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 = ArrayRef<SMRange>()); 205 virtual bool Error(SMLoc L, const Twine &Msg, 206 ArrayRef<SMRange> Ranges = ArrayRef<SMRange>()); 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 parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc); 225 virtual bool parseAbsoluteExpression(int64_t &Res); 226 227 /// parseIdentifier - Parse an identifier or string (as a quoted identifier) 228 /// and set \p Res to the identifier contents. 229 virtual bool parseIdentifier(StringRef &Res); 230 virtual void eatToEndOfStatement(); 231 232 virtual void checkForValidSection(); 233 /// } 234 235 private: 236 237 bool ParseStatement(ParseStatementInfo &Info); 238 void EatToEndOfLine(); 239 bool ParseCppHashLineFilenameComment(const SMLoc &L); 240 241 void CheckForBadMacro(SMLoc DirectiveLoc, StringRef Name, StringRef Body, 242 MCAsmMacroParameters Parameters); 243 bool expandMacro(raw_svector_ostream &OS, StringRef Body, 244 const MCAsmMacroParameters &Parameters, 245 const MCAsmMacroArguments &A, 246 const SMLoc &L); 247 248 /// \brief Are macros enabled in the parser? 249 bool MacrosEnabled() {return MacrosEnabledFlag;} 250 251 /// \brief Control a flag in the parser that enables or disables macros. 252 void SetMacrosEnabled(bool Flag) {MacrosEnabledFlag = Flag;} 253 254 /// \brief Lookup a previously defined macro. 255 /// \param Name Macro name. 256 /// \returns Pointer to macro. NULL if no such macro was defined. 257 const MCAsmMacro* LookupMacro(StringRef Name); 258 259 /// \brief Define a new macro with the given name and information. 260 void DefineMacro(StringRef Name, const MCAsmMacro& Macro); 261 262 /// \brief Undefine a macro. If no such macro was defined, it's a no-op. 263 void UndefineMacro(StringRef Name); 264 265 /// \brief Are we inside a macro instantiation? 266 bool InsideMacroInstantiation() {return !ActiveMacros.empty();} 267 268 /// \brief Handle entry to macro instantiation. 269 /// 270 /// \param M The macro. 271 /// \param NameLoc Instantiation location. 272 bool HandleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc); 273 274 /// \brief Handle exit from macro instantiation. 275 void HandleMacroExit(); 276 277 /// \brief Extract AsmTokens for a macro argument. If the argument delimiter 278 /// is initially unknown, set it to AsmToken::Eof. It will be set to the 279 /// correct delimiter by the method. 280 bool ParseMacroArgument(MCAsmMacroArgument &MA, 281 AsmToken::TokenKind &ArgumentDelimiter); 282 283 /// \brief Parse all macro arguments for a given macro. 284 bool ParseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A); 285 286 void PrintMacroInstantiations(); 287 void PrintMessage(SMLoc Loc, SourceMgr::DiagKind Kind, const Twine &Msg, 288 ArrayRef<SMRange> Ranges = ArrayRef<SMRange>()) const { 289 SrcMgr.PrintMessage(Loc, Kind, Msg, Ranges); 290 } 291 static void DiagHandler(const SMDiagnostic &Diag, void *Context); 292 293 /// EnterIncludeFile - Enter the specified file. This returns true on failure. 294 bool EnterIncludeFile(const std::string &Filename); 295 /// ProcessIncbinFile - Process the specified file for the .incbin directive. 296 /// This returns true on failure. 297 bool ProcessIncbinFile(const std::string &Filename); 298 299 /// \brief Reset the current lexer position to that given by \p Loc. The 300 /// current token is not set; clients should ensure Lex() is called 301 /// subsequently. 302 /// 303 /// \param InBuffer If not -1, should be the known buffer id that contains the 304 /// location. 305 void JumpToLoc(SMLoc Loc, int InBuffer=-1); 306 307 /// \brief Parse up to the end of statement and a return the contents from the 308 /// current token until the end of the statement; the current token on exit 309 /// will be either the EndOfStatement or EOF. 310 virtual StringRef parseStringToEndOfStatement(); 311 312 /// \brief Parse until the end of a statement or a comma is encountered, 313 /// return the contents from the current token up to the end or comma. 314 StringRef ParseStringToComma(); 315 316 bool ParseAssignment(StringRef Name, bool allow_redef, 317 bool NoDeadStrip = false); 318 319 bool ParsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc); 320 bool ParseBinOpRHS(unsigned Precedence, const MCExpr *&Res, SMLoc &EndLoc); 321 bool ParseParenExpr(const MCExpr *&Res, SMLoc &EndLoc); 322 bool ParseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc); 323 324 bool ParseRegisterOrRegisterNumber(int64_t &Register, SMLoc DirectiveLoc); 325 326 // Generic (target and platform independent) directive parsing. 327 enum DirectiveKind { 328 DK_NO_DIRECTIVE, // Placeholder 329 DK_SET, DK_EQU, DK_EQUIV, DK_ASCII, DK_ASCIZ, DK_STRING, DK_BYTE, DK_SHORT, 330 DK_VALUE, DK_2BYTE, DK_LONG, DK_INT, DK_4BYTE, DK_QUAD, DK_8BYTE, DK_SINGLE, 331 DK_FLOAT, DK_DOUBLE, DK_ALIGN, DK_ALIGN32, DK_BALIGN, DK_BALIGNW, 332 DK_BALIGNL, DK_P2ALIGN, DK_P2ALIGNW, DK_P2ALIGNL, DK_ORG, DK_FILL, DK_ENDR, 333 DK_BUNDLE_ALIGN_MODE, DK_BUNDLE_LOCK, DK_BUNDLE_UNLOCK, 334 DK_ZERO, DK_EXTERN, DK_GLOBL, DK_GLOBAL, DK_INDIRECT_SYMBOL, 335 DK_LAZY_REFERENCE, DK_NO_DEAD_STRIP, DK_SYMBOL_RESOLVER, DK_PRIVATE_EXTERN, 336 DK_REFERENCE, DK_WEAK_DEFINITION, DK_WEAK_REFERENCE, 337 DK_WEAK_DEF_CAN_BE_HIDDEN, DK_COMM, DK_COMMON, DK_LCOMM, DK_ABORT, 338 DK_INCLUDE, DK_INCBIN, DK_CODE16, DK_CODE16GCC, DK_REPT, DK_IRP, DK_IRPC, 339 DK_IF, DK_IFB, DK_IFNB, DK_IFC, DK_IFNC, DK_IFDEF, DK_IFNDEF, DK_IFNOTDEF, 340 DK_ELSEIF, DK_ELSE, DK_ENDIF, 341 DK_SPACE, DK_SKIP, DK_FILE, DK_LINE, DK_LOC, DK_STABS, 342 DK_CFI_SECTIONS, DK_CFI_STARTPROC, DK_CFI_ENDPROC, DK_CFI_DEF_CFA, 343 DK_CFI_DEF_CFA_OFFSET, DK_CFI_ADJUST_CFA_OFFSET, DK_CFI_DEF_CFA_REGISTER, 344 DK_CFI_OFFSET, DK_CFI_REL_OFFSET, DK_CFI_PERSONALITY, DK_CFI_LSDA, 345 DK_CFI_REMEMBER_STATE, DK_CFI_RESTORE_STATE, DK_CFI_SAME_VALUE, 346 DK_CFI_RESTORE, DK_CFI_ESCAPE, DK_CFI_SIGNAL_FRAME, DK_CFI_UNDEFINED, 347 DK_CFI_REGISTER, 348 DK_MACROS_ON, DK_MACROS_OFF, DK_MACRO, DK_ENDM, DK_ENDMACRO, DK_PURGEM, 349 DK_SLEB128, DK_ULEB128 350 }; 351 352 /// DirectiveKindMap - Maps directive name --> DirectiveKind enum, for 353 /// directives parsed by this class. 354 StringMap<DirectiveKind> DirectiveKindMap; 355 356 // ".ascii", ".asciz", ".string" 357 bool ParseDirectiveAscii(StringRef IDVal, bool ZeroTerminated); 358 bool ParseDirectiveValue(unsigned Size); // ".byte", ".long", ... 359 bool ParseDirectiveRealValue(const fltSemantics &); // ".single", ... 360 bool ParseDirectiveFill(); // ".fill" 361 bool ParseDirectiveZero(); // ".zero" 362 // ".set", ".equ", ".equiv" 363 bool ParseDirectiveSet(StringRef IDVal, bool allow_redef); 364 bool ParseDirectiveOrg(); // ".org" 365 // ".align{,32}", ".p2align{,w,l}" 366 bool ParseDirectiveAlign(bool IsPow2, unsigned ValueSize); 367 368 // ".file", ".line", ".loc", ".stabs" 369 bool ParseDirectiveFile(SMLoc DirectiveLoc); 370 bool ParseDirectiveLine(); 371 bool ParseDirectiveLoc(); 372 bool ParseDirectiveStabs(); 373 374 // .cfi directives 375 bool ParseDirectiveCFIRegister(SMLoc DirectiveLoc); 376 bool ParseDirectiveCFISections(); 377 bool ParseDirectiveCFIStartProc(); 378 bool ParseDirectiveCFIEndProc(); 379 bool ParseDirectiveCFIDefCfaOffset(); 380 bool ParseDirectiveCFIDefCfa(SMLoc DirectiveLoc); 381 bool ParseDirectiveCFIAdjustCfaOffset(); 382 bool ParseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc); 383 bool ParseDirectiveCFIOffset(SMLoc DirectiveLoc); 384 bool ParseDirectiveCFIRelOffset(SMLoc DirectiveLoc); 385 bool ParseDirectiveCFIPersonalityOrLsda(bool IsPersonality); 386 bool ParseDirectiveCFIRememberState(); 387 bool ParseDirectiveCFIRestoreState(); 388 bool ParseDirectiveCFISameValue(SMLoc DirectiveLoc); 389 bool ParseDirectiveCFIRestore(SMLoc DirectiveLoc); 390 bool ParseDirectiveCFIEscape(); 391 bool ParseDirectiveCFISignalFrame(); 392 bool ParseDirectiveCFIUndefined(SMLoc DirectiveLoc); 393 394 // macro directives 395 bool ParseDirectivePurgeMacro(SMLoc DirectiveLoc); 396 bool ParseDirectiveEndMacro(StringRef Directive); 397 bool ParseDirectiveMacro(SMLoc DirectiveLoc); 398 bool ParseDirectiveMacrosOnOff(StringRef Directive); 399 400 // ".bundle_align_mode" 401 bool ParseDirectiveBundleAlignMode(); 402 // ".bundle_lock" 403 bool ParseDirectiveBundleLock(); 404 // ".bundle_unlock" 405 bool ParseDirectiveBundleUnlock(); 406 407 // ".space", ".skip" 408 bool ParseDirectiveSpace(StringRef IDVal); 409 410 // .sleb128 (Signed=true) and .uleb128 (Signed=false) 411 bool ParseDirectiveLEB128(bool Signed); 412 413 /// ParseDirectiveSymbolAttribute - Parse a directive like ".globl" which 414 /// accepts a single symbol (which should be a label or an external). 415 bool ParseDirectiveSymbolAttribute(MCSymbolAttr Attr); 416 417 bool ParseDirectiveComm(bool IsLocal); // ".comm" and ".lcomm" 418 419 bool ParseDirectiveAbort(); // ".abort" 420 bool ParseDirectiveInclude(); // ".include" 421 bool ParseDirectiveIncbin(); // ".incbin" 422 423 bool ParseDirectiveIf(SMLoc DirectiveLoc); // ".if" 424 // ".ifb" or ".ifnb", depending on ExpectBlank. 425 bool ParseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank); 426 // ".ifc" or ".ifnc", depending on ExpectEqual. 427 bool ParseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual); 428 // ".ifdef" or ".ifndef", depending on expect_defined 429 bool ParseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined); 430 bool ParseDirectiveElseIf(SMLoc DirectiveLoc); // ".elseif" 431 bool ParseDirectiveElse(SMLoc DirectiveLoc); // ".else" 432 bool ParseDirectiveEndIf(SMLoc DirectiveLoc); // .endif 433 virtual bool parseEscapedString(std::string &Data); 434 435 const MCExpr *ApplyModifierToExpr(const MCExpr *E, 436 MCSymbolRefExpr::VariantKind Variant); 437 438 // Macro-like directives 439 MCAsmMacro *ParseMacroLikeBody(SMLoc DirectiveLoc); 440 void InstantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 441 raw_svector_ostream &OS); 442 bool ParseDirectiveRept(SMLoc DirectiveLoc); // ".rept" 443 bool ParseDirectiveIrp(SMLoc DirectiveLoc); // ".irp" 444 bool ParseDirectiveIrpc(SMLoc DirectiveLoc); // ".irpc" 445 bool ParseDirectiveEndr(SMLoc DirectiveLoc); // ".endr" 446 447 // "_emit" or "__emit" 448 bool ParseDirectiveMSEmit(SMLoc DirectiveLoc, ParseStatementInfo &Info, 449 size_t Len); 450 451 // "align" 452 bool ParseDirectiveMSAlign(SMLoc DirectiveLoc, ParseStatementInfo &Info); 453 454 void initializeDirectiveKindMap(); 455 }; 456 } 457 458 namespace llvm { 459 460 extern MCAsmParserExtension *createDarwinAsmParser(); 461 extern MCAsmParserExtension *createELFAsmParser(); 462 extern MCAsmParserExtension *createCOFFAsmParser(); 463 464 } 465 466 enum { DEFAULT_ADDRSPACE = 0 }; 467 468 AsmParser::AsmParser(SourceMgr &_SM, MCContext &_Ctx, 469 MCStreamer &_Out, const MCAsmInfo &_MAI) 470 : Lexer(_MAI), Ctx(_Ctx), Out(_Out), MAI(_MAI), SrcMgr(_SM), 471 PlatformParser(0), 472 CurBuffer(0), MacrosEnabledFlag(true), CppHashLineNumber(0), 473 AssemblerDialect(~0U), IsDarwin(false), ParsingInlineAsm(false) { 474 // Save the old handler. 475 SavedDiagHandler = SrcMgr.getDiagHandler(); 476 SavedDiagContext = SrcMgr.getDiagContext(); 477 // Set our own handler which calls the saved handler. 478 SrcMgr.setDiagHandler(DiagHandler, this); 479 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)); 480 481 // Initialize the platform / file format parser. 482 // 483 // FIXME: This is a hack, we need to (majorly) cleanup how these objects are 484 // created. 485 if (_MAI.hasMicrosoftFastStdCallMangling()) { 486 PlatformParser = createCOFFAsmParser(); 487 PlatformParser->Initialize(*this); 488 } else if (_MAI.hasSubsectionsViaSymbols()) { 489 PlatformParser = createDarwinAsmParser(); 490 PlatformParser->Initialize(*this); 491 IsDarwin = true; 492 } else { 493 PlatformParser = createELFAsmParser(); 494 PlatformParser->Initialize(*this); 495 } 496 497 initializeDirectiveKindMap(); 498 } 499 500 AsmParser::~AsmParser() { 501 assert(ActiveMacros.empty() && "Unexpected active macro instantiation!"); 502 503 // Destroy any macros. 504 for (StringMap<MCAsmMacro*>::iterator it = MacroMap.begin(), 505 ie = MacroMap.end(); it != ie; ++it) 506 delete it->getValue(); 507 508 delete PlatformParser; 509 } 510 511 void AsmParser::PrintMacroInstantiations() { 512 // Print the active macro instantiation stack. 513 for (std::vector<MacroInstantiation*>::const_reverse_iterator 514 it = ActiveMacros.rbegin(), ie = ActiveMacros.rend(); it != ie; ++it) 515 PrintMessage((*it)->InstantiationLoc, SourceMgr::DK_Note, 516 "while in macro instantiation"); 517 } 518 519 bool AsmParser::Warning(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) { 520 if (FatalAssemblerWarnings) 521 return Error(L, Msg, Ranges); 522 PrintMessage(L, SourceMgr::DK_Warning, Msg, Ranges); 523 PrintMacroInstantiations(); 524 return false; 525 } 526 527 bool AsmParser::Error(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) { 528 HadError = true; 529 PrintMessage(L, SourceMgr::DK_Error, Msg, Ranges); 530 PrintMacroInstantiations(); 531 return true; 532 } 533 534 bool AsmParser::EnterIncludeFile(const std::string &Filename) { 535 std::string IncludedFile; 536 int NewBuf = SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 537 if (NewBuf == -1) 538 return true; 539 540 CurBuffer = NewBuf; 541 542 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)); 543 544 return false; 545 } 546 547 /// Process the specified .incbin file by seaching for it in the include paths 548 /// then just emitting the byte contents of the file to the streamer. This 549 /// returns true on failure. 550 bool AsmParser::ProcessIncbinFile(const std::string &Filename) { 551 std::string IncludedFile; 552 int NewBuf = SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 553 if (NewBuf == -1) 554 return true; 555 556 // Pick up the bytes from the file and emit them. 557 getStreamer().EmitBytes(SrcMgr.getMemoryBuffer(NewBuf)->getBuffer(), 558 DEFAULT_ADDRSPACE); 559 return false; 560 } 561 562 void AsmParser::JumpToLoc(SMLoc Loc, int InBuffer) { 563 if (InBuffer != -1) { 564 CurBuffer = InBuffer; 565 } else { 566 CurBuffer = SrcMgr.FindBufferContainingLoc(Loc); 567 } 568 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer), Loc.getPointer()); 569 } 570 571 const AsmToken &AsmParser::Lex() { 572 const AsmToken *tok = &Lexer.Lex(); 573 574 if (tok->is(AsmToken::Eof)) { 575 // If this is the end of an included file, pop the parent file off the 576 // include stack. 577 SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer); 578 if (ParentIncludeLoc != SMLoc()) { 579 JumpToLoc(ParentIncludeLoc); 580 tok = &Lexer.Lex(); 581 } 582 } 583 584 if (tok->is(AsmToken::Error)) 585 Error(Lexer.getErrLoc(), Lexer.getErr()); 586 587 return *tok; 588 } 589 590 bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) { 591 // Create the initial section, if requested. 592 if (!NoInitialTextSection) 593 Out.InitSections(); 594 595 // Prime the lexer. 596 Lex(); 597 598 HadError = false; 599 AsmCond StartingCondState = TheCondState; 600 601 // If we are generating dwarf for assembly source files save the initial text 602 // section and generate a .file directive. 603 if (getContext().getGenDwarfForAssembly()) { 604 getContext().setGenDwarfSection(getStreamer().getCurrentSection()); 605 MCSymbol *SectionStartSym = getContext().CreateTempSymbol(); 606 getStreamer().EmitLabel(SectionStartSym); 607 getContext().setGenDwarfSectionStartSym(SectionStartSym); 608 getStreamer().EmitDwarfFileDirective(getContext().nextGenDwarfFileNumber(), 609 StringRef(), 610 getContext().getMainFileName()); 611 } 612 613 // While we have input, parse each statement. 614 while (Lexer.isNot(AsmToken::Eof)) { 615 ParseStatementInfo Info; 616 if (!ParseStatement(Info)) continue; 617 618 // We had an error, validate that one was emitted and recover by skipping to 619 // the next line. 620 assert(HadError && "Parse statement returned an error, but none emitted!"); 621 eatToEndOfStatement(); 622 } 623 624 if (TheCondState.TheCond != StartingCondState.TheCond || 625 TheCondState.Ignore != StartingCondState.Ignore) 626 return TokError("unmatched .ifs or .elses"); 627 628 // Check to see there are no empty DwarfFile slots. 629 const SmallVectorImpl<MCDwarfFile *> &MCDwarfFiles = 630 getContext().getMCDwarfFiles(); 631 for (unsigned i = 1; i < MCDwarfFiles.size(); i++) { 632 if (!MCDwarfFiles[i]) 633 TokError("unassigned file number: " + Twine(i) + " for .file directives"); 634 } 635 636 // Check to see that all assembler local symbols were actually defined. 637 // Targets that don't do subsections via symbols may not want this, though, 638 // so conservatively exclude them. Only do this if we're finalizing, though, 639 // as otherwise we won't necessarilly have seen everything yet. 640 if (!NoFinalize && MAI.hasSubsectionsViaSymbols()) { 641 const MCContext::SymbolTable &Symbols = getContext().getSymbols(); 642 for (MCContext::SymbolTable::const_iterator i = Symbols.begin(), 643 e = Symbols.end(); 644 i != e; ++i) { 645 MCSymbol *Sym = i->getValue(); 646 // Variable symbols may not be marked as defined, so check those 647 // explicitly. If we know it's a variable, we have a definition for 648 // the purposes of this check. 649 if (Sym->isTemporary() && !Sym->isVariable() && !Sym->isDefined()) 650 // FIXME: We would really like to refer back to where the symbol was 651 // first referenced for a source location. We need to add something 652 // to track that. Currently, we just point to the end of the file. 653 PrintMessage(getLexer().getLoc(), SourceMgr::DK_Error, 654 "assembler local symbol '" + Sym->getName() + 655 "' not defined"); 656 } 657 } 658 659 660 // Finalize the output stream if there are no errors and if the client wants 661 // us to. 662 if (!HadError && !NoFinalize) 663 Out.Finish(); 664 665 return HadError; 666 } 667 668 void AsmParser::checkForValidSection() { 669 if (!ParsingInlineAsm && !getStreamer().getCurrentSection()) { 670 TokError("expected section directive before assembly directive"); 671 Out.InitToTextSection(); 672 } 673 } 674 675 /// eatToEndOfStatement - Throw away the rest of the line for testing purposes. 676 void AsmParser::eatToEndOfStatement() { 677 while (Lexer.isNot(AsmToken::EndOfStatement) && 678 Lexer.isNot(AsmToken::Eof)) 679 Lex(); 680 681 // Eat EOL. 682 if (Lexer.is(AsmToken::EndOfStatement)) 683 Lex(); 684 } 685 686 StringRef AsmParser::parseStringToEndOfStatement() { 687 const char *Start = getTok().getLoc().getPointer(); 688 689 while (Lexer.isNot(AsmToken::EndOfStatement) && 690 Lexer.isNot(AsmToken::Eof)) 691 Lex(); 692 693 const char *End = getTok().getLoc().getPointer(); 694 return StringRef(Start, End - Start); 695 } 696 697 StringRef AsmParser::ParseStringToComma() { 698 const char *Start = getTok().getLoc().getPointer(); 699 700 while (Lexer.isNot(AsmToken::EndOfStatement) && 701 Lexer.isNot(AsmToken::Comma) && 702 Lexer.isNot(AsmToken::Eof)) 703 Lex(); 704 705 const char *End = getTok().getLoc().getPointer(); 706 return StringRef(Start, End - Start); 707 } 708 709 /// ParseParenExpr - Parse a paren expression and return it. 710 /// NOTE: This assumes the leading '(' has already been consumed. 711 /// 712 /// parenexpr ::= expr) 713 /// 714 bool AsmParser::ParseParenExpr(const MCExpr *&Res, SMLoc &EndLoc) { 715 if (parseExpression(Res)) return true; 716 if (Lexer.isNot(AsmToken::RParen)) 717 return TokError("expected ')' in parentheses expression"); 718 EndLoc = Lexer.getTok().getEndLoc(); 719 Lex(); 720 return false; 721 } 722 723 /// ParseBracketExpr - Parse a bracket expression and return it. 724 /// NOTE: This assumes the leading '[' has already been consumed. 725 /// 726 /// bracketexpr ::= expr] 727 /// 728 bool AsmParser::ParseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc) { 729 if (parseExpression(Res)) return true; 730 if (Lexer.isNot(AsmToken::RBrac)) 731 return TokError("expected ']' in brackets expression"); 732 EndLoc = Lexer.getTok().getEndLoc(); 733 Lex(); 734 return false; 735 } 736 737 /// ParsePrimaryExpr - Parse a primary expression and return it. 738 /// primaryexpr ::= (parenexpr 739 /// primaryexpr ::= symbol 740 /// primaryexpr ::= number 741 /// primaryexpr ::= '.' 742 /// primaryexpr ::= ~,+,- primaryexpr 743 bool AsmParser::ParsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) { 744 SMLoc FirstTokenLoc = getLexer().getLoc(); 745 AsmToken::TokenKind FirstTokenKind = Lexer.getKind(); 746 switch (FirstTokenKind) { 747 default: 748 return TokError("unknown token in expression"); 749 // If we have an error assume that we've already handled it. 750 case AsmToken::Error: 751 return true; 752 case AsmToken::Exclaim: 753 Lex(); // Eat the operator. 754 if (ParsePrimaryExpr(Res, EndLoc)) 755 return true; 756 Res = MCUnaryExpr::CreateLNot(Res, getContext()); 757 return false; 758 case AsmToken::Dollar: 759 case AsmToken::String: 760 case AsmToken::Identifier: { 761 StringRef Identifier; 762 if (parseIdentifier(Identifier)) { 763 if (FirstTokenKind == AsmToken::Dollar) 764 return Error(FirstTokenLoc, "invalid token in expression"); 765 return true; 766 } 767 768 EndLoc = SMLoc::getFromPointer(Identifier.end()); 769 770 // This is a symbol reference. 771 std::pair<StringRef, StringRef> Split = Identifier.split('@'); 772 MCSymbol *Sym = getContext().GetOrCreateSymbol(Split.first); 773 774 // Lookup the symbol variant if used. 775 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 776 if (Split.first.size() != Identifier.size()) { 777 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second); 778 if (Variant == MCSymbolRefExpr::VK_Invalid) { 779 Variant = MCSymbolRefExpr::VK_None; 780 return TokError("invalid variant '" + Split.second + "'"); 781 } 782 } 783 784 // If this is an absolute variable reference, substitute it now to preserve 785 // semantics in the face of reassignment. 786 if (Sym->isVariable() && isa<MCConstantExpr>(Sym->getVariableValue())) { 787 if (Variant) 788 return Error(EndLoc, "unexpected modifier on variable reference"); 789 790 Res = Sym->getVariableValue(); 791 return false; 792 } 793 794 // Otherwise create a symbol ref. 795 Res = MCSymbolRefExpr::Create(Sym, Variant, getContext()); 796 return false; 797 } 798 case AsmToken::Integer: { 799 SMLoc Loc = getTok().getLoc(); 800 int64_t IntVal = getTok().getIntVal(); 801 Res = MCConstantExpr::Create(IntVal, getContext()); 802 EndLoc = Lexer.getTok().getEndLoc(); 803 Lex(); // Eat token. 804 // Look for 'b' or 'f' following an Integer as a directional label 805 if (Lexer.getKind() == AsmToken::Identifier) { 806 StringRef IDVal = getTok().getString(); 807 if (IDVal == "f" || IDVal == "b"){ 808 MCSymbol *Sym = Ctx.GetDirectionalLocalSymbol(IntVal, 809 IDVal == "f" ? 1 : 0); 810 Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, 811 getContext()); 812 if (IDVal == "b" && Sym->isUndefined()) 813 return Error(Loc, "invalid reference to undefined symbol"); 814 EndLoc = Lexer.getTok().getEndLoc(); 815 Lex(); // Eat identifier. 816 } 817 } 818 return false; 819 } 820 case AsmToken::Real: { 821 APFloat RealVal(APFloat::IEEEdouble, getTok().getString()); 822 uint64_t IntVal = RealVal.bitcastToAPInt().getZExtValue(); 823 Res = MCConstantExpr::Create(IntVal, getContext()); 824 EndLoc = Lexer.getTok().getEndLoc(); 825 Lex(); // Eat token. 826 return false; 827 } 828 case AsmToken::Dot: { 829 // This is a '.' reference, which references the current PC. Emit a 830 // temporary label to the streamer and refer to it. 831 MCSymbol *Sym = Ctx.CreateTempSymbol(); 832 Out.EmitLabel(Sym); 833 Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, getContext()); 834 EndLoc = Lexer.getTok().getEndLoc(); 835 Lex(); // Eat identifier. 836 return false; 837 } 838 case AsmToken::LParen: 839 Lex(); // Eat the '('. 840 return ParseParenExpr(Res, EndLoc); 841 case AsmToken::LBrac: 842 if (!PlatformParser->HasBracketExpressions()) 843 return TokError("brackets expression not supported on this target"); 844 Lex(); // Eat the '['. 845 return ParseBracketExpr(Res, EndLoc); 846 case AsmToken::Minus: 847 Lex(); // Eat the operator. 848 if (ParsePrimaryExpr(Res, EndLoc)) 849 return true; 850 Res = MCUnaryExpr::CreateMinus(Res, getContext()); 851 return false; 852 case AsmToken::Plus: 853 Lex(); // Eat the operator. 854 if (ParsePrimaryExpr(Res, EndLoc)) 855 return true; 856 Res = MCUnaryExpr::CreatePlus(Res, getContext()); 857 return false; 858 case AsmToken::Tilde: 859 Lex(); // Eat the operator. 860 if (ParsePrimaryExpr(Res, EndLoc)) 861 return true; 862 Res = MCUnaryExpr::CreateNot(Res, getContext()); 863 return false; 864 } 865 } 866 867 bool AsmParser::parseExpression(const MCExpr *&Res) { 868 SMLoc EndLoc; 869 return parseExpression(Res, EndLoc); 870 } 871 872 const MCExpr * 873 AsmParser::ApplyModifierToExpr(const MCExpr *E, 874 MCSymbolRefExpr::VariantKind Variant) { 875 // Recurse over the given expression, rebuilding it to apply the given variant 876 // if there is exactly one symbol. 877 switch (E->getKind()) { 878 case MCExpr::Target: 879 case MCExpr::Constant: 880 return 0; 881 882 case MCExpr::SymbolRef: { 883 const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(E); 884 885 if (SRE->getKind() != MCSymbolRefExpr::VK_None) { 886 TokError("invalid variant on expression '" + 887 getTok().getIdentifier() + "' (already modified)"); 888 return E; 889 } 890 891 return MCSymbolRefExpr::Create(&SRE->getSymbol(), Variant, getContext()); 892 } 893 894 case MCExpr::Unary: { 895 const MCUnaryExpr *UE = cast<MCUnaryExpr>(E); 896 const MCExpr *Sub = ApplyModifierToExpr(UE->getSubExpr(), Variant); 897 if (!Sub) 898 return 0; 899 return MCUnaryExpr::Create(UE->getOpcode(), Sub, getContext()); 900 } 901 902 case MCExpr::Binary: { 903 const MCBinaryExpr *BE = cast<MCBinaryExpr>(E); 904 const MCExpr *LHS = ApplyModifierToExpr(BE->getLHS(), Variant); 905 const MCExpr *RHS = ApplyModifierToExpr(BE->getRHS(), Variant); 906 907 if (!LHS && !RHS) 908 return 0; 909 910 if (!LHS) LHS = BE->getLHS(); 911 if (!RHS) RHS = BE->getRHS(); 912 913 return MCBinaryExpr::Create(BE->getOpcode(), LHS, RHS, getContext()); 914 } 915 } 916 917 llvm_unreachable("Invalid expression kind!"); 918 } 919 920 /// parseExpression - Parse an expression and return it. 921 /// 922 /// expr ::= expr &&,|| expr -> lowest. 923 /// expr ::= expr |,^,&,! expr 924 /// expr ::= expr ==,!=,<>,<,<=,>,>= expr 925 /// expr ::= expr <<,>> expr 926 /// expr ::= expr +,- expr 927 /// expr ::= expr *,/,% expr -> highest. 928 /// expr ::= primaryexpr 929 /// 930 bool AsmParser::parseExpression(const MCExpr *&Res, SMLoc &EndLoc) { 931 // Parse the expression. 932 Res = 0; 933 if (ParsePrimaryExpr(Res, EndLoc) || ParseBinOpRHS(1, Res, EndLoc)) 934 return true; 935 936 // As a special case, we support 'a op b @ modifier' by rewriting the 937 // expression to include the modifier. This is inefficient, but in general we 938 // expect users to use 'a@modifier op b'. 939 if (Lexer.getKind() == AsmToken::At) { 940 Lex(); 941 942 if (Lexer.isNot(AsmToken::Identifier)) 943 return TokError("unexpected symbol modifier following '@'"); 944 945 MCSymbolRefExpr::VariantKind Variant = 946 MCSymbolRefExpr::getVariantKindForName(getTok().getIdentifier()); 947 if (Variant == MCSymbolRefExpr::VK_Invalid) 948 return TokError("invalid variant '" + getTok().getIdentifier() + "'"); 949 950 const MCExpr *ModifiedRes = ApplyModifierToExpr(Res, Variant); 951 if (!ModifiedRes) { 952 return TokError("invalid modifier '" + getTok().getIdentifier() + 953 "' (no symbols present)"); 954 } 955 956 Res = ModifiedRes; 957 Lex(); 958 } 959 960 // Try to constant fold it up front, if possible. 961 int64_t Value; 962 if (Res->EvaluateAsAbsolute(Value)) 963 Res = MCConstantExpr::Create(Value, getContext()); 964 965 return false; 966 } 967 968 bool AsmParser::parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) { 969 Res = 0; 970 return ParseParenExpr(Res, EndLoc) || 971 ParseBinOpRHS(1, Res, EndLoc); 972 } 973 974 bool AsmParser::parseAbsoluteExpression(int64_t &Res) { 975 const MCExpr *Expr; 976 977 SMLoc StartLoc = Lexer.getLoc(); 978 if (parseExpression(Expr)) 979 return true; 980 981 if (!Expr->EvaluateAsAbsolute(Res)) 982 return Error(StartLoc, "expected absolute expression"); 983 984 return false; 985 } 986 987 static unsigned getBinOpPrecedence(AsmToken::TokenKind K, 988 MCBinaryExpr::Opcode &Kind) { 989 switch (K) { 990 default: 991 return 0; // not a binop. 992 993 // Lowest Precedence: &&, || 994 case AsmToken::AmpAmp: 995 Kind = MCBinaryExpr::LAnd; 996 return 1; 997 case AsmToken::PipePipe: 998 Kind = MCBinaryExpr::LOr; 999 return 1; 1000 1001 1002 // Low Precedence: |, &, ^ 1003 // 1004 // FIXME: gas seems to support '!' as an infix operator? 1005 case AsmToken::Pipe: 1006 Kind = MCBinaryExpr::Or; 1007 return 2; 1008 case AsmToken::Caret: 1009 Kind = MCBinaryExpr::Xor; 1010 return 2; 1011 case AsmToken::Amp: 1012 Kind = MCBinaryExpr::And; 1013 return 2; 1014 1015 // Low Intermediate Precedence: ==, !=, <>, <, <=, >, >= 1016 case AsmToken::EqualEqual: 1017 Kind = MCBinaryExpr::EQ; 1018 return 3; 1019 case AsmToken::ExclaimEqual: 1020 case AsmToken::LessGreater: 1021 Kind = MCBinaryExpr::NE; 1022 return 3; 1023 case AsmToken::Less: 1024 Kind = MCBinaryExpr::LT; 1025 return 3; 1026 case AsmToken::LessEqual: 1027 Kind = MCBinaryExpr::LTE; 1028 return 3; 1029 case AsmToken::Greater: 1030 Kind = MCBinaryExpr::GT; 1031 return 3; 1032 case AsmToken::GreaterEqual: 1033 Kind = MCBinaryExpr::GTE; 1034 return 3; 1035 1036 // Intermediate Precedence: <<, >> 1037 case AsmToken::LessLess: 1038 Kind = MCBinaryExpr::Shl; 1039 return 4; 1040 case AsmToken::GreaterGreater: 1041 Kind = MCBinaryExpr::Shr; 1042 return 4; 1043 1044 // High Intermediate Precedence: +, - 1045 case AsmToken::Plus: 1046 Kind = MCBinaryExpr::Add; 1047 return 5; 1048 case AsmToken::Minus: 1049 Kind = MCBinaryExpr::Sub; 1050 return 5; 1051 1052 // Highest Precedence: *, /, % 1053 case AsmToken::Star: 1054 Kind = MCBinaryExpr::Mul; 1055 return 6; 1056 case AsmToken::Slash: 1057 Kind = MCBinaryExpr::Div; 1058 return 6; 1059 case AsmToken::Percent: 1060 Kind = MCBinaryExpr::Mod; 1061 return 6; 1062 } 1063 } 1064 1065 1066 /// ParseBinOpRHS - Parse all binary operators with precedence >= 'Precedence'. 1067 /// Res contains the LHS of the expression on input. 1068 bool AsmParser::ParseBinOpRHS(unsigned Precedence, const MCExpr *&Res, 1069 SMLoc &EndLoc) { 1070 while (1) { 1071 MCBinaryExpr::Opcode Kind = MCBinaryExpr::Add; 1072 unsigned TokPrec = getBinOpPrecedence(Lexer.getKind(), Kind); 1073 1074 // If the next token is lower precedence than we are allowed to eat, return 1075 // successfully with what we ate already. 1076 if (TokPrec < Precedence) 1077 return false; 1078 1079 Lex(); 1080 1081 // Eat the next primary expression. 1082 const MCExpr *RHS; 1083 if (ParsePrimaryExpr(RHS, EndLoc)) return true; 1084 1085 // If BinOp binds less tightly with RHS than the operator after RHS, let 1086 // the pending operator take RHS as its LHS. 1087 MCBinaryExpr::Opcode Dummy; 1088 unsigned NextTokPrec = getBinOpPrecedence(Lexer.getKind(), Dummy); 1089 if (TokPrec < NextTokPrec) { 1090 if (ParseBinOpRHS(Precedence+1, RHS, EndLoc)) return true; 1091 } 1092 1093 // Merge LHS and RHS according to operator. 1094 Res = MCBinaryExpr::Create(Kind, Res, RHS, getContext()); 1095 } 1096 } 1097 1098 /// ParseStatement: 1099 /// ::= EndOfStatement 1100 /// ::= Label* Directive ...Operands... EndOfStatement 1101 /// ::= Label* Identifier OperandList* EndOfStatement 1102 bool AsmParser::ParseStatement(ParseStatementInfo &Info) { 1103 if (Lexer.is(AsmToken::EndOfStatement)) { 1104 Out.AddBlankLine(); 1105 Lex(); 1106 return false; 1107 } 1108 1109 // Statements always start with an identifier or are a full line comment. 1110 AsmToken ID = getTok(); 1111 SMLoc IDLoc = ID.getLoc(); 1112 StringRef IDVal; 1113 int64_t LocalLabelVal = -1; 1114 // A full line comment is a '#' as the first token. 1115 if (Lexer.is(AsmToken::Hash)) 1116 return ParseCppHashLineFilenameComment(IDLoc); 1117 1118 // Allow an integer followed by a ':' as a directional local label. 1119 if (Lexer.is(AsmToken::Integer)) { 1120 LocalLabelVal = getTok().getIntVal(); 1121 if (LocalLabelVal < 0) { 1122 if (!TheCondState.Ignore) 1123 return TokError("unexpected token at start of statement"); 1124 IDVal = ""; 1125 } else { 1126 IDVal = getTok().getString(); 1127 Lex(); // Consume the integer token to be used as an identifier token. 1128 if (Lexer.getKind() != AsmToken::Colon) { 1129 if (!TheCondState.Ignore) 1130 return TokError("unexpected token at start of statement"); 1131 } 1132 } 1133 } else if (Lexer.is(AsmToken::Dot)) { 1134 // Treat '.' as a valid identifier in this context. 1135 Lex(); 1136 IDVal = "."; 1137 } else if (parseIdentifier(IDVal)) { 1138 if (!TheCondState.Ignore) 1139 return TokError("unexpected token at start of statement"); 1140 IDVal = ""; 1141 } 1142 1143 // Handle conditional assembly here before checking for skipping. We 1144 // have to do this so that .endif isn't skipped in a ".if 0" block for 1145 // example. 1146 StringMap<DirectiveKind>::const_iterator DirKindIt = 1147 DirectiveKindMap.find(IDVal); 1148 DirectiveKind DirKind = 1149 (DirKindIt == DirectiveKindMap.end()) ? DK_NO_DIRECTIVE : 1150 DirKindIt->getValue(); 1151 switch (DirKind) { 1152 default: 1153 break; 1154 case DK_IF: 1155 return ParseDirectiveIf(IDLoc); 1156 case DK_IFB: 1157 return ParseDirectiveIfb(IDLoc, true); 1158 case DK_IFNB: 1159 return ParseDirectiveIfb(IDLoc, false); 1160 case DK_IFC: 1161 return ParseDirectiveIfc(IDLoc, true); 1162 case DK_IFNC: 1163 return ParseDirectiveIfc(IDLoc, false); 1164 case DK_IFDEF: 1165 return ParseDirectiveIfdef(IDLoc, true); 1166 case DK_IFNDEF: 1167 case DK_IFNOTDEF: 1168 return ParseDirectiveIfdef(IDLoc, false); 1169 case DK_ELSEIF: 1170 return ParseDirectiveElseIf(IDLoc); 1171 case DK_ELSE: 1172 return ParseDirectiveElse(IDLoc); 1173 case DK_ENDIF: 1174 return ParseDirectiveEndIf(IDLoc); 1175 } 1176 1177 // Ignore the statement if in the middle of inactive conditional 1178 // (e.g. ".if 0"). 1179 if (TheCondState.Ignore) { 1180 eatToEndOfStatement(); 1181 return false; 1182 } 1183 1184 // FIXME: Recurse on local labels? 1185 1186 // See what kind of statement we have. 1187 switch (Lexer.getKind()) { 1188 case AsmToken::Colon: { 1189 checkForValidSection(); 1190 1191 // identifier ':' -> Label. 1192 Lex(); 1193 1194 // Diagnose attempt to use '.' as a label. 1195 if (IDVal == ".") 1196 return Error(IDLoc, "invalid use of pseudo-symbol '.' as a label"); 1197 1198 // Diagnose attempt to use a variable as a label. 1199 // 1200 // FIXME: Diagnostics. Note the location of the definition as a label. 1201 // FIXME: This doesn't diagnose assignment to a symbol which has been 1202 // implicitly marked as external. 1203 MCSymbol *Sym; 1204 if (LocalLabelVal == -1) 1205 Sym = getContext().GetOrCreateSymbol(IDVal); 1206 else 1207 Sym = Ctx.CreateDirectionalLocalSymbol(LocalLabelVal); 1208 if (!Sym->isUndefined() || Sym->isVariable()) 1209 return Error(IDLoc, "invalid symbol redefinition"); 1210 1211 // Emit the label. 1212 if (!ParsingInlineAsm) 1213 Out.EmitLabel(Sym); 1214 1215 // If we are generating dwarf for assembly source files then gather the 1216 // info to make a dwarf label entry for this label if needed. 1217 if (getContext().getGenDwarfForAssembly()) 1218 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(), 1219 IDLoc); 1220 1221 // Consume any end of statement token, if present, to avoid spurious 1222 // AddBlankLine calls(). 1223 if (Lexer.is(AsmToken::EndOfStatement)) { 1224 Lex(); 1225 if (Lexer.is(AsmToken::Eof)) 1226 return false; 1227 } 1228 1229 return false; 1230 } 1231 1232 case AsmToken::Equal: 1233 // identifier '=' ... -> assignment statement 1234 Lex(); 1235 1236 return ParseAssignment(IDVal, true); 1237 1238 default: // Normal instruction or directive. 1239 break; 1240 } 1241 1242 // If macros are enabled, check to see if this is a macro instantiation. 1243 if (MacrosEnabled()) 1244 if (const MCAsmMacro *M = LookupMacro(IDVal)) { 1245 return HandleMacroEntry(M, IDLoc); 1246 } 1247 1248 // Otherwise, we have a normal instruction or directive. 1249 1250 // Directives start with "." 1251 if (IDVal[0] == '.' && IDVal != ".") { 1252 // There are several entities interested in parsing directives: 1253 // 1254 // 1. The target-specific assembly parser. Some directives are target 1255 // specific or may potentially behave differently on certain targets. 1256 // 2. Asm parser extensions. For example, platform-specific parsers 1257 // (like the ELF parser) register themselves as extensions. 1258 // 3. The generic directive parser implemented by this class. These are 1259 // all the directives that behave in a target and platform independent 1260 // manner, or at least have a default behavior that's shared between 1261 // all targets and platforms. 1262 1263 // First query the target-specific parser. It will return 'true' if it 1264 // isn't interested in this directive. 1265 if (!getTargetParser().ParseDirective(ID)) 1266 return false; 1267 1268 // Next, check the extention directive map to see if any extension has 1269 // registered itself to parse this directive. 1270 std::pair<MCAsmParserExtension*, DirectiveHandler> Handler = 1271 ExtensionDirectiveMap.lookup(IDVal); 1272 if (Handler.first) 1273 return (*Handler.second)(Handler.first, IDVal, IDLoc); 1274 1275 // Finally, if no one else is interested in this directive, it must be 1276 // generic and familiar to this class. 1277 switch (DirKind) { 1278 default: 1279 break; 1280 case DK_SET: 1281 case DK_EQU: 1282 return ParseDirectiveSet(IDVal, true); 1283 case DK_EQUIV: 1284 return ParseDirectiveSet(IDVal, false); 1285 case DK_ASCII: 1286 return ParseDirectiveAscii(IDVal, false); 1287 case DK_ASCIZ: 1288 case DK_STRING: 1289 return ParseDirectiveAscii(IDVal, true); 1290 case DK_BYTE: 1291 return ParseDirectiveValue(1); 1292 case DK_SHORT: 1293 case DK_VALUE: 1294 case DK_2BYTE: 1295 return ParseDirectiveValue(2); 1296 case DK_LONG: 1297 case DK_INT: 1298 case DK_4BYTE: 1299 return ParseDirectiveValue(4); 1300 case DK_QUAD: 1301 case DK_8BYTE: 1302 return ParseDirectiveValue(8); 1303 case DK_SINGLE: 1304 case DK_FLOAT: 1305 return ParseDirectiveRealValue(APFloat::IEEEsingle); 1306 case DK_DOUBLE: 1307 return ParseDirectiveRealValue(APFloat::IEEEdouble); 1308 case DK_ALIGN: { 1309 bool IsPow2 = !getContext().getAsmInfo().getAlignmentIsInBytes(); 1310 return ParseDirectiveAlign(IsPow2, /*ExprSize=*/1); 1311 } 1312 case DK_ALIGN32: { 1313 bool IsPow2 = !getContext().getAsmInfo().getAlignmentIsInBytes(); 1314 return ParseDirectiveAlign(IsPow2, /*ExprSize=*/4); 1315 } 1316 case DK_BALIGN: 1317 return ParseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/1); 1318 case DK_BALIGNW: 1319 return ParseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/2); 1320 case DK_BALIGNL: 1321 return ParseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/4); 1322 case DK_P2ALIGN: 1323 return ParseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/1); 1324 case DK_P2ALIGNW: 1325 return ParseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/2); 1326 case DK_P2ALIGNL: 1327 return ParseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/4); 1328 case DK_ORG: 1329 return ParseDirectiveOrg(); 1330 case DK_FILL: 1331 return ParseDirectiveFill(); 1332 case DK_ZERO: 1333 return ParseDirectiveZero(); 1334 case DK_EXTERN: 1335 eatToEndOfStatement(); // .extern is the default, ignore it. 1336 return false; 1337 case DK_GLOBL: 1338 case DK_GLOBAL: 1339 return ParseDirectiveSymbolAttribute(MCSA_Global); 1340 case DK_INDIRECT_SYMBOL: 1341 return ParseDirectiveSymbolAttribute(MCSA_IndirectSymbol); 1342 case DK_LAZY_REFERENCE: 1343 return ParseDirectiveSymbolAttribute(MCSA_LazyReference); 1344 case DK_NO_DEAD_STRIP: 1345 return ParseDirectiveSymbolAttribute(MCSA_NoDeadStrip); 1346 case DK_SYMBOL_RESOLVER: 1347 return ParseDirectiveSymbolAttribute(MCSA_SymbolResolver); 1348 case DK_PRIVATE_EXTERN: 1349 return ParseDirectiveSymbolAttribute(MCSA_PrivateExtern); 1350 case DK_REFERENCE: 1351 return ParseDirectiveSymbolAttribute(MCSA_Reference); 1352 case DK_WEAK_DEFINITION: 1353 return ParseDirectiveSymbolAttribute(MCSA_WeakDefinition); 1354 case DK_WEAK_REFERENCE: 1355 return ParseDirectiveSymbolAttribute(MCSA_WeakReference); 1356 case DK_WEAK_DEF_CAN_BE_HIDDEN: 1357 return ParseDirectiveSymbolAttribute(MCSA_WeakDefAutoPrivate); 1358 case DK_COMM: 1359 case DK_COMMON: 1360 return ParseDirectiveComm(/*IsLocal=*/false); 1361 case DK_LCOMM: 1362 return ParseDirectiveComm(/*IsLocal=*/true); 1363 case DK_ABORT: 1364 return ParseDirectiveAbort(); 1365 case DK_INCLUDE: 1366 return ParseDirectiveInclude(); 1367 case DK_INCBIN: 1368 return ParseDirectiveIncbin(); 1369 case DK_CODE16: 1370 case DK_CODE16GCC: 1371 return TokError(Twine(IDVal) + " not supported yet"); 1372 case DK_REPT: 1373 return ParseDirectiveRept(IDLoc); 1374 case DK_IRP: 1375 return ParseDirectiveIrp(IDLoc); 1376 case DK_IRPC: 1377 return ParseDirectiveIrpc(IDLoc); 1378 case DK_ENDR: 1379 return ParseDirectiveEndr(IDLoc); 1380 case DK_BUNDLE_ALIGN_MODE: 1381 return ParseDirectiveBundleAlignMode(); 1382 case DK_BUNDLE_LOCK: 1383 return ParseDirectiveBundleLock(); 1384 case DK_BUNDLE_UNLOCK: 1385 return ParseDirectiveBundleUnlock(); 1386 case DK_SLEB128: 1387 return ParseDirectiveLEB128(true); 1388 case DK_ULEB128: 1389 return ParseDirectiveLEB128(false); 1390 case DK_SPACE: 1391 case DK_SKIP: 1392 return ParseDirectiveSpace(IDVal); 1393 case DK_FILE: 1394 return ParseDirectiveFile(IDLoc); 1395 case DK_LINE: 1396 return ParseDirectiveLine(); 1397 case DK_LOC: 1398 return ParseDirectiveLoc(); 1399 case DK_STABS: 1400 return ParseDirectiveStabs(); 1401 case DK_CFI_SECTIONS: 1402 return ParseDirectiveCFISections(); 1403 case DK_CFI_STARTPROC: 1404 return ParseDirectiveCFIStartProc(); 1405 case DK_CFI_ENDPROC: 1406 return ParseDirectiveCFIEndProc(); 1407 case DK_CFI_DEF_CFA: 1408 return ParseDirectiveCFIDefCfa(IDLoc); 1409 case DK_CFI_DEF_CFA_OFFSET: 1410 return ParseDirectiveCFIDefCfaOffset(); 1411 case DK_CFI_ADJUST_CFA_OFFSET: 1412 return ParseDirectiveCFIAdjustCfaOffset(); 1413 case DK_CFI_DEF_CFA_REGISTER: 1414 return ParseDirectiveCFIDefCfaRegister(IDLoc); 1415 case DK_CFI_OFFSET: 1416 return ParseDirectiveCFIOffset(IDLoc); 1417 case DK_CFI_REL_OFFSET: 1418 return ParseDirectiveCFIRelOffset(IDLoc); 1419 case DK_CFI_PERSONALITY: 1420 return ParseDirectiveCFIPersonalityOrLsda(true); 1421 case DK_CFI_LSDA: 1422 return ParseDirectiveCFIPersonalityOrLsda(false); 1423 case DK_CFI_REMEMBER_STATE: 1424 return ParseDirectiveCFIRememberState(); 1425 case DK_CFI_RESTORE_STATE: 1426 return ParseDirectiveCFIRestoreState(); 1427 case DK_CFI_SAME_VALUE: 1428 return ParseDirectiveCFISameValue(IDLoc); 1429 case DK_CFI_RESTORE: 1430 return ParseDirectiveCFIRestore(IDLoc); 1431 case DK_CFI_ESCAPE: 1432 return ParseDirectiveCFIEscape(); 1433 case DK_CFI_SIGNAL_FRAME: 1434 return ParseDirectiveCFISignalFrame(); 1435 case DK_CFI_UNDEFINED: 1436 return ParseDirectiveCFIUndefined(IDLoc); 1437 case DK_CFI_REGISTER: 1438 return ParseDirectiveCFIRegister(IDLoc); 1439 case DK_MACROS_ON: 1440 case DK_MACROS_OFF: 1441 return ParseDirectiveMacrosOnOff(IDVal); 1442 case DK_MACRO: 1443 return ParseDirectiveMacro(IDLoc); 1444 case DK_ENDM: 1445 case DK_ENDMACRO: 1446 return ParseDirectiveEndMacro(IDVal); 1447 case DK_PURGEM: 1448 return ParseDirectivePurgeMacro(IDLoc); 1449 } 1450 1451 return Error(IDLoc, "unknown directive"); 1452 } 1453 1454 // __asm _emit or __asm __emit 1455 if (ParsingInlineAsm && (IDVal == "_emit" || IDVal == "__emit" || 1456 IDVal == "_EMIT" || IDVal == "__EMIT")) 1457 return ParseDirectiveMSEmit(IDLoc, Info, IDVal.size()); 1458 1459 // __asm align 1460 if (ParsingInlineAsm && (IDVal == "align" || IDVal == "ALIGN")) 1461 return ParseDirectiveMSAlign(IDLoc, Info); 1462 1463 checkForValidSection(); 1464 1465 // Canonicalize the opcode to lower case. 1466 std::string OpcodeStr = IDVal.lower(); 1467 ParseInstructionInfo IInfo(Info.AsmRewrites); 1468 bool HadError = getTargetParser().ParseInstruction(IInfo, OpcodeStr, 1469 IDLoc, Info.ParsedOperands); 1470 Info.ParseError = HadError; 1471 1472 // Dump the parsed representation, if requested. 1473 if (getShowParsedOperands()) { 1474 SmallString<256> Str; 1475 raw_svector_ostream OS(Str); 1476 OS << "parsed instruction: ["; 1477 for (unsigned i = 0; i != Info.ParsedOperands.size(); ++i) { 1478 if (i != 0) 1479 OS << ", "; 1480 Info.ParsedOperands[i]->print(OS); 1481 } 1482 OS << "]"; 1483 1484 PrintMessage(IDLoc, SourceMgr::DK_Note, OS.str()); 1485 } 1486 1487 // If we are generating dwarf for assembly source files and the current 1488 // section is the initial text section then generate a .loc directive for 1489 // the instruction. 1490 if (!HadError && getContext().getGenDwarfForAssembly() && 1491 getContext().getGenDwarfSection() == getStreamer().getCurrentSection()) { 1492 1493 unsigned Line = SrcMgr.FindLineNumber(IDLoc, CurBuffer); 1494 1495 // If we previously parsed a cpp hash file line comment then make sure the 1496 // current Dwarf File is for the CppHashFilename if not then emit the 1497 // Dwarf File table for it and adjust the line number for the .loc. 1498 const SmallVectorImpl<MCDwarfFile *> &MCDwarfFiles = 1499 getContext().getMCDwarfFiles(); 1500 if (CppHashFilename.size() != 0) { 1501 if (MCDwarfFiles[getContext().getGenDwarfFileNumber()]->getName() != 1502 CppHashFilename) 1503 getStreamer().EmitDwarfFileDirective( 1504 getContext().nextGenDwarfFileNumber(), StringRef(), CppHashFilename); 1505 1506 unsigned CppHashLocLineNo = SrcMgr.FindLineNumber(CppHashLoc,CppHashBuf); 1507 Line = CppHashLineNumber - 1 + (Line - CppHashLocLineNo); 1508 } 1509 1510 getStreamer().EmitDwarfLocDirective(getContext().getGenDwarfFileNumber(), 1511 Line, 0, DWARF2_LINE_DEFAULT_IS_STMT ? 1512 DWARF2_FLAG_IS_STMT : 0, 0, 0, 1513 StringRef()); 1514 } 1515 1516 // If parsing succeeded, match the instruction. 1517 if (!HadError) { 1518 unsigned ErrorInfo; 1519 HadError = getTargetParser().MatchAndEmitInstruction(IDLoc, Info.Opcode, 1520 Info.ParsedOperands, 1521 Out, ErrorInfo, 1522 ParsingInlineAsm); 1523 } 1524 1525 // Don't skip the rest of the line, the instruction parser is responsible for 1526 // that. 1527 return false; 1528 } 1529 1530 /// EatToEndOfLine uses the Lexer to eat the characters to the end of the line 1531 /// since they may not be able to be tokenized to get to the end of line token. 1532 void AsmParser::EatToEndOfLine() { 1533 if (!Lexer.is(AsmToken::EndOfStatement)) 1534 Lexer.LexUntilEndOfLine(); 1535 // Eat EOL. 1536 Lex(); 1537 } 1538 1539 /// ParseCppHashLineFilenameComment as this: 1540 /// ::= # number "filename" 1541 /// or just as a full line comment if it doesn't have a number and a string. 1542 bool AsmParser::ParseCppHashLineFilenameComment(const SMLoc &L) { 1543 Lex(); // Eat the hash token. 1544 1545 if (getLexer().isNot(AsmToken::Integer)) { 1546 // Consume the line since in cases it is not a well-formed line directive, 1547 // as if were simply a full line comment. 1548 EatToEndOfLine(); 1549 return false; 1550 } 1551 1552 int64_t LineNumber = getTok().getIntVal(); 1553 Lex(); 1554 1555 if (getLexer().isNot(AsmToken::String)) { 1556 EatToEndOfLine(); 1557 return false; 1558 } 1559 1560 StringRef Filename = getTok().getString(); 1561 // Get rid of the enclosing quotes. 1562 Filename = Filename.substr(1, Filename.size()-2); 1563 1564 // Save the SMLoc, Filename and LineNumber for later use by diagnostics. 1565 CppHashLoc = L; 1566 CppHashFilename = Filename; 1567 CppHashLineNumber = LineNumber; 1568 CppHashBuf = CurBuffer; 1569 1570 // Ignore any trailing characters, they're just comment. 1571 EatToEndOfLine(); 1572 return false; 1573 } 1574 1575 /// DiagHandler - will use the last parsed cpp hash line filename comment 1576 /// for the Filename and LineNo if any in the diagnostic. 1577 void AsmParser::DiagHandler(const SMDiagnostic &Diag, void *Context) { 1578 const AsmParser *Parser = static_cast<const AsmParser*>(Context); 1579 raw_ostream &OS = errs(); 1580 1581 const SourceMgr &DiagSrcMgr = *Diag.getSourceMgr(); 1582 const SMLoc &DiagLoc = Diag.getLoc(); 1583 int DiagBuf = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 1584 int CppHashBuf = Parser->SrcMgr.FindBufferContainingLoc(Parser->CppHashLoc); 1585 1586 // Like SourceMgr::PrintMessage() we need to print the include stack if any 1587 // before printing the message. 1588 int DiagCurBuffer = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 1589 if (!Parser->SavedDiagHandler && DiagCurBuffer > 0) { 1590 SMLoc ParentIncludeLoc = DiagSrcMgr.getParentIncludeLoc(DiagCurBuffer); 1591 DiagSrcMgr.PrintIncludeStack(ParentIncludeLoc, OS); 1592 } 1593 1594 // If we have not parsed a cpp hash line filename comment or the source 1595 // manager changed or buffer changed (like in a nested include) then just 1596 // print the normal diagnostic using its Filename and LineNo. 1597 if (!Parser->CppHashLineNumber || 1598 &DiagSrcMgr != &Parser->SrcMgr || 1599 DiagBuf != CppHashBuf) { 1600 if (Parser->SavedDiagHandler) 1601 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext); 1602 else 1603 Diag.print(0, OS); 1604 return; 1605 } 1606 1607 // Use the CppHashFilename and calculate a line number based on the 1608 // CppHashLoc and CppHashLineNumber relative to this Diag's SMLoc for 1609 // the diagnostic. 1610 const std::string Filename = Parser->CppHashFilename; 1611 1612 int DiagLocLineNo = DiagSrcMgr.FindLineNumber(DiagLoc, DiagBuf); 1613 int CppHashLocLineNo = 1614 Parser->SrcMgr.FindLineNumber(Parser->CppHashLoc, CppHashBuf); 1615 int LineNo = Parser->CppHashLineNumber - 1 + 1616 (DiagLocLineNo - CppHashLocLineNo); 1617 1618 SMDiagnostic NewDiag(*Diag.getSourceMgr(), Diag.getLoc(), 1619 Filename, LineNo, Diag.getColumnNo(), 1620 Diag.getKind(), Diag.getMessage(), 1621 Diag.getLineContents(), Diag.getRanges()); 1622 1623 if (Parser->SavedDiagHandler) 1624 Parser->SavedDiagHandler(NewDiag, Parser->SavedDiagContext); 1625 else 1626 NewDiag.print(0, OS); 1627 } 1628 1629 // FIXME: This is mostly duplicated from the function in AsmLexer.cpp. The 1630 // difference being that that function accepts '@' as part of identifiers and 1631 // we can't do that. AsmLexer.cpp should probably be changed to handle 1632 // '@' as a special case when needed. 1633 static bool isIdentifierChar(char c) { 1634 return isalnum(static_cast<unsigned char>(c)) || c == '_' || c == '$' || 1635 c == '.'; 1636 } 1637 1638 bool AsmParser::expandMacro(raw_svector_ostream &OS, StringRef Body, 1639 const MCAsmMacroParameters &Parameters, 1640 const MCAsmMacroArguments &A, 1641 const SMLoc &L) { 1642 unsigned NParameters = Parameters.size(); 1643 if (NParameters != 0 && NParameters != A.size()) 1644 return Error(L, "Wrong number of arguments"); 1645 1646 // A macro without parameters is handled differently on Darwin: 1647 // gas accepts no arguments and does no substitutions 1648 while (!Body.empty()) { 1649 // Scan for the next substitution. 1650 std::size_t End = Body.size(), Pos = 0; 1651 for (; Pos != End; ++Pos) { 1652 // Check for a substitution or escape. 1653 if (!NParameters) { 1654 // This macro has no parameters, look for $0, $1, etc. 1655 if (Body[Pos] != '$' || Pos + 1 == End) 1656 continue; 1657 1658 char Next = Body[Pos + 1]; 1659 if (Next == '$' || Next == 'n' || 1660 isdigit(static_cast<unsigned char>(Next))) 1661 break; 1662 } else { 1663 // This macro has parameters, look for \foo, \bar, etc. 1664 if (Body[Pos] == '\\' && Pos + 1 != End) 1665 break; 1666 } 1667 } 1668 1669 // Add the prefix. 1670 OS << Body.slice(0, Pos); 1671 1672 // Check if we reached the end. 1673 if (Pos == End) 1674 break; 1675 1676 if (!NParameters) { 1677 switch (Body[Pos+1]) { 1678 // $$ => $ 1679 case '$': 1680 OS << '$'; 1681 break; 1682 1683 // $n => number of arguments 1684 case 'n': 1685 OS << A.size(); 1686 break; 1687 1688 // $[0-9] => argument 1689 default: { 1690 // Missing arguments are ignored. 1691 unsigned Index = Body[Pos+1] - '0'; 1692 if (Index >= A.size()) 1693 break; 1694 1695 // Otherwise substitute with the token values, with spaces eliminated. 1696 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(), 1697 ie = A[Index].end(); it != ie; ++it) 1698 OS << it->getString(); 1699 break; 1700 } 1701 } 1702 Pos += 2; 1703 } else { 1704 unsigned I = Pos + 1; 1705 while (isIdentifierChar(Body[I]) && I + 1 != End) 1706 ++I; 1707 1708 const char *Begin = Body.data() + Pos +1; 1709 StringRef Argument(Begin, I - (Pos +1)); 1710 unsigned Index = 0; 1711 for (; Index < NParameters; ++Index) 1712 if (Parameters[Index].first == Argument) 1713 break; 1714 1715 if (Index == NParameters) { 1716 if (Body[Pos+1] == '(' && Body[Pos+2] == ')') 1717 Pos += 3; 1718 else { 1719 OS << '\\' << Argument; 1720 Pos = I; 1721 } 1722 } else { 1723 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(), 1724 ie = A[Index].end(); it != ie; ++it) 1725 if (it->getKind() == AsmToken::String) 1726 OS << it->getStringContents(); 1727 else 1728 OS << it->getString(); 1729 1730 Pos += 1 + Argument.size(); 1731 } 1732 } 1733 // Update the scan point. 1734 Body = Body.substr(Pos); 1735 } 1736 1737 return false; 1738 } 1739 1740 MacroInstantiation::MacroInstantiation(const MCAsmMacro *M, SMLoc IL, 1741 int EB, SMLoc EL, 1742 MemoryBuffer *I) 1743 : TheMacro(M), Instantiation(I), InstantiationLoc(IL), ExitBuffer(EB), 1744 ExitLoc(EL) 1745 { 1746 } 1747 1748 static bool IsOperator(AsmToken::TokenKind kind) 1749 { 1750 switch (kind) 1751 { 1752 default: 1753 return false; 1754 case AsmToken::Plus: 1755 case AsmToken::Minus: 1756 case AsmToken::Tilde: 1757 case AsmToken::Slash: 1758 case AsmToken::Star: 1759 case AsmToken::Dot: 1760 case AsmToken::Equal: 1761 case AsmToken::EqualEqual: 1762 case AsmToken::Pipe: 1763 case AsmToken::PipePipe: 1764 case AsmToken::Caret: 1765 case AsmToken::Amp: 1766 case AsmToken::AmpAmp: 1767 case AsmToken::Exclaim: 1768 case AsmToken::ExclaimEqual: 1769 case AsmToken::Percent: 1770 case AsmToken::Less: 1771 case AsmToken::LessEqual: 1772 case AsmToken::LessLess: 1773 case AsmToken::LessGreater: 1774 case AsmToken::Greater: 1775 case AsmToken::GreaterEqual: 1776 case AsmToken::GreaterGreater: 1777 return true; 1778 } 1779 } 1780 1781 bool AsmParser::ParseMacroArgument(MCAsmMacroArgument &MA, 1782 AsmToken::TokenKind &ArgumentDelimiter) { 1783 unsigned ParenLevel = 0; 1784 unsigned AddTokens = 0; 1785 1786 // gas accepts arguments separated by whitespace, except on Darwin 1787 if (!IsDarwin) 1788 Lexer.setSkipSpace(false); 1789 1790 for (;;) { 1791 if (Lexer.is(AsmToken::Eof) || Lexer.is(AsmToken::Equal)) { 1792 Lexer.setSkipSpace(true); 1793 return TokError("unexpected token in macro instantiation"); 1794 } 1795 1796 if (ParenLevel == 0 && Lexer.is(AsmToken::Comma)) { 1797 // Spaces and commas cannot be mixed to delimit parameters 1798 if (ArgumentDelimiter == AsmToken::Eof) 1799 ArgumentDelimiter = AsmToken::Comma; 1800 else if (ArgumentDelimiter != AsmToken::Comma) { 1801 Lexer.setSkipSpace(true); 1802 return TokError("expected ' ' for macro argument separator"); 1803 } 1804 break; 1805 } 1806 1807 if (Lexer.is(AsmToken::Space)) { 1808 Lex(); // Eat spaces 1809 1810 // Spaces can delimit parameters, but could also be part an expression. 1811 // If the token after a space is an operator, add the token and the next 1812 // one into this argument 1813 if (ArgumentDelimiter == AsmToken::Space || 1814 ArgumentDelimiter == AsmToken::Eof) { 1815 if (IsOperator(Lexer.getKind())) { 1816 // Check to see whether the token is used as an operator, 1817 // or part of an identifier 1818 const char *NextChar = getTok().getEndLoc().getPointer(); 1819 if (*NextChar == ' ') 1820 AddTokens = 2; 1821 } 1822 1823 if (!AddTokens && ParenLevel == 0) { 1824 if (ArgumentDelimiter == AsmToken::Eof && 1825 !IsOperator(Lexer.getKind())) 1826 ArgumentDelimiter = AsmToken::Space; 1827 break; 1828 } 1829 } 1830 } 1831 1832 // HandleMacroEntry relies on not advancing the lexer here 1833 // to be able to fill in the remaining default parameter values 1834 if (Lexer.is(AsmToken::EndOfStatement)) 1835 break; 1836 1837 // Adjust the current parentheses level. 1838 if (Lexer.is(AsmToken::LParen)) 1839 ++ParenLevel; 1840 else if (Lexer.is(AsmToken::RParen) && ParenLevel) 1841 --ParenLevel; 1842 1843 // Append the token to the current argument list. 1844 MA.push_back(getTok()); 1845 if (AddTokens) 1846 AddTokens--; 1847 Lex(); 1848 } 1849 1850 Lexer.setSkipSpace(true); 1851 if (ParenLevel != 0) 1852 return TokError("unbalanced parentheses in macro argument"); 1853 return false; 1854 } 1855 1856 // Parse the macro instantiation arguments. 1857 bool AsmParser::ParseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A) { 1858 const unsigned NParameters = M ? M->Parameters.size() : 0; 1859 // Argument delimiter is initially unknown. It will be set by 1860 // ParseMacroArgument() 1861 AsmToken::TokenKind ArgumentDelimiter = AsmToken::Eof; 1862 1863 // Parse two kinds of macro invocations: 1864 // - macros defined without any parameters accept an arbitrary number of them 1865 // - macros defined with parameters accept at most that many of them 1866 for (unsigned Parameter = 0; !NParameters || Parameter < NParameters; 1867 ++Parameter) { 1868 MCAsmMacroArgument MA; 1869 1870 if (ParseMacroArgument(MA, ArgumentDelimiter)) 1871 return true; 1872 1873 if (!MA.empty() || !NParameters) 1874 A.push_back(MA); 1875 else if (NParameters) { 1876 if (!M->Parameters[Parameter].second.empty()) 1877 A.push_back(M->Parameters[Parameter].second); 1878 } 1879 1880 // At the end of the statement, fill in remaining arguments that have 1881 // default values. If there aren't any, then the next argument is 1882 // required but missing 1883 if (Lexer.is(AsmToken::EndOfStatement)) { 1884 if (NParameters && Parameter < NParameters - 1) { 1885 if (M->Parameters[Parameter + 1].second.empty()) 1886 return TokError("macro argument '" + 1887 Twine(M->Parameters[Parameter + 1].first) + 1888 "' is missing"); 1889 else 1890 continue; 1891 } 1892 return false; 1893 } 1894 1895 if (Lexer.is(AsmToken::Comma)) 1896 Lex(); 1897 } 1898 return TokError("Too many arguments"); 1899 } 1900 1901 const MCAsmMacro* AsmParser::LookupMacro(StringRef Name) { 1902 StringMap<MCAsmMacro*>::iterator I = MacroMap.find(Name); 1903 return (I == MacroMap.end()) ? NULL : I->getValue(); 1904 } 1905 1906 void AsmParser::DefineMacro(StringRef Name, const MCAsmMacro& Macro) { 1907 MacroMap[Name] = new MCAsmMacro(Macro); 1908 } 1909 1910 void AsmParser::UndefineMacro(StringRef Name) { 1911 StringMap<MCAsmMacro*>::iterator I = MacroMap.find(Name); 1912 if (I != MacroMap.end()) { 1913 delete I->getValue(); 1914 MacroMap.erase(I); 1915 } 1916 } 1917 1918 bool AsmParser::HandleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc) { 1919 // Arbitrarily limit macro nesting depth, to match 'as'. We can eliminate 1920 // this, although we should protect against infinite loops. 1921 if (ActiveMacros.size() == 20) 1922 return TokError("macros cannot be nested more than 20 levels deep"); 1923 1924 MCAsmMacroArguments A; 1925 if (ParseMacroArguments(M, A)) 1926 return true; 1927 1928 // Remove any trailing empty arguments. Do this after-the-fact as we have 1929 // to keep empty arguments in the middle of the list or positionality 1930 // gets off. e.g., "foo 1, , 2" vs. "foo 1, 2," 1931 while (!A.empty() && A.back().empty()) 1932 A.pop_back(); 1933 1934 // Macro instantiation is lexical, unfortunately. We construct a new buffer 1935 // to hold the macro body with substitutions. 1936 SmallString<256> Buf; 1937 StringRef Body = M->Body; 1938 raw_svector_ostream OS(Buf); 1939 1940 if (expandMacro(OS, Body, M->Parameters, A, getTok().getLoc())) 1941 return true; 1942 1943 // We include the .endmacro in the buffer as our cue to exit the macro 1944 // instantiation. 1945 OS << ".endmacro\n"; 1946 1947 MemoryBuffer *Instantiation = 1948 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 1949 1950 // Create the macro instantiation object and add to the current macro 1951 // instantiation stack. 1952 MacroInstantiation *MI = new MacroInstantiation(M, NameLoc, 1953 CurBuffer, 1954 getTok().getLoc(), 1955 Instantiation); 1956 ActiveMacros.push_back(MI); 1957 1958 // Jump to the macro instantiation and prime the lexer. 1959 CurBuffer = SrcMgr.AddNewSourceBuffer(MI->Instantiation, SMLoc()); 1960 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)); 1961 Lex(); 1962 1963 return false; 1964 } 1965 1966 void AsmParser::HandleMacroExit() { 1967 // Jump to the EndOfStatement we should return to, and consume it. 1968 JumpToLoc(ActiveMacros.back()->ExitLoc, ActiveMacros.back()->ExitBuffer); 1969 Lex(); 1970 1971 // Pop the instantiation entry. 1972 delete ActiveMacros.back(); 1973 ActiveMacros.pop_back(); 1974 } 1975 1976 static bool IsUsedIn(const MCSymbol *Sym, const MCExpr *Value) { 1977 switch (Value->getKind()) { 1978 case MCExpr::Binary: { 1979 const MCBinaryExpr *BE = static_cast<const MCBinaryExpr*>(Value); 1980 return IsUsedIn(Sym, BE->getLHS()) || IsUsedIn(Sym, BE->getRHS()); 1981 break; 1982 } 1983 case MCExpr::Target: 1984 case MCExpr::Constant: 1985 return false; 1986 case MCExpr::SymbolRef: { 1987 const MCSymbol &S = static_cast<const MCSymbolRefExpr*>(Value)->getSymbol(); 1988 if (S.isVariable()) 1989 return IsUsedIn(Sym, S.getVariableValue()); 1990 return &S == Sym; 1991 } 1992 case MCExpr::Unary: 1993 return IsUsedIn(Sym, static_cast<const MCUnaryExpr*>(Value)->getSubExpr()); 1994 } 1995 1996 llvm_unreachable("Unknown expr kind!"); 1997 } 1998 1999 bool AsmParser::ParseAssignment(StringRef Name, bool allow_redef, 2000 bool NoDeadStrip) { 2001 // FIXME: Use better location, we should use proper tokens. 2002 SMLoc EqualLoc = Lexer.getLoc(); 2003 2004 const MCExpr *Value; 2005 if (parseExpression(Value)) 2006 return true; 2007 2008 // Note: we don't count b as used in "a = b". This is to allow 2009 // a = b 2010 // b = c 2011 2012 if (Lexer.isNot(AsmToken::EndOfStatement)) 2013 return TokError("unexpected token in assignment"); 2014 2015 // Error on assignment to '.'. 2016 if (Name == ".") { 2017 return Error(EqualLoc, ("assignment to pseudo-symbol '.' is unsupported " 2018 "(use '.space' or '.org').)")); 2019 } 2020 2021 // Eat the end of statement marker. 2022 Lex(); 2023 2024 // Validate that the LHS is allowed to be a variable (either it has not been 2025 // used as a symbol, or it is an absolute symbol). 2026 MCSymbol *Sym = getContext().LookupSymbol(Name); 2027 if (Sym) { 2028 // Diagnose assignment to a label. 2029 // 2030 // FIXME: Diagnostics. Note the location of the definition as a label. 2031 // FIXME: Diagnose assignment to protected identifier (e.g., register name). 2032 if (IsUsedIn(Sym, Value)) 2033 return Error(EqualLoc, "Recursive use of '" + Name + "'"); 2034 else if (Sym->isUndefined() && !Sym->isUsed() && !Sym->isVariable()) 2035 ; // Allow redefinitions of undefined symbols only used in directives. 2036 else if (Sym->isVariable() && !Sym->isUsed() && allow_redef) 2037 ; // Allow redefinitions of variables that haven't yet been used. 2038 else if (!Sym->isUndefined() && (!Sym->isVariable() || !allow_redef)) 2039 return Error(EqualLoc, "redefinition of '" + Name + "'"); 2040 else if (!Sym->isVariable()) 2041 return Error(EqualLoc, "invalid assignment to '" + Name + "'"); 2042 else if (!isa<MCConstantExpr>(Sym->getVariableValue())) 2043 return Error(EqualLoc, "invalid reassignment of non-absolute variable '" + 2044 Name + "'"); 2045 2046 // Don't count these checks as uses. 2047 Sym->setUsed(false); 2048 } else 2049 Sym = getContext().GetOrCreateSymbol(Name); 2050 2051 // FIXME: Handle '.'. 2052 2053 // Do the assignment. 2054 Out.EmitAssignment(Sym, Value); 2055 if (NoDeadStrip) 2056 Out.EmitSymbolAttribute(Sym, MCSA_NoDeadStrip); 2057 2058 2059 return false; 2060 } 2061 2062 /// parseIdentifier: 2063 /// ::= identifier 2064 /// ::= string 2065 bool AsmParser::parseIdentifier(StringRef &Res) { 2066 // The assembler has relaxed rules for accepting identifiers, in particular we 2067 // allow things like '.globl $foo', which would normally be separate 2068 // tokens. At this level, we have already lexed so we cannot (currently) 2069 // handle this as a context dependent token, instead we detect adjacent tokens 2070 // and return the combined identifier. 2071 if (Lexer.is(AsmToken::Dollar)) { 2072 SMLoc DollarLoc = getLexer().getLoc(); 2073 2074 // Consume the dollar sign, and check for a following identifier. 2075 Lex(); 2076 if (Lexer.isNot(AsmToken::Identifier)) 2077 return true; 2078 2079 // We have a '$' followed by an identifier, make sure they are adjacent. 2080 if (DollarLoc.getPointer() + 1 != getTok().getLoc().getPointer()) 2081 return true; 2082 2083 // Construct the joined identifier and consume the token. 2084 Res = StringRef(DollarLoc.getPointer(), 2085 getTok().getIdentifier().size() + 1); 2086 Lex(); 2087 return false; 2088 } 2089 2090 if (Lexer.isNot(AsmToken::Identifier) && 2091 Lexer.isNot(AsmToken::String)) 2092 return true; 2093 2094 Res = getTok().getIdentifier(); 2095 2096 Lex(); // Consume the identifier token. 2097 2098 return false; 2099 } 2100 2101 /// ParseDirectiveSet: 2102 /// ::= .equ identifier ',' expression 2103 /// ::= .equiv identifier ',' expression 2104 /// ::= .set identifier ',' expression 2105 bool AsmParser::ParseDirectiveSet(StringRef IDVal, bool allow_redef) { 2106 StringRef Name; 2107 2108 if (parseIdentifier(Name)) 2109 return TokError("expected identifier after '" + Twine(IDVal) + "'"); 2110 2111 if (getLexer().isNot(AsmToken::Comma)) 2112 return TokError("unexpected token in '" + Twine(IDVal) + "'"); 2113 Lex(); 2114 2115 return ParseAssignment(Name, allow_redef, true); 2116 } 2117 2118 bool AsmParser::parseEscapedString(std::string &Data) { 2119 assert(getLexer().is(AsmToken::String) && "Unexpected current token!"); 2120 2121 Data = ""; 2122 StringRef Str = getTok().getStringContents(); 2123 for (unsigned i = 0, e = Str.size(); i != e; ++i) { 2124 if (Str[i] != '\\') { 2125 Data += Str[i]; 2126 continue; 2127 } 2128 2129 // Recognize escaped characters. Note that this escape semantics currently 2130 // loosely follows Darwin 'as'. Notably, it doesn't support hex escapes. 2131 ++i; 2132 if (i == e) 2133 return TokError("unexpected backslash at end of string"); 2134 2135 // Recognize octal sequences. 2136 if ((unsigned) (Str[i] - '0') <= 7) { 2137 // Consume up to three octal characters. 2138 unsigned Value = Str[i] - '0'; 2139 2140 if (i + 1 != e && ((unsigned) (Str[i + 1] - '0')) <= 7) { 2141 ++i; 2142 Value = Value * 8 + (Str[i] - '0'); 2143 2144 if (i + 1 != e && ((unsigned) (Str[i + 1] - '0')) <= 7) { 2145 ++i; 2146 Value = Value * 8 + (Str[i] - '0'); 2147 } 2148 } 2149 2150 if (Value > 255) 2151 return TokError("invalid octal escape sequence (out of range)"); 2152 2153 Data += (unsigned char) Value; 2154 continue; 2155 } 2156 2157 // Otherwise recognize individual escapes. 2158 switch (Str[i]) { 2159 default: 2160 // Just reject invalid escape sequences for now. 2161 return TokError("invalid escape sequence (unrecognized character)"); 2162 2163 case 'b': Data += '\b'; break; 2164 case 'f': Data += '\f'; break; 2165 case 'n': Data += '\n'; break; 2166 case 'r': Data += '\r'; break; 2167 case 't': Data += '\t'; break; 2168 case '"': Data += '"'; break; 2169 case '\\': Data += '\\'; break; 2170 } 2171 } 2172 2173 return false; 2174 } 2175 2176 /// ParseDirectiveAscii: 2177 /// ::= ( .ascii | .asciz | .string ) [ "string" ( , "string" )* ] 2178 bool AsmParser::ParseDirectiveAscii(StringRef IDVal, bool ZeroTerminated) { 2179 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2180 checkForValidSection(); 2181 2182 for (;;) { 2183 if (getLexer().isNot(AsmToken::String)) 2184 return TokError("expected string in '" + Twine(IDVal) + "' directive"); 2185 2186 std::string Data; 2187 if (parseEscapedString(Data)) 2188 return true; 2189 2190 getStreamer().EmitBytes(Data, DEFAULT_ADDRSPACE); 2191 if (ZeroTerminated) 2192 getStreamer().EmitBytes(StringRef("\0", 1), DEFAULT_ADDRSPACE); 2193 2194 Lex(); 2195 2196 if (getLexer().is(AsmToken::EndOfStatement)) 2197 break; 2198 2199 if (getLexer().isNot(AsmToken::Comma)) 2200 return TokError("unexpected token in '" + Twine(IDVal) + "' directive"); 2201 Lex(); 2202 } 2203 } 2204 2205 Lex(); 2206 return false; 2207 } 2208 2209 /// ParseDirectiveValue 2210 /// ::= (.byte | .short | ... ) [ expression (, expression)* ] 2211 bool AsmParser::ParseDirectiveValue(unsigned Size) { 2212 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2213 checkForValidSection(); 2214 2215 for (;;) { 2216 const MCExpr *Value; 2217 SMLoc ExprLoc = getLexer().getLoc(); 2218 if (parseExpression(Value)) 2219 return true; 2220 2221 // Special case constant expressions to match code generator. 2222 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2223 assert(Size <= 8 && "Invalid size"); 2224 uint64_t IntValue = MCE->getValue(); 2225 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue)) 2226 return Error(ExprLoc, "literal value out of range for directive"); 2227 getStreamer().EmitIntValue(IntValue, Size, DEFAULT_ADDRSPACE); 2228 } else 2229 getStreamer().EmitValue(Value, Size, DEFAULT_ADDRSPACE); 2230 2231 if (getLexer().is(AsmToken::EndOfStatement)) 2232 break; 2233 2234 // FIXME: Improve diagnostic. 2235 if (getLexer().isNot(AsmToken::Comma)) 2236 return TokError("unexpected token in directive"); 2237 Lex(); 2238 } 2239 } 2240 2241 Lex(); 2242 return false; 2243 } 2244 2245 /// ParseDirectiveRealValue 2246 /// ::= (.single | .double) [ expression (, expression)* ] 2247 bool AsmParser::ParseDirectiveRealValue(const fltSemantics &Semantics) { 2248 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2249 checkForValidSection(); 2250 2251 for (;;) { 2252 // We don't truly support arithmetic on floating point expressions, so we 2253 // have to manually parse unary prefixes. 2254 bool IsNeg = false; 2255 if (getLexer().is(AsmToken::Minus)) { 2256 Lex(); 2257 IsNeg = true; 2258 } else if (getLexer().is(AsmToken::Plus)) 2259 Lex(); 2260 2261 if (getLexer().isNot(AsmToken::Integer) && 2262 getLexer().isNot(AsmToken::Real) && 2263 getLexer().isNot(AsmToken::Identifier)) 2264 return TokError("unexpected token in directive"); 2265 2266 // Convert to an APFloat. 2267 APFloat Value(Semantics); 2268 StringRef IDVal = getTok().getString(); 2269 if (getLexer().is(AsmToken::Identifier)) { 2270 if (!IDVal.compare_lower("infinity") || !IDVal.compare_lower("inf")) 2271 Value = APFloat::getInf(Semantics); 2272 else if (!IDVal.compare_lower("nan")) 2273 Value = APFloat::getNaN(Semantics, false, ~0); 2274 else 2275 return TokError("invalid floating point literal"); 2276 } else if (Value.convertFromString(IDVal, APFloat::rmNearestTiesToEven) == 2277 APFloat::opInvalidOp) 2278 return TokError("invalid floating point literal"); 2279 if (IsNeg) 2280 Value.changeSign(); 2281 2282 // Consume the numeric token. 2283 Lex(); 2284 2285 // Emit the value as an integer. 2286 APInt AsInt = Value.bitcastToAPInt(); 2287 getStreamer().EmitIntValue(AsInt.getLimitedValue(), 2288 AsInt.getBitWidth() / 8, DEFAULT_ADDRSPACE); 2289 2290 if (getLexer().is(AsmToken::EndOfStatement)) 2291 break; 2292 2293 if (getLexer().isNot(AsmToken::Comma)) 2294 return TokError("unexpected token in directive"); 2295 Lex(); 2296 } 2297 } 2298 2299 Lex(); 2300 return false; 2301 } 2302 2303 /// ParseDirectiveZero 2304 /// ::= .zero expression 2305 bool AsmParser::ParseDirectiveZero() { 2306 checkForValidSection(); 2307 2308 int64_t NumBytes; 2309 if (parseAbsoluteExpression(NumBytes)) 2310 return true; 2311 2312 int64_t Val = 0; 2313 if (getLexer().is(AsmToken::Comma)) { 2314 Lex(); 2315 if (parseAbsoluteExpression(Val)) 2316 return true; 2317 } 2318 2319 if (getLexer().isNot(AsmToken::EndOfStatement)) 2320 return TokError("unexpected token in '.zero' directive"); 2321 2322 Lex(); 2323 2324 getStreamer().EmitFill(NumBytes, Val, DEFAULT_ADDRSPACE); 2325 2326 return false; 2327 } 2328 2329 /// ParseDirectiveFill 2330 /// ::= .fill expression , expression , expression 2331 bool AsmParser::ParseDirectiveFill() { 2332 checkForValidSection(); 2333 2334 int64_t NumValues; 2335 if (parseAbsoluteExpression(NumValues)) 2336 return true; 2337 2338 if (getLexer().isNot(AsmToken::Comma)) 2339 return TokError("unexpected token in '.fill' directive"); 2340 Lex(); 2341 2342 int64_t FillSize; 2343 if (parseAbsoluteExpression(FillSize)) 2344 return true; 2345 2346 if (getLexer().isNot(AsmToken::Comma)) 2347 return TokError("unexpected token in '.fill' directive"); 2348 Lex(); 2349 2350 int64_t FillExpr; 2351 if (parseAbsoluteExpression(FillExpr)) 2352 return true; 2353 2354 if (getLexer().isNot(AsmToken::EndOfStatement)) 2355 return TokError("unexpected token in '.fill' directive"); 2356 2357 Lex(); 2358 2359 if (FillSize != 1 && FillSize != 2 && FillSize != 4 && FillSize != 8) 2360 return TokError("invalid '.fill' size, expected 1, 2, 4, or 8"); 2361 2362 for (uint64_t i = 0, e = NumValues; i != e; ++i) 2363 getStreamer().EmitIntValue(FillExpr, FillSize, DEFAULT_ADDRSPACE); 2364 2365 return false; 2366 } 2367 2368 /// ParseDirectiveOrg 2369 /// ::= .org expression [ , expression ] 2370 bool AsmParser::ParseDirectiveOrg() { 2371 checkForValidSection(); 2372 2373 const MCExpr *Offset; 2374 SMLoc Loc = getTok().getLoc(); 2375 if (parseExpression(Offset)) 2376 return true; 2377 2378 // Parse optional fill expression. 2379 int64_t FillExpr = 0; 2380 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2381 if (getLexer().isNot(AsmToken::Comma)) 2382 return TokError("unexpected token in '.org' directive"); 2383 Lex(); 2384 2385 if (parseAbsoluteExpression(FillExpr)) 2386 return true; 2387 2388 if (getLexer().isNot(AsmToken::EndOfStatement)) 2389 return TokError("unexpected token in '.org' directive"); 2390 } 2391 2392 Lex(); 2393 2394 // Only limited forms of relocatable expressions are accepted here, it 2395 // has to be relative to the current section. The streamer will return 2396 // 'true' if the expression wasn't evaluatable. 2397 if (getStreamer().EmitValueToOffset(Offset, FillExpr)) 2398 return Error(Loc, "expected assembly-time absolute expression"); 2399 2400 return false; 2401 } 2402 2403 /// ParseDirectiveAlign 2404 /// ::= {.align, ...} expression [ , expression [ , expression ]] 2405 bool AsmParser::ParseDirectiveAlign(bool IsPow2, unsigned ValueSize) { 2406 checkForValidSection(); 2407 2408 SMLoc AlignmentLoc = getLexer().getLoc(); 2409 int64_t Alignment; 2410 if (parseAbsoluteExpression(Alignment)) 2411 return true; 2412 2413 SMLoc MaxBytesLoc; 2414 bool HasFillExpr = false; 2415 int64_t FillExpr = 0; 2416 int64_t MaxBytesToFill = 0; 2417 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2418 if (getLexer().isNot(AsmToken::Comma)) 2419 return TokError("unexpected token in directive"); 2420 Lex(); 2421 2422 // The fill expression can be omitted while specifying a maximum number of 2423 // alignment bytes, e.g: 2424 // .align 3,,4 2425 if (getLexer().isNot(AsmToken::Comma)) { 2426 HasFillExpr = true; 2427 if (parseAbsoluteExpression(FillExpr)) 2428 return true; 2429 } 2430 2431 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2432 if (getLexer().isNot(AsmToken::Comma)) 2433 return TokError("unexpected token in directive"); 2434 Lex(); 2435 2436 MaxBytesLoc = getLexer().getLoc(); 2437 if (parseAbsoluteExpression(MaxBytesToFill)) 2438 return true; 2439 2440 if (getLexer().isNot(AsmToken::EndOfStatement)) 2441 return TokError("unexpected token in directive"); 2442 } 2443 } 2444 2445 Lex(); 2446 2447 if (!HasFillExpr) 2448 FillExpr = 0; 2449 2450 // Compute alignment in bytes. 2451 if (IsPow2) { 2452 // FIXME: Diagnose overflow. 2453 if (Alignment >= 32) { 2454 Error(AlignmentLoc, "invalid alignment value"); 2455 Alignment = 31; 2456 } 2457 2458 Alignment = 1ULL << Alignment; 2459 } else { 2460 // Reject alignments that aren't a power of two, for gas compatibility. 2461 if (!isPowerOf2_64(Alignment)) 2462 Error(AlignmentLoc, "alignment must be a power of 2"); 2463 } 2464 2465 // Diagnose non-sensical max bytes to align. 2466 if (MaxBytesLoc.isValid()) { 2467 if (MaxBytesToFill < 1) { 2468 Error(MaxBytesLoc, "alignment directive can never be satisfied in this " 2469 "many bytes, ignoring maximum bytes expression"); 2470 MaxBytesToFill = 0; 2471 } 2472 2473 if (MaxBytesToFill >= Alignment) { 2474 Warning(MaxBytesLoc, "maximum bytes expression exceeds alignment and " 2475 "has no effect"); 2476 MaxBytesToFill = 0; 2477 } 2478 } 2479 2480 // Check whether we should use optimal code alignment for this .align 2481 // directive. 2482 bool UseCodeAlign = getStreamer().getCurrentSection()->UseCodeAlign(); 2483 if ((!HasFillExpr || Lexer.getMAI().getTextAlignFillValue() == FillExpr) && 2484 ValueSize == 1 && UseCodeAlign) { 2485 getStreamer().EmitCodeAlignment(Alignment, MaxBytesToFill); 2486 } else { 2487 // FIXME: Target specific behavior about how the "extra" bytes are filled. 2488 getStreamer().EmitValueToAlignment(Alignment, FillExpr, ValueSize, 2489 MaxBytesToFill); 2490 } 2491 2492 return false; 2493 } 2494 2495 /// ParseDirectiveFile 2496 /// ::= .file [number] filename 2497 /// ::= .file number directory filename 2498 bool AsmParser::ParseDirectiveFile(SMLoc DirectiveLoc) { 2499 // FIXME: I'm not sure what this is. 2500 int64_t FileNumber = -1; 2501 SMLoc FileNumberLoc = getLexer().getLoc(); 2502 if (getLexer().is(AsmToken::Integer)) { 2503 FileNumber = getTok().getIntVal(); 2504 Lex(); 2505 2506 if (FileNumber < 1) 2507 return TokError("file number less than one"); 2508 } 2509 2510 if (getLexer().isNot(AsmToken::String)) 2511 return TokError("unexpected token in '.file' directive"); 2512 2513 // Usually the directory and filename together, otherwise just the directory. 2514 StringRef Path = getTok().getString(); 2515 Path = Path.substr(1, Path.size()-2); 2516 Lex(); 2517 2518 StringRef Directory; 2519 StringRef Filename; 2520 if (getLexer().is(AsmToken::String)) { 2521 if (FileNumber == -1) 2522 return TokError("explicit path specified, but no file number"); 2523 Filename = getTok().getString(); 2524 Filename = Filename.substr(1, Filename.size()-2); 2525 Directory = Path; 2526 Lex(); 2527 } else { 2528 Filename = Path; 2529 } 2530 2531 if (getLexer().isNot(AsmToken::EndOfStatement)) 2532 return TokError("unexpected token in '.file' directive"); 2533 2534 if (FileNumber == -1) 2535 getStreamer().EmitFileDirective(Filename); 2536 else { 2537 if (getContext().getGenDwarfForAssembly() == true) 2538 Error(DirectiveLoc, "input can't have .file dwarf directives when -g is " 2539 "used to generate dwarf debug info for assembly code"); 2540 2541 if (getStreamer().EmitDwarfFileDirective(FileNumber, Directory, Filename)) 2542 Error(FileNumberLoc, "file number already allocated"); 2543 } 2544 2545 return false; 2546 } 2547 2548 /// ParseDirectiveLine 2549 /// ::= .line [number] 2550 bool AsmParser::ParseDirectiveLine() { 2551 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2552 if (getLexer().isNot(AsmToken::Integer)) 2553 return TokError("unexpected token in '.line' directive"); 2554 2555 int64_t LineNumber = getTok().getIntVal(); 2556 (void) LineNumber; 2557 Lex(); 2558 2559 // FIXME: Do something with the .line. 2560 } 2561 2562 if (getLexer().isNot(AsmToken::EndOfStatement)) 2563 return TokError("unexpected token in '.line' directive"); 2564 2565 return false; 2566 } 2567 2568 /// ParseDirectiveLoc 2569 /// ::= .loc FileNumber [LineNumber] [ColumnPos] [basic_block] [prologue_end] 2570 /// [epilogue_begin] [is_stmt VALUE] [isa VALUE] 2571 /// The first number is a file number, must have been previously assigned with 2572 /// a .file directive, the second number is the line number and optionally the 2573 /// third number is a column position (zero if not specified). The remaining 2574 /// optional items are .loc sub-directives. 2575 bool AsmParser::ParseDirectiveLoc() { 2576 if (getLexer().isNot(AsmToken::Integer)) 2577 return TokError("unexpected token in '.loc' directive"); 2578 int64_t FileNumber = getTok().getIntVal(); 2579 if (FileNumber < 1) 2580 return TokError("file number less than one in '.loc' directive"); 2581 if (!getContext().isValidDwarfFileNumber(FileNumber)) 2582 return TokError("unassigned file number in '.loc' directive"); 2583 Lex(); 2584 2585 int64_t LineNumber = 0; 2586 if (getLexer().is(AsmToken::Integer)) { 2587 LineNumber = getTok().getIntVal(); 2588 if (LineNumber < 1) 2589 return TokError("line number less than one in '.loc' directive"); 2590 Lex(); 2591 } 2592 2593 int64_t ColumnPos = 0; 2594 if (getLexer().is(AsmToken::Integer)) { 2595 ColumnPos = getTok().getIntVal(); 2596 if (ColumnPos < 0) 2597 return TokError("column position less than zero in '.loc' directive"); 2598 Lex(); 2599 } 2600 2601 unsigned Flags = DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0; 2602 unsigned Isa = 0; 2603 int64_t Discriminator = 0; 2604 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2605 for (;;) { 2606 if (getLexer().is(AsmToken::EndOfStatement)) 2607 break; 2608 2609 StringRef Name; 2610 SMLoc Loc = getTok().getLoc(); 2611 if (parseIdentifier(Name)) 2612 return TokError("unexpected token in '.loc' directive"); 2613 2614 if (Name == "basic_block") 2615 Flags |= DWARF2_FLAG_BASIC_BLOCK; 2616 else if (Name == "prologue_end") 2617 Flags |= DWARF2_FLAG_PROLOGUE_END; 2618 else if (Name == "epilogue_begin") 2619 Flags |= DWARF2_FLAG_EPILOGUE_BEGIN; 2620 else if (Name == "is_stmt") { 2621 Loc = getTok().getLoc(); 2622 const MCExpr *Value; 2623 if (parseExpression(Value)) 2624 return true; 2625 // The expression must be the constant 0 or 1. 2626 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2627 int Value = MCE->getValue(); 2628 if (Value == 0) 2629 Flags &= ~DWARF2_FLAG_IS_STMT; 2630 else if (Value == 1) 2631 Flags |= DWARF2_FLAG_IS_STMT; 2632 else 2633 return Error(Loc, "is_stmt value not 0 or 1"); 2634 } 2635 else { 2636 return Error(Loc, "is_stmt value not the constant value of 0 or 1"); 2637 } 2638 } 2639 else if (Name == "isa") { 2640 Loc = getTok().getLoc(); 2641 const MCExpr *Value; 2642 if (parseExpression(Value)) 2643 return true; 2644 // The expression must be a constant greater or equal to 0. 2645 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2646 int Value = MCE->getValue(); 2647 if (Value < 0) 2648 return Error(Loc, "isa number less than zero"); 2649 Isa = Value; 2650 } 2651 else { 2652 return Error(Loc, "isa number not a constant value"); 2653 } 2654 } 2655 else if (Name == "discriminator") { 2656 if (parseAbsoluteExpression(Discriminator)) 2657 return true; 2658 } 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 } 3629 else { 3630 int64_t ExprValue; 3631 if (parseAbsoluteExpression(ExprValue)) 3632 return true; 3633 3634 if (getLexer().isNot(AsmToken::EndOfStatement)) 3635 return TokError("unexpected token in '.elseif' directive"); 3636 3637 Lex(); 3638 TheCondState.CondMet = ExprValue; 3639 TheCondState.Ignore = !TheCondState.CondMet; 3640 } 3641 3642 return false; 3643 } 3644 3645 /// ParseDirectiveElse 3646 /// ::= .else 3647 bool AsmParser::ParseDirectiveElse(SMLoc DirectiveLoc) { 3648 if (getLexer().isNot(AsmToken::EndOfStatement)) 3649 return TokError("unexpected token in '.else' directive"); 3650 3651 Lex(); 3652 3653 if (TheCondState.TheCond != AsmCond::IfCond && 3654 TheCondState.TheCond != AsmCond::ElseIfCond) 3655 Error(DirectiveLoc, "Encountered a .else that doesn't follow a .if or an " 3656 ".elseif"); 3657 TheCondState.TheCond = AsmCond::ElseCond; 3658 bool LastIgnoreState = false; 3659 if (!TheCondStack.empty()) 3660 LastIgnoreState = TheCondStack.back().Ignore; 3661 if (LastIgnoreState || TheCondState.CondMet) 3662 TheCondState.Ignore = true; 3663 else 3664 TheCondState.Ignore = false; 3665 3666 return false; 3667 } 3668 3669 /// ParseDirectiveEndIf 3670 /// ::= .endif 3671 bool AsmParser::ParseDirectiveEndIf(SMLoc DirectiveLoc) { 3672 if (getLexer().isNot(AsmToken::EndOfStatement)) 3673 return TokError("unexpected token in '.endif' directive"); 3674 3675 Lex(); 3676 3677 if ((TheCondState.TheCond == AsmCond::NoCond) || 3678 TheCondStack.empty()) 3679 Error(DirectiveLoc, "Encountered a .endif that doesn't follow a .if or " 3680 ".else"); 3681 if (!TheCondStack.empty()) { 3682 TheCondState = TheCondStack.back(); 3683 TheCondStack.pop_back(); 3684 } 3685 3686 return false; 3687 } 3688 3689 void AsmParser::initializeDirectiveKindMap() { 3690 DirectiveKindMap[".set"] = DK_SET; 3691 DirectiveKindMap[".equ"] = DK_EQU; 3692 DirectiveKindMap[".equiv"] = DK_EQUIV; 3693 DirectiveKindMap[".ascii"] = DK_ASCII; 3694 DirectiveKindMap[".asciz"] = DK_ASCIZ; 3695 DirectiveKindMap[".string"] = DK_STRING; 3696 DirectiveKindMap[".byte"] = DK_BYTE; 3697 DirectiveKindMap[".short"] = DK_SHORT; 3698 DirectiveKindMap[".value"] = DK_VALUE; 3699 DirectiveKindMap[".2byte"] = DK_2BYTE; 3700 DirectiveKindMap[".long"] = DK_LONG; 3701 DirectiveKindMap[".int"] = DK_INT; 3702 DirectiveKindMap[".4byte"] = DK_4BYTE; 3703 DirectiveKindMap[".quad"] = DK_QUAD; 3704 DirectiveKindMap[".8byte"] = DK_8BYTE; 3705 DirectiveKindMap[".single"] = DK_SINGLE; 3706 DirectiveKindMap[".float"] = DK_FLOAT; 3707 DirectiveKindMap[".double"] = DK_DOUBLE; 3708 DirectiveKindMap[".align"] = DK_ALIGN; 3709 DirectiveKindMap[".align32"] = DK_ALIGN32; 3710 DirectiveKindMap[".balign"] = DK_BALIGN; 3711 DirectiveKindMap[".balignw"] = DK_BALIGNW; 3712 DirectiveKindMap[".balignl"] = DK_BALIGNL; 3713 DirectiveKindMap[".p2align"] = DK_P2ALIGN; 3714 DirectiveKindMap[".p2alignw"] = DK_P2ALIGNW; 3715 DirectiveKindMap[".p2alignl"] = DK_P2ALIGNL; 3716 DirectiveKindMap[".org"] = DK_ORG; 3717 DirectiveKindMap[".fill"] = DK_FILL; 3718 DirectiveKindMap[".zero"] = DK_ZERO; 3719 DirectiveKindMap[".extern"] = DK_EXTERN; 3720 DirectiveKindMap[".globl"] = DK_GLOBL; 3721 DirectiveKindMap[".global"] = DK_GLOBAL; 3722 DirectiveKindMap[".indirect_symbol"] = DK_INDIRECT_SYMBOL; 3723 DirectiveKindMap[".lazy_reference"] = DK_LAZY_REFERENCE; 3724 DirectiveKindMap[".no_dead_strip"] = DK_NO_DEAD_STRIP; 3725 DirectiveKindMap[".symbol_resolver"] = DK_SYMBOL_RESOLVER; 3726 DirectiveKindMap[".private_extern"] = DK_PRIVATE_EXTERN; 3727 DirectiveKindMap[".reference"] = DK_REFERENCE; 3728 DirectiveKindMap[".weak_definition"] = DK_WEAK_DEFINITION; 3729 DirectiveKindMap[".weak_reference"] = DK_WEAK_REFERENCE; 3730 DirectiveKindMap[".weak_def_can_be_hidden"] = DK_WEAK_DEF_CAN_BE_HIDDEN; 3731 DirectiveKindMap[".comm"] = DK_COMM; 3732 DirectiveKindMap[".common"] = DK_COMMON; 3733 DirectiveKindMap[".lcomm"] = DK_LCOMM; 3734 DirectiveKindMap[".abort"] = DK_ABORT; 3735 DirectiveKindMap[".include"] = DK_INCLUDE; 3736 DirectiveKindMap[".incbin"] = DK_INCBIN; 3737 DirectiveKindMap[".code16"] = DK_CODE16; 3738 DirectiveKindMap[".code16gcc"] = DK_CODE16GCC; 3739 DirectiveKindMap[".rept"] = DK_REPT; 3740 DirectiveKindMap[".irp"] = DK_IRP; 3741 DirectiveKindMap[".irpc"] = DK_IRPC; 3742 DirectiveKindMap[".endr"] = DK_ENDR; 3743 DirectiveKindMap[".bundle_align_mode"] = DK_BUNDLE_ALIGN_MODE; 3744 DirectiveKindMap[".bundle_lock"] = DK_BUNDLE_LOCK; 3745 DirectiveKindMap[".bundle_unlock"] = DK_BUNDLE_UNLOCK; 3746 DirectiveKindMap[".if"] = DK_IF; 3747 DirectiveKindMap[".ifb"] = DK_IFB; 3748 DirectiveKindMap[".ifnb"] = DK_IFNB; 3749 DirectiveKindMap[".ifc"] = DK_IFC; 3750 DirectiveKindMap[".ifnc"] = DK_IFNC; 3751 DirectiveKindMap[".ifdef"] = DK_IFDEF; 3752 DirectiveKindMap[".ifndef"] = DK_IFNDEF; 3753 DirectiveKindMap[".ifnotdef"] = DK_IFNOTDEF; 3754 DirectiveKindMap[".elseif"] = DK_ELSEIF; 3755 DirectiveKindMap[".else"] = DK_ELSE; 3756 DirectiveKindMap[".endif"] = DK_ENDIF; 3757 DirectiveKindMap[".skip"] = DK_SKIP; 3758 DirectiveKindMap[".space"] = DK_SPACE; 3759 DirectiveKindMap[".file"] = DK_FILE; 3760 DirectiveKindMap[".line"] = DK_LINE; 3761 DirectiveKindMap[".loc"] = DK_LOC; 3762 DirectiveKindMap[".stabs"] = DK_STABS; 3763 DirectiveKindMap[".sleb128"] = DK_SLEB128; 3764 DirectiveKindMap[".uleb128"] = DK_ULEB128; 3765 DirectiveKindMap[".cfi_sections"] = DK_CFI_SECTIONS; 3766 DirectiveKindMap[".cfi_startproc"] = DK_CFI_STARTPROC; 3767 DirectiveKindMap[".cfi_endproc"] = DK_CFI_ENDPROC; 3768 DirectiveKindMap[".cfi_def_cfa"] = DK_CFI_DEF_CFA; 3769 DirectiveKindMap[".cfi_def_cfa_offset"] = DK_CFI_DEF_CFA_OFFSET; 3770 DirectiveKindMap[".cfi_adjust_cfa_offset"] = DK_CFI_ADJUST_CFA_OFFSET; 3771 DirectiveKindMap[".cfi_def_cfa_register"] = DK_CFI_DEF_CFA_REGISTER; 3772 DirectiveKindMap[".cfi_offset"] = DK_CFI_OFFSET; 3773 DirectiveKindMap[".cfi_rel_offset"] = DK_CFI_REL_OFFSET; 3774 DirectiveKindMap[".cfi_personality"] = DK_CFI_PERSONALITY; 3775 DirectiveKindMap[".cfi_lsda"] = DK_CFI_LSDA; 3776 DirectiveKindMap[".cfi_remember_state"] = DK_CFI_REMEMBER_STATE; 3777 DirectiveKindMap[".cfi_restore_state"] = DK_CFI_RESTORE_STATE; 3778 DirectiveKindMap[".cfi_same_value"] = DK_CFI_SAME_VALUE; 3779 DirectiveKindMap[".cfi_restore"] = DK_CFI_RESTORE; 3780 DirectiveKindMap[".cfi_escape"] = DK_CFI_ESCAPE; 3781 DirectiveKindMap[".cfi_signal_frame"] = DK_CFI_SIGNAL_FRAME; 3782 DirectiveKindMap[".cfi_undefined"] = DK_CFI_UNDEFINED; 3783 DirectiveKindMap[".cfi_register"] = DK_CFI_REGISTER; 3784 DirectiveKindMap[".macros_on"] = DK_MACROS_ON; 3785 DirectiveKindMap[".macros_off"] = DK_MACROS_OFF; 3786 DirectiveKindMap[".macro"] = DK_MACRO; 3787 DirectiveKindMap[".endm"] = DK_ENDM; 3788 DirectiveKindMap[".endmacro"] = DK_ENDMACRO; 3789 DirectiveKindMap[".purgem"] = DK_PURGEM; 3790 } 3791 3792 3793 MCAsmMacro *AsmParser::ParseMacroLikeBody(SMLoc DirectiveLoc) { 3794 AsmToken EndToken, StartToken = getTok(); 3795 3796 unsigned NestLevel = 0; 3797 for (;;) { 3798 // Check whether we have reached the end of the file. 3799 if (getLexer().is(AsmToken::Eof)) { 3800 Error(DirectiveLoc, "no matching '.endr' in definition"); 3801 return 0; 3802 } 3803 3804 if (Lexer.is(AsmToken::Identifier) && 3805 (getTok().getIdentifier() == ".rept")) { 3806 ++NestLevel; 3807 } 3808 3809 // Otherwise, check whether we have reached the .endr. 3810 if (Lexer.is(AsmToken::Identifier) && 3811 getTok().getIdentifier() == ".endr") { 3812 if (NestLevel == 0) { 3813 EndToken = getTok(); 3814 Lex(); 3815 if (Lexer.isNot(AsmToken::EndOfStatement)) { 3816 TokError("unexpected token in '.endr' directive"); 3817 return 0; 3818 } 3819 break; 3820 } 3821 --NestLevel; 3822 } 3823 3824 // Otherwise, scan till the end of the statement. 3825 eatToEndOfStatement(); 3826 } 3827 3828 const char *BodyStart = StartToken.getLoc().getPointer(); 3829 const char *BodyEnd = EndToken.getLoc().getPointer(); 3830 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 3831 3832 // We Are Anonymous. 3833 StringRef Name; 3834 MCAsmMacroParameters Parameters; 3835 return new MCAsmMacro(Name, Body, Parameters); 3836 } 3837 3838 void AsmParser::InstantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 3839 raw_svector_ostream &OS) { 3840 OS << ".endr\n"; 3841 3842 MemoryBuffer *Instantiation = 3843 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 3844 3845 // Create the macro instantiation object and add to the current macro 3846 // instantiation stack. 3847 MacroInstantiation *MI = new MacroInstantiation(M, DirectiveLoc, 3848 CurBuffer, 3849 getTok().getLoc(), 3850 Instantiation); 3851 ActiveMacros.push_back(MI); 3852 3853 // Jump to the macro instantiation and prime the lexer. 3854 CurBuffer = SrcMgr.AddNewSourceBuffer(MI->Instantiation, SMLoc()); 3855 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)); 3856 Lex(); 3857 } 3858 3859 bool AsmParser::ParseDirectiveRept(SMLoc DirectiveLoc) { 3860 int64_t Count; 3861 if (parseAbsoluteExpression(Count)) 3862 return TokError("unexpected token in '.rept' directive"); 3863 3864 if (Count < 0) 3865 return TokError("Count is negative"); 3866 3867 if (Lexer.isNot(AsmToken::EndOfStatement)) 3868 return TokError("unexpected token in '.rept' directive"); 3869 3870 // Eat the end of statement. 3871 Lex(); 3872 3873 // Lex the rept definition. 3874 MCAsmMacro *M = ParseMacroLikeBody(DirectiveLoc); 3875 if (!M) 3876 return true; 3877 3878 // Macro instantiation is lexical, unfortunately. We construct a new buffer 3879 // to hold the macro body with substitutions. 3880 SmallString<256> Buf; 3881 MCAsmMacroParameters Parameters; 3882 MCAsmMacroArguments A; 3883 raw_svector_ostream OS(Buf); 3884 while (Count--) { 3885 if (expandMacro(OS, M->Body, Parameters, A, getTok().getLoc())) 3886 return true; 3887 } 3888 InstantiateMacroLikeBody(M, DirectiveLoc, OS); 3889 3890 return false; 3891 } 3892 3893 /// ParseDirectiveIrp 3894 /// ::= .irp symbol,values 3895 bool AsmParser::ParseDirectiveIrp(SMLoc DirectiveLoc) { 3896 MCAsmMacroParameters Parameters; 3897 MCAsmMacroParameter Parameter; 3898 3899 if (parseIdentifier(Parameter.first)) 3900 return TokError("expected identifier in '.irp' directive"); 3901 3902 Parameters.push_back(Parameter); 3903 3904 if (Lexer.isNot(AsmToken::Comma)) 3905 return TokError("expected comma in '.irp' directive"); 3906 3907 Lex(); 3908 3909 MCAsmMacroArguments A; 3910 if (ParseMacroArguments(0, A)) 3911 return true; 3912 3913 // Eat the end of statement. 3914 Lex(); 3915 3916 // Lex the irp definition. 3917 MCAsmMacro *M = ParseMacroLikeBody(DirectiveLoc); 3918 if (!M) 3919 return true; 3920 3921 // Macro instantiation is lexical, unfortunately. We construct a new buffer 3922 // to hold the macro body with substitutions. 3923 SmallString<256> Buf; 3924 raw_svector_ostream OS(Buf); 3925 3926 for (MCAsmMacroArguments::iterator i = A.begin(), e = A.end(); i != e; ++i) { 3927 MCAsmMacroArguments Args; 3928 Args.push_back(*i); 3929 3930 if (expandMacro(OS, M->Body, Parameters, Args, getTok().getLoc())) 3931 return true; 3932 } 3933 3934 InstantiateMacroLikeBody(M, DirectiveLoc, OS); 3935 3936 return false; 3937 } 3938 3939 /// ParseDirectiveIrpc 3940 /// ::= .irpc symbol,values 3941 bool AsmParser::ParseDirectiveIrpc(SMLoc DirectiveLoc) { 3942 MCAsmMacroParameters Parameters; 3943 MCAsmMacroParameter Parameter; 3944 3945 if (parseIdentifier(Parameter.first)) 3946 return TokError("expected identifier in '.irpc' directive"); 3947 3948 Parameters.push_back(Parameter); 3949 3950 if (Lexer.isNot(AsmToken::Comma)) 3951 return TokError("expected comma in '.irpc' directive"); 3952 3953 Lex(); 3954 3955 MCAsmMacroArguments A; 3956 if (ParseMacroArguments(0, A)) 3957 return true; 3958 3959 if (A.size() != 1 || A.front().size() != 1) 3960 return TokError("unexpected token in '.irpc' directive"); 3961 3962 // Eat the end of statement. 3963 Lex(); 3964 3965 // Lex the irpc definition. 3966 MCAsmMacro *M = ParseMacroLikeBody(DirectiveLoc); 3967 if (!M) 3968 return true; 3969 3970 // Macro instantiation is lexical, unfortunately. We construct a new buffer 3971 // to hold the macro body with substitutions. 3972 SmallString<256> Buf; 3973 raw_svector_ostream OS(Buf); 3974 3975 StringRef Values = A.front().front().getString(); 3976 std::size_t I, End = Values.size(); 3977 for (I = 0; I < End; ++I) { 3978 MCAsmMacroArgument Arg; 3979 Arg.push_back(AsmToken(AsmToken::Identifier, Values.slice(I, I+1))); 3980 3981 MCAsmMacroArguments Args; 3982 Args.push_back(Arg); 3983 3984 if (expandMacro(OS, M->Body, Parameters, Args, getTok().getLoc())) 3985 return true; 3986 } 3987 3988 InstantiateMacroLikeBody(M, DirectiveLoc, OS); 3989 3990 return false; 3991 } 3992 3993 bool AsmParser::ParseDirectiveEndr(SMLoc DirectiveLoc) { 3994 if (ActiveMacros.empty()) 3995 return TokError("unmatched '.endr' directive"); 3996 3997 // The only .repl that should get here are the ones created by 3998 // InstantiateMacroLikeBody. 3999 assert(getLexer().is(AsmToken::EndOfStatement)); 4000 4001 HandleMacroExit(); 4002 return false; 4003 } 4004 4005 bool AsmParser::ParseDirectiveMSEmit(SMLoc IDLoc, ParseStatementInfo &Info, 4006 size_t Len) { 4007 const MCExpr *Value; 4008 SMLoc ExprLoc = getLexer().getLoc(); 4009 if (parseExpression(Value)) 4010 return true; 4011 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 4012 if (!MCE) 4013 return Error(ExprLoc, "unexpected expression in _emit"); 4014 uint64_t IntValue = MCE->getValue(); 4015 if (!isUIntN(8, IntValue) && !isIntN(8, IntValue)) 4016 return Error(ExprLoc, "literal value out of range for directive"); 4017 4018 Info.AsmRewrites->push_back(AsmRewrite(AOK_Emit, IDLoc, Len)); 4019 return false; 4020 } 4021 4022 bool AsmParser::ParseDirectiveMSAlign(SMLoc IDLoc, ParseStatementInfo &Info) { 4023 const MCExpr *Value; 4024 SMLoc ExprLoc = getLexer().getLoc(); 4025 if (parseExpression(Value)) 4026 return true; 4027 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 4028 if (!MCE) 4029 return Error(ExprLoc, "unexpected expression in align"); 4030 uint64_t IntValue = MCE->getValue(); 4031 if (!isPowerOf2_64(IntValue)) 4032 return Error(ExprLoc, "literal value not a power of two greater then zero"); 4033 4034 Info.AsmRewrites->push_back(AsmRewrite(AOK_Align, IDLoc, 5, 4035 Log2_64(IntValue))); 4036 return false; 4037 } 4038 4039 // We are comparing pointers, but the pointers are relative to a single string. 4040 // Thus, this should always be deterministic. 4041 static int RewritesSort(const void *A, const void *B) { 4042 const AsmRewrite *AsmRewriteA = static_cast<const AsmRewrite *>(A); 4043 const AsmRewrite *AsmRewriteB = static_cast<const AsmRewrite *>(B); 4044 if (AsmRewriteA->Loc.getPointer() < AsmRewriteB->Loc.getPointer()) 4045 return -1; 4046 if (AsmRewriteB->Loc.getPointer() < AsmRewriteA->Loc.getPointer()) 4047 return 1; 4048 4049 // It's possible to have a SizeDirective rewrite and an Input/Output rewrite 4050 // to the same location. Make sure the SizeDirective rewrite is performed 4051 // first. This also ensure the sort algorithm is stable. 4052 if (AsmRewriteA->Kind == AOK_SizeDirective) { 4053 assert ((AsmRewriteB->Kind == AOK_Input || AsmRewriteB->Kind == AOK_Output) && 4054 "Expected an Input/Output rewrite!"); 4055 return -1; 4056 } 4057 if (AsmRewriteB->Kind == AOK_SizeDirective) { 4058 assert ((AsmRewriteA->Kind == AOK_Input || AsmRewriteA->Kind == AOK_Output) && 4059 "Expected an Input/Output rewrite!"); 4060 return 1; 4061 } 4062 llvm_unreachable ("Unstable rewrite sort."); 4063 } 4064 4065 bool 4066 AsmParser::parseMSInlineAsm(void *AsmLoc, std::string &AsmString, 4067 unsigned &NumOutputs, unsigned &NumInputs, 4068 SmallVectorImpl<std::pair<void *, bool> > &OpDecls, 4069 SmallVectorImpl<std::string> &Constraints, 4070 SmallVectorImpl<std::string> &Clobbers, 4071 const MCInstrInfo *MII, 4072 const MCInstPrinter *IP, 4073 MCAsmParserSemaCallback &SI) { 4074 SmallVector<void *, 4> InputDecls; 4075 SmallVector<void *, 4> OutputDecls; 4076 SmallVector<bool, 4> InputDeclsAddressOf; 4077 SmallVector<bool, 4> OutputDeclsAddressOf; 4078 SmallVector<std::string, 4> InputConstraints; 4079 SmallVector<std::string, 4> OutputConstraints; 4080 SmallVector<unsigned, 4> ClobberRegs; 4081 4082 SmallVector<AsmRewrite, 4> AsmStrRewrites; 4083 4084 // Prime the lexer. 4085 Lex(); 4086 4087 // While we have input, parse each statement. 4088 unsigned InputIdx = 0; 4089 unsigned OutputIdx = 0; 4090 while (getLexer().isNot(AsmToken::Eof)) { 4091 ParseStatementInfo Info(&AsmStrRewrites); 4092 if (ParseStatement(Info)) 4093 return true; 4094 4095 if (Info.ParseError) 4096 return true; 4097 4098 if (Info.Opcode == ~0U) 4099 continue; 4100 4101 const MCInstrDesc &Desc = MII->get(Info.Opcode); 4102 4103 // Build the list of clobbers, outputs and inputs. 4104 for (unsigned i = 1, e = Info.ParsedOperands.size(); i != e; ++i) { 4105 MCParsedAsmOperand *Operand = Info.ParsedOperands[i]; 4106 4107 // Immediate. 4108 if (Operand->isImm()) { 4109 if (Operand->needAsmRewrite()) 4110 AsmStrRewrites.push_back(AsmRewrite(AOK_ImmPrefix, 4111 Operand->getStartLoc())); 4112 continue; 4113 } 4114 4115 // Register operand. 4116 if (Operand->isReg() && !Operand->needAddressOf()) { 4117 unsigned NumDefs = Desc.getNumDefs(); 4118 // Clobber. 4119 if (NumDefs && Operand->getMCOperandNum() < NumDefs) 4120 ClobberRegs.push_back(Operand->getReg()); 4121 continue; 4122 } 4123 4124 // Expr/Input or Output. 4125 bool IsVarDecl; 4126 unsigned Length, Size, Type; 4127 void *OpDecl = SI.LookupInlineAsmIdentifier(Operand->getName(), AsmLoc, 4128 Length, Size, Type, 4129 IsVarDecl); 4130 if (!OpDecl) 4131 continue; 4132 4133 bool isOutput = (i == 1) && Desc.mayStore(); 4134 if (Operand->isMem() && Operand->needSizeDirective()) 4135 AsmStrRewrites.push_back(AsmRewrite(AOK_SizeDirective, 4136 Operand->getStartLoc(), /*Len*/0, 4137 Operand->getMemSize())); 4138 4139 if (isOutput) { 4140 ++InputIdx; 4141 OutputDecls.push_back(OpDecl); 4142 OutputDeclsAddressOf.push_back(Operand->needAddressOf()); 4143 OutputConstraints.push_back('=' + Operand->getConstraint().str()); 4144 AsmStrRewrites.push_back(AsmRewrite(AOK_Output, Operand->getStartLoc(), 4145 Operand->getNameLen())); 4146 } else { 4147 InputDecls.push_back(OpDecl); 4148 InputDeclsAddressOf.push_back(Operand->needAddressOf()); 4149 InputConstraints.push_back(Operand->getConstraint().str()); 4150 AsmStrRewrites.push_back(AsmRewrite(AOK_Input, Operand->getStartLoc(), 4151 Operand->getNameLen())); 4152 } 4153 } 4154 } 4155 4156 // Set the number of Outputs and Inputs. 4157 NumOutputs = OutputDecls.size(); 4158 NumInputs = InputDecls.size(); 4159 4160 // Set the unique clobbers. 4161 array_pod_sort(ClobberRegs.begin(), ClobberRegs.end()); 4162 ClobberRegs.erase(std::unique(ClobberRegs.begin(), ClobberRegs.end()), 4163 ClobberRegs.end()); 4164 Clobbers.assign(ClobberRegs.size(), std::string()); 4165 for (unsigned I = 0, E = ClobberRegs.size(); I != E; ++I) { 4166 raw_string_ostream OS(Clobbers[I]); 4167 IP->printRegName(OS, ClobberRegs[I]); 4168 } 4169 4170 // Merge the various outputs and inputs. Output are expected first. 4171 if (NumOutputs || NumInputs) { 4172 unsigned NumExprs = NumOutputs + NumInputs; 4173 OpDecls.resize(NumExprs); 4174 Constraints.resize(NumExprs); 4175 for (unsigned i = 0; i < NumOutputs; ++i) { 4176 OpDecls[i] = std::make_pair(OutputDecls[i], OutputDeclsAddressOf[i]); 4177 Constraints[i] = OutputConstraints[i]; 4178 } 4179 for (unsigned i = 0, j = NumOutputs; i < NumInputs; ++i, ++j) { 4180 OpDecls[j] = std::make_pair(InputDecls[i], InputDeclsAddressOf[i]); 4181 Constraints[j] = InputConstraints[i]; 4182 } 4183 } 4184 4185 // Build the IR assembly string. 4186 std::string AsmStringIR; 4187 AsmRewriteKind PrevKind = AOK_Imm; 4188 raw_string_ostream OS(AsmStringIR); 4189 const char *Start = SrcMgr.getMemoryBuffer(0)->getBufferStart(); 4190 array_pod_sort(AsmStrRewrites.begin(), AsmStrRewrites.end(), RewritesSort); 4191 for (SmallVectorImpl<AsmRewrite>::iterator I = AsmStrRewrites.begin(), 4192 E = AsmStrRewrites.end(); 4193 I != E; ++I) { 4194 const char *Loc = (*I).Loc.getPointer(); 4195 assert(Loc >= Start && "Expected Loc to be after Start!"); 4196 4197 unsigned AdditionalSkip = 0; 4198 AsmRewriteKind Kind = (*I).Kind; 4199 4200 // Emit everything up to the immediate/expression. If the previous rewrite 4201 // was a size directive, then this has already been done. 4202 if (PrevKind != AOK_SizeDirective) 4203 OS << StringRef(Start, Loc - Start); 4204 PrevKind = Kind; 4205 4206 // Skip the original expression. 4207 if (Kind == AOK_Skip) { 4208 Start = Loc + (*I).Len; 4209 continue; 4210 } 4211 4212 // Rewrite expressions in $N notation. 4213 switch (Kind) { 4214 default: break; 4215 case AOK_Imm: 4216 OS << "$$" << (*I).Val; 4217 break; 4218 case AOK_ImmPrefix: 4219 OS << "$$"; 4220 break; 4221 case AOK_Input: 4222 OS << '$' << InputIdx++; 4223 break; 4224 case AOK_Output: 4225 OS << '$' << OutputIdx++; 4226 break; 4227 case AOK_SizeDirective: 4228 switch ((*I).Val) { 4229 default: break; 4230 case 8: OS << "byte ptr "; break; 4231 case 16: OS << "word ptr "; break; 4232 case 32: OS << "dword ptr "; break; 4233 case 64: OS << "qword ptr "; break; 4234 case 80: OS << "xword ptr "; break; 4235 case 128: OS << "xmmword ptr "; break; 4236 case 256: OS << "ymmword ptr "; break; 4237 } 4238 break; 4239 case AOK_Emit: 4240 OS << ".byte"; 4241 break; 4242 case AOK_Align: { 4243 unsigned Val = (*I).Val; 4244 OS << ".align " << Val; 4245 4246 // Skip the original immediate. 4247 assert(Val < 10 && "Expected alignment less then 2^10."); 4248 AdditionalSkip = (Val < 4) ? 2 : Val < 7 ? 3 : 4; 4249 break; 4250 } 4251 case AOK_DotOperator: 4252 OS << (*I).Val; 4253 break; 4254 } 4255 4256 // Skip the original expression. 4257 if (Kind != AOK_SizeDirective) 4258 Start = Loc + (*I).Len + AdditionalSkip; 4259 } 4260 4261 // Emit the remainder of the asm string. 4262 const char *AsmEnd = SrcMgr.getMemoryBuffer(0)->getBufferEnd(); 4263 if (Start != AsmEnd) 4264 OS << StringRef(Start, AsmEnd - Start); 4265 4266 AsmString = OS.str(); 4267 return false; 4268 } 4269 4270 /// \brief Create an MCAsmParser instance. 4271 MCAsmParser *llvm::createMCAsmParser(SourceMgr &SM, 4272 MCContext &C, MCStreamer &Out, 4273 const MCAsmInfo &MAI) { 4274 return new AsmParser(SM, C, Out, MAI); 4275 } 4276