neon.isa revision 10037:5cac77888310
1// -*- mode:c++ -*- 2 3// Copyright (c) 2010-2011 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// Redistribution and use in source and binary forms, with or without 16// modification, are permitted provided that the following conditions are 17// met: redistributions of source code must retain the above copyright 18// notice, this list of conditions and the following disclaimer; 19// redistributions in binary form must reproduce the above copyright 20// notice, this list of conditions and the following disclaimer in the 21// documentation and/or other materials provided with the distribution; 22// neither the name of the copyright holders nor the names of its 23// contributors may be used to endorse or promote products derived from 24// this software without specific prior written permission. 25// 26// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 27// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 28// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 29// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 30// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 31// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 32// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 33// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 34// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 35// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 36// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 37// 38// Authors: Gabe Black 39 40output header {{ 41 template <template <typename T> class Base> 42 StaticInstPtr 43 decodeNeonUThreeUReg(unsigned size, 44 ExtMachInst machInst, IntRegIndex dest, 45 IntRegIndex op1, IntRegIndex op2) 46 { 47 switch (size) { 48 case 0: 49 return new Base<uint8_t>(machInst, dest, op1, op2); 50 case 1: 51 return new Base<uint16_t>(machInst, dest, op1, op2); 52 case 2: 53 return new Base<uint32_t>(machInst, dest, op1, op2); 54 case 3: 55 return new Base<uint64_t>(machInst, dest, op1, op2); 56 default: 57 return new Unknown(machInst); 58 } 59 } 60 61 template <template <typename T> class Base> 62 StaticInstPtr 63 decodeNeonSThreeUReg(unsigned size, 64 ExtMachInst machInst, IntRegIndex dest, 65 IntRegIndex op1, IntRegIndex op2) 66 { 67 switch (size) { 68 case 0: 69 return new Base<int8_t>(machInst, dest, op1, op2); 70 case 1: 71 return new Base<int16_t>(machInst, dest, op1, op2); 72 case 2: 73 return new Base<int32_t>(machInst, dest, op1, op2); 74 case 3: 75 return new Base<int64_t>(machInst, dest, op1, op2); 76 default: 77 return new Unknown(machInst); 78 } 79 } 80 81 template <template <typename T> class Base> 82 StaticInstPtr 83 decodeNeonUSThreeUReg(bool notSigned, unsigned size, 84 ExtMachInst machInst, IntRegIndex dest, 85 IntRegIndex op1, IntRegIndex op2) 86 { 87 if (notSigned) { 88 return decodeNeonUThreeUReg<Base>(size, machInst, dest, op1, op2); 89 } else { 90 return decodeNeonSThreeUReg<Base>(size, machInst, dest, op1, op2); 91 } 92 } 93 94 template <template <typename T> class Base> 95 StaticInstPtr 96 decodeNeonUThreeUSReg(unsigned size, 97 ExtMachInst machInst, IntRegIndex dest, 98 IntRegIndex op1, IntRegIndex op2) 99 { 100 switch (size) { 101 case 0: 102 return new Base<uint8_t>(machInst, dest, op1, op2); 103 case 1: 104 return new Base<uint16_t>(machInst, dest, op1, op2); 105 case 2: 106 return new Base<uint32_t>(machInst, dest, op1, op2); 107 default: 108 return new Unknown(machInst); 109 } 110 } 111 112 template <template <typename T> class Base> 113 StaticInstPtr 114 decodeNeonSThreeUSReg(unsigned size, 115 ExtMachInst machInst, IntRegIndex dest, 116 IntRegIndex op1, IntRegIndex op2) 117 { 118 switch (size) { 119 case 0: 120 return new Base<int8_t>(machInst, dest, op1, op2); 121 case 1: 122 return new Base<int16_t>(machInst, dest, op1, op2); 123 case 2: 124 return new Base<int32_t>(machInst, dest, op1, op2); 125 default: 126 return new Unknown(machInst); 127 } 128 } 129 130 template <template <typename T> class Base> 131 StaticInstPtr 132 decodeNeonSThreeHAndWReg(unsigned size, ExtMachInst machInst, 133 IntRegIndex dest, IntRegIndex op1, 134 IntRegIndex op2) 135 { 136 switch (size) { 137 case 1: 138 return new Base<int16_t>(machInst, dest, op1, op2); 139 case 2: 140 return new Base<int32_t>(machInst, dest, op1, op2); 141 default: 142 return new Unknown(machInst); 143 } 144 } 145 146 template <template <typename T> class Base> 147 StaticInstPtr 148 decodeNeonSThreeImmHAndWReg(unsigned size, ExtMachInst machInst, 149 IntRegIndex dest, IntRegIndex op1, 150 IntRegIndex op2, uint64_t imm) 151 { 152 switch (size) { 153 case 1: 154 return new Base<int16_t>(machInst, dest, op1, op2, imm); 155 case 2: 156 return new Base<int32_t>(machInst, dest, op1, op2, imm); 157 default: 158 return new Unknown(machInst); 159 } 160 } 161 162 template <template <typename T> class Base> 163 StaticInstPtr 164 decodeNeonUSThreeUSReg(bool notSigned, unsigned size, 165 ExtMachInst machInst, IntRegIndex dest, 166 IntRegIndex op1, IntRegIndex op2) 167 { 168 if (notSigned) { 169 return decodeNeonUThreeUSReg<Base>( 170 size, machInst, dest, op1, op2); 171 } else { 172 return decodeNeonSThreeUSReg<Base>( 173 size, machInst, dest, op1, op2); 174 } 175 } 176 177 template <template <typename T> class BaseD, 178 template <typename T> class BaseQ> 179 StaticInstPtr 180 decodeNeonUThreeSReg(bool q, unsigned size, 181 ExtMachInst machInst, IntRegIndex dest, 182 IntRegIndex op1, IntRegIndex op2) 183 { 184 if (q) { 185 return decodeNeonUThreeUSReg<BaseQ>( 186 size, machInst, dest, op1, op2); 187 } else { 188 return decodeNeonUThreeUSReg<BaseD>( 189 size, machInst, dest, op1, op2); 190 } 191 } 192 193 template <template <typename T> class BaseD, 194 template <typename T> class BaseQ> 195 StaticInstPtr 196 decodeNeonSThreeSReg(bool q, unsigned size, 197 ExtMachInst machInst, IntRegIndex dest, 198 IntRegIndex op1, IntRegIndex op2) 199 { 200 if (q) { 201 return decodeNeonSThreeUSReg<BaseQ>( 202 size, machInst, dest, op1, op2); 203 } else { 204 return decodeNeonSThreeUSReg<BaseD>( 205 size, machInst, dest, op1, op2); 206 } 207 } 208 209 template <template <typename T> class BaseD, 210 template <typename T> class BaseQ> 211 StaticInstPtr 212 decodeNeonSThreeXReg(bool q, unsigned size, 213 ExtMachInst machInst, IntRegIndex dest, 214 IntRegIndex op1, IntRegIndex op2) 215 { 216 if (q) { 217 return decodeNeonSThreeUReg<BaseQ>( 218 size, machInst, dest, op1, op2); 219 } else { 220 return decodeNeonSThreeUSReg<BaseD>( 221 size, machInst, dest, op1, op2); 222 } 223 } 224 225 template <template <typename T> class BaseD, 226 template <typename T> class BaseQ> 227 StaticInstPtr 228 decodeNeonUThreeXReg(bool q, unsigned size, 229 ExtMachInst machInst, IntRegIndex dest, 230 IntRegIndex op1, IntRegIndex op2) 231 { 232 if (q) { 233 return decodeNeonUThreeUReg<BaseQ>( 234 size, machInst, dest, op1, op2); 235 } else { 236 return decodeNeonUThreeUSReg<BaseD>( 237 size, machInst, dest, op1, op2); 238 } 239 } 240 241 template <template <typename T> class BaseD, 242 template <typename T> class BaseQ> 243 StaticInstPtr 244 decodeNeonUSThreeSReg(bool q, bool notSigned, unsigned size, 245 ExtMachInst machInst, IntRegIndex dest, 246 IntRegIndex op1, IntRegIndex op2) 247 { 248 if (notSigned) { 249 return decodeNeonUThreeSReg<BaseD, BaseQ>( 250 q, size, machInst, dest, op1, op2); 251 } else { 252 return decodeNeonSThreeSReg<BaseD, BaseQ>( 253 q, size, machInst, dest, op1, op2); 254 } 255 } 256 257 template <template <typename T> class BaseD, 258 template <typename T> class BaseQ> 259 StaticInstPtr 260 decodeNeonUThreeReg(bool q, unsigned size, 261 ExtMachInst machInst, IntRegIndex dest, 262 IntRegIndex op1, IntRegIndex op2) 263 { 264 if (q) { 265 return decodeNeonUThreeUReg<BaseQ>( 266 size, machInst, dest, op1, op2); 267 } else { 268 return decodeNeonUThreeUReg<BaseD>( 269 size, machInst, dest, op1, op2); 270 } 271 } 272 273 template <template <typename T> class BaseD, 274 template <typename T> class BaseQ> 275 StaticInstPtr 276 decodeNeonSThreeReg(bool q, unsigned size, 277 ExtMachInst machInst, IntRegIndex dest, 278 IntRegIndex op1, IntRegIndex op2) 279 { 280 if (q) { 281 return decodeNeonSThreeUReg<BaseQ>( 282 size, machInst, dest, op1, op2); 283 } else { 284 return decodeNeonSThreeUReg<BaseD>( 285 size, machInst, dest, op1, op2); 286 } 287 } 288 289 template <template <typename T> class BaseD, 290 template <typename T> class BaseQ> 291 StaticInstPtr 292 decodeNeonUSThreeReg(bool q, bool notSigned, unsigned size, 293 ExtMachInst machInst, IntRegIndex dest, 294 IntRegIndex op1, IntRegIndex op2) 295 { 296 if (notSigned) { 297 return decodeNeonUThreeReg<BaseD, BaseQ>( 298 q, size, machInst, dest, op1, op2); 299 } else { 300 return decodeNeonSThreeReg<BaseD, BaseQ>( 301 q, size, machInst, dest, op1, op2); 302 } 303 } 304 305 template <template <typename T> class BaseD, 306 template <typename T> class BaseQ> 307 StaticInstPtr 308 decodeNeonUThreeFpReg(bool q, unsigned size, ExtMachInst machInst, 309 IntRegIndex dest, IntRegIndex op1, IntRegIndex op2) 310 { 311 if (q) { 312 if (size) 313 return new BaseQ<uint64_t>(machInst, dest, op1, op2); 314 else 315 return new BaseQ<uint32_t>(machInst, dest, op1, op2); 316 } else { 317 if (size) 318 return new Unknown(machInst); 319 else 320 return new BaseD<uint32_t>(machInst, dest, op1, op2); 321 } 322 } 323 324 template <template <typename T> class Base> 325 StaticInstPtr 326 decodeNeonUThreeScFpReg(bool size, ExtMachInst machInst, 327 IntRegIndex dest, IntRegIndex op1, IntRegIndex op2) 328 { 329 if (size) 330 return new Base<uint64_t>(machInst, dest, op1, op2); 331 else 332 return new Base<uint32_t>(machInst, dest, op1, op2); 333 } 334 335 template <template <typename T> class Base> 336 StaticInstPtr 337 decodeNeonUThreeImmScFpReg(bool size, ExtMachInst machInst, 338 IntRegIndex dest, IntRegIndex op1, 339 IntRegIndex op2, uint64_t imm) 340 { 341 if (size) 342 return new Base<uint64_t>(machInst, dest, op1, op2, imm); 343 else 344 return new Base<uint32_t>(machInst, dest, op1, op2, imm); 345 } 346 347 template <template <typename T> class BaseD, 348 template <typename T> class BaseQ> 349 StaticInstPtr 350 decodeNeonUThreeImmHAndWReg(bool q, unsigned size, ExtMachInst machInst, 351 IntRegIndex dest, IntRegIndex op1, 352 IntRegIndex op2, uint64_t imm) 353 { 354 if (q) { 355 switch (size) { 356 case 1: 357 return new BaseQ<uint16_t>(machInst, dest, op1, op2, imm); 358 case 2: 359 return new BaseQ<uint32_t>(machInst, dest, op1, op2, imm); 360 default: 361 return new Unknown(machInst); 362 } 363 } else { 364 switch (size) { 365 case 1: 366 return new BaseD<uint16_t>(machInst, dest, op1, op2, imm); 367 case 2: 368 return new BaseD<uint32_t>(machInst, dest, op1, op2, imm); 369 default: 370 return new Unknown(machInst); 371 } 372 } 373 } 374 375 template <template <typename T> class BaseD, 376 template <typename T> class BaseQ> 377 StaticInstPtr 378 decodeNeonSThreeImmHAndWReg(bool q, unsigned size, ExtMachInst machInst, 379 IntRegIndex dest, IntRegIndex op1, 380 IntRegIndex op2, uint64_t imm) 381 { 382 if (q) { 383 switch (size) { 384 case 1: 385 return new BaseQ<int16_t>(machInst, dest, op1, op2, imm); 386 case 2: 387 return new BaseQ<int32_t>(machInst, dest, op1, op2, imm); 388 default: 389 return new Unknown(machInst); 390 } 391 } else { 392 switch (size) { 393 case 1: 394 return new BaseD<int16_t>(machInst, dest, op1, op2, imm); 395 case 2: 396 return new BaseD<int32_t>(machInst, dest, op1, op2, imm); 397 default: 398 return new Unknown(machInst); 399 } 400 } 401 } 402 403 template <template <typename T> class BaseD, 404 template <typename T> class BaseQ> 405 StaticInstPtr 406 decodeNeonUThreeImmFpReg(bool q, unsigned size, ExtMachInst machInst, 407 IntRegIndex dest, IntRegIndex op1, 408 IntRegIndex op2, uint64_t imm) 409 { 410 if (q) { 411 if (size) 412 return new BaseQ<uint64_t>(machInst, dest, op1, op2, imm); 413 else 414 return new BaseQ<uint32_t>(machInst, dest, op1, op2, imm); 415 } else { 416 if (size) 417 return new Unknown(machInst); 418 else 419 return new BaseD<uint32_t>(machInst, dest, op1, op2, imm); 420 } 421 } 422 423 template <template <typename T> class BaseD, 424 template <typename T> class BaseQ> 425 StaticInstPtr 426 decodeNeonUTwoShiftReg(bool q, unsigned size, 427 ExtMachInst machInst, IntRegIndex dest, 428 IntRegIndex op1, uint64_t imm) 429 { 430 if (q) { 431 switch (size) { 432 case 0: 433 return new BaseQ<uint8_t>(machInst, dest, op1, imm); 434 case 1: 435 return new BaseQ<uint16_t>(machInst, dest, op1, imm); 436 case 2: 437 return new BaseQ<uint32_t>(machInst, dest, op1, imm); 438 case 3: 439 return new BaseQ<uint64_t>(machInst, dest, op1, imm); 440 default: 441 return new Unknown(machInst); 442 } 443 } else { 444 switch (size) { 445 case 0: 446 return new BaseD<uint8_t>(machInst, dest, op1, imm); 447 case 1: 448 return new BaseD<uint16_t>(machInst, dest, op1, imm); 449 case 2: 450 return new BaseD<uint32_t>(machInst, dest, op1, imm); 451 case 3: 452 return new BaseD<uint64_t>(machInst, dest, op1, imm); 453 default: 454 return new Unknown(machInst); 455 } 456 } 457 } 458 459 template <template <typename T> class BaseD, 460 template <typename T> class BaseQ> 461 StaticInstPtr 462 decodeNeonSTwoShiftReg(bool q, unsigned size, 463 ExtMachInst machInst, IntRegIndex dest, 464 IntRegIndex op1, uint64_t imm) 465 { 466 if (q) { 467 switch (size) { 468 case 0: 469 return new BaseQ<int8_t>(machInst, dest, op1, imm); 470 case 1: 471 return new BaseQ<int16_t>(machInst, dest, op1, imm); 472 case 2: 473 return new BaseQ<int32_t>(machInst, dest, op1, imm); 474 case 3: 475 return new BaseQ<int64_t>(machInst, dest, op1, imm); 476 default: 477 return new Unknown(machInst); 478 } 479 } else { 480 switch (size) { 481 case 0: 482 return new BaseD<int8_t>(machInst, dest, op1, imm); 483 case 1: 484 return new BaseD<int16_t>(machInst, dest, op1, imm); 485 case 2: 486 return new BaseD<int32_t>(machInst, dest, op1, imm); 487 case 3: 488 return new BaseD<int64_t>(machInst, dest, op1, imm); 489 default: 490 return new Unknown(machInst); 491 } 492 } 493 } 494 495 496 template <template <typename T> class BaseD, 497 template <typename T> class BaseQ> 498 StaticInstPtr 499 decodeNeonUSTwoShiftReg(bool q, bool notSigned, unsigned size, 500 ExtMachInst machInst, IntRegIndex dest, 501 IntRegIndex op1, uint64_t imm) 502 { 503 if (notSigned) { 504 return decodeNeonUTwoShiftReg<BaseD, BaseQ>( 505 q, size, machInst, dest, op1, imm); 506 } else { 507 return decodeNeonSTwoShiftReg<BaseD, BaseQ>( 508 q, size, machInst, dest, op1, imm); 509 } 510 } 511 512 template <template <typename T> class Base> 513 StaticInstPtr 514 decodeNeonUTwoShiftUSReg(unsigned size, 515 ExtMachInst machInst, IntRegIndex dest, 516 IntRegIndex op1, uint64_t imm) 517 { 518 switch (size) { 519 case 0: 520 return new Base<uint8_t>(machInst, dest, op1, imm); 521 case 1: 522 return new Base<uint16_t>(machInst, dest, op1, imm); 523 case 2: 524 return new Base<uint32_t>(machInst, dest, op1, imm); 525 default: 526 return new Unknown(machInst); 527 } 528 } 529 530 template <template <typename T> class Base> 531 StaticInstPtr 532 decodeNeonUTwoShiftUReg(unsigned size, 533 ExtMachInst machInst, IntRegIndex dest, 534 IntRegIndex op1, uint64_t imm) 535 { 536 switch (size) { 537 case 0: 538 return new Base<uint8_t>(machInst, dest, op1, imm); 539 case 1: 540 return new Base<uint16_t>(machInst, dest, op1, imm); 541 case 2: 542 return new Base<uint32_t>(machInst, dest, op1, imm); 543 case 3: 544 return new Base<uint64_t>(machInst, dest, op1, imm); 545 default: 546 return new Unknown(machInst); 547 } 548 } 549 550 template <template <typename T> class Base> 551 StaticInstPtr 552 decodeNeonSTwoShiftUReg(unsigned size, 553 ExtMachInst machInst, IntRegIndex dest, 554 IntRegIndex op1, uint64_t imm) 555 { 556 switch (size) { 557 case 0: 558 return new Base<int8_t>(machInst, dest, op1, imm); 559 case 1: 560 return new Base<int16_t>(machInst, dest, op1, imm); 561 case 2: 562 return new Base<int32_t>(machInst, dest, op1, imm); 563 case 3: 564 return new Base<int64_t>(machInst, dest, op1, imm); 565 default: 566 return new Unknown(machInst); 567 } 568 } 569 570 template <template <typename T> class BaseD, 571 template <typename T> class BaseQ> 572 StaticInstPtr 573 decodeNeonUTwoShiftSReg(bool q, unsigned size, 574 ExtMachInst machInst, IntRegIndex dest, 575 IntRegIndex op1, uint64_t imm) 576 { 577 if (q) { 578 return decodeNeonUTwoShiftUSReg<BaseQ>( 579 size, machInst, dest, op1, imm); 580 } else { 581 return decodeNeonUTwoShiftUSReg<BaseD>( 582 size, machInst, dest, op1, imm); 583 } 584 } 585 586 template <template <typename T> class Base> 587 StaticInstPtr 588 decodeNeonSTwoShiftUSReg(unsigned size, 589 ExtMachInst machInst, IntRegIndex dest, 590 IntRegIndex op1, uint64_t imm) 591 { 592 switch (size) { 593 case 0: 594 return new Base<int8_t>(machInst, dest, op1, imm); 595 case 1: 596 return new Base<int16_t>(machInst, dest, op1, imm); 597 case 2: 598 return new Base<int32_t>(machInst, dest, op1, imm); 599 default: 600 return new Unknown(machInst); 601 } 602 } 603 604 template <template <typename T> class BaseD, 605 template <typename T> class BaseQ> 606 StaticInstPtr 607 decodeNeonSTwoShiftSReg(bool q, unsigned size, 608 ExtMachInst machInst, IntRegIndex dest, 609 IntRegIndex op1, uint64_t imm) 610 { 611 if (q) { 612 return decodeNeonSTwoShiftUSReg<BaseQ>( 613 size, machInst, dest, op1, imm); 614 } else { 615 return decodeNeonSTwoShiftUSReg<BaseD>( 616 size, machInst, dest, op1, imm); 617 } 618 } 619 620 template <template <typename T> class BaseD, 621 template <typename T> class BaseQ> 622 StaticInstPtr 623 decodeNeonUSTwoShiftSReg(bool q, bool notSigned, unsigned size, 624 ExtMachInst machInst, IntRegIndex dest, 625 IntRegIndex op1, uint64_t imm) 626 { 627 if (notSigned) { 628 return decodeNeonUTwoShiftSReg<BaseD, BaseQ>( 629 q, size, machInst, dest, op1, imm); 630 } else { 631 return decodeNeonSTwoShiftSReg<BaseD, BaseQ>( 632 q, size, machInst, dest, op1, imm); 633 } 634 } 635 636 template <template <typename T> class BaseD, 637 template <typename T> class BaseQ> 638 StaticInstPtr 639 decodeNeonUTwoShiftXReg(bool q, unsigned size, ExtMachInst machInst, 640 IntRegIndex dest, IntRegIndex op1, uint64_t imm) 641 { 642 if (q) { 643 return decodeNeonUTwoShiftUReg<BaseQ>( 644 size, machInst, dest, op1, imm); 645 } else { 646 return decodeNeonUTwoShiftUSReg<BaseD>( 647 size, machInst, dest, op1, imm); 648 } 649 } 650 651 template <template <typename T> class BaseD, 652 template <typename T> class BaseQ> 653 StaticInstPtr 654 decodeNeonSTwoShiftXReg(bool q, unsigned size, ExtMachInst machInst, 655 IntRegIndex dest, IntRegIndex op1, uint64_t imm) 656 { 657 if (q) { 658 return decodeNeonSTwoShiftUReg<BaseQ>( 659 size, machInst, dest, op1, imm); 660 } else { 661 return decodeNeonSTwoShiftUSReg<BaseD>( 662 size, machInst, dest, op1, imm); 663 } 664 } 665 666 template <template <typename T> class Base> 667 StaticInstPtr 668 decodeNeonUTwoShiftUFpReg(unsigned size, ExtMachInst machInst, 669 IntRegIndex dest, IntRegIndex op1, uint64_t imm) 670 { 671 if (size) 672 return new Base<uint64_t>(machInst, dest, op1, imm); 673 else 674 return new Base<uint32_t>(machInst, dest, op1, imm); 675 } 676 677 template <template <typename T> class BaseD, 678 template <typename T> class BaseQ> 679 StaticInstPtr 680 decodeNeonUTwoShiftFpReg(bool q, unsigned size, ExtMachInst machInst, 681 IntRegIndex dest, IntRegIndex op1, uint64_t imm) 682 { 683 if (q) { 684 if (size) 685 return new BaseQ<uint64_t>(machInst, dest, op1, imm); 686 else 687 return new BaseQ<uint32_t>(machInst, dest, op1, imm); 688 } else { 689 if (size) 690 return new Unknown(machInst); 691 else 692 return new BaseD<uint32_t>(machInst, dest, op1, imm); 693 } 694 } 695 696 template <template <typename T> class Base> 697 StaticInstPtr 698 decodeNeonUTwoMiscUSReg(unsigned size, 699 ExtMachInst machInst, IntRegIndex dest, 700 IntRegIndex op1) 701 { 702 switch (size) { 703 case 0: 704 return new Base<uint8_t>(machInst, dest, op1); 705 case 1: 706 return new Base<uint16_t>(machInst, dest, op1); 707 case 2: 708 return new Base<uint32_t>(machInst, dest, op1); 709 default: 710 return new Unknown(machInst); 711 } 712 } 713 714 template <template <typename T> class Base> 715 StaticInstPtr 716 decodeNeonSTwoMiscUSReg(unsigned size, 717 ExtMachInst machInst, IntRegIndex dest, 718 IntRegIndex op1) 719 { 720 switch (size) { 721 case 0: 722 return new Base<int8_t>(machInst, dest, op1); 723 case 1: 724 return new Base<int16_t>(machInst, dest, op1); 725 case 2: 726 return new Base<int32_t>(machInst, dest, op1); 727 default: 728 return new Unknown(machInst); 729 } 730 } 731 732 template <template <typename T> class BaseD, 733 template <typename T> class BaseQ> 734 StaticInstPtr 735 decodeNeonUTwoMiscSReg(bool q, unsigned size, 736 ExtMachInst machInst, IntRegIndex dest, 737 IntRegIndex op1) 738 { 739 if (q) { 740 return decodeNeonUTwoMiscUSReg<BaseQ>(size, machInst, dest, op1); 741 } else { 742 return decodeNeonUTwoMiscUSReg<BaseD>(size, machInst, dest, op1); 743 } 744 } 745 746 template <template <typename T> class BaseD, 747 template <typename T> class BaseQ> 748 StaticInstPtr 749 decodeNeonSTwoMiscSReg(bool q, unsigned size, 750 ExtMachInst machInst, IntRegIndex dest, 751 IntRegIndex op1) 752 { 753 if (q) { 754 return decodeNeonSTwoMiscUSReg<BaseQ>(size, machInst, dest, op1); 755 } else { 756 return decodeNeonSTwoMiscUSReg<BaseD>(size, machInst, dest, op1); 757 } 758 } 759 760 template <template <typename T> class Base> 761 StaticInstPtr 762 decodeNeonUTwoMiscUReg(unsigned size, 763 ExtMachInst machInst, IntRegIndex dest, 764 IntRegIndex op1) 765 { 766 switch (size) { 767 case 0: 768 return new Base<uint8_t>(machInst, dest, op1); 769 case 1: 770 return new Base<uint16_t>(machInst, dest, op1); 771 case 2: 772 return new Base<uint32_t>(machInst, dest, op1); 773 case 3: 774 return new Base<uint64_t>(machInst, dest, op1); 775 default: 776 return new Unknown(machInst); 777 } 778 } 779 780 template <template <typename T> class Base> 781 StaticInstPtr 782 decodeNeonSTwoMiscUReg(unsigned size, 783 ExtMachInst machInst, IntRegIndex dest, 784 IntRegIndex op1) 785 { 786 switch (size) { 787 case 0: 788 return new Base<int8_t>(machInst, dest, op1); 789 case 1: 790 return new Base<int16_t>(machInst, dest, op1); 791 case 2: 792 return new Base<int32_t>(machInst, dest, op1); 793 case 3: 794 return new Base<int64_t>(machInst, dest, op1); 795 default: 796 return new Unknown(machInst); 797 } 798 } 799 800 template <template <typename T> class BaseD, 801 template <typename T> class BaseQ> 802 StaticInstPtr 803 decodeNeonSTwoMiscReg(bool q, unsigned size, 804 ExtMachInst machInst, IntRegIndex dest, 805 IntRegIndex op1) 806 { 807 if (q) { 808 return decodeNeonSTwoMiscUReg<BaseQ>(size, machInst, dest, op1); 809 } else { 810 return decodeNeonSTwoMiscUReg<BaseD>(size, machInst, dest, op1); 811 } 812 } 813 814 template <template <typename T> class BaseD, 815 template <typename T> class BaseQ> 816 StaticInstPtr 817 decodeNeonUTwoMiscReg(bool q, unsigned size, 818 ExtMachInst machInst, IntRegIndex dest, 819 IntRegIndex op1) 820 { 821 if (q) { 822 return decodeNeonUTwoMiscUReg<BaseQ>(size, machInst, dest, op1); 823 } else { 824 return decodeNeonUTwoMiscUReg<BaseD>(size, machInst, dest, op1); 825 } 826 } 827 828 template <template <typename T> class BaseD, 829 template <typename T> class BaseQ> 830 StaticInstPtr 831 decodeNeonUSTwoMiscSReg(bool q, bool notSigned, unsigned size, 832 ExtMachInst machInst, IntRegIndex dest, 833 IntRegIndex op1) 834 { 835 if (notSigned) { 836 return decodeNeonUTwoShiftSReg<BaseD, BaseQ>( 837 q, size, machInst, dest, op1); 838 } else { 839 return decodeNeonSTwoShiftSReg<BaseD, BaseQ>( 840 q, size, machInst, dest, op1); 841 } 842 } 843 844 template <template <typename T> class BaseD, 845 template <typename T> class BaseQ> 846 StaticInstPtr 847 decodeNeonUTwoMiscXReg(bool q, unsigned size, ExtMachInst machInst, 848 IntRegIndex dest, IntRegIndex op1) 849 { 850 if (q) { 851 return decodeNeonUTwoMiscUReg<BaseQ>(size, machInst, dest, op1); 852 } else { 853 return decodeNeonUTwoMiscUSReg<BaseD>(size, machInst, dest, op1); 854 } 855 } 856 857 template <template <typename T> class BaseD, 858 template <typename T> class BaseQ> 859 StaticInstPtr 860 decodeNeonSTwoMiscXReg(bool q, unsigned size, ExtMachInst machInst, 861 IntRegIndex dest, IntRegIndex op1) 862 { 863 if (q) { 864 return decodeNeonSTwoMiscUReg<BaseQ>(size, machInst, dest, op1); 865 } else { 866 return decodeNeonSTwoMiscUSReg<BaseD>(size, machInst, dest, op1); 867 } 868 } 869 870 template <template <typename T> class BaseD, 871 template <typename T> class BaseQ> 872 StaticInstPtr 873 decodeNeonUTwoMiscFpReg(bool q, unsigned size, ExtMachInst machInst, 874 IntRegIndex dest, IntRegIndex op1) 875 { 876 if (q) { 877 if (size) 878 return new BaseQ<uint64_t>(machInst, dest, op1); 879 else 880 return new BaseQ<uint32_t>(machInst, dest, op1); 881 } else { 882 if (size) 883 return new Unknown(machInst); 884 else 885 return new BaseD<uint32_t>(machInst, dest, op1); 886 } 887 } 888 889 template <template <typename T> class BaseD, 890 template <typename T> class BaseQ> 891 StaticInstPtr 892 decodeNeonUTwoMiscPwiseScFpReg(unsigned size, ExtMachInst machInst, 893 IntRegIndex dest, IntRegIndex op1) 894 { 895 if (size) 896 return new BaseQ<uint64_t>(machInst, dest, op1); 897 else 898 return new BaseD<uint32_t>(machInst, dest, op1); 899 } 900 901 template <template <typename T> class Base> 902 StaticInstPtr 903 decodeNeonUTwoMiscScFpReg(unsigned size, ExtMachInst machInst, 904 IntRegIndex dest, IntRegIndex op1) 905 { 906 if (size) 907 return new Base<uint64_t>(machInst, dest, op1); 908 else 909 return new Base<uint32_t>(machInst, dest, op1); 910 } 911 912 template <template <typename T> class BaseD, 913 template <typename T> class BaseQ> 914 StaticInstPtr 915 decodeNeonUAcrossLanesReg(bool q, unsigned size, ExtMachInst machInst, 916 IntRegIndex dest, IntRegIndex op1) 917 { 918 if (q) { 919 switch (size) { 920 case 0x0: 921 return new BaseQ<uint8_t>(machInst, dest, op1); 922 case 0x1: 923 return new BaseQ<uint16_t>(machInst, dest, op1); 924 case 0x2: 925 return new BaseQ<uint32_t>(machInst, dest, op1); 926 default: 927 return new Unknown(machInst); 928 } 929 } else { 930 switch (size) { 931 case 0x0: 932 return new BaseD<uint8_t>(machInst, dest, op1); 933 case 0x1: 934 return new BaseD<uint16_t>(machInst, dest, op1); 935 default: 936 return new Unknown(machInst); 937 } 938 } 939 } 940 941 template <template <typename T> class BaseD, 942 template <typename T> class BaseQ, 943 template <typename T> class BaseBQ> 944 StaticInstPtr 945 decodeNeonUAcrossLanesReg(bool q, unsigned size, ExtMachInst machInst, 946 IntRegIndex dest, IntRegIndex op1) 947 { 948 if (q) { 949 switch (size) { 950 case 0x0: 951 return new BaseQ<uint8_t>(machInst, dest, op1); 952 case 0x1: 953 return new BaseQ<uint16_t>(machInst, dest, op1); 954 case 0x2: 955 return new BaseBQ<uint32_t>(machInst, dest, op1); 956 default: 957 return new Unknown(machInst); 958 } 959 } else { 960 switch (size) { 961 case 0x0: 962 return new BaseD<uint8_t>(machInst, dest, op1); 963 case 0x1: 964 return new BaseD<uint16_t>(machInst, dest, op1); 965 default: 966 return new Unknown(machInst); 967 } 968 } 969 } 970 971 template <template <typename T> class BaseD, 972 template <typename T> class BaseQ> 973 StaticInstPtr 974 decodeNeonSAcrossLanesReg(bool q, unsigned size, ExtMachInst machInst, 975 IntRegIndex dest, IntRegIndex op1) 976 { 977 if (q) { 978 switch (size) { 979 case 0x0: 980 return new BaseQ<int8_t>(machInst, dest, op1); 981 case 0x1: 982 return new BaseQ<int16_t>(machInst, dest, op1); 983 case 0x2: 984 return new BaseQ<int32_t>(machInst, dest, op1); 985 default: 986 return new Unknown(machInst); 987 } 988 } else { 989 switch (size) { 990 case 0x0: 991 return new BaseD<int8_t>(machInst, dest, op1); 992 case 0x1: 993 return new BaseD<int16_t>(machInst, dest, op1); 994 default: 995 return new Unknown(machInst); 996 } 997 } 998 } 999 1000 template <template <typename T> class BaseD, 1001 template <typename T> class BaseQ, 1002 template <typename T> class BaseBQ> 1003 StaticInstPtr 1004 decodeNeonUAcrossLanesLongReg(bool q, unsigned size, ExtMachInst machInst, 1005 IntRegIndex dest, IntRegIndex op1) 1006 { 1007 if (q) { 1008 switch (size) { 1009 case 0x0: 1010 return new BaseQ<uint8_t>(machInst, dest, op1); 1011 case 0x1: 1012 return new BaseQ<uint16_t>(machInst, dest, op1); 1013 case 0x2: 1014 return new BaseBQ<uint32_t>(machInst, dest, op1); 1015 default: 1016 return new Unknown(machInst); 1017 } 1018 } else { 1019 switch (size) { 1020 case 0x0: 1021 return new BaseD<uint8_t>(machInst, dest, op1); 1022 case 0x1: 1023 return new BaseD<uint16_t>(machInst, dest, op1); 1024 default: 1025 return new Unknown(machInst); 1026 } 1027 } 1028 } 1029 1030 template <template <typename T> class BaseD, 1031 template <typename T> class BaseQ, 1032 template <typename T> class BaseBQ> 1033 StaticInstPtr 1034 decodeNeonSAcrossLanesLongReg(bool q, unsigned size, ExtMachInst machInst, 1035 IntRegIndex dest, IntRegIndex op1) 1036 { 1037 if (q) { 1038 switch (size) { 1039 case 0x0: 1040 return new BaseQ<int8_t>(machInst, dest, op1); 1041 case 0x1: 1042 return new BaseQ<int16_t>(machInst, dest, op1); 1043 case 0x2: 1044 return new BaseBQ<int32_t>(machInst, dest, op1); 1045 default: 1046 return new Unknown(machInst); 1047 } 1048 } else { 1049 switch (size) { 1050 case 0x0: 1051 return new BaseD<int8_t>(machInst, dest, op1); 1052 case 0x1: 1053 return new BaseD<int16_t>(machInst, dest, op1); 1054 default: 1055 return new Unknown(machInst); 1056 } 1057 } 1058 } 1059}}; 1060 1061output exec {{ 1062 static float 1063 vcgtFunc(float op1, float op2) 1064 { 1065 if (std::isnan(op1) || std::isnan(op2)) 1066 return 2.0; 1067 return (op1 > op2) ? 0.0 : 1.0; 1068 } 1069 1070 static float 1071 vcgeFunc(float op1, float op2) 1072 { 1073 if (std::isnan(op1) || std::isnan(op2)) 1074 return 2.0; 1075 return (op1 >= op2) ? 0.0 : 1.0; 1076 } 1077 1078 static float 1079 vceqFunc(float op1, float op2) 1080 { 1081 if (isSnan(op1) || isSnan(op2)) 1082 return 2.0; 1083 return (op1 == op2) ? 0.0 : 1.0; 1084 } 1085 1086 static float 1087 vcleFunc(float op1, float op2) 1088 { 1089 if (std::isnan(op1) || std::isnan(op2)) 1090 return 2.0; 1091 return (op1 <= op2) ? 0.0 : 1.0; 1092 } 1093 1094 static float 1095 vcltFunc(float op1, float op2) 1096 { 1097 if (std::isnan(op1) || std::isnan(op2)) 1098 return 2.0; 1099 return (op1 < op2) ? 0.0 : 1.0; 1100 } 1101 1102 static float 1103 vacgtFunc(float op1, float op2) 1104 { 1105 if (std::isnan(op1) || std::isnan(op2)) 1106 return 2.0; 1107 return (fabsf(op1) > fabsf(op2)) ? 0.0 : 1.0; 1108 } 1109 1110 static float 1111 vacgeFunc(float op1, float op2) 1112 { 1113 if (std::isnan(op1) || std::isnan(op2)) 1114 return 2.0; 1115 return (fabsf(op1) >= fabsf(op2)) ? 0.0 : 1.0; 1116 } 1117}}; 1118 1119let {{ 1120 1121 header_output = "" 1122 exec_output = "" 1123 1124 smallUnsignedTypes = ("uint8_t", "uint16_t", "uint32_t") 1125 unsignedTypes = smallUnsignedTypes + ("uint64_t",) 1126 smallSignedTypes = ("int8_t", "int16_t", "int32_t") 1127 signedTypes = smallSignedTypes + ("int64_t",) 1128 smallTypes = smallUnsignedTypes + smallSignedTypes 1129 allTypes = unsignedTypes + signedTypes 1130 1131 def threeEqualRegInst(name, Name, opClass, types, rCount, op, 1132 readDest=False, pairwise=False): 1133 global header_output, exec_output 1134 eWalkCode = simdEnabledCheckCode + ''' 1135 RegVect srcReg1, srcReg2, destReg; 1136 ''' 1137 for reg in range(rCount): 1138 eWalkCode += ''' 1139 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1140 srcReg2.regs[%(reg)d] = htog(FpOp2P%(reg)d_uw); 1141 ''' % { "reg" : reg } 1142 if readDest: 1143 eWalkCode += ''' 1144 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1145 ''' % { "reg" : reg } 1146 readDestCode = '' 1147 if readDest: 1148 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1149 if pairwise: 1150 eWalkCode += ''' 1151 for (unsigned i = 0; i < eCount; i++) { 1152 Element srcElem1 = gtoh(2 * i < eCount ? 1153 srcReg1.elements[2 * i] : 1154 srcReg2.elements[2 * i - eCount]); 1155 Element srcElem2 = gtoh(2 * i < eCount ? 1156 srcReg1.elements[2 * i + 1] : 1157 srcReg2.elements[2 * i + 1 - eCount]); 1158 Element destElem; 1159 %(readDest)s 1160 %(op)s 1161 destReg.elements[i] = htog(destElem); 1162 } 1163 ''' % { "op" : op, "readDest" : readDestCode } 1164 else: 1165 eWalkCode += ''' 1166 for (unsigned i = 0; i < eCount; i++) { 1167 Element srcElem1 = gtoh(srcReg1.elements[i]); 1168 Element srcElem2 = gtoh(srcReg2.elements[i]); 1169 Element destElem; 1170 %(readDest)s 1171 %(op)s 1172 destReg.elements[i] = htog(destElem); 1173 } 1174 ''' % { "op" : op, "readDest" : readDestCode } 1175 for reg in range(rCount): 1176 eWalkCode += ''' 1177 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1178 ''' % { "reg" : reg } 1179 iop = InstObjParams(name, Name, 1180 "RegRegRegOp", 1181 { "code": eWalkCode, 1182 "r_count": rCount, 1183 "predicate_test": predicateTest, 1184 "op_class": opClass }, []) 1185 header_output += NeonRegRegRegOpDeclare.subst(iop) 1186 exec_output += NeonEqualRegExecute.subst(iop) 1187 for type in types: 1188 substDict = { "targs" : type, 1189 "class_name" : Name } 1190 exec_output += NeonExecDeclare.subst(substDict) 1191 1192 def threeEqualRegInstFp(name, Name, opClass, types, rCount, op, 1193 readDest=False, pairwise=False, toInt=False): 1194 global header_output, exec_output 1195 eWalkCode = simdEnabledCheckCode + ''' 1196 typedef FloatReg FloatVect[rCount]; 1197 FloatVect srcRegs1, srcRegs2; 1198 ''' 1199 if toInt: 1200 eWalkCode += 'RegVect destRegs;\n' 1201 else: 1202 eWalkCode += 'FloatVect destRegs;\n' 1203 for reg in range(rCount): 1204 eWalkCode += ''' 1205 srcRegs1[%(reg)d] = FpOp1P%(reg)d; 1206 srcRegs2[%(reg)d] = FpOp2P%(reg)d; 1207 ''' % { "reg" : reg } 1208 if readDest: 1209 if toInt: 1210 eWalkCode += ''' 1211 destRegs.regs[%(reg)d] = FpDestP%(reg)d.bits; 1212 ''' % { "reg" : reg } 1213 else: 1214 eWalkCode += ''' 1215 destRegs[%(reg)d] = FpDestP%(reg)d; 1216 ''' % { "reg" : reg } 1217 readDestCode = '' 1218 if readDest: 1219 readDestCode = 'destReg = destRegs[r];' 1220 destType = 'FloatReg' 1221 writeDest = 'destRegs[r] = destReg;' 1222 if toInt: 1223 destType = 'FloatRegBits' 1224 writeDest = 'destRegs.regs[r] = destReg;' 1225 if pairwise: 1226 eWalkCode += ''' 1227 for (unsigned r = 0; r < rCount; r++) { 1228 FloatReg srcReg1 = (2 * r < rCount) ? 1229 srcRegs1[2 * r] : srcRegs2[2 * r - rCount]; 1230 FloatReg srcReg2 = (2 * r < rCount) ? 1231 srcRegs1[2 * r + 1] : srcRegs2[2 * r + 1 - rCount]; 1232 %(destType)s destReg; 1233 %(readDest)s 1234 %(op)s 1235 %(writeDest)s 1236 } 1237 ''' % { "op" : op, 1238 "readDest" : readDestCode, 1239 "destType" : destType, 1240 "writeDest" : writeDest } 1241 else: 1242 eWalkCode += ''' 1243 for (unsigned r = 0; r < rCount; r++) { 1244 FloatReg srcReg1 = srcRegs1[r]; 1245 FloatReg srcReg2 = srcRegs2[r]; 1246 %(destType)s destReg; 1247 %(readDest)s 1248 %(op)s 1249 %(writeDest)s 1250 } 1251 ''' % { "op" : op, 1252 "readDest" : readDestCode, 1253 "destType" : destType, 1254 "writeDest" : writeDest } 1255 for reg in range(rCount): 1256 if toInt: 1257 eWalkCode += ''' 1258 FpDestP%(reg)d_uw = destRegs.regs[%(reg)d]; 1259 ''' % { "reg" : reg } 1260 else: 1261 eWalkCode += ''' 1262 FpDestP%(reg)d = destRegs[%(reg)d]; 1263 ''' % { "reg" : reg } 1264 iop = InstObjParams(name, Name, 1265 "FpRegRegRegOp", 1266 { "code": eWalkCode, 1267 "r_count": rCount, 1268 "predicate_test": predicateTest, 1269 "op_class": opClass }, []) 1270 header_output += NeonRegRegRegOpDeclare.subst(iop) 1271 exec_output += NeonEqualRegExecute.subst(iop) 1272 for type in types: 1273 substDict = { "targs" : type, 1274 "class_name" : Name } 1275 exec_output += NeonExecDeclare.subst(substDict) 1276 1277 def threeUnequalRegInst(name, Name, opClass, types, op, 1278 bigSrc1, bigSrc2, bigDest, readDest): 1279 global header_output, exec_output 1280 src1Cnt = src2Cnt = destCnt = 2 1281 src1Prefix = src2Prefix = destPrefix = '' 1282 if bigSrc1: 1283 src1Cnt = 4 1284 src1Prefix = 'Big' 1285 if bigSrc2: 1286 src2Cnt = 4 1287 src2Prefix = 'Big' 1288 if bigDest: 1289 destCnt = 4 1290 destPrefix = 'Big' 1291 eWalkCode = simdEnabledCheckCode + ''' 1292 %sRegVect srcReg1; 1293 %sRegVect srcReg2; 1294 %sRegVect destReg; 1295 ''' % (src1Prefix, src2Prefix, destPrefix) 1296 for reg in range(src1Cnt): 1297 eWalkCode += ''' 1298 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1299 ''' % { "reg" : reg } 1300 for reg in range(src2Cnt): 1301 eWalkCode += ''' 1302 srcReg2.regs[%(reg)d] = htog(FpOp2P%(reg)d_uw); 1303 ''' % { "reg" : reg } 1304 if readDest: 1305 for reg in range(destCnt): 1306 eWalkCode += ''' 1307 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1308 ''' % { "reg" : reg } 1309 readDestCode = '' 1310 if readDest: 1311 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1312 eWalkCode += ''' 1313 for (unsigned i = 0; i < eCount; i++) { 1314 %(src1Prefix)sElement srcElem1 = gtoh(srcReg1.elements[i]); 1315 %(src1Prefix)sElement srcElem2 = gtoh(srcReg2.elements[i]); 1316 %(destPrefix)sElement destElem; 1317 %(readDest)s 1318 %(op)s 1319 destReg.elements[i] = htog(destElem); 1320 } 1321 ''' % { "op" : op, "readDest" : readDestCode, 1322 "src1Prefix" : src1Prefix, "src2Prefix" : src2Prefix, 1323 "destPrefix" : destPrefix } 1324 for reg in range(destCnt): 1325 eWalkCode += ''' 1326 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1327 ''' % { "reg" : reg } 1328 iop = InstObjParams(name, Name, 1329 "RegRegRegOp", 1330 { "code": eWalkCode, 1331 "r_count": 2, 1332 "predicate_test": predicateTest, 1333 "op_class": opClass }, []) 1334 header_output += NeonRegRegRegOpDeclare.subst(iop) 1335 exec_output += NeonUnequalRegExecute.subst(iop) 1336 for type in types: 1337 substDict = { "targs" : type, 1338 "class_name" : Name } 1339 exec_output += NeonExecDeclare.subst(substDict) 1340 1341 def threeRegNarrowInst(name, Name, opClass, types, op, readDest=False): 1342 threeUnequalRegInst(name, Name, opClass, types, op, 1343 True, True, False, readDest) 1344 1345 def threeRegLongInst(name, Name, opClass, types, op, readDest=False): 1346 threeUnequalRegInst(name, Name, opClass, types, op, 1347 False, False, True, readDest) 1348 1349 def threeRegWideInst(name, Name, opClass, types, op, readDest=False): 1350 threeUnequalRegInst(name, Name, opClass, types, op, 1351 True, False, True, readDest) 1352 1353 def twoEqualRegInst(name, Name, opClass, types, rCount, op, readDest=False): 1354 global header_output, exec_output 1355 eWalkCode = simdEnabledCheckCode + ''' 1356 RegVect srcReg1, srcReg2, destReg; 1357 ''' 1358 for reg in range(rCount): 1359 eWalkCode += ''' 1360 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1361 srcReg2.regs[%(reg)d] = htog(FpOp2P%(reg)d_uw); 1362 ''' % { "reg" : reg } 1363 if readDest: 1364 eWalkCode += ''' 1365 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1366 ''' % { "reg" : reg } 1367 readDestCode = '' 1368 if readDest: 1369 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1370 eWalkCode += ''' 1371 if (imm < 0 && imm >= eCount) { 1372 fault = new UndefinedInstruction(machInst, false, mnemonic); 1373 } else { 1374 for (unsigned i = 0; i < eCount; i++) { 1375 Element srcElem1 = gtoh(srcReg1.elements[i]); 1376 Element srcElem2 = gtoh(srcReg2.elements[imm]); 1377 Element destElem; 1378 %(readDest)s 1379 %(op)s 1380 destReg.elements[i] = htog(destElem); 1381 } 1382 } 1383 ''' % { "op" : op, "readDest" : readDestCode } 1384 for reg in range(rCount): 1385 eWalkCode += ''' 1386 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1387 ''' % { "reg" : reg } 1388 iop = InstObjParams(name, Name, 1389 "RegRegRegImmOp", 1390 { "code": eWalkCode, 1391 "r_count": rCount, 1392 "predicate_test": predicateTest, 1393 "op_class": opClass }, []) 1394 header_output += NeonRegRegRegImmOpDeclare.subst(iop) 1395 exec_output += NeonEqualRegExecute.subst(iop) 1396 for type in types: 1397 substDict = { "targs" : type, 1398 "class_name" : Name } 1399 exec_output += NeonExecDeclare.subst(substDict) 1400 1401 def twoRegLongInst(name, Name, opClass, types, op, readDest=False): 1402 global header_output, exec_output 1403 rCount = 2 1404 eWalkCode = simdEnabledCheckCode + ''' 1405 RegVect srcReg1, srcReg2; 1406 BigRegVect destReg; 1407 ''' 1408 for reg in range(rCount): 1409 eWalkCode += ''' 1410 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1411 srcReg2.regs[%(reg)d] = htog(FpOp2P%(reg)d_uw);; 1412 ''' % { "reg" : reg } 1413 if readDest: 1414 for reg in range(2 * rCount): 1415 eWalkCode += ''' 1416 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1417 ''' % { "reg" : reg } 1418 readDestCode = '' 1419 if readDest: 1420 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1421 eWalkCode += ''' 1422 if (imm < 0 && imm >= eCount) { 1423 fault = new UndefinedInstruction(machInst, false, mnemonic); 1424 } else { 1425 for (unsigned i = 0; i < eCount; i++) { 1426 Element srcElem1 = gtoh(srcReg1.elements[i]); 1427 Element srcElem2 = gtoh(srcReg2.elements[imm]); 1428 BigElement destElem; 1429 %(readDest)s 1430 %(op)s 1431 destReg.elements[i] = htog(destElem); 1432 } 1433 } 1434 ''' % { "op" : op, "readDest" : readDestCode } 1435 for reg in range(2 * rCount): 1436 eWalkCode += ''' 1437 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1438 ''' % { "reg" : reg } 1439 iop = InstObjParams(name, Name, 1440 "RegRegRegImmOp", 1441 { "code": eWalkCode, 1442 "r_count": rCount, 1443 "predicate_test": predicateTest, 1444 "op_class": opClass }, []) 1445 header_output += NeonRegRegRegImmOpDeclare.subst(iop) 1446 exec_output += NeonUnequalRegExecute.subst(iop) 1447 for type in types: 1448 substDict = { "targs" : type, 1449 "class_name" : Name } 1450 exec_output += NeonExecDeclare.subst(substDict) 1451 1452 def twoEqualRegInstFp(name, Name, opClass, types, rCount, op, readDest=False): 1453 global header_output, exec_output 1454 eWalkCode = simdEnabledCheckCode + ''' 1455 typedef FloatReg FloatVect[rCount]; 1456 FloatVect srcRegs1, srcRegs2, destRegs; 1457 ''' 1458 for reg in range(rCount): 1459 eWalkCode += ''' 1460 srcRegs1[%(reg)d] = FpOp1P%(reg)d; 1461 srcRegs2[%(reg)d] = FpOp2P%(reg)d; 1462 ''' % { "reg" : reg } 1463 if readDest: 1464 eWalkCode += ''' 1465 destRegs[%(reg)d] = FpDestP%(reg)d; 1466 ''' % { "reg" : reg } 1467 readDestCode = '' 1468 if readDest: 1469 readDestCode = 'destReg = destRegs[i];' 1470 eWalkCode += ''' 1471 if (imm < 0 && imm >= eCount) { 1472 fault = new UndefinedInstruction(machInst, false, mnemonic); 1473 } else { 1474 for (unsigned i = 0; i < rCount; i++) { 1475 FloatReg srcReg1 = srcRegs1[i]; 1476 FloatReg srcReg2 = srcRegs2[imm]; 1477 FloatReg destReg; 1478 %(readDest)s 1479 %(op)s 1480 destRegs[i] = destReg; 1481 } 1482 } 1483 ''' % { "op" : op, "readDest" : readDestCode } 1484 for reg in range(rCount): 1485 eWalkCode += ''' 1486 FpDestP%(reg)d = destRegs[%(reg)d]; 1487 ''' % { "reg" : reg } 1488 iop = InstObjParams(name, Name, 1489 "FpRegRegRegImmOp", 1490 { "code": eWalkCode, 1491 "r_count": rCount, 1492 "predicate_test": predicateTest, 1493 "op_class": opClass }, []) 1494 header_output += NeonRegRegRegImmOpDeclare.subst(iop) 1495 exec_output += NeonEqualRegExecute.subst(iop) 1496 for type in types: 1497 substDict = { "targs" : type, 1498 "class_name" : Name } 1499 exec_output += NeonExecDeclare.subst(substDict) 1500 1501 def twoRegShiftInst(name, Name, opClass, types, rCount, op, 1502 readDest=False, toInt=False, fromInt=False): 1503 global header_output, exec_output 1504 eWalkCode = simdEnabledCheckCode + ''' 1505 RegVect srcRegs1, destRegs; 1506 ''' 1507 for reg in range(rCount): 1508 eWalkCode += ''' 1509 srcRegs1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1510 ''' % { "reg" : reg } 1511 if readDest: 1512 eWalkCode += ''' 1513 destRegs.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1514 ''' % { "reg" : reg } 1515 readDestCode = '' 1516 if readDest: 1517 readDestCode = 'destElem = gtoh(destRegs.elements[i]);' 1518 if toInt: 1519 readDestCode = 'destReg = gtoh(destRegs.regs[i]);' 1520 readOpCode = 'Element srcElem1 = gtoh(srcRegs1.elements[i]);' 1521 if fromInt: 1522 readOpCode = 'FloatRegBits srcReg1 = gtoh(srcRegs1.regs[i]);' 1523 declDest = 'Element destElem;' 1524 writeDestCode = 'destRegs.elements[i] = htog(destElem);' 1525 if toInt: 1526 declDest = 'FloatRegBits destReg;' 1527 writeDestCode = 'destRegs.regs[i] = htog(destReg);' 1528 eWalkCode += ''' 1529 for (unsigned i = 0; i < eCount; i++) { 1530 %(readOp)s 1531 %(declDest)s 1532 %(readDest)s 1533 %(op)s 1534 %(writeDest)s 1535 } 1536 ''' % { "readOp" : readOpCode, 1537 "declDest" : declDest, 1538 "readDest" : readDestCode, 1539 "op" : op, 1540 "writeDest" : writeDestCode } 1541 for reg in range(rCount): 1542 eWalkCode += ''' 1543 FpDestP%(reg)d_uw = gtoh(destRegs.regs[%(reg)d]); 1544 ''' % { "reg" : reg } 1545 iop = InstObjParams(name, Name, 1546 "RegRegImmOp", 1547 { "code": eWalkCode, 1548 "r_count": rCount, 1549 "predicate_test": predicateTest, 1550 "op_class": opClass }, []) 1551 header_output += NeonRegRegImmOpDeclare.subst(iop) 1552 exec_output += NeonEqualRegExecute.subst(iop) 1553 for type in types: 1554 substDict = { "targs" : type, 1555 "class_name" : Name } 1556 exec_output += NeonExecDeclare.subst(substDict) 1557 1558 def twoRegNarrowShiftInst(name, Name, opClass, types, op, readDest=False): 1559 global header_output, exec_output 1560 eWalkCode = simdEnabledCheckCode + ''' 1561 BigRegVect srcReg1; 1562 RegVect destReg; 1563 ''' 1564 for reg in range(4): 1565 eWalkCode += ''' 1566 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1567 ''' % { "reg" : reg } 1568 if readDest: 1569 for reg in range(2): 1570 eWalkCode += ''' 1571 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1572 ''' % { "reg" : reg } 1573 readDestCode = '' 1574 if readDest: 1575 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1576 eWalkCode += ''' 1577 for (unsigned i = 0; i < eCount; i++) { 1578 BigElement srcElem1 = gtoh(srcReg1.elements[i]); 1579 Element destElem; 1580 %(readDest)s 1581 %(op)s 1582 destReg.elements[i] = htog(destElem); 1583 } 1584 ''' % { "op" : op, "readDest" : readDestCode } 1585 for reg in range(2): 1586 eWalkCode += ''' 1587 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1588 ''' % { "reg" : reg } 1589 iop = InstObjParams(name, Name, 1590 "RegRegImmOp", 1591 { "code": eWalkCode, 1592 "r_count": 2, 1593 "predicate_test": predicateTest, 1594 "op_class": opClass }, []) 1595 header_output += NeonRegRegImmOpDeclare.subst(iop) 1596 exec_output += NeonUnequalRegExecute.subst(iop) 1597 for type in types: 1598 substDict = { "targs" : type, 1599 "class_name" : Name } 1600 exec_output += NeonExecDeclare.subst(substDict) 1601 1602 def twoRegLongShiftInst(name, Name, opClass, types, op, readDest=False): 1603 global header_output, exec_output 1604 eWalkCode = simdEnabledCheckCode + ''' 1605 RegVect srcReg1; 1606 BigRegVect destReg; 1607 ''' 1608 for reg in range(2): 1609 eWalkCode += ''' 1610 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1611 ''' % { "reg" : reg } 1612 if readDest: 1613 for reg in range(4): 1614 eWalkCode += ''' 1615 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1616 ''' % { "reg" : reg } 1617 readDestCode = '' 1618 if readDest: 1619 readDestCode = 'destReg = gtoh(destReg.elements[i]);' 1620 eWalkCode += ''' 1621 for (unsigned i = 0; i < eCount; i++) { 1622 Element srcElem1 = gtoh(srcReg1.elements[i]); 1623 BigElement destElem; 1624 %(readDest)s 1625 %(op)s 1626 destReg.elements[i] = htog(destElem); 1627 } 1628 ''' % { "op" : op, "readDest" : readDestCode } 1629 for reg in range(4): 1630 eWalkCode += ''' 1631 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1632 ''' % { "reg" : reg } 1633 iop = InstObjParams(name, Name, 1634 "RegRegImmOp", 1635 { "code": eWalkCode, 1636 "r_count": 2, 1637 "predicate_test": predicateTest, 1638 "op_class": opClass }, []) 1639 header_output += NeonRegRegImmOpDeclare.subst(iop) 1640 exec_output += NeonUnequalRegExecute.subst(iop) 1641 for type in types: 1642 substDict = { "targs" : type, 1643 "class_name" : Name } 1644 exec_output += NeonExecDeclare.subst(substDict) 1645 1646 def twoRegMiscInst(name, Name, opClass, types, rCount, op, readDest=False): 1647 global header_output, exec_output 1648 eWalkCode = simdEnabledCheckCode + ''' 1649 RegVect srcReg1, destReg; 1650 ''' 1651 for reg in range(rCount): 1652 eWalkCode += ''' 1653 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1654 ''' % { "reg" : reg } 1655 if readDest: 1656 eWalkCode += ''' 1657 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1658 ''' % { "reg" : reg } 1659 readDestCode = '' 1660 if readDest: 1661 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1662 eWalkCode += ''' 1663 for (unsigned i = 0; i < eCount; i++) { 1664 unsigned j = i; 1665 Element srcElem1 = gtoh(srcReg1.elements[i]); 1666 Element destElem; 1667 %(readDest)s 1668 %(op)s 1669 destReg.elements[j] = htog(destElem); 1670 } 1671 ''' % { "op" : op, "readDest" : readDestCode } 1672 for reg in range(rCount): 1673 eWalkCode += ''' 1674 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1675 ''' % { "reg" : reg } 1676 iop = InstObjParams(name, Name, 1677 "RegRegOp", 1678 { "code": eWalkCode, 1679 "r_count": rCount, 1680 "predicate_test": predicateTest, 1681 "op_class": opClass }, []) 1682 header_output += NeonRegRegOpDeclare.subst(iop) 1683 exec_output += NeonEqualRegExecute.subst(iop) 1684 for type in types: 1685 substDict = { "targs" : type, 1686 "class_name" : Name } 1687 exec_output += NeonExecDeclare.subst(substDict) 1688 1689 def twoRegMiscScInst(name, Name, opClass, types, rCount, op, readDest=False): 1690 global header_output, exec_output 1691 eWalkCode = simdEnabledCheckCode + ''' 1692 RegVect srcReg1, destReg; 1693 ''' 1694 for reg in range(rCount): 1695 eWalkCode += ''' 1696 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1697 ''' % { "reg" : reg } 1698 if readDest: 1699 eWalkCode += ''' 1700 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1701 ''' % { "reg" : reg } 1702 readDestCode = '' 1703 if readDest: 1704 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1705 eWalkCode += ''' 1706 for (unsigned i = 0; i < eCount; i++) { 1707 Element srcElem1 = gtoh(srcReg1.elements[imm]); 1708 Element destElem; 1709 %(readDest)s 1710 %(op)s 1711 destReg.elements[i] = htog(destElem); 1712 } 1713 ''' % { "op" : op, "readDest" : readDestCode } 1714 for reg in range(rCount): 1715 eWalkCode += ''' 1716 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1717 ''' % { "reg" : reg } 1718 iop = InstObjParams(name, Name, 1719 "RegRegImmOp", 1720 { "code": eWalkCode, 1721 "r_count": rCount, 1722 "predicate_test": predicateTest, 1723 "op_class": opClass }, []) 1724 header_output += NeonRegRegImmOpDeclare.subst(iop) 1725 exec_output += NeonEqualRegExecute.subst(iop) 1726 for type in types: 1727 substDict = { "targs" : type, 1728 "class_name" : Name } 1729 exec_output += NeonExecDeclare.subst(substDict) 1730 1731 def twoRegMiscScramble(name, Name, opClass, types, rCount, op, readDest=False): 1732 global header_output, exec_output 1733 eWalkCode = simdEnabledCheckCode + ''' 1734 RegVect srcReg1, destReg; 1735 ''' 1736 for reg in range(rCount): 1737 eWalkCode += ''' 1738 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1739 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1740 ''' % { "reg" : reg } 1741 if readDest: 1742 eWalkCode += ''' 1743 ''' % { "reg" : reg } 1744 readDestCode = '' 1745 if readDest: 1746 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1747 eWalkCode += op 1748 for reg in range(rCount): 1749 eWalkCode += ''' 1750 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1751 FpOp1P%(reg)d_uw = gtoh(srcReg1.regs[%(reg)d]); 1752 ''' % { "reg" : reg } 1753 iop = InstObjParams(name, Name, 1754 "RegRegOp", 1755 { "code": eWalkCode, 1756 "r_count": rCount, 1757 "predicate_test": predicateTest, 1758 "op_class": opClass }, []) 1759 header_output += NeonRegRegOpDeclare.subst(iop) 1760 exec_output += NeonEqualRegExecute.subst(iop) 1761 for type in types: 1762 substDict = { "targs" : type, 1763 "class_name" : Name } 1764 exec_output += NeonExecDeclare.subst(substDict) 1765 1766 def twoRegMiscInstFp(name, Name, opClass, types, rCount, op, 1767 readDest=False, toInt=False): 1768 global header_output, exec_output 1769 eWalkCode = simdEnabledCheckCode + ''' 1770 typedef FloatReg FloatVect[rCount]; 1771 FloatVect srcRegs1; 1772 ''' 1773 if toInt: 1774 eWalkCode += 'RegVect destRegs;\n' 1775 else: 1776 eWalkCode += 'FloatVect destRegs;\n' 1777 for reg in range(rCount): 1778 eWalkCode += ''' 1779 srcRegs1[%(reg)d] = FpOp1P%(reg)d; 1780 ''' % { "reg" : reg } 1781 if readDest: 1782 if toInt: 1783 eWalkCode += ''' 1784 destRegs.regs[%(reg)d] = FpDestP%(reg)d.bits; 1785 ''' % { "reg" : reg } 1786 else: 1787 eWalkCode += ''' 1788 destRegs[%(reg)d] = FpDestP%(reg)d; 1789 ''' % { "reg" : reg } 1790 readDestCode = '' 1791 if readDest: 1792 readDestCode = 'destReg = destRegs[i];' 1793 destType = 'FloatReg' 1794 writeDest = 'destRegs[r] = destReg;' 1795 if toInt: 1796 destType = 'FloatRegBits' 1797 writeDest = 'destRegs.regs[r] = destReg;' 1798 eWalkCode += ''' 1799 for (unsigned r = 0; r < rCount; r++) { 1800 FloatReg srcReg1 = srcRegs1[r]; 1801 %(destType)s destReg; 1802 %(readDest)s 1803 %(op)s 1804 %(writeDest)s 1805 } 1806 ''' % { "op" : op, 1807 "readDest" : readDestCode, 1808 "destType" : destType, 1809 "writeDest" : writeDest } 1810 for reg in range(rCount): 1811 if toInt: 1812 eWalkCode += ''' 1813 FpDestP%(reg)d_uw = destRegs.regs[%(reg)d]; 1814 ''' % { "reg" : reg } 1815 else: 1816 eWalkCode += ''' 1817 FpDestP%(reg)d = destRegs[%(reg)d]; 1818 ''' % { "reg" : reg } 1819 iop = InstObjParams(name, Name, 1820 "FpRegRegOp", 1821 { "code": eWalkCode, 1822 "r_count": rCount, 1823 "predicate_test": predicateTest, 1824 "op_class": opClass }, []) 1825 header_output += NeonRegRegOpDeclare.subst(iop) 1826 exec_output += NeonEqualRegExecute.subst(iop) 1827 for type in types: 1828 substDict = { "targs" : type, 1829 "class_name" : Name } 1830 exec_output += NeonExecDeclare.subst(substDict) 1831 1832 def twoRegCondenseInst(name, Name, opClass, types, rCount, op, readDest=False): 1833 global header_output, exec_output 1834 eWalkCode = simdEnabledCheckCode + ''' 1835 RegVect srcRegs; 1836 BigRegVect destReg; 1837 ''' 1838 for reg in range(rCount): 1839 eWalkCode += ''' 1840 srcRegs.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1841 ''' % { "reg" : reg } 1842 if readDest: 1843 eWalkCode += ''' 1844 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1845 ''' % { "reg" : reg } 1846 readDestCode = '' 1847 if readDest: 1848 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1849 eWalkCode += ''' 1850 for (unsigned i = 0; i < eCount / 2; i++) { 1851 Element srcElem1 = gtoh(srcRegs.elements[2 * i]); 1852 Element srcElem2 = gtoh(srcRegs.elements[2 * i + 1]); 1853 BigElement destElem; 1854 %(readDest)s 1855 %(op)s 1856 destReg.elements[i] = htog(destElem); 1857 } 1858 ''' % { "op" : op, "readDest" : readDestCode } 1859 for reg in range(rCount): 1860 eWalkCode += ''' 1861 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1862 ''' % { "reg" : reg } 1863 iop = InstObjParams(name, Name, 1864 "RegRegOp", 1865 { "code": eWalkCode, 1866 "r_count": rCount, 1867 "predicate_test": predicateTest, 1868 "op_class": opClass }, []) 1869 header_output += NeonRegRegOpDeclare.subst(iop) 1870 exec_output += NeonUnequalRegExecute.subst(iop) 1871 for type in types: 1872 substDict = { "targs" : type, 1873 "class_name" : Name } 1874 exec_output += NeonExecDeclare.subst(substDict) 1875 1876 def twoRegNarrowMiscInst(name, Name, opClass, types, op, readDest=False): 1877 global header_output, exec_output 1878 eWalkCode = simdEnabledCheckCode + ''' 1879 BigRegVect srcReg1; 1880 RegVect destReg; 1881 ''' 1882 for reg in range(4): 1883 eWalkCode += ''' 1884 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1885 ''' % { "reg" : reg } 1886 if readDest: 1887 for reg in range(2): 1888 eWalkCode += ''' 1889 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1890 ''' % { "reg" : reg } 1891 readDestCode = '' 1892 if readDest: 1893 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1894 eWalkCode += ''' 1895 for (unsigned i = 0; i < eCount; i++) { 1896 BigElement srcElem1 = gtoh(srcReg1.elements[i]); 1897 Element destElem; 1898 %(readDest)s 1899 %(op)s 1900 destReg.elements[i] = htog(destElem); 1901 } 1902 ''' % { "op" : op, "readDest" : readDestCode } 1903 for reg in range(2): 1904 eWalkCode += ''' 1905 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1906 ''' % { "reg" : reg } 1907 iop = InstObjParams(name, Name, 1908 "RegRegOp", 1909 { "code": eWalkCode, 1910 "r_count": 2, 1911 "predicate_test": predicateTest, 1912 "op_class": opClass }, []) 1913 header_output += NeonRegRegOpDeclare.subst(iop) 1914 exec_output += NeonUnequalRegExecute.subst(iop) 1915 for type in types: 1916 substDict = { "targs" : type, 1917 "class_name" : Name } 1918 exec_output += NeonExecDeclare.subst(substDict) 1919 1920 def oneRegImmInst(name, Name, opClass, types, rCount, op, readDest=False): 1921 global header_output, exec_output 1922 eWalkCode = simdEnabledCheckCode + ''' 1923 RegVect destReg; 1924 ''' 1925 if readDest: 1926 for reg in range(rCount): 1927 eWalkCode += ''' 1928 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1929 ''' % { "reg" : reg } 1930 readDestCode = '' 1931 if readDest: 1932 readDestCode = 'destElem = gtoh(destReg.elements[i]);' 1933 eWalkCode += ''' 1934 for (unsigned i = 0; i < eCount; i++) { 1935 Element destElem; 1936 %(readDest)s 1937 %(op)s 1938 destReg.elements[i] = htog(destElem); 1939 } 1940 ''' % { "op" : op, "readDest" : readDestCode } 1941 for reg in range(rCount): 1942 eWalkCode += ''' 1943 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1944 ''' % { "reg" : reg } 1945 iop = InstObjParams(name, Name, 1946 "RegImmOp", 1947 { "code": eWalkCode, 1948 "r_count": rCount, 1949 "predicate_test": predicateTest, 1950 "op_class": opClass }, []) 1951 header_output += NeonRegImmOpDeclare.subst(iop) 1952 exec_output += NeonEqualRegExecute.subst(iop) 1953 for type in types: 1954 substDict = { "targs" : type, 1955 "class_name" : Name } 1956 exec_output += NeonExecDeclare.subst(substDict) 1957 1958 def twoRegLongMiscInst(name, Name, opClass, types, op, readDest=False): 1959 global header_output, exec_output 1960 eWalkCode = simdEnabledCheckCode + ''' 1961 RegVect srcReg1; 1962 BigRegVect destReg; 1963 ''' 1964 for reg in range(2): 1965 eWalkCode += ''' 1966 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 1967 ''' % { "reg" : reg } 1968 if readDest: 1969 for reg in range(4): 1970 eWalkCode += ''' 1971 destReg.regs[%(reg)d] = htog(FpDestP%(reg)d_uw); 1972 ''' % { "reg" : reg } 1973 readDestCode = '' 1974 if readDest: 1975 readDestCode = 'destReg = gtoh(destReg.elements[i]);' 1976 eWalkCode += ''' 1977 for (unsigned i = 0; i < eCount; i++) { 1978 Element srcElem1 = gtoh(srcReg1.elements[i]); 1979 BigElement destElem; 1980 %(readDest)s 1981 %(op)s 1982 destReg.elements[i] = htog(destElem); 1983 } 1984 ''' % { "op" : op, "readDest" : readDestCode } 1985 for reg in range(4): 1986 eWalkCode += ''' 1987 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 1988 ''' % { "reg" : reg } 1989 iop = InstObjParams(name, Name, 1990 "RegRegOp", 1991 { "code": eWalkCode, 1992 "r_count": 2, 1993 "predicate_test": predicateTest, 1994 "op_class": opClass }, []) 1995 header_output += NeonRegRegOpDeclare.subst(iop) 1996 exec_output += NeonUnequalRegExecute.subst(iop) 1997 for type in types: 1998 substDict = { "targs" : type, 1999 "class_name" : Name } 2000 exec_output += NeonExecDeclare.subst(substDict) 2001 2002 vhaddCode = ''' 2003 Element carryBit = 2004 (((unsigned)srcElem1 & 0x1) + 2005 ((unsigned)srcElem2 & 0x1)) >> 1; 2006 // Use division instead of a shift to ensure the sign extension works 2007 // right. The compiler will figure out if it can be a shift. Mask the 2008 // inputs so they get truncated correctly. 2009 destElem = (((srcElem1 & ~(Element)1) / 2) + 2010 ((srcElem2 & ~(Element)1) / 2)) + carryBit; 2011 ''' 2012 threeEqualRegInst("vhadd", "VhaddD", "SimdAddOp", allTypes, 2, vhaddCode) 2013 threeEqualRegInst("vhadd", "VhaddQ", "SimdAddOp", allTypes, 4, vhaddCode) 2014 2015 vrhaddCode = ''' 2016 Element carryBit = 2017 (((unsigned)srcElem1 & 0x1) + 2018 ((unsigned)srcElem2 & 0x1) + 1) >> 1; 2019 // Use division instead of a shift to ensure the sign extension works 2020 // right. The compiler will figure out if it can be a shift. Mask the 2021 // inputs so they get truncated correctly. 2022 destElem = (((srcElem1 & ~(Element)1) / 2) + 2023 ((srcElem2 & ~(Element)1) / 2)) + carryBit; 2024 ''' 2025 threeEqualRegInst("vrhadd", "VrhaddD", "SimdAddOp", allTypes, 2, vrhaddCode) 2026 threeEqualRegInst("vrhadd", "VrhaddQ", "SimdAddOp", allTypes, 4, vrhaddCode) 2027 2028 vhsubCode = ''' 2029 Element barrowBit = 2030 (((srcElem1 & 0x1) - (srcElem2 & 0x1)) >> 1) & 0x1; 2031 // Use division instead of a shift to ensure the sign extension works 2032 // right. The compiler will figure out if it can be a shift. Mask the 2033 // inputs so they get truncated correctly. 2034 destElem = (((srcElem1 & ~(Element)1) / 2) - 2035 ((srcElem2 & ~(Element)1) / 2)) - barrowBit; 2036 ''' 2037 threeEqualRegInst("vhsub", "VhsubD", "SimdAddOp", allTypes, 2, vhsubCode) 2038 threeEqualRegInst("vhsub", "VhsubQ", "SimdAddOp", allTypes, 4, vhsubCode) 2039 2040 vandCode = ''' 2041 destElem = srcElem1 & srcElem2; 2042 ''' 2043 threeEqualRegInst("vand", "VandD", "SimdAluOp", unsignedTypes, 2, vandCode) 2044 threeEqualRegInst("vand", "VandQ", "SimdAluOp", unsignedTypes, 4, vandCode) 2045 2046 vbicCode = ''' 2047 destElem = srcElem1 & ~srcElem2; 2048 ''' 2049 threeEqualRegInst("vbic", "VbicD", "SimdAluOp", unsignedTypes, 2, vbicCode) 2050 threeEqualRegInst("vbic", "VbicQ", "SimdAluOp", unsignedTypes, 4, vbicCode) 2051 2052 vorrCode = ''' 2053 destElem = srcElem1 | srcElem2; 2054 ''' 2055 threeEqualRegInst("vorr", "VorrD", "SimdAluOp", unsignedTypes, 2, vorrCode) 2056 threeEqualRegInst("vorr", "VorrQ", "SimdAluOp", unsignedTypes, 4, vorrCode) 2057 2058 threeEqualRegInst("vmov", "VmovD", "SimdMiscOp", unsignedTypes, 2, vorrCode) 2059 threeEqualRegInst("vmov", "VmovQ", "SimdMiscOp", unsignedTypes, 4, vorrCode) 2060 2061 vornCode = ''' 2062 destElem = srcElem1 | ~srcElem2; 2063 ''' 2064 threeEqualRegInst("vorn", "VornD", "SimdAluOp", unsignedTypes, 2, vornCode) 2065 threeEqualRegInst("vorn", "VornQ", "SimdAluOp", unsignedTypes, 4, vornCode) 2066 2067 veorCode = ''' 2068 destElem = srcElem1 ^ srcElem2; 2069 ''' 2070 threeEqualRegInst("veor", "VeorD", "SimdAluOp", unsignedTypes, 2, veorCode) 2071 threeEqualRegInst("veor", "VeorQ", "SimdAluOp", unsignedTypes, 4, veorCode) 2072 2073 vbifCode = ''' 2074 destElem = (destElem & srcElem2) | (srcElem1 & ~srcElem2); 2075 ''' 2076 threeEqualRegInst("vbif", "VbifD", "SimdAluOp", unsignedTypes, 2, vbifCode, True) 2077 threeEqualRegInst("vbif", "VbifQ", "SimdAluOp", unsignedTypes, 4, vbifCode, True) 2078 vbitCode = ''' 2079 destElem = (srcElem1 & srcElem2) | (destElem & ~srcElem2); 2080 ''' 2081 threeEqualRegInst("vbit", "VbitD", "SimdAluOp", unsignedTypes, 2, vbitCode, True) 2082 threeEqualRegInst("vbit", "VbitQ", "SimdAluOp", unsignedTypes, 4, vbitCode, True) 2083 vbslCode = ''' 2084 destElem = (srcElem1 & destElem) | (srcElem2 & ~destElem); 2085 ''' 2086 threeEqualRegInst("vbsl", "VbslD", "SimdAluOp", unsignedTypes, 2, vbslCode, True) 2087 threeEqualRegInst("vbsl", "VbslQ", "SimdAluOp", unsignedTypes, 4, vbslCode, True) 2088 2089 vmaxCode = ''' 2090 destElem = (srcElem1 > srcElem2) ? srcElem1 : srcElem2; 2091 ''' 2092 threeEqualRegInst("vmax", "VmaxD", "SimdCmpOp", allTypes, 2, vmaxCode) 2093 threeEqualRegInst("vmax", "VmaxQ", "SimdCmpOp", allTypes, 4, vmaxCode) 2094 2095 vminCode = ''' 2096 destElem = (srcElem1 < srcElem2) ? srcElem1 : srcElem2; 2097 ''' 2098 threeEqualRegInst("vmin", "VminD", "SimdCmpOp", allTypes, 2, vminCode) 2099 threeEqualRegInst("vmin", "VminQ", "SimdCmpOp", allTypes, 4, vminCode) 2100 2101 vaddCode = ''' 2102 destElem = srcElem1 + srcElem2; 2103 ''' 2104 threeEqualRegInst("vadd", "NVaddD", "SimdAddOp", unsignedTypes, 2, vaddCode) 2105 threeEqualRegInst("vadd", "NVaddQ", "SimdAddOp", unsignedTypes, 4, vaddCode) 2106 2107 threeEqualRegInst("vpadd", "NVpaddD", "SimdAddOp", smallUnsignedTypes, 2108 2, vaddCode, pairwise=True) 2109 vaddlwCode = ''' 2110 destElem = (BigElement)srcElem1 + (BigElement)srcElem2; 2111 ''' 2112 threeRegLongInst("vaddl", "Vaddl", "SimdAddOp", smallTypes, vaddlwCode) 2113 threeRegWideInst("vaddw", "Vaddw", "SimdAddOp", smallTypes, vaddlwCode) 2114 vaddhnCode = ''' 2115 destElem = ((BigElement)srcElem1 + (BigElement)srcElem2) >> 2116 (sizeof(Element) * 8); 2117 ''' 2118 threeRegNarrowInst("vaddhn", "Vaddhn", "SimdAddOp", smallTypes, vaddhnCode) 2119 vraddhnCode = ''' 2120 destElem = ((BigElement)srcElem1 + (BigElement)srcElem2 + 2121 ((BigElement)1 << (sizeof(Element) * 8 - 1))) >> 2122 (sizeof(Element) * 8); 2123 ''' 2124 threeRegNarrowInst("vraddhn", "Vraddhn", "SimdAddOp", smallTypes, vraddhnCode) 2125 2126 vsubCode = ''' 2127 destElem = srcElem1 - srcElem2; 2128 ''' 2129 threeEqualRegInst("vsub", "NVsubD", "SimdAddOp", unsignedTypes, 2, vsubCode) 2130 threeEqualRegInst("vsub", "NVsubQ", "SimdAddOp", unsignedTypes, 4, vsubCode) 2131 vsublwCode = ''' 2132 destElem = (BigElement)srcElem1 - (BigElement)srcElem2; 2133 ''' 2134 threeRegLongInst("vsubl", "Vsubl", "SimdAddOp", smallTypes, vsublwCode) 2135 threeRegWideInst("vsubw", "Vsubw", "SimdAddOp", smallTypes, vsublwCode) 2136 2137 vqaddUCode = ''' 2138 destElem = srcElem1 + srcElem2; 2139 FPSCR fpscr = (FPSCR) FpscrQc; 2140 if (destElem < srcElem1 || destElem < srcElem2) { 2141 destElem = (Element)(-1); 2142 fpscr.qc = 1; 2143 } 2144 FpscrQc = fpscr; 2145 ''' 2146 threeEqualRegInst("vqadd", "VqaddUD", "SimdAddOp", unsignedTypes, 2, vqaddUCode) 2147 threeEqualRegInst("vqadd", "VqaddUQ", "SimdAddOp", unsignedTypes, 4, vqaddUCode) 2148 vsubhnCode = ''' 2149 destElem = ((BigElement)srcElem1 - (BigElement)srcElem2) >> 2150 (sizeof(Element) * 8); 2151 ''' 2152 threeRegNarrowInst("vsubhn", "Vsubhn", "SimdAddOp", smallTypes, vsubhnCode) 2153 vrsubhnCode = ''' 2154 destElem = ((BigElement)srcElem1 - (BigElement)srcElem2 + 2155 ((BigElement)1 << (sizeof(Element) * 8 - 1))) >> 2156 (sizeof(Element) * 8); 2157 ''' 2158 threeRegNarrowInst("vrsubhn", "Vrsubhn", "SimdAddOp", smallTypes, vrsubhnCode) 2159 2160 vqaddSCode = ''' 2161 destElem = srcElem1 + srcElem2; 2162 FPSCR fpscr = (FPSCR) FpscrQc; 2163 bool negDest = (destElem < 0); 2164 bool negSrc1 = (srcElem1 < 0); 2165 bool negSrc2 = (srcElem2 < 0); 2166 if ((negDest != negSrc1) && (negSrc1 == negSrc2)) { 2167 destElem = (Element)1 << (sizeof(Element) * 8 - 1); 2168 if (negDest) 2169 destElem -= 1; 2170 fpscr.qc = 1; 2171 } 2172 FpscrQc = fpscr; 2173 ''' 2174 threeEqualRegInst("vqadd", "VqaddSD", "SimdAddOp", signedTypes, 2, vqaddSCode) 2175 threeEqualRegInst("vqadd", "VqaddSQ", "SimdAddOp", signedTypes, 4, vqaddSCode) 2176 2177 vqsubUCode = ''' 2178 destElem = srcElem1 - srcElem2; 2179 FPSCR fpscr = (FPSCR) FpscrQc; 2180 if (destElem > srcElem1) { 2181 destElem = 0; 2182 fpscr.qc = 1; 2183 } 2184 FpscrQc = fpscr; 2185 ''' 2186 threeEqualRegInst("vqsub", "VqsubUD", "SimdAddOp", unsignedTypes, 2, vqsubUCode) 2187 threeEqualRegInst("vqsub", "VqsubUQ", "SimdAddOp", unsignedTypes, 4, vqsubUCode) 2188 2189 vqsubSCode = ''' 2190 destElem = srcElem1 - srcElem2; 2191 FPSCR fpscr = (FPSCR) FpscrQc; 2192 bool negDest = (destElem < 0); 2193 bool negSrc1 = (srcElem1 < 0); 2194 bool posSrc2 = (srcElem2 >= 0); 2195 if ((negDest != negSrc1) && (negSrc1 == posSrc2)) { 2196 destElem = (Element)1 << (sizeof(Element) * 8 - 1); 2197 if (negDest) 2198 destElem -= 1; 2199 fpscr.qc = 1; 2200 } 2201 FpscrQc = fpscr; 2202 ''' 2203 threeEqualRegInst("vqsub", "VqsubSD", "SimdAddOp", signedTypes, 2, vqsubSCode) 2204 threeEqualRegInst("vqsub", "VqsubSQ", "SimdAddOp", signedTypes, 4, vqsubSCode) 2205 2206 vcgtCode = ''' 2207 destElem = (srcElem1 > srcElem2) ? (Element)(-1) : 0; 2208 ''' 2209 threeEqualRegInst("vcgt", "VcgtD", "SimdCmpOp", allTypes, 2, vcgtCode) 2210 threeEqualRegInst("vcgt", "VcgtQ", "SimdCmpOp", allTypes, 4, vcgtCode) 2211 2212 vcgeCode = ''' 2213 destElem = (srcElem1 >= srcElem2) ? (Element)(-1) : 0; 2214 ''' 2215 threeEqualRegInst("vcge", "VcgeD", "SimdCmpOp", allTypes, 2, vcgeCode) 2216 threeEqualRegInst("vcge", "VcgeQ", "SimdCmpOp", allTypes, 4, vcgeCode) 2217 2218 vceqCode = ''' 2219 destElem = (srcElem1 == srcElem2) ? (Element)(-1) : 0; 2220 ''' 2221 threeEqualRegInst("vceq", "VceqD", "SimdCmpOp", unsignedTypes, 2, vceqCode) 2222 threeEqualRegInst("vceq", "VceqQ", "SimdCmpOp", unsignedTypes, 4, vceqCode) 2223 2224 vshlCode = ''' 2225 int16_t shiftAmt = (int8_t)srcElem2; 2226 if (shiftAmt < 0) { 2227 shiftAmt = -shiftAmt; 2228 if (shiftAmt >= sizeof(Element) * 8) { 2229 shiftAmt = sizeof(Element) * 8 - 1; 2230 destElem = 0; 2231 } else { 2232 destElem = (srcElem1 >> shiftAmt); 2233 } 2234 // Make sure the right shift sign extended when it should. 2235 if (ltz(srcElem1) && !ltz(destElem)) { 2236 destElem |= -((Element)1 << (sizeof(Element) * 8 - 2237 1 - shiftAmt)); 2238 } 2239 } else { 2240 if (shiftAmt >= sizeof(Element) * 8) { 2241 destElem = 0; 2242 } else { 2243 destElem = srcElem1 << shiftAmt; 2244 } 2245 } 2246 ''' 2247 threeEqualRegInst("vshl", "VshlD", "SimdShiftOp", allTypes, 2, vshlCode) 2248 threeEqualRegInst("vshl", "VshlQ", "SimdShiftOp", allTypes, 4, vshlCode) 2249 2250 vrshlCode = ''' 2251 int16_t shiftAmt = (int8_t)srcElem2; 2252 if (shiftAmt < 0) { 2253 shiftAmt = -shiftAmt; 2254 Element rBit = 0; 2255 if (shiftAmt <= sizeof(Element) * 8) 2256 rBit = bits(srcElem1, shiftAmt - 1); 2257 if (shiftAmt > sizeof(Element) * 8 && ltz(srcElem1)) 2258 rBit = 1; 2259 if (shiftAmt >= sizeof(Element) * 8) { 2260 shiftAmt = sizeof(Element) * 8 - 1; 2261 destElem = 0; 2262 } else { 2263 destElem = (srcElem1 >> shiftAmt); 2264 } 2265 // Make sure the right shift sign extended when it should. 2266 if (ltz(srcElem1) && !ltz(destElem)) { 2267 destElem |= -((Element)1 << (sizeof(Element) * 8 - 2268 1 - shiftAmt)); 2269 } 2270 destElem += rBit; 2271 } else if (shiftAmt > 0) { 2272 if (shiftAmt >= sizeof(Element) * 8) { 2273 destElem = 0; 2274 } else { 2275 destElem = srcElem1 << shiftAmt; 2276 } 2277 } else { 2278 destElem = srcElem1; 2279 } 2280 ''' 2281 threeEqualRegInst("vrshl", "VrshlD", "SimdAluOp", allTypes, 2, vrshlCode) 2282 threeEqualRegInst("vrshl", "VrshlQ", "SimdAluOp", allTypes, 4, vrshlCode) 2283 2284 vqshlUCode = ''' 2285 int16_t shiftAmt = (int8_t)srcElem2; 2286 FPSCR fpscr = (FPSCR) FpscrQc; 2287 if (shiftAmt < 0) { 2288 shiftAmt = -shiftAmt; 2289 if (shiftAmt >= sizeof(Element) * 8) { 2290 shiftAmt = sizeof(Element) * 8 - 1; 2291 destElem = 0; 2292 } else { 2293 destElem = (srcElem1 >> shiftAmt); 2294 } 2295 } else if (shiftAmt > 0) { 2296 if (shiftAmt >= sizeof(Element) * 8) { 2297 if (srcElem1 != 0) { 2298 destElem = mask(sizeof(Element) * 8); 2299 fpscr.qc = 1; 2300 } else { 2301 destElem = 0; 2302 } 2303 } else { 2304 if (bits(srcElem1, sizeof(Element) * 8 - 1, 2305 sizeof(Element) * 8 - shiftAmt)) { 2306 destElem = mask(sizeof(Element) * 8); 2307 fpscr.qc = 1; 2308 } else { 2309 destElem = srcElem1 << shiftAmt; 2310 } 2311 } 2312 } else { 2313 destElem = srcElem1; 2314 } 2315 FpscrQc = fpscr; 2316 ''' 2317 threeEqualRegInst("vqshl", "VqshlUD", "SimdAluOp", unsignedTypes, 2, vqshlUCode) 2318 threeEqualRegInst("vqshl", "VqshlUQ", "SimdAluOp", unsignedTypes, 4, vqshlUCode) 2319 2320 vqshlSCode = ''' 2321 int16_t shiftAmt = (int8_t)srcElem2; 2322 FPSCR fpscr = (FPSCR) FpscrQc; 2323 if (shiftAmt < 0) { 2324 shiftAmt = -shiftAmt; 2325 if (shiftAmt >= sizeof(Element) * 8) { 2326 shiftAmt = sizeof(Element) * 8 - 1; 2327 destElem = 0; 2328 } else { 2329 destElem = (srcElem1 >> shiftAmt); 2330 } 2331 // Make sure the right shift sign extended when it should. 2332 if (srcElem1 < 0 && destElem >= 0) { 2333 destElem |= -((Element)1 << (sizeof(Element) * 8 - 2334 1 - shiftAmt)); 2335 } 2336 } else if (shiftAmt > 0) { 2337 bool sat = false; 2338 if (shiftAmt >= sizeof(Element) * 8) { 2339 if (srcElem1 != 0) 2340 sat = true; 2341 else 2342 destElem = 0; 2343 } else { 2344 if (bits(srcElem1, sizeof(Element) * 8 - 1, 2345 sizeof(Element) * 8 - 1 - shiftAmt) != 2346 ((srcElem1 < 0) ? mask(shiftAmt + 1) : 0)) { 2347 sat = true; 2348 } else { 2349 destElem = srcElem1 << shiftAmt; 2350 } 2351 } 2352 if (sat) { 2353 fpscr.qc = 1; 2354 destElem = mask(sizeof(Element) * 8 - 1); 2355 if (srcElem1 < 0) 2356 destElem = ~destElem; 2357 } 2358 } else { 2359 destElem = srcElem1; 2360 } 2361 FpscrQc = fpscr; 2362 ''' 2363 threeEqualRegInst("vqshl", "VqshlSD", "SimdCmpOp", signedTypes, 2, vqshlSCode) 2364 threeEqualRegInst("vqshl", "VqshlSQ", "SimdCmpOp", signedTypes, 4, vqshlSCode) 2365 2366 vqrshlUCode = ''' 2367 int16_t shiftAmt = (int8_t)srcElem2; 2368 FPSCR fpscr = (FPSCR) FpscrQc; 2369 if (shiftAmt < 0) { 2370 shiftAmt = -shiftAmt; 2371 Element rBit = 0; 2372 if (shiftAmt <= sizeof(Element) * 8) 2373 rBit = bits(srcElem1, shiftAmt - 1); 2374 if (shiftAmt >= sizeof(Element) * 8) { 2375 shiftAmt = sizeof(Element) * 8 - 1; 2376 destElem = 0; 2377 } else { 2378 destElem = (srcElem1 >> shiftAmt); 2379 } 2380 destElem += rBit; 2381 } else { 2382 if (shiftAmt >= sizeof(Element) * 8) { 2383 if (srcElem1 != 0) { 2384 destElem = mask(sizeof(Element) * 8); 2385 fpscr.qc = 1; 2386 } else { 2387 destElem = 0; 2388 } 2389 } else { 2390 if (bits(srcElem1, sizeof(Element) * 8 - 1, 2391 sizeof(Element) * 8 - shiftAmt)) { 2392 destElem = mask(sizeof(Element) * 8); 2393 fpscr.qc = 1; 2394 } else { 2395 destElem = srcElem1 << shiftAmt; 2396 } 2397 } 2398 } 2399 FpscrQc = fpscr; 2400 ''' 2401 threeEqualRegInst("vqrshl", "VqrshlUD", "SimdCmpOp", unsignedTypes, 2, vqrshlUCode) 2402 threeEqualRegInst("vqrshl", "VqrshlUQ", "SimdCmpOp", unsignedTypes, 4, vqrshlUCode) 2403 2404 vqrshlSCode = ''' 2405 int16_t shiftAmt = (int8_t)srcElem2; 2406 FPSCR fpscr = (FPSCR) FpscrQc; 2407 if (shiftAmt < 0) { 2408 shiftAmt = -shiftAmt; 2409 Element rBit = 0; 2410 if (shiftAmt <= sizeof(Element) * 8) 2411 rBit = bits(srcElem1, shiftAmt - 1); 2412 if (shiftAmt > sizeof(Element) * 8 && srcElem1 < 0) 2413 rBit = 1; 2414 if (shiftAmt >= sizeof(Element) * 8) { 2415 shiftAmt = sizeof(Element) * 8 - 1; 2416 destElem = 0; 2417 } else { 2418 destElem = (srcElem1 >> shiftAmt); 2419 } 2420 // Make sure the right shift sign extended when it should. 2421 if (srcElem1 < 0 && destElem >= 0) { 2422 destElem |= -((Element)1 << (sizeof(Element) * 8 - 2423 1 - shiftAmt)); 2424 } 2425 destElem += rBit; 2426 } else if (shiftAmt > 0) { 2427 bool sat = false; 2428 if (shiftAmt >= sizeof(Element) * 8) { 2429 if (srcElem1 != 0) 2430 sat = true; 2431 else 2432 destElem = 0; 2433 } else { 2434 if (bits(srcElem1, sizeof(Element) * 8 - 1, 2435 sizeof(Element) * 8 - 1 - shiftAmt) != 2436 ((srcElem1 < 0) ? mask(shiftAmt + 1) : 0)) { 2437 sat = true; 2438 } else { 2439 destElem = srcElem1 << shiftAmt; 2440 } 2441 } 2442 if (sat) { 2443 fpscr.qc = 1; 2444 destElem = mask(sizeof(Element) * 8 - 1); 2445 if (srcElem1 < 0) 2446 destElem = ~destElem; 2447 } 2448 } else { 2449 destElem = srcElem1; 2450 } 2451 FpscrQc = fpscr; 2452 ''' 2453 threeEqualRegInst("vqrshl", "VqrshlSD", "SimdCmpOp", signedTypes, 2, vqrshlSCode) 2454 threeEqualRegInst("vqrshl", "VqrshlSQ", "SimdCmpOp", signedTypes, 4, vqrshlSCode) 2455 2456 vabaCode = ''' 2457 destElem += (srcElem1 > srcElem2) ? (srcElem1 - srcElem2) : 2458 (srcElem2 - srcElem1); 2459 ''' 2460 threeEqualRegInst("vaba", "VabaD", "SimdAddAccOp", allTypes, 2, vabaCode, True) 2461 threeEqualRegInst("vaba", "VabaQ", "SimdAddAccOp", allTypes, 4, vabaCode, True) 2462 vabalCode = ''' 2463 destElem += (srcElem1 > srcElem2) ? 2464 ((BigElement)srcElem1 - (BigElement)srcElem2) : 2465 ((BigElement)srcElem2 - (BigElement)srcElem1); 2466 ''' 2467 threeRegLongInst("vabal", "Vabal", "SimdAddAccOp", smallTypes, vabalCode, True) 2468 2469 vabdCode = ''' 2470 destElem = (srcElem1 > srcElem2) ? (srcElem1 - srcElem2) : 2471 (srcElem2 - srcElem1); 2472 ''' 2473 threeEqualRegInst("vabd", "VabdD", "SimdAddOp", allTypes, 2, vabdCode) 2474 threeEqualRegInst("vabd", "VabdQ", "SimdAddOp", allTypes, 4, vabdCode) 2475 vabdlCode = ''' 2476 destElem = (srcElem1 > srcElem2) ? 2477 ((BigElement)srcElem1 - (BigElement)srcElem2) : 2478 ((BigElement)srcElem2 - (BigElement)srcElem1); 2479 ''' 2480 threeRegLongInst("vabdl", "Vabdl", "SimdAddOp", smallTypes, vabdlCode) 2481 2482 vtstCode = ''' 2483 destElem = (srcElem1 & srcElem2) ? (Element)(-1) : 0; 2484 ''' 2485 threeEqualRegInst("vtst", "VtstD", "SimdAluOp", unsignedTypes, 2, vtstCode) 2486 threeEqualRegInst("vtst", "VtstQ", "SimdAluOp", unsignedTypes, 4, vtstCode) 2487 2488 vmulCode = ''' 2489 destElem = srcElem1 * srcElem2; 2490 ''' 2491 threeEqualRegInst("vmul", "NVmulD", "SimdMultOp", allTypes, 2, vmulCode) 2492 threeEqualRegInst("vmul", "NVmulQ", "SimdMultOp", allTypes, 4, vmulCode) 2493 vmullCode = ''' 2494 destElem = (BigElement)srcElem1 * (BigElement)srcElem2; 2495 ''' 2496 threeRegLongInst("vmull", "Vmull", "SimdMultOp", smallTypes, vmullCode) 2497 2498 vmlaCode = ''' 2499 destElem = destElem + srcElem1 * srcElem2; 2500 ''' 2501 threeEqualRegInst("vmla", "NVmlaD", "SimdMultAccOp", allTypes, 2, vmlaCode, True) 2502 threeEqualRegInst("vmla", "NVmlaQ", "SimdMultAccOp", allTypes, 4, vmlaCode, True) 2503 vmlalCode = ''' 2504 destElem = destElem + (BigElement)srcElem1 * (BigElement)srcElem2; 2505 ''' 2506 threeRegLongInst("vmlal", "Vmlal", "SimdMultAccOp", smallTypes, vmlalCode, True) 2507 2508 vqdmlalCode = ''' 2509 FPSCR fpscr = (FPSCR) FpscrQc; 2510 BigElement midElem = (2 * (int64_t)srcElem1 * (int64_t)srcElem2); 2511 Element maxNeg = (Element)1 << (sizeof(Element) * 8 - 1); 2512 Element halfNeg = maxNeg / 2; 2513 if ((srcElem1 == maxNeg && srcElem2 == maxNeg) || 2514 (srcElem1 == halfNeg && srcElem2 == maxNeg) || 2515 (srcElem1 == maxNeg && srcElem2 == halfNeg)) { 2516 midElem = ~((BigElement)maxNeg << (sizeof(Element) * 8)); 2517 fpscr.qc = 1; 2518 } 2519 bool negPreDest = ltz(destElem); 2520 destElem += midElem; 2521 bool negDest = ltz(destElem); 2522 bool negMid = ltz(midElem); 2523 if (negPreDest == negMid && negMid != negDest) { 2524 destElem = mask(sizeof(BigElement) * 8 - 1); 2525 if (negPreDest) 2526 destElem = ~destElem; 2527 fpscr.qc = 1; 2528 } 2529 FpscrQc = fpscr; 2530 ''' 2531 threeRegLongInst("vqdmlal", "Vqdmlal", "SimdMultAccOp", smallTypes, vqdmlalCode, True) 2532 2533 vqdmlslCode = ''' 2534 FPSCR fpscr = (FPSCR) FpscrQc; 2535 BigElement midElem = (2 * (int64_t)srcElem1 * (int64_t)srcElem2); 2536 Element maxNeg = (Element)1 << (sizeof(Element) * 8 - 1); 2537 Element halfNeg = maxNeg / 2; 2538 if ((srcElem1 == maxNeg && srcElem2 == maxNeg) || 2539 (srcElem1 == halfNeg && srcElem2 == maxNeg) || 2540 (srcElem1 == maxNeg && srcElem2 == halfNeg)) { 2541 midElem = ~((BigElement)maxNeg << (sizeof(Element) * 8)); 2542 fpscr.qc = 1; 2543 } 2544 bool negPreDest = ltz(destElem); 2545 destElem -= midElem; 2546 bool negDest = ltz(destElem); 2547 bool posMid = ltz((BigElement)-midElem); 2548 if (negPreDest == posMid && posMid != negDest) { 2549 destElem = mask(sizeof(BigElement) * 8 - 1); 2550 if (negPreDest) 2551 destElem = ~destElem; 2552 fpscr.qc = 1; 2553 } 2554 FpscrQc = fpscr; 2555 ''' 2556 threeRegLongInst("vqdmlsl", "Vqdmlsl", "SimdMultAccOp", smallTypes, vqdmlslCode, True) 2557 2558 vqdmullCode = ''' 2559 FPSCR fpscr = (FPSCR) FpscrQc; 2560 destElem = (2 * (int64_t)srcElem1 * (int64_t)srcElem2); 2561 if (srcElem1 == srcElem2 && 2562 srcElem1 == (Element)((Element)1 << 2563 (Element)(sizeof(Element) * 8 - 1))) { 2564 destElem = ~((BigElement)srcElem1 << (sizeof(Element) * 8)); 2565 fpscr.qc = 1; 2566 } 2567 FpscrQc = fpscr; 2568 ''' 2569 threeRegLongInst("vqdmull", "Vqdmull", "SimdMultAccOp", smallTypes, vqdmullCode) 2570 2571 vmlsCode = ''' 2572 destElem = destElem - srcElem1 * srcElem2; 2573 ''' 2574 threeEqualRegInst("vmls", "NVmlsD", "SimdMultAccOp", allTypes, 2, vmlsCode, True) 2575 threeEqualRegInst("vmls", "NVmlsQ", "SimdMultAccOp", allTypes, 4, vmlsCode, True) 2576 vmlslCode = ''' 2577 destElem = destElem - (BigElement)srcElem1 * (BigElement)srcElem2; 2578 ''' 2579 threeRegLongInst("vmlsl", "Vmlsl", "SimdMultAccOp", smallTypes, vmlslCode, True) 2580 2581 vmulpCode = ''' 2582 destElem = 0; 2583 for (unsigned j = 0; j < sizeof(Element) * 8; j++) { 2584 if (bits(srcElem2, j)) 2585 destElem ^= srcElem1 << j; 2586 } 2587 ''' 2588 threeEqualRegInst("vmul", "NVmulpD", "SimdMultOp", unsignedTypes, 2, vmulpCode) 2589 threeEqualRegInst("vmul", "NVmulpQ", "SimdMultOp", unsignedTypes, 4, vmulpCode) 2590 vmullpCode = ''' 2591 destElem = 0; 2592 for (unsigned j = 0; j < sizeof(Element) * 8; j++) { 2593 if (bits(srcElem2, j)) 2594 destElem ^= (BigElement)srcElem1 << j; 2595 } 2596 ''' 2597 threeRegLongInst("vmull", "Vmullp", "SimdMultOp", smallUnsignedTypes, vmullpCode) 2598 2599 threeEqualRegInst("vpmax", "VpmaxD", "SimdCmpOp", smallTypes, 2, vmaxCode, pairwise=True) 2600 2601 threeEqualRegInst("vpmin", "VpminD", "SimdCmpOp", smallTypes, 2, vminCode, pairwise=True) 2602 2603 vqdmulhCode = ''' 2604 FPSCR fpscr = (FPSCR) FpscrQc; 2605 destElem = (2 * (int64_t)srcElem1 * (int64_t)srcElem2) >> 2606 (sizeof(Element) * 8); 2607 if (srcElem1 == srcElem2 && 2608 srcElem1 == (Element)((Element)1 << 2609 (sizeof(Element) * 8 - 1))) { 2610 destElem = ~srcElem1; 2611 fpscr.qc = 1; 2612 } 2613 FpscrQc = fpscr; 2614 ''' 2615 threeEqualRegInst("vqdmulh", "VqdmulhD", "SimdMultOp", smallSignedTypes, 2, vqdmulhCode) 2616 threeEqualRegInst("vqdmulh", "VqdmulhQ", "SimdMultOp", smallSignedTypes, 4, vqdmulhCode) 2617 2618 vqrdmulhCode = ''' 2619 FPSCR fpscr = (FPSCR) FpscrQc; 2620 destElem = (2 * (int64_t)srcElem1 * (int64_t)srcElem2 + 2621 ((int64_t)1 << (sizeof(Element) * 8 - 1))) >> 2622 (sizeof(Element) * 8); 2623 Element maxNeg = (Element)1 << (sizeof(Element) * 8 - 1); 2624 Element halfNeg = maxNeg / 2; 2625 if ((srcElem1 == maxNeg && srcElem2 == maxNeg) || 2626 (srcElem1 == halfNeg && srcElem2 == maxNeg) || 2627 (srcElem1 == maxNeg && srcElem2 == halfNeg)) { 2628 if (destElem < 0) { 2629 destElem = mask(sizeof(Element) * 8 - 1); 2630 } else { 2631 destElem = (Element)1 << (sizeof(Element) * 8 - 1); 2632 } 2633 fpscr.qc = 1; 2634 } 2635 FpscrQc = fpscr; 2636 ''' 2637 threeEqualRegInst("vqrdmulh", "VqrdmulhD", 2638 "SimdMultOp", smallSignedTypes, 2, vqrdmulhCode) 2639 threeEqualRegInst("vqrdmulh", "VqrdmulhQ", 2640 "SimdMultOp", smallSignedTypes, 4, vqrdmulhCode) 2641 2642 vmaxfpCode = ''' 2643 FPSCR fpscr = (FPSCR) FpscrExc; 2644 bool done; 2645 destReg = processNans(fpscr, done, true, srcReg1, srcReg2); 2646 if (!done) { 2647 destReg = binaryOp(fpscr, srcReg1, srcReg2, fpMax<float>, 2648 true, true, VfpRoundNearest); 2649 } else if (flushToZero(srcReg1, srcReg2)) { 2650 fpscr.idc = 1; 2651 } 2652 FpscrExc = fpscr; 2653 ''' 2654 threeEqualRegInstFp("vmax", "VmaxDFp", "SimdFloatCmpOp", ("float",), 2, vmaxfpCode) 2655 threeEqualRegInstFp("vmax", "VmaxQFp", "SimdFloatCmpOp", ("float",), 4, vmaxfpCode) 2656 2657 vminfpCode = ''' 2658 FPSCR fpscr = (FPSCR) FpscrExc; 2659 bool done; 2660 destReg = processNans(fpscr, done, true, srcReg1, srcReg2); 2661 if (!done) { 2662 destReg = binaryOp(fpscr, srcReg1, srcReg2, fpMin<float>, 2663 true, true, VfpRoundNearest); 2664 } else if (flushToZero(srcReg1, srcReg2)) { 2665 fpscr.idc = 1; 2666 } 2667 FpscrExc = fpscr; 2668 ''' 2669 threeEqualRegInstFp("vmin", "VminDFp", "SimdFloatCmpOp", ("float",), 2, vminfpCode) 2670 threeEqualRegInstFp("vmin", "VminQFp", "SimdFloatCmpOp", ("float",), 4, vminfpCode) 2671 2672 threeEqualRegInstFp("vpmax", "VpmaxDFp", "SimdFloatCmpOp", ("float",), 2673 2, vmaxfpCode, pairwise=True) 2674 threeEqualRegInstFp("vpmax", "VpmaxQFp", "SimdFloatCmpOp", ("float",), 2675 4, vmaxfpCode, pairwise=True) 2676 2677 threeEqualRegInstFp("vpmin", "VpminDFp", "SimdFloatCmpOp", ("float",), 2678 2, vminfpCode, pairwise=True) 2679 threeEqualRegInstFp("vpmin", "VpminQFp", "SimdFloatCmpOp", ("float",), 2680 4, vminfpCode, pairwise=True) 2681 2682 vaddfpCode = ''' 2683 FPSCR fpscr = (FPSCR) FpscrExc; 2684 destReg = binaryOp(fpscr, srcReg1, srcReg2, fpAddS, 2685 true, true, VfpRoundNearest); 2686 FpscrExc = fpscr; 2687 ''' 2688 threeEqualRegInstFp("vadd", "VaddDFp", "SimdFloatAddOp", ("float",), 2, vaddfpCode) 2689 threeEqualRegInstFp("vadd", "VaddQFp", "SimdFloatAddOp", ("float",), 4, vaddfpCode) 2690 2691 threeEqualRegInstFp("vpadd", "VpaddDFp", "SimdFloatAddOp", ("float",), 2692 2, vaddfpCode, pairwise=True) 2693 threeEqualRegInstFp("vpadd", "VpaddQFp", "SimdFloatAddOp", ("float",), 2694 4, vaddfpCode, pairwise=True) 2695 2696 vsubfpCode = ''' 2697 FPSCR fpscr = (FPSCR) FpscrExc; 2698 destReg = binaryOp(fpscr, srcReg1, srcReg2, fpSubS, 2699 true, true, VfpRoundNearest); 2700 FpscrExc = fpscr; 2701 ''' 2702 threeEqualRegInstFp("vsub", "VsubDFp", "SimdFloatAddOp", ("float",), 2, vsubfpCode) 2703 threeEqualRegInstFp("vsub", "VsubQFp", "SimdFloatAddOp", ("float",), 4, vsubfpCode) 2704 2705 vmulfpCode = ''' 2706 FPSCR fpscr = (FPSCR) FpscrExc; 2707 destReg = binaryOp(fpscr, srcReg1, srcReg2, fpMulS, 2708 true, true, VfpRoundNearest); 2709 FpscrExc = fpscr; 2710 ''' 2711 threeEqualRegInstFp("vmul", "NVmulDFp", "SimdFloatMultOp", ("float",), 2, vmulfpCode) 2712 threeEqualRegInstFp("vmul", "NVmulQFp", "SimdFloatMultOp", ("float",), 4, vmulfpCode) 2713 2714 vmlafpCode = ''' 2715 FPSCR fpscr = (FPSCR) FpscrExc; 2716 float mid = binaryOp(fpscr, srcReg1, srcReg2, fpMulS, 2717 true, true, VfpRoundNearest); 2718 destReg = binaryOp(fpscr, mid, destReg, fpAddS, 2719 true, true, VfpRoundNearest); 2720 FpscrExc = fpscr; 2721 ''' 2722 threeEqualRegInstFp("vmla", "NVmlaDFp", "SimdFloatMultAccOp", ("float",), 2, vmlafpCode, True) 2723 threeEqualRegInstFp("vmla", "NVmlaQFp", "SimdFloatMultAccOp", ("float",), 4, vmlafpCode, True) 2724 2725 vfmafpCode = ''' 2726 FPSCR fpscr = (FPSCR) FpscrExc; 2727 destReg = ternaryOp(fpscr, srcReg1, srcReg2, destReg, fpMulAdd<float>, 2728 true, true, VfpRoundNearest); 2729 FpscrExc = fpscr; 2730 ''' 2731 threeEqualRegInstFp("vfma", "NVfmaDFp", "SimdFloatMultAccOp", ("float",), 2, vfmafpCode, True) 2732 threeEqualRegInstFp("vfma", "NVfmaQFp", "SimdFloatMultAccOp", ("float",), 4, vfmafpCode, True) 2733 2734 vfmsfpCode = ''' 2735 FPSCR fpscr = (FPSCR) FpscrExc; 2736 destReg = ternaryOp(fpscr, -srcReg1, srcReg2, destReg, fpMulAdd<float>, 2737 true, true, VfpRoundNearest); 2738 FpscrExc = fpscr; 2739 ''' 2740 threeEqualRegInstFp("vfms", "NVfmsDFp", "SimdFloatMultAccOp", ("float",), 2, vfmsfpCode, True) 2741 threeEqualRegInstFp("vfms", "NVfmsQFp", "SimdFloatMultAccOp", ("float",), 4, vfmsfpCode, True) 2742 2743 vmlsfpCode = ''' 2744 FPSCR fpscr = (FPSCR) FpscrExc; 2745 float mid = binaryOp(fpscr, srcReg1, srcReg2, fpMulS, 2746 true, true, VfpRoundNearest); 2747 destReg = binaryOp(fpscr, destReg, mid, fpSubS, 2748 true, true, VfpRoundNearest); 2749 FpscrExc = fpscr; 2750 ''' 2751 threeEqualRegInstFp("vmls", "NVmlsDFp", "SimdFloatMultAccOp", ("float",), 2, vmlsfpCode, True) 2752 threeEqualRegInstFp("vmls", "NVmlsQFp", "SimdFloatMultAccOp", ("float",), 4, vmlsfpCode, True) 2753 2754 vcgtfpCode = ''' 2755 FPSCR fpscr = (FPSCR) FpscrExc; 2756 float res = binaryOp(fpscr, srcReg1, srcReg2, vcgtFunc, 2757 true, true, VfpRoundNearest); 2758 destReg = (res == 0) ? -1 : 0; 2759 if (res == 2.0) 2760 fpscr.ioc = 1; 2761 FpscrExc = fpscr; 2762 ''' 2763 threeEqualRegInstFp("vcgt", "VcgtDFp", "SimdFloatCmpOp", ("float",), 2764 2, vcgtfpCode, toInt = True) 2765 threeEqualRegInstFp("vcgt", "VcgtQFp", "SimdFloatCmpOp", ("float",), 2766 4, vcgtfpCode, toInt = True) 2767 2768 vcgefpCode = ''' 2769 FPSCR fpscr = (FPSCR) FpscrExc; 2770 float res = binaryOp(fpscr, srcReg1, srcReg2, vcgeFunc, 2771 true, true, VfpRoundNearest); 2772 destReg = (res == 0) ? -1 : 0; 2773 if (res == 2.0) 2774 fpscr.ioc = 1; 2775 FpscrExc = fpscr; 2776 ''' 2777 threeEqualRegInstFp("vcge", "VcgeDFp", "SimdFloatCmpOp", ("float",), 2778 2, vcgefpCode, toInt = True) 2779 threeEqualRegInstFp("vcge", "VcgeQFp", "SimdFloatCmpOp", ("float",), 2780 4, vcgefpCode, toInt = True) 2781 2782 vacgtfpCode = ''' 2783 FPSCR fpscr = (FPSCR) FpscrExc; 2784 float res = binaryOp(fpscr, srcReg1, srcReg2, vacgtFunc, 2785 true, true, VfpRoundNearest); 2786 destReg = (res == 0) ? -1 : 0; 2787 if (res == 2.0) 2788 fpscr.ioc = 1; 2789 FpscrExc = fpscr; 2790 ''' 2791 threeEqualRegInstFp("vacgt", "VacgtDFp", "SimdFloatCmpOp", ("float",), 2792 2, vacgtfpCode, toInt = True) 2793 threeEqualRegInstFp("vacgt", "VacgtQFp", "SimdFloatCmpOp", ("float",), 2794 4, vacgtfpCode, toInt = True) 2795 2796 vacgefpCode = ''' 2797 FPSCR fpscr = (FPSCR) FpscrExc; 2798 float res = binaryOp(fpscr, srcReg1, srcReg2, vacgeFunc, 2799 true, true, VfpRoundNearest); 2800 destReg = (res == 0) ? -1 : 0; 2801 if (res == 2.0) 2802 fpscr.ioc = 1; 2803 FpscrExc = fpscr; 2804 ''' 2805 threeEqualRegInstFp("vacge", "VacgeDFp", "SimdFloatCmpOp", ("float",), 2806 2, vacgefpCode, toInt = True) 2807 threeEqualRegInstFp("vacge", "VacgeQFp", "SimdFloatCmpOp", ("float",), 2808 4, vacgefpCode, toInt = True) 2809 2810 vceqfpCode = ''' 2811 FPSCR fpscr = (FPSCR) FpscrExc; 2812 float res = binaryOp(fpscr, srcReg1, srcReg2, vceqFunc, 2813 true, true, VfpRoundNearest); 2814 destReg = (res == 0) ? -1 : 0; 2815 if (res == 2.0) 2816 fpscr.ioc = 1; 2817 FpscrExc = fpscr; 2818 ''' 2819 threeEqualRegInstFp("vceq", "VceqDFp", "SimdFloatCmpOp", ("float",), 2820 2, vceqfpCode, toInt = True) 2821 threeEqualRegInstFp("vceq", "VceqQFp", "SimdFloatCmpOp", ("float",), 2822 4, vceqfpCode, toInt = True) 2823 2824 vrecpsCode = ''' 2825 FPSCR fpscr = (FPSCR) FpscrExc; 2826 destReg = binaryOp(fpscr, srcReg1, srcReg2, fpRecpsS, 2827 true, true, VfpRoundNearest); 2828 FpscrExc = fpscr; 2829 ''' 2830 threeEqualRegInstFp("vrecps", "VrecpsDFp", "SimdFloatMultAccOp", ("float",), 2, vrecpsCode) 2831 threeEqualRegInstFp("vrecps", "VrecpsQFp", "SimdFloatMultAccOp", ("float",), 4, vrecpsCode) 2832 2833 vrsqrtsCode = ''' 2834 FPSCR fpscr = (FPSCR) FpscrExc; 2835 destReg = binaryOp(fpscr, srcReg1, srcReg2, fpRSqrtsS, 2836 true, true, VfpRoundNearest); 2837 FpscrExc = fpscr; 2838 ''' 2839 threeEqualRegInstFp("vrsqrts", "VrsqrtsDFp", "SimdFloatMiscOp", ("float",), 2, vrsqrtsCode) 2840 threeEqualRegInstFp("vrsqrts", "VrsqrtsQFp", "SimdFloatMiscOp", ("float",), 4, vrsqrtsCode) 2841 2842 vabdfpCode = ''' 2843 FPSCR fpscr = (FPSCR) FpscrExc; 2844 float mid = binaryOp(fpscr, srcReg1, srcReg2, fpSubS, 2845 true, true, VfpRoundNearest); 2846 destReg = fabs(mid); 2847 FpscrExc = fpscr; 2848 ''' 2849 threeEqualRegInstFp("vabd", "VabdDFp", "SimdFloatAddOp", ("float",), 2, vabdfpCode) 2850 threeEqualRegInstFp("vabd", "VabdQFp", "SimdFloatAddOp", ("float",), 4, vabdfpCode) 2851 2852 twoEqualRegInst("vmla", "VmlasD", "SimdMultAccOp", unsignedTypes, 2, vmlaCode, True) 2853 twoEqualRegInst("vmla", "VmlasQ", "SimdMultAccOp", unsignedTypes, 4, vmlaCode, True) 2854 twoEqualRegInstFp("vmla", "VmlasDFp", "SimdFloatMultAccOp", ("float",), 2, vmlafpCode, True) 2855 twoEqualRegInstFp("vmla", "VmlasQFp", "SimdFloatMultAccOp", ("float",), 4, vmlafpCode, True) 2856 twoRegLongInst("vmlal", "Vmlals", "SimdMultAccOp", smallTypes, vmlalCode, True) 2857 2858 twoEqualRegInst("vmls", "VmlssD", "SimdMultAccOp", allTypes, 2, vmlsCode, True) 2859 twoEqualRegInst("vmls", "VmlssQ", "SimdMultAccOp", allTypes, 4, vmlsCode, True) 2860 twoEqualRegInstFp("vmls", "VmlssDFp", "SimdFloatMultAccOp", ("float",), 2, vmlsfpCode, True) 2861 twoEqualRegInstFp("vmls", "VmlssQFp", "SimdFloatMultAccOp", ("float",), 4, vmlsfpCode, True) 2862 twoRegLongInst("vmlsl", "Vmlsls", "SimdMultAccOp", smallTypes, vmlslCode, True) 2863 2864 twoEqualRegInst("vmul", "VmulsD", "SimdMultOp", allTypes, 2, vmulCode) 2865 twoEqualRegInst("vmul", "VmulsQ", "SimdMultOp", allTypes, 4, vmulCode) 2866 twoEqualRegInstFp("vmul", "VmulsDFp", "SimdFloatMultOp", ("float",), 2, vmulfpCode) 2867 twoEqualRegInstFp("vmul", "VmulsQFp", "SimdFloatMultOp", ("float",), 4, vmulfpCode) 2868 twoRegLongInst("vmull", "Vmulls", "SimdMultOp", smallTypes, vmullCode) 2869 2870 twoRegLongInst("vqdmull", "Vqdmulls", "SimdMultOp", smallTypes, vqdmullCode) 2871 twoRegLongInst("vqdmlal", "Vqdmlals", "SimdMultAccOp", smallTypes, vqdmlalCode, True) 2872 twoRegLongInst("vqdmlsl", "Vqdmlsls", "SimdMultAccOp", smallTypes, vqdmlslCode, True) 2873 twoEqualRegInst("vqdmulh", "VqdmulhsD", "SimdMultOp", smallSignedTypes, 2, vqdmulhCode) 2874 twoEqualRegInst("vqdmulh", "VqdmulhsQ", "SimdMultOp", smallSignedTypes, 4, vqdmulhCode) 2875 twoEqualRegInst("vqrdmulh", "VqrdmulhsD", 2876 "SimdMultOp", smallSignedTypes, 2, vqrdmulhCode) 2877 twoEqualRegInst("vqrdmulh", "VqrdmulhsQ", 2878 "SimdMultOp", smallSignedTypes, 4, vqrdmulhCode) 2879 2880 vshrCode = ''' 2881 if (imm >= sizeof(srcElem1) * 8) { 2882 if (ltz(srcElem1)) 2883 destElem = -1; 2884 else 2885 destElem = 0; 2886 } else { 2887 destElem = srcElem1 >> imm; 2888 } 2889 ''' 2890 twoRegShiftInst("vshr", "NVshrD", "SimdShiftOp", allTypes, 2, vshrCode) 2891 twoRegShiftInst("vshr", "NVshrQ", "SimdShiftOp", allTypes, 4, vshrCode) 2892 2893 vsraCode = ''' 2894 Element mid;; 2895 if (imm >= sizeof(srcElem1) * 8) { 2896 mid = ltz(srcElem1) ? -1 : 0; 2897 } else { 2898 mid = srcElem1 >> imm; 2899 if (ltz(srcElem1) && !ltz(mid)) { 2900 mid |= -(mid & ((Element)1 << 2901 (sizeof(Element) * 8 - 1 - imm))); 2902 } 2903 } 2904 destElem += mid; 2905 ''' 2906 twoRegShiftInst("vsra", "NVsraD", "SimdShiftAccOp", allTypes, 2, vsraCode, True) 2907 twoRegShiftInst("vsra", "NVsraQ", "SimdShiftAccOp", allTypes, 4, vsraCode, True) 2908 2909 vrshrCode = ''' 2910 if (imm > sizeof(srcElem1) * 8) { 2911 destElem = 0; 2912 } else if (imm) { 2913 Element rBit = bits(srcElem1, imm - 1); 2914 destElem = ((srcElem1 >> (imm - 1)) >> 1) + rBit; 2915 } else { 2916 destElem = srcElem1; 2917 } 2918 ''' 2919 twoRegShiftInst("vrshr", "NVrshrD", "SimdShiftOp", allTypes, 2, vrshrCode) 2920 twoRegShiftInst("vrshr", "NVrshrQ", "SimdShiftOp", allTypes, 4, vrshrCode) 2921 2922 vrsraCode = ''' 2923 if (imm > sizeof(srcElem1) * 8) { 2924 destElem += 0; 2925 } else if (imm) { 2926 Element rBit = bits(srcElem1, imm - 1); 2927 destElem += ((srcElem1 >> (imm - 1)) >> 1) + rBit; 2928 } else { 2929 destElem += srcElem1; 2930 } 2931 ''' 2932 twoRegShiftInst("vrsra", "NVrsraD", "SimdShiftAccOp", allTypes, 2, vrsraCode, True) 2933 twoRegShiftInst("vrsra", "NVrsraQ", "SimdShiftAccOp", allTypes, 4, vrsraCode, True) 2934 2935 vsriCode = ''' 2936 if (imm >= sizeof(Element) * 8) 2937 destElem = destElem; 2938 else 2939 destElem = (srcElem1 >> imm) | 2940 (destElem & ~mask(sizeof(Element) * 8 - imm)); 2941 ''' 2942 twoRegShiftInst("vsri", "NVsriD", "SimdShiftOp", unsignedTypes, 2, vsriCode, True) 2943 twoRegShiftInst("vsri", "NVsriQ", "SimdShiftOp", unsignedTypes, 4, vsriCode, True) 2944 2945 vshlCode = ''' 2946 if (imm >= sizeof(Element) * 8) 2947 destElem = (srcElem1 << (sizeof(Element) * 8 - 1)) << 1; 2948 else 2949 destElem = srcElem1 << imm; 2950 ''' 2951 twoRegShiftInst("vshl", "NVshlD", "SimdShiftOp", unsignedTypes, 2, vshlCode) 2952 twoRegShiftInst("vshl", "NVshlQ", "SimdShiftOp", unsignedTypes, 4, vshlCode) 2953 2954 vsliCode = ''' 2955 if (imm >= sizeof(Element) * 8) 2956 destElem = destElem; 2957 else 2958 destElem = (srcElem1 << imm) | (destElem & mask(imm)); 2959 ''' 2960 twoRegShiftInst("vsli", "NVsliD", "SimdShiftOp", unsignedTypes, 2, vsliCode, True) 2961 twoRegShiftInst("vsli", "NVsliQ", "SimdShiftOp", unsignedTypes, 4, vsliCode, True) 2962 2963 vqshlCode = ''' 2964 FPSCR fpscr = (FPSCR) FpscrQc; 2965 if (imm >= sizeof(Element) * 8) { 2966 if (srcElem1 != 0) { 2967 destElem = (Element)1 << (sizeof(Element) * 8 - 1); 2968 if (srcElem1 > 0) 2969 destElem = ~destElem; 2970 fpscr.qc = 1; 2971 } else { 2972 destElem = 0; 2973 } 2974 } else if (imm) { 2975 destElem = (srcElem1 << imm); 2976 uint64_t topBits = bits((uint64_t)srcElem1, 2977 sizeof(Element) * 8 - 1, 2978 sizeof(Element) * 8 - 1 - imm); 2979 if (topBits != 0 && topBits != mask(imm + 1)) { 2980 destElem = (Element)1 << (sizeof(Element) * 8 - 1); 2981 if (srcElem1 > 0) 2982 destElem = ~destElem; 2983 fpscr.qc = 1; 2984 } 2985 } else { 2986 destElem = srcElem1; 2987 } 2988 FpscrQc = fpscr; 2989 ''' 2990 twoRegShiftInst("vqshl", "NVqshlD", "SimdShiftOp", signedTypes, 2, vqshlCode) 2991 twoRegShiftInst("vqshl", "NVqshlQ", "SimdShiftOp", signedTypes, 4, vqshlCode) 2992 2993 vqshluCode = ''' 2994 FPSCR fpscr = (FPSCR) FpscrQc; 2995 if (imm >= sizeof(Element) * 8) { 2996 if (srcElem1 != 0) { 2997 destElem = mask(sizeof(Element) * 8); 2998 fpscr.qc = 1; 2999 } else { 3000 destElem = 0; 3001 } 3002 } else if (imm) { 3003 destElem = (srcElem1 << imm); 3004 uint64_t topBits = bits((uint64_t)srcElem1, 3005 sizeof(Element) * 8 - 1, 3006 sizeof(Element) * 8 - imm); 3007 if (topBits != 0) { 3008 destElem = mask(sizeof(Element) * 8); 3009 fpscr.qc = 1; 3010 } 3011 } else { 3012 destElem = srcElem1; 3013 } 3014 FpscrQc = fpscr; 3015 ''' 3016 twoRegShiftInst("vqshlu", "NVqshluD", "SimdShiftOp", unsignedTypes, 2, vqshluCode) 3017 twoRegShiftInst("vqshlu", "NVqshluQ", "SimdShiftOp", unsignedTypes, 4, vqshluCode) 3018 3019 vqshlusCode = ''' 3020 FPSCR fpscr = (FPSCR) FpscrQc; 3021 if (imm >= sizeof(Element) * 8) { 3022 if (srcElem1 < 0) { 3023 destElem = 0; 3024 fpscr.qc = 1; 3025 } else if (srcElem1 > 0) { 3026 destElem = mask(sizeof(Element) * 8); 3027 fpscr.qc = 1; 3028 } else { 3029 destElem = 0; 3030 } 3031 } else if (imm) { 3032 destElem = (srcElem1 << imm); 3033 uint64_t topBits = bits((uint64_t)srcElem1, 3034 sizeof(Element) * 8 - 1, 3035 sizeof(Element) * 8 - imm); 3036 if (srcElem1 < 0) { 3037 destElem = 0; 3038 fpscr.qc = 1; 3039 } else if (topBits != 0) { 3040 destElem = mask(sizeof(Element) * 8); 3041 fpscr.qc = 1; 3042 } 3043 } else { 3044 if (srcElem1 < 0) { 3045 fpscr.qc = 1; 3046 destElem = 0; 3047 } else { 3048 destElem = srcElem1; 3049 } 3050 } 3051 FpscrQc = fpscr; 3052 ''' 3053 twoRegShiftInst("vqshlus", "NVqshlusD", "SimdShiftOp", signedTypes, 2, vqshlusCode) 3054 twoRegShiftInst("vqshlus", "NVqshlusQ", "SimdShiftOp", signedTypes, 4, vqshlusCode) 3055 3056 vshrnCode = ''' 3057 if (imm >= sizeof(srcElem1) * 8) { 3058 destElem = 0; 3059 } else { 3060 destElem = srcElem1 >> imm; 3061 } 3062 ''' 3063 twoRegNarrowShiftInst("vshrn", "NVshrn", "SimdShiftOp", smallUnsignedTypes, vshrnCode) 3064 3065 vrshrnCode = ''' 3066 if (imm > sizeof(srcElem1) * 8) { 3067 destElem = 0; 3068 } else if (imm) { 3069 Element rBit = bits(srcElem1, imm - 1); 3070 destElem = ((srcElem1 >> (imm - 1)) >> 1) + rBit; 3071 } else { 3072 destElem = srcElem1; 3073 } 3074 ''' 3075 twoRegNarrowShiftInst("vrshrn", "NVrshrn", "SimdShiftOp", smallUnsignedTypes, vrshrnCode) 3076 3077 vqshrnCode = ''' 3078 FPSCR fpscr = (FPSCR) FpscrQc; 3079 if (imm > sizeof(srcElem1) * 8) { 3080 if (srcElem1 != 0 && srcElem1 != -1) 3081 fpscr.qc = 1; 3082 destElem = 0; 3083 } else if (imm) { 3084 BigElement mid = ((srcElem1 >> (imm - 1)) >> 1); 3085 mid |= -(mid & ((BigElement)1 << 3086 (sizeof(BigElement) * 8 - 1 - imm))); 3087 if (mid != (Element)mid) { 3088 destElem = mask(sizeof(Element) * 8 - 1); 3089 if (srcElem1 < 0) 3090 destElem = ~destElem; 3091 fpscr.qc = 1; 3092 } else { 3093 destElem = mid; 3094 } 3095 } else { 3096 destElem = srcElem1; 3097 } 3098 FpscrQc = fpscr; 3099 ''' 3100 twoRegNarrowShiftInst("vqshrn", "NVqshrn", "SimdShiftOp", smallSignedTypes, vqshrnCode) 3101 3102 vqshrunCode = ''' 3103 FPSCR fpscr = (FPSCR) FpscrQc; 3104 if (imm > sizeof(srcElem1) * 8) { 3105 if (srcElem1 != 0) 3106 fpscr.qc = 1; 3107 destElem = 0; 3108 } else if (imm) { 3109 BigElement mid = ((srcElem1 >> (imm - 1)) >> 1); 3110 if (mid != (Element)mid) { 3111 destElem = mask(sizeof(Element) * 8); 3112 fpscr.qc = 1; 3113 } else { 3114 destElem = mid; 3115 } 3116 } else { 3117 destElem = srcElem1; 3118 } 3119 FpscrQc = fpscr; 3120 ''' 3121 twoRegNarrowShiftInst("vqshrun", "NVqshrun", 3122 "SimdShiftOp", smallUnsignedTypes, vqshrunCode) 3123 3124 vqshrunsCode = ''' 3125 FPSCR fpscr = (FPSCR) FpscrQc; 3126 if (imm > sizeof(srcElem1) * 8) { 3127 if (srcElem1 != 0) 3128 fpscr.qc = 1; 3129 destElem = 0; 3130 } else if (imm) { 3131 BigElement mid = ((srcElem1 >> (imm - 1)) >> 1); 3132 if (bits(mid, sizeof(BigElement) * 8 - 1, 3133 sizeof(Element) * 8) != 0) { 3134 if (srcElem1 < 0) { 3135 destElem = 0; 3136 } else { 3137 destElem = mask(sizeof(Element) * 8); 3138 } 3139 fpscr.qc = 1; 3140 } else { 3141 destElem = mid; 3142 } 3143 } else { 3144 destElem = srcElem1; 3145 } 3146 FpscrQc = fpscr; 3147 ''' 3148 twoRegNarrowShiftInst("vqshrun", "NVqshruns", 3149 "SimdShiftOp", smallSignedTypes, vqshrunsCode) 3150 3151 vqrshrnCode = ''' 3152 FPSCR fpscr = (FPSCR) FpscrQc; 3153 if (imm > sizeof(srcElem1) * 8) { 3154 if (srcElem1 != 0 && srcElem1 != -1) 3155 fpscr.qc = 1; 3156 destElem = 0; 3157 } else if (imm) { 3158 BigElement mid = (srcElem1 >> (imm - 1)); 3159 uint64_t rBit = mid & 0x1; 3160 mid >>= 1; 3161 mid |= -(mid & ((BigElement)1 << 3162 (sizeof(BigElement) * 8 - 1 - imm))); 3163 mid += rBit; 3164 if (mid != (Element)mid) { 3165 destElem = mask(sizeof(Element) * 8 - 1); 3166 if (srcElem1 < 0) 3167 destElem = ~destElem; 3168 fpscr.qc = 1; 3169 } else { 3170 destElem = mid; 3171 } 3172 } else { 3173 if (srcElem1 != (Element)srcElem1) { 3174 destElem = mask(sizeof(Element) * 8 - 1); 3175 if (srcElem1 < 0) 3176 destElem = ~destElem; 3177 fpscr.qc = 1; 3178 } else { 3179 destElem = srcElem1; 3180 } 3181 } 3182 FpscrQc = fpscr; 3183 ''' 3184 twoRegNarrowShiftInst("vqrshrn", "NVqrshrn", 3185 "SimdShiftOp", smallSignedTypes, vqrshrnCode) 3186 3187 vqrshrunCode = ''' 3188 FPSCR fpscr = (FPSCR) FpscrQc; 3189 if (imm > sizeof(srcElem1) * 8) { 3190 if (srcElem1 != 0) 3191 fpscr.qc = 1; 3192 destElem = 0; 3193 } else if (imm) { 3194 BigElement mid = (srcElem1 >> (imm - 1)); 3195 uint64_t rBit = mid & 0x1; 3196 mid >>= 1; 3197 mid += rBit; 3198 if (mid != (Element)mid) { 3199 destElem = mask(sizeof(Element) * 8); 3200 fpscr.qc = 1; 3201 } else { 3202 destElem = mid; 3203 } 3204 } else { 3205 if (srcElem1 != (Element)srcElem1) { 3206 destElem = mask(sizeof(Element) * 8 - 1); 3207 fpscr.qc = 1; 3208 } else { 3209 destElem = srcElem1; 3210 } 3211 } 3212 FpscrQc = fpscr; 3213 ''' 3214 twoRegNarrowShiftInst("vqrshrun", "NVqrshrun", 3215 "SimdShiftOp", smallUnsignedTypes, vqrshrunCode) 3216 3217 vqrshrunsCode = ''' 3218 FPSCR fpscr = (FPSCR) FpscrQc; 3219 if (imm > sizeof(srcElem1) * 8) { 3220 if (srcElem1 != 0) 3221 fpscr.qc = 1; 3222 destElem = 0; 3223 } else if (imm) { 3224 BigElement mid = (srcElem1 >> (imm - 1)); 3225 uint64_t rBit = mid & 0x1; 3226 mid >>= 1; 3227 mid |= -(mid & ((BigElement)1 << 3228 (sizeof(BigElement) * 8 - 1 - imm))); 3229 mid += rBit; 3230 if (bits(mid, sizeof(BigElement) * 8 - 1, 3231 sizeof(Element) * 8) != 0) { 3232 if (srcElem1 < 0) { 3233 destElem = 0; 3234 } else { 3235 destElem = mask(sizeof(Element) * 8); 3236 } 3237 fpscr.qc = 1; 3238 } else { 3239 destElem = mid; 3240 } 3241 } else { 3242 if (srcElem1 < 0) { 3243 fpscr.qc = 1; 3244 destElem = 0; 3245 } else { 3246 destElem = srcElem1; 3247 } 3248 } 3249 FpscrQc = fpscr; 3250 ''' 3251 twoRegNarrowShiftInst("vqrshrun", "NVqrshruns", 3252 "SimdShiftOp", smallSignedTypes, vqrshrunsCode) 3253 3254 vshllCode = ''' 3255 if (imm >= sizeof(destElem) * 8) { 3256 destElem = 0; 3257 } else { 3258 destElem = (BigElement)srcElem1 << imm; 3259 } 3260 ''' 3261 twoRegLongShiftInst("vshll", "NVshll", "SimdShiftOp", smallTypes, vshllCode) 3262 3263 vmovlCode = ''' 3264 destElem = srcElem1; 3265 ''' 3266 twoRegLongShiftInst("vmovl", "NVmovl", "SimdMiscOp", smallTypes, vmovlCode) 3267 3268 vcvt2ufxCode = ''' 3269 FPSCR fpscr = (FPSCR) FpscrExc; 3270 if (flushToZero(srcElem1)) 3271 fpscr.idc = 1; 3272 VfpSavedState state = prepFpState(VfpRoundNearest); 3273 __asm__ __volatile__("" : "=m" (srcElem1) : "m" (srcElem1)); 3274 destReg = vfpFpToFixed<float>(srcElem1, false, 32, imm); 3275 __asm__ __volatile__("" :: "m" (destReg)); 3276 finishVfp(fpscr, state, true); 3277 FpscrExc = fpscr; 3278 ''' 3279 twoRegShiftInst("vcvt", "NVcvt2ufxD", "SimdCvtOp", ("float",), 3280 2, vcvt2ufxCode, toInt = True) 3281 twoRegShiftInst("vcvt", "NVcvt2ufxQ", "SimdCvtOp", ("float",), 3282 4, vcvt2ufxCode, toInt = True) 3283 3284 vcvt2sfxCode = ''' 3285 FPSCR fpscr = (FPSCR) FpscrExc; 3286 if (flushToZero(srcElem1)) 3287 fpscr.idc = 1; 3288 VfpSavedState state = prepFpState(VfpRoundNearest); 3289 __asm__ __volatile__("" : "=m" (srcElem1) : "m" (srcElem1)); 3290 destReg = vfpFpToFixed<float>(srcElem1, true, 32, imm); 3291 __asm__ __volatile__("" :: "m" (destReg)); 3292 finishVfp(fpscr, state, true); 3293 FpscrExc = fpscr; 3294 ''' 3295 twoRegShiftInst("vcvt", "NVcvt2sfxD", "SimdCvtOp", ("float",), 3296 2, vcvt2sfxCode, toInt = True) 3297 twoRegShiftInst("vcvt", "NVcvt2sfxQ", "SimdCvtOp", ("float",), 3298 4, vcvt2sfxCode, toInt = True) 3299 3300 vcvtu2fpCode = ''' 3301 FPSCR fpscr = (FPSCR) FpscrExc; 3302 VfpSavedState state = prepFpState(VfpRoundNearest); 3303 __asm__ __volatile__("" : "=m" (srcReg1) : "m" (srcReg1)); 3304 destElem = vfpUFixedToFpS(true, true, srcReg1, 32, imm); 3305 __asm__ __volatile__("" :: "m" (destElem)); 3306 finishVfp(fpscr, state, true); 3307 FpscrExc = fpscr; 3308 ''' 3309 twoRegShiftInst("vcvt", "NVcvtu2fpD", "SimdCvtOp", ("float",), 3310 2, vcvtu2fpCode, fromInt = True) 3311 twoRegShiftInst("vcvt", "NVcvtu2fpQ", "SimdCvtOp", ("float",), 3312 4, vcvtu2fpCode, fromInt = True) 3313 3314 vcvts2fpCode = ''' 3315 FPSCR fpscr = (FPSCR) FpscrExc; 3316 VfpSavedState state = prepFpState(VfpRoundNearest); 3317 __asm__ __volatile__("" : "=m" (srcReg1) : "m" (srcReg1)); 3318 destElem = vfpSFixedToFpS(true, true, srcReg1, 32, imm); 3319 __asm__ __volatile__("" :: "m" (destElem)); 3320 finishVfp(fpscr, state, true); 3321 FpscrExc = fpscr; 3322 ''' 3323 twoRegShiftInst("vcvt", "NVcvts2fpD", "SimdCvtOp", ("float",), 3324 2, vcvts2fpCode, fromInt = True) 3325 twoRegShiftInst("vcvt", "NVcvts2fpQ", "SimdCvtOp", ("float",), 3326 4, vcvts2fpCode, fromInt = True) 3327 3328 vcvts2hCode = ''' 3329 destElem = 0; 3330 FPSCR fpscr = (FPSCR) FpscrExc; 3331 float srcFp1 = bitsToFp(srcElem1, (float)0.0); 3332 if (flushToZero(srcFp1)) 3333 fpscr.idc = 1; 3334 VfpSavedState state = prepFpState(VfpRoundNearest); 3335 __asm__ __volatile__("" : "=m" (srcFp1), "=m" (destElem) 3336 : "m" (srcFp1), "m" (destElem)); 3337 destElem = vcvtFpSFpH(fpscr, true, true, VfpRoundNearest, 3338 fpscr.ahp, srcFp1); 3339 __asm__ __volatile__("" :: "m" (destElem)); 3340 finishVfp(fpscr, state, true); 3341 FpscrExc = fpscr; 3342 ''' 3343 twoRegNarrowMiscInst("vcvt", "NVcvts2h", "SimdCvtOp", ("uint16_t",), vcvts2hCode) 3344 3345 vcvth2sCode = ''' 3346 destElem = 0; 3347 FPSCR fpscr = (FPSCR) FpscrExc; 3348 VfpSavedState state = prepFpState(VfpRoundNearest); 3349 __asm__ __volatile__("" : "=m" (srcElem1), "=m" (destElem) 3350 : "m" (srcElem1), "m" (destElem)); 3351 destElem = fpToBits(vcvtFpHFpS(fpscr, true, fpscr.ahp, srcElem1)); 3352 __asm__ __volatile__("" :: "m" (destElem)); 3353 finishVfp(fpscr, state, true); 3354 FpscrExc = fpscr; 3355 ''' 3356 twoRegLongMiscInst("vcvt", "NVcvth2s", "SimdCvtOp", ("uint16_t",), vcvth2sCode) 3357 3358 vrsqrteCode = ''' 3359 destElem = unsignedRSqrtEstimate(srcElem1); 3360 ''' 3361 twoRegMiscInst("vrsqrte", "NVrsqrteD", "SimdSqrtOp", ("uint32_t",), 2, vrsqrteCode) 3362 twoRegMiscInst("vrsqrte", "NVrsqrteQ", "SimdSqrtOp", ("uint32_t",), 4, vrsqrteCode) 3363 3364 vrsqrtefpCode = ''' 3365 FPSCR fpscr = (FPSCR) FpscrExc; 3366 if (flushToZero(srcReg1)) 3367 fpscr.idc = 1; 3368 destReg = fprSqrtEstimate(fpscr, srcReg1); 3369 FpscrExc = fpscr; 3370 ''' 3371 twoRegMiscInstFp("vrsqrte", "NVrsqrteDFp", "SimdFloatSqrtOp", ("float",), 2, vrsqrtefpCode) 3372 twoRegMiscInstFp("vrsqrte", "NVrsqrteQFp", "SimdFloatSqrtOp", ("float",), 4, vrsqrtefpCode) 3373 3374 vrecpeCode = ''' 3375 destElem = unsignedRecipEstimate(srcElem1); 3376 ''' 3377 twoRegMiscInst("vrecpe", "NVrecpeD", "SimdMultAccOp", ("uint32_t",), 2, vrecpeCode) 3378 twoRegMiscInst("vrecpe", "NVrecpeQ", "SimdMultAccOp", ("uint32_t",), 4, vrecpeCode) 3379 3380 vrecpefpCode = ''' 3381 FPSCR fpscr = (FPSCR) FpscrExc; 3382 if (flushToZero(srcReg1)) 3383 fpscr.idc = 1; 3384 destReg = fpRecipEstimate(fpscr, srcReg1); 3385 FpscrExc = fpscr; 3386 ''' 3387 twoRegMiscInstFp("vrecpe", "NVrecpeDFp", "SimdFloatMultAccOp", ("float",), 2, vrecpefpCode) 3388 twoRegMiscInstFp("vrecpe", "NVrecpeQFp", "SimdFloatMultAccOp", ("float",), 4, vrecpefpCode) 3389 3390 vrev16Code = ''' 3391 destElem = srcElem1; 3392 unsigned groupSize = ((1 << 1) / sizeof(Element)); 3393 unsigned reverseMask = (groupSize - 1); 3394 j = i ^ reverseMask; 3395 ''' 3396 twoRegMiscInst("vrev16", "NVrev16D", "SimdAluOp", ("uint8_t",), 2, vrev16Code) 3397 twoRegMiscInst("vrev16", "NVrev16Q", "SimdAluOp", ("uint8_t",), 4, vrev16Code) 3398 vrev32Code = ''' 3399 destElem = srcElem1; 3400 unsigned groupSize = ((1 << 2) / sizeof(Element)); 3401 unsigned reverseMask = (groupSize - 1); 3402 j = i ^ reverseMask; 3403 ''' 3404 twoRegMiscInst("vrev32", "NVrev32D", 3405 "SimdAluOp", ("uint8_t", "uint16_t"), 2, vrev32Code) 3406 twoRegMiscInst("vrev32", "NVrev32Q", 3407 "SimdAluOp", ("uint8_t", "uint16_t"), 4, vrev32Code) 3408 vrev64Code = ''' 3409 destElem = srcElem1; 3410 unsigned groupSize = ((1 << 3) / sizeof(Element)); 3411 unsigned reverseMask = (groupSize - 1); 3412 j = i ^ reverseMask; 3413 ''' 3414 twoRegMiscInst("vrev64", "NVrev64D", "SimdAluOp", smallUnsignedTypes, 2, vrev64Code) 3415 twoRegMiscInst("vrev64", "NVrev64Q", "SimdAluOp", smallUnsignedTypes, 4, vrev64Code) 3416 3417 vpaddlCode = ''' 3418 destElem = (BigElement)srcElem1 + (BigElement)srcElem2; 3419 ''' 3420 twoRegCondenseInst("vpaddl", "NVpaddlD", "SimdAddOp", smallTypes, 2, vpaddlCode) 3421 twoRegCondenseInst("vpaddl", "NVpaddlQ", "SimdAddOp", smallTypes, 4, vpaddlCode) 3422 3423 vpadalCode = ''' 3424 destElem += (BigElement)srcElem1 + (BigElement)srcElem2; 3425 ''' 3426 twoRegCondenseInst("vpadal", "NVpadalD", "SimdAddAccOp", smallTypes, 2, vpadalCode, True) 3427 twoRegCondenseInst("vpadal", "NVpadalQ", "SimdAddAccOp", smallTypes, 4, vpadalCode, True) 3428 3429 vclsCode = ''' 3430 unsigned count = 0; 3431 if (srcElem1 < 0) { 3432 srcElem1 <<= 1; 3433 while (srcElem1 < 0 && count < sizeof(Element) * 8 - 1) { 3434 count++; 3435 srcElem1 <<= 1; 3436 } 3437 } else { 3438 srcElem1 <<= 1; 3439 while (srcElem1 >= 0 && count < sizeof(Element) * 8 - 1) { 3440 count++; 3441 srcElem1 <<= 1; 3442 } 3443 } 3444 destElem = count; 3445 ''' 3446 twoRegMiscInst("vcls", "NVclsD", "SimdAluOp", signedTypes, 2, vclsCode) 3447 twoRegMiscInst("vcls", "NVclsQ", "SimdAluOp", signedTypes, 4, vclsCode) 3448 3449 vclzCode = ''' 3450 unsigned count = 0; 3451 while (srcElem1 >= 0 && count < sizeof(Element) * 8) { 3452 count++; 3453 srcElem1 <<= 1; 3454 } 3455 destElem = count; 3456 ''' 3457 twoRegMiscInst("vclz", "NVclzD", "SimdAluOp", signedTypes, 2, vclzCode) 3458 twoRegMiscInst("vclz", "NVclzQ", "SimdAluOp", signedTypes, 4, vclzCode) 3459 3460 vcntCode = ''' 3461 unsigned count = 0; 3462 while (srcElem1 && count < sizeof(Element) * 8) { 3463 count += srcElem1 & 0x1; 3464 srcElem1 >>= 1; 3465 } 3466 destElem = count; 3467 ''' 3468 3469 twoRegMiscInst("vcnt", "NVcntD", "SimdAluOp", unsignedTypes, 2, vcntCode) 3470 twoRegMiscInst("vcnt", "NVcntQ", "SimdAluOp", unsignedTypes, 4, vcntCode) 3471 3472 vmvnCode = ''' 3473 destElem = ~srcElem1; 3474 ''' 3475 twoRegMiscInst("vmvn", "NVmvnD", "SimdAluOp", ("uint64_t",), 2, vmvnCode) 3476 twoRegMiscInst("vmvn", "NVmvnQ", "SimdAluOp", ("uint64_t",), 4, vmvnCode) 3477 3478 vqabsCode = ''' 3479 FPSCR fpscr = (FPSCR) FpscrQc; 3480 if (srcElem1 == (Element)((Element)1 << (sizeof(Element) * 8 - 1))) { 3481 fpscr.qc = 1; 3482 destElem = ~srcElem1; 3483 } else if (srcElem1 < 0) { 3484 destElem = -srcElem1; 3485 } else { 3486 destElem = srcElem1; 3487 } 3488 FpscrQc = fpscr; 3489 ''' 3490 twoRegMiscInst("vqabs", "NVqabsD", "SimdAluOp", signedTypes, 2, vqabsCode) 3491 twoRegMiscInst("vqabs", "NVqabsQ", "SimdAluOp", signedTypes, 4, vqabsCode) 3492 3493 vqnegCode = ''' 3494 FPSCR fpscr = (FPSCR) FpscrQc; 3495 if (srcElem1 == (Element)((Element)1 << (sizeof(Element) * 8 - 1))) { 3496 fpscr.qc = 1; 3497 destElem = ~srcElem1; 3498 } else { 3499 destElem = -srcElem1; 3500 } 3501 FpscrQc = fpscr; 3502 ''' 3503 twoRegMiscInst("vqneg", "NVqnegD", "SimdAluOp", signedTypes, 2, vqnegCode) 3504 twoRegMiscInst("vqneg", "NVqnegQ", "SimdAluOp", signedTypes, 4, vqnegCode) 3505 3506 vabsCode = ''' 3507 if (srcElem1 < 0) { 3508 destElem = -srcElem1; 3509 } else { 3510 destElem = srcElem1; 3511 } 3512 ''' 3513 3514 twoRegMiscInst("vabs", "NVabsD", "SimdAluOp", signedTypes, 2, vabsCode) 3515 twoRegMiscInst("vabs", "NVabsQ", "SimdAluOp", signedTypes, 4, vabsCode) 3516 vabsfpCode = ''' 3517 union 3518 { 3519 uint32_t i; 3520 float f; 3521 } cStruct; 3522 cStruct.f = srcReg1; 3523 cStruct.i &= mask(sizeof(Element) * 8 - 1); 3524 destReg = cStruct.f; 3525 ''' 3526 twoRegMiscInstFp("vabs", "NVabsDFp", "SimdFloatAluOp", ("float",), 2, vabsfpCode) 3527 twoRegMiscInstFp("vabs", "NVabsQFp", "SimdFloatAluOp", ("float",), 4, vabsfpCode) 3528 3529 vnegCode = ''' 3530 destElem = -srcElem1; 3531 ''' 3532 twoRegMiscInst("vneg", "NVnegD", "SimdAluOp", signedTypes, 2, vnegCode) 3533 twoRegMiscInst("vneg", "NVnegQ", "SimdAluOp", signedTypes, 4, vnegCode) 3534 vnegfpCode = ''' 3535 destReg = -srcReg1; 3536 ''' 3537 twoRegMiscInstFp("vneg", "NVnegDFp", "SimdFloatAluOp", ("float",), 2, vnegfpCode) 3538 twoRegMiscInstFp("vneg", "NVnegQFp", "SimdFloatAluOp", ("float",), 4, vnegfpCode) 3539 3540 vcgtCode = 'destElem = (srcElem1 > 0) ? mask(sizeof(Element) * 8) : 0;' 3541 twoRegMiscInst("vcgt", "NVcgtD", "SimdCmpOp", signedTypes, 2, vcgtCode) 3542 twoRegMiscInst("vcgt", "NVcgtQ", "SimdCmpOp", signedTypes, 4, vcgtCode) 3543 vcgtfpCode = ''' 3544 FPSCR fpscr = (FPSCR) FpscrExc; 3545 float res = binaryOp(fpscr, srcReg1, (FloatReg)0.0, vcgtFunc, 3546 true, true, VfpRoundNearest); 3547 destReg = (res == 0) ? -1 : 0; 3548 if (res == 2.0) 3549 fpscr.ioc = 1; 3550 FpscrExc = fpscr; 3551 ''' 3552 twoRegMiscInstFp("vcgt", "NVcgtDFp", "SimdFloatCmpOp", ("float",), 3553 2, vcgtfpCode, toInt = True) 3554 twoRegMiscInstFp("vcgt", "NVcgtQFp", "SimdFloatCmpOp", ("float",), 3555 4, vcgtfpCode, toInt = True) 3556 3557 vcgeCode = 'destElem = (srcElem1 >= 0) ? mask(sizeof(Element) * 8) : 0;' 3558 twoRegMiscInst("vcge", "NVcgeD", "SimdCmpOp", signedTypes, 2, vcgeCode) 3559 twoRegMiscInst("vcge", "NVcgeQ", "SimdCmpOp", signedTypes, 4, vcgeCode) 3560 vcgefpCode = ''' 3561 FPSCR fpscr = (FPSCR) FpscrExc; 3562 float res = binaryOp(fpscr, srcReg1, (FloatReg)0.0, vcgeFunc, 3563 true, true, VfpRoundNearest); 3564 destReg = (res == 0) ? -1 : 0; 3565 if (res == 2.0) 3566 fpscr.ioc = 1; 3567 FpscrExc = fpscr; 3568 ''' 3569 twoRegMiscInstFp("vcge", "NVcgeDFp", "SimdFloatCmpOp", ("float",), 3570 2, vcgefpCode, toInt = True) 3571 twoRegMiscInstFp("vcge", "NVcgeQFp", "SimdFloatCmpOp", ("float",), 3572 4, vcgefpCode, toInt = True) 3573 3574 vceqCode = 'destElem = (srcElem1 == 0) ? mask(sizeof(Element) * 8) : 0;' 3575 twoRegMiscInst("vceq", "NVceqD", "SimdCmpOp", signedTypes, 2, vceqCode) 3576 twoRegMiscInst("vceq", "NVceqQ", "SimdCmpOp", signedTypes, 4, vceqCode) 3577 vceqfpCode = ''' 3578 FPSCR fpscr = (FPSCR) FpscrExc; 3579 float res = binaryOp(fpscr, srcReg1, (FloatReg)0.0, vceqFunc, 3580 true, true, VfpRoundNearest); 3581 destReg = (res == 0) ? -1 : 0; 3582 if (res == 2.0) 3583 fpscr.ioc = 1; 3584 FpscrExc = fpscr; 3585 ''' 3586 twoRegMiscInstFp("vceq", "NVceqDFp", "SimdFloatCmpOp", ("float",), 3587 2, vceqfpCode, toInt = True) 3588 twoRegMiscInstFp("vceq", "NVceqQFp", "SimdFloatCmpOp", ("float",), 3589 4, vceqfpCode, toInt = True) 3590 3591 vcleCode = 'destElem = (srcElem1 <= 0) ? mask(sizeof(Element) * 8) : 0;' 3592 twoRegMiscInst("vcle", "NVcleD", "SimdCmpOp", signedTypes, 2, vcleCode) 3593 twoRegMiscInst("vcle", "NVcleQ", "SimdCmpOp", signedTypes, 4, vcleCode) 3594 vclefpCode = ''' 3595 FPSCR fpscr = (FPSCR) FpscrExc; 3596 float res = binaryOp(fpscr, srcReg1, (FloatReg)0.0, vcleFunc, 3597 true, true, VfpRoundNearest); 3598 destReg = (res == 0) ? -1 : 0; 3599 if (res == 2.0) 3600 fpscr.ioc = 1; 3601 FpscrExc = fpscr; 3602 ''' 3603 twoRegMiscInstFp("vcle", "NVcleDFp", "SimdFloatCmpOp", ("float",), 3604 2, vclefpCode, toInt = True) 3605 twoRegMiscInstFp("vcle", "NVcleQFp", "SimdFloatCmpOp", ("float",), 3606 4, vclefpCode, toInt = True) 3607 3608 vcltCode = 'destElem = (srcElem1 < 0) ? mask(sizeof(Element) * 8) : 0;' 3609 twoRegMiscInst("vclt", "NVcltD", "SimdCmpOp", signedTypes, 2, vcltCode) 3610 twoRegMiscInst("vclt", "NVcltQ", "SimdCmpOp", signedTypes, 4, vcltCode) 3611 vcltfpCode = ''' 3612 FPSCR fpscr = (FPSCR) FpscrExc; 3613 float res = binaryOp(fpscr, srcReg1, (FloatReg)0.0, vcltFunc, 3614 true, true, VfpRoundNearest); 3615 destReg = (res == 0) ? -1 : 0; 3616 if (res == 2.0) 3617 fpscr.ioc = 1; 3618 FpscrExc = fpscr; 3619 ''' 3620 twoRegMiscInstFp("vclt", "NVcltDFp", "SimdFloatCmpOp", ("float",), 3621 2, vcltfpCode, toInt = True) 3622 twoRegMiscInstFp("vclt", "NVcltQFp", "SimdFloatCmpOp", ("float",), 3623 4, vcltfpCode, toInt = True) 3624 3625 vswpCode = ''' 3626 FloatRegBits mid; 3627 for (unsigned r = 0; r < rCount; r++) { 3628 mid = srcReg1.regs[r]; 3629 srcReg1.regs[r] = destReg.regs[r]; 3630 destReg.regs[r] = mid; 3631 } 3632 ''' 3633 twoRegMiscScramble("vswp", "NVswpD", "SimdAluOp", ("uint64_t",), 2, vswpCode) 3634 twoRegMiscScramble("vswp", "NVswpQ", "SimdAluOp", ("uint64_t",), 4, vswpCode) 3635 3636 vtrnCode = ''' 3637 Element mid; 3638 for (unsigned i = 0; i < eCount; i += 2) { 3639 mid = srcReg1.elements[i]; 3640 srcReg1.elements[i] = destReg.elements[i + 1]; 3641 destReg.elements[i + 1] = mid; 3642 } 3643 ''' 3644 twoRegMiscScramble("vtrn", "NVtrnD", "SimdAluOp", 3645 smallUnsignedTypes, 2, vtrnCode) 3646 twoRegMiscScramble("vtrn", "NVtrnQ", "SimdAluOp", 3647 smallUnsignedTypes, 4, vtrnCode) 3648 3649 vuzpCode = ''' 3650 Element mid[eCount]; 3651 memcpy(&mid, &srcReg1, sizeof(srcReg1)); 3652 for (unsigned i = 0; i < eCount / 2; i++) { 3653 srcReg1.elements[i] = destReg.elements[2 * i + 1]; 3654 srcReg1.elements[eCount / 2 + i] = mid[2 * i + 1]; 3655 destReg.elements[i] = destReg.elements[2 * i]; 3656 } 3657 for (unsigned i = 0; i < eCount / 2; i++) { 3658 destReg.elements[eCount / 2 + i] = mid[2 * i]; 3659 } 3660 ''' 3661 twoRegMiscScramble("vuzp", "NVuzpD", "SimdAluOp", unsignedTypes, 2, vuzpCode) 3662 twoRegMiscScramble("vuzp", "NVuzpQ", "SimdAluOp", unsignedTypes, 4, vuzpCode) 3663 3664 vzipCode = ''' 3665 Element mid[eCount]; 3666 memcpy(&mid, &destReg, sizeof(destReg)); 3667 for (unsigned i = 0; i < eCount / 2; i++) { 3668 destReg.elements[2 * i] = mid[i]; 3669 destReg.elements[2 * i + 1] = srcReg1.elements[i]; 3670 } 3671 for (int i = 0; i < eCount / 2; i++) { 3672 srcReg1.elements[2 * i] = mid[eCount / 2 + i]; 3673 srcReg1.elements[2 * i + 1] = srcReg1.elements[eCount / 2 + i]; 3674 } 3675 ''' 3676 twoRegMiscScramble("vzip", "NVzipD", "SimdAluOp", unsignedTypes, 2, vzipCode) 3677 twoRegMiscScramble("vzip", "NVzipQ", "SimdAluOp", unsignedTypes, 4, vzipCode) 3678 3679 vmovnCode = 'destElem = srcElem1;' 3680 twoRegNarrowMiscInst("vmovn", "NVmovn", "SimdMiscOp", smallUnsignedTypes, vmovnCode) 3681 3682 vdupCode = 'destElem = srcElem1;' 3683 twoRegMiscScInst("vdup", "NVdupD", "SimdAluOp", smallUnsignedTypes, 2, vdupCode) 3684 twoRegMiscScInst("vdup", "NVdupQ", "SimdAluOp", smallUnsignedTypes, 4, vdupCode) 3685 3686 def vdupGprInst(name, Name, opClass, types, rCount): 3687 global header_output, exec_output 3688 eWalkCode = ''' 3689 RegVect destReg; 3690 for (unsigned i = 0; i < eCount; i++) { 3691 destReg.elements[i] = htog((Element)Op1); 3692 } 3693 ''' 3694 for reg in range(rCount): 3695 eWalkCode += ''' 3696 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 3697 ''' % { "reg" : reg } 3698 iop = InstObjParams(name, Name, 3699 "RegRegOp", 3700 { "code": eWalkCode, 3701 "r_count": rCount, 3702 "predicate_test": predicateTest, 3703 "op_class": opClass }, []) 3704 header_output += NeonRegRegOpDeclare.subst(iop) 3705 exec_output += NeonEqualRegExecute.subst(iop) 3706 for type in types: 3707 substDict = { "targs" : type, 3708 "class_name" : Name } 3709 exec_output += NeonExecDeclare.subst(substDict) 3710 vdupGprInst("vdup", "NVdupDGpr", "SimdMiscOp", smallUnsignedTypes, 2) 3711 vdupGprInst("vdup", "NVdupQGpr", "SimdMiscOp", smallUnsignedTypes, 4) 3712 3713 vmovCode = 'destElem = imm;' 3714 oneRegImmInst("vmov", "NVmoviD", "SimdMiscOp", ("uint64_t",), 2, vmovCode) 3715 oneRegImmInst("vmov", "NVmoviQ", "SimdMiscOp", ("uint64_t",), 4, vmovCode) 3716 3717 vorrCode = 'destElem |= imm;' 3718 oneRegImmInst("vorr", "NVorriD", "SimdAluOp", ("uint64_t",), 2, vorrCode, True) 3719 oneRegImmInst("vorr", "NVorriQ", "SimdAluOp", ("uint64_t",), 4, vorrCode, True) 3720 3721 vmvnCode = 'destElem = ~imm;' 3722 oneRegImmInst("vmvn", "NVmvniD", "SimdAluOp", ("uint64_t",), 2, vmvnCode) 3723 oneRegImmInst("vmvn", "NVmvniQ", "SimdAluOp", ("uint64_t",), 4, vmvnCode) 3724 3725 vbicCode = 'destElem &= ~imm;' 3726 oneRegImmInst("vbic", "NVbiciD", "SimdAluOp", ("uint64_t",), 2, vbicCode, True) 3727 oneRegImmInst("vbic", "NVbiciQ", "SimdAluOp", ("uint64_t",), 4, vbicCode, True) 3728 3729 vqmovnCode = ''' 3730 FPSCR fpscr = (FPSCR) FpscrQc; 3731 destElem = srcElem1; 3732 if ((BigElement)destElem != srcElem1) { 3733 fpscr.qc = 1; 3734 destElem = mask(sizeof(Element) * 8 - 1); 3735 if (srcElem1 < 0) 3736 destElem = ~destElem; 3737 } 3738 FpscrQc = fpscr; 3739 ''' 3740 twoRegNarrowMiscInst("vqmovn", "NVqmovn", "SimdMiscOp", smallSignedTypes, vqmovnCode) 3741 3742 vqmovunCode = ''' 3743 FPSCR fpscr = (FPSCR) FpscrQc; 3744 destElem = srcElem1; 3745 if ((BigElement)destElem != srcElem1) { 3746 fpscr.qc = 1; 3747 destElem = mask(sizeof(Element) * 8); 3748 } 3749 FpscrQc = fpscr; 3750 ''' 3751 twoRegNarrowMiscInst("vqmovun", "NVqmovun", 3752 "SimdMiscOp", smallUnsignedTypes, vqmovunCode) 3753 3754 vqmovunsCode = ''' 3755 FPSCR fpscr = (FPSCR) FpscrQc; 3756 destElem = srcElem1; 3757 if (srcElem1 < 0 || 3758 ((BigElement)destElem & mask(sizeof(Element) * 8)) != srcElem1) { 3759 fpscr.qc = 1; 3760 destElem = mask(sizeof(Element) * 8); 3761 if (srcElem1 < 0) 3762 destElem = ~destElem; 3763 } 3764 FpscrQc = fpscr; 3765 ''' 3766 twoRegNarrowMiscInst("vqmovun", "NVqmovuns", 3767 "SimdMiscOp", smallSignedTypes, vqmovunsCode) 3768 3769 def buildVext(name, Name, opClass, types, rCount, op): 3770 global header_output, exec_output 3771 eWalkCode = ''' 3772 RegVect srcReg1, srcReg2, destReg; 3773 ''' 3774 for reg in range(rCount): 3775 eWalkCode += simdEnabledCheckCode + ''' 3776 srcReg1.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw); 3777 srcReg2.regs[%(reg)d] = htog(FpOp2P%(reg)d_uw); 3778 ''' % { "reg" : reg } 3779 eWalkCode += op 3780 for reg in range(rCount): 3781 eWalkCode += ''' 3782 FpDestP%(reg)d_uw = gtoh(destReg.regs[%(reg)d]); 3783 ''' % { "reg" : reg } 3784 iop = InstObjParams(name, Name, 3785 "RegRegRegImmOp", 3786 { "code": eWalkCode, 3787 "r_count": rCount, 3788 "predicate_test": predicateTest, 3789 "op_class": opClass }, []) 3790 header_output += NeonRegRegRegImmOpDeclare.subst(iop) 3791 exec_output += NeonEqualRegExecute.subst(iop) 3792 for type in types: 3793 substDict = { "targs" : type, 3794 "class_name" : Name } 3795 exec_output += NeonExecDeclare.subst(substDict) 3796 3797 vextCode = ''' 3798 for (unsigned i = 0; i < eCount; i++) { 3799 unsigned index = i + imm; 3800 if (index < eCount) { 3801 destReg.elements[i] = srcReg1.elements[index]; 3802 } else { 3803 index -= eCount; 3804 if (index >= eCount) { 3805 fault = new UndefinedInstruction(machInst, false, mnemonic); 3806 } else { 3807 destReg.elements[i] = srcReg2.elements[index]; 3808 } 3809 } 3810 } 3811 ''' 3812 buildVext("vext", "NVextD", "SimdMiscOp", ("uint8_t",), 2, vextCode) 3813 buildVext("vext", "NVextQ", "SimdMiscOp", ("uint8_t",), 4, vextCode) 3814 3815 def buildVtbxl(name, Name, opClass, length, isVtbl): 3816 global header_output, decoder_output, exec_output 3817 code = ''' 3818 union 3819 { 3820 uint8_t bytes[32]; 3821 FloatRegBits regs[8]; 3822 } table; 3823 3824 union 3825 { 3826 uint8_t bytes[8]; 3827 FloatRegBits regs[2]; 3828 } destReg, srcReg2; 3829 3830 const unsigned length = %(length)d; 3831 const bool isVtbl = %(isVtbl)s; 3832 3833 srcReg2.regs[0] = htog(FpOp2P0_uw); 3834 srcReg2.regs[1] = htog(FpOp2P1_uw); 3835 3836 destReg.regs[0] = htog(FpDestP0_uw); 3837 destReg.regs[1] = htog(FpDestP1_uw); 3838 ''' % { "length" : length, "isVtbl" : isVtbl } 3839 for reg in range(8): 3840 if reg < length * 2: 3841 code += 'table.regs[%(reg)d] = htog(FpOp1P%(reg)d_uw);\n' % \ 3842 { "reg" : reg } 3843 else: 3844 code += 'table.regs[%(reg)d] = 0;\n' % { "reg" : reg } 3845 code += ''' 3846 for (unsigned i = 0; i < sizeof(destReg); i++) { 3847 uint8_t index = srcReg2.bytes[i]; 3848 if (index < 8 * length) { 3849 destReg.bytes[i] = table.bytes[index]; 3850 } else { 3851 if (isVtbl) 3852 destReg.bytes[i] = 0; 3853 // else destReg.bytes[i] unchanged 3854 } 3855 } 3856 3857 FpDestP0_uw = gtoh(destReg.regs[0]); 3858 FpDestP1_uw = gtoh(destReg.regs[1]); 3859 ''' 3860 iop = InstObjParams(name, Name, 3861 "RegRegRegOp", 3862 { "code": code, 3863 "predicate_test": predicateTest, 3864 "op_class": opClass }, []) 3865 header_output += RegRegRegOpDeclare.subst(iop) 3866 decoder_output += RegRegRegOpConstructor.subst(iop) 3867 exec_output += PredOpExecute.subst(iop) 3868 3869 buildVtbxl("vtbl", "NVtbl1", "SimdMiscOp", 1, "true") 3870 buildVtbxl("vtbl", "NVtbl2", "SimdMiscOp", 2, "true") 3871 buildVtbxl("vtbl", "NVtbl3", "SimdMiscOp", 3, "true") 3872 buildVtbxl("vtbl", "NVtbl4", "SimdMiscOp", 4, "true") 3873 3874 buildVtbxl("vtbx", "NVtbx1", "SimdMiscOp", 1, "false") 3875 buildVtbxl("vtbx", "NVtbx2", "SimdMiscOp", 2, "false") 3876 buildVtbxl("vtbx", "NVtbx3", "SimdMiscOp", 3, "false") 3877 buildVtbxl("vtbx", "NVtbx4", "SimdMiscOp", 4, "false") 3878}}; 3879