X86AsmParser.cpp revision 5ea1ea8db810aaf51f68e32119006629a2f06f30
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 "llvm/ADT/APFloat.h" 12#include "llvm/ADT/SmallString.h" 13#include "llvm/ADT/SmallVector.h" 14#include "llvm/ADT/StringSwitch.h" 15#include "llvm/ADT/Twine.h" 16#include "llvm/MC/MCContext.h" 17#include "llvm/MC/MCExpr.h" 18#include "llvm/MC/MCInst.h" 19#include "llvm/MC/MCParser/MCAsmLexer.h" 20#include "llvm/MC/MCParser/MCAsmParser.h" 21#include "llvm/MC/MCParser/MCParsedAsmOperand.h" 22#include "llvm/MC/MCRegisterInfo.h" 23#include "llvm/MC/MCStreamer.h" 24#include "llvm/MC/MCSubtargetInfo.h" 25#include "llvm/MC/MCSymbol.h" 26#include "llvm/MC/MCTargetAsmParser.h" 27#include "llvm/Support/SourceMgr.h" 28#include "llvm/Support/TargetRegistry.h" 29#include "llvm/Support/raw_ostream.h" 30 31using namespace llvm; 32 33namespace { 34struct X86Operand; 35 36class X86AsmParser : public MCTargetAsmParser { 37 MCSubtargetInfo &STI; 38 MCAsmParser &Parser; 39 ParseInstructionInfo *InstInfo; 40private: 41 MCAsmParser &getParser() const { return Parser; } 42 43 MCAsmLexer &getLexer() const { return Parser.getLexer(); } 44 45 bool Error(SMLoc L, const Twine &Msg, 46 ArrayRef<SMRange> Ranges = ArrayRef<SMRange>(), 47 bool MatchingInlineAsm = false) { 48 if (MatchingInlineAsm) return true; 49 return Parser.Error(L, Msg, Ranges); 50 } 51 52 X86Operand *ErrorOperand(SMLoc Loc, StringRef Msg) { 53 Error(Loc, Msg); 54 return 0; 55 } 56 57 X86Operand *ParseOperand(); 58 X86Operand *ParseATTOperand(); 59 X86Operand *ParseIntelOperand(); 60 X86Operand *ParseIntelOffsetOfOperator(); 61 X86Operand *ParseIntelOperator(unsigned OpKind); 62 X86Operand *ParseIntelMemOperand(unsigned SegReg, uint64_t ImmDisp, 63 SMLoc StartLoc); 64 X86Operand *ParseIntelBracExpression(unsigned SegReg, SMLoc SizeDirLoc, 65 uint64_t ImmDisp, unsigned Size); 66 X86Operand *ParseIntelVarWithQualifier(const MCExpr *&Disp, 67 StringRef &Identifier); 68 X86Operand *ParseMemOperand(unsigned SegReg, SMLoc StartLoc); 69 70 X86Operand *CreateMemForInlineAsm(const MCExpr *Disp, SMLoc Start, SMLoc End, 71 SMLoc SizeDirLoc, unsigned Size, 72 StringRef SymName); 73 74 bool ParseIntelDotOperator(const MCExpr *Disp, const MCExpr **NewDisp, 75 SmallString<64> &Err); 76 77 bool ParseDirectiveWord(unsigned Size, SMLoc L); 78 bool ParseDirectiveCode(StringRef IDVal, SMLoc L); 79 80 bool processInstruction(MCInst &Inst, 81 const SmallVectorImpl<MCParsedAsmOperand*> &Ops); 82 83 bool MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode, 84 SmallVectorImpl<MCParsedAsmOperand*> &Operands, 85 MCStreamer &Out, unsigned &ErrorInfo, 86 bool MatchingInlineAsm); 87 88 /// isSrcOp - Returns true if operand is either (%rsi) or %ds:%(rsi) 89 /// in 64bit mode or (%esi) or %es:(%esi) in 32bit mode. 90 bool isSrcOp(X86Operand &Op); 91 92 /// isDstOp - Returns true if operand is either (%rdi) or %es:(%rdi) 93 /// in 64bit mode or (%edi) or %es:(%edi) in 32bit mode. 94 bool isDstOp(X86Operand &Op); 95 96 bool is64BitMode() const { 97 // FIXME: Can tablegen auto-generate this? 98 return (STI.getFeatureBits() & X86::Mode64Bit) != 0; 99 } 100 void SwitchMode() { 101 unsigned FB = ComputeAvailableFeatures(STI.ToggleFeature(X86::Mode64Bit)); 102 setAvailableFeatures(FB); 103 } 104 105 /// @name Auto-generated Matcher Functions 106 /// { 107 108#define GET_ASSEMBLER_HEADER 109#include "X86GenAsmMatcher.inc" 110 111 /// } 112 113public: 114 X86AsmParser(MCSubtargetInfo &sti, MCAsmParser &parser) 115 : MCTargetAsmParser(), STI(sti), Parser(parser), InstInfo(0) { 116 117 // Initialize the set of available features. 118 setAvailableFeatures(ComputeAvailableFeatures(STI.getFeatureBits())); 119 } 120 virtual bool ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &EndLoc); 121 122 virtual bool ParseInstruction(ParseInstructionInfo &Info, StringRef Name, 123 SMLoc NameLoc, 124 SmallVectorImpl<MCParsedAsmOperand*> &Operands); 125 126 virtual bool ParseDirective(AsmToken DirectiveID); 127 128 bool isParsingIntelSyntax() { 129 return getParser().getAssemblerDialect(); 130 } 131}; 132} // end anonymous namespace 133 134/// @name Auto-generated Match Functions 135/// { 136 137static unsigned MatchRegisterName(StringRef Name); 138 139/// } 140 141static bool isImmSExti16i8Value(uint64_t Value) { 142 return (( Value <= 0x000000000000007FULL)|| 143 (0x000000000000FF80ULL <= Value && Value <= 0x000000000000FFFFULL)|| 144 (0xFFFFFFFFFFFFFF80ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL)); 145} 146 147static bool isImmSExti32i8Value(uint64_t Value) { 148 return (( Value <= 0x000000000000007FULL)|| 149 (0x00000000FFFFFF80ULL <= Value && Value <= 0x00000000FFFFFFFFULL)|| 150 (0xFFFFFFFFFFFFFF80ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL)); 151} 152 153static bool isImmZExtu32u8Value(uint64_t Value) { 154 return (Value <= 0x00000000000000FFULL); 155} 156 157static bool isImmSExti64i8Value(uint64_t Value) { 158 return (( Value <= 0x000000000000007FULL)|| 159 (0xFFFFFFFFFFFFFF80ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL)); 160} 161 162static bool isImmSExti64i32Value(uint64_t Value) { 163 return (( Value <= 0x000000007FFFFFFFULL)|| 164 (0xFFFFFFFF80000000ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL)); 165} 166namespace { 167 168/// X86Operand - Instances of this class represent a parsed X86 machine 169/// instruction. 170struct X86Operand : public MCParsedAsmOperand { 171 enum KindTy { 172 Token, 173 Register, 174 Immediate, 175 Memory 176 } Kind; 177 178 SMLoc StartLoc, EndLoc; 179 SMLoc OffsetOfLoc; 180 StringRef SymName; 181 bool AddressOf; 182 183 struct TokOp { 184 const char *Data; 185 unsigned Length; 186 }; 187 188 struct RegOp { 189 unsigned RegNo; 190 }; 191 192 struct ImmOp { 193 const MCExpr *Val; 194 }; 195 196 struct MemOp { 197 unsigned SegReg; 198 const MCExpr *Disp; 199 unsigned BaseReg; 200 unsigned IndexReg; 201 unsigned Scale; 202 unsigned Size; 203 }; 204 205 union { 206 struct TokOp Tok; 207 struct RegOp Reg; 208 struct ImmOp Imm; 209 struct MemOp Mem; 210 }; 211 212 X86Operand(KindTy K, SMLoc Start, SMLoc End) 213 : Kind(K), StartLoc(Start), EndLoc(End) {} 214 215 StringRef getSymName() { return SymName; } 216 217 /// getStartLoc - Get the location of the first token of this operand. 218 SMLoc getStartLoc() const { return StartLoc; } 219 /// getEndLoc - Get the location of the last token of this operand. 220 SMLoc getEndLoc() const { return EndLoc; } 221 /// getLocRange - Get the range between the first and last token of this 222 /// operand. 223 SMRange getLocRange() const { return SMRange(StartLoc, EndLoc); } 224 /// getOffsetOfLoc - Get the location of the offset operator. 225 SMLoc getOffsetOfLoc() const { return OffsetOfLoc; } 226 227 virtual void print(raw_ostream &OS) const {} 228 229 StringRef getToken() const { 230 assert(Kind == Token && "Invalid access!"); 231 return StringRef(Tok.Data, Tok.Length); 232 } 233 void setTokenValue(StringRef Value) { 234 assert(Kind == Token && "Invalid access!"); 235 Tok.Data = Value.data(); 236 Tok.Length = Value.size(); 237 } 238 239 unsigned getReg() const { 240 assert(Kind == Register && "Invalid access!"); 241 return Reg.RegNo; 242 } 243 244 const MCExpr *getImm() const { 245 assert(Kind == Immediate && "Invalid access!"); 246 return Imm.Val; 247 } 248 249 const MCExpr *getMemDisp() const { 250 assert(Kind == Memory && "Invalid access!"); 251 return Mem.Disp; 252 } 253 unsigned getMemSegReg() const { 254 assert(Kind == Memory && "Invalid access!"); 255 return Mem.SegReg; 256 } 257 unsigned getMemBaseReg() const { 258 assert(Kind == Memory && "Invalid access!"); 259 return Mem.BaseReg; 260 } 261 unsigned getMemIndexReg() const { 262 assert(Kind == Memory && "Invalid access!"); 263 return Mem.IndexReg; 264 } 265 unsigned getMemScale() const { 266 assert(Kind == Memory && "Invalid access!"); 267 return Mem.Scale; 268 } 269 270 bool isToken() const {return Kind == Token; } 271 272 bool isImm() const { return Kind == Immediate; } 273 274 bool isImmSExti16i8() const { 275 if (!isImm()) 276 return false; 277 278 // If this isn't a constant expr, just assume it fits and let relaxation 279 // handle it. 280 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm()); 281 if (!CE) 282 return true; 283 284 // Otherwise, check the value is in a range that makes sense for this 285 // extension. 286 return isImmSExti16i8Value(CE->getValue()); 287 } 288 bool isImmSExti32i8() const { 289 if (!isImm()) 290 return false; 291 292 // If this isn't a constant expr, just assume it fits and let relaxation 293 // handle it. 294 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm()); 295 if (!CE) 296 return true; 297 298 // Otherwise, check the value is in a range that makes sense for this 299 // extension. 300 return isImmSExti32i8Value(CE->getValue()); 301 } 302 bool isImmZExtu32u8() const { 303 if (!isImm()) 304 return false; 305 306 // If this isn't a constant expr, just assume it fits and let relaxation 307 // handle it. 308 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm()); 309 if (!CE) 310 return true; 311 312 // Otherwise, check the value is in a range that makes sense for this 313 // extension. 314 return isImmZExtu32u8Value(CE->getValue()); 315 } 316 bool isImmSExti64i8() const { 317 if (!isImm()) 318 return false; 319 320 // If this isn't a constant expr, just assume it fits and let relaxation 321 // handle it. 322 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm()); 323 if (!CE) 324 return true; 325 326 // Otherwise, check the value is in a range that makes sense for this 327 // extension. 328 return isImmSExti64i8Value(CE->getValue()); 329 } 330 bool isImmSExti64i32() const { 331 if (!isImm()) 332 return false; 333 334 // If this isn't a constant expr, just assume it fits and let relaxation 335 // handle it. 336 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm()); 337 if (!CE) 338 return true; 339 340 // Otherwise, check the value is in a range that makes sense for this 341 // extension. 342 return isImmSExti64i32Value(CE->getValue()); 343 } 344 345 bool isOffsetOf() const { 346 return OffsetOfLoc.getPointer(); 347 } 348 349 bool needAddressOf() const { 350 return AddressOf; 351 } 352 353 bool isMem() const { return Kind == Memory; } 354 bool isMem8() const { 355 return Kind == Memory && (!Mem.Size || Mem.Size == 8); 356 } 357 bool isMem16() const { 358 return Kind == Memory && (!Mem.Size || Mem.Size == 16); 359 } 360 bool isMem32() const { 361 return Kind == Memory && (!Mem.Size || Mem.Size == 32); 362 } 363 bool isMem64() const { 364 return Kind == Memory && (!Mem.Size || Mem.Size == 64); 365 } 366 bool isMem80() const { 367 return Kind == Memory && (!Mem.Size || Mem.Size == 80); 368 } 369 bool isMem128() const { 370 return Kind == Memory && (!Mem.Size || Mem.Size == 128); 371 } 372 bool isMem256() const { 373 return Kind == Memory && (!Mem.Size || Mem.Size == 256); 374 } 375 376 bool isMemVX32() const { 377 return Kind == Memory && (!Mem.Size || Mem.Size == 32) && 378 getMemIndexReg() >= X86::XMM0 && getMemIndexReg() <= X86::XMM15; 379 } 380 bool isMemVY32() const { 381 return Kind == Memory && (!Mem.Size || Mem.Size == 32) && 382 getMemIndexReg() >= X86::YMM0 && getMemIndexReg() <= X86::YMM15; 383 } 384 bool isMemVX64() const { 385 return Kind == Memory && (!Mem.Size || Mem.Size == 64) && 386 getMemIndexReg() >= X86::XMM0 && getMemIndexReg() <= X86::XMM15; 387 } 388 bool isMemVY64() const { 389 return Kind == Memory && (!Mem.Size || Mem.Size == 64) && 390 getMemIndexReg() >= X86::YMM0 && getMemIndexReg() <= X86::YMM15; 391 } 392 393 bool isAbsMem() const { 394 return Kind == Memory && !getMemSegReg() && !getMemBaseReg() && 395 !getMemIndexReg() && getMemScale() == 1; 396 } 397 398 bool isReg() const { return Kind == Register; } 399 400 void addExpr(MCInst &Inst, const MCExpr *Expr) const { 401 // Add as immediates when possible. 402 if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Expr)) 403 Inst.addOperand(MCOperand::CreateImm(CE->getValue())); 404 else 405 Inst.addOperand(MCOperand::CreateExpr(Expr)); 406 } 407 408 void addRegOperands(MCInst &Inst, unsigned N) const { 409 assert(N == 1 && "Invalid number of operands!"); 410 Inst.addOperand(MCOperand::CreateReg(getReg())); 411 } 412 413 void addImmOperands(MCInst &Inst, unsigned N) const { 414 assert(N == 1 && "Invalid number of operands!"); 415 addExpr(Inst, getImm()); 416 } 417 418 void addMem8Operands(MCInst &Inst, unsigned N) const { 419 addMemOperands(Inst, N); 420 } 421 void addMem16Operands(MCInst &Inst, unsigned N) const { 422 addMemOperands(Inst, N); 423 } 424 void addMem32Operands(MCInst &Inst, unsigned N) const { 425 addMemOperands(Inst, N); 426 } 427 void addMem64Operands(MCInst &Inst, unsigned N) const { 428 addMemOperands(Inst, N); 429 } 430 void addMem80Operands(MCInst &Inst, unsigned N) const { 431 addMemOperands(Inst, N); 432 } 433 void addMem128Operands(MCInst &Inst, unsigned N) const { 434 addMemOperands(Inst, N); 435 } 436 void addMem256Operands(MCInst &Inst, unsigned N) const { 437 addMemOperands(Inst, N); 438 } 439 void addMemVX32Operands(MCInst &Inst, unsigned N) const { 440 addMemOperands(Inst, N); 441 } 442 void addMemVY32Operands(MCInst &Inst, unsigned N) const { 443 addMemOperands(Inst, N); 444 } 445 void addMemVX64Operands(MCInst &Inst, unsigned N) const { 446 addMemOperands(Inst, N); 447 } 448 void addMemVY64Operands(MCInst &Inst, unsigned N) const { 449 addMemOperands(Inst, N); 450 } 451 452 void addMemOperands(MCInst &Inst, unsigned N) const { 453 assert((N == 5) && "Invalid number of operands!"); 454 Inst.addOperand(MCOperand::CreateReg(getMemBaseReg())); 455 Inst.addOperand(MCOperand::CreateImm(getMemScale())); 456 Inst.addOperand(MCOperand::CreateReg(getMemIndexReg())); 457 addExpr(Inst, getMemDisp()); 458 Inst.addOperand(MCOperand::CreateReg(getMemSegReg())); 459 } 460 461 void addAbsMemOperands(MCInst &Inst, unsigned N) const { 462 assert((N == 1) && "Invalid number of operands!"); 463 // Add as immediates when possible. 464 if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getMemDisp())) 465 Inst.addOperand(MCOperand::CreateImm(CE->getValue())); 466 else 467 Inst.addOperand(MCOperand::CreateExpr(getMemDisp())); 468 } 469 470 static X86Operand *CreateToken(StringRef Str, SMLoc Loc) { 471 SMLoc EndLoc = SMLoc::getFromPointer(Loc.getPointer() + Str.size()); 472 X86Operand *Res = new X86Operand(Token, Loc, EndLoc); 473 Res->Tok.Data = Str.data(); 474 Res->Tok.Length = Str.size(); 475 return Res; 476 } 477 478 static X86Operand *CreateReg(unsigned RegNo, SMLoc StartLoc, SMLoc EndLoc, 479 bool AddressOf = false, 480 SMLoc OffsetOfLoc = SMLoc(), 481 StringRef SymName = StringRef()) { 482 X86Operand *Res = new X86Operand(Register, StartLoc, EndLoc); 483 Res->Reg.RegNo = RegNo; 484 Res->AddressOf = AddressOf; 485 Res->OffsetOfLoc = OffsetOfLoc; 486 Res->SymName = SymName; 487 return Res; 488 } 489 490 static X86Operand *CreateImm(const MCExpr *Val, SMLoc StartLoc, SMLoc EndLoc){ 491 X86Operand *Res = new X86Operand(Immediate, StartLoc, EndLoc); 492 Res->Imm.Val = Val; 493 return Res; 494 } 495 496 /// Create an absolute memory operand. 497 static X86Operand *CreateMem(const MCExpr *Disp, SMLoc StartLoc, SMLoc EndLoc, 498 unsigned Size = 0, 499 StringRef SymName = StringRef()) { 500 X86Operand *Res = new X86Operand(Memory, StartLoc, EndLoc); 501 Res->Mem.SegReg = 0; 502 Res->Mem.Disp = Disp; 503 Res->Mem.BaseReg = 0; 504 Res->Mem.IndexReg = 0; 505 Res->Mem.Scale = 1; 506 Res->Mem.Size = Size; 507 Res->SymName = SymName; 508 Res->AddressOf = false; 509 return Res; 510 } 511 512 /// Create a generalized memory operand. 513 static X86Operand *CreateMem(unsigned SegReg, const MCExpr *Disp, 514 unsigned BaseReg, unsigned IndexReg, 515 unsigned Scale, SMLoc StartLoc, SMLoc EndLoc, 516 unsigned Size = 0, 517 StringRef SymName = StringRef()) { 518 // We should never just have a displacement, that should be parsed as an 519 // absolute memory operand. 520 assert((SegReg || BaseReg || IndexReg) && "Invalid memory operand!"); 521 522 // The scale should always be one of {1,2,4,8}. 523 assert(((Scale == 1 || Scale == 2 || Scale == 4 || Scale == 8)) && 524 "Invalid scale!"); 525 X86Operand *Res = new X86Operand(Memory, StartLoc, EndLoc); 526 Res->Mem.SegReg = SegReg; 527 Res->Mem.Disp = Disp; 528 Res->Mem.BaseReg = BaseReg; 529 Res->Mem.IndexReg = IndexReg; 530 Res->Mem.Scale = Scale; 531 Res->Mem.Size = Size; 532 Res->SymName = SymName; 533 Res->AddressOf = false; 534 return Res; 535 } 536}; 537 538} // end anonymous namespace. 539 540bool X86AsmParser::isSrcOp(X86Operand &Op) { 541 unsigned basereg = is64BitMode() ? X86::RSI : X86::ESI; 542 543 return (Op.isMem() && 544 (Op.Mem.SegReg == 0 || Op.Mem.SegReg == X86::DS) && 545 isa<MCConstantExpr>(Op.Mem.Disp) && 546 cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 && 547 Op.Mem.BaseReg == basereg && Op.Mem.IndexReg == 0); 548} 549 550bool X86AsmParser::isDstOp(X86Operand &Op) { 551 unsigned basereg = is64BitMode() ? X86::RDI : X86::EDI; 552 553 return Op.isMem() && 554 (Op.Mem.SegReg == 0 || Op.Mem.SegReg == X86::ES) && 555 isa<MCConstantExpr>(Op.Mem.Disp) && 556 cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 && 557 Op.Mem.BaseReg == basereg && Op.Mem.IndexReg == 0; 558} 559 560bool X86AsmParser::ParseRegister(unsigned &RegNo, 561 SMLoc &StartLoc, SMLoc &EndLoc) { 562 RegNo = 0; 563 const AsmToken &PercentTok = Parser.getTok(); 564 StartLoc = PercentTok.getLoc(); 565 566 // If we encounter a %, ignore it. This code handles registers with and 567 // without the prefix, unprefixed registers can occur in cfi directives. 568 if (!isParsingIntelSyntax() && PercentTok.is(AsmToken::Percent)) 569 Parser.Lex(); // Eat percent token. 570 571 const AsmToken &Tok = Parser.getTok(); 572 EndLoc = Tok.getEndLoc(); 573 574 if (Tok.isNot(AsmToken::Identifier)) { 575 if (isParsingIntelSyntax()) return true; 576 return Error(StartLoc, "invalid register name", 577 SMRange(StartLoc, EndLoc)); 578 } 579 580 RegNo = MatchRegisterName(Tok.getString()); 581 582 // If the match failed, try the register name as lowercase. 583 if (RegNo == 0) 584 RegNo = MatchRegisterName(Tok.getString().lower()); 585 586 if (!is64BitMode()) { 587 // FIXME: This should be done using Requires<In32BitMode> and 588 // Requires<In64BitMode> so "eiz" usage in 64-bit instructions can be also 589 // checked. 590 // FIXME: Check AH, CH, DH, BH cannot be used in an instruction requiring a 591 // REX prefix. 592 if (RegNo == X86::RIZ || 593 X86MCRegisterClasses[X86::GR64RegClassID].contains(RegNo) || 594 X86II::isX86_64NonExtLowByteReg(RegNo) || 595 X86II::isX86_64ExtendedReg(RegNo)) 596 return Error(StartLoc, "register %" 597 + Tok.getString() + " is only available in 64-bit mode", 598 SMRange(StartLoc, EndLoc)); 599 } 600 601 // Parse "%st" as "%st(0)" and "%st(1)", which is multiple tokens. 602 if (RegNo == 0 && (Tok.getString() == "st" || Tok.getString() == "ST")) { 603 RegNo = X86::ST0; 604 Parser.Lex(); // Eat 'st' 605 606 // Check to see if we have '(4)' after %st. 607 if (getLexer().isNot(AsmToken::LParen)) 608 return false; 609 // Lex the paren. 610 getParser().Lex(); 611 612 const AsmToken &IntTok = Parser.getTok(); 613 if (IntTok.isNot(AsmToken::Integer)) 614 return Error(IntTok.getLoc(), "expected stack index"); 615 switch (IntTok.getIntVal()) { 616 case 0: RegNo = X86::ST0; break; 617 case 1: RegNo = X86::ST1; break; 618 case 2: RegNo = X86::ST2; break; 619 case 3: RegNo = X86::ST3; break; 620 case 4: RegNo = X86::ST4; break; 621 case 5: RegNo = X86::ST5; break; 622 case 6: RegNo = X86::ST6; break; 623 case 7: RegNo = X86::ST7; break; 624 default: return Error(IntTok.getLoc(), "invalid stack index"); 625 } 626 627 if (getParser().Lex().isNot(AsmToken::RParen)) 628 return Error(Parser.getTok().getLoc(), "expected ')'"); 629 630 EndLoc = Parser.getTok().getEndLoc(); 631 Parser.Lex(); // Eat ')' 632 return false; 633 } 634 635 EndLoc = Parser.getTok().getEndLoc(); 636 637 // If this is "db[0-7]", match it as an alias 638 // for dr[0-7]. 639 if (RegNo == 0 && Tok.getString().size() == 3 && 640 Tok.getString().startswith("db")) { 641 switch (Tok.getString()[2]) { 642 case '0': RegNo = X86::DR0; break; 643 case '1': RegNo = X86::DR1; break; 644 case '2': RegNo = X86::DR2; break; 645 case '3': RegNo = X86::DR3; break; 646 case '4': RegNo = X86::DR4; break; 647 case '5': RegNo = X86::DR5; break; 648 case '6': RegNo = X86::DR6; break; 649 case '7': RegNo = X86::DR7; break; 650 } 651 652 if (RegNo != 0) { 653 EndLoc = Parser.getTok().getEndLoc(); 654 Parser.Lex(); // Eat it. 655 return false; 656 } 657 } 658 659 if (RegNo == 0) { 660 if (isParsingIntelSyntax()) return true; 661 return Error(StartLoc, "invalid register name", 662 SMRange(StartLoc, EndLoc)); 663 } 664 665 Parser.Lex(); // Eat identifier token. 666 return false; 667} 668 669X86Operand *X86AsmParser::ParseOperand() { 670 if (isParsingIntelSyntax()) 671 return ParseIntelOperand(); 672 return ParseATTOperand(); 673} 674 675/// getIntelMemOperandSize - Return intel memory operand size. 676static unsigned getIntelMemOperandSize(StringRef OpStr) { 677 unsigned Size = StringSwitch<unsigned>(OpStr) 678 .Cases("BYTE", "byte", 8) 679 .Cases("WORD", "word", 16) 680 .Cases("DWORD", "dword", 32) 681 .Cases("QWORD", "qword", 64) 682 .Cases("XWORD", "xword", 80) 683 .Cases("XMMWORD", "xmmword", 128) 684 .Cases("YMMWORD", "ymmword", 256) 685 .Default(0); 686 return Size; 687} 688 689enum InfixCalculatorTok { 690 IC_PLUS = 0, 691 IC_MINUS, 692 IC_MULTIPLY, 693 IC_DIVIDE, 694 IC_RPAREN, 695 IC_LPAREN, 696 IC_IMM, 697 IC_REGISTER 698}; 699static const char OpPrecedence[] = { 700 0, // IC_PLUS 701 0, // IC_MINUS 702 1, // IC_MULTIPLY 703 1, // IC_DIVIDE 704 2, // IC_RPAREN 705 3, // IC_LPAREN 706 0, // IC_IMM 707 0 // IC_REGISTER 708}; 709 710class InfixCalculator { 711 typedef std::pair< InfixCalculatorTok, int64_t > ICToken; 712 SmallVector<InfixCalculatorTok, 4> InfixOperatorStack; 713 SmallVector<ICToken, 4> PostfixStack; 714 715public: 716 int64_t popOperand() { 717 assert (!PostfixStack.empty() && "Poped an empty stack!"); 718 ICToken Op = PostfixStack.pop_back_val(); 719 assert ((Op.first == IC_IMM || Op.first == IC_REGISTER) 720 && "Expected and immediate or register!"); 721 return Op.second; 722 } 723 void pushOperand(InfixCalculatorTok Op, int64_t Val = 0) { 724 assert ((Op == IC_IMM || Op == IC_REGISTER) && 725 "Unexpected operand!"); 726 PostfixStack.push_back(std::make_pair(Op, Val)); 727 } 728 729 void popOperator() { InfixOperatorStack.pop_back_val(); } 730 void pushOperator(InfixCalculatorTok Op) { 731 // Push the new operator if the stack is empty. 732 if (InfixOperatorStack.empty()) { 733 InfixOperatorStack.push_back(Op); 734 return; 735 } 736 737 // Push the new operator if it has a higher precedence than the operator on 738 // the top of the stack or the operator on the top of the stack is a left 739 // parentheses. 740 unsigned Idx = InfixOperatorStack.size() - 1; 741 InfixCalculatorTok StackOp = InfixOperatorStack[Idx]; 742 if (OpPrecedence[Op] > OpPrecedence[StackOp] || StackOp == IC_LPAREN) { 743 InfixOperatorStack.push_back(Op); 744 return; 745 } 746 747 // The operator on the top of the stack has higher precedence than the 748 // new operator. 749 unsigned ParenCount = 0; 750 while (1) { 751 // Nothing to process. 752 if (InfixOperatorStack.empty()) 753 break; 754 755 Idx = InfixOperatorStack.size() - 1; 756 StackOp = InfixOperatorStack[Idx]; 757 if (!(OpPrecedence[StackOp] >= OpPrecedence[Op] || ParenCount)) 758 break; 759 760 // If we have an even parentheses count and we see a left parentheses, 761 // then stop processing. 762 if (!ParenCount && StackOp == IC_LPAREN) 763 break; 764 765 if (StackOp == IC_RPAREN) { 766 ++ParenCount; 767 InfixOperatorStack.pop_back_val(); 768 } else if (StackOp == IC_LPAREN) { 769 --ParenCount; 770 InfixOperatorStack.pop_back_val(); 771 } else { 772 InfixOperatorStack.pop_back_val(); 773 PostfixStack.push_back(std::make_pair(StackOp, 0)); 774 } 775 } 776 // Push the new operator. 777 InfixOperatorStack.push_back(Op); 778 } 779 int64_t execute() { 780 // Push any remaining operators onto the postfix stack. 781 while (!InfixOperatorStack.empty()) { 782 InfixCalculatorTok StackOp = InfixOperatorStack.pop_back_val(); 783 if (StackOp != IC_LPAREN && StackOp != IC_RPAREN) 784 PostfixStack.push_back(std::make_pair(StackOp, 0)); 785 } 786 787 if (PostfixStack.empty()) 788 return 0; 789 790 SmallVector<ICToken, 16> OperandStack; 791 for (unsigned i = 0, e = PostfixStack.size(); i != e; ++i) { 792 ICToken Op = PostfixStack[i]; 793 if (Op.first == IC_IMM || Op.first == IC_REGISTER) { 794 OperandStack.push_back(Op); 795 } else { 796 assert (OperandStack.size() > 1 && "Too few operands."); 797 int64_t Val; 798 ICToken Op2 = OperandStack.pop_back_val(); 799 ICToken Op1 = OperandStack.pop_back_val(); 800 switch (Op.first) { 801 default: 802 report_fatal_error("Unexpected operator!"); 803 break; 804 case IC_PLUS: 805 Val = Op1.second + Op2.second; 806 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 807 break; 808 case IC_MINUS: 809 Val = Op1.second - Op2.second; 810 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 811 break; 812 case IC_MULTIPLY: 813 assert (Op1.first == IC_IMM && Op2.first == IC_IMM && 814 "Multiply operation with an immediate and a register!"); 815 Val = Op1.second * Op2.second; 816 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 817 break; 818 case IC_DIVIDE: 819 assert (Op1.first == IC_IMM && Op2.first == IC_IMM && 820 "Divide operation with an immediate and a register!"); 821 assert (Op2.second != 0 && "Division by zero!"); 822 Val = Op1.second / Op2.second; 823 OperandStack.push_back(std::make_pair(IC_IMM, Val)); 824 break; 825 } 826 } 827 } 828 assert (OperandStack.size() == 1 && "Expected a single result."); 829 return OperandStack.pop_back_val().second; 830 } 831}; 832 833enum IntelBracExprState { 834 IBES_PLUS, 835 IBES_MINUS, 836 IBES_MULTIPLY, 837 IBES_DIVIDE, 838 IBES_LBRAC, 839 IBES_RBRAC, 840 IBES_LPAREN, 841 IBES_RPAREN, 842 IBES_REGISTER, 843 IBES_REGISTER_STAR, 844 IBES_INTEGER, 845 IBES_INTEGER_STAR, 846 IBES_IDENTIFIER, 847 IBES_ERROR 848}; 849 850class IntelBracExprStateMachine { 851 IntelBracExprState State; 852 unsigned BaseReg, IndexReg, TmpReg, Scale; 853 int64_t Disp; 854 InfixCalculator IC; 855public: 856 IntelBracExprStateMachine(MCAsmParser &parser, int64_t disp) : 857 State(IBES_PLUS), BaseReg(0), IndexReg(0), TmpReg(0), Scale(1), Disp(disp){} 858 859 unsigned getBaseReg() { return BaseReg; } 860 unsigned getIndexReg() { return IndexReg; } 861 unsigned getScale() { return Scale; } 862 int64_t getDisp() { return Disp + IC.execute(); } 863 bool isValidEndState() { return State == IBES_RBRAC; } 864 865 void onPlus() { 866 switch (State) { 867 default: 868 State = IBES_ERROR; 869 break; 870 case IBES_INTEGER: 871 case IBES_RPAREN: 872 State = IBES_PLUS; 873 IC.pushOperator(IC_PLUS); 874 break; 875 case IBES_REGISTER: 876 State = IBES_PLUS; 877 // If we already have a BaseReg, then assume this is the IndexReg with a 878 // scale of 1. 879 if (!BaseReg) { 880 BaseReg = TmpReg; 881 } else { 882 assert (!IndexReg && "BaseReg/IndexReg already set!"); 883 IndexReg = TmpReg; 884 Scale = 1; 885 } 886 IC.pushOperator(IC_PLUS); 887 break; 888 } 889 } 890 void onMinus() { 891 switch (State) { 892 default: 893 State = IBES_ERROR; 894 break; 895 case IBES_PLUS: 896 case IBES_LPAREN: 897 IC.pushOperand(IC_IMM); 898 case IBES_INTEGER: 899 case IBES_RPAREN: 900 State = IBES_MINUS; 901 IC.pushOperator(IC_MINUS); 902 break; 903 case IBES_REGISTER: 904 State = IBES_MINUS; 905 // If we already have a BaseReg, then assume this is the IndexReg with a 906 // scale of 1. 907 if (!BaseReg) { 908 BaseReg = TmpReg; 909 } else { 910 assert (!IndexReg && "BaseReg/IndexReg already set!"); 911 IndexReg = TmpReg; 912 Scale = 1; 913 } 914 IC.pushOperator(IC_MINUS); 915 break; 916 } 917 } 918 void onRegister(unsigned Reg) { 919 switch (State) { 920 default: 921 State = IBES_ERROR; 922 break; 923 case IBES_PLUS: 924 case IBES_LPAREN: 925 State = IBES_REGISTER; 926 TmpReg = Reg; 927 IC.pushOperand(IC_REGISTER); 928 break; 929 case IBES_INTEGER_STAR: 930 assert (!IndexReg && "IndexReg already set!"); 931 State = IBES_INTEGER; 932 IndexReg = Reg; 933 Scale = IC.popOperand(); 934 IC.pushOperand(IC_IMM); 935 IC.popOperator(); 936 break; 937 } 938 } 939 void onDispExpr() { 940 switch (State) { 941 default: 942 State = IBES_ERROR; 943 break; 944 case IBES_PLUS: 945 case IBES_MINUS: 946 State = IBES_INTEGER; 947 IC.pushOperand(IC_IMM); 948 break; 949 } 950 } 951 void onInteger(int64_t TmpInt) { 952 switch (State) { 953 default: 954 State = IBES_ERROR; 955 break; 956 case IBES_PLUS: 957 case IBES_MINUS: 958 case IBES_MULTIPLY: 959 case IBES_DIVIDE: 960 case IBES_LPAREN: 961 case IBES_INTEGER_STAR: 962 State = IBES_INTEGER; 963 IC.pushOperand(IC_IMM, TmpInt); 964 break; 965 case IBES_REGISTER_STAR: 966 assert (!IndexReg && "IndexReg already set!"); 967 State = IBES_INTEGER; 968 IndexReg = TmpReg; 969 Scale = TmpInt; 970 IC.popOperator(); 971 break; 972 } 973 } 974 void onStar() { 975 switch (State) { 976 default: 977 State = IBES_ERROR; 978 break; 979 case IBES_INTEGER: 980 State = IBES_INTEGER_STAR; 981 IC.pushOperator(IC_MULTIPLY); 982 break; 983 case IBES_REGISTER: 984 State = IBES_REGISTER_STAR; 985 IC.pushOperator(IC_MULTIPLY); 986 break; 987 case IBES_RPAREN: 988 State = IBES_MULTIPLY; 989 IC.pushOperator(IC_MULTIPLY); 990 break; 991 } 992 } 993 void onDivide() { 994 switch (State) { 995 default: 996 State = IBES_ERROR; 997 break; 998 case IBES_INTEGER: 999 State = IBES_DIVIDE; 1000 IC.pushOperator(IC_DIVIDE); 1001 break; 1002 } 1003 } 1004 void onLBrac() { 1005 switch (State) { 1006 default: 1007 State = IBES_ERROR; 1008 break; 1009 case IBES_RBRAC: 1010 State = IBES_PLUS; 1011 IC.pushOperator(IC_PLUS); 1012 break; 1013 } 1014 } 1015 void onRBrac() { 1016 switch (State) { 1017 default: 1018 State = IBES_ERROR; 1019 break; 1020 case IBES_RPAREN: 1021 case IBES_INTEGER: 1022 State = IBES_RBRAC; 1023 break; 1024 case IBES_REGISTER: 1025 State = IBES_RBRAC; 1026 // If we already have a BaseReg, then assume this is the IndexReg with a 1027 // scale of 1. 1028 if (!BaseReg) { 1029 BaseReg = TmpReg; 1030 } else { 1031 assert (!IndexReg && "BaseReg/IndexReg already set!"); 1032 IndexReg = TmpReg; 1033 Scale = 1; 1034 } 1035 break; 1036 } 1037 } 1038 void onLParen() { 1039 switch (State) { 1040 default: 1041 State = IBES_ERROR; 1042 break; 1043 case IBES_PLUS: 1044 case IBES_MINUS: 1045 case IBES_MULTIPLY: 1046 case IBES_DIVIDE: 1047 case IBES_INTEGER_STAR: 1048 case IBES_LPAREN: 1049 State = IBES_LPAREN; 1050 IC.pushOperator(IC_LPAREN); 1051 break; 1052 } 1053 } 1054 void onRParen() { 1055 switch (State) { 1056 default: 1057 State = IBES_ERROR; 1058 break; 1059 case IBES_REGISTER: 1060 case IBES_INTEGER: 1061 case IBES_PLUS: 1062 case IBES_MINUS: 1063 case IBES_MULTIPLY: 1064 case IBES_DIVIDE: 1065 case IBES_RPAREN: 1066 State = IBES_RPAREN; 1067 IC.pushOperator(IC_RPAREN); 1068 break; 1069 } 1070 } 1071}; 1072 1073X86Operand *X86AsmParser::CreateMemForInlineAsm(const MCExpr *Disp, SMLoc Start, 1074 SMLoc End, SMLoc SizeDirLoc, 1075 unsigned Size, StringRef SymName) { 1076 bool NeedSizeDir = false; 1077 bool IsVarDecl = false; 1078 1079 if (const MCSymbolRefExpr *SymRef = dyn_cast<MCSymbolRefExpr>(Disp)) { 1080 const MCSymbol &Sym = SymRef->getSymbol(); 1081 // FIXME: The SemaLookup will fail if the name is anything other then an 1082 // identifier. 1083 // FIXME: Pass a valid SMLoc. 1084 unsigned tLength, tSize, tType; 1085 SemaCallback->LookupInlineAsmIdentifier(Sym.getName(), NULL, tLength, tSize, 1086 tType, IsVarDecl); 1087 if (!Size) { 1088 Size = tType * 8; // Size is in terms of bits in this context. 1089 NeedSizeDir = Size > 0; 1090 } 1091 } 1092 1093 // If this is not a VarDecl then assume it is a FuncDecl or some other label 1094 // reference. We need an 'r' constraint here, so we need to create register 1095 // operand to ensure proper matching. Just pick a GPR based on the size of 1096 // a pointer. 1097 if (!IsVarDecl) { 1098 unsigned RegNo = is64BitMode() ? X86::RBX : X86::EBX; 1099 return X86Operand::CreateReg(RegNo, Start, End, /*AddressOf=*/true, SMLoc(), 1100 SymName); 1101 } 1102 1103 if (NeedSizeDir) 1104 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_SizeDirective, SizeDirLoc, 1105 /*Len*/0, Size)); 1106 1107 // When parsing inline assembly we set the base register to a non-zero value 1108 // as we don't know the actual value at this time. This is necessary to 1109 // get the matching correct in some cases. 1110 return X86Operand::CreateMem(/*SegReg*/0, Disp, /*BaseReg*/1, /*IndexReg*/0, 1111 /*Scale*/1, Start, End, Size, SymName); 1112} 1113 1114X86Operand *X86AsmParser::ParseIntelBracExpression(unsigned SegReg, 1115 SMLoc SizeDirLoc, 1116 uint64_t ImmDisp, 1117 unsigned Size) { 1118 const AsmToken &Tok = Parser.getTok(); 1119 SMLoc Start = Tok.getLoc(), End = Tok.getEndLoc(); 1120 1121 // Eat '[' 1122 if (getLexer().isNot(AsmToken::LBrac)) 1123 return ErrorOperand(Start, "Expected '[' token!"); 1124 Parser.Lex(); 1125 1126 unsigned TmpReg = 0; 1127 1128 // Try to handle '[' 'Symbol' ']' 1129 if (getLexer().is(AsmToken::Identifier)) { 1130 SMLoc Loc = Tok.getLoc(); 1131 if (ParseRegister(TmpReg, Loc, End)) { 1132 const MCExpr *Disp; 1133 StringRef Identifier = Tok.getString(); 1134 if (getParser().parseExpression(Disp, End)) 1135 return 0; 1136 1137 if (X86Operand *Err = ParseIntelVarWithQualifier(Disp, Identifier)) 1138 return Err; 1139 1140 if (getLexer().isNot(AsmToken::RBrac)) 1141 return ErrorOperand(Tok.getLoc(), "Expected ']' token!"); 1142 1143 if (isParsingInlineAsm()) { 1144 // Remove the '[' and ']' from the IR string. 1145 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Skip, Start, 1)); 1146 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Skip, Tok.getLoc(), 1)); 1147 } 1148 Parser.Lex(); // Eat ']' 1149 if (!isParsingInlineAsm()) 1150 return X86Operand::CreateMem(Disp, Start, End, Size); 1151 return CreateMemForInlineAsm(Disp, Start, End, SizeDirLoc, Size, 1152 Identifier); 1153 } 1154 } 1155 1156 // Parse [ BaseReg + Scale*IndexReg + Disp ]. We may have already parsed an 1157 // immediate displacement before the bracketed expression. 1158 bool Done = false; 1159 IntelBracExprStateMachine SM(Parser, ImmDisp); 1160 1161 // If we parsed a register, then the end loc has already been set and 1162 // the identifier has already been lexed. We also need to update the 1163 // state. 1164 if (TmpReg) 1165 SM.onRegister(TmpReg); 1166 1167 const MCExpr *Disp = 0; 1168 while (!Done) { 1169 bool UpdateLocLex = true; 1170 1171 // The period in the dot operator (e.g., [ebx].foo.bar) is parsed as an 1172 // identifier. Don't try an parse it as a register. 1173 if (Tok.getString().startswith(".")) 1174 break; 1175 1176 switch (getLexer().getKind()) { 1177 default: { 1178 if (SM.isValidEndState()) { 1179 Done = true; 1180 break; 1181 } 1182 return ErrorOperand(Tok.getLoc(), "Unexpected token!"); 1183 } 1184 case AsmToken::Identifier: { 1185 // This could be a register or a displacement expression. 1186 SMLoc Loc = Tok.getLoc(); 1187 if(!ParseRegister(TmpReg, Loc, End)) { 1188 SM.onRegister(TmpReg); 1189 UpdateLocLex = false; 1190 break; 1191 } else if (!getParser().parsePrimaryExpr(Disp, End)) { 1192 SM.onDispExpr(); 1193 UpdateLocLex = false; 1194 break; 1195 } 1196 return ErrorOperand(Tok.getLoc(), "Unexpected identifier!"); 1197 } 1198 case AsmToken::Integer: 1199 if (isParsingInlineAsm()) 1200 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_ImmPrefix, 1201 Tok.getLoc())); 1202 SM.onInteger(Tok.getIntVal()); 1203 break; 1204 case AsmToken::Plus: SM.onPlus(); break; 1205 case AsmToken::Minus: SM.onMinus(); break; 1206 case AsmToken::Star: SM.onStar(); break; 1207 case AsmToken::Slash: SM.onDivide(); break; 1208 case AsmToken::LBrac: SM.onLBrac(); break; 1209 case AsmToken::RBrac: SM.onRBrac(); break; 1210 case AsmToken::LParen: SM.onLParen(); break; 1211 case AsmToken::RParen: SM.onRParen(); break; 1212 } 1213 if (!Done && UpdateLocLex) { 1214 End = Tok.getLoc(); 1215 Parser.Lex(); // Consume the token. 1216 } 1217 } 1218 if (isParsingInlineAsm() && Disp && isa<MCSymbolRefExpr>(Disp)) { 1219 // Remove the '[' and ']' from the IR string. 1220 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Skip, Start, 1)); 1221 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Skip, End, 1)); 1222 } 1223 1224 if (!Disp) 1225 Disp = MCConstantExpr::Create(SM.getDisp(), getContext()); 1226 1227 // Parse the dot operator (e.g., [ebx].foo.bar). 1228 if (Tok.getString().startswith(".")) { 1229 SmallString<64> Err; 1230 const MCExpr *NewDisp; 1231 if (ParseIntelDotOperator(Disp, &NewDisp, Err)) 1232 return ErrorOperand(Tok.getLoc(), Err); 1233 1234 End = Tok.getEndLoc(); 1235 Parser.Lex(); // Eat the field. 1236 Disp = NewDisp; 1237 } 1238 1239 int BaseReg = SM.getBaseReg(); 1240 int IndexReg = SM.getIndexReg(); 1241 1242 // handle [-42] 1243 if (!BaseReg && !IndexReg) { 1244 if (!SegReg) 1245 return X86Operand::CreateMem(Disp, Start, End, Size); 1246 else 1247 return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, Start, End, Size); 1248 } 1249 1250 int Scale = SM.getScale(); 1251 return X86Operand::CreateMem(SegReg, Disp, BaseReg, IndexReg, Scale, Start, 1252 End, Size); 1253} 1254 1255// Inline assembly may use variable names with namespace alias qualifiers. 1256X86Operand *X86AsmParser::ParseIntelVarWithQualifier(const MCExpr *&Disp, 1257 StringRef &Identifier) { 1258 // We should only see Foo::Bar if we're parsing inline assembly. 1259 if (!isParsingInlineAsm()) 1260 return 0; 1261 1262 // If we don't see a ':' then there can't be a qualifier. 1263 if (getLexer().isNot(AsmToken::Colon)) 1264 return 0; 1265 1266 bool Done = false; 1267 const AsmToken &Tok = Parser.getTok(); 1268 AsmToken IdentEnd = Tok; 1269 while (!Done) { 1270 switch (getLexer().getKind()) { 1271 default: 1272 Done = true; 1273 break; 1274 case AsmToken::Colon: 1275 getLexer().Lex(); // Consume ':'. 1276 if (getLexer().isNot(AsmToken::Colon)) 1277 return ErrorOperand(Tok.getLoc(), "Expected ':' token!"); 1278 getLexer().Lex(); // Consume second ':'. 1279 if (getLexer().isNot(AsmToken::Identifier)) 1280 return ErrorOperand(Tok.getLoc(), "Expected an identifier token!"); 1281 break; 1282 case AsmToken::Identifier: 1283 IdentEnd = Tok; 1284 getLexer().Lex(); // Consume the identifier. 1285 break; 1286 } 1287 } 1288 1289 unsigned Len = IdentEnd.getLoc().getPointer() - Identifier.data(); 1290 Identifier = StringRef(Identifier.data(), Len + IdentEnd.getString().size()); 1291 MCSymbol *Sym = getContext().GetOrCreateSymbol(Identifier); 1292 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 1293 Disp = MCSymbolRefExpr::Create(Sym, Variant, getParser().getContext()); 1294 return 0; 1295} 1296 1297/// ParseIntelMemOperand - Parse intel style memory operand. 1298X86Operand *X86AsmParser::ParseIntelMemOperand(unsigned SegReg, 1299 uint64_t ImmDisp, 1300 SMLoc Start) { 1301 const AsmToken &Tok = Parser.getTok(); 1302 SMLoc End; 1303 1304 unsigned Size = getIntelMemOperandSize(Tok.getString()); 1305 if (Size) { 1306 Parser.Lex(); 1307 assert ((Tok.getString() == "PTR" || Tok.getString() == "ptr") && 1308 "Unexpected token!"); 1309 Parser.Lex(); 1310 } 1311 1312 // Parse ImmDisp [ BaseReg + Scale*IndexReg + Disp ]. 1313 if (getLexer().is(AsmToken::Integer)) { 1314 if (isParsingInlineAsm()) 1315 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_ImmPrefix, 1316 Tok.getLoc())); 1317 uint64_t ImmDisp = Tok.getIntVal(); 1318 Parser.Lex(); // Eat the integer. 1319 if (getLexer().isNot(AsmToken::LBrac)) 1320 return ErrorOperand(Start, "Expected '[' token!"); 1321 return ParseIntelBracExpression(SegReg, Start, ImmDisp, Size); 1322 } 1323 1324 if (getLexer().is(AsmToken::LBrac)) 1325 return ParseIntelBracExpression(SegReg, Start, ImmDisp, Size); 1326 1327 if (!ParseRegister(SegReg, Start, End)) { 1328 // Handel SegReg : [ ... ] 1329 if (getLexer().isNot(AsmToken::Colon)) 1330 return ErrorOperand(Start, "Expected ':' token!"); 1331 Parser.Lex(); // Eat : 1332 if (getLexer().isNot(AsmToken::LBrac)) 1333 return ErrorOperand(Start, "Expected '[' token!"); 1334 return ParseIntelBracExpression(SegReg, Start, ImmDisp, Size); 1335 } 1336 1337 const MCExpr *Disp = 0; 1338 StringRef Identifier = Tok.getString(); 1339 if (getParser().parseExpression(Disp, End)) 1340 return 0; 1341 1342 if (!isParsingInlineAsm()) 1343 return X86Operand::CreateMem(Disp, Start, End, Size); 1344 1345 if (X86Operand *Err = ParseIntelVarWithQualifier(Disp, Identifier)) 1346 return Err; 1347 1348 return CreateMemForInlineAsm(Disp, Start, End, Start, Size, Identifier); 1349} 1350 1351/// Parse the '.' operator. 1352bool X86AsmParser::ParseIntelDotOperator(const MCExpr *Disp, 1353 const MCExpr **NewDisp, 1354 SmallString<64> &Err) { 1355 const AsmToken &Tok = Parser.getTok(); 1356 uint64_t OrigDispVal, DotDispVal; 1357 1358 // FIXME: Handle non-constant expressions. 1359 if (const MCConstantExpr *OrigDisp = dyn_cast<MCConstantExpr>(Disp)) { 1360 OrigDispVal = OrigDisp->getValue(); 1361 } else { 1362 Err = "Non-constant offsets are not supported!"; 1363 return true; 1364 } 1365 1366 // Drop the '.'. 1367 StringRef DotDispStr = Tok.getString().drop_front(1); 1368 1369 // .Imm gets lexed as a real. 1370 if (Tok.is(AsmToken::Real)) { 1371 APInt DotDisp; 1372 DotDispStr.getAsInteger(10, DotDisp); 1373 DotDispVal = DotDisp.getZExtValue(); 1374 } else if (Tok.is(AsmToken::Identifier)) { 1375 // We should only see an identifier when parsing the original inline asm. 1376 // The front-end should rewrite this in terms of immediates. 1377 assert (isParsingInlineAsm() && "Unexpected field name!"); 1378 1379 unsigned DotDisp; 1380 std::pair<StringRef, StringRef> BaseMember = DotDispStr.split('.'); 1381 if (SemaCallback->LookupInlineAsmField(BaseMember.first, BaseMember.second, 1382 DotDisp)) { 1383 Err = "Unable to lookup field reference!"; 1384 return true; 1385 } 1386 DotDispVal = DotDisp; 1387 } else { 1388 Err = "Unexpected token type!"; 1389 return true; 1390 } 1391 1392 if (isParsingInlineAsm() && Tok.is(AsmToken::Identifier)) { 1393 SMLoc Loc = SMLoc::getFromPointer(DotDispStr.data()); 1394 unsigned Len = DotDispStr.size(); 1395 unsigned Val = OrigDispVal + DotDispVal; 1396 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_DotOperator, Loc, Len, 1397 Val)); 1398 } 1399 1400 *NewDisp = MCConstantExpr::Create(OrigDispVal + DotDispVal, getContext()); 1401 return false; 1402} 1403 1404/// Parse the 'offset' operator. This operator is used to specify the 1405/// location rather then the content of a variable. 1406X86Operand *X86AsmParser::ParseIntelOffsetOfOperator() { 1407 const AsmToken &Tok = Parser.getTok(); 1408 SMLoc OffsetOfLoc = Tok.getLoc(); 1409 Parser.Lex(); // Eat offset. 1410 assert (Tok.is(AsmToken::Identifier) && "Expected an identifier"); 1411 1412 const MCExpr *Val; 1413 SMLoc Start = Tok.getLoc(), End; 1414 StringRef Identifier = Tok.getString(); 1415 if (getParser().parsePrimaryExpr(Val, End)) 1416 return ErrorOperand(Start, "Unable to parse expression!"); 1417 1418 const MCExpr *Disp = 0; 1419 if (X86Operand *Err = ParseIntelVarWithQualifier(Disp, Identifier)) 1420 return Err; 1421 1422 // Don't emit the offset operator. 1423 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Skip, OffsetOfLoc, 7)); 1424 1425 // The offset operator will have an 'r' constraint, thus we need to create 1426 // register operand to ensure proper matching. Just pick a GPR based on 1427 // the size of a pointer. 1428 unsigned RegNo = is64BitMode() ? X86::RBX : X86::EBX; 1429 return X86Operand::CreateReg(RegNo, Start, End, /*GetAddress=*/true, 1430 OffsetOfLoc, Identifier); 1431} 1432 1433enum IntelOperatorKind { 1434 IOK_LENGTH, 1435 IOK_SIZE, 1436 IOK_TYPE 1437}; 1438 1439/// Parse the 'LENGTH', 'TYPE' and 'SIZE' operators. The LENGTH operator 1440/// returns the number of elements in an array. It returns the value 1 for 1441/// non-array variables. The SIZE operator returns the size of a C or C++ 1442/// variable. A variable's size is the product of its LENGTH and TYPE. The 1443/// TYPE operator returns the size of a C or C++ type or variable. If the 1444/// variable is an array, TYPE returns the size of a single element. 1445X86Operand *X86AsmParser::ParseIntelOperator(unsigned OpKind) { 1446 const AsmToken &Tok = Parser.getTok(); 1447 SMLoc TypeLoc = Tok.getLoc(); 1448 Parser.Lex(); // Eat operator. 1449 assert (Tok.is(AsmToken::Identifier) && "Expected an identifier"); 1450 1451 const MCExpr *Val; 1452 SMLoc Start = Tok.getLoc(), End; 1453 if (getParser().parsePrimaryExpr(Val, End)) 1454 return 0; 1455 1456 unsigned Length = 0, Size = 0, Type = 0; 1457 if (const MCSymbolRefExpr *SymRef = dyn_cast<MCSymbolRefExpr>(Val)) { 1458 const MCSymbol &Sym = SymRef->getSymbol(); 1459 // FIXME: The SemaLookup will fail if the name is anything other then an 1460 // identifier. 1461 // FIXME: Pass a valid SMLoc. 1462 bool IsVarDecl; 1463 if (!SemaCallback->LookupInlineAsmIdentifier(Sym.getName(), NULL, Length, 1464 Size, Type, IsVarDecl)) 1465 return ErrorOperand(Start, "Unable to lookup expr!"); 1466 } 1467 unsigned CVal; 1468 switch(OpKind) { 1469 default: llvm_unreachable("Unexpected operand kind!"); 1470 case IOK_LENGTH: CVal = Length; break; 1471 case IOK_SIZE: CVal = Size; break; 1472 case IOK_TYPE: CVal = Type; break; 1473 } 1474 1475 // Rewrite the type operator and the C or C++ type or variable in terms of an 1476 // immediate. E.g. TYPE foo -> $$4 1477 unsigned Len = End.getPointer() - TypeLoc.getPointer(); 1478 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Imm, TypeLoc, Len, CVal)); 1479 1480 const MCExpr *Imm = MCConstantExpr::Create(CVal, getContext()); 1481 return X86Operand::CreateImm(Imm, Start, End); 1482} 1483 1484X86Operand *X86AsmParser::ParseIntelOperand() { 1485 const AsmToken &Tok = Parser.getTok(); 1486 SMLoc Start = Tok.getLoc(), End; 1487 StringRef AsmTokStr = Tok.getString(); 1488 1489 // Offset, length, type and size operators. 1490 if (isParsingInlineAsm()) { 1491 if (AsmTokStr == "offset" || AsmTokStr == "OFFSET") 1492 return ParseIntelOffsetOfOperator(); 1493 if (AsmTokStr == "length" || AsmTokStr == "LENGTH") 1494 return ParseIntelOperator(IOK_LENGTH); 1495 if (AsmTokStr == "size" || AsmTokStr == "SIZE") 1496 return ParseIntelOperator(IOK_SIZE); 1497 if (AsmTokStr == "type" || AsmTokStr == "TYPE") 1498 return ParseIntelOperator(IOK_TYPE); 1499 } 1500 1501 // Immediate. 1502 if (getLexer().is(AsmToken::Integer) || getLexer().is(AsmToken::Real) || 1503 getLexer().is(AsmToken::Minus)) { 1504 const MCExpr *Val; 1505 bool isInteger = getLexer().is(AsmToken::Integer); 1506 if (!getParser().parseExpression(Val, End)) { 1507 if (isParsingInlineAsm()) 1508 InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_ImmPrefix, Start)); 1509 // Immediate. 1510 if (getLexer().isNot(AsmToken::LBrac)) 1511 return X86Operand::CreateImm(Val, Start, End); 1512 1513 // Only positive immediates are valid. 1514 if (!isInteger) { 1515 Error(Tok.getLoc(), "expected a positive immediate " 1516 "displacement before bracketed expr."); 1517 return 0; 1518 } 1519 1520 // Parse ImmDisp [ BaseReg + Scale*IndexReg + Disp ]. 1521 if (uint64_t ImmDisp = dyn_cast<MCConstantExpr>(Val)->getValue()) 1522 return ParseIntelMemOperand(/*SegReg=*/0, ImmDisp, Start); 1523 } 1524 } 1525 1526 // Register. 1527 unsigned RegNo = 0; 1528 if (!ParseRegister(RegNo, Start, End)) { 1529 // If this is a segment register followed by a ':', then this is the start 1530 // of a memory reference, otherwise this is a normal register reference. 1531 if (getLexer().isNot(AsmToken::Colon)) 1532 return X86Operand::CreateReg(RegNo, Start, End); 1533 1534 getParser().Lex(); // Eat the colon. 1535 return ParseIntelMemOperand(/*SegReg=*/RegNo, /*Disp=*/0, Start); 1536 } 1537 1538 // Memory operand. 1539 return ParseIntelMemOperand(/*SegReg=*/0, /*Disp=*/0, Start); 1540} 1541 1542X86Operand *X86AsmParser::ParseATTOperand() { 1543 switch (getLexer().getKind()) { 1544 default: 1545 // Parse a memory operand with no segment register. 1546 return ParseMemOperand(0, Parser.getTok().getLoc()); 1547 case AsmToken::Percent: { 1548 // Read the register. 1549 unsigned RegNo; 1550 SMLoc Start, End; 1551 if (ParseRegister(RegNo, Start, End)) return 0; 1552 if (RegNo == X86::EIZ || RegNo == X86::RIZ) { 1553 Error(Start, "%eiz and %riz can only be used as index registers", 1554 SMRange(Start, End)); 1555 return 0; 1556 } 1557 1558 // If this is a segment register followed by a ':', then this is the start 1559 // of a memory reference, otherwise this is a normal register reference. 1560 if (getLexer().isNot(AsmToken::Colon)) 1561 return X86Operand::CreateReg(RegNo, Start, End); 1562 1563 getParser().Lex(); // Eat the colon. 1564 return ParseMemOperand(RegNo, Start); 1565 } 1566 case AsmToken::Dollar: { 1567 // $42 -> immediate. 1568 SMLoc Start = Parser.getTok().getLoc(), End; 1569 Parser.Lex(); 1570 const MCExpr *Val; 1571 if (getParser().parseExpression(Val, End)) 1572 return 0; 1573 return X86Operand::CreateImm(Val, Start, End); 1574 } 1575 } 1576} 1577 1578/// ParseMemOperand: segment: disp(basereg, indexreg, scale). The '%ds:' prefix 1579/// has already been parsed if present. 1580X86Operand *X86AsmParser::ParseMemOperand(unsigned SegReg, SMLoc MemStart) { 1581 1582 // We have to disambiguate a parenthesized expression "(4+5)" from the start 1583 // of a memory operand with a missing displacement "(%ebx)" or "(,%eax)". The 1584 // only way to do this without lookahead is to eat the '(' and see what is 1585 // after it. 1586 const MCExpr *Disp = MCConstantExpr::Create(0, getParser().getContext()); 1587 if (getLexer().isNot(AsmToken::LParen)) { 1588 SMLoc ExprEnd; 1589 if (getParser().parseExpression(Disp, ExprEnd)) return 0; 1590 1591 // After parsing the base expression we could either have a parenthesized 1592 // memory address or not. If not, return now. If so, eat the (. 1593 if (getLexer().isNot(AsmToken::LParen)) { 1594 // Unless we have a segment register, treat this as an immediate. 1595 if (SegReg == 0) 1596 return X86Operand::CreateMem(Disp, MemStart, ExprEnd); 1597 return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, MemStart, ExprEnd); 1598 } 1599 1600 // Eat the '('. 1601 Parser.Lex(); 1602 } else { 1603 // Okay, we have a '('. We don't know if this is an expression or not, but 1604 // so we have to eat the ( to see beyond it. 1605 SMLoc LParenLoc = Parser.getTok().getLoc(); 1606 Parser.Lex(); // Eat the '('. 1607 1608 if (getLexer().is(AsmToken::Percent) || getLexer().is(AsmToken::Comma)) { 1609 // Nothing to do here, fall into the code below with the '(' part of the 1610 // memory operand consumed. 1611 } else { 1612 SMLoc ExprEnd; 1613 1614 // It must be an parenthesized expression, parse it now. 1615 if (getParser().parseParenExpression(Disp, ExprEnd)) 1616 return 0; 1617 1618 // After parsing the base expression we could either have a parenthesized 1619 // memory address or not. If not, return now. If so, eat the (. 1620 if (getLexer().isNot(AsmToken::LParen)) { 1621 // Unless we have a segment register, treat this as an immediate. 1622 if (SegReg == 0) 1623 return X86Operand::CreateMem(Disp, LParenLoc, ExprEnd); 1624 return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, MemStart, ExprEnd); 1625 } 1626 1627 // Eat the '('. 1628 Parser.Lex(); 1629 } 1630 } 1631 1632 // If we reached here, then we just ate the ( of the memory operand. Process 1633 // the rest of the memory operand. 1634 unsigned BaseReg = 0, IndexReg = 0, Scale = 1; 1635 SMLoc IndexLoc; 1636 1637 if (getLexer().is(AsmToken::Percent)) { 1638 SMLoc StartLoc, EndLoc; 1639 if (ParseRegister(BaseReg, StartLoc, EndLoc)) return 0; 1640 if (BaseReg == X86::EIZ || BaseReg == X86::RIZ) { 1641 Error(StartLoc, "eiz and riz can only be used as index registers", 1642 SMRange(StartLoc, EndLoc)); 1643 return 0; 1644 } 1645 } 1646 1647 if (getLexer().is(AsmToken::Comma)) { 1648 Parser.Lex(); // Eat the comma. 1649 IndexLoc = Parser.getTok().getLoc(); 1650 1651 // Following the comma we should have either an index register, or a scale 1652 // value. We don't support the later form, but we want to parse it 1653 // correctly. 1654 // 1655 // Not that even though it would be completely consistent to support syntax 1656 // like "1(%eax,,1)", the assembler doesn't. Use "eiz" or "riz" for this. 1657 if (getLexer().is(AsmToken::Percent)) { 1658 SMLoc L; 1659 if (ParseRegister(IndexReg, L, L)) return 0; 1660 1661 if (getLexer().isNot(AsmToken::RParen)) { 1662 // Parse the scale amount: 1663 // ::= ',' [scale-expression] 1664 if (getLexer().isNot(AsmToken::Comma)) { 1665 Error(Parser.getTok().getLoc(), 1666 "expected comma in scale expression"); 1667 return 0; 1668 } 1669 Parser.Lex(); // Eat the comma. 1670 1671 if (getLexer().isNot(AsmToken::RParen)) { 1672 SMLoc Loc = Parser.getTok().getLoc(); 1673 1674 int64_t ScaleVal; 1675 if (getParser().parseAbsoluteExpression(ScaleVal)){ 1676 Error(Loc, "expected scale expression"); 1677 return 0; 1678 } 1679 1680 // Validate the scale amount. 1681 if (ScaleVal != 1 && ScaleVal != 2 && ScaleVal != 4 && ScaleVal != 8){ 1682 Error(Loc, "scale factor in address must be 1, 2, 4 or 8"); 1683 return 0; 1684 } 1685 Scale = (unsigned)ScaleVal; 1686 } 1687 } 1688 } else if (getLexer().isNot(AsmToken::RParen)) { 1689 // A scale amount without an index is ignored. 1690 // index. 1691 SMLoc Loc = Parser.getTok().getLoc(); 1692 1693 int64_t Value; 1694 if (getParser().parseAbsoluteExpression(Value)) 1695 return 0; 1696 1697 if (Value != 1) 1698 Warning(Loc, "scale factor without index register is ignored"); 1699 Scale = 1; 1700 } 1701 } 1702 1703 // Ok, we've eaten the memory operand, verify we have a ')' and eat it too. 1704 if (getLexer().isNot(AsmToken::RParen)) { 1705 Error(Parser.getTok().getLoc(), "unexpected token in memory operand"); 1706 return 0; 1707 } 1708 SMLoc MemEnd = Parser.getTok().getEndLoc(); 1709 Parser.Lex(); // Eat the ')'. 1710 1711 // If we have both a base register and an index register make sure they are 1712 // both 64-bit or 32-bit registers. 1713 // To support VSIB, IndexReg can be 128-bit or 256-bit registers. 1714 if (BaseReg != 0 && IndexReg != 0) { 1715 if (X86MCRegisterClasses[X86::GR64RegClassID].contains(BaseReg) && 1716 (X86MCRegisterClasses[X86::GR16RegClassID].contains(IndexReg) || 1717 X86MCRegisterClasses[X86::GR32RegClassID].contains(IndexReg)) && 1718 IndexReg != X86::RIZ) { 1719 Error(IndexLoc, "index register is 32-bit, but base register is 64-bit"); 1720 return 0; 1721 } 1722 if (X86MCRegisterClasses[X86::GR32RegClassID].contains(BaseReg) && 1723 (X86MCRegisterClasses[X86::GR16RegClassID].contains(IndexReg) || 1724 X86MCRegisterClasses[X86::GR64RegClassID].contains(IndexReg)) && 1725 IndexReg != X86::EIZ){ 1726 Error(IndexLoc, "index register is 64-bit, but base register is 32-bit"); 1727 return 0; 1728 } 1729 } 1730 1731 return X86Operand::CreateMem(SegReg, Disp, BaseReg, IndexReg, Scale, 1732 MemStart, MemEnd); 1733} 1734 1735bool X86AsmParser:: 1736ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc, 1737 SmallVectorImpl<MCParsedAsmOperand*> &Operands) { 1738 InstInfo = &Info; 1739 StringRef PatchedName = Name; 1740 1741 // FIXME: Hack to recognize setneb as setne. 1742 if (PatchedName.startswith("set") && PatchedName.endswith("b") && 1743 PatchedName != "setb" && PatchedName != "setnb") 1744 PatchedName = PatchedName.substr(0, Name.size()-1); 1745 1746 // FIXME: Hack to recognize cmp<comparison code>{ss,sd,ps,pd}. 1747 const MCExpr *ExtraImmOp = 0; 1748 if ((PatchedName.startswith("cmp") || PatchedName.startswith("vcmp")) && 1749 (PatchedName.endswith("ss") || PatchedName.endswith("sd") || 1750 PatchedName.endswith("ps") || PatchedName.endswith("pd"))) { 1751 bool IsVCMP = PatchedName[0] == 'v'; 1752 unsigned SSECCIdx = IsVCMP ? 4 : 3; 1753 unsigned SSEComparisonCode = StringSwitch<unsigned>( 1754 PatchedName.slice(SSECCIdx, PatchedName.size() - 2)) 1755 .Case("eq", 0x00) 1756 .Case("lt", 0x01) 1757 .Case("le", 0x02) 1758 .Case("unord", 0x03) 1759 .Case("neq", 0x04) 1760 .Case("nlt", 0x05) 1761 .Case("nle", 0x06) 1762 .Case("ord", 0x07) 1763 /* AVX only from here */ 1764 .Case("eq_uq", 0x08) 1765 .Case("nge", 0x09) 1766 .Case("ngt", 0x0A) 1767 .Case("false", 0x0B) 1768 .Case("neq_oq", 0x0C) 1769 .Case("ge", 0x0D) 1770 .Case("gt", 0x0E) 1771 .Case("true", 0x0F) 1772 .Case("eq_os", 0x10) 1773 .Case("lt_oq", 0x11) 1774 .Case("le_oq", 0x12) 1775 .Case("unord_s", 0x13) 1776 .Case("neq_us", 0x14) 1777 .Case("nlt_uq", 0x15) 1778 .Case("nle_uq", 0x16) 1779 .Case("ord_s", 0x17) 1780 .Case("eq_us", 0x18) 1781 .Case("nge_uq", 0x19) 1782 .Case("ngt_uq", 0x1A) 1783 .Case("false_os", 0x1B) 1784 .Case("neq_os", 0x1C) 1785 .Case("ge_oq", 0x1D) 1786 .Case("gt_oq", 0x1E) 1787 .Case("true_us", 0x1F) 1788 .Default(~0U); 1789 if (SSEComparisonCode != ~0U && (IsVCMP || SSEComparisonCode < 8)) { 1790 ExtraImmOp = MCConstantExpr::Create(SSEComparisonCode, 1791 getParser().getContext()); 1792 if (PatchedName.endswith("ss")) { 1793 PatchedName = IsVCMP ? "vcmpss" : "cmpss"; 1794 } else if (PatchedName.endswith("sd")) { 1795 PatchedName = IsVCMP ? "vcmpsd" : "cmpsd"; 1796 } else if (PatchedName.endswith("ps")) { 1797 PatchedName = IsVCMP ? "vcmpps" : "cmpps"; 1798 } else { 1799 assert(PatchedName.endswith("pd") && "Unexpected mnemonic!"); 1800 PatchedName = IsVCMP ? "vcmppd" : "cmppd"; 1801 } 1802 } 1803 } 1804 1805 Operands.push_back(X86Operand::CreateToken(PatchedName, NameLoc)); 1806 1807 if (ExtraImmOp && !isParsingIntelSyntax()) 1808 Operands.push_back(X86Operand::CreateImm(ExtraImmOp, NameLoc, NameLoc)); 1809 1810 // Determine whether this is an instruction prefix. 1811 bool isPrefix = 1812 Name == "lock" || Name == "rep" || 1813 Name == "repe" || Name == "repz" || 1814 Name == "repne" || Name == "repnz" || 1815 Name == "rex64" || Name == "data16"; 1816 1817 1818 // This does the actual operand parsing. Don't parse any more if we have a 1819 // prefix juxtaposed with an operation like "lock incl 4(%rax)", because we 1820 // just want to parse the "lock" as the first instruction and the "incl" as 1821 // the next one. 1822 if (getLexer().isNot(AsmToken::EndOfStatement) && !isPrefix) { 1823 1824 // Parse '*' modifier. 1825 if (getLexer().is(AsmToken::Star)) { 1826 SMLoc Loc = Parser.getTok().getLoc(); 1827 Operands.push_back(X86Operand::CreateToken("*", Loc)); 1828 Parser.Lex(); // Eat the star. 1829 } 1830 1831 // Read the first operand. 1832 if (X86Operand *Op = ParseOperand()) 1833 Operands.push_back(Op); 1834 else { 1835 Parser.eatToEndOfStatement(); 1836 return true; 1837 } 1838 1839 while (getLexer().is(AsmToken::Comma)) { 1840 Parser.Lex(); // Eat the comma. 1841 1842 // Parse and remember the operand. 1843 if (X86Operand *Op = ParseOperand()) 1844 Operands.push_back(Op); 1845 else { 1846 Parser.eatToEndOfStatement(); 1847 return true; 1848 } 1849 } 1850 1851 if (getLexer().isNot(AsmToken::EndOfStatement)) { 1852 SMLoc Loc = getLexer().getLoc(); 1853 Parser.eatToEndOfStatement(); 1854 return Error(Loc, "unexpected token in argument list"); 1855 } 1856 } 1857 1858 if (getLexer().is(AsmToken::EndOfStatement)) 1859 Parser.Lex(); // Consume the EndOfStatement 1860 else if (isPrefix && getLexer().is(AsmToken::Slash)) 1861 Parser.Lex(); // Consume the prefix separator Slash 1862 1863 if (ExtraImmOp && isParsingIntelSyntax()) 1864 Operands.push_back(X86Operand::CreateImm(ExtraImmOp, NameLoc, NameLoc)); 1865 1866 // This is a terrible hack to handle "out[bwl]? %al, (%dx)" -> 1867 // "outb %al, %dx". Out doesn't take a memory form, but this is a widely 1868 // documented form in various unofficial manuals, so a lot of code uses it. 1869 if ((Name == "outb" || Name == "outw" || Name == "outl" || Name == "out") && 1870 Operands.size() == 3) { 1871 X86Operand &Op = *(X86Operand*)Operands.back(); 1872 if (Op.isMem() && Op.Mem.SegReg == 0 && 1873 isa<MCConstantExpr>(Op.Mem.Disp) && 1874 cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 && 1875 Op.Mem.BaseReg == MatchRegisterName("dx") && Op.Mem.IndexReg == 0) { 1876 SMLoc Loc = Op.getEndLoc(); 1877 Operands.back() = X86Operand::CreateReg(Op.Mem.BaseReg, Loc, Loc); 1878 delete &Op; 1879 } 1880 } 1881 // Same hack for "in[bwl]? (%dx), %al" -> "inb %dx, %al". 1882 if ((Name == "inb" || Name == "inw" || Name == "inl" || Name == "in") && 1883 Operands.size() == 3) { 1884 X86Operand &Op = *(X86Operand*)Operands.begin()[1]; 1885 if (Op.isMem() && Op.Mem.SegReg == 0 && 1886 isa<MCConstantExpr>(Op.Mem.Disp) && 1887 cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 && 1888 Op.Mem.BaseReg == MatchRegisterName("dx") && Op.Mem.IndexReg == 0) { 1889 SMLoc Loc = Op.getEndLoc(); 1890 Operands.begin()[1] = X86Operand::CreateReg(Op.Mem.BaseReg, Loc, Loc); 1891 delete &Op; 1892 } 1893 } 1894 // Transform "ins[bwl] %dx, %es:(%edi)" into "ins[bwl]" 1895 if (Name.startswith("ins") && Operands.size() == 3 && 1896 (Name == "insb" || Name == "insw" || Name == "insl")) { 1897 X86Operand &Op = *(X86Operand*)Operands.begin()[1]; 1898 X86Operand &Op2 = *(X86Operand*)Operands.begin()[2]; 1899 if (Op.isReg() && Op.getReg() == X86::DX && isDstOp(Op2)) { 1900 Operands.pop_back(); 1901 Operands.pop_back(); 1902 delete &Op; 1903 delete &Op2; 1904 } 1905 } 1906 1907 // Transform "outs[bwl] %ds:(%esi), %dx" into "out[bwl]" 1908 if (Name.startswith("outs") && Operands.size() == 3 && 1909 (Name == "outsb" || Name == "outsw" || Name == "outsl")) { 1910 X86Operand &Op = *(X86Operand*)Operands.begin()[1]; 1911 X86Operand &Op2 = *(X86Operand*)Operands.begin()[2]; 1912 if (isSrcOp(Op) && Op2.isReg() && Op2.getReg() == X86::DX) { 1913 Operands.pop_back(); 1914 Operands.pop_back(); 1915 delete &Op; 1916 delete &Op2; 1917 } 1918 } 1919 1920 // Transform "movs[bwl] %ds:(%esi), %es:(%edi)" into "movs[bwl]" 1921 if (Name.startswith("movs") && Operands.size() == 3 && 1922 (Name == "movsb" || Name == "movsw" || Name == "movsl" || 1923 (is64BitMode() && Name == "movsq"))) { 1924 X86Operand &Op = *(X86Operand*)Operands.begin()[1]; 1925 X86Operand &Op2 = *(X86Operand*)Operands.begin()[2]; 1926 if (isSrcOp(Op) && isDstOp(Op2)) { 1927 Operands.pop_back(); 1928 Operands.pop_back(); 1929 delete &Op; 1930 delete &Op2; 1931 } 1932 } 1933 // Transform "lods[bwl] %ds:(%esi),{%al,%ax,%eax,%rax}" into "lods[bwl]" 1934 if (Name.startswith("lods") && Operands.size() == 3 && 1935 (Name == "lods" || Name == "lodsb" || Name == "lodsw" || 1936 Name == "lodsl" || (is64BitMode() && Name == "lodsq"))) { 1937 X86Operand *Op1 = static_cast<X86Operand*>(Operands[1]); 1938 X86Operand *Op2 = static_cast<X86Operand*>(Operands[2]); 1939 if (isSrcOp(*Op1) && Op2->isReg()) { 1940 const char *ins; 1941 unsigned reg = Op2->getReg(); 1942 bool isLods = Name == "lods"; 1943 if (reg == X86::AL && (isLods || Name == "lodsb")) 1944 ins = "lodsb"; 1945 else if (reg == X86::AX && (isLods || Name == "lodsw")) 1946 ins = "lodsw"; 1947 else if (reg == X86::EAX && (isLods || Name == "lodsl")) 1948 ins = "lodsl"; 1949 else if (reg == X86::RAX && (isLods || Name == "lodsq")) 1950 ins = "lodsq"; 1951 else 1952 ins = NULL; 1953 if (ins != NULL) { 1954 Operands.pop_back(); 1955 Operands.pop_back(); 1956 delete Op1; 1957 delete Op2; 1958 if (Name != ins) 1959 static_cast<X86Operand*>(Operands[0])->setTokenValue(ins); 1960 } 1961 } 1962 } 1963 // Transform "stos[bwl] {%al,%ax,%eax,%rax},%es:(%edi)" into "stos[bwl]" 1964 if (Name.startswith("stos") && Operands.size() == 3 && 1965 (Name == "stos" || Name == "stosb" || Name == "stosw" || 1966 Name == "stosl" || (is64BitMode() && Name == "stosq"))) { 1967 X86Operand *Op1 = static_cast<X86Operand*>(Operands[1]); 1968 X86Operand *Op2 = static_cast<X86Operand*>(Operands[2]); 1969 if (isDstOp(*Op2) && Op1->isReg()) { 1970 const char *ins; 1971 unsigned reg = Op1->getReg(); 1972 bool isStos = Name == "stos"; 1973 if (reg == X86::AL && (isStos || Name == "stosb")) 1974 ins = "stosb"; 1975 else if (reg == X86::AX && (isStos || Name == "stosw")) 1976 ins = "stosw"; 1977 else if (reg == X86::EAX && (isStos || Name == "stosl")) 1978 ins = "stosl"; 1979 else if (reg == X86::RAX && (isStos || Name == "stosq")) 1980 ins = "stosq"; 1981 else 1982 ins = NULL; 1983 if (ins != NULL) { 1984 Operands.pop_back(); 1985 Operands.pop_back(); 1986 delete Op1; 1987 delete Op2; 1988 if (Name != ins) 1989 static_cast<X86Operand*>(Operands[0])->setTokenValue(ins); 1990 } 1991 } 1992 } 1993 1994 // FIXME: Hack to handle recognize s{hr,ar,hl} $1, <op>. Canonicalize to 1995 // "shift <op>". 1996 if ((Name.startswith("shr") || Name.startswith("sar") || 1997 Name.startswith("shl") || Name.startswith("sal") || 1998 Name.startswith("rcl") || Name.startswith("rcr") || 1999 Name.startswith("rol") || Name.startswith("ror")) && 2000 Operands.size() == 3) { 2001 if (isParsingIntelSyntax()) { 2002 // Intel syntax 2003 X86Operand *Op1 = static_cast<X86Operand*>(Operands[2]); 2004 if (Op1->isImm() && isa<MCConstantExpr>(Op1->getImm()) && 2005 cast<MCConstantExpr>(Op1->getImm())->getValue() == 1) { 2006 delete Operands[2]; 2007 Operands.pop_back(); 2008 } 2009 } else { 2010 X86Operand *Op1 = static_cast<X86Operand*>(Operands[1]); 2011 if (Op1->isImm() && isa<MCConstantExpr>(Op1->getImm()) && 2012 cast<MCConstantExpr>(Op1->getImm())->getValue() == 1) { 2013 delete Operands[1]; 2014 Operands.erase(Operands.begin() + 1); 2015 } 2016 } 2017 } 2018 2019 // Transforms "int $3" into "int3" as a size optimization. We can't write an 2020 // instalias with an immediate operand yet. 2021 if (Name == "int" && Operands.size() == 2) { 2022 X86Operand *Op1 = static_cast<X86Operand*>(Operands[1]); 2023 if (Op1->isImm() && isa<MCConstantExpr>(Op1->getImm()) && 2024 cast<MCConstantExpr>(Op1->getImm())->getValue() == 3) { 2025 delete Operands[1]; 2026 Operands.erase(Operands.begin() + 1); 2027 static_cast<X86Operand*>(Operands[0])->setTokenValue("int3"); 2028 } 2029 } 2030 2031 return false; 2032} 2033 2034static bool convertToSExti8(MCInst &Inst, unsigned Opcode, unsigned Reg, 2035 bool isCmp) { 2036 MCInst TmpInst; 2037 TmpInst.setOpcode(Opcode); 2038 if (!isCmp) 2039 TmpInst.addOperand(MCOperand::CreateReg(Reg)); 2040 TmpInst.addOperand(MCOperand::CreateReg(Reg)); 2041 TmpInst.addOperand(Inst.getOperand(0)); 2042 Inst = TmpInst; 2043 return true; 2044} 2045 2046static bool convert16i16to16ri8(MCInst &Inst, unsigned Opcode, 2047 bool isCmp = false) { 2048 if (!Inst.getOperand(0).isImm() || 2049 !isImmSExti16i8Value(Inst.getOperand(0).getImm())) 2050 return false; 2051 2052 return convertToSExti8(Inst, Opcode, X86::AX, isCmp); 2053} 2054 2055static bool convert32i32to32ri8(MCInst &Inst, unsigned Opcode, 2056 bool isCmp = false) { 2057 if (!Inst.getOperand(0).isImm() || 2058 !isImmSExti32i8Value(Inst.getOperand(0).getImm())) 2059 return false; 2060 2061 return convertToSExti8(Inst, Opcode, X86::EAX, isCmp); 2062} 2063 2064static bool convert64i32to64ri8(MCInst &Inst, unsigned Opcode, 2065 bool isCmp = false) { 2066 if (!Inst.getOperand(0).isImm() || 2067 !isImmSExti64i8Value(Inst.getOperand(0).getImm())) 2068 return false; 2069 2070 return convertToSExti8(Inst, Opcode, X86::RAX, isCmp); 2071} 2072 2073bool X86AsmParser:: 2074processInstruction(MCInst &Inst, 2075 const SmallVectorImpl<MCParsedAsmOperand*> &Ops) { 2076 switch (Inst.getOpcode()) { 2077 default: return false; 2078 case X86::AND16i16: return convert16i16to16ri8(Inst, X86::AND16ri8); 2079 case X86::AND32i32: return convert32i32to32ri8(Inst, X86::AND32ri8); 2080 case X86::AND64i32: return convert64i32to64ri8(Inst, X86::AND64ri8); 2081 case X86::XOR16i16: return convert16i16to16ri8(Inst, X86::XOR16ri8); 2082 case X86::XOR32i32: return convert32i32to32ri8(Inst, X86::XOR32ri8); 2083 case X86::XOR64i32: return convert64i32to64ri8(Inst, X86::XOR64ri8); 2084 case X86::OR16i16: return convert16i16to16ri8(Inst, X86::OR16ri8); 2085 case X86::OR32i32: return convert32i32to32ri8(Inst, X86::OR32ri8); 2086 case X86::OR64i32: return convert64i32to64ri8(Inst, X86::OR64ri8); 2087 case X86::CMP16i16: return convert16i16to16ri8(Inst, X86::CMP16ri8, true); 2088 case X86::CMP32i32: return convert32i32to32ri8(Inst, X86::CMP32ri8, true); 2089 case X86::CMP64i32: return convert64i32to64ri8(Inst, X86::CMP64ri8, true); 2090 case X86::ADD16i16: return convert16i16to16ri8(Inst, X86::ADD16ri8); 2091 case X86::ADD32i32: return convert32i32to32ri8(Inst, X86::ADD32ri8); 2092 case X86::ADD64i32: return convert64i32to64ri8(Inst, X86::ADD64ri8); 2093 case X86::SUB16i16: return convert16i16to16ri8(Inst, X86::SUB16ri8); 2094 case X86::SUB32i32: return convert32i32to32ri8(Inst, X86::SUB32ri8); 2095 case X86::SUB64i32: return convert64i32to64ri8(Inst, X86::SUB64ri8); 2096 case X86::ADC16i16: return convert16i16to16ri8(Inst, X86::ADC16ri8); 2097 case X86::ADC32i32: return convert32i32to32ri8(Inst, X86::ADC32ri8); 2098 case X86::ADC64i32: return convert64i32to64ri8(Inst, X86::ADC64ri8); 2099 case X86::SBB16i16: return convert16i16to16ri8(Inst, X86::SBB16ri8); 2100 case X86::SBB32i32: return convert32i32to32ri8(Inst, X86::SBB32ri8); 2101 case X86::SBB64i32: return convert64i32to64ri8(Inst, X86::SBB64ri8); 2102 } 2103} 2104 2105static const char *getSubtargetFeatureName(unsigned Val); 2106bool X86AsmParser:: 2107MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode, 2108 SmallVectorImpl<MCParsedAsmOperand*> &Operands, 2109 MCStreamer &Out, unsigned &ErrorInfo, 2110 bool MatchingInlineAsm) { 2111 assert(!Operands.empty() && "Unexpect empty operand list!"); 2112 X86Operand *Op = static_cast<X86Operand*>(Operands[0]); 2113 assert(Op->isToken() && "Leading operand should always be a mnemonic!"); 2114 ArrayRef<SMRange> EmptyRanges = ArrayRef<SMRange>(); 2115 2116 // First, handle aliases that expand to multiple instructions. 2117 // FIXME: This should be replaced with a real .td file alias mechanism. 2118 // Also, MatchInstructionImpl should actually *do* the EmitInstruction 2119 // call. 2120 if (Op->getToken() == "fstsw" || Op->getToken() == "fstcw" || 2121 Op->getToken() == "fstsww" || Op->getToken() == "fstcww" || 2122 Op->getToken() == "finit" || Op->getToken() == "fsave" || 2123 Op->getToken() == "fstenv" || Op->getToken() == "fclex") { 2124 MCInst Inst; 2125 Inst.setOpcode(X86::WAIT); 2126 Inst.setLoc(IDLoc); 2127 if (!MatchingInlineAsm) 2128 Out.EmitInstruction(Inst); 2129 2130 const char *Repl = 2131 StringSwitch<const char*>(Op->getToken()) 2132 .Case("finit", "fninit") 2133 .Case("fsave", "fnsave") 2134 .Case("fstcw", "fnstcw") 2135 .Case("fstcww", "fnstcw") 2136 .Case("fstenv", "fnstenv") 2137 .Case("fstsw", "fnstsw") 2138 .Case("fstsww", "fnstsw") 2139 .Case("fclex", "fnclex") 2140 .Default(0); 2141 assert(Repl && "Unknown wait-prefixed instruction"); 2142 delete Operands[0]; 2143 Operands[0] = X86Operand::CreateToken(Repl, IDLoc); 2144 } 2145 2146 bool WasOriginallyInvalidOperand = false; 2147 MCInst Inst; 2148 2149 // First, try a direct match. 2150 switch (MatchInstructionImpl(Operands, Inst, 2151 ErrorInfo, MatchingInlineAsm, 2152 isParsingIntelSyntax())) { 2153 default: break; 2154 case Match_Success: 2155 // Some instructions need post-processing to, for example, tweak which 2156 // encoding is selected. Loop on it while changes happen so the 2157 // individual transformations can chain off each other. 2158 if (!MatchingInlineAsm) 2159 while (processInstruction(Inst, Operands)) 2160 ; 2161 2162 Inst.setLoc(IDLoc); 2163 if (!MatchingInlineAsm) 2164 Out.EmitInstruction(Inst); 2165 Opcode = Inst.getOpcode(); 2166 return false; 2167 case Match_MissingFeature: { 2168 assert(ErrorInfo && "Unknown missing feature!"); 2169 // Special case the error message for the very common case where only 2170 // a single subtarget feature is missing. 2171 std::string Msg = "instruction requires:"; 2172 unsigned Mask = 1; 2173 for (unsigned i = 0; i < (sizeof(ErrorInfo)*8-1); ++i) { 2174 if (ErrorInfo & Mask) { 2175 Msg += " "; 2176 Msg += getSubtargetFeatureName(ErrorInfo & Mask); 2177 } 2178 Mask <<= 1; 2179 } 2180 return Error(IDLoc, Msg, EmptyRanges, MatchingInlineAsm); 2181 } 2182 case Match_InvalidOperand: 2183 WasOriginallyInvalidOperand = true; 2184 break; 2185 case Match_MnemonicFail: 2186 break; 2187 } 2188 2189 // FIXME: Ideally, we would only attempt suffix matches for things which are 2190 // valid prefixes, and we could just infer the right unambiguous 2191 // type. However, that requires substantially more matcher support than the 2192 // following hack. 2193 2194 // Change the operand to point to a temporary token. 2195 StringRef Base = Op->getToken(); 2196 SmallString<16> Tmp; 2197 Tmp += Base; 2198 Tmp += ' '; 2199 Op->setTokenValue(Tmp.str()); 2200 2201 // If this instruction starts with an 'f', then it is a floating point stack 2202 // instruction. These come in up to three forms for 32-bit, 64-bit, and 2203 // 80-bit floating point, which use the suffixes s,l,t respectively. 2204 // 2205 // Otherwise, we assume that this may be an integer instruction, which comes 2206 // in 8/16/32/64-bit forms using the b,w,l,q suffixes respectively. 2207 const char *Suffixes = Base[0] != 'f' ? "bwlq" : "slt\0"; 2208 2209 // Check for the various suffix matches. 2210 Tmp[Base.size()] = Suffixes[0]; 2211 unsigned ErrorInfoIgnore; 2212 unsigned ErrorInfoMissingFeature = 0; // Init suppresses compiler warnings. 2213 unsigned Match1, Match2, Match3, Match4; 2214 2215 Match1 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore, 2216 isParsingIntelSyntax()); 2217 // If this returned as a missing feature failure, remember that. 2218 if (Match1 == Match_MissingFeature) 2219 ErrorInfoMissingFeature = ErrorInfoIgnore; 2220 Tmp[Base.size()] = Suffixes[1]; 2221 Match2 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore, 2222 isParsingIntelSyntax()); 2223 // If this returned as a missing feature failure, remember that. 2224 if (Match2 == Match_MissingFeature) 2225 ErrorInfoMissingFeature = ErrorInfoIgnore; 2226 Tmp[Base.size()] = Suffixes[2]; 2227 Match3 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore, 2228 isParsingIntelSyntax()); 2229 // If this returned as a missing feature failure, remember that. 2230 if (Match3 == Match_MissingFeature) 2231 ErrorInfoMissingFeature = ErrorInfoIgnore; 2232 Tmp[Base.size()] = Suffixes[3]; 2233 Match4 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore, 2234 isParsingIntelSyntax()); 2235 // If this returned as a missing feature failure, remember that. 2236 if (Match4 == Match_MissingFeature) 2237 ErrorInfoMissingFeature = ErrorInfoIgnore; 2238 2239 // Restore the old token. 2240 Op->setTokenValue(Base); 2241 2242 // If exactly one matched, then we treat that as a successful match (and the 2243 // instruction will already have been filled in correctly, since the failing 2244 // matches won't have modified it). 2245 unsigned NumSuccessfulMatches = 2246 (Match1 == Match_Success) + (Match2 == Match_Success) + 2247 (Match3 == Match_Success) + (Match4 == Match_Success); 2248 if (NumSuccessfulMatches == 1) { 2249 Inst.setLoc(IDLoc); 2250 if (!MatchingInlineAsm) 2251 Out.EmitInstruction(Inst); 2252 Opcode = Inst.getOpcode(); 2253 return false; 2254 } 2255 2256 // Otherwise, the match failed, try to produce a decent error message. 2257 2258 // If we had multiple suffix matches, then identify this as an ambiguous 2259 // match. 2260 if (NumSuccessfulMatches > 1) { 2261 char MatchChars[4]; 2262 unsigned NumMatches = 0; 2263 if (Match1 == Match_Success) MatchChars[NumMatches++] = Suffixes[0]; 2264 if (Match2 == Match_Success) MatchChars[NumMatches++] = Suffixes[1]; 2265 if (Match3 == Match_Success) MatchChars[NumMatches++] = Suffixes[2]; 2266 if (Match4 == Match_Success) MatchChars[NumMatches++] = Suffixes[3]; 2267 2268 SmallString<126> Msg; 2269 raw_svector_ostream OS(Msg); 2270 OS << "ambiguous instructions require an explicit suffix (could be "; 2271 for (unsigned i = 0; i != NumMatches; ++i) { 2272 if (i != 0) 2273 OS << ", "; 2274 if (i + 1 == NumMatches) 2275 OS << "or "; 2276 OS << "'" << Base << MatchChars[i] << "'"; 2277 } 2278 OS << ")"; 2279 Error(IDLoc, OS.str(), EmptyRanges, MatchingInlineAsm); 2280 return true; 2281 } 2282 2283 // Okay, we know that none of the variants matched successfully. 2284 2285 // If all of the instructions reported an invalid mnemonic, then the original 2286 // mnemonic was invalid. 2287 if ((Match1 == Match_MnemonicFail) && (Match2 == Match_MnemonicFail) && 2288 (Match3 == Match_MnemonicFail) && (Match4 == Match_MnemonicFail)) { 2289 if (!WasOriginallyInvalidOperand) { 2290 ArrayRef<SMRange> Ranges = MatchingInlineAsm ? EmptyRanges : 2291 Op->getLocRange(); 2292 return Error(IDLoc, "invalid instruction mnemonic '" + Base + "'", 2293 Ranges, MatchingInlineAsm); 2294 } 2295 2296 // Recover location info for the operand if we know which was the problem. 2297 if (ErrorInfo != ~0U) { 2298 if (ErrorInfo >= Operands.size()) 2299 return Error(IDLoc, "too few operands for instruction", 2300 EmptyRanges, MatchingInlineAsm); 2301 2302 X86Operand *Operand = (X86Operand*)Operands[ErrorInfo]; 2303 if (Operand->getStartLoc().isValid()) { 2304 SMRange OperandRange = Operand->getLocRange(); 2305 return Error(Operand->getStartLoc(), "invalid operand for instruction", 2306 OperandRange, MatchingInlineAsm); 2307 } 2308 } 2309 2310 return Error(IDLoc, "invalid operand for instruction", EmptyRanges, 2311 MatchingInlineAsm); 2312 } 2313 2314 // If one instruction matched with a missing feature, report this as a 2315 // missing feature. 2316 if ((Match1 == Match_MissingFeature) + (Match2 == Match_MissingFeature) + 2317 (Match3 == Match_MissingFeature) + (Match4 == Match_MissingFeature) == 1){ 2318 std::string Msg = "instruction requires:"; 2319 unsigned Mask = 1; 2320 for (unsigned i = 0; i < (sizeof(ErrorInfoMissingFeature)*8-1); ++i) { 2321 if (ErrorInfoMissingFeature & Mask) { 2322 Msg += " "; 2323 Msg += getSubtargetFeatureName(ErrorInfoMissingFeature & Mask); 2324 } 2325 Mask <<= 1; 2326 } 2327 return Error(IDLoc, Msg, EmptyRanges, MatchingInlineAsm); 2328 } 2329 2330 // If one instruction matched with an invalid operand, report this as an 2331 // operand failure. 2332 if ((Match1 == Match_InvalidOperand) + (Match2 == Match_InvalidOperand) + 2333 (Match3 == Match_InvalidOperand) + (Match4 == Match_InvalidOperand) == 1){ 2334 Error(IDLoc, "invalid operand for instruction", EmptyRanges, 2335 MatchingInlineAsm); 2336 return true; 2337 } 2338 2339 // If all of these were an outright failure, report it in a useless way. 2340 Error(IDLoc, "unknown use of instruction mnemonic without a size suffix", 2341 EmptyRanges, MatchingInlineAsm); 2342 return true; 2343} 2344 2345 2346bool X86AsmParser::ParseDirective(AsmToken DirectiveID) { 2347 StringRef IDVal = DirectiveID.getIdentifier(); 2348 if (IDVal == ".word") 2349 return ParseDirectiveWord(2, DirectiveID.getLoc()); 2350 else if (IDVal.startswith(".code")) 2351 return ParseDirectiveCode(IDVal, DirectiveID.getLoc()); 2352 else if (IDVal.startswith(".att_syntax")) { 2353 getParser().setAssemblerDialect(0); 2354 return false; 2355 } else if (IDVal.startswith(".intel_syntax")) { 2356 getParser().setAssemblerDialect(1); 2357 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2358 if(Parser.getTok().getString() == "noprefix") { 2359 // FIXME : Handle noprefix 2360 Parser.Lex(); 2361 } else 2362 return true; 2363 } 2364 return false; 2365 } 2366 return true; 2367} 2368 2369/// ParseDirectiveWord 2370/// ::= .word [ expression (, expression)* ] 2371bool X86AsmParser::ParseDirectiveWord(unsigned Size, SMLoc L) { 2372 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2373 for (;;) { 2374 const MCExpr *Value; 2375 if (getParser().parseExpression(Value)) 2376 return true; 2377 2378 getParser().getStreamer().EmitValue(Value, Size); 2379 2380 if (getLexer().is(AsmToken::EndOfStatement)) 2381 break; 2382 2383 // FIXME: Improve diagnostic. 2384 if (getLexer().isNot(AsmToken::Comma)) 2385 return Error(L, "unexpected token in directive"); 2386 Parser.Lex(); 2387 } 2388 } 2389 2390 Parser.Lex(); 2391 return false; 2392} 2393 2394/// ParseDirectiveCode 2395/// ::= .code32 | .code64 2396bool X86AsmParser::ParseDirectiveCode(StringRef IDVal, SMLoc L) { 2397 if (IDVal == ".code32") { 2398 Parser.Lex(); 2399 if (is64BitMode()) { 2400 SwitchMode(); 2401 getParser().getStreamer().EmitAssemblerFlag(MCAF_Code32); 2402 } 2403 } else if (IDVal == ".code64") { 2404 Parser.Lex(); 2405 if (!is64BitMode()) { 2406 SwitchMode(); 2407 getParser().getStreamer().EmitAssemblerFlag(MCAF_Code64); 2408 } 2409 } else { 2410 return Error(L, "unexpected directive " + IDVal); 2411 } 2412 2413 return false; 2414} 2415 2416// Force static initialization. 2417extern "C" void LLVMInitializeX86AsmParser() { 2418 RegisterMCAsmParser<X86AsmParser> X(TheX86_32Target); 2419 RegisterMCAsmParser<X86AsmParser> Y(TheX86_64Target); 2420} 2421 2422#define GET_REGISTER_MATCHER 2423#define GET_MATCHER_IMPLEMENTATION 2424#define GET_SUBTARGET_FEATURE_NAME 2425#include "X86GenAsmMatcher.inc" 2426