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, write to the Free Software
  19  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  */
  21 #if SHIFT == 0
  22 #define Reg MMXReg
  23 #define XMM_ONLY(x...)
  24 #define B(n) MMX_B(n)
  25 #define W(n) MMX_W(n)
  26 #define L(n) MMX_L(n)
  27 #define Q(n) q
  28 #define SUFFIX _mmx
  29 #else
  30 #define Reg XMMReg
  31 #define XMM_ONLY(x...) x
  32 #define B(n) XMM_B(n)
  33 #define W(n) XMM_W(n)
  34 #define L(n) XMM_L(n)
  35 #define Q(n) XMM_Q(n)
  36 #define SUFFIX _xmm
  37 #endif
  38
  39 void glue(helper_psrlw, SUFFIX)(Reg *d, Reg *s)
  40 {
  41     int shift;
  42
  43     if (s->Q(0) > 15) {
  44         d->Q(0) = 0;
  45 #if SHIFT == 1
  46         d->Q(1) = 0;
  47 #endif
  48     } else {
  49         shift = s->B(0);
  50         d->W(0) >>= shift;
  51         d->W(1) >>= shift;
  52         d->W(2) >>= shift;
  53         d->W(3) >>= shift;
  54 #if SHIFT == 1
  55         d->W(4) >>= shift;
  56         d->W(5) >>= shift;
  57         d->W(6) >>= shift;
  58         d->W(7) >>= shift;
  59 #endif
  60     }
  61     FORCE_RET();
  62 }
  63
  64 void glue(helper_psraw, SUFFIX)(Reg *d, Reg *s)
  65 {
  66     int shift;
  67
  68     if (s->Q(0) > 15) {
  69         shift = 15;
  70     } else {
  71         shift = s->B(0);
  72     }
  73     d->W(0) = (int16_t)d->W(0) >> shift;
  74     d->W(1) = (int16_t)d->W(1) >> shift;
  75     d->W(2) = (int16_t)d->W(2) >> shift;
  76     d->W(3) = (int16_t)d->W(3) >> shift;
  77 #if SHIFT == 1
  78     d->W(4) = (int16_t)d->W(4) >> shift;
  79     d->W(5) = (int16_t)d->W(5) >> shift;
  80     d->W(6) = (int16_t)d->W(6) >> shift;
  81     d->W(7) = (int16_t)d->W(7) >> shift;
  82 #endif
  83 }
  84
  85 void glue(helper_psllw, SUFFIX)(Reg *d, Reg *s)
  86 {
  87     int shift;
  88
  89     if (s->Q(0) > 15) {
  90         d->Q(0) = 0;
  91 #if SHIFT == 1
  92         d->Q(1) = 0;
  93 #endif
  94     } else {
  95         shift = s->B(0);
  96         d->W(0) <<= shift;
  97         d->W(1) <<= shift;
  98         d->W(2) <<= shift;
  99         d->W(3) <<= shift;
 100 #if SHIFT == 1
 101         d->W(4) <<= shift;
 102         d->W(5) <<= shift;
 103         d->W(6) <<= shift;
 104         d->W(7) <<= shift;
 105 #endif
 106     }
 107     FORCE_RET();
 108 }
 109
 110 void glue(helper_psrld, SUFFIX)(Reg *d, Reg *s)
 111 {
 112     int shift;
 113
 114     if (s->Q(0) > 31) {
 115         d->Q(0) = 0;
 116 #if SHIFT == 1
 117         d->Q(1) = 0;
 118 #endif
 119     } else {
 120         shift = s->B(0);
 121         d->L(0) >>= shift;
 122         d->L(1) >>= shift;
 123 #if SHIFT == 1
 124         d->L(2) >>= shift;
 125         d->L(3) >>= shift;
 126 #endif
 127     }
 128     FORCE_RET();
 129 }
 130
 131 void glue(helper_psrad, SUFFIX)(Reg *d, Reg *s)
 132 {
 133     int shift;
 134
 135     if (s->Q(0) > 31) {
 136         shift = 31;
 137     } else {
 138         shift = s->B(0);
 139     }
 140     d->L(0) = (int32_t)d->L(0) >> shift;
 141     d->L(1) = (int32_t)d->L(1) >> shift;
 142 #if SHIFT == 1
 143     d->L(2) = (int32_t)d->L(2) >> shift;
 144     d->L(3) = (int32_t)d->L(3) >> shift;
 145 #endif
 146 }
 147
 148 void glue(helper_pslld, SUFFIX)(Reg *d, Reg *s)
 149 {
 150     int shift;
 151
 152     if (s->Q(0) > 31) {
 153         d->Q(0) = 0;
 154 #if SHIFT == 1
 155         d->Q(1) = 0;
 156 #endif
 157     } else {
 158         shift = s->B(0);
 159         d->L(0) <<= shift;
 160         d->L(1) <<= shift;
 161 #if SHIFT == 1
 162         d->L(2) <<= shift;
 163         d->L(3) <<= shift;
 164 #endif
 165     }
 166     FORCE_RET();
 167 }
 168
 169 void glue(helper_psrlq, SUFFIX)(Reg *d, Reg *s)
 170 {
 171     int shift;
 172
 173     if (s->Q(0) > 63) {
 174         d->Q(0) = 0;
 175 #if SHIFT == 1
 176         d->Q(1) = 0;
 177 #endif
 178     } else {
 179         shift = s->B(0);
 180         d->Q(0) >>= shift;
 181 #if SHIFT == 1
 182         d->Q(1) >>= shift;
 183 #endif
 184     }
 185     FORCE_RET();
 186 }
 187
 188 void glue(helper_psllq, SUFFIX)(Reg *d, Reg *s)
 189 {
 190     int shift;
 191
 192     if (s->Q(0) > 63) {
 193         d->Q(0) = 0;
 194 #if SHIFT == 1
 195         d->Q(1) = 0;
 196 #endif
 197     } else {
 198         shift = s->B(0);
 199         d->Q(0) <<= shift;
 200 #if SHIFT == 1
 201         d->Q(1) <<= shift;
 202 #endif
 203     }
 204     FORCE_RET();
 205 }
 206
 207 #if SHIFT == 1
 208 void glue(helper_psrldq, SUFFIX)(Reg *d, Reg *s)
 209 {
 210     int shift, i;
 211
 212     shift = s->L(0);
 213     if (shift > 16)
 214         shift = 16;
 215     for(i = 0; i < 16 - shift; i++)
 216         d->B(i) = d->B(i + shift);
 217     for(i = 16 - shift; i < 16; i++)
 218         d->B(i) = 0;
 219     FORCE_RET();
 220 }
 221
 222 void glue(helper_pslldq, SUFFIX)(Reg *d, Reg *s)
 223 {
 224     int shift, i;
 225
 226     shift = s->L(0);
 227     if (shift > 16)
 228         shift = 16;
 229     for(i = 15; i >= shift; i--)
 230         d->B(i) = d->B(i - shift);
 231     for(i = 0; i < shift; i++)
 232         d->B(i) = 0;
 233     FORCE_RET();
 234 }
 235 #endif
 236
 237 #define SSE_HELPER_B(name, F)\
 238 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 239 {\
 240     d->B(0) = F(d->B(0), s->B(0));\
 241     d->B(1) = F(d->B(1), s->B(1));\
 242     d->B(2) = F(d->B(2), s->B(2));\
 243     d->B(3) = F(d->B(3), s->B(3));\
 244     d->B(4) = F(d->B(4), s->B(4));\
 245     d->B(5) = F(d->B(5), s->B(5));\
 246     d->B(6) = F(d->B(6), s->B(6));\
 247     d->B(7) = F(d->B(7), s->B(7));\
 248     XMM_ONLY(\
 249     d->B(8) = F(d->B(8), s->B(8));\
 250     d->B(9) = F(d->B(9), s->B(9));\
 251     d->B(10) = F(d->B(10), s->B(10));\
 252     d->B(11) = F(d->B(11), s->B(11));\
 253     d->B(12) = F(d->B(12), s->B(12));\
 254     d->B(13) = F(d->B(13), s->B(13));\
 255     d->B(14) = F(d->B(14), s->B(14));\
 256     d->B(15) = F(d->B(15), s->B(15));\
 257     )\
 258 }
 259
 260 #define SSE_HELPER_W(name, F)\
 261 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 262 {\
 263     d->W(0) = F(d->W(0), s->W(0));\
 264     d->W(1) = F(d->W(1), s->W(1));\
 265     d->W(2) = F(d->W(2), s->W(2));\
 266     d->W(3) = F(d->W(3), s->W(3));\
 267     XMM_ONLY(\
 268     d->W(4) = F(d->W(4), s->W(4));\
 269     d->W(5) = F(d->W(5), s->W(5));\
 270     d->W(6) = F(d->W(6), s->W(6));\
 271     d->W(7) = F(d->W(7), s->W(7));\
 272     )\
 273 }
 274
 275 #define SSE_HELPER_L(name, F)\
 276 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 277 {\
 278     d->L(0) = F(d->L(0), s->L(0));\
 279     d->L(1) = F(d->L(1), s->L(1));\
 280     XMM_ONLY(\
 281     d->L(2) = F(d->L(2), s->L(2));\
 282     d->L(3) = F(d->L(3), s->L(3));\
 283     )\
 284 }
 285
 286 #define SSE_HELPER_Q(name, F)\
 287 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 288 {\
 289     d->Q(0) = F(d->Q(0), s->Q(0));\
 290     XMM_ONLY(\
 291     d->Q(1) = F(d->Q(1), s->Q(1));\
 292     )\
 293 }
 294
 295 #if SHIFT == 0
 296 static inline int satub(int x)
 297 {
 298     if (x < 0)
 299         return 0;
 300     else if (x > 255)
 301         return 255;
 302     else
 303         return x;
 304 }
 305
 306 static inline int satuw(int x)
 307 {
 308     if (x < 0)
 309         return 0;
 310     else if (x > 65535)
 311         return 65535;
 312     else
 313         return x;
 314 }
 315
 316 static inline int satsb(int x)
 317 {
 318     if (x < -128)
 319         return -128;
 320     else if (x > 127)
 321         return 127;
 322     else
 323         return x;
 324 }
 325
 326 static inline int satsw(int x)
 327 {
 328     if (x < -32768)
 329         return -32768;
 330     else if (x > 32767)
 331         return 32767;
 332     else
 333         return x;
 334 }
 335
 336 #define FADD(a, b) ((a) + (b))
 337 #define FADDUB(a, b) satub((a) + (b))
 338 #define FADDUW(a, b) satuw((a) + (b))
 339 #define FADDSB(a, b) satsb((int8_t)(a) + (int8_t)(b))
 340 #define FADDSW(a, b) satsw((int16_t)(a) + (int16_t)(b))
 341
 342 #define FSUB(a, b) ((a) - (b))
 343 #define FSUBUB(a, b) satub((a) - (b))
 344 #define FSUBUW(a, b) satuw((a) - (b))
 345 #define FSUBSB(a, b) satsb((int8_t)(a) - (int8_t)(b))
 346 #define FSUBSW(a, b) satsw((int16_t)(a) - (int16_t)(b))
 347 #define FMINUB(a, b) ((a) < (b)) ? (a) : (b)
 348 #define FMINSW(a, b) ((int16_t)(a) < (int16_t)(b)) ? (a) : (b)
 349 #define FMAXUB(a, b) ((a) > (b)) ? (a) : (b)
 350 #define FMAXSW(a, b) ((int16_t)(a) > (int16_t)(b)) ? (a) : (b)
 351
 352 #define FAND(a, b) (a) & (b)
 353 #define FANDN(a, b) ((~(a)) & (b))
 354 #define FOR(a, b) (a) | (b)
 355 #define FXOR(a, b) (a) ^ (b)
 356
 357 #define FCMPGTB(a, b) (int8_t)(a) > (int8_t)(b) ? -1 : 0
 358 #define FCMPGTW(a, b) (int16_t)(a) > (int16_t)(b) ? -1 : 0
 359 #define FCMPGTL(a, b) (int32_t)(a) > (int32_t)(b) ? -1 : 0
 360 #define FCMPEQ(a, b) (a) == (b) ? -1 : 0
 361
 362 #define FMULLW(a, b) (a) * (b)
 363 #define FMULHRW(a, b) ((int16_t)(a) * (int16_t)(b) + 0x8000) >> 16
 364 #define FMULHUW(a, b) (a) * (b) >> 16
 365 #define FMULHW(a, b) (int16_t)(a) * (int16_t)(b) >> 16
 366
 367 #define FAVG(a, b) ((a) + (b) + 1) >> 1
 368 #endif
 369
 370 SSE_HELPER_B(helper_paddb, FADD)
 371 SSE_HELPER_W(helper_paddw, FADD)
 372 SSE_HELPER_L(helper_paddl, FADD)
 373 SSE_HELPER_Q(helper_paddq, FADD)
 374
 375 SSE_HELPER_B(helper_psubb, FSUB)
 376 SSE_HELPER_W(helper_psubw, FSUB)
 377 SSE_HELPER_L(helper_psubl, FSUB)
 378 SSE_HELPER_Q(helper_psubq, FSUB)
 379
 380 SSE_HELPER_B(helper_paddusb, FADDUB)
 381 SSE_HELPER_B(helper_paddsb, FADDSB)
 382 SSE_HELPER_B(helper_psubusb, FSUBUB)
 383 SSE_HELPER_B(helper_psubsb, FSUBSB)
 384
 385 SSE_HELPER_W(helper_paddusw, FADDUW)
 386 SSE_HELPER_W(helper_paddsw, FADDSW)
 387 SSE_HELPER_W(helper_psubusw, FSUBUW)
 388 SSE_HELPER_W(helper_psubsw, FSUBSW)
 389
 390 SSE_HELPER_B(helper_pminub, FMINUB)
 391 SSE_HELPER_B(helper_pmaxub, FMAXUB)
 392
 393 SSE_HELPER_W(helper_pminsw, FMINSW)
 394 SSE_HELPER_W(helper_pmaxsw, FMAXSW)
 395
 396 SSE_HELPER_Q(helper_pand, FAND)
 397 SSE_HELPER_Q(helper_pandn, FANDN)
 398 SSE_HELPER_Q(helper_por, FOR)
 399 SSE_HELPER_Q(helper_pxor, FXOR)
 400
 401 SSE_HELPER_B(helper_pcmpgtb, FCMPGTB)
 402 SSE_HELPER_W(helper_pcmpgtw, FCMPGTW)
 403 SSE_HELPER_L(helper_pcmpgtl, FCMPGTL)
 404
 405 SSE_HELPER_B(helper_pcmpeqb, FCMPEQ)
 406 SSE_HELPER_W(helper_pcmpeqw, FCMPEQ)
 407 SSE_HELPER_L(helper_pcmpeql, FCMPEQ)
 408
 409 SSE_HELPER_W(helper_pmullw, FMULLW)
 410 #if SHIFT == 0
 411 SSE_HELPER_W(helper_pmulhrw, FMULHRW)
 412 #endif
 413 SSE_HELPER_W(helper_pmulhuw, FMULHUW)
 414 SSE_HELPER_W(helper_pmulhw, FMULHW)
 415
 416 SSE_HELPER_B(helper_pavgb, FAVG)
 417 SSE_HELPER_W(helper_pavgw, FAVG)
 418
 419 void glue(helper_pmuludq, SUFFIX) (Reg *d, Reg *s)
 420 {
 421     d->Q(0) = (uint64_t)s->L(0) * (uint64_t)d->L(0);
 422 #if SHIFT == 1
 423     d->Q(1) = (uint64_t)s->L(2) * (uint64_t)d->L(2);
 424 #endif
 425 }
 426
 427 void glue(helper_pmaddwd, SUFFIX) (Reg *d, Reg *s)
 428 {
 429     int i;
 430
 431     for(i = 0; i < (2 << SHIFT); i++) {
 432         d->L(i) = (int16_t)s->W(2*i) * (int16_t)d->W(2*i) +
 433             (int16_t)s->W(2*i+1) * (int16_t)d->W(2*i+1);
 434     }
 435     FORCE_RET();
 436 }
 437
 438 #if SHIFT == 0
 439 static inline int abs1(int a)
 440 {
 441     if (a < 0)
 442         return -a;
 443     else
 444         return a;
 445 }
 446 #endif
 447 void glue(helper_psadbw, SUFFIX) (Reg *d, Reg *s)
 448 {
 449     unsigned int val;
 450
 451     val = 0;
 452     val += abs1(d->B(0) - s->B(0));
 453     val += abs1(d->B(1) - s->B(1));
 454     val += abs1(d->B(2) - s->B(2));
 455     val += abs1(d->B(3) - s->B(3));
 456     val += abs1(d->B(4) - s->B(4));
 457     val += abs1(d->B(5) - s->B(5));
 458     val += abs1(d->B(6) - s->B(6));
 459     val += abs1(d->B(7) - s->B(7));
 460     d->Q(0) = val;
 461 #if SHIFT == 1
 462     val = 0;
 463     val += abs1(d->B(8) - s->B(8));
 464     val += abs1(d->B(9) - s->B(9));
 465     val += abs1(d->B(10) - s->B(10));
 466     val += abs1(d->B(11) - s->B(11));
 467     val += abs1(d->B(12) - s->B(12));
 468     val += abs1(d->B(13) - s->B(13));
 469     val += abs1(d->B(14) - s->B(14));
 470     val += abs1(d->B(15) - s->B(15));
 471     d->Q(1) = val;
 472 #endif
 473 }
 474
 475 void glue(helper_maskmov, SUFFIX) (Reg *d, Reg *s, target_ulong a0)
 476 {
 477     int i;
 478     for(i = 0; i < (8 << SHIFT); i++) {
 479         if (s->B(i) & 0x80)
 480             stb(a0 + i, d->B(i));
 481     }
 482     FORCE_RET();
 483 }
 484
 485 void glue(helper_movl_mm_T0, SUFFIX) (Reg *d, uint32_t val)
 486 {
 487     d->L(0) = val;
 488     d->L(1) = 0;
 489 #if SHIFT == 1
 490     d->Q(1) = 0;
 491 #endif
 492 }
 493
 494 #ifdef TARGET_X86_64
 495 void glue(helper_movq_mm_T0, SUFFIX) (Reg *d, uint64_t val)
 496 {
 497     d->Q(0) = val;
 498 #if SHIFT == 1
 499     d->Q(1) = 0;
 500 #endif
 501 }
 502 #endif
 503
 504 #if SHIFT == 0
 505 void glue(helper_pshufw, SUFFIX) (Reg *d, Reg *s, int order)
 506 {
 507     Reg r;
 508     r.W(0) = s->W(order & 3);
 509     r.W(1) = s->W((order >> 2) & 3);
 510     r.W(2) = s->W((order >> 4) & 3);
 511     r.W(3) = s->W((order >> 6) & 3);
 512     *d = r;
 513 }
 514 #else
 515 void helper_shufps(Reg *d, Reg *s, int order)
 516 {
 517     Reg r;
 518     r.L(0) = d->L(order & 3);
 519     r.L(1) = d->L((order >> 2) & 3);
 520     r.L(2) = s->L((order >> 4) & 3);
 521     r.L(3) = s->L((order >> 6) & 3);
 522     *d = r;
 523 }
 524
 525 void helper_shufpd(Reg *d, Reg *s, int order)
 526 {
 527     Reg r;
 528     r.Q(0) = d->Q(order & 1);
 529     r.Q(1) = s->Q((order >> 1) & 1);
 530     *d = r;
 531 }
 532
 533 void glue(helper_pshufd, SUFFIX) (Reg *d, Reg *s, int order)
 534 {
 535     Reg r;
 536     r.L(0) = s->L(order & 3);
 537     r.L(1) = s->L((order >> 2) & 3);
 538     r.L(2) = s->L((order >> 4) & 3);
 539     r.L(3) = s->L((order >> 6) & 3);
 540     *d = r;
 541 }
 542
 543 void glue(helper_pshuflw, SUFFIX) (Reg *d, Reg *s, int order)
 544 {
 545     Reg r;
 546     r.W(0) = s->W(order & 3);
 547     r.W(1) = s->W((order >> 2) & 3);
 548     r.W(2) = s->W((order >> 4) & 3);
 549     r.W(3) = s->W((order >> 6) & 3);
 550     r.Q(1) = s->Q(1);
 551     *d = r;
 552 }
 553
 554 void glue(helper_pshufhw, SUFFIX) (Reg *d, Reg *s, int order)
 555 {
 556     Reg r;
 557     r.Q(0) = s->Q(0);
 558     r.W(4) = s->W(4 + (order & 3));
 559     r.W(5) = s->W(4 + ((order >> 2) & 3));
 560     r.W(6) = s->W(4 + ((order >> 4) & 3));
 561     r.W(7) = s->W(4 + ((order >> 6) & 3));
 562     *d = r;
 563 }
 564 #endif
 565
 566 #if SHIFT == 1
 567 /* FPU ops */
 568 /* XXX: not accurate */
 569
 570 #define SSE_HELPER_S(name, F)\
 571 void helper_ ## name ## ps (Reg *d, Reg *s)\
 572 {\
 573     d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 574     d->XMM_S(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
 575     d->XMM_S(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
 576     d->XMM_S(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
 577 }\
 578 \
 579 void helper_ ## name ## ss (Reg *d, Reg *s)\
 580 {\
 581     d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 582 }\
 583 void helper_ ## name ## pd (Reg *d, Reg *s)\
 584 {\
 585     d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 586     d->XMM_D(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
 587 }\
 588 \
 589 void helper_ ## name ## sd (Reg *d, Reg *s)\
 590 {\
 591     d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 592 }
 593
 594 #define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
 595 #define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
 596 #define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
 597 #define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
 598 #define FPU_MIN(size, a, b) (a) < (b) ? (a) : (b)
 599 #define FPU_MAX(size, a, b) (a) > (b) ? (a) : (b)
 600 #define FPU_SQRT(size, a, b) float ## size ## _sqrt(b, &env->sse_status)
 601
 602 SSE_HELPER_S(add, FPU_ADD)
 603 SSE_HELPER_S(sub, FPU_SUB)
 604 SSE_HELPER_S(mul, FPU_MUL)
 605 SSE_HELPER_S(div, FPU_DIV)
 606 SSE_HELPER_S(min, FPU_MIN)
 607 SSE_HELPER_S(max, FPU_MAX)
 608 SSE_HELPER_S(sqrt, FPU_SQRT)
 609
 610
 611 /* float to float conversions */
 612 void helper_cvtps2pd(Reg *d, Reg *s)
 613 {
 614     float32 s0, s1;
 615     s0 = s->XMM_S(0);
 616     s1 = s->XMM_S(1);
 617     d->XMM_D(0) = float32_to_float64(s0, &env->sse_status);
 618     d->XMM_D(1) = float32_to_float64(s1, &env->sse_status);
 619 }
 620
 621 void helper_cvtpd2ps(Reg *d, Reg *s)
 622 {
 623     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 624     d->XMM_S(1) = float64_to_float32(s->XMM_D(1), &env->sse_status);
 625     d->Q(1) = 0;
 626 }
 627
 628 void helper_cvtss2sd(Reg *d, Reg *s)
 629 {
 630     d->XMM_D(0) = float32_to_float64(s->XMM_S(0), &env->sse_status);
 631 }
 632
 633 void helper_cvtsd2ss(Reg *d, Reg *s)
 634 {
 635     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 636 }
 637
 638 /* integer to float */
 639 void helper_cvtdq2ps(Reg *d, Reg *s)
 640 {
 641     d->XMM_S(0) = int32_to_float32(s->XMM_L(0), &env->sse_status);
 642     d->XMM_S(1) = int32_to_float32(s->XMM_L(1), &env->sse_status);
 643     d->XMM_S(2) = int32_to_float32(s->XMM_L(2), &env->sse_status);
 644     d->XMM_S(3) = int32_to_float32(s->XMM_L(3), &env->sse_status);
 645 }
 646
 647 void helper_cvtdq2pd(Reg *d, Reg *s)
 648 {
 649     int32_t l0, l1;
 650     l0 = (int32_t)s->XMM_L(0);
 651     l1 = (int32_t)s->XMM_L(1);
 652     d->XMM_D(0) = int32_to_float64(l0, &env->sse_status);
 653     d->XMM_D(1) = int32_to_float64(l1, &env->sse_status);
 654 }
 655
 656 void helper_cvtpi2ps(XMMReg *d, MMXReg *s)
 657 {
 658     d->XMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
 659     d->XMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
 660 }
 661
 662 void helper_cvtpi2pd(XMMReg *d, MMXReg *s)
 663 {
 664     d->XMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
 665     d->XMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
 666 }
 667
 668 void helper_cvtsi2ss(XMMReg *d, uint32_t val)
 669 {
 670     d->XMM_S(0) = int32_to_float32(val, &env->sse_status);
 671 }
 672
 673 void helper_cvtsi2sd(XMMReg *d, uint32_t val)
 674 {
 675     d->XMM_D(0) = int32_to_float64(val, &env->sse_status);
 676 }
 677
 678 #ifdef TARGET_X86_64
 679 void helper_cvtsq2ss(XMMReg *d, uint64_t val)
 680 {
 681     d->XMM_S(0) = int64_to_float32(val, &env->sse_status);
 682 }
 683
 684 void helper_cvtsq2sd(XMMReg *d, uint64_t val)
 685 {
 686     d->XMM_D(0) = int64_to_float64(val, &env->sse_status);
 687 }
 688 #endif
 689
 690 /* float to integer */
 691 void helper_cvtps2dq(XMMReg *d, XMMReg *s)
 692 {
 693     d->XMM_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 694     d->XMM_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 695     d->XMM_L(2) = float32_to_int32(s->XMM_S(2), &env->sse_status);
 696     d->XMM_L(3) = float32_to_int32(s->XMM_S(3), &env->sse_status);
 697 }
 698
 699 void helper_cvtpd2dq(XMMReg *d, XMMReg *s)
 700 {
 701     d->XMM_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 702     d->XMM_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 703     d->XMM_Q(1) = 0;
 704 }
 705
 706 void helper_cvtps2pi(MMXReg *d, XMMReg *s)
 707 {
 708     d->MMX_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 709     d->MMX_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 710 }
 711
 712 void helper_cvtpd2pi(MMXReg *d, XMMReg *s)
 713 {
 714     d->MMX_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 715     d->MMX_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 716 }
 717
 718 int32_t helper_cvtss2si(XMMReg *s)
 719 {
 720     return float32_to_int32(s->XMM_S(0), &env->sse_status);
 721 }
 722
 723 int32_t helper_cvtsd2si(XMMReg *s)
 724 {
 725     return float64_to_int32(s->XMM_D(0), &env->sse_status);
 726 }
 727
 728 #ifdef TARGET_X86_64
 729 int64_t helper_cvtss2sq(XMMReg *s)
 730 {
 731     return float32_to_int64(s->XMM_S(0), &env->sse_status);
 732 }
 733
 734 int64_t helper_cvtsd2sq(XMMReg *s)
 735 {
 736     return float64_to_int64(s->XMM_D(0), &env->sse_status);
 737 }
 738 #endif
 739
 740 /* float to integer truncated */
 741 void helper_cvttps2dq(XMMReg *d, XMMReg *s)
 742 {
 743     d->XMM_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 744     d->XMM_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 745     d->XMM_L(2) = float32_to_int32_round_to_zero(s->XMM_S(2), &env->sse_status);
 746     d->XMM_L(3) = float32_to_int32_round_to_zero(s->XMM_S(3), &env->sse_status);
 747 }
 748
 749 void helper_cvttpd2dq(XMMReg *d, XMMReg *s)
 750 {
 751     d->XMM_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 752     d->XMM_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 753     d->XMM_Q(1) = 0;
 754 }
 755
 756 void helper_cvttps2pi(MMXReg *d, XMMReg *s)
 757 {
 758     d->MMX_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 759     d->MMX_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 760 }
 761
 762 void helper_cvttpd2pi(MMXReg *d, XMMReg *s)
 763 {
 764     d->MMX_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 765     d->MMX_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 766 }
 767
 768 int32_t helper_cvttss2si(XMMReg *s)
 769 {
 770     return float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 771 }
 772
 773 int32_t helper_cvttsd2si(XMMReg *s)
 774 {
 775     return float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 776 }
 777
 778 #ifdef TARGET_X86_64
 779 int64_t helper_cvttss2sq(XMMReg *s)
 780 {
 781     return float32_to_int64_round_to_zero(s->XMM_S(0), &env->sse_status);
 782 }
 783
 784 int64_t helper_cvttsd2sq(XMMReg *s)
 785 {
 786     return float64_to_int64_round_to_zero(s->XMM_D(0), &env->sse_status);
 787 }
 788 #endif
 789
 790 void helper_rsqrtps(XMMReg *d, XMMReg *s)
 791 {
 792     d->XMM_S(0) = approx_rsqrt(s->XMM_S(0));
 793     d->XMM_S(1) = approx_rsqrt(s->XMM_S(1));
 794     d->XMM_S(2) = approx_rsqrt(s->XMM_S(2));
 795     d->XMM_S(3) = approx_rsqrt(s->XMM_S(3));
 796 }
 797
 798 void helper_rsqrtss(XMMReg *d, XMMReg *s)
 799 {
 800     d->XMM_S(0) = approx_rsqrt(s->XMM_S(0));
 801 }
 802
 803 void helper_rcpps(XMMReg *d, XMMReg *s)
 804 {
 805     d->XMM_S(0) = approx_rcp(s->XMM_S(0));
 806     d->XMM_S(1) = approx_rcp(s->XMM_S(1));
 807     d->XMM_S(2) = approx_rcp(s->XMM_S(2));
 808     d->XMM_S(3) = approx_rcp(s->XMM_S(3));
 809 }
 810
 811 void helper_rcpss(XMMReg *d, XMMReg *s)
 812 {
 813     d->XMM_S(0) = approx_rcp(s->XMM_S(0));
 814 }
 815
 816 void helper_haddps(XMMReg *d, XMMReg *s)
 817 {
 818     XMMReg r;
 819     r.XMM_S(0) = d->XMM_S(0) + d->XMM_S(1);
 820     r.XMM_S(1) = d->XMM_S(2) + d->XMM_S(3);
 821     r.XMM_S(2) = s->XMM_S(0) + s->XMM_S(1);
 822     r.XMM_S(3) = s->XMM_S(2) + s->XMM_S(3);
 823     *d = r;
 824 }
 825
 826 void helper_haddpd(XMMReg *d, XMMReg *s)
 827 {
 828     XMMReg r;
 829     r.XMM_D(0) = d->XMM_D(0) + d->XMM_D(1);
 830     r.XMM_D(1) = s->XMM_D(0) + s->XMM_D(1);
 831     *d = r;
 832 }
 833
 834 void helper_hsubps(XMMReg *d, XMMReg *s)
 835 {
 836     XMMReg r;
 837     r.XMM_S(0) = d->XMM_S(0) - d->XMM_S(1);
 838     r.XMM_S(1) = d->XMM_S(2) - d->XMM_S(3);
 839     r.XMM_S(2) = s->XMM_S(0) - s->XMM_S(1);
 840     r.XMM_S(3) = s->XMM_S(2) - s->XMM_S(3);
 841     *d = r;
 842 }
 843
 844 void helper_hsubpd(XMMReg *d, XMMReg *s)
 845 {
 846     XMMReg r;
 847     r.XMM_D(0) = d->XMM_D(0) - d->XMM_D(1);
 848     r.XMM_D(1) = s->XMM_D(0) - s->XMM_D(1);
 849     *d = r;
 850 }
 851
 852 void helper_addsubps(XMMReg *d, XMMReg *s)
 853 {
 854     d->XMM_S(0) = d->XMM_S(0) - s->XMM_S(0);
 855     d->XMM_S(1) = d->XMM_S(1) + s->XMM_S(1);
 856     d->XMM_S(2) = d->XMM_S(2) - s->XMM_S(2);
 857     d->XMM_S(3) = d->XMM_S(3) + s->XMM_S(3);
 858 }
 859
 860 void helper_addsubpd(XMMReg *d, XMMReg *s)
 861 {
 862     d->XMM_D(0) = d->XMM_D(0) - s->XMM_D(0);
 863     d->XMM_D(1) = d->XMM_D(1) + s->XMM_D(1);
 864 }
 865
 866 /* XXX: unordered */
 867 #define SSE_HELPER_CMP(name, F)\
 868 void helper_ ## name ## ps (Reg *d, Reg *s)\
 869 {\
 870     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 871     d->XMM_L(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
 872     d->XMM_L(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
 873     d->XMM_L(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
 874 }\
 875 \
 876 void helper_ ## name ## ss (Reg *d, Reg *s)\
 877 {\
 878     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 879 }\
 880 void helper_ ## name ## pd (Reg *d, Reg *s)\
 881 {\
 882     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 883     d->XMM_Q(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
 884 }\
 885 \
 886 void helper_ ## name ## sd (Reg *d, Reg *s)\
 887 {\
 888     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 889 }
 890
 891 #define FPU_CMPEQ(size, a, b) float ## size ## _eq(a, b, &env->sse_status) ? -1 : 0
 892 #define FPU_CMPLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? -1 : 0
 893 #define FPU_CMPLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? -1 : 0
 894 #define FPU_CMPUNORD(size, a, b) float ## size ## _unordered(a, b, &env->sse_status) ? - 1 : 0
 895 #define FPU_CMPNEQ(size, a, b) float ## size ## _eq(a, b, &env->sse_status) ? 0 : -1
 896 #define FPU_CMPNLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? 0 : -1
 897 #define FPU_CMPNLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? 0 : -1
 898 #define FPU_CMPORD(size, a, b) float ## size ## _unordered(a, b, &env->sse_status) ? 0 : -1
 899
 900 SSE_HELPER_CMP(cmpeq, FPU_CMPEQ)
 901 SSE_HELPER_CMP(cmplt, FPU_CMPLT)
 902 SSE_HELPER_CMP(cmple, FPU_CMPLE)
 903 SSE_HELPER_CMP(cmpunord, FPU_CMPUNORD)
 904 SSE_HELPER_CMP(cmpneq, FPU_CMPNEQ)
 905 SSE_HELPER_CMP(cmpnlt, FPU_CMPNLT)
 906 SSE_HELPER_CMP(cmpnle, FPU_CMPNLE)
 907 SSE_HELPER_CMP(cmpord, FPU_CMPORD)
 908
 909 const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
 910
 911 void helper_ucomiss(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_quiet(s0, s1, &env->sse_status);
 919     CC_SRC = comis_eflags[ret + 1];
 920     FORCE_RET();
 921 }
 922
 923 void helper_comiss(Reg *d, Reg *s)
 924 {
 925     int ret;
 926     float32 s0, s1;
 927
 928     s0 = d->XMM_S(0);
 929     s1 = s->XMM_S(0);
 930     ret = float32_compare(s0, s1, &env->sse_status);
 931     CC_SRC = comis_eflags[ret + 1];
 932     FORCE_RET();
 933 }
 934
 935 void helper_ucomisd(Reg *d, Reg *s)
 936 {
 937     int ret;
 938     float64 d0, d1;
 939
 940     d0 = d->XMM_D(0);
 941     d1 = s->XMM_D(0);
 942     ret = float64_compare_quiet(d0, d1, &env->sse_status);
 943     CC_SRC = comis_eflags[ret + 1];
 944     FORCE_RET();
 945 }
 946
 947 void helper_comisd(Reg *d, Reg *s)
 948 {
 949     int ret;
 950     float64 d0, d1;
 951
 952     d0 = d->XMM_D(0);
 953     d1 = s->XMM_D(0);
 954     ret = float64_compare(d0, d1, &env->sse_status);
 955     CC_SRC = comis_eflags[ret + 1];
 956     FORCE_RET();
 957 }
 958
 959 uint32_t helper_movmskps(Reg *s)
 960 {
 961     int b0, b1, b2, b3;
 962     b0 = s->XMM_L(0) >> 31;
 963     b1 = s->XMM_L(1) >> 31;
 964     b2 = s->XMM_L(2) >> 31;
 965     b3 = s->XMM_L(3) >> 31;
 966     return b0 | (b1 << 1) | (b2 << 2) | (b3 << 3);
 967 }
 968
 969 uint32_t helper_movmskpd(Reg *s)
 970 {
 971     int b0, b1;
 972     b0 = s->XMM_L(1) >> 31;
 973     b1 = s->XMM_L(3) >> 31;
 974     return b0 | (b1 << 1);
 975 }
 976
 977 #endif
 978
 979 uint32_t glue(helper_pmovmskb, SUFFIX)(Reg *s)
 980 {
 981     uint32_t val;
 982     val = 0;
 983     val |= (s->XMM_B(0) >> 7);
 984     val |= (s->XMM_B(1) >> 6) & 0x02;
 985     val |= (s->XMM_B(2) >> 5) & 0x04;
 986     val |= (s->XMM_B(3) >> 4) & 0x08;
 987     val |= (s->XMM_B(4) >> 3) & 0x10;
 988     val |= (s->XMM_B(5) >> 2) & 0x20;
 989     val |= (s->XMM_B(6) >> 1) & 0x40;
 990     val |= (s->XMM_B(7)) & 0x80;
 991 #if SHIFT == 1
 992     val |= (s->XMM_B(8) << 1) & 0x0100;
 993     val |= (s->XMM_B(9) << 2) & 0x0200;
 994     val |= (s->XMM_B(10) << 3) & 0x0400;
 995     val |= (s->XMM_B(11) << 4) & 0x0800;
 996     val |= (s->XMM_B(12) << 5) & 0x1000;
 997     val |= (s->XMM_B(13) << 6) & 0x2000;
 998     val |= (s->XMM_B(14) << 7) & 0x4000;
 999     val |= (s->XMM_B(15) << 8) & 0x8000;
1000 #endif
1001     return val;
1002 }
1003
1004 void glue(helper_packsswb, SUFFIX) (Reg *d, Reg *s)
1005 {
1006     Reg r;
1007
1008     r.B(0) = satsb((int16_t)d->W(0));
1009     r.B(1) = satsb((int16_t)d->W(1));
1010     r.B(2) = satsb((int16_t)d->W(2));
1011     r.B(3) = satsb((int16_t)d->W(3));
1012 #if SHIFT == 1
1013     r.B(4) = satsb((int16_t)d->W(4));
1014     r.B(5) = satsb((int16_t)d->W(5));
1015     r.B(6) = satsb((int16_t)d->W(6));
1016     r.B(7) = satsb((int16_t)d->W(7));
1017 #endif
1018     r.B((4 << SHIFT) + 0) = satsb((int16_t)s->W(0));
1019     r.B((4 << SHIFT) + 1) = satsb((int16_t)s->W(1));
1020     r.B((4 << SHIFT) + 2) = satsb((int16_t)s->W(2));
1021     r.B((4 << SHIFT) + 3) = satsb((int16_t)s->W(3));
1022 #if SHIFT == 1
1023     r.B(12) = satsb((int16_t)s->W(4));
1024     r.B(13) = satsb((int16_t)s->W(5));
1025     r.B(14) = satsb((int16_t)s->W(6));
1026     r.B(15) = satsb((int16_t)s->W(7));
1027 #endif
1028     *d = r;
1029 }
1030
1031 void glue(helper_packuswb, SUFFIX) (Reg *d, Reg *s)
1032 {
1033     Reg r;
1034
1035     r.B(0) = satub((int16_t)d->W(0));
1036     r.B(1) = satub((int16_t)d->W(1));
1037     r.B(2) = satub((int16_t)d->W(2));
1038     r.B(3) = satub((int16_t)d->W(3));
1039 #if SHIFT == 1
1040     r.B(4) = satub((int16_t)d->W(4));
1041     r.B(5) = satub((int16_t)d->W(5));
1042     r.B(6) = satub((int16_t)d->W(6));
1043     r.B(7) = satub((int16_t)d->W(7));
1044 #endif
1045     r.B((4 << SHIFT) + 0) = satub((int16_t)s->W(0));
1046     r.B((4 << SHIFT) + 1) = satub((int16_t)s->W(1));
1047     r.B((4 << SHIFT) + 2) = satub((int16_t)s->W(2));
1048     r.B((4 << SHIFT) + 3) = satub((int16_t)s->W(3));
1049 #if SHIFT == 1
1050     r.B(12) = satub((int16_t)s->W(4));
1051     r.B(13) = satub((int16_t)s->W(5));
1052     r.B(14) = satub((int16_t)s->W(6));
1053     r.B(15) = satub((int16_t)s->W(7));
1054 #endif
1055     *d = r;
1056 }
1057
1058 void glue(helper_packssdw, SUFFIX) (Reg *d, Reg *s)
1059 {
1060     Reg r;
1061
1062     r.W(0) = satsw(d->L(0));
1063     r.W(1) = satsw(d->L(1));
1064 #if SHIFT == 1
1065     r.W(2) = satsw(d->L(2));
1066     r.W(3) = satsw(d->L(3));
1067 #endif
1068     r.W((2 << SHIFT) + 0) = satsw(s->L(0));
1069     r.W((2 << SHIFT) + 1) = satsw(s->L(1));
1070 #if SHIFT == 1
1071     r.W(6) = satsw(s->L(2));
1072     r.W(7) = satsw(s->L(3));
1073 #endif
1074     *d = r;
1075 }
1076
1077 #define UNPCK_OP(base_name, base)                               \
1078                                                                 \
1079 void glue(helper_punpck ## base_name ## bw, SUFFIX) (Reg *d, Reg *s)   \
1080 {                                                               \
1081     Reg r;                                              \
1082                                                                 \
1083     r.B(0) = d->B((base << (SHIFT + 2)) + 0);                   \
1084     r.B(1) = s->B((base << (SHIFT + 2)) + 0);                   \
1085     r.B(2) = d->B((base << (SHIFT + 2)) + 1);                   \
1086     r.B(3) = s->B((base << (SHIFT + 2)) + 1);                   \
1087     r.B(4) = d->B((base << (SHIFT + 2)) + 2);                   \
1088     r.B(5) = s->B((base << (SHIFT + 2)) + 2);                   \
1089     r.B(6) = d->B((base << (SHIFT + 2)) + 3);                   \
1090     r.B(7) = s->B((base << (SHIFT + 2)) + 3);                   \
1091 XMM_ONLY(                                                       \
1092     r.B(8) = d->B((base << (SHIFT + 2)) + 4);                   \
1093     r.B(9) = s->B((base << (SHIFT + 2)) + 4);                   \
1094     r.B(10) = d->B((base << (SHIFT + 2)) + 5);                  \
1095     r.B(11) = s->B((base << (SHIFT + 2)) + 5);                  \
1096     r.B(12) = d->B((base << (SHIFT + 2)) + 6);                  \
1097     r.B(13) = s->B((base << (SHIFT + 2)) + 6);                  \
1098     r.B(14) = d->B((base << (SHIFT + 2)) + 7);                  \
1099     r.B(15) = s->B((base << (SHIFT + 2)) + 7);                  \
1100 )                                                               \
1101     *d = r;                                                     \
1102 }                                                               \
1103                                                                 \
1104 void glue(helper_punpck ## base_name ## wd, SUFFIX) (Reg *d, Reg *s)   \
1105 {                                                               \
1106     Reg r;                                              \
1107                                                                 \
1108     r.W(0) = d->W((base << (SHIFT + 1)) + 0);                   \
1109     r.W(1) = s->W((base << (SHIFT + 1)) + 0);                   \
1110     r.W(2) = d->W((base << (SHIFT + 1)) + 1);                   \
1111     r.W(3) = s->W((base << (SHIFT + 1)) + 1);                   \
1112 XMM_ONLY(                                                       \
1113     r.W(4) = d->W((base << (SHIFT + 1)) + 2);                   \
1114     r.W(5) = s->W((base << (SHIFT + 1)) + 2);                   \
1115     r.W(6) = d->W((base << (SHIFT + 1)) + 3);                   \
1116     r.W(7) = s->W((base << (SHIFT + 1)) + 3);                   \
1117 )                                                               \
1118     *d = r;                                                     \
1119 }                                                               \
1120                                                                 \
1121 void glue(helper_punpck ## base_name ## dq, SUFFIX) (Reg *d, Reg *s)   \
1122 {                                                               \
1123     Reg r;                                              \
1124                                                                 \
1125     r.L(0) = d->L((base << SHIFT) + 0);                         \
1126     r.L(1) = s->L((base << SHIFT) + 0);                         \
1127 XMM_ONLY(                                                       \
1128     r.L(2) = d->L((base << SHIFT) + 1);                         \
1129     r.L(3) = s->L((base << SHIFT) + 1);                         \
1130 )                                                               \
1131     *d = r;                                                     \
1132 }                                                               \
1133                                                                 \
1134 XMM_ONLY(                                                       \
1135 void glue(helper_punpck ## base_name ## qdq, SUFFIX) (Reg *d, Reg *s)  \
1136 {                                                               \
1137     Reg r;                                              \
1138                                                                 \
1139     r.Q(0) = d->Q(base);                                        \
1140     r.Q(1) = s->Q(base);                                        \
1141     *d = r;                                                     \
1142 }                                                               \
1143 )
1144
1145 UNPCK_OP(l, 0)
1146 UNPCK_OP(h, 1)
1147
1148 /* 3DNow! float ops */
1149 #if SHIFT == 0
1150 void helper_pi2fd(MMXReg *d, MMXReg *s)
1151 {
1152     d->MMX_S(0) = int32_to_float32(s->MMX_L(0), &env->mmx_status);
1153     d->MMX_S(1) = int32_to_float32(s->MMX_L(1), &env->mmx_status);
1154 }
1155
1156 void helper_pi2fw(MMXReg *d, MMXReg *s)
1157 {
1158     d->MMX_S(0) = int32_to_float32((int16_t)s->MMX_W(0), &env->mmx_status);
1159     d->MMX_S(1) = int32_to_float32((int16_t)s->MMX_W(2), &env->mmx_status);
1160 }
1161
1162 void helper_pf2id(MMXReg *d, MMXReg *s)
1163 {
1164     d->MMX_L(0) = float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status);
1165     d->MMX_L(1) = float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status);
1166 }
1167
1168 void helper_pf2iw(MMXReg *d, MMXReg *s)
1169 {
1170     d->MMX_L(0) = satsw(float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status));
1171     d->MMX_L(1) = satsw(float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status));
1172 }
1173
1174 void helper_pfacc(MMXReg *d, MMXReg *s)
1175 {
1176     MMXReg r;
1177     r.MMX_S(0) = float32_add(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1178     r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1179     *d = r;
1180 }
1181
1182 void helper_pfadd(MMXReg *d, MMXReg *s)
1183 {
1184     d->MMX_S(0) = float32_add(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1185     d->MMX_S(1) = float32_add(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1186 }
1187
1188 void helper_pfcmpeq(MMXReg *d, MMXReg *s)
1189 {
1190     d->MMX_L(0) = float32_eq(d->MMX_S(0), s->MMX_S(0), &env->mmx_status) ? -1 : 0;
1191     d->MMX_L(1) = float32_eq(d->MMX_S(1), s->MMX_S(1), &env->mmx_status) ? -1 : 0;
1192 }
1193
1194 void helper_pfcmpge(MMXReg *d, MMXReg *s)
1195 {
1196     d->MMX_L(0) = float32_le(s->MMX_S(0), d->MMX_S(0), &env->mmx_status) ? -1 : 0;
1197     d->MMX_L(1) = float32_le(s->MMX_S(1), d->MMX_S(1), &env->mmx_status) ? -1 : 0;
1198 }
1199
1200 void helper_pfcmpgt(MMXReg *d, MMXReg *s)
1201 {
1202     d->MMX_L(0) = float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status) ? -1 : 0;
1203     d->MMX_L(1) = float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status) ? -1 : 0;
1204 }
1205
1206 void helper_pfmax(MMXReg *d, MMXReg *s)
1207 {
1208     if (float32_lt(d->MMX_S(0), s->MMX_S(0), &env->mmx_status))
1209         d->MMX_S(0) = s->MMX_S(0);
1210     if (float32_lt(d->MMX_S(1), s->MMX_S(1), &env->mmx_status))
1211         d->MMX_S(1) = s->MMX_S(1);
1212 }
1213
1214 void helper_pfmin(MMXReg *d, MMXReg *s)
1215 {
1216     if (float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status))
1217         d->MMX_S(0) = s->MMX_S(0);
1218     if (float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status))
1219         d->MMX_S(1) = s->MMX_S(1);
1220 }
1221
1222 void helper_pfmul(MMXReg *d, MMXReg *s)
1223 {
1224     d->MMX_S(0) = float32_mul(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1225     d->MMX_S(1) = float32_mul(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1226 }
1227
1228 void helper_pfnacc(MMXReg *d, MMXReg *s)
1229 {
1230     MMXReg r;
1231     r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1232     r.MMX_S(1) = float32_sub(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1233     *d = r;
1234 }
1235
1236 void helper_pfpnacc(MMXReg *d, MMXReg *s)
1237 {
1238     MMXReg r;
1239     r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1240     r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1241     *d = r;
1242 }
1243
1244 void helper_pfrcp(MMXReg *d, MMXReg *s)
1245 {
1246     d->MMX_S(0) = approx_rcp(s->MMX_S(0));
1247     d->MMX_S(1) = d->MMX_S(0);
1248 }
1249
1250 void helper_pfrsqrt(MMXReg *d, MMXReg *s)
1251 {
1252     d->MMX_L(1) = s->MMX_L(0) & 0x7fffffff;
1253     d->MMX_S(1) = approx_rsqrt(d->MMX_S(1));
1254     d->MMX_L(1) |= s->MMX_L(0) & 0x80000000;
1255     d->MMX_L(0) = d->MMX_L(1);
1256 }
1257
1258 void helper_pfsub(MMXReg *d, MMXReg *s)
1259 {
1260     d->MMX_S(0) = float32_sub(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1261     d->MMX_S(1) = float32_sub(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1262 }
1263
1264 void helper_pfsubr(MMXReg *d, MMXReg *s)
1265 {
1266     d->MMX_S(0) = float32_sub(s->MMX_S(0), d->MMX_S(0), &env->mmx_status);
1267     d->MMX_S(1) = float32_sub(s->MMX_S(1), d->MMX_S(1), &env->mmx_status);
1268 }
1269
1270 void helper_pswapd(MMXReg *d, MMXReg *s)
1271 {
1272     MMXReg r;
1273     r.MMX_L(0) = s->MMX_L(1);
1274     r.MMX_L(1) = s->MMX_L(0);
1275     *d = r;
1276 }
1277 #endif
1278
1279 /* SSSE3 op helpers */
1280 void glue(helper_pshufb, SUFFIX) (Reg *d, Reg *s)
1281 {
1282     int i;
1283     Reg r;
1284
1285     for (i = 0; i < (8 << SHIFT); i++)
1286         r.B(i) = (s->B(i) & 0x80) ? 0 : (d->B(s->B(i) & ((8 << SHIFT) - 1)));
1287
1288     *d = r;
1289 }
1290
1291 void glue(helper_phaddw, SUFFIX) (Reg *d, Reg *s)
1292 {
1293     d->W(0) = (int16_t)d->W(0) + (int16_t)d->W(1);
1294     d->W(1) = (int16_t)d->W(2) + (int16_t)d->W(3);
1295     XMM_ONLY(d->W(2) = (int16_t)d->W(4) + (int16_t)d->W(5));
1296     XMM_ONLY(d->W(3) = (int16_t)d->W(6) + (int16_t)d->W(7));
1297     d->W((2 << SHIFT) + 0) = (int16_t)s->W(0) + (int16_t)s->W(1);
1298     d->W((2 << SHIFT) + 1) = (int16_t)s->W(2) + (int16_t)s->W(3);
1299     XMM_ONLY(d->W(6) = (int16_t)s->W(4) + (int16_t)s->W(5));
1300     XMM_ONLY(d->W(7) = (int16_t)s->W(6) + (int16_t)s->W(7));
1301 }
1302
1303 void glue(helper_phaddd, SUFFIX) (Reg *d, Reg *s)
1304 {
1305     d->L(0) = (int32_t)d->L(0) + (int32_t)d->L(1);
1306     XMM_ONLY(d->L(1) = (int32_t)d->L(2) + (int32_t)d->L(3));
1307     d->L((1 << SHIFT) + 0) = (int32_t)s->L(0) + (int32_t)s->L(1);
1308     XMM_ONLY(d->L(3) = (int32_t)s->L(2) + (int32_t)s->L(3));
1309 }
1310
1311 void glue(helper_phaddsw, SUFFIX) (Reg *d, Reg *s)
1312 {
1313     d->W(0) = satsw((int16_t)d->W(0) + (int16_t)d->W(1));
1314     d->W(1) = satsw((int16_t)d->W(2) + (int16_t)d->W(3));
1315     XMM_ONLY(d->W(2) = satsw((int16_t)d->W(4) + (int16_t)d->W(5)));
1316     XMM_ONLY(d->W(3) = satsw((int16_t)d->W(6) + (int16_t)d->W(7)));
1317     d->W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) + (int16_t)s->W(1));
1318     d->W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) + (int16_t)s->W(3));
1319     XMM_ONLY(d->W(6) = satsw((int16_t)s->W(4) + (int16_t)s->W(5)));
1320     XMM_ONLY(d->W(7) = satsw((int16_t)s->W(6) + (int16_t)s->W(7)));
1321 }
1322
1323 void glue(helper_pmaddubsw, SUFFIX) (Reg *d, Reg *s)
1324 {
1325     d->W(0) = satsw((int8_t)s->B( 0) * (uint8_t)d->B( 0) +
1326                     (int8_t)s->B( 1) * (uint8_t)d->B( 1));
1327     d->W(1) = satsw((int8_t)s->B( 2) * (uint8_t)d->B( 2) +
1328                     (int8_t)s->B( 3) * (uint8_t)d->B( 3));
1329     d->W(2) = satsw((int8_t)s->B( 4) * (uint8_t)d->B( 4) +
1330                     (int8_t)s->B( 5) * (uint8_t)d->B( 5));
1331     d->W(3) = satsw((int8_t)s->B( 6) * (uint8_t)d->B( 6) +
1332                     (int8_t)s->B( 7) * (uint8_t)d->B( 7));
1333 #if SHIFT == 1
1334     d->W(4) = satsw((int8_t)s->B( 8) * (uint8_t)d->B( 8) +
1335                     (int8_t)s->B( 9) * (uint8_t)d->B( 9));
1336     d->W(5) = satsw((int8_t)s->B(10) * (uint8_t)d->B(10) +
1337                     (int8_t)s->B(11) * (uint8_t)d->B(11));
1338     d->W(6) = satsw((int8_t)s->B(12) * (uint8_t)d->B(12) +
1339                     (int8_t)s->B(13) * (uint8_t)d->B(13));
1340     d->W(7) = satsw((int8_t)s->B(14) * (uint8_t)d->B(14) +
1341                     (int8_t)s->B(15) * (uint8_t)d->B(15));
1342 #endif
1343 }
1344
1345 void glue(helper_phsubw, SUFFIX) (Reg *d, Reg *s)
1346 {
1347     d->W(0) = (int16_t)d->W(0) - (int16_t)d->W(1);
1348     d->W(1) = (int16_t)d->W(2) - (int16_t)d->W(3);
1349     XMM_ONLY(d->W(2) = (int16_t)d->W(4) - (int16_t)d->W(5));
1350     XMM_ONLY(d->W(3) = (int16_t)d->W(6) - (int16_t)d->W(7));
1351     d->W((2 << SHIFT) + 0) = (int16_t)s->W(0) - (int16_t)s->W(1);
1352     d->W((2 << SHIFT) + 1) = (int16_t)s->W(2) - (int16_t)s->W(3);
1353     XMM_ONLY(d->W(6) = (int16_t)s->W(4) - (int16_t)s->W(5));
1354     XMM_ONLY(d->W(7) = (int16_t)s->W(6) - (int16_t)s->W(7));
1355 }
1356
1357 void glue(helper_phsubd, SUFFIX) (Reg *d, Reg *s)
1358 {
1359     d->L(0) = (int32_t)d->L(0) - (int32_t)d->L(1);
1360     XMM_ONLY(d->L(1) = (int32_t)d->L(2) - (int32_t)d->L(3));
1361     d->L((1 << SHIFT) + 0) = (int32_t)s->L(0) - (int32_t)s->L(1);
1362     XMM_ONLY(d->L(3) = (int32_t)s->L(2) - (int32_t)s->L(3));
1363 }
1364
1365 void glue(helper_phsubsw, SUFFIX) (Reg *d, Reg *s)
1366 {
1367     d->W(0) = satsw((int16_t)d->W(0) - (int16_t)d->W(1));
1368     d->W(1) = satsw((int16_t)d->W(2) - (int16_t)d->W(3));
1369     XMM_ONLY(d->W(2) = satsw((int16_t)d->W(4) - (int16_t)d->W(5)));
1370     XMM_ONLY(d->W(3) = satsw((int16_t)d->W(6) - (int16_t)d->W(7)));
1371     d->W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) - (int16_t)s->W(1));
1372     d->W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) - (int16_t)s->W(3));
1373     XMM_ONLY(d->W(6) = satsw((int16_t)s->W(4) - (int16_t)s->W(5)));
1374     XMM_ONLY(d->W(7) = satsw((int16_t)s->W(6) - (int16_t)s->W(7)));
1375 }
1376
1377 #define FABSB(_, x) x > INT8_MAX  ? -(int8_t ) x : x
1378 #define FABSW(_, x) x > INT16_MAX ? -(int16_t) x : x
1379 #define FABSL(_, x) x > INT32_MAX ? -(int32_t) x : x
1380 SSE_HELPER_B(helper_pabsb, FABSB)
1381 SSE_HELPER_W(helper_pabsw, FABSW)
1382 SSE_HELPER_L(helper_pabsd, FABSL)
1383
1384 #define FMULHRSW(d, s) ((int16_t) d * (int16_t) s + 0x4000) >> 15
1385 SSE_HELPER_W(helper_pmulhrsw, FMULHRSW)
1386
1387 #define FSIGNB(d, s) s <= INT8_MAX  ? s ? d : 0 : -(int8_t ) d
1388 #define FSIGNW(d, s) s <= INT16_MAX ? s ? d : 0 : -(int16_t) d
1389 #define FSIGNL(d, s) s <= INT32_MAX ? s ? d : 0 : -(int32_t) d
1390 SSE_HELPER_B(helper_psignb, FSIGNB)
1391 SSE_HELPER_W(helper_psignw, FSIGNW)
1392 SSE_HELPER_L(helper_psignd, FSIGNL)
1393
1394 void glue(helper_palignr, SUFFIX) (Reg *d, Reg *s, int32_t shift)
1395 {
1396     Reg r;
1397
1398     /* XXX could be checked during translation */
1399     if (shift >= (16 << SHIFT)) {
1400         r.Q(0) = 0;
1401         XMM_ONLY(r.Q(1) = 0);
1402     } else {
1403         shift <<= 3;
1404 #define SHR(v, i) (i < 64 && i > -64 ? i > 0 ? v >> (i) : (v << -(i)) : 0)
1405 #if SHIFT == 0
1406         r.Q(0) = SHR(s->Q(0), shift -   0) |
1407                  SHR(d->Q(0), shift -  64);
1408 #else
1409         r.Q(0) = SHR(s->Q(0), shift -   0) |
1410                  SHR(s->Q(1), shift -  64) |
1411                  SHR(d->Q(0), shift - 128) |
1412                  SHR(d->Q(1), shift - 192);
1413         r.Q(1) = SHR(s->Q(0), shift +  64) |
1414                  SHR(s->Q(1), shift -   0) |
1415                  SHR(d->Q(0), shift -  64) |
1416                  SHR(d->Q(1), shift - 128);
1417 #endif
1418 #undef SHR
1419     }
1420
1421     *d = r;
1422 }
1423
1424 #define XMM0 env->xmm_regs[0]
1425
1426 #if SHIFT == 1
1427 #define SSE_HELPER_V(name, elem, num, F)\
1428 void glue(name, SUFFIX) (Reg *d, Reg *s)\
1429 {\
1430     d->elem(0) = F(d->elem(0), s->elem(0), XMM0.elem(0));\
1431     d->elem(1) = F(d->elem(1), s->elem(1), XMM0.elem(1));\
1432     if (num > 2) {\
1433         d->elem(2) = F(d->elem(2), s->elem(2), XMM0.elem(2));\
1434         d->elem(3) = F(d->elem(3), s->elem(3), XMM0.elem(3));\
1435         if (num > 4) {\
1436             d->elem(4) = F(d->elem(4), s->elem(4), XMM0.elem(4));\
1437             d->elem(5) = F(d->elem(5), s->elem(5), XMM0.elem(5));\
1438             d->elem(6) = F(d->elem(6), s->elem(6), XMM0.elem(6));\
1439             d->elem(7) = F(d->elem(7), s->elem(7), XMM0.elem(7));\
1440             if (num > 8) {\
1441                 d->elem(8) = F(d->elem(8), s->elem(8), XMM0.elem(8));\
1442                 d->elem(9) = F(d->elem(9), s->elem(9), XMM0.elem(9));\
1443                 d->elem(10) = F(d->elem(10), s->elem(10), XMM0.elem(10));\
1444                 d->elem(11) = F(d->elem(11), s->elem(11), XMM0.elem(11));\
1445                 d->elem(12) = F(d->elem(12), s->elem(12), XMM0.elem(12));\
1446                 d->elem(13) = F(d->elem(13), s->elem(13), XMM0.elem(13));\
1447                 d->elem(14) = F(d->elem(14), s->elem(14), XMM0.elem(14));\
1448                 d->elem(15) = F(d->elem(15), s->elem(15), XMM0.elem(15));\
1449             }\
1450         }\
1451     }\
1452 }
1453
1454 #define SSE_HELPER_I(name, elem, num, F)\
1455 void glue(name, SUFFIX) (Reg *d, Reg *s, uint32_t imm)\
1456 {\
1457     d->elem(0) = F(d->elem(0), s->elem(0), ((imm >> 0) & 1));\
1458     d->elem(1) = F(d->elem(1), s->elem(1), ((imm >> 1) & 1));\
1459     if (num > 2) {\
1460         d->elem(2) = F(d->elem(2), s->elem(2), ((imm >> 2) & 1));\
1461         d->elem(3) = F(d->elem(3), s->elem(3), ((imm >> 3) & 1));\
1462         if (num > 4) {\
1463             d->elem(4) = F(d->elem(4), s->elem(4), ((imm >> 4) & 1));\
1464             d->elem(5) = F(d->elem(5), s->elem(5), ((imm >> 5) & 1));\
1465             d->elem(6) = F(d->elem(6), s->elem(6), ((imm >> 6) & 1));\
1466             d->elem(7) = F(d->elem(7), s->elem(7), ((imm >> 7) & 1));\
1467             if (num > 8) {\
1468                 d->elem(8) = F(d->elem(8), s->elem(8), ((imm >> 8) & 1));\
1469                 d->elem(9) = F(d->elem(9), s->elem(9), ((imm >> 9) & 1));\
1470                 d->elem(10) = F(d->elem(10), s->elem(10), ((imm >> 10) & 1));\
1471                 d->elem(11) = F(d->elem(11), s->elem(11), ((imm >> 11) & 1));\
1472                 d->elem(12) = F(d->elem(12), s->elem(12), ((imm >> 12) & 1));\
1473                 d->elem(13) = F(d->elem(13), s->elem(13), ((imm >> 13) & 1));\
1474                 d->elem(14) = F(d->elem(14), s->elem(14), ((imm >> 14) & 1));\
1475                 d->elem(15) = F(d->elem(15), s->elem(15), ((imm >> 15) & 1));\
1476             }\
1477         }\
1478     }\
1479 }
1480
1481 /* SSE4.1 op helpers */
1482 #define FBLENDVB(d, s, m) (m & 0x80) ? s : d
1483 #define FBLENDVPS(d, s, m) (m & 0x80000000) ? s : d
1484 #define FBLENDVPD(d, s, m) (m & 0x8000000000000000LL) ? s : d
1485 SSE_HELPER_V(helper_pblendvb, B, 16, FBLENDVB)
1486 SSE_HELPER_V(helper_blendvps, L, 4, FBLENDVPS)
1487 SSE_HELPER_V(helper_blendvpd, Q, 2, FBLENDVPD)
1488
1489 void glue(helper_ptest, SUFFIX) (Reg *d, Reg *s)
1490 {
1491     uint64_t zf = (s->Q(0) &  d->Q(0)) | (s->Q(1) &  d->Q(1));
1492     uint64_t cf = (s->Q(0) & ~d->Q(0)) | (s->Q(1) & ~d->Q(1));
1493
1494     CC_SRC = (zf ? 0 : CC_Z) | (cf ? 0 : CC_C);
1495 }
1496
1497 #define SSE_HELPER_F(name, elem, num, F)\
1498 void glue(name, SUFFIX) (Reg *d, Reg *s)\
1499 {\
1500     d->elem(0) = F(0);\
1501     d->elem(1) = F(1);\
1502     d->elem(2) = F(2);\
1503     d->elem(3) = F(3);\
1504     if (num > 3) {\
1505         d->elem(4) = F(4);\
1506         d->elem(5) = F(5);\
1507         if (num > 5) {\
1508             d->elem(6) = F(6);\
1509             d->elem(7) = F(7);\
1510         }\
1511     }\
1512 }
1513
1514 SSE_HELPER_F(helper_pmovsxbw, W, 8, (int8_t) s->B)
1515 SSE_HELPER_F(helper_pmovsxbd, L, 4, (int8_t) s->B)
1516 SSE_HELPER_F(helper_pmovsxbq, Q, 2, (int8_t) s->B)
1517 SSE_HELPER_F(helper_pmovsxwd, L, 4, (int16_t) s->W)
1518 SSE_HELPER_F(helper_pmovsxwq, Q, 2, (int16_t) s->W)
1519 SSE_HELPER_F(helper_pmovsxdq, Q, 2, (int32_t) s->L)
1520 SSE_HELPER_F(helper_pmovzxbw, W, 8, s->B)
1521 SSE_HELPER_F(helper_pmovzxbd, L, 4, s->B)
1522 SSE_HELPER_F(helper_pmovzxbq, Q, 2, s->B)
1523 SSE_HELPER_F(helper_pmovzxwd, L, 4, s->W)
1524 SSE_HELPER_F(helper_pmovzxwq, Q, 2, s->W)
1525 SSE_HELPER_F(helper_pmovzxdq, Q, 2, s->L)
1526
1527 void glue(helper_pmuldq, SUFFIX) (Reg *d, Reg *s)
1528 {
1529     d->Q(0) = (int64_t) (int32_t) d->L(0) * (int32_t) s->L(0);
1530     d->Q(1) = (int64_t) (int32_t) d->L(2) * (int32_t) s->L(2);
1531 }
1532
1533 #define FCMPEQQ(d, s) d == s ? -1 : 0
1534 SSE_HELPER_Q(helper_pcmpeqq, FCMPEQQ)
1535
1536 void glue(helper_packusdw, SUFFIX) (Reg *d, Reg *s)
1537 {
1538     d->W(0) = satuw((int32_t) d->L(0));
1539     d->W(1) = satuw((int32_t) d->L(1));
1540     d->W(2) = satuw((int32_t) d->L(2));
1541     d->W(3) = satuw((int32_t) d->L(3));
1542     d->W(4) = satuw((int32_t) s->L(0));
1543     d->W(5) = satuw((int32_t) s->L(1));
1544     d->W(6) = satuw((int32_t) s->L(2));
1545     d->W(7) = satuw((int32_t) s->L(3));
1546 }
1547
1548 #define FMINSB(d, s) MIN((int8_t) d, (int8_t) s)
1549 #define FMINSD(d, s) MIN((int32_t) d, (int32_t) s)
1550 #define FMAXSB(d, s) MAX((int8_t) d, (int8_t) s)
1551 #define FMAXSD(d, s) MAX((int32_t) d, (int32_t) s)
1552 SSE_HELPER_B(helper_pminsb, FMINSB)
1553 SSE_HELPER_L(helper_pminsd, FMINSD)
1554 SSE_HELPER_W(helper_pminuw, MIN)
1555 SSE_HELPER_L(helper_pminud, MIN)
1556 SSE_HELPER_B(helper_pmaxsb, FMAXSB)
1557 SSE_HELPER_L(helper_pmaxsd, FMAXSD)
1558 SSE_HELPER_W(helper_pmaxuw, MAX)
1559 SSE_HELPER_L(helper_pmaxud, MAX)
1560
1561 #define FMULLD(d, s) (int32_t) d * (int32_t) s
1562 SSE_HELPER_L(helper_pmulld, FMULLD)
1563
1564 void glue(helper_phminposuw, SUFFIX) (Reg *d, Reg *s)
1565 {
1566     int idx = 0;
1567
1568     if (s->W(1) < s->W(idx))
1569         idx = 1;
1570     if (s->W(2) < s->W(idx))
1571         idx = 2;
1572     if (s->W(3) < s->W(idx))
1573         idx = 3;
1574     if (s->W(4) < s->W(idx))
1575         idx = 4;
1576     if (s->W(5) < s->W(idx))
1577         idx = 5;
1578     if (s->W(6) < s->W(idx))
1579         idx = 6;
1580     if (s->W(7) < s->W(idx))
1581         idx = 7;
1582
1583     d->Q(1) = 0;
1584     d->L(1) = 0;
1585     d->W(1) = idx;
1586     d->W(0) = s->W(idx);
1587 }
1588
1589 void glue(helper_roundps, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1590 {
1591     signed char prev_rounding_mode;
1592
1593     prev_rounding_mode = env->sse_status.float_rounding_mode;
1594     if (!(mode & (1 << 2)))
1595         switch (mode & 3) {
1596         case 0:
1597             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1598             break;
1599         case 1:
1600             set_float_rounding_mode(float_round_down, &env->sse_status);
1601             break;
1602         case 2:
1603             set_float_rounding_mode(float_round_up, &env->sse_status);
1604             break;
1605         case 3:
1606             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1607             break;
1608         }
1609
1610     d->L(0) = float64_round_to_int(s->L(0), &env->sse_status);
1611     d->L(1) = float64_round_to_int(s->L(1), &env->sse_status);
1612     d->L(2) = float64_round_to_int(s->L(2), &env->sse_status);
1613     d->L(3) = float64_round_to_int(s->L(3), &env->sse_status);
1614
1615 #if 0 /* TODO */
1616     if (mode & (1 << 3))
1617         set_float_exception_flags(
1618                         get_float_exception_flags(&env->sse_status) &
1619                         ~float_flag_inexact,
1620                         &env->sse_status);
1621 #endif
1622     env->sse_status.float_rounding_mode = prev_rounding_mode;
1623 }
1624
1625 void glue(helper_roundpd, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1626 {
1627     signed char prev_rounding_mode;
1628
1629     prev_rounding_mode = env->sse_status.float_rounding_mode;
1630     if (!(mode & (1 << 2)))
1631         switch (mode & 3) {
1632         case 0:
1633             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1634             break;
1635         case 1:
1636             set_float_rounding_mode(float_round_down, &env->sse_status);
1637             break;
1638         case 2:
1639             set_float_rounding_mode(float_round_up, &env->sse_status);
1640             break;
1641         case 3:
1642             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1643             break;
1644         }
1645
1646     d->Q(0) = float64_round_to_int(s->Q(0), &env->sse_status);
1647     d->Q(1) = float64_round_to_int(s->Q(1), &env->sse_status);
1648
1649 #if 0 /* TODO */
1650     if (mode & (1 << 3))
1651         set_float_exception_flags(
1652                         get_float_exception_flags(&env->sse_status) &
1653                         ~float_flag_inexact,
1654                         &env->sse_status);
1655 #endif
1656     env->sse_status.float_rounding_mode = prev_rounding_mode;
1657 }
1658
1659 void glue(helper_roundss, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1660 {
1661     signed char prev_rounding_mode;
1662
1663     prev_rounding_mode = env->sse_status.float_rounding_mode;
1664     if (!(mode & (1 << 2)))
1665         switch (mode & 3) {
1666         case 0:
1667             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1668             break;
1669         case 1:
1670             set_float_rounding_mode(float_round_down, &env->sse_status);
1671             break;
1672         case 2:
1673             set_float_rounding_mode(float_round_up, &env->sse_status);
1674             break;
1675         case 3:
1676             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1677             break;
1678         }
1679
1680     d->L(0) = float64_round_to_int(s->L(0), &env->sse_status);
1681
1682 #if 0 /* TODO */
1683     if (mode & (1 << 3))
1684         set_float_exception_flags(
1685                         get_float_exception_flags(&env->sse_status) &
1686                         ~float_flag_inexact,
1687                         &env->sse_status);
1688 #endif
1689     env->sse_status.float_rounding_mode = prev_rounding_mode;
1690 }
1691
1692 void glue(helper_roundsd, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1693 {
1694     signed char prev_rounding_mode;
1695
1696     prev_rounding_mode = env->sse_status.float_rounding_mode;
1697     if (!(mode & (1 << 2)))
1698         switch (mode & 3) {
1699         case 0:
1700             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1701             break;
1702         case 1:
1703             set_float_rounding_mode(float_round_down, &env->sse_status);
1704             break;
1705         case 2:
1706             set_float_rounding_mode(float_round_up, &env->sse_status);
1707             break;
1708         case 3:
1709             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1710             break;
1711         }
1712
1713     d->Q(0) = float64_round_to_int(s->Q(0), &env->sse_status);
1714
1715 #if 0 /* TODO */
1716     if (mode & (1 << 3))
1717         set_float_exception_flags(
1718                         get_float_exception_flags(&env->sse_status) &
1719                         ~float_flag_inexact,
1720                         &env->sse_status);
1721 #endif
1722     env->sse_status.float_rounding_mode = prev_rounding_mode;
1723 }
1724
1725 #define FBLENDP(d, s, m) m ? s : d
1726 SSE_HELPER_I(helper_blendps, L, 4, FBLENDP)
1727 SSE_HELPER_I(helper_blendpd, Q, 2, FBLENDP)
1728 SSE_HELPER_I(helper_pblendw, W, 8, FBLENDP)
1729
1730 void glue(helper_dpps, SUFFIX) (Reg *d, Reg *s, uint32_t mask)
1731 {
1732     float32 iresult = 0 /*float32_zero*/;
1733
1734     if (mask & (1 << 4))
1735         iresult = float32_add(iresult,
1736                         float32_mul(d->L(0), s->L(0), &env->sse_status),
1737                         &env->sse_status);
1738     if (mask & (1 << 5))
1739         iresult = float32_add(iresult,
1740                         float32_mul(d->L(1), s->L(1), &env->sse_status),
1741                         &env->sse_status);
1742     if (mask & (1 << 6))
1743         iresult = float32_add(iresult,
1744                         float32_mul(d->L(2), s->L(2), &env->sse_status),
1745                         &env->sse_status);
1746     if (mask & (1 << 7))
1747         iresult = float32_add(iresult,
1748                         float32_mul(d->L(3), s->L(3), &env->sse_status),
1749                         &env->sse_status);
1750     d->L(0) = (mask & (1 << 0)) ? iresult : 0 /*float32_zero*/;
1751     d->L(1) = (mask & (1 << 1)) ? iresult : 0 /*float32_zero*/;
1752     d->L(2) = (mask & (1 << 2)) ? iresult : 0 /*float32_zero*/;
1753     d->L(3) = (mask & (1 << 3)) ? iresult : 0 /*float32_zero*/;
1754 }
1755
1756 void glue(helper_dppd, SUFFIX) (Reg *d, Reg *s, uint32_t mask)
1757 {
1758     float64 iresult = 0 /*float64_zero*/;
1759
1760     if (mask & (1 << 4))
1761         iresult = float64_add(iresult,
1762                         float64_mul(d->Q(0), s->Q(0), &env->sse_status),
1763                         &env->sse_status);
1764     if (mask & (1 << 5))
1765         iresult = float64_add(iresult,
1766                         float64_mul(d->Q(1), s->Q(1), &env->sse_status),
1767                         &env->sse_status);
1768     d->Q(0) = (mask & (1 << 0)) ? iresult : 0 /*float64_zero*/;
1769     d->Q(1) = (mask & (1 << 1)) ? iresult : 0 /*float64_zero*/;
1770 }
1771
1772 void glue(helper_mpsadbw, SUFFIX) (Reg *d, Reg *s, uint32_t offset)
1773 {
1774     int s0 = (offset & 3) << 2;
1775     int d0 = (offset & 4) << 0;
1776     int i;
1777     Reg r;
1778
1779     for (i = 0; i < 8; i++, d0++) {
1780         r.W(i) = 0;
1781         r.W(i) += abs1(d->B(d0 + 0) - s->B(s0 + 0));
1782         r.W(i) += abs1(d->B(d0 + 1) - s->B(s0 + 1));
1783         r.W(i) += abs1(d->B(d0 + 2) - s->B(s0 + 2));
1784         r.W(i) += abs1(d->B(d0 + 3) - s->B(s0 + 3));
1785     }
1786
1787     *d = r;
1788 }
1789
1790 /* SSE4.2 op helpers */
1791 /* it's unclear whether signed or unsigned */
1792 #define FCMPGTQ(d, s) d > s ? -1 : 0
1793 SSE_HELPER_Q(helper_pcmpgtq, FCMPGTQ)
1794
1795 static inline int pcmp_elen(int reg, uint32_t ctrl)
1796 {
1797     int val;
1798
1799     /* Presence of REX.W is indicated by a bit higher than 7 set */
1800     if (ctrl >> 8)
1801         val = abs1((int64_t) env->regs[reg]);
1802     else
1803         val = abs1((int32_t) env->regs[reg]);
1804
1805     if (ctrl & 1) {
1806         if (val > 8)
1807             return 8;
1808     } else
1809         if (val > 16)
1810             return 16;
1811
1812     return val;
1813 }
1814
1815 static inline int pcmp_ilen(Reg *r, uint8_t ctrl)
1816 {
1817     int val = 0;
1818
1819     if (ctrl & 1) {
1820         while (val < 8 && r->W(val))
1821             val++;
1822     } else
1823         while (val < 16 && r->B(val))
1824             val++;
1825
1826     return val;
1827 }
1828
1829 static inline int pcmp_val(Reg *r, uint8_t ctrl, int i)
1830 {
1831     switch ((ctrl >> 0) & 3) {
1832     case 0:
1833         return r->B(i);
1834     case 1:
1835         return r->W(i);
1836     case 2:
1837         return (int8_t) r->B(i);
1838     case 3:
1839     default:
1840         return (int16_t) r->W(i);
1841     }
1842 }
1843
1844 static inline unsigned pcmpxstrx(Reg *d, Reg *s,
1845                 int8_t ctrl, int valids, int validd)
1846 {
1847     unsigned int res = 0;
1848     int v;
1849     int j, i;
1850     int upper = (ctrl & 1) ? 7 : 15;
1851
1852     valids--;
1853     validd--;
1854
1855     CC_SRC = (valids < upper ? CC_Z : 0) | (validd < upper ? CC_S : 0);
1856
1857     switch ((ctrl >> 2) & 3) {
1858     case 0:
1859         for (j = valids; j >= 0; j--) {
1860             res <<= 1;
1861             v = pcmp_val(s, ctrl, j);
1862             for (i = validd; i >= 0; i--)
1863                 res |= (v == pcmp_val(d, ctrl, i));
1864         }
1865         break;
1866     case 1:
1867         for (j = valids; j >= 0; j--) {
1868             res <<= 1;
1869             v = pcmp_val(s, ctrl, j);
1870             for (i = ((validd - 1) | 1); i >= 0; i -= 2)
1871                 res |= (pcmp_val(d, ctrl, i - 0) <= v &&
1872                         pcmp_val(d, ctrl, i - 1) >= v);
1873         }
1874         break;
1875     case 2:
1876         res = (2 << (upper - MAX(valids, validd))) - 1;
1877         res <<= MAX(valids, validd) - MIN(valids, validd);
1878         for (i = MIN(valids, validd); i >= 0; i--) {
1879             res <<= 1;
1880             v = pcmp_val(s, ctrl, i);
1881             res |= (v == pcmp_val(d, ctrl, i));
1882         }
1883         break;
1884     case 3:
1885         for (j = valids - validd; j >= 0; j--) {
1886             res <<= 1;
1887             res |= 1;
1888             for (i = MIN(upper - j, validd); i >= 0; i--)
1889                 res &= (pcmp_val(s, ctrl, i + j) == pcmp_val(d, ctrl, i));
1890         }
1891         break;
1892     }
1893
1894     switch ((ctrl >> 4) & 3) {
1895     case 1:
1896         res ^= (2 << upper) - 1;
1897         break;
1898     case 3:
1899         res ^= (2 << valids) - 1;
1900         break;
1901     }
1902
1903     if (res)
1904        CC_SRC |= CC_C;
1905     if (res & 1)
1906        CC_SRC |= CC_O;
1907
1908     return res;
1909 }
1910
1911 static inline int rffs1(unsigned int val)
1912 {
1913     int ret = 1, hi;
1914
1915     for (hi = sizeof(val) * 4; hi; hi /= 2)
1916         if (val >> hi) {
1917             val >>= hi;
1918             ret += hi;
1919         }
1920
1921     return ret;
1922 }
1923
1924 static inline int ffs1(unsigned int val)
1925 {
1926     int ret = 1, hi;
1927
1928     for (hi = sizeof(val) * 4; hi; hi /= 2)
1929         if (val << hi) {
1930             val <<= hi;
1931             ret += hi;
1932         }
1933
1934     return ret;
1935 }
1936
1937 void glue(helper_pcmpestri, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1938 {
1939     unsigned int res = pcmpxstrx(d, s, ctrl,
1940                     pcmp_elen(R_EDX, ctrl),
1941                     pcmp_elen(R_EAX, ctrl));
1942
1943     if (res)
1944         env->regs[R_ECX] = ((ctrl & (1 << 6)) ? rffs1 : ffs1)(res) - 1;
1945     else
1946         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
1947 }
1948
1949 void glue(helper_pcmpestrm, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1950 {
1951     int i;
1952     unsigned int res = pcmpxstrx(d, s, ctrl,
1953                     pcmp_elen(R_EDX, ctrl),
1954                     pcmp_elen(R_EAX, ctrl));
1955
1956     if ((ctrl >> 6) & 1) {
1957         if (ctrl & 1)
1958             for (i = 0; i <= 8; i--, res >>= 1)
1959                 d->W(i) = (res & 1) ? ~0 : 0;
1960         else
1961             for (i = 0; i <= 16; i--, res >>= 1)
1962                 d->B(i) = (res & 1) ? ~0 : 0;
1963     } else {
1964         d->Q(1) = 0;
1965         d->Q(0) = res;
1966     }
1967 }
1968
1969 void glue(helper_pcmpistri, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1970 {
1971     unsigned int res = pcmpxstrx(d, s, ctrl,
1972                     pcmp_ilen(s, ctrl),
1973                     pcmp_ilen(d, ctrl));
1974
1975     if (res)
1976         env->regs[R_ECX] = ((ctrl & (1 << 6)) ? rffs1 : ffs1)(res) - 1;
1977     else
1978         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
1979 }
1980
1981 void glue(helper_pcmpistrm, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1982 {
1983     int i;
1984     unsigned int res = pcmpxstrx(d, s, ctrl,
1985                     pcmp_ilen(s, ctrl),
1986                     pcmp_ilen(d, ctrl));
1987
1988     if ((ctrl >> 6) & 1) {
1989         if (ctrl & 1)
1990             for (i = 0; i <= 8; i--, res >>= 1)
1991                 d->W(i) = (res & 1) ? ~0 : 0;
1992         else
1993             for (i = 0; i <= 16; i--, res >>= 1)
1994                 d->B(i) = (res & 1) ? ~0 : 0;
1995     } else {
1996         d->Q(1) = 0;
1997         d->Q(0) = res;
1998     }
1999 }
2000
2001 #define CRCPOLY        0x1edc6f41
2002 #define CRCPOLY_BITREV 0x82f63b78
2003 target_ulong helper_crc32(uint32_t crc1, target_ulong msg, uint32_t len)
2004 {
2005     target_ulong crc = (msg & ((target_ulong) -1 >>
2006                             (TARGET_LONG_BITS - len))) ^ crc1;
2007
2008     while (len--)
2009         crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_BITREV : 0);
2010
2011     return crc;
2012 }
2013
2014 #define POPMASK(i)     ((target_ulong) -1 / ((1LL << (1 << i)) + 1))
2015 #define POPCOUNT(n, i) (n & POPMASK(i)) + ((n >> (1 << i)) & POPMASK(i))
2016 target_ulong helper_popcnt(target_ulong n, uint32_t type)
2017 {
2018     CC_SRC = n ? 0 : CC_Z;
2019
2020     n = POPCOUNT(n, 0);
2021     n = POPCOUNT(n, 1);
2022     n = POPCOUNT(n, 2);
2023     n = POPCOUNT(n, 3);
2024     if (type == 1)
2025         return n & 0xff;
2026
2027     n = POPCOUNT(n, 4);
2028 #ifndef TARGET_X86_64
2029     return n;
2030 #else
2031     if (type == 2)
2032         return n & 0xff;
2033
2034     return POPCOUNT(n, 5);
2035 #endif
2036 }
2037 #endif
2038
2039 #undef SHIFT
2040 #undef XMM_ONLY
2041 #undef Reg
2042 #undef B
2043 #undef W
2044 #undef L
2045 #undef Q
2046 #undef SUFFIX