1 //===- AsmParser.cpp - Parser for Assembly Files --------------------------===// 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 // This class implements the parser for assembly files. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "llvm/ADT/APFloat.h" 15 #include "llvm/ADT/SmallString.h" 16 #include "llvm/ADT/STLExtras.h" 17 #include "llvm/ADT/StringMap.h" 18 #include "llvm/ADT/Twine.h" 19 #include "llvm/MC/MCAsmInfo.h" 20 #include "llvm/MC/MCContext.h" 21 #include "llvm/MC/MCDwarf.h" 22 #include "llvm/MC/MCExpr.h" 23 #include "llvm/MC/MCInstPrinter.h" 24 #include "llvm/MC/MCInstrInfo.h" 25 #include "llvm/MC/MCParser/AsmCond.h" 26 #include "llvm/MC/MCParser/AsmLexer.h" 27 #include "llvm/MC/MCParser/MCAsmParser.h" 28 #include "llvm/MC/MCParser/MCParsedAsmOperand.h" 29 #include "llvm/MC/MCRegisterInfo.h" 30 #include "llvm/MC/MCSectionMachO.h" 31 #include "llvm/MC/MCStreamer.h" 32 #include "llvm/MC/MCSymbol.h" 33 #include "llvm/MC/MCTargetAsmParser.h" 34 #include "llvm/Support/CommandLine.h" 35 #include "llvm/Support/ErrorHandling.h" 36 #include "llvm/Support/MathExtras.h" 37 #include "llvm/Support/MemoryBuffer.h" 38 #include "llvm/Support/SourceMgr.h" 39 #include "llvm/Support/raw_ostream.h" 40 #include <cctype> 41 #include <set> 42 #include <string> 43 #include <vector> 44 using namespace llvm; 45 46 static cl::opt<bool> 47 FatalAssemblerWarnings("fatal-assembler-warnings", 48 cl::desc("Consider warnings as error")); 49 50 MCAsmParserSemaCallback::~MCAsmParserSemaCallback() {} 51 52 namespace { 53 54 /// \brief Helper types for tracking macro definitions. 55 typedef std::vector<AsmToken> MCAsmMacroArgument; 56 typedef std::vector<MCAsmMacroArgument> MCAsmMacroArguments; 57 typedef std::pair<StringRef, MCAsmMacroArgument> MCAsmMacroParameter; 58 typedef std::vector<MCAsmMacroParameter> MCAsmMacroParameters; 59 60 struct MCAsmMacro { 61 StringRef Name; 62 StringRef Body; 63 MCAsmMacroParameters Parameters; 64 65 public: 66 MCAsmMacro(StringRef N, StringRef B, const MCAsmMacroParameters &P) : 67 Name(N), Body(B), Parameters(P) {} 68 69 MCAsmMacro(const MCAsmMacro& Other) 70 : Name(Other.Name), Body(Other.Body), Parameters(Other.Parameters) {} 71 }; 72 73 /// \brief Helper class for storing information about an active macro 74 /// instantiation. 75 struct MacroInstantiation { 76 /// The macro being instantiated. 77 const MCAsmMacro *TheMacro; 78 79 /// The macro instantiation with substitutions. 80 MemoryBuffer *Instantiation; 81 82 /// The location of the instantiation. 83 SMLoc InstantiationLoc; 84 85 /// The buffer where parsing should resume upon instantiation completion. 86 int ExitBuffer; 87 88 /// The location where parsing should resume upon instantiation completion. 89 SMLoc ExitLoc; 90 91 public: 92 MacroInstantiation(const MCAsmMacro *M, SMLoc IL, int EB, SMLoc EL, 93 MemoryBuffer *I); 94 }; 95 96 struct ParseStatementInfo { 97 /// ParsedOperands - The parsed operands from the last parsed statement. 98 SmallVector<MCParsedAsmOperand*, 8> ParsedOperands; 99 100 /// Opcode - The opcode from the last parsed instruction. 101 unsigned Opcode; 102 103 /// Error - Was there an error parsing the inline assembly? 104 bool ParseError; 105 106 SmallVectorImpl<AsmRewrite> *AsmRewrites; 107 108 ParseStatementInfo() : Opcode(~0U), ParseError(false), AsmRewrites(0) {} 109 ParseStatementInfo(SmallVectorImpl<AsmRewrite> *rewrites) 110 : Opcode(~0), ParseError(false), AsmRewrites(rewrites) {} 111 112 ~ParseStatementInfo() { 113 // Free any parsed operands. 114 for (unsigned i = 0, e = ParsedOperands.size(); i != e; ++i) 115 delete ParsedOperands[i]; 116 ParsedOperands.clear(); 117 } 118 }; 119 120 /// \brief The concrete assembly parser instance. 121 class AsmParser : public MCAsmParser { 122 AsmParser(const AsmParser &) LLVM_DELETED_FUNCTION; 123 void operator=(const AsmParser &) LLVM_DELETED_FUNCTION; 124 private: 125 AsmLexer Lexer; 126 MCContext &Ctx; 127 MCStreamer &Out; 128 const MCAsmInfo &MAI; 129 SourceMgr &SrcMgr; 130 SourceMgr::DiagHandlerTy SavedDiagHandler; 131 void *SavedDiagContext; 132 MCAsmParserExtension *PlatformParser; 133 134 /// This is the current buffer index we're lexing from as managed by the 135 /// SourceMgr object. 136 int CurBuffer; 137 138 AsmCond TheCondState; 139 std::vector<AsmCond> TheCondStack; 140 141 /// ExtensionDirectiveMap - maps directive names to handler methods in parser 142 /// extensions. Extensions register themselves in this map by calling 143 /// addDirectiveHandler. 144 StringMap<ExtensionDirectiveHandler> ExtensionDirectiveMap; 145 146 /// MacroMap - Map of currently defined macros. 147 StringMap<MCAsmMacro*> MacroMap; 148 149 /// ActiveMacros - Stack of active macro instantiations. 150 std::vector<MacroInstantiation*> ActiveMacros; 151 152 /// Boolean tracking whether macro substitution is enabled. 153 unsigned MacrosEnabledFlag : 1; 154 155 /// Flag tracking whether any errors have been encountered. 156 unsigned HadError : 1; 157 158 /// The values from the last parsed cpp hash file line comment if any. 159 StringRef CppHashFilename; 160 int64_t CppHashLineNumber; 161 SMLoc CppHashLoc; 162 int CppHashBuf; 163 164 /// AssemblerDialect. ~OU means unset value and use value provided by MAI. 165 unsigned AssemblerDialect; 166 167 /// IsDarwin - is Darwin compatibility enabled? 168 bool IsDarwin; 169 170 /// ParsingInlineAsm - Are we parsing ms-style inline assembly? 171 bool ParsingInlineAsm; 172 173 public: 174 AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out, 175 const MCAsmInfo &MAI); 176 virtual ~AsmParser(); 177 178 virtual bool Run(bool NoInitialTextSection, bool NoFinalize = false); 179 180 virtual void addDirectiveHandler(StringRef Directive, 181 ExtensionDirectiveHandler Handler) { 182 ExtensionDirectiveMap[Directive] = Handler; 183 } 184 185 public: 186 /// @name MCAsmParser Interface 187 /// { 188 189 virtual SourceMgr &getSourceManager() { return SrcMgr; } 190 virtual MCAsmLexer &getLexer() { return Lexer; } 191 virtual MCContext &getContext() { return Ctx; } 192 virtual MCStreamer &getStreamer() { return Out; } 193 virtual unsigned getAssemblerDialect() { 194 if (AssemblerDialect == ~0U) 195 return MAI.getAssemblerDialect(); 196 else 197 return AssemblerDialect; 198 } 199 virtual void setAssemblerDialect(unsigned i) { 200 AssemblerDialect = i; 201 } 202 203 virtual bool Warning(SMLoc L, const Twine &Msg, 204 ArrayRef<SMRange> Ranges = None); 205 virtual bool Error(SMLoc L, const Twine &Msg, 206 ArrayRef<SMRange> Ranges = None); 207 208 virtual const AsmToken &Lex(); 209 210 void setParsingInlineAsm(bool V) { ParsingInlineAsm = V; } 211 bool isParsingInlineAsm() { return ParsingInlineAsm; } 212 213 bool parseMSInlineAsm(void *AsmLoc, std::string &AsmString, 214 unsigned &NumOutputs, unsigned &NumInputs, 215 SmallVectorImpl<std::pair<void *,bool> > &OpDecls, 216 SmallVectorImpl<std::string> &Constraints, 217 SmallVectorImpl<std::string> &Clobbers, 218 const MCInstrInfo *MII, 219 const MCInstPrinter *IP, 220 MCAsmParserSemaCallback &SI); 221 222 bool parseExpression(const MCExpr *&Res); 223 virtual bool parseExpression(const MCExpr *&Res, SMLoc &EndLoc); 224 virtual bool parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc); 225 virtual bool parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc); 226 virtual bool parseAbsoluteExpression(int64_t &Res); 227 228 /// parseIdentifier - Parse an identifier or string (as a quoted identifier) 229 /// and set \p Res to the identifier contents. 230 virtual bool parseIdentifier(StringRef &Res); 231 virtual void eatToEndOfStatement(); 232 233 virtual void checkForValidSection(); 234 /// } 235 236 private: 237 238 bool ParseStatement(ParseStatementInfo &Info); 239 void EatToEndOfLine(); 240 bool ParseCppHashLineFilenameComment(const SMLoc &L); 241 242 void CheckForBadMacro(SMLoc DirectiveLoc, StringRef Name, StringRef Body, 243 MCAsmMacroParameters Parameters); 244 bool expandMacro(raw_svector_ostream &OS, StringRef Body, 245 const MCAsmMacroParameters &Parameters, 246 const MCAsmMacroArguments &A, 247 const SMLoc &L); 248 249 /// \brief Are macros enabled in the parser? 250 bool MacrosEnabled() {return MacrosEnabledFlag;} 251 252 /// \brief Control a flag in the parser that enables or disables macros. 253 void SetMacrosEnabled(bool Flag) {MacrosEnabledFlag = Flag;} 254 255 /// \brief Lookup a previously defined macro. 256 /// \param Name Macro name. 257 /// \returns Pointer to macro. NULL if no such macro was defined. 258 const MCAsmMacro* LookupMacro(StringRef Name); 259 260 /// \brief Define a new macro with the given name and information. 261 void DefineMacro(StringRef Name, const MCAsmMacro& Macro); 262 263 /// \brief Undefine a macro. If no such macro was defined, it's a no-op. 264 void UndefineMacro(StringRef Name); 265 266 /// \brief Are we inside a macro instantiation? 267 bool InsideMacroInstantiation() {return !ActiveMacros.empty();} 268 269 /// \brief Handle entry to macro instantiation. 270 /// 271 /// \param M The macro. 272 /// \param NameLoc Instantiation location. 273 bool HandleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc); 274 275 /// \brief Handle exit from macro instantiation. 276 void HandleMacroExit(); 277 278 /// \brief Extract AsmTokens for a macro argument. If the argument delimiter 279 /// is initially unknown, set it to AsmToken::Eof. It will be set to the 280 /// correct delimiter by the method. 281 bool ParseMacroArgument(MCAsmMacroArgument &MA, 282 AsmToken::TokenKind &ArgumentDelimiter); 283 284 /// \brief Parse all macro arguments for a given macro. 285 bool ParseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A); 286 287 void PrintMacroInstantiations(); 288 void PrintMessage(SMLoc Loc, SourceMgr::DiagKind Kind, const Twine &Msg, 289 ArrayRef<SMRange> Ranges = None) const { 290 SrcMgr.PrintMessage(Loc, Kind, Msg, Ranges); 291 } 292 static void DiagHandler(const SMDiagnostic &Diag, void *Context); 293 294 /// EnterIncludeFile - Enter the specified file. This returns true on failure. 295 bool EnterIncludeFile(const std::string &Filename); 296 /// ProcessIncbinFile - Process the specified file for the .incbin directive. 297 /// This returns true on failure. 298 bool ProcessIncbinFile(const std::string &Filename); 299 300 /// \brief Reset the current lexer position to that given by \p Loc. The 301 /// current token is not set; clients should ensure Lex() is called 302 /// subsequently. 303 /// 304 /// \param InBuffer If not -1, should be the known buffer id that contains the 305 /// location. 306 void JumpToLoc(SMLoc Loc, int InBuffer=-1); 307 308 /// \brief Parse up to the end of statement and a return the contents from the 309 /// current token until the end of the statement; the current token on exit 310 /// will be either the EndOfStatement or EOF. 311 virtual StringRef parseStringToEndOfStatement(); 312 313 /// \brief Parse until the end of a statement or a comma is encountered, 314 /// return the contents from the current token up to the end or comma. 315 StringRef ParseStringToComma(); 316 317 bool ParseAssignment(StringRef Name, bool allow_redef, 318 bool NoDeadStrip = false); 319 320 bool ParsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc); 321 bool ParseBinOpRHS(unsigned Precedence, const MCExpr *&Res, SMLoc &EndLoc); 322 bool ParseParenExpr(const MCExpr *&Res, SMLoc &EndLoc); 323 bool ParseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc); 324 325 bool ParseRegisterOrRegisterNumber(int64_t &Register, SMLoc DirectiveLoc); 326 327 // Generic (target and platform independent) directive parsing. 328 enum DirectiveKind { 329 DK_NO_DIRECTIVE, // Placeholder 330 DK_SET, DK_EQU, DK_EQUIV, DK_ASCII, DK_ASCIZ, DK_STRING, DK_BYTE, DK_SHORT, 331 DK_VALUE, DK_2BYTE, DK_LONG, DK_INT, DK_4BYTE, DK_QUAD, DK_8BYTE, DK_SINGLE, 332 DK_FLOAT, DK_DOUBLE, DK_ALIGN, DK_ALIGN32, DK_BALIGN, DK_BALIGNW, 333 DK_BALIGNL, DK_P2ALIGN, DK_P2ALIGNW, DK_P2ALIGNL, DK_ORG, DK_FILL, DK_ENDR, 334 DK_BUNDLE_ALIGN_MODE, DK_BUNDLE_LOCK, DK_BUNDLE_UNLOCK, 335 DK_ZERO, DK_EXTERN, DK_GLOBL, DK_GLOBAL, DK_INDIRECT_SYMBOL, 336 DK_LAZY_REFERENCE, DK_NO_DEAD_STRIP, DK_SYMBOL_RESOLVER, DK_PRIVATE_EXTERN, 337 DK_REFERENCE, DK_WEAK_DEFINITION, DK_WEAK_REFERENCE, 338 DK_WEAK_DEF_CAN_BE_HIDDEN, DK_COMM, DK_COMMON, DK_LCOMM, DK_ABORT, 339 DK_INCLUDE, DK_INCBIN, DK_CODE16, DK_CODE16GCC, DK_REPT, DK_IRP, DK_IRPC, 340 DK_IF, DK_IFB, DK_IFNB, DK_IFC, DK_IFNC, DK_IFDEF, DK_IFNDEF, DK_IFNOTDEF, 341 DK_ELSEIF, DK_ELSE, DK_ENDIF, 342 DK_SPACE, DK_SKIP, DK_FILE, DK_LINE, DK_LOC, DK_STABS, 343 DK_CFI_SECTIONS, DK_CFI_STARTPROC, DK_CFI_ENDPROC, DK_CFI_DEF_CFA, 344 DK_CFI_DEF_CFA_OFFSET, DK_CFI_ADJUST_CFA_OFFSET, DK_CFI_DEF_CFA_REGISTER, 345 DK_CFI_OFFSET, DK_CFI_REL_OFFSET, DK_CFI_PERSONALITY, DK_CFI_LSDA, 346 DK_CFI_REMEMBER_STATE, DK_CFI_RESTORE_STATE, DK_CFI_SAME_VALUE, 347 DK_CFI_RESTORE, DK_CFI_ESCAPE, DK_CFI_SIGNAL_FRAME, DK_CFI_UNDEFINED, 348 DK_CFI_REGISTER, 349 DK_MACROS_ON, DK_MACROS_OFF, DK_MACRO, DK_ENDM, DK_ENDMACRO, DK_PURGEM, 350 DK_SLEB128, DK_ULEB128 351 }; 352 353 /// DirectiveKindMap - Maps directive name --> DirectiveKind enum, for 354 /// directives parsed by this class. 355 StringMap<DirectiveKind> DirectiveKindMap; 356 357 // ".ascii", ".asciz", ".string" 358 bool ParseDirectiveAscii(StringRef IDVal, bool ZeroTerminated); 359 bool ParseDirectiveValue(unsigned Size); // ".byte", ".long", ... 360 bool ParseDirectiveRealValue(const fltSemantics &); // ".single", ... 361 bool ParseDirectiveFill(); // ".fill" 362 bool ParseDirectiveZero(); // ".zero" 363 // ".set", ".equ", ".equiv" 364 bool ParseDirectiveSet(StringRef IDVal, bool allow_redef); 365 bool ParseDirectiveOrg(); // ".org" 366 // ".align{,32}", ".p2align{,w,l}" 367 bool ParseDirectiveAlign(bool IsPow2, unsigned ValueSize); 368 369 // ".file", ".line", ".loc", ".stabs" 370 bool ParseDirectiveFile(SMLoc DirectiveLoc); 371 bool ParseDirectiveLine(); 372 bool ParseDirectiveLoc(); 373 bool ParseDirectiveStabs(); 374 375 // .cfi directives 376 bool ParseDirectiveCFIRegister(SMLoc DirectiveLoc); 377 bool ParseDirectiveCFISections(); 378 bool ParseDirectiveCFIStartProc(); 379 bool ParseDirectiveCFIEndProc(); 380 bool ParseDirectiveCFIDefCfaOffset(); 381 bool ParseDirectiveCFIDefCfa(SMLoc DirectiveLoc); 382 bool ParseDirectiveCFIAdjustCfaOffset(); 383 bool ParseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc); 384 bool ParseDirectiveCFIOffset(SMLoc DirectiveLoc); 385 bool ParseDirectiveCFIRelOffset(SMLoc DirectiveLoc); 386 bool ParseDirectiveCFIPersonalityOrLsda(bool IsPersonality); 387 bool ParseDirectiveCFIRememberState(); 388 bool ParseDirectiveCFIRestoreState(); 389 bool ParseDirectiveCFISameValue(SMLoc DirectiveLoc); 390 bool ParseDirectiveCFIRestore(SMLoc DirectiveLoc); 391 bool ParseDirectiveCFIEscape(); 392 bool ParseDirectiveCFISignalFrame(); 393 bool ParseDirectiveCFIUndefined(SMLoc DirectiveLoc); 394 395 // macro directives 396 bool ParseDirectivePurgeMacro(SMLoc DirectiveLoc); 397 bool ParseDirectiveEndMacro(StringRef Directive); 398 bool ParseDirectiveMacro(SMLoc DirectiveLoc); 399 bool ParseDirectiveMacrosOnOff(StringRef Directive); 400 401 // ".bundle_align_mode" 402 bool ParseDirectiveBundleAlignMode(); 403 // ".bundle_lock" 404 bool ParseDirectiveBundleLock(); 405 // ".bundle_unlock" 406 bool ParseDirectiveBundleUnlock(); 407 408 // ".space", ".skip" 409 bool ParseDirectiveSpace(StringRef IDVal); 410 411 // .sleb128 (Signed=true) and .uleb128 (Signed=false) 412 bool ParseDirectiveLEB128(bool Signed); 413 414 /// ParseDirectiveSymbolAttribute - Parse a directive like ".globl" which 415 /// accepts a single symbol (which should be a label or an external). 416 bool ParseDirectiveSymbolAttribute(MCSymbolAttr Attr); 417 418 bool ParseDirectiveComm(bool IsLocal); // ".comm" and ".lcomm" 419 420 bool ParseDirectiveAbort(); // ".abort" 421 bool ParseDirectiveInclude(); // ".include" 422 bool ParseDirectiveIncbin(); // ".incbin" 423 424 bool ParseDirectiveIf(SMLoc DirectiveLoc); // ".if" 425 // ".ifb" or ".ifnb", depending on ExpectBlank. 426 bool ParseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank); 427 // ".ifc" or ".ifnc", depending on ExpectEqual. 428 bool ParseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual); 429 // ".ifdef" or ".ifndef", depending on expect_defined 430 bool ParseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined); 431 bool ParseDirectiveElseIf(SMLoc DirectiveLoc); // ".elseif" 432 bool ParseDirectiveElse(SMLoc DirectiveLoc); // ".else" 433 bool ParseDirectiveEndIf(SMLoc DirectiveLoc); // .endif 434 virtual bool parseEscapedString(std::string &Data); 435 436 const MCExpr *ApplyModifierToExpr(const MCExpr *E, 437 MCSymbolRefExpr::VariantKind Variant); 438 439 // Macro-like directives 440 MCAsmMacro *ParseMacroLikeBody(SMLoc DirectiveLoc); 441 void InstantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 442 raw_svector_ostream &OS); 443 bool ParseDirectiveRept(SMLoc DirectiveLoc); // ".rept" 444 bool ParseDirectiveIrp(SMLoc DirectiveLoc); // ".irp" 445 bool ParseDirectiveIrpc(SMLoc DirectiveLoc); // ".irpc" 446 bool ParseDirectiveEndr(SMLoc DirectiveLoc); // ".endr" 447 448 // "_emit" or "__emit" 449 bool ParseDirectiveMSEmit(SMLoc DirectiveLoc, ParseStatementInfo &Info, 450 size_t Len); 451 452 // "align" 453 bool ParseDirectiveMSAlign(SMLoc DirectiveLoc, ParseStatementInfo &Info); 454 455 void initializeDirectiveKindMap(); 456 }; 457 } 458 459 namespace llvm { 460 461 extern MCAsmParserExtension *createDarwinAsmParser(); 462 extern MCAsmParserExtension *createELFAsmParser(); 463 extern MCAsmParserExtension *createCOFFAsmParser(); 464 465 } 466 467 enum { DEFAULT_ADDRSPACE = 0 }; 468 469 AsmParser::AsmParser(SourceMgr &_SM, MCContext &_Ctx, 470 MCStreamer &_Out, const MCAsmInfo &_MAI) 471 : Lexer(_MAI), Ctx(_Ctx), Out(_Out), MAI(_MAI), SrcMgr(_SM), 472 PlatformParser(0), 473 CurBuffer(0), MacrosEnabledFlag(true), CppHashLineNumber(0), 474 AssemblerDialect(~0U), IsDarwin(false), ParsingInlineAsm(false) { 475 // Save the old handler. 476 SavedDiagHandler = SrcMgr.getDiagHandler(); 477 SavedDiagContext = SrcMgr.getDiagContext(); 478 // Set our own handler which calls the saved handler. 479 SrcMgr.setDiagHandler(DiagHandler, this); 480 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)); 481 482 // Initialize the platform / file format parser. 483 // 484 // FIXME: This is a hack, we need to (majorly) cleanup how these objects are 485 // created. 486 if (_MAI.hasMicrosoftFastStdCallMangling()) { 487 PlatformParser = createCOFFAsmParser(); 488 PlatformParser->Initialize(*this); 489 } else if (_MAI.hasSubsectionsViaSymbols()) { 490 PlatformParser = createDarwinAsmParser(); 491 PlatformParser->Initialize(*this); 492 IsDarwin = true; 493 } else { 494 PlatformParser = createELFAsmParser(); 495 PlatformParser->Initialize(*this); 496 } 497 498 initializeDirectiveKindMap(); 499 } 500 501 AsmParser::~AsmParser() { 502 assert(ActiveMacros.empty() && "Unexpected active macro instantiation!"); 503 504 // Destroy any macros. 505 for (StringMap<MCAsmMacro*>::iterator it = MacroMap.begin(), 506 ie = MacroMap.end(); it != ie; ++it) 507 delete it->getValue(); 508 509 delete PlatformParser; 510 } 511 512 void AsmParser::PrintMacroInstantiations() { 513 // Print the active macro instantiation stack. 514 for (std::vector<MacroInstantiation*>::const_reverse_iterator 515 it = ActiveMacros.rbegin(), ie = ActiveMacros.rend(); it != ie; ++it) 516 PrintMessage((*it)->InstantiationLoc, SourceMgr::DK_Note, 517 "while in macro instantiation"); 518 } 519 520 bool AsmParser::Warning(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) { 521 if (FatalAssemblerWarnings) 522 return Error(L, Msg, Ranges); 523 PrintMessage(L, SourceMgr::DK_Warning, Msg, Ranges); 524 PrintMacroInstantiations(); 525 return false; 526 } 527 528 bool AsmParser::Error(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) { 529 HadError = true; 530 PrintMessage(L, SourceMgr::DK_Error, Msg, Ranges); 531 PrintMacroInstantiations(); 532 return true; 533 } 534 535 bool AsmParser::EnterIncludeFile(const std::string &Filename) { 536 std::string IncludedFile; 537 int NewBuf = SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 538 if (NewBuf == -1) 539 return true; 540 541 CurBuffer = NewBuf; 542 543 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)); 544 545 return false; 546 } 547 548 /// Process the specified .incbin file by seaching for it in the include paths 549 /// then just emitting the byte contents of the file to the streamer. This 550 /// returns true on failure. 551 bool AsmParser::ProcessIncbinFile(const std::string &Filename) { 552 std::string IncludedFile; 553 int NewBuf = SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 554 if (NewBuf == -1) 555 return true; 556 557 // Pick up the bytes from the file and emit them. 558 getStreamer().EmitBytes(SrcMgr.getMemoryBuffer(NewBuf)->getBuffer(), 559 DEFAULT_ADDRSPACE); 560 return false; 561 } 562 563 void AsmParser::JumpToLoc(SMLoc Loc, int InBuffer) { 564 if (InBuffer != -1) { 565 CurBuffer = InBuffer; 566 } else { 567 CurBuffer = SrcMgr.FindBufferContainingLoc(Loc); 568 } 569 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer), Loc.getPointer()); 570 } 571 572 const AsmToken &AsmParser::Lex() { 573 const AsmToken *tok = &Lexer.Lex(); 574 575 if (tok->is(AsmToken::Eof)) { 576 // If this is the end of an included file, pop the parent file off the 577 // include stack. 578 SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer); 579 if (ParentIncludeLoc != SMLoc()) { 580 JumpToLoc(ParentIncludeLoc); 581 tok = &Lexer.Lex(); 582 } 583 } 584 585 if (tok->is(AsmToken::Error)) 586 Error(Lexer.getErrLoc(), Lexer.getErr()); 587 588 return *tok; 589 } 590 591 bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) { 592 // Create the initial section, if requested. 593 if (!NoInitialTextSection) 594 Out.InitSections(); 595 596 // Prime the lexer. 597 Lex(); 598 599 HadError = false; 600 AsmCond StartingCondState = TheCondState; 601 602 // If we are generating dwarf for assembly source files save the initial text 603 // section and generate a .file directive. 604 if (getContext().getGenDwarfForAssembly()) { 605 getContext().setGenDwarfSection(getStreamer().getCurrentSection().first); 606 MCSymbol *SectionStartSym = getContext().CreateTempSymbol(); 607 getStreamer().EmitLabel(SectionStartSym); 608 getContext().setGenDwarfSectionStartSym(SectionStartSym); 609 getStreamer().EmitDwarfFileDirective(getContext().nextGenDwarfFileNumber(), 610 StringRef(), 611 getContext().getMainFileName()); 612 } 613 614 // While we have input, parse each statement. 615 while (Lexer.isNot(AsmToken::Eof)) { 616 ParseStatementInfo Info; 617 if (!ParseStatement(Info)) continue; 618 619 // We had an error, validate that one was emitted and recover by skipping to 620 // the next line. 621 assert(HadError && "Parse statement returned an error, but none emitted!"); 622 eatToEndOfStatement(); 623 } 624 625 if (TheCondState.TheCond != StartingCondState.TheCond || 626 TheCondState.Ignore != StartingCondState.Ignore) 627 return TokError("unmatched .ifs or .elses"); 628 629 // Check to see there are no empty DwarfFile slots. 630 const SmallVectorImpl<MCDwarfFile *> &MCDwarfFiles = 631 getContext().getMCDwarfFiles(); 632 for (unsigned i = 1; i < MCDwarfFiles.size(); i++) { 633 if (!MCDwarfFiles[i]) 634 TokError("unassigned file number: " + Twine(i) + " for .file directives"); 635 } 636 637 // Check to see that all assembler local symbols were actually defined. 638 // Targets that don't do subsections via symbols may not want this, though, 639 // so conservatively exclude them. Only do this if we're finalizing, though, 640 // as otherwise we won't necessarilly have seen everything yet. 641 if (!NoFinalize && MAI.hasSubsectionsViaSymbols()) { 642 const MCContext::SymbolTable &Symbols = getContext().getSymbols(); 643 for (MCContext::SymbolTable::const_iterator i = Symbols.begin(), 644 e = Symbols.end(); 645 i != e; ++i) { 646 MCSymbol *Sym = i->getValue(); 647 // Variable symbols may not be marked as defined, so check those 648 // explicitly. If we know it's a variable, we have a definition for 649 // the purposes of this check. 650 if (Sym->isTemporary() && !Sym->isVariable() && !Sym->isDefined()) 651 // FIXME: We would really like to refer back to where the symbol was 652 // first referenced for a source location. We need to add something 653 // to track that. Currently, we just point to the end of the file. 654 PrintMessage(getLexer().getLoc(), SourceMgr::DK_Error, 655 "assembler local symbol '" + Sym->getName() + 656 "' not defined"); 657 } 658 } 659 660 661 // Finalize the output stream if there are no errors and if the client wants 662 // us to. 663 if (!HadError && !NoFinalize) 664 Out.Finish(); 665 666 return HadError; 667 } 668 669 void AsmParser::checkForValidSection() { 670 if (!ParsingInlineAsm && !getStreamer().getCurrentSection().first) { 671 TokError("expected section directive before assembly directive"); 672 Out.InitToTextSection(); 673 } 674 } 675 676 /// eatToEndOfStatement - Throw away the rest of the line for testing purposes. 677 void AsmParser::eatToEndOfStatement() { 678 while (Lexer.isNot(AsmToken::EndOfStatement) && 679 Lexer.isNot(AsmToken::Eof)) 680 Lex(); 681 682 // Eat EOL. 683 if (Lexer.is(AsmToken::EndOfStatement)) 684 Lex(); 685 } 686 687 StringRef AsmParser::parseStringToEndOfStatement() { 688 const char *Start = getTok().getLoc().getPointer(); 689 690 while (Lexer.isNot(AsmToken::EndOfStatement) && 691 Lexer.isNot(AsmToken::Eof)) 692 Lex(); 693 694 const char *End = getTok().getLoc().getPointer(); 695 return StringRef(Start, End - Start); 696 } 697 698 StringRef AsmParser::ParseStringToComma() { 699 const char *Start = getTok().getLoc().getPointer(); 700 701 while (Lexer.isNot(AsmToken::EndOfStatement) && 702 Lexer.isNot(AsmToken::Comma) && 703 Lexer.isNot(AsmToken::Eof)) 704 Lex(); 705 706 const char *End = getTok().getLoc().getPointer(); 707 return StringRef(Start, End - Start); 708 } 709 710 /// ParseParenExpr - Parse a paren expression and return it. 711 /// NOTE: This assumes the leading '(' has already been consumed. 712 /// 713 /// parenexpr ::= expr) 714 /// 715 bool AsmParser::ParseParenExpr(const MCExpr *&Res, SMLoc &EndLoc) { 716 if (parseExpression(Res)) return true; 717 if (Lexer.isNot(AsmToken::RParen)) 718 return TokError("expected ')' in parentheses expression"); 719 EndLoc = Lexer.getTok().getEndLoc(); 720 Lex(); 721 return false; 722 } 723 724 /// ParseBracketExpr - Parse a bracket expression and return it. 725 /// NOTE: This assumes the leading '[' has already been consumed. 726 /// 727 /// bracketexpr ::= expr] 728 /// 729 bool AsmParser::ParseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc) { 730 if (parseExpression(Res)) return true; 731 if (Lexer.isNot(AsmToken::RBrac)) 732 return TokError("expected ']' in brackets expression"); 733 EndLoc = Lexer.getTok().getEndLoc(); 734 Lex(); 735 return false; 736 } 737 738 /// ParsePrimaryExpr - Parse a primary expression and return it. 739 /// primaryexpr ::= (parenexpr 740 /// primaryexpr ::= symbol 741 /// primaryexpr ::= number 742 /// primaryexpr ::= '.' 743 /// primaryexpr ::= ~,+,- primaryexpr 744 bool AsmParser::ParsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) { 745 SMLoc FirstTokenLoc = getLexer().getLoc(); 746 AsmToken::TokenKind FirstTokenKind = Lexer.getKind(); 747 switch (FirstTokenKind) { 748 default: 749 return TokError("unknown token in expression"); 750 // If we have an error assume that we've already handled it. 751 case AsmToken::Error: 752 return true; 753 case AsmToken::Exclaim: 754 Lex(); // Eat the operator. 755 if (ParsePrimaryExpr(Res, EndLoc)) 756 return true; 757 Res = MCUnaryExpr::CreateLNot(Res, getContext()); 758 return false; 759 case AsmToken::Dollar: 760 case AsmToken::String: 761 case AsmToken::Identifier: { 762 StringRef Identifier; 763 if (parseIdentifier(Identifier)) { 764 if (FirstTokenKind == AsmToken::Dollar) 765 return Error(FirstTokenLoc, "invalid token in expression"); 766 return true; 767 } 768 769 EndLoc = SMLoc::getFromPointer(Identifier.end()); 770 771 // This is a symbol reference. 772 std::pair<StringRef, StringRef> Split = Identifier.split('@'); 773 MCSymbol *Sym = getContext().GetOrCreateSymbol(Split.first); 774 775 // Lookup the symbol variant if used. 776 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 777 if (Split.first.size() != Identifier.size()) { 778 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second); 779 if (Variant == MCSymbolRefExpr::VK_Invalid) { 780 Variant = MCSymbolRefExpr::VK_None; 781 return TokError("invalid variant '" + Split.second + "'"); 782 } 783 } 784 785 // If this is an absolute variable reference, substitute it now to preserve 786 // semantics in the face of reassignment. 787 if (Sym->isVariable() && isa<MCConstantExpr>(Sym->getVariableValue())) { 788 if (Variant) 789 return Error(EndLoc, "unexpected modifier on variable reference"); 790 791 Res = Sym->getVariableValue(); 792 return false; 793 } 794 795 // Otherwise create a symbol ref. 796 Res = MCSymbolRefExpr::Create(Sym, Variant, getContext()); 797 return false; 798 } 799 case AsmToken::Integer: { 800 SMLoc Loc = getTok().getLoc(); 801 int64_t IntVal = getTok().getIntVal(); 802 Res = MCConstantExpr::Create(IntVal, getContext()); 803 EndLoc = Lexer.getTok().getEndLoc(); 804 Lex(); // Eat token. 805 // Look for 'b' or 'f' following an Integer as a directional label 806 if (Lexer.getKind() == AsmToken::Identifier) { 807 StringRef IDVal = getTok().getString(); 808 if (IDVal == "f" || IDVal == "b"){ 809 MCSymbol *Sym = Ctx.GetDirectionalLocalSymbol(IntVal, 810 IDVal == "f" ? 1 : 0); 811 Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, 812 getContext()); 813 if (IDVal == "b" && Sym->isUndefined()) 814 return Error(Loc, "invalid reference to undefined symbol"); 815 EndLoc = Lexer.getTok().getEndLoc(); 816 Lex(); // Eat identifier. 817 } 818 } 819 return false; 820 } 821 case AsmToken::Real: { 822 APFloat RealVal(APFloat::IEEEdouble, getTok().getString()); 823 uint64_t IntVal = RealVal.bitcastToAPInt().getZExtValue(); 824 Res = MCConstantExpr::Create(IntVal, getContext()); 825 EndLoc = Lexer.getTok().getEndLoc(); 826 Lex(); // Eat token. 827 return false; 828 } 829 case AsmToken::Dot: { 830 // This is a '.' reference, which references the current PC. Emit a 831 // temporary label to the streamer and refer to it. 832 MCSymbol *Sym = Ctx.CreateTempSymbol(); 833 Out.EmitLabel(Sym); 834 Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, getContext()); 835 EndLoc = Lexer.getTok().getEndLoc(); 836 Lex(); // Eat identifier. 837 return false; 838 } 839 case AsmToken::LParen: 840 Lex(); // Eat the '('. 841 return ParseParenExpr(Res, EndLoc); 842 case AsmToken::LBrac: 843 if (!PlatformParser->HasBracketExpressions()) 844 return TokError("brackets expression not supported on this target"); 845 Lex(); // Eat the '['. 846 return ParseBracketExpr(Res, EndLoc); 847 case AsmToken::Minus: 848 Lex(); // Eat the operator. 849 if (ParsePrimaryExpr(Res, EndLoc)) 850 return true; 851 Res = MCUnaryExpr::CreateMinus(Res, getContext()); 852 return false; 853 case AsmToken::Plus: 854 Lex(); // Eat the operator. 855 if (ParsePrimaryExpr(Res, EndLoc)) 856 return true; 857 Res = MCUnaryExpr::CreatePlus(Res, getContext()); 858 return false; 859 case AsmToken::Tilde: 860 Lex(); // Eat the operator. 861 if (ParsePrimaryExpr(Res, EndLoc)) 862 return true; 863 Res = MCUnaryExpr::CreateNot(Res, getContext()); 864 return false; 865 } 866 } 867 868 bool AsmParser::parseExpression(const MCExpr *&Res) { 869 SMLoc EndLoc; 870 return parseExpression(Res, EndLoc); 871 } 872 873 bool AsmParser::parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) { 874 return ParsePrimaryExpr(Res, EndLoc); 875 } 876 877 const MCExpr * 878 AsmParser::ApplyModifierToExpr(const MCExpr *E, 879 MCSymbolRefExpr::VariantKind Variant) { 880 // Recurse over the given expression, rebuilding it to apply the given variant 881 // if there is exactly one symbol. 882 switch (E->getKind()) { 883 case MCExpr::Target: 884 case MCExpr::Constant: 885 return 0; 886 887 case MCExpr::SymbolRef: { 888 const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(E); 889 890 if (SRE->getKind() != MCSymbolRefExpr::VK_None) { 891 TokError("invalid variant on expression '" + 892 getTok().getIdentifier() + "' (already modified)"); 893 return E; 894 } 895 896 return MCSymbolRefExpr::Create(&SRE->getSymbol(), Variant, getContext()); 897 } 898 899 case MCExpr::Unary: { 900 const MCUnaryExpr *UE = cast<MCUnaryExpr>(E); 901 const MCExpr *Sub = ApplyModifierToExpr(UE->getSubExpr(), Variant); 902 if (!Sub) 903 return 0; 904 return MCUnaryExpr::Create(UE->getOpcode(), Sub, getContext()); 905 } 906 907 case MCExpr::Binary: { 908 const MCBinaryExpr *BE = cast<MCBinaryExpr>(E); 909 const MCExpr *LHS = ApplyModifierToExpr(BE->getLHS(), Variant); 910 const MCExpr *RHS = ApplyModifierToExpr(BE->getRHS(), Variant); 911 912 if (!LHS && !RHS) 913 return 0; 914 915 if (!LHS) LHS = BE->getLHS(); 916 if (!RHS) RHS = BE->getRHS(); 917 918 return MCBinaryExpr::Create(BE->getOpcode(), LHS, RHS, getContext()); 919 } 920 } 921 922 llvm_unreachable("Invalid expression kind!"); 923 } 924 925 /// parseExpression - Parse an expression and return it. 926 /// 927 /// expr ::= expr &&,|| expr -> lowest. 928 /// expr ::= expr |,^,&,! expr 929 /// expr ::= expr ==,!=,<>,<,<=,>,>= expr 930 /// expr ::= expr <<,>> expr 931 /// expr ::= expr +,- expr 932 /// expr ::= expr *,/,% expr -> highest. 933 /// expr ::= primaryexpr 934 /// 935 bool AsmParser::parseExpression(const MCExpr *&Res, SMLoc &EndLoc) { 936 // Parse the expression. 937 Res = 0; 938 if (ParsePrimaryExpr(Res, EndLoc) || ParseBinOpRHS(1, Res, EndLoc)) 939 return true; 940 941 // As a special case, we support 'a op b @ modifier' by rewriting the 942 // expression to include the modifier. This is inefficient, but in general we 943 // expect users to use 'a@modifier op b'. 944 if (Lexer.getKind() == AsmToken::At) { 945 Lex(); 946 947 if (Lexer.isNot(AsmToken::Identifier)) 948 return TokError("unexpected symbol modifier following '@'"); 949 950 MCSymbolRefExpr::VariantKind Variant = 951 MCSymbolRefExpr::getVariantKindForName(getTok().getIdentifier()); 952 if (Variant == MCSymbolRefExpr::VK_Invalid) 953 return TokError("invalid variant '" + getTok().getIdentifier() + "'"); 954 955 const MCExpr *ModifiedRes = ApplyModifierToExpr(Res, Variant); 956 if (!ModifiedRes) { 957 return TokError("invalid modifier '" + getTok().getIdentifier() + 958 "' (no symbols present)"); 959 } 960 961 Res = ModifiedRes; 962 Lex(); 963 } 964 965 // Try to constant fold it up front, if possible. 966 int64_t Value; 967 if (Res->EvaluateAsAbsolute(Value)) 968 Res = MCConstantExpr::Create(Value, getContext()); 969 970 return false; 971 } 972 973 bool AsmParser::parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) { 974 Res = 0; 975 return ParseParenExpr(Res, EndLoc) || 976 ParseBinOpRHS(1, Res, EndLoc); 977 } 978 979 bool AsmParser::parseAbsoluteExpression(int64_t &Res) { 980 const MCExpr *Expr; 981 982 SMLoc StartLoc = Lexer.getLoc(); 983 if (parseExpression(Expr)) 984 return true; 985 986 if (!Expr->EvaluateAsAbsolute(Res)) 987 return Error(StartLoc, "expected absolute expression"); 988 989 return false; 990 } 991 992 static unsigned getBinOpPrecedence(AsmToken::TokenKind K, 993 MCBinaryExpr::Opcode &Kind) { 994 switch (K) { 995 default: 996 return 0; // not a binop. 997 998 // Lowest Precedence: &&, || 999 case AsmToken::AmpAmp: 1000 Kind = MCBinaryExpr::LAnd; 1001 return 1; 1002 case AsmToken::PipePipe: 1003 Kind = MCBinaryExpr::LOr; 1004 return 1; 1005 1006 1007 // Low Precedence: |, &, ^ 1008 // 1009 // FIXME: gas seems to support '!' as an infix operator? 1010 case AsmToken::Pipe: 1011 Kind = MCBinaryExpr::Or; 1012 return 2; 1013 case AsmToken::Caret: 1014 Kind = MCBinaryExpr::Xor; 1015 return 2; 1016 case AsmToken::Amp: 1017 Kind = MCBinaryExpr::And; 1018 return 2; 1019 1020 // Low Intermediate Precedence: ==, !=, <>, <, <=, >, >= 1021 case AsmToken::EqualEqual: 1022 Kind = MCBinaryExpr::EQ; 1023 return 3; 1024 case AsmToken::ExclaimEqual: 1025 case AsmToken::LessGreater: 1026 Kind = MCBinaryExpr::NE; 1027 return 3; 1028 case AsmToken::Less: 1029 Kind = MCBinaryExpr::LT; 1030 return 3; 1031 case AsmToken::LessEqual: 1032 Kind = MCBinaryExpr::LTE; 1033 return 3; 1034 case AsmToken::Greater: 1035 Kind = MCBinaryExpr::GT; 1036 return 3; 1037 case AsmToken::GreaterEqual: 1038 Kind = MCBinaryExpr::GTE; 1039 return 3; 1040 1041 // Intermediate Precedence: <<, >> 1042 case AsmToken::LessLess: 1043 Kind = MCBinaryExpr::Shl; 1044 return 4; 1045 case AsmToken::GreaterGreater: 1046 Kind = MCBinaryExpr::Shr; 1047 return 4; 1048 1049 // High Intermediate Precedence: +, - 1050 case AsmToken::Plus: 1051 Kind = MCBinaryExpr::Add; 1052 return 5; 1053 case AsmToken::Minus: 1054 Kind = MCBinaryExpr::Sub; 1055 return 5; 1056 1057 // Highest Precedence: *, /, % 1058 case AsmToken::Star: 1059 Kind = MCBinaryExpr::Mul; 1060 return 6; 1061 case AsmToken::Slash: 1062 Kind = MCBinaryExpr::Div; 1063 return 6; 1064 case AsmToken::Percent: 1065 Kind = MCBinaryExpr::Mod; 1066 return 6; 1067 } 1068 } 1069 1070 1071 /// ParseBinOpRHS - Parse all binary operators with precedence >= 'Precedence'. 1072 /// Res contains the LHS of the expression on input. 1073 bool AsmParser::ParseBinOpRHS(unsigned Precedence, const MCExpr *&Res, 1074 SMLoc &EndLoc) { 1075 while (1) { 1076 MCBinaryExpr::Opcode Kind = MCBinaryExpr::Add; 1077 unsigned TokPrec = getBinOpPrecedence(Lexer.getKind(), Kind); 1078 1079 // If the next token is lower precedence than we are allowed to eat, return 1080 // successfully with what we ate already. 1081 if (TokPrec < Precedence) 1082 return false; 1083 1084 Lex(); 1085 1086 // Eat the next primary expression. 1087 const MCExpr *RHS; 1088 if (ParsePrimaryExpr(RHS, EndLoc)) return true; 1089 1090 // If BinOp binds less tightly with RHS than the operator after RHS, let 1091 // the pending operator take RHS as its LHS. 1092 MCBinaryExpr::Opcode Dummy; 1093 unsigned NextTokPrec = getBinOpPrecedence(Lexer.getKind(), Dummy); 1094 if (TokPrec < NextTokPrec) { 1095 if (ParseBinOpRHS(TokPrec+1, RHS, EndLoc)) return true; 1096 } 1097 1098 // Merge LHS and RHS according to operator. 1099 Res = MCBinaryExpr::Create(Kind, Res, RHS, getContext()); 1100 } 1101 } 1102 1103 /// ParseStatement: 1104 /// ::= EndOfStatement 1105 /// ::= Label* Directive ...Operands... EndOfStatement 1106 /// ::= Label* Identifier OperandList* EndOfStatement 1107 bool AsmParser::ParseStatement(ParseStatementInfo &Info) { 1108 if (Lexer.is(AsmToken::EndOfStatement)) { 1109 Out.AddBlankLine(); 1110 Lex(); 1111 return false; 1112 } 1113 1114 // Statements always start with an identifier or are a full line comment. 1115 AsmToken ID = getTok(); 1116 SMLoc IDLoc = ID.getLoc(); 1117 StringRef IDVal; 1118 int64_t LocalLabelVal = -1; 1119 // A full line comment is a '#' as the first token. 1120 if (Lexer.is(AsmToken::Hash)) 1121 return ParseCppHashLineFilenameComment(IDLoc); 1122 1123 // Allow an integer followed by a ':' as a directional local label. 1124 if (Lexer.is(AsmToken::Integer)) { 1125 LocalLabelVal = getTok().getIntVal(); 1126 if (LocalLabelVal < 0) { 1127 if (!TheCondState.Ignore) 1128 return TokError("unexpected token at start of statement"); 1129 IDVal = ""; 1130 } else { 1131 IDVal = getTok().getString(); 1132 Lex(); // Consume the integer token to be used as an identifier token. 1133 if (Lexer.getKind() != AsmToken::Colon) { 1134 if (!TheCondState.Ignore) 1135 return TokError("unexpected token at start of statement"); 1136 } 1137 } 1138 } else if (Lexer.is(AsmToken::Dot)) { 1139 // Treat '.' as a valid identifier in this context. 1140 Lex(); 1141 IDVal = "."; 1142 } else if (parseIdentifier(IDVal)) { 1143 if (!TheCondState.Ignore) 1144 return TokError("unexpected token at start of statement"); 1145 IDVal = ""; 1146 } 1147 1148 // Handle conditional assembly here before checking for skipping. We 1149 // have to do this so that .endif isn't skipped in a ".if 0" block for 1150 // example. 1151 StringMap<DirectiveKind>::const_iterator DirKindIt = 1152 DirectiveKindMap.find(IDVal); 1153 DirectiveKind DirKind = 1154 (DirKindIt == DirectiveKindMap.end()) ? DK_NO_DIRECTIVE : 1155 DirKindIt->getValue(); 1156 switch (DirKind) { 1157 default: 1158 break; 1159 case DK_IF: 1160 return ParseDirectiveIf(IDLoc); 1161 case DK_IFB: 1162 return ParseDirectiveIfb(IDLoc, true); 1163 case DK_IFNB: 1164 return ParseDirectiveIfb(IDLoc, false); 1165 case DK_IFC: 1166 return ParseDirectiveIfc(IDLoc, true); 1167 case DK_IFNC: 1168 return ParseDirectiveIfc(IDLoc, false); 1169 case DK_IFDEF: 1170 return ParseDirectiveIfdef(IDLoc, true); 1171 case DK_IFNDEF: 1172 case DK_IFNOTDEF: 1173 return ParseDirectiveIfdef(IDLoc, false); 1174 case DK_ELSEIF: 1175 return ParseDirectiveElseIf(IDLoc); 1176 case DK_ELSE: 1177 return ParseDirectiveElse(IDLoc); 1178 case DK_ENDIF: 1179 return ParseDirectiveEndIf(IDLoc); 1180 } 1181 1182 // Ignore the statement if in the middle of inactive conditional 1183 // (e.g. ".if 0"). 1184 if (TheCondState.Ignore) { 1185 eatToEndOfStatement(); 1186 return false; 1187 } 1188 1189 // FIXME: Recurse on local labels? 1190 1191 // See what kind of statement we have. 1192 switch (Lexer.getKind()) { 1193 case AsmToken::Colon: { 1194 checkForValidSection(); 1195 1196 // identifier ':' -> Label. 1197 Lex(); 1198 1199 // Diagnose attempt to use '.' as a label. 1200 if (IDVal == ".") 1201 return Error(IDLoc, "invalid use of pseudo-symbol '.' as a label"); 1202 1203 // Diagnose attempt to use a variable as a label. 1204 // 1205 // FIXME: Diagnostics. Note the location of the definition as a label. 1206 // FIXME: This doesn't diagnose assignment to a symbol which has been 1207 // implicitly marked as external. 1208 MCSymbol *Sym; 1209 if (LocalLabelVal == -1) 1210 Sym = getContext().GetOrCreateSymbol(IDVal); 1211 else 1212 Sym = Ctx.CreateDirectionalLocalSymbol(LocalLabelVal); 1213 if (!Sym->isUndefined() || Sym->isVariable()) 1214 return Error(IDLoc, "invalid symbol redefinition"); 1215 1216 // Emit the label. 1217 if (!ParsingInlineAsm) 1218 Out.EmitLabel(Sym); 1219 1220 // If we are generating dwarf for assembly source files then gather the 1221 // info to make a dwarf label entry for this label if needed. 1222 if (getContext().getGenDwarfForAssembly()) 1223 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(), 1224 IDLoc); 1225 1226 // Consume any end of statement token, if present, to avoid spurious 1227 // AddBlankLine calls(). 1228 if (Lexer.is(AsmToken::EndOfStatement)) { 1229 Lex(); 1230 if (Lexer.is(AsmToken::Eof)) 1231 return false; 1232 } 1233 1234 return false; 1235 } 1236 1237 case AsmToken::Equal: 1238 // identifier '=' ... -> assignment statement 1239 Lex(); 1240 1241 return ParseAssignment(IDVal, true); 1242 1243 default: // Normal instruction or directive. 1244 break; 1245 } 1246 1247 // If macros are enabled, check to see if this is a macro instantiation. 1248 if (MacrosEnabled()) 1249 if (const MCAsmMacro *M = LookupMacro(IDVal)) { 1250 return HandleMacroEntry(M, IDLoc); 1251 } 1252 1253 // Otherwise, we have a normal instruction or directive. 1254 1255 // Directives start with "." 1256 if (IDVal[0] == '.' && IDVal != ".") { 1257 // There are several entities interested in parsing directives: 1258 // 1259 // 1. The target-specific assembly parser. Some directives are target 1260 // specific or may potentially behave differently on certain targets. 1261 // 2. Asm parser extensions. For example, platform-specific parsers 1262 // (like the ELF parser) register themselves as extensions. 1263 // 3. The generic directive parser implemented by this class. These are 1264 // all the directives that behave in a target and platform independent 1265 // manner, or at least have a default behavior that's shared between 1266 // all targets and platforms. 1267 1268 // First query the target-specific parser. It will return 'true' if it 1269 // isn't interested in this directive. 1270 if (!getTargetParser().ParseDirective(ID)) 1271 return false; 1272 1273 // Next, check the extention directive map to see if any extension has 1274 // registered itself to parse this directive. 1275 std::pair<MCAsmParserExtension*, DirectiveHandler> Handler = 1276 ExtensionDirectiveMap.lookup(IDVal); 1277 if (Handler.first) 1278 return (*Handler.second)(Handler.first, IDVal, IDLoc); 1279 1280 // Finally, if no one else is interested in this directive, it must be 1281 // generic and familiar to this class. 1282 switch (DirKind) { 1283 default: 1284 break; 1285 case DK_SET: 1286 case DK_EQU: 1287 return ParseDirectiveSet(IDVal, true); 1288 case DK_EQUIV: 1289 return ParseDirectiveSet(IDVal, false); 1290 case DK_ASCII: 1291 return ParseDirectiveAscii(IDVal, false); 1292 case DK_ASCIZ: 1293 case DK_STRING: 1294 return ParseDirectiveAscii(IDVal, true); 1295 case DK_BYTE: 1296 return ParseDirectiveValue(1); 1297 case DK_SHORT: 1298 case DK_VALUE: 1299 case DK_2BYTE: 1300 return ParseDirectiveValue(2); 1301 case DK_LONG: 1302 case DK_INT: 1303 case DK_4BYTE: 1304 return ParseDirectiveValue(4); 1305 case DK_QUAD: 1306 case DK_8BYTE: 1307 return ParseDirectiveValue(8); 1308 case DK_SINGLE: 1309 case DK_FLOAT: 1310 return ParseDirectiveRealValue(APFloat::IEEEsingle); 1311 case DK_DOUBLE: 1312 return ParseDirectiveRealValue(APFloat::IEEEdouble); 1313 case DK_ALIGN: { 1314 bool IsPow2 = !getContext().getAsmInfo().getAlignmentIsInBytes(); 1315 return ParseDirectiveAlign(IsPow2, /*ExprSize=*/1); 1316 } 1317 case DK_ALIGN32: { 1318 bool IsPow2 = !getContext().getAsmInfo().getAlignmentIsInBytes(); 1319 return ParseDirectiveAlign(IsPow2, /*ExprSize=*/4); 1320 } 1321 case DK_BALIGN: 1322 return ParseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/1); 1323 case DK_BALIGNW: 1324 return ParseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/2); 1325 case DK_BALIGNL: 1326 return ParseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/4); 1327 case DK_P2ALIGN: 1328 return ParseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/1); 1329 case DK_P2ALIGNW: 1330 return ParseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/2); 1331 case DK_P2ALIGNL: 1332 return ParseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/4); 1333 case DK_ORG: 1334 return ParseDirectiveOrg(); 1335 case DK_FILL: 1336 return ParseDirectiveFill(); 1337 case DK_ZERO: 1338 return ParseDirectiveZero(); 1339 case DK_EXTERN: 1340 eatToEndOfStatement(); // .extern is the default, ignore it. 1341 return false; 1342 case DK_GLOBL: 1343 case DK_GLOBAL: 1344 return ParseDirectiveSymbolAttribute(MCSA_Global); 1345 case DK_INDIRECT_SYMBOL: 1346 return ParseDirectiveSymbolAttribute(MCSA_IndirectSymbol); 1347 case DK_LAZY_REFERENCE: 1348 return ParseDirectiveSymbolAttribute(MCSA_LazyReference); 1349 case DK_NO_DEAD_STRIP: 1350 return ParseDirectiveSymbolAttribute(MCSA_NoDeadStrip); 1351 case DK_SYMBOL_RESOLVER: 1352 return ParseDirectiveSymbolAttribute(MCSA_SymbolResolver); 1353 case DK_PRIVATE_EXTERN: 1354 return ParseDirectiveSymbolAttribute(MCSA_PrivateExtern); 1355 case DK_REFERENCE: 1356 return ParseDirectiveSymbolAttribute(MCSA_Reference); 1357 case DK_WEAK_DEFINITION: 1358 return ParseDirectiveSymbolAttribute(MCSA_WeakDefinition); 1359 case DK_WEAK_REFERENCE: 1360 return ParseDirectiveSymbolAttribute(MCSA_WeakReference); 1361 case DK_WEAK_DEF_CAN_BE_HIDDEN: 1362 return ParseDirectiveSymbolAttribute(MCSA_WeakDefAutoPrivate); 1363 case DK_COMM: 1364 case DK_COMMON: 1365 return ParseDirectiveComm(/*IsLocal=*/false); 1366 case DK_LCOMM: 1367 return ParseDirectiveComm(/*IsLocal=*/true); 1368 case DK_ABORT: 1369 return ParseDirectiveAbort(); 1370 case DK_INCLUDE: 1371 return ParseDirectiveInclude(); 1372 case DK_INCBIN: 1373 return ParseDirectiveIncbin(); 1374 case DK_CODE16: 1375 case DK_CODE16GCC: 1376 return TokError(Twine(IDVal) + " not supported yet"); 1377 case DK_REPT: 1378 return ParseDirectiveRept(IDLoc); 1379 case DK_IRP: 1380 return ParseDirectiveIrp(IDLoc); 1381 case DK_IRPC: 1382 return ParseDirectiveIrpc(IDLoc); 1383 case DK_ENDR: 1384 return ParseDirectiveEndr(IDLoc); 1385 case DK_BUNDLE_ALIGN_MODE: 1386 return ParseDirectiveBundleAlignMode(); 1387 case DK_BUNDLE_LOCK: 1388 return ParseDirectiveBundleLock(); 1389 case DK_BUNDLE_UNLOCK: 1390 return ParseDirectiveBundleUnlock(); 1391 case DK_SLEB128: 1392 return ParseDirectiveLEB128(true); 1393 case DK_ULEB128: 1394 return ParseDirectiveLEB128(false); 1395 case DK_SPACE: 1396 case DK_SKIP: 1397 return ParseDirectiveSpace(IDVal); 1398 case DK_FILE: 1399 return ParseDirectiveFile(IDLoc); 1400 case DK_LINE: 1401 return ParseDirectiveLine(); 1402 case DK_LOC: 1403 return ParseDirectiveLoc(); 1404 case DK_STABS: 1405 return ParseDirectiveStabs(); 1406 case DK_CFI_SECTIONS: 1407 return ParseDirectiveCFISections(); 1408 case DK_CFI_STARTPROC: 1409 return ParseDirectiveCFIStartProc(); 1410 case DK_CFI_ENDPROC: 1411 return ParseDirectiveCFIEndProc(); 1412 case DK_CFI_DEF_CFA: 1413 return ParseDirectiveCFIDefCfa(IDLoc); 1414 case DK_CFI_DEF_CFA_OFFSET: 1415 return ParseDirectiveCFIDefCfaOffset(); 1416 case DK_CFI_ADJUST_CFA_OFFSET: 1417 return ParseDirectiveCFIAdjustCfaOffset(); 1418 case DK_CFI_DEF_CFA_REGISTER: 1419 return ParseDirectiveCFIDefCfaRegister(IDLoc); 1420 case DK_CFI_OFFSET: 1421 return ParseDirectiveCFIOffset(IDLoc); 1422 case DK_CFI_REL_OFFSET: 1423 return ParseDirectiveCFIRelOffset(IDLoc); 1424 case DK_CFI_PERSONALITY: 1425 return ParseDirectiveCFIPersonalityOrLsda(true); 1426 case DK_CFI_LSDA: 1427 return ParseDirectiveCFIPersonalityOrLsda(false); 1428 case DK_CFI_REMEMBER_STATE: 1429 return ParseDirectiveCFIRememberState(); 1430 case DK_CFI_RESTORE_STATE: 1431 return ParseDirectiveCFIRestoreState(); 1432 case DK_CFI_SAME_VALUE: 1433 return ParseDirectiveCFISameValue(IDLoc); 1434 case DK_CFI_RESTORE: 1435 return ParseDirectiveCFIRestore(IDLoc); 1436 case DK_CFI_ESCAPE: 1437 return ParseDirectiveCFIEscape(); 1438 case DK_CFI_SIGNAL_FRAME: 1439 return ParseDirectiveCFISignalFrame(); 1440 case DK_CFI_UNDEFINED: 1441 return ParseDirectiveCFIUndefined(IDLoc); 1442 case DK_CFI_REGISTER: 1443 return ParseDirectiveCFIRegister(IDLoc); 1444 case DK_MACROS_ON: 1445 case DK_MACROS_OFF: 1446 return ParseDirectiveMacrosOnOff(IDVal); 1447 case DK_MACRO: 1448 return ParseDirectiveMacro(IDLoc); 1449 case DK_ENDM: 1450 case DK_ENDMACRO: 1451 return ParseDirectiveEndMacro(IDVal); 1452 case DK_PURGEM: 1453 return ParseDirectivePurgeMacro(IDLoc); 1454 } 1455 1456 return Error(IDLoc, "unknown directive"); 1457 } 1458 1459 // __asm _emit or __asm __emit 1460 if (ParsingInlineAsm && (IDVal == "_emit" || IDVal == "__emit" || 1461 IDVal == "_EMIT" || IDVal == "__EMIT")) 1462 return ParseDirectiveMSEmit(IDLoc, Info, IDVal.size()); 1463 1464 // __asm align 1465 if (ParsingInlineAsm && (IDVal == "align" || IDVal == "ALIGN")) 1466 return ParseDirectiveMSAlign(IDLoc, Info); 1467 1468 checkForValidSection(); 1469 1470 // Canonicalize the opcode to lower case. 1471 std::string OpcodeStr = IDVal.lower(); 1472 ParseInstructionInfo IInfo(Info.AsmRewrites); 1473 bool HadError = getTargetParser().ParseInstruction(IInfo, OpcodeStr, 1474 IDLoc, Info.ParsedOperands); 1475 Info.ParseError = HadError; 1476 1477 // Dump the parsed representation, if requested. 1478 if (getShowParsedOperands()) { 1479 SmallString<256> Str; 1480 raw_svector_ostream OS(Str); 1481 OS << "parsed instruction: ["; 1482 for (unsigned i = 0; i != Info.ParsedOperands.size(); ++i) { 1483 if (i != 0) 1484 OS << ", "; 1485 Info.ParsedOperands[i]->print(OS); 1486 } 1487 OS << "]"; 1488 1489 PrintMessage(IDLoc, SourceMgr::DK_Note, OS.str()); 1490 } 1491 1492 // If we are generating dwarf for assembly source files and the current 1493 // section is the initial text section then generate a .loc directive for 1494 // the instruction. 1495 if (!HadError && getContext().getGenDwarfForAssembly() && 1496 getContext().getGenDwarfSection() == 1497 getStreamer().getCurrentSection().first) { 1498 1499 unsigned Line = SrcMgr.FindLineNumber(IDLoc, CurBuffer); 1500 1501 // If we previously parsed a cpp hash file line comment then make sure the 1502 // current Dwarf File is for the CppHashFilename if not then emit the 1503 // Dwarf File table for it and adjust the line number for the .loc. 1504 const SmallVectorImpl<MCDwarfFile *> &MCDwarfFiles = 1505 getContext().getMCDwarfFiles(); 1506 if (CppHashFilename.size() != 0) { 1507 if (MCDwarfFiles[getContext().getGenDwarfFileNumber()]->getName() != 1508 CppHashFilename) 1509 getStreamer().EmitDwarfFileDirective( 1510 getContext().nextGenDwarfFileNumber(), StringRef(), CppHashFilename); 1511 1512 unsigned CppHashLocLineNo = SrcMgr.FindLineNumber(CppHashLoc,CppHashBuf); 1513 Line = CppHashLineNumber - 1 + (Line - CppHashLocLineNo); 1514 } 1515 1516 getStreamer().EmitDwarfLocDirective(getContext().getGenDwarfFileNumber(), 1517 Line, 0, DWARF2_LINE_DEFAULT_IS_STMT ? 1518 DWARF2_FLAG_IS_STMT : 0, 0, 0, 1519 StringRef()); 1520 } 1521 1522 // If parsing succeeded, match the instruction. 1523 if (!HadError) { 1524 unsigned ErrorInfo; 1525 HadError = getTargetParser().MatchAndEmitInstruction(IDLoc, Info.Opcode, 1526 Info.ParsedOperands, 1527 Out, ErrorInfo, 1528 ParsingInlineAsm); 1529 } 1530 1531 // Don't skip the rest of the line, the instruction parser is responsible for 1532 // that. 1533 return false; 1534 } 1535 1536 /// EatToEndOfLine uses the Lexer to eat the characters to the end of the line 1537 /// since they may not be able to be tokenized to get to the end of line token. 1538 void AsmParser::EatToEndOfLine() { 1539 if (!Lexer.is(AsmToken::EndOfStatement)) 1540 Lexer.LexUntilEndOfLine(); 1541 // Eat EOL. 1542 Lex(); 1543 } 1544 1545 /// ParseCppHashLineFilenameComment as this: 1546 /// ::= # number "filename" 1547 /// or just as a full line comment if it doesn't have a number and a string. 1548 bool AsmParser::ParseCppHashLineFilenameComment(const SMLoc &L) { 1549 Lex(); // Eat the hash token. 1550 1551 if (getLexer().isNot(AsmToken::Integer)) { 1552 // Consume the line since in cases it is not a well-formed line directive, 1553 // as if were simply a full line comment. 1554 EatToEndOfLine(); 1555 return false; 1556 } 1557 1558 int64_t LineNumber = getTok().getIntVal(); 1559 Lex(); 1560 1561 if (getLexer().isNot(AsmToken::String)) { 1562 EatToEndOfLine(); 1563 return false; 1564 } 1565 1566 StringRef Filename = getTok().getString(); 1567 // Get rid of the enclosing quotes. 1568 Filename = Filename.substr(1, Filename.size()-2); 1569 1570 // Save the SMLoc, Filename and LineNumber for later use by diagnostics. 1571 CppHashLoc = L; 1572 CppHashFilename = Filename; 1573 CppHashLineNumber = LineNumber; 1574 CppHashBuf = CurBuffer; 1575 1576 // Ignore any trailing characters, they're just comment. 1577 EatToEndOfLine(); 1578 return false; 1579 } 1580 1581 /// DiagHandler - will use the last parsed cpp hash line filename comment 1582 /// for the Filename and LineNo if any in the diagnostic. 1583 void AsmParser::DiagHandler(const SMDiagnostic &Diag, void *Context) { 1584 const AsmParser *Parser = static_cast<const AsmParser*>(Context); 1585 raw_ostream &OS = errs(); 1586 1587 const SourceMgr &DiagSrcMgr = *Diag.getSourceMgr(); 1588 const SMLoc &DiagLoc = Diag.getLoc(); 1589 int DiagBuf = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 1590 int CppHashBuf = Parser->SrcMgr.FindBufferContainingLoc(Parser->CppHashLoc); 1591 1592 // Like SourceMgr::PrintMessage() we need to print the include stack if any 1593 // before printing the message. 1594 int DiagCurBuffer = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 1595 if (!Parser->SavedDiagHandler && DiagCurBuffer > 0) { 1596 SMLoc ParentIncludeLoc = DiagSrcMgr.getParentIncludeLoc(DiagCurBuffer); 1597 DiagSrcMgr.PrintIncludeStack(ParentIncludeLoc, OS); 1598 } 1599 1600 // If we have not parsed a cpp hash line filename comment or the source 1601 // manager changed or buffer changed (like in a nested include) then just 1602 // print the normal diagnostic using its Filename and LineNo. 1603 if (!Parser->CppHashLineNumber || 1604 &DiagSrcMgr != &Parser->SrcMgr || 1605 DiagBuf != CppHashBuf) { 1606 if (Parser->SavedDiagHandler) 1607 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext); 1608 else 1609 Diag.print(0, OS); 1610 return; 1611 } 1612 1613 // Use the CppHashFilename and calculate a line number based on the 1614 // CppHashLoc and CppHashLineNumber relative to this Diag's SMLoc for 1615 // the diagnostic. 1616 const std::string Filename = Parser->CppHashFilename; 1617 1618 int DiagLocLineNo = DiagSrcMgr.FindLineNumber(DiagLoc, DiagBuf); 1619 int CppHashLocLineNo = 1620 Parser->SrcMgr.FindLineNumber(Parser->CppHashLoc, CppHashBuf); 1621 int LineNo = Parser->CppHashLineNumber - 1 + 1622 (DiagLocLineNo - CppHashLocLineNo); 1623 1624 SMDiagnostic NewDiag(*Diag.getSourceMgr(), Diag.getLoc(), 1625 Filename, LineNo, Diag.getColumnNo(), 1626 Diag.getKind(), Diag.getMessage(), 1627 Diag.getLineContents(), Diag.getRanges()); 1628 1629 if (Parser->SavedDiagHandler) 1630 Parser->SavedDiagHandler(NewDiag, Parser->SavedDiagContext); 1631 else 1632 NewDiag.print(0, OS); 1633 } 1634 1635 // FIXME: This is mostly duplicated from the function in AsmLexer.cpp. The 1636 // difference being that that function accepts '@' as part of identifiers and 1637 // we can't do that. AsmLexer.cpp should probably be changed to handle 1638 // '@' as a special case when needed. 1639 static bool isIdentifierChar(char c) { 1640 return isalnum(static_cast<unsigned char>(c)) || c == '_' || c == '$' || 1641 c == '.'; 1642 } 1643 1644 bool AsmParser::expandMacro(raw_svector_ostream &OS, StringRef Body, 1645 const MCAsmMacroParameters &Parameters, 1646 const MCAsmMacroArguments &A, 1647 const SMLoc &L) { 1648 unsigned NParameters = Parameters.size(); 1649 if (NParameters != 0 && NParameters != A.size()) 1650 return Error(L, "Wrong number of arguments"); 1651 1652 // A macro without parameters is handled differently on Darwin: 1653 // gas accepts no arguments and does no substitutions 1654 while (!Body.empty()) { 1655 // Scan for the next substitution. 1656 std::size_t End = Body.size(), Pos = 0; 1657 for (; Pos != End; ++Pos) { 1658 // Check for a substitution or escape. 1659 if (!NParameters) { 1660 // This macro has no parameters, look for $0, $1, etc. 1661 if (Body[Pos] != '$' || Pos + 1 == End) 1662 continue; 1663 1664 char Next = Body[Pos + 1]; 1665 if (Next == '$' || Next == 'n' || 1666 isdigit(static_cast<unsigned char>(Next))) 1667 break; 1668 } else { 1669 // This macro has parameters, look for \foo, \bar, etc. 1670 if (Body[Pos] == '\\' && Pos + 1 != End) 1671 break; 1672 } 1673 } 1674 1675 // Add the prefix. 1676 OS << Body.slice(0, Pos); 1677 1678 // Check if we reached the end. 1679 if (Pos == End) 1680 break; 1681 1682 if (!NParameters) { 1683 switch (Body[Pos+1]) { 1684 // $$ => $ 1685 case '$': 1686 OS << '$'; 1687 break; 1688 1689 // $n => number of arguments 1690 case 'n': 1691 OS << A.size(); 1692 break; 1693 1694 // $[0-9] => argument 1695 default: { 1696 // Missing arguments are ignored. 1697 unsigned Index = Body[Pos+1] - '0'; 1698 if (Index >= A.size()) 1699 break; 1700 1701 // Otherwise substitute with the token values, with spaces eliminated. 1702 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(), 1703 ie = A[Index].end(); it != ie; ++it) 1704 OS << it->getString(); 1705 break; 1706 } 1707 } 1708 Pos += 2; 1709 } else { 1710 unsigned I = Pos + 1; 1711 while (isIdentifierChar(Body[I]) && I + 1 != End) 1712 ++I; 1713 1714 const char *Begin = Body.data() + Pos +1; 1715 StringRef Argument(Begin, I - (Pos +1)); 1716 unsigned Index = 0; 1717 for (; Index < NParameters; ++Index) 1718 if (Parameters[Index].first == Argument) 1719 break; 1720 1721 if (Index == NParameters) { 1722 if (Body[Pos+1] == '(' && Body[Pos+2] == ')') 1723 Pos += 3; 1724 else { 1725 OS << '\\' << Argument; 1726 Pos = I; 1727 } 1728 } else { 1729 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(), 1730 ie = A[Index].end(); it != ie; ++it) 1731 if (it->getKind() == AsmToken::String) 1732 OS << it->getStringContents(); 1733 else 1734 OS << it->getString(); 1735 1736 Pos += 1 + Argument.size(); 1737 } 1738 } 1739 // Update the scan point. 1740 Body = Body.substr(Pos); 1741 } 1742 1743 return false; 1744 } 1745 1746 MacroInstantiation::MacroInstantiation(const MCAsmMacro *M, SMLoc IL, 1747 int EB, SMLoc EL, 1748 MemoryBuffer *I) 1749 : TheMacro(M), Instantiation(I), InstantiationLoc(IL), ExitBuffer(EB), 1750 ExitLoc(EL) 1751 { 1752 } 1753 1754 static bool IsOperator(AsmToken::TokenKind kind) 1755 { 1756 switch (kind) 1757 { 1758 default: 1759 return false; 1760 case AsmToken::Plus: 1761 case AsmToken::Minus: 1762 case AsmToken::Tilde: 1763 case AsmToken::Slash: 1764 case AsmToken::Star: 1765 case AsmToken::Dot: 1766 case AsmToken::Equal: 1767 case AsmToken::EqualEqual: 1768 case AsmToken::Pipe: 1769 case AsmToken::PipePipe: 1770 case AsmToken::Caret: 1771 case AsmToken::Amp: 1772 case AsmToken::AmpAmp: 1773 case AsmToken::Exclaim: 1774 case AsmToken::ExclaimEqual: 1775 case AsmToken::Percent: 1776 case AsmToken::Less: 1777 case AsmToken::LessEqual: 1778 case AsmToken::LessLess: 1779 case AsmToken::LessGreater: 1780 case AsmToken::Greater: 1781 case AsmToken::GreaterEqual: 1782 case AsmToken::GreaterGreater: 1783 return true; 1784 } 1785 } 1786 1787 bool AsmParser::ParseMacroArgument(MCAsmMacroArgument &MA, 1788 AsmToken::TokenKind &ArgumentDelimiter) { 1789 unsigned ParenLevel = 0; 1790 unsigned AddTokens = 0; 1791 1792 // gas accepts arguments separated by whitespace, except on Darwin 1793 if (!IsDarwin) 1794 Lexer.setSkipSpace(false); 1795 1796 for (;;) { 1797 if (Lexer.is(AsmToken::Eof) || Lexer.is(AsmToken::Equal)) { 1798 Lexer.setSkipSpace(true); 1799 return TokError("unexpected token in macro instantiation"); 1800 } 1801 1802 if (ParenLevel == 0 && Lexer.is(AsmToken::Comma)) { 1803 // Spaces and commas cannot be mixed to delimit parameters 1804 if (ArgumentDelimiter == AsmToken::Eof) 1805 ArgumentDelimiter = AsmToken::Comma; 1806 else if (ArgumentDelimiter != AsmToken::Comma) { 1807 Lexer.setSkipSpace(true); 1808 return TokError("expected ' ' for macro argument separator"); 1809 } 1810 break; 1811 } 1812 1813 if (Lexer.is(AsmToken::Space)) { 1814 Lex(); // Eat spaces 1815 1816 // Spaces can delimit parameters, but could also be part an expression. 1817 // If the token after a space is an operator, add the token and the next 1818 // one into this argument 1819 if (ArgumentDelimiter == AsmToken::Space || 1820 ArgumentDelimiter == AsmToken::Eof) { 1821 if (IsOperator(Lexer.getKind())) { 1822 // Check to see whether the token is used as an operator, 1823 // or part of an identifier 1824 const char *NextChar = getTok().getEndLoc().getPointer(); 1825 if (*NextChar == ' ') 1826 AddTokens = 2; 1827 } 1828 1829 if (!AddTokens && ParenLevel == 0) { 1830 if (ArgumentDelimiter == AsmToken::Eof && 1831 !IsOperator(Lexer.getKind())) 1832 ArgumentDelimiter = AsmToken::Space; 1833 break; 1834 } 1835 } 1836 } 1837 1838 // HandleMacroEntry relies on not advancing the lexer here 1839 // to be able to fill in the remaining default parameter values 1840 if (Lexer.is(AsmToken::EndOfStatement)) 1841 break; 1842 1843 // Adjust the current parentheses level. 1844 if (Lexer.is(AsmToken::LParen)) 1845 ++ParenLevel; 1846 else if (Lexer.is(AsmToken::RParen) && ParenLevel) 1847 --ParenLevel; 1848 1849 // Append the token to the current argument list. 1850 MA.push_back(getTok()); 1851 if (AddTokens) 1852 AddTokens--; 1853 Lex(); 1854 } 1855 1856 Lexer.setSkipSpace(true); 1857 if (ParenLevel != 0) 1858 return TokError("unbalanced parentheses in macro argument"); 1859 return false; 1860 } 1861 1862 // Parse the macro instantiation arguments. 1863 bool AsmParser::ParseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A) { 1864 const unsigned NParameters = M ? M->Parameters.size() : 0; 1865 // Argument delimiter is initially unknown. It will be set by 1866 // ParseMacroArgument() 1867 AsmToken::TokenKind ArgumentDelimiter = AsmToken::Eof; 1868 1869 // Parse two kinds of macro invocations: 1870 // - macros defined without any parameters accept an arbitrary number of them 1871 // - macros defined with parameters accept at most that many of them 1872 for (unsigned Parameter = 0; !NParameters || Parameter < NParameters; 1873 ++Parameter) { 1874 MCAsmMacroArgument MA; 1875 1876 if (ParseMacroArgument(MA, ArgumentDelimiter)) 1877 return true; 1878 1879 if (!MA.empty() || !NParameters) 1880 A.push_back(MA); 1881 else if (NParameters) { 1882 if (!M->Parameters[Parameter].second.empty()) 1883 A.push_back(M->Parameters[Parameter].second); 1884 } 1885 1886 // At the end of the statement, fill in remaining arguments that have 1887 // default values. If there aren't any, then the next argument is 1888 // required but missing 1889 if (Lexer.is(AsmToken::EndOfStatement)) { 1890 if (NParameters && Parameter < NParameters - 1) { 1891 if (M->Parameters[Parameter + 1].second.empty()) 1892 return TokError("macro argument '" + 1893 Twine(M->Parameters[Parameter + 1].first) + 1894 "' is missing"); 1895 else 1896 continue; 1897 } 1898 return false; 1899 } 1900 1901 if (Lexer.is(AsmToken::Comma)) 1902 Lex(); 1903 } 1904 return TokError("Too many arguments"); 1905 } 1906 1907 const MCAsmMacro* AsmParser::LookupMacro(StringRef Name) { 1908 StringMap<MCAsmMacro*>::iterator I = MacroMap.find(Name); 1909 return (I == MacroMap.end()) ? NULL : I->getValue(); 1910 } 1911 1912 void AsmParser::DefineMacro(StringRef Name, const MCAsmMacro& Macro) { 1913 MacroMap[Name] = new MCAsmMacro(Macro); 1914 } 1915 1916 void AsmParser::UndefineMacro(StringRef Name) { 1917 StringMap<MCAsmMacro*>::iterator I = MacroMap.find(Name); 1918 if (I != MacroMap.end()) { 1919 delete I->getValue(); 1920 MacroMap.erase(I); 1921 } 1922 } 1923 1924 bool AsmParser::HandleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc) { 1925 // Arbitrarily limit macro nesting depth, to match 'as'. We can eliminate 1926 // this, although we should protect against infinite loops. 1927 if (ActiveMacros.size() == 20) 1928 return TokError("macros cannot be nested more than 20 levels deep"); 1929 1930 MCAsmMacroArguments A; 1931 if (ParseMacroArguments(M, A)) 1932 return true; 1933 1934 // Remove any trailing empty arguments. Do this after-the-fact as we have 1935 // to keep empty arguments in the middle of the list or positionality 1936 // gets off. e.g., "foo 1, , 2" vs. "foo 1, 2," 1937 while (!A.empty() && A.back().empty()) 1938 A.pop_back(); 1939 1940 // Macro instantiation is lexical, unfortunately. We construct a new buffer 1941 // to hold the macro body with substitutions. 1942 SmallString<256> Buf; 1943 StringRef Body = M->Body; 1944 raw_svector_ostream OS(Buf); 1945 1946 if (expandMacro(OS, Body, M->Parameters, A, getTok().getLoc())) 1947 return true; 1948 1949 // We include the .endmacro in the buffer as our cue to exit the macro 1950 // instantiation. 1951 OS << ".endmacro\n"; 1952 1953 MemoryBuffer *Instantiation = 1954 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 1955 1956 // Create the macro instantiation object and add to the current macro 1957 // instantiation stack. 1958 MacroInstantiation *MI = new MacroInstantiation(M, NameLoc, 1959 CurBuffer, 1960 getTok().getLoc(), 1961 Instantiation); 1962 ActiveMacros.push_back(MI); 1963 1964 // Jump to the macro instantiation and prime the lexer. 1965 CurBuffer = SrcMgr.AddNewSourceBuffer(MI->Instantiation, SMLoc()); 1966 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)); 1967 Lex(); 1968 1969 return false; 1970 } 1971 1972 void AsmParser::HandleMacroExit() { 1973 // Jump to the EndOfStatement we should return to, and consume it. 1974 JumpToLoc(ActiveMacros.back()->ExitLoc, ActiveMacros.back()->ExitBuffer); 1975 Lex(); 1976 1977 // Pop the instantiation entry. 1978 delete ActiveMacros.back(); 1979 ActiveMacros.pop_back(); 1980 } 1981 1982 static bool IsUsedIn(const MCSymbol *Sym, const MCExpr *Value) { 1983 switch (Value->getKind()) { 1984 case MCExpr::Binary: { 1985 const MCBinaryExpr *BE = static_cast<const MCBinaryExpr*>(Value); 1986 return IsUsedIn(Sym, BE->getLHS()) || IsUsedIn(Sym, BE->getRHS()); 1987 } 1988 case MCExpr::Target: 1989 case MCExpr::Constant: 1990 return false; 1991 case MCExpr::SymbolRef: { 1992 const MCSymbol &S = static_cast<const MCSymbolRefExpr*>(Value)->getSymbol(); 1993 if (S.isVariable()) 1994 return IsUsedIn(Sym, S.getVariableValue()); 1995 return &S == Sym; 1996 } 1997 case MCExpr::Unary: 1998 return IsUsedIn(Sym, static_cast<const MCUnaryExpr*>(Value)->getSubExpr()); 1999 } 2000 2001 llvm_unreachable("Unknown expr kind!"); 2002 } 2003 2004 bool AsmParser::ParseAssignment(StringRef Name, bool allow_redef, 2005 bool NoDeadStrip) { 2006 // FIXME: Use better location, we should use proper tokens. 2007 SMLoc EqualLoc = Lexer.getLoc(); 2008 2009 const MCExpr *Value; 2010 if (parseExpression(Value)) 2011 return true; 2012 2013 // Note: we don't count b as used in "a = b". This is to allow 2014 // a = b 2015 // b = c 2016 2017 if (Lexer.isNot(AsmToken::EndOfStatement)) 2018 return TokError("unexpected token in assignment"); 2019 2020 // Error on assignment to '.'. 2021 if (Name == ".") { 2022 return Error(EqualLoc, ("assignment to pseudo-symbol '.' is unsupported " 2023 "(use '.space' or '.org').)")); 2024 } 2025 2026 // Eat the end of statement marker. 2027 Lex(); 2028 2029 // Validate that the LHS is allowed to be a variable (either it has not been 2030 // used as a symbol, or it is an absolute symbol). 2031 MCSymbol *Sym = getContext().LookupSymbol(Name); 2032 if (Sym) { 2033 // Diagnose assignment to a label. 2034 // 2035 // FIXME: Diagnostics. Note the location of the definition as a label. 2036 // FIXME: Diagnose assignment to protected identifier (e.g., register name). 2037 if (IsUsedIn(Sym, Value)) 2038 return Error(EqualLoc, "Recursive use of '" + Name + "'"); 2039 else if (Sym->isUndefined() && !Sym->isUsed() && !Sym->isVariable()) 2040 ; // Allow redefinitions of undefined symbols only used in directives. 2041 else if (Sym->isVariable() && !Sym->isUsed() && allow_redef) 2042 ; // Allow redefinitions of variables that haven't yet been used. 2043 else if (!Sym->isUndefined() && (!Sym->isVariable() || !allow_redef)) 2044 return Error(EqualLoc, "redefinition of '" + Name + "'"); 2045 else if (!Sym->isVariable()) 2046 return Error(EqualLoc, "invalid assignment to '" + Name + "'"); 2047 else if (!isa<MCConstantExpr>(Sym->getVariableValue())) 2048 return Error(EqualLoc, "invalid reassignment of non-absolute variable '" + 2049 Name + "'"); 2050 2051 // Don't count these checks as uses. 2052 Sym->setUsed(false); 2053 } else 2054 Sym = getContext().GetOrCreateSymbol(Name); 2055 2056 // FIXME: Handle '.'. 2057 2058 // Do the assignment. 2059 Out.EmitAssignment(Sym, Value); 2060 if (NoDeadStrip) 2061 Out.EmitSymbolAttribute(Sym, MCSA_NoDeadStrip); 2062 2063 2064 return false; 2065 } 2066 2067 /// parseIdentifier: 2068 /// ::= identifier 2069 /// ::= string 2070 bool AsmParser::parseIdentifier(StringRef &Res) { 2071 // The assembler has relaxed rules for accepting identifiers, in particular we 2072 // allow things like '.globl $foo', which would normally be separate 2073 // tokens. At this level, we have already lexed so we cannot (currently) 2074 // handle this as a context dependent token, instead we detect adjacent tokens 2075 // and return the combined identifier. 2076 if (Lexer.is(AsmToken::Dollar)) { 2077 SMLoc DollarLoc = getLexer().getLoc(); 2078 2079 // Consume the dollar sign, and check for a following identifier. 2080 Lex(); 2081 if (Lexer.isNot(AsmToken::Identifier)) 2082 return true; 2083 2084 // We have a '$' followed by an identifier, make sure they are adjacent. 2085 if (DollarLoc.getPointer() + 1 != getTok().getLoc().getPointer()) 2086 return true; 2087 2088 // Construct the joined identifier and consume the token. 2089 Res = StringRef(DollarLoc.getPointer(), 2090 getTok().getIdentifier().size() + 1); 2091 Lex(); 2092 return false; 2093 } 2094 2095 if (Lexer.isNot(AsmToken::Identifier) && 2096 Lexer.isNot(AsmToken::String)) 2097 return true; 2098 2099 Res = getTok().getIdentifier(); 2100 2101 Lex(); // Consume the identifier token. 2102 2103 return false; 2104 } 2105 2106 /// ParseDirectiveSet: 2107 /// ::= .equ identifier ',' expression 2108 /// ::= .equiv identifier ',' expression 2109 /// ::= .set identifier ',' expression 2110 bool AsmParser::ParseDirectiveSet(StringRef IDVal, bool allow_redef) { 2111 StringRef Name; 2112 2113 if (parseIdentifier(Name)) 2114 return TokError("expected identifier after '" + Twine(IDVal) + "'"); 2115 2116 if (getLexer().isNot(AsmToken::Comma)) 2117 return TokError("unexpected token in '" + Twine(IDVal) + "'"); 2118 Lex(); 2119 2120 return ParseAssignment(Name, allow_redef, true); 2121 } 2122 2123 bool AsmParser::parseEscapedString(std::string &Data) { 2124 assert(getLexer().is(AsmToken::String) && "Unexpected current token!"); 2125 2126 Data = ""; 2127 StringRef Str = getTok().getStringContents(); 2128 for (unsigned i = 0, e = Str.size(); i != e; ++i) { 2129 if (Str[i] != '\\') { 2130 Data += Str[i]; 2131 continue; 2132 } 2133 2134 // Recognize escaped characters. Note that this escape semantics currently 2135 // loosely follows Darwin 'as'. Notably, it doesn't support hex escapes. 2136 ++i; 2137 if (i == e) 2138 return TokError("unexpected backslash at end of string"); 2139 2140 // Recognize octal sequences. 2141 if ((unsigned) (Str[i] - '0') <= 7) { 2142 // Consume up to three octal characters. 2143 unsigned Value = Str[i] - '0'; 2144 2145 if (i + 1 != e && ((unsigned) (Str[i + 1] - '0')) <= 7) { 2146 ++i; 2147 Value = Value * 8 + (Str[i] - '0'); 2148 2149 if (i + 1 != e && ((unsigned) (Str[i + 1] - '0')) <= 7) { 2150 ++i; 2151 Value = Value * 8 + (Str[i] - '0'); 2152 } 2153 } 2154 2155 if (Value > 255) 2156 return TokError("invalid octal escape sequence (out of range)"); 2157 2158 Data += (unsigned char) Value; 2159 continue; 2160 } 2161 2162 // Otherwise recognize individual escapes. 2163 switch (Str[i]) { 2164 default: 2165 // Just reject invalid escape sequences for now. 2166 return TokError("invalid escape sequence (unrecognized character)"); 2167 2168 case 'b': Data += '\b'; break; 2169 case 'f': Data += '\f'; break; 2170 case 'n': Data += '\n'; break; 2171 case 'r': Data += '\r'; break; 2172 case 't': Data += '\t'; break; 2173 case '"': Data += '"'; break; 2174 case '\\': Data += '\\'; break; 2175 } 2176 } 2177 2178 return false; 2179 } 2180 2181 /// ParseDirectiveAscii: 2182 /// ::= ( .ascii | .asciz | .string ) [ "string" ( , "string" )* ] 2183 bool AsmParser::ParseDirectiveAscii(StringRef IDVal, bool ZeroTerminated) { 2184 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2185 checkForValidSection(); 2186 2187 for (;;) { 2188 if (getLexer().isNot(AsmToken::String)) 2189 return TokError("expected string in '" + Twine(IDVal) + "' directive"); 2190 2191 std::string Data; 2192 if (parseEscapedString(Data)) 2193 return true; 2194 2195 getStreamer().EmitBytes(Data, DEFAULT_ADDRSPACE); 2196 if (ZeroTerminated) 2197 getStreamer().EmitBytes(StringRef("\0", 1), DEFAULT_ADDRSPACE); 2198 2199 Lex(); 2200 2201 if (getLexer().is(AsmToken::EndOfStatement)) 2202 break; 2203 2204 if (getLexer().isNot(AsmToken::Comma)) 2205 return TokError("unexpected token in '" + Twine(IDVal) + "' directive"); 2206 Lex(); 2207 } 2208 } 2209 2210 Lex(); 2211 return false; 2212 } 2213 2214 /// ParseDirectiveValue 2215 /// ::= (.byte | .short | ... ) [ expression (, expression)* ] 2216 bool AsmParser::ParseDirectiveValue(unsigned Size) { 2217 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2218 checkForValidSection(); 2219 2220 for (;;) { 2221 const MCExpr *Value; 2222 SMLoc ExprLoc = getLexer().getLoc(); 2223 if (parseExpression(Value)) 2224 return true; 2225 2226 // Special case constant expressions to match code generator. 2227 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2228 assert(Size <= 8 && "Invalid size"); 2229 uint64_t IntValue = MCE->getValue(); 2230 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue)) 2231 return Error(ExprLoc, "literal value out of range for directive"); 2232 getStreamer().EmitIntValue(IntValue, Size, DEFAULT_ADDRSPACE); 2233 } else 2234 getStreamer().EmitValue(Value, Size, DEFAULT_ADDRSPACE); 2235 2236 if (getLexer().is(AsmToken::EndOfStatement)) 2237 break; 2238 2239 // FIXME: Improve diagnostic. 2240 if (getLexer().isNot(AsmToken::Comma)) 2241 return TokError("unexpected token in directive"); 2242 Lex(); 2243 } 2244 } 2245 2246 Lex(); 2247 return false; 2248 } 2249 2250 /// ParseDirectiveRealValue 2251 /// ::= (.single | .double) [ expression (, expression)* ] 2252 bool AsmParser::ParseDirectiveRealValue(const fltSemantics &Semantics) { 2253 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2254 checkForValidSection(); 2255 2256 for (;;) { 2257 // We don't truly support arithmetic on floating point expressions, so we 2258 // have to manually parse unary prefixes. 2259 bool IsNeg = false; 2260 if (getLexer().is(AsmToken::Minus)) { 2261 Lex(); 2262 IsNeg = true; 2263 } else if (getLexer().is(AsmToken::Plus)) 2264 Lex(); 2265 2266 if (getLexer().isNot(AsmToken::Integer) && 2267 getLexer().isNot(AsmToken::Real) && 2268 getLexer().isNot(AsmToken::Identifier)) 2269 return TokError("unexpected token in directive"); 2270 2271 // Convert to an APFloat. 2272 APFloat Value(Semantics); 2273 StringRef IDVal = getTok().getString(); 2274 if (getLexer().is(AsmToken::Identifier)) { 2275 if (!IDVal.compare_lower("infinity") || !IDVal.compare_lower("inf")) 2276 Value = APFloat::getInf(Semantics); 2277 else if (!IDVal.compare_lower("nan")) 2278 Value = APFloat::getNaN(Semantics, false, ~0); 2279 else 2280 return TokError("invalid floating point literal"); 2281 } else if (Value.convertFromString(IDVal, APFloat::rmNearestTiesToEven) == 2282 APFloat::opInvalidOp) 2283 return TokError("invalid floating point literal"); 2284 if (IsNeg) 2285 Value.changeSign(); 2286 2287 // Consume the numeric token. 2288 Lex(); 2289 2290 // Emit the value as an integer. 2291 APInt AsInt = Value.bitcastToAPInt(); 2292 getStreamer().EmitIntValue(AsInt.getLimitedValue(), 2293 AsInt.getBitWidth() / 8, DEFAULT_ADDRSPACE); 2294 2295 if (getLexer().is(AsmToken::EndOfStatement)) 2296 break; 2297 2298 if (getLexer().isNot(AsmToken::Comma)) 2299 return TokError("unexpected token in directive"); 2300 Lex(); 2301 } 2302 } 2303 2304 Lex(); 2305 return false; 2306 } 2307 2308 /// ParseDirectiveZero 2309 /// ::= .zero expression 2310 bool AsmParser::ParseDirectiveZero() { 2311 checkForValidSection(); 2312 2313 int64_t NumBytes; 2314 if (parseAbsoluteExpression(NumBytes)) 2315 return true; 2316 2317 int64_t Val = 0; 2318 if (getLexer().is(AsmToken::Comma)) { 2319 Lex(); 2320 if (parseAbsoluteExpression(Val)) 2321 return true; 2322 } 2323 2324 if (getLexer().isNot(AsmToken::EndOfStatement)) 2325 return TokError("unexpected token in '.zero' directive"); 2326 2327 Lex(); 2328 2329 getStreamer().EmitFill(NumBytes, Val, DEFAULT_ADDRSPACE); 2330 2331 return false; 2332 } 2333 2334 /// ParseDirectiveFill 2335 /// ::= .fill expression , expression , expression 2336 bool AsmParser::ParseDirectiveFill() { 2337 checkForValidSection(); 2338 2339 int64_t NumValues; 2340 if (parseAbsoluteExpression(NumValues)) 2341 return true; 2342 2343 if (getLexer().isNot(AsmToken::Comma)) 2344 return TokError("unexpected token in '.fill' directive"); 2345 Lex(); 2346 2347 int64_t FillSize; 2348 if (parseAbsoluteExpression(FillSize)) 2349 return true; 2350 2351 if (getLexer().isNot(AsmToken::Comma)) 2352 return TokError("unexpected token in '.fill' directive"); 2353 Lex(); 2354 2355 int64_t FillExpr; 2356 if (parseAbsoluteExpression(FillExpr)) 2357 return true; 2358 2359 if (getLexer().isNot(AsmToken::EndOfStatement)) 2360 return TokError("unexpected token in '.fill' directive"); 2361 2362 Lex(); 2363 2364 if (FillSize != 1 && FillSize != 2 && FillSize != 4 && FillSize != 8) 2365 return TokError("invalid '.fill' size, expected 1, 2, 4, or 8"); 2366 2367 for (uint64_t i = 0, e = NumValues; i != e; ++i) 2368 getStreamer().EmitIntValue(FillExpr, FillSize, DEFAULT_ADDRSPACE); 2369 2370 return false; 2371 } 2372 2373 /// ParseDirectiveOrg 2374 /// ::= .org expression [ , expression ] 2375 bool AsmParser::ParseDirectiveOrg() { 2376 checkForValidSection(); 2377 2378 const MCExpr *Offset; 2379 SMLoc Loc = getTok().getLoc(); 2380 if (parseExpression(Offset)) 2381 return true; 2382 2383 // Parse optional fill expression. 2384 int64_t FillExpr = 0; 2385 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2386 if (getLexer().isNot(AsmToken::Comma)) 2387 return TokError("unexpected token in '.org' directive"); 2388 Lex(); 2389 2390 if (parseAbsoluteExpression(FillExpr)) 2391 return true; 2392 2393 if (getLexer().isNot(AsmToken::EndOfStatement)) 2394 return TokError("unexpected token in '.org' directive"); 2395 } 2396 2397 Lex(); 2398 2399 // Only limited forms of relocatable expressions are accepted here, it 2400 // has to be relative to the current section. The streamer will return 2401 // 'true' if the expression wasn't evaluatable. 2402 if (getStreamer().EmitValueToOffset(Offset, FillExpr)) 2403 return Error(Loc, "expected assembly-time absolute expression"); 2404 2405 return false; 2406 } 2407 2408 /// ParseDirectiveAlign 2409 /// ::= {.align, ...} expression [ , expression [ , expression ]] 2410 bool AsmParser::ParseDirectiveAlign(bool IsPow2, unsigned ValueSize) { 2411 checkForValidSection(); 2412 2413 SMLoc AlignmentLoc = getLexer().getLoc(); 2414 int64_t Alignment; 2415 if (parseAbsoluteExpression(Alignment)) 2416 return true; 2417 2418 SMLoc MaxBytesLoc; 2419 bool HasFillExpr = false; 2420 int64_t FillExpr = 0; 2421 int64_t MaxBytesToFill = 0; 2422 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2423 if (getLexer().isNot(AsmToken::Comma)) 2424 return TokError("unexpected token in directive"); 2425 Lex(); 2426 2427 // The fill expression can be omitted while specifying a maximum number of 2428 // alignment bytes, e.g: 2429 // .align 3,,4 2430 if (getLexer().isNot(AsmToken::Comma)) { 2431 HasFillExpr = true; 2432 if (parseAbsoluteExpression(FillExpr)) 2433 return true; 2434 } 2435 2436 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2437 if (getLexer().isNot(AsmToken::Comma)) 2438 return TokError("unexpected token in directive"); 2439 Lex(); 2440 2441 MaxBytesLoc = getLexer().getLoc(); 2442 if (parseAbsoluteExpression(MaxBytesToFill)) 2443 return true; 2444 2445 if (getLexer().isNot(AsmToken::EndOfStatement)) 2446 return TokError("unexpected token in directive"); 2447 } 2448 } 2449 2450 Lex(); 2451 2452 if (!HasFillExpr) 2453 FillExpr = 0; 2454 2455 // Compute alignment in bytes. 2456 if (IsPow2) { 2457 // FIXME: Diagnose overflow. 2458 if (Alignment >= 32) { 2459 Error(AlignmentLoc, "invalid alignment value"); 2460 Alignment = 31; 2461 } 2462 2463 Alignment = 1ULL << Alignment; 2464 } else { 2465 // Reject alignments that aren't a power of two, for gas compatibility. 2466 if (!isPowerOf2_64(Alignment)) 2467 Error(AlignmentLoc, "alignment must be a power of 2"); 2468 } 2469 2470 // Diagnose non-sensical max bytes to align. 2471 if (MaxBytesLoc.isValid()) { 2472 if (MaxBytesToFill < 1) { 2473 Error(MaxBytesLoc, "alignment directive can never be satisfied in this " 2474 "many bytes, ignoring maximum bytes expression"); 2475 MaxBytesToFill = 0; 2476 } 2477 2478 if (MaxBytesToFill >= Alignment) { 2479 Warning(MaxBytesLoc, "maximum bytes expression exceeds alignment and " 2480 "has no effect"); 2481 MaxBytesToFill = 0; 2482 } 2483 } 2484 2485 // Check whether we should use optimal code alignment for this .align 2486 // directive. 2487 bool UseCodeAlign = getStreamer().getCurrentSection().first->UseCodeAlign(); 2488 if ((!HasFillExpr || Lexer.getMAI().getTextAlignFillValue() == FillExpr) && 2489 ValueSize == 1 && UseCodeAlign) { 2490 getStreamer().EmitCodeAlignment(Alignment, MaxBytesToFill); 2491 } else { 2492 // FIXME: Target specific behavior about how the "extra" bytes are filled. 2493 getStreamer().EmitValueToAlignment(Alignment, FillExpr, ValueSize, 2494 MaxBytesToFill); 2495 } 2496 2497 return false; 2498 } 2499 2500 /// ParseDirectiveFile 2501 /// ::= .file [number] filename 2502 /// ::= .file number directory filename 2503 bool AsmParser::ParseDirectiveFile(SMLoc DirectiveLoc) { 2504 // FIXME: I'm not sure what this is. 2505 int64_t FileNumber = -1; 2506 SMLoc FileNumberLoc = getLexer().getLoc(); 2507 if (getLexer().is(AsmToken::Integer)) { 2508 FileNumber = getTok().getIntVal(); 2509 Lex(); 2510 2511 if (FileNumber < 1) 2512 return TokError("file number less than one"); 2513 } 2514 2515 if (getLexer().isNot(AsmToken::String)) 2516 return TokError("unexpected token in '.file' directive"); 2517 2518 // Usually the directory and filename together, otherwise just the directory. 2519 StringRef Path = getTok().getString(); 2520 Path = Path.substr(1, Path.size()-2); 2521 Lex(); 2522 2523 StringRef Directory; 2524 StringRef Filename; 2525 if (getLexer().is(AsmToken::String)) { 2526 if (FileNumber == -1) 2527 return TokError("explicit path specified, but no file number"); 2528 Filename = getTok().getString(); 2529 Filename = Filename.substr(1, Filename.size()-2); 2530 Directory = Path; 2531 Lex(); 2532 } else { 2533 Filename = Path; 2534 } 2535 2536 if (getLexer().isNot(AsmToken::EndOfStatement)) 2537 return TokError("unexpected token in '.file' directive"); 2538 2539 if (FileNumber == -1) 2540 getStreamer().EmitFileDirective(Filename); 2541 else { 2542 if (getContext().getGenDwarfForAssembly() == true) 2543 Error(DirectiveLoc, "input can't have .file dwarf directives when -g is " 2544 "used to generate dwarf debug info for assembly code"); 2545 2546 if (getStreamer().EmitDwarfFileDirective(FileNumber, Directory, Filename)) 2547 Error(FileNumberLoc, "file number already allocated"); 2548 } 2549 2550 return false; 2551 } 2552 2553 /// ParseDirectiveLine 2554 /// ::= .line [number] 2555 bool AsmParser::ParseDirectiveLine() { 2556 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2557 if (getLexer().isNot(AsmToken::Integer)) 2558 return TokError("unexpected token in '.line' directive"); 2559 2560 int64_t LineNumber = getTok().getIntVal(); 2561 (void) LineNumber; 2562 Lex(); 2563 2564 // FIXME: Do something with the .line. 2565 } 2566 2567 if (getLexer().isNot(AsmToken::EndOfStatement)) 2568 return TokError("unexpected token in '.line' directive"); 2569 2570 return false; 2571 } 2572 2573 /// ParseDirectiveLoc 2574 /// ::= .loc FileNumber [LineNumber] [ColumnPos] [basic_block] [prologue_end] 2575 /// [epilogue_begin] [is_stmt VALUE] [isa VALUE] 2576 /// The first number is a file number, must have been previously assigned with 2577 /// a .file directive, the second number is the line number and optionally the 2578 /// third number is a column position (zero if not specified). The remaining 2579 /// optional items are .loc sub-directives. 2580 bool AsmParser::ParseDirectiveLoc() { 2581 if (getLexer().isNot(AsmToken::Integer)) 2582 return TokError("unexpected token in '.loc' directive"); 2583 int64_t FileNumber = getTok().getIntVal(); 2584 if (FileNumber < 1) 2585 return TokError("file number less than one in '.loc' directive"); 2586 if (!getContext().isValidDwarfFileNumber(FileNumber)) 2587 return TokError("unassigned file number in '.loc' directive"); 2588 Lex(); 2589 2590 int64_t LineNumber = 0; 2591 if (getLexer().is(AsmToken::Integer)) { 2592 LineNumber = getTok().getIntVal(); 2593 if (LineNumber < 1) 2594 return TokError("line number less than one in '.loc' directive"); 2595 Lex(); 2596 } 2597 2598 int64_t ColumnPos = 0; 2599 if (getLexer().is(AsmToken::Integer)) { 2600 ColumnPos = getTok().getIntVal(); 2601 if (ColumnPos < 0) 2602 return TokError("column position less than zero in '.loc' directive"); 2603 Lex(); 2604 } 2605 2606 unsigned Flags = DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0; 2607 unsigned Isa = 0; 2608 int64_t Discriminator = 0; 2609 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2610 for (;;) { 2611 if (getLexer().is(AsmToken::EndOfStatement)) 2612 break; 2613 2614 StringRef Name; 2615 SMLoc Loc = getTok().getLoc(); 2616 if (parseIdentifier(Name)) 2617 return TokError("unexpected token in '.loc' directive"); 2618 2619 if (Name == "basic_block") 2620 Flags |= DWARF2_FLAG_BASIC_BLOCK; 2621 else if (Name == "prologue_end") 2622 Flags |= DWARF2_FLAG_PROLOGUE_END; 2623 else if (Name == "epilogue_begin") 2624 Flags |= DWARF2_FLAG_EPILOGUE_BEGIN; 2625 else if (Name == "is_stmt") { 2626 Loc = getTok().getLoc(); 2627 const MCExpr *Value; 2628 if (parseExpression(Value)) 2629 return true; 2630 // The expression must be the constant 0 or 1. 2631 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2632 int Value = MCE->getValue(); 2633 if (Value == 0) 2634 Flags &= ~DWARF2_FLAG_IS_STMT; 2635 else if (Value == 1) 2636 Flags |= DWARF2_FLAG_IS_STMT; 2637 else 2638 return Error(Loc, "is_stmt value not 0 or 1"); 2639 } else { 2640 return Error(Loc, "is_stmt value not the constant value of 0 or 1"); 2641 } 2642 } else if (Name == "isa") { 2643 Loc = getTok().getLoc(); 2644 const MCExpr *Value; 2645 if (parseExpression(Value)) 2646 return true; 2647 // The expression must be a constant greater or equal to 0. 2648 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2649 int Value = MCE->getValue(); 2650 if (Value < 0) 2651 return Error(Loc, "isa number less than zero"); 2652 Isa = Value; 2653 } else { 2654 return Error(Loc, "isa number not a constant value"); 2655 } 2656 } else if (Name == "discriminator") { 2657 if (parseAbsoluteExpression(Discriminator)) 2658 return true; 2659 } else { 2660 return Error(Loc, "unknown sub-directive in '.loc' directive"); 2661 } 2662 2663 if (getLexer().is(AsmToken::EndOfStatement)) 2664 break; 2665 } 2666 } 2667 2668 getStreamer().EmitDwarfLocDirective(FileNumber, LineNumber, ColumnPos, Flags, 2669 Isa, Discriminator, StringRef()); 2670 2671 return false; 2672 } 2673 2674 /// ParseDirectiveStabs 2675 /// ::= .stabs string, number, number, number 2676 bool AsmParser::ParseDirectiveStabs() { 2677 return TokError("unsupported directive '.stabs'"); 2678 } 2679 2680 /// ParseDirectiveCFISections 2681 /// ::= .cfi_sections section [, section] 2682 bool AsmParser::ParseDirectiveCFISections() { 2683 StringRef Name; 2684 bool EH = false; 2685 bool Debug = false; 2686 2687 if (parseIdentifier(Name)) 2688 return TokError("Expected an identifier"); 2689 2690 if (Name == ".eh_frame") 2691 EH = true; 2692 else if (Name == ".debug_frame") 2693 Debug = true; 2694 2695 if (getLexer().is(AsmToken::Comma)) { 2696 Lex(); 2697 2698 if (parseIdentifier(Name)) 2699 return TokError("Expected an identifier"); 2700 2701 if (Name == ".eh_frame") 2702 EH = true; 2703 else if (Name == ".debug_frame") 2704 Debug = true; 2705 } 2706 2707 getStreamer().EmitCFISections(EH, Debug); 2708 return false; 2709 } 2710 2711 /// ParseDirectiveCFIStartProc 2712 /// ::= .cfi_startproc 2713 bool AsmParser::ParseDirectiveCFIStartProc() { 2714 getStreamer().EmitCFIStartProc(); 2715 return false; 2716 } 2717 2718 /// ParseDirectiveCFIEndProc 2719 /// ::= .cfi_endproc 2720 bool AsmParser::ParseDirectiveCFIEndProc() { 2721 getStreamer().EmitCFIEndProc(); 2722 return false; 2723 } 2724 2725 /// ParseRegisterOrRegisterNumber - parse register name or number. 2726 bool AsmParser::ParseRegisterOrRegisterNumber(int64_t &Register, 2727 SMLoc DirectiveLoc) { 2728 unsigned RegNo; 2729 2730 if (getLexer().isNot(AsmToken::Integer)) { 2731 if (getTargetParser().ParseRegister(RegNo, DirectiveLoc, DirectiveLoc)) 2732 return true; 2733 Register = getContext().getRegisterInfo().getDwarfRegNum(RegNo, true); 2734 } else 2735 return parseAbsoluteExpression(Register); 2736 2737 return false; 2738 } 2739 2740 /// ParseDirectiveCFIDefCfa 2741 /// ::= .cfi_def_cfa register, offset 2742 bool AsmParser::ParseDirectiveCFIDefCfa(SMLoc DirectiveLoc) { 2743 int64_t Register = 0; 2744 if (ParseRegisterOrRegisterNumber(Register, DirectiveLoc)) 2745 return true; 2746 2747 if (getLexer().isNot(AsmToken::Comma)) 2748 return TokError("unexpected token in directive"); 2749 Lex(); 2750 2751 int64_t Offset = 0; 2752 if (parseAbsoluteExpression(Offset)) 2753 return true; 2754 2755 getStreamer().EmitCFIDefCfa(Register, Offset); 2756 return false; 2757 } 2758 2759 /// ParseDirectiveCFIDefCfaOffset 2760 /// ::= .cfi_def_cfa_offset offset 2761 bool AsmParser::ParseDirectiveCFIDefCfaOffset() { 2762 int64_t Offset = 0; 2763 if (parseAbsoluteExpression(Offset)) 2764 return true; 2765 2766 getStreamer().EmitCFIDefCfaOffset(Offset); 2767 return false; 2768 } 2769 2770 /// ParseDirectiveCFIRegister 2771 /// ::= .cfi_register register, register 2772 bool AsmParser::ParseDirectiveCFIRegister(SMLoc DirectiveLoc) { 2773 int64_t Register1 = 0; 2774 if (ParseRegisterOrRegisterNumber(Register1, DirectiveLoc)) 2775 return true; 2776 2777 if (getLexer().isNot(AsmToken::Comma)) 2778 return TokError("unexpected token in directive"); 2779 Lex(); 2780 2781 int64_t Register2 = 0; 2782 if (ParseRegisterOrRegisterNumber(Register2, DirectiveLoc)) 2783 return true; 2784 2785 getStreamer().EmitCFIRegister(Register1, Register2); 2786 return false; 2787 } 2788 2789 /// ParseDirectiveCFIAdjustCfaOffset 2790 /// ::= .cfi_adjust_cfa_offset adjustment 2791 bool AsmParser::ParseDirectiveCFIAdjustCfaOffset() { 2792 int64_t Adjustment = 0; 2793 if (parseAbsoluteExpression(Adjustment)) 2794 return true; 2795 2796 getStreamer().EmitCFIAdjustCfaOffset(Adjustment); 2797 return false; 2798 } 2799 2800 /// ParseDirectiveCFIDefCfaRegister 2801 /// ::= .cfi_def_cfa_register register 2802 bool AsmParser::ParseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc) { 2803 int64_t Register = 0; 2804 if (ParseRegisterOrRegisterNumber(Register, DirectiveLoc)) 2805 return true; 2806 2807 getStreamer().EmitCFIDefCfaRegister(Register); 2808 return false; 2809 } 2810 2811 /// ParseDirectiveCFIOffset 2812 /// ::= .cfi_offset register, offset 2813 bool AsmParser::ParseDirectiveCFIOffset(SMLoc DirectiveLoc) { 2814 int64_t Register = 0; 2815 int64_t Offset = 0; 2816 2817 if (ParseRegisterOrRegisterNumber(Register, DirectiveLoc)) 2818 return true; 2819 2820 if (getLexer().isNot(AsmToken::Comma)) 2821 return TokError("unexpected token in directive"); 2822 Lex(); 2823 2824 if (parseAbsoluteExpression(Offset)) 2825 return true; 2826 2827 getStreamer().EmitCFIOffset(Register, Offset); 2828 return false; 2829 } 2830 2831 /// ParseDirectiveCFIRelOffset 2832 /// ::= .cfi_rel_offset register, offset 2833 bool AsmParser::ParseDirectiveCFIRelOffset(SMLoc DirectiveLoc) { 2834 int64_t Register = 0; 2835 2836 if (ParseRegisterOrRegisterNumber(Register, DirectiveLoc)) 2837 return true; 2838 2839 if (getLexer().isNot(AsmToken::Comma)) 2840 return TokError("unexpected token in directive"); 2841 Lex(); 2842 2843 int64_t Offset = 0; 2844 if (parseAbsoluteExpression(Offset)) 2845 return true; 2846 2847 getStreamer().EmitCFIRelOffset(Register, Offset); 2848 return false; 2849 } 2850 2851 static bool isValidEncoding(int64_t Encoding) { 2852 if (Encoding & ~0xff) 2853 return false; 2854 2855 if (Encoding == dwarf::DW_EH_PE_omit) 2856 return true; 2857 2858 const unsigned Format = Encoding & 0xf; 2859 if (Format != dwarf::DW_EH_PE_absptr && Format != dwarf::DW_EH_PE_udata2 && 2860 Format != dwarf::DW_EH_PE_udata4 && Format != dwarf::DW_EH_PE_udata8 && 2861 Format != dwarf::DW_EH_PE_sdata2 && Format != dwarf::DW_EH_PE_sdata4 && 2862 Format != dwarf::DW_EH_PE_sdata8 && Format != dwarf::DW_EH_PE_signed) 2863 return false; 2864 2865 const unsigned Application = Encoding & 0x70; 2866 if (Application != dwarf::DW_EH_PE_absptr && 2867 Application != dwarf::DW_EH_PE_pcrel) 2868 return false; 2869 2870 return true; 2871 } 2872 2873 /// ParseDirectiveCFIPersonalityOrLsda 2874 /// IsPersonality true for cfi_personality, false for cfi_lsda 2875 /// ::= .cfi_personality encoding, [symbol_name] 2876 /// ::= .cfi_lsda encoding, [symbol_name] 2877 bool AsmParser::ParseDirectiveCFIPersonalityOrLsda(bool IsPersonality) { 2878 int64_t Encoding = 0; 2879 if (parseAbsoluteExpression(Encoding)) 2880 return true; 2881 if (Encoding == dwarf::DW_EH_PE_omit) 2882 return false; 2883 2884 if (!isValidEncoding(Encoding)) 2885 return TokError("unsupported encoding."); 2886 2887 if (getLexer().isNot(AsmToken::Comma)) 2888 return TokError("unexpected token in directive"); 2889 Lex(); 2890 2891 StringRef Name; 2892 if (parseIdentifier(Name)) 2893 return TokError("expected identifier in directive"); 2894 2895 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name); 2896 2897 if (IsPersonality) 2898 getStreamer().EmitCFIPersonality(Sym, Encoding); 2899 else 2900 getStreamer().EmitCFILsda(Sym, Encoding); 2901 return false; 2902 } 2903 2904 /// ParseDirectiveCFIRememberState 2905 /// ::= .cfi_remember_state 2906 bool AsmParser::ParseDirectiveCFIRememberState() { 2907 getStreamer().EmitCFIRememberState(); 2908 return false; 2909 } 2910 2911 /// ParseDirectiveCFIRestoreState 2912 /// ::= .cfi_remember_state 2913 bool AsmParser::ParseDirectiveCFIRestoreState() { 2914 getStreamer().EmitCFIRestoreState(); 2915 return false; 2916 } 2917 2918 /// ParseDirectiveCFISameValue 2919 /// ::= .cfi_same_value register 2920 bool AsmParser::ParseDirectiveCFISameValue(SMLoc DirectiveLoc) { 2921 int64_t Register = 0; 2922 2923 if (ParseRegisterOrRegisterNumber(Register, DirectiveLoc)) 2924 return true; 2925 2926 getStreamer().EmitCFISameValue(Register); 2927 return false; 2928 } 2929 2930 /// ParseDirectiveCFIRestore 2931 /// ::= .cfi_restore register 2932 bool AsmParser::ParseDirectiveCFIRestore(SMLoc DirectiveLoc) { 2933 int64_t Register = 0; 2934 if (ParseRegisterOrRegisterNumber(Register, DirectiveLoc)) 2935 return true; 2936 2937 getStreamer().EmitCFIRestore(Register); 2938 return false; 2939 } 2940 2941 /// ParseDirectiveCFIEscape 2942 /// ::= .cfi_escape expression[,...] 2943 bool AsmParser::ParseDirectiveCFIEscape() { 2944 std::string Values; 2945 int64_t CurrValue; 2946 if (parseAbsoluteExpression(CurrValue)) 2947 return true; 2948 2949 Values.push_back((uint8_t)CurrValue); 2950 2951 while (getLexer().is(AsmToken::Comma)) { 2952 Lex(); 2953 2954 if (parseAbsoluteExpression(CurrValue)) 2955 return true; 2956 2957 Values.push_back((uint8_t)CurrValue); 2958 } 2959 2960 getStreamer().EmitCFIEscape(Values); 2961 return false; 2962 } 2963 2964 /// ParseDirectiveCFISignalFrame 2965 /// ::= .cfi_signal_frame 2966 bool AsmParser::ParseDirectiveCFISignalFrame() { 2967 if (getLexer().isNot(AsmToken::EndOfStatement)) 2968 return Error(getLexer().getLoc(), 2969 "unexpected token in '.cfi_signal_frame'"); 2970 2971 getStreamer().EmitCFISignalFrame(); 2972 return false; 2973 } 2974 2975 /// ParseDirectiveCFIUndefined 2976 /// ::= .cfi_undefined register 2977 bool AsmParser::ParseDirectiveCFIUndefined(SMLoc DirectiveLoc) { 2978 int64_t Register = 0; 2979 2980 if (ParseRegisterOrRegisterNumber(Register, DirectiveLoc)) 2981 return true; 2982 2983 getStreamer().EmitCFIUndefined(Register); 2984 return false; 2985 } 2986 2987 /// ParseDirectiveMacrosOnOff 2988 /// ::= .macros_on 2989 /// ::= .macros_off 2990 bool AsmParser::ParseDirectiveMacrosOnOff(StringRef Directive) { 2991 if (getLexer().isNot(AsmToken::EndOfStatement)) 2992 return Error(getLexer().getLoc(), 2993 "unexpected token in '" + Directive + "' directive"); 2994 2995 SetMacrosEnabled(Directive == ".macros_on"); 2996 return false; 2997 } 2998 2999 /// ParseDirectiveMacro 3000 /// ::= .macro name [parameters] 3001 bool AsmParser::ParseDirectiveMacro(SMLoc DirectiveLoc) { 3002 StringRef Name; 3003 if (parseIdentifier(Name)) 3004 return TokError("expected identifier in '.macro' directive"); 3005 3006 MCAsmMacroParameters Parameters; 3007 // Argument delimiter is initially unknown. It will be set by 3008 // ParseMacroArgument() 3009 AsmToken::TokenKind ArgumentDelimiter = AsmToken::Eof; 3010 if (getLexer().isNot(AsmToken::EndOfStatement)) { 3011 for (;;) { 3012 MCAsmMacroParameter Parameter; 3013 if (parseIdentifier(Parameter.first)) 3014 return TokError("expected identifier in '.macro' directive"); 3015 3016 if (getLexer().is(AsmToken::Equal)) { 3017 Lex(); 3018 if (ParseMacroArgument(Parameter.second, ArgumentDelimiter)) 3019 return true; 3020 } 3021 3022 Parameters.push_back(Parameter); 3023 3024 if (getLexer().is(AsmToken::Comma)) 3025 Lex(); 3026 else if (getLexer().is(AsmToken::EndOfStatement)) 3027 break; 3028 } 3029 } 3030 3031 // Eat the end of statement. 3032 Lex(); 3033 3034 AsmToken EndToken, StartToken = getTok(); 3035 3036 // Lex the macro definition. 3037 for (;;) { 3038 // Check whether we have reached the end of the file. 3039 if (getLexer().is(AsmToken::Eof)) 3040 return Error(DirectiveLoc, "no matching '.endmacro' in definition"); 3041 3042 // Otherwise, check whether we have reach the .endmacro. 3043 if (getLexer().is(AsmToken::Identifier) && 3044 (getTok().getIdentifier() == ".endm" || 3045 getTok().getIdentifier() == ".endmacro")) { 3046 EndToken = getTok(); 3047 Lex(); 3048 if (getLexer().isNot(AsmToken::EndOfStatement)) 3049 return TokError("unexpected token in '" + EndToken.getIdentifier() + 3050 "' directive"); 3051 break; 3052 } 3053 3054 // Otherwise, scan til the end of the statement. 3055 eatToEndOfStatement(); 3056 } 3057 3058 if (LookupMacro(Name)) { 3059 return Error(DirectiveLoc, "macro '" + Name + "' is already defined"); 3060 } 3061 3062 const char *BodyStart = StartToken.getLoc().getPointer(); 3063 const char *BodyEnd = EndToken.getLoc().getPointer(); 3064 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 3065 CheckForBadMacro(DirectiveLoc, Name, Body, Parameters); 3066 DefineMacro(Name, MCAsmMacro(Name, Body, Parameters)); 3067 return false; 3068 } 3069 3070 /// CheckForBadMacro 3071 /// 3072 /// With the support added for named parameters there may be code out there that 3073 /// is transitioning from positional parameters. In versions of gas that did 3074 /// not support named parameters they would be ignored on the macro defintion. 3075 /// But to support both styles of parameters this is not possible so if a macro 3076 /// defintion has named parameters but does not use them and has what appears 3077 /// to be positional parameters, strings like $1, $2, ... and $n, then issue a 3078 /// warning that the positional parameter found in body which have no effect. 3079 /// Hoping the developer will either remove the named parameters from the macro 3080 /// definiton so the positional parameters get used if that was what was 3081 /// intended or change the macro to use the named parameters. It is possible 3082 /// this warning will trigger when the none of the named parameters are used 3083 /// and the strings like $1 are infact to simply to be passed trough unchanged. 3084 void AsmParser::CheckForBadMacro(SMLoc DirectiveLoc, StringRef Name, 3085 StringRef Body, 3086 MCAsmMacroParameters Parameters) { 3087 // If this macro is not defined with named parameters the warning we are 3088 // checking for here doesn't apply. 3089 unsigned NParameters = Parameters.size(); 3090 if (NParameters == 0) 3091 return; 3092 3093 bool NamedParametersFound = false; 3094 bool PositionalParametersFound = false; 3095 3096 // Look at the body of the macro for use of both the named parameters and what 3097 // are likely to be positional parameters. This is what expandMacro() is 3098 // doing when it finds the parameters in the body. 3099 while (!Body.empty()) { 3100 // Scan for the next possible parameter. 3101 std::size_t End = Body.size(), Pos = 0; 3102 for (; Pos != End; ++Pos) { 3103 // Check for a substitution or escape. 3104 // This macro is defined with parameters, look for \foo, \bar, etc. 3105 if (Body[Pos] == '\\' && Pos + 1 != End) 3106 break; 3107 3108 // This macro should have parameters, but look for $0, $1, ..., $n too. 3109 if (Body[Pos] != '$' || Pos + 1 == End) 3110 continue; 3111 char Next = Body[Pos + 1]; 3112 if (Next == '$' || Next == 'n' || 3113 isdigit(static_cast<unsigned char>(Next))) 3114 break; 3115 } 3116 3117 // Check if we reached the end. 3118 if (Pos == End) 3119 break; 3120 3121 if (Body[Pos] == '$') { 3122 switch (Body[Pos+1]) { 3123 // $$ => $ 3124 case '$': 3125 break; 3126 3127 // $n => number of arguments 3128 case 'n': 3129 PositionalParametersFound = true; 3130 break; 3131 3132 // $[0-9] => argument 3133 default: { 3134 PositionalParametersFound = true; 3135 break; 3136 } 3137 } 3138 Pos += 2; 3139 } else { 3140 unsigned I = Pos + 1; 3141 while (isIdentifierChar(Body[I]) && I + 1 != End) 3142 ++I; 3143 3144 const char *Begin = Body.data() + Pos +1; 3145 StringRef Argument(Begin, I - (Pos +1)); 3146 unsigned Index = 0; 3147 for (; Index < NParameters; ++Index) 3148 if (Parameters[Index].first == Argument) 3149 break; 3150 3151 if (Index == NParameters) { 3152 if (Body[Pos+1] == '(' && Body[Pos+2] == ')') 3153 Pos += 3; 3154 else { 3155 Pos = I; 3156 } 3157 } else { 3158 NamedParametersFound = true; 3159 Pos += 1 + Argument.size(); 3160 } 3161 } 3162 // Update the scan point. 3163 Body = Body.substr(Pos); 3164 } 3165 3166 if (!NamedParametersFound && PositionalParametersFound) 3167 Warning(DirectiveLoc, "macro defined with named parameters which are not " 3168 "used in macro body, possible positional parameter " 3169 "found in body which will have no effect"); 3170 } 3171 3172 /// ParseDirectiveEndMacro 3173 /// ::= .endm 3174 /// ::= .endmacro 3175 bool AsmParser::ParseDirectiveEndMacro(StringRef Directive) { 3176 if (getLexer().isNot(AsmToken::EndOfStatement)) 3177 return TokError("unexpected token in '" + Directive + "' directive"); 3178 3179 // If we are inside a macro instantiation, terminate the current 3180 // instantiation. 3181 if (InsideMacroInstantiation()) { 3182 HandleMacroExit(); 3183 return false; 3184 } 3185 3186 // Otherwise, this .endmacro is a stray entry in the file; well formed 3187 // .endmacro directives are handled during the macro definition parsing. 3188 return TokError("unexpected '" + Directive + "' in file, " 3189 "no current macro definition"); 3190 } 3191 3192 /// ParseDirectivePurgeMacro 3193 /// ::= .purgem 3194 bool AsmParser::ParseDirectivePurgeMacro(SMLoc DirectiveLoc) { 3195 StringRef Name; 3196 if (parseIdentifier(Name)) 3197 return TokError("expected identifier in '.purgem' directive"); 3198 3199 if (getLexer().isNot(AsmToken::EndOfStatement)) 3200 return TokError("unexpected token in '.purgem' directive"); 3201 3202 if (!LookupMacro(Name)) 3203 return Error(DirectiveLoc, "macro '" + Name + "' is not defined"); 3204 3205 UndefineMacro(Name); 3206 return false; 3207 } 3208 3209 /// ParseDirectiveBundleAlignMode 3210 /// ::= {.bundle_align_mode} expression 3211 bool AsmParser::ParseDirectiveBundleAlignMode() { 3212 checkForValidSection(); 3213 3214 // Expect a single argument: an expression that evaluates to a constant 3215 // in the inclusive range 0-30. 3216 SMLoc ExprLoc = getLexer().getLoc(); 3217 int64_t AlignSizePow2; 3218 if (parseAbsoluteExpression(AlignSizePow2)) 3219 return true; 3220 else if (getLexer().isNot(AsmToken::EndOfStatement)) 3221 return TokError("unexpected token after expression in" 3222 " '.bundle_align_mode' directive"); 3223 else if (AlignSizePow2 < 0 || AlignSizePow2 > 30) 3224 return Error(ExprLoc, 3225 "invalid bundle alignment size (expected between 0 and 30)"); 3226 3227 Lex(); 3228 3229 // Because of AlignSizePow2's verified range we can safely truncate it to 3230 // unsigned. 3231 getStreamer().EmitBundleAlignMode(static_cast<unsigned>(AlignSizePow2)); 3232 return false; 3233 } 3234 3235 /// ParseDirectiveBundleLock 3236 /// ::= {.bundle_lock} [align_to_end] 3237 bool AsmParser::ParseDirectiveBundleLock() { 3238 checkForValidSection(); 3239 bool AlignToEnd = false; 3240 3241 if (getLexer().isNot(AsmToken::EndOfStatement)) { 3242 StringRef Option; 3243 SMLoc Loc = getTok().getLoc(); 3244 const char *kInvalidOptionError = 3245 "invalid option for '.bundle_lock' directive"; 3246 3247 if (parseIdentifier(Option)) 3248 return Error(Loc, kInvalidOptionError); 3249 3250 if (Option != "align_to_end") 3251 return Error(Loc, kInvalidOptionError); 3252 else if (getLexer().isNot(AsmToken::EndOfStatement)) 3253 return Error(Loc, 3254 "unexpected token after '.bundle_lock' directive option"); 3255 AlignToEnd = true; 3256 } 3257 3258 Lex(); 3259 3260 getStreamer().EmitBundleLock(AlignToEnd); 3261 return false; 3262 } 3263 3264 /// ParseDirectiveBundleLock 3265 /// ::= {.bundle_lock} 3266 bool AsmParser::ParseDirectiveBundleUnlock() { 3267 checkForValidSection(); 3268 3269 if (getLexer().isNot(AsmToken::EndOfStatement)) 3270 return TokError("unexpected token in '.bundle_unlock' directive"); 3271 Lex(); 3272 3273 getStreamer().EmitBundleUnlock(); 3274 return false; 3275 } 3276 3277 /// ParseDirectiveSpace 3278 /// ::= (.skip | .space) expression [ , expression ] 3279 bool AsmParser::ParseDirectiveSpace(StringRef IDVal) { 3280 checkForValidSection(); 3281 3282 int64_t NumBytes; 3283 if (parseAbsoluteExpression(NumBytes)) 3284 return true; 3285 3286 int64_t FillExpr = 0; 3287 if (getLexer().isNot(AsmToken::EndOfStatement)) { 3288 if (getLexer().isNot(AsmToken::Comma)) 3289 return TokError("unexpected token in '" + Twine(IDVal) + "' directive"); 3290 Lex(); 3291 3292 if (parseAbsoluteExpression(FillExpr)) 3293 return true; 3294 3295 if (getLexer().isNot(AsmToken::EndOfStatement)) 3296 return TokError("unexpected token in '" + Twine(IDVal) + "' directive"); 3297 } 3298 3299 Lex(); 3300 3301 if (NumBytes <= 0) 3302 return TokError("invalid number of bytes in '" + 3303 Twine(IDVal) + "' directive"); 3304 3305 // FIXME: Sometimes the fill expr is 'nop' if it isn't supplied, instead of 0. 3306 getStreamer().EmitFill(NumBytes, FillExpr, DEFAULT_ADDRSPACE); 3307 3308 return false; 3309 } 3310 3311 /// ParseDirectiveLEB128 3312 /// ::= (.sleb128 | .uleb128) expression 3313 bool AsmParser::ParseDirectiveLEB128(bool Signed) { 3314 checkForValidSection(); 3315 const MCExpr *Value; 3316 3317 if (parseExpression(Value)) 3318 return true; 3319 3320 if (getLexer().isNot(AsmToken::EndOfStatement)) 3321 return TokError("unexpected token in directive"); 3322 3323 if (Signed) 3324 getStreamer().EmitSLEB128Value(Value); 3325 else 3326 getStreamer().EmitULEB128Value(Value); 3327 3328 return false; 3329 } 3330 3331 /// ParseDirectiveSymbolAttribute 3332 /// ::= { ".globl", ".weak", ... } [ identifier ( , identifier )* ] 3333 bool AsmParser::ParseDirectiveSymbolAttribute(MCSymbolAttr Attr) { 3334 if (getLexer().isNot(AsmToken::EndOfStatement)) { 3335 for (;;) { 3336 StringRef Name; 3337 SMLoc Loc = getTok().getLoc(); 3338 3339 if (parseIdentifier(Name)) 3340 return Error(Loc, "expected identifier in directive"); 3341 3342 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name); 3343 3344 // Assembler local symbols don't make any sense here. Complain loudly. 3345 if (Sym->isTemporary()) 3346 return Error(Loc, "non-local symbol required in directive"); 3347 3348 getStreamer().EmitSymbolAttribute(Sym, Attr); 3349 3350 if (getLexer().is(AsmToken::EndOfStatement)) 3351 break; 3352 3353 if (getLexer().isNot(AsmToken::Comma)) 3354 return TokError("unexpected token in directive"); 3355 Lex(); 3356 } 3357 } 3358 3359 Lex(); 3360 return false; 3361 } 3362 3363 /// ParseDirectiveComm 3364 /// ::= ( .comm | .lcomm ) identifier , size_expression [ , align_expression ] 3365 bool AsmParser::ParseDirectiveComm(bool IsLocal) { 3366 checkForValidSection(); 3367 3368 SMLoc IDLoc = getLexer().getLoc(); 3369 StringRef Name; 3370 if (parseIdentifier(Name)) 3371 return TokError("expected identifier in directive"); 3372 3373 // Handle the identifier as the key symbol. 3374 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name); 3375 3376 if (getLexer().isNot(AsmToken::Comma)) 3377 return TokError("unexpected token in directive"); 3378 Lex(); 3379 3380 int64_t Size; 3381 SMLoc SizeLoc = getLexer().getLoc(); 3382 if (parseAbsoluteExpression(Size)) 3383 return true; 3384 3385 int64_t Pow2Alignment = 0; 3386 SMLoc Pow2AlignmentLoc; 3387 if (getLexer().is(AsmToken::Comma)) { 3388 Lex(); 3389 Pow2AlignmentLoc = getLexer().getLoc(); 3390 if (parseAbsoluteExpression(Pow2Alignment)) 3391 return true; 3392 3393 LCOMM::LCOMMType LCOMM = Lexer.getMAI().getLCOMMDirectiveAlignmentType(); 3394 if (IsLocal && LCOMM == LCOMM::NoAlignment) 3395 return Error(Pow2AlignmentLoc, "alignment not supported on this target"); 3396 3397 // If this target takes alignments in bytes (not log) validate and convert. 3398 if ((!IsLocal && Lexer.getMAI().getCOMMDirectiveAlignmentIsInBytes()) || 3399 (IsLocal && LCOMM == LCOMM::ByteAlignment)) { 3400 if (!isPowerOf2_64(Pow2Alignment)) 3401 return Error(Pow2AlignmentLoc, "alignment must be a power of 2"); 3402 Pow2Alignment = Log2_64(Pow2Alignment); 3403 } 3404 } 3405 3406 if (getLexer().isNot(AsmToken::EndOfStatement)) 3407 return TokError("unexpected token in '.comm' or '.lcomm' directive"); 3408 3409 Lex(); 3410 3411 // NOTE: a size of zero for a .comm should create a undefined symbol 3412 // but a size of .lcomm creates a bss symbol of size zero. 3413 if (Size < 0) 3414 return Error(SizeLoc, "invalid '.comm' or '.lcomm' directive size, can't " 3415 "be less than zero"); 3416 3417 // NOTE: The alignment in the directive is a power of 2 value, the assembler 3418 // may internally end up wanting an alignment in bytes. 3419 // FIXME: Diagnose overflow. 3420 if (Pow2Alignment < 0) 3421 return Error(Pow2AlignmentLoc, "invalid '.comm' or '.lcomm' directive " 3422 "alignment, can't be less than zero"); 3423 3424 if (!Sym->isUndefined()) 3425 return Error(IDLoc, "invalid symbol redefinition"); 3426 3427 // Create the Symbol as a common or local common with Size and Pow2Alignment 3428 if (IsLocal) { 3429 getStreamer().EmitLocalCommonSymbol(Sym, Size, 1 << Pow2Alignment); 3430 return false; 3431 } 3432 3433 getStreamer().EmitCommonSymbol(Sym, Size, 1 << Pow2Alignment); 3434 return false; 3435 } 3436 3437 /// ParseDirectiveAbort 3438 /// ::= .abort [... message ...] 3439 bool AsmParser::ParseDirectiveAbort() { 3440 // FIXME: Use loc from directive. 3441 SMLoc Loc = getLexer().getLoc(); 3442 3443 StringRef Str = parseStringToEndOfStatement(); 3444 if (getLexer().isNot(AsmToken::EndOfStatement)) 3445 return TokError("unexpected token in '.abort' directive"); 3446 3447 Lex(); 3448 3449 if (Str.empty()) 3450 Error(Loc, ".abort detected. Assembly stopping."); 3451 else 3452 Error(Loc, ".abort '" + Str + "' detected. Assembly stopping."); 3453 // FIXME: Actually abort assembly here. 3454 3455 return false; 3456 } 3457 3458 /// ParseDirectiveInclude 3459 /// ::= .include "filename" 3460 bool AsmParser::ParseDirectiveInclude() { 3461 if (getLexer().isNot(AsmToken::String)) 3462 return TokError("expected string in '.include' directive"); 3463 3464 std::string Filename = getTok().getString(); 3465 SMLoc IncludeLoc = getLexer().getLoc(); 3466 Lex(); 3467 3468 if (getLexer().isNot(AsmToken::EndOfStatement)) 3469 return TokError("unexpected token in '.include' directive"); 3470 3471 // Strip the quotes. 3472 Filename = Filename.substr(1, Filename.size()-2); 3473 3474 // Attempt to switch the lexer to the included file before consuming the end 3475 // of statement to avoid losing it when we switch. 3476 if (EnterIncludeFile(Filename)) { 3477 Error(IncludeLoc, "Could not find include file '" + Filename + "'"); 3478 return true; 3479 } 3480 3481 return false; 3482 } 3483 3484 /// ParseDirectiveIncbin 3485 /// ::= .incbin "filename" 3486 bool AsmParser::ParseDirectiveIncbin() { 3487 if (getLexer().isNot(AsmToken::String)) 3488 return TokError("expected string in '.incbin' directive"); 3489 3490 std::string Filename = getTok().getString(); 3491 SMLoc IncbinLoc = getLexer().getLoc(); 3492 Lex(); 3493 3494 if (getLexer().isNot(AsmToken::EndOfStatement)) 3495 return TokError("unexpected token in '.incbin' directive"); 3496 3497 // Strip the quotes. 3498 Filename = Filename.substr(1, Filename.size()-2); 3499 3500 // Attempt to process the included file. 3501 if (ProcessIncbinFile(Filename)) { 3502 Error(IncbinLoc, "Could not find incbin file '" + Filename + "'"); 3503 return true; 3504 } 3505 3506 return false; 3507 } 3508 3509 /// ParseDirectiveIf 3510 /// ::= .if expression 3511 bool AsmParser::ParseDirectiveIf(SMLoc DirectiveLoc) { 3512 TheCondStack.push_back(TheCondState); 3513 TheCondState.TheCond = AsmCond::IfCond; 3514 if (TheCondState.Ignore) { 3515 eatToEndOfStatement(); 3516 } else { 3517 int64_t ExprValue; 3518 if (parseAbsoluteExpression(ExprValue)) 3519 return true; 3520 3521 if (getLexer().isNot(AsmToken::EndOfStatement)) 3522 return TokError("unexpected token in '.if' directive"); 3523 3524 Lex(); 3525 3526 TheCondState.CondMet = ExprValue; 3527 TheCondState.Ignore = !TheCondState.CondMet; 3528 } 3529 3530 return false; 3531 } 3532 3533 /// ParseDirectiveIfb 3534 /// ::= .ifb string 3535 bool AsmParser::ParseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank) { 3536 TheCondStack.push_back(TheCondState); 3537 TheCondState.TheCond = AsmCond::IfCond; 3538 3539 if (TheCondState.Ignore) { 3540 eatToEndOfStatement(); 3541 } else { 3542 StringRef Str = parseStringToEndOfStatement(); 3543 3544 if (getLexer().isNot(AsmToken::EndOfStatement)) 3545 return TokError("unexpected token in '.ifb' directive"); 3546 3547 Lex(); 3548 3549 TheCondState.CondMet = ExpectBlank == Str.empty(); 3550 TheCondState.Ignore = !TheCondState.CondMet; 3551 } 3552 3553 return false; 3554 } 3555 3556 /// ParseDirectiveIfc 3557 /// ::= .ifc string1, string2 3558 bool AsmParser::ParseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual) { 3559 TheCondStack.push_back(TheCondState); 3560 TheCondState.TheCond = AsmCond::IfCond; 3561 3562 if (TheCondState.Ignore) { 3563 eatToEndOfStatement(); 3564 } else { 3565 StringRef Str1 = ParseStringToComma(); 3566 3567 if (getLexer().isNot(AsmToken::Comma)) 3568 return TokError("unexpected token in '.ifc' directive"); 3569 3570 Lex(); 3571 3572 StringRef Str2 = parseStringToEndOfStatement(); 3573 3574 if (getLexer().isNot(AsmToken::EndOfStatement)) 3575 return TokError("unexpected token in '.ifc' directive"); 3576 3577 Lex(); 3578 3579 TheCondState.CondMet = ExpectEqual == (Str1 == Str2); 3580 TheCondState.Ignore = !TheCondState.CondMet; 3581 } 3582 3583 return false; 3584 } 3585 3586 /// ParseDirectiveIfdef 3587 /// ::= .ifdef symbol 3588 bool AsmParser::ParseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined) { 3589 StringRef Name; 3590 TheCondStack.push_back(TheCondState); 3591 TheCondState.TheCond = AsmCond::IfCond; 3592 3593 if (TheCondState.Ignore) { 3594 eatToEndOfStatement(); 3595 } else { 3596 if (parseIdentifier(Name)) 3597 return TokError("expected identifier after '.ifdef'"); 3598 3599 Lex(); 3600 3601 MCSymbol *Sym = getContext().LookupSymbol(Name); 3602 3603 if (expect_defined) 3604 TheCondState.CondMet = (Sym != NULL && !Sym->isUndefined()); 3605 else 3606 TheCondState.CondMet = (Sym == NULL || Sym->isUndefined()); 3607 TheCondState.Ignore = !TheCondState.CondMet; 3608 } 3609 3610 return false; 3611 } 3612 3613 /// ParseDirectiveElseIf 3614 /// ::= .elseif expression 3615 bool AsmParser::ParseDirectiveElseIf(SMLoc DirectiveLoc) { 3616 if (TheCondState.TheCond != AsmCond::IfCond && 3617 TheCondState.TheCond != AsmCond::ElseIfCond) 3618 Error(DirectiveLoc, "Encountered a .elseif that doesn't follow a .if or " 3619 " an .elseif"); 3620 TheCondState.TheCond = AsmCond::ElseIfCond; 3621 3622 bool LastIgnoreState = false; 3623 if (!TheCondStack.empty()) 3624 LastIgnoreState = TheCondStack.back().Ignore; 3625 if (LastIgnoreState || TheCondState.CondMet) { 3626 TheCondState.Ignore = true; 3627 eatToEndOfStatement(); 3628 } else { 3629 int64_t ExprValue; 3630 if (parseAbsoluteExpression(ExprValue)) 3631 return true; 3632 3633 if (getLexer().isNot(AsmToken::EndOfStatement)) 3634 return TokError("unexpected token in '.elseif' directive"); 3635 3636 Lex(); 3637 TheCondState.CondMet = ExprValue; 3638 TheCondState.Ignore = !TheCondState.CondMet; 3639 } 3640 3641 return false; 3642 } 3643 3644 /// ParseDirectiveElse 3645 /// ::= .else 3646 bool AsmParser::ParseDirectiveElse(SMLoc DirectiveLoc) { 3647 if (getLexer().isNot(AsmToken::EndOfStatement)) 3648 return TokError("unexpected token in '.else' directive"); 3649 3650 Lex(); 3651 3652 if (TheCondState.TheCond != AsmCond::IfCond && 3653 TheCondState.TheCond != AsmCond::ElseIfCond) 3654 Error(DirectiveLoc, "Encountered a .else that doesn't follow a .if or an " 3655 ".elseif"); 3656 TheCondState.TheCond = AsmCond::ElseCond; 3657 bool LastIgnoreState = false; 3658 if (!TheCondStack.empty()) 3659 LastIgnoreState = TheCondStack.back().Ignore; 3660 if (LastIgnoreState || TheCondState.CondMet) 3661 TheCondState.Ignore = true; 3662 else 3663 TheCondState.Ignore = false; 3664 3665 return false; 3666 } 3667 3668 /// ParseDirectiveEndIf 3669 /// ::= .endif 3670 bool AsmParser::ParseDirectiveEndIf(SMLoc DirectiveLoc) { 3671 if (getLexer().isNot(AsmToken::EndOfStatement)) 3672 return TokError("unexpected token in '.endif' directive"); 3673 3674 Lex(); 3675 3676 if ((TheCondState.TheCond == AsmCond::NoCond) || 3677 TheCondStack.empty()) 3678 Error(DirectiveLoc, "Encountered a .endif that doesn't follow a .if or " 3679 ".else"); 3680 if (!TheCondStack.empty()) { 3681 TheCondState = TheCondStack.back(); 3682 TheCondStack.pop_back(); 3683 } 3684 3685 return false; 3686 } 3687 3688 void AsmParser::initializeDirectiveKindMap() { 3689 DirectiveKindMap[".set"] = DK_SET; 3690 DirectiveKindMap[".equ"] = DK_EQU; 3691 DirectiveKindMap[".equiv"] = DK_EQUIV; 3692 DirectiveKindMap[".ascii"] = DK_ASCII; 3693 DirectiveKindMap[".asciz"] = DK_ASCIZ; 3694 DirectiveKindMap[".string"] = DK_STRING; 3695 DirectiveKindMap[".byte"] = DK_BYTE; 3696 DirectiveKindMap[".short"] = DK_SHORT; 3697 DirectiveKindMap[".value"] = DK_VALUE; 3698 DirectiveKindMap[".2byte"] = DK_2BYTE; 3699 DirectiveKindMap[".long"] = DK_LONG; 3700 DirectiveKindMap[".int"] = DK_INT; 3701 DirectiveKindMap[".4byte"] = DK_4BYTE; 3702 DirectiveKindMap[".quad"] = DK_QUAD; 3703 DirectiveKindMap[".8byte"] = DK_8BYTE; 3704 DirectiveKindMap[".single"] = DK_SINGLE; 3705 DirectiveKindMap[".float"] = DK_FLOAT; 3706 DirectiveKindMap[".double"] = DK_DOUBLE; 3707 DirectiveKindMap[".align"] = DK_ALIGN; 3708 DirectiveKindMap[".align32"] = DK_ALIGN32; 3709 DirectiveKindMap[".balign"] = DK_BALIGN; 3710 DirectiveKindMap[".balignw"] = DK_BALIGNW; 3711 DirectiveKindMap[".balignl"] = DK_BALIGNL; 3712 DirectiveKindMap[".p2align"] = DK_P2ALIGN; 3713 DirectiveKindMap[".p2alignw"] = DK_P2ALIGNW; 3714 DirectiveKindMap[".p2alignl"] = DK_P2ALIGNL; 3715 DirectiveKindMap[".org"] = DK_ORG; 3716 DirectiveKindMap[".fill"] = DK_FILL; 3717 DirectiveKindMap[".zero"] = DK_ZERO; 3718 DirectiveKindMap[".extern"] = DK_EXTERN; 3719 DirectiveKindMap[".globl"] = DK_GLOBL; 3720 DirectiveKindMap[".global"] = DK_GLOBAL; 3721 DirectiveKindMap[".indirect_symbol"] = DK_INDIRECT_SYMBOL; 3722 DirectiveKindMap[".lazy_reference"] = DK_LAZY_REFERENCE; 3723 DirectiveKindMap[".no_dead_strip"] = DK_NO_DEAD_STRIP; 3724 DirectiveKindMap[".symbol_resolver"] = DK_SYMBOL_RESOLVER; 3725 DirectiveKindMap[".private_extern"] = DK_PRIVATE_EXTERN; 3726 DirectiveKindMap[".reference"] = DK_REFERENCE; 3727 DirectiveKindMap[".weak_definition"] = DK_WEAK_DEFINITION; 3728 DirectiveKindMap[".weak_reference"] = DK_WEAK_REFERENCE; 3729 DirectiveKindMap[".weak_def_can_be_hidden"] = DK_WEAK_DEF_CAN_BE_HIDDEN; 3730 DirectiveKindMap[".comm"] = DK_COMM; 3731 DirectiveKindMap[".common"] = DK_COMMON; 3732 DirectiveKindMap[".lcomm"] = DK_LCOMM; 3733 DirectiveKindMap[".abort"] = DK_ABORT; 3734 DirectiveKindMap[".include"] = DK_INCLUDE; 3735 DirectiveKindMap[".incbin"] = DK_INCBIN; 3736 DirectiveKindMap[".code16"] = DK_CODE16; 3737 DirectiveKindMap[".code16gcc"] = DK_CODE16GCC; 3738 DirectiveKindMap[".rept"] = DK_REPT; 3739 DirectiveKindMap[".irp"] = DK_IRP; 3740 DirectiveKindMap[".irpc"] = DK_IRPC; 3741 DirectiveKindMap[".endr"] = DK_ENDR; 3742 DirectiveKindMap[".bundle_align_mode"] = DK_BUNDLE_ALIGN_MODE; 3743 DirectiveKindMap[".bundle_lock"] = DK_BUNDLE_LOCK; 3744 DirectiveKindMap[".bundle_unlock"] = DK_BUNDLE_UNLOCK; 3745 DirectiveKindMap[".if"] = DK_IF; 3746 DirectiveKindMap[".ifb"] = DK_IFB; 3747 DirectiveKindMap[".ifnb"] = DK_IFNB; 3748 DirectiveKindMap[".ifc"] = DK_IFC; 3749 DirectiveKindMap[".ifnc"] = DK_IFNC; 3750 DirectiveKindMap[".ifdef"] = DK_IFDEF; 3751 DirectiveKindMap[".ifndef"] = DK_IFNDEF; 3752 DirectiveKindMap[".ifnotdef"] = DK_IFNOTDEF; 3753 DirectiveKindMap[".elseif"] = DK_ELSEIF; 3754 DirectiveKindMap[".else"] = DK_ELSE; 3755 DirectiveKindMap[".endif"] = DK_ENDIF; 3756 DirectiveKindMap[".skip"] = DK_SKIP; 3757 DirectiveKindMap[".space"] = DK_SPACE; 3758 DirectiveKindMap[".file"] = DK_FILE; 3759 DirectiveKindMap[".line"] = DK_LINE; 3760 DirectiveKindMap[".loc"] = DK_LOC; 3761 DirectiveKindMap[".stabs"] = DK_STABS; 3762 DirectiveKindMap[".sleb128"] = DK_SLEB128; 3763 DirectiveKindMap[".uleb128"] = DK_ULEB128; 3764 DirectiveKindMap[".cfi_sections"] = DK_CFI_SECTIONS; 3765 DirectiveKindMap[".cfi_startproc"] = DK_CFI_STARTPROC; 3766 DirectiveKindMap[".cfi_endproc"] = DK_CFI_ENDPROC; 3767 DirectiveKindMap[".cfi_def_cfa"] = DK_CFI_DEF_CFA; 3768 DirectiveKindMap[".cfi_def_cfa_offset"] = DK_CFI_DEF_CFA_OFFSET; 3769 DirectiveKindMap[".cfi_adjust_cfa_offset"] = DK_CFI_ADJUST_CFA_OFFSET; 3770 DirectiveKindMap[".cfi_def_cfa_register"] = DK_CFI_DEF_CFA_REGISTER; 3771 DirectiveKindMap[".cfi_offset"] = DK_CFI_OFFSET; 3772 DirectiveKindMap[".cfi_rel_offset"] = DK_CFI_REL_OFFSET; 3773 DirectiveKindMap[".cfi_personality"] = DK_CFI_PERSONALITY; 3774 DirectiveKindMap[".cfi_lsda"] = DK_CFI_LSDA; 3775 DirectiveKindMap[".cfi_remember_state"] = DK_CFI_REMEMBER_STATE; 3776 DirectiveKindMap[".cfi_restore_state"] = DK_CFI_RESTORE_STATE; 3777 DirectiveKindMap[".cfi_same_value"] = DK_CFI_SAME_VALUE; 3778 DirectiveKindMap[".cfi_restore"] = DK_CFI_RESTORE; 3779 DirectiveKindMap[".cfi_escape"] = DK_CFI_ESCAPE; 3780 DirectiveKindMap[".cfi_signal_frame"] = DK_CFI_SIGNAL_FRAME; 3781 DirectiveKindMap[".cfi_undefined"] = DK_CFI_UNDEFINED; 3782 DirectiveKindMap[".cfi_register"] = DK_CFI_REGISTER; 3783 DirectiveKindMap[".macros_on"] = DK_MACROS_ON; 3784 DirectiveKindMap[".macros_off"] = DK_MACROS_OFF; 3785 DirectiveKindMap[".macro"] = DK_MACRO; 3786 DirectiveKindMap[".endm"] = DK_ENDM; 3787 DirectiveKindMap[".endmacro"] = DK_ENDMACRO; 3788 DirectiveKindMap[".purgem"] = DK_PURGEM; 3789 } 3790 3791 3792 MCAsmMacro *AsmParser::ParseMacroLikeBody(SMLoc DirectiveLoc) { 3793 AsmToken EndToken, StartToken = getTok(); 3794 3795 unsigned NestLevel = 0; 3796 for (;;) { 3797 // Check whether we have reached the end of the file. 3798 if (getLexer().is(AsmToken::Eof)) { 3799 Error(DirectiveLoc, "no matching '.endr' in definition"); 3800 return 0; 3801 } 3802 3803 if (Lexer.is(AsmToken::Identifier) && 3804 (getTok().getIdentifier() == ".rept")) { 3805 ++NestLevel; 3806 } 3807 3808 // Otherwise, check whether we have reached the .endr. 3809 if (Lexer.is(AsmToken::Identifier) && 3810 getTok().getIdentifier() == ".endr") { 3811 if (NestLevel == 0) { 3812 EndToken = getTok(); 3813 Lex(); 3814 if (Lexer.isNot(AsmToken::EndOfStatement)) { 3815 TokError("unexpected token in '.endr' directive"); 3816 return 0; 3817 } 3818 break; 3819 } 3820 --NestLevel; 3821 } 3822 3823 // Otherwise, scan till the end of the statement. 3824 eatToEndOfStatement(); 3825 } 3826 3827 const char *BodyStart = StartToken.getLoc().getPointer(); 3828 const char *BodyEnd = EndToken.getLoc().getPointer(); 3829 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 3830 3831 // We Are Anonymous. 3832 StringRef Name; 3833 MCAsmMacroParameters Parameters; 3834 return new MCAsmMacro(Name, Body, Parameters); 3835 } 3836 3837 void AsmParser::InstantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 3838 raw_svector_ostream &OS) { 3839 OS << ".endr\n"; 3840 3841 MemoryBuffer *Instantiation = 3842 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 3843 3844 // Create the macro instantiation object and add to the current macro 3845 // instantiation stack. 3846 MacroInstantiation *MI = new MacroInstantiation(M, DirectiveLoc, 3847 CurBuffer, 3848 getTok().getLoc(), 3849 Instantiation); 3850 ActiveMacros.push_back(MI); 3851 3852 // Jump to the macro instantiation and prime the lexer. 3853 CurBuffer = SrcMgr.AddNewSourceBuffer(MI->Instantiation, SMLoc()); 3854 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)); 3855 Lex(); 3856 } 3857 3858 bool AsmParser::ParseDirectiveRept(SMLoc DirectiveLoc) { 3859 int64_t Count; 3860 if (parseAbsoluteExpression(Count)) 3861 return TokError("unexpected token in '.rept' directive"); 3862 3863 if (Count < 0) 3864 return TokError("Count is negative"); 3865 3866 if (Lexer.isNot(AsmToken::EndOfStatement)) 3867 return TokError("unexpected token in '.rept' directive"); 3868 3869 // Eat the end of statement. 3870 Lex(); 3871 3872 // Lex the rept definition. 3873 MCAsmMacro *M = ParseMacroLikeBody(DirectiveLoc); 3874 if (!M) 3875 return true; 3876 3877 // Macro instantiation is lexical, unfortunately. We construct a new buffer 3878 // to hold the macro body with substitutions. 3879 SmallString<256> Buf; 3880 MCAsmMacroParameters Parameters; 3881 MCAsmMacroArguments A; 3882 raw_svector_ostream OS(Buf); 3883 while (Count--) { 3884 if (expandMacro(OS, M->Body, Parameters, A, getTok().getLoc())) 3885 return true; 3886 } 3887 InstantiateMacroLikeBody(M, DirectiveLoc, OS); 3888 3889 return false; 3890 } 3891 3892 /// ParseDirectiveIrp 3893 /// ::= .irp symbol,values 3894 bool AsmParser::ParseDirectiveIrp(SMLoc DirectiveLoc) { 3895 MCAsmMacroParameters Parameters; 3896 MCAsmMacroParameter Parameter; 3897 3898 if (parseIdentifier(Parameter.first)) 3899 return TokError("expected identifier in '.irp' directive"); 3900 3901 Parameters.push_back(Parameter); 3902 3903 if (Lexer.isNot(AsmToken::Comma)) 3904 return TokError("expected comma in '.irp' directive"); 3905 3906 Lex(); 3907 3908 MCAsmMacroArguments A; 3909 if (ParseMacroArguments(0, A)) 3910 return true; 3911 3912 // Eat the end of statement. 3913 Lex(); 3914 3915 // Lex the irp definition. 3916 MCAsmMacro *M = ParseMacroLikeBody(DirectiveLoc); 3917 if (!M) 3918 return true; 3919 3920 // Macro instantiation is lexical, unfortunately. We construct a new buffer 3921 // to hold the macro body with substitutions. 3922 SmallString<256> Buf; 3923 raw_svector_ostream OS(Buf); 3924 3925 for (MCAsmMacroArguments::iterator i = A.begin(), e = A.end(); i != e; ++i) { 3926 MCAsmMacroArguments Args; 3927 Args.push_back(*i); 3928 3929 if (expandMacro(OS, M->Body, Parameters, Args, getTok().getLoc())) 3930 return true; 3931 } 3932 3933 InstantiateMacroLikeBody(M, DirectiveLoc, OS); 3934 3935 return false; 3936 } 3937 3938 /// ParseDirectiveIrpc 3939 /// ::= .irpc symbol,values 3940 bool AsmParser::ParseDirectiveIrpc(SMLoc DirectiveLoc) { 3941 MCAsmMacroParameters Parameters; 3942 MCAsmMacroParameter Parameter; 3943 3944 if (parseIdentifier(Parameter.first)) 3945 return TokError("expected identifier in '.irpc' directive"); 3946 3947 Parameters.push_back(Parameter); 3948 3949 if (Lexer.isNot(AsmToken::Comma)) 3950 return TokError("expected comma in '.irpc' directive"); 3951 3952 Lex(); 3953 3954 MCAsmMacroArguments A; 3955 if (ParseMacroArguments(0, A)) 3956 return true; 3957 3958 if (A.size() != 1 || A.front().size() != 1) 3959 return TokError("unexpected token in '.irpc' directive"); 3960 3961 // Eat the end of statement. 3962 Lex(); 3963 3964 // Lex the irpc definition. 3965 MCAsmMacro *M = ParseMacroLikeBody(DirectiveLoc); 3966 if (!M) 3967 return true; 3968 3969 // Macro instantiation is lexical, unfortunately. We construct a new buffer 3970 // to hold the macro body with substitutions. 3971 SmallString<256> Buf; 3972 raw_svector_ostream OS(Buf); 3973 3974 StringRef Values = A.front().front().getString(); 3975 std::size_t I, End = Values.size(); 3976 for (I = 0; I < End; ++I) { 3977 MCAsmMacroArgument Arg; 3978 Arg.push_back(AsmToken(AsmToken::Identifier, Values.slice(I, I+1))); 3979 3980 MCAsmMacroArguments Args; 3981 Args.push_back(Arg); 3982 3983 if (expandMacro(OS, M->Body, Parameters, Args, getTok().getLoc())) 3984 return true; 3985 } 3986 3987 InstantiateMacroLikeBody(M, DirectiveLoc, OS); 3988 3989 return false; 3990 } 3991 3992 bool AsmParser::ParseDirectiveEndr(SMLoc DirectiveLoc) { 3993 if (ActiveMacros.empty()) 3994 return TokError("unmatched '.endr' directive"); 3995 3996 // The only .repl that should get here are the ones created by 3997 // InstantiateMacroLikeBody. 3998 assert(getLexer().is(AsmToken::EndOfStatement)); 3999 4000 HandleMacroExit(); 4001 return false; 4002 } 4003 4004 bool AsmParser::ParseDirectiveMSEmit(SMLoc IDLoc, ParseStatementInfo &Info, 4005 size_t Len) { 4006 const MCExpr *Value; 4007 SMLoc ExprLoc = getLexer().getLoc(); 4008 if (parseExpression(Value)) 4009 return true; 4010 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 4011 if (!MCE) 4012 return Error(ExprLoc, "unexpected expression in _emit"); 4013 uint64_t IntValue = MCE->getValue(); 4014 if (!isUIntN(8, IntValue) && !isIntN(8, IntValue)) 4015 return Error(ExprLoc, "literal value out of range for directive"); 4016 4017 Info.AsmRewrites->push_back(AsmRewrite(AOK_Emit, IDLoc, Len)); 4018 return false; 4019 } 4020 4021 bool AsmParser::ParseDirectiveMSAlign(SMLoc IDLoc, ParseStatementInfo &Info) { 4022 const MCExpr *Value; 4023 SMLoc ExprLoc = getLexer().getLoc(); 4024 if (parseExpression(Value)) 4025 return true; 4026 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 4027 if (!MCE) 4028 return Error(ExprLoc, "unexpected expression in align"); 4029 uint64_t IntValue = MCE->getValue(); 4030 if (!isPowerOf2_64(IntValue)) 4031 return Error(ExprLoc, "literal value not a power of two greater then zero"); 4032 4033 Info.AsmRewrites->push_back(AsmRewrite(AOK_Align, IDLoc, 5, 4034 Log2_64(IntValue))); 4035 return false; 4036 } 4037 4038 // We are comparing pointers, but the pointers are relative to a single string. 4039 // Thus, this should always be deterministic. 4040 static int RewritesSort(const void *A, const void *B) { 4041 const AsmRewrite *AsmRewriteA = static_cast<const AsmRewrite *>(A); 4042 const AsmRewrite *AsmRewriteB = static_cast<const AsmRewrite *>(B); 4043 if (AsmRewriteA->Loc.getPointer() < AsmRewriteB->Loc.getPointer()) 4044 return -1; 4045 if (AsmRewriteB->Loc.getPointer() < AsmRewriteA->Loc.getPointer()) 4046 return 1; 4047 4048 // It's possible to have a SizeDirective, Imm/ImmPrefix and an Input/Output 4049 // rewrite to the same location. Make sure the SizeDirective rewrite is 4050 // performed first, then the Imm/ImmPrefix and finally the Input/Output. This 4051 // ensures the sort algorithm is stable. 4052 if (AsmRewritePrecedence [AsmRewriteA->Kind] > 4053 AsmRewritePrecedence [AsmRewriteB->Kind]) 4054 return -1; 4055 4056 if (AsmRewritePrecedence [AsmRewriteA->Kind] < 4057 AsmRewritePrecedence [AsmRewriteB->Kind]) 4058 return 1; 4059 llvm_unreachable ("Unstable rewrite sort."); 4060 } 4061 4062 bool 4063 AsmParser::parseMSInlineAsm(void *AsmLoc, std::string &AsmString, 4064 unsigned &NumOutputs, unsigned &NumInputs, 4065 SmallVectorImpl<std::pair<void *, bool> > &OpDecls, 4066 SmallVectorImpl<std::string> &Constraints, 4067 SmallVectorImpl<std::string> &Clobbers, 4068 const MCInstrInfo *MII, 4069 const MCInstPrinter *IP, 4070 MCAsmParserSemaCallback &SI) { 4071 SmallVector<void *, 4> InputDecls; 4072 SmallVector<void *, 4> OutputDecls; 4073 SmallVector<bool, 4> InputDeclsAddressOf; 4074 SmallVector<bool, 4> OutputDeclsAddressOf; 4075 SmallVector<std::string, 4> InputConstraints; 4076 SmallVector<std::string, 4> OutputConstraints; 4077 SmallVector<unsigned, 4> ClobberRegs; 4078 4079 SmallVector<AsmRewrite, 4> AsmStrRewrites; 4080 4081 // Prime the lexer. 4082 Lex(); 4083 4084 // While we have input, parse each statement. 4085 unsigned InputIdx = 0; 4086 unsigned OutputIdx = 0; 4087 while (getLexer().isNot(AsmToken::Eof)) { 4088 ParseStatementInfo Info(&AsmStrRewrites); 4089 if (ParseStatement(Info)) 4090 return true; 4091 4092 if (Info.ParseError) 4093 return true; 4094 4095 if (Info.Opcode == ~0U) 4096 continue; 4097 4098 const MCInstrDesc &Desc = MII->get(Info.Opcode); 4099 4100 // Build the list of clobbers, outputs and inputs. 4101 for (unsigned i = 1, e = Info.ParsedOperands.size(); i != e; ++i) { 4102 MCParsedAsmOperand *Operand = Info.ParsedOperands[i]; 4103 4104 // Immediate. 4105 if (Operand->isImm()) 4106 continue; 4107 4108 // Register operand. 4109 if (Operand->isReg() && !Operand->needAddressOf()) { 4110 unsigned NumDefs = Desc.getNumDefs(); 4111 // Clobber. 4112 if (NumDefs && Operand->getMCOperandNum() < NumDefs) 4113 ClobberRegs.push_back(Operand->getReg()); 4114 continue; 4115 } 4116 4117 // Expr/Input or Output. 4118 StringRef SymName = Operand->getSymName(); 4119 if (SymName.empty()) 4120 continue; 4121 4122 void *OpDecl = Operand->getOpDecl(); 4123 if (!OpDecl) 4124 continue; 4125 4126 bool isOutput = (i == 1) && Desc.mayStore(); 4127 SMLoc Start = SMLoc::getFromPointer(SymName.data()); 4128 if (isOutput) { 4129 ++InputIdx; 4130 OutputDecls.push_back(OpDecl); 4131 OutputDeclsAddressOf.push_back(Operand->needAddressOf()); 4132 OutputConstraints.push_back('=' + Operand->getConstraint().str()); 4133 AsmStrRewrites.push_back(AsmRewrite(AOK_Output, Start, SymName.size())); 4134 } else { 4135 InputDecls.push_back(OpDecl); 4136 InputDeclsAddressOf.push_back(Operand->needAddressOf()); 4137 InputConstraints.push_back(Operand->getConstraint().str()); 4138 AsmStrRewrites.push_back(AsmRewrite(AOK_Input, Start, SymName.size())); 4139 } 4140 } 4141 } 4142 4143 // Set the number of Outputs and Inputs. 4144 NumOutputs = OutputDecls.size(); 4145 NumInputs = InputDecls.size(); 4146 4147 // Set the unique clobbers. 4148 array_pod_sort(ClobberRegs.begin(), ClobberRegs.end()); 4149 ClobberRegs.erase(std::unique(ClobberRegs.begin(), ClobberRegs.end()), 4150 ClobberRegs.end()); 4151 Clobbers.assign(ClobberRegs.size(), std::string()); 4152 for (unsigned I = 0, E = ClobberRegs.size(); I != E; ++I) { 4153 raw_string_ostream OS(Clobbers[I]); 4154 IP->printRegName(OS, ClobberRegs[I]); 4155 } 4156 4157 // Merge the various outputs and inputs. Output are expected first. 4158 if (NumOutputs || NumInputs) { 4159 unsigned NumExprs = NumOutputs + NumInputs; 4160 OpDecls.resize(NumExprs); 4161 Constraints.resize(NumExprs); 4162 for (unsigned i = 0; i < NumOutputs; ++i) { 4163 OpDecls[i] = std::make_pair(OutputDecls[i], OutputDeclsAddressOf[i]); 4164 Constraints[i] = OutputConstraints[i]; 4165 } 4166 for (unsigned i = 0, j = NumOutputs; i < NumInputs; ++i, ++j) { 4167 OpDecls[j] = std::make_pair(InputDecls[i], InputDeclsAddressOf[i]); 4168 Constraints[j] = InputConstraints[i]; 4169 } 4170 } 4171 4172 // Build the IR assembly string. 4173 std::string AsmStringIR; 4174 raw_string_ostream OS(AsmStringIR); 4175 const char *AsmStart = SrcMgr.getMemoryBuffer(0)->getBufferStart(); 4176 const char *AsmEnd = SrcMgr.getMemoryBuffer(0)->getBufferEnd(); 4177 array_pod_sort(AsmStrRewrites.begin(), AsmStrRewrites.end(), RewritesSort); 4178 for (SmallVectorImpl<AsmRewrite>::iterator I = AsmStrRewrites.begin(), 4179 E = AsmStrRewrites.end(); 4180 I != E; ++I) { 4181 AsmRewriteKind Kind = (*I).Kind; 4182 if (Kind == AOK_Delete) 4183 continue; 4184 4185 const char *Loc = (*I).Loc.getPointer(); 4186 assert(Loc >= AsmStart && "Expected Loc to be at or after Start!"); 4187 4188 // Emit everything up to the immediate/expression. 4189 unsigned Len = Loc - AsmStart; 4190 if (Len) 4191 OS << StringRef(AsmStart, Len); 4192 4193 // Skip the original expression. 4194 if (Kind == AOK_Skip) { 4195 AsmStart = Loc + (*I).Len; 4196 continue; 4197 } 4198 4199 unsigned AdditionalSkip = 0; 4200 // Rewrite expressions in $N notation. 4201 switch (Kind) { 4202 default: break; 4203 case AOK_Imm: 4204 OS << "$$" << (*I).Val; 4205 break; 4206 case AOK_ImmPrefix: 4207 OS << "$$"; 4208 break; 4209 case AOK_Input: 4210 OS << '$' << InputIdx++; 4211 break; 4212 case AOK_Output: 4213 OS << '$' << OutputIdx++; 4214 break; 4215 case AOK_SizeDirective: 4216 switch ((*I).Val) { 4217 default: break; 4218 case 8: OS << "byte ptr "; break; 4219 case 16: OS << "word ptr "; break; 4220 case 32: OS << "dword ptr "; break; 4221 case 64: OS << "qword ptr "; break; 4222 case 80: OS << "xword ptr "; break; 4223 case 128: OS << "xmmword ptr "; break; 4224 case 256: OS << "ymmword ptr "; break; 4225 } 4226 break; 4227 case AOK_Emit: 4228 OS << ".byte"; 4229 break; 4230 case AOK_Align: { 4231 unsigned Val = (*I).Val; 4232 OS << ".align " << Val; 4233 4234 // Skip the original immediate. 4235 assert(Val < 10 && "Expected alignment less then 2^10."); 4236 AdditionalSkip = (Val < 4) ? 2 : Val < 7 ? 3 : 4; 4237 break; 4238 } 4239 case AOK_DotOperator: 4240 OS << (*I).Val; 4241 break; 4242 } 4243 4244 // Skip the original expression. 4245 AsmStart = Loc + (*I).Len + AdditionalSkip; 4246 } 4247 4248 // Emit the remainder of the asm string. 4249 if (AsmStart != AsmEnd) 4250 OS << StringRef(AsmStart, AsmEnd - AsmStart); 4251 4252 AsmString = OS.str(); 4253 return false; 4254 } 4255 4256 /// \brief Create an MCAsmParser instance. 4257 MCAsmParser *llvm::createMCAsmParser(SourceMgr &SM, 4258 MCContext &C, MCStreamer &Out, 4259 const MCAsmInfo &MAI) { 4260 return new AsmParser(SM, C, Out, MAI); 4261 } 4262