target-alpha/op_helper.c

   1 /*
   2  *  Alpha emulation cpu micro-operations helpers for qemu.
   3  *
   4  *  Copyright (c) 2007 Jocelyn Mayer
   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 "exec.h"
  21 #include "host-utils.h"
  22 #include "softfloat.h"
  23 #include "helper.h"
  24 #include "qemu-timer.h"
  25
  26 /*****************************************************************************/
  27 /* Exceptions processing helpers */
  28 void QEMU_NORETURN helper_excp (int excp, int error)
  29 {
  30     env->exception_index = excp;
  31     env->error_code = error;
  32     cpu_loop_exit();
  33 }
  34
  35 uint64_t helper_load_pcc (void)
  36 {
  37     /* ??? This isn't a timer for which we have any rate info.  */
  38     return (uint32_t)cpu_get_real_ticks();
  39 }
  40
  41 uint64_t helper_load_fpcr (void)
  42 {
  43     return cpu_alpha_load_fpcr (env);
  44 }
  45
  46 void helper_store_fpcr (uint64_t val)
  47 {
  48     cpu_alpha_store_fpcr (env, val);
  49 }
  50
  51 uint64_t helper_addqv (uint64_t op1, uint64_t op2)
  52 {
  53     uint64_t tmp = op1;
  54     op1 += op2;
  55     if (unlikely((tmp ^ op2 ^ (-1ULL)) & (tmp ^ op1) & (1ULL << 63))) {
  56         helper_excp(EXCP_ARITH, EXC_M_IOV);
  57     }
  58     return op1;
  59 }
  60
  61 uint64_t helper_addlv (uint64_t op1, uint64_t op2)
  62 {
  63     uint64_t tmp = op1;
  64     op1 = (uint32_t)(op1 + op2);
  65     if (unlikely((tmp ^ op2 ^ (-1UL)) & (tmp ^ op1) & (1UL << 31))) {
  66         helper_excp(EXCP_ARITH, EXC_M_IOV);
  67     }
  68     return op1;
  69 }
  70
  71 uint64_t helper_subqv (uint64_t op1, uint64_t op2)
  72 {
  73     uint64_t res;
  74     res = op1 - op2;
  75     if (unlikely((op1 ^ op2) & (res ^ op1) & (1ULL << 63))) {
  76         helper_excp(EXCP_ARITH, EXC_M_IOV);
  77     }
  78     return res;
  79 }
  80
  81 uint64_t helper_sublv (uint64_t op1, uint64_t op2)
  82 {
  83     uint32_t res;
  84     res = op1 - op2;
  85     if (unlikely((op1 ^ op2) & (res ^ op1) & (1UL << 31))) {
  86         helper_excp(EXCP_ARITH, EXC_M_IOV);
  87     }
  88     return res;
  89 }
  90
  91 uint64_t helper_mullv (uint64_t op1, uint64_t op2)
  92 {
  93     int64_t res = (int64_t)op1 * (int64_t)op2;
  94
  95     if (unlikely((int32_t)res != res)) {
  96         helper_excp(EXCP_ARITH, EXC_M_IOV);
  97     }
  98     return (int64_t)((int32_t)res);
  99 }
 100
 101 uint64_t helper_mulqv (uint64_t op1, uint64_t op2)
 102 {
 103     uint64_t tl, th;
 104
 105     muls64(&tl, &th, op1, op2);
 106     /* If th != 0 && th != -1, then we had an overflow */
 107     if (unlikely((th + 1) > 1)) {
 108         helper_excp(EXCP_ARITH, EXC_M_IOV);
 109     }
 110     return tl;
 111 }
 112
 113 uint64_t helper_umulh (uint64_t op1, uint64_t op2)
 114 {
 115     uint64_t tl, th;
 116
 117     mulu64(&tl, &th, op1, op2);
 118     return th;
 119 }
 120
 121 uint64_t helper_ctpop (uint64_t arg)
 122 {
 123     return ctpop64(arg);
 124 }
 125
 126 uint64_t helper_ctlz (uint64_t arg)
 127 {
 128     return clz64(arg);
 129 }
 130
 131 uint64_t helper_cttz (uint64_t arg)
 132 {
 133     return ctz64(arg);
 134 }
 135
 136 static inline uint64_t byte_zap(uint64_t op, uint8_t mskb)
 137 {
 138     uint64_t mask;
 139
 140     mask = 0;
 141     mask |= ((mskb >> 0) & 1) * 0x00000000000000FFULL;
 142     mask |= ((mskb >> 1) & 1) * 0x000000000000FF00ULL;
 143     mask |= ((mskb >> 2) & 1) * 0x0000000000FF0000ULL;
 144     mask |= ((mskb >> 3) & 1) * 0x00000000FF000000ULL;
 145     mask |= ((mskb >> 4) & 1) * 0x000000FF00000000ULL;
 146     mask |= ((mskb >> 5) & 1) * 0x0000FF0000000000ULL;
 147     mask |= ((mskb >> 6) & 1) * 0x00FF000000000000ULL;
 148     mask |= ((mskb >> 7) & 1) * 0xFF00000000000000ULL;
 149
 150     return op & ~mask;
 151 }
 152
 153 uint64_t helper_zap(uint64_t val, uint64_t mask)
 154 {
 155     return byte_zap(val, mask);
 156 }
 157
 158 uint64_t helper_zapnot(uint64_t val, uint64_t mask)
 159 {
 160     return byte_zap(val, ~mask);
 161 }
 162
 163 uint64_t helper_cmpbge (uint64_t op1, uint64_t op2)
 164 {
 165     uint8_t opa, opb, res;
 166     int i;
 167
 168     res = 0;
 169     for (i = 0; i < 8; i++) {
 170         opa = op1 >> (i * 8);
 171         opb = op2 >> (i * 8);
 172         if (opa >= opb)
 173             res |= 1 << i;
 174     }
 175     return res;
 176 }
 177
 178 uint64_t helper_minub8 (uint64_t op1, uint64_t op2)
 179 {
 180     uint64_t res = 0;
 181     uint8_t opa, opb, opr;
 182     int i;
 183
 184     for (i = 0; i < 8; ++i) {
 185         opa = op1 >> (i * 8);
 186         opb = op2 >> (i * 8);
 187         opr = opa < opb ? opa : opb;
 188         res |= (uint64_t)opr << (i * 8);
 189     }
 190     return res;
 191 }
 192
 193 uint64_t helper_minsb8 (uint64_t op1, uint64_t op2)
 194 {
 195     uint64_t res = 0;
 196     int8_t opa, opb;
 197     uint8_t opr;
 198     int i;
 199
 200     for (i = 0; i < 8; ++i) {
 201         opa = op1 >> (i * 8);
 202         opb = op2 >> (i * 8);
 203         opr = opa < opb ? opa : opb;
 204         res |= (uint64_t)opr << (i * 8);
 205     }
 206     return res;
 207 }
 208
 209 uint64_t helper_minuw4 (uint64_t op1, uint64_t op2)
 210 {
 211     uint64_t res = 0;
 212     uint16_t opa, opb, opr;
 213     int i;
 214
 215     for (i = 0; i < 4; ++i) {
 216         opa = op1 >> (i * 16);
 217         opb = op2 >> (i * 16);
 218         opr = opa < opb ? opa : opb;
 219         res |= (uint64_t)opr << (i * 16);
 220     }
 221     return res;
 222 }
 223
 224 uint64_t helper_minsw4 (uint64_t op1, uint64_t op2)
 225 {
 226     uint64_t res = 0;
 227     int16_t opa, opb;
 228     uint16_t opr;
 229     int i;
 230
 231     for (i = 0; i < 4; ++i) {
 232         opa = op1 >> (i * 16);
 233         opb = op2 >> (i * 16);
 234         opr = opa < opb ? opa : opb;
 235         res |= (uint64_t)opr << (i * 16);
 236     }
 237     return res;
 238 }
 239
 240 uint64_t helper_maxub8 (uint64_t op1, uint64_t op2)
 241 {
 242     uint64_t res = 0;
 243     uint8_t opa, opb, opr;
 244     int i;
 245
 246     for (i = 0; i < 8; ++i) {
 247         opa = op1 >> (i * 8);
 248         opb = op2 >> (i * 8);
 249         opr = opa > opb ? opa : opb;
 250         res |= (uint64_t)opr << (i * 8);
 251     }
 252     return res;
 253 }
 254
 255 uint64_t helper_maxsb8 (uint64_t op1, uint64_t op2)
 256 {
 257     uint64_t res = 0;
 258     int8_t opa, opb;
 259     uint8_t opr;
 260     int i;
 261
 262     for (i = 0; i < 8; ++i) {
 263         opa = op1 >> (i * 8);
 264         opb = op2 >> (i * 8);
 265         opr = opa > opb ? opa : opb;
 266         res |= (uint64_t)opr << (i * 8);
 267     }
 268     return res;
 269 }
 270
 271 uint64_t helper_maxuw4 (uint64_t op1, uint64_t op2)
 272 {
 273     uint64_t res = 0;
 274     uint16_t opa, opb, opr;
 275     int i;
 276
 277     for (i = 0; i < 4; ++i) {
 278         opa = op1 >> (i * 16);
 279         opb = op2 >> (i * 16);
 280         opr = opa > opb ? opa : opb;
 281         res |= (uint64_t)opr << (i * 16);
 282     }
 283     return res;
 284 }
 285
 286 uint64_t helper_maxsw4 (uint64_t op1, uint64_t op2)
 287 {
 288     uint64_t res = 0;
 289     int16_t opa, opb;
 290     uint16_t opr;
 291     int i;
 292
 293     for (i = 0; i < 4; ++i) {
 294         opa = op1 >> (i * 16);
 295         opb = op2 >> (i * 16);
 296         opr = opa > opb ? opa : opb;
 297         res |= (uint64_t)opr << (i * 16);
 298     }
 299     return res;
 300 }
 301
 302 uint64_t helper_perr (uint64_t op1, uint64_t op2)
 303 {
 304     uint64_t res = 0;
 305     uint8_t opa, opb, opr;
 306     int i;
 307
 308     for (i = 0; i < 8; ++i) {
 309         opa = op1 >> (i * 8);
 310         opb = op2 >> (i * 8);
 311         if (opa >= opb)
 312             opr = opa - opb;
 313         else
 314             opr = opb - opa;
 315         res += opr;
 316     }
 317     return res;
 318 }
 319
 320 uint64_t helper_pklb (uint64_t op1)
 321 {
 322     return (op1 & 0xff) | ((op1 >> 24) & 0xff00);
 323 }
 324
 325 uint64_t helper_pkwb (uint64_t op1)
 326 {
 327     return ((op1 & 0xff)
 328             | ((op1 >> 8) & 0xff00)
 329             | ((op1 >> 16) & 0xff0000)
 330             | ((op1 >> 24) & 0xff000000));
 331 }
 332
 333 uint64_t helper_unpkbl (uint64_t op1)
 334 {
 335     return (op1 & 0xff) | ((op1 & 0xff00) << 24);
 336 }
 337
 338 uint64_t helper_unpkbw (uint64_t op1)
 339 {
 340     return ((op1 & 0xff)
 341             | ((op1 & 0xff00) << 8)
 342             | ((op1 & 0xff0000) << 16)
 343             | ((op1 & 0xff000000) << 24));
 344 }
 345
 346 /* Floating point helpers */
 347
 348 void helper_setroundmode (uint32_t val)
 349 {
 350     set_float_rounding_mode(val, &FP_STATUS);
 351 }
 352
 353 void helper_setflushzero (uint32_t val)
 354 {
 355     set_flush_to_zero(val, &FP_STATUS);
 356 }
 357
 358 void helper_fp_exc_clear (void)
 359 {
 360     set_float_exception_flags(0, &FP_STATUS);
 361 }
 362
 363 uint32_t helper_fp_exc_get (void)
 364 {
 365     return get_float_exception_flags(&FP_STATUS);
 366 }
 367
 368 /* Raise exceptions for ieee fp insns without software completion.
 369    In that case there are no exceptions that don't trap; the mask
 370    doesn't apply.  */
 371 void helper_fp_exc_raise(uint32_t exc, uint32_t regno)
 372 {
 373     if (exc) {
 374         uint32_t hw_exc = 0;
 375
 376         env->trap_arg1 = 1ull << regno;
 377
 378         if (exc & float_flag_invalid) {
 379             hw_exc |= EXC_M_INV;
 380         }
 381         if (exc & float_flag_divbyzero) {
 382             hw_exc |= EXC_M_DZE;
 383         }
 384         if (exc & float_flag_overflow) {
 385             hw_exc |= EXC_M_FOV;
 386         }
 387         if (exc & float_flag_underflow) {
 388             hw_exc |= EXC_M_UNF;
 389         }
 390         if (exc & float_flag_inexact) {
 391             hw_exc |= EXC_M_INE;
 392         }
 393         helper_excp(EXCP_ARITH, hw_exc);
 394     }
 395 }
 396
 397 /* Raise exceptions for ieee fp insns with software completion.  */
 398 void helper_fp_exc_raise_s(uint32_t exc, uint32_t regno)
 399 {
 400     if (exc) {
 401         env->fpcr_exc_status |= exc;
 402
 403         exc &= ~env->fpcr_exc_mask;
 404         if (exc) {
 405             helper_fp_exc_raise(exc, regno);
 406         }
 407     }
 408 }
 409
 410 /* Input remapping without software completion.  Handle denormal-map-to-zero
 411    and trap for all other non-finite numbers.  */
 412 uint64_t helper_ieee_input(uint64_t val)
 413 {
 414     uint32_t exp = (uint32_t)(val >> 52) & 0x7ff;
 415     uint64_t frac = val & 0xfffffffffffffull;
 416
 417     if (exp == 0) {
 418         if (frac != 0) {
 419             /* If DNZ is set flush denormals to zero on input.  */
 420             if (env->fpcr_dnz) {
 421                 val &= 1ull << 63;
 422             } else {
 423                 helper_excp(EXCP_ARITH, EXC_M_UNF);
 424             }
 425         }
 426     } else if (exp == 0x7ff) {
 427         /* Infinity or NaN.  */
 428         /* ??? I'm not sure these exception bit flags are correct.  I do
 429            know that the Linux kernel, at least, doesn't rely on them and
 430            just emulates the insn to figure out what exception to use.  */
 431         helper_excp(EXCP_ARITH, frac ? EXC_M_INV : EXC_M_FOV);
 432     }
 433     return val;
 434 }
 435
 436 /* Similar, but does not trap for infinities.  Used for comparisons.  */
 437 uint64_t helper_ieee_input_cmp(uint64_t val)
 438 {
 439     uint32_t exp = (uint32_t)(val >> 52) & 0x7ff;
 440     uint64_t frac = val & 0xfffffffffffffull;
 441
 442     if (exp == 0) {
 443         if (frac != 0) {
 444             /* If DNZ is set flush denormals to zero on input.  */
 445             if (env->fpcr_dnz) {
 446                 val &= 1ull << 63;
 447             } else {
 448                 helper_excp(EXCP_ARITH, EXC_M_UNF);
 449             }
 450         }
 451     } else if (exp == 0x7ff && frac) {
 452         /* NaN.  */
 453         helper_excp(EXCP_ARITH, EXC_M_INV);
 454     }
 455     return val;
 456 }
 457
 458 /* Input remapping with software completion enabled.  All we have to do
 459    is handle denormal-map-to-zero; all other inputs get exceptions as
 460    needed from the actual operation.  */
 461 uint64_t helper_ieee_input_s(uint64_t val)
 462 {
 463     if (env->fpcr_dnz) {
 464         uint32_t exp = (uint32_t)(val >> 52) & 0x7ff;
 465         if (exp == 0) {
 466             val &= 1ull << 63;
 467         }
 468     }
 469     return val;
 470 }
 471
 472 /* F floating (VAX) */
 473 static inline uint64_t float32_to_f(float32 fa)
 474 {
 475     uint64_t r, exp, mant, sig;
 476     CPU_FloatU a;
 477
 478     a.f = fa;
 479     sig = ((uint64_t)a.l & 0x80000000) << 32;
 480     exp = (a.l >> 23) & 0xff;
 481     mant = ((uint64_t)a.l & 0x007fffff) << 29;
 482
 483     if (exp == 255) {
 484         /* NaN or infinity */
 485         r = 1; /* VAX dirty zero */
 486     } else if (exp == 0) {
 487         if (mant == 0) {
 488             /* Zero */
 489             r = 0;
 490         } else {
 491             /* Denormalized */
 492             r = sig | ((exp + 1) << 52) | mant;
 493         }
 494     } else {
 495         if (exp >= 253) {
 496             /* Overflow */
 497             r = 1; /* VAX dirty zero */
 498         } else {
 499             r = sig | ((exp + 2) << 52);
 500         }
 501     }
 502
 503     return r;
 504 }
 505
 506 static inline float32 f_to_float32(uint64_t a)
 507 {
 508     uint32_t exp, mant_sig;
 509     CPU_FloatU r;
 510
 511     exp = ((a >> 55) & 0x80) | ((a >> 52) & 0x7f);
 512     mant_sig = ((a >> 32) & 0x80000000) | ((a >> 29) & 0x007fffff);
 513
 514     if (unlikely(!exp && mant_sig)) {
 515         /* Reserved operands / Dirty zero */
 516         helper_excp(EXCP_OPCDEC, 0);
 517     }
 518
 519     if (exp < 3) {
 520         /* Underflow */
 521         r.l = 0;
 522     } else {
 523         r.l = ((exp - 2) << 23) | mant_sig;
 524     }
 525
 526     return r.f;
 527 }
 528
 529 uint32_t helper_f_to_memory (uint64_t a)
 530 {
 531     uint32_t r;
 532     r =  (a & 0x00001fffe0000000ull) >> 13;
 533     r |= (a & 0x07ffe00000000000ull) >> 45;
 534     r |= (a & 0xc000000000000000ull) >> 48;
 535     return r;
 536 }
 537
 538 uint64_t helper_memory_to_f (uint32_t a)
 539 {
 540     uint64_t r;
 541     r =  ((uint64_t)(a & 0x0000c000)) << 48;
 542     r |= ((uint64_t)(a & 0x003fffff)) << 45;
 543     r |= ((uint64_t)(a & 0xffff0000)) << 13;
 544     if (!(a & 0x00004000))
 545         r |= 0x7ll << 59;
 546     return r;
 547 }
 548
 549 /* ??? Emulating VAX arithmetic with IEEE arithmetic is wrong.  We should
 550    either implement VAX arithmetic properly or just signal invalid opcode.  */
 551
 552 uint64_t helper_addf (uint64_t a, uint64_t b)
 553 {
 554     float32 fa, fb, fr;
 555
 556     fa = f_to_float32(a);
 557     fb = f_to_float32(b);
 558     fr = float32_add(fa, fb, &FP_STATUS);
 559     return float32_to_f(fr);
 560 }
 561
 562 uint64_t helper_subf (uint64_t a, uint64_t b)
 563 {
 564     float32 fa, fb, fr;
 565
 566     fa = f_to_float32(a);
 567     fb = f_to_float32(b);
 568     fr = float32_sub(fa, fb, &FP_STATUS);
 569     return float32_to_f(fr);
 570 }
 571
 572 uint64_t helper_mulf (uint64_t a, uint64_t b)
 573 {
 574     float32 fa, fb, fr;
 575
 576     fa = f_to_float32(a);
 577     fb = f_to_float32(b);
 578     fr = float32_mul(fa, fb, &FP_STATUS);
 579     return float32_to_f(fr);
 580 }
 581
 582 uint64_t helper_divf (uint64_t a, uint64_t b)
 583 {
 584     float32 fa, fb, fr;
 585
 586     fa = f_to_float32(a);
 587     fb = f_to_float32(b);
 588     fr = float32_div(fa, fb, &FP_STATUS);
 589     return float32_to_f(fr);
 590 }
 591
 592 uint64_t helper_sqrtf (uint64_t t)
 593 {
 594     float32 ft, fr;
 595
 596     ft = f_to_float32(t);
 597     fr = float32_sqrt(ft, &FP_STATUS);
 598     return float32_to_f(fr);
 599 }
 600
 601
 602 /* G floating (VAX) */
 603 static inline uint64_t float64_to_g(float64 fa)
 604 {
 605     uint64_t r, exp, mant, sig;
 606     CPU_DoubleU a;
 607
 608     a.d = fa;
 609     sig = a.ll & 0x8000000000000000ull;
 610     exp = (a.ll >> 52) & 0x7ff;
 611     mant = a.ll & 0x000fffffffffffffull;
 612
 613     if (exp == 2047) {
 614         /* NaN or infinity */
 615         r = 1; /* VAX dirty zero */
 616     } else if (exp == 0) {
 617         if (mant == 0) {
 618             /* Zero */
 619             r = 0;
 620         } else {
 621             /* Denormalized */
 622             r = sig | ((exp + 1) << 52) | mant;
 623         }
 624     } else {
 625         if (exp >= 2045) {
 626             /* Overflow */
 627             r = 1; /* VAX dirty zero */
 628         } else {
 629             r = sig | ((exp + 2) << 52);
 630         }
 631     }
 632
 633     return r;
 634 }
 635
 636 static inline float64 g_to_float64(uint64_t a)
 637 {
 638     uint64_t exp, mant_sig;
 639     CPU_DoubleU r;
 640
 641     exp = (a >> 52) & 0x7ff;
 642     mant_sig = a & 0x800fffffffffffffull;
 643
 644     if (!exp && mant_sig) {
 645         /* Reserved operands / Dirty zero */
 646         helper_excp(EXCP_OPCDEC, 0);
 647     }
 648
 649     if (exp < 3) {
 650         /* Underflow */
 651         r.ll = 0;
 652     } else {
 653         r.ll = ((exp - 2) << 52) | mant_sig;
 654     }
 655
 656     return r.d;
 657 }
 658
 659 uint64_t helper_g_to_memory (uint64_t a)
 660 {
 661     uint64_t r;
 662     r =  (a & 0x000000000000ffffull) << 48;
 663     r |= (a & 0x00000000ffff0000ull) << 16;
 664     r |= (a & 0x0000ffff00000000ull) >> 16;
 665     r |= (a & 0xffff000000000000ull) >> 48;
 666     return r;
 667 }
 668
 669 uint64_t helper_memory_to_g (uint64_t a)
 670 {
 671     uint64_t r;
 672     r =  (a & 0x000000000000ffffull) << 48;
 673     r |= (a & 0x00000000ffff0000ull) << 16;
 674     r |= (a & 0x0000ffff00000000ull) >> 16;
 675     r |= (a & 0xffff000000000000ull) >> 48;
 676     return r;
 677 }
 678
 679 uint64_t helper_addg (uint64_t a, uint64_t b)
 680 {
 681     float64 fa, fb, fr;
 682
 683     fa = g_to_float64(a);
 684     fb = g_to_float64(b);
 685     fr = float64_add(fa, fb, &FP_STATUS);
 686     return float64_to_g(fr);
 687 }
 688
 689 uint64_t helper_subg (uint64_t a, uint64_t b)
 690 {
 691     float64 fa, fb, fr;
 692
 693     fa = g_to_float64(a);
 694     fb = g_to_float64(b);
 695     fr = float64_sub(fa, fb, &FP_STATUS);
 696     return float64_to_g(fr);
 697 }
 698
 699 uint64_t helper_mulg (uint64_t a, uint64_t b)
 700 {
 701     float64 fa, fb, fr;
 702
 703     fa = g_to_float64(a);
 704     fb = g_to_float64(b);
 705     fr = float64_mul(fa, fb, &FP_STATUS);
 706     return float64_to_g(fr);
 707 }
 708
 709 uint64_t helper_divg (uint64_t a, uint64_t b)
 710 {
 711     float64 fa, fb, fr;
 712
 713     fa = g_to_float64(a);
 714     fb = g_to_float64(b);
 715     fr = float64_div(fa, fb, &FP_STATUS);
 716     return float64_to_g(fr);
 717 }
 718
 719 uint64_t helper_sqrtg (uint64_t a)
 720 {
 721     float64 fa, fr;
 722
 723     fa = g_to_float64(a);
 724     fr = float64_sqrt(fa, &FP_STATUS);
 725     return float64_to_g(fr);
 726 }
 727
 728
 729 /* S floating (single) */
 730
 731 /* Taken from linux/arch/alpha/kernel/traps.c, s_mem_to_reg.  */
 732 static inline uint64_t float32_to_s_int(uint32_t fi)
 733 {
 734     uint32_t frac = fi & 0x7fffff;
 735     uint32_t sign = fi >> 31;
 736     uint32_t exp_msb = (fi >> 30) & 1;
 737     uint32_t exp_low = (fi >> 23) & 0x7f;
 738     uint32_t exp;
 739
 740     exp = (exp_msb << 10) | exp_low;
 741     if (exp_msb) {
 742         if (exp_low == 0x7f)
 743             exp = 0x7ff;
 744     } else {
 745         if (exp_low != 0x00)
 746             exp |= 0x380;
 747     }
 748
 749     return (((uint64_t)sign << 63)
 750             | ((uint64_t)exp << 52)
 751             | ((uint64_t)frac << 29));
 752 }
 753
 754 static inline uint64_t float32_to_s(float32 fa)
 755 {
 756     CPU_FloatU a;
 757     a.f = fa;
 758     return float32_to_s_int(a.l);
 759 }
 760
 761 static inline uint32_t s_to_float32_int(uint64_t a)
 762 {
 763     return ((a >> 32) & 0xc0000000) | ((a >> 29) & 0x3fffffff);
 764 }
 765
 766 static inline float32 s_to_float32(uint64_t a)
 767 {
 768     CPU_FloatU r;
 769     r.l = s_to_float32_int(a);
 770     return r.f;
 771 }
 772
 773 uint32_t helper_s_to_memory (uint64_t a)
 774 {
 775     return s_to_float32_int(a);
 776 }
 777
 778 uint64_t helper_memory_to_s (uint32_t a)
 779 {
 780     return float32_to_s_int(a);
 781 }
 782
 783 uint64_t helper_adds (uint64_t a, uint64_t b)
 784 {
 785     float32 fa, fb, fr;
 786
 787     fa = s_to_float32(a);
 788     fb = s_to_float32(b);
 789     fr = float32_add(fa, fb, &FP_STATUS);
 790     return float32_to_s(fr);
 791 }
 792
 793 uint64_t helper_subs (uint64_t a, uint64_t b)
 794 {
 795     float32 fa, fb, fr;
 796
 797     fa = s_to_float32(a);
 798     fb = s_to_float32(b);
 799     fr = float32_sub(fa, fb, &FP_STATUS);
 800     return float32_to_s(fr);
 801 }
 802
 803 uint64_t helper_muls (uint64_t a, uint64_t b)
 804 {
 805     float32 fa, fb, fr;
 806
 807     fa = s_to_float32(a);
 808     fb = s_to_float32(b);
 809     fr = float32_mul(fa, fb, &FP_STATUS);
 810     return float32_to_s(fr);
 811 }
 812
 813 uint64_t helper_divs (uint64_t a, uint64_t b)
 814 {
 815     float32 fa, fb, fr;
 816
 817     fa = s_to_float32(a);
 818     fb = s_to_float32(b);
 819     fr = float32_div(fa, fb, &FP_STATUS);
 820     return float32_to_s(fr);
 821 }
 822
 823 uint64_t helper_sqrts (uint64_t a)
 824 {
 825     float32 fa, fr;
 826
 827     fa = s_to_float32(a);
 828     fr = float32_sqrt(fa, &FP_STATUS);
 829     return float32_to_s(fr);
 830 }
 831
 832
 833 /* T floating (double) */
 834 static inline float64 t_to_float64(uint64_t a)
 835 {
 836     /* Memory format is the same as float64 */
 837     CPU_DoubleU r;
 838     r.ll = a;
 839     return r.d;
 840 }
 841
 842 static inline uint64_t float64_to_t(float64 fa)
 843 {
 844     /* Memory format is the same as float64 */
 845     CPU_DoubleU r;
 846     r.d = fa;
 847     return r.ll;
 848 }
 849
 850 uint64_t helper_addt (uint64_t a, uint64_t b)
 851 {
 852     float64 fa, fb, fr;
 853
 854     fa = t_to_float64(a);
 855     fb = t_to_float64(b);
 856     fr = float64_add(fa, fb, &FP_STATUS);
 857     return float64_to_t(fr);
 858 }
 859
 860 uint64_t helper_subt (uint64_t a, uint64_t b)
 861 {
 862     float64 fa, fb, fr;
 863
 864     fa = t_to_float64(a);
 865     fb = t_to_float64(b);
 866     fr = float64_sub(fa, fb, &FP_STATUS);
 867     return float64_to_t(fr);
 868 }
 869
 870 uint64_t helper_mult (uint64_t a, uint64_t b)
 871 {
 872     float64 fa, fb, fr;
 873
 874     fa = t_to_float64(a);
 875     fb = t_to_float64(b);
 876     fr = float64_mul(fa, fb, &FP_STATUS);
 877     return float64_to_t(fr);
 878 }
 879
 880 uint64_t helper_divt (uint64_t a, uint64_t b)
 881 {
 882     float64 fa, fb, fr;
 883
 884     fa = t_to_float64(a);
 885     fb = t_to_float64(b);
 886     fr = float64_div(fa, fb, &FP_STATUS);
 887     return float64_to_t(fr);
 888 }
 889
 890 uint64_t helper_sqrtt (uint64_t a)
 891 {
 892     float64 fa, fr;
 893
 894     fa = t_to_float64(a);
 895     fr = float64_sqrt(fa, &FP_STATUS);
 896     return float64_to_t(fr);
 897 }
 898
 899 /* Comparisons */
 900 uint64_t helper_cmptun (uint64_t a, uint64_t b)
 901 {
 902     float64 fa, fb;
 903
 904     fa = t_to_float64(a);
 905     fb = t_to_float64(b);
 906
 907     if (float64_unordered_quiet(fa, fb, &FP_STATUS)) {
 908         return 0x4000000000000000ULL;
 909     } else {
 910         return 0;
 911     }
 912 }
 913
 914 uint64_t helper_cmpteq(uint64_t a, uint64_t b)
 915 {
 916     float64 fa, fb;
 917
 918     fa = t_to_float64(a);
 919     fb = t_to_float64(b);
 920
 921     if (float64_eq_quiet(fa, fb, &FP_STATUS))
 922         return 0x4000000000000000ULL;
 923     else
 924         return 0;
 925 }
 926
 927 uint64_t helper_cmptle(uint64_t a, uint64_t b)
 928 {
 929     float64 fa, fb;
 930
 931     fa = t_to_float64(a);
 932     fb = t_to_float64(b);
 933
 934     if (float64_le(fa, fb, &FP_STATUS))
 935         return 0x4000000000000000ULL;
 936     else
 937         return 0;
 938 }
 939
 940 uint64_t helper_cmptlt(uint64_t a, uint64_t b)
 941 {
 942     float64 fa, fb;
 943
 944     fa = t_to_float64(a);
 945     fb = t_to_float64(b);
 946
 947     if (float64_lt(fa, fb, &FP_STATUS))
 948         return 0x4000000000000000ULL;
 949     else
 950         return 0;
 951 }
 952
 953 uint64_t helper_cmpgeq(uint64_t a, uint64_t b)
 954 {
 955     float64 fa, fb;
 956
 957     fa = g_to_float64(a);
 958     fb = g_to_float64(b);
 959
 960     if (float64_eq_quiet(fa, fb, &FP_STATUS))
 961         return 0x4000000000000000ULL;
 962     else
 963         return 0;
 964 }
 965
 966 uint64_t helper_cmpgle(uint64_t a, uint64_t b)
 967 {
 968     float64 fa, fb;
 969
 970     fa = g_to_float64(a);
 971     fb = g_to_float64(b);
 972
 973     if (float64_le(fa, fb, &FP_STATUS))
 974         return 0x4000000000000000ULL;
 975     else
 976         return 0;
 977 }
 978
 979 uint64_t helper_cmpglt(uint64_t a, uint64_t b)
 980 {
 981     float64 fa, fb;
 982
 983     fa = g_to_float64(a);
 984     fb = g_to_float64(b);
 985
 986     if (float64_lt(fa, fb, &FP_STATUS))
 987         return 0x4000000000000000ULL;
 988     else
 989         return 0;
 990 }
 991
 992 /* Floating point format conversion */
 993 uint64_t helper_cvtts (uint64_t a)
 994 {
 995     float64 fa;
 996     float32 fr;
 997
 998     fa = t_to_float64(a);
 999     fr = float64_to_float32(fa, &FP_STATUS);
1000     return float32_to_s(fr);
1001 }
1002
1003 uint64_t helper_cvtst (uint64_t a)
1004 {
1005     float32 fa;
1006     float64 fr;
1007
1008     fa = s_to_float32(a);
1009     fr = float32_to_float64(fa, &FP_STATUS);
1010     return float64_to_t(fr);
1011 }
1012
1013 uint64_t helper_cvtqs (uint64_t a)
1014 {
1015     float32 fr = int64_to_float32(a, &FP_STATUS);
1016     return float32_to_s(fr);
1017 }
1018
1019 /* Implement float64 to uint64 conversion without saturation -- we must
1020    supply the truncated result.  This behaviour is used by the compiler
1021    to get unsigned conversion for free with the same instruction.
1022
1023    The VI flag is set when overflow or inexact exceptions should be raised.  */
1024
1025 static inline uint64_t helper_cvttq_internal(uint64_t a, int roundmode, int VI)
1026 {
1027     uint64_t frac, ret = 0;
1028     uint32_t exp, sign, exc = 0;
1029     int shift;
1030
1031     sign = (a >> 63);
1032     exp = (uint32_t)(a >> 52) & 0x7ff;
1033     frac = a & 0xfffffffffffffull;
1034
1035     if (exp == 0) {
1036         if (unlikely(frac != 0)) {
1037             goto do_underflow;
1038         }
1039     } else if (exp == 0x7ff) {
1040         exc = (frac ? float_flag_invalid : VI ? float_flag_overflow : 0);
1041     } else {
1042         /* Restore implicit bit.  */
1043         frac |= 0x10000000000000ull;
1044
1045         shift = exp - 1023 - 52;
1046         if (shift >= 0) {
1047             /* In this case the number is so large that we must shift
1048                the fraction left.  There is no rounding to do.  */
1049             if (shift < 63) {
1050                 ret = frac << shift;
1051                 if (VI && (ret >> shift) != frac) {
1052                     exc = float_flag_overflow;
1053                 }
1054             }
1055         } else {
1056             uint64_t round;
1057
1058             /* In this case the number is smaller than the fraction as
1059                represented by the 52 bit number.  Here we must think
1060                about rounding the result.  Handle this by shifting the
1061                fractional part of the number into the high bits of ROUND.
1062                This will let us efficiently handle round-to-nearest.  */
1063             shift = -shift;
1064             if (shift < 63) {
1065                 ret = frac >> shift;
1066                 round = frac << (64 - shift);
1067             } else {
1068                 /* The exponent is so small we shift out everything.
1069                    Leave a sticky bit for proper rounding below.  */
1070             do_underflow:
1071                 round = 1;
1072             }
1073
1074             if (round) {
1075                 exc = (VI ? float_flag_inexact : 0);
1076                 switch (roundmode) {
1077                 case float_round_nearest_even:
1078                     if (round == (1ull << 63)) {
1079                         /* Fraction is exactly 0.5; round to even.  */
1080                         ret += (ret & 1);
1081                     } else if (round > (1ull << 63)) {
1082                         ret += 1;
1083                     }
1084                     break;
1085                 case float_round_to_zero:
1086                     break;
1087                 case float_round_up:
1088                     ret += 1 - sign;
1089                     break;
1090                 case float_round_down:
1091                     ret += sign;
1092                     break;
1093                 }
1094             }
1095         }
1096         if (sign) {
1097             ret = -ret;
1098         }
1099     }
1100     if (unlikely(exc)) {
1101         float_raise(exc, &FP_STATUS);
1102     }
1103
1104     return ret;
1105 }
1106
1107 uint64_t helper_cvttq(uint64_t a)
1108 {
1109     return helper_cvttq_internal(a, FP_STATUS.float_rounding_mode, 1);
1110 }
1111
1112 uint64_t helper_cvttq_c(uint64_t a)
1113 {
1114     return helper_cvttq_internal(a, float_round_to_zero, 0);
1115 }
1116
1117 uint64_t helper_cvttq_svic(uint64_t a)
1118 {
1119     return helper_cvttq_internal(a, float_round_to_zero, 1);
1120 }
1121
1122 uint64_t helper_cvtqt (uint64_t a)
1123 {
1124     float64 fr = int64_to_float64(a, &FP_STATUS);
1125     return float64_to_t(fr);
1126 }
1127
1128 uint64_t helper_cvtqf (uint64_t a)
1129 {
1130     float32 fr = int64_to_float32(a, &FP_STATUS);
1131     return float32_to_f(fr);
1132 }
1133
1134 uint64_t helper_cvtgf (uint64_t a)
1135 {
1136     float64 fa;
1137     float32 fr;
1138
1139     fa = g_to_float64(a);
1140     fr = float64_to_float32(fa, &FP_STATUS);
1141     return float32_to_f(fr);
1142 }
1143
1144 uint64_t helper_cvtgq (uint64_t a)
1145 {
1146     float64 fa = g_to_float64(a);
1147     return float64_to_int64_round_to_zero(fa, &FP_STATUS);
1148 }
1149
1150 uint64_t helper_cvtqg (uint64_t a)
1151 {
1152     float64 fr;
1153     fr = int64_to_float64(a, &FP_STATUS);
1154     return float64_to_g(fr);
1155 }
1156
1157 /* PALcode support special instructions */
1158 #if !defined (CONFIG_USER_ONLY)
1159 void helper_hw_ret (uint64_t a)
1160 {
1161     env->pc = a & ~3;
1162     env->pal_mode = a & 1;
1163     env->intr_flag = 0;
1164     env->lock_addr = -1;
1165 }
1166 #endif
1167
1168 /*****************************************************************************/
1169 /* Softmmu support */
1170 #if !defined (CONFIG_USER_ONLY)
1171 uint64_t helper_ldl_phys(uint64_t p)
1172 {
1173     return (int32_t)ldl_phys(p);
1174 }
1175
1176 uint64_t helper_ldq_phys(uint64_t p)
1177 {
1178     return ldq_phys(p);
1179 }
1180
1181 uint64_t helper_ldl_l_phys(uint64_t p)
1182 {
1183     env->lock_addr = p;
1184     return env->lock_value = (int32_t)ldl_phys(p);
1185 }
1186
1187 uint64_t helper_ldq_l_phys(uint64_t p)
1188 {
1189     env->lock_addr = p;
1190     return env->lock_value = ldl_phys(p);
1191 }
1192
1193 void helper_stl_phys(uint64_t p, uint64_t v)
1194 {
1195     stl_phys(p, v);
1196 }
1197
1198 void helper_stq_phys(uint64_t p, uint64_t v)
1199 {
1200     stq_phys(p, v);
1201 }
1202
1203 uint64_t helper_stl_c_phys(uint64_t p, uint64_t v)
1204 {
1205     uint64_t ret = 0;
1206
1207     if (p == env->lock_addr) {
1208         int32_t old = ldl_phys(p);
1209         if (old == (int32_t)env->lock_value) {
1210             stl_phys(p, v);
1211             ret = 1;
1212         }
1213     }
1214     env->lock_addr = -1;
1215
1216     return ret;
1217 }
1218
1219 uint64_t helper_stq_c_phys(uint64_t p, uint64_t v)
1220 {
1221     uint64_t ret = 0;
1222
1223     if (p == env->lock_addr) {
1224         uint64_t old = ldq_phys(p);
1225         if (old == env->lock_value) {
1226             stq_phys(p, v);
1227             ret = 1;
1228         }
1229     }
1230     env->lock_addr = -1;
1231
1232     return ret;
1233 }
1234
1235 #define MMUSUFFIX _mmu
1236
1237 #define SHIFT 0
1238 #include "softmmu_template.h"
1239
1240 #define SHIFT 1
1241 #include "softmmu_template.h"
1242
1243 #define SHIFT 2
1244 #include "softmmu_template.h"
1245
1246 #define SHIFT 3
1247 #include "softmmu_template.h"
1248
1249 /* try to fill the TLB and return an exception if error. If retaddr is
1250    NULL, it means that the function was called in C code (i.e. not
1251    from generated code or from helper.c) */
1252 /* XXX: fix it to restore all registers */
1253 void tlb_fill (target_ulong addr, int is_write, int mmu_idx, void *retaddr)
1254 {
1255     TranslationBlock *tb;
1256     CPUState *saved_env;
1257     unsigned long pc;
1258     int ret;
1259
1260     /* XXX: hack to restore env in all cases, even if not called from
1261        generated code */
1262     saved_env = env;
1263     env = cpu_single_env;
1264     ret = cpu_alpha_handle_mmu_fault(env, addr, is_write, mmu_idx, 1);
1265     if (!likely(ret == 0)) {
1266         if (likely(retaddr)) {
1267             /* now we have a real cpu fault */
1268             pc = (unsigned long)retaddr;
1269             tb = tb_find_pc(pc);
1270             if (likely(tb)) {
1271                 /* the PC is inside the translated code. It means that we have
1272                    a virtual CPU fault */
1273                 cpu_restore_state(tb, env, pc);
1274             }
1275         }
1276         /* Exception index and error code are already set */
1277         cpu_loop_exit();
1278     }
1279     env = saved_env;
1280 }
1281
1282 #endif