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
  29 /* This should only be called from translate, via gen_excp.
  30    We expect that ENV->PC has already been updated.  */
  31 void QEMU_NORETURN helper_excp(int excp, int error)
  32 {
  33     env->exception_index = excp;
  34     env->error_code = error;
  35     cpu_loop_exit(env);
  36 }
  37
  38 static void do_restore_state(void *retaddr)
  39 {
  40     unsigned long pc = (unsigned long)retaddr;
  41
  42     if (pc) {
  43         TranslationBlock *tb = tb_find_pc(pc);
  44         if (tb) {
  45             cpu_restore_state(tb, env, pc);
  46         }
  47     }
  48 }
  49
  50 /* This may be called from any of the helpers to set up EXCEPTION_INDEX.  */
  51 static void QEMU_NORETURN dynamic_excp(int excp, int error)
  52 {
  53     env->exception_index = excp;
  54     env->error_code = error;
  55     do_restore_state(GETPC());
  56     cpu_loop_exit(env);
  57 }
  58
  59 static void QEMU_NORETURN arith_excp(int exc, uint64_t mask)
  60 {
  61     env->trap_arg0 = exc;
  62     env->trap_arg1 = mask;
  63     dynamic_excp(EXCP_ARITH, 0);
  64 }
  65
  66 uint64_t helper_load_pcc (void)
  67 {
  68 #ifndef CONFIG_USER_ONLY
  69     /* In system mode we have access to a decent high-resolution clock.
  70        In order to make OS-level time accounting work with the RPCC,
  71        present it with a well-timed clock fixed at 250MHz.  */
  72     return (((uint64_t)env->pcc_ofs << 32)
  73             | (uint32_t)(qemu_get_clock_ns(vm_clock) >> 2));
  74 #else
  75     /* In user-mode, vm_clock doesn't exist.  Just pass through the host cpu
  76        clock ticks.  Also, don't bother taking PCC_OFS into account.  */
  77     return (uint32_t)cpu_get_real_ticks();
  78 #endif
  79 }
  80
  81 uint64_t helper_load_fpcr (void)
  82 {
  83     return cpu_alpha_load_fpcr (env);
  84 }
  85
  86 void helper_store_fpcr (uint64_t val)
  87 {
  88     cpu_alpha_store_fpcr (env, val);
  89 }
  90
  91 uint64_t helper_addqv (uint64_t op1, uint64_t op2)
  92 {
  93     uint64_t tmp = op1;
  94     op1 += op2;
  95     if (unlikely((tmp ^ op2 ^ (-1ULL)) & (tmp ^ op1) & (1ULL << 63))) {
  96         arith_excp(EXC_M_IOV, 0);
  97     }
  98     return op1;
  99 }
 100
 101 uint64_t helper_addlv (uint64_t op1, uint64_t op2)
 102 {
 103     uint64_t tmp = op1;
 104     op1 = (uint32_t)(op1 + op2);
 105     if (unlikely((tmp ^ op2 ^ (-1UL)) & (tmp ^ op1) & (1UL << 31))) {
 106         arith_excp(EXC_M_IOV, 0);
 107     }
 108     return op1;
 109 }
 110
 111 uint64_t helper_subqv (uint64_t op1, uint64_t op2)
 112 {
 113     uint64_t res;
 114     res = op1 - op2;
 115     if (unlikely((op1 ^ op2) & (res ^ op1) & (1ULL << 63))) {
 116         arith_excp(EXC_M_IOV, 0);
 117     }
 118     return res;
 119 }
 120
 121 uint64_t helper_sublv (uint64_t op1, uint64_t op2)
 122 {
 123     uint32_t res;
 124     res = op1 - op2;
 125     if (unlikely((op1 ^ op2) & (res ^ op1) & (1UL << 31))) {
 126         arith_excp(EXC_M_IOV, 0);
 127     }
 128     return res;
 129 }
 130
 131 uint64_t helper_mullv (uint64_t op1, uint64_t op2)
 132 {
 133     int64_t res = (int64_t)op1 * (int64_t)op2;
 134
 135     if (unlikely((int32_t)res != res)) {
 136         arith_excp(EXC_M_IOV, 0);
 137     }
 138     return (int64_t)((int32_t)res);
 139 }
 140
 141 uint64_t helper_mulqv (uint64_t op1, uint64_t op2)
 142 {
 143     uint64_t tl, th;
 144
 145     muls64(&tl, &th, op1, op2);
 146     /* If th != 0 && th != -1, then we had an overflow */
 147     if (unlikely((th + 1) > 1)) {
 148         arith_excp(EXC_M_IOV, 0);
 149     }
 150     return tl;
 151 }
 152
 153 uint64_t helper_umulh (uint64_t op1, uint64_t op2)
 154 {
 155     uint64_t tl, th;
 156
 157     mulu64(&tl, &th, op1, op2);
 158     return th;
 159 }
 160
 161 uint64_t helper_ctpop (uint64_t arg)
 162 {
 163     return ctpop64(arg);
 164 }
 165
 166 uint64_t helper_ctlz (uint64_t arg)
 167 {
 168     return clz64(arg);
 169 }
 170
 171 uint64_t helper_cttz (uint64_t arg)
 172 {
 173     return ctz64(arg);
 174 }
 175
 176 static inline uint64_t byte_zap(uint64_t op, uint8_t mskb)
 177 {
 178     uint64_t mask;
 179
 180     mask = 0;
 181     mask |= ((mskb >> 0) & 1) * 0x00000000000000FFULL;
 182     mask |= ((mskb >> 1) & 1) * 0x000000000000FF00ULL;
 183     mask |= ((mskb >> 2) & 1) * 0x0000000000FF0000ULL;
 184     mask |= ((mskb >> 3) & 1) * 0x00000000FF000000ULL;
 185     mask |= ((mskb >> 4) & 1) * 0x000000FF00000000ULL;
 186     mask |= ((mskb >> 5) & 1) * 0x0000FF0000000000ULL;
 187     mask |= ((mskb >> 6) & 1) * 0x00FF000000000000ULL;
 188     mask |= ((mskb >> 7) & 1) * 0xFF00000000000000ULL;
 189
 190     return op & ~mask;
 191 }
 192
 193 uint64_t helper_zap(uint64_t val, uint64_t mask)
 194 {
 195     return byte_zap(val, mask);
 196 }
 197
 198 uint64_t helper_zapnot(uint64_t val, uint64_t mask)
 199 {
 200     return byte_zap(val, ~mask);
 201 }
 202
 203 uint64_t helper_cmpbge (uint64_t op1, uint64_t op2)
 204 {
 205     uint8_t opa, opb, res;
 206     int i;
 207
 208     res = 0;
 209     for (i = 0; i < 8; i++) {
 210         opa = op1 >> (i * 8);
 211         opb = op2 >> (i * 8);
 212         if (opa >= opb)
 213             res |= 1 << i;
 214     }
 215     return res;
 216 }
 217
 218 uint64_t helper_minub8 (uint64_t op1, uint64_t op2)
 219 {
 220     uint64_t res = 0;
 221     uint8_t opa, opb, opr;
 222     int i;
 223
 224     for (i = 0; i < 8; ++i) {
 225         opa = op1 >> (i * 8);
 226         opb = op2 >> (i * 8);
 227         opr = opa < opb ? opa : opb;
 228         res |= (uint64_t)opr << (i * 8);
 229     }
 230     return res;
 231 }
 232
 233 uint64_t helper_minsb8 (uint64_t op1, uint64_t op2)
 234 {
 235     uint64_t res = 0;
 236     int8_t opa, opb;
 237     uint8_t opr;
 238     int i;
 239
 240     for (i = 0; i < 8; ++i) {
 241         opa = op1 >> (i * 8);
 242         opb = op2 >> (i * 8);
 243         opr = opa < opb ? opa : opb;
 244         res |= (uint64_t)opr << (i * 8);
 245     }
 246     return res;
 247 }
 248
 249 uint64_t helper_minuw4 (uint64_t op1, uint64_t op2)
 250 {
 251     uint64_t res = 0;
 252     uint16_t opa, opb, opr;
 253     int i;
 254
 255     for (i = 0; i < 4; ++i) {
 256         opa = op1 >> (i * 16);
 257         opb = op2 >> (i * 16);
 258         opr = opa < opb ? opa : opb;
 259         res |= (uint64_t)opr << (i * 16);
 260     }
 261     return res;
 262 }
 263
 264 uint64_t helper_minsw4 (uint64_t op1, uint64_t op2)
 265 {
 266     uint64_t res = 0;
 267     int16_t opa, opb;
 268     uint16_t opr;
 269     int i;
 270
 271     for (i = 0; i < 4; ++i) {
 272         opa = op1 >> (i * 16);
 273         opb = op2 >> (i * 16);
 274         opr = opa < opb ? opa : opb;
 275         res |= (uint64_t)opr << (i * 16);
 276     }
 277     return res;
 278 }
 279
 280 uint64_t helper_maxub8 (uint64_t op1, uint64_t op2)
 281 {
 282     uint64_t res = 0;
 283     uint8_t opa, opb, opr;
 284     int i;
 285
 286     for (i = 0; i < 8; ++i) {
 287         opa = op1 >> (i * 8);
 288         opb = op2 >> (i * 8);
 289         opr = opa > opb ? opa : opb;
 290         res |= (uint64_t)opr << (i * 8);
 291     }
 292     return res;
 293 }
 294
 295 uint64_t helper_maxsb8 (uint64_t op1, uint64_t op2)
 296 {
 297     uint64_t res = 0;
 298     int8_t opa, opb;
 299     uint8_t opr;
 300     int i;
 301
 302     for (i = 0; i < 8; ++i) {
 303         opa = op1 >> (i * 8);
 304         opb = op2 >> (i * 8);
 305         opr = opa > opb ? opa : opb;
 306         res |= (uint64_t)opr << (i * 8);
 307     }
 308     return res;
 309 }
 310
 311 uint64_t helper_maxuw4 (uint64_t op1, uint64_t op2)
 312 {
 313     uint64_t res = 0;
 314     uint16_t opa, opb, opr;
 315     int i;
 316
 317     for (i = 0; i < 4; ++i) {
 318         opa = op1 >> (i * 16);
 319         opb = op2 >> (i * 16);
 320         opr = opa > opb ? opa : opb;
 321         res |= (uint64_t)opr << (i * 16);
 322     }
 323     return res;
 324 }
 325
 326 uint64_t helper_maxsw4 (uint64_t op1, uint64_t op2)
 327 {
 328     uint64_t res = 0;
 329     int16_t opa, opb;
 330     uint16_t opr;
 331     int i;
 332
 333     for (i = 0; i < 4; ++i) {
 334         opa = op1 >> (i * 16);
 335         opb = op2 >> (i * 16);
 336         opr = opa > opb ? opa : opb;
 337         res |= (uint64_t)opr << (i * 16);
 338     }
 339     return res;
 340 }
 341
 342 uint64_t helper_perr (uint64_t op1, uint64_t op2)
 343 {
 344     uint64_t res = 0;
 345     uint8_t opa, opb, opr;
 346     int i;
 347
 348     for (i = 0; i < 8; ++i) {
 349         opa = op1 >> (i * 8);
 350         opb = op2 >> (i * 8);
 351         if (opa >= opb)
 352             opr = opa - opb;
 353         else
 354             opr = opb - opa;
 355         res += opr;
 356     }
 357     return res;
 358 }
 359
 360 uint64_t helper_pklb (uint64_t op1)
 361 {
 362     return (op1 & 0xff) | ((op1 >> 24) & 0xff00);
 363 }
 364
 365 uint64_t helper_pkwb (uint64_t op1)
 366 {
 367     return ((op1 & 0xff)
 368             | ((op1 >> 8) & 0xff00)
 369             | ((op1 >> 16) & 0xff0000)
 370             | ((op1 >> 24) & 0xff000000));
 371 }
 372
 373 uint64_t helper_unpkbl (uint64_t op1)
 374 {
 375     return (op1 & 0xff) | ((op1 & 0xff00) << 24);
 376 }
 377
 378 uint64_t helper_unpkbw (uint64_t op1)
 379 {
 380     return ((op1 & 0xff)
 381             | ((op1 & 0xff00) << 8)
 382             | ((op1 & 0xff0000) << 16)
 383             | ((op1 & 0xff000000) << 24));
 384 }
 385
 386 /* Floating point helpers */
 387
 388 void helper_setroundmode (uint32_t val)
 389 {
 390     set_float_rounding_mode(val, &FP_STATUS);
 391 }
 392
 393 void helper_setflushzero (uint32_t val)
 394 {
 395     set_flush_to_zero(val, &FP_STATUS);
 396 }
 397
 398 void helper_fp_exc_clear (void)
 399 {
 400     set_float_exception_flags(0, &FP_STATUS);
 401 }
 402
 403 uint32_t helper_fp_exc_get (void)
 404 {
 405     return get_float_exception_flags(&FP_STATUS);
 406 }
 407
 408 /* Raise exceptions for ieee fp insns without software completion.
 409    In that case there are no exceptions that don't trap; the mask
 410    doesn't apply.  */
 411 void helper_fp_exc_raise(uint32_t exc, uint32_t regno)
 412 {
 413     if (exc) {
 414         uint32_t hw_exc = 0;
 415
 416         if (exc & float_flag_invalid) {
 417             hw_exc |= EXC_M_INV;
 418         }
 419         if (exc & float_flag_divbyzero) {
 420             hw_exc |= EXC_M_DZE;
 421         }
 422         if (exc & float_flag_overflow) {
 423             hw_exc |= EXC_M_FOV;
 424         }
 425         if (exc & float_flag_underflow) {
 426             hw_exc |= EXC_M_UNF;
 427         }
 428         if (exc & float_flag_inexact) {
 429             hw_exc |= EXC_M_INE;
 430         }
 431
 432         arith_excp(hw_exc, 1ull << regno);
 433     }
 434 }
 435
 436 /* Raise exceptions for ieee fp insns with software completion.  */
 437 void helper_fp_exc_raise_s(uint32_t exc, uint32_t regno)
 438 {
 439     if (exc) {
 440         env->fpcr_exc_status |= exc;
 441
 442         exc &= ~env->fpcr_exc_mask;
 443         if (exc) {
 444             helper_fp_exc_raise(exc, regno);
 445         }
 446     }
 447 }
 448
 449 /* Input remapping without software completion.  Handle denormal-map-to-zero
 450    and trap for all other non-finite numbers.  */
 451 uint64_t helper_ieee_input(uint64_t val)
 452 {
 453     uint32_t exp = (uint32_t)(val >> 52) & 0x7ff;
 454     uint64_t frac = val & 0xfffffffffffffull;
 455
 456     if (exp == 0) {
 457         if (frac != 0) {
 458             /* If DNZ is set flush denormals to zero on input.  */
 459             if (env->fpcr_dnz) {
 460                 val &= 1ull << 63;
 461             } else {
 462                 arith_excp(EXC_M_UNF, 0);
 463             }
 464         }
 465     } else if (exp == 0x7ff) {
 466         /* Infinity or NaN.  */
 467         /* ??? I'm not sure these exception bit flags are correct.  I do
 468            know that the Linux kernel, at least, doesn't rely on them and
 469            just emulates the insn to figure out what exception to use.  */
 470         arith_excp(frac ? EXC_M_INV : EXC_M_FOV, 0);
 471     }
 472     return val;
 473 }
 474
 475 /* Similar, but does not trap for infinities.  Used for comparisons.  */
 476 uint64_t helper_ieee_input_cmp(uint64_t val)
 477 {
 478     uint32_t exp = (uint32_t)(val >> 52) & 0x7ff;
 479     uint64_t frac = val & 0xfffffffffffffull;
 480
 481     if (exp == 0) {
 482         if (frac != 0) {
 483             /* If DNZ is set flush denormals to zero on input.  */
 484             if (env->fpcr_dnz) {
 485                 val &= 1ull << 63;
 486             } else {
 487                 arith_excp(EXC_M_UNF, 0);
 488             }
 489         }
 490     } else if (exp == 0x7ff && frac) {
 491         /* NaN.  */
 492         arith_excp(EXC_M_INV, 0);
 493     }
 494     return val;
 495 }
 496
 497 /* Input remapping with software completion enabled.  All we have to do
 498    is handle denormal-map-to-zero; all other inputs get exceptions as
 499    needed from the actual operation.  */
 500 uint64_t helper_ieee_input_s(uint64_t val)
 501 {
 502     if (env->fpcr_dnz) {
 503         uint32_t exp = (uint32_t)(val >> 52) & 0x7ff;
 504         if (exp == 0) {
 505             val &= 1ull << 63;
 506         }
 507     }
 508     return val;
 509 }
 510
 511 /* F floating (VAX) */
 512 static inline uint64_t float32_to_f(float32 fa)
 513 {
 514     uint64_t r, exp, mant, sig;
 515     CPU_FloatU a;
 516
 517     a.f = fa;
 518     sig = ((uint64_t)a.l & 0x80000000) << 32;
 519     exp = (a.l >> 23) & 0xff;
 520     mant = ((uint64_t)a.l & 0x007fffff) << 29;
 521
 522     if (exp == 255) {
 523         /* NaN or infinity */
 524         r = 1; /* VAX dirty zero */
 525     } else if (exp == 0) {
 526         if (mant == 0) {
 527             /* Zero */
 528             r = 0;
 529         } else {
 530             /* Denormalized */
 531             r = sig | ((exp + 1) << 52) | mant;
 532         }
 533     } else {
 534         if (exp >= 253) {
 535             /* Overflow */
 536             r = 1; /* VAX dirty zero */
 537         } else {
 538             r = sig | ((exp + 2) << 52);
 539         }
 540     }
 541
 542     return r;
 543 }
 544
 545 static inline float32 f_to_float32(uint64_t a)
 546 {
 547     uint32_t exp, mant_sig;
 548     CPU_FloatU r;
 549
 550     exp = ((a >> 55) & 0x80) | ((a >> 52) & 0x7f);
 551     mant_sig = ((a >> 32) & 0x80000000) | ((a >> 29) & 0x007fffff);
 552
 553     if (unlikely(!exp && mant_sig)) {
 554         /* Reserved operands / Dirty zero */
 555         dynamic_excp(EXCP_OPCDEC, 0);
 556     }
 557
 558     if (exp < 3) {
 559         /* Underflow */
 560         r.l = 0;
 561     } else {
 562         r.l = ((exp - 2) << 23) | mant_sig;
 563     }
 564
 565     return r.f;
 566 }
 567
 568 uint32_t helper_f_to_memory (uint64_t a)
 569 {
 570     uint32_t r;
 571     r =  (a & 0x00001fffe0000000ull) >> 13;
 572     r |= (a & 0x07ffe00000000000ull) >> 45;
 573     r |= (a & 0xc000000000000000ull) >> 48;
 574     return r;
 575 }
 576
 577 uint64_t helper_memory_to_f (uint32_t a)
 578 {
 579     uint64_t r;
 580     r =  ((uint64_t)(a & 0x0000c000)) << 48;
 581     r |= ((uint64_t)(a & 0x003fffff)) << 45;
 582     r |= ((uint64_t)(a & 0xffff0000)) << 13;
 583     if (!(a & 0x00004000))
 584         r |= 0x7ll << 59;
 585     return r;
 586 }
 587
 588 /* ??? Emulating VAX arithmetic with IEEE arithmetic is wrong.  We should
 589    either implement VAX arithmetic properly or just signal invalid opcode.  */
 590
 591 uint64_t helper_addf (uint64_t a, uint64_t b)
 592 {
 593     float32 fa, fb, fr;
 594
 595     fa = f_to_float32(a);
 596     fb = f_to_float32(b);
 597     fr = float32_add(fa, fb, &FP_STATUS);
 598     return float32_to_f(fr);
 599 }
 600
 601 uint64_t helper_subf (uint64_t a, uint64_t b)
 602 {
 603     float32 fa, fb, fr;
 604
 605     fa = f_to_float32(a);
 606     fb = f_to_float32(b);
 607     fr = float32_sub(fa, fb, &FP_STATUS);
 608     return float32_to_f(fr);
 609 }
 610
 611 uint64_t helper_mulf (uint64_t a, uint64_t b)
 612 {
 613     float32 fa, fb, fr;
 614
 615     fa = f_to_float32(a);
 616     fb = f_to_float32(b);
 617     fr = float32_mul(fa, fb, &FP_STATUS);
 618     return float32_to_f(fr);
 619 }
 620
 621 uint64_t helper_divf (uint64_t a, uint64_t b)
 622 {
 623     float32 fa, fb, fr;
 624
 625     fa = f_to_float32(a);
 626     fb = f_to_float32(b);
 627     fr = float32_div(fa, fb, &FP_STATUS);
 628     return float32_to_f(fr);
 629 }
 630
 631 uint64_t helper_sqrtf (uint64_t t)
 632 {
 633     float32 ft, fr;
 634
 635     ft = f_to_float32(t);
 636     fr = float32_sqrt(ft, &FP_STATUS);
 637     return float32_to_f(fr);
 638 }
 639
 640
 641 /* G floating (VAX) */
 642 static inline uint64_t float64_to_g(float64 fa)
 643 {
 644     uint64_t r, exp, mant, sig;
 645     CPU_DoubleU a;
 646
 647     a.d = fa;
 648     sig = a.ll & 0x8000000000000000ull;
 649     exp = (a.ll >> 52) & 0x7ff;
 650     mant = a.ll & 0x000fffffffffffffull;
 651
 652     if (exp == 2047) {
 653         /* NaN or infinity */
 654         r = 1; /* VAX dirty zero */
 655     } else if (exp == 0) {
 656         if (mant == 0) {
 657             /* Zero */
 658             r = 0;
 659         } else {
 660             /* Denormalized */
 661             r = sig | ((exp + 1) << 52) | mant;
 662         }
 663     } else {
 664         if (exp >= 2045) {
 665             /* Overflow */
 666             r = 1; /* VAX dirty zero */
 667         } else {
 668             r = sig | ((exp + 2) << 52);
 669         }
 670     }
 671
 672     return r;
 673 }
 674
 675 static inline float64 g_to_float64(uint64_t a)
 676 {
 677     uint64_t exp, mant_sig;
 678     CPU_DoubleU r;
 679
 680     exp = (a >> 52) & 0x7ff;
 681     mant_sig = a & 0x800fffffffffffffull;
 682
 683     if (!exp && mant_sig) {
 684         /* Reserved operands / Dirty zero */
 685         dynamic_excp(EXCP_OPCDEC, 0);
 686     }
 687
 688     if (exp < 3) {
 689         /* Underflow */
 690         r.ll = 0;
 691     } else {
 692         r.ll = ((exp - 2) << 52) | mant_sig;
 693     }
 694
 695     return r.d;
 696 }
 697
 698 uint64_t helper_g_to_memory (uint64_t a)
 699 {
 700     uint64_t r;
 701     r =  (a & 0x000000000000ffffull) << 48;
 702     r |= (a & 0x00000000ffff0000ull) << 16;
 703     r |= (a & 0x0000ffff00000000ull) >> 16;
 704     r |= (a & 0xffff000000000000ull) >> 48;
 705     return r;
 706 }
 707
 708 uint64_t helper_memory_to_g (uint64_t a)
 709 {
 710     uint64_t r;
 711     r =  (a & 0x000000000000ffffull) << 48;
 712     r |= (a & 0x00000000ffff0000ull) << 16;
 713     r |= (a & 0x0000ffff00000000ull) >> 16;
 714     r |= (a & 0xffff000000000000ull) >> 48;
 715     return r;
 716 }
 717
 718 uint64_t helper_addg (uint64_t a, uint64_t b)
 719 {
 720     float64 fa, fb, fr;
 721
 722     fa = g_to_float64(a);
 723     fb = g_to_float64(b);
 724     fr = float64_add(fa, fb, &FP_STATUS);
 725     return float64_to_g(fr);
 726 }
 727
 728 uint64_t helper_subg (uint64_t a, uint64_t b)
 729 {
 730     float64 fa, fb, fr;
 731
 732     fa = g_to_float64(a);
 733     fb = g_to_float64(b);
 734     fr = float64_sub(fa, fb, &FP_STATUS);
 735     return float64_to_g(fr);
 736 }
 737
 738 uint64_t helper_mulg (uint64_t a, uint64_t b)
 739 {
 740     float64 fa, fb, fr;
 741
 742     fa = g_to_float64(a);
 743     fb = g_to_float64(b);
 744     fr = float64_mul(fa, fb, &FP_STATUS);
 745     return float64_to_g(fr);
 746 }
 747
 748 uint64_t helper_divg (uint64_t a, uint64_t b)
 749 {
 750     float64 fa, fb, fr;
 751
 752     fa = g_to_float64(a);
 753     fb = g_to_float64(b);
 754     fr = float64_div(fa, fb, &FP_STATUS);
 755     return float64_to_g(fr);
 756 }
 757
 758 uint64_t helper_sqrtg (uint64_t a)
 759 {
 760     float64 fa, fr;
 761
 762     fa = g_to_float64(a);
 763     fr = float64_sqrt(fa, &FP_STATUS);
 764     return float64_to_g(fr);
 765 }
 766
 767
 768 /* S floating (single) */
 769
 770 /* Taken from linux/arch/alpha/kernel/traps.c, s_mem_to_reg.  */
 771 static inline uint64_t float32_to_s_int(uint32_t fi)
 772 {
 773     uint32_t frac = fi & 0x7fffff;
 774     uint32_t sign = fi >> 31;
 775     uint32_t exp_msb = (fi >> 30) & 1;
 776     uint32_t exp_low = (fi >> 23) & 0x7f;
 777     uint32_t exp;
 778
 779     exp = (exp_msb << 10) | exp_low;
 780     if (exp_msb) {
 781         if (exp_low == 0x7f)
 782             exp = 0x7ff;
 783     } else {
 784         if (exp_low != 0x00)
 785             exp |= 0x380;
 786     }
 787
 788     return (((uint64_t)sign << 63)
 789             | ((uint64_t)exp << 52)
 790             | ((uint64_t)frac << 29));
 791 }
 792
 793 static inline uint64_t float32_to_s(float32 fa)
 794 {
 795     CPU_FloatU a;
 796     a.f = fa;
 797     return float32_to_s_int(a.l);
 798 }
 799
 800 static inline uint32_t s_to_float32_int(uint64_t a)
 801 {
 802     return ((a >> 32) & 0xc0000000) | ((a >> 29) & 0x3fffffff);
 803 }
 804
 805 static inline float32 s_to_float32(uint64_t a)
 806 {
 807     CPU_FloatU r;
 808     r.l = s_to_float32_int(a);
 809     return r.f;
 810 }
 811
 812 uint32_t helper_s_to_memory (uint64_t a)
 813 {
 814     return s_to_float32_int(a);
 815 }
 816
 817 uint64_t helper_memory_to_s (uint32_t a)
 818 {
 819     return float32_to_s_int(a);
 820 }
 821
 822 uint64_t helper_adds (uint64_t a, uint64_t b)
 823 {
 824     float32 fa, fb, fr;
 825
 826     fa = s_to_float32(a);
 827     fb = s_to_float32(b);
 828     fr = float32_add(fa, fb, &FP_STATUS);
 829     return float32_to_s(fr);
 830 }
 831
 832 uint64_t helper_subs (uint64_t a, uint64_t b)
 833 {
 834     float32 fa, fb, fr;
 835
 836     fa = s_to_float32(a);
 837     fb = s_to_float32(b);
 838     fr = float32_sub(fa, fb, &FP_STATUS);
 839     return float32_to_s(fr);
 840 }
 841
 842 uint64_t helper_muls (uint64_t a, uint64_t b)
 843 {
 844     float32 fa, fb, fr;
 845
 846     fa = s_to_float32(a);
 847     fb = s_to_float32(b);
 848     fr = float32_mul(fa, fb, &FP_STATUS);
 849     return float32_to_s(fr);
 850 }
 851
 852 uint64_t helper_divs (uint64_t a, uint64_t b)
 853 {
 854     float32 fa, fb, fr;
 855
 856     fa = s_to_float32(a);
 857     fb = s_to_float32(b);
 858     fr = float32_div(fa, fb, &FP_STATUS);
 859     return float32_to_s(fr);
 860 }
 861
 862 uint64_t helper_sqrts (uint64_t a)
 863 {
 864     float32 fa, fr;
 865
 866     fa = s_to_float32(a);
 867     fr = float32_sqrt(fa, &FP_STATUS);
 868     return float32_to_s(fr);
 869 }
 870
 871
 872 /* T floating (double) */
 873 static inline float64 t_to_float64(uint64_t a)
 874 {
 875     /* Memory format is the same as float64 */
 876     CPU_DoubleU r;
 877     r.ll = a;
 878     return r.d;
 879 }
 880
 881 static inline uint64_t float64_to_t(float64 fa)
 882 {
 883     /* Memory format is the same as float64 */
 884     CPU_DoubleU r;
 885     r.d = fa;
 886     return r.ll;
 887 }
 888
 889 uint64_t helper_addt (uint64_t a, uint64_t b)
 890 {
 891     float64 fa, fb, fr;
 892
 893     fa = t_to_float64(a);
 894     fb = t_to_float64(b);
 895     fr = float64_add(fa, fb, &FP_STATUS);
 896     return float64_to_t(fr);
 897 }
 898
 899 uint64_t helper_subt (uint64_t a, uint64_t b)
 900 {
 901     float64 fa, fb, fr;
 902
 903     fa = t_to_float64(a);
 904     fb = t_to_float64(b);
 905     fr = float64_sub(fa, fb, &FP_STATUS);
 906     return float64_to_t(fr);
 907 }
 908
 909 uint64_t helper_mult (uint64_t a, uint64_t b)
 910 {
 911     float64 fa, fb, fr;
 912
 913     fa = t_to_float64(a);
 914     fb = t_to_float64(b);
 915     fr = float64_mul(fa, fb, &FP_STATUS);
 916     return float64_to_t(fr);
 917 }
 918
 919 uint64_t helper_divt (uint64_t a, uint64_t b)
 920 {
 921     float64 fa, fb, fr;
 922
 923     fa = t_to_float64(a);
 924     fb = t_to_float64(b);
 925     fr = float64_div(fa, fb, &FP_STATUS);
 926     return float64_to_t(fr);
 927 }
 928
 929 uint64_t helper_sqrtt (uint64_t a)
 930 {
 931     float64 fa, fr;
 932
 933     fa = t_to_float64(a);
 934     fr = float64_sqrt(fa, &FP_STATUS);
 935     return float64_to_t(fr);
 936 }
 937
 938 /* Comparisons */
 939 uint64_t helper_cmptun (uint64_t a, uint64_t b)
 940 {
 941     float64 fa, fb;
 942
 943     fa = t_to_float64(a);
 944     fb = t_to_float64(b);
 945
 946     if (float64_unordered_quiet(fa, fb, &FP_STATUS)) {
 947         return 0x4000000000000000ULL;
 948     } else {
 949         return 0;
 950     }
 951 }
 952
 953 uint64_t helper_cmpteq(uint64_t a, uint64_t b)
 954 {
 955     float64 fa, fb;
 956
 957     fa = t_to_float64(a);
 958     fb = t_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_cmptle(uint64_t a, uint64_t b)
 967 {
 968     float64 fa, fb;
 969
 970     fa = t_to_float64(a);
 971     fb = t_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_cmptlt(uint64_t a, uint64_t b)
 980 {
 981     float64 fa, fb;
 982
 983     fa = t_to_float64(a);
 984     fb = t_to_float64(b);
 985
 986     if (float64_lt(fa, fb, &FP_STATUS))
 987         return 0x4000000000000000ULL;
 988     else
 989         return 0;
 990 }
 991
 992 uint64_t helper_cmpgeq(uint64_t a, uint64_t b)
 993 {
 994     float64 fa, fb;
 995
 996     fa = g_to_float64(a);
 997     fb = g_to_float64(b);
 998
 999     if (float64_eq_quiet(fa, fb, &FP_STATUS))
1000         return 0x4000000000000000ULL;
1001     else
1002         return 0;
1003 }
1004
1005 uint64_t helper_cmpgle(uint64_t a, uint64_t b)
1006 {
1007     float64 fa, fb;
1008
1009     fa = g_to_float64(a);
1010     fb = g_to_float64(b);
1011
1012     if (float64_le(fa, fb, &FP_STATUS))
1013         return 0x4000000000000000ULL;
1014     else
1015         return 0;
1016 }
1017
1018 uint64_t helper_cmpglt(uint64_t a, uint64_t b)
1019 {
1020     float64 fa, fb;
1021
1022     fa = g_to_float64(a);
1023     fb = g_to_float64(b);
1024
1025     if (float64_lt(fa, fb, &FP_STATUS))
1026         return 0x4000000000000000ULL;
1027     else
1028         return 0;
1029 }
1030
1031 /* Floating point format conversion */
1032 uint64_t helper_cvtts (uint64_t a)
1033 {
1034     float64 fa;
1035     float32 fr;
1036
1037     fa = t_to_float64(a);
1038     fr = float64_to_float32(fa, &FP_STATUS);
1039     return float32_to_s(fr);
1040 }
1041
1042 uint64_t helper_cvtst (uint64_t a)
1043 {
1044     float32 fa;
1045     float64 fr;
1046
1047     fa = s_to_float32(a);
1048     fr = float32_to_float64(fa, &FP_STATUS);
1049     return float64_to_t(fr);
1050 }
1051
1052 uint64_t helper_cvtqs (uint64_t a)
1053 {
1054     float32 fr = int64_to_float32(a, &FP_STATUS);
1055     return float32_to_s(fr);
1056 }
1057
1058 /* Implement float64 to uint64 conversion without saturation -- we must
1059    supply the truncated result.  This behaviour is used by the compiler
1060    to get unsigned conversion for free with the same instruction.
1061
1062    The VI flag is set when overflow or inexact exceptions should be raised.  */
1063
1064 static inline uint64_t helper_cvttq_internal(uint64_t a, int roundmode, int VI)
1065 {
1066     uint64_t frac, ret = 0;
1067     uint32_t exp, sign, exc = 0;
1068     int shift;
1069
1070     sign = (a >> 63);
1071     exp = (uint32_t)(a >> 52) & 0x7ff;
1072     frac = a & 0xfffffffffffffull;
1073
1074     if (exp == 0) {
1075         if (unlikely(frac != 0)) {
1076             goto do_underflow;
1077         }
1078     } else if (exp == 0x7ff) {
1079         exc = (frac ? float_flag_invalid : VI ? float_flag_overflow : 0);
1080     } else {
1081         /* Restore implicit bit.  */
1082         frac |= 0x10000000000000ull;
1083
1084         shift = exp - 1023 - 52;
1085         if (shift >= 0) {
1086             /* In this case the number is so large that we must shift
1087                the fraction left.  There is no rounding to do.  */
1088             if (shift < 63) {
1089                 ret = frac << shift;
1090                 if (VI && (ret >> shift) != frac) {
1091                     exc = float_flag_overflow;
1092                 }
1093             }
1094         } else {
1095             uint64_t round;
1096
1097             /* In this case the number is smaller than the fraction as
1098                represented by the 52 bit number.  Here we must think
1099                about rounding the result.  Handle this by shifting the
1100                fractional part of the number into the high bits of ROUND.
1101                This will let us efficiently handle round-to-nearest.  */
1102             shift = -shift;
1103             if (shift < 63) {
1104                 ret = frac >> shift;
1105                 round = frac << (64 - shift);
1106             } else {
1107                 /* The exponent is so small we shift out everything.
1108                    Leave a sticky bit for proper rounding below.  */
1109             do_underflow:
1110                 round = 1;
1111             }
1112
1113             if (round) {
1114                 exc = (VI ? float_flag_inexact : 0);
1115                 switch (roundmode) {
1116                 case float_round_nearest_even:
1117                     if (round == (1ull << 63)) {
1118                         /* Fraction is exactly 0.5; round to even.  */
1119                         ret += (ret & 1);
1120                     } else if (round > (1ull << 63)) {
1121                         ret += 1;
1122                     }
1123                     break;
1124                 case float_round_to_zero:
1125                     break;
1126                 case float_round_up:
1127                     ret += 1 - sign;
1128                     break;
1129                 case float_round_down:
1130                     ret += sign;
1131                     break;
1132                 }
1133             }
1134         }
1135         if (sign) {
1136             ret = -ret;
1137         }
1138     }
1139     if (unlikely(exc)) {
1140         float_raise(exc, &FP_STATUS);
1141     }
1142
1143     return ret;
1144 }
1145
1146 uint64_t helper_cvttq(uint64_t a)
1147 {
1148     return helper_cvttq_internal(a, FP_STATUS.float_rounding_mode, 1);
1149 }
1150
1151 uint64_t helper_cvttq_c(uint64_t a)
1152 {
1153     return helper_cvttq_internal(a, float_round_to_zero, 0);
1154 }
1155
1156 uint64_t helper_cvttq_svic(uint64_t a)
1157 {
1158     return helper_cvttq_internal(a, float_round_to_zero, 1);
1159 }
1160
1161 uint64_t helper_cvtqt (uint64_t a)
1162 {
1163     float64 fr = int64_to_float64(a, &FP_STATUS);
1164     return float64_to_t(fr);
1165 }
1166
1167 uint64_t helper_cvtqf (uint64_t a)
1168 {
1169     float32 fr = int64_to_float32(a, &FP_STATUS);
1170     return float32_to_f(fr);
1171 }
1172
1173 uint64_t helper_cvtgf (uint64_t a)
1174 {
1175     float64 fa;
1176     float32 fr;
1177
1178     fa = g_to_float64(a);
1179     fr = float64_to_float32(fa, &FP_STATUS);
1180     return float32_to_f(fr);
1181 }
1182
1183 uint64_t helper_cvtgq (uint64_t a)
1184 {
1185     float64 fa = g_to_float64(a);
1186     return float64_to_int64_round_to_zero(fa, &FP_STATUS);
1187 }
1188
1189 uint64_t helper_cvtqg (uint64_t a)
1190 {
1191     float64 fr;
1192     fr = int64_to_float64(a, &FP_STATUS);
1193     return float64_to_g(fr);
1194 }
1195
1196 /* PALcode support special instructions */
1197 #if !defined (CONFIG_USER_ONLY)
1198 void helper_hw_ret (uint64_t a)
1199 {
1200     env->pc = a & ~3;
1201     env->intr_flag = 0;
1202     env->lock_addr = -1;
1203     if ((a & 1) == 0) {
1204         env->pal_mode = 0;
1205         swap_shadow_regs(env);
1206     }
1207 }
1208
1209 void helper_tbia(void)
1210 {
1211     tlb_flush(env, 1);
1212 }
1213
1214 void helper_tbis(uint64_t p)
1215 {
1216     tlb_flush_page(env, p);
1217 }
1218 #endif
1219
1220 /*****************************************************************************/
1221 /* Softmmu support */
1222 #if !defined (CONFIG_USER_ONLY)
1223 uint64_t helper_ldl_phys(uint64_t p)
1224 {
1225     return (int32_t)ldl_phys(p);
1226 }
1227
1228 uint64_t helper_ldq_phys(uint64_t p)
1229 {
1230     return ldq_phys(p);
1231 }
1232
1233 uint64_t helper_ldl_l_phys(uint64_t p)
1234 {
1235     env->lock_addr = p;
1236     return env->lock_value = (int32_t)ldl_phys(p);
1237 }
1238
1239 uint64_t helper_ldq_l_phys(uint64_t p)
1240 {
1241     env->lock_addr = p;
1242     return env->lock_value = ldl_phys(p);
1243 }
1244
1245 void helper_stl_phys(uint64_t p, uint64_t v)
1246 {
1247     stl_phys(p, v);
1248 }
1249
1250 void helper_stq_phys(uint64_t p, uint64_t v)
1251 {
1252     stq_phys(p, v);
1253 }
1254
1255 uint64_t helper_stl_c_phys(uint64_t p, uint64_t v)
1256 {
1257     uint64_t ret = 0;
1258
1259     if (p == env->lock_addr) {
1260         int32_t old = ldl_phys(p);
1261         if (old == (int32_t)env->lock_value) {
1262             stl_phys(p, v);
1263             ret = 1;
1264         }
1265     }
1266     env->lock_addr = -1;
1267
1268     return ret;
1269 }
1270
1271 uint64_t helper_stq_c_phys(uint64_t p, uint64_t v)
1272 {
1273     uint64_t ret = 0;
1274
1275     if (p == env->lock_addr) {
1276         uint64_t old = ldq_phys(p);
1277         if (old == env->lock_value) {
1278             stq_phys(p, v);
1279             ret = 1;
1280         }
1281     }
1282     env->lock_addr = -1;
1283
1284     return ret;
1285 }
1286
1287 static void QEMU_NORETURN do_unaligned_access(target_ulong addr, int is_write,
1288                                               int is_user, void *retaddr)
1289 {
1290     uint64_t pc;
1291     uint32_t insn;
1292
1293     do_restore_state(retaddr);
1294
1295     pc = env->pc;
1296     insn = ldl_code(pc);
1297
1298     env->trap_arg0 = addr;
1299     env->trap_arg1 = insn >> 26;                /* opcode */
1300     env->trap_arg2 = (insn >> 21) & 31;         /* dest regno */
1301     helper_excp(EXCP_UNALIGN, 0);
1302 }
1303
1304 void QEMU_NORETURN cpu_unassigned_access(CPUState *env1,
1305                                          target_phys_addr_t addr, int is_write,
1306                                          int is_exec, int unused, int size)
1307 {
1308     env = env1;
1309     env->trap_arg0 = addr;
1310     env->trap_arg1 = is_write;
1311     dynamic_excp(EXCP_MCHK, 0);
1312 }
1313
1314 #define MMUSUFFIX _mmu
1315 #define ALIGNED_ONLY
1316
1317 #define SHIFT 0
1318 #include "softmmu_template.h"
1319
1320 #define SHIFT 1
1321 #include "softmmu_template.h"
1322
1323 #define SHIFT 2
1324 #include "softmmu_template.h"
1325
1326 #define SHIFT 3
1327 #include "softmmu_template.h"
1328
1329 /* try to fill the TLB and return an exception if error. If retaddr is
1330    NULL, it means that the function was called in C code (i.e. not
1331    from generated code or from helper.c) */
1332 /* XXX: fix it to restore all registers */
1333 void tlb_fill (target_ulong addr, int is_write, int mmu_idx, void *retaddr)
1334 {
1335     CPUState *saved_env;
1336     int ret;
1337
1338     /* XXX: hack to restore env in all cases, even if not called from
1339        generated code */
1340     saved_env = env;
1341     env = cpu_single_env;
1342     ret = cpu_alpha_handle_mmu_fault(env, addr, is_write, mmu_idx, 1);
1343     if (unlikely(ret != 0)) {
1344         do_restore_state(retaddr);
1345         /* Exception index and error code are already set */
1346         cpu_loop_exit(env);
1347     }
1348     env = saved_env;
1349 }
1350 #endif