270 }
1271
1272 void helper_pfnacc(MMXReg *d, MMXReg *s)
1273 {
1274     MMXReg r;
1275     r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1276     r.MMX_S(1) = float32_sub(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1277     *d = r;
1278 }
1279
1280 void helper_pfpnacc(MMXReg *d, MMXReg *s)
1281 {
1282     MMXReg r;
1283     r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1284     r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1285     *d = r;
1286 }
1287
1288 void helper_pfrcp(MMXReg *d, MMXReg *s)
1289 {
1290     d->MMX_S(0) = float32_div(float32_one, s->MMX_S(0), &env->mmx_status);
1291     d->MMX_S(1) = d->MMX_S(0);
1292 }
1293
1294 void helper_pfrsqrt(MMXReg *d, MMXReg *s)
1295 {
1296     d->MMX_L(1) = s->MMX_L(0) & 0x7fffffff;
1297     d->MMX_S(1) = float32_div(float32_one,
1298                               float32_sqrt(d->MMX_S(1), &env->mmx_status),
1299                               &env->mmx_status);
1300     d->MMX_L(1) |= s->MMX_L(0) & 0x80000000;
1301     d->MMX_L(0) = d->MMX_L(1);
1302 }
1303
1304 void helper_pfsub(MMXReg *d, MMXReg *s)
1305 {
1306     d->MMX_S(0) = float32_sub(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1307     d->MMX_S(1) = float32_sub(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1308 }
1309
1310 void helper_pfsubr(MMXReg *d, MMXReg *s)
1311 {
1312     d->MMX_S(0) = float32_sub(s->MMX_S(0), d->MMX_S(0), &env->mmx_status);
1313     d->MMX_S(1) = float32_sub(s->MMX_S(1), d->MMX_S(1), &env->mmx_status);
1314 }
1315
1316 void helper_pswapd(MMXReg *d, MMXReg *s)
1317 {
1318     MMXReg r;
1319     r.MMX_L(0) = s->MMX_L(1);
1320     r.MMX_L(1) = s->MMX_L(0);
1321     *d = r;
1322 }
1323 #endif
1324
1325 /* SSSE3 op helpers */
1326 void glue(helper_pshufb, SUFFIX) (Reg *d, Reg *s)
1327 {
1328     int i;
1329     Reg r;
1330
1331     for (i = 0; i < (8 << SHIFT); i++)
1332         r.B(i) = (s->B(i) & 0x80) ? 0 : (d->B(s->B(i) & ((8 << SHIFT) - 1)));
1333
1334     *d = r;
1335 }
1336
1337 void glue(helper_phaddw, SUFFIX) (Reg *d, Reg *s)
1338 {
1339     d->W(0) = (int16_t)d->W(0) + (int16_t)d->W(1);
1340     d->W(1) = (int16_t)d->W(2) + (int16_t)d->W(3);
1341     XMM_ONLY(d->W(2) = (int16_t)d->W(4) + (int16_t)d->W(5));
1342     XMM_ONLY(d->W(3) = (int16_t)d->W(6) + (int16_t)d->W(7));
1343     d->W((2 << SHIFT) + 0) = (int16_t)s->W(0) + (int16_t)s->W(1);
1344     d->W((2 << SHIFT) + 1) = (int16_t)s->W(2) + (int16_t)s->W(3);
1345     XMM_ONLY(d->W(6) = (int16_t)s->W(4) + (int16_t)s->W(5));
1346     XMM_ONLY(d->W(7) = (int16_t)s->W(6) + (int16_t)s->W(7));
1347 }
1348
1349 void glue(helper_phaddd, SUFFIX) (Reg *d, Reg *s)
1350 {
1351     d->L(0) = (int32_t)d->L(0) + (int32_t)d->L(1);
1352     XMM_ONLY(d->L(1) = (int32_t)d->L(2) + (int32_t)d->L(3));
1353     d->L((1 << SHIFT) + 0) = (int32_t)s->L(0) + (int32_t)s->L(1);
1354     XMM_ONLY(d->L(3) = (int32_t)s->L(2) + (int32_t)s->L(3));
1355 }
1356
1357 void glue(helper_phaddsw, SUFFIX) (Reg *d, Reg *s)
1358 {
1359     d->W(0) = satsw((int16_t)d->W(0) + (int16_t)d->W(1));
1360     d->W(1) = satsw((int16_t)d->W(2) + (int16_t)d->W(3));
1361     XMM_ONLY(d->W(2) = satsw((int16_t)d->W(4) + (int16_t)d->W(5)));
1362     XMM_ONLY(d->W(3) = satsw((int16_t)d->W(6) + (int16_t)d->W(7)));
1363     d->W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) + (int16_t)s->W(1));
1364     d->W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) + (int16_t)s->W(3));
1365     XMM_ONLY(d->W(6) = satsw((int16_t)s->W(4) + (int16_t)s->W(5)));
1366     XMM_ONLY(d->W(7) = satsw((int16_t)s->W(6) + (int16_t)s->W(7)));
1367 }
1368
1369 void glue(helper_pmaddubsw, SUFFIX) (Reg *d, Reg *s)
1370 {
1371     d->W(0) = satsw((int8_t)s->B( 0) * (uint8_t)d->B( 0) +
1372                     (int8_t)s->B( 1) * (uint8_t)d->B( 1));
1373     d->W(1) = satsw((int8_t)s->B( 2) * (uint8_t)d->B( 2) +
1374                     (int8_t)s->B( 3) * (uint8_t)d->B( 3));
1375     d->W(2) = satsw((int8_t)s->B( 4) * (uint8_t)d->B( 4) +
1376                     (int8_t)s->B( 5) * (uint8_t)d->B( 5));
1377     d->W(3) = satsw((int8_t)s->B( 6) * (uint8_t)d->B( 6) +
1378                     (int8_t)s->B( 7) * (uint8_t)d->B( 7));
1379 #if SHIFT == 1
1380     d->W(4) = satsw((int8_t)s->B( 8) * (uint8_t)d->B( 8) +
1381                     (int8_t)s->B( 9) * (uint8_t)d->B( 9));
1382     d->W(5) = satsw((int8_t)s->B(10) * (uint8_t)d->B(10) +
1383                     (int8_t)s->B(11) * (uint8_t)d->B(11));
1384     d->W(6) = satsw((int8_t)s->B(12) * (uint8_t)d->B(12) +
1385                     (int8_t)s->B(13) * (uint8_t)d->B(13));
1386     d->W(7) = satsw((int8_t)s->B(14) * (uint8_t)d->B(14) +
1387                     (int8_t)s->B(15) * (uint8_t)d->B(15));
1388 #endif
1389 }
1390
1391 void glue(helper_phsubw, SUFFIX) (Reg *d, Reg *s)
1392 {
1393     d->W(0) = (int16_t)d->W(0) - (int16_t)d->W(1);
1394     d->W(1) = (int16_t)d->W(2) - (int16_t)d->W(3);
1395     XMM_ONLY(d->W(2) = (int16_t)d->W(4) - (int16_t)d->W(5));
1396     XMM_ONLY(d->W(3) = (int16_t)d->W(6) - (int16_t)d->W(7));
1397     d->W((2 << SHIFT) + 0) = (int16_t)s->W(0) - (int16_t)s->W(1);
1398     d->W((2 << SHIFT) + 1) = (int16_t)s->W(2) - (int16_t)s->W(3);
1399     XMM_ONLY(d->W(6) = (int16_t)s->W(4) - (int16_t)s->W(5));
1400     XMM_ONLY(d->W(7) = (int16_t)s->W(6) - (int16_t)s->W(7));
1401 }
1402
1403 void glue(helper_phsubd, SUFFIX) (Reg *d, Reg *s)
1404 {
1405     d->L(0) = (int32_t)d->L(0) - (int32_t)d->L(1);
1406     XMM_ONLY(d->L(1) = (int32_t)d->L(2) - (int32_t)d->L(3));
1407     d->L((1 << SHIFT) + 0) = (int32_t)s->L(0) - (int32_t)s->L(1);
1408     XMM_ONLY(d->L(3) = (int32_t)s->L(2) - (int32_t)s->L(3));
1409 }
1410
1411 void glue(helper_phsubsw, SUFFIX) (Reg *d, Reg *s)
1412 {
1413     d->W(0) = satsw((int16_t)d->W(0) - (int16_t)d->W(1));
1414     d->W(1) = satsw((int16_t)d->W(2) - (int16_t)d->W(3));
1415     XMM_ONLY(d->W(2) = satsw((int16_t)d->W(4) - (int16_t)d->W(5)));
1416     XMM_ONLY(d->W(3) = satsw((int16_t)d->W(6) - (int16_t)d->W(7)));
1417     d->W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) - (int16_t)s->W(1));
1418     d->W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) - (int16_t)s->W(3));
1419     XMM_ONLY(d->W(6) = satsw((int16_t)s->W(4) - (int16_t)s->W(5)));
1420     XMM_ONLY(d->W(7) = satsw((int16_t)s->W(6) - (int16_t)s->W(7)));
1421 }
1422
1423 #define FABSB(_, x) x > INT8_MAX  ? -(int8_t ) x : x
1424 #define FABSW(_, x) x > INT16_MAX ? -(int16_t) x : x
1425 #define FABSL(_, x) x > INT32_MAX ? -(int32_t) x : x
1426 SSE_HELPER_B(helper_pabsb, FABSB)
1427 SSE_HELPER_W(helper_pabsw, FABSW)
1428 SSE_HELPER_L(helper_pabsd, FABSL)
1429
1430 #define FMULHRSW(d, s) ((int16_t) d * (int16_t) s + 0x4000) >> 15
1431 SSE_HELPER_W(helper_pmulhrsw, FMULHRSW)
1432
1433 #define FSIGNB(d, s) s <= INT8_MAX  ? s ? d : 0 : -(int8_t ) d
1434 #define FSIGNW(d, s) s <= INT16_MAX ? s ? d : 0 : -(int16_t) d
1435 #define FSIGNL(d, s) s <= INT32_MAX ? s ? d : 0 : -(int32_t) d
1436 SSE_HELPER_B(helper_psignb, FSIGNB)
1437 SSE_HELPER_W(helper_psignw, FSIGNW)
1438 SSE_HELPER_L(helper_psignd, FSIGNL)
1439
1440 void glue(helper_palignr, SUFFIX) (Reg *d, Reg *s, int32_t shift)
1441 {
1442     Reg r;
1443
1444     /* XXX could be checked during translation */
1445     if (shift >= (16 << SHIFT)) {
1446         r.Q(0) = 0;
1447         XMM_ONLY(r.Q(1) = 0);
1448     } else {
1449         shift <<= 3;
1450 #define SHR(v, i) (i < 64 && i > -64 ? i > 0 ? v >> (i) : (v << -(i)) : 0)
1451 #if SHIFT == 0
1452         r.Q(0) = SHR(s->Q(0), shift -   0) |
1453                  SHR(d->Q(0), shift -  64);
1454 #else
1455         r.Q(0) = SHR(s->Q(0), shift -   0) |
1456                  SHR(s->Q(1), shift -  64) |
1457                  SHR(d->Q(0), shift - 128) |
1458                  SHR(d->Q(1), shift - 192);
1459         r.Q(1) = SHR(s->Q(0), shift +  64) |
1460                  SHR(s->Q(1), shift -   0) |
1461                  SHR(d->Q(0), shift -  64) |
1462                  SHR(d->Q(1), shift - 128);
1463 #endif
1464 #undef SHR
1465     }
1466
1467     *d = r;
1468 }
1469
1470 #define XMM0 env->xmm_regs[0]
1471
1472 #if SHIFT == 1
1473 #define SSE_HELPER_V(name, elem, num, F)\
1474 void glue(name, SUFFIX) (Reg *d, Reg *s)\
1475 {\
1476     d->elem(0) = F(d->elem(0), s->elem(0), XMM0.elem(0));\
1477     d->elem(1) = F(d->elem(1), s->elem(1), XMM0.elem(1));\
1478     if (num > 2) {\
1479         d->elem(2) = F(d->elem(2), s->elem(2), XMM0.elem(2));\
1480         d->elem(3) = F(d->elem(3), s->elem(3), XMM0.elem(3));\
1481         if (num > 4) {\
1482             d->elem(4) = F(d->elem(4), s->elem(4), XMM0.elem(4));\
1483             d->elem(5) = F(d->elem(5), s->elem(5), XMM0.elem(5));\
1484             d->elem(6) = F(d->elem(6), s->elem(6), XMM0.elem(6));\
1485             d->elem(7) = F(d->elem(7), s->elem(7), XMM0.elem(7));\
1486             if (num > 8) {\
1487                 d->elem(8) = F(d->elem(8), s->elem(8), XMM0.elem(8));\
1488                 d->elem(9) = F(d->elem(9), s->elem(9), XMM0.elem(9));\
1489                 d->elem(10) = F(d->elem(10), s->elem(10), XMM0.elem(10));\
1490                 d->elem(11) = F(d->elem(11), s->elem(11), XMM0.elem(11));\
1491                 d->elem(12) = F(d->elem(12), s->elem(12), XMM0.elem(12));\
1492                 d->elem(13) = F(d->elem(13), s->elem(13), XMM0.elem(13));\
1493                 d->elem(14) = F(d->elem(14), s->elem(14), XMM0.elem(14));\
1494                 d->elem(15) = F(d->elem(15), s->elem(15), XMM0.elem(15));\
1495             }\
1496         }\
1497     }\
1498 }
1499
1500 #define SSE_HELPER_I(name, elem, num, F)\
1501 void glue(name, SUFFIX) (Reg *d, Reg *s, uint32_t imm)\
1502 {\
1503     d->elem(0) = F(d->elem(0), s->elem(0), ((imm >> 0) & 1));\
1504     d->elem(1) = F(d->elem(1), s->elem(1), ((imm >> 1) & 1));\
1505     if (num > 2) {\
1506         d->elem(2) = F(d->elem(2), s->elem(2), ((imm >> 2) & 1));\
1507         d->elem(3) = F(d->elem(3), s->elem(3), ((imm >> 3) & 1));\
1508         if (num > 4) {\
1509             d->elem(4) = F(d->elem(4), s->elem(4), ((imm >> 4) & 1));\
1510             d->elem(5) = F(d->elem(5), s->elem(5), ((imm >> 5) & 1));\
1511             d->elem(6) = F(d->elem(6), s->elem(6), ((imm >> 6) & 1));\
1512             d->elem(7) = F(d->elem(7), s->elem(7), ((imm >> 7) & 1));\
1513             if (num > 8) {\
1514                 d->elem(8) = F(d->elem(8), s->elem(8), ((imm >> 8) & 1));\
1515                 d->elem(9) = F(d->elem(9), s->elem(9), ((imm >> 9) & 1));\
1516                 d->elem(10) = F(d->elem(10), s->elem(10), ((imm >> 10) & 1));\
1517                 d->elem(11) = F(d->elem(11), s->elem(11), ((imm >> 11) & 1));\
1518                 d->elem(12) = F(d->elem(12), s->elem(12), ((imm >> 12) & 1));\
1519                 d->elem(13) = F(d->elem(13), s->elem(13), ((imm >> 13) & 1));\
1520                 d->elem(14) = F(d->elem(14), s->elem(14), ((imm >> 14) & 1));\
1521                 d->elem(15) = F(d->elem(15), s->elem(15), ((imm >> 15) & 1));\
1522             }\
1523         }\
1524     }\
1525 }
1526
1527 /* SSE4.1 op helpers */
1528 #define FBLENDVB(d, s, m) (m & 0x80) ? s : d
1529 #define FBLENDVPS(d, s, m) (m & 0x80000000) ? s : d
1530 #define FBLENDVPD(d, s, m) (m & 0x8000000000000000LL) ? s : d
1531 SSE_HELPER_V(helper_pblendvb, B, 16, FBLENDVB)
1532 SSE_HELPER_V(helper_blendvps, L, 4, FBLENDVPS)
1533 SSE_HELPER_V(helper_blendvpd, Q, 2, FBLENDVPD)
1534
1535 void glue(helper_ptest, SUFFIX) (Reg *d, Reg *s)
1536 {
1537     uint64_t zf = (s->Q(0) &  d->Q(0)) | (s->Q(1) &  d->Q(1));
1538     uint64_t cf = (s->Q(0) & ~d->Q(0)) | (s->Q(1) & ~d->Q(1));
1539
1540     CC_SRC = (zf ? 0 : CC_Z) | (cf ? 0 : CC_C);
1541 }
1542
1543 #define SSE_HELPER_F(name, elem, num, F)\
1544 void glue(name, SUFFIX) (Reg *d, Reg *s)\
1545 {\
1546     d->elem(0) = F(0);\
1547     d->elem(1) = F(1);\
1548     if (num > 2) {\
1549         d->elem(2) = F(2);\
1550         d->elem(3) = F(3);\
1551         if (num > 4) {\
1552             d->elem(4) = F(4);\
1553             d->elem(5) = F(5);\
1554             d->elem(6) = F(6);\
1555             d->elem(7) = F(7);\
1556         }\
1557     }\
1558 }
1559
1560 SSE_HELPER_F(helper_pmovsxbw, W, 8, (int8_t) s->B)
1561 SSE_HELPER_F(helper_pmovsxbd, L, 4, (int8_t) s->B)
1562 SSE_HELPER_F(helper_pmovsxbq, Q, 2, (int8_t) s->B)
1563 SSE_HELPER_F(helper_pmovsxwd, L, 4, (int16_t) s->W)
1564 SSE_HELPER_F(helper_pmovsxwq, Q, 2, (int16_t) s->W)
1565 SSE_HELPER_F(helper_pmovsxdq, Q, 2, (int32_t) s->L)
1566 SSE_HELPER_F(helper_pmovzxbw, W, 8, s->B)
1567 SSE_HELPER_F(helper_pmovzxbd, L, 4, s->B)
1568 SSE_HELPER_F(helper_pmovzxbq, Q, 2, s->B)
1569 SSE_HELPER_F(helper_pmovzxwd, L, 4, s->W)
1570 SSE_HELPER_F(helper_pmovzxwq, Q, 2, s->W)
1571 SSE_HELPER_F(helper_pmovzxdq, Q, 2, s->L)
1572
1573 void glue(helper_pmuldq, SUFFIX) (Reg *d, Reg *s)
1574 {
1575     d->Q(0) = (int64_t) (int32_t) d->L(0) * (int32_t) s->L(0);
1576     d->Q(1) = (int64_t) (int32_t) d->L(2) * (int32_t) s->L(2);
1577 }
1578
1579 #define FCMPEQQ(d, s) d == s ? -1 : 0
1580 SSE_HELPER_Q(helper_pcmpeqq, FCMPEQQ)
1581
1582 void glue(helper_packusdw, SUFFIX) (Reg *d, Reg *s)
1583 {
1584     d->W(0) = satuw((int32_t) d->L(0));
1585     d->W(1) = satuw((int32_t) d->L(1));
1586     d->W(2) = satuw((int32_t) d->L(2));
1587     d->W(3) = satuw((int32_t) d->L(3));
1588     d->W(4) = satuw((int32_t) s->L(0));
1589     d->W(5) = satuw((int32_t) s->L(1));
1590     d->W(6) = satuw((int32_t) s->L(2));
1591     d->W(7) = satuw((int32_t) s->L(3));
1592 }
1593
1594 #define FMINSB(d, s) MIN((int8_t) d, (int8_t) s)
1595 #define FMINSD(d, s) MIN((int32_t) d, (int32_t) s)
1596 #define FMAXSB(d, s) MAX((int8_t) d, (int8_t) s)
1597 #define FMAXSD(d, s) MAX((int32_t) d, (int32_t) s)
1598 SSE_HELPER_B(helper_pminsb, FMINSB)
1599 SSE_HELPER_L(helper_pminsd, FMINSD)
1600 SSE_HELPER_W(helper_pminuw, MIN)
1601 SSE_HELPER_L(helper_pminud, MIN)
1602 SSE_HELPER_B(helper_pmaxsb, FMAXSB)
1603 SSE_HELPER_L(helper_pmaxsd, FMAXSD)
1604 SSE_HELPER_W(helper_pmaxuw, MAX)
1605 SSE_HELPER_L(helper_pmaxud, MAX)
1606
1607 #define FMULLD(d, s) (int32_t) d * (int32_t) s
1608 SSE_HELPER_L(helper_pmulld, FMULLD)
1609
1610 void glue(helper_phminposuw, SUFFIX) (Reg *d, Reg *s)
1611 {
1612     int idx = 0;
1613
1614     if (s->W(1) < s->W(idx))
1615         idx = 1;
1616     if (s->W(2) < s->W(idx))
1617         idx = 2;
1618     if (s->W(3) < s->W(idx))
1619         idx = 3;
1620     if (s->W(4) < s->W(idx))
1621         idx = 4;
1622     if (s->W(5) < s->W(idx))
1623         idx = 5;
1624     if (s->W(6) < s->W(idx))
1625         idx = 6;
1626     if (s->W(7) < s->W(idx))
1627         idx = 7;
1628
1629     d->Q(1) = 0;
1630     d->L(1) = 0;
1631     d->W(1) = idx;
1632     d->W(0) = s->W(idx);
1633 }
1634
1635 void glue(helper_roundps, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1636 {
1637     signed char prev_rounding_mode;
1638
1639     prev_rounding_mode = env->sse_status.float_rounding_mode;
1640     if (!(mode & (1 << 2)))
1641         switch (mode & 3) {
1642         case 0:
1643             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1644             break;
1645         case 1:
1646             set_float_rounding_mode(float_round_down, &env->sse_status);
1647             break;
1648         case 2:
1649             set_float_rounding_mode(float_round_up, &env->sse_status);
1650             break;
1651         case 3:
1652             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1653             break;
1654         }
1655
1656     d->XMM_S(0) = float32_round_to_int(s->XMM_S(0), &env->sse_status);
1657     d->XMM_S(1) = float32_round_to_int(s->XMM_S(1), &env->sse_status);
1658     d->XMM_S(2) = float32_round_to_int(s->XMM_S(2), &env->sse_status);
1659     d->XMM_S(3) = float32_round_to_int(s->XMM_S(3), &env->sse_status);
1660
1661 #if 0 /* TODO */
1662     if (mode & (1 << 3))
1663         set_float_exception_flags(
1664                         get_float_exception_flags(&env->sse_status) &
1665                         ~float_flag_inexact,
1666                         &env->sse_status);
1667 #endif
1668     env->sse_status.float_rounding_mode = prev_rounding_mode;
1669 }
1670
1671 void glue(helper_roundpd, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1672 {
1673     signed char prev_rounding_mode;
1674
1675     prev_rounding_mode = env->sse_status.float_rounding_mode;
1676     if (!(mode & (1 << 2)))
1677         switch (mode & 3) {
1678         case 0:
1679             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1680             break;
1681         case 1:
1682             set_float_rounding_mode(float_round_down, &env->sse_status);
1683             break;
1684         case 2:
1685             set_float_rounding_mode(float_round_up, &env->sse_status);
1686             break;
1687         case 3:
1688             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1689             break;
1690         }
1691
1692     d->XMM_D(0) = float64_round_to_int(s->XMM_D(0), &env->sse_status);
1693     d->XMM_D(1) = float64_round_to_int(s->XMM_D(1), &env->sse_status);
1694
1695 #if 0 /* TODO */
1696     if (mode & (1 << 3))
1697         set_float_exception_flags(
1698                         get_float_exception_flags(&env->sse_status) &
1699                         ~float_flag_inexact,
1700                         &env->sse_status);
1701 #endif
1702     env->sse_status.float_rounding_mode = prev_rounding_mode;
1703 }
1704
1705 void glue(helper_roundss, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1706 {
1707     signed char prev_rounding_mode;
1708
1709     prev_rounding_mode = env->sse_status.float_rounding_mode;
1710     if (!(mode & (1 << 2)))
1711         switch (mode & 3) {
1712         case 0:
1713             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1714             break;
1715         case 1:
1716             set_float_rounding_mode(float_round_down, &env->sse_status);
1717             break;
1718         case 2:
1719             set_float_rounding_mode(float_round_up, &env->sse_status);
1720             break;
1721         case 3:
1722             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1723             break;
1724         }
1725
1726     d->XMM_S(0) = float32_round_to_int(s->XMM_S(0), &env->sse_status);
1727
1728 #if 0 /* TODO */
1729     if (mode & (1 << 3))
1730         set_float_exception_flags(
1731                         get_float_exception_flags(&env->sse_status) &
1732                         ~float_flag_inexact,
1733                         &env->sse_status);
1734 #endif
1735     env->sse_status.float_rounding_mode = prev_rounding_mode;
1736 }
1737
1738 void glue(helper_roundsd, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1739 {
1740     signed char prev_rounding_mode;
1741
1742     prev_rounding_mode = env->sse_status.float_rounding_mode;
1743     if (!(mode & (1 << 2)))
1744         switch (mode & 3) {
1745         case 0:
1746             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1747             break;
1748         case 1:
1749             set_float_rounding_mode(float_round_down, &env->sse_status);
1750             break;
1751         case 2:
1752             set_float_rounding_mode(float_round_up, &env->sse_status);
1753             break;
1754         case 3:
1755             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1756             break;
1757         }
1758
1759     d->XMM_D(0) = float64_round_to_int(s->XMM_D(0), &env->sse_status);
1760
1761 #if 0 /* TODO */
1762     if (mode & (1 << 3))
1763         set_float_exception_flags(
1764                         get_float_exception_flags(&env->sse_status) &
1765                         ~float_flag_inexact,
1766                         &env->sse_status);
1767 #endif
1768     env->sse_status.float_rounding_mode = prev_rounding_mode;
1769 }
1770
1771 #define FBLENDP(d, s, m) m ? s : d
1772 SSE_HELPER_I(helper_blendps, L, 4, FBLENDP)
1773 SSE_HELPER_I(helper_blendpd, Q, 2, FBLENDP)
1774 SSE_HELPER_I(helper_pblendw, W, 8, FBLENDP)
1775
1776 void glue(helper_dpps, SUFFIX) (Reg *d, Reg *s, uint32_t mask)
1777 {
1778     float32 iresult = float32_zero;
1779
1780     if (mask & (1 << 4))
1781         iresult = float32_add(iresult,
1782                         float32_mul(d->XMM_S(0), s->XMM_S(0), &env->sse_status),
1783                         &env->sse_status);
1784     if (mask & (1 << 5))
1785         iresult = float32_add(iresult,
1786                         float32_mul(d->XMM_S(1), s->XMM_S(1), &env->sse_status),
1787                         &env->sse_status);
1788     if (mask & (1 << 6))
1789         iresult = float32_add(iresult,
1790                         float32_mul(d->XMM_S(2), s->XMM_S(2), &env->sse_status),
1791                         &env->sse_status);
1792     if (mask & (1 << 7))
1793         iresult = float32_add(iresult,
1794                         float32_mul(d->XMM_S(3), s->XMM_S(3), &env->sse_status),
1795                         &env->sse_status);
1796     d->XMM_S(0) = (mask & (1 << 0)) ? iresult : float32_zero;
1797     d->XMM_S(1) = (mask & (1 << 1)) ? iresult : float32_zero;
1798     d->XMM_S(2) = (mask & (1 << 2)) ? iresult : float32_zero;
1799     d->XMM_S(3) = (mask & (1 << 3)) ? iresult : float32_zero;
1800 }
1801
1802 void glue(helper_dppd, SUFFIX) (Reg *d, Reg *s, uint32_t mask)
1803 {
1804     float64 iresult = float64_zero;
1805
1806     if (mask & (1 << 4))
1807         iresult = float64_add(iresult,
1808                         float64_mul(d->XMM_D(0), s->XMM_D(0), &env->sse_status),
1809                         &env->sse_status);
1810     if (mask & (1 << 5))
1811         iresult = float64_add(iresult,
1812                         float64_mul(d->XMM_D(1), s->XMM_D(1), &env->sse_status),
1813                         &env->sse_status);
1814     d->XMM_D(0) = (mask & (1 << 0)) ? iresult : float64_zero;
1815     d->XMM_D(1) = (mask & (1 << 1)) ? iresult : float64_zero;
1816 }
1817
1818 void glue(helper_mpsadbw, SUFFIX) (Reg *d, Reg *s, uint32_t offset)
1819 {
1820     int s0 = (offset & 3) << 2;
1821     int d0 = (offset & 4) << 0;
1822     int i;
1823     Reg r;
1824
1825     for (i = 0; i < 8; i++, d0++) {
1826         r.W(i) = 0;
1827         r.W(i) += abs1(d->B(d0 + 0) - s->B(s0 + 0));
1828         r.W(i) += abs1(d->B(d0 + 1) - s->B(s0 + 1));
1829         r.W(i) += abs1(d->B(d0 + 2) - s->B(s0 + 2));
1830         r.W(i) += abs1(d->B(d0 + 3) - s->B(s0 + 3));
1831     }
1832
1833     *d = r;
1834 }
1835
1836 /* SSE4.2 op helpers */
1837 /* it's unclear whether signed or unsigned */
1838 #define FCMPGTQ(d, s) d > s ? -1 : 0
1839 SSE_HELPER_Q(helper_pcmpgtq, FCMPGTQ)
1840
1841 static inline int pcmp_elen(int reg, uint32_t ctrl)
1842 {
1843     int val;
1844
1845     /* Presence of REX.W is indicated by a bit higher than 7 set */
1846     if (ctrl >> 8)
1847         val = abs1((int64_t) env->regs[reg]);
1848     else
1849         val = abs1((int32_t) env->regs[reg]);
1850
1851     if (ctrl & 1) {
1852         if (val > 8)
1853             return 8;
1854     } else
1855         if (val > 16)
1856             return 16;
1857
1858     return val;
1859 }
1860
1861 static inline int pcmp_ilen(Reg *r, uint8_t ctrl)
1862 {
1863     int val = 0;
1864
1865     if (ctrl & 1) {
1866         while (val < 8 && r->W(val))
1867             val++;
1868     } else
1869         while (val < 16 && r->B(val))
1870             val++;
1871
1872     return val;
1873 }
1874
1875 static inline int pcmp_val(Reg *r, uint8_t ctrl, int i)
1876 {
1877     switch ((ctrl >> 0) & 3) {
1878     case 0:
1879         return r->B(i);
1880     case 1:
1881         return r->W(i);
1882     case 2:
1883         return (int8_t) r->B(i);
1884     case 3:
1885     default:
1886         return (int16_t) r->W(i);
1887     }
1888 }
1889
1890 static inline unsigned pcmpxstrx(Reg *d, Reg *s,
1891                 int8_t ctrl, int valids, int validd)
1892 {
1893     unsigned int res = 0;
1894     int v;
1895     int j, i;
1896     int upper = (ctrl & 1) ? 7 : 15;
1897
1898     valids--;
1899     validd--;
1900
1901     CC_SRC = (valids < upper ? CC_Z : 0) | (validd < upper ? CC_S : 0);
1902
1903     switch ((ctrl >> 2) & 3) {
1904     case 0:
1905         for (j = valids; j >= 0; j--) {
1906             res <<= 1;
1907             v = pcmp_val(s, ctrl, j);
1908             for (i = validd; i >= 0; i--)
1909                 res |= (v == pcmp_val(d, ctrl, i));
1910         }
1911         break;
1912     case 1:
1913         for (j = valids; j >= 0; j--) {
1914             res <<= 1;
1915             v = pcmp_val(s, ctrl, j);
1916             for (i = ((validd - 1) | 1); i >= 0; i -= 2)
1917                 res |= (pcmp_val(d, ctrl, i - 0) <= v &&
1918                         pcmp_val(d, ctrl, i - 1) >= v);
1919         }
1920         break;
1921     case 2:
1922         res = (2 << (upper - MAX(valids, validd))) - 1;
1923         res <<= MAX(valids, validd) - MIN(valids, validd);
1924         for (i = MIN(valids, validd); i >= 0; i--) {
1925             res <<= 1;
1926             v = pcmp_val(s, ctrl, i);
1927             res |= (v == pcmp_val(d, ctrl, i));
1928         }
1929         break;
1930     case 3:
1931         for (j = valids - validd; j >= 0; j--) {
1932             res <<= 1;
1933             res |= 1;
1934             for (i = MIN(upper - j, validd); i >= 0; i--)
1935                 res &= (pcmp_val(s, ctrl, i + j) == pcmp_val(d, ctrl, i));
1936         }
1937         break;
1938     }
1939
1940     switch ((ctrl >> 4) & 3) {
1941     case 1:
1942         res ^= (2 << upper) - 1;
1943         break;
1944     case 3:
1945         res ^= (2 << valids) - 1;
1946         break;
1947     }
1948
1949     if (res)
1950        CC_SRC |= CC_C;
1951     if (res & 1)
1952        CC_SRC |= CC_O;
1953
1954     return res;
1955 }
1956
1957 static inline int rffs1(unsigned int val)
1958 {
1959     int ret = 1, hi;
1960
1961     for (hi = sizeof(val) * 4; hi; hi /= 2)
1962         if (val >> hi) {
1963             val >>= hi;
1964             ret += hi;
1965         }
1966
1967     return ret;
1968 }
1969
1970 static inline int ffs1(unsigned int val)
1971 {
1972     int ret = 1, hi;
1973
1974     for (hi = sizeof(val) * 4; hi; hi /= 2)
1975         if (val << hi) {
1976             val <<= hi;
1977             ret += hi;
1978         }
1979
1980     return ret;
1981 }
1982
1983 void glue(helper_pcmpestri, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1984 {
1985     unsigned int res = pcmpxstrx(d, s, ctrl,
1986                     pcmp_elen(R_EDX, ctrl),
1987                     pcmp_elen(R_EAX, ctrl));
1988
1989     if (res)
1990         env->regs[R_ECX] = ((ctrl & (1 << 6)) ? rffs1 : ffs1)(res) - 1;
1991     else
1992         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
1993 }
1994
1995 void glue(helper_pcmpestrm, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1996 {
1997     int i;
1998     unsigned int res = pcmpxstrx(d, s, ctrl,
1999                     pcmp_elen(R_EDX, ctrl),
2000                     pcmp_elen(R_EAX, ctrl));
2001
2002     if ((ctrl >> 6) & 1) {
2003         if (ctrl & 1)
2004             for (i = 0; i < 8; i++, res >>= 1) {
2005                 d->W(i) = (res & 1) ? ~0 : 0;
2006             }
2007         else
2008             for (i = 0; i < 16; i++, res >>= 1) {
2009                 d->B(i) = (res & 1) ? ~0 : 0;
2010             }
2011     } else {
2012         d->Q(1) = 0;
2013         d->Q(0) = res;
2014     }
2015 }
2016
2017 void glue(helper_pcmpistri, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
2018 {
2019     unsigned int res = pcmpxstrx(d, s, ctrl,
2020                     pcmp_ilen(s, ctrl),
2021                     pcmp_ilen(d, ctrl));
2022
2023     if (res)
2024         env->regs[R_ECX] = ((ctrl & (1 << 6)) ? rffs1 : ffs1)(res) - 1;
2025     else
2026         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2027 }
2028
2029 void glue(helper_pcmpistrm, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
2030 {
2031     int i;
2032     unsigned int res = pcmpxstrx(d, s, ctrl,
2033                     pcmp_ilen(s, ctrl),
2034                     pcmp_ilen(d, ctrl));
2035
2036     if ((ctrl >> 6) & 1) {
2037         if (ctrl & 1)
2038             for (i = 0; i < 8; i++, res >>= 1) {
2039                 d->W(i) = (res & 1) ? ~0 : 0;
2040             }
2041         else
2042             for (i = 0; i < 16; i++, res >>= 1) {
2043                 d->B(i) = (res & 1) ? ~0 : 0;
2044             }
2045     } else {
2046         d->Q(1) = 0;
2047         d->Q(0) = res;
2048     }
2049 }
2050
2051 #define CRCPOLY        0x1edc6f41
2052 #define CRCPOLY_BITREV 0x82f63b78
2053 target_ulong helper_crc32(uint32_t crc1, target_ulong msg, uint32_t len)
2054 {
2055     target_ulong crc = (msg & ((target_ulong) -1 >>
2056                             (TARGET_LONG_BITS - len))) ^ crc1;
2057
2058     while (len--)
2059         crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_BITREV : 0);
2060
2061     return crc;
2062 }
2063
2064 #define POPMASK(i)     ((target_ulong) -1 / ((1LL << (1 << i)) + 1))
2065 #define POPCOUNT(n, i) (n & POPMASK(i)) + ((n >> (1 << i)) & POPMASK(i))
2066 target_ulong helper_popcnt(target_ulong n, uint32_t type)
2067 {
2068     CC_SRC = n ? 0 : CC_Z;
2069
2070     n = POPCOUNT(n, 0);
2071     n = POPCOUNT(n, 1);
2072     n = POPCOUNT(n, 2);
2073     n = POPCOUNT(n, 3);
2074     if (type == 1)
2075         return n & 0xff;
2076
2077     n = POPCOUNT(n, 4);
2078 #ifndef TARGET_X86_64
2079     return n;
2080 #else
2081     if (type == 2)
2082         return n & 0xff;
2083
2084     return POPCOUNT(n, 5);
2085 #endif
2086 }
2087 #endif
2088
2089 #undef SHIFT
2090 #undef XMM_ONLY
2091 #undef Reg
2092 #undef B
2093 #undef W
2094 #undef L
2095 #undef Q
2096 #undef SUFFIX