1 /* 2 * PowerPC floating point and SPE emulation helpers for QEMU. 3 * 4 * Copyright (c) 2003-2007 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 #include "qemu/osdep.h" 20 #include "cpu.h" 21 #include "exec/helper-proto.h" 22 #include "exec/exec-all.h" 23 #include "internal.h" 24 #include "fpu/softfloat.h" 25 26 static inline float128 float128_snan_to_qnan(float128 x) 27 { 28 float128 r; 29 30 r.high = x.high | 0x0000800000000000; 31 r.low = x.low; 32 return r; 33 } 34 35 #define float64_snan_to_qnan(x) ((x) | 0x0008000000000000ULL) 36 #define float32_snan_to_qnan(x) ((x) | 0x00400000) 37 #define float16_snan_to_qnan(x) ((x) | 0x0200) 38 39 static inline bool fp_exceptions_enabled(CPUPPCState *env) 40 { 41 #ifdef CONFIG_USER_ONLY 42 return true; 43 #else 44 return (env->msr & ((1U << MSR_FE0) | (1U << MSR_FE1))) != 0; 45 #endif 46 } 47 48 /*****************************************************************************/ 49 /* Floating point operations helpers */ 50 51 /* 52 * This is the non-arithmatic conversion that happens e.g. on loads. 53 * In the Power ISA pseudocode, this is called DOUBLE. 54 */ 55 uint64_t helper_todouble(uint32_t arg) 56 { 57 uint32_t abs_arg = arg & 0x7fffffff; 58 uint64_t ret; 59 60 if (likely(abs_arg >= 0x00800000)) { 61 /* Normalized operand, or Inf, or NaN. */ 62 ret = (uint64_t)extract32(arg, 30, 2) << 62; 63 ret |= ((extract32(arg, 30, 1) ^ 1) * (uint64_t)7) << 59; 64 ret |= (uint64_t)extract32(arg, 0, 30) << 29; 65 } else { 66 /* Zero or Denormalized operand. */ 67 ret = (uint64_t)extract32(arg, 31, 1) << 63; 68 if (unlikely(abs_arg != 0)) { 69 /* Denormalized operand. */ 70 int shift = clz32(abs_arg) - 9; 71 int exp = -126 - shift + 1023; 72 ret |= (uint64_t)exp << 52; 73 ret |= abs_arg << (shift + 29); 74 } 75 } 76 return ret; 77 } 78 79 /* 80 * This is the non-arithmatic conversion that happens e.g. on stores. 81 * In the Power ISA pseudocode, this is called SINGLE. 82 */ 83 uint32_t helper_tosingle(uint64_t arg) 84 { 85 int exp = extract64(arg, 52, 11); 86 uint32_t ret; 87 88 if (likely(exp > 896)) { 89 /* No denormalization required (includes Inf, NaN). */ 90 ret = extract64(arg, 62, 2) << 30; 91 ret |= extract64(arg, 29, 30); 92 } else { 93 /* 94 * Zero or Denormal result. If the exponent is in bounds for 95 * a single-precision denormal result, extract the proper 96 * bits. If the input is not zero, and the exponent is out of 97 * bounds, then the result is undefined; this underflows to 98 * zero. 99 */ 100 ret = extract64(arg, 63, 1) << 31; 101 if (unlikely(exp >= 874)) { 102 /* Denormal result. */ 103 ret |= ((1ULL << 52) | extract64(arg, 0, 52)) >> (896 + 30 - exp); 104 } 105 } 106 return ret; 107 } 108 109 static inline int ppc_float32_get_unbiased_exp(float32 f) 110 { 111 return ((f >> 23) & 0xFF) - 127; 112 } 113 114 static inline int ppc_float64_get_unbiased_exp(float64 f) 115 { 116 return ((f >> 52) & 0x7FF) - 1023; 117 } 118 119 /* Classify a floating-point number. */ 120 enum { 121 is_normal = 1, 122 is_zero = 2, 123 is_denormal = 4, 124 is_inf = 8, 125 is_qnan = 16, 126 is_snan = 32, 127 is_neg = 64, 128 }; 129 130 #define COMPUTE_CLASS(tp) \ 131 static int tp##_classify(tp arg) \ 132 { \ 133 int ret = tp##_is_neg(arg) * is_neg; \ 134 if (unlikely(tp##_is_any_nan(arg))) { \ 135 float_status dummy = { }; /* snan_bit_is_one = 0 */ \ 136 ret |= (tp##_is_signaling_nan(arg, &dummy) \ 137 ? is_snan : is_qnan); \ 138 } else if (unlikely(tp##_is_infinity(arg))) { \ 139 ret |= is_inf; \ 140 } else if (tp##_is_zero(arg)) { \ 141 ret |= is_zero; \ 142 } else if (tp##_is_zero_or_denormal(arg)) { \ 143 ret |= is_denormal; \ 144 } else { \ 145 ret |= is_normal; \ 146 } \ 147 return ret; \ 148 } 149 150 COMPUTE_CLASS(float16) 151 COMPUTE_CLASS(float32) 152 COMPUTE_CLASS(float64) 153 COMPUTE_CLASS(float128) 154 155 static void set_fprf_from_class(CPUPPCState *env, int class) 156 { 157 static const uint8_t fprf[6][2] = { 158 { 0x04, 0x08 }, /* normalized */ 159 { 0x02, 0x12 }, /* zero */ 160 { 0x14, 0x18 }, /* denormalized */ 161 { 0x05, 0x09 }, /* infinity */ 162 { 0x11, 0x11 }, /* qnan */ 163 { 0x00, 0x00 }, /* snan -- flags are undefined */ 164 }; 165 bool isneg = class & is_neg; 166 167 env->fpscr &= ~(0x1F << FPSCR_FPRF); 168 env->fpscr |= fprf[ctz32(class)][isneg] << FPSCR_FPRF; 169 } 170 171 #define COMPUTE_FPRF(tp) \ 172 void helper_compute_fprf_##tp(CPUPPCState *env, tp arg) \ 173 { \ 174 set_fprf_from_class(env, tp##_classify(arg)); \ 175 } 176 177 COMPUTE_FPRF(float16) 178 COMPUTE_FPRF(float32) 179 COMPUTE_FPRF(float64) 180 COMPUTE_FPRF(float128) 181 182 /* Floating-point invalid operations exception */ 183 static void finish_invalid_op_excp(CPUPPCState *env, int op, uintptr_t retaddr) 184 { 185 /* Update the floating-point invalid operation summary */ 186 env->fpscr |= 1 << FPSCR_VX; 187 /* Update the floating-point exception summary */ 188 env->fpscr |= FP_FX; 189 if (fpscr_ve != 0) { 190 /* Update the floating-point enabled exception summary */ 191 env->fpscr |= 1 << FPSCR_FEX; 192 if (fp_exceptions_enabled(env)) { 193 raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM, 194 POWERPC_EXCP_FP | op, retaddr); 195 } 196 } 197 } 198 199 static void finish_invalid_op_arith(CPUPPCState *env, int op, 200 bool set_fpcc, uintptr_t retaddr) 201 { 202 env->fpscr &= ~((1 << FPSCR_FR) | (1 << FPSCR_FI)); 203 if (fpscr_ve == 0) { 204 if (set_fpcc) { 205 env->fpscr &= ~(0xF << FPSCR_FPCC); 206 env->fpscr |= 0x11 << FPSCR_FPCC; 207 } 208 } 209 finish_invalid_op_excp(env, op, retaddr); 210 } 211 212 /* Signalling NaN */ 213 static void float_invalid_op_vxsnan(CPUPPCState *env, uintptr_t retaddr) 214 { 215 env->fpscr |= 1 << FPSCR_VXSNAN; 216 finish_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, retaddr); 217 } 218 219 /* Magnitude subtraction of infinities */ 220 static void float_invalid_op_vxisi(CPUPPCState *env, bool set_fpcc, 221 uintptr_t retaddr) 222 { 223 env->fpscr |= 1 << FPSCR_VXISI; 224 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXISI, set_fpcc, retaddr); 225 } 226 227 /* Division of infinity by infinity */ 228 static void float_invalid_op_vxidi(CPUPPCState *env, bool set_fpcc, 229 uintptr_t retaddr) 230 { 231 env->fpscr |= 1 << FPSCR_VXIDI; 232 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXIDI, set_fpcc, retaddr); 233 } 234 235 /* Division of zero by zero */ 236 static void float_invalid_op_vxzdz(CPUPPCState *env, bool set_fpcc, 237 uintptr_t retaddr) 238 { 239 env->fpscr |= 1 << FPSCR_VXZDZ; 240 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXZDZ, set_fpcc, retaddr); 241 } 242 243 /* Multiplication of zero by infinity */ 244 static void float_invalid_op_vximz(CPUPPCState *env, bool set_fpcc, 245 uintptr_t retaddr) 246 { 247 env->fpscr |= 1 << FPSCR_VXIMZ; 248 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXIMZ, set_fpcc, retaddr); 249 } 250 251 /* Square root of a negative number */ 252 static void float_invalid_op_vxsqrt(CPUPPCState *env, bool set_fpcc, 253 uintptr_t retaddr) 254 { 255 env->fpscr |= 1 << FPSCR_VXSQRT; 256 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXSQRT, set_fpcc, retaddr); 257 } 258 259 /* Ordered comparison of NaN */ 260 static void float_invalid_op_vxvc(CPUPPCState *env, bool set_fpcc, 261 uintptr_t retaddr) 262 { 263 env->fpscr |= 1 << FPSCR_VXVC; 264 if (set_fpcc) { 265 env->fpscr &= ~(0xF << FPSCR_FPCC); 266 env->fpscr |= 0x11 << FPSCR_FPCC; 267 } 268 /* Update the floating-point invalid operation summary */ 269 env->fpscr |= 1 << FPSCR_VX; 270 /* Update the floating-point exception summary */ 271 env->fpscr |= FP_FX; 272 /* We must update the target FPR before raising the exception */ 273 if (fpscr_ve != 0) { 274 CPUState *cs = env_cpu(env); 275 276 cs->exception_index = POWERPC_EXCP_PROGRAM; 277 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_VXVC; 278 /* Update the floating-point enabled exception summary */ 279 env->fpscr |= 1 << FPSCR_FEX; 280 /* Exception is differed */ 281 } 282 } 283 284 /* Invalid conversion */ 285 static void float_invalid_op_vxcvi(CPUPPCState *env, bool set_fpcc, 286 uintptr_t retaddr) 287 { 288 env->fpscr |= 1 << FPSCR_VXCVI; 289 env->fpscr &= ~((1 << FPSCR_FR) | (1 << FPSCR_FI)); 290 if (fpscr_ve == 0) { 291 if (set_fpcc) { 292 env->fpscr &= ~(0xF << FPSCR_FPCC); 293 env->fpscr |= 0x11 << FPSCR_FPCC; 294 } 295 } 296 finish_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, retaddr); 297 } 298 299 static inline void float_zero_divide_excp(CPUPPCState *env, uintptr_t raddr) 300 { 301 env->fpscr |= 1 << FPSCR_ZX; 302 env->fpscr &= ~((1 << FPSCR_FR) | (1 << FPSCR_FI)); 303 /* Update the floating-point exception summary */ 304 env->fpscr |= FP_FX; 305 if (fpscr_ze != 0) { 306 /* Update the floating-point enabled exception summary */ 307 env->fpscr |= 1 << FPSCR_FEX; 308 if (fp_exceptions_enabled(env)) { 309 raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM, 310 POWERPC_EXCP_FP | POWERPC_EXCP_FP_ZX, 311 raddr); 312 } 313 } 314 } 315 316 static inline void float_overflow_excp(CPUPPCState *env) 317 { 318 CPUState *cs = env_cpu(env); 319 320 env->fpscr |= 1 << FPSCR_OX; 321 /* Update the floating-point exception summary */ 322 env->fpscr |= FP_FX; 323 if (fpscr_oe != 0) { 324 /* XXX: should adjust the result */ 325 /* Update the floating-point enabled exception summary */ 326 env->fpscr |= 1 << FPSCR_FEX; 327 /* We must update the target FPR before raising the exception */ 328 cs->exception_index = POWERPC_EXCP_PROGRAM; 329 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_OX; 330 } else { 331 env->fpscr |= 1 << FPSCR_XX; 332 env->fpscr |= 1 << FPSCR_FI; 333 } 334 } 335 336 static inline void float_underflow_excp(CPUPPCState *env) 337 { 338 CPUState *cs = env_cpu(env); 339 340 env->fpscr |= 1 << FPSCR_UX; 341 /* Update the floating-point exception summary */ 342 env->fpscr |= FP_FX; 343 if (fpscr_ue != 0) { 344 /* XXX: should adjust the result */ 345 /* Update the floating-point enabled exception summary */ 346 env->fpscr |= 1 << FPSCR_FEX; 347 /* We must update the target FPR before raising the exception */ 348 cs->exception_index = POWERPC_EXCP_PROGRAM; 349 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_UX; 350 } 351 } 352 353 static inline void float_inexact_excp(CPUPPCState *env) 354 { 355 CPUState *cs = env_cpu(env); 356 357 env->fpscr |= 1 << FPSCR_FI; 358 env->fpscr |= 1 << FPSCR_XX; 359 /* Update the floating-point exception summary */ 360 env->fpscr |= FP_FX; 361 if (fpscr_xe != 0) { 362 /* Update the floating-point enabled exception summary */ 363 env->fpscr |= 1 << FPSCR_FEX; 364 /* We must update the target FPR before raising the exception */ 365 cs->exception_index = POWERPC_EXCP_PROGRAM; 366 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_XX; 367 } 368 } 369 370 static inline void fpscr_set_rounding_mode(CPUPPCState *env) 371 { 372 int rnd_type; 373 374 /* Set rounding mode */ 375 switch (fpscr_rn) { 376 case 0: 377 /* Best approximation (round to nearest) */ 378 rnd_type = float_round_nearest_even; 379 break; 380 case 1: 381 /* Smaller magnitude (round toward zero) */ 382 rnd_type = float_round_to_zero; 383 break; 384 case 2: 385 /* Round toward +infinite */ 386 rnd_type = float_round_up; 387 break; 388 default: 389 case 3: 390 /* Round toward -infinite */ 391 rnd_type = float_round_down; 392 break; 393 } 394 set_float_rounding_mode(rnd_type, &env->fp_status); 395 } 396 397 void helper_fpscr_clrbit(CPUPPCState *env, uint32_t bit) 398 { 399 int prev; 400 401 prev = (env->fpscr >> bit) & 1; 402 env->fpscr &= ~(1 << bit); 403 if (prev == 1) { 404 switch (bit) { 405 case FPSCR_RN1: 406 case FPSCR_RN: 407 fpscr_set_rounding_mode(env); 408 break; 409 case FPSCR_VXSNAN: 410 case FPSCR_VXISI: 411 case FPSCR_VXIDI: 412 case FPSCR_VXZDZ: 413 case FPSCR_VXIMZ: 414 case FPSCR_VXVC: 415 case FPSCR_VXSOFT: 416 case FPSCR_VXSQRT: 417 case FPSCR_VXCVI: 418 if (!fpscr_ix) { 419 /* Set VX bit to zero */ 420 env->fpscr &= ~(1 << FPSCR_VX); 421 } 422 break; 423 case FPSCR_OX: 424 case FPSCR_UX: 425 case FPSCR_ZX: 426 case FPSCR_XX: 427 case FPSCR_VE: 428 case FPSCR_OE: 429 case FPSCR_UE: 430 case FPSCR_ZE: 431 case FPSCR_XE: 432 if (!fpscr_eex) { 433 /* Set the FEX bit */ 434 env->fpscr &= ~(1 << FPSCR_FEX); 435 } 436 break; 437 default: 438 break; 439 } 440 } 441 } 442 443 void helper_fpscr_setbit(CPUPPCState *env, uint32_t bit) 444 { 445 CPUState *cs = env_cpu(env); 446 int prev; 447 448 prev = (env->fpscr >> bit) & 1; 449 env->fpscr |= 1 << bit; 450 if (prev == 0) { 451 switch (bit) { 452 case FPSCR_VX: 453 env->fpscr |= FP_FX; 454 if (fpscr_ve) { 455 goto raise_ve; 456 } 457 break; 458 case FPSCR_OX: 459 env->fpscr |= FP_FX; 460 if (fpscr_oe) { 461 goto raise_oe; 462 } 463 break; 464 case FPSCR_UX: 465 env->fpscr |= FP_FX; 466 if (fpscr_ue) { 467 goto raise_ue; 468 } 469 break; 470 case FPSCR_ZX: 471 env->fpscr |= FP_FX; 472 if (fpscr_ze) { 473 goto raise_ze; 474 } 475 break; 476 case FPSCR_XX: 477 env->fpscr |= FP_FX; 478 if (fpscr_xe) { 479 goto raise_xe; 480 } 481 break; 482 case FPSCR_VXSNAN: 483 case FPSCR_VXISI: 484 case FPSCR_VXIDI: 485 case FPSCR_VXZDZ: 486 case FPSCR_VXIMZ: 487 case FPSCR_VXVC: 488 case FPSCR_VXSOFT: 489 case FPSCR_VXSQRT: 490 case FPSCR_VXCVI: 491 env->fpscr |= 1 << FPSCR_VX; 492 env->fpscr |= FP_FX; 493 if (fpscr_ve != 0) { 494 goto raise_ve; 495 } 496 break; 497 case FPSCR_VE: 498 if (fpscr_vx != 0) { 499 raise_ve: 500 env->error_code = POWERPC_EXCP_FP; 501 if (fpscr_vxsnan) { 502 env->error_code |= POWERPC_EXCP_FP_VXSNAN; 503 } 504 if (fpscr_vxisi) { 505 env->error_code |= POWERPC_EXCP_FP_VXISI; 506 } 507 if (fpscr_vxidi) { 508 env->error_code |= POWERPC_EXCP_FP_VXIDI; 509 } 510 if (fpscr_vxzdz) { 511 env->error_code |= POWERPC_EXCP_FP_VXZDZ; 512 } 513 if (fpscr_vximz) { 514 env->error_code |= POWERPC_EXCP_FP_VXIMZ; 515 } 516 if (fpscr_vxvc) { 517 env->error_code |= POWERPC_EXCP_FP_VXVC; 518 } 519 if (fpscr_vxsoft) { 520 env->error_code |= POWERPC_EXCP_FP_VXSOFT; 521 } 522 if (fpscr_vxsqrt) { 523 env->error_code |= POWERPC_EXCP_FP_VXSQRT; 524 } 525 if (fpscr_vxcvi) { 526 env->error_code |= POWERPC_EXCP_FP_VXCVI; 527 } 528 goto raise_excp; 529 } 530 break; 531 case FPSCR_OE: 532 if (fpscr_ox != 0) { 533 raise_oe: 534 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_OX; 535 goto raise_excp; 536 } 537 break; 538 case FPSCR_UE: 539 if (fpscr_ux != 0) { 540 raise_ue: 541 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_UX; 542 goto raise_excp; 543 } 544 break; 545 case FPSCR_ZE: 546 if (fpscr_zx != 0) { 547 raise_ze: 548 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_ZX; 549 goto raise_excp; 550 } 551 break; 552 case FPSCR_XE: 553 if (fpscr_xx != 0) { 554 raise_xe: 555 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_XX; 556 goto raise_excp; 557 } 558 break; 559 case FPSCR_RN1: 560 case FPSCR_RN: 561 fpscr_set_rounding_mode(env); 562 break; 563 default: 564 break; 565 raise_excp: 566 /* Update the floating-point enabled exception summary */ 567 env->fpscr |= 1 << FPSCR_FEX; 568 /* We have to update Rc1 before raising the exception */ 569 cs->exception_index = POWERPC_EXCP_PROGRAM; 570 break; 571 } 572 } 573 } 574 575 void helper_store_fpscr(CPUPPCState *env, uint64_t arg, uint32_t mask) 576 { 577 CPUState *cs = env_cpu(env); 578 target_ulong prev, new; 579 int i; 580 581 prev = env->fpscr; 582 new = (target_ulong)arg; 583 new &= ~0x60000000LL; 584 new |= prev & 0x60000000LL; 585 for (i = 0; i < sizeof(target_ulong) * 2; i++) { 586 if (mask & (1 << i)) { 587 env->fpscr &= ~(0xFLL << (4 * i)); 588 env->fpscr |= new & (0xFLL << (4 * i)); 589 } 590 } 591 /* Update VX and FEX */ 592 if (fpscr_ix != 0) { 593 env->fpscr |= 1 << FPSCR_VX; 594 } else { 595 env->fpscr &= ~(1 << FPSCR_VX); 596 } 597 if ((fpscr_ex & fpscr_eex) != 0) { 598 env->fpscr |= 1 << FPSCR_FEX; 599 cs->exception_index = POWERPC_EXCP_PROGRAM; 600 /* XXX: we should compute it properly */ 601 env->error_code = POWERPC_EXCP_FP; 602 } else { 603 env->fpscr &= ~(1 << FPSCR_FEX); 604 } 605 fpscr_set_rounding_mode(env); 606 } 607 608 void store_fpscr(CPUPPCState *env, uint64_t arg, uint32_t mask) 609 { 610 helper_store_fpscr(env, arg, mask); 611 } 612 613 static void do_float_check_status(CPUPPCState *env, uintptr_t raddr) 614 { 615 CPUState *cs = env_cpu(env); 616 int status = get_float_exception_flags(&env->fp_status); 617 bool inexact_happened = false; 618 619 if (status & float_flag_overflow) { 620 float_overflow_excp(env); 621 } else if (status & float_flag_underflow) { 622 float_underflow_excp(env); 623 } else if (status & float_flag_inexact) { 624 float_inexact_excp(env); 625 inexact_happened = true; 626 } 627 628 /* if the inexact flag was not set */ 629 if (inexact_happened == false) { 630 env->fpscr &= ~(1 << FPSCR_FI); /* clear the FPSCR[FI] bit */ 631 } 632 633 if (cs->exception_index == POWERPC_EXCP_PROGRAM && 634 (env->error_code & POWERPC_EXCP_FP)) { 635 /* Differred floating-point exception after target FPR update */ 636 if (fp_exceptions_enabled(env)) { 637 raise_exception_err_ra(env, cs->exception_index, 638 env->error_code, raddr); 639 } 640 } 641 } 642 643 void helper_float_check_status(CPUPPCState *env) 644 { 645 do_float_check_status(env, GETPC()); 646 } 647 648 void helper_reset_fpstatus(CPUPPCState *env) 649 { 650 set_float_exception_flags(0, &env->fp_status); 651 } 652 653 static void float_invalid_op_addsub(CPUPPCState *env, bool set_fpcc, 654 uintptr_t retaddr, int classes) 655 { 656 if ((classes & ~is_neg) == is_inf) { 657 /* Magnitude subtraction of infinities */ 658 float_invalid_op_vxisi(env, set_fpcc, retaddr); 659 } else if (classes & is_snan) { 660 float_invalid_op_vxsnan(env, retaddr); 661 } 662 } 663 664 /* fadd - fadd. */ 665 float64 helper_fadd(CPUPPCState *env, float64 arg1, float64 arg2) 666 { 667 float64 ret = float64_add(arg1, arg2, &env->fp_status); 668 int status = get_float_exception_flags(&env->fp_status); 669 670 if (unlikely(status & float_flag_invalid)) { 671 float_invalid_op_addsub(env, 1, GETPC(), 672 float64_classify(arg1) | 673 float64_classify(arg2)); 674 } 675 676 return ret; 677 } 678 679 /* fsub - fsub. */ 680 float64 helper_fsub(CPUPPCState *env, float64 arg1, float64 arg2) 681 { 682 float64 ret = float64_sub(arg1, arg2, &env->fp_status); 683 int status = get_float_exception_flags(&env->fp_status); 684 685 if (unlikely(status & float_flag_invalid)) { 686 float_invalid_op_addsub(env, 1, GETPC(), 687 float64_classify(arg1) | 688 float64_classify(arg2)); 689 } 690 691 return ret; 692 } 693 694 static void float_invalid_op_mul(CPUPPCState *env, bool set_fprc, 695 uintptr_t retaddr, int classes) 696 { 697 if ((classes & (is_zero | is_inf)) == (is_zero | is_inf)) { 698 /* Multiplication of zero by infinity */ 699 float_invalid_op_vximz(env, set_fprc, retaddr); 700 } else if (classes & is_snan) { 701 float_invalid_op_vxsnan(env, retaddr); 702 } 703 } 704 705 /* fmul - fmul. */ 706 float64 helper_fmul(CPUPPCState *env, float64 arg1, float64 arg2) 707 { 708 float64 ret = float64_mul(arg1, arg2, &env->fp_status); 709 int status = get_float_exception_flags(&env->fp_status); 710 711 if (unlikely(status & float_flag_invalid)) { 712 float_invalid_op_mul(env, 1, GETPC(), 713 float64_classify(arg1) | 714 float64_classify(arg2)); 715 } 716 717 return ret; 718 } 719 720 static void float_invalid_op_div(CPUPPCState *env, bool set_fprc, 721 uintptr_t retaddr, int classes) 722 { 723 classes &= ~is_neg; 724 if (classes == is_inf) { 725 /* Division of infinity by infinity */ 726 float_invalid_op_vxidi(env, set_fprc, retaddr); 727 } else if (classes == is_zero) { 728 /* Division of zero by zero */ 729 float_invalid_op_vxzdz(env, set_fprc, retaddr); 730 } else if (classes & is_snan) { 731 float_invalid_op_vxsnan(env, retaddr); 732 } 733 } 734 735 /* fdiv - fdiv. */ 736 float64 helper_fdiv(CPUPPCState *env, float64 arg1, float64 arg2) 737 { 738 float64 ret = float64_div(arg1, arg2, &env->fp_status); 739 int status = get_float_exception_flags(&env->fp_status); 740 741 if (unlikely(status)) { 742 if (status & float_flag_invalid) { 743 float_invalid_op_div(env, 1, GETPC(), 744 float64_classify(arg1) | 745 float64_classify(arg2)); 746 } 747 if (status & float_flag_divbyzero) { 748 float_zero_divide_excp(env, GETPC()); 749 } 750 } 751 752 return ret; 753 } 754 755 static void float_invalid_cvt(CPUPPCState *env, bool set_fprc, 756 uintptr_t retaddr, int class1) 757 { 758 float_invalid_op_vxcvi(env, set_fprc, retaddr); 759 if (class1 & is_snan) { 760 float_invalid_op_vxsnan(env, retaddr); 761 } 762 } 763 764 #define FPU_FCTI(op, cvt, nanval) \ 765 uint64_t helper_##op(CPUPPCState *env, float64 arg) \ 766 { \ 767 uint64_t ret = float64_to_##cvt(arg, &env->fp_status); \ 768 int status = get_float_exception_flags(&env->fp_status); \ 769 \ 770 if (unlikely(status)) { \ 771 if (status & float_flag_invalid) { \ 772 float_invalid_cvt(env, 1, GETPC(), float64_classify(arg)); \ 773 ret = nanval; \ 774 } \ 775 do_float_check_status(env, GETPC()); \ 776 } \ 777 return ret; \ 778 } 779 780 FPU_FCTI(fctiw, int32, 0x80000000U) 781 FPU_FCTI(fctiwz, int32_round_to_zero, 0x80000000U) 782 FPU_FCTI(fctiwu, uint32, 0x00000000U) 783 FPU_FCTI(fctiwuz, uint32_round_to_zero, 0x00000000U) 784 FPU_FCTI(fctid, int64, 0x8000000000000000ULL) 785 FPU_FCTI(fctidz, int64_round_to_zero, 0x8000000000000000ULL) 786 FPU_FCTI(fctidu, uint64, 0x0000000000000000ULL) 787 FPU_FCTI(fctiduz, uint64_round_to_zero, 0x0000000000000000ULL) 788 789 #define FPU_FCFI(op, cvtr, is_single) \ 790 uint64_t helper_##op(CPUPPCState *env, uint64_t arg) \ 791 { \ 792 CPU_DoubleU farg; \ 793 \ 794 if (is_single) { \ 795 float32 tmp = cvtr(arg, &env->fp_status); \ 796 farg.d = float32_to_float64(tmp, &env->fp_status); \ 797 } else { \ 798 farg.d = cvtr(arg, &env->fp_status); \ 799 } \ 800 do_float_check_status(env, GETPC()); \ 801 return farg.ll; \ 802 } 803 804 FPU_FCFI(fcfid, int64_to_float64, 0) 805 FPU_FCFI(fcfids, int64_to_float32, 1) 806 FPU_FCFI(fcfidu, uint64_to_float64, 0) 807 FPU_FCFI(fcfidus, uint64_to_float32, 1) 808 809 static inline uint64_t do_fri(CPUPPCState *env, uint64_t arg, 810 int rounding_mode) 811 { 812 CPU_DoubleU farg; 813 814 farg.ll = arg; 815 816 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) { 817 /* sNaN round */ 818 float_invalid_op_vxsnan(env, GETPC()); 819 farg.ll = arg | 0x0008000000000000ULL; 820 } else { 821 int inexact = get_float_exception_flags(&env->fp_status) & 822 float_flag_inexact; 823 set_float_rounding_mode(rounding_mode, &env->fp_status); 824 farg.ll = float64_round_to_int(farg.d, &env->fp_status); 825 /* Restore rounding mode from FPSCR */ 826 fpscr_set_rounding_mode(env); 827 828 /* fri* does not set FPSCR[XX] */ 829 if (!inexact) { 830 env->fp_status.float_exception_flags &= ~float_flag_inexact; 831 } 832 } 833 do_float_check_status(env, GETPC()); 834 return farg.ll; 835 } 836 837 uint64_t helper_frin(CPUPPCState *env, uint64_t arg) 838 { 839 return do_fri(env, arg, float_round_ties_away); 840 } 841 842 uint64_t helper_friz(CPUPPCState *env, uint64_t arg) 843 { 844 return do_fri(env, arg, float_round_to_zero); 845 } 846 847 uint64_t helper_frip(CPUPPCState *env, uint64_t arg) 848 { 849 return do_fri(env, arg, float_round_up); 850 } 851 852 uint64_t helper_frim(CPUPPCState *env, uint64_t arg) 853 { 854 return do_fri(env, arg, float_round_down); 855 } 856 857 #define FPU_MADDSUB_UPDATE(NAME, TP) \ 858 static void NAME(CPUPPCState *env, TP arg1, TP arg2, TP arg3, \ 859 unsigned int madd_flags, uintptr_t retaddr) \ 860 { \ 861 if (TP##_is_signaling_nan(arg1, &env->fp_status) || \ 862 TP##_is_signaling_nan(arg2, &env->fp_status) || \ 863 TP##_is_signaling_nan(arg3, &env->fp_status)) { \ 864 /* sNaN operation */ \ 865 float_invalid_op_vxsnan(env, retaddr); \ 866 } \ 867 if ((TP##_is_infinity(arg1) && TP##_is_zero(arg2)) || \ 868 (TP##_is_zero(arg1) && TP##_is_infinity(arg2))) { \ 869 /* Multiplication of zero by infinity */ \ 870 float_invalid_op_vximz(env, 1, retaddr); \ 871 } \ 872 if ((TP##_is_infinity(arg1) || TP##_is_infinity(arg2)) && \ 873 TP##_is_infinity(arg3)) { \ 874 uint8_t aSign, bSign, cSign; \ 875 \ 876 aSign = TP##_is_neg(arg1); \ 877 bSign = TP##_is_neg(arg2); \ 878 cSign = TP##_is_neg(arg3); \ 879 if (madd_flags & float_muladd_negate_c) { \ 880 cSign ^= 1; \ 881 } \ 882 if (aSign ^ bSign ^ cSign) { \ 883 float_invalid_op_vxisi(env, 1, retaddr); \ 884 } \ 885 } \ 886 } 887 FPU_MADDSUB_UPDATE(float32_maddsub_update_excp, float32) 888 FPU_MADDSUB_UPDATE(float64_maddsub_update_excp, float64) 889 890 #define FPU_FMADD(op, madd_flags) \ 891 uint64_t helper_##op(CPUPPCState *env, uint64_t arg1, \ 892 uint64_t arg2, uint64_t arg3) \ 893 { \ 894 uint32_t flags; \ 895 float64 ret = float64_muladd(arg1, arg2, arg3, madd_flags, \ 896 &env->fp_status); \ 897 flags = get_float_exception_flags(&env->fp_status); \ 898 if (flags) { \ 899 if (flags & float_flag_invalid) { \ 900 float64_maddsub_update_excp(env, arg1, arg2, arg3, \ 901 madd_flags, GETPC()); \ 902 } \ 903 do_float_check_status(env, GETPC()); \ 904 } \ 905 return ret; \ 906 } 907 908 #define MADD_FLGS 0 909 #define MSUB_FLGS float_muladd_negate_c 910 #define NMADD_FLGS float_muladd_negate_result 911 #define NMSUB_FLGS (float_muladd_negate_c | float_muladd_negate_result) 912 913 FPU_FMADD(fmadd, MADD_FLGS) 914 FPU_FMADD(fnmadd, NMADD_FLGS) 915 FPU_FMADD(fmsub, MSUB_FLGS) 916 FPU_FMADD(fnmsub, NMSUB_FLGS) 917 918 /* frsp - frsp. */ 919 uint64_t helper_frsp(CPUPPCState *env, uint64_t arg) 920 { 921 CPU_DoubleU farg; 922 float32 f32; 923 924 farg.ll = arg; 925 926 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) { 927 float_invalid_op_vxsnan(env, GETPC()); 928 } 929 f32 = float64_to_float32(farg.d, &env->fp_status); 930 farg.d = float32_to_float64(f32, &env->fp_status); 931 932 return farg.ll; 933 } 934 935 /* fsqrt - fsqrt. */ 936 float64 helper_fsqrt(CPUPPCState *env, float64 arg) 937 { 938 float64 ret = float64_sqrt(arg, &env->fp_status); 939 int status = get_float_exception_flags(&env->fp_status); 940 941 if (unlikely(status & float_flag_invalid)) { 942 if (unlikely(float64_is_any_nan(arg))) { 943 if (unlikely(float64_is_signaling_nan(arg, &env->fp_status))) { 944 /* sNaN square root */ 945 float_invalid_op_vxsnan(env, GETPC()); 946 } 947 } else { 948 /* Square root of a negative nonzero number */ 949 float_invalid_op_vxsqrt(env, 1, GETPC()); 950 } 951 } 952 953 return ret; 954 } 955 956 /* fre - fre. */ 957 float64 helper_fre(CPUPPCState *env, float64 arg) 958 { 959 /* "Estimate" the reciprocal with actual division. */ 960 float64 ret = float64_div(float64_one, arg, &env->fp_status); 961 int status = get_float_exception_flags(&env->fp_status); 962 963 if (unlikely(status)) { 964 if (status & float_flag_invalid) { 965 if (float64_is_signaling_nan(arg, &env->fp_status)) { 966 /* sNaN reciprocal */ 967 float_invalid_op_vxsnan(env, GETPC()); 968 } 969 } 970 if (status & float_flag_divbyzero) { 971 float_zero_divide_excp(env, GETPC()); 972 /* For FPSCR.ZE == 0, the result is 1/2. */ 973 ret = float64_set_sign(float64_half, float64_is_neg(arg)); 974 } 975 } 976 977 return ret; 978 } 979 980 /* fres - fres. */ 981 uint64_t helper_fres(CPUPPCState *env, uint64_t arg) 982 { 983 CPU_DoubleU farg; 984 float32 f32; 985 986 farg.ll = arg; 987 988 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) { 989 /* sNaN reciprocal */ 990 float_invalid_op_vxsnan(env, GETPC()); 991 } 992 farg.d = float64_div(float64_one, farg.d, &env->fp_status); 993 f32 = float64_to_float32(farg.d, &env->fp_status); 994 farg.d = float32_to_float64(f32, &env->fp_status); 995 996 return farg.ll; 997 } 998 999 /* frsqrte - frsqrte. */ 1000 float64 helper_frsqrte(CPUPPCState *env, float64 arg) 1001 { 1002 /* "Estimate" the reciprocal with actual division. */ 1003 float64 rets = float64_sqrt(arg, &env->fp_status); 1004 float64 retd = float64_div(float64_one, rets, &env->fp_status); 1005 int status = get_float_exception_flags(&env->fp_status); 1006 1007 if (unlikely(status)) { 1008 if (status & float_flag_invalid) { 1009 if (float64_is_signaling_nan(arg, &env->fp_status)) { 1010 /* sNaN reciprocal */ 1011 float_invalid_op_vxsnan(env, GETPC()); 1012 } else { 1013 /* Square root of a negative nonzero number */ 1014 float_invalid_op_vxsqrt(env, 1, GETPC()); 1015 } 1016 } 1017 if (status & float_flag_divbyzero) { 1018 /* Reciprocal of (square root of) zero. */ 1019 float_zero_divide_excp(env, GETPC()); 1020 } 1021 } 1022 1023 return retd; 1024 } 1025 1026 /* fsel - fsel. */ 1027 uint64_t helper_fsel(CPUPPCState *env, uint64_t arg1, uint64_t arg2, 1028 uint64_t arg3) 1029 { 1030 CPU_DoubleU farg1; 1031 1032 farg1.ll = arg1; 1033 1034 if ((!float64_is_neg(farg1.d) || float64_is_zero(farg1.d)) && 1035 !float64_is_any_nan(farg1.d)) { 1036 return arg2; 1037 } else { 1038 return arg3; 1039 } 1040 } 1041 1042 uint32_t helper_ftdiv(uint64_t fra, uint64_t frb) 1043 { 1044 int fe_flag = 0; 1045 int fg_flag = 0; 1046 1047 if (unlikely(float64_is_infinity(fra) || 1048 float64_is_infinity(frb) || 1049 float64_is_zero(frb))) { 1050 fe_flag = 1; 1051 fg_flag = 1; 1052 } else { 1053 int e_a = ppc_float64_get_unbiased_exp(fra); 1054 int e_b = ppc_float64_get_unbiased_exp(frb); 1055 1056 if (unlikely(float64_is_any_nan(fra) || 1057 float64_is_any_nan(frb))) { 1058 fe_flag = 1; 1059 } else if ((e_b <= -1022) || (e_b >= 1021)) { 1060 fe_flag = 1; 1061 } else if (!float64_is_zero(fra) && 1062 (((e_a - e_b) >= 1023) || 1063 ((e_a - e_b) <= -1021) || 1064 (e_a <= -970))) { 1065 fe_flag = 1; 1066 } 1067 1068 if (unlikely(float64_is_zero_or_denormal(frb))) { 1069 /* XB is not zero because of the above check and */ 1070 /* so must be denormalized. */ 1071 fg_flag = 1; 1072 } 1073 } 1074 1075 return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); 1076 } 1077 1078 uint32_t helper_ftsqrt(uint64_t frb) 1079 { 1080 int fe_flag = 0; 1081 int fg_flag = 0; 1082 1083 if (unlikely(float64_is_infinity(frb) || float64_is_zero(frb))) { 1084 fe_flag = 1; 1085 fg_flag = 1; 1086 } else { 1087 int e_b = ppc_float64_get_unbiased_exp(frb); 1088 1089 if (unlikely(float64_is_any_nan(frb))) { 1090 fe_flag = 1; 1091 } else if (unlikely(float64_is_zero(frb))) { 1092 fe_flag = 1; 1093 } else if (unlikely(float64_is_neg(frb))) { 1094 fe_flag = 1; 1095 } else if (!float64_is_zero(frb) && (e_b <= (-1022 + 52))) { 1096 fe_flag = 1; 1097 } 1098 1099 if (unlikely(float64_is_zero_or_denormal(frb))) { 1100 /* XB is not zero because of the above check and */ 1101 /* therefore must be denormalized. */ 1102 fg_flag = 1; 1103 } 1104 } 1105 1106 return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); 1107 } 1108 1109 void helper_fcmpu(CPUPPCState *env, uint64_t arg1, uint64_t arg2, 1110 uint32_t crfD) 1111 { 1112 CPU_DoubleU farg1, farg2; 1113 uint32_t ret = 0; 1114 1115 farg1.ll = arg1; 1116 farg2.ll = arg2; 1117 1118 if (unlikely(float64_is_any_nan(farg1.d) || 1119 float64_is_any_nan(farg2.d))) { 1120 ret = 0x01UL; 1121 } else if (float64_lt(farg1.d, farg2.d, &env->fp_status)) { 1122 ret = 0x08UL; 1123 } else if (!float64_le(farg1.d, farg2.d, &env->fp_status)) { 1124 ret = 0x04UL; 1125 } else { 1126 ret = 0x02UL; 1127 } 1128 1129 env->fpscr &= ~(0x0F << FPSCR_FPRF); 1130 env->fpscr |= ret << FPSCR_FPRF; 1131 env->crf[crfD] = ret; 1132 if (unlikely(ret == 0x01UL 1133 && (float64_is_signaling_nan(farg1.d, &env->fp_status) || 1134 float64_is_signaling_nan(farg2.d, &env->fp_status)))) { 1135 /* sNaN comparison */ 1136 float_invalid_op_vxsnan(env, GETPC()); 1137 } 1138 } 1139 1140 void helper_fcmpo(CPUPPCState *env, uint64_t arg1, uint64_t arg2, 1141 uint32_t crfD) 1142 { 1143 CPU_DoubleU farg1, farg2; 1144 uint32_t ret = 0; 1145 1146 farg1.ll = arg1; 1147 farg2.ll = arg2; 1148 1149 if (unlikely(float64_is_any_nan(farg1.d) || 1150 float64_is_any_nan(farg2.d))) { 1151 ret = 0x01UL; 1152 } else if (float64_lt(farg1.d, farg2.d, &env->fp_status)) { 1153 ret = 0x08UL; 1154 } else if (!float64_le(farg1.d, farg2.d, &env->fp_status)) { 1155 ret = 0x04UL; 1156 } else { 1157 ret = 0x02UL; 1158 } 1159 1160 env->fpscr &= ~(0x0F << FPSCR_FPRF); 1161 env->fpscr |= ret << FPSCR_FPRF; 1162 env->crf[crfD] = ret; 1163 if (unlikely(ret == 0x01UL)) { 1164 float_invalid_op_vxvc(env, 1, GETPC()); 1165 if (float64_is_signaling_nan(farg1.d, &env->fp_status) || 1166 float64_is_signaling_nan(farg2.d, &env->fp_status)) { 1167 /* sNaN comparison */ 1168 float_invalid_op_vxsnan(env, GETPC()); 1169 } 1170 } 1171 } 1172 1173 /* Single-precision floating-point conversions */ 1174 static inline uint32_t efscfsi(CPUPPCState *env, uint32_t val) 1175 { 1176 CPU_FloatU u; 1177 1178 u.f = int32_to_float32(val, &env->vec_status); 1179 1180 return u.l; 1181 } 1182 1183 static inline uint32_t efscfui(CPUPPCState *env, uint32_t val) 1184 { 1185 CPU_FloatU u; 1186 1187 u.f = uint32_to_float32(val, &env->vec_status); 1188 1189 return u.l; 1190 } 1191 1192 static inline int32_t efsctsi(CPUPPCState *env, uint32_t val) 1193 { 1194 CPU_FloatU u; 1195 1196 u.l = val; 1197 /* NaN are not treated the same way IEEE 754 does */ 1198 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) { 1199 return 0; 1200 } 1201 1202 return float32_to_int32(u.f, &env->vec_status); 1203 } 1204 1205 static inline uint32_t efsctui(CPUPPCState *env, uint32_t val) 1206 { 1207 CPU_FloatU u; 1208 1209 u.l = val; 1210 /* NaN are not treated the same way IEEE 754 does */ 1211 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) { 1212 return 0; 1213 } 1214 1215 return float32_to_uint32(u.f, &env->vec_status); 1216 } 1217 1218 static inline uint32_t efsctsiz(CPUPPCState *env, uint32_t val) 1219 { 1220 CPU_FloatU u; 1221 1222 u.l = val; 1223 /* NaN are not treated the same way IEEE 754 does */ 1224 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) { 1225 return 0; 1226 } 1227 1228 return float32_to_int32_round_to_zero(u.f, &env->vec_status); 1229 } 1230 1231 static inline uint32_t efsctuiz(CPUPPCState *env, uint32_t val) 1232 { 1233 CPU_FloatU u; 1234 1235 u.l = val; 1236 /* NaN are not treated the same way IEEE 754 does */ 1237 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) { 1238 return 0; 1239 } 1240 1241 return float32_to_uint32_round_to_zero(u.f, &env->vec_status); 1242 } 1243 1244 static inline uint32_t efscfsf(CPUPPCState *env, uint32_t val) 1245 { 1246 CPU_FloatU u; 1247 float32 tmp; 1248 1249 u.f = int32_to_float32(val, &env->vec_status); 1250 tmp = int64_to_float32(1ULL << 32, &env->vec_status); 1251 u.f = float32_div(u.f, tmp, &env->vec_status); 1252 1253 return u.l; 1254 } 1255 1256 static inline uint32_t efscfuf(CPUPPCState *env, uint32_t val) 1257 { 1258 CPU_FloatU u; 1259 float32 tmp; 1260 1261 u.f = uint32_to_float32(val, &env->vec_status); 1262 tmp = uint64_to_float32(1ULL << 32, &env->vec_status); 1263 u.f = float32_div(u.f, tmp, &env->vec_status); 1264 1265 return u.l; 1266 } 1267 1268 static inline uint32_t efsctsf(CPUPPCState *env, uint32_t val) 1269 { 1270 CPU_FloatU u; 1271 float32 tmp; 1272 1273 u.l = val; 1274 /* NaN are not treated the same way IEEE 754 does */ 1275 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) { 1276 return 0; 1277 } 1278 tmp = uint64_to_float32(1ULL << 32, &env->vec_status); 1279 u.f = float32_mul(u.f, tmp, &env->vec_status); 1280 1281 return float32_to_int32(u.f, &env->vec_status); 1282 } 1283 1284 static inline uint32_t efsctuf(CPUPPCState *env, uint32_t val) 1285 { 1286 CPU_FloatU u; 1287 float32 tmp; 1288 1289 u.l = val; 1290 /* NaN are not treated the same way IEEE 754 does */ 1291 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) { 1292 return 0; 1293 } 1294 tmp = uint64_to_float32(1ULL << 32, &env->vec_status); 1295 u.f = float32_mul(u.f, tmp, &env->vec_status); 1296 1297 return float32_to_uint32(u.f, &env->vec_status); 1298 } 1299 1300 #define HELPER_SPE_SINGLE_CONV(name) \ 1301 uint32_t helper_e##name(CPUPPCState *env, uint32_t val) \ 1302 { \ 1303 return e##name(env, val); \ 1304 } 1305 /* efscfsi */ 1306 HELPER_SPE_SINGLE_CONV(fscfsi); 1307 /* efscfui */ 1308 HELPER_SPE_SINGLE_CONV(fscfui); 1309 /* efscfuf */ 1310 HELPER_SPE_SINGLE_CONV(fscfuf); 1311 /* efscfsf */ 1312 HELPER_SPE_SINGLE_CONV(fscfsf); 1313 /* efsctsi */ 1314 HELPER_SPE_SINGLE_CONV(fsctsi); 1315 /* efsctui */ 1316 HELPER_SPE_SINGLE_CONV(fsctui); 1317 /* efsctsiz */ 1318 HELPER_SPE_SINGLE_CONV(fsctsiz); 1319 /* efsctuiz */ 1320 HELPER_SPE_SINGLE_CONV(fsctuiz); 1321 /* efsctsf */ 1322 HELPER_SPE_SINGLE_CONV(fsctsf); 1323 /* efsctuf */ 1324 HELPER_SPE_SINGLE_CONV(fsctuf); 1325 1326 #define HELPER_SPE_VECTOR_CONV(name) \ 1327 uint64_t helper_ev##name(CPUPPCState *env, uint64_t val) \ 1328 { \ 1329 return ((uint64_t)e##name(env, val >> 32) << 32) | \ 1330 (uint64_t)e##name(env, val); \ 1331 } 1332 /* evfscfsi */ 1333 HELPER_SPE_VECTOR_CONV(fscfsi); 1334 /* evfscfui */ 1335 HELPER_SPE_VECTOR_CONV(fscfui); 1336 /* evfscfuf */ 1337 HELPER_SPE_VECTOR_CONV(fscfuf); 1338 /* evfscfsf */ 1339 HELPER_SPE_VECTOR_CONV(fscfsf); 1340 /* evfsctsi */ 1341 HELPER_SPE_VECTOR_CONV(fsctsi); 1342 /* evfsctui */ 1343 HELPER_SPE_VECTOR_CONV(fsctui); 1344 /* evfsctsiz */ 1345 HELPER_SPE_VECTOR_CONV(fsctsiz); 1346 /* evfsctuiz */ 1347 HELPER_SPE_VECTOR_CONV(fsctuiz); 1348 /* evfsctsf */ 1349 HELPER_SPE_VECTOR_CONV(fsctsf); 1350 /* evfsctuf */ 1351 HELPER_SPE_VECTOR_CONV(fsctuf); 1352 1353 /* Single-precision floating-point arithmetic */ 1354 static inline uint32_t efsadd(CPUPPCState *env, uint32_t op1, uint32_t op2) 1355 { 1356 CPU_FloatU u1, u2; 1357 1358 u1.l = op1; 1359 u2.l = op2; 1360 u1.f = float32_add(u1.f, u2.f, &env->vec_status); 1361 return u1.l; 1362 } 1363 1364 static inline uint32_t efssub(CPUPPCState *env, uint32_t op1, uint32_t op2) 1365 { 1366 CPU_FloatU u1, u2; 1367 1368 u1.l = op1; 1369 u2.l = op2; 1370 u1.f = float32_sub(u1.f, u2.f, &env->vec_status); 1371 return u1.l; 1372 } 1373 1374 static inline uint32_t efsmul(CPUPPCState *env, uint32_t op1, uint32_t op2) 1375 { 1376 CPU_FloatU u1, u2; 1377 1378 u1.l = op1; 1379 u2.l = op2; 1380 u1.f = float32_mul(u1.f, u2.f, &env->vec_status); 1381 return u1.l; 1382 } 1383 1384 static inline uint32_t efsdiv(CPUPPCState *env, uint32_t op1, uint32_t op2) 1385 { 1386 CPU_FloatU u1, u2; 1387 1388 u1.l = op1; 1389 u2.l = op2; 1390 u1.f = float32_div(u1.f, u2.f, &env->vec_status); 1391 return u1.l; 1392 } 1393 1394 #define HELPER_SPE_SINGLE_ARITH(name) \ 1395 uint32_t helper_e##name(CPUPPCState *env, uint32_t op1, uint32_t op2) \ 1396 { \ 1397 return e##name(env, op1, op2); \ 1398 } 1399 /* efsadd */ 1400 HELPER_SPE_SINGLE_ARITH(fsadd); 1401 /* efssub */ 1402 HELPER_SPE_SINGLE_ARITH(fssub); 1403 /* efsmul */ 1404 HELPER_SPE_SINGLE_ARITH(fsmul); 1405 /* efsdiv */ 1406 HELPER_SPE_SINGLE_ARITH(fsdiv); 1407 1408 #define HELPER_SPE_VECTOR_ARITH(name) \ 1409 uint64_t helper_ev##name(CPUPPCState *env, uint64_t op1, uint64_t op2) \ 1410 { \ 1411 return ((uint64_t)e##name(env, op1 >> 32, op2 >> 32) << 32) | \ 1412 (uint64_t)e##name(env, op1, op2); \ 1413 } 1414 /* evfsadd */ 1415 HELPER_SPE_VECTOR_ARITH(fsadd); 1416 /* evfssub */ 1417 HELPER_SPE_VECTOR_ARITH(fssub); 1418 /* evfsmul */ 1419 HELPER_SPE_VECTOR_ARITH(fsmul); 1420 /* evfsdiv */ 1421 HELPER_SPE_VECTOR_ARITH(fsdiv); 1422 1423 /* Single-precision floating-point comparisons */ 1424 static inline uint32_t efscmplt(CPUPPCState *env, uint32_t op1, uint32_t op2) 1425 { 1426 CPU_FloatU u1, u2; 1427 1428 u1.l = op1; 1429 u2.l = op2; 1430 return float32_lt(u1.f, u2.f, &env->vec_status) ? 4 : 0; 1431 } 1432 1433 static inline uint32_t efscmpgt(CPUPPCState *env, uint32_t op1, uint32_t op2) 1434 { 1435 CPU_FloatU u1, u2; 1436 1437 u1.l = op1; 1438 u2.l = op2; 1439 return float32_le(u1.f, u2.f, &env->vec_status) ? 0 : 4; 1440 } 1441 1442 static inline uint32_t efscmpeq(CPUPPCState *env, uint32_t op1, uint32_t op2) 1443 { 1444 CPU_FloatU u1, u2; 1445 1446 u1.l = op1; 1447 u2.l = op2; 1448 return float32_eq(u1.f, u2.f, &env->vec_status) ? 4 : 0; 1449 } 1450 1451 static inline uint32_t efststlt(CPUPPCState *env, uint32_t op1, uint32_t op2) 1452 { 1453 /* XXX: TODO: ignore special values (NaN, infinites, ...) */ 1454 return efscmplt(env, op1, op2); 1455 } 1456 1457 static inline uint32_t efststgt(CPUPPCState *env, uint32_t op1, uint32_t op2) 1458 { 1459 /* XXX: TODO: ignore special values (NaN, infinites, ...) */ 1460 return efscmpgt(env, op1, op2); 1461 } 1462 1463 static inline uint32_t efststeq(CPUPPCState *env, uint32_t op1, uint32_t op2) 1464 { 1465 /* XXX: TODO: ignore special values (NaN, infinites, ...) */ 1466 return efscmpeq(env, op1, op2); 1467 } 1468 1469 #define HELPER_SINGLE_SPE_CMP(name) \ 1470 uint32_t helper_e##name(CPUPPCState *env, uint32_t op1, uint32_t op2) \ 1471 { \ 1472 return e##name(env, op1, op2); \ 1473 } 1474 /* efststlt */ 1475 HELPER_SINGLE_SPE_CMP(fststlt); 1476 /* efststgt */ 1477 HELPER_SINGLE_SPE_CMP(fststgt); 1478 /* efststeq */ 1479 HELPER_SINGLE_SPE_CMP(fststeq); 1480 /* efscmplt */ 1481 HELPER_SINGLE_SPE_CMP(fscmplt); 1482 /* efscmpgt */ 1483 HELPER_SINGLE_SPE_CMP(fscmpgt); 1484 /* efscmpeq */ 1485 HELPER_SINGLE_SPE_CMP(fscmpeq); 1486 1487 static inline uint32_t evcmp_merge(int t0, int t1) 1488 { 1489 return (t0 << 3) | (t1 << 2) | ((t0 | t1) << 1) | (t0 & t1); 1490 } 1491 1492 #define HELPER_VECTOR_SPE_CMP(name) \ 1493 uint32_t helper_ev##name(CPUPPCState *env, uint64_t op1, uint64_t op2) \ 1494 { \ 1495 return evcmp_merge(e##name(env, op1 >> 32, op2 >> 32), \ 1496 e##name(env, op1, op2)); \ 1497 } 1498 /* evfststlt */ 1499 HELPER_VECTOR_SPE_CMP(fststlt); 1500 /* evfststgt */ 1501 HELPER_VECTOR_SPE_CMP(fststgt); 1502 /* evfststeq */ 1503 HELPER_VECTOR_SPE_CMP(fststeq); 1504 /* evfscmplt */ 1505 HELPER_VECTOR_SPE_CMP(fscmplt); 1506 /* evfscmpgt */ 1507 HELPER_VECTOR_SPE_CMP(fscmpgt); 1508 /* evfscmpeq */ 1509 HELPER_VECTOR_SPE_CMP(fscmpeq); 1510 1511 /* Double-precision floating-point conversion */ 1512 uint64_t helper_efdcfsi(CPUPPCState *env, uint32_t val) 1513 { 1514 CPU_DoubleU u; 1515 1516 u.d = int32_to_float64(val, &env->vec_status); 1517 1518 return u.ll; 1519 } 1520 1521 uint64_t helper_efdcfsid(CPUPPCState *env, uint64_t val) 1522 { 1523 CPU_DoubleU u; 1524 1525 u.d = int64_to_float64(val, &env->vec_status); 1526 1527 return u.ll; 1528 } 1529 1530 uint64_t helper_efdcfui(CPUPPCState *env, uint32_t val) 1531 { 1532 CPU_DoubleU u; 1533 1534 u.d = uint32_to_float64(val, &env->vec_status); 1535 1536 return u.ll; 1537 } 1538 1539 uint64_t helper_efdcfuid(CPUPPCState *env, uint64_t val) 1540 { 1541 CPU_DoubleU u; 1542 1543 u.d = uint64_to_float64(val, &env->vec_status); 1544 1545 return u.ll; 1546 } 1547 1548 uint32_t helper_efdctsi(CPUPPCState *env, uint64_t val) 1549 { 1550 CPU_DoubleU u; 1551 1552 u.ll = val; 1553 /* NaN are not treated the same way IEEE 754 does */ 1554 if (unlikely(float64_is_any_nan(u.d))) { 1555 return 0; 1556 } 1557 1558 return float64_to_int32(u.d, &env->vec_status); 1559 } 1560 1561 uint32_t helper_efdctui(CPUPPCState *env, uint64_t val) 1562 { 1563 CPU_DoubleU u; 1564 1565 u.ll = val; 1566 /* NaN are not treated the same way IEEE 754 does */ 1567 if (unlikely(float64_is_any_nan(u.d))) { 1568 return 0; 1569 } 1570 1571 return float64_to_uint32(u.d, &env->vec_status); 1572 } 1573 1574 uint32_t helper_efdctsiz(CPUPPCState *env, uint64_t val) 1575 { 1576 CPU_DoubleU u; 1577 1578 u.ll = val; 1579 /* NaN are not treated the same way IEEE 754 does */ 1580 if (unlikely(float64_is_any_nan(u.d))) { 1581 return 0; 1582 } 1583 1584 return float64_to_int32_round_to_zero(u.d, &env->vec_status); 1585 } 1586 1587 uint64_t helper_efdctsidz(CPUPPCState *env, uint64_t val) 1588 { 1589 CPU_DoubleU u; 1590 1591 u.ll = val; 1592 /* NaN are not treated the same way IEEE 754 does */ 1593 if (unlikely(float64_is_any_nan(u.d))) { 1594 return 0; 1595 } 1596 1597 return float64_to_int64_round_to_zero(u.d, &env->vec_status); 1598 } 1599 1600 uint32_t helper_efdctuiz(CPUPPCState *env, uint64_t val) 1601 { 1602 CPU_DoubleU u; 1603 1604 u.ll = val; 1605 /* NaN are not treated the same way IEEE 754 does */ 1606 if (unlikely(float64_is_any_nan(u.d))) { 1607 return 0; 1608 } 1609 1610 return float64_to_uint32_round_to_zero(u.d, &env->vec_status); 1611 } 1612 1613 uint64_t helper_efdctuidz(CPUPPCState *env, uint64_t val) 1614 { 1615 CPU_DoubleU u; 1616 1617 u.ll = val; 1618 /* NaN are not treated the same way IEEE 754 does */ 1619 if (unlikely(float64_is_any_nan(u.d))) { 1620 return 0; 1621 } 1622 1623 return float64_to_uint64_round_to_zero(u.d, &env->vec_status); 1624 } 1625 1626 uint64_t helper_efdcfsf(CPUPPCState *env, uint32_t val) 1627 { 1628 CPU_DoubleU u; 1629 float64 tmp; 1630 1631 u.d = int32_to_float64(val, &env->vec_status); 1632 tmp = int64_to_float64(1ULL << 32, &env->vec_status); 1633 u.d = float64_div(u.d, tmp, &env->vec_status); 1634 1635 return u.ll; 1636 } 1637 1638 uint64_t helper_efdcfuf(CPUPPCState *env, uint32_t val) 1639 { 1640 CPU_DoubleU u; 1641 float64 tmp; 1642 1643 u.d = uint32_to_float64(val, &env->vec_status); 1644 tmp = int64_to_float64(1ULL << 32, &env->vec_status); 1645 u.d = float64_div(u.d, tmp, &env->vec_status); 1646 1647 return u.ll; 1648 } 1649 1650 uint32_t helper_efdctsf(CPUPPCState *env, uint64_t val) 1651 { 1652 CPU_DoubleU u; 1653 float64 tmp; 1654 1655 u.ll = val; 1656 /* NaN are not treated the same way IEEE 754 does */ 1657 if (unlikely(float64_is_any_nan(u.d))) { 1658 return 0; 1659 } 1660 tmp = uint64_to_float64(1ULL << 32, &env->vec_status); 1661 u.d = float64_mul(u.d, tmp, &env->vec_status); 1662 1663 return float64_to_int32(u.d, &env->vec_status); 1664 } 1665 1666 uint32_t helper_efdctuf(CPUPPCState *env, uint64_t val) 1667 { 1668 CPU_DoubleU u; 1669 float64 tmp; 1670 1671 u.ll = val; 1672 /* NaN are not treated the same way IEEE 754 does */ 1673 if (unlikely(float64_is_any_nan(u.d))) { 1674 return 0; 1675 } 1676 tmp = uint64_to_float64(1ULL << 32, &env->vec_status); 1677 u.d = float64_mul(u.d, tmp, &env->vec_status); 1678 1679 return float64_to_uint32(u.d, &env->vec_status); 1680 } 1681 1682 uint32_t helper_efscfd(CPUPPCState *env, uint64_t val) 1683 { 1684 CPU_DoubleU u1; 1685 CPU_FloatU u2; 1686 1687 u1.ll = val; 1688 u2.f = float64_to_float32(u1.d, &env->vec_status); 1689 1690 return u2.l; 1691 } 1692 1693 uint64_t helper_efdcfs(CPUPPCState *env, uint32_t val) 1694 { 1695 CPU_DoubleU u2; 1696 CPU_FloatU u1; 1697 1698 u1.l = val; 1699 u2.d = float32_to_float64(u1.f, &env->vec_status); 1700 1701 return u2.ll; 1702 } 1703 1704 /* Double precision fixed-point arithmetic */ 1705 uint64_t helper_efdadd(CPUPPCState *env, uint64_t op1, uint64_t op2) 1706 { 1707 CPU_DoubleU u1, u2; 1708 1709 u1.ll = op1; 1710 u2.ll = op2; 1711 u1.d = float64_add(u1.d, u2.d, &env->vec_status); 1712 return u1.ll; 1713 } 1714 1715 uint64_t helper_efdsub(CPUPPCState *env, uint64_t op1, uint64_t op2) 1716 { 1717 CPU_DoubleU u1, u2; 1718 1719 u1.ll = op1; 1720 u2.ll = op2; 1721 u1.d = float64_sub(u1.d, u2.d, &env->vec_status); 1722 return u1.ll; 1723 } 1724 1725 uint64_t helper_efdmul(CPUPPCState *env, uint64_t op1, uint64_t op2) 1726 { 1727 CPU_DoubleU u1, u2; 1728 1729 u1.ll = op1; 1730 u2.ll = op2; 1731 u1.d = float64_mul(u1.d, u2.d, &env->vec_status); 1732 return u1.ll; 1733 } 1734 1735 uint64_t helper_efddiv(CPUPPCState *env, uint64_t op1, uint64_t op2) 1736 { 1737 CPU_DoubleU u1, u2; 1738 1739 u1.ll = op1; 1740 u2.ll = op2; 1741 u1.d = float64_div(u1.d, u2.d, &env->vec_status); 1742 return u1.ll; 1743 } 1744 1745 /* Double precision floating point helpers */ 1746 uint32_t helper_efdtstlt(CPUPPCState *env, uint64_t op1, uint64_t op2) 1747 { 1748 CPU_DoubleU u1, u2; 1749 1750 u1.ll = op1; 1751 u2.ll = op2; 1752 return float64_lt(u1.d, u2.d, &env->vec_status) ? 4 : 0; 1753 } 1754 1755 uint32_t helper_efdtstgt(CPUPPCState *env, uint64_t op1, uint64_t op2) 1756 { 1757 CPU_DoubleU u1, u2; 1758 1759 u1.ll = op1; 1760 u2.ll = op2; 1761 return float64_le(u1.d, u2.d, &env->vec_status) ? 0 : 4; 1762 } 1763 1764 uint32_t helper_efdtsteq(CPUPPCState *env, uint64_t op1, uint64_t op2) 1765 { 1766 CPU_DoubleU u1, u2; 1767 1768 u1.ll = op1; 1769 u2.ll = op2; 1770 return float64_eq_quiet(u1.d, u2.d, &env->vec_status) ? 4 : 0; 1771 } 1772 1773 uint32_t helper_efdcmplt(CPUPPCState *env, uint64_t op1, uint64_t op2) 1774 { 1775 /* XXX: TODO: test special values (NaN, infinites, ...) */ 1776 return helper_efdtstlt(env, op1, op2); 1777 } 1778 1779 uint32_t helper_efdcmpgt(CPUPPCState *env, uint64_t op1, uint64_t op2) 1780 { 1781 /* XXX: TODO: test special values (NaN, infinites, ...) */ 1782 return helper_efdtstgt(env, op1, op2); 1783 } 1784 1785 uint32_t helper_efdcmpeq(CPUPPCState *env, uint64_t op1, uint64_t op2) 1786 { 1787 /* XXX: TODO: test special values (NaN, infinites, ...) */ 1788 return helper_efdtsteq(env, op1, op2); 1789 } 1790 1791 #define float64_to_float64(x, env) x 1792 1793 1794 /* 1795 * VSX_ADD_SUB - VSX floating point add/subract 1796 * name - instruction mnemonic 1797 * op - operation (add or sub) 1798 * nels - number of elements (1, 2 or 4) 1799 * tp - type (float32 or float64) 1800 * fld - vsr_t field (VsrD(*) or VsrW(*)) 1801 * sfprf - set FPRF 1802 */ 1803 #define VSX_ADD_SUB(name, op, nels, tp, fld, sfprf, r2sp) \ 1804 void helper_##name(CPUPPCState *env, ppc_vsr_t *xt, \ 1805 ppc_vsr_t *xa, ppc_vsr_t *xb) \ 1806 { \ 1807 ppc_vsr_t t = *xt; \ 1808 int i; \ 1809 \ 1810 helper_reset_fpstatus(env); \ 1811 \ 1812 for (i = 0; i < nels; i++) { \ 1813 float_status tstat = env->fp_status; \ 1814 set_float_exception_flags(0, &tstat); \ 1815 t.fld = tp##_##op(xa->fld, xb->fld, &tstat); \ 1816 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \ 1817 \ 1818 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \ 1819 float_invalid_op_addsub(env, sfprf, GETPC(), \ 1820 tp##_classify(xa->fld) | \ 1821 tp##_classify(xb->fld)); \ 1822 } \ 1823 \ 1824 if (r2sp) { \ 1825 t.fld = helper_frsp(env, t.fld); \ 1826 } \ 1827 \ 1828 if (sfprf) { \ 1829 helper_compute_fprf_float64(env, t.fld); \ 1830 } \ 1831 } \ 1832 *xt = t; \ 1833 do_float_check_status(env, GETPC()); \ 1834 } 1835 1836 VSX_ADD_SUB(xsadddp, add, 1, float64, VsrD(0), 1, 0) 1837 VSX_ADD_SUB(xsaddsp, add, 1, float64, VsrD(0), 1, 1) 1838 VSX_ADD_SUB(xvadddp, add, 2, float64, VsrD(i), 0, 0) 1839 VSX_ADD_SUB(xvaddsp, add, 4, float32, VsrW(i), 0, 0) 1840 VSX_ADD_SUB(xssubdp, sub, 1, float64, VsrD(0), 1, 0) 1841 VSX_ADD_SUB(xssubsp, sub, 1, float64, VsrD(0), 1, 1) 1842 VSX_ADD_SUB(xvsubdp, sub, 2, float64, VsrD(i), 0, 0) 1843 VSX_ADD_SUB(xvsubsp, sub, 4, float32, VsrW(i), 0, 0) 1844 1845 void helper_xsaddqp(CPUPPCState *env, uint32_t opcode, 1846 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb) 1847 { 1848 ppc_vsr_t t = *xt; 1849 float_status tstat; 1850 1851 helper_reset_fpstatus(env); 1852 1853 tstat = env->fp_status; 1854 if (unlikely(Rc(opcode) != 0)) { 1855 tstat.float_rounding_mode = float_round_to_odd; 1856 } 1857 1858 set_float_exception_flags(0, &tstat); 1859 t.f128 = float128_add(xa->f128, xb->f128, &tstat); 1860 env->fp_status.float_exception_flags |= tstat.float_exception_flags; 1861 1862 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { 1863 float_invalid_op_addsub(env, 1, GETPC(), 1864 float128_classify(xa->f128) | 1865 float128_classify(xb->f128)); 1866 } 1867 1868 helper_compute_fprf_float128(env, t.f128); 1869 1870 *xt = t; 1871 do_float_check_status(env, GETPC()); 1872 } 1873 1874 /* 1875 * VSX_MUL - VSX floating point multiply 1876 * op - instruction mnemonic 1877 * nels - number of elements (1, 2 or 4) 1878 * tp - type (float32 or float64) 1879 * fld - vsr_t field (VsrD(*) or VsrW(*)) 1880 * sfprf - set FPRF 1881 */ 1882 #define VSX_MUL(op, nels, tp, fld, sfprf, r2sp) \ 1883 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \ 1884 ppc_vsr_t *xa, ppc_vsr_t *xb) \ 1885 { \ 1886 ppc_vsr_t t = *xt; \ 1887 int i; \ 1888 \ 1889 helper_reset_fpstatus(env); \ 1890 \ 1891 for (i = 0; i < nels; i++) { \ 1892 float_status tstat = env->fp_status; \ 1893 set_float_exception_flags(0, &tstat); \ 1894 t.fld = tp##_mul(xa->fld, xb->fld, &tstat); \ 1895 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \ 1896 \ 1897 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \ 1898 float_invalid_op_mul(env, sfprf, GETPC(), \ 1899 tp##_classify(xa->fld) | \ 1900 tp##_classify(xb->fld)); \ 1901 } \ 1902 \ 1903 if (r2sp) { \ 1904 t.fld = helper_frsp(env, t.fld); \ 1905 } \ 1906 \ 1907 if (sfprf) { \ 1908 helper_compute_fprf_float64(env, t.fld); \ 1909 } \ 1910 } \ 1911 \ 1912 *xt = t; \ 1913 do_float_check_status(env, GETPC()); \ 1914 } 1915 1916 VSX_MUL(xsmuldp, 1, float64, VsrD(0), 1, 0) 1917 VSX_MUL(xsmulsp, 1, float64, VsrD(0), 1, 1) 1918 VSX_MUL(xvmuldp, 2, float64, VsrD(i), 0, 0) 1919 VSX_MUL(xvmulsp, 4, float32, VsrW(i), 0, 0) 1920 1921 void helper_xsmulqp(CPUPPCState *env, uint32_t opcode, 1922 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb) 1923 { 1924 ppc_vsr_t t = *xt; 1925 float_status tstat; 1926 1927 helper_reset_fpstatus(env); 1928 tstat = env->fp_status; 1929 if (unlikely(Rc(opcode) != 0)) { 1930 tstat.float_rounding_mode = float_round_to_odd; 1931 } 1932 1933 set_float_exception_flags(0, &tstat); 1934 t.f128 = float128_mul(xa->f128, xb->f128, &tstat); 1935 env->fp_status.float_exception_flags |= tstat.float_exception_flags; 1936 1937 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { 1938 float_invalid_op_mul(env, 1, GETPC(), 1939 float128_classify(xa->f128) | 1940 float128_classify(xb->f128)); 1941 } 1942 helper_compute_fprf_float128(env, t.f128); 1943 1944 *xt = t; 1945 do_float_check_status(env, GETPC()); 1946 } 1947 1948 /* 1949 * VSX_DIV - VSX floating point divide 1950 * op - instruction mnemonic 1951 * nels - number of elements (1, 2 or 4) 1952 * tp - type (float32 or float64) 1953 * fld - vsr_t field (VsrD(*) or VsrW(*)) 1954 * sfprf - set FPRF 1955 */ 1956 #define VSX_DIV(op, nels, tp, fld, sfprf, r2sp) \ 1957 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \ 1958 ppc_vsr_t *xa, ppc_vsr_t *xb) \ 1959 { \ 1960 ppc_vsr_t t = *xt; \ 1961 int i; \ 1962 \ 1963 helper_reset_fpstatus(env); \ 1964 \ 1965 for (i = 0; i < nels; i++) { \ 1966 float_status tstat = env->fp_status; \ 1967 set_float_exception_flags(0, &tstat); \ 1968 t.fld = tp##_div(xa->fld, xb->fld, &tstat); \ 1969 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \ 1970 \ 1971 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \ 1972 float_invalid_op_div(env, sfprf, GETPC(), \ 1973 tp##_classify(xa->fld) | \ 1974 tp##_classify(xb->fld)); \ 1975 } \ 1976 if (unlikely(tstat.float_exception_flags & float_flag_divbyzero)) { \ 1977 float_zero_divide_excp(env, GETPC()); \ 1978 } \ 1979 \ 1980 if (r2sp) { \ 1981 t.fld = helper_frsp(env, t.fld); \ 1982 } \ 1983 \ 1984 if (sfprf) { \ 1985 helper_compute_fprf_float64(env, t.fld); \ 1986 } \ 1987 } \ 1988 \ 1989 *xt = t; \ 1990 do_float_check_status(env, GETPC()); \ 1991 } 1992 1993 VSX_DIV(xsdivdp, 1, float64, VsrD(0), 1, 0) 1994 VSX_DIV(xsdivsp, 1, float64, VsrD(0), 1, 1) 1995 VSX_DIV(xvdivdp, 2, float64, VsrD(i), 0, 0) 1996 VSX_DIV(xvdivsp, 4, float32, VsrW(i), 0, 0) 1997 1998 void helper_xsdivqp(CPUPPCState *env, uint32_t opcode, 1999 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb) 2000 { 2001 ppc_vsr_t t = *xt; 2002 float_status tstat; 2003 2004 helper_reset_fpstatus(env); 2005 tstat = env->fp_status; 2006 if (unlikely(Rc(opcode) != 0)) { 2007 tstat.float_rounding_mode = float_round_to_odd; 2008 } 2009 2010 set_float_exception_flags(0, &tstat); 2011 t.f128 = float128_div(xa->f128, xb->f128, &tstat); 2012 env->fp_status.float_exception_flags |= tstat.float_exception_flags; 2013 2014 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { 2015 float_invalid_op_div(env, 1, GETPC(), 2016 float128_classify(xa->f128) | 2017 float128_classify(xb->f128)); 2018 } 2019 if (unlikely(tstat.float_exception_flags & float_flag_divbyzero)) { 2020 float_zero_divide_excp(env, GETPC()); 2021 } 2022 2023 helper_compute_fprf_float128(env, t.f128); 2024 *xt = t; 2025 do_float_check_status(env, GETPC()); 2026 } 2027 2028 /* 2029 * VSX_RE - VSX floating point reciprocal estimate 2030 * op - instruction mnemonic 2031 * nels - number of elements (1, 2 or 4) 2032 * tp - type (float32 or float64) 2033 * fld - vsr_t field (VsrD(*) or VsrW(*)) 2034 * sfprf - set FPRF 2035 */ 2036 #define VSX_RE(op, nels, tp, fld, sfprf, r2sp) \ 2037 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \ 2038 { \ 2039 ppc_vsr_t t = *xt; \ 2040 int i; \ 2041 \ 2042 helper_reset_fpstatus(env); \ 2043 \ 2044 for (i = 0; i < nels; i++) { \ 2045 if (unlikely(tp##_is_signaling_nan(xb->fld, &env->fp_status))) { \ 2046 float_invalid_op_vxsnan(env, GETPC()); \ 2047 } \ 2048 t.fld = tp##_div(tp##_one, xb->fld, &env->fp_status); \ 2049 \ 2050 if (r2sp) { \ 2051 t.fld = helper_frsp(env, t.fld); \ 2052 } \ 2053 \ 2054 if (sfprf) { \ 2055 helper_compute_fprf_float64(env, t.fld); \ 2056 } \ 2057 } \ 2058 \ 2059 *xt = t; \ 2060 do_float_check_status(env, GETPC()); \ 2061 } 2062 2063 VSX_RE(xsredp, 1, float64, VsrD(0), 1, 0) 2064 VSX_RE(xsresp, 1, float64, VsrD(0), 1, 1) 2065 VSX_RE(xvredp, 2, float64, VsrD(i), 0, 0) 2066 VSX_RE(xvresp, 4, float32, VsrW(i), 0, 0) 2067 2068 /* 2069 * VSX_SQRT - VSX floating point square root 2070 * op - instruction mnemonic 2071 * nels - number of elements (1, 2 or 4) 2072 * tp - type (float32 or float64) 2073 * fld - vsr_t field (VsrD(*) or VsrW(*)) 2074 * sfprf - set FPRF 2075 */ 2076 #define VSX_SQRT(op, nels, tp, fld, sfprf, r2sp) \ 2077 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \ 2078 { \ 2079 ppc_vsr_t t = *xt; \ 2080 int i; \ 2081 \ 2082 helper_reset_fpstatus(env); \ 2083 \ 2084 for (i = 0; i < nels; i++) { \ 2085 float_status tstat = env->fp_status; \ 2086 set_float_exception_flags(0, &tstat); \ 2087 t.fld = tp##_sqrt(xb->fld, &tstat); \ 2088 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \ 2089 \ 2090 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \ 2091 if (tp##_is_neg(xb->fld) && !tp##_is_zero(xb->fld)) { \ 2092 float_invalid_op_vxsqrt(env, sfprf, GETPC()); \ 2093 } else if (tp##_is_signaling_nan(xb->fld, &tstat)) { \ 2094 float_invalid_op_vxsnan(env, GETPC()); \ 2095 } \ 2096 } \ 2097 \ 2098 if (r2sp) { \ 2099 t.fld = helper_frsp(env, t.fld); \ 2100 } \ 2101 \ 2102 if (sfprf) { \ 2103 helper_compute_fprf_float64(env, t.fld); \ 2104 } \ 2105 } \ 2106 \ 2107 *xt = t; \ 2108 do_float_check_status(env, GETPC()); \ 2109 } 2110 2111 VSX_SQRT(xssqrtdp, 1, float64, VsrD(0), 1, 0) 2112 VSX_SQRT(xssqrtsp, 1, float64, VsrD(0), 1, 1) 2113 VSX_SQRT(xvsqrtdp, 2, float64, VsrD(i), 0, 0) 2114 VSX_SQRT(xvsqrtsp, 4, float32, VsrW(i), 0, 0) 2115 2116 /* 2117 *VSX_RSQRTE - VSX floating point reciprocal square root estimate 2118 * op - instruction mnemonic 2119 * nels - number of elements (1, 2 or 4) 2120 * tp - type (float32 or float64) 2121 * fld - vsr_t field (VsrD(*) or VsrW(*)) 2122 * sfprf - set FPRF 2123 */ 2124 #define VSX_RSQRTE(op, nels, tp, fld, sfprf, r2sp) \ 2125 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \ 2126 { \ 2127 ppc_vsr_t t = *xt; \ 2128 int i; \ 2129 \ 2130 helper_reset_fpstatus(env); \ 2131 \ 2132 for (i = 0; i < nels; i++) { \ 2133 float_status tstat = env->fp_status; \ 2134 set_float_exception_flags(0, &tstat); \ 2135 t.fld = tp##_sqrt(xb->fld, &tstat); \ 2136 t.fld = tp##_div(tp##_one, t.fld, &tstat); \ 2137 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \ 2138 \ 2139 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \ 2140 if (tp##_is_neg(xb->fld) && !tp##_is_zero(xb->fld)) { \ 2141 float_invalid_op_vxsqrt(env, sfprf, GETPC()); \ 2142 } else if (tp##_is_signaling_nan(xb->fld, &tstat)) { \ 2143 float_invalid_op_vxsnan(env, GETPC()); \ 2144 } \ 2145 } \ 2146 \ 2147 if (r2sp) { \ 2148 t.fld = helper_frsp(env, t.fld); \ 2149 } \ 2150 \ 2151 if (sfprf) { \ 2152 helper_compute_fprf_float64(env, t.fld); \ 2153 } \ 2154 } \ 2155 \ 2156 *xt = t; \ 2157 do_float_check_status(env, GETPC()); \ 2158 } 2159 2160 VSX_RSQRTE(xsrsqrtedp, 1, float64, VsrD(0), 1, 0) 2161 VSX_RSQRTE(xsrsqrtesp, 1, float64, VsrD(0), 1, 1) 2162 VSX_RSQRTE(xvrsqrtedp, 2, float64, VsrD(i), 0, 0) 2163 VSX_RSQRTE(xvrsqrtesp, 4, float32, VsrW(i), 0, 0) 2164 2165 /* 2166 * VSX_TDIV - VSX floating point test for divide 2167 * op - instruction mnemonic 2168 * nels - number of elements (1, 2 or 4) 2169 * tp - type (float32 or float64) 2170 * fld - vsr_t field (VsrD(*) or VsrW(*)) 2171 * emin - minimum unbiased exponent 2172 * emax - maximum unbiased exponent 2173 * nbits - number of fraction bits 2174 */ 2175 #define VSX_TDIV(op, nels, tp, fld, emin, emax, nbits) \ 2176 void helper_##op(CPUPPCState *env, uint32_t opcode, \ 2177 ppc_vsr_t *xa, ppc_vsr_t *xb) \ 2178 { \ 2179 int i; \ 2180 int fe_flag = 0; \ 2181 int fg_flag = 0; \ 2182 \ 2183 for (i = 0; i < nels; i++) { \ 2184 if (unlikely(tp##_is_infinity(xa->fld) || \ 2185 tp##_is_infinity(xb->fld) || \ 2186 tp##_is_zero(xb->fld))) { \ 2187 fe_flag = 1; \ 2188 fg_flag = 1; \ 2189 } else { \ 2190 int e_a = ppc_##tp##_get_unbiased_exp(xa->fld); \ 2191 int e_b = ppc_##tp##_get_unbiased_exp(xb->fld); \ 2192 \ 2193 if (unlikely(tp##_is_any_nan(xa->fld) || \ 2194 tp##_is_any_nan(xb->fld))) { \ 2195 fe_flag = 1; \ 2196 } else if ((e_b <= emin) || (e_b >= (emax - 2))) { \ 2197 fe_flag = 1; \ 2198 } else if (!tp##_is_zero(xa->fld) && \ 2199 (((e_a - e_b) >= emax) || \ 2200 ((e_a - e_b) <= (emin + 1)) || \ 2201 (e_a <= (emin + nbits)))) { \ 2202 fe_flag = 1; \ 2203 } \ 2204 \ 2205 if (unlikely(tp##_is_zero_or_denormal(xb->fld))) { \ 2206 /* \ 2207 * XB is not zero because of the above check and so \ 2208 * must be denormalized. \ 2209 */ \ 2210 fg_flag = 1; \ 2211 } \ 2212 } \ 2213 } \ 2214 \ 2215 env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \ 2216 } 2217 2218 VSX_TDIV(xstdivdp, 1, float64, VsrD(0), -1022, 1023, 52) 2219 VSX_TDIV(xvtdivdp, 2, float64, VsrD(i), -1022, 1023, 52) 2220 VSX_TDIV(xvtdivsp, 4, float32, VsrW(i), -126, 127, 23) 2221 2222 /* 2223 * VSX_TSQRT - VSX floating point test for square root 2224 * op - instruction mnemonic 2225 * nels - number of elements (1, 2 or 4) 2226 * tp - type (float32 or float64) 2227 * fld - vsr_t field (VsrD(*) or VsrW(*)) 2228 * emin - minimum unbiased exponent 2229 * emax - maximum unbiased exponent 2230 * nbits - number of fraction bits 2231 */ 2232 #define VSX_TSQRT(op, nels, tp, fld, emin, nbits) \ 2233 void helper_##op(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xb) \ 2234 { \ 2235 int i; \ 2236 int fe_flag = 0; \ 2237 int fg_flag = 0; \ 2238 \ 2239 for (i = 0; i < nels; i++) { \ 2240 if (unlikely(tp##_is_infinity(xb->fld) || \ 2241 tp##_is_zero(xb->fld))) { \ 2242 fe_flag = 1; \ 2243 fg_flag = 1; \ 2244 } else { \ 2245 int e_b = ppc_##tp##_get_unbiased_exp(xb->fld); \ 2246 \ 2247 if (unlikely(tp##_is_any_nan(xb->fld))) { \ 2248 fe_flag = 1; \ 2249 } else if (unlikely(tp##_is_zero(xb->fld))) { \ 2250 fe_flag = 1; \ 2251 } else if (unlikely(tp##_is_neg(xb->fld))) { \ 2252 fe_flag = 1; \ 2253 } else if (!tp##_is_zero(xb->fld) && \ 2254 (e_b <= (emin + nbits))) { \ 2255 fe_flag = 1; \ 2256 } \ 2257 \ 2258 if (unlikely(tp##_is_zero_or_denormal(xb->fld))) { \ 2259 /* \ 2260 * XB is not zero because of the above check and \ 2261 * therefore must be denormalized. \ 2262 */ \ 2263 fg_flag = 1; \ 2264 } \ 2265 } \ 2266 } \ 2267 \ 2268 env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \ 2269 } 2270 2271 VSX_TSQRT(xstsqrtdp, 1, float64, VsrD(0), -1022, 52) 2272 VSX_TSQRT(xvtsqrtdp, 2, float64, VsrD(i), -1022, 52) 2273 VSX_TSQRT(xvtsqrtsp, 4, float32, VsrW(i), -126, 23) 2274 2275 /* 2276 * VSX_MADD - VSX floating point muliply/add variations 2277 * op - instruction mnemonic 2278 * nels - number of elements (1, 2 or 4) 2279 * tp - type (float32 or float64) 2280 * fld - vsr_t field (VsrD(*) or VsrW(*)) 2281 * maddflgs - flags for the float*muladd routine that control the 2282 * various forms (madd, msub, nmadd, nmsub) 2283 * sfprf - set FPRF 2284 */ 2285 #define VSX_MADD(op, nels, tp, fld, maddflgs, sfprf, r2sp) \ 2286 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \ 2287 ppc_vsr_t *xa, ppc_vsr_t *b, ppc_vsr_t *c) \ 2288 { \ 2289 ppc_vsr_t t = *xt; \ 2290 int i; \ 2291 \ 2292 helper_reset_fpstatus(env); \ 2293 \ 2294 for (i = 0; i < nels; i++) { \ 2295 float_status tstat = env->fp_status; \ 2296 set_float_exception_flags(0, &tstat); \ 2297 if (r2sp && (tstat.float_rounding_mode == float_round_nearest_even)) {\ 2298 /* \ 2299 * Avoid double rounding errors by rounding the intermediate \ 2300 * result to odd. \ 2301 */ \ 2302 set_float_rounding_mode(float_round_to_zero, &tstat); \ 2303 t.fld = tp##_muladd(xa->fld, b->fld, c->fld, \ 2304 maddflgs, &tstat); \ 2305 t.fld |= (get_float_exception_flags(&tstat) & \ 2306 float_flag_inexact) != 0; \ 2307 } else { \ 2308 t.fld = tp##_muladd(xa->fld, b->fld, c->fld, \ 2309 maddflgs, &tstat); \ 2310 } \ 2311 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \ 2312 \ 2313 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \ 2314 tp##_maddsub_update_excp(env, xa->fld, b->fld, \ 2315 c->fld, maddflgs, GETPC()); \ 2316 } \ 2317 \ 2318 if (r2sp) { \ 2319 t.fld = helper_frsp(env, t.fld); \ 2320 } \ 2321 \ 2322 if (sfprf) { \ 2323 helper_compute_fprf_float64(env, t.fld); \ 2324 } \ 2325 } \ 2326 *xt = t; \ 2327 do_float_check_status(env, GETPC()); \ 2328 } 2329 2330 VSX_MADD(xsmadddp, 1, float64, VsrD(0), MADD_FLGS, 1, 0) 2331 VSX_MADD(xsmsubdp, 1, float64, VsrD(0), MSUB_FLGS, 1, 0) 2332 VSX_MADD(xsnmadddp, 1, float64, VsrD(0), NMADD_FLGS, 1, 0) 2333 VSX_MADD(xsnmsubdp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 0) 2334 VSX_MADD(xsmaddsp, 1, float64, VsrD(0), MADD_FLGS, 1, 1) 2335 VSX_MADD(xsmsubsp, 1, float64, VsrD(0), MSUB_FLGS, 1, 1) 2336 VSX_MADD(xsnmaddsp, 1, float64, VsrD(0), NMADD_FLGS, 1, 1) 2337 VSX_MADD(xsnmsubsp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 1) 2338 2339 VSX_MADD(xvmadddp, 2, float64, VsrD(i), MADD_FLGS, 0, 0) 2340 VSX_MADD(xvmsubdp, 2, float64, VsrD(i), MSUB_FLGS, 0, 0) 2341 VSX_MADD(xvnmadddp, 2, float64, VsrD(i), NMADD_FLGS, 0, 0) 2342 VSX_MADD(xvnmsubdp, 2, float64, VsrD(i), NMSUB_FLGS, 0, 0) 2343 2344 VSX_MADD(xvmaddsp, 4, float32, VsrW(i), MADD_FLGS, 0, 0) 2345 VSX_MADD(xvmsubsp, 4, float32, VsrW(i), MSUB_FLGS, 0, 0) 2346 VSX_MADD(xvnmaddsp, 4, float32, VsrW(i), NMADD_FLGS, 0, 0) 2347 VSX_MADD(xvnmsubsp, 4, float32, VsrW(i), NMSUB_FLGS, 0, 0) 2348 2349 /* 2350 * VSX_SCALAR_CMP_DP - VSX scalar floating point compare double precision 2351 * op - instruction mnemonic 2352 * cmp - comparison operation 2353 * exp - expected result of comparison 2354 * svxvc - set VXVC bit 2355 */ 2356 #define VSX_SCALAR_CMP_DP(op, cmp, exp, svxvc) \ 2357 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \ 2358 ppc_vsr_t *xa, ppc_vsr_t *xb) \ 2359 { \ 2360 ppc_vsr_t t = *xt; \ 2361 bool vxsnan_flag = false, vxvc_flag = false, vex_flag = false; \ 2362 \ 2363 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) || \ 2364 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \ 2365 vxsnan_flag = true; \ 2366 if (fpscr_ve == 0 && svxvc) { \ 2367 vxvc_flag = true; \ 2368 } \ 2369 } else if (svxvc) { \ 2370 vxvc_flag = float64_is_quiet_nan(xa->VsrD(0), &env->fp_status) || \ 2371 float64_is_quiet_nan(xb->VsrD(0), &env->fp_status); \ 2372 } \ 2373 if (vxsnan_flag) { \ 2374 float_invalid_op_vxsnan(env, GETPC()); \ 2375 } \ 2376 if (vxvc_flag) { \ 2377 float_invalid_op_vxvc(env, 0, GETPC()); \ 2378 } \ 2379 vex_flag = fpscr_ve && (vxvc_flag || vxsnan_flag); \ 2380 \ 2381 if (!vex_flag) { \ 2382 if (float64_##cmp(xb->VsrD(0), xa->VsrD(0), \ 2383 &env->fp_status) == exp) { \ 2384 t.VsrD(0) = -1; \ 2385 t.VsrD(1) = 0; \ 2386 } else { \ 2387 t.VsrD(0) = 0; \ 2388 t.VsrD(1) = 0; \ 2389 } \ 2390 } \ 2391 *xt = t; \ 2392 do_float_check_status(env, GETPC()); \ 2393 } 2394 2395 VSX_SCALAR_CMP_DP(xscmpeqdp, eq, 1, 0) 2396 VSX_SCALAR_CMP_DP(xscmpgedp, le, 1, 1) 2397 VSX_SCALAR_CMP_DP(xscmpgtdp, lt, 1, 1) 2398 VSX_SCALAR_CMP_DP(xscmpnedp, eq, 0, 0) 2399 2400 void helper_xscmpexpdp(CPUPPCState *env, uint32_t opcode, 2401 ppc_vsr_t *xa, ppc_vsr_t *xb) 2402 { 2403 int64_t exp_a, exp_b; 2404 uint32_t cc; 2405 2406 exp_a = extract64(xa->VsrD(0), 52, 11); 2407 exp_b = extract64(xb->VsrD(0), 52, 11); 2408 2409 if (unlikely(float64_is_any_nan(xa->VsrD(0)) || 2410 float64_is_any_nan(xb->VsrD(0)))) { 2411 cc = CRF_SO; 2412 } else { 2413 if (exp_a < exp_b) { 2414 cc = CRF_LT; 2415 } else if (exp_a > exp_b) { 2416 cc = CRF_GT; 2417 } else { 2418 cc = CRF_EQ; 2419 } 2420 } 2421 2422 env->fpscr &= ~(0x0F << FPSCR_FPRF); 2423 env->fpscr |= cc << FPSCR_FPRF; 2424 env->crf[BF(opcode)] = cc; 2425 2426 do_float_check_status(env, GETPC()); 2427 } 2428 2429 void helper_xscmpexpqp(CPUPPCState *env, uint32_t opcode, 2430 ppc_vsr_t *xa, ppc_vsr_t *xb) 2431 { 2432 int64_t exp_a, exp_b; 2433 uint32_t cc; 2434 2435 exp_a = extract64(xa->VsrD(0), 48, 15); 2436 exp_b = extract64(xb->VsrD(0), 48, 15); 2437 2438 if (unlikely(float128_is_any_nan(xa->f128) || 2439 float128_is_any_nan(xb->f128))) { 2440 cc = CRF_SO; 2441 } else { 2442 if (exp_a < exp_b) { 2443 cc = CRF_LT; 2444 } else if (exp_a > exp_b) { 2445 cc = CRF_GT; 2446 } else { 2447 cc = CRF_EQ; 2448 } 2449 } 2450 2451 env->fpscr &= ~(0x0F << FPSCR_FPRF); 2452 env->fpscr |= cc << FPSCR_FPRF; 2453 env->crf[BF(opcode)] = cc; 2454 2455 do_float_check_status(env, GETPC()); 2456 } 2457 2458 #define VSX_SCALAR_CMP(op, ordered) \ 2459 void helper_##op(CPUPPCState *env, uint32_t opcode, \ 2460 ppc_vsr_t *xa, ppc_vsr_t *xb) \ 2461 { \ 2462 uint32_t cc = 0; \ 2463 bool vxsnan_flag = false, vxvc_flag = false; \ 2464 \ 2465 helper_reset_fpstatus(env); \ 2466 \ 2467 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) || \ 2468 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \ 2469 vxsnan_flag = true; \ 2470 cc = CRF_SO; \ 2471 if (fpscr_ve == 0 && ordered) { \ 2472 vxvc_flag = true; \ 2473 } \ 2474 } else if (float64_is_quiet_nan(xa->VsrD(0), &env->fp_status) || \ 2475 float64_is_quiet_nan(xb->VsrD(0), &env->fp_status)) { \ 2476 cc = CRF_SO; \ 2477 if (ordered) { \ 2478 vxvc_flag = true; \ 2479 } \ 2480 } \ 2481 if (vxsnan_flag) { \ 2482 float_invalid_op_vxsnan(env, GETPC()); \ 2483 } \ 2484 if (vxvc_flag) { \ 2485 float_invalid_op_vxvc(env, 0, GETPC()); \ 2486 } \ 2487 \ 2488 if (float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) { \ 2489 cc |= CRF_LT; \ 2490 } else if (!float64_le(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) { \ 2491 cc |= CRF_GT; \ 2492 } else { \ 2493 cc |= CRF_EQ; \ 2494 } \ 2495 \ 2496 env->fpscr &= ~(0x0F << FPSCR_FPRF); \ 2497 env->fpscr |= cc << FPSCR_FPRF; \ 2498 env->crf[BF(opcode)] = cc; \ 2499 \ 2500 do_float_check_status(env, GETPC()); \ 2501 } 2502 2503 VSX_SCALAR_CMP(xscmpodp, 1) 2504 VSX_SCALAR_CMP(xscmpudp, 0) 2505 2506 #define VSX_SCALAR_CMPQ(op, ordered) \ 2507 void helper_##op(CPUPPCState *env, uint32_t opcode, \ 2508 ppc_vsr_t *xa, ppc_vsr_t *xb) \ 2509 { \ 2510 uint32_t cc = 0; \ 2511 bool vxsnan_flag = false, vxvc_flag = false; \ 2512 \ 2513 helper_reset_fpstatus(env); \ 2514 \ 2515 if (float128_is_signaling_nan(xa->f128, &env->fp_status) || \ 2516 float128_is_signaling_nan(xb->f128, &env->fp_status)) { \ 2517 vxsnan_flag = true; \ 2518 cc = CRF_SO; \ 2519 if (fpscr_ve == 0 && ordered) { \ 2520 vxvc_flag = true; \ 2521 } \ 2522 } else if (float128_is_quiet_nan(xa->f128, &env->fp_status) || \ 2523 float128_is_quiet_nan(xb->f128, &env->fp_status)) { \ 2524 cc = CRF_SO; \ 2525 if (ordered) { \ 2526 vxvc_flag = true; \ 2527 } \ 2528 } \ 2529 if (vxsnan_flag) { \ 2530 float_invalid_op_vxsnan(env, GETPC()); \ 2531 } \ 2532 if (vxvc_flag) { \ 2533 float_invalid_op_vxvc(env, 0, GETPC()); \ 2534 } \ 2535 \ 2536 if (float128_lt(xa->f128, xb->f128, &env->fp_status)) { \ 2537 cc |= CRF_LT; \ 2538 } else if (!float128_le(xa->f128, xb->f128, &env->fp_status)) { \ 2539 cc |= CRF_GT; \ 2540 } else { \ 2541 cc |= CRF_EQ; \ 2542 } \ 2543 \ 2544 env->fpscr &= ~(0x0F << FPSCR_FPRF); \ 2545 env->fpscr |= cc << FPSCR_FPRF; \ 2546 env->crf[BF(opcode)] = cc; \ 2547 \ 2548 do_float_check_status(env, GETPC()); \ 2549 } 2550 2551 VSX_SCALAR_CMPQ(xscmpoqp, 1) 2552 VSX_SCALAR_CMPQ(xscmpuqp, 0) 2553 2554 /* 2555 * VSX_MAX_MIN - VSX floating point maximum/minimum 2556 * name - instruction mnemonic 2557 * op - operation (max or min) 2558 * nels - number of elements (1, 2 or 4) 2559 * tp - type (float32 or float64) 2560 * fld - vsr_t field (VsrD(*) or VsrW(*)) 2561 */ 2562 #define VSX_MAX_MIN(name, op, nels, tp, fld) \ 2563 void helper_##name(CPUPPCState *env, ppc_vsr_t *xt, \ 2564 ppc_vsr_t *xa, ppc_vsr_t *xb) \ 2565 { \ 2566 ppc_vsr_t t = *xt; \ 2567 int i; \ 2568 \ 2569 for (i = 0; i < nels; i++) { \ 2570 t.fld = tp##_##op(xa->fld, xb->fld, &env->fp_status); \ 2571 if (unlikely(tp##_is_signaling_nan(xa->fld, &env->fp_status) || \ 2572 tp##_is_signaling_nan(xb->fld, &env->fp_status))) { \ 2573 float_invalid_op_vxsnan(env, GETPC()); \ 2574 } \ 2575 } \ 2576 \ 2577 *xt = t; \ 2578 do_float_check_status(env, GETPC()); \ 2579 } 2580 2581 VSX_MAX_MIN(xsmaxdp, maxnum, 1, float64, VsrD(0)) 2582 VSX_MAX_MIN(xvmaxdp, maxnum, 2, float64, VsrD(i)) 2583 VSX_MAX_MIN(xvmaxsp, maxnum, 4, float32, VsrW(i)) 2584 VSX_MAX_MIN(xsmindp, minnum, 1, float64, VsrD(0)) 2585 VSX_MAX_MIN(xvmindp, minnum, 2, float64, VsrD(i)) 2586 VSX_MAX_MIN(xvminsp, minnum, 4, float32, VsrW(i)) 2587 2588 #define VSX_MAX_MINC(name, max) \ 2589 void helper_##name(CPUPPCState *env, uint32_t opcode, \ 2590 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb) \ 2591 { \ 2592 ppc_vsr_t t = *xt; \ 2593 bool vxsnan_flag = false, vex_flag = false; \ 2594 \ 2595 if (unlikely(float64_is_any_nan(xa->VsrD(0)) || \ 2596 float64_is_any_nan(xb->VsrD(0)))) { \ 2597 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) || \ 2598 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \ 2599 vxsnan_flag = true; \ 2600 } \ 2601 t.VsrD(0) = xb->VsrD(0); \ 2602 } else if ((max && \ 2603 !float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) || \ 2604 (!max && \ 2605 float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status))) { \ 2606 t.VsrD(0) = xa->VsrD(0); \ 2607 } else { \ 2608 t.VsrD(0) = xb->VsrD(0); \ 2609 } \ 2610 \ 2611 vex_flag = fpscr_ve & vxsnan_flag; \ 2612 if (vxsnan_flag) { \ 2613 float_invalid_op_vxsnan(env, GETPC()); \ 2614 } \ 2615 if (!vex_flag) { \ 2616 *xt = t; \ 2617 } \ 2618 } \ 2619 2620 VSX_MAX_MINC(xsmaxcdp, 1); 2621 VSX_MAX_MINC(xsmincdp, 0); 2622 2623 #define VSX_MAX_MINJ(name, max) \ 2624 void helper_##name(CPUPPCState *env, uint32_t opcode, \ 2625 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb) \ 2626 { \ 2627 ppc_vsr_t t = *xt; \ 2628 bool vxsnan_flag = false, vex_flag = false; \ 2629 \ 2630 if (unlikely(float64_is_any_nan(xa->VsrD(0)))) { \ 2631 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status)) { \ 2632 vxsnan_flag = true; \ 2633 } \ 2634 t.VsrD(0) = xa->VsrD(0); \ 2635 } else if (unlikely(float64_is_any_nan(xb->VsrD(0)))) { \ 2636 if (float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \ 2637 vxsnan_flag = true; \ 2638 } \ 2639 t.VsrD(0) = xb->VsrD(0); \ 2640 } else if (float64_is_zero(xa->VsrD(0)) && \ 2641 float64_is_zero(xb->VsrD(0))) { \ 2642 if (max) { \ 2643 if (!float64_is_neg(xa->VsrD(0)) || \ 2644 !float64_is_neg(xb->VsrD(0))) { \ 2645 t.VsrD(0) = 0ULL; \ 2646 } else { \ 2647 t.VsrD(0) = 0x8000000000000000ULL; \ 2648 } \ 2649 } else { \ 2650 if (float64_is_neg(xa->VsrD(0)) || \ 2651 float64_is_neg(xb->VsrD(0))) { \ 2652 t.VsrD(0) = 0x8000000000000000ULL; \ 2653 } else { \ 2654 t.VsrD(0) = 0ULL; \ 2655 } \ 2656 } \ 2657 } else if ((max && \ 2658 !float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) || \ 2659 (!max && \ 2660 float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status))) { \ 2661 t.VsrD(0) = xa->VsrD(0); \ 2662 } else { \ 2663 t.VsrD(0) = xb->VsrD(0); \ 2664 } \ 2665 \ 2666 vex_flag = fpscr_ve & vxsnan_flag; \ 2667 if (vxsnan_flag) { \ 2668 float_invalid_op_vxsnan(env, GETPC()); \ 2669 } \ 2670 if (!vex_flag) { \ 2671 *xt = t; \ 2672 } \ 2673 } \ 2674 2675 VSX_MAX_MINJ(xsmaxjdp, 1); 2676 VSX_MAX_MINJ(xsminjdp, 0); 2677 2678 /* 2679 * VSX_CMP - VSX floating point compare 2680 * op - instruction mnemonic 2681 * nels - number of elements (1, 2 or 4) 2682 * tp - type (float32 or float64) 2683 * fld - vsr_t field (VsrD(*) or VsrW(*)) 2684 * cmp - comparison operation 2685 * svxvc - set VXVC bit 2686 * exp - expected result of comparison 2687 */ 2688 #define VSX_CMP(op, nels, tp, fld, cmp, svxvc, exp) \ 2689 uint32_t helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \ 2690 ppc_vsr_t *xa, ppc_vsr_t *xb) \ 2691 { \ 2692 ppc_vsr_t t = *xt; \ 2693 uint32_t crf6 = 0; \ 2694 int i; \ 2695 int all_true = 1; \ 2696 int all_false = 1; \ 2697 \ 2698 for (i = 0; i < nels; i++) { \ 2699 if (unlikely(tp##_is_any_nan(xa->fld) || \ 2700 tp##_is_any_nan(xb->fld))) { \ 2701 if (tp##_is_signaling_nan(xa->fld, &env->fp_status) || \ 2702 tp##_is_signaling_nan(xb->fld, &env->fp_status)) { \ 2703 float_invalid_op_vxsnan(env, GETPC()); \ 2704 } \ 2705 if (svxvc) { \ 2706 float_invalid_op_vxvc(env, 0, GETPC()); \ 2707 } \ 2708 t.fld = 0; \ 2709 all_true = 0; \ 2710 } else { \ 2711 if (tp##_##cmp(xb->fld, xa->fld, &env->fp_status) == exp) { \ 2712 t.fld = -1; \ 2713 all_false = 0; \ 2714 } else { \ 2715 t.fld = 0; \ 2716 all_true = 0; \ 2717 } \ 2718 } \ 2719 } \ 2720 \ 2721 *xt = t; \ 2722 crf6 = (all_true ? 0x8 : 0) | (all_false ? 0x2 : 0); \ 2723 return crf6; \ 2724 } 2725 2726 VSX_CMP(xvcmpeqdp, 2, float64, VsrD(i), eq, 0, 1) 2727 VSX_CMP(xvcmpgedp, 2, float64, VsrD(i), le, 1, 1) 2728 VSX_CMP(xvcmpgtdp, 2, float64, VsrD(i), lt, 1, 1) 2729 VSX_CMP(xvcmpnedp, 2, float64, VsrD(i), eq, 0, 0) 2730 VSX_CMP(xvcmpeqsp, 4, float32, VsrW(i), eq, 0, 1) 2731 VSX_CMP(xvcmpgesp, 4, float32, VsrW(i), le, 1, 1) 2732 VSX_CMP(xvcmpgtsp, 4, float32, VsrW(i), lt, 1, 1) 2733 VSX_CMP(xvcmpnesp, 4, float32, VsrW(i), eq, 0, 0) 2734 2735 /* 2736 * VSX_CVT_FP_TO_FP - VSX floating point/floating point conversion 2737 * op - instruction mnemonic 2738 * nels - number of elements (1, 2 or 4) 2739 * stp - source type (float32 or float64) 2740 * ttp - target type (float32 or float64) 2741 * sfld - source vsr_t field 2742 * tfld - target vsr_t field (f32 or f64) 2743 * sfprf - set FPRF 2744 */ 2745 #define VSX_CVT_FP_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf) \ 2746 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \ 2747 { \ 2748 ppc_vsr_t t = *xt; \ 2749 int i; \ 2750 \ 2751 for (i = 0; i < nels; i++) { \ 2752 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \ 2753 if (unlikely(stp##_is_signaling_nan(xb->sfld, \ 2754 &env->fp_status))) { \ 2755 float_invalid_op_vxsnan(env, GETPC()); \ 2756 t.tfld = ttp##_snan_to_qnan(t.tfld); \ 2757 } \ 2758 if (sfprf) { \ 2759 helper_compute_fprf_##ttp(env, t.tfld); \ 2760 } \ 2761 } \ 2762 \ 2763 *xt = t; \ 2764 do_float_check_status(env, GETPC()); \ 2765 } 2766 2767 VSX_CVT_FP_TO_FP(xscvdpsp, 1, float64, float32, VsrD(0), VsrW(0), 1) 2768 VSX_CVT_FP_TO_FP(xscvspdp, 1, float32, float64, VsrW(0), VsrD(0), 1) 2769 VSX_CVT_FP_TO_FP(xvcvdpsp, 2, float64, float32, VsrD(i), VsrW(2 * i), 0) 2770 VSX_CVT_FP_TO_FP(xvcvspdp, 2, float32, float64, VsrW(2 * i), VsrD(i), 0) 2771 2772 /* 2773 * VSX_CVT_FP_TO_FP_VECTOR - VSX floating point/floating point conversion 2774 * op - instruction mnemonic 2775 * nels - number of elements (1, 2 or 4) 2776 * stp - source type (float32 or float64) 2777 * ttp - target type (float32 or float64) 2778 * sfld - source vsr_t field 2779 * tfld - target vsr_t field (f32 or f64) 2780 * sfprf - set FPRF 2781 */ 2782 #define VSX_CVT_FP_TO_FP_VECTOR(op, nels, stp, ttp, sfld, tfld, sfprf) \ 2783 void helper_##op(CPUPPCState *env, uint32_t opcode, \ 2784 ppc_vsr_t *xt, ppc_vsr_t *xb) \ 2785 { \ 2786 ppc_vsr_t t = *xt; \ 2787 int i; \ 2788 \ 2789 for (i = 0; i < nels; i++) { \ 2790 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \ 2791 if (unlikely(stp##_is_signaling_nan(xb->sfld, \ 2792 &env->fp_status))) { \ 2793 float_invalid_op_vxsnan(env, GETPC()); \ 2794 t.tfld = ttp##_snan_to_qnan(t.tfld); \ 2795 } \ 2796 if (sfprf) { \ 2797 helper_compute_fprf_##ttp(env, t.tfld); \ 2798 } \ 2799 } \ 2800 \ 2801 *xt = t; \ 2802 do_float_check_status(env, GETPC()); \ 2803 } 2804 2805 VSX_CVT_FP_TO_FP_VECTOR(xscvdpqp, 1, float64, float128, VsrD(0), f128, 1) 2806 2807 /* 2808 * VSX_CVT_FP_TO_FP_HP - VSX floating point/floating point conversion 2809 * involving one half precision value 2810 * op - instruction mnemonic 2811 * nels - number of elements (1, 2 or 4) 2812 * stp - source type 2813 * ttp - target type 2814 * sfld - source vsr_t field 2815 * tfld - target vsr_t field 2816 * sfprf - set FPRF 2817 */ 2818 #define VSX_CVT_FP_TO_FP_HP(op, nels, stp, ttp, sfld, tfld, sfprf) \ 2819 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \ 2820 { \ 2821 ppc_vsr_t t = { }; \ 2822 int i; \ 2823 \ 2824 for (i = 0; i < nels; i++) { \ 2825 t.tfld = stp##_to_##ttp(xb->sfld, 1, &env->fp_status); \ 2826 if (unlikely(stp##_is_signaling_nan(xb->sfld, \ 2827 &env->fp_status))) { \ 2828 float_invalid_op_vxsnan(env, GETPC()); \ 2829 t.tfld = ttp##_snan_to_qnan(t.tfld); \ 2830 } \ 2831 if (sfprf) { \ 2832 helper_compute_fprf_##ttp(env, t.tfld); \ 2833 } \ 2834 } \ 2835 \ 2836 *xt = t; \ 2837 do_float_check_status(env, GETPC()); \ 2838 } 2839 2840 VSX_CVT_FP_TO_FP_HP(xscvdphp, 1, float64, float16, VsrD(0), VsrH(3), 1) 2841 VSX_CVT_FP_TO_FP_HP(xscvhpdp, 1, float16, float64, VsrH(3), VsrD(0), 1) 2842 VSX_CVT_FP_TO_FP_HP(xvcvsphp, 4, float32, float16, VsrW(i), VsrH(2 * i + 1), 0) 2843 VSX_CVT_FP_TO_FP_HP(xvcvhpsp, 4, float16, float32, VsrH(2 * i + 1), VsrW(i), 0) 2844 2845 /* 2846 * xscvqpdp isn't using VSX_CVT_FP_TO_FP() because xscvqpdpo will be 2847 * added to this later. 2848 */ 2849 void helper_xscvqpdp(CPUPPCState *env, uint32_t opcode, 2850 ppc_vsr_t *xt, ppc_vsr_t *xb) 2851 { 2852 ppc_vsr_t t = { }; 2853 float_status tstat; 2854 2855 tstat = env->fp_status; 2856 if (unlikely(Rc(opcode) != 0)) { 2857 tstat.float_rounding_mode = float_round_to_odd; 2858 } 2859 2860 t.VsrD(0) = float128_to_float64(xb->f128, &tstat); 2861 env->fp_status.float_exception_flags |= tstat.float_exception_flags; 2862 if (unlikely(float128_is_signaling_nan(xb->f128, &tstat))) { 2863 float_invalid_op_vxsnan(env, GETPC()); 2864 t.VsrD(0) = float64_snan_to_qnan(t.VsrD(0)); 2865 } 2866 helper_compute_fprf_float64(env, t.VsrD(0)); 2867 2868 *xt = t; 2869 do_float_check_status(env, GETPC()); 2870 } 2871 2872 uint64_t helper_xscvdpspn(CPUPPCState *env, uint64_t xb) 2873 { 2874 float_status tstat = env->fp_status; 2875 set_float_exception_flags(0, &tstat); 2876 2877 return (uint64_t)float64_to_float32(xb, &tstat) << 32; 2878 } 2879 2880 uint64_t helper_xscvspdpn(CPUPPCState *env, uint64_t xb) 2881 { 2882 float_status tstat = env->fp_status; 2883 set_float_exception_flags(0, &tstat); 2884 2885 return float32_to_float64(xb >> 32, &tstat); 2886 } 2887 2888 /* 2889 * VSX_CVT_FP_TO_INT - VSX floating point to integer conversion 2890 * op - instruction mnemonic 2891 * nels - number of elements (1, 2 or 4) 2892 * stp - source type (float32 or float64) 2893 * ttp - target type (int32, uint32, int64 or uint64) 2894 * sfld - source vsr_t field 2895 * tfld - target vsr_t field 2896 * rnan - resulting NaN 2897 */ 2898 #define VSX_CVT_FP_TO_INT(op, nels, stp, ttp, sfld, tfld, rnan) \ 2899 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \ 2900 { \ 2901 int all_flags = env->fp_status.float_exception_flags, flags; \ 2902 ppc_vsr_t t = *xt; \ 2903 int i; \ 2904 \ 2905 for (i = 0; i < nels; i++) { \ 2906 env->fp_status.float_exception_flags = 0; \ 2907 t.tfld = stp##_to_##ttp##_round_to_zero(xb->sfld, &env->fp_status); \ 2908 flags = env->fp_status.float_exception_flags; \ 2909 if (unlikely(flags & float_flag_invalid)) { \ 2910 float_invalid_cvt(env, 0, GETPC(), stp##_classify(xb->sfld)); \ 2911 t.tfld = rnan; \ 2912 } \ 2913 all_flags |= flags; \ 2914 } \ 2915 \ 2916 *xt = t; \ 2917 env->fp_status.float_exception_flags = all_flags; \ 2918 do_float_check_status(env, GETPC()); \ 2919 } 2920 2921 VSX_CVT_FP_TO_INT(xscvdpsxds, 1, float64, int64, VsrD(0), VsrD(0), \ 2922 0x8000000000000000ULL) 2923 VSX_CVT_FP_TO_INT(xscvdpsxws, 1, float64, int32, VsrD(0), VsrW(1), \ 2924 0x80000000U) 2925 VSX_CVT_FP_TO_INT(xscvdpuxds, 1, float64, uint64, VsrD(0), VsrD(0), 0ULL) 2926 VSX_CVT_FP_TO_INT(xscvdpuxws, 1, float64, uint32, VsrD(0), VsrW(1), 0U) 2927 VSX_CVT_FP_TO_INT(xvcvdpsxds, 2, float64, int64, VsrD(i), VsrD(i), \ 2928 0x8000000000000000ULL) 2929 VSX_CVT_FP_TO_INT(xvcvdpsxws, 2, float64, int32, VsrD(i), VsrW(2 * i), \ 2930 0x80000000U) 2931 VSX_CVT_FP_TO_INT(xvcvdpuxds, 2, float64, uint64, VsrD(i), VsrD(i), 0ULL) 2932 VSX_CVT_FP_TO_INT(xvcvdpuxws, 2, float64, uint32, VsrD(i), VsrW(2 * i), 0U) 2933 VSX_CVT_FP_TO_INT(xvcvspsxds, 2, float32, int64, VsrW(2 * i), VsrD(i), \ 2934 0x8000000000000000ULL) 2935 VSX_CVT_FP_TO_INT(xvcvspsxws, 4, float32, int32, VsrW(i), VsrW(i), 0x80000000U) 2936 VSX_CVT_FP_TO_INT(xvcvspuxds, 2, float32, uint64, VsrW(2 * i), VsrD(i), 0ULL) 2937 VSX_CVT_FP_TO_INT(xvcvspuxws, 4, float32, uint32, VsrW(i), VsrW(i), 0U) 2938 2939 /* 2940 * VSX_CVT_FP_TO_INT_VECTOR - VSX floating point to integer conversion 2941 * op - instruction mnemonic 2942 * stp - source type (float32 or float64) 2943 * ttp - target type (int32, uint32, int64 or uint64) 2944 * sfld - source vsr_t field 2945 * tfld - target vsr_t field 2946 * rnan - resulting NaN 2947 */ 2948 #define VSX_CVT_FP_TO_INT_VECTOR(op, stp, ttp, sfld, tfld, rnan) \ 2949 void helper_##op(CPUPPCState *env, uint32_t opcode, \ 2950 ppc_vsr_t *xt, ppc_vsr_t *xb) \ 2951 { \ 2952 ppc_vsr_t t = { }; \ 2953 \ 2954 t.tfld = stp##_to_##ttp##_round_to_zero(xb->sfld, &env->fp_status); \ 2955 if (env->fp_status.float_exception_flags & float_flag_invalid) { \ 2956 float_invalid_cvt(env, 0, GETPC(), stp##_classify(xb->sfld)); \ 2957 t.tfld = rnan; \ 2958 } \ 2959 \ 2960 *xt = t; \ 2961 do_float_check_status(env, GETPC()); \ 2962 } 2963 2964 VSX_CVT_FP_TO_INT_VECTOR(xscvqpsdz, float128, int64, f128, VsrD(0), \ 2965 0x8000000000000000ULL) 2966 2967 VSX_CVT_FP_TO_INT_VECTOR(xscvqpswz, float128, int32, f128, VsrD(0), \ 2968 0xffffffff80000000ULL) 2969 VSX_CVT_FP_TO_INT_VECTOR(xscvqpudz, float128, uint64, f128, VsrD(0), 0x0ULL) 2970 VSX_CVT_FP_TO_INT_VECTOR(xscvqpuwz, float128, uint32, f128, VsrD(0), 0x0ULL) 2971 2972 /* 2973 * VSX_CVT_INT_TO_FP - VSX integer to floating point conversion 2974 * op - instruction mnemonic 2975 * nels - number of elements (1, 2 or 4) 2976 * stp - source type (int32, uint32, int64 or uint64) 2977 * ttp - target type (float32 or float64) 2978 * sfld - source vsr_t field 2979 * tfld - target vsr_t field 2980 * jdef - definition of the j index (i or 2*i) 2981 * sfprf - set FPRF 2982 */ 2983 #define VSX_CVT_INT_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf, r2sp) \ 2984 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \ 2985 { \ 2986 ppc_vsr_t t = *xt; \ 2987 int i; \ 2988 \ 2989 for (i = 0; i < nels; i++) { \ 2990 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \ 2991 if (r2sp) { \ 2992 t.tfld = helper_frsp(env, t.tfld); \ 2993 } \ 2994 if (sfprf) { \ 2995 helper_compute_fprf_float64(env, t.tfld); \ 2996 } \ 2997 } \ 2998 \ 2999 *xt = t; \ 3000 do_float_check_status(env, GETPC()); \ 3001 } 3002 3003 VSX_CVT_INT_TO_FP(xscvsxddp, 1, int64, float64, VsrD(0), VsrD(0), 1, 0) 3004 VSX_CVT_INT_TO_FP(xscvuxddp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 0) 3005 VSX_CVT_INT_TO_FP(xscvsxdsp, 1, int64, float64, VsrD(0), VsrD(0), 1, 1) 3006 VSX_CVT_INT_TO_FP(xscvuxdsp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 1) 3007 VSX_CVT_INT_TO_FP(xvcvsxddp, 2, int64, float64, VsrD(i), VsrD(i), 0, 0) 3008 VSX_CVT_INT_TO_FP(xvcvuxddp, 2, uint64, float64, VsrD(i), VsrD(i), 0, 0) 3009 VSX_CVT_INT_TO_FP(xvcvsxwdp, 2, int32, float64, VsrW(2 * i), VsrD(i), 0, 0) 3010 VSX_CVT_INT_TO_FP(xvcvuxwdp, 2, uint64, float64, VsrW(2 * i), VsrD(i), 0, 0) 3011 VSX_CVT_INT_TO_FP(xvcvsxdsp, 2, int64, float32, VsrD(i), VsrW(2 * i), 0, 0) 3012 VSX_CVT_INT_TO_FP(xvcvuxdsp, 2, uint64, float32, VsrD(i), VsrW(2 * i), 0, 0) 3013 VSX_CVT_INT_TO_FP(xvcvsxwsp, 4, int32, float32, VsrW(i), VsrW(i), 0, 0) 3014 VSX_CVT_INT_TO_FP(xvcvuxwsp, 4, uint32, float32, VsrW(i), VsrW(i), 0, 0) 3015 3016 /* 3017 * VSX_CVT_INT_TO_FP_VECTOR - VSX integer to floating point conversion 3018 * op - instruction mnemonic 3019 * stp - source type (int32, uint32, int64 or uint64) 3020 * ttp - target type (float32 or float64) 3021 * sfld - source vsr_t field 3022 * tfld - target vsr_t field 3023 */ 3024 #define VSX_CVT_INT_TO_FP_VECTOR(op, stp, ttp, sfld, tfld) \ 3025 void helper_##op(CPUPPCState *env, uint32_t opcode, \ 3026 ppc_vsr_t *xt, ppc_vsr_t *xb) \ 3027 { \ 3028 ppc_vsr_t t = *xt; \ 3029 \ 3030 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \ 3031 helper_compute_fprf_##ttp(env, t.tfld); \ 3032 \ 3033 *xt = t; \ 3034 do_float_check_status(env, GETPC()); \ 3035 } 3036 3037 VSX_CVT_INT_TO_FP_VECTOR(xscvsdqp, int64, float128, VsrD(0), f128) 3038 VSX_CVT_INT_TO_FP_VECTOR(xscvudqp, uint64, float128, VsrD(0), f128) 3039 3040 /* 3041 * For "use current rounding mode", define a value that will not be 3042 * one of the existing rounding model enums. 3043 */ 3044 #define FLOAT_ROUND_CURRENT (float_round_nearest_even + float_round_down + \ 3045 float_round_up + float_round_to_zero) 3046 3047 /* 3048 * VSX_ROUND - VSX floating point round 3049 * op - instruction mnemonic 3050 * nels - number of elements (1, 2 or 4) 3051 * tp - type (float32 or float64) 3052 * fld - vsr_t field (VsrD(*) or VsrW(*)) 3053 * rmode - rounding mode 3054 * sfprf - set FPRF 3055 */ 3056 #define VSX_ROUND(op, nels, tp, fld, rmode, sfprf) \ 3057 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \ 3058 { \ 3059 ppc_vsr_t t = *xt; \ 3060 int i; \ 3061 \ 3062 if (rmode != FLOAT_ROUND_CURRENT) { \ 3063 set_float_rounding_mode(rmode, &env->fp_status); \ 3064 } \ 3065 \ 3066 for (i = 0; i < nels; i++) { \ 3067 if (unlikely(tp##_is_signaling_nan(xb->fld, \ 3068 &env->fp_status))) { \ 3069 float_invalid_op_vxsnan(env, GETPC()); \ 3070 t.fld = tp##_snan_to_qnan(xb->fld); \ 3071 } else { \ 3072 t.fld = tp##_round_to_int(xb->fld, &env->fp_status); \ 3073 } \ 3074 if (sfprf) { \ 3075 helper_compute_fprf_float64(env, t.fld); \ 3076 } \ 3077 } \ 3078 \ 3079 /* \ 3080 * If this is not a "use current rounding mode" instruction, \ 3081 * then inhibit setting of the XX bit and restore rounding \ 3082 * mode from FPSCR \ 3083 */ \ 3084 if (rmode != FLOAT_ROUND_CURRENT) { \ 3085 fpscr_set_rounding_mode(env); \ 3086 env->fp_status.float_exception_flags &= ~float_flag_inexact; \ 3087 } \ 3088 \ 3089 *xt = t; \ 3090 do_float_check_status(env, GETPC()); \ 3091 } 3092 3093 VSX_ROUND(xsrdpi, 1, float64, VsrD(0), float_round_ties_away, 1) 3094 VSX_ROUND(xsrdpic, 1, float64, VsrD(0), FLOAT_ROUND_CURRENT, 1) 3095 VSX_ROUND(xsrdpim, 1, float64, VsrD(0), float_round_down, 1) 3096 VSX_ROUND(xsrdpip, 1, float64, VsrD(0), float_round_up, 1) 3097 VSX_ROUND(xsrdpiz, 1, float64, VsrD(0), float_round_to_zero, 1) 3098 3099 VSX_ROUND(xvrdpi, 2, float64, VsrD(i), float_round_ties_away, 0) 3100 VSX_ROUND(xvrdpic, 2, float64, VsrD(i), FLOAT_ROUND_CURRENT, 0) 3101 VSX_ROUND(xvrdpim, 2, float64, VsrD(i), float_round_down, 0) 3102 VSX_ROUND(xvrdpip, 2, float64, VsrD(i), float_round_up, 0) 3103 VSX_ROUND(xvrdpiz, 2, float64, VsrD(i), float_round_to_zero, 0) 3104 3105 VSX_ROUND(xvrspi, 4, float32, VsrW(i), float_round_ties_away, 0) 3106 VSX_ROUND(xvrspic, 4, float32, VsrW(i), FLOAT_ROUND_CURRENT, 0) 3107 VSX_ROUND(xvrspim, 4, float32, VsrW(i), float_round_down, 0) 3108 VSX_ROUND(xvrspip, 4, float32, VsrW(i), float_round_up, 0) 3109 VSX_ROUND(xvrspiz, 4, float32, VsrW(i), float_round_to_zero, 0) 3110 3111 uint64_t helper_xsrsp(CPUPPCState *env, uint64_t xb) 3112 { 3113 helper_reset_fpstatus(env); 3114 3115 uint64_t xt = helper_frsp(env, xb); 3116 3117 helper_compute_fprf_float64(env, xt); 3118 do_float_check_status(env, GETPC()); 3119 return xt; 3120 } 3121 3122 #define VSX_XXPERM(op, indexed) \ 3123 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \ 3124 ppc_vsr_t *xa, ppc_vsr_t *pcv) \ 3125 { \ 3126 ppc_vsr_t t = *xt; \ 3127 int i, idx; \ 3128 \ 3129 for (i = 0; i < 16; i++) { \ 3130 idx = pcv->VsrB(i) & 0x1F; \ 3131 if (indexed) { \ 3132 idx = 31 - idx; \ 3133 } \ 3134 t.VsrB(i) = (idx <= 15) ? xa->VsrB(idx) \ 3135 : xt->VsrB(idx - 16); \ 3136 } \ 3137 *xt = t; \ 3138 } 3139 3140 VSX_XXPERM(xxperm, 0) 3141 VSX_XXPERM(xxpermr, 1) 3142 3143 void helper_xvxsigsp(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) 3144 { 3145 ppc_vsr_t t = { }; 3146 uint32_t exp, i, fraction; 3147 3148 for (i = 0; i < 4; i++) { 3149 exp = (xb->VsrW(i) >> 23) & 0xFF; 3150 fraction = xb->VsrW(i) & 0x7FFFFF; 3151 if (exp != 0 && exp != 255) { 3152 t.VsrW(i) = fraction | 0x00800000; 3153 } else { 3154 t.VsrW(i) = fraction; 3155 } 3156 } 3157 *xt = t; 3158 } 3159 3160 /* 3161 * VSX_TEST_DC - VSX floating point test data class 3162 * op - instruction mnemonic 3163 * nels - number of elements (1, 2 or 4) 3164 * xbn - VSR register number 3165 * tp - type (float32 or float64) 3166 * fld - vsr_t field (VsrD(*) or VsrW(*)) 3167 * tfld - target vsr_t field (VsrD(*) or VsrW(*)) 3168 * fld_max - target field max 3169 * scrf - set result in CR and FPCC 3170 */ 3171 #define VSX_TEST_DC(op, nels, xbn, tp, fld, tfld, fld_max, scrf) \ 3172 void helper_##op(CPUPPCState *env, uint32_t opcode) \ 3173 { \ 3174 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \ 3175 ppc_vsr_t *xb = &env->vsr[xbn]; \ 3176 ppc_vsr_t t = { }; \ 3177 uint32_t i, sign, dcmx; \ 3178 uint32_t cc, match = 0; \ 3179 \ 3180 if (!scrf) { \ 3181 dcmx = DCMX_XV(opcode); \ 3182 } else { \ 3183 t = *xt; \ 3184 dcmx = DCMX(opcode); \ 3185 } \ 3186 \ 3187 for (i = 0; i < nels; i++) { \ 3188 sign = tp##_is_neg(xb->fld); \ 3189 if (tp##_is_any_nan(xb->fld)) { \ 3190 match = extract32(dcmx, 6, 1); \ 3191 } else if (tp##_is_infinity(xb->fld)) { \ 3192 match = extract32(dcmx, 4 + !sign, 1); \ 3193 } else if (tp##_is_zero(xb->fld)) { \ 3194 match = extract32(dcmx, 2 + !sign, 1); \ 3195 } else if (tp##_is_zero_or_denormal(xb->fld)) { \ 3196 match = extract32(dcmx, 0 + !sign, 1); \ 3197 } \ 3198 \ 3199 if (scrf) { \ 3200 cc = sign << CRF_LT_BIT | match << CRF_EQ_BIT; \ 3201 env->fpscr &= ~(0x0F << FPSCR_FPRF); \ 3202 env->fpscr |= cc << FPSCR_FPRF; \ 3203 env->crf[BF(opcode)] = cc; \ 3204 } else { \ 3205 t.tfld = match ? fld_max : 0; \ 3206 } \ 3207 match = 0; \ 3208 } \ 3209 if (!scrf) { \ 3210 *xt = t; \ 3211 } \ 3212 } 3213 3214 VSX_TEST_DC(xvtstdcdp, 2, xB(opcode), float64, VsrD(i), VsrD(i), UINT64_MAX, 0) 3215 VSX_TEST_DC(xvtstdcsp, 4, xB(opcode), float32, VsrW(i), VsrW(i), UINT32_MAX, 0) 3216 VSX_TEST_DC(xststdcdp, 1, xB(opcode), float64, VsrD(0), VsrD(0), 0, 1) 3217 VSX_TEST_DC(xststdcqp, 1, (rB(opcode) + 32), float128, f128, VsrD(0), 0, 1) 3218 3219 void helper_xststdcsp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xb) 3220 { 3221 uint32_t dcmx, sign, exp; 3222 uint32_t cc, match = 0, not_sp = 0; 3223 3224 dcmx = DCMX(opcode); 3225 exp = (xb->VsrD(0) >> 52) & 0x7FF; 3226 3227 sign = float64_is_neg(xb->VsrD(0)); 3228 if (float64_is_any_nan(xb->VsrD(0))) { 3229 match = extract32(dcmx, 6, 1); 3230 } else if (float64_is_infinity(xb->VsrD(0))) { 3231 match = extract32(dcmx, 4 + !sign, 1); 3232 } else if (float64_is_zero(xb->VsrD(0))) { 3233 match = extract32(dcmx, 2 + !sign, 1); 3234 } else if (float64_is_zero_or_denormal(xb->VsrD(0)) || 3235 (exp > 0 && exp < 0x381)) { 3236 match = extract32(dcmx, 0 + !sign, 1); 3237 } 3238 3239 not_sp = !float64_eq(xb->VsrD(0), 3240 float32_to_float64( 3241 float64_to_float32(xb->VsrD(0), &env->fp_status), 3242 &env->fp_status), &env->fp_status); 3243 3244 cc = sign << CRF_LT_BIT | match << CRF_EQ_BIT | not_sp << CRF_SO_BIT; 3245 env->fpscr &= ~(0x0F << FPSCR_FPRF); 3246 env->fpscr |= cc << FPSCR_FPRF; 3247 env->crf[BF(opcode)] = cc; 3248 } 3249 3250 void helper_xsrqpi(CPUPPCState *env, uint32_t opcode, 3251 ppc_vsr_t *xt, ppc_vsr_t *xb) 3252 { 3253 ppc_vsr_t t = { }; 3254 uint8_t r = Rrm(opcode); 3255 uint8_t ex = Rc(opcode); 3256 uint8_t rmc = RMC(opcode); 3257 uint8_t rmode = 0; 3258 float_status tstat; 3259 3260 helper_reset_fpstatus(env); 3261 3262 if (r == 0 && rmc == 0) { 3263 rmode = float_round_ties_away; 3264 } else if (r == 0 && rmc == 0x3) { 3265 rmode = fpscr_rn; 3266 } else if (r == 1) { 3267 switch (rmc) { 3268 case 0: 3269 rmode = float_round_nearest_even; 3270 break; 3271 case 1: 3272 rmode = float_round_to_zero; 3273 break; 3274 case 2: 3275 rmode = float_round_up; 3276 break; 3277 case 3: 3278 rmode = float_round_down; 3279 break; 3280 default: 3281 abort(); 3282 } 3283 } 3284 3285 tstat = env->fp_status; 3286 set_float_exception_flags(0, &tstat); 3287 set_float_rounding_mode(rmode, &tstat); 3288 t.f128 = float128_round_to_int(xb->f128, &tstat); 3289 env->fp_status.float_exception_flags |= tstat.float_exception_flags; 3290 3291 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { 3292 if (float128_is_signaling_nan(xb->f128, &tstat)) { 3293 float_invalid_op_vxsnan(env, GETPC()); 3294 t.f128 = float128_snan_to_qnan(t.f128); 3295 } 3296 } 3297 3298 if (ex == 0 && (tstat.float_exception_flags & float_flag_inexact)) { 3299 env->fp_status.float_exception_flags &= ~float_flag_inexact; 3300 } 3301 3302 helper_compute_fprf_float128(env, t.f128); 3303 do_float_check_status(env, GETPC()); 3304 *xt = t; 3305 } 3306 3307 void helper_xsrqpxp(CPUPPCState *env, uint32_t opcode, 3308 ppc_vsr_t *xt, ppc_vsr_t *xb) 3309 { 3310 ppc_vsr_t t = { }; 3311 uint8_t r = Rrm(opcode); 3312 uint8_t rmc = RMC(opcode); 3313 uint8_t rmode = 0; 3314 floatx80 round_res; 3315 float_status tstat; 3316 3317 helper_reset_fpstatus(env); 3318 3319 if (r == 0 && rmc == 0) { 3320 rmode = float_round_ties_away; 3321 } else if (r == 0 && rmc == 0x3) { 3322 rmode = fpscr_rn; 3323 } else if (r == 1) { 3324 switch (rmc) { 3325 case 0: 3326 rmode = float_round_nearest_even; 3327 break; 3328 case 1: 3329 rmode = float_round_to_zero; 3330 break; 3331 case 2: 3332 rmode = float_round_up; 3333 break; 3334 case 3: 3335 rmode = float_round_down; 3336 break; 3337 default: 3338 abort(); 3339 } 3340 } 3341 3342 tstat = env->fp_status; 3343 set_float_exception_flags(0, &tstat); 3344 set_float_rounding_mode(rmode, &tstat); 3345 round_res = float128_to_floatx80(xb->f128, &tstat); 3346 t.f128 = floatx80_to_float128(round_res, &tstat); 3347 env->fp_status.float_exception_flags |= tstat.float_exception_flags; 3348 3349 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { 3350 if (float128_is_signaling_nan(xb->f128, &tstat)) { 3351 float_invalid_op_vxsnan(env, GETPC()); 3352 t.f128 = float128_snan_to_qnan(t.f128); 3353 } 3354 } 3355 3356 helper_compute_fprf_float128(env, t.f128); 3357 *xt = t; 3358 do_float_check_status(env, GETPC()); 3359 } 3360 3361 void helper_xssqrtqp(CPUPPCState *env, uint32_t opcode, 3362 ppc_vsr_t *xt, ppc_vsr_t *xb) 3363 { 3364 ppc_vsr_t t = { }; 3365 float_status tstat; 3366 3367 helper_reset_fpstatus(env); 3368 3369 tstat = env->fp_status; 3370 if (unlikely(Rc(opcode) != 0)) { 3371 tstat.float_rounding_mode = float_round_to_odd; 3372 } 3373 3374 set_float_exception_flags(0, &tstat); 3375 t.f128 = float128_sqrt(xb->f128, &tstat); 3376 env->fp_status.float_exception_flags |= tstat.float_exception_flags; 3377 3378 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { 3379 if (float128_is_signaling_nan(xb->f128, &tstat)) { 3380 float_invalid_op_vxsnan(env, GETPC()); 3381 t.f128 = float128_snan_to_qnan(xb->f128); 3382 } else if (float128_is_quiet_nan(xb->f128, &tstat)) { 3383 t.f128 = xb->f128; 3384 } else if (float128_is_neg(xb->f128) && !float128_is_zero(xb->f128)) { 3385 float_invalid_op_vxsqrt(env, 1, GETPC()); 3386 t.f128 = float128_default_nan(&env->fp_status); 3387 } 3388 } 3389 3390 helper_compute_fprf_float128(env, t.f128); 3391 *xt = t; 3392 do_float_check_status(env, GETPC()); 3393 } 3394 3395 void helper_xssubqp(CPUPPCState *env, uint32_t opcode, 3396 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb) 3397 { 3398 ppc_vsr_t t = *xt; 3399 float_status tstat; 3400 3401 helper_reset_fpstatus(env); 3402 3403 tstat = env->fp_status; 3404 if (unlikely(Rc(opcode) != 0)) { 3405 tstat.float_rounding_mode = float_round_to_odd; 3406 } 3407 3408 set_float_exception_flags(0, &tstat); 3409 t.f128 = float128_sub(xa->f128, xb->f128, &tstat); 3410 env->fp_status.float_exception_flags |= tstat.float_exception_flags; 3411 3412 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { 3413 float_invalid_op_addsub(env, 1, GETPC(), 3414 float128_classify(xa->f128) | 3415 float128_classify(xb->f128)); 3416 } 3417 3418 helper_compute_fprf_float128(env, t.f128); 3419 *xt = t; 3420 do_float_check_status(env, GETPC()); 3421 } 3422