1 //===-- X86AsmParser.cpp - Parse X86 assembly to MCInst instructions ------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 10 #include "MCTargetDesc/X86BaseInfo.h" 11 #include "X86AsmInstrumentation.h" 12 #include "X86AsmParserCommon.h" 13 #include "X86Operand.h" 14 #include "X86ISelLowering.h" 15 #include "llvm/ADT/APFloat.h" 16 #include "llvm/ADT/STLExtras.h" 17 #include "llvm/ADT/SmallString.h" 18 #include "llvm/ADT/SmallVector.h" 19 #include "llvm/ADT/StringSwitch.h" 20 #include "llvm/ADT/Twine.h" 21 #include "llvm/MC/MCContext.h" 22 #include "llvm/MC/MCExpr.h" 23 #include "llvm/MC/MCInst.h" 24 #include "llvm/MC/MCInstrInfo.h" 25 #include "llvm/MC/MCParser/MCAsmLexer.h" 26 #include "llvm/MC/MCParser/MCAsmParser.h" 27 #include "llvm/MC/MCParser/MCParsedAsmOperand.h" 28 #include "llvm/MC/MCRegisterInfo.h" 29 #include "llvm/MC/MCStreamer.h" 30 #include "llvm/MC/MCSubtargetInfo.h" 31 #include "llvm/MC/MCSymbol.h" 32 #include "llvm/MC/MCTargetAsmParser.h" 33 #include "llvm/Support/SourceMgr.h" 34 #include "llvm/Support/TargetRegistry.h" 35 #include "llvm/Support/raw_ostream.h" 36 #include <algorithm> 37 #include <memory> 38 39 using namespace llvm; 40 41 namespace { 42 43 static const char OpPrecedence[] = { 44 0, // IC_OR 45 1, // IC_XOR 46 2, // IC_AND 47 3, // IC_LSHIFT 48 3, // IC_RSHIFT 49 4, // IC_PLUS 50 4, // IC_MINUS 51 5, // IC_MULTIPLY 52 5, // IC_DIVIDE 53 6, // IC_RPAREN 54 7, // IC_LPAREN 55 0, // IC_IMM 56 0 // IC_REGISTER 57 }; 58 59 class X86AsmParser : public MCTargetAsmParser { 60 const MCInstrInfo &MII; 61 ParseInstructionInfo *InstInfo; 62 std::unique_ptr<X86AsmInstrumentation> Instrumentation; 63 64 private: 65 SMLoc consumeToken() { 66 MCAsmParser &Parser = getParser(); 67 SMLoc Result = Parser.getTok().getLoc(); 68 Parser.Lex(); 69 return Result; 70 } 71 72 enum InfixCalculatorTok { 73 IC_OR = 0, 74 IC_XOR, 75 IC_AND, 76 IC_LSHIFT, 77 IC_RSHIFT, 78 IC_PLUS, 79 IC_MINUS, 80 IC_MULTIPLY, 81 IC_DIVIDE, 82 IC_RPAREN, 83 IC_LPAREN, 84 IC_IMM, 85 IC_REGISTER 86 }; 87 88 class InfixCalculator { 89 typedef std::pair< InfixCalculatorTok, int64_t > ICToken; 90 SmallVector<InfixCalculatorTok, 4> InfixOperatorStack; 91 SmallVector<ICToken, 4> PostfixStack; 92 93 public: 94 int64_t popOperand() { 95 assert (!PostfixStack.empty() && "Poped an empty stack!"); 96 ICToken Op = PostfixStack.pop_back_val(); 97 assert ((Op.first == IC_IMM || Op.first == IC_REGISTER) 98 && "Expected and immediate or register!"); 99 return Op.second; 100 } 101 void pushOperand(InfixCalculatorTok Op, int64_t Val = 0) { 102 assert ((Op == IC_IMM || Op == IC_REGISTER) && 103 "Unexpected operand!"); 104 PostfixStack.push_back(std::make_pair(Op, Val)); 105 } 106 107 void popOperator() { InfixOperatorStack.pop_back(); } 108 void pushOperator(InfixCalculatorTok Op) { 109 // Push the new operator if the stack is empty. 110 if (InfixOperatorStack.empty()) { 111 InfixOperatorStack.push_back(Op); 112 return; 113 } 114 115 // Push the new operator if it has a higher precedence than the operator 116 // on the top of the stack or the operator on the top of the stack is a 117 // left parentheses. 118 unsigned Idx = InfixOperatorStack.size() - 1; 119 InfixCalculatorTok StackOp = InfixOperatorStack[Idx]; 120 if (OpPrecedence[Op] > OpPrecedence[StackOp] || StackOp == IC_LPAREN) { 121 InfixOperatorStack.push_back(Op); 122 return; 123 } 124 125 // The operator on the top of the stack has higher precedence than the 126 // new operator. 127 unsigned ParenCount = 0; 128 while (1) { 129 // Nothing to process. 130 if (InfixOperatorStack.empty()) 131 break; 132 133 Idx = InfixOperatorStack.size() - 1; 134 StackOp = InfixOperatorStack[Idx]; 135 if (!(OpPrecedence[StackOp] >= OpPrecedence[Op] || ParenCount)) 136 break; 137 138 // If we have an even parentheses count and we see a left parentheses, 139 // then stop processing. 140 if (!ParenCount && StackOp == IC_LPAREN) 141 break; 142 143 if (StackOp == IC_RPAREN) { 144 ++ParenCount; 145 InfixOperatorStack.pop_back(); 146 } else if (StackOp == IC_LPAREN) { 147 --ParenCount; 148 InfixOperatorStack.pop_back(); 149 } else { 150 InfixOperatorStack.pop_back(); 151 PostfixStack.push_back(std::make_pair(StackOp, 0)); 152 } 153 } 154 // Push the new operator. 155 InfixOperatorStack.push_back(Op); 156 } 157 158 int64_t execute() { 159 // Push any remaining operators onto the postfix stack. 160 while (!InfixOperatorStack.empty()) { 161 InfixCalculatorTok StackOp = InfixOperatorStack.pop_back_val(); 162 if (StackOp != IC_LPAREN && StackOp != IC_RPAREN) 163 PostfixStack.push_back(std::make_pair(StackOp, 0)); 164 } 165 166 if (PostfixStack.empty()) 167 return 0; 168 169 SmallVector<ICToken, 16> OperandStack; 170 for (unsigned i = 0, e = PostfixStack.size(); i != e; ++i) { 171 ICToken Op = PostfixStack[i]; 172 if (Op.first == IC_IMM || Op.first == IC_REGISTER) { 173 OperandStack.push_back(Op); 174 } else { 175 assert (OperandStack.size() > 1 && "Too few operands."); 176 int64_t Val; 177 ICToken Op2 = OperandStack.pop_back_val(); 178 ICToken Op1 = OperandStack.pop_back_val(); 179 switch (Op.first) { 180 default: 181 report_fatal_error("Unexpected operator!"); 182 break; 183 case IC_PLUS: 184 Val = Op1.second + Op2.second; 185 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 186 break; 187 case IC_MINUS: 188 Val = Op1.second - Op2.second; 189 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 190 break; 191 case IC_MULTIPLY: 192 assert (Op1.first == IC_IMM && Op2.first == IC_IMM && 193 "Multiply operation with an immediate and a register!"); 194 Val = Op1.second * Op2.second; 195 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 196 break; 197 case IC_DIVIDE: 198 assert (Op1.first == IC_IMM && Op2.first == IC_IMM && 199 "Divide operation with an immediate and a register!"); 200 assert (Op2.second != 0 && "Division by zero!"); 201 Val = Op1.second / Op2.second; 202 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 203 break; 204 case IC_OR: 205 assert (Op1.first == IC_IMM && Op2.first == IC_IMM && 206 "Or operation with an immediate and a register!"); 207 Val = Op1.second | Op2.second; 208 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 209 break; 210 case IC_XOR: 211 assert(Op1.first == IC_IMM && Op2.first == IC_IMM && 212 "Xor operation with an immediate and a register!"); 213 Val = Op1.second ^ Op2.second; 214 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 215 break; 216 case IC_AND: 217 assert (Op1.first == IC_IMM && Op2.first == IC_IMM && 218 "And operation with an immediate and a register!"); 219 Val = Op1.second & Op2.second; 220 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 221 break; 222 case IC_LSHIFT: 223 assert (Op1.first == IC_IMM && Op2.first == IC_IMM && 224 "Left shift operation with an immediate and a register!"); 225 Val = Op1.second << Op2.second; 226 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 227 break; 228 case IC_RSHIFT: 229 assert (Op1.first == IC_IMM && Op2.first == IC_IMM && 230 "Right shift operation with an immediate and a register!"); 231 Val = Op1.second >> Op2.second; 232 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 233 break; 234 } 235 } 236 } 237 assert (OperandStack.size() == 1 && "Expected a single result."); 238 return OperandStack.pop_back_val().second; 239 } 240 }; 241 242 enum IntelExprState { 243 IES_OR, 244 IES_XOR, 245 IES_AND, 246 IES_LSHIFT, 247 IES_RSHIFT, 248 IES_PLUS, 249 IES_MINUS, 250 IES_NOT, 251 IES_MULTIPLY, 252 IES_DIVIDE, 253 IES_LBRAC, 254 IES_RBRAC, 255 IES_LPAREN, 256 IES_RPAREN, 257 IES_REGISTER, 258 IES_INTEGER, 259 IES_IDENTIFIER, 260 IES_ERROR 261 }; 262 263 class IntelExprStateMachine { 264 IntelExprState State, PrevState; 265 unsigned BaseReg, IndexReg, TmpReg, Scale; 266 int64_t Imm; 267 const MCExpr *Sym; 268 StringRef SymName; 269 bool StopOnLBrac, AddImmPrefix; 270 InfixCalculator IC; 271 InlineAsmIdentifierInfo Info; 272 273 public: 274 IntelExprStateMachine(int64_t imm, bool stoponlbrac, bool addimmprefix) : 275 State(IES_PLUS), PrevState(IES_ERROR), BaseReg(0), IndexReg(0), TmpReg(0), 276 Scale(1), Imm(imm), Sym(nullptr), StopOnLBrac(stoponlbrac), 277 AddImmPrefix(addimmprefix) { Info.clear(); } 278 279 unsigned getBaseReg() { return BaseReg; } 280 unsigned getIndexReg() { return IndexReg; } 281 unsigned getScale() { return Scale; } 282 const MCExpr *getSym() { return Sym; } 283 StringRef getSymName() { return SymName; } 284 int64_t getImm() { return Imm + IC.execute(); } 285 bool isValidEndState() { 286 return State == IES_RBRAC || State == IES_INTEGER; 287 } 288 bool getStopOnLBrac() { return StopOnLBrac; } 289 bool getAddImmPrefix() { return AddImmPrefix; } 290 bool hadError() { return State == IES_ERROR; } 291 292 InlineAsmIdentifierInfo &getIdentifierInfo() { 293 return Info; 294 } 295 296 void onOr() { 297 IntelExprState CurrState = State; 298 switch (State) { 299 default: 300 State = IES_ERROR; 301 break; 302 case IES_INTEGER: 303 case IES_RPAREN: 304 case IES_REGISTER: 305 State = IES_OR; 306 IC.pushOperator(IC_OR); 307 break; 308 } 309 PrevState = CurrState; 310 } 311 void onXor() { 312 IntelExprState CurrState = State; 313 switch (State) { 314 default: 315 State = IES_ERROR; 316 break; 317 case IES_INTEGER: 318 case IES_RPAREN: 319 case IES_REGISTER: 320 State = IES_XOR; 321 IC.pushOperator(IC_XOR); 322 break; 323 } 324 PrevState = CurrState; 325 } 326 void onAnd() { 327 IntelExprState CurrState = State; 328 switch (State) { 329 default: 330 State = IES_ERROR; 331 break; 332 case IES_INTEGER: 333 case IES_RPAREN: 334 case IES_REGISTER: 335 State = IES_AND; 336 IC.pushOperator(IC_AND); 337 break; 338 } 339 PrevState = CurrState; 340 } 341 void onLShift() { 342 IntelExprState CurrState = State; 343 switch (State) { 344 default: 345 State = IES_ERROR; 346 break; 347 case IES_INTEGER: 348 case IES_RPAREN: 349 case IES_REGISTER: 350 State = IES_LSHIFT; 351 IC.pushOperator(IC_LSHIFT); 352 break; 353 } 354 PrevState = CurrState; 355 } 356 void onRShift() { 357 IntelExprState CurrState = State; 358 switch (State) { 359 default: 360 State = IES_ERROR; 361 break; 362 case IES_INTEGER: 363 case IES_RPAREN: 364 case IES_REGISTER: 365 State = IES_RSHIFT; 366 IC.pushOperator(IC_RSHIFT); 367 break; 368 } 369 PrevState = CurrState; 370 } 371 void onPlus() { 372 IntelExprState CurrState = State; 373 switch (State) { 374 default: 375 State = IES_ERROR; 376 break; 377 case IES_INTEGER: 378 case IES_RPAREN: 379 case IES_REGISTER: 380 State = IES_PLUS; 381 IC.pushOperator(IC_PLUS); 382 if (CurrState == IES_REGISTER && PrevState != IES_MULTIPLY) { 383 // If we already have a BaseReg, then assume this is the IndexReg with 384 // a scale of 1. 385 if (!BaseReg) { 386 BaseReg = TmpReg; 387 } else { 388 assert (!IndexReg && "BaseReg/IndexReg already set!"); 389 IndexReg = TmpReg; 390 Scale = 1; 391 } 392 } 393 break; 394 } 395 PrevState = CurrState; 396 } 397 void onMinus() { 398 IntelExprState CurrState = State; 399 switch (State) { 400 default: 401 State = IES_ERROR; 402 break; 403 case IES_PLUS: 404 case IES_NOT: 405 case IES_MULTIPLY: 406 case IES_DIVIDE: 407 case IES_LPAREN: 408 case IES_RPAREN: 409 case IES_LBRAC: 410 case IES_RBRAC: 411 case IES_INTEGER: 412 case IES_REGISTER: 413 State = IES_MINUS; 414 // Only push the minus operator if it is not a unary operator. 415 if (!(CurrState == IES_PLUS || CurrState == IES_MINUS || 416 CurrState == IES_MULTIPLY || CurrState == IES_DIVIDE || 417 CurrState == IES_LPAREN || CurrState == IES_LBRAC)) 418 IC.pushOperator(IC_MINUS); 419 if (CurrState == IES_REGISTER && PrevState != IES_MULTIPLY) { 420 // If we already have a BaseReg, then assume this is the IndexReg with 421 // a scale of 1. 422 if (!BaseReg) { 423 BaseReg = TmpReg; 424 } else { 425 assert (!IndexReg && "BaseReg/IndexReg already set!"); 426 IndexReg = TmpReg; 427 Scale = 1; 428 } 429 } 430 break; 431 } 432 PrevState = CurrState; 433 } 434 void onNot() { 435 IntelExprState CurrState = State; 436 switch (State) { 437 default: 438 State = IES_ERROR; 439 break; 440 case IES_PLUS: 441 case IES_NOT: 442 State = IES_NOT; 443 break; 444 } 445 PrevState = CurrState; 446 } 447 void onRegister(unsigned Reg) { 448 IntelExprState CurrState = State; 449 switch (State) { 450 default: 451 State = IES_ERROR; 452 break; 453 case IES_PLUS: 454 case IES_LPAREN: 455 State = IES_REGISTER; 456 TmpReg = Reg; 457 IC.pushOperand(IC_REGISTER); 458 break; 459 case IES_MULTIPLY: 460 // Index Register - Scale * Register 461 if (PrevState == IES_INTEGER) { 462 assert (!IndexReg && "IndexReg already set!"); 463 State = IES_REGISTER; 464 IndexReg = Reg; 465 // Get the scale and replace the 'Scale * Register' with '0'. 466 Scale = IC.popOperand(); 467 IC.pushOperand(IC_IMM); 468 IC.popOperator(); 469 } else { 470 State = IES_ERROR; 471 } 472 break; 473 } 474 PrevState = CurrState; 475 } 476 void onIdentifierExpr(const MCExpr *SymRef, StringRef SymRefName) { 477 PrevState = State; 478 switch (State) { 479 default: 480 State = IES_ERROR; 481 break; 482 case IES_PLUS: 483 case IES_MINUS: 484 case IES_NOT: 485 State = IES_INTEGER; 486 Sym = SymRef; 487 SymName = SymRefName; 488 IC.pushOperand(IC_IMM); 489 break; 490 } 491 } 492 bool onInteger(int64_t TmpInt, StringRef &ErrMsg) { 493 IntelExprState CurrState = State; 494 switch (State) { 495 default: 496 State = IES_ERROR; 497 break; 498 case IES_PLUS: 499 case IES_MINUS: 500 case IES_NOT: 501 case IES_OR: 502 case IES_XOR: 503 case IES_AND: 504 case IES_LSHIFT: 505 case IES_RSHIFT: 506 case IES_DIVIDE: 507 case IES_MULTIPLY: 508 case IES_LPAREN: 509 State = IES_INTEGER; 510 if (PrevState == IES_REGISTER && CurrState == IES_MULTIPLY) { 511 // Index Register - Register * Scale 512 assert (!IndexReg && "IndexReg already set!"); 513 IndexReg = TmpReg; 514 Scale = TmpInt; 515 if(Scale != 1 && Scale != 2 && Scale != 4 && Scale != 8) { 516 ErrMsg = "scale factor in address must be 1, 2, 4 or 8"; 517 return true; 518 } 519 // Get the scale and replace the 'Register * Scale' with '0'. 520 IC.popOperator(); 521 } else if ((PrevState == IES_PLUS || PrevState == IES_MINUS || 522 PrevState == IES_OR || PrevState == IES_AND || 523 PrevState == IES_LSHIFT || PrevState == IES_RSHIFT || 524 PrevState == IES_MULTIPLY || PrevState == IES_DIVIDE || 525 PrevState == IES_LPAREN || PrevState == IES_LBRAC || 526 PrevState == IES_NOT || PrevState == IES_XOR) && 527 CurrState == IES_MINUS) { 528 // Unary minus. No need to pop the minus operand because it was never 529 // pushed. 530 IC.pushOperand(IC_IMM, -TmpInt); // Push -Imm. 531 } else if ((PrevState == IES_PLUS || PrevState == IES_MINUS || 532 PrevState == IES_OR || PrevState == IES_AND || 533 PrevState == IES_LSHIFT || PrevState == IES_RSHIFT || 534 PrevState == IES_MULTIPLY || PrevState == IES_DIVIDE || 535 PrevState == IES_LPAREN || PrevState == IES_LBRAC || 536 PrevState == IES_NOT || PrevState == IES_XOR) && 537 CurrState == IES_NOT) { 538 // Unary not. No need to pop the not operand because it was never 539 // pushed. 540 IC.pushOperand(IC_IMM, ~TmpInt); // Push ~Imm. 541 } else { 542 IC.pushOperand(IC_IMM, TmpInt); 543 } 544 break; 545 } 546 PrevState = CurrState; 547 return false; 548 } 549 void onStar() { 550 PrevState = State; 551 switch (State) { 552 default: 553 State = IES_ERROR; 554 break; 555 case IES_INTEGER: 556 case IES_REGISTER: 557 case IES_RPAREN: 558 State = IES_MULTIPLY; 559 IC.pushOperator(IC_MULTIPLY); 560 break; 561 } 562 } 563 void onDivide() { 564 PrevState = State; 565 switch (State) { 566 default: 567 State = IES_ERROR; 568 break; 569 case IES_INTEGER: 570 case IES_RPAREN: 571 State = IES_DIVIDE; 572 IC.pushOperator(IC_DIVIDE); 573 break; 574 } 575 } 576 void onLBrac() { 577 PrevState = State; 578 switch (State) { 579 default: 580 State = IES_ERROR; 581 break; 582 case IES_RBRAC: 583 State = IES_PLUS; 584 IC.pushOperator(IC_PLUS); 585 break; 586 } 587 } 588 void onRBrac() { 589 IntelExprState CurrState = State; 590 switch (State) { 591 default: 592 State = IES_ERROR; 593 break; 594 case IES_INTEGER: 595 case IES_REGISTER: 596 case IES_RPAREN: 597 State = IES_RBRAC; 598 if (CurrState == IES_REGISTER && PrevState != IES_MULTIPLY) { 599 // If we already have a BaseReg, then assume this is the IndexReg with 600 // a scale of 1. 601 if (!BaseReg) { 602 BaseReg = TmpReg; 603 } else { 604 assert (!IndexReg && "BaseReg/IndexReg already set!"); 605 IndexReg = TmpReg; 606 Scale = 1; 607 } 608 } 609 break; 610 } 611 PrevState = CurrState; 612 } 613 void onLParen() { 614 IntelExprState CurrState = State; 615 switch (State) { 616 default: 617 State = IES_ERROR; 618 break; 619 case IES_PLUS: 620 case IES_MINUS: 621 case IES_NOT: 622 case IES_OR: 623 case IES_XOR: 624 case IES_AND: 625 case IES_LSHIFT: 626 case IES_RSHIFT: 627 case IES_MULTIPLY: 628 case IES_DIVIDE: 629 case IES_LPAREN: 630 // FIXME: We don't handle this type of unary minus or not, yet. 631 if ((PrevState == IES_PLUS || PrevState == IES_MINUS || 632 PrevState == IES_OR || PrevState == IES_AND || 633 PrevState == IES_LSHIFT || PrevState == IES_RSHIFT || 634 PrevState == IES_MULTIPLY || PrevState == IES_DIVIDE || 635 PrevState == IES_LPAREN || PrevState == IES_LBRAC || 636 PrevState == IES_NOT || PrevState == IES_XOR) && 637 (CurrState == IES_MINUS || CurrState == IES_NOT)) { 638 State = IES_ERROR; 639 break; 640 } 641 State = IES_LPAREN; 642 IC.pushOperator(IC_LPAREN); 643 break; 644 } 645 PrevState = CurrState; 646 } 647 void onRParen() { 648 PrevState = State; 649 switch (State) { 650 default: 651 State = IES_ERROR; 652 break; 653 case IES_INTEGER: 654 case IES_REGISTER: 655 case IES_RPAREN: 656 State = IES_RPAREN; 657 IC.pushOperator(IC_RPAREN); 658 break; 659 } 660 } 661 }; 662 663 bool Error(SMLoc L, const Twine &Msg, 664 ArrayRef<SMRange> Ranges = None, 665 bool MatchingInlineAsm = false) { 666 MCAsmParser &Parser = getParser(); 667 if (MatchingInlineAsm) return true; 668 return Parser.Error(L, Msg, Ranges); 669 } 670 671 bool ErrorAndEatStatement(SMLoc L, const Twine &Msg, 672 ArrayRef<SMRange> Ranges = None, 673 bool MatchingInlineAsm = false) { 674 MCAsmParser &Parser = getParser(); 675 Parser.eatToEndOfStatement(); 676 return Error(L, Msg, Ranges, MatchingInlineAsm); 677 } 678 679 std::nullptr_t ErrorOperand(SMLoc Loc, StringRef Msg) { 680 Error(Loc, Msg); 681 return nullptr; 682 } 683 684 std::unique_ptr<X86Operand> DefaultMemSIOperand(SMLoc Loc); 685 std::unique_ptr<X86Operand> DefaultMemDIOperand(SMLoc Loc); 686 void AddDefaultSrcDestOperands( 687 OperandVector& Operands, std::unique_ptr<llvm::MCParsedAsmOperand> &&Src, 688 std::unique_ptr<llvm::MCParsedAsmOperand> &&Dst); 689 std::unique_ptr<X86Operand> ParseOperand(); 690 std::unique_ptr<X86Operand> ParseATTOperand(); 691 std::unique_ptr<X86Operand> ParseIntelOperand(); 692 std::unique_ptr<X86Operand> ParseIntelOffsetOfOperator(); 693 bool ParseIntelDotOperator(const MCExpr *Disp, const MCExpr *&NewDisp); 694 std::unique_ptr<X86Operand> ParseIntelOperator(unsigned OpKind); 695 std::unique_ptr<X86Operand> 696 ParseIntelSegmentOverride(unsigned SegReg, SMLoc Start, unsigned Size); 697 std::unique_ptr<X86Operand> 698 ParseIntelMemOperand(int64_t ImmDisp, SMLoc StartLoc, unsigned Size); 699 std::unique_ptr<X86Operand> ParseRoundingModeOp(SMLoc Start, SMLoc End); 700 bool ParseIntelExpression(IntelExprStateMachine &SM, SMLoc &End); 701 std::unique_ptr<X86Operand> ParseIntelBracExpression(unsigned SegReg, 702 SMLoc Start, 703 int64_t ImmDisp, 704 unsigned Size); 705 bool ParseIntelIdentifier(const MCExpr *&Val, StringRef &Identifier, 706 InlineAsmIdentifierInfo &Info, 707 bool IsUnevaluatedOperand, SMLoc &End); 708 709 std::unique_ptr<X86Operand> ParseMemOperand(unsigned SegReg, SMLoc StartLoc); 710 711 std::unique_ptr<X86Operand> 712 CreateMemForInlineAsm(unsigned SegReg, const MCExpr *Disp, unsigned BaseReg, 713 unsigned IndexReg, unsigned Scale, SMLoc Start, 714 SMLoc End, unsigned Size, StringRef Identifier, 715 InlineAsmIdentifierInfo &Info); 716 717 bool ParseDirectiveWord(unsigned Size, SMLoc L); 718 bool ParseDirectiveCode(StringRef IDVal, SMLoc L); 719 720 bool processInstruction(MCInst &Inst, const OperandVector &Ops); 721 722 /// Wrapper around MCStreamer::EmitInstruction(). Possibly adds 723 /// instrumentation around Inst. 724 void EmitInstruction(MCInst &Inst, OperandVector &Operands, MCStreamer &Out); 725 726 bool MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode, 727 OperandVector &Operands, MCStreamer &Out, 728 uint64_t &ErrorInfo, 729 bool MatchingInlineAsm) override; 730 731 void MatchFPUWaitAlias(SMLoc IDLoc, X86Operand &Op, OperandVector &Operands, 732 MCStreamer &Out, bool MatchingInlineAsm); 733 734 bool ErrorMissingFeature(SMLoc IDLoc, uint64_t ErrorInfo, 735 bool MatchingInlineAsm); 736 737 bool MatchAndEmitATTInstruction(SMLoc IDLoc, unsigned &Opcode, 738 OperandVector &Operands, MCStreamer &Out, 739 uint64_t &ErrorInfo, 740 bool MatchingInlineAsm); 741 742 bool MatchAndEmitIntelInstruction(SMLoc IDLoc, unsigned &Opcode, 743 OperandVector &Operands, MCStreamer &Out, 744 uint64_t &ErrorInfo, 745 bool MatchingInlineAsm); 746 747 bool OmitRegisterFromClobberLists(unsigned RegNo) override; 748 749 /// doSrcDstMatch - Returns true if operands are matching in their 750 /// word size (%si and %di, %esi and %edi, etc.). Order depends on 751 /// the parsing mode (Intel vs. AT&T). 752 bool doSrcDstMatch(X86Operand &Op1, X86Operand &Op2); 753 754 /// Parses AVX512 specific operand primitives: masked registers ({%k<NUM>}, {z}) 755 /// and memory broadcasting ({1to<NUM>}) primitives, updating Operands vector if required. 756 /// \return \c true if no parsing errors occurred, \c false otherwise. 757 bool HandleAVX512Operand(OperandVector &Operands, 758 const MCParsedAsmOperand &Op); 759 760 bool is64BitMode() const { 761 // FIXME: Can tablegen auto-generate this? 762 return getSTI().getFeatureBits()[X86::Mode64Bit]; 763 } 764 bool is32BitMode() const { 765 // FIXME: Can tablegen auto-generate this? 766 return getSTI().getFeatureBits()[X86::Mode32Bit]; 767 } 768 bool is16BitMode() const { 769 // FIXME: Can tablegen auto-generate this? 770 return getSTI().getFeatureBits()[X86::Mode16Bit]; 771 } 772 void SwitchMode(unsigned mode) { 773 MCSubtargetInfo &STI = copySTI(); 774 FeatureBitset AllModes({X86::Mode64Bit, X86::Mode32Bit, X86::Mode16Bit}); 775 FeatureBitset OldMode = STI.getFeatureBits() & AllModes; 776 unsigned FB = ComputeAvailableFeatures( 777 STI.ToggleFeature(OldMode.flip(mode))); 778 setAvailableFeatures(FB); 779 780 assert(FeatureBitset({mode}) == (STI.getFeatureBits() & AllModes)); 781 } 782 783 unsigned getPointerWidth() { 784 if (is16BitMode()) return 16; 785 if (is32BitMode()) return 32; 786 if (is64BitMode()) return 64; 787 llvm_unreachable("invalid mode"); 788 } 789 790 bool isParsingIntelSyntax() { 791 return getParser().getAssemblerDialect(); 792 } 793 794 /// @name Auto-generated Matcher Functions 795 /// { 796 797 #define GET_ASSEMBLER_HEADER 798 #include "X86GenAsmMatcher.inc" 799 800 /// } 801 802 public: 803 X86AsmParser(const MCSubtargetInfo &sti, MCAsmParser &Parser, 804 const MCInstrInfo &mii, const MCTargetOptions &Options) 805 : MCTargetAsmParser(Options, sti), MII(mii), InstInfo(nullptr) { 806 807 // Initialize the set of available features. 808 setAvailableFeatures(ComputeAvailableFeatures(getSTI().getFeatureBits())); 809 Instrumentation.reset( 810 CreateX86AsmInstrumentation(Options, Parser.getContext(), STI)); 811 } 812 813 bool ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &EndLoc) override; 814 815 void SetFrameRegister(unsigned RegNo) override; 816 817 bool ParseInstruction(ParseInstructionInfo &Info, StringRef Name, 818 SMLoc NameLoc, OperandVector &Operands) override; 819 820 bool ParseDirective(AsmToken DirectiveID) override; 821 }; 822 } // end anonymous namespace 823 824 /// @name Auto-generated Match Functions 825 /// { 826 827 static unsigned MatchRegisterName(StringRef Name); 828 829 /// } 830 831 static bool CheckBaseRegAndIndexReg(unsigned BaseReg, unsigned IndexReg, 832 StringRef &ErrMsg) { 833 // If we have both a base register and an index register make sure they are 834 // both 64-bit or 32-bit registers. 835 // To support VSIB, IndexReg can be 128-bit or 256-bit registers. 836 if (BaseReg != 0 && IndexReg != 0) { 837 if (X86MCRegisterClasses[X86::GR64RegClassID].contains(BaseReg) && 838 (X86MCRegisterClasses[X86::GR16RegClassID].contains(IndexReg) || 839 X86MCRegisterClasses[X86::GR32RegClassID].contains(IndexReg)) && 840 IndexReg != X86::RIZ) { 841 ErrMsg = "base register is 64-bit, but index register is not"; 842 return true; 843 } 844 if (X86MCRegisterClasses[X86::GR32RegClassID].contains(BaseReg) && 845 (X86MCRegisterClasses[X86::GR16RegClassID].contains(IndexReg) || 846 X86MCRegisterClasses[X86::GR64RegClassID].contains(IndexReg)) && 847 IndexReg != X86::EIZ){ 848 ErrMsg = "base register is 32-bit, but index register is not"; 849 return true; 850 } 851 if (X86MCRegisterClasses[X86::GR16RegClassID].contains(BaseReg)) { 852 if (X86MCRegisterClasses[X86::GR32RegClassID].contains(IndexReg) || 853 X86MCRegisterClasses[X86::GR64RegClassID].contains(IndexReg)) { 854 ErrMsg = "base register is 16-bit, but index register is not"; 855 return true; 856 } 857 if (((BaseReg == X86::BX || BaseReg == X86::BP) && 858 IndexReg != X86::SI && IndexReg != X86::DI) || 859 ((BaseReg == X86::SI || BaseReg == X86::DI) && 860 IndexReg != X86::BX && IndexReg != X86::BP)) { 861 ErrMsg = "invalid 16-bit base/index register combination"; 862 return true; 863 } 864 } 865 } 866 return false; 867 } 868 869 bool X86AsmParser::doSrcDstMatch(X86Operand &Op1, X86Operand &Op2) 870 { 871 // Return true and let a normal complaint about bogus operands happen. 872 if (!Op1.isMem() || !Op2.isMem()) 873 return true; 874 875 // Actually these might be the other way round if Intel syntax is 876 // being used. It doesn't matter. 877 unsigned diReg = Op1.Mem.BaseReg; 878 unsigned siReg = Op2.Mem.BaseReg; 879 880 if (X86MCRegisterClasses[X86::GR16RegClassID].contains(siReg)) 881 return X86MCRegisterClasses[X86::GR16RegClassID].contains(diReg); 882 if (X86MCRegisterClasses[X86::GR32RegClassID].contains(siReg)) 883 return X86MCRegisterClasses[X86::GR32RegClassID].contains(diReg); 884 if (X86MCRegisterClasses[X86::GR64RegClassID].contains(siReg)) 885 return X86MCRegisterClasses[X86::GR64RegClassID].contains(diReg); 886 // Again, return true and let another error happen. 887 return true; 888 } 889 890 bool X86AsmParser::ParseRegister(unsigned &RegNo, 891 SMLoc &StartLoc, SMLoc &EndLoc) { 892 MCAsmParser &Parser = getParser(); 893 RegNo = 0; 894 const AsmToken &PercentTok = Parser.getTok(); 895 StartLoc = PercentTok.getLoc(); 896 897 // If we encounter a %, ignore it. This code handles registers with and 898 // without the prefix, unprefixed registers can occur in cfi directives. 899 if (!isParsingIntelSyntax() && PercentTok.is(AsmToken::Percent)) 900 Parser.Lex(); // Eat percent token. 901 902 const AsmToken &Tok = Parser.getTok(); 903 EndLoc = Tok.getEndLoc(); 904 905 if (Tok.isNot(AsmToken::Identifier)) { 906 if (isParsingIntelSyntax()) return true; 907 return Error(StartLoc, "invalid register name", 908 SMRange(StartLoc, EndLoc)); 909 } 910 911 RegNo = MatchRegisterName(Tok.getString()); 912 913 // If the match failed, try the register name as lowercase. 914 if (RegNo == 0) 915 RegNo = MatchRegisterName(Tok.getString().lower()); 916 917 // The "flags" register cannot be referenced directly. 918 // Treat it as an identifier instead. 919 if (isParsingInlineAsm() && isParsingIntelSyntax() && RegNo == X86::EFLAGS) 920 RegNo = 0; 921 922 if (!is64BitMode()) { 923 // FIXME: This should be done using Requires<Not64BitMode> and 924 // Requires<In64BitMode> so "eiz" usage in 64-bit instructions can be also 925 // checked. 926 // FIXME: Check AH, CH, DH, BH cannot be used in an instruction requiring a 927 // REX prefix. 928 if (RegNo == X86::RIZ || 929 X86MCRegisterClasses[X86::GR64RegClassID].contains(RegNo) || 930 X86II::isX86_64NonExtLowByteReg(RegNo) || 931 X86II::isX86_64ExtendedReg(RegNo)) 932 return Error(StartLoc, "register %" 933 + Tok.getString() + " is only available in 64-bit mode", 934 SMRange(StartLoc, EndLoc)); 935 } 936 937 // Parse "%st" as "%st(0)" and "%st(1)", which is multiple tokens. 938 if (RegNo == 0 && (Tok.getString() == "st" || Tok.getString() == "ST")) { 939 RegNo = X86::ST0; 940 Parser.Lex(); // Eat 'st' 941 942 // Check to see if we have '(4)' after %st. 943 if (getLexer().isNot(AsmToken::LParen)) 944 return false; 945 // Lex the paren. 946 getParser().Lex(); 947 948 const AsmToken &IntTok = Parser.getTok(); 949 if (IntTok.isNot(AsmToken::Integer)) 950 return Error(IntTok.getLoc(), "expected stack index"); 951 switch (IntTok.getIntVal()) { 952 case 0: RegNo = X86::ST0; break; 953 case 1: RegNo = X86::ST1; break; 954 case 2: RegNo = X86::ST2; break; 955 case 3: RegNo = X86::ST3; break; 956 case 4: RegNo = X86::ST4; break; 957 case 5: RegNo = X86::ST5; break; 958 case 6: RegNo = X86::ST6; break; 959 case 7: RegNo = X86::ST7; break; 960 default: return Error(IntTok.getLoc(), "invalid stack index"); 961 } 962 963 if (getParser().Lex().isNot(AsmToken::RParen)) 964 return Error(Parser.getTok().getLoc(), "expected ')'"); 965 966 EndLoc = Parser.getTok().getEndLoc(); 967 Parser.Lex(); // Eat ')' 968 return false; 969 } 970 971 EndLoc = Parser.getTok().getEndLoc(); 972 973 // If this is "db[0-7]", match it as an alias 974 // for dr[0-7]. 975 if (RegNo == 0 && Tok.getString().size() == 3 && 976 Tok.getString().startswith("db")) { 977 switch (Tok.getString()[2]) { 978 case '0': RegNo = X86::DR0; break; 979 case '1': RegNo = X86::DR1; break; 980 case '2': RegNo = X86::DR2; break; 981 case '3': RegNo = X86::DR3; break; 982 case '4': RegNo = X86::DR4; break; 983 case '5': RegNo = X86::DR5; break; 984 case '6': RegNo = X86::DR6; break; 985 case '7': RegNo = X86::DR7; break; 986 } 987 988 if (RegNo != 0) { 989 EndLoc = Parser.getTok().getEndLoc(); 990 Parser.Lex(); // Eat it. 991 return false; 992 } 993 } 994 995 if (RegNo == 0) { 996 if (isParsingIntelSyntax()) return true; 997 return Error(StartLoc, "invalid register name", 998 SMRange(StartLoc, EndLoc)); 999 } 1000 1001 Parser.Lex(); // Eat identifier token. 1002 return false; 1003 } 1004 1005 void X86AsmParser::SetFrameRegister(unsigned RegNo) { 1006 Instrumentation->SetInitialFrameRegister(RegNo); 1007 } 1008 1009 std::unique_ptr<X86Operand> X86AsmParser::DefaultMemSIOperand(SMLoc Loc) { 1010 unsigned basereg = 1011 is64BitMode() ? X86::RSI : (is32BitMode() ? X86::ESI : X86::SI); 1012 const MCExpr *Disp = MCConstantExpr::create(0, getContext()); 1013 return X86Operand::CreateMem(getPointerWidth(), /*SegReg=*/0, Disp, 1014 /*BaseReg=*/basereg, /*IndexReg=*/0, /*Scale=*/1, 1015 Loc, Loc, 0); 1016 } 1017 1018 std::unique_ptr<X86Operand> X86AsmParser::DefaultMemDIOperand(SMLoc Loc) { 1019 unsigned basereg = 1020 is64BitMode() ? X86::RDI : (is32BitMode() ? X86::EDI : X86::DI); 1021 const MCExpr *Disp = MCConstantExpr::create(0, getContext()); 1022 return X86Operand::CreateMem(getPointerWidth(), /*SegReg=*/0, Disp, 1023 /*BaseReg=*/basereg, /*IndexReg=*/0, /*Scale=*/1, 1024 Loc, Loc, 0); 1025 } 1026 1027 void X86AsmParser::AddDefaultSrcDestOperands( 1028 OperandVector& Operands, std::unique_ptr<llvm::MCParsedAsmOperand> &&Src, 1029 std::unique_ptr<llvm::MCParsedAsmOperand> &&Dst) { 1030 if (isParsingIntelSyntax()) { 1031 Operands.push_back(std::move(Dst)); 1032 Operands.push_back(std::move(Src)); 1033 } 1034 else { 1035 Operands.push_back(std::move(Src)); 1036 Operands.push_back(std::move(Dst)); 1037 } 1038 } 1039 1040 std::unique_ptr<X86Operand> X86AsmParser::ParseOperand() { 1041 if (isParsingIntelSyntax()) 1042 return ParseIntelOperand(); 1043 return ParseATTOperand(); 1044 } 1045 1046 /// getIntelMemOperandSize - Return intel memory operand size. 1047 static unsigned getIntelMemOperandSize(StringRef OpStr) { 1048 unsigned Size = StringSwitch<unsigned>(OpStr) 1049 .Cases("BYTE", "byte", 8) 1050 .Cases("WORD", "word", 16) 1051 .Cases("DWORD", "dword", 32) 1052 .Cases("QWORD", "qword", 64) 1053 .Cases("MMWORD","mmword", 64) 1054 .Cases("XWORD", "xword", 80) 1055 .Cases("TBYTE", "tbyte", 80) 1056 .Cases("XMMWORD", "xmmword", 128) 1057 .Cases("YMMWORD", "ymmword", 256) 1058 .Cases("ZMMWORD", "zmmword", 512) 1059 .Cases("OPAQUE", "opaque", -1U) // needs to be non-zero, but doesn't matter 1060 .Default(0); 1061 return Size; 1062 } 1063 1064 std::unique_ptr<X86Operand> X86AsmParser::CreateMemForInlineAsm( 1065 unsigned SegReg, const MCExpr *Disp, unsigned BaseReg, unsigned IndexReg, 1066 unsigned Scale, SMLoc Start, SMLoc End, unsigned Size, StringRef Identifier, 1067 InlineAsmIdentifierInfo &Info) { 1068 // If we found a decl other than a VarDecl, then assume it is a FuncDecl or 1069 // some other label reference. 1070 if (isa<MCSymbolRefExpr>(Disp) && Info.OpDecl && !Info.IsVarDecl) { 1071 // Insert an explicit size if the user didn't have one. 1072 if (!Size) { 1073 Size = getPointerWidth(); 1074 InstInfo->AsmRewrites->emplace_back(AOK_SizeDirective, Start, 1075 /*Len=*/0, Size); 1076 } 1077 1078 // Create an absolute memory reference in order to match against 1079 // instructions taking a PC relative operand. 1080 return X86Operand::CreateMem(getPointerWidth(), Disp, Start, End, Size, 1081 Identifier, Info.OpDecl); 1082 } 1083 1084 // We either have a direct symbol reference, or an offset from a symbol. The 1085 // parser always puts the symbol on the LHS, so look there for size 1086 // calculation purposes. 1087 const MCBinaryExpr *BinOp = dyn_cast<MCBinaryExpr>(Disp); 1088 bool IsSymRef = 1089 isa<MCSymbolRefExpr>(BinOp ? BinOp->getLHS() : Disp); 1090 if (IsSymRef) { 1091 if (!Size) { 1092 Size = Info.Type * 8; // Size is in terms of bits in this context. 1093 if (Size) 1094 InstInfo->AsmRewrites->emplace_back(AOK_SizeDirective, Start, 1095 /*Len=*/0, Size); 1096 } 1097 } 1098 1099 // When parsing inline assembly we set the base register to a non-zero value 1100 // if we don't know the actual value at this time. This is necessary to 1101 // get the matching correct in some cases. 1102 BaseReg = BaseReg ? BaseReg : 1; 1103 return X86Operand::CreateMem(getPointerWidth(), SegReg, Disp, BaseReg, 1104 IndexReg, Scale, Start, End, Size, Identifier, 1105 Info.OpDecl); 1106 } 1107 1108 static void 1109 RewriteIntelBracExpression(SmallVectorImpl<AsmRewrite> &AsmRewrites, 1110 StringRef SymName, int64_t ImmDisp, 1111 int64_t FinalImmDisp, SMLoc &BracLoc, 1112 SMLoc &StartInBrac, SMLoc &End) { 1113 // Remove the '[' and ']' from the IR string. 1114 AsmRewrites.emplace_back(AOK_Skip, BracLoc, 1); 1115 AsmRewrites.emplace_back(AOK_Skip, End, 1); 1116 1117 // If ImmDisp is non-zero, then we parsed a displacement before the 1118 // bracketed expression (i.e., ImmDisp [ BaseReg + Scale*IndexReg + Disp]) 1119 // If ImmDisp doesn't match the displacement computed by the state machine 1120 // then we have an additional displacement in the bracketed expression. 1121 if (ImmDisp != FinalImmDisp) { 1122 if (ImmDisp) { 1123 // We have an immediate displacement before the bracketed expression. 1124 // Adjust this to match the final immediate displacement. 1125 bool Found = false; 1126 for (AsmRewrite &AR : AsmRewrites) { 1127 if (AR.Loc.getPointer() > BracLoc.getPointer()) 1128 continue; 1129 if (AR.Kind == AOK_ImmPrefix || AR.Kind == AOK_Imm) { 1130 assert (!Found && "ImmDisp already rewritten."); 1131 AR.Kind = AOK_Imm; 1132 AR.Len = BracLoc.getPointer() - AR.Loc.getPointer(); 1133 AR.Val = FinalImmDisp; 1134 Found = true; 1135 break; 1136 } 1137 } 1138 assert (Found && "Unable to rewrite ImmDisp."); 1139 (void)Found; 1140 } else { 1141 // We have a symbolic and an immediate displacement, but no displacement 1142 // before the bracketed expression. Put the immediate displacement 1143 // before the bracketed expression. 1144 AsmRewrites.emplace_back(AOK_Imm, BracLoc, 0, FinalImmDisp); 1145 } 1146 } 1147 // Remove all the ImmPrefix rewrites within the brackets. 1148 for (AsmRewrite &AR : AsmRewrites) { 1149 if (AR.Loc.getPointer() < StartInBrac.getPointer()) 1150 continue; 1151 if (AR.Kind == AOK_ImmPrefix) 1152 AR.Kind = AOK_Delete; 1153 } 1154 const char *SymLocPtr = SymName.data(); 1155 // Skip everything before the symbol. 1156 if (unsigned Len = SymLocPtr - StartInBrac.getPointer()) { 1157 assert(Len > 0 && "Expected a non-negative length."); 1158 AsmRewrites.emplace_back(AOK_Skip, StartInBrac, Len); 1159 } 1160 // Skip everything after the symbol. 1161 if (unsigned Len = End.getPointer() - (SymLocPtr + SymName.size())) { 1162 SMLoc Loc = SMLoc::getFromPointer(SymLocPtr + SymName.size()); 1163 assert(Len > 0 && "Expected a non-negative length."); 1164 AsmRewrites.emplace_back(AOK_Skip, Loc, Len); 1165 } 1166 } 1167 1168 bool X86AsmParser::ParseIntelExpression(IntelExprStateMachine &SM, SMLoc &End) { 1169 MCAsmParser &Parser = getParser(); 1170 const AsmToken &Tok = Parser.getTok(); 1171 1172 bool Done = false; 1173 while (!Done) { 1174 bool UpdateLocLex = true; 1175 1176 // The period in the dot operator (e.g., [ebx].foo.bar) is parsed as an 1177 // identifier. Don't try an parse it as a register. 1178 if (Tok.getString().startswith(".")) 1179 break; 1180 1181 // If we're parsing an immediate expression, we don't expect a '['. 1182 if (SM.getStopOnLBrac() && getLexer().getKind() == AsmToken::LBrac) 1183 break; 1184 1185 AsmToken::TokenKind TK = getLexer().getKind(); 1186 switch (TK) { 1187 default: { 1188 if (SM.isValidEndState()) { 1189 Done = true; 1190 break; 1191 } 1192 return Error(Tok.getLoc(), "unknown token in expression"); 1193 } 1194 case AsmToken::EndOfStatement: { 1195 Done = true; 1196 break; 1197 } 1198 case AsmToken::String: 1199 case AsmToken::Identifier: { 1200 // This could be a register or a symbolic displacement. 1201 unsigned TmpReg; 1202 const MCExpr *Val; 1203 SMLoc IdentLoc = Tok.getLoc(); 1204 StringRef Identifier = Tok.getString(); 1205 if (TK != AsmToken::String && !ParseRegister(TmpReg, IdentLoc, End)) { 1206 SM.onRegister(TmpReg); 1207 UpdateLocLex = false; 1208 break; 1209 } else { 1210 if (!isParsingInlineAsm()) { 1211 if (getParser().parsePrimaryExpr(Val, End)) 1212 return Error(Tok.getLoc(), "Unexpected identifier!"); 1213 } else { 1214 // This is a dot operator, not an adjacent identifier. 1215 if (Identifier.find('.') != StringRef::npos) { 1216 return false; 1217 } else { 1218 InlineAsmIdentifierInfo &Info = SM.getIdentifierInfo(); 1219 if (ParseIntelIdentifier(Val, Identifier, Info, 1220 /*Unevaluated=*/false, End)) 1221 return true; 1222 } 1223 } 1224 SM.onIdentifierExpr(Val, Identifier); 1225 UpdateLocLex = false; 1226 break; 1227 } 1228 return Error(Tok.getLoc(), "Unexpected identifier!"); 1229 } 1230 case AsmToken::Integer: { 1231 StringRef ErrMsg; 1232 if (isParsingInlineAsm() && SM.getAddImmPrefix()) 1233 InstInfo->AsmRewrites->emplace_back(AOK_ImmPrefix, Tok.getLoc()); 1234 // Look for 'b' or 'f' following an Integer as a directional label 1235 SMLoc Loc = getTok().getLoc(); 1236 int64_t IntVal = getTok().getIntVal(); 1237 End = consumeToken(); 1238 UpdateLocLex = false; 1239 if (getLexer().getKind() == AsmToken::Identifier) { 1240 StringRef IDVal = getTok().getString(); 1241 if (IDVal == "f" || IDVal == "b") { 1242 MCSymbol *Sym = 1243 getContext().getDirectionalLocalSymbol(IntVal, IDVal == "b"); 1244 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 1245 const MCExpr *Val = 1246 MCSymbolRefExpr::create(Sym, Variant, getContext()); 1247 if (IDVal == "b" && Sym->isUndefined()) 1248 return Error(Loc, "invalid reference to undefined symbol"); 1249 StringRef Identifier = Sym->getName(); 1250 SM.onIdentifierExpr(Val, Identifier); 1251 End = consumeToken(); 1252 } else { 1253 if (SM.onInteger(IntVal, ErrMsg)) 1254 return Error(Loc, ErrMsg); 1255 } 1256 } else { 1257 if (SM.onInteger(IntVal, ErrMsg)) 1258 return Error(Loc, ErrMsg); 1259 } 1260 break; 1261 } 1262 case AsmToken::Plus: SM.onPlus(); break; 1263 case AsmToken::Minus: SM.onMinus(); break; 1264 case AsmToken::Tilde: SM.onNot(); break; 1265 case AsmToken::Star: SM.onStar(); break; 1266 case AsmToken::Slash: SM.onDivide(); break; 1267 case AsmToken::Pipe: SM.onOr(); break; 1268 case AsmToken::Caret: SM.onXor(); break; 1269 case AsmToken::Amp: SM.onAnd(); break; 1270 case AsmToken::LessLess: 1271 SM.onLShift(); break; 1272 case AsmToken::GreaterGreater: 1273 SM.onRShift(); break; 1274 case AsmToken::LBrac: SM.onLBrac(); break; 1275 case AsmToken::RBrac: SM.onRBrac(); break; 1276 case AsmToken::LParen: SM.onLParen(); break; 1277 case AsmToken::RParen: SM.onRParen(); break; 1278 } 1279 if (SM.hadError()) 1280 return Error(Tok.getLoc(), "unknown token in expression"); 1281 1282 if (!Done && UpdateLocLex) 1283 End = consumeToken(); 1284 } 1285 return false; 1286 } 1287 1288 std::unique_ptr<X86Operand> 1289 X86AsmParser::ParseIntelBracExpression(unsigned SegReg, SMLoc Start, 1290 int64_t ImmDisp, unsigned Size) { 1291 MCAsmParser &Parser = getParser(); 1292 const AsmToken &Tok = Parser.getTok(); 1293 SMLoc BracLoc = Tok.getLoc(), End = Tok.getEndLoc(); 1294 if (getLexer().isNot(AsmToken::LBrac)) 1295 return ErrorOperand(BracLoc, "Expected '[' token!"); 1296 Parser.Lex(); // Eat '[' 1297 1298 SMLoc StartInBrac = Tok.getLoc(); 1299 // Parse [ Symbol + ImmDisp ] and [ BaseReg + Scale*IndexReg + ImmDisp ]. We 1300 // may have already parsed an immediate displacement before the bracketed 1301 // expression. 1302 IntelExprStateMachine SM(ImmDisp, /*StopOnLBrac=*/false, /*AddImmPrefix=*/true); 1303 if (ParseIntelExpression(SM, End)) 1304 return nullptr; 1305 1306 const MCExpr *Disp = nullptr; 1307 if (const MCExpr *Sym = SM.getSym()) { 1308 // A symbolic displacement. 1309 Disp = Sym; 1310 if (isParsingInlineAsm()) 1311 RewriteIntelBracExpression(*InstInfo->AsmRewrites, SM.getSymName(), 1312 ImmDisp, SM.getImm(), BracLoc, StartInBrac, 1313 End); 1314 } 1315 1316 if (SM.getImm() || !Disp) { 1317 const MCExpr *Imm = MCConstantExpr::create(SM.getImm(), getContext()); 1318 if (Disp) 1319 Disp = MCBinaryExpr::createAdd(Disp, Imm, getContext()); 1320 else 1321 Disp = Imm; // An immediate displacement only. 1322 } 1323 1324 // Parse struct field access. Intel requires a dot, but MSVC doesn't. MSVC 1325 // will in fact do global lookup the field name inside all global typedefs, 1326 // but we don't emulate that. 1327 if (Tok.getString().find('.') != StringRef::npos) { 1328 const MCExpr *NewDisp; 1329 if (ParseIntelDotOperator(Disp, NewDisp)) 1330 return nullptr; 1331 1332 End = Tok.getEndLoc(); 1333 Parser.Lex(); // Eat the field. 1334 Disp = NewDisp; 1335 } 1336 1337 int BaseReg = SM.getBaseReg(); 1338 int IndexReg = SM.getIndexReg(); 1339 int Scale = SM.getScale(); 1340 if (!isParsingInlineAsm()) { 1341 // handle [-42] 1342 if (!BaseReg && !IndexReg) { 1343 if (!SegReg) 1344 return X86Operand::CreateMem(getPointerWidth(), Disp, Start, End, Size); 1345 return X86Operand::CreateMem(getPointerWidth(), SegReg, Disp, 0, 0, 1, 1346 Start, End, Size); 1347 } 1348 StringRef ErrMsg; 1349 if (CheckBaseRegAndIndexReg(BaseReg, IndexReg, ErrMsg)) { 1350 Error(StartInBrac, ErrMsg); 1351 return nullptr; 1352 } 1353 return X86Operand::CreateMem(getPointerWidth(), SegReg, Disp, BaseReg, 1354 IndexReg, Scale, Start, End, Size); 1355 } 1356 1357 InlineAsmIdentifierInfo &Info = SM.getIdentifierInfo(); 1358 return CreateMemForInlineAsm(SegReg, Disp, BaseReg, IndexReg, Scale, Start, 1359 End, Size, SM.getSymName(), Info); 1360 } 1361 1362 // Inline assembly may use variable names with namespace alias qualifiers. 1363 bool X86AsmParser::ParseIntelIdentifier(const MCExpr *&Val, 1364 StringRef &Identifier, 1365 InlineAsmIdentifierInfo &Info, 1366 bool IsUnevaluatedOperand, SMLoc &End) { 1367 MCAsmParser &Parser = getParser(); 1368 assert(isParsingInlineAsm() && "Expected to be parsing inline assembly."); 1369 Val = nullptr; 1370 1371 StringRef LineBuf(Identifier.data()); 1372 void *Result = 1373 SemaCallback->LookupInlineAsmIdentifier(LineBuf, Info, IsUnevaluatedOperand); 1374 1375 const AsmToken &Tok = Parser.getTok(); 1376 SMLoc Loc = Tok.getLoc(); 1377 1378 // Advance the token stream until the end of the current token is 1379 // after the end of what the frontend claimed. 1380 const char *EndPtr = Tok.getLoc().getPointer() + LineBuf.size(); 1381 do { 1382 End = Tok.getEndLoc(); 1383 getLexer().Lex(); 1384 } while (End.getPointer() < EndPtr); 1385 Identifier = LineBuf; 1386 1387 // The frontend should end parsing on an assembler token boundary, unless it 1388 // failed parsing. 1389 assert((End.getPointer() == EndPtr || !Result) && 1390 "frontend claimed part of a token?"); 1391 1392 // If the identifier lookup was unsuccessful, assume that we are dealing with 1393 // a label. 1394 if (!Result) { 1395 StringRef InternalName = 1396 SemaCallback->LookupInlineAsmLabel(Identifier, getSourceManager(), 1397 Loc, false); 1398 assert(InternalName.size() && "We should have an internal name here."); 1399 // Push a rewrite for replacing the identifier name with the internal name. 1400 InstInfo->AsmRewrites->emplace_back(AOK_Label, Loc, Identifier.size(), 1401 InternalName); 1402 } 1403 1404 // Create the symbol reference. 1405 MCSymbol *Sym = getContext().getOrCreateSymbol(Identifier); 1406 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 1407 Val = MCSymbolRefExpr::create(Sym, Variant, getParser().getContext()); 1408 return false; 1409 } 1410 1411 /// \brief Parse intel style segment override. 1412 std::unique_ptr<X86Operand> 1413 X86AsmParser::ParseIntelSegmentOverride(unsigned SegReg, SMLoc Start, 1414 unsigned Size) { 1415 MCAsmParser &Parser = getParser(); 1416 assert(SegReg != 0 && "Tried to parse a segment override without a segment!"); 1417 const AsmToken &Tok = Parser.getTok(); // Eat colon. 1418 if (Tok.isNot(AsmToken::Colon)) 1419 return ErrorOperand(Tok.getLoc(), "Expected ':' token!"); 1420 Parser.Lex(); // Eat ':' 1421 1422 int64_t ImmDisp = 0; 1423 if (getLexer().is(AsmToken::Integer)) { 1424 ImmDisp = Tok.getIntVal(); 1425 AsmToken ImmDispToken = Parser.Lex(); // Eat the integer. 1426 1427 if (isParsingInlineAsm()) 1428 InstInfo->AsmRewrites->emplace_back(AOK_ImmPrefix, ImmDispToken.getLoc()); 1429 1430 if (getLexer().isNot(AsmToken::LBrac)) { 1431 // An immediate following a 'segment register', 'colon' token sequence can 1432 // be followed by a bracketed expression. If it isn't we know we have our 1433 // final segment override. 1434 const MCExpr *Disp = MCConstantExpr::create(ImmDisp, getContext()); 1435 return X86Operand::CreateMem(getPointerWidth(), SegReg, Disp, 1436 /*BaseReg=*/0, /*IndexReg=*/0, /*Scale=*/1, 1437 Start, ImmDispToken.getEndLoc(), Size); 1438 } 1439 } 1440 1441 if (getLexer().is(AsmToken::LBrac)) 1442 return ParseIntelBracExpression(SegReg, Start, ImmDisp, Size); 1443 1444 const MCExpr *Val; 1445 SMLoc End; 1446 if (!isParsingInlineAsm()) { 1447 if (getParser().parsePrimaryExpr(Val, End)) 1448 return ErrorOperand(Tok.getLoc(), "unknown token in expression"); 1449 1450 return X86Operand::CreateMem(getPointerWidth(), Val, Start, End, Size); 1451 } 1452 1453 InlineAsmIdentifierInfo Info; 1454 StringRef Identifier = Tok.getString(); 1455 if (ParseIntelIdentifier(Val, Identifier, Info, 1456 /*Unevaluated=*/false, End)) 1457 return nullptr; 1458 return CreateMemForInlineAsm(/*SegReg=*/0, Val, /*BaseReg=*/0,/*IndexReg=*/0, 1459 /*Scale=*/1, Start, End, Size, Identifier, Info); 1460 } 1461 1462 //ParseRoundingModeOp - Parse AVX-512 rounding mode operand 1463 std::unique_ptr<X86Operand> 1464 X86AsmParser::ParseRoundingModeOp(SMLoc Start, SMLoc End) { 1465 MCAsmParser &Parser = getParser(); 1466 const AsmToken &Tok = Parser.getTok(); 1467 // Eat "{" and mark the current place. 1468 const SMLoc consumedToken = consumeToken(); 1469 if (Tok.getIdentifier().startswith("r")){ 1470 int rndMode = StringSwitch<int>(Tok.getIdentifier()) 1471 .Case("rn", X86::STATIC_ROUNDING::TO_NEAREST_INT) 1472 .Case("rd", X86::STATIC_ROUNDING::TO_NEG_INF) 1473 .Case("ru", X86::STATIC_ROUNDING::TO_POS_INF) 1474 .Case("rz", X86::STATIC_ROUNDING::TO_ZERO) 1475 .Default(-1); 1476 if (-1 == rndMode) 1477 return ErrorOperand(Tok.getLoc(), "Invalid rounding mode."); 1478 Parser.Lex(); // Eat "r*" of r*-sae 1479 if (!getLexer().is(AsmToken::Minus)) 1480 return ErrorOperand(Tok.getLoc(), "Expected - at this point"); 1481 Parser.Lex(); // Eat "-" 1482 Parser.Lex(); // Eat the sae 1483 if (!getLexer().is(AsmToken::RCurly)) 1484 return ErrorOperand(Tok.getLoc(), "Expected } at this point"); 1485 Parser.Lex(); // Eat "}" 1486 const MCExpr *RndModeOp = 1487 MCConstantExpr::create(rndMode, Parser.getContext()); 1488 return X86Operand::CreateImm(RndModeOp, Start, End); 1489 } 1490 if(Tok.getIdentifier().equals("sae")){ 1491 Parser.Lex(); // Eat the sae 1492 if (!getLexer().is(AsmToken::RCurly)) 1493 return ErrorOperand(Tok.getLoc(), "Expected } at this point"); 1494 Parser.Lex(); // Eat "}" 1495 return X86Operand::CreateToken("{sae}", consumedToken); 1496 } 1497 return ErrorOperand(Tok.getLoc(), "unknown token in expression"); 1498 } 1499 /// ParseIntelMemOperand - Parse intel style memory operand. 1500 std::unique_ptr<X86Operand> X86AsmParser::ParseIntelMemOperand(int64_t ImmDisp, 1501 SMLoc Start, 1502 unsigned Size) { 1503 MCAsmParser &Parser = getParser(); 1504 const AsmToken &Tok = Parser.getTok(); 1505 SMLoc End; 1506 1507 // Parse ImmDisp [ BaseReg + Scale*IndexReg + Disp ]. 1508 if (getLexer().is(AsmToken::LBrac)) 1509 return ParseIntelBracExpression(/*SegReg=*/0, Start, ImmDisp, Size); 1510 assert(ImmDisp == 0); 1511 1512 const MCExpr *Val; 1513 if (!isParsingInlineAsm()) { 1514 if (getParser().parsePrimaryExpr(Val, End)) 1515 return ErrorOperand(Tok.getLoc(), "unknown token in expression"); 1516 1517 return X86Operand::CreateMem(getPointerWidth(), Val, Start, End, Size); 1518 } 1519 1520 InlineAsmIdentifierInfo Info; 1521 StringRef Identifier = Tok.getString(); 1522 if (ParseIntelIdentifier(Val, Identifier, Info, 1523 /*Unevaluated=*/false, End)) 1524 return nullptr; 1525 1526 if (!getLexer().is(AsmToken::LBrac)) 1527 return CreateMemForInlineAsm(/*SegReg=*/0, Val, /*BaseReg=*/0, /*IndexReg=*/0, 1528 /*Scale=*/1, Start, End, Size, Identifier, Info); 1529 1530 Parser.Lex(); // Eat '[' 1531 1532 // Parse Identifier [ ImmDisp ] 1533 IntelExprStateMachine SM(/*ImmDisp=*/0, /*StopOnLBrac=*/true, 1534 /*AddImmPrefix=*/false); 1535 if (ParseIntelExpression(SM, End)) 1536 return nullptr; 1537 1538 if (SM.getSym()) { 1539 Error(Start, "cannot use more than one symbol in memory operand"); 1540 return nullptr; 1541 } 1542 if (SM.getBaseReg()) { 1543 Error(Start, "cannot use base register with variable reference"); 1544 return nullptr; 1545 } 1546 if (SM.getIndexReg()) { 1547 Error(Start, "cannot use index register with variable reference"); 1548 return nullptr; 1549 } 1550 1551 const MCExpr *Disp = MCConstantExpr::create(SM.getImm(), getContext()); 1552 // BaseReg is non-zero to avoid assertions. In the context of inline asm, 1553 // we're pointing to a local variable in memory, so the base register is 1554 // really the frame or stack pointer. 1555 return X86Operand::CreateMem(getPointerWidth(), /*SegReg=*/0, Disp, 1556 /*BaseReg=*/1, /*IndexReg=*/0, /*Scale=*/1, 1557 Start, End, Size, Identifier, Info.OpDecl); 1558 } 1559 1560 /// Parse the '.' operator. 1561 bool X86AsmParser::ParseIntelDotOperator(const MCExpr *Disp, 1562 const MCExpr *&NewDisp) { 1563 MCAsmParser &Parser = getParser(); 1564 const AsmToken &Tok = Parser.getTok(); 1565 int64_t OrigDispVal, DotDispVal; 1566 1567 // FIXME: Handle non-constant expressions. 1568 if (const MCConstantExpr *OrigDisp = dyn_cast<MCConstantExpr>(Disp)) 1569 OrigDispVal = OrigDisp->getValue(); 1570 else 1571 return Error(Tok.getLoc(), "Non-constant offsets are not supported!"); 1572 1573 // Drop the optional '.'. 1574 StringRef DotDispStr = Tok.getString(); 1575 if (DotDispStr.startswith(".")) 1576 DotDispStr = DotDispStr.drop_front(1); 1577 1578 // .Imm gets lexed as a real. 1579 if (Tok.is(AsmToken::Real)) { 1580 APInt DotDisp; 1581 DotDispStr.getAsInteger(10, DotDisp); 1582 DotDispVal = DotDisp.getZExtValue(); 1583 } else if (isParsingInlineAsm() && Tok.is(AsmToken::Identifier)) { 1584 unsigned DotDisp; 1585 std::pair<StringRef, StringRef> BaseMember = DotDispStr.split('.'); 1586 if (SemaCallback->LookupInlineAsmField(BaseMember.first, BaseMember.second, 1587 DotDisp)) 1588 return Error(Tok.getLoc(), "Unable to lookup field reference!"); 1589 DotDispVal = DotDisp; 1590 } else 1591 return Error(Tok.getLoc(), "Unexpected token type!"); 1592 1593 if (isParsingInlineAsm() && Tok.is(AsmToken::Identifier)) { 1594 SMLoc Loc = SMLoc::getFromPointer(DotDispStr.data()); 1595 unsigned Len = DotDispStr.size(); 1596 unsigned Val = OrigDispVal + DotDispVal; 1597 InstInfo->AsmRewrites->emplace_back(AOK_DotOperator, Loc, Len, Val); 1598 } 1599 1600 NewDisp = MCConstantExpr::create(OrigDispVal + DotDispVal, getContext()); 1601 return false; 1602 } 1603 1604 /// Parse the 'offset' operator. This operator is used to specify the 1605 /// location rather then the content of a variable. 1606 std::unique_ptr<X86Operand> X86AsmParser::ParseIntelOffsetOfOperator() { 1607 MCAsmParser &Parser = getParser(); 1608 const AsmToken &Tok = Parser.getTok(); 1609 SMLoc OffsetOfLoc = Tok.getLoc(); 1610 Parser.Lex(); // Eat offset. 1611 1612 const MCExpr *Val; 1613 InlineAsmIdentifierInfo Info; 1614 SMLoc Start = Tok.getLoc(), End; 1615 StringRef Identifier = Tok.getString(); 1616 if (ParseIntelIdentifier(Val, Identifier, Info, 1617 /*Unevaluated=*/false, End)) 1618 return nullptr; 1619 1620 // Don't emit the offset operator. 1621 InstInfo->AsmRewrites->emplace_back(AOK_Skip, OffsetOfLoc, 7); 1622 1623 // The offset operator will have an 'r' constraint, thus we need to create 1624 // register operand to ensure proper matching. Just pick a GPR based on 1625 // the size of a pointer. 1626 unsigned RegNo = 1627 is64BitMode() ? X86::RBX : (is32BitMode() ? X86::EBX : X86::BX); 1628 return X86Operand::CreateReg(RegNo, Start, End, /*GetAddress=*/true, 1629 OffsetOfLoc, Identifier, Info.OpDecl); 1630 } 1631 1632 enum IntelOperatorKind { 1633 IOK_LENGTH, 1634 IOK_SIZE, 1635 IOK_TYPE 1636 }; 1637 1638 /// Parse the 'LENGTH', 'TYPE' and 'SIZE' operators. The LENGTH operator 1639 /// returns the number of elements in an array. It returns the value 1 for 1640 /// non-array variables. The SIZE operator returns the size of a C or C++ 1641 /// variable. A variable's size is the product of its LENGTH and TYPE. The 1642 /// TYPE operator returns the size of a C or C++ type or variable. If the 1643 /// variable is an array, TYPE returns the size of a single element. 1644 std::unique_ptr<X86Operand> X86AsmParser::ParseIntelOperator(unsigned OpKind) { 1645 MCAsmParser &Parser = getParser(); 1646 const AsmToken &Tok = Parser.getTok(); 1647 SMLoc TypeLoc = Tok.getLoc(); 1648 Parser.Lex(); // Eat operator. 1649 1650 const MCExpr *Val = nullptr; 1651 InlineAsmIdentifierInfo Info; 1652 SMLoc Start = Tok.getLoc(), End; 1653 StringRef Identifier = Tok.getString(); 1654 if (ParseIntelIdentifier(Val, Identifier, Info, 1655 /*Unevaluated=*/true, End)) 1656 return nullptr; 1657 1658 if (!Info.OpDecl) 1659 return ErrorOperand(Start, "unable to lookup expression"); 1660 1661 unsigned CVal = 0; 1662 switch(OpKind) { 1663 default: llvm_unreachable("Unexpected operand kind!"); 1664 case IOK_LENGTH: CVal = Info.Length; break; 1665 case IOK_SIZE: CVal = Info.Size; break; 1666 case IOK_TYPE: CVal = Info.Type; break; 1667 } 1668 1669 // Rewrite the type operator and the C or C++ type or variable in terms of an 1670 // immediate. E.g. TYPE foo -> $$4 1671 unsigned Len = End.getPointer() - TypeLoc.getPointer(); 1672 InstInfo->AsmRewrites->emplace_back(AOK_Imm, TypeLoc, Len, CVal); 1673 1674 const MCExpr *Imm = MCConstantExpr::create(CVal, getContext()); 1675 return X86Operand::CreateImm(Imm, Start, End); 1676 } 1677 1678 std::unique_ptr<X86Operand> X86AsmParser::ParseIntelOperand() { 1679 MCAsmParser &Parser = getParser(); 1680 const AsmToken &Tok = Parser.getTok(); 1681 SMLoc Start, End; 1682 1683 // Offset, length, type and size operators. 1684 if (isParsingInlineAsm()) { 1685 StringRef AsmTokStr = Tok.getString(); 1686 if (AsmTokStr == "offset" || AsmTokStr == "OFFSET") 1687 return ParseIntelOffsetOfOperator(); 1688 if (AsmTokStr == "length" || AsmTokStr == "LENGTH") 1689 return ParseIntelOperator(IOK_LENGTH); 1690 if (AsmTokStr == "size" || AsmTokStr == "SIZE") 1691 return ParseIntelOperator(IOK_SIZE); 1692 if (AsmTokStr == "type" || AsmTokStr == "TYPE") 1693 return ParseIntelOperator(IOK_TYPE); 1694 } 1695 1696 unsigned Size = getIntelMemOperandSize(Tok.getString()); 1697 if (Size) { 1698 Parser.Lex(); // Eat operand size (e.g., byte, word). 1699 if (Tok.getString() != "PTR" && Tok.getString() != "ptr") 1700 return ErrorOperand(Tok.getLoc(), "Expected 'PTR' or 'ptr' token!"); 1701 Parser.Lex(); // Eat ptr. 1702 } 1703 Start = Tok.getLoc(); 1704 1705 // Immediate. 1706 if (getLexer().is(AsmToken::Integer) || getLexer().is(AsmToken::Minus) || 1707 getLexer().is(AsmToken::Tilde) || getLexer().is(AsmToken::LParen)) { 1708 AsmToken StartTok = Tok; 1709 IntelExprStateMachine SM(/*Imm=*/0, /*StopOnLBrac=*/true, 1710 /*AddImmPrefix=*/false); 1711 if (ParseIntelExpression(SM, End)) 1712 return nullptr; 1713 1714 int64_t Imm = SM.getImm(); 1715 if (isParsingInlineAsm()) { 1716 unsigned Len = Tok.getLoc().getPointer() - Start.getPointer(); 1717 if (StartTok.getString().size() == Len) 1718 // Just add a prefix if this wasn't a complex immediate expression. 1719 InstInfo->AsmRewrites->emplace_back(AOK_ImmPrefix, Start); 1720 else 1721 // Otherwise, rewrite the complex expression as a single immediate. 1722 InstInfo->AsmRewrites->emplace_back(AOK_Imm, Start, Len, Imm); 1723 } 1724 1725 if (getLexer().isNot(AsmToken::LBrac)) { 1726 // If a directional label (ie. 1f or 2b) was parsed above from 1727 // ParseIntelExpression() then SM.getSym() was set to a pointer to 1728 // to the MCExpr with the directional local symbol and this is a 1729 // memory operand not an immediate operand. 1730 if (SM.getSym()) 1731 return X86Operand::CreateMem(getPointerWidth(), SM.getSym(), Start, End, 1732 Size); 1733 1734 const MCExpr *ImmExpr = MCConstantExpr::create(Imm, getContext()); 1735 return X86Operand::CreateImm(ImmExpr, Start, End); 1736 } 1737 1738 // Only positive immediates are valid. 1739 if (Imm < 0) 1740 return ErrorOperand(Start, "expected a positive immediate displacement " 1741 "before bracketed expr."); 1742 1743 // Parse ImmDisp [ BaseReg + Scale*IndexReg + Disp ]. 1744 return ParseIntelMemOperand(Imm, Start, Size); 1745 } 1746 1747 // rounding mode token 1748 if (getSTI().getFeatureBits()[X86::FeatureAVX512] && 1749 getLexer().is(AsmToken::LCurly)) 1750 return ParseRoundingModeOp(Start, End); 1751 1752 // Register. 1753 unsigned RegNo = 0; 1754 if (!ParseRegister(RegNo, Start, End)) { 1755 // If this is a segment register followed by a ':', then this is the start 1756 // of a segment override, otherwise this is a normal register reference. 1757 if (getLexer().isNot(AsmToken::Colon)) 1758 return X86Operand::CreateReg(RegNo, Start, End); 1759 1760 return ParseIntelSegmentOverride(/*SegReg=*/RegNo, Start, Size); 1761 } 1762 1763 // Memory operand. 1764 return ParseIntelMemOperand(/*Disp=*/0, Start, Size); 1765 } 1766 1767 std::unique_ptr<X86Operand> X86AsmParser::ParseATTOperand() { 1768 MCAsmParser &Parser = getParser(); 1769 switch (getLexer().getKind()) { 1770 default: 1771 // Parse a memory operand with no segment register. 1772 return ParseMemOperand(0, Parser.getTok().getLoc()); 1773 case AsmToken::Percent: { 1774 // Read the register. 1775 unsigned RegNo; 1776 SMLoc Start, End; 1777 if (ParseRegister(RegNo, Start, End)) return nullptr; 1778 if (RegNo == X86::EIZ || RegNo == X86::RIZ) { 1779 Error(Start, "%eiz and %riz can only be used as index registers", 1780 SMRange(Start, End)); 1781 return nullptr; 1782 } 1783 1784 // If this is a segment register followed by a ':', then this is the start 1785 // of a memory reference, otherwise this is a normal register reference. 1786 if (getLexer().isNot(AsmToken::Colon)) 1787 return X86Operand::CreateReg(RegNo, Start, End); 1788 1789 if (!X86MCRegisterClasses[X86::SEGMENT_REGRegClassID].contains(RegNo)) 1790 return ErrorOperand(Start, "invalid segment register"); 1791 1792 getParser().Lex(); // Eat the colon. 1793 return ParseMemOperand(RegNo, Start); 1794 } 1795 case AsmToken::Dollar: { 1796 // $42 -> immediate. 1797 SMLoc Start = Parser.getTok().getLoc(), End; 1798 Parser.Lex(); 1799 const MCExpr *Val; 1800 if (getParser().parseExpression(Val, End)) 1801 return nullptr; 1802 return X86Operand::CreateImm(Val, Start, End); 1803 } 1804 case AsmToken::LCurly:{ 1805 SMLoc Start = Parser.getTok().getLoc(), End; 1806 if (getSTI().getFeatureBits()[X86::FeatureAVX512]) 1807 return ParseRoundingModeOp(Start, End); 1808 return ErrorOperand(Start, "unknown token in expression"); 1809 } 1810 } 1811 } 1812 1813 bool X86AsmParser::HandleAVX512Operand(OperandVector &Operands, 1814 const MCParsedAsmOperand &Op) { 1815 MCAsmParser &Parser = getParser(); 1816 if(getSTI().getFeatureBits()[X86::FeatureAVX512]) { 1817 if (getLexer().is(AsmToken::LCurly)) { 1818 // Eat "{" and mark the current place. 1819 const SMLoc consumedToken = consumeToken(); 1820 // Distinguish {1to<NUM>} from {%k<NUM>}. 1821 if(getLexer().is(AsmToken::Integer)) { 1822 // Parse memory broadcasting ({1to<NUM>}). 1823 if (getLexer().getTok().getIntVal() != 1) 1824 return !ErrorAndEatStatement(getLexer().getLoc(), 1825 "Expected 1to<NUM> at this point"); 1826 Parser.Lex(); // Eat "1" of 1to8 1827 if (!getLexer().is(AsmToken::Identifier) || 1828 !getLexer().getTok().getIdentifier().startswith("to")) 1829 return !ErrorAndEatStatement(getLexer().getLoc(), 1830 "Expected 1to<NUM> at this point"); 1831 // Recognize only reasonable suffixes. 1832 const char *BroadcastPrimitive = 1833 StringSwitch<const char*>(getLexer().getTok().getIdentifier()) 1834 .Case("to2", "{1to2}") 1835 .Case("to4", "{1to4}") 1836 .Case("to8", "{1to8}") 1837 .Case("to16", "{1to16}") 1838 .Default(nullptr); 1839 if (!BroadcastPrimitive) 1840 return !ErrorAndEatStatement(getLexer().getLoc(), 1841 "Invalid memory broadcast primitive."); 1842 Parser.Lex(); // Eat "toN" of 1toN 1843 if (!getLexer().is(AsmToken::RCurly)) 1844 return !ErrorAndEatStatement(getLexer().getLoc(), 1845 "Expected } at this point"); 1846 Parser.Lex(); // Eat "}" 1847 Operands.push_back(X86Operand::CreateToken(BroadcastPrimitive, 1848 consumedToken)); 1849 // No AVX512 specific primitives can pass 1850 // after memory broadcasting, so return. 1851 return true; 1852 } else { 1853 // Parse mask register {%k1} 1854 Operands.push_back(X86Operand::CreateToken("{", consumedToken)); 1855 if (std::unique_ptr<X86Operand> Op = ParseOperand()) { 1856 Operands.push_back(std::move(Op)); 1857 if (!getLexer().is(AsmToken::RCurly)) 1858 return !ErrorAndEatStatement(getLexer().getLoc(), 1859 "Expected } at this point"); 1860 Operands.push_back(X86Operand::CreateToken("}", consumeToken())); 1861 1862 // Parse "zeroing non-masked" semantic {z} 1863 if (getLexer().is(AsmToken::LCurly)) { 1864 Operands.push_back(X86Operand::CreateToken("{z}", consumeToken())); 1865 if (!getLexer().is(AsmToken::Identifier) || 1866 getLexer().getTok().getIdentifier() != "z") 1867 return !ErrorAndEatStatement(getLexer().getLoc(), 1868 "Expected z at this point"); 1869 Parser.Lex(); // Eat the z 1870 if (!getLexer().is(AsmToken::RCurly)) 1871 return !ErrorAndEatStatement(getLexer().getLoc(), 1872 "Expected } at this point"); 1873 Parser.Lex(); // Eat the } 1874 } 1875 } 1876 } 1877 } 1878 } 1879 return true; 1880 } 1881 1882 /// ParseMemOperand: segment: disp(basereg, indexreg, scale). The '%ds:' prefix 1883 /// has already been parsed if present. 1884 std::unique_ptr<X86Operand> X86AsmParser::ParseMemOperand(unsigned SegReg, 1885 SMLoc MemStart) { 1886 1887 MCAsmParser &Parser = getParser(); 1888 // We have to disambiguate a parenthesized expression "(4+5)" from the start 1889 // of a memory operand with a missing displacement "(%ebx)" or "(,%eax)". The 1890 // only way to do this without lookahead is to eat the '(' and see what is 1891 // after it. 1892 const MCExpr *Disp = MCConstantExpr::create(0, getParser().getContext()); 1893 if (getLexer().isNot(AsmToken::LParen)) { 1894 SMLoc ExprEnd; 1895 if (getParser().parseExpression(Disp, ExprEnd)) return nullptr; 1896 1897 // After parsing the base expression we could either have a parenthesized 1898 // memory address or not. If not, return now. If so, eat the (. 1899 if (getLexer().isNot(AsmToken::LParen)) { 1900 // Unless we have a segment register, treat this as an immediate. 1901 if (SegReg == 0) 1902 return X86Operand::CreateMem(getPointerWidth(), Disp, MemStart, ExprEnd); 1903 return X86Operand::CreateMem(getPointerWidth(), SegReg, Disp, 0, 0, 1, 1904 MemStart, ExprEnd); 1905 } 1906 1907 // Eat the '('. 1908 Parser.Lex(); 1909 } else { 1910 // Okay, we have a '('. We don't know if this is an expression or not, but 1911 // so we have to eat the ( to see beyond it. 1912 SMLoc LParenLoc = Parser.getTok().getLoc(); 1913 Parser.Lex(); // Eat the '('. 1914 1915 if (getLexer().is(AsmToken::Percent) || getLexer().is(AsmToken::Comma)) { 1916 // Nothing to do here, fall into the code below with the '(' part of the 1917 // memory operand consumed. 1918 } else { 1919 SMLoc ExprEnd; 1920 1921 // It must be an parenthesized expression, parse it now. 1922 if (getParser().parseParenExpression(Disp, ExprEnd)) 1923 return nullptr; 1924 1925 // After parsing the base expression we could either have a parenthesized 1926 // memory address or not. If not, return now. If so, eat the (. 1927 if (getLexer().isNot(AsmToken::LParen)) { 1928 // Unless we have a segment register, treat this as an immediate. 1929 if (SegReg == 0) 1930 return X86Operand::CreateMem(getPointerWidth(), Disp, LParenLoc, 1931 ExprEnd); 1932 return X86Operand::CreateMem(getPointerWidth(), SegReg, Disp, 0, 0, 1, 1933 MemStart, ExprEnd); 1934 } 1935 1936 // Eat the '('. 1937 Parser.Lex(); 1938 } 1939 } 1940 1941 // If we reached here, then we just ate the ( of the memory operand. Process 1942 // the rest of the memory operand. 1943 unsigned BaseReg = 0, IndexReg = 0, Scale = 1; 1944 SMLoc IndexLoc, BaseLoc; 1945 1946 if (getLexer().is(AsmToken::Percent)) { 1947 SMLoc StartLoc, EndLoc; 1948 BaseLoc = Parser.getTok().getLoc(); 1949 if (ParseRegister(BaseReg, StartLoc, EndLoc)) return nullptr; 1950 if (BaseReg == X86::EIZ || BaseReg == X86::RIZ) { 1951 Error(StartLoc, "eiz and riz can only be used as index registers", 1952 SMRange(StartLoc, EndLoc)); 1953 return nullptr; 1954 } 1955 } 1956 1957 if (getLexer().is(AsmToken::Comma)) { 1958 Parser.Lex(); // Eat the comma. 1959 IndexLoc = Parser.getTok().getLoc(); 1960 1961 // Following the comma we should have either an index register, or a scale 1962 // value. We don't support the later form, but we want to parse it 1963 // correctly. 1964 // 1965 // Not that even though it would be completely consistent to support syntax 1966 // like "1(%eax,,1)", the assembler doesn't. Use "eiz" or "riz" for this. 1967 if (getLexer().is(AsmToken::Percent)) { 1968 SMLoc L; 1969 if (ParseRegister(IndexReg, L, L)) return nullptr; 1970 1971 if (getLexer().isNot(AsmToken::RParen)) { 1972 // Parse the scale amount: 1973 // ::= ',' [scale-expression] 1974 if (getLexer().isNot(AsmToken::Comma)) { 1975 Error(Parser.getTok().getLoc(), 1976 "expected comma in scale expression"); 1977 return nullptr; 1978 } 1979 Parser.Lex(); // Eat the comma. 1980 1981 if (getLexer().isNot(AsmToken::RParen)) { 1982 SMLoc Loc = Parser.getTok().getLoc(); 1983 1984 int64_t ScaleVal; 1985 if (getParser().parseAbsoluteExpression(ScaleVal)){ 1986 Error(Loc, "expected scale expression"); 1987 return nullptr; 1988 } 1989 1990 // Validate the scale amount. 1991 if (X86MCRegisterClasses[X86::GR16RegClassID].contains(BaseReg) && 1992 ScaleVal != 1) { 1993 Error(Loc, "scale factor in 16-bit address must be 1"); 1994 return nullptr; 1995 } 1996 if (ScaleVal != 1 && ScaleVal != 2 && ScaleVal != 4 && 1997 ScaleVal != 8) { 1998 Error(Loc, "scale factor in address must be 1, 2, 4 or 8"); 1999 return nullptr; 2000 } 2001 Scale = (unsigned)ScaleVal; 2002 } 2003 } 2004 } else if (getLexer().isNot(AsmToken::RParen)) { 2005 // A scale amount without an index is ignored. 2006 // index. 2007 SMLoc Loc = Parser.getTok().getLoc(); 2008 2009 int64_t Value; 2010 if (getParser().parseAbsoluteExpression(Value)) 2011 return nullptr; 2012 2013 if (Value != 1) 2014 Warning(Loc, "scale factor without index register is ignored"); 2015 Scale = 1; 2016 } 2017 } 2018 2019 // Ok, we've eaten the memory operand, verify we have a ')' and eat it too. 2020 if (getLexer().isNot(AsmToken::RParen)) { 2021 Error(Parser.getTok().getLoc(), "unexpected token in memory operand"); 2022 return nullptr; 2023 } 2024 SMLoc MemEnd = Parser.getTok().getEndLoc(); 2025 Parser.Lex(); // Eat the ')'. 2026 2027 // Check for use of invalid 16-bit registers. Only BX/BP/SI/DI are allowed, 2028 // and then only in non-64-bit modes. Except for DX, which is a special case 2029 // because an unofficial form of in/out instructions uses it. 2030 if (X86MCRegisterClasses[X86::GR16RegClassID].contains(BaseReg) && 2031 (is64BitMode() || (BaseReg != X86::BX && BaseReg != X86::BP && 2032 BaseReg != X86::SI && BaseReg != X86::DI)) && 2033 BaseReg != X86::DX) { 2034 Error(BaseLoc, "invalid 16-bit base register"); 2035 return nullptr; 2036 } 2037 if (BaseReg == 0 && 2038 X86MCRegisterClasses[X86::GR16RegClassID].contains(IndexReg)) { 2039 Error(IndexLoc, "16-bit memory operand may not include only index register"); 2040 return nullptr; 2041 } 2042 2043 StringRef ErrMsg; 2044 if (CheckBaseRegAndIndexReg(BaseReg, IndexReg, ErrMsg)) { 2045 Error(BaseLoc, ErrMsg); 2046 return nullptr; 2047 } 2048 2049 if (SegReg || BaseReg || IndexReg) 2050 return X86Operand::CreateMem(getPointerWidth(), SegReg, Disp, BaseReg, 2051 IndexReg, Scale, MemStart, MemEnd); 2052 return X86Operand::CreateMem(getPointerWidth(), Disp, MemStart, MemEnd); 2053 } 2054 2055 bool X86AsmParser::ParseInstruction(ParseInstructionInfo &Info, StringRef Name, 2056 SMLoc NameLoc, OperandVector &Operands) { 2057 MCAsmParser &Parser = getParser(); 2058 InstInfo = &Info; 2059 StringRef PatchedName = Name; 2060 2061 // FIXME: Hack to recognize setneb as setne. 2062 if (PatchedName.startswith("set") && PatchedName.endswith("b") && 2063 PatchedName != "setb" && PatchedName != "setnb") 2064 PatchedName = PatchedName.substr(0, Name.size()-1); 2065 2066 // FIXME: Hack to recognize cmp<comparison code>{ss,sd,ps,pd}. 2067 if ((PatchedName.startswith("cmp") || PatchedName.startswith("vcmp")) && 2068 (PatchedName.endswith("ss") || PatchedName.endswith("sd") || 2069 PatchedName.endswith("ps") || PatchedName.endswith("pd"))) { 2070 bool IsVCMP = PatchedName[0] == 'v'; 2071 unsigned CCIdx = IsVCMP ? 4 : 3; 2072 unsigned ComparisonCode = StringSwitch<unsigned>( 2073 PatchedName.slice(CCIdx, PatchedName.size() - 2)) 2074 .Case("eq", 0x00) 2075 .Case("lt", 0x01) 2076 .Case("le", 0x02) 2077 .Case("unord", 0x03) 2078 .Case("neq", 0x04) 2079 .Case("nlt", 0x05) 2080 .Case("nle", 0x06) 2081 .Case("ord", 0x07) 2082 /* AVX only from here */ 2083 .Case("eq_uq", 0x08) 2084 .Case("nge", 0x09) 2085 .Case("ngt", 0x0A) 2086 .Case("false", 0x0B) 2087 .Case("neq_oq", 0x0C) 2088 .Case("ge", 0x0D) 2089 .Case("gt", 0x0E) 2090 .Case("true", 0x0F) 2091 .Case("eq_os", 0x10) 2092 .Case("lt_oq", 0x11) 2093 .Case("le_oq", 0x12) 2094 .Case("unord_s", 0x13) 2095 .Case("neq_us", 0x14) 2096 .Case("nlt_uq", 0x15) 2097 .Case("nle_uq", 0x16) 2098 .Case("ord_s", 0x17) 2099 .Case("eq_us", 0x18) 2100 .Case("nge_uq", 0x19) 2101 .Case("ngt_uq", 0x1A) 2102 .Case("false_os", 0x1B) 2103 .Case("neq_os", 0x1C) 2104 .Case("ge_oq", 0x1D) 2105 .Case("gt_oq", 0x1E) 2106 .Case("true_us", 0x1F) 2107 .Default(~0U); 2108 if (ComparisonCode != ~0U && (IsVCMP || ComparisonCode < 8)) { 2109 2110 Operands.push_back(X86Operand::CreateToken(PatchedName.slice(0, CCIdx), 2111 NameLoc)); 2112 2113 const MCExpr *ImmOp = MCConstantExpr::create(ComparisonCode, 2114 getParser().getContext()); 2115 Operands.push_back(X86Operand::CreateImm(ImmOp, NameLoc, NameLoc)); 2116 2117 PatchedName = PatchedName.substr(PatchedName.size() - 2); 2118 } 2119 } 2120 2121 // FIXME: Hack to recognize vpcmp<comparison code>{ub,uw,ud,uq,b,w,d,q}. 2122 if (PatchedName.startswith("vpcmp") && 2123 (PatchedName.endswith("b") || PatchedName.endswith("w") || 2124 PatchedName.endswith("d") || PatchedName.endswith("q"))) { 2125 unsigned CCIdx = PatchedName.drop_back().back() == 'u' ? 2 : 1; 2126 unsigned ComparisonCode = StringSwitch<unsigned>( 2127 PatchedName.slice(5, PatchedName.size() - CCIdx)) 2128 .Case("eq", 0x0) // Only allowed on unsigned. Checked below. 2129 .Case("lt", 0x1) 2130 .Case("le", 0x2) 2131 //.Case("false", 0x3) // Not a documented alias. 2132 .Case("neq", 0x4) 2133 .Case("nlt", 0x5) 2134 .Case("nle", 0x6) 2135 //.Case("true", 0x7) // Not a documented alias. 2136 .Default(~0U); 2137 if (ComparisonCode != ~0U && (ComparisonCode != 0 || CCIdx == 2)) { 2138 Operands.push_back(X86Operand::CreateToken("vpcmp", NameLoc)); 2139 2140 const MCExpr *ImmOp = MCConstantExpr::create(ComparisonCode, 2141 getParser().getContext()); 2142 Operands.push_back(X86Operand::CreateImm(ImmOp, NameLoc, NameLoc)); 2143 2144 PatchedName = PatchedName.substr(PatchedName.size() - CCIdx); 2145 } 2146 } 2147 2148 // FIXME: Hack to recognize vpcom<comparison code>{ub,uw,ud,uq,b,w,d,q}. 2149 if (PatchedName.startswith("vpcom") && 2150 (PatchedName.endswith("b") || PatchedName.endswith("w") || 2151 PatchedName.endswith("d") || PatchedName.endswith("q"))) { 2152 unsigned CCIdx = PatchedName.drop_back().back() == 'u' ? 2 : 1; 2153 unsigned ComparisonCode = StringSwitch<unsigned>( 2154 PatchedName.slice(5, PatchedName.size() - CCIdx)) 2155 .Case("lt", 0x0) 2156 .Case("le", 0x1) 2157 .Case("gt", 0x2) 2158 .Case("ge", 0x3) 2159 .Case("eq", 0x4) 2160 .Case("neq", 0x5) 2161 .Case("false", 0x6) 2162 .Case("true", 0x7) 2163 .Default(~0U); 2164 if (ComparisonCode != ~0U) { 2165 Operands.push_back(X86Operand::CreateToken("vpcom", NameLoc)); 2166 2167 const MCExpr *ImmOp = MCConstantExpr::create(ComparisonCode, 2168 getParser().getContext()); 2169 Operands.push_back(X86Operand::CreateImm(ImmOp, NameLoc, NameLoc)); 2170 2171 PatchedName = PatchedName.substr(PatchedName.size() - CCIdx); 2172 } 2173 } 2174 2175 Operands.push_back(X86Operand::CreateToken(PatchedName, NameLoc)); 2176 2177 // Determine whether this is an instruction prefix. 2178 bool isPrefix = 2179 Name == "lock" || Name == "rep" || 2180 Name == "repe" || Name == "repz" || 2181 Name == "repne" || Name == "repnz" || 2182 Name == "rex64" || Name == "data16"; 2183 2184 // This does the actual operand parsing. Don't parse any more if we have a 2185 // prefix juxtaposed with an operation like "lock incl 4(%rax)", because we 2186 // just want to parse the "lock" as the first instruction and the "incl" as 2187 // the next one. 2188 if (getLexer().isNot(AsmToken::EndOfStatement) && !isPrefix) { 2189 2190 // Parse '*' modifier. 2191 if (getLexer().is(AsmToken::Star)) 2192 Operands.push_back(X86Operand::CreateToken("*", consumeToken())); 2193 2194 // Read the operands. 2195 while(1) { 2196 if (std::unique_ptr<X86Operand> Op = ParseOperand()) { 2197 Operands.push_back(std::move(Op)); 2198 if (!HandleAVX512Operand(Operands, *Operands.back())) 2199 return true; 2200 } else { 2201 Parser.eatToEndOfStatement(); 2202 return true; 2203 } 2204 // check for comma and eat it 2205 if (getLexer().is(AsmToken::Comma)) 2206 Parser.Lex(); 2207 else 2208 break; 2209 } 2210 2211 if (getLexer().isNot(AsmToken::EndOfStatement)) 2212 return ErrorAndEatStatement(getLexer().getLoc(), 2213 "unexpected token in argument list"); 2214 } 2215 2216 // Consume the EndOfStatement or the prefix separator Slash 2217 if (getLexer().is(AsmToken::EndOfStatement) || 2218 (isPrefix && getLexer().is(AsmToken::Slash))) 2219 Parser.Lex(); 2220 2221 // This is for gas compatibility and cannot be done in td. 2222 // Adding "p" for some floating point with no argument. 2223 // For example: fsub --> fsubp 2224 bool IsFp = 2225 Name == "fsub" || Name == "fdiv" || Name == "fsubr" || Name == "fdivr"; 2226 if (IsFp && Operands.size() == 1) { 2227 const char *Repl = StringSwitch<const char *>(Name) 2228 .Case("fsub", "fsubp") 2229 .Case("fdiv", "fdivp") 2230 .Case("fsubr", "fsubrp") 2231 .Case("fdivr", "fdivrp"); 2232 static_cast<X86Operand &>(*Operands[0]).setTokenValue(Repl); 2233 } 2234 2235 // This is a terrible hack to handle "out[bwl]? %al, (%dx)" -> 2236 // "outb %al, %dx". Out doesn't take a memory form, but this is a widely 2237 // documented form in various unofficial manuals, so a lot of code uses it. 2238 if ((Name == "outb" || Name == "outw" || Name == "outl" || Name == "out") && 2239 Operands.size() == 3) { 2240 X86Operand &Op = (X86Operand &)*Operands.back(); 2241 if (Op.isMem() && Op.Mem.SegReg == 0 && 2242 isa<MCConstantExpr>(Op.Mem.Disp) && 2243 cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 && 2244 Op.Mem.BaseReg == MatchRegisterName("dx") && Op.Mem.IndexReg == 0) { 2245 SMLoc Loc = Op.getEndLoc(); 2246 Operands.back() = X86Operand::CreateReg(Op.Mem.BaseReg, Loc, Loc); 2247 } 2248 } 2249 // Same hack for "in[bwl]? (%dx), %al" -> "inb %dx, %al". 2250 if ((Name == "inb" || Name == "inw" || Name == "inl" || Name == "in") && 2251 Operands.size() == 3) { 2252 X86Operand &Op = (X86Operand &)*Operands[1]; 2253 if (Op.isMem() && Op.Mem.SegReg == 0 && 2254 isa<MCConstantExpr>(Op.Mem.Disp) && 2255 cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 && 2256 Op.Mem.BaseReg == MatchRegisterName("dx") && Op.Mem.IndexReg == 0) { 2257 SMLoc Loc = Op.getEndLoc(); 2258 Operands[1] = X86Operand::CreateReg(Op.Mem.BaseReg, Loc, Loc); 2259 } 2260 } 2261 2262 // Append default arguments to "ins[bwld]" 2263 if (Name.startswith("ins") && Operands.size() == 1 && 2264 (Name == "insb" || Name == "insw" || Name == "insl" || Name == "insd")) { 2265 AddDefaultSrcDestOperands(Operands, 2266 X86Operand::CreateReg(X86::DX, NameLoc, NameLoc), 2267 DefaultMemDIOperand(NameLoc)); 2268 } 2269 2270 // Append default arguments to "outs[bwld]" 2271 if (Name.startswith("outs") && Operands.size() == 1 && 2272 (Name == "outsb" || Name == "outsw" || Name == "outsl" || 2273 Name == "outsd" )) { 2274 AddDefaultSrcDestOperands(Operands, 2275 DefaultMemSIOperand(NameLoc), 2276 X86Operand::CreateReg(X86::DX, NameLoc, NameLoc)); 2277 } 2278 2279 // Transform "lods[bwlq]" into "lods[bwlq] ($SIREG)" for appropriate 2280 // values of $SIREG according to the mode. It would be nice if this 2281 // could be achieved with InstAlias in the tables. 2282 if (Name.startswith("lods") && Operands.size() == 1 && 2283 (Name == "lods" || Name == "lodsb" || Name == "lodsw" || 2284 Name == "lodsl" || Name == "lodsd" || Name == "lodsq")) 2285 Operands.push_back(DefaultMemSIOperand(NameLoc)); 2286 2287 // Transform "stos[bwlq]" into "stos[bwlq] ($DIREG)" for appropriate 2288 // values of $DIREG according to the mode. It would be nice if this 2289 // could be achieved with InstAlias in the tables. 2290 if (Name.startswith("stos") && Operands.size() == 1 && 2291 (Name == "stos" || Name == "stosb" || Name == "stosw" || 2292 Name == "stosl" || Name == "stosd" || Name == "stosq")) 2293 Operands.push_back(DefaultMemDIOperand(NameLoc)); 2294 2295 // Transform "scas[bwlq]" into "scas[bwlq] ($DIREG)" for appropriate 2296 // values of $DIREG according to the mode. It would be nice if this 2297 // could be achieved with InstAlias in the tables. 2298 if (Name.startswith("scas") && Operands.size() == 1 && 2299 (Name == "scas" || Name == "scasb" || Name == "scasw" || 2300 Name == "scasl" || Name == "scasd" || Name == "scasq")) 2301 Operands.push_back(DefaultMemDIOperand(NameLoc)); 2302 2303 // Add default SI and DI operands to "cmps[bwlq]". 2304 if (Name.startswith("cmps") && 2305 (Name == "cmps" || Name == "cmpsb" || Name == "cmpsw" || 2306 Name == "cmpsl" || Name == "cmpsd" || Name == "cmpsq")) { 2307 if (Operands.size() == 1) { 2308 AddDefaultSrcDestOperands(Operands, 2309 DefaultMemDIOperand(NameLoc), 2310 DefaultMemSIOperand(NameLoc)); 2311 } else if (Operands.size() == 3) { 2312 X86Operand &Op = (X86Operand &)*Operands[1]; 2313 X86Operand &Op2 = (X86Operand &)*Operands[2]; 2314 if (!doSrcDstMatch(Op, Op2)) 2315 return Error(Op.getStartLoc(), 2316 "mismatching source and destination index registers"); 2317 } 2318 } 2319 2320 // Add default SI and DI operands to "movs[bwlq]". 2321 if ((Name.startswith("movs") && 2322 (Name == "movs" || Name == "movsb" || Name == "movsw" || 2323 Name == "movsl" || Name == "movsd" || Name == "movsq")) || 2324 (Name.startswith("smov") && 2325 (Name == "smov" || Name == "smovb" || Name == "smovw" || 2326 Name == "smovl" || Name == "smovd" || Name == "smovq"))) { 2327 if (Operands.size() == 1) { 2328 if (Name == "movsd") 2329 Operands.back() = X86Operand::CreateToken("movsl", NameLoc); 2330 AddDefaultSrcDestOperands(Operands, 2331 DefaultMemSIOperand(NameLoc), 2332 DefaultMemDIOperand(NameLoc)); 2333 } else if (Operands.size() == 3) { 2334 X86Operand &Op = (X86Operand &)*Operands[1]; 2335 X86Operand &Op2 = (X86Operand &)*Operands[2]; 2336 if (!doSrcDstMatch(Op, Op2)) 2337 return Error(Op.getStartLoc(), 2338 "mismatching source and destination index registers"); 2339 } 2340 } 2341 2342 // FIXME: Hack to handle recognize s{hr,ar,hl} $1, <op>. Canonicalize to 2343 // "shift <op>". 2344 if ((Name.startswith("shr") || Name.startswith("sar") || 2345 Name.startswith("shl") || Name.startswith("sal") || 2346 Name.startswith("rcl") || Name.startswith("rcr") || 2347 Name.startswith("rol") || Name.startswith("ror")) && 2348 Operands.size() == 3) { 2349 if (isParsingIntelSyntax()) { 2350 // Intel syntax 2351 X86Operand &Op1 = static_cast<X86Operand &>(*Operands[2]); 2352 if (Op1.isImm() && isa<MCConstantExpr>(Op1.getImm()) && 2353 cast<MCConstantExpr>(Op1.getImm())->getValue() == 1) 2354 Operands.pop_back(); 2355 } else { 2356 X86Operand &Op1 = static_cast<X86Operand &>(*Operands[1]); 2357 if (Op1.isImm() && isa<MCConstantExpr>(Op1.getImm()) && 2358 cast<MCConstantExpr>(Op1.getImm())->getValue() == 1) 2359 Operands.erase(Operands.begin() + 1); 2360 } 2361 } 2362 2363 // Transforms "int $3" into "int3" as a size optimization. We can't write an 2364 // instalias with an immediate operand yet. 2365 if (Name == "int" && Operands.size() == 2) { 2366 X86Operand &Op1 = static_cast<X86Operand &>(*Operands[1]); 2367 if (Op1.isImm()) 2368 if (auto *CE = dyn_cast<MCConstantExpr>(Op1.getImm())) 2369 if (CE->getValue() == 3) { 2370 Operands.erase(Operands.begin() + 1); 2371 static_cast<X86Operand &>(*Operands[0]).setTokenValue("int3"); 2372 } 2373 } 2374 2375 return false; 2376 } 2377 2378 bool X86AsmParser::processInstruction(MCInst &Inst, const OperandVector &Ops) { 2379 switch (Inst.getOpcode()) { 2380 default: return false; 2381 case X86::VMOVZPQILo2PQIrr: 2382 case X86::VMOVAPDrr: 2383 case X86::VMOVAPDYrr: 2384 case X86::VMOVAPSrr: 2385 case X86::VMOVAPSYrr: 2386 case X86::VMOVDQArr: 2387 case X86::VMOVDQAYrr: 2388 case X86::VMOVDQUrr: 2389 case X86::VMOVDQUYrr: 2390 case X86::VMOVUPDrr: 2391 case X86::VMOVUPDYrr: 2392 case X86::VMOVUPSrr: 2393 case X86::VMOVUPSYrr: { 2394 if (X86II::isX86_64ExtendedReg(Inst.getOperand(0).getReg()) || 2395 !X86II::isX86_64ExtendedReg(Inst.getOperand(1).getReg())) 2396 return false; 2397 2398 unsigned NewOpc; 2399 switch (Inst.getOpcode()) { 2400 default: llvm_unreachable("Invalid opcode"); 2401 case X86::VMOVZPQILo2PQIrr: NewOpc = X86::VMOVPQI2QIrr; break; 2402 case X86::VMOVAPDrr: NewOpc = X86::VMOVAPDrr_REV; break; 2403 case X86::VMOVAPDYrr: NewOpc = X86::VMOVAPDYrr_REV; break; 2404 case X86::VMOVAPSrr: NewOpc = X86::VMOVAPSrr_REV; break; 2405 case X86::VMOVAPSYrr: NewOpc = X86::VMOVAPSYrr_REV; break; 2406 case X86::VMOVDQArr: NewOpc = X86::VMOVDQArr_REV; break; 2407 case X86::VMOVDQAYrr: NewOpc = X86::VMOVDQAYrr_REV; break; 2408 case X86::VMOVDQUrr: NewOpc = X86::VMOVDQUrr_REV; break; 2409 case X86::VMOVDQUYrr: NewOpc = X86::VMOVDQUYrr_REV; break; 2410 case X86::VMOVUPDrr: NewOpc = X86::VMOVUPDrr_REV; break; 2411 case X86::VMOVUPDYrr: NewOpc = X86::VMOVUPDYrr_REV; break; 2412 case X86::VMOVUPSrr: NewOpc = X86::VMOVUPSrr_REV; break; 2413 case X86::VMOVUPSYrr: NewOpc = X86::VMOVUPSYrr_REV; break; 2414 } 2415 Inst.setOpcode(NewOpc); 2416 return true; 2417 } 2418 case X86::VMOVSDrr: 2419 case X86::VMOVSSrr: { 2420 if (X86II::isX86_64ExtendedReg(Inst.getOperand(0).getReg()) || 2421 !X86II::isX86_64ExtendedReg(Inst.getOperand(2).getReg())) 2422 return false; 2423 unsigned NewOpc; 2424 switch (Inst.getOpcode()) { 2425 default: llvm_unreachable("Invalid opcode"); 2426 case X86::VMOVSDrr: NewOpc = X86::VMOVSDrr_REV; break; 2427 case X86::VMOVSSrr: NewOpc = X86::VMOVSSrr_REV; break; 2428 } 2429 Inst.setOpcode(NewOpc); 2430 return true; 2431 } 2432 } 2433 } 2434 2435 static const char *getSubtargetFeatureName(uint64_t Val); 2436 2437 void X86AsmParser::EmitInstruction(MCInst &Inst, OperandVector &Operands, 2438 MCStreamer &Out) { 2439 Instrumentation->InstrumentAndEmitInstruction(Inst, Operands, getContext(), 2440 MII, Out); 2441 } 2442 2443 bool X86AsmParser::MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode, 2444 OperandVector &Operands, 2445 MCStreamer &Out, uint64_t &ErrorInfo, 2446 bool MatchingInlineAsm) { 2447 if (isParsingIntelSyntax()) 2448 return MatchAndEmitIntelInstruction(IDLoc, Opcode, Operands, Out, ErrorInfo, 2449 MatchingInlineAsm); 2450 return MatchAndEmitATTInstruction(IDLoc, Opcode, Operands, Out, ErrorInfo, 2451 MatchingInlineAsm); 2452 } 2453 2454 void X86AsmParser::MatchFPUWaitAlias(SMLoc IDLoc, X86Operand &Op, 2455 OperandVector &Operands, MCStreamer &Out, 2456 bool MatchingInlineAsm) { 2457 // FIXME: This should be replaced with a real .td file alias mechanism. 2458 // Also, MatchInstructionImpl should actually *do* the EmitInstruction 2459 // call. 2460 const char *Repl = StringSwitch<const char *>(Op.getToken()) 2461 .Case("finit", "fninit") 2462 .Case("fsave", "fnsave") 2463 .Case("fstcw", "fnstcw") 2464 .Case("fstcww", "fnstcw") 2465 .Case("fstenv", "fnstenv") 2466 .Case("fstsw", "fnstsw") 2467 .Case("fstsww", "fnstsw") 2468 .Case("fclex", "fnclex") 2469 .Default(nullptr); 2470 if (Repl) { 2471 MCInst Inst; 2472 Inst.setOpcode(X86::WAIT); 2473 Inst.setLoc(IDLoc); 2474 if (!MatchingInlineAsm) 2475 EmitInstruction(Inst, Operands, Out); 2476 Operands[0] = X86Operand::CreateToken(Repl, IDLoc); 2477 } 2478 } 2479 2480 bool X86AsmParser::ErrorMissingFeature(SMLoc IDLoc, uint64_t ErrorInfo, 2481 bool MatchingInlineAsm) { 2482 assert(ErrorInfo && "Unknown missing feature!"); 2483 ArrayRef<SMRange> EmptyRanges = None; 2484 SmallString<126> Msg; 2485 raw_svector_ostream OS(Msg); 2486 OS << "instruction requires:"; 2487 uint64_t Mask = 1; 2488 for (unsigned i = 0; i < (sizeof(ErrorInfo)*8-1); ++i) { 2489 if (ErrorInfo & Mask) 2490 OS << ' ' << getSubtargetFeatureName(ErrorInfo & Mask); 2491 Mask <<= 1; 2492 } 2493 return Error(IDLoc, OS.str(), EmptyRanges, MatchingInlineAsm); 2494 } 2495 2496 bool X86AsmParser::MatchAndEmitATTInstruction(SMLoc IDLoc, unsigned &Opcode, 2497 OperandVector &Operands, 2498 MCStreamer &Out, 2499 uint64_t &ErrorInfo, 2500 bool MatchingInlineAsm) { 2501 assert(!Operands.empty() && "Unexpect empty operand list!"); 2502 X86Operand &Op = static_cast<X86Operand &>(*Operands[0]); 2503 assert(Op.isToken() && "Leading operand should always be a mnemonic!"); 2504 ArrayRef<SMRange> EmptyRanges = None; 2505 2506 // First, handle aliases that expand to multiple instructions. 2507 MatchFPUWaitAlias(IDLoc, Op, Operands, Out, MatchingInlineAsm); 2508 2509 bool WasOriginallyInvalidOperand = false; 2510 MCInst Inst; 2511 2512 // First, try a direct match. 2513 switch (MatchInstructionImpl(Operands, Inst, 2514 ErrorInfo, MatchingInlineAsm, 2515 isParsingIntelSyntax())) { 2516 default: llvm_unreachable("Unexpected match result!"); 2517 case Match_Success: 2518 // Some instructions need post-processing to, for example, tweak which 2519 // encoding is selected. Loop on it while changes happen so the 2520 // individual transformations can chain off each other. 2521 if (!MatchingInlineAsm) 2522 while (processInstruction(Inst, Operands)) 2523 ; 2524 2525 Inst.setLoc(IDLoc); 2526 if (!MatchingInlineAsm) 2527 EmitInstruction(Inst, Operands, Out); 2528 Opcode = Inst.getOpcode(); 2529 return false; 2530 case Match_MissingFeature: 2531 return ErrorMissingFeature(IDLoc, ErrorInfo, MatchingInlineAsm); 2532 case Match_InvalidOperand: 2533 WasOriginallyInvalidOperand = true; 2534 break; 2535 case Match_MnemonicFail: 2536 break; 2537 } 2538 2539 // FIXME: Ideally, we would only attempt suffix matches for things which are 2540 // valid prefixes, and we could just infer the right unambiguous 2541 // type. However, that requires substantially more matcher support than the 2542 // following hack. 2543 2544 // Change the operand to point to a temporary token. 2545 StringRef Base = Op.getToken(); 2546 SmallString<16> Tmp; 2547 Tmp += Base; 2548 Tmp += ' '; 2549 Op.setTokenValue(Tmp); 2550 2551 // If this instruction starts with an 'f', then it is a floating point stack 2552 // instruction. These come in up to three forms for 32-bit, 64-bit, and 2553 // 80-bit floating point, which use the suffixes s,l,t respectively. 2554 // 2555 // Otherwise, we assume that this may be an integer instruction, which comes 2556 // in 8/16/32/64-bit forms using the b,w,l,q suffixes respectively. 2557 const char *Suffixes = Base[0] != 'f' ? "bwlq" : "slt\0"; 2558 2559 // Check for the various suffix matches. 2560 uint64_t ErrorInfoIgnore; 2561 uint64_t ErrorInfoMissingFeature = 0; // Init suppresses compiler warnings. 2562 unsigned Match[4]; 2563 2564 for (unsigned I = 0, E = array_lengthof(Match); I != E; ++I) { 2565 Tmp.back() = Suffixes[I]; 2566 Match[I] = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore, 2567 MatchingInlineAsm, isParsingIntelSyntax()); 2568 // If this returned as a missing feature failure, remember that. 2569 if (Match[I] == Match_MissingFeature) 2570 ErrorInfoMissingFeature = ErrorInfoIgnore; 2571 } 2572 2573 // Restore the old token. 2574 Op.setTokenValue(Base); 2575 2576 // If exactly one matched, then we treat that as a successful match (and the 2577 // instruction will already have been filled in correctly, since the failing 2578 // matches won't have modified it). 2579 unsigned NumSuccessfulMatches = 2580 std::count(std::begin(Match), std::end(Match), Match_Success); 2581 if (NumSuccessfulMatches == 1) { 2582 Inst.setLoc(IDLoc); 2583 if (!MatchingInlineAsm) 2584 EmitInstruction(Inst, Operands, Out); 2585 Opcode = Inst.getOpcode(); 2586 return false; 2587 } 2588 2589 // Otherwise, the match failed, try to produce a decent error message. 2590 2591 // If we had multiple suffix matches, then identify this as an ambiguous 2592 // match. 2593 if (NumSuccessfulMatches > 1) { 2594 char MatchChars[4]; 2595 unsigned NumMatches = 0; 2596 for (unsigned I = 0, E = array_lengthof(Match); I != E; ++I) 2597 if (Match[I] == Match_Success) 2598 MatchChars[NumMatches++] = Suffixes[I]; 2599 2600 SmallString<126> Msg; 2601 raw_svector_ostream OS(Msg); 2602 OS << "ambiguous instructions require an explicit suffix (could be "; 2603 for (unsigned i = 0; i != NumMatches; ++i) { 2604 if (i != 0) 2605 OS << ", "; 2606 if (i + 1 == NumMatches) 2607 OS << "or "; 2608 OS << "'" << Base << MatchChars[i] << "'"; 2609 } 2610 OS << ")"; 2611 Error(IDLoc, OS.str(), EmptyRanges, MatchingInlineAsm); 2612 return true; 2613 } 2614 2615 // Okay, we know that none of the variants matched successfully. 2616 2617 // If all of the instructions reported an invalid mnemonic, then the original 2618 // mnemonic was invalid. 2619 if (std::count(std::begin(Match), std::end(Match), Match_MnemonicFail) == 4) { 2620 if (!WasOriginallyInvalidOperand) { 2621 ArrayRef<SMRange> Ranges = 2622 MatchingInlineAsm ? EmptyRanges : Op.getLocRange(); 2623 return Error(IDLoc, "invalid instruction mnemonic '" + Base + "'", 2624 Ranges, MatchingInlineAsm); 2625 } 2626 2627 // Recover location info for the operand if we know which was the problem. 2628 if (ErrorInfo != ~0ULL) { 2629 if (ErrorInfo >= Operands.size()) 2630 return Error(IDLoc, "too few operands for instruction", 2631 EmptyRanges, MatchingInlineAsm); 2632 2633 X86Operand &Operand = (X86Operand &)*Operands[ErrorInfo]; 2634 if (Operand.getStartLoc().isValid()) { 2635 SMRange OperandRange = Operand.getLocRange(); 2636 return Error(Operand.getStartLoc(), "invalid operand for instruction", 2637 OperandRange, MatchingInlineAsm); 2638 } 2639 } 2640 2641 return Error(IDLoc, "invalid operand for instruction", EmptyRanges, 2642 MatchingInlineAsm); 2643 } 2644 2645 // If one instruction matched with a missing feature, report this as a 2646 // missing feature. 2647 if (std::count(std::begin(Match), std::end(Match), 2648 Match_MissingFeature) == 1) { 2649 ErrorInfo = ErrorInfoMissingFeature; 2650 return ErrorMissingFeature(IDLoc, ErrorInfoMissingFeature, 2651 MatchingInlineAsm); 2652 } 2653 2654 // If one instruction matched with an invalid operand, report this as an 2655 // operand failure. 2656 if (std::count(std::begin(Match), std::end(Match), 2657 Match_InvalidOperand) == 1) { 2658 return Error(IDLoc, "invalid operand for instruction", EmptyRanges, 2659 MatchingInlineAsm); 2660 } 2661 2662 // If all of these were an outright failure, report it in a useless way. 2663 Error(IDLoc, "unknown use of instruction mnemonic without a size suffix", 2664 EmptyRanges, MatchingInlineAsm); 2665 return true; 2666 } 2667 2668 bool X86AsmParser::MatchAndEmitIntelInstruction(SMLoc IDLoc, unsigned &Opcode, 2669 OperandVector &Operands, 2670 MCStreamer &Out, 2671 uint64_t &ErrorInfo, 2672 bool MatchingInlineAsm) { 2673 assert(!Operands.empty() && "Unexpect empty operand list!"); 2674 X86Operand &Op = static_cast<X86Operand &>(*Operands[0]); 2675 assert(Op.isToken() && "Leading operand should always be a mnemonic!"); 2676 StringRef Mnemonic = Op.getToken(); 2677 ArrayRef<SMRange> EmptyRanges = None; 2678 2679 // First, handle aliases that expand to multiple instructions. 2680 MatchFPUWaitAlias(IDLoc, Op, Operands, Out, MatchingInlineAsm); 2681 2682 MCInst Inst; 2683 2684 // Find one unsized memory operand, if present. 2685 X86Operand *UnsizedMemOp = nullptr; 2686 for (const auto &Op : Operands) { 2687 X86Operand *X86Op = static_cast<X86Operand *>(Op.get()); 2688 if (X86Op->isMemUnsized()) 2689 UnsizedMemOp = X86Op; 2690 } 2691 2692 // Allow some instructions to have implicitly pointer-sized operands. This is 2693 // compatible with gas. 2694 if (UnsizedMemOp) { 2695 static const char *const PtrSizedInstrs[] = {"call", "jmp", "push"}; 2696 for (const char *Instr : PtrSizedInstrs) { 2697 if (Mnemonic == Instr) { 2698 UnsizedMemOp->Mem.Size = getPointerWidth(); 2699 break; 2700 } 2701 } 2702 } 2703 2704 // If an unsized memory operand is present, try to match with each memory 2705 // operand size. In Intel assembly, the size is not part of the instruction 2706 // mnemonic. 2707 SmallVector<unsigned, 8> Match; 2708 uint64_t ErrorInfoMissingFeature = 0; 2709 if (UnsizedMemOp && UnsizedMemOp->isMemUnsized()) { 2710 static const unsigned MopSizes[] = {8, 16, 32, 64, 80, 128, 256, 512}; 2711 for (unsigned Size : MopSizes) { 2712 UnsizedMemOp->Mem.Size = Size; 2713 uint64_t ErrorInfoIgnore; 2714 unsigned LastOpcode = Inst.getOpcode(); 2715 unsigned M = 2716 MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore, 2717 MatchingInlineAsm, isParsingIntelSyntax()); 2718 if (Match.empty() || LastOpcode != Inst.getOpcode()) 2719 Match.push_back(M); 2720 2721 // If this returned as a missing feature failure, remember that. 2722 if (Match.back() == Match_MissingFeature) 2723 ErrorInfoMissingFeature = ErrorInfoIgnore; 2724 } 2725 2726 // Restore the size of the unsized memory operand if we modified it. 2727 if (UnsizedMemOp) 2728 UnsizedMemOp->Mem.Size = 0; 2729 } 2730 2731 // If we haven't matched anything yet, this is not a basic integer or FPU 2732 // operation. There shouldn't be any ambiguity in our mnemonic table, so try 2733 // matching with the unsized operand. 2734 if (Match.empty()) { 2735 Match.push_back(MatchInstructionImpl(Operands, Inst, ErrorInfo, 2736 MatchingInlineAsm, 2737 isParsingIntelSyntax())); 2738 // If this returned as a missing feature failure, remember that. 2739 if (Match.back() == Match_MissingFeature) 2740 ErrorInfoMissingFeature = ErrorInfo; 2741 } 2742 2743 // Restore the size of the unsized memory operand if we modified it. 2744 if (UnsizedMemOp) 2745 UnsizedMemOp->Mem.Size = 0; 2746 2747 // If it's a bad mnemonic, all results will be the same. 2748 if (Match.back() == Match_MnemonicFail) { 2749 ArrayRef<SMRange> Ranges = 2750 MatchingInlineAsm ? EmptyRanges : Op.getLocRange(); 2751 return Error(IDLoc, "invalid instruction mnemonic '" + Mnemonic + "'", 2752 Ranges, MatchingInlineAsm); 2753 } 2754 2755 // If exactly one matched, then we treat that as a successful match (and the 2756 // instruction will already have been filled in correctly, since the failing 2757 // matches won't have modified it). 2758 unsigned NumSuccessfulMatches = 2759 std::count(std::begin(Match), std::end(Match), Match_Success); 2760 if (NumSuccessfulMatches == 1) { 2761 // Some instructions need post-processing to, for example, tweak which 2762 // encoding is selected. Loop on it while changes happen so the individual 2763 // transformations can chain off each other. 2764 if (!MatchingInlineAsm) 2765 while (processInstruction(Inst, Operands)) 2766 ; 2767 Inst.setLoc(IDLoc); 2768 if (!MatchingInlineAsm) 2769 EmitInstruction(Inst, Operands, Out); 2770 Opcode = Inst.getOpcode(); 2771 return false; 2772 } else if (NumSuccessfulMatches > 1) { 2773 assert(UnsizedMemOp && 2774 "multiple matches only possible with unsized memory operands"); 2775 ArrayRef<SMRange> Ranges = 2776 MatchingInlineAsm ? EmptyRanges : UnsizedMemOp->getLocRange(); 2777 return Error(UnsizedMemOp->getStartLoc(), 2778 "ambiguous operand size for instruction '" + Mnemonic + "\'", 2779 Ranges, MatchingInlineAsm); 2780 } 2781 2782 // If one instruction matched with a missing feature, report this as a 2783 // missing feature. 2784 if (std::count(std::begin(Match), std::end(Match), 2785 Match_MissingFeature) == 1) { 2786 ErrorInfo = ErrorInfoMissingFeature; 2787 return ErrorMissingFeature(IDLoc, ErrorInfoMissingFeature, 2788 MatchingInlineAsm); 2789 } 2790 2791 // If one instruction matched with an invalid operand, report this as an 2792 // operand failure. 2793 if (std::count(std::begin(Match), std::end(Match), 2794 Match_InvalidOperand) == 1) { 2795 return Error(IDLoc, "invalid operand for instruction", EmptyRanges, 2796 MatchingInlineAsm); 2797 } 2798 2799 // If all of these were an outright failure, report it in a useless way. 2800 return Error(IDLoc, "unknown instruction mnemonic", EmptyRanges, 2801 MatchingInlineAsm); 2802 } 2803 2804 bool X86AsmParser::OmitRegisterFromClobberLists(unsigned RegNo) { 2805 return X86MCRegisterClasses[X86::SEGMENT_REGRegClassID].contains(RegNo); 2806 } 2807 2808 bool X86AsmParser::ParseDirective(AsmToken DirectiveID) { 2809 MCAsmParser &Parser = getParser(); 2810 StringRef IDVal = DirectiveID.getIdentifier(); 2811 if (IDVal == ".word") 2812 return ParseDirectiveWord(2, DirectiveID.getLoc()); 2813 else if (IDVal.startswith(".code")) 2814 return ParseDirectiveCode(IDVal, DirectiveID.getLoc()); 2815 else if (IDVal.startswith(".att_syntax")) { 2816 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2817 if (Parser.getTok().getString() == "prefix") 2818 Parser.Lex(); 2819 else if (Parser.getTok().getString() == "noprefix") 2820 return Error(DirectiveID.getLoc(), "'.att_syntax noprefix' is not " 2821 "supported: registers must have a " 2822 "'%' prefix in .att_syntax"); 2823 } 2824 getParser().setAssemblerDialect(0); 2825 return false; 2826 } else if (IDVal.startswith(".intel_syntax")) { 2827 getParser().setAssemblerDialect(1); 2828 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2829 if (Parser.getTok().getString() == "noprefix") 2830 Parser.Lex(); 2831 else if (Parser.getTok().getString() == "prefix") 2832 return Error(DirectiveID.getLoc(), "'.intel_syntax prefix' is not " 2833 "supported: registers must not have " 2834 "a '%' prefix in .intel_syntax"); 2835 } 2836 return false; 2837 } 2838 return true; 2839 } 2840 2841 /// ParseDirectiveWord 2842 /// ::= .word [ expression (, expression)* ] 2843 bool X86AsmParser::ParseDirectiveWord(unsigned Size, SMLoc L) { 2844 MCAsmParser &Parser = getParser(); 2845 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2846 for (;;) { 2847 const MCExpr *Value; 2848 SMLoc ExprLoc = getLexer().getLoc(); 2849 if (getParser().parseExpression(Value)) 2850 return false; 2851 2852 if (const auto *MCE = dyn_cast<MCConstantExpr>(Value)) { 2853 assert(Size <= 8 && "Invalid size"); 2854 uint64_t IntValue = MCE->getValue(); 2855 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue)) 2856 return Error(ExprLoc, "literal value out of range for directive"); 2857 getStreamer().EmitIntValue(IntValue, Size); 2858 } else { 2859 getStreamer().EmitValue(Value, Size, ExprLoc); 2860 } 2861 2862 if (getLexer().is(AsmToken::EndOfStatement)) 2863 break; 2864 2865 // FIXME: Improve diagnostic. 2866 if (getLexer().isNot(AsmToken::Comma)) { 2867 Error(L, "unexpected token in directive"); 2868 return false; 2869 } 2870 Parser.Lex(); 2871 } 2872 } 2873 2874 Parser.Lex(); 2875 return false; 2876 } 2877 2878 /// ParseDirectiveCode 2879 /// ::= .code16 | .code32 | .code64 2880 bool X86AsmParser::ParseDirectiveCode(StringRef IDVal, SMLoc L) { 2881 MCAsmParser &Parser = getParser(); 2882 if (IDVal == ".code16") { 2883 Parser.Lex(); 2884 if (!is16BitMode()) { 2885 SwitchMode(X86::Mode16Bit); 2886 getParser().getStreamer().EmitAssemblerFlag(MCAF_Code16); 2887 } 2888 } else if (IDVal == ".code32") { 2889 Parser.Lex(); 2890 if (!is32BitMode()) { 2891 SwitchMode(X86::Mode32Bit); 2892 getParser().getStreamer().EmitAssemblerFlag(MCAF_Code32); 2893 } 2894 } else if (IDVal == ".code64") { 2895 Parser.Lex(); 2896 if (!is64BitMode()) { 2897 SwitchMode(X86::Mode64Bit); 2898 getParser().getStreamer().EmitAssemblerFlag(MCAF_Code64); 2899 } 2900 } else { 2901 Error(L, "unknown directive " + IDVal); 2902 return false; 2903 } 2904 2905 return false; 2906 } 2907 2908 // Force static initialization. 2909 extern "C" void LLVMInitializeX86AsmParser() { 2910 RegisterMCAsmParser<X86AsmParser> X(TheX86_32Target); 2911 RegisterMCAsmParser<X86AsmParser> Y(TheX86_64Target); 2912 } 2913 2914 #define GET_REGISTER_MATCHER 2915 #define GET_MATCHER_IMPLEMENTATION 2916 #define GET_SUBTARGET_FEATURE_NAME 2917 #include "X86GenAsmMatcher.inc" 2918