ARMAsmParser.cpp revision 1d6a26507bfd75758f5c8a29bccf577784ead751
1//===-- ARMAsmParser.cpp - Parse ARM 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 "ARM.h" 11#include "ARMAddressingModes.h" 12#include "ARMSubtarget.h" 13#include "llvm/MC/MCParser/MCAsmLexer.h" 14#include "llvm/MC/MCParser/MCAsmParser.h" 15#include "llvm/MC/MCParser/MCParsedAsmOperand.h" 16#include "llvm/MC/MCContext.h" 17#include "llvm/MC/MCStreamer.h" 18#include "llvm/MC/MCExpr.h" 19#include "llvm/MC/MCInst.h" 20#include "llvm/Target/TargetRegistry.h" 21#include "llvm/Target/TargetAsmParser.h" 22#include "llvm/Support/SourceMgr.h" 23#include "llvm/Support/raw_ostream.h" 24#include "llvm/ADT/SmallVector.h" 25#include "llvm/ADT/StringSwitch.h" 26#include "llvm/ADT/Twine.h" 27using namespace llvm; 28 29// The shift types for register controlled shifts in arm memory addressing 30enum ShiftType { 31 Lsl, 32 Lsr, 33 Asr, 34 Ror, 35 Rrx 36}; 37 38namespace { 39 struct ARMOperand; 40 41class ARMAsmParser : public TargetAsmParser { 42 MCAsmParser &Parser; 43 TargetMachine &TM; 44 45private: 46 MCAsmParser &getParser() const { return Parser; } 47 48 MCAsmLexer &getLexer() const { return Parser.getLexer(); } 49 50 void Warning(SMLoc L, const Twine &Msg) { Parser.Warning(L, Msg); } 51 52 bool Error(SMLoc L, const Twine &Msg) { return Parser.Error(L, Msg); } 53 54 int TryParseRegister(); 55 ARMOperand *TryParseRegisterWithWriteBack(); 56 ARMOperand *ParseRegisterList(); 57 ARMOperand *ParseMemory(); 58 59 bool ParseMemoryOffsetReg(bool &Negative, 60 bool &OffsetRegShifted, 61 enum ShiftType &ShiftType, 62 const MCExpr *&ShiftAmount, 63 const MCExpr *&Offset, 64 bool &OffsetIsReg, 65 int &OffsetRegNum, 66 SMLoc &E); 67 68 bool ParseShift(enum ShiftType &St, const MCExpr *&ShiftAmount, SMLoc &E); 69 70 ARMOperand *ParseOperand(); 71 72 bool ParseDirectiveWord(unsigned Size, SMLoc L); 73 74 bool ParseDirectiveThumb(SMLoc L); 75 76 bool ParseDirectiveThumbFunc(SMLoc L); 77 78 bool ParseDirectiveCode(SMLoc L); 79 80 bool ParseDirectiveSyntax(SMLoc L); 81 82 bool MatchAndEmitInstruction(SMLoc IDLoc, 83 SmallVectorImpl<MCParsedAsmOperand*> &Operands, 84 MCStreamer &Out); 85 86 /// @name Auto-generated Match Functions 87 /// { 88 89#define GET_ASSEMBLER_HEADER 90#include "ARMGenAsmMatcher.inc" 91 92 /// } 93 94 95public: 96 ARMAsmParser(const Target &T, MCAsmParser &_Parser, TargetMachine &_TM) 97 : TargetAsmParser(T), Parser(_Parser), TM(_TM) { 98 // Initialize the set of available features. 99 setAvailableFeatures(ComputeAvailableFeatures( 100 &TM.getSubtarget<ARMSubtarget>())); 101 } 102 103 virtual bool ParseInstruction(StringRef Name, SMLoc NameLoc, 104 SmallVectorImpl<MCParsedAsmOperand*> &Operands); 105 106 virtual bool ParseDirective(AsmToken DirectiveID); 107}; 108} // end anonymous namespace 109 110namespace { 111 112/// ARMOperand - Instances of this class represent a parsed ARM machine 113/// instruction. 114struct ARMOperand : public MCParsedAsmOperand { 115public: 116 enum KindTy { 117 CondCode, 118 Immediate, 119 Memory, 120 Register, 121 Token 122 } Kind; 123 124 SMLoc StartLoc, EndLoc; 125 126 union { 127 struct { 128 ARMCC::CondCodes Val; 129 } CC; 130 131 struct { 132 const char *Data; 133 unsigned Length; 134 } Tok; 135 136 struct { 137 unsigned RegNum; 138 bool Writeback; 139 } Reg; 140 141 struct { 142 const MCExpr *Val; 143 } Imm; 144 145 // This is for all forms of ARM address expressions 146 struct { 147 unsigned BaseRegNum; 148 unsigned OffsetRegNum; // used when OffsetIsReg is true 149 const MCExpr *Offset; // used when OffsetIsReg is false 150 const MCExpr *ShiftAmount; // used when OffsetRegShifted is true 151 enum ShiftType ShiftType; // used when OffsetRegShifted is true 152 unsigned 153 OffsetRegShifted : 1, // only used when OffsetIsReg is true 154 Preindexed : 1, 155 Postindexed : 1, 156 OffsetIsReg : 1, 157 Negative : 1, // only used when OffsetIsReg is true 158 Writeback : 1; 159 } Mem; 160 161 }; 162 163 ARMOperand(const ARMOperand &o) : MCParsedAsmOperand() { 164 Kind = o.Kind; 165 StartLoc = o.StartLoc; 166 EndLoc = o.EndLoc; 167 switch (Kind) { 168 case CondCode: 169 CC = o.CC; 170 break; 171 case Token: 172 Tok = o.Tok; 173 break; 174 case Register: 175 Reg = o.Reg; 176 break; 177 case Immediate: 178 Imm = o.Imm; 179 break; 180 case Memory: 181 Mem = o.Mem; 182 break; 183 } 184 } 185 186 /// getStartLoc - Get the location of the first token of this operand. 187 SMLoc getStartLoc() const { return StartLoc; } 188 /// getEndLoc - Get the location of the last token of this operand. 189 SMLoc getEndLoc() const { return EndLoc; } 190 191 ARMCC::CondCodes getCondCode() const { 192 assert(Kind == CondCode && "Invalid access!"); 193 return CC.Val; 194 } 195 196 StringRef getToken() const { 197 assert(Kind == Token && "Invalid access!"); 198 return StringRef(Tok.Data, Tok.Length); 199 } 200 201 unsigned getReg() const { 202 assert(Kind == Register && "Invalid access!"); 203 return Reg.RegNum; 204 } 205 206 const MCExpr *getImm() const { 207 assert(Kind == Immediate && "Invalid access!"); 208 return Imm.Val; 209 } 210 211 bool isCondCode() const { return Kind == CondCode; } 212 bool isImm() const { return Kind == Immediate; } 213 bool isReg() const { return Kind == Register; } 214 bool isToken() const { return Kind == Token; } 215 bool isMemory() const { return Kind == Memory; } 216 217 void addExpr(MCInst &Inst, const MCExpr *Expr) const { 218 // Add as immediates when possible. Null MCExpr = 0. 219 if (Expr == 0) 220 Inst.addOperand(MCOperand::CreateImm(0)); 221 else if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Expr)) 222 Inst.addOperand(MCOperand::CreateImm(CE->getValue())); 223 else 224 Inst.addOperand(MCOperand::CreateExpr(Expr)); 225 } 226 227 void addCondCodeOperands(MCInst &Inst, unsigned N) const { 228 assert(N == 2 && "Invalid number of operands!"); 229 Inst.addOperand(MCOperand::CreateImm(unsigned(getCondCode()))); 230 // FIXME: What belongs here? 231 Inst.addOperand(MCOperand::CreateReg(0)); 232 } 233 234 void addRegOperands(MCInst &Inst, unsigned N) const { 235 assert(N == 1 && "Invalid number of operands!"); 236 Inst.addOperand(MCOperand::CreateReg(getReg())); 237 } 238 239 void addImmOperands(MCInst &Inst, unsigned N) const { 240 assert(N == 1 && "Invalid number of operands!"); 241 addExpr(Inst, getImm()); 242 } 243 244 245 bool isMemMode5() const { 246 if (!isMemory() || Mem.OffsetIsReg || Mem.OffsetRegShifted || 247 Mem.Writeback || Mem.Negative) 248 return false; 249 // If there is an offset expression, make sure it's valid. 250 if (!Mem.Offset) 251 return true; 252 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Mem.Offset); 253 if (!CE) 254 return false; 255 // The offset must be a multiple of 4 in the range 0-1020. 256 int64_t Value = CE->getValue(); 257 return ((Value & 0x3) == 0 && Value <= 1020 && Value >= -1020); 258 } 259 260 void addMemMode5Operands(MCInst &Inst, unsigned N) const { 261 assert(N == 2 && isMemMode5() && "Invalid number of operands!"); 262 263 Inst.addOperand(MCOperand::CreateReg(Mem.BaseRegNum)); 264 assert(!Mem.OffsetIsReg && "invalid mode 5 operand"); 265 266 // FIXME: #-0 is encoded differently than #0. Does the parser preserve 267 // the difference? 268 if (Mem.Offset) { 269 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Mem.Offset); 270 assert(CE && "Non-constant mode 5 offset operand!"); 271 272 // The MCInst offset operand doesn't include the low two bits (like 273 // the instruction encoding). 274 int64_t Offset = CE->getValue() / 4; 275 if (Offset >= 0) 276 Inst.addOperand(MCOperand::CreateImm(ARM_AM::getAM5Opc(ARM_AM::add, 277 Offset))); 278 else 279 Inst.addOperand(MCOperand::CreateImm(ARM_AM::getAM5Opc(ARM_AM::sub, 280 -Offset))); 281 } else { 282 Inst.addOperand(MCOperand::CreateImm(0)); 283 } 284 } 285 286 virtual void dump(raw_ostream &OS) const; 287 288 static ARMOperand *CreateCondCode(ARMCC::CondCodes CC, SMLoc S) { 289 ARMOperand *Op = new ARMOperand(CondCode); 290 Op->CC.Val = CC; 291 Op->StartLoc = S; 292 Op->EndLoc = S; 293 return Op; 294 } 295 296 static ARMOperand *CreateToken(StringRef Str, SMLoc S) { 297 ARMOperand *Op = new ARMOperand(Token); 298 Op->Tok.Data = Str.data(); 299 Op->Tok.Length = Str.size(); 300 Op->StartLoc = S; 301 Op->EndLoc = S; 302 return Op; 303 } 304 305 static ARMOperand *CreateReg(unsigned RegNum, bool Writeback, SMLoc S, 306 SMLoc E) { 307 ARMOperand *Op = new ARMOperand(Register); 308 Op->Reg.RegNum = RegNum; 309 Op->Reg.Writeback = Writeback; 310 Op->StartLoc = S; 311 Op->EndLoc = E; 312 return Op; 313 } 314 315 static ARMOperand *CreateImm(const MCExpr *Val, SMLoc S, SMLoc E) { 316 ARMOperand *Op = new ARMOperand(Immediate); 317 Op->Imm.Val = Val; 318 Op->StartLoc = S; 319 Op->EndLoc = E; 320 return Op; 321 } 322 323 static ARMOperand *CreateMem(unsigned BaseRegNum, bool OffsetIsReg, 324 const MCExpr *Offset, unsigned OffsetRegNum, 325 bool OffsetRegShifted, enum ShiftType ShiftType, 326 const MCExpr *ShiftAmount, bool Preindexed, 327 bool Postindexed, bool Negative, bool Writeback, 328 SMLoc S, SMLoc E) { 329 ARMOperand *Op = new ARMOperand(Memory); 330 Op->Mem.BaseRegNum = BaseRegNum; 331 Op->Mem.OffsetIsReg = OffsetIsReg; 332 Op->Mem.Offset = Offset; 333 Op->Mem.OffsetRegNum = OffsetRegNum; 334 Op->Mem.OffsetRegShifted = OffsetRegShifted; 335 Op->Mem.ShiftType = ShiftType; 336 Op->Mem.ShiftAmount = ShiftAmount; 337 Op->Mem.Preindexed = Preindexed; 338 Op->Mem.Postindexed = Postindexed; 339 Op->Mem.Negative = Negative; 340 Op->Mem.Writeback = Writeback; 341 342 Op->StartLoc = S; 343 Op->EndLoc = E; 344 return Op; 345 } 346 347private: 348 ARMOperand(KindTy K) : Kind(K) {} 349}; 350 351} // end anonymous namespace. 352 353void ARMOperand::dump(raw_ostream &OS) const { 354 switch (Kind) { 355 case CondCode: 356 OS << ARMCondCodeToString(getCondCode()); 357 break; 358 case Immediate: 359 getImm()->print(OS); 360 break; 361 case Memory: 362 OS << "<memory>"; 363 break; 364 case Register: 365 OS << "<register " << getReg() << ">"; 366 break; 367 case Token: 368 OS << "'" << getToken() << "'"; 369 break; 370 } 371} 372 373/// @name Auto-generated Match Functions 374/// { 375 376static unsigned MatchRegisterName(StringRef Name); 377 378/// } 379 380/// Try to parse a register name. The token must be an Identifier when called, 381/// and if it is a register name the token is eaten and the register number is 382/// returned. Otherwise return -1. 383/// 384int ARMAsmParser::TryParseRegister() { 385 const AsmToken &Tok = Parser.getTok(); 386 assert(Tok.is(AsmToken::Identifier) && "Token is not an Identifier"); 387 388 // FIXME: Validate register for the current architecture; we have to do 389 // validation later, so maybe there is no need for this here. 390 int RegNum = MatchRegisterName(Tok.getString()); 391 if (RegNum == -1) 392 return -1; 393 Parser.Lex(); // Eat identifier token. 394 return RegNum; 395} 396 397 398/// Try to parse a register name. The token must be an Identifier when called, 399/// and if it is a register name the token is eaten and the register number is 400/// returned. Otherwise return -1. 401/// 402/// TODO this is likely to change to allow different register types and or to 403/// parse for a specific register type. 404ARMOperand *ARMAsmParser::TryParseRegisterWithWriteBack() { 405 SMLoc S = Parser.getTok().getLoc(); 406 int RegNo = TryParseRegister(); 407 if (RegNo == -1) return 0; 408 409 SMLoc E = Parser.getTok().getLoc(); 410 411 bool Writeback = false; 412 const AsmToken &ExclaimTok = Parser.getTok(); 413 if (ExclaimTok.is(AsmToken::Exclaim)) { 414 E = ExclaimTok.getLoc(); 415 Writeback = true; 416 Parser.Lex(); // Eat exclaim token 417 } 418 419 return ARMOperand::CreateReg(RegNo, Writeback, S, E); 420} 421 422/// Parse a register list, return it if successful else return null. The first 423/// token must be a '{' when called. 424ARMOperand *ARMAsmParser::ParseRegisterList() { 425 SMLoc S, E; 426 assert(Parser.getTok().is(AsmToken::LCurly) && 427 "Token is not an Left Curly Brace"); 428 S = Parser.getTok().getLoc(); 429 Parser.Lex(); // Eat left curly brace token. 430 431 const AsmToken &RegTok = Parser.getTok(); 432 SMLoc RegLoc = RegTok.getLoc(); 433 if (RegTok.isNot(AsmToken::Identifier)) { 434 Error(RegLoc, "register expected"); 435 return 0; 436 } 437 int RegNum = TryParseRegister(); 438 if (RegNum == -1) { 439 Error(RegLoc, "register expected"); 440 return 0; 441 } 442 443 unsigned RegList = 1 << RegNum; 444 445 int HighRegNum = RegNum; 446 // TODO ranges like "{Rn-Rm}" 447 while (Parser.getTok().is(AsmToken::Comma)) { 448 Parser.Lex(); // Eat comma token. 449 450 const AsmToken &RegTok = Parser.getTok(); 451 SMLoc RegLoc = RegTok.getLoc(); 452 if (RegTok.isNot(AsmToken::Identifier)) { 453 Error(RegLoc, "register expected"); 454 return 0; 455 } 456 int RegNum = TryParseRegister(); 457 if (RegNum == -1) { 458 Error(RegLoc, "register expected"); 459 return 0; 460 } 461 462 if (RegList & (1 << RegNum)) 463 Warning(RegLoc, "register duplicated in register list"); 464 else if (RegNum <= HighRegNum) 465 Warning(RegLoc, "register not in ascending order in register list"); 466 RegList |= 1 << RegNum; 467 HighRegNum = RegNum; 468 } 469 const AsmToken &RCurlyTok = Parser.getTok(); 470 if (RCurlyTok.isNot(AsmToken::RCurly)) { 471 Error(RCurlyTok.getLoc(), "'}' expected"); 472 return 0; 473 } 474 E = RCurlyTok.getLoc(); 475 Parser.Lex(); // Eat left curly brace token. 476 477 // FIXME: Need to return an operand! 478 Error(E, "FIXME: register list parsing not implemented"); 479 return 0; 480} 481 482/// Parse an arm memory expression, return false if successful else return true 483/// or an error. The first token must be a '[' when called. 484/// TODO Only preindexing and postindexing addressing are started, unindexed 485/// with option, etc are still to do. 486ARMOperand *ARMAsmParser::ParseMemory() { 487 SMLoc S, E; 488 assert(Parser.getTok().is(AsmToken::LBrac) && 489 "Token is not an Left Bracket"); 490 S = Parser.getTok().getLoc(); 491 Parser.Lex(); // Eat left bracket token. 492 493 const AsmToken &BaseRegTok = Parser.getTok(); 494 if (BaseRegTok.isNot(AsmToken::Identifier)) { 495 Error(BaseRegTok.getLoc(), "register expected"); 496 return 0; 497 } 498 int BaseRegNum = TryParseRegister(); 499 if (BaseRegNum == -1) { 500 Error(BaseRegTok.getLoc(), "register expected"); 501 return 0; 502 } 503 504 bool Preindexed = false; 505 bool Postindexed = false; 506 bool OffsetIsReg = false; 507 bool Negative = false; 508 bool Writeback = false; 509 510 // First look for preindexed address forms, that is after the "[Rn" we now 511 // have to see if the next token is a comma. 512 const AsmToken &Tok = Parser.getTok(); 513 if (Tok.is(AsmToken::Comma)) { 514 Preindexed = true; 515 Parser.Lex(); // Eat comma token. 516 int OffsetRegNum; 517 bool OffsetRegShifted; 518 enum ShiftType ShiftType; 519 const MCExpr *ShiftAmount; 520 const MCExpr *Offset; 521 if (ParseMemoryOffsetReg(Negative, OffsetRegShifted, ShiftType, ShiftAmount, 522 Offset, OffsetIsReg, OffsetRegNum, E)) 523 return 0; 524 const AsmToken &RBracTok = Parser.getTok(); 525 if (RBracTok.isNot(AsmToken::RBrac)) { 526 Error(RBracTok.getLoc(), "']' expected"); 527 return 0; 528 } 529 E = RBracTok.getLoc(); 530 Parser.Lex(); // Eat right bracket token. 531 532 const AsmToken &ExclaimTok = Parser.getTok(); 533 if (ExclaimTok.is(AsmToken::Exclaim)) { 534 E = ExclaimTok.getLoc(); 535 Writeback = true; 536 Parser.Lex(); // Eat exclaim token 537 } 538 return ARMOperand::CreateMem(BaseRegNum, OffsetIsReg, Offset, OffsetRegNum, 539 OffsetRegShifted, ShiftType, ShiftAmount, 540 Preindexed, Postindexed, Negative, Writeback, 541 S, E); 542 } 543 // The "[Rn" we have so far was not followed by a comma. 544 else if (Tok.is(AsmToken::RBrac)) { 545 // If there's anything other than the right brace, this is a post indexing 546 // addressing form. 547 E = Tok.getLoc(); 548 Parser.Lex(); // Eat right bracket token. 549 550 int OffsetRegNum = 0; 551 bool OffsetRegShifted = false; 552 enum ShiftType ShiftType; 553 const MCExpr *ShiftAmount; 554 const MCExpr *Offset = 0; 555 556 const AsmToken &NextTok = Parser.getTok(); 557 if (NextTok.isNot(AsmToken::EndOfStatement)) { 558 Postindexed = true; 559 Writeback = true; 560 if (NextTok.isNot(AsmToken::Comma)) { 561 Error(NextTok.getLoc(), "',' expected"); 562 return 0; 563 } 564 Parser.Lex(); // Eat comma token. 565 if (ParseMemoryOffsetReg(Negative, OffsetRegShifted, ShiftType, 566 ShiftAmount, Offset, OffsetIsReg, OffsetRegNum, 567 E)) 568 return 0; 569 } 570 571 return ARMOperand::CreateMem(BaseRegNum, OffsetIsReg, Offset, OffsetRegNum, 572 OffsetRegShifted, ShiftType, ShiftAmount, 573 Preindexed, Postindexed, Negative, Writeback, 574 S, E); 575 } 576 577 return 0; 578} 579 580/// Parse the offset of a memory operand after we have seen "[Rn," or "[Rn]," 581/// we will parse the following (were +/- means that a plus or minus is 582/// optional): 583/// +/-Rm 584/// +/-Rm, shift 585/// #offset 586/// we return false on success or an error otherwise. 587bool ARMAsmParser::ParseMemoryOffsetReg(bool &Negative, 588 bool &OffsetRegShifted, 589 enum ShiftType &ShiftType, 590 const MCExpr *&ShiftAmount, 591 const MCExpr *&Offset, 592 bool &OffsetIsReg, 593 int &OffsetRegNum, 594 SMLoc &E) { 595 Negative = false; 596 OffsetRegShifted = false; 597 OffsetIsReg = false; 598 OffsetRegNum = -1; 599 const AsmToken &NextTok = Parser.getTok(); 600 E = NextTok.getLoc(); 601 if (NextTok.is(AsmToken::Plus)) 602 Parser.Lex(); // Eat plus token. 603 else if (NextTok.is(AsmToken::Minus)) { 604 Negative = true; 605 Parser.Lex(); // Eat minus token 606 } 607 // See if there is a register following the "[Rn," or "[Rn]," we have so far. 608 const AsmToken &OffsetRegTok = Parser.getTok(); 609 if (OffsetRegTok.is(AsmToken::Identifier)) { 610 SMLoc CurLoc = OffsetRegTok.getLoc(); 611 OffsetRegNum = TryParseRegister(); 612 if (OffsetRegNum != -1) { 613 OffsetIsReg = true; 614 E = CurLoc; 615 } 616 } 617 618 // If we parsed a register as the offset then their can be a shift after that 619 if (OffsetRegNum != -1) { 620 // Look for a comma then a shift 621 const AsmToken &Tok = Parser.getTok(); 622 if (Tok.is(AsmToken::Comma)) { 623 Parser.Lex(); // Eat comma token. 624 625 const AsmToken &Tok = Parser.getTok(); 626 if (ParseShift(ShiftType, ShiftAmount, E)) 627 return Error(Tok.getLoc(), "shift expected"); 628 OffsetRegShifted = true; 629 } 630 } 631 else { // the "[Rn," or "[Rn,]" we have so far was not followed by "Rm" 632 // Look for #offset following the "[Rn," or "[Rn]," 633 const AsmToken &HashTok = Parser.getTok(); 634 if (HashTok.isNot(AsmToken::Hash)) 635 return Error(HashTok.getLoc(), "'#' expected"); 636 637 Parser.Lex(); // Eat hash token. 638 639 if (getParser().ParseExpression(Offset)) 640 return true; 641 E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1); 642 } 643 return false; 644} 645 646/// ParseShift as one of these two: 647/// ( lsl | lsr | asr | ror ) , # shift_amount 648/// rrx 649/// and returns true if it parses a shift otherwise it returns false. 650bool ARMAsmParser::ParseShift(ShiftType &St, const MCExpr *&ShiftAmount, 651 SMLoc &E) { 652 const AsmToken &Tok = Parser.getTok(); 653 if (Tok.isNot(AsmToken::Identifier)) 654 return true; 655 StringRef ShiftName = Tok.getString(); 656 if (ShiftName == "lsl" || ShiftName == "LSL") 657 St = Lsl; 658 else if (ShiftName == "lsr" || ShiftName == "LSR") 659 St = Lsr; 660 else if (ShiftName == "asr" || ShiftName == "ASR") 661 St = Asr; 662 else if (ShiftName == "ror" || ShiftName == "ROR") 663 St = Ror; 664 else if (ShiftName == "rrx" || ShiftName == "RRX") 665 St = Rrx; 666 else 667 return true; 668 Parser.Lex(); // Eat shift type token. 669 670 // Rrx stands alone. 671 if (St == Rrx) 672 return false; 673 674 // Otherwise, there must be a '#' and a shift amount. 675 const AsmToken &HashTok = Parser.getTok(); 676 if (HashTok.isNot(AsmToken::Hash)) 677 return Error(HashTok.getLoc(), "'#' expected"); 678 Parser.Lex(); // Eat hash token. 679 680 if (getParser().ParseExpression(ShiftAmount)) 681 return true; 682 683 return false; 684} 685 686/// Parse a arm instruction operand. For now this parses the operand regardless 687/// of the mnemonic. 688ARMOperand *ARMAsmParser::ParseOperand() { 689 SMLoc S, E; 690 691 switch (getLexer().getKind()) { 692 case AsmToken::Identifier: 693 if (ARMOperand *Op = TryParseRegisterWithWriteBack()) 694 return Op; 695 696 // This was not a register so parse other operands that start with an 697 // identifier (like labels) as expressions and create them as immediates. 698 const MCExpr *IdVal; 699 S = Parser.getTok().getLoc(); 700 if (getParser().ParseExpression(IdVal)) 701 return 0; 702 E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1); 703 return ARMOperand::CreateImm(IdVal, S, E); 704 case AsmToken::LBrac: 705 return ParseMemory(); 706 case AsmToken::LCurly: 707 return ParseRegisterList(); 708 case AsmToken::Hash: 709 // #42 -> immediate. 710 // TODO: ":lower16:" and ":upper16:" modifiers after # before immediate 711 S = Parser.getTok().getLoc(); 712 Parser.Lex(); 713 const MCExpr *ImmVal; 714 if (getParser().ParseExpression(ImmVal)) 715 return 0; 716 E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1); 717 return ARMOperand::CreateImm(ImmVal, S, E); 718 default: 719 Error(Parser.getTok().getLoc(), "unexpected token in operand"); 720 return 0; 721 } 722} 723 724/// Parse an arm instruction mnemonic followed by its operands. 725bool ARMAsmParser::ParseInstruction(StringRef Name, SMLoc NameLoc, 726 SmallVectorImpl<MCParsedAsmOperand*> &Operands) { 727 // Create the leading tokens for the mnemonic, split by '.' characters. 728 size_t Start = 0, Next = Name.find('.'); 729 StringRef Head = Name.slice(Start, Next); 730 731 // Determine the predicate, if any. 732 // 733 // FIXME: We need a way to check whether a prefix supports predication, 734 // otherwise we will end up with an ambiguity for instructions that happen to 735 // end with a predicate name. 736 // FIXME: Likewise, some arithmetic instructions have an 's' prefix which 737 // indicates to update the condition codes. Those instructions have an 738 // additional immediate operand which encodes the prefix as reg0 or CPSR. 739 // Just checking for a suffix of 's' definitely creates ambiguities; e.g, 740 // the SMMLS instruction. 741 unsigned CC = StringSwitch<unsigned>(Head.substr(Head.size()-2)) 742 .Case("eq", ARMCC::EQ) 743 .Case("ne", ARMCC::NE) 744 .Case("hs", ARMCC::HS) 745 .Case("lo", ARMCC::LO) 746 .Case("mi", ARMCC::MI) 747 .Case("pl", ARMCC::PL) 748 .Case("vs", ARMCC::VS) 749 .Case("vc", ARMCC::VC) 750 .Case("hi", ARMCC::HI) 751 .Case("ls", ARMCC::LS) 752 .Case("ge", ARMCC::GE) 753 .Case("lt", ARMCC::LT) 754 .Case("gt", ARMCC::GT) 755 .Case("le", ARMCC::LE) 756 .Case("al", ARMCC::AL) 757 .Default(~0U); 758 759 if (CC == ~0U || 760 (CC == ARMCC::LS && (Head == "vmls" || Head == "vnmls"))) { 761 CC = ARMCC::AL; 762 } else { 763 Head = Head.slice(0, Head.size() - 2); 764 } 765 766 Operands.push_back(ARMOperand::CreateToken(Head, NameLoc)); 767 // FIXME: Should only add this operand for predicated instructions 768 Operands.push_back(ARMOperand::CreateCondCode(ARMCC::CondCodes(CC), NameLoc)); 769 770 // Add the remaining tokens in the mnemonic. 771 while (Next != StringRef::npos) { 772 Start = Next; 773 Next = Name.find('.', Start + 1); 774 Head = Name.slice(Start, Next); 775 776 Operands.push_back(ARMOperand::CreateToken(Head, NameLoc)); 777 } 778 779 // Read the remaining operands. 780 if (getLexer().isNot(AsmToken::EndOfStatement)) { 781 // Read the first operand. 782 if (ARMOperand *Op = ParseOperand()) 783 Operands.push_back(Op); 784 else { 785 Parser.EatToEndOfStatement(); 786 return true; 787 } 788 789 while (getLexer().is(AsmToken::Comma)) { 790 Parser.Lex(); // Eat the comma. 791 792 // Parse and remember the operand. 793 if (ARMOperand *Op = ParseOperand()) 794 Operands.push_back(Op); 795 else { 796 Parser.EatToEndOfStatement(); 797 return true; 798 } 799 } 800 } 801 802 if (getLexer().isNot(AsmToken::EndOfStatement)) { 803 Parser.EatToEndOfStatement(); 804 return TokError("unexpected token in argument list"); 805 } 806 Parser.Lex(); // Consume the EndOfStatement 807 return false; 808} 809 810bool ARMAsmParser:: 811MatchAndEmitInstruction(SMLoc IDLoc, 812 SmallVectorImpl<MCParsedAsmOperand*> &Operands, 813 MCStreamer &Out) { 814 MCInst Inst; 815 unsigned ErrorInfo; 816 switch (MatchInstructionImpl(Operands, Inst, ErrorInfo)) { 817 case Match_Success: 818 Out.EmitInstruction(Inst); 819 return false; 820 821 case Match_MissingFeature: 822 Error(IDLoc, "instruction requires a CPU feature not currently enabled"); 823 return true; 824 case Match_InvalidOperand: { 825 SMLoc ErrorLoc = IDLoc; 826 if (ErrorInfo != ~0U) { 827 if (ErrorInfo >= Operands.size()) 828 return Error(IDLoc, "too few operands for instruction"); 829 830 ErrorLoc = ((ARMOperand*)Operands[ErrorInfo])->getStartLoc(); 831 if (ErrorLoc == SMLoc()) ErrorLoc = IDLoc; 832 } 833 834 return Error(ErrorLoc, "invalid operand for instruction"); 835 } 836 case Match_MnemonicFail: 837 return Error(IDLoc, "unrecognized instruction mnemonic"); 838 } 839 840 llvm_unreachable("Implement any new match types added!"); 841} 842 843 844 845/// ParseDirective parses the arm specific directives 846bool ARMAsmParser::ParseDirective(AsmToken DirectiveID) { 847 StringRef IDVal = DirectiveID.getIdentifier(); 848 if (IDVal == ".word") 849 return ParseDirectiveWord(4, DirectiveID.getLoc()); 850 else if (IDVal == ".thumb") 851 return ParseDirectiveThumb(DirectiveID.getLoc()); 852 else if (IDVal == ".thumb_func") 853 return ParseDirectiveThumbFunc(DirectiveID.getLoc()); 854 else if (IDVal == ".code") 855 return ParseDirectiveCode(DirectiveID.getLoc()); 856 else if (IDVal == ".syntax") 857 return ParseDirectiveSyntax(DirectiveID.getLoc()); 858 return true; 859} 860 861/// ParseDirectiveWord 862/// ::= .word [ expression (, expression)* ] 863bool ARMAsmParser::ParseDirectiveWord(unsigned Size, SMLoc L) { 864 if (getLexer().isNot(AsmToken::EndOfStatement)) { 865 for (;;) { 866 const MCExpr *Value; 867 if (getParser().ParseExpression(Value)) 868 return true; 869 870 getParser().getStreamer().EmitValue(Value, Size, 0/*addrspace*/); 871 872 if (getLexer().is(AsmToken::EndOfStatement)) 873 break; 874 875 // FIXME: Improve diagnostic. 876 if (getLexer().isNot(AsmToken::Comma)) 877 return Error(L, "unexpected token in directive"); 878 Parser.Lex(); 879 } 880 } 881 882 Parser.Lex(); 883 return false; 884} 885 886/// ParseDirectiveThumb 887/// ::= .thumb 888bool ARMAsmParser::ParseDirectiveThumb(SMLoc L) { 889 if (getLexer().isNot(AsmToken::EndOfStatement)) 890 return Error(L, "unexpected token in directive"); 891 Parser.Lex(); 892 893 // TODO: set thumb mode 894 // TODO: tell the MC streamer the mode 895 // getParser().getStreamer().Emit???(); 896 return false; 897} 898 899/// ParseDirectiveThumbFunc 900/// ::= .thumbfunc symbol_name 901bool ARMAsmParser::ParseDirectiveThumbFunc(SMLoc L) { 902 const AsmToken &Tok = Parser.getTok(); 903 if (Tok.isNot(AsmToken::Identifier) && Tok.isNot(AsmToken::String)) 904 return Error(L, "unexpected token in .thumb_func directive"); 905 StringRef Name = Tok.getString(); 906 Parser.Lex(); // Consume the identifier token. 907 if (getLexer().isNot(AsmToken::EndOfStatement)) 908 return Error(L, "unexpected token in directive"); 909 Parser.Lex(); 910 911 // Mark symbol as a thumb symbol. 912 MCSymbol *Func = getParser().getContext().GetOrCreateSymbol(Name); 913 getParser().getStreamer().EmitThumbFunc(Func); 914 return false; 915} 916 917/// ParseDirectiveSyntax 918/// ::= .syntax unified | divided 919bool ARMAsmParser::ParseDirectiveSyntax(SMLoc L) { 920 const AsmToken &Tok = Parser.getTok(); 921 if (Tok.isNot(AsmToken::Identifier)) 922 return Error(L, "unexpected token in .syntax directive"); 923 StringRef Mode = Tok.getString(); 924 if (Mode == "unified" || Mode == "UNIFIED") 925 Parser.Lex(); 926 else if (Mode == "divided" || Mode == "DIVIDED") 927 Parser.Lex(); 928 else 929 return Error(L, "unrecognized syntax mode in .syntax directive"); 930 931 if (getLexer().isNot(AsmToken::EndOfStatement)) 932 return Error(Parser.getTok().getLoc(), "unexpected token in directive"); 933 Parser.Lex(); 934 935 // TODO tell the MC streamer the mode 936 // getParser().getStreamer().Emit???(); 937 return false; 938} 939 940/// ParseDirectiveCode 941/// ::= .code 16 | 32 942bool ARMAsmParser::ParseDirectiveCode(SMLoc L) { 943 const AsmToken &Tok = Parser.getTok(); 944 if (Tok.isNot(AsmToken::Integer)) 945 return Error(L, "unexpected token in .code directive"); 946 int64_t Val = Parser.getTok().getIntVal(); 947 if (Val == 16) 948 Parser.Lex(); 949 else if (Val == 32) 950 Parser.Lex(); 951 else 952 return Error(L, "invalid operand to .code directive"); 953 954 if (getLexer().isNot(AsmToken::EndOfStatement)) 955 return Error(Parser.getTok().getLoc(), "unexpected token in directive"); 956 Parser.Lex(); 957 958 if (Val == 16) 959 getParser().getStreamer().EmitAssemblerFlag(MCAF_Code16); 960 else 961 getParser().getStreamer().EmitAssemblerFlag(MCAF_Code32); 962 963 return false; 964} 965 966extern "C" void LLVMInitializeARMAsmLexer(); 967 968/// Force static initialization. 969extern "C" void LLVMInitializeARMAsmParser() { 970 RegisterAsmParser<ARMAsmParser> X(TheARMTarget); 971 RegisterAsmParser<ARMAsmParser> Y(TheThumbTarget); 972 LLVMInitializeARMAsmLexer(); 973} 974 975#define GET_REGISTER_MATCHER 976#define GET_MATCHER_IMPLEMENTATION 977#include "ARMGenAsmMatcher.inc" 978