fp.isa revision 13169:eb3b2bea4231
1// -*- mode:c++ -*- 2 3// Copyright (c) 2010-2011, 2016-2018 ARM Limited 4// All rights reserved 5// 6// The license below extends only to copyright in the software and shall 7// not be construed as granting a license to any other intellectual 8// property including but not limited to intellectual property relating 9// to a hardware implementation of the functionality of the software 10// licensed hereunder. You may use the software subject to the license 11// terms below provided that you ensure that this notice is replicated 12// unmodified and in its entirety in all distributions of the software, 13// modified or unmodified, in source code or in binary form. 14// 15// Copyright (c) 2007-2008 The Florida State University 16// All rights reserved. 17// 18// Redistribution and use in source and binary forms, with or without 19// modification, are permitted provided that the following conditions are 20// met: redistributions of source code must retain the above copyright 21// notice, this list of conditions and the following disclaimer; 22// redistributions in binary form must reproduce the above copyright 23// notice, this list of conditions and the following disclaimer in the 24// documentation and/or other materials provided with the distribution; 25// neither the name of the copyright holders nor the names of its 26// contributors may be used to endorse or promote products derived from 27// this software without specific prior written permission. 28// 29// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 30// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 31// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 32// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 33// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 34// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 35// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 36// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 37// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 38// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 39// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 40// 41// Authors: Stephen Hines 42 43//////////////////////////////////////////////////////////////////// 44// 45// Floating Point operate instructions 46// 47 48output header {{ 49 50 template<template <typename T> class Base> 51 StaticInstPtr 52 newNeonMemInst(const unsigned size, 53 const ExtMachInst &machInst, 54 const RegIndex dest, const RegIndex ra, 55 const uint32_t imm, const unsigned extraMemFlags) 56 { 57 switch (size) { 58 case 0: 59 return new Base<uint8_t>(machInst, dest, ra, imm, extraMemFlags); 60 case 1: 61 return new Base<uint16_t>(machInst, dest, ra, imm, extraMemFlags); 62 case 2: 63 return new Base<uint32_t>(machInst, dest, ra, imm, extraMemFlags); 64 case 3: 65 return new Base<uint64_t>(machInst, dest, ra, imm, extraMemFlags); 66 default: 67 panic("Unrecognized width %d for Neon mem inst.\n", (1 << size)); 68 } 69 } 70 71 template<template <typename T> class Base> 72 StaticInstPtr 73 newNeonMixInst(const unsigned size, 74 const ExtMachInst &machInst, 75 const RegIndex dest, const RegIndex op1, 76 const uint32_t step) 77 { 78 switch (size) { 79 case 0: 80 return new Base<uint8_t>(machInst, dest, op1, step); 81 case 1: 82 return new Base<uint16_t>(machInst, dest, op1, step); 83 case 2: 84 return new Base<uint32_t>(machInst, dest, op1, step); 85 case 3: 86 return new Base<uint64_t>(machInst, dest, op1, step); 87 default: 88 panic("Unrecognized width %d for Neon mem inst.\n", (1 << size)); 89 } 90 } 91 92}}; 93 94let {{ 95 header_output = ''' 96 StaticInstPtr 97 decodeNeonMem(ExtMachInst machInst); 98 99 StaticInstPtr 100 decodeNeonData(ExtMachInst machInst); 101 ''' 102 103 decoder_output = ''' 104 StaticInstPtr 105 decodeNeonMem(ExtMachInst machInst) 106 { 107 const uint32_t b = bits(machInst, 11, 8); 108 const bool single = bits(machInst, 23); 109 const bool singleAll = single && (bits(b, 3, 2) == 3); 110 const bool load = bits(machInst, 21); 111 112 unsigned width = 0; 113 114 if (single) { 115 width = bits(b, 1, 0) + 1; 116 } else { 117 switch (bits(b, 3, 1)) { 118 case 0x0: width = 4; 119 break; 120 case 0x1: width = (b & 0x1) ? 2 : 1; 121 break; 122 case 0x2: width = 3; 123 break; 124 case 0x3: width = 1; 125 break; 126 case 0x4: width = 2; 127 break; 128 case 0x5: 129 if ((b & 0x1) == 0) { 130 width = 1; 131 break; 132 } 133 M5_FALLTHROUGH; 134 default: 135 return new Unknown(machInst); 136 } 137 } 138 assert(width > 0 && width <= 4); 139 140 const RegIndex rm = (RegIndex)(uint32_t)bits(machInst, 3, 0); 141 const RegIndex rn = (RegIndex)(uint32_t)bits(machInst, 19, 16); 142 const RegIndex vd = (RegIndex)(uint32_t)(bits(machInst, 15, 12) | 143 bits(machInst, 22) << 4); 144 const uint32_t type = bits(machInst, 11, 8); 145 uint32_t size = 0; 146 uint32_t align = TLB::MustBeOne; 147 unsigned inc = 1; 148 unsigned regs = 1; 149 unsigned lane = 0; 150 if (single) { 151 if (singleAll) { 152 size = bits(machInst, 7, 6); 153 bool t = bits(machInst, 5); 154 align = size | TLB::AllowUnaligned; 155 if (width == 1) { 156 regs = t ? 2 : 1; 157 inc = 1; 158 } else { 159 regs = width; 160 inc = t ? 2 : 1; 161 } 162 switch (width) { 163 case 1: 164 case 2: 165 if (bits(machInst, 4)) 166 align = size + width - 1; 167 break; 168 case 3: 169 break; 170 case 4: 171 if (size == 3) { 172 if (bits(machInst, 4) == 0) 173 return new Unknown(machInst); 174 size = 2; 175 align = 0x4; 176 } else if (size == 2) { 177 if (bits(machInst, 4)) 178 align = 0x3; 179 } else { 180 if (bits(machInst, 4)) 181 align = size + 2; 182 } 183 break; 184 } 185 } else { 186 size = bits(machInst, 11, 10); 187 align = size | TLB::AllowUnaligned; 188 regs = width; 189 unsigned indexAlign = bits(machInst, 7, 4); 190 // If width is 1, inc is always 1. That's overridden later. 191 switch (size) { 192 case 0: 193 inc = 1; 194 lane = bits(indexAlign, 3, 1); 195 break; 196 case 1: 197 inc = bits(indexAlign, 1) ? 2 : 1; 198 lane = bits(indexAlign, 3, 2); 199 break; 200 case 2: 201 inc = bits(indexAlign, 2) ? 2 : 1; 202 lane = bits(indexAlign, 3); 203 break; 204 } 205 // Override inc for width of 1. 206 if (width == 1) { 207 inc = 1; 208 } 209 switch (width) { 210 case 1: 211 switch (size) { 212 case 0: 213 break; 214 case 1: 215 if (bits(indexAlign, 0)) 216 align = 1; 217 break; 218 case 2: 219 if (bits(indexAlign, 1, 0)) 220 align = 2; 221 break; 222 } 223 break; 224 case 2: 225 if (bits(indexAlign, 0)) 226 align = size + 1; 227 break; 228 case 3: 229 break; 230 case 4: 231 switch (size) { 232 case 0: 233 case 1: 234 if (bits(indexAlign, 0)) 235 align = size + 2; 236 break; 237 case 2: 238 if (bits(indexAlign, 0)) 239 align = bits(indexAlign, 1, 0) + 2; 240 break; 241 } 242 break; 243 } 244 } 245 if (size == 0x3) { 246 return new Unknown(machInst); 247 } 248 } else { 249 size = bits(machInst, 7, 6); 250 align = bits(machInst, 5, 4); 251 if (align == 0) { 252 // @align wasn't specified, so alignment can be turned off. 253 align = size | TLB::AllowUnaligned; 254 } else { 255 align = align + 2; 256 } 257 switch (width) { 258 case 1: 259 switch (type) { 260 case 0x7: regs = 1; 261 break; 262 case 0xa: regs = 2; 263 break; 264 case 0x6: regs = 3; 265 break; 266 case 0x2: regs = 4; 267 break; 268 default: 269 return new Unknown(machInst); 270 } 271 break; 272 case 2: 273 // Regs doesn't behave exactly as it does in the manual 274 // because they loop over regs registers twice and we break 275 // it down in the macroop. 276 switch (type) { 277 case 0x8: regs = 2; inc = 1; 278 break; 279 case 0x9: regs = 2; inc = 2; 280 break; 281 case 0x3: regs = 4; inc = 2; 282 break; 283 default: 284 return new Unknown(machInst); 285 } 286 break; 287 case 3: 288 regs = 3; 289 switch (type) { 290 case 0x4: inc = 1; 291 break; 292 case 0x5: inc = 2;; 293 break; 294 default: 295 return new Unknown(machInst); 296 } 297 break; 298 case 4: 299 regs = 4; 300 switch (type) { 301 case 0: inc = 1; 302 break; 303 case 1: inc = 2; 304 break; 305 default: 306 return new Unknown(machInst); 307 } 308 break; 309 } 310 } 311 312 if (load) { 313 // Load instructions. 314 if (single) { 315 return new VldSingle(machInst, singleAll, width, rn, vd, 316 regs, inc, size, align, rm, lane); 317 } else { 318 return new VldMult(machInst, width, rn, vd, 319 regs, inc, size, align, rm); 320 } 321 } else { 322 // Store instructions. 323 if (single) { 324 if (singleAll) { 325 return new Unknown(machInst); 326 } else { 327 return new VstSingle(machInst, false, width, rn, vd, 328 regs, inc, size, align, rm, lane); 329 } 330 } else { 331 return new VstMult(machInst, width, rn, vd, 332 regs, inc, size, align, rm); 333 } 334 } 335 return new Unknown(machInst); 336 } 337 ''' 338 339 decoder_output += ''' 340 static StaticInstPtr 341 decodeNeonThreeRegistersSameLength(ExtMachInst machInst) 342 { 343 const bool u = THUMB ? bits(machInst, 28) : bits(machInst, 24); 344 const uint32_t a = bits(machInst, 11, 8); 345 const bool b = bits(machInst, 4); 346 const uint32_t c = bits(machInst, 21, 20); 347 const IntRegIndex vd = 348 (IntRegIndex)(2 * (bits(machInst, 15, 12) | 349 (bits(machInst, 22) << 4))); 350 const IntRegIndex vn = 351 (IntRegIndex)(2 * (bits(machInst, 19, 16) | 352 (bits(machInst, 7) << 4))); 353 const IntRegIndex vm = 354 (IntRegIndex)(2 * (bits(machInst, 3, 0) | 355 (bits(machInst, 5) << 4))); 356 const unsigned size = bits(machInst, 21, 20); 357 const bool q = bits(machInst, 6); 358 if (q && ((vd & 0x1) || (vn & 0x1) || (vm & 0x1))) 359 return new Unknown(machInst); 360 switch (a) { 361 case 0x0: 362 if (b) { 363 if (u) { 364 return decodeNeonUThreeReg<VqaddUD, VqaddUQ>( 365 q, size, machInst, vd, vn, vm); 366 } else { 367 return decodeNeonSThreeReg<VqaddSD, VqaddSQ>( 368 q, size, machInst, vd, vn, vm); 369 } 370 } else { 371 if (size == 3) 372 return new Unknown(machInst); 373 return decodeNeonUSThreeReg<VhaddD, VhaddQ>( 374 q, u, size, machInst, vd, vn, vm); 375 } 376 case 0x1: 377 if (!b) { 378 return decodeNeonUSThreeReg<VrhaddD, VrhaddQ>( 379 q, u, size, machInst, vd, vn, vm); 380 } else { 381 if (u) { 382 switch (c) { 383 case 0: 384 if (q) { 385 return new VeorQ<uint64_t>(machInst, vd, vn, vm); 386 } else { 387 return new VeorD<uint64_t>(machInst, vd, vn, vm); 388 } 389 case 1: 390 if (q) { 391 return new VbslQ<uint64_t>(machInst, vd, vn, vm); 392 } else { 393 return new VbslD<uint64_t>(machInst, vd, vn, vm); 394 } 395 case 2: 396 if (q) { 397 return new VbitQ<uint64_t>(machInst, vd, vn, vm); 398 } else { 399 return new VbitD<uint64_t>(machInst, vd, vn, vm); 400 } 401 case 3: 402 if (q) { 403 return new VbifQ<uint64_t>(machInst, vd, vn, vm); 404 } else { 405 return new VbifD<uint64_t>(machInst, vd, vn, vm); 406 } 407 default: 408 M5_UNREACHABLE; 409 } 410 } else { 411 switch (c) { 412 case 0: 413 if (q) { 414 return new VandQ<uint64_t>(machInst, vd, vn, vm); 415 } else { 416 return new VandD<uint64_t>(machInst, vd, vn, vm); 417 } 418 case 1: 419 if (q) { 420 return new VbicQ<uint64_t>(machInst, vd, vn, vm); 421 } else { 422 return new VbicD<uint64_t>(machInst, vd, vn, vm); 423 } 424 case 2: 425 if (vn == vm) { 426 if (q) { 427 return new VmovQ<uint64_t>( 428 machInst, vd, vn, vm); 429 } else { 430 return new VmovD<uint64_t>( 431 machInst, vd, vn, vm); 432 } 433 } else { 434 if (q) { 435 return new VorrQ<uint64_t>( 436 machInst, vd, vn, vm); 437 } else { 438 return new VorrD<uint64_t>( 439 machInst, vd, vn, vm); 440 } 441 } 442 case 3: 443 if (q) { 444 return new VornQ<uint64_t>( 445 machInst, vd, vn, vm); 446 } else { 447 return new VornD<uint64_t>( 448 machInst, vd, vn, vm); 449 } 450 default: 451 M5_UNREACHABLE; 452 } 453 } 454 } 455 case 0x2: 456 if (b) { 457 if (u) { 458 return decodeNeonUThreeReg<VqsubUD, VqsubUQ>( 459 q, size, machInst, vd, vn, vm); 460 } else { 461 return decodeNeonSThreeReg<VqsubSD, VqsubSQ>( 462 q, size, machInst, vd, vn, vm); 463 } 464 } else { 465 if (size == 3) 466 return new Unknown(machInst); 467 return decodeNeonUSThreeReg<VhsubD, VhsubQ>( 468 q, u, size, machInst, vd, vn, vm); 469 } 470 case 0x3: 471 if (b) { 472 return decodeNeonUSThreeReg<VcgeD, VcgeQ>( 473 q, u, size, machInst, vd, vn, vm); 474 } else { 475 return decodeNeonUSThreeReg<VcgtD, VcgtQ>( 476 q, u, size, machInst, vd, vn, vm); 477 } 478 case 0x4: 479 if (b) { 480 if (u) { 481 return decodeNeonUThreeReg<VqshlUD, VqshlUQ>( 482 q, size, machInst, vd, vm, vn); 483 } else { 484 return decodeNeonSThreeReg<VqshlSD, VqshlSQ>( 485 q, size, machInst, vd, vm, vn); 486 } 487 } else { 488 return decodeNeonUSThreeReg<VshlD, VshlQ>( 489 q, u, size, machInst, vd, vm, vn); 490 } 491 case 0x5: 492 if (b) { 493 if (u) { 494 return decodeNeonUThreeReg<VqrshlUD, VqrshlUQ>( 495 q, size, machInst, vd, vm, vn); 496 } else { 497 return decodeNeonSThreeReg<VqrshlSD, VqrshlSQ>( 498 q, size, machInst, vd, vm, vn); 499 } 500 } else { 501 return decodeNeonUSThreeReg<VrshlD, VrshlQ>( 502 q, u, size, machInst, vd, vm, vn); 503 } 504 case 0x6: 505 if (b) { 506 return decodeNeonUSThreeReg<VminD, VminQ>( 507 q, u, size, machInst, vd, vn, vm); 508 } else { 509 return decodeNeonUSThreeReg<VmaxD, VmaxQ>( 510 q, u, size, machInst, vd, vn, vm); 511 } 512 case 0x7: 513 if (b) { 514 return decodeNeonUSThreeReg<VabaD, VabaQ>( 515 q, u, size, machInst, vd, vn, vm); 516 } else { 517 if (bits(machInst, 23) == 1) { 518 if (q) { 519 return new Unknown(machInst); 520 } else { 521 return decodeNeonUSThreeUSReg<Vabdl>( 522 u, size, machInst, vd, vn, vm); 523 } 524 } else { 525 return decodeNeonUSThreeReg<VabdD, VabdQ>( 526 q, u, size, machInst, vd, vn, vm); 527 } 528 } 529 case 0x8: 530 if (b) { 531 if (u) { 532 return decodeNeonUThreeReg<VceqD, VceqQ>( 533 q, size, machInst, vd, vn, vm); 534 } else { 535 return decodeNeonUThreeReg<VtstD, VtstQ>( 536 q, size, machInst, vd, vn, vm); 537 } 538 } else { 539 if (u) { 540 return decodeNeonUThreeReg<NVsubD, NVsubQ>( 541 q, size, machInst, vd, vn, vm); 542 } else { 543 return decodeNeonUThreeReg<NVaddD, NVaddQ>( 544 q, size, machInst, vd, vn, vm); 545 } 546 } 547 case 0x9: 548 if (b) { 549 if (u) { 550 return decodeNeonUThreeReg<NVmulpD, NVmulpQ>( 551 q, size, machInst, vd, vn, vm); 552 } else { 553 return decodeNeonSThreeReg<NVmulD, NVmulQ>( 554 q, size, machInst, vd, vn, vm); 555 } 556 } else { 557 if (u) { 558 return decodeNeonUSThreeReg<NVmlsD, NVmlsQ>( 559 q, u, size, machInst, vd, vn, vm); 560 } else { 561 return decodeNeonUSThreeReg<NVmlaD, NVmlaQ>( 562 q, u, size, machInst, vd, vn, vm); 563 } 564 } 565 case 0xa: 566 if (q) 567 return new Unknown(machInst); 568 if (b) { 569 return decodeNeonUSThreeUSReg<VpminD>( 570 u, size, machInst, vd, vn, vm); 571 } else { 572 return decodeNeonUSThreeUSReg<VpmaxD>( 573 u, size, machInst, vd, vn, vm); 574 } 575 case 0xb: 576 if (b) { 577 if (u || q) { 578 return new Unknown(machInst); 579 } else { 580 return decodeNeonUThreeUSReg<NVpaddD>( 581 size, machInst, vd, vn, vm); 582 } 583 } else { 584 if (u) { 585 return decodeNeonSThreeSReg<VqrdmulhD, VqrdmulhQ>( 586 q, size, machInst, vd, vn, vm); 587 } else { 588 return decodeNeonSThreeSReg<VqdmulhD, VqdmulhQ>( 589 q, size, machInst, vd, vn, vm); 590 } 591 } 592 case 0xc: 593 if (b) { 594 if (!u) { 595 if (bits(c, 1) == 0) { 596 if (q) { 597 return new NVfmaQFp<float>(machInst, vd, vn, vm); 598 } else { 599 return new NVfmaDFp<float>(machInst, vd, vn, vm); 600 } 601 } else { 602 if (q) { 603 return new NVfmsQFp<float>(machInst, vd, vn, vm); 604 } else { 605 return new NVfmsDFp<float>(machInst, vd, vn, vm); 606 } 607 } 608 } 609 } else { 610 if (u) { 611 switch (c) { 612 case 0x0: 613 return new SHA256H(machInst, vd, vn, vm); 614 case 0x1: 615 return new SHA256H2(machInst, vd, vn, vm); 616 case 0x2: 617 return new SHA256SU1(machInst, vd, vn, vm); 618 case 0x3: 619 return new Unknown(machInst); 620 default: 621 M5_UNREACHABLE; 622 } 623 } else { 624 switch (c) { 625 case 0x0: 626 return new SHA1C(machInst, vd, vn, vm); 627 case 0x1: 628 return new SHA1P(machInst, vd, vn, vm); 629 case 0x2: 630 return new SHA1M(machInst, vd, vn, vm); 631 case 0x3: 632 return new SHA1SU0(machInst, vd, vn, vm); 633 default: 634 M5_UNREACHABLE; 635 } 636 } 637 } 638 return new Unknown(machInst); 639 case 0xd: 640 if (b) { 641 if (u) { 642 if (bits(c, 1) == 0) { 643 if (q) { 644 return new NVmulQFp<float>(machInst, vd, vn, vm); 645 } else { 646 return new NVmulDFp<float>(machInst, vd, vn, vm); 647 } 648 } else { 649 return new Unknown(machInst); 650 } 651 } else { 652 if (bits(c, 1) == 0) { 653 if (q) { 654 return new NVmlaQFp<float>(machInst, vd, vn, vm); 655 } else { 656 return new NVmlaDFp<float>(machInst, vd, vn, vm); 657 } 658 } else { 659 if (q) { 660 return new NVmlsQFp<float>(machInst, vd, vn, vm); 661 } else { 662 return new NVmlsDFp<float>(machInst, vd, vn, vm); 663 } 664 } 665 } 666 } else { 667 if (u) { 668 if (bits(c, 1) == 0) { 669 if (q) { 670 return new VpaddQFp<float>(machInst, vd, vn, vm); 671 } else { 672 return new VpaddDFp<float>(machInst, vd, vn, vm); 673 } 674 } else { 675 if (q) { 676 return new VabdQFp<float>(machInst, vd, vn, vm); 677 } else { 678 return new VabdDFp<float>(machInst, vd, vn, vm); 679 } 680 } 681 } else { 682 if (bits(c, 1) == 0) { 683 if (q) { 684 return new VaddQFp<float>(machInst, vd, vn, vm); 685 } else { 686 return new VaddDFp<float>(machInst, vd, vn, vm); 687 } 688 } else { 689 if (q) { 690 return new VsubQFp<float>(machInst, vd, vn, vm); 691 } else { 692 return new VsubDFp<float>(machInst, vd, vn, vm); 693 } 694 } 695 } 696 } 697 case 0xe: 698 if (b) { 699 if (u) { 700 if (bits(c, 1) == 0) { 701 if (q) { 702 return new VacgeQFp<float>(machInst, vd, vn, vm); 703 } else { 704 return new VacgeDFp<float>(machInst, vd, vn, vm); 705 } 706 } else { 707 if (q) { 708 return new VacgtQFp<float>(machInst, vd, vn, vm); 709 } else { 710 return new VacgtDFp<float>(machInst, vd, vn, vm); 711 } 712 } 713 } else { 714 return new Unknown(machInst); 715 } 716 } else { 717 if (u) { 718 if (bits(c, 1) == 0) { 719 if (q) { 720 return new VcgeQFp<float>(machInst, vd, vn, vm); 721 } else { 722 return new VcgeDFp<float>(machInst, vd, vn, vm); 723 } 724 } else { 725 if (q) { 726 return new VcgtQFp<float>(machInst, vd, vn, vm); 727 } else { 728 return new VcgtDFp<float>(machInst, vd, vn, vm); 729 } 730 } 731 } else { 732 if (bits(c, 1) == 0) { 733 if (q) { 734 return new VceqQFp<float>(machInst, vd, vn, vm); 735 } else { 736 return new VceqDFp<float>(machInst, vd, vn, vm); 737 } 738 } else { 739 return new Unknown(machInst); 740 } 741 } 742 } 743 case 0xf: 744 if (b) { 745 if (u) { 746 return new Unknown(machInst); 747 } else { 748 if (bits(c, 1) == 0) { 749 if (q) { 750 return new VrecpsQFp<float>(machInst, vd, vn, vm); 751 } else { 752 return new VrecpsDFp<float>(machInst, vd, vn, vm); 753 } 754 } else { 755 if (q) { 756 return new VrsqrtsQFp<float>(machInst, vd, vn, vm); 757 } else { 758 return new VrsqrtsDFp<float>(machInst, vd, vn, vm); 759 } 760 } 761 } 762 } else { 763 if (u) { 764 if (bits(c, 1) == 0) { 765 if (q) { 766 return new VpmaxQFp<float>(machInst, vd, vn, vm); 767 } else { 768 return new VpmaxDFp<float>(machInst, vd, vn, vm); 769 } 770 } else { 771 if (q) { 772 return new VpminQFp<float>(machInst, vd, vn, vm); 773 } else { 774 return new VpminDFp<float>(machInst, vd, vn, vm); 775 } 776 } 777 } else { 778 if (bits(c, 1) == 0) { 779 if (q) { 780 return new VmaxQFp<float>(machInst, vd, vn, vm); 781 } else { 782 return new VmaxDFp<float>(machInst, vd, vn, vm); 783 } 784 } else { 785 if (q) { 786 return new VminQFp<float>(machInst, vd, vn, vm); 787 } else { 788 return new VminDFp<float>(machInst, vd, vn, vm); 789 } 790 } 791 } 792 } 793 } 794 return new Unknown(machInst); 795 } 796 797 static StaticInstPtr 798 decodeNeonOneRegModImm(ExtMachInst machInst) 799 { 800 const IntRegIndex vd = 801 (IntRegIndex)(2 * (bits(machInst, 15, 12) | 802 (bits(machInst, 22) << 4))); 803 const bool q = bits(machInst, 6); 804 const bool op = bits(machInst, 5); 805 const uint8_t cmode = bits(machInst, 11, 8); 806 const uint8_t imm = ((THUMB ? bits(machInst, 28) : 807 bits(machInst, 24)) << 7) | 808 (bits(machInst, 18, 16) << 4) | 809 (bits(machInst, 3, 0) << 0); 810 811 // Check for invalid immediate encodings and return an unknown op 812 // if it happens 813 bool immValid = true; 814 const uint64_t bigImm = simd_modified_imm(op, cmode, imm, immValid); 815 if (!immValid) { 816 return new Unknown(machInst); 817 } 818 819 if (op) { 820 if (bits(cmode, 3) == 0) { 821 if (bits(cmode, 0) == 0) { 822 if (q) 823 return new NVmvniQ<uint64_t>(machInst, vd, bigImm); 824 else 825 return new NVmvniD<uint64_t>(machInst, vd, bigImm); 826 } else { 827 if (q) 828 return new NVbiciQ<uint64_t>(machInst, vd, bigImm); 829 else 830 return new NVbiciD<uint64_t>(machInst, vd, bigImm); 831 } 832 } else { 833 if (bits(cmode, 2) == 1) { 834 switch (bits(cmode, 1, 0)) { 835 case 0: 836 case 1: 837 if (q) 838 return new NVmvniQ<uint64_t>(machInst, vd, bigImm); 839 else 840 return new NVmvniD<uint64_t>(machInst, vd, bigImm); 841 case 2: 842 if (q) 843 return new NVmoviQ<uint64_t>(machInst, vd, bigImm); 844 else 845 return new NVmoviD<uint64_t>(machInst, vd, bigImm); 846 case 3: 847 if (q) 848 return new Unknown(machInst); 849 else 850 return new Unknown(machInst); 851 } 852 } else { 853 if (bits(cmode, 0) == 0) { 854 if (q) 855 return new NVmvniQ<uint64_t>(machInst, vd, bigImm); 856 else 857 return new NVmvniD<uint64_t>(machInst, vd, bigImm); 858 } else { 859 if (q) 860 return new NVbiciQ<uint64_t>(machInst, vd, bigImm); 861 else 862 return new NVbiciD<uint64_t>(machInst, vd, bigImm); 863 } 864 } 865 } 866 } else { 867 if (bits(cmode, 3) == 0) { 868 if (bits(cmode, 0) == 0) { 869 if (q) 870 return new NVmoviQ<uint64_t>(machInst, vd, bigImm); 871 else 872 return new NVmoviD<uint64_t>(machInst, vd, bigImm); 873 } else { 874 if (q) 875 return new NVorriQ<uint64_t>(machInst, vd, bigImm); 876 else 877 return new NVorriD<uint64_t>(machInst, vd, bigImm); 878 } 879 } else { 880 if (bits(cmode, 2) == 1) { 881 if (q) 882 return new NVmoviQ<uint64_t>(machInst, vd, bigImm); 883 else 884 return new NVmoviD<uint64_t>(machInst, vd, bigImm); 885 } else { 886 if (bits(cmode, 0) == 0) { 887 if (q) 888 return new NVmoviQ<uint64_t>(machInst, vd, bigImm); 889 else 890 return new NVmoviD<uint64_t>(machInst, vd, bigImm); 891 } else { 892 if (q) 893 return new NVorriQ<uint64_t>(machInst, vd, bigImm); 894 else 895 return new NVorriD<uint64_t>(machInst, vd, bigImm); 896 } 897 } 898 } 899 } 900 return new Unknown(machInst); 901 } 902 903 static StaticInstPtr 904 decodeNeonTwoRegAndShift(ExtMachInst machInst) 905 { 906 const uint32_t a = bits(machInst, 11, 8); 907 const bool u = THUMB ? bits(machInst, 28) : bits(machInst, 24); 908 const bool b = bits(machInst, 6); 909 const bool l = bits(machInst, 7); 910 const IntRegIndex vd = 911 (IntRegIndex)(2 * (bits(machInst, 15, 12) | 912 (bits(machInst, 22) << 4))); 913 const IntRegIndex vm = 914 (IntRegIndex)(2 * (bits(machInst, 3, 0) | 915 (bits(machInst, 5) << 4))); 916 unsigned imm6 = bits(machInst, 21, 16); 917 unsigned imm = ((l ? 1 : 0) << 6) | imm6; 918 unsigned size = 3; 919 unsigned lShiftAmt = 0; 920 unsigned bitSel; 921 for (bitSel = 1 << 6; true; bitSel >>= 1) { 922 if (bitSel & imm) 923 break; 924 else if (!size) 925 return new Unknown(machInst); 926 size--; 927 } 928 lShiftAmt = imm6 & ~bitSel; 929 unsigned rShiftAmt = 0; 930 if (a != 0xe && a != 0xf) { 931 if (size > 2) 932 rShiftAmt = 64 - imm6; 933 else 934 rShiftAmt = 2 * (8 << size) - imm6; 935 } 936 937 switch (a) { 938 case 0x0: 939 return decodeNeonUSTwoShiftReg<NVshrD, NVshrQ>( 940 b, u, size, machInst, vd, vm, rShiftAmt); 941 case 0x1: 942 return decodeNeonUSTwoShiftReg<NVsraD, NVsraQ>( 943 b, u, size, machInst, vd, vm, rShiftAmt); 944 case 0x2: 945 return decodeNeonUSTwoShiftReg<NVrshrD, NVrshrQ>( 946 b, u, size, machInst, vd, vm, rShiftAmt); 947 case 0x3: 948 return decodeNeonUSTwoShiftReg<NVrsraD, NVrsraQ>( 949 b, u, size, machInst, vd, vm, rShiftAmt); 950 case 0x4: 951 if (u) { 952 return decodeNeonUTwoShiftReg<NVsriD, NVsriQ>( 953 b, size, machInst, vd, vm, rShiftAmt); 954 } else { 955 return new Unknown(machInst); 956 } 957 case 0x5: 958 if (u) { 959 return decodeNeonUTwoShiftReg<NVsliD, NVsliQ>( 960 b, size, machInst, vd, vm, lShiftAmt); 961 } else { 962 return decodeNeonUTwoShiftReg<NVshlD, NVshlQ>( 963 b, size, machInst, vd, vm, lShiftAmt); 964 } 965 case 0x6: 966 case 0x7: 967 if (u) { 968 if (a == 0x6) { 969 return decodeNeonSTwoShiftReg<NVqshlusD, NVqshlusQ>( 970 b, size, machInst, vd, vm, lShiftAmt); 971 } else { 972 return decodeNeonUTwoShiftReg<NVqshluD, NVqshluQ>( 973 b, size, machInst, vd, vm, lShiftAmt); 974 } 975 } else { 976 return decodeNeonSTwoShiftReg<NVqshlD, NVqshlQ>( 977 b, size, machInst, vd, vm, lShiftAmt); 978 } 979 case 0x8: 980 if (l) { 981 return new Unknown(machInst); 982 } else if (u) { 983 return decodeNeonSTwoShiftSReg<NVqshruns, NVqrshruns>( 984 b, size, machInst, vd, vm, rShiftAmt); 985 } else { 986 return decodeNeonUTwoShiftSReg<NVshrn, NVrshrn>( 987 b, size, machInst, vd, vm, rShiftAmt); 988 } 989 case 0x9: 990 if (l) { 991 return new Unknown(machInst); 992 } else if (u) { 993 return decodeNeonUTwoShiftSReg<NVqshrun, NVqrshrun>( 994 b, size, machInst, vd, vm, rShiftAmt); 995 } else { 996 return decodeNeonSTwoShiftSReg<NVqshrn, NVqrshrn>( 997 b, size, machInst, vd, vm, rShiftAmt); 998 } 999 case 0xa: 1000 if (l || b) { 1001 return new Unknown(machInst); 1002 } else { 1003 return decodeNeonUSTwoShiftSReg<NVmovl, NVshll>( 1004 lShiftAmt, u, size, machInst, vd, vm, lShiftAmt); 1005 } 1006 case 0xe: 1007 if (l) { 1008 return new Unknown(machInst); 1009 } else { 1010 if (bits(imm6, 5) == 0) 1011 return new Unknown(machInst); 1012 if (u) { 1013 if (b) { 1014 return new NVcvtu2fpQ<float>( 1015 machInst, vd, vm, 64 - imm6); 1016 } else { 1017 return new NVcvtu2fpD<float>( 1018 machInst, vd, vm, 64 - imm6); 1019 } 1020 } else { 1021 if (b) { 1022 return new NVcvts2fpQ<float>( 1023 machInst, vd, vm, 64 - imm6); 1024 } else { 1025 return new NVcvts2fpD<float>( 1026 machInst, vd, vm, 64 - imm6); 1027 } 1028 } 1029 } 1030 case 0xf: 1031 if (l) { 1032 return new Unknown(machInst); 1033 } else { 1034 if (bits(imm6, 5) == 0) 1035 return new Unknown(machInst); 1036 if (u) { 1037 if (b) { 1038 return new NVcvt2ufxQ<float>( 1039 machInst, vd, vm, 64 - imm6); 1040 } else { 1041 return new NVcvt2ufxD<float>( 1042 machInst, vd, vm, 64 - imm6); 1043 } 1044 } else { 1045 if (b) { 1046 return new NVcvt2sfxQ<float>( 1047 machInst, vd, vm, 64 - imm6); 1048 } else { 1049 return new NVcvt2sfxD<float>( 1050 machInst, vd, vm, 64 - imm6); 1051 } 1052 } 1053 } 1054 } 1055 return new Unknown(machInst); 1056 } 1057 1058 static StaticInstPtr 1059 decodeNeonThreeRegDiffLengths(ExtMachInst machInst) 1060 { 1061 const bool u = THUMB ? bits(machInst, 28) : bits(machInst, 24); 1062 const uint32_t a = bits(machInst, 11, 8); 1063 const IntRegIndex vd = 1064 (IntRegIndex)(2 * (bits(machInst, 15, 12) | 1065 (bits(machInst, 22) << 4))); 1066 const IntRegIndex vn = 1067 (IntRegIndex)(2 * (bits(machInst, 19, 16) | 1068 (bits(machInst, 7) << 4))); 1069 const IntRegIndex vm = 1070 (IntRegIndex)(2 * (bits(machInst, 3, 0) | 1071 (bits(machInst, 5) << 4))); 1072 const unsigned size = bits(machInst, 21, 20); 1073 switch (a) { 1074 case 0x0: 1075 return decodeNeonUSThreeUSReg<Vaddl>( 1076 u, size, machInst, vd, vn, vm); 1077 case 0x1: 1078 return decodeNeonUSThreeUSReg<Vaddw>( 1079 u, size, machInst, vd, vn, vm); 1080 case 0x2: 1081 return decodeNeonUSThreeUSReg<Vsubl>( 1082 u, size, machInst, vd, vn, vm); 1083 case 0x3: 1084 return decodeNeonUSThreeUSReg<Vsubw>( 1085 u, size, machInst, vd, vn, vm); 1086 case 0x4: 1087 if (u) { 1088 return decodeNeonUThreeUSReg<Vraddhn>( 1089 size, machInst, vd, vn, vm); 1090 } else { 1091 return decodeNeonUThreeUSReg<Vaddhn>( 1092 size, machInst, vd, vn, vm); 1093 } 1094 case 0x5: 1095 return decodeNeonUSThreeUSReg<Vabal>( 1096 u, size, machInst, vd, vn, vm); 1097 case 0x6: 1098 if (u) { 1099 return decodeNeonUThreeUSReg<Vrsubhn>( 1100 size, machInst, vd, vn, vm); 1101 } else { 1102 return decodeNeonUThreeUSReg<Vsubhn>( 1103 size, machInst, vd, vn, vm); 1104 } 1105 case 0x7: 1106 if (bits(machInst, 23)) { 1107 return decodeNeonUSThreeUSReg<Vabdl>( 1108 u, size, machInst, vd, vn, vm); 1109 } else { 1110 return decodeNeonUSThreeReg<VabdD, VabdQ>( 1111 bits(machInst, 6), u, size, machInst, vd, vn, vm); 1112 } 1113 case 0x8: 1114 return decodeNeonUSThreeUSReg<Vmlal>( 1115 u, size, machInst, vd, vn, vm); 1116 case 0xa: 1117 return decodeNeonUSThreeUSReg<Vmlsl>( 1118 u, size, machInst, vd, vn, vm); 1119 case 0x9: 1120 if (u) { 1121 return new Unknown(machInst); 1122 } else { 1123 return decodeNeonSThreeUSReg<Vqdmlal>( 1124 size, machInst, vd, vn, vm); 1125 } 1126 case 0xb: 1127 if (u) { 1128 return new Unknown(machInst); 1129 } else { 1130 return decodeNeonSThreeUSReg<Vqdmlsl>( 1131 size, machInst, vd, vn, vm); 1132 } 1133 case 0xc: 1134 return decodeNeonUSThreeUSReg<Vmull>( 1135 u, size, machInst, vd, vn, vm); 1136 case 0xd: 1137 if (u) { 1138 return new Unknown(machInst); 1139 } else { 1140 return decodeNeonSThreeUSReg<Vqdmull>( 1141 size, machInst, vd, vn, vm); 1142 } 1143 case 0xe: 1144 return decodeNeonUThreeUSReg<Vmullp>( 1145 size, machInst, vd, vn, vm); 1146 } 1147 return new Unknown(machInst); 1148 } 1149 1150 static StaticInstPtr 1151 decodeNeonTwoRegScalar(ExtMachInst machInst) 1152 { 1153 const bool u = THUMB ? bits(machInst, 28) : bits(machInst, 24); 1154 const uint32_t a = bits(machInst, 11, 8); 1155 const unsigned size = bits(machInst, 21, 20); 1156 const IntRegIndex vd = 1157 (IntRegIndex)(2 * (bits(machInst, 15, 12) | 1158 (bits(machInst, 22) << 4))); 1159 const IntRegIndex vn = 1160 (IntRegIndex)(2 * (bits(machInst, 19, 16) | 1161 (bits(machInst, 7) << 4))); 1162 const IntRegIndex vm = (size == 2) ? 1163 (IntRegIndex)(2 * bits(machInst, 3, 0)) : 1164 (IntRegIndex)(2 * bits(machInst, 2, 0)); 1165 const unsigned index = (size == 2) ? (unsigned)bits(machInst, 5) : 1166 (bits(machInst, 3) | (bits(machInst, 5) << 1)); 1167 switch (a) { 1168 case 0x0: 1169 if (u) { 1170 switch (size) { 1171 case 1: 1172 return new VmlasQ<uint16_t>(machInst, vd, vn, vm, index); 1173 case 2: 1174 return new VmlasQ<uint32_t>(machInst, vd, vn, vm, index); 1175 default: 1176 return new Unknown(machInst); 1177 } 1178 } else { 1179 switch (size) { 1180 case 1: 1181 return new VmlasD<uint16_t>(machInst, vd, vn, vm, index); 1182 case 2: 1183 return new VmlasD<uint32_t>(machInst, vd, vn, vm, index); 1184 default: 1185 return new Unknown(machInst); 1186 } 1187 } 1188 case 0x1: 1189 if (u) 1190 return new VmlasQFp<float>(machInst, vd, vn, vm, index); 1191 else 1192 return new VmlasDFp<float>(machInst, vd, vn, vm, index); 1193 case 0x4: 1194 if (u) { 1195 switch (size) { 1196 case 1: 1197 return new VmlssQ<uint16_t>(machInst, vd, vn, vm, index); 1198 case 2: 1199 return new VmlssQ<uint32_t>(machInst, vd, vn, vm, index); 1200 default: 1201 return new Unknown(machInst); 1202 } 1203 } else { 1204 switch (size) { 1205 case 1: 1206 return new VmlssD<uint16_t>(machInst, vd, vn, vm, index); 1207 case 2: 1208 return new VmlssD<uint32_t>(machInst, vd, vn, vm, index); 1209 default: 1210 return new Unknown(machInst); 1211 } 1212 } 1213 case 0x5: 1214 if (u) 1215 return new VmlssQFp<float>(machInst, vd, vn, vm, index); 1216 else 1217 return new VmlssDFp<float>(machInst, vd, vn, vm, index); 1218 case 0x2: 1219 if (u) { 1220 switch (size) { 1221 case 1: 1222 return new Vmlals<uint16_t>(machInst, vd, vn, vm, index); 1223 case 2: 1224 return new Vmlals<uint32_t>(machInst, vd, vn, vm, index); 1225 default: 1226 return new Unknown(machInst); 1227 } 1228 } else { 1229 switch (size) { 1230 case 1: 1231 return new Vmlals<int16_t>(machInst, vd, vn, vm, index); 1232 case 2: 1233 return new Vmlals<int32_t>(machInst, vd, vn, vm, index); 1234 default: 1235 return new Unknown(machInst); 1236 } 1237 } 1238 case 0x6: 1239 if (u) { 1240 switch (size) { 1241 case 1: 1242 return new Vmlsls<uint16_t>(machInst, vd, vn, vm, index); 1243 case 2: 1244 return new Vmlsls<uint32_t>(machInst, vd, vn, vm, index); 1245 default: 1246 return new Unknown(machInst); 1247 } 1248 } else { 1249 switch (size) { 1250 case 1: 1251 return new Vmlsls<int16_t>(machInst, vd, vn, vm, index); 1252 case 2: 1253 return new Vmlsls<int32_t>(machInst, vd, vn, vm, index); 1254 default: 1255 return new Unknown(machInst); 1256 } 1257 } 1258 case 0x3: 1259 if (u) { 1260 return new Unknown(machInst); 1261 } else { 1262 switch (size) { 1263 case 1: 1264 return new Vqdmlals<int16_t>(machInst, vd, vn, vm, index); 1265 case 2: 1266 return new Vqdmlals<int32_t>(machInst, vd, vn, vm, index); 1267 default: 1268 return new Unknown(machInst); 1269 } 1270 } 1271 case 0x7: 1272 if (u) { 1273 return new Unknown(machInst); 1274 } else { 1275 switch (size) { 1276 case 1: 1277 return new Vqdmlsls<int16_t>(machInst, vd, vn, vm, index); 1278 case 2: 1279 return new Vqdmlsls<int32_t>(machInst, vd, vn, vm, index); 1280 default: 1281 return new Unknown(machInst); 1282 } 1283 } 1284 case 0x8: 1285 if (u) { 1286 switch (size) { 1287 case 1: 1288 return new VmulsQ<uint16_t>(machInst, vd, vn, vm, index); 1289 case 2: 1290 return new VmulsQ<uint32_t>(machInst, vd, vn, vm, index); 1291 default: 1292 return new Unknown(machInst); 1293 } 1294 } else { 1295 switch (size) { 1296 case 1: 1297 return new VmulsD<uint16_t>(machInst, vd, vn, vm, index); 1298 case 2: 1299 return new VmulsD<uint32_t>(machInst, vd, vn, vm, index); 1300 default: 1301 return new Unknown(machInst); 1302 } 1303 } 1304 case 0x9: 1305 if (u) 1306 return new VmulsQFp<float>(machInst, vd, vn, vm, index); 1307 else 1308 return new VmulsDFp<float>(machInst, vd, vn, vm, index); 1309 case 0xa: 1310 if (u) { 1311 switch (size) { 1312 case 1: 1313 return new Vmulls<uint16_t>(machInst, vd, vn, vm, index); 1314 case 2: 1315 return new Vmulls<uint32_t>(machInst, vd, vn, vm, index); 1316 default: 1317 return new Unknown(machInst); 1318 } 1319 } else { 1320 switch (size) { 1321 case 1: 1322 return new Vmulls<int16_t>(machInst, vd, vn, vm, index); 1323 case 2: 1324 return new Vmulls<int32_t>(machInst, vd, vn, vm, index); 1325 default: 1326 return new Unknown(machInst); 1327 } 1328 } 1329 case 0xb: 1330 if (u) { 1331 return new Unknown(machInst); 1332 } else { 1333 if (u) { 1334 switch (size) { 1335 case 1: 1336 return new Vqdmulls<uint16_t>( 1337 machInst, vd, vn, vm, index); 1338 case 2: 1339 return new Vqdmulls<uint32_t>( 1340 machInst, vd, vn, vm, index); 1341 default: 1342 return new Unknown(machInst); 1343 } 1344 } else { 1345 switch (size) { 1346 case 1: 1347 return new Vqdmulls<int16_t>( 1348 machInst, vd, vn, vm, index); 1349 case 2: 1350 return new Vqdmulls<int32_t>( 1351 machInst, vd, vn, vm, index); 1352 default: 1353 return new Unknown(machInst); 1354 } 1355 } 1356 } 1357 case 0xc: 1358 if (u) { 1359 switch (size) { 1360 case 1: 1361 return new VqdmulhsQ<int16_t>( 1362 machInst, vd, vn, vm, index); 1363 case 2: 1364 return new VqdmulhsQ<int32_t>( 1365 machInst, vd, vn, vm, index); 1366 default: 1367 return new Unknown(machInst); 1368 } 1369 } else { 1370 switch (size) { 1371 case 1: 1372 return new VqdmulhsD<int16_t>( 1373 machInst, vd, vn, vm, index); 1374 case 2: 1375 return new VqdmulhsD<int32_t>( 1376 machInst, vd, vn, vm, index); 1377 default: 1378 return new Unknown(machInst); 1379 } 1380 } 1381 case 0xd: 1382 if (u) { 1383 switch (size) { 1384 case 1: 1385 return new VqrdmulhsQ<int16_t>( 1386 machInst, vd, vn, vm, index); 1387 case 2: 1388 return new VqrdmulhsQ<int32_t>( 1389 machInst, vd, vn, vm, index); 1390 default: 1391 return new Unknown(machInst); 1392 } 1393 } else { 1394 switch (size) { 1395 case 1: 1396 return new VqrdmulhsD<int16_t>( 1397 machInst, vd, vn, vm, index); 1398 case 2: 1399 return new VqrdmulhsD<int32_t>( 1400 machInst, vd, vn, vm, index); 1401 default: 1402 return new Unknown(machInst); 1403 } 1404 } 1405 } 1406 return new Unknown(machInst); 1407 } 1408 1409 static StaticInstPtr 1410 decodeNeonTwoRegMisc(ExtMachInst machInst) 1411 { 1412 const uint32_t a = bits(machInst, 17, 16); 1413 const uint32_t b = bits(machInst, 10, 6); 1414 const bool q = bits(machInst, 6); 1415 const IntRegIndex vd = 1416 (IntRegIndex)(2 * (bits(machInst, 15, 12) | 1417 (bits(machInst, 22) << 4))); 1418 const IntRegIndex vm = 1419 (IntRegIndex)(2 * (bits(machInst, 3, 0) | 1420 (bits(machInst, 5) << 4))); 1421 const unsigned size = bits(machInst, 19, 18); 1422 switch (a) { 1423 case 0x0: 1424 switch (bits(b, 4, 1)) { 1425 case 0x0: 1426 switch (size) { 1427 case 0: 1428 if (q) { 1429 return new NVrev64Q<uint8_t>(machInst, vd, vm); 1430 } else { 1431 return new NVrev64D<uint8_t>(machInst, vd, vm); 1432 } 1433 case 1: 1434 if (q) { 1435 return new NVrev64Q<uint16_t>(machInst, vd, vm); 1436 } else { 1437 return new NVrev64D<uint16_t>(machInst, vd, vm); 1438 } 1439 case 2: 1440 if (q) { 1441 return new NVrev64Q<uint32_t>(machInst, vd, vm); 1442 } else { 1443 return new NVrev64D<uint32_t>(machInst, vd, vm); 1444 } 1445 default: 1446 return new Unknown(machInst); 1447 } 1448 case 0x1: 1449 switch (size) { 1450 case 0: 1451 if (q) { 1452 return new NVrev32Q<uint8_t>(machInst, vd, vm); 1453 } else { 1454 return new NVrev32D<uint8_t>(machInst, vd, vm); 1455 } 1456 case 1: 1457 if (q) { 1458 return new NVrev32Q<uint16_t>(machInst, vd, vm); 1459 } else { 1460 return new NVrev32D<uint16_t>(machInst, vd, vm); 1461 } 1462 default: 1463 return new Unknown(machInst); 1464 } 1465 case 0x2: 1466 if (size != 0) { 1467 return new Unknown(machInst); 1468 } else if (q) { 1469 return new NVrev16Q<uint8_t>(machInst, vd, vm); 1470 } else { 1471 return new NVrev16D<uint8_t>(machInst, vd, vm); 1472 } 1473 case 0x4: 1474 return decodeNeonSTwoMiscSReg<NVpaddlD, NVpaddlQ>( 1475 q, size, machInst, vd, vm); 1476 case 0x5: 1477 return decodeNeonUTwoMiscSReg<NVpaddlD, NVpaddlQ>( 1478 q, size, machInst, vd, vm); 1479 case 0x6: 1480 if (q == 0) { 1481 return new AESE(machInst, vd, vd, vm); 1482 } else { 1483 return new AESD(machInst, vd, vd, vm); 1484 } 1485 case 0x7: 1486 if (q == 0) { 1487 return new AESMC(machInst, vd, vm); 1488 } else { 1489 return new AESIMC(machInst, vd, vm); 1490 } 1491 case 0x8: 1492 return decodeNeonSTwoMiscReg<NVclsD, NVclsQ>( 1493 q, size, machInst, vd, vm); 1494 case 0x9: 1495 return decodeNeonSTwoMiscReg<NVclzD, NVclzQ>( 1496 q, size, machInst, vd, vm); 1497 case 0xa: 1498 return decodeNeonUTwoMiscReg<NVcntD, NVcntQ>( 1499 q, size, machInst, vd, vm); 1500 case 0xb: 1501 if (q) 1502 return new NVmvnQ<uint64_t>(machInst, vd, vm); 1503 else 1504 return new NVmvnD<uint64_t>(machInst, vd, vm); 1505 case 0xc: 1506 return decodeNeonSTwoMiscSReg<NVpadalD, NVpadalQ>( 1507 q, size, machInst, vd, vm); 1508 case 0xd: 1509 return decodeNeonUTwoMiscSReg<NVpadalD, NVpadalQ>( 1510 q, size, machInst, vd, vm); 1511 case 0xe: 1512 return decodeNeonSTwoMiscReg<NVqabsD, NVqabsQ>( 1513 q, size, machInst, vd, vm); 1514 case 0xf: 1515 return decodeNeonSTwoMiscReg<NVqnegD, NVqnegQ>( 1516 q, size, machInst, vd, vm); 1517 default: 1518 return new Unknown(machInst); 1519 } 1520 case 0x1: 1521 switch (bits(b, 3, 1)) { 1522 case 0x0: 1523 if (bits(b, 4)) { 1524 if (q) { 1525 return new NVcgtQFp<float>(machInst, vd, vm); 1526 } else { 1527 return new NVcgtDFp<float>(machInst, vd, vm); 1528 } 1529 } else { 1530 return decodeNeonSTwoMiscReg<NVcgtD, NVcgtQ>( 1531 q, size, machInst, vd, vm); 1532 } 1533 case 0x1: 1534 if (bits(b, 4)) { 1535 if (q) { 1536 return new NVcgeQFp<float>(machInst, vd, vm); 1537 } else { 1538 return new NVcgeDFp<float>(machInst, vd, vm); 1539 } 1540 } else { 1541 return decodeNeonSTwoMiscReg<NVcgeD, NVcgeQ>( 1542 q, size, machInst, vd, vm); 1543 } 1544 case 0x2: 1545 if (bits(b, 4)) { 1546 if (q) { 1547 return new NVceqQFp<float>(machInst, vd, vm); 1548 } else { 1549 return new NVceqDFp<float>(machInst, vd, vm); 1550 } 1551 } else { 1552 return decodeNeonSTwoMiscReg<NVceqD, NVceqQ>( 1553 q, size, machInst, vd, vm); 1554 } 1555 case 0x3: 1556 if (bits(b, 4)) { 1557 if (q) { 1558 return new NVcleQFp<float>(machInst, vd, vm); 1559 } else { 1560 return new NVcleDFp<float>(machInst, vd, vm); 1561 } 1562 } else { 1563 return decodeNeonSTwoMiscReg<NVcleD, NVcleQ>( 1564 q, size, machInst, vd, vm); 1565 } 1566 case 0x4: 1567 if (bits(b, 4)) { 1568 if (q) { 1569 return new NVcltQFp<float>(machInst, vd, vm); 1570 } else { 1571 return new NVcltDFp<float>(machInst, vd, vm); 1572 } 1573 } else { 1574 return decodeNeonSTwoMiscReg<NVcltD, NVcltQ>( 1575 q, size, machInst, vd, vm); 1576 } 1577 case 0x5: 1578 if (q) { 1579 return new SHA1H(machInst, vd, vm); 1580 } else { 1581 return new Unknown(machInst); 1582 } 1583 case 0x6: 1584 if (bits(machInst, 10)) { 1585 if (q) 1586 return new NVabsQFp<float>(machInst, vd, vm); 1587 else 1588 return new NVabsDFp<float>(machInst, vd, vm); 1589 } else { 1590 return decodeNeonSTwoMiscReg<NVabsD, NVabsQ>( 1591 q, size, machInst, vd, vm); 1592 } 1593 case 0x7: 1594 if (bits(machInst, 10)) { 1595 if (q) 1596 return new NVnegQFp<float>(machInst, vd, vm); 1597 else 1598 return new NVnegDFp<float>(machInst, vd, vm); 1599 } else { 1600 return decodeNeonSTwoMiscReg<NVnegD, NVnegQ>( 1601 q, size, machInst, vd, vm); 1602 } 1603 default: 1604 return new Unknown64(machInst); 1605 } 1606 case 0x2: 1607 switch (bits(b, 4, 1)) { 1608 case 0x0: 1609 if (q) 1610 return new NVswpQ<uint64_t>(machInst, vd, vm); 1611 else 1612 return new NVswpD<uint64_t>(machInst, vd, vm); 1613 case 0x1: 1614 return decodeNeonUTwoMiscSReg<NVtrnD, NVtrnQ>( 1615 q, size, machInst, vd, vm); 1616 case 0x2: 1617 return decodeNeonUTwoMiscReg<NVuzpD, NVuzpQ>( 1618 q, size, machInst, vd, vm); 1619 case 0x3: 1620 return decodeNeonUTwoMiscReg<NVzipD, NVzipQ>( 1621 q, size, machInst, vd, vm); 1622 case 0x4: 1623 if (b == 0x8) { 1624 return decodeNeonUTwoMiscUSReg<NVmovn>( 1625 size, machInst, vd, vm); 1626 } else { 1627 return decodeNeonSTwoMiscUSReg<NVqmovuns>( 1628 size, machInst, vd, vm); 1629 } 1630 case 0x5: 1631 if (q) { 1632 return decodeNeonUTwoMiscUSReg<NVqmovun>( 1633 size, machInst, vd, vm); 1634 } else { 1635 return decodeNeonSTwoMiscUSReg<NVqmovn>( 1636 size, machInst, vd, vm); 1637 } 1638 case 0x6: 1639 if (b == 0xc) { 1640 return decodeNeonSTwoShiftUSReg<NVshll>( 1641 size, machInst, vd, vm, 8 << size); 1642 } else { 1643 return new Unknown(machInst); 1644 } 1645 case 0x7: 1646 if (q) { 1647 return new SHA256SU0(machInst, vd, vm); 1648 } else { 1649 return new SHA1SU1(machInst, vd, vm); 1650 } 1651 case 0xc: 1652 case 0xe: 1653 if (b == 0x18) { 1654 if (size != 1 || (vm % 2)) 1655 return new Unknown(machInst); 1656 return new NVcvts2h<uint16_t>(machInst, vd, vm); 1657 } else if (b == 0x1c) { 1658 if (size != 1 || (vd % 2)) 1659 return new Unknown(machInst); 1660 return new NVcvth2s<uint16_t>(machInst, vd, vm); 1661 } else { 1662 return new Unknown(machInst); 1663 } 1664 default: 1665 return new Unknown(machInst); 1666 } 1667 case 0x3: 1668 if (bits(b, 4, 3) == 0x3) { 1669 if ((q && (vd % 2 || vm % 2)) || size != 2) { 1670 return new Unknown(machInst); 1671 } else { 1672 if (bits(b, 2)) { 1673 if (bits(b, 1)) { 1674 if (q) { 1675 return new NVcvt2ufxQ<float>( 1676 machInst, vd, vm, 0); 1677 } else { 1678 return new NVcvt2ufxD<float>( 1679 machInst, vd, vm, 0); 1680 } 1681 } else { 1682 if (q) { 1683 return new NVcvt2sfxQ<float>( 1684 machInst, vd, vm, 0); 1685 } else { 1686 return new NVcvt2sfxD<float>( 1687 machInst, vd, vm, 0); 1688 } 1689 } 1690 } else { 1691 if (bits(b, 1)) { 1692 if (q) { 1693 return new NVcvtu2fpQ<float>( 1694 machInst, vd, vm, 0); 1695 } else { 1696 return new NVcvtu2fpD<float>( 1697 machInst, vd, vm, 0); 1698 } 1699 } else { 1700 if (q) { 1701 return new NVcvts2fpQ<float>( 1702 machInst, vd, vm, 0); 1703 } else { 1704 return new NVcvts2fpD<float>( 1705 machInst, vd, vm, 0); 1706 } 1707 } 1708 } 1709 } 1710 } else if ((b & 0x1a) == 0x10) { 1711 if (bits(b, 2)) { 1712 if (q) { 1713 return new NVrecpeQFp<float>(machInst, vd, vm); 1714 } else { 1715 return new NVrecpeDFp<float>(machInst, vd, vm); 1716 } 1717 } else { 1718 if (q) { 1719 return new NVrecpeQ<uint32_t>(machInst, vd, vm); 1720 } else { 1721 return new NVrecpeD<uint32_t>(machInst, vd, vm); 1722 } 1723 } 1724 } else if ((b & 0x1a) == 0x12) { 1725 if (bits(b, 2)) { 1726 if (q) { 1727 return new NVrsqrteQFp<float>(machInst, vd, vm); 1728 } else { 1729 return new NVrsqrteDFp<float>(machInst, vd, vm); 1730 } 1731 } else { 1732 if (q) { 1733 return new NVrsqrteQ<uint32_t>(machInst, vd, vm); 1734 } else { 1735 return new NVrsqrteD<uint32_t>(machInst, vd, vm); 1736 } 1737 } 1738 } else { 1739 return new Unknown(machInst); 1740 } 1741 } 1742 return new Unknown(machInst); 1743 } 1744 1745 StaticInstPtr 1746 decodeNeonData(ExtMachInst machInst) 1747 { 1748 const bool u = THUMB ? bits(machInst, 28) : bits(machInst, 24); 1749 const uint32_t a = bits(machInst, 23, 19); 1750 const uint32_t b = bits(machInst, 11, 8); 1751 const uint32_t c = bits(machInst, 7, 4); 1752 if (bits(a, 4) == 0) { 1753 return decodeNeonThreeRegistersSameLength(machInst); 1754 } else if ((c & 0x9) == 1) { 1755 if ((a & 0x7) == 0) { 1756 return decodeNeonOneRegModImm(machInst); 1757 } else { 1758 return decodeNeonTwoRegAndShift(machInst); 1759 } 1760 } else if ((c & 0x9) == 9) { 1761 return decodeNeonTwoRegAndShift(machInst); 1762 } else if (bits(a, 2, 1) != 0x3) { 1763 if ((c & 0x5) == 0) { 1764 return decodeNeonThreeRegDiffLengths(machInst); 1765 } else if ((c & 0x5) == 4) { 1766 return decodeNeonTwoRegScalar(machInst); 1767 } 1768 } else if ((a & 0x16) == 0x16) { 1769 const IntRegIndex vd = 1770 (IntRegIndex)(2 * (bits(machInst, 15, 12) | 1771 (bits(machInst, 22) << 4))); 1772 const IntRegIndex vn = 1773 (IntRegIndex)(2 * (bits(machInst, 19, 16) | 1774 (bits(machInst, 7) << 4))); 1775 const IntRegIndex vm = 1776 (IntRegIndex)(2 * (bits(machInst, 3, 0) | 1777 (bits(machInst, 5) << 4))); 1778 if (!u) { 1779 if (bits(c, 0) == 0) { 1780 unsigned imm4 = bits(machInst, 11, 8); 1781 bool q = bits(machInst, 6); 1782 if (imm4 >= 16 && !q) 1783 return new Unknown(machInst); 1784 if (q) { 1785 return new NVextQ<uint8_t>(machInst, vd, vn, vm, imm4); 1786 } else { 1787 return new NVextD<uint8_t>(machInst, vd, vn, vm, imm4); 1788 } 1789 } 1790 } else if (bits(b, 3) == 0 && bits(c, 0) == 0) { 1791 return decodeNeonTwoRegMisc(machInst); 1792 } else if (bits(b, 3, 2) == 0x2 && bits(c, 0) == 0) { 1793 unsigned length = bits(machInst, 9, 8) + 1; 1794 if ((uint32_t)vn / 2 + length > 32) 1795 return new Unknown(machInst); 1796 if (bits(machInst, 6) == 0) { 1797 switch (length) { 1798 case 1: 1799 return new NVtbl1(machInst, vd, vn, vm); 1800 case 2: 1801 return new NVtbl2(machInst, vd, vn, vm); 1802 case 3: 1803 return new NVtbl3(machInst, vd, vn, vm); 1804 case 4: 1805 return new NVtbl4(machInst, vd, vn, vm); 1806 } 1807 } else { 1808 switch (length) { 1809 case 1: 1810 return new NVtbx1(machInst, vd, vn, vm); 1811 case 2: 1812 return new NVtbx2(machInst, vd, vn, vm); 1813 case 3: 1814 return new NVtbx3(machInst, vd, vn, vm); 1815 case 4: 1816 return new NVtbx4(machInst, vd, vn, vm); 1817 } 1818 } 1819 } else if (b == 0xc && (c & 0x9) == 0) { 1820 unsigned imm4 = bits(machInst, 19, 16); 1821 if (bits(imm4, 2, 0) == 0) 1822 return new Unknown(machInst); 1823 unsigned size = 0; 1824 while ((imm4 & 0x1) == 0) { 1825 size++; 1826 imm4 >>= 1; 1827 } 1828 unsigned index = imm4 >> 1; 1829 const bool q = bits(machInst, 6); 1830 return decodeNeonUTwoShiftSReg<NVdupD, NVdupQ>( 1831 q, size, machInst, vd, vm, index); 1832 } 1833 } 1834 return new Unknown(machInst); 1835 } 1836 ''' 1837}}; 1838 1839def format ThumbNeonMem() {{ 1840 decode_block = ''' 1841 return decodeNeonMem(machInst); 1842 ''' 1843}}; 1844 1845def format ThumbNeonData() {{ 1846 decode_block = ''' 1847 return decodeNeonData(machInst); 1848 ''' 1849}}; 1850 1851let {{ 1852 header_output = ''' 1853 StaticInstPtr 1854 decodeExtensionRegLoadStore(ExtMachInst machInst); 1855 ''' 1856 decoder_output = ''' 1857 StaticInstPtr 1858 decodeExtensionRegLoadStore(ExtMachInst machInst) 1859 { 1860 const uint32_t opcode = bits(machInst, 24, 20); 1861 const uint32_t offset = bits(machInst, 7, 0); 1862 const bool single = (bits(machInst, 8) == 0); 1863 const IntRegIndex rn = (IntRegIndex)(uint32_t)bits(machInst, 19, 16); 1864 RegIndex vd; 1865 if (single) { 1866 vd = (RegIndex)(uint32_t)((bits(machInst, 15, 12) << 1) | 1867 bits(machInst, 22)); 1868 } else { 1869 vd = (RegIndex)(uint32_t)((bits(machInst, 15, 12) << 1) | 1870 (bits(machInst, 22) << 5)); 1871 } 1872 switch (bits(opcode, 4, 3)) { 1873 case 0x0: 1874 if (bits(opcode, 4, 1) == 0x2 && 1875 !(machInst.thumb == 1 && bits(machInst, 28) == 1) && 1876 !(machInst.thumb == 0 && machInst.condCode == 0xf)) { 1877 if ((bits(machInst, 7, 4) & 0xd) != 1) { 1878 break; 1879 } 1880 const IntRegIndex rt = 1881 (IntRegIndex)(uint32_t)bits(machInst, 15, 12); 1882 const IntRegIndex rt2 = 1883 (IntRegIndex)(uint32_t)bits(machInst, 19, 16); 1884 const bool op = bits(machInst, 20); 1885 uint32_t vm; 1886 if (single) { 1887 vm = (bits(machInst, 3, 0) << 1) | bits(machInst, 5); 1888 } else { 1889 vm = (bits(machInst, 3, 0) << 1) | 1890 (bits(machInst, 5) << 5); 1891 } 1892 if (op) { 1893 return new Vmov2Core2Reg(machInst, rt, rt2, 1894 (IntRegIndex)vm); 1895 } else { 1896 return new Vmov2Reg2Core(machInst, (IntRegIndex)vm, 1897 rt, rt2); 1898 } 1899 } 1900 break; 1901 case 0x1: 1902 { 1903 if (offset == 0 || vd + offset/2 > NumFloatV7ArchRegs) { 1904 break; 1905 } 1906 switch (bits(opcode, 1, 0)) { 1907 case 0x0: 1908 return new VLdmStm(machInst, rn, vd, single, 1909 true, false, false, offset); 1910 case 0x1: 1911 return new VLdmStm(machInst, rn, vd, single, 1912 true, false, true, offset); 1913 case 0x2: 1914 return new VLdmStm(machInst, rn, vd, single, 1915 true, true, false, offset); 1916 case 0x3: 1917 // If rn == sp, then this is called vpop. 1918 return new VLdmStm(machInst, rn, vd, single, 1919 true, true, true, offset); 1920 default: 1921 M5_UNREACHABLE; 1922 } 1923 } 1924 case 0x2: 1925 if (bits(opcode, 1, 0) == 0x2) { 1926 // If rn == sp, then this is called vpush. 1927 return new VLdmStm(machInst, rn, vd, single, 1928 false, true, false, offset); 1929 } else if (bits(opcode, 1, 0) == 0x3) { 1930 return new VLdmStm(machInst, rn, vd, single, 1931 false, true, true, offset); 1932 } 1933 M5_FALLTHROUGH; 1934 case 0x3: 1935 const bool up = (bits(machInst, 23) == 1); 1936 const uint32_t imm = bits(machInst, 7, 0) << 2; 1937 if (single) { 1938 vd = (RegIndex)(uint32_t)((bits(machInst, 15, 12) << 1) | 1939 (bits(machInst, 22))); 1940 } else { 1941 vd = (RegIndex)(uint32_t)((bits(machInst, 15, 12) << 1) | 1942 (bits(machInst, 22) << 5)); 1943 } 1944 if (bits(opcode, 1, 0) == 0x0) { 1945 if (single) { 1946 if (up) { 1947 return new %(vstr_us)s(machInst, vd, rn, up, imm); 1948 } else { 1949 return new %(vstr_s)s(machInst, vd, rn, up, imm); 1950 } 1951 } else { 1952 if (up) { 1953 return new %(vstr_ud)s(machInst, vd, vd + 1, 1954 rn, up, imm); 1955 } else { 1956 return new %(vstr_d)s(machInst, vd, vd + 1, 1957 rn, up, imm); 1958 } 1959 } 1960 } else if (bits(opcode, 1, 0) == 0x1) { 1961 if (single) { 1962 if (up) { 1963 return new %(vldr_us)s(machInst, vd, rn, up, imm); 1964 } else { 1965 return new %(vldr_s)s(machInst, vd, rn, up, imm); 1966 } 1967 } else { 1968 if (up) { 1969 return new %(vldr_ud)s(machInst, vd, vd + 1, 1970 rn, up, imm); 1971 } else { 1972 return new %(vldr_d)s(machInst, vd, vd + 1, 1973 rn, up, imm); 1974 } 1975 } 1976 } 1977 } 1978 return new Unknown(machInst); 1979 } 1980 ''' % { 1981 "vldr_us" : "VLDR_" + loadImmClassName(False, True, False), 1982 "vldr_s" : "VLDR_" + loadImmClassName(False, False, False), 1983 "vldr_ud" : "VLDR_" + loadDoubleImmClassName(False, True, False), 1984 "vldr_d" : "VLDR_" + loadDoubleImmClassName(False, False, False), 1985 "vstr_us" : "VSTR_" + storeImmClassName(False, True, False), 1986 "vstr_s" : "VSTR_" + storeImmClassName(False, False, False), 1987 "vstr_ud" : "VSTR_" + storeDoubleImmClassName(False, True, False), 1988 "vstr_d" : "VSTR_" + storeDoubleImmClassName(False, False, False) 1989 } 1990}}; 1991 1992def format ExtensionRegLoadStore() {{ 1993 decode_block = ''' 1994 return decodeExtensionRegLoadStore(machInst); 1995 ''' 1996}}; 1997 1998let {{ 1999 header_output = ''' 2000 StaticInstPtr 2001 decodeShortFpTransfer(ExtMachInst machInst); 2002 ''' 2003 decoder_output = ''' 2004 StaticInstPtr 2005 decodeShortFpTransfer(ExtMachInst machInst) 2006 { 2007 const uint32_t l = bits(machInst, 20); 2008 const uint32_t c = bits(machInst, 8); 2009 const uint32_t a = bits(machInst, 23, 21); 2010 const uint32_t b = bits(machInst, 6, 5); 2011 if ((machInst.thumb == 1 && bits(machInst, 28) == 1) || 2012 (machInst.thumb == 0 && machInst.condCode == 0xf)) { 2013 // Determine if this is backported aarch64 FP instruction 2014 const bool b31_b24 = bits(machInst, 31, 24) == 0xFE; 2015 const bool b23 = bits(machInst, 23); 2016 const bool b21_b18 = bits(machInst, 21, 18) == 0xE; 2017 const bool b11_b9 = bits(machInst, 11, 9) == 0x5; 2018 const bool sz = bits(machInst, 8); 2019 const bool b7_b6 = bits(machInst, 7, 6) == 0x1; 2020 const bool b6 = bits(machInst, 6) == 0x0; 2021 const bool b4 = bits(machInst, 4) == 0x0; 2022 if (b31_b24 && b23 && b21_b18 && b11_b9 && b7_b6 && b4) { 2023 // VINT* Integer Rounding Instructon 2024 const uint32_t rm = bits(machInst, 17, 16); 2025 2026 if (sz) { 2027 const IntRegIndex vd = 2028 (IntRegIndex)((bits(machInst, 22) << 5) | 2029 (bits(machInst, 15, 12) << 1)); 2030 const IntRegIndex vm = 2031 (IntRegIndex)((bits(machInst, 5) << 5) | 2032 (bits(machInst, 3, 0) << 1)); 2033 switch(rm) { 2034 case 0x0: 2035 return decodeVfpRegRegOp<VRIntAD>(machInst, vd, vm, 2036 true); 2037 case 0x1: 2038 return decodeVfpRegRegOp<VRIntND>(machInst, vd, vm, 2039 true); 2040 case 0x2: 2041 return decodeVfpRegRegOp<VRIntPD>(machInst, vd, vm, 2042 true); 2043 case 0x3: 2044 return decodeVfpRegRegOp<VRIntMD>(machInst, vd, vm, 2045 true); 2046 default: return new Unknown(machInst); 2047 } 2048 } else { 2049 const IntRegIndex vd = 2050 (IntRegIndex)(bits(machInst, 22) | 2051 (bits(machInst, 15, 12) << 1)); 2052 const IntRegIndex vm = 2053 (IntRegIndex)(bits(machInst, 5) | 2054 (bits(machInst, 3, 0) << 1)); 2055 switch(rm) { 2056 case 0x0: 2057 return decodeVfpRegRegOp<VRIntAS>(machInst, vd, vm, 2058 false); 2059 case 0x1: 2060 return decodeVfpRegRegOp<VRIntNS>(machInst, vd, vm, 2061 false); 2062 case 0x2: 2063 return decodeVfpRegRegOp<VRIntPS>(machInst, vd, vm, 2064 false); 2065 case 0x3: 2066 return decodeVfpRegRegOp<VRIntMS>(machInst, vd, vm, 2067 false); 2068 default: return new Unknown(machInst); 2069 } 2070 } 2071 } else if (b31_b24 && !b23 && b11_b9 && b6 && b4){ 2072 // VSEL* floating point conditional select 2073 2074 ConditionCode cond; 2075 switch(bits(machInst, 21, 20)) { 2076 case 0x0: cond = COND_EQ; break; 2077 case 0x1: cond = COND_VS; break; 2078 case 0x2: cond = COND_GE; break; 2079 case 0x3: cond = COND_GT; break; 2080 } 2081 2082 if (sz) { 2083 const IntRegIndex vd = 2084 (IntRegIndex)((bits(machInst, 22) << 5) | 2085 (bits(machInst, 15, 12) << 1)); 2086 const IntRegIndex vm = 2087 (IntRegIndex)((bits(machInst, 5) << 5) | 2088 (bits(machInst, 3, 0) << 1)); 2089 const IntRegIndex vn = 2090 (IntRegIndex)((bits(machInst, 7) << 5) | 2091 (bits(machInst, 19, 16) << 1)); 2092 return new VselD(machInst, vd, vn, vm, cond); 2093 } else { 2094 const IntRegIndex vd = 2095 (IntRegIndex)(bits(machInst, 22) | 2096 (bits(machInst, 15, 12) << 1)); 2097 const IntRegIndex vm = 2098 (IntRegIndex)(bits(machInst, 5) | 2099 (bits(machInst, 3, 0) << 1)); 2100 const IntRegIndex vn = 2101 (IntRegIndex)((bits(machInst, 19, 16) << 1) | 2102 bits(machInst, 7)); 2103 return new VselS(machInst, vd, vn, vm, cond); 2104 } 2105 } else { 2106 return new Unknown(machInst); 2107 } 2108 } 2109 if (l == 0 && c == 0) { 2110 if (a == 0) { 2111 const uint32_t vn = (bits(machInst, 19, 16) << 1) | 2112 bits(machInst, 7); 2113 const IntRegIndex rt = 2114 (IntRegIndex)(uint32_t)bits(machInst, 15, 12); 2115 if (bits(machInst, 20) == 1) { 2116 return new VmovRegCoreW(machInst, rt, (IntRegIndex)vn); 2117 } else { 2118 return new VmovCoreRegW(machInst, (IntRegIndex)vn, rt); 2119 } 2120 } else if (a == 0x7) { 2121 const IntRegIndex rt = 2122 (IntRegIndex)(uint32_t)bits(machInst, 15, 12); 2123 uint32_t reg = bits(machInst, 19, 16); 2124 uint32_t specReg; 2125 switch (reg) { 2126 case 0: 2127 specReg = MISCREG_FPSID; 2128 break; 2129 case 1: 2130 specReg = MISCREG_FPSCR; 2131 break; 2132 case 6: 2133 specReg = MISCREG_MVFR1; 2134 break; 2135 case 7: 2136 specReg = MISCREG_MVFR0; 2137 break; 2138 case 8: 2139 specReg = MISCREG_FPEXC; 2140 break; 2141 default: 2142 return new Unknown(machInst); 2143 } 2144 if (specReg == MISCREG_FPSCR) { 2145 return new VmsrFpscr(machInst, (IntRegIndex)specReg, rt); 2146 } else { 2147 uint32_t iss = mcrMrcIssBuild(0, bits(machInst, 3, 0), rt, 2148 reg, a, bits(machInst, 7, 5)); 2149 return new Vmsr(machInst, (IntRegIndex)specReg, rt, iss); 2150 } 2151 } 2152 } else if (l == 0 && c == 1) { 2153 if (bits(a, 2) == 0) { 2154 uint32_t vd = (bits(machInst, 7) << 5) | 2155 (bits(machInst, 19, 16) << 1); 2156 // Handle accessing each single precision half of the vector. 2157 vd += bits(machInst, 21); 2158 const IntRegIndex rt = 2159 (IntRegIndex)(uint32_t)bits(machInst, 15, 12); 2160 if (bits(machInst, 22) == 1) { 2161 return new VmovCoreRegB(machInst, (IntRegIndex)vd, 2162 rt, bits(machInst, 6, 5)); 2163 } else if (bits(machInst, 5) == 1) { 2164 return new VmovCoreRegH(machInst, (IntRegIndex)vd, 2165 rt, bits(machInst, 6)); 2166 } else if (bits(machInst, 6) == 0) { 2167 return new VmovCoreRegW(machInst, (IntRegIndex)vd, rt); 2168 } else { 2169 return new Unknown(machInst); 2170 } 2171 } else if (bits(b, 1) == 0) { 2172 bool q = bits(machInst, 21); 2173 unsigned be = (bits(machInst, 22) << 1) | (bits(machInst, 5)); 2174 IntRegIndex vd = (IntRegIndex)(2 * (uint32_t) 2175 (bits(machInst, 19, 16) | (bits(machInst, 7) << 4))); 2176 IntRegIndex rt = (IntRegIndex)(uint32_t) 2177 bits(machInst, 15, 12); 2178 if (q) { 2179 switch (be) { 2180 case 0: 2181 return new NVdupQGpr<uint32_t>(machInst, vd, rt); 2182 case 1: 2183 return new NVdupQGpr<uint16_t>(machInst, vd, rt); 2184 case 2: 2185 return new NVdupQGpr<uint8_t>(machInst, vd, rt); 2186 case 3: 2187 return new Unknown(machInst); 2188 } 2189 } else { 2190 switch (be) { 2191 case 0: 2192 return new NVdupDGpr<uint32_t>(machInst, vd, rt); 2193 case 1: 2194 return new NVdupDGpr<uint16_t>(machInst, vd, rt); 2195 case 2: 2196 return new NVdupDGpr<uint8_t>(machInst, vd, rt); 2197 case 3: 2198 return new Unknown(machInst); 2199 } 2200 } 2201 } 2202 } else if (l == 1 && c == 0) { 2203 if (a == 0) { 2204 const uint32_t vn = (bits(machInst, 19, 16) << 1) | 2205 bits(machInst, 7); 2206 const IntRegIndex rt = 2207 (IntRegIndex)(uint32_t)bits(machInst, 15, 12); 2208 if (bits(machInst, 20) == 1) { 2209 return new VmovRegCoreW(machInst, rt, (IntRegIndex)vn); 2210 } else { 2211 return new VmovCoreRegW(machInst, (IntRegIndex)vn, rt); 2212 } 2213 } else if (a == 7) { 2214 const IntRegIndex rt = 2215 (IntRegIndex)(uint32_t)bits(machInst, 15, 12); 2216 uint32_t reg = bits(machInst, 19, 16); 2217 uint32_t specReg; 2218 switch (reg) { 2219 case 0: 2220 specReg = MISCREG_FPSID; 2221 break; 2222 case 1: 2223 specReg = MISCREG_FPSCR; 2224 break; 2225 case 6: 2226 specReg = MISCREG_MVFR1; 2227 break; 2228 case 7: 2229 specReg = MISCREG_MVFR0; 2230 break; 2231 case 8: 2232 specReg = MISCREG_FPEXC; 2233 break; 2234 default: 2235 return new Unknown(machInst); 2236 } 2237 if (rt == 0xf) { 2238 if (specReg == MISCREG_FPSCR) { 2239 return new VmrsApsrFpscr(machInst); 2240 } else { 2241 return new Unknown(machInst); 2242 } 2243 } else if (specReg == MISCREG_FPSCR) { 2244 return new VmrsFpscr(machInst, rt, (IntRegIndex)specReg); 2245 } else { 2246 uint32_t iss = mcrMrcIssBuild(l, bits(machInst, 3, 0), rt, 2247 reg, a, bits(machInst, 7, 5)); 2248 return new Vmrs(machInst, rt, (IntRegIndex)specReg, iss); 2249 } 2250 } 2251 } else { 2252 uint32_t vd = (bits(machInst, 7) << 5) | 2253 (bits(machInst, 19, 16) << 1); 2254 // Handle indexing into each single precision half of the vector. 2255 vd += bits(machInst, 21); 2256 uint32_t index; 2257 const IntRegIndex rt = 2258 (IntRegIndex)(uint32_t)bits(machInst, 15, 12); 2259 const bool u = (bits(machInst, 23) == 1); 2260 if (bits(machInst, 22) == 1) { 2261 index = bits(machInst, 6, 5); 2262 if (u) { 2263 return new VmovRegCoreUB(machInst, rt, 2264 (IntRegIndex)vd, index); 2265 } else { 2266 return new VmovRegCoreSB(machInst, rt, 2267 (IntRegIndex)vd, index); 2268 } 2269 } else if (bits(machInst, 5) == 1) { 2270 index = bits(machInst, 6); 2271 if (u) { 2272 return new VmovRegCoreUH(machInst, rt, 2273 (IntRegIndex)vd, index); 2274 } else { 2275 return new VmovRegCoreSH(machInst, rt, 2276 (IntRegIndex)vd, index); 2277 } 2278 } else if (bits(machInst, 6) == 0 && !u) { 2279 return new VmovRegCoreW(machInst, rt, (IntRegIndex)vd); 2280 } else { 2281 return new Unknown(machInst); 2282 } 2283 } 2284 return new Unknown(machInst); 2285 } 2286 ''' 2287}}; 2288 2289def format ShortFpTransfer() {{ 2290 decode_block = ''' 2291 return decodeShortFpTransfer(machInst); 2292 ''' 2293}}; 2294 2295let {{ 2296 header_output = ''' 2297 StaticInstPtr 2298 decodeVfpData(ExtMachInst machInst); 2299 ''' 2300 decoder_output = ''' 2301 StaticInstPtr 2302 decodeVfpData(ExtMachInst machInst) 2303 { 2304 const uint32_t opc1 = bits(machInst, 23, 20); 2305 const uint32_t opc2 = bits(machInst, 19, 16); 2306 const uint32_t opc3 = bits(machInst, 7, 6); 2307 //const uint32_t opc4 = bits(machInst, 3, 0); 2308 const bool single = (bits(machInst, 8) == 0); 2309 // Used to select between vcmp and vcmpe. 2310 const bool e = (bits(machInst, 7) == 1); 2311 IntRegIndex vd; 2312 IntRegIndex vm; 2313 IntRegIndex vn; 2314 if (single) { 2315 vd = (IntRegIndex)(bits(machInst, 22) | 2316 (bits(machInst, 15, 12) << 1)); 2317 vm = (IntRegIndex)(bits(machInst, 5) | 2318 (bits(machInst, 3, 0) << 1)); 2319 vn = (IntRegIndex)(bits(machInst, 7) | 2320 (bits(machInst, 19, 16) << 1)); 2321 } else { 2322 vd = (IntRegIndex)((bits(machInst, 22) << 5) | 2323 (bits(machInst, 15, 12) << 1)); 2324 vm = (IntRegIndex)((bits(machInst, 5) << 5) | 2325 (bits(machInst, 3, 0) << 1)); 2326 vn = (IntRegIndex)((bits(machInst, 7) << 5) | 2327 (bits(machInst, 19, 16) << 1)); 2328 } 2329 switch (opc1 & 0xb /* 1011 */) { 2330 case 0x0: 2331 if (bits(machInst, 6) == 0) { 2332 if (single) { 2333 return decodeVfpRegRegRegOp<VmlaS>( 2334 machInst, vd, vn, vm, false); 2335 } else { 2336 return decodeVfpRegRegRegOp<VmlaD>( 2337 machInst, vd, vn, vm, true); 2338 } 2339 } else { 2340 if (single) { 2341 return decodeVfpRegRegRegOp<VmlsS>( 2342 machInst, vd, vn, vm, false); 2343 } else { 2344 return decodeVfpRegRegRegOp<VmlsD>( 2345 machInst, vd, vn, vm, true); 2346 } 2347 } 2348 case 0x1: 2349 if (bits(machInst, 6) == 1) { 2350 if (single) { 2351 return decodeVfpRegRegRegOp<VnmlaS>( 2352 machInst, vd, vn, vm, false); 2353 } else { 2354 return decodeVfpRegRegRegOp<VnmlaD>( 2355 machInst, vd, vn, vm, true); 2356 } 2357 } else { 2358 if (single) { 2359 return decodeVfpRegRegRegOp<VnmlsS>( 2360 machInst, vd, vn, vm, false); 2361 } else { 2362 return decodeVfpRegRegRegOp<VnmlsD>( 2363 machInst, vd, vn, vm, true); 2364 } 2365 } 2366 case 0x2: 2367 if ((opc3 & 0x1) == 0) { 2368 if (single) { 2369 return decodeVfpRegRegRegOp<VmulS>( 2370 machInst, vd, vn, vm, false); 2371 } else { 2372 return decodeVfpRegRegRegOp<VmulD>( 2373 machInst, vd, vn, vm, true); 2374 } 2375 } else { 2376 if (single) { 2377 return decodeVfpRegRegRegOp<VnmulS>( 2378 machInst, vd, vn, vm, false); 2379 } else { 2380 return decodeVfpRegRegRegOp<VnmulD>( 2381 machInst, vd, vn, vm, true); 2382 } 2383 } 2384 case 0x3: 2385 if ((opc3 & 0x1) == 0) { 2386 if (single) { 2387 return decodeVfpRegRegRegOp<VaddS>( 2388 machInst, vd, vn, vm, false); 2389 } else { 2390 return decodeVfpRegRegRegOp<VaddD>( 2391 machInst, vd, vn, vm, true); 2392 } 2393 } else { 2394 if (single) { 2395 return decodeVfpRegRegRegOp<VsubS>( 2396 machInst, vd, vn, vm, false); 2397 } else { 2398 return decodeVfpRegRegRegOp<VsubD>( 2399 machInst, vd, vn, vm, true); 2400 } 2401 } 2402 case 0x8: 2403 if ((opc3 & 0x1) == 0) { 2404 if (single) { 2405 return decodeVfpRegRegRegOp<VdivS>( 2406 machInst, vd, vn, vm, false); 2407 } else { 2408 return decodeVfpRegRegRegOp<VdivD>( 2409 machInst, vd, vn, vm, true); 2410 } 2411 } 2412 break; 2413 case 0x9: 2414 if ((opc3 & 0x1) == 0) { 2415 if (single) { 2416 return decodeVfpRegRegRegOp<VfnmaS>( 2417 machInst, vd, vn, vm, false); 2418 } else { 2419 return decodeVfpRegRegRegOp<VfnmaD>( 2420 machInst, vd, vn, vm, true); 2421 } 2422 } else { 2423 if (single) { 2424 return decodeVfpRegRegRegOp<VfnmsS>( 2425 machInst, vd, vn, vm, false); 2426 } else { 2427 return decodeVfpRegRegRegOp<VfnmsD>( 2428 machInst, vd, vn, vm, true); 2429 } 2430 } 2431 break; 2432 case 0xa: 2433 if ((opc3 & 0x1) == 0) { 2434 if (single) { 2435 return decodeVfpRegRegRegOp<VfmaS>( 2436 machInst, vd, vn, vm, false); 2437 } else { 2438 return decodeVfpRegRegRegOp<VfmaD>( 2439 machInst, vd, vn, vm, true); 2440 } 2441 } else { 2442 if (single) { 2443 return decodeVfpRegRegRegOp<VfmsS>( 2444 machInst, vd, vn, vm, false); 2445 } else { 2446 return decodeVfpRegRegRegOp<VfmsD>( 2447 machInst, vd, vn, vm, true); 2448 } 2449 } 2450 break; 2451 case 0xb: 2452 if ((opc3 & 0x1) == 0) { 2453 const uint32_t baseImm = 2454 bits(machInst, 3, 0) | (bits(machInst, 19, 16) << 4); 2455 if (single) { 2456 uint32_t imm = vfp_modified_imm(baseImm, FpDataType::Fp32); 2457 return decodeVfpRegImmOp<VmovImmS>( 2458 machInst, vd, imm, false); 2459 } else { 2460 uint64_t imm = vfp_modified_imm(baseImm, FpDataType::Fp64); 2461 return decodeVfpRegImmOp<VmovImmD>( 2462 machInst, vd, imm, true); 2463 } 2464 } 2465 switch (opc2) { 2466 case 0x0: 2467 if (opc3 == 1) { 2468 if (single) { 2469 return decodeVfpRegRegOp<VmovRegS>( 2470 machInst, vd, vm, false); 2471 } else { 2472 return decodeVfpRegRegOp<VmovRegD>( 2473 machInst, vd, vm, true); 2474 } 2475 } else { 2476 if (single) { 2477 return decodeVfpRegRegOp<VabsS>( 2478 machInst, vd, vm, false); 2479 } else { 2480 return decodeVfpRegRegOp<VabsD>( 2481 machInst, vd, vm, true); 2482 } 2483 } 2484 case 0x1: 2485 if (opc3 == 1) { 2486 if (single) { 2487 return decodeVfpRegRegOp<VnegS>( 2488 machInst, vd, vm, false); 2489 } else { 2490 return decodeVfpRegRegOp<VnegD>( 2491 machInst, vd, vm, true); 2492 } 2493 } else { 2494 if (single) { 2495 return decodeVfpRegRegOp<VsqrtS>( 2496 machInst, vd, vm, false); 2497 } else { 2498 return decodeVfpRegRegOp<VsqrtD>( 2499 machInst, vd, vm, true); 2500 } 2501 } 2502 case 0x2: 2503 case 0x3: 2504 { 2505 const bool toHalf = bits(machInst, 16); 2506 const bool top = bits(machInst, 7); 2507 if (top) { 2508 if (toHalf) { 2509 return new VcvtFpSFpHT(machInst, vd, vm); 2510 } else { 2511 return new VcvtFpHTFpS(machInst, vd, vm); 2512 } 2513 } else { 2514 if (toHalf) { 2515 return new VcvtFpSFpHB(machInst, vd, vm); 2516 } else { 2517 return new VcvtFpHBFpS(machInst, vd, vm); 2518 } 2519 } 2520 } 2521 case 0x4: 2522 if (single) { 2523 if (e) { 2524 return new VcmpeS(machInst, vd, vm); 2525 } else { 2526 return new VcmpS(machInst, vd, vm); 2527 } 2528 } else { 2529 if (e) { 2530 return new VcmpeD(machInst, vd, vm); 2531 } else { 2532 return new VcmpD(machInst, vd, vm); 2533 } 2534 } 2535 case 0x5: 2536 if (single) { 2537 if (e) { 2538 return new VcmpeZeroS(machInst, vd, 0); 2539 } else { 2540 return new VcmpZeroS(machInst, vd, 0); 2541 } 2542 } else { 2543 if (e) { 2544 return new VcmpeZeroD(machInst, vd, 0); 2545 } else { 2546 return new VcmpZeroD(machInst, vd, 0); 2547 } 2548 } 2549 case 0x7: 2550 if (opc3 == 0x3) { 2551 if (single) { 2552 vd = (IntRegIndex)((bits(machInst, 22) << 5) | 2553 (bits(machInst, 15, 12) << 1)); 2554 return new VcvtFpSFpD(machInst, vd, vm); 2555 } else { 2556 vd = (IntRegIndex)(bits(machInst, 22) | 2557 (bits(machInst, 15, 12) << 1)); 2558 return new VcvtFpDFpS(machInst, vd, vm); 2559 } 2560 } 2561 break; 2562 case 0x8: 2563 if (bits(machInst, 7) == 0) { 2564 if (single) { 2565 return new VcvtUIntFpS(machInst, vd, vm); 2566 } else { 2567 vm = (IntRegIndex)(bits(machInst, 5) | 2568 (bits(machInst, 3, 0) << 1)); 2569 return new VcvtUIntFpD(machInst, vd, vm); 2570 } 2571 } else { 2572 if (single) { 2573 return new VcvtSIntFpS(machInst, vd, vm); 2574 } else { 2575 vm = (IntRegIndex)(bits(machInst, 5) | 2576 (bits(machInst, 3, 0) << 1)); 2577 return new VcvtSIntFpD(machInst, vd, vm); 2578 } 2579 } 2580 case 0xa: 2581 { 2582 const bool half = (bits(machInst, 7) == 0); 2583 const uint32_t imm = bits(machInst, 5) | 2584 (bits(machInst, 3, 0) << 1); 2585 const uint32_t size = 2586 (bits(machInst, 7) == 0 ? 16 : 32) - imm; 2587 if (single) { 2588 if (half) { 2589 return new VcvtSHFixedFpS(machInst, vd, vd, size); 2590 } else { 2591 return new VcvtSFixedFpS(machInst, vd, vd, size); 2592 } 2593 } else { 2594 if (half) { 2595 return new VcvtSHFixedFpD(machInst, vd, vd, size); 2596 } else { 2597 return new VcvtSFixedFpD(machInst, vd, vd, size); 2598 } 2599 } 2600 } 2601 case 0xb: 2602 { 2603 const bool half = (bits(machInst, 7) == 0); 2604 const uint32_t imm = bits(machInst, 5) | 2605 (bits(machInst, 3, 0) << 1); 2606 const uint32_t size = 2607 (bits(machInst, 7) == 0 ? 16 : 32) - imm; 2608 if (single) { 2609 if (half) { 2610 return new VcvtUHFixedFpS(machInst, vd, vd, size); 2611 } else { 2612 return new VcvtUFixedFpS(machInst, vd, vd, size); 2613 } 2614 } else { 2615 if (half) { 2616 return new VcvtUHFixedFpD(machInst, vd, vd, size); 2617 } else { 2618 return new VcvtUFixedFpD(machInst, vd, vd, size); 2619 } 2620 } 2621 } 2622 case 0xc: 2623 if (bits(machInst, 7) == 0) { 2624 if (single) { 2625 return new VcvtFpUIntSR(machInst, vd, vm); 2626 } else { 2627 vd = (IntRegIndex)(bits(machInst, 22) | 2628 (bits(machInst, 15, 12) << 1)); 2629 return new VcvtFpUIntDR(machInst, vd, vm); 2630 } 2631 } else { 2632 if (single) { 2633 return new VcvtFpUIntS(machInst, vd, vm); 2634 } else { 2635 vd = (IntRegIndex)(bits(machInst, 22) | 2636 (bits(machInst, 15, 12) << 1)); 2637 return new VcvtFpUIntD(machInst, vd, vm); 2638 } 2639 } 2640 case 0xd: 2641 if (bits(machInst, 7) == 0) { 2642 if (single) { 2643 return new VcvtFpSIntSR(machInst, vd, vm); 2644 } else { 2645 vd = (IntRegIndex)(bits(machInst, 22) | 2646 (bits(machInst, 15, 12) << 1)); 2647 return new VcvtFpSIntDR(machInst, vd, vm); 2648 } 2649 } else { 2650 if (single) { 2651 return new VcvtFpSIntS(machInst, vd, vm); 2652 } else { 2653 vd = (IntRegIndex)(bits(machInst, 22) | 2654 (bits(machInst, 15, 12) << 1)); 2655 return new VcvtFpSIntD(machInst, vd, vm); 2656 } 2657 } 2658 case 0xe: 2659 { 2660 const bool half = (bits(machInst, 7) == 0); 2661 const uint32_t imm = bits(machInst, 5) | 2662 (bits(machInst, 3, 0) << 1); 2663 const uint32_t size = 2664 (bits(machInst, 7) == 0 ? 16 : 32) - imm; 2665 if (single) { 2666 if (half) { 2667 return new VcvtFpSHFixedS(machInst, vd, vd, size); 2668 } else { 2669 return new VcvtFpSFixedS(machInst, vd, vd, size); 2670 } 2671 } else { 2672 if (half) { 2673 return new VcvtFpSHFixedD(machInst, vd, vd, size); 2674 } else { 2675 return new VcvtFpSFixedD(machInst, vd, vd, size); 2676 } 2677 } 2678 } 2679 case 0xf: 2680 { 2681 const bool half = (bits(machInst, 7) == 0); 2682 const uint32_t imm = bits(machInst, 5) | 2683 (bits(machInst, 3, 0) << 1); 2684 const uint32_t size = 2685 (bits(machInst, 7) == 0 ? 16 : 32) - imm; 2686 if (single) { 2687 if (half) { 2688 return new VcvtFpUHFixedS(machInst, vd, vd, size); 2689 } else { 2690 return new VcvtFpUFixedS(machInst, vd, vd, size); 2691 } 2692 } else { 2693 if (half) { 2694 return new VcvtFpUHFixedD(machInst, vd, vd, size); 2695 } else { 2696 return new VcvtFpUFixedD(machInst, vd, vd, size); 2697 } 2698 } 2699 } 2700 } 2701 break; 2702 } 2703 return new Unknown(machInst); 2704 } 2705 ''' 2706}}; 2707 2708def format VfpData() {{ 2709 decode_block = ''' 2710 return decodeVfpData(machInst); 2711 ''' 2712}}; 2713