ARMAsmParser.cpp revision e4e5e2aae7e1e0e84877061432e7b981a360a77d
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 "ARMMCExpr.h" 13#include "ARMBaseRegisterInfo.h" 14#include "ARMSubtarget.h" 15#include "llvm/MC/MCParser/MCAsmLexer.h" 16#include "llvm/MC/MCParser/MCAsmParser.h" 17#include "llvm/MC/MCParser/MCParsedAsmOperand.h" 18#include "llvm/MC/MCContext.h" 19#include "llvm/MC/MCStreamer.h" 20#include "llvm/MC/MCExpr.h" 21#include "llvm/MC/MCInst.h" 22#include "llvm/Target/TargetRegistry.h" 23#include "llvm/Target/TargetAsmParser.h" 24#include "llvm/Support/SourceMgr.h" 25#include "llvm/Support/raw_ostream.h" 26#include "llvm/ADT/SmallVector.h" 27#include "llvm/ADT/StringSwitch.h" 28#include "llvm/ADT/Twine.h" 29using namespace llvm; 30 31/// Shift types used for register controlled shifts in ARM memory addressing. 32enum ShiftType { 33 Lsl, 34 Lsr, 35 Asr, 36 Ror, 37 Rrx 38}; 39 40namespace { 41 42class ARMOperand; 43 44class ARMAsmParser : public TargetAsmParser { 45 MCAsmParser &Parser; 46 TargetMachine &TM; 47 48 MCAsmParser &getParser() const { return Parser; } 49 MCAsmLexer &getLexer() const { return Parser.getLexer(); } 50 51 void Warning(SMLoc L, const Twine &Msg) { Parser.Warning(L, Msg); } 52 bool Error(SMLoc L, const Twine &Msg) { return Parser.Error(L, Msg); } 53 54 int TryParseRegister(); 55 bool TryParseMCRName(SmallVectorImpl<MCParsedAsmOperand*>&); 56 bool TryParseRegisterWithWriteBack(SmallVectorImpl<MCParsedAsmOperand*> &); 57 bool ParseRegisterList(SmallVectorImpl<MCParsedAsmOperand*> &); 58 bool ParseMemory(SmallVectorImpl<MCParsedAsmOperand*> &); 59 bool ParseOperand(SmallVectorImpl<MCParsedAsmOperand*> &, bool isMCR); 60 bool ParsePrefix(ARMMCExpr::VariantKind &RefKind); 61 const MCExpr *ApplyPrefixToExpr(const MCExpr *E, 62 MCSymbolRefExpr::VariantKind Variant); 63 64 65 bool ParseMemoryOffsetReg(bool &Negative, 66 bool &OffsetRegShifted, 67 enum ShiftType &ShiftType, 68 const MCExpr *&ShiftAmount, 69 const MCExpr *&Offset, 70 bool &OffsetIsReg, 71 int &OffsetRegNum, 72 SMLoc &E); 73 bool ParseShift(enum ShiftType &St, const MCExpr *&ShiftAmount, SMLoc &E); 74 bool ParseDirectiveWord(unsigned Size, SMLoc L); 75 bool ParseDirectiveThumb(SMLoc L); 76 bool ParseDirectiveThumbFunc(SMLoc L); 77 bool ParseDirectiveCode(SMLoc L); 78 bool ParseDirectiveSyntax(SMLoc L); 79 80 bool MatchAndEmitInstruction(SMLoc IDLoc, 81 SmallVectorImpl<MCParsedAsmOperand*> &Operands, 82 MCStreamer &Out); 83 84 /// @name Auto-generated Match Functions 85 /// { 86 87#define GET_ASSEMBLER_HEADER 88#include "ARMGenAsmMatcher.inc" 89 90 /// } 91 92public: 93 ARMAsmParser(const Target &T, MCAsmParser &_Parser, TargetMachine &_TM) 94 : TargetAsmParser(T), Parser(_Parser), TM(_TM) { 95 // Initialize the set of available features. 96 setAvailableFeatures(ComputeAvailableFeatures( 97 &TM.getSubtarget<ARMSubtarget>())); 98 } 99 100 virtual bool ParseInstruction(StringRef Name, SMLoc NameLoc, 101 SmallVectorImpl<MCParsedAsmOperand*> &Operands); 102 virtual bool ParseDirective(AsmToken DirectiveID); 103}; 104} // end anonymous namespace 105 106namespace { 107 108/// ARMOperand - Instances of this class represent a parsed ARM machine 109/// instruction. 110class ARMOperand : public MCParsedAsmOperand { 111 enum KindTy { 112 CondCode, 113 CCOut, 114 Immediate, 115 Memory, 116 Register, 117 RegisterList, 118 DPRRegisterList, 119 SPRRegisterList, 120 Token 121 } Kind; 122 123 SMLoc StartLoc, EndLoc; 124 SmallVector<unsigned, 8> Registers; 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 } Reg; 139 140 struct { 141 const MCExpr *Val; 142 } Imm; 143 144 /// Combined record for all forms of ARM address expressions. 145 struct { 146 unsigned BaseRegNum; 147 unsigned OffsetRegNum; // used when OffsetIsReg is true 148 const MCExpr *Offset; // used when OffsetIsReg is false 149 const MCExpr *ShiftAmount; // used when OffsetRegShifted is true 150 enum ShiftType ShiftType; // used when OffsetRegShifted is true 151 unsigned OffsetRegShifted : 1; // only used when OffsetIsReg is true 152 unsigned Preindexed : 1; 153 unsigned Postindexed : 1; 154 unsigned OffsetIsReg : 1; 155 unsigned Negative : 1; // only used when OffsetIsReg is true 156 unsigned Writeback : 1; 157 } Mem; 158 }; 159 160 ARMOperand(KindTy K) : MCParsedAsmOperand(), Kind(K) {} 161public: 162 ARMOperand(const ARMOperand &o) : MCParsedAsmOperand() { 163 Kind = o.Kind; 164 StartLoc = o.StartLoc; 165 EndLoc = o.EndLoc; 166 switch (Kind) { 167 case CondCode: 168 CC = o.CC; 169 break; 170 case Token: 171 Tok = o.Tok; 172 break; 173 case CCOut: 174 case Register: 175 Reg = o.Reg; 176 break; 177 case RegisterList: 178 case DPRRegisterList: 179 case SPRRegisterList: 180 Registers = o.Registers; 181 break; 182 case Immediate: 183 Imm = o.Imm; 184 break; 185 case Memory: 186 Mem = o.Mem; 187 break; 188 } 189 } 190 191 /// getStartLoc - Get the location of the first token of this operand. 192 SMLoc getStartLoc() const { return StartLoc; } 193 /// getEndLoc - Get the location of the last token of this operand. 194 SMLoc getEndLoc() const { return EndLoc; } 195 196 ARMCC::CondCodes getCondCode() const { 197 assert(Kind == CondCode && "Invalid access!"); 198 return CC.Val; 199 } 200 201 StringRef getToken() const { 202 assert(Kind == Token && "Invalid access!"); 203 return StringRef(Tok.Data, Tok.Length); 204 } 205 206 unsigned getReg() const { 207 assert((Kind == Register || Kind == CCOut) && "Invalid access!"); 208 return Reg.RegNum; 209 } 210 211 const SmallVectorImpl<unsigned> &getRegList() const { 212 assert((Kind == RegisterList || Kind == DPRRegisterList || 213 Kind == SPRRegisterList) && "Invalid access!"); 214 return Registers; 215 } 216 217 const MCExpr *getImm() const { 218 assert(Kind == Immediate && "Invalid access!"); 219 return Imm.Val; 220 } 221 222 bool isCondCode() const { return Kind == CondCode; } 223 bool isCCOut() const { return Kind == CCOut; } 224 bool isImm() const { return Kind == Immediate; } 225 bool isReg() const { return Kind == Register; } 226 bool isRegList() const { return Kind == RegisterList; } 227 bool isDPRRegList() const { return Kind == DPRRegisterList; } 228 bool isSPRRegList() const { return Kind == SPRRegisterList; } 229 bool isToken() const { return Kind == Token; } 230 bool isMemory() const { return Kind == Memory; } 231 bool isMemMode5() const { 232 if (!isMemory() || Mem.OffsetIsReg || Mem.OffsetRegShifted || 233 Mem.Writeback || Mem.Negative) 234 return false; 235 236 // If there is an offset expression, make sure it's valid. 237 if (!Mem.Offset) return true; 238 239 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Mem.Offset); 240 if (!CE) return false; 241 242 // The offset must be a multiple of 4 in the range 0-1020. 243 int64_t Value = CE->getValue(); 244 return ((Value & 0x3) == 0 && Value <= 1020 && Value >= -1020); 245 } 246 bool isMemModeRegThumb() const { 247 if (!isMemory() || (!Mem.OffsetIsReg && !Mem.Offset) || Mem.Writeback) 248 return false; 249 return !Mem.Offset || !isa<MCConstantExpr>(Mem.Offset); 250 } 251 bool isMemModeImmThumb() const { 252 if (!isMemory() || (!Mem.OffsetIsReg && !Mem.Offset) || Mem.Writeback) 253 return false; 254 255 if (!Mem.Offset) return false; 256 257 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Mem.Offset); 258 if (!CE) return false; 259 260 // The offset must be a multiple of 4 in the range 0-124. 261 uint64_t Value = CE->getValue(); 262 return ((Value & 0x3) == 0 && Value <= 124); 263 } 264 265 void addExpr(MCInst &Inst, const MCExpr *Expr) const { 266 // Add as immediates when possible. Null MCExpr = 0. 267 if (Expr == 0) 268 Inst.addOperand(MCOperand::CreateImm(0)); 269 else if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Expr)) 270 Inst.addOperand(MCOperand::CreateImm(CE->getValue())); 271 else 272 Inst.addOperand(MCOperand::CreateExpr(Expr)); 273 } 274 275 void addCondCodeOperands(MCInst &Inst, unsigned N) const { 276 assert(N == 2 && "Invalid number of operands!"); 277 Inst.addOperand(MCOperand::CreateImm(unsigned(getCondCode()))); 278 unsigned RegNum = getCondCode() == ARMCC::AL ? 0: ARM::CPSR; 279 Inst.addOperand(MCOperand::CreateReg(RegNum)); 280 } 281 282 void addCCOutOperands(MCInst &Inst, unsigned N) const { 283 assert(N == 1 && "Invalid number of operands!"); 284 Inst.addOperand(MCOperand::CreateReg(getReg())); 285 } 286 287 void addRegOperands(MCInst &Inst, unsigned N) const { 288 assert(N == 1 && "Invalid number of operands!"); 289 Inst.addOperand(MCOperand::CreateReg(getReg())); 290 } 291 292 void addRegListOperands(MCInst &Inst, unsigned N) const { 293 assert(N == 1 && "Invalid number of operands!"); 294 const SmallVectorImpl<unsigned> &RegList = getRegList(); 295 for (SmallVectorImpl<unsigned>::const_iterator 296 I = RegList.begin(), E = RegList.end(); I != E; ++I) 297 Inst.addOperand(MCOperand::CreateReg(*I)); 298 } 299 300 void addDPRRegListOperands(MCInst &Inst, unsigned N) const { 301 addRegListOperands(Inst, N); 302 } 303 304 void addSPRRegListOperands(MCInst &Inst, unsigned N) const { 305 addRegListOperands(Inst, N); 306 } 307 308 void addImmOperands(MCInst &Inst, unsigned N) const { 309 assert(N == 1 && "Invalid number of operands!"); 310 addExpr(Inst, getImm()); 311 } 312 313 void addMemMode5Operands(MCInst &Inst, unsigned N) const { 314 assert(N == 2 && isMemMode5() && "Invalid number of operands!"); 315 316 Inst.addOperand(MCOperand::CreateReg(Mem.BaseRegNum)); 317 assert(!Mem.OffsetIsReg && "Invalid mode 5 operand"); 318 319 // FIXME: #-0 is encoded differently than #0. Does the parser preserve 320 // the difference? 321 if (Mem.Offset) { 322 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Mem.Offset); 323 assert(CE && "Non-constant mode 5 offset operand!"); 324 325 // The MCInst offset operand doesn't include the low two bits (like 326 // the instruction encoding). 327 int64_t Offset = CE->getValue() / 4; 328 if (Offset >= 0) 329 Inst.addOperand(MCOperand::CreateImm(ARM_AM::getAM5Opc(ARM_AM::add, 330 Offset))); 331 else 332 Inst.addOperand(MCOperand::CreateImm(ARM_AM::getAM5Opc(ARM_AM::sub, 333 -Offset))); 334 } else { 335 Inst.addOperand(MCOperand::CreateImm(0)); 336 } 337 } 338 339 void addMemModeRegThumbOperands(MCInst &Inst, unsigned N) const { 340 assert(N == 2 && isMemModeRegThumb() && "Invalid number of operands!"); 341 Inst.addOperand(MCOperand::CreateReg(Mem.BaseRegNum)); 342 Inst.addOperand(MCOperand::CreateReg(Mem.OffsetRegNum)); 343 } 344 345 void addMemModeImmThumbOperands(MCInst &Inst, unsigned N) const { 346 assert(N == 2 && isMemModeImmThumb() && "Invalid number of operands!"); 347 Inst.addOperand(MCOperand::CreateReg(Mem.BaseRegNum)); 348 const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Mem.Offset); 349 assert(CE && "Non-constant mode offset operand!"); 350 Inst.addOperand(MCOperand::CreateImm(CE->getValue())); 351 } 352 353 virtual void dump(raw_ostream &OS) const; 354 355 static ARMOperand *CreateCondCode(ARMCC::CondCodes CC, SMLoc S) { 356 ARMOperand *Op = new ARMOperand(CondCode); 357 Op->CC.Val = CC; 358 Op->StartLoc = S; 359 Op->EndLoc = S; 360 return Op; 361 } 362 363 static ARMOperand *CreateCCOut(unsigned RegNum, SMLoc S) { 364 ARMOperand *Op = new ARMOperand(CCOut); 365 Op->Reg.RegNum = RegNum; 366 Op->StartLoc = S; 367 Op->EndLoc = S; 368 return Op; 369 } 370 371 static ARMOperand *CreateToken(StringRef Str, SMLoc S) { 372 ARMOperand *Op = new ARMOperand(Token); 373 Op->Tok.Data = Str.data(); 374 Op->Tok.Length = Str.size(); 375 Op->StartLoc = S; 376 Op->EndLoc = S; 377 return Op; 378 } 379 380 static ARMOperand *CreateReg(unsigned RegNum, SMLoc S, SMLoc E) { 381 ARMOperand *Op = new ARMOperand(Register); 382 Op->Reg.RegNum = RegNum; 383 Op->StartLoc = S; 384 Op->EndLoc = E; 385 return Op; 386 } 387 388 static ARMOperand * 389 CreateRegList(const SmallVectorImpl<std::pair<unsigned, SMLoc> > &Regs, 390 SMLoc StartLoc, SMLoc EndLoc) { 391 KindTy Kind = RegisterList; 392 393 if (ARM::DPRRegClass.contains(Regs.front().first)) 394 Kind = DPRRegisterList; 395 else if (ARM::SPRRegClass.contains(Regs.front().first)) 396 Kind = SPRRegisterList; 397 398 ARMOperand *Op = new ARMOperand(Kind); 399 for (SmallVectorImpl<std::pair<unsigned, SMLoc> >::const_iterator 400 I = Regs.begin(), E = Regs.end(); I != E; ++I) 401 Op->Registers.push_back(I->first); 402 array_pod_sort(Op->Registers.begin(), Op->Registers.end()); 403 Op->StartLoc = StartLoc; 404 Op->EndLoc = EndLoc; 405 return Op; 406 } 407 408 static ARMOperand *CreateImm(const MCExpr *Val, SMLoc S, SMLoc E) { 409 ARMOperand *Op = new ARMOperand(Immediate); 410 Op->Imm.Val = Val; 411 Op->StartLoc = S; 412 Op->EndLoc = E; 413 return Op; 414 } 415 416 static ARMOperand *CreateMem(unsigned BaseRegNum, bool OffsetIsReg, 417 const MCExpr *Offset, unsigned OffsetRegNum, 418 bool OffsetRegShifted, enum ShiftType ShiftType, 419 const MCExpr *ShiftAmount, bool Preindexed, 420 bool Postindexed, bool Negative, bool Writeback, 421 SMLoc S, SMLoc E) { 422 ARMOperand *Op = new ARMOperand(Memory); 423 Op->Mem.BaseRegNum = BaseRegNum; 424 Op->Mem.OffsetIsReg = OffsetIsReg; 425 Op->Mem.Offset = Offset; 426 Op->Mem.OffsetRegNum = OffsetRegNum; 427 Op->Mem.OffsetRegShifted = OffsetRegShifted; 428 Op->Mem.ShiftType = ShiftType; 429 Op->Mem.ShiftAmount = ShiftAmount; 430 Op->Mem.Preindexed = Preindexed; 431 Op->Mem.Postindexed = Postindexed; 432 Op->Mem.Negative = Negative; 433 Op->Mem.Writeback = Writeback; 434 435 Op->StartLoc = S; 436 Op->EndLoc = E; 437 return Op; 438 } 439}; 440 441} // end anonymous namespace. 442 443void ARMOperand::dump(raw_ostream &OS) const { 444 switch (Kind) { 445 case CondCode: 446 OS << "<ARMCC::" << ARMCondCodeToString(getCondCode()) << ">"; 447 break; 448 case CCOut: 449 OS << "<ccout " << getReg() << ">"; 450 break; 451 case Immediate: 452 getImm()->print(OS); 453 break; 454 case Memory: 455 OS << "<memory>"; 456 break; 457 case Register: 458 OS << "<register " << getReg() << ">"; 459 break; 460 case RegisterList: 461 case DPRRegisterList: 462 case SPRRegisterList: { 463 OS << "<register_list "; 464 465 const SmallVectorImpl<unsigned> &RegList = getRegList(); 466 for (SmallVectorImpl<unsigned>::const_iterator 467 I = RegList.begin(), E = RegList.end(); I != E; ) { 468 OS << *I; 469 if (++I < E) OS << ", "; 470 } 471 472 OS << ">"; 473 break; 474 } 475 case Token: 476 OS << "'" << getToken() << "'"; 477 break; 478 } 479} 480 481/// @name Auto-generated Match Functions 482/// { 483 484static unsigned MatchRegisterName(StringRef Name); 485 486/// } 487 488/// Try to parse a register name. The token must be an Identifier when called, 489/// and if it is a register name the token is eaten and the register number is 490/// returned. Otherwise return -1. 491/// 492int ARMAsmParser::TryParseRegister() { 493 const AsmToken &Tok = Parser.getTok(); 494 assert(Tok.is(AsmToken::Identifier) && "Token is not an Identifier"); 495 496 // FIXME: Validate register for the current architecture; we have to do 497 // validation later, so maybe there is no need for this here. 498 unsigned RegNum = MatchRegisterName(Tok.getString()); 499 if (RegNum == 0) 500 return -1; 501 Parser.Lex(); // Eat identifier token. 502 return RegNum; 503} 504 505 506/// Try to parse a register name. The token must be an Identifier when called. 507/// If it's a register, an AsmOperand is created. Another AsmOperand is created 508/// if there is a "writeback". 'true' if it's not a register. 509/// 510/// TODO this is likely to change to allow different register types and or to 511/// parse for a specific register type. 512bool ARMAsmParser:: 513TryParseRegisterWithWriteBack(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { 514 SMLoc S = Parser.getTok().getLoc(); 515 int RegNo = TryParseRegister(); 516 if (RegNo == -1) 517 return true; 518 519 Operands.push_back(ARMOperand::CreateReg(RegNo, S, Parser.getTok().getLoc())); 520 521 const AsmToken &ExclaimTok = Parser.getTok(); 522 if (ExclaimTok.is(AsmToken::Exclaim)) { 523 Operands.push_back(ARMOperand::CreateToken(ExclaimTok.getString(), 524 ExclaimTok.getLoc())); 525 Parser.Lex(); // Eat exclaim token 526 } 527 528 return false; 529} 530 531static int MatchMCRName(StringRef Name) { 532 // Use the same layout as the tablegen'erated register name matcher. Ugly, 533 // but efficient. 534 switch (Name.size()) { 535 default: break; 536 case 2: 537 if (Name[0] != 'p' && Name[0] != 'c') 538 return -1; 539 switch (Name[1]) { 540 default: return -1; 541 case '0': return 0; 542 case '1': return 1; 543 case '2': return 2; 544 case '3': return 3; 545 case '4': return 4; 546 case '5': return 5; 547 case '6': return 6; 548 case '7': return 7; 549 case '8': return 8; 550 case '9': return 9; 551 } 552 break; 553 case 3: 554 if ((Name[0] != 'p' && Name[0] != 'c') || Name[1] != '1') 555 return -1; 556 switch (Name[2]) { 557 default: return -1; 558 case '0': return 10; 559 case '1': return 11; 560 case '2': return 12; 561 case '3': return 13; 562 case '4': return 14; 563 case '5': return 15; 564 } 565 break; 566 } 567 568 llvm_unreachable("Unhandled coprocessor operand string!"); 569 return -1; 570} 571 572/// TryParseMCRName - Try to parse an MCR/MRC symbolic operand 573/// name. The token must be an Identifier when called, and if it is a MCR 574/// operand name, the token is eaten and the operand is added to the 575/// operand list. 576bool ARMAsmParser:: 577TryParseMCRName(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { 578 SMLoc S = Parser.getTok().getLoc(); 579 const AsmToken &Tok = Parser.getTok(); 580 assert(Tok.is(AsmToken::Identifier) && "Token is not an Identifier"); 581 582 int Num = MatchMCRName(Tok.getString()); 583 if (Num == -1) 584 return true; 585 586 Parser.Lex(); // Eat identifier token. 587 Operands.push_back(ARMOperand::CreateImm( 588 MCConstantExpr::Create(Num, getContext()), S, Parser.getTok().getLoc())); 589 return false; 590} 591 592/// Parse a register list, return it if successful else return null. The first 593/// token must be a '{' when called. 594bool ARMAsmParser:: 595ParseRegisterList(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { 596 assert(Parser.getTok().is(AsmToken::LCurly) && 597 "Token is not a Left Curly Brace"); 598 SMLoc S = Parser.getTok().getLoc(); 599 600 // Read the rest of the registers in the list. 601 unsigned PrevRegNum = 0; 602 SmallVector<std::pair<unsigned, SMLoc>, 32> Registers; 603 604 do { 605 bool IsRange = Parser.getTok().is(AsmToken::Minus); 606 Parser.Lex(); // Eat non-identifier token. 607 608 const AsmToken &RegTok = Parser.getTok(); 609 SMLoc RegLoc = RegTok.getLoc(); 610 if (RegTok.isNot(AsmToken::Identifier)) { 611 Error(RegLoc, "register expected"); 612 return true; 613 } 614 615 int RegNum = TryParseRegister(); 616 if (RegNum == -1) { 617 Error(RegLoc, "register expected"); 618 return true; 619 } 620 621 if (IsRange) { 622 int Reg = PrevRegNum; 623 do { 624 ++Reg; 625 Registers.push_back(std::make_pair(Reg, RegLoc)); 626 } while (Reg != RegNum); 627 } else { 628 Registers.push_back(std::make_pair(RegNum, RegLoc)); 629 } 630 631 PrevRegNum = RegNum; 632 } while (Parser.getTok().is(AsmToken::Comma) || 633 Parser.getTok().is(AsmToken::Minus)); 634 635 // Process the right curly brace of the list. 636 const AsmToken &RCurlyTok = Parser.getTok(); 637 if (RCurlyTok.isNot(AsmToken::RCurly)) { 638 Error(RCurlyTok.getLoc(), "'}' expected"); 639 return true; 640 } 641 642 SMLoc E = RCurlyTok.getLoc(); 643 Parser.Lex(); // Eat right curly brace token. 644 645 // Verify the register list. 646 SmallVectorImpl<std::pair<unsigned, SMLoc> >::const_iterator 647 RI = Registers.begin(), RE = Registers.end(); 648 649 unsigned HighRegNum = getARMRegisterNumbering(RI->first); 650 bool EmittedWarning = false; 651 652 DenseMap<unsigned, bool> RegMap; 653 RegMap[HighRegNum] = true; 654 655 for (++RI; RI != RE; ++RI) { 656 const std::pair<unsigned, SMLoc> &RegInfo = *RI; 657 unsigned Reg = getARMRegisterNumbering(RegInfo.first); 658 659 if (RegMap[Reg]) { 660 Error(RegInfo.second, "register duplicated in register list"); 661 return true; 662 } 663 664 if (!EmittedWarning && Reg < HighRegNum) 665 Warning(RegInfo.second, 666 "register not in ascending order in register list"); 667 668 RegMap[Reg] = true; 669 HighRegNum = std::max(Reg, HighRegNum); 670 } 671 672 Operands.push_back(ARMOperand::CreateRegList(Registers, S, E)); 673 return false; 674} 675 676/// Parse an ARM memory expression, return false if successful else return true 677/// or an error. The first token must be a '[' when called. 678/// 679/// TODO Only preindexing and postindexing addressing are started, unindexed 680/// with option, etc are still to do. 681bool ARMAsmParser:: 682ParseMemory(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { 683 SMLoc S, E; 684 assert(Parser.getTok().is(AsmToken::LBrac) && 685 "Token is not a Left Bracket"); 686 S = Parser.getTok().getLoc(); 687 Parser.Lex(); // Eat left bracket token. 688 689 const AsmToken &BaseRegTok = Parser.getTok(); 690 if (BaseRegTok.isNot(AsmToken::Identifier)) { 691 Error(BaseRegTok.getLoc(), "register expected"); 692 return true; 693 } 694 int BaseRegNum = TryParseRegister(); 695 if (BaseRegNum == -1) { 696 Error(BaseRegTok.getLoc(), "register expected"); 697 return true; 698 } 699 700 bool Preindexed = false; 701 bool Postindexed = false; 702 bool OffsetIsReg = false; 703 bool Negative = false; 704 bool Writeback = false; 705 706 // First look for preindexed address forms, that is after the "[Rn" we now 707 // have to see if the next token is a comma. 708 const AsmToken &Tok = Parser.getTok(); 709 if (Tok.is(AsmToken::Comma)) { 710 Preindexed = true; 711 Parser.Lex(); // Eat comma token. 712 int OffsetRegNum; 713 bool OffsetRegShifted; 714 enum ShiftType ShiftType; 715 const MCExpr *ShiftAmount = 0; 716 const MCExpr *Offset = 0; 717 if (ParseMemoryOffsetReg(Negative, OffsetRegShifted, ShiftType, ShiftAmount, 718 Offset, OffsetIsReg, OffsetRegNum, E)) 719 return true; 720 const AsmToken &RBracTok = Parser.getTok(); 721 if (RBracTok.isNot(AsmToken::RBrac)) { 722 Error(RBracTok.getLoc(), "']' expected"); 723 return true; 724 } 725 E = RBracTok.getLoc(); 726 Parser.Lex(); // Eat right bracket token. 727 728 729 const AsmToken &ExclaimTok = Parser.getTok(); 730 ARMOperand *WBOp = 0; 731 if (ExclaimTok.is(AsmToken::Exclaim)) { 732 WBOp = ARMOperand::CreateToken(ExclaimTok.getString(), 733 ExclaimTok.getLoc()); 734 Writeback = true; 735 Parser.Lex(); // Eat exclaim token 736 } 737 738 Operands.push_back(ARMOperand::CreateMem(BaseRegNum, OffsetIsReg, Offset, 739 OffsetRegNum, OffsetRegShifted, 740 ShiftType, ShiftAmount, Preindexed, 741 Postindexed, Negative, Writeback, 742 S, E)); 743 if (WBOp) 744 Operands.push_back(WBOp); 745 746 return false; 747 } 748 // The "[Rn" we have so far was not followed by a comma. 749 else if (Tok.is(AsmToken::RBrac)) { 750 // If there's anything other than the right brace, this is a post indexing 751 // addressing form. 752 E = Tok.getLoc(); 753 Parser.Lex(); // Eat right bracket token. 754 755 int OffsetRegNum = 0; 756 bool OffsetRegShifted = false; 757 enum ShiftType ShiftType = Lsl; 758 const MCExpr *ShiftAmount = 0; 759 const MCExpr *Offset = 0; 760 761 const AsmToken &NextTok = Parser.getTok(); 762 763 if (NextTok.isNot(AsmToken::EndOfStatement)) { 764 Postindexed = true; 765 Writeback = true; 766 767 if (NextTok.isNot(AsmToken::Comma)) { 768 Error(NextTok.getLoc(), "',' expected"); 769 return true; 770 } 771 772 Parser.Lex(); // Eat comma token. 773 774 if (ParseMemoryOffsetReg(Negative, OffsetRegShifted, ShiftType, 775 ShiftAmount, Offset, OffsetIsReg, OffsetRegNum, 776 E)) 777 return true; 778 } 779 780 Operands.push_back(ARMOperand::CreateMem(BaseRegNum, OffsetIsReg, Offset, 781 OffsetRegNum, OffsetRegShifted, 782 ShiftType, ShiftAmount, Preindexed, 783 Postindexed, Negative, Writeback, 784 S, E)); 785 return false; 786 } 787 788 return true; 789} 790 791/// Parse the offset of a memory operand after we have seen "[Rn," or "[Rn]," 792/// we will parse the following (were +/- means that a plus or minus is 793/// optional): 794/// +/-Rm 795/// +/-Rm, shift 796/// #offset 797/// we return false on success or an error otherwise. 798bool ARMAsmParser::ParseMemoryOffsetReg(bool &Negative, 799 bool &OffsetRegShifted, 800 enum ShiftType &ShiftType, 801 const MCExpr *&ShiftAmount, 802 const MCExpr *&Offset, 803 bool &OffsetIsReg, 804 int &OffsetRegNum, 805 SMLoc &E) { 806 Negative = false; 807 OffsetRegShifted = false; 808 OffsetIsReg = false; 809 OffsetRegNum = -1; 810 const AsmToken &NextTok = Parser.getTok(); 811 E = NextTok.getLoc(); 812 if (NextTok.is(AsmToken::Plus)) 813 Parser.Lex(); // Eat plus token. 814 else if (NextTok.is(AsmToken::Minus)) { 815 Negative = true; 816 Parser.Lex(); // Eat minus token 817 } 818 // See if there is a register following the "[Rn," or "[Rn]," we have so far. 819 const AsmToken &OffsetRegTok = Parser.getTok(); 820 if (OffsetRegTok.is(AsmToken::Identifier)) { 821 SMLoc CurLoc = OffsetRegTok.getLoc(); 822 OffsetRegNum = TryParseRegister(); 823 if (OffsetRegNum != -1) { 824 OffsetIsReg = true; 825 E = CurLoc; 826 } 827 } 828 829 // If we parsed a register as the offset then there can be a shift after that. 830 if (OffsetRegNum != -1) { 831 // Look for a comma then a shift 832 const AsmToken &Tok = Parser.getTok(); 833 if (Tok.is(AsmToken::Comma)) { 834 Parser.Lex(); // Eat comma token. 835 836 const AsmToken &Tok = Parser.getTok(); 837 if (ParseShift(ShiftType, ShiftAmount, E)) 838 return Error(Tok.getLoc(), "shift expected"); 839 OffsetRegShifted = true; 840 } 841 } 842 else { // the "[Rn," or "[Rn,]" we have so far was not followed by "Rm" 843 // Look for #offset following the "[Rn," or "[Rn]," 844 const AsmToken &HashTok = Parser.getTok(); 845 if (HashTok.isNot(AsmToken::Hash)) 846 return Error(HashTok.getLoc(), "'#' expected"); 847 848 Parser.Lex(); // Eat hash token. 849 850 if (getParser().ParseExpression(Offset)) 851 return true; 852 E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1); 853 } 854 return false; 855} 856 857/// ParseShift as one of these two: 858/// ( lsl | lsr | asr | ror ) , # shift_amount 859/// rrx 860/// and returns true if it parses a shift otherwise it returns false. 861bool ARMAsmParser::ParseShift(ShiftType &St, const MCExpr *&ShiftAmount, 862 SMLoc &E) { 863 const AsmToken &Tok = Parser.getTok(); 864 if (Tok.isNot(AsmToken::Identifier)) 865 return true; 866 StringRef ShiftName = Tok.getString(); 867 if (ShiftName == "lsl" || ShiftName == "LSL") 868 St = Lsl; 869 else if (ShiftName == "lsr" || ShiftName == "LSR") 870 St = Lsr; 871 else if (ShiftName == "asr" || ShiftName == "ASR") 872 St = Asr; 873 else if (ShiftName == "ror" || ShiftName == "ROR") 874 St = Ror; 875 else if (ShiftName == "rrx" || ShiftName == "RRX") 876 St = Rrx; 877 else 878 return true; 879 Parser.Lex(); // Eat shift type token. 880 881 // Rrx stands alone. 882 if (St == Rrx) 883 return false; 884 885 // Otherwise, there must be a '#' and a shift amount. 886 const AsmToken &HashTok = Parser.getTok(); 887 if (HashTok.isNot(AsmToken::Hash)) 888 return Error(HashTok.getLoc(), "'#' expected"); 889 Parser.Lex(); // Eat hash token. 890 891 if (getParser().ParseExpression(ShiftAmount)) 892 return true; 893 894 return false; 895} 896 897/// Parse a arm instruction operand. For now this parses the operand regardless 898/// of the mnemonic. 899bool ARMAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands, 900 bool isMCR){ 901 SMLoc S, E; 902 switch (getLexer().getKind()) { 903 default: 904 Error(Parser.getTok().getLoc(), "unexpected token in operand"); 905 return true; 906 case AsmToken::Identifier: 907 if (!TryParseRegisterWithWriteBack(Operands)) 908 return false; 909 if (isMCR && !TryParseMCRName(Operands)) 910 return false; 911 912 // Fall though for the Identifier case that is not a register or a 913 // special name. 914 case AsmToken::Integer: // things like 1f and 2b as a branch targets 915 case AsmToken::Dot: { // . as a branch target 916 // This was not a register so parse other operands that start with an 917 // identifier (like labels) as expressions and create them as immediates. 918 const MCExpr *IdVal; 919 S = Parser.getTok().getLoc(); 920 if (getParser().ParseExpression(IdVal)) 921 return true; 922 E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1); 923 Operands.push_back(ARMOperand::CreateImm(IdVal, S, E)); 924 return false; 925 } 926 case AsmToken::LBrac: 927 return ParseMemory(Operands); 928 case AsmToken::LCurly: 929 return ParseRegisterList(Operands); 930 case AsmToken::Hash: 931 // #42 -> immediate. 932 // TODO: ":lower16:" and ":upper16:" modifiers after # before immediate 933 S = Parser.getTok().getLoc(); 934 Parser.Lex(); 935 const MCExpr *ImmVal; 936 if (getParser().ParseExpression(ImmVal)) 937 return true; 938 E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1); 939 Operands.push_back(ARMOperand::CreateImm(ImmVal, S, E)); 940 return false; 941 case AsmToken::Colon: { 942 // ":lower16:" and ":upper16:" expression prefixes 943 // FIXME: Check it's an expression prefix, 944 // e.g. (FOO - :lower16:BAR) isn't legal. 945 ARMMCExpr::VariantKind RefKind; 946 if (ParsePrefix(RefKind)) 947 return true; 948 949 const MCExpr *SubExprVal; 950 if (getParser().ParseExpression(SubExprVal)) 951 return true; 952 953 const MCExpr *ExprVal = ARMMCExpr::Create(RefKind, SubExprVal, 954 getContext()); 955 E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1); 956 Operands.push_back(ARMOperand::CreateImm(ExprVal, S, E)); 957 return false; 958 } 959 } 960} 961 962// ParsePrefix - Parse ARM 16-bit relocations expression prefix, i.e. 963// :lower16: and :upper16:. 964bool ARMAsmParser::ParsePrefix(ARMMCExpr::VariantKind &RefKind) { 965 RefKind = ARMMCExpr::VK_ARM_None; 966 967 // :lower16: and :upper16: modifiers 968 assert(getLexer().is(AsmToken::Colon) && "expected a :"); 969 Parser.Lex(); // Eat ':' 970 971 if (getLexer().isNot(AsmToken::Identifier)) { 972 Error(Parser.getTok().getLoc(), "expected prefix identifier in operand"); 973 return true; 974 } 975 976 StringRef IDVal = Parser.getTok().getIdentifier(); 977 if (IDVal == "lower16") { 978 RefKind = ARMMCExpr::VK_ARM_LO16; 979 } else if (IDVal == "upper16") { 980 RefKind = ARMMCExpr::VK_ARM_HI16; 981 } else { 982 Error(Parser.getTok().getLoc(), "unexpected prefix in operand"); 983 return true; 984 } 985 Parser.Lex(); 986 987 if (getLexer().isNot(AsmToken::Colon)) { 988 Error(Parser.getTok().getLoc(), "unexpected token after prefix"); 989 return true; 990 } 991 Parser.Lex(); // Eat the last ':' 992 return false; 993} 994 995const MCExpr * 996ARMAsmParser::ApplyPrefixToExpr(const MCExpr *E, 997 MCSymbolRefExpr::VariantKind Variant) { 998 // Recurse over the given expression, rebuilding it to apply the given variant 999 // to the leftmost symbol. 1000 if (Variant == MCSymbolRefExpr::VK_None) 1001 return E; 1002 1003 switch (E->getKind()) { 1004 case MCExpr::Target: 1005 llvm_unreachable("Can't handle target expr yet"); 1006 case MCExpr::Constant: 1007 llvm_unreachable("Can't handle lower16/upper16 of constant yet"); 1008 1009 case MCExpr::SymbolRef: { 1010 const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(E); 1011 1012 if (SRE->getKind() != MCSymbolRefExpr::VK_None) 1013 return 0; 1014 1015 return MCSymbolRefExpr::Create(&SRE->getSymbol(), Variant, getContext()); 1016 } 1017 1018 case MCExpr::Unary: 1019 llvm_unreachable("Can't handle unary expressions yet"); 1020 1021 case MCExpr::Binary: { 1022 const MCBinaryExpr *BE = cast<MCBinaryExpr>(E); 1023 const MCExpr *LHS = ApplyPrefixToExpr(BE->getLHS(), Variant); 1024 const MCExpr *RHS = BE->getRHS(); 1025 if (!LHS) 1026 return 0; 1027 1028 return MCBinaryExpr::Create(BE->getOpcode(), LHS, RHS, getContext()); 1029 } 1030 } 1031 1032 assert(0 && "Invalid expression kind!"); 1033 return 0; 1034} 1035 1036/// \brief Given a mnemonic, split out possible predication code and carry 1037/// setting letters to form a canonical mnemonic and flags. 1038// 1039// FIXME: Would be nice to autogen this. 1040static StringRef SplitMnemonicAndCC(StringRef Mnemonic, 1041 unsigned &PredicationCode, 1042 bool &CarrySetting) { 1043 PredicationCode = ARMCC::AL; 1044 CarrySetting = false; 1045 1046 // Ignore some mnemonics we know aren't predicated forms. 1047 // 1048 // FIXME: Would be nice to autogen this. 1049 if (Mnemonic == "teq" || Mnemonic == "vceq" || 1050 Mnemonic == "movs" || 1051 Mnemonic == "svc" || 1052 (Mnemonic == "mls" || Mnemonic == "smmls" || Mnemonic == "vcls" || 1053 Mnemonic == "vmls" || Mnemonic == "vnmls") || 1054 Mnemonic == "vacge" || Mnemonic == "vcge" || 1055 Mnemonic == "vclt" || 1056 Mnemonic == "vacgt" || Mnemonic == "vcgt" || 1057 Mnemonic == "vcle" || 1058 (Mnemonic == "smlal" || Mnemonic == "umaal" || Mnemonic == "umlal" || 1059 Mnemonic == "vabal" || Mnemonic == "vmlal" || Mnemonic == "vpadal" || 1060 Mnemonic == "vqdmlal")) 1061 return Mnemonic; 1062 1063 // First, split out any predication code. 1064 unsigned CC = StringSwitch<unsigned>(Mnemonic.substr(Mnemonic.size()-2)) 1065 .Case("eq", ARMCC::EQ) 1066 .Case("ne", ARMCC::NE) 1067 .Case("hs", ARMCC::HS) 1068 .Case("lo", ARMCC::LO) 1069 .Case("mi", ARMCC::MI) 1070 .Case("pl", ARMCC::PL) 1071 .Case("vs", ARMCC::VS) 1072 .Case("vc", ARMCC::VC) 1073 .Case("hi", ARMCC::HI) 1074 .Case("ls", ARMCC::LS) 1075 .Case("ge", ARMCC::GE) 1076 .Case("lt", ARMCC::LT) 1077 .Case("gt", ARMCC::GT) 1078 .Case("le", ARMCC::LE) 1079 .Case("al", ARMCC::AL) 1080 .Default(~0U); 1081 if (CC != ~0U) { 1082 Mnemonic = Mnemonic.slice(0, Mnemonic.size() - 2); 1083 PredicationCode = CC; 1084 } 1085 1086 // Next, determine if we have a carry setting bit. We explicitly ignore all 1087 // the instructions we know end in 's'. 1088 if (Mnemonic.endswith("s") && 1089 !(Mnemonic == "asrs" || Mnemonic == "cps" || Mnemonic == "mls" || 1090 Mnemonic == "movs" || Mnemonic == "mrs" || Mnemonic == "smmls" || 1091 Mnemonic == "vabs" || Mnemonic == "vcls" || Mnemonic == "vmls" || 1092 Mnemonic == "vmrs" || Mnemonic == "vnmls" || Mnemonic == "vqabs" || 1093 Mnemonic == "vrecps" || Mnemonic == "vrsqrts")) { 1094 Mnemonic = Mnemonic.slice(0, Mnemonic.size() - 1); 1095 CarrySetting = true; 1096 } 1097 1098 return Mnemonic; 1099} 1100 1101/// \brief Given a canonical mnemonic, determine if the instruction ever allows 1102/// inclusion of carry set or predication code operands. 1103// 1104// FIXME: It would be nice to autogen this. 1105static void GetMnemonicAcceptInfo(StringRef Mnemonic, bool &CanAcceptCarrySet, 1106 bool &CanAcceptPredicationCode) { 1107 if (Mnemonic == "and" || Mnemonic == "lsl" || Mnemonic == "lsr" || 1108 Mnemonic == "rrx" || Mnemonic == "ror" || Mnemonic == "sub" || 1109 Mnemonic == "smull" || Mnemonic == "add" || Mnemonic == "adc" || 1110 Mnemonic == "mul" || Mnemonic == "bic" || Mnemonic == "asr" || 1111 Mnemonic == "umlal" || Mnemonic == "orr" || Mnemonic == "mov" || 1112 Mnemonic == "rsb" || Mnemonic == "rsc" || Mnemonic == "orn" || 1113 Mnemonic == "sbc" || Mnemonic == "mla" || Mnemonic == "umull" || 1114 Mnemonic == "eor" || Mnemonic == "smlal" || Mnemonic == "mvn") { 1115 CanAcceptCarrySet = true; 1116 } else { 1117 CanAcceptCarrySet = false; 1118 } 1119 1120 if (Mnemonic == "cbnz" || Mnemonic == "setend" || Mnemonic == "dmb" || 1121 Mnemonic == "cps" || Mnemonic == "mcr2" || Mnemonic == "it" || 1122 Mnemonic == "mcrr2" || Mnemonic == "cbz" || Mnemonic == "cdp2" || 1123 Mnemonic == "trap" || Mnemonic == "mrc2" || Mnemonic == "mrrc2" || 1124 Mnemonic == "dsb" || Mnemonic == "movs") { 1125 CanAcceptPredicationCode = false; 1126 } else { 1127 CanAcceptPredicationCode = true; 1128 } 1129} 1130 1131/// Parse an arm instruction mnemonic followed by its operands. 1132bool ARMAsmParser::ParseInstruction(StringRef Name, SMLoc NameLoc, 1133 SmallVectorImpl<MCParsedAsmOperand*> &Operands) { 1134 // Create the leading tokens for the mnemonic, split by '.' characters. 1135 size_t Start = 0, Next = Name.find('.'); 1136 StringRef Head = Name.slice(Start, Next); 1137 1138 // Split out the predication code and carry setting flag from the mnemonic. 1139 unsigned PredicationCode; 1140 bool CarrySetting; 1141 Head = SplitMnemonicAndCC(Head, PredicationCode, CarrySetting); 1142 1143 Operands.push_back(ARMOperand::CreateToken(Head, NameLoc)); 1144 1145 // Next, add the CCOut and ConditionCode operands, if needed. 1146 // 1147 // For mnemonics which can ever incorporate a carry setting bit or predication 1148 // code, our matching model involves us always generating CCOut and 1149 // ConditionCode operands to match the mnemonic "as written" and then we let 1150 // the matcher deal with finding the right instruction or generating an 1151 // appropriate error. 1152 bool CanAcceptCarrySet, CanAcceptPredicationCode; 1153 GetMnemonicAcceptInfo(Head, CanAcceptCarrySet, CanAcceptPredicationCode); 1154 1155 // Add the carry setting operand, if necessary. 1156 // 1157 // FIXME: It would be awesome if we could somehow invent a location such that 1158 // match errors on this operand would print a nice diagnostic about how the 1159 // 's' character in the mnemonic resulted in a CCOut operand. 1160 if (CanAcceptCarrySet) { 1161 Operands.push_back(ARMOperand::CreateCCOut(CarrySetting ? ARM::CPSR : 0, 1162 NameLoc)); 1163 } else { 1164 // This mnemonic can't ever accept a carry set, but the user wrote one (or 1165 // misspelled another mnemonic). 1166 1167 // FIXME: Issue a nice error. 1168 } 1169 1170 // Add the predication code operand, if necessary. 1171 if (CanAcceptPredicationCode) { 1172 Operands.push_back(ARMOperand::CreateCondCode( 1173 ARMCC::CondCodes(PredicationCode), NameLoc)); 1174 } else { 1175 // This mnemonic can't ever accept a predication code, but the user wrote 1176 // one (or misspelled another mnemonic). 1177 1178 // FIXME: Issue a nice error. 1179 } 1180 1181 // Add the remaining tokens in the mnemonic. 1182 while (Next != StringRef::npos) { 1183 Start = Next; 1184 Next = Name.find('.', Start + 1); 1185 Head = Name.slice(Start, Next); 1186 1187 Operands.push_back(ARMOperand::CreateToken(Head, NameLoc)); 1188 } 1189 1190 bool isMCR = (Head == "mcr" || Head == "mcr2" || 1191 Head == "mcrr" || Head == "mcrr2" || 1192 Head == "mrc" || Head == "mrc2" || 1193 Head == "mrrc" || Head == "mrrc2"); 1194 1195 // Read the remaining operands. 1196 if (getLexer().isNot(AsmToken::EndOfStatement)) { 1197 // Read the first operand. 1198 if (ParseOperand(Operands, isMCR)) { 1199 Parser.EatToEndOfStatement(); 1200 return true; 1201 } 1202 1203 while (getLexer().is(AsmToken::Comma)) { 1204 Parser.Lex(); // Eat the comma. 1205 1206 // Parse and remember the operand. 1207 if (ParseOperand(Operands, isMCR)) { 1208 Parser.EatToEndOfStatement(); 1209 return true; 1210 } 1211 } 1212 } 1213 1214 if (getLexer().isNot(AsmToken::EndOfStatement)) { 1215 Parser.EatToEndOfStatement(); 1216 return TokError("unexpected token in argument list"); 1217 } 1218 1219 Parser.Lex(); // Consume the EndOfStatement 1220 return false; 1221} 1222 1223bool ARMAsmParser:: 1224MatchAndEmitInstruction(SMLoc IDLoc, 1225 SmallVectorImpl<MCParsedAsmOperand*> &Operands, 1226 MCStreamer &Out) { 1227 MCInst Inst; 1228 unsigned ErrorInfo; 1229 MatchResultTy MatchResult, MatchResult2; 1230 MatchResult = MatchInstructionImpl(Operands, Inst, ErrorInfo); 1231 if (MatchResult != Match_Success) { 1232 // If we get a Match_InvalidOperand it might be some arithmetic instruction 1233 // that does not update the condition codes. So try adding a CCOut operand 1234 // with a value of reg0. 1235 if (MatchResult == Match_InvalidOperand) { 1236 Operands.insert(Operands.begin() + 1, 1237 ARMOperand::CreateCCOut(0, 1238 ((ARMOperand*)Operands[0])->getStartLoc())); 1239 MatchResult2 = MatchInstructionImpl(Operands, Inst, ErrorInfo); 1240 if (MatchResult2 == Match_Success) 1241 MatchResult = Match_Success; 1242 else { 1243 ARMOperand *CCOut = ((ARMOperand*)Operands[1]); 1244 Operands.erase(Operands.begin() + 1); 1245 delete CCOut; 1246 } 1247 } 1248 // If we get a Match_MnemonicFail it might be some arithmetic instruction 1249 // that updates the condition codes if it ends in 's'. So see if the 1250 // mnemonic ends in 's' and if so try removing the 's' and adding a CCOut 1251 // operand with a value of CPSR. 1252 else if(MatchResult == Match_MnemonicFail) { 1253 // Get the instruction mnemonic, which is the first token. 1254 StringRef Mnemonic = ((ARMOperand*)Operands[0])->getToken(); 1255 if (Mnemonic.substr(Mnemonic.size()-1) == "s") { 1256 // removed the 's' from the mnemonic for matching. 1257 StringRef MnemonicNoS = Mnemonic.slice(0, Mnemonic.size() - 1); 1258 SMLoc NameLoc = ((ARMOperand*)Operands[0])->getStartLoc(); 1259 ARMOperand *OldMnemonic = ((ARMOperand*)Operands[0]); 1260 Operands.erase(Operands.begin()); 1261 delete OldMnemonic; 1262 Operands.insert(Operands.begin(), 1263 ARMOperand::CreateToken(MnemonicNoS, NameLoc)); 1264 Operands.insert(Operands.begin() + 1, 1265 ARMOperand::CreateCCOut(ARM::CPSR, NameLoc)); 1266 MatchResult2 = MatchInstructionImpl(Operands, Inst, ErrorInfo); 1267 if (MatchResult2 == Match_Success) 1268 MatchResult = Match_Success; 1269 else { 1270 ARMOperand *OldMnemonic = ((ARMOperand*)Operands[0]); 1271 Operands.erase(Operands.begin()); 1272 delete OldMnemonic; 1273 Operands.insert(Operands.begin(), 1274 ARMOperand::CreateToken(Mnemonic, NameLoc)); 1275 ARMOperand *CCOut = ((ARMOperand*)Operands[1]); 1276 Operands.erase(Operands.begin() + 1); 1277 delete CCOut; 1278 } 1279 } 1280 } 1281 } 1282 switch (MatchResult) { 1283 case Match_Success: 1284 Out.EmitInstruction(Inst); 1285 return false; 1286 case Match_MissingFeature: 1287 Error(IDLoc, "instruction requires a CPU feature not currently enabled"); 1288 return true; 1289 case Match_InvalidOperand: { 1290 SMLoc ErrorLoc = IDLoc; 1291 if (ErrorInfo != ~0U) { 1292 if (ErrorInfo >= Operands.size()) 1293 return Error(IDLoc, "too few operands for instruction"); 1294 1295 ErrorLoc = ((ARMOperand*)Operands[ErrorInfo])->getStartLoc(); 1296 if (ErrorLoc == SMLoc()) ErrorLoc = IDLoc; 1297 } 1298 1299 return Error(ErrorLoc, "invalid operand for instruction"); 1300 } 1301 case Match_MnemonicFail: 1302 return Error(IDLoc, "unrecognized instruction mnemonic"); 1303 } 1304 1305 llvm_unreachable("Implement any new match types added!"); 1306 return true; 1307} 1308 1309/// ParseDirective parses the arm specific directives 1310bool ARMAsmParser::ParseDirective(AsmToken DirectiveID) { 1311 StringRef IDVal = DirectiveID.getIdentifier(); 1312 if (IDVal == ".word") 1313 return ParseDirectiveWord(4, DirectiveID.getLoc()); 1314 else if (IDVal == ".thumb") 1315 return ParseDirectiveThumb(DirectiveID.getLoc()); 1316 else if (IDVal == ".thumb_func") 1317 return ParseDirectiveThumbFunc(DirectiveID.getLoc()); 1318 else if (IDVal == ".code") 1319 return ParseDirectiveCode(DirectiveID.getLoc()); 1320 else if (IDVal == ".syntax") 1321 return ParseDirectiveSyntax(DirectiveID.getLoc()); 1322 return true; 1323} 1324 1325/// ParseDirectiveWord 1326/// ::= .word [ expression (, expression)* ] 1327bool ARMAsmParser::ParseDirectiveWord(unsigned Size, SMLoc L) { 1328 if (getLexer().isNot(AsmToken::EndOfStatement)) { 1329 for (;;) { 1330 const MCExpr *Value; 1331 if (getParser().ParseExpression(Value)) 1332 return true; 1333 1334 getParser().getStreamer().EmitValue(Value, Size, 0/*addrspace*/); 1335 1336 if (getLexer().is(AsmToken::EndOfStatement)) 1337 break; 1338 1339 // FIXME: Improve diagnostic. 1340 if (getLexer().isNot(AsmToken::Comma)) 1341 return Error(L, "unexpected token in directive"); 1342 Parser.Lex(); 1343 } 1344 } 1345 1346 Parser.Lex(); 1347 return false; 1348} 1349 1350/// ParseDirectiveThumb 1351/// ::= .thumb 1352bool ARMAsmParser::ParseDirectiveThumb(SMLoc L) { 1353 if (getLexer().isNot(AsmToken::EndOfStatement)) 1354 return Error(L, "unexpected token in directive"); 1355 Parser.Lex(); 1356 1357 // TODO: set thumb mode 1358 // TODO: tell the MC streamer the mode 1359 // getParser().getStreamer().Emit???(); 1360 return false; 1361} 1362 1363/// ParseDirectiveThumbFunc 1364/// ::= .thumbfunc symbol_name 1365bool ARMAsmParser::ParseDirectiveThumbFunc(SMLoc L) { 1366 const AsmToken &Tok = Parser.getTok(); 1367 if (Tok.isNot(AsmToken::Identifier) && Tok.isNot(AsmToken::String)) 1368 return Error(L, "unexpected token in .thumb_func directive"); 1369 StringRef Name = Tok.getString(); 1370 Parser.Lex(); // Consume the identifier token. 1371 if (getLexer().isNot(AsmToken::EndOfStatement)) 1372 return Error(L, "unexpected token in directive"); 1373 Parser.Lex(); 1374 1375 // Mark symbol as a thumb symbol. 1376 MCSymbol *Func = getParser().getContext().GetOrCreateSymbol(Name); 1377 getParser().getStreamer().EmitThumbFunc(Func); 1378 return false; 1379} 1380 1381/// ParseDirectiveSyntax 1382/// ::= .syntax unified | divided 1383bool ARMAsmParser::ParseDirectiveSyntax(SMLoc L) { 1384 const AsmToken &Tok = Parser.getTok(); 1385 if (Tok.isNot(AsmToken::Identifier)) 1386 return Error(L, "unexpected token in .syntax directive"); 1387 StringRef Mode = Tok.getString(); 1388 if (Mode == "unified" || Mode == "UNIFIED") 1389 Parser.Lex(); 1390 else if (Mode == "divided" || Mode == "DIVIDED") 1391 Parser.Lex(); 1392 else 1393 return Error(L, "unrecognized syntax mode in .syntax directive"); 1394 1395 if (getLexer().isNot(AsmToken::EndOfStatement)) 1396 return Error(Parser.getTok().getLoc(), "unexpected token in directive"); 1397 Parser.Lex(); 1398 1399 // TODO tell the MC streamer the mode 1400 // getParser().getStreamer().Emit???(); 1401 return false; 1402} 1403 1404/// ParseDirectiveCode 1405/// ::= .code 16 | 32 1406bool ARMAsmParser::ParseDirectiveCode(SMLoc L) { 1407 const AsmToken &Tok = Parser.getTok(); 1408 if (Tok.isNot(AsmToken::Integer)) 1409 return Error(L, "unexpected token in .code directive"); 1410 int64_t Val = Parser.getTok().getIntVal(); 1411 if (Val == 16) 1412 Parser.Lex(); 1413 else if (Val == 32) 1414 Parser.Lex(); 1415 else 1416 return Error(L, "invalid operand to .code directive"); 1417 1418 if (getLexer().isNot(AsmToken::EndOfStatement)) 1419 return Error(Parser.getTok().getLoc(), "unexpected token in directive"); 1420 Parser.Lex(); 1421 1422 // FIXME: We need to be able switch subtargets at this point so that 1423 // MatchInstructionImpl() will work when it gets the AvailableFeatures which 1424 // includes Feature_IsThumb or not to match the right instructions. This is 1425 // blocked on the FIXME in llvm-mc.cpp when creating the TargetMachine. 1426 if (Val == 16){ 1427 assert(TM.getSubtarget<ARMSubtarget>().isThumb() && 1428 "switching between arm/thumb not yet suppported via .code 16)"); 1429 getParser().getStreamer().EmitAssemblerFlag(MCAF_Code16); 1430 } 1431 else{ 1432 assert(!TM.getSubtarget<ARMSubtarget>().isThumb() && 1433 "switching between thumb/arm not yet suppported via .code 32)"); 1434 getParser().getStreamer().EmitAssemblerFlag(MCAF_Code32); 1435 } 1436 1437 return false; 1438} 1439 1440extern "C" void LLVMInitializeARMAsmLexer(); 1441 1442/// Force static initialization. 1443extern "C" void LLVMInitializeARMAsmParser() { 1444 RegisterAsmParser<ARMAsmParser> X(TheARMTarget); 1445 RegisterAsmParser<ARMAsmParser> Y(TheThumbTarget); 1446 LLVMInitializeARMAsmLexer(); 1447} 1448 1449#define GET_REGISTER_MATCHER 1450#define GET_MATCHER_IMPLEMENTATION 1451#include "ARMGenAsmMatcher.inc" 1452