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/STLExtras.h" 16 #include "llvm/ADT/SmallString.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/MCObjectFileInfo.h" 26 #include "llvm/MC/MCParser/AsmCond.h" 27 #include "llvm/MC/MCParser/AsmLexer.h" 28 #include "llvm/MC/MCParser/MCAsmParser.h" 29 #include "llvm/MC/MCParser/MCParsedAsmOperand.h" 30 #include "llvm/MC/MCRegisterInfo.h" 31 #include "llvm/MC/MCSectionMachO.h" 32 #include "llvm/MC/MCStreamer.h" 33 #include "llvm/MC/MCSymbol.h" 34 #include "llvm/MC/MCTargetAsmParser.h" 35 #include "llvm/Support/CommandLine.h" 36 #include "llvm/Support/ErrorHandling.h" 37 #include "llvm/Support/MathExtras.h" 38 #include "llvm/Support/MemoryBuffer.h" 39 #include "llvm/Support/SourceMgr.h" 40 #include "llvm/Support/raw_ostream.h" 41 #include <cctype> 42 #include <deque> 43 #include <set> 44 #include <string> 45 #include <vector> 46 using namespace llvm; 47 48 static cl::opt<bool> 49 FatalAssemblerWarnings("fatal-assembler-warnings", 50 cl::desc("Consider warnings as error")); 51 52 MCAsmParserSemaCallback::~MCAsmParserSemaCallback() {} 53 54 namespace { 55 /// \brief Helper types for tracking macro definitions. 56 typedef std::vector<AsmToken> MCAsmMacroArgument; 57 typedef std::vector<MCAsmMacroArgument> MCAsmMacroArguments; 58 59 struct MCAsmMacroParameter { 60 StringRef Name; 61 MCAsmMacroArgument Value; 62 bool Required; 63 bool Vararg; 64 65 MCAsmMacroParameter() : Required(false), Vararg(false) {} 66 }; 67 68 typedef std::vector<MCAsmMacroParameter> MCAsmMacroParameters; 69 70 struct MCAsmMacro { 71 StringRef Name; 72 StringRef Body; 73 MCAsmMacroParameters Parameters; 74 75 public: 76 MCAsmMacro(StringRef N, StringRef B, ArrayRef<MCAsmMacroParameter> P) : 77 Name(N), Body(B), Parameters(P) {} 78 }; 79 80 /// \brief Helper class for storing information about an active macro 81 /// instantiation. 82 struct MacroInstantiation { 83 /// The macro being instantiated. 84 const MCAsmMacro *TheMacro; 85 86 /// The macro instantiation with substitutions. 87 MemoryBuffer *Instantiation; 88 89 /// The location of the instantiation. 90 SMLoc InstantiationLoc; 91 92 /// The buffer where parsing should resume upon instantiation completion. 93 int ExitBuffer; 94 95 /// The location where parsing should resume upon instantiation completion. 96 SMLoc ExitLoc; 97 98 public: 99 MacroInstantiation(const MCAsmMacro *M, SMLoc IL, int EB, SMLoc EL, 100 MemoryBuffer *I); 101 }; 102 103 struct ParseStatementInfo { 104 /// \brief The parsed operands from the last parsed statement. 105 SmallVector<std::unique_ptr<MCParsedAsmOperand>, 8> ParsedOperands; 106 107 /// \brief The opcode from the last parsed instruction. 108 unsigned Opcode; 109 110 /// \brief Was there an error parsing the inline assembly? 111 bool ParseError; 112 113 SmallVectorImpl<AsmRewrite> *AsmRewrites; 114 115 ParseStatementInfo() : Opcode(~0U), ParseError(false), AsmRewrites(nullptr) {} 116 ParseStatementInfo(SmallVectorImpl<AsmRewrite> *rewrites) 117 : Opcode(~0), ParseError(false), AsmRewrites(rewrites) {} 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 unsigned CurBuffer; 137 138 AsmCond TheCondState; 139 std::vector<AsmCond> TheCondStack; 140 141 /// \brief 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 /// \brief Map of currently defined macros. 147 StringMap<MCAsmMacro*> MacroMap; 148 149 /// \brief Stack of active macro instantiations. 150 std::vector<MacroInstantiation*> ActiveMacros; 151 152 /// \brief List of bodies of anonymous macros. 153 std::deque<MCAsmMacro> MacroLikeBodies; 154 155 /// Boolean tracking whether macro substitution is enabled. 156 unsigned MacrosEnabledFlag : 1; 157 158 /// Flag tracking whether any errors have been encountered. 159 unsigned HadError : 1; 160 161 /// The values from the last parsed cpp hash file line comment if any. 162 StringRef CppHashFilename; 163 int64_t CppHashLineNumber; 164 SMLoc CppHashLoc; 165 unsigned CppHashBuf; 166 /// When generating dwarf for assembly source files we need to calculate the 167 /// logical line number based on the last parsed cpp hash file line comment 168 /// and current line. Since this is slow and messes up the SourceMgr's 169 /// cache we save the last info we queried with SrcMgr.FindLineNumber(). 170 SMLoc LastQueryIDLoc; 171 unsigned LastQueryBuffer; 172 unsigned LastQueryLine; 173 174 /// AssemblerDialect. ~OU means unset value and use value provided by MAI. 175 unsigned AssemblerDialect; 176 177 /// \brief is Darwin compatibility enabled? 178 bool IsDarwin; 179 180 /// \brief Are we parsing ms-style inline assembly? 181 bool ParsingInlineAsm; 182 183 public: 184 AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out, 185 const MCAsmInfo &MAI); 186 virtual ~AsmParser(); 187 188 bool Run(bool NoInitialTextSection, bool NoFinalize = false) override; 189 190 void addDirectiveHandler(StringRef Directive, 191 ExtensionDirectiveHandler Handler) override { 192 ExtensionDirectiveMap[Directive] = Handler; 193 } 194 195 public: 196 /// @name MCAsmParser Interface 197 /// { 198 199 SourceMgr &getSourceManager() override { return SrcMgr; } 200 MCAsmLexer &getLexer() override { return Lexer; } 201 MCContext &getContext() override { return Ctx; } 202 MCStreamer &getStreamer() override { return Out; } 203 unsigned getAssemblerDialect() override { 204 if (AssemblerDialect == ~0U) 205 return MAI.getAssemblerDialect(); 206 else 207 return AssemblerDialect; 208 } 209 void setAssemblerDialect(unsigned i) override { 210 AssemblerDialect = i; 211 } 212 213 void Note(SMLoc L, const Twine &Msg, 214 ArrayRef<SMRange> Ranges = None) override; 215 bool Warning(SMLoc L, const Twine &Msg, 216 ArrayRef<SMRange> Ranges = None) override; 217 bool Error(SMLoc L, const Twine &Msg, 218 ArrayRef<SMRange> Ranges = None) override; 219 220 const AsmToken &Lex() override; 221 222 void setParsingInlineAsm(bool V) override { ParsingInlineAsm = V; } 223 bool isParsingInlineAsm() override { return ParsingInlineAsm; } 224 225 bool parseMSInlineAsm(void *AsmLoc, std::string &AsmString, 226 unsigned &NumOutputs, unsigned &NumInputs, 227 SmallVectorImpl<std::pair<void *,bool> > &OpDecls, 228 SmallVectorImpl<std::string> &Constraints, 229 SmallVectorImpl<std::string> &Clobbers, 230 const MCInstrInfo *MII, const MCInstPrinter *IP, 231 MCAsmParserSemaCallback &SI) override; 232 233 bool parseExpression(const MCExpr *&Res); 234 bool parseExpression(const MCExpr *&Res, SMLoc &EndLoc) override; 235 bool parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) override; 236 bool parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) override; 237 bool parseAbsoluteExpression(int64_t &Res) override; 238 239 /// \brief Parse an identifier or string (as a quoted identifier) 240 /// and set \p Res to the identifier contents. 241 bool parseIdentifier(StringRef &Res) override; 242 void eatToEndOfStatement() override; 243 244 void checkForValidSection() override; 245 /// } 246 247 private: 248 249 bool parseStatement(ParseStatementInfo &Info); 250 void eatToEndOfLine(); 251 bool parseCppHashLineFilenameComment(const SMLoc &L); 252 253 void checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, StringRef Body, 254 ArrayRef<MCAsmMacroParameter> Parameters); 255 bool expandMacro(raw_svector_ostream &OS, StringRef Body, 256 ArrayRef<MCAsmMacroParameter> Parameters, 257 ArrayRef<MCAsmMacroArgument> A, 258 const SMLoc &L); 259 260 /// \brief Are macros enabled in the parser? 261 bool areMacrosEnabled() {return MacrosEnabledFlag;} 262 263 /// \brief Control a flag in the parser that enables or disables macros. 264 void setMacrosEnabled(bool Flag) {MacrosEnabledFlag = Flag;} 265 266 /// \brief Lookup a previously defined macro. 267 /// \param Name Macro name. 268 /// \returns Pointer to macro. NULL if no such macro was defined. 269 const MCAsmMacro* lookupMacro(StringRef Name); 270 271 /// \brief Define a new macro with the given name and information. 272 void defineMacro(StringRef Name, const MCAsmMacro& Macro); 273 274 /// \brief Undefine a macro. If no such macro was defined, it's a no-op. 275 void undefineMacro(StringRef Name); 276 277 /// \brief Are we inside a macro instantiation? 278 bool isInsideMacroInstantiation() {return !ActiveMacros.empty();} 279 280 /// \brief Handle entry to macro instantiation. 281 /// 282 /// \param M The macro. 283 /// \param NameLoc Instantiation location. 284 bool handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc); 285 286 /// \brief Handle exit from macro instantiation. 287 void handleMacroExit(); 288 289 /// \brief Extract AsmTokens for a macro argument. 290 bool parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg); 291 292 /// \brief Parse all macro arguments for a given macro. 293 bool parseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A); 294 295 void printMacroInstantiations(); 296 void printMessage(SMLoc Loc, SourceMgr::DiagKind Kind, const Twine &Msg, 297 ArrayRef<SMRange> Ranges = None) const { 298 SrcMgr.PrintMessage(Loc, Kind, Msg, Ranges); 299 } 300 static void DiagHandler(const SMDiagnostic &Diag, void *Context); 301 302 /// \brief Enter the specified file. This returns true on failure. 303 bool enterIncludeFile(const std::string &Filename); 304 305 /// \brief Process the specified file for the .incbin directive. 306 /// This returns true on failure. 307 bool processIncbinFile(const std::string &Filename); 308 309 /// \brief Reset the current lexer position to that given by \p Loc. The 310 /// current token is not set; clients should ensure Lex() is called 311 /// subsequently. 312 /// 313 /// \param InBuffer If not 0, should be the known buffer id that contains the 314 /// location. 315 void jumpToLoc(SMLoc Loc, unsigned InBuffer = 0); 316 317 /// \brief Parse up to the end of statement and a return the contents from the 318 /// current token until the end of the statement; the current token on exit 319 /// will be either the EndOfStatement or EOF. 320 StringRef parseStringToEndOfStatement() override; 321 322 /// \brief Parse until the end of a statement or a comma is encountered, 323 /// return the contents from the current token up to the end or comma. 324 StringRef parseStringToComma(); 325 326 bool parseAssignment(StringRef Name, bool allow_redef, 327 bool NoDeadStrip = false); 328 329 bool parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, SMLoc &EndLoc); 330 bool parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc); 331 bool parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc); 332 333 bool parseRegisterOrRegisterNumber(int64_t &Register, SMLoc DirectiveLoc); 334 335 // Generic (target and platform independent) directive parsing. 336 enum DirectiveKind { 337 DK_NO_DIRECTIVE, // Placeholder 338 DK_SET, DK_EQU, DK_EQUIV, DK_ASCII, DK_ASCIZ, DK_STRING, DK_BYTE, DK_SHORT, 339 DK_VALUE, DK_2BYTE, DK_LONG, DK_INT, DK_4BYTE, DK_QUAD, DK_8BYTE, DK_OCTA, 340 DK_SINGLE, DK_FLOAT, DK_DOUBLE, DK_ALIGN, DK_ALIGN32, DK_BALIGN, DK_BALIGNW, 341 DK_BALIGNL, DK_P2ALIGN, DK_P2ALIGNW, DK_P2ALIGNL, DK_ORG, DK_FILL, DK_ENDR, 342 DK_BUNDLE_ALIGN_MODE, DK_BUNDLE_LOCK, DK_BUNDLE_UNLOCK, 343 DK_ZERO, DK_EXTERN, DK_GLOBL, DK_GLOBAL, 344 DK_LAZY_REFERENCE, DK_NO_DEAD_STRIP, DK_SYMBOL_RESOLVER, DK_PRIVATE_EXTERN, 345 DK_REFERENCE, DK_WEAK_DEFINITION, DK_WEAK_REFERENCE, 346 DK_WEAK_DEF_CAN_BE_HIDDEN, DK_COMM, DK_COMMON, DK_LCOMM, DK_ABORT, 347 DK_INCLUDE, DK_INCBIN, DK_CODE16, DK_CODE16GCC, DK_REPT, DK_IRP, DK_IRPC, 348 DK_IF, DK_IFEQ, DK_IFGE, DK_IFGT, DK_IFLE, DK_IFLT, DK_IFNE, DK_IFB, 349 DK_IFNB, DK_IFC, DK_IFEQS, DK_IFNC, DK_IFDEF, DK_IFNDEF, DK_IFNOTDEF, 350 DK_ELSEIF, DK_ELSE, DK_ENDIF, 351 DK_SPACE, DK_SKIP, DK_FILE, DK_LINE, DK_LOC, DK_STABS, 352 DK_CFI_SECTIONS, DK_CFI_STARTPROC, DK_CFI_ENDPROC, DK_CFI_DEF_CFA, 353 DK_CFI_DEF_CFA_OFFSET, DK_CFI_ADJUST_CFA_OFFSET, DK_CFI_DEF_CFA_REGISTER, 354 DK_CFI_OFFSET, DK_CFI_REL_OFFSET, DK_CFI_PERSONALITY, DK_CFI_LSDA, 355 DK_CFI_REMEMBER_STATE, DK_CFI_RESTORE_STATE, DK_CFI_SAME_VALUE, 356 DK_CFI_RESTORE, DK_CFI_ESCAPE, DK_CFI_SIGNAL_FRAME, DK_CFI_UNDEFINED, 357 DK_CFI_REGISTER, DK_CFI_WINDOW_SAVE, 358 DK_MACROS_ON, DK_MACROS_OFF, DK_MACRO, DK_ENDM, DK_ENDMACRO, DK_PURGEM, 359 DK_SLEB128, DK_ULEB128, 360 DK_ERR, DK_ERROR, 361 DK_END 362 }; 363 364 /// \brief Maps directive name --> DirectiveKind enum, for 365 /// directives parsed by this class. 366 StringMap<DirectiveKind> DirectiveKindMap; 367 368 // ".ascii", ".asciz", ".string" 369 bool parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated); 370 bool parseDirectiveValue(unsigned Size); // ".byte", ".long", ... 371 bool parseDirectiveOctaValue(); // ".octa" 372 bool parseDirectiveRealValue(const fltSemantics &); // ".single", ... 373 bool parseDirectiveFill(); // ".fill" 374 bool parseDirectiveZero(); // ".zero" 375 // ".set", ".equ", ".equiv" 376 bool parseDirectiveSet(StringRef IDVal, bool allow_redef); 377 bool parseDirectiveOrg(); // ".org" 378 // ".align{,32}", ".p2align{,w,l}" 379 bool parseDirectiveAlign(bool IsPow2, unsigned ValueSize); 380 381 // ".file", ".line", ".loc", ".stabs" 382 bool parseDirectiveFile(SMLoc DirectiveLoc); 383 bool parseDirectiveLine(); 384 bool parseDirectiveLoc(); 385 bool parseDirectiveStabs(); 386 387 // .cfi directives 388 bool parseDirectiveCFIRegister(SMLoc DirectiveLoc); 389 bool parseDirectiveCFIWindowSave(); 390 bool parseDirectiveCFISections(); 391 bool parseDirectiveCFIStartProc(); 392 bool parseDirectiveCFIEndProc(); 393 bool parseDirectiveCFIDefCfaOffset(); 394 bool parseDirectiveCFIDefCfa(SMLoc DirectiveLoc); 395 bool parseDirectiveCFIAdjustCfaOffset(); 396 bool parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc); 397 bool parseDirectiveCFIOffset(SMLoc DirectiveLoc); 398 bool parseDirectiveCFIRelOffset(SMLoc DirectiveLoc); 399 bool parseDirectiveCFIPersonalityOrLsda(bool IsPersonality); 400 bool parseDirectiveCFIRememberState(); 401 bool parseDirectiveCFIRestoreState(); 402 bool parseDirectiveCFISameValue(SMLoc DirectiveLoc); 403 bool parseDirectiveCFIRestore(SMLoc DirectiveLoc); 404 bool parseDirectiveCFIEscape(); 405 bool parseDirectiveCFISignalFrame(); 406 bool parseDirectiveCFIUndefined(SMLoc DirectiveLoc); 407 408 // macro directives 409 bool parseDirectivePurgeMacro(SMLoc DirectiveLoc); 410 bool parseDirectiveEndMacro(StringRef Directive); 411 bool parseDirectiveMacro(SMLoc DirectiveLoc); 412 bool parseDirectiveMacrosOnOff(StringRef Directive); 413 414 // ".bundle_align_mode" 415 bool parseDirectiveBundleAlignMode(); 416 // ".bundle_lock" 417 bool parseDirectiveBundleLock(); 418 // ".bundle_unlock" 419 bool parseDirectiveBundleUnlock(); 420 421 // ".space", ".skip" 422 bool parseDirectiveSpace(StringRef IDVal); 423 424 // .sleb128 (Signed=true) and .uleb128 (Signed=false) 425 bool parseDirectiveLEB128(bool Signed); 426 427 /// \brief Parse a directive like ".globl" which 428 /// accepts a single symbol (which should be a label or an external). 429 bool parseDirectiveSymbolAttribute(MCSymbolAttr Attr); 430 431 bool parseDirectiveComm(bool IsLocal); // ".comm" and ".lcomm" 432 433 bool parseDirectiveAbort(); // ".abort" 434 bool parseDirectiveInclude(); // ".include" 435 bool parseDirectiveIncbin(); // ".incbin" 436 437 // ".if", ".ifeq", ".ifge", ".ifgt" , ".ifle", ".iflt" or ".ifne" 438 bool parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind); 439 // ".ifb" or ".ifnb", depending on ExpectBlank. 440 bool parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank); 441 // ".ifc" or ".ifnc", depending on ExpectEqual. 442 bool parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual); 443 // ".ifeqs" 444 bool parseDirectiveIfeqs(SMLoc DirectiveLoc); 445 // ".ifdef" or ".ifndef", depending on expect_defined 446 bool parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined); 447 bool parseDirectiveElseIf(SMLoc DirectiveLoc); // ".elseif" 448 bool parseDirectiveElse(SMLoc DirectiveLoc); // ".else" 449 bool parseDirectiveEndIf(SMLoc DirectiveLoc); // .endif 450 bool parseEscapedString(std::string &Data) override; 451 452 const MCExpr *applyModifierToExpr(const MCExpr *E, 453 MCSymbolRefExpr::VariantKind Variant); 454 455 // Macro-like directives 456 MCAsmMacro *parseMacroLikeBody(SMLoc DirectiveLoc); 457 void instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 458 raw_svector_ostream &OS); 459 bool parseDirectiveRept(SMLoc DirectiveLoc, StringRef Directive); 460 bool parseDirectiveIrp(SMLoc DirectiveLoc); // ".irp" 461 bool parseDirectiveIrpc(SMLoc DirectiveLoc); // ".irpc" 462 bool parseDirectiveEndr(SMLoc DirectiveLoc); // ".endr" 463 464 // "_emit" or "__emit" 465 bool parseDirectiveMSEmit(SMLoc DirectiveLoc, ParseStatementInfo &Info, 466 size_t Len); 467 468 // "align" 469 bool parseDirectiveMSAlign(SMLoc DirectiveLoc, ParseStatementInfo &Info); 470 471 // "end" 472 bool parseDirectiveEnd(SMLoc DirectiveLoc); 473 474 // ".err" or ".error" 475 bool parseDirectiveError(SMLoc DirectiveLoc, bool WithMessage); 476 477 void initializeDirectiveKindMap(); 478 }; 479 } 480 481 namespace llvm { 482 483 extern MCAsmParserExtension *createDarwinAsmParser(); 484 extern MCAsmParserExtension *createELFAsmParser(); 485 extern MCAsmParserExtension *createCOFFAsmParser(); 486 487 } 488 489 enum { DEFAULT_ADDRSPACE = 0 }; 490 491 AsmParser::AsmParser(SourceMgr &_SM, MCContext &_Ctx, MCStreamer &_Out, 492 const MCAsmInfo &_MAI) 493 : Lexer(_MAI), Ctx(_Ctx), Out(_Out), MAI(_MAI), SrcMgr(_SM), 494 PlatformParser(nullptr), CurBuffer(_SM.getMainFileID()), 495 MacrosEnabledFlag(true), HadError(false), CppHashLineNumber(0), 496 AssemblerDialect(~0U), IsDarwin(false), ParsingInlineAsm(false) { 497 // Save the old handler. 498 SavedDiagHandler = SrcMgr.getDiagHandler(); 499 SavedDiagContext = SrcMgr.getDiagContext(); 500 // Set our own handler which calls the saved handler. 501 SrcMgr.setDiagHandler(DiagHandler, this); 502 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 503 504 // Initialize the platform / file format parser. 505 switch (_Ctx.getObjectFileInfo()->getObjectFileType()) { 506 case MCObjectFileInfo::IsCOFF: 507 PlatformParser = createCOFFAsmParser(); 508 PlatformParser->Initialize(*this); 509 break; 510 case MCObjectFileInfo::IsMachO: 511 PlatformParser = createDarwinAsmParser(); 512 PlatformParser->Initialize(*this); 513 IsDarwin = true; 514 break; 515 case MCObjectFileInfo::IsELF: 516 PlatformParser = createELFAsmParser(); 517 PlatformParser->Initialize(*this); 518 break; 519 } 520 521 initializeDirectiveKindMap(); 522 } 523 524 AsmParser::~AsmParser() { 525 assert((HadError || ActiveMacros.empty()) && 526 "Unexpected active macro instantiation!"); 527 528 // Destroy any macros. 529 for (StringMap<MCAsmMacro *>::iterator it = MacroMap.begin(), 530 ie = MacroMap.end(); 531 it != ie; ++it) 532 delete it->getValue(); 533 534 delete PlatformParser; 535 } 536 537 void AsmParser::printMacroInstantiations() { 538 // Print the active macro instantiation stack. 539 for (std::vector<MacroInstantiation *>::const_reverse_iterator 540 it = ActiveMacros.rbegin(), 541 ie = ActiveMacros.rend(); 542 it != ie; ++it) 543 printMessage((*it)->InstantiationLoc, SourceMgr::DK_Note, 544 "while in macro instantiation"); 545 } 546 547 void AsmParser::Note(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) { 548 printMessage(L, SourceMgr::DK_Note, Msg, Ranges); 549 printMacroInstantiations(); 550 } 551 552 bool AsmParser::Warning(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) { 553 if (FatalAssemblerWarnings) 554 return Error(L, Msg, Ranges); 555 printMessage(L, SourceMgr::DK_Warning, Msg, Ranges); 556 printMacroInstantiations(); 557 return false; 558 } 559 560 bool AsmParser::Error(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) { 561 HadError = true; 562 printMessage(L, SourceMgr::DK_Error, Msg, Ranges); 563 printMacroInstantiations(); 564 return true; 565 } 566 567 bool AsmParser::enterIncludeFile(const std::string &Filename) { 568 std::string IncludedFile; 569 unsigned NewBuf = 570 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 571 if (!NewBuf) 572 return true; 573 574 CurBuffer = NewBuf; 575 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 576 return false; 577 } 578 579 /// Process the specified .incbin file by searching for it in the include paths 580 /// then just emitting the byte contents of the file to the streamer. This 581 /// returns true on failure. 582 bool AsmParser::processIncbinFile(const std::string &Filename) { 583 std::string IncludedFile; 584 unsigned NewBuf = 585 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 586 if (!NewBuf) 587 return true; 588 589 // Pick up the bytes from the file and emit them. 590 getStreamer().EmitBytes(SrcMgr.getMemoryBuffer(NewBuf)->getBuffer()); 591 return false; 592 } 593 594 void AsmParser::jumpToLoc(SMLoc Loc, unsigned InBuffer) { 595 CurBuffer = InBuffer ? InBuffer : SrcMgr.FindBufferContainingLoc(Loc); 596 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer(), 597 Loc.getPointer()); 598 } 599 600 const AsmToken &AsmParser::Lex() { 601 const AsmToken *tok = &Lexer.Lex(); 602 603 if (tok->is(AsmToken::Eof)) { 604 // If this is the end of an included file, pop the parent file off the 605 // include stack. 606 SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer); 607 if (ParentIncludeLoc != SMLoc()) { 608 jumpToLoc(ParentIncludeLoc); 609 tok = &Lexer.Lex(); 610 } 611 } 612 613 if (tok->is(AsmToken::Error)) 614 Error(Lexer.getErrLoc(), Lexer.getErr()); 615 616 return *tok; 617 } 618 619 bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) { 620 // Create the initial section, if requested. 621 if (!NoInitialTextSection) 622 Out.InitSections(); 623 624 // Prime the lexer. 625 Lex(); 626 627 HadError = false; 628 AsmCond StartingCondState = TheCondState; 629 630 // If we are generating dwarf for assembly source files save the initial text 631 // section and generate a .file directive. 632 if (getContext().getGenDwarfForAssembly()) { 633 MCSymbol *SectionStartSym = getContext().CreateTempSymbol(); 634 getStreamer().EmitLabel(SectionStartSym); 635 auto InsertResult = getContext().addGenDwarfSection( 636 getStreamer().getCurrentSection().first); 637 assert(InsertResult.second && ".text section should not have debug info yet"); 638 InsertResult.first->second.first = SectionStartSym; 639 getContext().setGenDwarfFileNumber(getStreamer().EmitDwarfFileDirective( 640 0, StringRef(), getContext().getMainFileName())); 641 } 642 643 // While we have input, parse each statement. 644 while (Lexer.isNot(AsmToken::Eof)) { 645 ParseStatementInfo Info; 646 if (!parseStatement(Info)) 647 continue; 648 649 // We had an error, validate that one was emitted and recover by skipping to 650 // the next line. 651 assert(HadError && "Parse statement returned an error, but none emitted!"); 652 eatToEndOfStatement(); 653 } 654 655 if (TheCondState.TheCond != StartingCondState.TheCond || 656 TheCondState.Ignore != StartingCondState.Ignore) 657 return TokError("unmatched .ifs or .elses"); 658 659 // Check to see there are no empty DwarfFile slots. 660 const auto &LineTables = getContext().getMCDwarfLineTables(); 661 if (!LineTables.empty()) { 662 unsigned Index = 0; 663 for (const auto &File : LineTables.begin()->second.getMCDwarfFiles()) { 664 if (File.Name.empty() && Index != 0) 665 TokError("unassigned file number: " + Twine(Index) + 666 " for .file directives"); 667 ++Index; 668 } 669 } 670 671 // Check to see that all assembler local symbols were actually defined. 672 // Targets that don't do subsections via symbols may not want this, though, 673 // so conservatively exclude them. Only do this if we're finalizing, though, 674 // as otherwise we won't necessarilly have seen everything yet. 675 if (!NoFinalize && MAI.hasSubsectionsViaSymbols()) { 676 const MCContext::SymbolTable &Symbols = getContext().getSymbols(); 677 for (MCContext::SymbolTable::const_iterator i = Symbols.begin(), 678 e = Symbols.end(); 679 i != e; ++i) { 680 MCSymbol *Sym = i->getValue(); 681 // Variable symbols may not be marked as defined, so check those 682 // explicitly. If we know it's a variable, we have a definition for 683 // the purposes of this check. 684 if (Sym->isTemporary() && !Sym->isVariable() && !Sym->isDefined()) 685 // FIXME: We would really like to refer back to where the symbol was 686 // first referenced for a source location. We need to add something 687 // to track that. Currently, we just point to the end of the file. 688 printMessage( 689 getLexer().getLoc(), SourceMgr::DK_Error, 690 "assembler local symbol '" + Sym->getName() + "' not defined"); 691 } 692 } 693 694 // Finalize the output stream if there are no errors and if the client wants 695 // us to. 696 if (!HadError && !NoFinalize) 697 Out.Finish(); 698 699 return HadError; 700 } 701 702 void AsmParser::checkForValidSection() { 703 if (!ParsingInlineAsm && !getStreamer().getCurrentSection().first) { 704 TokError("expected section directive before assembly directive"); 705 Out.InitSections(); 706 } 707 } 708 709 /// \brief Throw away the rest of the line for testing purposes. 710 void AsmParser::eatToEndOfStatement() { 711 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof)) 712 Lex(); 713 714 // Eat EOL. 715 if (Lexer.is(AsmToken::EndOfStatement)) 716 Lex(); 717 } 718 719 StringRef AsmParser::parseStringToEndOfStatement() { 720 const char *Start = getTok().getLoc().getPointer(); 721 722 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof)) 723 Lex(); 724 725 const char *End = getTok().getLoc().getPointer(); 726 return StringRef(Start, End - Start); 727 } 728 729 StringRef AsmParser::parseStringToComma() { 730 const char *Start = getTok().getLoc().getPointer(); 731 732 while (Lexer.isNot(AsmToken::EndOfStatement) && 733 Lexer.isNot(AsmToken::Comma) && Lexer.isNot(AsmToken::Eof)) 734 Lex(); 735 736 const char *End = getTok().getLoc().getPointer(); 737 return StringRef(Start, End - Start); 738 } 739 740 /// \brief Parse a paren expression and return it. 741 /// NOTE: This assumes the leading '(' has already been consumed. 742 /// 743 /// parenexpr ::= expr) 744 /// 745 bool AsmParser::parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc) { 746 if (parseExpression(Res)) 747 return true; 748 if (Lexer.isNot(AsmToken::RParen)) 749 return TokError("expected ')' in parentheses expression"); 750 EndLoc = Lexer.getTok().getEndLoc(); 751 Lex(); 752 return false; 753 } 754 755 /// \brief Parse a bracket expression and return it. 756 /// NOTE: This assumes the leading '[' has already been consumed. 757 /// 758 /// bracketexpr ::= expr] 759 /// 760 bool AsmParser::parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc) { 761 if (parseExpression(Res)) 762 return true; 763 if (Lexer.isNot(AsmToken::RBrac)) 764 return TokError("expected ']' in brackets expression"); 765 EndLoc = Lexer.getTok().getEndLoc(); 766 Lex(); 767 return false; 768 } 769 770 /// \brief Parse a primary expression and return it. 771 /// primaryexpr ::= (parenexpr 772 /// primaryexpr ::= symbol 773 /// primaryexpr ::= number 774 /// primaryexpr ::= '.' 775 /// primaryexpr ::= ~,+,- primaryexpr 776 bool AsmParser::parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) { 777 SMLoc FirstTokenLoc = getLexer().getLoc(); 778 AsmToken::TokenKind FirstTokenKind = Lexer.getKind(); 779 switch (FirstTokenKind) { 780 default: 781 return TokError("unknown token in expression"); 782 // If we have an error assume that we've already handled it. 783 case AsmToken::Error: 784 return true; 785 case AsmToken::Exclaim: 786 Lex(); // Eat the operator. 787 if (parsePrimaryExpr(Res, EndLoc)) 788 return true; 789 Res = MCUnaryExpr::CreateLNot(Res, getContext()); 790 return false; 791 case AsmToken::Dollar: 792 case AsmToken::At: 793 case AsmToken::String: 794 case AsmToken::Identifier: { 795 StringRef Identifier; 796 if (parseIdentifier(Identifier)) { 797 if (FirstTokenKind == AsmToken::Dollar) { 798 if (Lexer.getMAI().getDollarIsPC()) { 799 // This is a '$' reference, which references the current PC. Emit a 800 // temporary label to the streamer and refer to it. 801 MCSymbol *Sym = Ctx.CreateTempSymbol(); 802 Out.EmitLabel(Sym); 803 Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, 804 getContext()); 805 EndLoc = FirstTokenLoc; 806 return false; 807 } 808 return Error(FirstTokenLoc, "invalid token in expression"); 809 } 810 } 811 // Parse symbol variant 812 std::pair<StringRef, StringRef> Split; 813 if (!MAI.useParensForSymbolVariant()) { 814 if (FirstTokenKind == AsmToken::String) { 815 if (Lexer.is(AsmToken::At)) { 816 Lexer.Lex(); // eat @ 817 SMLoc AtLoc = getLexer().getLoc(); 818 StringRef VName; 819 if (parseIdentifier(VName)) 820 return Error(AtLoc, "expected symbol variant after '@'"); 821 822 Split = std::make_pair(Identifier, VName); 823 } 824 } else { 825 Split = Identifier.split('@'); 826 } 827 } else if (Lexer.is(AsmToken::LParen)) { 828 Lexer.Lex(); // eat ( 829 StringRef VName; 830 parseIdentifier(VName); 831 if (Lexer.isNot(AsmToken::RParen)) { 832 return Error(Lexer.getTok().getLoc(), 833 "unexpected token in variant, expected ')'"); 834 } 835 Lexer.Lex(); // eat ) 836 Split = std::make_pair(Identifier, VName); 837 } 838 839 EndLoc = SMLoc::getFromPointer(Identifier.end()); 840 841 // This is a symbol reference. 842 StringRef SymbolName = Identifier; 843 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 844 845 // Lookup the symbol variant if used. 846 if (Split.second.size()) { 847 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second); 848 if (Variant != MCSymbolRefExpr::VK_Invalid) { 849 SymbolName = Split.first; 850 } else if (MAI.doesAllowAtInName() && !MAI.useParensForSymbolVariant()) { 851 Variant = MCSymbolRefExpr::VK_None; 852 } else { 853 return Error(SMLoc::getFromPointer(Split.second.begin()), 854 "invalid variant '" + Split.second + "'"); 855 } 856 } 857 858 MCSymbol *Sym = getContext().GetOrCreateSymbol(SymbolName); 859 860 // If this is an absolute variable reference, substitute it now to preserve 861 // semantics in the face of reassignment. 862 if (Sym->isVariable() && isa<MCConstantExpr>(Sym->getVariableValue())) { 863 if (Variant) 864 return Error(EndLoc, "unexpected modifier on variable reference"); 865 866 Res = Sym->getVariableValue(); 867 return false; 868 } 869 870 // Otherwise create a symbol ref. 871 Res = MCSymbolRefExpr::Create(Sym, Variant, getContext()); 872 return false; 873 } 874 case AsmToken::BigNum: 875 return TokError("literal value out of range for directive"); 876 case AsmToken::Integer: { 877 SMLoc Loc = getTok().getLoc(); 878 int64_t IntVal = getTok().getIntVal(); 879 Res = MCConstantExpr::Create(IntVal, getContext()); 880 EndLoc = Lexer.getTok().getEndLoc(); 881 Lex(); // Eat token. 882 // Look for 'b' or 'f' following an Integer as a directional label 883 if (Lexer.getKind() == AsmToken::Identifier) { 884 StringRef IDVal = getTok().getString(); 885 // Lookup the symbol variant if used. 886 std::pair<StringRef, StringRef> Split = IDVal.split('@'); 887 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 888 if (Split.first.size() != IDVal.size()) { 889 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second); 890 if (Variant == MCSymbolRefExpr::VK_Invalid) 891 return TokError("invalid variant '" + Split.second + "'"); 892 IDVal = Split.first; 893 } 894 if (IDVal == "f" || IDVal == "b") { 895 MCSymbol *Sym = 896 Ctx.GetDirectionalLocalSymbol(IntVal, IDVal == "b"); 897 Res = MCSymbolRefExpr::Create(Sym, Variant, getContext()); 898 if (IDVal == "b" && Sym->isUndefined()) 899 return Error(Loc, "invalid reference to undefined symbol"); 900 EndLoc = Lexer.getTok().getEndLoc(); 901 Lex(); // Eat identifier. 902 } 903 } 904 return false; 905 } 906 case AsmToken::Real: { 907 APFloat RealVal(APFloat::IEEEdouble, getTok().getString()); 908 uint64_t IntVal = RealVal.bitcastToAPInt().getZExtValue(); 909 Res = MCConstantExpr::Create(IntVal, getContext()); 910 EndLoc = Lexer.getTok().getEndLoc(); 911 Lex(); // Eat token. 912 return false; 913 } 914 case AsmToken::Dot: { 915 // This is a '.' reference, which references the current PC. Emit a 916 // temporary label to the streamer and refer to it. 917 MCSymbol *Sym = Ctx.CreateTempSymbol(); 918 Out.EmitLabel(Sym); 919 Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, getContext()); 920 EndLoc = Lexer.getTok().getEndLoc(); 921 Lex(); // Eat identifier. 922 return false; 923 } 924 case AsmToken::LParen: 925 Lex(); // Eat the '('. 926 return parseParenExpr(Res, EndLoc); 927 case AsmToken::LBrac: 928 if (!PlatformParser->HasBracketExpressions()) 929 return TokError("brackets expression not supported on this target"); 930 Lex(); // Eat the '['. 931 return parseBracketExpr(Res, EndLoc); 932 case AsmToken::Minus: 933 Lex(); // Eat the operator. 934 if (parsePrimaryExpr(Res, EndLoc)) 935 return true; 936 Res = MCUnaryExpr::CreateMinus(Res, getContext()); 937 return false; 938 case AsmToken::Plus: 939 Lex(); // Eat the operator. 940 if (parsePrimaryExpr(Res, EndLoc)) 941 return true; 942 Res = MCUnaryExpr::CreatePlus(Res, getContext()); 943 return false; 944 case AsmToken::Tilde: 945 Lex(); // Eat the operator. 946 if (parsePrimaryExpr(Res, EndLoc)) 947 return true; 948 Res = MCUnaryExpr::CreateNot(Res, getContext()); 949 return false; 950 } 951 } 952 953 bool AsmParser::parseExpression(const MCExpr *&Res) { 954 SMLoc EndLoc; 955 return parseExpression(Res, EndLoc); 956 } 957 958 const MCExpr * 959 AsmParser::applyModifierToExpr(const MCExpr *E, 960 MCSymbolRefExpr::VariantKind Variant) { 961 // Ask the target implementation about this expression first. 962 const MCExpr *NewE = getTargetParser().applyModifierToExpr(E, Variant, Ctx); 963 if (NewE) 964 return NewE; 965 // Recurse over the given expression, rebuilding it to apply the given variant 966 // if there is exactly one symbol. 967 switch (E->getKind()) { 968 case MCExpr::Target: 969 case MCExpr::Constant: 970 return nullptr; 971 972 case MCExpr::SymbolRef: { 973 const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(E); 974 975 if (SRE->getKind() != MCSymbolRefExpr::VK_None) { 976 TokError("invalid variant on expression '" + getTok().getIdentifier() + 977 "' (already modified)"); 978 return E; 979 } 980 981 return MCSymbolRefExpr::Create(&SRE->getSymbol(), Variant, getContext()); 982 } 983 984 case MCExpr::Unary: { 985 const MCUnaryExpr *UE = cast<MCUnaryExpr>(E); 986 const MCExpr *Sub = applyModifierToExpr(UE->getSubExpr(), Variant); 987 if (!Sub) 988 return nullptr; 989 return MCUnaryExpr::Create(UE->getOpcode(), Sub, getContext()); 990 } 991 992 case MCExpr::Binary: { 993 const MCBinaryExpr *BE = cast<MCBinaryExpr>(E); 994 const MCExpr *LHS = applyModifierToExpr(BE->getLHS(), Variant); 995 const MCExpr *RHS = applyModifierToExpr(BE->getRHS(), Variant); 996 997 if (!LHS && !RHS) 998 return nullptr; 999 1000 if (!LHS) 1001 LHS = BE->getLHS(); 1002 if (!RHS) 1003 RHS = BE->getRHS(); 1004 1005 return MCBinaryExpr::Create(BE->getOpcode(), LHS, RHS, getContext()); 1006 } 1007 } 1008 1009 llvm_unreachable("Invalid expression kind!"); 1010 } 1011 1012 /// \brief Parse an expression and return it. 1013 /// 1014 /// expr ::= expr &&,|| expr -> lowest. 1015 /// expr ::= expr |,^,&,! expr 1016 /// expr ::= expr ==,!=,<>,<,<=,>,>= expr 1017 /// expr ::= expr <<,>> expr 1018 /// expr ::= expr +,- expr 1019 /// expr ::= expr *,/,% expr -> highest. 1020 /// expr ::= primaryexpr 1021 /// 1022 bool AsmParser::parseExpression(const MCExpr *&Res, SMLoc &EndLoc) { 1023 // Parse the expression. 1024 Res = nullptr; 1025 if (parsePrimaryExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc)) 1026 return true; 1027 1028 // As a special case, we support 'a op b @ modifier' by rewriting the 1029 // expression to include the modifier. This is inefficient, but in general we 1030 // expect users to use 'a@modifier op b'. 1031 if (Lexer.getKind() == AsmToken::At) { 1032 Lex(); 1033 1034 if (Lexer.isNot(AsmToken::Identifier)) 1035 return TokError("unexpected symbol modifier following '@'"); 1036 1037 MCSymbolRefExpr::VariantKind Variant = 1038 MCSymbolRefExpr::getVariantKindForName(getTok().getIdentifier()); 1039 if (Variant == MCSymbolRefExpr::VK_Invalid) 1040 return TokError("invalid variant '" + getTok().getIdentifier() + "'"); 1041 1042 const MCExpr *ModifiedRes = applyModifierToExpr(Res, Variant); 1043 if (!ModifiedRes) { 1044 return TokError("invalid modifier '" + getTok().getIdentifier() + 1045 "' (no symbols present)"); 1046 } 1047 1048 Res = ModifiedRes; 1049 Lex(); 1050 } 1051 1052 // Try to constant fold it up front, if possible. 1053 int64_t Value; 1054 if (Res->EvaluateAsAbsolute(Value)) 1055 Res = MCConstantExpr::Create(Value, getContext()); 1056 1057 return false; 1058 } 1059 1060 bool AsmParser::parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) { 1061 Res = nullptr; 1062 return parseParenExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc); 1063 } 1064 1065 bool AsmParser::parseAbsoluteExpression(int64_t &Res) { 1066 const MCExpr *Expr; 1067 1068 SMLoc StartLoc = Lexer.getLoc(); 1069 if (parseExpression(Expr)) 1070 return true; 1071 1072 if (!Expr->EvaluateAsAbsolute(Res)) 1073 return Error(StartLoc, "expected absolute expression"); 1074 1075 return false; 1076 } 1077 1078 static unsigned getBinOpPrecedence(AsmToken::TokenKind K, 1079 MCBinaryExpr::Opcode &Kind) { 1080 switch (K) { 1081 default: 1082 return 0; // not a binop. 1083 1084 // Lowest Precedence: &&, || 1085 case AsmToken::AmpAmp: 1086 Kind = MCBinaryExpr::LAnd; 1087 return 1; 1088 case AsmToken::PipePipe: 1089 Kind = MCBinaryExpr::LOr; 1090 return 1; 1091 1092 // Low Precedence: |, &, ^ 1093 // 1094 // FIXME: gas seems to support '!' as an infix operator? 1095 case AsmToken::Pipe: 1096 Kind = MCBinaryExpr::Or; 1097 return 2; 1098 case AsmToken::Caret: 1099 Kind = MCBinaryExpr::Xor; 1100 return 2; 1101 case AsmToken::Amp: 1102 Kind = MCBinaryExpr::And; 1103 return 2; 1104 1105 // Low Intermediate Precedence: ==, !=, <>, <, <=, >, >= 1106 case AsmToken::EqualEqual: 1107 Kind = MCBinaryExpr::EQ; 1108 return 3; 1109 case AsmToken::ExclaimEqual: 1110 case AsmToken::LessGreater: 1111 Kind = MCBinaryExpr::NE; 1112 return 3; 1113 case AsmToken::Less: 1114 Kind = MCBinaryExpr::LT; 1115 return 3; 1116 case AsmToken::LessEqual: 1117 Kind = MCBinaryExpr::LTE; 1118 return 3; 1119 case AsmToken::Greater: 1120 Kind = MCBinaryExpr::GT; 1121 return 3; 1122 case AsmToken::GreaterEqual: 1123 Kind = MCBinaryExpr::GTE; 1124 return 3; 1125 1126 // Intermediate Precedence: <<, >> 1127 case AsmToken::LessLess: 1128 Kind = MCBinaryExpr::Shl; 1129 return 4; 1130 case AsmToken::GreaterGreater: 1131 Kind = MCBinaryExpr::Shr; 1132 return 4; 1133 1134 // High Intermediate Precedence: +, - 1135 case AsmToken::Plus: 1136 Kind = MCBinaryExpr::Add; 1137 return 5; 1138 case AsmToken::Minus: 1139 Kind = MCBinaryExpr::Sub; 1140 return 5; 1141 1142 // Highest Precedence: *, /, % 1143 case AsmToken::Star: 1144 Kind = MCBinaryExpr::Mul; 1145 return 6; 1146 case AsmToken::Slash: 1147 Kind = MCBinaryExpr::Div; 1148 return 6; 1149 case AsmToken::Percent: 1150 Kind = MCBinaryExpr::Mod; 1151 return 6; 1152 } 1153 } 1154 1155 /// \brief Parse all binary operators with precedence >= 'Precedence'. 1156 /// Res contains the LHS of the expression on input. 1157 bool AsmParser::parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, 1158 SMLoc &EndLoc) { 1159 while (1) { 1160 MCBinaryExpr::Opcode Kind = MCBinaryExpr::Add; 1161 unsigned TokPrec = getBinOpPrecedence(Lexer.getKind(), Kind); 1162 1163 // If the next token is lower precedence than we are allowed to eat, return 1164 // successfully with what we ate already. 1165 if (TokPrec < Precedence) 1166 return false; 1167 1168 Lex(); 1169 1170 // Eat the next primary expression. 1171 const MCExpr *RHS; 1172 if (parsePrimaryExpr(RHS, EndLoc)) 1173 return true; 1174 1175 // If BinOp binds less tightly with RHS than the operator after RHS, let 1176 // the pending operator take RHS as its LHS. 1177 MCBinaryExpr::Opcode Dummy; 1178 unsigned NextTokPrec = getBinOpPrecedence(Lexer.getKind(), Dummy); 1179 if (TokPrec < NextTokPrec && parseBinOpRHS(TokPrec + 1, RHS, EndLoc)) 1180 return true; 1181 1182 // Merge LHS and RHS according to operator. 1183 Res = MCBinaryExpr::Create(Kind, Res, RHS, getContext()); 1184 } 1185 } 1186 1187 /// ParseStatement: 1188 /// ::= EndOfStatement 1189 /// ::= Label* Directive ...Operands... EndOfStatement 1190 /// ::= Label* Identifier OperandList* EndOfStatement 1191 bool AsmParser::parseStatement(ParseStatementInfo &Info) { 1192 if (Lexer.is(AsmToken::EndOfStatement)) { 1193 Out.AddBlankLine(); 1194 Lex(); 1195 return false; 1196 } 1197 1198 // Statements always start with an identifier or are a full line comment. 1199 AsmToken ID = getTok(); 1200 SMLoc IDLoc = ID.getLoc(); 1201 StringRef IDVal; 1202 int64_t LocalLabelVal = -1; 1203 // A full line comment is a '#' as the first token. 1204 if (Lexer.is(AsmToken::Hash)) 1205 return parseCppHashLineFilenameComment(IDLoc); 1206 1207 // Allow an integer followed by a ':' as a directional local label. 1208 if (Lexer.is(AsmToken::Integer)) { 1209 LocalLabelVal = getTok().getIntVal(); 1210 if (LocalLabelVal < 0) { 1211 if (!TheCondState.Ignore) 1212 return TokError("unexpected token at start of statement"); 1213 IDVal = ""; 1214 } else { 1215 IDVal = getTok().getString(); 1216 Lex(); // Consume the integer token to be used as an identifier token. 1217 if (Lexer.getKind() != AsmToken::Colon) { 1218 if (!TheCondState.Ignore) 1219 return TokError("unexpected token at start of statement"); 1220 } 1221 } 1222 } else if (Lexer.is(AsmToken::Dot)) { 1223 // Treat '.' as a valid identifier in this context. 1224 Lex(); 1225 IDVal = "."; 1226 } else if (parseIdentifier(IDVal)) { 1227 if (!TheCondState.Ignore) 1228 return TokError("unexpected token at start of statement"); 1229 IDVal = ""; 1230 } 1231 1232 // Handle conditional assembly here before checking for skipping. We 1233 // have to do this so that .endif isn't skipped in a ".if 0" block for 1234 // example. 1235 StringMap<DirectiveKind>::const_iterator DirKindIt = 1236 DirectiveKindMap.find(IDVal); 1237 DirectiveKind DirKind = (DirKindIt == DirectiveKindMap.end()) 1238 ? DK_NO_DIRECTIVE 1239 : DirKindIt->getValue(); 1240 switch (DirKind) { 1241 default: 1242 break; 1243 case DK_IF: 1244 case DK_IFEQ: 1245 case DK_IFGE: 1246 case DK_IFGT: 1247 case DK_IFLE: 1248 case DK_IFLT: 1249 case DK_IFNE: 1250 return parseDirectiveIf(IDLoc, DirKind); 1251 case DK_IFB: 1252 return parseDirectiveIfb(IDLoc, true); 1253 case DK_IFNB: 1254 return parseDirectiveIfb(IDLoc, false); 1255 case DK_IFC: 1256 return parseDirectiveIfc(IDLoc, true); 1257 case DK_IFEQS: 1258 return parseDirectiveIfeqs(IDLoc); 1259 case DK_IFNC: 1260 return parseDirectiveIfc(IDLoc, false); 1261 case DK_IFDEF: 1262 return parseDirectiveIfdef(IDLoc, true); 1263 case DK_IFNDEF: 1264 case DK_IFNOTDEF: 1265 return parseDirectiveIfdef(IDLoc, false); 1266 case DK_ELSEIF: 1267 return parseDirectiveElseIf(IDLoc); 1268 case DK_ELSE: 1269 return parseDirectiveElse(IDLoc); 1270 case DK_ENDIF: 1271 return parseDirectiveEndIf(IDLoc); 1272 } 1273 1274 // Ignore the statement if in the middle of inactive conditional 1275 // (e.g. ".if 0"). 1276 if (TheCondState.Ignore) { 1277 eatToEndOfStatement(); 1278 return false; 1279 } 1280 1281 // FIXME: Recurse on local labels? 1282 1283 // See what kind of statement we have. 1284 switch (Lexer.getKind()) { 1285 case AsmToken::Colon: { 1286 checkForValidSection(); 1287 1288 // identifier ':' -> Label. 1289 Lex(); 1290 1291 // Diagnose attempt to use '.' as a label. 1292 if (IDVal == ".") 1293 return Error(IDLoc, "invalid use of pseudo-symbol '.' as a label"); 1294 1295 // Diagnose attempt to use a variable as a label. 1296 // 1297 // FIXME: Diagnostics. Note the location of the definition as a label. 1298 // FIXME: This doesn't diagnose assignment to a symbol which has been 1299 // implicitly marked as external. 1300 MCSymbol *Sym; 1301 if (LocalLabelVal == -1) 1302 Sym = getContext().GetOrCreateSymbol(IDVal); 1303 else 1304 Sym = Ctx.CreateDirectionalLocalSymbol(LocalLabelVal); 1305 if (!Sym->isUndefined() || Sym->isVariable()) 1306 return Error(IDLoc, "invalid symbol redefinition"); 1307 1308 // Emit the label. 1309 if (!ParsingInlineAsm) 1310 Out.EmitLabel(Sym); 1311 1312 // If we are generating dwarf for assembly source files then gather the 1313 // info to make a dwarf label entry for this label if needed. 1314 if (getContext().getGenDwarfForAssembly()) 1315 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(), 1316 IDLoc); 1317 1318 getTargetParser().onLabelParsed(Sym); 1319 1320 // Consume any end of statement token, if present, to avoid spurious 1321 // AddBlankLine calls(). 1322 if (Lexer.is(AsmToken::EndOfStatement)) { 1323 Lex(); 1324 if (Lexer.is(AsmToken::Eof)) 1325 return false; 1326 } 1327 1328 return false; 1329 } 1330 1331 case AsmToken::Equal: 1332 // identifier '=' ... -> assignment statement 1333 Lex(); 1334 1335 return parseAssignment(IDVal, true); 1336 1337 default: // Normal instruction or directive. 1338 break; 1339 } 1340 1341 // If macros are enabled, check to see if this is a macro instantiation. 1342 if (areMacrosEnabled()) 1343 if (const MCAsmMacro *M = lookupMacro(IDVal)) { 1344 return handleMacroEntry(M, IDLoc); 1345 } 1346 1347 // Otherwise, we have a normal instruction or directive. 1348 1349 // Directives start with "." 1350 if (IDVal[0] == '.' && IDVal != ".") { 1351 // There are several entities interested in parsing directives: 1352 // 1353 // 1. The target-specific assembly parser. Some directives are target 1354 // specific or may potentially behave differently on certain targets. 1355 // 2. Asm parser extensions. For example, platform-specific parsers 1356 // (like the ELF parser) register themselves as extensions. 1357 // 3. The generic directive parser implemented by this class. These are 1358 // all the directives that behave in a target and platform independent 1359 // manner, or at least have a default behavior that's shared between 1360 // all targets and platforms. 1361 1362 // First query the target-specific parser. It will return 'true' if it 1363 // isn't interested in this directive. 1364 if (!getTargetParser().ParseDirective(ID)) 1365 return false; 1366 1367 // Next, check the extension directive map to see if any extension has 1368 // registered itself to parse this directive. 1369 std::pair<MCAsmParserExtension *, DirectiveHandler> Handler = 1370 ExtensionDirectiveMap.lookup(IDVal); 1371 if (Handler.first) 1372 return (*Handler.second)(Handler.first, IDVal, IDLoc); 1373 1374 // Finally, if no one else is interested in this directive, it must be 1375 // generic and familiar to this class. 1376 switch (DirKind) { 1377 default: 1378 break; 1379 case DK_SET: 1380 case DK_EQU: 1381 return parseDirectiveSet(IDVal, true); 1382 case DK_EQUIV: 1383 return parseDirectiveSet(IDVal, false); 1384 case DK_ASCII: 1385 return parseDirectiveAscii(IDVal, false); 1386 case DK_ASCIZ: 1387 case DK_STRING: 1388 return parseDirectiveAscii(IDVal, true); 1389 case DK_BYTE: 1390 return parseDirectiveValue(1); 1391 case DK_SHORT: 1392 case DK_VALUE: 1393 case DK_2BYTE: 1394 return parseDirectiveValue(2); 1395 case DK_LONG: 1396 case DK_INT: 1397 case DK_4BYTE: 1398 return parseDirectiveValue(4); 1399 case DK_QUAD: 1400 case DK_8BYTE: 1401 return parseDirectiveValue(8); 1402 case DK_OCTA: 1403 return parseDirectiveOctaValue(); 1404 case DK_SINGLE: 1405 case DK_FLOAT: 1406 return parseDirectiveRealValue(APFloat::IEEEsingle); 1407 case DK_DOUBLE: 1408 return parseDirectiveRealValue(APFloat::IEEEdouble); 1409 case DK_ALIGN: { 1410 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes(); 1411 return parseDirectiveAlign(IsPow2, /*ExprSize=*/1); 1412 } 1413 case DK_ALIGN32: { 1414 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes(); 1415 return parseDirectiveAlign(IsPow2, /*ExprSize=*/4); 1416 } 1417 case DK_BALIGN: 1418 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/1); 1419 case DK_BALIGNW: 1420 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/2); 1421 case DK_BALIGNL: 1422 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/4); 1423 case DK_P2ALIGN: 1424 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/1); 1425 case DK_P2ALIGNW: 1426 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/2); 1427 case DK_P2ALIGNL: 1428 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/4); 1429 case DK_ORG: 1430 return parseDirectiveOrg(); 1431 case DK_FILL: 1432 return parseDirectiveFill(); 1433 case DK_ZERO: 1434 return parseDirectiveZero(); 1435 case DK_EXTERN: 1436 eatToEndOfStatement(); // .extern is the default, ignore it. 1437 return false; 1438 case DK_GLOBL: 1439 case DK_GLOBAL: 1440 return parseDirectiveSymbolAttribute(MCSA_Global); 1441 case DK_LAZY_REFERENCE: 1442 return parseDirectiveSymbolAttribute(MCSA_LazyReference); 1443 case DK_NO_DEAD_STRIP: 1444 return parseDirectiveSymbolAttribute(MCSA_NoDeadStrip); 1445 case DK_SYMBOL_RESOLVER: 1446 return parseDirectiveSymbolAttribute(MCSA_SymbolResolver); 1447 case DK_PRIVATE_EXTERN: 1448 return parseDirectiveSymbolAttribute(MCSA_PrivateExtern); 1449 case DK_REFERENCE: 1450 return parseDirectiveSymbolAttribute(MCSA_Reference); 1451 case DK_WEAK_DEFINITION: 1452 return parseDirectiveSymbolAttribute(MCSA_WeakDefinition); 1453 case DK_WEAK_REFERENCE: 1454 return parseDirectiveSymbolAttribute(MCSA_WeakReference); 1455 case DK_WEAK_DEF_CAN_BE_HIDDEN: 1456 return parseDirectiveSymbolAttribute(MCSA_WeakDefAutoPrivate); 1457 case DK_COMM: 1458 case DK_COMMON: 1459 return parseDirectiveComm(/*IsLocal=*/false); 1460 case DK_LCOMM: 1461 return parseDirectiveComm(/*IsLocal=*/true); 1462 case DK_ABORT: 1463 return parseDirectiveAbort(); 1464 case DK_INCLUDE: 1465 return parseDirectiveInclude(); 1466 case DK_INCBIN: 1467 return parseDirectiveIncbin(); 1468 case DK_CODE16: 1469 case DK_CODE16GCC: 1470 return TokError(Twine(IDVal) + " not supported yet"); 1471 case DK_REPT: 1472 return parseDirectiveRept(IDLoc, IDVal); 1473 case DK_IRP: 1474 return parseDirectiveIrp(IDLoc); 1475 case DK_IRPC: 1476 return parseDirectiveIrpc(IDLoc); 1477 case DK_ENDR: 1478 return parseDirectiveEndr(IDLoc); 1479 case DK_BUNDLE_ALIGN_MODE: 1480 return parseDirectiveBundleAlignMode(); 1481 case DK_BUNDLE_LOCK: 1482 return parseDirectiveBundleLock(); 1483 case DK_BUNDLE_UNLOCK: 1484 return parseDirectiveBundleUnlock(); 1485 case DK_SLEB128: 1486 return parseDirectiveLEB128(true); 1487 case DK_ULEB128: 1488 return parseDirectiveLEB128(false); 1489 case DK_SPACE: 1490 case DK_SKIP: 1491 return parseDirectiveSpace(IDVal); 1492 case DK_FILE: 1493 return parseDirectiveFile(IDLoc); 1494 case DK_LINE: 1495 return parseDirectiveLine(); 1496 case DK_LOC: 1497 return parseDirectiveLoc(); 1498 case DK_STABS: 1499 return parseDirectiveStabs(); 1500 case DK_CFI_SECTIONS: 1501 return parseDirectiveCFISections(); 1502 case DK_CFI_STARTPROC: 1503 return parseDirectiveCFIStartProc(); 1504 case DK_CFI_ENDPROC: 1505 return parseDirectiveCFIEndProc(); 1506 case DK_CFI_DEF_CFA: 1507 return parseDirectiveCFIDefCfa(IDLoc); 1508 case DK_CFI_DEF_CFA_OFFSET: 1509 return parseDirectiveCFIDefCfaOffset(); 1510 case DK_CFI_ADJUST_CFA_OFFSET: 1511 return parseDirectiveCFIAdjustCfaOffset(); 1512 case DK_CFI_DEF_CFA_REGISTER: 1513 return parseDirectiveCFIDefCfaRegister(IDLoc); 1514 case DK_CFI_OFFSET: 1515 return parseDirectiveCFIOffset(IDLoc); 1516 case DK_CFI_REL_OFFSET: 1517 return parseDirectiveCFIRelOffset(IDLoc); 1518 case DK_CFI_PERSONALITY: 1519 return parseDirectiveCFIPersonalityOrLsda(true); 1520 case DK_CFI_LSDA: 1521 return parseDirectiveCFIPersonalityOrLsda(false); 1522 case DK_CFI_REMEMBER_STATE: 1523 return parseDirectiveCFIRememberState(); 1524 case DK_CFI_RESTORE_STATE: 1525 return parseDirectiveCFIRestoreState(); 1526 case DK_CFI_SAME_VALUE: 1527 return parseDirectiveCFISameValue(IDLoc); 1528 case DK_CFI_RESTORE: 1529 return parseDirectiveCFIRestore(IDLoc); 1530 case DK_CFI_ESCAPE: 1531 return parseDirectiveCFIEscape(); 1532 case DK_CFI_SIGNAL_FRAME: 1533 return parseDirectiveCFISignalFrame(); 1534 case DK_CFI_UNDEFINED: 1535 return parseDirectiveCFIUndefined(IDLoc); 1536 case DK_CFI_REGISTER: 1537 return parseDirectiveCFIRegister(IDLoc); 1538 case DK_CFI_WINDOW_SAVE: 1539 return parseDirectiveCFIWindowSave(); 1540 case DK_MACROS_ON: 1541 case DK_MACROS_OFF: 1542 return parseDirectiveMacrosOnOff(IDVal); 1543 case DK_MACRO: 1544 return parseDirectiveMacro(IDLoc); 1545 case DK_ENDM: 1546 case DK_ENDMACRO: 1547 return parseDirectiveEndMacro(IDVal); 1548 case DK_PURGEM: 1549 return parseDirectivePurgeMacro(IDLoc); 1550 case DK_END: 1551 return parseDirectiveEnd(IDLoc); 1552 case DK_ERR: 1553 return parseDirectiveError(IDLoc, false); 1554 case DK_ERROR: 1555 return parseDirectiveError(IDLoc, true); 1556 } 1557 1558 return Error(IDLoc, "unknown directive"); 1559 } 1560 1561 // __asm _emit or __asm __emit 1562 if (ParsingInlineAsm && (IDVal == "_emit" || IDVal == "__emit" || 1563 IDVal == "_EMIT" || IDVal == "__EMIT")) 1564 return parseDirectiveMSEmit(IDLoc, Info, IDVal.size()); 1565 1566 // __asm align 1567 if (ParsingInlineAsm && (IDVal == "align" || IDVal == "ALIGN")) 1568 return parseDirectiveMSAlign(IDLoc, Info); 1569 1570 checkForValidSection(); 1571 1572 // Canonicalize the opcode to lower case. 1573 std::string OpcodeStr = IDVal.lower(); 1574 ParseInstructionInfo IInfo(Info.AsmRewrites); 1575 bool HadError = getTargetParser().ParseInstruction(IInfo, OpcodeStr, IDLoc, 1576 Info.ParsedOperands); 1577 Info.ParseError = HadError; 1578 1579 // Dump the parsed representation, if requested. 1580 if (getShowParsedOperands()) { 1581 SmallString<256> Str; 1582 raw_svector_ostream OS(Str); 1583 OS << "parsed instruction: ["; 1584 for (unsigned i = 0; i != Info.ParsedOperands.size(); ++i) { 1585 if (i != 0) 1586 OS << ", "; 1587 Info.ParsedOperands[i]->print(OS); 1588 } 1589 OS << "]"; 1590 1591 printMessage(IDLoc, SourceMgr::DK_Note, OS.str()); 1592 } 1593 1594 // If we are generating dwarf for the current section then generate a .loc 1595 // directive for the instruction. 1596 if (!HadError && getContext().getGenDwarfForAssembly() && 1597 getContext().getGenDwarfSectionSyms().count( 1598 getStreamer().getCurrentSection().first)) { 1599 1600 unsigned Line = SrcMgr.FindLineNumber(IDLoc, CurBuffer); 1601 1602 // If we previously parsed a cpp hash file line comment then make sure the 1603 // current Dwarf File is for the CppHashFilename if not then emit the 1604 // Dwarf File table for it and adjust the line number for the .loc. 1605 if (CppHashFilename.size() != 0) { 1606 unsigned FileNumber = getStreamer().EmitDwarfFileDirective( 1607 0, StringRef(), CppHashFilename); 1608 getContext().setGenDwarfFileNumber(FileNumber); 1609 1610 // Since SrcMgr.FindLineNumber() is slow and messes up the SourceMgr's 1611 // cache with the different Loc from the call above we save the last 1612 // info we queried here with SrcMgr.FindLineNumber(). 1613 unsigned CppHashLocLineNo; 1614 if (LastQueryIDLoc == CppHashLoc && LastQueryBuffer == CppHashBuf) 1615 CppHashLocLineNo = LastQueryLine; 1616 else { 1617 CppHashLocLineNo = SrcMgr.FindLineNumber(CppHashLoc, CppHashBuf); 1618 LastQueryLine = CppHashLocLineNo; 1619 LastQueryIDLoc = CppHashLoc; 1620 LastQueryBuffer = CppHashBuf; 1621 } 1622 Line = CppHashLineNumber - 1 + (Line - CppHashLocLineNo); 1623 } 1624 1625 getStreamer().EmitDwarfLocDirective( 1626 getContext().getGenDwarfFileNumber(), Line, 0, 1627 DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0, 0, 0, 1628 StringRef()); 1629 } 1630 1631 // If parsing succeeded, match the instruction. 1632 if (!HadError) { 1633 unsigned ErrorInfo; 1634 getTargetParser().MatchAndEmitInstruction(IDLoc, Info.Opcode, 1635 Info.ParsedOperands, Out, 1636 ErrorInfo, ParsingInlineAsm); 1637 } 1638 1639 // Don't skip the rest of the line, the instruction parser is responsible for 1640 // that. 1641 return false; 1642 } 1643 1644 /// eatToEndOfLine uses the Lexer to eat the characters to the end of the line 1645 /// since they may not be able to be tokenized to get to the end of line token. 1646 void AsmParser::eatToEndOfLine() { 1647 if (!Lexer.is(AsmToken::EndOfStatement)) 1648 Lexer.LexUntilEndOfLine(); 1649 // Eat EOL. 1650 Lex(); 1651 } 1652 1653 /// parseCppHashLineFilenameComment as this: 1654 /// ::= # number "filename" 1655 /// or just as a full line comment if it doesn't have a number and a string. 1656 bool AsmParser::parseCppHashLineFilenameComment(const SMLoc &L) { 1657 Lex(); // Eat the hash token. 1658 1659 if (getLexer().isNot(AsmToken::Integer)) { 1660 // Consume the line since in cases it is not a well-formed line directive, 1661 // as if were simply a full line comment. 1662 eatToEndOfLine(); 1663 return false; 1664 } 1665 1666 int64_t LineNumber = getTok().getIntVal(); 1667 Lex(); 1668 1669 if (getLexer().isNot(AsmToken::String)) { 1670 eatToEndOfLine(); 1671 return false; 1672 } 1673 1674 StringRef Filename = getTok().getString(); 1675 // Get rid of the enclosing quotes. 1676 Filename = Filename.substr(1, Filename.size() - 2); 1677 1678 // Save the SMLoc, Filename and LineNumber for later use by diagnostics. 1679 CppHashLoc = L; 1680 CppHashFilename = Filename; 1681 CppHashLineNumber = LineNumber; 1682 CppHashBuf = CurBuffer; 1683 1684 // Ignore any trailing characters, they're just comment. 1685 eatToEndOfLine(); 1686 return false; 1687 } 1688 1689 /// \brief will use the last parsed cpp hash line filename comment 1690 /// for the Filename and LineNo if any in the diagnostic. 1691 void AsmParser::DiagHandler(const SMDiagnostic &Diag, void *Context) { 1692 const AsmParser *Parser = static_cast<const AsmParser *>(Context); 1693 raw_ostream &OS = errs(); 1694 1695 const SourceMgr &DiagSrcMgr = *Diag.getSourceMgr(); 1696 const SMLoc &DiagLoc = Diag.getLoc(); 1697 unsigned DiagBuf = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 1698 unsigned CppHashBuf = 1699 Parser->SrcMgr.FindBufferContainingLoc(Parser->CppHashLoc); 1700 1701 // Like SourceMgr::printMessage() we need to print the include stack if any 1702 // before printing the message. 1703 unsigned DiagCurBuffer = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 1704 if (!Parser->SavedDiagHandler && DiagCurBuffer && 1705 DiagCurBuffer != DiagSrcMgr.getMainFileID()) { 1706 SMLoc ParentIncludeLoc = DiagSrcMgr.getParentIncludeLoc(DiagCurBuffer); 1707 DiagSrcMgr.PrintIncludeStack(ParentIncludeLoc, OS); 1708 } 1709 1710 // If we have not parsed a cpp hash line filename comment or the source 1711 // manager changed or buffer changed (like in a nested include) then just 1712 // print the normal diagnostic using its Filename and LineNo. 1713 if (!Parser->CppHashLineNumber || &DiagSrcMgr != &Parser->SrcMgr || 1714 DiagBuf != CppHashBuf) { 1715 if (Parser->SavedDiagHandler) 1716 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext); 1717 else 1718 Diag.print(nullptr, OS); 1719 return; 1720 } 1721 1722 // Use the CppHashFilename and calculate a line number based on the 1723 // CppHashLoc and CppHashLineNumber relative to this Diag's SMLoc for 1724 // the diagnostic. 1725 const std::string &Filename = Parser->CppHashFilename; 1726 1727 int DiagLocLineNo = DiagSrcMgr.FindLineNumber(DiagLoc, DiagBuf); 1728 int CppHashLocLineNo = 1729 Parser->SrcMgr.FindLineNumber(Parser->CppHashLoc, CppHashBuf); 1730 int LineNo = 1731 Parser->CppHashLineNumber - 1 + (DiagLocLineNo - CppHashLocLineNo); 1732 1733 SMDiagnostic NewDiag(*Diag.getSourceMgr(), Diag.getLoc(), Filename, LineNo, 1734 Diag.getColumnNo(), Diag.getKind(), Diag.getMessage(), 1735 Diag.getLineContents(), Diag.getRanges()); 1736 1737 if (Parser->SavedDiagHandler) 1738 Parser->SavedDiagHandler(NewDiag, Parser->SavedDiagContext); 1739 else 1740 NewDiag.print(nullptr, OS); 1741 } 1742 1743 // FIXME: This is mostly duplicated from the function in AsmLexer.cpp. The 1744 // difference being that that function accepts '@' as part of identifiers and 1745 // we can't do that. AsmLexer.cpp should probably be changed to handle 1746 // '@' as a special case when needed. 1747 static bool isIdentifierChar(char c) { 1748 return isalnum(static_cast<unsigned char>(c)) || c == '_' || c == '$' || 1749 c == '.'; 1750 } 1751 1752 bool AsmParser::expandMacro(raw_svector_ostream &OS, StringRef Body, 1753 ArrayRef<MCAsmMacroParameter> Parameters, 1754 ArrayRef<MCAsmMacroArgument> A, const SMLoc &L) { 1755 unsigned NParameters = Parameters.size(); 1756 bool HasVararg = NParameters ? Parameters.back().Vararg : false; 1757 if ((!IsDarwin || NParameters != 0) && NParameters != A.size()) 1758 return Error(L, "Wrong number of arguments"); 1759 1760 // A macro without parameters is handled differently on Darwin: 1761 // gas accepts no arguments and does no substitutions 1762 while (!Body.empty()) { 1763 // Scan for the next substitution. 1764 std::size_t End = Body.size(), Pos = 0; 1765 for (; Pos != End; ++Pos) { 1766 // Check for a substitution or escape. 1767 if (IsDarwin && !NParameters) { 1768 // This macro has no parameters, look for $0, $1, etc. 1769 if (Body[Pos] != '$' || Pos + 1 == End) 1770 continue; 1771 1772 char Next = Body[Pos + 1]; 1773 if (Next == '$' || Next == 'n' || 1774 isdigit(static_cast<unsigned char>(Next))) 1775 break; 1776 } else { 1777 // This macro has parameters, look for \foo, \bar, etc. 1778 if (Body[Pos] == '\\' && Pos + 1 != End) 1779 break; 1780 } 1781 } 1782 1783 // Add the prefix. 1784 OS << Body.slice(0, Pos); 1785 1786 // Check if we reached the end. 1787 if (Pos == End) 1788 break; 1789 1790 if (IsDarwin && !NParameters) { 1791 switch (Body[Pos + 1]) { 1792 // $$ => $ 1793 case '$': 1794 OS << '$'; 1795 break; 1796 1797 // $n => number of arguments 1798 case 'n': 1799 OS << A.size(); 1800 break; 1801 1802 // $[0-9] => argument 1803 default: { 1804 // Missing arguments are ignored. 1805 unsigned Index = Body[Pos + 1] - '0'; 1806 if (Index >= A.size()) 1807 break; 1808 1809 // Otherwise substitute with the token values, with spaces eliminated. 1810 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(), 1811 ie = A[Index].end(); 1812 it != ie; ++it) 1813 OS << it->getString(); 1814 break; 1815 } 1816 } 1817 Pos += 2; 1818 } else { 1819 unsigned I = Pos + 1; 1820 while (isIdentifierChar(Body[I]) && I + 1 != End) 1821 ++I; 1822 1823 const char *Begin = Body.data() + Pos + 1; 1824 StringRef Argument(Begin, I - (Pos + 1)); 1825 unsigned Index = 0; 1826 for (; Index < NParameters; ++Index) 1827 if (Parameters[Index].Name == Argument) 1828 break; 1829 1830 if (Index == NParameters) { 1831 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')') 1832 Pos += 3; 1833 else { 1834 OS << '\\' << Argument; 1835 Pos = I; 1836 } 1837 } else { 1838 bool VarargParameter = HasVararg && Index == (NParameters - 1); 1839 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(), 1840 ie = A[Index].end(); 1841 it != ie; ++it) 1842 // We expect no quotes around the string's contents when 1843 // parsing for varargs. 1844 if (it->getKind() != AsmToken::String || VarargParameter) 1845 OS << it->getString(); 1846 else 1847 OS << it->getStringContents(); 1848 1849 Pos += 1 + Argument.size(); 1850 } 1851 } 1852 // Update the scan point. 1853 Body = Body.substr(Pos); 1854 } 1855 1856 return false; 1857 } 1858 1859 MacroInstantiation::MacroInstantiation(const MCAsmMacro *M, SMLoc IL, int EB, 1860 SMLoc EL, MemoryBuffer *I) 1861 : TheMacro(M), Instantiation(I), InstantiationLoc(IL), ExitBuffer(EB), 1862 ExitLoc(EL) {} 1863 1864 static bool isOperator(AsmToken::TokenKind kind) { 1865 switch (kind) { 1866 default: 1867 return false; 1868 case AsmToken::Plus: 1869 case AsmToken::Minus: 1870 case AsmToken::Tilde: 1871 case AsmToken::Slash: 1872 case AsmToken::Star: 1873 case AsmToken::Dot: 1874 case AsmToken::Equal: 1875 case AsmToken::EqualEqual: 1876 case AsmToken::Pipe: 1877 case AsmToken::PipePipe: 1878 case AsmToken::Caret: 1879 case AsmToken::Amp: 1880 case AsmToken::AmpAmp: 1881 case AsmToken::Exclaim: 1882 case AsmToken::ExclaimEqual: 1883 case AsmToken::Percent: 1884 case AsmToken::Less: 1885 case AsmToken::LessEqual: 1886 case AsmToken::LessLess: 1887 case AsmToken::LessGreater: 1888 case AsmToken::Greater: 1889 case AsmToken::GreaterEqual: 1890 case AsmToken::GreaterGreater: 1891 return true; 1892 } 1893 } 1894 1895 namespace { 1896 class AsmLexerSkipSpaceRAII { 1897 public: 1898 AsmLexerSkipSpaceRAII(AsmLexer &Lexer, bool SkipSpace) : Lexer(Lexer) { 1899 Lexer.setSkipSpace(SkipSpace); 1900 } 1901 1902 ~AsmLexerSkipSpaceRAII() { 1903 Lexer.setSkipSpace(true); 1904 } 1905 1906 private: 1907 AsmLexer &Lexer; 1908 }; 1909 } 1910 1911 bool AsmParser::parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg) { 1912 1913 if (Vararg) { 1914 if (Lexer.isNot(AsmToken::EndOfStatement)) { 1915 StringRef Str = parseStringToEndOfStatement(); 1916 MA.push_back(AsmToken(AsmToken::String, Str)); 1917 } 1918 return false; 1919 } 1920 1921 unsigned ParenLevel = 0; 1922 unsigned AddTokens = 0; 1923 1924 // Darwin doesn't use spaces to delmit arguments. 1925 AsmLexerSkipSpaceRAII ScopedSkipSpace(Lexer, IsDarwin); 1926 1927 for (;;) { 1928 if (Lexer.is(AsmToken::Eof) || Lexer.is(AsmToken::Equal)) 1929 return TokError("unexpected token in macro instantiation"); 1930 1931 if (ParenLevel == 0 && Lexer.is(AsmToken::Comma)) 1932 break; 1933 1934 if (Lexer.is(AsmToken::Space)) { 1935 Lex(); // Eat spaces 1936 1937 // Spaces can delimit parameters, but could also be part an expression. 1938 // If the token after a space is an operator, add the token and the next 1939 // one into this argument 1940 if (!IsDarwin) { 1941 if (isOperator(Lexer.getKind())) { 1942 // Check to see whether the token is used as an operator, 1943 // or part of an identifier 1944 const char *NextChar = getTok().getEndLoc().getPointer(); 1945 if (*NextChar == ' ') 1946 AddTokens = 2; 1947 } 1948 1949 if (!AddTokens && ParenLevel == 0) { 1950 break; 1951 } 1952 } 1953 } 1954 1955 // handleMacroEntry relies on not advancing the lexer here 1956 // to be able to fill in the remaining default parameter values 1957 if (Lexer.is(AsmToken::EndOfStatement)) 1958 break; 1959 1960 // Adjust the current parentheses level. 1961 if (Lexer.is(AsmToken::LParen)) 1962 ++ParenLevel; 1963 else if (Lexer.is(AsmToken::RParen) && ParenLevel) 1964 --ParenLevel; 1965 1966 // Append the token to the current argument list. 1967 MA.push_back(getTok()); 1968 if (AddTokens) 1969 AddTokens--; 1970 Lex(); 1971 } 1972 1973 if (ParenLevel != 0) 1974 return TokError("unbalanced parentheses in macro argument"); 1975 return false; 1976 } 1977 1978 // Parse the macro instantiation arguments. 1979 bool AsmParser::parseMacroArguments(const MCAsmMacro *M, 1980 MCAsmMacroArguments &A) { 1981 const unsigned NParameters = M ? M->Parameters.size() : 0; 1982 bool NamedParametersFound = false; 1983 SmallVector<SMLoc, 4> FALocs; 1984 1985 A.resize(NParameters); 1986 FALocs.resize(NParameters); 1987 1988 // Parse two kinds of macro invocations: 1989 // - macros defined without any parameters accept an arbitrary number of them 1990 // - macros defined with parameters accept at most that many of them 1991 bool HasVararg = NParameters ? M->Parameters.back().Vararg : false; 1992 for (unsigned Parameter = 0; !NParameters || Parameter < NParameters; 1993 ++Parameter) { 1994 SMLoc IDLoc = Lexer.getLoc(); 1995 MCAsmMacroParameter FA; 1996 1997 if (Lexer.is(AsmToken::Identifier) && Lexer.peekTok().is(AsmToken::Equal)) { 1998 if (parseIdentifier(FA.Name)) { 1999 Error(IDLoc, "invalid argument identifier for formal argument"); 2000 eatToEndOfStatement(); 2001 return true; 2002 } 2003 2004 if (!Lexer.is(AsmToken::Equal)) { 2005 TokError("expected '=' after formal parameter identifier"); 2006 eatToEndOfStatement(); 2007 return true; 2008 } 2009 Lex(); 2010 2011 NamedParametersFound = true; 2012 } 2013 2014 if (NamedParametersFound && FA.Name.empty()) { 2015 Error(IDLoc, "cannot mix positional and keyword arguments"); 2016 eatToEndOfStatement(); 2017 return true; 2018 } 2019 2020 bool Vararg = HasVararg && Parameter == (NParameters - 1); 2021 if (parseMacroArgument(FA.Value, Vararg)) 2022 return true; 2023 2024 unsigned PI = Parameter; 2025 if (!FA.Name.empty()) { 2026 unsigned FAI = 0; 2027 for (FAI = 0; FAI < NParameters; ++FAI) 2028 if (M->Parameters[FAI].Name == FA.Name) 2029 break; 2030 2031 if (FAI >= NParameters) { 2032 assert(M && "expected macro to be defined"); 2033 Error(IDLoc, 2034 "parameter named '" + FA.Name + "' does not exist for macro '" + 2035 M->Name + "'"); 2036 return true; 2037 } 2038 PI = FAI; 2039 } 2040 2041 if (!FA.Value.empty()) { 2042 if (A.size() <= PI) 2043 A.resize(PI + 1); 2044 A[PI] = FA.Value; 2045 2046 if (FALocs.size() <= PI) 2047 FALocs.resize(PI + 1); 2048 2049 FALocs[PI] = Lexer.getLoc(); 2050 } 2051 2052 // At the end of the statement, fill in remaining arguments that have 2053 // default values. If there aren't any, then the next argument is 2054 // required but missing 2055 if (Lexer.is(AsmToken::EndOfStatement)) { 2056 bool Failure = false; 2057 for (unsigned FAI = 0; FAI < NParameters; ++FAI) { 2058 if (A[FAI].empty()) { 2059 if (M->Parameters[FAI].Required) { 2060 Error(FALocs[FAI].isValid() ? FALocs[FAI] : Lexer.getLoc(), 2061 "missing value for required parameter " 2062 "'" + M->Parameters[FAI].Name + "' in macro '" + M->Name + "'"); 2063 Failure = true; 2064 } 2065 2066 if (!M->Parameters[FAI].Value.empty()) 2067 A[FAI] = M->Parameters[FAI].Value; 2068 } 2069 } 2070 return Failure; 2071 } 2072 2073 if (Lexer.is(AsmToken::Comma)) 2074 Lex(); 2075 } 2076 2077 return TokError("too many positional arguments"); 2078 } 2079 2080 const MCAsmMacro *AsmParser::lookupMacro(StringRef Name) { 2081 StringMap<MCAsmMacro *>::iterator I = MacroMap.find(Name); 2082 return (I == MacroMap.end()) ? nullptr : I->getValue(); 2083 } 2084 2085 void AsmParser::defineMacro(StringRef Name, const MCAsmMacro &Macro) { 2086 MacroMap[Name] = new MCAsmMacro(Macro); 2087 } 2088 2089 void AsmParser::undefineMacro(StringRef Name) { 2090 StringMap<MCAsmMacro *>::iterator I = MacroMap.find(Name); 2091 if (I != MacroMap.end()) { 2092 delete I->getValue(); 2093 MacroMap.erase(I); 2094 } 2095 } 2096 2097 bool AsmParser::handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc) { 2098 // Arbitrarily limit macro nesting depth, to match 'as'. We can eliminate 2099 // this, although we should protect against infinite loops. 2100 if (ActiveMacros.size() == 20) 2101 return TokError("macros cannot be nested more than 20 levels deep"); 2102 2103 MCAsmMacroArguments A; 2104 if (parseMacroArguments(M, A)) 2105 return true; 2106 2107 // Macro instantiation is lexical, unfortunately. We construct a new buffer 2108 // to hold the macro body with substitutions. 2109 SmallString<256> Buf; 2110 StringRef Body = M->Body; 2111 raw_svector_ostream OS(Buf); 2112 2113 if (expandMacro(OS, Body, M->Parameters, A, getTok().getLoc())) 2114 return true; 2115 2116 // We include the .endmacro in the buffer as our cue to exit the macro 2117 // instantiation. 2118 OS << ".endmacro\n"; 2119 2120 MemoryBuffer *Instantiation = 2121 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 2122 2123 // Create the macro instantiation object and add to the current macro 2124 // instantiation stack. 2125 MacroInstantiation *MI = new MacroInstantiation( 2126 M, NameLoc, CurBuffer, getTok().getLoc(), Instantiation); 2127 ActiveMacros.push_back(MI); 2128 2129 // Jump to the macro instantiation and prime the lexer. 2130 CurBuffer = SrcMgr.AddNewSourceBuffer(MI->Instantiation, SMLoc()); 2131 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 2132 Lex(); 2133 2134 return false; 2135 } 2136 2137 void AsmParser::handleMacroExit() { 2138 // Jump to the EndOfStatement we should return to, and consume it. 2139 jumpToLoc(ActiveMacros.back()->ExitLoc, ActiveMacros.back()->ExitBuffer); 2140 Lex(); 2141 2142 // Pop the instantiation entry. 2143 delete ActiveMacros.back(); 2144 ActiveMacros.pop_back(); 2145 } 2146 2147 static bool isUsedIn(const MCSymbol *Sym, const MCExpr *Value) { 2148 switch (Value->getKind()) { 2149 case MCExpr::Binary: { 2150 const MCBinaryExpr *BE = static_cast<const MCBinaryExpr *>(Value); 2151 return isUsedIn(Sym, BE->getLHS()) || isUsedIn(Sym, BE->getRHS()); 2152 } 2153 case MCExpr::Target: 2154 case MCExpr::Constant: 2155 return false; 2156 case MCExpr::SymbolRef: { 2157 const MCSymbol &S = 2158 static_cast<const MCSymbolRefExpr *>(Value)->getSymbol(); 2159 if (S.isVariable()) 2160 return isUsedIn(Sym, S.getVariableValue()); 2161 return &S == Sym; 2162 } 2163 case MCExpr::Unary: 2164 return isUsedIn(Sym, static_cast<const MCUnaryExpr *>(Value)->getSubExpr()); 2165 } 2166 2167 llvm_unreachable("Unknown expr kind!"); 2168 } 2169 2170 bool AsmParser::parseAssignment(StringRef Name, bool allow_redef, 2171 bool NoDeadStrip) { 2172 // FIXME: Use better location, we should use proper tokens. 2173 SMLoc EqualLoc = Lexer.getLoc(); 2174 2175 const MCExpr *Value; 2176 if (parseExpression(Value)) 2177 return true; 2178 2179 // Note: we don't count b as used in "a = b". This is to allow 2180 // a = b 2181 // b = c 2182 2183 if (Lexer.isNot(AsmToken::EndOfStatement)) 2184 return TokError("unexpected token in assignment"); 2185 2186 // Eat the end of statement marker. 2187 Lex(); 2188 2189 // Validate that the LHS is allowed to be a variable (either it has not been 2190 // used as a symbol, or it is an absolute symbol). 2191 MCSymbol *Sym = getContext().LookupSymbol(Name); 2192 if (Sym) { 2193 // Diagnose assignment to a label. 2194 // 2195 // FIXME: Diagnostics. Note the location of the definition as a label. 2196 // FIXME: Diagnose assignment to protected identifier (e.g., register name). 2197 if (isUsedIn(Sym, Value)) 2198 return Error(EqualLoc, "Recursive use of '" + Name + "'"); 2199 else if (Sym->isUndefined() && !Sym->isUsed() && !Sym->isVariable()) 2200 ; // Allow redefinitions of undefined symbols only used in directives. 2201 else if (Sym->isVariable() && !Sym->isUsed() && allow_redef) 2202 ; // Allow redefinitions of variables that haven't yet been used. 2203 else if (!Sym->isUndefined() && (!Sym->isVariable() || !allow_redef)) 2204 return Error(EqualLoc, "redefinition of '" + Name + "'"); 2205 else if (!Sym->isVariable()) 2206 return Error(EqualLoc, "invalid assignment to '" + Name + "'"); 2207 else if (!isa<MCConstantExpr>(Sym->getVariableValue())) 2208 return Error(EqualLoc, "invalid reassignment of non-absolute variable '" + 2209 Name + "'"); 2210 2211 // Don't count these checks as uses. 2212 Sym->setUsed(false); 2213 } else if (Name == ".") { 2214 if (Out.EmitValueToOffset(Value, 0)) { 2215 Error(EqualLoc, "expected absolute expression"); 2216 eatToEndOfStatement(); 2217 } 2218 return false; 2219 } else 2220 Sym = getContext().GetOrCreateSymbol(Name); 2221 2222 // Do the assignment. 2223 Out.EmitAssignment(Sym, Value); 2224 if (NoDeadStrip) 2225 Out.EmitSymbolAttribute(Sym, MCSA_NoDeadStrip); 2226 2227 return false; 2228 } 2229 2230 /// parseIdentifier: 2231 /// ::= identifier 2232 /// ::= string 2233 bool AsmParser::parseIdentifier(StringRef &Res) { 2234 // The assembler has relaxed rules for accepting identifiers, in particular we 2235 // allow things like '.globl $foo' and '.def @feat.00', which would normally be 2236 // separate tokens. At this level, we have already lexed so we cannot (currently) 2237 // handle this as a context dependent token, instead we detect adjacent tokens 2238 // and return the combined identifier. 2239 if (Lexer.is(AsmToken::Dollar) || Lexer.is(AsmToken::At)) { 2240 SMLoc PrefixLoc = getLexer().getLoc(); 2241 2242 // Consume the prefix character, and check for a following identifier. 2243 Lex(); 2244 if (Lexer.isNot(AsmToken::Identifier)) 2245 return true; 2246 2247 // We have a '$' or '@' followed by an identifier, make sure they are adjacent. 2248 if (PrefixLoc.getPointer() + 1 != getTok().getLoc().getPointer()) 2249 return true; 2250 2251 // Construct the joined identifier and consume the token. 2252 Res = 2253 StringRef(PrefixLoc.getPointer(), getTok().getIdentifier().size() + 1); 2254 Lex(); 2255 return false; 2256 } 2257 2258 if (Lexer.isNot(AsmToken::Identifier) && Lexer.isNot(AsmToken::String)) 2259 return true; 2260 2261 Res = getTok().getIdentifier(); 2262 2263 Lex(); // Consume the identifier token. 2264 2265 return false; 2266 } 2267 2268 /// parseDirectiveSet: 2269 /// ::= .equ identifier ',' expression 2270 /// ::= .equiv identifier ',' expression 2271 /// ::= .set identifier ',' expression 2272 bool AsmParser::parseDirectiveSet(StringRef IDVal, bool allow_redef) { 2273 StringRef Name; 2274 2275 if (parseIdentifier(Name)) 2276 return TokError("expected identifier after '" + Twine(IDVal) + "'"); 2277 2278 if (getLexer().isNot(AsmToken::Comma)) 2279 return TokError("unexpected token in '" + Twine(IDVal) + "'"); 2280 Lex(); 2281 2282 return parseAssignment(Name, allow_redef, true); 2283 } 2284 2285 bool AsmParser::parseEscapedString(std::string &Data) { 2286 assert(getLexer().is(AsmToken::String) && "Unexpected current token!"); 2287 2288 Data = ""; 2289 StringRef Str = getTok().getStringContents(); 2290 for (unsigned i = 0, e = Str.size(); i != e; ++i) { 2291 if (Str[i] != '\\') { 2292 Data += Str[i]; 2293 continue; 2294 } 2295 2296 // Recognize escaped characters. Note that this escape semantics currently 2297 // loosely follows Darwin 'as'. Notably, it doesn't support hex escapes. 2298 ++i; 2299 if (i == e) 2300 return TokError("unexpected backslash at end of string"); 2301 2302 // Recognize octal sequences. 2303 if ((unsigned)(Str[i] - '0') <= 7) { 2304 // Consume up to three octal characters. 2305 unsigned Value = Str[i] - '0'; 2306 2307 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) { 2308 ++i; 2309 Value = Value * 8 + (Str[i] - '0'); 2310 2311 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) { 2312 ++i; 2313 Value = Value * 8 + (Str[i] - '0'); 2314 } 2315 } 2316 2317 if (Value > 255) 2318 return TokError("invalid octal escape sequence (out of range)"); 2319 2320 Data += (unsigned char)Value; 2321 continue; 2322 } 2323 2324 // Otherwise recognize individual escapes. 2325 switch (Str[i]) { 2326 default: 2327 // Just reject invalid escape sequences for now. 2328 return TokError("invalid escape sequence (unrecognized character)"); 2329 2330 case 'b': Data += '\b'; break; 2331 case 'f': Data += '\f'; break; 2332 case 'n': Data += '\n'; break; 2333 case 'r': Data += '\r'; break; 2334 case 't': Data += '\t'; break; 2335 case '"': Data += '"'; break; 2336 case '\\': Data += '\\'; break; 2337 } 2338 } 2339 2340 return false; 2341 } 2342 2343 /// parseDirectiveAscii: 2344 /// ::= ( .ascii | .asciz | .string ) [ "string" ( , "string" )* ] 2345 bool AsmParser::parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated) { 2346 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2347 checkForValidSection(); 2348 2349 for (;;) { 2350 if (getLexer().isNot(AsmToken::String)) 2351 return TokError("expected string in '" + Twine(IDVal) + "' directive"); 2352 2353 std::string Data; 2354 if (parseEscapedString(Data)) 2355 return true; 2356 2357 getStreamer().EmitBytes(Data); 2358 if (ZeroTerminated) 2359 getStreamer().EmitBytes(StringRef("\0", 1)); 2360 2361 Lex(); 2362 2363 if (getLexer().is(AsmToken::EndOfStatement)) 2364 break; 2365 2366 if (getLexer().isNot(AsmToken::Comma)) 2367 return TokError("unexpected token in '" + Twine(IDVal) + "' directive"); 2368 Lex(); 2369 } 2370 } 2371 2372 Lex(); 2373 return false; 2374 } 2375 2376 /// parseDirectiveValue 2377 /// ::= (.byte | .short | ... ) [ expression (, expression)* ] 2378 bool AsmParser::parseDirectiveValue(unsigned Size) { 2379 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2380 checkForValidSection(); 2381 2382 for (;;) { 2383 const MCExpr *Value; 2384 SMLoc ExprLoc = getLexer().getLoc(); 2385 if (parseExpression(Value)) 2386 return true; 2387 2388 // Special case constant expressions to match code generator. 2389 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2390 assert(Size <= 8 && "Invalid size"); 2391 uint64_t IntValue = MCE->getValue(); 2392 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue)) 2393 return Error(ExprLoc, "literal value out of range for directive"); 2394 getStreamer().EmitIntValue(IntValue, Size); 2395 } else 2396 getStreamer().EmitValue(Value, Size, ExprLoc); 2397 2398 if (getLexer().is(AsmToken::EndOfStatement)) 2399 break; 2400 2401 // FIXME: Improve diagnostic. 2402 if (getLexer().isNot(AsmToken::Comma)) 2403 return TokError("unexpected token in directive"); 2404 Lex(); 2405 } 2406 } 2407 2408 Lex(); 2409 return false; 2410 } 2411 2412 /// ParseDirectiveOctaValue 2413 /// ::= .octa [ hexconstant (, hexconstant)* ] 2414 bool AsmParser::parseDirectiveOctaValue() { 2415 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2416 checkForValidSection(); 2417 2418 for (;;) { 2419 if (Lexer.getKind() == AsmToken::Error) 2420 return true; 2421 if (Lexer.getKind() != AsmToken::Integer && 2422 Lexer.getKind() != AsmToken::BigNum) 2423 return TokError("unknown token in expression"); 2424 2425 SMLoc ExprLoc = getLexer().getLoc(); 2426 APInt IntValue = getTok().getAPIntVal(); 2427 Lex(); 2428 2429 uint64_t hi, lo; 2430 if (IntValue.isIntN(64)) { 2431 hi = 0; 2432 lo = IntValue.getZExtValue(); 2433 } else if (IntValue.isIntN(128)) { 2434 // It might actually have more than 128 bits, but the top ones are zero. 2435 hi = IntValue.getHiBits(IntValue.getBitWidth() - 64).getZExtValue(); 2436 lo = IntValue.getLoBits(64).getZExtValue(); 2437 } else 2438 return Error(ExprLoc, "literal value out of range for directive"); 2439 2440 if (MAI.isLittleEndian()) { 2441 getStreamer().EmitIntValue(lo, 8); 2442 getStreamer().EmitIntValue(hi, 8); 2443 } else { 2444 getStreamer().EmitIntValue(hi, 8); 2445 getStreamer().EmitIntValue(lo, 8); 2446 } 2447 2448 if (getLexer().is(AsmToken::EndOfStatement)) 2449 break; 2450 2451 // FIXME: Improve diagnostic. 2452 if (getLexer().isNot(AsmToken::Comma)) 2453 return TokError("unexpected token in directive"); 2454 Lex(); 2455 } 2456 } 2457 2458 Lex(); 2459 return false; 2460 } 2461 2462 /// parseDirectiveRealValue 2463 /// ::= (.single | .double) [ expression (, expression)* ] 2464 bool AsmParser::parseDirectiveRealValue(const fltSemantics &Semantics) { 2465 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2466 checkForValidSection(); 2467 2468 for (;;) { 2469 // We don't truly support arithmetic on floating point expressions, so we 2470 // have to manually parse unary prefixes. 2471 bool IsNeg = false; 2472 if (getLexer().is(AsmToken::Minus)) { 2473 Lex(); 2474 IsNeg = true; 2475 } else if (getLexer().is(AsmToken::Plus)) 2476 Lex(); 2477 2478 if (getLexer().isNot(AsmToken::Integer) && 2479 getLexer().isNot(AsmToken::Real) && 2480 getLexer().isNot(AsmToken::Identifier)) 2481 return TokError("unexpected token in directive"); 2482 2483 // Convert to an APFloat. 2484 APFloat Value(Semantics); 2485 StringRef IDVal = getTok().getString(); 2486 if (getLexer().is(AsmToken::Identifier)) { 2487 if (!IDVal.compare_lower("infinity") || !IDVal.compare_lower("inf")) 2488 Value = APFloat::getInf(Semantics); 2489 else if (!IDVal.compare_lower("nan")) 2490 Value = APFloat::getNaN(Semantics, false, ~0); 2491 else 2492 return TokError("invalid floating point literal"); 2493 } else if (Value.convertFromString(IDVal, APFloat::rmNearestTiesToEven) == 2494 APFloat::opInvalidOp) 2495 return TokError("invalid floating point literal"); 2496 if (IsNeg) 2497 Value.changeSign(); 2498 2499 // Consume the numeric token. 2500 Lex(); 2501 2502 // Emit the value as an integer. 2503 APInt AsInt = Value.bitcastToAPInt(); 2504 getStreamer().EmitIntValue(AsInt.getLimitedValue(), 2505 AsInt.getBitWidth() / 8); 2506 2507 if (getLexer().is(AsmToken::EndOfStatement)) 2508 break; 2509 2510 if (getLexer().isNot(AsmToken::Comma)) 2511 return TokError("unexpected token in directive"); 2512 Lex(); 2513 } 2514 } 2515 2516 Lex(); 2517 return false; 2518 } 2519 2520 /// parseDirectiveZero 2521 /// ::= .zero expression 2522 bool AsmParser::parseDirectiveZero() { 2523 checkForValidSection(); 2524 2525 int64_t NumBytes; 2526 if (parseAbsoluteExpression(NumBytes)) 2527 return true; 2528 2529 int64_t Val = 0; 2530 if (getLexer().is(AsmToken::Comma)) { 2531 Lex(); 2532 if (parseAbsoluteExpression(Val)) 2533 return true; 2534 } 2535 2536 if (getLexer().isNot(AsmToken::EndOfStatement)) 2537 return TokError("unexpected token in '.zero' directive"); 2538 2539 Lex(); 2540 2541 getStreamer().EmitFill(NumBytes, Val); 2542 2543 return false; 2544 } 2545 2546 /// parseDirectiveFill 2547 /// ::= .fill expression [ , expression [ , expression ] ] 2548 bool AsmParser::parseDirectiveFill() { 2549 checkForValidSection(); 2550 2551 SMLoc RepeatLoc = getLexer().getLoc(); 2552 int64_t NumValues; 2553 if (parseAbsoluteExpression(NumValues)) 2554 return true; 2555 2556 if (NumValues < 0) { 2557 Warning(RepeatLoc, 2558 "'.fill' directive with negative repeat count has no effect"); 2559 NumValues = 0; 2560 } 2561 2562 int64_t FillSize = 1; 2563 int64_t FillExpr = 0; 2564 2565 SMLoc SizeLoc, ExprLoc; 2566 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2567 if (getLexer().isNot(AsmToken::Comma)) 2568 return TokError("unexpected token in '.fill' directive"); 2569 Lex(); 2570 2571 SizeLoc = getLexer().getLoc(); 2572 if (parseAbsoluteExpression(FillSize)) 2573 return true; 2574 2575 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2576 if (getLexer().isNot(AsmToken::Comma)) 2577 return TokError("unexpected token in '.fill' directive"); 2578 Lex(); 2579 2580 ExprLoc = getLexer().getLoc(); 2581 if (parseAbsoluteExpression(FillExpr)) 2582 return true; 2583 2584 if (getLexer().isNot(AsmToken::EndOfStatement)) 2585 return TokError("unexpected token in '.fill' directive"); 2586 2587 Lex(); 2588 } 2589 } 2590 2591 if (FillSize < 0) { 2592 Warning(SizeLoc, "'.fill' directive with negative size has no effect"); 2593 NumValues = 0; 2594 } 2595 if (FillSize > 8) { 2596 Warning(SizeLoc, "'.fill' directive with size greater than 8 has been truncated to 8"); 2597 FillSize = 8; 2598 } 2599 2600 if (!isUInt<32>(FillExpr) && FillSize > 4) 2601 Warning(ExprLoc, "'.fill' directive pattern has been truncated to 32-bits"); 2602 2603 int64_t NonZeroFillSize = FillSize > 4 ? 4 : FillSize; 2604 FillExpr &= ~0ULL >> (64 - NonZeroFillSize * 8); 2605 2606 for (uint64_t i = 0, e = NumValues; i != e; ++i) { 2607 getStreamer().EmitIntValue(FillExpr, NonZeroFillSize); 2608 getStreamer().EmitIntValue(0, FillSize - NonZeroFillSize); 2609 } 2610 2611 return false; 2612 } 2613 2614 /// parseDirectiveOrg 2615 /// ::= .org expression [ , expression ] 2616 bool AsmParser::parseDirectiveOrg() { 2617 checkForValidSection(); 2618 2619 const MCExpr *Offset; 2620 SMLoc Loc = getTok().getLoc(); 2621 if (parseExpression(Offset)) 2622 return true; 2623 2624 // Parse optional fill expression. 2625 int64_t FillExpr = 0; 2626 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2627 if (getLexer().isNot(AsmToken::Comma)) 2628 return TokError("unexpected token in '.org' directive"); 2629 Lex(); 2630 2631 if (parseAbsoluteExpression(FillExpr)) 2632 return true; 2633 2634 if (getLexer().isNot(AsmToken::EndOfStatement)) 2635 return TokError("unexpected token in '.org' directive"); 2636 } 2637 2638 Lex(); 2639 2640 // Only limited forms of relocatable expressions are accepted here, it 2641 // has to be relative to the current section. The streamer will return 2642 // 'true' if the expression wasn't evaluatable. 2643 if (getStreamer().EmitValueToOffset(Offset, FillExpr)) 2644 return Error(Loc, "expected assembly-time absolute expression"); 2645 2646 return false; 2647 } 2648 2649 /// parseDirectiveAlign 2650 /// ::= {.align, ...} expression [ , expression [ , expression ]] 2651 bool AsmParser::parseDirectiveAlign(bool IsPow2, unsigned ValueSize) { 2652 checkForValidSection(); 2653 2654 SMLoc AlignmentLoc = getLexer().getLoc(); 2655 int64_t Alignment; 2656 if (parseAbsoluteExpression(Alignment)) 2657 return true; 2658 2659 SMLoc MaxBytesLoc; 2660 bool HasFillExpr = false; 2661 int64_t FillExpr = 0; 2662 int64_t MaxBytesToFill = 0; 2663 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2664 if (getLexer().isNot(AsmToken::Comma)) 2665 return TokError("unexpected token in directive"); 2666 Lex(); 2667 2668 // The fill expression can be omitted while specifying a maximum number of 2669 // alignment bytes, e.g: 2670 // .align 3,,4 2671 if (getLexer().isNot(AsmToken::Comma)) { 2672 HasFillExpr = true; 2673 if (parseAbsoluteExpression(FillExpr)) 2674 return true; 2675 } 2676 2677 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2678 if (getLexer().isNot(AsmToken::Comma)) 2679 return TokError("unexpected token in directive"); 2680 Lex(); 2681 2682 MaxBytesLoc = getLexer().getLoc(); 2683 if (parseAbsoluteExpression(MaxBytesToFill)) 2684 return true; 2685 2686 if (getLexer().isNot(AsmToken::EndOfStatement)) 2687 return TokError("unexpected token in directive"); 2688 } 2689 } 2690 2691 Lex(); 2692 2693 if (!HasFillExpr) 2694 FillExpr = 0; 2695 2696 // Compute alignment in bytes. 2697 if (IsPow2) { 2698 // FIXME: Diagnose overflow. 2699 if (Alignment >= 32) { 2700 Error(AlignmentLoc, "invalid alignment value"); 2701 Alignment = 31; 2702 } 2703 2704 Alignment = 1ULL << Alignment; 2705 } else { 2706 // Reject alignments that aren't a power of two, for gas compatibility. 2707 if (!isPowerOf2_64(Alignment)) 2708 Error(AlignmentLoc, "alignment must be a power of 2"); 2709 } 2710 2711 // Diagnose non-sensical max bytes to align. 2712 if (MaxBytesLoc.isValid()) { 2713 if (MaxBytesToFill < 1) { 2714 Error(MaxBytesLoc, "alignment directive can never be satisfied in this " 2715 "many bytes, ignoring maximum bytes expression"); 2716 MaxBytesToFill = 0; 2717 } 2718 2719 if (MaxBytesToFill >= Alignment) { 2720 Warning(MaxBytesLoc, "maximum bytes expression exceeds alignment and " 2721 "has no effect"); 2722 MaxBytesToFill = 0; 2723 } 2724 } 2725 2726 // Check whether we should use optimal code alignment for this .align 2727 // directive. 2728 const MCSection *Section = getStreamer().getCurrentSection().first; 2729 assert(Section && "must have section to emit alignment"); 2730 bool UseCodeAlign = Section->UseCodeAlign(); 2731 if ((!HasFillExpr || Lexer.getMAI().getTextAlignFillValue() == FillExpr) && 2732 ValueSize == 1 && UseCodeAlign) { 2733 getStreamer().EmitCodeAlignment(Alignment, MaxBytesToFill); 2734 } else { 2735 // FIXME: Target specific behavior about how the "extra" bytes are filled. 2736 getStreamer().EmitValueToAlignment(Alignment, FillExpr, ValueSize, 2737 MaxBytesToFill); 2738 } 2739 2740 return false; 2741 } 2742 2743 /// parseDirectiveFile 2744 /// ::= .file [number] filename 2745 /// ::= .file number directory filename 2746 bool AsmParser::parseDirectiveFile(SMLoc DirectiveLoc) { 2747 // FIXME: I'm not sure what this is. 2748 int64_t FileNumber = -1; 2749 SMLoc FileNumberLoc = getLexer().getLoc(); 2750 if (getLexer().is(AsmToken::Integer)) { 2751 FileNumber = getTok().getIntVal(); 2752 Lex(); 2753 2754 if (FileNumber < 1) 2755 return TokError("file number less than one"); 2756 } 2757 2758 if (getLexer().isNot(AsmToken::String)) 2759 return TokError("unexpected token in '.file' directive"); 2760 2761 // Usually the directory and filename together, otherwise just the directory. 2762 // Allow the strings to have escaped octal character sequence. 2763 std::string Path = getTok().getString(); 2764 if (parseEscapedString(Path)) 2765 return true; 2766 Lex(); 2767 2768 StringRef Directory; 2769 StringRef Filename; 2770 std::string FilenameData; 2771 if (getLexer().is(AsmToken::String)) { 2772 if (FileNumber == -1) 2773 return TokError("explicit path specified, but no file number"); 2774 if (parseEscapedString(FilenameData)) 2775 return true; 2776 Filename = FilenameData; 2777 Directory = Path; 2778 Lex(); 2779 } else { 2780 Filename = Path; 2781 } 2782 2783 if (getLexer().isNot(AsmToken::EndOfStatement)) 2784 return TokError("unexpected token in '.file' directive"); 2785 2786 if (FileNumber == -1) 2787 getStreamer().EmitFileDirective(Filename); 2788 else { 2789 if (getContext().getGenDwarfForAssembly() == true) 2790 Error(DirectiveLoc, 2791 "input can't have .file dwarf directives when -g is " 2792 "used to generate dwarf debug info for assembly code"); 2793 2794 if (getStreamer().EmitDwarfFileDirective(FileNumber, Directory, Filename) == 2795 0) 2796 Error(FileNumberLoc, "file number already allocated"); 2797 } 2798 2799 return false; 2800 } 2801 2802 /// parseDirectiveLine 2803 /// ::= .line [number] 2804 bool AsmParser::parseDirectiveLine() { 2805 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2806 if (getLexer().isNot(AsmToken::Integer)) 2807 return TokError("unexpected token in '.line' directive"); 2808 2809 int64_t LineNumber = getTok().getIntVal(); 2810 (void)LineNumber; 2811 Lex(); 2812 2813 // FIXME: Do something with the .line. 2814 } 2815 2816 if (getLexer().isNot(AsmToken::EndOfStatement)) 2817 return TokError("unexpected token in '.line' directive"); 2818 2819 return false; 2820 } 2821 2822 /// parseDirectiveLoc 2823 /// ::= .loc FileNumber [LineNumber] [ColumnPos] [basic_block] [prologue_end] 2824 /// [epilogue_begin] [is_stmt VALUE] [isa VALUE] 2825 /// The first number is a file number, must have been previously assigned with 2826 /// a .file directive, the second number is the line number and optionally the 2827 /// third number is a column position (zero if not specified). The remaining 2828 /// optional items are .loc sub-directives. 2829 bool AsmParser::parseDirectiveLoc() { 2830 if (getLexer().isNot(AsmToken::Integer)) 2831 return TokError("unexpected token in '.loc' directive"); 2832 int64_t FileNumber = getTok().getIntVal(); 2833 if (FileNumber < 1) 2834 return TokError("file number less than one in '.loc' directive"); 2835 if (!getContext().isValidDwarfFileNumber(FileNumber)) 2836 return TokError("unassigned file number in '.loc' directive"); 2837 Lex(); 2838 2839 int64_t LineNumber = 0; 2840 if (getLexer().is(AsmToken::Integer)) { 2841 LineNumber = getTok().getIntVal(); 2842 if (LineNumber < 0) 2843 return TokError("line number less than zero in '.loc' directive"); 2844 Lex(); 2845 } 2846 2847 int64_t ColumnPos = 0; 2848 if (getLexer().is(AsmToken::Integer)) { 2849 ColumnPos = getTok().getIntVal(); 2850 if (ColumnPos < 0) 2851 return TokError("column position less than zero in '.loc' directive"); 2852 Lex(); 2853 } 2854 2855 unsigned Flags = DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0; 2856 unsigned Isa = 0; 2857 int64_t Discriminator = 0; 2858 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2859 for (;;) { 2860 if (getLexer().is(AsmToken::EndOfStatement)) 2861 break; 2862 2863 StringRef Name; 2864 SMLoc Loc = getTok().getLoc(); 2865 if (parseIdentifier(Name)) 2866 return TokError("unexpected token in '.loc' directive"); 2867 2868 if (Name == "basic_block") 2869 Flags |= DWARF2_FLAG_BASIC_BLOCK; 2870 else if (Name == "prologue_end") 2871 Flags |= DWARF2_FLAG_PROLOGUE_END; 2872 else if (Name == "epilogue_begin") 2873 Flags |= DWARF2_FLAG_EPILOGUE_BEGIN; 2874 else if (Name == "is_stmt") { 2875 Loc = getTok().getLoc(); 2876 const MCExpr *Value; 2877 if (parseExpression(Value)) 2878 return true; 2879 // The expression must be the constant 0 or 1. 2880 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2881 int Value = MCE->getValue(); 2882 if (Value == 0) 2883 Flags &= ~DWARF2_FLAG_IS_STMT; 2884 else if (Value == 1) 2885 Flags |= DWARF2_FLAG_IS_STMT; 2886 else 2887 return Error(Loc, "is_stmt value not 0 or 1"); 2888 } else { 2889 return Error(Loc, "is_stmt value not the constant value of 0 or 1"); 2890 } 2891 } else if (Name == "isa") { 2892 Loc = getTok().getLoc(); 2893 const MCExpr *Value; 2894 if (parseExpression(Value)) 2895 return true; 2896 // The expression must be a constant greater or equal to 0. 2897 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2898 int Value = MCE->getValue(); 2899 if (Value < 0) 2900 return Error(Loc, "isa number less than zero"); 2901 Isa = Value; 2902 } else { 2903 return Error(Loc, "isa number not a constant value"); 2904 } 2905 } else if (Name == "discriminator") { 2906 if (parseAbsoluteExpression(Discriminator)) 2907 return true; 2908 } else { 2909 return Error(Loc, "unknown sub-directive in '.loc' directive"); 2910 } 2911 2912 if (getLexer().is(AsmToken::EndOfStatement)) 2913 break; 2914 } 2915 } 2916 2917 getStreamer().EmitDwarfLocDirective(FileNumber, LineNumber, ColumnPos, Flags, 2918 Isa, Discriminator, StringRef()); 2919 2920 return false; 2921 } 2922 2923 /// parseDirectiveStabs 2924 /// ::= .stabs string, number, number, number 2925 bool AsmParser::parseDirectiveStabs() { 2926 return TokError("unsupported directive '.stabs'"); 2927 } 2928 2929 /// parseDirectiveCFISections 2930 /// ::= .cfi_sections section [, section] 2931 bool AsmParser::parseDirectiveCFISections() { 2932 StringRef Name; 2933 bool EH = false; 2934 bool Debug = false; 2935 2936 if (parseIdentifier(Name)) 2937 return TokError("Expected an identifier"); 2938 2939 if (Name == ".eh_frame") 2940 EH = true; 2941 else if (Name == ".debug_frame") 2942 Debug = true; 2943 2944 if (getLexer().is(AsmToken::Comma)) { 2945 Lex(); 2946 2947 if (parseIdentifier(Name)) 2948 return TokError("Expected an identifier"); 2949 2950 if (Name == ".eh_frame") 2951 EH = true; 2952 else if (Name == ".debug_frame") 2953 Debug = true; 2954 } 2955 2956 getStreamer().EmitCFISections(EH, Debug); 2957 return false; 2958 } 2959 2960 /// parseDirectiveCFIStartProc 2961 /// ::= .cfi_startproc [simple] 2962 bool AsmParser::parseDirectiveCFIStartProc() { 2963 StringRef Simple; 2964 if (getLexer().isNot(AsmToken::EndOfStatement)) 2965 if (parseIdentifier(Simple) || Simple != "simple") 2966 return TokError("unexpected token in .cfi_startproc directive"); 2967 2968 getStreamer().EmitCFIStartProc(!Simple.empty()); 2969 return false; 2970 } 2971 2972 /// parseDirectiveCFIEndProc 2973 /// ::= .cfi_endproc 2974 bool AsmParser::parseDirectiveCFIEndProc() { 2975 getStreamer().EmitCFIEndProc(); 2976 return false; 2977 } 2978 2979 /// \brief parse register name or number. 2980 bool AsmParser::parseRegisterOrRegisterNumber(int64_t &Register, 2981 SMLoc DirectiveLoc) { 2982 unsigned RegNo; 2983 2984 if (getLexer().isNot(AsmToken::Integer)) { 2985 if (getTargetParser().ParseRegister(RegNo, DirectiveLoc, DirectiveLoc)) 2986 return true; 2987 Register = getContext().getRegisterInfo()->getDwarfRegNum(RegNo, true); 2988 } else 2989 return parseAbsoluteExpression(Register); 2990 2991 return false; 2992 } 2993 2994 /// parseDirectiveCFIDefCfa 2995 /// ::= .cfi_def_cfa register, offset 2996 bool AsmParser::parseDirectiveCFIDefCfa(SMLoc DirectiveLoc) { 2997 int64_t Register = 0; 2998 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 2999 return true; 3000 3001 if (getLexer().isNot(AsmToken::Comma)) 3002 return TokError("unexpected token in directive"); 3003 Lex(); 3004 3005 int64_t Offset = 0; 3006 if (parseAbsoluteExpression(Offset)) 3007 return true; 3008 3009 getStreamer().EmitCFIDefCfa(Register, Offset); 3010 return false; 3011 } 3012 3013 /// parseDirectiveCFIDefCfaOffset 3014 /// ::= .cfi_def_cfa_offset offset 3015 bool AsmParser::parseDirectiveCFIDefCfaOffset() { 3016 int64_t Offset = 0; 3017 if (parseAbsoluteExpression(Offset)) 3018 return true; 3019 3020 getStreamer().EmitCFIDefCfaOffset(Offset); 3021 return false; 3022 } 3023 3024 /// parseDirectiveCFIRegister 3025 /// ::= .cfi_register register, register 3026 bool AsmParser::parseDirectiveCFIRegister(SMLoc DirectiveLoc) { 3027 int64_t Register1 = 0; 3028 if (parseRegisterOrRegisterNumber(Register1, DirectiveLoc)) 3029 return true; 3030 3031 if (getLexer().isNot(AsmToken::Comma)) 3032 return TokError("unexpected token in directive"); 3033 Lex(); 3034 3035 int64_t Register2 = 0; 3036 if (parseRegisterOrRegisterNumber(Register2, DirectiveLoc)) 3037 return true; 3038 3039 getStreamer().EmitCFIRegister(Register1, Register2); 3040 return false; 3041 } 3042 3043 /// parseDirectiveCFIWindowSave 3044 /// ::= .cfi_window_save 3045 bool AsmParser::parseDirectiveCFIWindowSave() { 3046 getStreamer().EmitCFIWindowSave(); 3047 return false; 3048 } 3049 3050 /// parseDirectiveCFIAdjustCfaOffset 3051 /// ::= .cfi_adjust_cfa_offset adjustment 3052 bool AsmParser::parseDirectiveCFIAdjustCfaOffset() { 3053 int64_t Adjustment = 0; 3054 if (parseAbsoluteExpression(Adjustment)) 3055 return true; 3056 3057 getStreamer().EmitCFIAdjustCfaOffset(Adjustment); 3058 return false; 3059 } 3060 3061 /// parseDirectiveCFIDefCfaRegister 3062 /// ::= .cfi_def_cfa_register register 3063 bool AsmParser::parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc) { 3064 int64_t Register = 0; 3065 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3066 return true; 3067 3068 getStreamer().EmitCFIDefCfaRegister(Register); 3069 return false; 3070 } 3071 3072 /// parseDirectiveCFIOffset 3073 /// ::= .cfi_offset register, offset 3074 bool AsmParser::parseDirectiveCFIOffset(SMLoc DirectiveLoc) { 3075 int64_t Register = 0; 3076 int64_t Offset = 0; 3077 3078 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3079 return true; 3080 3081 if (getLexer().isNot(AsmToken::Comma)) 3082 return TokError("unexpected token in directive"); 3083 Lex(); 3084 3085 if (parseAbsoluteExpression(Offset)) 3086 return true; 3087 3088 getStreamer().EmitCFIOffset(Register, Offset); 3089 return false; 3090 } 3091 3092 /// parseDirectiveCFIRelOffset 3093 /// ::= .cfi_rel_offset register, offset 3094 bool AsmParser::parseDirectiveCFIRelOffset(SMLoc DirectiveLoc) { 3095 int64_t Register = 0; 3096 3097 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3098 return true; 3099 3100 if (getLexer().isNot(AsmToken::Comma)) 3101 return TokError("unexpected token in directive"); 3102 Lex(); 3103 3104 int64_t Offset = 0; 3105 if (parseAbsoluteExpression(Offset)) 3106 return true; 3107 3108 getStreamer().EmitCFIRelOffset(Register, Offset); 3109 return false; 3110 } 3111 3112 static bool isValidEncoding(int64_t Encoding) { 3113 if (Encoding & ~0xff) 3114 return false; 3115 3116 if (Encoding == dwarf::DW_EH_PE_omit) 3117 return true; 3118 3119 const unsigned Format = Encoding & 0xf; 3120 if (Format != dwarf::DW_EH_PE_absptr && Format != dwarf::DW_EH_PE_udata2 && 3121 Format != dwarf::DW_EH_PE_udata4 && Format != dwarf::DW_EH_PE_udata8 && 3122 Format != dwarf::DW_EH_PE_sdata2 && Format != dwarf::DW_EH_PE_sdata4 && 3123 Format != dwarf::DW_EH_PE_sdata8 && Format != dwarf::DW_EH_PE_signed) 3124 return false; 3125 3126 const unsigned Application = Encoding & 0x70; 3127 if (Application != dwarf::DW_EH_PE_absptr && 3128 Application != dwarf::DW_EH_PE_pcrel) 3129 return false; 3130 3131 return true; 3132 } 3133 3134 /// parseDirectiveCFIPersonalityOrLsda 3135 /// IsPersonality true for cfi_personality, false for cfi_lsda 3136 /// ::= .cfi_personality encoding, [symbol_name] 3137 /// ::= .cfi_lsda encoding, [symbol_name] 3138 bool AsmParser::parseDirectiveCFIPersonalityOrLsda(bool IsPersonality) { 3139 int64_t Encoding = 0; 3140 if (parseAbsoluteExpression(Encoding)) 3141 return true; 3142 if (Encoding == dwarf::DW_EH_PE_omit) 3143 return false; 3144 3145 if (!isValidEncoding(Encoding)) 3146 return TokError("unsupported encoding."); 3147 3148 if (getLexer().isNot(AsmToken::Comma)) 3149 return TokError("unexpected token in directive"); 3150 Lex(); 3151 3152 StringRef Name; 3153 if (parseIdentifier(Name)) 3154 return TokError("expected identifier in directive"); 3155 3156 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name); 3157 3158 if (IsPersonality) 3159 getStreamer().EmitCFIPersonality(Sym, Encoding); 3160 else 3161 getStreamer().EmitCFILsda(Sym, Encoding); 3162 return false; 3163 } 3164 3165 /// parseDirectiveCFIRememberState 3166 /// ::= .cfi_remember_state 3167 bool AsmParser::parseDirectiveCFIRememberState() { 3168 getStreamer().EmitCFIRememberState(); 3169 return false; 3170 } 3171 3172 /// parseDirectiveCFIRestoreState 3173 /// ::= .cfi_remember_state 3174 bool AsmParser::parseDirectiveCFIRestoreState() { 3175 getStreamer().EmitCFIRestoreState(); 3176 return false; 3177 } 3178 3179 /// parseDirectiveCFISameValue 3180 /// ::= .cfi_same_value register 3181 bool AsmParser::parseDirectiveCFISameValue(SMLoc DirectiveLoc) { 3182 int64_t Register = 0; 3183 3184 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3185 return true; 3186 3187 getStreamer().EmitCFISameValue(Register); 3188 return false; 3189 } 3190 3191 /// parseDirectiveCFIRestore 3192 /// ::= .cfi_restore register 3193 bool AsmParser::parseDirectiveCFIRestore(SMLoc DirectiveLoc) { 3194 int64_t Register = 0; 3195 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3196 return true; 3197 3198 getStreamer().EmitCFIRestore(Register); 3199 return false; 3200 } 3201 3202 /// parseDirectiveCFIEscape 3203 /// ::= .cfi_escape expression[,...] 3204 bool AsmParser::parseDirectiveCFIEscape() { 3205 std::string Values; 3206 int64_t CurrValue; 3207 if (parseAbsoluteExpression(CurrValue)) 3208 return true; 3209 3210 Values.push_back((uint8_t)CurrValue); 3211 3212 while (getLexer().is(AsmToken::Comma)) { 3213 Lex(); 3214 3215 if (parseAbsoluteExpression(CurrValue)) 3216 return true; 3217 3218 Values.push_back((uint8_t)CurrValue); 3219 } 3220 3221 getStreamer().EmitCFIEscape(Values); 3222 return false; 3223 } 3224 3225 /// parseDirectiveCFISignalFrame 3226 /// ::= .cfi_signal_frame 3227 bool AsmParser::parseDirectiveCFISignalFrame() { 3228 if (getLexer().isNot(AsmToken::EndOfStatement)) 3229 return Error(getLexer().getLoc(), 3230 "unexpected token in '.cfi_signal_frame'"); 3231 3232 getStreamer().EmitCFISignalFrame(); 3233 return false; 3234 } 3235 3236 /// parseDirectiveCFIUndefined 3237 /// ::= .cfi_undefined register 3238 bool AsmParser::parseDirectiveCFIUndefined(SMLoc DirectiveLoc) { 3239 int64_t Register = 0; 3240 3241 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3242 return true; 3243 3244 getStreamer().EmitCFIUndefined(Register); 3245 return false; 3246 } 3247 3248 /// parseDirectiveMacrosOnOff 3249 /// ::= .macros_on 3250 /// ::= .macros_off 3251 bool AsmParser::parseDirectiveMacrosOnOff(StringRef Directive) { 3252 if (getLexer().isNot(AsmToken::EndOfStatement)) 3253 return Error(getLexer().getLoc(), 3254 "unexpected token in '" + Directive + "' directive"); 3255 3256 setMacrosEnabled(Directive == ".macros_on"); 3257 return false; 3258 } 3259 3260 /// parseDirectiveMacro 3261 /// ::= .macro name[,] [parameters] 3262 bool AsmParser::parseDirectiveMacro(SMLoc DirectiveLoc) { 3263 StringRef Name; 3264 if (parseIdentifier(Name)) 3265 return TokError("expected identifier in '.macro' directive"); 3266 3267 if (getLexer().is(AsmToken::Comma)) 3268 Lex(); 3269 3270 MCAsmMacroParameters Parameters; 3271 while (getLexer().isNot(AsmToken::EndOfStatement)) { 3272 3273 if (Parameters.size() && Parameters.back().Vararg) 3274 return Error(Lexer.getLoc(), 3275 "Vararg parameter '" + Parameters.back().Name + 3276 "' should be last one in the list of parameters."); 3277 3278 MCAsmMacroParameter Parameter; 3279 if (parseIdentifier(Parameter.Name)) 3280 return TokError("expected identifier in '.macro' directive"); 3281 3282 if (Lexer.is(AsmToken::Colon)) { 3283 Lex(); // consume ':' 3284 3285 SMLoc QualLoc; 3286 StringRef Qualifier; 3287 3288 QualLoc = Lexer.getLoc(); 3289 if (parseIdentifier(Qualifier)) 3290 return Error(QualLoc, "missing parameter qualifier for " 3291 "'" + Parameter.Name + "' in macro '" + Name + "'"); 3292 3293 if (Qualifier == "req") 3294 Parameter.Required = true; 3295 else if (Qualifier == "vararg" && !IsDarwin) 3296 Parameter.Vararg = true; 3297 else 3298 return Error(QualLoc, Qualifier + " is not a valid parameter qualifier " 3299 "for '" + Parameter.Name + "' in macro '" + Name + "'"); 3300 } 3301 3302 if (getLexer().is(AsmToken::Equal)) { 3303 Lex(); 3304 3305 SMLoc ParamLoc; 3306 3307 ParamLoc = Lexer.getLoc(); 3308 if (parseMacroArgument(Parameter.Value, /*Vararg=*/false )) 3309 return true; 3310 3311 if (Parameter.Required) 3312 Warning(ParamLoc, "pointless default value for required parameter " 3313 "'" + Parameter.Name + "' in macro '" + Name + "'"); 3314 } 3315 3316 Parameters.push_back(Parameter); 3317 3318 if (getLexer().is(AsmToken::Comma)) 3319 Lex(); 3320 } 3321 3322 // Eat the end of statement. 3323 Lex(); 3324 3325 AsmToken EndToken, StartToken = getTok(); 3326 unsigned MacroDepth = 0; 3327 3328 // Lex the macro definition. 3329 for (;;) { 3330 // Check whether we have reached the end of the file. 3331 if (getLexer().is(AsmToken::Eof)) 3332 return Error(DirectiveLoc, "no matching '.endmacro' in definition"); 3333 3334 // Otherwise, check whether we have reach the .endmacro. 3335 if (getLexer().is(AsmToken::Identifier)) { 3336 if (getTok().getIdentifier() == ".endm" || 3337 getTok().getIdentifier() == ".endmacro") { 3338 if (MacroDepth == 0) { // Outermost macro. 3339 EndToken = getTok(); 3340 Lex(); 3341 if (getLexer().isNot(AsmToken::EndOfStatement)) 3342 return TokError("unexpected token in '" + EndToken.getIdentifier() + 3343 "' directive"); 3344 break; 3345 } else { 3346 // Otherwise we just found the end of an inner macro. 3347 --MacroDepth; 3348 } 3349 } else if (getTok().getIdentifier() == ".macro") { 3350 // We allow nested macros. Those aren't instantiated until the outermost 3351 // macro is expanded so just ignore them for now. 3352 ++MacroDepth; 3353 } 3354 } 3355 3356 // Otherwise, scan til the end of the statement. 3357 eatToEndOfStatement(); 3358 } 3359 3360 if (lookupMacro(Name)) { 3361 return Error(DirectiveLoc, "macro '" + Name + "' is already defined"); 3362 } 3363 3364 const char *BodyStart = StartToken.getLoc().getPointer(); 3365 const char *BodyEnd = EndToken.getLoc().getPointer(); 3366 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 3367 checkForBadMacro(DirectiveLoc, Name, Body, Parameters); 3368 defineMacro(Name, MCAsmMacro(Name, Body, Parameters)); 3369 return false; 3370 } 3371 3372 /// checkForBadMacro 3373 /// 3374 /// With the support added for named parameters there may be code out there that 3375 /// is transitioning from positional parameters. In versions of gas that did 3376 /// not support named parameters they would be ignored on the macro definition. 3377 /// But to support both styles of parameters this is not possible so if a macro 3378 /// definition has named parameters but does not use them and has what appears 3379 /// to be positional parameters, strings like $1, $2, ... and $n, then issue a 3380 /// warning that the positional parameter found in body which have no effect. 3381 /// Hoping the developer will either remove the named parameters from the macro 3382 /// definition so the positional parameters get used if that was what was 3383 /// intended or change the macro to use the named parameters. It is possible 3384 /// this warning will trigger when the none of the named parameters are used 3385 /// and the strings like $1 are infact to simply to be passed trough unchanged. 3386 void AsmParser::checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, 3387 StringRef Body, 3388 ArrayRef<MCAsmMacroParameter> Parameters) { 3389 // If this macro is not defined with named parameters the warning we are 3390 // checking for here doesn't apply. 3391 unsigned NParameters = Parameters.size(); 3392 if (NParameters == 0) 3393 return; 3394 3395 bool NamedParametersFound = false; 3396 bool PositionalParametersFound = false; 3397 3398 // Look at the body of the macro for use of both the named parameters and what 3399 // are likely to be positional parameters. This is what expandMacro() is 3400 // doing when it finds the parameters in the body. 3401 while (!Body.empty()) { 3402 // Scan for the next possible parameter. 3403 std::size_t End = Body.size(), Pos = 0; 3404 for (; Pos != End; ++Pos) { 3405 // Check for a substitution or escape. 3406 // This macro is defined with parameters, look for \foo, \bar, etc. 3407 if (Body[Pos] == '\\' && Pos + 1 != End) 3408 break; 3409 3410 // This macro should have parameters, but look for $0, $1, ..., $n too. 3411 if (Body[Pos] != '$' || Pos + 1 == End) 3412 continue; 3413 char Next = Body[Pos + 1]; 3414 if (Next == '$' || Next == 'n' || 3415 isdigit(static_cast<unsigned char>(Next))) 3416 break; 3417 } 3418 3419 // Check if we reached the end. 3420 if (Pos == End) 3421 break; 3422 3423 if (Body[Pos] == '$') { 3424 switch (Body[Pos + 1]) { 3425 // $$ => $ 3426 case '$': 3427 break; 3428 3429 // $n => number of arguments 3430 case 'n': 3431 PositionalParametersFound = true; 3432 break; 3433 3434 // $[0-9] => argument 3435 default: { 3436 PositionalParametersFound = true; 3437 break; 3438 } 3439 } 3440 Pos += 2; 3441 } else { 3442 unsigned I = Pos + 1; 3443 while (isIdentifierChar(Body[I]) && I + 1 != End) 3444 ++I; 3445 3446 const char *Begin = Body.data() + Pos + 1; 3447 StringRef Argument(Begin, I - (Pos + 1)); 3448 unsigned Index = 0; 3449 for (; Index < NParameters; ++Index) 3450 if (Parameters[Index].Name == Argument) 3451 break; 3452 3453 if (Index == NParameters) { 3454 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')') 3455 Pos += 3; 3456 else { 3457 Pos = I; 3458 } 3459 } else { 3460 NamedParametersFound = true; 3461 Pos += 1 + Argument.size(); 3462 } 3463 } 3464 // Update the scan point. 3465 Body = Body.substr(Pos); 3466 } 3467 3468 if (!NamedParametersFound && PositionalParametersFound) 3469 Warning(DirectiveLoc, "macro defined with named parameters which are not " 3470 "used in macro body, possible positional parameter " 3471 "found in body which will have no effect"); 3472 } 3473 3474 /// parseDirectiveEndMacro 3475 /// ::= .endm 3476 /// ::= .endmacro 3477 bool AsmParser::parseDirectiveEndMacro(StringRef Directive) { 3478 if (getLexer().isNot(AsmToken::EndOfStatement)) 3479 return TokError("unexpected token in '" + Directive + "' directive"); 3480 3481 // If we are inside a macro instantiation, terminate the current 3482 // instantiation. 3483 if (isInsideMacroInstantiation()) { 3484 handleMacroExit(); 3485 return false; 3486 } 3487 3488 // Otherwise, this .endmacro is a stray entry in the file; well formed 3489 // .endmacro directives are handled during the macro definition parsing. 3490 return TokError("unexpected '" + Directive + "' in file, " 3491 "no current macro definition"); 3492 } 3493 3494 /// parseDirectivePurgeMacro 3495 /// ::= .purgem 3496 bool AsmParser::parseDirectivePurgeMacro(SMLoc DirectiveLoc) { 3497 StringRef Name; 3498 if (parseIdentifier(Name)) 3499 return TokError("expected identifier in '.purgem' directive"); 3500 3501 if (getLexer().isNot(AsmToken::EndOfStatement)) 3502 return TokError("unexpected token in '.purgem' directive"); 3503 3504 if (!lookupMacro(Name)) 3505 return Error(DirectiveLoc, "macro '" + Name + "' is not defined"); 3506 3507 undefineMacro(Name); 3508 return false; 3509 } 3510 3511 /// parseDirectiveBundleAlignMode 3512 /// ::= {.bundle_align_mode} expression 3513 bool AsmParser::parseDirectiveBundleAlignMode() { 3514 checkForValidSection(); 3515 3516 // Expect a single argument: an expression that evaluates to a constant 3517 // in the inclusive range 0-30. 3518 SMLoc ExprLoc = getLexer().getLoc(); 3519 int64_t AlignSizePow2; 3520 if (parseAbsoluteExpression(AlignSizePow2)) 3521 return true; 3522 else if (getLexer().isNot(AsmToken::EndOfStatement)) 3523 return TokError("unexpected token after expression in" 3524 " '.bundle_align_mode' directive"); 3525 else if (AlignSizePow2 < 0 || AlignSizePow2 > 30) 3526 return Error(ExprLoc, 3527 "invalid bundle alignment size (expected between 0 and 30)"); 3528 3529 Lex(); 3530 3531 // Because of AlignSizePow2's verified range we can safely truncate it to 3532 // unsigned. 3533 getStreamer().EmitBundleAlignMode(static_cast<unsigned>(AlignSizePow2)); 3534 return false; 3535 } 3536 3537 /// parseDirectiveBundleLock 3538 /// ::= {.bundle_lock} [align_to_end] 3539 bool AsmParser::parseDirectiveBundleLock() { 3540 checkForValidSection(); 3541 bool AlignToEnd = false; 3542 3543 if (getLexer().isNot(AsmToken::EndOfStatement)) { 3544 StringRef Option; 3545 SMLoc Loc = getTok().getLoc(); 3546 const char *kInvalidOptionError = 3547 "invalid option for '.bundle_lock' directive"; 3548 3549 if (parseIdentifier(Option)) 3550 return Error(Loc, kInvalidOptionError); 3551 3552 if (Option != "align_to_end") 3553 return Error(Loc, kInvalidOptionError); 3554 else if (getLexer().isNot(AsmToken::EndOfStatement)) 3555 return Error(Loc, 3556 "unexpected token after '.bundle_lock' directive option"); 3557 AlignToEnd = true; 3558 } 3559 3560 Lex(); 3561 3562 getStreamer().EmitBundleLock(AlignToEnd); 3563 return false; 3564 } 3565 3566 /// parseDirectiveBundleLock 3567 /// ::= {.bundle_lock} 3568 bool AsmParser::parseDirectiveBundleUnlock() { 3569 checkForValidSection(); 3570 3571 if (getLexer().isNot(AsmToken::EndOfStatement)) 3572 return TokError("unexpected token in '.bundle_unlock' directive"); 3573 Lex(); 3574 3575 getStreamer().EmitBundleUnlock(); 3576 return false; 3577 } 3578 3579 /// parseDirectiveSpace 3580 /// ::= (.skip | .space) expression [ , expression ] 3581 bool AsmParser::parseDirectiveSpace(StringRef IDVal) { 3582 checkForValidSection(); 3583 3584 int64_t NumBytes; 3585 if (parseAbsoluteExpression(NumBytes)) 3586 return true; 3587 3588 int64_t FillExpr = 0; 3589 if (getLexer().isNot(AsmToken::EndOfStatement)) { 3590 if (getLexer().isNot(AsmToken::Comma)) 3591 return TokError("unexpected token in '" + Twine(IDVal) + "' directive"); 3592 Lex(); 3593 3594 if (parseAbsoluteExpression(FillExpr)) 3595 return true; 3596 3597 if (getLexer().isNot(AsmToken::EndOfStatement)) 3598 return TokError("unexpected token in '" + Twine(IDVal) + "' directive"); 3599 } 3600 3601 Lex(); 3602 3603 if (NumBytes <= 0) 3604 return TokError("invalid number of bytes in '" + Twine(IDVal) + 3605 "' directive"); 3606 3607 // FIXME: Sometimes the fill expr is 'nop' if it isn't supplied, instead of 0. 3608 getStreamer().EmitFill(NumBytes, FillExpr); 3609 3610 return false; 3611 } 3612 3613 /// parseDirectiveLEB128 3614 /// ::= (.sleb128 | .uleb128) expression 3615 bool AsmParser::parseDirectiveLEB128(bool Signed) { 3616 checkForValidSection(); 3617 const MCExpr *Value; 3618 3619 if (parseExpression(Value)) 3620 return true; 3621 3622 if (getLexer().isNot(AsmToken::EndOfStatement)) 3623 return TokError("unexpected token in directive"); 3624 3625 if (Signed) 3626 getStreamer().EmitSLEB128Value(Value); 3627 else 3628 getStreamer().EmitULEB128Value(Value); 3629 3630 return false; 3631 } 3632 3633 /// parseDirectiveSymbolAttribute 3634 /// ::= { ".globl", ".weak", ... } [ identifier ( , identifier )* ] 3635 bool AsmParser::parseDirectiveSymbolAttribute(MCSymbolAttr Attr) { 3636 if (getLexer().isNot(AsmToken::EndOfStatement)) { 3637 for (;;) { 3638 StringRef Name; 3639 SMLoc Loc = getTok().getLoc(); 3640 3641 if (parseIdentifier(Name)) 3642 return Error(Loc, "expected identifier in directive"); 3643 3644 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name); 3645 3646 // Assembler local symbols don't make any sense here. Complain loudly. 3647 if (Sym->isTemporary()) 3648 return Error(Loc, "non-local symbol required in directive"); 3649 3650 if (!getStreamer().EmitSymbolAttribute(Sym, Attr)) 3651 return Error(Loc, "unable to emit symbol attribute"); 3652 3653 if (getLexer().is(AsmToken::EndOfStatement)) 3654 break; 3655 3656 if (getLexer().isNot(AsmToken::Comma)) 3657 return TokError("unexpected token in directive"); 3658 Lex(); 3659 } 3660 } 3661 3662 Lex(); 3663 return false; 3664 } 3665 3666 /// parseDirectiveComm 3667 /// ::= ( .comm | .lcomm ) identifier , size_expression [ , align_expression ] 3668 bool AsmParser::parseDirectiveComm(bool IsLocal) { 3669 checkForValidSection(); 3670 3671 SMLoc IDLoc = getLexer().getLoc(); 3672 StringRef Name; 3673 if (parseIdentifier(Name)) 3674 return TokError("expected identifier in directive"); 3675 3676 // Handle the identifier as the key symbol. 3677 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name); 3678 3679 if (getLexer().isNot(AsmToken::Comma)) 3680 return TokError("unexpected token in directive"); 3681 Lex(); 3682 3683 int64_t Size; 3684 SMLoc SizeLoc = getLexer().getLoc(); 3685 if (parseAbsoluteExpression(Size)) 3686 return true; 3687 3688 int64_t Pow2Alignment = 0; 3689 SMLoc Pow2AlignmentLoc; 3690 if (getLexer().is(AsmToken::Comma)) { 3691 Lex(); 3692 Pow2AlignmentLoc = getLexer().getLoc(); 3693 if (parseAbsoluteExpression(Pow2Alignment)) 3694 return true; 3695 3696 LCOMM::LCOMMType LCOMM = Lexer.getMAI().getLCOMMDirectiveAlignmentType(); 3697 if (IsLocal && LCOMM == LCOMM::NoAlignment) 3698 return Error(Pow2AlignmentLoc, "alignment not supported on this target"); 3699 3700 // If this target takes alignments in bytes (not log) validate and convert. 3701 if ((!IsLocal && Lexer.getMAI().getCOMMDirectiveAlignmentIsInBytes()) || 3702 (IsLocal && LCOMM == LCOMM::ByteAlignment)) { 3703 if (!isPowerOf2_64(Pow2Alignment)) 3704 return Error(Pow2AlignmentLoc, "alignment must be a power of 2"); 3705 Pow2Alignment = Log2_64(Pow2Alignment); 3706 } 3707 } 3708 3709 if (getLexer().isNot(AsmToken::EndOfStatement)) 3710 return TokError("unexpected token in '.comm' or '.lcomm' directive"); 3711 3712 Lex(); 3713 3714 // NOTE: a size of zero for a .comm should create a undefined symbol 3715 // but a size of .lcomm creates a bss symbol of size zero. 3716 if (Size < 0) 3717 return Error(SizeLoc, "invalid '.comm' or '.lcomm' directive size, can't " 3718 "be less than zero"); 3719 3720 // NOTE: The alignment in the directive is a power of 2 value, the assembler 3721 // may internally end up wanting an alignment in bytes. 3722 // FIXME: Diagnose overflow. 3723 if (Pow2Alignment < 0) 3724 return Error(Pow2AlignmentLoc, "invalid '.comm' or '.lcomm' directive " 3725 "alignment, can't be less than zero"); 3726 3727 if (!Sym->isUndefined()) 3728 return Error(IDLoc, "invalid symbol redefinition"); 3729 3730 // Create the Symbol as a common or local common with Size and Pow2Alignment 3731 if (IsLocal) { 3732 getStreamer().EmitLocalCommonSymbol(Sym, Size, 1 << Pow2Alignment); 3733 return false; 3734 } 3735 3736 getStreamer().EmitCommonSymbol(Sym, Size, 1 << Pow2Alignment); 3737 return false; 3738 } 3739 3740 /// parseDirectiveAbort 3741 /// ::= .abort [... message ...] 3742 bool AsmParser::parseDirectiveAbort() { 3743 // FIXME: Use loc from directive. 3744 SMLoc Loc = getLexer().getLoc(); 3745 3746 StringRef Str = parseStringToEndOfStatement(); 3747 if (getLexer().isNot(AsmToken::EndOfStatement)) 3748 return TokError("unexpected token in '.abort' directive"); 3749 3750 Lex(); 3751 3752 if (Str.empty()) 3753 Error(Loc, ".abort detected. Assembly stopping."); 3754 else 3755 Error(Loc, ".abort '" + Str + "' detected. Assembly stopping."); 3756 // FIXME: Actually abort assembly here. 3757 3758 return false; 3759 } 3760 3761 /// parseDirectiveInclude 3762 /// ::= .include "filename" 3763 bool AsmParser::parseDirectiveInclude() { 3764 if (getLexer().isNot(AsmToken::String)) 3765 return TokError("expected string in '.include' directive"); 3766 3767 // Allow the strings to have escaped octal character sequence. 3768 std::string Filename; 3769 if (parseEscapedString(Filename)) 3770 return true; 3771 SMLoc IncludeLoc = getLexer().getLoc(); 3772 Lex(); 3773 3774 if (getLexer().isNot(AsmToken::EndOfStatement)) 3775 return TokError("unexpected token in '.include' directive"); 3776 3777 // Attempt to switch the lexer to the included file before consuming the end 3778 // of statement to avoid losing it when we switch. 3779 if (enterIncludeFile(Filename)) { 3780 Error(IncludeLoc, "Could not find include file '" + Filename + "'"); 3781 return true; 3782 } 3783 3784 return false; 3785 } 3786 3787 /// parseDirectiveIncbin 3788 /// ::= .incbin "filename" 3789 bool AsmParser::parseDirectiveIncbin() { 3790 if (getLexer().isNot(AsmToken::String)) 3791 return TokError("expected string in '.incbin' directive"); 3792 3793 // Allow the strings to have escaped octal character sequence. 3794 std::string Filename; 3795 if (parseEscapedString(Filename)) 3796 return true; 3797 SMLoc IncbinLoc = getLexer().getLoc(); 3798 Lex(); 3799 3800 if (getLexer().isNot(AsmToken::EndOfStatement)) 3801 return TokError("unexpected token in '.incbin' directive"); 3802 3803 // Attempt to process the included file. 3804 if (processIncbinFile(Filename)) { 3805 Error(IncbinLoc, "Could not find incbin file '" + Filename + "'"); 3806 return true; 3807 } 3808 3809 return false; 3810 } 3811 3812 /// parseDirectiveIf 3813 /// ::= .if{,eq,ge,gt,le,lt,ne} expression 3814 bool AsmParser::parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind) { 3815 TheCondStack.push_back(TheCondState); 3816 TheCondState.TheCond = AsmCond::IfCond; 3817 if (TheCondState.Ignore) { 3818 eatToEndOfStatement(); 3819 } else { 3820 int64_t ExprValue; 3821 if (parseAbsoluteExpression(ExprValue)) 3822 return true; 3823 3824 if (getLexer().isNot(AsmToken::EndOfStatement)) 3825 return TokError("unexpected token in '.if' directive"); 3826 3827 Lex(); 3828 3829 switch (DirKind) { 3830 default: 3831 llvm_unreachable("unsupported directive"); 3832 case DK_IF: 3833 case DK_IFNE: 3834 break; 3835 case DK_IFEQ: 3836 ExprValue = ExprValue == 0; 3837 break; 3838 case DK_IFGE: 3839 ExprValue = ExprValue >= 0; 3840 break; 3841 case DK_IFGT: 3842 ExprValue = ExprValue > 0; 3843 break; 3844 case DK_IFLE: 3845 ExprValue = ExprValue <= 0; 3846 break; 3847 case DK_IFLT: 3848 ExprValue = ExprValue < 0; 3849 break; 3850 } 3851 3852 TheCondState.CondMet = ExprValue; 3853 TheCondState.Ignore = !TheCondState.CondMet; 3854 } 3855 3856 return false; 3857 } 3858 3859 /// parseDirectiveIfb 3860 /// ::= .ifb string 3861 bool AsmParser::parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank) { 3862 TheCondStack.push_back(TheCondState); 3863 TheCondState.TheCond = AsmCond::IfCond; 3864 3865 if (TheCondState.Ignore) { 3866 eatToEndOfStatement(); 3867 } else { 3868 StringRef Str = parseStringToEndOfStatement(); 3869 3870 if (getLexer().isNot(AsmToken::EndOfStatement)) 3871 return TokError("unexpected token in '.ifb' directive"); 3872 3873 Lex(); 3874 3875 TheCondState.CondMet = ExpectBlank == Str.empty(); 3876 TheCondState.Ignore = !TheCondState.CondMet; 3877 } 3878 3879 return false; 3880 } 3881 3882 /// parseDirectiveIfc 3883 /// ::= .ifc string1, string2 3884 /// ::= .ifnc string1, string2 3885 bool AsmParser::parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual) { 3886 TheCondStack.push_back(TheCondState); 3887 TheCondState.TheCond = AsmCond::IfCond; 3888 3889 if (TheCondState.Ignore) { 3890 eatToEndOfStatement(); 3891 } else { 3892 StringRef Str1 = parseStringToComma(); 3893 3894 if (getLexer().isNot(AsmToken::Comma)) 3895 return TokError("unexpected token in '.ifc' directive"); 3896 3897 Lex(); 3898 3899 StringRef Str2 = parseStringToEndOfStatement(); 3900 3901 if (getLexer().isNot(AsmToken::EndOfStatement)) 3902 return TokError("unexpected token in '.ifc' directive"); 3903 3904 Lex(); 3905 3906 TheCondState.CondMet = ExpectEqual == (Str1.trim() == Str2.trim()); 3907 TheCondState.Ignore = !TheCondState.CondMet; 3908 } 3909 3910 return false; 3911 } 3912 3913 /// parseDirectiveIfeqs 3914 /// ::= .ifeqs string1, string2 3915 bool AsmParser::parseDirectiveIfeqs(SMLoc DirectiveLoc) { 3916 if (Lexer.isNot(AsmToken::String)) { 3917 TokError("expected string parameter for '.ifeqs' directive"); 3918 eatToEndOfStatement(); 3919 return true; 3920 } 3921 3922 StringRef String1 = getTok().getStringContents(); 3923 Lex(); 3924 3925 if (Lexer.isNot(AsmToken::Comma)) { 3926 TokError("expected comma after first string for '.ifeqs' directive"); 3927 eatToEndOfStatement(); 3928 return true; 3929 } 3930 3931 Lex(); 3932 3933 if (Lexer.isNot(AsmToken::String)) { 3934 TokError("expected string parameter for '.ifeqs' directive"); 3935 eatToEndOfStatement(); 3936 return true; 3937 } 3938 3939 StringRef String2 = getTok().getStringContents(); 3940 Lex(); 3941 3942 TheCondStack.push_back(TheCondState); 3943 TheCondState.TheCond = AsmCond::IfCond; 3944 TheCondState.CondMet = String1 == String2; 3945 TheCondState.Ignore = !TheCondState.CondMet; 3946 3947 return false; 3948 } 3949 3950 /// parseDirectiveIfdef 3951 /// ::= .ifdef symbol 3952 bool AsmParser::parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined) { 3953 StringRef Name; 3954 TheCondStack.push_back(TheCondState); 3955 TheCondState.TheCond = AsmCond::IfCond; 3956 3957 if (TheCondState.Ignore) { 3958 eatToEndOfStatement(); 3959 } else { 3960 if (parseIdentifier(Name)) 3961 return TokError("expected identifier after '.ifdef'"); 3962 3963 Lex(); 3964 3965 MCSymbol *Sym = getContext().LookupSymbol(Name); 3966 3967 if (expect_defined) 3968 TheCondState.CondMet = (Sym && !Sym->isUndefined()); 3969 else 3970 TheCondState.CondMet = (!Sym || Sym->isUndefined()); 3971 TheCondState.Ignore = !TheCondState.CondMet; 3972 } 3973 3974 return false; 3975 } 3976 3977 /// parseDirectiveElseIf 3978 /// ::= .elseif expression 3979 bool AsmParser::parseDirectiveElseIf(SMLoc DirectiveLoc) { 3980 if (TheCondState.TheCond != AsmCond::IfCond && 3981 TheCondState.TheCond != AsmCond::ElseIfCond) 3982 Error(DirectiveLoc, "Encountered a .elseif that doesn't follow a .if or " 3983 " an .elseif"); 3984 TheCondState.TheCond = AsmCond::ElseIfCond; 3985 3986 bool LastIgnoreState = false; 3987 if (!TheCondStack.empty()) 3988 LastIgnoreState = TheCondStack.back().Ignore; 3989 if (LastIgnoreState || TheCondState.CondMet) { 3990 TheCondState.Ignore = true; 3991 eatToEndOfStatement(); 3992 } else { 3993 int64_t ExprValue; 3994 if (parseAbsoluteExpression(ExprValue)) 3995 return true; 3996 3997 if (getLexer().isNot(AsmToken::EndOfStatement)) 3998 return TokError("unexpected token in '.elseif' directive"); 3999 4000 Lex(); 4001 TheCondState.CondMet = ExprValue; 4002 TheCondState.Ignore = !TheCondState.CondMet; 4003 } 4004 4005 return false; 4006 } 4007 4008 /// parseDirectiveElse 4009 /// ::= .else 4010 bool AsmParser::parseDirectiveElse(SMLoc DirectiveLoc) { 4011 if (getLexer().isNot(AsmToken::EndOfStatement)) 4012 return TokError("unexpected token in '.else' directive"); 4013 4014 Lex(); 4015 4016 if (TheCondState.TheCond != AsmCond::IfCond && 4017 TheCondState.TheCond != AsmCond::ElseIfCond) 4018 Error(DirectiveLoc, "Encountered a .else that doesn't follow a .if or an " 4019 ".elseif"); 4020 TheCondState.TheCond = AsmCond::ElseCond; 4021 bool LastIgnoreState = false; 4022 if (!TheCondStack.empty()) 4023 LastIgnoreState = TheCondStack.back().Ignore; 4024 if (LastIgnoreState || TheCondState.CondMet) 4025 TheCondState.Ignore = true; 4026 else 4027 TheCondState.Ignore = false; 4028 4029 return false; 4030 } 4031 4032 /// parseDirectiveEnd 4033 /// ::= .end 4034 bool AsmParser::parseDirectiveEnd(SMLoc DirectiveLoc) { 4035 if (getLexer().isNot(AsmToken::EndOfStatement)) 4036 return TokError("unexpected token in '.end' directive"); 4037 4038 Lex(); 4039 4040 while (Lexer.isNot(AsmToken::Eof)) 4041 Lex(); 4042 4043 return false; 4044 } 4045 4046 /// parseDirectiveError 4047 /// ::= .err 4048 /// ::= .error [string] 4049 bool AsmParser::parseDirectiveError(SMLoc L, bool WithMessage) { 4050 if (!TheCondStack.empty()) { 4051 if (TheCondStack.back().Ignore) { 4052 eatToEndOfStatement(); 4053 return false; 4054 } 4055 } 4056 4057 if (!WithMessage) 4058 return Error(L, ".err encountered"); 4059 4060 StringRef Message = ".error directive invoked in source file"; 4061 if (Lexer.isNot(AsmToken::EndOfStatement)) { 4062 if (Lexer.isNot(AsmToken::String)) { 4063 TokError(".error argument must be a string"); 4064 eatToEndOfStatement(); 4065 return true; 4066 } 4067 4068 Message = getTok().getStringContents(); 4069 Lex(); 4070 } 4071 4072 Error(L, Message); 4073 return true; 4074 } 4075 4076 /// parseDirectiveEndIf 4077 /// ::= .endif 4078 bool AsmParser::parseDirectiveEndIf(SMLoc DirectiveLoc) { 4079 if (getLexer().isNot(AsmToken::EndOfStatement)) 4080 return TokError("unexpected token in '.endif' directive"); 4081 4082 Lex(); 4083 4084 if ((TheCondState.TheCond == AsmCond::NoCond) || TheCondStack.empty()) 4085 Error(DirectiveLoc, "Encountered a .endif that doesn't follow a .if or " 4086 ".else"); 4087 if (!TheCondStack.empty()) { 4088 TheCondState = TheCondStack.back(); 4089 TheCondStack.pop_back(); 4090 } 4091 4092 return false; 4093 } 4094 4095 void AsmParser::initializeDirectiveKindMap() { 4096 DirectiveKindMap[".set"] = DK_SET; 4097 DirectiveKindMap[".equ"] = DK_EQU; 4098 DirectiveKindMap[".equiv"] = DK_EQUIV; 4099 DirectiveKindMap[".ascii"] = DK_ASCII; 4100 DirectiveKindMap[".asciz"] = DK_ASCIZ; 4101 DirectiveKindMap[".string"] = DK_STRING; 4102 DirectiveKindMap[".byte"] = DK_BYTE; 4103 DirectiveKindMap[".short"] = DK_SHORT; 4104 DirectiveKindMap[".value"] = DK_VALUE; 4105 DirectiveKindMap[".2byte"] = DK_2BYTE; 4106 DirectiveKindMap[".long"] = DK_LONG; 4107 DirectiveKindMap[".int"] = DK_INT; 4108 DirectiveKindMap[".4byte"] = DK_4BYTE; 4109 DirectiveKindMap[".quad"] = DK_QUAD; 4110 DirectiveKindMap[".8byte"] = DK_8BYTE; 4111 DirectiveKindMap[".octa"] = DK_OCTA; 4112 DirectiveKindMap[".single"] = DK_SINGLE; 4113 DirectiveKindMap[".float"] = DK_FLOAT; 4114 DirectiveKindMap[".double"] = DK_DOUBLE; 4115 DirectiveKindMap[".align"] = DK_ALIGN; 4116 DirectiveKindMap[".align32"] = DK_ALIGN32; 4117 DirectiveKindMap[".balign"] = DK_BALIGN; 4118 DirectiveKindMap[".balignw"] = DK_BALIGNW; 4119 DirectiveKindMap[".balignl"] = DK_BALIGNL; 4120 DirectiveKindMap[".p2align"] = DK_P2ALIGN; 4121 DirectiveKindMap[".p2alignw"] = DK_P2ALIGNW; 4122 DirectiveKindMap[".p2alignl"] = DK_P2ALIGNL; 4123 DirectiveKindMap[".org"] = DK_ORG; 4124 DirectiveKindMap[".fill"] = DK_FILL; 4125 DirectiveKindMap[".zero"] = DK_ZERO; 4126 DirectiveKindMap[".extern"] = DK_EXTERN; 4127 DirectiveKindMap[".globl"] = DK_GLOBL; 4128 DirectiveKindMap[".global"] = DK_GLOBAL; 4129 DirectiveKindMap[".lazy_reference"] = DK_LAZY_REFERENCE; 4130 DirectiveKindMap[".no_dead_strip"] = DK_NO_DEAD_STRIP; 4131 DirectiveKindMap[".symbol_resolver"] = DK_SYMBOL_RESOLVER; 4132 DirectiveKindMap[".private_extern"] = DK_PRIVATE_EXTERN; 4133 DirectiveKindMap[".reference"] = DK_REFERENCE; 4134 DirectiveKindMap[".weak_definition"] = DK_WEAK_DEFINITION; 4135 DirectiveKindMap[".weak_reference"] = DK_WEAK_REFERENCE; 4136 DirectiveKindMap[".weak_def_can_be_hidden"] = DK_WEAK_DEF_CAN_BE_HIDDEN; 4137 DirectiveKindMap[".comm"] = DK_COMM; 4138 DirectiveKindMap[".common"] = DK_COMMON; 4139 DirectiveKindMap[".lcomm"] = DK_LCOMM; 4140 DirectiveKindMap[".abort"] = DK_ABORT; 4141 DirectiveKindMap[".include"] = DK_INCLUDE; 4142 DirectiveKindMap[".incbin"] = DK_INCBIN; 4143 DirectiveKindMap[".code16"] = DK_CODE16; 4144 DirectiveKindMap[".code16gcc"] = DK_CODE16GCC; 4145 DirectiveKindMap[".rept"] = DK_REPT; 4146 DirectiveKindMap[".rep"] = DK_REPT; 4147 DirectiveKindMap[".irp"] = DK_IRP; 4148 DirectiveKindMap[".irpc"] = DK_IRPC; 4149 DirectiveKindMap[".endr"] = DK_ENDR; 4150 DirectiveKindMap[".bundle_align_mode"] = DK_BUNDLE_ALIGN_MODE; 4151 DirectiveKindMap[".bundle_lock"] = DK_BUNDLE_LOCK; 4152 DirectiveKindMap[".bundle_unlock"] = DK_BUNDLE_UNLOCK; 4153 DirectiveKindMap[".if"] = DK_IF; 4154 DirectiveKindMap[".ifeq"] = DK_IFEQ; 4155 DirectiveKindMap[".ifge"] = DK_IFGE; 4156 DirectiveKindMap[".ifgt"] = DK_IFGT; 4157 DirectiveKindMap[".ifle"] = DK_IFLE; 4158 DirectiveKindMap[".iflt"] = DK_IFLT; 4159 DirectiveKindMap[".ifne"] = DK_IFNE; 4160 DirectiveKindMap[".ifb"] = DK_IFB; 4161 DirectiveKindMap[".ifnb"] = DK_IFNB; 4162 DirectiveKindMap[".ifc"] = DK_IFC; 4163 DirectiveKindMap[".ifeqs"] = DK_IFEQS; 4164 DirectiveKindMap[".ifnc"] = DK_IFNC; 4165 DirectiveKindMap[".ifdef"] = DK_IFDEF; 4166 DirectiveKindMap[".ifndef"] = DK_IFNDEF; 4167 DirectiveKindMap[".ifnotdef"] = DK_IFNOTDEF; 4168 DirectiveKindMap[".elseif"] = DK_ELSEIF; 4169 DirectiveKindMap[".else"] = DK_ELSE; 4170 DirectiveKindMap[".end"] = DK_END; 4171 DirectiveKindMap[".endif"] = DK_ENDIF; 4172 DirectiveKindMap[".skip"] = DK_SKIP; 4173 DirectiveKindMap[".space"] = DK_SPACE; 4174 DirectiveKindMap[".file"] = DK_FILE; 4175 DirectiveKindMap[".line"] = DK_LINE; 4176 DirectiveKindMap[".loc"] = DK_LOC; 4177 DirectiveKindMap[".stabs"] = DK_STABS; 4178 DirectiveKindMap[".sleb128"] = DK_SLEB128; 4179 DirectiveKindMap[".uleb128"] = DK_ULEB128; 4180 DirectiveKindMap[".cfi_sections"] = DK_CFI_SECTIONS; 4181 DirectiveKindMap[".cfi_startproc"] = DK_CFI_STARTPROC; 4182 DirectiveKindMap[".cfi_endproc"] = DK_CFI_ENDPROC; 4183 DirectiveKindMap[".cfi_def_cfa"] = DK_CFI_DEF_CFA; 4184 DirectiveKindMap[".cfi_def_cfa_offset"] = DK_CFI_DEF_CFA_OFFSET; 4185 DirectiveKindMap[".cfi_adjust_cfa_offset"] = DK_CFI_ADJUST_CFA_OFFSET; 4186 DirectiveKindMap[".cfi_def_cfa_register"] = DK_CFI_DEF_CFA_REGISTER; 4187 DirectiveKindMap[".cfi_offset"] = DK_CFI_OFFSET; 4188 DirectiveKindMap[".cfi_rel_offset"] = DK_CFI_REL_OFFSET; 4189 DirectiveKindMap[".cfi_personality"] = DK_CFI_PERSONALITY; 4190 DirectiveKindMap[".cfi_lsda"] = DK_CFI_LSDA; 4191 DirectiveKindMap[".cfi_remember_state"] = DK_CFI_REMEMBER_STATE; 4192 DirectiveKindMap[".cfi_restore_state"] = DK_CFI_RESTORE_STATE; 4193 DirectiveKindMap[".cfi_same_value"] = DK_CFI_SAME_VALUE; 4194 DirectiveKindMap[".cfi_restore"] = DK_CFI_RESTORE; 4195 DirectiveKindMap[".cfi_escape"] = DK_CFI_ESCAPE; 4196 DirectiveKindMap[".cfi_signal_frame"] = DK_CFI_SIGNAL_FRAME; 4197 DirectiveKindMap[".cfi_undefined"] = DK_CFI_UNDEFINED; 4198 DirectiveKindMap[".cfi_register"] = DK_CFI_REGISTER; 4199 DirectiveKindMap[".cfi_window_save"] = DK_CFI_WINDOW_SAVE; 4200 DirectiveKindMap[".macros_on"] = DK_MACROS_ON; 4201 DirectiveKindMap[".macros_off"] = DK_MACROS_OFF; 4202 DirectiveKindMap[".macro"] = DK_MACRO; 4203 DirectiveKindMap[".endm"] = DK_ENDM; 4204 DirectiveKindMap[".endmacro"] = DK_ENDMACRO; 4205 DirectiveKindMap[".purgem"] = DK_PURGEM; 4206 DirectiveKindMap[".err"] = DK_ERR; 4207 DirectiveKindMap[".error"] = DK_ERROR; 4208 } 4209 4210 MCAsmMacro *AsmParser::parseMacroLikeBody(SMLoc DirectiveLoc) { 4211 AsmToken EndToken, StartToken = getTok(); 4212 4213 unsigned NestLevel = 0; 4214 for (;;) { 4215 // Check whether we have reached the end of the file. 4216 if (getLexer().is(AsmToken::Eof)) { 4217 Error(DirectiveLoc, "no matching '.endr' in definition"); 4218 return nullptr; 4219 } 4220 4221 if (Lexer.is(AsmToken::Identifier) && 4222 (getTok().getIdentifier() == ".rept")) { 4223 ++NestLevel; 4224 } 4225 4226 // Otherwise, check whether we have reached the .endr. 4227 if (Lexer.is(AsmToken::Identifier) && getTok().getIdentifier() == ".endr") { 4228 if (NestLevel == 0) { 4229 EndToken = getTok(); 4230 Lex(); 4231 if (Lexer.isNot(AsmToken::EndOfStatement)) { 4232 TokError("unexpected token in '.endr' directive"); 4233 return nullptr; 4234 } 4235 break; 4236 } 4237 --NestLevel; 4238 } 4239 4240 // Otherwise, scan till the end of the statement. 4241 eatToEndOfStatement(); 4242 } 4243 4244 const char *BodyStart = StartToken.getLoc().getPointer(); 4245 const char *BodyEnd = EndToken.getLoc().getPointer(); 4246 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 4247 4248 // We Are Anonymous. 4249 MacroLikeBodies.push_back(MCAsmMacro(StringRef(), Body, None)); 4250 return &MacroLikeBodies.back(); 4251 } 4252 4253 void AsmParser::instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 4254 raw_svector_ostream &OS) { 4255 OS << ".endr\n"; 4256 4257 MemoryBuffer *Instantiation = 4258 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 4259 4260 // Create the macro instantiation object and add to the current macro 4261 // instantiation stack. 4262 MacroInstantiation *MI = new MacroInstantiation( 4263 M, DirectiveLoc, CurBuffer, getTok().getLoc(), Instantiation); 4264 ActiveMacros.push_back(MI); 4265 4266 // Jump to the macro instantiation and prime the lexer. 4267 CurBuffer = SrcMgr.AddNewSourceBuffer(MI->Instantiation, SMLoc()); 4268 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 4269 Lex(); 4270 } 4271 4272 /// parseDirectiveRept 4273 /// ::= .rep | .rept count 4274 bool AsmParser::parseDirectiveRept(SMLoc DirectiveLoc, StringRef Dir) { 4275 const MCExpr *CountExpr; 4276 SMLoc CountLoc = getTok().getLoc(); 4277 if (parseExpression(CountExpr)) 4278 return true; 4279 4280 int64_t Count; 4281 if (!CountExpr->EvaluateAsAbsolute(Count)) { 4282 eatToEndOfStatement(); 4283 return Error(CountLoc, "unexpected token in '" + Dir + "' directive"); 4284 } 4285 4286 if (Count < 0) 4287 return Error(CountLoc, "Count is negative"); 4288 4289 if (Lexer.isNot(AsmToken::EndOfStatement)) 4290 return TokError("unexpected token in '" + Dir + "' directive"); 4291 4292 // Eat the end of statement. 4293 Lex(); 4294 4295 // Lex the rept definition. 4296 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 4297 if (!M) 4298 return true; 4299 4300 // Macro instantiation is lexical, unfortunately. We construct a new buffer 4301 // to hold the macro body with substitutions. 4302 SmallString<256> Buf; 4303 raw_svector_ostream OS(Buf); 4304 while (Count--) { 4305 if (expandMacro(OS, M->Body, None, None, getTok().getLoc())) 4306 return true; 4307 } 4308 instantiateMacroLikeBody(M, DirectiveLoc, OS); 4309 4310 return false; 4311 } 4312 4313 /// parseDirectiveIrp 4314 /// ::= .irp symbol,values 4315 bool AsmParser::parseDirectiveIrp(SMLoc DirectiveLoc) { 4316 MCAsmMacroParameter Parameter; 4317 4318 if (parseIdentifier(Parameter.Name)) 4319 return TokError("expected identifier in '.irp' directive"); 4320 4321 if (Lexer.isNot(AsmToken::Comma)) 4322 return TokError("expected comma in '.irp' directive"); 4323 4324 Lex(); 4325 4326 MCAsmMacroArguments A; 4327 if (parseMacroArguments(nullptr, A)) 4328 return true; 4329 4330 // Eat the end of statement. 4331 Lex(); 4332 4333 // Lex the irp definition. 4334 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 4335 if (!M) 4336 return true; 4337 4338 // Macro instantiation is lexical, unfortunately. We construct a new buffer 4339 // to hold the macro body with substitutions. 4340 SmallString<256> Buf; 4341 raw_svector_ostream OS(Buf); 4342 4343 for (MCAsmMacroArguments::iterator i = A.begin(), e = A.end(); i != e; ++i) { 4344 if (expandMacro(OS, M->Body, Parameter, *i, getTok().getLoc())) 4345 return true; 4346 } 4347 4348 instantiateMacroLikeBody(M, DirectiveLoc, OS); 4349 4350 return false; 4351 } 4352 4353 /// parseDirectiveIrpc 4354 /// ::= .irpc symbol,values 4355 bool AsmParser::parseDirectiveIrpc(SMLoc DirectiveLoc) { 4356 MCAsmMacroParameter Parameter; 4357 4358 if (parseIdentifier(Parameter.Name)) 4359 return TokError("expected identifier in '.irpc' directive"); 4360 4361 if (Lexer.isNot(AsmToken::Comma)) 4362 return TokError("expected comma in '.irpc' directive"); 4363 4364 Lex(); 4365 4366 MCAsmMacroArguments A; 4367 if (parseMacroArguments(nullptr, A)) 4368 return true; 4369 4370 if (A.size() != 1 || A.front().size() != 1) 4371 return TokError("unexpected token in '.irpc' directive"); 4372 4373 // Eat the end of statement. 4374 Lex(); 4375 4376 // Lex the irpc definition. 4377 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 4378 if (!M) 4379 return true; 4380 4381 // Macro instantiation is lexical, unfortunately. We construct a new buffer 4382 // to hold the macro body with substitutions. 4383 SmallString<256> Buf; 4384 raw_svector_ostream OS(Buf); 4385 4386 StringRef Values = A.front().front().getString(); 4387 for (std::size_t I = 0, End = Values.size(); I != End; ++I) { 4388 MCAsmMacroArgument Arg; 4389 Arg.push_back(AsmToken(AsmToken::Identifier, Values.slice(I, I + 1))); 4390 4391 if (expandMacro(OS, M->Body, Parameter, Arg, getTok().getLoc())) 4392 return true; 4393 } 4394 4395 instantiateMacroLikeBody(M, DirectiveLoc, OS); 4396 4397 return false; 4398 } 4399 4400 bool AsmParser::parseDirectiveEndr(SMLoc DirectiveLoc) { 4401 if (ActiveMacros.empty()) 4402 return TokError("unmatched '.endr' directive"); 4403 4404 // The only .repl that should get here are the ones created by 4405 // instantiateMacroLikeBody. 4406 assert(getLexer().is(AsmToken::EndOfStatement)); 4407 4408 handleMacroExit(); 4409 return false; 4410 } 4411 4412 bool AsmParser::parseDirectiveMSEmit(SMLoc IDLoc, ParseStatementInfo &Info, 4413 size_t Len) { 4414 const MCExpr *Value; 4415 SMLoc ExprLoc = getLexer().getLoc(); 4416 if (parseExpression(Value)) 4417 return true; 4418 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 4419 if (!MCE) 4420 return Error(ExprLoc, "unexpected expression in _emit"); 4421 uint64_t IntValue = MCE->getValue(); 4422 if (!isUIntN(8, IntValue) && !isIntN(8, IntValue)) 4423 return Error(ExprLoc, "literal value out of range for directive"); 4424 4425 Info.AsmRewrites->push_back(AsmRewrite(AOK_Emit, IDLoc, Len)); 4426 return false; 4427 } 4428 4429 bool AsmParser::parseDirectiveMSAlign(SMLoc IDLoc, ParseStatementInfo &Info) { 4430 const MCExpr *Value; 4431 SMLoc ExprLoc = getLexer().getLoc(); 4432 if (parseExpression(Value)) 4433 return true; 4434 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 4435 if (!MCE) 4436 return Error(ExprLoc, "unexpected expression in align"); 4437 uint64_t IntValue = MCE->getValue(); 4438 if (!isPowerOf2_64(IntValue)) 4439 return Error(ExprLoc, "literal value not a power of two greater then zero"); 4440 4441 Info.AsmRewrites->push_back( 4442 AsmRewrite(AOK_Align, IDLoc, 5, Log2_64(IntValue))); 4443 return false; 4444 } 4445 4446 // We are comparing pointers, but the pointers are relative to a single string. 4447 // Thus, this should always be deterministic. 4448 static int rewritesSort(const AsmRewrite *AsmRewriteA, 4449 const AsmRewrite *AsmRewriteB) { 4450 if (AsmRewriteA->Loc.getPointer() < AsmRewriteB->Loc.getPointer()) 4451 return -1; 4452 if (AsmRewriteB->Loc.getPointer() < AsmRewriteA->Loc.getPointer()) 4453 return 1; 4454 4455 // It's possible to have a SizeDirective, Imm/ImmPrefix and an Input/Output 4456 // rewrite to the same location. Make sure the SizeDirective rewrite is 4457 // performed first, then the Imm/ImmPrefix and finally the Input/Output. This 4458 // ensures the sort algorithm is stable. 4459 if (AsmRewritePrecedence[AsmRewriteA->Kind] > 4460 AsmRewritePrecedence[AsmRewriteB->Kind]) 4461 return -1; 4462 4463 if (AsmRewritePrecedence[AsmRewriteA->Kind] < 4464 AsmRewritePrecedence[AsmRewriteB->Kind]) 4465 return 1; 4466 llvm_unreachable("Unstable rewrite sort."); 4467 } 4468 4469 bool AsmParser::parseMSInlineAsm( 4470 void *AsmLoc, std::string &AsmString, unsigned &NumOutputs, 4471 unsigned &NumInputs, SmallVectorImpl<std::pair<void *, bool> > &OpDecls, 4472 SmallVectorImpl<std::string> &Constraints, 4473 SmallVectorImpl<std::string> &Clobbers, const MCInstrInfo *MII, 4474 const MCInstPrinter *IP, MCAsmParserSemaCallback &SI) { 4475 SmallVector<void *, 4> InputDecls; 4476 SmallVector<void *, 4> OutputDecls; 4477 SmallVector<bool, 4> InputDeclsAddressOf; 4478 SmallVector<bool, 4> OutputDeclsAddressOf; 4479 SmallVector<std::string, 4> InputConstraints; 4480 SmallVector<std::string, 4> OutputConstraints; 4481 SmallVector<unsigned, 4> ClobberRegs; 4482 4483 SmallVector<AsmRewrite, 4> AsmStrRewrites; 4484 4485 // Prime the lexer. 4486 Lex(); 4487 4488 // While we have input, parse each statement. 4489 unsigned InputIdx = 0; 4490 unsigned OutputIdx = 0; 4491 while (getLexer().isNot(AsmToken::Eof)) { 4492 ParseStatementInfo Info(&AsmStrRewrites); 4493 if (parseStatement(Info)) 4494 return true; 4495 4496 if (Info.ParseError) 4497 return true; 4498 4499 if (Info.Opcode == ~0U) 4500 continue; 4501 4502 const MCInstrDesc &Desc = MII->get(Info.Opcode); 4503 4504 // Build the list of clobbers, outputs and inputs. 4505 for (unsigned i = 1, e = Info.ParsedOperands.size(); i != e; ++i) { 4506 MCParsedAsmOperand &Operand = *Info.ParsedOperands[i]; 4507 4508 // Immediate. 4509 if (Operand.isImm()) 4510 continue; 4511 4512 // Register operand. 4513 if (Operand.isReg() && !Operand.needAddressOf()) { 4514 unsigned NumDefs = Desc.getNumDefs(); 4515 // Clobber. 4516 if (NumDefs && Operand.getMCOperandNum() < NumDefs) 4517 ClobberRegs.push_back(Operand.getReg()); 4518 continue; 4519 } 4520 4521 // Expr/Input or Output. 4522 StringRef SymName = Operand.getSymName(); 4523 if (SymName.empty()) 4524 continue; 4525 4526 void *OpDecl = Operand.getOpDecl(); 4527 if (!OpDecl) 4528 continue; 4529 4530 bool isOutput = (i == 1) && Desc.mayStore(); 4531 SMLoc Start = SMLoc::getFromPointer(SymName.data()); 4532 if (isOutput) { 4533 ++InputIdx; 4534 OutputDecls.push_back(OpDecl); 4535 OutputDeclsAddressOf.push_back(Operand.needAddressOf()); 4536 OutputConstraints.push_back('=' + Operand.getConstraint().str()); 4537 AsmStrRewrites.push_back(AsmRewrite(AOK_Output, Start, SymName.size())); 4538 } else { 4539 InputDecls.push_back(OpDecl); 4540 InputDeclsAddressOf.push_back(Operand.needAddressOf()); 4541 InputConstraints.push_back(Operand.getConstraint().str()); 4542 AsmStrRewrites.push_back(AsmRewrite(AOK_Input, Start, SymName.size())); 4543 } 4544 } 4545 4546 // Consider implicit defs to be clobbers. Think of cpuid and push. 4547 ArrayRef<uint16_t> ImpDefs(Desc.getImplicitDefs(), 4548 Desc.getNumImplicitDefs()); 4549 ClobberRegs.insert(ClobberRegs.end(), ImpDefs.begin(), ImpDefs.end()); 4550 } 4551 4552 // Set the number of Outputs and Inputs. 4553 NumOutputs = OutputDecls.size(); 4554 NumInputs = InputDecls.size(); 4555 4556 // Set the unique clobbers. 4557 array_pod_sort(ClobberRegs.begin(), ClobberRegs.end()); 4558 ClobberRegs.erase(std::unique(ClobberRegs.begin(), ClobberRegs.end()), 4559 ClobberRegs.end()); 4560 Clobbers.assign(ClobberRegs.size(), std::string()); 4561 for (unsigned I = 0, E = ClobberRegs.size(); I != E; ++I) { 4562 raw_string_ostream OS(Clobbers[I]); 4563 IP->printRegName(OS, ClobberRegs[I]); 4564 } 4565 4566 // Merge the various outputs and inputs. Output are expected first. 4567 if (NumOutputs || NumInputs) { 4568 unsigned NumExprs = NumOutputs + NumInputs; 4569 OpDecls.resize(NumExprs); 4570 Constraints.resize(NumExprs); 4571 for (unsigned i = 0; i < NumOutputs; ++i) { 4572 OpDecls[i] = std::make_pair(OutputDecls[i], OutputDeclsAddressOf[i]); 4573 Constraints[i] = OutputConstraints[i]; 4574 } 4575 for (unsigned i = 0, j = NumOutputs; i < NumInputs; ++i, ++j) { 4576 OpDecls[j] = std::make_pair(InputDecls[i], InputDeclsAddressOf[i]); 4577 Constraints[j] = InputConstraints[i]; 4578 } 4579 } 4580 4581 // Build the IR assembly string. 4582 std::string AsmStringIR; 4583 raw_string_ostream OS(AsmStringIR); 4584 StringRef ASMString = 4585 SrcMgr.getMemoryBuffer(SrcMgr.getMainFileID())->getBuffer(); 4586 const char *AsmStart = ASMString.begin(); 4587 const char *AsmEnd = ASMString.end(); 4588 array_pod_sort(AsmStrRewrites.begin(), AsmStrRewrites.end(), rewritesSort); 4589 for (const AsmRewrite &AR : AsmStrRewrites) { 4590 AsmRewriteKind Kind = AR.Kind; 4591 if (Kind == AOK_Delete) 4592 continue; 4593 4594 const char *Loc = AR.Loc.getPointer(); 4595 assert(Loc >= AsmStart && "Expected Loc to be at or after Start!"); 4596 4597 // Emit everything up to the immediate/expression. 4598 if (unsigned Len = Loc - AsmStart) 4599 OS << StringRef(AsmStart, Len); 4600 4601 // Skip the original expression. 4602 if (Kind == AOK_Skip) { 4603 AsmStart = Loc + AR.Len; 4604 continue; 4605 } 4606 4607 unsigned AdditionalSkip = 0; 4608 // Rewrite expressions in $N notation. 4609 switch (Kind) { 4610 default: 4611 break; 4612 case AOK_Imm: 4613 OS << "$$" << AR.Val; 4614 break; 4615 case AOK_ImmPrefix: 4616 OS << "$$"; 4617 break; 4618 case AOK_Input: 4619 OS << '$' << InputIdx++; 4620 break; 4621 case AOK_Output: 4622 OS << '$' << OutputIdx++; 4623 break; 4624 case AOK_SizeDirective: 4625 switch (AR.Val) { 4626 default: break; 4627 case 8: OS << "byte ptr "; break; 4628 case 16: OS << "word ptr "; break; 4629 case 32: OS << "dword ptr "; break; 4630 case 64: OS << "qword ptr "; break; 4631 case 80: OS << "xword ptr "; break; 4632 case 128: OS << "xmmword ptr "; break; 4633 case 256: OS << "ymmword ptr "; break; 4634 } 4635 break; 4636 case AOK_Emit: 4637 OS << ".byte"; 4638 break; 4639 case AOK_Align: { 4640 unsigned Val = AR.Val; 4641 OS << ".align " << Val; 4642 4643 // Skip the original immediate. 4644 assert(Val < 10 && "Expected alignment less then 2^10."); 4645 AdditionalSkip = (Val < 4) ? 2 : Val < 7 ? 3 : 4; 4646 break; 4647 } 4648 case AOK_DotOperator: 4649 // Insert the dot if the user omitted it. 4650 OS.flush(); 4651 if (AsmStringIR.back() != '.') 4652 OS << '.'; 4653 OS << AR.Val; 4654 break; 4655 } 4656 4657 // Skip the original expression. 4658 AsmStart = Loc + AR.Len + AdditionalSkip; 4659 } 4660 4661 // Emit the remainder of the asm string. 4662 if (AsmStart != AsmEnd) 4663 OS << StringRef(AsmStart, AsmEnd - AsmStart); 4664 4665 AsmString = OS.str(); 4666 return false; 4667 } 4668 4669 /// \brief Create an MCAsmParser instance. 4670 MCAsmParser *llvm::createMCAsmParser(SourceMgr &SM, MCContext &C, 4671 MCStreamer &Out, const MCAsmInfo &MAI) { 4672 return new AsmParser(SM, C, Out, MAI); 4673 } 4674