1// Copyright 2015, VIXL authors 2// All rights reserved. 3// 4// Redistribution and use in source and binary forms, with or without 5// modification, are permitted provided that the following conditions are met: 6// 7// * Redistributions of source code must retain the above copyright notice, 8// this list of conditions and the following disclaimer. 9// * Redistributions in binary form must reproduce the above copyright notice, 10// this list of conditions and the following disclaimer in the documentation 11// and/or other materials provided with the distribution. 12// * Neither the name of ARM Limited nor the names of its contributors may be 13// used to endorse or promote products derived from this software without 14// specific prior written permission. 15// 16// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS CONTRIBUTORS "AS IS" AND 17// ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 18// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 19// DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE 20// FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 22// SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 23// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 24// OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 25// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 27#include "assembler-aarch64.h" 28#include "instructions-aarch64.h" 29 30namespace vixl { 31namespace aarch64 { 32 33 34// Floating-point infinity values. 35const float16 kFP16PositiveInfinity = 0x7c00; 36const float16 kFP16NegativeInfinity = 0xfc00; 37const float kFP32PositiveInfinity = RawbitsToFloat(0x7f800000); 38const float kFP32NegativeInfinity = RawbitsToFloat(0xff800000); 39const double kFP64PositiveInfinity = 40 RawbitsToDouble(UINT64_C(0x7ff0000000000000)); 41const double kFP64NegativeInfinity = 42 RawbitsToDouble(UINT64_C(0xfff0000000000000)); 43 44 45// The default NaN values (for FPCR.DN=1). 46const double kFP64DefaultNaN = RawbitsToDouble(UINT64_C(0x7ff8000000000000)); 47const float kFP32DefaultNaN = RawbitsToFloat(0x7fc00000); 48const float16 kFP16DefaultNaN = 0x7e00; 49 50 51static uint64_t RepeatBitsAcrossReg(unsigned reg_size, 52 uint64_t value, 53 unsigned width) { 54 VIXL_ASSERT((width == 2) || (width == 4) || (width == 8) || (width == 16) || 55 (width == 32)); 56 VIXL_ASSERT((reg_size == kWRegSize) || (reg_size == kXRegSize)); 57 uint64_t result = value & ((UINT64_C(1) << width) - 1); 58 for (unsigned i = width; i < reg_size; i *= 2) { 59 result |= (result << i); 60 } 61 return result; 62} 63 64 65bool Instruction::IsLoad() const { 66 if (Mask(LoadStoreAnyFMask) != LoadStoreAnyFixed) { 67 return false; 68 } 69 70 if (Mask(LoadStorePairAnyFMask) == LoadStorePairAnyFixed) { 71 return Mask(LoadStorePairLBit) != 0; 72 } else { 73 LoadStoreOp op = static_cast<LoadStoreOp>(Mask(LoadStoreMask)); 74 switch (op) { 75 case LDRB_w: 76 case LDRH_w: 77 case LDR_w: 78 case LDR_x: 79 case LDRSB_w: 80 case LDRSB_x: 81 case LDRSH_w: 82 case LDRSH_x: 83 case LDRSW_x: 84 case LDR_b: 85 case LDR_h: 86 case LDR_s: 87 case LDR_d: 88 case LDR_q: 89 return true; 90 default: 91 return false; 92 } 93 } 94} 95 96 97bool Instruction::IsStore() const { 98 if (Mask(LoadStoreAnyFMask) != LoadStoreAnyFixed) { 99 return false; 100 } 101 102 if (Mask(LoadStorePairAnyFMask) == LoadStorePairAnyFixed) { 103 return Mask(LoadStorePairLBit) == 0; 104 } else { 105 LoadStoreOp op = static_cast<LoadStoreOp>(Mask(LoadStoreMask)); 106 switch (op) { 107 case STRB_w: 108 case STRH_w: 109 case STR_w: 110 case STR_x: 111 case STR_b: 112 case STR_h: 113 case STR_s: 114 case STR_d: 115 case STR_q: 116 return true; 117 default: 118 return false; 119 } 120 } 121} 122 123 124// Logical immediates can't encode zero, so a return value of zero is used to 125// indicate a failure case. Specifically, where the constraints on imm_s are 126// not met. 127uint64_t Instruction::GetImmLogical() const { 128 unsigned reg_size = GetSixtyFourBits() ? kXRegSize : kWRegSize; 129 int32_t n = GetBitN(); 130 int32_t imm_s = GetImmSetBits(); 131 int32_t imm_r = GetImmRotate(); 132 133 // An integer is constructed from the n, imm_s and imm_r bits according to 134 // the following table: 135 // 136 // N imms immr size S R 137 // 1 ssssss rrrrrr 64 UInt(ssssss) UInt(rrrrrr) 138 // 0 0sssss xrrrrr 32 UInt(sssss) UInt(rrrrr) 139 // 0 10ssss xxrrrr 16 UInt(ssss) UInt(rrrr) 140 // 0 110sss xxxrrr 8 UInt(sss) UInt(rrr) 141 // 0 1110ss xxxxrr 4 UInt(ss) UInt(rr) 142 // 0 11110s xxxxxr 2 UInt(s) UInt(r) 143 // (s bits must not be all set) 144 // 145 // A pattern is constructed of size bits, where the least significant S+1 146 // bits are set. The pattern is rotated right by R, and repeated across a 147 // 32 or 64-bit value, depending on destination register width. 148 // 149 150 if (n == 1) { 151 if (imm_s == 0x3f) { 152 return 0; 153 } 154 uint64_t bits = (UINT64_C(1) << (imm_s + 1)) - 1; 155 return RotateRight(bits, imm_r, 64); 156 } else { 157 if ((imm_s >> 1) == 0x1f) { 158 return 0; 159 } 160 for (int width = 0x20; width >= 0x2; width >>= 1) { 161 if ((imm_s & width) == 0) { 162 int mask = width - 1; 163 if ((imm_s & mask) == mask) { 164 return 0; 165 } 166 uint64_t bits = (UINT64_C(1) << ((imm_s & mask) + 1)) - 1; 167 return RepeatBitsAcrossReg(reg_size, 168 RotateRight(bits, imm_r & mask, width), 169 width); 170 } 171 } 172 } 173 VIXL_UNREACHABLE(); 174 return 0; 175} 176 177 178uint32_t Instruction::GetImmNEONabcdefgh() const { 179 return GetImmNEONabc() << 5 | GetImmNEONdefgh(); 180} 181 182 183float Instruction::Imm8ToFP32(uint32_t imm8) { 184 // Imm8: abcdefgh (8 bits) 185 // Single: aBbb.bbbc.defg.h000.0000.0000.0000.0000 (32 bits) 186 // where B is b ^ 1 187 uint32_t bits = imm8; 188 uint32_t bit7 = (bits >> 7) & 0x1; 189 uint32_t bit6 = (bits >> 6) & 0x1; 190 uint32_t bit5_to_0 = bits & 0x3f; 191 uint32_t result = (bit7 << 31) | ((32 - bit6) << 25) | (bit5_to_0 << 19); 192 193 return RawbitsToFloat(result); 194} 195 196 197float Instruction::GetImmFP32() const { return Imm8ToFP32(GetImmFP()); } 198 199 200double Instruction::Imm8ToFP64(uint32_t imm8) { 201 // Imm8: abcdefgh (8 bits) 202 // Double: aBbb.bbbb.bbcd.efgh.0000.0000.0000.0000 203 // 0000.0000.0000.0000.0000.0000.0000.0000 (64 bits) 204 // where B is b ^ 1 205 uint32_t bits = imm8; 206 uint64_t bit7 = (bits >> 7) & 0x1; 207 uint64_t bit6 = (bits >> 6) & 0x1; 208 uint64_t bit5_to_0 = bits & 0x3f; 209 uint64_t result = (bit7 << 63) | ((256 - bit6) << 54) | (bit5_to_0 << 48); 210 211 return RawbitsToDouble(result); 212} 213 214 215double Instruction::GetImmFP64() const { return Imm8ToFP64(GetImmFP()); } 216 217 218float Instruction::GetImmNEONFP32() const { 219 return Imm8ToFP32(GetImmNEONabcdefgh()); 220} 221 222 223double Instruction::GetImmNEONFP64() const { 224 return Imm8ToFP64(GetImmNEONabcdefgh()); 225} 226 227 228unsigned CalcLSDataSize(LoadStoreOp op) { 229 VIXL_ASSERT((LSSize_offset + LSSize_width) == (kInstructionSize * 8)); 230 unsigned size = static_cast<Instr>(op) >> LSSize_offset; 231 if ((op & LSVector_mask) != 0) { 232 // Vector register memory operations encode the access size in the "size" 233 // and "opc" fields. 234 if ((size == 0) && ((op & LSOpc_mask) >> LSOpc_offset) >= 2) { 235 size = kQRegSizeInBytesLog2; 236 } 237 } 238 return size; 239} 240 241 242unsigned CalcLSPairDataSize(LoadStorePairOp op) { 243 VIXL_STATIC_ASSERT(kXRegSizeInBytes == kDRegSizeInBytes); 244 VIXL_STATIC_ASSERT(kWRegSizeInBytes == kSRegSizeInBytes); 245 switch (op) { 246 case STP_q: 247 case LDP_q: 248 return kQRegSizeInBytesLog2; 249 case STP_x: 250 case LDP_x: 251 case STP_d: 252 case LDP_d: 253 return kXRegSizeInBytesLog2; 254 default: 255 return kWRegSizeInBytesLog2; 256 } 257} 258 259 260int Instruction::GetImmBranchRangeBitwidth(ImmBranchType branch_type) { 261 switch (branch_type) { 262 case UncondBranchType: 263 return ImmUncondBranch_width; 264 case CondBranchType: 265 return ImmCondBranch_width; 266 case CompareBranchType: 267 return ImmCmpBranch_width; 268 case TestBranchType: 269 return ImmTestBranch_width; 270 default: 271 VIXL_UNREACHABLE(); 272 return 0; 273 } 274} 275 276 277int32_t Instruction::GetImmBranchForwardRange(ImmBranchType branch_type) { 278 int32_t encoded_max = 1 << (GetImmBranchRangeBitwidth(branch_type) - 1); 279 return encoded_max * kInstructionSize; 280} 281 282 283bool Instruction::IsValidImmPCOffset(ImmBranchType branch_type, 284 int64_t offset) { 285 return IsIntN(GetImmBranchRangeBitwidth(branch_type), offset); 286} 287 288 289const Instruction* Instruction::GetImmPCOffsetTarget() const { 290 const Instruction* base = this; 291 ptrdiff_t offset; 292 if (IsPCRelAddressing()) { 293 // ADR and ADRP. 294 offset = GetImmPCRel(); 295 if (Mask(PCRelAddressingMask) == ADRP) { 296 base = AlignDown(base, kPageSize); 297 offset *= kPageSize; 298 } else { 299 VIXL_ASSERT(Mask(PCRelAddressingMask) == ADR); 300 } 301 } else { 302 // All PC-relative branches. 303 VIXL_ASSERT(GetBranchType() != UnknownBranchType); 304 // Relative branch offsets are instruction-size-aligned. 305 offset = GetImmBranch() * static_cast<int>(kInstructionSize); 306 } 307 return base + offset; 308} 309 310 311int Instruction::GetImmBranch() const { 312 switch (GetBranchType()) { 313 case CondBranchType: 314 return GetImmCondBranch(); 315 case UncondBranchType: 316 return GetImmUncondBranch(); 317 case CompareBranchType: 318 return GetImmCmpBranch(); 319 case TestBranchType: 320 return GetImmTestBranch(); 321 default: 322 VIXL_UNREACHABLE(); 323 } 324 return 0; 325} 326 327 328void Instruction::SetImmPCOffsetTarget(const Instruction* target) { 329 if (IsPCRelAddressing()) { 330 SetPCRelImmTarget(target); 331 } else { 332 SetBranchImmTarget(target); 333 } 334} 335 336 337void Instruction::SetPCRelImmTarget(const Instruction* target) { 338 ptrdiff_t imm21; 339 if ((Mask(PCRelAddressingMask) == ADR)) { 340 imm21 = target - this; 341 } else { 342 VIXL_ASSERT(Mask(PCRelAddressingMask) == ADRP); 343 uintptr_t this_page = reinterpret_cast<uintptr_t>(this) / kPageSize; 344 uintptr_t target_page = reinterpret_cast<uintptr_t>(target) / kPageSize; 345 imm21 = target_page - this_page; 346 } 347 Instr imm = Assembler::ImmPCRelAddress(static_cast<int32_t>(imm21)); 348 349 SetInstructionBits(Mask(~ImmPCRel_mask) | imm); 350} 351 352 353void Instruction::SetBranchImmTarget(const Instruction* target) { 354 VIXL_ASSERT(((target - this) & 3) == 0); 355 Instr branch_imm = 0; 356 uint32_t imm_mask = 0; 357 int offset = static_cast<int>((target - this) >> kInstructionSizeLog2); 358 switch (GetBranchType()) { 359 case CondBranchType: { 360 branch_imm = Assembler::ImmCondBranch(offset); 361 imm_mask = ImmCondBranch_mask; 362 break; 363 } 364 case UncondBranchType: { 365 branch_imm = Assembler::ImmUncondBranch(offset); 366 imm_mask = ImmUncondBranch_mask; 367 break; 368 } 369 case CompareBranchType: { 370 branch_imm = Assembler::ImmCmpBranch(offset); 371 imm_mask = ImmCmpBranch_mask; 372 break; 373 } 374 case TestBranchType: { 375 branch_imm = Assembler::ImmTestBranch(offset); 376 imm_mask = ImmTestBranch_mask; 377 break; 378 } 379 default: 380 VIXL_UNREACHABLE(); 381 } 382 SetInstructionBits(Mask(~imm_mask) | branch_imm); 383} 384 385 386void Instruction::SetImmLLiteral(const Instruction* source) { 387 VIXL_ASSERT(IsWordAligned(source)); 388 ptrdiff_t offset = (source - this) >> kLiteralEntrySizeLog2; 389 Instr imm = Assembler::ImmLLiteral(static_cast<int>(offset)); 390 Instr mask = ImmLLiteral_mask; 391 392 SetInstructionBits(Mask(~mask) | imm); 393} 394 395 396VectorFormat VectorFormatHalfWidth(VectorFormat vform) { 397 VIXL_ASSERT(vform == kFormat8H || vform == kFormat4S || vform == kFormat2D || 398 vform == kFormatH || vform == kFormatS || vform == kFormatD); 399 switch (vform) { 400 case kFormat8H: 401 return kFormat8B; 402 case kFormat4S: 403 return kFormat4H; 404 case kFormat2D: 405 return kFormat2S; 406 case kFormatH: 407 return kFormatB; 408 case kFormatS: 409 return kFormatH; 410 case kFormatD: 411 return kFormatS; 412 default: 413 VIXL_UNREACHABLE(); 414 return kFormatUndefined; 415 } 416} 417 418 419VectorFormat VectorFormatDoubleWidth(VectorFormat vform) { 420 VIXL_ASSERT(vform == kFormat8B || vform == kFormat4H || vform == kFormat2S || 421 vform == kFormatB || vform == kFormatH || vform == kFormatS); 422 switch (vform) { 423 case kFormat8B: 424 return kFormat8H; 425 case kFormat4H: 426 return kFormat4S; 427 case kFormat2S: 428 return kFormat2D; 429 case kFormatB: 430 return kFormatH; 431 case kFormatH: 432 return kFormatS; 433 case kFormatS: 434 return kFormatD; 435 default: 436 VIXL_UNREACHABLE(); 437 return kFormatUndefined; 438 } 439} 440 441 442VectorFormat VectorFormatFillQ(VectorFormat vform) { 443 switch (vform) { 444 case kFormatB: 445 case kFormat8B: 446 case kFormat16B: 447 return kFormat16B; 448 case kFormatH: 449 case kFormat4H: 450 case kFormat8H: 451 return kFormat8H; 452 case kFormatS: 453 case kFormat2S: 454 case kFormat4S: 455 return kFormat4S; 456 case kFormatD: 457 case kFormat1D: 458 case kFormat2D: 459 return kFormat2D; 460 default: 461 VIXL_UNREACHABLE(); 462 return kFormatUndefined; 463 } 464} 465 466VectorFormat VectorFormatHalfWidthDoubleLanes(VectorFormat vform) { 467 switch (vform) { 468 case kFormat4H: 469 return kFormat8B; 470 case kFormat8H: 471 return kFormat16B; 472 case kFormat2S: 473 return kFormat4H; 474 case kFormat4S: 475 return kFormat8H; 476 case kFormat1D: 477 return kFormat2S; 478 case kFormat2D: 479 return kFormat4S; 480 default: 481 VIXL_UNREACHABLE(); 482 return kFormatUndefined; 483 } 484} 485 486VectorFormat VectorFormatDoubleLanes(VectorFormat vform) { 487 VIXL_ASSERT(vform == kFormat8B || vform == kFormat4H || vform == kFormat2S); 488 switch (vform) { 489 case kFormat8B: 490 return kFormat16B; 491 case kFormat4H: 492 return kFormat8H; 493 case kFormat2S: 494 return kFormat4S; 495 default: 496 VIXL_UNREACHABLE(); 497 return kFormatUndefined; 498 } 499} 500 501 502VectorFormat VectorFormatHalfLanes(VectorFormat vform) { 503 VIXL_ASSERT(vform == kFormat16B || vform == kFormat8H || vform == kFormat4S); 504 switch (vform) { 505 case kFormat16B: 506 return kFormat8B; 507 case kFormat8H: 508 return kFormat4H; 509 case kFormat4S: 510 return kFormat2S; 511 default: 512 VIXL_UNREACHABLE(); 513 return kFormatUndefined; 514 } 515} 516 517 518VectorFormat ScalarFormatFromLaneSize(int laneSize) { 519 switch (laneSize) { 520 case 8: 521 return kFormatB; 522 case 16: 523 return kFormatH; 524 case 32: 525 return kFormatS; 526 case 64: 527 return kFormatD; 528 default: 529 VIXL_UNREACHABLE(); 530 return kFormatUndefined; 531 } 532} 533 534 535VectorFormat ScalarFormatFromFormat(VectorFormat vform) { 536 return ScalarFormatFromLaneSize(LaneSizeInBitsFromFormat(vform)); 537} 538 539 540unsigned RegisterSizeInBitsFromFormat(VectorFormat vform) { 541 VIXL_ASSERT(vform != kFormatUndefined); 542 switch (vform) { 543 case kFormatB: 544 return kBRegSize; 545 case kFormatH: 546 return kHRegSize; 547 case kFormatS: 548 return kSRegSize; 549 case kFormatD: 550 return kDRegSize; 551 case kFormat8B: 552 case kFormat4H: 553 case kFormat2S: 554 case kFormat1D: 555 return kDRegSize; 556 default: 557 return kQRegSize; 558 } 559} 560 561 562unsigned RegisterSizeInBytesFromFormat(VectorFormat vform) { 563 return RegisterSizeInBitsFromFormat(vform) / 8; 564} 565 566 567unsigned LaneSizeInBitsFromFormat(VectorFormat vform) { 568 VIXL_ASSERT(vform != kFormatUndefined); 569 switch (vform) { 570 case kFormatB: 571 case kFormat8B: 572 case kFormat16B: 573 return 8; 574 case kFormatH: 575 case kFormat4H: 576 case kFormat8H: 577 return 16; 578 case kFormatS: 579 case kFormat2S: 580 case kFormat4S: 581 return 32; 582 case kFormatD: 583 case kFormat1D: 584 case kFormat2D: 585 return 64; 586 default: 587 VIXL_UNREACHABLE(); 588 return 0; 589 } 590} 591 592 593int LaneSizeInBytesFromFormat(VectorFormat vform) { 594 return LaneSizeInBitsFromFormat(vform) / 8; 595} 596 597 598int LaneSizeInBytesLog2FromFormat(VectorFormat vform) { 599 VIXL_ASSERT(vform != kFormatUndefined); 600 switch (vform) { 601 case kFormatB: 602 case kFormat8B: 603 case kFormat16B: 604 return 0; 605 case kFormatH: 606 case kFormat4H: 607 case kFormat8H: 608 return 1; 609 case kFormatS: 610 case kFormat2S: 611 case kFormat4S: 612 return 2; 613 case kFormatD: 614 case kFormat1D: 615 case kFormat2D: 616 return 3; 617 default: 618 VIXL_UNREACHABLE(); 619 return 0; 620 } 621} 622 623 624int LaneCountFromFormat(VectorFormat vform) { 625 VIXL_ASSERT(vform != kFormatUndefined); 626 switch (vform) { 627 case kFormat16B: 628 return 16; 629 case kFormat8B: 630 case kFormat8H: 631 return 8; 632 case kFormat4H: 633 case kFormat4S: 634 return 4; 635 case kFormat2S: 636 case kFormat2D: 637 return 2; 638 case kFormat1D: 639 case kFormatB: 640 case kFormatH: 641 case kFormatS: 642 case kFormatD: 643 return 1; 644 default: 645 VIXL_UNREACHABLE(); 646 return 0; 647 } 648} 649 650 651int MaxLaneCountFromFormat(VectorFormat vform) { 652 VIXL_ASSERT(vform != kFormatUndefined); 653 switch (vform) { 654 case kFormatB: 655 case kFormat8B: 656 case kFormat16B: 657 return 16; 658 case kFormatH: 659 case kFormat4H: 660 case kFormat8H: 661 return 8; 662 case kFormatS: 663 case kFormat2S: 664 case kFormat4S: 665 return 4; 666 case kFormatD: 667 case kFormat1D: 668 case kFormat2D: 669 return 2; 670 default: 671 VIXL_UNREACHABLE(); 672 return 0; 673 } 674} 675 676 677// Does 'vform' indicate a vector format or a scalar format? 678bool IsVectorFormat(VectorFormat vform) { 679 VIXL_ASSERT(vform != kFormatUndefined); 680 switch (vform) { 681 case kFormatB: 682 case kFormatH: 683 case kFormatS: 684 case kFormatD: 685 return false; 686 default: 687 return true; 688 } 689} 690 691 692int64_t MaxIntFromFormat(VectorFormat vform) { 693 return INT64_MAX >> (64 - LaneSizeInBitsFromFormat(vform)); 694} 695 696 697int64_t MinIntFromFormat(VectorFormat vform) { 698 return INT64_MIN >> (64 - LaneSizeInBitsFromFormat(vform)); 699} 700 701 702uint64_t MaxUintFromFormat(VectorFormat vform) { 703 return UINT64_MAX >> (64 - LaneSizeInBitsFromFormat(vform)); 704} 705} // namespace aarch64 706} // namespace vixl 707