target-i386/ops_sse.h

   1 /*
   2  *  MMX/SSE/SSE2/PNI support
   3  *
   4  *  Copyright (c) 2005 Fabrice Bellard
   5  *
   6  * This library is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This library is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with this library; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  19  */
  20 #if SHIFT == 0
  21 #define Reg MMXReg
  22 #define XMM_ONLY(x...)
  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(x...) x
  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 OPPROTO glue(op_psrlw, SUFFIX)(void)
  39 {
  40     Reg *d, *s;
  41     int shift;
  42
  43     d = (Reg *)((char *)env + PARAM1);
  44     s = (Reg *)((char *)env + PARAM2);
  45
  46     if (s->Q(0) > 15) {
  47         d->Q(0) = 0;
  48 #if SHIFT == 1
  49         d->Q(1) = 0;
  50 #endif
  51     } else {
  52         shift = s->B(0);
  53         d->W(0) >>= shift;
  54         d->W(1) >>= shift;
  55         d->W(2) >>= shift;
  56         d->W(3) >>= shift;
  57 #if SHIFT == 1
  58         d->W(4) >>= shift;
  59         d->W(5) >>= shift;
  60         d->W(6) >>= shift;
  61         d->W(7) >>= shift;
  62 #endif
  63     }
  64     FORCE_RET();
  65 }
  66
  67 void OPPROTO glue(op_psraw, SUFFIX)(void)
  68 {
  69     Reg *d, *s;
  70     int shift;
  71
  72     d = (Reg *)((char *)env + PARAM1);
  73     s = (Reg *)((char *)env + PARAM2);
  74
  75     if (s->Q(0) > 15) {
  76         shift = 15;
  77     } else {
  78         shift = s->B(0);
  79     }
  80     d->W(0) = (int16_t)d->W(0) >> shift;
  81     d->W(1) = (int16_t)d->W(1) >> shift;
  82     d->W(2) = (int16_t)d->W(2) >> shift;
  83     d->W(3) = (int16_t)d->W(3) >> shift;
  84 #if SHIFT == 1
  85     d->W(4) = (int16_t)d->W(4) >> shift;
  86     d->W(5) = (int16_t)d->W(5) >> shift;
  87     d->W(6) = (int16_t)d->W(6) >> shift;
  88     d->W(7) = (int16_t)d->W(7) >> shift;
  89 #endif
  90 }
  91
  92 void OPPROTO glue(op_psllw, SUFFIX)(void)
  93 {
  94     Reg *d, *s;
  95     int shift;
  96
  97     d = (Reg *)((char *)env + PARAM1);
  98     s = (Reg *)((char *)env + PARAM2);
  99
 100     if (s->Q(0) > 15) {
 101         d->Q(0) = 0;
 102 #if SHIFT == 1
 103         d->Q(1) = 0;
 104 #endif
 105     } else {
 106         shift = s->B(0);
 107         d->W(0) <<= shift;
 108         d->W(1) <<= shift;
 109         d->W(2) <<= shift;
 110         d->W(3) <<= shift;
 111 #if SHIFT == 1
 112         d->W(4) <<= shift;
 113         d->W(5) <<= shift;
 114         d->W(6) <<= shift;
 115         d->W(7) <<= shift;
 116 #endif
 117     }
 118     FORCE_RET();
 119 }
 120
 121 void OPPROTO glue(op_psrld, SUFFIX)(void)
 122 {
 123     Reg *d, *s;
 124     int shift;
 125
 126     d = (Reg *)((char *)env + PARAM1);
 127     s = (Reg *)((char *)env + PARAM2);
 128
 129     if (s->Q(0) > 31) {
 130         d->Q(0) = 0;
 131 #if SHIFT == 1
 132         d->Q(1) = 0;
 133 #endif
 134     } else {
 135         shift = s->B(0);
 136         d->L(0) >>= shift;
 137         d->L(1) >>= shift;
 138 #if SHIFT == 1
 139         d->L(2) >>= shift;
 140         d->L(3) >>= shift;
 141 #endif
 142     }
 143     FORCE_RET();
 144 }
 145
 146 void OPPROTO glue(op_psrad, SUFFIX)(void)
 147 {
 148     Reg *d, *s;
 149     int shift;
 150
 151     d = (Reg *)((char *)env + PARAM1);
 152     s = (Reg *)((char *)env + PARAM2);
 153
 154     if (s->Q(0) > 31) {
 155         shift = 31;
 156     } else {
 157         shift = s->B(0);
 158     }
 159     d->L(0) = (int32_t)d->L(0) >> shift;
 160     d->L(1) = (int32_t)d->L(1) >> shift;
 161 #if SHIFT == 1
 162     d->L(2) = (int32_t)d->L(2) >> shift;
 163     d->L(3) = (int32_t)d->L(3) >> shift;
 164 #endif
 165 }
 166
 167 void OPPROTO glue(op_pslld, SUFFIX)(void)
 168 {
 169     Reg *d, *s;
 170     int shift;
 171
 172     d = (Reg *)((char *)env + PARAM1);
 173     s = (Reg *)((char *)env + PARAM2);
 174
 175     if (s->Q(0) > 31) {
 176         d->Q(0) = 0;
 177 #if SHIFT == 1
 178         d->Q(1) = 0;
 179 #endif
 180     } else {
 181         shift = s->B(0);
 182         d->L(0) <<= shift;
 183         d->L(1) <<= shift;
 184 #if SHIFT == 1
 185         d->L(2) <<= shift;
 186         d->L(3) <<= shift;
 187 #endif
 188     }
 189     FORCE_RET();
 190 }
 191
 192 void OPPROTO glue(op_psrlq, SUFFIX)(void)
 193 {
 194     Reg *d, *s;
 195     int shift;
 196
 197     d = (Reg *)((char *)env + PARAM1);
 198     s = (Reg *)((char *)env + PARAM2);
 199
 200     if (s->Q(0) > 63) {
 201         d->Q(0) = 0;
 202 #if SHIFT == 1
 203         d->Q(1) = 0;
 204 #endif
 205     } else {
 206         shift = s->B(0);
 207         d->Q(0) >>= shift;
 208 #if SHIFT == 1
 209         d->Q(1) >>= shift;
 210 #endif
 211     }
 212     FORCE_RET();
 213 }
 214
 215 void OPPROTO glue(op_psllq, SUFFIX)(void)
 216 {
 217     Reg *d, *s;
 218     int shift;
 219
 220     d = (Reg *)((char *)env + PARAM1);
 221     s = (Reg *)((char *)env + PARAM2);
 222
 223     if (s->Q(0) > 63) {
 224         d->Q(0) = 0;
 225 #if SHIFT == 1
 226         d->Q(1) = 0;
 227 #endif
 228     } else {
 229         shift = s->B(0);
 230         d->Q(0) <<= shift;
 231 #if SHIFT == 1
 232         d->Q(1) <<= shift;
 233 #endif
 234     }
 235     FORCE_RET();
 236 }
 237
 238 #if SHIFT == 1
 239 void OPPROTO glue(op_psrldq, SUFFIX)(void)
 240 {
 241     Reg *d, *s;
 242     int shift, i;
 243
 244     d = (Reg *)((char *)env + PARAM1);
 245     s = (Reg *)((char *)env + PARAM2);
 246     shift = s->L(0);
 247     if (shift > 16)
 248         shift = 16;
 249     for(i = 0; i < 16 - shift; i++)
 250         d->B(i) = d->B(i + shift);
 251     for(i = 16 - shift; i < 16; i++)
 252         d->B(i) = 0;
 253     FORCE_RET();
 254 }
 255
 256 void OPPROTO glue(op_pslldq, SUFFIX)(void)
 257 {
 258     Reg *d, *s;
 259     int shift, i;
 260
 261     d = (Reg *)((char *)env + PARAM1);
 262     s = (Reg *)((char *)env + PARAM2);
 263     shift = s->L(0);
 264     if (shift > 16)
 265         shift = 16;
 266     for(i = 15; i >= shift; i--)
 267         d->B(i) = d->B(i - shift);
 268     for(i = 0; i < shift; i++)
 269         d->B(i) = 0;
 270     FORCE_RET();
 271 }
 272 #endif
 273
 274 #define SSE_OP_B(name, F)\
 275 void OPPROTO glue(name, SUFFIX) (void)\
 276 {\
 277     Reg *d, *s;\
 278     d = (Reg *)((char *)env + PARAM1);\
 279     s = (Reg *)((char *)env + PARAM2);\
 280     d->B(0) = F(d->B(0), s->B(0));\
 281     d->B(1) = F(d->B(1), s->B(1));\
 282     d->B(2) = F(d->B(2), s->B(2));\
 283     d->B(3) = F(d->B(3), s->B(3));\
 284     d->B(4) = F(d->B(4), s->B(4));\
 285     d->B(5) = F(d->B(5), s->B(5));\
 286     d->B(6) = F(d->B(6), s->B(6));\
 287     d->B(7) = F(d->B(7), s->B(7));\
 288     XMM_ONLY(\
 289     d->B(8) = F(d->B(8), s->B(8));\
 290     d->B(9) = F(d->B(9), s->B(9));\
 291     d->B(10) = F(d->B(10), s->B(10));\
 292     d->B(11) = F(d->B(11), s->B(11));\
 293     d->B(12) = F(d->B(12), s->B(12));\
 294     d->B(13) = F(d->B(13), s->B(13));\
 295     d->B(14) = F(d->B(14), s->B(14));\
 296     d->B(15) = F(d->B(15), s->B(15));\
 297     )\
 298 }
 299
 300 #define SSE_OP_W(name, F)\
 301 void OPPROTO glue(name, SUFFIX) (void)\
 302 {\
 303     Reg *d, *s;\
 304     d = (Reg *)((char *)env + PARAM1);\
 305     s = (Reg *)((char *)env + PARAM2);\
 306     d->W(0) = F(d->W(0), s->W(0));\
 307     d->W(1) = F(d->W(1), s->W(1));\
 308     d->W(2) = F(d->W(2), s->W(2));\
 309     d->W(3) = F(d->W(3), s->W(3));\
 310     XMM_ONLY(\
 311     d->W(4) = F(d->W(4), s->W(4));\
 312     d->W(5) = F(d->W(5), s->W(5));\
 313     d->W(6) = F(d->W(6), s->W(6));\
 314     d->W(7) = F(d->W(7), s->W(7));\
 315     )\
 316 }
 317
 318 #define SSE_OP_L(name, F)\
 319 void OPPROTO glue(name, SUFFIX) (void)\
 320 {\
 321     Reg *d, *s;\
 322     d = (Reg *)((char *)env + PARAM1);\
 323     s = (Reg *)((char *)env + PARAM2);\
 324     d->L(0) = F(d->L(0), s->L(0));\
 325     d->L(1) = F(d->L(1), s->L(1));\
 326     XMM_ONLY(\
 327     d->L(2) = F(d->L(2), s->L(2));\
 328     d->L(3) = F(d->L(3), s->L(3));\
 329     )\
 330 }
 331
 332 #define SSE_OP_Q(name, F)\
 333 void OPPROTO glue(name, SUFFIX) (void)\
 334 {\
 335     Reg *d, *s;\
 336     d = (Reg *)((char *)env + PARAM1);\
 337     s = (Reg *)((char *)env + PARAM2);\
 338     d->Q(0) = F(d->Q(0), s->Q(0));\
 339     XMM_ONLY(\
 340     d->Q(1) = F(d->Q(1), s->Q(1));\
 341     )\
 342 }
 343
 344 #if SHIFT == 0
 345 static inline int satub(int x)
 346 {
 347     if (x < 0)
 348         return 0;
 349     else if (x > 255)
 350         return 255;
 351     else
 352         return x;
 353 }
 354
 355 static inline int satuw(int x)
 356 {
 357     if (x < 0)
 358         return 0;
 359     else if (x > 65535)
 360         return 65535;
 361     else
 362         return x;
 363 }
 364
 365 static inline int satsb(int x)
 366 {
 367     if (x < -128)
 368         return -128;
 369     else if (x > 127)
 370         return 127;
 371     else
 372         return x;
 373 }
 374
 375 static inline int satsw(int x)
 376 {
 377     if (x < -32768)
 378         return -32768;
 379     else if (x > 32767)
 380         return 32767;
 381     else
 382         return x;
 383 }
 384
 385 #define FADD(a, b) ((a) + (b))
 386 #define FADDUB(a, b) satub((a) + (b))
 387 #define FADDUW(a, b) satuw((a) + (b))
 388 #define FADDSB(a, b) satsb((int8_t)(a) + (int8_t)(b))
 389 #define FADDSW(a, b) satsw((int16_t)(a) + (int16_t)(b))
 390
 391 #define FSUB(a, b) ((a) - (b))
 392 #define FSUBUB(a, b) satub((a) - (b))
 393 #define FSUBUW(a, b) satuw((a) - (b))
 394 #define FSUBSB(a, b) satsb((int8_t)(a) - (int8_t)(b))
 395 #define FSUBSW(a, b) satsw((int16_t)(a) - (int16_t)(b))
 396 #define FMINUB(a, b) ((a) < (b)) ? (a) : (b)
 397 #define FMINSW(a, b) ((int16_t)(a) < (int16_t)(b)) ? (a) : (b)
 398 #define FMAXUB(a, b) ((a) > (b)) ? (a) : (b)
 399 #define FMAXSW(a, b) ((int16_t)(a) > (int16_t)(b)) ? (a) : (b)
 400
 401 #define FAND(a, b) (a) & (b)
 402 #define FANDN(a, b) ((~(a)) & (b))
 403 #define FOR(a, b) (a) | (b)
 404 #define FXOR(a, b) (a) ^ (b)
 405
 406 #define FCMPGTB(a, b) (int8_t)(a) > (int8_t)(b) ? -1 : 0
 407 #define FCMPGTW(a, b) (int16_t)(a) > (int16_t)(b) ? -1 : 0
 408 #define FCMPGTL(a, b) (int32_t)(a) > (int32_t)(b) ? -1 : 0
 409 #define FCMPEQ(a, b) (a) == (b) ? -1 : 0
 410
 411 #define FMULLW(a, b) (a) * (b)
 412 #define FMULHUW(a, b) (a) * (b) >> 16
 413 #define FMULHW(a, b) (int16_t)(a) * (int16_t)(b) >> 16
 414
 415 #define FAVG(a, b) ((a) + (b) + 1) >> 1
 416 #endif
 417
 418 SSE_OP_B(op_paddb, FADD)
 419 SSE_OP_W(op_paddw, FADD)
 420 SSE_OP_L(op_paddl, FADD)
 421 SSE_OP_Q(op_paddq, FADD)
 422
 423 SSE_OP_B(op_psubb, FSUB)
 424 SSE_OP_W(op_psubw, FSUB)
 425 SSE_OP_L(op_psubl, FSUB)
 426 SSE_OP_Q(op_psubq, FSUB)
 427
 428 SSE_OP_B(op_paddusb, FADDUB)
 429 SSE_OP_B(op_paddsb, FADDSB)
 430 SSE_OP_B(op_psubusb, FSUBUB)
 431 SSE_OP_B(op_psubsb, FSUBSB)
 432
 433 SSE_OP_W(op_paddusw, FADDUW)
 434 SSE_OP_W(op_paddsw, FADDSW)
 435 SSE_OP_W(op_psubusw, FSUBUW)
 436 SSE_OP_W(op_psubsw, FSUBSW)
 437
 438 SSE_OP_B(op_pminub, FMINUB)
 439 SSE_OP_B(op_pmaxub, FMAXUB)
 440
 441 SSE_OP_W(op_pminsw, FMINSW)
 442 SSE_OP_W(op_pmaxsw, FMAXSW)
 443
 444 SSE_OP_Q(op_pand, FAND)
 445 SSE_OP_Q(op_pandn, FANDN)
 446 SSE_OP_Q(op_por, FOR)
 447 SSE_OP_Q(op_pxor, FXOR)
 448
 449 SSE_OP_B(op_pcmpgtb, FCMPGTB)
 450 SSE_OP_W(op_pcmpgtw, FCMPGTW)
 451 SSE_OP_L(op_pcmpgtl, FCMPGTL)
 452
 453 SSE_OP_B(op_pcmpeqb, FCMPEQ)
 454 SSE_OP_W(op_pcmpeqw, FCMPEQ)
 455 SSE_OP_L(op_pcmpeql, FCMPEQ)
 456
 457 SSE_OP_W(op_pmullw, FMULLW)
 458 SSE_OP_W(op_pmulhuw, FMULHUW)
 459 SSE_OP_W(op_pmulhw, FMULHW)
 460
 461 SSE_OP_B(op_pavgb, FAVG)
 462 SSE_OP_W(op_pavgw, FAVG)
 463
 464 void OPPROTO glue(op_pmuludq, SUFFIX) (void)
 465 {
 466     Reg *d, *s;
 467     d = (Reg *)((char *)env + PARAM1);
 468     s = (Reg *)((char *)env + PARAM2);
 469
 470     d->Q(0) = (uint64_t)s->L(0) * (uint64_t)d->L(0);
 471 #if SHIFT == 1
 472     d->Q(1) = (uint64_t)s->L(2) * (uint64_t)d->L(2);
 473 #endif
 474 }
 475
 476 void OPPROTO glue(op_pmaddwd, SUFFIX) (void)
 477 {
 478     int i;
 479     Reg *d, *s;
 480     d = (Reg *)((char *)env + PARAM1);
 481     s = (Reg *)((char *)env + PARAM2);
 482
 483     for(i = 0; i < (2 << SHIFT); i++) {
 484         d->L(i) = (int16_t)s->W(2*i) * (int16_t)d->W(2*i) +
 485             (int16_t)s->W(2*i+1) * (int16_t)d->W(2*i+1);
 486     }
 487     FORCE_RET();
 488 }
 489
 490 #if SHIFT == 0
 491 static inline int abs1(int a)
 492 {
 493     if (a < 0)
 494         return -a;
 495     else
 496         return a;
 497 }
 498 #endif
 499 void OPPROTO glue(op_psadbw, SUFFIX) (void)
 500 {
 501     unsigned int val;
 502     Reg *d, *s;
 503     d = (Reg *)((char *)env + PARAM1);
 504     s = (Reg *)((char *)env + PARAM2);
 505
 506     val = 0;
 507     val += abs1(d->B(0) - s->B(0));
 508     val += abs1(d->B(1) - s->B(1));
 509     val += abs1(d->B(2) - s->B(2));
 510     val += abs1(d->B(3) - s->B(3));
 511     val += abs1(d->B(4) - s->B(4));
 512     val += abs1(d->B(5) - s->B(5));
 513     val += abs1(d->B(6) - s->B(6));
 514     val += abs1(d->B(7) - s->B(7));
 515     d->Q(0) = val;
 516 #if SHIFT == 1
 517     val = 0;
 518     val += abs1(d->B(8) - s->B(8));
 519     val += abs1(d->B(9) - s->B(9));
 520     val += abs1(d->B(10) - s->B(10));
 521     val += abs1(d->B(11) - s->B(11));
 522     val += abs1(d->B(12) - s->B(12));
 523     val += abs1(d->B(13) - s->B(13));
 524     val += abs1(d->B(14) - s->B(14));
 525     val += abs1(d->B(15) - s->B(15));
 526     d->Q(1) = val;
 527 #endif
 528 }
 529
 530 void OPPROTO glue(op_maskmov, SUFFIX) (void)
 531 {
 532     int i;
 533     Reg *d, *s;
 534     d = (Reg *)((char *)env + PARAM1);
 535     s = (Reg *)((char *)env + PARAM2);
 536     for(i = 0; i < (8 << SHIFT); i++) {
 537         if (s->B(i) & 0x80)
 538             stb(A0 + i, d->B(i));
 539     }
 540     FORCE_RET();
 541 }
 542
 543 void OPPROTO glue(op_movl_mm_T0, SUFFIX) (void)
 544 {
 545     Reg *d;
 546     d = (Reg *)((char *)env + PARAM1);
 547     d->L(0) = T0;
 548     d->L(1) = 0;
 549 #if SHIFT == 1
 550     d->Q(1) = 0;
 551 #endif
 552 }
 553
 554 void OPPROTO glue(op_movl_T0_mm, SUFFIX) (void)
 555 {
 556     Reg *s;
 557     s = (Reg *)((char *)env + PARAM1);
 558     T0 = s->L(0);
 559 }
 560
 561 #if SHIFT == 0
 562 void OPPROTO glue(op_pshufw, SUFFIX) (void)
 563 {
 564     Reg r, *d, *s;
 565     int order;
 566     d = (Reg *)((char *)env + PARAM1);
 567     s = (Reg *)((char *)env + PARAM2);
 568     order = PARAM3;
 569     r.W(0) = s->W(order & 3);
 570     r.W(1) = s->W((order >> 2) & 3);
 571     r.W(2) = s->W((order >> 4) & 3);
 572     r.W(3) = s->W((order >> 6) & 3);
 573     *d = r;
 574 }
 575 #else
 576 void OPPROTO op_shufps(void)
 577 {
 578     Reg r, *d, *s;
 579     int order;
 580     d = (Reg *)((char *)env + PARAM1);
 581     s = (Reg *)((char *)env + PARAM2);
 582     order = PARAM3;
 583     r.L(0) = d->L(order & 3);
 584     r.L(1) = d->L((order >> 2) & 3);
 585     r.L(2) = s->L((order >> 4) & 3);
 586     r.L(3) = s->L((order >> 6) & 3);
 587     *d = r;
 588 }
 589
 590 void OPPROTO op_shufpd(void)
 591 {
 592     Reg r, *d, *s;
 593     int order;
 594     d = (Reg *)((char *)env + PARAM1);
 595     s = (Reg *)((char *)env + PARAM2);
 596     order = PARAM3;
 597     r.Q(0) = d->Q(order & 1);
 598     r.Q(1) = s->Q((order >> 1) & 1);
 599     *d = r;
 600 }
 601
 602 void OPPROTO glue(op_pshufd, SUFFIX) (void)
 603 {
 604     Reg r, *d, *s;
 605     int order;
 606     d = (Reg *)((char *)env + PARAM1);
 607     s = (Reg *)((char *)env + PARAM2);
 608     order = PARAM3;
 609     r.L(0) = s->L(order & 3);
 610     r.L(1) = s->L((order >> 2) & 3);
 611     r.L(2) = s->L((order >> 4) & 3);
 612     r.L(3) = s->L((order >> 6) & 3);
 613     *d = r;
 614 }
 615
 616 void OPPROTO glue(op_pshuflw, SUFFIX) (void)
 617 {
 618     Reg r, *d, *s;
 619     int order;
 620     d = (Reg *)((char *)env + PARAM1);
 621     s = (Reg *)((char *)env + PARAM2);
 622     order = PARAM3;
 623     r.W(0) = s->W(order & 3);
 624     r.W(1) = s->W((order >> 2) & 3);
 625     r.W(2) = s->W((order >> 4) & 3);
 626     r.W(3) = s->W((order >> 6) & 3);
 627     r.Q(1) = s->Q(1);
 628     *d = r;
 629 }
 630
 631 void OPPROTO glue(op_pshufhw, SUFFIX) (void)
 632 {
 633     Reg r, *d, *s;
 634     int order;
 635     d = (Reg *)((char *)env + PARAM1);
 636     s = (Reg *)((char *)env + PARAM2);
 637     order = PARAM3;
 638     r.Q(0) = s->Q(0);
 639     r.W(4) = s->W(4 + (order & 3));
 640     r.W(5) = s->W(4 + ((order >> 2) & 3));
 641     r.W(6) = s->W(4 + ((order >> 4) & 3));
 642     r.W(7) = s->W(4 + ((order >> 6) & 3));
 643     *d = r;
 644 }
 645 #endif
 646
 647 #if SHIFT == 1
 648 /* FPU ops */
 649 /* XXX: not accurate */
 650
 651 #define SSE_OP_S(name, F)\
 652 void OPPROTO op_ ## name ## ps (void)\
 653 {\
 654     Reg *d, *s;\
 655     d = (Reg *)((char *)env + PARAM1);\
 656     s = (Reg *)((char *)env + PARAM2);\
 657     d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 658     d->XMM_S(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
 659     d->XMM_S(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
 660     d->XMM_S(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
 661 }\
 662 \
 663 void OPPROTO op_ ## name ## ss (void)\
 664 {\
 665     Reg *d, *s;\
 666     d = (Reg *)((char *)env + PARAM1);\
 667     s = (Reg *)((char *)env + PARAM2);\
 668     d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 669 }\
 670 void OPPROTO op_ ## name ## pd (void)\
 671 {\
 672     Reg *d, *s;\
 673     d = (Reg *)((char *)env + PARAM1);\
 674     s = (Reg *)((char *)env + PARAM2);\
 675     d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 676     d->XMM_D(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
 677 }\
 678 \
 679 void OPPROTO op_ ## name ## sd (void)\
 680 {\
 681     Reg *d, *s;\
 682     d = (Reg *)((char *)env + PARAM1);\
 683     s = (Reg *)((char *)env + PARAM2);\
 684     d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 685 }
 686
 687 #define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
 688 #define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
 689 #define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
 690 #define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
 691 #define FPU_MIN(size, a, b) (a) < (b) ? (a) : (b)
 692 #define FPU_MAX(size, a, b) (a) > (b) ? (a) : (b)
 693 #define FPU_SQRT(size, a, b) float ## size ## _sqrt(b, &env->sse_status)
 694
 695 SSE_OP_S(add, FPU_ADD)
 696 SSE_OP_S(sub, FPU_SUB)
 697 SSE_OP_S(mul, FPU_MUL)
 698 SSE_OP_S(div, FPU_DIV)
 699 SSE_OP_S(min, FPU_MIN)
 700 SSE_OP_S(max, FPU_MAX)
 701 SSE_OP_S(sqrt, FPU_SQRT)
 702
 703
 704 /* float to float conversions */
 705 void OPPROTO op_cvtps2pd(void)
 706 {
 707     float32 s0, s1;
 708     Reg *d, *s;
 709     d = (Reg *)((char *)env + PARAM1);
 710     s = (Reg *)((char *)env + PARAM2);
 711     s0 = s->XMM_S(0);
 712     s1 = s->XMM_S(1);
 713     d->XMM_D(0) = float32_to_float64(s0, &env->sse_status);
 714     d->XMM_D(1) = float32_to_float64(s1, &env->sse_status);
 715 }
 716
 717 void OPPROTO op_cvtpd2ps(void)
 718 {
 719     Reg *d, *s;
 720     d = (Reg *)((char *)env + PARAM1);
 721     s = (Reg *)((char *)env + PARAM2);
 722     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 723     d->XMM_S(1) = float64_to_float32(s->XMM_D(1), &env->sse_status);
 724     d->Q(1) = 0;
 725 }
 726
 727 void OPPROTO op_cvtss2sd(void)
 728 {
 729     Reg *d, *s;
 730     d = (Reg *)((char *)env + PARAM1);
 731     s = (Reg *)((char *)env + PARAM2);
 732     d->XMM_D(0) = float32_to_float64(s->XMM_S(0), &env->sse_status);
 733 }
 734
 735 void OPPROTO op_cvtsd2ss(void)
 736 {
 737     Reg *d, *s;
 738     d = (Reg *)((char *)env + PARAM1);
 739     s = (Reg *)((char *)env + PARAM2);
 740     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 741 }
 742
 743 /* integer to float */
 744 void OPPROTO op_cvtdq2ps(void)
 745 {
 746     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 747     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 748     d->XMM_S(0) = int32_to_float32(s->XMM_L(0), &env->sse_status);
 749     d->XMM_S(1) = int32_to_float32(s->XMM_L(1), &env->sse_status);
 750     d->XMM_S(2) = int32_to_float32(s->XMM_L(2), &env->sse_status);
 751     d->XMM_S(3) = int32_to_float32(s->XMM_L(3), &env->sse_status);
 752 }
 753
 754 void OPPROTO op_cvtdq2pd(void)
 755 {
 756     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 757     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 758     int32_t l0, l1;
 759     l0 = (int32_t)s->XMM_L(0);
 760     l1 = (int32_t)s->XMM_L(1);
 761     d->XMM_D(0) = int32_to_float64(l0, &env->sse_status);
 762     d->XMM_D(1) = int32_to_float64(l1, &env->sse_status);
 763 }
 764
 765 void OPPROTO op_cvtpi2ps(void)
 766 {
 767     XMMReg *d = (Reg *)((char *)env + PARAM1);
 768     MMXReg *s = (MMXReg *)((char *)env + PARAM2);
 769     d->XMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
 770     d->XMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
 771 }
 772
 773 void OPPROTO op_cvtpi2pd(void)
 774 {
 775     XMMReg *d = (Reg *)((char *)env + PARAM1);
 776     MMXReg *s = (MMXReg *)((char *)env + PARAM2);
 777     d->XMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
 778     d->XMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
 779 }
 780
 781 void OPPROTO op_cvtsi2ss(void)
 782 {
 783     XMMReg *d = (Reg *)((char *)env + PARAM1);
 784     d->XMM_S(0) = int32_to_float32(T0, &env->sse_status);
 785 }
 786
 787 void OPPROTO op_cvtsi2sd(void)
 788 {
 789     XMMReg *d = (Reg *)((char *)env + PARAM1);
 790     d->XMM_D(0) = int32_to_float64(T0, &env->sse_status);
 791 }
 792
 793 #ifdef TARGET_X86_64
 794 void OPPROTO op_cvtsq2ss(void)
 795 {
 796     XMMReg *d = (Reg *)((char *)env + PARAM1);
 797     d->XMM_S(0) = int64_to_float32(T0, &env->sse_status);
 798 }
 799
 800 void OPPROTO op_cvtsq2sd(void)
 801 {
 802     XMMReg *d = (Reg *)((char *)env + PARAM1);
 803     d->XMM_D(0) = int64_to_float64(T0, &env->sse_status);
 804 }
 805 #endif
 806
 807 /* float to integer */
 808 void OPPROTO op_cvtps2dq(void)
 809 {
 810     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 811     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 812     d->XMM_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 813     d->XMM_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 814     d->XMM_L(2) = float32_to_int32(s->XMM_S(2), &env->sse_status);
 815     d->XMM_L(3) = float32_to_int32(s->XMM_S(3), &env->sse_status);
 816 }
 817
 818 void OPPROTO op_cvtpd2dq(void)
 819 {
 820     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 821     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 822     d->XMM_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 823     d->XMM_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 824     d->XMM_Q(1) = 0;
 825 }
 826
 827 void OPPROTO op_cvtps2pi(void)
 828 {
 829     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
 830     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 831     d->MMX_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 832     d->MMX_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 833 }
 834
 835 void OPPROTO op_cvtpd2pi(void)
 836 {
 837     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
 838     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 839     d->MMX_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 840     d->MMX_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 841 }
 842
 843 void OPPROTO op_cvtss2si(void)
 844 {
 845     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 846     T0 = float32_to_int32(s->XMM_S(0), &env->sse_status);
 847 }
 848
 849 void OPPROTO op_cvtsd2si(void)
 850 {
 851     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 852     T0 = float64_to_int32(s->XMM_D(0), &env->sse_status);
 853 }
 854
 855 #ifdef TARGET_X86_64
 856 void OPPROTO op_cvtss2sq(void)
 857 {
 858     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 859     T0 = float32_to_int64(s->XMM_S(0), &env->sse_status);
 860 }
 861
 862 void OPPROTO op_cvtsd2sq(void)
 863 {
 864     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 865     T0 = float64_to_int64(s->XMM_D(0), &env->sse_status);
 866 }
 867 #endif
 868
 869 /* float to integer truncated */
 870 void OPPROTO op_cvttps2dq(void)
 871 {
 872     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 873     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 874     d->XMM_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 875     d->XMM_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 876     d->XMM_L(2) = float32_to_int32_round_to_zero(s->XMM_S(2), &env->sse_status);
 877     d->XMM_L(3) = float32_to_int32_round_to_zero(s->XMM_S(3), &env->sse_status);
 878 }
 879
 880 void OPPROTO op_cvttpd2dq(void)
 881 {
 882     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 883     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 884     d->XMM_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 885     d->XMM_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 886     d->XMM_Q(1) = 0;
 887 }
 888
 889 void OPPROTO op_cvttps2pi(void)
 890 {
 891     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
 892     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 893     d->MMX_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 894     d->MMX_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 895 }
 896
 897 void OPPROTO op_cvttpd2pi(void)
 898 {
 899     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
 900     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 901     d->MMX_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 902     d->MMX_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 903 }
 904
 905 void OPPROTO op_cvttss2si(void)
 906 {
 907     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 908     T0 = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 909 }
 910
 911 void OPPROTO op_cvttsd2si(void)
 912 {
 913     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 914     T0 = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 915 }
 916
 917 #ifdef TARGET_X86_64
 918 void OPPROTO op_cvttss2sq(void)
 919 {
 920     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 921     T0 = float32_to_int64_round_to_zero(s->XMM_S(0), &env->sse_status);
 922 }
 923
 924 void OPPROTO op_cvttsd2sq(void)
 925 {
 926     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 927     T0 = float64_to_int64_round_to_zero(s->XMM_D(0), &env->sse_status);
 928 }
 929 #endif
 930
 931 void OPPROTO op_rsqrtps(void)
 932 {
 933     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 934     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 935     d->XMM_S(0) = approx_rsqrt(s->XMM_S(0));
 936     d->XMM_S(1) = approx_rsqrt(s->XMM_S(1));
 937     d->XMM_S(2) = approx_rsqrt(s->XMM_S(2));
 938     d->XMM_S(3) = approx_rsqrt(s->XMM_S(3));
 939 }
 940
 941 void OPPROTO op_rsqrtss(void)
 942 {
 943     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 944     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 945     d->XMM_S(0) = approx_rsqrt(s->XMM_S(0));
 946 }
 947
 948 void OPPROTO op_rcpps(void)
 949 {
 950     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 951     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 952     d->XMM_S(0) = approx_rcp(s->XMM_S(0));
 953     d->XMM_S(1) = approx_rcp(s->XMM_S(1));
 954     d->XMM_S(2) = approx_rcp(s->XMM_S(2));
 955     d->XMM_S(3) = approx_rcp(s->XMM_S(3));
 956 }
 957
 958 void OPPROTO op_rcpss(void)
 959 {
 960     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 961     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 962     d->XMM_S(0) = approx_rcp(s->XMM_S(0));
 963 }
 964
 965 void OPPROTO op_haddps(void)
 966 {
 967     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 968     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 969     XMMReg r;
 970     r.XMM_S(0) = d->XMM_S(0) + d->XMM_S(1);
 971     r.XMM_S(1) = d->XMM_S(2) + d->XMM_S(3);
 972     r.XMM_S(2) = s->XMM_S(0) + s->XMM_S(1);
 973     r.XMM_S(3) = s->XMM_S(2) + s->XMM_S(3);
 974     *d = r;
 975 }
 976
 977 void OPPROTO op_haddpd(void)
 978 {
 979     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 980     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 981     XMMReg r;
 982     r.XMM_D(0) = d->XMM_D(0) + d->XMM_D(1);
 983     r.XMM_D(1) = s->XMM_D(0) + s->XMM_D(1);
 984     *d = r;
 985 }
 986
 987 void OPPROTO op_hsubps(void)
 988 {
 989     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 990     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 991     XMMReg r;
 992     r.XMM_S(0) = d->XMM_S(0) - d->XMM_S(1);
 993     r.XMM_S(1) = d->XMM_S(2) - d->XMM_S(3);
 994     r.XMM_S(2) = s->XMM_S(0) - s->XMM_S(1);
 995     r.XMM_S(3) = s->XMM_S(2) - s->XMM_S(3);
 996     *d = r;
 997 }
 998
 999 void OPPROTO op_hsubpd(void)
1000 {
1001     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
1002     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
1003     XMMReg r;
1004     r.XMM_D(0) = d->XMM_D(0) - d->XMM_D(1);
1005     r.XMM_D(1) = s->XMM_D(0) - s->XMM_D(1);
1006     *d = r;
1007 }
1008
1009 void OPPROTO op_addsubps(void)
1010 {
1011     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
1012     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
1013     d->XMM_S(0) = d->XMM_S(0) - s->XMM_S(0);
1014     d->XMM_S(1) = d->XMM_S(1) + s->XMM_S(1);
1015     d->XMM_S(2) = d->XMM_S(2) - s->XMM_S(2);
1016     d->XMM_S(3) = d->XMM_S(3) + s->XMM_S(3);
1017 }
1018
1019 void OPPROTO op_addsubpd(void)
1020 {
1021     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
1022     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
1023     d->XMM_D(0) = d->XMM_D(0) - s->XMM_D(0);
1024     d->XMM_D(1) = d->XMM_D(1) + s->XMM_D(1);
1025 }
1026
1027 /* XXX: unordered */
1028 #define SSE_OP_CMP(name, F)\
1029 void OPPROTO op_ ## name ## ps (void)\
1030 {\
1031     Reg *d, *s;\
1032     d = (Reg *)((char *)env + PARAM1);\
1033     s = (Reg *)((char *)env + PARAM2);\
1034     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
1035     d->XMM_L(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
1036     d->XMM_L(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
1037     d->XMM_L(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
1038 }\
1039 \
1040 void OPPROTO op_ ## name ## ss (void)\
1041 {\
1042     Reg *d, *s;\
1043     d = (Reg *)((char *)env + PARAM1);\
1044     s = (Reg *)((char *)env + PARAM2);\
1045     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
1046 }\
1047 void OPPROTO op_ ## name ## pd (void)\
1048 {\
1049     Reg *d, *s;\
1050     d = (Reg *)((char *)env + PARAM1);\
1051     s = (Reg *)((char *)env + PARAM2);\
1052     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
1053     d->XMM_Q(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
1054 }\
1055 \
1056 void OPPROTO op_ ## name ## sd (void)\
1057 {\
1058     Reg *d, *s;\
1059     d = (Reg *)((char *)env + PARAM1);\
1060     s = (Reg *)((char *)env + PARAM2);\
1061     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
1062 }
1063
1064 #define FPU_CMPEQ(size, a, b) float ## size ## _eq(a, b, &env->sse_status) ? -1 : 0
1065 #define FPU_CMPLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? -1 : 0
1066 #define FPU_CMPLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? -1 : 0
1067 #define FPU_CMPUNORD(size, a, b) float ## size ## _unordered(a, b, &env->sse_status) ? - 1 : 0
1068 #define FPU_CMPNEQ(size, a, b) float ## size ## _eq(a, b, &env->sse_status) ? 0 : -1
1069 #define FPU_CMPNLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? 0 : -1
1070 #define FPU_CMPNLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? 0 : -1
1071 #define FPU_CMPORD(size, a, b) float ## size ## _unordered(a, b, &env->sse_status) ? 0 : -1
1072
1073 SSE_OP_CMP(cmpeq, FPU_CMPEQ)
1074 SSE_OP_CMP(cmplt, FPU_CMPLT)
1075 SSE_OP_CMP(cmple, FPU_CMPLE)
1076 SSE_OP_CMP(cmpunord, FPU_CMPUNORD)
1077 SSE_OP_CMP(cmpneq, FPU_CMPNEQ)
1078 SSE_OP_CMP(cmpnlt, FPU_CMPNLT)
1079 SSE_OP_CMP(cmpnle, FPU_CMPNLE)
1080 SSE_OP_CMP(cmpord, FPU_CMPORD)
1081
1082 const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
1083
1084 void OPPROTO op_ucomiss(void)
1085 {
1086     int ret;
1087     float32 s0, s1;
1088     Reg *d, *s;
1089     d = (Reg *)((char *)env + PARAM1);
1090     s = (Reg *)((char *)env + PARAM2);
1091
1092     s0 = d->XMM_S(0);
1093     s1 = s->XMM_S(0);
1094     ret = float32_compare_quiet(s0, s1, &env->sse_status);
1095     CC_SRC = comis_eflags[ret + 1];
1096     FORCE_RET();
1097 }
1098
1099 void OPPROTO op_comiss(void)
1100 {
1101     int ret;
1102     float32 s0, s1;
1103     Reg *d, *s;
1104     d = (Reg *)((char *)env + PARAM1);
1105     s = (Reg *)((char *)env + PARAM2);
1106
1107     s0 = d->XMM_S(0);
1108     s1 = s->XMM_S(0);
1109     ret = float32_compare(s0, s1, &env->sse_status);
1110     CC_SRC = comis_eflags[ret + 1];
1111     FORCE_RET();
1112 }
1113
1114 void OPPROTO op_ucomisd(void)
1115 {
1116     int ret;
1117     float64 d0, d1;
1118     Reg *d, *s;
1119     d = (Reg *)((char *)env + PARAM1);
1120     s = (Reg *)((char *)env + PARAM2);
1121
1122     d0 = d->XMM_D(0);
1123     d1 = s->XMM_D(0);
1124     ret = float64_compare_quiet(d0, d1, &env->sse_status);
1125     CC_SRC = comis_eflags[ret + 1];
1126     FORCE_RET();
1127 }
1128
1129 void OPPROTO op_comisd(void)
1130 {
1131     int ret;
1132     float64 d0, d1;
1133     Reg *d, *s;
1134     d = (Reg *)((char *)env + PARAM1);
1135     s = (Reg *)((char *)env + PARAM2);
1136
1137     d0 = d->XMM_D(0);
1138     d1 = s->XMM_D(0);
1139     ret = float64_compare(d0, d1, &env->sse_status);
1140     CC_SRC = comis_eflags[ret + 1];
1141     FORCE_RET();
1142 }
1143
1144 void OPPROTO op_movmskps(void)
1145 {
1146     int b0, b1, b2, b3;
1147     Reg *s;
1148     s = (Reg *)((char *)env + PARAM1);
1149     b0 = s->XMM_L(0) >> 31;
1150     b1 = s->XMM_L(1) >> 31;
1151     b2 = s->XMM_L(2) >> 31;
1152     b3 = s->XMM_L(3) >> 31;
1153     T0 = b0 | (b1 << 1) | (b2 << 2) | (b3 << 3);
1154 }
1155
1156 void OPPROTO op_movmskpd(void)
1157 {
1158     int b0, b1;
1159     Reg *s;
1160     s = (Reg *)((char *)env + PARAM1);
1161     b0 = s->XMM_L(1) >> 31;
1162     b1 = s->XMM_L(3) >> 31;
1163     T0 = b0 | (b1 << 1);
1164 }
1165
1166 #endif
1167
1168 void OPPROTO glue(op_pmovmskb, SUFFIX)(void)
1169 {
1170     Reg *s;
1171     s = (Reg *)((char *)env + PARAM1);
1172     T0 = 0;
1173     T0 |= (s->XMM_B(0) >> 7);
1174     T0 |= (s->XMM_B(1) >> 6) & 0x02;
1175     T0 |= (s->XMM_B(2) >> 5) & 0x04;
1176     T0 |= (s->XMM_B(3) >> 4) & 0x08;
1177     T0 |= (s->XMM_B(4) >> 3) & 0x10;
1178     T0 |= (s->XMM_B(5) >> 2) & 0x20;
1179     T0 |= (s->XMM_B(6) >> 1) & 0x40;
1180     T0 |= (s->XMM_B(7)) & 0x80;
1181 #if SHIFT == 1
1182     T0 |= (s->XMM_B(8) << 1) & 0x0100;
1183     T0 |= (s->XMM_B(9) << 2) & 0x0200;
1184     T0 |= (s->XMM_B(10) << 3) & 0x0400;
1185     T0 |= (s->XMM_B(11) << 4) & 0x0800;
1186     T0 |= (s->XMM_B(12) << 5) & 0x1000;
1187     T0 |= (s->XMM_B(13) << 6) & 0x2000;
1188     T0 |= (s->XMM_B(14) << 7) & 0x4000;
1189     T0 |= (s->XMM_B(15) << 8) & 0x8000;
1190 #endif
1191 }
1192
1193 void OPPROTO glue(op_pinsrw, SUFFIX) (void)
1194 {
1195     Reg *d = (Reg *)((char *)env + PARAM1);
1196     int pos = PARAM2;
1197
1198     d->W(pos) = T0;
1199 }
1200
1201 void OPPROTO glue(op_pextrw, SUFFIX) (void)
1202 {
1203     Reg *s = (Reg *)((char *)env + PARAM1);
1204     int pos = PARAM2;
1205
1206     T0 = s->W(pos);
1207 }
1208
1209 void OPPROTO glue(op_packsswb, SUFFIX) (void)
1210 {
1211     Reg r, *d, *s;
1212     d = (Reg *)((char *)env + PARAM1);
1213     s = (Reg *)((char *)env + PARAM2);
1214
1215     r.B(0) = satsb((int16_t)d->W(0));
1216     r.B(1) = satsb((int16_t)d->W(1));
1217     r.B(2) = satsb((int16_t)d->W(2));
1218     r.B(3) = satsb((int16_t)d->W(3));
1219 #if SHIFT == 1
1220     r.B(4) = satsb((int16_t)d->W(4));
1221     r.B(5) = satsb((int16_t)d->W(5));
1222     r.B(6) = satsb((int16_t)d->W(6));
1223     r.B(7) = satsb((int16_t)d->W(7));
1224 #endif
1225     r.B((4 << SHIFT) + 0) = satsb((int16_t)s->W(0));
1226     r.B((4 << SHIFT) + 1) = satsb((int16_t)s->W(1));
1227     r.B((4 << SHIFT) + 2) = satsb((int16_t)s->W(2));
1228     r.B((4 << SHIFT) + 3) = satsb((int16_t)s->W(3));
1229 #if SHIFT == 1
1230     r.B(12) = satsb((int16_t)s->W(4));
1231     r.B(13) = satsb((int16_t)s->W(5));
1232     r.B(14) = satsb((int16_t)s->W(6));
1233     r.B(15) = satsb((int16_t)s->W(7));
1234 #endif
1235     *d = r;
1236 }
1237
1238 void OPPROTO glue(op_packuswb, SUFFIX) (void)
1239 {
1240     Reg r, *d, *s;
1241     d = (Reg *)((char *)env + PARAM1);
1242     s = (Reg *)((char *)env + PARAM2);
1243
1244     r.B(0) = satub((int16_t)d->W(0));
1245     r.B(1) = satub((int16_t)d->W(1));
1246     r.B(2) = satub((int16_t)d->W(2));
1247     r.B(3) = satub((int16_t)d->W(3));
1248 #if SHIFT == 1
1249     r.B(4) = satub((int16_t)d->W(4));
1250     r.B(5) = satub((int16_t)d->W(5));
1251     r.B(6) = satub((int16_t)d->W(6));
1252     r.B(7) = satub((int16_t)d->W(7));
1253 #endif
1254     r.B((4 << SHIFT) + 0) = satub((int16_t)s->W(0));
1255     r.B((4 << SHIFT) + 1) = satub((int16_t)s->W(1));
1256     r.B((4 << SHIFT) + 2) = satub((int16_t)s->W(2));
1257     r.B((4 << SHIFT) + 3) = satub((int16_t)s->W(3));
1258 #if SHIFT == 1
1259     r.B(12) = satub((int16_t)s->W(4));
1260     r.B(13) = satub((int16_t)s->W(5));
1261     r.B(14) = satub((int16_t)s->W(6));
1262     r.B(15) = satub((int16_t)s->W(7));
1263 #endif
1264     *d = r;
1265 }
1266
1267 void OPPROTO glue(op_packssdw, SUFFIX) (void)
1268 {
1269     Reg r, *d, *s;
1270     d = (Reg *)((char *)env + PARAM1);
1271     s = (Reg *)((char *)env + PARAM2);
1272
1273     r.W(0) = satsw(d->L(0));
1274     r.W(1) = satsw(d->L(1));
1275 #if SHIFT == 1
1276     r.W(2) = satsw(d->L(2));
1277     r.W(3) = satsw(d->L(3));
1278 #endif
1279     r.W((2 << SHIFT) + 0) = satsw(s->L(0));
1280     r.W((2 << SHIFT) + 1) = satsw(s->L(1));
1281 #if SHIFT == 1
1282     r.W(6) = satsw(s->L(2));
1283     r.W(7) = satsw(s->L(3));
1284 #endif
1285     *d = r;
1286 }
1287
1288 #define UNPCK_OP(base_name, base)                               \
1289                                                                 \
1290 void OPPROTO glue(op_punpck ## base_name ## bw, SUFFIX) (void)   \
1291 {                                                               \
1292     Reg r, *d, *s;                                              \
1293     d = (Reg *)((char *)env + PARAM1);                          \
1294     s = (Reg *)((char *)env + PARAM2);                          \
1295                                                                 \
1296     r.B(0) = d->B((base << (SHIFT + 2)) + 0);                   \
1297     r.B(1) = s->B((base << (SHIFT + 2)) + 0);                   \
1298     r.B(2) = d->B((base << (SHIFT + 2)) + 1);                   \
1299     r.B(3) = s->B((base << (SHIFT + 2)) + 1);                   \
1300     r.B(4) = d->B((base << (SHIFT + 2)) + 2);                   \
1301     r.B(5) = s->B((base << (SHIFT + 2)) + 2);                   \
1302     r.B(6) = d->B((base << (SHIFT + 2)) + 3);                   \
1303     r.B(7) = s->B((base << (SHIFT + 2)) + 3);                   \
1304 XMM_ONLY(                                                       \
1305     r.B(8) = d->B((base << (SHIFT + 2)) + 4);                   \
1306     r.B(9) = s->B((base << (SHIFT + 2)) + 4);                   \
1307     r.B(10) = d->B((base << (SHIFT + 2)) + 5);                  \
1308     r.B(11) = s->B((base << (SHIFT + 2)) + 5);                  \
1309     r.B(12) = d->B((base << (SHIFT + 2)) + 6);                  \
1310     r.B(13) = s->B((base << (SHIFT + 2)) + 6);                  \
1311     r.B(14) = d->B((base << (SHIFT + 2)) + 7);                  \
1312     r.B(15) = s->B((base << (SHIFT + 2)) + 7);                  \
1313 )                                                               \
1314     *d = r;                                                     \
1315 }                                                               \
1316                                                                 \
1317 void OPPROTO glue(op_punpck ## base_name ## wd, SUFFIX) (void)   \
1318 {                                                               \
1319     Reg r, *d, *s;                                              \
1320     d = (Reg *)((char *)env + PARAM1);                          \
1321     s = (Reg *)((char *)env + PARAM2);                          \
1322                                                                 \
1323     r.W(0) = d->W((base << (SHIFT + 1)) + 0);                   \
1324     r.W(1) = s->W((base << (SHIFT + 1)) + 0);                   \
1325     r.W(2) = d->W((base << (SHIFT + 1)) + 1);                   \
1326     r.W(3) = s->W((base << (SHIFT + 1)) + 1);                   \
1327 XMM_ONLY(                                                       \
1328     r.W(4) = d->W((base << (SHIFT + 1)) + 2);                   \
1329     r.W(5) = s->W((base << (SHIFT + 1)) + 2);                   \
1330     r.W(6) = d->W((base << (SHIFT + 1)) + 3);                   \
1331     r.W(7) = s->W((base << (SHIFT + 1)) + 3);                   \
1332 )                                                               \
1333     *d = r;                                                     \
1334 }                                                               \
1335                                                                 \
1336 void OPPROTO glue(op_punpck ## base_name ## dq, SUFFIX) (void)   \
1337 {                                                               \
1338     Reg r, *d, *s;                                              \
1339     d = (Reg *)((char *)env + PARAM1);                          \
1340     s = (Reg *)((char *)env + PARAM2);                          \
1341                                                                 \
1342     r.L(0) = d->L((base << SHIFT) + 0);                         \
1343     r.L(1) = s->L((base << SHIFT) + 0);                         \
1344 XMM_ONLY(                                                       \
1345     r.L(2) = d->L((base << SHIFT) + 1);                         \
1346     r.L(3) = s->L((base << SHIFT) + 1);                         \
1347 )                                                               \
1348     *d = r;                                                     \
1349 }                                                               \
1350                                                                 \
1351 XMM_ONLY(                                                       \
1352 void OPPROTO glue(op_punpck ## base_name ## qdq, SUFFIX) (void)  \
1353 {                                                               \
1354     Reg r, *d, *s;                                              \
1355     d = (Reg *)((char *)env + PARAM1);                          \
1356     s = (Reg *)((char *)env + PARAM2);                          \
1357                                                                 \
1358     r.Q(0) = d->Q(base);                                        \
1359     r.Q(1) = s->Q(base);                                        \
1360     *d = r;                                                     \
1361 }                                                               \
1362 )
1363
1364 UNPCK_OP(l, 0)
1365 UNPCK_OP(h, 1)
1366
1367 #undef SHIFT
1368 #undef XMM_ONLY
1369 #undef Reg
1370 #undef B
1371 #undef W
1372 #undef L
1373 #undef Q
1374 #undef SUFFIX