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 #ifdef TARGET_X86_64
 562 void OPPROTO glue(op_movq_mm_T0, SUFFIX) (void)
 563 {
 564     Reg *d;
 565     d = (Reg *)((char *)env + PARAM1);
 566     d->Q(0) = T0;
 567 #if SHIFT == 1
 568     d->Q(1) = 0;
 569 #endif
 570 }
 571
 572 void OPPROTO glue(op_movq_T0_mm, SUFFIX) (void)
 573 {
 574     Reg *s;
 575     s = (Reg *)((char *)env + PARAM1);
 576     T0 = s->Q(0);
 577 }
 578 #endif
 579
 580 #if SHIFT == 0
 581 void OPPROTO glue(op_pshufw, SUFFIX) (void)
 582 {
 583     Reg r, *d, *s;
 584     int order;
 585     d = (Reg *)((char *)env + PARAM1);
 586     s = (Reg *)((char *)env + PARAM2);
 587     order = PARAM3;
 588     r.W(0) = s->W(order & 3);
 589     r.W(1) = s->W((order >> 2) & 3);
 590     r.W(2) = s->W((order >> 4) & 3);
 591     r.W(3) = s->W((order >> 6) & 3);
 592     *d = r;
 593 }
 594 #else
 595 void OPPROTO op_shufps(void)
 596 {
 597     Reg r, *d, *s;
 598     int order;
 599     d = (Reg *)((char *)env + PARAM1);
 600     s = (Reg *)((char *)env + PARAM2);
 601     order = PARAM3;
 602     r.L(0) = d->L(order & 3);
 603     r.L(1) = d->L((order >> 2) & 3);
 604     r.L(2) = s->L((order >> 4) & 3);
 605     r.L(3) = s->L((order >> 6) & 3);
 606     *d = r;
 607 }
 608
 609 void OPPROTO op_shufpd(void)
 610 {
 611     Reg r, *d, *s;
 612     int order;
 613     d = (Reg *)((char *)env + PARAM1);
 614     s = (Reg *)((char *)env + PARAM2);
 615     order = PARAM3;
 616     r.Q(0) = d->Q(order & 1);
 617     r.Q(1) = s->Q((order >> 1) & 1);
 618     *d = r;
 619 }
 620
 621 void OPPROTO glue(op_pshufd, SUFFIX) (void)
 622 {
 623     Reg r, *d, *s;
 624     int order;
 625     d = (Reg *)((char *)env + PARAM1);
 626     s = (Reg *)((char *)env + PARAM2);
 627     order = PARAM3;
 628     r.L(0) = s->L(order & 3);
 629     r.L(1) = s->L((order >> 2) & 3);
 630     r.L(2) = s->L((order >> 4) & 3);
 631     r.L(3) = s->L((order >> 6) & 3);
 632     *d = r;
 633 }
 634
 635 void OPPROTO glue(op_pshuflw, SUFFIX) (void)
 636 {
 637     Reg r, *d, *s;
 638     int order;
 639     d = (Reg *)((char *)env + PARAM1);
 640     s = (Reg *)((char *)env + PARAM2);
 641     order = PARAM3;
 642     r.W(0) = s->W(order & 3);
 643     r.W(1) = s->W((order >> 2) & 3);
 644     r.W(2) = s->W((order >> 4) & 3);
 645     r.W(3) = s->W((order >> 6) & 3);
 646     r.Q(1) = s->Q(1);
 647     *d = r;
 648 }
 649
 650 void OPPROTO glue(op_pshufhw, SUFFIX) (void)
 651 {
 652     Reg r, *d, *s;
 653     int order;
 654     d = (Reg *)((char *)env + PARAM1);
 655     s = (Reg *)((char *)env + PARAM2);
 656     order = PARAM3;
 657     r.Q(0) = s->Q(0);
 658     r.W(4) = s->W(4 + (order & 3));
 659     r.W(5) = s->W(4 + ((order >> 2) & 3));
 660     r.W(6) = s->W(4 + ((order >> 4) & 3));
 661     r.W(7) = s->W(4 + ((order >> 6) & 3));
 662     *d = r;
 663 }
 664 #endif
 665
 666 #if SHIFT == 1
 667 /* FPU ops */
 668 /* XXX: not accurate */
 669
 670 #define SSE_OP_S(name, F)\
 671 void OPPROTO op_ ## name ## ps (void)\
 672 {\
 673     Reg *d, *s;\
 674     d = (Reg *)((char *)env + PARAM1);\
 675     s = (Reg *)((char *)env + PARAM2);\
 676     d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 677     d->XMM_S(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
 678     d->XMM_S(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
 679     d->XMM_S(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
 680 }\
 681 \
 682 void OPPROTO op_ ## name ## ss (void)\
 683 {\
 684     Reg *d, *s;\
 685     d = (Reg *)((char *)env + PARAM1);\
 686     s = (Reg *)((char *)env + PARAM2);\
 687     d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 688 }\
 689 void OPPROTO op_ ## name ## pd (void)\
 690 {\
 691     Reg *d, *s;\
 692     d = (Reg *)((char *)env + PARAM1);\
 693     s = (Reg *)((char *)env + PARAM2);\
 694     d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 695     d->XMM_D(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
 696 }\
 697 \
 698 void OPPROTO op_ ## name ## sd (void)\
 699 {\
 700     Reg *d, *s;\
 701     d = (Reg *)((char *)env + PARAM1);\
 702     s = (Reg *)((char *)env + PARAM2);\
 703     d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 704 }
 705
 706 #define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
 707 #define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
 708 #define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
 709 #define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
 710 #define FPU_MIN(size, a, b) (a) < (b) ? (a) : (b)
 711 #define FPU_MAX(size, a, b) (a) > (b) ? (a) : (b)
 712 #define FPU_SQRT(size, a, b) float ## size ## _sqrt(b, &env->sse_status)
 713
 714 SSE_OP_S(add, FPU_ADD)
 715 SSE_OP_S(sub, FPU_SUB)
 716 SSE_OP_S(mul, FPU_MUL)
 717 SSE_OP_S(div, FPU_DIV)
 718 SSE_OP_S(min, FPU_MIN)
 719 SSE_OP_S(max, FPU_MAX)
 720 SSE_OP_S(sqrt, FPU_SQRT)
 721
 722
 723 /* float to float conversions */
 724 void OPPROTO op_cvtps2pd(void)
 725 {
 726     float32 s0, s1;
 727     Reg *d, *s;
 728     d = (Reg *)((char *)env + PARAM1);
 729     s = (Reg *)((char *)env + PARAM2);
 730     s0 = s->XMM_S(0);
 731     s1 = s->XMM_S(1);
 732     d->XMM_D(0) = float32_to_float64(s0, &env->sse_status);
 733     d->XMM_D(1) = float32_to_float64(s1, &env->sse_status);
 734 }
 735
 736 void OPPROTO op_cvtpd2ps(void)
 737 {
 738     Reg *d, *s;
 739     d = (Reg *)((char *)env + PARAM1);
 740     s = (Reg *)((char *)env + PARAM2);
 741     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 742     d->XMM_S(1) = float64_to_float32(s->XMM_D(1), &env->sse_status);
 743     d->Q(1) = 0;
 744 }
 745
 746 void OPPROTO op_cvtss2sd(void)
 747 {
 748     Reg *d, *s;
 749     d = (Reg *)((char *)env + PARAM1);
 750     s = (Reg *)((char *)env + PARAM2);
 751     d->XMM_D(0) = float32_to_float64(s->XMM_S(0), &env->sse_status);
 752 }
 753
 754 void OPPROTO op_cvtsd2ss(void)
 755 {
 756     Reg *d, *s;
 757     d = (Reg *)((char *)env + PARAM1);
 758     s = (Reg *)((char *)env + PARAM2);
 759     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 760 }
 761
 762 /* integer to float */
 763 void OPPROTO op_cvtdq2ps(void)
 764 {
 765     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 766     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 767     d->XMM_S(0) = int32_to_float32(s->XMM_L(0), &env->sse_status);
 768     d->XMM_S(1) = int32_to_float32(s->XMM_L(1), &env->sse_status);
 769     d->XMM_S(2) = int32_to_float32(s->XMM_L(2), &env->sse_status);
 770     d->XMM_S(3) = int32_to_float32(s->XMM_L(3), &env->sse_status);
 771 }
 772
 773 void OPPROTO op_cvtdq2pd(void)
 774 {
 775     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 776     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 777     int32_t l0, l1;
 778     l0 = (int32_t)s->XMM_L(0);
 779     l1 = (int32_t)s->XMM_L(1);
 780     d->XMM_D(0) = int32_to_float64(l0, &env->sse_status);
 781     d->XMM_D(1) = int32_to_float64(l1, &env->sse_status);
 782 }
 783
 784 void OPPROTO op_cvtpi2ps(void)
 785 {
 786     XMMReg *d = (Reg *)((char *)env + PARAM1);
 787     MMXReg *s = (MMXReg *)((char *)env + PARAM2);
 788     d->XMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
 789     d->XMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
 790 }
 791
 792 void OPPROTO op_cvtpi2pd(void)
 793 {
 794     XMMReg *d = (Reg *)((char *)env + PARAM1);
 795     MMXReg *s = (MMXReg *)((char *)env + PARAM2);
 796     d->XMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
 797     d->XMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
 798 }
 799
 800 void OPPROTO op_cvtsi2ss(void)
 801 {
 802     XMMReg *d = (Reg *)((char *)env + PARAM1);
 803     d->XMM_S(0) = int32_to_float32(T0, &env->sse_status);
 804 }
 805
 806 void OPPROTO op_cvtsi2sd(void)
 807 {
 808     XMMReg *d = (Reg *)((char *)env + PARAM1);
 809     d->XMM_D(0) = int32_to_float64(T0, &env->sse_status);
 810 }
 811
 812 #ifdef TARGET_X86_64
 813 void OPPROTO op_cvtsq2ss(void)
 814 {
 815     XMMReg *d = (Reg *)((char *)env + PARAM1);
 816     d->XMM_S(0) = int64_to_float32(T0, &env->sse_status);
 817 }
 818
 819 void OPPROTO op_cvtsq2sd(void)
 820 {
 821     XMMReg *d = (Reg *)((char *)env + PARAM1);
 822     d->XMM_D(0) = int64_to_float64(T0, &env->sse_status);
 823 }
 824 #endif
 825
 826 /* float to integer */
 827 void OPPROTO op_cvtps2dq(void)
 828 {
 829     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 830     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 831     d->XMM_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 832     d->XMM_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 833     d->XMM_L(2) = float32_to_int32(s->XMM_S(2), &env->sse_status);
 834     d->XMM_L(3) = float32_to_int32(s->XMM_S(3), &env->sse_status);
 835 }
 836
 837 void OPPROTO op_cvtpd2dq(void)
 838 {
 839     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 840     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 841     d->XMM_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 842     d->XMM_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 843     d->XMM_Q(1) = 0;
 844 }
 845
 846 void OPPROTO op_cvtps2pi(void)
 847 {
 848     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
 849     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 850     d->MMX_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 851     d->MMX_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 852 }
 853
 854 void OPPROTO op_cvtpd2pi(void)
 855 {
 856     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
 857     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 858     d->MMX_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 859     d->MMX_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 860 }
 861
 862 void OPPROTO op_cvtss2si(void)
 863 {
 864     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 865     T0 = float32_to_int32(s->XMM_S(0), &env->sse_status);
 866 }
 867
 868 void OPPROTO op_cvtsd2si(void)
 869 {
 870     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 871     T0 = float64_to_int32(s->XMM_D(0), &env->sse_status);
 872 }
 873
 874 #ifdef TARGET_X86_64
 875 void OPPROTO op_cvtss2sq(void)
 876 {
 877     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 878     T0 = float32_to_int64(s->XMM_S(0), &env->sse_status);
 879 }
 880
 881 void OPPROTO op_cvtsd2sq(void)
 882 {
 883     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 884     T0 = float64_to_int64(s->XMM_D(0), &env->sse_status);
 885 }
 886 #endif
 887
 888 /* float to integer truncated */
 889 void OPPROTO op_cvttps2dq(void)
 890 {
 891     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 892     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 893     d->XMM_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 894     d->XMM_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 895     d->XMM_L(2) = float32_to_int32_round_to_zero(s->XMM_S(2), &env->sse_status);
 896     d->XMM_L(3) = float32_to_int32_round_to_zero(s->XMM_S(3), &env->sse_status);
 897 }
 898
 899 void OPPROTO op_cvttpd2dq(void)
 900 {
 901     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 902     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 903     d->XMM_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 904     d->XMM_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 905     d->XMM_Q(1) = 0;
 906 }
 907
 908 void OPPROTO op_cvttps2pi(void)
 909 {
 910     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
 911     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 912     d->MMX_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 913     d->MMX_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 914 }
 915
 916 void OPPROTO op_cvttpd2pi(void)
 917 {
 918     MMXReg *d = (MMXReg *)((char *)env + PARAM1);
 919     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 920     d->MMX_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 921     d->MMX_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 922 }
 923
 924 void OPPROTO op_cvttss2si(void)
 925 {
 926     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 927     T0 = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 928 }
 929
 930 void OPPROTO op_cvttsd2si(void)
 931 {
 932     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 933     T0 = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 934 }
 935
 936 #ifdef TARGET_X86_64
 937 void OPPROTO op_cvttss2sq(void)
 938 {
 939     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 940     T0 = float32_to_int64_round_to_zero(s->XMM_S(0), &env->sse_status);
 941 }
 942
 943 void OPPROTO op_cvttsd2sq(void)
 944 {
 945     XMMReg *s = (XMMReg *)((char *)env + PARAM1);
 946     T0 = float64_to_int64_round_to_zero(s->XMM_D(0), &env->sse_status);
 947 }
 948 #endif
 949
 950 void OPPROTO op_rsqrtps(void)
 951 {
 952     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 953     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 954     d->XMM_S(0) = approx_rsqrt(s->XMM_S(0));
 955     d->XMM_S(1) = approx_rsqrt(s->XMM_S(1));
 956     d->XMM_S(2) = approx_rsqrt(s->XMM_S(2));
 957     d->XMM_S(3) = approx_rsqrt(s->XMM_S(3));
 958 }
 959
 960 void OPPROTO op_rsqrtss(void)
 961 {
 962     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 963     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 964     d->XMM_S(0) = approx_rsqrt(s->XMM_S(0));
 965 }
 966
 967 void OPPROTO op_rcpps(void)
 968 {
 969     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 970     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 971     d->XMM_S(0) = approx_rcp(s->XMM_S(0));
 972     d->XMM_S(1) = approx_rcp(s->XMM_S(1));
 973     d->XMM_S(2) = approx_rcp(s->XMM_S(2));
 974     d->XMM_S(3) = approx_rcp(s->XMM_S(3));
 975 }
 976
 977 void OPPROTO op_rcpss(void)
 978 {
 979     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 980     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 981     d->XMM_S(0) = approx_rcp(s->XMM_S(0));
 982 }
 983
 984 void OPPROTO op_haddps(void)
 985 {
 986     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 987     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
 988     XMMReg r;
 989     r.XMM_S(0) = d->XMM_S(0) + d->XMM_S(1);
 990     r.XMM_S(1) = d->XMM_S(2) + d->XMM_S(3);
 991     r.XMM_S(2) = s->XMM_S(0) + s->XMM_S(1);
 992     r.XMM_S(3) = s->XMM_S(2) + s->XMM_S(3);
 993     *d = r;
 994 }
 995
 996 void OPPROTO op_haddpd(void)
 997 {
 998     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
 999     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
1000     XMMReg r;
1001     r.XMM_D(0) = d->XMM_D(0) + d->XMM_D(1);
1002     r.XMM_D(1) = s->XMM_D(0) + s->XMM_D(1);
1003     *d = r;
1004 }
1005
1006 void OPPROTO op_hsubps(void)
1007 {
1008     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
1009     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
1010     XMMReg r;
1011     r.XMM_S(0) = d->XMM_S(0) - d->XMM_S(1);
1012     r.XMM_S(1) = d->XMM_S(2) - d->XMM_S(3);
1013     r.XMM_S(2) = s->XMM_S(0) - s->XMM_S(1);
1014     r.XMM_S(3) = s->XMM_S(2) - s->XMM_S(3);
1015     *d = r;
1016 }
1017
1018 void OPPROTO op_hsubpd(void)
1019 {
1020     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
1021     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
1022     XMMReg r;
1023     r.XMM_D(0) = d->XMM_D(0) - d->XMM_D(1);
1024     r.XMM_D(1) = s->XMM_D(0) - s->XMM_D(1);
1025     *d = r;
1026 }
1027
1028 void OPPROTO op_addsubps(void)
1029 {
1030     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
1031     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
1032     d->XMM_S(0) = d->XMM_S(0) - s->XMM_S(0);
1033     d->XMM_S(1) = d->XMM_S(1) + s->XMM_S(1);
1034     d->XMM_S(2) = d->XMM_S(2) - s->XMM_S(2);
1035     d->XMM_S(3) = d->XMM_S(3) + s->XMM_S(3);
1036 }
1037
1038 void OPPROTO op_addsubpd(void)
1039 {
1040     XMMReg *d = (XMMReg *)((char *)env + PARAM1);
1041     XMMReg *s = (XMMReg *)((char *)env + PARAM2);
1042     d->XMM_D(0) = d->XMM_D(0) - s->XMM_D(0);
1043     d->XMM_D(1) = d->XMM_D(1) + s->XMM_D(1);
1044 }
1045
1046 /* XXX: unordered */
1047 #define SSE_OP_CMP(name, F)\
1048 void OPPROTO op_ ## name ## ps (void)\
1049 {\
1050     Reg *d, *s;\
1051     d = (Reg *)((char *)env + PARAM1);\
1052     s = (Reg *)((char *)env + PARAM2);\
1053     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
1054     d->XMM_L(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
1055     d->XMM_L(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
1056     d->XMM_L(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
1057 }\
1058 \
1059 void OPPROTO op_ ## name ## ss (void)\
1060 {\
1061     Reg *d, *s;\
1062     d = (Reg *)((char *)env + PARAM1);\
1063     s = (Reg *)((char *)env + PARAM2);\
1064     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
1065 }\
1066 void OPPROTO op_ ## name ## pd (void)\
1067 {\
1068     Reg *d, *s;\
1069     d = (Reg *)((char *)env + PARAM1);\
1070     s = (Reg *)((char *)env + PARAM2);\
1071     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
1072     d->XMM_Q(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
1073 }\
1074 \
1075 void OPPROTO op_ ## name ## sd (void)\
1076 {\
1077     Reg *d, *s;\
1078     d = (Reg *)((char *)env + PARAM1);\
1079     s = (Reg *)((char *)env + PARAM2);\
1080     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
1081 }
1082
1083 #define FPU_CMPEQ(size, a, b) float ## size ## _eq(a, b, &env->sse_status) ? -1 : 0
1084 #define FPU_CMPLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? -1 : 0
1085 #define FPU_CMPLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? -1 : 0
1086 #define FPU_CMPUNORD(size, a, b) float ## size ## _unordered(a, b, &env->sse_status) ? - 1 : 0
1087 #define FPU_CMPNEQ(size, a, b) float ## size ## _eq(a, b, &env->sse_status) ? 0 : -1
1088 #define FPU_CMPNLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? 0 : -1
1089 #define FPU_CMPNLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? 0 : -1
1090 #define FPU_CMPORD(size, a, b) float ## size ## _unordered(a, b, &env->sse_status) ? 0 : -1
1091
1092 SSE_OP_CMP(cmpeq, FPU_CMPEQ)
1093 SSE_OP_CMP(cmplt, FPU_CMPLT)
1094 SSE_OP_CMP(cmple, FPU_CMPLE)
1095 SSE_OP_CMP(cmpunord, FPU_CMPUNORD)
1096 SSE_OP_CMP(cmpneq, FPU_CMPNEQ)
1097 SSE_OP_CMP(cmpnlt, FPU_CMPNLT)
1098 SSE_OP_CMP(cmpnle, FPU_CMPNLE)
1099 SSE_OP_CMP(cmpord, FPU_CMPORD)
1100
1101 const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
1102
1103 void OPPROTO op_ucomiss(void)
1104 {
1105     int ret;
1106     float32 s0, s1;
1107     Reg *d, *s;
1108     d = (Reg *)((char *)env + PARAM1);
1109     s = (Reg *)((char *)env + PARAM2);
1110
1111     s0 = d->XMM_S(0);
1112     s1 = s->XMM_S(0);
1113     ret = float32_compare_quiet(s0, s1, &env->sse_status);
1114     CC_SRC = comis_eflags[ret + 1];
1115     FORCE_RET();
1116 }
1117
1118 void OPPROTO op_comiss(void)
1119 {
1120     int ret;
1121     float32 s0, s1;
1122     Reg *d, *s;
1123     d = (Reg *)((char *)env + PARAM1);
1124     s = (Reg *)((char *)env + PARAM2);
1125
1126     s0 = d->XMM_S(0);
1127     s1 = s->XMM_S(0);
1128     ret = float32_compare(s0, s1, &env->sse_status);
1129     CC_SRC = comis_eflags[ret + 1];
1130     FORCE_RET();
1131 }
1132
1133 void OPPROTO op_ucomisd(void)
1134 {
1135     int ret;
1136     float64 d0, d1;
1137     Reg *d, *s;
1138     d = (Reg *)((char *)env + PARAM1);
1139     s = (Reg *)((char *)env + PARAM2);
1140
1141     d0 = d->XMM_D(0);
1142     d1 = s->XMM_D(0);
1143     ret = float64_compare_quiet(d0, d1, &env->sse_status);
1144     CC_SRC = comis_eflags[ret + 1];
1145     FORCE_RET();
1146 }
1147
1148 void OPPROTO op_comisd(void)
1149 {
1150     int ret;
1151     float64 d0, d1;
1152     Reg *d, *s;
1153     d = (Reg *)((char *)env + PARAM1);
1154     s = (Reg *)((char *)env + PARAM2);
1155
1156     d0 = d->XMM_D(0);
1157     d1 = s->XMM_D(0);
1158     ret = float64_compare(d0, d1, &env->sse_status);
1159     CC_SRC = comis_eflags[ret + 1];
1160     FORCE_RET();
1161 }
1162
1163 void OPPROTO op_movmskps(void)
1164 {
1165     int b0, b1, b2, b3;
1166     Reg *s;
1167     s = (Reg *)((char *)env + PARAM1);
1168     b0 = s->XMM_L(0) >> 31;
1169     b1 = s->XMM_L(1) >> 31;
1170     b2 = s->XMM_L(2) >> 31;
1171     b3 = s->XMM_L(3) >> 31;
1172     T0 = b0 | (b1 << 1) | (b2 << 2) | (b3 << 3);
1173 }
1174
1175 void OPPROTO op_movmskpd(void)
1176 {
1177     int b0, b1;
1178     Reg *s;
1179     s = (Reg *)((char *)env + PARAM1);
1180     b0 = s->XMM_L(1) >> 31;
1181     b1 = s->XMM_L(3) >> 31;
1182     T0 = b0 | (b1 << 1);
1183 }
1184
1185 #endif
1186
1187 void OPPROTO glue(op_pmovmskb, SUFFIX)(void)
1188 {
1189     Reg *s;
1190     s = (Reg *)((char *)env + PARAM1);
1191     T0 = 0;
1192     T0 |= (s->XMM_B(0) >> 7);
1193     T0 |= (s->XMM_B(1) >> 6) & 0x02;
1194     T0 |= (s->XMM_B(2) >> 5) & 0x04;
1195     T0 |= (s->XMM_B(3) >> 4) & 0x08;
1196     T0 |= (s->XMM_B(4) >> 3) & 0x10;
1197     T0 |= (s->XMM_B(5) >> 2) & 0x20;
1198     T0 |= (s->XMM_B(6) >> 1) & 0x40;
1199     T0 |= (s->XMM_B(7)) & 0x80;
1200 #if SHIFT == 1
1201     T0 |= (s->XMM_B(8) << 1) & 0x0100;
1202     T0 |= (s->XMM_B(9) << 2) & 0x0200;
1203     T0 |= (s->XMM_B(10) << 3) & 0x0400;
1204     T0 |= (s->XMM_B(11) << 4) & 0x0800;
1205     T0 |= (s->XMM_B(12) << 5) & 0x1000;
1206     T0 |= (s->XMM_B(13) << 6) & 0x2000;
1207     T0 |= (s->XMM_B(14) << 7) & 0x4000;
1208     T0 |= (s->XMM_B(15) << 8) & 0x8000;
1209 #endif
1210 }
1211
1212 void OPPROTO glue(op_pinsrw, SUFFIX) (void)
1213 {
1214     Reg *d = (Reg *)((char *)env + PARAM1);
1215     int pos = PARAM2;
1216
1217     d->W(pos) = T0;
1218 }
1219
1220 void OPPROTO glue(op_pextrw, SUFFIX) (void)
1221 {
1222     Reg *s = (Reg *)((char *)env + PARAM1);
1223     int pos = PARAM2;
1224
1225     T0 = s->W(pos);
1226 }
1227
1228 void OPPROTO glue(op_packsswb, SUFFIX) (void)
1229 {
1230     Reg r, *d, *s;
1231     d = (Reg *)((char *)env + PARAM1);
1232     s = (Reg *)((char *)env + PARAM2);
1233
1234     r.B(0) = satsb((int16_t)d->W(0));
1235     r.B(1) = satsb((int16_t)d->W(1));
1236     r.B(2) = satsb((int16_t)d->W(2));
1237     r.B(3) = satsb((int16_t)d->W(3));
1238 #if SHIFT == 1
1239     r.B(4) = satsb((int16_t)d->W(4));
1240     r.B(5) = satsb((int16_t)d->W(5));
1241     r.B(6) = satsb((int16_t)d->W(6));
1242     r.B(7) = satsb((int16_t)d->W(7));
1243 #endif
1244     r.B((4 << SHIFT) + 0) = satsb((int16_t)s->W(0));
1245     r.B((4 << SHIFT) + 1) = satsb((int16_t)s->W(1));
1246     r.B((4 << SHIFT) + 2) = satsb((int16_t)s->W(2));
1247     r.B((4 << SHIFT) + 3) = satsb((int16_t)s->W(3));
1248 #if SHIFT == 1
1249     r.B(12) = satsb((int16_t)s->W(4));
1250     r.B(13) = satsb((int16_t)s->W(5));
1251     r.B(14) = satsb((int16_t)s->W(6));
1252     r.B(15) = satsb((int16_t)s->W(7));
1253 #endif
1254     *d = r;
1255 }
1256
1257 void OPPROTO glue(op_packuswb, SUFFIX) (void)
1258 {
1259     Reg r, *d, *s;
1260     d = (Reg *)((char *)env + PARAM1);
1261     s = (Reg *)((char *)env + PARAM2);
1262
1263     r.B(0) = satub((int16_t)d->W(0));
1264     r.B(1) = satub((int16_t)d->W(1));
1265     r.B(2) = satub((int16_t)d->W(2));
1266     r.B(3) = satub((int16_t)d->W(3));
1267 #if SHIFT == 1
1268     r.B(4) = satub((int16_t)d->W(4));
1269     r.B(5) = satub((int16_t)d->W(5));
1270     r.B(6) = satub((int16_t)d->W(6));
1271     r.B(7) = satub((int16_t)d->W(7));
1272 #endif
1273     r.B((4 << SHIFT) + 0) = satub((int16_t)s->W(0));
1274     r.B((4 << SHIFT) + 1) = satub((int16_t)s->W(1));
1275     r.B((4 << SHIFT) + 2) = satub((int16_t)s->W(2));
1276     r.B((4 << SHIFT) + 3) = satub((int16_t)s->W(3));
1277 #if SHIFT == 1
1278     r.B(12) = satub((int16_t)s->W(4));
1279     r.B(13) = satub((int16_t)s->W(5));
1280     r.B(14) = satub((int16_t)s->W(6));
1281     r.B(15) = satub((int16_t)s->W(7));
1282 #endif
1283     *d = r;
1284 }
1285
1286 void OPPROTO glue(op_packssdw, SUFFIX) (void)
1287 {
1288     Reg r, *d, *s;
1289     d = (Reg *)((char *)env + PARAM1);
1290     s = (Reg *)((char *)env + PARAM2);
1291
1292     r.W(0) = satsw(d->L(0));
1293     r.W(1) = satsw(d->L(1));
1294 #if SHIFT == 1
1295     r.W(2) = satsw(d->L(2));
1296     r.W(3) = satsw(d->L(3));
1297 #endif
1298     r.W((2 << SHIFT) + 0) = satsw(s->L(0));
1299     r.W((2 << SHIFT) + 1) = satsw(s->L(1));
1300 #if SHIFT == 1
1301     r.W(6) = satsw(s->L(2));
1302     r.W(7) = satsw(s->L(3));
1303 #endif
1304     *d = r;
1305 }
1306
1307 #define UNPCK_OP(base_name, base)                               \
1308                                                                 \
1309 void OPPROTO glue(op_punpck ## base_name ## bw, SUFFIX) (void)   \
1310 {                                                               \
1311     Reg r, *d, *s;                                              \
1312     d = (Reg *)((char *)env + PARAM1);                          \
1313     s = (Reg *)((char *)env + PARAM2);                          \
1314                                                                 \
1315     r.B(0) = d->B((base << (SHIFT + 2)) + 0);                   \
1316     r.B(1) = s->B((base << (SHIFT + 2)) + 0);                   \
1317     r.B(2) = d->B((base << (SHIFT + 2)) + 1);                   \
1318     r.B(3) = s->B((base << (SHIFT + 2)) + 1);                   \
1319     r.B(4) = d->B((base << (SHIFT + 2)) + 2);                   \
1320     r.B(5) = s->B((base << (SHIFT + 2)) + 2);                   \
1321     r.B(6) = d->B((base << (SHIFT + 2)) + 3);                   \
1322     r.B(7) = s->B((base << (SHIFT + 2)) + 3);                   \
1323 XMM_ONLY(                                                       \
1324     r.B(8) = d->B((base << (SHIFT + 2)) + 4);                   \
1325     r.B(9) = s->B((base << (SHIFT + 2)) + 4);                   \
1326     r.B(10) = d->B((base << (SHIFT + 2)) + 5);                  \
1327     r.B(11) = s->B((base << (SHIFT + 2)) + 5);                  \
1328     r.B(12) = d->B((base << (SHIFT + 2)) + 6);                  \
1329     r.B(13) = s->B((base << (SHIFT + 2)) + 6);                  \
1330     r.B(14) = d->B((base << (SHIFT + 2)) + 7);                  \
1331     r.B(15) = s->B((base << (SHIFT + 2)) + 7);                  \
1332 )                                                               \
1333     *d = r;                                                     \
1334 }                                                               \
1335                                                                 \
1336 void OPPROTO glue(op_punpck ## base_name ## wd, SUFFIX) (void)   \
1337 {                                                               \
1338     Reg r, *d, *s;                                              \
1339     d = (Reg *)((char *)env + PARAM1);                          \
1340     s = (Reg *)((char *)env + PARAM2);                          \
1341                                                                 \
1342     r.W(0) = d->W((base << (SHIFT + 1)) + 0);                   \
1343     r.W(1) = s->W((base << (SHIFT + 1)) + 0);                   \
1344     r.W(2) = d->W((base << (SHIFT + 1)) + 1);                   \
1345     r.W(3) = s->W((base << (SHIFT + 1)) + 1);                   \
1346 XMM_ONLY(                                                       \
1347     r.W(4) = d->W((base << (SHIFT + 1)) + 2);                   \
1348     r.W(5) = s->W((base << (SHIFT + 1)) + 2);                   \
1349     r.W(6) = d->W((base << (SHIFT + 1)) + 3);                   \
1350     r.W(7) = s->W((base << (SHIFT + 1)) + 3);                   \
1351 )                                                               \
1352     *d = r;                                                     \
1353 }                                                               \
1354                                                                 \
1355 void OPPROTO glue(op_punpck ## base_name ## dq, SUFFIX) (void)   \
1356 {                                                               \
1357     Reg r, *d, *s;                                              \
1358     d = (Reg *)((char *)env + PARAM1);                          \
1359     s = (Reg *)((char *)env + PARAM2);                          \
1360                                                                 \
1361     r.L(0) = d->L((base << SHIFT) + 0);                         \
1362     r.L(1) = s->L((base << SHIFT) + 0);                         \
1363 XMM_ONLY(                                                       \
1364     r.L(2) = d->L((base << SHIFT) + 1);                         \
1365     r.L(3) = s->L((base << SHIFT) + 1);                         \
1366 )                                                               \
1367     *d = r;                                                     \
1368 }                                                               \
1369                                                                 \
1370 XMM_ONLY(                                                       \
1371 void OPPROTO glue(op_punpck ## base_name ## qdq, SUFFIX) (void)  \
1372 {                                                               \
1373     Reg r, *d, *s;                                              \
1374     d = (Reg *)((char *)env + PARAM1);                          \
1375     s = (Reg *)((char *)env + PARAM2);                          \
1376                                                                 \
1377     r.Q(0) = d->Q(base);                                        \
1378     r.Q(1) = s->Q(base);                                        \
1379     *d = r;                                                     \
1380 }                                                               \
1381 )
1382
1383 UNPCK_OP(l, 0)
1384 UNPCK_OP(h, 1)
1385
1386 #undef SHIFT
1387 #undef XMM_ONLY
1388 #undef Reg
1389 #undef B
1390 #undef W
1391 #undef L
1392 #undef Q
1393 #undef SUFFIX