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 MCAsmParserSemaCallback::~MCAsmParserSemaCallback() {} 49 50 namespace { 51 /// \brief Helper types for tracking macro definitions. 52 typedef std::vector<AsmToken> MCAsmMacroArgument; 53 typedef std::vector<MCAsmMacroArgument> MCAsmMacroArguments; 54 55 struct MCAsmMacroParameter { 56 StringRef Name; 57 MCAsmMacroArgument Value; 58 bool Required; 59 bool Vararg; 60 61 MCAsmMacroParameter() : Required(false), Vararg(false) {} 62 }; 63 64 typedef std::vector<MCAsmMacroParameter> MCAsmMacroParameters; 65 66 struct MCAsmMacro { 67 StringRef Name; 68 StringRef Body; 69 MCAsmMacroParameters Parameters; 70 71 public: 72 MCAsmMacro(StringRef N, StringRef B, MCAsmMacroParameters P) 73 : Name(N), Body(B), Parameters(std::move(P)) {} 74 }; 75 76 /// \brief Helper class for storing information about an active macro 77 /// instantiation. 78 struct MacroInstantiation { 79 /// The location of the instantiation. 80 SMLoc InstantiationLoc; 81 82 /// The buffer where parsing should resume upon instantiation completion. 83 int ExitBuffer; 84 85 /// The location where parsing should resume upon instantiation completion. 86 SMLoc ExitLoc; 87 88 /// The depth of TheCondStack at the start of the instantiation. 89 size_t CondStackDepth; 90 91 public: 92 MacroInstantiation(SMLoc IL, int EB, SMLoc EL, size_t CondStackDepth); 93 }; 94 95 struct ParseStatementInfo { 96 /// \brief The parsed operands from the last parsed statement. 97 SmallVector<std::unique_ptr<MCParsedAsmOperand>, 8> ParsedOperands; 98 99 /// \brief The opcode from the last parsed instruction. 100 unsigned Opcode; 101 102 /// \brief Was there an error parsing the inline assembly? 103 bool ParseError; 104 105 SmallVectorImpl<AsmRewrite> *AsmRewrites; 106 107 ParseStatementInfo() : Opcode(~0U), ParseError(false), AsmRewrites(nullptr) {} 108 ParseStatementInfo(SmallVectorImpl<AsmRewrite> *rewrites) 109 : Opcode(~0), ParseError(false), AsmRewrites(rewrites) {} 110 }; 111 112 /// \brief The concrete assembly parser instance. 113 class AsmParser : public MCAsmParser { 114 AsmParser(const AsmParser &) = delete; 115 void operator=(const AsmParser &) = delete; 116 private: 117 AsmLexer Lexer; 118 MCContext &Ctx; 119 MCStreamer &Out; 120 const MCAsmInfo &MAI; 121 SourceMgr &SrcMgr; 122 SourceMgr::DiagHandlerTy SavedDiagHandler; 123 void *SavedDiagContext; 124 std::unique_ptr<MCAsmParserExtension> PlatformParser; 125 126 /// This is the current buffer index we're lexing from as managed by the 127 /// SourceMgr object. 128 unsigned CurBuffer; 129 130 AsmCond TheCondState; 131 std::vector<AsmCond> TheCondStack; 132 133 /// \brief maps directive names to handler methods in parser 134 /// extensions. Extensions register themselves in this map by calling 135 /// addDirectiveHandler. 136 StringMap<ExtensionDirectiveHandler> ExtensionDirectiveMap; 137 138 /// \brief Map of currently defined macros. 139 StringMap<MCAsmMacro> MacroMap; 140 141 /// \brief Stack of active macro instantiations. 142 std::vector<MacroInstantiation*> ActiveMacros; 143 144 /// \brief List of bodies of anonymous macros. 145 std::deque<MCAsmMacro> MacroLikeBodies; 146 147 /// Boolean tracking whether macro substitution is enabled. 148 unsigned MacrosEnabledFlag : 1; 149 150 /// Flag tracking whether any errors have been encountered. 151 unsigned HadError : 1; 152 153 /// The values from the last parsed cpp hash file line comment if any. 154 StringRef CppHashFilename; 155 int64_t CppHashLineNumber; 156 SMLoc CppHashLoc; 157 unsigned CppHashBuf; 158 /// When generating dwarf for assembly source files we need to calculate the 159 /// logical line number based on the last parsed cpp hash file line comment 160 /// and current line. Since this is slow and messes up the SourceMgr's 161 /// cache we save the last info we queried with SrcMgr.FindLineNumber(). 162 SMLoc LastQueryIDLoc; 163 unsigned LastQueryBuffer; 164 unsigned LastQueryLine; 165 166 /// AssemblerDialect. ~OU means unset value and use value provided by MAI. 167 unsigned AssemblerDialect; 168 169 /// \brief is Darwin compatibility enabled? 170 bool IsDarwin; 171 172 /// \brief Are we parsing ms-style inline assembly? 173 bool ParsingInlineAsm; 174 175 public: 176 AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out, 177 const MCAsmInfo &MAI); 178 ~AsmParser() override; 179 180 bool Run(bool NoInitialTextSection, bool NoFinalize = false) override; 181 182 void addDirectiveHandler(StringRef Directive, 183 ExtensionDirectiveHandler Handler) override { 184 ExtensionDirectiveMap[Directive] = Handler; 185 } 186 187 public: 188 /// @name MCAsmParser Interface 189 /// { 190 191 SourceMgr &getSourceManager() override { return SrcMgr; } 192 MCAsmLexer &getLexer() override { return Lexer; } 193 MCContext &getContext() override { return Ctx; } 194 MCStreamer &getStreamer() override { return Out; } 195 unsigned getAssemblerDialect() override { 196 if (AssemblerDialect == ~0U) 197 return MAI.getAssemblerDialect(); 198 else 199 return AssemblerDialect; 200 } 201 void setAssemblerDialect(unsigned i) override { 202 AssemblerDialect = i; 203 } 204 205 void Note(SMLoc L, const Twine &Msg, 206 ArrayRef<SMRange> Ranges = None) override; 207 bool Warning(SMLoc L, const Twine &Msg, 208 ArrayRef<SMRange> Ranges = None) override; 209 bool Error(SMLoc L, const Twine &Msg, 210 ArrayRef<SMRange> Ranges = None) override; 211 212 const AsmToken &Lex() override; 213 214 void setParsingInlineAsm(bool V) override { ParsingInlineAsm = V; } 215 bool isParsingInlineAsm() override { return ParsingInlineAsm; } 216 217 bool parseMSInlineAsm(void *AsmLoc, std::string &AsmString, 218 unsigned &NumOutputs, unsigned &NumInputs, 219 SmallVectorImpl<std::pair<void *,bool> > &OpDecls, 220 SmallVectorImpl<std::string> &Constraints, 221 SmallVectorImpl<std::string> &Clobbers, 222 const MCInstrInfo *MII, const MCInstPrinter *IP, 223 MCAsmParserSemaCallback &SI) override; 224 225 bool parseExpression(const MCExpr *&Res); 226 bool parseExpression(const MCExpr *&Res, SMLoc &EndLoc) override; 227 bool parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) override; 228 bool parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) override; 229 bool parseAbsoluteExpression(int64_t &Res) override; 230 231 /// \brief Parse an identifier or string (as a quoted identifier) 232 /// and set \p Res to the identifier contents. 233 bool parseIdentifier(StringRef &Res) override; 234 void eatToEndOfStatement() override; 235 236 void checkForValidSection() override; 237 /// } 238 239 private: 240 241 bool parseStatement(ParseStatementInfo &Info, 242 MCAsmParserSemaCallback *SI); 243 void eatToEndOfLine(); 244 bool parseCppHashLineFilenameComment(const SMLoc &L); 245 246 void checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, StringRef Body, 247 ArrayRef<MCAsmMacroParameter> Parameters); 248 bool expandMacro(raw_svector_ostream &OS, StringRef Body, 249 ArrayRef<MCAsmMacroParameter> Parameters, 250 ArrayRef<MCAsmMacroArgument> A, 251 const SMLoc &L); 252 253 /// \brief Are macros enabled in the parser? 254 bool areMacrosEnabled() {return MacrosEnabledFlag;} 255 256 /// \brief Control a flag in the parser that enables or disables macros. 257 void setMacrosEnabled(bool Flag) {MacrosEnabledFlag = Flag;} 258 259 /// \brief Lookup a previously defined macro. 260 /// \param Name Macro name. 261 /// \returns Pointer to macro. NULL if no such macro was defined. 262 const MCAsmMacro* lookupMacro(StringRef Name); 263 264 /// \brief Define a new macro with the given name and information. 265 void defineMacro(StringRef Name, MCAsmMacro Macro); 266 267 /// \brief Undefine a macro. If no such macro was defined, it's a no-op. 268 void undefineMacro(StringRef Name); 269 270 /// \brief Are we inside a macro instantiation? 271 bool isInsideMacroInstantiation() {return !ActiveMacros.empty();} 272 273 /// \brief Handle entry to macro instantiation. 274 /// 275 /// \param M The macro. 276 /// \param NameLoc Instantiation location. 277 bool handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc); 278 279 /// \brief Handle exit from macro instantiation. 280 void handleMacroExit(); 281 282 /// \brief Extract AsmTokens for a macro argument. 283 bool parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg); 284 285 /// \brief Parse all macro arguments for a given macro. 286 bool parseMacroArguments(const MCAsmMacro *M, MCAsmMacroArguments &A); 287 288 void printMacroInstantiations(); 289 void printMessage(SMLoc Loc, SourceMgr::DiagKind Kind, const Twine &Msg, 290 ArrayRef<SMRange> Ranges = None) const { 291 SrcMgr.PrintMessage(Loc, Kind, Msg, Ranges); 292 } 293 static void DiagHandler(const SMDiagnostic &Diag, void *Context); 294 295 /// \brief Enter the specified file. This returns true on failure. 296 bool enterIncludeFile(const std::string &Filename); 297 298 /// \brief Process the specified file for the .incbin directive. 299 /// This returns true on failure. 300 bool processIncbinFile(const std::string &Filename); 301 302 /// \brief Reset the current lexer position to that given by \p Loc. The 303 /// current token is not set; clients should ensure Lex() is called 304 /// subsequently. 305 /// 306 /// \param InBuffer If not 0, should be the known buffer id that contains the 307 /// location. 308 void jumpToLoc(SMLoc Loc, unsigned InBuffer = 0); 309 310 /// \brief Parse up to the end of statement and a return the contents from the 311 /// current token until the end of the statement; the current token on exit 312 /// will be either the EndOfStatement or EOF. 313 StringRef parseStringToEndOfStatement() override; 314 315 /// \brief Parse until the end of a statement or a comma is encountered, 316 /// return the contents from the current token up to the end or comma. 317 StringRef parseStringToComma(); 318 319 bool parseAssignment(StringRef Name, bool allow_redef, 320 bool NoDeadStrip = false); 321 322 bool parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, SMLoc &EndLoc); 323 bool parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc); 324 bool parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc); 325 326 bool parseRegisterOrRegisterNumber(int64_t &Register, SMLoc DirectiveLoc); 327 328 // Generic (target and platform independent) directive parsing. 329 enum DirectiveKind { 330 DK_NO_DIRECTIVE, // Placeholder 331 DK_SET, DK_EQU, DK_EQUIV, DK_ASCII, DK_ASCIZ, DK_STRING, DK_BYTE, DK_SHORT, 332 DK_VALUE, DK_2BYTE, DK_LONG, DK_INT, DK_4BYTE, DK_QUAD, DK_8BYTE, DK_OCTA, 333 DK_SINGLE, DK_FLOAT, DK_DOUBLE, DK_ALIGN, DK_ALIGN32, DK_BALIGN, DK_BALIGNW, 334 DK_BALIGNL, DK_P2ALIGN, DK_P2ALIGNW, DK_P2ALIGNL, DK_ORG, DK_FILL, DK_ENDR, 335 DK_BUNDLE_ALIGN_MODE, DK_BUNDLE_LOCK, DK_BUNDLE_UNLOCK, 336 DK_ZERO, DK_EXTERN, DK_GLOBL, DK_GLOBAL, 337 DK_LAZY_REFERENCE, DK_NO_DEAD_STRIP, DK_SYMBOL_RESOLVER, DK_PRIVATE_EXTERN, 338 DK_REFERENCE, DK_WEAK_DEFINITION, DK_WEAK_REFERENCE, 339 DK_WEAK_DEF_CAN_BE_HIDDEN, DK_COMM, DK_COMMON, DK_LCOMM, DK_ABORT, 340 DK_INCLUDE, DK_INCBIN, DK_CODE16, DK_CODE16GCC, DK_REPT, DK_IRP, DK_IRPC, 341 DK_IF, DK_IFEQ, DK_IFGE, DK_IFGT, DK_IFLE, DK_IFLT, DK_IFNE, DK_IFB, 342 DK_IFNB, DK_IFC, DK_IFEQS, DK_IFNC, DK_IFNES, DK_IFDEF, DK_IFNDEF, 343 DK_IFNOTDEF, DK_ELSEIF, DK_ELSE, DK_ENDIF, 344 DK_SPACE, DK_SKIP, DK_FILE, DK_LINE, DK_LOC, DK_STABS, 345 DK_CFI_SECTIONS, DK_CFI_STARTPROC, DK_CFI_ENDPROC, DK_CFI_DEF_CFA, 346 DK_CFI_DEF_CFA_OFFSET, DK_CFI_ADJUST_CFA_OFFSET, DK_CFI_DEF_CFA_REGISTER, 347 DK_CFI_OFFSET, DK_CFI_REL_OFFSET, DK_CFI_PERSONALITY, DK_CFI_LSDA, 348 DK_CFI_REMEMBER_STATE, DK_CFI_RESTORE_STATE, DK_CFI_SAME_VALUE, 349 DK_CFI_RESTORE, DK_CFI_ESCAPE, DK_CFI_SIGNAL_FRAME, DK_CFI_UNDEFINED, 350 DK_CFI_REGISTER, DK_CFI_WINDOW_SAVE, 351 DK_MACROS_ON, DK_MACROS_OFF, 352 DK_MACRO, DK_EXITM, DK_ENDM, DK_ENDMACRO, DK_PURGEM, 353 DK_SLEB128, DK_ULEB128, 354 DK_ERR, DK_ERROR, DK_WARNING, 355 DK_END 356 }; 357 358 /// \brief Maps directive name --> DirectiveKind enum, for 359 /// directives parsed by this class. 360 StringMap<DirectiveKind> DirectiveKindMap; 361 362 // ".ascii", ".asciz", ".string" 363 bool parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated); 364 bool parseDirectiveValue(unsigned Size); // ".byte", ".long", ... 365 bool parseDirectiveOctaValue(); // ".octa" 366 bool parseDirectiveRealValue(const fltSemantics &); // ".single", ... 367 bool parseDirectiveFill(); // ".fill" 368 bool parseDirectiveZero(); // ".zero" 369 // ".set", ".equ", ".equiv" 370 bool parseDirectiveSet(StringRef IDVal, bool allow_redef); 371 bool parseDirectiveOrg(); // ".org" 372 // ".align{,32}", ".p2align{,w,l}" 373 bool parseDirectiveAlign(bool IsPow2, unsigned ValueSize); 374 375 // ".file", ".line", ".loc", ".stabs" 376 bool parseDirectiveFile(SMLoc DirectiveLoc); 377 bool parseDirectiveLine(); 378 bool parseDirectiveLoc(); 379 bool parseDirectiveStabs(); 380 381 // .cfi directives 382 bool parseDirectiveCFIRegister(SMLoc DirectiveLoc); 383 bool parseDirectiveCFIWindowSave(); 384 bool parseDirectiveCFISections(); 385 bool parseDirectiveCFIStartProc(); 386 bool parseDirectiveCFIEndProc(); 387 bool parseDirectiveCFIDefCfaOffset(); 388 bool parseDirectiveCFIDefCfa(SMLoc DirectiveLoc); 389 bool parseDirectiveCFIAdjustCfaOffset(); 390 bool parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc); 391 bool parseDirectiveCFIOffset(SMLoc DirectiveLoc); 392 bool parseDirectiveCFIRelOffset(SMLoc DirectiveLoc); 393 bool parseDirectiveCFIPersonalityOrLsda(bool IsPersonality); 394 bool parseDirectiveCFIRememberState(); 395 bool parseDirectiveCFIRestoreState(); 396 bool parseDirectiveCFISameValue(SMLoc DirectiveLoc); 397 bool parseDirectiveCFIRestore(SMLoc DirectiveLoc); 398 bool parseDirectiveCFIEscape(); 399 bool parseDirectiveCFISignalFrame(); 400 bool parseDirectiveCFIUndefined(SMLoc DirectiveLoc); 401 402 // macro directives 403 bool parseDirectivePurgeMacro(SMLoc DirectiveLoc); 404 bool parseDirectiveExitMacro(StringRef Directive); 405 bool parseDirectiveEndMacro(StringRef Directive); 406 bool parseDirectiveMacro(SMLoc DirectiveLoc); 407 bool parseDirectiveMacrosOnOff(StringRef Directive); 408 409 // ".bundle_align_mode" 410 bool parseDirectiveBundleAlignMode(); 411 // ".bundle_lock" 412 bool parseDirectiveBundleLock(); 413 // ".bundle_unlock" 414 bool parseDirectiveBundleUnlock(); 415 416 // ".space", ".skip" 417 bool parseDirectiveSpace(StringRef IDVal); 418 419 // .sleb128 (Signed=true) and .uleb128 (Signed=false) 420 bool parseDirectiveLEB128(bool Signed); 421 422 /// \brief Parse a directive like ".globl" which 423 /// accepts a single symbol (which should be a label or an external). 424 bool parseDirectiveSymbolAttribute(MCSymbolAttr Attr); 425 426 bool parseDirectiveComm(bool IsLocal); // ".comm" and ".lcomm" 427 428 bool parseDirectiveAbort(); // ".abort" 429 bool parseDirectiveInclude(); // ".include" 430 bool parseDirectiveIncbin(); // ".incbin" 431 432 // ".if", ".ifeq", ".ifge", ".ifgt" , ".ifle", ".iflt" or ".ifne" 433 bool parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind); 434 // ".ifb" or ".ifnb", depending on ExpectBlank. 435 bool parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank); 436 // ".ifc" or ".ifnc", depending on ExpectEqual. 437 bool parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual); 438 // ".ifeqs" or ".ifnes", depending on ExpectEqual. 439 bool parseDirectiveIfeqs(SMLoc DirectiveLoc, bool ExpectEqual); 440 // ".ifdef" or ".ifndef", depending on expect_defined 441 bool parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined); 442 bool parseDirectiveElseIf(SMLoc DirectiveLoc); // ".elseif" 443 bool parseDirectiveElse(SMLoc DirectiveLoc); // ".else" 444 bool parseDirectiveEndIf(SMLoc DirectiveLoc); // .endif 445 bool parseEscapedString(std::string &Data) override; 446 447 const MCExpr *applyModifierToExpr(const MCExpr *E, 448 MCSymbolRefExpr::VariantKind Variant); 449 450 // Macro-like directives 451 MCAsmMacro *parseMacroLikeBody(SMLoc DirectiveLoc); 452 void instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 453 raw_svector_ostream &OS); 454 bool parseDirectiveRept(SMLoc DirectiveLoc, StringRef Directive); 455 bool parseDirectiveIrp(SMLoc DirectiveLoc); // ".irp" 456 bool parseDirectiveIrpc(SMLoc DirectiveLoc); // ".irpc" 457 bool parseDirectiveEndr(SMLoc DirectiveLoc); // ".endr" 458 459 // "_emit" or "__emit" 460 bool parseDirectiveMSEmit(SMLoc DirectiveLoc, ParseStatementInfo &Info, 461 size_t Len); 462 463 // "align" 464 bool parseDirectiveMSAlign(SMLoc DirectiveLoc, ParseStatementInfo &Info); 465 466 // "end" 467 bool parseDirectiveEnd(SMLoc DirectiveLoc); 468 469 // ".err" or ".error" 470 bool parseDirectiveError(SMLoc DirectiveLoc, bool WithMessage); 471 472 // ".warning" 473 bool parseDirectiveWarning(SMLoc DirectiveLoc); 474 475 void initializeDirectiveKindMap(); 476 }; 477 } 478 479 namespace llvm { 480 481 extern MCAsmParserExtension *createDarwinAsmParser(); 482 extern MCAsmParserExtension *createELFAsmParser(); 483 extern MCAsmParserExtension *createCOFFAsmParser(); 484 485 } 486 487 enum { DEFAULT_ADDRSPACE = 0 }; 488 489 AsmParser::AsmParser(SourceMgr &SM, MCContext &Ctx, MCStreamer &Out, 490 const MCAsmInfo &MAI) 491 : Lexer(MAI), Ctx(Ctx), Out(Out), MAI(MAI), SrcMgr(SM), 492 PlatformParser(nullptr), CurBuffer(SM.getMainFileID()), 493 MacrosEnabledFlag(true), HadError(false), CppHashLineNumber(0), 494 AssemblerDialect(~0U), IsDarwin(false), ParsingInlineAsm(false) { 495 // Save the old handler. 496 SavedDiagHandler = SrcMgr.getDiagHandler(); 497 SavedDiagContext = SrcMgr.getDiagContext(); 498 // Set our own handler which calls the saved handler. 499 SrcMgr.setDiagHandler(DiagHandler, this); 500 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 501 502 // Initialize the platform / file format parser. 503 switch (Ctx.getObjectFileInfo()->getObjectFileType()) { 504 case MCObjectFileInfo::IsCOFF: 505 PlatformParser.reset(createCOFFAsmParser()); 506 break; 507 case MCObjectFileInfo::IsMachO: 508 PlatformParser.reset(createDarwinAsmParser()); 509 IsDarwin = true; 510 break; 511 case MCObjectFileInfo::IsELF: 512 PlatformParser.reset(createELFAsmParser()); 513 break; 514 } 515 516 PlatformParser->Initialize(*this); 517 initializeDirectiveKindMap(); 518 } 519 520 AsmParser::~AsmParser() { 521 assert((HadError || ActiveMacros.empty()) && 522 "Unexpected active macro instantiation!"); 523 } 524 525 void AsmParser::printMacroInstantiations() { 526 // Print the active macro instantiation stack. 527 for (std::vector<MacroInstantiation *>::const_reverse_iterator 528 it = ActiveMacros.rbegin(), 529 ie = ActiveMacros.rend(); 530 it != ie; ++it) 531 printMessage((*it)->InstantiationLoc, SourceMgr::DK_Note, 532 "while in macro instantiation"); 533 } 534 535 void AsmParser::Note(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) { 536 printMessage(L, SourceMgr::DK_Note, Msg, Ranges); 537 printMacroInstantiations(); 538 } 539 540 bool AsmParser::Warning(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) { 541 if (getTargetParser().getTargetOptions().MCFatalWarnings) 542 return Error(L, Msg, Ranges); 543 printMessage(L, SourceMgr::DK_Warning, Msg, Ranges); 544 printMacroInstantiations(); 545 return false; 546 } 547 548 bool AsmParser::Error(SMLoc L, const Twine &Msg, ArrayRef<SMRange> Ranges) { 549 HadError = true; 550 printMessage(L, SourceMgr::DK_Error, Msg, Ranges); 551 printMacroInstantiations(); 552 return true; 553 } 554 555 bool AsmParser::enterIncludeFile(const std::string &Filename) { 556 std::string IncludedFile; 557 unsigned NewBuf = 558 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 559 if (!NewBuf) 560 return true; 561 562 CurBuffer = NewBuf; 563 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 564 return false; 565 } 566 567 /// Process the specified .incbin file by searching for it in the include paths 568 /// then just emitting the byte contents of the file to the streamer. This 569 /// returns true on failure. 570 bool AsmParser::processIncbinFile(const std::string &Filename) { 571 std::string IncludedFile; 572 unsigned NewBuf = 573 SrcMgr.AddIncludeFile(Filename, Lexer.getLoc(), IncludedFile); 574 if (!NewBuf) 575 return true; 576 577 // Pick up the bytes from the file and emit them. 578 getStreamer().EmitBytes(SrcMgr.getMemoryBuffer(NewBuf)->getBuffer()); 579 return false; 580 } 581 582 void AsmParser::jumpToLoc(SMLoc Loc, unsigned InBuffer) { 583 CurBuffer = InBuffer ? InBuffer : SrcMgr.FindBufferContainingLoc(Loc); 584 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer(), 585 Loc.getPointer()); 586 } 587 588 const AsmToken &AsmParser::Lex() { 589 const AsmToken *tok = &Lexer.Lex(); 590 591 if (tok->is(AsmToken::Eof)) { 592 // If this is the end of an included file, pop the parent file off the 593 // include stack. 594 SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer); 595 if (ParentIncludeLoc != SMLoc()) { 596 jumpToLoc(ParentIncludeLoc); 597 tok = &Lexer.Lex(); 598 } 599 } 600 601 if (tok->is(AsmToken::Error)) 602 Error(Lexer.getErrLoc(), Lexer.getErr()); 603 604 return *tok; 605 } 606 607 bool AsmParser::Run(bool NoInitialTextSection, bool NoFinalize) { 608 // Create the initial section, if requested. 609 if (!NoInitialTextSection) 610 Out.InitSections(false); 611 612 // Prime the lexer. 613 Lex(); 614 615 HadError = false; 616 AsmCond StartingCondState = TheCondState; 617 618 // If we are generating dwarf for assembly source files save the initial text 619 // section and generate a .file directive. 620 if (getContext().getGenDwarfForAssembly()) { 621 MCSymbol *SectionStartSym = getContext().CreateTempSymbol(); 622 getStreamer().EmitLabel(SectionStartSym); 623 auto InsertResult = getContext().addGenDwarfSection( 624 getStreamer().getCurrentSection().first); 625 assert(InsertResult.second && ".text section should not have debug info yet"); 626 InsertResult.first->second.first = SectionStartSym; 627 getContext().setGenDwarfFileNumber(getStreamer().EmitDwarfFileDirective( 628 0, StringRef(), getContext().getMainFileName())); 629 } 630 631 // While we have input, parse each statement. 632 while (Lexer.isNot(AsmToken::Eof)) { 633 ParseStatementInfo Info; 634 if (!parseStatement(Info, nullptr)) 635 continue; 636 637 // We had an error, validate that one was emitted and recover by skipping to 638 // the next line. 639 assert(HadError && "Parse statement returned an error, but none emitted!"); 640 eatToEndOfStatement(); 641 } 642 643 if (TheCondState.TheCond != StartingCondState.TheCond || 644 TheCondState.Ignore != StartingCondState.Ignore) 645 return TokError("unmatched .ifs or .elses"); 646 647 // Check to see there are no empty DwarfFile slots. 648 const auto &LineTables = getContext().getMCDwarfLineTables(); 649 if (!LineTables.empty()) { 650 unsigned Index = 0; 651 for (const auto &File : LineTables.begin()->second.getMCDwarfFiles()) { 652 if (File.Name.empty() && Index != 0) 653 TokError("unassigned file number: " + Twine(Index) + 654 " for .file directives"); 655 ++Index; 656 } 657 } 658 659 // Check to see that all assembler local symbols were actually defined. 660 // Targets that don't do subsections via symbols may not want this, though, 661 // so conservatively exclude them. Only do this if we're finalizing, though, 662 // as otherwise we won't necessarilly have seen everything yet. 663 if (!NoFinalize && MAI.hasSubsectionsViaSymbols()) { 664 const MCContext::SymbolTable &Symbols = getContext().getSymbols(); 665 for (MCContext::SymbolTable::const_iterator i = Symbols.begin(), 666 e = Symbols.end(); 667 i != e; ++i) { 668 MCSymbol *Sym = i->getValue(); 669 // Variable symbols may not be marked as defined, so check those 670 // explicitly. If we know it's a variable, we have a definition for 671 // the purposes of this check. 672 if (Sym->isTemporary() && !Sym->isVariable() && !Sym->isDefined()) 673 // FIXME: We would really like to refer back to where the symbol was 674 // first referenced for a source location. We need to add something 675 // to track that. Currently, we just point to the end of the file. 676 printMessage( 677 getLexer().getLoc(), SourceMgr::DK_Error, 678 "assembler local symbol '" + Sym->getName() + "' not defined"); 679 } 680 } 681 682 // Finalize the output stream if there are no errors and if the client wants 683 // us to. 684 if (!HadError && !NoFinalize) 685 Out.Finish(); 686 687 return HadError; 688 } 689 690 void AsmParser::checkForValidSection() { 691 if (!ParsingInlineAsm && !getStreamer().getCurrentSection().first) { 692 TokError("expected section directive before assembly directive"); 693 Out.InitSections(false); 694 } 695 } 696 697 /// \brief Throw away the rest of the line for testing purposes. 698 void AsmParser::eatToEndOfStatement() { 699 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof)) 700 Lex(); 701 702 // Eat EOL. 703 if (Lexer.is(AsmToken::EndOfStatement)) 704 Lex(); 705 } 706 707 StringRef AsmParser::parseStringToEndOfStatement() { 708 const char *Start = getTok().getLoc().getPointer(); 709 710 while (Lexer.isNot(AsmToken::EndOfStatement) && Lexer.isNot(AsmToken::Eof)) 711 Lex(); 712 713 const char *End = getTok().getLoc().getPointer(); 714 return StringRef(Start, End - Start); 715 } 716 717 StringRef AsmParser::parseStringToComma() { 718 const char *Start = getTok().getLoc().getPointer(); 719 720 while (Lexer.isNot(AsmToken::EndOfStatement) && 721 Lexer.isNot(AsmToken::Comma) && Lexer.isNot(AsmToken::Eof)) 722 Lex(); 723 724 const char *End = getTok().getLoc().getPointer(); 725 return StringRef(Start, End - Start); 726 } 727 728 /// \brief Parse a paren expression and return it. 729 /// NOTE: This assumes the leading '(' has already been consumed. 730 /// 731 /// parenexpr ::= expr) 732 /// 733 bool AsmParser::parseParenExpr(const MCExpr *&Res, SMLoc &EndLoc) { 734 if (parseExpression(Res)) 735 return true; 736 if (Lexer.isNot(AsmToken::RParen)) 737 return TokError("expected ')' in parentheses expression"); 738 EndLoc = Lexer.getTok().getEndLoc(); 739 Lex(); 740 return false; 741 } 742 743 /// \brief Parse a bracket expression and return it. 744 /// NOTE: This assumes the leading '[' has already been consumed. 745 /// 746 /// bracketexpr ::= expr] 747 /// 748 bool AsmParser::parseBracketExpr(const MCExpr *&Res, SMLoc &EndLoc) { 749 if (parseExpression(Res)) 750 return true; 751 if (Lexer.isNot(AsmToken::RBrac)) 752 return TokError("expected ']' in brackets expression"); 753 EndLoc = Lexer.getTok().getEndLoc(); 754 Lex(); 755 return false; 756 } 757 758 /// \brief Parse a primary expression and return it. 759 /// primaryexpr ::= (parenexpr 760 /// primaryexpr ::= symbol 761 /// primaryexpr ::= number 762 /// primaryexpr ::= '.' 763 /// primaryexpr ::= ~,+,- primaryexpr 764 bool AsmParser::parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) { 765 SMLoc FirstTokenLoc = getLexer().getLoc(); 766 AsmToken::TokenKind FirstTokenKind = Lexer.getKind(); 767 switch (FirstTokenKind) { 768 default: 769 return TokError("unknown token in expression"); 770 // If we have an error assume that we've already handled it. 771 case AsmToken::Error: 772 return true; 773 case AsmToken::Exclaim: 774 Lex(); // Eat the operator. 775 if (parsePrimaryExpr(Res, EndLoc)) 776 return true; 777 Res = MCUnaryExpr::CreateLNot(Res, getContext()); 778 return false; 779 case AsmToken::Dollar: 780 case AsmToken::At: 781 case AsmToken::String: 782 case AsmToken::Identifier: { 783 StringRef Identifier; 784 if (parseIdentifier(Identifier)) { 785 if (FirstTokenKind == AsmToken::Dollar) { 786 if (Lexer.getMAI().getDollarIsPC()) { 787 // This is a '$' reference, which references the current PC. Emit a 788 // temporary label to the streamer and refer to it. 789 MCSymbol *Sym = Ctx.CreateTempSymbol(); 790 Out.EmitLabel(Sym); 791 Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, 792 getContext()); 793 EndLoc = FirstTokenLoc; 794 return false; 795 } 796 return Error(FirstTokenLoc, "invalid token in expression"); 797 } 798 } 799 // Parse symbol variant 800 std::pair<StringRef, StringRef> Split; 801 if (!MAI.useParensForSymbolVariant()) { 802 if (FirstTokenKind == AsmToken::String) { 803 if (Lexer.is(AsmToken::At)) { 804 Lexer.Lex(); // eat @ 805 SMLoc AtLoc = getLexer().getLoc(); 806 StringRef VName; 807 if (parseIdentifier(VName)) 808 return Error(AtLoc, "expected symbol variant after '@'"); 809 810 Split = std::make_pair(Identifier, VName); 811 } 812 } else { 813 Split = Identifier.split('@'); 814 } 815 } else if (Lexer.is(AsmToken::LParen)) { 816 Lexer.Lex(); // eat ( 817 StringRef VName; 818 parseIdentifier(VName); 819 if (Lexer.isNot(AsmToken::RParen)) { 820 return Error(Lexer.getTok().getLoc(), 821 "unexpected token in variant, expected ')'"); 822 } 823 Lexer.Lex(); // eat ) 824 Split = std::make_pair(Identifier, VName); 825 } 826 827 EndLoc = SMLoc::getFromPointer(Identifier.end()); 828 829 // This is a symbol reference. 830 StringRef SymbolName = Identifier; 831 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 832 833 // Lookup the symbol variant if used. 834 if (Split.second.size()) { 835 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second); 836 if (Variant != MCSymbolRefExpr::VK_Invalid) { 837 SymbolName = Split.first; 838 } else if (MAI.doesAllowAtInName() && !MAI.useParensForSymbolVariant()) { 839 Variant = MCSymbolRefExpr::VK_None; 840 } else { 841 return Error(SMLoc::getFromPointer(Split.second.begin()), 842 "invalid variant '" + Split.second + "'"); 843 } 844 } 845 846 MCSymbol *Sym = getContext().GetOrCreateSymbol(SymbolName); 847 848 // If this is an absolute variable reference, substitute it now to preserve 849 // semantics in the face of reassignment. 850 if (Sym->isVariable() && isa<MCConstantExpr>(Sym->getVariableValue())) { 851 if (Variant) 852 return Error(EndLoc, "unexpected modifier on variable reference"); 853 854 Res = Sym->getVariableValue(); 855 return false; 856 } 857 858 // Otherwise create a symbol ref. 859 Res = MCSymbolRefExpr::Create(Sym, Variant, getContext()); 860 return false; 861 } 862 case AsmToken::BigNum: 863 return TokError("literal value out of range for directive"); 864 case AsmToken::Integer: { 865 SMLoc Loc = getTok().getLoc(); 866 int64_t IntVal = getTok().getIntVal(); 867 Res = MCConstantExpr::Create(IntVal, getContext()); 868 EndLoc = Lexer.getTok().getEndLoc(); 869 Lex(); // Eat token. 870 // Look for 'b' or 'f' following an Integer as a directional label 871 if (Lexer.getKind() == AsmToken::Identifier) { 872 StringRef IDVal = getTok().getString(); 873 // Lookup the symbol variant if used. 874 std::pair<StringRef, StringRef> Split = IDVal.split('@'); 875 MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; 876 if (Split.first.size() != IDVal.size()) { 877 Variant = MCSymbolRefExpr::getVariantKindForName(Split.second); 878 if (Variant == MCSymbolRefExpr::VK_Invalid) 879 return TokError("invalid variant '" + Split.second + "'"); 880 IDVal = Split.first; 881 } 882 if (IDVal == "f" || IDVal == "b") { 883 MCSymbol *Sym = 884 Ctx.GetDirectionalLocalSymbol(IntVal, IDVal == "b"); 885 Res = MCSymbolRefExpr::Create(Sym, Variant, getContext()); 886 if (IDVal == "b" && Sym->isUndefined()) 887 return Error(Loc, "invalid reference to undefined symbol"); 888 EndLoc = Lexer.getTok().getEndLoc(); 889 Lex(); // Eat identifier. 890 } 891 } 892 return false; 893 } 894 case AsmToken::Real: { 895 APFloat RealVal(APFloat::IEEEdouble, getTok().getString()); 896 uint64_t IntVal = RealVal.bitcastToAPInt().getZExtValue(); 897 Res = MCConstantExpr::Create(IntVal, getContext()); 898 EndLoc = Lexer.getTok().getEndLoc(); 899 Lex(); // Eat token. 900 return false; 901 } 902 case AsmToken::Dot: { 903 // This is a '.' reference, which references the current PC. Emit a 904 // temporary label to the streamer and refer to it. 905 MCSymbol *Sym = Ctx.CreateTempSymbol(); 906 Out.EmitLabel(Sym); 907 Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, getContext()); 908 EndLoc = Lexer.getTok().getEndLoc(); 909 Lex(); // Eat identifier. 910 return false; 911 } 912 case AsmToken::LParen: 913 Lex(); // Eat the '('. 914 return parseParenExpr(Res, EndLoc); 915 case AsmToken::LBrac: 916 if (!PlatformParser->HasBracketExpressions()) 917 return TokError("brackets expression not supported on this target"); 918 Lex(); // Eat the '['. 919 return parseBracketExpr(Res, EndLoc); 920 case AsmToken::Minus: 921 Lex(); // Eat the operator. 922 if (parsePrimaryExpr(Res, EndLoc)) 923 return true; 924 Res = MCUnaryExpr::CreateMinus(Res, getContext()); 925 return false; 926 case AsmToken::Plus: 927 Lex(); // Eat the operator. 928 if (parsePrimaryExpr(Res, EndLoc)) 929 return true; 930 Res = MCUnaryExpr::CreatePlus(Res, getContext()); 931 return false; 932 case AsmToken::Tilde: 933 Lex(); // Eat the operator. 934 if (parsePrimaryExpr(Res, EndLoc)) 935 return true; 936 Res = MCUnaryExpr::CreateNot(Res, getContext()); 937 return false; 938 } 939 } 940 941 bool AsmParser::parseExpression(const MCExpr *&Res) { 942 SMLoc EndLoc; 943 return parseExpression(Res, EndLoc); 944 } 945 946 const MCExpr * 947 AsmParser::applyModifierToExpr(const MCExpr *E, 948 MCSymbolRefExpr::VariantKind Variant) { 949 // Ask the target implementation about this expression first. 950 const MCExpr *NewE = getTargetParser().applyModifierToExpr(E, Variant, Ctx); 951 if (NewE) 952 return NewE; 953 // Recurse over the given expression, rebuilding it to apply the given variant 954 // if there is exactly one symbol. 955 switch (E->getKind()) { 956 case MCExpr::Target: 957 case MCExpr::Constant: 958 return nullptr; 959 960 case MCExpr::SymbolRef: { 961 const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(E); 962 963 if (SRE->getKind() != MCSymbolRefExpr::VK_None) { 964 TokError("invalid variant on expression '" + getTok().getIdentifier() + 965 "' (already modified)"); 966 return E; 967 } 968 969 return MCSymbolRefExpr::Create(&SRE->getSymbol(), Variant, getContext()); 970 } 971 972 case MCExpr::Unary: { 973 const MCUnaryExpr *UE = cast<MCUnaryExpr>(E); 974 const MCExpr *Sub = applyModifierToExpr(UE->getSubExpr(), Variant); 975 if (!Sub) 976 return nullptr; 977 return MCUnaryExpr::Create(UE->getOpcode(), Sub, getContext()); 978 } 979 980 case MCExpr::Binary: { 981 const MCBinaryExpr *BE = cast<MCBinaryExpr>(E); 982 const MCExpr *LHS = applyModifierToExpr(BE->getLHS(), Variant); 983 const MCExpr *RHS = applyModifierToExpr(BE->getRHS(), Variant); 984 985 if (!LHS && !RHS) 986 return nullptr; 987 988 if (!LHS) 989 LHS = BE->getLHS(); 990 if (!RHS) 991 RHS = BE->getRHS(); 992 993 return MCBinaryExpr::Create(BE->getOpcode(), LHS, RHS, getContext()); 994 } 995 } 996 997 llvm_unreachable("Invalid expression kind!"); 998 } 999 1000 /// \brief Parse an expression and return it. 1001 /// 1002 /// expr ::= expr &&,|| expr -> lowest. 1003 /// expr ::= expr |,^,&,! expr 1004 /// expr ::= expr ==,!=,<>,<,<=,>,>= expr 1005 /// expr ::= expr <<,>> expr 1006 /// expr ::= expr +,- expr 1007 /// expr ::= expr *,/,% expr -> highest. 1008 /// expr ::= primaryexpr 1009 /// 1010 bool AsmParser::parseExpression(const MCExpr *&Res, SMLoc &EndLoc) { 1011 // Parse the expression. 1012 Res = nullptr; 1013 if (parsePrimaryExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc)) 1014 return true; 1015 1016 // As a special case, we support 'a op b @ modifier' by rewriting the 1017 // expression to include the modifier. This is inefficient, but in general we 1018 // expect users to use 'a@modifier op b'. 1019 if (Lexer.getKind() == AsmToken::At) { 1020 Lex(); 1021 1022 if (Lexer.isNot(AsmToken::Identifier)) 1023 return TokError("unexpected symbol modifier following '@'"); 1024 1025 MCSymbolRefExpr::VariantKind Variant = 1026 MCSymbolRefExpr::getVariantKindForName(getTok().getIdentifier()); 1027 if (Variant == MCSymbolRefExpr::VK_Invalid) 1028 return TokError("invalid variant '" + getTok().getIdentifier() + "'"); 1029 1030 const MCExpr *ModifiedRes = applyModifierToExpr(Res, Variant); 1031 if (!ModifiedRes) { 1032 return TokError("invalid modifier '" + getTok().getIdentifier() + 1033 "' (no symbols present)"); 1034 } 1035 1036 Res = ModifiedRes; 1037 Lex(); 1038 } 1039 1040 // Try to constant fold it up front, if possible. 1041 int64_t Value; 1042 if (Res->EvaluateAsAbsolute(Value)) 1043 Res = MCConstantExpr::Create(Value, getContext()); 1044 1045 return false; 1046 } 1047 1048 bool AsmParser::parseParenExpression(const MCExpr *&Res, SMLoc &EndLoc) { 1049 Res = nullptr; 1050 return parseParenExpr(Res, EndLoc) || parseBinOpRHS(1, Res, EndLoc); 1051 } 1052 1053 bool AsmParser::parseAbsoluteExpression(int64_t &Res) { 1054 const MCExpr *Expr; 1055 1056 SMLoc StartLoc = Lexer.getLoc(); 1057 if (parseExpression(Expr)) 1058 return true; 1059 1060 if (!Expr->EvaluateAsAbsolute(Res)) 1061 return Error(StartLoc, "expected absolute expression"); 1062 1063 return false; 1064 } 1065 1066 static unsigned getBinOpPrecedence(AsmToken::TokenKind K, 1067 MCBinaryExpr::Opcode &Kind) { 1068 switch (K) { 1069 default: 1070 return 0; // not a binop. 1071 1072 // Lowest Precedence: &&, || 1073 case AsmToken::AmpAmp: 1074 Kind = MCBinaryExpr::LAnd; 1075 return 1; 1076 case AsmToken::PipePipe: 1077 Kind = MCBinaryExpr::LOr; 1078 return 1; 1079 1080 // Low Precedence: |, &, ^ 1081 // 1082 // FIXME: gas seems to support '!' as an infix operator? 1083 case AsmToken::Pipe: 1084 Kind = MCBinaryExpr::Or; 1085 return 2; 1086 case AsmToken::Caret: 1087 Kind = MCBinaryExpr::Xor; 1088 return 2; 1089 case AsmToken::Amp: 1090 Kind = MCBinaryExpr::And; 1091 return 2; 1092 1093 // Low Intermediate Precedence: ==, !=, <>, <, <=, >, >= 1094 case AsmToken::EqualEqual: 1095 Kind = MCBinaryExpr::EQ; 1096 return 3; 1097 case AsmToken::ExclaimEqual: 1098 case AsmToken::LessGreater: 1099 Kind = MCBinaryExpr::NE; 1100 return 3; 1101 case AsmToken::Less: 1102 Kind = MCBinaryExpr::LT; 1103 return 3; 1104 case AsmToken::LessEqual: 1105 Kind = MCBinaryExpr::LTE; 1106 return 3; 1107 case AsmToken::Greater: 1108 Kind = MCBinaryExpr::GT; 1109 return 3; 1110 case AsmToken::GreaterEqual: 1111 Kind = MCBinaryExpr::GTE; 1112 return 3; 1113 1114 // Intermediate Precedence: <<, >> 1115 case AsmToken::LessLess: 1116 Kind = MCBinaryExpr::Shl; 1117 return 4; 1118 case AsmToken::GreaterGreater: 1119 Kind = MCBinaryExpr::Shr; 1120 return 4; 1121 1122 // High Intermediate Precedence: +, - 1123 case AsmToken::Plus: 1124 Kind = MCBinaryExpr::Add; 1125 return 5; 1126 case AsmToken::Minus: 1127 Kind = MCBinaryExpr::Sub; 1128 return 5; 1129 1130 // Highest Precedence: *, /, % 1131 case AsmToken::Star: 1132 Kind = MCBinaryExpr::Mul; 1133 return 6; 1134 case AsmToken::Slash: 1135 Kind = MCBinaryExpr::Div; 1136 return 6; 1137 case AsmToken::Percent: 1138 Kind = MCBinaryExpr::Mod; 1139 return 6; 1140 } 1141 } 1142 1143 /// \brief Parse all binary operators with precedence >= 'Precedence'. 1144 /// Res contains the LHS of the expression on input. 1145 bool AsmParser::parseBinOpRHS(unsigned Precedence, const MCExpr *&Res, 1146 SMLoc &EndLoc) { 1147 while (1) { 1148 MCBinaryExpr::Opcode Kind = MCBinaryExpr::Add; 1149 unsigned TokPrec = getBinOpPrecedence(Lexer.getKind(), Kind); 1150 1151 // If the next token is lower precedence than we are allowed to eat, return 1152 // successfully with what we ate already. 1153 if (TokPrec < Precedence) 1154 return false; 1155 1156 Lex(); 1157 1158 // Eat the next primary expression. 1159 const MCExpr *RHS; 1160 if (parsePrimaryExpr(RHS, EndLoc)) 1161 return true; 1162 1163 // If BinOp binds less tightly with RHS than the operator after RHS, let 1164 // the pending operator take RHS as its LHS. 1165 MCBinaryExpr::Opcode Dummy; 1166 unsigned NextTokPrec = getBinOpPrecedence(Lexer.getKind(), Dummy); 1167 if (TokPrec < NextTokPrec && parseBinOpRHS(TokPrec + 1, RHS, EndLoc)) 1168 return true; 1169 1170 // Merge LHS and RHS according to operator. 1171 Res = MCBinaryExpr::Create(Kind, Res, RHS, getContext()); 1172 } 1173 } 1174 1175 /// ParseStatement: 1176 /// ::= EndOfStatement 1177 /// ::= Label* Directive ...Operands... EndOfStatement 1178 /// ::= Label* Identifier OperandList* EndOfStatement 1179 bool AsmParser::parseStatement(ParseStatementInfo &Info, 1180 MCAsmParserSemaCallback *SI) { 1181 if (Lexer.is(AsmToken::EndOfStatement)) { 1182 Out.AddBlankLine(); 1183 Lex(); 1184 return false; 1185 } 1186 1187 // Statements always start with an identifier or are a full line comment. 1188 AsmToken ID = getTok(); 1189 SMLoc IDLoc = ID.getLoc(); 1190 StringRef IDVal; 1191 int64_t LocalLabelVal = -1; 1192 // A full line comment is a '#' as the first token. 1193 if (Lexer.is(AsmToken::Hash)) 1194 return parseCppHashLineFilenameComment(IDLoc); 1195 1196 // Allow an integer followed by a ':' as a directional local label. 1197 if (Lexer.is(AsmToken::Integer)) { 1198 LocalLabelVal = getTok().getIntVal(); 1199 if (LocalLabelVal < 0) { 1200 if (!TheCondState.Ignore) 1201 return TokError("unexpected token at start of statement"); 1202 IDVal = ""; 1203 } else { 1204 IDVal = getTok().getString(); 1205 Lex(); // Consume the integer token to be used as an identifier token. 1206 if (Lexer.getKind() != AsmToken::Colon) { 1207 if (!TheCondState.Ignore) 1208 return TokError("unexpected token at start of statement"); 1209 } 1210 } 1211 } else if (Lexer.is(AsmToken::Dot)) { 1212 // Treat '.' as a valid identifier in this context. 1213 Lex(); 1214 IDVal = "."; 1215 } else if (parseIdentifier(IDVal)) { 1216 if (!TheCondState.Ignore) 1217 return TokError("unexpected token at start of statement"); 1218 IDVal = ""; 1219 } 1220 1221 // Handle conditional assembly here before checking for skipping. We 1222 // have to do this so that .endif isn't skipped in a ".if 0" block for 1223 // example. 1224 StringMap<DirectiveKind>::const_iterator DirKindIt = 1225 DirectiveKindMap.find(IDVal); 1226 DirectiveKind DirKind = (DirKindIt == DirectiveKindMap.end()) 1227 ? DK_NO_DIRECTIVE 1228 : DirKindIt->getValue(); 1229 switch (DirKind) { 1230 default: 1231 break; 1232 case DK_IF: 1233 case DK_IFEQ: 1234 case DK_IFGE: 1235 case DK_IFGT: 1236 case DK_IFLE: 1237 case DK_IFLT: 1238 case DK_IFNE: 1239 return parseDirectiveIf(IDLoc, DirKind); 1240 case DK_IFB: 1241 return parseDirectiveIfb(IDLoc, true); 1242 case DK_IFNB: 1243 return parseDirectiveIfb(IDLoc, false); 1244 case DK_IFC: 1245 return parseDirectiveIfc(IDLoc, true); 1246 case DK_IFEQS: 1247 return parseDirectiveIfeqs(IDLoc, true); 1248 case DK_IFNC: 1249 return parseDirectiveIfc(IDLoc, false); 1250 case DK_IFNES: 1251 return parseDirectiveIfeqs(IDLoc, false); 1252 case DK_IFDEF: 1253 return parseDirectiveIfdef(IDLoc, true); 1254 case DK_IFNDEF: 1255 case DK_IFNOTDEF: 1256 return parseDirectiveIfdef(IDLoc, false); 1257 case DK_ELSEIF: 1258 return parseDirectiveElseIf(IDLoc); 1259 case DK_ELSE: 1260 return parseDirectiveElse(IDLoc); 1261 case DK_ENDIF: 1262 return parseDirectiveEndIf(IDLoc); 1263 } 1264 1265 // Ignore the statement if in the middle of inactive conditional 1266 // (e.g. ".if 0"). 1267 if (TheCondState.Ignore) { 1268 eatToEndOfStatement(); 1269 return false; 1270 } 1271 1272 // FIXME: Recurse on local labels? 1273 1274 // See what kind of statement we have. 1275 switch (Lexer.getKind()) { 1276 case AsmToken::Colon: { 1277 checkForValidSection(); 1278 1279 // identifier ':' -> Label. 1280 Lex(); 1281 1282 // Diagnose attempt to use '.' as a label. 1283 if (IDVal == ".") 1284 return Error(IDLoc, "invalid use of pseudo-symbol '.' as a label"); 1285 1286 // Diagnose attempt to use a variable as a label. 1287 // 1288 // FIXME: Diagnostics. Note the location of the definition as a label. 1289 // FIXME: This doesn't diagnose assignment to a symbol which has been 1290 // implicitly marked as external. 1291 MCSymbol *Sym; 1292 if (LocalLabelVal == -1) { 1293 if (ParsingInlineAsm && SI) { 1294 StringRef RewrittenLabel = SI->LookupInlineAsmLabel(IDVal, getSourceManager(), IDLoc, true); 1295 assert(RewrittenLabel.size() && "We should have an internal name here."); 1296 Info.AsmRewrites->push_back(AsmRewrite(AOK_Label, IDLoc, 1297 IDVal.size(), RewrittenLabel)); 1298 IDVal = RewrittenLabel; 1299 } 1300 Sym = getContext().GetOrCreateSymbol(IDVal); 1301 } else 1302 Sym = Ctx.CreateDirectionalLocalSymbol(LocalLabelVal); 1303 1304 Sym->redefineIfPossible(); 1305 1306 if (!Sym->isUndefined() || Sym->isVariable()) 1307 return Error(IDLoc, "invalid symbol redefinition"); 1308 1309 // Emit the label. 1310 if (!ParsingInlineAsm) 1311 Out.EmitLabel(Sym); 1312 1313 // If we are generating dwarf for assembly source files then gather the 1314 // info to make a dwarf label entry for this label if needed. 1315 if (getContext().getGenDwarfForAssembly()) 1316 MCGenDwarfLabelEntry::Make(Sym, &getStreamer(), getSourceManager(), 1317 IDLoc); 1318 1319 getTargetParser().onLabelParsed(Sym); 1320 1321 // Consume any end of statement token, if present, to avoid spurious 1322 // AddBlankLine calls(). 1323 if (Lexer.is(AsmToken::EndOfStatement)) { 1324 Lex(); 1325 if (Lexer.is(AsmToken::Eof)) 1326 return false; 1327 } 1328 1329 return false; 1330 } 1331 1332 case AsmToken::Equal: 1333 // identifier '=' ... -> assignment statement 1334 Lex(); 1335 1336 return parseAssignment(IDVal, true); 1337 1338 default: // Normal instruction or directive. 1339 break; 1340 } 1341 1342 // If macros are enabled, check to see if this is a macro instantiation. 1343 if (areMacrosEnabled()) 1344 if (const MCAsmMacro *M = lookupMacro(IDVal)) { 1345 return handleMacroEntry(M, IDLoc); 1346 } 1347 1348 // Otherwise, we have a normal instruction or directive. 1349 1350 // Directives start with "." 1351 if (IDVal[0] == '.' && IDVal != ".") { 1352 // There are several entities interested in parsing directives: 1353 // 1354 // 1. The target-specific assembly parser. Some directives are target 1355 // specific or may potentially behave differently on certain targets. 1356 // 2. Asm parser extensions. For example, platform-specific parsers 1357 // (like the ELF parser) register themselves as extensions. 1358 // 3. The generic directive parser implemented by this class. These are 1359 // all the directives that behave in a target and platform independent 1360 // manner, or at least have a default behavior that's shared between 1361 // all targets and platforms. 1362 1363 // First query the target-specific parser. It will return 'true' if it 1364 // isn't interested in this directive. 1365 if (!getTargetParser().ParseDirective(ID)) 1366 return false; 1367 1368 // Next, check the extension directive map to see if any extension has 1369 // registered itself to parse this directive. 1370 std::pair<MCAsmParserExtension *, DirectiveHandler> Handler = 1371 ExtensionDirectiveMap.lookup(IDVal); 1372 if (Handler.first) 1373 return (*Handler.second)(Handler.first, IDVal, IDLoc); 1374 1375 // Finally, if no one else is interested in this directive, it must be 1376 // generic and familiar to this class. 1377 switch (DirKind) { 1378 default: 1379 break; 1380 case DK_SET: 1381 case DK_EQU: 1382 return parseDirectiveSet(IDVal, true); 1383 case DK_EQUIV: 1384 return parseDirectiveSet(IDVal, false); 1385 case DK_ASCII: 1386 return parseDirectiveAscii(IDVal, false); 1387 case DK_ASCIZ: 1388 case DK_STRING: 1389 return parseDirectiveAscii(IDVal, true); 1390 case DK_BYTE: 1391 return parseDirectiveValue(1); 1392 case DK_SHORT: 1393 case DK_VALUE: 1394 case DK_2BYTE: 1395 return parseDirectiveValue(2); 1396 case DK_LONG: 1397 case DK_INT: 1398 case DK_4BYTE: 1399 return parseDirectiveValue(4); 1400 case DK_QUAD: 1401 case DK_8BYTE: 1402 return parseDirectiveValue(8); 1403 case DK_OCTA: 1404 return parseDirectiveOctaValue(); 1405 case DK_SINGLE: 1406 case DK_FLOAT: 1407 return parseDirectiveRealValue(APFloat::IEEEsingle); 1408 case DK_DOUBLE: 1409 return parseDirectiveRealValue(APFloat::IEEEdouble); 1410 case DK_ALIGN: { 1411 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes(); 1412 return parseDirectiveAlign(IsPow2, /*ExprSize=*/1); 1413 } 1414 case DK_ALIGN32: { 1415 bool IsPow2 = !getContext().getAsmInfo()->getAlignmentIsInBytes(); 1416 return parseDirectiveAlign(IsPow2, /*ExprSize=*/4); 1417 } 1418 case DK_BALIGN: 1419 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/1); 1420 case DK_BALIGNW: 1421 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/2); 1422 case DK_BALIGNL: 1423 return parseDirectiveAlign(/*IsPow2=*/false, /*ExprSize=*/4); 1424 case DK_P2ALIGN: 1425 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/1); 1426 case DK_P2ALIGNW: 1427 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/2); 1428 case DK_P2ALIGNL: 1429 return parseDirectiveAlign(/*IsPow2=*/true, /*ExprSize=*/4); 1430 case DK_ORG: 1431 return parseDirectiveOrg(); 1432 case DK_FILL: 1433 return parseDirectiveFill(); 1434 case DK_ZERO: 1435 return parseDirectiveZero(); 1436 case DK_EXTERN: 1437 eatToEndOfStatement(); // .extern is the default, ignore it. 1438 return false; 1439 case DK_GLOBL: 1440 case DK_GLOBAL: 1441 return parseDirectiveSymbolAttribute(MCSA_Global); 1442 case DK_LAZY_REFERENCE: 1443 return parseDirectiveSymbolAttribute(MCSA_LazyReference); 1444 case DK_NO_DEAD_STRIP: 1445 return parseDirectiveSymbolAttribute(MCSA_NoDeadStrip); 1446 case DK_SYMBOL_RESOLVER: 1447 return parseDirectiveSymbolAttribute(MCSA_SymbolResolver); 1448 case DK_PRIVATE_EXTERN: 1449 return parseDirectiveSymbolAttribute(MCSA_PrivateExtern); 1450 case DK_REFERENCE: 1451 return parseDirectiveSymbolAttribute(MCSA_Reference); 1452 case DK_WEAK_DEFINITION: 1453 return parseDirectiveSymbolAttribute(MCSA_WeakDefinition); 1454 case DK_WEAK_REFERENCE: 1455 return parseDirectiveSymbolAttribute(MCSA_WeakReference); 1456 case DK_WEAK_DEF_CAN_BE_HIDDEN: 1457 return parseDirectiveSymbolAttribute(MCSA_WeakDefAutoPrivate); 1458 case DK_COMM: 1459 case DK_COMMON: 1460 return parseDirectiveComm(/*IsLocal=*/false); 1461 case DK_LCOMM: 1462 return parseDirectiveComm(/*IsLocal=*/true); 1463 case DK_ABORT: 1464 return parseDirectiveAbort(); 1465 case DK_INCLUDE: 1466 return parseDirectiveInclude(); 1467 case DK_INCBIN: 1468 return parseDirectiveIncbin(); 1469 case DK_CODE16: 1470 case DK_CODE16GCC: 1471 return TokError(Twine(IDVal) + " not supported yet"); 1472 case DK_REPT: 1473 return parseDirectiveRept(IDLoc, IDVal); 1474 case DK_IRP: 1475 return parseDirectiveIrp(IDLoc); 1476 case DK_IRPC: 1477 return parseDirectiveIrpc(IDLoc); 1478 case DK_ENDR: 1479 return parseDirectiveEndr(IDLoc); 1480 case DK_BUNDLE_ALIGN_MODE: 1481 return parseDirectiveBundleAlignMode(); 1482 case DK_BUNDLE_LOCK: 1483 return parseDirectiveBundleLock(); 1484 case DK_BUNDLE_UNLOCK: 1485 return parseDirectiveBundleUnlock(); 1486 case DK_SLEB128: 1487 return parseDirectiveLEB128(true); 1488 case DK_ULEB128: 1489 return parseDirectiveLEB128(false); 1490 case DK_SPACE: 1491 case DK_SKIP: 1492 return parseDirectiveSpace(IDVal); 1493 case DK_FILE: 1494 return parseDirectiveFile(IDLoc); 1495 case DK_LINE: 1496 return parseDirectiveLine(); 1497 case DK_LOC: 1498 return parseDirectiveLoc(); 1499 case DK_STABS: 1500 return parseDirectiveStabs(); 1501 case DK_CFI_SECTIONS: 1502 return parseDirectiveCFISections(); 1503 case DK_CFI_STARTPROC: 1504 return parseDirectiveCFIStartProc(); 1505 case DK_CFI_ENDPROC: 1506 return parseDirectiveCFIEndProc(); 1507 case DK_CFI_DEF_CFA: 1508 return parseDirectiveCFIDefCfa(IDLoc); 1509 case DK_CFI_DEF_CFA_OFFSET: 1510 return parseDirectiveCFIDefCfaOffset(); 1511 case DK_CFI_ADJUST_CFA_OFFSET: 1512 return parseDirectiveCFIAdjustCfaOffset(); 1513 case DK_CFI_DEF_CFA_REGISTER: 1514 return parseDirectiveCFIDefCfaRegister(IDLoc); 1515 case DK_CFI_OFFSET: 1516 return parseDirectiveCFIOffset(IDLoc); 1517 case DK_CFI_REL_OFFSET: 1518 return parseDirectiveCFIRelOffset(IDLoc); 1519 case DK_CFI_PERSONALITY: 1520 return parseDirectiveCFIPersonalityOrLsda(true); 1521 case DK_CFI_LSDA: 1522 return parseDirectiveCFIPersonalityOrLsda(false); 1523 case DK_CFI_REMEMBER_STATE: 1524 return parseDirectiveCFIRememberState(); 1525 case DK_CFI_RESTORE_STATE: 1526 return parseDirectiveCFIRestoreState(); 1527 case DK_CFI_SAME_VALUE: 1528 return parseDirectiveCFISameValue(IDLoc); 1529 case DK_CFI_RESTORE: 1530 return parseDirectiveCFIRestore(IDLoc); 1531 case DK_CFI_ESCAPE: 1532 return parseDirectiveCFIEscape(); 1533 case DK_CFI_SIGNAL_FRAME: 1534 return parseDirectiveCFISignalFrame(); 1535 case DK_CFI_UNDEFINED: 1536 return parseDirectiveCFIUndefined(IDLoc); 1537 case DK_CFI_REGISTER: 1538 return parseDirectiveCFIRegister(IDLoc); 1539 case DK_CFI_WINDOW_SAVE: 1540 return parseDirectiveCFIWindowSave(); 1541 case DK_MACROS_ON: 1542 case DK_MACROS_OFF: 1543 return parseDirectiveMacrosOnOff(IDVal); 1544 case DK_MACRO: 1545 return parseDirectiveMacro(IDLoc); 1546 case DK_EXITM: 1547 return parseDirectiveExitMacro(IDVal); 1548 case DK_ENDM: 1549 case DK_ENDMACRO: 1550 return parseDirectiveEndMacro(IDVal); 1551 case DK_PURGEM: 1552 return parseDirectivePurgeMacro(IDLoc); 1553 case DK_END: 1554 return parseDirectiveEnd(IDLoc); 1555 case DK_ERR: 1556 return parseDirectiveError(IDLoc, false); 1557 case DK_ERROR: 1558 return parseDirectiveError(IDLoc, true); 1559 case DK_WARNING: 1560 return parseDirectiveWarning(IDLoc); 1561 } 1562 1563 return Error(IDLoc, "unknown directive"); 1564 } 1565 1566 // __asm _emit or __asm __emit 1567 if (ParsingInlineAsm && (IDVal == "_emit" || IDVal == "__emit" || 1568 IDVal == "_EMIT" || IDVal == "__EMIT")) 1569 return parseDirectiveMSEmit(IDLoc, Info, IDVal.size()); 1570 1571 // __asm align 1572 if (ParsingInlineAsm && (IDVal == "align" || IDVal == "ALIGN")) 1573 return parseDirectiveMSAlign(IDLoc, Info); 1574 1575 checkForValidSection(); 1576 1577 // Canonicalize the opcode to lower case. 1578 std::string OpcodeStr = IDVal.lower(); 1579 ParseInstructionInfo IInfo(Info.AsmRewrites); 1580 bool HadError = getTargetParser().ParseInstruction(IInfo, OpcodeStr, IDLoc, 1581 Info.ParsedOperands); 1582 Info.ParseError = HadError; 1583 1584 // Dump the parsed representation, if requested. 1585 if (getShowParsedOperands()) { 1586 SmallString<256> Str; 1587 raw_svector_ostream OS(Str); 1588 OS << "parsed instruction: ["; 1589 for (unsigned i = 0; i != Info.ParsedOperands.size(); ++i) { 1590 if (i != 0) 1591 OS << ", "; 1592 Info.ParsedOperands[i]->print(OS); 1593 } 1594 OS << "]"; 1595 1596 printMessage(IDLoc, SourceMgr::DK_Note, OS.str()); 1597 } 1598 1599 // If we are generating dwarf for the current section then generate a .loc 1600 // directive for the instruction. 1601 if (!HadError && getContext().getGenDwarfForAssembly() && 1602 getContext().getGenDwarfSectionSyms().count( 1603 getStreamer().getCurrentSection().first)) { 1604 unsigned Line; 1605 if (ActiveMacros.empty()) 1606 Line = SrcMgr.FindLineNumber(IDLoc, CurBuffer); 1607 else 1608 Line = SrcMgr.FindLineNumber(ActiveMacros.back()->InstantiationLoc, 1609 ActiveMacros.back()->ExitBuffer); 1610 1611 // If we previously parsed a cpp hash file line comment then make sure the 1612 // current Dwarf File is for the CppHashFilename if not then emit the 1613 // Dwarf File table for it and adjust the line number for the .loc. 1614 if (CppHashFilename.size()) { 1615 unsigned FileNumber = getStreamer().EmitDwarfFileDirective( 1616 0, StringRef(), CppHashFilename); 1617 getContext().setGenDwarfFileNumber(FileNumber); 1618 1619 // Since SrcMgr.FindLineNumber() is slow and messes up the SourceMgr's 1620 // cache with the different Loc from the call above we save the last 1621 // info we queried here with SrcMgr.FindLineNumber(). 1622 unsigned CppHashLocLineNo; 1623 if (LastQueryIDLoc == CppHashLoc && LastQueryBuffer == CppHashBuf) 1624 CppHashLocLineNo = LastQueryLine; 1625 else { 1626 CppHashLocLineNo = SrcMgr.FindLineNumber(CppHashLoc, CppHashBuf); 1627 LastQueryLine = CppHashLocLineNo; 1628 LastQueryIDLoc = CppHashLoc; 1629 LastQueryBuffer = CppHashBuf; 1630 } 1631 Line = CppHashLineNumber - 1 + (Line - CppHashLocLineNo); 1632 } 1633 1634 getStreamer().EmitDwarfLocDirective( 1635 getContext().getGenDwarfFileNumber(), Line, 0, 1636 DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0, 0, 0, 1637 StringRef()); 1638 } 1639 1640 // If parsing succeeded, match the instruction. 1641 if (!HadError) { 1642 uint64_t ErrorInfo; 1643 getTargetParser().MatchAndEmitInstruction(IDLoc, Info.Opcode, 1644 Info.ParsedOperands, Out, 1645 ErrorInfo, ParsingInlineAsm); 1646 } 1647 1648 // Don't skip the rest of the line, the instruction parser is responsible for 1649 // that. 1650 return false; 1651 } 1652 1653 /// eatToEndOfLine uses the Lexer to eat the characters to the end of the line 1654 /// since they may not be able to be tokenized to get to the end of line token. 1655 void AsmParser::eatToEndOfLine() { 1656 if (!Lexer.is(AsmToken::EndOfStatement)) 1657 Lexer.LexUntilEndOfLine(); 1658 // Eat EOL. 1659 Lex(); 1660 } 1661 1662 /// parseCppHashLineFilenameComment as this: 1663 /// ::= # number "filename" 1664 /// or just as a full line comment if it doesn't have a number and a string. 1665 bool AsmParser::parseCppHashLineFilenameComment(const SMLoc &L) { 1666 Lex(); // Eat the hash token. 1667 1668 if (getLexer().isNot(AsmToken::Integer)) { 1669 // Consume the line since in cases it is not a well-formed line directive, 1670 // as if were simply a full line comment. 1671 eatToEndOfLine(); 1672 return false; 1673 } 1674 1675 int64_t LineNumber = getTok().getIntVal(); 1676 Lex(); 1677 1678 if (getLexer().isNot(AsmToken::String)) { 1679 eatToEndOfLine(); 1680 return false; 1681 } 1682 1683 StringRef Filename = getTok().getString(); 1684 // Get rid of the enclosing quotes. 1685 Filename = Filename.substr(1, Filename.size() - 2); 1686 1687 // Save the SMLoc, Filename and LineNumber for later use by diagnostics. 1688 CppHashLoc = L; 1689 CppHashFilename = Filename; 1690 CppHashLineNumber = LineNumber; 1691 CppHashBuf = CurBuffer; 1692 1693 // Ignore any trailing characters, they're just comment. 1694 eatToEndOfLine(); 1695 return false; 1696 } 1697 1698 /// \brief will use the last parsed cpp hash line filename comment 1699 /// for the Filename and LineNo if any in the diagnostic. 1700 void AsmParser::DiagHandler(const SMDiagnostic &Diag, void *Context) { 1701 const AsmParser *Parser = static_cast<const AsmParser *>(Context); 1702 raw_ostream &OS = errs(); 1703 1704 const SourceMgr &DiagSrcMgr = *Diag.getSourceMgr(); 1705 const SMLoc &DiagLoc = Diag.getLoc(); 1706 unsigned DiagBuf = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 1707 unsigned CppHashBuf = 1708 Parser->SrcMgr.FindBufferContainingLoc(Parser->CppHashLoc); 1709 1710 // Like SourceMgr::printMessage() we need to print the include stack if any 1711 // before printing the message. 1712 unsigned DiagCurBuffer = DiagSrcMgr.FindBufferContainingLoc(DiagLoc); 1713 if (!Parser->SavedDiagHandler && DiagCurBuffer && 1714 DiagCurBuffer != DiagSrcMgr.getMainFileID()) { 1715 SMLoc ParentIncludeLoc = DiagSrcMgr.getParentIncludeLoc(DiagCurBuffer); 1716 DiagSrcMgr.PrintIncludeStack(ParentIncludeLoc, OS); 1717 } 1718 1719 // If we have not parsed a cpp hash line filename comment or the source 1720 // manager changed or buffer changed (like in a nested include) then just 1721 // print the normal diagnostic using its Filename and LineNo. 1722 if (!Parser->CppHashLineNumber || &DiagSrcMgr != &Parser->SrcMgr || 1723 DiagBuf != CppHashBuf) { 1724 if (Parser->SavedDiagHandler) 1725 Parser->SavedDiagHandler(Diag, Parser->SavedDiagContext); 1726 else 1727 Diag.print(nullptr, OS); 1728 return; 1729 } 1730 1731 // Use the CppHashFilename and calculate a line number based on the 1732 // CppHashLoc and CppHashLineNumber relative to this Diag's SMLoc for 1733 // the diagnostic. 1734 const std::string &Filename = Parser->CppHashFilename; 1735 1736 int DiagLocLineNo = DiagSrcMgr.FindLineNumber(DiagLoc, DiagBuf); 1737 int CppHashLocLineNo = 1738 Parser->SrcMgr.FindLineNumber(Parser->CppHashLoc, CppHashBuf); 1739 int LineNo = 1740 Parser->CppHashLineNumber - 1 + (DiagLocLineNo - CppHashLocLineNo); 1741 1742 SMDiagnostic NewDiag(*Diag.getSourceMgr(), Diag.getLoc(), Filename, LineNo, 1743 Diag.getColumnNo(), Diag.getKind(), Diag.getMessage(), 1744 Diag.getLineContents(), Diag.getRanges()); 1745 1746 if (Parser->SavedDiagHandler) 1747 Parser->SavedDiagHandler(NewDiag, Parser->SavedDiagContext); 1748 else 1749 NewDiag.print(nullptr, OS); 1750 } 1751 1752 // FIXME: This is mostly duplicated from the function in AsmLexer.cpp. The 1753 // difference being that that function accepts '@' as part of identifiers and 1754 // we can't do that. AsmLexer.cpp should probably be changed to handle 1755 // '@' as a special case when needed. 1756 static bool isIdentifierChar(char c) { 1757 return isalnum(static_cast<unsigned char>(c)) || c == '_' || c == '$' || 1758 c == '.'; 1759 } 1760 1761 bool AsmParser::expandMacro(raw_svector_ostream &OS, StringRef Body, 1762 ArrayRef<MCAsmMacroParameter> Parameters, 1763 ArrayRef<MCAsmMacroArgument> A, const SMLoc &L) { 1764 unsigned NParameters = Parameters.size(); 1765 bool HasVararg = NParameters ? Parameters.back().Vararg : false; 1766 if ((!IsDarwin || NParameters != 0) && NParameters != A.size()) 1767 return Error(L, "Wrong number of arguments"); 1768 1769 // A macro without parameters is handled differently on Darwin: 1770 // gas accepts no arguments and does no substitutions 1771 while (!Body.empty()) { 1772 // Scan for the next substitution. 1773 std::size_t End = Body.size(), Pos = 0; 1774 for (; Pos != End; ++Pos) { 1775 // Check for a substitution or escape. 1776 if (IsDarwin && !NParameters) { 1777 // This macro has no parameters, look for $0, $1, etc. 1778 if (Body[Pos] != '$' || Pos + 1 == End) 1779 continue; 1780 1781 char Next = Body[Pos + 1]; 1782 if (Next == '$' || Next == 'n' || 1783 isdigit(static_cast<unsigned char>(Next))) 1784 break; 1785 } else { 1786 // This macro has parameters, look for \foo, \bar, etc. 1787 if (Body[Pos] == '\\' && Pos + 1 != End) 1788 break; 1789 } 1790 } 1791 1792 // Add the prefix. 1793 OS << Body.slice(0, Pos); 1794 1795 // Check if we reached the end. 1796 if (Pos == End) 1797 break; 1798 1799 if (IsDarwin && !NParameters) { 1800 switch (Body[Pos + 1]) { 1801 // $$ => $ 1802 case '$': 1803 OS << '$'; 1804 break; 1805 1806 // $n => number of arguments 1807 case 'n': 1808 OS << A.size(); 1809 break; 1810 1811 // $[0-9] => argument 1812 default: { 1813 // Missing arguments are ignored. 1814 unsigned Index = Body[Pos + 1] - '0'; 1815 if (Index >= A.size()) 1816 break; 1817 1818 // Otherwise substitute with the token values, with spaces eliminated. 1819 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(), 1820 ie = A[Index].end(); 1821 it != ie; ++it) 1822 OS << it->getString(); 1823 break; 1824 } 1825 } 1826 Pos += 2; 1827 } else { 1828 unsigned I = Pos + 1; 1829 while (isIdentifierChar(Body[I]) && I + 1 != End) 1830 ++I; 1831 1832 const char *Begin = Body.data() + Pos + 1; 1833 StringRef Argument(Begin, I - (Pos + 1)); 1834 unsigned Index = 0; 1835 for (; Index < NParameters; ++Index) 1836 if (Parameters[Index].Name == Argument) 1837 break; 1838 1839 if (Index == NParameters) { 1840 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')') 1841 Pos += 3; 1842 else { 1843 OS << '\\' << Argument; 1844 Pos = I; 1845 } 1846 } else { 1847 bool VarargParameter = HasVararg && Index == (NParameters - 1); 1848 for (MCAsmMacroArgument::const_iterator it = A[Index].begin(), 1849 ie = A[Index].end(); 1850 it != ie; ++it) 1851 // We expect no quotes around the string's contents when 1852 // parsing for varargs. 1853 if (it->getKind() != AsmToken::String || VarargParameter) 1854 OS << it->getString(); 1855 else 1856 OS << it->getStringContents(); 1857 1858 Pos += 1 + Argument.size(); 1859 } 1860 } 1861 // Update the scan point. 1862 Body = Body.substr(Pos); 1863 } 1864 1865 return false; 1866 } 1867 1868 MacroInstantiation::MacroInstantiation(SMLoc IL, int EB, SMLoc EL, 1869 size_t CondStackDepth) 1870 : InstantiationLoc(IL), ExitBuffer(EB), ExitLoc(EL), 1871 CondStackDepth(CondStackDepth) {} 1872 1873 static bool isOperator(AsmToken::TokenKind kind) { 1874 switch (kind) { 1875 default: 1876 return false; 1877 case AsmToken::Plus: 1878 case AsmToken::Minus: 1879 case AsmToken::Tilde: 1880 case AsmToken::Slash: 1881 case AsmToken::Star: 1882 case AsmToken::Dot: 1883 case AsmToken::Equal: 1884 case AsmToken::EqualEqual: 1885 case AsmToken::Pipe: 1886 case AsmToken::PipePipe: 1887 case AsmToken::Caret: 1888 case AsmToken::Amp: 1889 case AsmToken::AmpAmp: 1890 case AsmToken::Exclaim: 1891 case AsmToken::ExclaimEqual: 1892 case AsmToken::Percent: 1893 case AsmToken::Less: 1894 case AsmToken::LessEqual: 1895 case AsmToken::LessLess: 1896 case AsmToken::LessGreater: 1897 case AsmToken::Greater: 1898 case AsmToken::GreaterEqual: 1899 case AsmToken::GreaterGreater: 1900 return true; 1901 } 1902 } 1903 1904 namespace { 1905 class AsmLexerSkipSpaceRAII { 1906 public: 1907 AsmLexerSkipSpaceRAII(AsmLexer &Lexer, bool SkipSpace) : Lexer(Lexer) { 1908 Lexer.setSkipSpace(SkipSpace); 1909 } 1910 1911 ~AsmLexerSkipSpaceRAII() { 1912 Lexer.setSkipSpace(true); 1913 } 1914 1915 private: 1916 AsmLexer &Lexer; 1917 }; 1918 } 1919 1920 bool AsmParser::parseMacroArgument(MCAsmMacroArgument &MA, bool Vararg) { 1921 1922 if (Vararg) { 1923 if (Lexer.isNot(AsmToken::EndOfStatement)) { 1924 StringRef Str = parseStringToEndOfStatement(); 1925 MA.push_back(AsmToken(AsmToken::String, Str)); 1926 } 1927 return false; 1928 } 1929 1930 unsigned ParenLevel = 0; 1931 unsigned AddTokens = 0; 1932 1933 // Darwin doesn't use spaces to delmit arguments. 1934 AsmLexerSkipSpaceRAII ScopedSkipSpace(Lexer, IsDarwin); 1935 1936 for (;;) { 1937 if (Lexer.is(AsmToken::Eof) || Lexer.is(AsmToken::Equal)) 1938 return TokError("unexpected token in macro instantiation"); 1939 1940 if (ParenLevel == 0 && Lexer.is(AsmToken::Comma)) 1941 break; 1942 1943 if (Lexer.is(AsmToken::Space)) { 1944 Lex(); // Eat spaces 1945 1946 // Spaces can delimit parameters, but could also be part an expression. 1947 // If the token after a space is an operator, add the token and the next 1948 // one into this argument 1949 if (!IsDarwin) { 1950 if (isOperator(Lexer.getKind())) { 1951 // Check to see whether the token is used as an operator, 1952 // or part of an identifier 1953 const char *NextChar = getTok().getEndLoc().getPointer(); 1954 if (*NextChar == ' ') 1955 AddTokens = 2; 1956 } 1957 1958 if (!AddTokens && ParenLevel == 0) { 1959 break; 1960 } 1961 } 1962 } 1963 1964 // handleMacroEntry relies on not advancing the lexer here 1965 // to be able to fill in the remaining default parameter values 1966 if (Lexer.is(AsmToken::EndOfStatement)) 1967 break; 1968 1969 // Adjust the current parentheses level. 1970 if (Lexer.is(AsmToken::LParen)) 1971 ++ParenLevel; 1972 else if (Lexer.is(AsmToken::RParen) && ParenLevel) 1973 --ParenLevel; 1974 1975 // Append the token to the current argument list. 1976 MA.push_back(getTok()); 1977 if (AddTokens) 1978 AddTokens--; 1979 Lex(); 1980 } 1981 1982 if (ParenLevel != 0) 1983 return TokError("unbalanced parentheses in macro argument"); 1984 return false; 1985 } 1986 1987 // Parse the macro instantiation arguments. 1988 bool AsmParser::parseMacroArguments(const MCAsmMacro *M, 1989 MCAsmMacroArguments &A) { 1990 const unsigned NParameters = M ? M->Parameters.size() : 0; 1991 bool NamedParametersFound = false; 1992 SmallVector<SMLoc, 4> FALocs; 1993 1994 A.resize(NParameters); 1995 FALocs.resize(NParameters); 1996 1997 // Parse two kinds of macro invocations: 1998 // - macros defined without any parameters accept an arbitrary number of them 1999 // - macros defined with parameters accept at most that many of them 2000 bool HasVararg = NParameters ? M->Parameters.back().Vararg : false; 2001 for (unsigned Parameter = 0; !NParameters || Parameter < NParameters; 2002 ++Parameter) { 2003 SMLoc IDLoc = Lexer.getLoc(); 2004 MCAsmMacroParameter FA; 2005 2006 if (Lexer.is(AsmToken::Identifier) && Lexer.peekTok().is(AsmToken::Equal)) { 2007 if (parseIdentifier(FA.Name)) { 2008 Error(IDLoc, "invalid argument identifier for formal argument"); 2009 eatToEndOfStatement(); 2010 return true; 2011 } 2012 2013 if (!Lexer.is(AsmToken::Equal)) { 2014 TokError("expected '=' after formal parameter identifier"); 2015 eatToEndOfStatement(); 2016 return true; 2017 } 2018 Lex(); 2019 2020 NamedParametersFound = true; 2021 } 2022 2023 if (NamedParametersFound && FA.Name.empty()) { 2024 Error(IDLoc, "cannot mix positional and keyword arguments"); 2025 eatToEndOfStatement(); 2026 return true; 2027 } 2028 2029 bool Vararg = HasVararg && Parameter == (NParameters - 1); 2030 if (parseMacroArgument(FA.Value, Vararg)) 2031 return true; 2032 2033 unsigned PI = Parameter; 2034 if (!FA.Name.empty()) { 2035 unsigned FAI = 0; 2036 for (FAI = 0; FAI < NParameters; ++FAI) 2037 if (M->Parameters[FAI].Name == FA.Name) 2038 break; 2039 2040 if (FAI >= NParameters) { 2041 assert(M && "expected macro to be defined"); 2042 Error(IDLoc, 2043 "parameter named '" + FA.Name + "' does not exist for macro '" + 2044 M->Name + "'"); 2045 return true; 2046 } 2047 PI = FAI; 2048 } 2049 2050 if (!FA.Value.empty()) { 2051 if (A.size() <= PI) 2052 A.resize(PI + 1); 2053 A[PI] = FA.Value; 2054 2055 if (FALocs.size() <= PI) 2056 FALocs.resize(PI + 1); 2057 2058 FALocs[PI] = Lexer.getLoc(); 2059 } 2060 2061 // At the end of the statement, fill in remaining arguments that have 2062 // default values. If there aren't any, then the next argument is 2063 // required but missing 2064 if (Lexer.is(AsmToken::EndOfStatement)) { 2065 bool Failure = false; 2066 for (unsigned FAI = 0; FAI < NParameters; ++FAI) { 2067 if (A[FAI].empty()) { 2068 if (M->Parameters[FAI].Required) { 2069 Error(FALocs[FAI].isValid() ? FALocs[FAI] : Lexer.getLoc(), 2070 "missing value for required parameter " 2071 "'" + M->Parameters[FAI].Name + "' in macro '" + M->Name + "'"); 2072 Failure = true; 2073 } 2074 2075 if (!M->Parameters[FAI].Value.empty()) 2076 A[FAI] = M->Parameters[FAI].Value; 2077 } 2078 } 2079 return Failure; 2080 } 2081 2082 if (Lexer.is(AsmToken::Comma)) 2083 Lex(); 2084 } 2085 2086 return TokError("too many positional arguments"); 2087 } 2088 2089 const MCAsmMacro *AsmParser::lookupMacro(StringRef Name) { 2090 StringMap<MCAsmMacro>::iterator I = MacroMap.find(Name); 2091 return (I == MacroMap.end()) ? nullptr : &I->getValue(); 2092 } 2093 2094 void AsmParser::defineMacro(StringRef Name, MCAsmMacro Macro) { 2095 MacroMap.insert(std::make_pair(Name, std::move(Macro))); 2096 } 2097 2098 void AsmParser::undefineMacro(StringRef Name) { MacroMap.erase(Name); } 2099 2100 bool AsmParser::handleMacroEntry(const MCAsmMacro *M, SMLoc NameLoc) { 2101 // Arbitrarily limit macro nesting depth, to match 'as'. We can eliminate 2102 // this, although we should protect against infinite loops. 2103 if (ActiveMacros.size() == 20) 2104 return TokError("macros cannot be nested more than 20 levels deep"); 2105 2106 MCAsmMacroArguments A; 2107 if (parseMacroArguments(M, A)) 2108 return true; 2109 2110 // Macro instantiation is lexical, unfortunately. We construct a new buffer 2111 // to hold the macro body with substitutions. 2112 SmallString<256> Buf; 2113 StringRef Body = M->Body; 2114 raw_svector_ostream OS(Buf); 2115 2116 if (expandMacro(OS, Body, M->Parameters, A, getTok().getLoc())) 2117 return true; 2118 2119 // We include the .endmacro in the buffer as our cue to exit the macro 2120 // instantiation. 2121 OS << ".endmacro\n"; 2122 2123 std::unique_ptr<MemoryBuffer> Instantiation = 2124 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 2125 2126 // Create the macro instantiation object and add to the current macro 2127 // instantiation stack. 2128 MacroInstantiation *MI = new MacroInstantiation( 2129 NameLoc, CurBuffer, getTok().getLoc(), TheCondStack.size()); 2130 ActiveMacros.push_back(MI); 2131 2132 // Jump to the macro instantiation and prime the lexer. 2133 CurBuffer = SrcMgr.AddNewSourceBuffer(std::move(Instantiation), SMLoc()); 2134 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 2135 Lex(); 2136 2137 return false; 2138 } 2139 2140 void AsmParser::handleMacroExit() { 2141 // Jump to the EndOfStatement we should return to, and consume it. 2142 jumpToLoc(ActiveMacros.back()->ExitLoc, ActiveMacros.back()->ExitBuffer); 2143 Lex(); 2144 2145 // Pop the instantiation entry. 2146 delete ActiveMacros.back(); 2147 ActiveMacros.pop_back(); 2148 } 2149 2150 static bool isUsedIn(const MCSymbol *Sym, const MCExpr *Value) { 2151 switch (Value->getKind()) { 2152 case MCExpr::Binary: { 2153 const MCBinaryExpr *BE = static_cast<const MCBinaryExpr *>(Value); 2154 return isUsedIn(Sym, BE->getLHS()) || isUsedIn(Sym, BE->getRHS()); 2155 } 2156 case MCExpr::Target: 2157 case MCExpr::Constant: 2158 return false; 2159 case MCExpr::SymbolRef: { 2160 const MCSymbol &S = 2161 static_cast<const MCSymbolRefExpr *>(Value)->getSymbol(); 2162 if (S.isVariable()) 2163 return isUsedIn(Sym, S.getVariableValue()); 2164 return &S == Sym; 2165 } 2166 case MCExpr::Unary: 2167 return isUsedIn(Sym, static_cast<const MCUnaryExpr *>(Value)->getSubExpr()); 2168 } 2169 2170 llvm_unreachable("Unknown expr kind!"); 2171 } 2172 2173 bool AsmParser::parseAssignment(StringRef Name, bool allow_redef, 2174 bool NoDeadStrip) { 2175 // FIXME: Use better location, we should use proper tokens. 2176 SMLoc EqualLoc = Lexer.getLoc(); 2177 2178 const MCExpr *Value; 2179 if (parseExpression(Value)) 2180 return true; 2181 2182 // Note: we don't count b as used in "a = b". This is to allow 2183 // a = b 2184 // b = c 2185 2186 if (Lexer.isNot(AsmToken::EndOfStatement)) 2187 return TokError("unexpected token in assignment"); 2188 2189 // Eat the end of statement marker. 2190 Lex(); 2191 2192 // Validate that the LHS is allowed to be a variable (either it has not been 2193 // used as a symbol, or it is an absolute symbol). 2194 MCSymbol *Sym = getContext().LookupSymbol(Name); 2195 if (Sym) { 2196 // Diagnose assignment to a label. 2197 // 2198 // FIXME: Diagnostics. Note the location of the definition as a label. 2199 // FIXME: Diagnose assignment to protected identifier (e.g., register name). 2200 if (isUsedIn(Sym, Value)) 2201 return Error(EqualLoc, "Recursive use of '" + Name + "'"); 2202 else if (Sym->isUndefined() && !Sym->isUsed() && !Sym->isVariable()) 2203 ; // Allow redefinitions of undefined symbols only used in directives. 2204 else if (Sym->isVariable() && !Sym->isUsed() && allow_redef) 2205 ; // Allow redefinitions of variables that haven't yet been used. 2206 else if (!Sym->isUndefined() && (!Sym->isVariable() || !allow_redef)) 2207 return Error(EqualLoc, "redefinition of '" + Name + "'"); 2208 else if (!Sym->isVariable()) 2209 return Error(EqualLoc, "invalid assignment to '" + Name + "'"); 2210 else if (!isa<MCConstantExpr>(Sym->getVariableValue())) 2211 return Error(EqualLoc, "invalid reassignment of non-absolute variable '" + 2212 Name + "'"); 2213 2214 // Don't count these checks as uses. 2215 Sym->setUsed(false); 2216 } else if (Name == ".") { 2217 if (Out.EmitValueToOffset(Value, 0)) { 2218 Error(EqualLoc, "expected absolute expression"); 2219 eatToEndOfStatement(); 2220 } 2221 return false; 2222 } else 2223 Sym = getContext().GetOrCreateSymbol(Name); 2224 2225 Sym->setRedefinable(allow_redef); 2226 2227 // Do the assignment. 2228 Out.EmitAssignment(Sym, Value); 2229 if (NoDeadStrip) 2230 Out.EmitSymbolAttribute(Sym, MCSA_NoDeadStrip); 2231 2232 return false; 2233 } 2234 2235 /// parseIdentifier: 2236 /// ::= identifier 2237 /// ::= string 2238 bool AsmParser::parseIdentifier(StringRef &Res) { 2239 // The assembler has relaxed rules for accepting identifiers, in particular we 2240 // allow things like '.globl $foo' and '.def @feat.00', which would normally be 2241 // separate tokens. At this level, we have already lexed so we cannot (currently) 2242 // handle this as a context dependent token, instead we detect adjacent tokens 2243 // and return the combined identifier. 2244 if (Lexer.is(AsmToken::Dollar) || Lexer.is(AsmToken::At)) { 2245 SMLoc PrefixLoc = getLexer().getLoc(); 2246 2247 // Consume the prefix character, and check for a following identifier. 2248 Lex(); 2249 if (Lexer.isNot(AsmToken::Identifier)) 2250 return true; 2251 2252 // We have a '$' or '@' followed by an identifier, make sure they are adjacent. 2253 if (PrefixLoc.getPointer() + 1 != getTok().getLoc().getPointer()) 2254 return true; 2255 2256 // Construct the joined identifier and consume the token. 2257 Res = 2258 StringRef(PrefixLoc.getPointer(), getTok().getIdentifier().size() + 1); 2259 Lex(); 2260 return false; 2261 } 2262 2263 if (Lexer.isNot(AsmToken::Identifier) && Lexer.isNot(AsmToken::String)) 2264 return true; 2265 2266 Res = getTok().getIdentifier(); 2267 2268 Lex(); // Consume the identifier token. 2269 2270 return false; 2271 } 2272 2273 /// parseDirectiveSet: 2274 /// ::= .equ identifier ',' expression 2275 /// ::= .equiv identifier ',' expression 2276 /// ::= .set identifier ',' expression 2277 bool AsmParser::parseDirectiveSet(StringRef IDVal, bool allow_redef) { 2278 StringRef Name; 2279 2280 if (parseIdentifier(Name)) 2281 return TokError("expected identifier after '" + Twine(IDVal) + "'"); 2282 2283 if (getLexer().isNot(AsmToken::Comma)) 2284 return TokError("unexpected token in '" + Twine(IDVal) + "'"); 2285 Lex(); 2286 2287 return parseAssignment(Name, allow_redef, true); 2288 } 2289 2290 bool AsmParser::parseEscapedString(std::string &Data) { 2291 assert(getLexer().is(AsmToken::String) && "Unexpected current token!"); 2292 2293 Data = ""; 2294 StringRef Str = getTok().getStringContents(); 2295 for (unsigned i = 0, e = Str.size(); i != e; ++i) { 2296 if (Str[i] != '\\') { 2297 Data += Str[i]; 2298 continue; 2299 } 2300 2301 // Recognize escaped characters. Note that this escape semantics currently 2302 // loosely follows Darwin 'as'. Notably, it doesn't support hex escapes. 2303 ++i; 2304 if (i == e) 2305 return TokError("unexpected backslash at end of string"); 2306 2307 // Recognize octal sequences. 2308 if ((unsigned)(Str[i] - '0') <= 7) { 2309 // Consume up to three octal characters. 2310 unsigned Value = Str[i] - '0'; 2311 2312 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) { 2313 ++i; 2314 Value = Value * 8 + (Str[i] - '0'); 2315 2316 if (i + 1 != e && ((unsigned)(Str[i + 1] - '0')) <= 7) { 2317 ++i; 2318 Value = Value * 8 + (Str[i] - '0'); 2319 } 2320 } 2321 2322 if (Value > 255) 2323 return TokError("invalid octal escape sequence (out of range)"); 2324 2325 Data += (unsigned char)Value; 2326 continue; 2327 } 2328 2329 // Otherwise recognize individual escapes. 2330 switch (Str[i]) { 2331 default: 2332 // Just reject invalid escape sequences for now. 2333 return TokError("invalid escape sequence (unrecognized character)"); 2334 2335 case 'b': Data += '\b'; break; 2336 case 'f': Data += '\f'; break; 2337 case 'n': Data += '\n'; break; 2338 case 'r': Data += '\r'; break; 2339 case 't': Data += '\t'; break; 2340 case '"': Data += '"'; break; 2341 case '\\': Data += '\\'; break; 2342 } 2343 } 2344 2345 return false; 2346 } 2347 2348 /// parseDirectiveAscii: 2349 /// ::= ( .ascii | .asciz | .string ) [ "string" ( , "string" )* ] 2350 bool AsmParser::parseDirectiveAscii(StringRef IDVal, bool ZeroTerminated) { 2351 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2352 checkForValidSection(); 2353 2354 for (;;) { 2355 if (getLexer().isNot(AsmToken::String)) 2356 return TokError("expected string in '" + Twine(IDVal) + "' directive"); 2357 2358 std::string Data; 2359 if (parseEscapedString(Data)) 2360 return true; 2361 2362 getStreamer().EmitBytes(Data); 2363 if (ZeroTerminated) 2364 getStreamer().EmitBytes(StringRef("\0", 1)); 2365 2366 Lex(); 2367 2368 if (getLexer().is(AsmToken::EndOfStatement)) 2369 break; 2370 2371 if (getLexer().isNot(AsmToken::Comma)) 2372 return TokError("unexpected token in '" + Twine(IDVal) + "' directive"); 2373 Lex(); 2374 } 2375 } 2376 2377 Lex(); 2378 return false; 2379 } 2380 2381 /// parseDirectiveValue 2382 /// ::= (.byte | .short | ... ) [ expression (, expression)* ] 2383 bool AsmParser::parseDirectiveValue(unsigned Size) { 2384 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2385 checkForValidSection(); 2386 2387 for (;;) { 2388 const MCExpr *Value; 2389 SMLoc ExprLoc = getLexer().getLoc(); 2390 if (parseExpression(Value)) 2391 return true; 2392 2393 // Special case constant expressions to match code generator. 2394 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2395 assert(Size <= 8 && "Invalid size"); 2396 uint64_t IntValue = MCE->getValue(); 2397 if (!isUIntN(8 * Size, IntValue) && !isIntN(8 * Size, IntValue)) 2398 return Error(ExprLoc, "literal value out of range for directive"); 2399 getStreamer().EmitIntValue(IntValue, Size); 2400 } else 2401 getStreamer().EmitValue(Value, Size, ExprLoc); 2402 2403 if (getLexer().is(AsmToken::EndOfStatement)) 2404 break; 2405 2406 // FIXME: Improve diagnostic. 2407 if (getLexer().isNot(AsmToken::Comma)) 2408 return TokError("unexpected token in directive"); 2409 Lex(); 2410 } 2411 } 2412 2413 Lex(); 2414 return false; 2415 } 2416 2417 /// ParseDirectiveOctaValue 2418 /// ::= .octa [ hexconstant (, hexconstant)* ] 2419 bool AsmParser::parseDirectiveOctaValue() { 2420 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2421 checkForValidSection(); 2422 2423 for (;;) { 2424 if (Lexer.getKind() == AsmToken::Error) 2425 return true; 2426 if (Lexer.getKind() != AsmToken::Integer && 2427 Lexer.getKind() != AsmToken::BigNum) 2428 return TokError("unknown token in expression"); 2429 2430 SMLoc ExprLoc = getLexer().getLoc(); 2431 APInt IntValue = getTok().getAPIntVal(); 2432 Lex(); 2433 2434 uint64_t hi, lo; 2435 if (IntValue.isIntN(64)) { 2436 hi = 0; 2437 lo = IntValue.getZExtValue(); 2438 } else if (IntValue.isIntN(128)) { 2439 // It might actually have more than 128 bits, but the top ones are zero. 2440 hi = IntValue.getHiBits(IntValue.getBitWidth() - 64).getZExtValue(); 2441 lo = IntValue.getLoBits(64).getZExtValue(); 2442 } else 2443 return Error(ExprLoc, "literal value out of range for directive"); 2444 2445 if (MAI.isLittleEndian()) { 2446 getStreamer().EmitIntValue(lo, 8); 2447 getStreamer().EmitIntValue(hi, 8); 2448 } else { 2449 getStreamer().EmitIntValue(hi, 8); 2450 getStreamer().EmitIntValue(lo, 8); 2451 } 2452 2453 if (getLexer().is(AsmToken::EndOfStatement)) 2454 break; 2455 2456 // FIXME: Improve diagnostic. 2457 if (getLexer().isNot(AsmToken::Comma)) 2458 return TokError("unexpected token in directive"); 2459 Lex(); 2460 } 2461 } 2462 2463 Lex(); 2464 return false; 2465 } 2466 2467 /// parseDirectiveRealValue 2468 /// ::= (.single | .double) [ expression (, expression)* ] 2469 bool AsmParser::parseDirectiveRealValue(const fltSemantics &Semantics) { 2470 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2471 checkForValidSection(); 2472 2473 for (;;) { 2474 // We don't truly support arithmetic on floating point expressions, so we 2475 // have to manually parse unary prefixes. 2476 bool IsNeg = false; 2477 if (getLexer().is(AsmToken::Minus)) { 2478 Lex(); 2479 IsNeg = true; 2480 } else if (getLexer().is(AsmToken::Plus)) 2481 Lex(); 2482 2483 if (getLexer().isNot(AsmToken::Integer) && 2484 getLexer().isNot(AsmToken::Real) && 2485 getLexer().isNot(AsmToken::Identifier)) 2486 return TokError("unexpected token in directive"); 2487 2488 // Convert to an APFloat. 2489 APFloat Value(Semantics); 2490 StringRef IDVal = getTok().getString(); 2491 if (getLexer().is(AsmToken::Identifier)) { 2492 if (!IDVal.compare_lower("infinity") || !IDVal.compare_lower("inf")) 2493 Value = APFloat::getInf(Semantics); 2494 else if (!IDVal.compare_lower("nan")) 2495 Value = APFloat::getNaN(Semantics, false, ~0); 2496 else 2497 return TokError("invalid floating point literal"); 2498 } else if (Value.convertFromString(IDVal, APFloat::rmNearestTiesToEven) == 2499 APFloat::opInvalidOp) 2500 return TokError("invalid floating point literal"); 2501 if (IsNeg) 2502 Value.changeSign(); 2503 2504 // Consume the numeric token. 2505 Lex(); 2506 2507 // Emit the value as an integer. 2508 APInt AsInt = Value.bitcastToAPInt(); 2509 getStreamer().EmitIntValue(AsInt.getLimitedValue(), 2510 AsInt.getBitWidth() / 8); 2511 2512 if (getLexer().is(AsmToken::EndOfStatement)) 2513 break; 2514 2515 if (getLexer().isNot(AsmToken::Comma)) 2516 return TokError("unexpected token in directive"); 2517 Lex(); 2518 } 2519 } 2520 2521 Lex(); 2522 return false; 2523 } 2524 2525 /// parseDirectiveZero 2526 /// ::= .zero expression 2527 bool AsmParser::parseDirectiveZero() { 2528 checkForValidSection(); 2529 2530 int64_t NumBytes; 2531 if (parseAbsoluteExpression(NumBytes)) 2532 return true; 2533 2534 int64_t Val = 0; 2535 if (getLexer().is(AsmToken::Comma)) { 2536 Lex(); 2537 if (parseAbsoluteExpression(Val)) 2538 return true; 2539 } 2540 2541 if (getLexer().isNot(AsmToken::EndOfStatement)) 2542 return TokError("unexpected token in '.zero' directive"); 2543 2544 Lex(); 2545 2546 getStreamer().EmitFill(NumBytes, Val); 2547 2548 return false; 2549 } 2550 2551 /// parseDirectiveFill 2552 /// ::= .fill expression [ , expression [ , expression ] ] 2553 bool AsmParser::parseDirectiveFill() { 2554 checkForValidSection(); 2555 2556 SMLoc RepeatLoc = getLexer().getLoc(); 2557 int64_t NumValues; 2558 if (parseAbsoluteExpression(NumValues)) 2559 return true; 2560 2561 if (NumValues < 0) { 2562 Warning(RepeatLoc, 2563 "'.fill' directive with negative repeat count has no effect"); 2564 NumValues = 0; 2565 } 2566 2567 int64_t FillSize = 1; 2568 int64_t FillExpr = 0; 2569 2570 SMLoc SizeLoc, ExprLoc; 2571 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2572 if (getLexer().isNot(AsmToken::Comma)) 2573 return TokError("unexpected token in '.fill' directive"); 2574 Lex(); 2575 2576 SizeLoc = getLexer().getLoc(); 2577 if (parseAbsoluteExpression(FillSize)) 2578 return true; 2579 2580 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2581 if (getLexer().isNot(AsmToken::Comma)) 2582 return TokError("unexpected token in '.fill' directive"); 2583 Lex(); 2584 2585 ExprLoc = getLexer().getLoc(); 2586 if (parseAbsoluteExpression(FillExpr)) 2587 return true; 2588 2589 if (getLexer().isNot(AsmToken::EndOfStatement)) 2590 return TokError("unexpected token in '.fill' directive"); 2591 2592 Lex(); 2593 } 2594 } 2595 2596 if (FillSize < 0) { 2597 Warning(SizeLoc, "'.fill' directive with negative size has no effect"); 2598 NumValues = 0; 2599 } 2600 if (FillSize > 8) { 2601 Warning(SizeLoc, "'.fill' directive with size greater than 8 has been truncated to 8"); 2602 FillSize = 8; 2603 } 2604 2605 if (!isUInt<32>(FillExpr) && FillSize > 4) 2606 Warning(ExprLoc, "'.fill' directive pattern has been truncated to 32-bits"); 2607 2608 if (NumValues > 0) { 2609 int64_t NonZeroFillSize = FillSize > 4 ? 4 : FillSize; 2610 FillExpr &= ~0ULL >> (64 - NonZeroFillSize * 8); 2611 for (uint64_t i = 0, e = NumValues; i != e; ++i) { 2612 getStreamer().EmitIntValue(FillExpr, NonZeroFillSize); 2613 if (NonZeroFillSize < FillSize) 2614 getStreamer().EmitIntValue(0, FillSize - NonZeroFillSize); 2615 } 2616 } 2617 2618 return false; 2619 } 2620 2621 /// parseDirectiveOrg 2622 /// ::= .org expression [ , expression ] 2623 bool AsmParser::parseDirectiveOrg() { 2624 checkForValidSection(); 2625 2626 const MCExpr *Offset; 2627 SMLoc Loc = getTok().getLoc(); 2628 if (parseExpression(Offset)) 2629 return true; 2630 2631 // Parse optional fill expression. 2632 int64_t FillExpr = 0; 2633 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2634 if (getLexer().isNot(AsmToken::Comma)) 2635 return TokError("unexpected token in '.org' directive"); 2636 Lex(); 2637 2638 if (parseAbsoluteExpression(FillExpr)) 2639 return true; 2640 2641 if (getLexer().isNot(AsmToken::EndOfStatement)) 2642 return TokError("unexpected token in '.org' directive"); 2643 } 2644 2645 Lex(); 2646 2647 // Only limited forms of relocatable expressions are accepted here, it 2648 // has to be relative to the current section. The streamer will return 2649 // 'true' if the expression wasn't evaluatable. 2650 if (getStreamer().EmitValueToOffset(Offset, FillExpr)) 2651 return Error(Loc, "expected assembly-time absolute expression"); 2652 2653 return false; 2654 } 2655 2656 /// parseDirectiveAlign 2657 /// ::= {.align, ...} expression [ , expression [ , expression ]] 2658 bool AsmParser::parseDirectiveAlign(bool IsPow2, unsigned ValueSize) { 2659 checkForValidSection(); 2660 2661 SMLoc AlignmentLoc = getLexer().getLoc(); 2662 int64_t Alignment; 2663 if (parseAbsoluteExpression(Alignment)) 2664 return true; 2665 2666 SMLoc MaxBytesLoc; 2667 bool HasFillExpr = false; 2668 int64_t FillExpr = 0; 2669 int64_t MaxBytesToFill = 0; 2670 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2671 if (getLexer().isNot(AsmToken::Comma)) 2672 return TokError("unexpected token in directive"); 2673 Lex(); 2674 2675 // The fill expression can be omitted while specifying a maximum number of 2676 // alignment bytes, e.g: 2677 // .align 3,,4 2678 if (getLexer().isNot(AsmToken::Comma)) { 2679 HasFillExpr = true; 2680 if (parseAbsoluteExpression(FillExpr)) 2681 return true; 2682 } 2683 2684 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2685 if (getLexer().isNot(AsmToken::Comma)) 2686 return TokError("unexpected token in directive"); 2687 Lex(); 2688 2689 MaxBytesLoc = getLexer().getLoc(); 2690 if (parseAbsoluteExpression(MaxBytesToFill)) 2691 return true; 2692 2693 if (getLexer().isNot(AsmToken::EndOfStatement)) 2694 return TokError("unexpected token in directive"); 2695 } 2696 } 2697 2698 Lex(); 2699 2700 if (!HasFillExpr) 2701 FillExpr = 0; 2702 2703 // Compute alignment in bytes. 2704 if (IsPow2) { 2705 // FIXME: Diagnose overflow. 2706 if (Alignment >= 32) { 2707 Error(AlignmentLoc, "invalid alignment value"); 2708 Alignment = 31; 2709 } 2710 2711 Alignment = 1ULL << Alignment; 2712 } else { 2713 // Reject alignments that aren't a power of two, for gas compatibility. 2714 if (!isPowerOf2_64(Alignment)) 2715 Error(AlignmentLoc, "alignment must be a power of 2"); 2716 } 2717 2718 // Diagnose non-sensical max bytes to align. 2719 if (MaxBytesLoc.isValid()) { 2720 if (MaxBytesToFill < 1) { 2721 Error(MaxBytesLoc, "alignment directive can never be satisfied in this " 2722 "many bytes, ignoring maximum bytes expression"); 2723 MaxBytesToFill = 0; 2724 } 2725 2726 if (MaxBytesToFill >= Alignment) { 2727 Warning(MaxBytesLoc, "maximum bytes expression exceeds alignment and " 2728 "has no effect"); 2729 MaxBytesToFill = 0; 2730 } 2731 } 2732 2733 // Check whether we should use optimal code alignment for this .align 2734 // directive. 2735 const MCSection *Section = getStreamer().getCurrentSection().first; 2736 assert(Section && "must have section to emit alignment"); 2737 bool UseCodeAlign = Section->UseCodeAlign(); 2738 if ((!HasFillExpr || Lexer.getMAI().getTextAlignFillValue() == FillExpr) && 2739 ValueSize == 1 && UseCodeAlign) { 2740 getStreamer().EmitCodeAlignment(Alignment, MaxBytesToFill); 2741 } else { 2742 // FIXME: Target specific behavior about how the "extra" bytes are filled. 2743 getStreamer().EmitValueToAlignment(Alignment, FillExpr, ValueSize, 2744 MaxBytesToFill); 2745 } 2746 2747 return false; 2748 } 2749 2750 /// parseDirectiveFile 2751 /// ::= .file [number] filename 2752 /// ::= .file number directory filename 2753 bool AsmParser::parseDirectiveFile(SMLoc DirectiveLoc) { 2754 // FIXME: I'm not sure what this is. 2755 int64_t FileNumber = -1; 2756 SMLoc FileNumberLoc = getLexer().getLoc(); 2757 if (getLexer().is(AsmToken::Integer)) { 2758 FileNumber = getTok().getIntVal(); 2759 Lex(); 2760 2761 if (FileNumber < 1) 2762 return TokError("file number less than one"); 2763 } 2764 2765 if (getLexer().isNot(AsmToken::String)) 2766 return TokError("unexpected token in '.file' directive"); 2767 2768 // Usually the directory and filename together, otherwise just the directory. 2769 // Allow the strings to have escaped octal character sequence. 2770 std::string Path = getTok().getString(); 2771 if (parseEscapedString(Path)) 2772 return true; 2773 Lex(); 2774 2775 StringRef Directory; 2776 StringRef Filename; 2777 std::string FilenameData; 2778 if (getLexer().is(AsmToken::String)) { 2779 if (FileNumber == -1) 2780 return TokError("explicit path specified, but no file number"); 2781 if (parseEscapedString(FilenameData)) 2782 return true; 2783 Filename = FilenameData; 2784 Directory = Path; 2785 Lex(); 2786 } else { 2787 Filename = Path; 2788 } 2789 2790 if (getLexer().isNot(AsmToken::EndOfStatement)) 2791 return TokError("unexpected token in '.file' directive"); 2792 2793 if (FileNumber == -1) 2794 getStreamer().EmitFileDirective(Filename); 2795 else { 2796 if (getContext().getGenDwarfForAssembly()) 2797 Error(DirectiveLoc, 2798 "input can't have .file dwarf directives when -g is " 2799 "used to generate dwarf debug info for assembly code"); 2800 2801 if (getStreamer().EmitDwarfFileDirective(FileNumber, Directory, Filename) == 2802 0) 2803 Error(FileNumberLoc, "file number already allocated"); 2804 } 2805 2806 return false; 2807 } 2808 2809 /// parseDirectiveLine 2810 /// ::= .line [number] 2811 bool AsmParser::parseDirectiveLine() { 2812 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2813 if (getLexer().isNot(AsmToken::Integer)) 2814 return TokError("unexpected token in '.line' directive"); 2815 2816 int64_t LineNumber = getTok().getIntVal(); 2817 (void)LineNumber; 2818 Lex(); 2819 2820 // FIXME: Do something with the .line. 2821 } 2822 2823 if (getLexer().isNot(AsmToken::EndOfStatement)) 2824 return TokError("unexpected token in '.line' directive"); 2825 2826 return false; 2827 } 2828 2829 /// parseDirectiveLoc 2830 /// ::= .loc FileNumber [LineNumber] [ColumnPos] [basic_block] [prologue_end] 2831 /// [epilogue_begin] [is_stmt VALUE] [isa VALUE] 2832 /// The first number is a file number, must have been previously assigned with 2833 /// a .file directive, the second number is the line number and optionally the 2834 /// third number is a column position (zero if not specified). The remaining 2835 /// optional items are .loc sub-directives. 2836 bool AsmParser::parseDirectiveLoc() { 2837 if (getLexer().isNot(AsmToken::Integer)) 2838 return TokError("unexpected token in '.loc' directive"); 2839 int64_t FileNumber = getTok().getIntVal(); 2840 if (FileNumber < 1) 2841 return TokError("file number less than one in '.loc' directive"); 2842 if (!getContext().isValidDwarfFileNumber(FileNumber)) 2843 return TokError("unassigned file number in '.loc' directive"); 2844 Lex(); 2845 2846 int64_t LineNumber = 0; 2847 if (getLexer().is(AsmToken::Integer)) { 2848 LineNumber = getTok().getIntVal(); 2849 if (LineNumber < 0) 2850 return TokError("line number less than zero in '.loc' directive"); 2851 Lex(); 2852 } 2853 2854 int64_t ColumnPos = 0; 2855 if (getLexer().is(AsmToken::Integer)) { 2856 ColumnPos = getTok().getIntVal(); 2857 if (ColumnPos < 0) 2858 return TokError("column position less than zero in '.loc' directive"); 2859 Lex(); 2860 } 2861 2862 unsigned Flags = DWARF2_LINE_DEFAULT_IS_STMT ? DWARF2_FLAG_IS_STMT : 0; 2863 unsigned Isa = 0; 2864 int64_t Discriminator = 0; 2865 if (getLexer().isNot(AsmToken::EndOfStatement)) { 2866 for (;;) { 2867 if (getLexer().is(AsmToken::EndOfStatement)) 2868 break; 2869 2870 StringRef Name; 2871 SMLoc Loc = getTok().getLoc(); 2872 if (parseIdentifier(Name)) 2873 return TokError("unexpected token in '.loc' directive"); 2874 2875 if (Name == "basic_block") 2876 Flags |= DWARF2_FLAG_BASIC_BLOCK; 2877 else if (Name == "prologue_end") 2878 Flags |= DWARF2_FLAG_PROLOGUE_END; 2879 else if (Name == "epilogue_begin") 2880 Flags |= DWARF2_FLAG_EPILOGUE_BEGIN; 2881 else if (Name == "is_stmt") { 2882 Loc = getTok().getLoc(); 2883 const MCExpr *Value; 2884 if (parseExpression(Value)) 2885 return true; 2886 // The expression must be the constant 0 or 1. 2887 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2888 int Value = MCE->getValue(); 2889 if (Value == 0) 2890 Flags &= ~DWARF2_FLAG_IS_STMT; 2891 else if (Value == 1) 2892 Flags |= DWARF2_FLAG_IS_STMT; 2893 else 2894 return Error(Loc, "is_stmt value not 0 or 1"); 2895 } else { 2896 return Error(Loc, "is_stmt value not the constant value of 0 or 1"); 2897 } 2898 } else if (Name == "isa") { 2899 Loc = getTok().getLoc(); 2900 const MCExpr *Value; 2901 if (parseExpression(Value)) 2902 return true; 2903 // The expression must be a constant greater or equal to 0. 2904 if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value)) { 2905 int Value = MCE->getValue(); 2906 if (Value < 0) 2907 return Error(Loc, "isa number less than zero"); 2908 Isa = Value; 2909 } else { 2910 return Error(Loc, "isa number not a constant value"); 2911 } 2912 } else if (Name == "discriminator") { 2913 if (parseAbsoluteExpression(Discriminator)) 2914 return true; 2915 } else { 2916 return Error(Loc, "unknown sub-directive in '.loc' directive"); 2917 } 2918 2919 if (getLexer().is(AsmToken::EndOfStatement)) 2920 break; 2921 } 2922 } 2923 2924 getStreamer().EmitDwarfLocDirective(FileNumber, LineNumber, ColumnPos, Flags, 2925 Isa, Discriminator, StringRef()); 2926 2927 return false; 2928 } 2929 2930 /// parseDirectiveStabs 2931 /// ::= .stabs string, number, number, number 2932 bool AsmParser::parseDirectiveStabs() { 2933 return TokError("unsupported directive '.stabs'"); 2934 } 2935 2936 /// parseDirectiveCFISections 2937 /// ::= .cfi_sections section [, section] 2938 bool AsmParser::parseDirectiveCFISections() { 2939 StringRef Name; 2940 bool EH = false; 2941 bool Debug = false; 2942 2943 if (parseIdentifier(Name)) 2944 return TokError("Expected an identifier"); 2945 2946 if (Name == ".eh_frame") 2947 EH = true; 2948 else if (Name == ".debug_frame") 2949 Debug = true; 2950 2951 if (getLexer().is(AsmToken::Comma)) { 2952 Lex(); 2953 2954 if (parseIdentifier(Name)) 2955 return TokError("Expected an identifier"); 2956 2957 if (Name == ".eh_frame") 2958 EH = true; 2959 else if (Name == ".debug_frame") 2960 Debug = true; 2961 } 2962 2963 getStreamer().EmitCFISections(EH, Debug); 2964 return false; 2965 } 2966 2967 /// parseDirectiveCFIStartProc 2968 /// ::= .cfi_startproc [simple] 2969 bool AsmParser::parseDirectiveCFIStartProc() { 2970 StringRef Simple; 2971 if (getLexer().isNot(AsmToken::EndOfStatement)) 2972 if (parseIdentifier(Simple) || Simple != "simple") 2973 return TokError("unexpected token in .cfi_startproc directive"); 2974 2975 getStreamer().EmitCFIStartProc(!Simple.empty()); 2976 return false; 2977 } 2978 2979 /// parseDirectiveCFIEndProc 2980 /// ::= .cfi_endproc 2981 bool AsmParser::parseDirectiveCFIEndProc() { 2982 getStreamer().EmitCFIEndProc(); 2983 return false; 2984 } 2985 2986 /// \brief parse register name or number. 2987 bool AsmParser::parseRegisterOrRegisterNumber(int64_t &Register, 2988 SMLoc DirectiveLoc) { 2989 unsigned RegNo; 2990 2991 if (getLexer().isNot(AsmToken::Integer)) { 2992 if (getTargetParser().ParseRegister(RegNo, DirectiveLoc, DirectiveLoc)) 2993 return true; 2994 Register = getContext().getRegisterInfo()->getDwarfRegNum(RegNo, true); 2995 } else 2996 return parseAbsoluteExpression(Register); 2997 2998 return false; 2999 } 3000 3001 /// parseDirectiveCFIDefCfa 3002 /// ::= .cfi_def_cfa register, offset 3003 bool AsmParser::parseDirectiveCFIDefCfa(SMLoc DirectiveLoc) { 3004 int64_t Register = 0; 3005 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3006 return true; 3007 3008 if (getLexer().isNot(AsmToken::Comma)) 3009 return TokError("unexpected token in directive"); 3010 Lex(); 3011 3012 int64_t Offset = 0; 3013 if (parseAbsoluteExpression(Offset)) 3014 return true; 3015 3016 getStreamer().EmitCFIDefCfa(Register, Offset); 3017 return false; 3018 } 3019 3020 /// parseDirectiveCFIDefCfaOffset 3021 /// ::= .cfi_def_cfa_offset offset 3022 bool AsmParser::parseDirectiveCFIDefCfaOffset() { 3023 int64_t Offset = 0; 3024 if (parseAbsoluteExpression(Offset)) 3025 return true; 3026 3027 getStreamer().EmitCFIDefCfaOffset(Offset); 3028 return false; 3029 } 3030 3031 /// parseDirectiveCFIRegister 3032 /// ::= .cfi_register register, register 3033 bool AsmParser::parseDirectiveCFIRegister(SMLoc DirectiveLoc) { 3034 int64_t Register1 = 0; 3035 if (parseRegisterOrRegisterNumber(Register1, DirectiveLoc)) 3036 return true; 3037 3038 if (getLexer().isNot(AsmToken::Comma)) 3039 return TokError("unexpected token in directive"); 3040 Lex(); 3041 3042 int64_t Register2 = 0; 3043 if (parseRegisterOrRegisterNumber(Register2, DirectiveLoc)) 3044 return true; 3045 3046 getStreamer().EmitCFIRegister(Register1, Register2); 3047 return false; 3048 } 3049 3050 /// parseDirectiveCFIWindowSave 3051 /// ::= .cfi_window_save 3052 bool AsmParser::parseDirectiveCFIWindowSave() { 3053 getStreamer().EmitCFIWindowSave(); 3054 return false; 3055 } 3056 3057 /// parseDirectiveCFIAdjustCfaOffset 3058 /// ::= .cfi_adjust_cfa_offset adjustment 3059 bool AsmParser::parseDirectiveCFIAdjustCfaOffset() { 3060 int64_t Adjustment = 0; 3061 if (parseAbsoluteExpression(Adjustment)) 3062 return true; 3063 3064 getStreamer().EmitCFIAdjustCfaOffset(Adjustment); 3065 return false; 3066 } 3067 3068 /// parseDirectiveCFIDefCfaRegister 3069 /// ::= .cfi_def_cfa_register register 3070 bool AsmParser::parseDirectiveCFIDefCfaRegister(SMLoc DirectiveLoc) { 3071 int64_t Register = 0; 3072 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3073 return true; 3074 3075 getStreamer().EmitCFIDefCfaRegister(Register); 3076 return false; 3077 } 3078 3079 /// parseDirectiveCFIOffset 3080 /// ::= .cfi_offset register, offset 3081 bool AsmParser::parseDirectiveCFIOffset(SMLoc DirectiveLoc) { 3082 int64_t Register = 0; 3083 int64_t Offset = 0; 3084 3085 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3086 return true; 3087 3088 if (getLexer().isNot(AsmToken::Comma)) 3089 return TokError("unexpected token in directive"); 3090 Lex(); 3091 3092 if (parseAbsoluteExpression(Offset)) 3093 return true; 3094 3095 getStreamer().EmitCFIOffset(Register, Offset); 3096 return false; 3097 } 3098 3099 /// parseDirectiveCFIRelOffset 3100 /// ::= .cfi_rel_offset register, offset 3101 bool AsmParser::parseDirectiveCFIRelOffset(SMLoc DirectiveLoc) { 3102 int64_t Register = 0; 3103 3104 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3105 return true; 3106 3107 if (getLexer().isNot(AsmToken::Comma)) 3108 return TokError("unexpected token in directive"); 3109 Lex(); 3110 3111 int64_t Offset = 0; 3112 if (parseAbsoluteExpression(Offset)) 3113 return true; 3114 3115 getStreamer().EmitCFIRelOffset(Register, Offset); 3116 return false; 3117 } 3118 3119 static bool isValidEncoding(int64_t Encoding) { 3120 if (Encoding & ~0xff) 3121 return false; 3122 3123 if (Encoding == dwarf::DW_EH_PE_omit) 3124 return true; 3125 3126 const unsigned Format = Encoding & 0xf; 3127 if (Format != dwarf::DW_EH_PE_absptr && Format != dwarf::DW_EH_PE_udata2 && 3128 Format != dwarf::DW_EH_PE_udata4 && Format != dwarf::DW_EH_PE_udata8 && 3129 Format != dwarf::DW_EH_PE_sdata2 && Format != dwarf::DW_EH_PE_sdata4 && 3130 Format != dwarf::DW_EH_PE_sdata8 && Format != dwarf::DW_EH_PE_signed) 3131 return false; 3132 3133 const unsigned Application = Encoding & 0x70; 3134 if (Application != dwarf::DW_EH_PE_absptr && 3135 Application != dwarf::DW_EH_PE_pcrel) 3136 return false; 3137 3138 return true; 3139 } 3140 3141 /// parseDirectiveCFIPersonalityOrLsda 3142 /// IsPersonality true for cfi_personality, false for cfi_lsda 3143 /// ::= .cfi_personality encoding, [symbol_name] 3144 /// ::= .cfi_lsda encoding, [symbol_name] 3145 bool AsmParser::parseDirectiveCFIPersonalityOrLsda(bool IsPersonality) { 3146 int64_t Encoding = 0; 3147 if (parseAbsoluteExpression(Encoding)) 3148 return true; 3149 if (Encoding == dwarf::DW_EH_PE_omit) 3150 return false; 3151 3152 if (!isValidEncoding(Encoding)) 3153 return TokError("unsupported encoding."); 3154 3155 if (getLexer().isNot(AsmToken::Comma)) 3156 return TokError("unexpected token in directive"); 3157 Lex(); 3158 3159 StringRef Name; 3160 if (parseIdentifier(Name)) 3161 return TokError("expected identifier in directive"); 3162 3163 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name); 3164 3165 if (IsPersonality) 3166 getStreamer().EmitCFIPersonality(Sym, Encoding); 3167 else 3168 getStreamer().EmitCFILsda(Sym, Encoding); 3169 return false; 3170 } 3171 3172 /// parseDirectiveCFIRememberState 3173 /// ::= .cfi_remember_state 3174 bool AsmParser::parseDirectiveCFIRememberState() { 3175 getStreamer().EmitCFIRememberState(); 3176 return false; 3177 } 3178 3179 /// parseDirectiveCFIRestoreState 3180 /// ::= .cfi_remember_state 3181 bool AsmParser::parseDirectiveCFIRestoreState() { 3182 getStreamer().EmitCFIRestoreState(); 3183 return false; 3184 } 3185 3186 /// parseDirectiveCFISameValue 3187 /// ::= .cfi_same_value register 3188 bool AsmParser::parseDirectiveCFISameValue(SMLoc DirectiveLoc) { 3189 int64_t Register = 0; 3190 3191 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3192 return true; 3193 3194 getStreamer().EmitCFISameValue(Register); 3195 return false; 3196 } 3197 3198 /// parseDirectiveCFIRestore 3199 /// ::= .cfi_restore register 3200 bool AsmParser::parseDirectiveCFIRestore(SMLoc DirectiveLoc) { 3201 int64_t Register = 0; 3202 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3203 return true; 3204 3205 getStreamer().EmitCFIRestore(Register); 3206 return false; 3207 } 3208 3209 /// parseDirectiveCFIEscape 3210 /// ::= .cfi_escape expression[,...] 3211 bool AsmParser::parseDirectiveCFIEscape() { 3212 std::string Values; 3213 int64_t CurrValue; 3214 if (parseAbsoluteExpression(CurrValue)) 3215 return true; 3216 3217 Values.push_back((uint8_t)CurrValue); 3218 3219 while (getLexer().is(AsmToken::Comma)) { 3220 Lex(); 3221 3222 if (parseAbsoluteExpression(CurrValue)) 3223 return true; 3224 3225 Values.push_back((uint8_t)CurrValue); 3226 } 3227 3228 getStreamer().EmitCFIEscape(Values); 3229 return false; 3230 } 3231 3232 /// parseDirectiveCFISignalFrame 3233 /// ::= .cfi_signal_frame 3234 bool AsmParser::parseDirectiveCFISignalFrame() { 3235 if (getLexer().isNot(AsmToken::EndOfStatement)) 3236 return Error(getLexer().getLoc(), 3237 "unexpected token in '.cfi_signal_frame'"); 3238 3239 getStreamer().EmitCFISignalFrame(); 3240 return false; 3241 } 3242 3243 /// parseDirectiveCFIUndefined 3244 /// ::= .cfi_undefined register 3245 bool AsmParser::parseDirectiveCFIUndefined(SMLoc DirectiveLoc) { 3246 int64_t Register = 0; 3247 3248 if (parseRegisterOrRegisterNumber(Register, DirectiveLoc)) 3249 return true; 3250 3251 getStreamer().EmitCFIUndefined(Register); 3252 return false; 3253 } 3254 3255 /// parseDirectiveMacrosOnOff 3256 /// ::= .macros_on 3257 /// ::= .macros_off 3258 bool AsmParser::parseDirectiveMacrosOnOff(StringRef Directive) { 3259 if (getLexer().isNot(AsmToken::EndOfStatement)) 3260 return Error(getLexer().getLoc(), 3261 "unexpected token in '" + Directive + "' directive"); 3262 3263 setMacrosEnabled(Directive == ".macros_on"); 3264 return false; 3265 } 3266 3267 /// parseDirectiveMacro 3268 /// ::= .macro name[,] [parameters] 3269 bool AsmParser::parseDirectiveMacro(SMLoc DirectiveLoc) { 3270 StringRef Name; 3271 if (parseIdentifier(Name)) 3272 return TokError("expected identifier in '.macro' directive"); 3273 3274 if (getLexer().is(AsmToken::Comma)) 3275 Lex(); 3276 3277 MCAsmMacroParameters Parameters; 3278 while (getLexer().isNot(AsmToken::EndOfStatement)) { 3279 3280 if (!Parameters.empty() && Parameters.back().Vararg) 3281 return Error(Lexer.getLoc(), 3282 "Vararg parameter '" + Parameters.back().Name + 3283 "' should be last one in the list of parameters."); 3284 3285 MCAsmMacroParameter Parameter; 3286 if (parseIdentifier(Parameter.Name)) 3287 return TokError("expected identifier in '.macro' directive"); 3288 3289 if (Lexer.is(AsmToken::Colon)) { 3290 Lex(); // consume ':' 3291 3292 SMLoc QualLoc; 3293 StringRef Qualifier; 3294 3295 QualLoc = Lexer.getLoc(); 3296 if (parseIdentifier(Qualifier)) 3297 return Error(QualLoc, "missing parameter qualifier for " 3298 "'" + Parameter.Name + "' in macro '" + Name + "'"); 3299 3300 if (Qualifier == "req") 3301 Parameter.Required = true; 3302 else if (Qualifier == "vararg") 3303 Parameter.Vararg = true; 3304 else 3305 return Error(QualLoc, Qualifier + " is not a valid parameter qualifier " 3306 "for '" + Parameter.Name + "' in macro '" + Name + "'"); 3307 } 3308 3309 if (getLexer().is(AsmToken::Equal)) { 3310 Lex(); 3311 3312 SMLoc ParamLoc; 3313 3314 ParamLoc = Lexer.getLoc(); 3315 if (parseMacroArgument(Parameter.Value, /*Vararg=*/false )) 3316 return true; 3317 3318 if (Parameter.Required) 3319 Warning(ParamLoc, "pointless default value for required parameter " 3320 "'" + Parameter.Name + "' in macro '" + Name + "'"); 3321 } 3322 3323 Parameters.push_back(std::move(Parameter)); 3324 3325 if (getLexer().is(AsmToken::Comma)) 3326 Lex(); 3327 } 3328 3329 // Eat the end of statement. 3330 Lex(); 3331 3332 AsmToken EndToken, StartToken = getTok(); 3333 unsigned MacroDepth = 0; 3334 3335 // Lex the macro definition. 3336 for (;;) { 3337 // Check whether we have reached the end of the file. 3338 if (getLexer().is(AsmToken::Eof)) 3339 return Error(DirectiveLoc, "no matching '.endmacro' in definition"); 3340 3341 // Otherwise, check whether we have reach the .endmacro. 3342 if (getLexer().is(AsmToken::Identifier)) { 3343 if (getTok().getIdentifier() == ".endm" || 3344 getTok().getIdentifier() == ".endmacro") { 3345 if (MacroDepth == 0) { // Outermost macro. 3346 EndToken = getTok(); 3347 Lex(); 3348 if (getLexer().isNot(AsmToken::EndOfStatement)) 3349 return TokError("unexpected token in '" + EndToken.getIdentifier() + 3350 "' directive"); 3351 break; 3352 } else { 3353 // Otherwise we just found the end of an inner macro. 3354 --MacroDepth; 3355 } 3356 } else if (getTok().getIdentifier() == ".macro") { 3357 // We allow nested macros. Those aren't instantiated until the outermost 3358 // macro is expanded so just ignore them for now. 3359 ++MacroDepth; 3360 } 3361 } 3362 3363 // Otherwise, scan til the end of the statement. 3364 eatToEndOfStatement(); 3365 } 3366 3367 if (lookupMacro(Name)) { 3368 return Error(DirectiveLoc, "macro '" + Name + "' is already defined"); 3369 } 3370 3371 const char *BodyStart = StartToken.getLoc().getPointer(); 3372 const char *BodyEnd = EndToken.getLoc().getPointer(); 3373 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 3374 checkForBadMacro(DirectiveLoc, Name, Body, Parameters); 3375 defineMacro(Name, MCAsmMacro(Name, Body, std::move(Parameters))); 3376 return false; 3377 } 3378 3379 /// checkForBadMacro 3380 /// 3381 /// With the support added for named parameters there may be code out there that 3382 /// is transitioning from positional parameters. In versions of gas that did 3383 /// not support named parameters they would be ignored on the macro definition. 3384 /// But to support both styles of parameters this is not possible so if a macro 3385 /// definition has named parameters but does not use them and has what appears 3386 /// to be positional parameters, strings like $1, $2, ... and $n, then issue a 3387 /// warning that the positional parameter found in body which have no effect. 3388 /// Hoping the developer will either remove the named parameters from the macro 3389 /// definition so the positional parameters get used if that was what was 3390 /// intended or change the macro to use the named parameters. It is possible 3391 /// this warning will trigger when the none of the named parameters are used 3392 /// and the strings like $1 are infact to simply to be passed trough unchanged. 3393 void AsmParser::checkForBadMacro(SMLoc DirectiveLoc, StringRef Name, 3394 StringRef Body, 3395 ArrayRef<MCAsmMacroParameter> Parameters) { 3396 // If this macro is not defined with named parameters the warning we are 3397 // checking for here doesn't apply. 3398 unsigned NParameters = Parameters.size(); 3399 if (NParameters == 0) 3400 return; 3401 3402 bool NamedParametersFound = false; 3403 bool PositionalParametersFound = false; 3404 3405 // Look at the body of the macro for use of both the named parameters and what 3406 // are likely to be positional parameters. This is what expandMacro() is 3407 // doing when it finds the parameters in the body. 3408 while (!Body.empty()) { 3409 // Scan for the next possible parameter. 3410 std::size_t End = Body.size(), Pos = 0; 3411 for (; Pos != End; ++Pos) { 3412 // Check for a substitution or escape. 3413 // This macro is defined with parameters, look for \foo, \bar, etc. 3414 if (Body[Pos] == '\\' && Pos + 1 != End) 3415 break; 3416 3417 // This macro should have parameters, but look for $0, $1, ..., $n too. 3418 if (Body[Pos] != '$' || Pos + 1 == End) 3419 continue; 3420 char Next = Body[Pos + 1]; 3421 if (Next == '$' || Next == 'n' || 3422 isdigit(static_cast<unsigned char>(Next))) 3423 break; 3424 } 3425 3426 // Check if we reached the end. 3427 if (Pos == End) 3428 break; 3429 3430 if (Body[Pos] == '$') { 3431 switch (Body[Pos + 1]) { 3432 // $$ => $ 3433 case '$': 3434 break; 3435 3436 // $n => number of arguments 3437 case 'n': 3438 PositionalParametersFound = true; 3439 break; 3440 3441 // $[0-9] => argument 3442 default: { 3443 PositionalParametersFound = true; 3444 break; 3445 } 3446 } 3447 Pos += 2; 3448 } else { 3449 unsigned I = Pos + 1; 3450 while (isIdentifierChar(Body[I]) && I + 1 != End) 3451 ++I; 3452 3453 const char *Begin = Body.data() + Pos + 1; 3454 StringRef Argument(Begin, I - (Pos + 1)); 3455 unsigned Index = 0; 3456 for (; Index < NParameters; ++Index) 3457 if (Parameters[Index].Name == Argument) 3458 break; 3459 3460 if (Index == NParameters) { 3461 if (Body[Pos + 1] == '(' && Body[Pos + 2] == ')') 3462 Pos += 3; 3463 else { 3464 Pos = I; 3465 } 3466 } else { 3467 NamedParametersFound = true; 3468 Pos += 1 + Argument.size(); 3469 } 3470 } 3471 // Update the scan point. 3472 Body = Body.substr(Pos); 3473 } 3474 3475 if (!NamedParametersFound && PositionalParametersFound) 3476 Warning(DirectiveLoc, "macro defined with named parameters which are not " 3477 "used in macro body, possible positional parameter " 3478 "found in body which will have no effect"); 3479 } 3480 3481 /// parseDirectiveExitMacro 3482 /// ::= .exitm 3483 bool AsmParser::parseDirectiveExitMacro(StringRef Directive) { 3484 if (getLexer().isNot(AsmToken::EndOfStatement)) 3485 return TokError("unexpected token in '" + Directive + "' directive"); 3486 3487 if (!isInsideMacroInstantiation()) 3488 return TokError("unexpected '" + Directive + "' in file, " 3489 "no current macro definition"); 3490 3491 // Exit all conditionals that are active in the current macro. 3492 while (TheCondStack.size() != ActiveMacros.back()->CondStackDepth) { 3493 TheCondState = TheCondStack.back(); 3494 TheCondStack.pop_back(); 3495 } 3496 3497 handleMacroExit(); 3498 return false; 3499 } 3500 3501 /// parseDirectiveEndMacro 3502 /// ::= .endm 3503 /// ::= .endmacro 3504 bool AsmParser::parseDirectiveEndMacro(StringRef Directive) { 3505 if (getLexer().isNot(AsmToken::EndOfStatement)) 3506 return TokError("unexpected token in '" + Directive + "' directive"); 3507 3508 // If we are inside a macro instantiation, terminate the current 3509 // instantiation. 3510 if (isInsideMacroInstantiation()) { 3511 handleMacroExit(); 3512 return false; 3513 } 3514 3515 // Otherwise, this .endmacro is a stray entry in the file; well formed 3516 // .endmacro directives are handled during the macro definition parsing. 3517 return TokError("unexpected '" + Directive + "' in file, " 3518 "no current macro definition"); 3519 } 3520 3521 /// parseDirectivePurgeMacro 3522 /// ::= .purgem 3523 bool AsmParser::parseDirectivePurgeMacro(SMLoc DirectiveLoc) { 3524 StringRef Name; 3525 if (parseIdentifier(Name)) 3526 return TokError("expected identifier in '.purgem' directive"); 3527 3528 if (getLexer().isNot(AsmToken::EndOfStatement)) 3529 return TokError("unexpected token in '.purgem' directive"); 3530 3531 if (!lookupMacro(Name)) 3532 return Error(DirectiveLoc, "macro '" + Name + "' is not defined"); 3533 3534 undefineMacro(Name); 3535 return false; 3536 } 3537 3538 /// parseDirectiveBundleAlignMode 3539 /// ::= {.bundle_align_mode} expression 3540 bool AsmParser::parseDirectiveBundleAlignMode() { 3541 checkForValidSection(); 3542 3543 // Expect a single argument: an expression that evaluates to a constant 3544 // in the inclusive range 0-30. 3545 SMLoc ExprLoc = getLexer().getLoc(); 3546 int64_t AlignSizePow2; 3547 if (parseAbsoluteExpression(AlignSizePow2)) 3548 return true; 3549 else if (getLexer().isNot(AsmToken::EndOfStatement)) 3550 return TokError("unexpected token after expression in" 3551 " '.bundle_align_mode' directive"); 3552 else if (AlignSizePow2 < 0 || AlignSizePow2 > 30) 3553 return Error(ExprLoc, 3554 "invalid bundle alignment size (expected between 0 and 30)"); 3555 3556 Lex(); 3557 3558 // Because of AlignSizePow2's verified range we can safely truncate it to 3559 // unsigned. 3560 getStreamer().EmitBundleAlignMode(static_cast<unsigned>(AlignSizePow2)); 3561 return false; 3562 } 3563 3564 /// parseDirectiveBundleLock 3565 /// ::= {.bundle_lock} [align_to_end] 3566 bool AsmParser::parseDirectiveBundleLock() { 3567 checkForValidSection(); 3568 bool AlignToEnd = false; 3569 3570 if (getLexer().isNot(AsmToken::EndOfStatement)) { 3571 StringRef Option; 3572 SMLoc Loc = getTok().getLoc(); 3573 const char *kInvalidOptionError = 3574 "invalid option for '.bundle_lock' directive"; 3575 3576 if (parseIdentifier(Option)) 3577 return Error(Loc, kInvalidOptionError); 3578 3579 if (Option != "align_to_end") 3580 return Error(Loc, kInvalidOptionError); 3581 else if (getLexer().isNot(AsmToken::EndOfStatement)) 3582 return Error(Loc, 3583 "unexpected token after '.bundle_lock' directive option"); 3584 AlignToEnd = true; 3585 } 3586 3587 Lex(); 3588 3589 getStreamer().EmitBundleLock(AlignToEnd); 3590 return false; 3591 } 3592 3593 /// parseDirectiveBundleLock 3594 /// ::= {.bundle_lock} 3595 bool AsmParser::parseDirectiveBundleUnlock() { 3596 checkForValidSection(); 3597 3598 if (getLexer().isNot(AsmToken::EndOfStatement)) 3599 return TokError("unexpected token in '.bundle_unlock' directive"); 3600 Lex(); 3601 3602 getStreamer().EmitBundleUnlock(); 3603 return false; 3604 } 3605 3606 /// parseDirectiveSpace 3607 /// ::= (.skip | .space) expression [ , expression ] 3608 bool AsmParser::parseDirectiveSpace(StringRef IDVal) { 3609 checkForValidSection(); 3610 3611 int64_t NumBytes; 3612 if (parseAbsoluteExpression(NumBytes)) 3613 return true; 3614 3615 int64_t FillExpr = 0; 3616 if (getLexer().isNot(AsmToken::EndOfStatement)) { 3617 if (getLexer().isNot(AsmToken::Comma)) 3618 return TokError("unexpected token in '" + Twine(IDVal) + "' directive"); 3619 Lex(); 3620 3621 if (parseAbsoluteExpression(FillExpr)) 3622 return true; 3623 3624 if (getLexer().isNot(AsmToken::EndOfStatement)) 3625 return TokError("unexpected token in '" + Twine(IDVal) + "' directive"); 3626 } 3627 3628 Lex(); 3629 3630 if (NumBytes <= 0) 3631 return TokError("invalid number of bytes in '" + Twine(IDVal) + 3632 "' directive"); 3633 3634 // FIXME: Sometimes the fill expr is 'nop' if it isn't supplied, instead of 0. 3635 getStreamer().EmitFill(NumBytes, FillExpr); 3636 3637 return false; 3638 } 3639 3640 /// parseDirectiveLEB128 3641 /// ::= (.sleb128 | .uleb128) [ expression (, expression)* ] 3642 bool AsmParser::parseDirectiveLEB128(bool Signed) { 3643 checkForValidSection(); 3644 const MCExpr *Value; 3645 3646 for (;;) { 3647 if (parseExpression(Value)) 3648 return true; 3649 3650 if (Signed) 3651 getStreamer().EmitSLEB128Value(Value); 3652 else 3653 getStreamer().EmitULEB128Value(Value); 3654 3655 if (getLexer().is(AsmToken::EndOfStatement)) 3656 break; 3657 3658 if (getLexer().isNot(AsmToken::Comma)) 3659 return TokError("unexpected token in directive"); 3660 Lex(); 3661 } 3662 3663 return false; 3664 } 3665 3666 /// parseDirectiveSymbolAttribute 3667 /// ::= { ".globl", ".weak", ... } [ identifier ( , identifier )* ] 3668 bool AsmParser::parseDirectiveSymbolAttribute(MCSymbolAttr Attr) { 3669 if (getLexer().isNot(AsmToken::EndOfStatement)) { 3670 for (;;) { 3671 StringRef Name; 3672 SMLoc Loc = getTok().getLoc(); 3673 3674 if (parseIdentifier(Name)) 3675 return Error(Loc, "expected identifier in directive"); 3676 3677 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name); 3678 3679 // Assembler local symbols don't make any sense here. Complain loudly. 3680 if (Sym->isTemporary()) 3681 return Error(Loc, "non-local symbol required in directive"); 3682 3683 if (!getStreamer().EmitSymbolAttribute(Sym, Attr)) 3684 return Error(Loc, "unable to emit symbol attribute"); 3685 3686 if (getLexer().is(AsmToken::EndOfStatement)) 3687 break; 3688 3689 if (getLexer().isNot(AsmToken::Comma)) 3690 return TokError("unexpected token in directive"); 3691 Lex(); 3692 } 3693 } 3694 3695 Lex(); 3696 return false; 3697 } 3698 3699 /// parseDirectiveComm 3700 /// ::= ( .comm | .lcomm ) identifier , size_expression [ , align_expression ] 3701 bool AsmParser::parseDirectiveComm(bool IsLocal) { 3702 checkForValidSection(); 3703 3704 SMLoc IDLoc = getLexer().getLoc(); 3705 StringRef Name; 3706 if (parseIdentifier(Name)) 3707 return TokError("expected identifier in directive"); 3708 3709 // Handle the identifier as the key symbol. 3710 MCSymbol *Sym = getContext().GetOrCreateSymbol(Name); 3711 3712 if (getLexer().isNot(AsmToken::Comma)) 3713 return TokError("unexpected token in directive"); 3714 Lex(); 3715 3716 int64_t Size; 3717 SMLoc SizeLoc = getLexer().getLoc(); 3718 if (parseAbsoluteExpression(Size)) 3719 return true; 3720 3721 int64_t Pow2Alignment = 0; 3722 SMLoc Pow2AlignmentLoc; 3723 if (getLexer().is(AsmToken::Comma)) { 3724 Lex(); 3725 Pow2AlignmentLoc = getLexer().getLoc(); 3726 if (parseAbsoluteExpression(Pow2Alignment)) 3727 return true; 3728 3729 LCOMM::LCOMMType LCOMM = Lexer.getMAI().getLCOMMDirectiveAlignmentType(); 3730 if (IsLocal && LCOMM == LCOMM::NoAlignment) 3731 return Error(Pow2AlignmentLoc, "alignment not supported on this target"); 3732 3733 // If this target takes alignments in bytes (not log) validate and convert. 3734 if ((!IsLocal && Lexer.getMAI().getCOMMDirectiveAlignmentIsInBytes()) || 3735 (IsLocal && LCOMM == LCOMM::ByteAlignment)) { 3736 if (!isPowerOf2_64(Pow2Alignment)) 3737 return Error(Pow2AlignmentLoc, "alignment must be a power of 2"); 3738 Pow2Alignment = Log2_64(Pow2Alignment); 3739 } 3740 } 3741 3742 if (getLexer().isNot(AsmToken::EndOfStatement)) 3743 return TokError("unexpected token in '.comm' or '.lcomm' directive"); 3744 3745 Lex(); 3746 3747 // NOTE: a size of zero for a .comm should create a undefined symbol 3748 // but a size of .lcomm creates a bss symbol of size zero. 3749 if (Size < 0) 3750 return Error(SizeLoc, "invalid '.comm' or '.lcomm' directive size, can't " 3751 "be less than zero"); 3752 3753 // NOTE: The alignment in the directive is a power of 2 value, the assembler 3754 // may internally end up wanting an alignment in bytes. 3755 // FIXME: Diagnose overflow. 3756 if (Pow2Alignment < 0) 3757 return Error(Pow2AlignmentLoc, "invalid '.comm' or '.lcomm' directive " 3758 "alignment, can't be less than zero"); 3759 3760 if (!Sym->isUndefined()) 3761 return Error(IDLoc, "invalid symbol redefinition"); 3762 3763 // Create the Symbol as a common or local common with Size and Pow2Alignment 3764 if (IsLocal) { 3765 getStreamer().EmitLocalCommonSymbol(Sym, Size, 1 << Pow2Alignment); 3766 return false; 3767 } 3768 3769 getStreamer().EmitCommonSymbol(Sym, Size, 1 << Pow2Alignment); 3770 return false; 3771 } 3772 3773 /// parseDirectiveAbort 3774 /// ::= .abort [... message ...] 3775 bool AsmParser::parseDirectiveAbort() { 3776 // FIXME: Use loc from directive. 3777 SMLoc Loc = getLexer().getLoc(); 3778 3779 StringRef Str = parseStringToEndOfStatement(); 3780 if (getLexer().isNot(AsmToken::EndOfStatement)) 3781 return TokError("unexpected token in '.abort' directive"); 3782 3783 Lex(); 3784 3785 if (Str.empty()) 3786 Error(Loc, ".abort detected. Assembly stopping."); 3787 else 3788 Error(Loc, ".abort '" + Str + "' detected. Assembly stopping."); 3789 // FIXME: Actually abort assembly here. 3790 3791 return false; 3792 } 3793 3794 /// parseDirectiveInclude 3795 /// ::= .include "filename" 3796 bool AsmParser::parseDirectiveInclude() { 3797 if (getLexer().isNot(AsmToken::String)) 3798 return TokError("expected string in '.include' directive"); 3799 3800 // Allow the strings to have escaped octal character sequence. 3801 std::string Filename; 3802 if (parseEscapedString(Filename)) 3803 return true; 3804 SMLoc IncludeLoc = getLexer().getLoc(); 3805 Lex(); 3806 3807 if (getLexer().isNot(AsmToken::EndOfStatement)) 3808 return TokError("unexpected token in '.include' directive"); 3809 3810 // Attempt to switch the lexer to the included file before consuming the end 3811 // of statement to avoid losing it when we switch. 3812 if (enterIncludeFile(Filename)) { 3813 Error(IncludeLoc, "Could not find include file '" + Filename + "'"); 3814 return true; 3815 } 3816 3817 return false; 3818 } 3819 3820 /// parseDirectiveIncbin 3821 /// ::= .incbin "filename" 3822 bool AsmParser::parseDirectiveIncbin() { 3823 if (getLexer().isNot(AsmToken::String)) 3824 return TokError("expected string in '.incbin' directive"); 3825 3826 // Allow the strings to have escaped octal character sequence. 3827 std::string Filename; 3828 if (parseEscapedString(Filename)) 3829 return true; 3830 SMLoc IncbinLoc = getLexer().getLoc(); 3831 Lex(); 3832 3833 if (getLexer().isNot(AsmToken::EndOfStatement)) 3834 return TokError("unexpected token in '.incbin' directive"); 3835 3836 // Attempt to process the included file. 3837 if (processIncbinFile(Filename)) { 3838 Error(IncbinLoc, "Could not find incbin file '" + Filename + "'"); 3839 return true; 3840 } 3841 3842 return false; 3843 } 3844 3845 /// parseDirectiveIf 3846 /// ::= .if{,eq,ge,gt,le,lt,ne} expression 3847 bool AsmParser::parseDirectiveIf(SMLoc DirectiveLoc, DirectiveKind DirKind) { 3848 TheCondStack.push_back(TheCondState); 3849 TheCondState.TheCond = AsmCond::IfCond; 3850 if (TheCondState.Ignore) { 3851 eatToEndOfStatement(); 3852 } else { 3853 int64_t ExprValue; 3854 if (parseAbsoluteExpression(ExprValue)) 3855 return true; 3856 3857 if (getLexer().isNot(AsmToken::EndOfStatement)) 3858 return TokError("unexpected token in '.if' directive"); 3859 3860 Lex(); 3861 3862 switch (DirKind) { 3863 default: 3864 llvm_unreachable("unsupported directive"); 3865 case DK_IF: 3866 case DK_IFNE: 3867 break; 3868 case DK_IFEQ: 3869 ExprValue = ExprValue == 0; 3870 break; 3871 case DK_IFGE: 3872 ExprValue = ExprValue >= 0; 3873 break; 3874 case DK_IFGT: 3875 ExprValue = ExprValue > 0; 3876 break; 3877 case DK_IFLE: 3878 ExprValue = ExprValue <= 0; 3879 break; 3880 case DK_IFLT: 3881 ExprValue = ExprValue < 0; 3882 break; 3883 } 3884 3885 TheCondState.CondMet = ExprValue; 3886 TheCondState.Ignore = !TheCondState.CondMet; 3887 } 3888 3889 return false; 3890 } 3891 3892 /// parseDirectiveIfb 3893 /// ::= .ifb string 3894 bool AsmParser::parseDirectiveIfb(SMLoc DirectiveLoc, bool ExpectBlank) { 3895 TheCondStack.push_back(TheCondState); 3896 TheCondState.TheCond = AsmCond::IfCond; 3897 3898 if (TheCondState.Ignore) { 3899 eatToEndOfStatement(); 3900 } else { 3901 StringRef Str = parseStringToEndOfStatement(); 3902 3903 if (getLexer().isNot(AsmToken::EndOfStatement)) 3904 return TokError("unexpected token in '.ifb' directive"); 3905 3906 Lex(); 3907 3908 TheCondState.CondMet = ExpectBlank == Str.empty(); 3909 TheCondState.Ignore = !TheCondState.CondMet; 3910 } 3911 3912 return false; 3913 } 3914 3915 /// parseDirectiveIfc 3916 /// ::= .ifc string1, string2 3917 /// ::= .ifnc string1, string2 3918 bool AsmParser::parseDirectiveIfc(SMLoc DirectiveLoc, bool ExpectEqual) { 3919 TheCondStack.push_back(TheCondState); 3920 TheCondState.TheCond = AsmCond::IfCond; 3921 3922 if (TheCondState.Ignore) { 3923 eatToEndOfStatement(); 3924 } else { 3925 StringRef Str1 = parseStringToComma(); 3926 3927 if (getLexer().isNot(AsmToken::Comma)) 3928 return TokError("unexpected token in '.ifc' directive"); 3929 3930 Lex(); 3931 3932 StringRef Str2 = parseStringToEndOfStatement(); 3933 3934 if (getLexer().isNot(AsmToken::EndOfStatement)) 3935 return TokError("unexpected token in '.ifc' directive"); 3936 3937 Lex(); 3938 3939 TheCondState.CondMet = ExpectEqual == (Str1.trim() == Str2.trim()); 3940 TheCondState.Ignore = !TheCondState.CondMet; 3941 } 3942 3943 return false; 3944 } 3945 3946 /// parseDirectiveIfeqs 3947 /// ::= .ifeqs string1, string2 3948 bool AsmParser::parseDirectiveIfeqs(SMLoc DirectiveLoc, bool ExpectEqual) { 3949 if (Lexer.isNot(AsmToken::String)) { 3950 if (ExpectEqual) 3951 TokError("expected string parameter for '.ifeqs' directive"); 3952 else 3953 TokError("expected string parameter for '.ifnes' directive"); 3954 eatToEndOfStatement(); 3955 return true; 3956 } 3957 3958 StringRef String1 = getTok().getStringContents(); 3959 Lex(); 3960 3961 if (Lexer.isNot(AsmToken::Comma)) { 3962 if (ExpectEqual) 3963 TokError("expected comma after first string for '.ifeqs' directive"); 3964 else 3965 TokError("expected comma after first string for '.ifnes' directive"); 3966 eatToEndOfStatement(); 3967 return true; 3968 } 3969 3970 Lex(); 3971 3972 if (Lexer.isNot(AsmToken::String)) { 3973 if (ExpectEqual) 3974 TokError("expected string parameter for '.ifeqs' directive"); 3975 else 3976 TokError("expected string parameter for '.ifnes' directive"); 3977 eatToEndOfStatement(); 3978 return true; 3979 } 3980 3981 StringRef String2 = getTok().getStringContents(); 3982 Lex(); 3983 3984 TheCondStack.push_back(TheCondState); 3985 TheCondState.TheCond = AsmCond::IfCond; 3986 TheCondState.CondMet = ExpectEqual == (String1 == String2); 3987 TheCondState.Ignore = !TheCondState.CondMet; 3988 3989 return false; 3990 } 3991 3992 /// parseDirectiveIfdef 3993 /// ::= .ifdef symbol 3994 bool AsmParser::parseDirectiveIfdef(SMLoc DirectiveLoc, bool expect_defined) { 3995 StringRef Name; 3996 TheCondStack.push_back(TheCondState); 3997 TheCondState.TheCond = AsmCond::IfCond; 3998 3999 if (TheCondState.Ignore) { 4000 eatToEndOfStatement(); 4001 } else { 4002 if (parseIdentifier(Name)) 4003 return TokError("expected identifier after '.ifdef'"); 4004 4005 Lex(); 4006 4007 MCSymbol *Sym = getContext().LookupSymbol(Name); 4008 4009 if (expect_defined) 4010 TheCondState.CondMet = (Sym && !Sym->isUndefined()); 4011 else 4012 TheCondState.CondMet = (!Sym || Sym->isUndefined()); 4013 TheCondState.Ignore = !TheCondState.CondMet; 4014 } 4015 4016 return false; 4017 } 4018 4019 /// parseDirectiveElseIf 4020 /// ::= .elseif expression 4021 bool AsmParser::parseDirectiveElseIf(SMLoc DirectiveLoc) { 4022 if (TheCondState.TheCond != AsmCond::IfCond && 4023 TheCondState.TheCond != AsmCond::ElseIfCond) 4024 Error(DirectiveLoc, "Encountered a .elseif that doesn't follow a .if or " 4025 " an .elseif"); 4026 TheCondState.TheCond = AsmCond::ElseIfCond; 4027 4028 bool LastIgnoreState = false; 4029 if (!TheCondStack.empty()) 4030 LastIgnoreState = TheCondStack.back().Ignore; 4031 if (LastIgnoreState || TheCondState.CondMet) { 4032 TheCondState.Ignore = true; 4033 eatToEndOfStatement(); 4034 } else { 4035 int64_t ExprValue; 4036 if (parseAbsoluteExpression(ExprValue)) 4037 return true; 4038 4039 if (getLexer().isNot(AsmToken::EndOfStatement)) 4040 return TokError("unexpected token in '.elseif' directive"); 4041 4042 Lex(); 4043 TheCondState.CondMet = ExprValue; 4044 TheCondState.Ignore = !TheCondState.CondMet; 4045 } 4046 4047 return false; 4048 } 4049 4050 /// parseDirectiveElse 4051 /// ::= .else 4052 bool AsmParser::parseDirectiveElse(SMLoc DirectiveLoc) { 4053 if (getLexer().isNot(AsmToken::EndOfStatement)) 4054 return TokError("unexpected token in '.else' directive"); 4055 4056 Lex(); 4057 4058 if (TheCondState.TheCond != AsmCond::IfCond && 4059 TheCondState.TheCond != AsmCond::ElseIfCond) 4060 Error(DirectiveLoc, "Encountered a .else that doesn't follow a .if or an " 4061 ".elseif"); 4062 TheCondState.TheCond = AsmCond::ElseCond; 4063 bool LastIgnoreState = false; 4064 if (!TheCondStack.empty()) 4065 LastIgnoreState = TheCondStack.back().Ignore; 4066 if (LastIgnoreState || TheCondState.CondMet) 4067 TheCondState.Ignore = true; 4068 else 4069 TheCondState.Ignore = false; 4070 4071 return false; 4072 } 4073 4074 /// parseDirectiveEnd 4075 /// ::= .end 4076 bool AsmParser::parseDirectiveEnd(SMLoc DirectiveLoc) { 4077 if (getLexer().isNot(AsmToken::EndOfStatement)) 4078 return TokError("unexpected token in '.end' directive"); 4079 4080 Lex(); 4081 4082 while (Lexer.isNot(AsmToken::Eof)) 4083 Lex(); 4084 4085 return false; 4086 } 4087 4088 /// parseDirectiveError 4089 /// ::= .err 4090 /// ::= .error [string] 4091 bool AsmParser::parseDirectiveError(SMLoc L, bool WithMessage) { 4092 if (!TheCondStack.empty()) { 4093 if (TheCondStack.back().Ignore) { 4094 eatToEndOfStatement(); 4095 return false; 4096 } 4097 } 4098 4099 if (!WithMessage) 4100 return Error(L, ".err encountered"); 4101 4102 StringRef Message = ".error directive invoked in source file"; 4103 if (Lexer.isNot(AsmToken::EndOfStatement)) { 4104 if (Lexer.isNot(AsmToken::String)) { 4105 TokError(".error argument must be a string"); 4106 eatToEndOfStatement(); 4107 return true; 4108 } 4109 4110 Message = getTok().getStringContents(); 4111 Lex(); 4112 } 4113 4114 Error(L, Message); 4115 return true; 4116 } 4117 4118 /// parseDirectiveWarning 4119 /// ::= .warning [string] 4120 bool AsmParser::parseDirectiveWarning(SMLoc L) { 4121 if (!TheCondStack.empty()) { 4122 if (TheCondStack.back().Ignore) { 4123 eatToEndOfStatement(); 4124 return false; 4125 } 4126 } 4127 4128 StringRef Message = ".warning directive invoked in source file"; 4129 if (Lexer.isNot(AsmToken::EndOfStatement)) { 4130 if (Lexer.isNot(AsmToken::String)) { 4131 TokError(".warning argument must be a string"); 4132 eatToEndOfStatement(); 4133 return true; 4134 } 4135 4136 Message = getTok().getStringContents(); 4137 Lex(); 4138 } 4139 4140 Warning(L, Message); 4141 return false; 4142 } 4143 4144 /// parseDirectiveEndIf 4145 /// ::= .endif 4146 bool AsmParser::parseDirectiveEndIf(SMLoc DirectiveLoc) { 4147 if (getLexer().isNot(AsmToken::EndOfStatement)) 4148 return TokError("unexpected token in '.endif' directive"); 4149 4150 Lex(); 4151 4152 if ((TheCondState.TheCond == AsmCond::NoCond) || TheCondStack.empty()) 4153 Error(DirectiveLoc, "Encountered a .endif that doesn't follow a .if or " 4154 ".else"); 4155 if (!TheCondStack.empty()) { 4156 TheCondState = TheCondStack.back(); 4157 TheCondStack.pop_back(); 4158 } 4159 4160 return false; 4161 } 4162 4163 void AsmParser::initializeDirectiveKindMap() { 4164 DirectiveKindMap[".set"] = DK_SET; 4165 DirectiveKindMap[".equ"] = DK_EQU; 4166 DirectiveKindMap[".equiv"] = DK_EQUIV; 4167 DirectiveKindMap[".ascii"] = DK_ASCII; 4168 DirectiveKindMap[".asciz"] = DK_ASCIZ; 4169 DirectiveKindMap[".string"] = DK_STRING; 4170 DirectiveKindMap[".byte"] = DK_BYTE; 4171 DirectiveKindMap[".short"] = DK_SHORT; 4172 DirectiveKindMap[".value"] = DK_VALUE; 4173 DirectiveKindMap[".2byte"] = DK_2BYTE; 4174 DirectiveKindMap[".long"] = DK_LONG; 4175 DirectiveKindMap[".int"] = DK_INT; 4176 DirectiveKindMap[".4byte"] = DK_4BYTE; 4177 DirectiveKindMap[".quad"] = DK_QUAD; 4178 DirectiveKindMap[".8byte"] = DK_8BYTE; 4179 DirectiveKindMap[".octa"] = DK_OCTA; 4180 DirectiveKindMap[".single"] = DK_SINGLE; 4181 DirectiveKindMap[".float"] = DK_FLOAT; 4182 DirectiveKindMap[".double"] = DK_DOUBLE; 4183 DirectiveKindMap[".align"] = DK_ALIGN; 4184 DirectiveKindMap[".align32"] = DK_ALIGN32; 4185 DirectiveKindMap[".balign"] = DK_BALIGN; 4186 DirectiveKindMap[".balignw"] = DK_BALIGNW; 4187 DirectiveKindMap[".balignl"] = DK_BALIGNL; 4188 DirectiveKindMap[".p2align"] = DK_P2ALIGN; 4189 DirectiveKindMap[".p2alignw"] = DK_P2ALIGNW; 4190 DirectiveKindMap[".p2alignl"] = DK_P2ALIGNL; 4191 DirectiveKindMap[".org"] = DK_ORG; 4192 DirectiveKindMap[".fill"] = DK_FILL; 4193 DirectiveKindMap[".zero"] = DK_ZERO; 4194 DirectiveKindMap[".extern"] = DK_EXTERN; 4195 DirectiveKindMap[".globl"] = DK_GLOBL; 4196 DirectiveKindMap[".global"] = DK_GLOBAL; 4197 DirectiveKindMap[".lazy_reference"] = DK_LAZY_REFERENCE; 4198 DirectiveKindMap[".no_dead_strip"] = DK_NO_DEAD_STRIP; 4199 DirectiveKindMap[".symbol_resolver"] = DK_SYMBOL_RESOLVER; 4200 DirectiveKindMap[".private_extern"] = DK_PRIVATE_EXTERN; 4201 DirectiveKindMap[".reference"] = DK_REFERENCE; 4202 DirectiveKindMap[".weak_definition"] = DK_WEAK_DEFINITION; 4203 DirectiveKindMap[".weak_reference"] = DK_WEAK_REFERENCE; 4204 DirectiveKindMap[".weak_def_can_be_hidden"] = DK_WEAK_DEF_CAN_BE_HIDDEN; 4205 DirectiveKindMap[".comm"] = DK_COMM; 4206 DirectiveKindMap[".common"] = DK_COMMON; 4207 DirectiveKindMap[".lcomm"] = DK_LCOMM; 4208 DirectiveKindMap[".abort"] = DK_ABORT; 4209 DirectiveKindMap[".include"] = DK_INCLUDE; 4210 DirectiveKindMap[".incbin"] = DK_INCBIN; 4211 DirectiveKindMap[".code16"] = DK_CODE16; 4212 DirectiveKindMap[".code16gcc"] = DK_CODE16GCC; 4213 DirectiveKindMap[".rept"] = DK_REPT; 4214 DirectiveKindMap[".rep"] = DK_REPT; 4215 DirectiveKindMap[".irp"] = DK_IRP; 4216 DirectiveKindMap[".irpc"] = DK_IRPC; 4217 DirectiveKindMap[".endr"] = DK_ENDR; 4218 DirectiveKindMap[".bundle_align_mode"] = DK_BUNDLE_ALIGN_MODE; 4219 DirectiveKindMap[".bundle_lock"] = DK_BUNDLE_LOCK; 4220 DirectiveKindMap[".bundle_unlock"] = DK_BUNDLE_UNLOCK; 4221 DirectiveKindMap[".if"] = DK_IF; 4222 DirectiveKindMap[".ifeq"] = DK_IFEQ; 4223 DirectiveKindMap[".ifge"] = DK_IFGE; 4224 DirectiveKindMap[".ifgt"] = DK_IFGT; 4225 DirectiveKindMap[".ifle"] = DK_IFLE; 4226 DirectiveKindMap[".iflt"] = DK_IFLT; 4227 DirectiveKindMap[".ifne"] = DK_IFNE; 4228 DirectiveKindMap[".ifb"] = DK_IFB; 4229 DirectiveKindMap[".ifnb"] = DK_IFNB; 4230 DirectiveKindMap[".ifc"] = DK_IFC; 4231 DirectiveKindMap[".ifeqs"] = DK_IFEQS; 4232 DirectiveKindMap[".ifnc"] = DK_IFNC; 4233 DirectiveKindMap[".ifnes"] = DK_IFNES; 4234 DirectiveKindMap[".ifdef"] = DK_IFDEF; 4235 DirectiveKindMap[".ifndef"] = DK_IFNDEF; 4236 DirectiveKindMap[".ifnotdef"] = DK_IFNOTDEF; 4237 DirectiveKindMap[".elseif"] = DK_ELSEIF; 4238 DirectiveKindMap[".else"] = DK_ELSE; 4239 DirectiveKindMap[".end"] = DK_END; 4240 DirectiveKindMap[".endif"] = DK_ENDIF; 4241 DirectiveKindMap[".skip"] = DK_SKIP; 4242 DirectiveKindMap[".space"] = DK_SPACE; 4243 DirectiveKindMap[".file"] = DK_FILE; 4244 DirectiveKindMap[".line"] = DK_LINE; 4245 DirectiveKindMap[".loc"] = DK_LOC; 4246 DirectiveKindMap[".stabs"] = DK_STABS; 4247 DirectiveKindMap[".sleb128"] = DK_SLEB128; 4248 DirectiveKindMap[".uleb128"] = DK_ULEB128; 4249 DirectiveKindMap[".cfi_sections"] = DK_CFI_SECTIONS; 4250 DirectiveKindMap[".cfi_startproc"] = DK_CFI_STARTPROC; 4251 DirectiveKindMap[".cfi_endproc"] = DK_CFI_ENDPROC; 4252 DirectiveKindMap[".cfi_def_cfa"] = DK_CFI_DEF_CFA; 4253 DirectiveKindMap[".cfi_def_cfa_offset"] = DK_CFI_DEF_CFA_OFFSET; 4254 DirectiveKindMap[".cfi_adjust_cfa_offset"] = DK_CFI_ADJUST_CFA_OFFSET; 4255 DirectiveKindMap[".cfi_def_cfa_register"] = DK_CFI_DEF_CFA_REGISTER; 4256 DirectiveKindMap[".cfi_offset"] = DK_CFI_OFFSET; 4257 DirectiveKindMap[".cfi_rel_offset"] = DK_CFI_REL_OFFSET; 4258 DirectiveKindMap[".cfi_personality"] = DK_CFI_PERSONALITY; 4259 DirectiveKindMap[".cfi_lsda"] = DK_CFI_LSDA; 4260 DirectiveKindMap[".cfi_remember_state"] = DK_CFI_REMEMBER_STATE; 4261 DirectiveKindMap[".cfi_restore_state"] = DK_CFI_RESTORE_STATE; 4262 DirectiveKindMap[".cfi_same_value"] = DK_CFI_SAME_VALUE; 4263 DirectiveKindMap[".cfi_restore"] = DK_CFI_RESTORE; 4264 DirectiveKindMap[".cfi_escape"] = DK_CFI_ESCAPE; 4265 DirectiveKindMap[".cfi_signal_frame"] = DK_CFI_SIGNAL_FRAME; 4266 DirectiveKindMap[".cfi_undefined"] = DK_CFI_UNDEFINED; 4267 DirectiveKindMap[".cfi_register"] = DK_CFI_REGISTER; 4268 DirectiveKindMap[".cfi_window_save"] = DK_CFI_WINDOW_SAVE; 4269 DirectiveKindMap[".macros_on"] = DK_MACROS_ON; 4270 DirectiveKindMap[".macros_off"] = DK_MACROS_OFF; 4271 DirectiveKindMap[".macro"] = DK_MACRO; 4272 DirectiveKindMap[".exitm"] = DK_EXITM; 4273 DirectiveKindMap[".endm"] = DK_ENDM; 4274 DirectiveKindMap[".endmacro"] = DK_ENDMACRO; 4275 DirectiveKindMap[".purgem"] = DK_PURGEM; 4276 DirectiveKindMap[".err"] = DK_ERR; 4277 DirectiveKindMap[".error"] = DK_ERROR; 4278 DirectiveKindMap[".warning"] = DK_WARNING; 4279 } 4280 4281 MCAsmMacro *AsmParser::parseMacroLikeBody(SMLoc DirectiveLoc) { 4282 AsmToken EndToken, StartToken = getTok(); 4283 4284 unsigned NestLevel = 0; 4285 for (;;) { 4286 // Check whether we have reached the end of the file. 4287 if (getLexer().is(AsmToken::Eof)) { 4288 Error(DirectiveLoc, "no matching '.endr' in definition"); 4289 return nullptr; 4290 } 4291 4292 if (Lexer.is(AsmToken::Identifier) && 4293 (getTok().getIdentifier() == ".rept")) { 4294 ++NestLevel; 4295 } 4296 4297 // Otherwise, check whether we have reached the .endr. 4298 if (Lexer.is(AsmToken::Identifier) && getTok().getIdentifier() == ".endr") { 4299 if (NestLevel == 0) { 4300 EndToken = getTok(); 4301 Lex(); 4302 if (Lexer.isNot(AsmToken::EndOfStatement)) { 4303 TokError("unexpected token in '.endr' directive"); 4304 return nullptr; 4305 } 4306 break; 4307 } 4308 --NestLevel; 4309 } 4310 4311 // Otherwise, scan till the end of the statement. 4312 eatToEndOfStatement(); 4313 } 4314 4315 const char *BodyStart = StartToken.getLoc().getPointer(); 4316 const char *BodyEnd = EndToken.getLoc().getPointer(); 4317 StringRef Body = StringRef(BodyStart, BodyEnd - BodyStart); 4318 4319 // We Are Anonymous. 4320 MacroLikeBodies.push_back( 4321 MCAsmMacro(StringRef(), Body, MCAsmMacroParameters())); 4322 return &MacroLikeBodies.back(); 4323 } 4324 4325 void AsmParser::instantiateMacroLikeBody(MCAsmMacro *M, SMLoc DirectiveLoc, 4326 raw_svector_ostream &OS) { 4327 OS << ".endr\n"; 4328 4329 std::unique_ptr<MemoryBuffer> Instantiation = 4330 MemoryBuffer::getMemBufferCopy(OS.str(), "<instantiation>"); 4331 4332 // Create the macro instantiation object and add to the current macro 4333 // instantiation stack. 4334 MacroInstantiation *MI = new MacroInstantiation( 4335 DirectiveLoc, CurBuffer, getTok().getLoc(), TheCondStack.size()); 4336 ActiveMacros.push_back(MI); 4337 4338 // Jump to the macro instantiation and prime the lexer. 4339 CurBuffer = SrcMgr.AddNewSourceBuffer(std::move(Instantiation), SMLoc()); 4340 Lexer.setBuffer(SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer()); 4341 Lex(); 4342 } 4343 4344 /// parseDirectiveRept 4345 /// ::= .rep | .rept count 4346 bool AsmParser::parseDirectiveRept(SMLoc DirectiveLoc, StringRef Dir) { 4347 const MCExpr *CountExpr; 4348 SMLoc CountLoc = getTok().getLoc(); 4349 if (parseExpression(CountExpr)) 4350 return true; 4351 4352 int64_t Count; 4353 if (!CountExpr->EvaluateAsAbsolute(Count)) { 4354 eatToEndOfStatement(); 4355 return Error(CountLoc, "unexpected token in '" + Dir + "' directive"); 4356 } 4357 4358 if (Count < 0) 4359 return Error(CountLoc, "Count is negative"); 4360 4361 if (Lexer.isNot(AsmToken::EndOfStatement)) 4362 return TokError("unexpected token in '" + Dir + "' directive"); 4363 4364 // Eat the end of statement. 4365 Lex(); 4366 4367 // Lex the rept definition. 4368 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 4369 if (!M) 4370 return true; 4371 4372 // Macro instantiation is lexical, unfortunately. We construct a new buffer 4373 // to hold the macro body with substitutions. 4374 SmallString<256> Buf; 4375 raw_svector_ostream OS(Buf); 4376 while (Count--) { 4377 if (expandMacro(OS, M->Body, None, None, getTok().getLoc())) 4378 return true; 4379 } 4380 instantiateMacroLikeBody(M, DirectiveLoc, OS); 4381 4382 return false; 4383 } 4384 4385 /// parseDirectiveIrp 4386 /// ::= .irp symbol,values 4387 bool AsmParser::parseDirectiveIrp(SMLoc DirectiveLoc) { 4388 MCAsmMacroParameter Parameter; 4389 4390 if (parseIdentifier(Parameter.Name)) 4391 return TokError("expected identifier in '.irp' directive"); 4392 4393 if (Lexer.isNot(AsmToken::Comma)) 4394 return TokError("expected comma in '.irp' directive"); 4395 4396 Lex(); 4397 4398 MCAsmMacroArguments A; 4399 if (parseMacroArguments(nullptr, A)) 4400 return true; 4401 4402 // Eat the end of statement. 4403 Lex(); 4404 4405 // Lex the irp definition. 4406 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 4407 if (!M) 4408 return true; 4409 4410 // Macro instantiation is lexical, unfortunately. We construct a new buffer 4411 // to hold the macro body with substitutions. 4412 SmallString<256> Buf; 4413 raw_svector_ostream OS(Buf); 4414 4415 for (MCAsmMacroArguments::iterator i = A.begin(), e = A.end(); i != e; ++i) { 4416 if (expandMacro(OS, M->Body, Parameter, *i, getTok().getLoc())) 4417 return true; 4418 } 4419 4420 instantiateMacroLikeBody(M, DirectiveLoc, OS); 4421 4422 return false; 4423 } 4424 4425 /// parseDirectiveIrpc 4426 /// ::= .irpc symbol,values 4427 bool AsmParser::parseDirectiveIrpc(SMLoc DirectiveLoc) { 4428 MCAsmMacroParameter Parameter; 4429 4430 if (parseIdentifier(Parameter.Name)) 4431 return TokError("expected identifier in '.irpc' directive"); 4432 4433 if (Lexer.isNot(AsmToken::Comma)) 4434 return TokError("expected comma in '.irpc' directive"); 4435 4436 Lex(); 4437 4438 MCAsmMacroArguments A; 4439 if (parseMacroArguments(nullptr, A)) 4440 return true; 4441 4442 if (A.size() != 1 || A.front().size() != 1) 4443 return TokError("unexpected token in '.irpc' directive"); 4444 4445 // Eat the end of statement. 4446 Lex(); 4447 4448 // Lex the irpc definition. 4449 MCAsmMacro *M = parseMacroLikeBody(DirectiveLoc); 4450 if (!M) 4451 return true; 4452 4453 // Macro instantiation is lexical, unfortunately. We construct a new buffer 4454 // to hold the macro body with substitutions. 4455 SmallString<256> Buf; 4456 raw_svector_ostream OS(Buf); 4457 4458 StringRef Values = A.front().front().getString(); 4459 for (std::size_t I = 0, End = Values.size(); I != End; ++I) { 4460 MCAsmMacroArgument Arg; 4461 Arg.push_back(AsmToken(AsmToken::Identifier, Values.slice(I, I + 1))); 4462 4463 if (expandMacro(OS, M->Body, Parameter, Arg, getTok().getLoc())) 4464 return true; 4465 } 4466 4467 instantiateMacroLikeBody(M, DirectiveLoc, OS); 4468 4469 return false; 4470 } 4471 4472 bool AsmParser::parseDirectiveEndr(SMLoc DirectiveLoc) { 4473 if (ActiveMacros.empty()) 4474 return TokError("unmatched '.endr' directive"); 4475 4476 // The only .repl that should get here are the ones created by 4477 // instantiateMacroLikeBody. 4478 assert(getLexer().is(AsmToken::EndOfStatement)); 4479 4480 handleMacroExit(); 4481 return false; 4482 } 4483 4484 bool AsmParser::parseDirectiveMSEmit(SMLoc IDLoc, ParseStatementInfo &Info, 4485 size_t Len) { 4486 const MCExpr *Value; 4487 SMLoc ExprLoc = getLexer().getLoc(); 4488 if (parseExpression(Value)) 4489 return true; 4490 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 4491 if (!MCE) 4492 return Error(ExprLoc, "unexpected expression in _emit"); 4493 uint64_t IntValue = MCE->getValue(); 4494 if (!isUIntN(8, IntValue) && !isIntN(8, IntValue)) 4495 return Error(ExprLoc, "literal value out of range for directive"); 4496 4497 Info.AsmRewrites->push_back(AsmRewrite(AOK_Emit, IDLoc, Len)); 4498 return false; 4499 } 4500 4501 bool AsmParser::parseDirectiveMSAlign(SMLoc IDLoc, ParseStatementInfo &Info) { 4502 const MCExpr *Value; 4503 SMLoc ExprLoc = getLexer().getLoc(); 4504 if (parseExpression(Value)) 4505 return true; 4506 const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Value); 4507 if (!MCE) 4508 return Error(ExprLoc, "unexpected expression in align"); 4509 uint64_t IntValue = MCE->getValue(); 4510 if (!isPowerOf2_64(IntValue)) 4511 return Error(ExprLoc, "literal value not a power of two greater then zero"); 4512 4513 Info.AsmRewrites->push_back( 4514 AsmRewrite(AOK_Align, IDLoc, 5, Log2_64(IntValue))); 4515 return false; 4516 } 4517 4518 // We are comparing pointers, but the pointers are relative to a single string. 4519 // Thus, this should always be deterministic. 4520 static int rewritesSort(const AsmRewrite *AsmRewriteA, 4521 const AsmRewrite *AsmRewriteB) { 4522 if (AsmRewriteA->Loc.getPointer() < AsmRewriteB->Loc.getPointer()) 4523 return -1; 4524 if (AsmRewriteB->Loc.getPointer() < AsmRewriteA->Loc.getPointer()) 4525 return 1; 4526 4527 // It's possible to have a SizeDirective, Imm/ImmPrefix and an Input/Output 4528 // rewrite to the same location. Make sure the SizeDirective rewrite is 4529 // performed first, then the Imm/ImmPrefix and finally the Input/Output. This 4530 // ensures the sort algorithm is stable. 4531 if (AsmRewritePrecedence[AsmRewriteA->Kind] > 4532 AsmRewritePrecedence[AsmRewriteB->Kind]) 4533 return -1; 4534 4535 if (AsmRewritePrecedence[AsmRewriteA->Kind] < 4536 AsmRewritePrecedence[AsmRewriteB->Kind]) 4537 return 1; 4538 llvm_unreachable("Unstable rewrite sort."); 4539 } 4540 4541 bool AsmParser::parseMSInlineAsm( 4542 void *AsmLoc, std::string &AsmString, unsigned &NumOutputs, 4543 unsigned &NumInputs, SmallVectorImpl<std::pair<void *, bool> > &OpDecls, 4544 SmallVectorImpl<std::string> &Constraints, 4545 SmallVectorImpl<std::string> &Clobbers, const MCInstrInfo *MII, 4546 const MCInstPrinter *IP, MCAsmParserSemaCallback &SI) { 4547 SmallVector<void *, 4> InputDecls; 4548 SmallVector<void *, 4> OutputDecls; 4549 SmallVector<bool, 4> InputDeclsAddressOf; 4550 SmallVector<bool, 4> OutputDeclsAddressOf; 4551 SmallVector<std::string, 4> InputConstraints; 4552 SmallVector<std::string, 4> OutputConstraints; 4553 SmallVector<unsigned, 4> ClobberRegs; 4554 4555 SmallVector<AsmRewrite, 4> AsmStrRewrites; 4556 4557 // Prime the lexer. 4558 Lex(); 4559 4560 // While we have input, parse each statement. 4561 unsigned InputIdx = 0; 4562 unsigned OutputIdx = 0; 4563 while (getLexer().isNot(AsmToken::Eof)) { 4564 ParseStatementInfo Info(&AsmStrRewrites); 4565 if (parseStatement(Info, &SI)) 4566 return true; 4567 4568 if (Info.ParseError) 4569 return true; 4570 4571 if (Info.Opcode == ~0U) 4572 continue; 4573 4574 const MCInstrDesc &Desc = MII->get(Info.Opcode); 4575 4576 // Build the list of clobbers, outputs and inputs. 4577 for (unsigned i = 1, e = Info.ParsedOperands.size(); i != e; ++i) { 4578 MCParsedAsmOperand &Operand = *Info.ParsedOperands[i]; 4579 4580 // Immediate. 4581 if (Operand.isImm()) 4582 continue; 4583 4584 // Register operand. 4585 if (Operand.isReg() && !Operand.needAddressOf() && 4586 !getTargetParser().OmitRegisterFromClobberLists(Operand.getReg())) { 4587 unsigned NumDefs = Desc.getNumDefs(); 4588 // Clobber. 4589 if (NumDefs && Operand.getMCOperandNum() < NumDefs) 4590 ClobberRegs.push_back(Operand.getReg()); 4591 continue; 4592 } 4593 4594 // Expr/Input or Output. 4595 StringRef SymName = Operand.getSymName(); 4596 if (SymName.empty()) 4597 continue; 4598 4599 void *OpDecl = Operand.getOpDecl(); 4600 if (!OpDecl) 4601 continue; 4602 4603 bool isOutput = (i == 1) && Desc.mayStore(); 4604 SMLoc Start = SMLoc::getFromPointer(SymName.data()); 4605 if (isOutput) { 4606 ++InputIdx; 4607 OutputDecls.push_back(OpDecl); 4608 OutputDeclsAddressOf.push_back(Operand.needAddressOf()); 4609 OutputConstraints.push_back(("=" + Operand.getConstraint()).str()); 4610 AsmStrRewrites.push_back(AsmRewrite(AOK_Output, Start, SymName.size())); 4611 } else { 4612 InputDecls.push_back(OpDecl); 4613 InputDeclsAddressOf.push_back(Operand.needAddressOf()); 4614 InputConstraints.push_back(Operand.getConstraint().str()); 4615 AsmStrRewrites.push_back(AsmRewrite(AOK_Input, Start, SymName.size())); 4616 } 4617 } 4618 4619 // Consider implicit defs to be clobbers. Think of cpuid and push. 4620 ArrayRef<uint16_t> ImpDefs(Desc.getImplicitDefs(), 4621 Desc.getNumImplicitDefs()); 4622 ClobberRegs.insert(ClobberRegs.end(), ImpDefs.begin(), ImpDefs.end()); 4623 } 4624 4625 // Set the number of Outputs and Inputs. 4626 NumOutputs = OutputDecls.size(); 4627 NumInputs = InputDecls.size(); 4628 4629 // Set the unique clobbers. 4630 array_pod_sort(ClobberRegs.begin(), ClobberRegs.end()); 4631 ClobberRegs.erase(std::unique(ClobberRegs.begin(), ClobberRegs.end()), 4632 ClobberRegs.end()); 4633 Clobbers.assign(ClobberRegs.size(), std::string()); 4634 for (unsigned I = 0, E = ClobberRegs.size(); I != E; ++I) { 4635 raw_string_ostream OS(Clobbers[I]); 4636 IP->printRegName(OS, ClobberRegs[I]); 4637 } 4638 4639 // Merge the various outputs and inputs. Output are expected first. 4640 if (NumOutputs || NumInputs) { 4641 unsigned NumExprs = NumOutputs + NumInputs; 4642 OpDecls.resize(NumExprs); 4643 Constraints.resize(NumExprs); 4644 for (unsigned i = 0; i < NumOutputs; ++i) { 4645 OpDecls[i] = std::make_pair(OutputDecls[i], OutputDeclsAddressOf[i]); 4646 Constraints[i] = OutputConstraints[i]; 4647 } 4648 for (unsigned i = 0, j = NumOutputs; i < NumInputs; ++i, ++j) { 4649 OpDecls[j] = std::make_pair(InputDecls[i], InputDeclsAddressOf[i]); 4650 Constraints[j] = InputConstraints[i]; 4651 } 4652 } 4653 4654 // Build the IR assembly string. 4655 std::string AsmStringIR; 4656 raw_string_ostream OS(AsmStringIR); 4657 StringRef ASMString = 4658 SrcMgr.getMemoryBuffer(SrcMgr.getMainFileID())->getBuffer(); 4659 const char *AsmStart = ASMString.begin(); 4660 const char *AsmEnd = ASMString.end(); 4661 array_pod_sort(AsmStrRewrites.begin(), AsmStrRewrites.end(), rewritesSort); 4662 for (const AsmRewrite &AR : AsmStrRewrites) { 4663 AsmRewriteKind Kind = AR.Kind; 4664 if (Kind == AOK_Delete) 4665 continue; 4666 4667 const char *Loc = AR.Loc.getPointer(); 4668 assert(Loc >= AsmStart && "Expected Loc to be at or after Start!"); 4669 4670 // Emit everything up to the immediate/expression. 4671 if (unsigned Len = Loc - AsmStart) 4672 OS << StringRef(AsmStart, Len); 4673 4674 // Skip the original expression. 4675 if (Kind == AOK_Skip) { 4676 AsmStart = Loc + AR.Len; 4677 continue; 4678 } 4679 4680 unsigned AdditionalSkip = 0; 4681 // Rewrite expressions in $N notation. 4682 switch (Kind) { 4683 default: 4684 break; 4685 case AOK_Imm: 4686 OS << "$$" << AR.Val; 4687 break; 4688 case AOK_ImmPrefix: 4689 OS << "$$"; 4690 break; 4691 case AOK_Label: 4692 OS << Ctx.getAsmInfo()->getPrivateLabelPrefix() << AR.Label; 4693 break; 4694 case AOK_Input: 4695 OS << '$' << InputIdx++; 4696 break; 4697 case AOK_Output: 4698 OS << '$' << OutputIdx++; 4699 break; 4700 case AOK_SizeDirective: 4701 switch (AR.Val) { 4702 default: break; 4703 case 8: OS << "byte ptr "; break; 4704 case 16: OS << "word ptr "; break; 4705 case 32: OS << "dword ptr "; break; 4706 case 64: OS << "qword ptr "; break; 4707 case 80: OS << "xword ptr "; break; 4708 case 128: OS << "xmmword ptr "; break; 4709 case 256: OS << "ymmword ptr "; break; 4710 } 4711 break; 4712 case AOK_Emit: 4713 OS << ".byte"; 4714 break; 4715 case AOK_Align: { 4716 unsigned Val = AR.Val; 4717 OS << ".align " << Val; 4718 4719 // Skip the original immediate. 4720 assert(Val < 10 && "Expected alignment less then 2^10."); 4721 AdditionalSkip = (Val < 4) ? 2 : Val < 7 ? 3 : 4; 4722 break; 4723 } 4724 case AOK_DotOperator: 4725 // Insert the dot if the user omitted it. 4726 OS.flush(); 4727 if (AsmStringIR.back() != '.') 4728 OS << '.'; 4729 OS << AR.Val; 4730 break; 4731 } 4732 4733 // Skip the original expression. 4734 AsmStart = Loc + AR.Len + AdditionalSkip; 4735 } 4736 4737 // Emit the remainder of the asm string. 4738 if (AsmStart != AsmEnd) 4739 OS << StringRef(AsmStart, AsmEnd - AsmStart); 4740 4741 AsmString = OS.str(); 4742 return false; 4743 } 4744 4745 /// \brief Create an MCAsmParser instance. 4746 MCAsmParser *llvm::createMCAsmParser(SourceMgr &SM, MCContext &C, 4747 MCStreamer &Out, const MCAsmInfo &MAI) { 4748 return new AsmParser(SM, C, Out, MAI); 4749 } 4750