target-i386/fpu_helper.c

   1 /*
   2  *  x86 FPU, MMX/3DNow!/SSE/SSE2/SSE3/SSSE3/SSE4/PNI helpers
   3  *
   4  *  Copyright (c) 2003 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, see <http://www.gnu.org/licenses/>.
  18  */
  19
  20 #include <math.h>
  21 #include "cpu.h"
  22 #include "dyngen-exec.h"
  23 #include "helper.h"
  24
  25 #if !defined(CONFIG_USER_ONLY)
  26 #include "softmmu_exec.h"
  27 #endif /* !defined(CONFIG_USER_ONLY) */
  28
  29 #define FPU_RC_MASK         0xc00
  30 #define FPU_RC_NEAR         0x000
  31 #define FPU_RC_DOWN         0x400
  32 #define FPU_RC_UP           0x800
  33 #define FPU_RC_CHOP         0xc00
  34
  35 #define MAXTAN 9223372036854775808.0
  36
  37 /* the following deal with x86 long double-precision numbers */
  38 #define MAXEXPD 0x7fff
  39 #define EXPBIAS 16383
  40 #define EXPD(fp)        (fp.l.upper & 0x7fff)
  41 #define SIGND(fp)       ((fp.l.upper) & 0x8000)
  42 #define MANTD(fp)       (fp.l.lower)
  43 #define BIASEXPONENT(fp) fp.l.upper = (fp.l.upper & ~(0x7fff)) | EXPBIAS
  44
  45 #define FPUS_IE (1 << 0)
  46 #define FPUS_DE (1 << 1)
  47 #define FPUS_ZE (1 << 2)
  48 #define FPUS_OE (1 << 3)
  49 #define FPUS_UE (1 << 4)
  50 #define FPUS_PE (1 << 5)
  51 #define FPUS_SF (1 << 6)
  52 #define FPUS_SE (1 << 7)
  53 #define FPUS_B  (1 << 15)
  54
  55 #define FPUC_EM 0x3f
  56
  57 #define floatx80_lg2 make_floatx80(0x3ffd, 0x9a209a84fbcff799LL)
  58 #define floatx80_l2e make_floatx80(0x3fff, 0xb8aa3b295c17f0bcLL)
  59 #define floatx80_l2t make_floatx80(0x4000, 0xd49a784bcd1b8afeLL)
  60
  61 static inline void fpush(void)
  62 {
  63     env->fpstt = (env->fpstt - 1) & 7;
  64     env->fptags[env->fpstt] = 0; /* validate stack entry */
  65 }
  66
  67 static inline void fpop(void)
  68 {
  69     env->fptags[env->fpstt] = 1; /* invalidate stack entry */
  70     env->fpstt = (env->fpstt + 1) & 7;
  71 }
  72
  73 static inline floatx80 helper_fldt(target_ulong ptr)
  74 {
  75     CPU_LDoubleU temp;
  76
  77     temp.l.lower = ldq(ptr);
  78     temp.l.upper = lduw(ptr + 8);
  79     return temp.d;
  80 }
  81
  82 static inline void helper_fstt(floatx80 f, target_ulong ptr)
  83 {
  84     CPU_LDoubleU temp;
  85
  86     temp.d = f;
  87     stq(ptr, temp.l.lower);
  88     stw(ptr + 8, temp.l.upper);
  89 }
  90
  91 /* x87 FPU helpers */
  92
  93 static inline double floatx80_to_double(floatx80 a)
  94 {
  95     union {
  96         float64 f64;
  97         double d;
  98     } u;
  99
 100     u.f64 = floatx80_to_float64(a, &env->fp_status);
 101     return u.d;
 102 }
 103
 104 static inline floatx80 double_to_floatx80(double a)
 105 {
 106     union {
 107         float64 f64;
 108         double d;
 109     } u;
 110
 111     u.d = a;
 112     return float64_to_floatx80(u.f64, &env->fp_status);
 113 }
 114
 115 static void fpu_set_exception(int mask)
 116 {
 117     env->fpus |= mask;
 118     if (env->fpus & (~env->fpuc & FPUC_EM)) {
 119         env->fpus |= FPUS_SE | FPUS_B;
 120     }
 121 }
 122
 123 static inline floatx80 helper_fdiv(floatx80 a, floatx80 b)
 124 {
 125     if (floatx80_is_zero(b)) {
 126         fpu_set_exception(FPUS_ZE);
 127     }
 128     return floatx80_div(a, b, &env->fp_status);
 129 }
 130
 131 static void fpu_raise_exception(void)
 132 {
 133     if (env->cr[0] & CR0_NE_MASK) {
 134         raise_exception(env, EXCP10_COPR);
 135     }
 136 #if !defined(CONFIG_USER_ONLY)
 137     else {
 138         cpu_set_ferr(env);
 139     }
 140 #endif
 141 }
 142
 143 void helper_flds_FT0(uint32_t val)
 144 {
 145     union {
 146         float32 f;
 147         uint32_t i;
 148     } u;
 149
 150     u.i = val;
 151     FT0 = float32_to_floatx80(u.f, &env->fp_status);
 152 }
 153
 154 void helper_fldl_FT0(uint64_t val)
 155 {
 156     union {
 157         float64 f;
 158         uint64_t i;
 159     } u;
 160
 161     u.i = val;
 162     FT0 = float64_to_floatx80(u.f, &env->fp_status);
 163 }
 164
 165 void helper_fildl_FT0(int32_t val)
 166 {
 167     FT0 = int32_to_floatx80(val, &env->fp_status);
 168 }
 169
 170 void helper_flds_ST0(uint32_t val)
 171 {
 172     int new_fpstt;
 173     union {
 174         float32 f;
 175         uint32_t i;
 176     } u;
 177
 178     new_fpstt = (env->fpstt - 1) & 7;
 179     u.i = val;
 180     env->fpregs[new_fpstt].d = float32_to_floatx80(u.f, &env->fp_status);
 181     env->fpstt = new_fpstt;
 182     env->fptags[new_fpstt] = 0; /* validate stack entry */
 183 }
 184
 185 void helper_fldl_ST0(uint64_t val)
 186 {
 187     int new_fpstt;
 188     union {
 189         float64 f;
 190         uint64_t i;
 191     } u;
 192
 193     new_fpstt = (env->fpstt - 1) & 7;
 194     u.i = val;
 195     env->fpregs[new_fpstt].d = float64_to_floatx80(u.f, &env->fp_status);
 196     env->fpstt = new_fpstt;
 197     env->fptags[new_fpstt] = 0; /* validate stack entry */
 198 }
 199
 200 void helper_fildl_ST0(int32_t val)
 201 {
 202     int new_fpstt;
 203
 204     new_fpstt = (env->fpstt - 1) & 7;
 205     env->fpregs[new_fpstt].d = int32_to_floatx80(val, &env->fp_status);
 206     env->fpstt = new_fpstt;
 207     env->fptags[new_fpstt] = 0; /* validate stack entry */
 208 }
 209
 210 void helper_fildll_ST0(int64_t val)
 211 {
 212     int new_fpstt;
 213
 214     new_fpstt = (env->fpstt - 1) & 7;
 215     env->fpregs[new_fpstt].d = int64_to_floatx80(val, &env->fp_status);
 216     env->fpstt = new_fpstt;
 217     env->fptags[new_fpstt] = 0; /* validate stack entry */
 218 }
 219
 220 uint32_t helper_fsts_ST0(void)
 221 {
 222     union {
 223         float32 f;
 224         uint32_t i;
 225     } u;
 226
 227     u.f = floatx80_to_float32(ST0, &env->fp_status);
 228     return u.i;
 229 }
 230
 231 uint64_t helper_fstl_ST0(void)
 232 {
 233     union {
 234         float64 f;
 235         uint64_t i;
 236     } u;
 237
 238     u.f = floatx80_to_float64(ST0, &env->fp_status);
 239     return u.i;
 240 }
 241
 242 int32_t helper_fist_ST0(void)
 243 {
 244     int32_t val;
 245
 246     val = floatx80_to_int32(ST0, &env->fp_status);
 247     if (val != (int16_t)val) {
 248         val = -32768;
 249     }
 250     return val;
 251 }
 252
 253 int32_t helper_fistl_ST0(void)
 254 {
 255     int32_t val;
 256
 257     val = floatx80_to_int32(ST0, &env->fp_status);
 258     return val;
 259 }
 260
 261 int64_t helper_fistll_ST0(void)
 262 {
 263     int64_t val;
 264
 265     val = floatx80_to_int64(ST0, &env->fp_status);
 266     return val;
 267 }
 268
 269 int32_t helper_fistt_ST0(void)
 270 {
 271     int32_t val;
 272
 273     val = floatx80_to_int32_round_to_zero(ST0, &env->fp_status);
 274     if (val != (int16_t)val) {
 275         val = -32768;
 276     }
 277     return val;
 278 }
 279
 280 int32_t helper_fisttl_ST0(void)
 281 {
 282     int32_t val;
 283
 284     val = floatx80_to_int32_round_to_zero(ST0, &env->fp_status);
 285     return val;
 286 }
 287
 288 int64_t helper_fisttll_ST0(void)
 289 {
 290     int64_t val;
 291
 292     val = floatx80_to_int64_round_to_zero(ST0, &env->fp_status);
 293     return val;
 294 }
 295
 296 void helper_fldt_ST0(target_ulong ptr)
 297 {
 298     int new_fpstt;
 299
 300     new_fpstt = (env->fpstt - 1) & 7;
 301     env->fpregs[new_fpstt].d = helper_fldt(ptr);
 302     env->fpstt = new_fpstt;
 303     env->fptags[new_fpstt] = 0; /* validate stack entry */
 304 }
 305
 306 void helper_fstt_ST0(target_ulong ptr)
 307 {
 308     helper_fstt(ST0, ptr);
 309 }
 310
 311 void helper_fpush(void)
 312 {
 313     fpush();
 314 }
 315
 316 void helper_fpop(void)
 317 {
 318     fpop();
 319 }
 320
 321 void helper_fdecstp(void)
 322 {
 323     env->fpstt = (env->fpstt - 1) & 7;
 324     env->fpus &= ~0x4700;
 325 }
 326
 327 void helper_fincstp(void)
 328 {
 329     env->fpstt = (env->fpstt + 1) & 7;
 330     env->fpus &= ~0x4700;
 331 }
 332
 333 /* FPU move */
 334
 335 void helper_ffree_STN(int st_index)
 336 {
 337     env->fptags[(env->fpstt + st_index) & 7] = 1;
 338 }
 339
 340 void helper_fmov_ST0_FT0(void)
 341 {
 342     ST0 = FT0;
 343 }
 344
 345 void helper_fmov_FT0_STN(int st_index)
 346 {
 347     FT0 = ST(st_index);
 348 }
 349
 350 void helper_fmov_ST0_STN(int st_index)
 351 {
 352     ST0 = ST(st_index);
 353 }
 354
 355 void helper_fmov_STN_ST0(int st_index)
 356 {
 357     ST(st_index) = ST0;
 358 }
 359
 360 void helper_fxchg_ST0_STN(int st_index)
 361 {
 362     floatx80 tmp;
 363
 364     tmp = ST(st_index);
 365     ST(st_index) = ST0;
 366     ST0 = tmp;
 367 }
 368
 369 /* FPU operations */
 370
 371 static const int fcom_ccval[4] = {0x0100, 0x4000, 0x0000, 0x4500};
 372
 373 void helper_fcom_ST0_FT0(void)
 374 {
 375     int ret;
 376
 377     ret = floatx80_compare(ST0, FT0, &env->fp_status);
 378     env->fpus = (env->fpus & ~0x4500) | fcom_ccval[ret + 1];
 379 }
 380
 381 void helper_fucom_ST0_FT0(void)
 382 {
 383     int ret;
 384
 385     ret = floatx80_compare_quiet(ST0, FT0, &env->fp_status);
 386     env->fpus = (env->fpus & ~0x4500) | fcom_ccval[ret + 1];
 387 }
 388
 389 static const int fcomi_ccval[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
 390
 391 void helper_fcomi_ST0_FT0(void)
 392 {
 393     int eflags;
 394     int ret;
 395
 396     ret = floatx80_compare(ST0, FT0, &env->fp_status);
 397     eflags = helper_cc_compute_all(CC_OP);
 398     eflags = (eflags & ~(CC_Z | CC_P | CC_C)) | fcomi_ccval[ret + 1];
 399     CC_SRC = eflags;
 400 }
 401
 402 void helper_fucomi_ST0_FT0(void)
 403 {
 404     int eflags;
 405     int ret;
 406
 407     ret = floatx80_compare_quiet(ST0, FT0, &env->fp_status);
 408     eflags = helper_cc_compute_all(CC_OP);
 409     eflags = (eflags & ~(CC_Z | CC_P | CC_C)) | fcomi_ccval[ret + 1];
 410     CC_SRC = eflags;
 411 }
 412
 413 void helper_fadd_ST0_FT0(void)
 414 {
 415     ST0 = floatx80_add(ST0, FT0, &env->fp_status);
 416 }
 417
 418 void helper_fmul_ST0_FT0(void)
 419 {
 420     ST0 = floatx80_mul(ST0, FT0, &env->fp_status);
 421 }
 422
 423 void helper_fsub_ST0_FT0(void)
 424 {
 425     ST0 = floatx80_sub(ST0, FT0, &env->fp_status);
 426 }
 427
 428 void helper_fsubr_ST0_FT0(void)
 429 {
 430     ST0 = floatx80_sub(FT0, ST0, &env->fp_status);
 431 }
 432
 433 void helper_fdiv_ST0_FT0(void)
 434 {
 435     ST0 = helper_fdiv(ST0, FT0);
 436 }
 437
 438 void helper_fdivr_ST0_FT0(void)
 439 {
 440     ST0 = helper_fdiv(FT0, ST0);
 441 }
 442
 443 /* fp operations between STN and ST0 */
 444
 445 void helper_fadd_STN_ST0(int st_index)
 446 {
 447     ST(st_index) = floatx80_add(ST(st_index), ST0, &env->fp_status);
 448 }
 449
 450 void helper_fmul_STN_ST0(int st_index)
 451 {
 452     ST(st_index) = floatx80_mul(ST(st_index), ST0, &env->fp_status);
 453 }
 454
 455 void helper_fsub_STN_ST0(int st_index)
 456 {
 457     ST(st_index) = floatx80_sub(ST(st_index), ST0, &env->fp_status);
 458 }
 459
 460 void helper_fsubr_STN_ST0(int st_index)
 461 {
 462     ST(st_index) = floatx80_sub(ST0, ST(st_index), &env->fp_status);
 463 }
 464
 465 void helper_fdiv_STN_ST0(int st_index)
 466 {
 467     floatx80 *p;
 468
 469     p = &ST(st_index);
 470     *p = helper_fdiv(*p, ST0);
 471 }
 472
 473 void helper_fdivr_STN_ST0(int st_index)
 474 {
 475     floatx80 *p;
 476
 477     p = &ST(st_index);
 478     *p = helper_fdiv(ST0, *p);
 479 }
 480
 481 /* misc FPU operations */
 482 void helper_fchs_ST0(void)
 483 {
 484     ST0 = floatx80_chs(ST0);
 485 }
 486
 487 void helper_fabs_ST0(void)
 488 {
 489     ST0 = floatx80_abs(ST0);
 490 }
 491
 492 void helper_fld1_ST0(void)
 493 {
 494     ST0 = floatx80_one;
 495 }
 496
 497 void helper_fldl2t_ST0(void)
 498 {
 499     ST0 = floatx80_l2t;
 500 }
 501
 502 void helper_fldl2e_ST0(void)
 503 {
 504     ST0 = floatx80_l2e;
 505 }
 506
 507 void helper_fldpi_ST0(void)
 508 {
 509     ST0 = floatx80_pi;
 510 }
 511
 512 void helper_fldlg2_ST0(void)
 513 {
 514     ST0 = floatx80_lg2;
 515 }
 516
 517 void helper_fldln2_ST0(void)
 518 {
 519     ST0 = floatx80_ln2;
 520 }
 521
 522 void helper_fldz_ST0(void)
 523 {
 524     ST0 = floatx80_zero;
 525 }
 526
 527 void helper_fldz_FT0(void)
 528 {
 529     FT0 = floatx80_zero;
 530 }
 531
 532 uint32_t helper_fnstsw(void)
 533 {
 534     return (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
 535 }
 536
 537 uint32_t helper_fnstcw(void)
 538 {
 539     return env->fpuc;
 540 }
 541
 542 static void update_fp_status(void)
 543 {
 544     int rnd_type;
 545
 546     /* set rounding mode */
 547     switch (env->fpuc & FPU_RC_MASK) {
 548     default:
 549     case FPU_RC_NEAR:
 550         rnd_type = float_round_nearest_even;
 551         break;
 552     case FPU_RC_DOWN:
 553         rnd_type = float_round_down;
 554         break;
 555     case FPU_RC_UP:
 556         rnd_type = float_round_up;
 557         break;
 558     case FPU_RC_CHOP:
 559         rnd_type = float_round_to_zero;
 560         break;
 561     }
 562     set_float_rounding_mode(rnd_type, &env->fp_status);
 563     switch ((env->fpuc >> 8) & 3) {
 564     case 0:
 565         rnd_type = 32;
 566         break;
 567     case 2:
 568         rnd_type = 64;
 569         break;
 570     case 3:
 571     default:
 572         rnd_type = 80;
 573         break;
 574     }
 575     set_floatx80_rounding_precision(rnd_type, &env->fp_status);
 576 }
 577
 578 void helper_fldcw(uint32_t val)
 579 {
 580     env->fpuc = val;
 581     update_fp_status();
 582 }
 583
 584 void helper_fclex(void)
 585 {
 586     env->fpus &= 0x7f00;
 587 }
 588
 589 void helper_fwait(void)
 590 {
 591     if (env->fpus & FPUS_SE) {
 592         fpu_raise_exception();
 593     }
 594 }
 595
 596 void helper_fninit(void)
 597 {
 598     env->fpus = 0;
 599     env->fpstt = 0;
 600     env->fpuc = 0x37f;
 601     env->fptags[0] = 1;
 602     env->fptags[1] = 1;
 603     env->fptags[2] = 1;
 604     env->fptags[3] = 1;
 605     env->fptags[4] = 1;
 606     env->fptags[5] = 1;
 607     env->fptags[6] = 1;
 608     env->fptags[7] = 1;
 609 }
 610
 611 /* BCD ops */
 612
 613 void helper_fbld_ST0(target_ulong ptr)
 614 {
 615     floatx80 tmp;
 616     uint64_t val;
 617     unsigned int v;
 618     int i;
 619
 620     val = 0;
 621     for (i = 8; i >= 0; i--) {
 622         v = ldub(ptr + i);
 623         val = (val * 100) + ((v >> 4) * 10) + (v & 0xf);
 624     }
 625     tmp = int64_to_floatx80(val, &env->fp_status);
 626     if (ldub(ptr + 9) & 0x80) {
 627         floatx80_chs(tmp);
 628     }
 629     fpush();
 630     ST0 = tmp;
 631 }
 632
 633 void helper_fbst_ST0(target_ulong ptr)
 634 {
 635     int v;
 636     target_ulong mem_ref, mem_end;
 637     int64_t val;
 638
 639     val = floatx80_to_int64(ST0, &env->fp_status);
 640     mem_ref = ptr;
 641     mem_end = mem_ref + 9;
 642     if (val < 0) {
 643         stb(mem_end, 0x80);
 644         val = -val;
 645     } else {
 646         stb(mem_end, 0x00);
 647     }
 648     while (mem_ref < mem_end) {
 649         if (val == 0) {
 650             break;
 651         }
 652         v = val % 100;
 653         val = val / 100;
 654         v = ((v / 10) << 4) | (v % 10);
 655         stb(mem_ref++, v);
 656     }
 657     while (mem_ref < mem_end) {
 658         stb(mem_ref++, 0);
 659     }
 660 }
 661
 662 void helper_f2xm1(void)
 663 {
 664     double val = floatx80_to_double(ST0);
 665
 666     val = pow(2.0, val) - 1.0;
 667     ST0 = double_to_floatx80(val);
 668 }
 669
 670 void helper_fyl2x(void)
 671 {
 672     double fptemp = floatx80_to_double(ST0);
 673
 674     if (fptemp > 0.0) {
 675         fptemp = log(fptemp) / log(2.0); /* log2(ST) */
 676         fptemp *= floatx80_to_double(ST1);
 677         ST1 = double_to_floatx80(fptemp);
 678         fpop();
 679     } else {
 680         env->fpus &= ~0x4700;
 681         env->fpus |= 0x400;
 682     }
 683 }
 684
 685 void helper_fptan(void)
 686 {
 687     double fptemp = floatx80_to_double(ST0);
 688
 689     if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) {
 690         env->fpus |= 0x400;
 691     } else {
 692         fptemp = tan(fptemp);
 693         ST0 = double_to_floatx80(fptemp);
 694         fpush();
 695         ST0 = floatx80_one;
 696         env->fpus &= ~0x400; /* C2 <-- 0 */
 697         /* the above code is for |arg| < 2**52 only */
 698     }
 699 }
 700
 701 void helper_fpatan(void)
 702 {
 703     double fptemp, fpsrcop;
 704
 705     fpsrcop = floatx80_to_double(ST1);
 706     fptemp = floatx80_to_double(ST0);
 707     ST1 = double_to_floatx80(atan2(fpsrcop, fptemp));
 708     fpop();
 709 }
 710
 711 void helper_fxtract(void)
 712 {
 713     CPU_LDoubleU temp;
 714
 715     temp.d = ST0;
 716
 717     if (floatx80_is_zero(ST0)) {
 718         /* Easy way to generate -inf and raising division by 0 exception */
 719         ST0 = floatx80_div(floatx80_chs(floatx80_one), floatx80_zero,
 720                            &env->fp_status);
 721         fpush();
 722         ST0 = temp.d;
 723     } else {
 724         int expdif;
 725
 726         expdif = EXPD(temp) - EXPBIAS;
 727         /* DP exponent bias */
 728         ST0 = int32_to_floatx80(expdif, &env->fp_status);
 729         fpush();
 730         BIASEXPONENT(temp);
 731         ST0 = temp.d;
 732     }
 733 }
 734
 735 void helper_fprem1(void)
 736 {
 737     double st0, st1, dblq, fpsrcop, fptemp;
 738     CPU_LDoubleU fpsrcop1, fptemp1;
 739     int expdif;
 740     signed long long int q;
 741
 742     st0 = floatx80_to_double(ST0);
 743     st1 = floatx80_to_double(ST1);
 744
 745     if (isinf(st0) || isnan(st0) || isnan(st1) || (st1 == 0.0)) {
 746         ST0 = double_to_floatx80(0.0 / 0.0); /* NaN */
 747         env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
 748         return;
 749     }
 750
 751     fpsrcop = st0;
 752     fptemp = st1;
 753     fpsrcop1.d = ST0;
 754     fptemp1.d = ST1;
 755     expdif = EXPD(fpsrcop1) - EXPD(fptemp1);
 756
 757     if (expdif < 0) {
 758         /* optimisation? taken from the AMD docs */
 759         env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
 760         /* ST0 is unchanged */
 761         return;
 762     }
 763
 764     if (expdif < 53) {
 765         dblq = fpsrcop / fptemp;
 766         /* round dblq towards nearest integer */
 767         dblq = rint(dblq);
 768         st0 = fpsrcop - fptemp * dblq;
 769
 770         /* convert dblq to q by truncating towards zero */
 771         if (dblq < 0.0) {
 772             q = (signed long long int)(-dblq);
 773         } else {
 774             q = (signed long long int)dblq;
 775         }
 776
 777         env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
 778         /* (C0,C3,C1) <-- (q2,q1,q0) */
 779         env->fpus |= (q & 0x4) << (8 - 2);  /* (C0) <-- q2 */
 780         env->fpus |= (q & 0x2) << (14 - 1); /* (C3) <-- q1 */
 781         env->fpus |= (q & 0x1) << (9 - 0);  /* (C1) <-- q0 */
 782     } else {
 783         env->fpus |= 0x400;  /* C2 <-- 1 */
 784         fptemp = pow(2.0, expdif - 50);
 785         fpsrcop = (st0 / st1) / fptemp;
 786         /* fpsrcop = integer obtained by chopping */
 787         fpsrcop = (fpsrcop < 0.0) ?
 788                   -(floor(fabs(fpsrcop))) : floor(fpsrcop);
 789         st0 -= (st1 * fpsrcop * fptemp);
 790     }
 791     ST0 = double_to_floatx80(st0);
 792 }
 793
 794 void helper_fprem(void)
 795 {
 796     double st0, st1, dblq, fpsrcop, fptemp;
 797     CPU_LDoubleU fpsrcop1, fptemp1;
 798     int expdif;
 799     signed long long int q;
 800
 801     st0 = floatx80_to_double(ST0);
 802     st1 = floatx80_to_double(ST1);
 803
 804     if (isinf(st0) || isnan(st0) || isnan(st1) || (st1 == 0.0)) {
 805         ST0 = double_to_floatx80(0.0 / 0.0); /* NaN */
 806         env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
 807         return;
 808     }
 809
 810     fpsrcop = st0;
 811     fptemp = st1;
 812     fpsrcop1.d = ST0;
 813     fptemp1.d = ST1;
 814     expdif = EXPD(fpsrcop1) - EXPD(fptemp1);
 815
 816     if (expdif < 0) {
 817         /* optimisation? taken from the AMD docs */
 818         env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
 819         /* ST0 is unchanged */
 820         return;
 821     }
 822
 823     if (expdif < 53) {
 824         dblq = fpsrcop / fptemp; /* ST0 / ST1 */
 825         /* round dblq towards zero */
 826         dblq = (dblq < 0.0) ? ceil(dblq) : floor(dblq);
 827         st0 = fpsrcop - fptemp * dblq; /* fpsrcop is ST0 */
 828
 829         /* convert dblq to q by truncating towards zero */
 830         if (dblq < 0.0) {
 831             q = (signed long long int)(-dblq);
 832         } else {
 833             q = (signed long long int)dblq;
 834         }
 835
 836         env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
 837         /* (C0,C3,C1) <-- (q2,q1,q0) */
 838         env->fpus |= (q & 0x4) << (8 - 2);  /* (C0) <-- q2 */
 839         env->fpus |= (q & 0x2) << (14 - 1); /* (C3) <-- q1 */
 840         env->fpus |= (q & 0x1) << (9 - 0);  /* (C1) <-- q0 */
 841     } else {
 842         int N = 32 + (expdif % 32); /* as per AMD docs */
 843
 844         env->fpus |= 0x400;  /* C2 <-- 1 */
 845         fptemp = pow(2.0, (double)(expdif - N));
 846         fpsrcop = (st0 / st1) / fptemp;
 847         /* fpsrcop = integer obtained by chopping */
 848         fpsrcop = (fpsrcop < 0.0) ?
 849                   -(floor(fabs(fpsrcop))) : floor(fpsrcop);
 850         st0 -= (st1 * fpsrcop * fptemp);
 851     }
 852     ST0 = double_to_floatx80(st0);
 853 }
 854
 855 void helper_fyl2xp1(void)
 856 {
 857     double fptemp = floatx80_to_double(ST0);
 858
 859     if ((fptemp + 1.0) > 0.0) {
 860         fptemp = log(fptemp + 1.0) / log(2.0); /* log2(ST + 1.0) */
 861         fptemp *= floatx80_to_double(ST1);
 862         ST1 = double_to_floatx80(fptemp);
 863         fpop();
 864     } else {
 865         env->fpus &= ~0x4700;
 866         env->fpus |= 0x400;
 867     }
 868 }
 869
 870 void helper_fsqrt(void)
 871 {
 872     if (floatx80_is_neg(ST0)) {
 873         env->fpus &= ~0x4700;  /* (C3,C2,C1,C0) <-- 0000 */
 874         env->fpus |= 0x400;
 875     }
 876     ST0 = floatx80_sqrt(ST0, &env->fp_status);
 877 }
 878
 879 void helper_fsincos(void)
 880 {
 881     double fptemp = floatx80_to_double(ST0);
 882
 883     if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) {
 884         env->fpus |= 0x400;
 885     } else {
 886         ST0 = double_to_floatx80(sin(fptemp));
 887         fpush();
 888         ST0 = double_to_floatx80(cos(fptemp));
 889         env->fpus &= ~0x400;  /* C2 <-- 0 */
 890         /* the above code is for |arg| < 2**63 only */
 891     }
 892 }
 893
 894 void helper_frndint(void)
 895 {
 896     ST0 = floatx80_round_to_int(ST0, &env->fp_status);
 897 }
 898
 899 void helper_fscale(void)
 900 {
 901     if (floatx80_is_any_nan(ST1)) {
 902         ST0 = ST1;
 903     } else {
 904         int n = floatx80_to_int32_round_to_zero(ST1, &env->fp_status);
 905         ST0 = floatx80_scalbn(ST0, n, &env->fp_status);
 906     }
 907 }
 908
 909 void helper_fsin(void)
 910 {
 911     double fptemp = floatx80_to_double(ST0);
 912
 913     if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) {
 914         env->fpus |= 0x400;
 915     } else {
 916         ST0 = double_to_floatx80(sin(fptemp));
 917         env->fpus &= ~0x400;  /* C2 <-- 0 */
 918         /* the above code is for |arg| < 2**53 only */
 919     }
 920 }
 921
 922 void helper_fcos(void)
 923 {
 924     double fptemp = floatx80_to_double(ST0);
 925
 926     if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) {
 927         env->fpus |= 0x400;
 928     } else {
 929         ST0 = double_to_floatx80(cos(fptemp));
 930         env->fpus &= ~0x400;  /* C2 <-- 0 */
 931         /* the above code is for |arg| < 2**63 only */
 932     }
 933 }
 934
 935 void helper_fxam_ST0(void)
 936 {
 937     CPU_LDoubleU temp;
 938     int expdif;
 939
 940     temp.d = ST0;
 941
 942     env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
 943     if (SIGND(temp)) {
 944         env->fpus |= 0x200; /* C1 <-- 1 */
 945     }
 946
 947     /* XXX: test fptags too */
 948     expdif = EXPD(temp);
 949     if (expdif == MAXEXPD) {
 950         if (MANTD(temp) == 0x8000000000000000ULL) {
 951             env->fpus |= 0x500; /* Infinity */
 952         } else {
 953             env->fpus |= 0x100; /* NaN */
 954         }
 955     } else if (expdif == 0) {
 956         if (MANTD(temp) == 0) {
 957             env->fpus |=  0x4000; /* Zero */
 958         } else {
 959             env->fpus |= 0x4400; /* Denormal */
 960         }
 961     } else {
 962         env->fpus |= 0x400;
 963     }
 964 }
 965
 966 void helper_fstenv(target_ulong ptr, int data32)
 967 {
 968     int fpus, fptag, exp, i;
 969     uint64_t mant;
 970     CPU_LDoubleU tmp;
 971
 972     fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
 973     fptag = 0;
 974     for (i = 7; i >= 0; i--) {
 975         fptag <<= 2;
 976         if (env->fptags[i]) {
 977             fptag |= 3;
 978         } else {
 979             tmp.d = env->fpregs[i].d;
 980             exp = EXPD(tmp);
 981             mant = MANTD(tmp);
 982             if (exp == 0 && mant == 0) {
 983                 /* zero */
 984                 fptag |= 1;
 985             } else if (exp == 0 || exp == MAXEXPD
 986                        || (mant & (1LL << 63)) == 0) {
 987                 /* NaNs, infinity, denormal */
 988                 fptag |= 2;
 989             }
 990         }
 991     }
 992     if (data32) {
 993         /* 32 bit */
 994         stl(ptr, env->fpuc);
 995         stl(ptr + 4, fpus);
 996         stl(ptr + 8, fptag);
 997         stl(ptr + 12, 0); /* fpip */
 998         stl(ptr + 16, 0); /* fpcs */
 999         stl(ptr + 20, 0); /* fpoo */
1000         stl(ptr + 24, 0); /* fpos */
1001     } else {
1002         /* 16 bit */
1003         stw(ptr, env->fpuc);
1004         stw(ptr + 2, fpus);
1005         stw(ptr + 4, fptag);
1006         stw(ptr + 6, 0);
1007         stw(ptr + 8, 0);
1008         stw(ptr + 10, 0);
1009         stw(ptr + 12, 0);
1010     }
1011 }
1012
1013 void helper_fldenv(target_ulong ptr, int data32)
1014 {
1015     int i, fpus, fptag;
1016
1017     if (data32) {
1018         env->fpuc = lduw(ptr);
1019         fpus = lduw(ptr + 4);
1020         fptag = lduw(ptr + 8);
1021     } else {
1022         env->fpuc = lduw(ptr);
1023         fpus = lduw(ptr + 2);
1024         fptag = lduw(ptr + 4);
1025     }
1026     env->fpstt = (fpus >> 11) & 7;
1027     env->fpus = fpus & ~0x3800;
1028     for (i = 0; i < 8; i++) {
1029         env->fptags[i] = ((fptag & 3) == 3);
1030         fptag >>= 2;
1031     }
1032 }
1033
1034 void helper_fsave(target_ulong ptr, int data32)
1035 {
1036     floatx80 tmp;
1037     int i;
1038
1039     helper_fstenv(ptr, data32);
1040
1041     ptr += (14 << data32);
1042     for (i = 0; i < 8; i++) {
1043         tmp = ST(i);
1044         helper_fstt(tmp, ptr);
1045         ptr += 10;
1046     }
1047
1048     /* fninit */
1049     env->fpus = 0;
1050     env->fpstt = 0;
1051     env->fpuc = 0x37f;
1052     env->fptags[0] = 1;
1053     env->fptags[1] = 1;
1054     env->fptags[2] = 1;
1055     env->fptags[3] = 1;
1056     env->fptags[4] = 1;
1057     env->fptags[5] = 1;
1058     env->fptags[6] = 1;
1059     env->fptags[7] = 1;
1060 }
1061
1062 void helper_frstor(target_ulong ptr, int data32)
1063 {
1064     floatx80 tmp;
1065     int i;
1066
1067     helper_fldenv(ptr, data32);
1068     ptr += (14 << data32);
1069
1070     for (i = 0; i < 8; i++) {
1071         tmp = helper_fldt(ptr);
1072         ST(i) = tmp;
1073         ptr += 10;
1074     }
1075 }
1076
1077 #if defined(CONFIG_USER_ONLY)
1078 void cpu_x86_fsave(CPUX86State *s, target_ulong ptr, int data32)
1079 {
1080     CPUX86State *saved_env;
1081
1082     saved_env = env;
1083     env = s;
1084
1085     helper_fsave(ptr, data32);
1086
1087     env = saved_env;
1088 }
1089
1090 void cpu_x86_frstor(CPUX86State *s, target_ulong ptr, int data32)
1091 {
1092     CPUX86State *saved_env;
1093
1094     saved_env = env;
1095     env = s;
1096
1097     helper_frstor(ptr, data32);
1098
1099     env = saved_env;
1100 }
1101 #endif
1102
1103 void helper_fxsave(target_ulong ptr, int data64)
1104 {
1105     int fpus, fptag, i, nb_xmm_regs;
1106     floatx80 tmp;
1107     target_ulong addr;
1108
1109     /* The operand must be 16 byte aligned */
1110     if (ptr & 0xf) {
1111         raise_exception(env, EXCP0D_GPF);
1112     }
1113
1114     fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
1115     fptag = 0;
1116     for (i = 0; i < 8; i++) {
1117         fptag |= (env->fptags[i] << i);
1118     }
1119     stw(ptr, env->fpuc);
1120     stw(ptr + 2, fpus);
1121     stw(ptr + 4, fptag ^ 0xff);
1122 #ifdef TARGET_X86_64
1123     if (data64) {
1124         stq(ptr + 0x08, 0); /* rip */
1125         stq(ptr + 0x10, 0); /* rdp */
1126     } else
1127 #endif
1128     {
1129         stl(ptr + 0x08, 0); /* eip */
1130         stl(ptr + 0x0c, 0); /* sel  */
1131         stl(ptr + 0x10, 0); /* dp */
1132         stl(ptr + 0x14, 0); /* sel  */
1133     }
1134
1135     addr = ptr + 0x20;
1136     for (i = 0; i < 8; i++) {
1137         tmp = ST(i);
1138         helper_fstt(tmp, addr);
1139         addr += 16;
1140     }
1141
1142     if (env->cr[4] & CR4_OSFXSR_MASK) {
1143         /* XXX: finish it */
1144         stl(ptr + 0x18, env->mxcsr); /* mxcsr */
1145         stl(ptr + 0x1c, 0x0000ffff); /* mxcsr_mask */
1146         if (env->hflags & HF_CS64_MASK) {
1147             nb_xmm_regs = 16;
1148         } else {
1149             nb_xmm_regs = 8;
1150         }
1151         addr = ptr + 0xa0;
1152         /* Fast FXSAVE leaves out the XMM registers */
1153         if (!(env->efer & MSR_EFER_FFXSR)
1154             || (env->hflags & HF_CPL_MASK)
1155             || !(env->hflags & HF_LMA_MASK)) {
1156             for (i = 0; i < nb_xmm_regs; i++) {
1157                 stq(addr, env->xmm_regs[i].XMM_Q(0));
1158                 stq(addr + 8, env->xmm_regs[i].XMM_Q(1));
1159                 addr += 16;
1160             }
1161         }
1162     }
1163 }
1164
1165 void helper_fxrstor(target_ulong ptr, int data64)
1166 {
1167     int i, fpus, fptag, nb_xmm_regs;
1168     floatx80 tmp;
1169     target_ulong addr;
1170
1171     /* The operand must be 16 byte aligned */
1172     if (ptr & 0xf) {
1173         raise_exception(env, EXCP0D_GPF);
1174     }
1175
1176     env->fpuc = lduw(ptr);
1177     fpus = lduw(ptr + 2);
1178     fptag = lduw(ptr + 4);
1179     env->fpstt = (fpus >> 11) & 7;
1180     env->fpus = fpus & ~0x3800;
1181     fptag ^= 0xff;
1182     for (i = 0; i < 8; i++) {
1183         env->fptags[i] = ((fptag >> i) & 1);
1184     }
1185
1186     addr = ptr + 0x20;
1187     for (i = 0; i < 8; i++) {
1188         tmp = helper_fldt(addr);
1189         ST(i) = tmp;
1190         addr += 16;
1191     }
1192
1193     if (env->cr[4] & CR4_OSFXSR_MASK) {
1194         /* XXX: finish it */
1195         env->mxcsr = ldl(ptr + 0x18);
1196         /* ldl(ptr + 0x1c); */
1197         if (env->hflags & HF_CS64_MASK) {
1198             nb_xmm_regs = 16;
1199         } else {
1200             nb_xmm_regs = 8;
1201         }
1202         addr = ptr + 0xa0;
1203         /* Fast FXRESTORE leaves out the XMM registers */
1204         if (!(env->efer & MSR_EFER_FFXSR)
1205             || (env->hflags & HF_CPL_MASK)
1206             || !(env->hflags & HF_LMA_MASK)) {
1207             for (i = 0; i < nb_xmm_regs; i++) {
1208                 env->xmm_regs[i].XMM_Q(0) = ldq(addr);
1209                 env->xmm_regs[i].XMM_Q(1) = ldq(addr + 8);
1210                 addr += 16;
1211             }
1212         }
1213     }
1214 }
1215
1216 void cpu_get_fp80(uint64_t *pmant, uint16_t *pexp, floatx80 f)
1217 {
1218     CPU_LDoubleU temp;
1219
1220     temp.d = f;
1221     *pmant = temp.l.lower;
1222     *pexp = temp.l.upper;
1223 }
1224
1225 floatx80 cpu_set_fp80(uint64_t mant, uint16_t upper)
1226 {
1227     CPU_LDoubleU temp;
1228
1229     temp.l.upper = upper;
1230     temp.l.lower = mant;
1231     return temp.d;
1232 }
1233
1234 /* MMX/SSE */
1235 /* XXX: optimize by storing fptt and fptags in the static cpu state */
1236
1237 #define SSE_DAZ             0x0040
1238 #define SSE_RC_MASK         0x6000
1239 #define SSE_RC_NEAR         0x0000
1240 #define SSE_RC_DOWN         0x2000
1241 #define SSE_RC_UP           0x4000
1242 #define SSE_RC_CHOP         0x6000
1243 #define SSE_FZ              0x8000
1244
1245 static void update_sse_status(void)
1246 {
1247     int rnd_type;
1248
1249     /* set rounding mode */
1250     switch (env->mxcsr & SSE_RC_MASK) {
1251     default:
1252     case SSE_RC_NEAR:
1253         rnd_type = float_round_nearest_even;
1254         break;
1255     case SSE_RC_DOWN:
1256         rnd_type = float_round_down;
1257         break;
1258     case SSE_RC_UP:
1259         rnd_type = float_round_up;
1260         break;
1261     case SSE_RC_CHOP:
1262         rnd_type = float_round_to_zero;
1263         break;
1264     }
1265     set_float_rounding_mode(rnd_type, &env->sse_status);
1266
1267     /* set denormals are zero */
1268     set_flush_inputs_to_zero((env->mxcsr & SSE_DAZ) ? 1 : 0, &env->sse_status);
1269
1270     /* set flush to zero */
1271     set_flush_to_zero((env->mxcsr & SSE_FZ) ? 1 : 0, &env->fp_status);
1272 }
1273
1274 void helper_ldmxcsr(uint32_t val)
1275 {
1276     env->mxcsr = val;
1277     update_sse_status();
1278 }
1279
1280 void helper_enter_mmx(void)
1281 {
1282     env->fpstt = 0;
1283     *(uint32_t *)(env->fptags) = 0;
1284     *(uint32_t *)(env->fptags + 4) = 0;
1285 }
1286
1287 void helper_emms(void)
1288 {
1289     /* set to empty state */
1290     *(uint32_t *)(env->fptags) = 0x01010101;
1291     *(uint32_t *)(env->fptags + 4) = 0x01010101;
1292 }
1293
1294 /* XXX: suppress */
1295 void helper_movq(void *d, void *s)
1296 {
1297     *(uint64_t *)d = *(uint64_t *)s;
1298 }
1299
1300 #define SHIFT 0
1301 #include "ops_sse.h"
1302
1303 #define SHIFT 1
1304 #include "ops_sse.h"